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"
20 #include "promisor-remote.h"
22 char *odb_pack_name(struct strbuf *buf,
23 const unsigned char *sha1,
27 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
28 sha1_to_hex(sha1), ext);
32 char *sha1_pack_name(const unsigned char *sha1)
34 static struct strbuf buf = STRBUF_INIT;
35 return odb_pack_name(&buf, sha1, "pack");
38 char *sha1_pack_index_name(const unsigned char *sha1)
40 static struct strbuf buf = STRBUF_INIT;
41 return odb_pack_name(&buf, sha1, "idx");
44 static unsigned int pack_used_ctr;
45 static unsigned int pack_mmap_calls;
46 static unsigned int peak_pack_open_windows;
47 static unsigned int pack_open_windows;
48 static unsigned int pack_open_fds;
49 static unsigned int pack_max_fds;
50 static size_t peak_pack_mapped;
51 static size_t pack_mapped;
53 #define SZ_FMT PRIuMAX
54 static inline uintmax_t sz_fmt(size_t s) { return s; }
56 void pack_report(void)
59 "pack_report: getpagesize() = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
61 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
62 sz_fmt(getpagesize()),
63 sz_fmt(packed_git_window_size),
64 sz_fmt(packed_git_limit));
66 "pack_report: pack_used_ctr = %10u\n"
67 "pack_report: pack_mmap_calls = %10u\n"
68 "pack_report: pack_open_windows = %10u / %10u\n"
69 "pack_report: pack_mapped = "
70 "%10" SZ_FMT " / %10" SZ_FMT "\n",
73 pack_open_windows, peak_pack_open_windows,
74 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
78 * Open and mmap the index file at path, perform a couple of
79 * consistency checks, then record its information to p. Return 0 on
82 static int check_packed_git_idx(const char *path, struct packed_git *p)
86 int fd = git_open(path), ret;
88 const unsigned int hashsz = the_hash_algo->rawsz;
96 idx_size = xsize_t(st.st_size);
97 if (idx_size < 4 * 256 + hashsz + hashsz) {
99 return error("index file %s is too small", path);
101 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
104 ret = load_idx(path, hashsz, idx_map, idx_size, p);
107 munmap(idx_map, idx_size);
112 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
113 size_t idx_size, struct packed_git *p)
115 struct pack_idx_header *hdr = idx_map;
116 uint32_t version, nr, i, *index;
118 if (idx_size < 4 * 256 + hashsz + hashsz)
119 return error("index file %s is too small", path);
121 return error("empty data");
123 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
124 version = ntohl(hdr->idx_version);
125 if (version < 2 || version > 2)
126 return error("index file %s is version %"PRIu32
127 " and is not supported by this binary"
128 " (try upgrading GIT to a newer version)",
136 index += 2; /* skip index header */
137 for (i = 0; i < 256; i++) {
138 uint32_t n = ntohl(index[i]);
140 return error("non-monotonic index %s", path);
147 * - 256 index entries 4 bytes each
148 * - 24-byte entries * nr (object ID + 4-byte offset)
149 * - hash of the packfile
152 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
153 return error("wrong index v1 file size in %s", path);
154 } else if (version == 2) {
157 * - 8 bytes of header
158 * - 256 index entries 4 bytes each
159 * - object ID entry * nr
160 * - 4-byte crc entry * nr
161 * - 4-byte offset entry * nr
162 * - hash of the packfile
164 * And after the 4-byte offset table might be a
165 * variable sized table containing 8-byte entries
166 * for offsets larger than 2^31.
168 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
169 unsigned long max_size = min_size;
171 max_size += (nr - 1)*8;
172 if (idx_size < min_size || idx_size > max_size)
173 return error("wrong index v2 file size in %s", path);
174 if (idx_size != min_size &&
176 * make sure we can deal with large pack offsets.
177 * 31-bit signed offset won't be enough, neither
178 * 32-bit unsigned one will be.
180 (sizeof(off_t) <= 4))
181 return error("pack too large for current definition of off_t in %s", path);
184 p->index_version = version;
185 p->index_data = idx_map;
186 p->index_size = idx_size;
191 int open_pack_index(struct packed_git *p)
200 if (!strip_suffix(p->pack_name, ".pack", &len))
201 BUG("pack_name does not end in .pack");
202 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
203 ret = check_packed_git_idx(idx_name, p);
208 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
210 const uint32_t *level1_ofs = p->index_data;
213 if (open_pack_index(p))
215 level1_ofs = p->index_data;
218 if (p->index_version > 1) {
222 return ntohl(level1_ofs[value]);
225 static struct packed_git *alloc_packed_git(int extra)
227 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
228 memset(p, 0, sizeof(*p));
233 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
235 const char *path = sha1_pack_name(sha1);
236 size_t alloc = st_add(strlen(path), 1);
237 struct packed_git *p = alloc_packed_git(alloc);
239 memcpy(p->pack_name, path, alloc); /* includes NUL */
240 hashcpy(p->hash, sha1);
241 if (check_packed_git_idx(idx_path, p)) {
249 static void scan_windows(struct packed_git *p,
250 struct packed_git **lru_p,
251 struct pack_window **lru_w,
252 struct pack_window **lru_l)
254 struct pack_window *w, *w_l;
256 for (w_l = NULL, w = p->windows; w; w = w->next) {
258 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
268 static int unuse_one_window(struct packed_git *current)
270 struct packed_git *p, *lru_p = NULL;
271 struct pack_window *lru_w = NULL, *lru_l = NULL;
274 scan_windows(current, &lru_p, &lru_w, &lru_l);
275 for (p = the_repository->objects->packed_git; p; p = p->next)
276 scan_windows(p, &lru_p, &lru_w, &lru_l);
278 munmap(lru_w->base, lru_w->len);
279 pack_mapped -= lru_w->len;
281 lru_l->next = lru_w->next;
283 lru_p->windows = lru_w->next;
291 void close_pack_windows(struct packed_git *p)
294 struct pack_window *w = p->windows;
297 die("pack '%s' still has open windows to it",
299 munmap(w->base, w->len);
300 pack_mapped -= w->len;
302 p->windows = w->next;
307 int close_pack_fd(struct packed_git *p)
319 void close_pack_index(struct packed_git *p)
322 munmap((void *)p->index_data, p->index_size);
323 p->index_data = NULL;
327 void close_pack(struct packed_git *p)
329 close_pack_windows(p);
334 void close_object_store(struct raw_object_store *o)
336 struct packed_git *p;
338 for (p = o->packed_git; p; p = p->next)
340 BUG("want to close pack marked 'do-not-close'");
344 if (o->multi_pack_index) {
345 close_midx(o->multi_pack_index);
346 o->multi_pack_index = NULL;
349 close_commit_graph(o);
352 void unlink_pack_path(const char *pack_name, int force_delete)
354 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
356 struct strbuf buf = STRBUF_INIT;
359 strbuf_addstr(&buf, pack_name);
360 strip_suffix_mem(buf.buf, &buf.len, ".pack");
364 strbuf_addstr(&buf, ".keep");
365 if (!access(buf.buf, F_OK)) {
366 strbuf_release(&buf);
371 for (i = 0; i < ARRAY_SIZE(exts); i++) {
372 strbuf_setlen(&buf, plen);
373 strbuf_addstr(&buf, exts[i]);
377 strbuf_release(&buf);
381 * The LRU pack is the one with the oldest MRU window, preferring packs
382 * with no used windows, or the oldest mtime if it has no windows allocated.
384 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
386 struct pack_window *w, *this_mru_w;
387 int has_windows_inuse = 0;
390 * Reject this pack if it has windows and the previously selected
391 * one does not. If this pack does not have windows, reject
392 * it if the pack file is newer than the previously selected one.
394 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
397 for (w = this_mru_w = p->windows; w; w = w->next) {
399 * Reject this pack if any of its windows are in use,
400 * but the previously selected pack did not have any
401 * inuse windows. Otherwise, record that this pack
402 * has windows in use.
405 if (*accept_windows_inuse)
406 has_windows_inuse = 1;
411 if (w->last_used > this_mru_w->last_used)
415 * Reject this pack if it has windows that have been
416 * used more recently than the previously selected pack.
417 * If the previously selected pack had windows inuse and
418 * we have not encountered a window in this pack that is
419 * inuse, skip this check since we prefer a pack with no
420 * inuse windows to one that has inuse windows.
422 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
423 this_mru_w->last_used > (*mru_w)->last_used)
432 *accept_windows_inuse = has_windows_inuse;
435 static int close_one_pack(void)
437 struct packed_git *p, *lru_p = NULL;
438 struct pack_window *mru_w = NULL;
439 int accept_windows_inuse = 1;
441 for (p = the_repository->objects->packed_git; p; p = p->next) {
442 if (p->pack_fd == -1)
444 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
448 return close_pack_fd(lru_p);
453 static unsigned int get_max_fd_limit(void)
459 if (!getrlimit(RLIMIT_NOFILE, &lim))
466 long open_max = sysconf(_SC_OPEN_MAX);
470 * Otherwise, we got -1 for one of the two
473 * (1) sysconf() did not understand _SC_OPEN_MAX
474 * and signaled an error with -1; or
475 * (2) sysconf() said there is no limit.
477 * We _could_ clear errno before calling sysconf() to
478 * tell these two cases apart and return a huge number
479 * in the latter case to let the caller cap it to a
480 * value that is not so selfish, but letting the
481 * fallback OPEN_MAX codepath take care of these cases
490 return 1; /* see the caller ;-) */
494 const char *pack_basename(struct packed_git *p)
496 const char *ret = strrchr(p->pack_name, '/');
498 ret = ret + 1; /* skip past slash */
500 ret = p->pack_name; /* we only have a base */
505 * Do not call this directly as this leaks p->pack_fd on error return;
506 * call open_packed_git() instead.
508 static int open_packed_git_1(struct packed_git *p)
511 struct pack_header hdr;
512 unsigned char hash[GIT_MAX_RAWSZ];
513 unsigned char *idx_hash;
516 const unsigned hashsz = the_hash_algo->rawsz;
518 if (!p->index_data) {
519 struct multi_pack_index *m;
520 const char *pack_name = pack_basename(p);
522 for (m = the_repository->objects->multi_pack_index;
524 if (midx_contains_pack(m, pack_name))
528 if (!m && open_pack_index(p))
529 return error("packfile %s index unavailable", p->pack_name);
533 unsigned int max_fds = get_max_fd_limit();
535 /* Save 3 for stdin/stdout/stderr, 22 for work */
537 pack_max_fds = max_fds - 25;
542 while (pack_max_fds <= pack_open_fds && close_one_pack())
545 p->pack_fd = git_open(p->pack_name);
546 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
550 /* If we created the struct before we had the pack we lack size. */
552 if (!S_ISREG(st.st_mode))
553 return error("packfile %s not a regular file", p->pack_name);
554 p->pack_size = st.st_size;
555 } else if (p->pack_size != st.st_size)
556 return error("packfile %s size changed", p->pack_name);
558 /* We leave these file descriptors open with sliding mmap;
559 * there is no point keeping them open across exec(), though.
561 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
563 return error("cannot determine file descriptor flags");
564 fd_flag |= FD_CLOEXEC;
565 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
566 return error("cannot set FD_CLOEXEC");
568 /* Verify we recognize this pack file format. */
569 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
571 return error_errno("error reading from %s", p->pack_name);
572 if (read_result != sizeof(hdr))
573 return error("file %s is far too short to be a packfile", p->pack_name);
574 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
575 return error("file %s is not a GIT packfile", p->pack_name);
576 if (!pack_version_ok(hdr.hdr_version))
577 return error("packfile %s is version %"PRIu32" and not"
578 " supported (try upgrading GIT to a newer version)",
579 p->pack_name, ntohl(hdr.hdr_version));
581 /* Skip index checking if in multi-pack-index */
585 /* Verify the pack matches its index. */
586 if (p->num_objects != ntohl(hdr.hdr_entries))
587 return error("packfile %s claims to have %"PRIu32" objects"
588 " while index indicates %"PRIu32" objects",
589 p->pack_name, ntohl(hdr.hdr_entries),
591 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
592 return error("end of packfile %s is unavailable", p->pack_name);
593 read_result = read_in_full(p->pack_fd, hash, hashsz);
595 return error_errno("error reading from %s", p->pack_name);
596 if (read_result != hashsz)
597 return error("packfile %s signature is unavailable", p->pack_name);
598 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
599 if (!hasheq(hash, idx_hash))
600 return error("packfile %s does not match index", p->pack_name);
604 static int open_packed_git(struct packed_git *p)
606 if (!open_packed_git_1(p))
612 static int in_window(struct pack_window *win, off_t offset)
614 /* We must promise at least one full hash after the
615 * offset is available from this window, otherwise the offset
616 * is not actually in this window and a different window (which
617 * has that one hash excess) must be used. This is to support
618 * the object header and delta base parsing routines below.
620 off_t win_off = win->offset;
621 return win_off <= offset
622 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
625 unsigned char *use_pack(struct packed_git *p,
626 struct pack_window **w_cursor,
630 struct pack_window *win = *w_cursor;
632 /* Since packfiles end in a hash of their content and it's
633 * pointless to ask for an offset into the middle of that
634 * hash, and the in_window function above wouldn't match
635 * don't allow an offset too close to the end of the file.
637 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
638 die("packfile %s cannot be accessed", p->pack_name);
639 if (offset > (p->pack_size - the_hash_algo->rawsz))
640 die("offset beyond end of packfile (truncated pack?)");
642 die(_("offset before end of packfile (broken .idx?)"));
644 if (!win || !in_window(win, offset)) {
647 for (win = p->windows; win; win = win->next) {
648 if (in_window(win, offset))
652 size_t window_align = packed_git_window_size / 2;
655 if (p->pack_fd == -1 && open_packed_git(p))
656 die("packfile %s cannot be accessed", p->pack_name);
658 win = xcalloc(1, sizeof(*win));
659 win->offset = (offset / window_align) * window_align;
660 len = p->pack_size - win->offset;
661 if (len > packed_git_window_size)
662 len = packed_git_window_size;
663 win->len = (size_t)len;
664 pack_mapped += win->len;
665 while (packed_git_limit < pack_mapped
666 && unuse_one_window(p))
668 win->base = xmmap_gently(NULL, win->len,
669 PROT_READ, MAP_PRIVATE,
670 p->pack_fd, win->offset);
671 if (win->base == MAP_FAILED)
672 die_errno("packfile %s cannot be mapped",
674 if (!win->offset && win->len == p->pack_size
679 if (pack_mapped > peak_pack_mapped)
680 peak_pack_mapped = pack_mapped;
681 if (pack_open_windows > peak_pack_open_windows)
682 peak_pack_open_windows = pack_open_windows;
683 win->next = p->windows;
687 if (win != *w_cursor) {
688 win->last_used = pack_used_ctr++;
692 offset -= win->offset;
694 *left = win->len - xsize_t(offset);
695 return win->base + offset;
698 void unuse_pack(struct pack_window **w_cursor)
700 struct pack_window *w = *w_cursor;
707 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
711 struct packed_git *p;
714 * Make sure a corresponding .pack file exists and that
715 * the index looks sane.
717 if (!strip_suffix_mem(path, &path_len, ".idx"))
721 * ".promisor" is long enough to hold any suffix we're adding (and
722 * the use xsnprintf double-checks that)
724 alloc = st_add3(path_len, strlen(".promisor"), 1);
725 p = alloc_packed_git(alloc);
726 memcpy(p->pack_name, path, path_len);
728 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
729 if (!access(p->pack_name, F_OK))
732 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
733 if (!access(p->pack_name, F_OK))
734 p->pack_promisor = 1;
736 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
737 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
742 /* ok, it looks sane as far as we can check without
743 * actually mapping the pack file.
745 p->pack_size = st.st_size;
746 p->pack_local = local;
747 p->mtime = st.st_mtime;
748 if (path_len < the_hash_algo->hexsz ||
749 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
754 void install_packed_git(struct repository *r, struct packed_git *pack)
756 if (pack->pack_fd != -1)
759 pack->next = r->objects->packed_git;
760 r->objects->packed_git = pack;
763 void (*report_garbage)(unsigned seen_bits, const char *path);
765 static void report_helper(const struct string_list *list,
766 int seen_bits, int first, int last)
768 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
771 for (; first < last; first++)
772 report_garbage(seen_bits, list->items[first].string);
775 static void report_pack_garbage(struct string_list *list)
777 int i, baselen = -1, first = 0, seen_bits = 0;
782 string_list_sort(list);
784 for (i = 0; i < list->nr; i++) {
785 const char *path = list->items[i].string;
787 strncmp(path, list->items[first].string, baselen)) {
788 report_helper(list, seen_bits, first, i);
793 const char *dot = strrchr(path, '.');
795 report_garbage(PACKDIR_FILE_GARBAGE, path);
798 baselen = dot - path + 1;
801 if (!strcmp(path + baselen, "pack"))
803 else if (!strcmp(path + baselen, "idx"))
806 report_helper(list, seen_bits, first, list->nr);
809 void for_each_file_in_pack_dir(const char *objdir,
810 each_file_in_pack_dir_fn fn,
813 struct strbuf path = STRBUF_INIT;
818 strbuf_addstr(&path, objdir);
819 strbuf_addstr(&path, "/pack");
820 dir = opendir(path.buf);
823 error_errno("unable to open object pack directory: %s",
825 strbuf_release(&path);
828 strbuf_addch(&path, '/');
829 dirnamelen = path.len;
830 while ((de = readdir(dir)) != NULL) {
831 if (is_dot_or_dotdot(de->d_name))
834 strbuf_setlen(&path, dirnamelen);
835 strbuf_addstr(&path, de->d_name);
837 fn(path.buf, path.len, de->d_name, data);
841 strbuf_release(&path);
844 struct prepare_pack_data {
845 struct repository *r;
846 struct string_list *garbage;
848 struct multi_pack_index *m;
851 static void prepare_pack(const char *full_name, size_t full_name_len,
852 const char *file_name, void *_data)
854 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
855 struct packed_git *p;
856 size_t base_len = full_name_len;
858 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
859 !(data->m && midx_contains_pack(data->m, file_name))) {
860 /* Don't reopen a pack we already have. */
861 for (p = data->r->objects->packed_git; p; p = p->next) {
863 if (strip_suffix(p->pack_name, ".pack", &len) &&
865 !memcmp(p->pack_name, full_name, len))
870 p = add_packed_git(full_name, full_name_len, data->local);
872 install_packed_git(data->r, p);
879 if (!strcmp(file_name, "multi-pack-index"))
881 if (ends_with(file_name, ".idx") ||
882 ends_with(file_name, ".pack") ||
883 ends_with(file_name, ".bitmap") ||
884 ends_with(file_name, ".keep") ||
885 ends_with(file_name, ".promisor"))
886 string_list_append(data->garbage, full_name);
888 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
891 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
893 struct prepare_pack_data data;
894 struct string_list garbage = STRING_LIST_INIT_DUP;
896 data.m = r->objects->multi_pack_index;
898 /* look for the multi-pack-index for this object directory */
899 while (data.m && strcmp(data.m->object_dir, objdir))
900 data.m = data.m->next;
903 data.garbage = &garbage;
906 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
908 report_pack_garbage(data.garbage);
909 string_list_clear(data.garbage, 0);
912 static void prepare_packed_git(struct repository *r);
914 * Give a fast, rough count of the number of objects in the repository. This
915 * ignores loose objects completely. If you have a lot of them, then either
916 * you should repack because your performance will be awful, or they are
917 * all unreachable objects about to be pruned, in which case they're not really
918 * interesting as a measure of repo size in the first place.
920 unsigned long repo_approximate_object_count(struct repository *r)
922 if (!r->objects->approximate_object_count_valid) {
924 struct multi_pack_index *m;
925 struct packed_git *p;
927 prepare_packed_git(r);
929 for (m = get_multi_pack_index(r); m; m = m->next)
930 count += m->num_objects;
931 for (p = r->objects->packed_git; p; p = p->next) {
932 if (open_pack_index(p))
934 count += p->num_objects;
936 r->objects->approximate_object_count = count;
938 return r->objects->approximate_object_count;
941 static void *get_next_packed_git(const void *p)
943 return ((const struct packed_git *)p)->next;
946 static void set_next_packed_git(void *p, void *next)
948 ((struct packed_git *)p)->next = next;
951 static int sort_pack(const void *a_, const void *b_)
953 const struct packed_git *a = a_;
954 const struct packed_git *b = b_;
958 * Local packs tend to contain objects specific to our
959 * variant of the project than remote ones. In addition,
960 * remote ones could be on a network mounted filesystem.
961 * Favor local ones for these reasons.
963 st = a->pack_local - b->pack_local;
968 * Younger packs tend to contain more recent objects,
969 * and more recent objects tend to get accessed more
972 if (a->mtime < b->mtime)
974 else if (a->mtime == b->mtime)
979 static void rearrange_packed_git(struct repository *r)
981 r->objects->packed_git = llist_mergesort(
982 r->objects->packed_git, get_next_packed_git,
983 set_next_packed_git, sort_pack);
986 static void prepare_packed_git_mru(struct repository *r)
988 struct packed_git *p;
990 INIT_LIST_HEAD(&r->objects->packed_git_mru);
992 for (p = r->objects->packed_git; p; p = p->next)
993 list_add_tail(&p->mru, &r->objects->packed_git_mru);
996 static void prepare_packed_git(struct repository *r)
998 struct object_directory *odb;
1000 if (r->objects->packed_git_initialized)
1004 for (odb = r->objects->odb; odb; odb = odb->next) {
1005 int local = (odb == r->objects->odb);
1006 prepare_multi_pack_index_one(r, odb->path, local);
1007 prepare_packed_git_one(r, odb->path, local);
1009 rearrange_packed_git(r);
1011 prepare_packed_git_mru(r);
1012 r->objects->packed_git_initialized = 1;
1015 void reprepare_packed_git(struct repository *r)
1017 struct object_directory *odb;
1019 for (odb = r->objects->odb; odb; odb = odb->next)
1020 odb_clear_loose_cache(odb);
1022 r->objects->approximate_object_count_valid = 0;
1023 r->objects->packed_git_initialized = 0;
1024 prepare_packed_git(r);
1027 struct packed_git *get_packed_git(struct repository *r)
1029 prepare_packed_git(r);
1030 return r->objects->packed_git;
1033 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1035 prepare_packed_git(r);
1036 return r->objects->multi_pack_index;
1039 struct packed_git *get_all_packs(struct repository *r)
1041 struct multi_pack_index *m;
1043 prepare_packed_git(r);
1044 for (m = r->objects->multi_pack_index; m; m = m->next) {
1046 for (i = 0; i < m->num_packs; i++)
1047 prepare_midx_pack(r, m, i);
1050 return r->objects->packed_git;
1053 struct list_head *get_packed_git_mru(struct repository *r)
1055 prepare_packed_git(r);
1056 return &r->objects->packed_git_mru;
1059 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1060 unsigned long len, enum object_type *type, unsigned long *sizep)
1063 unsigned long size, c;
1064 unsigned long used = 0;
1067 *type = (c >> 4) & 7;
1071 if (len <= used || bitsizeof(long) <= shift) {
1072 error("bad object header");
1077 size += (c & 0x7f) << shift;
1084 unsigned long get_size_from_delta(struct packed_git *p,
1085 struct pack_window **w_curs,
1088 const unsigned char *data;
1089 unsigned char delta_head[20], *in;
1093 memset(&stream, 0, sizeof(stream));
1094 stream.next_out = delta_head;
1095 stream.avail_out = sizeof(delta_head);
1097 git_inflate_init(&stream);
1099 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1100 stream.next_in = in;
1101 st = git_inflate(&stream, Z_FINISH);
1102 curpos += stream.next_in - in;
1103 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1104 stream.total_out < sizeof(delta_head));
1105 git_inflate_end(&stream);
1106 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1107 error("delta data unpack-initial failed");
1111 /* Examine the initial part of the delta to figure out
1116 /* ignore base size */
1117 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1119 /* Read the result size */
1120 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1123 int unpack_object_header(struct packed_git *p,
1124 struct pack_window **w_curs,
1126 unsigned long *sizep)
1128 unsigned char *base;
1131 enum object_type type;
1133 /* use_pack() assures us we have [base, base + 20) available
1134 * as a range that we can look at. (Its actually the hash
1135 * size that is assured.) With our object header encoding
1136 * the maximum deflated object size is 2^137, which is just
1137 * insane, so we know won't exceed what we have been given.
1139 base = use_pack(p, w_curs, *curpos, &left);
1140 used = unpack_object_header_buffer(base, left, &type, sizep);
1149 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1152 const unsigned hashsz = the_hash_algo->rawsz;
1153 for (i = 0; i < p->num_bad_objects; i++)
1154 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1156 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1157 st_mult(GIT_MAX_RAWSZ,
1158 st_add(p->num_bad_objects, 1)));
1159 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1160 p->num_bad_objects++;
1163 const struct packed_git *has_packed_and_bad(struct repository *r,
1164 const unsigned char *sha1)
1166 struct packed_git *p;
1169 for (p = r->objects->packed_git; p; p = p->next)
1170 for (i = 0; i < p->num_bad_objects; i++)
1172 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1177 static off_t get_delta_base(struct packed_git *p,
1178 struct pack_window **w_curs,
1180 enum object_type type,
1181 off_t delta_obj_offset)
1183 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1186 /* use_pack() assured us we have [base_info, base_info + 20)
1187 * as a range that we can look at without walking off the
1188 * end of the mapped window. Its actually the hash size
1189 * that is assured. An OFS_DELTA longer than the hash size
1190 * is stupid, as then a REF_DELTA would be smaller to store.
1192 if (type == OBJ_OFS_DELTA) {
1194 unsigned char c = base_info[used++];
1195 base_offset = c & 127;
1198 if (!base_offset || MSB(base_offset, 7))
1199 return 0; /* overflow */
1200 c = base_info[used++];
1201 base_offset = (base_offset << 7) + (c & 127);
1203 base_offset = delta_obj_offset - base_offset;
1204 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1205 return 0; /* out of bound */
1207 } else if (type == OBJ_REF_DELTA) {
1208 /* The base entry _must_ be in the same pack */
1209 base_offset = find_pack_entry_one(base_info, p);
1210 *curpos += the_hash_algo->rawsz;
1212 die("I am totally screwed");
1217 * Like get_delta_base above, but we return the sha1 instead of the pack
1218 * offset. This means it is cheaper for REF deltas (we do not have to do
1219 * the final object lookup), but more expensive for OFS deltas (we
1220 * have to load the revidx to convert the offset back into a sha1).
1222 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1223 struct pack_window **w_curs,
1225 enum object_type type,
1226 off_t delta_obj_offset)
1228 if (type == OBJ_REF_DELTA) {
1229 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1231 } else if (type == OBJ_OFS_DELTA) {
1232 struct revindex_entry *revidx;
1233 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1234 type, delta_obj_offset);
1239 revidx = find_pack_revindex(p, base_offset);
1243 return nth_packed_object_sha1(p, revidx->nr);
1248 static int retry_bad_packed_offset(struct repository *r,
1249 struct packed_git *p,
1253 struct revindex_entry *revidx;
1254 struct object_id oid;
1255 revidx = find_pack_revindex(p, obj_offset);
1258 nth_packed_object_oid(&oid, p, revidx->nr);
1259 mark_bad_packed_object(p, oid.hash);
1260 type = oid_object_info(r, &oid, NULL);
1261 if (type <= OBJ_NONE)
1266 #define POI_STACK_PREALLOC 64
1268 static enum object_type packed_to_object_type(struct repository *r,
1269 struct packed_git *p,
1271 enum object_type type,
1272 struct pack_window **w_curs,
1275 off_t small_poi_stack[POI_STACK_PREALLOC];
1276 off_t *poi_stack = small_poi_stack;
1277 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1279 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1282 /* Push the object we're going to leave behind */
1283 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1284 poi_stack_alloc = alloc_nr(poi_stack_nr);
1285 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1286 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1288 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1290 poi_stack[poi_stack_nr++] = obj_offset;
1291 /* If parsing the base offset fails, just unwind */
1292 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1295 curpos = obj_offset = base_offset;
1296 type = unpack_object_header(p, w_curs, &curpos, &size);
1297 if (type <= OBJ_NONE) {
1298 /* If getting the base itself fails, we first
1299 * retry the base, otherwise unwind */
1300 type = retry_bad_packed_offset(r, p, base_offset);
1301 if (type > OBJ_NONE)
1315 error("unknown object type %i at offset %"PRIuMAX" in %s",
1316 type, (uintmax_t)obj_offset, p->pack_name);
1321 if (poi_stack != small_poi_stack)
1326 while (poi_stack_nr) {
1327 obj_offset = poi_stack[--poi_stack_nr];
1328 type = retry_bad_packed_offset(r, p, obj_offset);
1329 if (type > OBJ_NONE)
1336 static struct hashmap delta_base_cache;
1337 static size_t delta_base_cached;
1339 static LIST_HEAD(delta_base_cache_lru);
1341 struct delta_base_cache_key {
1342 struct packed_git *p;
1346 struct delta_base_cache_entry {
1347 struct hashmap hash;
1348 struct delta_base_cache_key key;
1349 struct list_head lru;
1352 enum object_type type;
1355 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1359 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1360 hash += (hash >> 8) + (hash >> 16);
1364 static struct delta_base_cache_entry *
1365 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1367 struct hashmap_entry entry;
1368 struct delta_base_cache_key key;
1370 if (!delta_base_cache.cmpfn)
1373 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1375 key.base_offset = base_offset;
1376 return hashmap_get(&delta_base_cache, &entry, &key);
1379 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1380 const struct delta_base_cache_key *b)
1382 return a->p == b->p && a->base_offset == b->base_offset;
1385 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1386 const void *va, const void *vb,
1389 const struct delta_base_cache_entry *a = va, *b = vb;
1390 const struct delta_base_cache_key *key = vkey;
1392 return !delta_base_cache_key_eq(&a->key, key);
1394 return !delta_base_cache_key_eq(&a->key, &b->key);
1397 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1399 return !!get_delta_base_cache_entry(p, base_offset);
1403 * Remove the entry from the cache, but do _not_ free the associated
1404 * entry data. The caller takes ownership of the "data" buffer, and
1405 * should copy out any fields it wants before detaching.
1407 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1409 hashmap_remove(&delta_base_cache, ent, &ent->key);
1410 list_del(&ent->lru);
1411 delta_base_cached -= ent->size;
1415 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1416 off_t base_offset, unsigned long *base_size,
1417 enum object_type *type)
1419 struct delta_base_cache_entry *ent;
1421 ent = get_delta_base_cache_entry(p, base_offset);
1423 return unpack_entry(r, p, base_offset, type, base_size);
1428 *base_size = ent->size;
1429 return xmemdupz(ent->data, ent->size);
1432 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1435 detach_delta_base_cache_entry(ent);
1438 void clear_delta_base_cache(void)
1440 struct list_head *lru, *tmp;
1441 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1442 struct delta_base_cache_entry *entry =
1443 list_entry(lru, struct delta_base_cache_entry, lru);
1444 release_delta_base_cache(entry);
1448 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1449 void *base, unsigned long base_size, enum object_type type)
1451 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1452 struct list_head *lru, *tmp;
1454 delta_base_cached += base_size;
1456 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1457 struct delta_base_cache_entry *f =
1458 list_entry(lru, struct delta_base_cache_entry, lru);
1459 if (delta_base_cached <= delta_base_cache_limit)
1461 release_delta_base_cache(f);
1465 ent->key.base_offset = base_offset;
1468 ent->size = base_size;
1469 list_add_tail(&ent->lru, &delta_base_cache_lru);
1471 if (!delta_base_cache.cmpfn)
1472 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1473 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1474 hashmap_add(&delta_base_cache, ent);
1477 int packed_object_info(struct repository *r, struct packed_git *p,
1478 off_t obj_offset, struct object_info *oi)
1480 struct pack_window *w_curs = NULL;
1482 off_t curpos = obj_offset;
1483 enum object_type type;
1486 * We always get the representation type, but only convert it to
1487 * a "real" type later if the caller is interested.
1490 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1495 type = unpack_object_header(p, &w_curs, &curpos, &size);
1498 if (!oi->contentp && oi->sizep) {
1499 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1500 off_t tmp_pos = curpos;
1501 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1507 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1508 if (*oi->sizep == 0) {
1517 if (oi->disk_sizep) {
1518 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1519 *oi->disk_sizep = revidx[1].offset - obj_offset;
1522 if (oi->typep || oi->type_name) {
1523 enum object_type ptot;
1524 ptot = packed_to_object_type(r, p, obj_offset,
1525 type, &w_curs, curpos);
1528 if (oi->type_name) {
1529 const char *tn = type_name(ptot);
1531 strbuf_addstr(oi->type_name, tn);
1539 if (oi->delta_base_sha1) {
1540 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1541 const unsigned char *base;
1543 base = get_delta_base_sha1(p, &w_curs, curpos,
1550 hashcpy(oi->delta_base_sha1, base);
1552 hashclr(oi->delta_base_sha1);
1555 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1559 unuse_pack(&w_curs);
1563 static void *unpack_compressed_entry(struct packed_git *p,
1564 struct pack_window **w_curs,
1570 unsigned char *buffer, *in;
1572 buffer = xmallocz_gently(size);
1575 memset(&stream, 0, sizeof(stream));
1576 stream.next_out = buffer;
1577 stream.avail_out = size + 1;
1579 git_inflate_init(&stream);
1581 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1582 stream.next_in = in;
1583 st = git_inflate(&stream, Z_FINISH);
1584 if (!stream.avail_out)
1585 break; /* the payload is larger than it should be */
1586 curpos += stream.next_in - in;
1587 } while (st == Z_OK || st == Z_BUF_ERROR);
1588 git_inflate_end(&stream);
1589 if ((st != Z_STREAM_END) || stream.total_out != size) {
1594 /* versions of zlib can clobber unconsumed portion of outbuf */
1595 buffer[size] = '\0';
1600 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1602 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1603 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1604 p->pack_name, (uintmax_t)obj_offset);
1607 int do_check_packed_object_crc;
1609 #define UNPACK_ENTRY_STACK_PREALLOC 64
1610 struct unpack_entry_stack_ent {
1616 static void *read_object(struct repository *r,
1617 const struct object_id *oid,
1618 enum object_type *type,
1619 unsigned long *size)
1621 struct object_info oi = OBJECT_INFO_INIT;
1625 oi.contentp = &content;
1627 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1632 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1633 enum object_type *final_type, unsigned long *final_size)
1635 struct pack_window *w_curs = NULL;
1636 off_t curpos = obj_offset;
1639 enum object_type type;
1640 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1641 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1642 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1643 int base_from_cache = 0;
1645 write_pack_access_log(p, obj_offset);
1647 /* PHASE 1: drill down to the innermost base object */
1651 struct delta_base_cache_entry *ent;
1653 ent = get_delta_base_cache_entry(p, curpos);
1658 detach_delta_base_cache_entry(ent);
1659 base_from_cache = 1;
1663 if (do_check_packed_object_crc && p->index_version > 1) {
1664 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1665 off_t len = revidx[1].offset - obj_offset;
1666 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1667 struct object_id oid;
1668 nth_packed_object_oid(&oid, p, revidx->nr);
1669 error("bad packed object CRC for %s",
1671 mark_bad_packed_object(p, oid.hash);
1677 type = unpack_object_header(p, &w_curs, &curpos, &size);
1678 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1681 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1683 error("failed to validate delta base reference "
1684 "at offset %"PRIuMAX" from %s",
1685 (uintmax_t)curpos, p->pack_name);
1686 /* bail to phase 2, in hopes of recovery */
1691 /* push object, proceed to base */
1692 if (delta_stack_nr >= delta_stack_alloc
1693 && delta_stack == small_delta_stack) {
1694 delta_stack_alloc = alloc_nr(delta_stack_nr);
1695 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1696 COPY_ARRAY(delta_stack, small_delta_stack,
1699 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1701 i = delta_stack_nr++;
1702 delta_stack[i].obj_offset = obj_offset;
1703 delta_stack[i].curpos = curpos;
1704 delta_stack[i].size = size;
1706 curpos = obj_offset = base_offset;
1709 /* PHASE 2: handle the base */
1714 BUG("unpack_entry: left loop at a valid delta");
1720 if (!base_from_cache)
1721 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1725 error("unknown object type %i at offset %"PRIuMAX" in %s",
1726 type, (uintmax_t)obj_offset, p->pack_name);
1729 /* PHASE 3: apply deltas in order */
1732 * 'data' holds the base data, or NULL if there was corruption
1734 while (delta_stack_nr) {
1737 void *external_base = NULL;
1738 unsigned long delta_size, base_size = size;
1744 add_delta_base_cache(p, obj_offset, base, base_size, type);
1748 * We're probably in deep shit, but let's try to fetch
1749 * the required base anyway from another pack or loose.
1750 * This is costly but should happen only in the presence
1751 * of a corrupted pack, and is better than failing outright.
1753 struct revindex_entry *revidx;
1754 struct object_id base_oid;
1755 revidx = find_pack_revindex(p, obj_offset);
1757 nth_packed_object_oid(&base_oid, p, revidx->nr);
1758 error("failed to read delta base object %s"
1759 " at offset %"PRIuMAX" from %s",
1760 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1762 mark_bad_packed_object(p, base_oid.hash);
1763 base = read_object(r, &base_oid, &type, &base_size);
1764 external_base = base;
1768 i = --delta_stack_nr;
1769 obj_offset = delta_stack[i].obj_offset;
1770 curpos = delta_stack[i].curpos;
1771 delta_size = delta_stack[i].size;
1776 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1779 error("failed to unpack compressed delta "
1780 "at offset %"PRIuMAX" from %s",
1781 (uintmax_t)curpos, p->pack_name);
1783 free(external_base);
1787 data = patch_delta(base, base_size,
1788 delta_data, delta_size,
1792 * We could not apply the delta; warn the user, but keep going.
1793 * Our failure will be noticed either in the next iteration of
1794 * the loop, or if this is the final delta, in the caller when
1795 * we return NULL. Those code paths will take care of making
1796 * a more explicit warning and retrying with another copy of
1800 error("failed to apply delta");
1803 free(external_base);
1812 unuse_pack(&w_curs);
1814 if (delta_stack != small_delta_stack)
1820 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1822 const unsigned char *index_fanout = p->index_data;
1823 const unsigned char *index_lookup;
1824 const unsigned int hashsz = the_hash_algo->rawsz;
1825 int index_lookup_width;
1828 BUG("bsearch_pack called without a valid pack-index");
1830 index_lookup = index_fanout + 4 * 256;
1831 if (p->index_version == 1) {
1832 index_lookup_width = hashsz + 4;
1835 index_lookup_width = hashsz;
1840 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1841 index_lookup, index_lookup_width, result);
1844 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1847 const unsigned char *index = p->index_data;
1848 const unsigned int hashsz = the_hash_algo->rawsz;
1850 if (open_pack_index(p))
1852 index = p->index_data;
1854 if (n >= p->num_objects)
1857 if (p->index_version == 1) {
1858 return index + (hashsz + 4) * n + 4;
1861 return index + hashsz * n;
1865 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1866 struct packed_git *p,
1869 const unsigned char *hash = nth_packed_object_sha1(p, n);
1872 hashcpy(oid->hash, hash);
1876 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1878 const unsigned char *ptr = vptr;
1879 const unsigned char *start = p->index_data;
1880 const unsigned char *end = start + p->index_size;
1882 die(_("offset before start of pack index for %s (corrupt index?)"),
1884 /* No need to check for underflow; .idx files must be at least 8 bytes */
1886 die(_("offset beyond end of pack index for %s (truncated index?)"),
1890 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1892 const unsigned char *index = p->index_data;
1893 const unsigned int hashsz = the_hash_algo->rawsz;
1895 if (p->index_version == 1) {
1896 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1899 index += 8 + p->num_objects * (hashsz + 4);
1900 off = ntohl(*((uint32_t *)(index + 4 * n)));
1901 if (!(off & 0x80000000))
1903 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1904 check_pack_index_ptr(p, index);
1905 return get_be64(index);
1909 off_t find_pack_entry_one(const unsigned char *sha1,
1910 struct packed_git *p)
1912 const unsigned char *index = p->index_data;
1913 struct object_id oid;
1917 if (open_pack_index(p))
1921 hashcpy(oid.hash, sha1);
1922 if (bsearch_pack(&oid, p, &result))
1923 return nth_packed_object_offset(p, result);
1927 int is_pack_valid(struct packed_git *p)
1929 /* An already open pack is known to be valid. */
1930 if (p->pack_fd != -1)
1933 /* If the pack has one window completely covering the
1934 * file size, the pack is known to be valid even if
1935 * the descriptor is not currently open.
1938 struct pack_window *w = p->windows;
1940 if (!w->offset && w->len == p->pack_size)
1944 /* Force the pack to open to prove its valid. */
1945 return !open_packed_git(p);
1948 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1949 struct packed_git *packs)
1951 struct packed_git *p;
1953 for (p = packs; p; p = p->next) {
1954 if (find_pack_entry_one(sha1, p))
1961 static int fill_pack_entry(const struct object_id *oid,
1962 struct pack_entry *e,
1963 struct packed_git *p)
1967 if (p->num_bad_objects) {
1969 for (i = 0; i < p->num_bad_objects; i++)
1970 if (hasheq(oid->hash,
1971 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1975 offset = find_pack_entry_one(oid->hash, p);
1980 * We are about to tell the caller where they can locate the
1981 * requested object. We better make sure the packfile is
1982 * still here and can be accessed before supplying that
1983 * answer, as it may have been deleted since the index was
1986 if (!is_pack_valid(p))
1993 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1995 struct list_head *pos;
1996 struct multi_pack_index *m;
1998 prepare_packed_git(r);
1999 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2002 for (m = r->objects->multi_pack_index; m; m = m->next) {
2003 if (fill_midx_entry(r, oid, e, m))
2007 list_for_each(pos, &r->objects->packed_git_mru) {
2008 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2009 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2010 list_move(&p->mru, &r->objects->packed_git_mru);
2017 int has_object_pack(const struct object_id *oid)
2019 struct pack_entry e;
2020 return find_pack_entry(the_repository, oid, &e);
2023 int has_pack_index(const unsigned char *sha1)
2026 if (stat(sha1_pack_index_name(sha1), &st))
2031 int for_each_object_in_pack(struct packed_git *p,
2032 each_packed_object_fn cb, void *data,
2033 enum for_each_object_flags flags)
2038 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2039 if (load_pack_revindex(p))
2043 for (i = 0; i < p->num_objects; i++) {
2045 struct object_id oid;
2047 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2048 pos = p->revindex[i].nr;
2052 if (!nth_packed_object_oid(&oid, p, pos))
2053 return error("unable to get sha1 of object %u in %s",
2056 r = cb(&oid, p, pos, data);
2063 int for_each_packed_object(each_packed_object_fn cb, void *data,
2064 enum for_each_object_flags flags)
2066 struct packed_git *p;
2068 int pack_errors = 0;
2070 prepare_packed_git(the_repository);
2071 for (p = get_all_packs(the_repository); p; p = p->next) {
2072 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2074 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2077 if (open_pack_index(p)) {
2081 r = for_each_object_in_pack(p, cb, data, flags);
2085 return r ? r : pack_errors;
2088 static int add_promisor_object(const struct object_id *oid,
2089 struct packed_git *pack,
2093 struct oidset *set = set_;
2094 struct object *obj = parse_object(the_repository, oid);
2098 oidset_insert(set, oid);
2101 * If this is a tree, commit, or tag, the objects it refers
2102 * to are also promisor objects. (Blobs refer to no objects->)
2104 if (obj->type == OBJ_TREE) {
2105 struct tree *tree = (struct tree *)obj;
2106 struct tree_desc desc;
2107 struct name_entry entry;
2108 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2110 * Error messages are given when packs are
2111 * verified, so do not print any here.
2114 while (tree_entry_gently(&desc, &entry))
2115 oidset_insert(set, &entry.oid);
2116 } else if (obj->type == OBJ_COMMIT) {
2117 struct commit *commit = (struct commit *) obj;
2118 struct commit_list *parents = commit->parents;
2120 oidset_insert(set, get_commit_tree_oid(commit));
2121 for (; parents; parents = parents->next)
2122 oidset_insert(set, &parents->item->object.oid);
2123 } else if (obj->type == OBJ_TAG) {
2124 struct tag *tag = (struct tag *) obj;
2125 oidset_insert(set, get_tagged_oid(tag));
2130 int is_promisor_object(const struct object_id *oid)
2132 static struct oidset promisor_objects;
2133 static int promisor_objects_prepared;
2135 if (!promisor_objects_prepared) {
2136 if (has_promisor_remote()) {
2137 for_each_packed_object(add_promisor_object,
2139 FOR_EACH_OBJECT_PROMISOR_ONLY);
2141 promisor_objects_prepared = 1;
2143 return oidset_contains(&promisor_objects, oid);