]> git.ipfire.org Git - thirdparty/git.git/blob - builtin/index-pack.c
submodule-config.c: strengthen URL fsck check
[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 "tag.h"
12 #include "tree.h"
13 #include "progress.h"
14 #include "fsck.h"
15 #include "exec-cmd.h"
16 #include "strbuf.h"
17 #include "streaming.h"
18 #include "thread-utils.h"
19 #include "packfile.h"
20 #include "pack-revindex.h"
21 #include "object-file.h"
22 #include "object-store-ll.h"
23 #include "oid-array.h"
24 #include "replace-object.h"
25 #include "promisor-remote.h"
26 #include "setup.h"
27
28 static const char index_pack_usage[] =
29 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--[no-]rev-index] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
30
31 struct object_entry {
32 struct pack_idx_entry idx;
33 unsigned long size;
34 unsigned char hdr_size;
35 signed char type;
36 signed char real_type;
37 };
38
39 struct object_stat {
40 unsigned delta_depth;
41 int base_object_no;
42 };
43
44 struct base_data {
45 /* Initialized by make_base(). */
46 struct base_data *base;
47 struct object_entry *obj;
48 int ref_first, ref_last;
49 int ofs_first, ofs_last;
50 /*
51 * Threads should increment retain_data if they are about to call
52 * patch_delta() using this struct's data as a base, and decrement this
53 * when they are done. While retain_data is nonzero, this struct's data
54 * will not be freed even if the delta base cache limit is exceeded.
55 */
56 int retain_data;
57 /*
58 * The number of direct children that have not been fully processed
59 * (entered work_head, entered done_head, left done_head). When this
60 * number reaches zero, this struct base_data can be freed.
61 */
62 int children_remaining;
63
64 /* Not initialized by make_base(). */
65 struct list_head list;
66 void *data;
67 unsigned long size;
68 };
69
70 /*
71 * Stack of struct base_data that have unprocessed children.
72 * threaded_second_pass() uses this as a source of work (the other being the
73 * objects array).
74 *
75 * Guarded by work_mutex.
76 */
77 static LIST_HEAD(work_head);
78
79 /*
80 * Stack of struct base_data that have children, all of whom have been
81 * processed or are being processed, and at least one child is being processed.
82 * These struct base_data must be kept around until the last child is
83 * processed.
84 *
85 * Guarded by work_mutex.
86 */
87 static LIST_HEAD(done_head);
88
89 /*
90 * All threads share one delta base cache.
91 *
92 * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only
93 * in a thread.
94 */
95 static size_t base_cache_used;
96 static size_t base_cache_limit;
97
98 struct thread_local {
99 pthread_t thread;
100 int pack_fd;
101 };
102
103 /* Remember to update object flag allocation in object.h */
104 #define FLAG_LINK (1u<<20)
105 #define FLAG_CHECKED (1u<<21)
106
107 struct ofs_delta_entry {
108 off_t offset;
109 int obj_no;
110 };
111
112 struct ref_delta_entry {
113 struct object_id oid;
114 int obj_no;
115 };
116
117 static struct object_entry *objects;
118 static struct object_stat *obj_stat;
119 static struct ofs_delta_entry *ofs_deltas;
120 static struct ref_delta_entry *ref_deltas;
121 static struct thread_local nothread_data;
122 static int nr_objects;
123 static int nr_ofs_deltas;
124 static int nr_ref_deltas;
125 static int ref_deltas_alloc;
126 static int nr_resolved_deltas;
127 static int nr_threads;
128
129 static int from_stdin;
130 static int strict;
131 static int do_fsck_object;
132 static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES;
133 static int verbose;
134 static const char *progress_title;
135 static int show_resolving_progress;
136 static int show_stat;
137 static int check_self_contained_and_connected;
138
139 static struct progress *progress;
140
141 /* We always read in 4kB chunks. */
142 static unsigned char input_buffer[4096];
143 static unsigned int input_offset, input_len;
144 static off_t consumed_bytes;
145 static off_t max_input_size;
146 static unsigned deepest_delta;
147 static git_hash_ctx input_ctx;
148 static uint32_t input_crc32;
149 static int input_fd, output_fd;
150 static const char *curr_pack;
151
152 static struct thread_local *thread_data;
153 static int nr_dispatched;
154 static int threads_active;
155
156 static pthread_mutex_t read_mutex;
157 #define read_lock() lock_mutex(&read_mutex)
158 #define read_unlock() unlock_mutex(&read_mutex)
159
160 static pthread_mutex_t counter_mutex;
161 #define counter_lock() lock_mutex(&counter_mutex)
162 #define counter_unlock() unlock_mutex(&counter_mutex)
163
164 static pthread_mutex_t work_mutex;
165 #define work_lock() lock_mutex(&work_mutex)
166 #define work_unlock() unlock_mutex(&work_mutex)
167
168 static pthread_mutex_t deepest_delta_mutex;
169 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
170 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
171
172 static pthread_key_t key;
173
174 static inline void lock_mutex(pthread_mutex_t *mutex)
175 {
176 if (threads_active)
177 pthread_mutex_lock(mutex);
178 }
179
180 static inline void unlock_mutex(pthread_mutex_t *mutex)
181 {
182 if (threads_active)
183 pthread_mutex_unlock(mutex);
184 }
185
186 /*
187 * Mutex and conditional variable can't be statically-initialized on Windows.
188 */
189 static void init_thread(void)
190 {
191 int i;
192 init_recursive_mutex(&read_mutex);
193 pthread_mutex_init(&counter_mutex, NULL);
194 pthread_mutex_init(&work_mutex, NULL);
195 if (show_stat)
196 pthread_mutex_init(&deepest_delta_mutex, NULL);
197 pthread_key_create(&key, NULL);
198 CALLOC_ARRAY(thread_data, nr_threads);
199 for (i = 0; i < nr_threads; i++) {
200 thread_data[i].pack_fd = xopen(curr_pack, O_RDONLY);
201 }
202
203 threads_active = 1;
204 }
205
206 static void cleanup_thread(void)
207 {
208 int i;
209 if (!threads_active)
210 return;
211 threads_active = 0;
212 pthread_mutex_destroy(&read_mutex);
213 pthread_mutex_destroy(&counter_mutex);
214 pthread_mutex_destroy(&work_mutex);
215 if (show_stat)
216 pthread_mutex_destroy(&deepest_delta_mutex);
217 for (i = 0; i < nr_threads; i++)
218 close(thread_data[i].pack_fd);
219 pthread_key_delete(key);
220 free(thread_data);
221 }
222
223 static int mark_link(struct object *obj, enum object_type type,
224 void *data UNUSED,
225 struct fsck_options *options UNUSED)
226 {
227 if (!obj)
228 return -1;
229
230 if (type != OBJ_ANY && obj->type != type)
231 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
232
233 obj->flags |= FLAG_LINK;
234 return 0;
235 }
236
237 /* The content of each linked object must have been checked
238 or it must be already present in the object database */
239 static unsigned check_object(struct object *obj)
240 {
241 if (!obj)
242 return 0;
243
244 if (!(obj->flags & FLAG_LINK))
245 return 0;
246
247 if (!(obj->flags & FLAG_CHECKED)) {
248 unsigned long size;
249 int type = oid_object_info(the_repository, &obj->oid, &size);
250 if (type <= 0)
251 die(_("did not receive expected object %s"),
252 oid_to_hex(&obj->oid));
253 if (type != obj->type)
254 die(_("object %s: expected type %s, found %s"),
255 oid_to_hex(&obj->oid),
256 type_name(obj->type), type_name(type));
257 obj->flags |= FLAG_CHECKED;
258 return 1;
259 }
260
261 return 0;
262 }
263
264 static unsigned check_objects(void)
265 {
266 unsigned i, max, foreign_nr = 0;
267
268 max = get_max_object_index();
269
270 if (verbose)
271 progress = start_delayed_progress(_("Checking objects"), max);
272
273 for (i = 0; i < max; i++) {
274 foreign_nr += check_object(get_indexed_object(i));
275 display_progress(progress, i + 1);
276 }
277
278 stop_progress(&progress);
279 return foreign_nr;
280 }
281
282
283 /* Discard current buffer used content. */
284 static void flush(void)
285 {
286 if (input_offset) {
287 if (output_fd >= 0)
288 write_or_die(output_fd, input_buffer, input_offset);
289 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
290 memmove(input_buffer, input_buffer + input_offset, input_len);
291 input_offset = 0;
292 }
293 }
294
295 /*
296 * Make sure at least "min" bytes are available in the buffer, and
297 * return the pointer to the buffer.
298 */
299 static void *fill(int min)
300 {
301 if (min <= input_len)
302 return input_buffer + input_offset;
303 if (min > sizeof(input_buffer))
304 die(Q_("cannot fill %d byte",
305 "cannot fill %d bytes",
306 min),
307 min);
308 flush();
309 do {
310 ssize_t ret = xread(input_fd, input_buffer + input_len,
311 sizeof(input_buffer) - input_len);
312 if (ret <= 0) {
313 if (!ret)
314 die(_("early EOF"));
315 die_errno(_("read error on input"));
316 }
317 input_len += ret;
318 if (from_stdin)
319 display_throughput(progress, consumed_bytes + input_len);
320 } while (input_len < min);
321 return input_buffer;
322 }
323
324 static void use(int bytes)
325 {
326 if (bytes > input_len)
327 die(_("used more bytes than were available"));
328 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
329 input_len -= bytes;
330 input_offset += bytes;
331
332 /* make sure off_t is sufficiently large not to wrap */
333 if (signed_add_overflows(consumed_bytes, bytes))
334 die(_("pack too large for current definition of off_t"));
335 consumed_bytes += bytes;
336 if (max_input_size && consumed_bytes > max_input_size) {
337 struct strbuf size_limit = STRBUF_INIT;
338 strbuf_humanise_bytes(&size_limit, max_input_size);
339 die(_("pack exceeds maximum allowed size (%s)"),
340 size_limit.buf);
341 }
342 }
343
344 static const char *open_pack_file(const char *pack_name)
345 {
346 if (from_stdin) {
347 input_fd = 0;
348 if (!pack_name) {
349 struct strbuf tmp_file = STRBUF_INIT;
350 output_fd = odb_mkstemp(&tmp_file,
351 "pack/tmp_pack_XXXXXX");
352 pack_name = strbuf_detach(&tmp_file, NULL);
353 } else {
354 output_fd = xopen(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
355 }
356 nothread_data.pack_fd = output_fd;
357 } else {
358 input_fd = xopen(pack_name, O_RDONLY);
359 output_fd = -1;
360 nothread_data.pack_fd = input_fd;
361 }
362 the_hash_algo->init_fn(&input_ctx);
363 return pack_name;
364 }
365
366 static void parse_pack_header(void)
367 {
368 struct pack_header *hdr = fill(sizeof(struct pack_header));
369
370 /* Header consistency check */
371 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
372 die(_("pack signature mismatch"));
373 if (!pack_version_ok(hdr->hdr_version))
374 die(_("pack version %"PRIu32" unsupported"),
375 ntohl(hdr->hdr_version));
376
377 nr_objects = ntohl(hdr->hdr_entries);
378 use(sizeof(struct pack_header));
379 }
380
381 __attribute__((format (printf, 2, 3)))
382 static NORETURN void bad_object(off_t offset, const char *format, ...)
383 {
384 va_list params;
385 char buf[1024];
386
387 va_start(params, format);
388 vsnprintf(buf, sizeof(buf), format, params);
389 va_end(params);
390 die(_("pack has bad object at offset %"PRIuMAX": %s"),
391 (uintmax_t)offset, buf);
392 }
393
394 static inline struct thread_local *get_thread_data(void)
395 {
396 if (HAVE_THREADS) {
397 if (threads_active)
398 return pthread_getspecific(key);
399 assert(!threads_active &&
400 "This should only be reached when all threads are gone");
401 }
402 return &nothread_data;
403 }
404
405 static void set_thread_data(struct thread_local *data)
406 {
407 if (threads_active)
408 pthread_setspecific(key, data);
409 }
410
411 static void free_base_data(struct base_data *c)
412 {
413 if (c->data) {
414 FREE_AND_NULL(c->data);
415 base_cache_used -= c->size;
416 }
417 }
418
419 static void prune_base_data(struct base_data *retain)
420 {
421 struct list_head *pos;
422
423 if (base_cache_used <= base_cache_limit)
424 return;
425
426 list_for_each_prev(pos, &done_head) {
427 struct base_data *b = list_entry(pos, struct base_data, list);
428 if (b->retain_data || b == retain)
429 continue;
430 if (b->data) {
431 free_base_data(b);
432 if (base_cache_used <= base_cache_limit)
433 return;
434 }
435 }
436
437 list_for_each_prev(pos, &work_head) {
438 struct base_data *b = list_entry(pos, struct base_data, list);
439 if (b->retain_data || b == retain)
440 continue;
441 if (b->data) {
442 free_base_data(b);
443 if (base_cache_used <= base_cache_limit)
444 return;
445 }
446 }
447 }
448
449 static int is_delta_type(enum object_type type)
450 {
451 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
452 }
453
454 static void *unpack_entry_data(off_t offset, unsigned long size,
455 enum object_type type, struct object_id *oid)
456 {
457 static char fixed_buf[8192];
458 int status;
459 git_zstream stream;
460 void *buf;
461 git_hash_ctx c;
462 char hdr[32];
463 int hdrlen;
464
465 if (!is_delta_type(type)) {
466 hdrlen = format_object_header(hdr, sizeof(hdr), type, size);
467 the_hash_algo->init_fn(&c);
468 the_hash_algo->update_fn(&c, hdr, hdrlen);
469 } else
470 oid = NULL;
471 if (type == OBJ_BLOB && size > big_file_threshold)
472 buf = fixed_buf;
473 else
474 buf = xmallocz(size);
475
476 memset(&stream, 0, sizeof(stream));
477 git_inflate_init(&stream);
478 stream.next_out = buf;
479 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
480
481 do {
482 unsigned char *last_out = stream.next_out;
483 stream.next_in = fill(1);
484 stream.avail_in = input_len;
485 status = git_inflate(&stream, 0);
486 use(input_len - stream.avail_in);
487 if (oid)
488 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
489 if (buf == fixed_buf) {
490 stream.next_out = buf;
491 stream.avail_out = sizeof(fixed_buf);
492 }
493 } while (status == Z_OK);
494 if (stream.total_out != size || status != Z_STREAM_END)
495 bad_object(offset, _("inflate returned %d"), status);
496 git_inflate_end(&stream);
497 if (oid)
498 the_hash_algo->final_oid_fn(oid, &c);
499 return buf == fixed_buf ? NULL : buf;
500 }
501
502 static void *unpack_raw_entry(struct object_entry *obj,
503 off_t *ofs_offset,
504 struct object_id *ref_oid,
505 struct object_id *oid)
506 {
507 unsigned char *p;
508 unsigned long size, c;
509 off_t base_offset;
510 unsigned shift;
511 void *data;
512
513 obj->idx.offset = consumed_bytes;
514 input_crc32 = crc32(0, NULL, 0);
515
516 p = fill(1);
517 c = *p;
518 use(1);
519 obj->type = (c >> 4) & 7;
520 size = (c & 15);
521 shift = 4;
522 while (c & 0x80) {
523 p = fill(1);
524 c = *p;
525 use(1);
526 size += (c & 0x7f) << shift;
527 shift += 7;
528 }
529 obj->size = size;
530
531 switch (obj->type) {
532 case OBJ_REF_DELTA:
533 oidread(ref_oid, fill(the_hash_algo->rawsz));
534 use(the_hash_algo->rawsz);
535 break;
536 case OBJ_OFS_DELTA:
537 p = fill(1);
538 c = *p;
539 use(1);
540 base_offset = c & 127;
541 while (c & 128) {
542 base_offset += 1;
543 if (!base_offset || MSB(base_offset, 7))
544 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
545 p = fill(1);
546 c = *p;
547 use(1);
548 base_offset = (base_offset << 7) + (c & 127);
549 }
550 *ofs_offset = obj->idx.offset - base_offset;
551 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
552 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
553 break;
554 case OBJ_COMMIT:
555 case OBJ_TREE:
556 case OBJ_BLOB:
557 case OBJ_TAG:
558 break;
559 default:
560 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
561 }
562 obj->hdr_size = consumed_bytes - obj->idx.offset;
563
564 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
565 obj->idx.crc32 = input_crc32;
566 return data;
567 }
568
569 static void *unpack_data(struct object_entry *obj,
570 int (*consume)(const unsigned char *, unsigned long, void *),
571 void *cb_data)
572 {
573 off_t from = obj[0].idx.offset + obj[0].hdr_size;
574 off_t len = obj[1].idx.offset - from;
575 unsigned char *data, *inbuf;
576 git_zstream stream;
577 int status;
578
579 data = xmallocz(consume ? 64*1024 : obj->size);
580 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
581
582 memset(&stream, 0, sizeof(stream));
583 git_inflate_init(&stream);
584 stream.next_out = data;
585 stream.avail_out = consume ? 64*1024 : obj->size;
586
587 do {
588 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
589 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
590 if (n < 0)
591 die_errno(_("cannot pread pack file"));
592 if (!n)
593 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
594 "premature end of pack file, %"PRIuMAX" bytes missing",
595 len),
596 (uintmax_t)len);
597 from += n;
598 len -= n;
599 stream.next_in = inbuf;
600 stream.avail_in = n;
601 if (!consume)
602 status = git_inflate(&stream, 0);
603 else {
604 do {
605 status = git_inflate(&stream, 0);
606 if (consume(data, stream.next_out - data, cb_data)) {
607 free(inbuf);
608 free(data);
609 return NULL;
610 }
611 stream.next_out = data;
612 stream.avail_out = 64*1024;
613 } while (status == Z_OK && stream.avail_in);
614 }
615 } while (len && status == Z_OK && !stream.avail_in);
616
617 /* This has been inflated OK when first encountered, so... */
618 if (status != Z_STREAM_END || stream.total_out != obj->size)
619 die(_("serious inflate inconsistency"));
620
621 git_inflate_end(&stream);
622 free(inbuf);
623 if (consume) {
624 FREE_AND_NULL(data);
625 }
626 return data;
627 }
628
629 static void *get_data_from_pack(struct object_entry *obj)
630 {
631 return unpack_data(obj, NULL, NULL);
632 }
633
634 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
635 enum object_type type1,
636 enum object_type type2)
637 {
638 int cmp = type1 - type2;
639 if (cmp)
640 return cmp;
641 return offset1 < offset2 ? -1 :
642 offset1 > offset2 ? 1 :
643 0;
644 }
645
646 static int find_ofs_delta(const off_t offset)
647 {
648 int first = 0, last = nr_ofs_deltas;
649
650 while (first < last) {
651 int next = first + (last - first) / 2;
652 struct ofs_delta_entry *delta = &ofs_deltas[next];
653 int cmp;
654
655 cmp = compare_ofs_delta_bases(offset, delta->offset,
656 OBJ_OFS_DELTA,
657 objects[delta->obj_no].type);
658 if (!cmp)
659 return next;
660 if (cmp < 0) {
661 last = next;
662 continue;
663 }
664 first = next+1;
665 }
666 return -first-1;
667 }
668
669 static void find_ofs_delta_children(off_t offset,
670 int *first_index, int *last_index)
671 {
672 int first = find_ofs_delta(offset);
673 int last = first;
674 int end = nr_ofs_deltas - 1;
675
676 if (first < 0) {
677 *first_index = 0;
678 *last_index = -1;
679 return;
680 }
681 while (first > 0 && ofs_deltas[first - 1].offset == offset)
682 --first;
683 while (last < end && ofs_deltas[last + 1].offset == offset)
684 ++last;
685 *first_index = first;
686 *last_index = last;
687 }
688
689 static int compare_ref_delta_bases(const struct object_id *oid1,
690 const struct object_id *oid2,
691 enum object_type type1,
692 enum object_type type2)
693 {
694 int cmp = type1 - type2;
695 if (cmp)
696 return cmp;
697 return oidcmp(oid1, oid2);
698 }
699
700 static int find_ref_delta(const struct object_id *oid)
701 {
702 int first = 0, last = nr_ref_deltas;
703
704 while (first < last) {
705 int next = first + (last - first) / 2;
706 struct ref_delta_entry *delta = &ref_deltas[next];
707 int cmp;
708
709 cmp = compare_ref_delta_bases(oid, &delta->oid,
710 OBJ_REF_DELTA,
711 objects[delta->obj_no].type);
712 if (!cmp)
713 return next;
714 if (cmp < 0) {
715 last = next;
716 continue;
717 }
718 first = next+1;
719 }
720 return -first-1;
721 }
722
723 static void find_ref_delta_children(const struct object_id *oid,
724 int *first_index, int *last_index)
725 {
726 int first = find_ref_delta(oid);
727 int last = first;
728 int end = nr_ref_deltas - 1;
729
730 if (first < 0) {
731 *first_index = 0;
732 *last_index = -1;
733 return;
734 }
735 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
736 --first;
737 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
738 ++last;
739 *first_index = first;
740 *last_index = last;
741 }
742
743 struct compare_data {
744 struct object_entry *entry;
745 struct git_istream *st;
746 unsigned char *buf;
747 unsigned long buf_size;
748 };
749
750 static int compare_objects(const unsigned char *buf, unsigned long size,
751 void *cb_data)
752 {
753 struct compare_data *data = cb_data;
754
755 if (data->buf_size < size) {
756 free(data->buf);
757 data->buf = xmalloc(size);
758 data->buf_size = size;
759 }
760
761 while (size) {
762 ssize_t len = read_istream(data->st, data->buf, size);
763 if (len == 0)
764 die(_("SHA1 COLLISION FOUND WITH %s !"),
765 oid_to_hex(&data->entry->idx.oid));
766 if (len < 0)
767 die(_("unable to read %s"),
768 oid_to_hex(&data->entry->idx.oid));
769 if (memcmp(buf, data->buf, len))
770 die(_("SHA1 COLLISION FOUND WITH %s !"),
771 oid_to_hex(&data->entry->idx.oid));
772 size -= len;
773 buf += len;
774 }
775 return 0;
776 }
777
778 static int check_collison(struct object_entry *entry)
779 {
780 struct compare_data data;
781 enum object_type type;
782 unsigned long size;
783
784 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
785 return -1;
786
787 memset(&data, 0, sizeof(data));
788 data.entry = entry;
789 data.st = open_istream(the_repository, &entry->idx.oid, &type, &size,
790 NULL);
791 if (!data.st)
792 return -1;
793 if (size != entry->size || type != entry->type)
794 die(_("SHA1 COLLISION FOUND WITH %s !"),
795 oid_to_hex(&entry->idx.oid));
796 unpack_data(entry, compare_objects, &data);
797 close_istream(data.st);
798 free(data.buf);
799 return 0;
800 }
801
802 static void sha1_object(const void *data, struct object_entry *obj_entry,
803 unsigned long size, enum object_type type,
804 const struct object_id *oid)
805 {
806 void *new_data = NULL;
807 int collision_test_needed = 0;
808
809 assert(data || obj_entry);
810
811 if (startup_info->have_repository) {
812 read_lock();
813 collision_test_needed =
814 repo_has_object_file_with_flags(the_repository, oid,
815 OBJECT_INFO_QUICK);
816 read_unlock();
817 }
818
819 if (collision_test_needed && !data) {
820 read_lock();
821 if (!check_collison(obj_entry))
822 collision_test_needed = 0;
823 read_unlock();
824 }
825 if (collision_test_needed) {
826 void *has_data;
827 enum object_type has_type;
828 unsigned long has_size;
829 read_lock();
830 has_type = oid_object_info(the_repository, oid, &has_size);
831 if (has_type < 0)
832 die(_("cannot read existing object info %s"), oid_to_hex(oid));
833 if (has_type != type || has_size != size)
834 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
835 has_data = repo_read_object_file(the_repository, oid,
836 &has_type, &has_size);
837 read_unlock();
838 if (!data)
839 data = new_data = get_data_from_pack(obj_entry);
840 if (!has_data)
841 die(_("cannot read existing object %s"), oid_to_hex(oid));
842 if (size != has_size || type != has_type ||
843 memcmp(data, has_data, size) != 0)
844 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
845 free(has_data);
846 }
847
848 if (strict || do_fsck_object) {
849 read_lock();
850 if (type == OBJ_BLOB) {
851 struct blob *blob = lookup_blob(the_repository, oid);
852 if (blob)
853 blob->object.flags |= FLAG_CHECKED;
854 else
855 die(_("invalid blob object %s"), oid_to_hex(oid));
856 if (do_fsck_object &&
857 fsck_object(&blob->object, (void *)data, size, &fsck_options))
858 die(_("fsck error in packed object"));
859 } else {
860 struct object *obj;
861 int eaten;
862 void *buf = (void *) data;
863
864 assert(data && "data can only be NULL for large _blobs_");
865
866 /*
867 * we do not need to free the memory here, as the
868 * buf is deleted by the caller.
869 */
870 obj = parse_object_buffer(the_repository, oid, type,
871 size, buf,
872 &eaten);
873 if (!obj)
874 die(_("invalid %s"), type_name(type));
875 if (do_fsck_object &&
876 fsck_object(obj, buf, size, &fsck_options))
877 die(_("fsck error in packed object"));
878 if (strict && fsck_walk(obj, NULL, &fsck_options))
879 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
880
881 if (obj->type == OBJ_TREE) {
882 struct tree *item = (struct tree *) obj;
883 item->buffer = NULL;
884 obj->parsed = 0;
885 }
886 if (obj->type == OBJ_COMMIT) {
887 struct commit *commit = (struct commit *) obj;
888 if (detach_commit_buffer(commit, NULL) != data)
889 BUG("parse_object_buffer transmogrified our buffer");
890 }
891 obj->flags |= FLAG_CHECKED;
892 }
893 read_unlock();
894 }
895
896 free(new_data);
897 }
898
899 /*
900 * Ensure that this node has been reconstructed and return its contents.
901 *
902 * In the typical and best case, this node would already be reconstructed
903 * (through the invocation to resolve_delta() in threaded_second_pass()) and it
904 * would not be pruned. However, if pruning of this node was necessary due to
905 * reaching delta_base_cache_limit, this function will find the closest
906 * ancestor with reconstructed data that has not been pruned (or if there is
907 * none, the ultimate base object), and reconstruct each node in the delta
908 * chain in order to generate the reconstructed data for this node.
909 */
910 static void *get_base_data(struct base_data *c)
911 {
912 if (!c->data) {
913 struct object_entry *obj = c->obj;
914 struct base_data **delta = NULL;
915 int delta_nr = 0, delta_alloc = 0;
916
917 while (is_delta_type(c->obj->type) && !c->data) {
918 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
919 delta[delta_nr++] = c;
920 c = c->base;
921 }
922 if (!delta_nr) {
923 c->data = get_data_from_pack(obj);
924 c->size = obj->size;
925 base_cache_used += c->size;
926 prune_base_data(c);
927 }
928 for (; delta_nr > 0; delta_nr--) {
929 void *base, *raw;
930 c = delta[delta_nr - 1];
931 obj = c->obj;
932 base = get_base_data(c->base);
933 raw = get_data_from_pack(obj);
934 c->data = patch_delta(
935 base, c->base->size,
936 raw, obj->size,
937 &c->size);
938 free(raw);
939 if (!c->data)
940 bad_object(obj->idx.offset, _("failed to apply delta"));
941 base_cache_used += c->size;
942 prune_base_data(c);
943 }
944 free(delta);
945 }
946 return c->data;
947 }
948
949 static struct base_data *make_base(struct object_entry *obj,
950 struct base_data *parent)
951 {
952 struct base_data *base = xcalloc(1, sizeof(struct base_data));
953 base->base = parent;
954 base->obj = obj;
955 find_ref_delta_children(&obj->idx.oid,
956 &base->ref_first, &base->ref_last);
957 find_ofs_delta_children(obj->idx.offset,
958 &base->ofs_first, &base->ofs_last);
959 base->children_remaining = base->ref_last - base->ref_first +
960 base->ofs_last - base->ofs_first + 2;
961 return base;
962 }
963
964 static struct base_data *resolve_delta(struct object_entry *delta_obj,
965 struct base_data *base)
966 {
967 void *delta_data, *result_data;
968 struct base_data *result;
969 unsigned long result_size;
970
971 if (show_stat) {
972 int i = delta_obj - objects;
973 int j = base->obj - objects;
974 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
975 deepest_delta_lock();
976 if (deepest_delta < obj_stat[i].delta_depth)
977 deepest_delta = obj_stat[i].delta_depth;
978 deepest_delta_unlock();
979 obj_stat[i].base_object_no = j;
980 }
981 delta_data = get_data_from_pack(delta_obj);
982 assert(base->data);
983 result_data = patch_delta(base->data, base->size,
984 delta_data, delta_obj->size, &result_size);
985 free(delta_data);
986 if (!result_data)
987 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
988 hash_object_file(the_hash_algo, result_data, result_size,
989 delta_obj->real_type, &delta_obj->idx.oid);
990 sha1_object(result_data, NULL, result_size, delta_obj->real_type,
991 &delta_obj->idx.oid);
992
993 result = make_base(delta_obj, base);
994 result->data = result_data;
995 result->size = result_size;
996
997 counter_lock();
998 nr_resolved_deltas++;
999 counter_unlock();
1000
1001 return result;
1002 }
1003
1004 static int compare_ofs_delta_entry(const void *a, const void *b)
1005 {
1006 const struct ofs_delta_entry *delta_a = a;
1007 const struct ofs_delta_entry *delta_b = b;
1008
1009 return delta_a->offset < delta_b->offset ? -1 :
1010 delta_a->offset > delta_b->offset ? 1 :
1011 0;
1012 }
1013
1014 static int compare_ref_delta_entry(const void *a, const void *b)
1015 {
1016 const struct ref_delta_entry *delta_a = a;
1017 const struct ref_delta_entry *delta_b = b;
1018
1019 return oidcmp(&delta_a->oid, &delta_b->oid);
1020 }
1021
1022 static void *threaded_second_pass(void *data)
1023 {
1024 if (data)
1025 set_thread_data(data);
1026 for (;;) {
1027 struct base_data *parent = NULL;
1028 struct object_entry *child_obj;
1029 struct base_data *child;
1030
1031 counter_lock();
1032 display_progress(progress, nr_resolved_deltas);
1033 counter_unlock();
1034
1035 work_lock();
1036 if (list_empty(&work_head)) {
1037 /*
1038 * Take an object from the object array.
1039 */
1040 while (nr_dispatched < nr_objects &&
1041 is_delta_type(objects[nr_dispatched].type))
1042 nr_dispatched++;
1043 if (nr_dispatched >= nr_objects) {
1044 work_unlock();
1045 break;
1046 }
1047 child_obj = &objects[nr_dispatched++];
1048 } else {
1049 /*
1050 * Peek at the top of the stack, and take a child from
1051 * it.
1052 */
1053 parent = list_first_entry(&work_head, struct base_data,
1054 list);
1055
1056 if (parent->ref_first <= parent->ref_last) {
1057 int offset = ref_deltas[parent->ref_first++].obj_no;
1058 child_obj = objects + offset;
1059 if (child_obj->real_type != OBJ_REF_DELTA)
1060 die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)",
1061 (uintmax_t) child_obj->idx.offset,
1062 oid_to_hex(&parent->obj->idx.oid));
1063 child_obj->real_type = parent->obj->real_type;
1064 } else {
1065 child_obj = objects +
1066 ofs_deltas[parent->ofs_first++].obj_no;
1067 assert(child_obj->real_type == OBJ_OFS_DELTA);
1068 child_obj->real_type = parent->obj->real_type;
1069 }
1070
1071 if (parent->ref_first > parent->ref_last &&
1072 parent->ofs_first > parent->ofs_last) {
1073 /*
1074 * This parent has run out of children, so move
1075 * it to done_head.
1076 */
1077 list_del(&parent->list);
1078 list_add(&parent->list, &done_head);
1079 }
1080
1081 /*
1082 * Ensure that the parent has data, since we will need
1083 * it later.
1084 *
1085 * NEEDSWORK: If parent data needs to be reloaded, this
1086 * prolongs the time that the current thread spends in
1087 * the mutex. A mitigating factor is that parent data
1088 * needs to be reloaded only if the delta base cache
1089 * limit is exceeded, so in the typical case, this does
1090 * not happen.
1091 */
1092 get_base_data(parent);
1093 parent->retain_data++;
1094 }
1095 work_unlock();
1096
1097 if (parent) {
1098 child = resolve_delta(child_obj, parent);
1099 if (!child->children_remaining)
1100 FREE_AND_NULL(child->data);
1101 } else {
1102 child = make_base(child_obj, NULL);
1103 if (child->children_remaining) {
1104 /*
1105 * Since this child has its own delta children,
1106 * we will need this data in the future.
1107 * Inflate now so that future iterations will
1108 * have access to this object's data while
1109 * outside the work mutex.
1110 */
1111 child->data = get_data_from_pack(child_obj);
1112 child->size = child_obj->size;
1113 }
1114 }
1115
1116 work_lock();
1117 if (parent)
1118 parent->retain_data--;
1119 if (child->data) {
1120 /*
1121 * This child has its own children, so add it to
1122 * work_head.
1123 */
1124 list_add(&child->list, &work_head);
1125 base_cache_used += child->size;
1126 prune_base_data(NULL);
1127 free_base_data(child);
1128 } else {
1129 /*
1130 * This child does not have its own children. It may be
1131 * the last descendant of its ancestors; free those
1132 * that we can.
1133 */
1134 struct base_data *p = parent;
1135
1136 while (p) {
1137 struct base_data *next_p;
1138
1139 p->children_remaining--;
1140 if (p->children_remaining)
1141 break;
1142
1143 next_p = p->base;
1144 free_base_data(p);
1145 list_del(&p->list);
1146 free(p);
1147
1148 p = next_p;
1149 }
1150 FREE_AND_NULL(child);
1151 }
1152 work_unlock();
1153 }
1154 return NULL;
1155 }
1156
1157 /*
1158 * First pass:
1159 * - find locations of all objects;
1160 * - calculate SHA1 of all non-delta objects;
1161 * - remember base (SHA1 or offset) for all deltas.
1162 */
1163 static void parse_pack_objects(unsigned char *hash)
1164 {
1165 int i, nr_delays = 0;
1166 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1167 struct object_id ref_delta_oid;
1168 struct stat st;
1169 git_hash_ctx tmp_ctx;
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->init_fn(&tmp_ctx);
1207 the_hash_algo->clone_fn(&tmp_ctx, &input_ctx);
1208 the_hash_algo->final_fn(hash, &tmp_ctx);
1209 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1210 die(_("pack is corrupted (SHA1 mismatch)"));
1211 use(the_hash_algo->rawsz);
1212
1213 /* If input_fd is a file, we should have reached its end now. */
1214 if (fstat(input_fd, &st))
1215 die_errno(_("cannot fstat packfile"));
1216 if (S_ISREG(st.st_mode) &&
1217 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1218 die(_("pack has junk at the end"));
1219
1220 for (i = 0; i < nr_objects; i++) {
1221 struct object_entry *obj = &objects[i];
1222 if (obj->real_type != OBJ_BAD)
1223 continue;
1224 obj->real_type = obj->type;
1225 sha1_object(NULL, obj, obj->size, obj->type,
1226 &obj->idx.oid);
1227 nr_delays--;
1228 }
1229 if (nr_delays)
1230 die(_("confusion beyond insanity in parse_pack_objects()"));
1231 }
1232
1233 /*
1234 * Second pass:
1235 * - for all non-delta objects, look if it is used as a base for
1236 * deltas;
1237 * - if used as a base, uncompress the object and apply all deltas,
1238 * recursively checking if the resulting object is used as a base
1239 * for some more deltas.
1240 */
1241 static void resolve_deltas(void)
1242 {
1243 int i;
1244
1245 if (!nr_ofs_deltas && !nr_ref_deltas)
1246 return;
1247
1248 /* Sort deltas by base SHA1/offset for fast searching */
1249 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1250 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1251
1252 if (verbose || show_resolving_progress)
1253 progress = start_progress(_("Resolving deltas"),
1254 nr_ref_deltas + nr_ofs_deltas);
1255
1256 nr_dispatched = 0;
1257 base_cache_limit = delta_base_cache_limit * nr_threads;
1258 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1259 init_thread();
1260 for (i = 0; i < nr_threads; i++) {
1261 int ret = pthread_create(&thread_data[i].thread, NULL,
1262 threaded_second_pass, thread_data + i);
1263 if (ret)
1264 die(_("unable to create thread: %s"),
1265 strerror(ret));
1266 }
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 }