2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
33 static size_t sz_fmt(size_t s) { return s; }
36 const unsigned char null_sha1[20];
38 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
86 static inline int offset_1st_component(const char *path)
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
93 int safe_create_leading_directories(char *path)
95 char *pos = path + offset_1st_component(path);
99 pos = strchr(pos, '/');
103 if (!stat(path, &st)) {
105 if (!S_ISDIR(st.st_mode)) {
110 else if (mkdir(path, 0777)) {
114 else if (adjust_shared_perm(path)) {
123 int safe_create_leading_directories_const(const char *path)
125 /* path points to cache entries, so xstrdup before messing with it */
126 char *buf = xstrdup(path);
127 int result = safe_create_leading_directories(buf);
132 char *sha1_to_hex(const unsigned char *sha1)
135 static char hexbuffer[4][50];
136 static const char hex[] = "0123456789abcdef";
137 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
150 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
153 for (i = 0; i < 20; i++) {
154 static char hex[] = "0123456789abcdef";
155 unsigned int val = sha1[i];
156 char *pos = pathbuf + i*2 + (i > 0);
157 *pos++ = hex[val >> 4];
158 *pos = hex[val & 0xf];
163 * NOTE! This returns a statically allocated buffer, so you have to be
164 * careful about using it. Do an "xstrdup()" if you need to save the
167 * Also note that this returns the location for creating. Reading
168 * SHA1 file can happen from any alternate directory listed in the
169 * DB_ENVIRONMENT environment variable if it is not found in
170 * the primary object database.
172 char *sha1_file_name(const unsigned char *sha1)
174 static char *name, *base;
177 const char *sha1_file_directory = get_object_directory();
178 int len = strlen(sha1_file_directory);
179 base = xmalloc(len + 60);
180 memcpy(base, sha1_file_directory, len);
181 memset(base+len, 0, 60);
184 name = base + len + 1;
186 fill_sha1_path(name, sha1);
190 static char *sha1_get_pack_name(const unsigned char *sha1,
191 char **name, char **base, const char *which)
193 static const char hex[] = "0123456789abcdef";
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 *base = xmalloc(len + 60);
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
203 *name = *base + len + 11;
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
217 char *sha1_pack_name(const unsigned char *sha1)
219 static char *name, *base;
221 return sha1_get_pack_name(sha1, &name, &base, "pack");
224 char *sha1_pack_index_name(const unsigned char *sha1)
226 static char *name, *base;
228 return sha1_get_pack_name(sha1, &name, &base, "idx");
231 struct alternate_object_database *alt_odb_list;
232 static struct alternate_object_database **alt_odb_tail;
234 static void read_info_alternates(const char * alternates, int depth);
237 * Prepare alternate object database registry.
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
251 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
254 const char *objdir = get_object_directory();
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 /* 43 = 40-byte + 2 '/' + terminating NUL */
259 int entlen = pfxlen + 43;
262 if (!is_absolute_path(entry) && relative_base) {
263 /* Relative alt-odb */
265 base_len = strlen(relative_base) + 1;
269 ent = xmalloc(sizeof(*ent) + entlen);
271 if (!is_absolute_path(entry) && relative_base) {
272 memcpy(ent->base, relative_base, base_len - 1);
273 ent->base[base_len - 1] = '/';
274 memcpy(ent->base + base_len, entry, len);
277 memcpy(ent->base, entry, pfxlen);
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283 /* Detect cases where alternate disappeared */
284 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
301 if (!memcmp(ent->base, objdir, pfxlen)) {
306 /* add the alternate entry */
308 alt_odb_tail = &(ent->next);
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
314 ent->base[pfxlen] = '/';
319 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
322 const char *cp, *last;
325 error("%s: ignoring alternate object stores, nesting too deep.",
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
339 while (cp < ep && *cp != sep)
342 if (!is_absolute_path(last) && depth) {
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
350 while (cp < ep && *cp == sep)
356 static void read_info_alternates(const char * relative_base, int depth)
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = open(path, O_RDONLY);
372 if (fstat(fd, &st) || (st.st_size == 0)) {
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
385 void add_to_alternates_file(const char *reference)
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
389 char *alt = mkpath("%s/objects\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
397 void prepare_alt_odb(void)
404 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
407 alt_odb_tail = &alt_odb_list;
408 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
410 read_info_alternates(get_object_directory(), 0);
413 static int has_loose_object(const unsigned char *sha1)
415 char *name = sha1_file_name(sha1);
416 struct alternate_object_database *alt;
418 if (!access(name, F_OK))
421 for (alt = alt_odb_list; alt; alt = alt->next) {
423 fill_sha1_path(name, sha1);
424 if (!access(alt->base, F_OK))
430 static unsigned int pack_used_ctr;
431 static unsigned int pack_mmap_calls;
432 static unsigned int peak_pack_open_windows;
433 static unsigned int pack_open_windows;
434 static size_t peak_pack_mapped;
435 static size_t pack_mapped;
436 struct packed_git *packed_git;
438 void pack_report(void)
441 "pack_report: getpagesize() = %10" SZ_FMT "\n"
442 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
443 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
444 sz_fmt(getpagesize()),
445 sz_fmt(packed_git_window_size),
446 sz_fmt(packed_git_limit));
448 "pack_report: pack_used_ctr = %10u\n"
449 "pack_report: pack_mmap_calls = %10u\n"
450 "pack_report: pack_open_windows = %10u / %10u\n"
451 "pack_report: pack_mapped = "
452 "%10" SZ_FMT " / %10" SZ_FMT "\n",
455 pack_open_windows, peak_pack_open_windows,
456 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
459 static int check_packed_git_idx(const char *path, struct packed_git *p)
462 struct pack_idx_header *hdr;
464 uint32_t version, nr, i, *index;
465 int fd = open(path, O_RDONLY);
470 if (fstat(fd, &st)) {
474 idx_size = xsize_t(st.st_size);
475 if (idx_size < 4 * 256 + 20 + 20) {
477 return error("index file %s is too small", path);
479 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
483 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
484 version = ntohl(hdr->idx_version);
485 if (version < 2 || version > 2) {
486 munmap(idx_map, idx_size);
487 return error("index file %s is version %"PRIu32
488 " and is not supported by this binary"
489 " (try upgrading GIT to a newer version)",
498 index += 2; /* skip index header */
499 for (i = 0; i < 256; i++) {
500 uint32_t n = ntohl(index[i]);
502 munmap(idx_map, idx_size);
503 return error("non-monotonic index %s", path);
511 * - 256 index entries 4 bytes each
512 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
513 * - 20-byte SHA1 of the packfile
514 * - 20-byte SHA1 file checksum
516 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
517 munmap(idx_map, idx_size);
518 return error("wrong index v1 file size in %s", path);
520 } else if (version == 2) {
523 * - 8 bytes of header
524 * - 256 index entries 4 bytes each
525 * - 20-byte sha1 entry * nr
526 * - 4-byte crc entry * nr
527 * - 4-byte offset entry * nr
528 * - 20-byte SHA1 of the packfile
529 * - 20-byte SHA1 file checksum
530 * And after the 4-byte offset table might be a
531 * variable sized table containing 8-byte entries
532 * for offsets larger than 2^31.
534 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
535 unsigned long max_size = min_size;
537 max_size += (nr - 1)*8;
538 if (idx_size < min_size || idx_size > max_size) {
539 munmap(idx_map, idx_size);
540 return error("wrong index v2 file size in %s", path);
542 if (idx_size != min_size &&
544 * make sure we can deal with large pack offsets.
545 * 31-bit signed offset won't be enough, neither
546 * 32-bit unsigned one will be.
548 (sizeof(off_t) <= 4)) {
549 munmap(idx_map, idx_size);
550 return error("pack too large for current definition of off_t in %s", path);
554 p->index_version = version;
555 p->index_data = idx_map;
556 p->index_size = idx_size;
561 int open_pack_index(struct packed_git *p)
569 idx_name = xstrdup(p->pack_name);
570 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
571 ret = check_packed_git_idx(idx_name, p);
576 static void scan_windows(struct packed_git *p,
577 struct packed_git **lru_p,
578 struct pack_window **lru_w,
579 struct pack_window **lru_l)
581 struct pack_window *w, *w_l;
583 for (w_l = NULL, w = p->windows; w; w = w->next) {
585 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
595 static int unuse_one_window(struct packed_git *current, int keep_fd)
597 struct packed_git *p, *lru_p = NULL;
598 struct pack_window *lru_w = NULL, *lru_l = NULL;
601 scan_windows(current, &lru_p, &lru_w, &lru_l);
602 for (p = packed_git; p; p = p->next)
603 scan_windows(p, &lru_p, &lru_w, &lru_l);
605 munmap(lru_w->base, lru_w->len);
606 pack_mapped -= lru_w->len;
608 lru_l->next = lru_w->next;
610 lru_p->windows = lru_w->next;
611 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
612 close(lru_p->pack_fd);
623 void release_pack_memory(size_t need, int fd)
625 size_t cur = pack_mapped;
626 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
630 void close_pack_windows(struct packed_git *p)
633 struct pack_window *w = p->windows;
636 die("pack '%s' still has open windows to it",
638 munmap(w->base, w->len);
639 pack_mapped -= w->len;
641 p->windows = w->next;
646 void unuse_pack(struct pack_window **w_cursor)
648 struct pack_window *w = *w_cursor;
656 * Do not call this directly as this leaks p->pack_fd on error return;
657 * call open_packed_git() instead.
659 static int open_packed_git_1(struct packed_git *p)
662 struct pack_header hdr;
663 unsigned char sha1[20];
664 unsigned char *idx_sha1;
667 if (!p->index_data && open_pack_index(p))
668 return error("packfile %s index unavailable", p->pack_name);
670 p->pack_fd = open(p->pack_name, O_RDONLY);
671 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
674 /* If we created the struct before we had the pack we lack size. */
676 if (!S_ISREG(st.st_mode))
677 return error("packfile %s not a regular file", p->pack_name);
678 p->pack_size = st.st_size;
679 } else if (p->pack_size != st.st_size)
680 return error("packfile %s size changed", p->pack_name);
682 /* We leave these file descriptors open with sliding mmap;
683 * there is no point keeping them open across exec(), though.
685 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
687 return error("cannot determine file descriptor flags");
688 fd_flag |= FD_CLOEXEC;
689 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
690 return error("cannot set FD_CLOEXEC");
692 /* Verify we recognize this pack file format. */
693 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
694 return error("file %s is far too short to be a packfile", p->pack_name);
695 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
696 return error("file %s is not a GIT packfile", p->pack_name);
697 if (!pack_version_ok(hdr.hdr_version))
698 return error("packfile %s is version %"PRIu32" and not"
699 " supported (try upgrading GIT to a newer version)",
700 p->pack_name, ntohl(hdr.hdr_version));
702 /* Verify the pack matches its index. */
703 if (p->num_objects != ntohl(hdr.hdr_entries))
704 return error("packfile %s claims to have %"PRIu32" objects"
705 " while index indicates %"PRIu32" objects",
706 p->pack_name, ntohl(hdr.hdr_entries),
708 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
709 return error("end of packfile %s is unavailable", p->pack_name);
710 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
711 return error("packfile %s signature is unavailable", p->pack_name);
712 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
713 if (hashcmp(sha1, idx_sha1))
714 return error("packfile %s does not match index", p->pack_name);
718 static int open_packed_git(struct packed_git *p)
720 if (!open_packed_git_1(p))
722 if (p->pack_fd != -1) {
729 static int in_window(struct pack_window *win, off_t offset)
731 /* We must promise at least 20 bytes (one hash) after the
732 * offset is available from this window, otherwise the offset
733 * is not actually in this window and a different window (which
734 * has that one hash excess) must be used. This is to support
735 * the object header and delta base parsing routines below.
737 off_t win_off = win->offset;
738 return win_off <= offset
739 && (offset + 20) <= (win_off + win->len);
742 unsigned char* use_pack(struct packed_git *p,
743 struct pack_window **w_cursor,
747 struct pack_window *win = *w_cursor;
749 if (p->pack_fd == -1 && open_packed_git(p))
750 die("packfile %s cannot be accessed", p->pack_name);
752 /* Since packfiles end in a hash of their content and its
753 * pointless to ask for an offset into the middle of that
754 * hash, and the in_window function above wouldn't match
755 * don't allow an offset too close to the end of the file.
757 if (offset > (p->pack_size - 20))
758 die("offset beyond end of packfile (truncated pack?)");
760 if (!win || !in_window(win, offset)) {
763 for (win = p->windows; win; win = win->next) {
764 if (in_window(win, offset))
768 size_t window_align = packed_git_window_size / 2;
770 win = xcalloc(1, sizeof(*win));
771 win->offset = (offset / window_align) * window_align;
772 len = p->pack_size - win->offset;
773 if (len > packed_git_window_size)
774 len = packed_git_window_size;
775 win->len = (size_t)len;
776 pack_mapped += win->len;
777 while (packed_git_limit < pack_mapped
778 && unuse_one_window(p, p->pack_fd))
780 win->base = xmmap(NULL, win->len,
781 PROT_READ, MAP_PRIVATE,
782 p->pack_fd, win->offset);
783 if (win->base == MAP_FAILED)
784 die("packfile %s cannot be mapped: %s",
789 if (pack_mapped > peak_pack_mapped)
790 peak_pack_mapped = pack_mapped;
791 if (pack_open_windows > peak_pack_open_windows)
792 peak_pack_open_windows = pack_open_windows;
793 win->next = p->windows;
797 if (win != *w_cursor) {
798 win->last_used = pack_used_ctr++;
802 offset -= win->offset;
804 *left = win->len - xsize_t(offset);
805 return win->base + offset;
808 static struct packed_git *alloc_packed_git(int extra)
810 struct packed_git *p = xmalloc(sizeof(*p) + extra);
811 memset(p, 0, sizeof(*p));
816 struct packed_git *add_packed_git(const char *path, int path_len, int local)
819 struct packed_git *p = alloc_packed_git(path_len + 2);
822 * Make sure a corresponding .pack file exists and that
823 * the index looks sane.
825 path_len -= strlen(".idx");
830 memcpy(p->pack_name, path, path_len);
831 strcpy(p->pack_name + path_len, ".pack");
832 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
837 /* ok, it looks sane as far as we can check without
838 * actually mapping the pack file.
840 p->pack_size = st.st_size;
841 p->pack_local = local;
842 p->mtime = st.st_mtime;
843 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
848 struct packed_git *parse_pack_index(unsigned char *sha1)
850 const char *idx_path = sha1_pack_index_name(sha1);
851 const char *path = sha1_pack_name(sha1);
852 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
854 strcpy(p->pack_name, path);
855 hashcpy(p->sha1, sha1);
856 if (check_packed_git_idx(idx_path, p)) {
864 void install_packed_git(struct packed_git *pack)
866 pack->next = packed_git;
870 static void prepare_packed_git_one(char *objdir, int local)
872 /* Ensure that this buffer is large enough so that we can
873 append "/pack/" without clobbering the stack even if
874 strlen(objdir) were PATH_MAX. */
875 char path[PATH_MAX + 1 + 4 + 1 + 1];
880 sprintf(path, "%s/pack", objdir);
885 error("unable to open object pack directory: %s: %s",
886 path, strerror(errno));
890 while ((de = readdir(dir)) != NULL) {
891 int namelen = strlen(de->d_name);
892 struct packed_git *p;
894 if (!has_extension(de->d_name, ".idx"))
897 if (len + namelen + 1 > sizeof(path))
900 /* Don't reopen a pack we already have. */
901 strcpy(path + len, de->d_name);
902 for (p = packed_git; p; p = p->next) {
903 if (!memcmp(path, p->pack_name, len + namelen - 4))
908 /* See if it really is a valid .idx file with corresponding
909 * .pack file that we can map.
911 p = add_packed_git(path, len + namelen, local);
914 install_packed_git(p);
919 static int sort_pack(const void *a_, const void *b_)
921 struct packed_git *a = *((struct packed_git **)a_);
922 struct packed_git *b = *((struct packed_git **)b_);
926 * Local packs tend to contain objects specific to our
927 * variant of the project than remote ones. In addition,
928 * remote ones could be on a network mounted filesystem.
929 * Favor local ones for these reasons.
931 st = a->pack_local - b->pack_local;
936 * Younger packs tend to contain more recent objects,
937 * and more recent objects tend to get accessed more
940 if (a->mtime < b->mtime)
942 else if (a->mtime == b->mtime)
947 static void rearrange_packed_git(void)
949 struct packed_git **ary, *p;
952 for (n = 0, p = packed_git; p; p = p->next)
957 /* prepare an array of packed_git for easier sorting */
958 ary = xcalloc(n, sizeof(struct packed_git *));
959 for (n = 0, p = packed_git; p; p = p->next)
962 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
964 /* link them back again */
965 for (i = 0; i < n - 1; i++)
966 ary[i]->next = ary[i + 1];
967 ary[n - 1]->next = NULL;
973 static int prepare_packed_git_run_once = 0;
974 void prepare_packed_git(void)
976 struct alternate_object_database *alt;
978 if (prepare_packed_git_run_once)
980 prepare_packed_git_one(get_object_directory(), 1);
982 for (alt = alt_odb_list; alt; alt = alt->next) {
984 prepare_packed_git_one(alt->base, 0);
987 rearrange_packed_git();
988 prepare_packed_git_run_once = 1;
991 void reprepare_packed_git(void)
993 prepare_packed_git_run_once = 0;
994 prepare_packed_git();
997 static void mark_bad_packed_object(struct packed_git *p,
998 const unsigned char *sha1)
1001 for (i = 0; i < p->num_bad_objects; i++)
1002 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1004 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1005 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1006 p->num_bad_objects++;
1009 static int has_packed_and_bad(const unsigned char *sha1)
1011 struct packed_git *p;
1014 for (p = packed_git; p; p = p->next)
1015 for (i = 0; i < p->num_bad_objects; i++)
1016 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1021 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1023 unsigned char real_sha1[20];
1024 hash_sha1_file(map, size, type, real_sha1);
1025 return hashcmp(sha1, real_sha1) ? -1 : 0;
1028 static int git_open_noatime(const char *name)
1030 static int sha1_file_open_flag = O_NOATIME;
1031 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1033 /* Might the failure be due to O_NOATIME? */
1034 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1035 fd = open(name, O_RDONLY);
1037 sha1_file_open_flag = 0;
1042 static int open_sha1_file(const unsigned char *sha1)
1045 char *name = sha1_file_name(sha1);
1046 struct alternate_object_database *alt;
1048 fd = git_open_noatime(name);
1054 for (alt = alt_odb_list; alt; alt = alt->next) {
1056 fill_sha1_path(name, sha1);
1057 fd = git_open_noatime(alt->base);
1064 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1069 fd = open_sha1_file(sha1);
1074 if (!fstat(fd, &st)) {
1075 *size = xsize_t(st.st_size);
1076 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1083 static int legacy_loose_object(unsigned char *map)
1088 * Is it a zlib-compressed buffer? If so, the first byte
1089 * must be 0x78 (15-bit window size, deflated), and the
1090 * first 16-bit word is evenly divisible by 31
1092 word = (map[0] << 8) + map[1];
1093 if (map[0] == 0x78 && !(word % 31))
1099 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1104 unsigned long used = 0;
1107 *type = (c >> 4) & 7;
1113 if (sizeof(long) * 8 <= shift)
1116 size += (c & 0x7f) << shift;
1123 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1125 unsigned long size, used;
1126 static const char valid_loose_object_type[8] = {
1128 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1129 0, /* "delta" and others are invalid in a loose object */
1131 enum object_type type;
1133 /* Get the data stream */
1134 memset(stream, 0, sizeof(*stream));
1135 stream->next_in = map;
1136 stream->avail_in = mapsize;
1137 stream->next_out = buffer;
1138 stream->avail_out = bufsiz;
1140 if (legacy_loose_object(map)) {
1141 inflateInit(stream);
1142 return inflate(stream, 0);
1147 * There used to be a second loose object header format which
1148 * was meant to mimic the in-pack format, allowing for direct
1149 * copy of the object data. This format turned up not to be
1150 * really worth it and we don't write it any longer. But we
1151 * can still read it.
1153 used = unpack_object_header_gently(map, mapsize, &type, &size);
1154 if (!used || !valid_loose_object_type[type])
1159 /* Set up the stream for the rest.. */
1160 stream->next_in = map;
1161 stream->avail_in = mapsize;
1162 inflateInit(stream);
1164 /* And generate the fake traditional header */
1165 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1166 typename(type), size);
1170 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1172 int bytes = strlen(buffer) + 1;
1173 unsigned char *buf = xmalloc(1+size);
1177 n = stream->total_out - bytes;
1180 memcpy(buf, (char *) buffer + bytes, n);
1182 if (bytes <= size) {
1184 * The above condition must be (bytes <= size), not
1185 * (bytes < size). In other words, even though we
1186 * expect no more output and set avail_out to zer0,
1187 * the input zlib stream may have bytes that express
1188 * "this concludes the stream", and we *do* want to
1191 * Otherwise we would not be able to test that we
1192 * consumed all the input to reach the expected size;
1193 * we also want to check that zlib tells us that all
1194 * went well with status == Z_STREAM_END at the end.
1196 stream->next_out = buf + bytes;
1197 stream->avail_out = size - bytes;
1198 while (status == Z_OK)
1199 status = inflate(stream, Z_FINISH);
1202 if (status == Z_STREAM_END && !stream->avail_in) {
1208 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1209 else if (stream->avail_in)
1210 error("garbage at end of loose object '%s'",
1217 * We used to just use "sscanf()", but that's actually way
1218 * too permissive for what we want to check. So do an anal
1219 * object header parse by hand.
1221 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1228 * The type can be at most ten bytes (including the
1229 * terminating '\0' that we add), and is followed by
1238 if (i >= sizeof(type))
1244 * The length must follow immediately, and be in canonical
1245 * decimal format (ie "010" is not valid).
1247 size = *hdr++ - '0';
1252 unsigned long c = *hdr - '0';
1256 size = size * 10 + c;
1262 * The length must be followed by a zero byte
1264 return *hdr ? -1 : type_from_string(type);
1267 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1273 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1274 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1277 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1280 unsigned long get_size_from_delta(struct packed_git *p,
1281 struct pack_window **w_curs,
1284 const unsigned char *data;
1285 unsigned char delta_head[20], *in;
1289 memset(&stream, 0, sizeof(stream));
1290 stream.next_out = delta_head;
1291 stream.avail_out = sizeof(delta_head);
1293 inflateInit(&stream);
1295 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1296 stream.next_in = in;
1297 st = inflate(&stream, Z_FINISH);
1298 curpos += stream.next_in - in;
1299 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1300 stream.total_out < sizeof(delta_head));
1301 inflateEnd(&stream);
1302 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1303 die("delta data unpack-initial failed");
1305 /* Examine the initial part of the delta to figure out
1310 /* ignore base size */
1311 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1313 /* Read the result size */
1314 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1317 static off_t get_delta_base(struct packed_git *p,
1318 struct pack_window **w_curs,
1320 enum object_type type,
1321 off_t delta_obj_offset)
1323 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1326 /* use_pack() assured us we have [base_info, base_info + 20)
1327 * as a range that we can look at without walking off the
1328 * end of the mapped window. Its actually the hash size
1329 * that is assured. An OFS_DELTA longer than the hash size
1330 * is stupid, as then a REF_DELTA would be smaller to store.
1332 if (type == OBJ_OFS_DELTA) {
1334 unsigned char c = base_info[used++];
1335 base_offset = c & 127;
1338 if (!base_offset || MSB(base_offset, 7))
1339 return 0; /* overflow */
1340 c = base_info[used++];
1341 base_offset = (base_offset << 7) + (c & 127);
1343 base_offset = delta_obj_offset - base_offset;
1344 if (base_offset >= delta_obj_offset)
1345 return 0; /* out of bound */
1347 } else if (type == OBJ_REF_DELTA) {
1348 /* The base entry _must_ be in the same pack */
1349 base_offset = find_pack_entry_one(base_info, p);
1352 die("I am totally screwed");
1356 /* forward declaration for a mutually recursive function */
1357 static int packed_object_info(struct packed_git *p, off_t offset,
1358 unsigned long *sizep);
1360 static int packed_delta_info(struct packed_git *p,
1361 struct pack_window **w_curs,
1363 enum object_type type,
1365 unsigned long *sizep)
1369 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1370 type = packed_object_info(p, base_offset, NULL);
1372 /* We choose to only get the type of the base object and
1373 * ignore potentially corrupt pack file that expects the delta
1374 * based on a base with a wrong size. This saves tons of
1378 *sizep = get_size_from_delta(p, w_curs, curpos);
1383 static int unpack_object_header(struct packed_git *p,
1384 struct pack_window **w_curs,
1386 unsigned long *sizep)
1388 unsigned char *base;
1391 enum object_type type;
1393 /* use_pack() assures us we have [base, base + 20) available
1394 * as a range that we can look at at. (Its actually the hash
1395 * size that is assured.) With our object header encoding
1396 * the maximum deflated object size is 2^137, which is just
1397 * insane, so we know won't exceed what we have been given.
1399 base = use_pack(p, w_curs, *curpos, &left);
1400 used = unpack_object_header_gently(base, left, &type, sizep);
1402 die("object offset outside of pack file");
1408 const char *packed_object_info_detail(struct packed_git *p,
1410 unsigned long *size,
1411 unsigned long *store_size,
1412 unsigned int *delta_chain_length,
1413 unsigned char *base_sha1)
1415 struct pack_window *w_curs = NULL;
1417 unsigned long dummy;
1418 unsigned char *next_sha1;
1419 enum object_type type;
1420 struct revindex_entry *revidx;
1422 *delta_chain_length = 0;
1423 curpos = obj_offset;
1424 type = unpack_object_header(p, &w_curs, &curpos, size);
1426 revidx = find_pack_revindex(p, obj_offset);
1427 *store_size = revidx[1].offset - obj_offset;
1432 die("pack %s contains unknown object type %d",
1433 p->pack_name, type);
1438 unuse_pack(&w_curs);
1439 return typename(type);
1441 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1443 die("pack %s contains bad delta base reference of type %s",
1444 p->pack_name, typename(type));
1445 if (*delta_chain_length == 0) {
1446 revidx = find_pack_revindex(p, obj_offset);
1447 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1451 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1452 if (*delta_chain_length == 0)
1453 hashcpy(base_sha1, next_sha1);
1454 obj_offset = find_pack_entry_one(next_sha1, p);
1457 (*delta_chain_length)++;
1458 curpos = obj_offset;
1459 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1463 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1464 unsigned long *sizep)
1466 struct pack_window *w_curs = NULL;
1468 off_t curpos = obj_offset;
1469 enum object_type type;
1471 type = unpack_object_header(p, &w_curs, &curpos, &size);
1476 type = packed_delta_info(p, &w_curs, curpos,
1477 type, obj_offset, sizep);
1487 die("pack %s contains unknown object type %d",
1488 p->pack_name, type);
1490 unuse_pack(&w_curs);
1494 static void *unpack_compressed_entry(struct packed_git *p,
1495 struct pack_window **w_curs,
1501 unsigned char *buffer, *in;
1503 buffer = xmalloc(size + 1);
1505 memset(&stream, 0, sizeof(stream));
1506 stream.next_out = buffer;
1507 stream.avail_out = size;
1509 inflateInit(&stream);
1511 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1512 stream.next_in = in;
1513 st = inflate(&stream, Z_FINISH);
1514 curpos += stream.next_in - in;
1515 } while (st == Z_OK || st == Z_BUF_ERROR);
1516 inflateEnd(&stream);
1517 if ((st != Z_STREAM_END) || stream.total_out != size) {
1525 #define MAX_DELTA_CACHE (256)
1527 static size_t delta_base_cached;
1529 static struct delta_base_cache_lru_list {
1530 struct delta_base_cache_lru_list *prev;
1531 struct delta_base_cache_lru_list *next;
1532 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1534 static struct delta_base_cache_entry {
1535 struct delta_base_cache_lru_list lru;
1537 struct packed_git *p;
1540 enum object_type type;
1541 } delta_base_cache[MAX_DELTA_CACHE];
1543 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1547 hash = (unsigned long)p + (unsigned long)base_offset;
1548 hash += (hash >> 8) + (hash >> 16);
1549 return hash % MAX_DELTA_CACHE;
1552 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1553 unsigned long *base_size, enum object_type *type, int keep_cache)
1556 unsigned long hash = pack_entry_hash(p, base_offset);
1557 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1560 if (ret && ent->p == p && ent->base_offset == base_offset)
1561 goto found_cache_entry;
1562 return unpack_entry(p, base_offset, type, base_size);
1567 ent->lru.next->prev = ent->lru.prev;
1568 ent->lru.prev->next = ent->lru.next;
1569 delta_base_cached -= ent->size;
1571 ret = xmemdupz(ent->data, ent->size);
1574 *base_size = ent->size;
1578 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1583 ent->lru.next->prev = ent->lru.prev;
1584 ent->lru.prev->next = ent->lru.next;
1585 delta_base_cached -= ent->size;
1589 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1590 void *base, unsigned long base_size, enum object_type type)
1592 unsigned long hash = pack_entry_hash(p, base_offset);
1593 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1594 struct delta_base_cache_lru_list *lru;
1596 release_delta_base_cache(ent);
1597 delta_base_cached += base_size;
1599 for (lru = delta_base_cache_lru.next;
1600 delta_base_cached > delta_base_cache_limit
1601 && lru != &delta_base_cache_lru;
1603 struct delta_base_cache_entry *f = (void *)lru;
1604 if (f->type == OBJ_BLOB)
1605 release_delta_base_cache(f);
1607 for (lru = delta_base_cache_lru.next;
1608 delta_base_cached > delta_base_cache_limit
1609 && lru != &delta_base_cache_lru;
1611 struct delta_base_cache_entry *f = (void *)lru;
1612 release_delta_base_cache(f);
1616 ent->base_offset = base_offset;
1619 ent->size = base_size;
1620 ent->lru.next = &delta_base_cache_lru;
1621 ent->lru.prev = delta_base_cache_lru.prev;
1622 delta_base_cache_lru.prev->next = &ent->lru;
1623 delta_base_cache_lru.prev = &ent->lru;
1626 static void *unpack_delta_entry(struct packed_git *p,
1627 struct pack_window **w_curs,
1629 unsigned long delta_size,
1631 enum object_type *type,
1632 unsigned long *sizep)
1634 void *delta_data, *result, *base;
1635 unsigned long base_size;
1638 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1640 error("failed to validate delta base reference "
1641 "at offset %"PRIuMAX" from %s",
1642 (uintmax_t)curpos, p->pack_name);
1646 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1649 * We're probably in deep shit, but let's try to fetch
1650 * the required base anyway from another pack or loose.
1651 * This is costly but should happen only in the presence
1652 * of a corrupted pack, and is better than failing outright.
1654 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1655 const unsigned char *base_sha1 =
1656 nth_packed_object_sha1(p, revidx->nr);
1657 error("failed to read delta base object %s"
1658 " at offset %"PRIuMAX" from %s",
1659 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1661 mark_bad_packed_object(p, base_sha1);
1662 base = read_object(base_sha1, type, &base_size);
1667 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1669 error("failed to unpack compressed delta "
1670 "at offset %"PRIuMAX" from %s",
1671 (uintmax_t)curpos, p->pack_name);
1675 result = patch_delta(base, base_size,
1676 delta_data, delta_size,
1679 die("failed to apply delta");
1681 add_delta_base_cache(p, base_offset, base, base_size, *type);
1685 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1686 enum object_type *type, unsigned long *sizep)
1688 struct pack_window *w_curs = NULL;
1689 off_t curpos = obj_offset;
1692 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1696 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1697 obj_offset, type, sizep);
1703 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1707 error("unknown object type %i at offset %"PRIuMAX" in %s",
1708 *type, (uintmax_t)obj_offset, p->pack_name);
1710 unuse_pack(&w_curs);
1714 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1717 const unsigned char *index = p->index_data;
1719 if (open_pack_index(p))
1721 index = p->index_data;
1723 if (n >= p->num_objects)
1726 if (p->index_version == 1) {
1727 return index + 24 * n + 4;
1730 return index + 20 * n;
1734 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1736 const unsigned char *index = p->index_data;
1738 if (p->index_version == 1) {
1739 return ntohl(*((uint32_t *)(index + 24 * n)));
1742 index += 8 + p->num_objects * (20 + 4);
1743 off = ntohl(*((uint32_t *)(index + 4 * n)));
1744 if (!(off & 0x80000000))
1746 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1747 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1748 ntohl(*((uint32_t *)(index + 4)));
1752 off_t find_pack_entry_one(const unsigned char *sha1,
1753 struct packed_git *p)
1755 const uint32_t *level1_ofs = p->index_data;
1756 const unsigned char *index = p->index_data;
1757 unsigned hi, lo, stride;
1758 static int use_lookup = -1;
1759 static int debug_lookup = -1;
1761 if (debug_lookup < 0)
1762 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1765 if (open_pack_index(p))
1767 level1_ofs = p->index_data;
1768 index = p->index_data;
1770 if (p->index_version > 1) {
1775 hi = ntohl(level1_ofs[*sha1]);
1776 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1777 if (p->index_version > 1) {
1785 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1786 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1789 use_lookup = !!getenv("GIT_USE_LOOKUP");
1791 int pos = sha1_entry_pos(index, stride, 0,
1792 lo, hi, p->num_objects, sha1);
1795 return nth_packed_object_offset(p, pos);
1799 unsigned mi = (lo + hi) / 2;
1800 int cmp = hashcmp(index + mi * stride, sha1);
1803 printf("lo %u hi %u rg %u mi %u\n",
1804 lo, hi, hi - lo, mi);
1806 return nth_packed_object_offset(p, mi);
1815 int matches_pack_name(struct packed_git *p, const char *name)
1817 const char *last_c, *c;
1819 if (!strcmp(p->pack_name, name))
1822 for (c = p->pack_name, last_c = c; *c;)
1827 if (!strcmp(last_c, name))
1833 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1835 static struct packed_git *last_found = (void *)1;
1836 struct packed_git *p;
1839 prepare_packed_git();
1842 p = (last_found == (void *)1) ? packed_git : last_found;
1845 if (ignore_packed) {
1847 for (ig = ignore_packed; *ig; ig++)
1848 if (matches_pack_name(p, *ig))
1854 if (p->num_bad_objects) {
1856 for (i = 0; i < p->num_bad_objects; i++)
1857 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1861 offset = find_pack_entry_one(sha1, p);
1864 * We are about to tell the caller where they can
1865 * locate the requested object. We better make
1866 * sure the packfile is still here and can be
1867 * accessed before supplying that answer, as
1868 * it may have been deleted since the index
1871 if (p->pack_fd == -1 && open_packed_git(p)) {
1872 error("packfile %s cannot be accessed", p->pack_name);
1877 hashcpy(e->sha1, sha1);
1883 if (p == last_found)
1887 if (p == last_found)
1893 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1894 struct packed_git *packs)
1896 struct packed_git *p;
1898 for (p = packs; p; p = p->next) {
1899 if (find_pack_entry_one(sha1, p))
1906 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1909 unsigned long mapsize, size;
1914 map = map_sha1_file(sha1, &mapsize);
1916 return error("unable to find %s", sha1_to_hex(sha1));
1917 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1918 status = error("unable to unpack %s header",
1920 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1921 status = error("unable to parse %s header", sha1_to_hex(sha1));
1924 inflateEnd(&stream);
1925 munmap(map, mapsize);
1929 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1931 struct pack_entry e;
1934 if (!find_pack_entry(sha1, &e, NULL)) {
1935 /* Most likely it's a loose object. */
1936 status = sha1_loose_object_info(sha1, sizep);
1940 /* Not a loose object; someone else may have just packed it. */
1941 reprepare_packed_git();
1942 if (!find_pack_entry(sha1, &e, NULL))
1945 return packed_object_info(e.p, e.offset, sizep);
1948 static void *read_packed_sha1(const unsigned char *sha1,
1949 enum object_type *type, unsigned long *size)
1951 struct pack_entry e;
1954 if (!find_pack_entry(sha1, &e, NULL))
1956 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1959 * We're probably in deep shit, but let's try to fetch
1960 * the required object anyway from another pack or loose.
1961 * This should happen only in the presence of a corrupted
1962 * pack, and is better than failing outright.
1964 error("failed to read object %s at offset %"PRIuMAX" from %s",
1965 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1966 mark_bad_packed_object(e.p, sha1);
1967 data = read_object(sha1, type, size);
1973 * This is meant to hold a *small* number of objects that you would
1974 * want read_sha1_file() to be able to return, but yet you do not want
1975 * to write them into the object store (e.g. a browse-only
1978 static struct cached_object {
1979 unsigned char sha1[20];
1980 enum object_type type;
1984 static int cached_object_nr, cached_object_alloc;
1986 static struct cached_object empty_tree = {
1987 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1988 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1989 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1995 static struct cached_object *find_cached_object(const unsigned char *sha1)
1998 struct cached_object *co = cached_objects;
2000 for (i = 0; i < cached_object_nr; i++, co++) {
2001 if (!hashcmp(co->sha1, sha1))
2004 if (!hashcmp(sha1, empty_tree.sha1))
2009 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2010 unsigned char *sha1)
2012 struct cached_object *co;
2014 hash_sha1_file(buf, len, typename(type), sha1);
2015 if (has_sha1_file(sha1) || find_cached_object(sha1))
2017 if (cached_object_alloc <= cached_object_nr) {
2018 cached_object_alloc = alloc_nr(cached_object_alloc);
2019 cached_objects = xrealloc(cached_objects,
2020 sizeof(*cached_objects) *
2021 cached_object_alloc);
2023 co = &cached_objects[cached_object_nr++];
2026 co->buf = xmalloc(len);
2027 memcpy(co->buf, buf, len);
2028 hashcpy(co->sha1, sha1);
2032 void *read_object(const unsigned char *sha1, enum object_type *type,
2033 unsigned long *size)
2035 unsigned long mapsize;
2037 struct cached_object *co;
2039 co = find_cached_object(sha1);
2043 return xmemdupz(co->buf, co->size);
2046 buf = read_packed_sha1(sha1, type, size);
2049 map = map_sha1_file(sha1, &mapsize);
2051 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2052 munmap(map, mapsize);
2055 reprepare_packed_git();
2056 return read_packed_sha1(sha1, type, size);
2059 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2060 unsigned long *size)
2062 void *data = read_object(sha1, type, size);
2063 /* legacy behavior is to die on corrupted objects */
2064 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2065 die("object %s is corrupted", sha1_to_hex(sha1));
2069 void *read_object_with_reference(const unsigned char *sha1,
2070 const char *required_type_name,
2071 unsigned long *size,
2072 unsigned char *actual_sha1_return)
2074 enum object_type type, required_type;
2076 unsigned long isize;
2077 unsigned char actual_sha1[20];
2079 required_type = type_from_string(required_type_name);
2080 hashcpy(actual_sha1, sha1);
2082 int ref_length = -1;
2083 const char *ref_type = NULL;
2085 buffer = read_sha1_file(actual_sha1, &type, &isize);
2088 if (type == required_type) {
2090 if (actual_sha1_return)
2091 hashcpy(actual_sha1_return, actual_sha1);
2094 /* Handle references */
2095 else if (type == OBJ_COMMIT)
2097 else if (type == OBJ_TAG)
2098 ref_type = "object ";
2103 ref_length = strlen(ref_type);
2105 if (ref_length + 40 > isize ||
2106 memcmp(buffer, ref_type, ref_length) ||
2107 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2112 /* Now we have the ID of the referred-to object in
2113 * actual_sha1. Check again. */
2117 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2118 const char *type, unsigned char *sha1,
2119 char *hdr, int *hdrlen)
2123 /* Generate the header */
2124 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2128 SHA1_Update(&c, hdr, *hdrlen);
2129 SHA1_Update(&c, buf, len);
2130 SHA1_Final(sha1, &c);
2134 * Move the just written object into its final resting place
2136 int move_temp_to_file(const char *tmpfile, const char *filename)
2138 int ret = link(tmpfile, filename);
2141 * Coda hack - coda doesn't like cross-directory links,
2142 * so we fall back to a rename, which will mean that it
2143 * won't be able to check collisions, but that's not a
2146 * The same holds for FAT formatted media.
2148 * When this succeeds, we just return 0. We have nothing
2151 if (ret && ret != EEXIST) {
2152 if (!rename(tmpfile, filename))
2158 if (ret != EEXIST) {
2159 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2161 /* FIXME!!! Collision check here ? */
2167 static int write_buffer(int fd, const void *buf, size_t len)
2169 if (write_in_full(fd, buf, len) < 0)
2170 return error("file write error (%s)", strerror(errno));
2174 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2175 unsigned char *sha1)
2179 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2183 /* Finalize a file on disk, and close it. */
2184 static void close_sha1_file(int fd)
2186 if (fsync_object_files)
2187 fsync_or_die(fd, "sha1 file");
2190 die("unable to write sha1 file");
2193 /* Size of directory component, including the ending '/' */
2194 static inline int directory_size(const char *filename)
2196 const char *s = strrchr(filename, '/');
2199 return s - filename + 1;
2203 * This creates a temporary file in the same directory as the final
2206 * We want to avoid cross-directory filename renames, because those
2207 * can have problems on various filesystems (FAT, NFS, Coda).
2209 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2211 int fd, dirlen = directory_size(filename);
2213 if (dirlen + 20 > bufsiz) {
2214 errno = ENAMETOOLONG;
2217 memcpy(buffer, filename, dirlen);
2218 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2219 fd = mkstemp(buffer);
2220 if (fd < 0 && dirlen) {
2221 /* Make sure the directory exists */
2222 memcpy(buffer, filename, dirlen);
2223 buffer[dirlen-1] = 0;
2224 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2228 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2229 fd = mkstemp(buffer);
2234 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2235 void *buf, unsigned long len, time_t mtime)
2238 unsigned char *compressed;
2241 static char tmpfile[PATH_MAX];
2243 filename = sha1_file_name(sha1);
2244 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2247 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2249 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2253 memset(&stream, 0, sizeof(stream));
2254 deflateInit(&stream, zlib_compression_level);
2255 size = 8 + deflateBound(&stream, len+hdrlen);
2256 compressed = xmalloc(size);
2259 stream.next_out = compressed;
2260 stream.avail_out = size;
2262 /* First header.. */
2263 stream.next_in = (unsigned char *)hdr;
2264 stream.avail_in = hdrlen;
2265 while (deflate(&stream, 0) == Z_OK)
2268 /* Then the data itself.. */
2269 stream.next_in = buf;
2270 stream.avail_in = len;
2271 ret = deflate(&stream, Z_FINISH);
2272 if (ret != Z_STREAM_END)
2273 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2275 ret = deflateEnd(&stream);
2277 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2279 size = stream.total_out;
2281 if (write_buffer(fd, compressed, size) < 0)
2282 die("unable to write sha1 file");
2283 close_sha1_file(fd);
2289 utb.modtime = mtime;
2290 if (utime(tmpfile, &utb) < 0)
2291 warning("failed utime() on %s: %s",
2292 tmpfile, strerror(errno));
2295 return move_temp_to_file(tmpfile, filename);
2298 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2300 unsigned char sha1[20];
2304 /* Normally if we have it in the pack then we do not bother writing
2305 * it out into .git/objects/??/?{38} file.
2307 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2309 hashcpy(returnsha1, sha1);
2310 if (has_sha1_file(sha1))
2312 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2315 int force_object_loose(const unsigned char *sha1, time_t mtime)
2319 enum object_type type;
2323 if (has_loose_object(sha1))
2325 buf = read_packed_sha1(sha1, &type, &len);
2327 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2328 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2329 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2332 int has_pack_index(const unsigned char *sha1)
2335 if (stat(sha1_pack_index_name(sha1), &st))
2340 int has_pack_file(const unsigned char *sha1)
2343 if (stat(sha1_pack_name(sha1), &st))
2348 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2350 struct pack_entry e;
2351 return find_pack_entry(sha1, &e, ignore_packed);
2354 int has_sha1_file(const unsigned char *sha1)
2356 struct pack_entry e;
2358 if (find_pack_entry(sha1, &e, NULL))
2360 return has_loose_object(sha1);
2363 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2364 int write_object, enum object_type type, const char *path)
2366 int ret, re_allocated = 0;
2372 * Convert blobs to git internal format
2374 if ((type == OBJ_BLOB) && path) {
2376 strbuf_init(&nbuf, 0);
2377 if (convert_to_git(path, buf, size, &nbuf,
2378 write_object ? safe_crlf : 0)) {
2379 buf = strbuf_detach(&nbuf, &size);
2385 ret = write_sha1_file(buf, size, typename(type), sha1);
2387 ret = hash_sha1_file(buf, size, typename(type), sha1);
2393 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2394 enum object_type type, const char *path)
2397 size_t size = xsize_t(st->st_size);
2399 if (!S_ISREG(st->st_mode)) {
2401 strbuf_init(&sbuf, 0);
2402 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2403 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2407 strbuf_release(&sbuf);
2409 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2410 ret = index_mem(sha1, buf, size, write_object, type, path);
2413 ret = index_mem(sha1, NULL, size, write_object, type, path);
2418 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2424 switch (st->st_mode & S_IFMT) {
2426 fd = open(path, O_RDONLY);
2428 return error("open(\"%s\"): %s", path,
2430 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2431 return error("%s: failed to insert into database",
2435 len = xsize_t(st->st_size);
2436 target = xmalloc(len + 1);
2437 if (readlink(path, target, len + 1) != st->st_size) {
2438 char *errstr = strerror(errno);
2440 return error("readlink(\"%s\"): %s", path,
2444 hash_sha1_file(target, len, blob_type, sha1);
2445 else if (write_sha1_file(target, len, blob_type, sha1))
2446 return error("%s: failed to insert into database",
2451 return resolve_gitlink_ref(path, "HEAD", sha1);
2453 return error("%s: unsupported file type", path);
2458 int read_pack_header(int fd, struct pack_header *header)
2460 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2461 /* "eof before pack header was fully read" */
2462 return PH_ERROR_EOF;
2464 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2465 /* "protocol error (pack signature mismatch detected)" */
2466 return PH_ERROR_PACK_SIGNATURE;
2467 if (!pack_version_ok(header->hdr_version))
2468 /* "protocol error (pack version unsupported)" */
2469 return PH_ERROR_PROTOCOL;