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
25 const unsigned char null_sha1[20];
27 static unsigned int sha1_file_open_flag = O_NOATIME;
29 signed char hexval_table[256] = {
30 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
31 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
32 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
33 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
34 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
35 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
36 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
37 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
38 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
42 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
64 int get_sha1_hex(const char *hex, unsigned char *sha1)
67 for (i = 0; i < 20; i++) {
68 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
77 int safe_create_leading_directories(char *path)
86 pos = strchr(pos, '/');
90 if (!stat(path, &st)) {
92 if (!S_ISDIR(st.st_mode)) {
97 else if (mkdir(path, 0777)) {
101 else if (adjust_shared_perm(path)) {
110 char * sha1_to_hex(const unsigned char *sha1)
113 static char hexbuffer[4][50];
114 static const char hex[] = "0123456789abcdef";
115 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
118 for (i = 0; i < 20; i++) {
119 unsigned int val = *sha1++;
120 *buf++ = hex[val >> 4];
121 *buf++ = hex[val & 0xf];
128 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
131 for (i = 0; i < 20; i++) {
132 static char hex[] = "0123456789abcdef";
133 unsigned int val = sha1[i];
134 char *pos = pathbuf + i*2 + (i > 0);
135 *pos++ = hex[val >> 4];
136 *pos = hex[val & 0xf];
141 * NOTE! This returns a statically allocated buffer, so you have to be
142 * careful about using it. Do a "xstrdup()" if you need to save the
145 * Also note that this returns the location for creating. Reading
146 * SHA1 file can happen from any alternate directory listed in the
147 * DB_ENVIRONMENT environment variable if it is not found in
148 * the primary object database.
150 char *sha1_file_name(const unsigned char *sha1)
152 static char *name, *base;
155 const char *sha1_file_directory = get_object_directory();
156 int len = strlen(sha1_file_directory);
157 base = xmalloc(len + 60);
158 memcpy(base, sha1_file_directory, len);
159 memset(base+len, 0, 60);
162 name = base + len + 1;
164 fill_sha1_path(name, sha1);
168 char *sha1_pack_name(const unsigned char *sha1)
170 static const char hex[] = "0123456789abcdef";
171 static char *name, *base, *buf;
175 const char *sha1_file_directory = get_object_directory();
176 int len = strlen(sha1_file_directory);
177 base = xmalloc(len + 60);
178 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
179 name = base + len + 11;
184 for (i = 0; i < 20; i++) {
185 unsigned int val = *sha1++;
186 *buf++ = hex[val >> 4];
187 *buf++ = hex[val & 0xf];
193 char *sha1_pack_index_name(const unsigned char *sha1)
195 static const char hex[] = "0123456789abcdef";
196 static char *name, *base, *buf;
200 const char *sha1_file_directory = get_object_directory();
201 int len = strlen(sha1_file_directory);
202 base = xmalloc(len + 60);
203 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
204 name = base + len + 11;
209 for (i = 0; i < 20; i++) {
210 unsigned int val = *sha1++;
211 *buf++ = hex[val >> 4];
212 *buf++ = hex[val & 0xf];
218 struct alternate_object_database *alt_odb_list;
219 static struct alternate_object_database **alt_odb_tail;
221 static void read_info_alternates(const char * alternates, int depth);
224 * Prepare alternate object database registry.
226 * The variable alt_odb_list points at the list of struct
227 * alternate_object_database. The elements on this list come from
228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
230 * whose contents is similar to that environment variable but can be
231 * LF separated. Its base points at a statically allocated buffer that
232 * contains "/the/directory/corresponding/to/.git/objects/...", while
233 * its name points just after the slash at the end of ".git/objects/"
234 * in the example above, and has enough space to hold 40-byte hex
235 * SHA1, an extra slash for the first level indirection, and the
238 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
241 const char *objdir = get_object_directory();
242 struct alternate_object_database *ent;
243 struct alternate_object_database *alt;
244 /* 43 = 40-byte + 2 '/' + terminating NUL */
246 int entlen = pfxlen + 43;
249 if (*entry != '/' && relative_base) {
250 /* Relative alt-odb */
252 base_len = strlen(relative_base) + 1;
256 ent = xmalloc(sizeof(*ent) + entlen);
258 if (*entry != '/' && relative_base) {
259 memcpy(ent->base, relative_base, base_len - 1);
260 ent->base[base_len - 1] = '/';
261 memcpy(ent->base + base_len, entry, len);
264 memcpy(ent->base, entry, pfxlen);
266 ent->name = ent->base + pfxlen + 1;
267 ent->base[pfxlen + 3] = '/';
268 ent->base[pfxlen] = ent->base[entlen-1] = 0;
270 /* Detect cases where alternate disappeared */
271 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
272 error("object directory %s does not exist; "
273 "check .git/objects/info/alternates.",
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
282 for (alt = alt_odb_list; alt; alt = alt->next) {
283 if (!memcmp(ent->base, alt->base, pfxlen)) {
288 if (!memcmp(ent->base, objdir, pfxlen)) {
293 /* add the alternate entry */
295 alt_odb_tail = &(ent->next);
298 /* recursively add alternates */
299 read_info_alternates(ent->base, depth + 1);
301 ent->base[pfxlen] = '/';
306 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307 const char *relative_base, int depth)
309 const char *cp, *last;
312 error("%s: ignoring alternate object stores, nesting too deep.",
320 if (cp < ep && *cp == '#') {
321 while (cp < ep && *cp != sep)
326 while (cp < ep && *cp != sep)
329 if ((*last != '/') && depth) {
330 error("%s: ignoring relative alternate object store %s",
331 relative_base, last);
333 link_alt_odb_entry(last, cp - last,
334 relative_base, depth);
337 while (cp < ep && *cp == sep)
343 static void read_info_alternates(const char * relative_base, int depth)
350 sprintf(path, "%s/info/alternates", relative_base);
351 fd = open(path, O_RDONLY);
354 if (fstat(fd, &st) || (st.st_size == 0)) {
358 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
360 if (map == MAP_FAILED)
363 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
365 munmap(map, st.st_size);
368 void prepare_alt_odb(void)
372 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 alt_odb_tail = &alt_odb_list;
378 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
380 read_info_alternates(get_object_directory(), 0);
383 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
385 char *name = sha1_file_name(sha1);
386 struct alternate_object_database *alt;
391 for (alt = alt_odb_list; alt; alt = alt->next) {
393 fill_sha1_path(name, sha1);
394 if (!stat(alt->base, st))
400 static unsigned int pack_used_ctr;
401 static size_t pack_mapped;
402 static size_t page_size;
403 struct packed_git *packed_git;
405 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
410 unsigned long idx_size;
412 int fd = open(path, O_RDONLY);
416 if (fstat(fd, &st)) {
420 idx_size = st.st_size;
421 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
423 if (idx_map == MAP_FAILED)
428 *idx_size_ = idx_size;
430 /* check index map */
431 if (idx_size < 4*256 + 20 + 20)
432 return error("index file too small");
434 for (i = 0; i < 256; i++) {
435 unsigned int n = ntohl(index[i]);
437 return error("non-monotonic index");
443 * - 256 index entries 4 bytes each
444 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
445 * - 20-byte SHA1 of the packfile
446 * - 20-byte SHA1 file checksum
448 if (idx_size != 4*256 + nr * 24 + 20 + 20)
449 return error("wrong index file size");
454 static void scan_windows(struct packed_git *p,
455 struct packed_git **lru_p,
456 struct pack_window **lru_w,
457 struct pack_window **lru_l)
459 struct pack_window *w, *w_l;
461 for (w_l = NULL, w = p->windows; w; w = w->next) {
463 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
473 static int unuse_one_window(struct packed_git *current)
475 struct packed_git *p, *lru_p = NULL;
476 struct pack_window *lru_w = NULL, *lru_l = NULL;
479 scan_windows(current, &lru_p, &lru_w, &lru_l);
480 for (p = packed_git; p; p = p->next)
481 scan_windows(p, &lru_p, &lru_w, &lru_l);
483 munmap(lru_w->base, lru_w->len);
484 pack_mapped -= lru_w->len;
486 lru_l->next = lru_w->next;
488 lru_p->windows = lru_w->next;
489 if (!lru_p->windows && lru_p != current) {
490 close(lru_p->pack_fd);
500 void unuse_pack(struct pack_window **w_cursor)
502 struct pack_window *w = *w_cursor;
509 static void open_packed_git(struct packed_git *p)
512 struct pack_header hdr;
513 unsigned char sha1[20];
514 unsigned char *idx_sha1;
516 p->pack_fd = open(p->pack_name, O_RDONLY);
517 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
518 die("packfile %s cannot be opened", p->pack_name);
520 /* If we created the struct before we had the pack we lack size. */
522 if (!S_ISREG(st.st_mode))
523 die("packfile %s not a regular file", p->pack_name);
524 p->pack_size = st.st_size;
525 } else if (p->pack_size != st.st_size)
526 die("packfile %s size changed", p->pack_name);
528 /* Verify we recognize this pack file format. */
529 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
530 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
531 die("file %s is not a GIT packfile", p->pack_name);
532 if (!pack_version_ok(hdr.hdr_version))
533 die("packfile %s is version %u and not supported"
534 " (try upgrading GIT to a newer version)",
535 p->pack_name, ntohl(hdr.hdr_version));
537 /* Verify the pack matches its index. */
538 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
539 die("packfile %s claims to have %u objects"
540 " while index size indicates %u objects",
541 p->pack_name, ntohl(hdr.hdr_entries),
542 num_packed_objects(p));
543 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
544 die("end of packfile %s is unavailable", p->pack_name);
545 read_or_die(p->pack_fd, sha1, sizeof(sha1));
546 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
547 if (hashcmp(sha1, idx_sha1))
548 die("packfile %s does not match index", p->pack_name);
551 static int in_window(struct pack_window *win, unsigned long offset)
553 /* We must promise at least 20 bytes (one hash) after the
554 * offset is available from this window, otherwise the offset
555 * is not actually in this window and a different window (which
556 * has that one hash excess) must be used. This is to support
557 * the object header and delta base parsing routines below.
559 off_t win_off = win->offset;
560 return win_off <= offset
561 && (offset + 20) <= (win_off + win->len);
564 unsigned char* use_pack(struct packed_git *p,
565 struct pack_window **w_cursor,
566 unsigned long offset,
569 struct pack_window *win = *w_cursor;
571 if (p->pack_fd == -1)
574 /* Since packfiles end in a hash of their content and its
575 * pointless to ask for an offset into the middle of that
576 * hash, and the in_window function above wouldn't match
577 * don't allow an offset too close to the end of the file.
579 if (offset > (p->pack_size - 20))
580 die("offset beyond end of packfile (truncated pack?)");
582 if (!win || !in_window(win, offset)) {
585 for (win = p->windows; win; win = win->next) {
586 if (in_window(win, offset))
591 page_size = getpagesize();
592 win = xcalloc(1, sizeof(*win));
593 win->offset = (offset / page_size) * page_size;
594 win->len = p->pack_size - win->offset;
595 if (win->len > packed_git_window_size)
596 win->len = packed_git_window_size;
597 pack_mapped += win->len;
598 while (packed_git_limit < pack_mapped
599 && unuse_one_window(p))
601 win->base = mmap(NULL, win->len,
602 PROT_READ, MAP_PRIVATE,
603 p->pack_fd, win->offset);
604 if (win->base == MAP_FAILED)
605 die("packfile %s cannot be mapped: %s",
608 win->next = p->windows;
612 if (win != *w_cursor) {
613 win->last_used = pack_used_ctr++;
617 offset -= win->offset;
619 *left = win->len - offset;
620 return win->base + offset;
623 struct packed_git *add_packed_git(char *path, int path_len, int local)
626 struct packed_git *p;
627 unsigned long idx_size;
629 unsigned char sha1[20];
631 if (check_packed_git_idx(path, &idx_size, &idx_map))
634 /* do we have a corresponding .pack file? */
635 strcpy(path + path_len - 4, ".pack");
636 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
637 munmap(idx_map, idx_size);
640 /* ok, it looks sane as far as we can check without
641 * actually mapping the pack file.
643 p = xmalloc(sizeof(*p) + path_len + 2);
644 strcpy(p->pack_name, path);
645 p->index_size = idx_size;
646 p->pack_size = st.st_size;
647 p->index_base = idx_map;
651 p->pack_local = local;
652 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
653 hashcpy(p->sha1, sha1);
657 struct packed_git *parse_pack_index(unsigned char *sha1)
659 char *path = sha1_pack_index_name(sha1);
660 return parse_pack_index_file(sha1, path);
663 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
665 struct packed_git *p;
666 unsigned long idx_size;
670 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
673 path = sha1_pack_name(sha1);
675 p = xmalloc(sizeof(*p) + strlen(path) + 2);
676 strcpy(p->pack_name, path);
677 p->index_size = idx_size;
679 p->index_base = idx_map;
683 hashcpy(p->sha1, sha1);
687 void install_packed_git(struct packed_git *pack)
689 pack->next = packed_git;
693 static void prepare_packed_git_one(char *objdir, int local)
700 sprintf(path, "%s/pack", objdir);
705 error("unable to open object pack directory: %s: %s",
706 path, strerror(errno));
710 while ((de = readdir(dir)) != NULL) {
711 int namelen = strlen(de->d_name);
712 struct packed_git *p;
714 if (!has_extension(de->d_name, ".idx"))
717 /* we have .idx. Is it a file we can map? */
718 strcpy(path + len, de->d_name);
719 for (p = packed_git; p; p = p->next) {
720 if (!memcmp(path, p->pack_name, len + namelen - 4))
725 p = add_packed_git(path, len + namelen, local);
728 p->next = packed_git;
734 static int prepare_packed_git_run_once = 0;
735 void prepare_packed_git(void)
737 struct alternate_object_database *alt;
739 if (prepare_packed_git_run_once)
741 prepare_packed_git_one(get_object_directory(), 1);
743 for (alt = alt_odb_list; alt; alt = alt->next) {
745 prepare_packed_git_one(alt->base, 0);
748 prepare_packed_git_run_once = 1;
751 void reprepare_packed_git(void)
753 prepare_packed_git_run_once = 0;
754 prepare_packed_git();
757 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
759 unsigned char real_sha1[20];
760 hash_sha1_file(map, size, type, real_sha1);
761 return hashcmp(sha1, real_sha1) ? -1 : 0;
764 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
769 char *filename = find_sha1_file(sha1, &st);
775 fd = open(filename, O_RDONLY | sha1_file_open_flag);
777 /* See if it works without O_NOATIME */
778 switch (sha1_file_open_flag) {
780 fd = open(filename, O_RDONLY);
788 /* If it failed once, it will probably fail again.
789 * Stop using O_NOATIME
791 sha1_file_open_flag = 0;
793 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
795 if (map == MAP_FAILED)
801 int legacy_loose_object(unsigned char *map)
806 * Is it a zlib-compressed buffer? If so, the first byte
807 * must be 0x78 (15-bit window size, deflated), and the
808 * first 16-bit word is evenly divisible by 31
810 word = (map[0] << 8) + map[1];
811 if (map[0] == 0x78 && !(word % 31))
817 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
822 unsigned long used = 0;
825 *type = (c >> 4) & 7;
831 if (sizeof(long) * 8 <= shift)
834 size += (c & 0x7f) << shift;
841 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
843 unsigned long size, used;
844 static const char valid_loose_object_type[8] = {
846 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
847 0, /* "delta" and others are invalid in a loose object */
849 enum object_type type;
851 /* Get the data stream */
852 memset(stream, 0, sizeof(*stream));
853 stream->next_in = map;
854 stream->avail_in = mapsize;
855 stream->next_out = buffer;
856 stream->avail_out = bufsiz;
858 if (legacy_loose_object(map)) {
860 return inflate(stream, 0);
863 used = unpack_object_header_gently(map, mapsize, &type, &size);
864 if (!used || !valid_loose_object_type[type])
869 /* Set up the stream for the rest.. */
870 stream->next_in = map;
871 stream->avail_in = mapsize;
874 /* And generate the fake traditional header */
875 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
876 type_names[type], size);
880 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
882 int bytes = strlen(buffer) + 1;
883 unsigned char *buf = xmalloc(1+size);
886 n = stream->total_out - bytes;
889 memcpy(buf, (char *) buffer + bytes, n);
892 stream->next_out = buf + bytes;
893 stream->avail_out = size - bytes;
894 while (inflate(stream, Z_FINISH) == Z_OK)
903 * We used to just use "sscanf()", but that's actually way
904 * too permissive for what we want to check. So do an anal
905 * object header parse by hand.
907 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
913 * The type can be at most ten bytes (including the
914 * terminating '\0' that we add), and is followed by
929 * The length must follow immediately, and be in canonical
930 * decimal format (ie "010" is not valid).
937 unsigned long c = *hdr - '0';
941 size = size * 10 + c;
947 * The length must be followed by a zero byte
949 return *hdr ? -1 : 0;
952 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
958 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
959 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
962 return unpack_sha1_rest(&stream, hdr, *size);
965 static unsigned long get_delta_base(struct packed_git *p,
966 struct pack_window **w_curs,
967 unsigned long offset,
968 enum object_type kind,
969 unsigned long delta_obj_offset,
970 unsigned long *base_obj_offset)
972 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
973 unsigned long base_offset;
975 /* use_pack() assured us we have [base_info, base_info + 20)
976 * as a range that we can look at without walking off the
977 * end of the mapped window. Its actually the hash size
978 * that is assured. An OFS_DELTA longer than the hash size
979 * is stupid, as then a REF_DELTA would be smaller to store.
981 if (kind == OBJ_OFS_DELTA) {
983 unsigned char c = base_info[used++];
984 base_offset = c & 127;
987 if (!base_offset || base_offset & ~(~0UL >> 7))
988 die("offset value overflow for delta base object");
989 c = base_info[used++];
990 base_offset = (base_offset << 7) + (c & 127);
992 base_offset = delta_obj_offset - base_offset;
993 if (base_offset >= delta_obj_offset)
994 die("delta base offset out of bound");
996 } else if (kind == OBJ_REF_DELTA) {
997 /* The base entry _must_ be in the same pack */
998 base_offset = find_pack_entry_one(base_info, p);
1000 die("failed to find delta-pack base object %s",
1001 sha1_to_hex(base_info));
1004 die("I am totally screwed");
1005 *base_obj_offset = base_offset;
1009 /* forward declaration for a mutually recursive function */
1010 static int packed_object_info(struct packed_git *p, unsigned long offset,
1011 char *type, unsigned long *sizep);
1013 static int packed_delta_info(struct packed_git *p,
1014 struct pack_window **w_curs,
1015 unsigned long offset,
1016 enum object_type kind,
1017 unsigned long obj_offset,
1019 unsigned long *sizep)
1021 unsigned long base_offset;
1023 offset = get_delta_base(p, w_curs, offset, kind,
1024 obj_offset, &base_offset);
1026 /* We choose to only get the type of the base object and
1027 * ignore potentially corrupt pack file that expects the delta
1028 * based on a base with a wrong size. This saves tons of
1031 if (packed_object_info(p, base_offset, type, NULL))
1032 die("cannot get info for delta-pack base");
1035 const unsigned char *data;
1036 unsigned char delta_head[20], *in;
1037 unsigned long result_size;
1041 memset(&stream, 0, sizeof(stream));
1042 stream.next_out = delta_head;
1043 stream.avail_out = sizeof(delta_head);
1045 inflateInit(&stream);
1047 in = use_pack(p, w_curs, offset, &stream.avail_in);
1048 stream.next_in = in;
1049 st = inflate(&stream, Z_FINISH);
1050 offset += stream.next_in - in;
1051 } while ((st == Z_OK || st == Z_BUF_ERROR)
1052 && stream.total_out < sizeof(delta_head));
1053 inflateEnd(&stream);
1054 if ((st != Z_STREAM_END) &&
1055 stream.total_out != sizeof(delta_head))
1056 die("delta data unpack-initial failed");
1058 /* Examine the initial part of the delta to figure out
1063 /* ignore base size */
1064 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1066 /* Read the result size */
1067 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1068 *sizep = result_size;
1073 static unsigned long unpack_object_header(struct packed_git *p,
1074 struct pack_window **w_curs,
1075 unsigned long offset,
1076 enum object_type *type,
1077 unsigned long *sizep)
1079 unsigned char *base;
1083 /* use_pack() assures us we have [base, base + 20) available
1084 * as a range that we can look at at. (Its actually the hash
1085 * size that is assurred.) With our object header encoding
1086 * the maximum deflated object size is 2^137, which is just
1087 * insane, so we know won't exceed what we have been given.
1089 base = use_pack(p, w_curs, offset, &left);
1090 used = unpack_object_header_gently(base, left, type, sizep);
1092 die("object offset outside of pack file");
1094 return offset + used;
1097 void packed_object_info_detail(struct packed_git *p,
1098 unsigned long offset,
1100 unsigned long *size,
1101 unsigned long *store_size,
1102 unsigned int *delta_chain_length,
1103 unsigned char *base_sha1)
1105 struct pack_window *w_curs = NULL;
1106 unsigned long obj_offset, val;
1107 unsigned char *next_sha1;
1108 enum object_type kind;
1110 *delta_chain_length = 0;
1111 obj_offset = offset;
1112 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1117 die("pack %s contains unknown object type %d",
1118 p->pack_name, kind);
1123 strcpy(type, type_names[kind]);
1124 *store_size = 0; /* notyet */
1125 unuse_pack(&w_curs);
1128 get_delta_base(p, &w_curs, offset, kind,
1129 obj_offset, &offset);
1130 if (*delta_chain_length == 0) {
1131 /* TODO: find base_sha1 as pointed by offset */
1135 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1136 if (*delta_chain_length == 0)
1137 hashcpy(base_sha1, next_sha1);
1138 offset = find_pack_entry_one(next_sha1, p);
1141 obj_offset = offset;
1142 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1143 (*delta_chain_length)++;
1147 static int packed_object_info(struct packed_git *p, unsigned long offset,
1148 char *type, unsigned long *sizep)
1150 struct pack_window *w_curs = NULL;
1151 unsigned long size, obj_offset = offset;
1152 enum object_type kind;
1155 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1160 r = packed_delta_info(p, &w_curs, offset, kind,
1161 obj_offset, type, sizep);
1162 unuse_pack(&w_curs);
1168 strcpy(type, type_names[kind]);
1169 unuse_pack(&w_curs);
1172 die("pack %s contains unknown object type %d",
1173 p->pack_name, kind);
1180 static void *unpack_compressed_entry(struct packed_git *p,
1181 struct pack_window **w_curs,
1182 unsigned long offset,
1187 unsigned char *buffer, *in;
1189 buffer = xmalloc(size + 1);
1191 memset(&stream, 0, sizeof(stream));
1192 stream.next_out = buffer;
1193 stream.avail_out = size;
1195 inflateInit(&stream);
1197 in = use_pack(p, w_curs, offset, &stream.avail_in);
1198 stream.next_in = in;
1199 st = inflate(&stream, Z_FINISH);
1200 offset += stream.next_in - in;
1201 } while (st == Z_OK || st == Z_BUF_ERROR);
1202 inflateEnd(&stream);
1203 if ((st != Z_STREAM_END) || stream.total_out != size) {
1211 static void *unpack_delta_entry(struct packed_git *p,
1212 struct pack_window **w_curs,
1213 unsigned long offset,
1214 unsigned long delta_size,
1215 enum object_type kind,
1216 unsigned long obj_offset,
1218 unsigned long *sizep)
1220 void *delta_data, *result, *base;
1221 unsigned long result_size, base_size, base_offset;
1223 offset = get_delta_base(p, w_curs, offset, kind,
1224 obj_offset, &base_offset);
1225 base = unpack_entry(p, base_offset, type, &base_size);
1227 die("failed to read delta base object at %lu from %s",
1228 base_offset, p->pack_name);
1230 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1231 result = patch_delta(base, base_size,
1232 delta_data, delta_size,
1235 die("failed to apply delta");
1238 *sizep = result_size;
1242 void *unpack_entry(struct packed_git *p, unsigned long offset,
1243 char *type, unsigned long *sizep)
1245 struct pack_window *w_curs = NULL;
1246 unsigned long size, obj_offset = offset;
1247 enum object_type kind;
1250 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1254 retval = unpack_delta_entry(p, &w_curs, offset, size,
1255 kind, obj_offset, type, sizep);
1261 strcpy(type, type_names[kind]);
1263 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1266 die("unknown object type %i in %s", kind, p->pack_name);
1268 unuse_pack(&w_curs);
1272 int num_packed_objects(const struct packed_git *p)
1274 /* See check_packed_git_idx() */
1275 return (p->index_size - 20 - 20 - 4*256) / 24;
1278 int nth_packed_object_sha1(const struct packed_git *p, int n,
1279 unsigned char* sha1)
1281 void *index = p->index_base + 256;
1282 if (n < 0 || num_packed_objects(p) <= n)
1284 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1288 unsigned long find_pack_entry_one(const unsigned char *sha1,
1289 struct packed_git *p)
1291 unsigned int *level1_ofs = p->index_base;
1292 int hi = ntohl(level1_ofs[*sha1]);
1293 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1294 void *index = p->index_base + 256;
1297 int mi = (lo + hi) / 2;
1298 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1300 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1309 static int matches_pack_name(struct packed_git *p, const char *ig)
1311 const char *last_c, *c;
1313 if (!strcmp(p->pack_name, ig))
1316 for (c = p->pack_name, last_c = c; *c;)
1321 if (!strcmp(last_c, ig))
1327 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1329 struct packed_git *p;
1330 unsigned long offset;
1332 prepare_packed_git();
1334 for (p = packed_git; p; p = p->next) {
1335 if (ignore_packed) {
1337 for (ig = ignore_packed; *ig; ig++)
1338 if (!matches_pack_name(p, *ig))
1343 offset = find_pack_entry_one(sha1, p);
1347 hashcpy(e->sha1, sha1);
1354 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1355 struct packed_git *packs)
1357 struct packed_git *p;
1359 for (p = packs; p; p = p->next) {
1360 if (find_pack_entry_one(sha1, p))
1367 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1370 unsigned long mapsize, size;
1375 map = map_sha1_file(sha1, &mapsize);
1377 return error("unable to find %s", sha1_to_hex(sha1));
1378 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1379 status = error("unable to unpack %s header",
1381 if (parse_sha1_header(hdr, type, &size) < 0)
1382 status = error("unable to parse %s header", sha1_to_hex(sha1));
1388 inflateEnd(&stream);
1389 munmap(map, mapsize);
1393 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1395 struct pack_entry e;
1397 if (!find_pack_entry(sha1, &e, NULL)) {
1398 reprepare_packed_git();
1399 if (!find_pack_entry(sha1, &e, NULL))
1400 return sha1_loose_object_info(sha1, type, sizep);
1402 return packed_object_info(e.p, e.offset, type, sizep);
1405 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1407 struct pack_entry e;
1409 if (!find_pack_entry(sha1, &e, NULL)) {
1410 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1413 return unpack_entry(e.p, e.offset, type, size);
1416 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1418 unsigned long mapsize;
1420 struct pack_entry e;
1422 if (find_pack_entry(sha1, &e, NULL))
1423 return read_packed_sha1(sha1, type, size);
1424 map = map_sha1_file(sha1, &mapsize);
1426 buf = unpack_sha1_file(map, mapsize, type, size);
1427 munmap(map, mapsize);
1430 reprepare_packed_git();
1431 if (find_pack_entry(sha1, &e, NULL))
1432 return read_packed_sha1(sha1, type, size);
1436 void *read_object_with_reference(const unsigned char *sha1,
1437 const char *required_type,
1438 unsigned long *size,
1439 unsigned char *actual_sha1_return)
1443 unsigned long isize;
1444 unsigned char actual_sha1[20];
1446 hashcpy(actual_sha1, sha1);
1448 int ref_length = -1;
1449 const char *ref_type = NULL;
1451 buffer = read_sha1_file(actual_sha1, type, &isize);
1454 if (!strcmp(type, required_type)) {
1456 if (actual_sha1_return)
1457 hashcpy(actual_sha1_return, actual_sha1);
1460 /* Handle references */
1461 else if (!strcmp(type, commit_type))
1463 else if (!strcmp(type, tag_type))
1464 ref_type = "object ";
1469 ref_length = strlen(ref_type);
1471 if (memcmp(buffer, ref_type, ref_length) ||
1472 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1477 /* Now we have the ID of the referred-to object in
1478 * actual_sha1. Check again. */
1482 static void write_sha1_file_prepare(void *buf, unsigned long len,
1483 const char *type, unsigned char *sha1,
1484 unsigned char *hdr, int *hdrlen)
1488 /* Generate the header */
1489 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1493 SHA1_Update(&c, hdr, *hdrlen);
1494 SHA1_Update(&c, buf, len);
1495 SHA1_Final(sha1, &c);
1499 * Link the tempfile to the final place, possibly creating the
1500 * last directory level as you do so.
1502 * Returns the errno on failure, 0 on success.
1504 static int link_temp_to_file(const char *tmpfile, const char *filename)
1509 if (!link(tmpfile, filename))
1513 * Try to mkdir the last path component if that failed.
1515 * Re-try the "link()" regardless of whether the mkdir
1516 * succeeds, since a race might mean that somebody
1520 dir = strrchr(filename, '/');
1523 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1528 if (!link(tmpfile, filename))
1536 * Move the just written object into its final resting place
1538 int move_temp_to_file(const char *tmpfile, const char *filename)
1540 int ret = link_temp_to_file(tmpfile, filename);
1543 * Coda hack - coda doesn't like cross-directory links,
1544 * so we fall back to a rename, which will mean that it
1545 * won't be able to check collisions, but that's not a
1548 * The same holds for FAT formatted media.
1550 * When this succeeds, we just return 0. We have nothing
1553 if (ret && ret != EEXIST) {
1554 if (!rename(tmpfile, filename))
1560 if (ret != EEXIST) {
1561 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1563 /* FIXME!!! Collision check here ? */
1569 static int write_buffer(int fd, const void *buf, size_t len)
1574 size = write(fd, buf, len);
1576 return error("file write: disk full");
1578 if (errno == EINTR || errno == EAGAIN)
1580 return error("file write error (%s)", strerror(errno));
1583 buf = (char *) buf + size;
1588 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1593 c = (type << 4) | (len & 15);
1606 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1610 if (use_legacy_headers) {
1611 while (deflate(stream, 0) == Z_OK)
1615 if (!strcmp(type, blob_type))
1616 obj_type = OBJ_BLOB;
1617 else if (!strcmp(type, tree_type))
1618 obj_type = OBJ_TREE;
1619 else if (!strcmp(type, commit_type))
1620 obj_type = OBJ_COMMIT;
1621 else if (!strcmp(type, tag_type))
1624 die("trying to generate bogus object of type '%s'", type);
1625 hdr = write_binary_header(stream->next_out, obj_type, len);
1626 stream->total_out = hdr;
1627 stream->next_out += hdr;
1628 stream->avail_out -= hdr;
1631 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1632 unsigned char *sha1)
1634 unsigned char hdr[50];
1636 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1640 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1643 unsigned char *compressed;
1645 unsigned char sha1[20];
1647 static char tmpfile[PATH_MAX];
1648 unsigned char hdr[50];
1651 /* Normally if we have it in the pack then we do not bother writing
1652 * it out into .git/objects/??/?{38} file.
1654 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1655 filename = sha1_file_name(sha1);
1657 hashcpy(returnsha1, sha1);
1658 if (has_sha1_file(sha1))
1660 fd = open(filename, O_RDONLY);
1663 * FIXME!!! We might do collision checking here, but we'd
1664 * need to uncompress the old file and check it. Later.
1670 if (errno != ENOENT) {
1671 return error("sha1 file %s: %s\n", filename, strerror(errno));
1674 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1676 fd = mkstemp(tmpfile);
1679 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1681 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1685 memset(&stream, 0, sizeof(stream));
1686 deflateInit(&stream, zlib_compression_level);
1687 size = 8 + deflateBound(&stream, len+hdrlen);
1688 compressed = xmalloc(size);
1691 stream.next_out = compressed;
1692 stream.avail_out = size;
1694 /* First header.. */
1695 stream.next_in = hdr;
1696 stream.avail_in = hdrlen;
1697 setup_object_header(&stream, type, len);
1699 /* Then the data itself.. */
1700 stream.next_in = buf;
1701 stream.avail_in = len;
1702 while (deflate(&stream, Z_FINISH) == Z_OK)
1704 deflateEnd(&stream);
1705 size = stream.total_out;
1707 if (write_buffer(fd, compressed, size) < 0)
1708 die("unable to write sha1 file");
1713 return move_temp_to_file(tmpfile, filename);
1717 * We need to unpack and recompress the object for writing
1718 * it out to a different file.
1720 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1724 unsigned char *unpacked;
1731 /* need to unpack and recompress it by itself */
1732 unpacked = read_packed_sha1(sha1, type, &len);
1734 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1737 memset(&stream, 0, sizeof(stream));
1738 deflateInit(&stream, zlib_compression_level);
1739 size = deflateBound(&stream, len + hdrlen);
1740 buf = xmalloc(size);
1743 stream.next_out = buf;
1744 stream.avail_out = size;
1746 /* First header.. */
1747 stream.next_in = (void *)hdr;
1748 stream.avail_in = hdrlen;
1749 while (deflate(&stream, 0) == Z_OK)
1752 /* Then the data itself.. */
1753 stream.next_in = unpacked;
1754 stream.avail_in = len;
1755 while (deflate(&stream, Z_FINISH) == Z_OK)
1757 deflateEnd(&stream);
1760 *objsize = stream.total_out;
1764 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1767 unsigned long objsize;
1768 void *buf = map_sha1_file(sha1, &objsize);
1771 retval = write_buffer(fd, buf, objsize);
1772 munmap(buf, objsize);
1776 buf = repack_object(sha1, &objsize);
1777 retval = write_buffer(fd, buf, objsize);
1782 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1783 size_t bufsize, size_t *bufposn)
1785 char tmpfile[PATH_MAX];
1788 unsigned char real_sha1[20];
1789 unsigned char discard[4096];
1793 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1795 local = mkstemp(tmpfile);
1798 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1800 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1803 memset(&stream, 0, sizeof(stream));
1805 inflateInit(&stream);
1812 stream.avail_in = *bufposn;
1813 stream.next_in = (unsigned char *) buffer;
1815 stream.next_out = discard;
1816 stream.avail_out = sizeof(discard);
1817 ret = inflate(&stream, Z_SYNC_FLUSH);
1818 SHA1_Update(&c, discard, sizeof(discard) -
1820 } while (stream.avail_in && ret == Z_OK);
1821 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1822 die("unable to write sha1 file");
1823 memmove(buffer, buffer + *bufposn - stream.avail_in,
1825 *bufposn = stream.avail_in;
1829 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1834 return error("Connection closed?");
1835 perror("Reading from connection");
1840 inflateEnd(&stream);
1843 SHA1_Final(real_sha1, &c);
1844 if (ret != Z_STREAM_END) {
1846 return error("File %s corrupted", sha1_to_hex(sha1));
1848 if (hashcmp(sha1, real_sha1)) {
1850 return error("File %s has bad hash", sha1_to_hex(sha1));
1853 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1856 int has_pack_index(const unsigned char *sha1)
1859 if (stat(sha1_pack_index_name(sha1), &st))
1864 int has_pack_file(const unsigned char *sha1)
1867 if (stat(sha1_pack_name(sha1), &st))
1872 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1874 struct pack_entry e;
1875 return find_pack_entry(sha1, &e, ignore_packed);
1878 int has_sha1_file(const unsigned char *sha1)
1881 struct pack_entry e;
1883 if (find_pack_entry(sha1, &e, NULL))
1885 return find_sha1_file(sha1, &st) ? 1 : 0;
1889 * reads from fd as long as possible into a supplied buffer of size bytes.
1890 * If necessary the buffer's size is increased using realloc()
1892 * returns 0 if anything went fine and -1 otherwise
1894 * NOTE: both buf and size may change, but even when -1 is returned
1895 * you still have to free() it yourself.
1897 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1899 char* buf = *return_buf;
1900 unsigned long size = *return_size;
1902 unsigned long off = 0;
1905 iret = xread(fd, buf + off, size - off);
1910 buf = xrealloc(buf, size);
1923 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1925 unsigned long size = 4096;
1926 char *buf = xmalloc(size);
1929 if (read_pipe(fd, &buf, &size)) {
1937 ret = write_sha1_file(buf, size, type, sha1);
1939 ret = hash_sha1_file(buf, size, type, sha1);
1944 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1946 unsigned long size = st->st_size;
1952 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1954 if (buf == MAP_FAILED)
1960 ret = write_sha1_file(buf, size, type, sha1);
1962 ret = hash_sha1_file(buf, size, type, sha1);
1968 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1973 switch (st->st_mode & S_IFMT) {
1975 fd = open(path, O_RDONLY);
1977 return error("open(\"%s\"): %s", path,
1979 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1980 return error("%s: failed to insert into database",
1984 target = xmalloc(st->st_size+1);
1985 if (readlink(path, target, st->st_size+1) != st->st_size) {
1986 char *errstr = strerror(errno);
1988 return error("readlink(\"%s\"): %s", path,
1992 hash_sha1_file(target, st->st_size, blob_type, sha1);
1993 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1994 return error("%s: failed to insert into database",
1999 return error("%s: unsupported file type", path);