13 #include "streaming.h"
14 #include "thread-utils.h"
16 static const char index_pack_usage[] =
17 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
20 struct pack_idx_entry idx;
22 unsigned char hdr_size;
24 signed char real_type;
33 struct base_data *base;
34 struct base_data *child;
35 struct object_entry *obj;
38 int ref_first, ref_last;
39 int ofs_first, ofs_last;
46 struct base_data *base_cache;
47 size_t base_cache_used;
51 #define FLAG_LINK (1u<<20)
52 #define FLAG_CHECKED (1u<<21)
54 struct ofs_delta_entry {
59 struct ref_delta_entry {
60 unsigned char sha1[20];
64 static struct object_entry *objects;
65 static struct object_stat *obj_stat;
66 static struct ofs_delta_entry *ofs_deltas;
67 static struct ref_delta_entry *ref_deltas;
68 static struct thread_local nothread_data;
69 static int nr_objects;
70 static int nr_ofs_deltas;
71 static int nr_ref_deltas;
72 static int ref_deltas_alloc;
73 static int nr_resolved_deltas;
74 static int nr_threads;
76 static int from_stdin;
78 static int do_fsck_object;
79 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
81 static int show_resolving_progress;
83 static int check_self_contained_and_connected;
85 static struct progress *progress;
87 /* We always read in 4kB chunks. */
88 static unsigned char input_buffer[4096];
89 static unsigned int input_offset, input_len;
90 static off_t consumed_bytes;
91 static off_t max_input_size;
92 static unsigned deepest_delta;
93 static git_SHA_CTX input_ctx;
94 static uint32_t input_crc32;
95 static int input_fd, output_fd;
96 static const char *curr_pack;
100 static struct thread_local *thread_data;
101 static int nr_dispatched;
102 static int threads_active;
104 static pthread_mutex_t read_mutex;
105 #define read_lock() lock_mutex(&read_mutex)
106 #define read_unlock() unlock_mutex(&read_mutex)
108 static pthread_mutex_t counter_mutex;
109 #define counter_lock() lock_mutex(&counter_mutex)
110 #define counter_unlock() unlock_mutex(&counter_mutex)
112 static pthread_mutex_t work_mutex;
113 #define work_lock() lock_mutex(&work_mutex)
114 #define work_unlock() unlock_mutex(&work_mutex)
116 static pthread_mutex_t deepest_delta_mutex;
117 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
118 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
120 static pthread_mutex_t type_cas_mutex;
121 #define type_cas_lock() lock_mutex(&type_cas_mutex)
122 #define type_cas_unlock() unlock_mutex(&type_cas_mutex)
124 static pthread_key_t key;
126 static inline void lock_mutex(pthread_mutex_t *mutex)
129 pthread_mutex_lock(mutex);
132 static inline void unlock_mutex(pthread_mutex_t *mutex)
135 pthread_mutex_unlock(mutex);
139 * Mutex and conditional variable can't be statically-initialized on Windows.
141 static void init_thread(void)
144 init_recursive_mutex(&read_mutex);
145 pthread_mutex_init(&counter_mutex, NULL);
146 pthread_mutex_init(&work_mutex, NULL);
147 pthread_mutex_init(&type_cas_mutex, NULL);
149 pthread_mutex_init(&deepest_delta_mutex, NULL);
150 pthread_key_create(&key, NULL);
151 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
152 for (i = 0; i < nr_threads; i++) {
153 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
154 if (thread_data[i].pack_fd == -1)
155 die_errno(_("unable to open %s"), curr_pack);
161 static void cleanup_thread(void)
167 pthread_mutex_destroy(&read_mutex);
168 pthread_mutex_destroy(&counter_mutex);
169 pthread_mutex_destroy(&work_mutex);
170 pthread_mutex_destroy(&type_cas_mutex);
172 pthread_mutex_destroy(&deepest_delta_mutex);
173 for (i = 0; i < nr_threads; i++)
174 close(thread_data[i].pack_fd);
175 pthread_key_delete(key);
182 #define read_unlock()
184 #define counter_lock()
185 #define counter_unlock()
188 #define work_unlock()
190 #define deepest_delta_lock()
191 #define deepest_delta_unlock()
193 #define type_cas_lock()
194 #define type_cas_unlock()
199 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
204 if (type != OBJ_ANY && obj->type != type)
205 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
207 obj->flags |= FLAG_LINK;
211 /* The content of each linked object must have been checked
212 or it must be already present in the object database */
213 static unsigned check_object(struct object *obj)
218 if (!(obj->flags & FLAG_LINK))
221 if (!(obj->flags & FLAG_CHECKED)) {
223 int type = sha1_object_info(obj->oid.hash, &size);
225 die(_("did not receive expected object %s"),
226 oid_to_hex(&obj->oid));
227 if (type != obj->type)
228 die(_("object %s: expected type %s, found %s"),
229 oid_to_hex(&obj->oid),
230 typename(obj->type), typename(type));
231 obj->flags |= FLAG_CHECKED;
238 static unsigned check_objects(void)
240 unsigned i, max, foreign_nr = 0;
242 max = get_max_object_index();
243 for (i = 0; i < max; i++)
244 foreign_nr += check_object(get_indexed_object(i));
249 /* Discard current buffer used content. */
250 static void flush(void)
254 write_or_die(output_fd, input_buffer, input_offset);
255 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
256 memmove(input_buffer, input_buffer + input_offset, input_len);
262 * Make sure at least "min" bytes are available in the buffer, and
263 * return the pointer to the buffer.
265 static void *fill(int min)
267 if (min <= input_len)
268 return input_buffer + input_offset;
269 if (min > sizeof(input_buffer))
270 die(Q_("cannot fill %d byte",
271 "cannot fill %d bytes",
276 ssize_t ret = xread(input_fd, input_buffer + input_len,
277 sizeof(input_buffer) - input_len);
281 die_errno(_("read error on input"));
285 display_throughput(progress, consumed_bytes + input_len);
286 } while (input_len < min);
290 static void use(int bytes)
292 if (bytes > input_len)
293 die(_("used more bytes than were available"));
294 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
296 input_offset += bytes;
298 /* make sure off_t is sufficiently large not to wrap */
299 if (signed_add_overflows(consumed_bytes, bytes))
300 die(_("pack too large for current definition of off_t"));
301 consumed_bytes += bytes;
302 if (max_input_size && consumed_bytes > max_input_size)
303 die(_("pack exceeds maximum allowed size"));
306 static const char *open_pack_file(const char *pack_name)
311 struct strbuf tmp_file = STRBUF_INIT;
312 output_fd = odb_mkstemp(&tmp_file,
313 "pack/tmp_pack_XXXXXX");
314 pack_name = strbuf_detach(&tmp_file, NULL);
316 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
318 die_errno(_("unable to create '%s'"), pack_name);
320 nothread_data.pack_fd = output_fd;
322 input_fd = open(pack_name, O_RDONLY);
324 die_errno(_("cannot open packfile '%s'"), pack_name);
326 nothread_data.pack_fd = input_fd;
328 git_SHA1_Init(&input_ctx);
332 static void parse_pack_header(void)
334 struct pack_header *hdr = fill(sizeof(struct pack_header));
336 /* Header consistency check */
337 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
338 die(_("pack signature mismatch"));
339 if (!pack_version_ok(hdr->hdr_version))
340 die(_("pack version %"PRIu32" unsupported"),
341 ntohl(hdr->hdr_version));
343 nr_objects = ntohl(hdr->hdr_entries);
344 use(sizeof(struct pack_header));
347 static NORETURN void bad_object(off_t offset, const char *format,
348 ...) __attribute__((format (printf, 2, 3)));
350 static NORETURN void bad_object(off_t offset, const char *format, ...)
355 va_start(params, format);
356 vsnprintf(buf, sizeof(buf), format, params);
358 die(_("pack has bad object at offset %"PRIuMAX": %s"),
359 (uintmax_t)offset, buf);
362 static inline struct thread_local *get_thread_data(void)
366 return pthread_getspecific(key);
367 assert(!threads_active &&
368 "This should only be reached when all threads are gone");
370 return ¬hread_data;
374 static void set_thread_data(struct thread_local *data)
377 pthread_setspecific(key, data);
381 static struct base_data *alloc_base_data(void)
383 struct base_data *base = xcalloc(1, sizeof(struct base_data));
389 static void free_base_data(struct base_data *c)
392 FREE_AND_NULL(c->data);
393 get_thread_data()->base_cache_used -= c->size;
397 static void prune_base_data(struct base_data *retain)
400 struct thread_local *data = get_thread_data();
401 for (b = data->base_cache;
402 data->base_cache_used > delta_base_cache_limit && b;
404 if (b->data && b != retain)
409 static void link_base_data(struct base_data *base, struct base_data *c)
414 get_thread_data()->base_cache = c;
419 get_thread_data()->base_cache_used += c->size;
423 static void unlink_base_data(struct base_data *c)
425 struct base_data *base = c->base;
429 get_thread_data()->base_cache = NULL;
433 static int is_delta_type(enum object_type type)
435 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
438 static void *unpack_entry_data(off_t offset, unsigned long size,
439 enum object_type type, unsigned char *sha1)
441 static char fixed_buf[8192];
449 if (!is_delta_type(type)) {
450 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), size) + 1;
452 git_SHA1_Update(&c, hdr, hdrlen);
455 if (type == OBJ_BLOB && size > big_file_threshold)
458 buf = xmallocz(size);
460 memset(&stream, 0, sizeof(stream));
461 git_inflate_init(&stream);
462 stream.next_out = buf;
463 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
466 unsigned char *last_out = stream.next_out;
467 stream.next_in = fill(1);
468 stream.avail_in = input_len;
469 status = git_inflate(&stream, 0);
470 use(input_len - stream.avail_in);
472 git_SHA1_Update(&c, last_out, stream.next_out - last_out);
473 if (buf == fixed_buf) {
474 stream.next_out = buf;
475 stream.avail_out = sizeof(fixed_buf);
477 } while (status == Z_OK);
478 if (stream.total_out != size || status != Z_STREAM_END)
479 bad_object(offset, _("inflate returned %d"), status);
480 git_inflate_end(&stream);
482 git_SHA1_Final(sha1, &c);
483 return buf == fixed_buf ? NULL : buf;
486 static void *unpack_raw_entry(struct object_entry *obj,
488 unsigned char *ref_sha1,
492 unsigned long size, c;
497 obj->idx.offset = consumed_bytes;
498 input_crc32 = crc32(0, NULL, 0);
503 obj->type = (c >> 4) & 7;
510 size += (c & 0x7f) << shift;
517 hashcpy(ref_sha1, fill(20));
524 base_offset = c & 127;
527 if (!base_offset || MSB(base_offset, 7))
528 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
532 base_offset = (base_offset << 7) + (c & 127);
534 *ofs_offset = obj->idx.offset - base_offset;
535 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
536 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
544 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
546 obj->hdr_size = consumed_bytes - obj->idx.offset;
548 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
549 obj->idx.crc32 = input_crc32;
553 static void *unpack_data(struct object_entry *obj,
554 int (*consume)(const unsigned char *, unsigned long, void *),
557 off_t from = obj[0].idx.offset + obj[0].hdr_size;
558 off_t len = obj[1].idx.offset - from;
559 unsigned char *data, *inbuf;
563 data = xmallocz(consume ? 64*1024 : obj->size);
564 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
566 memset(&stream, 0, sizeof(stream));
567 git_inflate_init(&stream);
568 stream.next_out = data;
569 stream.avail_out = consume ? 64*1024 : obj->size;
572 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
573 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
575 die_errno(_("cannot pread pack file"));
577 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
578 "premature end of pack file, %"PRIuMAX" bytes missing",
583 stream.next_in = inbuf;
586 status = git_inflate(&stream, 0);
589 status = git_inflate(&stream, 0);
590 if (consume(data, stream.next_out - data, cb_data)) {
595 stream.next_out = data;
596 stream.avail_out = 64*1024;
597 } while (status == Z_OK && stream.avail_in);
599 } while (len && status == Z_OK && !stream.avail_in);
601 /* This has been inflated OK when first encountered, so... */
602 if (status != Z_STREAM_END || stream.total_out != obj->size)
603 die(_("serious inflate inconsistency"));
605 git_inflate_end(&stream);
613 static void *get_data_from_pack(struct object_entry *obj)
615 return unpack_data(obj, NULL, NULL);
618 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
619 enum object_type type1,
620 enum object_type type2)
622 int cmp = type1 - type2;
625 return offset1 < offset2 ? -1 :
626 offset1 > offset2 ? 1 :
630 static int find_ofs_delta(const off_t offset, enum object_type type)
632 int first = 0, last = nr_ofs_deltas;
634 while (first < last) {
635 int next = (first + last) / 2;
636 struct ofs_delta_entry *delta = &ofs_deltas[next];
639 cmp = compare_ofs_delta_bases(offset, delta->offset,
640 type, objects[delta->obj_no].type);
652 static void find_ofs_delta_children(off_t offset,
653 int *first_index, int *last_index,
654 enum object_type type)
656 int first = find_ofs_delta(offset, type);
658 int end = nr_ofs_deltas - 1;
665 while (first > 0 && ofs_deltas[first - 1].offset == offset)
667 while (last < end && ofs_deltas[last + 1].offset == offset)
669 *first_index = first;
673 static int compare_ref_delta_bases(const unsigned char *sha1,
674 const unsigned char *sha2,
675 enum object_type type1,
676 enum object_type type2)
678 int cmp = type1 - type2;
681 return hashcmp(sha1, sha2);
684 static int find_ref_delta(const unsigned char *sha1, enum object_type type)
686 int first = 0, last = nr_ref_deltas;
688 while (first < last) {
689 int next = (first + last) / 2;
690 struct ref_delta_entry *delta = &ref_deltas[next];
693 cmp = compare_ref_delta_bases(sha1, delta->sha1,
694 type, objects[delta->obj_no].type);
706 static void find_ref_delta_children(const unsigned char *sha1,
707 int *first_index, int *last_index,
708 enum object_type type)
710 int first = find_ref_delta(sha1, type);
712 int end = nr_ref_deltas - 1;
719 while (first > 0 && !hashcmp(ref_deltas[first - 1].sha1, sha1))
721 while (last < end && !hashcmp(ref_deltas[last + 1].sha1, sha1))
723 *first_index = first;
727 struct compare_data {
728 struct object_entry *entry;
729 struct git_istream *st;
731 unsigned long buf_size;
734 static int compare_objects(const unsigned char *buf, unsigned long size,
737 struct compare_data *data = cb_data;
739 if (data->buf_size < size) {
741 data->buf = xmalloc(size);
742 data->buf_size = size;
746 ssize_t len = read_istream(data->st, data->buf, size);
748 die(_("SHA1 COLLISION FOUND WITH %s !"),
749 oid_to_hex(&data->entry->idx.oid));
751 die(_("unable to read %s"),
752 oid_to_hex(&data->entry->idx.oid));
753 if (memcmp(buf, data->buf, len))
754 die(_("SHA1 COLLISION FOUND WITH %s !"),
755 oid_to_hex(&data->entry->idx.oid));
762 static int check_collison(struct object_entry *entry)
764 struct compare_data data;
765 enum object_type type;
768 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
771 memset(&data, 0, sizeof(data));
773 data.st = open_istream(entry->idx.oid.hash, &type, &size, NULL);
776 if (size != entry->size || type != entry->type)
777 die(_("SHA1 COLLISION FOUND WITH %s !"),
778 oid_to_hex(&entry->idx.oid));
779 unpack_data(entry, compare_objects, &data);
780 close_istream(data.st);
785 static void sha1_object(const void *data, struct object_entry *obj_entry,
786 unsigned long size, enum object_type type,
787 const struct object_id *oid)
789 void *new_data = NULL;
790 int collision_test_needed = 0;
792 assert(data || obj_entry);
794 if (startup_info->have_repository) {
796 collision_test_needed = has_sha1_file_with_flags(oid->hash, HAS_SHA1_QUICK);
800 if (collision_test_needed && !data) {
802 if (!check_collison(obj_entry))
803 collision_test_needed = 0;
806 if (collision_test_needed) {
808 enum object_type has_type;
809 unsigned long has_size;
811 has_type = sha1_object_info(oid->hash, &has_size);
813 die(_("cannot read existing object info %s"), oid_to_hex(oid));
814 if (has_type != type || has_size != size)
815 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
816 has_data = read_sha1_file(oid->hash, &has_type, &has_size);
819 data = new_data = get_data_from_pack(obj_entry);
821 die(_("cannot read existing object %s"), oid_to_hex(oid));
822 if (size != has_size || type != has_type ||
823 memcmp(data, has_data, size) != 0)
824 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
830 if (type == OBJ_BLOB) {
831 struct blob *blob = lookup_blob(oid);
833 blob->object.flags |= FLAG_CHECKED;
835 die(_("invalid blob object %s"), oid_to_hex(oid));
839 void *buf = (void *) data;
841 assert(data && "data can only be NULL for large _blobs_");
844 * we do not need to free the memory here, as the
845 * buf is deleted by the caller.
847 obj = parse_object_buffer(oid, type, size, buf,
850 die(_("invalid %s"), typename(type));
851 if (do_fsck_object &&
852 fsck_object(obj, buf, size, &fsck_options))
853 die(_("Error in object"));
854 if (fsck_walk(obj, NULL, &fsck_options))
855 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
857 if (obj->type == OBJ_TREE) {
858 struct tree *item = (struct tree *) obj;
862 if (obj->type == OBJ_COMMIT) {
863 struct commit *commit = (struct commit *) obj;
864 if (detach_commit_buffer(commit, NULL) != data)
865 die("BUG: parse_object_buffer transmogrified our buffer");
867 obj->flags |= FLAG_CHECKED;
876 * This function is part of find_unresolved_deltas(). There are two
877 * walkers going in the opposite ways.
879 * The first one in find_unresolved_deltas() traverses down from
880 * parent node to children, deflating nodes along the way. However,
881 * memory for deflated nodes is limited by delta_base_cache_limit, so
882 * at some point parent node's deflated content may be freed.
884 * The second walker is this function, which goes from current node up
885 * to top parent if necessary to deflate the node. In normal
886 * situation, its parent node would be already deflated, so it just
887 * needs to apply delta.
889 * In the worst case scenario, parent node is no longer deflated because
890 * we're running out of delta_base_cache_limit; we need to re-deflate
891 * parents, possibly up to the top base.
893 * All deflated objects here are subject to be freed if we exceed
894 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
895 * just need to make sure the last node is not freed.
897 static void *get_base_data(struct base_data *c)
900 struct object_entry *obj = c->obj;
901 struct base_data **delta = NULL;
902 int delta_nr = 0, delta_alloc = 0;
904 while (is_delta_type(c->obj->type) && !c->data) {
905 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
906 delta[delta_nr++] = c;
910 c->data = get_data_from_pack(obj);
912 get_thread_data()->base_cache_used += c->size;
915 for (; delta_nr > 0; delta_nr--) {
917 c = delta[delta_nr - 1];
919 base = get_base_data(c->base);
920 raw = get_data_from_pack(obj);
921 c->data = patch_delta(
927 bad_object(obj->idx.offset, _("failed to apply delta"));
928 get_thread_data()->base_cache_used += c->size;
936 static void resolve_delta(struct object_entry *delta_obj,
937 struct base_data *base, struct base_data *result)
939 void *base_data, *delta_data;
942 int i = delta_obj - objects;
943 int j = base->obj - objects;
944 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
945 deepest_delta_lock();
946 if (deepest_delta < obj_stat[i].delta_depth)
947 deepest_delta = obj_stat[i].delta_depth;
948 deepest_delta_unlock();
949 obj_stat[i].base_object_no = j;
951 delta_data = get_data_from_pack(delta_obj);
952 base_data = get_base_data(base);
953 result->obj = delta_obj;
954 result->data = patch_delta(base_data, base->size,
955 delta_data, delta_obj->size, &result->size);
958 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
959 hash_sha1_file(result->data, result->size,
960 typename(delta_obj->real_type),
961 delta_obj->idx.oid.hash);
962 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
963 &delta_obj->idx.oid);
965 nr_resolved_deltas++;
970 * Standard boolean compare-and-swap: atomically check whether "*type" is
971 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
974 static int compare_and_swap_type(signed char *type,
975 enum object_type want,
976 enum object_type set)
978 enum object_type old;
989 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
990 struct base_data *prev_base)
992 if (base->ref_last == -1 && base->ofs_last == -1) {
993 find_ref_delta_children(base->obj->idx.oid.hash,
994 &base->ref_first, &base->ref_last,
997 find_ofs_delta_children(base->obj->idx.offset,
998 &base->ofs_first, &base->ofs_last,
1001 if (base->ref_last == -1 && base->ofs_last == -1) {
1006 link_base_data(prev_base, base);
1009 if (base->ref_first <= base->ref_last) {
1010 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1011 struct base_data *result = alloc_base_data();
1013 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1014 base->obj->real_type))
1015 die("BUG: child->real_type != OBJ_REF_DELTA");
1017 resolve_delta(child, base, result);
1018 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1019 free_base_data(base);
1025 if (base->ofs_first <= base->ofs_last) {
1026 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1027 struct base_data *result = alloc_base_data();
1029 assert(child->real_type == OBJ_OFS_DELTA);
1030 child->real_type = base->obj->real_type;
1031 resolve_delta(child, base, result);
1032 if (base->ofs_first == base->ofs_last)
1033 free_base_data(base);
1039 unlink_base_data(base);
1043 static void find_unresolved_deltas(struct base_data *base)
1045 struct base_data *new_base, *prev_base = NULL;
1047 new_base = find_unresolved_deltas_1(base, prev_base);
1057 prev_base = base->base;
1062 static int compare_ofs_delta_entry(const void *a, const void *b)
1064 const struct ofs_delta_entry *delta_a = a;
1065 const struct ofs_delta_entry *delta_b = b;
1067 return delta_a->offset < delta_b->offset ? -1 :
1068 delta_a->offset > delta_b->offset ? 1 :
1072 static int compare_ref_delta_entry(const void *a, const void *b)
1074 const struct ref_delta_entry *delta_a = a;
1075 const struct ref_delta_entry *delta_b = b;
1077 return hashcmp(delta_a->sha1, delta_b->sha1);
1080 static void resolve_base(struct object_entry *obj)
1082 struct base_data *base_obj = alloc_base_data();
1083 base_obj->obj = obj;
1084 base_obj->data = NULL;
1085 find_unresolved_deltas(base_obj);
1089 static void *threaded_second_pass(void *data)
1091 set_thread_data(data);
1095 display_progress(progress, nr_resolved_deltas);
1098 while (nr_dispatched < nr_objects &&
1099 is_delta_type(objects[nr_dispatched].type))
1101 if (nr_dispatched >= nr_objects) {
1105 i = nr_dispatched++;
1108 resolve_base(&objects[i]);
1116 * - find locations of all objects;
1117 * - calculate SHA1 of all non-delta objects;
1118 * - remember base (SHA1 or offset) for all deltas.
1120 static void parse_pack_objects(unsigned char *sha1)
1122 int i, nr_delays = 0;
1123 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1124 unsigned char ref_delta_sha1[20];
1128 progress = start_progress(
1129 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1131 for (i = 0; i < nr_objects; i++) {
1132 struct object_entry *obj = &objects[i];
1133 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1136 obj->real_type = obj->type;
1137 if (obj->type == OBJ_OFS_DELTA) {
1139 ofs_delta->obj_no = i;
1141 } else if (obj->type == OBJ_REF_DELTA) {
1142 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1143 hashcpy(ref_deltas[nr_ref_deltas].sha1, ref_delta_sha1);
1144 ref_deltas[nr_ref_deltas].obj_no = i;
1147 /* large blobs, check later */
1148 obj->real_type = OBJ_BAD;
1151 sha1_object(data, NULL, obj->size, obj->type,
1154 display_progress(progress, i+1);
1156 objects[i].idx.offset = consumed_bytes;
1157 stop_progress(&progress);
1159 /* Check pack integrity */
1161 git_SHA1_Final(sha1, &input_ctx);
1162 if (hashcmp(fill(20), sha1))
1163 die(_("pack is corrupted (SHA1 mismatch)"));
1166 /* If input_fd is a file, we should have reached its end now. */
1167 if (fstat(input_fd, &st))
1168 die_errno(_("cannot fstat packfile"));
1169 if (S_ISREG(st.st_mode) &&
1170 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1171 die(_("pack has junk at the end"));
1173 for (i = 0; i < nr_objects; i++) {
1174 struct object_entry *obj = &objects[i];
1175 if (obj->real_type != OBJ_BAD)
1177 obj->real_type = obj->type;
1178 sha1_object(NULL, obj, obj->size, obj->type,
1183 die(_("confusion beyond insanity in parse_pack_objects()"));
1188 * - for all non-delta objects, look if it is used as a base for
1190 * - if used as a base, uncompress the object and apply all deltas,
1191 * recursively checking if the resulting object is used as a base
1192 * for some more deltas.
1194 static void resolve_deltas(void)
1198 if (!nr_ofs_deltas && !nr_ref_deltas)
1201 /* Sort deltas by base SHA1/offset for fast searching */
1202 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1203 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1205 if (verbose || show_resolving_progress)
1206 progress = start_progress(_("Resolving deltas"),
1207 nr_ref_deltas + nr_ofs_deltas);
1211 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1213 for (i = 0; i < nr_threads; i++) {
1214 int ret = pthread_create(&thread_data[i].thread, NULL,
1215 threaded_second_pass, thread_data + i);
1217 die(_("unable to create thread: %s"),
1220 for (i = 0; i < nr_threads; i++)
1221 pthread_join(thread_data[i].thread, NULL);
1227 for (i = 0; i < nr_objects; i++) {
1228 struct object_entry *obj = &objects[i];
1230 if (is_delta_type(obj->type))
1233 display_progress(progress, nr_resolved_deltas);
1239 * - append objects to convert thin pack to full pack if required
1240 * - write the final 20-byte SHA-1
1242 static void fix_unresolved_deltas(struct sha1file *f);
1243 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1245 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1246 stop_progress(&progress);
1247 /* Flush remaining pack final 20-byte SHA1. */
1252 if (fix_thin_pack) {
1254 unsigned char read_sha1[20], tail_sha1[20];
1255 struct strbuf msg = STRBUF_INIT;
1256 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1257 int nr_objects_initial = nr_objects;
1258 if (nr_unresolved <= 0)
1259 die(_("confusion beyond insanity"));
1260 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1261 memset(objects + nr_objects + 1, 0,
1262 nr_unresolved * sizeof(*objects));
1263 f = sha1fd(output_fd, curr_pack);
1264 fix_unresolved_deltas(f);
1265 strbuf_addf(&msg, Q_("completed with %d local object",
1266 "completed with %d local objects",
1267 nr_objects - nr_objects_initial),
1268 nr_objects - nr_objects_initial);
1269 stop_progress_msg(&progress, msg.buf);
1270 strbuf_release(&msg);
1271 sha1close(f, tail_sha1, 0);
1272 hashcpy(read_sha1, pack_sha1);
1273 fixup_pack_header_footer(output_fd, pack_sha1,
1274 curr_pack, nr_objects,
1275 read_sha1, consumed_bytes-20);
1276 if (hashcmp(read_sha1, tail_sha1) != 0)
1277 die(_("Unexpected tail checksum for %s "
1278 "(disk corruption?)"), curr_pack);
1280 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1281 die(Q_("pack has %d unresolved delta",
1282 "pack has %d unresolved deltas",
1283 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1284 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1287 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1291 unsigned char outbuf[4096];
1293 git_deflate_init(&stream, zlib_compression_level);
1294 stream.next_in = in;
1295 stream.avail_in = size;
1298 stream.next_out = outbuf;
1299 stream.avail_out = sizeof(outbuf);
1300 status = git_deflate(&stream, Z_FINISH);
1301 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1302 } while (status == Z_OK);
1304 if (status != Z_STREAM_END)
1305 die(_("unable to deflate appended object (%d)"), status);
1306 size = stream.total_out;
1307 git_deflate_end(&stream);
1311 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1312 const unsigned char *sha1, void *buf,
1313 unsigned long size, enum object_type type)
1315 struct object_entry *obj = &objects[nr_objects++];
1316 unsigned char header[10];
1317 unsigned long s = size;
1319 unsigned char c = (type << 4) | (s & 15);
1322 header[n++] = c | 0x80;
1328 sha1write(f, header, n);
1330 obj[0].hdr_size = n;
1332 obj[0].real_type = type;
1333 obj[1].idx.offset = obj[0].idx.offset + n;
1334 obj[1].idx.offset += write_compressed(f, buf, size);
1335 obj[0].idx.crc32 = crc32_end(f);
1337 hashcpy(obj->idx.oid.hash, sha1);
1341 static int delta_pos_compare(const void *_a, const void *_b)
1343 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1344 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1345 return a->obj_no - b->obj_no;
1348 static void fix_unresolved_deltas(struct sha1file *f)
1350 struct ref_delta_entry **sorted_by_pos;
1354 * Since many unresolved deltas may well be themselves base objects
1355 * for more unresolved deltas, we really want to include the
1356 * smallest number of base objects that would cover as much delta
1357 * as possible by picking the
1358 * trunc deltas first, allowing for other deltas to resolve without
1359 * additional base objects. Since most base objects are to be found
1360 * before deltas depending on them, a good heuristic is to start
1361 * resolving deltas in the same order as their position in the pack.
1363 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1364 for (i = 0; i < nr_ref_deltas; i++)
1365 sorted_by_pos[i] = &ref_deltas[i];
1366 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1368 for (i = 0; i < nr_ref_deltas; i++) {
1369 struct ref_delta_entry *d = sorted_by_pos[i];
1370 enum object_type type;
1371 struct base_data *base_obj = alloc_base_data();
1373 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1375 base_obj->data = read_sha1_file(d->sha1, &type, &base_obj->size);
1376 if (!base_obj->data)
1379 if (check_sha1_signature(d->sha1, base_obj->data,
1380 base_obj->size, typename(type)))
1381 die(_("local object %s is corrupt"), sha1_to_hex(d->sha1));
1382 base_obj->obj = append_obj_to_pack(f, d->sha1,
1383 base_obj->data, base_obj->size, type);
1384 find_unresolved_deltas(base_obj);
1385 display_progress(progress, nr_resolved_deltas);
1387 free(sorted_by_pos);
1390 static void final(const char *final_pack_name, const char *curr_pack_name,
1391 const char *final_index_name, const char *curr_index_name,
1392 const char *keep_name, const char *keep_msg,
1393 unsigned char *sha1)
1395 const char *report = "pack";
1396 struct strbuf pack_name = STRBUF_INIT;
1397 struct strbuf index_name = STRBUF_INIT;
1398 struct strbuf keep_name_buf = STRBUF_INIT;
1404 fsync_or_die(output_fd, curr_pack_name);
1405 err = close(output_fd);
1407 die_errno(_("error while closing pack file"));
1411 int keep_fd, keep_msg_len = strlen(keep_msg);
1414 keep_name = odb_pack_name(&keep_name_buf, sha1, "keep");
1416 keep_fd = odb_pack_keep(keep_name);
1418 if (errno != EEXIST)
1419 die_errno(_("cannot write keep file '%s'"),
1422 if (keep_msg_len > 0) {
1423 write_or_die(keep_fd, keep_msg, keep_msg_len);
1424 write_or_die(keep_fd, "\n", 1);
1426 if (close(keep_fd) != 0)
1427 die_errno(_("cannot close written keep file '%s'"),
1433 if (final_pack_name != curr_pack_name) {
1434 if (!final_pack_name)
1435 final_pack_name = odb_pack_name(&pack_name, sha1, "pack");
1436 if (finalize_object_file(curr_pack_name, final_pack_name))
1437 die(_("cannot store pack file"));
1438 } else if (from_stdin)
1439 chmod(final_pack_name, 0444);
1441 if (final_index_name != curr_index_name) {
1442 if (!final_index_name)
1443 final_index_name = odb_pack_name(&index_name, sha1, "idx");
1444 if (finalize_object_file(curr_index_name, final_index_name))
1445 die(_("cannot store index file"));
1447 chmod(final_index_name, 0444);
1450 printf("%s\n", sha1_to_hex(sha1));
1452 struct strbuf buf = STRBUF_INIT;
1454 strbuf_addf(&buf, "%s\t%s\n", report, sha1_to_hex(sha1));
1455 write_or_die(1, buf.buf, buf.len);
1456 strbuf_release(&buf);
1459 * Let's just mimic git-unpack-objects here and write
1460 * the last part of the input buffer to stdout.
1463 err = xwrite(1, input_buffer + input_offset, input_len);
1467 input_offset += err;
1471 strbuf_release(&index_name);
1472 strbuf_release(&pack_name);
1473 strbuf_release(&keep_name_buf);
1476 static int git_index_pack_config(const char *k, const char *v, void *cb)
1478 struct pack_idx_option *opts = cb;
1480 if (!strcmp(k, "pack.indexversion")) {
1481 opts->version = git_config_int(k, v);
1482 if (opts->version > 2)
1483 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1486 if (!strcmp(k, "pack.threads")) {
1487 nr_threads = git_config_int(k, v);
1489 die(_("invalid number of threads specified (%d)"),
1492 if (nr_threads != 1)
1493 warning(_("no threads support, ignoring %s"), k);
1498 return git_default_config(k, v, cb);
1501 static int cmp_uint32(const void *a_, const void *b_)
1503 uint32_t a = *((uint32_t *)a_);
1504 uint32_t b = *((uint32_t *)b_);
1506 return (a < b) ? -1 : (a != b);
1509 static void read_v2_anomalous_offsets(struct packed_git *p,
1510 struct pack_idx_option *opts)
1512 const uint32_t *idx1, *idx2;
1515 /* The address of the 4-byte offset table */
1516 idx1 = (((const uint32_t *)p->index_data)
1517 + 2 /* 8-byte header */
1519 + 5 * p->num_objects /* 20-byte SHA-1 table */
1520 + p->num_objects /* CRC32 table */
1523 /* The address of the 8-byte offset table */
1524 idx2 = idx1 + p->num_objects;
1526 for (i = 0; i < p->num_objects; i++) {
1527 uint32_t off = ntohl(idx1[i]);
1528 if (!(off & 0x80000000))
1530 off = off & 0x7fffffff;
1531 check_pack_index_ptr(p, &idx2[off * 2]);
1535 * The real offset is ntohl(idx2[off * 2]) in high 4
1536 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1537 * octets. But idx2[off * 2] is Zero!!!
1539 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1540 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1543 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1546 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1548 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1551 die(_("Cannot open existing pack file '%s'"), pack_name);
1552 if (open_pack_index(p))
1553 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1555 /* Read the attributes from the existing idx file */
1556 opts->version = p->index_version;
1558 if (opts->version == 2)
1559 read_v2_anomalous_offsets(p, opts);
1562 * Get rid of the idx file as we do not need it anymore.
1563 * NEEDSWORK: extract this bit from free_pack_by_name() in
1564 * sha1_file.c, perhaps? It shouldn't matter very much as we
1565 * know we haven't installed this pack (hence we never have
1566 * read anything from it).
1568 close_pack_index(p);
1572 static void show_pack_info(int stat_only)
1574 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1575 unsigned long *chain_histogram = NULL;
1578 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1580 for (i = 0; i < nr_objects; i++) {
1581 struct object_entry *obj = &objects[i];
1583 if (is_delta_type(obj->type))
1584 chain_histogram[obj_stat[i].delta_depth - 1]++;
1587 printf("%s %-6s %lu %lu %"PRIuMAX,
1588 oid_to_hex(&obj->idx.oid),
1589 typename(obj->real_type), obj->size,
1590 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1591 (uintmax_t)obj->idx.offset);
1592 if (is_delta_type(obj->type)) {
1593 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1594 printf(" %u %s", obj_stat[i].delta_depth,
1595 oid_to_hex(&bobj->idx.oid));
1601 printf_ln(Q_("non delta: %d object",
1602 "non delta: %d objects",
1605 for (i = 0; i < deepest_delta; i++) {
1606 if (!chain_histogram[i])
1608 printf_ln(Q_("chain length = %d: %lu object",
1609 "chain length = %d: %lu objects",
1610 chain_histogram[i]),
1612 chain_histogram[i]);
1616 static const char *derive_filename(const char *pack_name, const char *suffix,
1620 if (!strip_suffix(pack_name, ".pack", &len))
1621 die(_("packfile name '%s' does not end with '.pack'"),
1623 strbuf_add(buf, pack_name, len);
1624 strbuf_addstr(buf, suffix);
1628 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1630 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1631 const char *curr_index;
1632 const char *index_name = NULL, *pack_name = NULL;
1633 const char *keep_name = NULL, *keep_msg = NULL;
1634 struct strbuf index_name_buf = STRBUF_INIT,
1635 keep_name_buf = STRBUF_INIT;
1636 struct pack_idx_entry **idx_objects;
1637 struct pack_idx_option opts;
1638 unsigned char pack_sha1[20];
1639 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1640 int report_end_of_input = 0;
1642 if (argc == 2 && !strcmp(argv[1], "-h"))
1643 usage(index_pack_usage);
1645 check_replace_refs = 0;
1646 fsck_options.walk = mark_link;
1648 reset_pack_idx_option(&opts);
1649 git_config(git_index_pack_config, &opts);
1650 if (prefix && chdir(prefix))
1651 die(_("Cannot come back to cwd"));
1653 for (i = 1; i < argc; i++) {
1654 const char *arg = argv[i];
1657 if (!strcmp(arg, "--stdin")) {
1659 } else if (!strcmp(arg, "--fix-thin")) {
1661 } else if (!strcmp(arg, "--strict")) {
1664 } else if (skip_prefix(arg, "--strict=", &arg)) {
1667 fsck_set_msg_types(&fsck_options, arg);
1668 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1670 check_self_contained_and_connected = 1;
1671 } else if (!strcmp(arg, "--verify")) {
1673 } else if (!strcmp(arg, "--verify-stat")) {
1676 } else if (!strcmp(arg, "--verify-stat-only")) {
1680 } else if (!strcmp(arg, "--keep")) {
1682 } else if (starts_with(arg, "--keep=")) {
1684 } else if (starts_with(arg, "--threads=")) {
1686 nr_threads = strtoul(arg+10, &end, 0);
1687 if (!arg[10] || *end || nr_threads < 0)
1688 usage(index_pack_usage);
1690 if (nr_threads != 1)
1691 warning(_("no threads support, "
1692 "ignoring %s"), arg);
1695 } else if (starts_with(arg, "--pack_header=")) {
1696 struct pack_header *hdr;
1699 hdr = (struct pack_header *)input_buffer;
1700 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1701 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1703 die(_("bad %s"), arg);
1704 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1706 die(_("bad %s"), arg);
1707 input_len = sizeof(*hdr);
1708 } else if (!strcmp(arg, "-v")) {
1710 } else if (!strcmp(arg, "--show-resolving-progress")) {
1711 show_resolving_progress = 1;
1712 } else if (!strcmp(arg, "--report-end-of-input")) {
1713 report_end_of_input = 1;
1714 } else if (!strcmp(arg, "-o")) {
1715 if (index_name || (i+1) >= argc)
1716 usage(index_pack_usage);
1717 index_name = argv[++i];
1718 } else if (starts_with(arg, "--index-version=")) {
1720 opts.version = strtoul(arg + 16, &c, 10);
1721 if (opts.version > 2)
1722 die(_("bad %s"), arg);
1724 opts.off32_limit = strtoul(c+1, &c, 0);
1725 if (*c || opts.off32_limit & 0x80000000)
1726 die(_("bad %s"), arg);
1727 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1728 max_input_size = strtoumax(arg, NULL, 10);
1730 usage(index_pack_usage);
1735 usage(index_pack_usage);
1739 if (!pack_name && !from_stdin)
1740 usage(index_pack_usage);
1741 if (fix_thin_pack && !from_stdin)
1742 die(_("--fix-thin cannot be used without --stdin"));
1743 if (from_stdin && !startup_info->have_repository)
1744 die(_("--stdin requires a git repository"));
1745 if (!index_name && pack_name)
1746 index_name = derive_filename(pack_name, ".idx", &index_name_buf);
1747 if (keep_msg && !keep_name && pack_name)
1748 keep_name = derive_filename(pack_name, ".keep", &keep_name_buf);
1752 die(_("--verify with no packfile name given"));
1753 read_idx_option(&opts, index_name);
1754 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1757 opts.flags |= WRITE_IDX_STRICT;
1761 nr_threads = online_cpus();
1762 /* An experiment showed that more threads does not mean faster */
1768 curr_pack = open_pack_file(pack_name);
1769 parse_pack_header();
1770 objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1772 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1773 ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1774 parse_pack_objects(pack_sha1);
1775 if (report_end_of_input)
1776 write_in_full(2, "\0", 1);
1778 conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1782 foreign_nr = check_objects();
1785 show_pack_info(stat_only);
1787 ALLOC_ARRAY(idx_objects, nr_objects);
1788 for (i = 0; i < nr_objects; i++)
1789 idx_objects[i] = &objects[i].idx;
1790 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1794 final(pack_name, curr_pack,
1795 index_name, curr_index,
1796 keep_name, keep_msg,
1801 strbuf_release(&index_name_buf);
1802 strbuf_release(&keep_name_buf);
1803 if (pack_name == NULL)
1804 free((void *) curr_pack);
1805 if (index_name == NULL)
1806 free((void *) curr_index);
1809 * Let the caller know this pack is not self contained
1811 if (check_self_contained_and_connected && foreign_nr)