4 #include "repository.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
15 #include "tree-walk.h"
17 #include "object-store.h"
19 #include "commit-graph.h"
21 char *odb_pack_name(struct strbuf *buf,
22 const unsigned char *sha1,
26 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27 sha1_to_hex(sha1), ext);
31 char *sha1_pack_name(const unsigned char *sha1)
33 static struct strbuf buf = STRBUF_INIT;
34 return odb_pack_name(&buf, sha1, "pack");
37 char *sha1_pack_index_name(const unsigned char *sha1)
39 static struct strbuf buf = STRBUF_INIT;
40 return odb_pack_name(&buf, sha1, "idx");
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
55 void pack_report(void)
58 "pack_report: getpagesize() = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size),
63 sz_fmt(packed_git_limit));
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT " / %10" SZ_FMT "\n",
72 pack_open_windows, peak_pack_open_windows,
73 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
85 int fd = git_open(path), ret;
87 const unsigned int hashsz = the_hash_algo->rawsz;
95 idx_size = xsize_t(st.st_size);
96 if (idx_size < 4 * 256 + hashsz + hashsz) {
98 return error("index file %s is too small", path);
100 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
106 munmap(idx_map, idx_size);
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112 size_t idx_size, struct packed_git *p)
114 struct pack_idx_header *hdr = idx_map;
115 uint32_t version, nr, i, *index;
117 if (idx_size < 4 * 256 + hashsz + hashsz)
118 return error("index file %s is too small", path);
120 return error("empty data");
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123 version = ntohl(hdr->idx_version);
124 if (version < 2 || version > 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
135 index += 2; /* skip index header */
136 for (i = 0; i < 256; i++) {
137 uint32_t n = ntohl(index[i]);
139 return error("non-monotonic index %s", path);
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
151 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
152 return error("wrong index v1 file size in %s", path);
153 } else if (version == 2) {
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
167 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
168 unsigned long max_size = min_size;
170 max_size += (nr - 1)*8;
171 if (idx_size < min_size || idx_size > max_size)
172 return error("wrong index v2 file size in %s", path);
173 if (idx_size != min_size &&
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
179 (sizeof(off_t) <= 4))
180 return error("pack too large for current definition of off_t in %s", path);
183 p->index_version = version;
184 p->index_data = idx_map;
185 p->index_size = idx_size;
190 int open_pack_index(struct packed_git *p)
199 if (!strip_suffix(p->pack_name, ".pack", &len))
200 BUG("pack_name does not end in .pack");
201 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
202 ret = check_packed_git_idx(idx_name, p);
207 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
209 const uint32_t *level1_ofs = p->index_data;
212 if (open_pack_index(p))
214 level1_ofs = p->index_data;
217 if (p->index_version > 1) {
221 return ntohl(level1_ofs[value]);
224 static struct packed_git *alloc_packed_git(int extra)
226 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
227 memset(p, 0, sizeof(*p));
232 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
234 const char *path = sha1_pack_name(sha1);
235 size_t alloc = st_add(strlen(path), 1);
236 struct packed_git *p = alloc_packed_git(alloc);
238 memcpy(p->pack_name, path, alloc); /* includes NUL */
239 hashcpy(p->hash, sha1);
240 if (check_packed_git_idx(idx_path, p)) {
248 static void scan_windows(struct packed_git *p,
249 struct packed_git **lru_p,
250 struct pack_window **lru_w,
251 struct pack_window **lru_l)
253 struct pack_window *w, *w_l;
255 for (w_l = NULL, w = p->windows; w; w = w->next) {
257 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
267 static int unuse_one_window(struct packed_git *current)
269 struct packed_git *p, *lru_p = NULL;
270 struct pack_window *lru_w = NULL, *lru_l = NULL;
273 scan_windows(current, &lru_p, &lru_w, &lru_l);
274 for (p = the_repository->objects->packed_git; p; p = p->next)
275 scan_windows(p, &lru_p, &lru_w, &lru_l);
277 munmap(lru_w->base, lru_w->len);
278 pack_mapped -= lru_w->len;
280 lru_l->next = lru_w->next;
282 lru_p->windows = lru_w->next;
290 void release_pack_memory(size_t need)
292 size_t cur = pack_mapped;
293 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
297 void close_pack_windows(struct packed_git *p)
300 struct pack_window *w = p->windows;
303 die("pack '%s' still has open windows to it",
305 munmap(w->base, w->len);
306 pack_mapped -= w->len;
308 p->windows = w->next;
313 int close_pack_fd(struct packed_git *p)
325 void close_pack_index(struct packed_git *p)
328 munmap((void *)p->index_data, p->index_size);
329 p->index_data = NULL;
333 void close_pack(struct packed_git *p)
335 close_pack_windows(p);
340 void close_object_store(struct raw_object_store *o)
342 struct packed_git *p;
344 for (p = o->packed_git; p; p = p->next)
346 BUG("want to close pack marked 'do-not-close'");
350 if (o->multi_pack_index) {
351 close_midx(o->multi_pack_index);
352 o->multi_pack_index = NULL;
355 close_commit_graph(o);
358 void unlink_pack_path(const char *pack_name, int force_delete)
360 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
362 struct strbuf buf = STRBUF_INIT;
365 strbuf_addstr(&buf, pack_name);
366 strip_suffix_mem(buf.buf, &buf.len, ".pack");
370 strbuf_addstr(&buf, ".keep");
371 if (!access(buf.buf, F_OK)) {
372 strbuf_release(&buf);
377 for (i = 0; i < ARRAY_SIZE(exts); i++) {
378 strbuf_setlen(&buf, plen);
379 strbuf_addstr(&buf, exts[i]);
383 strbuf_release(&buf);
387 * The LRU pack is the one with the oldest MRU window, preferring packs
388 * with no used windows, or the oldest mtime if it has no windows allocated.
390 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
392 struct pack_window *w, *this_mru_w;
393 int has_windows_inuse = 0;
396 * Reject this pack if it has windows and the previously selected
397 * one does not. If this pack does not have windows, reject
398 * it if the pack file is newer than the previously selected one.
400 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
403 for (w = this_mru_w = p->windows; w; w = w->next) {
405 * Reject this pack if any of its windows are in use,
406 * but the previously selected pack did not have any
407 * inuse windows. Otherwise, record that this pack
408 * has windows in use.
411 if (*accept_windows_inuse)
412 has_windows_inuse = 1;
417 if (w->last_used > this_mru_w->last_used)
421 * Reject this pack if it has windows that have been
422 * used more recently than the previously selected pack.
423 * If the previously selected pack had windows inuse and
424 * we have not encountered a window in this pack that is
425 * inuse, skip this check since we prefer a pack with no
426 * inuse windows to one that has inuse windows.
428 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
429 this_mru_w->last_used > (*mru_w)->last_used)
438 *accept_windows_inuse = has_windows_inuse;
441 static int close_one_pack(void)
443 struct packed_git *p, *lru_p = NULL;
444 struct pack_window *mru_w = NULL;
445 int accept_windows_inuse = 1;
447 for (p = the_repository->objects->packed_git; p; p = p->next) {
448 if (p->pack_fd == -1)
450 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
454 return close_pack_fd(lru_p);
459 static unsigned int get_max_fd_limit(void)
465 if (!getrlimit(RLIMIT_NOFILE, &lim))
472 long open_max = sysconf(_SC_OPEN_MAX);
476 * Otherwise, we got -1 for one of the two
479 * (1) sysconf() did not understand _SC_OPEN_MAX
480 * and signaled an error with -1; or
481 * (2) sysconf() said there is no limit.
483 * We _could_ clear errno before calling sysconf() to
484 * tell these two cases apart and return a huge number
485 * in the latter case to let the caller cap it to a
486 * value that is not so selfish, but letting the
487 * fallback OPEN_MAX codepath take care of these cases
496 return 1; /* see the caller ;-) */
500 const char *pack_basename(struct packed_git *p)
502 const char *ret = strrchr(p->pack_name, '/');
504 ret = ret + 1; /* skip past slash */
506 ret = p->pack_name; /* we only have a base */
511 * Do not call this directly as this leaks p->pack_fd on error return;
512 * call open_packed_git() instead.
514 static int open_packed_git_1(struct packed_git *p)
517 struct pack_header hdr;
518 unsigned char hash[GIT_MAX_RAWSZ];
519 unsigned char *idx_hash;
522 const unsigned hashsz = the_hash_algo->rawsz;
524 if (!p->index_data) {
525 struct multi_pack_index *m;
526 const char *pack_name = pack_basename(p);
528 for (m = the_repository->objects->multi_pack_index;
530 if (midx_contains_pack(m, pack_name))
534 if (!m && open_pack_index(p))
535 return error("packfile %s index unavailable", p->pack_name);
539 unsigned int max_fds = get_max_fd_limit();
541 /* Save 3 for stdin/stdout/stderr, 22 for work */
543 pack_max_fds = max_fds - 25;
548 while (pack_max_fds <= pack_open_fds && close_one_pack())
551 p->pack_fd = git_open(p->pack_name);
552 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
556 /* If we created the struct before we had the pack we lack size. */
558 if (!S_ISREG(st.st_mode))
559 return error("packfile %s not a regular file", p->pack_name);
560 p->pack_size = st.st_size;
561 } else if (p->pack_size != st.st_size)
562 return error("packfile %s size changed", p->pack_name);
564 /* We leave these file descriptors open with sliding mmap;
565 * there is no point keeping them open across exec(), though.
567 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
569 return error("cannot determine file descriptor flags");
570 fd_flag |= FD_CLOEXEC;
571 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
572 return error("cannot set FD_CLOEXEC");
574 /* Verify we recognize this pack file format. */
575 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
577 return error_errno("error reading from %s", p->pack_name);
578 if (read_result != sizeof(hdr))
579 return error("file %s is far too short to be a packfile", p->pack_name);
580 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
581 return error("file %s is not a GIT packfile", p->pack_name);
582 if (!pack_version_ok(hdr.hdr_version))
583 return error("packfile %s is version %"PRIu32" and not"
584 " supported (try upgrading GIT to a newer version)",
585 p->pack_name, ntohl(hdr.hdr_version));
587 /* Skip index checking if in multi-pack-index */
591 /* Verify the pack matches its index. */
592 if (p->num_objects != ntohl(hdr.hdr_entries))
593 return error("packfile %s claims to have %"PRIu32" objects"
594 " while index indicates %"PRIu32" objects",
595 p->pack_name, ntohl(hdr.hdr_entries),
597 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
598 return error("end of packfile %s is unavailable", p->pack_name);
599 read_result = read_in_full(p->pack_fd, hash, hashsz);
601 return error_errno("error reading from %s", p->pack_name);
602 if (read_result != hashsz)
603 return error("packfile %s signature is unavailable", p->pack_name);
604 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
605 if (!hasheq(hash, idx_hash))
606 return error("packfile %s does not match index", p->pack_name);
610 static int open_packed_git(struct packed_git *p)
612 if (!open_packed_git_1(p))
618 static int in_window(struct pack_window *win, off_t offset)
620 /* We must promise at least one full hash after the
621 * offset is available from this window, otherwise the offset
622 * is not actually in this window and a different window (which
623 * has that one hash excess) must be used. This is to support
624 * the object header and delta base parsing routines below.
626 off_t win_off = win->offset;
627 return win_off <= offset
628 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
631 unsigned char *use_pack(struct packed_git *p,
632 struct pack_window **w_cursor,
636 struct pack_window *win = *w_cursor;
638 /* Since packfiles end in a hash of their content and it's
639 * pointless to ask for an offset into the middle of that
640 * hash, and the in_window function above wouldn't match
641 * don't allow an offset too close to the end of the file.
643 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
644 die("packfile %s cannot be accessed", p->pack_name);
645 if (offset > (p->pack_size - the_hash_algo->rawsz))
646 die("offset beyond end of packfile (truncated pack?)");
648 die(_("offset before end of packfile (broken .idx?)"));
650 if (!win || !in_window(win, offset)) {
653 for (win = p->windows; win; win = win->next) {
654 if (in_window(win, offset))
658 size_t window_align = packed_git_window_size / 2;
661 if (p->pack_fd == -1 && open_packed_git(p))
662 die("packfile %s cannot be accessed", p->pack_name);
664 win = xcalloc(1, sizeof(*win));
665 win->offset = (offset / window_align) * window_align;
666 len = p->pack_size - win->offset;
667 if (len > packed_git_window_size)
668 len = packed_git_window_size;
669 win->len = (size_t)len;
670 pack_mapped += win->len;
671 while (packed_git_limit < pack_mapped
672 && unuse_one_window(p))
674 win->base = xmmap_gently(NULL, win->len,
675 PROT_READ, MAP_PRIVATE,
676 p->pack_fd, win->offset);
677 if (win->base == MAP_FAILED)
678 die_errno("packfile %s cannot be mapped",
680 if (!win->offset && win->len == p->pack_size
685 if (pack_mapped > peak_pack_mapped)
686 peak_pack_mapped = pack_mapped;
687 if (pack_open_windows > peak_pack_open_windows)
688 peak_pack_open_windows = pack_open_windows;
689 win->next = p->windows;
693 if (win != *w_cursor) {
694 win->last_used = pack_used_ctr++;
698 offset -= win->offset;
700 *left = win->len - xsize_t(offset);
701 return win->base + offset;
704 void unuse_pack(struct pack_window **w_cursor)
706 struct pack_window *w = *w_cursor;
713 static void try_to_free_pack_memory(size_t size)
715 release_pack_memory(size);
718 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
720 static int have_set_try_to_free_routine;
723 struct packed_git *p;
725 if (!have_set_try_to_free_routine) {
726 have_set_try_to_free_routine = 1;
727 set_try_to_free_routine(try_to_free_pack_memory);
731 * Make sure a corresponding .pack file exists and that
732 * the index looks sane.
734 if (!strip_suffix_mem(path, &path_len, ".idx"))
738 * ".promisor" is long enough to hold any suffix we're adding (and
739 * the use xsnprintf double-checks that)
741 alloc = st_add3(path_len, strlen(".promisor"), 1);
742 p = alloc_packed_git(alloc);
743 memcpy(p->pack_name, path, path_len);
745 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
746 if (!access(p->pack_name, F_OK))
749 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
750 if (!access(p->pack_name, F_OK))
751 p->pack_promisor = 1;
753 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
754 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
759 /* ok, it looks sane as far as we can check without
760 * actually mapping the pack file.
762 p->pack_size = st.st_size;
763 p->pack_local = local;
764 p->mtime = st.st_mtime;
765 if (path_len < the_hash_algo->hexsz ||
766 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
771 void install_packed_git(struct repository *r, struct packed_git *pack)
773 if (pack->pack_fd != -1)
776 pack->next = r->objects->packed_git;
777 r->objects->packed_git = pack;
780 void (*report_garbage)(unsigned seen_bits, const char *path);
782 static void report_helper(const struct string_list *list,
783 int seen_bits, int first, int last)
785 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
788 for (; first < last; first++)
789 report_garbage(seen_bits, list->items[first].string);
792 static void report_pack_garbage(struct string_list *list)
794 int i, baselen = -1, first = 0, seen_bits = 0;
799 string_list_sort(list);
801 for (i = 0; i < list->nr; i++) {
802 const char *path = list->items[i].string;
804 strncmp(path, list->items[first].string, baselen)) {
805 report_helper(list, seen_bits, first, i);
810 const char *dot = strrchr(path, '.');
812 report_garbage(PACKDIR_FILE_GARBAGE, path);
815 baselen = dot - path + 1;
818 if (!strcmp(path + baselen, "pack"))
820 else if (!strcmp(path + baselen, "idx"))
823 report_helper(list, seen_bits, first, list->nr);
826 void for_each_file_in_pack_dir(const char *objdir,
827 each_file_in_pack_dir_fn fn,
830 struct strbuf path = STRBUF_INIT;
835 strbuf_addstr(&path, objdir);
836 strbuf_addstr(&path, "/pack");
837 dir = opendir(path.buf);
840 error_errno("unable to open object pack directory: %s",
842 strbuf_release(&path);
845 strbuf_addch(&path, '/');
846 dirnamelen = path.len;
847 while ((de = readdir(dir)) != NULL) {
848 if (is_dot_or_dotdot(de->d_name))
851 strbuf_setlen(&path, dirnamelen);
852 strbuf_addstr(&path, de->d_name);
854 fn(path.buf, path.len, de->d_name, data);
858 strbuf_release(&path);
861 struct prepare_pack_data {
862 struct repository *r;
863 struct string_list *garbage;
865 struct multi_pack_index *m;
868 static void prepare_pack(const char *full_name, size_t full_name_len,
869 const char *file_name, void *_data)
871 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
872 struct packed_git *p;
873 size_t base_len = full_name_len;
875 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
876 !(data->m && midx_contains_pack(data->m, file_name))) {
877 /* Don't reopen a pack we already have. */
878 for (p = data->r->objects->packed_git; p; p = p->next) {
880 if (strip_suffix(p->pack_name, ".pack", &len) &&
882 !memcmp(p->pack_name, full_name, len))
887 p = add_packed_git(full_name, full_name_len, data->local);
889 install_packed_git(data->r, p);
896 if (!strcmp(file_name, "multi-pack-index"))
898 if (ends_with(file_name, ".idx") ||
899 ends_with(file_name, ".pack") ||
900 ends_with(file_name, ".bitmap") ||
901 ends_with(file_name, ".keep") ||
902 ends_with(file_name, ".promisor"))
903 string_list_append(data->garbage, full_name);
905 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
908 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
910 struct prepare_pack_data data;
911 struct string_list garbage = STRING_LIST_INIT_DUP;
913 data.m = r->objects->multi_pack_index;
915 /* look for the multi-pack-index for this object directory */
916 while (data.m && strcmp(data.m->object_dir, objdir))
917 data.m = data.m->next;
920 data.garbage = &garbage;
923 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
925 report_pack_garbage(data.garbage);
926 string_list_clear(data.garbage, 0);
929 static void prepare_packed_git(struct repository *r);
931 * Give a fast, rough count of the number of objects in the repository. This
932 * ignores loose objects completely. If you have a lot of them, then either
933 * you should repack because your performance will be awful, or they are
934 * all unreachable objects about to be pruned, in which case they're not really
935 * interesting as a measure of repo size in the first place.
937 unsigned long repo_approximate_object_count(struct repository *r)
939 if (!r->objects->approximate_object_count_valid) {
941 struct multi_pack_index *m;
942 struct packed_git *p;
944 prepare_packed_git(r);
946 for (m = get_multi_pack_index(r); m; m = m->next)
947 count += m->num_objects;
948 for (p = r->objects->packed_git; p; p = p->next) {
949 if (open_pack_index(p))
951 count += p->num_objects;
953 r->objects->approximate_object_count = count;
955 return r->objects->approximate_object_count;
958 static void *get_next_packed_git(const void *p)
960 return ((const struct packed_git *)p)->next;
963 static void set_next_packed_git(void *p, void *next)
965 ((struct packed_git *)p)->next = next;
968 static int sort_pack(const void *a_, const void *b_)
970 const struct packed_git *a = a_;
971 const struct packed_git *b = b_;
975 * Local packs tend to contain objects specific to our
976 * variant of the project than remote ones. In addition,
977 * remote ones could be on a network mounted filesystem.
978 * Favor local ones for these reasons.
980 st = a->pack_local - b->pack_local;
985 * Younger packs tend to contain more recent objects,
986 * and more recent objects tend to get accessed more
989 if (a->mtime < b->mtime)
991 else if (a->mtime == b->mtime)
996 static void rearrange_packed_git(struct repository *r)
998 r->objects->packed_git = llist_mergesort(
999 r->objects->packed_git, get_next_packed_git,
1000 set_next_packed_git, sort_pack);
1003 static void prepare_packed_git_mru(struct repository *r)
1005 struct packed_git *p;
1007 INIT_LIST_HEAD(&r->objects->packed_git_mru);
1009 for (p = r->objects->packed_git; p; p = p->next)
1010 list_add_tail(&p->mru, &r->objects->packed_git_mru);
1013 static void prepare_packed_git(struct repository *r)
1015 struct object_directory *odb;
1017 if (r->objects->packed_git_initialized)
1021 for (odb = r->objects->odb; odb; odb = odb->next) {
1022 int local = (odb == r->objects->odb);
1023 prepare_multi_pack_index_one(r, odb->path, local);
1024 prepare_packed_git_one(r, odb->path, local);
1026 rearrange_packed_git(r);
1028 prepare_packed_git_mru(r);
1029 r->objects->packed_git_initialized = 1;
1032 void reprepare_packed_git(struct repository *r)
1034 struct object_directory *odb;
1036 for (odb = r->objects->odb; odb; odb = odb->next)
1037 odb_clear_loose_cache(odb);
1039 r->objects->approximate_object_count_valid = 0;
1040 r->objects->packed_git_initialized = 0;
1041 prepare_packed_git(r);
1044 struct packed_git *get_packed_git(struct repository *r)
1046 prepare_packed_git(r);
1047 return r->objects->packed_git;
1050 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1052 prepare_packed_git(r);
1053 return r->objects->multi_pack_index;
1056 struct packed_git *get_all_packs(struct repository *r)
1058 struct multi_pack_index *m;
1060 prepare_packed_git(r);
1061 for (m = r->objects->multi_pack_index; m; m = m->next) {
1063 for (i = 0; i < m->num_packs; i++)
1064 prepare_midx_pack(r, m, i);
1067 return r->objects->packed_git;
1070 struct list_head *get_packed_git_mru(struct repository *r)
1072 prepare_packed_git(r);
1073 return &r->objects->packed_git_mru;
1076 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1077 unsigned long len, enum object_type *type, unsigned long *sizep)
1080 unsigned long size, c;
1081 unsigned long used = 0;
1084 *type = (c >> 4) & 7;
1088 if (len <= used || bitsizeof(long) <= shift) {
1089 error("bad object header");
1094 size += (c & 0x7f) << shift;
1101 unsigned long get_size_from_delta(struct packed_git *p,
1102 struct pack_window **w_curs,
1105 const unsigned char *data;
1106 unsigned char delta_head[20], *in;
1110 memset(&stream, 0, sizeof(stream));
1111 stream.next_out = delta_head;
1112 stream.avail_out = sizeof(delta_head);
1114 git_inflate_init(&stream);
1116 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1117 stream.next_in = in;
1118 st = git_inflate(&stream, Z_FINISH);
1119 curpos += stream.next_in - in;
1120 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1121 stream.total_out < sizeof(delta_head));
1122 git_inflate_end(&stream);
1123 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1124 error("delta data unpack-initial failed");
1128 /* Examine the initial part of the delta to figure out
1133 /* ignore base size */
1134 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1136 /* Read the result size */
1137 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1140 int unpack_object_header(struct packed_git *p,
1141 struct pack_window **w_curs,
1143 unsigned long *sizep)
1145 unsigned char *base;
1148 enum object_type type;
1150 /* use_pack() assures us we have [base, base + 20) available
1151 * as a range that we can look at. (Its actually the hash
1152 * size that is assured.) With our object header encoding
1153 * the maximum deflated object size is 2^137, which is just
1154 * insane, so we know won't exceed what we have been given.
1156 base = use_pack(p, w_curs, *curpos, &left);
1157 used = unpack_object_header_buffer(base, left, &type, sizep);
1166 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1169 const unsigned hashsz = the_hash_algo->rawsz;
1170 for (i = 0; i < p->num_bad_objects; i++)
1171 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1173 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1174 st_mult(GIT_MAX_RAWSZ,
1175 st_add(p->num_bad_objects, 1)));
1176 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1177 p->num_bad_objects++;
1180 const struct packed_git *has_packed_and_bad(struct repository *r,
1181 const unsigned char *sha1)
1183 struct packed_git *p;
1186 for (p = r->objects->packed_git; p; p = p->next)
1187 for (i = 0; i < p->num_bad_objects; i++)
1189 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1194 static off_t get_delta_base(struct packed_git *p,
1195 struct pack_window **w_curs,
1197 enum object_type type,
1198 off_t delta_obj_offset)
1200 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1203 /* use_pack() assured us we have [base_info, base_info + 20)
1204 * as a range that we can look at without walking off the
1205 * end of the mapped window. Its actually the hash size
1206 * that is assured. An OFS_DELTA longer than the hash size
1207 * is stupid, as then a REF_DELTA would be smaller to store.
1209 if (type == OBJ_OFS_DELTA) {
1211 unsigned char c = base_info[used++];
1212 base_offset = c & 127;
1215 if (!base_offset || MSB(base_offset, 7))
1216 return 0; /* overflow */
1217 c = base_info[used++];
1218 base_offset = (base_offset << 7) + (c & 127);
1220 base_offset = delta_obj_offset - base_offset;
1221 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1222 return 0; /* out of bound */
1224 } else if (type == OBJ_REF_DELTA) {
1225 /* The base entry _must_ be in the same pack */
1226 base_offset = find_pack_entry_one(base_info, p);
1227 *curpos += the_hash_algo->rawsz;
1229 die("I am totally screwed");
1234 * Like get_delta_base above, but we return the sha1 instead of the pack
1235 * offset. This means it is cheaper for REF deltas (we do not have to do
1236 * the final object lookup), but more expensive for OFS deltas (we
1237 * have to load the revidx to convert the offset back into a sha1).
1239 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1240 struct pack_window **w_curs,
1242 enum object_type type,
1243 off_t delta_obj_offset)
1245 if (type == OBJ_REF_DELTA) {
1246 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1248 } else if (type == OBJ_OFS_DELTA) {
1249 struct revindex_entry *revidx;
1250 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1251 type, delta_obj_offset);
1256 revidx = find_pack_revindex(p, base_offset);
1260 return nth_packed_object_sha1(p, revidx->nr);
1265 static int retry_bad_packed_offset(struct repository *r,
1266 struct packed_git *p,
1270 struct revindex_entry *revidx;
1271 struct object_id oid;
1272 revidx = find_pack_revindex(p, obj_offset);
1275 nth_packed_object_oid(&oid, p, revidx->nr);
1276 mark_bad_packed_object(p, oid.hash);
1277 type = oid_object_info(r, &oid, NULL);
1278 if (type <= OBJ_NONE)
1283 #define POI_STACK_PREALLOC 64
1285 static enum object_type packed_to_object_type(struct repository *r,
1286 struct packed_git *p,
1288 enum object_type type,
1289 struct pack_window **w_curs,
1292 off_t small_poi_stack[POI_STACK_PREALLOC];
1293 off_t *poi_stack = small_poi_stack;
1294 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1296 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1299 /* Push the object we're going to leave behind */
1300 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1301 poi_stack_alloc = alloc_nr(poi_stack_nr);
1302 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1303 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1305 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1307 poi_stack[poi_stack_nr++] = obj_offset;
1308 /* If parsing the base offset fails, just unwind */
1309 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1312 curpos = obj_offset = base_offset;
1313 type = unpack_object_header(p, w_curs, &curpos, &size);
1314 if (type <= OBJ_NONE) {
1315 /* If getting the base itself fails, we first
1316 * retry the base, otherwise unwind */
1317 type = retry_bad_packed_offset(r, p, base_offset);
1318 if (type > OBJ_NONE)
1332 error("unknown object type %i at offset %"PRIuMAX" in %s",
1333 type, (uintmax_t)obj_offset, p->pack_name);
1338 if (poi_stack != small_poi_stack)
1343 while (poi_stack_nr) {
1344 obj_offset = poi_stack[--poi_stack_nr];
1345 type = retry_bad_packed_offset(r, p, obj_offset);
1346 if (type > OBJ_NONE)
1353 static struct hashmap delta_base_cache;
1354 static size_t delta_base_cached;
1356 static LIST_HEAD(delta_base_cache_lru);
1358 struct delta_base_cache_key {
1359 struct packed_git *p;
1363 struct delta_base_cache_entry {
1364 struct hashmap_entry ent;
1365 struct delta_base_cache_key key;
1366 struct list_head lru;
1369 enum object_type type;
1372 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1376 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1377 hash += (hash >> 8) + (hash >> 16);
1381 static struct delta_base_cache_entry *
1382 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1384 struct hashmap_entry entry, *e;
1385 struct delta_base_cache_key key;
1387 if (!delta_base_cache.cmpfn)
1390 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1392 key.base_offset = base_offset;
1393 e = hashmap_get(&delta_base_cache, &entry, &key);
1394 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1397 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1398 const struct delta_base_cache_key *b)
1400 return a->p == b->p && a->base_offset == b->base_offset;
1403 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1404 const struct hashmap_entry *va,
1405 const struct hashmap_entry *vb,
1408 const struct delta_base_cache_entry *a, *b;
1409 const struct delta_base_cache_key *key = vkey;
1411 a = container_of(va, const struct delta_base_cache_entry, ent);
1412 b = container_of(vb, const struct delta_base_cache_entry, ent);
1415 return !delta_base_cache_key_eq(&a->key, key);
1417 return !delta_base_cache_key_eq(&a->key, &b->key);
1420 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1422 return !!get_delta_base_cache_entry(p, base_offset);
1426 * Remove the entry from the cache, but do _not_ free the associated
1427 * entry data. The caller takes ownership of the "data" buffer, and
1428 * should copy out any fields it wants before detaching.
1430 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1432 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1433 list_del(&ent->lru);
1434 delta_base_cached -= ent->size;
1438 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1439 off_t base_offset, unsigned long *base_size,
1440 enum object_type *type)
1442 struct delta_base_cache_entry *ent;
1444 ent = get_delta_base_cache_entry(p, base_offset);
1446 return unpack_entry(r, p, base_offset, type, base_size);
1451 *base_size = ent->size;
1452 return xmemdupz(ent->data, ent->size);
1455 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1458 detach_delta_base_cache_entry(ent);
1461 void clear_delta_base_cache(void)
1463 struct list_head *lru, *tmp;
1464 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1465 struct delta_base_cache_entry *entry =
1466 list_entry(lru, struct delta_base_cache_entry, lru);
1467 release_delta_base_cache(entry);
1471 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1472 void *base, unsigned long base_size, enum object_type type)
1474 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1475 struct list_head *lru, *tmp;
1477 delta_base_cached += base_size;
1479 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1480 struct delta_base_cache_entry *f =
1481 list_entry(lru, struct delta_base_cache_entry, lru);
1482 if (delta_base_cached <= delta_base_cache_limit)
1484 release_delta_base_cache(f);
1488 ent->key.base_offset = base_offset;
1491 ent->size = base_size;
1492 list_add_tail(&ent->lru, &delta_base_cache_lru);
1494 if (!delta_base_cache.cmpfn)
1495 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1496 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1497 hashmap_add(&delta_base_cache, &ent->ent);
1500 int packed_object_info(struct repository *r, struct packed_git *p,
1501 off_t obj_offset, struct object_info *oi)
1503 struct pack_window *w_curs = NULL;
1505 off_t curpos = obj_offset;
1506 enum object_type type;
1509 * We always get the representation type, but only convert it to
1510 * a "real" type later if the caller is interested.
1513 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1518 type = unpack_object_header(p, &w_curs, &curpos, &size);
1521 if (!oi->contentp && oi->sizep) {
1522 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1523 off_t tmp_pos = curpos;
1524 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1530 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1531 if (*oi->sizep == 0) {
1540 if (oi->disk_sizep) {
1541 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1542 *oi->disk_sizep = revidx[1].offset - obj_offset;
1545 if (oi->typep || oi->type_name) {
1546 enum object_type ptot;
1547 ptot = packed_to_object_type(r, p, obj_offset,
1548 type, &w_curs, curpos);
1551 if (oi->type_name) {
1552 const char *tn = type_name(ptot);
1554 strbuf_addstr(oi->type_name, tn);
1562 if (oi->delta_base_sha1) {
1563 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1564 const unsigned char *base;
1566 base = get_delta_base_sha1(p, &w_curs, curpos,
1573 hashcpy(oi->delta_base_sha1, base);
1575 hashclr(oi->delta_base_sha1);
1578 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1582 unuse_pack(&w_curs);
1586 static void *unpack_compressed_entry(struct packed_git *p,
1587 struct pack_window **w_curs,
1593 unsigned char *buffer, *in;
1595 buffer = xmallocz_gently(size);
1598 memset(&stream, 0, sizeof(stream));
1599 stream.next_out = buffer;
1600 stream.avail_out = size + 1;
1602 git_inflate_init(&stream);
1604 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1605 stream.next_in = in;
1606 st = git_inflate(&stream, Z_FINISH);
1607 if (!stream.avail_out)
1608 break; /* the payload is larger than it should be */
1609 curpos += stream.next_in - in;
1610 } while (st == Z_OK || st == Z_BUF_ERROR);
1611 git_inflate_end(&stream);
1612 if ((st != Z_STREAM_END) || stream.total_out != size) {
1617 /* versions of zlib can clobber unconsumed portion of outbuf */
1618 buffer[size] = '\0';
1623 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1625 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1626 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1627 p->pack_name, (uintmax_t)obj_offset);
1630 int do_check_packed_object_crc;
1632 #define UNPACK_ENTRY_STACK_PREALLOC 64
1633 struct unpack_entry_stack_ent {
1639 static void *read_object(struct repository *r,
1640 const struct object_id *oid,
1641 enum object_type *type,
1642 unsigned long *size)
1644 struct object_info oi = OBJECT_INFO_INIT;
1648 oi.contentp = &content;
1650 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1655 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1656 enum object_type *final_type, unsigned long *final_size)
1658 struct pack_window *w_curs = NULL;
1659 off_t curpos = obj_offset;
1662 enum object_type type;
1663 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1664 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1665 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1666 int base_from_cache = 0;
1668 write_pack_access_log(p, obj_offset);
1670 /* PHASE 1: drill down to the innermost base object */
1674 struct delta_base_cache_entry *ent;
1676 ent = get_delta_base_cache_entry(p, curpos);
1681 detach_delta_base_cache_entry(ent);
1682 base_from_cache = 1;
1686 if (do_check_packed_object_crc && p->index_version > 1) {
1687 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1688 off_t len = revidx[1].offset - obj_offset;
1689 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1690 struct object_id oid;
1691 nth_packed_object_oid(&oid, p, revidx->nr);
1692 error("bad packed object CRC for %s",
1694 mark_bad_packed_object(p, oid.hash);
1700 type = unpack_object_header(p, &w_curs, &curpos, &size);
1701 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1704 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1706 error("failed to validate delta base reference "
1707 "at offset %"PRIuMAX" from %s",
1708 (uintmax_t)curpos, p->pack_name);
1709 /* bail to phase 2, in hopes of recovery */
1714 /* push object, proceed to base */
1715 if (delta_stack_nr >= delta_stack_alloc
1716 && delta_stack == small_delta_stack) {
1717 delta_stack_alloc = alloc_nr(delta_stack_nr);
1718 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1719 COPY_ARRAY(delta_stack, small_delta_stack,
1722 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1724 i = delta_stack_nr++;
1725 delta_stack[i].obj_offset = obj_offset;
1726 delta_stack[i].curpos = curpos;
1727 delta_stack[i].size = size;
1729 curpos = obj_offset = base_offset;
1732 /* PHASE 2: handle the base */
1737 BUG("unpack_entry: left loop at a valid delta");
1743 if (!base_from_cache)
1744 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1748 error("unknown object type %i at offset %"PRIuMAX" in %s",
1749 type, (uintmax_t)obj_offset, p->pack_name);
1752 /* PHASE 3: apply deltas in order */
1755 * 'data' holds the base data, or NULL if there was corruption
1757 while (delta_stack_nr) {
1760 void *external_base = NULL;
1761 unsigned long delta_size, base_size = size;
1767 add_delta_base_cache(p, obj_offset, base, base_size, type);
1771 * We're probably in deep shit, but let's try to fetch
1772 * the required base anyway from another pack or loose.
1773 * This is costly but should happen only in the presence
1774 * of a corrupted pack, and is better than failing outright.
1776 struct revindex_entry *revidx;
1777 struct object_id base_oid;
1778 revidx = find_pack_revindex(p, obj_offset);
1780 nth_packed_object_oid(&base_oid, p, revidx->nr);
1781 error("failed to read delta base object %s"
1782 " at offset %"PRIuMAX" from %s",
1783 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1785 mark_bad_packed_object(p, base_oid.hash);
1786 base = read_object(r, &base_oid, &type, &base_size);
1787 external_base = base;
1791 i = --delta_stack_nr;
1792 obj_offset = delta_stack[i].obj_offset;
1793 curpos = delta_stack[i].curpos;
1794 delta_size = delta_stack[i].size;
1799 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1802 error("failed to unpack compressed delta "
1803 "at offset %"PRIuMAX" from %s",
1804 (uintmax_t)curpos, p->pack_name);
1806 free(external_base);
1810 data = patch_delta(base, base_size,
1811 delta_data, delta_size,
1815 * We could not apply the delta; warn the user, but keep going.
1816 * Our failure will be noticed either in the next iteration of
1817 * the loop, or if this is the final delta, in the caller when
1818 * we return NULL. Those code paths will take care of making
1819 * a more explicit warning and retrying with another copy of
1823 error("failed to apply delta");
1826 free(external_base);
1835 unuse_pack(&w_curs);
1837 if (delta_stack != small_delta_stack)
1843 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1845 const unsigned char *index_fanout = p->index_data;
1846 const unsigned char *index_lookup;
1847 const unsigned int hashsz = the_hash_algo->rawsz;
1848 int index_lookup_width;
1851 BUG("bsearch_pack called without a valid pack-index");
1853 index_lookup = index_fanout + 4 * 256;
1854 if (p->index_version == 1) {
1855 index_lookup_width = hashsz + 4;
1858 index_lookup_width = hashsz;
1863 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1864 index_lookup, index_lookup_width, result);
1867 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1870 const unsigned char *index = p->index_data;
1871 const unsigned int hashsz = the_hash_algo->rawsz;
1873 if (open_pack_index(p))
1875 index = p->index_data;
1877 if (n >= p->num_objects)
1880 if (p->index_version == 1) {
1881 return index + (hashsz + 4) * n + 4;
1884 return index + hashsz * n;
1888 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1889 struct packed_git *p,
1892 const unsigned char *hash = nth_packed_object_sha1(p, n);
1895 hashcpy(oid->hash, hash);
1899 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1901 const unsigned char *ptr = vptr;
1902 const unsigned char *start = p->index_data;
1903 const unsigned char *end = start + p->index_size;
1905 die(_("offset before start of pack index for %s (corrupt index?)"),
1907 /* No need to check for underflow; .idx files must be at least 8 bytes */
1909 die(_("offset beyond end of pack index for %s (truncated index?)"),
1913 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1915 const unsigned char *index = p->index_data;
1916 const unsigned int hashsz = the_hash_algo->rawsz;
1918 if (p->index_version == 1) {
1919 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1922 index += 8 + p->num_objects * (hashsz + 4);
1923 off = ntohl(*((uint32_t *)(index + 4 * n)));
1924 if (!(off & 0x80000000))
1926 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1927 check_pack_index_ptr(p, index);
1928 return get_be64(index);
1932 off_t find_pack_entry_one(const unsigned char *sha1,
1933 struct packed_git *p)
1935 const unsigned char *index = p->index_data;
1936 struct object_id oid;
1940 if (open_pack_index(p))
1944 hashcpy(oid.hash, sha1);
1945 if (bsearch_pack(&oid, p, &result))
1946 return nth_packed_object_offset(p, result);
1950 int is_pack_valid(struct packed_git *p)
1952 /* An already open pack is known to be valid. */
1953 if (p->pack_fd != -1)
1956 /* If the pack has one window completely covering the
1957 * file size, the pack is known to be valid even if
1958 * the descriptor is not currently open.
1961 struct pack_window *w = p->windows;
1963 if (!w->offset && w->len == p->pack_size)
1967 /* Force the pack to open to prove its valid. */
1968 return !open_packed_git(p);
1971 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1972 struct packed_git *packs)
1974 struct packed_git *p;
1976 for (p = packs; p; p = p->next) {
1977 if (find_pack_entry_one(sha1, p))
1984 static int fill_pack_entry(const struct object_id *oid,
1985 struct pack_entry *e,
1986 struct packed_git *p)
1990 if (p->num_bad_objects) {
1992 for (i = 0; i < p->num_bad_objects; i++)
1993 if (hasheq(oid->hash,
1994 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1998 offset = find_pack_entry_one(oid->hash, p);
2003 * We are about to tell the caller where they can locate the
2004 * requested object. We better make sure the packfile is
2005 * still here and can be accessed before supplying that
2006 * answer, as it may have been deleted since the index was
2009 if (!is_pack_valid(p))
2016 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2018 struct list_head *pos;
2019 struct multi_pack_index *m;
2021 prepare_packed_git(r);
2022 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2025 for (m = r->objects->multi_pack_index; m; m = m->next) {
2026 if (fill_midx_entry(r, oid, e, m))
2030 list_for_each(pos, &r->objects->packed_git_mru) {
2031 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2032 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2033 list_move(&p->mru, &r->objects->packed_git_mru);
2040 int has_object_pack(const struct object_id *oid)
2042 struct pack_entry e;
2043 return find_pack_entry(the_repository, oid, &e);
2046 int has_pack_index(const unsigned char *sha1)
2049 if (stat(sha1_pack_index_name(sha1), &st))
2054 int for_each_object_in_pack(struct packed_git *p,
2055 each_packed_object_fn cb, void *data,
2056 enum for_each_object_flags flags)
2061 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2062 if (load_pack_revindex(p))
2066 for (i = 0; i < p->num_objects; i++) {
2068 struct object_id oid;
2070 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2071 pos = p->revindex[i].nr;
2075 if (!nth_packed_object_oid(&oid, p, pos))
2076 return error("unable to get sha1 of object %u in %s",
2079 r = cb(&oid, p, pos, data);
2086 int for_each_packed_object(each_packed_object_fn cb, void *data,
2087 enum for_each_object_flags flags)
2089 struct packed_git *p;
2091 int pack_errors = 0;
2093 prepare_packed_git(the_repository);
2094 for (p = get_all_packs(the_repository); p; p = p->next) {
2095 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2097 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2100 if (open_pack_index(p)) {
2104 r = for_each_object_in_pack(p, cb, data, flags);
2108 return r ? r : pack_errors;
2111 static int add_promisor_object(const struct object_id *oid,
2112 struct packed_git *pack,
2116 struct oidset *set = set_;
2117 struct object *obj = parse_object(the_repository, oid);
2121 oidset_insert(set, oid);
2124 * If this is a tree, commit, or tag, the objects it refers
2125 * to are also promisor objects. (Blobs refer to no objects->)
2127 if (obj->type == OBJ_TREE) {
2128 struct tree *tree = (struct tree *)obj;
2129 struct tree_desc desc;
2130 struct name_entry entry;
2131 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2133 * Error messages are given when packs are
2134 * verified, so do not print any here.
2137 while (tree_entry_gently(&desc, &entry))
2138 oidset_insert(set, &entry.oid);
2139 } else if (obj->type == OBJ_COMMIT) {
2140 struct commit *commit = (struct commit *) obj;
2141 struct commit_list *parents = commit->parents;
2143 oidset_insert(set, get_commit_tree_oid(commit));
2144 for (; parents; parents = parents->next)
2145 oidset_insert(set, &parents->item->object.oid);
2146 } else if (obj->type == OBJ_TAG) {
2147 struct tag *tag = (struct tag *) obj;
2148 oidset_insert(set, &tag->tagged->oid);
2153 int is_promisor_object(const struct object_id *oid)
2155 static struct oidset promisor_objects;
2156 static int promisor_objects_prepared;
2158 if (!promisor_objects_prepared) {
2159 if (repository_format_partial_clone) {
2160 for_each_packed_object(add_promisor_object,
2162 FOR_EACH_OBJECT_PROMISOR_ONLY);
2164 promisor_objects_prepared = 1;
2166 return oidset_contains(&promisor_objects, oid);