2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
33 static size_t sz_fmt(size_t s) { return s; }
36 const unsigned char null_sha1[20];
38 static unsigned int sha1_file_open_flag = O_NOATIME;
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
88 int safe_create_leading_directories(char *path)
93 if (is_absolute_path(path))
97 pos = strchr(pos, '/');
101 if (!stat(path, &st)) {
103 if (!S_ISDIR(st.st_mode)) {
108 else if (mkdir(path, 0777)) {
112 else if (adjust_shared_perm(path)) {
121 char * sha1_to_hex(const unsigned char *sha1)
124 static char hexbuffer[4][50];
125 static const char hex[] = "0123456789abcdef";
126 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
129 for (i = 0; i < 20; i++) {
130 unsigned int val = *sha1++;
131 *buf++ = hex[val >> 4];
132 *buf++ = hex[val & 0xf];
139 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
142 for (i = 0; i < 20; i++) {
143 static char hex[] = "0123456789abcdef";
144 unsigned int val = sha1[i];
145 char *pos = pathbuf + i*2 + (i > 0);
146 *pos++ = hex[val >> 4];
147 *pos = hex[val & 0xf];
152 * NOTE! This returns a statically allocated buffer, so you have to be
153 * careful about using it. Do an "xstrdup()" if you need to save the
156 * Also note that this returns the location for creating. Reading
157 * SHA1 file can happen from any alternate directory listed in the
158 * DB_ENVIRONMENT environment variable if it is not found in
159 * the primary object database.
161 char *sha1_file_name(const unsigned char *sha1)
163 static char *name, *base;
166 const char *sha1_file_directory = get_object_directory();
167 int len = strlen(sha1_file_directory);
168 base = xmalloc(len + 60);
169 memcpy(base, sha1_file_directory, len);
170 memset(base+len, 0, 60);
173 name = base + len + 1;
175 fill_sha1_path(name, sha1);
179 static char *sha1_get_pack_name(const unsigned char *sha1,
180 char **name, char **base)
182 static const char hex[] = "0123456789abcdef";
187 const char *sha1_file_directory = get_object_directory();
188 int len = strlen(sha1_file_directory);
189 *base = xmalloc(len + 60);
190 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
191 *name = *base + len + 11;
196 for (i = 0; i < 20; i++) {
197 unsigned int val = *sha1++;
198 *buf++ = hex[val >> 4];
199 *buf++ = hex[val & 0xf];
205 char *sha1_pack_name(const unsigned char *sha1)
207 static char *name, *base;
209 return sha1_get_pack_name(sha1, &name, &base);
212 char *sha1_pack_index_name(const unsigned char *sha1)
214 static char *name, *base;
216 return sha1_get_pack_name(sha1, &name, &base);
219 struct alternate_object_database *alt_odb_list;
220 static struct alternate_object_database **alt_odb_tail;
222 static void read_info_alternates(const char * alternates, int depth);
225 * Prepare alternate object database registry.
227 * The variable alt_odb_list points at the list of struct
228 * alternate_object_database. The elements on this list come from
229 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
230 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
231 * whose contents is similar to that environment variable but can be
232 * LF separated. Its base points at a statically allocated buffer that
233 * contains "/the/directory/corresponding/to/.git/objects/...", while
234 * its name points just after the slash at the end of ".git/objects/"
235 * in the example above, and has enough space to hold 40-byte hex
236 * SHA1, an extra slash for the first level indirection, and the
239 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
242 const char *objdir = get_object_directory();
243 struct alternate_object_database *ent;
244 struct alternate_object_database *alt;
245 /* 43 = 40-byte + 2 '/' + terminating NUL */
247 int entlen = pfxlen + 43;
250 if (!is_absolute_path(entry) && relative_base) {
251 /* Relative alt-odb */
253 base_len = strlen(relative_base) + 1;
257 ent = xmalloc(sizeof(*ent) + entlen);
259 if (!is_absolute_path(entry) && relative_base) {
260 memcpy(ent->base, relative_base, base_len - 1);
261 ent->base[base_len - 1] = '/';
262 memcpy(ent->base + base_len, entry, len);
265 memcpy(ent->base, entry, pfxlen);
267 ent->name = ent->base + pfxlen + 1;
268 ent->base[pfxlen + 3] = '/';
269 ent->base[pfxlen] = ent->base[entlen-1] = 0;
271 /* Detect cases where alternate disappeared */
272 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
273 error("object directory %s does not exist; "
274 "check .git/objects/info/alternates.",
280 /* Prevent the common mistake of listing the same
281 * thing twice, or object directory itself.
283 for (alt = alt_odb_list; alt; alt = alt->next) {
284 if (!memcmp(ent->base, alt->base, pfxlen)) {
289 if (!memcmp(ent->base, objdir, pfxlen)) {
294 /* add the alternate entry */
296 alt_odb_tail = &(ent->next);
299 /* recursively add alternates */
300 read_info_alternates(ent->base, depth + 1);
302 ent->base[pfxlen] = '/';
307 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
308 const char *relative_base, int depth)
310 const char *cp, *last;
313 error("%s: ignoring alternate object stores, nesting too deep.",
321 if (cp < ep && *cp == '#') {
322 while (cp < ep && *cp != sep)
327 while (cp < ep && *cp != sep)
330 if (!is_absolute_path(last) && depth) {
331 error("%s: ignoring relative alternate object store %s",
332 relative_base, last);
334 link_alt_odb_entry(last, cp - last,
335 relative_base, depth);
338 while (cp < ep && *cp == sep)
344 static void read_info_alternates(const char * relative_base, int depth)
349 const char alt_file_name[] = "info/alternates";
350 /* Given that relative_base is no longer than PATH_MAX,
351 ensure that "path" has enough space to append "/", the
352 file name, "info/alternates", and a trailing NUL. */
353 char path[PATH_MAX + 1 + sizeof alt_file_name];
356 sprintf(path, "%s/%s", relative_base, alt_file_name);
357 fd = open(path, O_RDONLY);
360 if (fstat(fd, &st) || (st.st_size == 0)) {
364 mapsz = xsize_t(st.st_size);
365 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
368 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
373 void prepare_alt_odb(void)
380 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
386 read_info_alternates(get_object_directory(), 0);
389 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
391 char *name = sha1_file_name(sha1);
392 struct alternate_object_database *alt;
397 for (alt = alt_odb_list; alt; alt = alt->next) {
399 fill_sha1_path(name, sha1);
400 if (!stat(alt->base, st))
406 static unsigned int pack_used_ctr;
407 static unsigned int pack_mmap_calls;
408 static unsigned int peak_pack_open_windows;
409 static unsigned int pack_open_windows;
410 static size_t peak_pack_mapped;
411 static size_t pack_mapped;
412 struct packed_git *packed_git;
414 void pack_report(void)
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
420 sz_fmt(getpagesize()),
421 sz_fmt(packed_git_window_size),
422 sz_fmt(packed_git_limit));
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
431 pack_open_windows, peak_pack_open_windows,
432 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
435 static int check_packed_git_idx(const char *path, struct packed_git *p)
438 struct pack_idx_header *hdr;
440 uint32_t version, nr, i, *index;
441 int fd = open(path, O_RDONLY);
446 if (fstat(fd, &st)) {
450 idx_size = xsize_t(st.st_size);
451 if (idx_size < 4 * 256 + 20 + 20) {
453 return error("index file %s is too small", path);
455 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
459 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
460 version = ntohl(hdr->idx_version);
461 if (version < 2 || version > 2) {
462 munmap(idx_map, idx_size);
463 return error("index file %s is version %d"
464 " and is not supported by this binary"
465 " (try upgrading GIT to a newer version)",
474 index += 2; /* skip index header */
475 for (i = 0; i < 256; i++) {
476 uint32_t n = ntohl(index[i]);
478 munmap(idx_map, idx_size);
479 return error("non-monotonic index %s", path);
487 * - 256 index entries 4 bytes each
488 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
489 * - 20-byte SHA1 of the packfile
490 * - 20-byte SHA1 file checksum
492 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
493 munmap(idx_map, idx_size);
494 return error("wrong index v1 file size in %s", path);
496 } else if (version == 2) {
499 * - 8 bytes of header
500 * - 256 index entries 4 bytes each
501 * - 20-byte sha1 entry * nr
502 * - 4-byte crc entry * nr
503 * - 4-byte offset entry * nr
504 * - 20-byte SHA1 of the packfile
505 * - 20-byte SHA1 file checksum
506 * And after the 4-byte offset table might be a
507 * variable sized table containing 8-byte entries
508 * for offsets larger than 2^31.
510 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
511 unsigned long max_size = min_size;
513 max_size += (nr - 1)*8;
514 if (idx_size < min_size || idx_size > max_size) {
515 munmap(idx_map, idx_size);
516 return error("wrong index v2 file size in %s", path);
518 if (idx_size != min_size &&
520 * make sure we can deal with large pack offsets.
521 * 31-bit signed offset won't be enough, neither
522 * 32-bit unsigned one will be.
524 (sizeof(off_t) <= 4)) {
525 munmap(idx_map, idx_size);
526 return error("pack too large for current definition of off_t in %s", path);
530 p->index_version = version;
531 p->index_data = idx_map;
532 p->index_size = idx_size;
537 int open_pack_index(struct packed_git *p)
545 idx_name = xstrdup(p->pack_name);
546 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
547 ret = check_packed_git_idx(idx_name, p);
552 static void scan_windows(struct packed_git *p,
553 struct packed_git **lru_p,
554 struct pack_window **lru_w,
555 struct pack_window **lru_l)
557 struct pack_window *w, *w_l;
559 for (w_l = NULL, w = p->windows; w; w = w->next) {
561 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
571 static int unuse_one_window(struct packed_git *current, int keep_fd)
573 struct packed_git *p, *lru_p = NULL;
574 struct pack_window *lru_w = NULL, *lru_l = NULL;
577 scan_windows(current, &lru_p, &lru_w, &lru_l);
578 for (p = packed_git; p; p = p->next)
579 scan_windows(p, &lru_p, &lru_w, &lru_l);
581 munmap(lru_w->base, lru_w->len);
582 pack_mapped -= lru_w->len;
584 lru_l->next = lru_w->next;
586 lru_p->windows = lru_w->next;
587 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
588 close(lru_p->pack_fd);
599 void release_pack_memory(size_t need, int fd)
601 size_t cur = pack_mapped;
602 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
606 void close_pack_windows(struct packed_git *p)
609 struct pack_window *w = p->windows;
612 die("pack '%s' still has open windows to it",
614 munmap(w->base, w->len);
615 pack_mapped -= w->len;
617 p->windows = w->next;
622 void unuse_pack(struct pack_window **w_cursor)
624 struct pack_window *w = *w_cursor;
632 * Do not call this directly as this leaks p->pack_fd on error return;
633 * call open_packed_git() instead.
635 static int open_packed_git_1(struct packed_git *p)
638 struct pack_header hdr;
639 unsigned char sha1[20];
640 unsigned char *idx_sha1;
643 if (!p->index_data && open_pack_index(p))
644 return error("packfile %s index unavailable", p->pack_name);
646 p->pack_fd = open(p->pack_name, O_RDONLY);
647 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
650 /* If we created the struct before we had the pack we lack size. */
652 if (!S_ISREG(st.st_mode))
653 return error("packfile %s not a regular file", p->pack_name);
654 p->pack_size = st.st_size;
655 } else if (p->pack_size != st.st_size)
656 return error("packfile %s size changed", p->pack_name);
658 /* We leave these file descriptors open with sliding mmap;
659 * there is no point keeping them open across exec(), though.
661 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
663 return error("cannot determine file descriptor flags");
664 fd_flag |= FD_CLOEXEC;
665 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
666 return error("cannot set FD_CLOEXEC");
668 /* Verify we recognize this pack file format. */
669 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
670 return error("file %s is far too short to be a packfile", p->pack_name);
671 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
672 return error("file %s is not a GIT packfile", p->pack_name);
673 if (!pack_version_ok(hdr.hdr_version))
674 return error("packfile %s is version %u and not supported"
675 " (try upgrading GIT to a newer version)",
676 p->pack_name, ntohl(hdr.hdr_version));
678 /* Verify the pack matches its index. */
679 if (p->num_objects != ntohl(hdr.hdr_entries))
680 return error("packfile %s claims to have %u objects"
681 " while index indicates %u objects",
682 p->pack_name, ntohl(hdr.hdr_entries),
684 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
685 return error("end of packfile %s is unavailable", p->pack_name);
686 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
687 return error("packfile %s signature is unavailable", p->pack_name);
688 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
689 if (hashcmp(sha1, idx_sha1))
690 return error("packfile %s does not match index", p->pack_name);
694 static int open_packed_git(struct packed_git *p)
696 if (!open_packed_git_1(p))
698 if (p->pack_fd != -1) {
705 static int in_window(struct pack_window *win, off_t offset)
707 /* We must promise at least 20 bytes (one hash) after the
708 * offset is available from this window, otherwise the offset
709 * is not actually in this window and a different window (which
710 * has that one hash excess) must be used. This is to support
711 * the object header and delta base parsing routines below.
713 off_t win_off = win->offset;
714 return win_off <= offset
715 && (offset + 20) <= (win_off + win->len);
718 unsigned char* use_pack(struct packed_git *p,
719 struct pack_window **w_cursor,
723 struct pack_window *win = *w_cursor;
725 if (p->pack_fd == -1 && open_packed_git(p))
726 die("packfile %s cannot be accessed", p->pack_name);
728 /* Since packfiles end in a hash of their content and its
729 * pointless to ask for an offset into the middle of that
730 * hash, and the in_window function above wouldn't match
731 * don't allow an offset too close to the end of the file.
733 if (offset > (p->pack_size - 20))
734 die("offset beyond end of packfile (truncated pack?)");
736 if (!win || !in_window(win, offset)) {
739 for (win = p->windows; win; win = win->next) {
740 if (in_window(win, offset))
744 size_t window_align = packed_git_window_size / 2;
746 win = xcalloc(1, sizeof(*win));
747 win->offset = (offset / window_align) * window_align;
748 len = p->pack_size - win->offset;
749 if (len > packed_git_window_size)
750 len = packed_git_window_size;
751 win->len = (size_t)len;
752 pack_mapped += win->len;
753 while (packed_git_limit < pack_mapped
754 && unuse_one_window(p, p->pack_fd))
756 win->base = xmmap(NULL, win->len,
757 PROT_READ, MAP_PRIVATE,
758 p->pack_fd, win->offset);
759 if (win->base == MAP_FAILED)
760 die("packfile %s cannot be mapped: %s",
765 if (pack_mapped > peak_pack_mapped)
766 peak_pack_mapped = pack_mapped;
767 if (pack_open_windows > peak_pack_open_windows)
768 peak_pack_open_windows = pack_open_windows;
769 win->next = p->windows;
773 if (win != *w_cursor) {
774 win->last_used = pack_used_ctr++;
778 offset -= win->offset;
780 *left = win->len - xsize_t(offset);
781 return win->base + offset;
784 struct packed_git *add_packed_git(const char *path, int path_len, int local)
787 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
790 * Make sure a corresponding .pack file exists and that
791 * the index looks sane.
793 path_len -= strlen(".idx");
796 memcpy(p->pack_name, path, path_len);
797 strcpy(p->pack_name + path_len, ".pack");
798 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
803 /* ok, it looks sane as far as we can check without
804 * actually mapping the pack file.
806 p->index_version = 0;
807 p->index_data = NULL;
810 p->pack_size = st.st_size;
814 p->pack_local = local;
815 p->mtime = st.st_mtime;
816 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
821 struct packed_git *parse_pack_index(unsigned char *sha1)
823 char *path = sha1_pack_index_name(sha1);
824 return parse_pack_index_file(sha1, path);
827 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
828 const char *idx_path)
830 const char *path = sha1_pack_name(sha1);
831 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
833 if (check_packed_git_idx(idx_path, p)) {
838 strcpy(p->pack_name, path);
843 hashcpy(p->sha1, sha1);
847 void install_packed_git(struct packed_git *pack)
849 pack->next = packed_git;
853 static void prepare_packed_git_one(char *objdir, int local)
855 /* Ensure that this buffer is large enough so that we can
856 append "/pack/" without clobbering the stack even if
857 strlen(objdir) were PATH_MAX. */
858 char path[PATH_MAX + 1 + 4 + 1 + 1];
863 sprintf(path, "%s/pack", objdir);
868 error("unable to open object pack directory: %s: %s",
869 path, strerror(errno));
873 while ((de = readdir(dir)) != NULL) {
874 int namelen = strlen(de->d_name);
875 struct packed_git *p;
877 if (!has_extension(de->d_name, ".idx"))
880 if (len + namelen + 1 > sizeof(path))
883 /* Don't reopen a pack we already have. */
884 strcpy(path + len, de->d_name);
885 for (p = packed_git; p; p = p->next) {
886 if (!memcmp(path, p->pack_name, len + namelen - 4))
891 /* See if it really is a valid .idx file with corresponding
892 * .pack file that we can map.
894 p = add_packed_git(path, len + namelen, local);
897 install_packed_git(p);
902 static int sort_pack(const void *a_, const void *b_)
904 struct packed_git *a = *((struct packed_git **)a_);
905 struct packed_git *b = *((struct packed_git **)b_);
909 * Local packs tend to contain objects specific to our
910 * variant of the project than remote ones. In addition,
911 * remote ones could be on a network mounted filesystem.
912 * Favor local ones for these reasons.
914 st = a->pack_local - b->pack_local;
919 * Younger packs tend to contain more recent objects,
920 * and more recent objects tend to get accessed more
923 if (a->mtime < b->mtime)
925 else if (a->mtime == b->mtime)
930 static void rearrange_packed_git(void)
932 struct packed_git **ary, *p;
935 for (n = 0, p = packed_git; p; p = p->next)
940 /* prepare an array of packed_git for easier sorting */
941 ary = xcalloc(n, sizeof(struct packed_git *));
942 for (n = 0, p = packed_git; p; p = p->next)
945 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
947 /* link them back again */
948 for (i = 0; i < n - 1; i++)
949 ary[i]->next = ary[i + 1];
950 ary[n - 1]->next = NULL;
956 static int prepare_packed_git_run_once = 0;
957 void prepare_packed_git(void)
959 struct alternate_object_database *alt;
961 if (prepare_packed_git_run_once)
963 prepare_packed_git_one(get_object_directory(), 1);
965 for (alt = alt_odb_list; alt; alt = alt->next) {
967 prepare_packed_git_one(alt->base, 0);
970 rearrange_packed_git();
971 prepare_packed_git_run_once = 1;
974 void reprepare_packed_git(void)
976 prepare_packed_git_run_once = 0;
977 prepare_packed_git();
980 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
982 unsigned char real_sha1[20];
983 hash_sha1_file(map, size, type, real_sha1);
984 return hashcmp(sha1, real_sha1) ? -1 : 0;
987 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
992 char *filename = find_sha1_file(sha1, &st);
998 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1000 /* See if it works without O_NOATIME */
1001 switch (sha1_file_open_flag) {
1003 fd = open(filename, O_RDONLY);
1011 /* If it failed once, it will probably fail again.
1012 * Stop using O_NOATIME
1014 sha1_file_open_flag = 0;
1016 *size = xsize_t(st.st_size);
1017 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1022 static int legacy_loose_object(unsigned char *map)
1027 * Is it a zlib-compressed buffer? If so, the first byte
1028 * must be 0x78 (15-bit window size, deflated), and the
1029 * first 16-bit word is evenly divisible by 31
1031 word = (map[0] << 8) + map[1];
1032 if (map[0] == 0x78 && !(word % 31))
1038 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1043 unsigned long used = 0;
1046 *type = (c >> 4) & 7;
1052 if (sizeof(long) * 8 <= shift)
1055 size += (c & 0x7f) << shift;
1062 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1064 unsigned long size, used;
1065 static const char valid_loose_object_type[8] = {
1067 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1068 0, /* "delta" and others are invalid in a loose object */
1070 enum object_type type;
1072 /* Get the data stream */
1073 memset(stream, 0, sizeof(*stream));
1074 stream->next_in = map;
1075 stream->avail_in = mapsize;
1076 stream->next_out = buffer;
1077 stream->avail_out = bufsiz;
1079 if (legacy_loose_object(map)) {
1080 inflateInit(stream);
1081 return inflate(stream, 0);
1086 * There used to be a second loose object header format which
1087 * was meant to mimic the in-pack format, allowing for direct
1088 * copy of the object data. This format turned up not to be
1089 * really worth it and we don't write it any longer. But we
1090 * can still read it.
1092 used = unpack_object_header_gently(map, mapsize, &type, &size);
1093 if (!used || !valid_loose_object_type[type])
1098 /* Set up the stream for the rest.. */
1099 stream->next_in = map;
1100 stream->avail_in = mapsize;
1101 inflateInit(stream);
1103 /* And generate the fake traditional header */
1104 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1105 typename(type), size);
1109 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1111 int bytes = strlen(buffer) + 1;
1112 unsigned char *buf = xmalloc(1+size);
1116 n = stream->total_out - bytes;
1119 memcpy(buf, (char *) buffer + bytes, n);
1121 if (bytes <= size) {
1123 * The above condition must be (bytes <= size), not
1124 * (bytes < size). In other words, even though we
1125 * expect no more output and set avail_out to zer0,
1126 * the input zlib stream may have bytes that express
1127 * "this concludes the stream", and we *do* want to
1130 * Otherwise we would not be able to test that we
1131 * consumed all the input to reach the expected size;
1132 * we also want to check that zlib tells us that all
1133 * went well with status == Z_STREAM_END at the end.
1135 stream->next_out = buf + bytes;
1136 stream->avail_out = size - bytes;
1137 while (status == Z_OK)
1138 status = inflate(stream, Z_FINISH);
1141 if (status == Z_STREAM_END && !stream->avail_in) {
1147 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1148 else if (stream->avail_in)
1149 error("garbage at end of loose object '%s'",
1156 * We used to just use "sscanf()", but that's actually way
1157 * too permissive for what we want to check. So do an anal
1158 * object header parse by hand.
1160 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1167 * The type can be at most ten bytes (including the
1168 * terminating '\0' that we add), and is followed by
1177 if (i >= sizeof(type))
1183 * The length must follow immediately, and be in canonical
1184 * decimal format (ie "010" is not valid).
1186 size = *hdr++ - '0';
1191 unsigned long c = *hdr - '0';
1195 size = size * 10 + c;
1201 * The length must be followed by a zero byte
1203 return *hdr ? -1 : type_from_string(type);
1206 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1212 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1213 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1216 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1219 unsigned long get_size_from_delta(struct packed_git *p,
1220 struct pack_window **w_curs,
1223 const unsigned char *data;
1224 unsigned char delta_head[20], *in;
1228 memset(&stream, 0, sizeof(stream));
1229 stream.next_out = delta_head;
1230 stream.avail_out = sizeof(delta_head);
1232 inflateInit(&stream);
1234 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1235 stream.next_in = in;
1236 st = inflate(&stream, Z_FINISH);
1237 curpos += stream.next_in - in;
1238 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1239 stream.total_out < sizeof(delta_head));
1240 inflateEnd(&stream);
1241 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1242 die("delta data unpack-initial failed");
1244 /* Examine the initial part of the delta to figure out
1249 /* ignore base size */
1250 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1252 /* Read the result size */
1253 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1256 static off_t get_delta_base(struct packed_git *p,
1257 struct pack_window **w_curs,
1259 enum object_type type,
1260 off_t delta_obj_offset)
1262 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1265 /* use_pack() assured us we have [base_info, base_info + 20)
1266 * as a range that we can look at without walking off the
1267 * end of the mapped window. Its actually the hash size
1268 * that is assured. An OFS_DELTA longer than the hash size
1269 * is stupid, as then a REF_DELTA would be smaller to store.
1271 if (type == OBJ_OFS_DELTA) {
1273 unsigned char c = base_info[used++];
1274 base_offset = c & 127;
1277 if (!base_offset || MSB(base_offset, 7))
1278 die("offset value overflow for delta base object");
1279 c = base_info[used++];
1280 base_offset = (base_offset << 7) + (c & 127);
1282 base_offset = delta_obj_offset - base_offset;
1283 if (base_offset >= delta_obj_offset)
1284 die("delta base offset out of bound");
1286 } else if (type == OBJ_REF_DELTA) {
1287 /* The base entry _must_ be in the same pack */
1288 base_offset = find_pack_entry_one(base_info, p);
1290 die("failed to find delta-pack base object %s",
1291 sha1_to_hex(base_info));
1294 die("I am totally screwed");
1298 /* forward declaration for a mutually recursive function */
1299 static int packed_object_info(struct packed_git *p, off_t offset,
1300 unsigned long *sizep);
1302 static int packed_delta_info(struct packed_git *p,
1303 struct pack_window **w_curs,
1305 enum object_type type,
1307 unsigned long *sizep)
1311 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1312 type = packed_object_info(p, base_offset, NULL);
1314 /* We choose to only get the type of the base object and
1315 * ignore potentially corrupt pack file that expects the delta
1316 * based on a base with a wrong size. This saves tons of
1320 *sizep = get_size_from_delta(p, w_curs, curpos);
1325 static int unpack_object_header(struct packed_git *p,
1326 struct pack_window **w_curs,
1328 unsigned long *sizep)
1330 unsigned char *base;
1333 enum object_type type;
1335 /* use_pack() assures us we have [base, base + 20) available
1336 * as a range that we can look at at. (Its actually the hash
1337 * size that is assured.) With our object header encoding
1338 * the maximum deflated object size is 2^137, which is just
1339 * insane, so we know won't exceed what we have been given.
1341 base = use_pack(p, w_curs, *curpos, &left);
1342 used = unpack_object_header_gently(base, left, &type, sizep);
1344 die("object offset outside of pack file");
1350 const char *packed_object_info_detail(struct packed_git *p,
1352 unsigned long *size,
1353 unsigned long *store_size,
1354 unsigned int *delta_chain_length,
1355 unsigned char *base_sha1)
1357 struct pack_window *w_curs = NULL;
1359 unsigned long dummy;
1360 unsigned char *next_sha1;
1361 enum object_type type;
1362 struct revindex_entry *revidx;
1364 *delta_chain_length = 0;
1365 curpos = obj_offset;
1366 type = unpack_object_header(p, &w_curs, &curpos, size);
1368 revidx = find_pack_revindex(p, obj_offset);
1369 *store_size = revidx[1].offset - obj_offset;
1374 die("pack %s contains unknown object type %d",
1375 p->pack_name, type);
1380 unuse_pack(&w_curs);
1381 return typename(type);
1383 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1384 if (*delta_chain_length == 0) {
1385 revidx = find_pack_revindex(p, obj_offset);
1386 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1390 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1391 if (*delta_chain_length == 0)
1392 hashcpy(base_sha1, next_sha1);
1393 obj_offset = find_pack_entry_one(next_sha1, p);
1396 (*delta_chain_length)++;
1397 curpos = obj_offset;
1398 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1402 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1403 unsigned long *sizep)
1405 struct pack_window *w_curs = NULL;
1407 off_t curpos = obj_offset;
1408 enum object_type type;
1410 type = unpack_object_header(p, &w_curs, &curpos, &size);
1415 type = packed_delta_info(p, &w_curs, curpos,
1416 type, obj_offset, sizep);
1426 die("pack %s contains unknown object type %d",
1427 p->pack_name, type);
1429 unuse_pack(&w_curs);
1433 static void *unpack_compressed_entry(struct packed_git *p,
1434 struct pack_window **w_curs,
1440 unsigned char *buffer, *in;
1442 buffer = xmalloc(size + 1);
1444 memset(&stream, 0, sizeof(stream));
1445 stream.next_out = buffer;
1446 stream.avail_out = size;
1448 inflateInit(&stream);
1450 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1451 stream.next_in = in;
1452 st = inflate(&stream, Z_FINISH);
1453 curpos += stream.next_in - in;
1454 } while (st == Z_OK || st == Z_BUF_ERROR);
1455 inflateEnd(&stream);
1456 if ((st != Z_STREAM_END) || stream.total_out != size) {
1464 #define MAX_DELTA_CACHE (256)
1466 static size_t delta_base_cached;
1468 static struct delta_base_cache_lru_list {
1469 struct delta_base_cache_lru_list *prev;
1470 struct delta_base_cache_lru_list *next;
1471 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1473 static struct delta_base_cache_entry {
1474 struct delta_base_cache_lru_list lru;
1476 struct packed_git *p;
1479 enum object_type type;
1480 } delta_base_cache[MAX_DELTA_CACHE];
1482 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1486 hash = (unsigned long)p + (unsigned long)base_offset;
1487 hash += (hash >> 8) + (hash >> 16);
1488 return hash % MAX_DELTA_CACHE;
1491 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1492 unsigned long *base_size, enum object_type *type, int keep_cache)
1495 unsigned long hash = pack_entry_hash(p, base_offset);
1496 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1499 if (ret && ent->p == p && ent->base_offset == base_offset)
1500 goto found_cache_entry;
1501 return unpack_entry(p, base_offset, type, base_size);
1506 ent->lru.next->prev = ent->lru.prev;
1507 ent->lru.prev->next = ent->lru.next;
1508 delta_base_cached -= ent->size;
1510 ret = xmemdupz(ent->data, ent->size);
1513 *base_size = ent->size;
1517 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1522 ent->lru.next->prev = ent->lru.prev;
1523 ent->lru.prev->next = ent->lru.next;
1524 delta_base_cached -= ent->size;
1528 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1529 void *base, unsigned long base_size, enum object_type type)
1531 unsigned long hash = pack_entry_hash(p, base_offset);
1532 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1533 struct delta_base_cache_lru_list *lru;
1535 release_delta_base_cache(ent);
1536 delta_base_cached += base_size;
1538 for (lru = delta_base_cache_lru.next;
1539 delta_base_cached > delta_base_cache_limit
1540 && lru != &delta_base_cache_lru;
1542 struct delta_base_cache_entry *f = (void *)lru;
1543 if (f->type == OBJ_BLOB)
1544 release_delta_base_cache(f);
1546 for (lru = delta_base_cache_lru.next;
1547 delta_base_cached > delta_base_cache_limit
1548 && lru != &delta_base_cache_lru;
1550 struct delta_base_cache_entry *f = (void *)lru;
1551 release_delta_base_cache(f);
1555 ent->base_offset = base_offset;
1558 ent->size = base_size;
1559 ent->lru.next = &delta_base_cache_lru;
1560 ent->lru.prev = delta_base_cache_lru.prev;
1561 delta_base_cache_lru.prev->next = &ent->lru;
1562 delta_base_cache_lru.prev = &ent->lru;
1565 static void *unpack_delta_entry(struct packed_git *p,
1566 struct pack_window **w_curs,
1568 unsigned long delta_size,
1570 enum object_type *type,
1571 unsigned long *sizep)
1573 void *delta_data, *result, *base;
1574 unsigned long base_size;
1577 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1578 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1580 die("failed to read delta base object"
1581 " at %"PRIuMAX" from %s",
1582 (uintmax_t)base_offset, p->pack_name);
1584 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1586 die("failed to unpack compressed delta"
1587 " at %"PRIuMAX" from %s",
1588 (uintmax_t)curpos, p->pack_name);
1589 result = patch_delta(base, base_size,
1590 delta_data, delta_size,
1593 die("failed to apply delta");
1595 add_delta_base_cache(p, base_offset, base, base_size, *type);
1599 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1600 enum object_type *type, unsigned long *sizep)
1602 struct pack_window *w_curs = NULL;
1603 off_t curpos = obj_offset;
1606 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1610 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1611 obj_offset, type, sizep);
1617 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1620 die("unknown object type %i in %s", *type, p->pack_name);
1622 unuse_pack(&w_curs);
1626 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1629 const unsigned char *index = p->index_data;
1631 if (open_pack_index(p))
1633 index = p->index_data;
1635 if (n >= p->num_objects)
1638 if (p->index_version == 1) {
1639 return index + 24 * n + 4;
1642 return index + 20 * n;
1646 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1648 const unsigned char *index = p->index_data;
1650 if (p->index_version == 1) {
1651 return ntohl(*((uint32_t *)(index + 24 * n)));
1654 index += 8 + p->num_objects * (20 + 4);
1655 off = ntohl(*((uint32_t *)(index + 4 * n)));
1656 if (!(off & 0x80000000))
1658 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1659 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1660 ntohl(*((uint32_t *)(index + 4)));
1664 off_t find_pack_entry_one(const unsigned char *sha1,
1665 struct packed_git *p)
1667 const uint32_t *level1_ofs = p->index_data;
1668 const unsigned char *index = p->index_data;
1669 unsigned hi, lo, stride;
1670 static int use_lookup = -1;
1671 static int debug_lookup = -1;
1673 if (debug_lookup < 0)
1674 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1677 if (open_pack_index(p))
1679 level1_ofs = p->index_data;
1680 index = p->index_data;
1682 if (p->index_version > 1) {
1687 hi = ntohl(level1_ofs[*sha1]);
1688 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1689 if (p->index_version > 1) {
1697 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1698 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1701 use_lookup = !!getenv("GIT_USE_LOOKUP");
1703 int pos = sha1_entry_pos(index, stride, 0,
1704 lo, hi, p->num_objects, sha1);
1707 return nth_packed_object_offset(p, pos);
1711 unsigned mi = (lo + hi) / 2;
1712 int cmp = hashcmp(index + mi * stride, sha1);
1715 printf("lo %u hi %u rg %u mi %u\n",
1716 lo, hi, hi - lo, mi);
1718 return nth_packed_object_offset(p, mi);
1727 int matches_pack_name(struct packed_git *p, const char *name)
1729 const char *last_c, *c;
1731 if (!strcmp(p->pack_name, name))
1734 for (c = p->pack_name, last_c = c; *c;)
1739 if (!strcmp(last_c, name))
1745 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1747 static struct packed_git *last_found = (void *)1;
1748 struct packed_git *p;
1751 prepare_packed_git();
1754 p = (last_found == (void *)1) ? packed_git : last_found;
1757 if (ignore_packed) {
1759 for (ig = ignore_packed; *ig; ig++)
1760 if (matches_pack_name(p, *ig))
1766 offset = find_pack_entry_one(sha1, p);
1769 * We are about to tell the caller where they can
1770 * locate the requested object. We better make
1771 * sure the packfile is still here and can be
1772 * accessed before supplying that answer, as
1773 * it may have been deleted since the index
1776 if (p->pack_fd == -1 && open_packed_git(p)) {
1777 error("packfile %s cannot be accessed", p->pack_name);
1782 hashcpy(e->sha1, sha1);
1788 if (p == last_found)
1792 if (p == last_found)
1798 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1799 struct packed_git *packs)
1801 struct packed_git *p;
1803 for (p = packs; p; p = p->next) {
1804 if (find_pack_entry_one(sha1, p))
1811 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1814 unsigned long mapsize, size;
1819 map = map_sha1_file(sha1, &mapsize);
1821 return error("unable to find %s", sha1_to_hex(sha1));
1822 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1823 status = error("unable to unpack %s header",
1825 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1826 status = error("unable to parse %s header", sha1_to_hex(sha1));
1829 inflateEnd(&stream);
1830 munmap(map, mapsize);
1834 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1836 struct pack_entry e;
1838 if (!find_pack_entry(sha1, &e, NULL)) {
1839 reprepare_packed_git();
1840 if (!find_pack_entry(sha1, &e, NULL))
1841 return sha1_loose_object_info(sha1, sizep);
1843 return packed_object_info(e.p, e.offset, sizep);
1846 static void *read_packed_sha1(const unsigned char *sha1,
1847 enum object_type *type, unsigned long *size)
1849 struct pack_entry e;
1851 if (!find_pack_entry(sha1, &e, NULL))
1854 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1858 * This is meant to hold a *small* number of objects that you would
1859 * want read_sha1_file() to be able to return, but yet you do not want
1860 * to write them into the object store (e.g. a browse-only
1863 static struct cached_object {
1864 unsigned char sha1[20];
1865 enum object_type type;
1869 static int cached_object_nr, cached_object_alloc;
1871 static struct cached_object empty_tree = {
1872 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1873 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1874 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1880 static struct cached_object *find_cached_object(const unsigned char *sha1)
1883 struct cached_object *co = cached_objects;
1885 for (i = 0; i < cached_object_nr; i++, co++) {
1886 if (!hashcmp(co->sha1, sha1))
1889 if (!hashcmp(sha1, empty_tree.sha1))
1894 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1895 unsigned char *sha1)
1897 struct cached_object *co;
1899 hash_sha1_file(buf, len, typename(type), sha1);
1900 if (has_sha1_file(sha1) || find_cached_object(sha1))
1902 if (cached_object_alloc <= cached_object_nr) {
1903 cached_object_alloc = alloc_nr(cached_object_alloc);
1904 cached_objects = xrealloc(cached_objects,
1905 sizeof(*cached_objects) *
1906 cached_object_alloc);
1908 co = &cached_objects[cached_object_nr++];
1911 co->buf = xmalloc(len);
1912 memcpy(co->buf, buf, len);
1913 hashcpy(co->sha1, sha1);
1917 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1918 unsigned long *size)
1920 unsigned long mapsize;
1922 struct cached_object *co;
1924 co = find_cached_object(sha1);
1928 return xmemdupz(co->buf, co->size);
1931 buf = read_packed_sha1(sha1, type, size);
1934 map = map_sha1_file(sha1, &mapsize);
1936 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1937 munmap(map, mapsize);
1940 reprepare_packed_git();
1941 return read_packed_sha1(sha1, type, size);
1944 void *read_object_with_reference(const unsigned char *sha1,
1945 const char *required_type_name,
1946 unsigned long *size,
1947 unsigned char *actual_sha1_return)
1949 enum object_type type, required_type;
1951 unsigned long isize;
1952 unsigned char actual_sha1[20];
1954 required_type = type_from_string(required_type_name);
1955 hashcpy(actual_sha1, sha1);
1957 int ref_length = -1;
1958 const char *ref_type = NULL;
1960 buffer = read_sha1_file(actual_sha1, &type, &isize);
1963 if (type == required_type) {
1965 if (actual_sha1_return)
1966 hashcpy(actual_sha1_return, actual_sha1);
1969 /* Handle references */
1970 else if (type == OBJ_COMMIT)
1972 else if (type == OBJ_TAG)
1973 ref_type = "object ";
1978 ref_length = strlen(ref_type);
1980 if (ref_length + 40 > isize ||
1981 memcmp(buffer, ref_type, ref_length) ||
1982 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1987 /* Now we have the ID of the referred-to object in
1988 * actual_sha1. Check again. */
1992 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1993 const char *type, unsigned char *sha1,
1994 char *hdr, int *hdrlen)
1998 /* Generate the header */
1999 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2003 SHA1_Update(&c, hdr, *hdrlen);
2004 SHA1_Update(&c, buf, len);
2005 SHA1_Final(sha1, &c);
2009 * Link the tempfile to the final place, possibly creating the
2010 * last directory level as you do so.
2012 * Returns the errno on failure, 0 on success.
2014 static int link_temp_to_file(const char *tmpfile, const char *filename)
2019 if (!link(tmpfile, filename))
2023 * Try to mkdir the last path component if that failed.
2025 * Re-try the "link()" regardless of whether the mkdir
2026 * succeeds, since a race might mean that somebody
2030 dir = strrchr(filename, '/');
2033 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2038 if (!link(tmpfile, filename))
2046 * Move the just written object into its final resting place
2048 int move_temp_to_file(const char *tmpfile, const char *filename)
2050 int ret = link_temp_to_file(tmpfile, filename);
2053 * Coda hack - coda doesn't like cross-directory links,
2054 * so we fall back to a rename, which will mean that it
2055 * won't be able to check collisions, but that's not a
2058 * The same holds for FAT formatted media.
2060 * When this succeeds, we just return 0. We have nothing
2063 if (ret && ret != EEXIST) {
2064 if (!rename(tmpfile, filename))
2070 if (ret != EEXIST) {
2071 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2073 /* FIXME!!! Collision check here ? */
2079 static int write_buffer(int fd, const void *buf, size_t len)
2081 if (write_in_full(fd, buf, len) < 0)
2082 return error("file write error (%s)", strerror(errno));
2086 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2087 unsigned char *sha1)
2091 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2095 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2096 void *buf, unsigned long len, time_t mtime)
2099 unsigned char *compressed;
2102 static char tmpfile[PATH_MAX];
2104 filename = sha1_file_name(sha1);
2105 fd = open(filename, O_RDONLY);
2108 * FIXME!!! We might do collision checking here, but we'd
2109 * need to uncompress the old file and check it. Later.
2115 if (errno != ENOENT) {
2116 return error("sha1 file %s: %s\n", filename, strerror(errno));
2119 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2121 fd = mkstemp(tmpfile);
2124 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2126 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2130 memset(&stream, 0, sizeof(stream));
2131 deflateInit(&stream, zlib_compression_level);
2132 size = 8 + deflateBound(&stream, len+hdrlen);
2133 compressed = xmalloc(size);
2136 stream.next_out = compressed;
2137 stream.avail_out = size;
2139 /* First header.. */
2140 stream.next_in = (unsigned char *)hdr;
2141 stream.avail_in = hdrlen;
2142 while (deflate(&stream, 0) == Z_OK)
2145 /* Then the data itself.. */
2146 stream.next_in = buf;
2147 stream.avail_in = len;
2148 ret = deflate(&stream, Z_FINISH);
2149 if (ret != Z_STREAM_END)
2150 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2152 ret = deflateEnd(&stream);
2154 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2156 size = stream.total_out;
2158 if (write_buffer(fd, compressed, size) < 0)
2159 die("unable to write sha1 file");
2162 die("unable to write sha1 file");
2168 utb.modtime = mtime;
2169 if (utime(tmpfile, &utb) < 0)
2170 warning("failed utime() on %s: %s",
2171 tmpfile, strerror(errno));
2174 return move_temp_to_file(tmpfile, filename);
2177 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2179 unsigned char sha1[20];
2183 /* Normally if we have it in the pack then we do not bother writing
2184 * it out into .git/objects/??/?{38} file.
2186 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2188 hashcpy(returnsha1, sha1);
2189 if (has_sha1_file(sha1))
2191 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2194 int force_object_loose(const unsigned char *sha1, time_t mtime)
2199 enum object_type type;
2203 if (find_sha1_file(sha1, &st))
2205 buf = read_packed_sha1(sha1, &type, &len);
2207 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2208 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2209 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2213 * We need to unpack and recompress the object for writing
2214 * it out to a different file.
2216 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2220 unsigned char *unpacked;
2222 enum object_type type;
2227 /* need to unpack and recompress it by itself */
2228 unpacked = read_packed_sha1(sha1, &type, &len);
2230 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2232 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2235 memset(&stream, 0, sizeof(stream));
2236 deflateInit(&stream, zlib_compression_level);
2237 size = deflateBound(&stream, len + hdrlen);
2238 buf = xmalloc(size);
2241 stream.next_out = buf;
2242 stream.avail_out = size;
2244 /* First header.. */
2245 stream.next_in = (void *)hdr;
2246 stream.avail_in = hdrlen;
2247 while (deflate(&stream, 0) == Z_OK)
2250 /* Then the data itself.. */
2251 stream.next_in = unpacked;
2252 stream.avail_in = len;
2253 while (deflate(&stream, Z_FINISH) == Z_OK)
2255 deflateEnd(&stream);
2258 *objsize = stream.total_out;
2262 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2265 unsigned long objsize;
2266 void *buf = map_sha1_file(sha1, &objsize);
2269 retval = write_buffer(fd, buf, objsize);
2270 munmap(buf, objsize);
2274 buf = repack_object(sha1, &objsize);
2275 retval = write_buffer(fd, buf, objsize);
2280 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2281 size_t bufsize, size_t *bufposn)
2283 char tmpfile[PATH_MAX];
2286 unsigned char real_sha1[20];
2287 unsigned char discard[4096];
2291 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2293 local = mkstemp(tmpfile);
2296 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2298 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2301 memset(&stream, 0, sizeof(stream));
2303 inflateInit(&stream);
2310 stream.avail_in = *bufposn;
2311 stream.next_in = (unsigned char *) buffer;
2313 stream.next_out = discard;
2314 stream.avail_out = sizeof(discard);
2315 ret = inflate(&stream, Z_SYNC_FLUSH);
2316 SHA1_Update(&c, discard, sizeof(discard) -
2318 } while (stream.avail_in && ret == Z_OK);
2319 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2320 die("unable to write sha1 file");
2321 memmove(buffer, buffer + *bufposn - stream.avail_in,
2323 *bufposn = stream.avail_in;
2327 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2332 return error("Connection closed?");
2333 perror("Reading from connection");
2338 inflateEnd(&stream);
2340 fchmod(local, 0444);
2341 if (close(local) != 0)
2342 die("unable to write sha1 file");
2343 SHA1_Final(real_sha1, &c);
2344 if (ret != Z_STREAM_END) {
2346 return error("File %s corrupted", sha1_to_hex(sha1));
2348 if (hashcmp(sha1, real_sha1)) {
2350 return error("File %s has bad hash", sha1_to_hex(sha1));
2353 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2356 int has_pack_index(const unsigned char *sha1)
2359 if (stat(sha1_pack_index_name(sha1), &st))
2364 int has_pack_file(const unsigned char *sha1)
2367 if (stat(sha1_pack_name(sha1), &st))
2372 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2374 struct pack_entry e;
2375 return find_pack_entry(sha1, &e, ignore_packed);
2378 int has_sha1_file(const unsigned char *sha1)
2381 struct pack_entry e;
2383 if (find_pack_entry(sha1, &e, NULL))
2385 return find_sha1_file(sha1, &st) ? 1 : 0;
2388 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2393 strbuf_init(&buf, 0);
2394 if (strbuf_read(&buf, fd, 4096) < 0) {
2395 strbuf_release(&buf);
2402 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2404 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2405 strbuf_release(&buf);
2410 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2411 enum object_type type, const char *path)
2413 size_t size = xsize_t(st->st_size);
2415 int ret, re_allocated = 0;
2418 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2425 * Convert blobs to git internal format
2427 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2429 strbuf_init(&nbuf, 0);
2430 if (convert_to_git(path, buf, size, &nbuf,
2431 write_object ? safe_crlf : 0)) {
2433 buf = strbuf_detach(&nbuf, &size);
2439 ret = write_sha1_file(buf, size, typename(type), sha1);
2441 ret = hash_sha1_file(buf, size, typename(type), sha1);
2451 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2457 switch (st->st_mode & S_IFMT) {
2459 fd = open(path, O_RDONLY);
2461 return error("open(\"%s\"): %s", path,
2463 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2464 return error("%s: failed to insert into database",
2468 len = xsize_t(st->st_size);
2469 target = xmalloc(len + 1);
2470 if (readlink(path, target, len + 1) != st->st_size) {
2471 char *errstr = strerror(errno);
2473 return error("readlink(\"%s\"): %s", path,
2477 hash_sha1_file(target, len, blob_type, sha1);
2478 else if (write_sha1_file(target, len, blob_type, sha1))
2479 return error("%s: failed to insert into database",
2484 return resolve_gitlink_ref(path, "HEAD", sha1);
2486 return error("%s: unsupported file type", path);
2491 int read_pack_header(int fd, struct pack_header *header)
2493 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2494 /* "eof before pack header was fully read" */
2495 return PH_ERROR_EOF;
2497 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2498 /* "protocol error (pack signature mismatch detected)" */
2499 return PH_ERROR_PACK_SIGNATURE;
2500 if (!pack_version_ok(header->hdr_version))
2501 /* "protocol error (pack version unsupported)" */
2502 return PH_ERROR_PROTOCOL;