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 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);
459 /* a future index format would start with this, as older git
460 * binaries would fail the non-monotonic index check below.
461 * give a nicer warning to the user if we can.
464 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
465 munmap(idx_map, idx_size);
466 return error("index file %s is a newer version"
467 " and is not supported by this binary"
468 " (try upgrading GIT to a newer version)",
474 for (i = 0; i < 256; i++) {
475 uint32_t n = ntohl(index[i]);
477 munmap(idx_map, idx_size);
478 return error("non-monotonic index %s", path);
485 * - 256 index entries 4 bytes each
486 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
487 * - 20-byte SHA1 of the packfile
488 * - 20-byte SHA1 file checksum
490 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
491 munmap(idx_map, idx_size);
492 return error("wrong index file size in %s", path);
495 p->index_version = 1;
496 p->index_data = idx_map;
497 p->index_size = idx_size;
501 static void scan_windows(struct packed_git *p,
502 struct packed_git **lru_p,
503 struct pack_window **lru_w,
504 struct pack_window **lru_l)
506 struct pack_window *w, *w_l;
508 for (w_l = NULL, w = p->windows; w; w = w->next) {
510 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
520 static int unuse_one_window(struct packed_git *current)
522 struct packed_git *p, *lru_p = NULL;
523 struct pack_window *lru_w = NULL, *lru_l = NULL;
526 scan_windows(current, &lru_p, &lru_w, &lru_l);
527 for (p = packed_git; p; p = p->next)
528 scan_windows(p, &lru_p, &lru_w, &lru_l);
530 munmap(lru_w->base, lru_w->len);
531 pack_mapped -= lru_w->len;
533 lru_l->next = lru_w->next;
535 lru_p->windows = lru_w->next;
536 if (!lru_p->windows && lru_p != current) {
537 close(lru_p->pack_fd);
548 void release_pack_memory(size_t need)
550 size_t cur = pack_mapped;
551 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
555 void unuse_pack(struct pack_window **w_cursor)
557 struct pack_window *w = *w_cursor;
565 * Do not call this directly as this leaks p->pack_fd on error return;
566 * call open_packed_git() instead.
568 static int open_packed_git_1(struct packed_git *p)
571 struct pack_header hdr;
572 unsigned char sha1[20];
573 unsigned char *idx_sha1;
576 p->pack_fd = open(p->pack_name, O_RDONLY);
577 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
580 /* If we created the struct before we had the pack we lack size. */
582 if (!S_ISREG(st.st_mode))
583 return error("packfile %s not a regular file", p->pack_name);
584 p->pack_size = st.st_size;
585 } else if (p->pack_size != st.st_size)
586 return error("packfile %s size changed", p->pack_name);
588 /* We leave these file descriptors open with sliding mmap;
589 * there is no point keeping them open across exec(), though.
591 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
593 return error("cannot determine file descriptor flags");
594 fd_flag |= FD_CLOEXEC;
595 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
596 return error("cannot set FD_CLOEXEC");
598 /* Verify we recognize this pack file format. */
599 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
600 return error("file %s is far too short to be a packfile", p->pack_name);
601 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
602 return error("file %s is not a GIT packfile", p->pack_name);
603 if (!pack_version_ok(hdr.hdr_version))
604 return error("packfile %s is version %u and not supported"
605 " (try upgrading GIT to a newer version)",
606 p->pack_name, ntohl(hdr.hdr_version));
608 /* Verify the pack matches its index. */
609 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
610 return error("packfile %s claims to have %u objects"
611 " while index size indicates %u objects",
612 p->pack_name, ntohl(hdr.hdr_entries),
613 num_packed_objects(p));
614 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
615 return error("end of packfile %s is unavailable", p->pack_name);
616 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
617 return error("packfile %s signature is unavailable", p->pack_name);
618 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
619 if (hashcmp(sha1, idx_sha1))
620 return error("packfile %s does not match index", p->pack_name);
624 static int open_packed_git(struct packed_git *p)
626 if (!open_packed_git_1(p))
628 if (p->pack_fd != -1) {
635 static int in_window(struct pack_window *win, off_t offset)
637 /* We must promise at least 20 bytes (one hash) after the
638 * offset is available from this window, otherwise the offset
639 * is not actually in this window and a different window (which
640 * has that one hash excess) must be used. This is to support
641 * the object header and delta base parsing routines below.
643 off_t win_off = win->offset;
644 return win_off <= offset
645 && (offset + 20) <= (win_off + win->len);
648 unsigned char* use_pack(struct packed_git *p,
649 struct pack_window **w_cursor,
653 struct pack_window *win = *w_cursor;
655 if (p->pack_fd == -1 && open_packed_git(p))
656 die("packfile %s cannot be accessed", p->pack_name);
658 /* Since packfiles end in a hash of their content and its
659 * pointless to ask for an offset into the middle of that
660 * hash, and the in_window function above wouldn't match
661 * don't allow an offset too close to the end of the file.
663 if (offset > (p->pack_size - 20))
664 die("offset beyond end of packfile (truncated pack?)");
666 if (!win || !in_window(win, offset)) {
669 for (win = p->windows; win; win = win->next) {
670 if (in_window(win, offset))
674 size_t window_align = packed_git_window_size / 2;
676 win = xcalloc(1, sizeof(*win));
677 win->offset = (offset / window_align) * window_align;
678 len = p->pack_size - win->offset;
679 if (len > packed_git_window_size)
680 len = packed_git_window_size;
681 win->len = (size_t)len;
682 pack_mapped += win->len;
683 while (packed_git_limit < pack_mapped
684 && unuse_one_window(p))
686 win->base = xmmap(NULL, win->len,
687 PROT_READ, MAP_PRIVATE,
688 p->pack_fd, win->offset);
689 if (win->base == MAP_FAILED)
690 die("packfile %s cannot be mapped: %s",
695 if (pack_mapped > peak_pack_mapped)
696 peak_pack_mapped = pack_mapped;
697 if (pack_open_windows > peak_pack_open_windows)
698 peak_pack_open_windows = pack_open_windows;
699 win->next = p->windows;
703 if (win != *w_cursor) {
704 win->last_used = pack_used_ctr++;
708 offset -= win->offset;
710 *left = win->len - xsize_t(offset);
711 return win->base + offset;
714 struct packed_git *add_packed_git(const char *path, int path_len, int local)
717 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
720 * Make sure a corresponding .pack file exists and that
721 * the index looks sane.
723 path_len -= strlen(".idx");
726 memcpy(p->pack_name, path, path_len);
727 strcpy(p->pack_name + path_len, ".pack");
728 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
729 check_packed_git_idx(path, p)) {
734 /* ok, it looks sane as far as we can check without
735 * actually mapping the pack file.
737 p->pack_size = st.st_size;
741 p->pack_local = local;
742 p->mtime = st.st_mtime;
743 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
748 struct packed_git *parse_pack_index(unsigned char *sha1)
750 char *path = sha1_pack_index_name(sha1);
751 return parse_pack_index_file(sha1, path);
754 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
755 const char *idx_path)
757 const char *path = sha1_pack_name(sha1);
758 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
760 if (check_packed_git_idx(idx_path, p)) {
765 strcpy(p->pack_name, path);
770 hashcpy(p->sha1, sha1);
774 void install_packed_git(struct packed_git *pack)
776 pack->next = packed_git;
780 static void prepare_packed_git_one(char *objdir, int local)
787 sprintf(path, "%s/pack", objdir);
792 error("unable to open object pack directory: %s: %s",
793 path, strerror(errno));
797 while ((de = readdir(dir)) != NULL) {
798 int namelen = strlen(de->d_name);
799 struct packed_git *p;
801 if (!has_extension(de->d_name, ".idx"))
804 /* Don't reopen a pack we already have. */
805 strcpy(path + len, de->d_name);
806 for (p = packed_git; p; p = p->next) {
807 if (!memcmp(path, p->pack_name, len + namelen - 4))
812 /* See if it really is a valid .idx file with corresponding
813 * .pack file that we can map.
815 p = add_packed_git(path, len + namelen, local);
818 install_packed_git(p);
823 static int sort_pack(const void *a_, const void *b_)
825 struct packed_git *a = *((struct packed_git **)a_);
826 struct packed_git *b = *((struct packed_git **)b_);
830 * Local packs tend to contain objects specific to our
831 * variant of the project than remote ones. In addition,
832 * remote ones could be on a network mounted filesystem.
833 * Favor local ones for these reasons.
835 st = a->pack_local - b->pack_local;
840 * Younger packs tend to contain more recent objects,
841 * and more recent objects tend to get accessed more
844 if (a->mtime < b->mtime)
846 else if (a->mtime == b->mtime)
851 static void rearrange_packed_git(void)
853 struct packed_git **ary, *p;
856 for (n = 0, p = packed_git; p; p = p->next)
861 /* prepare an array of packed_git for easier sorting */
862 ary = xcalloc(n, sizeof(struct packed_git *));
863 for (n = 0, p = packed_git; p; p = p->next)
866 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
868 /* link them back again */
869 for (i = 0; i < n - 1; i++)
870 ary[i]->next = ary[i + 1];
871 ary[n - 1]->next = NULL;
877 static int prepare_packed_git_run_once = 0;
878 void prepare_packed_git(void)
880 struct alternate_object_database *alt;
882 if (prepare_packed_git_run_once)
884 prepare_packed_git_one(get_object_directory(), 1);
886 for (alt = alt_odb_list; alt; alt = alt->next) {
888 prepare_packed_git_one(alt->base, 0);
891 rearrange_packed_git();
892 prepare_packed_git_run_once = 1;
895 void reprepare_packed_git(void)
897 prepare_packed_git_run_once = 0;
898 prepare_packed_git();
901 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
903 unsigned char real_sha1[20];
904 hash_sha1_file(map, size, type, real_sha1);
905 return hashcmp(sha1, real_sha1) ? -1 : 0;
908 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
913 char *filename = find_sha1_file(sha1, &st);
919 fd = open(filename, O_RDONLY | sha1_file_open_flag);
921 /* See if it works without O_NOATIME */
922 switch (sha1_file_open_flag) {
924 fd = open(filename, O_RDONLY);
932 /* If it failed once, it will probably fail again.
933 * Stop using O_NOATIME
935 sha1_file_open_flag = 0;
937 *size = xsize_t(st.st_size);
938 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
943 int legacy_loose_object(unsigned char *map)
948 * Is it a zlib-compressed buffer? If so, the first byte
949 * must be 0x78 (15-bit window size, deflated), and the
950 * first 16-bit word is evenly divisible by 31
952 word = (map[0] << 8) + map[1];
953 if (map[0] == 0x78 && !(word % 31))
959 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
964 unsigned long used = 0;
967 *type = (c >> 4) & 7;
973 if (sizeof(long) * 8 <= shift)
976 size += (c & 0x7f) << shift;
983 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
985 unsigned long size, used;
986 static const char valid_loose_object_type[8] = {
988 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
989 0, /* "delta" and others are invalid in a loose object */
991 enum object_type type;
993 /* Get the data stream */
994 memset(stream, 0, sizeof(*stream));
995 stream->next_in = map;
996 stream->avail_in = mapsize;
997 stream->next_out = buffer;
998 stream->avail_out = bufsiz;
1000 if (legacy_loose_object(map)) {
1001 inflateInit(stream);
1002 return inflate(stream, 0);
1005 used = unpack_object_header_gently(map, mapsize, &type, &size);
1006 if (!used || !valid_loose_object_type[type])
1011 /* Set up the stream for the rest.. */
1012 stream->next_in = map;
1013 stream->avail_in = mapsize;
1014 inflateInit(stream);
1016 /* And generate the fake traditional header */
1017 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1018 typename(type), size);
1022 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1024 int bytes = strlen(buffer) + 1;
1025 unsigned char *buf = xmalloc(1+size);
1029 n = stream->total_out - bytes;
1032 memcpy(buf, (char *) buffer + bytes, n);
1034 if (bytes <= size) {
1036 * The above condition must be (bytes <= size), not
1037 * (bytes < size). In other words, even though we
1038 * expect no more output and set avail_out to zer0,
1039 * the input zlib stream may have bytes that express
1040 * "this concludes the stream", and we *do* want to
1043 * Otherwise we would not be able to test that we
1044 * consumed all the input to reach the expected size;
1045 * we also want to check that zlib tells us that all
1046 * went well with status == Z_STREAM_END at the end.
1048 stream->next_out = buf + bytes;
1049 stream->avail_out = size - bytes;
1050 while (status == Z_OK)
1051 status = inflate(stream, Z_FINISH);
1054 if (status == Z_STREAM_END && !stream->avail_in) {
1060 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1061 else if (stream->avail_in)
1062 error("garbage at end of loose object '%s'",
1069 * We used to just use "sscanf()", but that's actually way
1070 * too permissive for what we want to check. So do an anal
1071 * object header parse by hand.
1073 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1080 * The type can be at most ten bytes (including the
1081 * terminating '\0' that we add), and is followed by
1090 if (i >= sizeof(type))
1096 * The length must follow immediately, and be in canonical
1097 * decimal format (ie "010" is not valid).
1099 size = *hdr++ - '0';
1104 unsigned long c = *hdr - '0';
1108 size = size * 10 + c;
1114 * The length must be followed by a zero byte
1116 return *hdr ? -1 : type_from_string(type);
1119 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1125 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1126 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1129 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1132 static off_t get_delta_base(struct packed_git *p,
1133 struct pack_window **w_curs,
1135 enum object_type type,
1136 off_t delta_obj_offset)
1138 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1141 /* use_pack() assured us we have [base_info, base_info + 20)
1142 * as a range that we can look at without walking off the
1143 * end of the mapped window. Its actually the hash size
1144 * that is assured. An OFS_DELTA longer than the hash size
1145 * is stupid, as then a REF_DELTA would be smaller to store.
1147 if (type == OBJ_OFS_DELTA) {
1149 unsigned char c = base_info[used++];
1150 base_offset = c & 127;
1153 if (!base_offset || base_offset & ~(~0UL >> 7))
1154 die("offset value overflow for delta base object");
1155 c = base_info[used++];
1156 base_offset = (base_offset << 7) + (c & 127);
1158 base_offset = delta_obj_offset - base_offset;
1159 if (base_offset >= delta_obj_offset)
1160 die("delta base offset out of bound");
1162 } else if (type == OBJ_REF_DELTA) {
1163 /* The base entry _must_ be in the same pack */
1164 base_offset = find_pack_entry_one(base_info, p);
1166 die("failed to find delta-pack base object %s",
1167 sha1_to_hex(base_info));
1170 die("I am totally screwed");
1174 /* forward declaration for a mutually recursive function */
1175 static int packed_object_info(struct packed_git *p, off_t offset,
1176 unsigned long *sizep);
1178 static int packed_delta_info(struct packed_git *p,
1179 struct pack_window **w_curs,
1181 enum object_type type,
1183 unsigned long *sizep)
1187 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1188 type = packed_object_info(p, base_offset, NULL);
1190 /* We choose to only get the type of the base object and
1191 * ignore potentially corrupt pack file that expects the delta
1192 * based on a base with a wrong size. This saves tons of
1196 const unsigned char *data;
1197 unsigned char delta_head[20], *in;
1201 memset(&stream, 0, sizeof(stream));
1202 stream.next_out = delta_head;
1203 stream.avail_out = sizeof(delta_head);
1205 inflateInit(&stream);
1207 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1208 stream.next_in = in;
1209 st = inflate(&stream, Z_FINISH);
1210 curpos += stream.next_in - in;
1211 } while ((st == Z_OK || st == Z_BUF_ERROR)
1212 && stream.total_out < sizeof(delta_head));
1213 inflateEnd(&stream);
1214 if ((st != Z_STREAM_END) &&
1215 stream.total_out != sizeof(delta_head))
1216 die("delta data unpack-initial failed");
1218 /* Examine the initial part of the delta to figure out
1223 /* ignore base size */
1224 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1226 /* Read the result size */
1227 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1233 static int unpack_object_header(struct packed_git *p,
1234 struct pack_window **w_curs,
1236 unsigned long *sizep)
1238 unsigned char *base;
1241 enum object_type type;
1243 /* use_pack() assures us we have [base, base + 20) available
1244 * as a range that we can look at at. (Its actually the hash
1245 * size that is assured.) With our object header encoding
1246 * the maximum deflated object size is 2^137, which is just
1247 * insane, so we know won't exceed what we have been given.
1249 base = use_pack(p, w_curs, *curpos, &left);
1250 used = unpack_object_header_gently(base, left, &type, sizep);
1252 die("object offset outside of pack file");
1258 const char *packed_object_info_detail(struct packed_git *p,
1260 unsigned long *size,
1261 unsigned long *store_size,
1262 unsigned int *delta_chain_length,
1263 unsigned char *base_sha1)
1265 struct pack_window *w_curs = NULL;
1267 unsigned long dummy;
1268 unsigned char *next_sha1;
1269 enum object_type type;
1271 *delta_chain_length = 0;
1272 curpos = obj_offset;
1273 type = unpack_object_header(p, &w_curs, &curpos, size);
1278 die("pack %s contains unknown object type %d",
1279 p->pack_name, type);
1284 *store_size = 0; /* notyet */
1285 unuse_pack(&w_curs);
1286 return typename(type);
1288 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1289 if (*delta_chain_length == 0) {
1290 /* TODO: find base_sha1 as pointed by curpos */
1295 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1296 if (*delta_chain_length == 0)
1297 hashcpy(base_sha1, next_sha1);
1298 obj_offset = find_pack_entry_one(next_sha1, p);
1301 (*delta_chain_length)++;
1302 curpos = obj_offset;
1303 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1307 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1308 unsigned long *sizep)
1310 struct pack_window *w_curs = NULL;
1312 off_t curpos = obj_offset;
1313 enum object_type type;
1315 type = unpack_object_header(p, &w_curs, &curpos, &size);
1320 type = packed_delta_info(p, &w_curs, curpos,
1321 type, obj_offset, sizep);
1331 die("pack %s contains unknown object type %d",
1332 p->pack_name, type);
1334 unuse_pack(&w_curs);
1338 static void *unpack_compressed_entry(struct packed_git *p,
1339 struct pack_window **w_curs,
1345 unsigned char *buffer, *in;
1347 buffer = xmalloc(size + 1);
1349 memset(&stream, 0, sizeof(stream));
1350 stream.next_out = buffer;
1351 stream.avail_out = size;
1353 inflateInit(&stream);
1355 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1356 stream.next_in = in;
1357 st = inflate(&stream, Z_FINISH);
1358 curpos += stream.next_in - in;
1359 } while (st == Z_OK || st == Z_BUF_ERROR);
1360 inflateEnd(&stream);
1361 if ((st != Z_STREAM_END) || stream.total_out != size) {
1369 #define MAX_DELTA_CACHE (256)
1371 static size_t delta_base_cached;
1373 static struct delta_base_cache_lru_list {
1374 struct delta_base_cache_lru_list *prev;
1375 struct delta_base_cache_lru_list *next;
1376 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1378 static struct delta_base_cache_entry {
1379 struct delta_base_cache_lru_list lru;
1381 struct packed_git *p;
1384 enum object_type type;
1385 } delta_base_cache[MAX_DELTA_CACHE];
1387 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1391 hash = (unsigned long)p + (unsigned long)base_offset;
1392 hash += (hash >> 8) + (hash >> 16);
1393 return hash % MAX_DELTA_CACHE;
1396 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1397 unsigned long *base_size, enum object_type *type, int keep_cache)
1400 unsigned long hash = pack_entry_hash(p, base_offset);
1401 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1404 if (ret && ent->p == p && ent->base_offset == base_offset)
1405 goto found_cache_entry;
1406 return unpack_entry(p, base_offset, type, base_size);
1411 ent->lru.next->prev = ent->lru.prev;
1412 ent->lru.prev->next = ent->lru.next;
1413 delta_base_cached -= ent->size;
1416 ret = xmalloc(ent->size + 1);
1417 memcpy(ret, ent->data, ent->size);
1418 ((char *)ret)[ent->size] = 0;
1421 *base_size = ent->size;
1425 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1430 ent->lru.next->prev = ent->lru.prev;
1431 ent->lru.prev->next = ent->lru.next;
1432 delta_base_cached -= ent->size;
1436 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1437 void *base, unsigned long base_size, enum object_type type)
1439 unsigned long hash = pack_entry_hash(p, base_offset);
1440 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1441 struct delta_base_cache_lru_list *lru;
1443 release_delta_base_cache(ent);
1444 delta_base_cached += base_size;
1446 for (lru = delta_base_cache_lru.next;
1447 delta_base_cached > delta_base_cache_limit
1448 && lru != &delta_base_cache_lru;
1450 struct delta_base_cache_entry *f = (void *)lru;
1451 if (f->type == OBJ_BLOB)
1452 release_delta_base_cache(f);
1454 for (lru = delta_base_cache_lru.next;
1455 delta_base_cached > delta_base_cache_limit
1456 && lru != &delta_base_cache_lru;
1458 struct delta_base_cache_entry *f = (void *)lru;
1459 release_delta_base_cache(f);
1463 ent->base_offset = base_offset;
1466 ent->size = base_size;
1467 ent->lru.next = &delta_base_cache_lru;
1468 ent->lru.prev = delta_base_cache_lru.prev;
1469 delta_base_cache_lru.prev->next = &ent->lru;
1470 delta_base_cache_lru.prev = &ent->lru;
1473 static void *unpack_delta_entry(struct packed_git *p,
1474 struct pack_window **w_curs,
1476 unsigned long delta_size,
1478 enum object_type *type,
1479 unsigned long *sizep)
1481 void *delta_data, *result, *base;
1482 unsigned long base_size;
1485 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1486 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1488 die("failed to read delta base object"
1489 " at %"PRIuMAX" from %s",
1490 (uintmax_t)base_offset, p->pack_name);
1492 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1493 result = patch_delta(base, base_size,
1494 delta_data, delta_size,
1497 die("failed to apply delta");
1499 add_delta_base_cache(p, base_offset, base, base_size, *type);
1503 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1504 enum object_type *type, unsigned long *sizep)
1506 struct pack_window *w_curs = NULL;
1507 off_t curpos = obj_offset;
1510 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1514 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1515 obj_offset, type, sizep);
1521 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1524 die("unknown object type %i in %s", *type, p->pack_name);
1526 unuse_pack(&w_curs);
1530 uint32_t num_packed_objects(const struct packed_git *p)
1532 /* See check_packed_git_idx() */
1533 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1536 const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1539 const unsigned char *index = p->index_data;
1541 if (num_packed_objects(p) <= n)
1543 return index + 24 * n + 4;
1546 off_t find_pack_entry_one(const unsigned char *sha1,
1547 struct packed_git *p)
1549 const uint32_t *level1_ofs = p->index_data;
1550 int hi = ntohl(level1_ofs[*sha1]);
1551 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1552 const unsigned char *index = p->index_data;
1557 int mi = (lo + hi) / 2;
1558 int cmp = hashcmp(index + 24 * mi + 4, sha1);
1560 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1569 static int matches_pack_name(struct packed_git *p, const char *ig)
1571 const char *last_c, *c;
1573 if (!strcmp(p->pack_name, ig))
1576 for (c = p->pack_name, last_c = c; *c;)
1581 if (!strcmp(last_c, ig))
1587 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1589 struct packed_git *p;
1592 prepare_packed_git();
1594 for (p = packed_git; p; p = p->next) {
1595 if (ignore_packed) {
1597 for (ig = ignore_packed; *ig; ig++)
1598 if (!matches_pack_name(p, *ig))
1603 offset = find_pack_entry_one(sha1, p);
1606 * We are about to tell the caller where they can
1607 * locate the requested object. We better make
1608 * sure the packfile is still here and can be
1609 * accessed before supplying that answer, as
1610 * it may have been deleted since the index
1613 if (p->pack_fd == -1 && open_packed_git(p)) {
1614 error("packfile %s cannot be accessed", p->pack_name);
1619 hashcpy(e->sha1, sha1);
1626 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1627 struct packed_git *packs)
1629 struct packed_git *p;
1631 for (p = packs; p; p = p->next) {
1632 if (find_pack_entry_one(sha1, p))
1639 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1642 unsigned long mapsize, size;
1647 map = map_sha1_file(sha1, &mapsize);
1649 return error("unable to find %s", sha1_to_hex(sha1));
1650 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1651 status = error("unable to unpack %s header",
1653 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1654 status = error("unable to parse %s header", sha1_to_hex(sha1));
1657 inflateEnd(&stream);
1658 munmap(map, mapsize);
1662 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1664 struct pack_entry e;
1666 if (!find_pack_entry(sha1, &e, NULL)) {
1667 reprepare_packed_git();
1668 if (!find_pack_entry(sha1, &e, NULL))
1669 return sha1_loose_object_info(sha1, sizep);
1671 return packed_object_info(e.p, e.offset, sizep);
1674 static void *read_packed_sha1(const unsigned char *sha1,
1675 enum object_type *type, unsigned long *size)
1677 struct pack_entry e;
1679 if (!find_pack_entry(sha1, &e, NULL))
1682 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1686 * This is meant to hold a *small* number of objects that you would
1687 * want read_sha1_file() to be able to return, but yet you do not want
1688 * to write them into the object store (e.g. a browse-only
1691 static struct cached_object {
1692 unsigned char sha1[20];
1693 enum object_type type;
1697 static int cached_object_nr, cached_object_alloc;
1699 static struct cached_object *find_cached_object(const unsigned char *sha1)
1702 struct cached_object *co = cached_objects;
1704 for (i = 0; i < cached_object_nr; i++, co++) {
1705 if (!hashcmp(co->sha1, sha1))
1711 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1712 unsigned char *sha1)
1714 struct cached_object *co;
1716 hash_sha1_file(buf, len, typename(type), sha1);
1717 if (has_sha1_file(sha1) || find_cached_object(sha1))
1719 if (cached_object_alloc <= cached_object_nr) {
1720 cached_object_alloc = alloc_nr(cached_object_alloc);
1721 cached_objects = xrealloc(cached_objects,
1722 sizeof(*cached_objects) *
1723 cached_object_alloc);
1725 co = &cached_objects[cached_object_nr++];
1728 co->buf = xmalloc(len);
1729 memcpy(co->buf, buf, len);
1730 hashcpy(co->sha1, sha1);
1734 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1735 unsigned long *size)
1737 unsigned long mapsize;
1739 struct cached_object *co;
1741 co = find_cached_object(sha1);
1743 buf = xmalloc(co->size + 1);
1744 memcpy(buf, co->buf, co->size);
1745 ((char*)buf)[co->size] = 0;
1751 buf = read_packed_sha1(sha1, type, size);
1754 map = map_sha1_file(sha1, &mapsize);
1756 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1757 munmap(map, mapsize);
1760 reprepare_packed_git();
1761 return read_packed_sha1(sha1, type, size);
1764 void *read_object_with_reference(const unsigned char *sha1,
1765 const char *required_type_name,
1766 unsigned long *size,
1767 unsigned char *actual_sha1_return)
1769 enum object_type type, required_type;
1771 unsigned long isize;
1772 unsigned char actual_sha1[20];
1774 required_type = type_from_string(required_type_name);
1775 hashcpy(actual_sha1, sha1);
1777 int ref_length = -1;
1778 const char *ref_type = NULL;
1780 buffer = read_sha1_file(actual_sha1, &type, &isize);
1783 if (type == required_type) {
1785 if (actual_sha1_return)
1786 hashcpy(actual_sha1_return, actual_sha1);
1789 /* Handle references */
1790 else if (type == OBJ_COMMIT)
1792 else if (type == OBJ_TAG)
1793 ref_type = "object ";
1798 ref_length = strlen(ref_type);
1800 if (memcmp(buffer, ref_type, ref_length) ||
1801 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1806 /* Now we have the ID of the referred-to object in
1807 * actual_sha1. Check again. */
1811 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1812 const char *type, unsigned char *sha1,
1813 char *hdr, int *hdrlen)
1817 /* Generate the header */
1818 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1822 SHA1_Update(&c, hdr, *hdrlen);
1823 SHA1_Update(&c, buf, len);
1824 SHA1_Final(sha1, &c);
1828 * Link the tempfile to the final place, possibly creating the
1829 * last directory level as you do so.
1831 * Returns the errno on failure, 0 on success.
1833 static int link_temp_to_file(const char *tmpfile, const char *filename)
1838 if (!link(tmpfile, filename))
1842 * Try to mkdir the last path component if that failed.
1844 * Re-try the "link()" regardless of whether the mkdir
1845 * succeeds, since a race might mean that somebody
1849 dir = strrchr(filename, '/');
1852 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1857 if (!link(tmpfile, filename))
1865 * Move the just written object into its final resting place
1867 int move_temp_to_file(const char *tmpfile, const char *filename)
1869 int ret = link_temp_to_file(tmpfile, filename);
1872 * Coda hack - coda doesn't like cross-directory links,
1873 * so we fall back to a rename, which will mean that it
1874 * won't be able to check collisions, but that's not a
1877 * The same holds for FAT formatted media.
1879 * When this succeeds, we just return 0. We have nothing
1882 if (ret && ret != EEXIST) {
1883 if (!rename(tmpfile, filename))
1889 if (ret != EEXIST) {
1890 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1892 /* FIXME!!! Collision check here ? */
1898 static int write_buffer(int fd, const void *buf, size_t len)
1900 if (write_in_full(fd, buf, len) < 0)
1901 return error("file write error (%s)", strerror(errno));
1905 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1910 c = (type << 4) | (len & 15);
1923 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1925 int obj_type, hdrlen;
1927 if (use_legacy_headers) {
1928 while (deflate(stream, 0) == Z_OK)
1932 obj_type = type_from_string(type);
1933 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1934 stream->total_out = hdrlen;
1935 stream->next_out += hdrlen;
1936 stream->avail_out -= hdrlen;
1939 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1940 unsigned char *sha1)
1944 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1948 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1951 unsigned char *compressed;
1953 unsigned char sha1[20];
1955 static char tmpfile[PATH_MAX];
1959 /* Normally if we have it in the pack then we do not bother writing
1960 * it out into .git/objects/??/?{38} file.
1962 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1963 filename = sha1_file_name(sha1);
1965 hashcpy(returnsha1, sha1);
1966 if (has_sha1_file(sha1))
1968 fd = open(filename, O_RDONLY);
1971 * FIXME!!! We might do collision checking here, but we'd
1972 * need to uncompress the old file and check it. Later.
1978 if (errno != ENOENT) {
1979 return error("sha1 file %s: %s\n", filename, strerror(errno));
1982 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
1984 fd = mkstemp(tmpfile);
1987 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1989 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1993 memset(&stream, 0, sizeof(stream));
1994 deflateInit(&stream, zlib_compression_level);
1995 size = 8 + deflateBound(&stream, len+hdrlen);
1996 compressed = xmalloc(size);
1999 stream.next_out = compressed;
2000 stream.avail_out = size;
2002 /* First header.. */
2003 stream.next_in = (unsigned char *)hdr;
2004 stream.avail_in = hdrlen;
2005 setup_object_header(&stream, type, len);
2007 /* Then the data itself.. */
2008 stream.next_in = buf;
2009 stream.avail_in = len;
2010 ret = deflate(&stream, Z_FINISH);
2011 if (ret != Z_STREAM_END)
2012 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2014 ret = deflateEnd(&stream);
2016 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2018 size = stream.total_out;
2020 if (write_buffer(fd, compressed, size) < 0)
2021 die("unable to write sha1 file");
2024 die("unable to write sha1 file");
2027 return move_temp_to_file(tmpfile, filename);
2031 * We need to unpack and recompress the object for writing
2032 * it out to a different file.
2034 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2038 unsigned char *unpacked;
2040 enum object_type type;
2045 /* need to unpack and recompress it by itself */
2046 unpacked = read_packed_sha1(sha1, &type, &len);
2048 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2050 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2053 memset(&stream, 0, sizeof(stream));
2054 deflateInit(&stream, zlib_compression_level);
2055 size = deflateBound(&stream, len + hdrlen);
2056 buf = xmalloc(size);
2059 stream.next_out = buf;
2060 stream.avail_out = size;
2062 /* First header.. */
2063 stream.next_in = (void *)hdr;
2064 stream.avail_in = hdrlen;
2065 while (deflate(&stream, 0) == Z_OK)
2068 /* Then the data itself.. */
2069 stream.next_in = unpacked;
2070 stream.avail_in = len;
2071 while (deflate(&stream, Z_FINISH) == Z_OK)
2073 deflateEnd(&stream);
2076 *objsize = stream.total_out;
2080 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2083 unsigned long objsize;
2084 void *buf = map_sha1_file(sha1, &objsize);
2087 retval = write_buffer(fd, buf, objsize);
2088 munmap(buf, objsize);
2092 buf = repack_object(sha1, &objsize);
2093 retval = write_buffer(fd, buf, objsize);
2098 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2099 size_t bufsize, size_t *bufposn)
2101 char tmpfile[PATH_MAX];
2104 unsigned char real_sha1[20];
2105 unsigned char discard[4096];
2109 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2111 local = mkstemp(tmpfile);
2114 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2116 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2119 memset(&stream, 0, sizeof(stream));
2121 inflateInit(&stream);
2128 stream.avail_in = *bufposn;
2129 stream.next_in = (unsigned char *) buffer;
2131 stream.next_out = discard;
2132 stream.avail_out = sizeof(discard);
2133 ret = inflate(&stream, Z_SYNC_FLUSH);
2134 SHA1_Update(&c, discard, sizeof(discard) -
2136 } while (stream.avail_in && ret == Z_OK);
2137 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2138 die("unable to write sha1 file");
2139 memmove(buffer, buffer + *bufposn - stream.avail_in,
2141 *bufposn = stream.avail_in;
2145 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2150 return error("Connection closed?");
2151 perror("Reading from connection");
2156 inflateEnd(&stream);
2158 fchmod(local, 0444);
2159 if (close(local) != 0)
2160 die("unable to write sha1 file");
2161 SHA1_Final(real_sha1, &c);
2162 if (ret != Z_STREAM_END) {
2164 return error("File %s corrupted", sha1_to_hex(sha1));
2166 if (hashcmp(sha1, real_sha1)) {
2168 return error("File %s has bad hash", sha1_to_hex(sha1));
2171 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2174 int has_pack_index(const unsigned char *sha1)
2177 if (stat(sha1_pack_index_name(sha1), &st))
2182 int has_pack_file(const unsigned char *sha1)
2185 if (stat(sha1_pack_name(sha1), &st))
2190 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2192 struct pack_entry e;
2193 return find_pack_entry(sha1, &e, ignore_packed);
2196 int has_sha1_file(const unsigned char *sha1)
2199 struct pack_entry e;
2201 if (find_pack_entry(sha1, &e, NULL))
2203 return find_sha1_file(sha1, &st) ? 1 : 0;
2207 * reads from fd as long as possible into a supplied buffer of size bytes.
2208 * If necessary the buffer's size is increased using realloc()
2210 * returns 0 if anything went fine and -1 otherwise
2212 * NOTE: both buf and size may change, but even when -1 is returned
2213 * you still have to free() it yourself.
2215 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2217 char* buf = *return_buf;
2218 unsigned long size = *return_size;
2220 unsigned long off = 0;
2223 iret = xread(fd, buf + off, size - off);
2228 buf = xrealloc(buf, size);
2241 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2243 unsigned long size = 4096;
2244 char *buf = xmalloc(size);
2247 if (read_pipe(fd, &buf, &size)) {
2255 ret = write_sha1_file(buf, size, type, sha1);
2257 ret = hash_sha1_file(buf, size, type, sha1);
2262 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2263 enum object_type type, const char *path)
2265 size_t size = xsize_t(st->st_size);
2267 int ret, re_allocated = 0;
2270 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2277 * Convert blobs to git internal format
2279 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2280 unsigned long nsize = size;
2282 if (convert_to_git(path, &nbuf, &nsize)) {
2292 ret = write_sha1_file(buf, size, typename(type), sha1);
2294 ret = hash_sha1_file(buf, size, typename(type), sha1);
2304 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2310 switch (st->st_mode & S_IFMT) {
2312 fd = open(path, O_RDONLY);
2314 return error("open(\"%s\"): %s", path,
2316 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2317 return error("%s: failed to insert into database",
2321 len = xsize_t(st->st_size);
2322 target = xmalloc(len + 1);
2323 if (readlink(path, target, len + 1) != st->st_size) {
2324 char *errstr = strerror(errno);
2326 return error("readlink(\"%s\"): %s", path,
2330 hash_sha1_file(target, len, blob_type, sha1);
2331 else if (write_sha1_file(target, len, blob_type, sha1))
2332 return error("%s: failed to insert into database",
2337 return resolve_gitlink_ref(path, "HEAD", sha1);
2339 return error("%s: unsupported file type", path);
2344 int read_pack_header(int fd, struct pack_header *header)
2346 char *c = (char*)header;
2347 ssize_t remaining = sizeof(struct pack_header);
2349 ssize_t r = xread(fd, c, remaining);
2351 /* "eof before pack header was fully read" */
2352 return PH_ERROR_EOF;
2355 } while (remaining > 0);
2356 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2357 /* "protocol error (pack signature mismatch detected)" */
2358 return PH_ERROR_PACK_SIGNATURE;
2359 if (!pack_version_ok(header->hdr_version))
2360 /* "protocol error (pack version unsupported)" */
2361 return PH_ERROR_PROTOCOL;