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 const 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)
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = open(path, O_RDONLY);
366 if (fstat(fd, &st) || (st.st_size == 0)) {
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
379 void prepare_alt_odb(void)
383 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
388 alt_odb_tail = &alt_odb_list;
389 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
391 read_info_alternates(get_object_directory(), 0);
394 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
396 char *name = sha1_file_name(sha1);
397 struct alternate_object_database *alt;
402 for (alt = alt_odb_list; alt; alt = alt->next) {
404 fill_sha1_path(name, sha1);
405 if (!stat(alt->base, st))
411 static unsigned int pack_used_ctr;
412 static unsigned int pack_mmap_calls;
413 static unsigned int peak_pack_open_windows;
414 static unsigned int pack_open_windows;
415 static size_t peak_pack_mapped;
416 static size_t pack_mapped;
417 struct packed_git *packed_git;
422 "pack_report: getpagesize() = %10" SZ_FMT "\n"
423 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
424 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
425 (size_t) getpagesize(),
426 packed_git_window_size,
429 "pack_report: pack_used_ctr = %10u\n"
430 "pack_report: pack_mmap_calls = %10u\n"
431 "pack_report: pack_open_windows = %10u / %10u\n"
432 "pack_report: pack_mapped = "
433 "%10" SZ_FMT " / %10" SZ_FMT "\n",
436 pack_open_windows, peak_pack_open_windows,
437 pack_mapped, peak_pack_mapped);
440 static int check_packed_git_idx(const char *path, struct packed_git *p)
443 struct pack_idx_header *hdr;
445 uint32_t version, nr, i, *index;
446 int fd = open(path, O_RDONLY);
451 if (fstat(fd, &st)) {
455 idx_size = xsize_t(st.st_size);
456 if (idx_size < 4 * 256 + 20 + 20) {
458 return error("index file %s is too small", path);
460 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
464 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
465 version = ntohl(hdr->idx_version);
466 if (version < 2 || version > 2) {
467 munmap(idx_map, idx_size);
468 return error("index file %s is version %d"
469 " and is not supported by this binary"
470 " (try upgrading GIT to a newer version)",
479 index += 2; /* skip index header */
480 for (i = 0; i < 256; i++) {
481 uint32_t n = ntohl(index[i]);
483 munmap(idx_map, idx_size);
484 return error("non-monotonic index %s", path);
492 * - 256 index entries 4 bytes each
493 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
494 * - 20-byte SHA1 of the packfile
495 * - 20-byte SHA1 file checksum
497 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
498 munmap(idx_map, idx_size);
499 return error("wrong index file size in %s", path);
501 } else if (version == 2) {
504 * - 8 bytes of header
505 * - 256 index entries 4 bytes each
506 * - 20-byte sha1 entry * nr
507 * - 4-byte crc entry * nr
508 * - 4-byte offset entry * nr
509 * - 20-byte SHA1 of the packfile
510 * - 20-byte SHA1 file checksum
511 * And after the 4-byte offset table might be a
512 * variable sized table containing 8-byte entries
513 * for offsets larger than 2^31.
515 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
516 unsigned long max_size = min_size;
518 max_size += (nr - 1)*8;
519 if (idx_size < min_size || idx_size > max_size) {
520 munmap(idx_map, idx_size);
521 return error("wrong index file size in %s", path);
523 if (idx_size != min_size) {
524 /* make sure we can deal with large pack offsets */
525 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
526 if (x > (x + 1) || y > (y + 1)) {
527 munmap(idx_map, idx_size);
528 return error("pack too large for current definition of off_t in %s", path);
533 p->index_version = version;
534 p->index_data = idx_map;
535 p->index_size = idx_size;
540 static void scan_windows(struct packed_git *p,
541 struct packed_git **lru_p,
542 struct pack_window **lru_w,
543 struct pack_window **lru_l)
545 struct pack_window *w, *w_l;
547 for (w_l = NULL, w = p->windows; w; w = w->next) {
549 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
559 static int unuse_one_window(struct packed_git *current, int keep_fd)
561 struct packed_git *p, *lru_p = NULL;
562 struct pack_window *lru_w = NULL, *lru_l = NULL;
565 scan_windows(current, &lru_p, &lru_w, &lru_l);
566 for (p = packed_git; p; p = p->next)
567 scan_windows(p, &lru_p, &lru_w, &lru_l);
569 munmap(lru_w->base, lru_w->len);
570 pack_mapped -= lru_w->len;
572 lru_l->next = lru_w->next;
574 lru_p->windows = lru_w->next;
575 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
576 close(lru_p->pack_fd);
587 void release_pack_memory(size_t need, int fd)
589 size_t cur = pack_mapped;
590 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
594 void unuse_pack(struct pack_window **w_cursor)
596 struct pack_window *w = *w_cursor;
604 * Do not call this directly as this leaks p->pack_fd on error return;
605 * call open_packed_git() instead.
607 static int open_packed_git_1(struct packed_git *p)
610 struct pack_header hdr;
611 unsigned char sha1[20];
612 unsigned char *idx_sha1;
615 p->pack_fd = open(p->pack_name, O_RDONLY);
616 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
619 /* If we created the struct before we had the pack we lack size. */
621 if (!S_ISREG(st.st_mode))
622 return error("packfile %s not a regular file", p->pack_name);
623 p->pack_size = st.st_size;
624 } else if (p->pack_size != st.st_size)
625 return error("packfile %s size changed", p->pack_name);
627 /* We leave these file descriptors open with sliding mmap;
628 * there is no point keeping them open across exec(), though.
630 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
632 return error("cannot determine file descriptor flags");
633 fd_flag |= FD_CLOEXEC;
634 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
635 return error("cannot set FD_CLOEXEC");
637 /* Verify we recognize this pack file format. */
638 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
639 return error("file %s is far too short to be a packfile", p->pack_name);
640 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
641 return error("file %s is not a GIT packfile", p->pack_name);
642 if (!pack_version_ok(hdr.hdr_version))
643 return error("packfile %s is version %u and not supported"
644 " (try upgrading GIT to a newer version)",
645 p->pack_name, ntohl(hdr.hdr_version));
647 /* Verify the pack matches its index. */
648 if (p->num_objects != ntohl(hdr.hdr_entries))
649 return error("packfile %s claims to have %u objects"
650 " while index indicates %u objects",
651 p->pack_name, ntohl(hdr.hdr_entries),
653 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
654 return error("end of packfile %s is unavailable", p->pack_name);
655 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
656 return error("packfile %s signature is unavailable", p->pack_name);
657 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
658 if (hashcmp(sha1, idx_sha1))
659 return error("packfile %s does not match index", p->pack_name);
663 static int open_packed_git(struct packed_git *p)
665 if (!open_packed_git_1(p))
667 if (p->pack_fd != -1) {
674 static int in_window(struct pack_window *win, off_t offset)
676 /* We must promise at least 20 bytes (one hash) after the
677 * offset is available from this window, otherwise the offset
678 * is not actually in this window and a different window (which
679 * has that one hash excess) must be used. This is to support
680 * the object header and delta base parsing routines below.
682 off_t win_off = win->offset;
683 return win_off <= offset
684 && (offset + 20) <= (win_off + win->len);
687 unsigned char* use_pack(struct packed_git *p,
688 struct pack_window **w_cursor,
692 struct pack_window *win = *w_cursor;
694 if (p->pack_fd == -1 && open_packed_git(p))
695 die("packfile %s cannot be accessed", p->pack_name);
697 /* Since packfiles end in a hash of their content and its
698 * pointless to ask for an offset into the middle of that
699 * hash, and the in_window function above wouldn't match
700 * don't allow an offset too close to the end of the file.
702 if (offset > (p->pack_size - 20))
703 die("offset beyond end of packfile (truncated pack?)");
705 if (!win || !in_window(win, offset)) {
708 for (win = p->windows; win; win = win->next) {
709 if (in_window(win, offset))
713 size_t window_align = packed_git_window_size / 2;
715 win = xcalloc(1, sizeof(*win));
716 win->offset = (offset / window_align) * window_align;
717 len = p->pack_size - win->offset;
718 if (len > packed_git_window_size)
719 len = packed_git_window_size;
720 win->len = (size_t)len;
721 pack_mapped += win->len;
722 while (packed_git_limit < pack_mapped
723 && unuse_one_window(p, p->pack_fd))
725 win->base = xmmap(NULL, win->len,
726 PROT_READ, MAP_PRIVATE,
727 p->pack_fd, win->offset);
728 if (win->base == MAP_FAILED)
729 die("packfile %s cannot be mapped: %s",
734 if (pack_mapped > peak_pack_mapped)
735 peak_pack_mapped = pack_mapped;
736 if (pack_open_windows > peak_pack_open_windows)
737 peak_pack_open_windows = pack_open_windows;
738 win->next = p->windows;
742 if (win != *w_cursor) {
743 win->last_used = pack_used_ctr++;
747 offset -= win->offset;
749 *left = win->len - xsize_t(offset);
750 return win->base + offset;
753 struct packed_git *add_packed_git(const char *path, int path_len, int local)
756 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
759 * Make sure a corresponding .pack file exists and that
760 * the index looks sane.
762 path_len -= strlen(".idx");
765 memcpy(p->pack_name, path, path_len);
766 strcpy(p->pack_name + path_len, ".pack");
767 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
768 check_packed_git_idx(path, p)) {
773 /* ok, it looks sane as far as we can check without
774 * actually mapping the pack file.
776 p->pack_size = st.st_size;
780 p->pack_local = local;
781 p->mtime = st.st_mtime;
782 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
787 struct packed_git *parse_pack_index(unsigned char *sha1)
789 char *path = sha1_pack_index_name(sha1);
790 return parse_pack_index_file(sha1, path);
793 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
794 const char *idx_path)
796 const char *path = sha1_pack_name(sha1);
797 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
799 if (check_packed_git_idx(idx_path, p)) {
804 strcpy(p->pack_name, path);
809 hashcpy(p->sha1, sha1);
813 void install_packed_git(struct packed_git *pack)
815 pack->next = packed_git;
819 static void prepare_packed_git_one(char *objdir, int local)
821 /* Ensure that this buffer is large enough so that we can
822 append "/pack/" without clobbering the stack even if
823 strlen(objdir) were PATH_MAX. */
824 char path[PATH_MAX + 1 + 4 + 1 + 1];
829 sprintf(path, "%s/pack", objdir);
834 error("unable to open object pack directory: %s: %s",
835 path, strerror(errno));
839 while ((de = readdir(dir)) != NULL) {
840 int namelen = strlen(de->d_name);
841 struct packed_git *p;
843 if (!has_extension(de->d_name, ".idx"))
846 if (len + namelen + 1 > sizeof(path))
849 /* Don't reopen a pack we already have. */
850 strcpy(path + len, de->d_name);
851 for (p = packed_git; p; p = p->next) {
852 if (!memcmp(path, p->pack_name, len + namelen - 4))
857 /* See if it really is a valid .idx file with corresponding
858 * .pack file that we can map.
860 p = add_packed_git(path, len + namelen, local);
863 install_packed_git(p);
868 static int sort_pack(const void *a_, const void *b_)
870 struct packed_git *a = *((struct packed_git **)a_);
871 struct packed_git *b = *((struct packed_git **)b_);
875 * Local packs tend to contain objects specific to our
876 * variant of the project than remote ones. In addition,
877 * remote ones could be on a network mounted filesystem.
878 * Favor local ones for these reasons.
880 st = a->pack_local - b->pack_local;
885 * Younger packs tend to contain more recent objects,
886 * and more recent objects tend to get accessed more
889 if (a->mtime < b->mtime)
891 else if (a->mtime == b->mtime)
896 static void rearrange_packed_git(void)
898 struct packed_git **ary, *p;
901 for (n = 0, p = packed_git; p; p = p->next)
906 /* prepare an array of packed_git for easier sorting */
907 ary = xcalloc(n, sizeof(struct packed_git *));
908 for (n = 0, p = packed_git; p; p = p->next)
911 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
913 /* link them back again */
914 for (i = 0; i < n - 1; i++)
915 ary[i]->next = ary[i + 1];
916 ary[n - 1]->next = NULL;
922 static int prepare_packed_git_run_once = 0;
923 void prepare_packed_git(void)
925 struct alternate_object_database *alt;
927 if (prepare_packed_git_run_once)
929 prepare_packed_git_one(get_object_directory(), 1);
931 for (alt = alt_odb_list; alt; alt = alt->next) {
933 prepare_packed_git_one(alt->base, 0);
936 rearrange_packed_git();
937 prepare_packed_git_run_once = 1;
940 void reprepare_packed_git(void)
942 prepare_packed_git_run_once = 0;
943 prepare_packed_git();
946 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
948 unsigned char real_sha1[20];
949 hash_sha1_file(map, size, type, real_sha1);
950 return hashcmp(sha1, real_sha1) ? -1 : 0;
953 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
958 char *filename = find_sha1_file(sha1, &st);
964 fd = open(filename, O_RDONLY | sha1_file_open_flag);
966 /* See if it works without O_NOATIME */
967 switch (sha1_file_open_flag) {
969 fd = open(filename, O_RDONLY);
977 /* If it failed once, it will probably fail again.
978 * Stop using O_NOATIME
980 sha1_file_open_flag = 0;
982 *size = xsize_t(st.st_size);
983 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
988 int legacy_loose_object(unsigned char *map)
993 * Is it a zlib-compressed buffer? If so, the first byte
994 * must be 0x78 (15-bit window size, deflated), and the
995 * first 16-bit word is evenly divisible by 31
997 word = (map[0] << 8) + map[1];
998 if (map[0] == 0x78 && !(word % 31))
1004 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1009 unsigned long used = 0;
1012 *type = (c >> 4) & 7;
1018 if (sizeof(long) * 8 <= shift)
1021 size += (c & 0x7f) << shift;
1028 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1030 unsigned long size, used;
1031 static const char valid_loose_object_type[8] = {
1033 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1034 0, /* "delta" and others are invalid in a loose object */
1036 enum object_type type;
1038 /* Get the data stream */
1039 memset(stream, 0, sizeof(*stream));
1040 stream->next_in = map;
1041 stream->avail_in = mapsize;
1042 stream->next_out = buffer;
1043 stream->avail_out = bufsiz;
1045 if (legacy_loose_object(map)) {
1046 inflateInit(stream);
1047 return inflate(stream, 0);
1050 used = unpack_object_header_gently(map, mapsize, &type, &size);
1051 if (!used || !valid_loose_object_type[type])
1056 /* Set up the stream for the rest.. */
1057 stream->next_in = map;
1058 stream->avail_in = mapsize;
1059 inflateInit(stream);
1061 /* And generate the fake traditional header */
1062 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1063 typename(type), size);
1067 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1069 int bytes = strlen(buffer) + 1;
1070 unsigned char *buf = xmalloc(1+size);
1074 n = stream->total_out - bytes;
1077 memcpy(buf, (char *) buffer + bytes, n);
1079 if (bytes <= size) {
1081 * The above condition must be (bytes <= size), not
1082 * (bytes < size). In other words, even though we
1083 * expect no more output and set avail_out to zer0,
1084 * the input zlib stream may have bytes that express
1085 * "this concludes the stream", and we *do* want to
1088 * Otherwise we would not be able to test that we
1089 * consumed all the input to reach the expected size;
1090 * we also want to check that zlib tells us that all
1091 * went well with status == Z_STREAM_END at the end.
1093 stream->next_out = buf + bytes;
1094 stream->avail_out = size - bytes;
1095 while (status == Z_OK)
1096 status = inflate(stream, Z_FINISH);
1099 if (status == Z_STREAM_END && !stream->avail_in) {
1105 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1106 else if (stream->avail_in)
1107 error("garbage at end of loose object '%s'",
1114 * We used to just use "sscanf()", but that's actually way
1115 * too permissive for what we want to check. So do an anal
1116 * object header parse by hand.
1118 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1125 * The type can be at most ten bytes (including the
1126 * terminating '\0' that we add), and is followed by
1135 if (i >= sizeof(type))
1141 * The length must follow immediately, and be in canonical
1142 * decimal format (ie "010" is not valid).
1144 size = *hdr++ - '0';
1149 unsigned long c = *hdr - '0';
1153 size = size * 10 + c;
1159 * The length must be followed by a zero byte
1161 return *hdr ? -1 : type_from_string(type);
1164 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1170 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1171 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1174 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1177 unsigned long get_size_from_delta(struct packed_git *p,
1178 struct pack_window **w_curs,
1181 const unsigned char *data;
1182 unsigned char delta_head[20], *in;
1186 memset(&stream, 0, sizeof(stream));
1187 stream.next_out = delta_head;
1188 stream.avail_out = sizeof(delta_head);
1190 inflateInit(&stream);
1192 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1193 stream.next_in = in;
1194 st = inflate(&stream, Z_FINISH);
1195 curpos += stream.next_in - in;
1196 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1197 stream.total_out < sizeof(delta_head));
1198 inflateEnd(&stream);
1199 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1200 die("delta data unpack-initial failed");
1202 /* Examine the initial part of the delta to figure out
1207 /* ignore base size */
1208 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1210 /* Read the result size */
1211 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1214 static off_t get_delta_base(struct packed_git *p,
1215 struct pack_window **w_curs,
1217 enum object_type type,
1218 off_t delta_obj_offset)
1220 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1223 /* use_pack() assured us we have [base_info, base_info + 20)
1224 * as a range that we can look at without walking off the
1225 * end of the mapped window. Its actually the hash size
1226 * that is assured. An OFS_DELTA longer than the hash size
1227 * is stupid, as then a REF_DELTA would be smaller to store.
1229 if (type == OBJ_OFS_DELTA) {
1231 unsigned char c = base_info[used++];
1232 base_offset = c & 127;
1235 if (!base_offset || MSB(base_offset, 7))
1236 die("offset value overflow for delta base object");
1237 c = base_info[used++];
1238 base_offset = (base_offset << 7) + (c & 127);
1240 base_offset = delta_obj_offset - base_offset;
1241 if (base_offset >= delta_obj_offset)
1242 die("delta base offset out of bound");
1244 } else if (type == OBJ_REF_DELTA) {
1245 /* The base entry _must_ be in the same pack */
1246 base_offset = find_pack_entry_one(base_info, p);
1248 die("failed to find delta-pack base object %s",
1249 sha1_to_hex(base_info));
1252 die("I am totally screwed");
1256 /* forward declaration for a mutually recursive function */
1257 static int packed_object_info(struct packed_git *p, off_t offset,
1258 unsigned long *sizep);
1260 static int packed_delta_info(struct packed_git *p,
1261 struct pack_window **w_curs,
1263 enum object_type type,
1265 unsigned long *sizep)
1269 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1270 type = packed_object_info(p, base_offset, NULL);
1272 /* We choose to only get the type of the base object and
1273 * ignore potentially corrupt pack file that expects the delta
1274 * based on a base with a wrong size. This saves tons of
1278 *sizep = get_size_from_delta(p, w_curs, curpos);
1283 static int unpack_object_header(struct packed_git *p,
1284 struct pack_window **w_curs,
1286 unsigned long *sizep)
1288 unsigned char *base;
1291 enum object_type type;
1293 /* use_pack() assures us we have [base, base + 20) available
1294 * as a range that we can look at at. (Its actually the hash
1295 * size that is assured.) With our object header encoding
1296 * the maximum deflated object size is 2^137, which is just
1297 * insane, so we know won't exceed what we have been given.
1299 base = use_pack(p, w_curs, *curpos, &left);
1300 used = unpack_object_header_gently(base, left, &type, sizep);
1302 die("object offset outside of pack file");
1308 const char *packed_object_info_detail(struct packed_git *p,
1310 unsigned long *size,
1311 unsigned long *store_size,
1312 unsigned int *delta_chain_length,
1313 unsigned char *base_sha1)
1315 struct pack_window *w_curs = NULL;
1317 unsigned long dummy;
1318 unsigned char *next_sha1;
1319 enum object_type type;
1321 *delta_chain_length = 0;
1322 curpos = obj_offset;
1323 type = unpack_object_header(p, &w_curs, &curpos, size);
1328 die("pack %s contains unknown object type %d",
1329 p->pack_name, type);
1334 *store_size = 0; /* notyet */
1335 unuse_pack(&w_curs);
1336 return typename(type);
1338 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1339 if (*delta_chain_length == 0) {
1340 /* TODO: find base_sha1 as pointed by curpos */
1345 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1346 if (*delta_chain_length == 0)
1347 hashcpy(base_sha1, next_sha1);
1348 obj_offset = find_pack_entry_one(next_sha1, p);
1351 (*delta_chain_length)++;
1352 curpos = obj_offset;
1353 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1357 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1358 unsigned long *sizep)
1360 struct pack_window *w_curs = NULL;
1362 off_t curpos = obj_offset;
1363 enum object_type type;
1365 type = unpack_object_header(p, &w_curs, &curpos, &size);
1370 type = packed_delta_info(p, &w_curs, curpos,
1371 type, obj_offset, sizep);
1381 die("pack %s contains unknown object type %d",
1382 p->pack_name, type);
1384 unuse_pack(&w_curs);
1388 static void *unpack_compressed_entry(struct packed_git *p,
1389 struct pack_window **w_curs,
1395 unsigned char *buffer, *in;
1397 buffer = xmalloc(size + 1);
1399 memset(&stream, 0, sizeof(stream));
1400 stream.next_out = buffer;
1401 stream.avail_out = size;
1403 inflateInit(&stream);
1405 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1406 stream.next_in = in;
1407 st = inflate(&stream, Z_FINISH);
1408 curpos += stream.next_in - in;
1409 } while (st == Z_OK || st == Z_BUF_ERROR);
1410 inflateEnd(&stream);
1411 if ((st != Z_STREAM_END) || stream.total_out != size) {
1419 #define MAX_DELTA_CACHE (256)
1421 static size_t delta_base_cached;
1423 static struct delta_base_cache_lru_list {
1424 struct delta_base_cache_lru_list *prev;
1425 struct delta_base_cache_lru_list *next;
1426 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1428 static struct delta_base_cache_entry {
1429 struct delta_base_cache_lru_list lru;
1431 struct packed_git *p;
1434 enum object_type type;
1435 } delta_base_cache[MAX_DELTA_CACHE];
1437 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1441 hash = (unsigned long)p + (unsigned long)base_offset;
1442 hash += (hash >> 8) + (hash >> 16);
1443 return hash % MAX_DELTA_CACHE;
1446 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1447 unsigned long *base_size, enum object_type *type, int keep_cache)
1450 unsigned long hash = pack_entry_hash(p, base_offset);
1451 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1454 if (ret && ent->p == p && ent->base_offset == base_offset)
1455 goto found_cache_entry;
1456 return unpack_entry(p, base_offset, type, base_size);
1461 ent->lru.next->prev = ent->lru.prev;
1462 ent->lru.prev->next = ent->lru.next;
1463 delta_base_cached -= ent->size;
1466 ret = xmalloc(ent->size + 1);
1467 memcpy(ret, ent->data, ent->size);
1468 ((char *)ret)[ent->size] = 0;
1471 *base_size = ent->size;
1475 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1480 ent->lru.next->prev = ent->lru.prev;
1481 ent->lru.prev->next = ent->lru.next;
1482 delta_base_cached -= ent->size;
1486 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1487 void *base, unsigned long base_size, enum object_type type)
1489 unsigned long hash = pack_entry_hash(p, base_offset);
1490 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1491 struct delta_base_cache_lru_list *lru;
1493 release_delta_base_cache(ent);
1494 delta_base_cached += base_size;
1496 for (lru = delta_base_cache_lru.next;
1497 delta_base_cached > delta_base_cache_limit
1498 && lru != &delta_base_cache_lru;
1500 struct delta_base_cache_entry *f = (void *)lru;
1501 if (f->type == OBJ_BLOB)
1502 release_delta_base_cache(f);
1504 for (lru = delta_base_cache_lru.next;
1505 delta_base_cached > delta_base_cache_limit
1506 && lru != &delta_base_cache_lru;
1508 struct delta_base_cache_entry *f = (void *)lru;
1509 release_delta_base_cache(f);
1513 ent->base_offset = base_offset;
1516 ent->size = base_size;
1517 ent->lru.next = &delta_base_cache_lru;
1518 ent->lru.prev = delta_base_cache_lru.prev;
1519 delta_base_cache_lru.prev->next = &ent->lru;
1520 delta_base_cache_lru.prev = &ent->lru;
1523 static void *unpack_delta_entry(struct packed_git *p,
1524 struct pack_window **w_curs,
1526 unsigned long delta_size,
1528 enum object_type *type,
1529 unsigned long *sizep)
1531 void *delta_data, *result, *base;
1532 unsigned long base_size;
1535 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1536 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1538 die("failed to read delta base object"
1539 " at %"PRIuMAX" from %s",
1540 (uintmax_t)base_offset, p->pack_name);
1542 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1543 result = patch_delta(base, base_size,
1544 delta_data, delta_size,
1547 die("failed to apply delta");
1549 add_delta_base_cache(p, base_offset, base, base_size, *type);
1553 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1554 enum object_type *type, unsigned long *sizep)
1556 struct pack_window *w_curs = NULL;
1557 off_t curpos = obj_offset;
1560 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1564 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1565 obj_offset, type, sizep);
1571 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1574 die("unknown object type %i in %s", *type, p->pack_name);
1576 unuse_pack(&w_curs);
1580 const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1583 const unsigned char *index = p->index_data;
1584 if (n >= p->num_objects)
1587 if (p->index_version == 1) {
1588 return index + 24 * n + 4;
1591 return index + 20 * n;
1595 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1597 const unsigned char *index = p->index_data;
1599 if (p->index_version == 1) {
1600 return ntohl(*((uint32_t *)(index + 24 * n)));
1603 index += 8 + p->num_objects * (20 + 4);
1604 off = ntohl(*((uint32_t *)(index + 4 * n)));
1605 if (!(off & 0x80000000))
1607 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1608 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1609 ntohl(*((uint32_t *)(index + 4)));
1613 off_t find_pack_entry_one(const unsigned char *sha1,
1614 struct packed_git *p)
1616 const uint32_t *level1_ofs = p->index_data;
1617 const unsigned char *index = p->index_data;
1620 if (p->index_version > 1) {
1625 hi = ntohl(level1_ofs[*sha1]);
1626 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1629 unsigned mi = (lo + hi) / 2;
1630 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1631 int cmp = hashcmp(index + x, sha1);
1633 return nth_packed_object_offset(p, mi);
1642 static int matches_pack_name(struct packed_git *p, const char *ig)
1644 const char *last_c, *c;
1646 if (!strcmp(p->pack_name, ig))
1649 for (c = p->pack_name, last_c = c; *c;)
1654 if (!strcmp(last_c, ig))
1660 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1662 struct packed_git *p;
1665 prepare_packed_git();
1667 for (p = packed_git; p; p = p->next) {
1668 if (ignore_packed) {
1670 for (ig = ignore_packed; *ig; ig++)
1671 if (!matches_pack_name(p, *ig))
1676 offset = find_pack_entry_one(sha1, p);
1679 * We are about to tell the caller where they can
1680 * locate the requested object. We better make
1681 * sure the packfile is still here and can be
1682 * accessed before supplying that answer, as
1683 * it may have been deleted since the index
1686 if (p->pack_fd == -1 && open_packed_git(p)) {
1687 error("packfile %s cannot be accessed", p->pack_name);
1692 hashcpy(e->sha1, sha1);
1699 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1700 struct packed_git *packs)
1702 struct packed_git *p;
1704 for (p = packs; p; p = p->next) {
1705 if (find_pack_entry_one(sha1, p))
1712 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1715 unsigned long mapsize, size;
1720 map = map_sha1_file(sha1, &mapsize);
1722 return error("unable to find %s", sha1_to_hex(sha1));
1723 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1724 status = error("unable to unpack %s header",
1726 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1727 status = error("unable to parse %s header", sha1_to_hex(sha1));
1730 inflateEnd(&stream);
1731 munmap(map, mapsize);
1735 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1737 struct pack_entry e;
1739 if (!find_pack_entry(sha1, &e, NULL)) {
1740 reprepare_packed_git();
1741 if (!find_pack_entry(sha1, &e, NULL))
1742 return sha1_loose_object_info(sha1, sizep);
1744 return packed_object_info(e.p, e.offset, sizep);
1747 static void *read_packed_sha1(const unsigned char *sha1,
1748 enum object_type *type, unsigned long *size)
1750 struct pack_entry e;
1752 if (!find_pack_entry(sha1, &e, NULL))
1755 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1759 * This is meant to hold a *small* number of objects that you would
1760 * want read_sha1_file() to be able to return, but yet you do not want
1761 * to write them into the object store (e.g. a browse-only
1764 static struct cached_object {
1765 unsigned char sha1[20];
1766 enum object_type type;
1770 static int cached_object_nr, cached_object_alloc;
1772 static struct cached_object *find_cached_object(const unsigned char *sha1)
1775 struct cached_object *co = cached_objects;
1777 for (i = 0; i < cached_object_nr; i++, co++) {
1778 if (!hashcmp(co->sha1, sha1))
1784 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1785 unsigned char *sha1)
1787 struct cached_object *co;
1789 hash_sha1_file(buf, len, typename(type), sha1);
1790 if (has_sha1_file(sha1) || find_cached_object(sha1))
1792 if (cached_object_alloc <= cached_object_nr) {
1793 cached_object_alloc = alloc_nr(cached_object_alloc);
1794 cached_objects = xrealloc(cached_objects,
1795 sizeof(*cached_objects) *
1796 cached_object_alloc);
1798 co = &cached_objects[cached_object_nr++];
1801 co->buf = xmalloc(len);
1802 memcpy(co->buf, buf, len);
1803 hashcpy(co->sha1, sha1);
1807 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1808 unsigned long *size)
1810 unsigned long mapsize;
1812 struct cached_object *co;
1814 co = find_cached_object(sha1);
1816 buf = xmalloc(co->size + 1);
1817 memcpy(buf, co->buf, co->size);
1818 ((char*)buf)[co->size] = 0;
1824 buf = read_packed_sha1(sha1, type, size);
1827 map = map_sha1_file(sha1, &mapsize);
1829 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1830 munmap(map, mapsize);
1833 reprepare_packed_git();
1834 return read_packed_sha1(sha1, type, size);
1837 void *read_object_with_reference(const unsigned char *sha1,
1838 const char *required_type_name,
1839 unsigned long *size,
1840 unsigned char *actual_sha1_return)
1842 enum object_type type, required_type;
1844 unsigned long isize;
1845 unsigned char actual_sha1[20];
1847 required_type = type_from_string(required_type_name);
1848 hashcpy(actual_sha1, sha1);
1850 int ref_length = -1;
1851 const char *ref_type = NULL;
1853 buffer = read_sha1_file(actual_sha1, &type, &isize);
1856 if (type == required_type) {
1858 if (actual_sha1_return)
1859 hashcpy(actual_sha1_return, actual_sha1);
1862 /* Handle references */
1863 else if (type == OBJ_COMMIT)
1865 else if (type == OBJ_TAG)
1866 ref_type = "object ";
1871 ref_length = strlen(ref_type);
1873 if (memcmp(buffer, ref_type, ref_length) ||
1874 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1879 /* Now we have the ID of the referred-to object in
1880 * actual_sha1. Check again. */
1884 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1885 const char *type, unsigned char *sha1,
1886 char *hdr, int *hdrlen)
1890 /* Generate the header */
1891 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1895 SHA1_Update(&c, hdr, *hdrlen);
1896 SHA1_Update(&c, buf, len);
1897 SHA1_Final(sha1, &c);
1901 * Link the tempfile to the final place, possibly creating the
1902 * last directory level as you do so.
1904 * Returns the errno on failure, 0 on success.
1906 static int link_temp_to_file(const char *tmpfile, const char *filename)
1911 if (!link(tmpfile, filename))
1915 * Try to mkdir the last path component if that failed.
1917 * Re-try the "link()" regardless of whether the mkdir
1918 * succeeds, since a race might mean that somebody
1922 dir = strrchr(filename, '/');
1925 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1930 if (!link(tmpfile, filename))
1938 * Move the just written object into its final resting place
1940 int move_temp_to_file(const char *tmpfile, const char *filename)
1942 int ret = link_temp_to_file(tmpfile, filename);
1945 * Coda hack - coda doesn't like cross-directory links,
1946 * so we fall back to a rename, which will mean that it
1947 * won't be able to check collisions, but that's not a
1950 * The same holds for FAT formatted media.
1952 * When this succeeds, we just return 0. We have nothing
1955 if (ret && ret != EEXIST) {
1956 if (!rename(tmpfile, filename))
1962 if (ret != EEXIST) {
1963 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1965 /* FIXME!!! Collision check here ? */
1971 static int write_buffer(int fd, const void *buf, size_t len)
1973 if (write_in_full(fd, buf, len) < 0)
1974 return error("file write error (%s)", strerror(errno));
1978 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1983 c = (type << 4) | (len & 15);
1996 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1998 int obj_type, hdrlen;
2000 if (use_legacy_headers) {
2001 while (deflate(stream, 0) == Z_OK)
2005 obj_type = type_from_string(type);
2006 hdrlen = write_binary_header(stream->next_out, obj_type, len);
2007 stream->total_out = hdrlen;
2008 stream->next_out += hdrlen;
2009 stream->avail_out -= hdrlen;
2012 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2013 unsigned char *sha1)
2017 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2021 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2024 unsigned char *compressed;
2026 unsigned char sha1[20];
2028 static char tmpfile[PATH_MAX];
2032 /* Normally if we have it in the pack then we do not bother writing
2033 * it out into .git/objects/??/?{38} file.
2035 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2036 filename = sha1_file_name(sha1);
2038 hashcpy(returnsha1, sha1);
2039 if (has_sha1_file(sha1))
2041 fd = open(filename, O_RDONLY);
2044 * FIXME!!! We might do collision checking here, but we'd
2045 * need to uncompress the old file and check it. Later.
2051 if (errno != ENOENT) {
2052 return error("sha1 file %s: %s\n", filename, strerror(errno));
2055 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2057 fd = mkstemp(tmpfile);
2060 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2062 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2066 memset(&stream, 0, sizeof(stream));
2067 deflateInit(&stream, zlib_compression_level);
2068 size = 8 + deflateBound(&stream, len+hdrlen);
2069 compressed = xmalloc(size);
2072 stream.next_out = compressed;
2073 stream.avail_out = size;
2075 /* First header.. */
2076 stream.next_in = (unsigned char *)hdr;
2077 stream.avail_in = hdrlen;
2078 setup_object_header(&stream, type, len);
2080 /* Then the data itself.. */
2081 stream.next_in = buf;
2082 stream.avail_in = len;
2083 ret = deflate(&stream, Z_FINISH);
2084 if (ret != Z_STREAM_END)
2085 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2087 ret = deflateEnd(&stream);
2089 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2091 size = stream.total_out;
2093 if (write_buffer(fd, compressed, size) < 0)
2094 die("unable to write sha1 file");
2097 die("unable to write sha1 file");
2100 return move_temp_to_file(tmpfile, filename);
2104 * We need to unpack and recompress the object for writing
2105 * it out to a different file.
2107 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2111 unsigned char *unpacked;
2113 enum object_type type;
2118 /* need to unpack and recompress it by itself */
2119 unpacked = read_packed_sha1(sha1, &type, &len);
2121 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2123 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2126 memset(&stream, 0, sizeof(stream));
2127 deflateInit(&stream, zlib_compression_level);
2128 size = deflateBound(&stream, len + hdrlen);
2129 buf = xmalloc(size);
2132 stream.next_out = buf;
2133 stream.avail_out = size;
2135 /* First header.. */
2136 stream.next_in = (void *)hdr;
2137 stream.avail_in = hdrlen;
2138 while (deflate(&stream, 0) == Z_OK)
2141 /* Then the data itself.. */
2142 stream.next_in = unpacked;
2143 stream.avail_in = len;
2144 while (deflate(&stream, Z_FINISH) == Z_OK)
2146 deflateEnd(&stream);
2149 *objsize = stream.total_out;
2153 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2156 unsigned long objsize;
2157 void *buf = map_sha1_file(sha1, &objsize);
2160 retval = write_buffer(fd, buf, objsize);
2161 munmap(buf, objsize);
2165 buf = repack_object(sha1, &objsize);
2166 retval = write_buffer(fd, buf, objsize);
2171 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2172 size_t bufsize, size_t *bufposn)
2174 char tmpfile[PATH_MAX];
2177 unsigned char real_sha1[20];
2178 unsigned char discard[4096];
2182 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2184 local = mkstemp(tmpfile);
2187 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2189 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2192 memset(&stream, 0, sizeof(stream));
2194 inflateInit(&stream);
2201 stream.avail_in = *bufposn;
2202 stream.next_in = (unsigned char *) buffer;
2204 stream.next_out = discard;
2205 stream.avail_out = sizeof(discard);
2206 ret = inflate(&stream, Z_SYNC_FLUSH);
2207 SHA1_Update(&c, discard, sizeof(discard) -
2209 } while (stream.avail_in && ret == Z_OK);
2210 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2211 die("unable to write sha1 file");
2212 memmove(buffer, buffer + *bufposn - stream.avail_in,
2214 *bufposn = stream.avail_in;
2218 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2223 return error("Connection closed?");
2224 perror("Reading from connection");
2229 inflateEnd(&stream);
2231 fchmod(local, 0444);
2232 if (close(local) != 0)
2233 die("unable to write sha1 file");
2234 SHA1_Final(real_sha1, &c);
2235 if (ret != Z_STREAM_END) {
2237 return error("File %s corrupted", sha1_to_hex(sha1));
2239 if (hashcmp(sha1, real_sha1)) {
2241 return error("File %s has bad hash", sha1_to_hex(sha1));
2244 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2247 int has_pack_index(const unsigned char *sha1)
2250 if (stat(sha1_pack_index_name(sha1), &st))
2255 int has_pack_file(const unsigned char *sha1)
2258 if (stat(sha1_pack_name(sha1), &st))
2263 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2265 struct pack_entry e;
2266 return find_pack_entry(sha1, &e, ignore_packed);
2269 int has_sha1_file(const unsigned char *sha1)
2272 struct pack_entry e;
2274 if (find_pack_entry(sha1, &e, NULL))
2276 return find_sha1_file(sha1, &st) ? 1 : 0;
2280 * reads from fd as long as possible into a supplied buffer of size bytes.
2281 * If necessary the buffer's size is increased using realloc()
2283 * returns 0 if anything went fine and -1 otherwise
2285 * NOTE: both buf and size may change, but even when -1 is returned
2286 * you still have to free() it yourself.
2288 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2290 char* buf = *return_buf;
2291 unsigned long size = *return_size;
2293 unsigned long off = 0;
2296 iret = xread(fd, buf + off, size - off);
2301 buf = xrealloc(buf, size);
2314 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2316 unsigned long size = 4096;
2317 char *buf = xmalloc(size);
2320 if (read_pipe(fd, &buf, &size)) {
2328 ret = write_sha1_file(buf, size, type, sha1);
2330 ret = hash_sha1_file(buf, size, type, sha1);
2335 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2336 enum object_type type, const char *path)
2338 size_t size = xsize_t(st->st_size);
2340 int ret, re_allocated = 0;
2343 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2350 * Convert blobs to git internal format
2352 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2353 unsigned long nsize = size;
2354 char *nbuf = convert_to_git(path, buf, &nsize);
2364 ret = write_sha1_file(buf, size, typename(type), sha1);
2366 ret = hash_sha1_file(buf, size, typename(type), sha1);
2376 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2382 switch (st->st_mode & S_IFMT) {
2384 fd = open(path, O_RDONLY);
2386 return error("open(\"%s\"): %s", path,
2388 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2389 return error("%s: failed to insert into database",
2393 len = xsize_t(st->st_size);
2394 target = xmalloc(len + 1);
2395 if (readlink(path, target, len + 1) != st->st_size) {
2396 char *errstr = strerror(errno);
2398 return error("readlink(\"%s\"): %s", path,
2402 hash_sha1_file(target, len, blob_type, sha1);
2403 else if (write_sha1_file(target, len, blob_type, sha1))
2404 return error("%s: failed to insert into database",
2409 return resolve_gitlink_ref(path, "HEAD", sha1);
2411 return error("%s: unsupported file type", path);
2416 int read_pack_header(int fd, struct pack_header *header)
2418 char *c = (char*)header;
2419 ssize_t remaining = sizeof(struct pack_header);
2421 ssize_t r = xread(fd, c, remaining);
2423 /* "eof before pack header was fully read" */
2424 return PH_ERROR_EOF;
2427 } while (remaining > 0);
2428 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2429 /* "protocol error (pack signature mismatch detected)" */
2430 return PH_ERROR_PACK_SIGNATURE;
2431 if (!pack_version_ok(header->hdr_version))
2432 /* "protocol error (pack version unsupported)" */
2433 return PH_ERROR_PROTOCOL;