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"
23 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
24 #define O_NOATIME 01000000
30 #define SZ_FMT PRIuMAX
31 static inline uintmax_t sz_fmt(size_t s) { return s; }
33 const unsigned char null_sha1[20];
36 * This is meant to hold a *small* number of objects that you would
37 * want read_sha1_file() to be able to return, but yet you do not want
38 * to write them into the object store (e.g. a browse-only
41 static struct cached_object {
42 unsigned char sha1[20];
43 enum object_type type;
47 static int cached_object_nr, cached_object_alloc;
49 static struct cached_object empty_tree = {
50 EMPTY_TREE_SHA1_BIN_LITERAL,
56 static struct cached_object *find_cached_object(const unsigned char *sha1)
59 struct cached_object *co = cached_objects;
61 for (i = 0; i < cached_object_nr; i++, co++) {
62 if (!hashcmp(co->sha1, sha1))
65 if (!hashcmp(sha1, empty_tree.sha1))
70 int mkdir_in_gitdir(const char *path)
72 if (mkdir(path, 0777)) {
73 int saved_errno = errno;
75 struct strbuf sb = STRBUF_INIT;
80 * Are we looking at a path in a symlinked worktree
81 * whose original repository does not yet have it?
82 * e.g. .git/rr-cache pointing at its original
83 * repository in which the user hasn't performed any
84 * conflict resolution yet?
86 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
87 strbuf_readlink(&sb, path, st.st_size) ||
88 !is_absolute_path(sb.buf) ||
89 mkdir(sb.buf, 0777)) {
96 return adjust_shared_perm(path);
99 int safe_create_leading_directories(char *path)
101 char *pos = path + offset_1st_component(path);
105 pos = strchr(pos, '/');
108 while (*++pos == '/')
113 if (!stat(path, &st)) {
115 if (!S_ISDIR(st.st_mode)) {
120 else if (mkdir(path, 0777)) {
124 else if (adjust_shared_perm(path)) {
133 int safe_create_leading_directories_const(const char *path)
135 /* path points to cache entries, so xstrdup before messing with it */
136 char *buf = xstrdup(path);
137 int result = safe_create_leading_directories(buf);
142 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 for (i = 0; i < 20; i++) {
146 static char hex[] = "0123456789abcdef";
147 unsigned int val = sha1[i];
148 char *pos = pathbuf + i*2 + (i > 0);
149 *pos++ = hex[val >> 4];
150 *pos = hex[val & 0xf];
155 * NOTE! This returns a statically allocated buffer, so you have to be
156 * careful about using it. Do an "xstrdup()" if you need to save the
159 * Also note that this returns the location for creating. Reading
160 * SHA1 file can happen from any alternate directory listed in the
161 * DB_ENVIRONMENT environment variable if it is not found in
162 * the primary object database.
164 char *sha1_file_name(const unsigned char *sha1)
166 static char buf[PATH_MAX];
170 objdir = get_object_directory();
171 len = strlen(objdir);
173 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
174 if (len + 43 > PATH_MAX)
175 die("insanely long object directory %s", objdir);
176 memcpy(buf, objdir, len);
180 fill_sha1_path(buf + len + 1, sha1);
184 static char *sha1_get_pack_name(const unsigned char *sha1,
185 char **name, char **base, const char *which)
187 static const char hex[] = "0123456789abcdef";
192 const char *sha1_file_directory = get_object_directory();
193 int len = strlen(sha1_file_directory);
194 *base = xmalloc(len + 60);
195 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
196 sha1_file_directory, which);
197 *name = *base + len + 11;
202 for (i = 0; i < 20; i++) {
203 unsigned int val = *sha1++;
204 *buf++ = hex[val >> 4];
205 *buf++ = hex[val & 0xf];
211 char *sha1_pack_name(const unsigned char *sha1)
213 static char *name, *base;
215 return sha1_get_pack_name(sha1, &name, &base, "pack");
218 char *sha1_pack_index_name(const unsigned char *sha1)
220 static char *name, *base;
222 return sha1_get_pack_name(sha1, &name, &base, "idx");
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
229 static int git_open_noatime(const char *name);
232 * Prepare alternate object database registry.
234 * The variable alt_odb_list points at the list of struct
235 * alternate_object_database. The elements on this list come from
236 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
238 * whose contents is similar to that environment variable but can be
239 * LF separated. Its base points at a statically allocated buffer that
240 * contains "/the/directory/corresponding/to/.git/objects/...", while
241 * its name points just after the slash at the end of ".git/objects/"
242 * in the example above, and has enough space to hold 40-byte hex
243 * SHA1, an extra slash for the first level indirection, and the
246 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
253 int entlen = pfxlen + 43;
256 if (!is_absolute_path(entry) && relative_base) {
257 /* Relative alt-odb */
259 base_len = strlen(relative_base) + 1;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (!is_absolute_path(entry) && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (!is_directory(ent->base)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
295 if (!memcmp(ent->base, objdir, pfxlen)) {
300 /* add the alternate entry */
302 alt_odb_tail = &(ent->next);
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
319 error("%s: ignoring alternate object stores, nesting too deep.",
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
333 while (cp < ep && *cp != sep)
336 if (!is_absolute_path(last) && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
350 static void read_info_alternates(const char * relative_base, int depth)
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = git_open_noatime(path);
366 if (fstat(fd, &st) || (st.st_size == 0)) {
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
379 void add_to_alternates_file(const char *reference)
381 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
382 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
383 char *alt = mkpath("%s/objects\n", reference);
384 write_or_die(fd, alt, strlen(alt));
385 if (commit_lock_file(lock))
386 die("could not close alternates file");
388 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
391 void foreach_alt_odb(alt_odb_fn fn, void *cb)
393 struct alternate_object_database *ent;
396 for (ent = alt_odb_list; ent; ent = ent->next)
401 void prepare_alt_odb(void)
408 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
411 alt_odb_tail = &alt_odb_list;
412 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
414 read_info_alternates(get_object_directory(), 0);
417 static int has_loose_object_local(const unsigned char *sha1)
419 char *name = sha1_file_name(sha1);
420 return !access(name, F_OK);
423 int has_loose_object_nonlocal(const unsigned char *sha1)
425 struct alternate_object_database *alt;
427 for (alt = alt_odb_list; alt; alt = alt->next) {
428 fill_sha1_path(alt->name, sha1);
429 if (!access(alt->base, F_OK))
435 static int has_loose_object(const unsigned char *sha1)
437 return has_loose_object_local(sha1) ||
438 has_loose_object_nonlocal(sha1);
441 static unsigned int pack_used_ctr;
442 static unsigned int pack_mmap_calls;
443 static unsigned int peak_pack_open_windows;
444 static unsigned int pack_open_windows;
445 static unsigned int pack_open_fds;
446 static unsigned int pack_max_fds;
447 static size_t peak_pack_mapped;
448 static size_t pack_mapped;
449 struct packed_git *packed_git;
451 void pack_report(void)
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
468 pack_open_windows, peak_pack_open_windows,
469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
472 static int check_packed_git_idx(const char *path, struct packed_git *p)
475 struct pack_idx_header *hdr;
477 uint32_t version, nr, i, *index;
478 int fd = git_open_noatime(path);
483 if (fstat(fd, &st)) {
487 idx_size = xsize_t(st.st_size);
488 if (idx_size < 4 * 256 + 20 + 20) {
490 return error("index file %s is too small", path);
492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
511 index += 2; /* skip index header */
512 for (i = 0; i < 256; i++) {
513 uint32_t n = ntohl(index[i]);
515 munmap(idx_map, idx_size);
516 return error("non-monotonic index %s", path);
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
531 return error("wrong index v1 file size in %s", path);
533 } else if (version == 2) {
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size = min_size;
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
552 munmap(idx_map, idx_size);
553 return error("wrong index v2 file size in %s", path);
555 if (idx_size != min_size &&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
567 p->index_version = version;
568 p->index_data = idx_map;
569 p->index_size = idx_size;
574 int open_pack_index(struct packed_git *p)
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
589 static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
594 struct pack_window *w, *w_l;
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
608 static int unuse_one_window(struct packed_git *current, int keep_fd)
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
621 lru_l->next = lru_w->next;
623 lru_p->windows = lru_w->next;
624 if (!lru_p->windows && lru_p->pack_fd != -1
625 && lru_p->pack_fd != keep_fd) {
626 close(lru_p->pack_fd);
638 void release_pack_memory(size_t need, int fd)
640 size_t cur = pack_mapped;
641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
645 void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
652 release_pack_memory(length, fd);
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
660 void close_pack_windows(struct packed_git *p)
663 struct pack_window *w = p->windows;
666 die("pack '%s' still has open windows to it",
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
671 p->windows = w->next;
676 void unuse_pack(struct pack_window **w_cursor)
678 struct pack_window *w = *w_cursor;
685 void close_pack_index(struct packed_git *p)
688 munmap((void *)p->index_data, p->index_size);
689 p->index_data = NULL;
694 * This is used by git-repack in case a newly created pack happens to
695 * contain the same set of objects as an existing one. In that case
696 * the resulting file might be different even if its name would be the
697 * same. It is best to close any reference to the old pack before it is
698 * replaced on disk. Of course no index pointers nor windows for given pack
699 * must subsist at this point. If ever objects from this pack are requested
700 * again, the new version of the pack will be reinitialized through
701 * reprepare_packed_git().
703 void free_pack_by_name(const char *pack_name)
705 struct packed_git *p, **pp = &packed_git;
709 if (strcmp(pack_name, p->pack_name) == 0) {
710 clear_delta_base_cache();
711 close_pack_windows(p);
712 if (p->pack_fd != -1) {
717 free(p->bad_object_sha1);
727 * Do not call this directly as this leaks p->pack_fd on error return;
728 * call open_packed_git() instead.
730 static int open_packed_git_1(struct packed_git *p)
733 struct pack_header hdr;
734 unsigned char sha1[20];
735 unsigned char *idx_sha1;
738 if (!p->index_data && open_pack_index(p))
739 return error("packfile %s index unavailable", p->pack_name);
743 unsigned int max_fds;
745 if (getrlimit(RLIMIT_NOFILE, &lim))
746 die_errno("cannot get RLIMIT_NOFILE");
748 max_fds = lim.rlim_cur;
750 /* Save 3 for stdin/stdout/stderr, 22 for work */
752 pack_max_fds = max_fds - 25;
757 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
760 p->pack_fd = git_open_noatime(p->pack_name);
761 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
765 /* If we created the struct before we had the pack we lack size. */
767 if (!S_ISREG(st.st_mode))
768 return error("packfile %s not a regular file", p->pack_name);
769 p->pack_size = st.st_size;
770 } else if (p->pack_size != st.st_size)
771 return error("packfile %s size changed", p->pack_name);
773 /* We leave these file descriptors open with sliding mmap;
774 * there is no point keeping them open across exec(), though.
776 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
778 return error("cannot determine file descriptor flags");
779 fd_flag |= FD_CLOEXEC;
780 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
781 return error("cannot set FD_CLOEXEC");
783 /* Verify we recognize this pack file format. */
784 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
785 return error("file %s is far too short to be a packfile", p->pack_name);
786 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
787 return error("file %s is not a GIT packfile", p->pack_name);
788 if (!pack_version_ok(hdr.hdr_version))
789 return error("packfile %s is version %"PRIu32" and not"
790 " supported (try upgrading GIT to a newer version)",
791 p->pack_name, ntohl(hdr.hdr_version));
793 /* Verify the pack matches its index. */
794 if (p->num_objects != ntohl(hdr.hdr_entries))
795 return error("packfile %s claims to have %"PRIu32" objects"
796 " while index indicates %"PRIu32" objects",
797 p->pack_name, ntohl(hdr.hdr_entries),
799 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
800 return error("end of packfile %s is unavailable", p->pack_name);
801 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
802 return error("packfile %s signature is unavailable", p->pack_name);
803 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
804 if (hashcmp(sha1, idx_sha1))
805 return error("packfile %s does not match index", p->pack_name);
809 static int open_packed_git(struct packed_git *p)
811 if (!open_packed_git_1(p))
813 if (p->pack_fd != -1) {
821 static int in_window(struct pack_window *win, off_t offset)
823 /* We must promise at least 20 bytes (one hash) after the
824 * offset is available from this window, otherwise the offset
825 * is not actually in this window and a different window (which
826 * has that one hash excess) must be used. This is to support
827 * the object header and delta base parsing routines below.
829 off_t win_off = win->offset;
830 return win_off <= offset
831 && (offset + 20) <= (win_off + win->len);
834 unsigned char *use_pack(struct packed_git *p,
835 struct pack_window **w_cursor,
839 struct pack_window *win = *w_cursor;
841 /* Since packfiles end in a hash of their content and it's
842 * pointless to ask for an offset into the middle of that
843 * hash, and the in_window function above wouldn't match
844 * don't allow an offset too close to the end of the file.
846 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
847 die("packfile %s cannot be accessed", p->pack_name);
848 if (offset > (p->pack_size - 20))
849 die("offset beyond end of packfile (truncated pack?)");
851 if (!win || !in_window(win, offset)) {
854 for (win = p->windows; win; win = win->next) {
855 if (in_window(win, offset))
859 size_t window_align = packed_git_window_size / 2;
862 if (p->pack_fd == -1 && open_packed_git(p))
863 die("packfile %s cannot be accessed", p->pack_name);
865 win = xcalloc(1, sizeof(*win));
866 win->offset = (offset / window_align) * window_align;
867 len = p->pack_size - win->offset;
868 if (len > packed_git_window_size)
869 len = packed_git_window_size;
870 win->len = (size_t)len;
871 pack_mapped += win->len;
872 while (packed_git_limit < pack_mapped
873 && unuse_one_window(p, p->pack_fd))
875 win->base = xmmap(NULL, win->len,
876 PROT_READ, MAP_PRIVATE,
877 p->pack_fd, win->offset);
878 if (win->base == MAP_FAILED)
879 die("packfile %s cannot be mapped: %s",
882 if (!win->offset && win->len == p->pack_size
883 && !p->do_not_close) {
890 if (pack_mapped > peak_pack_mapped)
891 peak_pack_mapped = pack_mapped;
892 if (pack_open_windows > peak_pack_open_windows)
893 peak_pack_open_windows = pack_open_windows;
894 win->next = p->windows;
898 if (win != *w_cursor) {
899 win->last_used = pack_used_ctr++;
903 offset -= win->offset;
905 *left = win->len - xsize_t(offset);
906 return win->base + offset;
909 static struct packed_git *alloc_packed_git(int extra)
911 struct packed_git *p = xmalloc(sizeof(*p) + extra);
912 memset(p, 0, sizeof(*p));
917 static void try_to_free_pack_memory(size_t size)
919 release_pack_memory(size, -1);
922 struct packed_git *add_packed_git(const char *path, int path_len, int local)
924 static int have_set_try_to_free_routine;
926 struct packed_git *p = alloc_packed_git(path_len + 2);
928 if (!have_set_try_to_free_routine) {
929 have_set_try_to_free_routine = 1;
930 set_try_to_free_routine(try_to_free_pack_memory);
934 * Make sure a corresponding .pack file exists and that
935 * the index looks sane.
937 path_len -= strlen(".idx");
942 memcpy(p->pack_name, path, path_len);
944 strcpy(p->pack_name + path_len, ".keep");
945 if (!access(p->pack_name, F_OK))
948 strcpy(p->pack_name + path_len, ".pack");
949 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
954 /* ok, it looks sane as far as we can check without
955 * actually mapping the pack file.
957 p->pack_size = st.st_size;
958 p->pack_local = local;
959 p->mtime = st.st_mtime;
960 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
965 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
967 const char *path = sha1_pack_name(sha1);
968 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
970 strcpy(p->pack_name, path);
971 hashcpy(p->sha1, sha1);
972 if (check_packed_git_idx(idx_path, p)) {
980 void install_packed_git(struct packed_git *pack)
982 if (pack->pack_fd != -1)
985 pack->next = packed_git;
989 static void prepare_packed_git_one(char *objdir, int local)
991 /* Ensure that this buffer is large enough so that we can
992 append "/pack/" without clobbering the stack even if
993 strlen(objdir) were PATH_MAX. */
994 char path[PATH_MAX + 1 + 4 + 1 + 1];
999 sprintf(path, "%s/pack", objdir);
1001 dir = opendir(path);
1003 if (errno != ENOENT)
1004 error("unable to open object pack directory: %s: %s",
1005 path, strerror(errno));
1009 while ((de = readdir(dir)) != NULL) {
1010 int namelen = strlen(de->d_name);
1011 struct packed_git *p;
1013 if (!has_extension(de->d_name, ".idx"))
1016 if (len + namelen + 1 > sizeof(path))
1019 /* Don't reopen a pack we already have. */
1020 strcpy(path + len, de->d_name);
1021 for (p = packed_git; p; p = p->next) {
1022 if (!memcmp(path, p->pack_name, len + namelen - 4))
1027 /* See if it really is a valid .idx file with corresponding
1028 * .pack file that we can map.
1030 p = add_packed_git(path, len + namelen, local);
1033 install_packed_git(p);
1038 static int sort_pack(const void *a_, const void *b_)
1040 struct packed_git *a = *((struct packed_git **)a_);
1041 struct packed_git *b = *((struct packed_git **)b_);
1045 * Local packs tend to contain objects specific to our
1046 * variant of the project than remote ones. In addition,
1047 * remote ones could be on a network mounted filesystem.
1048 * Favor local ones for these reasons.
1050 st = a->pack_local - b->pack_local;
1055 * Younger packs tend to contain more recent objects,
1056 * and more recent objects tend to get accessed more
1059 if (a->mtime < b->mtime)
1061 else if (a->mtime == b->mtime)
1066 static void rearrange_packed_git(void)
1068 struct packed_git **ary, *p;
1071 for (n = 0, p = packed_git; p; p = p->next)
1076 /* prepare an array of packed_git for easier sorting */
1077 ary = xcalloc(n, sizeof(struct packed_git *));
1078 for (n = 0, p = packed_git; p; p = p->next)
1081 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1083 /* link them back again */
1084 for (i = 0; i < n - 1; i++)
1085 ary[i]->next = ary[i + 1];
1086 ary[n - 1]->next = NULL;
1087 packed_git = ary[0];
1092 static int prepare_packed_git_run_once = 0;
1093 void prepare_packed_git(void)
1095 struct alternate_object_database *alt;
1097 if (prepare_packed_git_run_once)
1099 prepare_packed_git_one(get_object_directory(), 1);
1101 for (alt = alt_odb_list; alt; alt = alt->next) {
1103 prepare_packed_git_one(alt->base, 0);
1104 alt->name[-1] = '/';
1106 rearrange_packed_git();
1107 prepare_packed_git_run_once = 1;
1110 void reprepare_packed_git(void)
1113 prepare_packed_git_run_once = 0;
1114 prepare_packed_git();
1117 static void mark_bad_packed_object(struct packed_git *p,
1118 const unsigned char *sha1)
1121 for (i = 0; i < p->num_bad_objects; i++)
1122 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1124 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1125 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1126 p->num_bad_objects++;
1129 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1131 struct packed_git *p;
1134 for (p = packed_git; p; p = p->next)
1135 for (i = 0; i < p->num_bad_objects; i++)
1136 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1141 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1143 unsigned char real_sha1[20];
1144 hash_sha1_file(map, size, type, real_sha1);
1145 return hashcmp(sha1, real_sha1) ? -1 : 0;
1148 static int git_open_noatime(const char *name)
1150 static int sha1_file_open_flag = O_NOATIME;
1153 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1157 /* Might the failure be due to O_NOATIME? */
1158 if (errno != ENOENT && sha1_file_open_flag) {
1159 sha1_file_open_flag = 0;
1167 static int open_sha1_file(const unsigned char *sha1)
1170 char *name = sha1_file_name(sha1);
1171 struct alternate_object_database *alt;
1173 fd = git_open_noatime(name);
1179 for (alt = alt_odb_list; alt; alt = alt->next) {
1181 fill_sha1_path(name, sha1);
1182 fd = git_open_noatime(alt->base);
1189 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1194 fd = open_sha1_file(sha1);
1199 if (!fstat(fd, &st)) {
1200 *size = xsize_t(st.st_size);
1201 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1209 * There used to be a second loose object header format which
1210 * was meant to mimic the in-pack format, allowing for direct
1211 * copy of the object data. This format turned up not to be
1212 * really worth it and we no longer write loose objects in that
1215 static int experimental_loose_object(unsigned char *map)
1220 * We must determine if the buffer contains the standard
1221 * zlib-deflated stream or the experimental format based
1222 * on the in-pack object format. Compare the header byte
1225 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1226 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1228 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1229 * in standard loose-object format, UNLESS it is a Git-pack
1230 * format object *exactly* 8 bytes in size when inflated.
1232 * However, RFC1950 also specifies that the 1st 16-bit word
1233 * must be divisible by 31 - this checksum tells us our buffer
1234 * is in the standard format, giving a false positive only if
1235 * the 1st word of the Git-pack format object happens to be
1236 * divisible by 31, ie:
1237 * ((byte0 * 256) + byte1) % 31 = 0
1238 * => 0ttt10000www1000 % 31 = 0
1240 * As it happens, this case can only arise for www=3 & ttt=1
1241 * - ie, a Commit object, which would have to be 8 bytes in
1242 * size. As no Commit can be that small, we find that the
1243 * combination of these two criteria (bitmask & checksum)
1244 * can always correctly determine the buffer format.
1246 word = (map[0] << 8) + map[1];
1247 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1253 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1254 unsigned long len, enum object_type *type, unsigned long *sizep)
1257 unsigned long size, c;
1258 unsigned long used = 0;
1261 *type = (c >> 4) & 7;
1265 if (len <= used || bitsizeof(long) <= shift) {
1266 error("bad object header");
1270 size += (c & 0x7f) << shift;
1277 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1279 unsigned long size, used;
1280 static const char valid_loose_object_type[8] = {
1282 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1283 0, /* "delta" and others are invalid in a loose object */
1285 enum object_type type;
1287 /* Get the data stream */
1288 memset(stream, 0, sizeof(*stream));
1289 stream->next_in = map;
1290 stream->avail_in = mapsize;
1291 stream->next_out = buffer;
1292 stream->avail_out = bufsiz;
1294 if (experimental_loose_object(map)) {
1296 * The old experimental format we no longer produce;
1297 * we can still read it.
1299 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1300 if (!used || !valid_loose_object_type[type])
1305 /* Set up the stream for the rest.. */
1306 stream->next_in = map;
1307 stream->avail_in = mapsize;
1308 git_inflate_init(stream);
1310 /* And generate the fake traditional header */
1311 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1312 typename(type), size);
1315 git_inflate_init(stream);
1316 return git_inflate(stream, 0);
1319 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1321 int bytes = strlen(buffer) + 1;
1322 unsigned char *buf = xmallocz(size);
1326 n = stream->total_out - bytes;
1329 memcpy(buf, (char *) buffer + bytes, n);
1331 if (bytes <= size) {
1333 * The above condition must be (bytes <= size), not
1334 * (bytes < size). In other words, even though we
1335 * expect no more output and set avail_out to zero,
1336 * the input zlib stream may have bytes that express
1337 * "this concludes the stream", and we *do* want to
1340 * Otherwise we would not be able to test that we
1341 * consumed all the input to reach the expected size;
1342 * we also want to check that zlib tells us that all
1343 * went well with status == Z_STREAM_END at the end.
1345 stream->next_out = buf + bytes;
1346 stream->avail_out = size - bytes;
1347 while (status == Z_OK)
1348 status = git_inflate(stream, Z_FINISH);
1350 if (status == Z_STREAM_END && !stream->avail_in) {
1351 git_inflate_end(stream);
1356 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1357 else if (stream->avail_in)
1358 error("garbage at end of loose object '%s'",
1365 * We used to just use "sscanf()", but that's actually way
1366 * too permissive for what we want to check. So do an anal
1367 * object header parse by hand.
1369 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1376 * The type can be at most ten bytes (including the
1377 * terminating '\0' that we add), and is followed by
1386 if (i >= sizeof(type))
1392 * The length must follow immediately, and be in canonical
1393 * decimal format (ie "010" is not valid).
1395 size = *hdr++ - '0';
1400 unsigned long c = *hdr - '0';
1404 size = size * 10 + c;
1410 * The length must be followed by a zero byte
1412 return *hdr ? -1 : type_from_string(type);
1415 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1421 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1422 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1425 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1428 unsigned long get_size_from_delta(struct packed_git *p,
1429 struct pack_window **w_curs,
1432 const unsigned char *data;
1433 unsigned char delta_head[20], *in;
1437 memset(&stream, 0, sizeof(stream));
1438 stream.next_out = delta_head;
1439 stream.avail_out = sizeof(delta_head);
1441 git_inflate_init(&stream);
1443 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1444 stream.next_in = in;
1445 st = git_inflate(&stream, Z_FINISH);
1446 curpos += stream.next_in - in;
1447 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1448 stream.total_out < sizeof(delta_head));
1449 git_inflate_end(&stream);
1450 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1451 error("delta data unpack-initial failed");
1455 /* Examine the initial part of the delta to figure out
1460 /* ignore base size */
1461 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1463 /* Read the result size */
1464 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1467 static off_t get_delta_base(struct packed_git *p,
1468 struct pack_window **w_curs,
1470 enum object_type type,
1471 off_t delta_obj_offset)
1473 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1476 /* use_pack() assured us we have [base_info, base_info + 20)
1477 * as a range that we can look at without walking off the
1478 * end of the mapped window. Its actually the hash size
1479 * that is assured. An OFS_DELTA longer than the hash size
1480 * is stupid, as then a REF_DELTA would be smaller to store.
1482 if (type == OBJ_OFS_DELTA) {
1484 unsigned char c = base_info[used++];
1485 base_offset = c & 127;
1488 if (!base_offset || MSB(base_offset, 7))
1489 return 0; /* overflow */
1490 c = base_info[used++];
1491 base_offset = (base_offset << 7) + (c & 127);
1493 base_offset = delta_obj_offset - base_offset;
1494 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1495 return 0; /* out of bound */
1497 } else if (type == OBJ_REF_DELTA) {
1498 /* The base entry _must_ be in the same pack */
1499 base_offset = find_pack_entry_one(base_info, p);
1502 die("I am totally screwed");
1506 /* forward declaration for a mutually recursive function */
1507 static int packed_object_info(struct packed_git *p, off_t offset,
1508 unsigned long *sizep);
1510 static int packed_delta_info(struct packed_git *p,
1511 struct pack_window **w_curs,
1513 enum object_type type,
1515 unsigned long *sizep)
1519 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1522 type = packed_object_info(p, base_offset, NULL);
1523 if (type <= OBJ_NONE) {
1524 struct revindex_entry *revidx;
1525 const unsigned char *base_sha1;
1526 revidx = find_pack_revindex(p, base_offset);
1529 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1530 mark_bad_packed_object(p, base_sha1);
1531 type = sha1_object_info(base_sha1, NULL);
1532 if (type <= OBJ_NONE)
1536 /* We choose to only get the type of the base object and
1537 * ignore potentially corrupt pack file that expects the delta
1538 * based on a base with a wrong size. This saves tons of
1542 *sizep = get_size_from_delta(p, w_curs, curpos);
1550 static int unpack_object_header(struct packed_git *p,
1551 struct pack_window **w_curs,
1553 unsigned long *sizep)
1555 unsigned char *base;
1558 enum object_type type;
1560 /* use_pack() assures us we have [base, base + 20) available
1561 * as a range that we can look at. (Its actually the hash
1562 * size that is assured.) With our object header encoding
1563 * the maximum deflated object size is 2^137, which is just
1564 * insane, so we know won't exceed what we have been given.
1566 base = use_pack(p, w_curs, *curpos, &left);
1567 used = unpack_object_header_buffer(base, left, &type, sizep);
1576 const char *packed_object_info_detail(struct packed_git *p,
1578 unsigned long *size,
1579 unsigned long *store_size,
1580 unsigned int *delta_chain_length,
1581 unsigned char *base_sha1)
1583 struct pack_window *w_curs = NULL;
1585 unsigned long dummy;
1586 unsigned char *next_sha1;
1587 enum object_type type;
1588 struct revindex_entry *revidx;
1590 *delta_chain_length = 0;
1591 curpos = obj_offset;
1592 type = unpack_object_header(p, &w_curs, &curpos, size);
1594 revidx = find_pack_revindex(p, obj_offset);
1595 *store_size = revidx[1].offset - obj_offset;
1600 die("pack %s contains unknown object type %d",
1601 p->pack_name, type);
1606 unuse_pack(&w_curs);
1607 return typename(type);
1609 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1611 die("pack %s contains bad delta base reference of type %s",
1612 p->pack_name, typename(type));
1613 if (*delta_chain_length == 0) {
1614 revidx = find_pack_revindex(p, obj_offset);
1615 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1619 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1620 if (*delta_chain_length == 0)
1621 hashcpy(base_sha1, next_sha1);
1622 obj_offset = find_pack_entry_one(next_sha1, p);
1625 (*delta_chain_length)++;
1626 curpos = obj_offset;
1627 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1631 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1632 unsigned long *sizep)
1634 struct pack_window *w_curs = NULL;
1636 off_t curpos = obj_offset;
1637 enum object_type type;
1639 type = unpack_object_header(p, &w_curs, &curpos, &size);
1644 type = packed_delta_info(p, &w_curs, curpos,
1645 type, obj_offset, sizep);
1655 error("unknown object type %i at offset %"PRIuMAX" in %s",
1656 type, (uintmax_t)obj_offset, p->pack_name);
1659 unuse_pack(&w_curs);
1663 static void *unpack_compressed_entry(struct packed_git *p,
1664 struct pack_window **w_curs,
1670 unsigned char *buffer, *in;
1672 buffer = xmallocz(size);
1673 memset(&stream, 0, sizeof(stream));
1674 stream.next_out = buffer;
1675 stream.avail_out = size + 1;
1677 git_inflate_init(&stream);
1679 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1680 stream.next_in = in;
1681 st = git_inflate(&stream, Z_FINISH);
1682 if (!stream.avail_out)
1683 break; /* the payload is larger than it should be */
1684 curpos += stream.next_in - in;
1685 } while (st == Z_OK || st == Z_BUF_ERROR);
1686 git_inflate_end(&stream);
1687 if ((st != Z_STREAM_END) || stream.total_out != size) {
1695 #define MAX_DELTA_CACHE (256)
1697 static size_t delta_base_cached;
1699 static struct delta_base_cache_lru_list {
1700 struct delta_base_cache_lru_list *prev;
1701 struct delta_base_cache_lru_list *next;
1702 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1704 static struct delta_base_cache_entry {
1705 struct delta_base_cache_lru_list lru;
1707 struct packed_git *p;
1710 enum object_type type;
1711 } delta_base_cache[MAX_DELTA_CACHE];
1713 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1717 hash = (unsigned long)p + (unsigned long)base_offset;
1718 hash += (hash >> 8) + (hash >> 16);
1719 return hash % MAX_DELTA_CACHE;
1722 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1723 unsigned long *base_size, enum object_type *type, int keep_cache)
1726 unsigned long hash = pack_entry_hash(p, base_offset);
1727 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1730 if (!ret || ent->p != p || ent->base_offset != base_offset)
1731 return unpack_entry(p, base_offset, type, base_size);
1735 ent->lru.next->prev = ent->lru.prev;
1736 ent->lru.prev->next = ent->lru.next;
1737 delta_base_cached -= ent->size;
1739 ret = xmemdupz(ent->data, ent->size);
1742 *base_size = ent->size;
1746 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1751 ent->lru.next->prev = ent->lru.prev;
1752 ent->lru.prev->next = ent->lru.next;
1753 delta_base_cached -= ent->size;
1757 void clear_delta_base_cache(void)
1760 for (p = 0; p < MAX_DELTA_CACHE; p++)
1761 release_delta_base_cache(&delta_base_cache[p]);
1764 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1765 void *base, unsigned long base_size, enum object_type type)
1767 unsigned long hash = pack_entry_hash(p, base_offset);
1768 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1769 struct delta_base_cache_lru_list *lru;
1771 release_delta_base_cache(ent);
1772 delta_base_cached += base_size;
1774 for (lru = delta_base_cache_lru.next;
1775 delta_base_cached > delta_base_cache_limit
1776 && lru != &delta_base_cache_lru;
1778 struct delta_base_cache_entry *f = (void *)lru;
1779 if (f->type == OBJ_BLOB)
1780 release_delta_base_cache(f);
1782 for (lru = delta_base_cache_lru.next;
1783 delta_base_cached > delta_base_cache_limit
1784 && lru != &delta_base_cache_lru;
1786 struct delta_base_cache_entry *f = (void *)lru;
1787 release_delta_base_cache(f);
1791 ent->base_offset = base_offset;
1794 ent->size = base_size;
1795 ent->lru.next = &delta_base_cache_lru;
1796 ent->lru.prev = delta_base_cache_lru.prev;
1797 delta_base_cache_lru.prev->next = &ent->lru;
1798 delta_base_cache_lru.prev = &ent->lru;
1801 static void *read_object(const unsigned char *sha1, enum object_type *type,
1802 unsigned long *size);
1804 static void *unpack_delta_entry(struct packed_git *p,
1805 struct pack_window **w_curs,
1807 unsigned long delta_size,
1809 enum object_type *type,
1810 unsigned long *sizep)
1812 void *delta_data, *result, *base;
1813 unsigned long base_size;
1816 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1818 error("failed to validate delta base reference "
1819 "at offset %"PRIuMAX" from %s",
1820 (uintmax_t)curpos, p->pack_name);
1824 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1827 * We're probably in deep shit, but let's try to fetch
1828 * the required base anyway from another pack or loose.
1829 * This is costly but should happen only in the presence
1830 * of a corrupted pack, and is better than failing outright.
1832 struct revindex_entry *revidx;
1833 const unsigned char *base_sha1;
1834 revidx = find_pack_revindex(p, base_offset);
1837 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1838 error("failed to read delta base object %s"
1839 " at offset %"PRIuMAX" from %s",
1840 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1842 mark_bad_packed_object(p, base_sha1);
1843 base = read_object(base_sha1, type, &base_size);
1848 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1850 error("failed to unpack compressed delta "
1851 "at offset %"PRIuMAX" from %s",
1852 (uintmax_t)curpos, p->pack_name);
1856 result = patch_delta(base, base_size,
1857 delta_data, delta_size,
1860 die("failed to apply delta");
1862 add_delta_base_cache(p, base_offset, base, base_size, *type);
1866 int do_check_packed_object_crc;
1868 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1869 enum object_type *type, unsigned long *sizep)
1871 struct pack_window *w_curs = NULL;
1872 off_t curpos = obj_offset;
1875 if (do_check_packed_object_crc && p->index_version > 1) {
1876 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1877 unsigned long len = revidx[1].offset - obj_offset;
1878 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1879 const unsigned char *sha1 =
1880 nth_packed_object_sha1(p, revidx->nr);
1881 error("bad packed object CRC for %s",
1883 mark_bad_packed_object(p, sha1);
1884 unuse_pack(&w_curs);
1889 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1893 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1894 obj_offset, type, sizep);
1900 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1904 error("unknown object type %i at offset %"PRIuMAX" in %s",
1905 *type, (uintmax_t)obj_offset, p->pack_name);
1907 unuse_pack(&w_curs);
1911 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1914 const unsigned char *index = p->index_data;
1916 if (open_pack_index(p))
1918 index = p->index_data;
1920 if (n >= p->num_objects)
1923 if (p->index_version == 1) {
1924 return index + 24 * n + 4;
1927 return index + 20 * n;
1931 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1933 const unsigned char *index = p->index_data;
1935 if (p->index_version == 1) {
1936 return ntohl(*((uint32_t *)(index + 24 * n)));
1939 index += 8 + p->num_objects * (20 + 4);
1940 off = ntohl(*((uint32_t *)(index + 4 * n)));
1941 if (!(off & 0x80000000))
1943 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1944 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1945 ntohl(*((uint32_t *)(index + 4)));
1949 off_t find_pack_entry_one(const unsigned char *sha1,
1950 struct packed_git *p)
1952 const uint32_t *level1_ofs = p->index_data;
1953 const unsigned char *index = p->index_data;
1954 unsigned hi, lo, stride;
1955 static int use_lookup = -1;
1956 static int debug_lookup = -1;
1958 if (debug_lookup < 0)
1959 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1962 if (open_pack_index(p))
1964 level1_ofs = p->index_data;
1965 index = p->index_data;
1967 if (p->index_version > 1) {
1972 hi = ntohl(level1_ofs[*sha1]);
1973 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1974 if (p->index_version > 1) {
1982 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1983 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1986 use_lookup = !!getenv("GIT_USE_LOOKUP");
1988 int pos = sha1_entry_pos(index, stride, 0,
1989 lo, hi, p->num_objects, sha1);
1992 return nth_packed_object_offset(p, pos);
1996 unsigned mi = (lo + hi) / 2;
1997 int cmp = hashcmp(index + mi * stride, sha1);
2000 printf("lo %u hi %u rg %u mi %u\n",
2001 lo, hi, hi - lo, mi);
2003 return nth_packed_object_offset(p, mi);
2012 static int is_pack_valid(struct packed_git *p)
2014 /* An already open pack is known to be valid. */
2015 if (p->pack_fd != -1)
2018 /* If the pack has one window completely covering the
2019 * file size, the pack is known to be valid even if
2020 * the descriptor is not currently open.
2023 struct pack_window *w = p->windows;
2025 if (!w->offset && w->len == p->pack_size)
2029 /* Force the pack to open to prove its valid. */
2030 return !open_packed_git(p);
2033 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2035 static struct packed_git *last_found = (void *)1;
2036 struct packed_git *p;
2039 prepare_packed_git();
2042 p = (last_found == (void *)1) ? packed_git : last_found;
2045 if (p->num_bad_objects) {
2047 for (i = 0; i < p->num_bad_objects; i++)
2048 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2052 offset = find_pack_entry_one(sha1, p);
2055 * We are about to tell the caller where they can
2056 * locate the requested object. We better make
2057 * sure the packfile is still here and can be
2058 * accessed before supplying that answer, as
2059 * it may have been deleted since the index
2062 if (!is_pack_valid(p)) {
2063 error("packfile %s cannot be accessed", p->pack_name);
2068 hashcpy(e->sha1, sha1);
2074 if (p == last_found)
2078 if (p == last_found)
2084 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2085 struct packed_git *packs)
2087 struct packed_git *p;
2089 for (p = packs; p; p = p->next) {
2090 if (find_pack_entry_one(sha1, p))
2097 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2100 unsigned long mapsize, size;
2105 map = map_sha1_file(sha1, &mapsize);
2107 return error("unable to find %s", sha1_to_hex(sha1));
2108 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2109 status = error("unable to unpack %s header",
2111 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2112 status = error("unable to parse %s header", sha1_to_hex(sha1));
2115 git_inflate_end(&stream);
2116 munmap(map, mapsize);
2120 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2122 struct cached_object *co;
2123 struct pack_entry e;
2126 co = find_cached_object(sha1);
2133 if (!find_pack_entry(sha1, &e)) {
2134 /* Most likely it's a loose object. */
2135 status = sha1_loose_object_info(sha1, sizep);
2139 /* Not a loose object; someone else may have just packed it. */
2140 reprepare_packed_git();
2141 if (!find_pack_entry(sha1, &e))
2145 status = packed_object_info(e.p, e.offset, sizep);
2147 mark_bad_packed_object(e.p, sha1);
2148 status = sha1_object_info(sha1, sizep);
2154 static void *read_packed_sha1(const unsigned char *sha1,
2155 enum object_type *type, unsigned long *size)
2157 struct pack_entry e;
2160 if (!find_pack_entry(sha1, &e))
2162 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2165 * We're probably in deep shit, but let's try to fetch
2166 * the required object anyway from another pack or loose.
2167 * This should happen only in the presence of a corrupted
2168 * pack, and is better than failing outright.
2170 error("failed to read object %s at offset %"PRIuMAX" from %s",
2171 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2172 mark_bad_packed_object(e.p, sha1);
2173 data = read_object(sha1, type, size);
2178 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2179 unsigned char *sha1)
2181 struct cached_object *co;
2183 hash_sha1_file(buf, len, typename(type), sha1);
2184 if (has_sha1_file(sha1) || find_cached_object(sha1))
2186 if (cached_object_alloc <= cached_object_nr) {
2187 cached_object_alloc = alloc_nr(cached_object_alloc);
2188 cached_objects = xrealloc(cached_objects,
2189 sizeof(*cached_objects) *
2190 cached_object_alloc);
2192 co = &cached_objects[cached_object_nr++];
2195 co->buf = xmalloc(len);
2196 memcpy(co->buf, buf, len);
2197 hashcpy(co->sha1, sha1);
2201 static void *read_object(const unsigned char *sha1, enum object_type *type,
2202 unsigned long *size)
2204 unsigned long mapsize;
2206 struct cached_object *co;
2208 co = find_cached_object(sha1);
2212 return xmemdupz(co->buf, co->size);
2215 buf = read_packed_sha1(sha1, type, size);
2218 map = map_sha1_file(sha1, &mapsize);
2220 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2221 munmap(map, mapsize);
2224 reprepare_packed_git();
2225 return read_packed_sha1(sha1, type, size);
2229 * This function dies on corrupt objects; the callers who want to
2230 * deal with them should arrange to call read_object() and give error
2231 * messages themselves.
2233 void *read_sha1_file_extended(const unsigned char *sha1,
2234 enum object_type *type,
2235 unsigned long *size,
2240 const struct packed_git *p;
2241 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2242 ? lookup_replace_object(sha1) : sha1;
2245 data = read_object(repl, type, size);
2249 if (errno && errno != ENOENT)
2250 die_errno("failed to read object %s", sha1_to_hex(sha1));
2252 /* die if we replaced an object with one that does not exist */
2254 die("replacement %s not found for %s",
2255 sha1_to_hex(repl), sha1_to_hex(sha1));
2257 if (has_loose_object(repl)) {
2258 path = sha1_file_name(sha1);
2259 die("loose object %s (stored in %s) is corrupt",
2260 sha1_to_hex(repl), path);
2263 if ((p = has_packed_and_bad(repl)) != NULL)
2264 die("packed object %s (stored in %s) is corrupt",
2265 sha1_to_hex(repl), p->pack_name);
2270 void *read_object_with_reference(const unsigned char *sha1,
2271 const char *required_type_name,
2272 unsigned long *size,
2273 unsigned char *actual_sha1_return)
2275 enum object_type type, required_type;
2277 unsigned long isize;
2278 unsigned char actual_sha1[20];
2280 required_type = type_from_string(required_type_name);
2281 hashcpy(actual_sha1, sha1);
2283 int ref_length = -1;
2284 const char *ref_type = NULL;
2286 buffer = read_sha1_file(actual_sha1, &type, &isize);
2289 if (type == required_type) {
2291 if (actual_sha1_return)
2292 hashcpy(actual_sha1_return, actual_sha1);
2295 /* Handle references */
2296 else if (type == OBJ_COMMIT)
2298 else if (type == OBJ_TAG)
2299 ref_type = "object ";
2304 ref_length = strlen(ref_type);
2306 if (ref_length + 40 > isize ||
2307 memcmp(buffer, ref_type, ref_length) ||
2308 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2313 /* Now we have the ID of the referred-to object in
2314 * actual_sha1. Check again. */
2318 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2319 const char *type, unsigned char *sha1,
2320 char *hdr, int *hdrlen)
2324 /* Generate the header */
2325 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2329 git_SHA1_Update(&c, hdr, *hdrlen);
2330 git_SHA1_Update(&c, buf, len);
2331 git_SHA1_Final(sha1, &c);
2335 * Move the just written object into its final resting place.
2336 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2337 * "moving" is only a part of what it does, when no patch between
2338 * master to pu changes the call sites of this function.
2340 int move_temp_to_file(const char *tmpfile, const char *filename)
2344 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2346 else if (link(tmpfile, filename))
2350 * Coda hack - coda doesn't like cross-directory links,
2351 * so we fall back to a rename, which will mean that it
2352 * won't be able to check collisions, but that's not a
2355 * The same holds for FAT formatted media.
2357 * When this succeeds, we just return. We have nothing
2360 if (ret && ret != EEXIST) {
2362 if (!rename(tmpfile, filename))
2366 unlink_or_warn(tmpfile);
2368 if (ret != EEXIST) {
2369 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2371 /* FIXME!!! Collision check here ? */
2375 if (adjust_shared_perm(filename))
2376 return error("unable to set permission to '%s'", filename);
2380 static int write_buffer(int fd, const void *buf, size_t len)
2382 if (write_in_full(fd, buf, len) < 0)
2383 return error("file write error (%s)", strerror(errno));
2387 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2388 unsigned char *sha1)
2392 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2396 /* Finalize a file on disk, and close it. */
2397 static void close_sha1_file(int fd)
2399 if (fsync_object_files)
2400 fsync_or_die(fd, "sha1 file");
2402 die_errno("error when closing sha1 file");
2405 /* Size of directory component, including the ending '/' */
2406 static inline int directory_size(const char *filename)
2408 const char *s = strrchr(filename, '/');
2411 return s - filename + 1;
2415 * This creates a temporary file in the same directory as the final
2418 * We want to avoid cross-directory filename renames, because those
2419 * can have problems on various filesystems (FAT, NFS, Coda).
2421 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2423 int fd, dirlen = directory_size(filename);
2425 if (dirlen + 20 > bufsiz) {
2426 errno = ENAMETOOLONG;
2429 memcpy(buffer, filename, dirlen);
2430 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2431 fd = git_mkstemp_mode(buffer, 0444);
2432 if (fd < 0 && dirlen && errno == ENOENT) {
2433 /* Make sure the directory exists */
2434 memcpy(buffer, filename, dirlen);
2435 buffer[dirlen-1] = 0;
2436 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2440 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2441 fd = git_mkstemp_mode(buffer, 0444);
2446 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2447 const void *buf, unsigned long len, time_t mtime)
2450 unsigned char compressed[4096];
2453 unsigned char parano_sha1[20];
2455 static char tmpfile[PATH_MAX];
2457 filename = sha1_file_name(sha1);
2458 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2460 if (errno == EACCES)
2461 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2463 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2467 memset(&stream, 0, sizeof(stream));
2468 deflateInit(&stream, zlib_compression_level);
2469 stream.next_out = compressed;
2470 stream.avail_out = sizeof(compressed);
2473 /* First header.. */
2474 stream.next_in = (unsigned char *)hdr;
2475 stream.avail_in = hdrlen;
2476 while (deflate(&stream, 0) == Z_OK)
2478 git_SHA1_Update(&c, hdr, hdrlen);
2480 /* Then the data itself.. */
2481 stream.next_in = (void *)buf;
2482 stream.avail_in = len;
2484 unsigned char *in0 = stream.next_in;
2485 ret = deflate(&stream, Z_FINISH);
2486 git_SHA1_Update(&c, in0, stream.next_in - in0);
2487 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2488 die("unable to write sha1 file");
2489 stream.next_out = compressed;
2490 stream.avail_out = sizeof(compressed);
2491 } while (ret == Z_OK);
2493 if (ret != Z_STREAM_END)
2494 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2495 ret = deflateEnd(&stream);
2497 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2498 git_SHA1_Final(parano_sha1, &c);
2499 if (hashcmp(sha1, parano_sha1) != 0)
2500 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2502 close_sha1_file(fd);
2507 utb.modtime = mtime;
2508 if (utime(tmpfile, &utb) < 0)
2509 warning("failed utime() on %s: %s",
2510 tmpfile, strerror(errno));
2513 return move_temp_to_file(tmpfile, filename);
2516 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2518 unsigned char sha1[20];
2522 /* Normally if we have it in the pack then we do not bother writing
2523 * it out into .git/objects/??/?{38} file.
2525 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2527 hashcpy(returnsha1, sha1);
2528 if (has_sha1_file(sha1))
2530 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2533 int force_object_loose(const unsigned char *sha1, time_t mtime)
2537 enum object_type type;
2542 if (has_loose_object(sha1))
2544 buf = read_packed_sha1(sha1, &type, &len);
2546 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2547 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2548 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2554 int has_pack_index(const unsigned char *sha1)
2557 if (stat(sha1_pack_index_name(sha1), &st))
2562 int has_sha1_pack(const unsigned char *sha1)
2564 struct pack_entry e;
2565 return find_pack_entry(sha1, &e);
2568 int has_sha1_file(const unsigned char *sha1)
2570 struct pack_entry e;
2572 if (find_pack_entry(sha1, &e))
2574 return has_loose_object(sha1);
2577 static void check_tree(const void *buf, size_t size)
2579 struct tree_desc desc;
2580 struct name_entry entry;
2582 init_tree_desc(&desc, buf, size);
2583 while (tree_entry(&desc, &entry))
2585 * tree_entry() will die() on malformed entries */
2589 static void check_commit(const void *buf, size_t size)
2592 memset(&c, 0, sizeof(c));
2593 if (parse_commit_buffer(&c, buf, size))
2594 die("corrupt commit");
2597 static void check_tag(const void *buf, size_t size)
2600 memset(&t, 0, sizeof(t));
2601 if (parse_tag_buffer(&t, buf, size))
2605 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2606 enum object_type type,
2607 const char *path, unsigned flags)
2609 int ret, re_allocated = 0;
2610 int write_object = flags & HASH_WRITE_OBJECT;
2616 * Convert blobs to git internal format
2618 if ((type == OBJ_BLOB) && path) {
2619 struct strbuf nbuf = STRBUF_INIT;
2620 if (convert_to_git(path, buf, size, &nbuf,
2621 write_object ? safe_crlf : 0)) {
2622 buf = strbuf_detach(&nbuf, &size);
2626 if (flags & HASH_FORMAT_CHECK) {
2627 if (type == OBJ_TREE)
2628 check_tree(buf, size);
2629 if (type == OBJ_COMMIT)
2630 check_commit(buf, size);
2631 if (type == OBJ_TAG)
2632 check_tag(buf, size);
2636 ret = write_sha1_file(buf, size, typename(type), sha1);
2638 ret = hash_sha1_file(buf, size, typename(type), sha1);
2644 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2645 const char *path, unsigned flags)
2647 struct strbuf sbuf = STRBUF_INIT;
2650 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2651 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2654 strbuf_release(&sbuf);
2658 #define SMALL_FILE_SIZE (32*1024)
2660 static int index_core(unsigned char *sha1, int fd, size_t size,
2661 enum object_type type, const char *path,
2667 ret = index_mem(sha1, NULL, size, type, path, flags);
2668 } else if (size <= SMALL_FILE_SIZE) {
2669 char *buf = xmalloc(size);
2670 if (size == read_in_full(fd, buf, size))
2671 ret = index_mem(sha1, buf, size, type, path, flags);
2673 ret = error("short read %s", strerror(errno));
2676 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2677 ret = index_mem(sha1, buf, size, type, path, flags);
2684 * This creates one packfile per large blob, because the caller
2685 * immediately wants the result sha1, and fast-import can report the
2686 * object name via marks mechanism only by closing the created
2689 * This also bypasses the usual "convert-to-git" dance, and that is on
2690 * purpose. We could write a streaming version of the converting
2691 * functions and insert that before feeding the data to fast-import
2692 * (or equivalent in-core API described above), but the primary
2693 * motivation for trying to stream from the working tree file and to
2694 * avoid mmaping it in core is to deal with large binary blobs, and
2695 * by definition they do _not_ want to get any conversion.
2697 static int index_stream(unsigned char *sha1, int fd, size_t size,
2698 enum object_type type, const char *path,
2701 struct child_process fast_import;
2702 char export_marks[512];
2703 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2705 char fast_import_cmd[512];
2709 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2710 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2712 die_errno("cannot create tempfile: %s", tmpfile);
2714 die_errno("cannot close tempfile: %s", tmpfile);
2715 sprintf(export_marks, "--export-marks=%s", tmpfile);
2717 memset(&fast_import, 0, sizeof(fast_import));
2718 fast_import.in = -1;
2719 fast_import.argv = argv;
2720 fast_import.git_cmd = 1;
2721 if (start_command(&fast_import))
2722 die_errno("index-stream: git fast-import failed");
2724 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2725 (unsigned long) size);
2726 write_or_whine(fast_import.in, fast_import_cmd, len,
2727 "index-stream: feeding fast-import");
2730 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2733 actual = read_in_full(fd, buf, sz);
2735 die_errno("index-stream: reading input");
2736 if (write_in_full(fast_import.in, buf, actual) != actual)
2737 die_errno("index-stream: feeding fast-import");
2740 if (close(fast_import.in))
2741 die_errno("index-stream: closing fast-import");
2742 if (finish_command(&fast_import))
2743 die_errno("index-stream: finishing fast-import");
2745 tmpfd = open(tmpfile, O_RDONLY);
2747 die_errno("index-stream: cannot open fast-import mark");
2748 len = read(tmpfd, buf, sizeof(buf));
2750 die_errno("index-stream: reading fast-import mark");
2751 if (close(tmpfd) < 0)
2752 die_errno("index-stream: closing fast-import mark");
2753 if (unlink(tmpfile))
2754 die_errno("index-stream: unlinking fast-import mark");
2756 memcmp(":1 ", buf, 3) ||
2757 get_sha1_hex(buf + 3, sha1))
2758 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2762 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2763 enum object_type type, const char *path, unsigned flags)
2766 size_t size = xsize_t(st->st_size);
2768 if (!S_ISREG(st->st_mode))
2769 ret = index_pipe(sha1, fd, type, path, flags);
2770 else if (size <= big_file_threshold || type != OBJ_BLOB)
2771 ret = index_core(sha1, fd, size, type, path, flags);
2773 ret = index_stream(sha1, fd, size, type, path, flags);
2778 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2781 struct strbuf sb = STRBUF_INIT;
2783 switch (st->st_mode & S_IFMT) {
2785 fd = open(path, O_RDONLY);
2787 return error("open(\"%s\"): %s", path,
2789 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2790 return error("%s: failed to insert into database",
2794 if (strbuf_readlink(&sb, path, st->st_size)) {
2795 char *errstr = strerror(errno);
2796 return error("readlink(\"%s\"): %s", path,
2799 if (!(flags & HASH_WRITE_OBJECT))
2800 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2801 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2802 return error("%s: failed to insert into database",
2804 strbuf_release(&sb);
2807 return resolve_gitlink_ref(path, "HEAD", sha1);
2809 return error("%s: unsupported file type", path);
2814 int read_pack_header(int fd, struct pack_header *header)
2816 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2817 /* "eof before pack header was fully read" */
2818 return PH_ERROR_EOF;
2820 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2821 /* "protocol error (pack signature mismatch detected)" */
2822 return PH_ERROR_PACK_SIGNATURE;
2823 if (!pack_version_ok(header->hdr_version))
2824 /* "protocol error (pack version unsupported)" */
2825 return PH_ERROR_PROTOCOL;
2829 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2831 enum object_type type = sha1_object_info(sha1, NULL);
2833 die("%s is not a valid object", sha1_to_hex(sha1));
2835 die("%s is not a valid '%s' object", sha1_to_hex(sha1),