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 = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
361 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
363 munmap(map, st.st_size);
366 void prepare_alt_odb(void)
370 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
375 alt_odb_tail = &alt_odb_list;
376 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
378 read_info_alternates(get_object_directory(), 0);
381 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
383 char *name = sha1_file_name(sha1);
384 struct alternate_object_database *alt;
389 for (alt = alt_odb_list; alt; alt = alt->next) {
391 fill_sha1_path(name, sha1);
392 if (!stat(alt->base, st))
398 static unsigned int pack_used_ctr;
399 static unsigned int pack_mmap_calls;
400 static unsigned int peak_pack_open_windows;
401 static unsigned int pack_open_windows;
402 static size_t peak_pack_mapped;
403 static size_t pack_mapped;
404 static size_t page_size;
405 struct packed_git *packed_git;
410 "pack_report: getpagesize() = %10lu\n"
411 "pack_report: core.packedGitWindowSize = %10lu\n"
412 "pack_report: core.packedGitLimit = %10lu\n",
414 packed_git_window_size,
417 "pack_report: pack_used_ctr = %10u\n"
418 "pack_report: pack_mmap_calls = %10u\n"
419 "pack_report: pack_open_windows = %10u / %10u\n"
420 "pack_report: pack_mapped = %10lu / %10lu\n",
423 pack_open_windows, peak_pack_open_windows,
424 pack_mapped, peak_pack_mapped);
427 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
432 unsigned long idx_size;
434 int fd = open(path, O_RDONLY);
438 if (fstat(fd, &st)) {
442 idx_size = st.st_size;
443 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
448 *idx_size_ = idx_size;
450 /* check index map */
451 if (idx_size < 4*256 + 20 + 20)
452 return error("index file too small");
454 for (i = 0; i < 256; i++) {
455 unsigned int n = ntohl(index[i]);
457 return error("non-monotonic index");
463 * - 256 index entries 4 bytes each
464 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
465 * - 20-byte SHA1 of the packfile
466 * - 20-byte SHA1 file checksum
468 if (idx_size != 4*256 + nr * 24 + 20 + 20)
469 return error("wrong index file size");
474 static void scan_windows(struct packed_git *p,
475 struct packed_git **lru_p,
476 struct pack_window **lru_w,
477 struct pack_window **lru_l)
479 struct pack_window *w, *w_l;
481 for (w_l = NULL, w = p->windows; w; w = w->next) {
483 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
493 static int unuse_one_window(struct packed_git *current)
495 struct packed_git *p, *lru_p = NULL;
496 struct pack_window *lru_w = NULL, *lru_l = NULL;
499 scan_windows(current, &lru_p, &lru_w, &lru_l);
500 for (p = packed_git; p; p = p->next)
501 scan_windows(p, &lru_p, &lru_w, &lru_l);
503 munmap(lru_w->base, lru_w->len);
504 pack_mapped -= lru_w->len;
506 lru_l->next = lru_w->next;
508 lru_p->windows = lru_w->next;
509 if (!lru_p->windows && lru_p != current) {
510 close(lru_p->pack_fd);
521 void release_pack_memory(size_t need)
523 size_t cur = pack_mapped;
524 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
528 void unuse_pack(struct pack_window **w_cursor)
530 struct pack_window *w = *w_cursor;
537 static void open_packed_git(struct packed_git *p)
540 struct pack_header hdr;
541 unsigned char sha1[20];
542 unsigned char *idx_sha1;
545 p->pack_fd = open(p->pack_name, O_RDONLY);
546 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
547 die("packfile %s cannot be opened", p->pack_name);
549 /* If we created the struct before we had the pack we lack size. */
551 if (!S_ISREG(st.st_mode))
552 die("packfile %s not a regular file", p->pack_name);
553 p->pack_size = st.st_size;
554 } else if (p->pack_size != st.st_size)
555 die("packfile %s size changed", p->pack_name);
557 /* We leave these file descriptors open with sliding mmap;
558 * there is no point keeping them open across exec(), though.
560 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
562 die("cannot determine file descriptor flags");
563 fd_flag |= FD_CLOEXEC;
564 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
565 die("cannot set FD_CLOEXEC");
567 /* Verify we recognize this pack file format. */
568 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
569 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
570 die("file %s is not a GIT packfile", p->pack_name);
571 if (!pack_version_ok(hdr.hdr_version))
572 die("packfile %s is version %u and not supported"
573 " (try upgrading GIT to a newer version)",
574 p->pack_name, ntohl(hdr.hdr_version));
576 /* Verify the pack matches its index. */
577 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
578 die("packfile %s claims to have %u objects"
579 " while index size indicates %u objects",
580 p->pack_name, ntohl(hdr.hdr_entries),
581 num_packed_objects(p));
582 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
583 die("end of packfile %s is unavailable", p->pack_name);
584 read_or_die(p->pack_fd, sha1, sizeof(sha1));
585 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
586 if (hashcmp(sha1, idx_sha1))
587 die("packfile %s does not match index", p->pack_name);
590 static int in_window(struct pack_window *win, unsigned long offset)
592 /* We must promise at least 20 bytes (one hash) after the
593 * offset is available from this window, otherwise the offset
594 * is not actually in this window and a different window (which
595 * has that one hash excess) must be used. This is to support
596 * the object header and delta base parsing routines below.
598 off_t win_off = win->offset;
599 return win_off <= offset
600 && (offset + 20) <= (win_off + win->len);
603 unsigned char* use_pack(struct packed_git *p,
604 struct pack_window **w_cursor,
605 unsigned long offset,
608 struct pack_window *win = *w_cursor;
610 if (p->pack_fd == -1)
613 /* Since packfiles end in a hash of their content and its
614 * pointless to ask for an offset into the middle of that
615 * hash, and the in_window function above wouldn't match
616 * don't allow an offset too close to the end of the file.
618 if (offset > (p->pack_size - 20))
619 die("offset beyond end of packfile (truncated pack?)");
621 if (!win || !in_window(win, offset)) {
624 for (win = p->windows; win; win = win->next) {
625 if (in_window(win, offset))
630 page_size = getpagesize();
631 win = xcalloc(1, sizeof(*win));
632 win->offset = (offset / page_size) * page_size;
633 win->len = p->pack_size - win->offset;
634 if (win->len > packed_git_window_size)
635 win->len = packed_git_window_size;
636 pack_mapped += win->len;
637 while (packed_git_limit < pack_mapped
638 && unuse_one_window(p))
640 win->base = xmmap(NULL, win->len,
641 PROT_READ, MAP_PRIVATE,
642 p->pack_fd, win->offset);
643 if (win->base == MAP_FAILED)
644 die("packfile %s cannot be mapped: %s",
649 if (pack_mapped > peak_pack_mapped)
650 peak_pack_mapped = pack_mapped;
651 if (pack_open_windows > peak_pack_open_windows)
652 peak_pack_open_windows = pack_open_windows;
653 win->next = p->windows;
657 if (win != *w_cursor) {
658 win->last_used = pack_used_ctr++;
662 offset -= win->offset;
664 *left = win->len - offset;
665 return win->base + offset;
668 struct packed_git *add_packed_git(char *path, int path_len, int local)
671 struct packed_git *p;
672 unsigned long idx_size;
674 unsigned char sha1[20];
676 if (check_packed_git_idx(path, &idx_size, &idx_map))
679 /* do we have a corresponding .pack file? */
680 strcpy(path + path_len - 4, ".pack");
681 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
682 munmap(idx_map, idx_size);
685 /* ok, it looks sane as far as we can check without
686 * actually mapping the pack file.
688 p = xmalloc(sizeof(*p) + path_len + 2);
689 strcpy(p->pack_name, path);
690 p->index_size = idx_size;
691 p->pack_size = st.st_size;
692 p->index_base = idx_map;
696 p->pack_local = local;
697 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
698 hashcpy(p->sha1, sha1);
702 struct packed_git *parse_pack_index(unsigned char *sha1)
704 char *path = sha1_pack_index_name(sha1);
705 return parse_pack_index_file(sha1, path);
708 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
710 struct packed_git *p;
711 unsigned long idx_size;
715 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
718 path = sha1_pack_name(sha1);
720 p = xmalloc(sizeof(*p) + strlen(path) + 2);
721 strcpy(p->pack_name, path);
722 p->index_size = idx_size;
724 p->index_base = idx_map;
728 hashcpy(p->sha1, sha1);
732 void install_packed_git(struct packed_git *pack)
734 pack->next = packed_git;
738 static void prepare_packed_git_one(char *objdir, int local)
745 sprintf(path, "%s/pack", objdir);
750 error("unable to open object pack directory: %s: %s",
751 path, strerror(errno));
755 while ((de = readdir(dir)) != NULL) {
756 int namelen = strlen(de->d_name);
757 struct packed_git *p;
759 if (!has_extension(de->d_name, ".idx"))
762 /* we have .idx. Is it a file we can map? */
763 strcpy(path + len, de->d_name);
764 for (p = packed_git; p; p = p->next) {
765 if (!memcmp(path, p->pack_name, len + namelen - 4))
770 p = add_packed_git(path, len + namelen, local);
773 p->next = packed_git;
779 static int prepare_packed_git_run_once = 0;
780 void prepare_packed_git(void)
782 struct alternate_object_database *alt;
784 if (prepare_packed_git_run_once)
786 prepare_packed_git_one(get_object_directory(), 1);
788 for (alt = alt_odb_list; alt; alt = alt->next) {
790 prepare_packed_git_one(alt->base, 0);
793 prepare_packed_git_run_once = 1;
796 void reprepare_packed_git(void)
798 prepare_packed_git_run_once = 0;
799 prepare_packed_git();
802 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
804 unsigned char real_sha1[20];
805 hash_sha1_file(map, size, type, real_sha1);
806 return hashcmp(sha1, real_sha1) ? -1 : 0;
809 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
814 char *filename = find_sha1_file(sha1, &st);
820 fd = open(filename, O_RDONLY | sha1_file_open_flag);
822 /* See if it works without O_NOATIME */
823 switch (sha1_file_open_flag) {
825 fd = open(filename, O_RDONLY);
833 /* If it failed once, it will probably fail again.
834 * Stop using O_NOATIME
836 sha1_file_open_flag = 0;
838 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
844 int legacy_loose_object(unsigned char *map)
849 * Is it a zlib-compressed buffer? If so, the first byte
850 * must be 0x78 (15-bit window size, deflated), and the
851 * first 16-bit word is evenly divisible by 31
853 word = (map[0] << 8) + map[1];
854 if (map[0] == 0x78 && !(word % 31))
860 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
865 unsigned long used = 0;
868 *type = (c >> 4) & 7;
874 if (sizeof(long) * 8 <= shift)
877 size += (c & 0x7f) << shift;
884 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
886 unsigned long size, used;
887 static const char valid_loose_object_type[8] = {
889 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
890 0, /* "delta" and others are invalid in a loose object */
892 enum object_type type;
894 /* Get the data stream */
895 memset(stream, 0, sizeof(*stream));
896 stream->next_in = map;
897 stream->avail_in = mapsize;
898 stream->next_out = buffer;
899 stream->avail_out = bufsiz;
901 if (legacy_loose_object(map)) {
903 return inflate(stream, 0);
906 used = unpack_object_header_gently(map, mapsize, &type, &size);
907 if (!used || !valid_loose_object_type[type])
912 /* Set up the stream for the rest.. */
913 stream->next_in = map;
914 stream->avail_in = mapsize;
917 /* And generate the fake traditional header */
918 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
919 type_names[type], size);
923 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
925 int bytes = strlen(buffer) + 1;
926 unsigned char *buf = xmalloc(1+size);
929 n = stream->total_out - bytes;
932 memcpy(buf, (char *) buffer + bytes, n);
935 stream->next_out = buf + bytes;
936 stream->avail_out = size - bytes;
937 while (inflate(stream, Z_FINISH) == Z_OK)
946 * We used to just use "sscanf()", but that's actually way
947 * too permissive for what we want to check. So do an anal
948 * object header parse by hand.
950 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
956 * The type can be at most ten bytes (including the
957 * terminating '\0' that we add), and is followed by
972 * The length must follow immediately, and be in canonical
973 * decimal format (ie "010" is not valid).
980 unsigned long c = *hdr - '0';
984 size = size * 10 + c;
990 * The length must be followed by a zero byte
992 return *hdr ? -1 : 0;
995 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1001 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1002 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1005 return unpack_sha1_rest(&stream, hdr, *size);
1008 static unsigned long get_delta_base(struct packed_git *p,
1009 struct pack_window **w_curs,
1010 unsigned long offset,
1011 enum object_type kind,
1012 unsigned long delta_obj_offset,
1013 unsigned long *base_obj_offset)
1015 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1016 unsigned long base_offset;
1018 /* use_pack() assured us we have [base_info, base_info + 20)
1019 * as a range that we can look at without walking off the
1020 * end of the mapped window. Its actually the hash size
1021 * that is assured. An OFS_DELTA longer than the hash size
1022 * is stupid, as then a REF_DELTA would be smaller to store.
1024 if (kind == OBJ_OFS_DELTA) {
1026 unsigned char c = base_info[used++];
1027 base_offset = c & 127;
1030 if (!base_offset || base_offset & ~(~0UL >> 7))
1031 die("offset value overflow for delta base object");
1032 c = base_info[used++];
1033 base_offset = (base_offset << 7) + (c & 127);
1035 base_offset = delta_obj_offset - base_offset;
1036 if (base_offset >= delta_obj_offset)
1037 die("delta base offset out of bound");
1039 } else if (kind == OBJ_REF_DELTA) {
1040 /* The base entry _must_ be in the same pack */
1041 base_offset = find_pack_entry_one(base_info, p);
1043 die("failed to find delta-pack base object %s",
1044 sha1_to_hex(base_info));
1047 die("I am totally screwed");
1048 *base_obj_offset = base_offset;
1052 /* forward declaration for a mutually recursive function */
1053 static int packed_object_info(struct packed_git *p, unsigned long offset,
1054 char *type, unsigned long *sizep);
1056 static int packed_delta_info(struct packed_git *p,
1057 struct pack_window **w_curs,
1058 unsigned long offset,
1059 enum object_type kind,
1060 unsigned long obj_offset,
1062 unsigned long *sizep)
1064 unsigned long base_offset;
1066 offset = get_delta_base(p, w_curs, offset, kind,
1067 obj_offset, &base_offset);
1069 /* We choose to only get the type of the base object and
1070 * ignore potentially corrupt pack file that expects the delta
1071 * based on a base with a wrong size. This saves tons of
1074 if (packed_object_info(p, base_offset, type, NULL))
1075 die("cannot get info for delta-pack base");
1078 const unsigned char *data;
1079 unsigned char delta_head[20], *in;
1080 unsigned long result_size;
1084 memset(&stream, 0, sizeof(stream));
1085 stream.next_out = delta_head;
1086 stream.avail_out = sizeof(delta_head);
1088 inflateInit(&stream);
1090 in = use_pack(p, w_curs, offset, &stream.avail_in);
1091 stream.next_in = in;
1092 st = inflate(&stream, Z_FINISH);
1093 offset += stream.next_in - in;
1094 } while ((st == Z_OK || st == Z_BUF_ERROR)
1095 && stream.total_out < sizeof(delta_head));
1096 inflateEnd(&stream);
1097 if ((st != Z_STREAM_END) &&
1098 stream.total_out != sizeof(delta_head))
1099 die("delta data unpack-initial failed");
1101 /* Examine the initial part of the delta to figure out
1106 /* ignore base size */
1107 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1109 /* Read the result size */
1110 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1111 *sizep = result_size;
1116 static unsigned long unpack_object_header(struct packed_git *p,
1117 struct pack_window **w_curs,
1118 unsigned long offset,
1119 enum object_type *type,
1120 unsigned long *sizep)
1122 unsigned char *base;
1126 /* use_pack() assures us we have [base, base + 20) available
1127 * as a range that we can look at at. (Its actually the hash
1128 * size that is assurred.) With our object header encoding
1129 * the maximum deflated object size is 2^137, which is just
1130 * insane, so we know won't exceed what we have been given.
1132 base = use_pack(p, w_curs, offset, &left);
1133 used = unpack_object_header_gently(base, left, type, sizep);
1135 die("object offset outside of pack file");
1137 return offset + used;
1140 void packed_object_info_detail(struct packed_git *p,
1141 unsigned long offset,
1143 unsigned long *size,
1144 unsigned long *store_size,
1145 unsigned int *delta_chain_length,
1146 unsigned char *base_sha1)
1148 struct pack_window *w_curs = NULL;
1149 unsigned long obj_offset, val;
1150 unsigned char *next_sha1;
1151 enum object_type kind;
1153 *delta_chain_length = 0;
1154 obj_offset = offset;
1155 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1160 die("pack %s contains unknown object type %d",
1161 p->pack_name, kind);
1166 strcpy(type, type_names[kind]);
1167 *store_size = 0; /* notyet */
1168 unuse_pack(&w_curs);
1171 get_delta_base(p, &w_curs, offset, kind,
1172 obj_offset, &offset);
1173 if (*delta_chain_length == 0) {
1174 /* TODO: find base_sha1 as pointed by offset */
1178 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1179 if (*delta_chain_length == 0)
1180 hashcpy(base_sha1, next_sha1);
1181 offset = find_pack_entry_one(next_sha1, p);
1184 obj_offset = offset;
1185 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1186 (*delta_chain_length)++;
1190 static int packed_object_info(struct packed_git *p, unsigned long offset,
1191 char *type, unsigned long *sizep)
1193 struct pack_window *w_curs = NULL;
1194 unsigned long size, obj_offset = offset;
1195 enum object_type kind;
1198 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1203 r = packed_delta_info(p, &w_curs, offset, kind,
1204 obj_offset, type, sizep);
1205 unuse_pack(&w_curs);
1211 strcpy(type, type_names[kind]);
1212 unuse_pack(&w_curs);
1215 die("pack %s contains unknown object type %d",
1216 p->pack_name, kind);
1223 static void *unpack_compressed_entry(struct packed_git *p,
1224 struct pack_window **w_curs,
1225 unsigned long offset,
1230 unsigned char *buffer, *in;
1232 buffer = xmalloc(size + 1);
1234 memset(&stream, 0, sizeof(stream));
1235 stream.next_out = buffer;
1236 stream.avail_out = size;
1238 inflateInit(&stream);
1240 in = use_pack(p, w_curs, offset, &stream.avail_in);
1241 stream.next_in = in;
1242 st = inflate(&stream, Z_FINISH);
1243 offset += stream.next_in - in;
1244 } while (st == Z_OK || st == Z_BUF_ERROR);
1245 inflateEnd(&stream);
1246 if ((st != Z_STREAM_END) || stream.total_out != size) {
1254 static void *unpack_delta_entry(struct packed_git *p,
1255 struct pack_window **w_curs,
1256 unsigned long offset,
1257 unsigned long delta_size,
1258 enum object_type kind,
1259 unsigned long obj_offset,
1261 unsigned long *sizep)
1263 void *delta_data, *result, *base;
1264 unsigned long result_size, base_size, base_offset;
1266 offset = get_delta_base(p, w_curs, offset, kind,
1267 obj_offset, &base_offset);
1268 base = unpack_entry(p, base_offset, type, &base_size);
1270 die("failed to read delta base object at %lu from %s",
1271 base_offset, p->pack_name);
1273 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1274 result = patch_delta(base, base_size,
1275 delta_data, delta_size,
1278 die("failed to apply delta");
1281 *sizep = result_size;
1285 void *unpack_entry(struct packed_git *p, unsigned long offset,
1286 char *type, unsigned long *sizep)
1288 struct pack_window *w_curs = NULL;
1289 unsigned long size, obj_offset = offset;
1290 enum object_type kind;
1293 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1297 retval = unpack_delta_entry(p, &w_curs, offset, size,
1298 kind, obj_offset, type, sizep);
1304 strcpy(type, type_names[kind]);
1306 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1309 die("unknown object type %i in %s", kind, p->pack_name);
1311 unuse_pack(&w_curs);
1315 int num_packed_objects(const struct packed_git *p)
1317 /* See check_packed_git_idx() */
1318 return (p->index_size - 20 - 20 - 4*256) / 24;
1321 int nth_packed_object_sha1(const struct packed_git *p, int n,
1322 unsigned char* sha1)
1324 void *index = p->index_base + 256;
1325 if (n < 0 || num_packed_objects(p) <= n)
1327 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1331 unsigned long find_pack_entry_one(const unsigned char *sha1,
1332 struct packed_git *p)
1334 unsigned int *level1_ofs = p->index_base;
1335 int hi = ntohl(level1_ofs[*sha1]);
1336 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1337 void *index = p->index_base + 256;
1340 int mi = (lo + hi) / 2;
1341 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1343 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1352 static int matches_pack_name(struct packed_git *p, const char *ig)
1354 const char *last_c, *c;
1356 if (!strcmp(p->pack_name, ig))
1359 for (c = p->pack_name, last_c = c; *c;)
1364 if (!strcmp(last_c, ig))
1370 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1372 struct packed_git *p;
1373 unsigned long offset;
1375 prepare_packed_git();
1377 for (p = packed_git; p; p = p->next) {
1378 if (ignore_packed) {
1380 for (ig = ignore_packed; *ig; ig++)
1381 if (!matches_pack_name(p, *ig))
1386 offset = find_pack_entry_one(sha1, p);
1390 hashcpy(e->sha1, sha1);
1397 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1398 struct packed_git *packs)
1400 struct packed_git *p;
1402 for (p = packs; p; p = p->next) {
1403 if (find_pack_entry_one(sha1, p))
1410 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1413 unsigned long mapsize, size;
1418 map = map_sha1_file(sha1, &mapsize);
1420 return error("unable to find %s", sha1_to_hex(sha1));
1421 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1422 status = error("unable to unpack %s header",
1424 if (parse_sha1_header(hdr, type, &size) < 0)
1425 status = error("unable to parse %s header", sha1_to_hex(sha1));
1431 inflateEnd(&stream);
1432 munmap(map, mapsize);
1436 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1438 struct pack_entry e;
1440 if (!find_pack_entry(sha1, &e, NULL)) {
1441 reprepare_packed_git();
1442 if (!find_pack_entry(sha1, &e, NULL))
1443 return sha1_loose_object_info(sha1, type, sizep);
1445 return packed_object_info(e.p, e.offset, type, sizep);
1448 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1450 struct pack_entry e;
1452 if (!find_pack_entry(sha1, &e, NULL)) {
1453 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1456 return unpack_entry(e.p, e.offset, type, size);
1459 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1461 unsigned long mapsize;
1463 struct pack_entry e;
1465 if (find_pack_entry(sha1, &e, NULL))
1466 return read_packed_sha1(sha1, type, size);
1467 map = map_sha1_file(sha1, &mapsize);
1469 buf = unpack_sha1_file(map, mapsize, type, size);
1470 munmap(map, mapsize);
1473 reprepare_packed_git();
1474 if (find_pack_entry(sha1, &e, NULL))
1475 return read_packed_sha1(sha1, type, size);
1479 void *read_object_with_reference(const unsigned char *sha1,
1480 const char *required_type,
1481 unsigned long *size,
1482 unsigned char *actual_sha1_return)
1486 unsigned long isize;
1487 unsigned char actual_sha1[20];
1489 hashcpy(actual_sha1, sha1);
1491 int ref_length = -1;
1492 const char *ref_type = NULL;
1494 buffer = read_sha1_file(actual_sha1, type, &isize);
1497 if (!strcmp(type, required_type)) {
1499 if (actual_sha1_return)
1500 hashcpy(actual_sha1_return, actual_sha1);
1503 /* Handle references */
1504 else if (!strcmp(type, commit_type))
1506 else if (!strcmp(type, tag_type))
1507 ref_type = "object ";
1512 ref_length = strlen(ref_type);
1514 if (memcmp(buffer, ref_type, ref_length) ||
1515 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1520 /* Now we have the ID of the referred-to object in
1521 * actual_sha1. Check again. */
1525 static void write_sha1_file_prepare(void *buf, unsigned long len,
1526 const char *type, unsigned char *sha1,
1527 unsigned char *hdr, int *hdrlen)
1531 /* Generate the header */
1532 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1536 SHA1_Update(&c, hdr, *hdrlen);
1537 SHA1_Update(&c, buf, len);
1538 SHA1_Final(sha1, &c);
1542 * Link the tempfile to the final place, possibly creating the
1543 * last directory level as you do so.
1545 * Returns the errno on failure, 0 on success.
1547 static int link_temp_to_file(const char *tmpfile, const char *filename)
1552 if (!link(tmpfile, filename))
1556 * Try to mkdir the last path component if that failed.
1558 * Re-try the "link()" regardless of whether the mkdir
1559 * succeeds, since a race might mean that somebody
1563 dir = strrchr(filename, '/');
1566 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1571 if (!link(tmpfile, filename))
1579 * Move the just written object into its final resting place
1581 int move_temp_to_file(const char *tmpfile, const char *filename)
1583 int ret = link_temp_to_file(tmpfile, filename);
1586 * Coda hack - coda doesn't like cross-directory links,
1587 * so we fall back to a rename, which will mean that it
1588 * won't be able to check collisions, but that's not a
1591 * The same holds for FAT formatted media.
1593 * When this succeeds, we just return 0. We have nothing
1596 if (ret && ret != EEXIST) {
1597 if (!rename(tmpfile, filename))
1603 if (ret != EEXIST) {
1604 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1606 /* FIXME!!! Collision check here ? */
1612 static int write_buffer(int fd, const void *buf, size_t len)
1617 size = write(fd, buf, len);
1619 return error("file write: disk full");
1621 if (errno == EINTR || errno == EAGAIN)
1623 return error("file write error (%s)", strerror(errno));
1626 buf = (char *) buf + size;
1631 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1636 c = (type << 4) | (len & 15);
1649 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1653 if (use_legacy_headers) {
1654 while (deflate(stream, 0) == Z_OK)
1658 if (!strcmp(type, blob_type))
1659 obj_type = OBJ_BLOB;
1660 else if (!strcmp(type, tree_type))
1661 obj_type = OBJ_TREE;
1662 else if (!strcmp(type, commit_type))
1663 obj_type = OBJ_COMMIT;
1664 else if (!strcmp(type, tag_type))
1667 die("trying to generate bogus object of type '%s'", type);
1668 hdr = write_binary_header(stream->next_out, obj_type, len);
1669 stream->total_out = hdr;
1670 stream->next_out += hdr;
1671 stream->avail_out -= hdr;
1674 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1675 unsigned char *sha1)
1677 unsigned char hdr[50];
1679 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1683 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1686 unsigned char *compressed;
1688 unsigned char sha1[20];
1690 static char tmpfile[PATH_MAX];
1691 unsigned char hdr[50];
1694 /* Normally if we have it in the pack then we do not bother writing
1695 * it out into .git/objects/??/?{38} file.
1697 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1698 filename = sha1_file_name(sha1);
1700 hashcpy(returnsha1, sha1);
1701 if (has_sha1_file(sha1))
1703 fd = open(filename, O_RDONLY);
1706 * FIXME!!! We might do collision checking here, but we'd
1707 * need to uncompress the old file and check it. Later.
1713 if (errno != ENOENT) {
1714 return error("sha1 file %s: %s\n", filename, strerror(errno));
1717 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1719 fd = mkstemp(tmpfile);
1722 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1724 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1728 memset(&stream, 0, sizeof(stream));
1729 deflateInit(&stream, zlib_compression_level);
1730 size = 8 + deflateBound(&stream, len+hdrlen);
1731 compressed = xmalloc(size);
1734 stream.next_out = compressed;
1735 stream.avail_out = size;
1737 /* First header.. */
1738 stream.next_in = hdr;
1739 stream.avail_in = hdrlen;
1740 setup_object_header(&stream, type, len);
1742 /* Then the data itself.. */
1743 stream.next_in = buf;
1744 stream.avail_in = len;
1745 while (deflate(&stream, Z_FINISH) == Z_OK)
1747 deflateEnd(&stream);
1748 size = stream.total_out;
1750 if (write_buffer(fd, compressed, size) < 0)
1751 die("unable to write sha1 file");
1756 return move_temp_to_file(tmpfile, filename);
1760 * We need to unpack and recompress the object for writing
1761 * it out to a different file.
1763 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1767 unsigned char *unpacked;
1774 /* need to unpack and recompress it by itself */
1775 unpacked = read_packed_sha1(sha1, type, &len);
1777 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1780 memset(&stream, 0, sizeof(stream));
1781 deflateInit(&stream, zlib_compression_level);
1782 size = deflateBound(&stream, len + hdrlen);
1783 buf = xmalloc(size);
1786 stream.next_out = buf;
1787 stream.avail_out = size;
1789 /* First header.. */
1790 stream.next_in = (void *)hdr;
1791 stream.avail_in = hdrlen;
1792 while (deflate(&stream, 0) == Z_OK)
1795 /* Then the data itself.. */
1796 stream.next_in = unpacked;
1797 stream.avail_in = len;
1798 while (deflate(&stream, Z_FINISH) == Z_OK)
1800 deflateEnd(&stream);
1803 *objsize = stream.total_out;
1807 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1810 unsigned long objsize;
1811 void *buf = map_sha1_file(sha1, &objsize);
1814 retval = write_buffer(fd, buf, objsize);
1815 munmap(buf, objsize);
1819 buf = repack_object(sha1, &objsize);
1820 retval = write_buffer(fd, buf, objsize);
1825 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1826 size_t bufsize, size_t *bufposn)
1828 char tmpfile[PATH_MAX];
1831 unsigned char real_sha1[20];
1832 unsigned char discard[4096];
1836 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1838 local = mkstemp(tmpfile);
1841 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1843 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1846 memset(&stream, 0, sizeof(stream));
1848 inflateInit(&stream);
1855 stream.avail_in = *bufposn;
1856 stream.next_in = (unsigned char *) buffer;
1858 stream.next_out = discard;
1859 stream.avail_out = sizeof(discard);
1860 ret = inflate(&stream, Z_SYNC_FLUSH);
1861 SHA1_Update(&c, discard, sizeof(discard) -
1863 } while (stream.avail_in && ret == Z_OK);
1864 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1865 die("unable to write sha1 file");
1866 memmove(buffer, buffer + *bufposn - stream.avail_in,
1868 *bufposn = stream.avail_in;
1872 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1877 return error("Connection closed?");
1878 perror("Reading from connection");
1883 inflateEnd(&stream);
1886 SHA1_Final(real_sha1, &c);
1887 if (ret != Z_STREAM_END) {
1889 return error("File %s corrupted", sha1_to_hex(sha1));
1891 if (hashcmp(sha1, real_sha1)) {
1893 return error("File %s has bad hash", sha1_to_hex(sha1));
1896 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1899 int has_pack_index(const unsigned char *sha1)
1902 if (stat(sha1_pack_index_name(sha1), &st))
1907 int has_pack_file(const unsigned char *sha1)
1910 if (stat(sha1_pack_name(sha1), &st))
1915 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1917 struct pack_entry e;
1918 return find_pack_entry(sha1, &e, ignore_packed);
1921 int has_sha1_file(const unsigned char *sha1)
1924 struct pack_entry e;
1926 if (find_pack_entry(sha1, &e, NULL))
1928 return find_sha1_file(sha1, &st) ? 1 : 0;
1932 * reads from fd as long as possible into a supplied buffer of size bytes.
1933 * If necessary the buffer's size is increased using realloc()
1935 * returns 0 if anything went fine and -1 otherwise
1937 * NOTE: both buf and size may change, but even when -1 is returned
1938 * you still have to free() it yourself.
1940 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1942 char* buf = *return_buf;
1943 unsigned long size = *return_size;
1945 unsigned long off = 0;
1948 iret = xread(fd, buf + off, size - off);
1953 buf = xrealloc(buf, size);
1966 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1968 unsigned long size = 4096;
1969 char *buf = xmalloc(size);
1972 if (read_pipe(fd, &buf, &size)) {
1980 ret = write_sha1_file(buf, size, type, sha1);
1982 ret = hash_sha1_file(buf, size, type, sha1);
1987 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1989 unsigned long size = st->st_size;
1995 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2001 ret = write_sha1_file(buf, size, type, sha1);
2003 ret = hash_sha1_file(buf, size, type, sha1);
2009 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2014 switch (st->st_mode & S_IFMT) {
2016 fd = open(path, O_RDONLY);
2018 return error("open(\"%s\"): %s", path,
2020 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2021 return error("%s: failed to insert into database",
2025 target = xmalloc(st->st_size+1);
2026 if (readlink(path, target, st->st_size+1) != st->st_size) {
2027 char *errstr = strerror(errno);
2029 return error("readlink(\"%s\"): %s", path,
2033 hash_sha1_file(target, st->st_size, blob_type, sha1);
2034 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2035 return error("%s: failed to insert into database",
2040 return error("%s: unsupported file type", path);