Sync with maint
[git/git.git] / builtin / index-pack.c
1 #include "builtin.h"
2 #include "delta.h"
3 #include "pack.h"
4 #include "csum-file.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "progress.h"
10 #include "fsck.h"
11 #include "exec_cmd.h"
12 #include "streaming.h"
13 #include "thread-utils.h"
14
15 static const char index_pack_usage[] =
16 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
17
18 struct object_entry {
19 struct pack_idx_entry idx;
20 unsigned long size;
21 unsigned int hdr_size;
22 enum object_type type;
23 enum object_type real_type;
24 unsigned delta_depth;
25 int base_object_no;
26 };
27
28 union delta_base {
29 unsigned char sha1[20];
30 off_t offset;
31 };
32
33 struct base_data {
34 struct base_data *base;
35 struct base_data *child;
36 struct object_entry *obj;
37 void *data;
38 unsigned long size;
39 int ref_first, ref_last;
40 int ofs_first, ofs_last;
41 };
42
43 #if !defined(NO_PTHREADS) && defined(NO_THREAD_SAFE_PREAD)
44 /* pread() emulation is not thread-safe. Disable threading. */
45 #define NO_PTHREADS
46 #endif
47
48 struct thread_local {
49 #ifndef NO_PTHREADS
50 pthread_t thread;
51 #endif
52 struct base_data *base_cache;
53 size_t base_cache_used;
54 };
55
56 /*
57 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
58 * to memcmp() only the first 20 bytes.
59 */
60 #define UNION_BASE_SZ 20
61
62 #define FLAG_LINK (1u<<20)
63 #define FLAG_CHECKED (1u<<21)
64
65 struct delta_entry {
66 union delta_base base;
67 int obj_no;
68 };
69
70 static struct object_entry *objects;
71 static struct delta_entry *deltas;
72 static struct thread_local nothread_data;
73 static int nr_objects;
74 static int nr_deltas;
75 static int nr_resolved_deltas;
76 static int nr_threads;
77
78 static int from_stdin;
79 static int strict;
80 static int do_fsck_object;
81 static int verbose;
82 static int show_stat;
83 static int check_self_contained_and_connected;
84
85 static struct progress *progress;
86
87 /* We always read in 4kB chunks. */
88 static unsigned char input_buffer[4096];
89 static unsigned int input_offset, input_len;
90 static off_t consumed_bytes;
91 static unsigned deepest_delta;
92 static git_SHA_CTX input_ctx;
93 static uint32_t input_crc32;
94 static int input_fd, output_fd, pack_fd;
95
96 #ifndef NO_PTHREADS
97
98 static struct thread_local *thread_data;
99 static int nr_dispatched;
100 static int threads_active;
101
102 static pthread_mutex_t read_mutex;
103 #define read_lock() lock_mutex(&read_mutex)
104 #define read_unlock() unlock_mutex(&read_mutex)
105
106 static pthread_mutex_t counter_mutex;
107 #define counter_lock() lock_mutex(&counter_mutex)
108 #define counter_unlock() unlock_mutex(&counter_mutex)
109
110 static pthread_mutex_t work_mutex;
111 #define work_lock() lock_mutex(&work_mutex)
112 #define work_unlock() unlock_mutex(&work_mutex)
113
114 static pthread_mutex_t deepest_delta_mutex;
115 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
116 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
117
118 static pthread_key_t key;
119
120 static inline void lock_mutex(pthread_mutex_t *mutex)
121 {
122 if (threads_active)
123 pthread_mutex_lock(mutex);
124 }
125
126 static inline void unlock_mutex(pthread_mutex_t *mutex)
127 {
128 if (threads_active)
129 pthread_mutex_unlock(mutex);
130 }
131
132 /*
133 * Mutex and conditional variable can't be statically-initialized on Windows.
134 */
135 static void init_thread(void)
136 {
137 init_recursive_mutex(&read_mutex);
138 pthread_mutex_init(&counter_mutex, NULL);
139 pthread_mutex_init(&work_mutex, NULL);
140 if (show_stat)
141 pthread_mutex_init(&deepest_delta_mutex, NULL);
142 pthread_key_create(&key, NULL);
143 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
144 threads_active = 1;
145 }
146
147 static void cleanup_thread(void)
148 {
149 if (!threads_active)
150 return;
151 threads_active = 0;
152 pthread_mutex_destroy(&read_mutex);
153 pthread_mutex_destroy(&counter_mutex);
154 pthread_mutex_destroy(&work_mutex);
155 if (show_stat)
156 pthread_mutex_destroy(&deepest_delta_mutex);
157 pthread_key_delete(key);
158 free(thread_data);
159 }
160
161 #else
162
163 #define read_lock()
164 #define read_unlock()
165
166 #define counter_lock()
167 #define counter_unlock()
168
169 #define work_lock()
170 #define work_unlock()
171
172 #define deepest_delta_lock()
173 #define deepest_delta_unlock()
174
175 #endif
176
177
178 static int mark_link(struct object *obj, int type, void *data)
179 {
180 if (!obj)
181 return -1;
182
183 if (type != OBJ_ANY && obj->type != type)
184 die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
185
186 obj->flags |= FLAG_LINK;
187 return 0;
188 }
189
190 /* The content of each linked object must have been checked
191 or it must be already present in the object database */
192 static unsigned check_object(struct object *obj)
193 {
194 if (!obj)
195 return 0;
196
197 if (!(obj->flags & FLAG_LINK))
198 return 0;
199
200 if (!(obj->flags & FLAG_CHECKED)) {
201 unsigned long size;
202 int type = sha1_object_info(obj->sha1, &size);
203 if (type != obj->type || type <= 0)
204 die(_("object of unexpected type"));
205 obj->flags |= FLAG_CHECKED;
206 return 1;
207 }
208
209 return 0;
210 }
211
212 static unsigned check_objects(void)
213 {
214 unsigned i, max, foreign_nr = 0;
215
216 max = get_max_object_index();
217 for (i = 0; i < max; i++)
218 foreign_nr += check_object(get_indexed_object(i));
219 return foreign_nr;
220 }
221
222
223 /* Discard current buffer used content. */
224 static void flush(void)
225 {
226 if (input_offset) {
227 if (output_fd >= 0)
228 write_or_die(output_fd, input_buffer, input_offset);
229 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
230 memmove(input_buffer, input_buffer + input_offset, input_len);
231 input_offset = 0;
232 }
233 }
234
235 /*
236 * Make sure at least "min" bytes are available in the buffer, and
237 * return the pointer to the buffer.
238 */
239 static void *fill(int min)
240 {
241 if (min <= input_len)
242 return input_buffer + input_offset;
243 if (min > sizeof(input_buffer))
244 die(Q_("cannot fill %d byte",
245 "cannot fill %d bytes",
246 min),
247 min);
248 flush();
249 do {
250 ssize_t ret = xread(input_fd, input_buffer + input_len,
251 sizeof(input_buffer) - input_len);
252 if (ret <= 0) {
253 if (!ret)
254 die(_("early EOF"));
255 die_errno(_("read error on input"));
256 }
257 input_len += ret;
258 if (from_stdin)
259 display_throughput(progress, consumed_bytes + input_len);
260 } while (input_len < min);
261 return input_buffer;
262 }
263
264 static void use(int bytes)
265 {
266 if (bytes > input_len)
267 die(_("used more bytes than were available"));
268 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
269 input_len -= bytes;
270 input_offset += bytes;
271
272 /* make sure off_t is sufficiently large not to wrap */
273 if (signed_add_overflows(consumed_bytes, bytes))
274 die(_("pack too large for current definition of off_t"));
275 consumed_bytes += bytes;
276 }
277
278 static const char *open_pack_file(const char *pack_name)
279 {
280 if (from_stdin) {
281 input_fd = 0;
282 if (!pack_name) {
283 static char tmp_file[PATH_MAX];
284 output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
285 "pack/tmp_pack_XXXXXX");
286 pack_name = xstrdup(tmp_file);
287 } else
288 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
289 if (output_fd < 0)
290 die_errno(_("unable to create '%s'"), pack_name);
291 pack_fd = output_fd;
292 } else {
293 input_fd = open(pack_name, O_RDONLY);
294 if (input_fd < 0)
295 die_errno(_("cannot open packfile '%s'"), pack_name);
296 output_fd = -1;
297 pack_fd = input_fd;
298 }
299 git_SHA1_Init(&input_ctx);
300 return pack_name;
301 }
302
303 static void parse_pack_header(void)
304 {
305 struct pack_header *hdr = fill(sizeof(struct pack_header));
306
307 /* Header consistency check */
308 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
309 die(_("pack signature mismatch"));
310 if (!pack_version_ok(hdr->hdr_version))
311 die(_("pack version %"PRIu32" unsupported"),
312 ntohl(hdr->hdr_version));
313
314 nr_objects = ntohl(hdr->hdr_entries);
315 use(sizeof(struct pack_header));
316 }
317
318 static NORETURN void bad_object(unsigned long offset, const char *format,
319 ...) __attribute__((format (printf, 2, 3)));
320
321 static NORETURN void bad_object(unsigned long offset, const char *format, ...)
322 {
323 va_list params;
324 char buf[1024];
325
326 va_start(params, format);
327 vsnprintf(buf, sizeof(buf), format, params);
328 va_end(params);
329 die(_("pack has bad object at offset %lu: %s"), offset, buf);
330 }
331
332 static inline struct thread_local *get_thread_data(void)
333 {
334 #ifndef NO_PTHREADS
335 if (threads_active)
336 return pthread_getspecific(key);
337 assert(!threads_active &&
338 "This should only be reached when all threads are gone");
339 #endif
340 return &nothread_data;
341 }
342
343 #ifndef NO_PTHREADS
344 static void set_thread_data(struct thread_local *data)
345 {
346 if (threads_active)
347 pthread_setspecific(key, data);
348 }
349 #endif
350
351 static struct base_data *alloc_base_data(void)
352 {
353 struct base_data *base = xmalloc(sizeof(struct base_data));
354 memset(base, 0, sizeof(*base));
355 base->ref_last = -1;
356 base->ofs_last = -1;
357 return base;
358 }
359
360 static void free_base_data(struct base_data *c)
361 {
362 if (c->data) {
363 free(c->data);
364 c->data = NULL;
365 get_thread_data()->base_cache_used -= c->size;
366 }
367 }
368
369 static void prune_base_data(struct base_data *retain)
370 {
371 struct base_data *b;
372 struct thread_local *data = get_thread_data();
373 for (b = data->base_cache;
374 data->base_cache_used > delta_base_cache_limit && b;
375 b = b->child) {
376 if (b->data && b != retain)
377 free_base_data(b);
378 }
379 }
380
381 static void link_base_data(struct base_data *base, struct base_data *c)
382 {
383 if (base)
384 base->child = c;
385 else
386 get_thread_data()->base_cache = c;
387
388 c->base = base;
389 c->child = NULL;
390 if (c->data)
391 get_thread_data()->base_cache_used += c->size;
392 prune_base_data(c);
393 }
394
395 static void unlink_base_data(struct base_data *c)
396 {
397 struct base_data *base = c->base;
398 if (base)
399 base->child = NULL;
400 else
401 get_thread_data()->base_cache = NULL;
402 free_base_data(c);
403 }
404
405 static int is_delta_type(enum object_type type)
406 {
407 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
408 }
409
410 static void *unpack_entry_data(unsigned long offset, unsigned long size,
411 enum object_type type, unsigned char *sha1)
412 {
413 static char fixed_buf[8192];
414 int status;
415 git_zstream stream;
416 void *buf;
417 git_SHA_CTX c;
418 char hdr[32];
419 int hdrlen;
420
421 if (!is_delta_type(type)) {
422 hdrlen = sprintf(hdr, "%s %lu", typename(type), size) + 1;
423 git_SHA1_Init(&c);
424 git_SHA1_Update(&c, hdr, hdrlen);
425 } else
426 sha1 = NULL;
427 if (type == OBJ_BLOB && size > big_file_threshold)
428 buf = fixed_buf;
429 else
430 buf = xmalloc(size);
431
432 memset(&stream, 0, sizeof(stream));
433 git_inflate_init(&stream);
434 stream.next_out = buf;
435 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
436
437 do {
438 unsigned char *last_out = stream.next_out;
439 stream.next_in = fill(1);
440 stream.avail_in = input_len;
441 status = git_inflate(&stream, 0);
442 use(input_len - stream.avail_in);
443 if (sha1)
444 git_SHA1_Update(&c, last_out, stream.next_out - last_out);
445 if (buf == fixed_buf) {
446 stream.next_out = buf;
447 stream.avail_out = sizeof(fixed_buf);
448 }
449 } while (status == Z_OK);
450 if (stream.total_out != size || status != Z_STREAM_END)
451 bad_object(offset, _("inflate returned %d"), status);
452 git_inflate_end(&stream);
453 if (sha1)
454 git_SHA1_Final(sha1, &c);
455 return buf == fixed_buf ? NULL : buf;
456 }
457
458 static void *unpack_raw_entry(struct object_entry *obj,
459 union delta_base *delta_base,
460 unsigned char *sha1)
461 {
462 unsigned char *p;
463 unsigned long size, c;
464 off_t base_offset;
465 unsigned shift;
466 void *data;
467
468 obj->idx.offset = consumed_bytes;
469 input_crc32 = crc32(0, NULL, 0);
470
471 p = fill(1);
472 c = *p;
473 use(1);
474 obj->type = (c >> 4) & 7;
475 size = (c & 15);
476 shift = 4;
477 while (c & 0x80) {
478 p = fill(1);
479 c = *p;
480 use(1);
481 size += (c & 0x7f) << shift;
482 shift += 7;
483 }
484 obj->size = size;
485
486 switch (obj->type) {
487 case OBJ_REF_DELTA:
488 hashcpy(delta_base->sha1, fill(20));
489 use(20);
490 break;
491 case OBJ_OFS_DELTA:
492 memset(delta_base, 0, sizeof(*delta_base));
493 p = fill(1);
494 c = *p;
495 use(1);
496 base_offset = c & 127;
497 while (c & 128) {
498 base_offset += 1;
499 if (!base_offset || MSB(base_offset, 7))
500 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
501 p = fill(1);
502 c = *p;
503 use(1);
504 base_offset = (base_offset << 7) + (c & 127);
505 }
506 delta_base->offset = obj->idx.offset - base_offset;
507 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
508 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
509 break;
510 case OBJ_COMMIT:
511 case OBJ_TREE:
512 case OBJ_BLOB:
513 case OBJ_TAG:
514 break;
515 default:
516 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
517 }
518 obj->hdr_size = consumed_bytes - obj->idx.offset;
519
520 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
521 obj->idx.crc32 = input_crc32;
522 return data;
523 }
524
525 static void *unpack_data(struct object_entry *obj,
526 int (*consume)(const unsigned char *, unsigned long, void *),
527 void *cb_data)
528 {
529 off_t from = obj[0].idx.offset + obj[0].hdr_size;
530 unsigned long len = obj[1].idx.offset - from;
531 unsigned char *data, *inbuf;
532 git_zstream stream;
533 int status;
534
535 data = xmalloc(consume ? 64*1024 : obj->size);
536 inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
537
538 memset(&stream, 0, sizeof(stream));
539 git_inflate_init(&stream);
540 stream.next_out = data;
541 stream.avail_out = consume ? 64*1024 : obj->size;
542
543 do {
544 ssize_t n = (len < 64*1024) ? len : 64*1024;
545 n = pread(pack_fd, inbuf, n, from);
546 if (n < 0)
547 die_errno(_("cannot pread pack file"));
548 if (!n)
549 die(Q_("premature end of pack file, %lu byte missing",
550 "premature end of pack file, %lu bytes missing",
551 len),
552 len);
553 from += n;
554 len -= n;
555 stream.next_in = inbuf;
556 stream.avail_in = n;
557 if (!consume)
558 status = git_inflate(&stream, 0);
559 else {
560 do {
561 status = git_inflate(&stream, 0);
562 if (consume(data, stream.next_out - data, cb_data)) {
563 free(inbuf);
564 free(data);
565 return NULL;
566 }
567 stream.next_out = data;
568 stream.avail_out = 64*1024;
569 } while (status == Z_OK && stream.avail_in);
570 }
571 } while (len && status == Z_OK && !stream.avail_in);
572
573 /* This has been inflated OK when first encountered, so... */
574 if (status != Z_STREAM_END || stream.total_out != obj->size)
575 die(_("serious inflate inconsistency"));
576
577 git_inflate_end(&stream);
578 free(inbuf);
579 if (consume) {
580 free(data);
581 data = NULL;
582 }
583 return data;
584 }
585
586 static void *get_data_from_pack(struct object_entry *obj)
587 {
588 return unpack_data(obj, NULL, NULL);
589 }
590
591 static int compare_delta_bases(const union delta_base *base1,
592 const union delta_base *base2,
593 enum object_type type1,
594 enum object_type type2)
595 {
596 int cmp = type1 - type2;
597 if (cmp)
598 return cmp;
599 return memcmp(base1, base2, UNION_BASE_SZ);
600 }
601
602 static int find_delta(const union delta_base *base, enum object_type type)
603 {
604 int first = 0, last = nr_deltas;
605
606 while (first < last) {
607 int next = (first + last) / 2;
608 struct delta_entry *delta = &deltas[next];
609 int cmp;
610
611 cmp = compare_delta_bases(base, &delta->base,
612 type, objects[delta->obj_no].type);
613 if (!cmp)
614 return next;
615 if (cmp < 0) {
616 last = next;
617 continue;
618 }
619 first = next+1;
620 }
621 return -first-1;
622 }
623
624 static void find_delta_children(const union delta_base *base,
625 int *first_index, int *last_index,
626 enum object_type type)
627 {
628 int first = find_delta(base, type);
629 int last = first;
630 int end = nr_deltas - 1;
631
632 if (first < 0) {
633 *first_index = 0;
634 *last_index = -1;
635 return;
636 }
637 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
638 --first;
639 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
640 ++last;
641 *first_index = first;
642 *last_index = last;
643 }
644
645 struct compare_data {
646 struct object_entry *entry;
647 struct git_istream *st;
648 unsigned char *buf;
649 unsigned long buf_size;
650 };
651
652 static int compare_objects(const unsigned char *buf, unsigned long size,
653 void *cb_data)
654 {
655 struct compare_data *data = cb_data;
656
657 if (data->buf_size < size) {
658 free(data->buf);
659 data->buf = xmalloc(size);
660 data->buf_size = size;
661 }
662
663 while (size) {
664 ssize_t len = read_istream(data->st, data->buf, size);
665 if (len == 0)
666 die(_("SHA1 COLLISION FOUND WITH %s !"),
667 sha1_to_hex(data->entry->idx.sha1));
668 if (len < 0)
669 die(_("unable to read %s"),
670 sha1_to_hex(data->entry->idx.sha1));
671 if (memcmp(buf, data->buf, len))
672 die(_("SHA1 COLLISION FOUND WITH %s !"),
673 sha1_to_hex(data->entry->idx.sha1));
674 size -= len;
675 buf += len;
676 }
677 return 0;
678 }
679
680 static int check_collison(struct object_entry *entry)
681 {
682 struct compare_data data;
683 enum object_type type;
684 unsigned long size;
685
686 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
687 return -1;
688
689 memset(&data, 0, sizeof(data));
690 data.entry = entry;
691 data.st = open_istream(entry->idx.sha1, &type, &size, NULL);
692 if (!data.st)
693 return -1;
694 if (size != entry->size || type != entry->type)
695 die(_("SHA1 COLLISION FOUND WITH %s !"),
696 sha1_to_hex(entry->idx.sha1));
697 unpack_data(entry, compare_objects, &data);
698 close_istream(data.st);
699 free(data.buf);
700 return 0;
701 }
702
703 static void sha1_object(const void *data, struct object_entry *obj_entry,
704 unsigned long size, enum object_type type,
705 const unsigned char *sha1)
706 {
707 void *new_data = NULL;
708 int collision_test_needed;
709
710 assert(data || obj_entry);
711
712 read_lock();
713 collision_test_needed = has_sha1_file(sha1);
714 read_unlock();
715
716 if (collision_test_needed && !data) {
717 read_lock();
718 if (!check_collison(obj_entry))
719 collision_test_needed = 0;
720 read_unlock();
721 }
722 if (collision_test_needed) {
723 void *has_data;
724 enum object_type has_type;
725 unsigned long has_size;
726 read_lock();
727 has_type = sha1_object_info(sha1, &has_size);
728 if (has_type != type || has_size != size)
729 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
730 has_data = read_sha1_file(sha1, &has_type, &has_size);
731 read_unlock();
732 if (!data)
733 data = new_data = get_data_from_pack(obj_entry);
734 if (!has_data)
735 die(_("cannot read existing object %s"), sha1_to_hex(sha1));
736 if (size != has_size || type != has_type ||
737 memcmp(data, has_data, size) != 0)
738 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
739 free(has_data);
740 }
741
742 if (strict) {
743 read_lock();
744 if (type == OBJ_BLOB) {
745 struct blob *blob = lookup_blob(sha1);
746 if (blob)
747 blob->object.flags |= FLAG_CHECKED;
748 else
749 die(_("invalid blob object %s"), sha1_to_hex(sha1));
750 } else {
751 struct object *obj;
752 int eaten;
753 void *buf = (void *) data;
754
755 assert(data && "data can only be NULL for large _blobs_");
756
757 /*
758 * we do not need to free the memory here, as the
759 * buf is deleted by the caller.
760 */
761 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
762 if (!obj)
763 die(_("invalid %s"), typename(type));
764 if (do_fsck_object &&
765 fsck_object(obj, 1, fsck_error_function))
766 die(_("Error in object"));
767 if (fsck_walk(obj, mark_link, NULL))
768 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
769
770 if (obj->type == OBJ_TREE) {
771 struct tree *item = (struct tree *) obj;
772 item->buffer = NULL;
773 }
774 if (obj->type == OBJ_COMMIT) {
775 struct commit *commit = (struct commit *) obj;
776 commit->buffer = NULL;
777 }
778 obj->flags |= FLAG_CHECKED;
779 }
780 read_unlock();
781 }
782
783 free(new_data);
784 }
785
786 /*
787 * This function is part of find_unresolved_deltas(). There are two
788 * walkers going in the opposite ways.
789 *
790 * The first one in find_unresolved_deltas() traverses down from
791 * parent node to children, deflating nodes along the way. However,
792 * memory for deflated nodes is limited by delta_base_cache_limit, so
793 * at some point parent node's deflated content may be freed.
794 *
795 * The second walker is this function, which goes from current node up
796 * to top parent if necessary to deflate the node. In normal
797 * situation, its parent node would be already deflated, so it just
798 * needs to apply delta.
799 *
800 * In the worst case scenario, parent node is no longer deflated because
801 * we're running out of delta_base_cache_limit; we need to re-deflate
802 * parents, possibly up to the top base.
803 *
804 * All deflated objects here are subject to be freed if we exceed
805 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
806 * just need to make sure the last node is not freed.
807 */
808 static void *get_base_data(struct base_data *c)
809 {
810 if (!c->data) {
811 struct object_entry *obj = c->obj;
812 struct base_data **delta = NULL;
813 int delta_nr = 0, delta_alloc = 0;
814
815 while (is_delta_type(c->obj->type) && !c->data) {
816 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
817 delta[delta_nr++] = c;
818 c = c->base;
819 }
820 if (!delta_nr) {
821 c->data = get_data_from_pack(obj);
822 c->size = obj->size;
823 get_thread_data()->base_cache_used += c->size;
824 prune_base_data(c);
825 }
826 for (; delta_nr > 0; delta_nr--) {
827 void *base, *raw;
828 c = delta[delta_nr - 1];
829 obj = c->obj;
830 base = get_base_data(c->base);
831 raw = get_data_from_pack(obj);
832 c->data = patch_delta(
833 base, c->base->size,
834 raw, obj->size,
835 &c->size);
836 free(raw);
837 if (!c->data)
838 bad_object(obj->idx.offset, _("failed to apply delta"));
839 get_thread_data()->base_cache_used += c->size;
840 prune_base_data(c);
841 }
842 free(delta);
843 }
844 return c->data;
845 }
846
847 static void resolve_delta(struct object_entry *delta_obj,
848 struct base_data *base, struct base_data *result)
849 {
850 void *base_data, *delta_data;
851
852 delta_obj->real_type = base->obj->real_type;
853 if (show_stat) {
854 delta_obj->delta_depth = base->obj->delta_depth + 1;
855 deepest_delta_lock();
856 if (deepest_delta < delta_obj->delta_depth)
857 deepest_delta = delta_obj->delta_depth;
858 deepest_delta_unlock();
859 }
860 delta_obj->base_object_no = base->obj - objects;
861 delta_data = get_data_from_pack(delta_obj);
862 base_data = get_base_data(base);
863 result->obj = delta_obj;
864 result->data = patch_delta(base_data, base->size,
865 delta_data, delta_obj->size, &result->size);
866 free(delta_data);
867 if (!result->data)
868 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
869 hash_sha1_file(result->data, result->size,
870 typename(delta_obj->real_type), delta_obj->idx.sha1);
871 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
872 delta_obj->idx.sha1);
873 counter_lock();
874 nr_resolved_deltas++;
875 counter_unlock();
876 }
877
878 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
879 struct base_data *prev_base)
880 {
881 if (base->ref_last == -1 && base->ofs_last == -1) {
882 union delta_base base_spec;
883
884 hashcpy(base_spec.sha1, base->obj->idx.sha1);
885 find_delta_children(&base_spec,
886 &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
887
888 memset(&base_spec, 0, sizeof(base_spec));
889 base_spec.offset = base->obj->idx.offset;
890 find_delta_children(&base_spec,
891 &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
892
893 if (base->ref_last == -1 && base->ofs_last == -1) {
894 free(base->data);
895 return NULL;
896 }
897
898 link_base_data(prev_base, base);
899 }
900
901 if (base->ref_first <= base->ref_last) {
902 struct object_entry *child = objects + deltas[base->ref_first].obj_no;
903 struct base_data *result = alloc_base_data();
904
905 assert(child->real_type == OBJ_REF_DELTA);
906 resolve_delta(child, base, result);
907 if (base->ref_first == base->ref_last && base->ofs_last == -1)
908 free_base_data(base);
909
910 base->ref_first++;
911 return result;
912 }
913
914 if (base->ofs_first <= base->ofs_last) {
915 struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
916 struct base_data *result = alloc_base_data();
917
918 assert(child->real_type == OBJ_OFS_DELTA);
919 resolve_delta(child, base, result);
920 if (base->ofs_first == base->ofs_last)
921 free_base_data(base);
922
923 base->ofs_first++;
924 return result;
925 }
926
927 unlink_base_data(base);
928 return NULL;
929 }
930
931 static void find_unresolved_deltas(struct base_data *base)
932 {
933 struct base_data *new_base, *prev_base = NULL;
934 for (;;) {
935 new_base = find_unresolved_deltas_1(base, prev_base);
936
937 if (new_base) {
938 prev_base = base;
939 base = new_base;
940 } else {
941 free(base);
942 base = prev_base;
943 if (!base)
944 return;
945 prev_base = base->base;
946 }
947 }
948 }
949
950 static int compare_delta_entry(const void *a, const void *b)
951 {
952 const struct delta_entry *delta_a = a;
953 const struct delta_entry *delta_b = b;
954
955 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
956 return compare_delta_bases(&delta_a->base, &delta_b->base,
957 objects[delta_a->obj_no].type,
958 objects[delta_b->obj_no].type);
959 }
960
961 static void resolve_base(struct object_entry *obj)
962 {
963 struct base_data *base_obj = alloc_base_data();
964 base_obj->obj = obj;
965 base_obj->data = NULL;
966 find_unresolved_deltas(base_obj);
967 }
968
969 #ifndef NO_PTHREADS
970 static void *threaded_second_pass(void *data)
971 {
972 set_thread_data(data);
973 for (;;) {
974 int i;
975 counter_lock();
976 display_progress(progress, nr_resolved_deltas);
977 counter_unlock();
978 work_lock();
979 while (nr_dispatched < nr_objects &&
980 is_delta_type(objects[nr_dispatched].type))
981 nr_dispatched++;
982 if (nr_dispatched >= nr_objects) {
983 work_unlock();
984 break;
985 }
986 i = nr_dispatched++;
987 work_unlock();
988
989 resolve_base(&objects[i]);
990 }
991 return NULL;
992 }
993 #endif
994
995 /*
996 * First pass:
997 * - find locations of all objects;
998 * - calculate SHA1 of all non-delta objects;
999 * - remember base (SHA1 or offset) for all deltas.
1000 */
1001 static void parse_pack_objects(unsigned char *sha1)
1002 {
1003 int i, nr_delays = 0;
1004 struct delta_entry *delta = deltas;
1005 struct stat st;
1006
1007 if (verbose)
1008 progress = start_progress(
1009 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1010 nr_objects);
1011 for (i = 0; i < nr_objects; i++) {
1012 struct object_entry *obj = &objects[i];
1013 void *data = unpack_raw_entry(obj, &delta->base, obj->idx.sha1);
1014 obj->real_type = obj->type;
1015 if (is_delta_type(obj->type)) {
1016 nr_deltas++;
1017 delta->obj_no = i;
1018 delta++;
1019 } else if (!data) {
1020 /* large blobs, check later */
1021 obj->real_type = OBJ_BAD;
1022 nr_delays++;
1023 } else
1024 sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
1025 free(data);
1026 display_progress(progress, i+1);
1027 }
1028 objects[i].idx.offset = consumed_bytes;
1029 stop_progress(&progress);
1030
1031 /* Check pack integrity */
1032 flush();
1033 git_SHA1_Final(sha1, &input_ctx);
1034 if (hashcmp(fill(20), sha1))
1035 die(_("pack is corrupted (SHA1 mismatch)"));
1036 use(20);
1037
1038 /* If input_fd is a file, we should have reached its end now. */
1039 if (fstat(input_fd, &st))
1040 die_errno(_("cannot fstat packfile"));
1041 if (S_ISREG(st.st_mode) &&
1042 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1043 die(_("pack has junk at the end"));
1044
1045 for (i = 0; i < nr_objects; i++) {
1046 struct object_entry *obj = &objects[i];
1047 if (obj->real_type != OBJ_BAD)
1048 continue;
1049 obj->real_type = obj->type;
1050 sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
1051 nr_delays--;
1052 }
1053 if (nr_delays)
1054 die(_("confusion beyond insanity in parse_pack_objects()"));
1055 }
1056
1057 /*
1058 * Second pass:
1059 * - for all non-delta objects, look if it is used as a base for
1060 * deltas;
1061 * - if used as a base, uncompress the object and apply all deltas,
1062 * recursively checking if the resulting object is used as a base
1063 * for some more deltas.
1064 */
1065 static void resolve_deltas(void)
1066 {
1067 int i;
1068
1069 if (!nr_deltas)
1070 return;
1071
1072 /* Sort deltas by base SHA1/offset for fast searching */
1073 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
1074 compare_delta_entry);
1075
1076 if (verbose)
1077 progress = start_progress(_("Resolving deltas"), nr_deltas);
1078
1079 #ifndef NO_PTHREADS
1080 nr_dispatched = 0;
1081 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1082 init_thread();
1083 for (i = 0; i < nr_threads; i++) {
1084 int ret = pthread_create(&thread_data[i].thread, NULL,
1085 threaded_second_pass, thread_data + i);
1086 if (ret)
1087 die(_("unable to create thread: %s"),
1088 strerror(ret));
1089 }
1090 for (i = 0; i < nr_threads; i++)
1091 pthread_join(thread_data[i].thread, NULL);
1092 cleanup_thread();
1093 return;
1094 }
1095 #endif
1096
1097 for (i = 0; i < nr_objects; i++) {
1098 struct object_entry *obj = &objects[i];
1099
1100 if (is_delta_type(obj->type))
1101 continue;
1102 resolve_base(obj);
1103 display_progress(progress, nr_resolved_deltas);
1104 }
1105 }
1106
1107 /*
1108 * Third pass:
1109 * - append objects to convert thin pack to full pack if required
1110 * - write the final 20-byte SHA-1
1111 */
1112 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
1113 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1114 {
1115 if (nr_deltas == nr_resolved_deltas) {
1116 stop_progress(&progress);
1117 /* Flush remaining pack final 20-byte SHA1. */
1118 flush();
1119 return;
1120 }
1121
1122 if (fix_thin_pack) {
1123 struct sha1file *f;
1124 unsigned char read_sha1[20], tail_sha1[20];
1125 struct strbuf msg = STRBUF_INIT;
1126 int nr_unresolved = nr_deltas - nr_resolved_deltas;
1127 int nr_objects_initial = nr_objects;
1128 if (nr_unresolved <= 0)
1129 die(_("confusion beyond insanity"));
1130 objects = xrealloc(objects,
1131 (nr_objects + nr_unresolved + 1)
1132 * sizeof(*objects));
1133 memset(objects + nr_objects + 1, 0,
1134 nr_unresolved * sizeof(*objects));
1135 f = sha1fd(output_fd, curr_pack);
1136 fix_unresolved_deltas(f, nr_unresolved);
1137 strbuf_addf(&msg, _("completed with %d local objects"),
1138 nr_objects - nr_objects_initial);
1139 stop_progress_msg(&progress, msg.buf);
1140 strbuf_release(&msg);
1141 sha1close(f, tail_sha1, 0);
1142 hashcpy(read_sha1, pack_sha1);
1143 fixup_pack_header_footer(output_fd, pack_sha1,
1144 curr_pack, nr_objects,
1145 read_sha1, consumed_bytes-20);
1146 if (hashcmp(read_sha1, tail_sha1) != 0)
1147 die(_("Unexpected tail checksum for %s "
1148 "(disk corruption?)"), curr_pack);
1149 }
1150 if (nr_deltas != nr_resolved_deltas)
1151 die(Q_("pack has %d unresolved delta",
1152 "pack has %d unresolved deltas",
1153 nr_deltas - nr_resolved_deltas),
1154 nr_deltas - nr_resolved_deltas);
1155 }
1156
1157 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1158 {
1159 git_zstream stream;
1160 int status;
1161 unsigned char outbuf[4096];
1162
1163 memset(&stream, 0, sizeof(stream));
1164 git_deflate_init(&stream, zlib_compression_level);
1165 stream.next_in = in;
1166 stream.avail_in = size;
1167
1168 do {
1169 stream.next_out = outbuf;
1170 stream.avail_out = sizeof(outbuf);
1171 status = git_deflate(&stream, Z_FINISH);
1172 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1173 } while (status == Z_OK);
1174
1175 if (status != Z_STREAM_END)
1176 die(_("unable to deflate appended object (%d)"), status);
1177 size = stream.total_out;
1178 git_deflate_end(&stream);
1179 return size;
1180 }
1181
1182 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1183 const unsigned char *sha1, void *buf,
1184 unsigned long size, enum object_type type)
1185 {
1186 struct object_entry *obj = &objects[nr_objects++];
1187 unsigned char header[10];
1188 unsigned long s = size;
1189 int n = 0;
1190 unsigned char c = (type << 4) | (s & 15);
1191 s >>= 4;
1192 while (s) {
1193 header[n++] = c | 0x80;
1194 c = s & 0x7f;
1195 s >>= 7;
1196 }
1197 header[n++] = c;
1198 crc32_begin(f);
1199 sha1write(f, header, n);
1200 obj[0].size = size;
1201 obj[0].hdr_size = n;
1202 obj[0].type = type;
1203 obj[0].real_type = type;
1204 obj[1].idx.offset = obj[0].idx.offset + n;
1205 obj[1].idx.offset += write_compressed(f, buf, size);
1206 obj[0].idx.crc32 = crc32_end(f);
1207 sha1flush(f);
1208 hashcpy(obj->idx.sha1, sha1);
1209 return obj;
1210 }
1211
1212 static int delta_pos_compare(const void *_a, const void *_b)
1213 {
1214 struct delta_entry *a = *(struct delta_entry **)_a;
1215 struct delta_entry *b = *(struct delta_entry **)_b;
1216 return a->obj_no - b->obj_no;
1217 }
1218
1219 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
1220 {
1221 struct delta_entry **sorted_by_pos;
1222 int i, n = 0;
1223
1224 /*
1225 * Since many unresolved deltas may well be themselves base objects
1226 * for more unresolved deltas, we really want to include the
1227 * smallest number of base objects that would cover as much delta
1228 * as possible by picking the
1229 * trunc deltas first, allowing for other deltas to resolve without
1230 * additional base objects. Since most base objects are to be found
1231 * before deltas depending on them, a good heuristic is to start
1232 * resolving deltas in the same order as their position in the pack.
1233 */
1234 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
1235 for (i = 0; i < nr_deltas; i++) {
1236 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1237 continue;
1238 sorted_by_pos[n++] = &deltas[i];
1239 }
1240 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1241
1242 for (i = 0; i < n; i++) {
1243 struct delta_entry *d = sorted_by_pos[i];
1244 enum object_type type;
1245 struct base_data *base_obj = alloc_base_data();
1246
1247 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1248 continue;
1249 base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1250 if (!base_obj->data)
1251 continue;
1252
1253 if (check_sha1_signature(d->base.sha1, base_obj->data,
1254 base_obj->size, typename(type)))
1255 die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
1256 base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1257 base_obj->data, base_obj->size, type);
1258 find_unresolved_deltas(base_obj);
1259 display_progress(progress, nr_resolved_deltas);
1260 }
1261 free(sorted_by_pos);
1262 }
1263
1264 static void final(const char *final_pack_name, const char *curr_pack_name,
1265 const char *final_index_name, const char *curr_index_name,
1266 const char *keep_name, const char *keep_msg,
1267 unsigned char *sha1)
1268 {
1269 const char *report = "pack";
1270 char name[PATH_MAX];
1271 int err;
1272
1273 if (!from_stdin) {
1274 close(input_fd);
1275 } else {
1276 fsync_or_die(output_fd, curr_pack_name);
1277 err = close(output_fd);
1278 if (err)
1279 die_errno(_("error while closing pack file"));
1280 }
1281
1282 if (keep_msg) {
1283 int keep_fd, keep_msg_len = strlen(keep_msg);
1284
1285 if (!keep_name)
1286 keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1287 else
1288 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1289
1290 if (keep_fd < 0) {
1291 if (errno != EEXIST)
1292 die_errno(_("cannot write keep file '%s'"),
1293 keep_name);
1294 } else {
1295 if (keep_msg_len > 0) {
1296 write_or_die(keep_fd, keep_msg, keep_msg_len);
1297 write_or_die(keep_fd, "\n", 1);
1298 }
1299 if (close(keep_fd) != 0)
1300 die_errno(_("cannot close written keep file '%s'"),
1301 keep_name);
1302 report = "keep";
1303 }
1304 }
1305
1306 if (final_pack_name != curr_pack_name) {
1307 if (!final_pack_name) {
1308 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1309 get_object_directory(), sha1_to_hex(sha1));
1310 final_pack_name = name;
1311 }
1312 if (move_temp_to_file(curr_pack_name, final_pack_name))
1313 die(_("cannot store pack file"));
1314 } else if (from_stdin)
1315 chmod(final_pack_name, 0444);
1316
1317 if (final_index_name != curr_index_name) {
1318 if (!final_index_name) {
1319 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1320 get_object_directory(), sha1_to_hex(sha1));
1321 final_index_name = name;
1322 }
1323 if (move_temp_to_file(curr_index_name, final_index_name))
1324 die(_("cannot store index file"));
1325 } else
1326 chmod(final_index_name, 0444);
1327
1328 if (!from_stdin) {
1329 printf("%s\n", sha1_to_hex(sha1));
1330 } else {
1331 char buf[48];
1332 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1333 report, sha1_to_hex(sha1));
1334 write_or_die(1, buf, len);
1335
1336 /*
1337 * Let's just mimic git-unpack-objects here and write
1338 * the last part of the input buffer to stdout.
1339 */
1340 while (input_len) {
1341 err = xwrite(1, input_buffer + input_offset, input_len);
1342 if (err <= 0)
1343 break;
1344 input_len -= err;
1345 input_offset += err;
1346 }
1347 }
1348 }
1349
1350 static int git_index_pack_config(const char *k, const char *v, void *cb)
1351 {
1352 struct pack_idx_option *opts = cb;
1353
1354 if (!strcmp(k, "pack.indexversion")) {
1355 opts->version = git_config_int(k, v);
1356 if (opts->version > 2)
1357 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1358 return 0;
1359 }
1360 if (!strcmp(k, "pack.threads")) {
1361 nr_threads = git_config_int(k, v);
1362 if (nr_threads < 0)
1363 die(_("invalid number of threads specified (%d)"),
1364 nr_threads);
1365 #ifdef NO_PTHREADS
1366 if (nr_threads != 1)
1367 warning(_("no threads support, ignoring %s"), k);
1368 nr_threads = 1;
1369 #endif
1370 return 0;
1371 }
1372 return git_default_config(k, v, cb);
1373 }
1374
1375 static int cmp_uint32(const void *a_, const void *b_)
1376 {
1377 uint32_t a = *((uint32_t *)a_);
1378 uint32_t b = *((uint32_t *)b_);
1379
1380 return (a < b) ? -1 : (a != b);
1381 }
1382
1383 static void read_v2_anomalous_offsets(struct packed_git *p,
1384 struct pack_idx_option *opts)
1385 {
1386 const uint32_t *idx1, *idx2;
1387 uint32_t i;
1388
1389 /* The address of the 4-byte offset table */
1390 idx1 = (((const uint32_t *)p->index_data)
1391 + 2 /* 8-byte header */
1392 + 256 /* fan out */
1393 + 5 * p->num_objects /* 20-byte SHA-1 table */
1394 + p->num_objects /* CRC32 table */
1395 );
1396
1397 /* The address of the 8-byte offset table */
1398 idx2 = idx1 + p->num_objects;
1399
1400 for (i = 0; i < p->num_objects; i++) {
1401 uint32_t off = ntohl(idx1[i]);
1402 if (!(off & 0x80000000))
1403 continue;
1404 off = off & 0x7fffffff;
1405 if (idx2[off * 2])
1406 continue;
1407 /*
1408 * The real offset is ntohl(idx2[off * 2]) in high 4
1409 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1410 * octets. But idx2[off * 2] is Zero!!!
1411 */
1412 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1413 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1414 }
1415
1416 if (1 < opts->anomaly_nr)
1417 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1418 }
1419
1420 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1421 {
1422 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1423
1424 if (!p)
1425 die(_("Cannot open existing pack file '%s'"), pack_name);
1426 if (open_pack_index(p))
1427 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1428
1429 /* Read the attributes from the existing idx file */
1430 opts->version = p->index_version;
1431
1432 if (opts->version == 2)
1433 read_v2_anomalous_offsets(p, opts);
1434
1435 /*
1436 * Get rid of the idx file as we do not need it anymore.
1437 * NEEDSWORK: extract this bit from free_pack_by_name() in
1438 * sha1_file.c, perhaps? It shouldn't matter very much as we
1439 * know we haven't installed this pack (hence we never have
1440 * read anything from it).
1441 */
1442 close_pack_index(p);
1443 free(p);
1444 }
1445
1446 static void show_pack_info(int stat_only)
1447 {
1448 int i, baseobjects = nr_objects - nr_deltas;
1449 unsigned long *chain_histogram = NULL;
1450
1451 if (deepest_delta)
1452 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1453
1454 for (i = 0; i < nr_objects; i++) {
1455 struct object_entry *obj = &objects[i];
1456
1457 if (is_delta_type(obj->type))
1458 chain_histogram[obj->delta_depth - 1]++;
1459 if (stat_only)
1460 continue;
1461 printf("%s %-6s %lu %lu %"PRIuMAX,
1462 sha1_to_hex(obj->idx.sha1),
1463 typename(obj->real_type), obj->size,
1464 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1465 (uintmax_t)obj->idx.offset);
1466 if (is_delta_type(obj->type)) {
1467 struct object_entry *bobj = &objects[obj->base_object_no];
1468 printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1469 }
1470 putchar('\n');
1471 }
1472
1473 if (baseobjects)
1474 printf_ln(Q_("non delta: %d object",
1475 "non delta: %d objects",
1476 baseobjects),
1477 baseobjects);
1478 for (i = 0; i < deepest_delta; i++) {
1479 if (!chain_histogram[i])
1480 continue;
1481 printf_ln(Q_("chain length = %d: %lu object",
1482 "chain length = %d: %lu objects",
1483 chain_histogram[i]),
1484 i + 1,
1485 chain_histogram[i]);
1486 }
1487 }
1488
1489 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1490 {
1491 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1492 const char *curr_pack, *curr_index;
1493 const char *index_name = NULL, *pack_name = NULL;
1494 const char *keep_name = NULL, *keep_msg = NULL;
1495 char *index_name_buf = NULL, *keep_name_buf = NULL;
1496 struct pack_idx_entry **idx_objects;
1497 struct pack_idx_option opts;
1498 unsigned char pack_sha1[20];
1499 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1500
1501 if (argc == 2 && !strcmp(argv[1], "-h"))
1502 usage(index_pack_usage);
1503
1504 read_replace_refs = 0;
1505
1506 reset_pack_idx_option(&opts);
1507 git_config(git_index_pack_config, &opts);
1508 if (prefix && chdir(prefix))
1509 die(_("Cannot come back to cwd"));
1510
1511 for (i = 1; i < argc; i++) {
1512 const char *arg = argv[i];
1513
1514 if (*arg == '-') {
1515 if (!strcmp(arg, "--stdin")) {
1516 from_stdin = 1;
1517 } else if (!strcmp(arg, "--fix-thin")) {
1518 fix_thin_pack = 1;
1519 } else if (!strcmp(arg, "--strict")) {
1520 strict = 1;
1521 do_fsck_object = 1;
1522 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1523 strict = 1;
1524 check_self_contained_and_connected = 1;
1525 } else if (!strcmp(arg, "--verify")) {
1526 verify = 1;
1527 } else if (!strcmp(arg, "--verify-stat")) {
1528 verify = 1;
1529 show_stat = 1;
1530 } else if (!strcmp(arg, "--verify-stat-only")) {
1531 verify = 1;
1532 show_stat = 1;
1533 stat_only = 1;
1534 } else if (!strcmp(arg, "--keep")) {
1535 keep_msg = "";
1536 } else if (!prefixcmp(arg, "--keep=")) {
1537 keep_msg = arg + 7;
1538 } else if (!prefixcmp(arg, "--threads=")) {
1539 char *end;
1540 nr_threads = strtoul(arg+10, &end, 0);
1541 if (!arg[10] || *end || nr_threads < 0)
1542 usage(index_pack_usage);
1543 #ifdef NO_PTHREADS
1544 if (nr_threads != 1)
1545 warning(_("no threads support, "
1546 "ignoring %s"), arg);
1547 nr_threads = 1;
1548 #endif
1549 } else if (!prefixcmp(arg, "--pack_header=")) {
1550 struct pack_header *hdr;
1551 char *c;
1552
1553 hdr = (struct pack_header *)input_buffer;
1554 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1555 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1556 if (*c != ',')
1557 die(_("bad %s"), arg);
1558 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1559 if (*c)
1560 die(_("bad %s"), arg);
1561 input_len = sizeof(*hdr);
1562 } else if (!strcmp(arg, "-v")) {
1563 verbose = 1;
1564 } else if (!strcmp(arg, "-o")) {
1565 if (index_name || (i+1) >= argc)
1566 usage(index_pack_usage);
1567 index_name = argv[++i];
1568 } else if (!prefixcmp(arg, "--index-version=")) {
1569 char *c;
1570 opts.version = strtoul(arg + 16, &c, 10);
1571 if (opts.version > 2)
1572 die(_("bad %s"), arg);
1573 if (*c == ',')
1574 opts.off32_limit = strtoul(c+1, &c, 0);
1575 if (*c || opts.off32_limit & 0x80000000)
1576 die(_("bad %s"), arg);
1577 } else
1578 usage(index_pack_usage);
1579 continue;
1580 }
1581
1582 if (pack_name)
1583 usage(index_pack_usage);
1584 pack_name = arg;
1585 }
1586
1587 if (!pack_name && !from_stdin)
1588 usage(index_pack_usage);
1589 if (fix_thin_pack && !from_stdin)
1590 die(_("--fix-thin cannot be used without --stdin"));
1591 if (!index_name && pack_name) {
1592 int len = strlen(pack_name);
1593 if (!has_extension(pack_name, ".pack"))
1594 die(_("packfile name '%s' does not end with '.pack'"),
1595 pack_name);
1596 index_name_buf = xmalloc(len);
1597 memcpy(index_name_buf, pack_name, len - 5);
1598 strcpy(index_name_buf + len - 5, ".idx");
1599 index_name = index_name_buf;
1600 }
1601 if (keep_msg && !keep_name && pack_name) {
1602 int len = strlen(pack_name);
1603 if (!has_extension(pack_name, ".pack"))
1604 die(_("packfile name '%s' does not end with '.pack'"),
1605 pack_name);
1606 keep_name_buf = xmalloc(len);
1607 memcpy(keep_name_buf, pack_name, len - 5);
1608 strcpy(keep_name_buf + len - 5, ".keep");
1609 keep_name = keep_name_buf;
1610 }
1611 if (verify) {
1612 if (!index_name)
1613 die(_("--verify with no packfile name given"));
1614 read_idx_option(&opts, index_name);
1615 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1616 }
1617 if (strict)
1618 opts.flags |= WRITE_IDX_STRICT;
1619
1620 #ifndef NO_PTHREADS
1621 if (!nr_threads) {
1622 nr_threads = online_cpus();
1623 /* An experiment showed that more threads does not mean faster */
1624 if (nr_threads > 3)
1625 nr_threads = 3;
1626 }
1627 #endif
1628
1629 curr_pack = open_pack_file(pack_name);
1630 parse_pack_header();
1631 objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1632 deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1633 parse_pack_objects(pack_sha1);
1634 resolve_deltas();
1635 conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1636 free(deltas);
1637 if (strict)
1638 foreign_nr = check_objects();
1639
1640 if (show_stat)
1641 show_pack_info(stat_only);
1642
1643 idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1644 for (i = 0; i < nr_objects; i++)
1645 idx_objects[i] = &objects[i].idx;
1646 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1647 free(idx_objects);
1648
1649 if (!verify)
1650 final(pack_name, curr_pack,
1651 index_name, curr_index,
1652 keep_name, keep_msg,
1653 pack_sha1);
1654 else
1655 close(input_fd);
1656 free(objects);
1657 free(index_name_buf);
1658 free(keep_name_buf);
1659 if (pack_name == NULL)
1660 free((void *) curr_pack);
1661 if (index_name == NULL)
1662 free((void *) curr_index);
1663
1664 /*
1665 * Let the caller know this pack is not self contained
1666 */
1667 if (check_self_contained_and_connected && foreign_nr)
1668 return 1;
1669
1670 return 0;
1671 }