?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
t0301: fixes for windows compatibility
[git/debian.git] / builtin / index-pack.c
blob6cc48902170e32c578f50d27bba5af6ea8568ab8
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>] [--[no-]rev-index] [--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 /* Initialized by make_base(). */
37 struct base_data *base;
38 struct object_entry *obj;
39 int ref_first, ref_last;
40 int ofs_first, ofs_last;
41 /*
42 * Threads should increment retain_data if they are about to call
43 * patch_delta() using this struct's data as a base, and decrement this
44 * when they are done. While retain_data is nonzero, this struct's data
45 * will not be freed even if the delta base cache limit is exceeded.
46 */
47 int retain_data;
48 /*
49 * The number of direct children that have not been fully processed
50 * (entered work_head, entered done_head, left done_head). When this
51 * number reaches zero, this struct base_data can be freed.
52 */
53 int children_remaining;
54
55 /* Not initialized by make_base(). */
56 struct list_head list;
57 void *data;
58 unsigned long size;
59 };
60
61 /*
62 * Stack of struct base_data that have unprocessed children.
63 * threaded_second_pass() uses this as a source of work (the other being the
64 * objects array).
65 *
66 * Guarded by work_mutex.
67 */
68 static LIST_HEAD(work_head);
69
70 /*
71 * Stack of struct base_data that have children, all of whom have been
72 * processed or are being processed, and at least one child is being processed.
73 * These struct base_data must be kept around until the last child is
74 * processed.
75 *
76 * Guarded by work_mutex.
77 */
78 static LIST_HEAD(done_head);
79
80 /*
81 * All threads share one delta base cache.
82 *
83 * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only
84 * in a thread.
85 */
86 static size_t base_cache_used;
87 static size_t base_cache_limit;
88
89 struct thread_local {
90 pthread_t thread;
91 int pack_fd;
92 };
93
94 /* Remember to update object flag allocation in object.h */
95 #define FLAG_LINK (1u<<20)
96 #define FLAG_CHECKED (1u<<21)
97
98 struct ofs_delta_entry {
99 off_t offset;
100 int obj_no;
101 };
102
103 struct ref_delta_entry {
104 struct object_id oid;
105 int obj_no;
106 };
107
108 static struct object_entry *objects;
109 static struct object_stat *obj_stat;
110 static struct ofs_delta_entry *ofs_deltas;
111 static struct ref_delta_entry *ref_deltas;
112 static struct thread_local nothread_data;
113 static int nr_objects;
114 static int nr_ofs_deltas;
115 static int nr_ref_deltas;
116 static int ref_deltas_alloc;
117 static int nr_resolved_deltas;
118 static int nr_threads;
119
120 static int from_stdin;
121 static int strict;
122 static int do_fsck_object;
123 static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES;
124 static int verbose;
125 static int show_resolving_progress;
126 static int show_stat;
127 static int check_self_contained_and_connected;
128
129 static struct progress *progress;
130
131 /* We always read in 4kB chunks. */
132 static unsigned char input_buffer[4096];
133 static unsigned int input_offset, input_len;
134 static off_t consumed_bytes;
135 static off_t max_input_size;
136 static unsigned deepest_delta;
137 static git_hash_ctx input_ctx;
138 static uint32_t input_crc32;
139 static int input_fd, output_fd;
140 static const char *curr_pack;
141
142 static struct thread_local *thread_data;
143 static int nr_dispatched;
144 static int threads_active;
145
146 static pthread_mutex_t read_mutex;
147 #define read_lock() lock_mutex(&read_mutex)
148 #define read_unlock() unlock_mutex(&read_mutex)
149
150 static pthread_mutex_t counter_mutex;
151 #define counter_lock() lock_mutex(&counter_mutex)
152 #define counter_unlock() unlock_mutex(&counter_mutex)
153
154 static pthread_mutex_t work_mutex;
155 #define work_lock() lock_mutex(&work_mutex)
156 #define work_unlock() unlock_mutex(&work_mutex)
157
158 static pthread_mutex_t deepest_delta_mutex;
159 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
160 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
161
162 static pthread_key_t key;
163
164 static inline void lock_mutex(pthread_mutex_t *mutex)
165 {
166 if (threads_active)
167 pthread_mutex_lock(mutex);
168 }
169
170 static inline void unlock_mutex(pthread_mutex_t *mutex)
171 {
172 if (threads_active)
173 pthread_mutex_unlock(mutex);
174 }
175
176 /*
177 * Mutex and conditional variable can't be statically-initialized on Windows.
178 */
179 static void init_thread(void)
180 {
181 int i;
182 init_recursive_mutex(&read_mutex);
183 pthread_mutex_init(&counter_mutex, NULL);
184 pthread_mutex_init(&work_mutex, NULL);
185 if (show_stat)
186 pthread_mutex_init(&deepest_delta_mutex, NULL);
187 pthread_key_create(&key, NULL);
188 CALLOC_ARRAY(thread_data, nr_threads);
189 for (i = 0; i < nr_threads; i++) {
190 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
191 if (thread_data[i].pack_fd == -1)
192 die_errno(_("unable to open %s"), curr_pack);
193 }
194
195 threads_active = 1;
196 }
197
198 static void cleanup_thread(void)
199 {
200 int i;
201 if (!threads_active)
202 return;
203 threads_active = 0;
204 pthread_mutex_destroy(&read_mutex);
205 pthread_mutex_destroy(&counter_mutex);
206 pthread_mutex_destroy(&work_mutex);
207 if (show_stat)
208 pthread_mutex_destroy(&deepest_delta_mutex);
209 for (i = 0; i < nr_threads; i++)
210 close(thread_data[i].pack_fd);
211 pthread_key_delete(key);
212 free(thread_data);
213 }
214
215 static int mark_link(struct object *obj, enum object_type type,
216 void *data, struct fsck_options *options)
217 {
218 if (!obj)
219 return -1;
220
221 if (type != OBJ_ANY && obj->type != type)
222 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
223
224 obj->flags |= FLAG_LINK;
225 return 0;
226 }
227
228 /* The content of each linked object must have been checked
229 or it must be already present in the object database */
230 static unsigned check_object(struct object *obj)
231 {
232 if (!obj)
233 return 0;
234
235 if (!(obj->flags & FLAG_LINK))
236 return 0;
237
238 if (!(obj->flags & FLAG_CHECKED)) {
239 unsigned long size;
240 int type = oid_object_info(the_repository, &obj->oid, &size);
241 if (type <= 0)
242 die(_("did not receive expected object %s"),
243 oid_to_hex(&obj->oid));
244 if (type != obj->type)
245 die(_("object %s: expected type %s, found %s"),
246 oid_to_hex(&obj->oid),
247 type_name(obj->type), type_name(type));
248 obj->flags |= FLAG_CHECKED;
249 return 1;
250 }
251
252 return 0;
253 }
254
255 static unsigned check_objects(void)
256 {
257 unsigned i, max, foreign_nr = 0;
258
259 max = get_max_object_index();
260
261 if (verbose)
262 progress = start_delayed_progress(_("Checking objects"), max);
263
264 for (i = 0; i < max; i++) {
265 foreign_nr += check_object(get_indexed_object(i));
266 display_progress(progress, i + 1);
267 }
268
269 stop_progress(&progress);
270 return foreign_nr;
271 }
272
273
274 /* Discard current buffer used content. */
275 static void flush(void)
276 {
277 if (input_offset) {
278 if (output_fd >= 0)
279 write_or_die(output_fd, input_buffer, input_offset);
280 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
281 memmove(input_buffer, input_buffer + input_offset, input_len);
282 input_offset = 0;
283 }
284 }
285
286 /*
287 * Make sure at least "min" bytes are available in the buffer, and
288 * return the pointer to the buffer.
289 */
290 static void *fill(int min)
291 {
292 if (min <= input_len)
293 return input_buffer + input_offset;
294 if (min > sizeof(input_buffer))
295 die(Q_("cannot fill %d byte",
296 "cannot fill %d bytes",
297 min),
298 min);
299 flush();
300 do {
301 ssize_t ret = xread(input_fd, input_buffer + input_len,
302 sizeof(input_buffer) - input_len);
303 if (ret <= 0) {
304 if (!ret)
305 die(_("early EOF"));
306 die_errno(_("read error on input"));
307 }
308 input_len += ret;
309 if (from_stdin)
310 display_throughput(progress, consumed_bytes + input_len);
311 } while (input_len < min);
312 return input_buffer;
313 }
314
315 static void use(int bytes)
316 {
317 if (bytes > input_len)
318 die(_("used more bytes than were available"));
319 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
320 input_len -= bytes;
321 input_offset += bytes;
322
323 /* make sure off_t is sufficiently large not to wrap */
324 if (signed_add_overflows(consumed_bytes, bytes))
325 die(_("pack too large for current definition of off_t"));
326 consumed_bytes += bytes;
327 if (max_input_size && consumed_bytes > max_input_size)
328 die(_("pack exceeds maximum allowed size"));
329 }
330
331 static const char *open_pack_file(const char *pack_name)
332 {
333 if (from_stdin) {
334 input_fd = 0;
335 if (!pack_name) {
336 struct strbuf tmp_file = STRBUF_INIT;
337 output_fd = odb_mkstemp(&tmp_file,
338 "pack/tmp_pack_XXXXXX");
339 pack_name = strbuf_detach(&tmp_file, NULL);
340 } else {
341 output_fd = xopen(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
342 }
343 nothread_data.pack_fd = output_fd;
344 } else {
345 input_fd = xopen(pack_name, O_RDONLY);
346 output_fd = -1;
347 nothread_data.pack_fd = input_fd;
348 }
349 the_hash_algo->init_fn(&input_ctx);
350 return pack_name;
351 }
352
353 static void parse_pack_header(void)
354 {
355 struct pack_header *hdr = fill(sizeof(struct pack_header));
356
357 /* Header consistency check */
358 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
359 die(_("pack signature mismatch"));
360 if (!pack_version_ok(hdr->hdr_version))
361 die(_("pack version %"PRIu32" unsupported"),
362 ntohl(hdr->hdr_version));
363
364 nr_objects = ntohl(hdr->hdr_entries);
365 use(sizeof(struct pack_header));
366 }
367
368 __attribute__((format (printf, 2, 3)))
369 static NORETURN void bad_object(off_t offset, const char *format, ...)
370 {
371 va_list params;
372 char buf[1024];
373
374 va_start(params, format);
375 vsnprintf(buf, sizeof(buf), format, params);
376 va_end(params);
377 die(_("pack has bad object at offset %"PRIuMAX": %s"),
378 (uintmax_t)offset, buf);
379 }
380
381 static inline struct thread_local *get_thread_data(void)
382 {
383 if (HAVE_THREADS) {
384 if (threads_active)
385 return pthread_getspecific(key);
386 assert(!threads_active &&
387 "This should only be reached when all threads are gone");
388 }
389 return &nothread_data;
390 }
391
392 static void set_thread_data(struct thread_local *data)
393 {
394 if (threads_active)
395 pthread_setspecific(key, data);
396 }
397
398 static void free_base_data(struct base_data *c)
399 {
400 if (c->data) {
401 FREE_AND_NULL(c->data);
402 base_cache_used -= c->size;
403 }
404 }
405
406 static void prune_base_data(struct base_data *retain)
407 {
408 struct list_head *pos;
409
410 if (base_cache_used <= base_cache_limit)
411 return;
412
413 list_for_each_prev(pos, &done_head) {
414 struct base_data *b = list_entry(pos, struct base_data, list);
415 if (b->retain_data || b == retain)
416 continue;
417 if (b->data) {
418 free_base_data(b);
419 if (base_cache_used <= base_cache_limit)
420 return;
421 }
422 }
423
424 list_for_each_prev(pos, &work_head) {
425 struct base_data *b = list_entry(pos, struct base_data, list);
426 if (b->retain_data || b == retain)
427 continue;
428 if (b->data) {
429 free_base_data(b);
430 if (base_cache_used <= base_cache_limit)
431 return;
432 }
433 }
434 }
435
436 static int is_delta_type(enum object_type type)
437 {
438 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
439 }
440
441 static void *unpack_entry_data(off_t offset, unsigned long size,
442 enum object_type type, struct object_id *oid)
443 {
444 static char fixed_buf[8192];
445 int status;
446 git_zstream stream;
447 void *buf;
448 git_hash_ctx c;
449 char hdr[32];
450 int hdrlen;
451
452 if (!is_delta_type(type)) {
453 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX,
454 type_name(type),(uintmax_t)size) + 1;
455 the_hash_algo->init_fn(&c);
456 the_hash_algo->update_fn(&c, hdr, hdrlen);
457 } else
458 oid = NULL;
459 if (type == OBJ_BLOB && size > big_file_threshold)
460 buf = fixed_buf;
461 else
462 buf = xmallocz(size);
463
464 memset(&stream, 0, sizeof(stream));
465 git_inflate_init(&stream);
466 stream.next_out = buf;
467 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
468
469 do {
470 unsigned char *last_out = stream.next_out;
471 stream.next_in = fill(1);
472 stream.avail_in = input_len;
473 status = git_inflate(&stream, 0);
474 use(input_len - stream.avail_in);
475 if (oid)
476 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
477 if (buf == fixed_buf) {
478 stream.next_out = buf;
479 stream.avail_out = sizeof(fixed_buf);
480 }
481 } while (status == Z_OK);
482 if (stream.total_out != size || status != Z_STREAM_END)
483 bad_object(offset, _("inflate returned %d"), status);
484 git_inflate_end(&stream);
485 if (oid)
486 the_hash_algo->final_oid_fn(oid, &c);
487 return buf == fixed_buf ? NULL : buf;
488 }
489
490 static void *unpack_raw_entry(struct object_entry *obj,
491 off_t *ofs_offset,
492 struct object_id *ref_oid,
493 struct object_id *oid)
494 {
495 unsigned char *p;
496 unsigned long size, c;
497 off_t base_offset;
498 unsigned shift;
499 void *data;
500
501 obj->idx.offset = consumed_bytes;
502 input_crc32 = crc32(0, NULL, 0);
503
504 p = fill(1);
505 c = *p;
506 use(1);
507 obj->type = (c >> 4) & 7;
508 size = (c & 15);
509 shift = 4;
510 while (c & 0x80) {
511 p = fill(1);
512 c = *p;
513 use(1);
514 size += (c & 0x7f) << shift;
515 shift += 7;
516 }
517 obj->size = size;
518
519 switch (obj->type) {
520 case OBJ_REF_DELTA:
521 oidread(ref_oid, fill(the_hash_algo->rawsz));
522 use(the_hash_algo->rawsz);
523 break;
524 case OBJ_OFS_DELTA:
525 p = fill(1);
526 c = *p;
527 use(1);
528 base_offset = c & 127;
529 while (c & 128) {
530 base_offset += 1;
531 if (!base_offset || MSB(base_offset, 7))
532 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
533 p = fill(1);
534 c = *p;
535 use(1);
536 base_offset = (base_offset << 7) + (c & 127);
537 }
538 *ofs_offset = obj->idx.offset - base_offset;
539 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
540 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
541 break;
542 case OBJ_COMMIT:
543 case OBJ_TREE:
544 case OBJ_BLOB:
545 case OBJ_TAG:
546 break;
547 default:
548 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
549 }
550 obj->hdr_size = consumed_bytes - obj->idx.offset;
551
552 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
553 obj->idx.crc32 = input_crc32;
554 return data;
555 }
556
557 static void *unpack_data(struct object_entry *obj,
558 int (*consume)(const unsigned char *, unsigned long, void *),
559 void *cb_data)
560 {
561 off_t from = obj[0].idx.offset + obj[0].hdr_size;
562 off_t len = obj[1].idx.offset - from;
563 unsigned char *data, *inbuf;
564 git_zstream stream;
565 int status;
566
567 data = xmallocz(consume ? 64*1024 : obj->size);
568 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
569
570 memset(&stream, 0, sizeof(stream));
571 git_inflate_init(&stream);
572 stream.next_out = data;
573 stream.avail_out = consume ? 64*1024 : obj->size;
574
575 do {
576 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
577 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
578 if (n < 0)
579 die_errno(_("cannot pread pack file"));
580 if (!n)
581 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
582 "premature end of pack file, %"PRIuMAX" bytes missing",
583 (unsigned int)len),
584 (uintmax_t)len);
585 from += n;
586 len -= n;
587 stream.next_in = inbuf;
588 stream.avail_in = n;
589 if (!consume)
590 status = git_inflate(&stream, 0);
591 else {
592 do {
593 status = git_inflate(&stream, 0);
594 if (consume(data, stream.next_out - data, cb_data)) {
595 free(inbuf);
596 free(data);
597 return NULL;
598 }
599 stream.next_out = data;
600 stream.avail_out = 64*1024;
601 } while (status == Z_OK && stream.avail_in);
602 }
603 } while (len && status == Z_OK && !stream.avail_in);
604
605 /* This has been inflated OK when first encountered, so... */
606 if (status != Z_STREAM_END || stream.total_out != obj->size)
607 die(_("serious inflate inconsistency"));
608
609 git_inflate_end(&stream);
610 free(inbuf);
611 if (consume) {
612 FREE_AND_NULL(data);
613 }
614 return data;
615 }
616
617 static void *get_data_from_pack(struct object_entry *obj)
618 {
619 return unpack_data(obj, NULL, NULL);
620 }
621
622 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
623 enum object_type type1,
624 enum object_type type2)
625 {
626 int cmp = type1 - type2;
627 if (cmp)
628 return cmp;
629 return offset1 < offset2 ? -1 :
630 offset1 > offset2 ? 1 :
631 0;
632 }
633
634 static int find_ofs_delta(const off_t offset)
635 {
636 int first = 0, last = nr_ofs_deltas;
637
638 while (first < last) {
639 int next = first + (last - first) / 2;
640 struct ofs_delta_entry *delta = &ofs_deltas[next];
641 int cmp;
642
643 cmp = compare_ofs_delta_bases(offset, delta->offset,
644 OBJ_OFS_DELTA,
645 objects[delta->obj_no].type);
646 if (!cmp)
647 return next;
648 if (cmp < 0) {
649 last = next;
650 continue;
651 }
652 first = next+1;
653 }
654 return -first-1;
655 }
656
657 static void find_ofs_delta_children(off_t offset,
658 int *first_index, int *last_index)
659 {
660 int first = find_ofs_delta(offset);
661 int last = first;
662 int end = nr_ofs_deltas - 1;
663
664 if (first < 0) {
665 *first_index = 0;
666 *last_index = -1;
667 return;
668 }
669 while (first > 0 && ofs_deltas[first - 1].offset == offset)
670 --first;
671 while (last < end && ofs_deltas[last + 1].offset == offset)
672 ++last;
673 *first_index = first;
674 *last_index = last;
675 }
676
677 static int compare_ref_delta_bases(const struct object_id *oid1,
678 const struct object_id *oid2,
679 enum object_type type1,
680 enum object_type type2)
681 {
682 int cmp = type1 - type2;
683 if (cmp)
684 return cmp;
685 return oidcmp(oid1, oid2);
686 }
687
688 static int find_ref_delta(const struct object_id *oid)
689 {
690 int first = 0, last = nr_ref_deltas;
691
692 while (first < last) {
693 int next = first + (last - first) / 2;
694 struct ref_delta_entry *delta = &ref_deltas[next];
695 int cmp;
696
697 cmp = compare_ref_delta_bases(oid, &delta->oid,
698 OBJ_REF_DELTA,
699 objects[delta->obj_no].type);
700 if (!cmp)
701 return next;
702 if (cmp < 0) {
703 last = next;
704 continue;
705 }
706 first = next+1;
707 }
708 return -first-1;
709 }
710
711 static void find_ref_delta_children(const struct object_id *oid,
712 int *first_index, int *last_index)
713 {
714 int first = find_ref_delta(oid);
715 int last = first;
716 int end = nr_ref_deltas - 1;
717
718 if (first < 0) {
719 *first_index = 0;
720 *last_index = -1;
721 return;
722 }
723 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
724 --first;
725 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
726 ++last;
727 *first_index = first;
728 *last_index = last;
729 }
730
731 struct compare_data {
732 struct object_entry *entry;
733 struct git_istream *st;
734 unsigned char *buf;
735 unsigned long buf_size;
736 };
737
738 static int compare_objects(const unsigned char *buf, unsigned long size,
739 void *cb_data)
740 {
741 struct compare_data *data = cb_data;
742
743 if (data->buf_size < size) {
744 free(data->buf);
745 data->buf = xmalloc(size);
746 data->buf_size = size;
747 }
748
749 while (size) {
750 ssize_t len = read_istream(data->st, data->buf, size);
751 if (len == 0)
752 die(_("SHA1 COLLISION FOUND WITH %s !"),
753 oid_to_hex(&data->entry->idx.oid));
754 if (len < 0)
755 die(_("unable to read %s"),
756 oid_to_hex(&data->entry->idx.oid));
757 if (memcmp(buf, data->buf, len))
758 die(_("SHA1 COLLISION FOUND WITH %s !"),
759 oid_to_hex(&data->entry->idx.oid));
760 size -= len;
761 buf += len;
762 }
763 return 0;
764 }
765
766 static int check_collison(struct object_entry *entry)
767 {
768 struct compare_data data;
769 enum object_type type;
770 unsigned long size;
771
772 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
773 return -1;
774
775 memset(&data, 0, sizeof(data));
776 data.entry = entry;
777 data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
778 NULL);
779 if (!data.st)
780 return -1;
781 if (size != entry->size || type != entry->type)
782 die(_("SHA1 COLLISION FOUND WITH %s !"),
783 oid_to_hex(&entry->idx.oid));
784 unpack_data(entry, compare_objects, &data);
785 close_istream(data.st);
786 free(data.buf);
787 return 0;
788 }
789
790 static void sha1_object(const void *data, struct object_entry *obj_entry,
791 unsigned long size, enum object_type type,
792 const struct object_id *oid)
793 {
794 void *new_data = NULL;
795 int collision_test_needed = 0;
796
797 assert(data || obj_entry);
798
799 if (startup_info->have_repository) {
800 read_lock();
801 collision_test_needed =
802 has_object_file_with_flags(oid, OBJECT_INFO_QUICK);
803 read_unlock();
804 }
805
806 if (collision_test_needed && !data) {
807 read_lock();
808 if (!check_collison(obj_entry))
809 collision_test_needed = 0;
810 read_unlock();
811 }
812 if (collision_test_needed) {
813 void *has_data;
814 enum object_type has_type;
815 unsigned long has_size;
816 read_lock();
817 has_type = oid_object_info(the_repository, oid, &has_size);
818 if (has_type < 0)
819 die(_("cannot read existing object info %s"), oid_to_hex(oid));
820 if (has_type != type || has_size != size)
821 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
822 has_data = read_object_file(oid, &has_type, &has_size);
823 read_unlock();
824 if (!data)
825 data = new_data = get_data_from_pack(obj_entry);
826 if (!has_data)
827 die(_("cannot read existing object %s"), oid_to_hex(oid));
828 if (size != has_size || type != has_type ||
829 memcmp(data, has_data, size) != 0)
830 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
831 free(has_data);
832 }
833
834 if (strict || do_fsck_object) {
835 read_lock();
836 if (type == OBJ_BLOB) {
837 struct blob *blob = lookup_blob(the_repository, oid);
838 if (blob)
839 blob->object.flags |= FLAG_CHECKED;
840 else
841 die(_("invalid blob object %s"), oid_to_hex(oid));
842 if (do_fsck_object &&
843 fsck_object(&blob->object, (void *)data, size, &fsck_options))
844 die(_("fsck error in packed object"));
845 } else {
846 struct object *obj;
847 int eaten;
848 void *buf = (void *) data;
849
850 assert(data && "data can only be NULL for large _blobs_");
851
852 /*
853 * we do not need to free the memory here, as the
854 * buf is deleted by the caller.
855 */
856 obj = parse_object_buffer(the_repository, oid, type,
857 size, buf,
858 &eaten);
859 if (!obj)
860 die(_("invalid %s"), type_name(type));
861 if (do_fsck_object &&
862 fsck_object(obj, buf, size, &fsck_options))
863 die(_("fsck error in packed object"));
864 if (strict && fsck_walk(obj, NULL, &fsck_options))
865 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
866
867 if (obj->type == OBJ_TREE) {
868 struct tree *item = (struct tree *) obj;
869 item->buffer = NULL;
870 obj->parsed = 0;
871 }
872 if (obj->type == OBJ_COMMIT) {
873 struct commit *commit = (struct commit *) obj;
874 if (detach_commit_buffer(commit, NULL) != data)
875 BUG("parse_object_buffer transmogrified our buffer");
876 }
877 obj->flags |= FLAG_CHECKED;
878 }
879 read_unlock();
880 }
881
882 free(new_data);
883 }
884
885 /*
886 * Ensure that this node has been reconstructed and return its contents.
887 *
888 * In the typical and best case, this node would already be reconstructed
889 * (through the invocation to resolve_delta() in threaded_second_pass()) and it
890 * would not be pruned. However, if pruning of this node was necessary due to
891 * reaching delta_base_cache_limit, this function will find the closest
892 * ancestor with reconstructed data that has not been pruned (or if there is
893 * none, the ultimate base object), and reconstruct each node in the delta
894 * chain in order to generate the reconstructed data for this node.
895 */
896 static void *get_base_data(struct base_data *c)
897 {
898 if (!c->data) {
899 struct object_entry *obj = c->obj;
900 struct base_data **delta = NULL;
901 int delta_nr = 0, delta_alloc = 0;
902
903 while (is_delta_type(c->obj->type) && !c->data) {
904 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
905 delta[delta_nr++] = c;
906 c = c->base;
907 }
908 if (!delta_nr) {
909 c->data = get_data_from_pack(obj);
910 c->size = obj->size;
911 base_cache_used += c->size;
912 prune_base_data(c);
913 }
914 for (; delta_nr > 0; delta_nr--) {
915 void *base, *raw;
916 c = delta[delta_nr - 1];
917 obj = c->obj;
918 base = get_base_data(c->base);
919 raw = get_data_from_pack(obj);
920 c->data = patch_delta(
921 base, c->base->size,
922 raw, obj->size,
923 &c->size);
924 free(raw);
925 if (!c->data)
926 bad_object(obj->idx.offset, _("failed to apply delta"));
927 base_cache_used += c->size;
928 prune_base_data(c);
929 }
930 free(delta);
931 }
932 return c->data;
933 }
934
935 static struct base_data *make_base(struct object_entry *obj,
936 struct base_data *parent)
937 {
938 struct base_data *base = xcalloc(1, sizeof(struct base_data));
939 base->base = parent;
940 base->obj = obj;
941 find_ref_delta_children(&obj->idx.oid,
942 &base->ref_first, &base->ref_last);
943 find_ofs_delta_children(obj->idx.offset,
944 &base->ofs_first, &base->ofs_last);
945 base->children_remaining = base->ref_last - base->ref_first +
946 base->ofs_last - base->ofs_first + 2;
947 return base;
948 }
949
950 static struct base_data *resolve_delta(struct object_entry *delta_obj,
951 struct base_data *base)
952 {
953 void *delta_data, *result_data;
954 struct base_data *result;
955 unsigned long result_size;
956
957 if (show_stat) {
958 int i = delta_obj - objects;
959 int j = base->obj - objects;
960 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
961 deepest_delta_lock();
962 if (deepest_delta < obj_stat[i].delta_depth)
963 deepest_delta = obj_stat[i].delta_depth;
964 deepest_delta_unlock();
965 obj_stat[i].base_object_no = j;
966 }
967 delta_data = get_data_from_pack(delta_obj);
968 assert(base->data);
969 result_data = patch_delta(base->data, base->size,
970 delta_data, delta_obj->size, &result_size);
971 free(delta_data);
972 if (!result_data)
973 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
974 hash_object_file(the_hash_algo, result_data, result_size,
975 type_name(delta_obj->real_type), &delta_obj->idx.oid);
976 sha1_object(result_data, NULL, result_size, delta_obj->real_type,
977 &delta_obj->idx.oid);
978
979 result = make_base(delta_obj, base);
980 result->data = result_data;
981 result->size = result_size;
982
983 counter_lock();
984 nr_resolved_deltas++;
985 counter_unlock();
986
987 return result;
988 }
989
990 static int compare_ofs_delta_entry(const void *a, const void *b)
991 {
992 const struct ofs_delta_entry *delta_a = a;
993 const struct ofs_delta_entry *delta_b = b;
994
995 return delta_a->offset < delta_b->offset ? -1 :
996 delta_a->offset > delta_b->offset ? 1 :
997 0;
998 }
999
1000 static int compare_ref_delta_entry(const void *a, const void *b)
1001 {
1002 const struct ref_delta_entry *delta_a = a;
1003 const struct ref_delta_entry *delta_b = b;
1004
1005 return oidcmp(&delta_a->oid, &delta_b->oid);
1006 }
1007
1008 static void *threaded_second_pass(void *data)
1009 {
1010 if (data)
1011 set_thread_data(data);
1012 for (;;) {
1013 struct base_data *parent = NULL;
1014 struct object_entry *child_obj;
1015 struct base_data *child;
1016
1017 counter_lock();
1018 display_progress(progress, nr_resolved_deltas);
1019 counter_unlock();
1020
1021 work_lock();
1022 if (list_empty(&work_head)) {
1023 /*
1024 * Take an object from the object array.
1025 */
1026 while (nr_dispatched < nr_objects &&
1027 is_delta_type(objects[nr_dispatched].type))
1028 nr_dispatched++;
1029 if (nr_dispatched >= nr_objects) {
1030 work_unlock();
1031 break;
1032 }
1033 child_obj = &objects[nr_dispatched++];
1034 } else {
1035 /*
1036 * Peek at the top of the stack, and take a child from
1037 * it.
1038 */
1039 parent = list_first_entry(&work_head, struct base_data,
1040 list);
1041
1042 if (parent->ref_first <= parent->ref_last) {
1043 int offset = ref_deltas[parent->ref_first++].obj_no;
1044 child_obj = objects + offset;
1045 if (child_obj->real_type != OBJ_REF_DELTA)
1046 die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)",
1047 (uintmax_t) child_obj->idx.offset,
1048 oid_to_hex(&parent->obj->idx.oid));
1049 child_obj->real_type = parent->obj->real_type;
1050 } else {
1051 child_obj = objects +
1052 ofs_deltas[parent->ofs_first++].obj_no;
1053 assert(child_obj->real_type == OBJ_OFS_DELTA);
1054 child_obj->real_type = parent->obj->real_type;
1055 }
1056
1057 if (parent->ref_first > parent->ref_last &&
1058 parent->ofs_first > parent->ofs_last) {
1059 /*
1060 * This parent has run out of children, so move
1061 * it to done_head.
1062 */
1063 list_del(&parent->list);
1064 list_add(&parent->list, &done_head);
1065 }
1066
1067 /*
1068 * Ensure that the parent has data, since we will need
1069 * it later.
1070 *
1071 * NEEDSWORK: If parent data needs to be reloaded, this
1072 * prolongs the time that the current thread spends in
1073 * the mutex. A mitigating factor is that parent data
1074 * needs to be reloaded only if the delta base cache
1075 * limit is exceeded, so in the typical case, this does
1076 * not happen.
1077 */
1078 get_base_data(parent);
1079 parent->retain_data++;
1080 }
1081 work_unlock();
1082
1083 if (parent) {
1084 child = resolve_delta(child_obj, parent);
1085 if (!child->children_remaining)
1086 FREE_AND_NULL(child->data);
1087 } else {
1088 child = make_base(child_obj, NULL);
1089 if (child->children_remaining) {
1090 /*
1091 * Since this child has its own delta children,
1092 * we will need this data in the future.
1093 * Inflate now so that future iterations will
1094 * have access to this object's data while
1095 * outside the work mutex.
1096 */
1097 child->data = get_data_from_pack(child_obj);
1098 child->size = child_obj->size;
1099 }
1100 }
1101
1102 work_lock();
1103 if (parent)
1104 parent->retain_data--;
1105 if (child->data) {
1106 /*
1107 * This child has its own children, so add it to
1108 * work_head.
1109 */
1110 list_add(&child->list, &work_head);
1111 base_cache_used += child->size;
1112 prune_base_data(NULL);
1113 } else {
1114 /*
1115 * This child does not have its own children. It may be
1116 * the last descendant of its ancestors; free those
1117 * that we can.
1118 */
1119 struct base_data *p = parent;
1120
1121 while (p) {
1122 struct base_data *next_p;
1123
1124 p->children_remaining--;
1125 if (p->children_remaining)
1126 break;
1127
1128 next_p = p->base;
1129 free_base_data(p);
1130 list_del(&p->list);
1131 free(p);
1132
1133 p = next_p;
1134 }
1135 }
1136 work_unlock();
1137 }
1138 return NULL;
1139 }
1140
1141 /*
1142 * First pass:
1143 * - find locations of all objects;
1144 * - calculate SHA1 of all non-delta objects;
1145 * - remember base (SHA1 or offset) for all deltas.
1146 */
1147 static void parse_pack_objects(unsigned char *hash)
1148 {
1149 int i, nr_delays = 0;
1150 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1151 struct object_id ref_delta_oid;
1152 struct stat st;
1153
1154 if (verbose)
1155 progress = start_progress(
1156 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1157 nr_objects);
1158 for (i = 0; i < nr_objects; i++) {
1159 struct object_entry *obj = &objects[i];
1160 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1161 &ref_delta_oid,
1162 &obj->idx.oid);
1163 obj->real_type = obj->type;
1164 if (obj->type == OBJ_OFS_DELTA) {
1165 nr_ofs_deltas++;
1166 ofs_delta->obj_no = i;
1167 ofs_delta++;
1168 } else if (obj->type == OBJ_REF_DELTA) {
1169 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1170 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1171 ref_deltas[nr_ref_deltas].obj_no = i;
1172 nr_ref_deltas++;
1173 } else if (!data) {
1174 /* large blobs, check later */
1175 obj->real_type = OBJ_BAD;
1176 nr_delays++;
1177 } else
1178 sha1_object(data, NULL, obj->size, obj->type,
1179 &obj->idx.oid);
1180 free(data);
1181 display_progress(progress, i+1);
1182 }
1183 objects[i].idx.offset = consumed_bytes;
1184 stop_progress(&progress);
1185
1186 /* Check pack integrity */
1187 flush();
1188 the_hash_algo->final_fn(hash, &input_ctx);
1189 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1190 die(_("pack is corrupted (SHA1 mismatch)"));
1191 use(the_hash_algo->rawsz);
1192
1193 /* If input_fd is a file, we should have reached its end now. */
1194 if (fstat(input_fd, &st))
1195 die_errno(_("cannot fstat packfile"));
1196 if (S_ISREG(st.st_mode) &&
1197 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1198 die(_("pack has junk at the end"));
1199
1200 for (i = 0; i < nr_objects; i++) {
1201 struct object_entry *obj = &objects[i];
1202 if (obj->real_type != OBJ_BAD)
1203 continue;
1204 obj->real_type = obj->type;
1205 sha1_object(NULL, obj, obj->size, obj->type,
1206 &obj->idx.oid);
1207 nr_delays--;
1208 }
1209 if (nr_delays)
1210 die(_("confusion beyond insanity in parse_pack_objects()"));
1211 }
1212
1213 /*
1214 * Second pass:
1215 * - for all non-delta objects, look if it is used as a base for
1216 * deltas;
1217 * - if used as a base, uncompress the object and apply all deltas,
1218 * recursively checking if the resulting object is used as a base
1219 * for some more deltas.
1220 */
1221 static void resolve_deltas(void)
1222 {
1223 int i;
1224
1225 if (!nr_ofs_deltas && !nr_ref_deltas)
1226 return;
1227
1228 /* Sort deltas by base SHA1/offset for fast searching */
1229 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1230 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1231
1232 if (verbose || show_resolving_progress)
1233 progress = start_progress(_("Resolving deltas"),
1234 nr_ref_deltas + nr_ofs_deltas);
1235
1236 nr_dispatched = 0;
1237 base_cache_limit = delta_base_cache_limit * nr_threads;
1238 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1239 init_thread();
1240 for (i = 0; i < nr_threads; i++) {
1241 int ret = pthread_create(&thread_data[i].thread, NULL,
1242 threaded_second_pass, thread_data + i);
1243 if (ret)
1244 die(_("unable to create thread: %s"),
1245 strerror(ret));
1246 }
1247 for (i = 0; i < nr_threads; i++)
1248 pthread_join(thread_data[i].thread, NULL);
1249 cleanup_thread();
1250 return;
1251 }
1252 threaded_second_pass(&nothread_data);
1253 }
1254
1255 /*
1256 * Third pass:
1257 * - append objects to convert thin pack to full pack if required
1258 * - write the final pack hash
1259 */
1260 static void fix_unresolved_deltas(struct hashfile *f);
1261 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1262 {
1263 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1264 stop_progress(&progress);
1265 /* Flush remaining pack final hash. */
1266 flush();
1267 return;
1268 }
1269
1270 if (fix_thin_pack) {
1271 struct hashfile *f;
1272 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1273 struct strbuf msg = STRBUF_INIT;
1274 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1275 int nr_objects_initial = nr_objects;
1276 if (nr_unresolved <= 0)
1277 die(_("confusion beyond insanity"));
1278 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1279 memset(objects + nr_objects + 1, 0,
1280 nr_unresolved * sizeof(*objects));
1281 f = hashfd(output_fd, curr_pack);
1282 fix_unresolved_deltas(f);
1283 strbuf_addf(&msg, Q_("completed with %d local object",
1284 "completed with %d local objects",
1285 nr_objects - nr_objects_initial),
1286 nr_objects - nr_objects_initial);
1287 stop_progress_msg(&progress, msg.buf);
1288 strbuf_release(&msg);
1289 finalize_hashfile(f, tail_hash, 0);
1290 hashcpy(read_hash, pack_hash);
1291 fixup_pack_header_footer(output_fd, pack_hash,
1292 curr_pack, nr_objects,
1293 read_hash, consumed_bytes-the_hash_algo->rawsz);
1294 if (!hasheq(read_hash, tail_hash))
1295 die(_("Unexpected tail checksum for %s "
1296 "(disk corruption?)"), curr_pack);
1297 }
1298 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1299 die(Q_("pack has %d unresolved delta",
1300 "pack has %d unresolved deltas",
1301 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1302 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1303 }
1304
1305 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1306 {
1307 git_zstream stream;
1308 int status;
1309 unsigned char outbuf[4096];
1310
1311 git_deflate_init(&stream, zlib_compression_level);
1312 stream.next_in = in;
1313 stream.avail_in = size;
1314
1315 do {
1316 stream.next_out = outbuf;
1317 stream.avail_out = sizeof(outbuf);
1318 status = git_deflate(&stream, Z_FINISH);
1319 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1320 } while (status == Z_OK);
1321
1322 if (status != Z_STREAM_END)
1323 die(_("unable to deflate appended object (%d)"), status);
1324 size = stream.total_out;
1325 git_deflate_end(&stream);
1326 return size;
1327 }
1328
1329 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1330 const unsigned char *sha1, void *buf,
1331 unsigned long size, enum object_type type)
1332 {
1333 struct object_entry *obj = &objects[nr_objects++];
1334 unsigned char header[10];
1335 unsigned long s = size;
1336 int n = 0;
1337 unsigned char c = (type << 4) | (s & 15);
1338 s >>= 4;
1339 while (s) {
1340 header[n++] = c | 0x80;
1341 c = s & 0x7f;
1342 s >>= 7;
1343 }
1344 header[n++] = c;
1345 crc32_begin(f);
1346 hashwrite(f, header, n);
1347 obj[0].size = size;
1348 obj[0].hdr_size = n;
1349 obj[0].type = type;
1350 obj[0].real_type = type;
1351 obj[1].idx.offset = obj[0].idx.offset + n;
1352 obj[1].idx.offset += write_compressed(f, buf, size);
1353 obj[0].idx.crc32 = crc32_end(f);
1354 hashflush(f);
1355 oidread(&obj->idx.oid, sha1);
1356 return obj;
1357 }
1358
1359 static int delta_pos_compare(const void *_a, const void *_b)
1360 {
1361 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1362 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1363 return a->obj_no - b->obj_no;
1364 }
1365
1366 static void fix_unresolved_deltas(struct hashfile *f)
1367 {
1368 struct ref_delta_entry **sorted_by_pos;
1369 int i;
1370
1371 /*
1372 * Since many unresolved deltas may well be themselves base objects
1373 * for more unresolved deltas, we really want to include the
1374 * smallest number of base objects that would cover as much delta
1375 * as possible by picking the
1376 * trunc deltas first, allowing for other deltas to resolve without
1377 * additional base objects. Since most base objects are to be found
1378 * before deltas depending on them, a good heuristic is to start
1379 * resolving deltas in the same order as their position in the pack.
1380 */
1381 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1382 for (i = 0; i < nr_ref_deltas; i++)
1383 sorted_by_pos[i] = &ref_deltas[i];
1384 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1385
1386 if (has_promisor_remote()) {
1387 /*
1388 * Prefetch the delta bases.
1389 */
1390 struct oid_array to_fetch = OID_ARRAY_INIT;
1391 for (i = 0; i < nr_ref_deltas; i++) {
1392 struct ref_delta_entry *d = sorted_by_pos[i];
1393 if (!oid_object_info_extended(the_repository, &d->oid,
1394 NULL,
1395 OBJECT_INFO_FOR_PREFETCH))
1396 continue;
1397 oid_array_append(&to_fetch, &d->oid);
1398 }
1399 promisor_remote_get_direct(the_repository,
1400 to_fetch.oid, to_fetch.nr);
1401 oid_array_clear(&to_fetch);
1402 }
1403
1404 for (i = 0; i < nr_ref_deltas; i++) {
1405 struct ref_delta_entry *d = sorted_by_pos[i];
1406 enum object_type type;
1407 void *data;
1408 unsigned long size;
1409
1410 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1411 continue;
1412 data = read_object_file(&d->oid, &type, &size);
1413 if (!data)
1414 continue;
1415
1416 if (check_object_signature(the_repository, &d->oid,
1417 data, size,
1418 type_name(type)))
1419 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1420
1421 /*
1422 * Add this as an object to the objects array and call
1423 * threaded_second_pass() (which will pick up the added
1424 * object).
1425 */
1426 append_obj_to_pack(f, d->oid.hash, data, size, type);
1427 threaded_second_pass(NULL);
1428
1429 display_progress(progress, nr_resolved_deltas);
1430 }
1431 free(sorted_by_pos);
1432 }
1433
1434 static const char *derive_filename(const char *pack_name, const char *strip,
1435 const char *suffix, struct strbuf *buf)
1436 {
1437 size_t len;
1438 if (!strip_suffix(pack_name, strip, &len) || !len ||
1439 pack_name[len - 1] != '.')
1440 die(_("packfile name '%s' does not end with '.%s'"),
1441 pack_name, strip);
1442 strbuf_add(buf, pack_name, len);
1443 strbuf_addstr(buf, suffix);
1444 return buf->buf;
1445 }
1446
1447 static void write_special_file(const char *suffix, const char *msg,
1448 const char *pack_name, const unsigned char *hash,
1449 const char **report)
1450 {
1451 struct strbuf name_buf = STRBUF_INIT;
1452 const char *filename;
1453 int fd;
1454 int msg_len = strlen(msg);
1455
1456 if (pack_name)
1457 filename = derive_filename(pack_name, "pack", suffix, &name_buf);
1458 else
1459 filename = odb_pack_name(&name_buf, hash, suffix);
1460
1461 fd = odb_pack_keep(filename);
1462 if (fd < 0) {
1463 if (errno != EEXIST)
1464 die_errno(_("cannot write %s file '%s'"),
1465 suffix, filename);
1466 } else {
1467 if (msg_len > 0) {
1468 write_or_die(fd, msg, msg_len);
1469 write_or_die(fd, "\n", 1);
1470 }
1471 if (close(fd) != 0)
1472 die_errno(_("cannot close written %s file '%s'"),
1473 suffix, filename);
1474 if (report)
1475 *report = suffix;
1476 }
1477 strbuf_release(&name_buf);
1478 }
1479
1480 static void final(const char *final_pack_name, const char *curr_pack_name,
1481 const char *final_index_name, const char *curr_index_name,
1482 const char *final_rev_index_name, const char *curr_rev_index_name,
1483 const char *keep_msg, const char *promisor_msg,
1484 unsigned char *hash)
1485 {
1486 const char *report = "pack";
1487 struct strbuf pack_name = STRBUF_INIT;
1488 struct strbuf index_name = STRBUF_INIT;
1489 struct strbuf rev_index_name = STRBUF_INIT;
1490 int err;
1491
1492 if (!from_stdin) {
1493 close(input_fd);
1494 } else {
1495 fsync_or_die(output_fd, curr_pack_name);
1496 err = close(output_fd);
1497 if (err)
1498 die_errno(_("error while closing pack file"));
1499 }
1500
1501 if (keep_msg)
1502 write_special_file("keep", keep_msg, final_pack_name, hash,
1503 &report);
1504 if (promisor_msg)
1505 write_special_file("promisor", promisor_msg, final_pack_name,
1506 hash, NULL);
1507
1508 if (final_pack_name != curr_pack_name) {
1509 if (!final_pack_name)
1510 final_pack_name = odb_pack_name(&pack_name, hash, "pack");
1511 if (finalize_object_file(curr_pack_name, final_pack_name))
1512 die(_("cannot store pack file"));
1513 } else if (from_stdin)
1514 chmod(final_pack_name, 0444);
1515
1516 if (final_index_name != curr_index_name) {
1517 if (!final_index_name)
1518 final_index_name = odb_pack_name(&index_name, hash, "idx");
1519 if (finalize_object_file(curr_index_name, final_index_name))
1520 die(_("cannot store index file"));
1521 } else
1522 chmod(final_index_name, 0444);
1523
1524 if (curr_rev_index_name) {
1525 if (final_rev_index_name != curr_rev_index_name) {
1526 if (!final_rev_index_name)
1527 final_rev_index_name = odb_pack_name(&rev_index_name, hash, "rev");
1528 if (finalize_object_file(curr_rev_index_name, final_rev_index_name))
1529 die(_("cannot store reverse index file"));
1530 } else
1531 chmod(final_rev_index_name, 0444);
1532 }
1533
1534 if (do_fsck_object) {
1535 struct packed_git *p;
1536 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1537 if (p)
1538 install_packed_git(the_repository, p);
1539 }
1540
1541 if (!from_stdin) {
1542 printf("%s\n", hash_to_hex(hash));
1543 } else {
1544 struct strbuf buf = STRBUF_INIT;
1545
1546 strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1547 write_or_die(1, buf.buf, buf.len);
1548 strbuf_release(&buf);
1549
1550 /*
1551 * Let's just mimic git-unpack-objects here and write
1552 * the last part of the input buffer to stdout.
1553 */
1554 while (input_len) {
1555 err = xwrite(1, input_buffer + input_offset, input_len);
1556 if (err <= 0)
1557 break;
1558 input_len -= err;
1559 input_offset += err;
1560 }
1561 }
1562
1563 strbuf_release(&rev_index_name);
1564 strbuf_release(&index_name);
1565 strbuf_release(&pack_name);
1566 }
1567
1568 static int git_index_pack_config(const char *k, const char *v, void *cb)
1569 {
1570 struct pack_idx_option *opts = cb;
1571
1572 if (!strcmp(k, "pack.indexversion")) {
1573 opts->version = git_config_int(k, v);
1574 if (opts->version > 2)
1575 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1576 return 0;
1577 }
1578 if (!strcmp(k, "pack.threads")) {
1579 nr_threads = git_config_int(k, v);
1580 if (nr_threads < 0)
1581 die(_("invalid number of threads specified (%d)"),
1582 nr_threads);
1583 if (!HAVE_THREADS && nr_threads != 1) {
1584 warning(_("no threads support, ignoring %s"), k);
1585 nr_threads = 1;
1586 }
1587 return 0;
1588 }
1589 if (!strcmp(k, "pack.writereverseindex")) {
1590 if (git_config_bool(k, v))
1591 opts->flags |= WRITE_REV;
1592 else
1593 opts->flags &= ~WRITE_REV;
1594 }
1595 return git_default_config(k, v, cb);
1596 }
1597
1598 static int cmp_uint32(const void *a_, const void *b_)
1599 {
1600 uint32_t a = *((uint32_t *)a_);
1601 uint32_t b = *((uint32_t *)b_);
1602
1603 return (a < b) ? -1 : (a != b);
1604 }
1605
1606 static void read_v2_anomalous_offsets(struct packed_git *p,
1607 struct pack_idx_option *opts)
1608 {
1609 const uint32_t *idx1, *idx2;
1610 uint32_t i;
1611
1612 /* The address of the 4-byte offset table */
1613 idx1 = (((const uint32_t *)((const uint8_t *)p->index_data + p->crc_offset))
1614 + (size_t)p->num_objects /* CRC32 table */
1615 );
1616
1617 /* The address of the 8-byte offset table */
1618 idx2 = idx1 + p->num_objects;
1619
1620 for (i = 0; i < p->num_objects; i++) {
1621 uint32_t off = ntohl(idx1[i]);
1622 if (!(off & 0x80000000))
1623 continue;
1624 off = off & 0x7fffffff;
1625 check_pack_index_ptr(p, &idx2[off * 2]);
1626 if (idx2[off * 2])
1627 continue;
1628 /*
1629 * The real offset is ntohl(idx2[off * 2]) in high 4
1630 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1631 * octets. But idx2[off * 2] is Zero!!!
1632 */
1633 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1634 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1635 }
1636
1637 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1638 }
1639
1640 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1641 {
1642 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1643
1644 if (!p)
1645 die(_("Cannot open existing pack file '%s'"), pack_name);
1646 if (open_pack_index(p))
1647 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1648
1649 /* Read the attributes from the existing idx file */
1650 opts->version = p->index_version;
1651
1652 if (opts->version == 2)
1653 read_v2_anomalous_offsets(p, opts);
1654
1655 /*
1656 * Get rid of the idx file as we do not need it anymore.
1657 * NEEDSWORK: extract this bit from free_pack_by_name() in
1658 * object-file.c, perhaps? It shouldn't matter very much as we
1659 * know we haven't installed this pack (hence we never have
1660 * read anything from it).
1661 */
1662 close_pack_index(p);
1663 free(p);
1664 }
1665
1666 static void show_pack_info(int stat_only)
1667 {
1668 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1669 unsigned long *chain_histogram = NULL;
1670
1671 if (deepest_delta)
1672 CALLOC_ARRAY(chain_histogram, deepest_delta);
1673
1674 for (i = 0; i < nr_objects; i++) {
1675 struct object_entry *obj = &objects[i];
1676
1677 if (is_delta_type(obj->type))
1678 chain_histogram[obj_stat[i].delta_depth - 1]++;
1679 if (stat_only)
1680 continue;
1681 printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1682 oid_to_hex(&obj->idx.oid),
1683 type_name(obj->real_type), (uintmax_t)obj->size,
1684 (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1685 (uintmax_t)obj->idx.offset);
1686 if (is_delta_type(obj->type)) {
1687 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1688 printf(" %u %s", obj_stat[i].delta_depth,
1689 oid_to_hex(&bobj->idx.oid));
1690 }
1691 putchar('\n');
1692 }
1693
1694 if (baseobjects)
1695 printf_ln(Q_("non delta: %d object",
1696 "non delta: %d objects",
1697 baseobjects),
1698 baseobjects);
1699 for (i = 0; i < deepest_delta; i++) {
1700 if (!chain_histogram[i])
1701 continue;
1702 printf_ln(Q_("chain length = %d: %lu object",
1703 "chain length = %d: %lu objects",
1704 chain_histogram[i]),
1705 i + 1,
1706 chain_histogram[i]);
1707 }
1708 }
1709
1710 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1711 {
1712 int i, fix_thin_pack = 0, verify = 0, stat_only = 0, rev_index;
1713 const char *curr_index;
1714 const char *curr_rev_index = NULL;
1715 const char *index_name = NULL, *pack_name = NULL, *rev_index_name = NULL;
1716 const char *keep_msg = NULL;
1717 const char *promisor_msg = NULL;
1718 struct strbuf index_name_buf = STRBUF_INIT;
1719 struct strbuf rev_index_name_buf = STRBUF_INIT;
1720 struct pack_idx_entry **idx_objects;
1721 struct pack_idx_option opts;
1722 unsigned char pack_hash[GIT_MAX_RAWSZ];
1723 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1724 int report_end_of_input = 0;
1725 int hash_algo = 0;
1726
1727 /*
1728 * index-pack never needs to fetch missing objects except when
1729 * REF_DELTA bases are missing (which are explicitly handled). It only
1730 * accesses the repo to do hash collision checks and to check which
1731 * REF_DELTA bases need to be fetched.
1732 */
1733 fetch_if_missing = 0;
1734
1735 if (argc == 2 && !strcmp(argv[1], "-h"))
1736 usage(index_pack_usage);
1737
1738 read_replace_refs = 0;
1739 fsck_options.walk = mark_link;
1740
1741 reset_pack_idx_option(&opts);
1742 git_config(git_index_pack_config, &opts);
1743 if (prefix && chdir(prefix))
1744 die(_("Cannot come back to cwd"));
1745
1746 if (git_env_bool(GIT_TEST_WRITE_REV_INDEX, 0))
1747 rev_index = 1;
1748 else
1749 rev_index = !!(opts.flags & (WRITE_REV_VERIFY | WRITE_REV));
1750
1751 for (i = 1; i < argc; i++) {
1752 const char *arg = argv[i];
1753
1754 if (*arg == '-') {
1755 if (!strcmp(arg, "--stdin")) {
1756 from_stdin = 1;
1757 } else if (!strcmp(arg, "--fix-thin")) {
1758 fix_thin_pack = 1;
1759 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1760 strict = 1;
1761 do_fsck_object = 1;
1762 fsck_set_msg_types(&fsck_options, arg);
1763 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1764 strict = 1;
1765 check_self_contained_and_connected = 1;
1766 } else if (!strcmp(arg, "--fsck-objects")) {
1767 do_fsck_object = 1;
1768 } else if (!strcmp(arg, "--verify")) {
1769 verify = 1;
1770 } else if (!strcmp(arg, "--verify-stat")) {
1771 verify = 1;
1772 show_stat = 1;
1773 } else if (!strcmp(arg, "--verify-stat-only")) {
1774 verify = 1;
1775 show_stat = 1;
1776 stat_only = 1;
1777 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1778 ; /* nothing to do */
1779 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1780 ; /* already parsed */
1781 } else if (starts_with(arg, "--threads=")) {
1782 char *end;
1783 nr_threads = strtoul(arg+10, &end, 0);
1784 if (!arg[10] || *end || nr_threads < 0)
1785 usage(index_pack_usage);
1786 if (!HAVE_THREADS && nr_threads != 1) {
1787 warning(_("no threads support, ignoring %s"), arg);
1788 nr_threads = 1;
1789 }
1790 } else if (starts_with(arg, "--pack_header=")) {
1791 struct pack_header *hdr;
1792 char *c;
1793
1794 hdr = (struct pack_header *)input_buffer;
1795 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1796 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1797 if (*c != ',')
1798 die(_("bad %s"), arg);
1799 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1800 if (*c)
1801 die(_("bad %s"), arg);
1802 input_len = sizeof(*hdr);
1803 } else if (!strcmp(arg, "-v")) {
1804 verbose = 1;
1805 } else if (!strcmp(arg, "--show-resolving-progress")) {
1806 show_resolving_progress = 1;
1807 } else if (!strcmp(arg, "--report-end-of-input")) {
1808 report_end_of_input = 1;
1809 } else if (!strcmp(arg, "-o")) {
1810 if (index_name || (i+1) >= argc)
1811 usage(index_pack_usage);
1812 index_name = argv[++i];
1813 } else if (starts_with(arg, "--index-version=")) {
1814 char *c;
1815 opts.version = strtoul(arg + 16, &c, 10);
1816 if (opts.version > 2)
1817 die(_("bad %s"), arg);
1818 if (*c == ',')
1819 opts.off32_limit = strtoul(c+1, &c, 0);
1820 if (*c || opts.off32_limit & 0x80000000)
1821 die(_("bad %s"), arg);
1822 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1823 max_input_size = strtoumax(arg, NULL, 10);
1824 } else if (skip_prefix(arg, "--object-format=", &arg)) {
1825 hash_algo = hash_algo_by_name(arg);
1826 if (hash_algo == GIT_HASH_UNKNOWN)
1827 die(_("unknown hash algorithm '%s'"), arg);
1828 repo_set_hash_algo(the_repository, hash_algo);
1829 } else if (!strcmp(arg, "--rev-index")) {
1830 rev_index = 1;
1831 } else if (!strcmp(arg, "--no-rev-index")) {
1832 rev_index = 0;
1833 } else
1834 usage(index_pack_usage);
1835 continue;
1836 }
1837
1838 if (pack_name)
1839 usage(index_pack_usage);
1840 pack_name = arg;
1841 }
1842
1843 if (!pack_name && !from_stdin)
1844 usage(index_pack_usage);
1845 if (fix_thin_pack && !from_stdin)
1846 die(_("--fix-thin cannot be used without --stdin"));
1847 if (from_stdin && !startup_info->have_repository)
1848 die(_("--stdin requires a git repository"));
1849 if (from_stdin && hash_algo)
1850 die(_("--object-format cannot be used with --stdin"));
1851 if (!index_name && pack_name)
1852 index_name = derive_filename(pack_name, "pack", "idx", &index_name_buf);
1853
1854 opts.flags &= ~(WRITE_REV | WRITE_REV_VERIFY);
1855 if (rev_index) {
1856 opts.flags |= verify ? WRITE_REV_VERIFY : WRITE_REV;
1857 if (index_name)
1858 rev_index_name = derive_filename(index_name,
1859 "idx", "rev",
1860 &rev_index_name_buf);
1861 }
1862
1863 if (verify) {
1864 if (!index_name)
1865 die(_("--verify with no packfile name given"));
1866 read_idx_option(&opts, index_name);
1867 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1868 }
1869 if (strict)
1870 opts.flags |= WRITE_IDX_STRICT;
1871
1872 if (HAVE_THREADS && !nr_threads) {
1873 nr_threads = online_cpus();
1874 /*
1875 * Experiments show that going above 20 threads doesn't help,
1876 * no matter how many cores you have. Below that, we tend to
1877 * max at half the number of online_cpus(), presumably because
1878 * half of those are hyperthreads rather than full cores. We'll
1879 * never reduce the level below "3", though, to match a
1880 * historical value that nobody complained about.
1881 */
1882 if (nr_threads < 4)
1883 ; /* too few cores to consider capping */
1884 else if (nr_threads < 6)
1885 nr_threads = 3; /* historic cap */
1886 else if (nr_threads < 40)
1887 nr_threads /= 2;
1888 else
1889 nr_threads = 20; /* hard cap */
1890 }
1891
1892 curr_pack = open_pack_file(pack_name);
1893 parse_pack_header();
1894 CALLOC_ARRAY(objects, st_add(nr_objects, 1));
1895 if (show_stat)
1896 CALLOC_ARRAY(obj_stat, st_add(nr_objects, 1));
1897 CALLOC_ARRAY(ofs_deltas, nr_objects);
1898 parse_pack_objects(pack_hash);
1899 if (report_end_of_input)
1900 write_in_full(2, "\0", 1);
1901 resolve_deltas();
1902 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1903 free(ofs_deltas);
1904 free(ref_deltas);
1905 if (strict)
1906 foreign_nr = check_objects();
1907
1908 if (show_stat)
1909 show_pack_info(stat_only);
1910
1911 ALLOC_ARRAY(idx_objects, nr_objects);
1912 for (i = 0; i < nr_objects; i++)
1913 idx_objects[i] = &objects[i].idx;
1914 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1915 if (rev_index)
1916 curr_rev_index = write_rev_file(rev_index_name, idx_objects,
1917 nr_objects, pack_hash,
1918 opts.flags);
1919 free(idx_objects);
1920
1921 if (!verify)
1922 final(pack_name, curr_pack,
1923 index_name, curr_index,
1924 rev_index_name, curr_rev_index,
1925 keep_msg, promisor_msg,
1926 pack_hash);
1927 else
1928 close(input_fd);
1929
1930 if (do_fsck_object && fsck_finish(&fsck_options))
1931 die(_("fsck error in pack objects"));
1932
1933 free(objects);
1934 strbuf_release(&index_name_buf);
1935 strbuf_release(&rev_index_name_buf);
1936 if (pack_name == NULL)
1937 free((void *) curr_pack);
1938 if (index_name == NULL)
1939 free((void *) curr_index);
1940 if (rev_index_name == NULL)
1941 free((void *) curr_rev_index);
1942
1943 /*
1944 * Let the caller know this pack is not self contained
1945 */
1946 if (check_self_contained_and_connected && foreign_nr)
1947 return 1;
1948
1949 return 0;
1950 }
git for Debian