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"
23 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
24 #define O_NOATIME 01000000
32 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
35 static size_t sz_fmt(size_t s) { return s; }
38 const unsigned char null_sha1[20];
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
88 static inline int offset_1st_component(const char *path)
90 if (has_dos_drive_prefix(path))
91 return 2 + (path[2] == '/');
95 int safe_create_leading_directories(char *path)
97 char *pos = path + offset_1st_component(path);
101 pos = strchr(pos, '/');
105 if (!stat(path, &st)) {
107 if (!S_ISDIR(st.st_mode)) {
112 else if (mkdir(path, 0777)) {
116 else if (adjust_shared_perm(path)) {
125 int safe_create_leading_directories_const(const char *path)
127 /* path points to cache entries, so xstrdup before messing with it */
128 char *buf = xstrdup(path);
129 int result = safe_create_leading_directories(buf);
134 char *sha1_to_hex(const unsigned char *sha1)
137 static char hexbuffer[4][50];
138 static const char hex[] = "0123456789abcdef";
139 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
142 for (i = 0; i < 20; i++) {
143 unsigned int val = *sha1++;
144 *buf++ = hex[val >> 4];
145 *buf++ = hex[val & 0xf];
152 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155 for (i = 0; i < 20; i++) {
156 static char hex[] = "0123456789abcdef";
157 unsigned int val = sha1[i];
158 char *pos = pathbuf + i*2 + (i > 0);
159 *pos++ = hex[val >> 4];
160 *pos = hex[val & 0xf];
165 * NOTE! This returns a statically allocated buffer, so you have to be
166 * careful about using it. Do an "xstrdup()" if you need to save the
169 * Also note that this returns the location for creating. Reading
170 * SHA1 file can happen from any alternate directory listed in the
171 * DB_ENVIRONMENT environment variable if it is not found in
172 * the primary object database.
174 char *sha1_file_name(const unsigned char *sha1)
176 static char *name, *base;
179 const char *sha1_file_directory = get_object_directory();
180 int len = strlen(sha1_file_directory);
181 base = xmalloc(len + 60);
182 memcpy(base, sha1_file_directory, len);
183 memset(base+len, 0, 60);
186 name = base + len + 1;
188 fill_sha1_path(name, sha1);
192 static char *sha1_get_pack_name(const unsigned char *sha1,
193 char **name, char **base, const char *which)
195 static const char hex[] = "0123456789abcdef";
200 const char *sha1_file_directory = get_object_directory();
201 int len = strlen(sha1_file_directory);
202 *base = xmalloc(len + 60);
203 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
204 sha1_file_directory, which);
205 *name = *base + len + 11;
210 for (i = 0; i < 20; i++) {
211 unsigned int val = *sha1++;
212 *buf++ = hex[val >> 4];
213 *buf++ = hex[val & 0xf];
219 char *sha1_pack_name(const unsigned char *sha1)
221 static char *name, *base;
223 return sha1_get_pack_name(sha1, &name, &base, "pack");
226 char *sha1_pack_index_name(const unsigned char *sha1)
228 static char *name, *base;
230 return sha1_get_pack_name(sha1, &name, &base, "idx");
233 struct alternate_object_database *alt_odb_list;
234 static struct alternate_object_database **alt_odb_tail;
236 static void read_info_alternates(const char * alternates, int depth);
239 * Prepare alternate object database registry.
241 * The variable alt_odb_list points at the list of struct
242 * alternate_object_database. The elements on this list come from
243 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
244 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
245 * whose contents is similar to that environment variable but can be
246 * LF separated. Its base points at a statically allocated buffer that
247 * contains "/the/directory/corresponding/to/.git/objects/...", while
248 * its name points just after the slash at the end of ".git/objects/"
249 * in the example above, and has enough space to hold 40-byte hex
250 * SHA1, an extra slash for the first level indirection, and the
253 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
256 const char *objdir = get_object_directory();
257 struct alternate_object_database *ent;
258 struct alternate_object_database *alt;
259 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int entlen = pfxlen + 43;
264 if (!is_absolute_path(entry) && relative_base) {
265 /* Relative alt-odb */
267 base_len = strlen(relative_base) + 1;
271 ent = xmalloc(sizeof(*ent) + entlen);
273 if (!is_absolute_path(entry) && relative_base) {
274 memcpy(ent->base, relative_base, base_len - 1);
275 ent->base[base_len - 1] = '/';
276 memcpy(ent->base + base_len, entry, len);
279 memcpy(ent->base, entry, pfxlen);
281 ent->name = ent->base + pfxlen + 1;
282 ent->base[pfxlen + 3] = '/';
283 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285 /* Detect cases where alternate disappeared */
286 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt = alt_odb_list; alt; alt = alt->next) {
298 if (!memcmp(ent->base, alt->base, pfxlen)) {
303 if (!memcmp(ent->base, objdir, pfxlen)) {
308 /* add the alternate entry */
310 alt_odb_tail = &(ent->next);
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
316 ent->base[pfxlen] = '/';
321 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
322 const char *relative_base, int depth)
324 const char *cp, *last;
327 error("%s: ignoring alternate object stores, nesting too deep.",
335 if (cp < ep && *cp == '#') {
336 while (cp < ep && *cp != sep)
341 while (cp < ep && *cp != sep)
344 if (!is_absolute_path(last) && depth) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base, last);
348 link_alt_odb_entry(last, cp - last,
349 relative_base, depth);
352 while (cp < ep && *cp == sep)
358 static void read_info_alternates(const char * relative_base, int depth)
363 const char alt_file_name[] = "info/alternates";
364 /* Given that relative_base is no longer than PATH_MAX,
365 ensure that "path" has enough space to append "/", the
366 file name, "info/alternates", and a trailing NUL. */
367 char path[PATH_MAX + 1 + sizeof alt_file_name];
370 sprintf(path, "%s/%s", relative_base, alt_file_name);
371 fd = open(path, O_RDONLY);
374 if (fstat(fd, &st) || (st.st_size == 0)) {
378 mapsz = xsize_t(st.st_size);
379 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
382 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
387 void add_to_alternates_file(const char *reference)
389 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
390 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
391 char *alt = mkpath("%s/objects\n", reference);
392 write_or_die(fd, alt, strlen(alt));
393 if (commit_lock_file(lock))
394 die("could not close alternates file");
396 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
399 void prepare_alt_odb(void)
406 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409 alt_odb_tail = &alt_odb_list;
410 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
412 read_info_alternates(get_object_directory(), 0);
415 static int has_loose_object_local(const unsigned char *sha1)
417 char *name = sha1_file_name(sha1);
418 return !access(name, F_OK);
421 int has_loose_object_nonlocal(const unsigned char *sha1)
423 struct alternate_object_database *alt;
425 for (alt = alt_odb_list; alt; alt = alt->next) {
426 fill_sha1_path(alt->name, sha1);
427 if (!access(alt->base, F_OK))
433 static int has_loose_object(const unsigned char *sha1)
435 return has_loose_object_local(sha1) ||
436 has_loose_object_nonlocal(sha1);
439 static unsigned int pack_used_ctr;
440 static unsigned int pack_mmap_calls;
441 static unsigned int peak_pack_open_windows;
442 static unsigned int pack_open_windows;
443 static size_t peak_pack_mapped;
444 static size_t pack_mapped;
445 struct packed_git *packed_git;
447 void pack_report(void)
450 "pack_report: getpagesize() = %10" SZ_FMT "\n"
451 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
452 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
453 sz_fmt(getpagesize()),
454 sz_fmt(packed_git_window_size),
455 sz_fmt(packed_git_limit));
457 "pack_report: pack_used_ctr = %10u\n"
458 "pack_report: pack_mmap_calls = %10u\n"
459 "pack_report: pack_open_windows = %10u / %10u\n"
460 "pack_report: pack_mapped = "
461 "%10" SZ_FMT " / %10" SZ_FMT "\n",
464 pack_open_windows, peak_pack_open_windows,
465 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
468 static int check_packed_git_idx(const char *path, struct packed_git *p)
471 struct pack_idx_header *hdr;
473 uint32_t version, nr, i, *index;
474 int fd = open(path, O_RDONLY);
479 if (fstat(fd, &st)) {
483 idx_size = xsize_t(st.st_size);
484 if (idx_size < 4 * 256 + 20 + 20) {
486 return error("index file %s is too small", path);
488 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
492 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
493 version = ntohl(hdr->idx_version);
494 if (version < 2 || version > 2) {
495 munmap(idx_map, idx_size);
496 return error("index file %s is version %"PRIu32
497 " and is not supported by this binary"
498 " (try upgrading GIT to a newer version)",
507 index += 2; /* skip index header */
508 for (i = 0; i < 256; i++) {
509 uint32_t n = ntohl(index[i]);
511 munmap(idx_map, idx_size);
512 return error("non-monotonic index %s", path);
520 * - 256 index entries 4 bytes each
521 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
522 * - 20-byte SHA1 of the packfile
523 * - 20-byte SHA1 file checksum
525 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
526 munmap(idx_map, idx_size);
527 return error("wrong index v1 file size in %s", path);
529 } else if (version == 2) {
532 * - 8 bytes of header
533 * - 256 index entries 4 bytes each
534 * - 20-byte sha1 entry * nr
535 * - 4-byte crc entry * nr
536 * - 4-byte offset entry * nr
537 * - 20-byte SHA1 of the packfile
538 * - 20-byte SHA1 file checksum
539 * And after the 4-byte offset table might be a
540 * variable sized table containing 8-byte entries
541 * for offsets larger than 2^31.
543 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
544 unsigned long max_size = min_size;
546 max_size += (nr - 1)*8;
547 if (idx_size < min_size || idx_size > max_size) {
548 munmap(idx_map, idx_size);
549 return error("wrong index v2 file size in %s", path);
551 if (idx_size != min_size &&
553 * make sure we can deal with large pack offsets.
554 * 31-bit signed offset won't be enough, neither
555 * 32-bit unsigned one will be.
557 (sizeof(off_t) <= 4)) {
558 munmap(idx_map, idx_size);
559 return error("pack too large for current definition of off_t in %s", path);
563 p->index_version = version;
564 p->index_data = idx_map;
565 p->index_size = idx_size;
570 int open_pack_index(struct packed_git *p)
578 idx_name = xstrdup(p->pack_name);
579 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
580 ret = check_packed_git_idx(idx_name, p);
585 static void scan_windows(struct packed_git *p,
586 struct packed_git **lru_p,
587 struct pack_window **lru_w,
588 struct pack_window **lru_l)
590 struct pack_window *w, *w_l;
592 for (w_l = NULL, w = p->windows; w; w = w->next) {
594 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
604 static int unuse_one_window(struct packed_git *current, int keep_fd)
606 struct packed_git *p, *lru_p = NULL;
607 struct pack_window *lru_w = NULL, *lru_l = NULL;
610 scan_windows(current, &lru_p, &lru_w, &lru_l);
611 for (p = packed_git; p; p = p->next)
612 scan_windows(p, &lru_p, &lru_w, &lru_l);
614 munmap(lru_w->base, lru_w->len);
615 pack_mapped -= lru_w->len;
617 lru_l->next = lru_w->next;
619 lru_p->windows = lru_w->next;
620 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
621 close(lru_p->pack_fd);
632 void release_pack_memory(size_t need, int fd)
634 size_t cur = pack_mapped;
635 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
639 void close_pack_windows(struct packed_git *p)
642 struct pack_window *w = p->windows;
645 die("pack '%s' still has open windows to it",
647 munmap(w->base, w->len);
648 pack_mapped -= w->len;
650 p->windows = w->next;
655 void unuse_pack(struct pack_window **w_cursor)
657 struct pack_window *w = *w_cursor;
665 * This is used by git-repack in case a newly created pack happens to
666 * contain the same set of objects as an existing one. In that case
667 * the resulting file might be different even if its name would be the
668 * same. It is best to close any reference to the old pack before it is
669 * replaced on disk. Of course no index pointers nor windows for given pack
670 * must subsist at this point. If ever objects from this pack are requested
671 * again, the new version of the pack will be reinitialized through
672 * reprepare_packed_git().
674 void free_pack_by_name(const char *pack_name)
676 struct packed_git *p, **pp = &packed_git;
680 if (strcmp(pack_name, p->pack_name) == 0) {
681 close_pack_windows(p);
682 if (p->pack_fd != -1)
685 munmap((void *)p->index_data, p->index_size);
686 free(p->bad_object_sha1);
696 * Do not call this directly as this leaks p->pack_fd on error return;
697 * call open_packed_git() instead.
699 static int open_packed_git_1(struct packed_git *p)
702 struct pack_header hdr;
703 unsigned char sha1[20];
704 unsigned char *idx_sha1;
707 if (!p->index_data && open_pack_index(p))
708 return error("packfile %s index unavailable", p->pack_name);
710 p->pack_fd = open(p->pack_name, O_RDONLY);
711 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
714 /* If we created the struct before we had the pack we lack size. */
716 if (!S_ISREG(st.st_mode))
717 return error("packfile %s not a regular file", p->pack_name);
718 p->pack_size = st.st_size;
719 } else if (p->pack_size != st.st_size)
720 return error("packfile %s size changed", p->pack_name);
722 /* We leave these file descriptors open with sliding mmap;
723 * there is no point keeping them open across exec(), though.
725 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
727 return error("cannot determine file descriptor flags");
728 fd_flag |= FD_CLOEXEC;
729 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
730 return error("cannot set FD_CLOEXEC");
732 /* Verify we recognize this pack file format. */
733 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
734 return error("file %s is far too short to be a packfile", p->pack_name);
735 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
736 return error("file %s is not a GIT packfile", p->pack_name);
737 if (!pack_version_ok(hdr.hdr_version))
738 return error("packfile %s is version %"PRIu32" and not"
739 " supported (try upgrading GIT to a newer version)",
740 p->pack_name, ntohl(hdr.hdr_version));
742 /* Verify the pack matches its index. */
743 if (p->num_objects != ntohl(hdr.hdr_entries))
744 return error("packfile %s claims to have %"PRIu32" objects"
745 " while index indicates %"PRIu32" objects",
746 p->pack_name, ntohl(hdr.hdr_entries),
748 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
749 return error("end of packfile %s is unavailable", p->pack_name);
750 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
751 return error("packfile %s signature is unavailable", p->pack_name);
752 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
753 if (hashcmp(sha1, idx_sha1))
754 return error("packfile %s does not match index", p->pack_name);
758 static int open_packed_git(struct packed_git *p)
760 if (!open_packed_git_1(p))
762 if (p->pack_fd != -1) {
769 static int in_window(struct pack_window *win, off_t offset)
771 /* We must promise at least 20 bytes (one hash) after the
772 * offset is available from this window, otherwise the offset
773 * is not actually in this window and a different window (which
774 * has that one hash excess) must be used. This is to support
775 * the object header and delta base parsing routines below.
777 off_t win_off = win->offset;
778 return win_off <= offset
779 && (offset + 20) <= (win_off + win->len);
782 unsigned char* use_pack(struct packed_git *p,
783 struct pack_window **w_cursor,
787 struct pack_window *win = *w_cursor;
789 if (p->pack_fd == -1 && open_packed_git(p))
790 die("packfile %s cannot be accessed", p->pack_name);
792 /* Since packfiles end in a hash of their content and its
793 * pointless to ask for an offset into the middle of that
794 * hash, and the in_window function above wouldn't match
795 * don't allow an offset too close to the end of the file.
797 if (offset > (p->pack_size - 20))
798 die("offset beyond end of packfile (truncated pack?)");
800 if (!win || !in_window(win, offset)) {
803 for (win = p->windows; win; win = win->next) {
804 if (in_window(win, offset))
808 size_t window_align = packed_git_window_size / 2;
810 win = xcalloc(1, sizeof(*win));
811 win->offset = (offset / window_align) * window_align;
812 len = p->pack_size - win->offset;
813 if (len > packed_git_window_size)
814 len = packed_git_window_size;
815 win->len = (size_t)len;
816 pack_mapped += win->len;
817 while (packed_git_limit < pack_mapped
818 && unuse_one_window(p, p->pack_fd))
820 win->base = xmmap(NULL, win->len,
821 PROT_READ, MAP_PRIVATE,
822 p->pack_fd, win->offset);
823 if (win->base == MAP_FAILED)
824 die("packfile %s cannot be mapped: %s",
829 if (pack_mapped > peak_pack_mapped)
830 peak_pack_mapped = pack_mapped;
831 if (pack_open_windows > peak_pack_open_windows)
832 peak_pack_open_windows = pack_open_windows;
833 win->next = p->windows;
837 if (win != *w_cursor) {
838 win->last_used = pack_used_ctr++;
842 offset -= win->offset;
844 *left = win->len - xsize_t(offset);
845 return win->base + offset;
848 static struct packed_git *alloc_packed_git(int extra)
850 struct packed_git *p = xmalloc(sizeof(*p) + extra);
851 memset(p, 0, sizeof(*p));
856 struct packed_git *add_packed_git(const char *path, int path_len, int local)
859 struct packed_git *p = alloc_packed_git(path_len + 2);
862 * Make sure a corresponding .pack file exists and that
863 * the index looks sane.
865 path_len -= strlen(".idx");
870 memcpy(p->pack_name, path, path_len);
872 strcpy(p->pack_name + path_len, ".keep");
873 if (!access(p->pack_name, F_OK))
876 strcpy(p->pack_name + path_len, ".pack");
877 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
882 /* ok, it looks sane as far as we can check without
883 * actually mapping the pack file.
885 p->pack_size = st.st_size;
886 p->pack_local = local;
887 p->mtime = st.st_mtime;
888 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
893 struct packed_git *parse_pack_index(unsigned char *sha1)
895 const char *idx_path = sha1_pack_index_name(sha1);
896 const char *path = sha1_pack_name(sha1);
897 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
899 strcpy(p->pack_name, path);
900 hashcpy(p->sha1, sha1);
901 if (check_packed_git_idx(idx_path, p)) {
909 void install_packed_git(struct packed_git *pack)
911 pack->next = packed_git;
915 static void prepare_packed_git_one(char *objdir, int local)
917 /* Ensure that this buffer is large enough so that we can
918 append "/pack/" without clobbering the stack even if
919 strlen(objdir) were PATH_MAX. */
920 char path[PATH_MAX + 1 + 4 + 1 + 1];
925 sprintf(path, "%s/pack", objdir);
930 error("unable to open object pack directory: %s: %s",
931 path, strerror(errno));
935 while ((de = readdir(dir)) != NULL) {
936 int namelen = strlen(de->d_name);
937 struct packed_git *p;
939 if (!has_extension(de->d_name, ".idx"))
942 if (len + namelen + 1 > sizeof(path))
945 /* Don't reopen a pack we already have. */
946 strcpy(path + len, de->d_name);
947 for (p = packed_git; p; p = p->next) {
948 if (!memcmp(path, p->pack_name, len + namelen - 4))
953 /* See if it really is a valid .idx file with corresponding
954 * .pack file that we can map.
956 p = add_packed_git(path, len + namelen, local);
959 install_packed_git(p);
964 static int sort_pack(const void *a_, const void *b_)
966 struct packed_git *a = *((struct packed_git **)a_);
967 struct packed_git *b = *((struct packed_git **)b_);
971 * Local packs tend to contain objects specific to our
972 * variant of the project than remote ones. In addition,
973 * remote ones could be on a network mounted filesystem.
974 * Favor local ones for these reasons.
976 st = a->pack_local - b->pack_local;
981 * Younger packs tend to contain more recent objects,
982 * and more recent objects tend to get accessed more
985 if (a->mtime < b->mtime)
987 else if (a->mtime == b->mtime)
992 static void rearrange_packed_git(void)
994 struct packed_git **ary, *p;
997 for (n = 0, p = packed_git; p; p = p->next)
1002 /* prepare an array of packed_git for easier sorting */
1003 ary = xcalloc(n, sizeof(struct packed_git *));
1004 for (n = 0, p = packed_git; p; p = p->next)
1007 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1009 /* link them back again */
1010 for (i = 0; i < n - 1; i++)
1011 ary[i]->next = ary[i + 1];
1012 ary[n - 1]->next = NULL;
1013 packed_git = ary[0];
1018 static int prepare_packed_git_run_once = 0;
1019 void prepare_packed_git(void)
1021 struct alternate_object_database *alt;
1023 if (prepare_packed_git_run_once)
1025 prepare_packed_git_one(get_object_directory(), 1);
1027 for (alt = alt_odb_list; alt; alt = alt->next) {
1029 prepare_packed_git_one(alt->base, 0);
1030 alt->name[-1] = '/';
1032 rearrange_packed_git();
1033 prepare_packed_git_run_once = 1;
1036 void reprepare_packed_git(void)
1039 prepare_packed_git_run_once = 0;
1040 prepare_packed_git();
1043 static void mark_bad_packed_object(struct packed_git *p,
1044 const unsigned char *sha1)
1047 for (i = 0; i < p->num_bad_objects; i++)
1048 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1050 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1051 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1052 p->num_bad_objects++;
1055 static int has_packed_and_bad(const unsigned char *sha1)
1057 struct packed_git *p;
1060 for (p = packed_git; p; p = p->next)
1061 for (i = 0; i < p->num_bad_objects; i++)
1062 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1067 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1069 unsigned char real_sha1[20];
1070 hash_sha1_file(map, size, type, real_sha1);
1071 return hashcmp(sha1, real_sha1) ? -1 : 0;
1074 static int git_open_noatime(const char *name)
1076 static int sha1_file_open_flag = O_NOATIME;
1077 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1079 /* Might the failure be due to O_NOATIME? */
1080 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1081 fd = open(name, O_RDONLY);
1083 sha1_file_open_flag = 0;
1088 static int open_sha1_file(const unsigned char *sha1)
1091 char *name = sha1_file_name(sha1);
1092 struct alternate_object_database *alt;
1094 fd = git_open_noatime(name);
1100 for (alt = alt_odb_list; alt; alt = alt->next) {
1102 fill_sha1_path(name, sha1);
1103 fd = git_open_noatime(alt->base);
1110 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1115 fd = open_sha1_file(sha1);
1120 if (!fstat(fd, &st)) {
1121 *size = xsize_t(st.st_size);
1122 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1129 static int legacy_loose_object(unsigned char *map)
1134 * Is it a zlib-compressed buffer? If so, the first byte
1135 * must be 0x78 (15-bit window size, deflated), and the
1136 * first 16-bit word is evenly divisible by 31
1138 word = (map[0] << 8) + map[1];
1139 if (map[0] == 0x78 && !(word % 31))
1145 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1150 unsigned long used = 0;
1153 *type = (c >> 4) & 7;
1159 if (sizeof(long) * 8 <= shift)
1162 size += (c & 0x7f) << shift;
1169 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1171 unsigned long size, used;
1172 static const char valid_loose_object_type[8] = {
1174 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1175 0, /* "delta" and others are invalid in a loose object */
1177 enum object_type type;
1179 /* Get the data stream */
1180 memset(stream, 0, sizeof(*stream));
1181 stream->next_in = map;
1182 stream->avail_in = mapsize;
1183 stream->next_out = buffer;
1184 stream->avail_out = bufsiz;
1186 if (legacy_loose_object(map)) {
1187 inflateInit(stream);
1188 return inflate(stream, 0);
1193 * There used to be a second loose object header format which
1194 * was meant to mimic the in-pack format, allowing for direct
1195 * copy of the object data. This format turned up not to be
1196 * really worth it and we don't write it any longer. But we
1197 * can still read it.
1199 used = unpack_object_header_gently(map, mapsize, &type, &size);
1200 if (!used || !valid_loose_object_type[type])
1205 /* Set up the stream for the rest.. */
1206 stream->next_in = map;
1207 stream->avail_in = mapsize;
1208 inflateInit(stream);
1210 /* And generate the fake traditional header */
1211 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1212 typename(type), size);
1216 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1218 int bytes = strlen(buffer) + 1;
1219 unsigned char *buf = xmalloc(1+size);
1223 n = stream->total_out - bytes;
1226 memcpy(buf, (char *) buffer + bytes, n);
1228 if (bytes <= size) {
1230 * The above condition must be (bytes <= size), not
1231 * (bytes < size). In other words, even though we
1232 * expect no more output and set avail_out to zer0,
1233 * the input zlib stream may have bytes that express
1234 * "this concludes the stream", and we *do* want to
1237 * Otherwise we would not be able to test that we
1238 * consumed all the input to reach the expected size;
1239 * we also want to check that zlib tells us that all
1240 * went well with status == Z_STREAM_END at the end.
1242 stream->next_out = buf + bytes;
1243 stream->avail_out = size - bytes;
1244 while (status == Z_OK)
1245 status = inflate(stream, Z_FINISH);
1248 if (status == Z_STREAM_END && !stream->avail_in) {
1254 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1255 else if (stream->avail_in)
1256 error("garbage at end of loose object '%s'",
1263 * We used to just use "sscanf()", but that's actually way
1264 * too permissive for what we want to check. So do an anal
1265 * object header parse by hand.
1267 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1274 * The type can be at most ten bytes (including the
1275 * terminating '\0' that we add), and is followed by
1284 if (i >= sizeof(type))
1290 * The length must follow immediately, and be in canonical
1291 * decimal format (ie "010" is not valid).
1293 size = *hdr++ - '0';
1298 unsigned long c = *hdr - '0';
1302 size = size * 10 + c;
1308 * The length must be followed by a zero byte
1310 return *hdr ? -1 : type_from_string(type);
1313 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1319 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1320 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1323 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1326 unsigned long get_size_from_delta(struct packed_git *p,
1327 struct pack_window **w_curs,
1330 const unsigned char *data;
1331 unsigned char delta_head[20], *in;
1335 memset(&stream, 0, sizeof(stream));
1336 stream.next_out = delta_head;
1337 stream.avail_out = sizeof(delta_head);
1339 inflateInit(&stream);
1341 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1342 stream.next_in = in;
1343 st = inflate(&stream, Z_FINISH);
1344 curpos += stream.next_in - in;
1345 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1346 stream.total_out < sizeof(delta_head));
1347 inflateEnd(&stream);
1348 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1349 die("delta data unpack-initial failed");
1351 /* Examine the initial part of the delta to figure out
1356 /* ignore base size */
1357 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1359 /* Read the result size */
1360 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1363 static off_t get_delta_base(struct packed_git *p,
1364 struct pack_window **w_curs,
1366 enum object_type type,
1367 off_t delta_obj_offset)
1369 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1372 /* use_pack() assured us we have [base_info, base_info + 20)
1373 * as a range that we can look at without walking off the
1374 * end of the mapped window. Its actually the hash size
1375 * that is assured. An OFS_DELTA longer than the hash size
1376 * is stupid, as then a REF_DELTA would be smaller to store.
1378 if (type == OBJ_OFS_DELTA) {
1380 unsigned char c = base_info[used++];
1381 base_offset = c & 127;
1384 if (!base_offset || MSB(base_offset, 7))
1385 return 0; /* overflow */
1386 c = base_info[used++];
1387 base_offset = (base_offset << 7) + (c & 127);
1389 base_offset = delta_obj_offset - base_offset;
1390 if (base_offset >= delta_obj_offset)
1391 return 0; /* out of bound */
1393 } else if (type == OBJ_REF_DELTA) {
1394 /* The base entry _must_ be in the same pack */
1395 base_offset = find_pack_entry_one(base_info, p);
1398 die("I am totally screwed");
1402 /* forward declaration for a mutually recursive function */
1403 static int packed_object_info(struct packed_git *p, off_t offset,
1404 unsigned long *sizep);
1406 static int packed_delta_info(struct packed_git *p,
1407 struct pack_window **w_curs,
1409 enum object_type type,
1411 unsigned long *sizep)
1415 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1416 type = packed_object_info(p, base_offset, NULL);
1418 /* We choose to only get the type of the base object and
1419 * ignore potentially corrupt pack file that expects the delta
1420 * based on a base with a wrong size. This saves tons of
1424 *sizep = get_size_from_delta(p, w_curs, curpos);
1429 static int unpack_object_header(struct packed_git *p,
1430 struct pack_window **w_curs,
1432 unsigned long *sizep)
1434 unsigned char *base;
1437 enum object_type type;
1439 /* use_pack() assures us we have [base, base + 20) available
1440 * as a range that we can look at at. (Its actually the hash
1441 * size that is assured.) With our object header encoding
1442 * the maximum deflated object size is 2^137, which is just
1443 * insane, so we know won't exceed what we have been given.
1445 base = use_pack(p, w_curs, *curpos, &left);
1446 used = unpack_object_header_gently(base, left, &type, sizep);
1448 die("object offset outside of pack file");
1454 const char *packed_object_info_detail(struct packed_git *p,
1456 unsigned long *size,
1457 unsigned long *store_size,
1458 unsigned int *delta_chain_length,
1459 unsigned char *base_sha1)
1461 struct pack_window *w_curs = NULL;
1463 unsigned long dummy;
1464 unsigned char *next_sha1;
1465 enum object_type type;
1466 struct revindex_entry *revidx;
1468 *delta_chain_length = 0;
1469 curpos = obj_offset;
1470 type = unpack_object_header(p, &w_curs, &curpos, size);
1472 revidx = find_pack_revindex(p, obj_offset);
1473 *store_size = revidx[1].offset - obj_offset;
1478 die("pack %s contains unknown object type %d",
1479 p->pack_name, type);
1484 unuse_pack(&w_curs);
1485 return typename(type);
1487 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1489 die("pack %s contains bad delta base reference of type %s",
1490 p->pack_name, typename(type));
1491 if (*delta_chain_length == 0) {
1492 revidx = find_pack_revindex(p, obj_offset);
1493 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1497 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1498 if (*delta_chain_length == 0)
1499 hashcpy(base_sha1, next_sha1);
1500 obj_offset = find_pack_entry_one(next_sha1, p);
1503 (*delta_chain_length)++;
1504 curpos = obj_offset;
1505 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1509 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1510 unsigned long *sizep)
1512 struct pack_window *w_curs = NULL;
1514 off_t curpos = obj_offset;
1515 enum object_type type;
1517 type = unpack_object_header(p, &w_curs, &curpos, &size);
1522 type = packed_delta_info(p, &w_curs, curpos,
1523 type, obj_offset, sizep);
1533 die("pack %s contains unknown object type %d",
1534 p->pack_name, type);
1536 unuse_pack(&w_curs);
1540 static void *unpack_compressed_entry(struct packed_git *p,
1541 struct pack_window **w_curs,
1547 unsigned char *buffer, *in;
1549 buffer = xmalloc(size + 1);
1551 memset(&stream, 0, sizeof(stream));
1552 stream.next_out = buffer;
1553 stream.avail_out = size;
1555 inflateInit(&stream);
1557 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1558 stream.next_in = in;
1559 st = inflate(&stream, Z_FINISH);
1560 curpos += stream.next_in - in;
1561 } while (st == Z_OK || st == Z_BUF_ERROR);
1562 inflateEnd(&stream);
1563 if ((st != Z_STREAM_END) || stream.total_out != size) {
1571 #define MAX_DELTA_CACHE (256)
1573 static size_t delta_base_cached;
1575 static struct delta_base_cache_lru_list {
1576 struct delta_base_cache_lru_list *prev;
1577 struct delta_base_cache_lru_list *next;
1578 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1580 static struct delta_base_cache_entry {
1581 struct delta_base_cache_lru_list lru;
1583 struct packed_git *p;
1586 enum object_type type;
1587 } delta_base_cache[MAX_DELTA_CACHE];
1589 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1593 hash = (unsigned long)p + (unsigned long)base_offset;
1594 hash += (hash >> 8) + (hash >> 16);
1595 return hash % MAX_DELTA_CACHE;
1598 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1599 unsigned long *base_size, enum object_type *type, int keep_cache)
1602 unsigned long hash = pack_entry_hash(p, base_offset);
1603 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1606 if (ret && ent->p == p && ent->base_offset == base_offset)
1607 goto found_cache_entry;
1608 return unpack_entry(p, base_offset, type, base_size);
1613 ent->lru.next->prev = ent->lru.prev;
1614 ent->lru.prev->next = ent->lru.next;
1615 delta_base_cached -= ent->size;
1617 ret = xmemdupz(ent->data, ent->size);
1620 *base_size = ent->size;
1624 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1629 ent->lru.next->prev = ent->lru.prev;
1630 ent->lru.prev->next = ent->lru.next;
1631 delta_base_cached -= ent->size;
1635 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1636 void *base, unsigned long base_size, enum object_type type)
1638 unsigned long hash = pack_entry_hash(p, base_offset);
1639 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1640 struct delta_base_cache_lru_list *lru;
1642 release_delta_base_cache(ent);
1643 delta_base_cached += base_size;
1645 for (lru = delta_base_cache_lru.next;
1646 delta_base_cached > delta_base_cache_limit
1647 && lru != &delta_base_cache_lru;
1649 struct delta_base_cache_entry *f = (void *)lru;
1650 if (f->type == OBJ_BLOB)
1651 release_delta_base_cache(f);
1653 for (lru = delta_base_cache_lru.next;
1654 delta_base_cached > delta_base_cache_limit
1655 && lru != &delta_base_cache_lru;
1657 struct delta_base_cache_entry *f = (void *)lru;
1658 release_delta_base_cache(f);
1662 ent->base_offset = base_offset;
1665 ent->size = base_size;
1666 ent->lru.next = &delta_base_cache_lru;
1667 ent->lru.prev = delta_base_cache_lru.prev;
1668 delta_base_cache_lru.prev->next = &ent->lru;
1669 delta_base_cache_lru.prev = &ent->lru;
1672 static void *unpack_delta_entry(struct packed_git *p,
1673 struct pack_window **w_curs,
1675 unsigned long delta_size,
1677 enum object_type *type,
1678 unsigned long *sizep)
1680 void *delta_data, *result, *base;
1681 unsigned long base_size;
1684 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1686 error("failed to validate delta base reference "
1687 "at offset %"PRIuMAX" from %s",
1688 (uintmax_t)curpos, p->pack_name);
1692 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1695 * We're probably in deep shit, but let's try to fetch
1696 * the required base anyway from another pack or loose.
1697 * This is costly but should happen only in the presence
1698 * of a corrupted pack, and is better than failing outright.
1700 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1701 const unsigned char *base_sha1 =
1702 nth_packed_object_sha1(p, revidx->nr);
1703 error("failed to read delta base object %s"
1704 " at offset %"PRIuMAX" from %s",
1705 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1707 mark_bad_packed_object(p, base_sha1);
1708 base = read_object(base_sha1, type, &base_size);
1713 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1715 error("failed to unpack compressed delta "
1716 "at offset %"PRIuMAX" from %s",
1717 (uintmax_t)curpos, p->pack_name);
1721 result = patch_delta(base, base_size,
1722 delta_data, delta_size,
1725 die("failed to apply delta");
1727 add_delta_base_cache(p, base_offset, base, base_size, *type);
1731 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1732 enum object_type *type, unsigned long *sizep)
1734 struct pack_window *w_curs = NULL;
1735 off_t curpos = obj_offset;
1738 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1742 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1743 obj_offset, type, sizep);
1749 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1753 error("unknown object type %i at offset %"PRIuMAX" in %s",
1754 *type, (uintmax_t)obj_offset, p->pack_name);
1756 unuse_pack(&w_curs);
1760 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1763 const unsigned char *index = p->index_data;
1765 if (open_pack_index(p))
1767 index = p->index_data;
1769 if (n >= p->num_objects)
1772 if (p->index_version == 1) {
1773 return index + 24 * n + 4;
1776 return index + 20 * n;
1780 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1782 const unsigned char *index = p->index_data;
1784 if (p->index_version == 1) {
1785 return ntohl(*((uint32_t *)(index + 24 * n)));
1788 index += 8 + p->num_objects * (20 + 4);
1789 off = ntohl(*((uint32_t *)(index + 4 * n)));
1790 if (!(off & 0x80000000))
1792 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1793 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1794 ntohl(*((uint32_t *)(index + 4)));
1798 off_t find_pack_entry_one(const unsigned char *sha1,
1799 struct packed_git *p)
1801 const uint32_t *level1_ofs = p->index_data;
1802 const unsigned char *index = p->index_data;
1803 unsigned hi, lo, stride;
1804 static int use_lookup = -1;
1805 static int debug_lookup = -1;
1807 if (debug_lookup < 0)
1808 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1811 if (open_pack_index(p))
1813 level1_ofs = p->index_data;
1814 index = p->index_data;
1816 if (p->index_version > 1) {
1821 hi = ntohl(level1_ofs[*sha1]);
1822 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1823 if (p->index_version > 1) {
1831 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1832 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1835 use_lookup = !!getenv("GIT_USE_LOOKUP");
1837 int pos = sha1_entry_pos(index, stride, 0,
1838 lo, hi, p->num_objects, sha1);
1841 return nth_packed_object_offset(p, pos);
1845 unsigned mi = (lo + hi) / 2;
1846 int cmp = hashcmp(index + mi * stride, sha1);
1849 printf("lo %u hi %u rg %u mi %u\n",
1850 lo, hi, hi - lo, mi);
1852 return nth_packed_object_offset(p, mi);
1861 static int matches_pack_name(const struct packed_git *p, const char *name)
1863 const char *last_c, *c;
1865 if (!strcmp(p->pack_name, name))
1868 for (c = p->pack_name, last_c = c; *c;)
1873 if (!strcmp(last_c, name))
1879 int is_kept_pack(const struct packed_git *p, const struct rev_info *revs)
1883 for (i = 0; i < revs->num_ignore_packed; i++) {
1884 if (matches_pack_name(p, revs->ignore_packed[i]))
1890 static int find_pack_ent(const unsigned char *sha1, struct pack_entry *e,
1891 const struct rev_info *revs)
1893 static struct packed_git *last_found = (void *)1;
1894 struct packed_git *p;
1897 prepare_packed_git();
1900 p = (last_found == (void *)1) ? packed_git : last_found;
1903 if (revs->ignore_packed && !is_kept_pack(p, revs))
1905 if (p->num_bad_objects) {
1907 for (i = 0; i < p->num_bad_objects; i++)
1908 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1912 offset = find_pack_entry_one(sha1, p);
1915 * We are about to tell the caller where they can
1916 * locate the requested object. We better make
1917 * sure the packfile is still here and can be
1918 * accessed before supplying that answer, as
1919 * it may have been deleted since the index
1922 if (p->pack_fd == -1 && open_packed_git(p)) {
1923 error("packfile %s cannot be accessed", p->pack_name);
1928 hashcpy(e->sha1, sha1);
1934 if (p == last_found)
1938 if (p == last_found)
1944 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1946 return find_pack_ent(sha1, e, NULL);
1949 static int find_kept_pack_entry(const unsigned char *sha1, struct pack_entry *e,
1950 const struct rev_info *revs)
1952 return find_pack_ent(sha1, e, revs);
1955 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1956 struct packed_git *packs)
1958 struct packed_git *p;
1960 for (p = packs; p; p = p->next) {
1961 if (find_pack_entry_one(sha1, p))
1968 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1971 unsigned long mapsize, size;
1976 map = map_sha1_file(sha1, &mapsize);
1978 return error("unable to find %s", sha1_to_hex(sha1));
1979 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1980 status = error("unable to unpack %s header",
1982 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1983 status = error("unable to parse %s header", sha1_to_hex(sha1));
1986 inflateEnd(&stream);
1987 munmap(map, mapsize);
1991 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1993 struct pack_entry e;
1996 if (!find_pack_entry(sha1, &e)) {
1997 /* Most likely it's a loose object. */
1998 status = sha1_loose_object_info(sha1, sizep);
2002 /* Not a loose object; someone else may have just packed it. */
2003 reprepare_packed_git();
2004 if (!find_pack_entry(sha1, &e))
2007 return packed_object_info(e.p, e.offset, sizep);
2010 static void *read_packed_sha1(const unsigned char *sha1,
2011 enum object_type *type, unsigned long *size)
2013 struct pack_entry e;
2016 if (!find_pack_entry(sha1, &e))
2018 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2021 * We're probably in deep shit, but let's try to fetch
2022 * the required object anyway from another pack or loose.
2023 * This should happen only in the presence of a corrupted
2024 * pack, and is better than failing outright.
2026 error("failed to read object %s at offset %"PRIuMAX" from %s",
2027 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2028 mark_bad_packed_object(e.p, sha1);
2029 data = read_object(sha1, type, size);
2035 * This is meant to hold a *small* number of objects that you would
2036 * want read_sha1_file() to be able to return, but yet you do not want
2037 * to write them into the object store (e.g. a browse-only
2040 static struct cached_object {
2041 unsigned char sha1[20];
2042 enum object_type type;
2046 static int cached_object_nr, cached_object_alloc;
2048 static struct cached_object empty_tree = {
2049 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2050 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2051 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2057 static struct cached_object *find_cached_object(const unsigned char *sha1)
2060 struct cached_object *co = cached_objects;
2062 for (i = 0; i < cached_object_nr; i++, co++) {
2063 if (!hashcmp(co->sha1, sha1))
2066 if (!hashcmp(sha1, empty_tree.sha1))
2071 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2072 unsigned char *sha1)
2074 struct cached_object *co;
2076 hash_sha1_file(buf, len, typename(type), sha1);
2077 if (has_sha1_file(sha1) || find_cached_object(sha1))
2079 if (cached_object_alloc <= cached_object_nr) {
2080 cached_object_alloc = alloc_nr(cached_object_alloc);
2081 cached_objects = xrealloc(cached_objects,
2082 sizeof(*cached_objects) *
2083 cached_object_alloc);
2085 co = &cached_objects[cached_object_nr++];
2088 co->buf = xmalloc(len);
2089 memcpy(co->buf, buf, len);
2090 hashcpy(co->sha1, sha1);
2094 void *read_object(const unsigned char *sha1, enum object_type *type,
2095 unsigned long *size)
2097 unsigned long mapsize;
2099 struct cached_object *co;
2101 co = find_cached_object(sha1);
2105 return xmemdupz(co->buf, co->size);
2108 buf = read_packed_sha1(sha1, type, size);
2111 map = map_sha1_file(sha1, &mapsize);
2113 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2114 munmap(map, mapsize);
2117 reprepare_packed_git();
2118 return read_packed_sha1(sha1, type, size);
2121 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2122 unsigned long *size)
2124 void *data = read_object(sha1, type, size);
2125 /* legacy behavior is to die on corrupted objects */
2126 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2127 die("object %s is corrupted", sha1_to_hex(sha1));
2131 void *read_object_with_reference(const unsigned char *sha1,
2132 const char *required_type_name,
2133 unsigned long *size,
2134 unsigned char *actual_sha1_return)
2136 enum object_type type, required_type;
2138 unsigned long isize;
2139 unsigned char actual_sha1[20];
2141 required_type = type_from_string(required_type_name);
2142 hashcpy(actual_sha1, sha1);
2144 int ref_length = -1;
2145 const char *ref_type = NULL;
2147 buffer = read_sha1_file(actual_sha1, &type, &isize);
2150 if (type == required_type) {
2152 if (actual_sha1_return)
2153 hashcpy(actual_sha1_return, actual_sha1);
2156 /* Handle references */
2157 else if (type == OBJ_COMMIT)
2159 else if (type == OBJ_TAG)
2160 ref_type = "object ";
2165 ref_length = strlen(ref_type);
2167 if (ref_length + 40 > isize ||
2168 memcmp(buffer, ref_type, ref_length) ||
2169 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2174 /* Now we have the ID of the referred-to object in
2175 * actual_sha1. Check again. */
2179 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2180 const char *type, unsigned char *sha1,
2181 char *hdr, int *hdrlen)
2185 /* Generate the header */
2186 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2190 SHA1_Update(&c, hdr, *hdrlen);
2191 SHA1_Update(&c, buf, len);
2192 SHA1_Final(sha1, &c);
2196 * Move the just written object into its final resting place
2198 int move_temp_to_file(const char *tmpfile, const char *filename)
2201 if (link(tmpfile, filename))
2205 * Coda hack - coda doesn't like cross-directory links,
2206 * so we fall back to a rename, which will mean that it
2207 * won't be able to check collisions, but that's not a
2210 * The same holds for FAT formatted media.
2212 * When this succeeds, we just return 0. We have nothing
2215 if (ret && ret != EEXIST) {
2216 if (!rename(tmpfile, filename))
2222 if (ret != EEXIST) {
2223 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2225 /* FIXME!!! Collision check here ? */
2231 static int write_buffer(int fd, const void *buf, size_t len)
2233 if (write_in_full(fd, buf, len) < 0)
2234 return error("file write error (%s)", strerror(errno));
2238 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2239 unsigned char *sha1)
2243 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2247 /* Finalize a file on disk, and close it. */
2248 static void close_sha1_file(int fd)
2250 if (fsync_object_files)
2251 fsync_or_die(fd, "sha1 file");
2254 die("unable to write sha1 file");
2257 /* Size of directory component, including the ending '/' */
2258 static inline int directory_size(const char *filename)
2260 const char *s = strrchr(filename, '/');
2263 return s - filename + 1;
2267 * This creates a temporary file in the same directory as the final
2270 * We want to avoid cross-directory filename renames, because those
2271 * can have problems on various filesystems (FAT, NFS, Coda).
2273 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2275 int fd, dirlen = directory_size(filename);
2277 if (dirlen + 20 > bufsiz) {
2278 errno = ENAMETOOLONG;
2281 memcpy(buffer, filename, dirlen);
2282 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2283 fd = mkstemp(buffer);
2284 if (fd < 0 && dirlen && errno == ENOENT) {
2285 /* Make sure the directory exists */
2286 memcpy(buffer, filename, dirlen);
2287 buffer[dirlen-1] = 0;
2288 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2292 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2293 fd = mkstemp(buffer);
2298 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2299 void *buf, unsigned long len, time_t mtime)
2302 unsigned char *compressed;
2305 static char tmpfile[PATH_MAX];
2307 filename = sha1_file_name(sha1);
2308 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2310 if (errno == EACCES)
2311 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2313 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2317 memset(&stream, 0, sizeof(stream));
2318 deflateInit(&stream, zlib_compression_level);
2319 size = 8 + deflateBound(&stream, len+hdrlen);
2320 compressed = xmalloc(size);
2323 stream.next_out = compressed;
2324 stream.avail_out = size;
2326 /* First header.. */
2327 stream.next_in = (unsigned char *)hdr;
2328 stream.avail_in = hdrlen;
2329 while (deflate(&stream, 0) == Z_OK)
2332 /* Then the data itself.. */
2333 stream.next_in = buf;
2334 stream.avail_in = len;
2335 ret = deflate(&stream, Z_FINISH);
2336 if (ret != Z_STREAM_END)
2337 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2339 ret = deflateEnd(&stream);
2341 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2343 size = stream.total_out;
2345 if (write_buffer(fd, compressed, size) < 0)
2346 die("unable to write sha1 file");
2347 close_sha1_file(fd);
2353 utb.modtime = mtime;
2354 if (utime(tmpfile, &utb) < 0)
2355 warning("failed utime() on %s: %s",
2356 tmpfile, strerror(errno));
2359 return move_temp_to_file(tmpfile, filename);
2362 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2364 unsigned char sha1[20];
2368 /* Normally if we have it in the pack then we do not bother writing
2369 * it out into .git/objects/??/?{38} file.
2371 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2373 hashcpy(returnsha1, sha1);
2374 if (has_sha1_file(sha1))
2376 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2379 int force_object_loose(const unsigned char *sha1, time_t mtime)
2383 enum object_type type;
2388 if (has_loose_object(sha1))
2390 buf = read_packed_sha1(sha1, &type, &len);
2392 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2393 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2394 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2400 int has_pack_index(const unsigned char *sha1)
2403 if (stat(sha1_pack_index_name(sha1), &st))
2408 int has_pack_file(const unsigned char *sha1)
2411 if (stat(sha1_pack_name(sha1), &st))
2416 int has_sha1_pack(const unsigned char *sha1)
2418 struct pack_entry e;
2419 return find_pack_entry(sha1, &e);
2422 int has_sha1_kept_pack(const unsigned char *sha1, const struct rev_info *revs)
2424 struct pack_entry e;
2425 return find_kept_pack_entry(sha1, &e, revs);
2428 int has_sha1_file(const unsigned char *sha1)
2430 struct pack_entry e;
2432 if (find_pack_entry(sha1, &e))
2434 return has_loose_object(sha1);
2437 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2442 strbuf_init(&buf, 0);
2443 if (strbuf_read(&buf, fd, 4096) < 0) {
2444 strbuf_release(&buf);
2451 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2453 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2454 strbuf_release(&buf);
2459 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2460 enum object_type type, const char *path)
2462 size_t size = xsize_t(st->st_size);
2464 int ret, re_allocated = 0;
2467 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2474 * Convert blobs to git internal format
2476 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2478 strbuf_init(&nbuf, 0);
2479 if (convert_to_git(path, buf, size, &nbuf,
2480 write_object ? safe_crlf : 0)) {
2482 buf = strbuf_detach(&nbuf, &size);
2488 ret = write_sha1_file(buf, size, typename(type), sha1);
2490 ret = hash_sha1_file(buf, size, typename(type), sha1);
2500 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2506 switch (st->st_mode & S_IFMT) {
2508 fd = open(path, O_RDONLY);
2510 return error("open(\"%s\"): %s", path,
2512 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2513 return error("%s: failed to insert into database",
2517 len = xsize_t(st->st_size);
2518 target = xmalloc(len + 1);
2519 if (readlink(path, target, len + 1) != st->st_size) {
2520 char *errstr = strerror(errno);
2522 return error("readlink(\"%s\"): %s", path,
2526 hash_sha1_file(target, len, blob_type, sha1);
2527 else if (write_sha1_file(target, len, blob_type, sha1))
2528 return error("%s: failed to insert into database",
2533 return resolve_gitlink_ref(path, "HEAD", sha1);
2535 return error("%s: unsupported file type", path);
2540 int read_pack_header(int fd, struct pack_header *header)
2542 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2543 /* "eof before pack header was fully read" */
2544 return PH_ERROR_EOF;
2546 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2547 /* "protocol error (pack signature mismatch detected)" */
2548 return PH_ERROR_PACK_SIGNATURE;
2549 if (!pack_version_ok(header->hdr_version))
2550 /* "protocol error (pack version unsupported)" */
2551 return PH_ERROR_PROTOCOL;