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 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
86 static inline int offset_1st_component(const char *path)
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
93 int safe_create_leading_directories(char *path)
95 char *pos = path + offset_1st_component(path);
99 pos = strchr(pos, '/');
103 if (!stat(path, &st)) {
105 if (!S_ISDIR(st.st_mode)) {
110 else if (mkdir(path, 0777)) {
114 else if (adjust_shared_perm(path)) {
123 int safe_create_leading_directories_const(const char *path)
125 /* path points to cache entries, so xstrdup before messing with it */
126 char *buf = xstrdup(path);
127 int result = safe_create_leading_directories(buf);
132 char *sha1_to_hex(const unsigned char *sha1)
135 static char hexbuffer[4][50];
136 static const char hex[] = "0123456789abcdef";
137 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
150 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
153 for (i = 0; i < 20; i++) {
154 static char hex[] = "0123456789abcdef";
155 unsigned int val = sha1[i];
156 char *pos = pathbuf + i*2 + (i > 0);
157 *pos++ = hex[val >> 4];
158 *pos = hex[val & 0xf];
163 * NOTE! This returns a statically allocated buffer, so you have to be
164 * careful about using it. Do an "xstrdup()" if you need to save the
167 * Also note that this returns the location for creating. Reading
168 * SHA1 file can happen from any alternate directory listed in the
169 * DB_ENVIRONMENT environment variable if it is not found in
170 * the primary object database.
172 char *sha1_file_name(const unsigned char *sha1)
174 static char *name, *base;
177 const char *sha1_file_directory = get_object_directory();
178 int len = strlen(sha1_file_directory);
179 base = xmalloc(len + 60);
180 memcpy(base, sha1_file_directory, len);
181 memset(base+len, 0, 60);
184 name = base + len + 1;
186 fill_sha1_path(name, sha1);
190 static char *sha1_get_pack_name(const unsigned char *sha1,
191 char **name, char **base, const char *which)
193 static const char hex[] = "0123456789abcdef";
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 *base = xmalloc(len + 60);
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
203 *name = *base + len + 11;
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
217 char *sha1_pack_name(const unsigned char *sha1)
219 static char *name, *base;
221 return sha1_get_pack_name(sha1, &name, &base, "pack");
224 char *sha1_pack_index_name(const unsigned char *sha1)
226 static char *name, *base;
228 return sha1_get_pack_name(sha1, &name, &base, "idx");
231 struct alternate_object_database *alt_odb_list;
232 static struct alternate_object_database **alt_odb_tail;
234 static void read_info_alternates(const char * alternates, int depth);
237 * Prepare alternate object database registry.
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
251 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
254 const char *objdir = get_object_directory();
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 /* 43 = 40-byte + 2 '/' + terminating NUL */
259 int entlen = pfxlen + 43;
262 if (!is_absolute_path(entry) && relative_base) {
263 /* Relative alt-odb */
265 base_len = strlen(relative_base) + 1;
269 ent = xmalloc(sizeof(*ent) + entlen);
271 if (!is_absolute_path(entry) && relative_base) {
272 memcpy(ent->base, relative_base, base_len - 1);
273 ent->base[base_len - 1] = '/';
274 memcpy(ent->base + base_len, entry, len);
277 memcpy(ent->base, entry, pfxlen);
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283 /* Detect cases where alternate disappeared */
284 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
301 if (!memcmp(ent->base, objdir, pfxlen)) {
306 /* add the alternate entry */
308 alt_odb_tail = &(ent->next);
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
314 ent->base[pfxlen] = '/';
319 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
322 const char *cp, *last;
325 error("%s: ignoring alternate object stores, nesting too deep.",
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
339 while (cp < ep && *cp != sep)
342 if (!is_absolute_path(last) && depth) {
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
350 while (cp < ep && *cp == sep)
356 static void read_info_alternates(const char * relative_base, int depth)
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = open(path, O_RDONLY);
372 if (fstat(fd, &st) || (st.st_size == 0)) {
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
385 void add_to_alternates_file(const char *reference)
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
389 char *alt = mkpath("%s/objects\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
397 void prepare_alt_odb(void)
404 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
407 alt_odb_tail = &alt_odb_list;
408 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
410 read_info_alternates(get_object_directory(), 0);
413 static int has_loose_object_local(const unsigned char *sha1)
415 char *name = sha1_file_name(sha1);
416 return !access(name, F_OK);
419 int has_loose_object_nonlocal(const unsigned char *sha1)
421 struct alternate_object_database *alt;
423 for (alt = alt_odb_list; alt; alt = alt->next) {
424 fill_sha1_path(alt->name, sha1);
425 if (!access(alt->base, F_OK))
431 static int has_loose_object(const unsigned char *sha1)
433 return has_loose_object_local(sha1) ||
434 has_loose_object_nonlocal(sha1);
437 static unsigned int pack_used_ctr;
438 static unsigned int pack_mmap_calls;
439 static unsigned int peak_pack_open_windows;
440 static unsigned int pack_open_windows;
441 static size_t peak_pack_mapped;
442 static size_t pack_mapped;
443 struct packed_git *packed_git;
445 void pack_report(void)
448 "pack_report: getpagesize() = %10" SZ_FMT "\n"
449 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
450 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
451 sz_fmt(getpagesize()),
452 sz_fmt(packed_git_window_size),
453 sz_fmt(packed_git_limit));
455 "pack_report: pack_used_ctr = %10u\n"
456 "pack_report: pack_mmap_calls = %10u\n"
457 "pack_report: pack_open_windows = %10u / %10u\n"
458 "pack_report: pack_mapped = "
459 "%10" SZ_FMT " / %10" SZ_FMT "\n",
462 pack_open_windows, peak_pack_open_windows,
463 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
466 static int check_packed_git_idx(const char *path, struct packed_git *p)
469 struct pack_idx_header *hdr;
471 uint32_t version, nr, i, *index;
472 int fd = open(path, O_RDONLY);
477 if (fstat(fd, &st)) {
481 idx_size = xsize_t(st.st_size);
482 if (idx_size < 4 * 256 + 20 + 20) {
484 return error("index file %s is too small", path);
486 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
490 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
491 version = ntohl(hdr->idx_version);
492 if (version < 2 || version > 2) {
493 munmap(idx_map, idx_size);
494 return error("index file %s is version %"PRIu32
495 " and is not supported by this binary"
496 " (try upgrading GIT to a newer version)",
505 index += 2; /* skip index header */
506 for (i = 0; i < 256; i++) {
507 uint32_t n = ntohl(index[i]);
509 munmap(idx_map, idx_size);
510 return error("non-monotonic index %s", path);
518 * - 256 index entries 4 bytes each
519 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
520 * - 20-byte SHA1 of the packfile
521 * - 20-byte SHA1 file checksum
523 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
524 munmap(idx_map, idx_size);
525 return error("wrong index v1 file size in %s", path);
527 } else if (version == 2) {
530 * - 8 bytes of header
531 * - 256 index entries 4 bytes each
532 * - 20-byte sha1 entry * nr
533 * - 4-byte crc entry * nr
534 * - 4-byte offset entry * nr
535 * - 20-byte SHA1 of the packfile
536 * - 20-byte SHA1 file checksum
537 * And after the 4-byte offset table might be a
538 * variable sized table containing 8-byte entries
539 * for offsets larger than 2^31.
541 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
542 unsigned long max_size = min_size;
544 max_size += (nr - 1)*8;
545 if (idx_size < min_size || idx_size > max_size) {
546 munmap(idx_map, idx_size);
547 return error("wrong index v2 file size in %s", path);
549 if (idx_size != min_size &&
551 * make sure we can deal with large pack offsets.
552 * 31-bit signed offset won't be enough, neither
553 * 32-bit unsigned one will be.
555 (sizeof(off_t) <= 4)) {
556 munmap(idx_map, idx_size);
557 return error("pack too large for current definition of off_t in %s", path);
561 p->index_version = version;
562 p->index_data = idx_map;
563 p->index_size = idx_size;
568 int open_pack_index(struct packed_git *p)
576 idx_name = xstrdup(p->pack_name);
577 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
578 ret = check_packed_git_idx(idx_name, p);
583 static void scan_windows(struct packed_git *p,
584 struct packed_git **lru_p,
585 struct pack_window **lru_w,
586 struct pack_window **lru_l)
588 struct pack_window *w, *w_l;
590 for (w_l = NULL, w = p->windows; w; w = w->next) {
592 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
602 static int unuse_one_window(struct packed_git *current, int keep_fd)
604 struct packed_git *p, *lru_p = NULL;
605 struct pack_window *lru_w = NULL, *lru_l = NULL;
608 scan_windows(current, &lru_p, &lru_w, &lru_l);
609 for (p = packed_git; p; p = p->next)
610 scan_windows(p, &lru_p, &lru_w, &lru_l);
612 munmap(lru_w->base, lru_w->len);
613 pack_mapped -= lru_w->len;
615 lru_l->next = lru_w->next;
617 lru_p->windows = lru_w->next;
618 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
619 close(lru_p->pack_fd);
630 void release_pack_memory(size_t need, int fd)
632 size_t cur = pack_mapped;
633 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
637 void close_pack_windows(struct packed_git *p)
640 struct pack_window *w = p->windows;
643 die("pack '%s' still has open windows to it",
645 munmap(w->base, w->len);
646 pack_mapped -= w->len;
648 p->windows = w->next;
653 void unuse_pack(struct pack_window **w_cursor)
655 struct pack_window *w = *w_cursor;
663 * Do not call this directly as this leaks p->pack_fd on error return;
664 * call open_packed_git() instead.
666 static int open_packed_git_1(struct packed_git *p)
669 struct pack_header hdr;
670 unsigned char sha1[20];
671 unsigned char *idx_sha1;
674 if (!p->index_data && open_pack_index(p))
675 return error("packfile %s index unavailable", p->pack_name);
677 p->pack_fd = open(p->pack_name, O_RDONLY);
678 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
681 /* If we created the struct before we had the pack we lack size. */
683 if (!S_ISREG(st.st_mode))
684 return error("packfile %s not a regular file", p->pack_name);
685 p->pack_size = st.st_size;
686 } else if (p->pack_size != st.st_size)
687 return error("packfile %s size changed", p->pack_name);
689 /* We leave these file descriptors open with sliding mmap;
690 * there is no point keeping them open across exec(), though.
692 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
694 return error("cannot determine file descriptor flags");
695 fd_flag |= FD_CLOEXEC;
696 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
697 return error("cannot set FD_CLOEXEC");
699 /* Verify we recognize this pack file format. */
700 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
701 return error("file %s is far too short to be a packfile", p->pack_name);
702 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
703 return error("file %s is not a GIT packfile", p->pack_name);
704 if (!pack_version_ok(hdr.hdr_version))
705 return error("packfile %s is version %"PRIu32" and not"
706 " supported (try upgrading GIT to a newer version)",
707 p->pack_name, ntohl(hdr.hdr_version));
709 /* Verify the pack matches its index. */
710 if (p->num_objects != ntohl(hdr.hdr_entries))
711 return error("packfile %s claims to have %"PRIu32" objects"
712 " while index indicates %"PRIu32" objects",
713 p->pack_name, ntohl(hdr.hdr_entries),
715 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
716 return error("end of packfile %s is unavailable", p->pack_name);
717 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
718 return error("packfile %s signature is unavailable", p->pack_name);
719 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
720 if (hashcmp(sha1, idx_sha1))
721 return error("packfile %s does not match index", p->pack_name);
725 static int open_packed_git(struct packed_git *p)
727 if (!open_packed_git_1(p))
729 if (p->pack_fd != -1) {
736 static int in_window(struct pack_window *win, off_t offset)
738 /* We must promise at least 20 bytes (one hash) after the
739 * offset is available from this window, otherwise the offset
740 * is not actually in this window and a different window (which
741 * has that one hash excess) must be used. This is to support
742 * the object header and delta base parsing routines below.
744 off_t win_off = win->offset;
745 return win_off <= offset
746 && (offset + 20) <= (win_off + win->len);
749 unsigned char* use_pack(struct packed_git *p,
750 struct pack_window **w_cursor,
754 struct pack_window *win = *w_cursor;
756 if (p->pack_fd == -1 && open_packed_git(p))
757 die("packfile %s cannot be accessed", p->pack_name);
759 /* Since packfiles end in a hash of their content and its
760 * pointless to ask for an offset into the middle of that
761 * hash, and the in_window function above wouldn't match
762 * don't allow an offset too close to the end of the file.
764 if (offset > (p->pack_size - 20))
765 die("offset beyond end of packfile (truncated pack?)");
767 if (!win || !in_window(win, offset)) {
770 for (win = p->windows; win; win = win->next) {
771 if (in_window(win, offset))
775 size_t window_align = packed_git_window_size / 2;
777 win = xcalloc(1, sizeof(*win));
778 win->offset = (offset / window_align) * window_align;
779 len = p->pack_size - win->offset;
780 if (len > packed_git_window_size)
781 len = packed_git_window_size;
782 win->len = (size_t)len;
783 pack_mapped += win->len;
784 while (packed_git_limit < pack_mapped
785 && unuse_one_window(p, p->pack_fd))
787 win->base = xmmap(NULL, win->len,
788 PROT_READ, MAP_PRIVATE,
789 p->pack_fd, win->offset);
790 if (win->base == MAP_FAILED)
791 die("packfile %s cannot be mapped: %s",
796 if (pack_mapped > peak_pack_mapped)
797 peak_pack_mapped = pack_mapped;
798 if (pack_open_windows > peak_pack_open_windows)
799 peak_pack_open_windows = pack_open_windows;
800 win->next = p->windows;
804 if (win != *w_cursor) {
805 win->last_used = pack_used_ctr++;
809 offset -= win->offset;
811 *left = win->len - xsize_t(offset);
812 return win->base + offset;
815 static struct packed_git *alloc_packed_git(int extra)
817 struct packed_git *p = xmalloc(sizeof(*p) + extra);
818 memset(p, 0, sizeof(*p));
823 struct packed_git *add_packed_git(const char *path, int path_len, int local)
826 struct packed_git *p = alloc_packed_git(path_len + 2);
829 * Make sure a corresponding .pack file exists and that
830 * the index looks sane.
832 path_len -= strlen(".idx");
837 memcpy(p->pack_name, path, path_len);
839 strcpy(p->pack_name + path_len, ".keep");
840 if (!access(p->pack_name, F_OK))
843 strcpy(p->pack_name + path_len, ".pack");
844 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
849 /* ok, it looks sane as far as we can check without
850 * actually mapping the pack file.
852 p->pack_size = st.st_size;
853 p->pack_local = local;
854 p->mtime = st.st_mtime;
855 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
860 struct packed_git *parse_pack_index(unsigned char *sha1)
862 const char *idx_path = sha1_pack_index_name(sha1);
863 const char *path = sha1_pack_name(sha1);
864 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
866 strcpy(p->pack_name, path);
867 hashcpy(p->sha1, sha1);
868 if (check_packed_git_idx(idx_path, p)) {
876 void install_packed_git(struct packed_git *pack)
878 pack->next = packed_git;
882 static void prepare_packed_git_one(char *objdir, int local)
884 /* Ensure that this buffer is large enough so that we can
885 append "/pack/" without clobbering the stack even if
886 strlen(objdir) were PATH_MAX. */
887 char path[PATH_MAX + 1 + 4 + 1 + 1];
892 sprintf(path, "%s/pack", objdir);
897 error("unable to open object pack directory: %s: %s",
898 path, strerror(errno));
902 while ((de = readdir(dir)) != NULL) {
903 int namelen = strlen(de->d_name);
904 struct packed_git *p;
906 if (!has_extension(de->d_name, ".idx"))
909 if (len + namelen + 1 > sizeof(path))
912 /* Don't reopen a pack we already have. */
913 strcpy(path + len, de->d_name);
914 for (p = packed_git; p; p = p->next) {
915 if (!memcmp(path, p->pack_name, len + namelen - 4))
920 /* See if it really is a valid .idx file with corresponding
921 * .pack file that we can map.
923 p = add_packed_git(path, len + namelen, local);
926 install_packed_git(p);
931 static int sort_pack(const void *a_, const void *b_)
933 struct packed_git *a = *((struct packed_git **)a_);
934 struct packed_git *b = *((struct packed_git **)b_);
938 * Local packs tend to contain objects specific to our
939 * variant of the project than remote ones. In addition,
940 * remote ones could be on a network mounted filesystem.
941 * Favor local ones for these reasons.
943 st = a->pack_local - b->pack_local;
948 * Younger packs tend to contain more recent objects,
949 * and more recent objects tend to get accessed more
952 if (a->mtime < b->mtime)
954 else if (a->mtime == b->mtime)
959 static void rearrange_packed_git(void)
961 struct packed_git **ary, *p;
964 for (n = 0, p = packed_git; p; p = p->next)
969 /* prepare an array of packed_git for easier sorting */
970 ary = xcalloc(n, sizeof(struct packed_git *));
971 for (n = 0, p = packed_git; p; p = p->next)
974 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
976 /* link them back again */
977 for (i = 0; i < n - 1; i++)
978 ary[i]->next = ary[i + 1];
979 ary[n - 1]->next = NULL;
985 static int prepare_packed_git_run_once = 0;
986 void prepare_packed_git(void)
988 struct alternate_object_database *alt;
990 if (prepare_packed_git_run_once)
992 prepare_packed_git_one(get_object_directory(), 1);
994 for (alt = alt_odb_list; alt; alt = alt->next) {
996 prepare_packed_git_one(alt->base, 0);
999 rearrange_packed_git();
1000 prepare_packed_git_run_once = 1;
1003 void reprepare_packed_git(void)
1006 prepare_packed_git_run_once = 0;
1007 prepare_packed_git();
1010 static void mark_bad_packed_object(struct packed_git *p,
1011 const unsigned char *sha1)
1014 for (i = 0; i < p->num_bad_objects; i++)
1015 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1017 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1018 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1019 p->num_bad_objects++;
1022 static int has_packed_and_bad(const unsigned char *sha1)
1024 struct packed_git *p;
1027 for (p = packed_git; p; p = p->next)
1028 for (i = 0; i < p->num_bad_objects; i++)
1029 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1034 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1036 unsigned char real_sha1[20];
1037 hash_sha1_file(map, size, type, real_sha1);
1038 return hashcmp(sha1, real_sha1) ? -1 : 0;
1041 static int git_open_noatime(const char *name)
1043 static int sha1_file_open_flag = O_NOATIME;
1044 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1046 /* Might the failure be due to O_NOATIME? */
1047 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1048 fd = open(name, O_RDONLY);
1050 sha1_file_open_flag = 0;
1055 static int open_sha1_file(const unsigned char *sha1)
1058 char *name = sha1_file_name(sha1);
1059 struct alternate_object_database *alt;
1061 fd = git_open_noatime(name);
1067 for (alt = alt_odb_list; alt; alt = alt->next) {
1069 fill_sha1_path(name, sha1);
1070 fd = git_open_noatime(alt->base);
1077 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1082 fd = open_sha1_file(sha1);
1087 if (!fstat(fd, &st)) {
1088 *size = xsize_t(st.st_size);
1089 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1096 static int legacy_loose_object(unsigned char *map)
1101 * Is it a zlib-compressed buffer? If so, the first byte
1102 * must be 0x78 (15-bit window size, deflated), and the
1103 * first 16-bit word is evenly divisible by 31
1105 word = (map[0] << 8) + map[1];
1106 if (map[0] == 0x78 && !(word % 31))
1112 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1117 unsigned long used = 0;
1120 *type = (c >> 4) & 7;
1126 if (sizeof(long) * 8 <= shift)
1129 size += (c & 0x7f) << shift;
1136 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1138 unsigned long size, used;
1139 static const char valid_loose_object_type[8] = {
1141 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1142 0, /* "delta" and others are invalid in a loose object */
1144 enum object_type type;
1146 /* Get the data stream */
1147 memset(stream, 0, sizeof(*stream));
1148 stream->next_in = map;
1149 stream->avail_in = mapsize;
1150 stream->next_out = buffer;
1151 stream->avail_out = bufsiz;
1153 if (legacy_loose_object(map)) {
1154 inflateInit(stream);
1155 return inflate(stream, 0);
1160 * There used to be a second loose object header format which
1161 * was meant to mimic the in-pack format, allowing for direct
1162 * copy of the object data. This format turned up not to be
1163 * really worth it and we don't write it any longer. But we
1164 * can still read it.
1166 used = unpack_object_header_gently(map, mapsize, &type, &size);
1167 if (!used || !valid_loose_object_type[type])
1172 /* Set up the stream for the rest.. */
1173 stream->next_in = map;
1174 stream->avail_in = mapsize;
1175 inflateInit(stream);
1177 /* And generate the fake traditional header */
1178 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1179 typename(type), size);
1183 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1185 int bytes = strlen(buffer) + 1;
1186 unsigned char *buf = xmalloc(1+size);
1190 n = stream->total_out - bytes;
1193 memcpy(buf, (char *) buffer + bytes, n);
1195 if (bytes <= size) {
1197 * The above condition must be (bytes <= size), not
1198 * (bytes < size). In other words, even though we
1199 * expect no more output and set avail_out to zer0,
1200 * the input zlib stream may have bytes that express
1201 * "this concludes the stream", and we *do* want to
1204 * Otherwise we would not be able to test that we
1205 * consumed all the input to reach the expected size;
1206 * we also want to check that zlib tells us that all
1207 * went well with status == Z_STREAM_END at the end.
1209 stream->next_out = buf + bytes;
1210 stream->avail_out = size - bytes;
1211 while (status == Z_OK)
1212 status = inflate(stream, Z_FINISH);
1215 if (status == Z_STREAM_END && !stream->avail_in) {
1221 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1222 else if (stream->avail_in)
1223 error("garbage at end of loose object '%s'",
1230 * We used to just use "sscanf()", but that's actually way
1231 * too permissive for what we want to check. So do an anal
1232 * object header parse by hand.
1234 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1241 * The type can be at most ten bytes (including the
1242 * terminating '\0' that we add), and is followed by
1251 if (i >= sizeof(type))
1257 * The length must follow immediately, and be in canonical
1258 * decimal format (ie "010" is not valid).
1260 size = *hdr++ - '0';
1265 unsigned long c = *hdr - '0';
1269 size = size * 10 + c;
1275 * The length must be followed by a zero byte
1277 return *hdr ? -1 : type_from_string(type);
1280 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1286 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1287 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1290 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1293 unsigned long get_size_from_delta(struct packed_git *p,
1294 struct pack_window **w_curs,
1297 const unsigned char *data;
1298 unsigned char delta_head[20], *in;
1302 memset(&stream, 0, sizeof(stream));
1303 stream.next_out = delta_head;
1304 stream.avail_out = sizeof(delta_head);
1306 inflateInit(&stream);
1308 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1309 stream.next_in = in;
1310 st = inflate(&stream, Z_FINISH);
1311 curpos += stream.next_in - in;
1312 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1313 stream.total_out < sizeof(delta_head));
1314 inflateEnd(&stream);
1315 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1316 die("delta data unpack-initial failed");
1318 /* Examine the initial part of the delta to figure out
1323 /* ignore base size */
1324 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1326 /* Read the result size */
1327 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1330 static off_t get_delta_base(struct packed_git *p,
1331 struct pack_window **w_curs,
1333 enum object_type type,
1334 off_t delta_obj_offset)
1336 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1339 /* use_pack() assured us we have [base_info, base_info + 20)
1340 * as a range that we can look at without walking off the
1341 * end of the mapped window. Its actually the hash size
1342 * that is assured. An OFS_DELTA longer than the hash size
1343 * is stupid, as then a REF_DELTA would be smaller to store.
1345 if (type == OBJ_OFS_DELTA) {
1347 unsigned char c = base_info[used++];
1348 base_offset = c & 127;
1351 if (!base_offset || MSB(base_offset, 7))
1352 return 0; /* overflow */
1353 c = base_info[used++];
1354 base_offset = (base_offset << 7) + (c & 127);
1356 base_offset = delta_obj_offset - base_offset;
1357 if (base_offset >= delta_obj_offset)
1358 return 0; /* out of bound */
1360 } else if (type == OBJ_REF_DELTA) {
1361 /* The base entry _must_ be in the same pack */
1362 base_offset = find_pack_entry_one(base_info, p);
1365 die("I am totally screwed");
1369 /* forward declaration for a mutually recursive function */
1370 static int packed_object_info(struct packed_git *p, off_t offset,
1371 unsigned long *sizep);
1373 static int packed_delta_info(struct packed_git *p,
1374 struct pack_window **w_curs,
1376 enum object_type type,
1378 unsigned long *sizep)
1382 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1383 type = packed_object_info(p, base_offset, NULL);
1385 /* We choose to only get the type of the base object and
1386 * ignore potentially corrupt pack file that expects the delta
1387 * based on a base with a wrong size. This saves tons of
1391 *sizep = get_size_from_delta(p, w_curs, curpos);
1396 static int unpack_object_header(struct packed_git *p,
1397 struct pack_window **w_curs,
1399 unsigned long *sizep)
1401 unsigned char *base;
1404 enum object_type type;
1406 /* use_pack() assures us we have [base, base + 20) available
1407 * as a range that we can look at at. (Its actually the hash
1408 * size that is assured.) With our object header encoding
1409 * the maximum deflated object size is 2^137, which is just
1410 * insane, so we know won't exceed what we have been given.
1412 base = use_pack(p, w_curs, *curpos, &left);
1413 used = unpack_object_header_gently(base, left, &type, sizep);
1415 die("object offset outside of pack file");
1421 const char *packed_object_info_detail(struct packed_git *p,
1423 unsigned long *size,
1424 unsigned long *store_size,
1425 unsigned int *delta_chain_length,
1426 unsigned char *base_sha1)
1428 struct pack_window *w_curs = NULL;
1430 unsigned long dummy;
1431 unsigned char *next_sha1;
1432 enum object_type type;
1433 struct revindex_entry *revidx;
1435 *delta_chain_length = 0;
1436 curpos = obj_offset;
1437 type = unpack_object_header(p, &w_curs, &curpos, size);
1439 revidx = find_pack_revindex(p, obj_offset);
1440 *store_size = revidx[1].offset - obj_offset;
1445 die("pack %s contains unknown object type %d",
1446 p->pack_name, type);
1451 unuse_pack(&w_curs);
1452 return typename(type);
1454 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1456 die("pack %s contains bad delta base reference of type %s",
1457 p->pack_name, typename(type));
1458 if (*delta_chain_length == 0) {
1459 revidx = find_pack_revindex(p, obj_offset);
1460 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1464 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1465 if (*delta_chain_length == 0)
1466 hashcpy(base_sha1, next_sha1);
1467 obj_offset = find_pack_entry_one(next_sha1, p);
1470 (*delta_chain_length)++;
1471 curpos = obj_offset;
1472 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1476 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1477 unsigned long *sizep)
1479 struct pack_window *w_curs = NULL;
1481 off_t curpos = obj_offset;
1482 enum object_type type;
1484 type = unpack_object_header(p, &w_curs, &curpos, &size);
1489 type = packed_delta_info(p, &w_curs, curpos,
1490 type, obj_offset, sizep);
1500 die("pack %s contains unknown object type %d",
1501 p->pack_name, type);
1503 unuse_pack(&w_curs);
1507 static void *unpack_compressed_entry(struct packed_git *p,
1508 struct pack_window **w_curs,
1514 unsigned char *buffer, *in;
1516 buffer = xmalloc(size + 1);
1518 memset(&stream, 0, sizeof(stream));
1519 stream.next_out = buffer;
1520 stream.avail_out = size;
1522 inflateInit(&stream);
1524 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1525 stream.next_in = in;
1526 st = inflate(&stream, Z_FINISH);
1527 curpos += stream.next_in - in;
1528 } while (st == Z_OK || st == Z_BUF_ERROR);
1529 inflateEnd(&stream);
1530 if ((st != Z_STREAM_END) || stream.total_out != size) {
1538 #define MAX_DELTA_CACHE (256)
1540 static size_t delta_base_cached;
1542 static struct delta_base_cache_lru_list {
1543 struct delta_base_cache_lru_list *prev;
1544 struct delta_base_cache_lru_list *next;
1545 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1547 static struct delta_base_cache_entry {
1548 struct delta_base_cache_lru_list lru;
1550 struct packed_git *p;
1553 enum object_type type;
1554 } delta_base_cache[MAX_DELTA_CACHE];
1556 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1560 hash = (unsigned long)p + (unsigned long)base_offset;
1561 hash += (hash >> 8) + (hash >> 16);
1562 return hash % MAX_DELTA_CACHE;
1565 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1566 unsigned long *base_size, enum object_type *type, int keep_cache)
1569 unsigned long hash = pack_entry_hash(p, base_offset);
1570 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1573 if (ret && ent->p == p && ent->base_offset == base_offset)
1574 goto found_cache_entry;
1575 return unpack_entry(p, base_offset, type, base_size);
1580 ent->lru.next->prev = ent->lru.prev;
1581 ent->lru.prev->next = ent->lru.next;
1582 delta_base_cached -= ent->size;
1584 ret = xmemdupz(ent->data, ent->size);
1587 *base_size = ent->size;
1591 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1596 ent->lru.next->prev = ent->lru.prev;
1597 ent->lru.prev->next = ent->lru.next;
1598 delta_base_cached -= ent->size;
1602 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1603 void *base, unsigned long base_size, enum object_type type)
1605 unsigned long hash = pack_entry_hash(p, base_offset);
1606 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1607 struct delta_base_cache_lru_list *lru;
1609 release_delta_base_cache(ent);
1610 delta_base_cached += base_size;
1612 for (lru = delta_base_cache_lru.next;
1613 delta_base_cached > delta_base_cache_limit
1614 && lru != &delta_base_cache_lru;
1616 struct delta_base_cache_entry *f = (void *)lru;
1617 if (f->type == OBJ_BLOB)
1618 release_delta_base_cache(f);
1620 for (lru = delta_base_cache_lru.next;
1621 delta_base_cached > delta_base_cache_limit
1622 && lru != &delta_base_cache_lru;
1624 struct delta_base_cache_entry *f = (void *)lru;
1625 release_delta_base_cache(f);
1629 ent->base_offset = base_offset;
1632 ent->size = base_size;
1633 ent->lru.next = &delta_base_cache_lru;
1634 ent->lru.prev = delta_base_cache_lru.prev;
1635 delta_base_cache_lru.prev->next = &ent->lru;
1636 delta_base_cache_lru.prev = &ent->lru;
1639 static void *unpack_delta_entry(struct packed_git *p,
1640 struct pack_window **w_curs,
1642 unsigned long delta_size,
1644 enum object_type *type,
1645 unsigned long *sizep)
1647 void *delta_data, *result, *base;
1648 unsigned long base_size;
1651 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1653 error("failed to validate delta base reference "
1654 "at offset %"PRIuMAX" from %s",
1655 (uintmax_t)curpos, p->pack_name);
1659 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1662 * We're probably in deep shit, but let's try to fetch
1663 * the required base anyway from another pack or loose.
1664 * This is costly but should happen only in the presence
1665 * of a corrupted pack, and is better than failing outright.
1667 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1668 const unsigned char *base_sha1 =
1669 nth_packed_object_sha1(p, revidx->nr);
1670 error("failed to read delta base object %s"
1671 " at offset %"PRIuMAX" from %s",
1672 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1674 mark_bad_packed_object(p, base_sha1);
1675 base = read_object(base_sha1, type, &base_size);
1680 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1682 error("failed to unpack compressed delta "
1683 "at offset %"PRIuMAX" from %s",
1684 (uintmax_t)curpos, p->pack_name);
1688 result = patch_delta(base, base_size,
1689 delta_data, delta_size,
1692 die("failed to apply delta");
1694 add_delta_base_cache(p, base_offset, base, base_size, *type);
1698 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1699 enum object_type *type, unsigned long *sizep)
1701 struct pack_window *w_curs = NULL;
1702 off_t curpos = obj_offset;
1705 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1709 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1710 obj_offset, type, sizep);
1716 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1720 error("unknown object type %i at offset %"PRIuMAX" in %s",
1721 *type, (uintmax_t)obj_offset, p->pack_name);
1723 unuse_pack(&w_curs);
1727 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1730 const unsigned char *index = p->index_data;
1732 if (open_pack_index(p))
1734 index = p->index_data;
1736 if (n >= p->num_objects)
1739 if (p->index_version == 1) {
1740 return index + 24 * n + 4;
1743 return index + 20 * n;
1747 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1749 const unsigned char *index = p->index_data;
1751 if (p->index_version == 1) {
1752 return ntohl(*((uint32_t *)(index + 24 * n)));
1755 index += 8 + p->num_objects * (20 + 4);
1756 off = ntohl(*((uint32_t *)(index + 4 * n)));
1757 if (!(off & 0x80000000))
1759 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1760 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1761 ntohl(*((uint32_t *)(index + 4)));
1765 off_t find_pack_entry_one(const unsigned char *sha1,
1766 struct packed_git *p)
1768 const uint32_t *level1_ofs = p->index_data;
1769 const unsigned char *index = p->index_data;
1770 unsigned hi, lo, stride;
1771 static int use_lookup = -1;
1772 static int debug_lookup = -1;
1774 if (debug_lookup < 0)
1775 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1778 if (open_pack_index(p))
1780 level1_ofs = p->index_data;
1781 index = p->index_data;
1783 if (p->index_version > 1) {
1788 hi = ntohl(level1_ofs[*sha1]);
1789 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1790 if (p->index_version > 1) {
1798 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1799 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1802 use_lookup = !!getenv("GIT_USE_LOOKUP");
1804 int pos = sha1_entry_pos(index, stride, 0,
1805 lo, hi, p->num_objects, sha1);
1808 return nth_packed_object_offset(p, pos);
1812 unsigned mi = (lo + hi) / 2;
1813 int cmp = hashcmp(index + mi * stride, sha1);
1816 printf("lo %u hi %u rg %u mi %u\n",
1817 lo, hi, hi - lo, mi);
1819 return nth_packed_object_offset(p, mi);
1828 int matches_pack_name(struct packed_git *p, const char *name)
1830 const char *last_c, *c;
1832 if (!strcmp(p->pack_name, name))
1835 for (c = p->pack_name, last_c = c; *c;)
1840 if (!strcmp(last_c, name))
1846 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1848 static struct packed_git *last_found = (void *)1;
1849 struct packed_git *p;
1852 prepare_packed_git();
1855 p = (last_found == (void *)1) ? packed_git : last_found;
1858 if (ignore_packed) {
1860 for (ig = ignore_packed; *ig; ig++)
1861 if (matches_pack_name(p, *ig))
1867 if (p->num_bad_objects) {
1869 for (i = 0; i < p->num_bad_objects; i++)
1870 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1874 offset = find_pack_entry_one(sha1, p);
1877 * We are about to tell the caller where they can
1878 * locate the requested object. We better make
1879 * sure the packfile is still here and can be
1880 * accessed before supplying that answer, as
1881 * it may have been deleted since the index
1884 if (p->pack_fd == -1 && open_packed_git(p)) {
1885 error("packfile %s cannot be accessed", p->pack_name);
1890 hashcpy(e->sha1, sha1);
1896 if (p == last_found)
1900 if (p == last_found)
1906 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1907 struct packed_git *packs)
1909 struct packed_git *p;
1911 for (p = packs; p; p = p->next) {
1912 if (find_pack_entry_one(sha1, p))
1919 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1922 unsigned long mapsize, size;
1927 map = map_sha1_file(sha1, &mapsize);
1929 return error("unable to find %s", sha1_to_hex(sha1));
1930 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1931 status = error("unable to unpack %s header",
1933 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1934 status = error("unable to parse %s header", sha1_to_hex(sha1));
1937 inflateEnd(&stream);
1938 munmap(map, mapsize);
1942 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1944 struct pack_entry e;
1947 if (!find_pack_entry(sha1, &e, NULL)) {
1948 /* Most likely it's a loose object. */
1949 status = sha1_loose_object_info(sha1, sizep);
1953 /* Not a loose object; someone else may have just packed it. */
1954 reprepare_packed_git();
1955 if (!find_pack_entry(sha1, &e, NULL))
1958 return packed_object_info(e.p, e.offset, sizep);
1961 static void *read_packed_sha1(const unsigned char *sha1,
1962 enum object_type *type, unsigned long *size)
1964 struct pack_entry e;
1967 if (!find_pack_entry(sha1, &e, NULL))
1969 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1972 * We're probably in deep shit, but let's try to fetch
1973 * the required object anyway from another pack or loose.
1974 * This should happen only in the presence of a corrupted
1975 * pack, and is better than failing outright.
1977 error("failed to read object %s at offset %"PRIuMAX" from %s",
1978 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1979 mark_bad_packed_object(e.p, sha1);
1980 data = read_object(sha1, type, size);
1986 * This is meant to hold a *small* number of objects that you would
1987 * want read_sha1_file() to be able to return, but yet you do not want
1988 * to write them into the object store (e.g. a browse-only
1991 static struct cached_object {
1992 unsigned char sha1[20];
1993 enum object_type type;
1997 static int cached_object_nr, cached_object_alloc;
1999 static struct cached_object empty_tree = {
2000 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2001 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2002 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2008 static struct cached_object *find_cached_object(const unsigned char *sha1)
2011 struct cached_object *co = cached_objects;
2013 for (i = 0; i < cached_object_nr; i++, co++) {
2014 if (!hashcmp(co->sha1, sha1))
2017 if (!hashcmp(sha1, empty_tree.sha1))
2022 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2023 unsigned char *sha1)
2025 struct cached_object *co;
2027 hash_sha1_file(buf, len, typename(type), sha1);
2028 if (has_sha1_file(sha1) || find_cached_object(sha1))
2030 if (cached_object_alloc <= cached_object_nr) {
2031 cached_object_alloc = alloc_nr(cached_object_alloc);
2032 cached_objects = xrealloc(cached_objects,
2033 sizeof(*cached_objects) *
2034 cached_object_alloc);
2036 co = &cached_objects[cached_object_nr++];
2039 co->buf = xmalloc(len);
2040 memcpy(co->buf, buf, len);
2041 hashcpy(co->sha1, sha1);
2045 void *read_object(const unsigned char *sha1, enum object_type *type,
2046 unsigned long *size)
2048 unsigned long mapsize;
2050 struct cached_object *co;
2052 co = find_cached_object(sha1);
2056 return xmemdupz(co->buf, co->size);
2059 buf = read_packed_sha1(sha1, type, size);
2062 map = map_sha1_file(sha1, &mapsize);
2064 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2065 munmap(map, mapsize);
2068 reprepare_packed_git();
2069 return read_packed_sha1(sha1, type, size);
2072 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2073 unsigned long *size)
2075 void *data = read_object(sha1, type, size);
2076 /* legacy behavior is to die on corrupted objects */
2077 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2078 die("object %s is corrupted", sha1_to_hex(sha1));
2082 void *read_object_with_reference(const unsigned char *sha1,
2083 const char *required_type_name,
2084 unsigned long *size,
2085 unsigned char *actual_sha1_return)
2087 enum object_type type, required_type;
2089 unsigned long isize;
2090 unsigned char actual_sha1[20];
2092 required_type = type_from_string(required_type_name);
2093 hashcpy(actual_sha1, sha1);
2095 int ref_length = -1;
2096 const char *ref_type = NULL;
2098 buffer = read_sha1_file(actual_sha1, &type, &isize);
2101 if (type == required_type) {
2103 if (actual_sha1_return)
2104 hashcpy(actual_sha1_return, actual_sha1);
2107 /* Handle references */
2108 else if (type == OBJ_COMMIT)
2110 else if (type == OBJ_TAG)
2111 ref_type = "object ";
2116 ref_length = strlen(ref_type);
2118 if (ref_length + 40 > isize ||
2119 memcmp(buffer, ref_type, ref_length) ||
2120 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2125 /* Now we have the ID of the referred-to object in
2126 * actual_sha1. Check again. */
2130 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2131 const char *type, unsigned char *sha1,
2132 char *hdr, int *hdrlen)
2136 /* Generate the header */
2137 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2141 SHA1_Update(&c, hdr, *hdrlen);
2142 SHA1_Update(&c, buf, len);
2143 SHA1_Final(sha1, &c);
2147 * Move the just written object into its final resting place
2149 int move_temp_to_file(const char *tmpfile, const char *filename)
2152 if (link(tmpfile, filename))
2156 * Coda hack - coda doesn't like cross-directory links,
2157 * so we fall back to a rename, which will mean that it
2158 * won't be able to check collisions, but that's not a
2161 * The same holds for FAT formatted media.
2163 * When this succeeds, we just return 0. We have nothing
2166 if (ret && ret != EEXIST) {
2167 if (!rename(tmpfile, filename))
2173 if (ret != EEXIST) {
2174 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2176 /* FIXME!!! Collision check here ? */
2182 static int write_buffer(int fd, const void *buf, size_t len)
2184 if (write_in_full(fd, buf, len) < 0)
2185 return error("file write error (%s)", strerror(errno));
2189 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2190 unsigned char *sha1)
2194 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2198 /* Finalize a file on disk, and close it. */
2199 static void close_sha1_file(int fd)
2201 if (fsync_object_files)
2202 fsync_or_die(fd, "sha1 file");
2205 die("unable to write sha1 file");
2208 /* Size of directory component, including the ending '/' */
2209 static inline int directory_size(const char *filename)
2211 const char *s = strrchr(filename, '/');
2214 return s - filename + 1;
2218 * This creates a temporary file in the same directory as the final
2221 * We want to avoid cross-directory filename renames, because those
2222 * can have problems on various filesystems (FAT, NFS, Coda).
2224 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2226 int fd, dirlen = directory_size(filename);
2228 if (dirlen + 20 > bufsiz) {
2229 errno = ENAMETOOLONG;
2232 memcpy(buffer, filename, dirlen);
2233 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2234 fd = mkstemp(buffer);
2235 if (fd < 0 && dirlen && errno == ENOENT) {
2236 /* Make sure the directory exists */
2237 memcpy(buffer, filename, dirlen);
2238 buffer[dirlen-1] = 0;
2239 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2243 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2244 fd = mkstemp(buffer);
2249 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2250 void *buf, unsigned long len, time_t mtime)
2253 unsigned char *compressed;
2256 static char tmpfile[PATH_MAX];
2258 filename = sha1_file_name(sha1);
2259 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2261 if (errno == EACCES)
2262 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2264 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2268 memset(&stream, 0, sizeof(stream));
2269 deflateInit(&stream, zlib_compression_level);
2270 size = 8 + deflateBound(&stream, len+hdrlen);
2271 compressed = xmalloc(size);
2274 stream.next_out = compressed;
2275 stream.avail_out = size;
2277 /* First header.. */
2278 stream.next_in = (unsigned char *)hdr;
2279 stream.avail_in = hdrlen;
2280 while (deflate(&stream, 0) == Z_OK)
2283 /* Then the data itself.. */
2284 stream.next_in = buf;
2285 stream.avail_in = len;
2286 ret = deflate(&stream, Z_FINISH);
2287 if (ret != Z_STREAM_END)
2288 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2290 ret = deflateEnd(&stream);
2292 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2294 size = stream.total_out;
2296 if (write_buffer(fd, compressed, size) < 0)
2297 die("unable to write sha1 file");
2298 close_sha1_file(fd);
2304 utb.modtime = mtime;
2305 if (utime(tmpfile, &utb) < 0)
2306 warning("failed utime() on %s: %s",
2307 tmpfile, strerror(errno));
2310 return move_temp_to_file(tmpfile, filename);
2313 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2315 unsigned char sha1[20];
2319 /* Normally if we have it in the pack then we do not bother writing
2320 * it out into .git/objects/??/?{38} file.
2322 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2324 hashcpy(returnsha1, sha1);
2325 if (has_sha1_file(sha1))
2327 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2330 int force_object_loose(const unsigned char *sha1, time_t mtime)
2334 enum object_type type;
2339 if (has_loose_object(sha1))
2341 buf = read_packed_sha1(sha1, &type, &len);
2343 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2344 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2345 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2351 int has_pack_index(const unsigned char *sha1)
2354 if (stat(sha1_pack_index_name(sha1), &st))
2359 int has_pack_file(const unsigned char *sha1)
2362 if (stat(sha1_pack_name(sha1), &st))
2367 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2369 struct pack_entry e;
2370 return find_pack_entry(sha1, &e, ignore_packed);
2373 int has_sha1_file(const unsigned char *sha1)
2375 struct pack_entry e;
2377 if (find_pack_entry(sha1, &e, NULL))
2379 return has_loose_object(sha1);
2382 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2387 strbuf_init(&buf, 0);
2388 if (strbuf_read(&buf, fd, 4096) < 0) {
2389 strbuf_release(&buf);
2396 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2398 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2399 strbuf_release(&buf);
2404 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2405 enum object_type type, const char *path)
2407 size_t size = xsize_t(st->st_size);
2409 int ret, re_allocated = 0;
2412 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2419 * Convert blobs to git internal format
2421 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2423 strbuf_init(&nbuf, 0);
2424 if (convert_to_git(path, buf, size, &nbuf,
2425 write_object ? safe_crlf : 0)) {
2427 buf = strbuf_detach(&nbuf, &size);
2433 ret = write_sha1_file(buf, size, typename(type), sha1);
2435 ret = hash_sha1_file(buf, size, typename(type), sha1);
2445 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2451 switch (st->st_mode & S_IFMT) {
2453 fd = open(path, O_RDONLY);
2455 return error("open(\"%s\"): %s", path,
2457 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2458 return error("%s: failed to insert into database",
2462 len = xsize_t(st->st_size);
2463 target = xmalloc(len + 1);
2464 if (readlink(path, target, len + 1) != st->st_size) {
2465 char *errstr = strerror(errno);
2467 return error("readlink(\"%s\"): %s", path,
2471 hash_sha1_file(target, len, blob_type, sha1);
2472 else if (write_sha1_file(target, len, blob_type, sha1))
2473 return error("%s: failed to insert into database",
2478 return resolve_gitlink_ref(path, "HEAD", sha1);
2480 return error("%s: unsupported file type", path);
2485 int read_pack_header(int fd, struct pack_header *header)
2487 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2488 /* "eof before pack header was fully read" */
2489 return PH_ERROR_EOF;
2491 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2492 /* "protocol error (pack signature mismatch detected)" */
2493 return PH_ERROR_PACK_SIGNATURE;
2494 if (!pack_version_ok(header->hdr_version))
2495 /* "protocol error (pack version unsupported)" */
2496 return PH_ERROR_PROTOCOL;