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, '/');
102 while (*++pos == '/')
107 if (!stat(path, &st)) {
109 if (!S_ISDIR(st.st_mode)) {
114 else if (mkdir(path, 0777)) {
118 else if (adjust_shared_perm(path)) {
127 int safe_create_leading_directories_const(const char *path)
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf = xstrdup(path);
131 int result = safe_create_leading_directories(buf);
136 char *sha1_to_hex(const unsigned char *sha1)
139 static char hexbuffer[4][50];
140 static const char hex[] = "0123456789abcdef";
141 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
144 for (i = 0; i < 20; i++) {
145 unsigned int val = *sha1++;
146 *buf++ = hex[val >> 4];
147 *buf++ = hex[val & 0xf];
154 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
176 char *sha1_file_name(const unsigned char *sha1)
178 static char *name, *base;
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 memcpy(base, sha1_file_directory, len);
185 memset(base+len, 0, 60);
188 name = base + len + 1;
190 fill_sha1_path(name, sha1);
194 static char *sha1_get_pack_name(const unsigned char *sha1,
195 char **name, char **base, const char *which)
197 static const char hex[] = "0123456789abcdef";
202 const char *sha1_file_directory = get_object_directory();
203 int len = strlen(sha1_file_directory);
204 *base = xmalloc(len + 60);
205 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory, which);
207 *name = *base + len + 11;
212 for (i = 0; i < 20; i++) {
213 unsigned int val = *sha1++;
214 *buf++ = hex[val >> 4];
215 *buf++ = hex[val & 0xf];
221 char *sha1_pack_name(const unsigned char *sha1)
223 static char *name, *base;
225 return sha1_get_pack_name(sha1, &name, &base, "pack");
228 char *sha1_pack_index_name(const unsigned char *sha1)
230 static char *name, *base;
232 return sha1_get_pack_name(sha1, &name, &base, "idx");
235 struct alternate_object_database *alt_odb_list;
236 static struct alternate_object_database **alt_odb_tail;
238 static void read_info_alternates(const char * alternates, int depth);
241 * Prepare alternate object database registry.
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
255 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
257 const char *objdir = get_object_directory();
258 struct alternate_object_database *ent;
259 struct alternate_object_database *alt;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
262 int entlen = pfxlen + 43;
265 if (!is_absolute_path(entry) && relative_base) {
266 /* Relative alt-odb */
268 base_len = strlen(relative_base) + 1;
272 ent = xmalloc(sizeof(*ent) + entlen);
274 if (!is_absolute_path(entry) && relative_base) {
275 memcpy(ent->base, relative_base, base_len - 1);
276 ent->base[base_len - 1] = '/';
277 memcpy(ent->base + base_len, entry, len);
280 memcpy(ent->base, entry, pfxlen);
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent->base)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (!memcmp(ent->base, alt->base, pfxlen)) {
304 if (!memcmp(ent->base, objdir, pfxlen)) {
309 /* add the alternate entry */
311 alt_odb_tail = &(ent->next);
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
317 ent->base[pfxlen] = '/';
322 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
323 const char *relative_base, int depth)
325 const char *cp, *last;
328 error("%s: ignoring alternate object stores, nesting too deep.",
336 if (cp < ep && *cp == '#') {
337 while (cp < ep && *cp != sep)
342 while (cp < ep && *cp != sep)
345 if (!is_absolute_path(last) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, last);
349 link_alt_odb_entry(last, cp - last,
350 relative_base, depth);
353 while (cp < ep && *cp == sep)
359 static void read_info_alternates(const char * relative_base, int depth)
364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
371 sprintf(path, "%s/%s", relative_base, alt_file_name);
372 fd = open(path, O_RDONLY);
375 if (fstat(fd, &st) || (st.st_size == 0)) {
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
383 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
388 void add_to_alternates_file(const char *reference)
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
392 char *alt = mkpath("%s/objects\n", reference);
393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
397 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
400 void foreach_alt_odb(alt_odb_fn fn, void *cb)
402 struct alternate_object_database *ent;
405 for (ent = alt_odb_list; ent; ent = ent->next)
410 void prepare_alt_odb(void)
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
420 alt_odb_tail = &alt_odb_list;
421 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object(const unsigned char *sha1)
428 char *name = sha1_file_name(sha1);
429 struct alternate_object_database *alt;
431 if (!access(name, F_OK))
434 for (alt = alt_odb_list; alt; alt = alt->next) {
436 fill_sha1_path(name, sha1);
437 if (!access(alt->base, F_OK))
443 static unsigned int pack_used_ctr;
444 static unsigned int pack_mmap_calls;
445 static unsigned int peak_pack_open_windows;
446 static unsigned int pack_open_windows;
447 static size_t peak_pack_mapped;
448 static size_t pack_mapped;
449 struct packed_git *packed_git;
451 void pack_report(void)
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
468 pack_open_windows, peak_pack_open_windows,
469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
472 static int check_packed_git_idx(const char *path, struct packed_git *p)
475 struct pack_idx_header *hdr;
477 uint32_t version, nr, i, *index;
478 int fd = open(path, O_RDONLY);
483 if (fstat(fd, &st)) {
487 idx_size = xsize_t(st.st_size);
488 if (idx_size < 4 * 256 + 20 + 20) {
490 return error("index file %s is too small", path);
492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
511 index += 2; /* skip index header */
512 for (i = 0; i < 256; i++) {
513 uint32_t n = ntohl(index[i]);
515 munmap(idx_map, idx_size);
516 return error("non-monotonic index %s", path);
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
531 return error("wrong index v1 file size in %s", path);
533 } else if (version == 2) {
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size = min_size;
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
552 munmap(idx_map, idx_size);
553 return error("wrong index v2 file size in %s", path);
555 if (idx_size != min_size &&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
567 p->index_version = version;
568 p->index_data = idx_map;
569 p->index_size = idx_size;
574 int open_pack_index(struct packed_git *p)
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
589 static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
594 struct pack_window *w, *w_l;
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
608 static int unuse_one_window(struct packed_git *current, int keep_fd)
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
621 lru_l->next = lru_w->next;
623 lru_p->windows = lru_w->next;
624 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
625 close(lru_p->pack_fd);
636 void release_pack_memory(size_t need, int fd)
638 size_t cur = pack_mapped;
639 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
643 void close_pack_windows(struct packed_git *p)
646 struct pack_window *w = p->windows;
649 die("pack '%s' still has open windows to it",
651 munmap(w->base, w->len);
652 pack_mapped -= w->len;
654 p->windows = w->next;
659 void unuse_pack(struct pack_window **w_cursor)
661 struct pack_window *w = *w_cursor;
669 * Do not call this directly as this leaks p->pack_fd on error return;
670 * call open_packed_git() instead.
672 static int open_packed_git_1(struct packed_git *p)
675 struct pack_header hdr;
676 unsigned char sha1[20];
677 unsigned char *idx_sha1;
680 if (!p->index_data && open_pack_index(p))
681 return error("packfile %s index unavailable", p->pack_name);
683 p->pack_fd = open(p->pack_name, O_RDONLY);
684 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
687 /* If we created the struct before we had the pack we lack size. */
689 if (!S_ISREG(st.st_mode))
690 return error("packfile %s not a regular file", p->pack_name);
691 p->pack_size = st.st_size;
692 } else if (p->pack_size != st.st_size)
693 return error("packfile %s size changed", p->pack_name);
695 /* We leave these file descriptors open with sliding mmap;
696 * there is no point keeping them open across exec(), though.
698 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
700 return error("cannot determine file descriptor flags");
701 fd_flag |= FD_CLOEXEC;
702 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
703 return error("cannot set FD_CLOEXEC");
705 /* Verify we recognize this pack file format. */
706 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
707 return error("file %s is far too short to be a packfile", p->pack_name);
708 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
709 return error("file %s is not a GIT packfile", p->pack_name);
710 if (!pack_version_ok(hdr.hdr_version))
711 return error("packfile %s is version %"PRIu32" and not"
712 " supported (try upgrading GIT to a newer version)",
713 p->pack_name, ntohl(hdr.hdr_version));
715 /* Verify the pack matches its index. */
716 if (p->num_objects != ntohl(hdr.hdr_entries))
717 return error("packfile %s claims to have %"PRIu32" objects"
718 " while index indicates %"PRIu32" objects",
719 p->pack_name, ntohl(hdr.hdr_entries),
721 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
722 return error("end of packfile %s is unavailable", p->pack_name);
723 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
724 return error("packfile %s signature is unavailable", p->pack_name);
725 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
726 if (hashcmp(sha1, idx_sha1))
727 return error("packfile %s does not match index", p->pack_name);
731 static int open_packed_git(struct packed_git *p)
733 if (!open_packed_git_1(p))
735 if (p->pack_fd != -1) {
742 static int in_window(struct pack_window *win, off_t offset)
744 /* We must promise at least 20 bytes (one hash) after the
745 * offset is available from this window, otherwise the offset
746 * is not actually in this window and a different window (which
747 * has that one hash excess) must be used. This is to support
748 * the object header and delta base parsing routines below.
750 off_t win_off = win->offset;
751 return win_off <= offset
752 && (offset + 20) <= (win_off + win->len);
755 unsigned char* use_pack(struct packed_git *p,
756 struct pack_window **w_cursor,
760 struct pack_window *win = *w_cursor;
762 if (p->pack_fd == -1 && open_packed_git(p))
763 die("packfile %s cannot be accessed", p->pack_name);
765 /* Since packfiles end in a hash of their content and its
766 * pointless to ask for an offset into the middle of that
767 * hash, and the in_window function above wouldn't match
768 * don't allow an offset too close to the end of the file.
770 if (offset > (p->pack_size - 20))
771 die("offset beyond end of packfile (truncated pack?)");
773 if (!win || !in_window(win, offset)) {
776 for (win = p->windows; win; win = win->next) {
777 if (in_window(win, offset))
781 size_t window_align = packed_git_window_size / 2;
783 win = xcalloc(1, sizeof(*win));
784 win->offset = (offset / window_align) * window_align;
785 len = p->pack_size - win->offset;
786 if (len > packed_git_window_size)
787 len = packed_git_window_size;
788 win->len = (size_t)len;
789 pack_mapped += win->len;
790 while (packed_git_limit < pack_mapped
791 && unuse_one_window(p, p->pack_fd))
793 win->base = xmmap(NULL, win->len,
794 PROT_READ, MAP_PRIVATE,
795 p->pack_fd, win->offset);
796 if (win->base == MAP_FAILED)
797 die("packfile %s cannot be mapped: %s",
802 if (pack_mapped > peak_pack_mapped)
803 peak_pack_mapped = pack_mapped;
804 if (pack_open_windows > peak_pack_open_windows)
805 peak_pack_open_windows = pack_open_windows;
806 win->next = p->windows;
810 if (win != *w_cursor) {
811 win->last_used = pack_used_ctr++;
815 offset -= win->offset;
817 *left = win->len - xsize_t(offset);
818 return win->base + offset;
821 static struct packed_git *alloc_packed_git(int extra)
823 struct packed_git *p = xmalloc(sizeof(*p) + extra);
824 memset(p, 0, sizeof(*p));
829 struct packed_git *add_packed_git(const char *path, int path_len, int local)
832 struct packed_git *p = alloc_packed_git(path_len + 2);
835 * Make sure a corresponding .pack file exists and that
836 * the index looks sane.
838 path_len -= strlen(".idx");
843 memcpy(p->pack_name, path, path_len);
844 strcpy(p->pack_name + path_len, ".pack");
845 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
850 /* ok, it looks sane as far as we can check without
851 * actually mapping the pack file.
853 p->pack_size = st.st_size;
854 p->pack_local = local;
855 p->mtime = st.st_mtime;
856 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
861 struct packed_git *parse_pack_index(unsigned char *sha1)
863 const char *idx_path = sha1_pack_index_name(sha1);
864 const char *path = sha1_pack_name(sha1);
865 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
867 strcpy(p->pack_name, path);
868 hashcpy(p->sha1, sha1);
869 if (check_packed_git_idx(idx_path, p)) {
877 void install_packed_git(struct packed_git *pack)
879 pack->next = packed_git;
883 static void prepare_packed_git_one(char *objdir, int local)
885 /* Ensure that this buffer is large enough so that we can
886 append "/pack/" without clobbering the stack even if
887 strlen(objdir) were PATH_MAX. */
888 char path[PATH_MAX + 1 + 4 + 1 + 1];
893 sprintf(path, "%s/pack", objdir);
898 error("unable to open object pack directory: %s: %s",
899 path, strerror(errno));
903 while ((de = readdir(dir)) != NULL) {
904 int namelen = strlen(de->d_name);
905 struct packed_git *p;
907 if (!has_extension(de->d_name, ".idx"))
910 if (len + namelen + 1 > sizeof(path))
913 /* Don't reopen a pack we already have. */
914 strcpy(path + len, de->d_name);
915 for (p = packed_git; p; p = p->next) {
916 if (!memcmp(path, p->pack_name, len + namelen - 4))
921 /* See if it really is a valid .idx file with corresponding
922 * .pack file that we can map.
924 p = add_packed_git(path, len + namelen, local);
927 install_packed_git(p);
932 static int sort_pack(const void *a_, const void *b_)
934 struct packed_git *a = *((struct packed_git **)a_);
935 struct packed_git *b = *((struct packed_git **)b_);
939 * Local packs tend to contain objects specific to our
940 * variant of the project than remote ones. In addition,
941 * remote ones could be on a network mounted filesystem.
942 * Favor local ones for these reasons.
944 st = a->pack_local - b->pack_local;
949 * Younger packs tend to contain more recent objects,
950 * and more recent objects tend to get accessed more
953 if (a->mtime < b->mtime)
955 else if (a->mtime == b->mtime)
960 static void rearrange_packed_git(void)
962 struct packed_git **ary, *p;
965 for (n = 0, p = packed_git; p; p = p->next)
970 /* prepare an array of packed_git for easier sorting */
971 ary = xcalloc(n, sizeof(struct packed_git *));
972 for (n = 0, p = packed_git; p; p = p->next)
975 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
977 /* link them back again */
978 for (i = 0; i < n - 1; i++)
979 ary[i]->next = ary[i + 1];
980 ary[n - 1]->next = NULL;
986 static int prepare_packed_git_run_once = 0;
987 void prepare_packed_git(void)
989 struct alternate_object_database *alt;
991 if (prepare_packed_git_run_once)
993 prepare_packed_git_one(get_object_directory(), 1);
995 for (alt = alt_odb_list; alt; alt = alt->next) {
997 prepare_packed_git_one(alt->base, 0);
1000 rearrange_packed_git();
1001 prepare_packed_git_run_once = 1;
1004 void reprepare_packed_git(void)
1007 prepare_packed_git_run_once = 0;
1008 prepare_packed_git();
1011 static void mark_bad_packed_object(struct packed_git *p,
1012 const unsigned char *sha1)
1015 for (i = 0; i < p->num_bad_objects; i++)
1016 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1018 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1019 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1020 p->num_bad_objects++;
1023 static int has_packed_and_bad(const unsigned char *sha1)
1025 struct packed_git *p;
1028 for (p = packed_git; p; p = p->next)
1029 for (i = 0; i < p->num_bad_objects; i++)
1030 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1035 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1037 unsigned char real_sha1[20];
1038 hash_sha1_file(map, size, type, real_sha1);
1039 return hashcmp(sha1, real_sha1) ? -1 : 0;
1042 static int git_open_noatime(const char *name)
1044 static int sha1_file_open_flag = O_NOATIME;
1045 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1047 /* Might the failure be due to O_NOATIME? */
1048 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1049 fd = open(name, O_RDONLY);
1051 sha1_file_open_flag = 0;
1056 static int open_sha1_file(const unsigned char *sha1)
1059 char *name = sha1_file_name(sha1);
1060 struct alternate_object_database *alt;
1062 fd = git_open_noatime(name);
1068 for (alt = alt_odb_list; alt; alt = alt->next) {
1070 fill_sha1_path(name, sha1);
1071 fd = git_open_noatime(alt->base);
1078 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1083 fd = open_sha1_file(sha1);
1088 if (!fstat(fd, &st)) {
1089 *size = xsize_t(st.st_size);
1090 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1097 static int legacy_loose_object(unsigned char *map)
1102 * Is it a zlib-compressed buffer? If so, the first byte
1103 * must be 0x78 (15-bit window size, deflated), and the
1104 * first 16-bit word is evenly divisible by 31
1106 word = (map[0] << 8) + map[1];
1107 if (map[0] == 0x78 && !(word % 31))
1113 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1114 unsigned long len, enum object_type *type, unsigned long *sizep)
1119 unsigned long used = 0;
1122 *type = (c >> 4) & 7;
1126 if (len <= used || sizeof(long) * 8 <= shift) {
1127 error("bad object header");
1131 size += (c & 0x7f) << shift;
1138 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1140 unsigned long size, used;
1141 static const char valid_loose_object_type[8] = {
1143 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1144 0, /* "delta" and others are invalid in a loose object */
1146 enum object_type type;
1148 /* Get the data stream */
1149 memset(stream, 0, sizeof(*stream));
1150 stream->next_in = map;
1151 stream->avail_in = mapsize;
1152 stream->next_out = buffer;
1153 stream->avail_out = bufsiz;
1155 if (legacy_loose_object(map)) {
1156 inflateInit(stream);
1157 return inflate(stream, 0);
1162 * There used to be a second loose object header format which
1163 * was meant to mimic the in-pack format, allowing for direct
1164 * copy of the object data. This format turned up not to be
1165 * really worth it and we don't write it any longer. But we
1166 * can still read it.
1168 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1169 if (!used || !valid_loose_object_type[type])
1174 /* Set up the stream for the rest.. */
1175 stream->next_in = map;
1176 stream->avail_in = mapsize;
1177 inflateInit(stream);
1179 /* And generate the fake traditional header */
1180 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1181 typename(type), size);
1185 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1187 int bytes = strlen(buffer) + 1;
1188 unsigned char *buf = xmalloc(1+size);
1192 n = stream->total_out - bytes;
1195 memcpy(buf, (char *) buffer + bytes, n);
1197 if (bytes <= size) {
1199 * The above condition must be (bytes <= size), not
1200 * (bytes < size). In other words, even though we
1201 * expect no more output and set avail_out to zer0,
1202 * the input zlib stream may have bytes that express
1203 * "this concludes the stream", and we *do* want to
1206 * Otherwise we would not be able to test that we
1207 * consumed all the input to reach the expected size;
1208 * we also want to check that zlib tells us that all
1209 * went well with status == Z_STREAM_END at the end.
1211 stream->next_out = buf + bytes;
1212 stream->avail_out = size - bytes;
1213 while (status == Z_OK)
1214 status = inflate(stream, Z_FINISH);
1217 if (status == Z_STREAM_END && !stream->avail_in) {
1223 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1224 else if (stream->avail_in)
1225 error("garbage at end of loose object '%s'",
1232 * We used to just use "sscanf()", but that's actually way
1233 * too permissive for what we want to check. So do an anal
1234 * object header parse by hand.
1236 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1243 * The type can be at most ten bytes (including the
1244 * terminating '\0' that we add), and is followed by
1253 if (i >= sizeof(type))
1259 * The length must follow immediately, and be in canonical
1260 * decimal format (ie "010" is not valid).
1262 size = *hdr++ - '0';
1267 unsigned long c = *hdr - '0';
1271 size = size * 10 + c;
1277 * The length must be followed by a zero byte
1279 return *hdr ? -1 : type_from_string(type);
1282 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1288 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1289 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1292 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1295 unsigned long get_size_from_delta(struct packed_git *p,
1296 struct pack_window **w_curs,
1299 const unsigned char *data;
1300 unsigned char delta_head[20], *in;
1304 memset(&stream, 0, sizeof(stream));
1305 stream.next_out = delta_head;
1306 stream.avail_out = sizeof(delta_head);
1308 inflateInit(&stream);
1310 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1311 stream.next_in = in;
1312 st = inflate(&stream, Z_FINISH);
1313 curpos += stream.next_in - in;
1314 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1315 stream.total_out < sizeof(delta_head));
1316 inflateEnd(&stream);
1317 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1318 die("delta data unpack-initial failed");
1320 /* Examine the initial part of the delta to figure out
1325 /* ignore base size */
1326 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1328 /* Read the result size */
1329 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1332 static off_t get_delta_base(struct packed_git *p,
1333 struct pack_window **w_curs,
1335 enum object_type type,
1336 off_t delta_obj_offset)
1338 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1341 /* use_pack() assured us we have [base_info, base_info + 20)
1342 * as a range that we can look at without walking off the
1343 * end of the mapped window. Its actually the hash size
1344 * that is assured. An OFS_DELTA longer than the hash size
1345 * is stupid, as then a REF_DELTA would be smaller to store.
1347 if (type == OBJ_OFS_DELTA) {
1349 unsigned char c = base_info[used++];
1350 base_offset = c & 127;
1353 if (!base_offset || MSB(base_offset, 7))
1354 return 0; /* overflow */
1355 c = base_info[used++];
1356 base_offset = (base_offset << 7) + (c & 127);
1358 base_offset = delta_obj_offset - base_offset;
1359 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1360 return 0; /* out of bound */
1362 } else if (type == OBJ_REF_DELTA) {
1363 /* The base entry _must_ be in the same pack */
1364 base_offset = find_pack_entry_one(base_info, p);
1367 die("I am totally screwed");
1371 /* forward declaration for a mutually recursive function */
1372 static int packed_object_info(struct packed_git *p, off_t offset,
1373 unsigned long *sizep);
1375 static int packed_delta_info(struct packed_git *p,
1376 struct pack_window **w_curs,
1378 enum object_type type,
1380 unsigned long *sizep)
1384 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1385 type = packed_object_info(p, base_offset, NULL);
1387 /* We choose to only get the type of the base object and
1388 * ignore potentially corrupt pack file that expects the delta
1389 * based on a base with a wrong size. This saves tons of
1393 *sizep = get_size_from_delta(p, w_curs, curpos);
1398 static int unpack_object_header(struct packed_git *p,
1399 struct pack_window **w_curs,
1401 unsigned long *sizep)
1403 unsigned char *base;
1406 enum object_type type;
1408 /* use_pack() assures us we have [base, base + 20) available
1409 * as a range that we can look at at. (Its actually the hash
1410 * size that is assured.) With our object header encoding
1411 * the maximum deflated object size is 2^137, which is just
1412 * insane, so we know won't exceed what we have been given.
1414 base = use_pack(p, w_curs, *curpos, &left);
1415 used = unpack_object_header_buffer(base, left, &type, sizep);
1424 const char *packed_object_info_detail(struct packed_git *p,
1426 unsigned long *size,
1427 unsigned long *store_size,
1428 unsigned int *delta_chain_length,
1429 unsigned char *base_sha1)
1431 struct pack_window *w_curs = NULL;
1433 unsigned long dummy;
1434 unsigned char *next_sha1;
1435 enum object_type type;
1436 struct revindex_entry *revidx;
1438 *delta_chain_length = 0;
1439 curpos = obj_offset;
1440 type = unpack_object_header(p, &w_curs, &curpos, size);
1442 revidx = find_pack_revindex(p, obj_offset);
1443 *store_size = revidx[1].offset - obj_offset;
1448 die("pack %s contains unknown object type %d",
1449 p->pack_name, type);
1454 unuse_pack(&w_curs);
1455 return typename(type);
1457 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1459 die("pack %s contains bad delta base reference of type %s",
1460 p->pack_name, typename(type));
1461 if (*delta_chain_length == 0) {
1462 revidx = find_pack_revindex(p, obj_offset);
1463 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1467 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1468 if (*delta_chain_length == 0)
1469 hashcpy(base_sha1, next_sha1);
1470 obj_offset = find_pack_entry_one(next_sha1, p);
1473 (*delta_chain_length)++;
1474 curpos = obj_offset;
1475 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1479 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1480 unsigned long *sizep)
1482 struct pack_window *w_curs = NULL;
1484 off_t curpos = obj_offset;
1485 enum object_type type;
1487 type = unpack_object_header(p, &w_curs, &curpos, &size);
1492 type = packed_delta_info(p, &w_curs, curpos,
1493 type, obj_offset, sizep);
1503 die("pack %s contains unknown object type %d",
1504 p->pack_name, type);
1506 unuse_pack(&w_curs);
1510 static void *unpack_compressed_entry(struct packed_git *p,
1511 struct pack_window **w_curs,
1517 unsigned char *buffer, *in;
1519 buffer = xmalloc(size + 1);
1521 memset(&stream, 0, sizeof(stream));
1522 stream.next_out = buffer;
1523 stream.avail_out = size;
1525 inflateInit(&stream);
1527 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1528 stream.next_in = in;
1529 st = inflate(&stream, Z_FINISH);
1530 curpos += stream.next_in - in;
1531 } while (st == Z_OK || st == Z_BUF_ERROR);
1532 inflateEnd(&stream);
1533 if ((st != Z_STREAM_END) || stream.total_out != size) {
1541 #define MAX_DELTA_CACHE (256)
1543 static size_t delta_base_cached;
1545 static struct delta_base_cache_lru_list {
1546 struct delta_base_cache_lru_list *prev;
1547 struct delta_base_cache_lru_list *next;
1548 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1550 static struct delta_base_cache_entry {
1551 struct delta_base_cache_lru_list lru;
1553 struct packed_git *p;
1556 enum object_type type;
1557 } delta_base_cache[MAX_DELTA_CACHE];
1559 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1563 hash = (unsigned long)p + (unsigned long)base_offset;
1564 hash += (hash >> 8) + (hash >> 16);
1565 return hash % MAX_DELTA_CACHE;
1568 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1569 unsigned long *base_size, enum object_type *type, int keep_cache)
1572 unsigned long hash = pack_entry_hash(p, base_offset);
1573 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1576 if (!ret || ent->p != p || ent->base_offset != base_offset)
1577 return unpack_entry(p, base_offset, type, base_size);
1581 ent->lru.next->prev = ent->lru.prev;
1582 ent->lru.prev->next = ent->lru.next;
1583 delta_base_cached -= ent->size;
1585 ret = xmemdupz(ent->data, ent->size);
1588 *base_size = ent->size;
1592 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1597 ent->lru.next->prev = ent->lru.prev;
1598 ent->lru.prev->next = ent->lru.next;
1599 delta_base_cached -= ent->size;
1603 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1604 void *base, unsigned long base_size, enum object_type type)
1606 unsigned long hash = pack_entry_hash(p, base_offset);
1607 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1608 struct delta_base_cache_lru_list *lru;
1610 release_delta_base_cache(ent);
1611 delta_base_cached += base_size;
1613 for (lru = delta_base_cache_lru.next;
1614 delta_base_cached > delta_base_cache_limit
1615 && lru != &delta_base_cache_lru;
1617 struct delta_base_cache_entry *f = (void *)lru;
1618 if (f->type == OBJ_BLOB)
1619 release_delta_base_cache(f);
1621 for (lru = delta_base_cache_lru.next;
1622 delta_base_cached > delta_base_cache_limit
1623 && lru != &delta_base_cache_lru;
1625 struct delta_base_cache_entry *f = (void *)lru;
1626 release_delta_base_cache(f);
1630 ent->base_offset = base_offset;
1633 ent->size = base_size;
1634 ent->lru.next = &delta_base_cache_lru;
1635 ent->lru.prev = delta_base_cache_lru.prev;
1636 delta_base_cache_lru.prev->next = &ent->lru;
1637 delta_base_cache_lru.prev = &ent->lru;
1640 static void *unpack_delta_entry(struct packed_git *p,
1641 struct pack_window **w_curs,
1643 unsigned long delta_size,
1645 enum object_type *type,
1646 unsigned long *sizep)
1648 void *delta_data, *result, *base;
1649 unsigned long base_size;
1652 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1654 error("failed to validate delta base reference "
1655 "at offset %"PRIuMAX" from %s",
1656 (uintmax_t)curpos, p->pack_name);
1660 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1663 * We're probably in deep shit, but let's try to fetch
1664 * the required base anyway from another pack or loose.
1665 * This is costly but should happen only in the presence
1666 * of a corrupted pack, and is better than failing outright.
1668 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1669 const unsigned char *base_sha1 =
1670 nth_packed_object_sha1(p, revidx->nr);
1671 error("failed to read delta base object %s"
1672 " at offset %"PRIuMAX" from %s",
1673 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1675 mark_bad_packed_object(p, base_sha1);
1676 base = read_object(base_sha1, type, &base_size);
1681 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1683 error("failed to unpack compressed delta "
1684 "at offset %"PRIuMAX" from %s",
1685 (uintmax_t)curpos, p->pack_name);
1689 result = patch_delta(base, base_size,
1690 delta_data, delta_size,
1693 die("failed to apply delta");
1695 add_delta_base_cache(p, base_offset, base, base_size, *type);
1699 int do_check_packed_object_crc;
1701 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1702 enum object_type *type, unsigned long *sizep)
1704 struct pack_window *w_curs = NULL;
1705 off_t curpos = obj_offset;
1708 if (do_check_packed_object_crc && p->index_version > 1) {
1709 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1710 unsigned long len = revidx[1].offset - obj_offset;
1711 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1712 const unsigned char *sha1 =
1713 nth_packed_object_sha1(p, revidx->nr);
1714 error("bad packed object CRC for %s",
1716 mark_bad_packed_object(p, sha1);
1721 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1725 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1726 obj_offset, type, sizep);
1732 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1736 error("unknown object type %i at offset %"PRIuMAX" in %s",
1737 *type, (uintmax_t)obj_offset, p->pack_name);
1739 unuse_pack(&w_curs);
1743 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1746 const unsigned char *index = p->index_data;
1748 if (open_pack_index(p))
1750 index = p->index_data;
1752 if (n >= p->num_objects)
1755 if (p->index_version == 1) {
1756 return index + 24 * n + 4;
1759 return index + 20 * n;
1763 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1765 const unsigned char *index = p->index_data;
1767 if (p->index_version == 1) {
1768 return ntohl(*((uint32_t *)(index + 24 * n)));
1771 index += 8 + p->num_objects * (20 + 4);
1772 off = ntohl(*((uint32_t *)(index + 4 * n)));
1773 if (!(off & 0x80000000))
1775 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1776 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1777 ntohl(*((uint32_t *)(index + 4)));
1781 off_t find_pack_entry_one(const unsigned char *sha1,
1782 struct packed_git *p)
1784 const uint32_t *level1_ofs = p->index_data;
1785 const unsigned char *index = p->index_data;
1786 unsigned hi, lo, stride;
1787 static int use_lookup = -1;
1788 static int debug_lookup = -1;
1790 if (debug_lookup < 0)
1791 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1794 if (open_pack_index(p))
1796 level1_ofs = p->index_data;
1797 index = p->index_data;
1799 if (p->index_version > 1) {
1804 hi = ntohl(level1_ofs[*sha1]);
1805 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1806 if (p->index_version > 1) {
1814 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1815 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1818 use_lookup = !!getenv("GIT_USE_LOOKUP");
1820 int pos = sha1_entry_pos(index, stride, 0,
1821 lo, hi, p->num_objects, sha1);
1824 return nth_packed_object_offset(p, pos);
1828 unsigned mi = (lo + hi) / 2;
1829 int cmp = hashcmp(index + mi * stride, sha1);
1832 printf("lo %u hi %u rg %u mi %u\n",
1833 lo, hi, hi - lo, mi);
1835 return nth_packed_object_offset(p, mi);
1844 int matches_pack_name(struct packed_git *p, const char *name)
1846 const char *last_c, *c;
1848 if (!strcmp(p->pack_name, name))
1851 for (c = p->pack_name, last_c = c; *c;)
1856 if (!strcmp(last_c, name))
1862 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1864 static struct packed_git *last_found = (void *)1;
1865 struct packed_git *p;
1868 prepare_packed_git();
1871 p = (last_found == (void *)1) ? packed_git : last_found;
1874 if (ignore_packed) {
1876 for (ig = ignore_packed; *ig; ig++)
1877 if (matches_pack_name(p, *ig))
1883 if (p->num_bad_objects) {
1885 for (i = 0; i < p->num_bad_objects; i++)
1886 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1890 offset = find_pack_entry_one(sha1, p);
1893 * We are about to tell the caller where they can
1894 * locate the requested object. We better make
1895 * sure the packfile is still here and can be
1896 * accessed before supplying that answer, as
1897 * it may have been deleted since the index
1900 if (p->pack_fd == -1 && open_packed_git(p)) {
1901 error("packfile %s cannot be accessed", p->pack_name);
1906 hashcpy(e->sha1, sha1);
1912 if (p == last_found)
1916 if (p == last_found)
1922 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1923 struct packed_git *packs)
1925 struct packed_git *p;
1927 for (p = packs; p; p = p->next) {
1928 if (find_pack_entry_one(sha1, p))
1935 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1938 unsigned long mapsize, size;
1943 map = map_sha1_file(sha1, &mapsize);
1945 return error("unable to find %s", sha1_to_hex(sha1));
1946 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1947 status = error("unable to unpack %s header",
1949 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1950 status = error("unable to parse %s header", sha1_to_hex(sha1));
1953 inflateEnd(&stream);
1954 munmap(map, mapsize);
1958 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1960 struct pack_entry e;
1963 if (!find_pack_entry(sha1, &e, NULL)) {
1964 /* Most likely it's a loose object. */
1965 status = sha1_loose_object_info(sha1, sizep);
1969 /* Not a loose object; someone else may have just packed it. */
1970 reprepare_packed_git();
1971 if (!find_pack_entry(sha1, &e, NULL))
1974 return packed_object_info(e.p, e.offset, sizep);
1977 static void *read_packed_sha1(const unsigned char *sha1,
1978 enum object_type *type, unsigned long *size)
1980 struct pack_entry e;
1983 if (!find_pack_entry(sha1, &e, NULL))
1985 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1988 * We're probably in deep shit, but let's try to fetch
1989 * the required object anyway from another pack or loose.
1990 * This should happen only in the presence of a corrupted
1991 * pack, and is better than failing outright.
1993 error("failed to read object %s at offset %"PRIuMAX" from %s",
1994 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1995 mark_bad_packed_object(e.p, sha1);
1996 data = read_object(sha1, type, size);
2002 * This is meant to hold a *small* number of objects that you would
2003 * want read_sha1_file() to be able to return, but yet you do not want
2004 * to write them into the object store (e.g. a browse-only
2007 static struct cached_object {
2008 unsigned char sha1[20];
2009 enum object_type type;
2013 static int cached_object_nr, cached_object_alloc;
2015 static struct cached_object empty_tree = {
2016 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2017 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2018 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2024 static struct cached_object *find_cached_object(const unsigned char *sha1)
2027 struct cached_object *co = cached_objects;
2029 for (i = 0; i < cached_object_nr; i++, co++) {
2030 if (!hashcmp(co->sha1, sha1))
2033 if (!hashcmp(sha1, empty_tree.sha1))
2038 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2039 unsigned char *sha1)
2041 struct cached_object *co;
2043 hash_sha1_file(buf, len, typename(type), sha1);
2044 if (has_sha1_file(sha1) || find_cached_object(sha1))
2046 if (cached_object_alloc <= cached_object_nr) {
2047 cached_object_alloc = alloc_nr(cached_object_alloc);
2048 cached_objects = xrealloc(cached_objects,
2049 sizeof(*cached_objects) *
2050 cached_object_alloc);
2052 co = &cached_objects[cached_object_nr++];
2055 co->buf = xmalloc(len);
2056 memcpy(co->buf, buf, len);
2057 hashcpy(co->sha1, sha1);
2061 void *read_object(const unsigned char *sha1, enum object_type *type,
2062 unsigned long *size)
2064 unsigned long mapsize;
2066 struct cached_object *co;
2068 co = find_cached_object(sha1);
2072 return xmemdupz(co->buf, co->size);
2075 buf = read_packed_sha1(sha1, type, size);
2078 map = map_sha1_file(sha1, &mapsize);
2080 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2081 munmap(map, mapsize);
2084 reprepare_packed_git();
2085 return read_packed_sha1(sha1, type, size);
2088 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2089 unsigned long *size)
2091 void *data = read_object(sha1, type, size);
2092 /* legacy behavior is to die on corrupted objects */
2093 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2094 die("object %s is corrupted", sha1_to_hex(sha1));
2098 void *read_object_with_reference(const unsigned char *sha1,
2099 const char *required_type_name,
2100 unsigned long *size,
2101 unsigned char *actual_sha1_return)
2103 enum object_type type, required_type;
2105 unsigned long isize;
2106 unsigned char actual_sha1[20];
2108 required_type = type_from_string(required_type_name);
2109 hashcpy(actual_sha1, sha1);
2111 int ref_length = -1;
2112 const char *ref_type = NULL;
2114 buffer = read_sha1_file(actual_sha1, &type, &isize);
2117 if (type == required_type) {
2119 if (actual_sha1_return)
2120 hashcpy(actual_sha1_return, actual_sha1);
2123 /* Handle references */
2124 else if (type == OBJ_COMMIT)
2126 else if (type == OBJ_TAG)
2127 ref_type = "object ";
2132 ref_length = strlen(ref_type);
2134 if (ref_length + 40 > isize ||
2135 memcmp(buffer, ref_type, ref_length) ||
2136 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2141 /* Now we have the ID of the referred-to object in
2142 * actual_sha1. Check again. */
2146 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2147 const char *type, unsigned char *sha1,
2148 char *hdr, int *hdrlen)
2152 /* Generate the header */
2153 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2157 git_SHA1_Update(&c, hdr, *hdrlen);
2158 git_SHA1_Update(&c, buf, len);
2159 git_SHA1_Final(sha1, &c);
2163 * Move the just written object into its final resting place
2165 int move_temp_to_file(const char *tmpfile, const char *filename)
2168 if (link(tmpfile, filename))
2172 * Coda hack - coda doesn't like cross-directory links,
2173 * so we fall back to a rename, which will mean that it
2174 * won't be able to check collisions, but that's not a
2177 * The same holds for FAT formatted media.
2179 * When this succeeds, we just return 0. We have nothing
2182 if (ret && ret != EEXIST) {
2183 if (!rename(tmpfile, filename))
2189 if (ret != EEXIST) {
2190 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2192 /* FIXME!!! Collision check here ? */
2198 static int write_buffer(int fd, const void *buf, size_t len)
2200 if (write_in_full(fd, buf, len) < 0)
2201 return error("file write error (%s)", strerror(errno));
2205 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2206 unsigned char *sha1)
2210 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2214 /* Finalize a file on disk, and close it. */
2215 static void close_sha1_file(int fd)
2217 if (fsync_object_files)
2218 fsync_or_die(fd, "sha1 file");
2221 die("unable to write sha1 file");
2224 /* Size of directory component, including the ending '/' */
2225 static inline int directory_size(const char *filename)
2227 const char *s = strrchr(filename, '/');
2230 return s - filename + 1;
2234 * This creates a temporary file in the same directory as the final
2237 * We want to avoid cross-directory filename renames, because those
2238 * can have problems on various filesystems (FAT, NFS, Coda).
2240 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2242 int fd, dirlen = directory_size(filename);
2244 if (dirlen + 20 > bufsiz) {
2245 errno = ENAMETOOLONG;
2248 memcpy(buffer, filename, dirlen);
2249 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2250 fd = mkstemp(buffer);
2251 if (fd < 0 && dirlen) {
2252 /* Make sure the directory exists */
2253 memcpy(buffer, filename, dirlen);
2254 buffer[dirlen-1] = 0;
2255 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2259 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2260 fd = mkstemp(buffer);
2265 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2266 void *buf, unsigned long len, time_t mtime)
2269 unsigned char *compressed;
2272 static char tmpfile[PATH_MAX];
2274 filename = sha1_file_name(sha1);
2275 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2278 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2280 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2284 memset(&stream, 0, sizeof(stream));
2285 deflateInit(&stream, zlib_compression_level);
2286 size = 8 + deflateBound(&stream, len+hdrlen);
2287 compressed = xmalloc(size);
2290 stream.next_out = compressed;
2291 stream.avail_out = size;
2293 /* First header.. */
2294 stream.next_in = (unsigned char *)hdr;
2295 stream.avail_in = hdrlen;
2296 while (deflate(&stream, 0) == Z_OK)
2299 /* Then the data itself.. */
2300 stream.next_in = buf;
2301 stream.avail_in = len;
2302 ret = deflate(&stream, Z_FINISH);
2303 if (ret != Z_STREAM_END)
2304 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2306 ret = deflateEnd(&stream);
2308 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2310 size = stream.total_out;
2312 if (write_buffer(fd, compressed, size) < 0)
2313 die("unable to write sha1 file");
2314 close_sha1_file(fd);
2320 utb.modtime = mtime;
2321 if (utime(tmpfile, &utb) < 0)
2322 warning("failed utime() on %s: %s",
2323 tmpfile, strerror(errno));
2326 return move_temp_to_file(tmpfile, filename);
2329 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2331 unsigned char sha1[20];
2335 /* Normally if we have it in the pack then we do not bother writing
2336 * it out into .git/objects/??/?{38} file.
2338 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2340 hashcpy(returnsha1, sha1);
2341 if (has_sha1_file(sha1))
2343 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2346 int force_object_loose(const unsigned char *sha1, time_t mtime)
2350 enum object_type type;
2355 if (has_loose_object(sha1))
2357 buf = read_packed_sha1(sha1, &type, &len);
2359 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2360 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2361 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2367 int has_pack_index(const unsigned char *sha1)
2370 if (stat(sha1_pack_index_name(sha1), &st))
2375 int has_pack_file(const unsigned char *sha1)
2378 if (stat(sha1_pack_name(sha1), &st))
2383 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2385 struct pack_entry e;
2386 return find_pack_entry(sha1, &e, ignore_packed);
2389 int has_sha1_file(const unsigned char *sha1)
2391 struct pack_entry e;
2393 if (find_pack_entry(sha1, &e, NULL))
2395 return has_loose_object(sha1);
2398 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2399 int write_object, enum object_type type, const char *path)
2401 int ret, re_allocated = 0;
2407 * Convert blobs to git internal format
2409 if ((type == OBJ_BLOB) && path) {
2410 struct strbuf nbuf = STRBUF_INIT;
2411 if (convert_to_git(path, buf, size, &nbuf,
2412 write_object ? safe_crlf : 0)) {
2413 buf = strbuf_detach(&nbuf, &size);
2419 ret = write_sha1_file(buf, size, typename(type), sha1);
2421 ret = hash_sha1_file(buf, size, typename(type), sha1);
2427 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2428 enum object_type type, const char *path)
2431 size_t size = xsize_t(st->st_size);
2433 if (!S_ISREG(st->st_mode)) {
2434 struct strbuf sbuf = STRBUF_INIT;
2435 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2436 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2440 strbuf_release(&sbuf);
2442 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2443 ret = index_mem(sha1, buf, size, write_object, type, path);
2446 ret = index_mem(sha1, NULL, size, write_object, type, path);
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;