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 char *sha1_to_hex(const unsigned char *sha1)
126 static char hexbuffer[4][50];
127 static const char hex[] = "0123456789abcdef";
128 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
131 for (i = 0; i < 20; i++) {
132 unsigned int val = *sha1++;
133 *buf++ = hex[val >> 4];
134 *buf++ = hex[val & 0xf];
141 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
144 for (i = 0; i < 20; i++) {
145 static char hex[] = "0123456789abcdef";
146 unsigned int val = sha1[i];
147 char *pos = pathbuf + i*2 + (i > 0);
148 *pos++ = hex[val >> 4];
149 *pos = hex[val & 0xf];
154 * NOTE! This returns a statically allocated buffer, so you have to be
155 * careful about using it. Do an "xstrdup()" if you need to save the
158 * Also note that this returns the location for creating. Reading
159 * SHA1 file can happen from any alternate directory listed in the
160 * DB_ENVIRONMENT environment variable if it is not found in
161 * the primary object database.
163 char *sha1_file_name(const unsigned char *sha1)
165 static char *name, *base;
168 const char *sha1_file_directory = get_object_directory();
169 int len = strlen(sha1_file_directory);
170 base = xmalloc(len + 60);
171 memcpy(base, sha1_file_directory, len);
172 memset(base+len, 0, 60);
175 name = base + len + 1;
177 fill_sha1_path(name, sha1);
181 static char *sha1_get_pack_name(const unsigned char *sha1,
182 char **name, char **base, const char *which)
184 static const char hex[] = "0123456789abcdef";
189 const char *sha1_file_directory = get_object_directory();
190 int len = strlen(sha1_file_directory);
191 *base = xmalloc(len + 60);
192 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
193 sha1_file_directory, which);
194 *name = *base + len + 11;
199 for (i = 0; i < 20; i++) {
200 unsigned int val = *sha1++;
201 *buf++ = hex[val >> 4];
202 *buf++ = hex[val & 0xf];
208 char *sha1_pack_name(const unsigned char *sha1)
210 static char *name, *base;
212 return sha1_get_pack_name(sha1, &name, &base, "pack");
215 char *sha1_pack_index_name(const unsigned char *sha1)
217 static char *name, *base;
219 return sha1_get_pack_name(sha1, &name, &base, "idx");
222 struct alternate_object_database *alt_odb_list;
223 static struct alternate_object_database **alt_odb_tail;
225 static void read_info_alternates(const char * alternates, int depth);
228 * Prepare alternate object database registry.
230 * The variable alt_odb_list points at the list of struct
231 * alternate_object_database. The elements on this list come from
232 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
233 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
234 * whose contents is similar to that environment variable but can be
235 * LF separated. Its base points at a statically allocated buffer that
236 * contains "/the/directory/corresponding/to/.git/objects/...", while
237 * its name points just after the slash at the end of ".git/objects/"
238 * in the example above, and has enough space to hold 40-byte hex
239 * SHA1, an extra slash for the first level indirection, and the
242 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
245 const char *objdir = get_object_directory();
246 struct alternate_object_database *ent;
247 struct alternate_object_database *alt;
248 /* 43 = 40-byte + 2 '/' + terminating NUL */
250 int entlen = pfxlen + 43;
253 if (!is_absolute_path(entry) && relative_base) {
254 /* Relative alt-odb */
256 base_len = strlen(relative_base) + 1;
260 ent = xmalloc(sizeof(*ent) + entlen);
262 if (!is_absolute_path(entry) && relative_base) {
263 memcpy(ent->base, relative_base, base_len - 1);
264 ent->base[base_len - 1] = '/';
265 memcpy(ent->base + base_len, entry, len);
268 memcpy(ent->base, entry, pfxlen);
270 ent->name = ent->base + pfxlen + 1;
271 ent->base[pfxlen + 3] = '/';
272 ent->base[pfxlen] = ent->base[entlen-1] = 0;
274 /* Detect cases where alternate disappeared */
275 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
276 error("object directory %s does not exist; "
277 "check .git/objects/info/alternates.",
283 /* Prevent the common mistake of listing the same
284 * thing twice, or object directory itself.
286 for (alt = alt_odb_list; alt; alt = alt->next) {
287 if (!memcmp(ent->base, alt->base, pfxlen)) {
292 if (!memcmp(ent->base, objdir, pfxlen)) {
297 /* add the alternate entry */
299 alt_odb_tail = &(ent->next);
302 /* recursively add alternates */
303 read_info_alternates(ent->base, depth + 1);
305 ent->base[pfxlen] = '/';
310 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
311 const char *relative_base, int depth)
313 const char *cp, *last;
316 error("%s: ignoring alternate object stores, nesting too deep.",
324 if (cp < ep && *cp == '#') {
325 while (cp < ep && *cp != sep)
330 while (cp < ep && *cp != sep)
333 if (!is_absolute_path(last) && depth) {
334 error("%s: ignoring relative alternate object store %s",
335 relative_base, last);
337 link_alt_odb_entry(last, cp - last,
338 relative_base, depth);
341 while (cp < ep && *cp == sep)
347 static void read_info_alternates(const char * relative_base, int depth)
352 const char alt_file_name[] = "info/alternates";
353 /* Given that relative_base is no longer than PATH_MAX,
354 ensure that "path" has enough space to append "/", the
355 file name, "info/alternates", and a trailing NUL. */
356 char path[PATH_MAX + 1 + sizeof alt_file_name];
359 sprintf(path, "%s/%s", relative_base, alt_file_name);
360 fd = open(path, O_RDONLY);
363 if (fstat(fd, &st) || (st.st_size == 0)) {
367 mapsz = xsize_t(st.st_size);
368 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
371 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376 void add_to_alternates_file(const char *reference)
378 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
379 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
380 char *alt = mkpath("%s/objects\n", reference);
381 write_or_die(fd, alt, strlen(alt));
382 if (commit_lock_file(lock))
383 die("could not close alternates file");
385 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
388 void prepare_alt_odb(void)
395 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
398 alt_odb_tail = &alt_odb_list;
399 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
401 read_info_alternates(get_object_directory(), 0);
404 static int has_loose_object(const unsigned char *sha1)
406 char *name = sha1_file_name(sha1);
407 struct alternate_object_database *alt;
409 if (!access(name, F_OK))
412 for (alt = alt_odb_list; alt; alt = alt->next) {
414 fill_sha1_path(name, sha1);
415 if (!access(alt->base, F_OK))
421 static unsigned int pack_used_ctr;
422 static unsigned int pack_mmap_calls;
423 static unsigned int peak_pack_open_windows;
424 static unsigned int pack_open_windows;
425 static size_t peak_pack_mapped;
426 static size_t pack_mapped;
427 struct packed_git *packed_git;
429 void pack_report(void)
432 "pack_report: getpagesize() = %10" SZ_FMT "\n"
433 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
434 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
435 sz_fmt(getpagesize()),
436 sz_fmt(packed_git_window_size),
437 sz_fmt(packed_git_limit));
439 "pack_report: pack_used_ctr = %10u\n"
440 "pack_report: pack_mmap_calls = %10u\n"
441 "pack_report: pack_open_windows = %10u / %10u\n"
442 "pack_report: pack_mapped = "
443 "%10" SZ_FMT " / %10" SZ_FMT "\n",
446 pack_open_windows, peak_pack_open_windows,
447 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
450 static int check_packed_git_idx(const char *path, struct packed_git *p)
453 struct pack_idx_header *hdr;
455 uint32_t version, nr, i, *index;
456 int fd = open(path, O_RDONLY);
461 if (fstat(fd, &st)) {
465 idx_size = xsize_t(st.st_size);
466 if (idx_size < 4 * 256 + 20 + 20) {
468 return error("index file %s is too small", path);
470 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
474 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
475 version = ntohl(hdr->idx_version);
476 if (version < 2 || version > 2) {
477 munmap(idx_map, idx_size);
478 return error("index file %s is version %d"
479 " and is not supported by this binary"
480 " (try upgrading GIT to a newer version)",
489 index += 2; /* skip index header */
490 for (i = 0; i < 256; i++) {
491 uint32_t n = ntohl(index[i]);
493 munmap(idx_map, idx_size);
494 return error("non-monotonic index %s", path);
502 * - 256 index entries 4 bytes each
503 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
504 * - 20-byte SHA1 of the packfile
505 * - 20-byte SHA1 file checksum
507 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
508 munmap(idx_map, idx_size);
509 return error("wrong index v1 file size in %s", path);
511 } else if (version == 2) {
514 * - 8 bytes of header
515 * - 256 index entries 4 bytes each
516 * - 20-byte sha1 entry * nr
517 * - 4-byte crc entry * nr
518 * - 4-byte offset entry * nr
519 * - 20-byte SHA1 of the packfile
520 * - 20-byte SHA1 file checksum
521 * And after the 4-byte offset table might be a
522 * variable sized table containing 8-byte entries
523 * for offsets larger than 2^31.
525 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
526 unsigned long max_size = min_size;
528 max_size += (nr - 1)*8;
529 if (idx_size < min_size || idx_size > max_size) {
530 munmap(idx_map, idx_size);
531 return error("wrong index v2 file size in %s", path);
533 if (idx_size != min_size &&
535 * make sure we can deal with large pack offsets.
536 * 31-bit signed offset won't be enough, neither
537 * 32-bit unsigned one will be.
539 (sizeof(off_t) <= 4)) {
540 munmap(idx_map, idx_size);
541 return error("pack too large for current definition of off_t in %s", path);
545 p->index_version = version;
546 p->index_data = idx_map;
547 p->index_size = idx_size;
552 int open_pack_index(struct packed_git *p)
560 idx_name = xstrdup(p->pack_name);
561 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
562 ret = check_packed_git_idx(idx_name, p);
567 static void scan_windows(struct packed_git *p,
568 struct packed_git **lru_p,
569 struct pack_window **lru_w,
570 struct pack_window **lru_l)
572 struct pack_window *w, *w_l;
574 for (w_l = NULL, w = p->windows; w; w = w->next) {
576 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
586 static int unuse_one_window(struct packed_git *current, int keep_fd)
588 struct packed_git *p, *lru_p = NULL;
589 struct pack_window *lru_w = NULL, *lru_l = NULL;
592 scan_windows(current, &lru_p, &lru_w, &lru_l);
593 for (p = packed_git; p; p = p->next)
594 scan_windows(p, &lru_p, &lru_w, &lru_l);
596 munmap(lru_w->base, lru_w->len);
597 pack_mapped -= lru_w->len;
599 lru_l->next = lru_w->next;
601 lru_p->windows = lru_w->next;
602 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
603 close(lru_p->pack_fd);
614 void release_pack_memory(size_t need, int fd)
616 size_t cur = pack_mapped;
617 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
621 void close_pack_windows(struct packed_git *p)
624 struct pack_window *w = p->windows;
627 die("pack '%s' still has open windows to it",
629 munmap(w->base, w->len);
630 pack_mapped -= w->len;
632 p->windows = w->next;
637 void unuse_pack(struct pack_window **w_cursor)
639 struct pack_window *w = *w_cursor;
647 * Do not call this directly as this leaks p->pack_fd on error return;
648 * call open_packed_git() instead.
650 static int open_packed_git_1(struct packed_git *p)
653 struct pack_header hdr;
654 unsigned char sha1[20];
655 unsigned char *idx_sha1;
658 if (!p->index_data && open_pack_index(p))
659 return error("packfile %s index unavailable", p->pack_name);
661 p->pack_fd = open(p->pack_name, O_RDONLY);
662 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
665 /* If we created the struct before we had the pack we lack size. */
667 if (!S_ISREG(st.st_mode))
668 return error("packfile %s not a regular file", p->pack_name);
669 p->pack_size = st.st_size;
670 } else if (p->pack_size != st.st_size)
671 return error("packfile %s size changed", p->pack_name);
673 /* We leave these file descriptors open with sliding mmap;
674 * there is no point keeping them open across exec(), though.
676 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
678 return error("cannot determine file descriptor flags");
679 fd_flag |= FD_CLOEXEC;
680 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
681 return error("cannot set FD_CLOEXEC");
683 /* Verify we recognize this pack file format. */
684 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
685 return error("file %s is far too short to be a packfile", p->pack_name);
686 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
687 return error("file %s is not a GIT packfile", p->pack_name);
688 if (!pack_version_ok(hdr.hdr_version))
689 return error("packfile %s is version %u and not supported"
690 " (try upgrading GIT to a newer version)",
691 p->pack_name, ntohl(hdr.hdr_version));
693 /* Verify the pack matches its index. */
694 if (p->num_objects != ntohl(hdr.hdr_entries))
695 return error("packfile %s claims to have %u objects"
696 " while index indicates %u objects",
697 p->pack_name, ntohl(hdr.hdr_entries),
699 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
700 return error("end of packfile %s is unavailable", p->pack_name);
701 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
702 return error("packfile %s signature is unavailable", p->pack_name);
703 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
704 if (hashcmp(sha1, idx_sha1))
705 return error("packfile %s does not match index", p->pack_name);
709 static int open_packed_git(struct packed_git *p)
711 if (!open_packed_git_1(p))
713 if (p->pack_fd != -1) {
720 static int in_window(struct pack_window *win, off_t offset)
722 /* We must promise at least 20 bytes (one hash) after the
723 * offset is available from this window, otherwise the offset
724 * is not actually in this window and a different window (which
725 * has that one hash excess) must be used. This is to support
726 * the object header and delta base parsing routines below.
728 off_t win_off = win->offset;
729 return win_off <= offset
730 && (offset + 20) <= (win_off + win->len);
733 unsigned char* use_pack(struct packed_git *p,
734 struct pack_window **w_cursor,
738 struct pack_window *win = *w_cursor;
740 if (p->pack_fd == -1 && open_packed_git(p))
741 die("packfile %s cannot be accessed", p->pack_name);
743 /* Since packfiles end in a hash of their content and its
744 * pointless to ask for an offset into the middle of that
745 * hash, and the in_window function above wouldn't match
746 * don't allow an offset too close to the end of the file.
748 if (offset > (p->pack_size - 20))
749 die("offset beyond end of packfile (truncated pack?)");
751 if (!win || !in_window(win, offset)) {
754 for (win = p->windows; win; win = win->next) {
755 if (in_window(win, offset))
759 size_t window_align = packed_git_window_size / 2;
761 win = xcalloc(1, sizeof(*win));
762 win->offset = (offset / window_align) * window_align;
763 len = p->pack_size - win->offset;
764 if (len > packed_git_window_size)
765 len = packed_git_window_size;
766 win->len = (size_t)len;
767 pack_mapped += win->len;
768 while (packed_git_limit < pack_mapped
769 && unuse_one_window(p, p->pack_fd))
771 win->base = xmmap(NULL, win->len,
772 PROT_READ, MAP_PRIVATE,
773 p->pack_fd, win->offset);
774 if (win->base == MAP_FAILED)
775 die("packfile %s cannot be mapped: %s",
780 if (pack_mapped > peak_pack_mapped)
781 peak_pack_mapped = pack_mapped;
782 if (pack_open_windows > peak_pack_open_windows)
783 peak_pack_open_windows = pack_open_windows;
784 win->next = p->windows;
788 if (win != *w_cursor) {
789 win->last_used = pack_used_ctr++;
793 offset -= win->offset;
795 *left = win->len - xsize_t(offset);
796 return win->base + offset;
799 struct packed_git *add_packed_git(const char *path, int path_len, int local)
802 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
805 * Make sure a corresponding .pack file exists and that
806 * the index looks sane.
808 path_len -= strlen(".idx");
811 memcpy(p->pack_name, path, path_len);
812 strcpy(p->pack_name + path_len, ".pack");
813 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
818 /* ok, it looks sane as far as we can check without
819 * actually mapping the pack file.
821 p->index_version = 0;
822 p->index_data = NULL;
825 p->pack_size = st.st_size;
829 p->pack_local = local;
830 p->mtime = st.st_mtime;
831 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
836 struct packed_git *parse_pack_index(unsigned char *sha1)
838 const char *idx_path = sha1_pack_index_name(sha1);
839 const char *path = sha1_pack_name(sha1);
840 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
842 if (check_packed_git_idx(idx_path, p)) {
847 strcpy(p->pack_name, path);
852 hashcpy(p->sha1, sha1);
856 void install_packed_git(struct packed_git *pack)
858 pack->next = packed_git;
862 static void prepare_packed_git_one(char *objdir, int local)
864 /* Ensure that this buffer is large enough so that we can
865 append "/pack/" without clobbering the stack even if
866 strlen(objdir) were PATH_MAX. */
867 char path[PATH_MAX + 1 + 4 + 1 + 1];
872 sprintf(path, "%s/pack", objdir);
877 error("unable to open object pack directory: %s: %s",
878 path, strerror(errno));
882 while ((de = readdir(dir)) != NULL) {
883 int namelen = strlen(de->d_name);
884 struct packed_git *p;
886 if (!has_extension(de->d_name, ".idx"))
889 if (len + namelen + 1 > sizeof(path))
892 /* Don't reopen a pack we already have. */
893 strcpy(path + len, de->d_name);
894 for (p = packed_git; p; p = p->next) {
895 if (!memcmp(path, p->pack_name, len + namelen - 4))
900 /* See if it really is a valid .idx file with corresponding
901 * .pack file that we can map.
903 p = add_packed_git(path, len + namelen, local);
906 install_packed_git(p);
911 static int sort_pack(const void *a_, const void *b_)
913 struct packed_git *a = *((struct packed_git **)a_);
914 struct packed_git *b = *((struct packed_git **)b_);
918 * Local packs tend to contain objects specific to our
919 * variant of the project than remote ones. In addition,
920 * remote ones could be on a network mounted filesystem.
921 * Favor local ones for these reasons.
923 st = a->pack_local - b->pack_local;
928 * Younger packs tend to contain more recent objects,
929 * and more recent objects tend to get accessed more
932 if (a->mtime < b->mtime)
934 else if (a->mtime == b->mtime)
939 static void rearrange_packed_git(void)
941 struct packed_git **ary, *p;
944 for (n = 0, p = packed_git; p; p = p->next)
949 /* prepare an array of packed_git for easier sorting */
950 ary = xcalloc(n, sizeof(struct packed_git *));
951 for (n = 0, p = packed_git; p; p = p->next)
954 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
956 /* link them back again */
957 for (i = 0; i < n - 1; i++)
958 ary[i]->next = ary[i + 1];
959 ary[n - 1]->next = NULL;
965 static int prepare_packed_git_run_once = 0;
966 void prepare_packed_git(void)
968 struct alternate_object_database *alt;
970 if (prepare_packed_git_run_once)
972 prepare_packed_git_one(get_object_directory(), 1);
974 for (alt = alt_odb_list; alt; alt = alt->next) {
976 prepare_packed_git_one(alt->base, 0);
979 rearrange_packed_git();
980 prepare_packed_git_run_once = 1;
983 void reprepare_packed_git(void)
985 prepare_packed_git_run_once = 0;
986 prepare_packed_git();
989 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
991 unsigned char real_sha1[20];
992 hash_sha1_file(map, size, type, real_sha1);
993 return hashcmp(sha1, real_sha1) ? -1 : 0;
996 static int git_open_noatime(const char *name)
998 static int sha1_file_open_flag = O_NOATIME;
999 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1001 /* Might the failure be due to O_NOATIME? */
1002 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1003 fd = open(name, O_RDONLY);
1005 sha1_file_open_flag = 0;
1010 static int open_sha1_file(const unsigned char *sha1)
1013 char *name = sha1_file_name(sha1);
1014 struct alternate_object_database *alt;
1016 fd = git_open_noatime(name);
1022 for (alt = alt_odb_list; alt; alt = alt->next) {
1024 fill_sha1_path(name, sha1);
1025 fd = git_open_noatime(alt->base);
1032 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1037 fd = open_sha1_file(sha1);
1042 if (!fstat(fd, &st)) {
1043 *size = xsize_t(st.st_size);
1044 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1051 static int legacy_loose_object(unsigned char *map)
1056 * Is it a zlib-compressed buffer? If so, the first byte
1057 * must be 0x78 (15-bit window size, deflated), and the
1058 * first 16-bit word is evenly divisible by 31
1060 word = (map[0] << 8) + map[1];
1061 if (map[0] == 0x78 && !(word % 31))
1067 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1072 unsigned long used = 0;
1075 *type = (c >> 4) & 7;
1081 if (sizeof(long) * 8 <= shift)
1084 size += (c & 0x7f) << shift;
1091 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1093 unsigned long size, used;
1094 static const char valid_loose_object_type[8] = {
1096 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1097 0, /* "delta" and others are invalid in a loose object */
1099 enum object_type type;
1101 /* Get the data stream */
1102 memset(stream, 0, sizeof(*stream));
1103 stream->next_in = map;
1104 stream->avail_in = mapsize;
1105 stream->next_out = buffer;
1106 stream->avail_out = bufsiz;
1108 if (legacy_loose_object(map)) {
1109 inflateInit(stream);
1110 return inflate(stream, 0);
1115 * There used to be a second loose object header format which
1116 * was meant to mimic the in-pack format, allowing for direct
1117 * copy of the object data. This format turned up not to be
1118 * really worth it and we don't write it any longer. But we
1119 * can still read it.
1121 used = unpack_object_header_gently(map, mapsize, &type, &size);
1122 if (!used || !valid_loose_object_type[type])
1127 /* Set up the stream for the rest.. */
1128 stream->next_in = map;
1129 stream->avail_in = mapsize;
1130 inflateInit(stream);
1132 /* And generate the fake traditional header */
1133 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1134 typename(type), size);
1138 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1140 int bytes = strlen(buffer) + 1;
1141 unsigned char *buf = xmalloc(1+size);
1145 n = stream->total_out - bytes;
1148 memcpy(buf, (char *) buffer + bytes, n);
1150 if (bytes <= size) {
1152 * The above condition must be (bytes <= size), not
1153 * (bytes < size). In other words, even though we
1154 * expect no more output and set avail_out to zer0,
1155 * the input zlib stream may have bytes that express
1156 * "this concludes the stream", and we *do* want to
1159 * Otherwise we would not be able to test that we
1160 * consumed all the input to reach the expected size;
1161 * we also want to check that zlib tells us that all
1162 * went well with status == Z_STREAM_END at the end.
1164 stream->next_out = buf + bytes;
1165 stream->avail_out = size - bytes;
1166 while (status == Z_OK)
1167 status = inflate(stream, Z_FINISH);
1170 if (status == Z_STREAM_END && !stream->avail_in) {
1176 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1177 else if (stream->avail_in)
1178 error("garbage at end of loose object '%s'",
1185 * We used to just use "sscanf()", but that's actually way
1186 * too permissive for what we want to check. So do an anal
1187 * object header parse by hand.
1189 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1196 * The type can be at most ten bytes (including the
1197 * terminating '\0' that we add), and is followed by
1206 if (i >= sizeof(type))
1212 * The length must follow immediately, and be in canonical
1213 * decimal format (ie "010" is not valid).
1215 size = *hdr++ - '0';
1220 unsigned long c = *hdr - '0';
1224 size = size * 10 + c;
1230 * The length must be followed by a zero byte
1232 return *hdr ? -1 : type_from_string(type);
1235 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1241 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1242 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1245 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1248 unsigned long get_size_from_delta(struct packed_git *p,
1249 struct pack_window **w_curs,
1252 const unsigned char *data;
1253 unsigned char delta_head[20], *in;
1257 memset(&stream, 0, sizeof(stream));
1258 stream.next_out = delta_head;
1259 stream.avail_out = sizeof(delta_head);
1261 inflateInit(&stream);
1263 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1264 stream.next_in = in;
1265 st = inflate(&stream, Z_FINISH);
1266 curpos += stream.next_in - in;
1267 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1268 stream.total_out < sizeof(delta_head));
1269 inflateEnd(&stream);
1270 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1271 die("delta data unpack-initial failed");
1273 /* Examine the initial part of the delta to figure out
1278 /* ignore base size */
1279 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1281 /* Read the result size */
1282 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1285 static off_t get_delta_base(struct packed_git *p,
1286 struct pack_window **w_curs,
1288 enum object_type type,
1289 off_t delta_obj_offset)
1291 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1294 /* use_pack() assured us we have [base_info, base_info + 20)
1295 * as a range that we can look at without walking off the
1296 * end of the mapped window. Its actually the hash size
1297 * that is assured. An OFS_DELTA longer than the hash size
1298 * is stupid, as then a REF_DELTA would be smaller to store.
1300 if (type == OBJ_OFS_DELTA) {
1302 unsigned char c = base_info[used++];
1303 base_offset = c & 127;
1306 if (!base_offset || MSB(base_offset, 7))
1307 die("offset value overflow for delta base object");
1308 c = base_info[used++];
1309 base_offset = (base_offset << 7) + (c & 127);
1311 base_offset = delta_obj_offset - base_offset;
1312 if (base_offset >= delta_obj_offset)
1313 die("delta base offset out of bound");
1315 } else if (type == OBJ_REF_DELTA) {
1316 /* The base entry _must_ be in the same pack */
1317 base_offset = find_pack_entry_one(base_info, p);
1319 die("failed to find delta-pack base object %s",
1320 sha1_to_hex(base_info));
1323 die("I am totally screwed");
1327 /* forward declaration for a mutually recursive function */
1328 static int packed_object_info(struct packed_git *p, off_t offset,
1329 unsigned long *sizep);
1331 static int packed_delta_info(struct packed_git *p,
1332 struct pack_window **w_curs,
1334 enum object_type type,
1336 unsigned long *sizep)
1340 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1341 type = packed_object_info(p, base_offset, NULL);
1343 /* We choose to only get the type of the base object and
1344 * ignore potentially corrupt pack file that expects the delta
1345 * based on a base with a wrong size. This saves tons of
1349 *sizep = get_size_from_delta(p, w_curs, curpos);
1354 static int unpack_object_header(struct packed_git *p,
1355 struct pack_window **w_curs,
1357 unsigned long *sizep)
1359 unsigned char *base;
1362 enum object_type type;
1364 /* use_pack() assures us we have [base, base + 20) available
1365 * as a range that we can look at at. (Its actually the hash
1366 * size that is assured.) With our object header encoding
1367 * the maximum deflated object size is 2^137, which is just
1368 * insane, so we know won't exceed what we have been given.
1370 base = use_pack(p, w_curs, *curpos, &left);
1371 used = unpack_object_header_gently(base, left, &type, sizep);
1373 die("object offset outside of pack file");
1379 const char *packed_object_info_detail(struct packed_git *p,
1381 unsigned long *size,
1382 unsigned long *store_size,
1383 unsigned int *delta_chain_length,
1384 unsigned char *base_sha1)
1386 struct pack_window *w_curs = NULL;
1388 unsigned long dummy;
1389 unsigned char *next_sha1;
1390 enum object_type type;
1391 struct revindex_entry *revidx;
1393 *delta_chain_length = 0;
1394 curpos = obj_offset;
1395 type = unpack_object_header(p, &w_curs, &curpos, size);
1397 revidx = find_pack_revindex(p, obj_offset);
1398 *store_size = revidx[1].offset - obj_offset;
1403 die("pack %s contains unknown object type %d",
1404 p->pack_name, type);
1409 unuse_pack(&w_curs);
1410 return typename(type);
1412 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1413 if (*delta_chain_length == 0) {
1414 revidx = find_pack_revindex(p, obj_offset);
1415 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1419 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1420 if (*delta_chain_length == 0)
1421 hashcpy(base_sha1, next_sha1);
1422 obj_offset = find_pack_entry_one(next_sha1, p);
1425 (*delta_chain_length)++;
1426 curpos = obj_offset;
1427 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1431 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1432 unsigned long *sizep)
1434 struct pack_window *w_curs = NULL;
1436 off_t curpos = obj_offset;
1437 enum object_type type;
1439 type = unpack_object_header(p, &w_curs, &curpos, &size);
1444 type = packed_delta_info(p, &w_curs, curpos,
1445 type, obj_offset, sizep);
1455 die("pack %s contains unknown object type %d",
1456 p->pack_name, type);
1458 unuse_pack(&w_curs);
1462 static void *unpack_compressed_entry(struct packed_git *p,
1463 struct pack_window **w_curs,
1469 unsigned char *buffer, *in;
1471 buffer = xmalloc(size + 1);
1473 memset(&stream, 0, sizeof(stream));
1474 stream.next_out = buffer;
1475 stream.avail_out = size;
1477 inflateInit(&stream);
1479 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1480 stream.next_in = in;
1481 st = inflate(&stream, Z_FINISH);
1482 curpos += stream.next_in - in;
1483 } while (st == Z_OK || st == Z_BUF_ERROR);
1484 inflateEnd(&stream);
1485 if ((st != Z_STREAM_END) || stream.total_out != size) {
1493 #define MAX_DELTA_CACHE (256)
1495 static size_t delta_base_cached;
1497 static struct delta_base_cache_lru_list {
1498 struct delta_base_cache_lru_list *prev;
1499 struct delta_base_cache_lru_list *next;
1500 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1502 static struct delta_base_cache_entry {
1503 struct delta_base_cache_lru_list lru;
1505 struct packed_git *p;
1508 enum object_type type;
1509 } delta_base_cache[MAX_DELTA_CACHE];
1511 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1515 hash = (unsigned long)p + (unsigned long)base_offset;
1516 hash += (hash >> 8) + (hash >> 16);
1517 return hash % MAX_DELTA_CACHE;
1520 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1521 unsigned long *base_size, enum object_type *type, int keep_cache)
1524 unsigned long hash = pack_entry_hash(p, base_offset);
1525 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1528 if (ret && ent->p == p && ent->base_offset == base_offset)
1529 goto found_cache_entry;
1530 return unpack_entry(p, base_offset, type, base_size);
1535 ent->lru.next->prev = ent->lru.prev;
1536 ent->lru.prev->next = ent->lru.next;
1537 delta_base_cached -= ent->size;
1539 ret = xmemdupz(ent->data, ent->size);
1542 *base_size = ent->size;
1546 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1551 ent->lru.next->prev = ent->lru.prev;
1552 ent->lru.prev->next = ent->lru.next;
1553 delta_base_cached -= ent->size;
1557 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1558 void *base, unsigned long base_size, enum object_type type)
1560 unsigned long hash = pack_entry_hash(p, base_offset);
1561 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1562 struct delta_base_cache_lru_list *lru;
1564 release_delta_base_cache(ent);
1565 delta_base_cached += base_size;
1567 for (lru = delta_base_cache_lru.next;
1568 delta_base_cached > delta_base_cache_limit
1569 && lru != &delta_base_cache_lru;
1571 struct delta_base_cache_entry *f = (void *)lru;
1572 if (f->type == OBJ_BLOB)
1573 release_delta_base_cache(f);
1575 for (lru = delta_base_cache_lru.next;
1576 delta_base_cached > delta_base_cache_limit
1577 && lru != &delta_base_cache_lru;
1579 struct delta_base_cache_entry *f = (void *)lru;
1580 release_delta_base_cache(f);
1584 ent->base_offset = base_offset;
1587 ent->size = base_size;
1588 ent->lru.next = &delta_base_cache_lru;
1589 ent->lru.prev = delta_base_cache_lru.prev;
1590 delta_base_cache_lru.prev->next = &ent->lru;
1591 delta_base_cache_lru.prev = &ent->lru;
1594 static void *unpack_delta_entry(struct packed_git *p,
1595 struct pack_window **w_curs,
1597 unsigned long delta_size,
1599 enum object_type *type,
1600 unsigned long *sizep)
1602 void *delta_data, *result, *base;
1603 unsigned long base_size;
1606 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1607 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1609 die("failed to read delta base object"
1610 " at %"PRIuMAX" from %s",
1611 (uintmax_t)base_offset, p->pack_name);
1613 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1615 die("failed to unpack compressed delta"
1616 " at %"PRIuMAX" from %s",
1617 (uintmax_t)curpos, p->pack_name);
1618 result = patch_delta(base, base_size,
1619 delta_data, delta_size,
1622 die("failed to apply delta");
1624 add_delta_base_cache(p, base_offset, base, base_size, *type);
1628 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1629 enum object_type *type, unsigned long *sizep)
1631 struct pack_window *w_curs = NULL;
1632 off_t curpos = obj_offset;
1635 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1639 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1640 obj_offset, type, sizep);
1646 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1649 die("unknown object type %i in %s", *type, p->pack_name);
1651 unuse_pack(&w_curs);
1655 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1658 const unsigned char *index = p->index_data;
1660 if (open_pack_index(p))
1662 index = p->index_data;
1664 if (n >= p->num_objects)
1667 if (p->index_version == 1) {
1668 return index + 24 * n + 4;
1671 return index + 20 * n;
1675 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1677 const unsigned char *index = p->index_data;
1679 if (p->index_version == 1) {
1680 return ntohl(*((uint32_t *)(index + 24 * n)));
1683 index += 8 + p->num_objects * (20 + 4);
1684 off = ntohl(*((uint32_t *)(index + 4 * n)));
1685 if (!(off & 0x80000000))
1687 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1688 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1689 ntohl(*((uint32_t *)(index + 4)));
1693 off_t find_pack_entry_one(const unsigned char *sha1,
1694 struct packed_git *p)
1696 const uint32_t *level1_ofs = p->index_data;
1697 const unsigned char *index = p->index_data;
1698 unsigned hi, lo, stride;
1699 static int use_lookup = -1;
1700 static int debug_lookup = -1;
1702 if (debug_lookup < 0)
1703 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1706 if (open_pack_index(p))
1708 level1_ofs = p->index_data;
1709 index = p->index_data;
1711 if (p->index_version > 1) {
1716 hi = ntohl(level1_ofs[*sha1]);
1717 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1718 if (p->index_version > 1) {
1726 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1727 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1730 use_lookup = !!getenv("GIT_USE_LOOKUP");
1732 int pos = sha1_entry_pos(index, stride, 0,
1733 lo, hi, p->num_objects, sha1);
1736 return nth_packed_object_offset(p, pos);
1740 unsigned mi = (lo + hi) / 2;
1741 int cmp = hashcmp(index + mi * stride, sha1);
1744 printf("lo %u hi %u rg %u mi %u\n",
1745 lo, hi, hi - lo, mi);
1747 return nth_packed_object_offset(p, mi);
1756 int matches_pack_name(struct packed_git *p, const char *name)
1758 const char *last_c, *c;
1760 if (!strcmp(p->pack_name, name))
1763 for (c = p->pack_name, last_c = c; *c;)
1768 if (!strcmp(last_c, name))
1774 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1776 static struct packed_git *last_found = (void *)1;
1777 struct packed_git *p;
1780 prepare_packed_git();
1783 p = (last_found == (void *)1) ? packed_git : last_found;
1786 if (ignore_packed) {
1788 for (ig = ignore_packed; *ig; ig++)
1789 if (matches_pack_name(p, *ig))
1795 offset = find_pack_entry_one(sha1, p);
1798 * We are about to tell the caller where they can
1799 * locate the requested object. We better make
1800 * sure the packfile is still here and can be
1801 * accessed before supplying that answer, as
1802 * it may have been deleted since the index
1805 if (p->pack_fd == -1 && open_packed_git(p)) {
1806 error("packfile %s cannot be accessed", p->pack_name);
1811 hashcpy(e->sha1, sha1);
1817 if (p == last_found)
1821 if (p == last_found)
1827 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1828 struct packed_git *packs)
1830 struct packed_git *p;
1832 for (p = packs; p; p = p->next) {
1833 if (find_pack_entry_one(sha1, p))
1840 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1843 unsigned long mapsize, size;
1848 map = map_sha1_file(sha1, &mapsize);
1850 return error("unable to find %s", sha1_to_hex(sha1));
1851 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1852 status = error("unable to unpack %s header",
1854 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1855 status = error("unable to parse %s header", sha1_to_hex(sha1));
1858 inflateEnd(&stream);
1859 munmap(map, mapsize);
1863 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1865 struct pack_entry e;
1867 if (!find_pack_entry(sha1, &e, NULL)) {
1868 reprepare_packed_git();
1869 if (!find_pack_entry(sha1, &e, NULL))
1870 return sha1_loose_object_info(sha1, sizep);
1872 return packed_object_info(e.p, e.offset, sizep);
1875 static void *read_packed_sha1(const unsigned char *sha1,
1876 enum object_type *type, unsigned long *size)
1878 struct pack_entry e;
1880 if (!find_pack_entry(sha1, &e, NULL))
1883 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1887 * This is meant to hold a *small* number of objects that you would
1888 * want read_sha1_file() to be able to return, but yet you do not want
1889 * to write them into the object store (e.g. a browse-only
1892 static struct cached_object {
1893 unsigned char sha1[20];
1894 enum object_type type;
1898 static int cached_object_nr, cached_object_alloc;
1900 static struct cached_object empty_tree = {
1901 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1902 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1903 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1909 static struct cached_object *find_cached_object(const unsigned char *sha1)
1912 struct cached_object *co = cached_objects;
1914 for (i = 0; i < cached_object_nr; i++, co++) {
1915 if (!hashcmp(co->sha1, sha1))
1918 if (!hashcmp(sha1, empty_tree.sha1))
1923 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1924 unsigned char *sha1)
1926 struct cached_object *co;
1928 hash_sha1_file(buf, len, typename(type), sha1);
1929 if (has_sha1_file(sha1) || find_cached_object(sha1))
1931 if (cached_object_alloc <= cached_object_nr) {
1932 cached_object_alloc = alloc_nr(cached_object_alloc);
1933 cached_objects = xrealloc(cached_objects,
1934 sizeof(*cached_objects) *
1935 cached_object_alloc);
1937 co = &cached_objects[cached_object_nr++];
1940 co->buf = xmalloc(len);
1941 memcpy(co->buf, buf, len);
1942 hashcpy(co->sha1, sha1);
1946 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1947 unsigned long *size)
1949 unsigned long mapsize;
1951 struct cached_object *co;
1953 co = find_cached_object(sha1);
1957 return xmemdupz(co->buf, co->size);
1960 buf = read_packed_sha1(sha1, type, size);
1963 map = map_sha1_file(sha1, &mapsize);
1965 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1966 munmap(map, mapsize);
1969 reprepare_packed_git();
1970 return read_packed_sha1(sha1, type, size);
1973 void *read_object_with_reference(const unsigned char *sha1,
1974 const char *required_type_name,
1975 unsigned long *size,
1976 unsigned char *actual_sha1_return)
1978 enum object_type type, required_type;
1980 unsigned long isize;
1981 unsigned char actual_sha1[20];
1983 required_type = type_from_string(required_type_name);
1984 hashcpy(actual_sha1, sha1);
1986 int ref_length = -1;
1987 const char *ref_type = NULL;
1989 buffer = read_sha1_file(actual_sha1, &type, &isize);
1992 if (type == required_type) {
1994 if (actual_sha1_return)
1995 hashcpy(actual_sha1_return, actual_sha1);
1998 /* Handle references */
1999 else if (type == OBJ_COMMIT)
2001 else if (type == OBJ_TAG)
2002 ref_type = "object ";
2007 ref_length = strlen(ref_type);
2009 if (ref_length + 40 > isize ||
2010 memcmp(buffer, ref_type, ref_length) ||
2011 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2016 /* Now we have the ID of the referred-to object in
2017 * actual_sha1. Check again. */
2021 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2022 const char *type, unsigned char *sha1,
2023 char *hdr, int *hdrlen)
2027 /* Generate the header */
2028 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2032 SHA1_Update(&c, hdr, *hdrlen);
2033 SHA1_Update(&c, buf, len);
2034 SHA1_Final(sha1, &c);
2038 * Move the just written object into its final resting place
2040 int move_temp_to_file(const char *tmpfile, const char *filename)
2042 int ret = link(tmpfile, filename);
2045 * Coda hack - coda doesn't like cross-directory links,
2046 * so we fall back to a rename, which will mean that it
2047 * won't be able to check collisions, but that's not a
2050 * The same holds for FAT formatted media.
2052 * When this succeeds, we just return 0. We have nothing
2055 if (ret && ret != EEXIST) {
2056 if (!rename(tmpfile, filename))
2062 if (ret != EEXIST) {
2063 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2065 /* FIXME!!! Collision check here ? */
2071 static int write_buffer(int fd, const void *buf, size_t len)
2073 if (write_in_full(fd, buf, len) < 0)
2074 return error("file write error (%s)", strerror(errno));
2078 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2079 unsigned char *sha1)
2083 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2087 /* Finalize a file on disk, and close it. */
2088 static void close_sha1_file(int fd)
2090 /* For safe-mode, we could fsync_or_die(fd, "sha1 file") here */
2093 die("unable to write sha1 file");
2096 /* Size of directory component, including the ending '/' */
2097 static inline int directory_size(const char *filename)
2099 const char *s = strrchr(filename, '/');
2102 return s - filename + 1;
2106 * This creates a temporary file in the same directory as the final
2109 * We want to avoid cross-directory filename renames, because those
2110 * can have problems on various filesystems (FAT, NFS, Coda).
2112 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2114 int fd, dirlen = directory_size(filename);
2116 if (dirlen + 20 > bufsiz) {
2117 errno = ENAMETOOLONG;
2120 memcpy(buffer, filename, dirlen);
2121 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2122 fd = mkstemp(buffer);
2123 if (fd < 0 && dirlen) {
2124 /* Make sure the directory exists */
2125 buffer[dirlen-1] = 0;
2126 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2130 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2131 fd = mkstemp(buffer);
2136 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2137 void *buf, unsigned long len, time_t mtime)
2140 unsigned char *compressed;
2143 static char tmpfile[PATH_MAX];
2145 filename = sha1_file_name(sha1);
2146 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2149 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2151 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2155 memset(&stream, 0, sizeof(stream));
2156 deflateInit(&stream, zlib_compression_level);
2157 size = 8 + deflateBound(&stream, len+hdrlen);
2158 compressed = xmalloc(size);
2161 stream.next_out = compressed;
2162 stream.avail_out = size;
2164 /* First header.. */
2165 stream.next_in = (unsigned char *)hdr;
2166 stream.avail_in = hdrlen;
2167 while (deflate(&stream, 0) == Z_OK)
2170 /* Then the data itself.. */
2171 stream.next_in = buf;
2172 stream.avail_in = len;
2173 ret = deflate(&stream, Z_FINISH);
2174 if (ret != Z_STREAM_END)
2175 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2177 ret = deflateEnd(&stream);
2179 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2181 size = stream.total_out;
2183 if (write_buffer(fd, compressed, size) < 0)
2184 die("unable to write sha1 file");
2185 close_sha1_file(fd);
2191 utb.modtime = mtime;
2192 if (utime(tmpfile, &utb) < 0)
2193 warning("failed utime() on %s: %s",
2194 tmpfile, strerror(errno));
2197 return move_temp_to_file(tmpfile, filename);
2200 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2202 unsigned char sha1[20];
2206 /* Normally if we have it in the pack then we do not bother writing
2207 * it out into .git/objects/??/?{38} file.
2209 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2211 hashcpy(returnsha1, sha1);
2212 if (has_sha1_file(sha1))
2214 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2217 int force_object_loose(const unsigned char *sha1, time_t mtime)
2221 enum object_type type;
2225 if (has_loose_object(sha1))
2227 buf = read_packed_sha1(sha1, &type, &len);
2229 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2230 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2231 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2234 int has_pack_index(const unsigned char *sha1)
2237 if (stat(sha1_pack_index_name(sha1), &st))
2242 int has_pack_file(const unsigned char *sha1)
2245 if (stat(sha1_pack_name(sha1), &st))
2250 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2252 struct pack_entry e;
2253 return find_pack_entry(sha1, &e, ignore_packed);
2256 int has_sha1_file(const unsigned char *sha1)
2258 struct pack_entry e;
2260 if (find_pack_entry(sha1, &e, NULL))
2262 return has_loose_object(sha1);
2265 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2270 strbuf_init(&buf, 0);
2271 if (strbuf_read(&buf, fd, 4096) < 0) {
2272 strbuf_release(&buf);
2279 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2281 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2282 strbuf_release(&buf);
2287 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2288 enum object_type type, const char *path)
2290 size_t size = xsize_t(st->st_size);
2292 int ret, re_allocated = 0;
2295 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2302 * Convert blobs to git internal format
2304 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2306 strbuf_init(&nbuf, 0);
2307 if (convert_to_git(path, buf, size, &nbuf,
2308 write_object ? safe_crlf : 0)) {
2310 buf = strbuf_detach(&nbuf, &size);
2316 ret = write_sha1_file(buf, size, typename(type), sha1);
2318 ret = hash_sha1_file(buf, size, typename(type), sha1);
2328 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2334 switch (st->st_mode & S_IFMT) {
2336 fd = open(path, O_RDONLY);
2338 return error("open(\"%s\"): %s", path,
2340 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2341 return error("%s: failed to insert into database",
2345 len = xsize_t(st->st_size);
2346 target = xmalloc(len + 1);
2347 if (readlink(path, target, len + 1) != st->st_size) {
2348 char *errstr = strerror(errno);
2350 return error("readlink(\"%s\"): %s", path,
2354 hash_sha1_file(target, len, blob_type, sha1);
2355 else if (write_sha1_file(target, len, blob_type, sha1))
2356 return error("%s: failed to insert into database",
2361 return resolve_gitlink_ref(path, "HEAD", sha1);
2363 return error("%s: unsupported file type", path);
2368 int read_pack_header(int fd, struct pack_header *header)
2370 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2371 /* "eof before pack header was fully read" */
2372 return PH_ERROR_EOF;
2374 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2375 /* "protocol error (pack signature mismatch detected)" */
2376 return PH_ERROR_PACK_SIGNATURE;
2377 if (!pack_version_ok(header->hdr_version))
2378 /* "protocol error (pack version unsupported)" */
2379 return PH_ERROR_PROTOCOL;