?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'en/header-split-cache-h-part-3'
[git/debian.git] / builtin / index-pack.c
blobc1250b070f5e7409ee8f98f6e4dcd1109e85e3c8
1 #include "builtin.h"
2 #include "alloc.h"
3 #include "config.h"
4 #include "delta.h"
5 #include "environment.h"
6 #include "gettext.h"
7 #include "hex.h"
8 #include "pack.h"
9 #include "csum-file.h"
10 #include "blob.h"
11 #include "commit.h"
12 #include "tag.h"
13 #include "tree.h"
14 #include "progress.h"
15 #include "fsck.h"
16 #include "exec-cmd.h"
17 #include "strbuf.h"
18 #include "streaming.h"
19 #include "thread-utils.h"
20 #include "packfile.h"
21 #include "pack-revindex.h"
22 #include "object-file.h"
23 #include "object-store-ll.h"
24 #include "oid-array.h"
25 #include "replace-object.h"
26 #include "promisor-remote.h"
27 #include "setup.h"
28 #include "wrapper.h"
29
30 static const char index_pack_usage[] =
31 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--[no-]rev-index] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
32
33 struct object_entry {
34 struct pack_idx_entry idx;
35 unsigned long size;
36 unsigned char hdr_size;
37 signed char type;
38 signed char real_type;
39 };
40
41 struct object_stat {
42 unsigned delta_depth;
43 int base_object_no;
44 };
45
46 struct base_data {
47 /* Initialized by make_base(). */
48 struct base_data *base;
49 struct object_entry *obj;
50 int ref_first, ref_last;
51 int ofs_first, ofs_last;
52 /*
53 * Threads should increment retain_data if they are about to call
54 * patch_delta() using this struct's data as a base, and decrement this
55 * when they are done. While retain_data is nonzero, this struct's data
56 * will not be freed even if the delta base cache limit is exceeded.
57 */
58 int retain_data;
59 /*
60 * The number of direct children that have not been fully processed
61 * (entered work_head, entered done_head, left done_head). When this
62 * number reaches zero, this struct base_data can be freed.
63 */
64 int children_remaining;
65
66 /* Not initialized by make_base(). */
67 struct list_head list;
68 void *data;
69 unsigned long size;
70 };
71
72 /*
73 * Stack of struct base_data that have unprocessed children.
74 * threaded_second_pass() uses this as a source of work (the other being the
75 * objects array).
76 *
77 * Guarded by work_mutex.
78 */
79 static LIST_HEAD(work_head);
80
81 /*
82 * Stack of struct base_data that have children, all of whom have been
83 * processed or are being processed, and at least one child is being processed.
84 * These struct base_data must be kept around until the last child is
85 * processed.
86 *
87 * Guarded by work_mutex.
88 */
89 static LIST_HEAD(done_head);
90
91 /*
92 * All threads share one delta base cache.
93 *
94 * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only
95 * in a thread.
96 */
97 static size_t base_cache_used;
98 static size_t base_cache_limit;
99
100 struct thread_local {
101 pthread_t thread;
102 int pack_fd;
103 };
104
105 /* Remember to update object flag allocation in object.h */
106 #define FLAG_LINK (1u<<20)
107 #define FLAG_CHECKED (1u<<21)
108
109 struct ofs_delta_entry {
110 off_t offset;
111 int obj_no;
112 };
113
114 struct ref_delta_entry {
115 struct object_id oid;
116 int obj_no;
117 };
118
119 static struct object_entry *objects;
120 static struct object_stat *obj_stat;
121 static struct ofs_delta_entry *ofs_deltas;
122 static struct ref_delta_entry *ref_deltas;
123 static struct thread_local nothread_data;
124 static int nr_objects;
125 static int nr_ofs_deltas;
126 static int nr_ref_deltas;
127 static int ref_deltas_alloc;
128 static int nr_resolved_deltas;
129 static int nr_threads;
130
131 static int from_stdin;
132 static int strict;
133 static int do_fsck_object;
134 static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES;
135 static int verbose;
136 static const char *progress_title;
137 static int show_resolving_progress;
138 static int show_stat;
139 static int check_self_contained_and_connected;
140
141 static struct progress *progress;
142
143 /* We always read in 4kB chunks. */
144 static unsigned char input_buffer[4096];
145 static unsigned int input_offset, input_len;
146 static off_t consumed_bytes;
147 static off_t max_input_size;
148 static unsigned deepest_delta;
149 static git_hash_ctx input_ctx;
150 static uint32_t input_crc32;
151 static int input_fd, output_fd;
152 static const char *curr_pack;
153
154 static struct thread_local *thread_data;
155 static int nr_dispatched;
156 static int threads_active;
157
158 static pthread_mutex_t read_mutex;
159 #define read_lock() lock_mutex(&read_mutex)
160 #define read_unlock() unlock_mutex(&read_mutex)
161
162 static pthread_mutex_t counter_mutex;
163 #define counter_lock() lock_mutex(&counter_mutex)
164 #define counter_unlock() unlock_mutex(&counter_mutex)
165
166 static pthread_mutex_t work_mutex;
167 #define work_lock() lock_mutex(&work_mutex)
168 #define work_unlock() unlock_mutex(&work_mutex)
169
170 static pthread_mutex_t deepest_delta_mutex;
171 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
172 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
173
174 static pthread_key_t key;
175
176 static inline void lock_mutex(pthread_mutex_t *mutex)
177 {
178 if (threads_active)
179 pthread_mutex_lock(mutex);
180 }
181
182 static inline void unlock_mutex(pthread_mutex_t *mutex)
183 {
184 if (threads_active)
185 pthread_mutex_unlock(mutex);
186 }
187
188 /*
189 * Mutex and conditional variable can't be statically-initialized on Windows.
190 */
191 static void init_thread(void)
192 {
193 int i;
194 init_recursive_mutex(&read_mutex);
195 pthread_mutex_init(&counter_mutex, NULL);
196 pthread_mutex_init(&work_mutex, NULL);
197 if (show_stat)
198 pthread_mutex_init(&deepest_delta_mutex, NULL);
199 pthread_key_create(&key, NULL);
200 CALLOC_ARRAY(thread_data, nr_threads);
201 for (i = 0; i < nr_threads; i++) {
202 thread_data[i].pack_fd = xopen(curr_pack, O_RDONLY);
203 }
204
205 threads_active = 1;
206 }
207
208 static void cleanup_thread(void)
209 {
210 int i;
211 if (!threads_active)
212 return;
213 threads_active = 0;
214 pthread_mutex_destroy(&read_mutex);
215 pthread_mutex_destroy(&counter_mutex);
216 pthread_mutex_destroy(&work_mutex);
217 if (show_stat)
218 pthread_mutex_destroy(&deepest_delta_mutex);
219 for (i = 0; i < nr_threads; i++)
220 close(thread_data[i].pack_fd);
221 pthread_key_delete(key);
222 free(thread_data);
223 }
224
225 static int mark_link(struct object *obj, enum object_type type,
226 void *data, struct fsck_options *options)
227 {
228 if (!obj)
229 return -1;
230
231 if (type != OBJ_ANY && obj->type != type)
232 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
233
234 obj->flags |= FLAG_LINK;
235 return 0;
236 }
237
238 /* The content of each linked object must have been checked
239 or it must be already present in the object database */
240 static unsigned check_object(struct object *obj)
241 {
242 if (!obj)
243 return 0;
244
245 if (!(obj->flags & FLAG_LINK))
246 return 0;
247
248 if (!(obj->flags & FLAG_CHECKED)) {
249 unsigned long size;
250 int type = oid_object_info(the_repository, &obj->oid, &size);
251 if (type <= 0)
252 die(_("did not receive expected object %s"),
253 oid_to_hex(&obj->oid));
254 if (type != obj->type)
255 die(_("object %s: expected type %s, found %s"),
256 oid_to_hex(&obj->oid),
257 type_name(obj->type), type_name(type));
258 obj->flags |= FLAG_CHECKED;
259 return 1;
260 }
261
262 return 0;
263 }
264
265 static unsigned check_objects(void)
266 {
267 unsigned i, max, foreign_nr = 0;
268
269 max = get_max_object_index();
270
271 if (verbose)
272 progress = start_delayed_progress(_("Checking objects"), max);
273
274 for (i = 0; i < max; i++) {
275 foreign_nr += check_object(get_indexed_object(i));
276 display_progress(progress, i + 1);
277 }
278
279 stop_progress(&progress);
280 return foreign_nr;
281 }
282
283
284 /* Discard current buffer used content. */
285 static void flush(void)
286 {
287 if (input_offset) {
288 if (output_fd >= 0)
289 write_or_die(output_fd, input_buffer, input_offset);
290 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
291 memmove(input_buffer, input_buffer + input_offset, input_len);
292 input_offset = 0;
293 }
294 }
295
296 /*
297 * Make sure at least "min" bytes are available in the buffer, and
298 * return the pointer to the buffer.
299 */
300 static void *fill(int min)
301 {
302 if (min <= input_len)
303 return input_buffer + input_offset;
304 if (min > sizeof(input_buffer))
305 die(Q_("cannot fill %d byte",
306 "cannot fill %d bytes",
307 min),
308 min);
309 flush();
310 do {
311 ssize_t ret = xread(input_fd, input_buffer + input_len,
312 sizeof(input_buffer) - input_len);
313 if (ret <= 0) {
314 if (!ret)
315 die(_("early EOF"));
316 die_errno(_("read error on input"));
317 }
318 input_len += ret;
319 if (from_stdin)
320 display_throughput(progress, consumed_bytes + input_len);
321 } while (input_len < min);
322 return input_buffer;
323 }
324
325 static void use(int bytes)
326 {
327 if (bytes > input_len)
328 die(_("used more bytes than were available"));
329 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
330 input_len -= bytes;
331 input_offset += bytes;
332
333 /* make sure off_t is sufficiently large not to wrap */
334 if (signed_add_overflows(consumed_bytes, bytes))
335 die(_("pack too large for current definition of off_t"));
336 consumed_bytes += bytes;
337 if (max_input_size && consumed_bytes > max_input_size) {
338 struct strbuf size_limit = STRBUF_INIT;
339 strbuf_humanise_bytes(&size_limit, max_input_size);
340 die(_("pack exceeds maximum allowed size (%s)"),
341 size_limit.buf);
342 }
343 }
344
345 static const char *open_pack_file(const char *pack_name)
346 {
347 if (from_stdin) {
348 input_fd = 0;
349 if (!pack_name) {
350 struct strbuf tmp_file = STRBUF_INIT;
351 output_fd = odb_mkstemp(&tmp_file,
352 "pack/tmp_pack_XXXXXX");
353 pack_name = strbuf_detach(&tmp_file, NULL);
354 } else {
355 output_fd = xopen(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
356 }
357 nothread_data.pack_fd = output_fd;
358 } else {
359 input_fd = xopen(pack_name, O_RDONLY);
360 output_fd = -1;
361 nothread_data.pack_fd = input_fd;
362 }
363 the_hash_algo->init_fn(&input_ctx);
364 return pack_name;
365 }
366
367 static void parse_pack_header(void)
368 {
369 struct pack_header *hdr = fill(sizeof(struct pack_header));
370
371 /* Header consistency check */
372 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
373 die(_("pack signature mismatch"));
374 if (!pack_version_ok(hdr->hdr_version))
375 die(_("pack version %"PRIu32" unsupported"),
376 ntohl(hdr->hdr_version));
377
378 nr_objects = ntohl(hdr->hdr_entries);
379 use(sizeof(struct pack_header));
380 }
381
382 __attribute__((format (printf, 2, 3)))
383 static NORETURN void bad_object(off_t offset, const char *format, ...)
384 {
385 va_list params;
386 char buf[1024];
387
388 va_start(params, format);
389 vsnprintf(buf, sizeof(buf), format, params);
390 va_end(params);
391 die(_("pack has bad object at offset %"PRIuMAX": %s"),
392 (uintmax_t)offset, buf);
393 }
394
395 static inline struct thread_local *get_thread_data(void)
396 {
397 if (HAVE_THREADS) {
398 if (threads_active)
399 return pthread_getspecific(key);
400 assert(!threads_active &&
401 "This should only be reached when all threads are gone");
402 }
403 return &nothread_data;
404 }
405
406 static void set_thread_data(struct thread_local *data)
407 {
408 if (threads_active)
409 pthread_setspecific(key, data);
410 }
411
412 static void free_base_data(struct base_data *c)
413 {
414 if (c->data) {
415 FREE_AND_NULL(c->data);
416 base_cache_used -= c->size;
417 }
418 }
419
420 static void prune_base_data(struct base_data *retain)
421 {
422 struct list_head *pos;
423
424 if (base_cache_used <= base_cache_limit)
425 return;
426
427 list_for_each_prev(pos, &done_head) {
428 struct base_data *b = list_entry(pos, struct base_data, list);
429 if (b->retain_data || b == retain)
430 continue;
431 if (b->data) {
432 free_base_data(b);
433 if (base_cache_used <= base_cache_limit)
434 return;
435 }
436 }
437
438 list_for_each_prev(pos, &work_head) {
439 struct base_data *b = list_entry(pos, struct base_data, list);
440 if (b->retain_data || b == retain)
441 continue;
442 if (b->data) {
443 free_base_data(b);
444 if (base_cache_used <= base_cache_limit)
445 return;
446 }
447 }
448 }
449
450 static int is_delta_type(enum object_type type)
451 {
452 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
453 }
454
455 static void *unpack_entry_data(off_t offset, unsigned long size,
456 enum object_type type, struct object_id *oid)
457 {
458 static char fixed_buf[8192];
459 int status;
460 git_zstream stream;
461 void *buf;
462 git_hash_ctx c;
463 char hdr[32];
464 int hdrlen;
465
466 if (!is_delta_type(type)) {
467 hdrlen = format_object_header(hdr, sizeof(hdr), type, size);
468 the_hash_algo->init_fn(&c);
469 the_hash_algo->update_fn(&c, hdr, hdrlen);
470 } else
471 oid = NULL;
472 if (type == OBJ_BLOB && size > big_file_threshold)
473 buf = fixed_buf;
474 else
475 buf = xmallocz(size);
476
477 memset(&stream, 0, sizeof(stream));
478 git_inflate_init(&stream);
479 stream.next_out = buf;
480 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
481
482 do {
483 unsigned char *last_out = stream.next_out;
484 stream.next_in = fill(1);
485 stream.avail_in = input_len;
486 status = git_inflate(&stream, 0);
487 use(input_len - stream.avail_in);
488 if (oid)
489 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
490 if (buf == fixed_buf) {
491 stream.next_out = buf;
492 stream.avail_out = sizeof(fixed_buf);
493 }
494 } while (status == Z_OK);
495 if (stream.total_out != size || status != Z_STREAM_END)
496 bad_object(offset, _("inflate returned %d"), status);
497 git_inflate_end(&stream);
498 if (oid)
499 the_hash_algo->final_oid_fn(oid, &c);
500 return buf == fixed_buf ? NULL : buf;
501 }
502
503 static void *unpack_raw_entry(struct object_entry *obj,
504 off_t *ofs_offset,
505 struct object_id *ref_oid,
506 struct object_id *oid)
507 {
508 unsigned char *p;
509 unsigned long size, c;
510 off_t base_offset;
511 unsigned shift;
512 void *data;
513
514 obj->idx.offset = consumed_bytes;
515 input_crc32 = crc32(0, NULL, 0);
516
517 p = fill(1);
518 c = *p;
519 use(1);
520 obj->type = (c >> 4) & 7;
521 size = (c & 15);
522 shift = 4;
523 while (c & 0x80) {
524 p = fill(1);
525 c = *p;
526 use(1);
527 size += (c & 0x7f) << shift;
528 shift += 7;
529 }
530 obj->size = size;
531
532 switch (obj->type) {
533 case OBJ_REF_DELTA:
534 oidread(ref_oid, fill(the_hash_algo->rawsz));
535 use(the_hash_algo->rawsz);
536 break;
537 case OBJ_OFS_DELTA:
538 p = fill(1);
539 c = *p;
540 use(1);
541 base_offset = c & 127;
542 while (c & 128) {
543 base_offset += 1;
544 if (!base_offset || MSB(base_offset, 7))
545 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
546 p = fill(1);
547 c = *p;
548 use(1);
549 base_offset = (base_offset << 7) + (c & 127);
550 }
551 *ofs_offset = obj->idx.offset - base_offset;
552 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
553 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
554 break;
555 case OBJ_COMMIT:
556 case OBJ_TREE:
557 case OBJ_BLOB:
558 case OBJ_TAG:
559 break;
560 default:
561 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
562 }
563 obj->hdr_size = consumed_bytes - obj->idx.offset;
564
565 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
566 obj->idx.crc32 = input_crc32;
567 return data;
568 }
569
570 static void *unpack_data(struct object_entry *obj,
571 int (*consume)(const unsigned char *, unsigned long, void *),
572 void *cb_data)
573 {
574 off_t from = obj[0].idx.offset + obj[0].hdr_size;
575 off_t len = obj[1].idx.offset - from;
576 unsigned char *data, *inbuf;
577 git_zstream stream;
578 int status;
579
580 data = xmallocz(consume ? 64*1024 : obj->size);
581 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
582
583 memset(&stream, 0, sizeof(stream));
584 git_inflate_init(&stream);
585 stream.next_out = data;
586 stream.avail_out = consume ? 64*1024 : obj->size;
587
588 do {
589 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
590 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
591 if (n < 0)
592 die_errno(_("cannot pread pack file"));
593 if (!n)
594 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
595 "premature end of pack file, %"PRIuMAX" bytes missing",
596 len),
597 (uintmax_t)len);
598 from += n;
599 len -= n;
600 stream.next_in = inbuf;
601 stream.avail_in = n;
602 if (!consume)
603 status = git_inflate(&stream, 0);
604 else {
605 do {
606 status = git_inflate(&stream, 0);
607 if (consume(data, stream.next_out - data, cb_data)) {
608 free(inbuf);
609 free(data);
610 return NULL;
611 }
612 stream.next_out = data;
613 stream.avail_out = 64*1024;
614 } while (status == Z_OK && stream.avail_in);
615 }
616 } while (len && status == Z_OK && !stream.avail_in);
617
618 /* This has been inflated OK when first encountered, so... */
619 if (status != Z_STREAM_END || stream.total_out != obj->size)
620 die(_("serious inflate inconsistency"));
621
622 git_inflate_end(&stream);
623 free(inbuf);
624 if (consume) {
625 FREE_AND_NULL(data);
626 }
627 return data;
628 }
629
630 static void *get_data_from_pack(struct object_entry *obj)
631 {
632 return unpack_data(obj, NULL, NULL);
633 }
634
635 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
636 enum object_type type1,
637 enum object_type type2)
638 {
639 int cmp = type1 - type2;
640 if (cmp)
641 return cmp;
642 return offset1 < offset2 ? -1 :
643 offset1 > offset2 ? 1 :
644 0;
645 }
646
647 static int find_ofs_delta(const off_t offset)
648 {
649 int first = 0, last = nr_ofs_deltas;
650
651 while (first < last) {
652 int next = first + (last - first) / 2;
653 struct ofs_delta_entry *delta = &ofs_deltas[next];
654 int cmp;
655
656 cmp = compare_ofs_delta_bases(offset, delta->offset,
657 OBJ_OFS_DELTA,
658 objects[delta->obj_no].type);
659 if (!cmp)
660 return next;
661 if (cmp < 0) {
662 last = next;
663 continue;
664 }
665 first = next+1;
666 }
667 return -first-1;
668 }
669
670 static void find_ofs_delta_children(off_t offset,
671 int *first_index, int *last_index)
672 {
673 int first = find_ofs_delta(offset);
674 int last = first;
675 int end = nr_ofs_deltas - 1;
676
677 if (first < 0) {
678 *first_index = 0;
679 *last_index = -1;
680 return;
681 }
682 while (first > 0 && ofs_deltas[first - 1].offset == offset)
683 --first;
684 while (last < end && ofs_deltas[last + 1].offset == offset)
685 ++last;
686 *first_index = first;
687 *last_index = last;
688 }
689
690 static int compare_ref_delta_bases(const struct object_id *oid1,
691 const struct object_id *oid2,
692 enum object_type type1,
693 enum object_type type2)
694 {
695 int cmp = type1 - type2;
696 if (cmp)
697 return cmp;
698 return oidcmp(oid1, oid2);
699 }
700
701 static int find_ref_delta(const struct object_id *oid)
702 {
703 int first = 0, last = nr_ref_deltas;
704
705 while (first < last) {
706 int next = first + (last - first) / 2;
707 struct ref_delta_entry *delta = &ref_deltas[next];
708 int cmp;
709
710 cmp = compare_ref_delta_bases(oid, &delta->oid,
711 OBJ_REF_DELTA,
712 objects[delta->obj_no].type);
713 if (!cmp)
714 return next;
715 if (cmp < 0) {
716 last = next;
717 continue;
718 }
719 first = next+1;
720 }
721 return -first-1;
722 }
723
724 static void find_ref_delta_children(const struct object_id *oid,
725 int *first_index, int *last_index)
726 {
727 int first = find_ref_delta(oid);
728 int last = first;
729 int end = nr_ref_deltas - 1;
730
731 if (first < 0) {
732 *first_index = 0;
733 *last_index = -1;
734 return;
735 }
736 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
737 --first;
738 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
739 ++last;
740 *first_index = first;
741 *last_index = last;
742 }
743
744 struct compare_data {
745 struct object_entry *entry;
746 struct git_istream *st;
747 unsigned char *buf;
748 unsigned long buf_size;
749 };
750
751 static int compare_objects(const unsigned char *buf, unsigned long size,
752 void *cb_data)
753 {
754 struct compare_data *data = cb_data;
755
756 if (data->buf_size < size) {
757 free(data->buf);
758 data->buf = xmalloc(size);
759 data->buf_size = size;
760 }
761
762 while (size) {
763 ssize_t len = read_istream(data->st, data->buf, size);
764 if (len == 0)
765 die(_("SHA1 COLLISION FOUND WITH %s !"),
766 oid_to_hex(&data->entry->idx.oid));
767 if (len < 0)
768 die(_("unable to read %s"),
769 oid_to_hex(&data->entry->idx.oid));
770 if (memcmp(buf, data->buf, len))
771 die(_("SHA1 COLLISION FOUND WITH %s !"),
772 oid_to_hex(&data->entry->idx.oid));
773 size -= len;
774 buf += len;
775 }
776 return 0;
777 }
778
779 static int check_collison(struct object_entry *entry)
780 {
781 struct compare_data data;
782 enum object_type type;
783 unsigned long size;
784
785 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
786 return -1;
787
788 memset(&data, 0, sizeof(data));
789 data.entry = entry;
790 data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
791 NULL);
792 if (!data.st)
793 return -1;
794 if (size != entry->size || type != entry->type)
795 die(_("SHA1 COLLISION FOUND WITH %s !"),
796 oid_to_hex(&entry->idx.oid));
797 unpack_data(entry, compare_objects, &data);
798 close_istream(data.st);
799 free(data.buf);
800 return 0;
801 }
802
803 static void sha1_object(const void *data, struct object_entry *obj_entry,
804 unsigned long size, enum object_type type,
805 const struct object_id *oid)
806 {
807 void *new_data = NULL;
808 int collision_test_needed = 0;
809
810 assert(data || obj_entry);
811
812 if (startup_info->have_repository) {
813 read_lock();
814 collision_test_needed =
815 repo_has_object_file_with_flags(the_repository, oid,
816 OBJECT_INFO_QUICK);
817 read_unlock();
818 }
819
820 if (collision_test_needed && !data) {
821 read_lock();
822 if (!check_collison(obj_entry))
823 collision_test_needed = 0;
824 read_unlock();
825 }
826 if (collision_test_needed) {
827 void *has_data;
828 enum object_type has_type;
829 unsigned long has_size;
830 read_lock();
831 has_type = oid_object_info(the_repository, oid, &has_size);
832 if (has_type < 0)
833 die(_("cannot read existing object info %s"), oid_to_hex(oid));
834 if (has_type != type || has_size != size)
835 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
836 has_data = repo_read_object_file(the_repository, oid,
837 &has_type, &has_size);
838 read_unlock();
839 if (!data)
840 data = new_data = get_data_from_pack(obj_entry);
841 if (!has_data)
842 die(_("cannot read existing object %s"), oid_to_hex(oid));
843 if (size != has_size || type != has_type ||
844 memcmp(data, has_data, size) != 0)
845 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
846 free(has_data);
847 }
848
849 if (strict || do_fsck_object) {
850 read_lock();
851 if (type == OBJ_BLOB) {
852 struct blob *blob = lookup_blob(the_repository, oid);
853 if (blob)
854 blob->object.flags |= FLAG_CHECKED;
855 else
856 die(_("invalid blob object %s"), oid_to_hex(oid));
857 if (do_fsck_object &&
858 fsck_object(&blob->object, (void *)data, size, &fsck_options))
859 die(_("fsck error in packed object"));
860 } else {
861 struct object *obj;
862 int eaten;
863 void *buf = (void *) data;
864
865 assert(data && "data can only be NULL for large _blobs_");
866
867 /*
868 * we do not need to free the memory here, as the
869 * buf is deleted by the caller.
870 */
871 obj = parse_object_buffer(the_repository, oid, type,
872 size, buf,
873 &eaten);
874 if (!obj)
875 die(_("invalid %s"), type_name(type));
876 if (do_fsck_object &&
877 fsck_object(obj, buf, size, &fsck_options))
878 die(_("fsck error in packed object"));
879 if (strict && fsck_walk(obj, NULL, &fsck_options))
880 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
881
882 if (obj->type == OBJ_TREE) {
883 struct tree *item = (struct tree *) obj;
884 item->buffer = NULL;
885 obj->parsed = 0;
886 }
887 if (obj->type == OBJ_COMMIT) {
888 struct commit *commit = (struct commit *) obj;
889 if (detach_commit_buffer(commit, NULL) != data)
890 BUG("parse_object_buffer transmogrified our buffer");
891 }
892 obj->flags |= FLAG_CHECKED;
893 }
894 read_unlock();
895 }
896
897 free(new_data);
898 }
899
900 /*
901 * Ensure that this node has been reconstructed and return its contents.
902 *
903 * In the typical and best case, this node would already be reconstructed
904 * (through the invocation to resolve_delta() in threaded_second_pass()) and it
905 * would not be pruned. However, if pruning of this node was necessary due to
906 * reaching delta_base_cache_limit, this function will find the closest
907 * ancestor with reconstructed data that has not been pruned (or if there is
908 * none, the ultimate base object), and reconstruct each node in the delta
909 * chain in order to generate the reconstructed data for this node.
910 */
911 static void *get_base_data(struct base_data *c)
912 {
913 if (!c->data) {
914 struct object_entry *obj = c->obj;
915 struct base_data **delta = NULL;
916 int delta_nr = 0, delta_alloc = 0;
917
918 while (is_delta_type(c->obj->type) && !c->data) {
919 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
920 delta[delta_nr++] = c;
921 c = c->base;
922 }
923 if (!delta_nr) {
924 c->data = get_data_from_pack(obj);
925 c->size = obj->size;
926 base_cache_used += c->size;
927 prune_base_data(c);
928 }
929 for (; delta_nr > 0; delta_nr--) {
930 void *base, *raw;
931 c = delta[delta_nr - 1];
932 obj = c->obj;
933 base = get_base_data(c->base);
934 raw = get_data_from_pack(obj);
935 c->data = patch_delta(
936 base, c->base->size,
937 raw, obj->size,
938 &c->size);
939 free(raw);
940 if (!c->data)
941 bad_object(obj->idx.offset, _("failed to apply delta"));
942 base_cache_used += c->size;
943 prune_base_data(c);
944 }
945 free(delta);
946 }
947 return c->data;
948 }
949
950 static struct base_data *make_base(struct object_entry *obj,
951 struct base_data *parent)
952 {
953 struct base_data *base = xcalloc(1, sizeof(struct base_data));
954 base->base = parent;
955 base->obj = obj;
956 find_ref_delta_children(&obj->idx.oid,
957 &base->ref_first, &base->ref_last);
958 find_ofs_delta_children(obj->idx.offset,
959 &base->ofs_first, &base->ofs_last);
960 base->children_remaining = base->ref_last - base->ref_first +
961 base->ofs_last - base->ofs_first + 2;
962 return base;
963 }
964
965 static struct base_data *resolve_delta(struct object_entry *delta_obj,
966 struct base_data *base)
967 {
968 void *delta_data, *result_data;
969 struct base_data *result;
970 unsigned long result_size;
971
972 if (show_stat) {
973 int i = delta_obj - objects;
974 int j = base->obj - objects;
975 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
976 deepest_delta_lock();
977 if (deepest_delta < obj_stat[i].delta_depth)
978 deepest_delta = obj_stat[i].delta_depth;
979 deepest_delta_unlock();
980 obj_stat[i].base_object_no = j;
981 }
982 delta_data = get_data_from_pack(delta_obj);
983 assert(base->data);
984 result_data = patch_delta(base->data, base->size,
985 delta_data, delta_obj->size, &result_size);
986 free(delta_data);
987 if (!result_data)
988 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
989 hash_object_file(the_hash_algo, result_data, result_size,
990 delta_obj->real_type, &delta_obj->idx.oid);
991 sha1_object(result_data, NULL, result_size, delta_obj->real_type,
992 &delta_obj->idx.oid);
993
994 result = make_base(delta_obj, base);
995 result->data = result_data;
996 result->size = result_size;
997
998 counter_lock();
999 nr_resolved_deltas++;
1000 counter_unlock();
1001
1002 return result;
1003 }
1004
1005 static int compare_ofs_delta_entry(const void *a, const void *b)
1006 {
1007 const struct ofs_delta_entry *delta_a = a;
1008 const struct ofs_delta_entry *delta_b = b;
1009
1010 return delta_a->offset < delta_b->offset ? -1 :
1011 delta_a->offset > delta_b->offset ? 1 :
1012 0;
1013 }
1014
1015 static int compare_ref_delta_entry(const void *a, const void *b)
1016 {
1017 const struct ref_delta_entry *delta_a = a;
1018 const struct ref_delta_entry *delta_b = b;
1019
1020 return oidcmp(&delta_a->oid, &delta_b->oid);
1021 }
1022
1023 static void *threaded_second_pass(void *data)
1024 {
1025 if (data)
1026 set_thread_data(data);
1027 for (;;) {
1028 struct base_data *parent = NULL;
1029 struct object_entry *child_obj;
1030 struct base_data *child;
1031
1032 counter_lock();
1033 display_progress(progress, nr_resolved_deltas);
1034 counter_unlock();
1035
1036 work_lock();
1037 if (list_empty(&work_head)) {
1038 /*
1039 * Take an object from the object array.
1040 */
1041 while (nr_dispatched < nr_objects &&
1042 is_delta_type(objects[nr_dispatched].type))
1043 nr_dispatched++;
1044 if (nr_dispatched >= nr_objects) {
1045 work_unlock();
1046 break;
1047 }
1048 child_obj = &objects[nr_dispatched++];
1049 } else {
1050 /*
1051 * Peek at the top of the stack, and take a child from
1052 * it.
1053 */
1054 parent = list_first_entry(&work_head, struct base_data,
1055 list);
1056
1057 if (parent->ref_first <= parent->ref_last) {
1058 int offset = ref_deltas[parent->ref_first++].obj_no;
1059 child_obj = objects + offset;
1060 if (child_obj->real_type != OBJ_REF_DELTA)
1061 die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)",
1062 (uintmax_t) child_obj->idx.offset,
1063 oid_to_hex(&parent->obj->idx.oid));
1064 child_obj->real_type = parent->obj->real_type;
1065 } else {
1066 child_obj = objects +
1067 ofs_deltas[parent->ofs_first++].obj_no;
1068 assert(child_obj->real_type == OBJ_OFS_DELTA);
1069 child_obj->real_type = parent->obj->real_type;
1070 }
1071
1072 if (parent->ref_first > parent->ref_last &&
1073 parent->ofs_first > parent->ofs_last) {
1074 /*
1075 * This parent has run out of children, so move
1076 * it to done_head.
1077 */
1078 list_del(&parent->list);
1079 list_add(&parent->list, &done_head);
1080 }
1081
1082 /*
1083 * Ensure that the parent has data, since we will need
1084 * it later.
1085 *
1086 * NEEDSWORK: If parent data needs to be reloaded, this
1087 * prolongs the time that the current thread spends in
1088 * the mutex. A mitigating factor is that parent data
1089 * needs to be reloaded only if the delta base cache
1090 * limit is exceeded, so in the typical case, this does
1091 * not happen.
1092 */
1093 get_base_data(parent);
1094 parent->retain_data++;
1095 }
1096 work_unlock();
1097
1098 if (parent) {
1099 child = resolve_delta(child_obj, parent);
1100 if (!child->children_remaining)
1101 FREE_AND_NULL(child->data);
1102 } else {
1103 child = make_base(child_obj, NULL);
1104 if (child->children_remaining) {
1105 /*
1106 * Since this child has its own delta children,
1107 * we will need this data in the future.
1108 * Inflate now so that future iterations will
1109 * have access to this object's data while
1110 * outside the work mutex.
1111 */
1112 child->data = get_data_from_pack(child_obj);
1113 child->size = child_obj->size;
1114 }
1115 }
1116
1117 work_lock();
1118 if (parent)
1119 parent->retain_data--;
1120 if (child->data) {
1121 /*
1122 * This child has its own children, so add it to
1123 * work_head.
1124 */
1125 list_add(&child->list, &work_head);
1126 base_cache_used += child->size;
1127 prune_base_data(NULL);
1128 free_base_data(child);
1129 } else {
1130 /*
1131 * This child does not have its own children. It may be
1132 * the last descendant of its ancestors; free those
1133 * that we can.
1134 */
1135 struct base_data *p = parent;
1136
1137 while (p) {
1138 struct base_data *next_p;
1139
1140 p->children_remaining--;
1141 if (p->children_remaining)
1142 break;
1143
1144 next_p = p->base;
1145 free_base_data(p);
1146 list_del(&p->list);
1147 free(p);
1148
1149 p = next_p;
1150 }
1151 FREE_AND_NULL(child);
1152 }
1153 work_unlock();
1154 }
1155 return NULL;
1156 }
1157
1158 /*
1159 * First pass:
1160 * - find locations of all objects;
1161 * - calculate SHA1 of all non-delta objects;
1162 * - remember base (SHA1 or offset) for all deltas.
1163 */
1164 static void parse_pack_objects(unsigned char *hash)
1165 {
1166 int i, nr_delays = 0;
1167 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1168 struct object_id ref_delta_oid;
1169 struct stat st;
1170
1171 if (verbose)
1172 progress = start_progress(
1173 progress_title ? progress_title :
1174 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1175 nr_objects);
1176 for (i = 0; i < nr_objects; i++) {
1177 struct object_entry *obj = &objects[i];
1178 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1179 &ref_delta_oid,
1180 &obj->idx.oid);
1181 obj->real_type = obj->type;
1182 if (obj->type == OBJ_OFS_DELTA) {
1183 nr_ofs_deltas++;
1184 ofs_delta->obj_no = i;
1185 ofs_delta++;
1186 } else if (obj->type == OBJ_REF_DELTA) {
1187 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1188 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1189 ref_deltas[nr_ref_deltas].obj_no = i;
1190 nr_ref_deltas++;
1191 } else if (!data) {
1192 /* large blobs, check later */
1193 obj->real_type = OBJ_BAD;
1194 nr_delays++;
1195 } else
1196 sha1_object(data, NULL, obj->size, obj->type,
1197 &obj->idx.oid);
1198 free(data);
1199 display_progress(progress, i+1);
1200 }
1201 objects[i].idx.offset = consumed_bytes;
1202 stop_progress(&progress);
1203
1204 /* Check pack integrity */
1205 flush();
1206 the_hash_algo->final_fn(hash, &input_ctx);
1207 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1208 die(_("pack is corrupted (SHA1 mismatch)"));
1209 use(the_hash_algo->rawsz);
1210
1211 /* If input_fd is a file, we should have reached its end now. */
1212 if (fstat(input_fd, &st))
1213 die_errno(_("cannot fstat packfile"));
1214 if (S_ISREG(st.st_mode) &&
1215 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1216 die(_("pack has junk at the end"));
1217
1218 for (i = 0; i < nr_objects; i++) {
1219 struct object_entry *obj = &objects[i];
1220 if (obj->real_type != OBJ_BAD)
1221 continue;
1222 obj->real_type = obj->type;
1223 sha1_object(NULL, obj, obj->size, obj->type,
1224 &obj->idx.oid);
1225 nr_delays--;
1226 }
1227 if (nr_delays)
1228 die(_("confusion beyond insanity in parse_pack_objects()"));
1229 }
1230
1231 /*
1232 * Second pass:
1233 * - for all non-delta objects, look if it is used as a base for
1234 * deltas;
1235 * - if used as a base, uncompress the object and apply all deltas,
1236 * recursively checking if the resulting object is used as a base
1237 * for some more deltas.
1238 */
1239 static void resolve_deltas(void)
1240 {
1241 int i;
1242
1243 if (!nr_ofs_deltas && !nr_ref_deltas)
1244 return;
1245
1246 /* Sort deltas by base SHA1/offset for fast searching */
1247 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1248 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1249
1250 if (verbose || show_resolving_progress)
1251 progress = start_progress(_("Resolving deltas"),
1252 nr_ref_deltas + nr_ofs_deltas);
1253
1254 nr_dispatched = 0;
1255 base_cache_limit = delta_base_cache_limit * nr_threads;
1256 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1257 init_thread();
1258 for (i = 0; i < nr_threads; i++) {
1259 int ret = pthread_create(&thread_data[i].thread, NULL,
1260 threaded_second_pass, thread_data + i);
1261 if (ret)
1262 die(_("unable to create thread: %s"),
1263 strerror(ret));
1264 }
1265 for (i = 0; i < nr_threads; i++)
1266 pthread_join(thread_data[i].thread, NULL);
1267 cleanup_thread();
1268 return;
1269 }
1270 threaded_second_pass(&nothread_data);
1271 }
1272
1273 /*
1274 * Third pass:
1275 * - append objects to convert thin pack to full pack if required
1276 * - write the final pack hash
1277 */
1278 static void fix_unresolved_deltas(struct hashfile *f);
1279 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1280 {
1281 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1282 stop_progress(&progress);
1283 /* Flush remaining pack final hash. */
1284 flush();
1285 return;
1286 }
1287
1288 if (fix_thin_pack) {
1289 struct hashfile *f;
1290 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1291 struct strbuf msg = STRBUF_INIT;
1292 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1293 int nr_objects_initial = nr_objects;
1294 if (nr_unresolved <= 0)
1295 die(_("confusion beyond insanity"));
1296 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1297 memset(objects + nr_objects + 1, 0,
1298 nr_unresolved * sizeof(*objects));
1299 f = hashfd(output_fd, curr_pack);
1300 fix_unresolved_deltas(f);
1301 strbuf_addf(&msg, Q_("completed with %d local object",
1302 "completed with %d local objects",
1303 nr_objects - nr_objects_initial),
1304 nr_objects - nr_objects_initial);
1305 stop_progress_msg(&progress, msg.buf);
1306 strbuf_release(&msg);
1307 finalize_hashfile(f, tail_hash, FSYNC_COMPONENT_PACK, 0);
1308 hashcpy(read_hash, pack_hash);
1309 fixup_pack_header_footer(output_fd, pack_hash,
1310 curr_pack, nr_objects,
1311 read_hash, consumed_bytes-the_hash_algo->rawsz);
1312 if (!hasheq(read_hash, tail_hash))
1313 die(_("Unexpected tail checksum for %s "
1314 "(disk corruption?)"), curr_pack);
1315 }
1316 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1317 die(Q_("pack has %d unresolved delta",
1318 "pack has %d unresolved deltas",
1319 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1320 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1321 }
1322
1323 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1324 {
1325 git_zstream stream;
1326 int status;
1327 unsigned char outbuf[4096];
1328
1329 git_deflate_init(&stream, zlib_compression_level);
1330 stream.next_in = in;
1331 stream.avail_in = size;
1332
1333 do {
1334 stream.next_out = outbuf;
1335 stream.avail_out = sizeof(outbuf);
1336 status = git_deflate(&stream, Z_FINISH);
1337 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1338 } while (status == Z_OK);
1339
1340 if (status != Z_STREAM_END)
1341 die(_("unable to deflate appended object (%d)"), status);
1342 size = stream.total_out;
1343 git_deflate_end(&stream);
1344 return size;
1345 }
1346
1347 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1348 const unsigned char *sha1, void *buf,
1349 unsigned long size, enum object_type type)
1350 {
1351 struct object_entry *obj = &objects[nr_objects++];
1352 unsigned char header[10];
1353 unsigned long s = size;
1354 int n = 0;
1355 unsigned char c = (type << 4) | (s & 15);
1356 s >>= 4;
1357 while (s) {
1358 header[n++] = c | 0x80;
1359 c = s & 0x7f;
1360 s >>= 7;
1361 }
1362 header[n++] = c;
1363 crc32_begin(f);
1364 hashwrite(f, header, n);
1365 obj[0].size = size;
1366 obj[0].hdr_size = n;
1367 obj[0].type = type;
1368 obj[0].real_type = type;
1369 obj[1].idx.offset = obj[0].idx.offset + n;
1370 obj[1].idx.offset += write_compressed(f, buf, size);
1371 obj[0].idx.crc32 = crc32_end(f);
1372 hashflush(f);
1373 oidread(&obj->idx.oid, sha1);
1374 return obj;
1375 }
1376
1377 static int delta_pos_compare(const void *_a, const void *_b)
1378 {
1379 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1380 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1381 return a->obj_no - b->obj_no;
1382 }
1383
1384 static void fix_unresolved_deltas(struct hashfile *f)
1385 {
1386 struct ref_delta_entry **sorted_by_pos;
1387 int i;
1388
1389 /*
1390 * Since many unresolved deltas may well be themselves base objects
1391 * for more unresolved deltas, we really want to include the
1392 * smallest number of base objects that would cover as much delta
1393 * as possible by picking the
1394 * trunc deltas first, allowing for other deltas to resolve without
1395 * additional base objects. Since most base objects are to be found
1396 * before deltas depending on them, a good heuristic is to start
1397 * resolving deltas in the same order as their position in the pack.
1398 */
1399 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1400 for (i = 0; i < nr_ref_deltas; i++)
1401 sorted_by_pos[i] = &ref_deltas[i];
1402 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1403
1404 if (repo_has_promisor_remote(the_repository)) {
1405 /*
1406 * Prefetch the delta bases.
1407 */
1408 struct oid_array to_fetch = OID_ARRAY_INIT;
1409 for (i = 0; i < nr_ref_deltas; i++) {
1410 struct ref_delta_entry *d = sorted_by_pos[i];
1411 if (!oid_object_info_extended(the_repository, &d->oid,
1412 NULL,
1413 OBJECT_INFO_FOR_PREFETCH))
1414 continue;
1415 oid_array_append(&to_fetch, &d->oid);
1416 }
1417 promisor_remote_get_direct(the_repository,
1418 to_fetch.oid, to_fetch.nr);
1419 oid_array_clear(&to_fetch);
1420 }
1421
1422 for (i = 0; i < nr_ref_deltas; i++) {
1423 struct ref_delta_entry *d = sorted_by_pos[i];
1424 enum object_type type;
1425 void *data;
1426 unsigned long size;
1427
1428 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1429 continue;
1430 data = repo_read_object_file(the_repository, &d->oid, &type,
1431 &size);
1432 if (!data)
1433 continue;
1434
1435 if (check_object_signature(the_repository, &d->oid, data, size,
1436 type) < 0)
1437 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1438
1439 /*
1440 * Add this as an object to the objects array and call
1441 * threaded_second_pass() (which will pick up the added
1442 * object).
1443 */
1444 append_obj_to_pack(f, d->oid.hash, data, size, type);
1445 free(data);
1446 threaded_second_pass(NULL);
1447
1448 display_progress(progress, nr_resolved_deltas);
1449 }
1450 free(sorted_by_pos);
1451 }
1452
1453 static const char *derive_filename(const char *pack_name, const char *strip,
1454 const char *suffix, struct strbuf *buf)
1455 {
1456 size_t len;
1457 if (!strip_suffix(pack_name, strip, &len) || !len ||
1458 pack_name[len - 1] != '.')
1459 die(_("packfile name '%s' does not end with '.%s'"),
1460 pack_name, strip);
1461 strbuf_add(buf, pack_name, len);
1462 strbuf_addstr(buf, suffix);
1463 return buf->buf;
1464 }
1465
1466 static void write_special_file(const char *suffix, const char *msg,
1467 const char *pack_name, const unsigned char *hash,
1468 const char **report)
1469 {
1470 struct strbuf name_buf = STRBUF_INIT;
1471 const char *filename;
1472 int fd;
1473 int msg_len = strlen(msg);
1474
1475 if (pack_name)
1476 filename = derive_filename(pack_name, "pack", suffix, &name_buf);
1477 else
1478 filename = odb_pack_name(&name_buf, hash, suffix);
1479
1480 fd = odb_pack_keep(filename);
1481 if (fd < 0) {
1482 if (errno != EEXIST)
1483 die_errno(_("cannot write %s file '%s'"),
1484 suffix, filename);
1485 } else {
1486 if (msg_len > 0) {
1487 write_or_die(fd, msg, msg_len);
1488 write_or_die(fd, "\n", 1);
1489 }
1490 if (close(fd) != 0)
1491 die_errno(_("cannot close written %s file '%s'"),
1492 suffix, filename);
1493 if (report)
1494 *report = suffix;
1495 }
1496 strbuf_release(&name_buf);
1497 }
1498
1499 static void rename_tmp_packfile(const char **final_name,
1500 const char *curr_name,
1501 struct strbuf *name, unsigned char *hash,
1502 const char *ext, int make_read_only_if_same)
1503 {
1504 if (*final_name != curr_name) {
1505 if (!*final_name)
1506 *final_name = odb_pack_name(name, hash, ext);
1507 if (finalize_object_file(curr_name, *final_name))
1508 die(_("unable to rename temporary '*.%s' file to '%s'"),
1509 ext, *final_name);
1510 } else if (make_read_only_if_same) {
1511 chmod(*final_name, 0444);
1512 }
1513 }
1514
1515 static void final(const char *final_pack_name, const char *curr_pack_name,
1516 const char *final_index_name, const char *curr_index_name,
1517 const char *final_rev_index_name, const char *curr_rev_index_name,
1518 const char *keep_msg, const char *promisor_msg,
1519 unsigned char *hash)
1520 {
1521 const char *report = "pack";
1522 struct strbuf pack_name = STRBUF_INIT;
1523 struct strbuf index_name = STRBUF_INIT;
1524 struct strbuf rev_index_name = STRBUF_INIT;
1525 int err;
1526
1527 if (!from_stdin) {
1528 close(input_fd);
1529 } else {
1530 fsync_component_or_die(FSYNC_COMPONENT_PACK, output_fd, curr_pack_name);
1531 err = close(output_fd);
1532 if (err)
1533 die_errno(_("error while closing pack file"));
1534 }
1535
1536 if (keep_msg)
1537 write_special_file("keep", keep_msg, final_pack_name, hash,
1538 &report);
1539 if (promisor_msg)
1540 write_special_file("promisor", promisor_msg, final_pack_name,
1541 hash, NULL);
1542
1543 rename_tmp_packfile(&final_pack_name, curr_pack_name, &pack_name,
1544 hash, "pack", from_stdin);
1545 if (curr_rev_index_name)
1546 rename_tmp_packfile(&final_rev_index_name, curr_rev_index_name,
1547 &rev_index_name, hash, "rev", 1);
1548 rename_tmp_packfile(&final_index_name, curr_index_name, &index_name,
1549 hash, "idx", 1);
1550
1551 if (do_fsck_object) {
1552 struct packed_git *p;
1553 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1554 if (p)
1555 install_packed_git(the_repository, p);
1556 }
1557
1558 if (!from_stdin) {
1559 printf("%s\n", hash_to_hex(hash));
1560 } else {
1561 struct strbuf buf = STRBUF_INIT;
1562
1563 strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1564 write_or_die(1, buf.buf, buf.len);
1565 strbuf_release(&buf);
1566
1567 /*
1568 * Let's just mimic git-unpack-objects here and write
1569 * the last part of the input buffer to stdout.
1570 */
1571 while (input_len) {
1572 err = xwrite(1, input_buffer + input_offset, input_len);
1573 if (err <= 0)
1574 break;
1575 input_len -= err;
1576 input_offset += err;
1577 }
1578 }
1579
1580 strbuf_release(&rev_index_name);
1581 strbuf_release(&index_name);
1582 strbuf_release(&pack_name);
1583 }
1584
1585 static int git_index_pack_config(const char *k, const char *v, void *cb)
1586 {
1587 struct pack_idx_option *opts = cb;
1588
1589 if (!strcmp(k, "pack.indexversion")) {
1590 opts->version = git_config_int(k, v);
1591 if (opts->version > 2)
1592 die(_("bad pack.indexVersion=%"PRIu32), opts->version);
1593 return 0;
1594 }
1595 if (!strcmp(k, "pack.threads")) {
1596 nr_threads = git_config_int(k, v);
1597 if (nr_threads < 0)
1598 die(_("invalid number of threads specified (%d)"),
1599 nr_threads);
1600 if (!HAVE_THREADS && nr_threads != 1) {
1601 warning(_("no threads support, ignoring %s"), k);
1602 nr_threads = 1;
1603 }
1604 return 0;
1605 }
1606 if (!strcmp(k, "pack.writereverseindex")) {
1607 if (git_config_bool(k, v))
1608 opts->flags |= WRITE_REV;
1609 else
1610 opts->flags &= ~WRITE_REV;
1611 }
1612 return git_default_config(k, v, cb);
1613 }
1614
1615 static int cmp_uint32(const void *a_, const void *b_)
1616 {
1617 uint32_t a = *((uint32_t *)a_);
1618 uint32_t b = *((uint32_t *)b_);
1619
1620 return (a < b) ? -1 : (a != b);
1621 }
1622
1623 static void read_v2_anomalous_offsets(struct packed_git *p,
1624 struct pack_idx_option *opts)
1625 {
1626 const uint32_t *idx1, *idx2;
1627 uint32_t i;
1628
1629 /* The address of the 4-byte offset table */
1630 idx1 = (((const uint32_t *)((const uint8_t *)p->index_data + p->crc_offset))
1631 + (size_t)p->num_objects /* CRC32 table */
1632 );
1633
1634 /* The address of the 8-byte offset table */
1635 idx2 = idx1 + p->num_objects;
1636
1637 for (i = 0; i < p->num_objects; i++) {
1638 uint32_t off = ntohl(idx1[i]);
1639 if (!(off & 0x80000000))
1640 continue;
1641 off = off & 0x7fffffff;
1642 check_pack_index_ptr(p, &idx2[off * 2]);
1643 if (idx2[off * 2])
1644 continue;
1645 /*
1646 * The real offset is ntohl(idx2[off * 2]) in high 4
1647 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1648 * octets. But idx2[off * 2] is Zero!!!
1649 */
1650 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1651 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1652 }
1653
1654 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1655 }
1656
1657 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1658 {
1659 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1660
1661 if (!p)
1662 die(_("Cannot open existing pack file '%s'"), pack_name);
1663 if (open_pack_index(p))
1664 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1665
1666 /* Read the attributes from the existing idx file */
1667 opts->version = p->index_version;
1668
1669 if (opts->version == 2)
1670 read_v2_anomalous_offsets(p, opts);
1671
1672 /*
1673 * Get rid of the idx file as we do not need it anymore.
1674 * NEEDSWORK: extract this bit from free_pack_by_name() in
1675 * object-file.c, perhaps? It shouldn't matter very much as we
1676 * know we haven't installed this pack (hence we never have
1677 * read anything from it).
1678 */
1679 close_pack_index(p);
1680 free(p);
1681 }
1682
1683 static void show_pack_info(int stat_only)
1684 {
1685 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1686 unsigned long *chain_histogram = NULL;
1687
1688 if (deepest_delta)
1689 CALLOC_ARRAY(chain_histogram, deepest_delta);
1690
1691 for (i = 0; i < nr_objects; i++) {
1692 struct object_entry *obj = &objects[i];
1693
1694 if (is_delta_type(obj->type))
1695 chain_histogram[obj_stat[i].delta_depth - 1]++;
1696 if (stat_only)
1697 continue;
1698 printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1699 oid_to_hex(&obj->idx.oid),
1700 type_name(obj->real_type), (uintmax_t)obj->size,
1701 (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1702 (uintmax_t)obj->idx.offset);
1703 if (is_delta_type(obj->type)) {
1704 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1705 printf(" %u %s", obj_stat[i].delta_depth,
1706 oid_to_hex(&bobj->idx.oid));
1707 }
1708 putchar('\n');
1709 }
1710
1711 if (baseobjects)
1712 printf_ln(Q_("non delta: %d object",
1713 "non delta: %d objects",
1714 baseobjects),
1715 baseobjects);
1716 for (i = 0; i < deepest_delta; i++) {
1717 if (!chain_histogram[i])
1718 continue;
1719 printf_ln(Q_("chain length = %d: %lu object",
1720 "chain length = %d: %lu objects",
1721 chain_histogram[i]),
1722 i + 1,
1723 chain_histogram[i]);
1724 }
1725 free(chain_histogram);
1726 }
1727
1728 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1729 {
1730 int i, fix_thin_pack = 0, verify = 0, stat_only = 0, rev_index;
1731 const char *curr_index;
1732 const char *curr_rev_index = NULL;
1733 const char *index_name = NULL, *pack_name = NULL, *rev_index_name = NULL;
1734 const char *keep_msg = NULL;
1735 const char *promisor_msg = NULL;
1736 struct strbuf index_name_buf = STRBUF_INIT;
1737 struct strbuf rev_index_name_buf = STRBUF_INIT;
1738 struct pack_idx_entry **idx_objects;
1739 struct pack_idx_option opts;
1740 unsigned char pack_hash[GIT_MAX_RAWSZ];
1741 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1742 int report_end_of_input = 0;
1743 int hash_algo = 0;
1744
1745 /*
1746 * index-pack never needs to fetch missing objects except when
1747 * REF_DELTA bases are missing (which are explicitly handled). It only
1748 * accesses the repo to do hash collision checks and to check which
1749 * REF_DELTA bases need to be fetched.
1750 */
1751 fetch_if_missing = 0;
1752
1753 if (argc == 2 && !strcmp(argv[1], "-h"))
1754 usage(index_pack_usage);
1755
1756 disable_replace_refs();
1757 fsck_options.walk = mark_link;
1758
1759 reset_pack_idx_option(&opts);
1760 opts.flags |= WRITE_REV;
1761 git_config(git_index_pack_config, &opts);
1762 if (prefix && chdir(prefix))
1763 die(_("Cannot come back to cwd"));
1764
1765 if (git_env_bool(GIT_TEST_NO_WRITE_REV_INDEX, 0))
1766 rev_index = 0;
1767 else
1768 rev_index = !!(opts.flags & (WRITE_REV_VERIFY | WRITE_REV));
1769
1770 for (i = 1; i < argc; i++) {
1771 const char *arg = argv[i];
1772
1773 if (*arg == '-') {
1774 if (!strcmp(arg, "--stdin")) {
1775 from_stdin = 1;
1776 } else if (!strcmp(arg, "--fix-thin")) {
1777 fix_thin_pack = 1;
1778 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1779 strict = 1;
1780 do_fsck_object = 1;
1781 fsck_set_msg_types(&fsck_options, arg);
1782 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1783 strict = 1;
1784 check_self_contained_and_connected = 1;
1785 } else if (!strcmp(arg, "--fsck-objects")) {
1786 do_fsck_object = 1;
1787 } else if (!strcmp(arg, "--verify")) {
1788 verify = 1;
1789 } else if (!strcmp(arg, "--verify-stat")) {
1790 verify = 1;
1791 show_stat = 1;
1792 } else if (!strcmp(arg, "--verify-stat-only")) {
1793 verify = 1;
1794 show_stat = 1;
1795 stat_only = 1;
1796 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1797 ; /* nothing to do */
1798 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1799 ; /* already parsed */
1800 } else if (starts_with(arg, "--threads=")) {
1801 char *end;
1802 nr_threads = strtoul(arg+10, &end, 0);
1803 if (!arg[10] || *end || nr_threads < 0)
1804 usage(index_pack_usage);
1805 if (!HAVE_THREADS && nr_threads != 1) {
1806 warning(_("no threads support, ignoring %s"), arg);
1807 nr_threads = 1;
1808 }
1809 } else if (starts_with(arg, "--pack_header=")) {
1810 struct pack_header *hdr;
1811 char *c;
1812
1813 hdr = (struct pack_header *)input_buffer;
1814 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1815 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1816 if (*c != ',')
1817 die(_("bad %s"), arg);
1818 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1819 if (*c)
1820 die(_("bad %s"), arg);
1821 input_len = sizeof(*hdr);
1822 } else if (!strcmp(arg, "-v")) {
1823 verbose = 1;
1824 } else if (!strcmp(arg, "--progress-title")) {
1825 if (progress_title || (i+1) >= argc)
1826 usage(index_pack_usage);
1827 progress_title = argv[++i];
1828 } else if (!strcmp(arg, "--show-resolving-progress")) {
1829 show_resolving_progress = 1;
1830 } else if (!strcmp(arg, "--report-end-of-input")) {
1831 report_end_of_input = 1;
1832 } else if (!strcmp(arg, "-o")) {
1833 if (index_name || (i+1) >= argc)
1834 usage(index_pack_usage);
1835 index_name = argv[++i];
1836 } else if (starts_with(arg, "--index-version=")) {
1837 char *c;
1838 opts.version = strtoul(arg + 16, &c, 10);
1839 if (opts.version > 2)
1840 die(_("bad %s"), arg);
1841 if (*c == ',')
1842 opts.off32_limit = strtoul(c+1, &c, 0);
1843 if (*c || opts.off32_limit & 0x80000000)
1844 die(_("bad %s"), arg);
1845 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1846 max_input_size = strtoumax(arg, NULL, 10);
1847 } else if (skip_prefix(arg, "--object-format=", &arg)) {
1848 hash_algo = hash_algo_by_name(arg);
1849 if (hash_algo == GIT_HASH_UNKNOWN)
1850 die(_("unknown hash algorithm '%s'"), arg);
1851 repo_set_hash_algo(the_repository, hash_algo);
1852 } else if (!strcmp(arg, "--rev-index")) {
1853 rev_index = 1;
1854 } else if (!strcmp(arg, "--no-rev-index")) {
1855 rev_index = 0;
1856 } else
1857 usage(index_pack_usage);
1858 continue;
1859 }
1860
1861 if (pack_name)
1862 usage(index_pack_usage);
1863 pack_name = arg;
1864 }
1865
1866 if (!pack_name && !from_stdin)
1867 usage(index_pack_usage);
1868 if (fix_thin_pack && !from_stdin)
1869 die(_("the option '%s' requires '%s'"), "--fix-thin", "--stdin");
1870 if (from_stdin && !startup_info->have_repository)
1871 die(_("--stdin requires a git repository"));
1872 if (from_stdin && hash_algo)
1873 die(_("options '%s' and '%s' cannot be used together"), "--object-format", "--stdin");
1874 if (!index_name && pack_name)
1875 index_name = derive_filename(pack_name, "pack", "idx", &index_name_buf);
1876
1877 opts.flags &= ~(WRITE_REV | WRITE_REV_VERIFY);
1878 if (rev_index) {
1879 opts.flags |= verify ? WRITE_REV_VERIFY : WRITE_REV;
1880 if (index_name)
1881 rev_index_name = derive_filename(index_name,
1882 "idx", "rev",
1883 &rev_index_name_buf);
1884 }
1885
1886 if (verify) {
1887 if (!index_name)
1888 die(_("--verify with no packfile name given"));
1889 read_idx_option(&opts, index_name);
1890 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1891 }
1892 if (strict)
1893 opts.flags |= WRITE_IDX_STRICT;
1894
1895 if (HAVE_THREADS && !nr_threads) {
1896 nr_threads = online_cpus();
1897 /*
1898 * Experiments show that going above 20 threads doesn't help,
1899 * no matter how many cores you have. Below that, we tend to
1900 * max at half the number of online_cpus(), presumably because
1901 * half of those are hyperthreads rather than full cores. We'll
1902 * never reduce the level below "3", though, to match a
1903 * historical value that nobody complained about.
1904 */
1905 if (nr_threads < 4)
1906 ; /* too few cores to consider capping */
1907 else if (nr_threads < 6)
1908 nr_threads = 3; /* historic cap */
1909 else if (nr_threads < 40)
1910 nr_threads /= 2;
1911 else
1912 nr_threads = 20; /* hard cap */
1913 }
1914
1915 curr_pack = open_pack_file(pack_name);
1916 parse_pack_header();
1917 CALLOC_ARRAY(objects, st_add(nr_objects, 1));
1918 if (show_stat)
1919 CALLOC_ARRAY(obj_stat, st_add(nr_objects, 1));
1920 CALLOC_ARRAY(ofs_deltas, nr_objects);
1921 parse_pack_objects(pack_hash);
1922 if (report_end_of_input)
1923 write_in_full(2, "\0", 1);
1924 resolve_deltas();
1925 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1926 free(ofs_deltas);
1927 free(ref_deltas);
1928 if (strict)
1929 foreign_nr = check_objects();
1930
1931 if (show_stat)
1932 show_pack_info(stat_only);
1933
1934 ALLOC_ARRAY(idx_objects, nr_objects);
1935 for (i = 0; i < nr_objects; i++)
1936 idx_objects[i] = &objects[i].idx;
1937 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1938 if (rev_index)
1939 curr_rev_index = write_rev_file(rev_index_name, idx_objects,
1940 nr_objects, pack_hash,
1941 opts.flags);
1942 free(idx_objects);
1943
1944 if (!verify)
1945 final(pack_name, curr_pack,
1946 index_name, curr_index,
1947 rev_index_name, curr_rev_index,
1948 keep_msg, promisor_msg,
1949 pack_hash);
1950 else
1951 close(input_fd);
1952
1953 if (do_fsck_object && fsck_finish(&fsck_options))
1954 die(_("fsck error in pack objects"));
1955
1956 free(opts.anomaly);
1957 free(objects);
1958 strbuf_release(&index_name_buf);
1959 strbuf_release(&rev_index_name_buf);
1960 if (!pack_name)
1961 free((void *) curr_pack);
1962 if (!index_name)
1963 free((void *) curr_index);
1964 if (!rev_index_name)
1965 free((void *) curr_rev_index);
1966
1967 /*
1968 * Let the caller know this pack is not self contained
1969 */
1970 if (check_self_contained_and_connected && foreign_nr)
1971 return 1;
1972
1973 return 0;
1974 }
git for Debian