2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
33 static size_t sz_fmt(size_t s) { return s; }
36 const unsigned char null_sha1[20];
38 static unsigned int sha1_file_open_flag = O_NOATIME;
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
88 int safe_create_leading_directories(char *path)
93 if (is_absolute_path(path))
97 pos = strchr(pos, '/');
101 if (!stat(path, &st)) {
103 if (!S_ISDIR(st.st_mode)) {
108 else if (mkdir(path, 0777)) {
112 else if (adjust_shared_perm(path)) {
121 char * sha1_to_hex(const unsigned char *sha1)
124 static char hexbuffer[4][50];
125 static const char hex[] = "0123456789abcdef";
126 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
129 for (i = 0; i < 20; i++) {
130 unsigned int val = *sha1++;
131 *buf++ = hex[val >> 4];
132 *buf++ = hex[val & 0xf];
139 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
142 for (i = 0; i < 20; i++) {
143 static char hex[] = "0123456789abcdef";
144 unsigned int val = sha1[i];
145 char *pos = pathbuf + i*2 + (i > 0);
146 *pos++ = hex[val >> 4];
147 *pos = hex[val & 0xf];
152 * NOTE! This returns a statically allocated buffer, so you have to be
153 * careful about using it. Do an "xstrdup()" if you need to save the
156 * Also note that this returns the location for creating. Reading
157 * SHA1 file can happen from any alternate directory listed in the
158 * DB_ENVIRONMENT environment variable if it is not found in
159 * the primary object database.
161 char *sha1_file_name(const unsigned char *sha1)
163 static char *name, *base;
166 const char *sha1_file_directory = get_object_directory();
167 int len = strlen(sha1_file_directory);
168 base = xmalloc(len + 60);
169 memcpy(base, sha1_file_directory, len);
170 memset(base+len, 0, 60);
173 name = base + len + 1;
175 fill_sha1_path(name, sha1);
179 char *sha1_pack_name(const unsigned char *sha1)
181 static const char hex[] = "0123456789abcdef";
182 static char *name, *base, *buf;
186 const char *sha1_file_directory = get_object_directory();
187 int len = strlen(sha1_file_directory);
188 base = xmalloc(len + 60);
189 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
190 name = base + len + 11;
195 for (i = 0; i < 20; i++) {
196 unsigned int val = *sha1++;
197 *buf++ = hex[val >> 4];
198 *buf++ = hex[val & 0xf];
204 char *sha1_pack_index_name(const unsigned char *sha1)
206 static const char hex[] = "0123456789abcdef";
207 static char *name, *base, *buf;
211 const char *sha1_file_directory = get_object_directory();
212 int len = strlen(sha1_file_directory);
213 base = xmalloc(len + 60);
214 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
215 name = base + len + 11;
220 for (i = 0; i < 20; i++) {
221 unsigned int val = *sha1++;
222 *buf++ = hex[val >> 4];
223 *buf++ = hex[val & 0xf];
229 struct alternate_object_database *alt_odb_list;
230 static struct alternate_object_database **alt_odb_tail;
232 static void read_info_alternates(const char * alternates, int depth);
235 * Prepare alternate object database registry.
237 * The variable alt_odb_list points at the list of struct
238 * alternate_object_database. The elements on this list come from
239 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
240 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
241 * whose contents is similar to that environment variable but can be
242 * LF separated. Its base points at a statically allocated buffer that
243 * contains "/the/directory/corresponding/to/.git/objects/...", while
244 * its name points just after the slash at the end of ".git/objects/"
245 * in the example above, and has enough space to hold 40-byte hex
246 * SHA1, an extra slash for the first level indirection, and the
249 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
252 const char *objdir = get_object_directory();
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 /* 43 = 40-byte + 2 '/' + terminating NUL */
257 int entlen = pfxlen + 43;
260 if (!is_absolute_path(entry) && relative_base) {
261 /* Relative alt-odb */
263 base_len = strlen(relative_base) + 1;
267 ent = xmalloc(sizeof(*ent) + entlen);
269 if (!is_absolute_path(entry) && relative_base) {
270 memcpy(ent->base, relative_base, base_len - 1);
271 ent->base[base_len - 1] = '/';
272 memcpy(ent->base + base_len, entry, len);
275 memcpy(ent->base, entry, pfxlen);
277 ent->name = ent->base + pfxlen + 1;
278 ent->base[pfxlen + 3] = '/';
279 ent->base[pfxlen] = ent->base[entlen-1] = 0;
281 /* Detect cases where alternate disappeared */
282 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
283 error("object directory %s does not exist; "
284 "check .git/objects/info/alternates.",
290 /* Prevent the common mistake of listing the same
291 * thing twice, or object directory itself.
293 for (alt = alt_odb_list; alt; alt = alt->next) {
294 if (!memcmp(ent->base, alt->base, pfxlen)) {
299 if (!memcmp(ent->base, objdir, pfxlen)) {
304 /* add the alternate entry */
306 alt_odb_tail = &(ent->next);
309 /* recursively add alternates */
310 read_info_alternates(ent->base, depth + 1);
312 ent->base[pfxlen] = '/';
317 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
318 const char *relative_base, int depth)
320 const char *cp, *last;
323 error("%s: ignoring alternate object stores, nesting too deep.",
331 if (cp < ep && *cp == '#') {
332 while (cp < ep && *cp != sep)
337 while (cp < ep && *cp != sep)
340 if (!is_absolute_path(last) && depth) {
341 error("%s: ignoring relative alternate object store %s",
342 relative_base, last);
344 link_alt_odb_entry(last, cp - last,
345 relative_base, depth);
348 while (cp < ep && *cp == sep)
354 static void read_info_alternates(const char * relative_base, int depth)
359 const char alt_file_name[] = "info/alternates";
360 /* Given that relative_base is no longer than PATH_MAX,
361 ensure that "path" has enough space to append "/", the
362 file name, "info/alternates", and a trailing NUL. */
363 char path[PATH_MAX + 1 + sizeof alt_file_name];
366 sprintf(path, "%s/%s", relative_base, alt_file_name);
367 fd = open(path, O_RDONLY);
370 if (fstat(fd, &st) || (st.st_size == 0)) {
374 mapsz = xsize_t(st.st_size);
375 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
378 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
383 void add_to_alternates_file(const char *reference)
385 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
386 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
387 char *alt = mkpath("%s/objects\n", reference);
388 write_or_die(fd, alt, strlen(alt));
389 if (commit_lock_file(lock))
390 die("could not close alternates file");
392 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
395 void prepare_alt_odb(void)
402 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
405 alt_odb_tail = &alt_odb_list;
406 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
408 read_info_alternates(get_object_directory(), 0);
411 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
413 char *name = sha1_file_name(sha1);
414 struct alternate_object_database *alt;
419 for (alt = alt_odb_list; alt; alt = alt->next) {
421 fill_sha1_path(name, sha1);
422 if (!stat(alt->base, st))
428 static unsigned int pack_used_ctr;
429 static unsigned int pack_mmap_calls;
430 static unsigned int peak_pack_open_windows;
431 static unsigned int pack_open_windows;
432 static size_t peak_pack_mapped;
433 static size_t pack_mapped;
434 struct packed_git *packed_git;
436 void pack_report(void)
439 "pack_report: getpagesize() = %10" SZ_FMT "\n"
440 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
441 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
442 sz_fmt(getpagesize()),
443 sz_fmt(packed_git_window_size),
444 sz_fmt(packed_git_limit));
446 "pack_report: pack_used_ctr = %10u\n"
447 "pack_report: pack_mmap_calls = %10u\n"
448 "pack_report: pack_open_windows = %10u / %10u\n"
449 "pack_report: pack_mapped = "
450 "%10" SZ_FMT " / %10" SZ_FMT "\n",
453 pack_open_windows, peak_pack_open_windows,
454 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
457 static int check_packed_git_idx(const char *path, struct packed_git *p)
460 struct pack_idx_header *hdr;
462 uint32_t version, nr, i, *index;
463 int fd = open(path, O_RDONLY);
468 if (fstat(fd, &st)) {
472 idx_size = xsize_t(st.st_size);
473 if (idx_size < 4 * 256 + 20 + 20) {
475 return error("index file %s is too small", path);
477 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
481 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
482 version = ntohl(hdr->idx_version);
483 if (version < 2 || version > 2) {
484 munmap(idx_map, idx_size);
485 return error("index file %s is version %d"
486 " and is not supported by this binary"
487 " (try upgrading GIT to a newer version)",
496 index += 2; /* skip index header */
497 for (i = 0; i < 256; i++) {
498 uint32_t n = ntohl(index[i]);
500 munmap(idx_map, idx_size);
501 return error("non-monotonic index %s", path);
509 * - 256 index entries 4 bytes each
510 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
511 * - 20-byte SHA1 of the packfile
512 * - 20-byte SHA1 file checksum
514 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
515 munmap(idx_map, idx_size);
516 return error("wrong index v1 file size in %s", path);
518 } else if (version == 2) {
521 * - 8 bytes of header
522 * - 256 index entries 4 bytes each
523 * - 20-byte sha1 entry * nr
524 * - 4-byte crc entry * nr
525 * - 4-byte offset entry * nr
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
528 * And after the 4-byte offset table might be a
529 * variable sized table containing 8-byte entries
530 * for offsets larger than 2^31.
532 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
533 unsigned long max_size = min_size;
535 max_size += (nr - 1)*8;
536 if (idx_size < min_size || idx_size > max_size) {
537 munmap(idx_map, idx_size);
538 return error("wrong index v2 file size in %s", path);
540 if (idx_size != min_size &&
542 * make sure we can deal with large pack offsets.
543 * 31-bit signed offset won't be enough, neither
544 * 32-bit unsigned one will be.
546 (sizeof(off_t) <= 4)) {
547 munmap(idx_map, idx_size);
548 return error("pack too large for current definition of off_t in %s", path);
552 p->index_version = version;
553 p->index_data = idx_map;
554 p->index_size = idx_size;
559 int open_pack_index(struct packed_git *p)
567 idx_name = xstrdup(p->pack_name);
568 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
569 ret = check_packed_git_idx(idx_name, p);
574 static void scan_windows(struct packed_git *p,
575 struct packed_git **lru_p,
576 struct pack_window **lru_w,
577 struct pack_window **lru_l)
579 struct pack_window *w, *w_l;
581 for (w_l = NULL, w = p->windows; w; w = w->next) {
583 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
593 static int unuse_one_window(struct packed_git *current, int keep_fd)
595 struct packed_git *p, *lru_p = NULL;
596 struct pack_window *lru_w = NULL, *lru_l = NULL;
599 scan_windows(current, &lru_p, &lru_w, &lru_l);
600 for (p = packed_git; p; p = p->next)
601 scan_windows(p, &lru_p, &lru_w, &lru_l);
603 munmap(lru_w->base, lru_w->len);
604 pack_mapped -= lru_w->len;
606 lru_l->next = lru_w->next;
608 lru_p->windows = lru_w->next;
609 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
610 close(lru_p->pack_fd);
621 void release_pack_memory(size_t need, int fd)
623 size_t cur = pack_mapped;
624 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
628 void close_pack_windows(struct packed_git *p)
631 struct pack_window *w = p->windows;
634 die("pack '%s' still has open windows to it",
636 munmap(w->base, w->len);
637 pack_mapped -= w->len;
639 p->windows = w->next;
644 void unuse_pack(struct pack_window **w_cursor)
646 struct pack_window *w = *w_cursor;
654 * Do not call this directly as this leaks p->pack_fd on error return;
655 * call open_packed_git() instead.
657 static int open_packed_git_1(struct packed_git *p)
660 struct pack_header hdr;
661 unsigned char sha1[20];
662 unsigned char *idx_sha1;
665 if (!p->index_data && open_pack_index(p))
666 return error("packfile %s index unavailable", p->pack_name);
668 p->pack_fd = open(p->pack_name, O_RDONLY);
669 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
672 /* If we created the struct before we had the pack we lack size. */
674 if (!S_ISREG(st.st_mode))
675 return error("packfile %s not a regular file", p->pack_name);
676 p->pack_size = st.st_size;
677 } else if (p->pack_size != st.st_size)
678 return error("packfile %s size changed", p->pack_name);
680 /* We leave these file descriptors open with sliding mmap;
681 * there is no point keeping them open across exec(), though.
683 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
685 return error("cannot determine file descriptor flags");
686 fd_flag |= FD_CLOEXEC;
687 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
688 return error("cannot set FD_CLOEXEC");
690 /* Verify we recognize this pack file format. */
691 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
692 return error("file %s is far too short to be a packfile", p->pack_name);
693 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
694 return error("file %s is not a GIT packfile", p->pack_name);
695 if (!pack_version_ok(hdr.hdr_version))
696 return error("packfile %s is version %u and not supported"
697 " (try upgrading GIT to a newer version)",
698 p->pack_name, ntohl(hdr.hdr_version));
700 /* Verify the pack matches its index. */
701 if (p->num_objects != ntohl(hdr.hdr_entries))
702 return error("packfile %s claims to have %u objects"
703 " while index indicates %u objects",
704 p->pack_name, ntohl(hdr.hdr_entries),
706 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
707 return error("end of packfile %s is unavailable", p->pack_name);
708 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
709 return error("packfile %s signature is unavailable", p->pack_name);
710 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
711 if (hashcmp(sha1, idx_sha1))
712 return error("packfile %s does not match index", p->pack_name);
716 static int open_packed_git(struct packed_git *p)
718 if (!open_packed_git_1(p))
720 if (p->pack_fd != -1) {
727 static int in_window(struct pack_window *win, off_t offset)
729 /* We must promise at least 20 bytes (one hash) after the
730 * offset is available from this window, otherwise the offset
731 * is not actually in this window and a different window (which
732 * has that one hash excess) must be used. This is to support
733 * the object header and delta base parsing routines below.
735 off_t win_off = win->offset;
736 return win_off <= offset
737 && (offset + 20) <= (win_off + win->len);
740 unsigned char* use_pack(struct packed_git *p,
741 struct pack_window **w_cursor,
745 struct pack_window *win = *w_cursor;
747 if (p->pack_fd == -1 && open_packed_git(p))
748 die("packfile %s cannot be accessed", p->pack_name);
750 /* Since packfiles end in a hash of their content and its
751 * pointless to ask for an offset into the middle of that
752 * hash, and the in_window function above wouldn't match
753 * don't allow an offset too close to the end of the file.
755 if (offset > (p->pack_size - 20))
756 die("offset beyond end of packfile (truncated pack?)");
758 if (!win || !in_window(win, offset)) {
761 for (win = p->windows; win; win = win->next) {
762 if (in_window(win, offset))
766 size_t window_align = packed_git_window_size / 2;
768 win = xcalloc(1, sizeof(*win));
769 win->offset = (offset / window_align) * window_align;
770 len = p->pack_size - win->offset;
771 if (len > packed_git_window_size)
772 len = packed_git_window_size;
773 win->len = (size_t)len;
774 pack_mapped += win->len;
775 while (packed_git_limit < pack_mapped
776 && unuse_one_window(p, p->pack_fd))
778 win->base = xmmap(NULL, win->len,
779 PROT_READ, MAP_PRIVATE,
780 p->pack_fd, win->offset);
781 if (win->base == MAP_FAILED)
782 die("packfile %s cannot be mapped: %s",
787 if (pack_mapped > peak_pack_mapped)
788 peak_pack_mapped = pack_mapped;
789 if (pack_open_windows > peak_pack_open_windows)
790 peak_pack_open_windows = pack_open_windows;
791 win->next = p->windows;
795 if (win != *w_cursor) {
796 win->last_used = pack_used_ctr++;
800 offset -= win->offset;
802 *left = win->len - xsize_t(offset);
803 return win->base + offset;
806 struct packed_git *add_packed_git(const char *path, int path_len, int local)
809 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
812 * Make sure a corresponding .pack file exists and that
813 * the index looks sane.
815 path_len -= strlen(".idx");
818 memcpy(p->pack_name, path, path_len);
819 strcpy(p->pack_name + path_len, ".pack");
820 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
825 /* ok, it looks sane as far as we can check without
826 * actually mapping the pack file.
828 p->index_version = 0;
829 p->index_data = NULL;
832 p->pack_size = st.st_size;
836 p->pack_local = local;
837 p->mtime = st.st_mtime;
838 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
843 struct packed_git *parse_pack_index(unsigned char *sha1)
845 char *path = sha1_pack_index_name(sha1);
846 return parse_pack_index_file(sha1, path);
849 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
850 const char *idx_path)
852 const char *path = sha1_pack_name(sha1);
853 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
855 if (check_packed_git_idx(idx_path, p)) {
860 strcpy(p->pack_name, path);
865 hashcpy(p->sha1, sha1);
869 void install_packed_git(struct packed_git *pack)
871 pack->next = packed_git;
875 static void prepare_packed_git_one(char *objdir, int local)
877 /* Ensure that this buffer is large enough so that we can
878 append "/pack/" without clobbering the stack even if
879 strlen(objdir) were PATH_MAX. */
880 char path[PATH_MAX + 1 + 4 + 1 + 1];
885 sprintf(path, "%s/pack", objdir);
890 error("unable to open object pack directory: %s: %s",
891 path, strerror(errno));
895 while ((de = readdir(dir)) != NULL) {
896 int namelen = strlen(de->d_name);
897 struct packed_git *p;
899 if (!has_extension(de->d_name, ".idx"))
902 if (len + namelen + 1 > sizeof(path))
905 /* Don't reopen a pack we already have. */
906 strcpy(path + len, de->d_name);
907 for (p = packed_git; p; p = p->next) {
908 if (!memcmp(path, p->pack_name, len + namelen - 4))
913 /* See if it really is a valid .idx file with corresponding
914 * .pack file that we can map.
916 p = add_packed_git(path, len + namelen, local);
919 install_packed_git(p);
924 static int sort_pack(const void *a_, const void *b_)
926 struct packed_git *a = *((struct packed_git **)a_);
927 struct packed_git *b = *((struct packed_git **)b_);
931 * Local packs tend to contain objects specific to our
932 * variant of the project than remote ones. In addition,
933 * remote ones could be on a network mounted filesystem.
934 * Favor local ones for these reasons.
936 st = a->pack_local - b->pack_local;
941 * Younger packs tend to contain more recent objects,
942 * and more recent objects tend to get accessed more
945 if (a->mtime < b->mtime)
947 else if (a->mtime == b->mtime)
952 static void rearrange_packed_git(void)
954 struct packed_git **ary, *p;
957 for (n = 0, p = packed_git; p; p = p->next)
962 /* prepare an array of packed_git for easier sorting */
963 ary = xcalloc(n, sizeof(struct packed_git *));
964 for (n = 0, p = packed_git; p; p = p->next)
967 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
969 /* link them back again */
970 for (i = 0; i < n - 1; i++)
971 ary[i]->next = ary[i + 1];
972 ary[n - 1]->next = NULL;
978 static int prepare_packed_git_run_once = 0;
979 void prepare_packed_git(void)
981 struct alternate_object_database *alt;
983 if (prepare_packed_git_run_once)
985 prepare_packed_git_one(get_object_directory(), 1);
987 for (alt = alt_odb_list; alt; alt = alt->next) {
989 prepare_packed_git_one(alt->base, 0);
992 rearrange_packed_git();
993 prepare_packed_git_run_once = 1;
996 void reprepare_packed_git(void)
998 prepare_packed_git_run_once = 0;
999 prepare_packed_git();
1002 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1004 unsigned char real_sha1[20];
1005 hash_sha1_file(map, size, type, real_sha1);
1006 return hashcmp(sha1, real_sha1) ? -1 : 0;
1009 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1014 char *filename = find_sha1_file(sha1, &st);
1020 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1022 /* See if it works without O_NOATIME */
1023 switch (sha1_file_open_flag) {
1025 fd = open(filename, O_RDONLY);
1033 /* If it failed once, it will probably fail again.
1034 * Stop using O_NOATIME
1036 sha1_file_open_flag = 0;
1038 *size = xsize_t(st.st_size);
1039 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1044 static int legacy_loose_object(unsigned char *map)
1049 * Is it a zlib-compressed buffer? If so, the first byte
1050 * must be 0x78 (15-bit window size, deflated), and the
1051 * first 16-bit word is evenly divisible by 31
1053 word = (map[0] << 8) + map[1];
1054 if (map[0] == 0x78 && !(word % 31))
1060 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1065 unsigned long used = 0;
1068 *type = (c >> 4) & 7;
1074 if (sizeof(long) * 8 <= shift)
1077 size += (c & 0x7f) << shift;
1084 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1086 unsigned long size, used;
1087 static const char valid_loose_object_type[8] = {
1089 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1090 0, /* "delta" and others are invalid in a loose object */
1092 enum object_type type;
1094 /* Get the data stream */
1095 memset(stream, 0, sizeof(*stream));
1096 stream->next_in = map;
1097 stream->avail_in = mapsize;
1098 stream->next_out = buffer;
1099 stream->avail_out = bufsiz;
1101 if (legacy_loose_object(map)) {
1102 inflateInit(stream);
1103 return inflate(stream, 0);
1108 * There used to be a second loose object header format which
1109 * was meant to mimic the in-pack format, allowing for direct
1110 * copy of the object data. This format turned up not to be
1111 * really worth it and we don't write it any longer. But we
1112 * can still read it.
1114 used = unpack_object_header_gently(map, mapsize, &type, &size);
1115 if (!used || !valid_loose_object_type[type])
1120 /* Set up the stream for the rest.. */
1121 stream->next_in = map;
1122 stream->avail_in = mapsize;
1123 inflateInit(stream);
1125 /* And generate the fake traditional header */
1126 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1127 typename(type), size);
1131 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1133 int bytes = strlen(buffer) + 1;
1134 unsigned char *buf = xmalloc(1+size);
1138 n = stream->total_out - bytes;
1141 memcpy(buf, (char *) buffer + bytes, n);
1143 if (bytes <= size) {
1145 * The above condition must be (bytes <= size), not
1146 * (bytes < size). In other words, even though we
1147 * expect no more output and set avail_out to zer0,
1148 * the input zlib stream may have bytes that express
1149 * "this concludes the stream", and we *do* want to
1152 * Otherwise we would not be able to test that we
1153 * consumed all the input to reach the expected size;
1154 * we also want to check that zlib tells us that all
1155 * went well with status == Z_STREAM_END at the end.
1157 stream->next_out = buf + bytes;
1158 stream->avail_out = size - bytes;
1159 while (status == Z_OK)
1160 status = inflate(stream, Z_FINISH);
1163 if (status == Z_STREAM_END && !stream->avail_in) {
1169 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1170 else if (stream->avail_in)
1171 error("garbage at end of loose object '%s'",
1178 * We used to just use "sscanf()", but that's actually way
1179 * too permissive for what we want to check. So do an anal
1180 * object header parse by hand.
1182 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1189 * The type can be at most ten bytes (including the
1190 * terminating '\0' that we add), and is followed by
1199 if (i >= sizeof(type))
1205 * The length must follow immediately, and be in canonical
1206 * decimal format (ie "010" is not valid).
1208 size = *hdr++ - '0';
1213 unsigned long c = *hdr - '0';
1217 size = size * 10 + c;
1223 * The length must be followed by a zero byte
1225 return *hdr ? -1 : type_from_string(type);
1228 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1234 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1235 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1238 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1241 unsigned long get_size_from_delta(struct packed_git *p,
1242 struct pack_window **w_curs,
1245 const unsigned char *data;
1246 unsigned char delta_head[20], *in;
1250 memset(&stream, 0, sizeof(stream));
1251 stream.next_out = delta_head;
1252 stream.avail_out = sizeof(delta_head);
1254 inflateInit(&stream);
1256 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1257 stream.next_in = in;
1258 st = inflate(&stream, Z_FINISH);
1259 curpos += stream.next_in - in;
1260 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1261 stream.total_out < sizeof(delta_head));
1262 inflateEnd(&stream);
1263 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1264 die("delta data unpack-initial failed");
1266 /* Examine the initial part of the delta to figure out
1271 /* ignore base size */
1272 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1274 /* Read the result size */
1275 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1278 static off_t get_delta_base(struct packed_git *p,
1279 struct pack_window **w_curs,
1281 enum object_type type,
1282 off_t delta_obj_offset)
1284 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1287 /* use_pack() assured us we have [base_info, base_info + 20)
1288 * as a range that we can look at without walking off the
1289 * end of the mapped window. Its actually the hash size
1290 * that is assured. An OFS_DELTA longer than the hash size
1291 * is stupid, as then a REF_DELTA would be smaller to store.
1293 if (type == OBJ_OFS_DELTA) {
1295 unsigned char c = base_info[used++];
1296 base_offset = c & 127;
1299 if (!base_offset || MSB(base_offset, 7))
1300 die("offset value overflow for delta base object");
1301 c = base_info[used++];
1302 base_offset = (base_offset << 7) + (c & 127);
1304 base_offset = delta_obj_offset - base_offset;
1305 if (base_offset >= delta_obj_offset)
1306 die("delta base offset out of bound");
1308 } else if (type == OBJ_REF_DELTA) {
1309 /* The base entry _must_ be in the same pack */
1310 base_offset = find_pack_entry_one(base_info, p);
1312 die("failed to find delta-pack base object %s",
1313 sha1_to_hex(base_info));
1316 die("I am totally screwed");
1320 /* forward declaration for a mutually recursive function */
1321 static int packed_object_info(struct packed_git *p, off_t offset,
1322 unsigned long *sizep);
1324 static int packed_delta_info(struct packed_git *p,
1325 struct pack_window **w_curs,
1327 enum object_type type,
1329 unsigned long *sizep)
1333 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1334 type = packed_object_info(p, base_offset, NULL);
1336 /* We choose to only get the type of the base object and
1337 * ignore potentially corrupt pack file that expects the delta
1338 * based on a base with a wrong size. This saves tons of
1342 *sizep = get_size_from_delta(p, w_curs, curpos);
1347 static int unpack_object_header(struct packed_git *p,
1348 struct pack_window **w_curs,
1350 unsigned long *sizep)
1352 unsigned char *base;
1355 enum object_type type;
1357 /* use_pack() assures us we have [base, base + 20) available
1358 * as a range that we can look at at. (Its actually the hash
1359 * size that is assured.) With our object header encoding
1360 * the maximum deflated object size is 2^137, which is just
1361 * insane, so we know won't exceed what we have been given.
1363 base = use_pack(p, w_curs, *curpos, &left);
1364 used = unpack_object_header_gently(base, left, &type, sizep);
1366 die("object offset outside of pack file");
1372 const char *packed_object_info_detail(struct packed_git *p,
1374 unsigned long *size,
1375 unsigned long *store_size,
1376 unsigned int *delta_chain_length,
1377 unsigned char *base_sha1)
1379 struct pack_window *w_curs = NULL;
1381 unsigned long dummy;
1382 unsigned char *next_sha1;
1383 enum object_type type;
1384 struct revindex_entry *revidx;
1386 *delta_chain_length = 0;
1387 curpos = obj_offset;
1388 type = unpack_object_header(p, &w_curs, &curpos, size);
1390 revidx = find_pack_revindex(p, obj_offset);
1391 *store_size = revidx[1].offset - obj_offset;
1396 die("pack %s contains unknown object type %d",
1397 p->pack_name, type);
1402 unuse_pack(&w_curs);
1403 return typename(type);
1405 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1406 if (*delta_chain_length == 0) {
1407 revidx = find_pack_revindex(p, obj_offset);
1408 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1412 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1413 if (*delta_chain_length == 0)
1414 hashcpy(base_sha1, next_sha1);
1415 obj_offset = find_pack_entry_one(next_sha1, p);
1418 (*delta_chain_length)++;
1419 curpos = obj_offset;
1420 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1424 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1425 unsigned long *sizep)
1427 struct pack_window *w_curs = NULL;
1429 off_t curpos = obj_offset;
1430 enum object_type type;
1432 type = unpack_object_header(p, &w_curs, &curpos, &size);
1437 type = packed_delta_info(p, &w_curs, curpos,
1438 type, obj_offset, sizep);
1448 die("pack %s contains unknown object type %d",
1449 p->pack_name, type);
1451 unuse_pack(&w_curs);
1455 static void *unpack_compressed_entry(struct packed_git *p,
1456 struct pack_window **w_curs,
1462 unsigned char *buffer, *in;
1464 buffer = xmalloc(size + 1);
1466 memset(&stream, 0, sizeof(stream));
1467 stream.next_out = buffer;
1468 stream.avail_out = size;
1470 inflateInit(&stream);
1472 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1473 stream.next_in = in;
1474 st = inflate(&stream, Z_FINISH);
1475 curpos += stream.next_in - in;
1476 } while (st == Z_OK || st == Z_BUF_ERROR);
1477 inflateEnd(&stream);
1478 if ((st != Z_STREAM_END) || stream.total_out != size) {
1486 #define MAX_DELTA_CACHE (256)
1488 static size_t delta_base_cached;
1490 static struct delta_base_cache_lru_list {
1491 struct delta_base_cache_lru_list *prev;
1492 struct delta_base_cache_lru_list *next;
1493 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1495 static struct delta_base_cache_entry {
1496 struct delta_base_cache_lru_list lru;
1498 struct packed_git *p;
1501 enum object_type type;
1502 } delta_base_cache[MAX_DELTA_CACHE];
1504 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1508 hash = (unsigned long)p + (unsigned long)base_offset;
1509 hash += (hash >> 8) + (hash >> 16);
1510 return hash % MAX_DELTA_CACHE;
1513 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1514 unsigned long *base_size, enum object_type *type, int keep_cache)
1517 unsigned long hash = pack_entry_hash(p, base_offset);
1518 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1521 if (ret && ent->p == p && ent->base_offset == base_offset)
1522 goto found_cache_entry;
1523 return unpack_entry(p, base_offset, type, base_size);
1528 ent->lru.next->prev = ent->lru.prev;
1529 ent->lru.prev->next = ent->lru.next;
1530 delta_base_cached -= ent->size;
1532 ret = xmemdupz(ent->data, ent->size);
1535 *base_size = ent->size;
1539 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1544 ent->lru.next->prev = ent->lru.prev;
1545 ent->lru.prev->next = ent->lru.next;
1546 delta_base_cached -= ent->size;
1550 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1551 void *base, unsigned long base_size, enum object_type type)
1553 unsigned long hash = pack_entry_hash(p, base_offset);
1554 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1555 struct delta_base_cache_lru_list *lru;
1557 release_delta_base_cache(ent);
1558 delta_base_cached += base_size;
1560 for (lru = delta_base_cache_lru.next;
1561 delta_base_cached > delta_base_cache_limit
1562 && lru != &delta_base_cache_lru;
1564 struct delta_base_cache_entry *f = (void *)lru;
1565 if (f->type == OBJ_BLOB)
1566 release_delta_base_cache(f);
1568 for (lru = delta_base_cache_lru.next;
1569 delta_base_cached > delta_base_cache_limit
1570 && lru != &delta_base_cache_lru;
1572 struct delta_base_cache_entry *f = (void *)lru;
1573 release_delta_base_cache(f);
1577 ent->base_offset = base_offset;
1580 ent->size = base_size;
1581 ent->lru.next = &delta_base_cache_lru;
1582 ent->lru.prev = delta_base_cache_lru.prev;
1583 delta_base_cache_lru.prev->next = &ent->lru;
1584 delta_base_cache_lru.prev = &ent->lru;
1587 static void *unpack_delta_entry(struct packed_git *p,
1588 struct pack_window **w_curs,
1590 unsigned long delta_size,
1592 enum object_type *type,
1593 unsigned long *sizep)
1595 void *delta_data, *result, *base;
1596 unsigned long base_size;
1599 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1600 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1602 die("failed to read delta base object"
1603 " at %"PRIuMAX" from %s",
1604 (uintmax_t)base_offset, p->pack_name);
1606 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1608 die("failed to unpack compressed delta"
1609 " at %"PRIuMAX" from %s",
1610 (uintmax_t)curpos, p->pack_name);
1611 result = patch_delta(base, base_size,
1612 delta_data, delta_size,
1615 die("failed to apply delta");
1617 add_delta_base_cache(p, base_offset, base, base_size, *type);
1621 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1622 enum object_type *type, unsigned long *sizep)
1624 struct pack_window *w_curs = NULL;
1625 off_t curpos = obj_offset;
1628 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1632 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1633 obj_offset, type, sizep);
1639 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1642 die("unknown object type %i in %s", *type, p->pack_name);
1644 unuse_pack(&w_curs);
1648 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1651 const unsigned char *index = p->index_data;
1653 if (open_pack_index(p))
1655 index = p->index_data;
1657 if (n >= p->num_objects)
1660 if (p->index_version == 1) {
1661 return index + 24 * n + 4;
1664 return index + 20 * n;
1668 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1670 const unsigned char *index = p->index_data;
1672 if (p->index_version == 1) {
1673 return ntohl(*((uint32_t *)(index + 24 * n)));
1676 index += 8 + p->num_objects * (20 + 4);
1677 off = ntohl(*((uint32_t *)(index + 4 * n)));
1678 if (!(off & 0x80000000))
1680 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1681 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1682 ntohl(*((uint32_t *)(index + 4)));
1686 off_t find_pack_entry_one(const unsigned char *sha1,
1687 struct packed_git *p)
1689 const uint32_t *level1_ofs = p->index_data;
1690 const unsigned char *index = p->index_data;
1691 unsigned hi, lo, stride;
1692 static int use_lookup = -1;
1693 static int debug_lookup = -1;
1695 if (debug_lookup < 0)
1696 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1699 if (open_pack_index(p))
1701 level1_ofs = p->index_data;
1702 index = p->index_data;
1704 if (p->index_version > 1) {
1709 hi = ntohl(level1_ofs[*sha1]);
1710 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1711 if (p->index_version > 1) {
1719 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1720 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1723 use_lookup = !!getenv("GIT_USE_LOOKUP");
1725 int pos = sha1_entry_pos(index, stride, 0,
1726 lo, hi, p->num_objects, sha1);
1729 return nth_packed_object_offset(p, pos);
1733 unsigned mi = (lo + hi) / 2;
1734 int cmp = hashcmp(index + mi * stride, sha1);
1737 printf("lo %u hi %u rg %u mi %u\n",
1738 lo, hi, hi - lo, mi);
1740 return nth_packed_object_offset(p, mi);
1749 int matches_pack_name(struct packed_git *p, const char *name)
1751 const char *last_c, *c;
1753 if (!strcmp(p->pack_name, name))
1756 for (c = p->pack_name, last_c = c; *c;)
1761 if (!strcmp(last_c, name))
1767 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1769 static struct packed_git *last_found = (void *)1;
1770 struct packed_git *p;
1773 prepare_packed_git();
1776 p = (last_found == (void *)1) ? packed_git : last_found;
1779 if (ignore_packed) {
1781 for (ig = ignore_packed; *ig; ig++)
1782 if (matches_pack_name(p, *ig))
1788 offset = find_pack_entry_one(sha1, p);
1791 * We are about to tell the caller where they can
1792 * locate the requested object. We better make
1793 * sure the packfile is still here and can be
1794 * accessed before supplying that answer, as
1795 * it may have been deleted since the index
1798 if (p->pack_fd == -1 && open_packed_git(p)) {
1799 error("packfile %s cannot be accessed", p->pack_name);
1804 hashcpy(e->sha1, sha1);
1810 if (p == last_found)
1814 if (p == last_found)
1820 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1821 struct packed_git *packs)
1823 struct packed_git *p;
1825 for (p = packs; p; p = p->next) {
1826 if (find_pack_entry_one(sha1, p))
1833 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1836 unsigned long mapsize, size;
1841 map = map_sha1_file(sha1, &mapsize);
1843 return error("unable to find %s", sha1_to_hex(sha1));
1844 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1845 status = error("unable to unpack %s header",
1847 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1848 status = error("unable to parse %s header", sha1_to_hex(sha1));
1851 inflateEnd(&stream);
1852 munmap(map, mapsize);
1856 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1858 struct pack_entry e;
1860 if (!find_pack_entry(sha1, &e, NULL)) {
1861 reprepare_packed_git();
1862 if (!find_pack_entry(sha1, &e, NULL))
1863 return sha1_loose_object_info(sha1, sizep);
1865 return packed_object_info(e.p, e.offset, sizep);
1868 static void *read_packed_sha1(const unsigned char *sha1,
1869 enum object_type *type, unsigned long *size)
1871 struct pack_entry e;
1873 if (!find_pack_entry(sha1, &e, NULL))
1876 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1880 * This is meant to hold a *small* number of objects that you would
1881 * want read_sha1_file() to be able to return, but yet you do not want
1882 * to write them into the object store (e.g. a browse-only
1885 static struct cached_object {
1886 unsigned char sha1[20];
1887 enum object_type type;
1891 static int cached_object_nr, cached_object_alloc;
1893 static struct cached_object empty_tree = {
1894 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1895 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1896 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1902 static struct cached_object *find_cached_object(const unsigned char *sha1)
1905 struct cached_object *co = cached_objects;
1907 for (i = 0; i < cached_object_nr; i++, co++) {
1908 if (!hashcmp(co->sha1, sha1))
1911 if (!hashcmp(sha1, empty_tree.sha1))
1916 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1917 unsigned char *sha1)
1919 struct cached_object *co;
1921 hash_sha1_file(buf, len, typename(type), sha1);
1922 if (has_sha1_file(sha1) || find_cached_object(sha1))
1924 if (cached_object_alloc <= cached_object_nr) {
1925 cached_object_alloc = alloc_nr(cached_object_alloc);
1926 cached_objects = xrealloc(cached_objects,
1927 sizeof(*cached_objects) *
1928 cached_object_alloc);
1930 co = &cached_objects[cached_object_nr++];
1933 co->buf = xmalloc(len);
1934 memcpy(co->buf, buf, len);
1935 hashcpy(co->sha1, sha1);
1939 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1940 unsigned long *size)
1942 unsigned long mapsize;
1944 struct cached_object *co;
1946 co = find_cached_object(sha1);
1950 return xmemdupz(co->buf, co->size);
1953 buf = read_packed_sha1(sha1, type, size);
1956 map = map_sha1_file(sha1, &mapsize);
1958 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1959 munmap(map, mapsize);
1962 reprepare_packed_git();
1963 return read_packed_sha1(sha1, type, size);
1966 void *read_object_with_reference(const unsigned char *sha1,
1967 const char *required_type_name,
1968 unsigned long *size,
1969 unsigned char *actual_sha1_return)
1971 enum object_type type, required_type;
1973 unsigned long isize;
1974 unsigned char actual_sha1[20];
1976 required_type = type_from_string(required_type_name);
1977 hashcpy(actual_sha1, sha1);
1979 int ref_length = -1;
1980 const char *ref_type = NULL;
1982 buffer = read_sha1_file(actual_sha1, &type, &isize);
1985 if (type == required_type) {
1987 if (actual_sha1_return)
1988 hashcpy(actual_sha1_return, actual_sha1);
1991 /* Handle references */
1992 else if (type == OBJ_COMMIT)
1994 else if (type == OBJ_TAG)
1995 ref_type = "object ";
2000 ref_length = strlen(ref_type);
2002 if (ref_length + 40 > isize ||
2003 memcmp(buffer, ref_type, ref_length) ||
2004 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2009 /* Now we have the ID of the referred-to object in
2010 * actual_sha1. Check again. */
2014 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2015 const char *type, unsigned char *sha1,
2016 char *hdr, int *hdrlen)
2020 /* Generate the header */
2021 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2025 SHA1_Update(&c, hdr, *hdrlen);
2026 SHA1_Update(&c, buf, len);
2027 SHA1_Final(sha1, &c);
2031 * Link the tempfile to the final place, possibly creating the
2032 * last directory level as you do so.
2034 * Returns the errno on failure, 0 on success.
2036 static int link_temp_to_file(const char *tmpfile, const char *filename)
2041 if (!link(tmpfile, filename))
2045 * Try to mkdir the last path component if that failed.
2047 * Re-try the "link()" regardless of whether the mkdir
2048 * succeeds, since a race might mean that somebody
2052 dir = strrchr(filename, '/');
2055 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2060 if (!link(tmpfile, filename))
2068 * Move the just written object into its final resting place
2070 int move_temp_to_file(const char *tmpfile, const char *filename)
2072 int ret = link_temp_to_file(tmpfile, filename);
2075 * Coda hack - coda doesn't like cross-directory links,
2076 * so we fall back to a rename, which will mean that it
2077 * won't be able to check collisions, but that's not a
2080 * The same holds for FAT formatted media.
2082 * When this succeeds, we just return 0. We have nothing
2085 if (ret && ret != EEXIST) {
2086 if (!rename(tmpfile, filename))
2092 if (ret != EEXIST) {
2093 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2095 /* FIXME!!! Collision check here ? */
2101 static int write_buffer(int fd, const void *buf, size_t len)
2103 if (write_in_full(fd, buf, len) < 0)
2104 return error("file write error (%s)", strerror(errno));
2108 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2109 unsigned char *sha1)
2113 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2117 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2120 unsigned char *compressed;
2122 unsigned char sha1[20];
2124 static char tmpfile[PATH_MAX];
2128 /* Normally if we have it in the pack then we do not bother writing
2129 * it out into .git/objects/??/?{38} file.
2131 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2132 filename = sha1_file_name(sha1);
2134 hashcpy(returnsha1, sha1);
2135 if (has_sha1_file(sha1))
2137 fd = open(filename, O_RDONLY);
2140 * FIXME!!! We might do collision checking here, but we'd
2141 * need to uncompress the old file and check it. Later.
2147 if (errno != ENOENT) {
2148 return error("sha1 file %s: %s\n", filename, strerror(errno));
2151 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2153 fd = mkstemp(tmpfile);
2156 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2158 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2162 memset(&stream, 0, sizeof(stream));
2163 deflateInit(&stream, zlib_compression_level);
2164 size = 8 + deflateBound(&stream, len+hdrlen);
2165 compressed = xmalloc(size);
2168 stream.next_out = compressed;
2169 stream.avail_out = size;
2171 /* First header.. */
2172 stream.next_in = (unsigned char *)hdr;
2173 stream.avail_in = hdrlen;
2174 while (deflate(&stream, 0) == Z_OK)
2177 /* Then the data itself.. */
2178 stream.next_in = buf;
2179 stream.avail_in = len;
2180 ret = deflate(&stream, Z_FINISH);
2181 if (ret != Z_STREAM_END)
2182 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2184 ret = deflateEnd(&stream);
2186 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2188 size = stream.total_out;
2190 if (write_buffer(fd, compressed, size) < 0)
2191 die("unable to write sha1 file");
2194 die("unable to write sha1 file");
2197 return move_temp_to_file(tmpfile, filename);
2201 * We need to unpack and recompress the object for writing
2202 * it out to a different file.
2204 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2208 unsigned char *unpacked;
2210 enum object_type type;
2215 /* need to unpack and recompress it by itself */
2216 unpacked = read_packed_sha1(sha1, &type, &len);
2218 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2220 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2223 memset(&stream, 0, sizeof(stream));
2224 deflateInit(&stream, zlib_compression_level);
2225 size = deflateBound(&stream, len + hdrlen);
2226 buf = xmalloc(size);
2229 stream.next_out = buf;
2230 stream.avail_out = size;
2232 /* First header.. */
2233 stream.next_in = (void *)hdr;
2234 stream.avail_in = hdrlen;
2235 while (deflate(&stream, 0) == Z_OK)
2238 /* Then the data itself.. */
2239 stream.next_in = unpacked;
2240 stream.avail_in = len;
2241 while (deflate(&stream, Z_FINISH) == Z_OK)
2243 deflateEnd(&stream);
2246 *objsize = stream.total_out;
2250 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2253 unsigned long objsize;
2254 void *buf = map_sha1_file(sha1, &objsize);
2257 retval = write_buffer(fd, buf, objsize);
2258 munmap(buf, objsize);
2262 buf = repack_object(sha1, &objsize);
2263 retval = write_buffer(fd, buf, objsize);
2268 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2269 size_t bufsize, size_t *bufposn)
2271 char tmpfile[PATH_MAX];
2274 unsigned char real_sha1[20];
2275 unsigned char discard[4096];
2279 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2281 local = mkstemp(tmpfile);
2284 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2286 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2289 memset(&stream, 0, sizeof(stream));
2291 inflateInit(&stream);
2298 stream.avail_in = *bufposn;
2299 stream.next_in = (unsigned char *) buffer;
2301 stream.next_out = discard;
2302 stream.avail_out = sizeof(discard);
2303 ret = inflate(&stream, Z_SYNC_FLUSH);
2304 SHA1_Update(&c, discard, sizeof(discard) -
2306 } while (stream.avail_in && ret == Z_OK);
2307 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2308 die("unable to write sha1 file");
2309 memmove(buffer, buffer + *bufposn - stream.avail_in,
2311 *bufposn = stream.avail_in;
2315 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2320 return error("Connection closed?");
2321 perror("Reading from connection");
2326 inflateEnd(&stream);
2328 fchmod(local, 0444);
2329 if (close(local) != 0)
2330 die("unable to write sha1 file");
2331 SHA1_Final(real_sha1, &c);
2332 if (ret != Z_STREAM_END) {
2334 return error("File %s corrupted", sha1_to_hex(sha1));
2336 if (hashcmp(sha1, real_sha1)) {
2338 return error("File %s has bad hash", sha1_to_hex(sha1));
2341 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2344 int has_pack_index(const unsigned char *sha1)
2347 if (stat(sha1_pack_index_name(sha1), &st))
2352 int has_pack_file(const unsigned char *sha1)
2355 if (stat(sha1_pack_name(sha1), &st))
2360 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2362 struct pack_entry e;
2363 return find_pack_entry(sha1, &e, ignore_packed);
2366 int has_sha1_file(const unsigned char *sha1)
2369 struct pack_entry e;
2371 if (find_pack_entry(sha1, &e, NULL))
2373 return find_sha1_file(sha1, &st) ? 1 : 0;
2376 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2381 strbuf_init(&buf, 0);
2382 if (strbuf_read(&buf, fd, 4096) < 0) {
2383 strbuf_release(&buf);
2390 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2392 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2393 strbuf_release(&buf);
2398 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2399 enum object_type type, const char *path)
2401 size_t size = xsize_t(st->st_size);
2403 int ret, re_allocated = 0;
2406 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2413 * Convert blobs to git internal format
2415 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2417 strbuf_init(&nbuf, 0);
2418 if (convert_to_git(path, buf, size, &nbuf,
2419 write_object ? safe_crlf : 0)) {
2421 buf = strbuf_detach(&nbuf, &size);
2427 ret = write_sha1_file(buf, size, typename(type), sha1);
2429 ret = hash_sha1_file(buf, size, typename(type), sha1);
2439 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2445 switch (st->st_mode & S_IFMT) {
2447 fd = open(path, O_RDONLY);
2449 return error("open(\"%s\"): %s", path,
2451 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2452 return error("%s: failed to insert into database",
2456 len = xsize_t(st->st_size);
2457 target = xmalloc(len + 1);
2458 if (readlink(path, target, len + 1) != st->st_size) {
2459 char *errstr = strerror(errno);
2461 return error("readlink(\"%s\"): %s", path,
2465 hash_sha1_file(target, len, blob_type, sha1);
2466 else if (write_sha1_file(target, len, blob_type, sha1))
2467 return error("%s: failed to insert into database",
2472 return resolve_gitlink_ref(path, "HEAD", sha1);
2474 return error("%s: unsupported file type", path);
2479 int read_pack_header(int fd, struct pack_header *header)
2481 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2482 /* "eof before pack header was fully read" */
2483 return PH_ERROR_EOF;
2485 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2486 /* "protocol error (pack signature mismatch detected)" */
2487 return PH_ERROR_PACK_SIGNATURE;
2488 if (!pack_version_ok(header->hdr_version))
2489 /* "protocol error (pack version unsupported)" */
2490 return PH_ERROR_PROTOCOL;