sha1-file: pass git_hash_algo to hash_object_file()
[git/debian.git] / builtin / index-pack.c
blob0183610a7686cb9b85937e921839ab1cf28999ac
1 #include "builtin.h"
2 #include "config.h"
3 #include "delta.h"
4 #include "pack.h"
5 #include "csum-file.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "progress.h"
11 #include "fsck.h"
12 #include "exec-cmd.h"
13 #include "streaming.h"
14 #include "thread-utils.h"
15 #include "packfile.h"
16 #include "object-store.h"
17 #include "promisor-remote.h"
19 static const char index_pack_usage[] =
20 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
22 struct object_entry {
23 struct pack_idx_entry idx;
24 unsigned long size;
25 unsigned char hdr_size;
26 signed char type;
27 signed char real_type;
30 struct object_stat {
31 unsigned delta_depth;
32 int base_object_no;
35 struct base_data {
36 struct base_data *base;
37 struct base_data *child;
38 struct object_entry *obj;
39 void *data;
40 unsigned long size;
41 int ref_first, ref_last;
42 int ofs_first, ofs_last;
45 struct thread_local {
46 pthread_t thread;
47 struct base_data *base_cache;
48 size_t base_cache_used;
49 int pack_fd;
52 /* Remember to update object flag allocation in object.h */
53 #define FLAG_LINK (1u<<20)
54 #define FLAG_CHECKED (1u<<21)
56 struct ofs_delta_entry {
57 off_t offset;
58 int obj_no;
61 struct ref_delta_entry {
62 struct object_id oid;
63 int obj_no;
66 static struct object_entry *objects;
67 static struct object_stat *obj_stat;
68 static struct ofs_delta_entry *ofs_deltas;
69 static struct ref_delta_entry *ref_deltas;
70 static struct thread_local nothread_data;
71 static int nr_objects;
72 static int nr_ofs_deltas;
73 static int nr_ref_deltas;
74 static int ref_deltas_alloc;
75 static int nr_resolved_deltas;
76 static int nr_threads;
78 static int from_stdin;
79 static int strict;
80 static int do_fsck_object;
81 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
82 static int verbose;
83 static int show_resolving_progress;
84 static int show_stat;
85 static int check_self_contained_and_connected;
87 static struct progress *progress;
89 /* We always read in 4kB chunks. */
90 static unsigned char input_buffer[4096];
91 static unsigned int input_offset, input_len;
92 static off_t consumed_bytes;
93 static off_t max_input_size;
94 static unsigned deepest_delta;
95 static git_hash_ctx input_ctx;
96 static uint32_t input_crc32;
97 static int input_fd, output_fd;
98 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)
128 if (threads_active)
129 pthread_mutex_lock(mutex);
132 static inline void unlock_mutex(pthread_mutex_t *mutex)
134 if (threads_active)
135 pthread_mutex_unlock(mutex);
139 * Mutex and conditional variable can't be statically-initialized on Windows.
141 static void init_thread(void)
143 int i;
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);
148 if (show_stat)
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);
158 threads_active = 1;
161 static void cleanup_thread(void)
163 int i;
164 if (!threads_active)
165 return;
166 threads_active = 0;
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);
171 if (show_stat)
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);
176 free(thread_data);
179 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
181 if (!obj)
182 return -1;
184 if (type != OBJ_ANY && obj->type != type)
185 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
187 obj->flags |= FLAG_LINK;
188 return 0;
191 /* The content of each linked object must have been checked
192 or it must be already present in the object database */
193 static unsigned check_object(struct object *obj)
195 if (!obj)
196 return 0;
198 if (!(obj->flags & FLAG_LINK))
199 return 0;
201 if (!(obj->flags & FLAG_CHECKED)) {
202 unsigned long size;
203 int type = oid_object_info(the_repository, &obj->oid, &size);
204 if (type <= 0)
205 die(_("did not receive expected object %s"),
206 oid_to_hex(&obj->oid));
207 if (type != obj->type)
208 die(_("object %s: expected type %s, found %s"),
209 oid_to_hex(&obj->oid),
210 type_name(obj->type), type_name(type));
211 obj->flags |= FLAG_CHECKED;
212 return 1;
215 return 0;
218 static unsigned check_objects(void)
220 unsigned i, max, foreign_nr = 0;
222 max = get_max_object_index();
224 if (verbose)
225 progress = start_delayed_progress(_("Checking objects"), max);
227 for (i = 0; i < max; i++) {
228 foreign_nr += check_object(get_indexed_object(i));
229 display_progress(progress, i + 1);
232 stop_progress(&progress);
233 return foreign_nr;
237 /* Discard current buffer used content. */
238 static void flush(void)
240 if (input_offset) {
241 if (output_fd >= 0)
242 write_or_die(output_fd, input_buffer, input_offset);
243 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
244 memmove(input_buffer, input_buffer + input_offset, input_len);
245 input_offset = 0;
250 * Make sure at least "min" bytes are available in the buffer, and
251 * return the pointer to the buffer.
253 static void *fill(int min)
255 if (min <= input_len)
256 return input_buffer + input_offset;
257 if (min > sizeof(input_buffer))
258 die(Q_("cannot fill %d byte",
259 "cannot fill %d bytes",
260 min),
261 min);
262 flush();
263 do {
264 ssize_t ret = xread(input_fd, input_buffer + input_len,
265 sizeof(input_buffer) - input_len);
266 if (ret <= 0) {
267 if (!ret)
268 die(_("early EOF"));
269 die_errno(_("read error on input"));
271 input_len += ret;
272 if (from_stdin)
273 display_throughput(progress, consumed_bytes + input_len);
274 } while (input_len < min);
275 return input_buffer;
278 static void use(int bytes)
280 if (bytes > input_len)
281 die(_("used more bytes than were available"));
282 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
283 input_len -= bytes;
284 input_offset += bytes;
286 /* make sure off_t is sufficiently large not to wrap */
287 if (signed_add_overflows(consumed_bytes, bytes))
288 die(_("pack too large for current definition of off_t"));
289 consumed_bytes += bytes;
290 if (max_input_size && consumed_bytes > max_input_size)
291 die(_("pack exceeds maximum allowed size"));
294 static const char *open_pack_file(const char *pack_name)
296 if (from_stdin) {
297 input_fd = 0;
298 if (!pack_name) {
299 struct strbuf tmp_file = STRBUF_INIT;
300 output_fd = odb_mkstemp(&tmp_file,
301 "pack/tmp_pack_XXXXXX");
302 pack_name = strbuf_detach(&tmp_file, NULL);
303 } else {
304 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
305 if (output_fd < 0)
306 die_errno(_("unable to create '%s'"), pack_name);
308 nothread_data.pack_fd = output_fd;
309 } else {
310 input_fd = open(pack_name, O_RDONLY);
311 if (input_fd < 0)
312 die_errno(_("cannot open packfile '%s'"), pack_name);
313 output_fd = -1;
314 nothread_data.pack_fd = input_fd;
316 the_hash_algo->init_fn(&input_ctx);
317 return pack_name;
320 static void parse_pack_header(void)
322 struct pack_header *hdr = fill(sizeof(struct pack_header));
324 /* Header consistency check */
325 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
326 die(_("pack signature mismatch"));
327 if (!pack_version_ok(hdr->hdr_version))
328 die(_("pack version %"PRIu32" unsupported"),
329 ntohl(hdr->hdr_version));
331 nr_objects = ntohl(hdr->hdr_entries);
332 use(sizeof(struct pack_header));
335 static NORETURN void bad_object(off_t offset, const char *format,
336 ...) __attribute__((format (printf, 2, 3)));
338 static NORETURN void bad_object(off_t offset, const char *format, ...)
340 va_list params;
341 char buf[1024];
343 va_start(params, format);
344 vsnprintf(buf, sizeof(buf), format, params);
345 va_end(params);
346 die(_("pack has bad object at offset %"PRIuMAX": %s"),
347 (uintmax_t)offset, buf);
350 static inline struct thread_local *get_thread_data(void)
352 if (HAVE_THREADS) {
353 if (threads_active)
354 return pthread_getspecific(key);
355 assert(!threads_active &&
356 "This should only be reached when all threads are gone");
358 return &nothread_data;
361 static void set_thread_data(struct thread_local *data)
363 if (threads_active)
364 pthread_setspecific(key, data);
367 static struct base_data *alloc_base_data(void)
369 struct base_data *base = xcalloc(1, sizeof(struct base_data));
370 base->ref_last = -1;
371 base->ofs_last = -1;
372 return base;
375 static void free_base_data(struct base_data *c)
377 if (c->data) {
378 FREE_AND_NULL(c->data);
379 get_thread_data()->base_cache_used -= c->size;
383 static void prune_base_data(struct base_data *retain)
385 struct base_data *b;
386 struct thread_local *data = get_thread_data();
387 for (b = data->base_cache;
388 data->base_cache_used > delta_base_cache_limit && b;
389 b = b->child) {
390 if (b->data && b != retain)
391 free_base_data(b);
395 static void link_base_data(struct base_data *base, struct base_data *c)
397 if (base)
398 base->child = c;
399 else
400 get_thread_data()->base_cache = c;
402 c->base = base;
403 c->child = NULL;
404 if (c->data)
405 get_thread_data()->base_cache_used += c->size;
406 prune_base_data(c);
409 static void unlink_base_data(struct base_data *c)
411 struct base_data *base = c->base;
412 if (base)
413 base->child = NULL;
414 else
415 get_thread_data()->base_cache = NULL;
416 free_base_data(c);
419 static int is_delta_type(enum object_type type)
421 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
424 static void *unpack_entry_data(off_t offset, unsigned long size,
425 enum object_type type, struct object_id *oid)
427 static char fixed_buf[8192];
428 int status;
429 git_zstream stream;
430 void *buf;
431 git_hash_ctx c;
432 char hdr[32];
433 int hdrlen;
435 if (!is_delta_type(type)) {
436 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX,
437 type_name(type),(uintmax_t)size) + 1;
438 the_hash_algo->init_fn(&c);
439 the_hash_algo->update_fn(&c, hdr, hdrlen);
440 } else
441 oid = NULL;
442 if (type == OBJ_BLOB && size > big_file_threshold)
443 buf = fixed_buf;
444 else
445 buf = xmallocz(size);
447 memset(&stream, 0, sizeof(stream));
448 git_inflate_init(&stream);
449 stream.next_out = buf;
450 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
452 do {
453 unsigned char *last_out = stream.next_out;
454 stream.next_in = fill(1);
455 stream.avail_in = input_len;
456 status = git_inflate(&stream, 0);
457 use(input_len - stream.avail_in);
458 if (oid)
459 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
460 if (buf == fixed_buf) {
461 stream.next_out = buf;
462 stream.avail_out = sizeof(fixed_buf);
464 } while (status == Z_OK);
465 if (stream.total_out != size || status != Z_STREAM_END)
466 bad_object(offset, _("inflate returned %d"), status);
467 git_inflate_end(&stream);
468 if (oid)
469 the_hash_algo->final_fn(oid->hash, &c);
470 return buf == fixed_buf ? NULL : buf;
473 static void *unpack_raw_entry(struct object_entry *obj,
474 off_t *ofs_offset,
475 struct object_id *ref_oid,
476 struct object_id *oid)
478 unsigned char *p;
479 unsigned long size, c;
480 off_t base_offset;
481 unsigned shift;
482 void *data;
484 obj->idx.offset = consumed_bytes;
485 input_crc32 = crc32(0, NULL, 0);
487 p = fill(1);
488 c = *p;
489 use(1);
490 obj->type = (c >> 4) & 7;
491 size = (c & 15);
492 shift = 4;
493 while (c & 0x80) {
494 p = fill(1);
495 c = *p;
496 use(1);
497 size += (c & 0x7f) << shift;
498 shift += 7;
500 obj->size = size;
502 switch (obj->type) {
503 case OBJ_REF_DELTA:
504 hashcpy(ref_oid->hash, fill(the_hash_algo->rawsz));
505 use(the_hash_algo->rawsz);
506 break;
507 case OBJ_OFS_DELTA:
508 p = fill(1);
509 c = *p;
510 use(1);
511 base_offset = c & 127;
512 while (c & 128) {
513 base_offset += 1;
514 if (!base_offset || MSB(base_offset, 7))
515 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
516 p = fill(1);
517 c = *p;
518 use(1);
519 base_offset = (base_offset << 7) + (c & 127);
521 *ofs_offset = obj->idx.offset - base_offset;
522 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
523 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
524 break;
525 case OBJ_COMMIT:
526 case OBJ_TREE:
527 case OBJ_BLOB:
528 case OBJ_TAG:
529 break;
530 default:
531 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
533 obj->hdr_size = consumed_bytes - obj->idx.offset;
535 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
536 obj->idx.crc32 = input_crc32;
537 return data;
540 static void *unpack_data(struct object_entry *obj,
541 int (*consume)(const unsigned char *, unsigned long, void *),
542 void *cb_data)
544 off_t from = obj[0].idx.offset + obj[0].hdr_size;
545 off_t len = obj[1].idx.offset - from;
546 unsigned char *data, *inbuf;
547 git_zstream stream;
548 int status;
550 data = xmallocz(consume ? 64*1024 : obj->size);
551 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
553 memset(&stream, 0, sizeof(stream));
554 git_inflate_init(&stream);
555 stream.next_out = data;
556 stream.avail_out = consume ? 64*1024 : obj->size;
558 do {
559 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
560 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
561 if (n < 0)
562 die_errno(_("cannot pread pack file"));
563 if (!n)
564 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
565 "premature end of pack file, %"PRIuMAX" bytes missing",
566 (unsigned int)len),
567 (uintmax_t)len);
568 from += n;
569 len -= n;
570 stream.next_in = inbuf;
571 stream.avail_in = n;
572 if (!consume)
573 status = git_inflate(&stream, 0);
574 else {
575 do {
576 status = git_inflate(&stream, 0);
577 if (consume(data, stream.next_out - data, cb_data)) {
578 free(inbuf);
579 free(data);
580 return NULL;
582 stream.next_out = data;
583 stream.avail_out = 64*1024;
584 } while (status == Z_OK && stream.avail_in);
586 } while (len && status == Z_OK && !stream.avail_in);
588 /* This has been inflated OK when first encountered, so... */
589 if (status != Z_STREAM_END || stream.total_out != obj->size)
590 die(_("serious inflate inconsistency"));
592 git_inflate_end(&stream);
593 free(inbuf);
594 if (consume) {
595 FREE_AND_NULL(data);
597 return data;
600 static void *get_data_from_pack(struct object_entry *obj)
602 return unpack_data(obj, NULL, NULL);
605 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
606 enum object_type type1,
607 enum object_type type2)
609 int cmp = type1 - type2;
610 if (cmp)
611 return cmp;
612 return offset1 < offset2 ? -1 :
613 offset1 > offset2 ? 1 :
617 static int find_ofs_delta(const off_t offset, enum object_type type)
619 int first = 0, last = nr_ofs_deltas;
621 while (first < last) {
622 int next = first + (last - first) / 2;
623 struct ofs_delta_entry *delta = &ofs_deltas[next];
624 int cmp;
626 cmp = compare_ofs_delta_bases(offset, delta->offset,
627 type, objects[delta->obj_no].type);
628 if (!cmp)
629 return next;
630 if (cmp < 0) {
631 last = next;
632 continue;
634 first = next+1;
636 return -first-1;
639 static void find_ofs_delta_children(off_t offset,
640 int *first_index, int *last_index,
641 enum object_type type)
643 int first = find_ofs_delta(offset, type);
644 int last = first;
645 int end = nr_ofs_deltas - 1;
647 if (first < 0) {
648 *first_index = 0;
649 *last_index = -1;
650 return;
652 while (first > 0 && ofs_deltas[first - 1].offset == offset)
653 --first;
654 while (last < end && ofs_deltas[last + 1].offset == offset)
655 ++last;
656 *first_index = first;
657 *last_index = last;
660 static int compare_ref_delta_bases(const struct object_id *oid1,
661 const struct object_id *oid2,
662 enum object_type type1,
663 enum object_type type2)
665 int cmp = type1 - type2;
666 if (cmp)
667 return cmp;
668 return oidcmp(oid1, oid2);
671 static int find_ref_delta(const struct object_id *oid, enum object_type type)
673 int first = 0, last = nr_ref_deltas;
675 while (first < last) {
676 int next = first + (last - first) / 2;
677 struct ref_delta_entry *delta = &ref_deltas[next];
678 int cmp;
680 cmp = compare_ref_delta_bases(oid, &delta->oid,
681 type, objects[delta->obj_no].type);
682 if (!cmp)
683 return next;
684 if (cmp < 0) {
685 last = next;
686 continue;
688 first = next+1;
690 return -first-1;
693 static void find_ref_delta_children(const struct object_id *oid,
694 int *first_index, int *last_index,
695 enum object_type type)
697 int first = find_ref_delta(oid, type);
698 int last = first;
699 int end = nr_ref_deltas - 1;
701 if (first < 0) {
702 *first_index = 0;
703 *last_index = -1;
704 return;
706 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
707 --first;
708 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
709 ++last;
710 *first_index = first;
711 *last_index = last;
714 struct compare_data {
715 struct object_entry *entry;
716 struct git_istream *st;
717 unsigned char *buf;
718 unsigned long buf_size;
721 static int compare_objects(const unsigned char *buf, unsigned long size,
722 void *cb_data)
724 struct compare_data *data = cb_data;
726 if (data->buf_size < size) {
727 free(data->buf);
728 data->buf = xmalloc(size);
729 data->buf_size = size;
732 while (size) {
733 ssize_t len = read_istream(data->st, data->buf, size);
734 if (len == 0)
735 die(_("SHA1 COLLISION FOUND WITH %s !"),
736 oid_to_hex(&data->entry->idx.oid));
737 if (len < 0)
738 die(_("unable to read %s"),
739 oid_to_hex(&data->entry->idx.oid));
740 if (memcmp(buf, data->buf, len))
741 die(_("SHA1 COLLISION FOUND WITH %s !"),
742 oid_to_hex(&data->entry->idx.oid));
743 size -= len;
744 buf += len;
746 return 0;
749 static int check_collison(struct object_entry *entry)
751 struct compare_data data;
752 enum object_type type;
753 unsigned long size;
755 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
756 return -1;
758 memset(&data, 0, sizeof(data));
759 data.entry = entry;
760 data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
761 NULL);
762 if (!data.st)
763 return -1;
764 if (size != entry->size || type != entry->type)
765 die(_("SHA1 COLLISION FOUND WITH %s !"),
766 oid_to_hex(&entry->idx.oid));
767 unpack_data(entry, compare_objects, &data);
768 close_istream(data.st);
769 free(data.buf);
770 return 0;
773 static void sha1_object(const void *data, struct object_entry *obj_entry,
774 unsigned long size, enum object_type type,
775 const struct object_id *oid)
777 void *new_data = NULL;
778 int collision_test_needed = 0;
780 assert(data || obj_entry);
782 if (startup_info->have_repository) {
783 read_lock();
784 collision_test_needed =
785 has_object_file_with_flags(oid, OBJECT_INFO_QUICK);
786 read_unlock();
789 if (collision_test_needed && !data) {
790 read_lock();
791 if (!check_collison(obj_entry))
792 collision_test_needed = 0;
793 read_unlock();
795 if (collision_test_needed) {
796 void *has_data;
797 enum object_type has_type;
798 unsigned long has_size;
799 read_lock();
800 has_type = oid_object_info(the_repository, oid, &has_size);
801 if (has_type < 0)
802 die(_("cannot read existing object info %s"), oid_to_hex(oid));
803 if (has_type != type || has_size != size)
804 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
805 has_data = read_object_file(oid, &has_type, &has_size);
806 read_unlock();
807 if (!data)
808 data = new_data = get_data_from_pack(obj_entry);
809 if (!has_data)
810 die(_("cannot read existing object %s"), oid_to_hex(oid));
811 if (size != has_size || type != has_type ||
812 memcmp(data, has_data, size) != 0)
813 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
814 free(has_data);
817 if (strict || do_fsck_object) {
818 read_lock();
819 if (type == OBJ_BLOB) {
820 struct blob *blob = lookup_blob(the_repository, oid);
821 if (blob)
822 blob->object.flags |= FLAG_CHECKED;
823 else
824 die(_("invalid blob object %s"), oid_to_hex(oid));
825 if (do_fsck_object &&
826 fsck_object(&blob->object, (void *)data, size, &fsck_options))
827 die(_("fsck error in packed object"));
828 } else {
829 struct object *obj;
830 int eaten;
831 void *buf = (void *) data;
833 assert(data && "data can only be NULL for large _blobs_");
836 * we do not need to free the memory here, as the
837 * buf is deleted by the caller.
839 obj = parse_object_buffer(the_repository, oid, type,
840 size, buf,
841 &eaten);
842 if (!obj)
843 die(_("invalid %s"), type_name(type));
844 if (do_fsck_object &&
845 fsck_object(obj, buf, size, &fsck_options))
846 die(_("fsck error in packed object"));
847 if (strict && fsck_walk(obj, NULL, &fsck_options))
848 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
850 if (obj->type == OBJ_TREE) {
851 struct tree *item = (struct tree *) obj;
852 item->buffer = NULL;
853 obj->parsed = 0;
855 if (obj->type == OBJ_COMMIT) {
856 struct commit *commit = (struct commit *) obj;
857 if (detach_commit_buffer(commit, NULL) != data)
858 BUG("parse_object_buffer transmogrified our buffer");
860 obj->flags |= FLAG_CHECKED;
862 read_unlock();
865 free(new_data);
869 * This function is part of find_unresolved_deltas(). There are two
870 * walkers going in the opposite ways.
872 * The first one in find_unresolved_deltas() traverses down from
873 * parent node to children, deflating nodes along the way. However,
874 * memory for deflated nodes is limited by delta_base_cache_limit, so
875 * at some point parent node's deflated content may be freed.
877 * The second walker is this function, which goes from current node up
878 * to top parent if necessary to deflate the node. In normal
879 * situation, its parent node would be already deflated, so it just
880 * needs to apply delta.
882 * In the worst case scenario, parent node is no longer deflated because
883 * we're running out of delta_base_cache_limit; we need to re-deflate
884 * parents, possibly up to the top base.
886 * All deflated objects here are subject to be freed if we exceed
887 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
888 * just need to make sure the last node is not freed.
890 static void *get_base_data(struct base_data *c)
892 if (!c->data) {
893 struct object_entry *obj = c->obj;
894 struct base_data **delta = NULL;
895 int delta_nr = 0, delta_alloc = 0;
897 while (is_delta_type(c->obj->type) && !c->data) {
898 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
899 delta[delta_nr++] = c;
900 c = c->base;
902 if (!delta_nr) {
903 c->data = get_data_from_pack(obj);
904 c->size = obj->size;
905 get_thread_data()->base_cache_used += c->size;
906 prune_base_data(c);
908 for (; delta_nr > 0; delta_nr--) {
909 void *base, *raw;
910 c = delta[delta_nr - 1];
911 obj = c->obj;
912 base = get_base_data(c->base);
913 raw = get_data_from_pack(obj);
914 c->data = patch_delta(
915 base, c->base->size,
916 raw, obj->size,
917 &c->size);
918 free(raw);
919 if (!c->data)
920 bad_object(obj->idx.offset, _("failed to apply delta"));
921 get_thread_data()->base_cache_used += c->size;
922 prune_base_data(c);
924 free(delta);
926 return c->data;
929 static void resolve_delta(struct object_entry *delta_obj,
930 struct base_data *base, struct base_data *result)
932 void *base_data, *delta_data;
934 if (show_stat) {
935 int i = delta_obj - objects;
936 int j = base->obj - objects;
937 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
938 deepest_delta_lock();
939 if (deepest_delta < obj_stat[i].delta_depth)
940 deepest_delta = obj_stat[i].delta_depth;
941 deepest_delta_unlock();
942 obj_stat[i].base_object_no = j;
944 delta_data = get_data_from_pack(delta_obj);
945 base_data = get_base_data(base);
946 result->obj = delta_obj;
947 result->data = patch_delta(base_data, base->size,
948 delta_data, delta_obj->size, &result->size);
949 free(delta_data);
950 if (!result->data)
951 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
952 hash_object_file(the_hash_algo, result->data, result->size,
953 type_name(delta_obj->real_type), &delta_obj->idx.oid);
954 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
955 &delta_obj->idx.oid);
956 counter_lock();
957 nr_resolved_deltas++;
958 counter_unlock();
962 * Standard boolean compare-and-swap: atomically check whether "*type" is
963 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
964 * and return false.
966 static int compare_and_swap_type(signed char *type,
967 enum object_type want,
968 enum object_type set)
970 enum object_type old;
972 type_cas_lock();
973 old = *type;
974 if (old == want)
975 *type = set;
976 type_cas_unlock();
978 return old == want;
981 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
982 struct base_data *prev_base)
984 if (base->ref_last == -1 && base->ofs_last == -1) {
985 find_ref_delta_children(&base->obj->idx.oid,
986 &base->ref_first, &base->ref_last,
987 OBJ_REF_DELTA);
989 find_ofs_delta_children(base->obj->idx.offset,
990 &base->ofs_first, &base->ofs_last,
991 OBJ_OFS_DELTA);
993 if (base->ref_last == -1 && base->ofs_last == -1) {
994 free(base->data);
995 return NULL;
998 link_base_data(prev_base, base);
1001 if (base->ref_first <= base->ref_last) {
1002 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1003 struct base_data *result = alloc_base_data();
1005 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1006 base->obj->real_type))
1007 BUG("child->real_type != OBJ_REF_DELTA");
1009 resolve_delta(child, base, result);
1010 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1011 free_base_data(base);
1013 base->ref_first++;
1014 return result;
1017 if (base->ofs_first <= base->ofs_last) {
1018 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1019 struct base_data *result = alloc_base_data();
1021 assert(child->real_type == OBJ_OFS_DELTA);
1022 child->real_type = base->obj->real_type;
1023 resolve_delta(child, base, result);
1024 if (base->ofs_first == base->ofs_last)
1025 free_base_data(base);
1027 base->ofs_first++;
1028 return result;
1031 unlink_base_data(base);
1032 return NULL;
1035 static void find_unresolved_deltas(struct base_data *base)
1037 struct base_data *new_base, *prev_base = NULL;
1038 for (;;) {
1039 new_base = find_unresolved_deltas_1(base, prev_base);
1041 if (new_base) {
1042 prev_base = base;
1043 base = new_base;
1044 } else {
1045 free(base);
1046 base = prev_base;
1047 if (!base)
1048 return;
1049 prev_base = base->base;
1054 static int compare_ofs_delta_entry(const void *a, const void *b)
1056 const struct ofs_delta_entry *delta_a = a;
1057 const struct ofs_delta_entry *delta_b = b;
1059 return delta_a->offset < delta_b->offset ? -1 :
1060 delta_a->offset > delta_b->offset ? 1 :
1064 static int compare_ref_delta_entry(const void *a, const void *b)
1066 const struct ref_delta_entry *delta_a = a;
1067 const struct ref_delta_entry *delta_b = b;
1069 return oidcmp(&delta_a->oid, &delta_b->oid);
1072 static void resolve_base(struct object_entry *obj)
1074 struct base_data *base_obj = alloc_base_data();
1075 base_obj->obj = obj;
1076 base_obj->data = NULL;
1077 find_unresolved_deltas(base_obj);
1080 static void *threaded_second_pass(void *data)
1082 set_thread_data(data);
1083 for (;;) {
1084 int i;
1085 counter_lock();
1086 display_progress(progress, nr_resolved_deltas);
1087 counter_unlock();
1088 work_lock();
1089 while (nr_dispatched < nr_objects &&
1090 is_delta_type(objects[nr_dispatched].type))
1091 nr_dispatched++;
1092 if (nr_dispatched >= nr_objects) {
1093 work_unlock();
1094 break;
1096 i = nr_dispatched++;
1097 work_unlock();
1099 resolve_base(&objects[i]);
1101 return NULL;
1105 * First pass:
1106 * - find locations of all objects;
1107 * - calculate SHA1 of all non-delta objects;
1108 * - remember base (SHA1 or offset) for all deltas.
1110 static void parse_pack_objects(unsigned char *hash)
1112 int i, nr_delays = 0;
1113 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1114 struct object_id ref_delta_oid;
1115 struct stat st;
1117 if (verbose)
1118 progress = start_progress(
1119 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1120 nr_objects);
1121 for (i = 0; i < nr_objects; i++) {
1122 struct object_entry *obj = &objects[i];
1123 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1124 &ref_delta_oid,
1125 &obj->idx.oid);
1126 obj->real_type = obj->type;
1127 if (obj->type == OBJ_OFS_DELTA) {
1128 nr_ofs_deltas++;
1129 ofs_delta->obj_no = i;
1130 ofs_delta++;
1131 } else if (obj->type == OBJ_REF_DELTA) {
1132 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1133 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1134 ref_deltas[nr_ref_deltas].obj_no = i;
1135 nr_ref_deltas++;
1136 } else if (!data) {
1137 /* large blobs, check later */
1138 obj->real_type = OBJ_BAD;
1139 nr_delays++;
1140 } else
1141 sha1_object(data, NULL, obj->size, obj->type,
1142 &obj->idx.oid);
1143 free(data);
1144 display_progress(progress, i+1);
1146 objects[i].idx.offset = consumed_bytes;
1147 stop_progress(&progress);
1149 /* Check pack integrity */
1150 flush();
1151 the_hash_algo->final_fn(hash, &input_ctx);
1152 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1153 die(_("pack is corrupted (SHA1 mismatch)"));
1154 use(the_hash_algo->rawsz);
1156 /* If input_fd is a file, we should have reached its end now. */
1157 if (fstat(input_fd, &st))
1158 die_errno(_("cannot fstat packfile"));
1159 if (S_ISREG(st.st_mode) &&
1160 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1161 die(_("pack has junk at the end"));
1163 for (i = 0; i < nr_objects; i++) {
1164 struct object_entry *obj = &objects[i];
1165 if (obj->real_type != OBJ_BAD)
1166 continue;
1167 obj->real_type = obj->type;
1168 sha1_object(NULL, obj, obj->size, obj->type,
1169 &obj->idx.oid);
1170 nr_delays--;
1172 if (nr_delays)
1173 die(_("confusion beyond insanity in parse_pack_objects()"));
1177 * Second pass:
1178 * - for all non-delta objects, look if it is used as a base for
1179 * deltas;
1180 * - if used as a base, uncompress the object and apply all deltas,
1181 * recursively checking if the resulting object is used as a base
1182 * for some more deltas.
1184 static void resolve_deltas(void)
1186 int i;
1188 if (!nr_ofs_deltas && !nr_ref_deltas)
1189 return;
1191 /* Sort deltas by base SHA1/offset for fast searching */
1192 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1193 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1195 if (verbose || show_resolving_progress)
1196 progress = start_progress(_("Resolving deltas"),
1197 nr_ref_deltas + nr_ofs_deltas);
1199 nr_dispatched = 0;
1200 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1201 init_thread();
1202 for (i = 0; i < nr_threads; i++) {
1203 int ret = pthread_create(&thread_data[i].thread, NULL,
1204 threaded_second_pass, thread_data + i);
1205 if (ret)
1206 die(_("unable to create thread: %s"),
1207 strerror(ret));
1209 for (i = 0; i < nr_threads; i++)
1210 pthread_join(thread_data[i].thread, NULL);
1211 cleanup_thread();
1212 return;
1215 for (i = 0; i < nr_objects; i++) {
1216 struct object_entry *obj = &objects[i];
1218 if (is_delta_type(obj->type))
1219 continue;
1220 resolve_base(obj);
1221 display_progress(progress, nr_resolved_deltas);
1226 * Third pass:
1227 * - append objects to convert thin pack to full pack if required
1228 * - write the final pack hash
1230 static void fix_unresolved_deltas(struct hashfile *f);
1231 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1233 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1234 stop_progress(&progress);
1235 /* Flush remaining pack final hash. */
1236 flush();
1237 return;
1240 if (fix_thin_pack) {
1241 struct hashfile *f;
1242 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1243 struct strbuf msg = STRBUF_INIT;
1244 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1245 int nr_objects_initial = nr_objects;
1246 if (nr_unresolved <= 0)
1247 die(_("confusion beyond insanity"));
1248 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1249 memset(objects + nr_objects + 1, 0,
1250 nr_unresolved * sizeof(*objects));
1251 f = hashfd(output_fd, curr_pack);
1252 fix_unresolved_deltas(f);
1253 strbuf_addf(&msg, Q_("completed with %d local object",
1254 "completed with %d local objects",
1255 nr_objects - nr_objects_initial),
1256 nr_objects - nr_objects_initial);
1257 stop_progress_msg(&progress, msg.buf);
1258 strbuf_release(&msg);
1259 finalize_hashfile(f, tail_hash, 0);
1260 hashcpy(read_hash, pack_hash);
1261 fixup_pack_header_footer(output_fd, pack_hash,
1262 curr_pack, nr_objects,
1263 read_hash, consumed_bytes-the_hash_algo->rawsz);
1264 if (!hasheq(read_hash, tail_hash))
1265 die(_("Unexpected tail checksum for %s "
1266 "(disk corruption?)"), curr_pack);
1268 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1269 die(Q_("pack has %d unresolved delta",
1270 "pack has %d unresolved deltas",
1271 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1272 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1275 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1277 git_zstream stream;
1278 int status;
1279 unsigned char outbuf[4096];
1281 git_deflate_init(&stream, zlib_compression_level);
1282 stream.next_in = in;
1283 stream.avail_in = size;
1285 do {
1286 stream.next_out = outbuf;
1287 stream.avail_out = sizeof(outbuf);
1288 status = git_deflate(&stream, Z_FINISH);
1289 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1290 } while (status == Z_OK);
1292 if (status != Z_STREAM_END)
1293 die(_("unable to deflate appended object (%d)"), status);
1294 size = stream.total_out;
1295 git_deflate_end(&stream);
1296 return size;
1299 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1300 const unsigned char *sha1, void *buf,
1301 unsigned long size, enum object_type type)
1303 struct object_entry *obj = &objects[nr_objects++];
1304 unsigned char header[10];
1305 unsigned long s = size;
1306 int n = 0;
1307 unsigned char c = (type << 4) | (s & 15);
1308 s >>= 4;
1309 while (s) {
1310 header[n++] = c | 0x80;
1311 c = s & 0x7f;
1312 s >>= 7;
1314 header[n++] = c;
1315 crc32_begin(f);
1316 hashwrite(f, header, n);
1317 obj[0].size = size;
1318 obj[0].hdr_size = n;
1319 obj[0].type = type;
1320 obj[0].real_type = type;
1321 obj[1].idx.offset = obj[0].idx.offset + n;
1322 obj[1].idx.offset += write_compressed(f, buf, size);
1323 obj[0].idx.crc32 = crc32_end(f);
1324 hashflush(f);
1325 hashcpy(obj->idx.oid.hash, sha1);
1326 return obj;
1329 static int delta_pos_compare(const void *_a, const void *_b)
1331 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1332 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1333 return a->obj_no - b->obj_no;
1336 static void fix_unresolved_deltas(struct hashfile *f)
1338 struct ref_delta_entry **sorted_by_pos;
1339 int i;
1342 * Since many unresolved deltas may well be themselves base objects
1343 * for more unresolved deltas, we really want to include the
1344 * smallest number of base objects that would cover as much delta
1345 * as possible by picking the
1346 * trunc deltas first, allowing for other deltas to resolve without
1347 * additional base objects. Since most base objects are to be found
1348 * before deltas depending on them, a good heuristic is to start
1349 * resolving deltas in the same order as their position in the pack.
1351 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1352 for (i = 0; i < nr_ref_deltas; i++)
1353 sorted_by_pos[i] = &ref_deltas[i];
1354 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1356 if (has_promisor_remote()) {
1358 * Prefetch the delta bases.
1360 struct oid_array to_fetch = OID_ARRAY_INIT;
1361 for (i = 0; i < nr_ref_deltas; i++) {
1362 struct ref_delta_entry *d = sorted_by_pos[i];
1363 if (!oid_object_info_extended(the_repository, &d->oid,
1364 NULL,
1365 OBJECT_INFO_FOR_PREFETCH))
1366 continue;
1367 oid_array_append(&to_fetch, &d->oid);
1369 if (to_fetch.nr)
1370 promisor_remote_get_direct(the_repository,
1371 to_fetch.oid, to_fetch.nr);
1372 oid_array_clear(&to_fetch);
1375 for (i = 0; i < nr_ref_deltas; i++) {
1376 struct ref_delta_entry *d = sorted_by_pos[i];
1377 enum object_type type;
1378 struct base_data *base_obj = alloc_base_data();
1380 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1381 continue;
1382 base_obj->data = read_object_file(&d->oid, &type,
1383 &base_obj->size);
1384 if (!base_obj->data)
1385 continue;
1387 if (check_object_signature(&d->oid, base_obj->data,
1388 base_obj->size, type_name(type)))
1389 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1390 base_obj->obj = append_obj_to_pack(f, d->oid.hash,
1391 base_obj->data, base_obj->size, type);
1392 find_unresolved_deltas(base_obj);
1393 display_progress(progress, nr_resolved_deltas);
1395 free(sorted_by_pos);
1398 static const char *derive_filename(const char *pack_name, const char *suffix,
1399 struct strbuf *buf)
1401 size_t len;
1402 if (!strip_suffix(pack_name, ".pack", &len))
1403 die(_("packfile name '%s' does not end with '.pack'"),
1404 pack_name);
1405 strbuf_add(buf, pack_name, len);
1406 strbuf_addch(buf, '.');
1407 strbuf_addstr(buf, suffix);
1408 return buf->buf;
1411 static void write_special_file(const char *suffix, const char *msg,
1412 const char *pack_name, const unsigned char *hash,
1413 const char **report)
1415 struct strbuf name_buf = STRBUF_INIT;
1416 const char *filename;
1417 int fd;
1418 int msg_len = strlen(msg);
1420 if (pack_name)
1421 filename = derive_filename(pack_name, suffix, &name_buf);
1422 else
1423 filename = odb_pack_name(&name_buf, hash, suffix);
1425 fd = odb_pack_keep(filename);
1426 if (fd < 0) {
1427 if (errno != EEXIST)
1428 die_errno(_("cannot write %s file '%s'"),
1429 suffix, filename);
1430 } else {
1431 if (msg_len > 0) {
1432 write_or_die(fd, msg, msg_len);
1433 write_or_die(fd, "\n", 1);
1435 if (close(fd) != 0)
1436 die_errno(_("cannot close written %s file '%s'"),
1437 suffix, filename);
1438 if (report)
1439 *report = suffix;
1441 strbuf_release(&name_buf);
1444 static void final(const char *final_pack_name, const char *curr_pack_name,
1445 const char *final_index_name, const char *curr_index_name,
1446 const char *keep_msg, const char *promisor_msg,
1447 unsigned char *hash)
1449 const char *report = "pack";
1450 struct strbuf pack_name = STRBUF_INIT;
1451 struct strbuf index_name = STRBUF_INIT;
1452 int err;
1454 if (!from_stdin) {
1455 close(input_fd);
1456 } else {
1457 fsync_or_die(output_fd, curr_pack_name);
1458 err = close(output_fd);
1459 if (err)
1460 die_errno(_("error while closing pack file"));
1463 if (keep_msg)
1464 write_special_file("keep", keep_msg, final_pack_name, hash,
1465 &report);
1466 if (promisor_msg)
1467 write_special_file("promisor", promisor_msg, final_pack_name,
1468 hash, NULL);
1470 if (final_pack_name != curr_pack_name) {
1471 if (!final_pack_name)
1472 final_pack_name = odb_pack_name(&pack_name, hash, "pack");
1473 if (finalize_object_file(curr_pack_name, final_pack_name))
1474 die(_("cannot store pack file"));
1475 } else if (from_stdin)
1476 chmod(final_pack_name, 0444);
1478 if (final_index_name != curr_index_name) {
1479 if (!final_index_name)
1480 final_index_name = odb_pack_name(&index_name, hash, "idx");
1481 if (finalize_object_file(curr_index_name, final_index_name))
1482 die(_("cannot store index file"));
1483 } else
1484 chmod(final_index_name, 0444);
1486 if (do_fsck_object) {
1487 struct packed_git *p;
1488 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1489 if (p)
1490 install_packed_git(the_repository, p);
1493 if (!from_stdin) {
1494 printf("%s\n", hash_to_hex(hash));
1495 } else {
1496 struct strbuf buf = STRBUF_INIT;
1498 strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1499 write_or_die(1, buf.buf, buf.len);
1500 strbuf_release(&buf);
1503 * Let's just mimic git-unpack-objects here and write
1504 * the last part of the input buffer to stdout.
1506 while (input_len) {
1507 err = xwrite(1, input_buffer + input_offset, input_len);
1508 if (err <= 0)
1509 break;
1510 input_len -= err;
1511 input_offset += err;
1515 strbuf_release(&index_name);
1516 strbuf_release(&pack_name);
1519 static int git_index_pack_config(const char *k, const char *v, void *cb)
1521 struct pack_idx_option *opts = cb;
1523 if (!strcmp(k, "pack.indexversion")) {
1524 opts->version = git_config_int(k, v);
1525 if (opts->version > 2)
1526 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1527 return 0;
1529 if (!strcmp(k, "pack.threads")) {
1530 nr_threads = git_config_int(k, v);
1531 if (nr_threads < 0)
1532 die(_("invalid number of threads specified (%d)"),
1533 nr_threads);
1534 if (!HAVE_THREADS && nr_threads != 1) {
1535 warning(_("no threads support, ignoring %s"), k);
1536 nr_threads = 1;
1538 return 0;
1540 return git_default_config(k, v, cb);
1543 static int cmp_uint32(const void *a_, const void *b_)
1545 uint32_t a = *((uint32_t *)a_);
1546 uint32_t b = *((uint32_t *)b_);
1548 return (a < b) ? -1 : (a != b);
1551 static void read_v2_anomalous_offsets(struct packed_git *p,
1552 struct pack_idx_option *opts)
1554 const uint32_t *idx1, *idx2;
1555 uint32_t i;
1556 const uint32_t hashwords = the_hash_algo->rawsz / sizeof(uint32_t);
1558 /* The address of the 4-byte offset table */
1559 idx1 = (((const uint32_t *)p->index_data)
1560 + 2 /* 8-byte header */
1561 + 256 /* fan out */
1562 + hashwords * p->num_objects /* object ID table */
1563 + p->num_objects /* CRC32 table */
1566 /* The address of the 8-byte offset table */
1567 idx2 = idx1 + p->num_objects;
1569 for (i = 0; i < p->num_objects; i++) {
1570 uint32_t off = ntohl(idx1[i]);
1571 if (!(off & 0x80000000))
1572 continue;
1573 off = off & 0x7fffffff;
1574 check_pack_index_ptr(p, &idx2[off * 2]);
1575 if (idx2[off * 2])
1576 continue;
1578 * The real offset is ntohl(idx2[off * 2]) in high 4
1579 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1580 * octets. But idx2[off * 2] is Zero!!!
1582 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1583 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1586 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1589 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1591 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1593 if (!p)
1594 die(_("Cannot open existing pack file '%s'"), pack_name);
1595 if (open_pack_index(p))
1596 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1598 /* Read the attributes from the existing idx file */
1599 opts->version = p->index_version;
1601 if (opts->version == 2)
1602 read_v2_anomalous_offsets(p, opts);
1605 * Get rid of the idx file as we do not need it anymore.
1606 * NEEDSWORK: extract this bit from free_pack_by_name() in
1607 * sha1-file.c, perhaps? It shouldn't matter very much as we
1608 * know we haven't installed this pack (hence we never have
1609 * read anything from it).
1611 close_pack_index(p);
1612 free(p);
1615 static void show_pack_info(int stat_only)
1617 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1618 unsigned long *chain_histogram = NULL;
1620 if (deepest_delta)
1621 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1623 for (i = 0; i < nr_objects; i++) {
1624 struct object_entry *obj = &objects[i];
1626 if (is_delta_type(obj->type))
1627 chain_histogram[obj_stat[i].delta_depth - 1]++;
1628 if (stat_only)
1629 continue;
1630 printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1631 oid_to_hex(&obj->idx.oid),
1632 type_name(obj->real_type), (uintmax_t)obj->size,
1633 (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1634 (uintmax_t)obj->idx.offset);
1635 if (is_delta_type(obj->type)) {
1636 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1637 printf(" %u %s", obj_stat[i].delta_depth,
1638 oid_to_hex(&bobj->idx.oid));
1640 putchar('\n');
1643 if (baseobjects)
1644 printf_ln(Q_("non delta: %d object",
1645 "non delta: %d objects",
1646 baseobjects),
1647 baseobjects);
1648 for (i = 0; i < deepest_delta; i++) {
1649 if (!chain_histogram[i])
1650 continue;
1651 printf_ln(Q_("chain length = %d: %lu object",
1652 "chain length = %d: %lu objects",
1653 chain_histogram[i]),
1654 i + 1,
1655 chain_histogram[i]);
1659 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1661 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1662 const char *curr_index;
1663 const char *index_name = NULL, *pack_name = NULL;
1664 const char *keep_msg = NULL;
1665 const char *promisor_msg = NULL;
1666 struct strbuf index_name_buf = STRBUF_INIT;
1667 struct pack_idx_entry **idx_objects;
1668 struct pack_idx_option opts;
1669 unsigned char pack_hash[GIT_MAX_RAWSZ];
1670 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1671 int report_end_of_input = 0;
1674 * index-pack never needs to fetch missing objects except when
1675 * REF_DELTA bases are missing (which are explicitly handled). It only
1676 * accesses the repo to do hash collision checks and to check which
1677 * REF_DELTA bases need to be fetched.
1679 fetch_if_missing = 0;
1681 if (argc == 2 && !strcmp(argv[1], "-h"))
1682 usage(index_pack_usage);
1684 read_replace_refs = 0;
1685 fsck_options.walk = mark_link;
1687 reset_pack_idx_option(&opts);
1688 git_config(git_index_pack_config, &opts);
1689 if (prefix && chdir(prefix))
1690 die(_("Cannot come back to cwd"));
1692 for (i = 1; i < argc; i++) {
1693 const char *arg = argv[i];
1695 if (*arg == '-') {
1696 if (!strcmp(arg, "--stdin")) {
1697 from_stdin = 1;
1698 } else if (!strcmp(arg, "--fix-thin")) {
1699 fix_thin_pack = 1;
1700 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1701 strict = 1;
1702 do_fsck_object = 1;
1703 fsck_set_msg_types(&fsck_options, arg);
1704 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1705 strict = 1;
1706 check_self_contained_and_connected = 1;
1707 } else if (!strcmp(arg, "--fsck-objects")) {
1708 do_fsck_object = 1;
1709 } else if (!strcmp(arg, "--verify")) {
1710 verify = 1;
1711 } else if (!strcmp(arg, "--verify-stat")) {
1712 verify = 1;
1713 show_stat = 1;
1714 } else if (!strcmp(arg, "--verify-stat-only")) {
1715 verify = 1;
1716 show_stat = 1;
1717 stat_only = 1;
1718 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1719 ; /* nothing to do */
1720 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1721 ; /* already parsed */
1722 } else if (starts_with(arg, "--threads=")) {
1723 char *end;
1724 nr_threads = strtoul(arg+10, &end, 0);
1725 if (!arg[10] || *end || nr_threads < 0)
1726 usage(index_pack_usage);
1727 if (!HAVE_THREADS && nr_threads != 1) {
1728 warning(_("no threads support, ignoring %s"), arg);
1729 nr_threads = 1;
1731 } else if (starts_with(arg, "--pack_header=")) {
1732 struct pack_header *hdr;
1733 char *c;
1735 hdr = (struct pack_header *)input_buffer;
1736 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1737 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1738 if (*c != ',')
1739 die(_("bad %s"), arg);
1740 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1741 if (*c)
1742 die(_("bad %s"), arg);
1743 input_len = sizeof(*hdr);
1744 } else if (!strcmp(arg, "-v")) {
1745 verbose = 1;
1746 } else if (!strcmp(arg, "--show-resolving-progress")) {
1747 show_resolving_progress = 1;
1748 } else if (!strcmp(arg, "--report-end-of-input")) {
1749 report_end_of_input = 1;
1750 } else if (!strcmp(arg, "-o")) {
1751 if (index_name || (i+1) >= argc)
1752 usage(index_pack_usage);
1753 index_name = argv[++i];
1754 } else if (starts_with(arg, "--index-version=")) {
1755 char *c;
1756 opts.version = strtoul(arg + 16, &c, 10);
1757 if (opts.version > 2)
1758 die(_("bad %s"), arg);
1759 if (*c == ',')
1760 opts.off32_limit = strtoul(c+1, &c, 0);
1761 if (*c || opts.off32_limit & 0x80000000)
1762 die(_("bad %s"), arg);
1763 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1764 max_input_size = strtoumax(arg, NULL, 10);
1765 } else
1766 usage(index_pack_usage);
1767 continue;
1770 if (pack_name)
1771 usage(index_pack_usage);
1772 pack_name = arg;
1775 if (!pack_name && !from_stdin)
1776 usage(index_pack_usage);
1777 if (fix_thin_pack && !from_stdin)
1778 die(_("--fix-thin cannot be used without --stdin"));
1779 if (from_stdin && !startup_info->have_repository)
1780 die(_("--stdin requires a git repository"));
1781 if (!index_name && pack_name)
1782 index_name = derive_filename(pack_name, "idx", &index_name_buf);
1784 if (verify) {
1785 if (!index_name)
1786 die(_("--verify with no packfile name given"));
1787 read_idx_option(&opts, index_name);
1788 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1790 if (strict)
1791 opts.flags |= WRITE_IDX_STRICT;
1793 if (HAVE_THREADS && !nr_threads) {
1794 nr_threads = online_cpus();
1795 /* An experiment showed that more threads does not mean faster */
1796 if (nr_threads > 3)
1797 nr_threads = 3;
1800 curr_pack = open_pack_file(pack_name);
1801 parse_pack_header();
1802 objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1803 if (show_stat)
1804 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1805 ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1806 parse_pack_objects(pack_hash);
1807 if (report_end_of_input)
1808 write_in_full(2, "\0", 1);
1809 resolve_deltas();
1810 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1811 free(ofs_deltas);
1812 free(ref_deltas);
1813 if (strict)
1814 foreign_nr = check_objects();
1816 if (show_stat)
1817 show_pack_info(stat_only);
1819 ALLOC_ARRAY(idx_objects, nr_objects);
1820 for (i = 0; i < nr_objects; i++)
1821 idx_objects[i] = &objects[i].idx;
1822 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1823 free(idx_objects);
1825 if (!verify)
1826 final(pack_name, curr_pack,
1827 index_name, curr_index,
1828 keep_msg, promisor_msg,
1829 pack_hash);
1830 else
1831 close(input_fd);
1833 if (do_fsck_object && fsck_finish(&fsck_options))
1834 die(_("fsck error in pack objects"));
1836 free(objects);
1837 strbuf_release(&index_name_buf);
1838 if (pack_name == NULL)
1839 free((void *) curr_pack);
1840 if (index_name == NULL)
1841 free((void *) curr_index);
1844 * Let the caller know this pack is not self contained
1846 if (check_self_contained_and_connected && foreign_nr)
1847 return 1;
1849 return 0;