2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
28 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 static size_t sz_fmt(size_t s) { return s; }
34 const unsigned char null_sha1[20];
36 static unsigned int sha1_file_open_flag = O_NOATIME;
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 int safe_create_leading_directories(char *path)
91 if (is_absolute_path(path))
95 pos = strchr(pos, '/');
99 if (!stat(path, &st)) {
101 if (!S_ISDIR(st.st_mode)) {
106 else if (mkdir(path, 0777)) {
110 else if (adjust_shared_perm(path)) {
119 char * sha1_to_hex(const unsigned char *sha1)
122 static char hexbuffer[4][50];
123 static const char hex[] = "0123456789abcdef";
124 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
127 for (i = 0; i < 20; i++) {
128 unsigned int val = *sha1++;
129 *buf++ = hex[val >> 4];
130 *buf++ = hex[val & 0xf];
137 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
140 for (i = 0; i < 20; i++) {
141 static char hex[] = "0123456789abcdef";
142 unsigned int val = sha1[i];
143 char *pos = pathbuf + i*2 + (i > 0);
144 *pos++ = hex[val >> 4];
145 *pos = hex[val & 0xf];
150 * NOTE! This returns a statically allocated buffer, so you have to be
151 * careful about using it. Do an "xstrdup()" if you need to save the
154 * Also note that this returns the location for creating. Reading
155 * SHA1 file can happen from any alternate directory listed in the
156 * DB_ENVIRONMENT environment variable if it is not found in
157 * the primary object database.
159 char *sha1_file_name(const unsigned char *sha1)
161 static char *name, *base;
164 const char *sha1_file_directory = get_object_directory();
165 int len = strlen(sha1_file_directory);
166 base = xmalloc(len + 60);
167 memcpy(base, sha1_file_directory, len);
168 memset(base+len, 0, 60);
171 name = base + len + 1;
173 fill_sha1_path(name, sha1);
177 char *sha1_pack_name(const unsigned char *sha1)
179 static const char hex[] = "0123456789abcdef";
180 static char *name, *base, *buf;
184 const char *sha1_file_directory = get_object_directory();
185 int len = strlen(sha1_file_directory);
186 base = xmalloc(len + 60);
187 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
188 name = base + len + 11;
193 for (i = 0; i < 20; i++) {
194 unsigned int val = *sha1++;
195 *buf++ = hex[val >> 4];
196 *buf++ = hex[val & 0xf];
202 char *sha1_pack_index_name(const unsigned char *sha1)
204 static const char hex[] = "0123456789abcdef";
205 static char *name, *base, *buf;
209 const char *sha1_file_directory = get_object_directory();
210 int len = strlen(sha1_file_directory);
211 base = xmalloc(len + 60);
212 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
213 name = base + len + 11;
218 for (i = 0; i < 20; i++) {
219 unsigned int val = *sha1++;
220 *buf++ = hex[val >> 4];
221 *buf++ = hex[val & 0xf];
227 struct alternate_object_database *alt_odb_list;
228 static struct alternate_object_database **alt_odb_tail;
230 static void read_info_alternates(const char * alternates, int depth);
233 * Prepare alternate object database registry.
235 * The variable alt_odb_list points at the list of struct
236 * alternate_object_database. The elements on this list come from
237 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
238 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
239 * whose contents is similar to that environment variable but can be
240 * LF separated. Its base points at a statically allocated buffer that
241 * contains "/the/directory/corresponding/to/.git/objects/...", while
242 * its name points just after the slash at the end of ".git/objects/"
243 * in the example above, and has enough space to hold 40-byte hex
244 * SHA1, an extra slash for the first level indirection, and the
247 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
250 const char *objdir = get_object_directory();
251 struct alternate_object_database *ent;
252 struct alternate_object_database *alt;
253 /* 43 = 40-byte + 2 '/' + terminating NUL */
255 int entlen = pfxlen + 43;
258 if (!is_absolute_path(entry) && relative_base) {
259 /* Relative alt-odb */
261 base_len = strlen(relative_base) + 1;
265 ent = xmalloc(sizeof(*ent) + entlen);
267 if (!is_absolute_path(entry) && relative_base) {
268 memcpy(ent->base, relative_base, base_len - 1);
269 ent->base[base_len - 1] = '/';
270 memcpy(ent->base + base_len, entry, len);
273 memcpy(ent->base, entry, pfxlen);
275 ent->name = ent->base + pfxlen + 1;
276 ent->base[pfxlen + 3] = '/';
277 ent->base[pfxlen] = ent->base[entlen-1] = 0;
279 /* Detect cases where alternate disappeared */
280 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
291 for (alt = alt_odb_list; alt; alt = alt->next) {
292 if (!memcmp(ent->base, alt->base, pfxlen)) {
297 if (!memcmp(ent->base, objdir, pfxlen)) {
302 /* add the alternate entry */
304 alt_odb_tail = &(ent->next);
307 /* recursively add alternates */
308 read_info_alternates(ent->base, depth + 1);
310 ent->base[pfxlen] = '/';
315 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
316 const char *relative_base, int depth)
318 const char *cp, *last;
321 error("%s: ignoring alternate object stores, nesting too deep.",
329 if (cp < ep && *cp == '#') {
330 while (cp < ep && *cp != sep)
335 while (cp < ep && *cp != sep)
338 if (!is_absolute_path(last) && depth) {
339 error("%s: ignoring relative alternate object store %s",
340 relative_base, last);
342 link_alt_odb_entry(last, cp - last,
343 relative_base, depth);
346 while (cp < ep && *cp == sep)
352 static void read_info_alternates(const char * relative_base, int depth)
357 const char alt_file_name[] = "info/alternates";
358 /* Given that relative_base is no longer than PATH_MAX,
359 ensure that "path" has enough space to append "/", the
360 file name, "info/alternates", and a trailing NUL. */
361 char path[PATH_MAX + 1 + sizeof alt_file_name];
364 sprintf(path, "%s/%s", relative_base, alt_file_name);
365 fd = open(path, O_RDONLY);
368 if (fstat(fd, &st) || (st.st_size == 0)) {
372 mapsz = xsize_t(st.st_size);
373 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
376 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
381 void prepare_alt_odb(void)
388 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
391 alt_odb_tail = &alt_odb_list;
392 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
394 read_info_alternates(get_object_directory(), 0);
397 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
399 char *name = sha1_file_name(sha1);
400 struct alternate_object_database *alt;
405 for (alt = alt_odb_list; alt; alt = alt->next) {
407 fill_sha1_path(name, sha1);
408 if (!stat(alt->base, st))
414 static unsigned int pack_used_ctr;
415 static unsigned int pack_mmap_calls;
416 static unsigned int peak_pack_open_windows;
417 static unsigned int pack_open_windows;
418 static size_t peak_pack_mapped;
419 static size_t pack_mapped;
420 struct packed_git *packed_git;
422 void pack_report(void)
425 "pack_report: getpagesize() = %10" SZ_FMT "\n"
426 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
427 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
428 sz_fmt(getpagesize()),
429 sz_fmt(packed_git_window_size),
430 sz_fmt(packed_git_limit));
432 "pack_report: pack_used_ctr = %10u\n"
433 "pack_report: pack_mmap_calls = %10u\n"
434 "pack_report: pack_open_windows = %10u / %10u\n"
435 "pack_report: pack_mapped = "
436 "%10" SZ_FMT " / %10" SZ_FMT "\n",
439 pack_open_windows, peak_pack_open_windows,
440 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
443 static int check_packed_git_idx(const char *path, struct packed_git *p)
446 struct pack_idx_header *hdr;
448 uint32_t version, nr, i, *index;
449 int fd = open(path, O_RDONLY);
454 if (fstat(fd, &st)) {
458 idx_size = xsize_t(st.st_size);
459 if (idx_size < 4 * 256 + 20 + 20) {
461 return error("index file %s is too small", path);
463 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
467 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
468 version = ntohl(hdr->idx_version);
469 if (version < 2 || version > 2) {
470 munmap(idx_map, idx_size);
471 return error("index file %s is version %d"
472 " and is not supported by this binary"
473 " (try upgrading GIT to a newer version)",
482 index += 2; /* skip index header */
483 for (i = 0; i < 256; i++) {
484 uint32_t n = ntohl(index[i]);
486 munmap(idx_map, idx_size);
487 return error("non-monotonic index %s", path);
495 * - 256 index entries 4 bytes each
496 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
497 * - 20-byte SHA1 of the packfile
498 * - 20-byte SHA1 file checksum
500 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
501 munmap(idx_map, idx_size);
502 return error("wrong index v1 file size in %s", path);
504 } else if (version == 2) {
507 * - 8 bytes of header
508 * - 256 index entries 4 bytes each
509 * - 20-byte sha1 entry * nr
510 * - 4-byte crc entry * nr
511 * - 4-byte offset entry * nr
512 * - 20-byte SHA1 of the packfile
513 * - 20-byte SHA1 file checksum
514 * And after the 4-byte offset table might be a
515 * variable sized table containing 8-byte entries
516 * for offsets larger than 2^31.
518 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
519 unsigned long max_size = min_size;
521 max_size += (nr - 1)*8;
522 if (idx_size < min_size || idx_size > max_size) {
523 munmap(idx_map, idx_size);
524 return error("wrong index v2 file size in %s", path);
526 if (idx_size != min_size &&
528 * make sure we can deal with large pack offsets.
529 * 31-bit signed offset won't be enough, neither
530 * 32-bit unsigned one will be.
532 (sizeof(off_t) <= 4)) {
533 munmap(idx_map, idx_size);
534 return error("pack too large for current definition of off_t in %s", path);
538 p->index_version = version;
539 p->index_data = idx_map;
540 p->index_size = idx_size;
545 int open_pack_index(struct packed_git *p)
553 idx_name = xstrdup(p->pack_name);
554 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
555 ret = check_packed_git_idx(idx_name, p);
560 static void scan_windows(struct packed_git *p,
561 struct packed_git **lru_p,
562 struct pack_window **lru_w,
563 struct pack_window **lru_l)
565 struct pack_window *w, *w_l;
567 for (w_l = NULL, w = p->windows; w; w = w->next) {
569 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
579 static int unuse_one_window(struct packed_git *current, int keep_fd)
581 struct packed_git *p, *lru_p = NULL;
582 struct pack_window *lru_w = NULL, *lru_l = NULL;
585 scan_windows(current, &lru_p, &lru_w, &lru_l);
586 for (p = packed_git; p; p = p->next)
587 scan_windows(p, &lru_p, &lru_w, &lru_l);
589 munmap(lru_w->base, lru_w->len);
590 pack_mapped -= lru_w->len;
592 lru_l->next = lru_w->next;
594 lru_p->windows = lru_w->next;
595 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
596 close(lru_p->pack_fd);
607 void release_pack_memory(size_t need, int fd)
609 size_t cur = pack_mapped;
610 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
614 void close_pack_windows(struct packed_git *p)
617 struct pack_window *w = p->windows;
620 die("pack '%s' still has open windows to it",
622 munmap(w->base, w->len);
623 pack_mapped -= w->len;
625 p->windows = w->next;
630 void unuse_pack(struct pack_window **w_cursor)
632 struct pack_window *w = *w_cursor;
640 * Do not call this directly as this leaks p->pack_fd on error return;
641 * call open_packed_git() instead.
643 static int open_packed_git_1(struct packed_git *p)
646 struct pack_header hdr;
647 unsigned char sha1[20];
648 unsigned char *idx_sha1;
651 if (!p->index_data && open_pack_index(p))
652 return error("packfile %s index unavailable", p->pack_name);
654 p->pack_fd = open(p->pack_name, O_RDONLY);
655 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
658 /* If we created the struct before we had the pack we lack size. */
660 if (!S_ISREG(st.st_mode))
661 return error("packfile %s not a regular file", p->pack_name);
662 p->pack_size = st.st_size;
663 } else if (p->pack_size != st.st_size)
664 return error("packfile %s size changed", p->pack_name);
666 /* We leave these file descriptors open with sliding mmap;
667 * there is no point keeping them open across exec(), though.
669 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
671 return error("cannot determine file descriptor flags");
672 fd_flag |= FD_CLOEXEC;
673 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
674 return error("cannot set FD_CLOEXEC");
676 /* Verify we recognize this pack file format. */
677 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
678 return error("file %s is far too short to be a packfile", p->pack_name);
679 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
680 return error("file %s is not a GIT packfile", p->pack_name);
681 if (!pack_version_ok(hdr.hdr_version))
682 return error("packfile %s is version %u and not supported"
683 " (try upgrading GIT to a newer version)",
684 p->pack_name, ntohl(hdr.hdr_version));
686 /* Verify the pack matches its index. */
687 if (p->num_objects != ntohl(hdr.hdr_entries))
688 return error("packfile %s claims to have %u objects"
689 " while index indicates %u objects",
690 p->pack_name, ntohl(hdr.hdr_entries),
692 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
693 return error("end of packfile %s is unavailable", p->pack_name);
694 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
695 return error("packfile %s signature is unavailable", p->pack_name);
696 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
697 if (hashcmp(sha1, idx_sha1))
698 return error("packfile %s does not match index", p->pack_name);
702 static int open_packed_git(struct packed_git *p)
704 if (!open_packed_git_1(p))
706 if (p->pack_fd != -1) {
713 static int in_window(struct pack_window *win, off_t offset)
715 /* We must promise at least 20 bytes (one hash) after the
716 * offset is available from this window, otherwise the offset
717 * is not actually in this window and a different window (which
718 * has that one hash excess) must be used. This is to support
719 * the object header and delta base parsing routines below.
721 off_t win_off = win->offset;
722 return win_off <= offset
723 && (offset + 20) <= (win_off + win->len);
726 unsigned char* use_pack(struct packed_git *p,
727 struct pack_window **w_cursor,
731 struct pack_window *win = *w_cursor;
733 if (p->pack_fd == -1 && open_packed_git(p))
734 die("packfile %s cannot be accessed", p->pack_name);
736 /* Since packfiles end in a hash of their content and its
737 * pointless to ask for an offset into the middle of that
738 * hash, and the in_window function above wouldn't match
739 * don't allow an offset too close to the end of the file.
741 if (offset > (p->pack_size - 20))
742 die("offset beyond end of packfile (truncated pack?)");
744 if (!win || !in_window(win, offset)) {
747 for (win = p->windows; win; win = win->next) {
748 if (in_window(win, offset))
752 size_t window_align = packed_git_window_size / 2;
754 win = xcalloc(1, sizeof(*win));
755 win->offset = (offset / window_align) * window_align;
756 len = p->pack_size - win->offset;
757 if (len > packed_git_window_size)
758 len = packed_git_window_size;
759 win->len = (size_t)len;
760 pack_mapped += win->len;
761 while (packed_git_limit < pack_mapped
762 && unuse_one_window(p, p->pack_fd))
764 win->base = xmmap(NULL, win->len,
765 PROT_READ, MAP_PRIVATE,
766 p->pack_fd, win->offset);
767 if (win->base == MAP_FAILED)
768 die("packfile %s cannot be mapped: %s",
773 if (pack_mapped > peak_pack_mapped)
774 peak_pack_mapped = pack_mapped;
775 if (pack_open_windows > peak_pack_open_windows)
776 peak_pack_open_windows = pack_open_windows;
777 win->next = p->windows;
781 if (win != *w_cursor) {
782 win->last_used = pack_used_ctr++;
786 offset -= win->offset;
788 *left = win->len - xsize_t(offset);
789 return win->base + offset;
792 struct packed_git *add_packed_git(const char *path, int path_len, int local)
795 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
798 * Make sure a corresponding .pack file exists and that
799 * the index looks sane.
801 path_len -= strlen(".idx");
804 memcpy(p->pack_name, path, path_len);
805 strcpy(p->pack_name + path_len, ".pack");
806 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
811 /* ok, it looks sane as far as we can check without
812 * actually mapping the pack file.
814 p->index_version = 0;
815 p->index_data = NULL;
818 p->pack_size = st.st_size;
822 p->pack_local = local;
823 p->mtime = st.st_mtime;
824 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
829 struct packed_git *parse_pack_index(unsigned char *sha1)
831 char *path = sha1_pack_index_name(sha1);
832 return parse_pack_index_file(sha1, path);
835 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
836 const char *idx_path)
838 const char *path = sha1_pack_name(sha1);
839 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
841 if (check_packed_git_idx(idx_path, p)) {
846 strcpy(p->pack_name, path);
851 hashcpy(p->sha1, sha1);
855 void install_packed_git(struct packed_git *pack)
857 pack->next = packed_git;
861 static void prepare_packed_git_one(char *objdir, int local)
863 /* Ensure that this buffer is large enough so that we can
864 append "/pack/" without clobbering the stack even if
865 strlen(objdir) were PATH_MAX. */
866 char path[PATH_MAX + 1 + 4 + 1 + 1];
871 sprintf(path, "%s/pack", objdir);
876 error("unable to open object pack directory: %s: %s",
877 path, strerror(errno));
881 while ((de = readdir(dir)) != NULL) {
882 int namelen = strlen(de->d_name);
883 struct packed_git *p;
885 if (!has_extension(de->d_name, ".idx"))
888 if (len + namelen + 1 > sizeof(path))
891 /* Don't reopen a pack we already have. */
892 strcpy(path + len, de->d_name);
893 for (p = packed_git; p; p = p->next) {
894 if (!memcmp(path, p->pack_name, len + namelen - 4))
899 /* See if it really is a valid .idx file with corresponding
900 * .pack file that we can map.
902 p = add_packed_git(path, len + namelen, local);
905 install_packed_git(p);
910 static int sort_pack(const void *a_, const void *b_)
912 struct packed_git *a = *((struct packed_git **)a_);
913 struct packed_git *b = *((struct packed_git **)b_);
917 * Local packs tend to contain objects specific to our
918 * variant of the project than remote ones. In addition,
919 * remote ones could be on a network mounted filesystem.
920 * Favor local ones for these reasons.
922 st = a->pack_local - b->pack_local;
927 * Younger packs tend to contain more recent objects,
928 * and more recent objects tend to get accessed more
931 if (a->mtime < b->mtime)
933 else if (a->mtime == b->mtime)
938 static void rearrange_packed_git(void)
940 struct packed_git **ary, *p;
943 for (n = 0, p = packed_git; p; p = p->next)
948 /* prepare an array of packed_git for easier sorting */
949 ary = xcalloc(n, sizeof(struct packed_git *));
950 for (n = 0, p = packed_git; p; p = p->next)
953 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
955 /* link them back again */
956 for (i = 0; i < n - 1; i++)
957 ary[i]->next = ary[i + 1];
958 ary[n - 1]->next = NULL;
964 static int prepare_packed_git_run_once = 0;
965 void prepare_packed_git(void)
967 struct alternate_object_database *alt;
969 if (prepare_packed_git_run_once)
971 prepare_packed_git_one(get_object_directory(), 1);
973 for (alt = alt_odb_list; alt; alt = alt->next) {
975 prepare_packed_git_one(alt->base, 0);
978 rearrange_packed_git();
979 prepare_packed_git_run_once = 1;
982 void reprepare_packed_git(void)
984 prepare_packed_git_run_once = 0;
985 prepare_packed_git();
988 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
990 unsigned char real_sha1[20];
991 hash_sha1_file(map, size, type, real_sha1);
992 return hashcmp(sha1, real_sha1) ? -1 : 0;
995 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1000 char *filename = find_sha1_file(sha1, &st);
1006 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1008 /* See if it works without O_NOATIME */
1009 switch (sha1_file_open_flag) {
1011 fd = open(filename, O_RDONLY);
1019 /* If it failed once, it will probably fail again.
1020 * Stop using O_NOATIME
1022 sha1_file_open_flag = 0;
1024 *size = xsize_t(st.st_size);
1025 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1030 static int legacy_loose_object(unsigned char *map)
1035 * Is it a zlib-compressed buffer? If so, the first byte
1036 * must be 0x78 (15-bit window size, deflated), and the
1037 * first 16-bit word is evenly divisible by 31
1039 word = (map[0] << 8) + map[1];
1040 if (map[0] == 0x78 && !(word % 31))
1046 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1051 unsigned long used = 0;
1054 *type = (c >> 4) & 7;
1060 if (sizeof(long) * 8 <= shift)
1063 size += (c & 0x7f) << shift;
1070 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1072 unsigned long size, used;
1073 static const char valid_loose_object_type[8] = {
1075 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1076 0, /* "delta" and others are invalid in a loose object */
1078 enum object_type type;
1080 /* Get the data stream */
1081 memset(stream, 0, sizeof(*stream));
1082 stream->next_in = map;
1083 stream->avail_in = mapsize;
1084 stream->next_out = buffer;
1085 stream->avail_out = bufsiz;
1087 if (legacy_loose_object(map)) {
1088 inflateInit(stream);
1089 return inflate(stream, 0);
1094 * There used to be a second loose object header format which
1095 * was meant to mimic the in-pack format, allowing for direct
1096 * copy of the object data. This format turned up not to be
1097 * really worth it and we don't write it any longer. But we
1098 * can still read it.
1100 used = unpack_object_header_gently(map, mapsize, &type, &size);
1101 if (!used || !valid_loose_object_type[type])
1106 /* Set up the stream for the rest.. */
1107 stream->next_in = map;
1108 stream->avail_in = mapsize;
1109 inflateInit(stream);
1111 /* And generate the fake traditional header */
1112 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1113 typename(type), size);
1117 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1119 int bytes = strlen(buffer) + 1;
1120 unsigned char *buf = xmalloc(1+size);
1124 n = stream->total_out - bytes;
1127 memcpy(buf, (char *) buffer + bytes, n);
1129 if (bytes <= size) {
1131 * The above condition must be (bytes <= size), not
1132 * (bytes < size). In other words, even though we
1133 * expect no more output and set avail_out to zer0,
1134 * the input zlib stream may have bytes that express
1135 * "this concludes the stream", and we *do* want to
1138 * Otherwise we would not be able to test that we
1139 * consumed all the input to reach the expected size;
1140 * we also want to check that zlib tells us that all
1141 * went well with status == Z_STREAM_END at the end.
1143 stream->next_out = buf + bytes;
1144 stream->avail_out = size - bytes;
1145 while (status == Z_OK)
1146 status = inflate(stream, Z_FINISH);
1149 if (status == Z_STREAM_END && !stream->avail_in) {
1155 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1156 else if (stream->avail_in)
1157 error("garbage at end of loose object '%s'",
1164 * We used to just use "sscanf()", but that's actually way
1165 * too permissive for what we want to check. So do an anal
1166 * object header parse by hand.
1168 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1175 * The type can be at most ten bytes (including the
1176 * terminating '\0' that we add), and is followed by
1185 if (i >= sizeof(type))
1191 * The length must follow immediately, and be in canonical
1192 * decimal format (ie "010" is not valid).
1194 size = *hdr++ - '0';
1199 unsigned long c = *hdr - '0';
1203 size = size * 10 + c;
1209 * The length must be followed by a zero byte
1211 return *hdr ? -1 : type_from_string(type);
1214 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1220 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1221 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1224 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1227 unsigned long get_size_from_delta(struct packed_git *p,
1228 struct pack_window **w_curs,
1231 const unsigned char *data;
1232 unsigned char delta_head[20], *in;
1236 memset(&stream, 0, sizeof(stream));
1237 stream.next_out = delta_head;
1238 stream.avail_out = sizeof(delta_head);
1240 inflateInit(&stream);
1242 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1243 stream.next_in = in;
1244 st = inflate(&stream, Z_FINISH);
1245 curpos += stream.next_in - in;
1246 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1247 stream.total_out < sizeof(delta_head));
1248 inflateEnd(&stream);
1249 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1250 die("delta data unpack-initial failed");
1252 /* Examine the initial part of the delta to figure out
1257 /* ignore base size */
1258 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1260 /* Read the result size */
1261 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1264 static off_t get_delta_base(struct packed_git *p,
1265 struct pack_window **w_curs,
1267 enum object_type type,
1268 off_t delta_obj_offset)
1270 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1273 /* use_pack() assured us we have [base_info, base_info + 20)
1274 * as a range that we can look at without walking off the
1275 * end of the mapped window. Its actually the hash size
1276 * that is assured. An OFS_DELTA longer than the hash size
1277 * is stupid, as then a REF_DELTA would be smaller to store.
1279 if (type == OBJ_OFS_DELTA) {
1281 unsigned char c = base_info[used++];
1282 base_offset = c & 127;
1285 if (!base_offset || MSB(base_offset, 7))
1286 die("offset value overflow for delta base object");
1287 c = base_info[used++];
1288 base_offset = (base_offset << 7) + (c & 127);
1290 base_offset = delta_obj_offset - base_offset;
1291 if (base_offset >= delta_obj_offset)
1292 die("delta base offset out of bound");
1294 } else if (type == OBJ_REF_DELTA) {
1295 /* The base entry _must_ be in the same pack */
1296 base_offset = find_pack_entry_one(base_info, p);
1298 die("failed to find delta-pack base object %s",
1299 sha1_to_hex(base_info));
1302 die("I am totally screwed");
1306 /* forward declaration for a mutually recursive function */
1307 static int packed_object_info(struct packed_git *p, off_t offset,
1308 unsigned long *sizep);
1310 static int packed_delta_info(struct packed_git *p,
1311 struct pack_window **w_curs,
1313 enum object_type type,
1315 unsigned long *sizep)
1319 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1320 type = packed_object_info(p, base_offset, NULL);
1322 /* We choose to only get the type of the base object and
1323 * ignore potentially corrupt pack file that expects the delta
1324 * based on a base with a wrong size. This saves tons of
1328 *sizep = get_size_from_delta(p, w_curs, curpos);
1333 static int unpack_object_header(struct packed_git *p,
1334 struct pack_window **w_curs,
1336 unsigned long *sizep)
1338 unsigned char *base;
1341 enum object_type type;
1343 /* use_pack() assures us we have [base, base + 20) available
1344 * as a range that we can look at at. (Its actually the hash
1345 * size that is assured.) With our object header encoding
1346 * the maximum deflated object size is 2^137, which is just
1347 * insane, so we know won't exceed what we have been given.
1349 base = use_pack(p, w_curs, *curpos, &left);
1350 used = unpack_object_header_gently(base, left, &type, sizep);
1352 die("object offset outside of pack file");
1358 const char *packed_object_info_detail(struct packed_git *p,
1360 unsigned long *size,
1361 unsigned long *store_size,
1362 unsigned int *delta_chain_length,
1363 unsigned char *base_sha1)
1365 struct pack_window *w_curs = NULL;
1367 unsigned long dummy;
1368 unsigned char *next_sha1;
1369 enum object_type type;
1371 *delta_chain_length = 0;
1372 curpos = obj_offset;
1373 type = unpack_object_header(p, &w_curs, &curpos, size);
1378 die("pack %s contains unknown object type %d",
1379 p->pack_name, type);
1384 *store_size = 0; /* notyet */
1385 unuse_pack(&w_curs);
1386 return typename(type);
1388 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1389 if (*delta_chain_length == 0) {
1390 /* TODO: find base_sha1 as pointed by curpos */
1395 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1396 if (*delta_chain_length == 0)
1397 hashcpy(base_sha1, next_sha1);
1398 obj_offset = find_pack_entry_one(next_sha1, p);
1401 (*delta_chain_length)++;
1402 curpos = obj_offset;
1403 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1407 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1408 unsigned long *sizep)
1410 struct pack_window *w_curs = NULL;
1412 off_t curpos = obj_offset;
1413 enum object_type type;
1415 type = unpack_object_header(p, &w_curs, &curpos, &size);
1420 type = packed_delta_info(p, &w_curs, curpos,
1421 type, obj_offset, sizep);
1431 die("pack %s contains unknown object type %d",
1432 p->pack_name, type);
1434 unuse_pack(&w_curs);
1438 static void *unpack_compressed_entry(struct packed_git *p,
1439 struct pack_window **w_curs,
1445 unsigned char *buffer, *in;
1447 buffer = xmalloc(size + 1);
1449 memset(&stream, 0, sizeof(stream));
1450 stream.next_out = buffer;
1451 stream.avail_out = size;
1453 inflateInit(&stream);
1455 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1456 stream.next_in = in;
1457 st = inflate(&stream, Z_FINISH);
1458 curpos += stream.next_in - in;
1459 } while (st == Z_OK || st == Z_BUF_ERROR);
1460 inflateEnd(&stream);
1461 if ((st != Z_STREAM_END) || stream.total_out != size) {
1469 #define MAX_DELTA_CACHE (256)
1471 static size_t delta_base_cached;
1473 static struct delta_base_cache_lru_list {
1474 struct delta_base_cache_lru_list *prev;
1475 struct delta_base_cache_lru_list *next;
1476 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1478 static struct delta_base_cache_entry {
1479 struct delta_base_cache_lru_list lru;
1481 struct packed_git *p;
1484 enum object_type type;
1485 } delta_base_cache[MAX_DELTA_CACHE];
1487 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1491 hash = (unsigned long)p + (unsigned long)base_offset;
1492 hash += (hash >> 8) + (hash >> 16);
1493 return hash % MAX_DELTA_CACHE;
1496 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1497 unsigned long *base_size, enum object_type *type, int keep_cache)
1500 unsigned long hash = pack_entry_hash(p, base_offset);
1501 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1504 if (ret && ent->p == p && ent->base_offset == base_offset)
1505 goto found_cache_entry;
1506 return unpack_entry(p, base_offset, type, base_size);
1511 ent->lru.next->prev = ent->lru.prev;
1512 ent->lru.prev->next = ent->lru.next;
1513 delta_base_cached -= ent->size;
1515 ret = xmemdupz(ent->data, ent->size);
1518 *base_size = ent->size;
1522 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1527 ent->lru.next->prev = ent->lru.prev;
1528 ent->lru.prev->next = ent->lru.next;
1529 delta_base_cached -= ent->size;
1533 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1534 void *base, unsigned long base_size, enum object_type type)
1536 unsigned long hash = pack_entry_hash(p, base_offset);
1537 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1538 struct delta_base_cache_lru_list *lru;
1540 release_delta_base_cache(ent);
1541 delta_base_cached += base_size;
1543 for (lru = delta_base_cache_lru.next;
1544 delta_base_cached > delta_base_cache_limit
1545 && lru != &delta_base_cache_lru;
1547 struct delta_base_cache_entry *f = (void *)lru;
1548 if (f->type == OBJ_BLOB)
1549 release_delta_base_cache(f);
1551 for (lru = delta_base_cache_lru.next;
1552 delta_base_cached > delta_base_cache_limit
1553 && lru != &delta_base_cache_lru;
1555 struct delta_base_cache_entry *f = (void *)lru;
1556 release_delta_base_cache(f);
1560 ent->base_offset = base_offset;
1563 ent->size = base_size;
1564 ent->lru.next = &delta_base_cache_lru;
1565 ent->lru.prev = delta_base_cache_lru.prev;
1566 delta_base_cache_lru.prev->next = &ent->lru;
1567 delta_base_cache_lru.prev = &ent->lru;
1570 static void *unpack_delta_entry(struct packed_git *p,
1571 struct pack_window **w_curs,
1573 unsigned long delta_size,
1575 enum object_type *type,
1576 unsigned long *sizep)
1578 void *delta_data, *result, *base;
1579 unsigned long base_size;
1582 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1583 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1585 die("failed to read delta base object"
1586 " at %"PRIuMAX" from %s",
1587 (uintmax_t)base_offset, p->pack_name);
1589 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1591 die("failed to unpack compressed delta"
1592 " at %"PRIuMAX" from %s",
1593 (uintmax_t)curpos, p->pack_name);
1594 result = patch_delta(base, base_size,
1595 delta_data, delta_size,
1598 die("failed to apply delta");
1600 add_delta_base_cache(p, base_offset, base, base_size, *type);
1604 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1605 enum object_type *type, unsigned long *sizep)
1607 struct pack_window *w_curs = NULL;
1608 off_t curpos = obj_offset;
1611 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1615 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1616 obj_offset, type, sizep);
1622 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1625 die("unknown object type %i in %s", *type, p->pack_name);
1627 unuse_pack(&w_curs);
1631 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1634 const unsigned char *index = p->index_data;
1636 if (open_pack_index(p))
1638 index = p->index_data;
1640 if (n >= p->num_objects)
1643 if (p->index_version == 1) {
1644 return index + 24 * n + 4;
1647 return index + 20 * n;
1651 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1653 const unsigned char *index = p->index_data;
1655 if (p->index_version == 1) {
1656 return ntohl(*((uint32_t *)(index + 24 * n)));
1659 index += 8 + p->num_objects * (20 + 4);
1660 off = ntohl(*((uint32_t *)(index + 4 * n)));
1661 if (!(off & 0x80000000))
1663 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1664 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1665 ntohl(*((uint32_t *)(index + 4)));
1669 off_t find_pack_entry_one(const unsigned char *sha1,
1670 struct packed_git *p)
1672 const uint32_t *level1_ofs = p->index_data;
1673 const unsigned char *index = p->index_data;
1677 if (open_pack_index(p))
1679 level1_ofs = p->index_data;
1680 index = p->index_data;
1682 if (p->index_version > 1) {
1687 hi = ntohl(level1_ofs[*sha1]);
1688 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1691 unsigned mi = (lo + hi) / 2;
1692 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1693 int cmp = hashcmp(index + x, sha1);
1695 return nth_packed_object_offset(p, mi);
1704 int matches_pack_name(struct packed_git *p, const char *name)
1706 const char *last_c, *c;
1708 if (!strcmp(p->pack_name, name))
1711 for (c = p->pack_name, last_c = c; *c;)
1716 if (!strcmp(last_c, name))
1722 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1724 static struct packed_git *last_found = (void *)1;
1725 struct packed_git *p;
1728 prepare_packed_git();
1731 p = (last_found == (void *)1) ? packed_git : last_found;
1734 if (ignore_packed) {
1736 for (ig = ignore_packed; *ig; ig++)
1737 if (matches_pack_name(p, *ig))
1743 offset = find_pack_entry_one(sha1, p);
1746 * We are about to tell the caller where they can
1747 * locate the requested object. We better make
1748 * sure the packfile is still here and can be
1749 * accessed before supplying that answer, as
1750 * it may have been deleted since the index
1753 if (p->pack_fd == -1 && open_packed_git(p)) {
1754 error("packfile %s cannot be accessed", p->pack_name);
1759 hashcpy(e->sha1, sha1);
1765 if (p == last_found)
1769 if (p == last_found)
1775 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1776 struct packed_git *packs)
1778 struct packed_git *p;
1780 for (p = packs; p; p = p->next) {
1781 if (find_pack_entry_one(sha1, p))
1788 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1791 unsigned long mapsize, size;
1796 map = map_sha1_file(sha1, &mapsize);
1798 return error("unable to find %s", sha1_to_hex(sha1));
1799 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1800 status = error("unable to unpack %s header",
1802 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1803 status = error("unable to parse %s header", sha1_to_hex(sha1));
1806 inflateEnd(&stream);
1807 munmap(map, mapsize);
1811 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1813 struct pack_entry e;
1815 if (!find_pack_entry(sha1, &e, NULL)) {
1816 reprepare_packed_git();
1817 if (!find_pack_entry(sha1, &e, NULL))
1818 return sha1_loose_object_info(sha1, sizep);
1820 return packed_object_info(e.p, e.offset, sizep);
1823 static void *read_packed_sha1(const unsigned char *sha1,
1824 enum object_type *type, unsigned long *size)
1826 struct pack_entry e;
1828 if (!find_pack_entry(sha1, &e, NULL))
1831 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1835 * This is meant to hold a *small* number of objects that you would
1836 * want read_sha1_file() to be able to return, but yet you do not want
1837 * to write them into the object store (e.g. a browse-only
1840 static struct cached_object {
1841 unsigned char sha1[20];
1842 enum object_type type;
1846 static int cached_object_nr, cached_object_alloc;
1848 static struct cached_object *find_cached_object(const unsigned char *sha1)
1851 struct cached_object *co = cached_objects;
1853 for (i = 0; i < cached_object_nr; i++, co++) {
1854 if (!hashcmp(co->sha1, sha1))
1860 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1861 unsigned char *sha1)
1863 struct cached_object *co;
1865 hash_sha1_file(buf, len, typename(type), sha1);
1866 if (has_sha1_file(sha1) || find_cached_object(sha1))
1868 if (cached_object_alloc <= cached_object_nr) {
1869 cached_object_alloc = alloc_nr(cached_object_alloc);
1870 cached_objects = xrealloc(cached_objects,
1871 sizeof(*cached_objects) *
1872 cached_object_alloc);
1874 co = &cached_objects[cached_object_nr++];
1877 co->buf = xmalloc(len);
1878 memcpy(co->buf, buf, len);
1879 hashcpy(co->sha1, sha1);
1883 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1884 unsigned long *size)
1886 unsigned long mapsize;
1888 struct cached_object *co;
1890 co = find_cached_object(sha1);
1894 return xmemdupz(co->buf, co->size);
1897 buf = read_packed_sha1(sha1, type, size);
1900 map = map_sha1_file(sha1, &mapsize);
1902 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1903 munmap(map, mapsize);
1906 reprepare_packed_git();
1907 return read_packed_sha1(sha1, type, size);
1910 void *read_object_with_reference(const unsigned char *sha1,
1911 const char *required_type_name,
1912 unsigned long *size,
1913 unsigned char *actual_sha1_return)
1915 enum object_type type, required_type;
1917 unsigned long isize;
1918 unsigned char actual_sha1[20];
1920 required_type = type_from_string(required_type_name);
1921 hashcpy(actual_sha1, sha1);
1923 int ref_length = -1;
1924 const char *ref_type = NULL;
1926 buffer = read_sha1_file(actual_sha1, &type, &isize);
1929 if (type == required_type) {
1931 if (actual_sha1_return)
1932 hashcpy(actual_sha1_return, actual_sha1);
1935 /* Handle references */
1936 else if (type == OBJ_COMMIT)
1938 else if (type == OBJ_TAG)
1939 ref_type = "object ";
1944 ref_length = strlen(ref_type);
1946 if (memcmp(buffer, ref_type, ref_length) ||
1947 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1952 /* Now we have the ID of the referred-to object in
1953 * actual_sha1. Check again. */
1957 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1958 const char *type, unsigned char *sha1,
1959 char *hdr, int *hdrlen)
1963 /* Generate the header */
1964 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1968 SHA1_Update(&c, hdr, *hdrlen);
1969 SHA1_Update(&c, buf, len);
1970 SHA1_Final(sha1, &c);
1974 * Link the tempfile to the final place, possibly creating the
1975 * last directory level as you do so.
1977 * Returns the errno on failure, 0 on success.
1979 static int link_temp_to_file(const char *tmpfile, const char *filename)
1984 if (!link(tmpfile, filename))
1988 * Try to mkdir the last path component if that failed.
1990 * Re-try the "link()" regardless of whether the mkdir
1991 * succeeds, since a race might mean that somebody
1995 dir = strrchr(filename, '/');
1998 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2003 if (!link(tmpfile, filename))
2011 * Move the just written object into its final resting place
2013 int move_temp_to_file(const char *tmpfile, const char *filename)
2015 int ret = link_temp_to_file(tmpfile, filename);
2018 * Coda hack - coda doesn't like cross-directory links,
2019 * so we fall back to a rename, which will mean that it
2020 * won't be able to check collisions, but that's not a
2023 * The same holds for FAT formatted media.
2025 * When this succeeds, we just return 0. We have nothing
2028 if (ret && ret != EEXIST) {
2029 if (!rename(tmpfile, filename))
2035 if (ret != EEXIST) {
2036 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2038 /* FIXME!!! Collision check here ? */
2044 static int write_buffer(int fd, const void *buf, size_t len)
2046 if (write_in_full(fd, buf, len) < 0)
2047 return error("file write error (%s)", strerror(errno));
2051 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2052 unsigned char *sha1)
2056 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2060 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2063 unsigned char *compressed;
2065 unsigned char sha1[20];
2067 static char tmpfile[PATH_MAX];
2071 /* Normally if we have it in the pack then we do not bother writing
2072 * it out into .git/objects/??/?{38} file.
2074 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2075 filename = sha1_file_name(sha1);
2077 hashcpy(returnsha1, sha1);
2078 if (has_sha1_file(sha1))
2080 fd = open(filename, O_RDONLY);
2083 * FIXME!!! We might do collision checking here, but we'd
2084 * need to uncompress the old file and check it. Later.
2090 if (errno != ENOENT) {
2091 return error("sha1 file %s: %s\n", filename, strerror(errno));
2094 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2096 fd = mkstemp(tmpfile);
2099 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2101 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2105 memset(&stream, 0, sizeof(stream));
2106 deflateInit(&stream, zlib_compression_level);
2107 size = 8 + deflateBound(&stream, len+hdrlen);
2108 compressed = xmalloc(size);
2111 stream.next_out = compressed;
2112 stream.avail_out = size;
2114 /* First header.. */
2115 stream.next_in = (unsigned char *)hdr;
2116 stream.avail_in = hdrlen;
2117 while (deflate(&stream, 0) == Z_OK)
2120 /* Then the data itself.. */
2121 stream.next_in = buf;
2122 stream.avail_in = len;
2123 ret = deflate(&stream, Z_FINISH);
2124 if (ret != Z_STREAM_END)
2125 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2127 ret = deflateEnd(&stream);
2129 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2131 size = stream.total_out;
2133 if (write_buffer(fd, compressed, size) < 0)
2134 die("unable to write sha1 file");
2137 die("unable to write sha1 file");
2140 return move_temp_to_file(tmpfile, filename);
2144 * We need to unpack and recompress the object for writing
2145 * it out to a different file.
2147 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2151 unsigned char *unpacked;
2153 enum object_type type;
2158 /* need to unpack and recompress it by itself */
2159 unpacked = read_packed_sha1(sha1, &type, &len);
2161 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2163 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2166 memset(&stream, 0, sizeof(stream));
2167 deflateInit(&stream, zlib_compression_level);
2168 size = deflateBound(&stream, len + hdrlen);
2169 buf = xmalloc(size);
2172 stream.next_out = buf;
2173 stream.avail_out = size;
2175 /* First header.. */
2176 stream.next_in = (void *)hdr;
2177 stream.avail_in = hdrlen;
2178 while (deflate(&stream, 0) == Z_OK)
2181 /* Then the data itself.. */
2182 stream.next_in = unpacked;
2183 stream.avail_in = len;
2184 while (deflate(&stream, Z_FINISH) == Z_OK)
2186 deflateEnd(&stream);
2189 *objsize = stream.total_out;
2193 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2196 unsigned long objsize;
2197 void *buf = map_sha1_file(sha1, &objsize);
2200 retval = write_buffer(fd, buf, objsize);
2201 munmap(buf, objsize);
2205 buf = repack_object(sha1, &objsize);
2206 retval = write_buffer(fd, buf, objsize);
2211 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2212 size_t bufsize, size_t *bufposn)
2214 char tmpfile[PATH_MAX];
2217 unsigned char real_sha1[20];
2218 unsigned char discard[4096];
2222 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2224 local = mkstemp(tmpfile);
2227 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2229 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2232 memset(&stream, 0, sizeof(stream));
2234 inflateInit(&stream);
2241 stream.avail_in = *bufposn;
2242 stream.next_in = (unsigned char *) buffer;
2244 stream.next_out = discard;
2245 stream.avail_out = sizeof(discard);
2246 ret = inflate(&stream, Z_SYNC_FLUSH);
2247 SHA1_Update(&c, discard, sizeof(discard) -
2249 } while (stream.avail_in && ret == Z_OK);
2250 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2251 die("unable to write sha1 file");
2252 memmove(buffer, buffer + *bufposn - stream.avail_in,
2254 *bufposn = stream.avail_in;
2258 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2263 return error("Connection closed?");
2264 perror("Reading from connection");
2269 inflateEnd(&stream);
2271 fchmod(local, 0444);
2272 if (close(local) != 0)
2273 die("unable to write sha1 file");
2274 SHA1_Final(real_sha1, &c);
2275 if (ret != Z_STREAM_END) {
2277 return error("File %s corrupted", sha1_to_hex(sha1));
2279 if (hashcmp(sha1, real_sha1)) {
2281 return error("File %s has bad hash", sha1_to_hex(sha1));
2284 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2287 int has_pack_index(const unsigned char *sha1)
2290 if (stat(sha1_pack_index_name(sha1), &st))
2295 int has_pack_file(const unsigned char *sha1)
2298 if (stat(sha1_pack_name(sha1), &st))
2303 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2305 struct pack_entry e;
2306 return find_pack_entry(sha1, &e, ignore_packed);
2309 int has_sha1_file(const unsigned char *sha1)
2312 struct pack_entry e;
2314 if (find_pack_entry(sha1, &e, NULL))
2316 return find_sha1_file(sha1, &st) ? 1 : 0;
2319 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2324 strbuf_init(&buf, 0);
2325 if (strbuf_read(&buf, fd, 4096) < 0) {
2326 strbuf_release(&buf);
2333 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2335 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2336 strbuf_release(&buf);
2341 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2342 enum object_type type, const char *path)
2344 size_t size = xsize_t(st->st_size);
2346 int ret, re_allocated = 0;
2349 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2356 * Convert blobs to git internal format
2358 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2360 strbuf_init(&nbuf, 0);
2361 if (convert_to_git(path, buf, size, &nbuf,
2362 write_object ? safe_crlf : 0)) {
2364 buf = strbuf_detach(&nbuf, &size);
2370 ret = write_sha1_file(buf, size, typename(type), sha1);
2372 ret = hash_sha1_file(buf, size, typename(type), sha1);
2382 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2388 switch (st->st_mode & S_IFMT) {
2390 fd = open(path, O_RDONLY);
2392 return error("open(\"%s\"): %s", path,
2394 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2395 return error("%s: failed to insert into database",
2399 len = xsize_t(st->st_size);
2400 target = xmalloc(len + 1);
2401 if (readlink(path, target, len + 1) != st->st_size) {
2402 char *errstr = strerror(errno);
2404 return error("readlink(\"%s\"): %s", path,
2408 hash_sha1_file(target, len, blob_type, sha1);
2409 else if (write_sha1_file(target, len, blob_type, sha1))
2410 return error("%s: failed to insert into database",
2415 return resolve_gitlink_ref(path, "HEAD", sha1);
2417 return error("%s: unsupported file type", path);
2422 int read_pack_header(int fd, struct pack_header *header)
2424 char *c = (char*)header;
2425 ssize_t remaining = sizeof(struct pack_header);
2427 ssize_t r = xread(fd, c, remaining);
2429 /* "eof before pack header was fully read" */
2430 return PH_ERROR_EOF;
2433 } while (remaining > 0);
2434 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2435 /* "protocol error (pack signature mismatch detected)" */
2436 return PH_ERROR_PACK_SIGNATURE;
2437 if (!pack_version_ok(header->hdr_version))
2438 /* "protocol error (pack version unsupported)" */
2439 return PH_ERROR_PROTOCOL;