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)
812 if (getrlimit(RLIMIT_NOFILE, &lim))
813 die_errno("cannot get RLIMIT_NOFILE");
816 #elif defined(_SC_OPEN_MAX)
817 return sysconf(_SC_OPEN_MAX);
818 #elif defined(OPEN_MAX)
821 return 1; /* see the caller ;-) */
826 * Do not call this directly as this leaks p->pack_fd on error return;
827 * call open_packed_git() instead.
829 static int open_packed_git_1(struct packed_git *p)
832 struct pack_header hdr;
833 unsigned char sha1[20];
834 unsigned char *idx_sha1;
837 if (!p->index_data && open_pack_index(p))
838 return error("packfile %s index unavailable", p->pack_name);
841 unsigned int max_fds = get_max_fd_limit();
843 /* Save 3 for stdin/stdout/stderr, 22 for work */
845 pack_max_fds = max_fds - 25;
850 while (pack_max_fds <= pack_open_fds && close_one_pack())
853 p->pack_fd = git_open_noatime(p->pack_name);
854 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
858 /* If we created the struct before we had the pack we lack size. */
860 if (!S_ISREG(st.st_mode))
861 return error("packfile %s not a regular file", p->pack_name);
862 p->pack_size = st.st_size;
863 } else if (p->pack_size != st.st_size)
864 return error("packfile %s size changed", p->pack_name);
866 /* We leave these file descriptors open with sliding mmap;
867 * there is no point keeping them open across exec(), though.
869 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
871 return error("cannot determine file descriptor flags");
872 fd_flag |= FD_CLOEXEC;
873 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
874 return error("cannot set FD_CLOEXEC");
876 /* Verify we recognize this pack file format. */
877 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
878 return error("file %s is far too short to be a packfile", p->pack_name);
879 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
880 return error("file %s is not a GIT packfile", p->pack_name);
881 if (!pack_version_ok(hdr.hdr_version))
882 return error("packfile %s is version %"PRIu32" and not"
883 " supported (try upgrading GIT to a newer version)",
884 p->pack_name, ntohl(hdr.hdr_version));
886 /* Verify the pack matches its index. */
887 if (p->num_objects != ntohl(hdr.hdr_entries))
888 return error("packfile %s claims to have %"PRIu32" objects"
889 " while index indicates %"PRIu32" objects",
890 p->pack_name, ntohl(hdr.hdr_entries),
892 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
893 return error("end of packfile %s is unavailable", p->pack_name);
894 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
895 return error("packfile %s signature is unavailable", p->pack_name);
896 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
897 if (hashcmp(sha1, idx_sha1))
898 return error("packfile %s does not match index", p->pack_name);
902 static int open_packed_git(struct packed_git *p)
904 if (!open_packed_git_1(p))
906 if (p->pack_fd != -1) {
914 static int in_window(struct pack_window *win, off_t offset)
916 /* We must promise at least 20 bytes (one hash) after the
917 * offset is available from this window, otherwise the offset
918 * is not actually in this window and a different window (which
919 * has that one hash excess) must be used. This is to support
920 * the object header and delta base parsing routines below.
922 off_t win_off = win->offset;
923 return win_off <= offset
924 && (offset + 20) <= (win_off + win->len);
927 unsigned char *use_pack(struct packed_git *p,
928 struct pack_window **w_cursor,
932 struct pack_window *win = *w_cursor;
934 /* Since packfiles end in a hash of their content and it's
935 * pointless to ask for an offset into the middle of that
936 * hash, and the in_window function above wouldn't match
937 * don't allow an offset too close to the end of the file.
939 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
940 die("packfile %s cannot be accessed", p->pack_name);
941 if (offset > (p->pack_size - 20))
942 die("offset beyond end of packfile (truncated pack?)");
944 if (!win || !in_window(win, offset)) {
947 for (win = p->windows; win; win = win->next) {
948 if (in_window(win, offset))
952 size_t window_align = packed_git_window_size / 2;
955 if (p->pack_fd == -1 && open_packed_git(p))
956 die("packfile %s cannot be accessed", p->pack_name);
958 win = xcalloc(1, sizeof(*win));
959 win->offset = (offset / window_align) * window_align;
960 len = p->pack_size - win->offset;
961 if (len > packed_git_window_size)
962 len = packed_git_window_size;
963 win->len = (size_t)len;
964 pack_mapped += win->len;
965 while (packed_git_limit < pack_mapped
966 && unuse_one_window(p))
968 win->base = xmmap(NULL, win->len,
969 PROT_READ, MAP_PRIVATE,
970 p->pack_fd, win->offset);
971 if (win->base == MAP_FAILED)
972 die("packfile %s cannot be mapped: %s",
975 if (!win->offset && win->len == p->pack_size
976 && !p->do_not_close) {
983 if (pack_mapped > peak_pack_mapped)
984 peak_pack_mapped = pack_mapped;
985 if (pack_open_windows > peak_pack_open_windows)
986 peak_pack_open_windows = pack_open_windows;
987 win->next = p->windows;
991 if (win != *w_cursor) {
992 win->last_used = pack_used_ctr++;
996 offset -= win->offset;
998 *left = win->len - xsize_t(offset);
999 return win->base + offset;
1002 static struct packed_git *alloc_packed_git(int extra)
1004 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1005 memset(p, 0, sizeof(*p));
1010 static void try_to_free_pack_memory(size_t size)
1012 release_pack_memory(size);
1015 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1017 static int have_set_try_to_free_routine;
1019 struct packed_git *p = alloc_packed_git(path_len + 2);
1021 if (!have_set_try_to_free_routine) {
1022 have_set_try_to_free_routine = 1;
1023 set_try_to_free_routine(try_to_free_pack_memory);
1027 * Make sure a corresponding .pack file exists and that
1028 * the index looks sane.
1030 path_len -= strlen(".idx");
1035 memcpy(p->pack_name, path, path_len);
1037 strcpy(p->pack_name + path_len, ".keep");
1038 if (!access(p->pack_name, F_OK))
1041 strcpy(p->pack_name + path_len, ".pack");
1042 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1047 /* ok, it looks sane as far as we can check without
1048 * actually mapping the pack file.
1050 p->pack_size = st.st_size;
1051 p->pack_local = local;
1052 p->mtime = st.st_mtime;
1053 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1058 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1060 const char *path = sha1_pack_name(sha1);
1061 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1063 strcpy(p->pack_name, path);
1064 hashcpy(p->sha1, sha1);
1065 if (check_packed_git_idx(idx_path, p)) {
1073 void install_packed_git(struct packed_git *pack)
1075 if (pack->pack_fd != -1)
1078 pack->next = packed_git;
1082 void (*report_garbage)(const char *desc, const char *path);
1084 static void report_helper(const struct string_list *list,
1085 int seen_bits, int first, int last)
1088 switch (seen_bits) {
1090 msg = "no corresponding .idx nor .pack";
1093 msg = "no corresponding .idx";
1096 msg = "no corresponding .pack";
1101 for (; first < last; first++)
1102 report_garbage(msg, list->items[first].string);
1105 static void report_pack_garbage(struct string_list *list)
1107 int i, baselen = -1, first = 0, seen_bits = 0;
1109 if (!report_garbage)
1112 sort_string_list(list);
1114 for (i = 0; i < list->nr; i++) {
1115 const char *path = list->items[i].string;
1116 if (baselen != -1 &&
1117 strncmp(path, list->items[first].string, baselen)) {
1118 report_helper(list, seen_bits, first, i);
1122 if (baselen == -1) {
1123 const char *dot = strrchr(path, '.');
1125 report_garbage("garbage found", path);
1128 baselen = dot - path + 1;
1131 if (!strcmp(path + baselen, "pack"))
1133 else if (!strcmp(path + baselen, "idx"))
1136 report_helper(list, seen_bits, first, list->nr);
1139 static void prepare_packed_git_one(char *objdir, int local)
1141 /* Ensure that this buffer is large enough so that we can
1142 append "/pack/" without clobbering the stack even if
1143 strlen(objdir) were PATH_MAX. */
1144 char path[PATH_MAX + 1 + 4 + 1 + 1];
1148 struct string_list garbage = STRING_LIST_INIT_DUP;
1150 sprintf(path, "%s/pack", objdir);
1152 dir = opendir(path);
1154 if (errno != ENOENT)
1155 error("unable to open object pack directory: %s: %s",
1156 path, strerror(errno));
1160 while ((de = readdir(dir)) != NULL) {
1161 int namelen = strlen(de->d_name);
1162 struct packed_git *p;
1164 if (len + namelen + 1 > sizeof(path)) {
1165 if (report_garbage) {
1166 struct strbuf sb = STRBUF_INIT;
1167 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1168 report_garbage("path too long", sb.buf);
1169 strbuf_release(&sb);
1174 if (is_dot_or_dotdot(de->d_name))
1177 strcpy(path + len, de->d_name);
1179 if (has_extension(de->d_name, ".idx")) {
1180 /* Don't reopen a pack we already have. */
1181 for (p = packed_git; p; p = p->next) {
1182 if (!memcmp(path, p->pack_name, len + namelen - 4))
1187 * See if it really is a valid .idx file with
1188 * corresponding .pack file that we can map.
1190 (p = add_packed_git(path, len + namelen, local)) != NULL)
1191 install_packed_git(p);
1194 if (!report_garbage)
1197 if (has_extension(de->d_name, ".idx") ||
1198 has_extension(de->d_name, ".pack") ||
1199 has_extension(de->d_name, ".keep"))
1200 string_list_append(&garbage, path);
1202 report_garbage("garbage found", path);
1205 report_pack_garbage(&garbage);
1206 string_list_clear(&garbage, 0);
1209 static int sort_pack(const void *a_, const void *b_)
1211 struct packed_git *a = *((struct packed_git **)a_);
1212 struct packed_git *b = *((struct packed_git **)b_);
1216 * Local packs tend to contain objects specific to our
1217 * variant of the project than remote ones. In addition,
1218 * remote ones could be on a network mounted filesystem.
1219 * Favor local ones for these reasons.
1221 st = a->pack_local - b->pack_local;
1226 * Younger packs tend to contain more recent objects,
1227 * and more recent objects tend to get accessed more
1230 if (a->mtime < b->mtime)
1232 else if (a->mtime == b->mtime)
1237 static void rearrange_packed_git(void)
1239 struct packed_git **ary, *p;
1242 for (n = 0, p = packed_git; p; p = p->next)
1247 /* prepare an array of packed_git for easier sorting */
1248 ary = xcalloc(n, sizeof(struct packed_git *));
1249 for (n = 0, p = packed_git; p; p = p->next)
1252 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1254 /* link them back again */
1255 for (i = 0; i < n - 1; i++)
1256 ary[i]->next = ary[i + 1];
1257 ary[n - 1]->next = NULL;
1258 packed_git = ary[0];
1263 static int prepare_packed_git_run_once = 0;
1264 void prepare_packed_git(void)
1266 struct alternate_object_database *alt;
1268 if (prepare_packed_git_run_once)
1270 prepare_packed_git_one(get_object_directory(), 1);
1272 for (alt = alt_odb_list; alt; alt = alt->next) {
1274 prepare_packed_git_one(alt->base, 0);
1275 alt->name[-1] = '/';
1277 rearrange_packed_git();
1278 prepare_packed_git_run_once = 1;
1281 void reprepare_packed_git(void)
1284 prepare_packed_git_run_once = 0;
1285 prepare_packed_git();
1288 static void mark_bad_packed_object(struct packed_git *p,
1289 const unsigned char *sha1)
1292 for (i = 0; i < p->num_bad_objects; i++)
1293 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1295 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1296 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1297 p->num_bad_objects++;
1300 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1302 struct packed_git *p;
1305 for (p = packed_git; p; p = p->next)
1306 for (i = 0; i < p->num_bad_objects; i++)
1307 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1313 * With an in-core object data in "map", rehash it to make sure the
1314 * object name actually matches "sha1" to detect object corruption.
1315 * With "map" == NULL, try reading the object named with "sha1" using
1316 * the streaming interface and rehash it to do the same.
1318 int check_sha1_signature(const unsigned char *sha1, void *map,
1319 unsigned long size, const char *type)
1321 unsigned char real_sha1[20];
1322 enum object_type obj_type;
1323 struct git_istream *st;
1329 hash_sha1_file(map, size, type, real_sha1);
1330 return hashcmp(sha1, real_sha1) ? -1 : 0;
1333 st = open_istream(sha1, &obj_type, &size, NULL);
1337 /* Generate the header */
1338 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1342 git_SHA1_Update(&c, hdr, hdrlen);
1344 char buf[1024 * 16];
1345 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1353 git_SHA1_Update(&c, buf, readlen);
1355 git_SHA1_Final(real_sha1, &c);
1357 return hashcmp(sha1, real_sha1) ? -1 : 0;
1360 static int git_open_noatime(const char *name)
1362 static int sha1_file_open_flag = O_NOATIME;
1365 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1369 /* Might the failure be due to O_NOATIME? */
1370 if (errno != ENOENT && sha1_file_open_flag) {
1371 sha1_file_open_flag = 0;
1379 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1381 char *name = sha1_file_name(sha1);
1382 struct alternate_object_database *alt;
1384 if (!lstat(name, st))
1389 for (alt = alt_odb_list; alt; alt = alt->next) {
1391 fill_sha1_path(name, sha1);
1392 if (!lstat(alt->base, st))
1399 static int open_sha1_file(const unsigned char *sha1)
1402 char *name = sha1_file_name(sha1);
1403 struct alternate_object_database *alt;
1405 fd = git_open_noatime(name);
1411 for (alt = alt_odb_list; alt; alt = alt->next) {
1413 fill_sha1_path(name, sha1);
1414 fd = git_open_noatime(alt->base);
1421 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1426 fd = open_sha1_file(sha1);
1431 if (!fstat(fd, &st)) {
1432 *size = xsize_t(st.st_size);
1434 /* mmap() is forbidden on empty files */
1435 error("object file %s is empty", sha1_file_name(sha1));
1438 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1445 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1446 unsigned long len, enum object_type *type, unsigned long *sizep)
1449 unsigned long size, c;
1450 unsigned long used = 0;
1453 *type = (c >> 4) & 7;
1457 if (len <= used || bitsizeof(long) <= shift) {
1458 error("bad object header");
1463 size += (c & 0x7f) << shift;
1470 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1472 /* Get the data stream */
1473 memset(stream, 0, sizeof(*stream));
1474 stream->next_in = map;
1475 stream->avail_in = mapsize;
1476 stream->next_out = buffer;
1477 stream->avail_out = bufsiz;
1479 git_inflate_init(stream);
1480 return git_inflate(stream, 0);
1483 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1485 int bytes = strlen(buffer) + 1;
1486 unsigned char *buf = xmallocz(size);
1490 n = stream->total_out - bytes;
1493 memcpy(buf, (char *) buffer + bytes, n);
1495 if (bytes <= size) {
1497 * The above condition must be (bytes <= size), not
1498 * (bytes < size). In other words, even though we
1499 * expect no more output and set avail_out to zero,
1500 * the input zlib stream may have bytes that express
1501 * "this concludes the stream", and we *do* want to
1504 * Otherwise we would not be able to test that we
1505 * consumed all the input to reach the expected size;
1506 * we also want to check that zlib tells us that all
1507 * went well with status == Z_STREAM_END at the end.
1509 stream->next_out = buf + bytes;
1510 stream->avail_out = size - bytes;
1511 while (status == Z_OK)
1512 status = git_inflate(stream, Z_FINISH);
1514 if (status == Z_STREAM_END && !stream->avail_in) {
1515 git_inflate_end(stream);
1520 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1521 else if (stream->avail_in)
1522 error("garbage at end of loose object '%s'",
1529 * We used to just use "sscanf()", but that's actually way
1530 * too permissive for what we want to check. So do an anal
1531 * object header parse by hand.
1533 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1540 * The type can be at most ten bytes (including the
1541 * terminating '\0' that we add), and is followed by
1550 if (i >= sizeof(type))
1556 * The length must follow immediately, and be in canonical
1557 * decimal format (ie "010" is not valid).
1559 size = *hdr++ - '0';
1564 unsigned long c = *hdr - '0';
1568 size = size * 10 + c;
1574 * The length must be followed by a zero byte
1576 return *hdr ? -1 : type_from_string(type);
1579 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1585 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1586 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1589 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1592 unsigned long get_size_from_delta(struct packed_git *p,
1593 struct pack_window **w_curs,
1596 const unsigned char *data;
1597 unsigned char delta_head[20], *in;
1601 memset(&stream, 0, sizeof(stream));
1602 stream.next_out = delta_head;
1603 stream.avail_out = sizeof(delta_head);
1605 git_inflate_init(&stream);
1607 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1608 stream.next_in = in;
1609 st = git_inflate(&stream, Z_FINISH);
1610 curpos += stream.next_in - in;
1611 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1612 stream.total_out < sizeof(delta_head));
1613 git_inflate_end(&stream);
1614 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1615 error("delta data unpack-initial failed");
1619 /* Examine the initial part of the delta to figure out
1624 /* ignore base size */
1625 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1627 /* Read the result size */
1628 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1631 static off_t get_delta_base(struct packed_git *p,
1632 struct pack_window **w_curs,
1634 enum object_type type,
1635 off_t delta_obj_offset)
1637 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1640 /* use_pack() assured us we have [base_info, base_info + 20)
1641 * as a range that we can look at without walking off the
1642 * end of the mapped window. Its actually the hash size
1643 * that is assured. An OFS_DELTA longer than the hash size
1644 * is stupid, as then a REF_DELTA would be smaller to store.
1646 if (type == OBJ_OFS_DELTA) {
1648 unsigned char c = base_info[used++];
1649 base_offset = c & 127;
1652 if (!base_offset || MSB(base_offset, 7))
1653 return 0; /* overflow */
1654 c = base_info[used++];
1655 base_offset = (base_offset << 7) + (c & 127);
1657 base_offset = delta_obj_offset - base_offset;
1658 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1659 return 0; /* out of bound */
1661 } else if (type == OBJ_REF_DELTA) {
1662 /* The base entry _must_ be in the same pack */
1663 base_offset = find_pack_entry_one(base_info, p);
1666 die("I am totally screwed");
1671 * Like get_delta_base above, but we return the sha1 instead of the pack
1672 * offset. This means it is cheaper for REF deltas (we do not have to do
1673 * the final object lookup), but more expensive for OFS deltas (we
1674 * have to load the revidx to convert the offset back into a sha1).
1676 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1677 struct pack_window **w_curs,
1679 enum object_type type,
1680 off_t delta_obj_offset)
1682 if (type == OBJ_REF_DELTA) {
1683 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1685 } else if (type == OBJ_OFS_DELTA) {
1686 struct revindex_entry *revidx;
1687 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1688 type, delta_obj_offset);
1693 revidx = find_pack_revindex(p, base_offset);
1697 return nth_packed_object_sha1(p, revidx->nr);
1702 int unpack_object_header(struct packed_git *p,
1703 struct pack_window **w_curs,
1705 unsigned long *sizep)
1707 unsigned char *base;
1710 enum object_type type;
1712 /* use_pack() assures us we have [base, base + 20) available
1713 * as a range that we can look at. (Its actually the hash
1714 * size that is assured.) With our object header encoding
1715 * the maximum deflated object size is 2^137, which is just
1716 * insane, so we know won't exceed what we have been given.
1718 base = use_pack(p, w_curs, *curpos, &left);
1719 used = unpack_object_header_buffer(base, left, &type, sizep);
1728 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1731 struct revindex_entry *revidx;
1732 const unsigned char *sha1;
1733 revidx = find_pack_revindex(p, obj_offset);
1736 sha1 = nth_packed_object_sha1(p, revidx->nr);
1737 mark_bad_packed_object(p, sha1);
1738 type = sha1_object_info(sha1, NULL);
1739 if (type <= OBJ_NONE)
1744 #define POI_STACK_PREALLOC 64
1746 static enum object_type packed_to_object_type(struct packed_git *p,
1748 enum object_type type,
1749 struct pack_window **w_curs,
1752 off_t small_poi_stack[POI_STACK_PREALLOC];
1753 off_t *poi_stack = small_poi_stack;
1754 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1756 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1759 /* Push the object we're going to leave behind */
1760 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1761 poi_stack_alloc = alloc_nr(poi_stack_nr);
1762 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1763 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1765 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1767 poi_stack[poi_stack_nr++] = obj_offset;
1768 /* If parsing the base offset fails, just unwind */
1769 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1772 curpos = obj_offset = base_offset;
1773 type = unpack_object_header(p, w_curs, &curpos, &size);
1774 if (type <= OBJ_NONE) {
1775 /* If getting the base itself fails, we first
1776 * retry the base, otherwise unwind */
1777 type = retry_bad_packed_offset(p, base_offset);
1778 if (type > OBJ_NONE)
1792 error("unknown object type %i at offset %"PRIuMAX" in %s",
1793 type, (uintmax_t)obj_offset, p->pack_name);
1798 if (poi_stack != small_poi_stack)
1803 while (poi_stack_nr) {
1804 obj_offset = poi_stack[--poi_stack_nr];
1805 type = retry_bad_packed_offset(p, obj_offset);
1806 if (type > OBJ_NONE)
1813 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1814 struct object_info *oi)
1816 struct pack_window *w_curs = NULL;
1818 off_t curpos = obj_offset;
1819 enum object_type type;
1822 * We always get the representation type, but only convert it to
1823 * a "real" type later if the caller is interested.
1825 type = unpack_object_header(p, &w_curs, &curpos, &size);
1828 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1829 off_t tmp_pos = curpos;
1830 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1836 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1837 if (*oi->sizep == 0) {
1846 if (oi->disk_sizep) {
1847 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1848 *oi->disk_sizep = revidx[1].offset - obj_offset;
1852 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1853 if (*oi->typep < 0) {
1859 if (oi->delta_base_sha1) {
1860 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1861 const unsigned char *base;
1863 base = get_delta_base_sha1(p, &w_curs, curpos,
1870 hashcpy(oi->delta_base_sha1, base);
1872 hashclr(oi->delta_base_sha1);
1876 unuse_pack(&w_curs);
1880 static void *unpack_compressed_entry(struct packed_git *p,
1881 struct pack_window **w_curs,
1887 unsigned char *buffer, *in;
1889 buffer = xmallocz(size);
1890 memset(&stream, 0, sizeof(stream));
1891 stream.next_out = buffer;
1892 stream.avail_out = size + 1;
1894 git_inflate_init(&stream);
1896 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1897 stream.next_in = in;
1898 st = git_inflate(&stream, Z_FINISH);
1899 if (!stream.avail_out)
1900 break; /* the payload is larger than it should be */
1901 curpos += stream.next_in - in;
1902 } while (st == Z_OK || st == Z_BUF_ERROR);
1903 git_inflate_end(&stream);
1904 if ((st != Z_STREAM_END) || stream.total_out != size) {
1912 #define MAX_DELTA_CACHE (256)
1914 static size_t delta_base_cached;
1916 static struct delta_base_cache_lru_list {
1917 struct delta_base_cache_lru_list *prev;
1918 struct delta_base_cache_lru_list *next;
1919 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1921 static struct delta_base_cache_entry {
1922 struct delta_base_cache_lru_list lru;
1924 struct packed_git *p;
1927 enum object_type type;
1928 } delta_base_cache[MAX_DELTA_CACHE];
1930 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1934 hash = (unsigned long)p + (unsigned long)base_offset;
1935 hash += (hash >> 8) + (hash >> 16);
1936 return hash % MAX_DELTA_CACHE;
1939 static struct delta_base_cache_entry *
1940 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1942 unsigned long hash = pack_entry_hash(p, base_offset);
1943 return delta_base_cache + hash;
1946 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1947 struct packed_git *p, off_t base_offset)
1949 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1952 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1954 struct delta_base_cache_entry *ent;
1955 ent = get_delta_base_cache_entry(p, base_offset);
1956 return eq_delta_base_cache_entry(ent, p, base_offset);
1959 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1962 ent->lru.next->prev = ent->lru.prev;
1963 ent->lru.prev->next = ent->lru.next;
1964 delta_base_cached -= ent->size;
1967 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1968 unsigned long *base_size, enum object_type *type, int keep_cache)
1970 struct delta_base_cache_entry *ent;
1973 ent = get_delta_base_cache_entry(p, base_offset);
1975 if (!eq_delta_base_cache_entry(ent, p, base_offset))
1976 return unpack_entry(p, base_offset, type, base_size);
1981 clear_delta_base_cache_entry(ent);
1983 ret = xmemdupz(ent->data, ent->size);
1985 *base_size = ent->size;
1989 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1994 ent->lru.next->prev = ent->lru.prev;
1995 ent->lru.prev->next = ent->lru.next;
1996 delta_base_cached -= ent->size;
2000 void clear_delta_base_cache(void)
2003 for (p = 0; p < MAX_DELTA_CACHE; p++)
2004 release_delta_base_cache(&delta_base_cache[p]);
2007 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2008 void *base, unsigned long base_size, enum object_type type)
2010 unsigned long hash = pack_entry_hash(p, base_offset);
2011 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2012 struct delta_base_cache_lru_list *lru;
2014 release_delta_base_cache(ent);
2015 delta_base_cached += base_size;
2017 for (lru = delta_base_cache_lru.next;
2018 delta_base_cached > delta_base_cache_limit
2019 && lru != &delta_base_cache_lru;
2021 struct delta_base_cache_entry *f = (void *)lru;
2022 if (f->type == OBJ_BLOB)
2023 release_delta_base_cache(f);
2025 for (lru = delta_base_cache_lru.next;
2026 delta_base_cached > delta_base_cache_limit
2027 && lru != &delta_base_cache_lru;
2029 struct delta_base_cache_entry *f = (void *)lru;
2030 release_delta_base_cache(f);
2034 ent->base_offset = base_offset;
2037 ent->size = base_size;
2038 ent->lru.next = &delta_base_cache_lru;
2039 ent->lru.prev = delta_base_cache_lru.prev;
2040 delta_base_cache_lru.prev->next = &ent->lru;
2041 delta_base_cache_lru.prev = &ent->lru;
2044 static void *read_object(const unsigned char *sha1, enum object_type *type,
2045 unsigned long *size);
2047 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2049 static FILE *log_file;
2051 if (!log_pack_access)
2052 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2053 if (!log_pack_access)
2054 log_pack_access = no_log_pack_access;
2055 if (log_pack_access == no_log_pack_access)
2059 log_file = fopen(log_pack_access, "w");
2061 error("cannot open pack access log '%s' for writing: %s",
2062 log_pack_access, strerror(errno));
2063 log_pack_access = no_log_pack_access;
2067 fprintf(log_file, "%s %"PRIuMAX"\n",
2068 p->pack_name, (uintmax_t)obj_offset);
2072 int do_check_packed_object_crc;
2074 #define UNPACK_ENTRY_STACK_PREALLOC 64
2075 struct unpack_entry_stack_ent {
2081 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2082 enum object_type *final_type, unsigned long *final_size)
2084 struct pack_window *w_curs = NULL;
2085 off_t curpos = obj_offset;
2088 enum object_type type;
2089 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2090 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2091 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2092 int base_from_cache = 0;
2094 if (log_pack_access != no_log_pack_access)
2095 write_pack_access_log(p, obj_offset);
2097 /* PHASE 1: drill down to the innermost base object */
2101 struct delta_base_cache_entry *ent;
2103 ent = get_delta_base_cache_entry(p, curpos);
2104 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2108 clear_delta_base_cache_entry(ent);
2109 base_from_cache = 1;
2113 if (do_check_packed_object_crc && p->index_version > 1) {
2114 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2115 unsigned long len = revidx[1].offset - obj_offset;
2116 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2117 const unsigned char *sha1 =
2118 nth_packed_object_sha1(p, revidx->nr);
2119 error("bad packed object CRC for %s",
2121 mark_bad_packed_object(p, sha1);
2122 unuse_pack(&w_curs);
2127 type = unpack_object_header(p, &w_curs, &curpos, &size);
2128 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2131 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2133 error("failed to validate delta base reference "
2134 "at offset %"PRIuMAX" from %s",
2135 (uintmax_t)curpos, p->pack_name);
2136 /* bail to phase 2, in hopes of recovery */
2141 /* push object, proceed to base */
2142 if (delta_stack_nr >= delta_stack_alloc
2143 && delta_stack == small_delta_stack) {
2144 delta_stack_alloc = alloc_nr(delta_stack_nr);
2145 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2146 memcpy(delta_stack, small_delta_stack,
2147 sizeof(*delta_stack)*delta_stack_nr);
2149 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2151 i = delta_stack_nr++;
2152 delta_stack[i].obj_offset = obj_offset;
2153 delta_stack[i].curpos = curpos;
2154 delta_stack[i].size = size;
2156 curpos = obj_offset = base_offset;
2159 /* PHASE 2: handle the base */
2164 die("BUG in unpack_entry: left loop at a valid delta");
2170 if (!base_from_cache)
2171 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2175 error("unknown object type %i at offset %"PRIuMAX" in %s",
2176 type, (uintmax_t)obj_offset, p->pack_name);
2179 /* PHASE 3: apply deltas in order */
2182 * 'data' holds the base data, or NULL if there was corruption
2184 while (delta_stack_nr) {
2187 unsigned long delta_size, base_size = size;
2193 add_delta_base_cache(p, obj_offset, base, base_size, type);
2197 * We're probably in deep shit, but let's try to fetch
2198 * the required base anyway from another pack or loose.
2199 * This is costly but should happen only in the presence
2200 * of a corrupted pack, and is better than failing outright.
2202 struct revindex_entry *revidx;
2203 const unsigned char *base_sha1;
2204 revidx = find_pack_revindex(p, obj_offset);
2206 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2207 error("failed to read delta base object %s"
2208 " at offset %"PRIuMAX" from %s",
2209 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2211 mark_bad_packed_object(p, base_sha1);
2212 base = read_object(base_sha1, &type, &base_size);
2216 i = --delta_stack_nr;
2217 obj_offset = delta_stack[i].obj_offset;
2218 curpos = delta_stack[i].curpos;
2219 delta_size = delta_stack[i].size;
2224 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2227 error("failed to unpack compressed delta "
2228 "at offset %"PRIuMAX" from %s",
2229 (uintmax_t)curpos, p->pack_name);
2234 data = patch_delta(base, base_size,
2235 delta_data, delta_size,
2239 * We could not apply the delta; warn the user, but keep going.
2240 * Our failure will be noticed either in the next iteration of
2241 * the loop, or if this is the final delta, in the caller when
2242 * we return NULL. Those code paths will take care of making
2243 * a more explicit warning and retrying with another copy of
2247 error("failed to apply delta");
2255 unuse_pack(&w_curs);
2259 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2262 const unsigned char *index = p->index_data;
2264 if (open_pack_index(p))
2266 index = p->index_data;
2268 if (n >= p->num_objects)
2271 if (p->index_version == 1) {
2272 return index + 24 * n + 4;
2275 return index + 20 * n;
2279 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2281 const unsigned char *index = p->index_data;
2283 if (p->index_version == 1) {
2284 return ntohl(*((uint32_t *)(index + 24 * n)));
2287 index += 8 + p->num_objects * (20 + 4);
2288 off = ntohl(*((uint32_t *)(index + 4 * n)));
2289 if (!(off & 0x80000000))
2291 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2292 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2293 ntohl(*((uint32_t *)(index + 4)));
2297 off_t find_pack_entry_one(const unsigned char *sha1,
2298 struct packed_git *p)
2300 const uint32_t *level1_ofs = p->index_data;
2301 const unsigned char *index = p->index_data;
2302 unsigned hi, lo, stride;
2303 static int use_lookup = -1;
2304 static int debug_lookup = -1;
2306 if (debug_lookup < 0)
2307 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2310 if (open_pack_index(p))
2312 level1_ofs = p->index_data;
2313 index = p->index_data;
2315 if (p->index_version > 1) {
2320 hi = ntohl(level1_ofs[*sha1]);
2321 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2322 if (p->index_version > 1) {
2330 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2331 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2334 use_lookup = !!getenv("GIT_USE_LOOKUP");
2336 int pos = sha1_entry_pos(index, stride, 0,
2337 lo, hi, p->num_objects, sha1);
2340 return nth_packed_object_offset(p, pos);
2344 unsigned mi = (lo + hi) / 2;
2345 int cmp = hashcmp(index + mi * stride, sha1);
2348 printf("lo %u hi %u rg %u mi %u\n",
2349 lo, hi, hi - lo, mi);
2351 return nth_packed_object_offset(p, mi);
2360 int is_pack_valid(struct packed_git *p)
2362 /* An already open pack is known to be valid. */
2363 if (p->pack_fd != -1)
2366 /* If the pack has one window completely covering the
2367 * file size, the pack is known to be valid even if
2368 * the descriptor is not currently open.
2371 struct pack_window *w = p->windows;
2373 if (!w->offset && w->len == p->pack_size)
2377 /* Force the pack to open to prove its valid. */
2378 return !open_packed_git(p);
2381 static int fill_pack_entry(const unsigned char *sha1,
2382 struct pack_entry *e,
2383 struct packed_git *p)
2387 if (p->num_bad_objects) {
2389 for (i = 0; i < p->num_bad_objects; i++)
2390 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2394 offset = find_pack_entry_one(sha1, p);
2399 * We are about to tell the caller where they can locate the
2400 * requested object. We better make sure the packfile is
2401 * still here and can be accessed before supplying that
2402 * answer, as it may have been deleted since the index was
2405 if (!is_pack_valid(p)) {
2406 warning("packfile %s cannot be accessed", p->pack_name);
2411 hashcpy(e->sha1, sha1);
2415 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2417 struct packed_git *p;
2419 prepare_packed_git();
2423 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2426 for (p = packed_git; p; p = p->next) {
2427 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2430 last_found_pack = p;
2436 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2437 struct packed_git *packs)
2439 struct packed_git *p;
2441 for (p = packs; p; p = p->next) {
2442 if (find_pack_entry_one(sha1, p))
2449 static int sha1_loose_object_info(const unsigned char *sha1,
2450 struct object_info *oi)
2453 unsigned long mapsize, size;
2458 if (oi->delta_base_sha1)
2459 hashclr(oi->delta_base_sha1);
2462 * If we don't care about type or size, then we don't
2463 * need to look inside the object at all. Note that we
2464 * do not optimize out the stat call, even if the
2465 * caller doesn't care about the disk-size, since our
2466 * return value implicitly indicates whether the
2467 * object even exists.
2469 if (!oi->typep && !oi->sizep) {
2471 if (stat_sha1_file(sha1, &st) < 0)
2474 *oi->disk_sizep = st.st_size;
2478 map = map_sha1_file(sha1, &mapsize);
2482 *oi->disk_sizep = mapsize;
2483 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2484 status = error("unable to unpack %s header",
2486 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2487 status = error("unable to parse %s header", sha1_to_hex(sha1));
2490 git_inflate_end(&stream);
2491 munmap(map, mapsize);
2493 *oi->typep = status;
2497 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2499 struct cached_object *co;
2500 struct pack_entry e;
2503 co = find_cached_object(sha1);
2506 *(oi->typep) = co->type;
2508 *(oi->sizep) = co->size;
2510 *(oi->disk_sizep) = 0;
2511 if (oi->delta_base_sha1)
2512 hashclr(oi->delta_base_sha1);
2513 oi->whence = OI_CACHED;
2517 if (!find_pack_entry(sha1, &e)) {
2518 /* Most likely it's a loose object. */
2519 if (!sha1_loose_object_info(sha1, oi)) {
2520 oi->whence = OI_LOOSE;
2524 /* Not a loose object; someone else may have just packed it. */
2525 reprepare_packed_git();
2526 if (!find_pack_entry(sha1, &e))
2530 rtype = packed_object_info(e.p, e.offset, oi);
2532 mark_bad_packed_object(e.p, sha1);
2533 return sha1_object_info_extended(sha1, oi);
2534 } else if (in_delta_base_cache(e.p, e.offset)) {
2535 oi->whence = OI_DBCACHED;
2537 oi->whence = OI_PACKED;
2538 oi->u.packed.offset = e.offset;
2539 oi->u.packed.pack = e.p;
2540 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2541 rtype == OBJ_OFS_DELTA);
2547 /* returns enum object_type or negative */
2548 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2550 enum object_type type;
2551 struct object_info oi = {NULL};
2555 if (sha1_object_info_extended(sha1, &oi) < 0)
2560 static void *read_packed_sha1(const unsigned char *sha1,
2561 enum object_type *type, unsigned long *size)
2563 struct pack_entry e;
2566 if (!find_pack_entry(sha1, &e))
2568 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2571 * We're probably in deep shit, but let's try to fetch
2572 * the required object anyway from another pack or loose.
2573 * This should happen only in the presence of a corrupted
2574 * pack, and is better than failing outright.
2576 error("failed to read object %s at offset %"PRIuMAX" from %s",
2577 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2578 mark_bad_packed_object(e.p, sha1);
2579 data = read_object(sha1, type, size);
2584 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2585 unsigned char *sha1)
2587 struct cached_object *co;
2589 hash_sha1_file(buf, len, typename(type), sha1);
2590 if (has_sha1_file(sha1) || find_cached_object(sha1))
2592 if (cached_object_alloc <= cached_object_nr) {
2593 cached_object_alloc = alloc_nr(cached_object_alloc);
2594 cached_objects = xrealloc(cached_objects,
2595 sizeof(*cached_objects) *
2596 cached_object_alloc);
2598 co = &cached_objects[cached_object_nr++];
2601 co->buf = xmalloc(len);
2602 memcpy(co->buf, buf, len);
2603 hashcpy(co->sha1, sha1);
2607 static void *read_object(const unsigned char *sha1, enum object_type *type,
2608 unsigned long *size)
2610 unsigned long mapsize;
2612 struct cached_object *co;
2614 co = find_cached_object(sha1);
2618 return xmemdupz(co->buf, co->size);
2621 buf = read_packed_sha1(sha1, type, size);
2624 map = map_sha1_file(sha1, &mapsize);
2626 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2627 munmap(map, mapsize);
2630 reprepare_packed_git();
2631 return read_packed_sha1(sha1, type, size);
2635 * This function dies on corrupt objects; the callers who want to
2636 * deal with them should arrange to call read_object() and give error
2637 * messages themselves.
2639 void *read_sha1_file_extended(const unsigned char *sha1,
2640 enum object_type *type,
2641 unsigned long *size,
2646 const struct packed_git *p;
2647 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2648 ? lookup_replace_object(sha1) : sha1;
2651 data = read_object(repl, type, size);
2655 if (errno && errno != ENOENT)
2656 die_errno("failed to read object %s", sha1_to_hex(sha1));
2658 /* die if we replaced an object with one that does not exist */
2660 die("replacement %s not found for %s",
2661 sha1_to_hex(repl), sha1_to_hex(sha1));
2663 if (has_loose_object(repl)) {
2664 path = sha1_file_name(sha1);
2665 die("loose object %s (stored in %s) is corrupt",
2666 sha1_to_hex(repl), path);
2669 if ((p = has_packed_and_bad(repl)) != NULL)
2670 die("packed object %s (stored in %s) is corrupt",
2671 sha1_to_hex(repl), p->pack_name);
2676 void *read_object_with_reference(const unsigned char *sha1,
2677 const char *required_type_name,
2678 unsigned long *size,
2679 unsigned char *actual_sha1_return)
2681 enum object_type type, required_type;
2683 unsigned long isize;
2684 unsigned char actual_sha1[20];
2686 required_type = type_from_string(required_type_name);
2687 hashcpy(actual_sha1, sha1);
2689 int ref_length = -1;
2690 const char *ref_type = NULL;
2692 buffer = read_sha1_file(actual_sha1, &type, &isize);
2695 if (type == required_type) {
2697 if (actual_sha1_return)
2698 hashcpy(actual_sha1_return, actual_sha1);
2701 /* Handle references */
2702 else if (type == OBJ_COMMIT)
2704 else if (type == OBJ_TAG)
2705 ref_type = "object ";
2710 ref_length = strlen(ref_type);
2712 if (ref_length + 40 > isize ||
2713 memcmp(buffer, ref_type, ref_length) ||
2714 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2719 /* Now we have the ID of the referred-to object in
2720 * actual_sha1. Check again. */
2724 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2725 const char *type, unsigned char *sha1,
2726 char *hdr, int *hdrlen)
2730 /* Generate the header */
2731 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2735 git_SHA1_Update(&c, hdr, *hdrlen);
2736 git_SHA1_Update(&c, buf, len);
2737 git_SHA1_Final(sha1, &c);
2741 * Move the just written object into its final resting place.
2742 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2743 * "moving" is only a part of what it does, when no patch between
2744 * master to pu changes the call sites of this function.
2746 int move_temp_to_file(const char *tmpfile, const char *filename)
2750 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2752 else if (link(tmpfile, filename))
2756 * Coda hack - coda doesn't like cross-directory links,
2757 * so we fall back to a rename, which will mean that it
2758 * won't be able to check collisions, but that's not a
2761 * The same holds for FAT formatted media.
2763 * When this succeeds, we just return. We have nothing
2766 if (ret && ret != EEXIST) {
2768 if (!rename(tmpfile, filename))
2772 unlink_or_warn(tmpfile);
2774 if (ret != EEXIST) {
2775 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2777 /* FIXME!!! Collision check here ? */
2781 if (adjust_shared_perm(filename))
2782 return error("unable to set permission to '%s'", filename);
2786 static int write_buffer(int fd, const void *buf, size_t len)
2788 if (write_in_full(fd, buf, len) < 0)
2789 return error("file write error (%s)", strerror(errno));
2793 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2794 unsigned char *sha1)
2798 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2802 /* Finalize a file on disk, and close it. */
2803 static void close_sha1_file(int fd)
2805 if (fsync_object_files)
2806 fsync_or_die(fd, "sha1 file");
2808 die_errno("error when closing sha1 file");
2811 /* Size of directory component, including the ending '/' */
2812 static inline int directory_size(const char *filename)
2814 const char *s = strrchr(filename, '/');
2817 return s - filename + 1;
2821 * This creates a temporary file in the same directory as the final
2824 * We want to avoid cross-directory filename renames, because those
2825 * can have problems on various filesystems (FAT, NFS, Coda).
2827 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2829 int fd, dirlen = directory_size(filename);
2831 if (dirlen + 20 > bufsiz) {
2832 errno = ENAMETOOLONG;
2835 memcpy(buffer, filename, dirlen);
2836 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2837 fd = git_mkstemp_mode(buffer, 0444);
2838 if (fd < 0 && dirlen && errno == ENOENT) {
2839 /* Make sure the directory exists */
2840 memcpy(buffer, filename, dirlen);
2841 buffer[dirlen-1] = 0;
2842 if (mkdir(buffer, 0777) && errno != EEXIST)
2844 if (adjust_shared_perm(buffer))
2848 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2849 fd = git_mkstemp_mode(buffer, 0444);
2854 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2855 const void *buf, unsigned long len, time_t mtime)
2858 unsigned char compressed[4096];
2861 unsigned char parano_sha1[20];
2863 static char tmp_file[PATH_MAX];
2865 filename = sha1_file_name(sha1);
2866 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2868 if (errno == EACCES)
2869 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2871 return error("unable to create temporary file: %s", strerror(errno));
2875 memset(&stream, 0, sizeof(stream));
2876 git_deflate_init(&stream, zlib_compression_level);
2877 stream.next_out = compressed;
2878 stream.avail_out = sizeof(compressed);
2881 /* First header.. */
2882 stream.next_in = (unsigned char *)hdr;
2883 stream.avail_in = hdrlen;
2884 while (git_deflate(&stream, 0) == Z_OK)
2886 git_SHA1_Update(&c, hdr, hdrlen);
2888 /* Then the data itself.. */
2889 stream.next_in = (void *)buf;
2890 stream.avail_in = len;
2892 unsigned char *in0 = stream.next_in;
2893 ret = git_deflate(&stream, Z_FINISH);
2894 git_SHA1_Update(&c, in0, stream.next_in - in0);
2895 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2896 die("unable to write sha1 file");
2897 stream.next_out = compressed;
2898 stream.avail_out = sizeof(compressed);
2899 } while (ret == Z_OK);
2901 if (ret != Z_STREAM_END)
2902 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2903 ret = git_deflate_end_gently(&stream);
2905 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2906 git_SHA1_Final(parano_sha1, &c);
2907 if (hashcmp(sha1, parano_sha1) != 0)
2908 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2910 close_sha1_file(fd);
2915 utb.modtime = mtime;
2916 if (utime(tmp_file, &utb) < 0)
2917 warning("failed utime() on %s: %s",
2918 tmp_file, strerror(errno));
2921 return move_temp_to_file(tmp_file, filename);
2924 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2926 unsigned char sha1[20];
2930 /* Normally if we have it in the pack then we do not bother writing
2931 * it out into .git/objects/??/?{38} file.
2933 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2935 hashcpy(returnsha1, sha1);
2936 if (has_sha1_file(sha1))
2938 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2941 int force_object_loose(const unsigned char *sha1, time_t mtime)
2945 enum object_type type;
2950 if (has_loose_object(sha1))
2952 buf = read_packed_sha1(sha1, &type, &len);
2954 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2955 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2956 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2962 int has_pack_index(const unsigned char *sha1)
2965 if (stat(sha1_pack_index_name(sha1), &st))
2970 int has_sha1_pack(const unsigned char *sha1)
2972 struct pack_entry e;
2973 return find_pack_entry(sha1, &e);
2976 int has_sha1_file(const unsigned char *sha1)
2978 struct pack_entry e;
2980 if (find_pack_entry(sha1, &e))
2982 if (has_loose_object(sha1))
2984 reprepare_packed_git();
2985 return find_pack_entry(sha1, &e);
2988 static void check_tree(const void *buf, size_t size)
2990 struct tree_desc desc;
2991 struct name_entry entry;
2993 init_tree_desc(&desc, buf, size);
2994 while (tree_entry(&desc, &entry))
2996 * tree_entry() will die() on malformed entries */
3000 static void check_commit(const void *buf, size_t size)
3003 memset(&c, 0, sizeof(c));
3004 if (parse_commit_buffer(&c, buf, size))
3005 die("corrupt commit");
3008 static void check_tag(const void *buf, size_t size)
3011 memset(&t, 0, sizeof(t));
3012 if (parse_tag_buffer(&t, buf, size))
3016 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3017 enum object_type type,
3018 const char *path, unsigned flags)
3020 int ret, re_allocated = 0;
3021 int write_object = flags & HASH_WRITE_OBJECT;
3027 * Convert blobs to git internal format
3029 if ((type == OBJ_BLOB) && path) {
3030 struct strbuf nbuf = STRBUF_INIT;
3031 if (convert_to_git(path, buf, size, &nbuf,
3032 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3033 buf = strbuf_detach(&nbuf, &size);
3037 if (flags & HASH_FORMAT_CHECK) {
3038 if (type == OBJ_TREE)
3039 check_tree(buf, size);
3040 if (type == OBJ_COMMIT)
3041 check_commit(buf, size);
3042 if (type == OBJ_TAG)
3043 check_tag(buf, size);
3047 ret = write_sha1_file(buf, size, typename(type), sha1);
3049 ret = hash_sha1_file(buf, size, typename(type), sha1);
3055 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3056 const char *path, unsigned flags)
3058 struct strbuf sbuf = STRBUF_INIT;
3061 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3062 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3065 strbuf_release(&sbuf);
3069 #define SMALL_FILE_SIZE (32*1024)
3071 static int index_core(unsigned char *sha1, int fd, size_t size,
3072 enum object_type type, const char *path,
3078 ret = index_mem(sha1, NULL, size, type, path, flags);
3079 } else if (size <= SMALL_FILE_SIZE) {
3080 char *buf = xmalloc(size);
3081 if (size == read_in_full(fd, buf, size))
3082 ret = index_mem(sha1, buf, size, type, path, flags);
3084 ret = error("short read %s", strerror(errno));
3087 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3088 ret = index_mem(sha1, buf, size, type, path, flags);
3095 * This creates one packfile per large blob unless bulk-checkin
3096 * machinery is "plugged".
3098 * This also bypasses the usual "convert-to-git" dance, and that is on
3099 * purpose. We could write a streaming version of the converting
3100 * functions and insert that before feeding the data to fast-import
3101 * (or equivalent in-core API described above). However, that is
3102 * somewhat complicated, as we do not know the size of the filter
3103 * result, which we need to know beforehand when writing a git object.
3104 * Since the primary motivation for trying to stream from the working
3105 * tree file and to avoid mmaping it in core is to deal with large
3106 * binary blobs, they generally do not want to get any conversion, and
3107 * callers should avoid this code path when filters are requested.
3109 static int index_stream(unsigned char *sha1, int fd, size_t size,
3110 enum object_type type, const char *path,
3113 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3116 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3117 enum object_type type, const char *path, unsigned flags)
3120 size_t size = xsize_t(st->st_size);
3122 if (!S_ISREG(st->st_mode))
3123 ret = index_pipe(sha1, fd, type, path, flags);
3124 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3125 (path && would_convert_to_git(path, NULL, 0, 0)))
3126 ret = index_core(sha1, fd, size, type, path, flags);
3128 ret = index_stream(sha1, fd, size, type, path, flags);
3133 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3136 struct strbuf sb = STRBUF_INIT;
3138 switch (st->st_mode & S_IFMT) {
3140 fd = open(path, O_RDONLY);
3142 return error("open(\"%s\"): %s", path,
3144 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3145 return error("%s: failed to insert into database",
3149 if (strbuf_readlink(&sb, path, st->st_size)) {
3150 char *errstr = strerror(errno);
3151 return error("readlink(\"%s\"): %s", path,
3154 if (!(flags & HASH_WRITE_OBJECT))
3155 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3156 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3157 return error("%s: failed to insert into database",
3159 strbuf_release(&sb);
3162 return resolve_gitlink_ref(path, "HEAD", sha1);
3164 return error("%s: unsupported file type", path);
3169 int read_pack_header(int fd, struct pack_header *header)
3171 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3172 /* "eof before pack header was fully read" */
3173 return PH_ERROR_EOF;
3175 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3176 /* "protocol error (pack signature mismatch detected)" */
3177 return PH_ERROR_PACK_SIGNATURE;
3178 if (!pack_version_ok(header->hdr_version))
3179 /* "protocol error (pack version unsupported)" */
3180 return PH_ERROR_PROTOCOL;
3184 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3186 enum object_type type = sha1_object_info(sha1, NULL);
3188 die("%s is not a valid object", sha1_to_hex(sha1));
3190 die("%s is not a valid '%s' object", sha1_to_hex(sha1),