]> git.ipfire.org Git - thirdparty/git.git/blob - builtin/index-pack.c
Merge branch 'vd/fsck-submodule-url-test'
[thirdparty/git.git] / builtin / index-pack.c
1 #include "builtin.h"
2 #include "config.h"
3 #include "delta.h"
4 #include "environment.h"
5 #include "gettext.h"
6 #include "hex.h"
7 #include "pack.h"
8 #include "csum-file.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tree.h"
12 #include "progress.h"
13 #include "fsck.h"
14 #include "strbuf.h"
15 #include "streaming.h"
16 #include "thread-utils.h"
17 #include "packfile.h"
18 #include "pack-revindex.h"
19 #include "object-file.h"
20 #include "object-store-ll.h"
21 #include "oid-array.h"
22 #include "replace-object.h"
23 #include "promisor-remote.h"
24 #include "setup.h"
25
26 static const char index_pack_usage[] =
27 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--[no-]rev-index] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
28
29 struct object_entry {
30 struct pack_idx_entry idx;
31 unsigned long size;
32 unsigned char hdr_size;
33 signed char type;
34 signed char real_type;
35 };
36
37 struct object_stat {
38 unsigned delta_depth;
39 int base_object_no;
40 };
41
42 struct base_data {
43 /* Initialized by make_base(). */
44 struct base_data *base;
45 struct object_entry *obj;
46 int ref_first, ref_last;
47 int ofs_first, ofs_last;
48 /*
49 * Threads should increment retain_data if they are about to call
50 * patch_delta() using this struct's data as a base, and decrement this
51 * when they are done. While retain_data is nonzero, this struct's data
52 * will not be freed even if the delta base cache limit is exceeded.
53 */
54 int retain_data;
55 /*
56 * The number of direct children that have not been fully processed
57 * (entered work_head, entered done_head, left done_head). When this
58 * number reaches zero, this struct base_data can be freed.
59 */
60 int children_remaining;
61
62 /* Not initialized by make_base(). */
63 struct list_head list;
64 void *data;
65 unsigned long size;
66 };
67
68 /*
69 * Stack of struct base_data that have unprocessed children.
70 * threaded_second_pass() uses this as a source of work (the other being the
71 * objects array).
72 *
73 * Guarded by work_mutex.
74 */
75 static LIST_HEAD(work_head);
76
77 /*
78 * Stack of struct base_data that have children, all of whom have been
79 * processed or are being processed, and at least one child is being processed.
80 * These struct base_data must be kept around until the last child is
81 * processed.
82 *
83 * Guarded by work_mutex.
84 */
85 static LIST_HEAD(done_head);
86
87 /*
88 * All threads share one delta base cache.
89 *
90 * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only
91 * in a thread.
92 */
93 static size_t base_cache_used;
94 static size_t base_cache_limit;
95
96 struct thread_local {
97 pthread_t thread;
98 int pack_fd;
99 };
100
101 /* Remember to update object flag allocation in object.h */
102 #define FLAG_LINK (1u<<20)
103 #define FLAG_CHECKED (1u<<21)
104
105 struct ofs_delta_entry {
106 off_t offset;
107 int obj_no;
108 };
109
110 struct ref_delta_entry {
111 struct object_id oid;
112 int obj_no;
113 };
114
115 static struct object_entry *objects;
116 static struct object_stat *obj_stat;
117 static struct ofs_delta_entry *ofs_deltas;
118 static struct ref_delta_entry *ref_deltas;
119 static struct thread_local nothread_data;
120 static int nr_objects;
121 static int nr_ofs_deltas;
122 static int nr_ref_deltas;
123 static int ref_deltas_alloc;
124 static int nr_resolved_deltas;
125 static int nr_threads;
126
127 static int from_stdin;
128 static int strict;
129 static int do_fsck_object;
130 static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES;
131 static int verbose;
132 static const char *progress_title;
133 static int show_resolving_progress;
134 static int show_stat;
135 static int check_self_contained_and_connected;
136
137 static struct progress *progress;
138
139 /* We always read in 4kB chunks. */
140 static unsigned char input_buffer[4096];
141 static unsigned int input_offset, input_len;
142 static off_t consumed_bytes;
143 static off_t max_input_size;
144 static unsigned deepest_delta;
145 static git_hash_ctx input_ctx;
146 static uint32_t input_crc32;
147 static int input_fd, output_fd;
148 static const char *curr_pack;
149
150 static struct thread_local *thread_data;
151 static int nr_dispatched;
152 static int threads_active;
153
154 static pthread_mutex_t read_mutex;
155 #define read_lock() lock_mutex(&read_mutex)
156 #define read_unlock() unlock_mutex(&read_mutex)
157
158 static pthread_mutex_t counter_mutex;
159 #define counter_lock() lock_mutex(&counter_mutex)
160 #define counter_unlock() unlock_mutex(&counter_mutex)
161
162 static pthread_mutex_t work_mutex;
163 #define work_lock() lock_mutex(&work_mutex)
164 #define work_unlock() unlock_mutex(&work_mutex)
165
166 static pthread_mutex_t deepest_delta_mutex;
167 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
168 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
169
170 static pthread_key_t key;
171
172 static inline void lock_mutex(pthread_mutex_t *mutex)
173 {
174 if (threads_active)
175 pthread_mutex_lock(mutex);
176 }
177
178 static inline void unlock_mutex(pthread_mutex_t *mutex)
179 {
180 if (threads_active)
181 pthread_mutex_unlock(mutex);
182 }
183
184 /*
185 * Mutex and conditional variable can't be statically-initialized on Windows.
186 */
187 static void init_thread(void)
188 {
189 int i;
190 init_recursive_mutex(&read_mutex);
191 pthread_mutex_init(&counter_mutex, NULL);
192 pthread_mutex_init(&work_mutex, NULL);
193 if (show_stat)
194 pthread_mutex_init(&deepest_delta_mutex, NULL);
195 pthread_key_create(&key, NULL);
196 CALLOC_ARRAY(thread_data, nr_threads);
197 for (i = 0; i < nr_threads; i++) {
198 thread_data[i].pack_fd = xopen(curr_pack, O_RDONLY);
199 }
200
201 threads_active = 1;
202 }
203
204 static void cleanup_thread(void)
205 {
206 int i;
207 if (!threads_active)
208 return;
209 threads_active = 0;
210 pthread_mutex_destroy(&read_mutex);
211 pthread_mutex_destroy(&counter_mutex);
212 pthread_mutex_destroy(&work_mutex);
213 if (show_stat)
214 pthread_mutex_destroy(&deepest_delta_mutex);
215 for (i = 0; i < nr_threads; i++)
216 close(thread_data[i].pack_fd);
217 pthread_key_delete(key);
218 free(thread_data);
219 }
220
221 static int mark_link(struct object *obj, enum object_type type,
222 void *data UNUSED,
223 struct fsck_options *options UNUSED)
224 {
225 if (!obj)
226 return -1;
227
228 if (type != OBJ_ANY && obj->type != type)
229 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
230
231 obj->flags |= FLAG_LINK;
232 return 0;
233 }
234
235 /* The content of each linked object must have been checked
236 or it must be already present in the object database */
237 static unsigned check_object(struct object *obj)
238 {
239 if (!obj)
240 return 0;
241
242 if (!(obj->flags & FLAG_LINK))
243 return 0;
244
245 if (!(obj->flags & FLAG_CHECKED)) {
246 unsigned long size;
247 int type = oid_object_info(the_repository, &obj->oid, &size);
248 if (type <= 0)
249 die(_("did not receive expected object %s"),
250 oid_to_hex(&obj->oid));
251 if (type != obj->type)
252 die(_("object %s: expected type %s, found %s"),
253 oid_to_hex(&obj->oid),
254 type_name(obj->type), type_name(type));
255 obj->flags |= FLAG_CHECKED;
256 return 1;
257 }
258
259 return 0;
260 }
261
262 static unsigned check_objects(void)
263 {
264 unsigned i, max, foreign_nr = 0;
265
266 max = get_max_object_index();
267
268 if (verbose)
269 progress = start_delayed_progress(_("Checking objects"), max);
270
271 for (i = 0; i < max; i++) {
272 foreign_nr += check_object(get_indexed_object(i));
273 display_progress(progress, i + 1);
274 }
275
276 stop_progress(&progress);
277 return foreign_nr;
278 }
279
280
281 /* Discard current buffer used content. */
282 static void flush(void)
283 {
284 if (input_offset) {
285 if (output_fd >= 0)
286 write_or_die(output_fd, input_buffer, input_offset);
287 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
288 memmove(input_buffer, input_buffer + input_offset, input_len);
289 input_offset = 0;
290 }
291 }
292
293 /*
294 * Make sure at least "min" bytes are available in the buffer, and
295 * return the pointer to the buffer.
296 */
297 static void *fill(int min)
298 {
299 if (min <= input_len)
300 return input_buffer + input_offset;
301 if (min > sizeof(input_buffer))
302 die(Q_("cannot fill %d byte",
303 "cannot fill %d bytes",
304 min),
305 min);
306 flush();
307 do {
308 ssize_t ret = xread(input_fd, input_buffer + input_len,
309 sizeof(input_buffer) - input_len);
310 if (ret <= 0) {
311 if (!ret)
312 die(_("early EOF"));
313 die_errno(_("read error on input"));
314 }
315 input_len += ret;
316 if (from_stdin)
317 display_throughput(progress, consumed_bytes + input_len);
318 } while (input_len < min);
319 return input_buffer;
320 }
321
322 static void use(int bytes)
323 {
324 if (bytes > input_len)
325 die(_("used more bytes than were available"));
326 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
327 input_len -= bytes;
328 input_offset += bytes;
329
330 /* make sure off_t is sufficiently large not to wrap */
331 if (signed_add_overflows(consumed_bytes, bytes))
332 die(_("pack too large for current definition of off_t"));
333 consumed_bytes += bytes;
334 if (max_input_size && consumed_bytes > max_input_size) {
335 struct strbuf size_limit = STRBUF_INIT;
336 strbuf_humanise_bytes(&size_limit, max_input_size);
337 die(_("pack exceeds maximum allowed size (%s)"),
338 size_limit.buf);
339 }
340 }
341
342 static const char *open_pack_file(const char *pack_name)
343 {
344 if (from_stdin) {
345 input_fd = 0;
346 if (!pack_name) {
347 struct strbuf tmp_file = STRBUF_INIT;
348 output_fd = odb_mkstemp(&tmp_file,
349 "pack/tmp_pack_XXXXXX");
350 pack_name = strbuf_detach(&tmp_file, NULL);
351 } else {
352 output_fd = xopen(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
353 }
354 nothread_data.pack_fd = output_fd;
355 } else {
356 input_fd = xopen(pack_name, O_RDONLY);
357 output_fd = -1;
358 nothread_data.pack_fd = input_fd;
359 }
360 the_hash_algo->init_fn(&input_ctx);
361 return pack_name;
362 }
363
364 static void parse_pack_header(void)
365 {
366 struct pack_header *hdr = fill(sizeof(struct pack_header));
367
368 /* Header consistency check */
369 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
370 die(_("pack signature mismatch"));
371 if (!pack_version_ok(hdr->hdr_version))
372 die(_("pack version %"PRIu32" unsupported"),
373 ntohl(hdr->hdr_version));
374
375 nr_objects = ntohl(hdr->hdr_entries);
376 use(sizeof(struct pack_header));
377 }
378
379 __attribute__((format (printf, 2, 3)))
380 static NORETURN void bad_object(off_t offset, const char *format, ...)
381 {
382 va_list params;
383 char buf[1024];
384
385 va_start(params, format);
386 vsnprintf(buf, sizeof(buf), format, params);
387 va_end(params);
388 die(_("pack has bad object at offset %"PRIuMAX": %s"),
389 (uintmax_t)offset, buf);
390 }
391
392 static inline struct thread_local *get_thread_data(void)
393 {
394 if (HAVE_THREADS) {
395 if (threads_active)
396 return pthread_getspecific(key);
397 assert(!threads_active &&
398 "This should only be reached when all threads are gone");
399 }
400 return &nothread_data;
401 }
402
403 static void set_thread_data(struct thread_local *data)
404 {
405 if (threads_active)
406 pthread_setspecific(key, data);
407 }
408
409 static void free_base_data(struct base_data *c)
410 {
411 if (c->data) {
412 FREE_AND_NULL(c->data);
413 base_cache_used -= c->size;
414 }
415 }
416
417 static void prune_base_data(struct base_data *retain)
418 {
419 struct list_head *pos;
420
421 if (base_cache_used <= base_cache_limit)
422 return;
423
424 list_for_each_prev(pos, &done_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 list_for_each_prev(pos, &work_head) {
436 struct base_data *b = list_entry(pos, struct base_data, list);
437 if (b->retain_data || b == retain)
438 continue;
439 if (b->data) {
440 free_base_data(b);
441 if (base_cache_used <= base_cache_limit)
442 return;
443 }
444 }
445 }
446
447 static int is_delta_type(enum object_type type)
448 {
449 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
450 }
451
452 static void *unpack_entry_data(off_t offset, unsigned long size,
453 enum object_type type, struct object_id *oid)
454 {
455 static char fixed_buf[8192];
456 int status;
457 git_zstream stream;
458 void *buf;
459 git_hash_ctx c;
460 char hdr[32];
461 int hdrlen;
462
463 if (!is_delta_type(type)) {
464 hdrlen = format_object_header(hdr, sizeof(hdr), type, size);
465 the_hash_algo->init_fn(&c);
466 the_hash_algo->update_fn(&c, hdr, hdrlen);
467 } else
468 oid = NULL;
469 if (type == OBJ_BLOB && size > big_file_threshold)
470 buf = fixed_buf;
471 else
472 buf = xmallocz(size);
473
474 memset(&stream, 0, sizeof(stream));
475 git_inflate_init(&stream);
476 stream.next_out = buf;
477 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
478
479 do {
480 unsigned char *last_out = stream.next_out;
481 stream.next_in = fill(1);
482 stream.avail_in = input_len;
483 status = git_inflate(&stream, 0);
484 use(input_len - stream.avail_in);
485 if (oid)
486 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
487 if (buf == fixed_buf) {
488 stream.next_out = buf;
489 stream.avail_out = sizeof(fixed_buf);
490 }
491 } while (status == Z_OK);
492 if (stream.total_out != size || status != Z_STREAM_END)
493 bad_object(offset, _("inflate returned %d"), status);
494 git_inflate_end(&stream);
495 if (oid)
496 the_hash_algo->final_oid_fn(oid, &c);
497 return buf == fixed_buf ? NULL : buf;
498 }
499
500 static void *unpack_raw_entry(struct object_entry *obj,
501 off_t *ofs_offset,
502 struct object_id *ref_oid,
503 struct object_id *oid)
504 {
505 unsigned char *p;
506 unsigned long size, c;
507 off_t base_offset;
508 unsigned shift;
509 void *data;
510
511 obj->idx.offset = consumed_bytes;
512 input_crc32 = crc32(0, NULL, 0);
513
514 p = fill(1);
515 c = *p;
516 use(1);
517 obj->type = (c >> 4) & 7;
518 size = (c & 15);
519 shift = 4;
520 while (c & 0x80) {
521 p = fill(1);
522 c = *p;
523 use(1);
524 size += (c & 0x7f) << shift;
525 shift += 7;
526 }
527 obj->size = size;
528
529 switch (obj->type) {
530 case OBJ_REF_DELTA:
531 oidread(ref_oid, fill(the_hash_algo->rawsz));
532 use(the_hash_algo->rawsz);
533 break;
534 case OBJ_OFS_DELTA:
535 p = fill(1);
536 c = *p;
537 use(1);
538 base_offset = c & 127;
539 while (c & 128) {
540 base_offset += 1;
541 if (!base_offset || MSB(base_offset, 7))
542 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
543 p = fill(1);
544 c = *p;
545 use(1);
546 base_offset = (base_offset << 7) + (c & 127);
547 }
548 *ofs_offset = obj->idx.offset - base_offset;
549 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
550 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
551 break;
552 case OBJ_COMMIT:
553 case OBJ_TREE:
554 case OBJ_BLOB:
555 case OBJ_TAG:
556 break;
557 default:
558 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
559 }
560 obj->hdr_size = consumed_bytes - obj->idx.offset;
561
562 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
563 obj->idx.crc32 = input_crc32;
564 return data;
565 }
566
567 static void *unpack_data(struct object_entry *obj,
568 int (*consume)(const unsigned char *, unsigned long, void *),
569 void *cb_data)
570 {
571 off_t from = obj[0].idx.offset + obj[0].hdr_size;
572 off_t len = obj[1].idx.offset - from;
573 unsigned char *data, *inbuf;
574 git_zstream stream;
575 int status;
576
577 data = xmallocz(consume ? 64*1024 : obj->size);
578 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
579
580 memset(&stream, 0, sizeof(stream));
581 git_inflate_init(&stream);
582 stream.next_out = data;
583 stream.avail_out = consume ? 64*1024 : obj->size;
584
585 do {
586 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
587 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
588 if (n < 0)
589 die_errno(_("cannot pread pack file"));
590 if (!n)
591 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
592 "premature end of pack file, %"PRIuMAX" bytes missing",
593 len),
594 (uintmax_t)len);
595 from += n;
596 len -= n;
597 stream.next_in = inbuf;
598 stream.avail_in = n;
599 if (!consume)
600 status = git_inflate(&stream, 0);
601 else {
602 do {
603 status = git_inflate(&stream, 0);
604 if (consume(data, stream.next_out - data, cb_data)) {
605 free(inbuf);
606 free(data);
607 return NULL;
608 }
609 stream.next_out = data;
610 stream.avail_out = 64*1024;
611 } while (status == Z_OK && stream.avail_in);
612 }
613 } while (len && status == Z_OK && !stream.avail_in);
614
615 /* This has been inflated OK when first encountered, so... */
616 if (status != Z_STREAM_END || stream.total_out != obj->size)
617 die(_("serious inflate inconsistency"));
618
619 git_inflate_end(&stream);
620 free(inbuf);
621 if (consume) {
622 FREE_AND_NULL(data);
623 }
624 return data;
625 }
626
627 static void *get_data_from_pack(struct object_entry *obj)
628 {
629 return unpack_data(obj, NULL, NULL);
630 }
631
632 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
633 enum object_type type1,
634 enum object_type type2)
635 {
636 int cmp = type1 - type2;
637 if (cmp)
638 return cmp;
639 return offset1 < offset2 ? -1 :
640 offset1 > offset2 ? 1 :
641 0;
642 }
643
644 static int find_ofs_delta(const off_t offset)
645 {
646 int first = 0, last = nr_ofs_deltas;
647
648 while (first < last) {
649 int next = first + (last - first) / 2;
650 struct ofs_delta_entry *delta = &ofs_deltas[next];
651 int cmp;
652
653 cmp = compare_ofs_delta_bases(offset, delta->offset,
654 OBJ_OFS_DELTA,
655 objects[delta->obj_no].type);
656 if (!cmp)
657 return next;
658 if (cmp < 0) {
659 last = next;
660 continue;
661 }
662 first = next+1;
663 }
664 return -first-1;
665 }
666
667 static void find_ofs_delta_children(off_t offset,
668 int *first_index, int *last_index)
669 {
670 int first = find_ofs_delta(offset);
671 int last = first;
672 int end = nr_ofs_deltas - 1;
673
674 if (first < 0) {
675 *first_index = 0;
676 *last_index = -1;
677 return;
678 }
679 while (first > 0 && ofs_deltas[first - 1].offset == offset)
680 --first;
681 while (last < end && ofs_deltas[last + 1].offset == offset)
682 ++last;
683 *first_index = first;
684 *last_index = last;
685 }
686
687 static int compare_ref_delta_bases(const struct object_id *oid1,
688 const struct object_id *oid2,
689 enum object_type type1,
690 enum object_type type2)
691 {
692 int cmp = type1 - type2;
693 if (cmp)
694 return cmp;
695 return oidcmp(oid1, oid2);
696 }
697
698 static int find_ref_delta(const struct object_id *oid)
699 {
700 int first = 0, last = nr_ref_deltas;
701
702 while (first < last) {
703 int next = first + (last - first) / 2;
704 struct ref_delta_entry *delta = &ref_deltas[next];
705 int cmp;
706
707 cmp = compare_ref_delta_bases(oid, &delta->oid,
708 OBJ_REF_DELTA,
709 objects[delta->obj_no].type);
710 if (!cmp)
711 return next;
712 if (cmp < 0) {
713 last = next;
714 continue;
715 }
716 first = next+1;
717 }
718 return -first-1;
719 }
720
721 static void find_ref_delta_children(const struct object_id *oid,
722 int *first_index, int *last_index)
723 {
724 int first = find_ref_delta(oid);
725 int last = first;
726 int end = nr_ref_deltas - 1;
727
728 if (first < 0) {
729 *first_index = 0;
730 *last_index = -1;
731 return;
732 }
733 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
734 --first;
735 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
736 ++last;
737 *first_index = first;
738 *last_index = last;
739 }
740
741 struct compare_data {
742 struct object_entry *entry;
743 struct git_istream *st;
744 unsigned char *buf;
745 unsigned long buf_size;
746 };
747
748 static int compare_objects(const unsigned char *buf, unsigned long size,
749 void *cb_data)
750 {
751 struct compare_data *data = cb_data;
752
753 if (data->buf_size < size) {
754 free(data->buf);
755 data->buf = xmalloc(size);
756 data->buf_size = size;
757 }
758
759 while (size) {
760 ssize_t len = read_istream(data->st, data->buf, size);
761 if (len == 0)
762 die(_("SHA1 COLLISION FOUND WITH %s !"),
763 oid_to_hex(&data->entry->idx.oid));
764 if (len < 0)
765 die(_("unable to read %s"),
766 oid_to_hex(&data->entry->idx.oid));
767 if (memcmp(buf, data->buf, len))
768 die(_("SHA1 COLLISION FOUND WITH %s !"),
769 oid_to_hex(&data->entry->idx.oid));
770 size -= len;
771 buf += len;
772 }
773 return 0;
774 }
775
776 static int check_collison(struct object_entry *entry)
777 {
778 struct compare_data data;
779 enum object_type type;
780 unsigned long size;
781
782 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
783 return -1;
784
785 memset(&data, 0, sizeof(data));
786 data.entry = entry;
787 data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
788 NULL);
789 if (!data.st)
790 return -1;
791 if (size != entry->size || type != entry->type)
792 die(_("SHA1 COLLISION FOUND WITH %s !"),
793 oid_to_hex(&entry->idx.oid));
794 unpack_data(entry, compare_objects, &data);
795 close_istream(data.st);
796 free(data.buf);
797 return 0;
798 }
799
800 static void sha1_object(const void *data, struct object_entry *obj_entry,
801 unsigned long size, enum object_type type,
802 const struct object_id *oid)
803 {
804 void *new_data = NULL;
805 int collision_test_needed = 0;
806
807 assert(data || obj_entry);
808
809 if (startup_info->have_repository) {
810 read_lock();
811 collision_test_needed =
812 repo_has_object_file_with_flags(the_repository, oid,
813 OBJECT_INFO_QUICK);
814 read_unlock();
815 }
816
817 if (collision_test_needed && !data) {
818 read_lock();
819 if (!check_collison(obj_entry))
820 collision_test_needed = 0;
821 read_unlock();
822 }
823 if (collision_test_needed) {
824 void *has_data;
825 enum object_type has_type;
826 unsigned long has_size;
827 read_lock();
828 has_type = oid_object_info(the_repository, oid, &has_size);
829 if (has_type < 0)
830 die(_("cannot read existing object info %s"), oid_to_hex(oid));
831 if (has_type != type || has_size != size)
832 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
833 has_data = repo_read_object_file(the_repository, oid,
834 &has_type, &has_size);
835 read_unlock();
836 if (!data)
837 data = new_data = get_data_from_pack(obj_entry);
838 if (!has_data)
839 die(_("cannot read existing object %s"), oid_to_hex(oid));
840 if (size != has_size || type != has_type ||
841 memcmp(data, has_data, size) != 0)
842 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
843 free(has_data);
844 }
845
846 if (strict || do_fsck_object) {
847 read_lock();
848 if (type == OBJ_BLOB) {
849 struct blob *blob = lookup_blob(the_repository, oid);
850 if (blob)
851 blob->object.flags |= FLAG_CHECKED;
852 else
853 die(_("invalid blob object %s"), oid_to_hex(oid));
854 if (do_fsck_object &&
855 fsck_object(&blob->object, (void *)data, size, &fsck_options))
856 die(_("fsck error in packed object"));
857 } else {
858 struct object *obj;
859 int eaten;
860 void *buf = (void *) data;
861
862 assert(data && "data can only be NULL for large _blobs_");
863
864 /*
865 * we do not need to free the memory here, as the
866 * buf is deleted by the caller.
867 */
868 obj = parse_object_buffer(the_repository, oid, type,
869 size, buf,
870 &eaten);
871 if (!obj)
872 die(_("invalid %s"), type_name(type));
873 if (do_fsck_object &&
874 fsck_object(obj, buf, size, &fsck_options))
875 die(_("fsck error in packed object"));
876 if (strict && fsck_walk(obj, NULL, &fsck_options))
877 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
878
879 if (obj->type == OBJ_TREE) {
880 struct tree *item = (struct tree *) obj;
881 item->buffer = NULL;
882 obj->parsed = 0;
883 }
884 if (obj->type == OBJ_COMMIT) {
885 struct commit *commit = (struct commit *) obj;
886 if (detach_commit_buffer(commit, NULL) != data)
887 BUG("parse_object_buffer transmogrified our buffer");
888 }
889 obj->flags |= FLAG_CHECKED;
890 }
891 read_unlock();
892 }
893
894 free(new_data);
895 }
896
897 /*
898 * Ensure that this node has been reconstructed and return its contents.
899 *
900 * In the typical and best case, this node would already be reconstructed
901 * (through the invocation to resolve_delta() in threaded_second_pass()) and it
902 * would not be pruned. However, if pruning of this node was necessary due to
903 * reaching delta_base_cache_limit, this function will find the closest
904 * ancestor with reconstructed data that has not been pruned (or if there is
905 * none, the ultimate base object), and reconstruct each node in the delta
906 * chain in order to generate the reconstructed data for this node.
907 */
908 static void *get_base_data(struct base_data *c)
909 {
910 if (!c->data) {
911 struct object_entry *obj = c->obj;
912 struct base_data **delta = NULL;
913 int delta_nr = 0, delta_alloc = 0;
914
915 while (is_delta_type(c->obj->type) && !c->data) {
916 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
917 delta[delta_nr++] = c;
918 c = c->base;
919 }
920 if (!delta_nr) {
921 c->data = get_data_from_pack(obj);
922 c->size = obj->size;
923 base_cache_used += c->size;
924 prune_base_data(c);
925 }
926 for (; delta_nr > 0; delta_nr--) {
927 void *base, *raw;
928 c = delta[delta_nr - 1];
929 obj = c->obj;
930 base = get_base_data(c->base);
931 raw = get_data_from_pack(obj);
932 c->data = patch_delta(
933 base, c->base->size,
934 raw, obj->size,
935 &c->size);
936 free(raw);
937 if (!c->data)
938 bad_object(obj->idx.offset, _("failed to apply delta"));
939 base_cache_used += c->size;
940 prune_base_data(c);
941 }
942 free(delta);
943 }
944 return c->data;
945 }
946
947 static struct base_data *make_base(struct object_entry *obj,
948 struct base_data *parent)
949 {
950 struct base_data *base = xcalloc(1, sizeof(struct base_data));
951 base->base = parent;
952 base->obj = obj;
953 find_ref_delta_children(&obj->idx.oid,
954 &base->ref_first, &base->ref_last);
955 find_ofs_delta_children(obj->idx.offset,
956 &base->ofs_first, &base->ofs_last);
957 base->children_remaining = base->ref_last - base->ref_first +
958 base->ofs_last - base->ofs_first + 2;
959 return base;
960 }
961
962 static struct base_data *resolve_delta(struct object_entry *delta_obj,
963 struct base_data *base)
964 {
965 void *delta_data, *result_data;
966 struct base_data *result;
967 unsigned long result_size;
968
969 if (show_stat) {
970 int i = delta_obj - objects;
971 int j = base->obj - objects;
972 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
973 deepest_delta_lock();
974 if (deepest_delta < obj_stat[i].delta_depth)
975 deepest_delta = obj_stat[i].delta_depth;
976 deepest_delta_unlock();
977 obj_stat[i].base_object_no = j;
978 }
979 delta_data = get_data_from_pack(delta_obj);
980 assert(base->data);
981 result_data = patch_delta(base->data, base->size,
982 delta_data, delta_obj->size, &result_size);
983 free(delta_data);
984 if (!result_data)
985 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
986 hash_object_file(the_hash_algo, result_data, result_size,
987 delta_obj->real_type, &delta_obj->idx.oid);
988 sha1_object(result_data, NULL, result_size, delta_obj->real_type,
989 &delta_obj->idx.oid);
990
991 result = make_base(delta_obj, base);
992 result->data = result_data;
993 result->size = result_size;
994
995 counter_lock();
996 nr_resolved_deltas++;
997 counter_unlock();
998
999 return result;
1000 }
1001
1002 static int compare_ofs_delta_entry(const void *a, const void *b)
1003 {
1004 const struct ofs_delta_entry *delta_a = a;
1005 const struct ofs_delta_entry *delta_b = b;
1006
1007 return delta_a->offset < delta_b->offset ? -1 :
1008 delta_a->offset > delta_b->offset ? 1 :
1009 0;
1010 }
1011
1012 static int compare_ref_delta_entry(const void *a, const void *b)
1013 {
1014 const struct ref_delta_entry *delta_a = a;
1015 const struct ref_delta_entry *delta_b = b;
1016
1017 return oidcmp(&delta_a->oid, &delta_b->oid);
1018 }
1019
1020 static void *threaded_second_pass(void *data)
1021 {
1022 if (data)
1023 set_thread_data(data);
1024 for (;;) {
1025 struct base_data *parent = NULL;
1026 struct object_entry *child_obj;
1027 struct base_data *child;
1028
1029 counter_lock();
1030 display_progress(progress, nr_resolved_deltas);
1031 counter_unlock();
1032
1033 work_lock();
1034 if (list_empty(&work_head)) {
1035 /*
1036 * Take an object from the object array.
1037 */
1038 while (nr_dispatched < nr_objects &&
1039 is_delta_type(objects[nr_dispatched].type))
1040 nr_dispatched++;
1041 if (nr_dispatched >= nr_objects) {
1042 work_unlock();
1043 break;
1044 }
1045 child_obj = &objects[nr_dispatched++];
1046 } else {
1047 /*
1048 * Peek at the top of the stack, and take a child from
1049 * it.
1050 */
1051 parent = list_first_entry(&work_head, struct base_data,
1052 list);
1053
1054 if (parent->ref_first <= parent->ref_last) {
1055 int offset = ref_deltas[parent->ref_first++].obj_no;
1056 child_obj = objects + offset;
1057 if (child_obj->real_type != OBJ_REF_DELTA)
1058 die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)",
1059 (uintmax_t) child_obj->idx.offset,
1060 oid_to_hex(&parent->obj->idx.oid));
1061 child_obj->real_type = parent->obj->real_type;
1062 } else {
1063 child_obj = objects +
1064 ofs_deltas[parent->ofs_first++].obj_no;
1065 assert(child_obj->real_type == OBJ_OFS_DELTA);
1066 child_obj->real_type = parent->obj->real_type;
1067 }
1068
1069 if (parent->ref_first > parent->ref_last &&
1070 parent->ofs_first > parent->ofs_last) {
1071 /*
1072 * This parent has run out of children, so move
1073 * it to done_head.
1074 */
1075 list_del(&parent->list);
1076 list_add(&parent->list, &done_head);
1077 }
1078
1079 /*
1080 * Ensure that the parent has data, since we will need
1081 * it later.
1082 *
1083 * NEEDSWORK: If parent data needs to be reloaded, this
1084 * prolongs the time that the current thread spends in
1085 * the mutex. A mitigating factor is that parent data
1086 * needs to be reloaded only if the delta base cache
1087 * limit is exceeded, so in the typical case, this does
1088 * not happen.
1089 */
1090 get_base_data(parent);
1091 parent->retain_data++;
1092 }
1093 work_unlock();
1094
1095 if (parent) {
1096 child = resolve_delta(child_obj, parent);
1097 if (!child->children_remaining)
1098 FREE_AND_NULL(child->data);
1099 } else {
1100 child = make_base(child_obj, NULL);
1101 if (child->children_remaining) {
1102 /*
1103 * Since this child has its own delta children,
1104 * we will need this data in the future.
1105 * Inflate now so that future iterations will
1106 * have access to this object's data while
1107 * outside the work mutex.
1108 */
1109 child->data = get_data_from_pack(child_obj);
1110 child->size = child_obj->size;
1111 }
1112 }
1113
1114 work_lock();
1115 if (parent)
1116 parent->retain_data--;
1117 if (child->data) {
1118 /*
1119 * This child has its own children, so add it to
1120 * work_head.
1121 */
1122 list_add(&child->list, &work_head);
1123 base_cache_used += child->size;
1124 prune_base_data(NULL);
1125 free_base_data(child);
1126 } else {
1127 /*
1128 * This child does not have its own children. It may be
1129 * the last descendant of its ancestors; free those
1130 * that we can.
1131 */
1132 struct base_data *p = parent;
1133
1134 while (p) {
1135 struct base_data *next_p;
1136
1137 p->children_remaining--;
1138 if (p->children_remaining)
1139 break;
1140
1141 next_p = p->base;
1142 free_base_data(p);
1143 list_del(&p->list);
1144 free(p);
1145
1146 p = next_p;
1147 }
1148 FREE_AND_NULL(child);
1149 }
1150 work_unlock();
1151 }
1152 return NULL;
1153 }
1154
1155 /*
1156 * First pass:
1157 * - find locations of all objects;
1158 * - calculate SHA1 of all non-delta objects;
1159 * - remember base (SHA1 or offset) for all deltas.
1160 */
1161 static void parse_pack_objects(unsigned char *hash)
1162 {
1163 int i, nr_delays = 0;
1164 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1165 struct object_id ref_delta_oid;
1166 struct stat st;
1167 git_hash_ctx tmp_ctx;
1168
1169 if (verbose)
1170 progress = start_progress(
1171 progress_title ? progress_title :
1172 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1173 nr_objects);
1174 for (i = 0; i < nr_objects; i++) {
1175 struct object_entry *obj = &objects[i];
1176 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1177 &ref_delta_oid,
1178 &obj->idx.oid);
1179 obj->real_type = obj->type;
1180 if (obj->type == OBJ_OFS_DELTA) {
1181 nr_ofs_deltas++;
1182 ofs_delta->obj_no = i;
1183 ofs_delta++;
1184 } else if (obj->type == OBJ_REF_DELTA) {
1185 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1186 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1187 ref_deltas[nr_ref_deltas].obj_no = i;
1188 nr_ref_deltas++;
1189 } else if (!data) {
1190 /* large blobs, check later */
1191 obj->real_type = OBJ_BAD;
1192 nr_delays++;
1193 } else
1194 sha1_object(data, NULL, obj->size, obj->type,
1195 &obj->idx.oid);
1196 free(data);
1197 display_progress(progress, i+1);
1198 }
1199 objects[i].idx.offset = consumed_bytes;
1200 stop_progress(&progress);
1201
1202 /* Check pack integrity */
1203 flush();
1204 the_hash_algo->init_fn(&tmp_ctx);
1205 the_hash_algo->clone_fn(&tmp_ctx, &input_ctx);
1206 the_hash_algo->final_fn(hash, &tmp_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 work_lock();
1259 for (i = 0; i < nr_threads; i++) {
1260 int ret = pthread_create(&thread_data[i].thread, NULL,
1261 threaded_second_pass, thread_data + i);
1262 if (ret)
1263 die(_("unable to create thread: %s"),
1264 strerror(ret));
1265 }
1266 work_unlock();
1267 for (i = 0; i < nr_threads; i++)
1268 pthread_join(thread_data[i].thread, NULL);
1269 cleanup_thread();
1270 return;
1271 }
1272 threaded_second_pass(&nothread_data);
1273 }
1274
1275 /*
1276 * Third pass:
1277 * - append objects to convert thin pack to full pack if required
1278 * - write the final pack hash
1279 */
1280 static void fix_unresolved_deltas(struct hashfile *f);
1281 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1282 {
1283 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1284 stop_progress(&progress);
1285 /* Flush remaining pack final hash. */
1286 flush();
1287 return;
1288 }
1289
1290 if (fix_thin_pack) {
1291 struct hashfile *f;
1292 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1293 struct strbuf msg = STRBUF_INIT;
1294 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1295 int nr_objects_initial = nr_objects;
1296 if (nr_unresolved <= 0)
1297 die(_("confusion beyond insanity"));
1298 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1299 memset(objects + nr_objects + 1, 0,
1300 nr_unresolved * sizeof(*objects));
1301 f = hashfd(output_fd, curr_pack);
1302 fix_unresolved_deltas(f);
1303 strbuf_addf(&msg, Q_("completed with %d local object",
1304 "completed with %d local objects",
1305 nr_objects - nr_objects_initial),
1306 nr_objects - nr_objects_initial);
1307 stop_progress_msg(&progress, msg.buf);
1308 strbuf_release(&msg);
1309 finalize_hashfile(f, tail_hash, FSYNC_COMPONENT_PACK, 0);
1310 hashcpy(read_hash, pack_hash);
1311 fixup_pack_header_footer(output_fd, pack_hash,
1312 curr_pack, nr_objects,
1313 read_hash, consumed_bytes-the_hash_algo->rawsz);
1314 if (!hasheq(read_hash, tail_hash))
1315 die(_("Unexpected tail checksum for %s "
1316 "(disk corruption?)"), curr_pack);
1317 }
1318 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1319 die(Q_("pack has %d unresolved delta",
1320 "pack has %d unresolved deltas",
1321 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1322 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1323 }
1324
1325 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1326 {
1327 git_zstream stream;
1328 int status;
1329 unsigned char outbuf[4096];
1330
1331 git_deflate_init(&stream, zlib_compression_level);
1332 stream.next_in = in;
1333 stream.avail_in = size;
1334
1335 do {
1336 stream.next_out = outbuf;
1337 stream.avail_out = sizeof(outbuf);
1338 status = git_deflate(&stream, Z_FINISH);
1339 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1340 } while (status == Z_OK);
1341
1342 if (status != Z_STREAM_END)
1343 die(_("unable to deflate appended object (%d)"), status);
1344 size = stream.total_out;
1345 git_deflate_end(&stream);
1346 return size;
1347 }
1348
1349 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1350 const unsigned char *sha1, void *buf,
1351 unsigned long size, enum object_type type)
1352 {
1353 struct object_entry *obj = &objects[nr_objects++];
1354 unsigned char header[10];
1355 unsigned long s = size;
1356 int n = 0;
1357 unsigned char c = (type << 4) | (s & 15);
1358 s >>= 4;
1359 while (s) {
1360 header[n++] = c | 0x80;
1361 c = s & 0x7f;
1362 s >>= 7;
1363 }
1364 header[n++] = c;
1365 crc32_begin(f);
1366 hashwrite(f, header, n);
1367 obj[0].size = size;
1368 obj[0].hdr_size = n;
1369 obj[0].type = type;
1370 obj[0].real_type = type;
1371 obj[1].idx.offset = obj[0].idx.offset + n;
1372 obj[1].idx.offset += write_compressed(f, buf, size);
1373 obj[0].idx.crc32 = crc32_end(f);
1374 hashflush(f);
1375 oidread(&obj->idx.oid, sha1);
1376 return obj;
1377 }
1378
1379 static int delta_pos_compare(const void *_a, const void *_b)
1380 {
1381 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1382 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1383 return a->obj_no - b->obj_no;
1384 }
1385
1386 static void fix_unresolved_deltas(struct hashfile *f)
1387 {
1388 struct ref_delta_entry **sorted_by_pos;
1389 int i;
1390
1391 /*
1392 * Since many unresolved deltas may well be themselves base objects
1393 * for more unresolved deltas, we really want to include the
1394 * smallest number of base objects that would cover as much delta
1395 * as possible by picking the
1396 * trunc deltas first, allowing for other deltas to resolve without
1397 * additional base objects. Since most base objects are to be found
1398 * before deltas depending on them, a good heuristic is to start
1399 * resolving deltas in the same order as their position in the pack.
1400 */
1401 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1402 for (i = 0; i < nr_ref_deltas; i++)
1403 sorted_by_pos[i] = &ref_deltas[i];
1404 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1405
1406 if (repo_has_promisor_remote(the_repository)) {
1407 /*
1408 * Prefetch the delta bases.
1409 */
1410 struct oid_array to_fetch = OID_ARRAY_INIT;
1411 for (i = 0; i < nr_ref_deltas; i++) {
1412 struct ref_delta_entry *d = sorted_by_pos[i];
1413 if (!oid_object_info_extended(the_repository, &d->oid,
1414 NULL,
1415 OBJECT_INFO_FOR_PREFETCH))
1416 continue;
1417 oid_array_append(&to_fetch, &d->oid);
1418 }
1419 promisor_remote_get_direct(the_repository,
1420 to_fetch.oid, to_fetch.nr);
1421 oid_array_clear(&to_fetch);
1422 }
1423
1424 for (i = 0; i < nr_ref_deltas; i++) {
1425 struct ref_delta_entry *d = sorted_by_pos[i];
1426 enum object_type type;
1427 void *data;
1428 unsigned long size;
1429
1430 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1431 continue;
1432 data = repo_read_object_file(the_repository, &d->oid, &type,
1433 &size);
1434 if (!data)
1435 continue;
1436
1437 if (check_object_signature(the_repository, &d->oid, data, size,
1438 type) < 0)
1439 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1440
1441 /*
1442 * Add this as an object to the objects array and call
1443 * threaded_second_pass() (which will pick up the added
1444 * object).
1445 */
1446 append_obj_to_pack(f, d->oid.hash, data, size, type);
1447 free(data);
1448 threaded_second_pass(NULL);
1449
1450 display_progress(progress, nr_resolved_deltas);
1451 }
1452 free(sorted_by_pos);
1453 }
1454
1455 static const char *derive_filename(const char *pack_name, const char *strip,
1456 const char *suffix, struct strbuf *buf)
1457 {
1458 size_t len;
1459 if (!strip_suffix(pack_name, strip, &len) || !len ||
1460 pack_name[len - 1] != '.')
1461 die(_("packfile name '%s' does not end with '.%s'"),
1462 pack_name, strip);
1463 strbuf_add(buf, pack_name, len);
1464 strbuf_addstr(buf, suffix);
1465 return buf->buf;
1466 }
1467
1468 static void write_special_file(const char *suffix, const char *msg,
1469 const char *pack_name, const unsigned char *hash,
1470 const char **report)
1471 {
1472 struct strbuf name_buf = STRBUF_INIT;
1473 const char *filename;
1474 int fd;
1475 int msg_len = strlen(msg);
1476
1477 if (pack_name)
1478 filename = derive_filename(pack_name, "pack", suffix, &name_buf);
1479 else
1480 filename = odb_pack_name(&name_buf, hash, suffix);
1481
1482 fd = odb_pack_keep(filename);
1483 if (fd < 0) {
1484 if (errno != EEXIST)
1485 die_errno(_("cannot write %s file '%s'"),
1486 suffix, filename);
1487 } else {
1488 if (msg_len > 0) {
1489 write_or_die(fd, msg, msg_len);
1490 write_or_die(fd, "\n", 1);
1491 }
1492 if (close(fd) != 0)
1493 die_errno(_("cannot close written %s file '%s'"),
1494 suffix, filename);
1495 if (report)
1496 *report = suffix;
1497 }
1498 strbuf_release(&name_buf);
1499 }
1500
1501 static void rename_tmp_packfile(const char **final_name,
1502 const char *curr_name,
1503 struct strbuf *name, unsigned char *hash,
1504 const char *ext, int make_read_only_if_same)
1505 {
1506 if (*final_name != curr_name) {
1507 if (!*final_name)
1508 *final_name = odb_pack_name(name, hash, ext);
1509 if (finalize_object_file(curr_name, *final_name))
1510 die(_("unable to rename temporary '*.%s' file to '%s'"),
1511 ext, *final_name);
1512 } else if (make_read_only_if_same) {
1513 chmod(*final_name, 0444);
1514 }
1515 }
1516
1517 static void final(const char *final_pack_name, const char *curr_pack_name,
1518 const char *final_index_name, const char *curr_index_name,
1519 const char *final_rev_index_name, const char *curr_rev_index_name,
1520 const char *keep_msg, const char *promisor_msg,
1521 unsigned char *hash)
1522 {
1523 const char *report = "pack";
1524 struct strbuf pack_name = STRBUF_INIT;
1525 struct strbuf index_name = STRBUF_INIT;
1526 struct strbuf rev_index_name = STRBUF_INIT;
1527 int err;
1528
1529 if (!from_stdin) {
1530 close(input_fd);
1531 } else {
1532 fsync_component_or_die(FSYNC_COMPONENT_PACK, output_fd, curr_pack_name);
1533 err = close(output_fd);
1534 if (err)
1535 die_errno(_("error while closing pack file"));
1536 }
1537
1538 if (keep_msg)
1539 write_special_file("keep", keep_msg, final_pack_name, hash,
1540 &report);
1541 if (promisor_msg)
1542 write_special_file("promisor", promisor_msg, final_pack_name,
1543 hash, NULL);
1544
1545 rename_tmp_packfile(&final_pack_name, curr_pack_name, &pack_name,
1546 hash, "pack", from_stdin);
1547 if (curr_rev_index_name)
1548 rename_tmp_packfile(&final_rev_index_name, curr_rev_index_name,
1549 &rev_index_name, hash, "rev", 1);
1550 rename_tmp_packfile(&final_index_name, curr_index_name, &index_name,
1551 hash, "idx", 1);
1552
1553 if (do_fsck_object) {
1554 struct packed_git *p;
1555 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1556 if (p)
1557 install_packed_git(the_repository, p);
1558 }
1559
1560 if (!from_stdin) {
1561 printf("%s\n", hash_to_hex(hash));
1562 } else {
1563 struct strbuf buf = STRBUF_INIT;
1564
1565 strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash));
1566 write_or_die(1, buf.buf, buf.len);
1567 strbuf_release(&buf);
1568
1569 /*
1570 * Let's just mimic git-unpack-objects here and write
1571 * the last part of the input buffer to stdout.
1572 */
1573 while (input_len) {
1574 err = xwrite(1, input_buffer + input_offset, input_len);
1575 if (err <= 0)
1576 break;
1577 input_len -= err;
1578 input_offset += err;
1579 }
1580 }
1581
1582 strbuf_release(&rev_index_name);
1583 strbuf_release(&index_name);
1584 strbuf_release(&pack_name);
1585 }
1586
1587 static int git_index_pack_config(const char *k, const char *v,
1588 const struct config_context *ctx, void *cb)
1589 {
1590 struct pack_idx_option *opts = cb;
1591
1592 if (!strcmp(k, "pack.indexversion")) {
1593 opts->version = git_config_int(k, v, ctx->kvi);
1594 if (opts->version > 2)
1595 die(_("bad pack.indexVersion=%"PRIu32), opts->version);
1596 return 0;
1597 }
1598 if (!strcmp(k, "pack.threads")) {
1599 nr_threads = git_config_int(k, v, ctx->kvi);
1600 if (nr_threads < 0)
1601 die(_("invalid number of threads specified (%d)"),
1602 nr_threads);
1603 if (!HAVE_THREADS && nr_threads != 1) {
1604 warning(_("no threads support, ignoring %s"), k);
1605 nr_threads = 1;
1606 }
1607 return 0;
1608 }
1609 if (!strcmp(k, "pack.writereverseindex")) {
1610 if (git_config_bool(k, v))
1611 opts->flags |= WRITE_REV;
1612 else
1613 opts->flags &= ~WRITE_REV;
1614 }
1615 return git_default_config(k, v, ctx, cb);
1616 }
1617
1618 static int cmp_uint32(const void *a_, const void *b_)
1619 {
1620 uint32_t a = *((uint32_t *)a_);
1621 uint32_t b = *((uint32_t *)b_);
1622
1623 return (a < b) ? -1 : (a != b);
1624 }
1625
1626 static void read_v2_anomalous_offsets(struct packed_git *p,
1627 struct pack_idx_option *opts)
1628 {
1629 const uint32_t *idx1, *idx2;
1630 uint32_t i;
1631
1632 /* The address of the 4-byte offset table */
1633 idx1 = (((const uint32_t *)((const uint8_t *)p->index_data + p->crc_offset))
1634 + (size_t)p->num_objects /* CRC32 table */
1635 );
1636
1637 /* The address of the 8-byte offset table */
1638 idx2 = idx1 + p->num_objects;
1639
1640 for (i = 0; i < p->num_objects; i++) {
1641 uint32_t off = ntohl(idx1[i]);
1642 if (!(off & 0x80000000))
1643 continue;
1644 off = off & 0x7fffffff;
1645 check_pack_index_ptr(p, &idx2[off * 2]);
1646 if (idx2[off * 2])
1647 continue;
1648 /*
1649 * The real offset is ntohl(idx2[off * 2]) in high 4
1650 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1651 * octets. But idx2[off * 2] is Zero!!!
1652 */
1653 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1654 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1655 }
1656
1657 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1658 }
1659
1660 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1661 {
1662 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1663
1664 if (!p)
1665 die(_("Cannot open existing pack file '%s'"), pack_name);
1666 if (open_pack_index(p))
1667 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1668
1669 /* Read the attributes from the existing idx file */
1670 opts->version = p->index_version;
1671
1672 if (opts->version == 2)
1673 read_v2_anomalous_offsets(p, opts);
1674
1675 /*
1676 * Get rid of the idx file as we do not need it anymore.
1677 * NEEDSWORK: extract this bit from free_pack_by_name() in
1678 * object-file.c, perhaps? It shouldn't matter very much as we
1679 * know we haven't installed this pack (hence we never have
1680 * read anything from it).
1681 */
1682 close_pack_index(p);
1683 free(p);
1684 }
1685
1686 static void show_pack_info(int stat_only)
1687 {
1688 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1689 unsigned long *chain_histogram = NULL;
1690
1691 if (deepest_delta)
1692 CALLOC_ARRAY(chain_histogram, deepest_delta);
1693
1694 for (i = 0; i < nr_objects; i++) {
1695 struct object_entry *obj = &objects[i];
1696
1697 if (is_delta_type(obj->type))
1698 chain_histogram[obj_stat[i].delta_depth - 1]++;
1699 if (stat_only)
1700 continue;
1701 printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX,
1702 oid_to_hex(&obj->idx.oid),
1703 type_name(obj->real_type), (uintmax_t)obj->size,
1704 (uintmax_t)(obj[1].idx.offset - obj->idx.offset),
1705 (uintmax_t)obj->idx.offset);
1706 if (is_delta_type(obj->type)) {
1707 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1708 printf(" %u %s", obj_stat[i].delta_depth,
1709 oid_to_hex(&bobj->idx.oid));
1710 }
1711 putchar('\n');
1712 }
1713
1714 if (baseobjects)
1715 printf_ln(Q_("non delta: %d object",
1716 "non delta: %d objects",
1717 baseobjects),
1718 baseobjects);
1719 for (i = 0; i < deepest_delta; i++) {
1720 if (!chain_histogram[i])
1721 continue;
1722 printf_ln(Q_("chain length = %d: %lu object",
1723 "chain length = %d: %lu objects",
1724 chain_histogram[i]),
1725 i + 1,
1726 chain_histogram[i]);
1727 }
1728 free(chain_histogram);
1729 }
1730
1731 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1732 {
1733 int i, fix_thin_pack = 0, verify = 0, stat_only = 0, rev_index;
1734 const char *curr_index;
1735 const char *curr_rev_index = NULL;
1736 const char *index_name = NULL, *pack_name = NULL, *rev_index_name = NULL;
1737 const char *keep_msg = NULL;
1738 const char *promisor_msg = NULL;
1739 struct strbuf index_name_buf = STRBUF_INIT;
1740 struct strbuf rev_index_name_buf = STRBUF_INIT;
1741 struct pack_idx_entry **idx_objects;
1742 struct pack_idx_option opts;
1743 unsigned char pack_hash[GIT_MAX_RAWSZ];
1744 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1745 int report_end_of_input = 0;
1746 int hash_algo = 0;
1747
1748 /*
1749 * index-pack never needs to fetch missing objects except when
1750 * REF_DELTA bases are missing (which are explicitly handled). It only
1751 * accesses the repo to do hash collision checks and to check which
1752 * REF_DELTA bases need to be fetched.
1753 */
1754 fetch_if_missing = 0;
1755
1756 if (argc == 2 && !strcmp(argv[1], "-h"))
1757 usage(index_pack_usage);
1758
1759 disable_replace_refs();
1760 fsck_options.walk = mark_link;
1761
1762 reset_pack_idx_option(&opts);
1763 opts.flags |= WRITE_REV;
1764 git_config(git_index_pack_config, &opts);
1765 if (prefix && chdir(prefix))
1766 die(_("Cannot come back to cwd"));
1767
1768 if (git_env_bool(GIT_TEST_NO_WRITE_REV_INDEX, 0))
1769 rev_index = 0;
1770 else
1771 rev_index = !!(opts.flags & (WRITE_REV_VERIFY | WRITE_REV));
1772
1773 for (i = 1; i < argc; i++) {
1774 const char *arg = argv[i];
1775
1776 if (*arg == '-') {
1777 if (!strcmp(arg, "--stdin")) {
1778 from_stdin = 1;
1779 } else if (!strcmp(arg, "--fix-thin")) {
1780 fix_thin_pack = 1;
1781 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1782 strict = 1;
1783 do_fsck_object = 1;
1784 fsck_set_msg_types(&fsck_options, arg);
1785 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1786 strict = 1;
1787 check_self_contained_and_connected = 1;
1788 } else if (!strcmp(arg, "--fsck-objects")) {
1789 do_fsck_object = 1;
1790 } else if (!strcmp(arg, "--verify")) {
1791 verify = 1;
1792 } else if (!strcmp(arg, "--verify-stat")) {
1793 verify = 1;
1794 show_stat = 1;
1795 } else if (!strcmp(arg, "--verify-stat-only")) {
1796 verify = 1;
1797 show_stat = 1;
1798 stat_only = 1;
1799 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1800 ; /* nothing to do */
1801 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1802 ; /* already parsed */
1803 } else if (starts_with(arg, "--threads=")) {
1804 char *end;
1805 nr_threads = strtoul(arg+10, &end, 0);
1806 if (!arg[10] || *end || nr_threads < 0)
1807 usage(index_pack_usage);
1808 if (!HAVE_THREADS && nr_threads != 1) {
1809 warning(_("no threads support, ignoring %s"), arg);
1810 nr_threads = 1;
1811 }
1812 } else if (starts_with(arg, "--pack_header=")) {
1813 struct pack_header *hdr;
1814 char *c;
1815
1816 hdr = (struct pack_header *)input_buffer;
1817 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1818 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1819 if (*c != ',')
1820 die(_("bad %s"), arg);
1821 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1822 if (*c)
1823 die(_("bad %s"), arg);
1824 input_len = sizeof(*hdr);
1825 } else if (!strcmp(arg, "-v")) {
1826 verbose = 1;
1827 } else if (!strcmp(arg, "--progress-title")) {
1828 if (progress_title || (i+1) >= argc)
1829 usage(index_pack_usage);
1830 progress_title = argv[++i];
1831 } else if (!strcmp(arg, "--show-resolving-progress")) {
1832 show_resolving_progress = 1;
1833 } else if (!strcmp(arg, "--report-end-of-input")) {
1834 report_end_of_input = 1;
1835 } else if (!strcmp(arg, "-o")) {
1836 if (index_name || (i+1) >= argc)
1837 usage(index_pack_usage);
1838 index_name = argv[++i];
1839 } else if (starts_with(arg, "--index-version=")) {
1840 char *c;
1841 opts.version = strtoul(arg + 16, &c, 10);
1842 if (opts.version > 2)
1843 die(_("bad %s"), arg);
1844 if (*c == ',')
1845 opts.off32_limit = strtoul(c+1, &c, 0);
1846 if (*c || opts.off32_limit & 0x80000000)
1847 die(_("bad %s"), arg);
1848 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1849 max_input_size = strtoumax(arg, NULL, 10);
1850 } else if (skip_prefix(arg, "--object-format=", &arg)) {
1851 hash_algo = hash_algo_by_name(arg);
1852 if (hash_algo == GIT_HASH_UNKNOWN)
1853 die(_("unknown hash algorithm '%s'"), arg);
1854 repo_set_hash_algo(the_repository, hash_algo);
1855 } else if (!strcmp(arg, "--rev-index")) {
1856 rev_index = 1;
1857 } else if (!strcmp(arg, "--no-rev-index")) {
1858 rev_index = 0;
1859 } else
1860 usage(index_pack_usage);
1861 continue;
1862 }
1863
1864 if (pack_name)
1865 usage(index_pack_usage);
1866 pack_name = arg;
1867 }
1868
1869 if (!pack_name && !from_stdin)
1870 usage(index_pack_usage);
1871 if (fix_thin_pack && !from_stdin)
1872 die(_("the option '%s' requires '%s'"), "--fix-thin", "--stdin");
1873 if (from_stdin && !startup_info->have_repository)
1874 die(_("--stdin requires a git repository"));
1875 if (from_stdin && hash_algo)
1876 die(_("options '%s' and '%s' cannot be used together"), "--object-format", "--stdin");
1877 if (!index_name && pack_name)
1878 index_name = derive_filename(pack_name, "pack", "idx", &index_name_buf);
1879
1880 opts.flags &= ~(WRITE_REV | WRITE_REV_VERIFY);
1881 if (rev_index) {
1882 opts.flags |= verify ? WRITE_REV_VERIFY : WRITE_REV;
1883 if (index_name)
1884 rev_index_name = derive_filename(index_name,
1885 "idx", "rev",
1886 &rev_index_name_buf);
1887 }
1888
1889 if (verify) {
1890 if (!index_name)
1891 die(_("--verify with no packfile name given"));
1892 read_idx_option(&opts, index_name);
1893 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1894 }
1895 if (strict)
1896 opts.flags |= WRITE_IDX_STRICT;
1897
1898 if (HAVE_THREADS && !nr_threads) {
1899 nr_threads = online_cpus();
1900 /*
1901 * Experiments show that going above 20 threads doesn't help,
1902 * no matter how many cores you have. Below that, we tend to
1903 * max at half the number of online_cpus(), presumably because
1904 * half of those are hyperthreads rather than full cores. We'll
1905 * never reduce the level below "3", though, to match a
1906 * historical value that nobody complained about.
1907 */
1908 if (nr_threads < 4)
1909 ; /* too few cores to consider capping */
1910 else if (nr_threads < 6)
1911 nr_threads = 3; /* historic cap */
1912 else if (nr_threads < 40)
1913 nr_threads /= 2;
1914 else
1915 nr_threads = 20; /* hard cap */
1916 }
1917
1918 curr_pack = open_pack_file(pack_name);
1919 parse_pack_header();
1920 CALLOC_ARRAY(objects, st_add(nr_objects, 1));
1921 if (show_stat)
1922 CALLOC_ARRAY(obj_stat, st_add(nr_objects, 1));
1923 CALLOC_ARRAY(ofs_deltas, nr_objects);
1924 parse_pack_objects(pack_hash);
1925 if (report_end_of_input)
1926 write_in_full(2, "\0", 1);
1927 resolve_deltas();
1928 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1929 free(ofs_deltas);
1930 free(ref_deltas);
1931 if (strict)
1932 foreign_nr = check_objects();
1933
1934 if (show_stat)
1935 show_pack_info(stat_only);
1936
1937 ALLOC_ARRAY(idx_objects, nr_objects);
1938 for (i = 0; i < nr_objects; i++)
1939 idx_objects[i] = &objects[i].idx;
1940 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1941 if (rev_index)
1942 curr_rev_index = write_rev_file(rev_index_name, idx_objects,
1943 nr_objects, pack_hash,
1944 opts.flags);
1945 free(idx_objects);
1946
1947 if (!verify)
1948 final(pack_name, curr_pack,
1949 index_name, curr_index,
1950 rev_index_name, curr_rev_index,
1951 keep_msg, promisor_msg,
1952 pack_hash);
1953 else
1954 close(input_fd);
1955
1956 if (do_fsck_object && fsck_finish(&fsck_options))
1957 die(_("fsck error in pack objects"));
1958
1959 free(opts.anomaly);
1960 free(objects);
1961 strbuf_release(&index_name_buf);
1962 strbuf_release(&rev_index_name_buf);
1963 if (!pack_name)
1964 free((void *) curr_pack);
1965 if (!index_name)
1966 free((void *) curr_index);
1967 if (!rev_index_name)
1968 free((void *) curr_rev_index);
1969
1970 /*
1971 * Let the caller know this pack is not self contained
1972 */
1973 if (check_self_contained_and_connected && foreign_nr)
1974 return 1;
1975
1976 return 0;
1977 }