12 static const char index_pack_usage[] =
13 "git-index-pack [-v] [-o <index-file>] [{ ---keep | --keep=<msg> }] [--strict] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";
17 struct pack_idx_entry idx;
19 unsigned int hdr_size;
20 enum object_type type;
21 enum object_type real_type;
25 unsigned char sha1[20];
35 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
36 * to memcmp() only the first 20 bytes.
38 #define UNION_BASE_SZ 20
40 #define FLAG_LINK (1u<<20)
41 #define FLAG_CHECKED (1u<<21)
45 union delta_base base;
49 static struct object_entry *objects;
50 static struct delta_entry *deltas;
51 static int nr_objects;
53 static int nr_resolved_deltas;
55 static int from_stdin;
59 static struct progress *progress;
61 /* We always read in 4kB chunks. */
62 static unsigned char input_buffer[4096];
63 static unsigned int input_offset, input_len;
64 static off_t consumed_bytes;
65 static SHA_CTX input_ctx;
66 static uint32_t input_crc32;
67 static int input_fd, output_fd, pack_fd;
69 static int mark_link(struct object *obj, int type, void *data)
74 if (type != OBJ_ANY && obj->type != type)
75 die("object type mismatch at %s", sha1_to_hex(obj->sha1));
77 obj->flags |= FLAG_LINK;
81 /* The content of each linked object must have been checked
82 or it must be already present in the object database */
83 static void check_object(struct object *obj)
88 if (!(obj->flags & FLAG_LINK))
91 if (!(obj->flags & FLAG_CHECKED)) {
93 int type = sha1_object_info(obj->sha1, &size);
94 if (type != obj->type || type <= 0)
95 die("object of unexpected type");
96 obj->flags |= FLAG_CHECKED;
101 static void check_objects(void)
105 max = get_max_object_index();
106 for (i = 0; i < max; i++)
107 check_object(get_indexed_object(i));
111 /* Discard current buffer used content. */
112 static void flush(void)
116 write_or_die(output_fd, input_buffer, input_offset);
117 SHA1_Update(&input_ctx, input_buffer, input_offset);
118 memmove(input_buffer, input_buffer + input_offset, input_len);
124 * Make sure at least "min" bytes are available in the buffer, and
125 * return the pointer to the buffer.
127 static void *fill(int min)
129 if (min <= input_len)
130 return input_buffer + input_offset;
131 if (min > sizeof(input_buffer))
132 die("cannot fill %d bytes", min);
135 ssize_t ret = xread(input_fd, input_buffer + input_len,
136 sizeof(input_buffer) - input_len);
140 die("read error on input: %s", strerror(errno));
144 display_throughput(progress, consumed_bytes + input_len);
145 } while (input_len < min);
149 static void use(int bytes)
151 if (bytes > input_len)
152 die("used more bytes than were available");
153 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
155 input_offset += bytes;
157 /* make sure off_t is sufficiently large not to wrap */
158 if (consumed_bytes > consumed_bytes + bytes)
159 die("pack too large for current definition of off_t");
160 consumed_bytes += bytes;
163 static char *open_pack_file(char *pack_name)
168 static char tmpfile[PATH_MAX];
169 snprintf(tmpfile, sizeof(tmpfile),
170 "%s/tmp_pack_XXXXXX", get_object_directory());
171 output_fd = xmkstemp(tmpfile);
172 pack_name = xstrdup(tmpfile);
174 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
176 die("unable to create %s: %s\n", pack_name, strerror(errno));
179 input_fd = open(pack_name, O_RDONLY);
181 die("cannot open packfile '%s': %s",
182 pack_name, strerror(errno));
186 SHA1_Init(&input_ctx);
190 static void parse_pack_header(void)
192 struct pack_header *hdr = fill(sizeof(struct pack_header));
194 /* Header consistency check */
195 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
196 die("pack signature mismatch");
197 if (!pack_version_ok(hdr->hdr_version))
198 die("pack version %d unsupported", ntohl(hdr->hdr_version));
200 nr_objects = ntohl(hdr->hdr_entries);
201 use(sizeof(struct pack_header));
204 static void bad_object(unsigned long offset, const char *format,
205 ...) NORETURN __attribute__((format (printf, 2, 3)));
207 static void bad_object(unsigned long offset, const char *format, ...)
212 va_start(params, format);
213 vsnprintf(buf, sizeof(buf), format, params);
215 die("pack has bad object at offset %lu: %s", offset, buf);
218 static void *unpack_entry_data(unsigned long offset, unsigned long size)
221 void *buf = xmalloc(size);
223 memset(&stream, 0, sizeof(stream));
224 stream.next_out = buf;
225 stream.avail_out = size;
226 stream.next_in = fill(1);
227 stream.avail_in = input_len;
228 inflateInit(&stream);
231 int ret = inflate(&stream, 0);
232 use(input_len - stream.avail_in);
233 if (stream.total_out == size && ret == Z_STREAM_END)
236 bad_object(offset, "inflate returned %d", ret);
237 stream.next_in = fill(1);
238 stream.avail_in = input_len;
244 static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
252 obj->idx.offset = consumed_bytes;
253 input_crc32 = crc32(0, Z_NULL, 0);
258 obj->type = (c >> 4) & 7;
265 size += (c & 0x7fUL) << shift;
272 hashcpy(delta_base->sha1, fill(20));
276 memset(delta_base, 0, sizeof(*delta_base));
280 base_offset = c & 127;
283 if (!base_offset || MSB(base_offset, 7))
284 bad_object(obj->idx.offset, "offset value overflow for delta base object");
288 base_offset = (base_offset << 7) + (c & 127);
290 delta_base->offset = obj->idx.offset - base_offset;
291 if (delta_base->offset >= obj->idx.offset)
292 bad_object(obj->idx.offset, "delta base offset is out of bound");
300 bad_object(obj->idx.offset, "unknown object type %d", obj->type);
302 obj->hdr_size = consumed_bytes - obj->idx.offset;
304 data = unpack_entry_data(obj->idx.offset, obj->size);
305 obj->idx.crc32 = input_crc32;
309 static void *get_data_from_pack(struct object_entry *obj)
311 off_t from = obj[0].idx.offset + obj[0].hdr_size;
312 unsigned long len = obj[1].idx.offset - from;
313 unsigned long rdy = 0;
314 unsigned char *src, *data;
321 ssize_t n = pread(pack_fd, data + rdy, len - rdy, from + rdy);
323 die("cannot pread pack file: %s", strerror(errno));
326 data = xmalloc(obj->size);
327 memset(&stream, 0, sizeof(stream));
328 stream.next_out = data;
329 stream.avail_out = obj->size;
330 stream.next_in = src;
331 stream.avail_in = len;
332 inflateInit(&stream);
333 while ((st = inflate(&stream, Z_FINISH)) == Z_OK);
335 if (st != Z_STREAM_END || stream.total_out != obj->size)
336 die("serious inflate inconsistency");
341 static int find_delta(const union delta_base *base)
343 int first = 0, last = nr_deltas;
345 while (first < last) {
346 int next = (first + last) / 2;
347 struct delta_entry *delta = &deltas[next];
350 cmp = memcmp(base, &delta->base, UNION_BASE_SZ);
362 static int find_delta_children(const union delta_base *base,
363 int *first_index, int *last_index)
365 int first = find_delta(base);
367 int end = nr_deltas - 1;
371 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
373 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
375 *first_index = first;
380 static void sha1_object(const void *data, unsigned long size,
381 enum object_type type, unsigned char *sha1)
383 hash_sha1_file(data, size, typename(type), sha1);
384 if (has_sha1_file(sha1)) {
386 enum object_type has_type;
387 unsigned long has_size;
388 has_data = read_sha1_file(sha1, &has_type, &has_size);
390 die("cannot read existing object %s", sha1_to_hex(sha1));
391 if (size != has_size || type != has_type ||
392 memcmp(data, has_data, size) != 0)
393 die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
397 if (type == OBJ_BLOB) {
398 struct blob *blob = lookup_blob(sha1);
400 blob->object.flags |= FLAG_CHECKED;
402 die("invalid blob object %s", sha1_to_hex(sha1));
406 void *buf = (void *) data;
409 * we do not need to free the memory here, as the
410 * buf is deleted by the caller.
412 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
414 die("invalid %s", typename(type));
415 if (fsck_object(obj, 1, fsck_error_function))
416 die("Error in object");
417 if (fsck_walk(obj, mark_link, 0))
418 die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1));
420 if (obj->type == OBJ_TREE) {
421 struct tree *item = (struct tree *) obj;
424 if (obj->type == OBJ_COMMIT) {
425 struct commit *commit = (struct commit *) obj;
426 commit->buffer = NULL;
428 obj->flags |= FLAG_CHECKED;
433 static void resolve_delta(struct object_entry *delta_obj,
434 struct base_data *base_obj, enum object_type type)
437 unsigned long delta_size;
438 union delta_base delta_base;
440 struct base_data result;
442 delta_obj->real_type = type;
443 delta_data = get_data_from_pack(delta_obj);
444 delta_size = delta_obj->size;
445 result.data = patch_delta(base_obj->data, base_obj->size,
446 delta_data, delta_size,
450 bad_object(delta_obj->idx.offset, "failed to apply delta");
451 sha1_object(result.data, result.size, type, delta_obj->idx.sha1);
452 nr_resolved_deltas++;
454 hashcpy(delta_base.sha1, delta_obj->idx.sha1);
455 if (!find_delta_children(&delta_base, &first, &last)) {
456 for (j = first; j <= last; j++) {
457 struct object_entry *child = objects + deltas[j].obj_no;
458 if (child->real_type == OBJ_REF_DELTA)
459 resolve_delta(child, &result, type);
463 memset(&delta_base, 0, sizeof(delta_base));
464 delta_base.offset = delta_obj->idx.offset;
465 if (!find_delta_children(&delta_base, &first, &last)) {
466 for (j = first; j <= last; j++) {
467 struct object_entry *child = objects + deltas[j].obj_no;
468 if (child->real_type == OBJ_OFS_DELTA)
469 resolve_delta(child, &result, type);
476 static int compare_delta_entry(const void *a, const void *b)
478 const struct delta_entry *delta_a = a;
479 const struct delta_entry *delta_b = b;
480 return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ);
483 /* Parse all objects and return the pack content SHA1 hash */
484 static void parse_pack_objects(unsigned char *sha1)
487 struct delta_entry *delta = deltas;
492 * - find locations of all objects;
493 * - calculate SHA1 of all non-delta objects;
494 * - remember base (SHA1 or offset) for all deltas.
497 progress = start_progress(
498 from_stdin ? "Receiving objects" : "Indexing objects",
500 for (i = 0; i < nr_objects; i++) {
501 struct object_entry *obj = &objects[i];
502 void *data = unpack_raw_entry(obj, &delta->base);
503 obj->real_type = obj->type;
504 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
509 sha1_object(data, obj->size, obj->type, obj->idx.sha1);
511 display_progress(progress, i+1);
513 objects[i].idx.offset = consumed_bytes;
514 stop_progress(&progress);
516 /* Check pack integrity */
518 SHA1_Final(sha1, &input_ctx);
519 if (hashcmp(fill(20), sha1))
520 die("pack is corrupted (SHA1 mismatch)");
523 /* If input_fd is a file, we should have reached its end now. */
524 if (fstat(input_fd, &st))
525 die("cannot fstat packfile: %s", strerror(errno));
526 if (S_ISREG(st.st_mode) &&
527 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
528 die("pack has junk at the end");
533 /* Sort deltas by base SHA1/offset for fast searching */
534 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
535 compare_delta_entry);
539 * - for all non-delta objects, look if it is used as a base for
541 * - if used as a base, uncompress the object and apply all deltas,
542 * recursively checking if the resulting object is used as a base
543 * for some more deltas.
546 progress = start_progress("Resolving deltas", nr_deltas);
547 for (i = 0; i < nr_objects; i++) {
548 struct object_entry *obj = &objects[i];
549 union delta_base base;
550 int j, ref, ref_first, ref_last, ofs, ofs_first, ofs_last;
551 struct base_data base_obj;
553 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA)
555 hashcpy(base.sha1, obj->idx.sha1);
556 ref = !find_delta_children(&base, &ref_first, &ref_last);
557 memset(&base, 0, sizeof(base));
558 base.offset = obj->idx.offset;
559 ofs = !find_delta_children(&base, &ofs_first, &ofs_last);
562 base_obj.data = get_data_from_pack(obj);
563 base_obj.size = obj->size;
566 for (j = ref_first; j <= ref_last; j++) {
567 struct object_entry *child = objects + deltas[j].obj_no;
568 if (child->real_type == OBJ_REF_DELTA)
569 resolve_delta(child, &base_obj, obj->type);
572 for (j = ofs_first; j <= ofs_last; j++) {
573 struct object_entry *child = objects + deltas[j].obj_no;
574 if (child->real_type == OBJ_OFS_DELTA)
575 resolve_delta(child, &base_obj, obj->type);
578 display_progress(progress, nr_resolved_deltas);
582 static int write_compressed(int fd, void *in, unsigned int size, uint32_t *obj_crc)
585 unsigned long maxsize;
588 memset(&stream, 0, sizeof(stream));
589 deflateInit(&stream, zlib_compression_level);
590 maxsize = deflateBound(&stream, size);
591 out = xmalloc(maxsize);
595 stream.avail_in = size;
596 stream.next_out = out;
597 stream.avail_out = maxsize;
598 while (deflate(&stream, Z_FINISH) == Z_OK);
601 size = stream.total_out;
602 write_or_die(fd, out, size);
603 *obj_crc = crc32(*obj_crc, out, size);
608 static void append_obj_to_pack(const unsigned char *sha1, void *buf,
609 unsigned long size, enum object_type type)
611 struct object_entry *obj = &objects[nr_objects++];
612 unsigned char header[10];
613 unsigned long s = size;
615 unsigned char c = (type << 4) | (s & 15);
618 header[n++] = c | 0x80;
623 write_or_die(output_fd, header, n);
624 obj[0].idx.crc32 = crc32(0, Z_NULL, 0);
625 obj[0].idx.crc32 = crc32(obj[0].idx.crc32, header, n);
626 obj[1].idx.offset = obj[0].idx.offset + n;
627 obj[1].idx.offset += write_compressed(output_fd, buf, size, &obj[0].idx.crc32);
628 hashcpy(obj->idx.sha1, sha1);
631 static int delta_pos_compare(const void *_a, const void *_b)
633 struct delta_entry *a = *(struct delta_entry **)_a;
634 struct delta_entry *b = *(struct delta_entry **)_b;
635 return a->obj_no - b->obj_no;
638 static void fix_unresolved_deltas(int nr_unresolved)
640 struct delta_entry **sorted_by_pos;
644 * Since many unresolved deltas may well be themselves base objects
645 * for more unresolved deltas, we really want to include the
646 * smallest number of base objects that would cover as much delta
647 * as possible by picking the
648 * trunc deltas first, allowing for other deltas to resolve without
649 * additional base objects. Since most base objects are to be found
650 * before deltas depending on them, a good heuristic is to start
651 * resolving deltas in the same order as their position in the pack.
653 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
654 for (i = 0; i < nr_deltas; i++) {
655 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
657 sorted_by_pos[n++] = &deltas[i];
659 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
661 for (i = 0; i < n; i++) {
662 struct delta_entry *d = sorted_by_pos[i];
663 enum object_type type;
665 struct base_data base_obj;
667 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
669 base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size);
673 find_delta_children(&d->base, &first, &last);
674 for (j = first; j <= last; j++) {
675 struct object_entry *child = objects + deltas[j].obj_no;
676 if (child->real_type == OBJ_REF_DELTA)
677 resolve_delta(child, &base_obj, type);
680 if (check_sha1_signature(d->base.sha1, base_obj.data,
681 base_obj.size, typename(type)))
682 die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
683 append_obj_to_pack(d->base.sha1, base_obj.data,
684 base_obj.size, type);
686 display_progress(progress, nr_resolved_deltas);
691 static void final(const char *final_pack_name, const char *curr_pack_name,
692 const char *final_index_name, const char *curr_index_name,
693 const char *keep_name, const char *keep_msg,
696 const char *report = "pack";
703 fsync_or_die(output_fd, curr_pack_name);
704 err = close(output_fd);
706 die("error while closing pack file: %s", strerror(errno));
707 chmod(curr_pack_name, 0444);
711 int keep_fd, keep_msg_len = strlen(keep_msg);
713 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
714 get_object_directory(), sha1_to_hex(sha1));
717 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
720 die("cannot write keep file");
722 if (keep_msg_len > 0) {
723 write_or_die(keep_fd, keep_msg, keep_msg_len);
724 write_or_die(keep_fd, "\n", 1);
726 if (close(keep_fd) != 0)
727 die("cannot write keep file");
732 if (final_pack_name != curr_pack_name) {
733 if (!final_pack_name) {
734 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
735 get_object_directory(), sha1_to_hex(sha1));
736 final_pack_name = name;
738 if (move_temp_to_file(curr_pack_name, final_pack_name))
739 die("cannot store pack file");
742 chmod(curr_index_name, 0444);
743 if (final_index_name != curr_index_name) {
744 if (!final_index_name) {
745 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
746 get_object_directory(), sha1_to_hex(sha1));
747 final_index_name = name;
749 if (move_temp_to_file(curr_index_name, final_index_name))
750 die("cannot store index file");
754 printf("%s\n", sha1_to_hex(sha1));
757 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
758 report, sha1_to_hex(sha1));
759 write_or_die(1, buf, len);
762 * Let's just mimic git-unpack-objects here and write
763 * the last part of the input buffer to stdout.
766 err = xwrite(1, input_buffer + input_offset, input_len);
775 static int git_index_pack_config(const char *k, const char *v, void *cb)
777 if (!strcmp(k, "pack.indexversion")) {
778 pack_idx_default_version = git_config_int(k, v);
779 if (pack_idx_default_version > 2)
780 die("bad pack.indexversion=%d", pack_idx_default_version);
783 return git_default_config(k, v, cb);
786 int main(int argc, char **argv)
788 int i, fix_thin_pack = 0;
789 char *curr_pack, *pack_name = NULL;
790 char *curr_index, *index_name = NULL;
791 const char *keep_name = NULL, *keep_msg = NULL;
792 char *index_name_buf = NULL, *keep_name_buf = NULL;
793 struct pack_idx_entry **idx_objects;
794 unsigned char sha1[20];
796 git_config(git_index_pack_config, NULL);
798 for (i = 1; i < argc; i++) {
802 if (!strcmp(arg, "--stdin")) {
804 } else if (!strcmp(arg, "--fix-thin")) {
806 } else if (!strcmp(arg, "--strict")) {
808 } else if (!strcmp(arg, "--keep")) {
810 } else if (!prefixcmp(arg, "--keep=")) {
812 } else if (!prefixcmp(arg, "--pack_header=")) {
813 struct pack_header *hdr;
816 hdr = (struct pack_header *)input_buffer;
817 hdr->hdr_signature = htonl(PACK_SIGNATURE);
818 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
821 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
824 input_len = sizeof(*hdr);
825 } else if (!strcmp(arg, "-v")) {
827 } else if (!strcmp(arg, "-o")) {
828 if (index_name || (i+1) >= argc)
829 usage(index_pack_usage);
830 index_name = argv[++i];
831 } else if (!prefixcmp(arg, "--index-version=")) {
833 pack_idx_default_version = strtoul(arg + 16, &c, 10);
834 if (pack_idx_default_version > 2)
837 pack_idx_off32_limit = strtoul(c+1, &c, 0);
838 if (*c || pack_idx_off32_limit & 0x80000000)
841 usage(index_pack_usage);
846 usage(index_pack_usage);
850 if (!pack_name && !from_stdin)
851 usage(index_pack_usage);
852 if (fix_thin_pack && !from_stdin)
853 die("--fix-thin cannot be used without --stdin");
854 if (!index_name && pack_name) {
855 int len = strlen(pack_name);
856 if (!has_extension(pack_name, ".pack"))
857 die("packfile name '%s' does not end with '.pack'",
859 index_name_buf = xmalloc(len);
860 memcpy(index_name_buf, pack_name, len - 5);
861 strcpy(index_name_buf + len - 5, ".idx");
862 index_name = index_name_buf;
864 if (keep_msg && !keep_name && pack_name) {
865 int len = strlen(pack_name);
866 if (!has_extension(pack_name, ".pack"))
867 die("packfile name '%s' does not end with '.pack'",
869 keep_name_buf = xmalloc(len);
870 memcpy(keep_name_buf, pack_name, len - 5);
871 strcpy(keep_name_buf + len - 5, ".keep");
872 keep_name = keep_name_buf;
875 curr_pack = open_pack_file(pack_name);
877 objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry));
878 deltas = xmalloc(nr_objects * sizeof(struct delta_entry));
879 parse_pack_objects(sha1);
880 if (nr_deltas == nr_resolved_deltas) {
881 stop_progress(&progress);
882 /* Flush remaining pack final 20-byte SHA1. */
887 int nr_unresolved = nr_deltas - nr_resolved_deltas;
888 int nr_objects_initial = nr_objects;
889 if (nr_unresolved <= 0)
890 die("confusion beyond insanity");
891 objects = xrealloc(objects,
892 (nr_objects + nr_unresolved + 1)
894 fix_unresolved_deltas(nr_unresolved);
895 sprintf(msg, "completed with %d local objects",
896 nr_objects - nr_objects_initial);
897 stop_progress_msg(&progress, msg);
898 fixup_pack_header_footer(output_fd, sha1,
899 curr_pack, nr_objects);
901 if (nr_deltas != nr_resolved_deltas)
902 die("pack has %d unresolved deltas",
903 nr_deltas - nr_resolved_deltas);
909 idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
910 for (i = 0; i < nr_objects; i++)
911 idx_objects[i] = &objects[i].idx;
912 curr_index = write_idx_file(index_name, idx_objects, nr_objects, sha1);
915 final(pack_name, curr_pack,
916 index_name, curr_index,
920 free(index_name_buf);
922 if (pack_name == NULL)
924 if (index_name == NULL)