2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
15 #include "run-command.h"
18 #include "tree-walk.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s) { return s; }
37 const unsigned char null_sha1[20];
39 static const char *no_log_pack_access = "no_log_pack_access";
40 static const char *log_pack_access;
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
48 static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
54 static int cached_object_nr, cached_object_alloc;
56 static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
63 static struct packed_git *last_found_pack;
65 static struct cached_object *find_cached_object(const unsigned char *sha1)
68 struct cached_object *co = cached_objects;
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
74 if (!hashcmp(sha1, empty_tree.sha1))
79 int mkdir_in_gitdir(const char *path)
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
84 struct strbuf sb = STRBUF_INIT;
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
105 return adjust_shared_perm(path);
108 int safe_create_leading_directories(char *path)
110 char *pos = path + offset_1st_component(path);
114 pos = strchr(pos, '/');
117 while (*++pos == '/')
122 if (!stat(path, &st)) {
124 if (!S_ISDIR(st.st_mode)) {
129 else if (mkdir(path, 0777)) {
130 if (errno == EEXIST &&
131 !stat(path, &st) && S_ISDIR(st.st_mode)) {
132 ; /* somebody created it since we checked */
138 else if (adjust_shared_perm(path)) {
147 int safe_create_leading_directories_const(const char *path)
149 /* path points to cache entries, so xstrdup before messing with it */
150 char *buf = xstrdup(path);
151 int result = safe_create_leading_directories(buf);
156 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
159 for (i = 0; i < 20; i++) {
160 static char hex[] = "0123456789abcdef";
161 unsigned int val = sha1[i];
162 char *pos = pathbuf + i*2 + (i > 0);
163 *pos++ = hex[val >> 4];
164 *pos = hex[val & 0xf];
169 * NOTE! This returns a statically allocated buffer, so you have to be
170 * careful about using it. Do an "xstrdup()" if you need to save the
173 * Also note that this returns the location for creating. Reading
174 * SHA1 file can happen from any alternate directory listed in the
175 * DB_ENVIRONMENT environment variable if it is not found in
176 * the primary object database.
178 char *sha1_file_name(const unsigned char *sha1)
180 static char buf[PATH_MAX];
184 objdir = get_object_directory();
185 len = strlen(objdir);
187 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
188 if (len + 43 > PATH_MAX)
189 die("insanely long object directory %s", objdir);
190 memcpy(buf, objdir, len);
194 fill_sha1_path(buf + len + 1, sha1);
198 static char *sha1_get_pack_name(const unsigned char *sha1,
199 char **name, char **base, const char *which)
201 static const char hex[] = "0123456789abcdef";
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 *base = xmalloc(len + 60);
209 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
210 sha1_file_directory, which);
211 *name = *base + len + 11;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
225 char *sha1_pack_name(const unsigned char *sha1)
227 static char *name, *base;
229 return sha1_get_pack_name(sha1, &name, &base, "pack");
232 char *sha1_pack_index_name(const unsigned char *sha1)
234 static char *name, *base;
236 return sha1_get_pack_name(sha1, &name, &base, "idx");
239 struct alternate_object_database *alt_odb_list;
240 static struct alternate_object_database **alt_odb_tail;
242 static int git_open_noatime(const char *name);
245 * Prepare alternate object database registry.
247 * The variable alt_odb_list points at the list of struct
248 * alternate_object_database. The elements on this list come from
249 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
250 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
251 * whose contents is similar to that environment variable but can be
252 * LF separated. Its base points at a statically allocated buffer that
253 * contains "/the/directory/corresponding/to/.git/objects/...", while
254 * its name points just after the slash at the end of ".git/objects/"
255 * in the example above, and has enough space to hold 40-byte hex
256 * SHA1, an extra slash for the first level indirection, and the
259 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
261 const char *objdir = get_object_directory();
262 struct alternate_object_database *ent;
263 struct alternate_object_database *alt;
265 struct strbuf pathbuf = STRBUF_INIT;
267 if (!is_absolute_path(entry) && relative_base) {
268 strbuf_addstr(&pathbuf, real_path(relative_base));
269 strbuf_addch(&pathbuf, '/');
271 strbuf_addstr(&pathbuf, entry);
273 normalize_path_copy(pathbuf.buf, pathbuf.buf);
275 pfxlen = strlen(pathbuf.buf);
278 * The trailing slash after the directory name is given by
279 * this function at the end. Remove duplicates.
281 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
284 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
285 ent = xmalloc(sizeof(*ent) + entlen);
286 memcpy(ent->base, pathbuf.buf, pfxlen);
287 strbuf_release(&pathbuf);
289 ent->name = ent->base + pfxlen + 1;
290 ent->base[pfxlen + 3] = '/';
291 ent->base[pfxlen] = ent->base[entlen-1] = 0;
293 /* Detect cases where alternate disappeared */
294 if (!is_directory(ent->base)) {
295 error("object directory %s does not exist; "
296 "check .git/objects/info/alternates.",
302 /* Prevent the common mistake of listing the same
303 * thing twice, or object directory itself.
305 for (alt = alt_odb_list; alt; alt = alt->next) {
306 if (!memcmp(ent->base, alt->base, pfxlen)) {
311 if (!strcmp(ent->base, objdir)) {
316 /* add the alternate entry */
318 alt_odb_tail = &(ent->next);
321 /* recursively add alternates */
322 read_info_alternates(ent->base, depth + 1);
324 ent->base[pfxlen] = '/';
329 static void link_alt_odb_entries(const char *alt, int len, int sep,
330 const char *relative_base, int depth)
332 struct string_list entries = STRING_LIST_INIT_NODUP;
337 error("%s: ignoring alternate object stores, nesting too deep.",
342 alt_copy = xmemdupz(alt, len);
343 string_list_split_in_place(&entries, alt_copy, sep, -1);
344 for (i = 0; i < entries.nr; i++) {
345 const char *entry = entries.items[i].string;
346 if (entry[0] == '\0' || entry[0] == '#')
348 if (!is_absolute_path(entry) && depth) {
349 error("%s: ignoring relative alternate object store %s",
350 relative_base, entry);
352 link_alt_odb_entry(entry, relative_base, depth);
355 string_list_clear(&entries, 0);
359 void read_info_alternates(const char * relative_base, int depth)
364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
371 sprintf(path, "%s/%s", relative_base, alt_file_name);
372 fd = git_open_noatime(path);
375 if (fstat(fd, &st) || (st.st_size == 0)) {
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
383 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
388 void add_to_alternates_file(const char *reference)
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
392 char *alt = mkpath("%s\n", reference);
393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
397 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
400 void foreach_alt_odb(alt_odb_fn fn, void *cb)
402 struct alternate_object_database *ent;
405 for (ent = alt_odb_list; ent; ent = ent->next)
410 void prepare_alt_odb(void)
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
420 alt_odb_tail = &alt_odb_list;
421 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object_local(const unsigned char *sha1)
428 char *name = sha1_file_name(sha1);
429 return !access(name, F_OK);
432 int has_loose_object_nonlocal(const unsigned char *sha1)
434 struct alternate_object_database *alt;
436 for (alt = alt_odb_list; alt; alt = alt->next) {
437 fill_sha1_path(alt->name, sha1);
438 if (!access(alt->base, F_OK))
444 static int has_loose_object(const unsigned char *sha1)
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
450 static unsigned int pack_used_ctr;
451 static unsigned int pack_mmap_calls;
452 static unsigned int peak_pack_open_windows;
453 static unsigned int pack_open_windows;
454 static unsigned int pack_open_fds;
455 static unsigned int pack_max_fds;
456 static size_t peak_pack_mapped;
457 static size_t pack_mapped;
458 struct packed_git *packed_git;
460 void pack_report(void)
463 "pack_report: getpagesize() = %10" SZ_FMT "\n"
464 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
465 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
466 sz_fmt(getpagesize()),
467 sz_fmt(packed_git_window_size),
468 sz_fmt(packed_git_limit));
470 "pack_report: pack_used_ctr = %10u\n"
471 "pack_report: pack_mmap_calls = %10u\n"
472 "pack_report: pack_open_windows = %10u / %10u\n"
473 "pack_report: pack_mapped = "
474 "%10" SZ_FMT " / %10" SZ_FMT "\n",
477 pack_open_windows, peak_pack_open_windows,
478 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
481 static int check_packed_git_idx(const char *path, struct packed_git *p)
484 struct pack_idx_header *hdr;
486 uint32_t version, nr, i, *index;
487 int fd = git_open_noatime(path);
492 if (fstat(fd, &st)) {
496 idx_size = xsize_t(st.st_size);
497 if (idx_size < 4 * 256 + 20 + 20) {
499 return error("index file %s is too small", path);
501 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
505 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
506 version = ntohl(hdr->idx_version);
507 if (version < 2 || version > 2) {
508 munmap(idx_map, idx_size);
509 return error("index file %s is version %"PRIu32
510 " and is not supported by this binary"
511 " (try upgrading GIT to a newer version)",
520 index += 2; /* skip index header */
521 for (i = 0; i < 256; i++) {
522 uint32_t n = ntohl(index[i]);
524 munmap(idx_map, idx_size);
525 return error("non-monotonic index %s", path);
533 * - 256 index entries 4 bytes each
534 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
535 * - 20-byte SHA1 of the packfile
536 * - 20-byte SHA1 file checksum
538 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
539 munmap(idx_map, idx_size);
540 return error("wrong index v1 file size in %s", path);
542 } else if (version == 2) {
545 * - 8 bytes of header
546 * - 256 index entries 4 bytes each
547 * - 20-byte sha1 entry * nr
548 * - 4-byte crc entry * nr
549 * - 4-byte offset entry * nr
550 * - 20-byte SHA1 of the packfile
551 * - 20-byte SHA1 file checksum
552 * And after the 4-byte offset table might be a
553 * variable sized table containing 8-byte entries
554 * for offsets larger than 2^31.
556 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
557 unsigned long max_size = min_size;
559 max_size += (nr - 1)*8;
560 if (idx_size < min_size || idx_size > max_size) {
561 munmap(idx_map, idx_size);
562 return error("wrong index v2 file size in %s", path);
564 if (idx_size != min_size &&
566 * make sure we can deal with large pack offsets.
567 * 31-bit signed offset won't be enough, neither
568 * 32-bit unsigned one will be.
570 (sizeof(off_t) <= 4)) {
571 munmap(idx_map, idx_size);
572 return error("pack too large for current definition of off_t in %s", path);
576 p->index_version = version;
577 p->index_data = idx_map;
578 p->index_size = idx_size;
583 int open_pack_index(struct packed_git *p)
591 idx_name = xstrdup(p->pack_name);
592 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
593 ret = check_packed_git_idx(idx_name, p);
598 static void scan_windows(struct packed_git *p,
599 struct packed_git **lru_p,
600 struct pack_window **lru_w,
601 struct pack_window **lru_l)
603 struct pack_window *w, *w_l;
605 for (w_l = NULL, w = p->windows; w; w = w->next) {
607 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
617 static int unuse_one_window(struct packed_git *current)
619 struct packed_git *p, *lru_p = NULL;
620 struct pack_window *lru_w = NULL, *lru_l = NULL;
623 scan_windows(current, &lru_p, &lru_w, &lru_l);
624 for (p = packed_git; p; p = p->next)
625 scan_windows(p, &lru_p, &lru_w, &lru_l);
627 munmap(lru_w->base, lru_w->len);
628 pack_mapped -= lru_w->len;
630 lru_l->next = lru_w->next;
632 lru_p->windows = lru_w->next;
640 void release_pack_memory(size_t need)
642 size_t cur = pack_mapped;
643 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
647 void *xmmap(void *start, size_t length,
648 int prot, int flags, int fd, off_t offset)
650 void *ret = mmap(start, length, prot, flags, fd, offset);
651 if (ret == MAP_FAILED) {
654 release_pack_memory(length);
655 ret = mmap(start, length, prot, flags, fd, offset);
656 if (ret == MAP_FAILED)
657 die_errno("Out of memory? mmap failed");
662 void close_pack_windows(struct packed_git *p)
665 struct pack_window *w = p->windows;
668 die("pack '%s' still has open windows to it",
670 munmap(w->base, w->len);
671 pack_mapped -= w->len;
673 p->windows = w->next;
679 * The LRU pack is the one with the oldest MRU window, preferring packs
680 * with no used windows, or the oldest mtime if it has no windows allocated.
682 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
684 struct pack_window *w, *this_mru_w;
685 int has_windows_inuse = 0;
688 * Reject this pack if it has windows and the previously selected
689 * one does not. If this pack does not have windows, reject
690 * it if the pack file is newer than the previously selected one.
692 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
695 for (w = this_mru_w = p->windows; w; w = w->next) {
697 * Reject this pack if any of its windows are in use,
698 * but the previously selected pack did not have any
699 * inuse windows. Otherwise, record that this pack
700 * has windows in use.
703 if (*accept_windows_inuse)
704 has_windows_inuse = 1;
709 if (w->last_used > this_mru_w->last_used)
713 * Reject this pack if it has windows that have been
714 * used more recently than the previously selected pack.
715 * If the previously selected pack had windows inuse and
716 * we have not encountered a window in this pack that is
717 * inuse, skip this check since we prefer a pack with no
718 * inuse windows to one that has inuse windows.
720 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
721 this_mru_w->last_used > (*mru_w)->last_used)
730 *accept_windows_inuse = has_windows_inuse;
733 static int close_one_pack(void)
735 struct packed_git *p, *lru_p = NULL;
736 struct pack_window *mru_w = NULL;
737 int accept_windows_inuse = 1;
739 for (p = packed_git; p; p = p->next) {
740 if (p->pack_fd == -1)
742 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
746 close(lru_p->pack_fd);
755 void unuse_pack(struct pack_window **w_cursor)
757 struct pack_window *w = *w_cursor;
764 void close_pack_index(struct packed_git *p)
767 munmap((void *)p->index_data, p->index_size);
768 p->index_data = NULL;
773 * This is used by git-repack in case a newly created pack happens to
774 * contain the same set of objects as an existing one. In that case
775 * the resulting file might be different even if its name would be the
776 * same. It is best to close any reference to the old pack before it is
777 * replaced on disk. Of course no index pointers nor windows for given pack
778 * must subsist at this point. If ever objects from this pack are requested
779 * again, the new version of the pack will be reinitialized through
780 * reprepare_packed_git().
782 void free_pack_by_name(const char *pack_name)
784 struct packed_git *p, **pp = &packed_git;
788 if (strcmp(pack_name, p->pack_name) == 0) {
789 clear_delta_base_cache();
790 close_pack_windows(p);
791 if (p->pack_fd != -1) {
796 free(p->bad_object_sha1);
798 if (last_found_pack == p)
799 last_found_pack = NULL;
807 static unsigned int get_max_fd_limit(void)
813 if (!getrlimit(RLIMIT_NOFILE, &lim))
820 long open_max = sysconf(_SC_OPEN_MAX);
824 * Otherwise, we got -1 for one of the two
827 * (1) sysconf() did not understand _SC_OPEN_MAX
828 * and signaled an error with -1; or
829 * (2) sysconf() said there is no limit.
831 * We _could_ clear errno before calling sysconf() to
832 * tell these two cases apart and return a huge number
833 * in the latter case to let the caller cap it to a
834 * value that is not so selfish, but letting the
835 * fallback OPEN_MAX codepath take care of these cases
844 return 1; /* see the caller ;-) */
849 * Do not call this directly as this leaks p->pack_fd on error return;
850 * call open_packed_git() instead.
852 static int open_packed_git_1(struct packed_git *p)
855 struct pack_header hdr;
856 unsigned char sha1[20];
857 unsigned char *idx_sha1;
860 if (!p->index_data && open_pack_index(p))
861 return error("packfile %s index unavailable", p->pack_name);
864 unsigned int max_fds = get_max_fd_limit();
866 /* Save 3 for stdin/stdout/stderr, 22 for work */
868 pack_max_fds = max_fds - 25;
873 while (pack_max_fds <= pack_open_fds && close_one_pack())
876 p->pack_fd = git_open_noatime(p->pack_name);
877 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
881 /* If we created the struct before we had the pack we lack size. */
883 if (!S_ISREG(st.st_mode))
884 return error("packfile %s not a regular file", p->pack_name);
885 p->pack_size = st.st_size;
886 } else if (p->pack_size != st.st_size)
887 return error("packfile %s size changed", p->pack_name);
889 /* We leave these file descriptors open with sliding mmap;
890 * there is no point keeping them open across exec(), though.
892 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
894 return error("cannot determine file descriptor flags");
895 fd_flag |= FD_CLOEXEC;
896 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
897 return error("cannot set FD_CLOEXEC");
899 /* Verify we recognize this pack file format. */
900 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
901 return error("file %s is far too short to be a packfile", p->pack_name);
902 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
903 return error("file %s is not a GIT packfile", p->pack_name);
904 if (!pack_version_ok(hdr.hdr_version))
905 return error("packfile %s is version %"PRIu32" and not"
906 " supported (try upgrading GIT to a newer version)",
907 p->pack_name, ntohl(hdr.hdr_version));
909 /* Verify the pack matches its index. */
910 if (p->num_objects != ntohl(hdr.hdr_entries))
911 return error("packfile %s claims to have %"PRIu32" objects"
912 " while index indicates %"PRIu32" objects",
913 p->pack_name, ntohl(hdr.hdr_entries),
915 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
916 return error("end of packfile %s is unavailable", p->pack_name);
917 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
918 return error("packfile %s signature is unavailable", p->pack_name);
919 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
920 if (hashcmp(sha1, idx_sha1))
921 return error("packfile %s does not match index", p->pack_name);
925 static int open_packed_git(struct packed_git *p)
927 if (!open_packed_git_1(p))
929 if (p->pack_fd != -1) {
937 static int in_window(struct pack_window *win, off_t offset)
939 /* We must promise at least 20 bytes (one hash) after the
940 * offset is available from this window, otherwise the offset
941 * is not actually in this window and a different window (which
942 * has that one hash excess) must be used. This is to support
943 * the object header and delta base parsing routines below.
945 off_t win_off = win->offset;
946 return win_off <= offset
947 && (offset + 20) <= (win_off + win->len);
950 unsigned char *use_pack(struct packed_git *p,
951 struct pack_window **w_cursor,
955 struct pack_window *win = *w_cursor;
957 /* Since packfiles end in a hash of their content and it's
958 * pointless to ask for an offset into the middle of that
959 * hash, and the in_window function above wouldn't match
960 * don't allow an offset too close to the end of the file.
962 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
963 die("packfile %s cannot be accessed", p->pack_name);
964 if (offset > (p->pack_size - 20))
965 die("offset beyond end of packfile (truncated pack?)");
967 if (!win || !in_window(win, offset)) {
970 for (win = p->windows; win; win = win->next) {
971 if (in_window(win, offset))
975 size_t window_align = packed_git_window_size / 2;
978 if (p->pack_fd == -1 && open_packed_git(p))
979 die("packfile %s cannot be accessed", p->pack_name);
981 win = xcalloc(1, sizeof(*win));
982 win->offset = (offset / window_align) * window_align;
983 len = p->pack_size - win->offset;
984 if (len > packed_git_window_size)
985 len = packed_git_window_size;
986 win->len = (size_t)len;
987 pack_mapped += win->len;
988 while (packed_git_limit < pack_mapped
989 && unuse_one_window(p))
991 win->base = xmmap(NULL, win->len,
992 PROT_READ, MAP_PRIVATE,
993 p->pack_fd, win->offset);
994 if (win->base == MAP_FAILED)
995 die("packfile %s cannot be mapped: %s",
998 if (!win->offset && win->len == p->pack_size
999 && !p->do_not_close) {
1005 pack_open_windows++;
1006 if (pack_mapped > peak_pack_mapped)
1007 peak_pack_mapped = pack_mapped;
1008 if (pack_open_windows > peak_pack_open_windows)
1009 peak_pack_open_windows = pack_open_windows;
1010 win->next = p->windows;
1014 if (win != *w_cursor) {
1015 win->last_used = pack_used_ctr++;
1019 offset -= win->offset;
1021 *left = win->len - xsize_t(offset);
1022 return win->base + offset;
1025 static struct packed_git *alloc_packed_git(int extra)
1027 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1028 memset(p, 0, sizeof(*p));
1033 static void try_to_free_pack_memory(size_t size)
1035 release_pack_memory(size);
1038 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1040 static int have_set_try_to_free_routine;
1042 struct packed_git *p = alloc_packed_git(path_len + 2);
1044 if (!have_set_try_to_free_routine) {
1045 have_set_try_to_free_routine = 1;
1046 set_try_to_free_routine(try_to_free_pack_memory);
1050 * Make sure a corresponding .pack file exists and that
1051 * the index looks sane.
1053 path_len -= strlen(".idx");
1058 memcpy(p->pack_name, path, path_len);
1060 strcpy(p->pack_name + path_len, ".keep");
1061 if (!access(p->pack_name, F_OK))
1064 strcpy(p->pack_name + path_len, ".pack");
1065 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1070 /* ok, it looks sane as far as we can check without
1071 * actually mapping the pack file.
1073 p->pack_size = st.st_size;
1074 p->pack_local = local;
1075 p->mtime = st.st_mtime;
1076 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1081 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1083 const char *path = sha1_pack_name(sha1);
1084 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1086 strcpy(p->pack_name, path);
1087 hashcpy(p->sha1, sha1);
1088 if (check_packed_git_idx(idx_path, p)) {
1096 void install_packed_git(struct packed_git *pack)
1098 if (pack->pack_fd != -1)
1101 pack->next = packed_git;
1105 void (*report_garbage)(const char *desc, const char *path);
1107 static void report_helper(const struct string_list *list,
1108 int seen_bits, int first, int last)
1111 switch (seen_bits) {
1113 msg = "no corresponding .idx nor .pack";
1116 msg = "no corresponding .idx";
1119 msg = "no corresponding .pack";
1124 for (; first < last; first++)
1125 report_garbage(msg, list->items[first].string);
1128 static void report_pack_garbage(struct string_list *list)
1130 int i, baselen = -1, first = 0, seen_bits = 0;
1132 if (!report_garbage)
1135 sort_string_list(list);
1137 for (i = 0; i < list->nr; i++) {
1138 const char *path = list->items[i].string;
1139 if (baselen != -1 &&
1140 strncmp(path, list->items[first].string, baselen)) {
1141 report_helper(list, seen_bits, first, i);
1145 if (baselen == -1) {
1146 const char *dot = strrchr(path, '.');
1148 report_garbage("garbage found", path);
1151 baselen = dot - path + 1;
1154 if (!strcmp(path + baselen, "pack"))
1156 else if (!strcmp(path + baselen, "idx"))
1159 report_helper(list, seen_bits, first, list->nr);
1162 static void prepare_packed_git_one(char *objdir, int local)
1164 /* Ensure that this buffer is large enough so that we can
1165 append "/pack/" without clobbering the stack even if
1166 strlen(objdir) were PATH_MAX. */
1167 char path[PATH_MAX + 1 + 4 + 1 + 1];
1171 struct string_list garbage = STRING_LIST_INIT_DUP;
1173 sprintf(path, "%s/pack", objdir);
1175 dir = opendir(path);
1177 if (errno != ENOENT)
1178 error("unable to open object pack directory: %s: %s",
1179 path, strerror(errno));
1183 while ((de = readdir(dir)) != NULL) {
1184 int namelen = strlen(de->d_name);
1185 struct packed_git *p;
1187 if (len + namelen + 1 > sizeof(path)) {
1188 if (report_garbage) {
1189 struct strbuf sb = STRBUF_INIT;
1190 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1191 report_garbage("path too long", sb.buf);
1192 strbuf_release(&sb);
1197 if (is_dot_or_dotdot(de->d_name))
1200 strcpy(path + len, de->d_name);
1202 if (has_extension(de->d_name, ".idx")) {
1203 /* Don't reopen a pack we already have. */
1204 for (p = packed_git; p; p = p->next) {
1205 if (!memcmp(path, p->pack_name, len + namelen - 4))
1210 * See if it really is a valid .idx file with
1211 * corresponding .pack file that we can map.
1213 (p = add_packed_git(path, len + namelen, local)) != NULL)
1214 install_packed_git(p);
1217 if (!report_garbage)
1220 if (has_extension(de->d_name, ".idx") ||
1221 has_extension(de->d_name, ".pack") ||
1222 has_extension(de->d_name, ".keep"))
1223 string_list_append(&garbage, path);
1225 report_garbage("garbage found", path);
1228 report_pack_garbage(&garbage);
1229 string_list_clear(&garbage, 0);
1232 static int sort_pack(const void *a_, const void *b_)
1234 struct packed_git *a = *((struct packed_git **)a_);
1235 struct packed_git *b = *((struct packed_git **)b_);
1239 * Local packs tend to contain objects specific to our
1240 * variant of the project than remote ones. In addition,
1241 * remote ones could be on a network mounted filesystem.
1242 * Favor local ones for these reasons.
1244 st = a->pack_local - b->pack_local;
1249 * Younger packs tend to contain more recent objects,
1250 * and more recent objects tend to get accessed more
1253 if (a->mtime < b->mtime)
1255 else if (a->mtime == b->mtime)
1260 static void rearrange_packed_git(void)
1262 struct packed_git **ary, *p;
1265 for (n = 0, p = packed_git; p; p = p->next)
1270 /* prepare an array of packed_git for easier sorting */
1271 ary = xcalloc(n, sizeof(struct packed_git *));
1272 for (n = 0, p = packed_git; p; p = p->next)
1275 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1277 /* link them back again */
1278 for (i = 0; i < n - 1; i++)
1279 ary[i]->next = ary[i + 1];
1280 ary[n - 1]->next = NULL;
1281 packed_git = ary[0];
1286 static int prepare_packed_git_run_once = 0;
1287 void prepare_packed_git(void)
1289 struct alternate_object_database *alt;
1291 if (prepare_packed_git_run_once)
1293 prepare_packed_git_one(get_object_directory(), 1);
1295 for (alt = alt_odb_list; alt; alt = alt->next) {
1297 prepare_packed_git_one(alt->base, 0);
1298 alt->name[-1] = '/';
1300 rearrange_packed_git();
1301 prepare_packed_git_run_once = 1;
1304 void reprepare_packed_git(void)
1307 prepare_packed_git_run_once = 0;
1308 prepare_packed_git();
1311 static void mark_bad_packed_object(struct packed_git *p,
1312 const unsigned char *sha1)
1315 for (i = 0; i < p->num_bad_objects; i++)
1316 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1318 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1319 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1320 p->num_bad_objects++;
1323 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1325 struct packed_git *p;
1328 for (p = packed_git; p; p = p->next)
1329 for (i = 0; i < p->num_bad_objects; i++)
1330 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1336 * With an in-core object data in "map", rehash it to make sure the
1337 * object name actually matches "sha1" to detect object corruption.
1338 * With "map" == NULL, try reading the object named with "sha1" using
1339 * the streaming interface and rehash it to do the same.
1341 int check_sha1_signature(const unsigned char *sha1, void *map,
1342 unsigned long size, const char *type)
1344 unsigned char real_sha1[20];
1345 enum object_type obj_type;
1346 struct git_istream *st;
1352 hash_sha1_file(map, size, type, real_sha1);
1353 return hashcmp(sha1, real_sha1) ? -1 : 0;
1356 st = open_istream(sha1, &obj_type, &size, NULL);
1360 /* Generate the header */
1361 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1365 git_SHA1_Update(&c, hdr, hdrlen);
1367 char buf[1024 * 16];
1368 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1376 git_SHA1_Update(&c, buf, readlen);
1378 git_SHA1_Final(real_sha1, &c);
1380 return hashcmp(sha1, real_sha1) ? -1 : 0;
1383 static int git_open_noatime(const char *name)
1385 static int sha1_file_open_flag = O_NOATIME;
1388 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1392 /* Might the failure be due to O_NOATIME? */
1393 if (errno != ENOENT && sha1_file_open_flag) {
1394 sha1_file_open_flag = 0;
1402 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1404 char *name = sha1_file_name(sha1);
1405 struct alternate_object_database *alt;
1407 if (!lstat(name, st))
1412 for (alt = alt_odb_list; alt; alt = alt->next) {
1414 fill_sha1_path(name, sha1);
1415 if (!lstat(alt->base, st))
1422 static int open_sha1_file(const unsigned char *sha1)
1425 char *name = sha1_file_name(sha1);
1426 struct alternate_object_database *alt;
1428 fd = git_open_noatime(name);
1434 for (alt = alt_odb_list; alt; alt = alt->next) {
1436 fill_sha1_path(name, sha1);
1437 fd = git_open_noatime(alt->base);
1444 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1449 fd = open_sha1_file(sha1);
1454 if (!fstat(fd, &st)) {
1455 *size = xsize_t(st.st_size);
1457 /* mmap() is forbidden on empty files */
1458 error("object file %s is empty", sha1_file_name(sha1));
1461 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1469 * There used to be a second loose object header format which
1470 * was meant to mimic the in-pack format, allowing for direct
1471 * copy of the object data. This format turned up not to be
1472 * really worth it and we no longer write loose objects in that
1475 static int experimental_loose_object(unsigned char *map)
1480 * We must determine if the buffer contains the standard
1481 * zlib-deflated stream or the experimental format based
1482 * on the in-pack object format. Compare the header byte
1485 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1486 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1488 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1489 * in standard loose-object format, UNLESS it is a Git-pack
1490 * format object *exactly* 8 bytes in size when inflated.
1492 * However, RFC1950 also specifies that the 1st 16-bit word
1493 * must be divisible by 31 - this checksum tells us our buffer
1494 * is in the standard format, giving a false positive only if
1495 * the 1st word of the Git-pack format object happens to be
1496 * divisible by 31, ie:
1497 * ((byte0 * 256) + byte1) % 31 = 0
1498 * => 0ttt10000www1000 % 31 = 0
1500 * As it happens, this case can only arise for www=3 & ttt=1
1501 * - ie, a Commit object, which would have to be 8 bytes in
1502 * size. As no Commit can be that small, we find that the
1503 * combination of these two criteria (bitmask & checksum)
1504 * can always correctly determine the buffer format.
1506 word = (map[0] << 8) + map[1];
1507 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1513 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1514 unsigned long len, enum object_type *type, unsigned long *sizep)
1517 unsigned long size, c;
1518 unsigned long used = 0;
1521 *type = (c >> 4) & 7;
1525 if (len <= used || bitsizeof(long) <= shift) {
1526 error("bad object header");
1531 size += (c & 0x7f) << shift;
1538 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1540 unsigned long size, used;
1541 static const char valid_loose_object_type[8] = {
1543 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1544 0, /* "delta" and others are invalid in a loose object */
1546 enum object_type type;
1548 /* Get the data stream */
1549 memset(stream, 0, sizeof(*stream));
1550 stream->next_in = map;
1551 stream->avail_in = mapsize;
1552 stream->next_out = buffer;
1553 stream->avail_out = bufsiz;
1555 if (experimental_loose_object(map)) {
1557 * The old experimental format we no longer produce;
1558 * we can still read it.
1560 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1561 if (!used || !valid_loose_object_type[type])
1566 /* Set up the stream for the rest.. */
1567 stream->next_in = map;
1568 stream->avail_in = mapsize;
1569 git_inflate_init(stream);
1571 /* And generate the fake traditional header */
1572 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1573 typename(type), size);
1576 git_inflate_init(stream);
1577 return git_inflate(stream, 0);
1580 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1582 int bytes = strlen(buffer) + 1;
1583 unsigned char *buf = xmallocz(size);
1587 n = stream->total_out - bytes;
1590 memcpy(buf, (char *) buffer + bytes, n);
1592 if (bytes <= size) {
1594 * The above condition must be (bytes <= size), not
1595 * (bytes < size). In other words, even though we
1596 * expect no more output and set avail_out to zero,
1597 * the input zlib stream may have bytes that express
1598 * "this concludes the stream", and we *do* want to
1601 * Otherwise we would not be able to test that we
1602 * consumed all the input to reach the expected size;
1603 * we also want to check that zlib tells us that all
1604 * went well with status == Z_STREAM_END at the end.
1606 stream->next_out = buf + bytes;
1607 stream->avail_out = size - bytes;
1608 while (status == Z_OK)
1609 status = git_inflate(stream, Z_FINISH);
1611 if (status == Z_STREAM_END && !stream->avail_in) {
1612 git_inflate_end(stream);
1617 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1618 else if (stream->avail_in)
1619 error("garbage at end of loose object '%s'",
1626 * We used to just use "sscanf()", but that's actually way
1627 * too permissive for what we want to check. So do an anal
1628 * object header parse by hand.
1630 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1637 * The type can be at most ten bytes (including the
1638 * terminating '\0' that we add), and is followed by
1647 if (i >= sizeof(type))
1653 * The length must follow immediately, and be in canonical
1654 * decimal format (ie "010" is not valid).
1656 size = *hdr++ - '0';
1661 unsigned long c = *hdr - '0';
1665 size = size * 10 + c;
1671 * The length must be followed by a zero byte
1673 return *hdr ? -1 : type_from_string(type);
1676 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1682 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1683 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1686 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1689 unsigned long get_size_from_delta(struct packed_git *p,
1690 struct pack_window **w_curs,
1693 const unsigned char *data;
1694 unsigned char delta_head[20], *in;
1698 memset(&stream, 0, sizeof(stream));
1699 stream.next_out = delta_head;
1700 stream.avail_out = sizeof(delta_head);
1702 git_inflate_init(&stream);
1704 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1705 stream.next_in = in;
1706 st = git_inflate(&stream, Z_FINISH);
1707 curpos += stream.next_in - in;
1708 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1709 stream.total_out < sizeof(delta_head));
1710 git_inflate_end(&stream);
1711 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1712 error("delta data unpack-initial failed");
1716 /* Examine the initial part of the delta to figure out
1721 /* ignore base size */
1722 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1724 /* Read the result size */
1725 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1728 static off_t get_delta_base(struct packed_git *p,
1729 struct pack_window **w_curs,
1731 enum object_type type,
1732 off_t delta_obj_offset)
1734 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1737 /* use_pack() assured us we have [base_info, base_info + 20)
1738 * as a range that we can look at without walking off the
1739 * end of the mapped window. Its actually the hash size
1740 * that is assured. An OFS_DELTA longer than the hash size
1741 * is stupid, as then a REF_DELTA would be smaller to store.
1743 if (type == OBJ_OFS_DELTA) {
1745 unsigned char c = base_info[used++];
1746 base_offset = c & 127;
1749 if (!base_offset || MSB(base_offset, 7))
1750 return 0; /* overflow */
1751 c = base_info[used++];
1752 base_offset = (base_offset << 7) + (c & 127);
1754 base_offset = delta_obj_offset - base_offset;
1755 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1756 return 0; /* out of bound */
1758 } else if (type == OBJ_REF_DELTA) {
1759 /* The base entry _must_ be in the same pack */
1760 base_offset = find_pack_entry_one(base_info, p);
1763 die("I am totally screwed");
1767 int unpack_object_header(struct packed_git *p,
1768 struct pack_window **w_curs,
1770 unsigned long *sizep)
1772 unsigned char *base;
1775 enum object_type type;
1777 /* use_pack() assures us we have [base, base + 20) available
1778 * as a range that we can look at. (Its actually the hash
1779 * size that is assured.) With our object header encoding
1780 * the maximum deflated object size is 2^137, which is just
1781 * insane, so we know won't exceed what we have been given.
1783 base = use_pack(p, w_curs, *curpos, &left);
1784 used = unpack_object_header_buffer(base, left, &type, sizep);
1793 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1796 struct revindex_entry *revidx;
1797 const unsigned char *sha1;
1798 revidx = find_pack_revindex(p, obj_offset);
1801 sha1 = nth_packed_object_sha1(p, revidx->nr);
1802 mark_bad_packed_object(p, sha1);
1803 type = sha1_object_info(sha1, NULL);
1804 if (type <= OBJ_NONE)
1809 #define POI_STACK_PREALLOC 64
1811 static enum object_type packed_to_object_type(struct packed_git *p,
1813 enum object_type type,
1814 struct pack_window **w_curs,
1817 off_t small_poi_stack[POI_STACK_PREALLOC];
1818 off_t *poi_stack = small_poi_stack;
1819 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1821 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1824 /* Push the object we're going to leave behind */
1825 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1826 poi_stack_alloc = alloc_nr(poi_stack_nr);
1827 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1828 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1830 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1832 poi_stack[poi_stack_nr++] = obj_offset;
1833 /* If parsing the base offset fails, just unwind */
1834 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1837 curpos = obj_offset = base_offset;
1838 type = unpack_object_header(p, w_curs, &curpos, &size);
1839 if (type <= OBJ_NONE) {
1840 /* If getting the base itself fails, we first
1841 * retry the base, otherwise unwind */
1842 type = retry_bad_packed_offset(p, base_offset);
1843 if (type > OBJ_NONE)
1857 error("unknown object type %i at offset %"PRIuMAX" in %s",
1858 type, (uintmax_t)obj_offset, p->pack_name);
1863 if (poi_stack != small_poi_stack)
1868 while (poi_stack_nr) {
1869 obj_offset = poi_stack[--poi_stack_nr];
1870 type = retry_bad_packed_offset(p, obj_offset);
1871 if (type > OBJ_NONE)
1878 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1879 struct object_info *oi)
1881 struct pack_window *w_curs = NULL;
1883 off_t curpos = obj_offset;
1884 enum object_type type;
1887 * We always get the representation type, but only convert it to
1888 * a "real" type later if the caller is interested.
1890 type = unpack_object_header(p, &w_curs, &curpos, &size);
1893 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1894 off_t tmp_pos = curpos;
1895 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1901 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1902 if (*oi->sizep == 0) {
1911 if (oi->disk_sizep) {
1912 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1913 *oi->disk_sizep = revidx[1].offset - obj_offset;
1917 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1918 if (*oi->typep < 0) {
1925 unuse_pack(&w_curs);
1929 static void *unpack_compressed_entry(struct packed_git *p,
1930 struct pack_window **w_curs,
1936 unsigned char *buffer, *in;
1938 buffer = xmallocz(size);
1939 memset(&stream, 0, sizeof(stream));
1940 stream.next_out = buffer;
1941 stream.avail_out = size + 1;
1943 git_inflate_init(&stream);
1945 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1946 stream.next_in = in;
1947 st = git_inflate(&stream, Z_FINISH);
1948 if (!stream.avail_out)
1949 break; /* the payload is larger than it should be */
1950 curpos += stream.next_in - in;
1951 } while (st == Z_OK || st == Z_BUF_ERROR);
1952 git_inflate_end(&stream);
1953 if ((st != Z_STREAM_END) || stream.total_out != size) {
1961 #define MAX_DELTA_CACHE (256)
1963 static size_t delta_base_cached;
1965 static struct delta_base_cache_lru_list {
1966 struct delta_base_cache_lru_list *prev;
1967 struct delta_base_cache_lru_list *next;
1968 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1970 static struct delta_base_cache_entry {
1971 struct delta_base_cache_lru_list lru;
1973 struct packed_git *p;
1976 enum object_type type;
1977 } delta_base_cache[MAX_DELTA_CACHE];
1979 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1983 hash = (unsigned long)p + (unsigned long)base_offset;
1984 hash += (hash >> 8) + (hash >> 16);
1985 return hash % MAX_DELTA_CACHE;
1988 static struct delta_base_cache_entry *
1989 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1991 unsigned long hash = pack_entry_hash(p, base_offset);
1992 return delta_base_cache + hash;
1995 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1996 struct packed_git *p, off_t base_offset)
1998 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2001 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2003 struct delta_base_cache_entry *ent;
2004 ent = get_delta_base_cache_entry(p, base_offset);
2005 return eq_delta_base_cache_entry(ent, p, base_offset);
2008 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2011 ent->lru.next->prev = ent->lru.prev;
2012 ent->lru.prev->next = ent->lru.next;
2013 delta_base_cached -= ent->size;
2016 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2017 unsigned long *base_size, enum object_type *type, int keep_cache)
2019 struct delta_base_cache_entry *ent;
2022 ent = get_delta_base_cache_entry(p, base_offset);
2024 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2025 return unpack_entry(p, base_offset, type, base_size);
2030 clear_delta_base_cache_entry(ent);
2032 ret = xmemdupz(ent->data, ent->size);
2034 *base_size = ent->size;
2038 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2043 ent->lru.next->prev = ent->lru.prev;
2044 ent->lru.prev->next = ent->lru.next;
2045 delta_base_cached -= ent->size;
2049 void clear_delta_base_cache(void)
2052 for (p = 0; p < MAX_DELTA_CACHE; p++)
2053 release_delta_base_cache(&delta_base_cache[p]);
2056 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2057 void *base, unsigned long base_size, enum object_type type)
2059 unsigned long hash = pack_entry_hash(p, base_offset);
2060 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2061 struct delta_base_cache_lru_list *lru;
2063 release_delta_base_cache(ent);
2064 delta_base_cached += base_size;
2066 for (lru = delta_base_cache_lru.next;
2067 delta_base_cached > delta_base_cache_limit
2068 && lru != &delta_base_cache_lru;
2070 struct delta_base_cache_entry *f = (void *)lru;
2071 if (f->type == OBJ_BLOB)
2072 release_delta_base_cache(f);
2074 for (lru = delta_base_cache_lru.next;
2075 delta_base_cached > delta_base_cache_limit
2076 && lru != &delta_base_cache_lru;
2078 struct delta_base_cache_entry *f = (void *)lru;
2079 release_delta_base_cache(f);
2083 ent->base_offset = base_offset;
2086 ent->size = base_size;
2087 ent->lru.next = &delta_base_cache_lru;
2088 ent->lru.prev = delta_base_cache_lru.prev;
2089 delta_base_cache_lru.prev->next = &ent->lru;
2090 delta_base_cache_lru.prev = &ent->lru;
2093 static void *read_object(const unsigned char *sha1, enum object_type *type,
2094 unsigned long *size);
2096 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2098 static FILE *log_file;
2100 if (!log_pack_access)
2101 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2102 if (!log_pack_access)
2103 log_pack_access = no_log_pack_access;
2104 if (log_pack_access == no_log_pack_access)
2108 log_file = fopen(log_pack_access, "w");
2110 error("cannot open pack access log '%s' for writing: %s",
2111 log_pack_access, strerror(errno));
2112 log_pack_access = no_log_pack_access;
2116 fprintf(log_file, "%s %"PRIuMAX"\n",
2117 p->pack_name, (uintmax_t)obj_offset);
2121 int do_check_packed_object_crc;
2123 #define UNPACK_ENTRY_STACK_PREALLOC 64
2124 struct unpack_entry_stack_ent {
2130 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2131 enum object_type *final_type, unsigned long *final_size)
2133 struct pack_window *w_curs = NULL;
2134 off_t curpos = obj_offset;
2137 enum object_type type;
2138 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2139 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2140 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2141 int base_from_cache = 0;
2143 if (log_pack_access != no_log_pack_access)
2144 write_pack_access_log(p, obj_offset);
2146 /* PHASE 1: drill down to the innermost base object */
2150 struct delta_base_cache_entry *ent;
2152 ent = get_delta_base_cache_entry(p, curpos);
2153 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2157 clear_delta_base_cache_entry(ent);
2158 base_from_cache = 1;
2162 if (do_check_packed_object_crc && p->index_version > 1) {
2163 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2164 unsigned long len = revidx[1].offset - obj_offset;
2165 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2166 const unsigned char *sha1 =
2167 nth_packed_object_sha1(p, revidx->nr);
2168 error("bad packed object CRC for %s",
2170 mark_bad_packed_object(p, sha1);
2171 unuse_pack(&w_curs);
2176 type = unpack_object_header(p, &w_curs, &curpos, &size);
2177 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2180 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2182 error("failed to validate delta base reference "
2183 "at offset %"PRIuMAX" from %s",
2184 (uintmax_t)curpos, p->pack_name);
2185 /* bail to phase 2, in hopes of recovery */
2190 /* push object, proceed to base */
2191 if (delta_stack_nr >= delta_stack_alloc
2192 && delta_stack == small_delta_stack) {
2193 delta_stack_alloc = alloc_nr(delta_stack_nr);
2194 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2195 memcpy(delta_stack, small_delta_stack,
2196 sizeof(*delta_stack)*delta_stack_nr);
2198 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2200 i = delta_stack_nr++;
2201 delta_stack[i].obj_offset = obj_offset;
2202 delta_stack[i].curpos = curpos;
2203 delta_stack[i].size = size;
2205 curpos = obj_offset = base_offset;
2208 /* PHASE 2: handle the base */
2213 die("BUG in unpack_entry: left loop at a valid delta");
2219 if (!base_from_cache)
2220 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2224 error("unknown object type %i at offset %"PRIuMAX" in %s",
2225 type, (uintmax_t)obj_offset, p->pack_name);
2228 /* PHASE 3: apply deltas in order */
2231 * 'data' holds the base data, or NULL if there was corruption
2233 while (delta_stack_nr) {
2236 unsigned long delta_size, base_size = size;
2242 add_delta_base_cache(p, obj_offset, base, base_size, type);
2246 * We're probably in deep shit, but let's try to fetch
2247 * the required base anyway from another pack or loose.
2248 * This is costly but should happen only in the presence
2249 * of a corrupted pack, and is better than failing outright.
2251 struct revindex_entry *revidx;
2252 const unsigned char *base_sha1;
2253 revidx = find_pack_revindex(p, obj_offset);
2255 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2256 error("failed to read delta base object %s"
2257 " at offset %"PRIuMAX" from %s",
2258 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2260 mark_bad_packed_object(p, base_sha1);
2261 base = read_object(base_sha1, &type, &base_size);
2265 i = --delta_stack_nr;
2266 obj_offset = delta_stack[i].obj_offset;
2267 curpos = delta_stack[i].curpos;
2268 delta_size = delta_stack[i].size;
2273 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2276 error("failed to unpack compressed delta "
2277 "at offset %"PRIuMAX" from %s",
2278 (uintmax_t)curpos, p->pack_name);
2283 data = patch_delta(base, base_size,
2284 delta_data, delta_size,
2288 * We could not apply the delta; warn the user, but keep going.
2289 * Our failure will be noticed either in the next iteration of
2290 * the loop, or if this is the final delta, in the caller when
2291 * we return NULL. Those code paths will take care of making
2292 * a more explicit warning and retrying with another copy of
2296 error("failed to apply delta");
2304 unuse_pack(&w_curs);
2308 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2311 const unsigned char *index = p->index_data;
2313 if (open_pack_index(p))
2315 index = p->index_data;
2317 if (n >= p->num_objects)
2320 if (p->index_version == 1) {
2321 return index + 24 * n + 4;
2324 return index + 20 * n;
2328 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2330 const unsigned char *index = p->index_data;
2332 if (p->index_version == 1) {
2333 return ntohl(*((uint32_t *)(index + 24 * n)));
2336 index += 8 + p->num_objects * (20 + 4);
2337 off = ntohl(*((uint32_t *)(index + 4 * n)));
2338 if (!(off & 0x80000000))
2340 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2341 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2342 ntohl(*((uint32_t *)(index + 4)));
2346 off_t find_pack_entry_one(const unsigned char *sha1,
2347 struct packed_git *p)
2349 const uint32_t *level1_ofs = p->index_data;
2350 const unsigned char *index = p->index_data;
2351 unsigned hi, lo, stride;
2352 static int use_lookup = -1;
2353 static int debug_lookup = -1;
2355 if (debug_lookup < 0)
2356 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2359 if (open_pack_index(p))
2361 level1_ofs = p->index_data;
2362 index = p->index_data;
2364 if (p->index_version > 1) {
2369 hi = ntohl(level1_ofs[*sha1]);
2370 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2371 if (p->index_version > 1) {
2379 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2380 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2383 use_lookup = !!getenv("GIT_USE_LOOKUP");
2385 int pos = sha1_entry_pos(index, stride, 0,
2386 lo, hi, p->num_objects, sha1);
2389 return nth_packed_object_offset(p, pos);
2393 unsigned mi = (lo + hi) / 2;
2394 int cmp = hashcmp(index + mi * stride, sha1);
2397 printf("lo %u hi %u rg %u mi %u\n",
2398 lo, hi, hi - lo, mi);
2400 return nth_packed_object_offset(p, mi);
2409 int is_pack_valid(struct packed_git *p)
2411 /* An already open pack is known to be valid. */
2412 if (p->pack_fd != -1)
2415 /* If the pack has one window completely covering the
2416 * file size, the pack is known to be valid even if
2417 * the descriptor is not currently open.
2420 struct pack_window *w = p->windows;
2422 if (!w->offset && w->len == p->pack_size)
2426 /* Force the pack to open to prove its valid. */
2427 return !open_packed_git(p);
2430 static int fill_pack_entry(const unsigned char *sha1,
2431 struct pack_entry *e,
2432 struct packed_git *p)
2436 if (p->num_bad_objects) {
2438 for (i = 0; i < p->num_bad_objects; i++)
2439 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2443 offset = find_pack_entry_one(sha1, p);
2448 * We are about to tell the caller where they can locate the
2449 * requested object. We better make sure the packfile is
2450 * still here and can be accessed before supplying that
2451 * answer, as it may have been deleted since the index was
2454 if (!is_pack_valid(p)) {
2455 warning("packfile %s cannot be accessed", p->pack_name);
2460 hashcpy(e->sha1, sha1);
2464 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2466 struct packed_git *p;
2468 prepare_packed_git();
2472 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2475 for (p = packed_git; p; p = p->next) {
2476 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2479 last_found_pack = p;
2485 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2486 struct packed_git *packs)
2488 struct packed_git *p;
2490 for (p = packs; p; p = p->next) {
2491 if (find_pack_entry_one(sha1, p))
2498 static int sha1_loose_object_info(const unsigned char *sha1,
2499 struct object_info *oi)
2502 unsigned long mapsize, size;
2508 * If we don't care about type or size, then we don't
2509 * need to look inside the object at all. Note that we
2510 * do not optimize out the stat call, even if the
2511 * caller doesn't care about the disk-size, since our
2512 * return value implicitly indicates whether the
2513 * object even exists.
2515 if (!oi->typep && !oi->sizep) {
2517 if (stat_sha1_file(sha1, &st) < 0)
2520 *oi->disk_sizep = st.st_size;
2524 map = map_sha1_file(sha1, &mapsize);
2528 *oi->disk_sizep = mapsize;
2529 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2530 status = error("unable to unpack %s header",
2532 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2533 status = error("unable to parse %s header", sha1_to_hex(sha1));
2536 git_inflate_end(&stream);
2537 munmap(map, mapsize);
2539 *oi->typep = status;
2543 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2545 struct cached_object *co;
2546 struct pack_entry e;
2549 co = find_cached_object(sha1);
2552 *(oi->typep) = co->type;
2554 *(oi->sizep) = co->size;
2556 *(oi->disk_sizep) = 0;
2557 oi->whence = OI_CACHED;
2561 if (!find_pack_entry(sha1, &e)) {
2562 /* Most likely it's a loose object. */
2563 if (!sha1_loose_object_info(sha1, oi)) {
2564 oi->whence = OI_LOOSE;
2568 /* Not a loose object; someone else may have just packed it. */
2569 reprepare_packed_git();
2570 if (!find_pack_entry(sha1, &e))
2574 rtype = packed_object_info(e.p, e.offset, oi);
2576 mark_bad_packed_object(e.p, sha1);
2577 return sha1_object_info_extended(sha1, oi);
2578 } else if (in_delta_base_cache(e.p, e.offset)) {
2579 oi->whence = OI_DBCACHED;
2581 oi->whence = OI_PACKED;
2582 oi->u.packed.offset = e.offset;
2583 oi->u.packed.pack = e.p;
2584 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2585 rtype == OBJ_OFS_DELTA);
2591 /* returns enum object_type or negative */
2592 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2594 enum object_type type;
2595 struct object_info oi = {NULL};
2599 if (sha1_object_info_extended(sha1, &oi) < 0)
2604 static void *read_packed_sha1(const unsigned char *sha1,
2605 enum object_type *type, unsigned long *size)
2607 struct pack_entry e;
2610 if (!find_pack_entry(sha1, &e))
2612 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2615 * We're probably in deep shit, but let's try to fetch
2616 * the required object anyway from another pack or loose.
2617 * This should happen only in the presence of a corrupted
2618 * pack, and is better than failing outright.
2620 error("failed to read object %s at offset %"PRIuMAX" from %s",
2621 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2622 mark_bad_packed_object(e.p, sha1);
2623 data = read_object(sha1, type, size);
2628 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2629 unsigned char *sha1)
2631 struct cached_object *co;
2633 hash_sha1_file(buf, len, typename(type), sha1);
2634 if (has_sha1_file(sha1) || find_cached_object(sha1))
2636 if (cached_object_alloc <= cached_object_nr) {
2637 cached_object_alloc = alloc_nr(cached_object_alloc);
2638 cached_objects = xrealloc(cached_objects,
2639 sizeof(*cached_objects) *
2640 cached_object_alloc);
2642 co = &cached_objects[cached_object_nr++];
2645 co->buf = xmalloc(len);
2646 memcpy(co->buf, buf, len);
2647 hashcpy(co->sha1, sha1);
2651 static void *read_object(const unsigned char *sha1, enum object_type *type,
2652 unsigned long *size)
2654 unsigned long mapsize;
2656 struct cached_object *co;
2658 co = find_cached_object(sha1);
2662 return xmemdupz(co->buf, co->size);
2665 buf = read_packed_sha1(sha1, type, size);
2668 map = map_sha1_file(sha1, &mapsize);
2670 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2671 munmap(map, mapsize);
2674 reprepare_packed_git();
2675 return read_packed_sha1(sha1, type, size);
2679 * This function dies on corrupt objects; the callers who want to
2680 * deal with them should arrange to call read_object() and give error
2681 * messages themselves.
2683 void *read_sha1_file_extended(const unsigned char *sha1,
2684 enum object_type *type,
2685 unsigned long *size,
2690 const struct packed_git *p;
2691 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2692 ? lookup_replace_object(sha1) : sha1;
2695 data = read_object(repl, type, size);
2699 if (errno && errno != ENOENT)
2700 die_errno("failed to read object %s", sha1_to_hex(sha1));
2702 /* die if we replaced an object with one that does not exist */
2704 die("replacement %s not found for %s",
2705 sha1_to_hex(repl), sha1_to_hex(sha1));
2707 if (has_loose_object(repl)) {
2708 path = sha1_file_name(sha1);
2709 die("loose object %s (stored in %s) is corrupt",
2710 sha1_to_hex(repl), path);
2713 if ((p = has_packed_and_bad(repl)) != NULL)
2714 die("packed object %s (stored in %s) is corrupt",
2715 sha1_to_hex(repl), p->pack_name);
2720 void *read_object_with_reference(const unsigned char *sha1,
2721 const char *required_type_name,
2722 unsigned long *size,
2723 unsigned char *actual_sha1_return)
2725 enum object_type type, required_type;
2727 unsigned long isize;
2728 unsigned char actual_sha1[20];
2730 required_type = type_from_string(required_type_name);
2731 hashcpy(actual_sha1, sha1);
2733 int ref_length = -1;
2734 const char *ref_type = NULL;
2736 buffer = read_sha1_file(actual_sha1, &type, &isize);
2739 if (type == required_type) {
2741 if (actual_sha1_return)
2742 hashcpy(actual_sha1_return, actual_sha1);
2745 /* Handle references */
2746 else if (type == OBJ_COMMIT)
2748 else if (type == OBJ_TAG)
2749 ref_type = "object ";
2754 ref_length = strlen(ref_type);
2756 if (ref_length + 40 > isize ||
2757 memcmp(buffer, ref_type, ref_length) ||
2758 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2763 /* Now we have the ID of the referred-to object in
2764 * actual_sha1. Check again. */
2768 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2769 const char *type, unsigned char *sha1,
2770 char *hdr, int *hdrlen)
2774 /* Generate the header */
2775 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2779 git_SHA1_Update(&c, hdr, *hdrlen);
2780 git_SHA1_Update(&c, buf, len);
2781 git_SHA1_Final(sha1, &c);
2785 * Move the just written object into its final resting place.
2786 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2787 * "moving" is only a part of what it does, when no patch between
2788 * master to pu changes the call sites of this function.
2790 int move_temp_to_file(const char *tmpfile, const char *filename)
2794 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2796 else if (link(tmpfile, filename))
2800 * Coda hack - coda doesn't like cross-directory links,
2801 * so we fall back to a rename, which will mean that it
2802 * won't be able to check collisions, but that's not a
2805 * The same holds for FAT formatted media.
2807 * When this succeeds, we just return. We have nothing
2810 if (ret && ret != EEXIST) {
2812 if (!rename(tmpfile, filename))
2816 unlink_or_warn(tmpfile);
2818 if (ret != EEXIST) {
2819 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2821 /* FIXME!!! Collision check here ? */
2825 if (adjust_shared_perm(filename))
2826 return error("unable to set permission to '%s'", filename);
2830 static int write_buffer(int fd, const void *buf, size_t len)
2832 if (write_in_full(fd, buf, len) < 0)
2833 return error("file write error (%s)", strerror(errno));
2837 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2838 unsigned char *sha1)
2842 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2846 /* Finalize a file on disk, and close it. */
2847 static void close_sha1_file(int fd)
2849 if (fsync_object_files)
2850 fsync_or_die(fd, "sha1 file");
2852 die_errno("error when closing sha1 file");
2855 /* Size of directory component, including the ending '/' */
2856 static inline int directory_size(const char *filename)
2858 const char *s = strrchr(filename, '/');
2861 return s - filename + 1;
2865 * This creates a temporary file in the same directory as the final
2868 * We want to avoid cross-directory filename renames, because those
2869 * can have problems on various filesystems (FAT, NFS, Coda).
2871 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2873 int fd, dirlen = directory_size(filename);
2875 if (dirlen + 20 > bufsiz) {
2876 errno = ENAMETOOLONG;
2879 memcpy(buffer, filename, dirlen);
2880 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2881 fd = git_mkstemp_mode(buffer, 0444);
2882 if (fd < 0 && dirlen && errno == ENOENT) {
2883 /* Make sure the directory exists */
2884 memcpy(buffer, filename, dirlen);
2885 buffer[dirlen-1] = 0;
2886 if (mkdir(buffer, 0777) && errno != EEXIST)
2888 if (adjust_shared_perm(buffer))
2892 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2893 fd = git_mkstemp_mode(buffer, 0444);
2898 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2899 const void *buf, unsigned long len, time_t mtime)
2902 unsigned char compressed[4096];
2905 unsigned char parano_sha1[20];
2907 static char tmp_file[PATH_MAX];
2909 filename = sha1_file_name(sha1);
2910 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2912 if (errno == EACCES)
2913 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2915 return error("unable to create temporary file: %s", strerror(errno));
2919 memset(&stream, 0, sizeof(stream));
2920 git_deflate_init(&stream, zlib_compression_level);
2921 stream.next_out = compressed;
2922 stream.avail_out = sizeof(compressed);
2925 /* First header.. */
2926 stream.next_in = (unsigned char *)hdr;
2927 stream.avail_in = hdrlen;
2928 while (git_deflate(&stream, 0) == Z_OK)
2930 git_SHA1_Update(&c, hdr, hdrlen);
2932 /* Then the data itself.. */
2933 stream.next_in = (void *)buf;
2934 stream.avail_in = len;
2936 unsigned char *in0 = stream.next_in;
2937 ret = git_deflate(&stream, Z_FINISH);
2938 git_SHA1_Update(&c, in0, stream.next_in - in0);
2939 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2940 die("unable to write sha1 file");
2941 stream.next_out = compressed;
2942 stream.avail_out = sizeof(compressed);
2943 } while (ret == Z_OK);
2945 if (ret != Z_STREAM_END)
2946 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2947 ret = git_deflate_end_gently(&stream);
2949 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2950 git_SHA1_Final(parano_sha1, &c);
2951 if (hashcmp(sha1, parano_sha1) != 0)
2952 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2954 close_sha1_file(fd);
2959 utb.modtime = mtime;
2960 if (utime(tmp_file, &utb) < 0)
2961 warning("failed utime() on %s: %s",
2962 tmp_file, strerror(errno));
2965 return move_temp_to_file(tmp_file, filename);
2968 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2970 unsigned char sha1[20];
2974 /* Normally if we have it in the pack then we do not bother writing
2975 * it out into .git/objects/??/?{38} file.
2977 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2979 hashcpy(returnsha1, sha1);
2980 if (has_sha1_file(sha1))
2982 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2985 int force_object_loose(const unsigned char *sha1, time_t mtime)
2989 enum object_type type;
2994 if (has_loose_object(sha1))
2996 buf = read_packed_sha1(sha1, &type, &len);
2998 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2999 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3000 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3006 int has_pack_index(const unsigned char *sha1)
3009 if (stat(sha1_pack_index_name(sha1), &st))
3014 int has_sha1_pack(const unsigned char *sha1)
3016 struct pack_entry e;
3017 return find_pack_entry(sha1, &e);
3020 int has_sha1_file(const unsigned char *sha1)
3022 struct pack_entry e;
3024 if (find_pack_entry(sha1, &e))
3026 if (has_loose_object(sha1))
3028 reprepare_packed_git();
3029 return find_pack_entry(sha1, &e);
3032 static void check_tree(const void *buf, size_t size)
3034 struct tree_desc desc;
3035 struct name_entry entry;
3037 init_tree_desc(&desc, buf, size);
3038 while (tree_entry(&desc, &entry))
3040 * tree_entry() will die() on malformed entries */
3044 static void check_commit(const void *buf, size_t size)
3047 memset(&c, 0, sizeof(c));
3048 if (parse_commit_buffer(&c, buf, size))
3049 die("corrupt commit");
3052 static void check_tag(const void *buf, size_t size)
3055 memset(&t, 0, sizeof(t));
3056 if (parse_tag_buffer(&t, buf, size))
3060 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3061 enum object_type type,
3062 const char *path, unsigned flags)
3064 int ret, re_allocated = 0;
3065 int write_object = flags & HASH_WRITE_OBJECT;
3071 * Convert blobs to git internal format
3073 if ((type == OBJ_BLOB) && path) {
3074 struct strbuf nbuf = STRBUF_INIT;
3075 if (convert_to_git(path, buf, size, &nbuf,
3076 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3077 buf = strbuf_detach(&nbuf, &size);
3081 if (flags & HASH_FORMAT_CHECK) {
3082 if (type == OBJ_TREE)
3083 check_tree(buf, size);
3084 if (type == OBJ_COMMIT)
3085 check_commit(buf, size);
3086 if (type == OBJ_TAG)
3087 check_tag(buf, size);
3091 ret = write_sha1_file(buf, size, typename(type), sha1);
3093 ret = hash_sha1_file(buf, size, typename(type), sha1);
3099 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3100 const char *path, unsigned flags)
3102 struct strbuf sbuf = STRBUF_INIT;
3105 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3106 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3109 strbuf_release(&sbuf);
3113 #define SMALL_FILE_SIZE (32*1024)
3115 static int index_core(unsigned char *sha1, int fd, size_t size,
3116 enum object_type type, const char *path,
3122 ret = index_mem(sha1, NULL, size, type, path, flags);
3123 } else if (size <= SMALL_FILE_SIZE) {
3124 char *buf = xmalloc(size);
3125 if (size == read_in_full(fd, buf, size))
3126 ret = index_mem(sha1, buf, size, type, path, flags);
3128 ret = error("short read %s", strerror(errno));
3131 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3132 ret = index_mem(sha1, buf, size, type, path, flags);
3139 * This creates one packfile per large blob unless bulk-checkin
3140 * machinery is "plugged".
3142 * This also bypasses the usual "convert-to-git" dance, and that is on
3143 * purpose. We could write a streaming version of the converting
3144 * functions and insert that before feeding the data to fast-import
3145 * (or equivalent in-core API described above). However, that is
3146 * somewhat complicated, as we do not know the size of the filter
3147 * result, which we need to know beforehand when writing a git object.
3148 * Since the primary motivation for trying to stream from the working
3149 * tree file and to avoid mmaping it in core is to deal with large
3150 * binary blobs, they generally do not want to get any conversion, and
3151 * callers should avoid this code path when filters are requested.
3153 static int index_stream(unsigned char *sha1, int fd, size_t size,
3154 enum object_type type, const char *path,
3157 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3160 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3161 enum object_type type, const char *path, unsigned flags)
3164 size_t size = xsize_t(st->st_size);
3166 if (!S_ISREG(st->st_mode))
3167 ret = index_pipe(sha1, fd, type, path, flags);
3168 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3169 (path && would_convert_to_git(path, NULL, 0, 0)))
3170 ret = index_core(sha1, fd, size, type, path, flags);
3172 ret = index_stream(sha1, fd, size, type, path, flags);
3177 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3180 struct strbuf sb = STRBUF_INIT;
3182 switch (st->st_mode & S_IFMT) {
3184 fd = open(path, O_RDONLY);
3186 return error("open(\"%s\"): %s", path,
3188 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3189 return error("%s: failed to insert into database",
3193 if (strbuf_readlink(&sb, path, st->st_size)) {
3194 char *errstr = strerror(errno);
3195 return error("readlink(\"%s\"): %s", path,
3198 if (!(flags & HASH_WRITE_OBJECT))
3199 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3200 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3201 return error("%s: failed to insert into database",
3203 strbuf_release(&sb);
3206 return resolve_gitlink_ref(path, "HEAD", sha1);
3208 return error("%s: unsupported file type", path);
3213 int read_pack_header(int fd, struct pack_header *header)
3215 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3216 /* "eof before pack header was fully read" */
3217 return PH_ERROR_EOF;
3219 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3220 /* "protocol error (pack signature mismatch detected)" */
3221 return PH_ERROR_PACK_SIGNATURE;
3222 if (!pack_version_ok(header->hdr_version))
3223 /* "protocol error (pack version unsupported)" */
3224 return PH_ERROR_PROTOCOL;
3228 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3230 enum object_type type = sha1_object_info(sha1, NULL);
3232 die("%s is not a valid object", sha1_to_hex(sha1));
3234 die("%s is not a valid '%s' object", sha1_to_hex(sha1),