?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
clone: add --sparse mode
[git.git] / builtin / index-pack.c
blob60a5591039abbdb9e0050bfda53123c577b41864
1 #include "builtin.h"
2 #include "config.h"
3 #include "delta.h"
4 #include "pack.h"
5 #include "csum-file.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "progress.h"
11 #include "fsck.h"
12 #include "exec-cmd.h"
13 #include "streaming.h"
14 #include "thread-utils.h"
15 #include "packfile.h"
16 #include "object-store.h"
17 #include "promisor-remote.h"
18
19 static const char index_pack_usage[] =
20 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
21
22 struct object_entry {
23 struct pack_idx_entry idx;
24 unsigned long size;
25 unsigned char hdr_size;
26 signed char type;
27 signed char real_type;
28 };
29
30 struct object_stat {
31 unsigned delta_depth;
32 int base_object_no;
33 };
34
35 struct base_data {
36 struct base_data *base;
37 struct base_data *child;
38 struct object_entry *obj;
39 void *data;
40 unsigned long size;
41 int ref_first, ref_last;
42 int ofs_first, ofs_last;
43 };
44
45 struct thread_local {
46 pthread_t thread;
47 struct base_data *base_cache;
48 size_t base_cache_used;
49 int pack_fd;
50 };
51
52 /* Remember to update object flag allocation in object.h */
53 #define FLAG_LINK (1u<<20)
54 #define FLAG_CHECKED (1u<<21)
55
56 struct ofs_delta_entry {
57 off_t offset;
58 int obj_no;
59 };
60
61 struct ref_delta_entry {
62 struct object_id oid;
63 int obj_no;
64 };
65
66 static struct object_entry *objects;
67 static struct object_stat *obj_stat;
68 static struct ofs_delta_entry *ofs_deltas;
69 static struct ref_delta_entry *ref_deltas;
70 static struct thread_local nothread_data;
71 static int nr_objects;
72 static int nr_ofs_deltas;
73 static int nr_ref_deltas;
74 static int ref_deltas_alloc;
75 static int nr_resolved_deltas;
76 static int nr_threads;
77
78 static int from_stdin;
79 static int strict;
80 static int do_fsck_object;
81 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
82 static int verbose;
83 static int show_resolving_progress;
84 static int show_stat;
85 static int check_self_contained_and_connected;
86
87 static struct progress *progress;
88
89 /* We always read in 4kB chunks. */
90 static unsigned char input_buffer[4096];
91 static unsigned int input_offset, input_len;
92 static off_t consumed_bytes;
93 static off_t max_input_size;
94 static unsigned deepest_delta;
95 static git_hash_ctx input_ctx;
96 static uint32_t input_crc32;
97 static int input_fd, output_fd;
98 static const char *curr_pack;
99
100 static struct thread_local *thread_data;
101 static int nr_dispatched;
102 static int threads_active;
103
104 static pthread_mutex_t read_mutex;
105 #define read_lock() lock_mutex(&read_mutex)
106 #define read_unlock() unlock_mutex(&read_mutex)
107
108 static pthread_mutex_t counter_mutex;
109 #define counter_lock() lock_mutex(&counter_mutex)
110 #define counter_unlock() unlock_mutex(&counter_mutex)
111
112 static pthread_mutex_t work_mutex;
113 #define work_lock() lock_mutex(&work_mutex)
114 #define work_unlock() unlock_mutex(&work_mutex)
115
116 static pthread_mutex_t deepest_delta_mutex;
117 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
118 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
119
120 static pthread_mutex_t type_cas_mutex;
121 #define type_cas_lock() lock_mutex(&type_cas_mutex)
122 #define type_cas_unlock() unlock_mutex(&type_cas_mutex)
123
124 static pthread_key_t key;
125
126 static inline void lock_mutex(pthread_mutex_t *mutex)
127 {
128 if (threads_active)
129 pthread_mutex_lock(mutex);
130 }
131
132 static inline void unlock_mutex(pthread_mutex_t *mutex)
133 {
134 if (threads_active)
135 pthread_mutex_unlock(mutex);
136 }
137
138 /*
139 * Mutex and conditional variable can't be statically-initialized on Windows.
140 */
141 static void init_thread(void)
142 {
143 int i;
144 init_recursive_mutex(&read_mutex);
145 pthread_mutex_init(&counter_mutex, NULL);
146 pthread_mutex_init(&work_mutex, NULL);
147 pthread_mutex_init(&type_cas_mutex, NULL);
148 if (show_stat)
149 pthread_mutex_init(&deepest_delta_mutex, NULL);
150 pthread_key_create(&key, NULL);
151 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
152 for (i = 0; i < nr_threads; i++) {
153 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
154 if (thread_data[i].pack_fd == -1)
155 die_errno(_("unable to open %s"), curr_pack);
156 }
157
158 threads_active = 1;
159 }
160
161 static void cleanup_thread(void)
162 {
163 int i;
164 if (!threads_active)
165 return;
166 threads_active = 0;
167 pthread_mutex_destroy(&read_mutex);
168 pthread_mutex_destroy(&counter_mutex);
169 pthread_mutex_destroy(&work_mutex);
170 pthread_mutex_destroy(&type_cas_mutex);
171 if (show_stat)
172 pthread_mutex_destroy(&deepest_delta_mutex);
173 for (i = 0; i < nr_threads; i++)
174 close(thread_data[i].pack_fd);
175 pthread_key_delete(key);
176 free(thread_data);
177 }
178
179 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
180 {
181 if (!obj)
182 return -1;
183
184 if (type != OBJ_ANY && obj->type != type)
185 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
186
187 obj->flags |= FLAG_LINK;
188 return 0;
189 }
190
191 /* The content of each linked object must have been checked
192 or it must be already present in the object database */
193 static unsigned check_object(struct object *obj)
194 {
195 if (!obj)
196 return 0;
197
198 if (!(obj->flags & FLAG_LINK))
199 return 0;
200
201 if (!(obj->flags & FLAG_CHECKED)) {
202 unsigned long size;
203 int type = oid_object_info(the_repository, &obj->oid, &size);
204 if (type <= 0)
205 die(_("did not receive expected object %s"),
206 oid_to_hex(&obj->oid));
207 if (type != obj->type)
208 die(_("object %s: expected type %s, found %s"),
209 oid_to_hex(&obj->oid),
210 type_name(obj->type), type_name(type));
211 obj->flags |= FLAG_CHECKED;
212 return 1;
213 }
214
215 return 0;
216 }
217
218 static unsigned check_objects(void)
219 {
220 unsigned i, max, foreign_nr = 0;
221
222 max = get_max_object_index();
223
224 if (verbose)
225 progress = start_delayed_progress(_("Checking objects"), max);
226
227 for (i = 0; i < max; i++) {
228 foreign_nr += check_object(get_indexed_object(i));
229 display_progress(progress, i + 1);
230 }
231
232 stop_progress(&progress);
233 return foreign_nr;
234 }
235
236
237 /* Discard current buffer used content. */
238 static void flush(void)
239 {
240 if (input_offset) {
241 if (output_fd >= 0)
242 write_or_die(output_fd, input_buffer, input_offset);
243 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
244 memmove(input_buffer, input_buffer + input_offset, input_len);
245 input_offset = 0;
246 }
247 }
248
249 /*
250 * Make sure at least "min" bytes are available in the buffer, and
251 * return the pointer to the buffer.
252 */
253 static void *fill(int min)
254 {
255 if (min <= input_len)
256 return input_buffer + input_offset;
257 if (min > sizeof(input_buffer))
258 die(Q_("cannot fill %d byte",
259 "cannot fill %d bytes",
260 min),
261 min);
262 flush();
263 do {
264 ssize_t ret = xread(input_fd, input_buffer + input_len,
265 sizeof(input_buffer) - input_len);
266 if (ret <= 0) {
267 if (!ret)
268 die(_("early EOF"));
269 die_errno(_("read error on input"));
270 }
271 input_len += ret;
272 if (from_stdin)
273 display_throughput(progress, consumed_bytes + input_len);
274 } while (input_len < min);
275 return input_buffer;
276 }
277
278 static void use(int bytes)
279 {
280 if (bytes > input_len)
281 die(_("used more bytes than were available"));
282 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
283 input_len -= bytes;
284 input_offset += bytes;
285
286 /* make sure off_t is sufficiently large not to wrap */
287 if (signed_add_overflows(consumed_bytes, bytes))
288 die(_("pack too large for current definition of off_t"));
289 consumed_bytes += bytes;
290 if (max_input_size && consumed_bytes > max_input_size)
291 die(_("pack exceeds maximum allowed size"));
292 }
293
294 static const char *open_pack_file(const char *pack_name)
295 {
296 if (from_stdin) {
297 input_fd = 0;
298 if (!pack_name) {
299 struct strbuf tmp_file = STRBUF_INIT;
300 output_fd = odb_mkstemp(&tmp_file,
301 "pack/tmp_pack_XXXXXX");
302 pack_name = strbuf_detach(&tmp_file, NULL);
303 } else {
304 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
305 if (output_fd < 0)
306 die_errno(_("unable to create '%s'"), pack_name);
307 }
308 nothread_data.pack_fd = output_fd;
309 } else {
310 input_fd = open(pack_name, O_RDONLY);
311 if (input_fd < 0)
312 die_errno(_("cannot open packfile '%s'"), pack_name);
313 output_fd = -1;
314 nothread_data.pack_fd = input_fd;
315 }
316 the_hash_algo->init_fn(&input_ctx);
317 return pack_name;
318 }
319
320 static void parse_pack_header(void)
321 {
322 struct pack_header *hdr = fill(sizeof(struct pack_header));
323
324 /* Header consistency check */
325 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
326 die(_("pack signature mismatch"));
327 if (!pack_version_ok(hdr->hdr_version))
328 die(_("pack version %"PRIu32" unsupported"),
329 ntohl(hdr->hdr_version));
330
331 nr_objects = ntohl(hdr->hdr_entries);
332 use(sizeof(struct pack_header));
333 }
334
335 static NORETURN void bad_object(off_t offset, const char *format,
336 ...) __attribute__((format (printf, 2, 3)));
337
338 static NORETURN void bad_object(off_t offset, const char *format, ...)
339 {
340 va_list params;
341 char buf[1024];
342
343 va_start(params, format);
344 vsnprintf(buf, sizeof(buf), format, params);
345 va_end(params);
346 die(_("pack has bad object at offset %"PRIuMAX": %s"),
347 (uintmax_t)offset, buf);
348 }
349
350 static inline struct thread_local *get_thread_data(void)
351 {
352 if (HAVE_THREADS) {
353 if (threads_active)
354 return pthread_getspecific(key);
355 assert(!threads_active &&
356 "This should only be reached when all threads are gone");
357 }
358 return &nothread_data;
359 }
360
361 static void set_thread_data(struct thread_local *data)
362 {
363 if (threads_active)
364 pthread_setspecific(key, data);
365 }
366
367 static struct base_data *alloc_base_data(void)
368 {
369 struct base_data *base = xcalloc(1, sizeof(struct base_data));
370 base->ref_last = -1;
371 base->ofs_last = -1;
372 return base;
373 }
374
375 static void free_base_data(struct base_data *c)
376 {
377 if (c->data) {
378 FREE_AND_NULL(c->data);
379 get_thread_data()->base_cache_used -= c->size;
380 }
381 }
382
383 static void prune_base_data(struct base_data *retain)
384 {
385 struct base_data *b;
386 struct thread_local *data = get_thread_data();
387 for (b = data->base_cache;
388 data->base_cache_used > delta_base_cache_limit && b;
389 b = b->child) {
390 if (b->data && b != retain)
391 free_base_data(b);
392 }
393 }
394
395 static void link_base_data(struct base_data *base, struct base_data *c)
396 {
397 if (base)
398 base->child = c;
399 else
400 get_thread_data()->base_cache = c;
401
402 c->base = base;
403 c->child = NULL;
404 if (c->data)
405 get_thread_data()->base_cache_used += c->size;
406 prune_base_data(c);
407 }
408
409 static void unlink_base_data(struct base_data *c)
410 {
411 struct base_data *base = c->base;
412 if (base)
413 base->child = NULL;
414 else
415 get_thread_data()->base_cache = NULL;
416 free_base_data(c);
417 }
418
419 static int is_delta_type(enum object_type type)
420 {
421 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
422 }
423
424 static void *unpack_entry_data(off_t offset, unsigned long size,
425 enum object_type type, struct object_id *oid)
426 {
427 static char fixed_buf[8192];
428 int status;
429 git_zstream stream;
430 void *buf;
431 git_hash_ctx c;
432 char hdr[32];
433 int hdrlen;
434
435 if (!is_delta_type(type)) {
436 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX,
437 type_name(type),(uintmax_t)size) + 1;
438 the_hash_algo->init_fn(&c);
439 the_hash_algo->update_fn(&c, hdr, hdrlen);
440 } else
441 oid = NULL;
442 if (type == OBJ_BLOB && size > big_file_threshold)
443 buf = fixed_buf;
444 else
445 buf = xmallocz(size);
446
447 memset(&stream, 0, sizeof(stream));
448 git_inflate_init(&stream);
449 stream.next_out = buf;
450 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
451
452 do {
453 unsigned char *last_out = stream.next_out;
454 stream.next_in = fill(1);
455 stream.avail_in = input_len;
456 status = git_inflate(&stream, 0);
457 use(input_len - stream.avail_in);
458 if (oid)
459 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
460 if (buf == fixed_buf) {
461 stream.next_out = buf;
462 stream.avail_out = sizeof(fixed_buf);
463 }
464 } while (status == Z_OK);
465 if (stream.total_out != size || status != Z_STREAM_END)
466 bad_object(offset, _("inflate returned %d"), status);
467 git_inflate_end(&stream);
468 if (oid)
469 the_hash_algo->final_fn(oid->hash, &c);
470 return buf == fixed_buf ? NULL : buf;
471 }
472
473 static void *unpack_raw_entry(struct object_entry *obj,
474 off_t *ofs_offset,
475 struct object_id *ref_oid,
476 struct object_id *oid)
477 {
478 unsigned char *p;
479 unsigned long size, c;
480 off_t base_offset;
481 unsigned shift;
482 void *data;
483
484 obj->idx.offset = consumed_bytes;
485 input_crc32 = crc32(0, NULL, 0);
486
487 p = fill(1);
488 c = *p;
489 use(1);
490 obj->type = (c >> 4) & 7;
491 size = (c & 15);
492 shift = 4;
493 while (c & 0x80) {
494 p = fill(1);
495 c = *p;
496 use(1);
497 size += (c & 0x7f) << shift;
498 shift += 7;
499 }
500 obj->size = size;
501
502 switch (obj->type) {
503 case OBJ_REF_DELTA:
504 hashcpy(ref_oid->hash, fill(the_hash_algo->rawsz));
505 use(the_hash_algo->rawsz);
506 break;
507 case OBJ_OFS_DELTA:
508 p = fill(1);
509 c = *p;
510 use(1);
511 base_offset = c & 127;
512 while (c & 128) {
513 base_offset += 1;
514 if (!base_offset || MSB(base_offset, 7))
515 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
516 p = fill(1);
517 c = *p;
518 use(1);
519 base_offset = (base_offset << 7) + (c & 127);
520 }
521 *ofs_offset = obj->idx.offset - base_offset;
522 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
523 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
524 break;
525 case OBJ_COMMIT:
526 case OBJ_TREE:
527 case OBJ_BLOB:
528 case OBJ_TAG:
529 break;
530 default:
531 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
532 }
533 obj->hdr_size = consumed_bytes - obj->idx.offset;
534
535 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
536 obj->idx.crc32 = input_crc32;
537 return data;
538 }
539
540 static void *unpack_data(struct object_entry *obj,
541 int (*consume)(const unsigned char *, unsigned long, void *),
542 void *cb_data)
543 {
544 off_t from = obj[0].idx.offset + obj[0].hdr_size;
545 off_t len = obj[1].idx.offset - from;
546 unsigned char *data, *inbuf;
547 git_zstream stream;
548 int status;
549
550 data = xmallocz(consume ? 64*1024 : obj->size);
551 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
552
553 memset(&stream, 0, sizeof(stream));
554 git_inflate_init(&stream);
555 stream.next_out = data;
556 stream.avail_out = consume ? 64*1024 : obj->size;
557
558 do {
559 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
560 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
561 if (n < 0)
562 die_errno(_("cannot pread pack file"));
563 if (!n)
564 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
565 "premature end of pack file, %"PRIuMAX" bytes missing",
566 (unsigned int)len),
567 (uintmax_t)len);
568 from += n;
569 len -= n;
570 stream.next_in = inbuf;
571 stream.avail_in = n;
572 if (!consume)
573 status = git_inflate(&stream, 0);
574 else {
575 do {
576 status = git_inflate(&stream, 0);
577 if (consume(data, stream.next_out - data, cb_data)) {
578 free(inbuf);
579 free(data);
580 return NULL;
581 }
582 stream.next_out = data;
583 stream.avail_out = 64*1024;
584 } while (status == Z_OK && stream.avail_in);
585 }
586 } while (len && status == Z_OK && !stream.avail_in);
587
588 /* This has been inflated OK when first encountered, so... */
589 if (status != Z_STREAM_END || stream.total_out != obj->size)
590 die(_("serious inflate inconsistency"));
591
592 git_inflate_end(&stream);
593 free(inbuf);
594 if (consume) {
595 FREE_AND_NULL(data);
596 }
597 return data;
598 }
599
600 static void *get_data_from_pack(struct object_entry *obj)
601 {
602 return unpack_data(obj, NULL, NULL);
603 }
604
605 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
606 enum object_type type1,
607 enum object_type type2)
608 {
609 int cmp = type1 - type2;
610 if (cmp)
611 return cmp;
612 return offset1 < offset2 ? -1 :
613 offset1 > offset2 ? 1 :
614 0;
615 }
616
617 static int find_ofs_delta(const off_t offset, enum object_type type)
618 {
619 int first = 0, last = nr_ofs_deltas;
620
621 while (first < last) {
622 int next = first + (last - first) / 2;
623 struct ofs_delta_entry *delta = &ofs_deltas[next];
624 int cmp;
625
626 cmp = compare_ofs_delta_bases(offset, delta->offset,
627 type, objects[delta->obj_no].type);
628 if (!cmp)
629 return next;
630 if (cmp < 0) {
631 last = next;
632 continue;
633 }
634 first = next+1;
635 }
636 return -first-1;
637 }
638
639 static void find_ofs_delta_children(off_t offset,
640 int *first_index, int *last_index,
641 enum object_type type)
642 {
643 int first = find_ofs_delta(offset, type);
644 int last = first;
645 int end = nr_ofs_deltas - 1;
646
647 if (first < 0) {
648 *first_index = 0;
649 *last_index = -1;
650 return;
651 }
652 while (first > 0 && ofs_deltas[first - 1].offset == offset)
653 --first;
654 while (last < end && ofs_deltas[last + 1].offset == offset)
655 ++last;
656 *first_index = first;
657 *last_index = last;
658 }
659
660 static int compare_ref_delta_bases(const struct object_id *oid1,
661 const struct object_id *oid2,
662 enum object_type type1,
663 enum object_type type2)
664 {
665 int cmp = type1 - type2;
666 if (cmp)
667 return cmp;
668 return oidcmp(oid1, oid2);
669 }
670
671 static int find_ref_delta(const struct object_id *oid, enum object_type type)
672 {
673 int first = 0, last = nr_ref_deltas;
674
675 while (first < last) {
676 int next = first + (last - first) / 2;
677 struct ref_delta_entry *delta = &ref_deltas[next];
678 int cmp;
679
680 cmp = compare_ref_delta_bases(oid, &delta->oid,
681 type, objects[delta->obj_no].type);
682 if (!cmp)
683 return next;
684 if (cmp < 0) {
685 last = next;
686 continue;
687 }
688 first = next+1;
689 }
690 return -first-1;
691 }
692
693 static void find_ref_delta_children(const struct object_id *oid,
694 int *first_index, int *last_index,
695 enum object_type type)
696 {
697 int first = find_ref_delta(oid, type);
698 int last = first;
699 int end = nr_ref_deltas - 1;
700
701 if (first < 0) {
702 *first_index = 0;
703 *last_index = -1;
704 return;
705 }
706 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
707 --first;
708 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
709 ++last;
710 *first_index = first;
711 *last_index = last;
712 }
713
714 struct compare_data {
715 struct object_entry *entry;
716 struct git_istream *st;
717 unsigned char *buf;
718 unsigned long buf_size;
719 };
720
721 static int compare_objects(const unsigned char *buf, unsigned long size,
722 void *cb_data)
723 {
724 struct compare_data *data = cb_data;
725
726 if (data->buf_size < size) {
727 free(data->buf);
728 data->buf = xmalloc(size);
729 data->buf_size = size;
730 }
731
732 while (size) {
733 ssize_t len = read_istream(data->st, data->buf, size);
734 if (len == 0)
735 die(_("SHA1 COLLISION FOUND WITH %s !"),
736 oid_to_hex(&data->entry->idx.oid));
737 if (len < 0)
738 die(_("unable to read %s"),
739 oid_to_hex(&data->entry->idx.oid));
740 if (memcmp(buf, data->buf, len))
741 die(_("SHA1 COLLISION FOUND WITH %s !"),
742 oid_to_hex(&data->entry->idx.oid));
743 size -= len;
744 buf += len;
745 }
746 return 0;
747 }
748
749 static int check_collison(struct object_entry *entry)
750 {
751 struct compare_data data;
752 enum object_type type;
753 unsigned long size;
754
755 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
756 return -1;
757
758 memset(&data, 0, sizeof(data));
759 data.entry = entry;
760 data.st = open_istream(&entry->idx.oid, &type, &size, NULL);
761 if (!data.st)
762 return -1;
763 if (size != entry->size || type != entry->type)
764 die(_("SHA1 COLLISION FOUND WITH %s !"),
765 oid_to_hex(&entry->idx.oid));
766 unpack_data(entry, compare_objects, &data);
767 close_istream(data.st);
768 free(data.buf);
769 return 0;
770 }
771
772 static void sha1_object(const void *data, struct object_entry *obj_entry,
773 unsigned long size, enum object_type type,
774 const struct object_id *oid)
775 {
776 void *new_data = NULL;
777 int collision_test_needed = 0;
778
779 assert(data || obj_entry);
780
781 if (startup_info->have_repository) {
782 read_lock();
783 collision_test_needed =
784 has_object_file_with_flags(oid, OBJECT_INFO_QUICK);
785 read_unlock();
786 }
787
788 if (collision_test_needed && !data) {
789 read_lock();
790 if (!check_collison(obj_entry))
791 collision_test_needed = 0;
792 read_unlock();
793 }
794 if (collision_test_needed) {
795 void *has_data;
796 enum object_type has_type;
797 unsigned long has_size;
798 read_lock();
799 has_type = oid_object_info(the_repository, oid, &has_size);
800 if (has_type < 0)
801 die(_("cannot read existing object info %s"), oid_to_hex(oid));
802 if (has_type != type || has_size != size)
803 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
804 has_data = read_object_file(oid, &has_type, &has_size);
805 read_unlock();
806 if (!data)
807 data = new_data = get_data_from_pack(obj_entry);
808 if (!has_data)
809 die(_("cannot read existing object %s"), oid_to_hex(oid));
810 if (size != has_size || type != has_type ||
811 memcmp(data, has_data, size) != 0)
812 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
813 free(has_data);
814 }
815
816 if (strict || do_fsck_object) {
817 read_lock();
818 if (type == OBJ_BLOB) {
819 struct blob *blob = lookup_blob(the_repository, oid);
820 if (blob)
821 blob->object.flags |= FLAG_CHECKED;
822 else
823 die(_("invalid blob object %s"), oid_to_hex(oid));
824 if (do_fsck_object &&
825 fsck_object(&blob->object, (void *)data, size, &fsck_options))
826 die(_("fsck error in packed object"));
827 } else {
828 struct object *obj;
829 int eaten;
830 void *buf = (void *) data;
831
832 assert(data && "data can only be NULL for large _blobs_");
833
834 /*
835 * we do not need to free the memory here, as the
836 * buf is deleted by the caller.
837 */
838 obj = parse_object_buffer(the_repository, oid, type,
839 size, buf,
840 &eaten);
841 if (!obj)
842 die(_("invalid %s"), type_name(type));
843 if (do_fsck_object &&
844 fsck_object(obj, buf, size, &fsck_options))
845 die(_("fsck error in packed object"));
846 if (strict && fsck_walk(obj, NULL, &fsck_options))
847 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
848
849 if (obj->type == OBJ_TREE) {
850 struct tree *item = (struct tree *) obj;
851 item->buffer = NULL;
852 obj->parsed = 0;
853 }
854 if (obj->type == OBJ_COMMIT) {
855 struct commit *commit = (struct commit *) obj;
856 if (detach_commit_buffer(commit, NULL) != data)
857 BUG("parse_object_buffer transmogrified our buffer");
858 }
859 obj->flags |= FLAG_CHECKED;
860 }
861 read_unlock();
862 }
863
864 free(new_data);
865 }
866
867 /*
868 * This function is part of find_unresolved_deltas(). There are two
869 * walkers going in the opposite ways.
870 *
871 * The first one in find_unresolved_deltas() traverses down from
872 * parent node to children, deflating nodes along the way. However,
873 * memory for deflated nodes is limited by delta_base_cache_limit, so
874 * at some point parent node's deflated content may be freed.
875 *
876 * The second walker is this function, which goes from current node up
877 * to top parent if necessary to deflate the node. In normal
878 * situation, its parent node would be already deflated, so it just
879 * needs to apply delta.
880 *
881 * In the worst case scenario, parent node is no longer deflated because
882 * we're running out of delta_base_cache_limit; we need to re-deflate
883 * parents, possibly up to the top base.
884 *
885 * All deflated objects here are subject to be freed if we exceed
886 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
887 * just need to make sure the last node is not freed.
888 */
889 static void *get_base_data(struct base_data *c)
890 {
891 if (!c->data) {
892 struct object_entry *obj = c->obj;
893 struct base_data **delta = NULL;
894 int delta_nr = 0, delta_alloc = 0;
895
896 while (is_delta_type(c->obj->type) && !c->data) {
897 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
898 delta[delta_nr++] = c;
899 c = c->base;
900 }
901 if (!delta_nr) {
902 c->data = get_data_from_pack(obj);
903 c->size = obj->size;
904 get_thread_data()->base_cache_used += c->size;
905 prune_base_data(c);
906 }
907 for (; delta_nr > 0; delta_nr--) {
908 void *base, *raw;
909 c = delta[delta_nr - 1];
910 obj = c->obj;
911 base = get_base_data(c->base);
912 raw = get_data_from_pack(obj);
913 c->data = patch_delta(
914 base, c->base->size,
915 raw, obj->size,
916 &c->size);
917 free(raw);
918 if (!c->data)
919 bad_object(obj->idx.offset, _("failed to apply delta"));
920 get_thread_data()->base_cache_used += c->size;
921 prune_base_data(c);
922 }
923 free(delta);
924 }
925 return c->data;
926 }
927
928 static void resolve_delta(struct object_entry *delta_obj,
929 struct base_data *base, struct base_data *result)
930 {
931 void *base_data, *delta_data;
932
933 if (show_stat) {
934 int i = delta_obj - objects;
935 int j = base->obj - objects;
936 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
937 deepest_delta_lock();
938 if (deepest_delta < obj_stat[i].delta_depth)
939 deepest_delta = obj_stat[i].delta_depth;
940 deepest_delta_unlock();
941 obj_stat[i].base_object_no = j;
942 }
943 delta_data = get_data_from_pack(delta_obj);
944 base_data = get_base_data(base);
945 result->obj = delta_obj;
946 result->data = patch_delta(base_data, base->size,
947 delta_data, delta_obj->size, &result->size);
948 free(delta_data);
949 if (!result->data)
950 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
951 hash_object_file(result->data, result->size,
952 type_name(delta_obj->real_type), &delta_obj->idx.oid);
953 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
954 &delta_obj->idx.oid);
955 counter_lock();
956 nr_resolved_deltas++;
957 counter_unlock();
958 }
959
960 /*
961 * Standard boolean compare-and-swap: atomically check whether "*type" is
962 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
963 * and return false.
964 */
965 static int compare_and_swap_type(signed char *type,
966 enum object_type want,
967 enum object_type set)
968 {
969 enum object_type old;
970
971 type_cas_lock();
972 old = *type;
973 if (old == want)
974 *type = set;
975 type_cas_unlock();
976
977 return old == want;
978 }
979
980 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
981 struct base_data *prev_base)
982 {
983 if (base->ref_last == -1 && base->ofs_last == -1) {
984 find_ref_delta_children(&base->obj->idx.oid,
985 &base->ref_first, &base->ref_last,
986 OBJ_REF_DELTA);
987
988 find_ofs_delta_children(base->obj->idx.offset,
989 &base->ofs_first, &base->ofs_last,
990 OBJ_OFS_DELTA);
991
992 if (base->ref_last == -1 && base->ofs_last == -1) {
993 free(base->data);
994 return NULL;
995 }
996
997 link_base_data(prev_base, base);
998 }
999
1000 if (base->ref_first <= base->ref_last) {
1001 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1002 struct base_data *result = alloc_base_data();
1003
1004 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1005 base->obj->real_type))
1006 BUG("child->real_type != OBJ_REF_DELTA");
1007
1008 resolve_delta(child, base, result);
1009 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1010 free_base_data(base);
1011
1012 base->ref_first++;
1013 return result;
1014 }
1015
1016 if (base->ofs_first <= base->ofs_last) {
1017 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1018 struct base_data *result = alloc_base_data();
1019
1020 assert(child->real_type == OBJ_OFS_DELTA);
1021 child->real_type = base->obj->real_type;
1022 resolve_delta(child, base, result);
1023 if (base->ofs_first == base->ofs_last)
1024 free_base_data(base);
1025
1026 base->ofs_first++;
1027 return result;
1028 }
1029
1030 unlink_base_data(base);
1031 return NULL;
1032 }
1033
1034 static void find_unresolved_deltas(struct base_data *base)
1035 {
1036 struct base_data *new_base, *prev_base = NULL;
1037 for (;;) {
1038 new_base = find_unresolved_deltas_1(base, prev_base);
1039
1040 if (new_base) {
1041 prev_base = base;
1042 base = new_base;
1043 } else {
1044 free(base);
1045 base = prev_base;
1046 if (!base)
1047 return;
1048 prev_base = base->base;
1049 }
1050 }
1051 }
1052
1053 static int compare_ofs_delta_entry(const void *a, const void *b)
1054 {
1055 const struct ofs_delta_entry *delta_a = a;
1056 const struct ofs_delta_entry *delta_b = b;
1057
1058 return delta_a->offset < delta_b->offset ? -1 :
1059 delta_a->offset > delta_b->offset ? 1 :
1060 0;
1061 }
1062
1063 static int compare_ref_delta_entry(const void *a, const void *b)
1064 {
1065 const struct ref_delta_entry *delta_a = a;
1066 const struct ref_delta_entry *delta_b = b;
1067
1068 return oidcmp(&delta_a->oid, &delta_b->oid);
1069 }
1070
1071 static void resolve_base(struct object_entry *obj)
1072 {
1073 struct base_data *base_obj = alloc_base_data();
1074 base_obj->obj = obj;
1075 base_obj->data = NULL;
1076 find_unresolved_deltas(base_obj);
1077 }
1078
1079 static void *threaded_second_pass(void *data)
1080 {
1081 set_thread_data(data);
1082 for (;;) {
1083 int i;
1084 counter_lock();
1085 display_progress(progress, nr_resolved_deltas);
1086 counter_unlock();
1087 work_lock();
1088 while (nr_dispatched < nr_objects &&
1089 is_delta_type(objects[nr_dispatched].type))
1090 nr_dispatched++;
1091 if (nr_dispatched >= nr_objects) {
1092 work_unlock();
1093 break;
1094 }
1095 i = nr_dispatched++;
1096 work_unlock();
1097
1098 resolve_base(&objects[i]);
1099 }
1100 return NULL;
1101 }
1102
1103 /*
1104 * First pass:
1105 * - find locations of all objects;
1106 * - calculate SHA1 of all non-delta objects;
1107 * - remember base (SHA1 or offset) for all deltas.
1108 */
1109 static void parse_pack_objects(unsigned char *hash)
1110 {
1111 int i, nr_delays = 0;
1112 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1113 struct object_id ref_delta_oid;
1114 struct stat st;
1115
1116 if (verbose)
1117 progress = start_progress(
1118 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1119 nr_objects);
1120 for (i = 0; i < nr_objects; i++) {
1121 struct object_entry *obj = &objects[i];
1122 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1123 &ref_delta_oid,
1124 &obj->idx.oid);
1125 obj->real_type = obj->type;
1126 if (obj->type == OBJ_OFS_DELTA) {
1127 nr_ofs_deltas++;
1128 ofs_delta->obj_no = i;
1129 ofs_delta++;
1130 } else if (obj->type == OBJ_REF_DELTA) {
1131 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1132 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1133 ref_deltas[nr_ref_deltas].obj_no = i;
1134 nr_ref_deltas++;
1135 } else if (!data) {
1136 /* large blobs, check later */
1137 obj->real_type = OBJ_BAD;
1138 nr_delays++;
1139 } else
1140 sha1_object(data, NULL, obj->size, obj->type,
1141 &obj->idx.oid);
1142 free(data);
1143 display_progress(progress, i+1);
1144 }
1145 objects[i].idx.offset = consumed_bytes;
1146 stop_progress(&progress);
1147
1148 /* Check pack integrity */
1149 flush();
1150 the_hash_algo->final_fn(hash, &input_ctx);
1151 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1152 die(_("pack is corrupted (SHA1 mismatch)"));
1153 use(the_hash_algo->rawsz);
1154
1155 /* If input_fd is a file, we should have reached its end now. */
1156 if (fstat(input_fd, &st))
1157 die_errno(_("cannot fstat packfile"));
1158 if (S_ISREG(st.st_mode) &&
1159 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1160 die(_("pack has junk at the end"));
1161
1162 for (i = 0; i < nr_objects; i++) {
1163 struct object_entry *obj = &objects[i];
1164 if (obj->real_type != OBJ_BAD)
1165 continue;
1166 obj->real_type = obj->type;
1167 sha1_object(NULL, obj, obj->size, obj->type,
1168 &obj->idx.oid);
1169 nr_delays--;
1170 }
1171 if (nr_delays)
1172 die(_("confusion beyond insanity in parse_pack_objects()"));
1173 }
1174
1175 /*
1176 * Second pass:
1177 * - for all non-delta objects, look if it is used as a base for
1178 * deltas;
1179 * - if used as a base, uncompress the object and apply all deltas,
1180 * recursively checking if the resulting object is used as a base
1181 * for some more deltas.
1182 */
1183 static void resolve_deltas(void)
1184 {
1185 int i;
1186
1187 if (!nr_ofs_deltas && !nr_ref_deltas)
1188 return;
1189
1190 /* Sort deltas by base SHA1/offset for fast searching */
1191 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1192 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1193
1194 if (verbose || show_resolving_progress)
1195 progress = start_progress(_("Resolving deltas"),
1196 nr_ref_deltas + nr_ofs_deltas);
1197
1198 nr_dispatched = 0;
1199 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1200 init_thread();
1201 for (i = 0; i < nr_threads; i++) {
1202 int ret = pthread_create(&thread_data[i].thread, NULL,
1203 threaded_second_pass, thread_data + i);
1204 if (ret)
1205 die(_("unable to create thread: %s"),
1206 strerror(ret));
1207 }
1208 for (i = 0; i < nr_threads; i++)
1209 pthread_join(thread_data[i].thread, NULL);
1210 cleanup_thread();
1211 return;
1212 }
1213
1214 for (i = 0; i < nr_objects; i++) {
1215 struct object_entry *obj = &objects[i];
1216
1217 if (is_delta_type(obj->type))
1218 continue;
1219 resolve_base(obj);
1220 display_progress(progress, nr_resolved_deltas);
1221 }
1222 }
1223
1224 /*
1225 * Third pass:
1226 * - append objects to convert thin pack to full pack if required
1227 * - write the final pack hash
1228 */
1229 static void fix_unresolved_deltas(struct hashfile *f);
1230 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1231 {
1232 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1233 stop_progress(&progress);
1234 /* Flush remaining pack final hash. */
1235 flush();
1236 return;
1237 }
1238
1239 if (fix_thin_pack) {
1240 struct hashfile *f;
1241 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1242 struct strbuf msg = STRBUF_INIT;
1243 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1244 int nr_objects_initial = nr_objects;
1245 if (nr_unresolved <= 0)
1246 die(_("confusion beyond insanity"));
1247 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1248 memset(objects + nr_objects + 1, 0,
1249 nr_unresolved * sizeof(*objects));
1250 f = hashfd(output_fd, curr_pack);
1251 fix_unresolved_deltas(f);
1252 strbuf_addf(&msg, Q_("completed with %d local object",
1253 "completed with %d local objects",
1254 nr_objects - nr_objects_initial),
1255 nr_objects - nr_objects_initial);
1256 stop_progress_msg(&progress, msg.buf);
1257 strbuf_release(&msg);
1258 finalize_hashfile(f, tail_hash, 0);
1259 hashcpy(read_hash, pack_hash);
1260 fixup_pack_header_footer(output_fd, pack_hash,
1261 curr_pack, nr_objects,
1262 read_hash, consumed_bytes-the_hash_algo->rawsz);
1263 if (!hasheq(read_hash, tail_hash))
1264 die(_("Unexpected tail checksum for %s "
1265 "(disk corruption?)"), curr_pack);
1266 }
1267 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1268 die(Q_("pack has %d unresolved delta",
1269 "pack has %d unresolved deltas",
1270 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1271 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1272 }
1273
1274 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1275 {
1276 git_zstream stream;
1277 int status;
1278 unsigned char outbuf[4096];
1279
1280 git_deflate_init(&stream, zlib_compression_level);
1281 stream.next_in = in;
1282 stream.avail_in = size;
1283
1284 do {
1285 stream.next_out = outbuf;
1286 stream.avail_out = sizeof(outbuf);
1287 status = git_deflate(&stream, Z_FINISH);
1288 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1289 } while (status == Z_OK);
1290
1291 if (status != Z_STREAM_END)
1292 die(_("unable to deflate appended object (%d)"), status);
1293 size = stream.total_out;
1294 git_deflate_end(&stream);
1295 return size;
1296 }
1297
1298 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1299 const unsigned char *sha1, void *buf,
1300 unsigned long size, enum object_type type)
1301 {
1302 struct object_entry *obj = &objects[nr_objects++];
1303 unsigned char header[10];
1304 unsigned long s = size;
1305 int n = 0;
1306 unsigned char c = (type << 4) | (s & 15);
1307 s >>= 4;
1308 while (s) {
1309 header[n++] = c | 0x80;
1310 c = s & 0x7f;
1311 s >>= 7;
1312 }
1313 header[n++] = c;
1314 crc32_begin(f);
1315 hashwrite(f, header, n);
1316 obj[0].size = size;
1317 obj[0].hdr_size = n;
1318 obj[0].type = type;
1319 obj[0].real_type = type;
1320 obj[1].idx.offset = obj[0].idx.offset + n;
1321 obj[1].idx.offset += write_compressed(f, buf, size);
1322 obj[0].idx.crc32 = crc32_end(f);
1323 hashflush(f);
1324 hashcpy(obj->idx.oid.hash, sha1);
1325 return obj;
1326 }
1327
1328 static int delta_pos_compare(const void *_a, const void *_b)
1329 {
1330 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1331 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1332 return a->obj_no - b->obj_no;
1333 }
1334
1335 static void fix_unresolved_deltas(struct hashfile *f)
1336 {
1337 struct ref_delta_entry **sorted_by_pos;
1338 int i;
1339
1340 /*
1341 * Since many unresolved deltas may well be themselves base objects
1342 * for more unresolved deltas, we really want to include the
1343 * smallest number of base objects that would cover as much delta
1344 * as possible by picking the
1345 * trunc deltas first, allowing for other deltas to resolve without
1346 * additional base objects. Since most base objects are to be found
1347 * before deltas depending on them, a good heuristic is to start
1348 * resolving deltas in the same order as their position in the pack.
1349 */
1350 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1351 for (i = 0; i < nr_ref_deltas; i++)
1352 sorted_by_pos[i] = &ref_deltas[i];
1353 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1354
1355 if (has_promisor_remote()) {
1356 /*
1357 * Prefetch the delta bases.
1358 */
1359 struct oid_array to_fetch = OID_ARRAY_INIT;
1360 for (i = 0; i < nr_ref_deltas; i++) {
1361 struct ref_delta_entry *d = sorted_by_pos[i];
1362 if (!oid_object_info_extended(the_repository, &d->oid,
1363 NULL,
1364 OBJECT_INFO_FOR_PREFETCH))
1365 continue;
1366 oid_array_append(&to_fetch, &d->oid);
1367 }
1368 if (to_fetch.nr)
1369 promisor_remote_get_direct(the_repository,
1370 to_fetch.oid, to_fetch.nr);
1371 oid_array_clear(&to_fetch);
1372 }
1373
1374 for (i = 0; i < nr_ref_deltas; i++) {
1375 struct ref_delta_entry *d = sorted_by_pos[i];
1376 enum object_type type;
1377 struct base_data *base_obj = alloc_base_data();
1378
1379 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1380 continue;
1381 base_obj->data = read_object_file(&d->oid, &type,
1382 &base_obj->size);
1383 if (!base_obj->data)
1384 continue;
1385
1386 if (check_object_signature(&d->oid, base_obj->data,
1387 base_obj->size, type_name(type)))
1388 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1389 base_obj->obj = append_obj_to_pack(f, d->oid.hash,
1390 base_obj->data, base_obj->size, type);
1391 find_unresolved_deltas(base_obj);
1392 display_progress(progress, nr_resolved_deltas);
1393 }
1394 free(sorted_by_pos);
1395 }
1396
1397 static const char *derive_filename(const char *pack_name, const char *suffix,
1398 struct strbuf *buf)
1399 {
1400 size_t len;
1401 if (!strip_suffix(pack_name, ".pack", &len))
1402 die(_("packfile name '%s' does not end with '.pack'"),
1403 pack_name);
1404 strbuf_add(buf, pack_name, len);
1405 strbuf_addch(buf, '.');
1406 strbuf_addstr(buf, suffix);
1407 return buf->buf;
1408 }
1409
1410 static void write_special_file(const char *suffix, const char *msg,
1411 const char *pack_name, const unsigned char *hash,
1412 const char **report)
1413 {
1414 struct strbuf name_buf = STRBUF_INIT;
1415 const char *filename;
1416 int fd;
1417 int msg_len = strlen(msg);
1418
1419 if (pack_name)
1420 filename = derive_filename(pack_name, suffix, &name_buf);
1421 else
1422 filename = odb_pack_name(&name_buf, hash, suffix);
1423
1424 fd = odb_pack_keep(filename);
1425 if (fd < 0) {
1426 if (errno != EEXIST)
1427 die_errno(_("cannot write %s file '%s'"),
1428 suffix, filename);
1429 } else {
1430 if (msg_len > 0) {
1431 write_or_die(fd, msg, msg_len);
1432 write_or_die(fd, "\n", 1);
1433 }
1434 if (close(fd) != 0)
1435 die_errno(_("cannot close written %s file '%s'"),
1436 suffix, filename);
1437 if (report)
1438 *report = suffix;
1439 }
1440 strbuf_release(&name_buf);
1441 }
1442
1443 static void final(const char *final_pack_name, const char *curr_pack_name,
1444 const char *final_index_name, const char *curr_index_name,
1445 const char *keep_msg, const char *promisor_msg,
1446 unsigned char *hash)
1447 {
1448 const char *report = "pack";
1449 struct strbuf pack_name = STRBUF_INIT;
1450 struct strbuf index_name = STRBUF_INIT;
1451 int err;
1452
1453 if (!from_stdin) {
1454 close(input_fd);
1455 } else {
1456 fsync_or_die(output_fd, curr_pack_name);
1457 err = close(output_fd);
1458 if (err)
1459 die_errno(_("error while closing pack file"));
1460 }
1461
1462 if (keep_msg)
1463 write_special_file("keep", keep_msg, final_pack_name, hash,
1464 &report);
1465 if (promisor_msg)
1466 write_special_file("promisor", promisor_msg, final_pack_name,
1467 hash, NULL);
1468
1469 if (final_pack_name != curr_pack_name) {
1470 if (!final_pack_name)
1471 final_pack_name = odb_pack_name(&pack_name, hash, "pack");
1472 if (finalize_object_file(curr_pack_name, final_pack_name))
1473 die(_("cannot store pack file"));
1474 } else if (from_stdin)
1475 chmod(final_pack_name, 0444);
1476
1477 if (final_index_name != curr_index_name) {
1478 if (!final_index_name)
1479 final_index_name = odb_pack_name(&index_name, hash, "idx");
1480 if (finalize_object_file(curr_index_name, final_index_name))
1481 die(_("cannot store index file"));
1482 } else
1483 chmod(final_index_name, 0444);
1484
1485 if (do_fsck_object) {
1486 struct packed_git *p;
1487 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1488 if (p)
1489 install_packed_git(the_repository, p);
1490 }
1491
1492 if (!from_stdin) {
1493 printf("%s\n", hash_to_hex(hash));
1494 } else {
1495 struct strbuf buf = STRBUF_INIT;
1496
1497 strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1498 write_or_die(1, buf.buf, buf.len);
1499 strbuf_release(&buf);
1500
1501 /*
1502 * Let's just mimic git-unpack-objects here and write
1503 * the last part of the input buffer to stdout.
1504 */
1505 while (input_len) {
1506 err = xwrite(1, input_buffer + input_offset, input_len);
1507 if (err <= 0)
1508 break;
1509 input_len -= err;
1510 input_offset += err;
1511 }
1512 }
1513
1514 strbuf_release(&index_name);
1515 strbuf_release(&pack_name);
1516 }
1517
1518 static int git_index_pack_config(const char *k, const char *v, void *cb)
1519 {
1520 struct pack_idx_option *opts = cb;
1521
1522 if (!strcmp(k, "pack.indexversion")) {
1523 opts->version = git_config_int(k, v);
1524 if (opts->version > 2)
1525 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1526 return 0;
1527 }
1528 if (!strcmp(k, "pack.threads")) {
1529 nr_threads = git_config_int(k, v);
1530 if (nr_threads < 0)
1531 die(_("invalid number of threads specified (%d)"),
1532 nr_threads);
1533 if (!HAVE_THREADS && nr_threads != 1) {
1534 warning(_("no threads support, ignoring %s"), k);
1535 nr_threads = 1;
1536 }
1537 return 0;
1538 }
1539 return git_default_config(k, v, cb);
1540 }
1541
1542 static int cmp_uint32(const void *a_, const void *b_)
1543 {
1544 uint32_t a = *((uint32_t *)a_);
1545 uint32_t b = *((uint32_t *)b_);
1546
1547 return (a < b) ? -1 : (a != b);
1548 }
1549
1550 static void read_v2_anomalous_offsets(struct packed_git *p,
1551 struct pack_idx_option *opts)
1552 {
1553 const uint32_t *idx1, *idx2;
1554 uint32_t i;
1555 const uint32_t hashwords = the_hash_algo->rawsz / sizeof(uint32_t);
1556
1557 /* The address of the 4-byte offset table */
1558 idx1 = (((const uint32_t *)p->index_data)
1559 + 2 /* 8-byte header */
1560 + 256 /* fan out */
1561 + hashwords * p->num_objects /* object ID table */
1562 + p->num_objects /* CRC32 table */
1563 );
1564
1565 /* The address of the 8-byte offset table */
1566 idx2 = idx1 + p->num_objects;
1567
1568 for (i = 0; i < p->num_objects; i++) {
1569 uint32_t off = ntohl(idx1[i]);
1570 if (!(off & 0x80000000))
1571 continue;
1572 off = off & 0x7fffffff;
1573 check_pack_index_ptr(p, &idx2[off * 2]);
1574 if (idx2[off * 2])
1575 continue;
1576 /*
1577 * The real offset is ntohl(idx2[off * 2]) in high 4
1578 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1579 * octets. But idx2[off * 2] is Zero!!!
1580 */
1581 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1582 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1583 }
1584
1585 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1586 }
1587
1588 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1589 {
1590 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1591
1592 if (!p)
1593 die(_("Cannot open existing pack file '%s'"), pack_name);
1594 if (open_pack_index(p))
1595 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1596
1597 /* Read the attributes from the existing idx file */
1598 opts->version = p->index_version;
1599
1600 if (opts->version == 2)
1601 read_v2_anomalous_offsets(p, opts);
1602
1603 /*
1604 * Get rid of the idx file as we do not need it anymore.
1605 * NEEDSWORK: extract this bit from free_pack_by_name() in
1606 * sha1-file.c, perhaps? It shouldn't matter very much as we
1607 * know we haven't installed this pack (hence we never have
1608 * read anything from it).
1609 */
1610 close_pack_index(p);
1611 free(p);
1612 }
1613
1614 static void show_pack_info(int stat_only)
1615 {
1616 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1617 unsigned long *chain_histogram = NULL;
1618
1619 if (deepest_delta)
1620 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1621
1622 for (i = 0; i < nr_objects; i++) {
1623 struct object_entry *obj = &objects[i];
1624
1625 if (is_delta_type(obj->type))
1626 chain_histogram[obj_stat[i].delta_depth - 1]++;
1627 if (stat_only)
1628 continue;
1629 printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1630 oid_to_hex(&obj->idx.oid),
1631 type_name(obj->real_type), (uintmax_t)obj->size,
1632 (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1633 (uintmax_t)obj->idx.offset);
1634 if (is_delta_type(obj->type)) {
1635 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1636 printf(" %u %s", obj_stat[i].delta_depth,
1637 oid_to_hex(&bobj->idx.oid));
1638 }
1639 putchar('\n');
1640 }
1641
1642 if (baseobjects)
1643 printf_ln(Q_("non delta: %d object",
1644 "non delta: %d objects",
1645 baseobjects),
1646 baseobjects);
1647 for (i = 0; i < deepest_delta; i++) {
1648 if (!chain_histogram[i])
1649 continue;
1650 printf_ln(Q_("chain length = %d: %lu object",
1651 "chain length = %d: %lu objects",
1652 chain_histogram[i]),
1653 i + 1,
1654 chain_histogram[i]);
1655 }
1656 }
1657
1658 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1659 {
1660 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1661 const char *curr_index;
1662 const char *index_name = NULL, *pack_name = NULL;
1663 const char *keep_msg = NULL;
1664 const char *promisor_msg = NULL;
1665 struct strbuf index_name_buf = STRBUF_INIT;
1666 struct pack_idx_entry **idx_objects;
1667 struct pack_idx_option opts;
1668 unsigned char pack_hash[GIT_MAX_RAWSZ];
1669 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1670 int report_end_of_input = 0;
1671
1672 /*
1673 * index-pack never needs to fetch missing objects except when
1674 * REF_DELTA bases are missing (which are explicitly handled). It only
1675 * accesses the repo to do hash collision checks and to check which
1676 * REF_DELTA bases need to be fetched.
1677 */
1678 fetch_if_missing = 0;
1679
1680 if (argc == 2 && !strcmp(argv[1], "-h"))
1681 usage(index_pack_usage);
1682
1683 read_replace_refs = 0;
1684 fsck_options.walk = mark_link;
1685
1686 reset_pack_idx_option(&opts);
1687 git_config(git_index_pack_config, &opts);
1688 if (prefix && chdir(prefix))
1689 die(_("Cannot come back to cwd"));
1690
1691 for (i = 1; i < argc; i++) {
1692 const char *arg = argv[i];
1693
1694 if (*arg == '-') {
1695 if (!strcmp(arg, "--stdin")) {
1696 from_stdin = 1;
1697 } else if (!strcmp(arg, "--fix-thin")) {
1698 fix_thin_pack = 1;
1699 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1700 strict = 1;
1701 do_fsck_object = 1;
1702 fsck_set_msg_types(&fsck_options, arg);
1703 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1704 strict = 1;
1705 check_self_contained_and_connected = 1;
1706 } else if (!strcmp(arg, "--fsck-objects")) {
1707 do_fsck_object = 1;
1708 } else if (!strcmp(arg, "--verify")) {
1709 verify = 1;
1710 } else if (!strcmp(arg, "--verify-stat")) {
1711 verify = 1;
1712 show_stat = 1;
1713 } else if (!strcmp(arg, "--verify-stat-only")) {
1714 verify = 1;
1715 show_stat = 1;
1716 stat_only = 1;
1717 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1718 ; /* nothing to do */
1719 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1720 ; /* already parsed */
1721 } else if (starts_with(arg, "--threads=")) {
1722 char *end;
1723 nr_threads = strtoul(arg+10, &end, 0);
1724 if (!arg[10] || *end || nr_threads < 0)
1725 usage(index_pack_usage);
1726 if (!HAVE_THREADS && nr_threads != 1) {
1727 warning(_("no threads support, ignoring %s"), arg);
1728 nr_threads = 1;
1729 }
1730 } else if (starts_with(arg, "--pack_header=")) {
1731 struct pack_header *hdr;
1732 char *c;
1733
1734 hdr = (struct pack_header *)input_buffer;
1735 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1736 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1737 if (*c != ',')
1738 die(_("bad %s"), arg);
1739 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1740 if (*c)
1741 die(_("bad %s"), arg);
1742 input_len = sizeof(*hdr);
1743 } else if (!strcmp(arg, "-v")) {
1744 verbose = 1;
1745 } else if (!strcmp(arg, "--show-resolving-progress")) {
1746 show_resolving_progress = 1;
1747 } else if (!strcmp(arg, "--report-end-of-input")) {
1748 report_end_of_input = 1;
1749 } else if (!strcmp(arg, "-o")) {
1750 if (index_name || (i+1) >= argc)
1751 usage(index_pack_usage);
1752 index_name = argv[++i];
1753 } else if (starts_with(arg, "--index-version=")) {
1754 char *c;
1755 opts.version = strtoul(arg + 16, &c, 10);
1756 if (opts.version > 2)
1757 die(_("bad %s"), arg);
1758 if (*c == ',')
1759 opts.off32_limit = strtoul(c+1, &c, 0);
1760 if (*c || opts.off32_limit & 0x80000000)
1761 die(_("bad %s"), arg);
1762 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1763 max_input_size = strtoumax(arg, NULL, 10);
1764 } else
1765 usage(index_pack_usage);
1766 continue;
1767 }
1768
1769 if (pack_name)
1770 usage(index_pack_usage);
1771 pack_name = arg;
1772 }
1773
1774 if (!pack_name && !from_stdin)
1775 usage(index_pack_usage);
1776 if (fix_thin_pack && !from_stdin)
1777 die(_("--fix-thin cannot be used without --stdin"));
1778 if (from_stdin && !startup_info->have_repository)
1779 die(_("--stdin requires a git repository"));
1780 if (!index_name && pack_name)
1781 index_name = derive_filename(pack_name, "idx", &index_name_buf);
1782
1783 if (verify) {
1784 if (!index_name)
1785 die(_("--verify with no packfile name given"));
1786 read_idx_option(&opts, index_name);
1787 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1788 }
1789 if (strict)
1790 opts.flags |= WRITE_IDX_STRICT;
1791
1792 if (HAVE_THREADS && !nr_threads) {
1793 nr_threads = online_cpus();
1794 /* An experiment showed that more threads does not mean faster */
1795 if (nr_threads > 3)
1796 nr_threads = 3;
1797 }
1798
1799 curr_pack = open_pack_file(pack_name);
1800 parse_pack_header();
1801 objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1802 if (show_stat)
1803 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1804 ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1805 parse_pack_objects(pack_hash);
1806 if (report_end_of_input)
1807 write_in_full(2, "\0", 1);
1808 resolve_deltas();
1809 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1810 free(ofs_deltas);
1811 free(ref_deltas);
1812 if (strict)
1813 foreign_nr = check_objects();
1814
1815 if (show_stat)
1816 show_pack_info(stat_only);
1817
1818 ALLOC_ARRAY(idx_objects, nr_objects);
1819 for (i = 0; i < nr_objects; i++)
1820 idx_objects[i] = &objects[i].idx;
1821 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1822 free(idx_objects);
1823
1824 if (!verify)
1825 final(pack_name, curr_pack,
1826 index_name, curr_index,
1827 keep_msg, promisor_msg,
1828 pack_hash);
1829 else
1830 close(input_fd);
1831
1832 if (do_fsck_object && fsck_finish(&fsck_options))
1833 die(_("fsck error in pack objects"));
1834
1835 free(objects);
1836 strbuf_release(&index_name_buf);
1837 if (pack_name == NULL)
1838 free((void *) curr_pack);
1839 if (index_name == NULL)
1840 free((void *) curr_index);
1841
1842 /*
1843 * Let the caller know this pack is not self contained
1844 */
1845 if (check_self_contained_and_connected && foreign_nr)
1846 return 1;
1847
1848 return 0;
1849 }
The Best VCS Around