pack: convert struct pack_idx_entry to struct object_id
[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 char hdr_size;
22 signed char type;
23 signed char real_type;
24 };
25
26 struct object_stat {
27 unsigned delta_depth;
28 int base_object_no;
29 };
30
31 struct base_data {
32 struct base_data *base;
33 struct base_data *child;
34 struct object_entry *obj;
35 void *data;
36 unsigned long size;
37 int ref_first, ref_last;
38 int ofs_first, ofs_last;
39 };
40
41 struct thread_local {
42 #ifndef NO_PTHREADS
43 pthread_t thread;
44 #endif
45 struct base_data *base_cache;
46 size_t base_cache_used;
47 int pack_fd;
48 };
49
50 #define FLAG_LINK (1u<<20)
51 #define FLAG_CHECKED (1u<<21)
52
53 struct ofs_delta_entry {
54 off_t offset;
55 int obj_no;
56 };
57
58 struct ref_delta_entry {
59 unsigned char sha1[20];
60 int obj_no;
61 };
62
63 static struct object_entry *objects;
64 static struct object_stat *obj_stat;
65 static struct ofs_delta_entry *ofs_deltas;
66 static struct ref_delta_entry *ref_deltas;
67 static struct thread_local nothread_data;
68 static int nr_objects;
69 static int nr_ofs_deltas;
70 static int nr_ref_deltas;
71 static int ref_deltas_alloc;
72 static int nr_resolved_deltas;
73 static int nr_threads;
74
75 static int from_stdin;
76 static int strict;
77 static int do_fsck_object;
78 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
79 static int verbose;
80 static int show_resolving_progress;
81 static int show_stat;
82 static int check_self_contained_and_connected;
83
84 static struct progress *progress;
85
86 /* We always read in 4kB chunks. */
87 static unsigned char input_buffer[4096];
88 static unsigned int input_offset, input_len;
89 static off_t consumed_bytes;
90 static off_t max_input_size;
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;
95 static const char *curr_pack;
96
97 #ifndef NO_PTHREADS
98
99 static struct thread_local *thread_data;
100 static int nr_dispatched;
101 static int threads_active;
102
103 static pthread_mutex_t read_mutex;
104 #define read_lock() lock_mutex(&read_mutex)
105 #define read_unlock() unlock_mutex(&read_mutex)
106
107 static pthread_mutex_t counter_mutex;
108 #define counter_lock() lock_mutex(&counter_mutex)
109 #define counter_unlock() unlock_mutex(&counter_mutex)
110
111 static pthread_mutex_t work_mutex;
112 #define work_lock() lock_mutex(&work_mutex)
113 #define work_unlock() unlock_mutex(&work_mutex)
114
115 static pthread_mutex_t deepest_delta_mutex;
116 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
117 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
118
119 static pthread_mutex_t type_cas_mutex;
120 #define type_cas_lock() lock_mutex(&type_cas_mutex)
121 #define type_cas_unlock() unlock_mutex(&type_cas_mutex)
122
123 static pthread_key_t key;
124
125 static inline void lock_mutex(pthread_mutex_t *mutex)
126 {
127 if (threads_active)
128 pthread_mutex_lock(mutex);
129 }
130
131 static inline void unlock_mutex(pthread_mutex_t *mutex)
132 {
133 if (threads_active)
134 pthread_mutex_unlock(mutex);
135 }
136
137 /*
138 * Mutex and conditional variable can't be statically-initialized on Windows.
139 */
140 static void init_thread(void)
141 {
142 int i;
143 init_recursive_mutex(&read_mutex);
144 pthread_mutex_init(&counter_mutex, NULL);
145 pthread_mutex_init(&work_mutex, NULL);
146 pthread_mutex_init(&type_cas_mutex, NULL);
147 if (show_stat)
148 pthread_mutex_init(&deepest_delta_mutex, NULL);
149 pthread_key_create(&key, NULL);
150 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
151 for (i = 0; i < nr_threads; i++) {
152 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
153 if (thread_data[i].pack_fd == -1)
154 die_errno(_("unable to open %s"), curr_pack);
155 }
156
157 threads_active = 1;
158 }
159
160 static void cleanup_thread(void)
161 {
162 int i;
163 if (!threads_active)
164 return;
165 threads_active = 0;
166 pthread_mutex_destroy(&read_mutex);
167 pthread_mutex_destroy(&counter_mutex);
168 pthread_mutex_destroy(&work_mutex);
169 pthread_mutex_destroy(&type_cas_mutex);
170 if (show_stat)
171 pthread_mutex_destroy(&deepest_delta_mutex);
172 for (i = 0; i < nr_threads; i++)
173 close(thread_data[i].pack_fd);
174 pthread_key_delete(key);
175 free(thread_data);
176 }
177
178 #else
179
180 #define read_lock()
181 #define read_unlock()
182
183 #define counter_lock()
184 #define counter_unlock()
185
186 #define work_lock()
187 #define work_unlock()
188
189 #define deepest_delta_lock()
190 #define deepest_delta_unlock()
191
192 #define type_cas_lock()
193 #define type_cas_unlock()
194
195 #endif
196
197
198 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
199 {
200 if (!obj)
201 return -1;
202
203 if (type != OBJ_ANY && obj->type != type)
204 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
205
206 obj->flags |= FLAG_LINK;
207 return 0;
208 }
209
210 /* The content of each linked object must have been checked
211 or it must be already present in the object database */
212 static unsigned check_object(struct object *obj)
213 {
214 if (!obj)
215 return 0;
216
217 if (!(obj->flags & FLAG_LINK))
218 return 0;
219
220 if (!(obj->flags & FLAG_CHECKED)) {
221 unsigned long size;
222 int type = sha1_object_info(obj->oid.hash, &size);
223 if (type <= 0)
224 die(_("did not receive expected object %s"),
225 oid_to_hex(&obj->oid));
226 if (type != obj->type)
227 die(_("object %s: expected type %s, found %s"),
228 oid_to_hex(&obj->oid),
229 typename(obj->type), typename(type));
230 obj->flags |= FLAG_CHECKED;
231 return 1;
232 }
233
234 return 0;
235 }
236
237 static unsigned check_objects(void)
238 {
239 unsigned i, max, foreign_nr = 0;
240
241 max = get_max_object_index();
242 for (i = 0; i < max; i++)
243 foreign_nr += check_object(get_indexed_object(i));
244 return foreign_nr;
245 }
246
247
248 /* Discard current buffer used content. */
249 static void flush(void)
250 {
251 if (input_offset) {
252 if (output_fd >= 0)
253 write_or_die(output_fd, input_buffer, input_offset);
254 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
255 memmove(input_buffer, input_buffer + input_offset, input_len);
256 input_offset = 0;
257 }
258 }
259
260 /*
261 * Make sure at least "min" bytes are available in the buffer, and
262 * return the pointer to the buffer.
263 */
264 static void *fill(int min)
265 {
266 if (min <= input_len)
267 return input_buffer + input_offset;
268 if (min > sizeof(input_buffer))
269 die(Q_("cannot fill %d byte",
270 "cannot fill %d bytes",
271 min),
272 min);
273 flush();
274 do {
275 ssize_t ret = xread(input_fd, input_buffer + input_len,
276 sizeof(input_buffer) - input_len);
277 if (ret <= 0) {
278 if (!ret)
279 die(_("early EOF"));
280 die_errno(_("read error on input"));
281 }
282 input_len += ret;
283 if (from_stdin)
284 display_throughput(progress, consumed_bytes + input_len);
285 } while (input_len < min);
286 return input_buffer;
287 }
288
289 static void use(int bytes)
290 {
291 if (bytes > input_len)
292 die(_("used more bytes than were available"));
293 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
294 input_len -= bytes;
295 input_offset += bytes;
296
297 /* make sure off_t is sufficiently large not to wrap */
298 if (signed_add_overflows(consumed_bytes, bytes))
299 die(_("pack too large for current definition of off_t"));
300 consumed_bytes += bytes;
301 if (max_input_size && consumed_bytes > max_input_size)
302 die(_("pack exceeds maximum allowed size"));
303 }
304
305 static const char *open_pack_file(const char *pack_name)
306 {
307 if (from_stdin) {
308 input_fd = 0;
309 if (!pack_name) {
310 struct strbuf tmp_file = STRBUF_INIT;
311 output_fd = odb_mkstemp(&tmp_file,
312 "pack/tmp_pack_XXXXXX");
313 pack_name = strbuf_detach(&tmp_file, NULL);
314 } else {
315 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
316 if (output_fd < 0)
317 die_errno(_("unable to create '%s'"), pack_name);
318 }
319 nothread_data.pack_fd = output_fd;
320 } else {
321 input_fd = open(pack_name, O_RDONLY);
322 if (input_fd < 0)
323 die_errno(_("cannot open packfile '%s'"), pack_name);
324 output_fd = -1;
325 nothread_data.pack_fd = input_fd;
326 }
327 git_SHA1_Init(&input_ctx);
328 return pack_name;
329 }
330
331 static void parse_pack_header(void)
332 {
333 struct pack_header *hdr = fill(sizeof(struct pack_header));
334
335 /* Header consistency check */
336 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
337 die(_("pack signature mismatch"));
338 if (!pack_version_ok(hdr->hdr_version))
339 die(_("pack version %"PRIu32" unsupported"),
340 ntohl(hdr->hdr_version));
341
342 nr_objects = ntohl(hdr->hdr_entries);
343 use(sizeof(struct pack_header));
344 }
345
346 static NORETURN void bad_object(off_t offset, const char *format,
347 ...) __attribute__((format (printf, 2, 3)));
348
349 static NORETURN void bad_object(off_t offset, const char *format, ...)
350 {
351 va_list params;
352 char buf[1024];
353
354 va_start(params, format);
355 vsnprintf(buf, sizeof(buf), format, params);
356 va_end(params);
357 die(_("pack has bad object at offset %"PRIuMAX": %s"),
358 (uintmax_t)offset, buf);
359 }
360
361 static inline struct thread_local *get_thread_data(void)
362 {
363 #ifndef NO_PTHREADS
364 if (threads_active)
365 return pthread_getspecific(key);
366 assert(!threads_active &&
367 "This should only be reached when all threads are gone");
368 #endif
369 return &nothread_data;
370 }
371
372 #ifndef NO_PTHREADS
373 static void set_thread_data(struct thread_local *data)
374 {
375 if (threads_active)
376 pthread_setspecific(key, data);
377 }
378 #endif
379
380 static struct base_data *alloc_base_data(void)
381 {
382 struct base_data *base = xcalloc(1, sizeof(struct base_data));
383 base->ref_last = -1;
384 base->ofs_last = -1;
385 return base;
386 }
387
388 static void free_base_data(struct base_data *c)
389 {
390 if (c->data) {
391 free(c->data);
392 c->data = NULL;
393 get_thread_data()->base_cache_used -= c->size;
394 }
395 }
396
397 static void prune_base_data(struct base_data *retain)
398 {
399 struct base_data *b;
400 struct thread_local *data = get_thread_data();
401 for (b = data->base_cache;
402 data->base_cache_used > delta_base_cache_limit && b;
403 b = b->child) {
404 if (b->data && b != retain)
405 free_base_data(b);
406 }
407 }
408
409 static void link_base_data(struct base_data *base, struct base_data *c)
410 {
411 if (base)
412 base->child = c;
413 else
414 get_thread_data()->base_cache = c;
415
416 c->base = base;
417 c->child = NULL;
418 if (c->data)
419 get_thread_data()->base_cache_used += c->size;
420 prune_base_data(c);
421 }
422
423 static void unlink_base_data(struct base_data *c)
424 {
425 struct base_data *base = c->base;
426 if (base)
427 base->child = NULL;
428 else
429 get_thread_data()->base_cache = NULL;
430 free_base_data(c);
431 }
432
433 static int is_delta_type(enum object_type type)
434 {
435 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
436 }
437
438 static void *unpack_entry_data(off_t offset, unsigned long size,
439 enum object_type type, unsigned char *sha1)
440 {
441 static char fixed_buf[8192];
442 int status;
443 git_zstream stream;
444 void *buf;
445 git_SHA_CTX c;
446 char hdr[32];
447 int hdrlen;
448
449 if (!is_delta_type(type)) {
450 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), size) + 1;
451 git_SHA1_Init(&c);
452 git_SHA1_Update(&c, hdr, hdrlen);
453 } else
454 sha1 = NULL;
455 if (type == OBJ_BLOB && size > big_file_threshold)
456 buf = fixed_buf;
457 else
458 buf = xmallocz(size);
459
460 memset(&stream, 0, sizeof(stream));
461 git_inflate_init(&stream);
462 stream.next_out = buf;
463 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
464
465 do {
466 unsigned char *last_out = stream.next_out;
467 stream.next_in = fill(1);
468 stream.avail_in = input_len;
469 status = git_inflate(&stream, 0);
470 use(input_len - stream.avail_in);
471 if (sha1)
472 git_SHA1_Update(&c, last_out, stream.next_out - last_out);
473 if (buf == fixed_buf) {
474 stream.next_out = buf;
475 stream.avail_out = sizeof(fixed_buf);
476 }
477 } while (status == Z_OK);
478 if (stream.total_out != size || status != Z_STREAM_END)
479 bad_object(offset, _("inflate returned %d"), status);
480 git_inflate_end(&stream);
481 if (sha1)
482 git_SHA1_Final(sha1, &c);
483 return buf == fixed_buf ? NULL : buf;
484 }
485
486 static void *unpack_raw_entry(struct object_entry *obj,
487 off_t *ofs_offset,
488 unsigned char *ref_sha1,
489 unsigned char *sha1)
490 {
491 unsigned char *p;
492 unsigned long size, c;
493 off_t base_offset;
494 unsigned shift;
495 void *data;
496
497 obj->idx.offset = consumed_bytes;
498 input_crc32 = crc32(0, NULL, 0);
499
500 p = fill(1);
501 c = *p;
502 use(1);
503 obj->type = (c >> 4) & 7;
504 size = (c & 15);
505 shift = 4;
506 while (c & 0x80) {
507 p = fill(1);
508 c = *p;
509 use(1);
510 size += (c & 0x7f) << shift;
511 shift += 7;
512 }
513 obj->size = size;
514
515 switch (obj->type) {
516 case OBJ_REF_DELTA:
517 hashcpy(ref_sha1, fill(20));
518 use(20);
519 break;
520 case OBJ_OFS_DELTA:
521 p = fill(1);
522 c = *p;
523 use(1);
524 base_offset = c & 127;
525 while (c & 128) {
526 base_offset += 1;
527 if (!base_offset || MSB(base_offset, 7))
528 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
529 p = fill(1);
530 c = *p;
531 use(1);
532 base_offset = (base_offset << 7) + (c & 127);
533 }
534 *ofs_offset = obj->idx.offset - base_offset;
535 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
536 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
537 break;
538 case OBJ_COMMIT:
539 case OBJ_TREE:
540 case OBJ_BLOB:
541 case OBJ_TAG:
542 break;
543 default:
544 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
545 }
546 obj->hdr_size = consumed_bytes - obj->idx.offset;
547
548 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
549 obj->idx.crc32 = input_crc32;
550 return data;
551 }
552
553 static void *unpack_data(struct object_entry *obj,
554 int (*consume)(const unsigned char *, unsigned long, void *),
555 void *cb_data)
556 {
557 off_t from = obj[0].idx.offset + obj[0].hdr_size;
558 off_t len = obj[1].idx.offset - from;
559 unsigned char *data, *inbuf;
560 git_zstream stream;
561 int status;
562
563 data = xmallocz(consume ? 64*1024 : obj->size);
564 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
565
566 memset(&stream, 0, sizeof(stream));
567 git_inflate_init(&stream);
568 stream.next_out = data;
569 stream.avail_out = consume ? 64*1024 : obj->size;
570
571 do {
572 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
573 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
574 if (n < 0)
575 die_errno(_("cannot pread pack file"));
576 if (!n)
577 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
578 "premature end of pack file, %"PRIuMAX" bytes missing",
579 (unsigned int)len),
580 (uintmax_t)len);
581 from += n;
582 len -= n;
583 stream.next_in = inbuf;
584 stream.avail_in = n;
585 if (!consume)
586 status = git_inflate(&stream, 0);
587 else {
588 do {
589 status = git_inflate(&stream, 0);
590 if (consume(data, stream.next_out - data, cb_data)) {
591 free(inbuf);
592 free(data);
593 return NULL;
594 }
595 stream.next_out = data;
596 stream.avail_out = 64*1024;
597 } while (status == Z_OK && stream.avail_in);
598 }
599 } while (len && status == Z_OK && !stream.avail_in);
600
601 /* This has been inflated OK when first encountered, so... */
602 if (status != Z_STREAM_END || stream.total_out != obj->size)
603 die(_("serious inflate inconsistency"));
604
605 git_inflate_end(&stream);
606 free(inbuf);
607 if (consume) {
608 free(data);
609 data = NULL;
610 }
611 return data;
612 }
613
614 static void *get_data_from_pack(struct object_entry *obj)
615 {
616 return unpack_data(obj, NULL, NULL);
617 }
618
619 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
620 enum object_type type1,
621 enum object_type type2)
622 {
623 int cmp = type1 - type2;
624 if (cmp)
625 return cmp;
626 return offset1 < offset2 ? -1 :
627 offset1 > offset2 ? 1 :
628 0;
629 }
630
631 static int find_ofs_delta(const off_t offset, enum object_type type)
632 {
633 int first = 0, last = nr_ofs_deltas;
634
635 while (first < last) {
636 int next = (first + last) / 2;
637 struct ofs_delta_entry *delta = &ofs_deltas[next];
638 int cmp;
639
640 cmp = compare_ofs_delta_bases(offset, delta->offset,
641 type, objects[delta->obj_no].type);
642 if (!cmp)
643 return next;
644 if (cmp < 0) {
645 last = next;
646 continue;
647 }
648 first = next+1;
649 }
650 return -first-1;
651 }
652
653 static void find_ofs_delta_children(off_t offset,
654 int *first_index, int *last_index,
655 enum object_type type)
656 {
657 int first = find_ofs_delta(offset, type);
658 int last = first;
659 int end = nr_ofs_deltas - 1;
660
661 if (first < 0) {
662 *first_index = 0;
663 *last_index = -1;
664 return;
665 }
666 while (first > 0 && ofs_deltas[first - 1].offset == offset)
667 --first;
668 while (last < end && ofs_deltas[last + 1].offset == offset)
669 ++last;
670 *first_index = first;
671 *last_index = last;
672 }
673
674 static int compare_ref_delta_bases(const unsigned char *sha1,
675 const unsigned char *sha2,
676 enum object_type type1,
677 enum object_type type2)
678 {
679 int cmp = type1 - type2;
680 if (cmp)
681 return cmp;
682 return hashcmp(sha1, sha2);
683 }
684
685 static int find_ref_delta(const unsigned char *sha1, enum object_type type)
686 {
687 int first = 0, last = nr_ref_deltas;
688
689 while (first < last) {
690 int next = (first + last) / 2;
691 struct ref_delta_entry *delta = &ref_deltas[next];
692 int cmp;
693
694 cmp = compare_ref_delta_bases(sha1, delta->sha1,
695 type, objects[delta->obj_no].type);
696 if (!cmp)
697 return next;
698 if (cmp < 0) {
699 last = next;
700 continue;
701 }
702 first = next+1;
703 }
704 return -first-1;
705 }
706
707 static void find_ref_delta_children(const unsigned char *sha1,
708 int *first_index, int *last_index,
709 enum object_type type)
710 {
711 int first = find_ref_delta(sha1, type);
712 int last = first;
713 int end = nr_ref_deltas - 1;
714
715 if (first < 0) {
716 *first_index = 0;
717 *last_index = -1;
718 return;
719 }
720 while (first > 0 && !hashcmp(ref_deltas[first - 1].sha1, sha1))
721 --first;
722 while (last < end && !hashcmp(ref_deltas[last + 1].sha1, sha1))
723 ++last;
724 *first_index = first;
725 *last_index = last;
726 }
727
728 struct compare_data {
729 struct object_entry *entry;
730 struct git_istream *st;
731 unsigned char *buf;
732 unsigned long buf_size;
733 };
734
735 static int compare_objects(const unsigned char *buf, unsigned long size,
736 void *cb_data)
737 {
738 struct compare_data *data = cb_data;
739
740 if (data->buf_size < size) {
741 free(data->buf);
742 data->buf = xmalloc(size);
743 data->buf_size = size;
744 }
745
746 while (size) {
747 ssize_t len = read_istream(data->st, data->buf, size);
748 if (len == 0)
749 die(_("SHA1 COLLISION FOUND WITH %s !"),
750 oid_to_hex(&data->entry->idx.oid));
751 if (len < 0)
752 die(_("unable to read %s"),
753 oid_to_hex(&data->entry->idx.oid));
754 if (memcmp(buf, data->buf, len))
755 die(_("SHA1 COLLISION FOUND WITH %s !"),
756 oid_to_hex(&data->entry->idx.oid));
757 size -= len;
758 buf += len;
759 }
760 return 0;
761 }
762
763 static int check_collison(struct object_entry *entry)
764 {
765 struct compare_data data;
766 enum object_type type;
767 unsigned long size;
768
769 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
770 return -1;
771
772 memset(&data, 0, sizeof(data));
773 data.entry = entry;
774 data.st = open_istream(entry->idx.oid.hash, &type, &size, NULL);
775 if (!data.st)
776 return -1;
777 if (size != entry->size || type != entry->type)
778 die(_("SHA1 COLLISION FOUND WITH %s !"),
779 oid_to_hex(&entry->idx.oid));
780 unpack_data(entry, compare_objects, &data);
781 close_istream(data.st);
782 free(data.buf);
783 return 0;
784 }
785
786 static void sha1_object(const void *data, struct object_entry *obj_entry,
787 unsigned long size, enum object_type type,
788 const unsigned char *sha1)
789 {
790 void *new_data = NULL;
791 int collision_test_needed = 0;
792
793 assert(data || obj_entry);
794
795 if (startup_info->have_repository) {
796 read_lock();
797 collision_test_needed = has_sha1_file_with_flags(sha1, HAS_SHA1_QUICK);
798 read_unlock();
799 }
800
801 if (collision_test_needed && !data) {
802 read_lock();
803 if (!check_collison(obj_entry))
804 collision_test_needed = 0;
805 read_unlock();
806 }
807 if (collision_test_needed) {
808 void *has_data;
809 enum object_type has_type;
810 unsigned long has_size;
811 read_lock();
812 has_type = sha1_object_info(sha1, &has_size);
813 if (has_type < 0)
814 die(_("cannot read existing object info %s"), sha1_to_hex(sha1));
815 if (has_type != type || has_size != size)
816 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
817 has_data = read_sha1_file(sha1, &has_type, &has_size);
818 read_unlock();
819 if (!data)
820 data = new_data = get_data_from_pack(obj_entry);
821 if (!has_data)
822 die(_("cannot read existing object %s"), sha1_to_hex(sha1));
823 if (size != has_size || type != has_type ||
824 memcmp(data, has_data, size) != 0)
825 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
826 free(has_data);
827 }
828
829 if (strict) {
830 read_lock();
831 if (type == OBJ_BLOB) {
832 struct blob *blob = lookup_blob(sha1);
833 if (blob)
834 blob->object.flags |= FLAG_CHECKED;
835 else
836 die(_("invalid blob object %s"), sha1_to_hex(sha1));
837 } else {
838 struct object *obj;
839 int eaten;
840 void *buf = (void *) data;
841
842 assert(data && "data can only be NULL for large _blobs_");
843
844 /*
845 * we do not need to free the memory here, as the
846 * buf is deleted by the caller.
847 */
848 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
849 if (!obj)
850 die(_("invalid %s"), typename(type));
851 if (do_fsck_object &&
852 fsck_object(obj, buf, size, &fsck_options))
853 die(_("Error in object"));
854 if (fsck_walk(obj, NULL, &fsck_options))
855 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
856
857 if (obj->type == OBJ_TREE) {
858 struct tree *item = (struct tree *) obj;
859 item->buffer = NULL;
860 obj->parsed = 0;
861 }
862 if (obj->type == OBJ_COMMIT) {
863 struct commit *commit = (struct commit *) obj;
864 if (detach_commit_buffer(commit, NULL) != data)
865 die("BUG: parse_object_buffer transmogrified our buffer");
866 }
867 obj->flags |= FLAG_CHECKED;
868 }
869 read_unlock();
870 }
871
872 free(new_data);
873 }
874
875 /*
876 * This function is part of find_unresolved_deltas(). There are two
877 * walkers going in the opposite ways.
878 *
879 * The first one in find_unresolved_deltas() traverses down from
880 * parent node to children, deflating nodes along the way. However,
881 * memory for deflated nodes is limited by delta_base_cache_limit, so
882 * at some point parent node's deflated content may be freed.
883 *
884 * The second walker is this function, which goes from current node up
885 * to top parent if necessary to deflate the node. In normal
886 * situation, its parent node would be already deflated, so it just
887 * needs to apply delta.
888 *
889 * In the worst case scenario, parent node is no longer deflated because
890 * we're running out of delta_base_cache_limit; we need to re-deflate
891 * parents, possibly up to the top base.
892 *
893 * All deflated objects here are subject to be freed if we exceed
894 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
895 * just need to make sure the last node is not freed.
896 */
897 static void *get_base_data(struct base_data *c)
898 {
899 if (!c->data) {
900 struct object_entry *obj = c->obj;
901 struct base_data **delta = NULL;
902 int delta_nr = 0, delta_alloc = 0;
903
904 while (is_delta_type(c->obj->type) && !c->data) {
905 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
906 delta[delta_nr++] = c;
907 c = c->base;
908 }
909 if (!delta_nr) {
910 c->data = get_data_from_pack(obj);
911 c->size = obj->size;
912 get_thread_data()->base_cache_used += c->size;
913 prune_base_data(c);
914 }
915 for (; delta_nr > 0; delta_nr--) {
916 void *base, *raw;
917 c = delta[delta_nr - 1];
918 obj = c->obj;
919 base = get_base_data(c->base);
920 raw = get_data_from_pack(obj);
921 c->data = patch_delta(
922 base, c->base->size,
923 raw, obj->size,
924 &c->size);
925 free(raw);
926 if (!c->data)
927 bad_object(obj->idx.offset, _("failed to apply delta"));
928 get_thread_data()->base_cache_used += c->size;
929 prune_base_data(c);
930 }
931 free(delta);
932 }
933 return c->data;
934 }
935
936 static void resolve_delta(struct object_entry *delta_obj,
937 struct base_data *base, struct base_data *result)
938 {
939 void *base_data, *delta_data;
940
941 if (show_stat) {
942 int i = delta_obj - objects;
943 int j = base->obj - objects;
944 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
945 deepest_delta_lock();
946 if (deepest_delta < obj_stat[i].delta_depth)
947 deepest_delta = obj_stat[i].delta_depth;
948 deepest_delta_unlock();
949 obj_stat[i].base_object_no = j;
950 }
951 delta_data = get_data_from_pack(delta_obj);
952 base_data = get_base_data(base);
953 result->obj = delta_obj;
954 result->data = patch_delta(base_data, base->size,
955 delta_data, delta_obj->size, &result->size);
956 free(delta_data);
957 if (!result->data)
958 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
959 hash_sha1_file(result->data, result->size,
960 typename(delta_obj->real_type),
961 delta_obj->idx.oid.hash);
962 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
963 delta_obj->idx.oid.hash);
964 counter_lock();
965 nr_resolved_deltas++;
966 counter_unlock();
967 }
968
969 /*
970 * Standard boolean compare-and-swap: atomically check whether "*type" is
971 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
972 * and return false.
973 */
974 static int compare_and_swap_type(signed char *type,
975 enum object_type want,
976 enum object_type set)
977 {
978 enum object_type old;
979
980 type_cas_lock();
981 old = *type;
982 if (old == want)
983 *type = set;
984 type_cas_unlock();
985
986 return old == want;
987 }
988
989 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
990 struct base_data *prev_base)
991 {
992 if (base->ref_last == -1 && base->ofs_last == -1) {
993 find_ref_delta_children(base->obj->idx.oid.hash,
994 &base->ref_first, &base->ref_last,
995 OBJ_REF_DELTA);
996
997 find_ofs_delta_children(base->obj->idx.offset,
998 &base->ofs_first, &base->ofs_last,
999 OBJ_OFS_DELTA);
1000
1001 if (base->ref_last == -1 && base->ofs_last == -1) {
1002 free(base->data);
1003 return NULL;
1004 }
1005
1006 link_base_data(prev_base, base);
1007 }
1008
1009 if (base->ref_first <= base->ref_last) {
1010 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1011 struct base_data *result = alloc_base_data();
1012
1013 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1014 base->obj->real_type))
1015 die("BUG: child->real_type != OBJ_REF_DELTA");
1016
1017 resolve_delta(child, base, result);
1018 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1019 free_base_data(base);
1020
1021 base->ref_first++;
1022 return result;
1023 }
1024
1025 if (base->ofs_first <= base->ofs_last) {
1026 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1027 struct base_data *result = alloc_base_data();
1028
1029 assert(child->real_type == OBJ_OFS_DELTA);
1030 child->real_type = base->obj->real_type;
1031 resolve_delta(child, base, result);
1032 if (base->ofs_first == base->ofs_last)
1033 free_base_data(base);
1034
1035 base->ofs_first++;
1036 return result;
1037 }
1038
1039 unlink_base_data(base);
1040 return NULL;
1041 }
1042
1043 static void find_unresolved_deltas(struct base_data *base)
1044 {
1045 struct base_data *new_base, *prev_base = NULL;
1046 for (;;) {
1047 new_base = find_unresolved_deltas_1(base, prev_base);
1048
1049 if (new_base) {
1050 prev_base = base;
1051 base = new_base;
1052 } else {
1053 free(base);
1054 base = prev_base;
1055 if (!base)
1056 return;
1057 prev_base = base->base;
1058 }
1059 }
1060 }
1061
1062 static int compare_ofs_delta_entry(const void *a, const void *b)
1063 {
1064 const struct ofs_delta_entry *delta_a = a;
1065 const struct ofs_delta_entry *delta_b = b;
1066
1067 return delta_a->offset < delta_b->offset ? -1 :
1068 delta_a->offset > delta_b->offset ? 1 :
1069 0;
1070 }
1071
1072 static int compare_ref_delta_entry(const void *a, const void *b)
1073 {
1074 const struct ref_delta_entry *delta_a = a;
1075 const struct ref_delta_entry *delta_b = b;
1076
1077 return hashcmp(delta_a->sha1, delta_b->sha1);
1078 }
1079
1080 static void resolve_base(struct object_entry *obj)
1081 {
1082 struct base_data *base_obj = alloc_base_data();
1083 base_obj->obj = obj;
1084 base_obj->data = NULL;
1085 find_unresolved_deltas(base_obj);
1086 }
1087
1088 #ifndef NO_PTHREADS
1089 static void *threaded_second_pass(void *data)
1090 {
1091 set_thread_data(data);
1092 for (;;) {
1093 int i;
1094 counter_lock();
1095 display_progress(progress, nr_resolved_deltas);
1096 counter_unlock();
1097 work_lock();
1098 while (nr_dispatched < nr_objects &&
1099 is_delta_type(objects[nr_dispatched].type))
1100 nr_dispatched++;
1101 if (nr_dispatched >= nr_objects) {
1102 work_unlock();
1103 break;
1104 }
1105 i = nr_dispatched++;
1106 work_unlock();
1107
1108 resolve_base(&objects[i]);
1109 }
1110 return NULL;
1111 }
1112 #endif
1113
1114 /*
1115 * First pass:
1116 * - find locations of all objects;
1117 * - calculate SHA1 of all non-delta objects;
1118 * - remember base (SHA1 or offset) for all deltas.
1119 */
1120 static void parse_pack_objects(unsigned char *sha1)
1121 {
1122 int i, nr_delays = 0;
1123 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1124 unsigned char ref_delta_sha1[20];
1125 struct stat st;
1126
1127 if (verbose)
1128 progress = start_progress(
1129 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1130 nr_objects);
1131 for (i = 0; i < nr_objects; i++) {
1132 struct object_entry *obj = &objects[i];
1133 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1134 ref_delta_sha1,
1135 obj->idx.oid.hash);
1136 obj->real_type = obj->type;
1137 if (obj->type == OBJ_OFS_DELTA) {
1138 nr_ofs_deltas++;
1139 ofs_delta->obj_no = i;
1140 ofs_delta++;
1141 } else if (obj->type == OBJ_REF_DELTA) {
1142 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1143 hashcpy(ref_deltas[nr_ref_deltas].sha1, ref_delta_sha1);
1144 ref_deltas[nr_ref_deltas].obj_no = i;
1145 nr_ref_deltas++;
1146 } else if (!data) {
1147 /* large blobs, check later */
1148 obj->real_type = OBJ_BAD;
1149 nr_delays++;
1150 } else
1151 sha1_object(data, NULL, obj->size, obj->type,
1152 obj->idx.oid.hash);
1153 free(data);
1154 display_progress(progress, i+1);
1155 }
1156 objects[i].idx.offset = consumed_bytes;
1157 stop_progress(&progress);
1158
1159 /* Check pack integrity */
1160 flush();
1161 git_SHA1_Final(sha1, &input_ctx);
1162 if (hashcmp(fill(20), sha1))
1163 die(_("pack is corrupted (SHA1 mismatch)"));
1164 use(20);
1165
1166 /* If input_fd is a file, we should have reached its end now. */
1167 if (fstat(input_fd, &st))
1168 die_errno(_("cannot fstat packfile"));
1169 if (S_ISREG(st.st_mode) &&
1170 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1171 die(_("pack has junk at the end"));
1172
1173 for (i = 0; i < nr_objects; i++) {
1174 struct object_entry *obj = &objects[i];
1175 if (obj->real_type != OBJ_BAD)
1176 continue;
1177 obj->real_type = obj->type;
1178 sha1_object(NULL, obj, obj->size, obj->type,
1179 obj->idx.oid.hash);
1180 nr_delays--;
1181 }
1182 if (nr_delays)
1183 die(_("confusion beyond insanity in parse_pack_objects()"));
1184 }
1185
1186 /*
1187 * Second pass:
1188 * - for all non-delta objects, look if it is used as a base for
1189 * deltas;
1190 * - if used as a base, uncompress the object and apply all deltas,
1191 * recursively checking if the resulting object is used as a base
1192 * for some more deltas.
1193 */
1194 static void resolve_deltas(void)
1195 {
1196 int i;
1197
1198 if (!nr_ofs_deltas && !nr_ref_deltas)
1199 return;
1200
1201 /* Sort deltas by base SHA1/offset for fast searching */
1202 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1203 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1204
1205 if (verbose || show_resolving_progress)
1206 progress = start_progress(_("Resolving deltas"),
1207 nr_ref_deltas + nr_ofs_deltas);
1208
1209 #ifndef NO_PTHREADS
1210 nr_dispatched = 0;
1211 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1212 init_thread();
1213 for (i = 0; i < nr_threads; i++) {
1214 int ret = pthread_create(&thread_data[i].thread, NULL,
1215 threaded_second_pass, thread_data + i);
1216 if (ret)
1217 die(_("unable to create thread: %s"),
1218 strerror(ret));
1219 }
1220 for (i = 0; i < nr_threads; i++)
1221 pthread_join(thread_data[i].thread, NULL);
1222 cleanup_thread();
1223 return;
1224 }
1225 #endif
1226
1227 for (i = 0; i < nr_objects; i++) {
1228 struct object_entry *obj = &objects[i];
1229
1230 if (is_delta_type(obj->type))
1231 continue;
1232 resolve_base(obj);
1233 display_progress(progress, nr_resolved_deltas);
1234 }
1235 }
1236
1237 /*
1238 * Third pass:
1239 * - append objects to convert thin pack to full pack if required
1240 * - write the final 20-byte SHA-1
1241 */
1242 static void fix_unresolved_deltas(struct sha1file *f);
1243 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1244 {
1245 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1246 stop_progress(&progress);
1247 /* Flush remaining pack final 20-byte SHA1. */
1248 flush();
1249 return;
1250 }
1251
1252 if (fix_thin_pack) {
1253 struct sha1file *f;
1254 unsigned char read_sha1[20], tail_sha1[20];
1255 struct strbuf msg = STRBUF_INIT;
1256 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1257 int nr_objects_initial = nr_objects;
1258 if (nr_unresolved <= 0)
1259 die(_("confusion beyond insanity"));
1260 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1261 memset(objects + nr_objects + 1, 0,
1262 nr_unresolved * sizeof(*objects));
1263 f = sha1fd(output_fd, curr_pack);
1264 fix_unresolved_deltas(f);
1265 strbuf_addf(&msg, Q_("completed with %d local object",
1266 "completed with %d local objects",
1267 nr_objects - nr_objects_initial),
1268 nr_objects - nr_objects_initial);
1269 stop_progress_msg(&progress, msg.buf);
1270 strbuf_release(&msg);
1271 sha1close(f, tail_sha1, 0);
1272 hashcpy(read_sha1, pack_sha1);
1273 fixup_pack_header_footer(output_fd, pack_sha1,
1274 curr_pack, nr_objects,
1275 read_sha1, consumed_bytes-20);
1276 if (hashcmp(read_sha1, tail_sha1) != 0)
1277 die(_("Unexpected tail checksum for %s "
1278 "(disk corruption?)"), curr_pack);
1279 }
1280 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1281 die(Q_("pack has %d unresolved delta",
1282 "pack has %d unresolved deltas",
1283 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1284 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1285 }
1286
1287 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1288 {
1289 git_zstream stream;
1290 int status;
1291 unsigned char outbuf[4096];
1292
1293 git_deflate_init(&stream, zlib_compression_level);
1294 stream.next_in = in;
1295 stream.avail_in = size;
1296
1297 do {
1298 stream.next_out = outbuf;
1299 stream.avail_out = sizeof(outbuf);
1300 status = git_deflate(&stream, Z_FINISH);
1301 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1302 } while (status == Z_OK);
1303
1304 if (status != Z_STREAM_END)
1305 die(_("unable to deflate appended object (%d)"), status);
1306 size = stream.total_out;
1307 git_deflate_end(&stream);
1308 return size;
1309 }
1310
1311 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1312 const unsigned char *sha1, void *buf,
1313 unsigned long size, enum object_type type)
1314 {
1315 struct object_entry *obj = &objects[nr_objects++];
1316 unsigned char header[10];
1317 unsigned long s = size;
1318 int n = 0;
1319 unsigned char c = (type << 4) | (s & 15);
1320 s >>= 4;
1321 while (s) {
1322 header[n++] = c | 0x80;
1323 c = s & 0x7f;
1324 s >>= 7;
1325 }
1326 header[n++] = c;
1327 crc32_begin(f);
1328 sha1write(f, header, n);
1329 obj[0].size = size;
1330 obj[0].hdr_size = n;
1331 obj[0].type = type;
1332 obj[0].real_type = type;
1333 obj[1].idx.offset = obj[0].idx.offset + n;
1334 obj[1].idx.offset += write_compressed(f, buf, size);
1335 obj[0].idx.crc32 = crc32_end(f);
1336 sha1flush(f);
1337 hashcpy(obj->idx.oid.hash, sha1);
1338 return obj;
1339 }
1340
1341 static int delta_pos_compare(const void *_a, const void *_b)
1342 {
1343 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1344 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1345 return a->obj_no - b->obj_no;
1346 }
1347
1348 static void fix_unresolved_deltas(struct sha1file *f)
1349 {
1350 struct ref_delta_entry **sorted_by_pos;
1351 int i;
1352
1353 /*
1354 * Since many unresolved deltas may well be themselves base objects
1355 * for more unresolved deltas, we really want to include the
1356 * smallest number of base objects that would cover as much delta
1357 * as possible by picking the
1358 * trunc deltas first, allowing for other deltas to resolve without
1359 * additional base objects. Since most base objects are to be found
1360 * before deltas depending on them, a good heuristic is to start
1361 * resolving deltas in the same order as their position in the pack.
1362 */
1363 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1364 for (i = 0; i < nr_ref_deltas; i++)
1365 sorted_by_pos[i] = &ref_deltas[i];
1366 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1367
1368 for (i = 0; i < nr_ref_deltas; i++) {
1369 struct ref_delta_entry *d = sorted_by_pos[i];
1370 enum object_type type;
1371 struct base_data *base_obj = alloc_base_data();
1372
1373 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1374 continue;
1375 base_obj->data = read_sha1_file(d->sha1, &type, &base_obj->size);
1376 if (!base_obj->data)
1377 continue;
1378
1379 if (check_sha1_signature(d->sha1, base_obj->data,
1380 base_obj->size, typename(type)))
1381 die(_("local object %s is corrupt"), sha1_to_hex(d->sha1));
1382 base_obj->obj = append_obj_to_pack(f, d->sha1,
1383 base_obj->data, base_obj->size, type);
1384 find_unresolved_deltas(base_obj);
1385 display_progress(progress, nr_resolved_deltas);
1386 }
1387 free(sorted_by_pos);
1388 }
1389
1390 static void final(const char *final_pack_name, const char *curr_pack_name,
1391 const char *final_index_name, const char *curr_index_name,
1392 const char *keep_name, const char *keep_msg,
1393 unsigned char *sha1)
1394 {
1395 const char *report = "pack";
1396 struct strbuf pack_name = STRBUF_INIT;
1397 struct strbuf index_name = STRBUF_INIT;
1398 struct strbuf keep_name_buf = STRBUF_INIT;
1399 int err;
1400
1401 if (!from_stdin) {
1402 close(input_fd);
1403 } else {
1404 fsync_or_die(output_fd, curr_pack_name);
1405 err = close(output_fd);
1406 if (err)
1407 die_errno(_("error while closing pack file"));
1408 }
1409
1410 if (keep_msg) {
1411 int keep_fd, keep_msg_len = strlen(keep_msg);
1412
1413 if (!keep_name)
1414 keep_name = odb_pack_name(&keep_name_buf, sha1, "keep");
1415
1416 keep_fd = odb_pack_keep(keep_name);
1417 if (keep_fd < 0) {
1418 if (errno != EEXIST)
1419 die_errno(_("cannot write keep file '%s'"),
1420 keep_name);
1421 } else {
1422 if (keep_msg_len > 0) {
1423 write_or_die(keep_fd, keep_msg, keep_msg_len);
1424 write_or_die(keep_fd, "\n", 1);
1425 }
1426 if (close(keep_fd) != 0)
1427 die_errno(_("cannot close written keep file '%s'"),
1428 keep_name);
1429 report = "keep";
1430 }
1431 }
1432
1433 if (final_pack_name != curr_pack_name) {
1434 if (!final_pack_name)
1435 final_pack_name = odb_pack_name(&pack_name, sha1, "pack");
1436 if (finalize_object_file(curr_pack_name, final_pack_name))
1437 die(_("cannot store pack file"));
1438 } else if (from_stdin)
1439 chmod(final_pack_name, 0444);
1440
1441 if (final_index_name != curr_index_name) {
1442 if (!final_index_name)
1443 final_index_name = odb_pack_name(&index_name, sha1, "idx");
1444 if (finalize_object_file(curr_index_name, final_index_name))
1445 die(_("cannot store index file"));
1446 } else
1447 chmod(final_index_name, 0444);
1448
1449 if (!from_stdin) {
1450 printf("%s\n", sha1_to_hex(sha1));
1451 } else {
1452 struct strbuf buf = STRBUF_INIT;
1453
1454 strbuf_addf(&buf, "%s\t%s\n", report, sha1_to_hex(sha1));
1455 write_or_die(1, buf.buf, buf.len);
1456 strbuf_release(&buf);
1457
1458 /*
1459 * Let's just mimic git-unpack-objects here and write
1460 * the last part of the input buffer to stdout.
1461 */
1462 while (input_len) {
1463 err = xwrite(1, input_buffer + input_offset, input_len);
1464 if (err <= 0)
1465 break;
1466 input_len -= err;
1467 input_offset += err;
1468 }
1469 }
1470
1471 strbuf_release(&index_name);
1472 strbuf_release(&pack_name);
1473 strbuf_release(&keep_name_buf);
1474 }
1475
1476 static int git_index_pack_config(const char *k, const char *v, void *cb)
1477 {
1478 struct pack_idx_option *opts = cb;
1479
1480 if (!strcmp(k, "pack.indexversion")) {
1481 opts->version = git_config_int(k, v);
1482 if (opts->version > 2)
1483 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1484 return 0;
1485 }
1486 if (!strcmp(k, "pack.threads")) {
1487 nr_threads = git_config_int(k, v);
1488 if (nr_threads < 0)
1489 die(_("invalid number of threads specified (%d)"),
1490 nr_threads);
1491 #ifdef NO_PTHREADS
1492 if (nr_threads != 1)
1493 warning(_("no threads support, ignoring %s"), k);
1494 nr_threads = 1;
1495 #endif
1496 return 0;
1497 }
1498 return git_default_config(k, v, cb);
1499 }
1500
1501 static int cmp_uint32(const void *a_, const void *b_)
1502 {
1503 uint32_t a = *((uint32_t *)a_);
1504 uint32_t b = *((uint32_t *)b_);
1505
1506 return (a < b) ? -1 : (a != b);
1507 }
1508
1509 static void read_v2_anomalous_offsets(struct packed_git *p,
1510 struct pack_idx_option *opts)
1511 {
1512 const uint32_t *idx1, *idx2;
1513 uint32_t i;
1514
1515 /* The address of the 4-byte offset table */
1516 idx1 = (((const uint32_t *)p->index_data)
1517 + 2 /* 8-byte header */
1518 + 256 /* fan out */
1519 + 5 * p->num_objects /* 20-byte SHA-1 table */
1520 + p->num_objects /* CRC32 table */
1521 );
1522
1523 /* The address of the 8-byte offset table */
1524 idx2 = idx1 + p->num_objects;
1525
1526 for (i = 0; i < p->num_objects; i++) {
1527 uint32_t off = ntohl(idx1[i]);
1528 if (!(off & 0x80000000))
1529 continue;
1530 off = off & 0x7fffffff;
1531 check_pack_index_ptr(p, &idx2[off * 2]);
1532 if (idx2[off * 2])
1533 continue;
1534 /*
1535 * The real offset is ntohl(idx2[off * 2]) in high 4
1536 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1537 * octets. But idx2[off * 2] is Zero!!!
1538 */
1539 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1540 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1541 }
1542
1543 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1544 }
1545
1546 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1547 {
1548 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1549
1550 if (!p)
1551 die(_("Cannot open existing pack file '%s'"), pack_name);
1552 if (open_pack_index(p))
1553 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1554
1555 /* Read the attributes from the existing idx file */
1556 opts->version = p->index_version;
1557
1558 if (opts->version == 2)
1559 read_v2_anomalous_offsets(p, opts);
1560
1561 /*
1562 * Get rid of the idx file as we do not need it anymore.
1563 * NEEDSWORK: extract this bit from free_pack_by_name() in
1564 * sha1_file.c, perhaps? It shouldn't matter very much as we
1565 * know we haven't installed this pack (hence we never have
1566 * read anything from it).
1567 */
1568 close_pack_index(p);
1569 free(p);
1570 }
1571
1572 static void show_pack_info(int stat_only)
1573 {
1574 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1575 unsigned long *chain_histogram = NULL;
1576
1577 if (deepest_delta)
1578 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1579
1580 for (i = 0; i < nr_objects; i++) {
1581 struct object_entry *obj = &objects[i];
1582
1583 if (is_delta_type(obj->type))
1584 chain_histogram[obj_stat[i].delta_depth - 1]++;
1585 if (stat_only)
1586 continue;
1587 printf("%s %-6s %lu %lu %"PRIuMAX,
1588 oid_to_hex(&obj->idx.oid),
1589 typename(obj->real_type), obj->size,
1590 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1591 (uintmax_t)obj->idx.offset);
1592 if (is_delta_type(obj->type)) {
1593 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1594 printf(" %u %s", obj_stat[i].delta_depth,
1595 oid_to_hex(&bobj->idx.oid));
1596 }
1597 putchar('\n');
1598 }
1599
1600 if (baseobjects)
1601 printf_ln(Q_("non delta: %d object",
1602 "non delta: %d objects",
1603 baseobjects),
1604 baseobjects);
1605 for (i = 0; i < deepest_delta; i++) {
1606 if (!chain_histogram[i])
1607 continue;
1608 printf_ln(Q_("chain length = %d: %lu object",
1609 "chain length = %d: %lu objects",
1610 chain_histogram[i]),
1611 i + 1,
1612 chain_histogram[i]);
1613 }
1614 }
1615
1616 static const char *derive_filename(const char *pack_name, const char *suffix,
1617 struct strbuf *buf)
1618 {
1619 size_t len;
1620 if (!strip_suffix(pack_name, ".pack", &len))
1621 die(_("packfile name '%s' does not end with '.pack'"),
1622 pack_name);
1623 strbuf_add(buf, pack_name, len);
1624 strbuf_addstr(buf, suffix);
1625 return buf->buf;
1626 }
1627
1628 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1629 {
1630 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1631 const char *curr_index;
1632 const char *index_name = NULL, *pack_name = NULL;
1633 const char *keep_name = NULL, *keep_msg = NULL;
1634 struct strbuf index_name_buf = STRBUF_INIT,
1635 keep_name_buf = STRBUF_INIT;
1636 struct pack_idx_entry **idx_objects;
1637 struct pack_idx_option opts;
1638 unsigned char pack_sha1[20];
1639 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1640 int report_end_of_input = 0;
1641
1642 if (argc == 2 && !strcmp(argv[1], "-h"))
1643 usage(index_pack_usage);
1644
1645 check_replace_refs = 0;
1646 fsck_options.walk = mark_link;
1647
1648 reset_pack_idx_option(&opts);
1649 git_config(git_index_pack_config, &opts);
1650 if (prefix && chdir(prefix))
1651 die(_("Cannot come back to cwd"));
1652
1653 for (i = 1; i < argc; i++) {
1654 const char *arg = argv[i];
1655
1656 if (*arg == '-') {
1657 if (!strcmp(arg, "--stdin")) {
1658 from_stdin = 1;
1659 } else if (!strcmp(arg, "--fix-thin")) {
1660 fix_thin_pack = 1;
1661 } else if (!strcmp(arg, "--strict")) {
1662 strict = 1;
1663 do_fsck_object = 1;
1664 } else if (skip_prefix(arg, "--strict=", &arg)) {
1665 strict = 1;
1666 do_fsck_object = 1;
1667 fsck_set_msg_types(&fsck_options, arg);
1668 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1669 strict = 1;
1670 check_self_contained_and_connected = 1;
1671 } else if (!strcmp(arg, "--verify")) {
1672 verify = 1;
1673 } else if (!strcmp(arg, "--verify-stat")) {
1674 verify = 1;
1675 show_stat = 1;
1676 } else if (!strcmp(arg, "--verify-stat-only")) {
1677 verify = 1;
1678 show_stat = 1;
1679 stat_only = 1;
1680 } else if (!strcmp(arg, "--keep")) {
1681 keep_msg = "";
1682 } else if (starts_with(arg, "--keep=")) {
1683 keep_msg = arg + 7;
1684 } else if (starts_with(arg, "--threads=")) {
1685 char *end;
1686 nr_threads = strtoul(arg+10, &end, 0);
1687 if (!arg[10] || *end || nr_threads < 0)
1688 usage(index_pack_usage);
1689 #ifdef NO_PTHREADS
1690 if (nr_threads != 1)
1691 warning(_("no threads support, "
1692 "ignoring %s"), arg);
1693 nr_threads = 1;
1694 #endif
1695 } else if (starts_with(arg, "--pack_header=")) {
1696 struct pack_header *hdr;
1697 char *c;
1698
1699 hdr = (struct pack_header *)input_buffer;
1700 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1701 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1702 if (*c != ',')
1703 die(_("bad %s"), arg);
1704 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1705 if (*c)
1706 die(_("bad %s"), arg);
1707 input_len = sizeof(*hdr);
1708 } else if (!strcmp(arg, "-v")) {
1709 verbose = 1;
1710 } else if (!strcmp(arg, "--show-resolving-progress")) {
1711 show_resolving_progress = 1;
1712 } else if (!strcmp(arg, "--report-end-of-input")) {
1713 report_end_of_input = 1;
1714 } else if (!strcmp(arg, "-o")) {
1715 if (index_name || (i+1) >= argc)
1716 usage(index_pack_usage);
1717 index_name = argv[++i];
1718 } else if (starts_with(arg, "--index-version=")) {
1719 char *c;
1720 opts.version = strtoul(arg + 16, &c, 10);
1721 if (opts.version > 2)
1722 die(_("bad %s"), arg);
1723 if (*c == ',')
1724 opts.off32_limit = strtoul(c+1, &c, 0);
1725 if (*c || opts.off32_limit & 0x80000000)
1726 die(_("bad %s"), arg);
1727 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1728 max_input_size = strtoumax(arg, NULL, 10);
1729 } else
1730 usage(index_pack_usage);
1731 continue;
1732 }
1733
1734 if (pack_name)
1735 usage(index_pack_usage);
1736 pack_name = arg;
1737 }
1738
1739 if (!pack_name && !from_stdin)
1740 usage(index_pack_usage);
1741 if (fix_thin_pack && !from_stdin)
1742 die(_("--fix-thin cannot be used without --stdin"));
1743 if (from_stdin && !startup_info->have_repository)
1744 die(_("--stdin requires a git repository"));
1745 if (!index_name && pack_name)
1746 index_name = derive_filename(pack_name, ".idx", &index_name_buf);
1747 if (keep_msg && !keep_name && pack_name)
1748 keep_name = derive_filename(pack_name, ".keep", &keep_name_buf);
1749
1750 if (verify) {
1751 if (!index_name)
1752 die(_("--verify with no packfile name given"));
1753 read_idx_option(&opts, index_name);
1754 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1755 }
1756 if (strict)
1757 opts.flags |= WRITE_IDX_STRICT;
1758
1759 #ifndef NO_PTHREADS
1760 if (!nr_threads) {
1761 nr_threads = online_cpus();
1762 /* An experiment showed that more threads does not mean faster */
1763 if (nr_threads > 3)
1764 nr_threads = 3;
1765 }
1766 #endif
1767
1768 curr_pack = open_pack_file(pack_name);
1769 parse_pack_header();
1770 objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1771 if (show_stat)
1772 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1773 ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1774 parse_pack_objects(pack_sha1);
1775 if (report_end_of_input)
1776 write_in_full(2, "\0", 1);
1777 resolve_deltas();
1778 conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1779 free(ofs_deltas);
1780 free(ref_deltas);
1781 if (strict)
1782 foreign_nr = check_objects();
1783
1784 if (show_stat)
1785 show_pack_info(stat_only);
1786
1787 ALLOC_ARRAY(idx_objects, nr_objects);
1788 for (i = 0; i < nr_objects; i++)
1789 idx_objects[i] = &objects[i].idx;
1790 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1791 free(idx_objects);
1792
1793 if (!verify)
1794 final(pack_name, curr_pack,
1795 index_name, curr_index,
1796 keep_name, keep_msg,
1797 pack_sha1);
1798 else
1799 close(input_fd);
1800 free(objects);
1801 strbuf_release(&index_name_buf);
1802 strbuf_release(&keep_name_buf);
1803 if (pack_name == NULL)
1804 free((void *) curr_pack);
1805 if (index_name == NULL)
1806 free((void *) curr_index);
1807
1808 /*
1809 * Let the caller know this pack is not self contained
1810 */
1811 if (check_self_contained_and_connected && foreign_nr)
1812 return 1;
1813
1814 return 0;
1815 }