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 int safe_create_leading_directories(char *path)
40 char *pos = path + offset_1st_component(path);
44 pos = strchr(pos, '/');
52 if (!stat(path, &st)) {
54 if (!S_ISDIR(st.st_mode)) {
59 else if (mkdir(path, 0777)) {
63 else if (adjust_shared_perm(path)) {
72 int safe_create_leading_directories_const(const char *path)
74 /* path points to cache entries, so xstrdup before messing with it */
75 char *buf = xstrdup(path);
76 int result = safe_create_leading_directories(buf);
81 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
84 for (i = 0; i < 20; i++) {
85 static char hex[] = "0123456789abcdef";
86 unsigned int val = sha1[i];
87 char *pos = pathbuf + i*2 + (i > 0);
88 *pos++ = hex[val >> 4];
89 *pos = hex[val & 0xf];
94 * NOTE! This returns a statically allocated buffer, so you have to be
95 * careful about using it. Do an "xstrdup()" if you need to save the
98 * Also note that this returns the location for creating. Reading
99 * SHA1 file can happen from any alternate directory listed in the
100 * DB_ENVIRONMENT environment variable if it is not found in
101 * the primary object database.
103 char *sha1_file_name(const unsigned char *sha1)
105 static char buf[PATH_MAX];
109 objdir = get_object_directory();
110 len = strlen(objdir);
112 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
113 if (len + 43 > PATH_MAX)
114 die("insanely long object directory %s", objdir);
115 memcpy(buf, objdir, len);
119 fill_sha1_path(buf + len + 1, sha1);
123 static char *sha1_get_pack_name(const unsigned char *sha1,
124 char **name, char **base, const char *which)
126 static const char hex[] = "0123456789abcdef";
131 const char *sha1_file_directory = get_object_directory();
132 int len = strlen(sha1_file_directory);
133 *base = xmalloc(len + 60);
134 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
135 sha1_file_directory, which);
136 *name = *base + len + 11;
141 for (i = 0; i < 20; i++) {
142 unsigned int val = *sha1++;
143 *buf++ = hex[val >> 4];
144 *buf++ = hex[val & 0xf];
150 char *sha1_pack_name(const unsigned char *sha1)
152 static char *name, *base;
154 return sha1_get_pack_name(sha1, &name, &base, "pack");
157 char *sha1_pack_index_name(const unsigned char *sha1)
159 static char *name, *base;
161 return sha1_get_pack_name(sha1, &name, &base, "idx");
164 struct alternate_object_database *alt_odb_list;
165 static struct alternate_object_database **alt_odb_tail;
167 static void read_info_alternates(const char * alternates, int depth);
170 * Prepare alternate object database registry.
172 * The variable alt_odb_list points at the list of struct
173 * alternate_object_database. The elements on this list come from
174 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
175 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
176 * whose contents is similar to that environment variable but can be
177 * LF separated. Its base points at a statically allocated buffer that
178 * contains "/the/directory/corresponding/to/.git/objects/...", while
179 * its name points just after the slash at the end of ".git/objects/"
180 * in the example above, and has enough space to hold 40-byte hex
181 * SHA1, an extra slash for the first level indirection, and the
184 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
186 const char *objdir = get_object_directory();
187 struct alternate_object_database *ent;
188 struct alternate_object_database *alt;
189 /* 43 = 40-byte + 2 '/' + terminating NUL */
191 int entlen = pfxlen + 43;
194 if (!is_absolute_path(entry) && relative_base) {
195 /* Relative alt-odb */
197 base_len = strlen(relative_base) + 1;
201 ent = xmalloc(sizeof(*ent) + entlen);
203 if (!is_absolute_path(entry) && relative_base) {
204 memcpy(ent->base, relative_base, base_len - 1);
205 ent->base[base_len - 1] = '/';
206 memcpy(ent->base + base_len, entry, len);
209 memcpy(ent->base, entry, pfxlen);
211 ent->name = ent->base + pfxlen + 1;
212 ent->base[pfxlen + 3] = '/';
213 ent->base[pfxlen] = ent->base[entlen-1] = 0;
215 /* Detect cases where alternate disappeared */
216 if (!is_directory(ent->base)) {
217 error("object directory %s does not exist; "
218 "check .git/objects/info/alternates.",
224 /* Prevent the common mistake of listing the same
225 * thing twice, or object directory itself.
227 for (alt = alt_odb_list; alt; alt = alt->next) {
228 if (!memcmp(ent->base, alt->base, pfxlen)) {
233 if (!memcmp(ent->base, objdir, pfxlen)) {
238 /* add the alternate entry */
240 alt_odb_tail = &(ent->next);
243 /* recursively add alternates */
244 read_info_alternates(ent->base, depth + 1);
246 ent->base[pfxlen] = '/';
251 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
252 const char *relative_base, int depth)
254 const char *cp, *last;
257 error("%s: ignoring alternate object stores, nesting too deep.",
265 if (cp < ep && *cp == '#') {
266 while (cp < ep && *cp != sep)
271 while (cp < ep && *cp != sep)
274 if (!is_absolute_path(last) && depth) {
275 error("%s: ignoring relative alternate object store %s",
276 relative_base, last);
278 link_alt_odb_entry(last, cp - last,
279 relative_base, depth);
282 while (cp < ep && *cp == sep)
288 static void read_info_alternates(const char * relative_base, int depth)
293 const char alt_file_name[] = "info/alternates";
294 /* Given that relative_base is no longer than PATH_MAX,
295 ensure that "path" has enough space to append "/", the
296 file name, "info/alternates", and a trailing NUL. */
297 char path[PATH_MAX + 1 + sizeof alt_file_name];
300 sprintf(path, "%s/%s", relative_base, alt_file_name);
301 fd = open(path, O_RDONLY);
304 if (fstat(fd, &st) || (st.st_size == 0)) {
308 mapsz = xsize_t(st.st_size);
309 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
312 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
317 void add_to_alternates_file(const char *reference)
319 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
320 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
321 char *alt = mkpath("%s/objects\n", reference);
322 write_or_die(fd, alt, strlen(alt));
323 if (commit_lock_file(lock))
324 die("could not close alternates file");
326 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
329 void foreach_alt_odb(alt_odb_fn fn, void *cb)
331 struct alternate_object_database *ent;
334 for (ent = alt_odb_list; ent; ent = ent->next)
339 void prepare_alt_odb(void)
346 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
349 alt_odb_tail = &alt_odb_list;
350 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
352 read_info_alternates(get_object_directory(), 0);
355 static int has_loose_object_local(const unsigned char *sha1)
357 char *name = sha1_file_name(sha1);
358 return !access(name, F_OK);
361 int has_loose_object_nonlocal(const unsigned char *sha1)
363 struct alternate_object_database *alt;
365 for (alt = alt_odb_list; alt; alt = alt->next) {
366 fill_sha1_path(alt->name, sha1);
367 if (!access(alt->base, F_OK))
373 static int has_loose_object(const unsigned char *sha1)
375 return has_loose_object_local(sha1) ||
376 has_loose_object_nonlocal(sha1);
379 static unsigned int pack_used_ctr;
380 static unsigned int pack_mmap_calls;
381 static unsigned int peak_pack_open_windows;
382 static unsigned int pack_open_windows;
383 static size_t peak_pack_mapped;
384 static size_t pack_mapped;
385 struct packed_git *packed_git;
387 void pack_report(void)
390 "pack_report: getpagesize() = %10" SZ_FMT "\n"
391 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
392 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
393 sz_fmt(getpagesize()),
394 sz_fmt(packed_git_window_size),
395 sz_fmt(packed_git_limit));
397 "pack_report: pack_used_ctr = %10u\n"
398 "pack_report: pack_mmap_calls = %10u\n"
399 "pack_report: pack_open_windows = %10u / %10u\n"
400 "pack_report: pack_mapped = "
401 "%10" SZ_FMT " / %10" SZ_FMT "\n",
404 pack_open_windows, peak_pack_open_windows,
405 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
408 static int check_packed_git_idx(const char *path, struct packed_git *p)
411 struct pack_idx_header *hdr;
413 uint32_t version, nr, i, *index;
414 int fd = open(path, O_RDONLY);
419 if (fstat(fd, &st)) {
423 idx_size = xsize_t(st.st_size);
424 if (idx_size < 4 * 256 + 20 + 20) {
426 return error("index file %s is too small", path);
428 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
432 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
433 version = ntohl(hdr->idx_version);
434 if (version < 2 || version > 2) {
435 munmap(idx_map, idx_size);
436 return error("index file %s is version %"PRIu32
437 " and is not supported by this binary"
438 " (try upgrading GIT to a newer version)",
447 index += 2; /* skip index header */
448 for (i = 0; i < 256; i++) {
449 uint32_t n = ntohl(index[i]);
451 munmap(idx_map, idx_size);
452 return error("non-monotonic index %s", path);
460 * - 256 index entries 4 bytes each
461 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
462 * - 20-byte SHA1 of the packfile
463 * - 20-byte SHA1 file checksum
465 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
466 munmap(idx_map, idx_size);
467 return error("wrong index v1 file size in %s", path);
469 } else if (version == 2) {
472 * - 8 bytes of header
473 * - 256 index entries 4 bytes each
474 * - 20-byte sha1 entry * nr
475 * - 4-byte crc entry * nr
476 * - 4-byte offset entry * nr
477 * - 20-byte SHA1 of the packfile
478 * - 20-byte SHA1 file checksum
479 * And after the 4-byte offset table might be a
480 * variable sized table containing 8-byte entries
481 * for offsets larger than 2^31.
483 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
484 unsigned long max_size = min_size;
486 max_size += (nr - 1)*8;
487 if (idx_size < min_size || idx_size > max_size) {
488 munmap(idx_map, idx_size);
489 return error("wrong index v2 file size in %s", path);
491 if (idx_size != min_size &&
493 * make sure we can deal with large pack offsets.
494 * 31-bit signed offset won't be enough, neither
495 * 32-bit unsigned one will be.
497 (sizeof(off_t) <= 4)) {
498 munmap(idx_map, idx_size);
499 return error("pack too large for current definition of off_t in %s", path);
503 p->index_version = version;
504 p->index_data = idx_map;
505 p->index_size = idx_size;
510 int open_pack_index(struct packed_git *p)
518 idx_name = xstrdup(p->pack_name);
519 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
520 ret = check_packed_git_idx(idx_name, p);
525 static void scan_windows(struct packed_git *p,
526 struct packed_git **lru_p,
527 struct pack_window **lru_w,
528 struct pack_window **lru_l)
530 struct pack_window *w, *w_l;
532 for (w_l = NULL, w = p->windows; w; w = w->next) {
534 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
544 static int unuse_one_window(struct packed_git *current, int keep_fd)
546 struct packed_git *p, *lru_p = NULL;
547 struct pack_window *lru_w = NULL, *lru_l = NULL;
550 scan_windows(current, &lru_p, &lru_w, &lru_l);
551 for (p = packed_git; p; p = p->next)
552 scan_windows(p, &lru_p, &lru_w, &lru_l);
554 munmap(lru_w->base, lru_w->len);
555 pack_mapped -= lru_w->len;
557 lru_l->next = lru_w->next;
559 lru_p->windows = lru_w->next;
560 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
561 close(lru_p->pack_fd);
572 void release_pack_memory(size_t need, int fd)
574 size_t cur = pack_mapped;
575 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
579 void *xmmap(void *start, size_t length,
580 int prot, int flags, int fd, off_t offset)
582 void *ret = mmap(start, length, prot, flags, fd, offset);
583 if (ret == MAP_FAILED) {
586 release_pack_memory(length, fd);
587 ret = mmap(start, length, prot, flags, fd, offset);
588 if (ret == MAP_FAILED)
589 die_errno("Out of memory? mmap failed");
594 void close_pack_windows(struct packed_git *p)
597 struct pack_window *w = p->windows;
600 die("pack '%s' still has open windows to it",
602 munmap(w->base, w->len);
603 pack_mapped -= w->len;
605 p->windows = w->next;
610 void unuse_pack(struct pack_window **w_cursor)
612 struct pack_window *w = *w_cursor;
619 void close_pack_index(struct packed_git *p)
622 munmap((void *)p->index_data, p->index_size);
623 p->index_data = NULL;
628 * This is used by git-repack in case a newly created pack happens to
629 * contain the same set of objects as an existing one. In that case
630 * the resulting file might be different even if its name would be the
631 * same. It is best to close any reference to the old pack before it is
632 * replaced on disk. Of course no index pointers nor windows for given pack
633 * must subsist at this point. If ever objects from this pack are requested
634 * again, the new version of the pack will be reinitialized through
635 * reprepare_packed_git().
637 void free_pack_by_name(const char *pack_name)
639 struct packed_git *p, **pp = &packed_git;
643 if (strcmp(pack_name, p->pack_name) == 0) {
644 clear_delta_base_cache();
645 close_pack_windows(p);
646 if (p->pack_fd != -1)
649 free(p->bad_object_sha1);
659 * Do not call this directly as this leaks p->pack_fd on error return;
660 * call open_packed_git() instead.
662 static int open_packed_git_1(struct packed_git *p)
665 struct pack_header hdr;
666 unsigned char sha1[20];
667 unsigned char *idx_sha1;
670 if (!p->index_data && open_pack_index(p))
671 return error("packfile %s index unavailable", p->pack_name);
673 p->pack_fd = open(p->pack_name, O_RDONLY);
674 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
675 p->pack_fd = open(p->pack_name, O_RDONLY);
676 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
679 /* If we created the struct before we had the pack we lack size. */
681 if (!S_ISREG(st.st_mode))
682 return error("packfile %s not a regular file", p->pack_name);
683 p->pack_size = st.st_size;
684 } else if (p->pack_size != st.st_size)
685 return error("packfile %s size changed", p->pack_name);
687 /* We leave these file descriptors open with sliding mmap;
688 * there is no point keeping them open across exec(), though.
690 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
692 return error("cannot determine file descriptor flags");
693 fd_flag |= FD_CLOEXEC;
694 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
695 return error("cannot set FD_CLOEXEC");
697 /* Verify we recognize this pack file format. */
698 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
699 return error("file %s is far too short to be a packfile", p->pack_name);
700 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
701 return error("file %s is not a GIT packfile", p->pack_name);
702 if (!pack_version_ok(hdr.hdr_version))
703 return error("packfile %s is version %"PRIu32" and not"
704 " supported (try upgrading GIT to a newer version)",
705 p->pack_name, ntohl(hdr.hdr_version));
707 /* Verify the pack matches its index. */
708 if (p->num_objects != ntohl(hdr.hdr_entries))
709 return error("packfile %s claims to have %"PRIu32" objects"
710 " while index indicates %"PRIu32" objects",
711 p->pack_name, ntohl(hdr.hdr_entries),
713 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
714 return error("end of packfile %s is unavailable", p->pack_name);
715 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
716 return error("packfile %s signature is unavailable", p->pack_name);
717 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
718 if (hashcmp(sha1, idx_sha1))
719 return error("packfile %s does not match index", p->pack_name);
723 static int open_packed_git(struct packed_git *p)
725 if (!open_packed_git_1(p))
727 if (p->pack_fd != -1) {
734 static int in_window(struct pack_window *win, off_t offset)
736 /* We must promise at least 20 bytes (one hash) after the
737 * offset is available from this window, otherwise the offset
738 * is not actually in this window and a different window (which
739 * has that one hash excess) must be used. This is to support
740 * the object header and delta base parsing routines below.
742 off_t win_off = win->offset;
743 return win_off <= offset
744 && (offset + 20) <= (win_off + win->len);
747 unsigned char *use_pack(struct packed_git *p,
748 struct pack_window **w_cursor,
752 struct pack_window *win = *w_cursor;
754 if (p->pack_fd == -1 && open_packed_git(p))
755 die("packfile %s cannot be accessed", p->pack_name);
757 /* Since packfiles end in a hash of their content and it's
758 * pointless to ask for an offset into the middle of that
759 * hash, and the in_window function above wouldn't match
760 * don't allow an offset too close to the end of the file.
762 if (offset > (p->pack_size - 20))
763 die("offset beyond end of packfile (truncated pack?)");
765 if (!win || !in_window(win, offset)) {
768 for (win = p->windows; win; win = win->next) {
769 if (in_window(win, offset))
773 size_t window_align = packed_git_window_size / 2;
775 win = xcalloc(1, sizeof(*win));
776 win->offset = (offset / window_align) * window_align;
777 len = p->pack_size - win->offset;
778 if (len > packed_git_window_size)
779 len = packed_git_window_size;
780 win->len = (size_t)len;
781 pack_mapped += win->len;
782 while (packed_git_limit < pack_mapped
783 && unuse_one_window(p, p->pack_fd))
785 win->base = xmmap(NULL, win->len,
786 PROT_READ, MAP_PRIVATE,
787 p->pack_fd, win->offset);
788 if (win->base == MAP_FAILED)
789 die("packfile %s cannot be mapped: %s",
794 if (pack_mapped > peak_pack_mapped)
795 peak_pack_mapped = pack_mapped;
796 if (pack_open_windows > peak_pack_open_windows)
797 peak_pack_open_windows = pack_open_windows;
798 win->next = p->windows;
802 if (win != *w_cursor) {
803 win->last_used = pack_used_ctr++;
807 offset -= win->offset;
809 *left = win->len - xsize_t(offset);
810 return win->base + offset;
813 static struct packed_git *alloc_packed_git(int extra)
815 struct packed_git *p = xmalloc(sizeof(*p) + extra);
816 memset(p, 0, sizeof(*p));
821 static void try_to_free_pack_memory(size_t size)
823 release_pack_memory(size, -1);
826 struct packed_git *add_packed_git(const char *path, int path_len, int local)
828 static int have_set_try_to_free_routine;
830 struct packed_git *p = alloc_packed_git(path_len + 2);
832 if (!have_set_try_to_free_routine) {
833 have_set_try_to_free_routine = 1;
834 set_try_to_free_routine(try_to_free_pack_memory);
838 * Make sure a corresponding .pack file exists and that
839 * the index looks sane.
841 path_len -= strlen(".idx");
846 memcpy(p->pack_name, path, path_len);
848 strcpy(p->pack_name + path_len, ".keep");
849 if (!access(p->pack_name, F_OK))
852 strcpy(p->pack_name + path_len, ".pack");
853 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
858 /* ok, it looks sane as far as we can check without
859 * actually mapping the pack file.
861 p->pack_size = st.st_size;
862 p->pack_local = local;
863 p->mtime = st.st_mtime;
864 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
869 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
871 const char *path = sha1_pack_name(sha1);
872 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
874 strcpy(p->pack_name, path);
875 hashcpy(p->sha1, sha1);
876 if (check_packed_git_idx(idx_path, p)) {
884 void install_packed_git(struct packed_git *pack)
886 pack->next = packed_git;
890 static void prepare_packed_git_one(char *objdir, int local)
892 /* Ensure that this buffer is large enough so that we can
893 append "/pack/" without clobbering the stack even if
894 strlen(objdir) were PATH_MAX. */
895 char path[PATH_MAX + 1 + 4 + 1 + 1];
900 sprintf(path, "%s/pack", objdir);
903 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
907 error("unable to open object pack directory: %s: %s",
908 path, strerror(errno));
912 while ((de = readdir(dir)) != NULL) {
913 int namelen = strlen(de->d_name);
914 struct packed_git *p;
916 if (!has_extension(de->d_name, ".idx"))
919 if (len + namelen + 1 > sizeof(path))
922 /* Don't reopen a pack we already have. */
923 strcpy(path + len, de->d_name);
924 for (p = packed_git; p; p = p->next) {
925 if (!memcmp(path, p->pack_name, len + namelen - 4))
930 /* See if it really is a valid .idx file with corresponding
931 * .pack file that we can map.
933 p = add_packed_git(path, len + namelen, local);
936 install_packed_git(p);
941 static int sort_pack(const void *a_, const void *b_)
943 struct packed_git *a = *((struct packed_git **)a_);
944 struct packed_git *b = *((struct packed_git **)b_);
948 * Local packs tend to contain objects specific to our
949 * variant of the project than remote ones. In addition,
950 * remote ones could be on a network mounted filesystem.
951 * Favor local ones for these reasons.
953 st = a->pack_local - b->pack_local;
958 * Younger packs tend to contain more recent objects,
959 * and more recent objects tend to get accessed more
962 if (a->mtime < b->mtime)
964 else if (a->mtime == b->mtime)
969 static void rearrange_packed_git(void)
971 struct packed_git **ary, *p;
974 for (n = 0, p = packed_git; p; p = p->next)
979 /* prepare an array of packed_git for easier sorting */
980 ary = xcalloc(n, sizeof(struct packed_git *));
981 for (n = 0, p = packed_git; p; p = p->next)
984 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
986 /* link them back again */
987 for (i = 0; i < n - 1; i++)
988 ary[i]->next = ary[i + 1];
989 ary[n - 1]->next = NULL;
995 static int prepare_packed_git_run_once = 0;
996 void prepare_packed_git(void)
998 struct alternate_object_database *alt;
1000 if (prepare_packed_git_run_once)
1002 prepare_packed_git_one(get_object_directory(), 1);
1004 for (alt = alt_odb_list; alt; alt = alt->next) {
1006 prepare_packed_git_one(alt->base, 0);
1007 alt->name[-1] = '/';
1009 rearrange_packed_git();
1010 prepare_packed_git_run_once = 1;
1013 void reprepare_packed_git(void)
1016 prepare_packed_git_run_once = 0;
1017 prepare_packed_git();
1020 static void mark_bad_packed_object(struct packed_git *p,
1021 const unsigned char *sha1)
1024 for (i = 0; i < p->num_bad_objects; i++)
1025 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1027 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1028 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1029 p->num_bad_objects++;
1032 static int has_packed_and_bad(const unsigned char *sha1)
1034 struct packed_git *p;
1037 for (p = packed_git; p; p = p->next)
1038 for (i = 0; i < p->num_bad_objects; i++)
1039 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1044 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1046 unsigned char real_sha1[20];
1047 hash_sha1_file(map, size, type, real_sha1);
1048 return hashcmp(sha1, real_sha1) ? -1 : 0;
1051 static int git_open_noatime(const char *name)
1053 static int sha1_file_open_flag = O_NOATIME;
1054 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1056 /* Might the failure be due to O_NOATIME? */
1057 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1058 fd = open(name, O_RDONLY);
1060 sha1_file_open_flag = 0;
1065 static int open_sha1_file(const unsigned char *sha1)
1068 char *name = sha1_file_name(sha1);
1069 struct alternate_object_database *alt;
1071 fd = git_open_noatime(name);
1077 for (alt = alt_odb_list; alt; alt = alt->next) {
1079 fill_sha1_path(name, sha1);
1080 fd = git_open_noatime(alt->base);
1087 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1092 fd = open_sha1_file(sha1);
1097 if (!fstat(fd, &st)) {
1098 *size = xsize_t(st.st_size);
1099 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1106 static int legacy_loose_object(unsigned char *map)
1111 * Is it a zlib-compressed buffer? If so, the first byte
1112 * must be 0x78 (15-bit window size, deflated), and the
1113 * first 16-bit word is evenly divisible by 31
1115 word = (map[0] << 8) + map[1];
1116 if (map[0] == 0x78 && !(word % 31))
1122 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1123 unsigned long len, enum object_type *type, unsigned long *sizep)
1126 unsigned long size, c;
1127 unsigned long used = 0;
1130 *type = (c >> 4) & 7;
1134 if (len <= used || bitsizeof(long) <= shift) {
1135 error("bad object header");
1139 size += (c & 0x7f) << shift;
1146 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1148 unsigned long size, used;
1149 static const char valid_loose_object_type[8] = {
1151 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1152 0, /* "delta" and others are invalid in a loose object */
1154 enum object_type type;
1156 /* Get the data stream */
1157 memset(stream, 0, sizeof(*stream));
1158 stream->next_in = map;
1159 stream->avail_in = mapsize;
1160 stream->next_out = buffer;
1161 stream->avail_out = bufsiz;
1163 if (legacy_loose_object(map)) {
1164 git_inflate_init(stream);
1165 return git_inflate(stream, 0);
1170 * There used to be a second loose object header format which
1171 * was meant to mimic the in-pack format, allowing for direct
1172 * copy of the object data. This format turned up not to be
1173 * really worth it and we don't write it any longer. But we
1174 * can still read it.
1176 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1177 if (!used || !valid_loose_object_type[type])
1182 /* Set up the stream for the rest.. */
1183 stream->next_in = map;
1184 stream->avail_in = mapsize;
1185 git_inflate_init(stream);
1187 /* And generate the fake traditional header */
1188 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1189 typename(type), size);
1193 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1195 int bytes = strlen(buffer) + 1;
1196 unsigned char *buf = xmallocz(size);
1200 n = stream->total_out - bytes;
1203 memcpy(buf, (char *) buffer + bytes, n);
1205 if (bytes <= size) {
1207 * The above condition must be (bytes <= size), not
1208 * (bytes < size). In other words, even though we
1209 * expect no more output and set avail_out to zer0,
1210 * the input zlib stream may have bytes that express
1211 * "this concludes the stream", and we *do* want to
1214 * Otherwise we would not be able to test that we
1215 * consumed all the input to reach the expected size;
1216 * we also want to check that zlib tells us that all
1217 * went well with status == Z_STREAM_END at the end.
1219 stream->next_out = buf + bytes;
1220 stream->avail_out = size - bytes;
1221 while (status == Z_OK)
1222 status = git_inflate(stream, Z_FINISH);
1224 if (status == Z_STREAM_END && !stream->avail_in) {
1225 git_inflate_end(stream);
1230 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1231 else if (stream->avail_in)
1232 error("garbage at end of loose object '%s'",
1239 * We used to just use "sscanf()", but that's actually way
1240 * too permissive for what we want to check. So do an anal
1241 * object header parse by hand.
1243 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1250 * The type can be at most ten bytes (including the
1251 * terminating '\0' that we add), and is followed by
1260 if (i >= sizeof(type))
1266 * The length must follow immediately, and be in canonical
1267 * decimal format (ie "010" is not valid).
1269 size = *hdr++ - '0';
1274 unsigned long c = *hdr - '0';
1278 size = size * 10 + c;
1284 * The length must be followed by a zero byte
1286 return *hdr ? -1 : type_from_string(type);
1289 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1295 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1296 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1299 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1302 unsigned long get_size_from_delta(struct packed_git *p,
1303 struct pack_window **w_curs,
1306 const unsigned char *data;
1307 unsigned char delta_head[20], *in;
1311 memset(&stream, 0, sizeof(stream));
1312 stream.next_out = delta_head;
1313 stream.avail_out = sizeof(delta_head);
1315 git_inflate_init(&stream);
1317 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1318 stream.next_in = in;
1319 st = git_inflate(&stream, Z_FINISH);
1320 curpos += stream.next_in - in;
1321 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1322 stream.total_out < sizeof(delta_head));
1323 git_inflate_end(&stream);
1324 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1325 error("delta data unpack-initial failed");
1329 /* Examine the initial part of the delta to figure out
1334 /* ignore base size */
1335 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1337 /* Read the result size */
1338 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1341 static off_t get_delta_base(struct packed_git *p,
1342 struct pack_window **w_curs,
1344 enum object_type type,
1345 off_t delta_obj_offset)
1347 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1350 /* use_pack() assured us we have [base_info, base_info + 20)
1351 * as a range that we can look at without walking off the
1352 * end of the mapped window. Its actually the hash size
1353 * that is assured. An OFS_DELTA longer than the hash size
1354 * is stupid, as then a REF_DELTA would be smaller to store.
1356 if (type == OBJ_OFS_DELTA) {
1358 unsigned char c = base_info[used++];
1359 base_offset = c & 127;
1362 if (!base_offset || MSB(base_offset, 7))
1363 return 0; /* overflow */
1364 c = base_info[used++];
1365 base_offset = (base_offset << 7) + (c & 127);
1367 base_offset = delta_obj_offset - base_offset;
1368 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1369 return 0; /* out of bound */
1371 } else if (type == OBJ_REF_DELTA) {
1372 /* The base entry _must_ be in the same pack */
1373 base_offset = find_pack_entry_one(base_info, p);
1376 die("I am totally screwed");
1380 /* forward declaration for a mutually recursive function */
1381 static int packed_object_info(struct packed_git *p, off_t offset,
1382 unsigned long *sizep);
1384 static int packed_delta_info(struct packed_git *p,
1385 struct pack_window **w_curs,
1387 enum object_type type,
1389 unsigned long *sizep)
1393 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1396 type = packed_object_info(p, base_offset, NULL);
1397 if (type <= OBJ_NONE) {
1398 struct revindex_entry *revidx;
1399 const unsigned char *base_sha1;
1400 revidx = find_pack_revindex(p, base_offset);
1403 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1404 mark_bad_packed_object(p, base_sha1);
1405 type = sha1_object_info(base_sha1, NULL);
1406 if (type <= OBJ_NONE)
1410 /* We choose to only get the type of the base object and
1411 * ignore potentially corrupt pack file that expects the delta
1412 * based on a base with a wrong size. This saves tons of
1416 *sizep = get_size_from_delta(p, w_curs, curpos);
1424 static int unpack_object_header(struct packed_git *p,
1425 struct pack_window **w_curs,
1427 unsigned long *sizep)
1429 unsigned char *base;
1432 enum object_type type;
1434 /* use_pack() assures us we have [base, base + 20) available
1435 * as a range that we can look at at. (Its actually the hash
1436 * size that is assured.) With our object header encoding
1437 * the maximum deflated object size is 2^137, which is just
1438 * insane, so we know won't exceed what we have been given.
1440 base = use_pack(p, w_curs, *curpos, &left);
1441 used = unpack_object_header_buffer(base, left, &type, sizep);
1450 const char *packed_object_info_detail(struct packed_git *p,
1452 unsigned long *size,
1453 unsigned long *store_size,
1454 unsigned int *delta_chain_length,
1455 unsigned char *base_sha1)
1457 struct pack_window *w_curs = NULL;
1459 unsigned long dummy;
1460 unsigned char *next_sha1;
1461 enum object_type type;
1462 struct revindex_entry *revidx;
1464 *delta_chain_length = 0;
1465 curpos = obj_offset;
1466 type = unpack_object_header(p, &w_curs, &curpos, size);
1468 revidx = find_pack_revindex(p, obj_offset);
1469 *store_size = revidx[1].offset - obj_offset;
1474 die("pack %s contains unknown object type %d",
1475 p->pack_name, type);
1480 unuse_pack(&w_curs);
1481 return typename(type);
1483 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1485 die("pack %s contains bad delta base reference of type %s",
1486 p->pack_name, typename(type));
1487 if (*delta_chain_length == 0) {
1488 revidx = find_pack_revindex(p, obj_offset);
1489 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1493 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1494 if (*delta_chain_length == 0)
1495 hashcpy(base_sha1, next_sha1);
1496 obj_offset = find_pack_entry_one(next_sha1, p);
1499 (*delta_chain_length)++;
1500 curpos = obj_offset;
1501 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1505 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1506 unsigned long *sizep)
1508 struct pack_window *w_curs = NULL;
1510 off_t curpos = obj_offset;
1511 enum object_type type;
1513 type = unpack_object_header(p, &w_curs, &curpos, &size);
1518 type = packed_delta_info(p, &w_curs, curpos,
1519 type, obj_offset, sizep);
1529 error("unknown object type %i at offset %"PRIuMAX" in %s",
1530 type, (uintmax_t)obj_offset, p->pack_name);
1533 unuse_pack(&w_curs);
1537 static void *unpack_compressed_entry(struct packed_git *p,
1538 struct pack_window **w_curs,
1544 unsigned char *buffer, *in;
1546 buffer = xmallocz(size);
1547 memset(&stream, 0, sizeof(stream));
1548 stream.next_out = buffer;
1549 stream.avail_out = size + 1;
1551 git_inflate_init(&stream);
1553 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1554 stream.next_in = in;
1555 st = git_inflate(&stream, Z_FINISH);
1556 if (!stream.avail_out)
1557 break; /* the payload is larger than it should be */
1558 curpos += stream.next_in - in;
1559 } while (st == Z_OK || st == Z_BUF_ERROR);
1560 git_inflate_end(&stream);
1561 if ((st != Z_STREAM_END) || stream.total_out != size) {
1569 #define MAX_DELTA_CACHE (256)
1571 static size_t delta_base_cached;
1573 static struct delta_base_cache_lru_list {
1574 struct delta_base_cache_lru_list *prev;
1575 struct delta_base_cache_lru_list *next;
1576 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1578 static struct delta_base_cache_entry {
1579 struct delta_base_cache_lru_list lru;
1581 struct packed_git *p;
1584 enum object_type type;
1585 } delta_base_cache[MAX_DELTA_CACHE];
1587 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1591 hash = (unsigned long)p + (unsigned long)base_offset;
1592 hash += (hash >> 8) + (hash >> 16);
1593 return hash % MAX_DELTA_CACHE;
1596 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1597 unsigned long *base_size, enum object_type *type, int keep_cache)
1600 unsigned long hash = pack_entry_hash(p, base_offset);
1601 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1604 if (!ret || ent->p != p || ent->base_offset != base_offset)
1605 return unpack_entry(p, base_offset, type, base_size);
1609 ent->lru.next->prev = ent->lru.prev;
1610 ent->lru.prev->next = ent->lru.next;
1611 delta_base_cached -= ent->size;
1613 ret = xmemdupz(ent->data, ent->size);
1616 *base_size = ent->size;
1620 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1625 ent->lru.next->prev = ent->lru.prev;
1626 ent->lru.prev->next = ent->lru.next;
1627 delta_base_cached -= ent->size;
1631 void clear_delta_base_cache(void)
1634 for (p = 0; p < MAX_DELTA_CACHE; p++)
1635 release_delta_base_cache(&delta_base_cache[p]);
1638 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1639 void *base, unsigned long base_size, enum object_type type)
1641 unsigned long hash = pack_entry_hash(p, base_offset);
1642 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1643 struct delta_base_cache_lru_list *lru;
1645 release_delta_base_cache(ent);
1646 delta_base_cached += base_size;
1648 for (lru = delta_base_cache_lru.next;
1649 delta_base_cached > delta_base_cache_limit
1650 && lru != &delta_base_cache_lru;
1652 struct delta_base_cache_entry *f = (void *)lru;
1653 if (f->type == OBJ_BLOB)
1654 release_delta_base_cache(f);
1656 for (lru = delta_base_cache_lru.next;
1657 delta_base_cached > delta_base_cache_limit
1658 && lru != &delta_base_cache_lru;
1660 struct delta_base_cache_entry *f = (void *)lru;
1661 release_delta_base_cache(f);
1665 ent->base_offset = base_offset;
1668 ent->size = base_size;
1669 ent->lru.next = &delta_base_cache_lru;
1670 ent->lru.prev = delta_base_cache_lru.prev;
1671 delta_base_cache_lru.prev->next = &ent->lru;
1672 delta_base_cache_lru.prev = &ent->lru;
1675 static void *read_object(const unsigned char *sha1, enum object_type *type,
1676 unsigned long *size);
1678 static void *unpack_delta_entry(struct packed_git *p,
1679 struct pack_window **w_curs,
1681 unsigned long delta_size,
1683 enum object_type *type,
1684 unsigned long *sizep)
1686 void *delta_data, *result, *base;
1687 unsigned long base_size;
1690 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1692 error("failed to validate delta base reference "
1693 "at offset %"PRIuMAX" from %s",
1694 (uintmax_t)curpos, p->pack_name);
1698 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1701 * We're probably in deep shit, but let's try to fetch
1702 * the required base anyway from another pack or loose.
1703 * This is costly but should happen only in the presence
1704 * of a corrupted pack, and is better than failing outright.
1706 struct revindex_entry *revidx;
1707 const unsigned char *base_sha1;
1708 revidx = find_pack_revindex(p, base_offset);
1711 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1712 error("failed to read delta base object %s"
1713 " at offset %"PRIuMAX" from %s",
1714 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1716 mark_bad_packed_object(p, base_sha1);
1717 base = read_object(base_sha1, type, &base_size);
1722 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1724 error("failed to unpack compressed delta "
1725 "at offset %"PRIuMAX" from %s",
1726 (uintmax_t)curpos, p->pack_name);
1730 result = patch_delta(base, base_size,
1731 delta_data, delta_size,
1734 die("failed to apply delta");
1736 add_delta_base_cache(p, base_offset, base, base_size, *type);
1740 int do_check_packed_object_crc;
1742 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1743 enum object_type *type, unsigned long *sizep)
1745 struct pack_window *w_curs = NULL;
1746 off_t curpos = obj_offset;
1749 if (do_check_packed_object_crc && p->index_version > 1) {
1750 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1751 unsigned long len = revidx[1].offset - obj_offset;
1752 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1753 const unsigned char *sha1 =
1754 nth_packed_object_sha1(p, revidx->nr);
1755 error("bad packed object CRC for %s",
1757 mark_bad_packed_object(p, sha1);
1758 unuse_pack(&w_curs);
1763 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1767 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1768 obj_offset, type, sizep);
1774 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1778 error("unknown object type %i at offset %"PRIuMAX" in %s",
1779 *type, (uintmax_t)obj_offset, p->pack_name);
1781 unuse_pack(&w_curs);
1785 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1788 const unsigned char *index = p->index_data;
1790 if (open_pack_index(p))
1792 index = p->index_data;
1794 if (n >= p->num_objects)
1797 if (p->index_version == 1) {
1798 return index + 24 * n + 4;
1801 return index + 20 * n;
1805 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1807 const unsigned char *index = p->index_data;
1809 if (p->index_version == 1) {
1810 return ntohl(*((uint32_t *)(index + 24 * n)));
1813 index += 8 + p->num_objects * (20 + 4);
1814 off = ntohl(*((uint32_t *)(index + 4 * n)));
1815 if (!(off & 0x80000000))
1817 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1818 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1819 ntohl(*((uint32_t *)(index + 4)));
1823 off_t find_pack_entry_one(const unsigned char *sha1,
1824 struct packed_git *p)
1826 const uint32_t *level1_ofs = p->index_data;
1827 const unsigned char *index = p->index_data;
1828 unsigned hi, lo, stride;
1829 static int use_lookup = -1;
1830 static int debug_lookup = -1;
1832 if (debug_lookup < 0)
1833 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1836 if (open_pack_index(p))
1838 level1_ofs = p->index_data;
1839 index = p->index_data;
1841 if (p->index_version > 1) {
1846 hi = ntohl(level1_ofs[*sha1]);
1847 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1848 if (p->index_version > 1) {
1856 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1857 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1860 use_lookup = !!getenv("GIT_USE_LOOKUP");
1862 int pos = sha1_entry_pos(index, stride, 0,
1863 lo, hi, p->num_objects, sha1);
1866 return nth_packed_object_offset(p, pos);
1870 unsigned mi = (lo + hi) / 2;
1871 int cmp = hashcmp(index + mi * stride, sha1);
1874 printf("lo %u hi %u rg %u mi %u\n",
1875 lo, hi, hi - lo, mi);
1877 return nth_packed_object_offset(p, mi);
1886 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1888 static struct packed_git *last_found = (void *)1;
1889 struct packed_git *p;
1892 prepare_packed_git();
1895 p = (last_found == (void *)1) ? packed_git : last_found;
1898 if (p->num_bad_objects) {
1900 for (i = 0; i < p->num_bad_objects; i++)
1901 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1905 offset = find_pack_entry_one(sha1, p);
1908 * We are about to tell the caller where they can
1909 * locate the requested object. We better make
1910 * sure the packfile is still here and can be
1911 * accessed before supplying that answer, as
1912 * it may have been deleted since the index
1915 if (p->pack_fd == -1 && open_packed_git(p)) {
1916 error("packfile %s cannot be accessed", p->pack_name);
1921 hashcpy(e->sha1, sha1);
1927 if (p == last_found)
1931 if (p == last_found)
1937 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1938 struct packed_git *packs)
1940 struct packed_git *p;
1942 for (p = packs; p; p = p->next) {
1943 if (find_pack_entry_one(sha1, p))
1950 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1953 unsigned long mapsize, size;
1958 map = map_sha1_file(sha1, &mapsize);
1960 return error("unable to find %s", sha1_to_hex(sha1));
1961 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1962 status = error("unable to unpack %s header",
1964 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1965 status = error("unable to parse %s header", sha1_to_hex(sha1));
1968 git_inflate_end(&stream);
1969 munmap(map, mapsize);
1973 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1975 struct pack_entry e;
1978 if (!find_pack_entry(sha1, &e)) {
1979 /* Most likely it's a loose object. */
1980 status = sha1_loose_object_info(sha1, sizep);
1984 /* Not a loose object; someone else may have just packed it. */
1985 reprepare_packed_git();
1986 if (!find_pack_entry(sha1, &e))
1990 status = packed_object_info(e.p, e.offset, sizep);
1992 mark_bad_packed_object(e.p, sha1);
1993 status = sha1_object_info(sha1, sizep);
1999 static void *read_packed_sha1(const unsigned char *sha1,
2000 enum object_type *type, unsigned long *size)
2002 struct pack_entry e;
2005 if (!find_pack_entry(sha1, &e))
2007 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2010 * We're probably in deep shit, but let's try to fetch
2011 * the required object anyway from another pack or loose.
2012 * This should happen only in the presence of a corrupted
2013 * pack, and is better than failing outright.
2015 error("failed to read object %s at offset %"PRIuMAX" from %s",
2016 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2017 mark_bad_packed_object(e.p, sha1);
2018 data = read_object(sha1, type, size);
2024 * This is meant to hold a *small* number of objects that you would
2025 * want read_sha1_file() to be able to return, but yet you do not want
2026 * to write them into the object store (e.g. a browse-only
2029 static struct cached_object {
2030 unsigned char sha1[20];
2031 enum object_type type;
2035 static int cached_object_nr, cached_object_alloc;
2037 static struct cached_object empty_tree = {
2038 EMPTY_TREE_SHA1_BIN,
2044 static struct cached_object *find_cached_object(const unsigned char *sha1)
2047 struct cached_object *co = cached_objects;
2049 for (i = 0; i < cached_object_nr; i++, co++) {
2050 if (!hashcmp(co->sha1, sha1))
2053 if (!hashcmp(sha1, empty_tree.sha1))
2058 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2059 unsigned char *sha1)
2061 struct cached_object *co;
2063 hash_sha1_file(buf, len, typename(type), sha1);
2064 if (has_sha1_file(sha1) || find_cached_object(sha1))
2066 if (cached_object_alloc <= cached_object_nr) {
2067 cached_object_alloc = alloc_nr(cached_object_alloc);
2068 cached_objects = xrealloc(cached_objects,
2069 sizeof(*cached_objects) *
2070 cached_object_alloc);
2072 co = &cached_objects[cached_object_nr++];
2075 co->buf = xmalloc(len);
2076 memcpy(co->buf, buf, len);
2077 hashcpy(co->sha1, sha1);
2081 static void *read_object(const unsigned char *sha1, enum object_type *type,
2082 unsigned long *size)
2084 unsigned long mapsize;
2086 struct cached_object *co;
2088 co = find_cached_object(sha1);
2092 return xmemdupz(co->buf, co->size);
2095 buf = read_packed_sha1(sha1, type, size);
2098 map = map_sha1_file(sha1, &mapsize);
2100 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2101 munmap(map, mapsize);
2104 reprepare_packed_git();
2105 return read_packed_sha1(sha1, type, size);
2108 void *read_sha1_file_repl(const unsigned char *sha1,
2109 enum object_type *type,
2110 unsigned long *size,
2111 const unsigned char **replacement)
2113 const unsigned char *repl = lookup_replace_object(sha1);
2114 void *data = read_object(repl, type, size);
2117 /* die if we replaced an object with one that does not exist */
2118 if (!data && repl != sha1)
2119 die("replacement %s not found for %s",
2120 sha1_to_hex(repl), sha1_to_hex(sha1));
2122 /* legacy behavior is to die on corrupted objects */
2124 if (has_loose_object(repl)) {
2125 path = sha1_file_name(sha1);
2126 die("loose object %s (stored in %s) is corrupted", sha1_to_hex(repl), path);
2128 if (has_packed_and_bad(repl)) {
2129 path = sha1_pack_name(sha1);
2130 die("packed object %s (stored in %s) is corrupted", sha1_to_hex(repl), path);
2135 *replacement = repl;
2140 void *read_object_with_reference(const unsigned char *sha1,
2141 const char *required_type_name,
2142 unsigned long *size,
2143 unsigned char *actual_sha1_return)
2145 enum object_type type, required_type;
2147 unsigned long isize;
2148 unsigned char actual_sha1[20];
2150 required_type = type_from_string(required_type_name);
2151 hashcpy(actual_sha1, sha1);
2153 int ref_length = -1;
2154 const char *ref_type = NULL;
2156 buffer = read_sha1_file(actual_sha1, &type, &isize);
2159 if (type == required_type) {
2161 if (actual_sha1_return)
2162 hashcpy(actual_sha1_return, actual_sha1);
2165 /* Handle references */
2166 else if (type == OBJ_COMMIT)
2168 else if (type == OBJ_TAG)
2169 ref_type = "object ";
2174 ref_length = strlen(ref_type);
2176 if (ref_length + 40 > isize ||
2177 memcmp(buffer, ref_type, ref_length) ||
2178 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2183 /* Now we have the ID of the referred-to object in
2184 * actual_sha1. Check again. */
2188 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2189 const char *type, unsigned char *sha1,
2190 char *hdr, int *hdrlen)
2194 /* Generate the header */
2195 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2199 git_SHA1_Update(&c, hdr, *hdrlen);
2200 git_SHA1_Update(&c, buf, len);
2201 git_SHA1_Final(sha1, &c);
2205 * Move the just written object into its final resting place.
2206 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2207 * "moving" is only a part of what it does, when no patch between
2208 * master to pu changes the call sites of this function.
2210 int move_temp_to_file(const char *tmpfile, const char *filename)
2214 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2216 else if (link(tmpfile, filename))
2220 * Coda hack - coda doesn't like cross-directory links,
2221 * so we fall back to a rename, which will mean that it
2222 * won't be able to check collisions, but that's not a
2225 * The same holds for FAT formatted media.
2227 * When this succeeds, we just return. We have nothing
2230 if (ret && ret != EEXIST) {
2232 if (!rename(tmpfile, filename))
2236 unlink_or_warn(tmpfile);
2238 if (ret != EEXIST) {
2239 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2241 /* FIXME!!! Collision check here ? */
2245 if (adjust_shared_perm(filename))
2246 return error("unable to set permission to '%s'", filename);
2250 static int write_buffer(int fd, const void *buf, size_t len)
2252 if (write_in_full(fd, buf, len) < 0)
2253 return error("file write error (%s)", strerror(errno));
2257 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2258 unsigned char *sha1)
2262 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2266 /* Finalize a file on disk, and close it. */
2267 static void close_sha1_file(int fd)
2269 if (fsync_object_files)
2270 fsync_or_die(fd, "sha1 file");
2272 die_errno("error when closing sha1 file");
2275 /* Size of directory component, including the ending '/' */
2276 static inline int directory_size(const char *filename)
2278 const char *s = strrchr(filename, '/');
2281 return s - filename + 1;
2285 * This creates a temporary file in the same directory as the final
2288 * We want to avoid cross-directory filename renames, because those
2289 * can have problems on various filesystems (FAT, NFS, Coda).
2291 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2293 int fd, dirlen = directory_size(filename);
2295 if (dirlen + 20 > bufsiz) {
2296 errno = ENAMETOOLONG;
2299 memcpy(buffer, filename, dirlen);
2300 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2301 fd = git_mkstemp_mode(buffer, 0444);
2302 if (fd < 0 && dirlen && errno == ENOENT) {
2303 /* Make sure the directory exists */
2304 memcpy(buffer, filename, dirlen);
2305 buffer[dirlen-1] = 0;
2306 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2310 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2311 fd = git_mkstemp_mode(buffer, 0444);
2316 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2317 const void *buf, unsigned long len, time_t mtime)
2320 unsigned char compressed[4096];
2323 unsigned char parano_sha1[20];
2325 static char tmpfile[PATH_MAX];
2327 filename = sha1_file_name(sha1);
2328 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2329 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2330 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2332 if (errno == EACCES)
2333 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2335 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2339 memset(&stream, 0, sizeof(stream));
2340 deflateInit(&stream, zlib_compression_level);
2341 stream.next_out = compressed;
2342 stream.avail_out = sizeof(compressed);
2345 /* First header.. */
2346 stream.next_in = (unsigned char *)hdr;
2347 stream.avail_in = hdrlen;
2348 while (deflate(&stream, 0) == Z_OK)
2350 git_SHA1_Update(&c, hdr, hdrlen);
2352 /* Then the data itself.. */
2353 stream.next_in = (void *)buf;
2354 stream.avail_in = len;
2356 unsigned char *in0 = stream.next_in;
2357 ret = deflate(&stream, Z_FINISH);
2358 git_SHA1_Update(&c, in0, stream.next_in - in0);
2359 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2360 die("unable to write sha1 file");
2361 stream.next_out = compressed;
2362 stream.avail_out = sizeof(compressed);
2363 } while (ret == Z_OK);
2365 if (ret != Z_STREAM_END)
2366 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2367 ret = deflateEnd(&stream);
2369 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2370 git_SHA1_Final(parano_sha1, &c);
2371 if (hashcmp(sha1, parano_sha1) != 0)
2372 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2374 close_sha1_file(fd);
2379 utb.modtime = mtime;
2380 if (utime(tmpfile, &utb) < 0)
2381 warning("failed utime() on %s: %s",
2382 tmpfile, strerror(errno));
2385 return move_temp_to_file(tmpfile, filename);
2388 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2390 unsigned char sha1[20];
2394 /* Normally if we have it in the pack then we do not bother writing
2395 * it out into .git/objects/??/?{38} file.
2397 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2399 hashcpy(returnsha1, sha1);
2400 if (has_sha1_file(sha1))
2402 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2405 int force_object_loose(const unsigned char *sha1, time_t mtime)
2409 enum object_type type;
2414 if (has_loose_object(sha1))
2416 buf = read_packed_sha1(sha1, &type, &len);
2418 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2419 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2420 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2426 int has_pack_index(const unsigned char *sha1)
2429 if (stat(sha1_pack_index_name(sha1), &st))
2434 int has_sha1_pack(const unsigned char *sha1)
2436 struct pack_entry e;
2437 return find_pack_entry(sha1, &e);
2440 int has_sha1_file(const unsigned char *sha1)
2442 struct pack_entry e;
2444 if (find_pack_entry(sha1, &e))
2446 return has_loose_object(sha1);
2449 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2450 int write_object, enum object_type type, const char *path)
2452 int ret, re_allocated = 0;
2458 * Convert blobs to git internal format
2460 if ((type == OBJ_BLOB) && path) {
2461 struct strbuf nbuf = STRBUF_INIT;
2462 if (convert_to_git(path, buf, size, &nbuf,
2463 write_object ? safe_crlf : 0)) {
2464 buf = strbuf_detach(&nbuf, &size);
2470 ret = write_sha1_file(buf, size, typename(type), sha1);
2472 ret = hash_sha1_file(buf, size, typename(type), sha1);
2478 #define SMALL_FILE_SIZE (32*1024)
2480 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2481 enum object_type type, const char *path)
2484 size_t size = xsize_t(st->st_size);
2486 if (!S_ISREG(st->st_mode)) {
2487 struct strbuf sbuf = STRBUF_INIT;
2488 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2489 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2493 strbuf_release(&sbuf);
2495 ret = index_mem(sha1, NULL, size, write_object, type, path);
2496 } else if (size <= SMALL_FILE_SIZE) {
2497 char *buf = xmalloc(size);
2498 if (size == read_in_full(fd, buf, size))
2499 ret = index_mem(sha1, buf, size, write_object, type,
2502 ret = error("short read %s", strerror(errno));
2505 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2506 ret = index_mem(sha1, buf, size, write_object, type, path);
2513 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2516 struct strbuf sb = STRBUF_INIT;
2518 switch (st->st_mode & S_IFMT) {
2520 fd = open(path, O_RDONLY);
2522 return error("open(\"%s\"): %s", path,
2524 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2525 return error("%s: failed to insert into database",
2529 if (strbuf_readlink(&sb, path, st->st_size)) {
2530 char *errstr = strerror(errno);
2531 return error("readlink(\"%s\"): %s", path,
2535 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2536 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2537 return error("%s: failed to insert into database",
2539 strbuf_release(&sb);
2542 return resolve_gitlink_ref(path, "HEAD", sha1);
2544 return error("%s: unsupported file type", path);
2549 int read_pack_header(int fd, struct pack_header *header)
2551 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2552 /* "eof before pack header was fully read" */
2553 return PH_ERROR_EOF;
2555 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2556 /* "protocol error (pack signature mismatch detected)" */
2557 return PH_ERROR_PACK_SIGNATURE;
2558 if (!pack_version_ok(header->hdr_version))
2559 /* "protocol error (pack version unsupported)" */
2560 return PH_ERROR_PROTOCOL;
2564 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2566 enum object_type type = sha1_object_info(sha1, NULL);
2568 die("%s is not a valid object", sha1_to_hex(sha1));
2570 die("%s is not a valid '%s' object", sha1_to_hex(sha1),