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 int pack_used_ctr;
401 static unsigned long pack_mapped;
402 struct packed_git *packed_git;
404 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
409 unsigned long idx_size;
411 int fd = open(path, O_RDONLY);
415 if (fstat(fd, &st)) {
419 idx_size = st.st_size;
420 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
422 if (idx_map == MAP_FAILED)
427 *idx_size_ = idx_size;
429 /* check index map */
430 if (idx_size < 4*256 + 20 + 20)
431 return error("index file too small");
433 for (i = 0; i < 256; i++) {
434 unsigned int n = ntohl(index[i]);
436 return error("non-monotonic index");
442 * - 256 index entries 4 bytes each
443 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
444 * - 20-byte SHA1 of the packfile
445 * - 20-byte SHA1 file checksum
447 if (idx_size != 4*256 + nr * 24 + 20 + 20)
448 return error("wrong index file size");
453 static int unuse_one_packed_git(void)
455 struct packed_git *p, *lru = NULL;
457 for (p = packed_git; p; p = p->next) {
458 if (!p->windows || p->windows->inuse_cnt)
460 if (!lru || p->windows->last_used < lru->windows->last_used)
465 munmap(lru->windows->base, lru->windows->len);
473 void unuse_pack(struct pack_window **w_cursor)
475 struct pack_window *w = *w_cursor;
482 static void open_packed_git(struct packed_git *p)
485 struct pack_header hdr;
486 unsigned char sha1[20];
487 unsigned char *idx_sha1;
489 p->pack_fd = open(p->pack_name, O_RDONLY);
490 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
491 die("packfile %s cannot be opened", p->pack_name);
493 /* If we created the struct before we had the pack we lack size. */
495 if (!S_ISREG(st.st_mode))
496 die("packfile %s not a regular file", p->pack_name);
497 p->pack_size = st.st_size;
498 } else if (p->pack_size != st.st_size)
499 die("packfile %s size changed", p->pack_name);
501 /* Verify we recognize this pack file format. */
502 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
503 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
504 die("file %s is not a GIT packfile", p->pack_name);
505 if (!pack_version_ok(hdr.hdr_version))
506 die("packfile %s is version %u and not supported"
507 " (try upgrading GIT to a newer version)",
508 p->pack_name, ntohl(hdr.hdr_version));
510 /* Verify the pack matches its index. */
511 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
512 die("packfile %s claims to have %u objects"
513 " while index size indicates %u objects",
514 p->pack_name, ntohl(hdr.hdr_entries),
515 num_packed_objects(p));
516 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
517 die("end of packfile %s is unavailable", p->pack_name);
518 read_or_die(p->pack_fd, sha1, sizeof(sha1));
519 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
520 if (hashcmp(sha1, idx_sha1))
521 die("packfile %s does not match index", p->pack_name);
524 unsigned char* use_pack(struct packed_git *p,
525 struct pack_window **w_cursor,
526 unsigned long offset,
529 struct pack_window *win = p->windows;
531 if (p->pack_fd == -1)
534 pack_mapped += p->pack_size;
535 while (packed_git_limit < pack_mapped && unuse_one_packed_git())
537 win = xcalloc(1, sizeof(*win));
538 win->len = p->pack_size;
539 win->base = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, p->pack_fd, 0);
540 if (win->base == MAP_FAILED)
541 die("packfile %s cannot be mapped.", p->pack_name);
544 if (win != *w_cursor) {
545 win->last_used = pack_used_ctr++;
550 *left = win->len - offset;
551 return win->base + offset;
554 struct packed_git *add_packed_git(char *path, int path_len, int local)
557 struct packed_git *p;
558 unsigned long idx_size;
560 unsigned char sha1[20];
562 if (check_packed_git_idx(path, &idx_size, &idx_map))
565 /* do we have a corresponding .pack file? */
566 strcpy(path + path_len - 4, ".pack");
567 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
568 munmap(idx_map, idx_size);
571 /* ok, it looks sane as far as we can check without
572 * actually mapping the pack file.
574 p = xmalloc(sizeof(*p) + path_len + 2);
575 strcpy(p->pack_name, path);
576 p->index_size = idx_size;
577 p->pack_size = st.st_size;
578 p->index_base = idx_map;
582 p->pack_local = local;
583 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
584 hashcpy(p->sha1, sha1);
588 struct packed_git *parse_pack_index(unsigned char *sha1)
590 char *path = sha1_pack_index_name(sha1);
591 return parse_pack_index_file(sha1, path);
594 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
596 struct packed_git *p;
597 unsigned long idx_size;
601 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
604 path = sha1_pack_name(sha1);
606 p = xmalloc(sizeof(*p) + strlen(path) + 2);
607 strcpy(p->pack_name, path);
608 p->index_size = idx_size;
610 p->index_base = idx_map;
614 hashcpy(p->sha1, sha1);
618 void install_packed_git(struct packed_git *pack)
620 pack->next = packed_git;
624 static void prepare_packed_git_one(char *objdir, int local)
631 sprintf(path, "%s/pack", objdir);
636 error("unable to open object pack directory: %s: %s",
637 path, strerror(errno));
641 while ((de = readdir(dir)) != NULL) {
642 int namelen = strlen(de->d_name);
643 struct packed_git *p;
645 if (!has_extension(de->d_name, ".idx"))
648 /* we have .idx. Is it a file we can map? */
649 strcpy(path + len, de->d_name);
650 for (p = packed_git; p; p = p->next) {
651 if (!memcmp(path, p->pack_name, len + namelen - 4))
656 p = add_packed_git(path, len + namelen, local);
659 p->next = packed_git;
665 static int prepare_packed_git_run_once = 0;
666 void prepare_packed_git(void)
668 struct alternate_object_database *alt;
670 if (prepare_packed_git_run_once)
672 prepare_packed_git_one(get_object_directory(), 1);
674 for (alt = alt_odb_list; alt; alt = alt->next) {
676 prepare_packed_git_one(alt->base, 0);
679 prepare_packed_git_run_once = 1;
682 void reprepare_packed_git(void)
684 prepare_packed_git_run_once = 0;
685 prepare_packed_git();
688 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
690 unsigned char real_sha1[20];
691 hash_sha1_file(map, size, type, real_sha1);
692 return hashcmp(sha1, real_sha1) ? -1 : 0;
695 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
700 char *filename = find_sha1_file(sha1, &st);
706 fd = open(filename, O_RDONLY | sha1_file_open_flag);
708 /* See if it works without O_NOATIME */
709 switch (sha1_file_open_flag) {
711 fd = open(filename, O_RDONLY);
719 /* If it failed once, it will probably fail again.
720 * Stop using O_NOATIME
722 sha1_file_open_flag = 0;
724 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
726 if (map == MAP_FAILED)
732 int legacy_loose_object(unsigned char *map)
737 * Is it a zlib-compressed buffer? If so, the first byte
738 * must be 0x78 (15-bit window size, deflated), and the
739 * first 16-bit word is evenly divisible by 31
741 word = (map[0] << 8) + map[1];
742 if (map[0] == 0x78 && !(word % 31))
748 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
753 unsigned long used = 0;
756 *type = (c >> 4) & 7;
762 if (sizeof(long) * 8 <= shift)
765 size += (c & 0x7f) << shift;
772 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
774 unsigned long size, used;
775 static const char valid_loose_object_type[8] = {
777 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
778 0, /* "delta" and others are invalid in a loose object */
780 enum object_type type;
782 /* Get the data stream */
783 memset(stream, 0, sizeof(*stream));
784 stream->next_in = map;
785 stream->avail_in = mapsize;
786 stream->next_out = buffer;
787 stream->avail_out = bufsiz;
789 if (legacy_loose_object(map)) {
791 return inflate(stream, 0);
794 used = unpack_object_header_gently(map, mapsize, &type, &size);
795 if (!used || !valid_loose_object_type[type])
800 /* Set up the stream for the rest.. */
801 stream->next_in = map;
802 stream->avail_in = mapsize;
805 /* And generate the fake traditional header */
806 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
807 type_names[type], size);
811 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
813 int bytes = strlen(buffer) + 1;
814 unsigned char *buf = xmalloc(1+size);
817 n = stream->total_out - bytes;
820 memcpy(buf, (char *) buffer + bytes, n);
823 stream->next_out = buf + bytes;
824 stream->avail_out = size - bytes;
825 while (inflate(stream, Z_FINISH) == Z_OK)
834 * We used to just use "sscanf()", but that's actually way
835 * too permissive for what we want to check. So do an anal
836 * object header parse by hand.
838 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
844 * The type can be at most ten bytes (including the
845 * terminating '\0' that we add), and is followed by
860 * The length must follow immediately, and be in canonical
861 * decimal format (ie "010" is not valid).
868 unsigned long c = *hdr - '0';
872 size = size * 10 + c;
878 * The length must be followed by a zero byte
880 return *hdr ? -1 : 0;
883 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
889 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
890 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
893 return unpack_sha1_rest(&stream, hdr, *size);
896 static unsigned long get_delta_base(struct packed_git *p,
897 struct pack_window **w_curs,
898 unsigned long offset,
899 enum object_type kind,
900 unsigned long delta_obj_offset,
901 unsigned long *base_obj_offset)
903 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
904 unsigned long base_offset;
906 /* there must be at least 20 bytes left regardless of delta type */
907 if (p->pack_size <= offset + 20)
908 die("truncated pack file");
910 if (kind == OBJ_OFS_DELTA) {
912 unsigned char c = base_info[used++];
913 base_offset = c & 127;
916 if (!base_offset || base_offset & ~(~0UL >> 7))
917 die("offset value overflow for delta base object");
918 c = base_info[used++];
919 base_offset = (base_offset << 7) + (c & 127);
921 base_offset = delta_obj_offset - base_offset;
922 if (base_offset >= delta_obj_offset)
923 die("delta base offset out of bound");
925 } else if (kind == OBJ_REF_DELTA) {
926 /* The base entry _must_ be in the same pack */
927 base_offset = find_pack_entry_one(base_info, p);
929 die("failed to find delta-pack base object %s",
930 sha1_to_hex(base_info));
933 die("I am totally screwed");
934 *base_obj_offset = base_offset;
938 /* forward declaration for a mutually recursive function */
939 static int packed_object_info(struct packed_git *p, unsigned long offset,
940 char *type, unsigned long *sizep);
942 static int packed_delta_info(struct packed_git *p,
943 struct pack_window **w_curs,
944 unsigned long offset,
945 enum object_type kind,
946 unsigned long obj_offset,
948 unsigned long *sizep)
950 unsigned long base_offset;
952 offset = get_delta_base(p, w_curs, offset, kind,
953 obj_offset, &base_offset);
955 /* We choose to only get the type of the base object and
956 * ignore potentially corrupt pack file that expects the delta
957 * based on a base with a wrong size. This saves tons of
960 if (packed_object_info(p, base_offset, type, NULL))
961 die("cannot get info for delta-pack base");
964 const unsigned char *data;
965 unsigned char delta_head[20];
966 unsigned long result_size;
970 memset(&stream, 0, sizeof(stream));
972 stream.next_in = use_pack(p, w_curs, offset, &stream.avail_in);
973 stream.next_out = delta_head;
974 stream.avail_out = sizeof(delta_head);
976 inflateInit(&stream);
977 st = inflate(&stream, Z_FINISH);
979 if ((st != Z_STREAM_END) &&
980 stream.total_out != sizeof(delta_head))
981 die("delta data unpack-initial failed");
983 /* Examine the initial part of the delta to figure out
988 /* ignore base size */
989 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
991 /* Read the result size */
992 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
993 *sizep = result_size;
998 static unsigned long unpack_object_header(struct packed_git *p,
999 struct pack_window **w_curs,
1000 unsigned long offset,
1001 enum object_type *type,
1002 unsigned long *sizep)
1004 unsigned char *base;
1008 base = use_pack(p, w_curs, offset, &left);
1009 used = unpack_object_header_gently(base, left, type, sizep);
1011 die("object offset outside of pack file");
1013 return offset + used;
1016 void packed_object_info_detail(struct packed_git *p,
1017 unsigned long offset,
1019 unsigned long *size,
1020 unsigned long *store_size,
1021 unsigned int *delta_chain_length,
1022 unsigned char *base_sha1)
1024 struct pack_window *w_curs = NULL;
1025 unsigned long obj_offset, val;
1026 unsigned char *next_sha1;
1027 enum object_type kind;
1029 *delta_chain_length = 0;
1030 obj_offset = offset;
1031 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1036 die("pack %s contains unknown object type %d",
1037 p->pack_name, kind);
1042 strcpy(type, type_names[kind]);
1043 *store_size = 0; /* notyet */
1044 unuse_pack(&w_curs);
1047 get_delta_base(p, &w_curs, offset, kind,
1048 obj_offset, &offset);
1049 if (*delta_chain_length == 0) {
1050 /* TODO: find base_sha1 as pointed by offset */
1054 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1055 if (*delta_chain_length == 0)
1056 hashcpy(base_sha1, next_sha1);
1057 offset = find_pack_entry_one(next_sha1, p);
1060 obj_offset = offset;
1061 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1062 (*delta_chain_length)++;
1066 static int packed_object_info(struct packed_git *p, unsigned long offset,
1067 char *type, unsigned long *sizep)
1069 struct pack_window *w_curs = NULL;
1070 unsigned long size, obj_offset = offset;
1071 enum object_type kind;
1074 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1079 r = packed_delta_info(p, &w_curs, offset, kind,
1080 obj_offset, type, sizep);
1081 unuse_pack(&w_curs);
1087 strcpy(type, type_names[kind]);
1088 unuse_pack(&w_curs);
1091 die("pack %s contains unknown object type %d",
1092 p->pack_name, kind);
1099 static void *unpack_compressed_entry(struct packed_git *p,
1100 struct pack_window **w_curs,
1101 unsigned long offset,
1106 unsigned char *buffer;
1108 buffer = xmalloc(size + 1);
1110 memset(&stream, 0, sizeof(stream));
1111 stream.next_in = use_pack(p, w_curs, offset, &stream.avail_in);
1112 stream.next_out = buffer;
1113 stream.avail_out = size;
1115 inflateInit(&stream);
1116 st = inflate(&stream, Z_FINISH);
1117 inflateEnd(&stream);
1118 if ((st != Z_STREAM_END) || stream.total_out != size) {
1126 static void *unpack_delta_entry(struct packed_git *p,
1127 struct pack_window **w_curs,
1128 unsigned long offset,
1129 unsigned long delta_size,
1130 enum object_type kind,
1131 unsigned long obj_offset,
1133 unsigned long *sizep)
1135 void *delta_data, *result, *base;
1136 unsigned long result_size, base_size, base_offset;
1138 offset = get_delta_base(p, w_curs, offset, kind,
1139 obj_offset, &base_offset);
1140 base = unpack_entry(p, base_offset, type, &base_size);
1142 die("failed to read delta base object at %lu from %s",
1143 base_offset, p->pack_name);
1145 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1146 result = patch_delta(base, base_size,
1147 delta_data, delta_size,
1150 die("failed to apply delta");
1153 *sizep = result_size;
1157 void *unpack_entry(struct packed_git *p, unsigned long offset,
1158 char *type, unsigned long *sizep)
1160 struct pack_window *w_curs = NULL;
1161 unsigned long size, obj_offset = offset;
1162 enum object_type kind;
1165 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1169 retval = unpack_delta_entry(p, &w_curs, offset, size,
1170 kind, obj_offset, type, sizep);
1176 strcpy(type, type_names[kind]);
1178 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1181 die("unknown object type %i in %s", kind, p->pack_name);
1183 unuse_pack(&w_curs);
1187 int num_packed_objects(const struct packed_git *p)
1189 /* See check_packed_git_idx() */
1190 return (p->index_size - 20 - 20 - 4*256) / 24;
1193 int nth_packed_object_sha1(const struct packed_git *p, int n,
1194 unsigned char* sha1)
1196 void *index = p->index_base + 256;
1197 if (n < 0 || num_packed_objects(p) <= n)
1199 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1203 unsigned long find_pack_entry_one(const unsigned char *sha1,
1204 struct packed_git *p)
1206 unsigned int *level1_ofs = p->index_base;
1207 int hi = ntohl(level1_ofs[*sha1]);
1208 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1209 void *index = p->index_base + 256;
1212 int mi = (lo + hi) / 2;
1213 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1215 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1224 static int matches_pack_name(struct packed_git *p, const char *ig)
1226 const char *last_c, *c;
1228 if (!strcmp(p->pack_name, ig))
1231 for (c = p->pack_name, last_c = c; *c;)
1236 if (!strcmp(last_c, ig))
1242 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1244 struct packed_git *p;
1245 unsigned long offset;
1247 prepare_packed_git();
1249 for (p = packed_git; p; p = p->next) {
1250 if (ignore_packed) {
1252 for (ig = ignore_packed; *ig; ig++)
1253 if (!matches_pack_name(p, *ig))
1258 offset = find_pack_entry_one(sha1, p);
1262 hashcpy(e->sha1, sha1);
1269 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1270 struct packed_git *packs)
1272 struct packed_git *p;
1274 for (p = packs; p; p = p->next) {
1275 if (find_pack_entry_one(sha1, p))
1282 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1285 unsigned long mapsize, size;
1290 map = map_sha1_file(sha1, &mapsize);
1292 return error("unable to find %s", sha1_to_hex(sha1));
1293 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1294 status = error("unable to unpack %s header",
1296 if (parse_sha1_header(hdr, type, &size) < 0)
1297 status = error("unable to parse %s header", sha1_to_hex(sha1));
1303 inflateEnd(&stream);
1304 munmap(map, mapsize);
1308 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1310 struct pack_entry e;
1312 if (!find_pack_entry(sha1, &e, NULL)) {
1313 reprepare_packed_git();
1314 if (!find_pack_entry(sha1, &e, NULL))
1315 return sha1_loose_object_info(sha1, type, sizep);
1317 return packed_object_info(e.p, e.offset, type, sizep);
1320 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1322 struct pack_entry e;
1324 if (!find_pack_entry(sha1, &e, NULL)) {
1325 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1328 return unpack_entry(e.p, e.offset, type, size);
1331 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1333 unsigned long mapsize;
1335 struct pack_entry e;
1337 if (find_pack_entry(sha1, &e, NULL))
1338 return read_packed_sha1(sha1, type, size);
1339 map = map_sha1_file(sha1, &mapsize);
1341 buf = unpack_sha1_file(map, mapsize, type, size);
1342 munmap(map, mapsize);
1345 reprepare_packed_git();
1346 if (find_pack_entry(sha1, &e, NULL))
1347 return read_packed_sha1(sha1, type, size);
1351 void *read_object_with_reference(const unsigned char *sha1,
1352 const char *required_type,
1353 unsigned long *size,
1354 unsigned char *actual_sha1_return)
1358 unsigned long isize;
1359 unsigned char actual_sha1[20];
1361 hashcpy(actual_sha1, sha1);
1363 int ref_length = -1;
1364 const char *ref_type = NULL;
1366 buffer = read_sha1_file(actual_sha1, type, &isize);
1369 if (!strcmp(type, required_type)) {
1371 if (actual_sha1_return)
1372 hashcpy(actual_sha1_return, actual_sha1);
1375 /* Handle references */
1376 else if (!strcmp(type, commit_type))
1378 else if (!strcmp(type, tag_type))
1379 ref_type = "object ";
1384 ref_length = strlen(ref_type);
1386 if (memcmp(buffer, ref_type, ref_length) ||
1387 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1392 /* Now we have the ID of the referred-to object in
1393 * actual_sha1. Check again. */
1397 static void write_sha1_file_prepare(void *buf, unsigned long len,
1398 const char *type, unsigned char *sha1,
1399 unsigned char *hdr, int *hdrlen)
1403 /* Generate the header */
1404 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1408 SHA1_Update(&c, hdr, *hdrlen);
1409 SHA1_Update(&c, buf, len);
1410 SHA1_Final(sha1, &c);
1414 * Link the tempfile to the final place, possibly creating the
1415 * last directory level as you do so.
1417 * Returns the errno on failure, 0 on success.
1419 static int link_temp_to_file(const char *tmpfile, const char *filename)
1424 if (!link(tmpfile, filename))
1428 * Try to mkdir the last path component if that failed.
1430 * Re-try the "link()" regardless of whether the mkdir
1431 * succeeds, since a race might mean that somebody
1435 dir = strrchr(filename, '/');
1438 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1443 if (!link(tmpfile, filename))
1451 * Move the just written object into its final resting place
1453 int move_temp_to_file(const char *tmpfile, const char *filename)
1455 int ret = link_temp_to_file(tmpfile, filename);
1458 * Coda hack - coda doesn't like cross-directory links,
1459 * so we fall back to a rename, which will mean that it
1460 * won't be able to check collisions, but that's not a
1463 * The same holds for FAT formatted media.
1465 * When this succeeds, we just return 0. We have nothing
1468 if (ret && ret != EEXIST) {
1469 if (!rename(tmpfile, filename))
1475 if (ret != EEXIST) {
1476 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1478 /* FIXME!!! Collision check here ? */
1484 static int write_buffer(int fd, const void *buf, size_t len)
1489 size = write(fd, buf, len);
1491 return error("file write: disk full");
1493 if (errno == EINTR || errno == EAGAIN)
1495 return error("file write error (%s)", strerror(errno));
1498 buf = (char *) buf + size;
1503 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1508 c = (type << 4) | (len & 15);
1521 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1525 if (use_legacy_headers) {
1526 while (deflate(stream, 0) == Z_OK)
1530 if (!strcmp(type, blob_type))
1531 obj_type = OBJ_BLOB;
1532 else if (!strcmp(type, tree_type))
1533 obj_type = OBJ_TREE;
1534 else if (!strcmp(type, commit_type))
1535 obj_type = OBJ_COMMIT;
1536 else if (!strcmp(type, tag_type))
1539 die("trying to generate bogus object of type '%s'", type);
1540 hdr = write_binary_header(stream->next_out, obj_type, len);
1541 stream->total_out = hdr;
1542 stream->next_out += hdr;
1543 stream->avail_out -= hdr;
1546 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1547 unsigned char *sha1)
1549 unsigned char hdr[50];
1551 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1555 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1558 unsigned char *compressed;
1560 unsigned char sha1[20];
1562 static char tmpfile[PATH_MAX];
1563 unsigned char hdr[50];
1566 /* Normally if we have it in the pack then we do not bother writing
1567 * it out into .git/objects/??/?{38} file.
1569 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1570 filename = sha1_file_name(sha1);
1572 hashcpy(returnsha1, sha1);
1573 if (has_sha1_file(sha1))
1575 fd = open(filename, O_RDONLY);
1578 * FIXME!!! We might do collision checking here, but we'd
1579 * need to uncompress the old file and check it. Later.
1585 if (errno != ENOENT) {
1586 return error("sha1 file %s: %s\n", filename, strerror(errno));
1589 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1591 fd = mkstemp(tmpfile);
1594 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1596 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1600 memset(&stream, 0, sizeof(stream));
1601 deflateInit(&stream, zlib_compression_level);
1602 size = 8 + deflateBound(&stream, len+hdrlen);
1603 compressed = xmalloc(size);
1606 stream.next_out = compressed;
1607 stream.avail_out = size;
1609 /* First header.. */
1610 stream.next_in = hdr;
1611 stream.avail_in = hdrlen;
1612 setup_object_header(&stream, type, len);
1614 /* Then the data itself.. */
1615 stream.next_in = buf;
1616 stream.avail_in = len;
1617 while (deflate(&stream, Z_FINISH) == Z_OK)
1619 deflateEnd(&stream);
1620 size = stream.total_out;
1622 if (write_buffer(fd, compressed, size) < 0)
1623 die("unable to write sha1 file");
1628 return move_temp_to_file(tmpfile, filename);
1632 * We need to unpack and recompress the object for writing
1633 * it out to a different file.
1635 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1639 unsigned char *unpacked;
1646 /* need to unpack and recompress it by itself */
1647 unpacked = read_packed_sha1(sha1, type, &len);
1649 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1652 memset(&stream, 0, sizeof(stream));
1653 deflateInit(&stream, zlib_compression_level);
1654 size = deflateBound(&stream, len + hdrlen);
1655 buf = xmalloc(size);
1658 stream.next_out = buf;
1659 stream.avail_out = size;
1661 /* First header.. */
1662 stream.next_in = (void *)hdr;
1663 stream.avail_in = hdrlen;
1664 while (deflate(&stream, 0) == Z_OK)
1667 /* Then the data itself.. */
1668 stream.next_in = unpacked;
1669 stream.avail_in = len;
1670 while (deflate(&stream, Z_FINISH) == Z_OK)
1672 deflateEnd(&stream);
1675 *objsize = stream.total_out;
1679 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1682 unsigned long objsize;
1683 void *buf = map_sha1_file(sha1, &objsize);
1686 retval = write_buffer(fd, buf, objsize);
1687 munmap(buf, objsize);
1691 buf = repack_object(sha1, &objsize);
1692 retval = write_buffer(fd, buf, objsize);
1697 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1698 size_t bufsize, size_t *bufposn)
1700 char tmpfile[PATH_MAX];
1703 unsigned char real_sha1[20];
1704 unsigned char discard[4096];
1708 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1710 local = mkstemp(tmpfile);
1713 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1715 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1718 memset(&stream, 0, sizeof(stream));
1720 inflateInit(&stream);
1727 stream.avail_in = *bufposn;
1728 stream.next_in = (unsigned char *) buffer;
1730 stream.next_out = discard;
1731 stream.avail_out = sizeof(discard);
1732 ret = inflate(&stream, Z_SYNC_FLUSH);
1733 SHA1_Update(&c, discard, sizeof(discard) -
1735 } while (stream.avail_in && ret == Z_OK);
1736 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1737 die("unable to write sha1 file");
1738 memmove(buffer, buffer + *bufposn - stream.avail_in,
1740 *bufposn = stream.avail_in;
1744 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1749 return error("Connection closed?");
1750 perror("Reading from connection");
1755 inflateEnd(&stream);
1758 SHA1_Final(real_sha1, &c);
1759 if (ret != Z_STREAM_END) {
1761 return error("File %s corrupted", sha1_to_hex(sha1));
1763 if (hashcmp(sha1, real_sha1)) {
1765 return error("File %s has bad hash", sha1_to_hex(sha1));
1768 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1771 int has_pack_index(const unsigned char *sha1)
1774 if (stat(sha1_pack_index_name(sha1), &st))
1779 int has_pack_file(const unsigned char *sha1)
1782 if (stat(sha1_pack_name(sha1), &st))
1787 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1789 struct pack_entry e;
1790 return find_pack_entry(sha1, &e, ignore_packed);
1793 int has_sha1_file(const unsigned char *sha1)
1796 struct pack_entry e;
1798 if (find_pack_entry(sha1, &e, NULL))
1800 return find_sha1_file(sha1, &st) ? 1 : 0;
1804 * reads from fd as long as possible into a supplied buffer of size bytes.
1805 * If necessary the buffer's size is increased using realloc()
1807 * returns 0 if anything went fine and -1 otherwise
1809 * NOTE: both buf and size may change, but even when -1 is returned
1810 * you still have to free() it yourself.
1812 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1814 char* buf = *return_buf;
1815 unsigned long size = *return_size;
1817 unsigned long off = 0;
1820 iret = xread(fd, buf + off, size - off);
1825 buf = xrealloc(buf, size);
1838 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1840 unsigned long size = 4096;
1841 char *buf = xmalloc(size);
1844 if (read_pipe(fd, &buf, &size)) {
1852 ret = write_sha1_file(buf, size, type, sha1);
1854 ret = hash_sha1_file(buf, size, type, sha1);
1859 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1861 unsigned long size = st->st_size;
1867 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1869 if (buf == MAP_FAILED)
1875 ret = write_sha1_file(buf, size, type, sha1);
1877 ret = hash_sha1_file(buf, size, type, sha1);
1883 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1888 switch (st->st_mode & S_IFMT) {
1890 fd = open(path, O_RDONLY);
1892 return error("open(\"%s\"): %s", path,
1894 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1895 return error("%s: failed to insert into database",
1899 target = xmalloc(st->st_size+1);
1900 if (readlink(path, target, st->st_size+1) != st->st_size) {
1901 char *errstr = strerror(errno);
1903 return error("readlink(\"%s\"): %s", path,
1907 hash_sha1_file(target, st->st_size, blob_type, sha1);
1908 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1909 return error("%s: failed to insert into database",
1914 return error("%s: unsupported file type", path);