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 unsigned int pack_mmap_calls;
402 static unsigned int peak_pack_open_windows;
403 static unsigned int pack_open_windows;
404 static size_t peak_pack_mapped;
405 static size_t pack_mapped;
406 static size_t page_size;
407 struct packed_git *packed_git;
412 "pack_report: getpagesize() = %10lu\n"
413 "pack_report: core.packedGitWindowSize = %10lu\n"
414 "pack_report: core.packedGitLimit = %10lu\n",
416 packed_git_window_size,
419 "pack_report: pack_used_ctr = %10u\n"
420 "pack_report: pack_mmap_calls = %10u\n"
421 "pack_report: pack_open_windows = %10u / %10u\n"
422 "pack_report: pack_mapped = %10lu / %10lu\n",
425 pack_open_windows, peak_pack_open_windows,
426 pack_mapped, peak_pack_mapped);
429 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
434 unsigned long idx_size;
436 int fd = open(path, O_RDONLY);
440 if (fstat(fd, &st)) {
444 idx_size = st.st_size;
445 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
447 if (idx_map == MAP_FAILED)
452 *idx_size_ = idx_size;
454 /* check index map */
455 if (idx_size < 4*256 + 20 + 20)
456 return error("index file too small");
458 for (i = 0; i < 256; i++) {
459 unsigned int n = ntohl(index[i]);
461 return error("non-monotonic index");
467 * - 256 index entries 4 bytes each
468 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
469 * - 20-byte SHA1 of the packfile
470 * - 20-byte SHA1 file checksum
472 if (idx_size != 4*256 + nr * 24 + 20 + 20)
473 return error("wrong index file size");
478 static void scan_windows(struct packed_git *p,
479 struct packed_git **lru_p,
480 struct pack_window **lru_w,
481 struct pack_window **lru_l)
483 struct pack_window *w, *w_l;
485 for (w_l = NULL, w = p->windows; w; w = w->next) {
487 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
497 static int unuse_one_window(struct packed_git *current)
499 struct packed_git *p, *lru_p = NULL;
500 struct pack_window *lru_w = NULL, *lru_l = NULL;
503 scan_windows(current, &lru_p, &lru_w, &lru_l);
504 for (p = packed_git; p; p = p->next)
505 scan_windows(p, &lru_p, &lru_w, &lru_l);
507 munmap(lru_w->base, lru_w->len);
508 pack_mapped -= lru_w->len;
510 lru_l->next = lru_w->next;
512 lru_p->windows = lru_w->next;
513 if (!lru_p->windows && lru_p != current) {
514 close(lru_p->pack_fd);
525 void release_pack_memory(size_t need)
527 size_t cur = pack_mapped;
528 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
532 void unuse_pack(struct pack_window **w_cursor)
534 struct pack_window *w = *w_cursor;
541 static void open_packed_git(struct packed_git *p)
544 struct pack_header hdr;
545 unsigned char sha1[20];
546 unsigned char *idx_sha1;
548 p->pack_fd = open(p->pack_name, O_RDONLY);
549 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
550 die("packfile %s cannot be opened", p->pack_name);
552 /* If we created the struct before we had the pack we lack size. */
554 if (!S_ISREG(st.st_mode))
555 die("packfile %s not a regular file", p->pack_name);
556 p->pack_size = st.st_size;
557 } else if (p->pack_size != st.st_size)
558 die("packfile %s size changed", p->pack_name);
560 /* Verify we recognize this pack file format. */
561 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
562 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
563 die("file %s is not a GIT packfile", p->pack_name);
564 if (!pack_version_ok(hdr.hdr_version))
565 die("packfile %s is version %u and not supported"
566 " (try upgrading GIT to a newer version)",
567 p->pack_name, ntohl(hdr.hdr_version));
569 /* Verify the pack matches its index. */
570 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
571 die("packfile %s claims to have %u objects"
572 " while index size indicates %u objects",
573 p->pack_name, ntohl(hdr.hdr_entries),
574 num_packed_objects(p));
575 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
576 die("end of packfile %s is unavailable", p->pack_name);
577 read_or_die(p->pack_fd, sha1, sizeof(sha1));
578 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
579 if (hashcmp(sha1, idx_sha1))
580 die("packfile %s does not match index", p->pack_name);
583 static int in_window(struct pack_window *win, unsigned long offset)
585 /* We must promise at least 20 bytes (one hash) after the
586 * offset is available from this window, otherwise the offset
587 * is not actually in this window and a different window (which
588 * has that one hash excess) must be used. This is to support
589 * the object header and delta base parsing routines below.
591 off_t win_off = win->offset;
592 return win_off <= offset
593 && (offset + 20) <= (win_off + win->len);
596 unsigned char* use_pack(struct packed_git *p,
597 struct pack_window **w_cursor,
598 unsigned long offset,
601 struct pack_window *win = *w_cursor;
603 if (p->pack_fd == -1)
606 /* Since packfiles end in a hash of their content and its
607 * pointless to ask for an offset into the middle of that
608 * hash, and the in_window function above wouldn't match
609 * don't allow an offset too close to the end of the file.
611 if (offset > (p->pack_size - 20))
612 die("offset beyond end of packfile (truncated pack?)");
614 if (!win || !in_window(win, offset)) {
617 for (win = p->windows; win; win = win->next) {
618 if (in_window(win, offset))
623 page_size = getpagesize();
624 win = xcalloc(1, sizeof(*win));
625 win->offset = (offset / page_size) * page_size;
626 win->len = p->pack_size - win->offset;
627 if (win->len > packed_git_window_size)
628 win->len = packed_git_window_size;
629 pack_mapped += win->len;
630 while (packed_git_limit < pack_mapped
631 && unuse_one_window(p))
633 win->base = mmap(NULL, win->len,
634 PROT_READ, MAP_PRIVATE,
635 p->pack_fd, win->offset);
636 if (win->base == MAP_FAILED)
637 die("packfile %s cannot be mapped: %s",
642 if (pack_mapped > peak_pack_mapped)
643 peak_pack_mapped = pack_mapped;
644 if (pack_open_windows > peak_pack_open_windows)
645 peak_pack_open_windows = pack_open_windows;
646 win->next = p->windows;
650 if (win != *w_cursor) {
651 win->last_used = pack_used_ctr++;
655 offset -= win->offset;
657 *left = win->len - offset;
658 return win->base + offset;
661 struct packed_git *add_packed_git(char *path, int path_len, int local)
664 struct packed_git *p;
665 unsigned long idx_size;
667 unsigned char sha1[20];
669 if (check_packed_git_idx(path, &idx_size, &idx_map))
672 /* do we have a corresponding .pack file? */
673 strcpy(path + path_len - 4, ".pack");
674 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
675 munmap(idx_map, idx_size);
678 /* ok, it looks sane as far as we can check without
679 * actually mapping the pack file.
681 p = xmalloc(sizeof(*p) + path_len + 2);
682 strcpy(p->pack_name, path);
683 p->index_size = idx_size;
684 p->pack_size = st.st_size;
685 p->index_base = idx_map;
689 p->pack_local = local;
690 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
691 hashcpy(p->sha1, sha1);
695 struct packed_git *parse_pack_index(unsigned char *sha1)
697 char *path = sha1_pack_index_name(sha1);
698 return parse_pack_index_file(sha1, path);
701 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
703 struct packed_git *p;
704 unsigned long idx_size;
708 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
711 path = sha1_pack_name(sha1);
713 p = xmalloc(sizeof(*p) + strlen(path) + 2);
714 strcpy(p->pack_name, path);
715 p->index_size = idx_size;
717 p->index_base = idx_map;
721 hashcpy(p->sha1, sha1);
725 void install_packed_git(struct packed_git *pack)
727 pack->next = packed_git;
731 static void prepare_packed_git_one(char *objdir, int local)
738 sprintf(path, "%s/pack", objdir);
743 error("unable to open object pack directory: %s: %s",
744 path, strerror(errno));
748 while ((de = readdir(dir)) != NULL) {
749 int namelen = strlen(de->d_name);
750 struct packed_git *p;
752 if (!has_extension(de->d_name, ".idx"))
755 /* we have .idx. Is it a file we can map? */
756 strcpy(path + len, de->d_name);
757 for (p = packed_git; p; p = p->next) {
758 if (!memcmp(path, p->pack_name, len + namelen - 4))
763 p = add_packed_git(path, len + namelen, local);
766 p->next = packed_git;
772 static int prepare_packed_git_run_once = 0;
773 void prepare_packed_git(void)
775 struct alternate_object_database *alt;
777 if (prepare_packed_git_run_once)
779 prepare_packed_git_one(get_object_directory(), 1);
781 for (alt = alt_odb_list; alt; alt = alt->next) {
783 prepare_packed_git_one(alt->base, 0);
786 prepare_packed_git_run_once = 1;
789 void reprepare_packed_git(void)
791 prepare_packed_git_run_once = 0;
792 prepare_packed_git();
795 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
797 unsigned char real_sha1[20];
798 hash_sha1_file(map, size, type, real_sha1);
799 return hashcmp(sha1, real_sha1) ? -1 : 0;
802 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
807 char *filename = find_sha1_file(sha1, &st);
813 fd = open(filename, O_RDONLY | sha1_file_open_flag);
815 /* See if it works without O_NOATIME */
816 switch (sha1_file_open_flag) {
818 fd = open(filename, O_RDONLY);
826 /* If it failed once, it will probably fail again.
827 * Stop using O_NOATIME
829 sha1_file_open_flag = 0;
831 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
833 if (map == MAP_FAILED)
839 int legacy_loose_object(unsigned char *map)
844 * Is it a zlib-compressed buffer? If so, the first byte
845 * must be 0x78 (15-bit window size, deflated), and the
846 * first 16-bit word is evenly divisible by 31
848 word = (map[0] << 8) + map[1];
849 if (map[0] == 0x78 && !(word % 31))
855 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
860 unsigned long used = 0;
863 *type = (c >> 4) & 7;
869 if (sizeof(long) * 8 <= shift)
872 size += (c & 0x7f) << shift;
879 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
881 unsigned long size, used;
882 static const char valid_loose_object_type[8] = {
884 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
885 0, /* "delta" and others are invalid in a loose object */
887 enum object_type type;
889 /* Get the data stream */
890 memset(stream, 0, sizeof(*stream));
891 stream->next_in = map;
892 stream->avail_in = mapsize;
893 stream->next_out = buffer;
894 stream->avail_out = bufsiz;
896 if (legacy_loose_object(map)) {
898 return inflate(stream, 0);
901 used = unpack_object_header_gently(map, mapsize, &type, &size);
902 if (!used || !valid_loose_object_type[type])
907 /* Set up the stream for the rest.. */
908 stream->next_in = map;
909 stream->avail_in = mapsize;
912 /* And generate the fake traditional header */
913 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
914 type_names[type], size);
918 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
920 int bytes = strlen(buffer) + 1;
921 unsigned char *buf = xmalloc(1+size);
924 n = stream->total_out - bytes;
927 memcpy(buf, (char *) buffer + bytes, n);
930 stream->next_out = buf + bytes;
931 stream->avail_out = size - bytes;
932 while (inflate(stream, Z_FINISH) == Z_OK)
941 * We used to just use "sscanf()", but that's actually way
942 * too permissive for what we want to check. So do an anal
943 * object header parse by hand.
945 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
951 * The type can be at most ten bytes (including the
952 * terminating '\0' that we add), and is followed by
967 * The length must follow immediately, and be in canonical
968 * decimal format (ie "010" is not valid).
975 unsigned long c = *hdr - '0';
979 size = size * 10 + c;
985 * The length must be followed by a zero byte
987 return *hdr ? -1 : 0;
990 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
996 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
997 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1000 return unpack_sha1_rest(&stream, hdr, *size);
1003 static unsigned long get_delta_base(struct packed_git *p,
1004 struct pack_window **w_curs,
1005 unsigned long offset,
1006 enum object_type kind,
1007 unsigned long delta_obj_offset,
1008 unsigned long *base_obj_offset)
1010 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1011 unsigned long base_offset;
1013 /* use_pack() assured us we have [base_info, base_info + 20)
1014 * as a range that we can look at without walking off the
1015 * end of the mapped window. Its actually the hash size
1016 * that is assured. An OFS_DELTA longer than the hash size
1017 * is stupid, as then a REF_DELTA would be smaller to store.
1019 if (kind == OBJ_OFS_DELTA) {
1021 unsigned char c = base_info[used++];
1022 base_offset = c & 127;
1025 if (!base_offset || base_offset & ~(~0UL >> 7))
1026 die("offset value overflow for delta base object");
1027 c = base_info[used++];
1028 base_offset = (base_offset << 7) + (c & 127);
1030 base_offset = delta_obj_offset - base_offset;
1031 if (base_offset >= delta_obj_offset)
1032 die("delta base offset out of bound");
1034 } else if (kind == OBJ_REF_DELTA) {
1035 /* The base entry _must_ be in the same pack */
1036 base_offset = find_pack_entry_one(base_info, p);
1038 die("failed to find delta-pack base object %s",
1039 sha1_to_hex(base_info));
1042 die("I am totally screwed");
1043 *base_obj_offset = base_offset;
1047 /* forward declaration for a mutually recursive function */
1048 static int packed_object_info(struct packed_git *p, unsigned long offset,
1049 char *type, unsigned long *sizep);
1051 static int packed_delta_info(struct packed_git *p,
1052 struct pack_window **w_curs,
1053 unsigned long offset,
1054 enum object_type kind,
1055 unsigned long obj_offset,
1057 unsigned long *sizep)
1059 unsigned long base_offset;
1061 offset = get_delta_base(p, w_curs, offset, kind,
1062 obj_offset, &base_offset);
1064 /* We choose to only get the type of the base object and
1065 * ignore potentially corrupt pack file that expects the delta
1066 * based on a base with a wrong size. This saves tons of
1069 if (packed_object_info(p, base_offset, type, NULL))
1070 die("cannot get info for delta-pack base");
1073 const unsigned char *data;
1074 unsigned char delta_head[20], *in;
1075 unsigned long result_size;
1079 memset(&stream, 0, sizeof(stream));
1080 stream.next_out = delta_head;
1081 stream.avail_out = sizeof(delta_head);
1083 inflateInit(&stream);
1085 in = use_pack(p, w_curs, offset, &stream.avail_in);
1086 stream.next_in = in;
1087 st = inflate(&stream, Z_FINISH);
1088 offset += stream.next_in - in;
1089 } while ((st == Z_OK || st == Z_BUF_ERROR)
1090 && stream.total_out < sizeof(delta_head));
1091 inflateEnd(&stream);
1092 if ((st != Z_STREAM_END) &&
1093 stream.total_out != sizeof(delta_head))
1094 die("delta data unpack-initial failed");
1096 /* Examine the initial part of the delta to figure out
1101 /* ignore base size */
1102 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1104 /* Read the result size */
1105 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1106 *sizep = result_size;
1111 static unsigned long unpack_object_header(struct packed_git *p,
1112 struct pack_window **w_curs,
1113 unsigned long offset,
1114 enum object_type *type,
1115 unsigned long *sizep)
1117 unsigned char *base;
1121 /* use_pack() assures us we have [base, base + 20) available
1122 * as a range that we can look at at. (Its actually the hash
1123 * size that is assurred.) With our object header encoding
1124 * the maximum deflated object size is 2^137, which is just
1125 * insane, so we know won't exceed what we have been given.
1127 base = use_pack(p, w_curs, offset, &left);
1128 used = unpack_object_header_gently(base, left, type, sizep);
1130 die("object offset outside of pack file");
1132 return offset + used;
1135 void packed_object_info_detail(struct packed_git *p,
1136 unsigned long offset,
1138 unsigned long *size,
1139 unsigned long *store_size,
1140 unsigned int *delta_chain_length,
1141 unsigned char *base_sha1)
1143 struct pack_window *w_curs = NULL;
1144 unsigned long obj_offset, val;
1145 unsigned char *next_sha1;
1146 enum object_type kind;
1148 *delta_chain_length = 0;
1149 obj_offset = offset;
1150 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1155 die("pack %s contains unknown object type %d",
1156 p->pack_name, kind);
1161 strcpy(type, type_names[kind]);
1162 *store_size = 0; /* notyet */
1163 unuse_pack(&w_curs);
1166 get_delta_base(p, &w_curs, offset, kind,
1167 obj_offset, &offset);
1168 if (*delta_chain_length == 0) {
1169 /* TODO: find base_sha1 as pointed by offset */
1173 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1174 if (*delta_chain_length == 0)
1175 hashcpy(base_sha1, next_sha1);
1176 offset = find_pack_entry_one(next_sha1, p);
1179 obj_offset = offset;
1180 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1181 (*delta_chain_length)++;
1185 static int packed_object_info(struct packed_git *p, unsigned long offset,
1186 char *type, unsigned long *sizep)
1188 struct pack_window *w_curs = NULL;
1189 unsigned long size, obj_offset = offset;
1190 enum object_type kind;
1193 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1198 r = packed_delta_info(p, &w_curs, offset, kind,
1199 obj_offset, type, sizep);
1200 unuse_pack(&w_curs);
1206 strcpy(type, type_names[kind]);
1207 unuse_pack(&w_curs);
1210 die("pack %s contains unknown object type %d",
1211 p->pack_name, kind);
1218 static void *unpack_compressed_entry(struct packed_git *p,
1219 struct pack_window **w_curs,
1220 unsigned long offset,
1225 unsigned char *buffer, *in;
1227 buffer = xmalloc(size + 1);
1229 memset(&stream, 0, sizeof(stream));
1230 stream.next_out = buffer;
1231 stream.avail_out = size;
1233 inflateInit(&stream);
1235 in = use_pack(p, w_curs, offset, &stream.avail_in);
1236 stream.next_in = in;
1237 st = inflate(&stream, Z_FINISH);
1238 offset += stream.next_in - in;
1239 } while (st == Z_OK || st == Z_BUF_ERROR);
1240 inflateEnd(&stream);
1241 if ((st != Z_STREAM_END) || stream.total_out != size) {
1249 static void *unpack_delta_entry(struct packed_git *p,
1250 struct pack_window **w_curs,
1251 unsigned long offset,
1252 unsigned long delta_size,
1253 enum object_type kind,
1254 unsigned long obj_offset,
1256 unsigned long *sizep)
1258 void *delta_data, *result, *base;
1259 unsigned long result_size, base_size, base_offset;
1261 offset = get_delta_base(p, w_curs, offset, kind,
1262 obj_offset, &base_offset);
1263 base = unpack_entry(p, base_offset, type, &base_size);
1265 die("failed to read delta base object at %lu from %s",
1266 base_offset, p->pack_name);
1268 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1269 result = patch_delta(base, base_size,
1270 delta_data, delta_size,
1273 die("failed to apply delta");
1276 *sizep = result_size;
1280 void *unpack_entry(struct packed_git *p, unsigned long offset,
1281 char *type, unsigned long *sizep)
1283 struct pack_window *w_curs = NULL;
1284 unsigned long size, obj_offset = offset;
1285 enum object_type kind;
1288 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1292 retval = unpack_delta_entry(p, &w_curs, offset, size,
1293 kind, obj_offset, type, sizep);
1299 strcpy(type, type_names[kind]);
1301 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1304 die("unknown object type %i in %s", kind, p->pack_name);
1306 unuse_pack(&w_curs);
1310 int num_packed_objects(const struct packed_git *p)
1312 /* See check_packed_git_idx() */
1313 return (p->index_size - 20 - 20 - 4*256) / 24;
1316 int nth_packed_object_sha1(const struct packed_git *p, int n,
1317 unsigned char* sha1)
1319 void *index = p->index_base + 256;
1320 if (n < 0 || num_packed_objects(p) <= n)
1322 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1326 unsigned long find_pack_entry_one(const unsigned char *sha1,
1327 struct packed_git *p)
1329 unsigned int *level1_ofs = p->index_base;
1330 int hi = ntohl(level1_ofs[*sha1]);
1331 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1332 void *index = p->index_base + 256;
1335 int mi = (lo + hi) / 2;
1336 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1338 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1347 static int matches_pack_name(struct packed_git *p, const char *ig)
1349 const char *last_c, *c;
1351 if (!strcmp(p->pack_name, ig))
1354 for (c = p->pack_name, last_c = c; *c;)
1359 if (!strcmp(last_c, ig))
1365 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1367 struct packed_git *p;
1368 unsigned long offset;
1370 prepare_packed_git();
1372 for (p = packed_git; p; p = p->next) {
1373 if (ignore_packed) {
1375 for (ig = ignore_packed; *ig; ig++)
1376 if (!matches_pack_name(p, *ig))
1381 offset = find_pack_entry_one(sha1, p);
1385 hashcpy(e->sha1, sha1);
1392 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1393 struct packed_git *packs)
1395 struct packed_git *p;
1397 for (p = packs; p; p = p->next) {
1398 if (find_pack_entry_one(sha1, p))
1405 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1408 unsigned long mapsize, size;
1413 map = map_sha1_file(sha1, &mapsize);
1415 return error("unable to find %s", sha1_to_hex(sha1));
1416 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1417 status = error("unable to unpack %s header",
1419 if (parse_sha1_header(hdr, type, &size) < 0)
1420 status = error("unable to parse %s header", sha1_to_hex(sha1));
1426 inflateEnd(&stream);
1427 munmap(map, mapsize);
1431 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1433 struct pack_entry e;
1435 if (!find_pack_entry(sha1, &e, NULL)) {
1436 reprepare_packed_git();
1437 if (!find_pack_entry(sha1, &e, NULL))
1438 return sha1_loose_object_info(sha1, type, sizep);
1440 return packed_object_info(e.p, e.offset, type, sizep);
1443 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1445 struct pack_entry e;
1447 if (!find_pack_entry(sha1, &e, NULL)) {
1448 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1451 return unpack_entry(e.p, e.offset, type, size);
1454 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1456 unsigned long mapsize;
1458 struct pack_entry e;
1460 if (find_pack_entry(sha1, &e, NULL))
1461 return read_packed_sha1(sha1, type, size);
1462 map = map_sha1_file(sha1, &mapsize);
1464 buf = unpack_sha1_file(map, mapsize, type, size);
1465 munmap(map, mapsize);
1468 reprepare_packed_git();
1469 if (find_pack_entry(sha1, &e, NULL))
1470 return read_packed_sha1(sha1, type, size);
1474 void *read_object_with_reference(const unsigned char *sha1,
1475 const char *required_type,
1476 unsigned long *size,
1477 unsigned char *actual_sha1_return)
1481 unsigned long isize;
1482 unsigned char actual_sha1[20];
1484 hashcpy(actual_sha1, sha1);
1486 int ref_length = -1;
1487 const char *ref_type = NULL;
1489 buffer = read_sha1_file(actual_sha1, type, &isize);
1492 if (!strcmp(type, required_type)) {
1494 if (actual_sha1_return)
1495 hashcpy(actual_sha1_return, actual_sha1);
1498 /* Handle references */
1499 else if (!strcmp(type, commit_type))
1501 else if (!strcmp(type, tag_type))
1502 ref_type = "object ";
1507 ref_length = strlen(ref_type);
1509 if (memcmp(buffer, ref_type, ref_length) ||
1510 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1515 /* Now we have the ID of the referred-to object in
1516 * actual_sha1. Check again. */
1520 static void write_sha1_file_prepare(void *buf, unsigned long len,
1521 const char *type, unsigned char *sha1,
1522 unsigned char *hdr, int *hdrlen)
1526 /* Generate the header */
1527 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1531 SHA1_Update(&c, hdr, *hdrlen);
1532 SHA1_Update(&c, buf, len);
1533 SHA1_Final(sha1, &c);
1537 * Link the tempfile to the final place, possibly creating the
1538 * last directory level as you do so.
1540 * Returns the errno on failure, 0 on success.
1542 static int link_temp_to_file(const char *tmpfile, const char *filename)
1547 if (!link(tmpfile, filename))
1551 * Try to mkdir the last path component if that failed.
1553 * Re-try the "link()" regardless of whether the mkdir
1554 * succeeds, since a race might mean that somebody
1558 dir = strrchr(filename, '/');
1561 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1566 if (!link(tmpfile, filename))
1574 * Move the just written object into its final resting place
1576 int move_temp_to_file(const char *tmpfile, const char *filename)
1578 int ret = link_temp_to_file(tmpfile, filename);
1581 * Coda hack - coda doesn't like cross-directory links,
1582 * so we fall back to a rename, which will mean that it
1583 * won't be able to check collisions, but that's not a
1586 * The same holds for FAT formatted media.
1588 * When this succeeds, we just return 0. We have nothing
1591 if (ret && ret != EEXIST) {
1592 if (!rename(tmpfile, filename))
1598 if (ret != EEXIST) {
1599 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1601 /* FIXME!!! Collision check here ? */
1607 static int write_buffer(int fd, const void *buf, size_t len)
1612 size = write(fd, buf, len);
1614 return error("file write: disk full");
1616 if (errno == EINTR || errno == EAGAIN)
1618 return error("file write error (%s)", strerror(errno));
1621 buf = (char *) buf + size;
1626 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1631 c = (type << 4) | (len & 15);
1644 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1648 if (use_legacy_headers) {
1649 while (deflate(stream, 0) == Z_OK)
1653 if (!strcmp(type, blob_type))
1654 obj_type = OBJ_BLOB;
1655 else if (!strcmp(type, tree_type))
1656 obj_type = OBJ_TREE;
1657 else if (!strcmp(type, commit_type))
1658 obj_type = OBJ_COMMIT;
1659 else if (!strcmp(type, tag_type))
1662 die("trying to generate bogus object of type '%s'", type);
1663 hdr = write_binary_header(stream->next_out, obj_type, len);
1664 stream->total_out = hdr;
1665 stream->next_out += hdr;
1666 stream->avail_out -= hdr;
1669 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1670 unsigned char *sha1)
1672 unsigned char hdr[50];
1674 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1678 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1681 unsigned char *compressed;
1683 unsigned char sha1[20];
1685 static char tmpfile[PATH_MAX];
1686 unsigned char hdr[50];
1689 /* Normally if we have it in the pack then we do not bother writing
1690 * it out into .git/objects/??/?{38} file.
1692 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1693 filename = sha1_file_name(sha1);
1695 hashcpy(returnsha1, sha1);
1696 if (has_sha1_file(sha1))
1698 fd = open(filename, O_RDONLY);
1701 * FIXME!!! We might do collision checking here, but we'd
1702 * need to uncompress the old file and check it. Later.
1708 if (errno != ENOENT) {
1709 return error("sha1 file %s: %s\n", filename, strerror(errno));
1712 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1714 fd = mkstemp(tmpfile);
1717 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1719 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1723 memset(&stream, 0, sizeof(stream));
1724 deflateInit(&stream, zlib_compression_level);
1725 size = 8 + deflateBound(&stream, len+hdrlen);
1726 compressed = xmalloc(size);
1729 stream.next_out = compressed;
1730 stream.avail_out = size;
1732 /* First header.. */
1733 stream.next_in = hdr;
1734 stream.avail_in = hdrlen;
1735 setup_object_header(&stream, type, len);
1737 /* Then the data itself.. */
1738 stream.next_in = buf;
1739 stream.avail_in = len;
1740 while (deflate(&stream, Z_FINISH) == Z_OK)
1742 deflateEnd(&stream);
1743 size = stream.total_out;
1745 if (write_buffer(fd, compressed, size) < 0)
1746 die("unable to write sha1 file");
1751 return move_temp_to_file(tmpfile, filename);
1755 * We need to unpack and recompress the object for writing
1756 * it out to a different file.
1758 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1762 unsigned char *unpacked;
1769 /* need to unpack and recompress it by itself */
1770 unpacked = read_packed_sha1(sha1, type, &len);
1772 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1775 memset(&stream, 0, sizeof(stream));
1776 deflateInit(&stream, zlib_compression_level);
1777 size = deflateBound(&stream, len + hdrlen);
1778 buf = xmalloc(size);
1781 stream.next_out = buf;
1782 stream.avail_out = size;
1784 /* First header.. */
1785 stream.next_in = (void *)hdr;
1786 stream.avail_in = hdrlen;
1787 while (deflate(&stream, 0) == Z_OK)
1790 /* Then the data itself.. */
1791 stream.next_in = unpacked;
1792 stream.avail_in = len;
1793 while (deflate(&stream, Z_FINISH) == Z_OK)
1795 deflateEnd(&stream);
1798 *objsize = stream.total_out;
1802 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1805 unsigned long objsize;
1806 void *buf = map_sha1_file(sha1, &objsize);
1809 retval = write_buffer(fd, buf, objsize);
1810 munmap(buf, objsize);
1814 buf = repack_object(sha1, &objsize);
1815 retval = write_buffer(fd, buf, objsize);
1820 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1821 size_t bufsize, size_t *bufposn)
1823 char tmpfile[PATH_MAX];
1826 unsigned char real_sha1[20];
1827 unsigned char discard[4096];
1831 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1833 local = mkstemp(tmpfile);
1836 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1838 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1841 memset(&stream, 0, sizeof(stream));
1843 inflateInit(&stream);
1850 stream.avail_in = *bufposn;
1851 stream.next_in = (unsigned char *) buffer;
1853 stream.next_out = discard;
1854 stream.avail_out = sizeof(discard);
1855 ret = inflate(&stream, Z_SYNC_FLUSH);
1856 SHA1_Update(&c, discard, sizeof(discard) -
1858 } while (stream.avail_in && ret == Z_OK);
1859 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1860 die("unable to write sha1 file");
1861 memmove(buffer, buffer + *bufposn - stream.avail_in,
1863 *bufposn = stream.avail_in;
1867 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1872 return error("Connection closed?");
1873 perror("Reading from connection");
1878 inflateEnd(&stream);
1881 SHA1_Final(real_sha1, &c);
1882 if (ret != Z_STREAM_END) {
1884 return error("File %s corrupted", sha1_to_hex(sha1));
1886 if (hashcmp(sha1, real_sha1)) {
1888 return error("File %s has bad hash", sha1_to_hex(sha1));
1891 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1894 int has_pack_index(const unsigned char *sha1)
1897 if (stat(sha1_pack_index_name(sha1), &st))
1902 int has_pack_file(const unsigned char *sha1)
1905 if (stat(sha1_pack_name(sha1), &st))
1910 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1912 struct pack_entry e;
1913 return find_pack_entry(sha1, &e, ignore_packed);
1916 int has_sha1_file(const unsigned char *sha1)
1919 struct pack_entry e;
1921 if (find_pack_entry(sha1, &e, NULL))
1923 return find_sha1_file(sha1, &st) ? 1 : 0;
1927 * reads from fd as long as possible into a supplied buffer of size bytes.
1928 * If necessary the buffer's size is increased using realloc()
1930 * returns 0 if anything went fine and -1 otherwise
1932 * NOTE: both buf and size may change, but even when -1 is returned
1933 * you still have to free() it yourself.
1935 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1937 char* buf = *return_buf;
1938 unsigned long size = *return_size;
1940 unsigned long off = 0;
1943 iret = xread(fd, buf + off, size - off);
1948 buf = xrealloc(buf, size);
1961 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1963 unsigned long size = 4096;
1964 char *buf = xmalloc(size);
1967 if (read_pipe(fd, &buf, &size)) {
1975 ret = write_sha1_file(buf, size, type, sha1);
1977 ret = hash_sha1_file(buf, size, type, sha1);
1982 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1984 unsigned long size = st->st_size;
1990 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1992 if (buf == MAP_FAILED)
1998 ret = write_sha1_file(buf, size, type, sha1);
2000 ret = hash_sha1_file(buf, size, type, sha1);
2006 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2011 switch (st->st_mode & S_IFMT) {
2013 fd = open(path, O_RDONLY);
2015 return error("open(\"%s\"): %s", path,
2017 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2018 return error("%s: failed to insert into database",
2022 target = xmalloc(st->st_size+1);
2023 if (readlink(path, target, st->st_size+1) != st->st_size) {
2024 char *errstr = strerror(errno);
2026 return error("readlink(\"%s\"): %s", path,
2030 hash_sha1_file(target, st->st_size, blob_type, sha1);
2031 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2032 return error("%s: failed to insert into database",
2037 return error("%s: unsupported file type", path);