4 #include "repository.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
15 #include "tree-walk.h"
17 #include "object-store.h"
20 char *odb_pack_name(struct strbuf *buf,
21 const unsigned char *sha1,
25 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1), ext);
30 char *sha1_pack_name(const unsigned char *sha1)
32 static struct strbuf buf = STRBUF_INIT;
33 return odb_pack_name(&buf, sha1, "pack");
36 char *sha1_pack_index_name(const unsigned char *sha1)
38 static struct strbuf buf = STRBUF_INIT;
39 return odb_pack_name(&buf, sha1, "idx");
42 static unsigned int pack_used_ctr;
43 static unsigned int pack_mmap_calls;
44 static unsigned int peak_pack_open_windows;
45 static unsigned int pack_open_windows;
46 static unsigned int pack_open_fds;
47 static unsigned int pack_max_fds;
48 static size_t peak_pack_mapped;
49 static size_t pack_mapped;
51 #define SZ_FMT PRIuMAX
52 static inline uintmax_t sz_fmt(size_t s) { return s; }
54 void pack_report(void)
57 "pack_report: getpagesize() = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size),
62 sz_fmt(packed_git_limit));
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT " / %10" SZ_FMT "\n",
71 pack_open_windows, peak_pack_open_windows,
72 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
80 static int check_packed_git_idx(const char *path, struct packed_git *p)
84 int fd = git_open(path), ret;
86 const unsigned int hashsz = the_hash_algo->rawsz;
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + hashsz + hashsz) {
97 return error("index file %s is too small", path);
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
102 ret = load_idx(path, hashsz, idx_map, idx_size, p);
105 munmap(idx_map, idx_size);
110 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
111 size_t idx_size, struct packed_git *p)
113 struct pack_idx_header *hdr = idx_map;
114 uint32_t version, nr, i, *index;
116 if (idx_size < 4 * 256 + hashsz + hashsz)
117 return error("index file %s is too small", path);
119 return error("empty data");
121 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
122 version = ntohl(hdr->idx_version);
123 if (version < 2 || version > 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
134 index += 2; /* skip index header */
135 for (i = 0; i < 256; i++) {
136 uint32_t n = ntohl(index[i]);
138 return error("non-monotonic index %s", path);
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
150 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
151 return error("wrong index v1 file size in %s", path);
152 } else if (version == 2) {
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
166 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
167 unsigned long max_size = min_size;
169 max_size += (nr - 1)*8;
170 if (idx_size < min_size || idx_size > max_size)
171 return error("wrong index v2 file size in %s", path);
172 if (idx_size != min_size &&
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
178 (sizeof(off_t) <= 4))
179 return error("pack too large for current definition of off_t in %s", path);
182 p->index_version = version;
183 p->index_data = idx_map;
184 p->index_size = idx_size;
189 int open_pack_index(struct packed_git *p)
198 if (!strip_suffix(p->pack_name, ".pack", &len))
199 BUG("pack_name does not end in .pack");
200 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
201 ret = check_packed_git_idx(idx_name, p);
206 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
208 const uint32_t *level1_ofs = p->index_data;
211 if (open_pack_index(p))
213 level1_ofs = p->index_data;
216 if (p->index_version > 1) {
220 return ntohl(level1_ofs[value]);
223 static struct packed_git *alloc_packed_git(int extra)
225 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
226 memset(p, 0, sizeof(*p));
231 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
233 const char *path = sha1_pack_name(sha1);
234 size_t alloc = st_add(strlen(path), 1);
235 struct packed_git *p = alloc_packed_git(alloc);
237 memcpy(p->pack_name, path, alloc); /* includes NUL */
238 hashcpy(p->sha1, sha1);
239 if (check_packed_git_idx(idx_path, p)) {
247 static void scan_windows(struct packed_git *p,
248 struct packed_git **lru_p,
249 struct pack_window **lru_w,
250 struct pack_window **lru_l)
252 struct pack_window *w, *w_l;
254 for (w_l = NULL, w = p->windows; w; w = w->next) {
256 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
266 static int unuse_one_window(struct packed_git *current)
268 struct packed_git *p, *lru_p = NULL;
269 struct pack_window *lru_w = NULL, *lru_l = NULL;
272 scan_windows(current, &lru_p, &lru_w, &lru_l);
273 for (p = the_repository->objects->packed_git; p; p = p->next)
274 scan_windows(p, &lru_p, &lru_w, &lru_l);
276 munmap(lru_w->base, lru_w->len);
277 pack_mapped -= lru_w->len;
279 lru_l->next = lru_w->next;
281 lru_p->windows = lru_w->next;
289 void release_pack_memory(size_t need)
291 size_t cur = pack_mapped;
292 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
296 void close_pack_windows(struct packed_git *p)
299 struct pack_window *w = p->windows;
302 die("pack '%s' still has open windows to it",
304 munmap(w->base, w->len);
305 pack_mapped -= w->len;
307 p->windows = w->next;
312 static int close_pack_fd(struct packed_git *p)
324 void close_pack_index(struct packed_git *p)
327 munmap((void *)p->index_data, p->index_size);
328 p->index_data = NULL;
332 void close_pack(struct packed_git *p)
334 close_pack_windows(p);
339 void close_all_packs(struct raw_object_store *o)
341 struct packed_git *p;
343 for (p = o->packed_git; p; p = p->next)
345 BUG("want to close pack marked 'do-not-close'");
351 * The LRU pack is the one with the oldest MRU window, preferring packs
352 * with no used windows, or the oldest mtime if it has no windows allocated.
354 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
356 struct pack_window *w, *this_mru_w;
357 int has_windows_inuse = 0;
360 * Reject this pack if it has windows and the previously selected
361 * one does not. If this pack does not have windows, reject
362 * it if the pack file is newer than the previously selected one.
364 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
367 for (w = this_mru_w = p->windows; w; w = w->next) {
369 * Reject this pack if any of its windows are in use,
370 * but the previously selected pack did not have any
371 * inuse windows. Otherwise, record that this pack
372 * has windows in use.
375 if (*accept_windows_inuse)
376 has_windows_inuse = 1;
381 if (w->last_used > this_mru_w->last_used)
385 * Reject this pack if it has windows that have been
386 * used more recently than the previously selected pack.
387 * If the previously selected pack had windows inuse and
388 * we have not encountered a window in this pack that is
389 * inuse, skip this check since we prefer a pack with no
390 * inuse windows to one that has inuse windows.
392 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
393 this_mru_w->last_used > (*mru_w)->last_used)
402 *accept_windows_inuse = has_windows_inuse;
405 static int close_one_pack(void)
407 struct packed_git *p, *lru_p = NULL;
408 struct pack_window *mru_w = NULL;
409 int accept_windows_inuse = 1;
411 for (p = the_repository->objects->packed_git; p; p = p->next) {
412 if (p->pack_fd == -1)
414 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
418 return close_pack_fd(lru_p);
423 static unsigned int get_max_fd_limit(void)
429 if (!getrlimit(RLIMIT_NOFILE, &lim))
436 long open_max = sysconf(_SC_OPEN_MAX);
440 * Otherwise, we got -1 for one of the two
443 * (1) sysconf() did not understand _SC_OPEN_MAX
444 * and signaled an error with -1; or
445 * (2) sysconf() said there is no limit.
447 * We _could_ clear errno before calling sysconf() to
448 * tell these two cases apart and return a huge number
449 * in the latter case to let the caller cap it to a
450 * value that is not so selfish, but letting the
451 * fallback OPEN_MAX codepath take care of these cases
460 return 1; /* see the caller ;-) */
465 * Do not call this directly as this leaks p->pack_fd on error return;
466 * call open_packed_git() instead.
468 static int open_packed_git_1(struct packed_git *p)
471 struct pack_header hdr;
472 unsigned char hash[GIT_MAX_RAWSZ];
473 unsigned char *idx_hash;
476 const unsigned hashsz = the_hash_algo->rawsz;
478 if (!p->index_data) {
479 struct multi_pack_index *m;
480 const char *pack_name = strrchr(p->pack_name, '/');
482 for (m = the_repository->objects->multi_pack_index;
484 if (midx_contains_pack(m, pack_name))
488 if (!m && open_pack_index(p))
489 return error("packfile %s index unavailable", p->pack_name);
493 unsigned int max_fds = get_max_fd_limit();
495 /* Save 3 for stdin/stdout/stderr, 22 for work */
497 pack_max_fds = max_fds - 25;
502 while (pack_max_fds <= pack_open_fds && close_one_pack())
505 p->pack_fd = git_open(p->pack_name);
506 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
510 /* If we created the struct before we had the pack we lack size. */
512 if (!S_ISREG(st.st_mode))
513 return error("packfile %s not a regular file", p->pack_name);
514 p->pack_size = st.st_size;
515 } else if (p->pack_size != st.st_size)
516 return error("packfile %s size changed", p->pack_name);
518 /* We leave these file descriptors open with sliding mmap;
519 * there is no point keeping them open across exec(), though.
521 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
523 return error("cannot determine file descriptor flags");
524 fd_flag |= FD_CLOEXEC;
525 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
526 return error("cannot set FD_CLOEXEC");
528 /* Verify we recognize this pack file format. */
529 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
531 return error_errno("error reading from %s", p->pack_name);
532 if (read_result != sizeof(hdr))
533 return error("file %s is far too short to be a packfile", p->pack_name);
534 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
535 return error("file %s is not a GIT packfile", p->pack_name);
536 if (!pack_version_ok(hdr.hdr_version))
537 return error("packfile %s is version %"PRIu32" and not"
538 " supported (try upgrading GIT to a newer version)",
539 p->pack_name, ntohl(hdr.hdr_version));
541 /* Skip index checking if in multi-pack-index */
545 /* Verify the pack matches its index. */
546 if (p->num_objects != ntohl(hdr.hdr_entries))
547 return error("packfile %s claims to have %"PRIu32" objects"
548 " while index indicates %"PRIu32" objects",
549 p->pack_name, ntohl(hdr.hdr_entries),
551 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
552 return error("end of packfile %s is unavailable", p->pack_name);
553 read_result = read_in_full(p->pack_fd, hash, hashsz);
555 return error_errno("error reading from %s", p->pack_name);
556 if (read_result != hashsz)
557 return error("packfile %s signature is unavailable", p->pack_name);
558 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
559 if (!hasheq(hash, idx_hash))
560 return error("packfile %s does not match index", p->pack_name);
564 static int open_packed_git(struct packed_git *p)
566 if (!open_packed_git_1(p))
572 static int in_window(struct pack_window *win, off_t offset)
574 /* We must promise at least one full hash after the
575 * offset is available from this window, otherwise the offset
576 * is not actually in this window and a different window (which
577 * has that one hash excess) must be used. This is to support
578 * the object header and delta base parsing routines below.
580 off_t win_off = win->offset;
581 return win_off <= offset
582 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
585 unsigned char *use_pack(struct packed_git *p,
586 struct pack_window **w_cursor,
590 struct pack_window *win = *w_cursor;
592 /* Since packfiles end in a hash of their content and it's
593 * pointless to ask for an offset into the middle of that
594 * hash, and the in_window function above wouldn't match
595 * don't allow an offset too close to the end of the file.
597 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
598 die("packfile %s cannot be accessed", p->pack_name);
599 if (offset > (p->pack_size - the_hash_algo->rawsz))
600 die("offset beyond end of packfile (truncated pack?)");
602 die(_("offset before end of packfile (broken .idx?)"));
604 if (!win || !in_window(win, offset)) {
607 for (win = p->windows; win; win = win->next) {
608 if (in_window(win, offset))
612 size_t window_align = packed_git_window_size / 2;
615 if (p->pack_fd == -1 && open_packed_git(p))
616 die("packfile %s cannot be accessed", p->pack_name);
618 win = xcalloc(1, sizeof(*win));
619 win->offset = (offset / window_align) * window_align;
620 len = p->pack_size - win->offset;
621 if (len > packed_git_window_size)
622 len = packed_git_window_size;
623 win->len = (size_t)len;
624 pack_mapped += win->len;
625 while (packed_git_limit < pack_mapped
626 && unuse_one_window(p))
628 win->base = xmmap(NULL, win->len,
629 PROT_READ, MAP_PRIVATE,
630 p->pack_fd, win->offset);
631 if (win->base == MAP_FAILED)
632 die_errno("packfile %s cannot be mapped",
634 if (!win->offset && win->len == p->pack_size
639 if (pack_mapped > peak_pack_mapped)
640 peak_pack_mapped = pack_mapped;
641 if (pack_open_windows > peak_pack_open_windows)
642 peak_pack_open_windows = pack_open_windows;
643 win->next = p->windows;
647 if (win != *w_cursor) {
648 win->last_used = pack_used_ctr++;
652 offset -= win->offset;
654 *left = win->len - xsize_t(offset);
655 return win->base + offset;
658 void unuse_pack(struct pack_window **w_cursor)
660 struct pack_window *w = *w_cursor;
667 static void try_to_free_pack_memory(size_t size)
669 release_pack_memory(size);
672 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
674 static int have_set_try_to_free_routine;
677 struct packed_git *p;
679 if (!have_set_try_to_free_routine) {
680 have_set_try_to_free_routine = 1;
681 set_try_to_free_routine(try_to_free_pack_memory);
685 * Make sure a corresponding .pack file exists and that
686 * the index looks sane.
688 if (!strip_suffix_mem(path, &path_len, ".idx"))
692 * ".promisor" is long enough to hold any suffix we're adding (and
693 * the use xsnprintf double-checks that)
695 alloc = st_add3(path_len, strlen(".promisor"), 1);
696 p = alloc_packed_git(alloc);
697 memcpy(p->pack_name, path, path_len);
699 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
700 if (!access(p->pack_name, F_OK))
703 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
704 if (!access(p->pack_name, F_OK))
705 p->pack_promisor = 1;
707 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
708 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
713 /* ok, it looks sane as far as we can check without
714 * actually mapping the pack file.
716 p->pack_size = st.st_size;
717 p->pack_local = local;
718 p->mtime = st.st_mtime;
719 if (path_len < the_hash_algo->hexsz ||
720 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->sha1))
725 void install_packed_git(struct repository *r, struct packed_git *pack)
727 if (pack->pack_fd != -1)
730 pack->next = r->objects->packed_git;
731 r->objects->packed_git = pack;
734 void (*report_garbage)(unsigned seen_bits, const char *path);
736 static void report_helper(const struct string_list *list,
737 int seen_bits, int first, int last)
739 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
742 for (; first < last; first++)
743 report_garbage(seen_bits, list->items[first].string);
746 static void report_pack_garbage(struct string_list *list)
748 int i, baselen = -1, first = 0, seen_bits = 0;
753 string_list_sort(list);
755 for (i = 0; i < list->nr; i++) {
756 const char *path = list->items[i].string;
758 strncmp(path, list->items[first].string, baselen)) {
759 report_helper(list, seen_bits, first, i);
764 const char *dot = strrchr(path, '.');
766 report_garbage(PACKDIR_FILE_GARBAGE, path);
769 baselen = dot - path + 1;
772 if (!strcmp(path + baselen, "pack"))
774 else if (!strcmp(path + baselen, "idx"))
777 report_helper(list, seen_bits, first, list->nr);
780 void for_each_file_in_pack_dir(const char *objdir,
781 each_file_in_pack_dir_fn fn,
784 struct strbuf path = STRBUF_INIT;
789 strbuf_addstr(&path, objdir);
790 strbuf_addstr(&path, "/pack");
791 dir = opendir(path.buf);
794 error_errno("unable to open object pack directory: %s",
796 strbuf_release(&path);
799 strbuf_addch(&path, '/');
800 dirnamelen = path.len;
801 while ((de = readdir(dir)) != NULL) {
802 if (is_dot_or_dotdot(de->d_name))
805 strbuf_setlen(&path, dirnamelen);
806 strbuf_addstr(&path, de->d_name);
808 fn(path.buf, path.len, de->d_name, data);
812 strbuf_release(&path);
815 struct prepare_pack_data {
816 struct repository *r;
817 struct string_list *garbage;
819 struct multi_pack_index *m;
822 static void prepare_pack(const char *full_name, size_t full_name_len,
823 const char *file_name, void *_data)
825 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
826 struct packed_git *p;
827 size_t base_len = full_name_len;
829 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
830 !(data->m && midx_contains_pack(data->m, file_name))) {
831 /* Don't reopen a pack we already have. */
832 for (p = data->r->objects->packed_git; p; p = p->next) {
834 if (strip_suffix(p->pack_name, ".pack", &len) &&
836 !memcmp(p->pack_name, full_name, len))
841 p = add_packed_git(full_name, full_name_len, data->local);
843 install_packed_git(data->r, p);
850 if (!strcmp(file_name, "multi-pack-index"))
852 if (ends_with(file_name, ".idx") ||
853 ends_with(file_name, ".pack") ||
854 ends_with(file_name, ".bitmap") ||
855 ends_with(file_name, ".keep") ||
856 ends_with(file_name, ".promisor"))
857 string_list_append(data->garbage, full_name);
859 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
862 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
864 struct prepare_pack_data data;
865 struct string_list garbage = STRING_LIST_INIT_DUP;
867 data.m = r->objects->multi_pack_index;
869 /* look for the multi-pack-index for this object directory */
870 while (data.m && strcmp(data.m->object_dir, objdir))
871 data.m = data.m->next;
874 data.garbage = &garbage;
877 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
879 report_pack_garbage(data.garbage);
880 string_list_clear(data.garbage, 0);
883 static void prepare_packed_git(struct repository *r);
885 * Give a fast, rough count of the number of objects in the repository. This
886 * ignores loose objects completely. If you have a lot of them, then either
887 * you should repack because your performance will be awful, or they are
888 * all unreachable objects about to be pruned, in which case they're not really
889 * interesting as a measure of repo size in the first place.
891 unsigned long approximate_object_count(void)
893 if (!the_repository->objects->approximate_object_count_valid) {
895 struct multi_pack_index *m;
896 struct packed_git *p;
898 prepare_packed_git(the_repository);
900 for (m = get_multi_pack_index(the_repository); m; m = m->next)
901 count += m->num_objects;
902 for (p = the_repository->objects->packed_git; p; p = p->next) {
903 if (open_pack_index(p))
905 count += p->num_objects;
907 the_repository->objects->approximate_object_count = count;
909 return the_repository->objects->approximate_object_count;
912 static void *get_next_packed_git(const void *p)
914 return ((const struct packed_git *)p)->next;
917 static void set_next_packed_git(void *p, void *next)
919 ((struct packed_git *)p)->next = next;
922 static int sort_pack(const void *a_, const void *b_)
924 const struct packed_git *a = a_;
925 const struct packed_git *b = b_;
929 * Local packs tend to contain objects specific to our
930 * variant of the project than remote ones. In addition,
931 * remote ones could be on a network mounted filesystem.
932 * Favor local ones for these reasons.
934 st = a->pack_local - b->pack_local;
939 * Younger packs tend to contain more recent objects,
940 * and more recent objects tend to get accessed more
943 if (a->mtime < b->mtime)
945 else if (a->mtime == b->mtime)
950 static void rearrange_packed_git(struct repository *r)
952 r->objects->packed_git = llist_mergesort(
953 r->objects->packed_git, get_next_packed_git,
954 set_next_packed_git, sort_pack);
957 static void prepare_packed_git_mru(struct repository *r)
959 struct packed_git *p;
961 INIT_LIST_HEAD(&r->objects->packed_git_mru);
963 for (p = r->objects->packed_git; p; p = p->next)
964 list_add_tail(&p->mru, &r->objects->packed_git_mru);
967 static void prepare_packed_git(struct repository *r)
969 struct alternate_object_database *alt;
971 if (r->objects->packed_git_initialized)
973 prepare_multi_pack_index_one(r, r->objects->objectdir, 1);
974 prepare_packed_git_one(r, r->objects->objectdir, 1);
976 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
977 prepare_multi_pack_index_one(r, alt->path, 0);
978 prepare_packed_git_one(r, alt->path, 0);
980 rearrange_packed_git(r);
982 r->objects->all_packs = NULL;
984 prepare_packed_git_mru(r);
985 r->objects->packed_git_initialized = 1;
988 void reprepare_packed_git(struct repository *r)
990 r->objects->approximate_object_count_valid = 0;
991 r->objects->packed_git_initialized = 0;
992 prepare_packed_git(r);
995 struct packed_git *get_packed_git(struct repository *r)
997 prepare_packed_git(r);
998 return r->objects->packed_git;
1001 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1003 prepare_packed_git(r);
1004 return r->objects->multi_pack_index;
1007 struct packed_git *get_all_packs(struct repository *r)
1009 prepare_packed_git(r);
1011 if (!r->objects->all_packs) {
1012 struct packed_git *p = r->objects->packed_git;
1013 struct multi_pack_index *m;
1015 for (m = r->objects->multi_pack_index; m; m = m->next) {
1017 for (i = 0; i < m->num_packs; i++) {
1018 if (!prepare_midx_pack(m, i)) {
1019 m->packs[i]->next = p;
1025 r->objects->all_packs = p;
1028 return r->objects->all_packs;
1031 struct list_head *get_packed_git_mru(struct repository *r)
1033 prepare_packed_git(r);
1034 return &r->objects->packed_git_mru;
1037 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1038 unsigned long len, enum object_type *type, unsigned long *sizep)
1041 unsigned long size, c;
1042 unsigned long used = 0;
1045 *type = (c >> 4) & 7;
1049 if (len <= used || bitsizeof(long) <= shift) {
1050 error("bad object header");
1055 size += (c & 0x7f) << shift;
1062 unsigned long get_size_from_delta(struct packed_git *p,
1063 struct pack_window **w_curs,
1066 const unsigned char *data;
1067 unsigned char delta_head[20], *in;
1071 memset(&stream, 0, sizeof(stream));
1072 stream.next_out = delta_head;
1073 stream.avail_out = sizeof(delta_head);
1075 git_inflate_init(&stream);
1077 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1078 stream.next_in = in;
1079 st = git_inflate(&stream, Z_FINISH);
1080 curpos += stream.next_in - in;
1081 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1082 stream.total_out < sizeof(delta_head));
1083 git_inflate_end(&stream);
1084 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1085 error("delta data unpack-initial failed");
1089 /* Examine the initial part of the delta to figure out
1094 /* ignore base size */
1095 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1097 /* Read the result size */
1098 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1101 int unpack_object_header(struct packed_git *p,
1102 struct pack_window **w_curs,
1104 unsigned long *sizep)
1106 unsigned char *base;
1109 enum object_type type;
1111 /* use_pack() assures us we have [base, base + 20) available
1112 * as a range that we can look at. (Its actually the hash
1113 * size that is assured.) With our object header encoding
1114 * the maximum deflated object size is 2^137, which is just
1115 * insane, so we know won't exceed what we have been given.
1117 base = use_pack(p, w_curs, *curpos, &left);
1118 used = unpack_object_header_buffer(base, left, &type, sizep);
1127 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1130 const unsigned hashsz = the_hash_algo->rawsz;
1131 for (i = 0; i < p->num_bad_objects; i++)
1132 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1134 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1135 st_mult(GIT_MAX_RAWSZ,
1136 st_add(p->num_bad_objects, 1)));
1137 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1138 p->num_bad_objects++;
1141 const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1143 struct packed_git *p;
1146 for (p = the_repository->objects->packed_git; p; p = p->next)
1147 for (i = 0; i < p->num_bad_objects; i++)
1149 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1154 static off_t get_delta_base(struct packed_git *p,
1155 struct pack_window **w_curs,
1157 enum object_type type,
1158 off_t delta_obj_offset)
1160 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1163 /* use_pack() assured us we have [base_info, base_info + 20)
1164 * as a range that we can look at without walking off the
1165 * end of the mapped window. Its actually the hash size
1166 * that is assured. An OFS_DELTA longer than the hash size
1167 * is stupid, as then a REF_DELTA would be smaller to store.
1169 if (type == OBJ_OFS_DELTA) {
1171 unsigned char c = base_info[used++];
1172 base_offset = c & 127;
1175 if (!base_offset || MSB(base_offset, 7))
1176 return 0; /* overflow */
1177 c = base_info[used++];
1178 base_offset = (base_offset << 7) + (c & 127);
1180 base_offset = delta_obj_offset - base_offset;
1181 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1182 return 0; /* out of bound */
1184 } else if (type == OBJ_REF_DELTA) {
1185 /* The base entry _must_ be in the same pack */
1186 base_offset = find_pack_entry_one(base_info, p);
1187 *curpos += the_hash_algo->rawsz;
1189 die("I am totally screwed");
1194 * Like get_delta_base above, but we return the sha1 instead of the pack
1195 * offset. This means it is cheaper for REF deltas (we do not have to do
1196 * the final object lookup), but more expensive for OFS deltas (we
1197 * have to load the revidx to convert the offset back into a sha1).
1199 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1200 struct pack_window **w_curs,
1202 enum object_type type,
1203 off_t delta_obj_offset)
1205 if (type == OBJ_REF_DELTA) {
1206 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1208 } else if (type == OBJ_OFS_DELTA) {
1209 struct revindex_entry *revidx;
1210 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1211 type, delta_obj_offset);
1216 revidx = find_pack_revindex(p, base_offset);
1220 return nth_packed_object_sha1(p, revidx->nr);
1225 static int retry_bad_packed_offset(struct repository *r,
1226 struct packed_git *p,
1230 struct revindex_entry *revidx;
1231 struct object_id oid;
1232 revidx = find_pack_revindex(p, obj_offset);
1235 nth_packed_object_oid(&oid, p, revidx->nr);
1236 mark_bad_packed_object(p, oid.hash);
1237 type = oid_object_info(r, &oid, NULL);
1238 if (type <= OBJ_NONE)
1243 #define POI_STACK_PREALLOC 64
1245 static enum object_type packed_to_object_type(struct repository *r,
1246 struct packed_git *p,
1248 enum object_type type,
1249 struct pack_window **w_curs,
1252 off_t small_poi_stack[POI_STACK_PREALLOC];
1253 off_t *poi_stack = small_poi_stack;
1254 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1256 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1259 /* Push the object we're going to leave behind */
1260 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1261 poi_stack_alloc = alloc_nr(poi_stack_nr);
1262 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1263 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1265 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1267 poi_stack[poi_stack_nr++] = obj_offset;
1268 /* If parsing the base offset fails, just unwind */
1269 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1272 curpos = obj_offset = base_offset;
1273 type = unpack_object_header(p, w_curs, &curpos, &size);
1274 if (type <= OBJ_NONE) {
1275 /* If getting the base itself fails, we first
1276 * retry the base, otherwise unwind */
1277 type = retry_bad_packed_offset(r, p, base_offset);
1278 if (type > OBJ_NONE)
1292 error("unknown object type %i at offset %"PRIuMAX" in %s",
1293 type, (uintmax_t)obj_offset, p->pack_name);
1298 if (poi_stack != small_poi_stack)
1303 while (poi_stack_nr) {
1304 obj_offset = poi_stack[--poi_stack_nr];
1305 type = retry_bad_packed_offset(r, p, obj_offset);
1306 if (type > OBJ_NONE)
1313 static struct hashmap delta_base_cache;
1314 static size_t delta_base_cached;
1316 static LIST_HEAD(delta_base_cache_lru);
1318 struct delta_base_cache_key {
1319 struct packed_git *p;
1323 struct delta_base_cache_entry {
1324 struct hashmap hash;
1325 struct delta_base_cache_key key;
1326 struct list_head lru;
1329 enum object_type type;
1332 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1336 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1337 hash += (hash >> 8) + (hash >> 16);
1341 static struct delta_base_cache_entry *
1342 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1344 struct hashmap_entry entry;
1345 struct delta_base_cache_key key;
1347 if (!delta_base_cache.cmpfn)
1350 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1352 key.base_offset = base_offset;
1353 return hashmap_get(&delta_base_cache, &entry, &key);
1356 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1357 const struct delta_base_cache_key *b)
1359 return a->p == b->p && a->base_offset == b->base_offset;
1362 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1363 const void *va, const void *vb,
1366 const struct delta_base_cache_entry *a = va, *b = vb;
1367 const struct delta_base_cache_key *key = vkey;
1369 return !delta_base_cache_key_eq(&a->key, key);
1371 return !delta_base_cache_key_eq(&a->key, &b->key);
1374 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1376 return !!get_delta_base_cache_entry(p, base_offset);
1380 * Remove the entry from the cache, but do _not_ free the associated
1381 * entry data. The caller takes ownership of the "data" buffer, and
1382 * should copy out any fields it wants before detaching.
1384 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1386 hashmap_remove(&delta_base_cache, ent, &ent->key);
1387 list_del(&ent->lru);
1388 delta_base_cached -= ent->size;
1392 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1393 off_t base_offset, unsigned long *base_size,
1394 enum object_type *type)
1396 struct delta_base_cache_entry *ent;
1398 ent = get_delta_base_cache_entry(p, base_offset);
1400 return unpack_entry(r, p, base_offset, type, base_size);
1405 *base_size = ent->size;
1406 return xmemdupz(ent->data, ent->size);
1409 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1412 detach_delta_base_cache_entry(ent);
1415 void clear_delta_base_cache(void)
1417 struct list_head *lru, *tmp;
1418 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1419 struct delta_base_cache_entry *entry =
1420 list_entry(lru, struct delta_base_cache_entry, lru);
1421 release_delta_base_cache(entry);
1425 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1426 void *base, unsigned long base_size, enum object_type type)
1428 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1429 struct list_head *lru, *tmp;
1431 delta_base_cached += base_size;
1433 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1434 struct delta_base_cache_entry *f =
1435 list_entry(lru, struct delta_base_cache_entry, lru);
1436 if (delta_base_cached <= delta_base_cache_limit)
1438 release_delta_base_cache(f);
1442 ent->key.base_offset = base_offset;
1445 ent->size = base_size;
1446 list_add_tail(&ent->lru, &delta_base_cache_lru);
1448 if (!delta_base_cache.cmpfn)
1449 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1450 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1451 hashmap_add(&delta_base_cache, ent);
1454 int packed_object_info(struct repository *r, struct packed_git *p,
1455 off_t obj_offset, struct object_info *oi)
1457 struct pack_window *w_curs = NULL;
1459 off_t curpos = obj_offset;
1460 enum object_type type;
1463 * We always get the representation type, but only convert it to
1464 * a "real" type later if the caller is interested.
1467 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1472 type = unpack_object_header(p, &w_curs, &curpos, &size);
1475 if (!oi->contentp && oi->sizep) {
1476 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1477 off_t tmp_pos = curpos;
1478 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1484 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1485 if (*oi->sizep == 0) {
1494 if (oi->disk_sizep) {
1495 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1496 *oi->disk_sizep = revidx[1].offset - obj_offset;
1499 if (oi->typep || oi->type_name) {
1500 enum object_type ptot;
1501 ptot = packed_to_object_type(r, p, obj_offset,
1502 type, &w_curs, curpos);
1505 if (oi->type_name) {
1506 const char *tn = type_name(ptot);
1508 strbuf_addstr(oi->type_name, tn);
1516 if (oi->delta_base_sha1) {
1517 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1518 const unsigned char *base;
1520 base = get_delta_base_sha1(p, &w_curs, curpos,
1527 hashcpy(oi->delta_base_sha1, base);
1529 hashclr(oi->delta_base_sha1);
1532 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1536 unuse_pack(&w_curs);
1540 static void *unpack_compressed_entry(struct packed_git *p,
1541 struct pack_window **w_curs,
1547 unsigned char *buffer, *in;
1549 buffer = xmallocz_gently(size);
1552 memset(&stream, 0, sizeof(stream));
1553 stream.next_out = buffer;
1554 stream.avail_out = size + 1;
1556 git_inflate_init(&stream);
1558 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1559 stream.next_in = in;
1560 st = git_inflate(&stream, Z_FINISH);
1561 if (!stream.avail_out)
1562 break; /* the payload is larger than it should be */
1563 curpos += stream.next_in - in;
1564 } while (st == Z_OK || st == Z_BUF_ERROR);
1565 git_inflate_end(&stream);
1566 if ((st != Z_STREAM_END) || stream.total_out != size) {
1571 /* versions of zlib can clobber unconsumed portion of outbuf */
1572 buffer[size] = '\0';
1577 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1579 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1580 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1581 p->pack_name, (uintmax_t)obj_offset);
1584 int do_check_packed_object_crc;
1586 #define UNPACK_ENTRY_STACK_PREALLOC 64
1587 struct unpack_entry_stack_ent {
1593 static void *read_object(struct repository *r,
1594 const struct object_id *oid,
1595 enum object_type *type,
1596 unsigned long *size)
1598 struct object_info oi = OBJECT_INFO_INIT;
1602 oi.contentp = &content;
1604 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1609 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1610 enum object_type *final_type, unsigned long *final_size)
1612 struct pack_window *w_curs = NULL;
1613 off_t curpos = obj_offset;
1616 enum object_type type;
1617 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1618 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1619 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1620 int base_from_cache = 0;
1622 write_pack_access_log(p, obj_offset);
1624 /* PHASE 1: drill down to the innermost base object */
1628 struct delta_base_cache_entry *ent;
1630 ent = get_delta_base_cache_entry(p, curpos);
1635 detach_delta_base_cache_entry(ent);
1636 base_from_cache = 1;
1640 if (do_check_packed_object_crc && p->index_version > 1) {
1641 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1642 off_t len = revidx[1].offset - obj_offset;
1643 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1644 struct object_id oid;
1645 nth_packed_object_oid(&oid, p, revidx->nr);
1646 error("bad packed object CRC for %s",
1648 mark_bad_packed_object(p, oid.hash);
1654 type = unpack_object_header(p, &w_curs, &curpos, &size);
1655 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1658 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1660 error("failed to validate delta base reference "
1661 "at offset %"PRIuMAX" from %s",
1662 (uintmax_t)curpos, p->pack_name);
1663 /* bail to phase 2, in hopes of recovery */
1668 /* push object, proceed to base */
1669 if (delta_stack_nr >= delta_stack_alloc
1670 && delta_stack == small_delta_stack) {
1671 delta_stack_alloc = alloc_nr(delta_stack_nr);
1672 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1673 memcpy(delta_stack, small_delta_stack,
1674 sizeof(*delta_stack)*delta_stack_nr);
1676 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1678 i = delta_stack_nr++;
1679 delta_stack[i].obj_offset = obj_offset;
1680 delta_stack[i].curpos = curpos;
1681 delta_stack[i].size = size;
1683 curpos = obj_offset = base_offset;
1686 /* PHASE 2: handle the base */
1691 BUG("unpack_entry: left loop at a valid delta");
1697 if (!base_from_cache)
1698 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1702 error("unknown object type %i at offset %"PRIuMAX" in %s",
1703 type, (uintmax_t)obj_offset, p->pack_name);
1706 /* PHASE 3: apply deltas in order */
1709 * 'data' holds the base data, or NULL if there was corruption
1711 while (delta_stack_nr) {
1714 void *external_base = NULL;
1715 unsigned long delta_size, base_size = size;
1721 add_delta_base_cache(p, obj_offset, base, base_size, type);
1725 * We're probably in deep shit, but let's try to fetch
1726 * the required base anyway from another pack or loose.
1727 * This is costly but should happen only in the presence
1728 * of a corrupted pack, and is better than failing outright.
1730 struct revindex_entry *revidx;
1731 struct object_id base_oid;
1732 revidx = find_pack_revindex(p, obj_offset);
1734 nth_packed_object_oid(&base_oid, p, revidx->nr);
1735 error("failed to read delta base object %s"
1736 " at offset %"PRIuMAX" from %s",
1737 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1739 mark_bad_packed_object(p, base_oid.hash);
1740 base = read_object(r, &base_oid, &type, &base_size);
1741 external_base = base;
1745 i = --delta_stack_nr;
1746 obj_offset = delta_stack[i].obj_offset;
1747 curpos = delta_stack[i].curpos;
1748 delta_size = delta_stack[i].size;
1753 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1756 error("failed to unpack compressed delta "
1757 "at offset %"PRIuMAX" from %s",
1758 (uintmax_t)curpos, p->pack_name);
1760 free(external_base);
1764 data = patch_delta(base, base_size,
1765 delta_data, delta_size,
1769 * We could not apply the delta; warn the user, but keep going.
1770 * Our failure will be noticed either in the next iteration of
1771 * the loop, or if this is the final delta, in the caller when
1772 * we return NULL. Those code paths will take care of making
1773 * a more explicit warning and retrying with another copy of
1777 error("failed to apply delta");
1780 free(external_base);
1789 unuse_pack(&w_curs);
1791 if (delta_stack != small_delta_stack)
1797 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1799 const unsigned char *index_fanout = p->index_data;
1800 const unsigned char *index_lookup;
1801 const unsigned int hashsz = the_hash_algo->rawsz;
1802 int index_lookup_width;
1805 BUG("bsearch_pack called without a valid pack-index");
1807 index_lookup = index_fanout + 4 * 256;
1808 if (p->index_version == 1) {
1809 index_lookup_width = hashsz + 4;
1812 index_lookup_width = hashsz;
1817 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1818 index_lookup, index_lookup_width, result);
1821 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1824 const unsigned char *index = p->index_data;
1825 const unsigned int hashsz = the_hash_algo->rawsz;
1827 if (open_pack_index(p))
1829 index = p->index_data;
1831 if (n >= p->num_objects)
1834 if (p->index_version == 1) {
1835 return index + (hashsz + 4) * n + 4;
1838 return index + hashsz * n;
1842 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1843 struct packed_git *p,
1846 const unsigned char *hash = nth_packed_object_sha1(p, n);
1849 hashcpy(oid->hash, hash);
1853 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1855 const unsigned char *ptr = vptr;
1856 const unsigned char *start = p->index_data;
1857 const unsigned char *end = start + p->index_size;
1859 die(_("offset before start of pack index for %s (corrupt index?)"),
1861 /* No need to check for underflow; .idx files must be at least 8 bytes */
1863 die(_("offset beyond end of pack index for %s (truncated index?)"),
1867 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1869 const unsigned char *index = p->index_data;
1870 const unsigned int hashsz = the_hash_algo->rawsz;
1872 if (p->index_version == 1) {
1873 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1876 index += 8 + p->num_objects * (hashsz + 4);
1877 off = ntohl(*((uint32_t *)(index + 4 * n)));
1878 if (!(off & 0x80000000))
1880 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1881 check_pack_index_ptr(p, index);
1882 return get_be64(index);
1886 off_t find_pack_entry_one(const unsigned char *sha1,
1887 struct packed_git *p)
1889 const unsigned char *index = p->index_data;
1890 struct object_id oid;
1894 if (open_pack_index(p))
1898 hashcpy(oid.hash, sha1);
1899 if (bsearch_pack(&oid, p, &result))
1900 return nth_packed_object_offset(p, result);
1904 int is_pack_valid(struct packed_git *p)
1906 /* An already open pack is known to be valid. */
1907 if (p->pack_fd != -1)
1910 /* If the pack has one window completely covering the
1911 * file size, the pack is known to be valid even if
1912 * the descriptor is not currently open.
1915 struct pack_window *w = p->windows;
1917 if (!w->offset && w->len == p->pack_size)
1921 /* Force the pack to open to prove its valid. */
1922 return !open_packed_git(p);
1925 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1926 struct packed_git *packs)
1928 struct packed_git *p;
1930 for (p = packs; p; p = p->next) {
1931 if (find_pack_entry_one(sha1, p))
1938 static int fill_pack_entry(const struct object_id *oid,
1939 struct pack_entry *e,
1940 struct packed_git *p)
1944 if (p->num_bad_objects) {
1946 for (i = 0; i < p->num_bad_objects; i++)
1947 if (hasheq(oid->hash,
1948 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1952 offset = find_pack_entry_one(oid->hash, p);
1957 * We are about to tell the caller where they can locate the
1958 * requested object. We better make sure the packfile is
1959 * still here and can be accessed before supplying that
1960 * answer, as it may have been deleted since the index was
1963 if (!is_pack_valid(p))
1970 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1972 struct list_head *pos;
1973 struct multi_pack_index *m;
1975 prepare_packed_git(r);
1976 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1979 for (m = r->objects->multi_pack_index; m; m = m->next) {
1980 if (fill_midx_entry(oid, e, m))
1984 list_for_each(pos, &r->objects->packed_git_mru) {
1985 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1986 if (fill_pack_entry(oid, e, p)) {
1987 list_move(&p->mru, &r->objects->packed_git_mru);
1994 int has_object_pack(const struct object_id *oid)
1996 struct pack_entry e;
1997 return find_pack_entry(the_repository, oid, &e);
2000 int has_pack_index(const unsigned char *sha1)
2003 if (stat(sha1_pack_index_name(sha1), &st))
2008 int for_each_object_in_pack(struct packed_git *p,
2009 each_packed_object_fn cb, void *data,
2010 enum for_each_object_flags flags)
2015 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2016 load_pack_revindex(p);
2018 for (i = 0; i < p->num_objects; i++) {
2020 struct object_id oid;
2022 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2023 pos = p->revindex[i].nr;
2027 if (!nth_packed_object_oid(&oid, p, pos))
2028 return error("unable to get sha1 of object %u in %s",
2031 r = cb(&oid, p, pos, data);
2038 int for_each_packed_object(each_packed_object_fn cb, void *data,
2039 enum for_each_object_flags flags)
2041 struct packed_git *p;
2043 int pack_errors = 0;
2045 prepare_packed_git(the_repository);
2046 for (p = get_all_packs(the_repository); p; p = p->next) {
2047 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2049 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2052 if (open_pack_index(p)) {
2056 r = for_each_object_in_pack(p, cb, data, flags);
2060 return r ? r : pack_errors;
2063 static int add_promisor_object(const struct object_id *oid,
2064 struct packed_git *pack,
2068 struct oidset *set = set_;
2069 struct object *obj = parse_object(the_repository, oid);
2073 oidset_insert(set, oid);
2076 * If this is a tree, commit, or tag, the objects it refers
2077 * to are also promisor objects. (Blobs refer to no objects->)
2079 if (obj->type == OBJ_TREE) {
2080 struct tree *tree = (struct tree *)obj;
2081 struct tree_desc desc;
2082 struct name_entry entry;
2083 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2085 * Error messages are given when packs are
2086 * verified, so do not print any here.
2089 while (tree_entry_gently(&desc, &entry))
2090 oidset_insert(set, entry.oid);
2091 } else if (obj->type == OBJ_COMMIT) {
2092 struct commit *commit = (struct commit *) obj;
2093 struct commit_list *parents = commit->parents;
2095 oidset_insert(set, get_commit_tree_oid(commit));
2096 for (; parents; parents = parents->next)
2097 oidset_insert(set, &parents->item->object.oid);
2098 } else if (obj->type == OBJ_TAG) {
2099 struct tag *tag = (struct tag *) obj;
2100 oidset_insert(set, &tag->tagged->oid);
2105 int is_promisor_object(const struct object_id *oid)
2107 static struct oidset promisor_objects;
2108 static int promisor_objects_prepared;
2110 if (!promisor_objects_prepared) {
2111 if (repository_format_partial_clone) {
2112 for_each_packed_object(add_promisor_object,
2114 FOR_EACH_OBJECT_PROMISOR_ONLY);
2116 promisor_objects_prepared = 1;
2118 return oidset_contains(&promisor_objects, oid);