2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
32 const unsigned char null_sha1[20];
34 static unsigned int sha1_file_open_flag = O_NOATIME;
36 signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 int get_sha1_hex(const char *hex, unsigned char *sha1)
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
84 int safe_create_leading_directories(char *path)
93 pos = strchr(pos, '/');
97 if (!stat(path, &st)) {
99 if (!S_ISDIR(st.st_mode)) {
104 else if (mkdir(path, 0777)) {
108 else if (adjust_shared_perm(path)) {
117 char * sha1_to_hex(const unsigned char *sha1)
120 static char hexbuffer[4][50];
121 static const char hex[] = "0123456789abcdef";
122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1)
159 static char *name, *base;
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
169 name = base + len + 1;
171 fill_sha1_path(name, sha1);
175 char *sha1_pack_name(const unsigned char *sha1)
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
200 char *sha1_pack_index_name(const unsigned char *sha1)
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
245 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
253 int entlen = pfxlen + 43;
256 if (*entry != '/' && relative_base) {
257 /* Relative alt-odb */
259 base_len = strlen(relative_base) + 1;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (*entry != '/' && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
295 if (!memcmp(ent->base, objdir, pfxlen)) {
300 /* add the alternate entry */
302 alt_odb_tail = &(ent->next);
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
319 error("%s: ignoring alternate object stores, nesting too deep.",
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
333 while (cp < ep && *cp != sep)
336 if ((*last != '/') && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
350 static void read_info_alternates(const char * relative_base, int depth)
358 sprintf(path, "%s/info/alternates", relative_base);
359 fd = open(path, O_RDONLY);
362 if (fstat(fd, &st) || (st.st_size == 0)) {
366 mapsz = xsize_t(st.st_size);
367 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
370 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
375 void prepare_alt_odb(void)
379 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
384 alt_odb_tail = &alt_odb_list;
385 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
387 read_info_alternates(get_object_directory(), 0);
390 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
392 char *name = sha1_file_name(sha1);
393 struct alternate_object_database *alt;
398 for (alt = alt_odb_list; alt; alt = alt->next) {
400 fill_sha1_path(name, sha1);
401 if (!stat(alt->base, st))
407 static unsigned int pack_used_ctr;
408 static unsigned int pack_mmap_calls;
409 static unsigned int peak_pack_open_windows;
410 static unsigned int pack_open_windows;
411 static size_t peak_pack_mapped;
412 static size_t pack_mapped;
413 struct packed_git *packed_git;
418 "pack_report: getpagesize() = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
420 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
421 (size_t) getpagesize(),
422 packed_git_window_size,
425 "pack_report: pack_used_ctr = %10u\n"
426 "pack_report: pack_mmap_calls = %10u\n"
427 "pack_report: pack_open_windows = %10u / %10u\n"
428 "pack_report: pack_mapped = "
429 "%10" SZ_FMT " / %10" SZ_FMT "\n",
432 pack_open_windows, peak_pack_open_windows,
433 pack_mapped, peak_pack_mapped);
436 static int check_packed_git_idx(const char *path, struct packed_git *p)
439 struct pack_idx_header *hdr;
441 uint32_t version, nr, i, *index;
442 int fd = open(path, O_RDONLY);
447 if (fstat(fd, &st)) {
451 idx_size = xsize_t(st.st_size);
452 if (idx_size < 4 * 256 + 20 + 20) {
454 return error("index file %s is too small", path);
456 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
460 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
461 version = ntohl(hdr->idx_version);
462 if (version < 2 || version > 2) {
463 munmap(idx_map, idx_size);
464 return error("index file %s is version %d"
465 " and is not supported by this binary"
466 " (try upgrading GIT to a newer version)",
475 index += 2; /* skip index header */
476 for (i = 0; i < 256; i++) {
477 uint32_t n = ntohl(index[i]);
479 munmap(idx_map, idx_size);
480 return error("non-monotonic index %s", path);
488 * - 256 index entries 4 bytes each
489 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
490 * - 20-byte SHA1 of the packfile
491 * - 20-byte SHA1 file checksum
493 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
494 munmap(idx_map, idx_size);
495 return error("wrong index file size in %s", path);
497 } else if (version == 2) {
500 * - 8 bytes of header
501 * - 256 index entries 4 bytes each
502 * - 20-byte sha1 entry * nr
503 * - 4-byte crc entry * nr
504 * - 4-byte offset entry * nr
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
507 * And after the 4-byte offset table might be a
508 * variable sized table containing 8-byte entries
509 * for offsets larger than 2^31.
511 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
512 if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
513 munmap(idx_map, idx_size);
514 return error("wrong index file size in %s", path);
516 if (idx_size != min_size) {
517 /* make sure we can deal with large pack offsets */
518 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
519 if (x > (x + 1) || y > (y + 1)) {
520 munmap(idx_map, idx_size);
521 return error("pack too large for current definition of off_t in %s", path);
526 p->index_version = version;
527 p->index_data = idx_map;
528 p->index_size = idx_size;
533 static void scan_windows(struct packed_git *p,
534 struct packed_git **lru_p,
535 struct pack_window **lru_w,
536 struct pack_window **lru_l)
538 struct pack_window *w, *w_l;
540 for (w_l = NULL, w = p->windows; w; w = w->next) {
542 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
552 static int unuse_one_window(struct packed_git *current, int keep_fd)
554 struct packed_git *p, *lru_p = NULL;
555 struct pack_window *lru_w = NULL, *lru_l = NULL;
558 scan_windows(current, &lru_p, &lru_w, &lru_l);
559 for (p = packed_git; p; p = p->next)
560 scan_windows(p, &lru_p, &lru_w, &lru_l);
562 munmap(lru_w->base, lru_w->len);
563 pack_mapped -= lru_w->len;
565 lru_l->next = lru_w->next;
567 lru_p->windows = lru_w->next;
568 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
569 close(lru_p->pack_fd);
580 void release_pack_memory(size_t need, int fd)
582 size_t cur = pack_mapped;
583 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
587 void unuse_pack(struct pack_window **w_cursor)
589 struct pack_window *w = *w_cursor;
597 * Do not call this directly as this leaks p->pack_fd on error return;
598 * call open_packed_git() instead.
600 static int open_packed_git_1(struct packed_git *p)
603 struct pack_header hdr;
604 unsigned char sha1[20];
605 unsigned char *idx_sha1;
608 p->pack_fd = open(p->pack_name, O_RDONLY);
609 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
612 /* If we created the struct before we had the pack we lack size. */
614 if (!S_ISREG(st.st_mode))
615 return error("packfile %s not a regular file", p->pack_name);
616 p->pack_size = st.st_size;
617 } else if (p->pack_size != st.st_size)
618 return error("packfile %s size changed", p->pack_name);
620 /* We leave these file descriptors open with sliding mmap;
621 * there is no point keeping them open across exec(), though.
623 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
625 return error("cannot determine file descriptor flags");
626 fd_flag |= FD_CLOEXEC;
627 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
628 return error("cannot set FD_CLOEXEC");
630 /* Verify we recognize this pack file format. */
631 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
632 return error("file %s is far too short to be a packfile", p->pack_name);
633 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
634 return error("file %s is not a GIT packfile", p->pack_name);
635 if (!pack_version_ok(hdr.hdr_version))
636 return error("packfile %s is version %u and not supported"
637 " (try upgrading GIT to a newer version)",
638 p->pack_name, ntohl(hdr.hdr_version));
640 /* Verify the pack matches its index. */
641 if (p->num_objects != ntohl(hdr.hdr_entries))
642 return error("packfile %s claims to have %u objects"
643 " while index indicates %u objects",
644 p->pack_name, ntohl(hdr.hdr_entries),
646 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
647 return error("end of packfile %s is unavailable", p->pack_name);
648 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
649 return error("packfile %s signature is unavailable", p->pack_name);
650 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
651 if (hashcmp(sha1, idx_sha1))
652 return error("packfile %s does not match index", p->pack_name);
656 static int open_packed_git(struct packed_git *p)
658 if (!open_packed_git_1(p))
660 if (p->pack_fd != -1) {
667 static int in_window(struct pack_window *win, off_t offset)
669 /* We must promise at least 20 bytes (one hash) after the
670 * offset is available from this window, otherwise the offset
671 * is not actually in this window and a different window (which
672 * has that one hash excess) must be used. This is to support
673 * the object header and delta base parsing routines below.
675 off_t win_off = win->offset;
676 return win_off <= offset
677 && (offset + 20) <= (win_off + win->len);
680 unsigned char* use_pack(struct packed_git *p,
681 struct pack_window **w_cursor,
685 struct pack_window *win = *w_cursor;
687 if (p->pack_fd == -1 && open_packed_git(p))
688 die("packfile %s cannot be accessed", p->pack_name);
690 /* Since packfiles end in a hash of their content and its
691 * pointless to ask for an offset into the middle of that
692 * hash, and the in_window function above wouldn't match
693 * don't allow an offset too close to the end of the file.
695 if (offset > (p->pack_size - 20))
696 die("offset beyond end of packfile (truncated pack?)");
698 if (!win || !in_window(win, offset)) {
701 for (win = p->windows; win; win = win->next) {
702 if (in_window(win, offset))
706 size_t window_align = packed_git_window_size / 2;
708 win = xcalloc(1, sizeof(*win));
709 win->offset = (offset / window_align) * window_align;
710 len = p->pack_size - win->offset;
711 if (len > packed_git_window_size)
712 len = packed_git_window_size;
713 win->len = (size_t)len;
714 pack_mapped += win->len;
715 while (packed_git_limit < pack_mapped
716 && unuse_one_window(p, p->pack_fd))
718 win->base = xmmap(NULL, win->len,
719 PROT_READ, MAP_PRIVATE,
720 p->pack_fd, win->offset);
721 if (win->base == MAP_FAILED)
722 die("packfile %s cannot be mapped: %s",
727 if (pack_mapped > peak_pack_mapped)
728 peak_pack_mapped = pack_mapped;
729 if (pack_open_windows > peak_pack_open_windows)
730 peak_pack_open_windows = pack_open_windows;
731 win->next = p->windows;
735 if (win != *w_cursor) {
736 win->last_used = pack_used_ctr++;
740 offset -= win->offset;
742 *left = win->len - xsize_t(offset);
743 return win->base + offset;
746 struct packed_git *add_packed_git(const char *path, int path_len, int local)
749 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
752 * Make sure a corresponding .pack file exists and that
753 * the index looks sane.
755 path_len -= strlen(".idx");
758 memcpy(p->pack_name, path, path_len);
759 strcpy(p->pack_name + path_len, ".pack");
760 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
761 check_packed_git_idx(path, p)) {
766 /* ok, it looks sane as far as we can check without
767 * actually mapping the pack file.
769 p->pack_size = st.st_size;
773 p->pack_local = local;
774 p->mtime = st.st_mtime;
775 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
780 struct packed_git *parse_pack_index(unsigned char *sha1)
782 char *path = sha1_pack_index_name(sha1);
783 return parse_pack_index_file(sha1, path);
786 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
787 const char *idx_path)
789 const char *path = sha1_pack_name(sha1);
790 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
792 if (check_packed_git_idx(idx_path, p)) {
797 strcpy(p->pack_name, path);
802 hashcpy(p->sha1, sha1);
806 void install_packed_git(struct packed_git *pack)
808 pack->next = packed_git;
812 static void prepare_packed_git_one(char *objdir, int local)
819 sprintf(path, "%s/pack", objdir);
824 error("unable to open object pack directory: %s: %s",
825 path, strerror(errno));
829 while ((de = readdir(dir)) != NULL) {
830 int namelen = strlen(de->d_name);
831 struct packed_git *p;
833 if (!has_extension(de->d_name, ".idx"))
836 /* Don't reopen a pack we already have. */
837 strcpy(path + len, de->d_name);
838 for (p = packed_git; p; p = p->next) {
839 if (!memcmp(path, p->pack_name, len + namelen - 4))
844 /* See if it really is a valid .idx file with corresponding
845 * .pack file that we can map.
847 p = add_packed_git(path, len + namelen, local);
850 install_packed_git(p);
855 static int sort_pack(const void *a_, const void *b_)
857 struct packed_git *a = *((struct packed_git **)a_);
858 struct packed_git *b = *((struct packed_git **)b_);
862 * Local packs tend to contain objects specific to our
863 * variant of the project than remote ones. In addition,
864 * remote ones could be on a network mounted filesystem.
865 * Favor local ones for these reasons.
867 st = a->pack_local - b->pack_local;
872 * Younger packs tend to contain more recent objects,
873 * and more recent objects tend to get accessed more
876 if (a->mtime < b->mtime)
878 else if (a->mtime == b->mtime)
883 static void rearrange_packed_git(void)
885 struct packed_git **ary, *p;
888 for (n = 0, p = packed_git; p; p = p->next)
893 /* prepare an array of packed_git for easier sorting */
894 ary = xcalloc(n, sizeof(struct packed_git *));
895 for (n = 0, p = packed_git; p; p = p->next)
898 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
900 /* link them back again */
901 for (i = 0; i < n - 1; i++)
902 ary[i]->next = ary[i + 1];
903 ary[n - 1]->next = NULL;
909 static int prepare_packed_git_run_once = 0;
910 void prepare_packed_git(void)
912 struct alternate_object_database *alt;
914 if (prepare_packed_git_run_once)
916 prepare_packed_git_one(get_object_directory(), 1);
918 for (alt = alt_odb_list; alt; alt = alt->next) {
920 prepare_packed_git_one(alt->base, 0);
923 rearrange_packed_git();
924 prepare_packed_git_run_once = 1;
927 void reprepare_packed_git(void)
929 prepare_packed_git_run_once = 0;
930 prepare_packed_git();
933 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
935 unsigned char real_sha1[20];
936 hash_sha1_file(map, size, type, real_sha1);
937 return hashcmp(sha1, real_sha1) ? -1 : 0;
940 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
945 char *filename = find_sha1_file(sha1, &st);
951 fd = open(filename, O_RDONLY | sha1_file_open_flag);
953 /* See if it works without O_NOATIME */
954 switch (sha1_file_open_flag) {
956 fd = open(filename, O_RDONLY);
964 /* If it failed once, it will probably fail again.
965 * Stop using O_NOATIME
967 sha1_file_open_flag = 0;
969 *size = xsize_t(st.st_size);
970 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
975 static int legacy_loose_object(unsigned char *map)
980 * Is it a zlib-compressed buffer? If so, the first byte
981 * must be 0x78 (15-bit window size, deflated), and the
982 * first 16-bit word is evenly divisible by 31
984 word = (map[0] << 8) + map[1];
985 if (map[0] == 0x78 && !(word % 31))
991 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
996 unsigned long used = 0;
999 *type = (c >> 4) & 7;
1005 if (sizeof(long) * 8 <= shift)
1008 size += (c & 0x7f) << shift;
1015 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1017 unsigned long size, used;
1018 static const char valid_loose_object_type[8] = {
1020 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1021 0, /* "delta" and others are invalid in a loose object */
1023 enum object_type type;
1025 /* Get the data stream */
1026 memset(stream, 0, sizeof(*stream));
1027 stream->next_in = map;
1028 stream->avail_in = mapsize;
1029 stream->next_out = buffer;
1030 stream->avail_out = bufsiz;
1032 if (legacy_loose_object(map)) {
1033 inflateInit(stream);
1034 return inflate(stream, 0);
1039 * There used to be a second loose object header format which
1040 * was meant to mimic the in-pack format, allowing for direct
1041 * copy of the object data. This format turned up not to be
1042 * really worth it and we don't write it any longer. But we
1043 * can still read it.
1045 used = unpack_object_header_gently(map, mapsize, &type, &size);
1046 if (!used || !valid_loose_object_type[type])
1051 /* Set up the stream for the rest.. */
1052 stream->next_in = map;
1053 stream->avail_in = mapsize;
1054 inflateInit(stream);
1056 /* And generate the fake traditional header */
1057 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1058 typename(type), size);
1062 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1064 int bytes = strlen(buffer) + 1;
1065 unsigned char *buf = xmalloc(1+size);
1069 n = stream->total_out - bytes;
1072 memcpy(buf, (char *) buffer + bytes, n);
1074 if (bytes <= size) {
1076 * The above condition must be (bytes <= size), not
1077 * (bytes < size). In other words, even though we
1078 * expect no more output and set avail_out to zer0,
1079 * the input zlib stream may have bytes that express
1080 * "this concludes the stream", and we *do* want to
1083 * Otherwise we would not be able to test that we
1084 * consumed all the input to reach the expected size;
1085 * we also want to check that zlib tells us that all
1086 * went well with status == Z_STREAM_END at the end.
1088 stream->next_out = buf + bytes;
1089 stream->avail_out = size - bytes;
1090 while (status == Z_OK)
1091 status = inflate(stream, Z_FINISH);
1094 if (status == Z_STREAM_END && !stream->avail_in) {
1100 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1101 else if (stream->avail_in)
1102 error("garbage at end of loose object '%s'",
1109 * We used to just use "sscanf()", but that's actually way
1110 * too permissive for what we want to check. So do an anal
1111 * object header parse by hand.
1113 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1120 * The type can be at most ten bytes (including the
1121 * terminating '\0' that we add), and is followed by
1130 if (i >= sizeof(type))
1136 * The length must follow immediately, and be in canonical
1137 * decimal format (ie "010" is not valid).
1139 size = *hdr++ - '0';
1144 unsigned long c = *hdr - '0';
1148 size = size * 10 + c;
1154 * The length must be followed by a zero byte
1156 return *hdr ? -1 : type_from_string(type);
1159 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1165 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1166 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1169 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1172 unsigned long get_size_from_delta(struct packed_git *p,
1173 struct pack_window **w_curs,
1176 const unsigned char *data;
1177 unsigned char delta_head[20], *in;
1181 memset(&stream, 0, sizeof(stream));
1182 stream.next_out = delta_head;
1183 stream.avail_out = sizeof(delta_head);
1185 inflateInit(&stream);
1187 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1188 stream.next_in = in;
1189 st = inflate(&stream, Z_FINISH);
1190 curpos += stream.next_in - in;
1191 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1192 stream.total_out < sizeof(delta_head));
1193 inflateEnd(&stream);
1194 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1195 die("delta data unpack-initial failed");
1197 /* Examine the initial part of the delta to figure out
1202 /* ignore base size */
1203 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1205 /* Read the result size */
1206 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1209 static off_t get_delta_base(struct packed_git *p,
1210 struct pack_window **w_curs,
1212 enum object_type type,
1213 off_t delta_obj_offset)
1215 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1218 /* use_pack() assured us we have [base_info, base_info + 20)
1219 * as a range that we can look at without walking off the
1220 * end of the mapped window. Its actually the hash size
1221 * that is assured. An OFS_DELTA longer than the hash size
1222 * is stupid, as then a REF_DELTA would be smaller to store.
1224 if (type == OBJ_OFS_DELTA) {
1226 unsigned char c = base_info[used++];
1227 base_offset = c & 127;
1230 if (!base_offset || MSB(base_offset, 7))
1231 die("offset value overflow for delta base object");
1232 c = base_info[used++];
1233 base_offset = (base_offset << 7) + (c & 127);
1235 base_offset = delta_obj_offset - base_offset;
1236 if (base_offset >= delta_obj_offset)
1237 die("delta base offset out of bound");
1239 } else if (type == OBJ_REF_DELTA) {
1240 /* The base entry _must_ be in the same pack */
1241 base_offset = find_pack_entry_one(base_info, p);
1243 die("failed to find delta-pack base object %s",
1244 sha1_to_hex(base_info));
1247 die("I am totally screwed");
1251 /* forward declaration for a mutually recursive function */
1252 static int packed_object_info(struct packed_git *p, off_t offset,
1253 unsigned long *sizep);
1255 static int packed_delta_info(struct packed_git *p,
1256 struct pack_window **w_curs,
1258 enum object_type type,
1260 unsigned long *sizep)
1264 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1265 type = packed_object_info(p, base_offset, NULL);
1267 /* We choose to only get the type of the base object and
1268 * ignore potentially corrupt pack file that expects the delta
1269 * based on a base with a wrong size. This saves tons of
1273 *sizep = get_size_from_delta(p, w_curs, curpos);
1278 static int unpack_object_header(struct packed_git *p,
1279 struct pack_window **w_curs,
1281 unsigned long *sizep)
1283 unsigned char *base;
1286 enum object_type type;
1288 /* use_pack() assures us we have [base, base + 20) available
1289 * as a range that we can look at at. (Its actually the hash
1290 * size that is assured.) With our object header encoding
1291 * the maximum deflated object size is 2^137, which is just
1292 * insane, so we know won't exceed what we have been given.
1294 base = use_pack(p, w_curs, *curpos, &left);
1295 used = unpack_object_header_gently(base, left, &type, sizep);
1297 die("object offset outside of pack file");
1303 const char *packed_object_info_detail(struct packed_git *p,
1305 unsigned long *size,
1306 unsigned long *store_size,
1307 unsigned int *delta_chain_length,
1308 unsigned char *base_sha1)
1310 struct pack_window *w_curs = NULL;
1312 unsigned long dummy;
1313 unsigned char *next_sha1;
1314 enum object_type type;
1316 *delta_chain_length = 0;
1317 curpos = obj_offset;
1318 type = unpack_object_header(p, &w_curs, &curpos, size);
1323 die("pack %s contains unknown object type %d",
1324 p->pack_name, type);
1329 *store_size = 0; /* notyet */
1330 unuse_pack(&w_curs);
1331 return typename(type);
1333 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1334 if (*delta_chain_length == 0) {
1335 /* TODO: find base_sha1 as pointed by curpos */
1340 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1341 if (*delta_chain_length == 0)
1342 hashcpy(base_sha1, next_sha1);
1343 obj_offset = find_pack_entry_one(next_sha1, p);
1346 (*delta_chain_length)++;
1347 curpos = obj_offset;
1348 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1352 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1353 unsigned long *sizep)
1355 struct pack_window *w_curs = NULL;
1357 off_t curpos = obj_offset;
1358 enum object_type type;
1360 type = unpack_object_header(p, &w_curs, &curpos, &size);
1365 type = packed_delta_info(p, &w_curs, curpos,
1366 type, obj_offset, sizep);
1376 die("pack %s contains unknown object type %d",
1377 p->pack_name, type);
1379 unuse_pack(&w_curs);
1383 static void *unpack_compressed_entry(struct packed_git *p,
1384 struct pack_window **w_curs,
1390 unsigned char *buffer, *in;
1392 buffer = xmalloc(size + 1);
1394 memset(&stream, 0, sizeof(stream));
1395 stream.next_out = buffer;
1396 stream.avail_out = size;
1398 inflateInit(&stream);
1400 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1401 stream.next_in = in;
1402 st = inflate(&stream, Z_FINISH);
1403 curpos += stream.next_in - in;
1404 } while (st == Z_OK || st == Z_BUF_ERROR);
1405 inflateEnd(&stream);
1406 if ((st != Z_STREAM_END) || stream.total_out != size) {
1414 #define MAX_DELTA_CACHE (256)
1416 static size_t delta_base_cached;
1418 static struct delta_base_cache_lru_list {
1419 struct delta_base_cache_lru_list *prev;
1420 struct delta_base_cache_lru_list *next;
1421 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1423 static struct delta_base_cache_entry {
1424 struct delta_base_cache_lru_list lru;
1426 struct packed_git *p;
1429 enum object_type type;
1430 } delta_base_cache[MAX_DELTA_CACHE];
1432 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1436 hash = (unsigned long)p + (unsigned long)base_offset;
1437 hash += (hash >> 8) + (hash >> 16);
1438 return hash % MAX_DELTA_CACHE;
1441 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1442 unsigned long *base_size, enum object_type *type, int keep_cache)
1445 unsigned long hash = pack_entry_hash(p, base_offset);
1446 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1449 if (ret && ent->p == p && ent->base_offset == base_offset)
1450 goto found_cache_entry;
1451 return unpack_entry(p, base_offset, type, base_size);
1456 ent->lru.next->prev = ent->lru.prev;
1457 ent->lru.prev->next = ent->lru.next;
1458 delta_base_cached -= ent->size;
1461 ret = xmalloc(ent->size + 1);
1462 memcpy(ret, ent->data, ent->size);
1463 ((char *)ret)[ent->size] = 0;
1466 *base_size = ent->size;
1470 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1475 ent->lru.next->prev = ent->lru.prev;
1476 ent->lru.prev->next = ent->lru.next;
1477 delta_base_cached -= ent->size;
1481 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1482 void *base, unsigned long base_size, enum object_type type)
1484 unsigned long hash = pack_entry_hash(p, base_offset);
1485 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1486 struct delta_base_cache_lru_list *lru;
1488 release_delta_base_cache(ent);
1489 delta_base_cached += base_size;
1491 for (lru = delta_base_cache_lru.next;
1492 delta_base_cached > delta_base_cache_limit
1493 && lru != &delta_base_cache_lru;
1495 struct delta_base_cache_entry *f = (void *)lru;
1496 if (f->type == OBJ_BLOB)
1497 release_delta_base_cache(f);
1499 for (lru = delta_base_cache_lru.next;
1500 delta_base_cached > delta_base_cache_limit
1501 && lru != &delta_base_cache_lru;
1503 struct delta_base_cache_entry *f = (void *)lru;
1504 release_delta_base_cache(f);
1508 ent->base_offset = base_offset;
1511 ent->size = base_size;
1512 ent->lru.next = &delta_base_cache_lru;
1513 ent->lru.prev = delta_base_cache_lru.prev;
1514 delta_base_cache_lru.prev->next = &ent->lru;
1515 delta_base_cache_lru.prev = &ent->lru;
1518 static void *unpack_delta_entry(struct packed_git *p,
1519 struct pack_window **w_curs,
1521 unsigned long delta_size,
1523 enum object_type *type,
1524 unsigned long *sizep)
1526 void *delta_data, *result, *base;
1527 unsigned long base_size;
1530 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1531 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1533 die("failed to read delta base object"
1534 " at %"PRIuMAX" from %s",
1535 (uintmax_t)base_offset, p->pack_name);
1537 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1538 result = patch_delta(base, base_size,
1539 delta_data, delta_size,
1542 die("failed to apply delta");
1544 add_delta_base_cache(p, base_offset, base, base_size, *type);
1548 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1549 enum object_type *type, unsigned long *sizep)
1551 struct pack_window *w_curs = NULL;
1552 off_t curpos = obj_offset;
1555 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1559 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1560 obj_offset, type, sizep);
1566 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1569 die("unknown object type %i in %s", *type, p->pack_name);
1571 unuse_pack(&w_curs);
1575 const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1578 const unsigned char *index = p->index_data;
1579 if (n >= p->num_objects)
1582 if (p->index_version == 1) {
1583 return index + 24 * n + 4;
1586 return index + 20 * n;
1590 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1592 const unsigned char *index = p->index_data;
1594 if (p->index_version == 1) {
1595 return ntohl(*((uint32_t *)(index + 24 * n)));
1598 index += 8 + p->num_objects * (20 + 4);
1599 off = ntohl(*((uint32_t *)(index + 4 * n)));
1600 if (!(off & 0x80000000))
1602 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1603 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1604 ntohl(*((uint32_t *)(index + 4)));
1608 off_t find_pack_entry_one(const unsigned char *sha1,
1609 struct packed_git *p)
1611 const uint32_t *level1_ofs = p->index_data;
1612 const unsigned char *index = p->index_data;
1615 if (p->index_version > 1) {
1620 hi = ntohl(level1_ofs[*sha1]);
1621 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1624 unsigned mi = (lo + hi) / 2;
1625 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1626 int cmp = hashcmp(index + x, sha1);
1628 return nth_packed_object_offset(p, mi);
1637 static int matches_pack_name(struct packed_git *p, const char *ig)
1639 const char *last_c, *c;
1641 if (!strcmp(p->pack_name, ig))
1644 for (c = p->pack_name, last_c = c; *c;)
1649 if (!strcmp(last_c, ig))
1655 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1657 struct packed_git *p;
1660 prepare_packed_git();
1662 for (p = packed_git; p; p = p->next) {
1663 if (ignore_packed) {
1665 for (ig = ignore_packed; *ig; ig++)
1666 if (!matches_pack_name(p, *ig))
1671 offset = find_pack_entry_one(sha1, p);
1674 * We are about to tell the caller where they can
1675 * locate the requested object. We better make
1676 * sure the packfile is still here and can be
1677 * accessed before supplying that answer, as
1678 * it may have been deleted since the index
1681 if (p->pack_fd == -1 && open_packed_git(p)) {
1682 error("packfile %s cannot be accessed", p->pack_name);
1687 hashcpy(e->sha1, sha1);
1694 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1695 struct packed_git *packs)
1697 struct packed_git *p;
1699 for (p = packs; p; p = p->next) {
1700 if (find_pack_entry_one(sha1, p))
1707 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1710 unsigned long mapsize, size;
1715 map = map_sha1_file(sha1, &mapsize);
1717 return error("unable to find %s", sha1_to_hex(sha1));
1718 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1719 status = error("unable to unpack %s header",
1721 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1722 status = error("unable to parse %s header", sha1_to_hex(sha1));
1725 inflateEnd(&stream);
1726 munmap(map, mapsize);
1730 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1732 struct pack_entry e;
1734 if (!find_pack_entry(sha1, &e, NULL)) {
1735 reprepare_packed_git();
1736 if (!find_pack_entry(sha1, &e, NULL))
1737 return sha1_loose_object_info(sha1, sizep);
1739 return packed_object_info(e.p, e.offset, sizep);
1742 static void *read_packed_sha1(const unsigned char *sha1,
1743 enum object_type *type, unsigned long *size)
1745 struct pack_entry e;
1747 if (!find_pack_entry(sha1, &e, NULL))
1750 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1754 * This is meant to hold a *small* number of objects that you would
1755 * want read_sha1_file() to be able to return, but yet you do not want
1756 * to write them into the object store (e.g. a browse-only
1759 static struct cached_object {
1760 unsigned char sha1[20];
1761 enum object_type type;
1765 static int cached_object_nr, cached_object_alloc;
1767 static struct cached_object *find_cached_object(const unsigned char *sha1)
1770 struct cached_object *co = cached_objects;
1772 for (i = 0; i < cached_object_nr; i++, co++) {
1773 if (!hashcmp(co->sha1, sha1))
1779 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1780 unsigned char *sha1)
1782 struct cached_object *co;
1784 hash_sha1_file(buf, len, typename(type), sha1);
1785 if (has_sha1_file(sha1) || find_cached_object(sha1))
1787 if (cached_object_alloc <= cached_object_nr) {
1788 cached_object_alloc = alloc_nr(cached_object_alloc);
1789 cached_objects = xrealloc(cached_objects,
1790 sizeof(*cached_objects) *
1791 cached_object_alloc);
1793 co = &cached_objects[cached_object_nr++];
1796 co->buf = xmalloc(len);
1797 memcpy(co->buf, buf, len);
1798 hashcpy(co->sha1, sha1);
1802 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1803 unsigned long *size)
1805 unsigned long mapsize;
1807 struct cached_object *co;
1809 co = find_cached_object(sha1);
1811 buf = xmalloc(co->size + 1);
1812 memcpy(buf, co->buf, co->size);
1813 ((char*)buf)[co->size] = 0;
1819 buf = read_packed_sha1(sha1, type, size);
1822 map = map_sha1_file(sha1, &mapsize);
1824 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1825 munmap(map, mapsize);
1828 reprepare_packed_git();
1829 return read_packed_sha1(sha1, type, size);
1832 void *read_object_with_reference(const unsigned char *sha1,
1833 const char *required_type_name,
1834 unsigned long *size,
1835 unsigned char *actual_sha1_return)
1837 enum object_type type, required_type;
1839 unsigned long isize;
1840 unsigned char actual_sha1[20];
1842 required_type = type_from_string(required_type_name);
1843 hashcpy(actual_sha1, sha1);
1845 int ref_length = -1;
1846 const char *ref_type = NULL;
1848 buffer = read_sha1_file(actual_sha1, &type, &isize);
1851 if (type == required_type) {
1853 if (actual_sha1_return)
1854 hashcpy(actual_sha1_return, actual_sha1);
1857 /* Handle references */
1858 else if (type == OBJ_COMMIT)
1860 else if (type == OBJ_TAG)
1861 ref_type = "object ";
1866 ref_length = strlen(ref_type);
1868 if (memcmp(buffer, ref_type, ref_length) ||
1869 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1874 /* Now we have the ID of the referred-to object in
1875 * actual_sha1. Check again. */
1879 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1880 const char *type, unsigned char *sha1,
1881 char *hdr, int *hdrlen)
1885 /* Generate the header */
1886 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1890 SHA1_Update(&c, hdr, *hdrlen);
1891 SHA1_Update(&c, buf, len);
1892 SHA1_Final(sha1, &c);
1896 * Link the tempfile to the final place, possibly creating the
1897 * last directory level as you do so.
1899 * Returns the errno on failure, 0 on success.
1901 static int link_temp_to_file(const char *tmpfile, const char *filename)
1906 if (!link(tmpfile, filename))
1910 * Try to mkdir the last path component if that failed.
1912 * Re-try the "link()" regardless of whether the mkdir
1913 * succeeds, since a race might mean that somebody
1917 dir = strrchr(filename, '/');
1920 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1925 if (!link(tmpfile, filename))
1933 * Move the just written object into its final resting place
1935 int move_temp_to_file(const char *tmpfile, const char *filename)
1937 int ret = link_temp_to_file(tmpfile, filename);
1940 * Coda hack - coda doesn't like cross-directory links,
1941 * so we fall back to a rename, which will mean that it
1942 * won't be able to check collisions, but that's not a
1945 * The same holds for FAT formatted media.
1947 * When this succeeds, we just return 0. We have nothing
1950 if (ret && ret != EEXIST) {
1951 if (!rename(tmpfile, filename))
1957 if (ret != EEXIST) {
1958 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1960 /* FIXME!!! Collision check here ? */
1966 static int write_buffer(int fd, const void *buf, size_t len)
1968 if (write_in_full(fd, buf, len) < 0)
1969 return error("file write error (%s)", strerror(errno));
1973 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1974 unsigned char *sha1)
1978 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1982 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1985 unsigned char *compressed;
1987 unsigned char sha1[20];
1989 static char tmpfile[PATH_MAX];
1993 /* Normally if we have it in the pack then we do not bother writing
1994 * it out into .git/objects/??/?{38} file.
1996 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1997 filename = sha1_file_name(sha1);
1999 hashcpy(returnsha1, sha1);
2000 if (has_sha1_file(sha1))
2002 fd = open(filename, O_RDONLY);
2005 * FIXME!!! We might do collision checking here, but we'd
2006 * need to uncompress the old file and check it. Later.
2012 if (errno != ENOENT) {
2013 return error("sha1 file %s: %s\n", filename, strerror(errno));
2016 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2018 fd = mkstemp(tmpfile);
2021 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2023 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2027 memset(&stream, 0, sizeof(stream));
2028 deflateInit(&stream, zlib_compression_level);
2029 size = 8 + deflateBound(&stream, len+hdrlen);
2030 compressed = xmalloc(size);
2033 stream.next_out = compressed;
2034 stream.avail_out = size;
2036 /* First header.. */
2037 stream.next_in = (unsigned char *)hdr;
2038 stream.avail_in = hdrlen;
2039 while (deflate(&stream, 0) == Z_OK)
2042 /* Then the data itself.. */
2043 stream.next_in = buf;
2044 stream.avail_in = len;
2045 ret = deflate(&stream, Z_FINISH);
2046 if (ret != Z_STREAM_END)
2047 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2049 ret = deflateEnd(&stream);
2051 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2053 size = stream.total_out;
2055 if (write_buffer(fd, compressed, size) < 0)
2056 die("unable to write sha1 file");
2059 die("unable to write sha1 file");
2062 return move_temp_to_file(tmpfile, filename);
2066 * We need to unpack and recompress the object for writing
2067 * it out to a different file.
2069 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2073 unsigned char *unpacked;
2075 enum object_type type;
2080 /* need to unpack and recompress it by itself */
2081 unpacked = read_packed_sha1(sha1, &type, &len);
2083 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2085 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2088 memset(&stream, 0, sizeof(stream));
2089 deflateInit(&stream, zlib_compression_level);
2090 size = deflateBound(&stream, len + hdrlen);
2091 buf = xmalloc(size);
2094 stream.next_out = buf;
2095 stream.avail_out = size;
2097 /* First header.. */
2098 stream.next_in = (void *)hdr;
2099 stream.avail_in = hdrlen;
2100 while (deflate(&stream, 0) == Z_OK)
2103 /* Then the data itself.. */
2104 stream.next_in = unpacked;
2105 stream.avail_in = len;
2106 while (deflate(&stream, Z_FINISH) == Z_OK)
2108 deflateEnd(&stream);
2111 *objsize = stream.total_out;
2115 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2118 unsigned long objsize;
2119 void *buf = map_sha1_file(sha1, &objsize);
2122 retval = write_buffer(fd, buf, objsize);
2123 munmap(buf, objsize);
2127 buf = repack_object(sha1, &objsize);
2128 retval = write_buffer(fd, buf, objsize);
2133 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2134 size_t bufsize, size_t *bufposn)
2136 char tmpfile[PATH_MAX];
2139 unsigned char real_sha1[20];
2140 unsigned char discard[4096];
2144 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2146 local = mkstemp(tmpfile);
2149 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2151 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2154 memset(&stream, 0, sizeof(stream));
2156 inflateInit(&stream);
2163 stream.avail_in = *bufposn;
2164 stream.next_in = (unsigned char *) buffer;
2166 stream.next_out = discard;
2167 stream.avail_out = sizeof(discard);
2168 ret = inflate(&stream, Z_SYNC_FLUSH);
2169 SHA1_Update(&c, discard, sizeof(discard) -
2171 } while (stream.avail_in && ret == Z_OK);
2172 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2173 die("unable to write sha1 file");
2174 memmove(buffer, buffer + *bufposn - stream.avail_in,
2176 *bufposn = stream.avail_in;
2180 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2185 return error("Connection closed?");
2186 perror("Reading from connection");
2191 inflateEnd(&stream);
2193 fchmod(local, 0444);
2194 if (close(local) != 0)
2195 die("unable to write sha1 file");
2196 SHA1_Final(real_sha1, &c);
2197 if (ret != Z_STREAM_END) {
2199 return error("File %s corrupted", sha1_to_hex(sha1));
2201 if (hashcmp(sha1, real_sha1)) {
2203 return error("File %s has bad hash", sha1_to_hex(sha1));
2206 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2209 int has_pack_index(const unsigned char *sha1)
2212 if (stat(sha1_pack_index_name(sha1), &st))
2217 int has_pack_file(const unsigned char *sha1)
2220 if (stat(sha1_pack_name(sha1), &st))
2225 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2227 struct pack_entry e;
2228 return find_pack_entry(sha1, &e, ignore_packed);
2231 int has_sha1_file(const unsigned char *sha1)
2234 struct pack_entry e;
2236 if (find_pack_entry(sha1, &e, NULL))
2238 return find_sha1_file(sha1, &st) ? 1 : 0;
2242 * reads from fd as long as possible into a supplied buffer of size bytes.
2243 * If necessary the buffer's size is increased using realloc()
2245 * returns 0 if anything went fine and -1 otherwise
2247 * NOTE: both buf and size may change, but even when -1 is returned
2248 * you still have to free() it yourself.
2250 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2252 char* buf = *return_buf;
2253 unsigned long size = *return_size;
2255 unsigned long off = 0;
2258 iret = xread(fd, buf + off, size - off);
2263 buf = xrealloc(buf, size);
2276 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2278 unsigned long size = 4096;
2279 char *buf = xmalloc(size);
2282 if (read_pipe(fd, &buf, &size)) {
2290 ret = write_sha1_file(buf, size, type, sha1);
2292 ret = hash_sha1_file(buf, size, type, sha1);
2297 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2298 enum object_type type, const char *path)
2300 size_t size = xsize_t(st->st_size);
2302 int ret, re_allocated = 0;
2305 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2312 * Convert blobs to git internal format
2314 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2315 unsigned long nsize = size;
2316 char *nbuf = convert_to_git(path, buf, &nsize);
2326 ret = write_sha1_file(buf, size, typename(type), sha1);
2328 ret = hash_sha1_file(buf, size, typename(type), sha1);
2338 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2344 switch (st->st_mode & S_IFMT) {
2346 fd = open(path, O_RDONLY);
2348 return error("open(\"%s\"): %s", path,
2350 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2351 return error("%s: failed to insert into database",
2355 len = xsize_t(st->st_size);
2356 target = xmalloc(len + 1);
2357 if (readlink(path, target, len + 1) != st->st_size) {
2358 char *errstr = strerror(errno);
2360 return error("readlink(\"%s\"): %s", path,
2364 hash_sha1_file(target, len, blob_type, sha1);
2365 else if (write_sha1_file(target, len, blob_type, sha1))
2366 return error("%s: failed to insert into database",
2371 return resolve_gitlink_ref(path, "HEAD", sha1);
2373 return error("%s: unsupported file type", path);
2378 int read_pack_header(int fd, struct pack_header *header)
2380 char *c = (char*)header;
2381 ssize_t remaining = sizeof(struct pack_header);
2383 ssize_t r = xread(fd, c, remaining);
2385 /* "eof before pack header was fully read" */
2386 return PH_ERROR_EOF;
2389 } while (remaining > 0);
2390 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2391 /* "protocol error (pack signature mismatch detected)" */
2392 return PH_ERROR_PACK_SIGNATURE;
2393 if (!pack_version_ok(header->hdr_version))
2394 /* "protocol error (pack version unsupported)" */
2395 return PH_ERROR_PROTOCOL;