2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
14 #include "run-command.h"
17 #include "tree-walk.h"
19 #include "pack-revindex.h"
20 #include "sha1-lookup.h"
21 #include "bulk-checkin.h"
24 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
25 #define O_NOATIME 01000000
31 #define SZ_FMT PRIuMAX
32 static inline uintmax_t sz_fmt(size_t s) { return s; }
34 const unsigned char null_sha1[20];
37 * This is meant to hold a *small* number of objects that you would
38 * want read_sha1_file() to be able to return, but yet you do not want
39 * to write them into the object store (e.g. a browse-only
42 static struct cached_object {
43 unsigned char sha1[20];
44 enum object_type type;
48 static int cached_object_nr, cached_object_alloc;
50 static struct cached_object empty_tree = {
51 EMPTY_TREE_SHA1_BIN_LITERAL,
57 static struct packed_git *last_found_pack;
59 static struct cached_object *find_cached_object(const unsigned char *sha1)
62 struct cached_object *co = cached_objects;
64 for (i = 0; i < cached_object_nr; i++, co++) {
65 if (!hashcmp(co->sha1, sha1))
68 if (!hashcmp(sha1, empty_tree.sha1))
73 int mkdir_in_gitdir(const char *path)
75 if (mkdir(path, 0777)) {
76 int saved_errno = errno;
78 struct strbuf sb = STRBUF_INIT;
83 * Are we looking at a path in a symlinked worktree
84 * whose original repository does not yet have it?
85 * e.g. .git/rr-cache pointing at its original
86 * repository in which the user hasn't performed any
87 * conflict resolution yet?
89 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
90 strbuf_readlink(&sb, path, st.st_size) ||
91 !is_absolute_path(sb.buf) ||
92 mkdir(sb.buf, 0777)) {
99 return adjust_shared_perm(path);
102 int safe_create_leading_directories(char *path)
104 char *pos = path + offset_1st_component(path);
108 pos = strchr(pos, '/');
111 while (*++pos == '/')
116 if (!stat(path, &st)) {
118 if (!S_ISDIR(st.st_mode)) {
123 else if (mkdir(path, 0777)) {
127 else if (adjust_shared_perm(path)) {
136 int safe_create_leading_directories_const(const char *path)
138 /* path points to cache entries, so xstrdup before messing with it */
139 char *buf = xstrdup(path);
140 int result = safe_create_leading_directories(buf);
145 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
148 for (i = 0; i < 20; i++) {
149 static char hex[] = "0123456789abcdef";
150 unsigned int val = sha1[i];
151 char *pos = pathbuf + i*2 + (i > 0);
152 *pos++ = hex[val >> 4];
153 *pos = hex[val & 0xf];
158 * NOTE! This returns a statically allocated buffer, so you have to be
159 * careful about using it. Do an "xstrdup()" if you need to save the
162 * Also note that this returns the location for creating. Reading
163 * SHA1 file can happen from any alternate directory listed in the
164 * DB_ENVIRONMENT environment variable if it is not found in
165 * the primary object database.
167 char *sha1_file_name(const unsigned char *sha1)
169 static char buf[PATH_MAX];
173 objdir = get_object_directory();
174 len = strlen(objdir);
176 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
177 if (len + 43 > PATH_MAX)
178 die("insanely long object directory %s", objdir);
179 memcpy(buf, objdir, len);
183 fill_sha1_path(buf + len + 1, sha1);
187 static char *sha1_get_pack_name(const unsigned char *sha1,
188 char **name, char **base, const char *which)
190 static const char hex[] = "0123456789abcdef";
195 const char *sha1_file_directory = get_object_directory();
196 int len = strlen(sha1_file_directory);
197 *base = xmalloc(len + 60);
198 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
199 sha1_file_directory, which);
200 *name = *base + len + 11;
205 for (i = 0; i < 20; i++) {
206 unsigned int val = *sha1++;
207 *buf++ = hex[val >> 4];
208 *buf++ = hex[val & 0xf];
214 char *sha1_pack_name(const unsigned char *sha1)
216 static char *name, *base;
218 return sha1_get_pack_name(sha1, &name, &base, "pack");
221 char *sha1_pack_index_name(const unsigned char *sha1)
223 static char *name, *base;
225 return sha1_get_pack_name(sha1, &name, &base, "idx");
228 struct alternate_object_database *alt_odb_list;
229 static struct alternate_object_database **alt_odb_tail;
231 static int git_open_noatime(const char *name);
234 * Prepare alternate object database registry.
236 * The variable alt_odb_list points at the list of struct
237 * alternate_object_database. The elements on this list come from
238 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
239 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
240 * whose contents is similar to that environment variable but can be
241 * LF separated. Its base points at a statically allocated buffer that
242 * contains "/the/directory/corresponding/to/.git/objects/...", while
243 * its name points just after the slash at the end of ".git/objects/"
244 * in the example above, and has enough space to hold 40-byte hex
245 * SHA1, an extra slash for the first level indirection, and the
248 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
250 const char *objdir = get_object_directory();
251 struct alternate_object_database *ent;
252 struct alternate_object_database *alt;
254 struct strbuf pathbuf = STRBUF_INIT;
256 if (!is_absolute_path(entry) && relative_base) {
257 strbuf_addstr(&pathbuf, real_path(relative_base));
258 strbuf_addch(&pathbuf, '/');
260 strbuf_add(&pathbuf, entry, len);
262 normalize_path_copy(pathbuf.buf, pathbuf.buf);
264 pfxlen = strlen(pathbuf.buf);
267 * The trailing slash after the directory name is given by
268 * this function at the end. Remove duplicates.
270 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
274 ent = xmalloc(sizeof(*ent) + entlen);
275 memcpy(ent->base, pathbuf.buf, pfxlen);
276 strbuf_release(&pathbuf);
278 ent->name = ent->base + pfxlen + 1;
279 ent->base[pfxlen + 3] = '/';
280 ent->base[pfxlen] = ent->base[entlen-1] = 0;
282 /* Detect cases where alternate disappeared */
283 if (!is_directory(ent->base)) {
284 error("object directory %s does not exist; "
285 "check .git/objects/info/alternates.",
291 /* Prevent the common mistake of listing the same
292 * thing twice, or object directory itself.
294 for (alt = alt_odb_list; alt; alt = alt->next) {
295 if (!memcmp(ent->base, alt->base, pfxlen)) {
300 if (!memcmp(ent->base, objdir, pfxlen)) {
305 /* add the alternate entry */
307 alt_odb_tail = &(ent->next);
310 /* recursively add alternates */
311 read_info_alternates(ent->base, depth + 1);
313 ent->base[pfxlen] = '/';
318 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
319 const char *relative_base, int depth)
321 const char *cp, *last;
324 error("%s: ignoring alternate object stores, nesting too deep.",
332 if (cp < ep && *cp == '#') {
333 while (cp < ep && *cp != sep)
338 while (cp < ep && *cp != sep)
341 if (!is_absolute_path(last) && depth) {
342 error("%s: ignoring relative alternate object store %s",
343 relative_base, last);
345 link_alt_odb_entry(last, cp - last,
346 relative_base, depth);
349 while (cp < ep && *cp == sep)
355 void read_info_alternates(const char * relative_base, int depth)
360 const char alt_file_name[] = "info/alternates";
361 /* Given that relative_base is no longer than PATH_MAX,
362 ensure that "path" has enough space to append "/", the
363 file name, "info/alternates", and a trailing NUL. */
364 char path[PATH_MAX + 1 + sizeof alt_file_name];
367 sprintf(path, "%s/%s", relative_base, alt_file_name);
368 fd = git_open_noatime(path);
371 if (fstat(fd, &st) || (st.st_size == 0)) {
375 mapsz = xsize_t(st.st_size);
376 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
379 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
384 void add_to_alternates_file(const char *reference)
386 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
387 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
388 char *alt = mkpath("%s\n", reference);
389 write_or_die(fd, alt, strlen(alt));
390 if (commit_lock_file(lock))
391 die("could not close alternates file");
393 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
396 void foreach_alt_odb(alt_odb_fn fn, void *cb)
398 struct alternate_object_database *ent;
401 for (ent = alt_odb_list; ent; ent = ent->next)
406 void prepare_alt_odb(void)
413 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
416 alt_odb_tail = &alt_odb_list;
417 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
419 read_info_alternates(get_object_directory(), 0);
422 static int has_loose_object_local(const unsigned char *sha1)
424 char *name = sha1_file_name(sha1);
425 return !access(name, F_OK);
428 int has_loose_object_nonlocal(const unsigned char *sha1)
430 struct alternate_object_database *alt;
432 for (alt = alt_odb_list; alt; alt = alt->next) {
433 fill_sha1_path(alt->name, sha1);
434 if (!access(alt->base, F_OK))
440 static int has_loose_object(const unsigned char *sha1)
442 return has_loose_object_local(sha1) ||
443 has_loose_object_nonlocal(sha1);
446 static unsigned int pack_used_ctr;
447 static unsigned int pack_mmap_calls;
448 static unsigned int peak_pack_open_windows;
449 static unsigned int pack_open_windows;
450 static unsigned int pack_open_fds;
451 static unsigned int pack_max_fds;
452 static size_t peak_pack_mapped;
453 static size_t pack_mapped;
454 struct packed_git *packed_git;
456 void pack_report(void)
459 "pack_report: getpagesize() = %10" SZ_FMT "\n"
460 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
461 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
462 sz_fmt(getpagesize()),
463 sz_fmt(packed_git_window_size),
464 sz_fmt(packed_git_limit));
466 "pack_report: pack_used_ctr = %10u\n"
467 "pack_report: pack_mmap_calls = %10u\n"
468 "pack_report: pack_open_windows = %10u / %10u\n"
469 "pack_report: pack_mapped = "
470 "%10" SZ_FMT " / %10" SZ_FMT "\n",
473 pack_open_windows, peak_pack_open_windows,
474 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
477 static int check_packed_git_idx(const char *path, struct packed_git *p)
480 struct pack_idx_header *hdr;
482 uint32_t version, nr, i, *index;
483 int fd = git_open_noatime(path);
488 if (fstat(fd, &st)) {
492 idx_size = xsize_t(st.st_size);
493 if (idx_size < 4 * 256 + 20 + 20) {
495 return error("index file %s is too small", path);
497 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
501 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
502 version = ntohl(hdr->idx_version);
503 if (version < 2 || version > 2) {
504 munmap(idx_map, idx_size);
505 return error("index file %s is version %"PRIu32
506 " and is not supported by this binary"
507 " (try upgrading GIT to a newer version)",
516 index += 2; /* skip index header */
517 for (i = 0; i < 256; i++) {
518 uint32_t n = ntohl(index[i]);
520 munmap(idx_map, idx_size);
521 return error("non-monotonic index %s", path);
529 * - 256 index entries 4 bytes each
530 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
531 * - 20-byte SHA1 of the packfile
532 * - 20-byte SHA1 file checksum
534 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
535 munmap(idx_map, idx_size);
536 return error("wrong index v1 file size in %s", path);
538 } else if (version == 2) {
541 * - 8 bytes of header
542 * - 256 index entries 4 bytes each
543 * - 20-byte sha1 entry * nr
544 * - 4-byte crc entry * nr
545 * - 4-byte offset entry * nr
546 * - 20-byte SHA1 of the packfile
547 * - 20-byte SHA1 file checksum
548 * And after the 4-byte offset table might be a
549 * variable sized table containing 8-byte entries
550 * for offsets larger than 2^31.
552 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
553 unsigned long max_size = min_size;
555 max_size += (nr - 1)*8;
556 if (idx_size < min_size || idx_size > max_size) {
557 munmap(idx_map, idx_size);
558 return error("wrong index v2 file size in %s", path);
560 if (idx_size != min_size &&
562 * make sure we can deal with large pack offsets.
563 * 31-bit signed offset won't be enough, neither
564 * 32-bit unsigned one will be.
566 (sizeof(off_t) <= 4)) {
567 munmap(idx_map, idx_size);
568 return error("pack too large for current definition of off_t in %s", path);
572 p->index_version = version;
573 p->index_data = idx_map;
574 p->index_size = idx_size;
579 int open_pack_index(struct packed_git *p)
587 idx_name = xstrdup(p->pack_name);
588 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
589 ret = check_packed_git_idx(idx_name, p);
594 static void scan_windows(struct packed_git *p,
595 struct packed_git **lru_p,
596 struct pack_window **lru_w,
597 struct pack_window **lru_l)
599 struct pack_window *w, *w_l;
601 for (w_l = NULL, w = p->windows; w; w = w->next) {
603 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
613 static int unuse_one_window(struct packed_git *current, int keep_fd)
615 struct packed_git *p, *lru_p = NULL;
616 struct pack_window *lru_w = NULL, *lru_l = NULL;
619 scan_windows(current, &lru_p, &lru_w, &lru_l);
620 for (p = packed_git; p; p = p->next)
621 scan_windows(p, &lru_p, &lru_w, &lru_l);
623 munmap(lru_w->base, lru_w->len);
624 pack_mapped -= lru_w->len;
626 lru_l->next = lru_w->next;
628 lru_p->windows = lru_w->next;
629 if (!lru_p->windows && lru_p->pack_fd != -1
630 && lru_p->pack_fd != keep_fd) {
631 close(lru_p->pack_fd);
643 void release_pack_memory(size_t need, int fd)
645 size_t cur = pack_mapped;
646 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
650 void *xmmap(void *start, size_t length,
651 int prot, int flags, int fd, off_t offset)
653 void *ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED) {
657 release_pack_memory(length, fd);
658 ret = mmap(start, length, prot, flags, fd, offset);
659 if (ret == MAP_FAILED)
660 die_errno("Out of memory? mmap failed");
665 void close_pack_windows(struct packed_git *p)
668 struct pack_window *w = p->windows;
671 die("pack '%s' still has open windows to it",
673 munmap(w->base, w->len);
674 pack_mapped -= w->len;
676 p->windows = w->next;
681 void unuse_pack(struct pack_window **w_cursor)
683 struct pack_window *w = *w_cursor;
690 void close_pack_index(struct packed_git *p)
693 munmap((void *)p->index_data, p->index_size);
694 p->index_data = NULL;
699 * This is used by git-repack in case a newly created pack happens to
700 * contain the same set of objects as an existing one. In that case
701 * the resulting file might be different even if its name would be the
702 * same. It is best to close any reference to the old pack before it is
703 * replaced on disk. Of course no index pointers nor windows for given pack
704 * must subsist at this point. If ever objects from this pack are requested
705 * again, the new version of the pack will be reinitialized through
706 * reprepare_packed_git().
708 void free_pack_by_name(const char *pack_name)
710 struct packed_git *p, **pp = &packed_git;
714 if (strcmp(pack_name, p->pack_name) == 0) {
715 clear_delta_base_cache();
716 close_pack_windows(p);
717 if (p->pack_fd != -1) {
722 free(p->bad_object_sha1);
724 if (last_found_pack == p)
725 last_found_pack = NULL;
734 * Do not call this directly as this leaks p->pack_fd on error return;
735 * call open_packed_git() instead.
737 static int open_packed_git_1(struct packed_git *p)
740 struct pack_header hdr;
741 unsigned char sha1[20];
742 unsigned char *idx_sha1;
745 if (!p->index_data && open_pack_index(p))
746 return error("packfile %s index unavailable", p->pack_name);
750 unsigned int max_fds;
752 if (getrlimit(RLIMIT_NOFILE, &lim))
753 die_errno("cannot get RLIMIT_NOFILE");
755 max_fds = lim.rlim_cur;
757 /* Save 3 for stdin/stdout/stderr, 22 for work */
759 pack_max_fds = max_fds - 25;
764 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
767 p->pack_fd = git_open_noatime(p->pack_name);
768 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
772 /* If we created the struct before we had the pack we lack size. */
774 if (!S_ISREG(st.st_mode))
775 return error("packfile %s not a regular file", p->pack_name);
776 p->pack_size = st.st_size;
777 } else if (p->pack_size != st.st_size)
778 return error("packfile %s size changed", p->pack_name);
780 /* We leave these file descriptors open with sliding mmap;
781 * there is no point keeping them open across exec(), though.
783 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
785 return error("cannot determine file descriptor flags");
786 fd_flag |= FD_CLOEXEC;
787 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
788 return error("cannot set FD_CLOEXEC");
790 /* Verify we recognize this pack file format. */
791 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
792 return error("file %s is far too short to be a packfile", p->pack_name);
793 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
794 return error("file %s is not a GIT packfile", p->pack_name);
795 if (!pack_version_ok(hdr.hdr_version))
796 return error("packfile %s is version %"PRIu32" and not"
797 " supported (try upgrading GIT to a newer version)",
798 p->pack_name, ntohl(hdr.hdr_version));
800 /* Verify the pack matches its index. */
801 if (p->num_objects != ntohl(hdr.hdr_entries))
802 return error("packfile %s claims to have %"PRIu32" objects"
803 " while index indicates %"PRIu32" objects",
804 p->pack_name, ntohl(hdr.hdr_entries),
806 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
807 return error("end of packfile %s is unavailable", p->pack_name);
808 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
809 return error("packfile %s signature is unavailable", p->pack_name);
810 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
811 if (hashcmp(sha1, idx_sha1))
812 return error("packfile %s does not match index", p->pack_name);
816 static int open_packed_git(struct packed_git *p)
818 if (!open_packed_git_1(p))
820 if (p->pack_fd != -1) {
828 static int in_window(struct pack_window *win, off_t offset)
830 /* We must promise at least 20 bytes (one hash) after the
831 * offset is available from this window, otherwise the offset
832 * is not actually in this window and a different window (which
833 * has that one hash excess) must be used. This is to support
834 * the object header and delta base parsing routines below.
836 off_t win_off = win->offset;
837 return win_off <= offset
838 && (offset + 20) <= (win_off + win->len);
841 unsigned char *use_pack(struct packed_git *p,
842 struct pack_window **w_cursor,
846 struct pack_window *win = *w_cursor;
848 /* Since packfiles end in a hash of their content and it's
849 * pointless to ask for an offset into the middle of that
850 * hash, and the in_window function above wouldn't match
851 * don't allow an offset too close to the end of the file.
853 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
854 die("packfile %s cannot be accessed", p->pack_name);
855 if (offset > (p->pack_size - 20))
856 die("offset beyond end of packfile (truncated pack?)");
858 if (!win || !in_window(win, offset)) {
861 for (win = p->windows; win; win = win->next) {
862 if (in_window(win, offset))
866 size_t window_align = packed_git_window_size / 2;
869 if (p->pack_fd == -1 && open_packed_git(p))
870 die("packfile %s cannot be accessed", p->pack_name);
872 win = xcalloc(1, sizeof(*win));
873 win->offset = (offset / window_align) * window_align;
874 len = p->pack_size - win->offset;
875 if (len > packed_git_window_size)
876 len = packed_git_window_size;
877 win->len = (size_t)len;
878 pack_mapped += win->len;
879 while (packed_git_limit < pack_mapped
880 && unuse_one_window(p, p->pack_fd))
882 win->base = xmmap(NULL, win->len,
883 PROT_READ, MAP_PRIVATE,
884 p->pack_fd, win->offset);
885 if (win->base == MAP_FAILED)
886 die("packfile %s cannot be mapped: %s",
889 if (!win->offset && win->len == p->pack_size
890 && !p->do_not_close) {
897 if (pack_mapped > peak_pack_mapped)
898 peak_pack_mapped = pack_mapped;
899 if (pack_open_windows > peak_pack_open_windows)
900 peak_pack_open_windows = pack_open_windows;
901 win->next = p->windows;
905 if (win != *w_cursor) {
906 win->last_used = pack_used_ctr++;
910 offset -= win->offset;
912 *left = win->len - xsize_t(offset);
913 return win->base + offset;
916 static struct packed_git *alloc_packed_git(int extra)
918 struct packed_git *p = xmalloc(sizeof(*p) + extra);
919 memset(p, 0, sizeof(*p));
924 static void try_to_free_pack_memory(size_t size)
926 release_pack_memory(size, -1);
929 struct packed_git *add_packed_git(const char *path, int path_len, int local)
931 static int have_set_try_to_free_routine;
933 struct packed_git *p = alloc_packed_git(path_len + 2);
935 if (!have_set_try_to_free_routine) {
936 have_set_try_to_free_routine = 1;
937 set_try_to_free_routine(try_to_free_pack_memory);
941 * Make sure a corresponding .pack file exists and that
942 * the index looks sane.
944 path_len -= strlen(".idx");
949 memcpy(p->pack_name, path, path_len);
951 strcpy(p->pack_name + path_len, ".keep");
952 if (!access(p->pack_name, F_OK))
955 strcpy(p->pack_name + path_len, ".pack");
956 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
961 /* ok, it looks sane as far as we can check without
962 * actually mapping the pack file.
964 p->pack_size = st.st_size;
965 p->pack_local = local;
966 p->mtime = st.st_mtime;
967 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
972 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
974 const char *path = sha1_pack_name(sha1);
975 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
977 strcpy(p->pack_name, path);
978 hashcpy(p->sha1, sha1);
979 if (check_packed_git_idx(idx_path, p)) {
987 void install_packed_git(struct packed_git *pack)
989 if (pack->pack_fd != -1)
992 pack->next = packed_git;
996 static void prepare_packed_git_one(char *objdir, int local)
998 /* Ensure that this buffer is large enough so that we can
999 append "/pack/" without clobbering the stack even if
1000 strlen(objdir) were PATH_MAX. */
1001 char path[PATH_MAX + 1 + 4 + 1 + 1];
1006 sprintf(path, "%s/pack", objdir);
1008 dir = opendir(path);
1010 if (errno != ENOENT)
1011 error("unable to open object pack directory: %s: %s",
1012 path, strerror(errno));
1016 while ((de = readdir(dir)) != NULL) {
1017 int namelen = strlen(de->d_name);
1018 struct packed_git *p;
1020 if (!has_extension(de->d_name, ".idx"))
1023 if (len + namelen + 1 > sizeof(path))
1026 /* Don't reopen a pack we already have. */
1027 strcpy(path + len, de->d_name);
1028 for (p = packed_git; p; p = p->next) {
1029 if (!memcmp(path, p->pack_name, len + namelen - 4))
1034 /* See if it really is a valid .idx file with corresponding
1035 * .pack file that we can map.
1037 p = add_packed_git(path, len + namelen, local);
1040 install_packed_git(p);
1045 static int sort_pack(const void *a_, const void *b_)
1047 struct packed_git *a = *((struct packed_git **)a_);
1048 struct packed_git *b = *((struct packed_git **)b_);
1052 * Local packs tend to contain objects specific to our
1053 * variant of the project than remote ones. In addition,
1054 * remote ones could be on a network mounted filesystem.
1055 * Favor local ones for these reasons.
1057 st = a->pack_local - b->pack_local;
1062 * Younger packs tend to contain more recent objects,
1063 * and more recent objects tend to get accessed more
1066 if (a->mtime < b->mtime)
1068 else if (a->mtime == b->mtime)
1073 static void rearrange_packed_git(void)
1075 struct packed_git **ary, *p;
1078 for (n = 0, p = packed_git; p; p = p->next)
1083 /* prepare an array of packed_git for easier sorting */
1084 ary = xcalloc(n, sizeof(struct packed_git *));
1085 for (n = 0, p = packed_git; p; p = p->next)
1088 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1090 /* link them back again */
1091 for (i = 0; i < n - 1; i++)
1092 ary[i]->next = ary[i + 1];
1093 ary[n - 1]->next = NULL;
1094 packed_git = ary[0];
1099 static int prepare_packed_git_run_once = 0;
1100 void prepare_packed_git(void)
1102 struct alternate_object_database *alt;
1104 if (prepare_packed_git_run_once)
1106 prepare_packed_git_one(get_object_directory(), 1);
1108 for (alt = alt_odb_list; alt; alt = alt->next) {
1110 prepare_packed_git_one(alt->base, 0);
1111 alt->name[-1] = '/';
1113 rearrange_packed_git();
1114 prepare_packed_git_run_once = 1;
1117 void reprepare_packed_git(void)
1120 prepare_packed_git_run_once = 0;
1121 prepare_packed_git();
1124 static void mark_bad_packed_object(struct packed_git *p,
1125 const unsigned char *sha1)
1128 for (i = 0; i < p->num_bad_objects; i++)
1129 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1131 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1132 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1133 p->num_bad_objects++;
1136 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1138 struct packed_git *p;
1141 for (p = packed_git; p; p = p->next)
1142 for (i = 0; i < p->num_bad_objects; i++)
1143 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1148 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1150 unsigned char real_sha1[20];
1151 hash_sha1_file(map, size, type, real_sha1);
1152 return hashcmp(sha1, real_sha1) ? -1 : 0;
1155 static int git_open_noatime(const char *name)
1157 static int sha1_file_open_flag = O_NOATIME;
1160 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1164 /* Might the failure be due to O_NOATIME? */
1165 if (errno != ENOENT && sha1_file_open_flag) {
1166 sha1_file_open_flag = 0;
1174 static int open_sha1_file(const unsigned char *sha1)
1177 char *name = sha1_file_name(sha1);
1178 struct alternate_object_database *alt;
1180 fd = git_open_noatime(name);
1186 for (alt = alt_odb_list; alt; alt = alt->next) {
1188 fill_sha1_path(name, sha1);
1189 fd = git_open_noatime(alt->base);
1196 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1201 fd = open_sha1_file(sha1);
1206 if (!fstat(fd, &st)) {
1207 *size = xsize_t(st.st_size);
1209 /* mmap() is forbidden on empty files */
1210 error("object file %s is empty", sha1_file_name(sha1));
1213 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1221 * There used to be a second loose object header format which
1222 * was meant to mimic the in-pack format, allowing for direct
1223 * copy of the object data. This format turned up not to be
1224 * really worth it and we no longer write loose objects in that
1227 static int experimental_loose_object(unsigned char *map)
1232 * We must determine if the buffer contains the standard
1233 * zlib-deflated stream or the experimental format based
1234 * on the in-pack object format. Compare the header byte
1237 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1238 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1240 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1241 * in standard loose-object format, UNLESS it is a Git-pack
1242 * format object *exactly* 8 bytes in size when inflated.
1244 * However, RFC1950 also specifies that the 1st 16-bit word
1245 * must be divisible by 31 - this checksum tells us our buffer
1246 * is in the standard format, giving a false positive only if
1247 * the 1st word of the Git-pack format object happens to be
1248 * divisible by 31, ie:
1249 * ((byte0 * 256) + byte1) % 31 = 0
1250 * => 0ttt10000www1000 % 31 = 0
1252 * As it happens, this case can only arise for www=3 & ttt=1
1253 * - ie, a Commit object, which would have to be 8 bytes in
1254 * size. As no Commit can be that small, we find that the
1255 * combination of these two criteria (bitmask & checksum)
1256 * can always correctly determine the buffer format.
1258 word = (map[0] << 8) + map[1];
1259 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1265 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1266 unsigned long len, enum object_type *type, unsigned long *sizep)
1269 unsigned long size, c;
1270 unsigned long used = 0;
1273 *type = (c >> 4) & 7;
1277 if (len <= used || bitsizeof(long) <= shift) {
1278 error("bad object header");
1283 size += (c & 0x7f) << shift;
1290 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1292 unsigned long size, used;
1293 static const char valid_loose_object_type[8] = {
1295 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1296 0, /* "delta" and others are invalid in a loose object */
1298 enum object_type type;
1300 /* Get the data stream */
1301 memset(stream, 0, sizeof(*stream));
1302 stream->next_in = map;
1303 stream->avail_in = mapsize;
1304 stream->next_out = buffer;
1305 stream->avail_out = bufsiz;
1307 if (experimental_loose_object(map)) {
1309 * The old experimental format we no longer produce;
1310 * we can still read it.
1312 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1313 if (!used || !valid_loose_object_type[type])
1318 /* Set up the stream for the rest.. */
1319 stream->next_in = map;
1320 stream->avail_in = mapsize;
1321 git_inflate_init(stream);
1323 /* And generate the fake traditional header */
1324 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1325 typename(type), size);
1328 git_inflate_init(stream);
1329 return git_inflate(stream, 0);
1332 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1334 int bytes = strlen(buffer) + 1;
1335 unsigned char *buf = xmallocz(size);
1339 n = stream->total_out - bytes;
1342 memcpy(buf, (char *) buffer + bytes, n);
1344 if (bytes <= size) {
1346 * The above condition must be (bytes <= size), not
1347 * (bytes < size). In other words, even though we
1348 * expect no more output and set avail_out to zero,
1349 * the input zlib stream may have bytes that express
1350 * "this concludes the stream", and we *do* want to
1353 * Otherwise we would not be able to test that we
1354 * consumed all the input to reach the expected size;
1355 * we also want to check that zlib tells us that all
1356 * went well with status == Z_STREAM_END at the end.
1358 stream->next_out = buf + bytes;
1359 stream->avail_out = size - bytes;
1360 while (status == Z_OK)
1361 status = git_inflate(stream, Z_FINISH);
1363 if (status == Z_STREAM_END && !stream->avail_in) {
1364 git_inflate_end(stream);
1369 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1370 else if (stream->avail_in)
1371 error("garbage at end of loose object '%s'",
1378 * We used to just use "sscanf()", but that's actually way
1379 * too permissive for what we want to check. So do an anal
1380 * object header parse by hand.
1382 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1389 * The type can be at most ten bytes (including the
1390 * terminating '\0' that we add), and is followed by
1399 if (i >= sizeof(type))
1405 * The length must follow immediately, and be in canonical
1406 * decimal format (ie "010" is not valid).
1408 size = *hdr++ - '0';
1413 unsigned long c = *hdr - '0';
1417 size = size * 10 + c;
1423 * The length must be followed by a zero byte
1425 return *hdr ? -1 : type_from_string(type);
1428 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1434 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1435 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1438 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1441 unsigned long get_size_from_delta(struct packed_git *p,
1442 struct pack_window **w_curs,
1445 const unsigned char *data;
1446 unsigned char delta_head[20], *in;
1450 memset(&stream, 0, sizeof(stream));
1451 stream.next_out = delta_head;
1452 stream.avail_out = sizeof(delta_head);
1454 git_inflate_init(&stream);
1456 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1457 stream.next_in = in;
1458 st = git_inflate(&stream, Z_FINISH);
1459 curpos += stream.next_in - in;
1460 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1461 stream.total_out < sizeof(delta_head));
1462 git_inflate_end(&stream);
1463 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1464 error("delta data unpack-initial failed");
1468 /* Examine the initial part of the delta to figure out
1473 /* ignore base size */
1474 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1476 /* Read the result size */
1477 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1480 static off_t get_delta_base(struct packed_git *p,
1481 struct pack_window **w_curs,
1483 enum object_type type,
1484 off_t delta_obj_offset)
1486 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1489 /* use_pack() assured us we have [base_info, base_info + 20)
1490 * as a range that we can look at without walking off the
1491 * end of the mapped window. Its actually the hash size
1492 * that is assured. An OFS_DELTA longer than the hash size
1493 * is stupid, as then a REF_DELTA would be smaller to store.
1495 if (type == OBJ_OFS_DELTA) {
1497 unsigned char c = base_info[used++];
1498 base_offset = c & 127;
1501 if (!base_offset || MSB(base_offset, 7))
1502 return 0; /* overflow */
1503 c = base_info[used++];
1504 base_offset = (base_offset << 7) + (c & 127);
1506 base_offset = delta_obj_offset - base_offset;
1507 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1508 return 0; /* out of bound */
1510 } else if (type == OBJ_REF_DELTA) {
1511 /* The base entry _must_ be in the same pack */
1512 base_offset = find_pack_entry_one(base_info, p);
1515 die("I am totally screwed");
1519 /* forward declaration for a mutually recursive function */
1520 static int packed_object_info(struct packed_git *p, off_t offset,
1521 unsigned long *sizep, int *rtype);
1523 static int packed_delta_info(struct packed_git *p,
1524 struct pack_window **w_curs,
1526 enum object_type type,
1528 unsigned long *sizep)
1532 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1535 type = packed_object_info(p, base_offset, NULL, NULL);
1536 if (type <= OBJ_NONE) {
1537 struct revindex_entry *revidx;
1538 const unsigned char *base_sha1;
1539 revidx = find_pack_revindex(p, base_offset);
1542 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1543 mark_bad_packed_object(p, base_sha1);
1544 type = sha1_object_info(base_sha1, NULL);
1545 if (type <= OBJ_NONE)
1549 /* We choose to only get the type of the base object and
1550 * ignore potentially corrupt pack file that expects the delta
1551 * based on a base with a wrong size. This saves tons of
1555 *sizep = get_size_from_delta(p, w_curs, curpos);
1563 int unpack_object_header(struct packed_git *p,
1564 struct pack_window **w_curs,
1566 unsigned long *sizep)
1568 unsigned char *base;
1571 enum object_type type;
1573 /* use_pack() assures us we have [base, base + 20) available
1574 * as a range that we can look at. (Its actually the hash
1575 * size that is assured.) With our object header encoding
1576 * the maximum deflated object size is 2^137, which is just
1577 * insane, so we know won't exceed what we have been given.
1579 base = use_pack(p, w_curs, *curpos, &left);
1580 used = unpack_object_header_buffer(base, left, &type, sizep);
1589 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1590 unsigned long *sizep, int *rtype)
1592 struct pack_window *w_curs = NULL;
1594 off_t curpos = obj_offset;
1595 enum object_type type;
1597 type = unpack_object_header(p, &w_curs, &curpos, &size);
1599 *rtype = type; /* representation type */
1604 type = packed_delta_info(p, &w_curs, curpos,
1605 type, obj_offset, sizep);
1615 error("unknown object type %i at offset %"PRIuMAX" in %s",
1616 type, (uintmax_t)obj_offset, p->pack_name);
1619 unuse_pack(&w_curs);
1623 static void *unpack_compressed_entry(struct packed_git *p,
1624 struct pack_window **w_curs,
1630 unsigned char *buffer, *in;
1632 buffer = xmallocz(size);
1633 memset(&stream, 0, sizeof(stream));
1634 stream.next_out = buffer;
1635 stream.avail_out = size + 1;
1637 git_inflate_init(&stream);
1639 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1640 stream.next_in = in;
1641 st = git_inflate(&stream, Z_FINISH);
1642 if (!stream.avail_out)
1643 break; /* the payload is larger than it should be */
1644 curpos += stream.next_in - in;
1645 } while (st == Z_OK || st == Z_BUF_ERROR);
1646 git_inflate_end(&stream);
1647 if ((st != Z_STREAM_END) || stream.total_out != size) {
1655 #define MAX_DELTA_CACHE (256)
1657 static size_t delta_base_cached;
1659 static struct delta_base_cache_lru_list {
1660 struct delta_base_cache_lru_list *prev;
1661 struct delta_base_cache_lru_list *next;
1662 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1664 static struct delta_base_cache_entry {
1665 struct delta_base_cache_lru_list lru;
1667 struct packed_git *p;
1670 enum object_type type;
1671 } delta_base_cache[MAX_DELTA_CACHE];
1673 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1677 hash = (unsigned long)p + (unsigned long)base_offset;
1678 hash += (hash >> 8) + (hash >> 16);
1679 return hash % MAX_DELTA_CACHE;
1682 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1684 unsigned long hash = pack_entry_hash(p, base_offset);
1685 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1686 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1689 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1690 unsigned long *base_size, enum object_type *type, int keep_cache)
1693 unsigned long hash = pack_entry_hash(p, base_offset);
1694 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1697 if (!ret || ent->p != p || ent->base_offset != base_offset)
1698 return unpack_entry(p, base_offset, type, base_size);
1702 ent->lru.next->prev = ent->lru.prev;
1703 ent->lru.prev->next = ent->lru.next;
1704 delta_base_cached -= ent->size;
1706 ret = xmemdupz(ent->data, ent->size);
1709 *base_size = ent->size;
1713 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1718 ent->lru.next->prev = ent->lru.prev;
1719 ent->lru.prev->next = ent->lru.next;
1720 delta_base_cached -= ent->size;
1724 void clear_delta_base_cache(void)
1727 for (p = 0; p < MAX_DELTA_CACHE; p++)
1728 release_delta_base_cache(&delta_base_cache[p]);
1731 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1732 void *base, unsigned long base_size, enum object_type type)
1734 unsigned long hash = pack_entry_hash(p, base_offset);
1735 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1736 struct delta_base_cache_lru_list *lru;
1738 release_delta_base_cache(ent);
1739 delta_base_cached += base_size;
1741 for (lru = delta_base_cache_lru.next;
1742 delta_base_cached > delta_base_cache_limit
1743 && lru != &delta_base_cache_lru;
1745 struct delta_base_cache_entry *f = (void *)lru;
1746 if (f->type == OBJ_BLOB)
1747 release_delta_base_cache(f);
1749 for (lru = delta_base_cache_lru.next;
1750 delta_base_cached > delta_base_cache_limit
1751 && lru != &delta_base_cache_lru;
1753 struct delta_base_cache_entry *f = (void *)lru;
1754 release_delta_base_cache(f);
1758 ent->base_offset = base_offset;
1761 ent->size = base_size;
1762 ent->lru.next = &delta_base_cache_lru;
1763 ent->lru.prev = delta_base_cache_lru.prev;
1764 delta_base_cache_lru.prev->next = &ent->lru;
1765 delta_base_cache_lru.prev = &ent->lru;
1768 static void *read_object(const unsigned char *sha1, enum object_type *type,
1769 unsigned long *size);
1771 static void *unpack_delta_entry(struct packed_git *p,
1772 struct pack_window **w_curs,
1774 unsigned long delta_size,
1776 enum object_type *type,
1777 unsigned long *sizep)
1779 void *delta_data, *result, *base;
1780 unsigned long base_size;
1783 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1785 error("failed to validate delta base reference "
1786 "at offset %"PRIuMAX" from %s",
1787 (uintmax_t)curpos, p->pack_name);
1791 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1794 * We're probably in deep shit, but let's try to fetch
1795 * the required base anyway from another pack or loose.
1796 * This is costly but should happen only in the presence
1797 * of a corrupted pack, and is better than failing outright.
1799 struct revindex_entry *revidx;
1800 const unsigned char *base_sha1;
1801 revidx = find_pack_revindex(p, base_offset);
1804 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1805 error("failed to read delta base object %s"
1806 " at offset %"PRIuMAX" from %s",
1807 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1809 mark_bad_packed_object(p, base_sha1);
1810 base = read_object(base_sha1, type, &base_size);
1815 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1817 error("failed to unpack compressed delta "
1818 "at offset %"PRIuMAX" from %s",
1819 (uintmax_t)curpos, p->pack_name);
1823 result = patch_delta(base, base_size,
1824 delta_data, delta_size,
1827 die("failed to apply delta");
1829 add_delta_base_cache(p, base_offset, base, base_size, *type);
1833 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1835 static FILE *log_file;
1838 log_file = fopen(log_pack_access, "w");
1840 error("cannot open pack access log '%s' for writing: %s",
1841 log_pack_access, strerror(errno));
1842 log_pack_access = NULL;
1846 fprintf(log_file, "%s %"PRIuMAX"\n",
1847 p->pack_name, (uintmax_t)obj_offset);
1851 int do_check_packed_object_crc;
1853 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1854 enum object_type *type, unsigned long *sizep)
1856 struct pack_window *w_curs = NULL;
1857 off_t curpos = obj_offset;
1860 if (log_pack_access)
1861 write_pack_access_log(p, obj_offset);
1863 if (do_check_packed_object_crc && p->index_version > 1) {
1864 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1865 unsigned long len = revidx[1].offset - obj_offset;
1866 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1867 const unsigned char *sha1 =
1868 nth_packed_object_sha1(p, revidx->nr);
1869 error("bad packed object CRC for %s",
1871 mark_bad_packed_object(p, sha1);
1872 unuse_pack(&w_curs);
1877 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1881 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1882 obj_offset, type, sizep);
1888 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1892 error("unknown object type %i at offset %"PRIuMAX" in %s",
1893 *type, (uintmax_t)obj_offset, p->pack_name);
1895 unuse_pack(&w_curs);
1899 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1902 const unsigned char *index = p->index_data;
1904 if (open_pack_index(p))
1906 index = p->index_data;
1908 if (n >= p->num_objects)
1911 if (p->index_version == 1) {
1912 return index + 24 * n + 4;
1915 return index + 20 * n;
1919 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1921 const unsigned char *index = p->index_data;
1923 if (p->index_version == 1) {
1924 return ntohl(*((uint32_t *)(index + 24 * n)));
1927 index += 8 + p->num_objects * (20 + 4);
1928 off = ntohl(*((uint32_t *)(index + 4 * n)));
1929 if (!(off & 0x80000000))
1931 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1932 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1933 ntohl(*((uint32_t *)(index + 4)));
1937 off_t find_pack_entry_one(const unsigned char *sha1,
1938 struct packed_git *p)
1940 const uint32_t *level1_ofs = p->index_data;
1941 const unsigned char *index = p->index_data;
1942 unsigned hi, lo, stride;
1943 static int use_lookup = -1;
1944 static int debug_lookup = -1;
1946 if (debug_lookup < 0)
1947 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1950 if (open_pack_index(p))
1952 level1_ofs = p->index_data;
1953 index = p->index_data;
1955 if (p->index_version > 1) {
1960 hi = ntohl(level1_ofs[*sha1]);
1961 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1962 if (p->index_version > 1) {
1970 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1971 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1974 use_lookup = !!getenv("GIT_USE_LOOKUP");
1976 int pos = sha1_entry_pos(index, stride, 0,
1977 lo, hi, p->num_objects, sha1);
1980 return nth_packed_object_offset(p, pos);
1984 unsigned mi = (lo + hi) / 2;
1985 int cmp = hashcmp(index + mi * stride, sha1);
1988 printf("lo %u hi %u rg %u mi %u\n",
1989 lo, hi, hi - lo, mi);
1991 return nth_packed_object_offset(p, mi);
2000 int is_pack_valid(struct packed_git *p)
2002 /* An already open pack is known to be valid. */
2003 if (p->pack_fd != -1)
2006 /* If the pack has one window completely covering the
2007 * file size, the pack is known to be valid even if
2008 * the descriptor is not currently open.
2011 struct pack_window *w = p->windows;
2013 if (!w->offset && w->len == p->pack_size)
2017 /* Force the pack to open to prove its valid. */
2018 return !open_packed_git(p);
2021 static int fill_pack_entry(const unsigned char *sha1,
2022 struct pack_entry *e,
2023 struct packed_git *p)
2027 if (p->num_bad_objects) {
2029 for (i = 0; i < p->num_bad_objects; i++)
2030 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2034 offset = find_pack_entry_one(sha1, p);
2039 * We are about to tell the caller where they can locate the
2040 * requested object. We better make sure the packfile is
2041 * still here and can be accessed before supplying that
2042 * answer, as it may have been deleted since the index was
2045 if (!is_pack_valid(p)) {
2046 warning("packfile %s cannot be accessed", p->pack_name);
2051 hashcpy(e->sha1, sha1);
2055 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2057 struct packed_git *p;
2059 prepare_packed_git();
2063 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2066 for (p = packed_git; p; p = p->next) {
2067 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2070 last_found_pack = p;
2076 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2077 struct packed_git *packs)
2079 struct packed_git *p;
2081 for (p = packs; p; p = p->next) {
2082 if (find_pack_entry_one(sha1, p))
2089 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2092 unsigned long mapsize, size;
2097 map = map_sha1_file(sha1, &mapsize);
2099 return error("unable to find %s", sha1_to_hex(sha1));
2100 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2101 status = error("unable to unpack %s header",
2103 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2104 status = error("unable to parse %s header", sha1_to_hex(sha1));
2107 git_inflate_end(&stream);
2108 munmap(map, mapsize);
2112 /* returns enum object_type or negative */
2113 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2115 struct cached_object *co;
2116 struct pack_entry e;
2119 co = find_cached_object(sha1);
2122 *(oi->sizep) = co->size;
2123 oi->whence = OI_CACHED;
2127 if (!find_pack_entry(sha1, &e)) {
2128 /* Most likely it's a loose object. */
2129 status = sha1_loose_object_info(sha1, oi->sizep);
2131 oi->whence = OI_LOOSE;
2135 /* Not a loose object; someone else may have just packed it. */
2136 reprepare_packed_git();
2137 if (!find_pack_entry(sha1, &e))
2141 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2143 mark_bad_packed_object(e.p, sha1);
2144 status = sha1_object_info_extended(sha1, oi);
2145 } else if (in_delta_base_cache(e.p, e.offset)) {
2146 oi->whence = OI_DBCACHED;
2148 oi->whence = OI_PACKED;
2149 oi->u.packed.offset = e.offset;
2150 oi->u.packed.pack = e.p;
2151 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2152 rtype == OBJ_OFS_DELTA);
2158 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2160 struct object_info oi;
2163 return sha1_object_info_extended(sha1, &oi);
2166 static void *read_packed_sha1(const unsigned char *sha1,
2167 enum object_type *type, unsigned long *size)
2169 struct pack_entry e;
2172 if (!find_pack_entry(sha1, &e))
2174 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2177 * We're probably in deep shit, but let's try to fetch
2178 * the required object anyway from another pack or loose.
2179 * This should happen only in the presence of a corrupted
2180 * pack, and is better than failing outright.
2182 error("failed to read object %s at offset %"PRIuMAX" from %s",
2183 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2184 mark_bad_packed_object(e.p, sha1);
2185 data = read_object(sha1, type, size);
2190 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2191 unsigned char *sha1)
2193 struct cached_object *co;
2195 hash_sha1_file(buf, len, typename(type), sha1);
2196 if (has_sha1_file(sha1) || find_cached_object(sha1))
2198 if (cached_object_alloc <= cached_object_nr) {
2199 cached_object_alloc = alloc_nr(cached_object_alloc);
2200 cached_objects = xrealloc(cached_objects,
2201 sizeof(*cached_objects) *
2202 cached_object_alloc);
2204 co = &cached_objects[cached_object_nr++];
2207 co->buf = xmalloc(len);
2208 memcpy(co->buf, buf, len);
2209 hashcpy(co->sha1, sha1);
2213 static void *read_object(const unsigned char *sha1, enum object_type *type,
2214 unsigned long *size)
2216 unsigned long mapsize;
2218 struct cached_object *co;
2220 co = find_cached_object(sha1);
2224 return xmemdupz(co->buf, co->size);
2227 buf = read_packed_sha1(sha1, type, size);
2230 map = map_sha1_file(sha1, &mapsize);
2232 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2233 munmap(map, mapsize);
2236 reprepare_packed_git();
2237 return read_packed_sha1(sha1, type, size);
2241 * This function dies on corrupt objects; the callers who want to
2242 * deal with them should arrange to call read_object() and give error
2243 * messages themselves.
2245 void *read_sha1_file_extended(const unsigned char *sha1,
2246 enum object_type *type,
2247 unsigned long *size,
2252 const struct packed_git *p;
2253 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2254 ? lookup_replace_object(sha1) : sha1;
2257 data = read_object(repl, type, size);
2261 if (errno && errno != ENOENT)
2262 die_errno("failed to read object %s", sha1_to_hex(sha1));
2264 /* die if we replaced an object with one that does not exist */
2266 die("replacement %s not found for %s",
2267 sha1_to_hex(repl), sha1_to_hex(sha1));
2269 if (has_loose_object(repl)) {
2270 path = sha1_file_name(sha1);
2271 die("loose object %s (stored in %s) is corrupt",
2272 sha1_to_hex(repl), path);
2275 if ((p = has_packed_and_bad(repl)) != NULL)
2276 die("packed object %s (stored in %s) is corrupt",
2277 sha1_to_hex(repl), p->pack_name);
2282 void *read_object_with_reference(const unsigned char *sha1,
2283 const char *required_type_name,
2284 unsigned long *size,
2285 unsigned char *actual_sha1_return)
2287 enum object_type type, required_type;
2289 unsigned long isize;
2290 unsigned char actual_sha1[20];
2292 required_type = type_from_string(required_type_name);
2293 hashcpy(actual_sha1, sha1);
2295 int ref_length = -1;
2296 const char *ref_type = NULL;
2298 buffer = read_sha1_file(actual_sha1, &type, &isize);
2301 if (type == required_type) {
2303 if (actual_sha1_return)
2304 hashcpy(actual_sha1_return, actual_sha1);
2307 /* Handle references */
2308 else if (type == OBJ_COMMIT)
2310 else if (type == OBJ_TAG)
2311 ref_type = "object ";
2316 ref_length = strlen(ref_type);
2318 if (ref_length + 40 > isize ||
2319 memcmp(buffer, ref_type, ref_length) ||
2320 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2325 /* Now we have the ID of the referred-to object in
2326 * actual_sha1. Check again. */
2330 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2331 const char *type, unsigned char *sha1,
2332 char *hdr, int *hdrlen)
2336 /* Generate the header */
2337 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2341 git_SHA1_Update(&c, hdr, *hdrlen);
2342 git_SHA1_Update(&c, buf, len);
2343 git_SHA1_Final(sha1, &c);
2347 * Move the just written object into its final resting place.
2348 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2349 * "moving" is only a part of what it does, when no patch between
2350 * master to pu changes the call sites of this function.
2352 int move_temp_to_file(const char *tmpfile, const char *filename)
2356 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2358 else if (link(tmpfile, filename))
2362 * Coda hack - coda doesn't like cross-directory links,
2363 * so we fall back to a rename, which will mean that it
2364 * won't be able to check collisions, but that's not a
2367 * The same holds for FAT formatted media.
2369 * When this succeeds, we just return. We have nothing
2372 if (ret && ret != EEXIST) {
2374 if (!rename(tmpfile, filename))
2378 unlink_or_warn(tmpfile);
2380 if (ret != EEXIST) {
2381 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2383 /* FIXME!!! Collision check here ? */
2387 if (adjust_shared_perm(filename))
2388 return error("unable to set permission to '%s'", filename);
2392 static int write_buffer(int fd, const void *buf, size_t len)
2394 if (write_in_full(fd, buf, len) < 0)
2395 return error("file write error (%s)", strerror(errno));
2399 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2400 unsigned char *sha1)
2404 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2408 /* Finalize a file on disk, and close it. */
2409 static void close_sha1_file(int fd)
2411 if (fsync_object_files)
2412 fsync_or_die(fd, "sha1 file");
2414 die_errno("error when closing sha1 file");
2417 /* Size of directory component, including the ending '/' */
2418 static inline int directory_size(const char *filename)
2420 const char *s = strrchr(filename, '/');
2423 return s - filename + 1;
2427 * This creates a temporary file in the same directory as the final
2430 * We want to avoid cross-directory filename renames, because those
2431 * can have problems on various filesystems (FAT, NFS, Coda).
2433 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2435 int fd, dirlen = directory_size(filename);
2437 if (dirlen + 20 > bufsiz) {
2438 errno = ENAMETOOLONG;
2441 memcpy(buffer, filename, dirlen);
2442 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2443 fd = git_mkstemp_mode(buffer, 0444);
2444 if (fd < 0 && dirlen && errno == ENOENT) {
2445 /* Make sure the directory exists */
2446 memcpy(buffer, filename, dirlen);
2447 buffer[dirlen-1] = 0;
2448 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2452 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2453 fd = git_mkstemp_mode(buffer, 0444);
2458 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2459 const void *buf, unsigned long len, time_t mtime)
2462 unsigned char compressed[4096];
2465 unsigned char parano_sha1[20];
2467 static char tmp_file[PATH_MAX];
2469 filename = sha1_file_name(sha1);
2470 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2472 if (errno == EACCES)
2473 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2475 return error("unable to create temporary sha1 filename %s: %s\n", tmp_file, strerror(errno));
2479 memset(&stream, 0, sizeof(stream));
2480 git_deflate_init(&stream, zlib_compression_level);
2481 stream.next_out = compressed;
2482 stream.avail_out = sizeof(compressed);
2485 /* First header.. */
2486 stream.next_in = (unsigned char *)hdr;
2487 stream.avail_in = hdrlen;
2488 while (git_deflate(&stream, 0) == Z_OK)
2490 git_SHA1_Update(&c, hdr, hdrlen);
2492 /* Then the data itself.. */
2493 stream.next_in = (void *)buf;
2494 stream.avail_in = len;
2496 unsigned char *in0 = stream.next_in;
2497 ret = git_deflate(&stream, Z_FINISH);
2498 git_SHA1_Update(&c, in0, stream.next_in - in0);
2499 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2500 die("unable to write sha1 file");
2501 stream.next_out = compressed;
2502 stream.avail_out = sizeof(compressed);
2503 } while (ret == Z_OK);
2505 if (ret != Z_STREAM_END)
2506 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2507 ret = git_deflate_end_gently(&stream);
2509 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2510 git_SHA1_Final(parano_sha1, &c);
2511 if (hashcmp(sha1, parano_sha1) != 0)
2512 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2514 close_sha1_file(fd);
2519 utb.modtime = mtime;
2520 if (utime(tmp_file, &utb) < 0)
2521 warning("failed utime() on %s: %s",
2522 tmp_file, strerror(errno));
2525 return move_temp_to_file(tmp_file, filename);
2528 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2530 unsigned char sha1[20];
2534 /* Normally if we have it in the pack then we do not bother writing
2535 * it out into .git/objects/??/?{38} file.
2537 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2539 hashcpy(returnsha1, sha1);
2540 if (has_sha1_file(sha1))
2542 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2545 int force_object_loose(const unsigned char *sha1, time_t mtime)
2549 enum object_type type;
2554 if (has_loose_object(sha1))
2556 buf = read_packed_sha1(sha1, &type, &len);
2558 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2559 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2560 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2566 int has_pack_index(const unsigned char *sha1)
2569 if (stat(sha1_pack_index_name(sha1), &st))
2574 int has_sha1_pack(const unsigned char *sha1)
2576 struct pack_entry e;
2577 return find_pack_entry(sha1, &e);
2580 int has_sha1_file(const unsigned char *sha1)
2582 struct pack_entry e;
2584 if (find_pack_entry(sha1, &e))
2586 return has_loose_object(sha1);
2589 static void check_tree(const void *buf, size_t size)
2591 struct tree_desc desc;
2592 struct name_entry entry;
2594 init_tree_desc(&desc, buf, size);
2595 while (tree_entry(&desc, &entry))
2597 * tree_entry() will die() on malformed entries */
2601 static void check_commit(const void *buf, size_t size)
2604 memset(&c, 0, sizeof(c));
2605 if (parse_commit_buffer(&c, buf, size))
2606 die("corrupt commit");
2609 static void check_tag(const void *buf, size_t size)
2612 memset(&t, 0, sizeof(t));
2613 if (parse_tag_buffer(&t, buf, size))
2617 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2618 enum object_type type,
2619 const char *path, unsigned flags)
2621 int ret, re_allocated = 0;
2622 int write_object = flags & HASH_WRITE_OBJECT;
2628 * Convert blobs to git internal format
2630 if ((type == OBJ_BLOB) && path) {
2631 struct strbuf nbuf = STRBUF_INIT;
2632 if (convert_to_git(path, buf, size, &nbuf,
2633 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2634 buf = strbuf_detach(&nbuf, &size);
2638 if (flags & HASH_FORMAT_CHECK) {
2639 if (type == OBJ_TREE)
2640 check_tree(buf, size);
2641 if (type == OBJ_COMMIT)
2642 check_commit(buf, size);
2643 if (type == OBJ_TAG)
2644 check_tag(buf, size);
2648 ret = write_sha1_file(buf, size, typename(type), sha1);
2650 ret = hash_sha1_file(buf, size, typename(type), sha1);
2656 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2657 const char *path, unsigned flags)
2659 struct strbuf sbuf = STRBUF_INIT;
2662 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2663 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2666 strbuf_release(&sbuf);
2670 #define SMALL_FILE_SIZE (32*1024)
2672 static int index_core(unsigned char *sha1, int fd, size_t size,
2673 enum object_type type, const char *path,
2679 ret = index_mem(sha1, NULL, size, type, path, flags);
2680 } else if (size <= SMALL_FILE_SIZE) {
2681 char *buf = xmalloc(size);
2682 if (size == read_in_full(fd, buf, size))
2683 ret = index_mem(sha1, buf, size, type, path, flags);
2685 ret = error("short read %s", strerror(errno));
2688 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2689 ret = index_mem(sha1, buf, size, type, path, flags);
2696 * This creates one packfile per large blob unless bulk-checkin
2697 * machinery is "plugged".
2699 * This also bypasses the usual "convert-to-git" dance, and that is on
2700 * purpose. We could write a streaming version of the converting
2701 * functions and insert that before feeding the data to fast-import
2702 * (or equivalent in-core API described above). However, that is
2703 * somewhat complicated, as we do not know the size of the filter
2704 * result, which we need to know beforehand when writing a git object.
2705 * Since the primary motivation for trying to stream from the working
2706 * tree file and to avoid mmaping it in core is to deal with large
2707 * binary blobs, they generally do not want to get any conversion, and
2708 * callers should avoid this code path when filters are requested.
2710 static int index_stream(unsigned char *sha1, int fd, size_t size,
2711 enum object_type type, const char *path,
2714 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2717 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2718 enum object_type type, const char *path, unsigned flags)
2721 size_t size = xsize_t(st->st_size);
2723 if (!S_ISREG(st->st_mode))
2724 ret = index_pipe(sha1, fd, type, path, flags);
2725 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2726 (path && would_convert_to_git(path, NULL, 0, 0)))
2727 ret = index_core(sha1, fd, size, type, path, flags);
2729 ret = index_stream(sha1, fd, size, type, path, flags);
2734 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2737 struct strbuf sb = STRBUF_INIT;
2739 switch (st->st_mode & S_IFMT) {
2741 fd = open(path, O_RDONLY);
2743 return error("open(\"%s\"): %s", path,
2745 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2746 return error("%s: failed to insert into database",
2750 if (strbuf_readlink(&sb, path, st->st_size)) {
2751 char *errstr = strerror(errno);
2752 return error("readlink(\"%s\"): %s", path,
2755 if (!(flags & HASH_WRITE_OBJECT))
2756 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2757 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2758 return error("%s: failed to insert into database",
2760 strbuf_release(&sb);
2763 return resolve_gitlink_ref(path, "HEAD", sha1);
2765 return error("%s: unsupported file type", path);
2770 int read_pack_header(int fd, struct pack_header *header)
2772 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2773 /* "eof before pack header was fully read" */
2774 return PH_ERROR_EOF;
2776 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2777 /* "protocol error (pack signature mismatch detected)" */
2778 return PH_ERROR_PACK_SIGNATURE;
2779 if (!pack_version_ok(header->hdr_version))
2780 /* "protocol error (pack version unsupported)" */
2781 return PH_ERROR_PROTOCOL;
2785 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2787 enum object_type type = sha1_object_info(sha1, NULL);
2789 die("%s is not a valid object", sha1_to_hex(sha1));
2791 die("%s is not a valid '%s' object", sha1_to_hex(sha1),