git

1
#include "builtin.h"
2
#include "config.h"
3
#include "delta.h"
4
#include "environment.h"
5
#include "gettext.h"
6
#include "hex.h"
7
#include "pack.h"
8
#include "csum-file.h"
9
#include "blob.h"
10
#include "commit.h"
11
#include "tree.h"
12
#include "progress.h"
13
#include "fsck.h"
14
#include "strbuf.h"
15
#include "streaming.h"
16
#include "thread-utils.h"
17
#include "packfile.h"
18
#include "pack-revindex.h"
19
#include "object-file.h"
20
#include "object-store-ll.h"
21
#include "oid-array.h"
22
#include "replace-object.h"
23
#include "promisor-remote.h"
24
#include "setup.h"
25

26
static const char index_pack_usage[] =
27
"git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--[no-]rev-index] [--verify] [--strict[=<msg-id>=<severity>...]] [--fsck-objects[=<msg-id>=<severity>...]] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
28

29
struct object_entry {
30
	struct pack_idx_entry idx;
31
	unsigned long size;
32
	unsigned char hdr_size;
33
	signed char type;
34
	signed char real_type;
35
};
36

37
struct object_stat {
38
	unsigned delta_depth;
39
	int base_object_no;
40
};
41

42
struct base_data {
43
	/* Initialized by make_base(). */
44
	struct base_data *base;
45
	struct object_entry *obj;
46
	int ref_first, ref_last;
47
	int ofs_first, ofs_last;
48
	/*
49
	 * Threads should increment retain_data if they are about to call
50
	 * patch_delta() using this struct's data as a base, and decrement this
51
	 * when they are done. While retain_data is nonzero, this struct's data
52
	 * will not be freed even if the delta base cache limit is exceeded.
53
	 */
54
	int retain_data;
55
	/*
56
	 * The number of direct children that have not been fully processed
57
	 * (entered work_head, entered done_head, left done_head). When this
58
	 * number reaches zero, this struct base_data can be freed.
59
	 */
60
	int children_remaining;
61

62
	/* Not initialized by make_base(). */
63
	struct list_head list;
64
	void *data;
65
	unsigned long size;
66
};
67

68
/*
69
 * Stack of struct base_data that have unprocessed children.
70
 * threaded_second_pass() uses this as a source of work (the other being the
71
 * objects array).
72
 *
73
 * Guarded by work_mutex.
74
 */
75
static LIST_HEAD(work_head);
76

77
/*
78
 * Stack of struct base_data that have children, all of whom have been
79
 * processed or are being processed, and at least one child is being processed.
80
 * These struct base_data must be kept around until the last child is
81
 * processed.
82
 *
83
 * Guarded by work_mutex.
84
 */
85
static LIST_HEAD(done_head);
86

87
/*
88
 * All threads share one delta base cache.
89
 *
90
 * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only
91
 * in a thread.
92
 */
93
static size_t base_cache_used;
94
static size_t base_cache_limit;
95

96
struct thread_local {
97
	pthread_t thread;
98
	int pack_fd;
99
};
100

101
/* Remember to update object flag allocation in object.h */
102
#define FLAG_LINK (1u<<20)
103
#define FLAG_CHECKED (1u<<21)
104

105
struct ofs_delta_entry {
106
	off_t offset;
107
	int obj_no;
108
};
109

110
struct ref_delta_entry {
111
	struct object_id oid;
112
	int obj_no;
113
};
114

115
static struct object_entry *objects;
116
static struct object_stat *obj_stat;
117
static struct ofs_delta_entry *ofs_deltas;
118
static struct ref_delta_entry *ref_deltas;
119
static struct thread_local nothread_data;
120
static int nr_objects;
121
static int nr_ofs_deltas;
122
static int nr_ref_deltas;
123
static int ref_deltas_alloc;
124
static int nr_resolved_deltas;
125
static int nr_threads;
126

127
static int from_stdin;
128
static int strict;
129
static int do_fsck_object;
130
static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES;
131
static int verbose;
132
static const char *progress_title;
133
static int show_resolving_progress;
134
static int show_stat;
135
static int check_self_contained_and_connected;
136

137
static struct progress *progress;
138

139
/* We always read in 4kB chunks. */
140
static unsigned char input_buffer[4096];
141
static unsigned int input_offset, input_len;
142
static off_t consumed_bytes;
143
static off_t max_input_size;
144
static unsigned deepest_delta;
145
static git_hash_ctx input_ctx;
146
static uint32_t input_crc32;
147
static int input_fd, output_fd;
148
static const char *curr_pack;
149

150
static struct thread_local *thread_data;
151
static int nr_dispatched;
152
static int threads_active;
153

154
static pthread_mutex_t read_mutex;
155
#define read_lock()		lock_mutex(&read_mutex)
156
#define read_unlock()		unlock_mutex(&read_mutex)
157

158
static pthread_mutex_t counter_mutex;
159
#define counter_lock()		lock_mutex(&counter_mutex)
160
#define counter_unlock()	unlock_mutex(&counter_mutex)
161

162
static pthread_mutex_t work_mutex;
163
#define work_lock()		lock_mutex(&work_mutex)
164
#define work_unlock()		unlock_mutex(&work_mutex)
165

166
static pthread_mutex_t deepest_delta_mutex;
167
#define deepest_delta_lock()	lock_mutex(&deepest_delta_mutex)
168
#define deepest_delta_unlock()	unlock_mutex(&deepest_delta_mutex)
169

170
static pthread_key_t key;
171

172
static inline void lock_mutex(pthread_mutex_t *mutex)
173
{
174
	if (threads_active)
175
		pthread_mutex_lock(mutex);
176
}
177

178
static inline void unlock_mutex(pthread_mutex_t *mutex)
179
{
180
	if (threads_active)
181
		pthread_mutex_unlock(mutex);
182
}
183

184
/*
185
 * Mutex and conditional variable can't be statically-initialized on Windows.
186
 */
187
static void init_thread(void)
188
{
189
	int i;
190
	init_recursive_mutex(&read_mutex);
191
	pthread_mutex_init(&counter_mutex, NULL);
192
	pthread_mutex_init(&work_mutex, NULL);
193
	if (show_stat)
194
		pthread_mutex_init(&deepest_delta_mutex, NULL);
195
	pthread_key_create(&key, NULL);
196
	CALLOC_ARRAY(thread_data, nr_threads);
197
	for (i = 0; i < nr_threads; i++) {
198
		thread_data[i].pack_fd = xopen(curr_pack, O_RDONLY);
199
	}
200

201
	threads_active = 1;
202
}
203

204
static void cleanup_thread(void)
205
{
206
	int i;
207
	if (!threads_active)
208
		return;
209
	threads_active = 0;
210
	pthread_mutex_destroy(&read_mutex);
211
	pthread_mutex_destroy(&counter_mutex);
212
	pthread_mutex_destroy(&work_mutex);
213
	if (show_stat)
214
		pthread_mutex_destroy(&deepest_delta_mutex);
215
	for (i = 0; i < nr_threads; i++)
216
		close(thread_data[i].pack_fd);
217
	pthread_key_delete(key);
218
	free(thread_data);
219
}
220

221
static int mark_link(struct object *obj, enum object_type type,
222
		     void *data UNUSED,
223
		     struct fsck_options *options UNUSED)
224
{
225
	if (!obj)
226
		return -1;
227

228
	if (type != OBJ_ANY && obj->type != type)
229
		die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
230

231
	obj->flags |= FLAG_LINK;
232
	return 0;
233
}
234

235
/* The content of each linked object must have been checked
236
   or it must be already present in the object database */
237
static unsigned check_object(struct object *obj)
238
{
239
	if (!obj)
240
		return 0;
241

242
	if (!(obj->flags & FLAG_LINK))
243
		return 0;
244

245
	if (!(obj->flags & FLAG_CHECKED)) {
246
		unsigned long size;
247
		int type = oid_object_info(the_repository, &obj->oid, &size);
248
		if (type <= 0)
249
			die(_("did not receive expected object %s"),
250
			      oid_to_hex(&obj->oid));
251
		if (type != obj->type)
252
			die(_("object %s: expected type %s, found %s"),
253
			    oid_to_hex(&obj->oid),
254
			    type_name(obj->type), type_name(type));
255
		obj->flags |= FLAG_CHECKED;
256
		return 1;
257
	}
258

259
	return 0;
260
}
261

262
static unsigned check_objects(void)
263
{
264
	unsigned i, max, foreign_nr = 0;
265

266
	max = get_max_object_index();
267

268
	if (verbose)
269
		progress = start_delayed_progress(_("Checking objects"), max);
270

271
	for (i = 0; i < max; i++) {
272
		foreign_nr += check_object(get_indexed_object(i));
273
		display_progress(progress, i + 1);
274
	}
275

276
	stop_progress(&progress);
277
	return foreign_nr;
278
}
279

280

281
/* Discard current buffer used content. */
282
static void flush(void)
283
{
284
	if (input_offset) {
285
		if (output_fd >= 0)
286
			write_or_die(output_fd, input_buffer, input_offset);
287
		the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
288
		memmove(input_buffer, input_buffer + input_offset, input_len);
289
		input_offset = 0;
290
	}
291
}
292

293
/*
294
 * Make sure at least "min" bytes are available in the buffer, and
295
 * return the pointer to the buffer.
296
 */
297
static void *fill(int min)
298
{
299
	if (min <= input_len)
300
		return input_buffer + input_offset;
301
	if (min > sizeof(input_buffer))
302
		die(Q_("cannot fill %d byte",
303
		       "cannot fill %d bytes",
304
		       min),
305
		    min);
306
	flush();
307
	do {
308
		ssize_t ret = xread(input_fd, input_buffer + input_len,
309
				sizeof(input_buffer) - input_len);
310
		if (ret <= 0) {
311
			if (!ret)
312
				die(_("early EOF"));
313
			die_errno(_("read error on input"));
314
		}
315
		input_len += ret;
316
		if (from_stdin)
317
			display_throughput(progress, consumed_bytes + input_len);
318
	} while (input_len < min);
319
	return input_buffer;
320
}
321

322
static void use(int bytes)
323
{
324
	if (bytes > input_len)
325
		die(_("used more bytes than were available"));
326
	input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
327
	input_len -= bytes;
328
	input_offset += bytes;
329

330
	/* make sure off_t is sufficiently large not to wrap */
331
	if (signed_add_overflows(consumed_bytes, bytes))
332
		die(_("pack too large for current definition of off_t"));
333
	consumed_bytes += bytes;
334
	if (max_input_size && consumed_bytes > max_input_size) {
335
		struct strbuf size_limit = STRBUF_INIT;
336
		strbuf_humanise_bytes(&size_limit, max_input_size);
337
		die(_("pack exceeds maximum allowed size (%s)"),
338
		    size_limit.buf);
339
	}
340
}
341

342
static const char *open_pack_file(const char *pack_name)
343
{
344
	if (from_stdin) {
345
		input_fd = 0;
346
		if (!pack_name) {
347
			struct strbuf tmp_file = STRBUF_INIT;
348
			output_fd = odb_mkstemp(&tmp_file,
349
						"pack/tmp_pack_XXXXXX");
350
			pack_name = strbuf_detach(&tmp_file, NULL);
351
		} else {
352
			output_fd = xopen(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
353
		}
354
		nothread_data.pack_fd = output_fd;
355
	} else {
356
		input_fd = xopen(pack_name, O_RDONLY);
357
		output_fd = -1;
358
		nothread_data.pack_fd = input_fd;
359
	}
360
	the_hash_algo->init_fn(&input_ctx);
361
	return pack_name;
362
}
363

364
static void parse_pack_header(void)
365
{
366
	struct pack_header *hdr = fill(sizeof(struct pack_header));
367

368
	/* Header consistency check */
369
	if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
370
		die(_("pack signature mismatch"));
371
	if (!pack_version_ok(hdr->hdr_version))
372
		die(_("pack version %"PRIu32" unsupported"),
373
			ntohl(hdr->hdr_version));
374

375
	nr_objects = ntohl(hdr->hdr_entries);
376
	use(sizeof(struct pack_header));
377
}
378

379
__attribute__((format (printf, 2, 3)))
380
static NORETURN void bad_object(off_t offset, const char *format, ...)
381
{
382
	va_list params;
383
	char buf[1024];
384

385
	va_start(params, format);
386
	vsnprintf(buf, sizeof(buf), format, params);
387
	va_end(params);
388
	die(_("pack has bad object at offset %"PRIuMAX": %s"),
389
	    (uintmax_t)offset, buf);
390
}
391

392
static inline struct thread_local *get_thread_data(void)
393
{
394
	if (HAVE_THREADS) {
395
		if (threads_active)
396
			return pthread_getspecific(key);
397
		assert(!threads_active &&
398
		       "This should only be reached when all threads are gone");
399
	}
400
	return &nothread_data;
401
}
402

403
static void set_thread_data(struct thread_local *data)
404
{
405
	if (threads_active)
406
		pthread_setspecific(key, data);
407
}
408

409
static void free_base_data(struct base_data *c)
410
{
411
	if (c->data) {
412
		FREE_AND_NULL(c->data);
413
		base_cache_used -= c->size;
414
	}
415
}
416

417
static void prune_base_data(struct base_data *retain)
418
{
419
	struct list_head *pos;
420

421
	if (base_cache_used <= base_cache_limit)
422
		return;
423

424
	list_for_each_prev(pos, &done_head) {
425
		struct base_data *b = list_entry(pos, struct base_data, list);
426
		if (b->retain_data || b == retain)
427
			continue;
428
		if (b->data) {
429
			free_base_data(b);
430
			if (base_cache_used <= base_cache_limit)
431
				return;
432
		}
433
	}
434

435
	list_for_each_prev(pos, &work_head) {
436
		struct base_data *b = list_entry(pos, struct base_data, list);
437
		if (b->retain_data || b == retain)
438
			continue;
439
		if (b->data) {
440
			free_base_data(b);
441
			if (base_cache_used <= base_cache_limit)
442
				return;
443
		}
444
	}
445
}
446

447
static int is_delta_type(enum object_type type)
448
{
449
	return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
450
}
451

452
static void *unpack_entry_data(off_t offset, unsigned long size,
453
			       enum object_type type, struct object_id *oid)
454
{
455
	static char fixed_buf[8192];
456
	int status;
457
	git_zstream stream;
458
	void *buf;
459
	git_hash_ctx c;
460
	char hdr[32];
461
	int hdrlen;
462

463
	if (!is_delta_type(type)) {
464
		hdrlen = format_object_header(hdr, sizeof(hdr), type, size);
465
		the_hash_algo->init_fn(&c);
466
		the_hash_algo->update_fn(&c, hdr, hdrlen);
467
	} else
468
		oid = NULL;
469
	if (type == OBJ_BLOB && size > big_file_threshold)
470
		buf = fixed_buf;
471
	else
472
		buf = xmallocz(size);
473

474
	memset(&stream, 0, sizeof(stream));
475
	git_inflate_init(&stream);
476
	stream.next_out = buf;
477
	stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
478

479
	do {
480
		unsigned char *last_out = stream.next_out;
481
		stream.next_in = fill(1);
482
		stream.avail_in = input_len;
483
		status = git_inflate(&stream, 0);
484
		use(input_len - stream.avail_in);
485
		if (oid)
486
			the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
487
		if (buf == fixed_buf) {
488
			stream.next_out = buf;
489
			stream.avail_out = sizeof(fixed_buf);
490
		}
491
	} while (status == Z_OK);
492
	if (stream.total_out != size || status != Z_STREAM_END)
493
		bad_object(offset, _("inflate returned %d"), status);
494
	git_inflate_end(&stream);
495
	if (oid)
496
		the_hash_algo->final_oid_fn(oid, &c);
497
	return buf == fixed_buf ? NULL : buf;
498
}
499

500
static void *unpack_raw_entry(struct object_entry *obj,
501
			      off_t *ofs_offset,
502
			      struct object_id *ref_oid,
503
			      struct object_id *oid)
504
{
505
	unsigned char *p;
506
	unsigned long size, c;
507
	off_t base_offset;
508
	unsigned shift;
509
	void *data;
510

511
	obj->idx.offset = consumed_bytes;
512
	input_crc32 = crc32(0, NULL, 0);
513

514
	p = fill(1);
515
	c = *p;
516
	use(1);
517
	obj->type = (c >> 4) & 7;
518
	size = (c & 15);
519
	shift = 4;
520
	while (c & 0x80) {
521
		p = fill(1);
522
		c = *p;
523
		use(1);
524
		size += (c & 0x7f) << shift;
525
		shift += 7;
526
	}
527
	obj->size = size;
528

529
	switch (obj->type) {
530
	case OBJ_REF_DELTA:
531
		oidread(ref_oid, fill(the_hash_algo->rawsz),
532
			the_repository->hash_algo);
533
		use(the_hash_algo->rawsz);
534
		break;
535
	case OBJ_OFS_DELTA:
536
		p = fill(1);
537
		c = *p;
538
		use(1);
539
		base_offset = c & 127;
540
		while (c & 128) {
541
			base_offset += 1;
542
			if (!base_offset || MSB(base_offset, 7))
543
				bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
544
			p = fill(1);
545
			c = *p;
546
			use(1);
547
			base_offset = (base_offset << 7) + (c & 127);
548
		}
549
		*ofs_offset = obj->idx.offset - base_offset;
550
		if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
551
			bad_object(obj->idx.offset, _("delta base offset is out of bound"));
552
		break;
553
	case OBJ_COMMIT:
554
	case OBJ_TREE:
555
	case OBJ_BLOB:
556
	case OBJ_TAG:
557
		break;
558
	default:
559
		bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
560
	}
561
	obj->hdr_size = consumed_bytes - obj->idx.offset;
562

563
	data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
564
	obj->idx.crc32 = input_crc32;
565
	return data;
566
}
567

568
static void *unpack_data(struct object_entry *obj,
569
			 int (*consume)(const unsigned char *, unsigned long, void *),
570
			 void *cb_data)
571
{
572
	off_t from = obj[0].idx.offset + obj[0].hdr_size;
573
	off_t len = obj[1].idx.offset - from;
574
	unsigned char *data, *inbuf;
575
	git_zstream stream;
576
	int status;
577

578
	data = xmallocz(consume ? 64*1024 : obj->size);
579
	inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
580

581
	memset(&stream, 0, sizeof(stream));
582
	git_inflate_init(&stream);
583
	stream.next_out = data;
584
	stream.avail_out = consume ? 64*1024 : obj->size;
585

586
	do {
587
		ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
588
		n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
589
		if (n < 0)
590
			die_errno(_("cannot pread pack file"));
591
		if (!n)
592
			die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
593
			       "premature end of pack file, %"PRIuMAX" bytes missing",
594
			       len),
595
			    (uintmax_t)len);
596
		from += n;
597
		len -= n;
598
		stream.next_in = inbuf;
599
		stream.avail_in = n;
600
		if (!consume)
601
			status = git_inflate(&stream, 0);
602
		else {
603
			do {
604
				status = git_inflate(&stream, 0);
605
				if (consume(data, stream.next_out - data, cb_data)) {
606
					free(inbuf);
607
					free(data);
608
					return NULL;
609
				}
610
				stream.next_out = data;
611
				stream.avail_out = 64*1024;
612
			} while (status == Z_OK && stream.avail_in);
613
		}
614
	} while (len && status == Z_OK && !stream.avail_in);
615

616
	/* This has been inflated OK when first encountered, so... */
617
	if (status != Z_STREAM_END || stream.total_out != obj->size)
618
		die(_("serious inflate inconsistency"));
619

620
	git_inflate_end(&stream);
621
	free(inbuf);
622
	if (consume) {
623
		FREE_AND_NULL(data);
624
	}
625
	return data;
626
}
627

628
static void *get_data_from_pack(struct object_entry *obj)
629
{
630
	return unpack_data(obj, NULL, NULL);
631
}
632

633
static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
634
				   enum object_type type1,
635
				   enum object_type type2)
636
{
637
	int cmp = type1 - type2;
638
	if (cmp)
639
		return cmp;
640
	return offset1 < offset2 ? -1 :
641
	       offset1 > offset2 ?  1 :
642
	       0;
643
}
644

645
static int find_ofs_delta(const off_t offset)
646
{
647
	int first = 0, last = nr_ofs_deltas;
648

649
	while (first < last) {
650
		int next = first + (last - first) / 2;
651
		struct ofs_delta_entry *delta = &ofs_deltas[next];
652
		int cmp;
653

654
		cmp = compare_ofs_delta_bases(offset, delta->offset,
655
					      OBJ_OFS_DELTA,
656
					      objects[delta->obj_no].type);
657
		if (!cmp)
658
			return next;
659
		if (cmp < 0) {
660
			last = next;
661
			continue;
662
		}
663
		first = next+1;
664
	}
665
	return -first-1;
666
}
667

668
static void find_ofs_delta_children(off_t offset,
669
				    int *first_index, int *last_index)
670
{
671
	int first = find_ofs_delta(offset);
672
	int last = first;
673
	int end = nr_ofs_deltas - 1;
674

675
	if (first < 0) {
676
		*first_index = 0;
677
		*last_index = -1;
678
		return;
679
	}
680
	while (first > 0 && ofs_deltas[first - 1].offset == offset)
681
		--first;
682
	while (last < end && ofs_deltas[last + 1].offset == offset)
683
		++last;
684
	*first_index = first;
685
	*last_index = last;
686
}
687

688
static int compare_ref_delta_bases(const struct object_id *oid1,
689
				   const struct object_id *oid2,
690
				   enum object_type type1,
691
				   enum object_type type2)
692
{
693
	int cmp = type1 - type2;
694
	if (cmp)
695
		return cmp;
696
	return oidcmp(oid1, oid2);
697
}
698

699
static int find_ref_delta(const struct object_id *oid)
700
{
701
	int first = 0, last = nr_ref_deltas;
702

703
	while (first < last) {
704
		int next = first + (last - first) / 2;
705
		struct ref_delta_entry *delta = &ref_deltas[next];
706
		int cmp;
707

708
		cmp = compare_ref_delta_bases(oid, &delta->oid,
709
					      OBJ_REF_DELTA,
710
					      objects[delta->obj_no].type);
711
		if (!cmp)
712
			return next;
713
		if (cmp < 0) {
714
			last = next;
715
			continue;
716
		}
717
		first = next+1;
718
	}
719
	return -first-1;
720
}
721

722
static void find_ref_delta_children(const struct object_id *oid,
723
				    int *first_index, int *last_index)
724
{
725
	int first = find_ref_delta(oid);
726
	int last = first;
727
	int end = nr_ref_deltas - 1;
728

729
	if (first < 0) {
730
		*first_index = 0;
731
		*last_index = -1;
732
		return;
733
	}
734
	while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
735
		--first;
736
	while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
737
		++last;
738
	*first_index = first;
739
	*last_index = last;
740
}
741

742
struct compare_data {
743
	struct object_entry *entry;
744
	struct git_istream *st;
745
	unsigned char *buf;
746
	unsigned long buf_size;
747
};
748

749
static int compare_objects(const unsigned char *buf, unsigned long size,
750
			   void *cb_data)
751
{
752
	struct compare_data *data = cb_data;
753

754
	if (data->buf_size < size) {
755
		free(data->buf);
756
		data->buf = xmalloc(size);
757
		data->buf_size = size;
758
	}
759

760
	while (size) {
761
		ssize_t len = read_istream(data->st, data->buf, size);
762
		if (len == 0)
763
			die(_("SHA1 COLLISION FOUND WITH %s !"),
764
			    oid_to_hex(&data->entry->idx.oid));
765
		if (len < 0)
766
			die(_("unable to read %s"),
767
			    oid_to_hex(&data->entry->idx.oid));
768
		if (memcmp(buf, data->buf, len))
769
			die(_("SHA1 COLLISION FOUND WITH %s !"),
770
			    oid_to_hex(&data->entry->idx.oid));
771
		size -= len;
772
		buf += len;
773
	}
774
	return 0;
775
}
776

777
static int check_collison(struct object_entry *entry)
778
{
779
	struct compare_data data;
780
	enum object_type type;
781
	unsigned long size;
782

783
	if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
784
		return -1;
785

786
	memset(&data, 0, sizeof(data));
787
	data.entry = entry;
788
	data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
789
			       NULL);
790
	if (!data.st)
791
		return -1;
792
	if (size != entry->size || type != entry->type)
793
		die(_("SHA1 COLLISION FOUND WITH %s !"),
794
		    oid_to_hex(&entry->idx.oid));
795
	unpack_data(entry, compare_objects, &data);
796
	close_istream(data.st);
797
	free(data.buf);
798
	return 0;
799
}
800

801
static void sha1_object(const void *data, struct object_entry *obj_entry,
802
			unsigned long size, enum object_type type,
803
			const struct object_id *oid)
804
{
805
	void *new_data = NULL;
806
	int collision_test_needed = 0;
807

808
	assert(data || obj_entry);
809

810
	if (startup_info->have_repository) {
811
		read_lock();
812
		collision_test_needed =
813
			repo_has_object_file_with_flags(the_repository, oid,
814
							OBJECT_INFO_QUICK);
815
		read_unlock();
816
	}
817

818
	if (collision_test_needed && !data) {
819
		read_lock();
820
		if (!check_collison(obj_entry))
821
			collision_test_needed = 0;
822
		read_unlock();
823
	}
824
	if (collision_test_needed) {
825
		void *has_data;
826
		enum object_type has_type;
827
		unsigned long has_size;
828
		read_lock();
829
		has_type = oid_object_info(the_repository, oid, &has_size);
830
		if (has_type < 0)
831
			die(_("cannot read existing object info %s"), oid_to_hex(oid));
832
		if (has_type != type || has_size != size)
833
			die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
834
		has_data = repo_read_object_file(the_repository, oid,
835
						 &has_type, &has_size);
836
		read_unlock();
837
		if (!data)
838
			data = new_data = get_data_from_pack(obj_entry);
839
		if (!has_data)
840
			die(_("cannot read existing object %s"), oid_to_hex(oid));
841
		if (size != has_size || type != has_type ||
842
		    memcmp(data, has_data, size) != 0)
843
			die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
844
		free(has_data);
845
	}
846

847
	if (strict || do_fsck_object) {
848
		read_lock();
849
		if (type == OBJ_BLOB) {
850
			struct blob *blob = lookup_blob(the_repository, oid);
851
			if (blob)
852
				blob->object.flags |= FLAG_CHECKED;
853
			else
854
				die(_("invalid blob object %s"), oid_to_hex(oid));
855
			if (do_fsck_object &&
856
			    fsck_object(&blob->object, (void *)data, size, &fsck_options))
857
				die(_("fsck error in packed object"));
858
		} else {
859
			struct object *obj;
860
			int eaten;
861
			void *buf = (void *) data;
862

863
			assert(data && "data can only be NULL for large _blobs_");
864

865
			/*
866
			 * we do not need to free the memory here, as the
867
			 * buf is deleted by the caller.
868
			 */
869
			obj = parse_object_buffer(the_repository, oid, type,
870
						  size, buf,
871
						  &eaten);
872
			if (!obj)
873
				die(_("invalid %s"), type_name(type));
874
			if (do_fsck_object &&
875
			    fsck_object(obj, buf, size, &fsck_options))
876
				die(_("fsck error in packed object"));
877
			if (strict && fsck_walk(obj, NULL, &fsck_options))
878
				die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
879

880
			if (obj->type == OBJ_TREE) {
881
				struct tree *item = (struct tree *) obj;
882
				item->buffer = NULL;
883
				obj->parsed = 0;
884
			}
885
			if (obj->type == OBJ_COMMIT) {
886
				struct commit *commit = (struct commit *) obj;
887
				if (detach_commit_buffer(commit, NULL) != data)
888
					BUG("parse_object_buffer transmogrified our buffer");
889
			}
890
			obj->flags |= FLAG_CHECKED;
891
		}
892
		read_unlock();
893
	}
894

895
	free(new_data);
896
}
897

898
/*
899
 * Ensure that this node has been reconstructed and return its contents.
900
 *
901
 * In the typical and best case, this node would already be reconstructed
902
 * (through the invocation to resolve_delta() in threaded_second_pass()) and it
903
 * would not be pruned. However, if pruning of this node was necessary due to
904
 * reaching delta_base_cache_limit, this function will find the closest
905
 * ancestor with reconstructed data that has not been pruned (or if there is
906
 * none, the ultimate base object), and reconstruct each node in the delta
907
 * chain in order to generate the reconstructed data for this node.
908
 */
909
static void *get_base_data(struct base_data *c)
910
{
911
	if (!c->data) {
912
		struct object_entry *obj = c->obj;
913
		struct base_data **delta = NULL;
914
		int delta_nr = 0, delta_alloc = 0;
915

916
		while (is_delta_type(c->obj->type) && !c->data) {
917
			ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
918
			delta[delta_nr++] = c;
919
			c = c->base;
920
		}
921
		if (!delta_nr) {
922
			c->data = get_data_from_pack(obj);
923
			c->size = obj->size;
924
			base_cache_used += c->size;
925
			prune_base_data(c);
926
		}
927
		for (; delta_nr > 0; delta_nr--) {
928
			void *base, *raw;
929
			c = delta[delta_nr - 1];
930
			obj = c->obj;
931
			base = get_base_data(c->base);
932
			raw = get_data_from_pack(obj);
933
			c->data = patch_delta(
934
				base, c->base->size,
935
				raw, obj->size,
936
				&c->size);
937
			free(raw);
938
			if (!c->data)
939
				bad_object(obj->idx.offset, _("failed to apply delta"));
940
			base_cache_used += c->size;
941
			prune_base_data(c);
942
		}
943
		free(delta);
944
	}
945
	return c->data;
946
}
947

948
static struct base_data *make_base(struct object_entry *obj,
949
				   struct base_data *parent)
950
{
951
	struct base_data *base = xcalloc(1, sizeof(struct base_data));
952
	base->base = parent;
953
	base->obj = obj;
954
	find_ref_delta_children(&obj->idx.oid,
955
				&base->ref_first, &base->ref_last);
956
	find_ofs_delta_children(obj->idx.offset,
957
				&base->ofs_first, &base->ofs_last);
958
	base->children_remaining = base->ref_last - base->ref_first +
959
		base->ofs_last - base->ofs_first + 2;
960
	return base;
961
}
962

963
static struct base_data *resolve_delta(struct object_entry *delta_obj,
964
				       struct base_data *base)
965
{
966
	void *delta_data, *result_data;
967
	struct base_data *result;
968
	unsigned long result_size;
969

970
	if (show_stat) {
971
		int i = delta_obj - objects;
972
		int j = base->obj - objects;
973
		obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
974
		deepest_delta_lock();
975
		if (deepest_delta < obj_stat[i].delta_depth)
976
			deepest_delta = obj_stat[i].delta_depth;
977
		deepest_delta_unlock();
978
		obj_stat[i].base_object_no = j;
979
	}
980
	delta_data = get_data_from_pack(delta_obj);
981
	assert(base->data);
982
	result_data = patch_delta(base->data, base->size,
983
				  delta_data, delta_obj->size, &result_size);
984
	free(delta_data);
985
	if (!result_data)
986
		bad_object(delta_obj->idx.offset, _("failed to apply delta"));
987
	hash_object_file(the_hash_algo, result_data, result_size,
988
			 delta_obj->real_type, &delta_obj->idx.oid);
989
	sha1_object(result_data, NULL, result_size, delta_obj->real_type,
990
		    &delta_obj->idx.oid);
991

992
	result = make_base(delta_obj, base);
993
	result->data = result_data;
994
	result->size = result_size;
995

996
	counter_lock();
997
	nr_resolved_deltas++;
998
	counter_unlock();
999

1000
	return result;
1001
}
1002

1003
static int compare_ofs_delta_entry(const void *a, const void *b)
1004
{
1005
	const struct ofs_delta_entry *delta_a = a;
1006
	const struct ofs_delta_entry *delta_b = b;
1007

1008
	return delta_a->offset < delta_b->offset ? -1 :
1009
	       delta_a->offset > delta_b->offset ?  1 :
1010
	       0;
1011
}
1012

1013
static int compare_ref_delta_entry(const void *a, const void *b)
1014
{
1015
	const struct ref_delta_entry *delta_a = a;
1016
	const struct ref_delta_entry *delta_b = b;
1017

1018
	return oidcmp(&delta_a->oid, &delta_b->oid);
1019
}
1020

1021
static void *threaded_second_pass(void *data)
1022
{
1023
	if (data)
1024
		set_thread_data(data);
1025
	for (;;) {
1026
		struct base_data *parent = NULL;
1027
		struct object_entry *child_obj;
1028
		struct base_data *child;
1029

1030
		counter_lock();
1031
		display_progress(progress, nr_resolved_deltas);
1032
		counter_unlock();
1033

1034
		work_lock();
1035
		if (list_empty(&work_head)) {
1036
			/*
1037
			 * Take an object from the object array.
1038
			 */
1039
			while (nr_dispatched < nr_objects &&
1040
			       is_delta_type(objects[nr_dispatched].type))
1041
				nr_dispatched++;
1042
			if (nr_dispatched >= nr_objects) {
1043
				work_unlock();
1044
				break;
1045
			}
1046
			child_obj = &objects[nr_dispatched++];
1047
		} else {
1048
			/*
1049
			 * Peek at the top of the stack, and take a child from
1050
			 * it.
1051
			 */
1052
			parent = list_first_entry(&work_head, struct base_data,
1053
						  list);
1054

1055
			if (parent->ref_first <= parent->ref_last) {
1056
				int offset = ref_deltas[parent->ref_first++].obj_no;
1057
				child_obj = objects + offset;
1058
				if (child_obj->real_type != OBJ_REF_DELTA)
1059
					die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)",
1060
					    (uintmax_t) child_obj->idx.offset,
1061
					    oid_to_hex(&parent->obj->idx.oid));
1062
				child_obj->real_type = parent->obj->real_type;
1063
			} else {
1064
				child_obj = objects +
1065
					ofs_deltas[parent->ofs_first++].obj_no;
1066
				assert(child_obj->real_type == OBJ_OFS_DELTA);
1067
				child_obj->real_type = parent->obj->real_type;
1068
			}
1069

1070
			if (parent->ref_first > parent->ref_last &&
1071
			    parent->ofs_first > parent->ofs_last) {
1072
				/*
1073
				 * This parent has run out of children, so move
1074
				 * it to done_head.
1075
				 */
1076
				list_del(&parent->list);
1077
				list_add(&parent->list, &done_head);
1078
			}
1079

1080
			/*
1081
			 * Ensure that the parent has data, since we will need
1082
			 * it later.
1083
			 *
1084
			 * NEEDSWORK: If parent data needs to be reloaded, this
1085
			 * prolongs the time that the current thread spends in
1086
			 * the mutex. A mitigating factor is that parent data
1087
			 * needs to be reloaded only if the delta base cache
1088
			 * limit is exceeded, so in the typical case, this does
1089
			 * not happen.
1090
			 */
1091
			get_base_data(parent);
1092
			parent->retain_data++;
1093
		}
1094
		work_unlock();
1095

1096
		if (parent) {
1097
			child = resolve_delta(child_obj, parent);
1098
			if (!child->children_remaining)
1099
				FREE_AND_NULL(child->data);
1100
		} else {
1101
			child = make_base(child_obj, NULL);
1102
			if (child->children_remaining) {
1103
				/*
1104
				 * Since this child has its own delta children,
1105
				 * we will need this data in the future.
1106
				 * Inflate now so that future iterations will
1107
				 * have access to this object's data while
1108
				 * outside the work mutex.
1109
				 */
1110
				child->data = get_data_from_pack(child_obj);
1111
				child->size = child_obj->size;
1112
			}
1113
		}
1114

1115
		work_lock();
1116
		if (parent)
1117
			parent->retain_data--;
1118
		if (child->data) {
1119
			/*
1120
			 * This child has its own children, so add it to
1121
			 * work_head.
1122
			 */
1123
			list_add(&child->list, &work_head);
1124
			base_cache_used += child->size;
1125
			prune_base_data(NULL);
1126
			free_base_data(child);
1127
		} else {
1128
			/*
1129
			 * This child does not have its own children. It may be
1130
			 * the last descendant of its ancestors; free those
1131
			 * that we can.
1132
			 */
1133
			struct base_data *p = parent;
1134

1135
			while (p) {
1136
				struct base_data *next_p;
1137

1138
				p->children_remaining--;
1139
				if (p->children_remaining)
1140
					break;
1141

1142
				next_p = p->base;
1143
				free_base_data(p);
1144
				list_del(&p->list);
1145
				free(p);
1146

1147
				p = next_p;
1148
			}
1149
			FREE_AND_NULL(child);
1150
		}
1151
		work_unlock();
1152
	}
1153
	return NULL;
1154
}
1155

1156
/*
1157
 * First pass:
1158
 * - find locations of all objects;
1159
 * - calculate SHA1 of all non-delta objects;
1160
 * - remember base (SHA1 or offset) for all deltas.
1161
 */
1162
static void parse_pack_objects(unsigned char *hash)
1163
{
1164
	int i, nr_delays = 0;
1165
	struct ofs_delta_entry *ofs_delta = ofs_deltas;
1166
	struct object_id ref_delta_oid;
1167
	struct stat st;
1168
	git_hash_ctx tmp_ctx;
1169

1170
	if (verbose)
1171
		progress = start_progress(
1172
				progress_title ? progress_title :
1173
				from_stdin ? _("Receiving objects") : _("Indexing objects"),
1174
				nr_objects);
1175
	for (i = 0; i < nr_objects; i++) {
1176
		struct object_entry *obj = &objects[i];
1177
		void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1178
					      &ref_delta_oid,
1179
					      &obj->idx.oid);
1180
		obj->real_type = obj->type;
1181
		if (obj->type == OBJ_OFS_DELTA) {
1182
			nr_ofs_deltas++;
1183
			ofs_delta->obj_no = i;
1184
			ofs_delta++;
1185
		} else if (obj->type == OBJ_REF_DELTA) {
1186
			ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1187
			oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1188
			ref_deltas[nr_ref_deltas].obj_no = i;
1189
			nr_ref_deltas++;
1190
		} else if (!data) {
1191
			/* large blobs, check later */
1192
			obj->real_type = OBJ_BAD;
1193
			nr_delays++;
1194
		} else
1195
			sha1_object(data, NULL, obj->size, obj->type,
1196
				    &obj->idx.oid);
1197
		free(data);
1198
		display_progress(progress, i+1);
1199
	}
1200
	objects[i].idx.offset = consumed_bytes;
1201
	stop_progress(&progress);
1202

1203
	/* Check pack integrity */
1204
	flush();
1205
	the_hash_algo->init_fn(&tmp_ctx);
1206
	the_hash_algo->clone_fn(&tmp_ctx, &input_ctx);
1207
	the_hash_algo->final_fn(hash, &tmp_ctx);
1208
	if (!hasheq(fill(the_hash_algo->rawsz), hash, the_repository->hash_algo))
1209
		die(_("pack is corrupted (SHA1 mismatch)"));
1210
	use(the_hash_algo->rawsz);
1211

1212
	/* If input_fd is a file, we should have reached its end now. */
1213
	if (fstat(input_fd, &st))
1214
		die_errno(_("cannot fstat packfile"));
1215
	if (S_ISREG(st.st_mode) &&
1216
			lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1217
		die(_("pack has junk at the end"));
1218

1219
	for (i = 0; i < nr_objects; i++) {
1220
		struct object_entry *obj = &objects[i];
1221
		if (obj->real_type != OBJ_BAD)
1222
			continue;
1223
		obj->real_type = obj->type;
1224
		sha1_object(NULL, obj, obj->size, obj->type,
1225
			    &obj->idx.oid);
1226
		nr_delays--;
1227
	}
1228
	if (nr_delays)
1229
		die(_("confusion beyond insanity in parse_pack_objects()"));
1230
}
1231

1232
/*
1233
 * Second pass:
1234
 * - for all non-delta objects, look if it is used as a base for
1235
 *   deltas;
1236
 * - if used as a base, uncompress the object and apply all deltas,
1237
 *   recursively checking if the resulting object is used as a base
1238
 *   for some more deltas.
1239
 */
1240
static void resolve_deltas(void)
1241
{
1242
	int i;
1243

1244
	if (!nr_ofs_deltas && !nr_ref_deltas)
1245
		return;
1246

1247
	/* Sort deltas by base SHA1/offset for fast searching */
1248
	QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1249
	QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1250

1251
	if (verbose || show_resolving_progress)
1252
		progress = start_progress(_("Resolving deltas"),
1253
					  nr_ref_deltas + nr_ofs_deltas);
1254

1255
	nr_dispatched = 0;
1256
	base_cache_limit = delta_base_cache_limit * nr_threads;
1257
	if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1258
		init_thread();
1259
		work_lock();
1260
		for (i = 0; i < nr_threads; i++) {
1261
			int ret = pthread_create(&thread_data[i].thread, NULL,
1262
						 threaded_second_pass, thread_data + i);
1263
			if (ret)
1264
				die(_("unable to create thread: %s"),
1265
				    strerror(ret));
1266
		}
1267
		work_unlock();
1268
		for (i = 0; i < nr_threads; i++)
1269
			pthread_join(thread_data[i].thread, NULL);
1270
		cleanup_thread();
1271
		return;
1272
	}
1273
	threaded_second_pass(&nothread_data);
1274
}
1275

1276
/*
1277
 * Third pass:
1278
 * - append objects to convert thin pack to full pack if required
1279
 * - write the final pack hash
1280
 */
1281
static void fix_unresolved_deltas(struct hashfile *f);
1282
static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1283
{
1284
	if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1285
		stop_progress(&progress);
1286
		/* Flush remaining pack final hash. */
1287
		flush();
1288
		return;
1289
	}
1290

1291
	if (fix_thin_pack) {
1292
		struct hashfile *f;
1293
		unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1294
		struct strbuf msg = STRBUF_INIT;
1295
		int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1296
		int nr_objects_initial = nr_objects;
1297
		if (nr_unresolved <= 0)
1298
			die(_("confusion beyond insanity"));
1299
		REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1300
		memset(objects + nr_objects + 1, 0,
1301
		       nr_unresolved * sizeof(*objects));
1302
		f = hashfd(output_fd, curr_pack);
1303
		fix_unresolved_deltas(f);
1304
		strbuf_addf(&msg, Q_("completed with %d local object",
1305
				     "completed with %d local objects",
1306
				     nr_objects - nr_objects_initial),
1307
			    nr_objects - nr_objects_initial);
1308
		stop_progress_msg(&progress, msg.buf);
1309
		strbuf_release(&msg);
1310
		finalize_hashfile(f, tail_hash, FSYNC_COMPONENT_PACK, 0);
1311
		hashcpy(read_hash, pack_hash, the_repository->hash_algo);
1312
		fixup_pack_header_footer(output_fd, pack_hash,
1313
					 curr_pack, nr_objects,
1314
					 read_hash, consumed_bytes-the_hash_algo->rawsz);
1315
		if (!hasheq(read_hash, tail_hash, the_repository->hash_algo))
1316
			die(_("Unexpected tail checksum for %s "
1317
			      "(disk corruption?)"), curr_pack);
1318
	}
1319
	if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1320
		die(Q_("pack has %d unresolved delta",
1321
		       "pack has %d unresolved deltas",
1322
		       nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1323
		    nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1324
}
1325

1326
static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1327
{
1328
	git_zstream stream;
1329
	int status;
1330
	unsigned char outbuf[4096];
1331

1332
	git_deflate_init(&stream, zlib_compression_level);
1333
	stream.next_in = in;
1334
	stream.avail_in = size;
1335

1336
	do {
1337
		stream.next_out = outbuf;
1338
		stream.avail_out = sizeof(outbuf);
1339
		status = git_deflate(&stream, Z_FINISH);
1340
		hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1341
	} while (status == Z_OK);
1342

1343
	if (status != Z_STREAM_END)
1344
		die(_("unable to deflate appended object (%d)"), status);
1345
	size = stream.total_out;
1346
	git_deflate_end(&stream);
1347
	return size;
1348
}
1349

1350
static struct object_entry *append_obj_to_pack(struct hashfile *f,
1351
			       const unsigned char *sha1, void *buf,
1352
			       unsigned long size, enum object_type type)
1353
{
1354
	struct object_entry *obj = &objects[nr_objects++];
1355
	unsigned char header[10];
1356
	unsigned long s = size;
1357
	int n = 0;
1358
	unsigned char c = (type << 4) | (s & 15);
1359
	s >>= 4;
1360
	while (s) {
1361
		header[n++] = c | 0x80;
1362
		c = s & 0x7f;
1363
		s >>= 7;
1364
	}
1365
	header[n++] = c;
1366
	crc32_begin(f);
1367
	hashwrite(f, header, n);
1368
	obj[0].size = size;
1369
	obj[0].hdr_size = n;
1370
	obj[0].type = type;
1371
	obj[0].real_type = type;
1372
	obj[1].idx.offset = obj[0].idx.offset + n;
1373
	obj[1].idx.offset += write_compressed(f, buf, size);
1374
	obj[0].idx.crc32 = crc32_end(f);
1375
	hashflush(f);
1376
	oidread(&obj->idx.oid, sha1, the_repository->hash_algo);
1377
	return obj;
1378
}
1379

1380
static int delta_pos_compare(const void *_a, const void *_b)
1381
{
1382
	struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1383
	struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1384
	return a->obj_no - b->obj_no;
1385
}
1386

1387
static void fix_unresolved_deltas(struct hashfile *f)
1388
{
1389
	struct ref_delta_entry **sorted_by_pos;
1390
	int i;
1391

1392
	/*
1393
	 * Since many unresolved deltas may well be themselves base objects
1394
	 * for more unresolved deltas, we really want to include the
1395
	 * smallest number of base objects that would cover as much delta
1396
	 * as possible by picking the
1397
	 * trunc deltas first, allowing for other deltas to resolve without
1398
	 * additional base objects.  Since most base objects are to be found
1399
	 * before deltas depending on them, a good heuristic is to start
1400
	 * resolving deltas in the same order as their position in the pack.
1401
	 */
1402
	ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1403
	for (i = 0; i < nr_ref_deltas; i++)
1404
		sorted_by_pos[i] = &ref_deltas[i];
1405
	QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1406

1407
	if (repo_has_promisor_remote(the_repository)) {
1408
		/*
1409
		 * Prefetch the delta bases.
1410
		 */
1411
		struct oid_array to_fetch = OID_ARRAY_INIT;
1412
		for (i = 0; i < nr_ref_deltas; i++) {
1413
			struct ref_delta_entry *d = sorted_by_pos[i];
1414
			if (!oid_object_info_extended(the_repository, &d->oid,
1415
						      NULL,
1416
						      OBJECT_INFO_FOR_PREFETCH))
1417
				continue;
1418
			oid_array_append(&to_fetch, &d->oid);
1419
		}
1420
		promisor_remote_get_direct(the_repository,
1421
					   to_fetch.oid, to_fetch.nr);
1422
		oid_array_clear(&to_fetch);
1423
	}
1424

1425
	for (i = 0; i < nr_ref_deltas; i++) {
1426
		struct ref_delta_entry *d = sorted_by_pos[i];
1427
		enum object_type type;
1428
		void *data;
1429
		unsigned long size;
1430

1431
		if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1432
			continue;
1433
		data = repo_read_object_file(the_repository, &d->oid, &type,
1434
					     &size);
1435
		if (!data)
1436
			continue;
1437

1438
		if (check_object_signature(the_repository, &d->oid, data, size,
1439
					   type) < 0)
1440
			die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1441

1442
		/*
1443
		 * Add this as an object to the objects array and call
1444
		 * threaded_second_pass() (which will pick up the added
1445
		 * object).
1446
		 */
1447
		append_obj_to_pack(f, d->oid.hash, data, size, type);
1448
		free(data);
1449
		threaded_second_pass(NULL);
1450

1451
		display_progress(progress, nr_resolved_deltas);
1452
	}
1453
	free(sorted_by_pos);
1454
}
1455

1456
static const char *derive_filename(const char *pack_name, const char *strip,
1457
				   const char *suffix, struct strbuf *buf)
1458
{
1459
	size_t len;
1460
	if (!strip_suffix(pack_name, strip, &len) || !len ||
1461
	    pack_name[len - 1] != '.')
1462
		die(_("packfile name '%s' does not end with '.%s'"),
1463
		    pack_name, strip);
1464
	strbuf_add(buf, pack_name, len);
1465
	strbuf_addstr(buf, suffix);
1466
	return buf->buf;
1467
}
1468

1469
static void write_special_file(const char *suffix, const char *msg,
1470
			       const char *pack_name, const unsigned char *hash,
1471
			       const char **report)
1472
{
1473
	struct strbuf name_buf = STRBUF_INIT;
1474
	const char *filename;
1475
	int fd;
1476
	int msg_len = strlen(msg);
1477

1478
	if (pack_name)
1479
		filename = derive_filename(pack_name, "pack", suffix, &name_buf);
1480
	else
1481
		filename = odb_pack_name(&name_buf, hash, suffix);
1482

1483
	fd = odb_pack_keep(filename);
1484
	if (fd < 0) {
1485
		if (errno != EEXIST)
1486
			die_errno(_("cannot write %s file '%s'"),
1487
				  suffix, filename);
1488
	} else {
1489
		if (msg_len > 0) {
1490
			write_or_die(fd, msg, msg_len);
1491
			write_or_die(fd, "\n", 1);
1492
		}
1493
		if (close(fd) != 0)
1494
			die_errno(_("cannot close written %s file '%s'"),
1495
				  suffix, filename);
1496
		if (report)
1497
			*report = suffix;
1498
	}
1499
	strbuf_release(&name_buf);
1500
}
1501

1502
static void rename_tmp_packfile(const char **final_name,
1503
				const char *curr_name,
1504
				struct strbuf *name, unsigned char *hash,
1505
				const char *ext, int make_read_only_if_same)
1506
{
1507
	if (*final_name != curr_name) {
1508
		if (!*final_name)
1509
			*final_name = odb_pack_name(name, hash, ext);
1510
		if (finalize_object_file(curr_name, *final_name))
1511
			die(_("unable to rename temporary '*.%s' file to '%s'"),
1512
			    ext, *final_name);
1513
	} else if (make_read_only_if_same) {
1514
		chmod(*final_name, 0444);
1515
	}
1516
}
1517

1518
static void final(const char *final_pack_name, const char *curr_pack_name,
1519
		  const char *final_index_name, const char *curr_index_name,
1520
		  const char *final_rev_index_name, const char *curr_rev_index_name,
1521
		  const char *keep_msg, const char *promisor_msg,
1522
		  unsigned char *hash)
1523
{
1524
	const char *report = "pack";
1525
	struct strbuf pack_name = STRBUF_INIT;
1526
	struct strbuf index_name = STRBUF_INIT;
1527
	struct strbuf rev_index_name = STRBUF_INIT;
1528

1529
	if (!from_stdin) {
1530
		close(input_fd);
1531
	} else {
1532
		fsync_component_or_die(FSYNC_COMPONENT_PACK, output_fd, curr_pack_name);
1533
		if (close(output_fd))
1534
			die_errno(_("error while closing pack file"));
1535
	}
1536

1537
	if (keep_msg)
1538
		write_special_file("keep", keep_msg, final_pack_name, hash,
1539
				   &report);
1540
	if (promisor_msg)
1541
		write_special_file("promisor", promisor_msg, final_pack_name,
1542
				   hash, NULL);
1543

1544
	rename_tmp_packfile(&final_pack_name, curr_pack_name, &pack_name,
1545
			    hash, "pack", from_stdin);
1546
	if (curr_rev_index_name)
1547
		rename_tmp_packfile(&final_rev_index_name, curr_rev_index_name,
1548
				    &rev_index_name, hash, "rev", 1);
1549
	rename_tmp_packfile(&final_index_name, curr_index_name, &index_name,
1550
			    hash, "idx", 1);
1551

1552
	if (do_fsck_object) {
1553
		struct packed_git *p;
1554
		p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1555
		if (p)
1556
			install_packed_git(the_repository, p);
1557
	}
1558

1559
	if (!from_stdin) {
1560
		printf("%s\n", hash_to_hex(hash));
1561
	} else {
1562
		struct strbuf buf = STRBUF_INIT;
1563

1564
		strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1565
		write_or_die(1, buf.buf, buf.len);
1566
		strbuf_release(&buf);
1567

1568
		/* Write the last part of the buffer to stdout */
1569
		write_in_full(1, input_buffer + input_offset, input_len);
1570
	}
1571

1572
	strbuf_release(&rev_index_name);
1573
	strbuf_release(&index_name);
1574
	strbuf_release(&pack_name);
1575
}
1576

1577
static int git_index_pack_config(const char *k, const char *v,
1578
				 const struct config_context *ctx, void *cb)
1579
{
1580
	struct pack_idx_option *opts = cb;
1581

1582
	if (!strcmp(k, "pack.indexversion")) {
1583
		opts->version = git_config_int(k, v, ctx->kvi);
1584
		if (opts->version > 2)
1585
			die(_("bad pack.indexVersion=%"PRIu32), opts->version);
1586
		return 0;
1587
	}
1588
	if (!strcmp(k, "pack.threads")) {
1589
		nr_threads = git_config_int(k, v, ctx->kvi);
1590
		if (nr_threads < 0)
1591
			die(_("invalid number of threads specified (%d)"),
1592
			    nr_threads);
1593
		if (!HAVE_THREADS && nr_threads != 1) {
1594
			warning(_("no threads support, ignoring %s"), k);
1595
			nr_threads = 1;
1596
		}
1597
		return 0;
1598
	}
1599
	if (!strcmp(k, "pack.writereverseindex")) {
1600
		if (git_config_bool(k, v))
1601
			opts->flags |= WRITE_REV;
1602
		else
1603
			opts->flags &= ~WRITE_REV;
1604
	}
1605
	return git_default_config(k, v, ctx, cb);
1606
}
1607

1608
static int cmp_uint32(const void *a_, const void *b_)
1609
{
1610
	uint32_t a = *((uint32_t *)a_);
1611
	uint32_t b = *((uint32_t *)b_);
1612

1613
	return (a < b) ? -1 : (a != b);
1614
}
1615

1616
static void read_v2_anomalous_offsets(struct packed_git *p,
1617
				      struct pack_idx_option *opts)
1618
{
1619
	const uint32_t *idx1, *idx2;
1620
	uint32_t i;
1621

1622
	/* The address of the 4-byte offset table */
1623
	idx1 = (((const uint32_t *)((const uint8_t *)p->index_data + p->crc_offset))
1624
		+ (size_t)p->num_objects /* CRC32 table */
1625
		);
1626

1627
	/* The address of the 8-byte offset table */
1628
	idx2 = idx1 + p->num_objects;
1629

1630
	for (i = 0; i < p->num_objects; i++) {
1631
		uint32_t off = ntohl(idx1[i]);
1632
		if (!(off & 0x80000000))
1633
			continue;
1634
		off = off & 0x7fffffff;
1635
		check_pack_index_ptr(p, &idx2[off * 2]);
1636
		if (idx2[off * 2])
1637
			continue;
1638
		/*
1639
		 * The real offset is ntohl(idx2[off * 2]) in high 4
1640
		 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1641
		 * octets.  But idx2[off * 2] is Zero!!!
1642
		 */
1643
		ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1644
		opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1645
	}
1646

1647
	QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1648
}
1649

1650
static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1651
{
1652
	struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1653

1654
	if (!p)
1655
		die(_("Cannot open existing pack file '%s'"), pack_name);
1656
	if (open_pack_index(p))
1657
		die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1658

1659
	/* Read the attributes from the existing idx file */
1660
	opts->version = p->index_version;
1661

1662
	if (opts->version == 2)
1663
		read_v2_anomalous_offsets(p, opts);
1664

1665
	/*
1666
	 * Get rid of the idx file as we do not need it anymore.
1667
	 * NEEDSWORK: extract this bit from free_pack_by_name() in
1668
	 * object-file.c, perhaps?  It shouldn't matter very much as we
1669
	 * know we haven't installed this pack (hence we never have
1670
	 * read anything from it).
1671
	 */
1672
	close_pack_index(p);
1673
	free(p);
1674
}
1675

1676
static void show_pack_info(int stat_only)
1677
{
1678
	int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1679
	unsigned long *chain_histogram = NULL;
1680

1681
	if (deepest_delta)
1682
		CALLOC_ARRAY(chain_histogram, deepest_delta);
1683

1684
	for (i = 0; i < nr_objects; i++) {
1685
		struct object_entry *obj = &objects[i];
1686

1687
		if (is_delta_type(obj->type))
1688
			chain_histogram[obj_stat[i].delta_depth - 1]++;
1689
		if (stat_only)
1690
			continue;
1691
		printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1692
		       oid_to_hex(&obj->idx.oid),
1693
		       type_name(obj->real_type), (uintmax_t)obj->size,
1694
		       (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1695
		       (uintmax_t)obj->idx.offset);
1696
		if (is_delta_type(obj->type)) {
1697
			struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1698
			printf(" %u %s", obj_stat[i].delta_depth,
1699
			       oid_to_hex(&bobj->idx.oid));
1700
		}
1701
		putchar('\n');
1702
	}
1703

1704
	if (baseobjects)
1705
		printf_ln(Q_("non delta: %d object",
1706
			     "non delta: %d objects",
1707
			     baseobjects),
1708
			  baseobjects);
1709
	for (i = 0; i < deepest_delta; i++) {
1710
		if (!chain_histogram[i])
1711
			continue;
1712
		printf_ln(Q_("chain length = %d: %lu object",
1713
			     "chain length = %d: %lu objects",
1714
			     chain_histogram[i]),
1715
			  i + 1,
1716
			  chain_histogram[i]);
1717
	}
1718
	free(chain_histogram);
1719
}
1720

1721
int cmd_index_pack(int argc, const char **argv, const char *prefix)
1722
{
1723
	int i, fix_thin_pack = 0, verify = 0, stat_only = 0, rev_index;
1724
	const char *curr_index;
1725
	const char *curr_rev_index = NULL;
1726
	const char *index_name = NULL, *pack_name = NULL, *rev_index_name = NULL;
1727
	const char *keep_msg = NULL;
1728
	const char *promisor_msg = NULL;
1729
	struct strbuf index_name_buf = STRBUF_INIT;
1730
	struct strbuf rev_index_name_buf = STRBUF_INIT;
1731
	struct pack_idx_entry **idx_objects;
1732
	struct pack_idx_option opts;
1733
	unsigned char pack_hash[GIT_MAX_RAWSZ];
1734
	unsigned foreign_nr = 1;	/* zero is a "good" value, assume bad */
1735
	int report_end_of_input = 0;
1736
	int hash_algo = 0;
1737

1738
	/*
1739
	 * index-pack never needs to fetch missing objects except when
1740
	 * REF_DELTA bases are missing (which are explicitly handled). It only
1741
	 * accesses the repo to do hash collision checks and to check which
1742
	 * REF_DELTA bases need to be fetched.
1743
	 */
1744
	fetch_if_missing = 0;
1745

1746
	if (argc == 2 && !strcmp(argv[1], "-h"))
1747
		usage(index_pack_usage);
1748

1749
	disable_replace_refs();
1750
	fsck_options.walk = mark_link;
1751

1752
	reset_pack_idx_option(&opts);
1753
	opts.flags |= WRITE_REV;
1754
	git_config(git_index_pack_config, &opts);
1755
	if (prefix && chdir(prefix))
1756
		die(_("Cannot come back to cwd"));
1757

1758
	if (git_env_bool(GIT_TEST_NO_WRITE_REV_INDEX, 0))
1759
		rev_index = 0;
1760
	else
1761
		rev_index = !!(opts.flags & (WRITE_REV_VERIFY | WRITE_REV));
1762

1763
	for (i = 1; i < argc; i++) {
1764
		const char *arg = argv[i];
1765

1766
		if (*arg == '-') {
1767
			if (!strcmp(arg, "--stdin")) {
1768
				from_stdin = 1;
1769
			} else if (!strcmp(arg, "--fix-thin")) {
1770
				fix_thin_pack = 1;
1771
			} else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1772
				strict = 1;
1773
				do_fsck_object = 1;
1774
				fsck_set_msg_types(&fsck_options, arg);
1775
			} else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1776
				strict = 1;
1777
				check_self_contained_and_connected = 1;
1778
			} else if (skip_to_optional_arg(arg, "--fsck-objects", &arg)) {
1779
				do_fsck_object = 1;
1780
				fsck_set_msg_types(&fsck_options, arg);
1781
			} else if (!strcmp(arg, "--verify")) {
1782
				verify = 1;
1783
			} else if (!strcmp(arg, "--verify-stat")) {
1784
				verify = 1;
1785
				show_stat = 1;
1786
			} else if (!strcmp(arg, "--verify-stat-only")) {
1787
				verify = 1;
1788
				show_stat = 1;
1789
				stat_only = 1;
1790
			} else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1791
				; /* nothing to do */
1792
			} else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1793
				; /* already parsed */
1794
			} else if (starts_with(arg, "--threads=")) {
1795
				char *end;
1796
				nr_threads = strtoul(arg+10, &end, 0);
1797
				if (!arg[10] || *end || nr_threads < 0)
1798
					usage(index_pack_usage);
1799
				if (!HAVE_THREADS && nr_threads != 1) {
1800
					warning(_("no threads support, ignoring %s"), arg);
1801
					nr_threads = 1;
1802
				}
1803
			} else if (starts_with(arg, "--pack_header=")) {
1804
				struct pack_header *hdr;
1805
				char *c;
1806

1807
				hdr = (struct pack_header *)input_buffer;
1808
				hdr->hdr_signature = htonl(PACK_SIGNATURE);
1809
				hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1810
				if (*c != ',')
1811
					die(_("bad %s"), arg);
1812
				hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1813
				if (*c)
1814
					die(_("bad %s"), arg);
1815
				input_len = sizeof(*hdr);
1816
			} else if (!strcmp(arg, "-v")) {
1817
				verbose = 1;
1818
			} else if (!strcmp(arg, "--progress-title")) {
1819
				if (progress_title || (i+1) >= argc)
1820
					usage(index_pack_usage);
1821
				progress_title = argv[++i];
1822
			} else if (!strcmp(arg, "--show-resolving-progress")) {
1823
				show_resolving_progress = 1;
1824
			} else if (!strcmp(arg, "--report-end-of-input")) {
1825
				report_end_of_input = 1;
1826
			} else if (!strcmp(arg, "-o")) {
1827
				if (index_name || (i+1) >= argc)
1828
					usage(index_pack_usage);
1829
				index_name = argv[++i];
1830
			} else if (starts_with(arg, "--index-version=")) {
1831
				char *c;
1832
				opts.version = strtoul(arg + 16, &c, 10);
1833
				if (opts.version > 2)
1834
					die(_("bad %s"), arg);
1835
				if (*c == ',')
1836
					opts.off32_limit = strtoul(c+1, &c, 0);
1837
				if (*c || opts.off32_limit & 0x80000000)
1838
					die(_("bad %s"), arg);
1839
			} else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1840
				max_input_size = strtoumax(arg, NULL, 10);
1841
			} else if (skip_prefix(arg, "--object-format=", &arg)) {
1842
				hash_algo = hash_algo_by_name(arg);
1843
				if (hash_algo == GIT_HASH_UNKNOWN)
1844
					die(_("unknown hash algorithm '%s'"), arg);
1845
				repo_set_hash_algo(the_repository, hash_algo);
1846
			} else if (!strcmp(arg, "--rev-index")) {
1847
				rev_index = 1;
1848
			} else if (!strcmp(arg, "--no-rev-index")) {
1849
				rev_index = 0;
1850
			} else
1851
				usage(index_pack_usage);
1852
			continue;
1853
		}
1854

1855
		if (pack_name)
1856
			usage(index_pack_usage);
1857
		pack_name = arg;
1858
	}
1859

1860
	if (!pack_name && !from_stdin)
1861
		usage(index_pack_usage);
1862
	if (fix_thin_pack && !from_stdin)
1863
		die(_("the option '%s' requires '%s'"), "--fix-thin", "--stdin");
1864
	if (from_stdin && !startup_info->have_repository)
1865
		die(_("--stdin requires a git repository"));
1866
	if (from_stdin && hash_algo)
1867
		die(_("options '%s' and '%s' cannot be used together"), "--object-format", "--stdin");
1868
	if (!index_name && pack_name)
1869
		index_name = derive_filename(pack_name, "pack", "idx", &index_name_buf);
1870

1871
	opts.flags &= ~(WRITE_REV | WRITE_REV_VERIFY);
1872
	if (rev_index) {
1873
		opts.flags |= verify ? WRITE_REV_VERIFY : WRITE_REV;
1874
		if (index_name)
1875
			rev_index_name = derive_filename(index_name,
1876
							 "idx", "rev",
1877
							 &rev_index_name_buf);
1878
	}
1879

1880
	if (verify) {
1881
		if (!index_name)
1882
			die(_("--verify with no packfile name given"));
1883
		read_idx_option(&opts, index_name);
1884
		opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1885
	}
1886
	if (strict)
1887
		opts.flags |= WRITE_IDX_STRICT;
1888

1889
	if (HAVE_THREADS && !nr_threads) {
1890
		nr_threads = online_cpus();
1891
		/*
1892
		 * Experiments show that going above 20 threads doesn't help,
1893
		 * no matter how many cores you have. Below that, we tend to
1894
		 * max at half the number of online_cpus(), presumably because
1895
		 * half of those are hyperthreads rather than full cores. We'll
1896
		 * never reduce the level below "3", though, to match a
1897
		 * historical value that nobody complained about.
1898
		 */
1899
		if (nr_threads < 4)
1900
			; /* too few cores to consider capping */
1901
		else if (nr_threads < 6)
1902
			nr_threads = 3; /* historic cap */
1903
		else if (nr_threads < 40)
1904
			nr_threads /= 2;
1905
		else
1906
			nr_threads = 20; /* hard cap */
1907
	}
1908

1909
	curr_pack = open_pack_file(pack_name);
1910
	parse_pack_header();
1911
	CALLOC_ARRAY(objects, st_add(nr_objects, 1));
1912
	if (show_stat)
1913
		CALLOC_ARRAY(obj_stat, st_add(nr_objects, 1));
1914
	CALLOC_ARRAY(ofs_deltas, nr_objects);
1915
	parse_pack_objects(pack_hash);
1916
	if (report_end_of_input)
1917
		write_in_full(2, "\0", 1);
1918
	resolve_deltas();
1919
	conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1920
	free(ofs_deltas);
1921
	free(ref_deltas);
1922
	if (strict)
1923
		foreign_nr = check_objects();
1924

1925
	if (show_stat)
1926
		show_pack_info(stat_only);
1927

1928
	ALLOC_ARRAY(idx_objects, nr_objects);
1929
	for (i = 0; i < nr_objects; i++)
1930
		idx_objects[i] = &objects[i].idx;
1931
	curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1932
	if (rev_index)
1933
		curr_rev_index = write_rev_file(rev_index_name, idx_objects,
1934
						nr_objects, pack_hash,
1935
						opts.flags);
1936
	free(idx_objects);
1937

1938
	if (!verify)
1939
		final(pack_name, curr_pack,
1940
		      index_name, curr_index,
1941
		      rev_index_name, curr_rev_index,
1942
		      keep_msg, promisor_msg,
1943
		      pack_hash);
1944
	else
1945
		close(input_fd);
1946

1947
	if (do_fsck_object && fsck_finish(&fsck_options))
1948
		die(_("fsck error in pack objects"));
1949

1950
	free(opts.anomaly);
1951
	free(objects);
1952
	strbuf_release(&index_name_buf);
1953
	strbuf_release(&rev_index_name_buf);
1954
	if (!pack_name)
1955
		free((void *) curr_pack);
1956
	if (!index_name)
1957
		free((void *) curr_index);
1958
	if (!rev_index_name)
1959
		free((void *) curr_rev_index);
1960

1961
	/*
1962
	 * Let the caller know this pack is not self contained
1963
	 */
1964
	if (check_self_contained_and_connected && foreign_nr)
1965
		return 1;
1966

1967
	return 0;
1968
}
1969