1#include "builtin.h"
2#include "config.h"
3#include "delta.h"
4#include "pack.h"
5#include "csum-file.h"
6#include "blob.h"
7#include "commit.h"
8#include "tag.h"
9#include "tree.h"
10#include "progress.h"
11#include "fsck.h"
12#include "exec-cmd.h"
13#include "streaming.h"
14#include "thread-utils.h"
15#include "packfile.h"
16#include "object-store.h"
17static const char index_pack_usage[] =
19"git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
20struct object_entry {
22 struct pack_idx_entry idx;
23 unsigned long size;
24 unsigned char hdr_size;
25 signed char type;
26 signed char real_type;
27};
28struct object_stat {
30 unsigned delta_depth;
31 int base_object_no;
32};
33struct base_data {
35 struct base_data *base;
36 struct base_data *child;
37 struct object_entry *obj;
38 void *data;
39 unsigned long size;
40 int ref_first, ref_last;
41 int ofs_first, ofs_last;
42};
43struct thread_local {
45 pthread_t thread;
46 struct base_data *base_cache;
47 size_t base_cache_used;
48 int pack_fd;
49};
50/* Remember to update object flag allocation in object.h */
52#define FLAG_LINK (1u<<20)
53#define FLAG_CHECKED (1u<<21)
54struct ofs_delta_entry {
56 off_t offset;
57 int obj_no;
58};
59struct ref_delta_entry {
61 struct object_id oid;
62 int obj_no;
63};
64static struct object_entry *objects;
66static struct object_stat *obj_stat;
67static struct ofs_delta_entry *ofs_deltas;
68static struct ref_delta_entry *ref_deltas;
69static struct thread_local nothread_data;
70static int nr_objects;
71static int nr_ofs_deltas;
72static int nr_ref_deltas;
73static int ref_deltas_alloc;
74static int nr_resolved_deltas;
75static int nr_threads;
76static int from_stdin;
78static int strict;
79static int do_fsck_object;
80static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
81static int verbose;
82static int show_resolving_progress;
83static int show_stat;
84static int check_self_contained_and_connected;
85static struct progress *progress;
87/* We always read in 4kB chunks. */
89static unsigned char input_buffer[4096];
90static unsigned int input_offset, input_len;
91static off_t consumed_bytes;
92static off_t max_input_size;
93static unsigned deepest_delta;
94static git_hash_ctx input_ctx;
95static uint32_t input_crc32;
96static int input_fd, output_fd;
97static const char *curr_pack;
98static struct thread_local *thread_data;
100static int nr_dispatched;
101static int threads_active;
102static pthread_mutex_t read_mutex;
104#define read_lock() lock_mutex(&read_mutex)
105#define read_unlock() unlock_mutex(&read_mutex)
106static pthread_mutex_t counter_mutex;
108#define counter_lock() lock_mutex(&counter_mutex)
109#define counter_unlock() unlock_mutex(&counter_mutex)
110static pthread_mutex_t work_mutex;
112#define work_lock() lock_mutex(&work_mutex)
113#define work_unlock() unlock_mutex(&work_mutex)
114static pthread_mutex_t deepest_delta_mutex;
116#define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
117#define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
118static pthread_mutex_t type_cas_mutex;
120#define type_cas_lock() lock_mutex(&type_cas_mutex)
121#define type_cas_unlock() unlock_mutex(&type_cas_mutex)
122static pthread_key_t key;
124static inline void lock_mutex(pthread_mutex_t *mutex)
126{
127 if (threads_active)
128 pthread_mutex_lock(mutex);
129}
130static inline void unlock_mutex(pthread_mutex_t *mutex)
132{
133 if (threads_active)
134 pthread_mutex_unlock(mutex);
135}
136/*
138 * Mutex and conditional variable can't be statically-initialized on Windows.
139 */
140static void init_thread(void)
141{
142 int i;
143 init_recursive_mutex(&read_mutex);
144 pthread_mutex_init(&counter_mutex, NULL);
145 pthread_mutex_init(&work_mutex, NULL);
146 pthread_mutex_init(&type_cas_mutex, NULL);
147 if (show_stat)
148 pthread_mutex_init(&deepest_delta_mutex, NULL);
149 pthread_key_create(&key, NULL);
150 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
151 for (i = 0; i < nr_threads; i++) {
152 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
153 if (thread_data[i].pack_fd == -1)
154 die_errno(_("unable to open %s"), curr_pack);
155 }
156 threads_active = 1;
158}
159static void cleanup_thread(void)
161{
162 int i;
163 if (!threads_active)
164 return;
165 threads_active = 0;
166 pthread_mutex_destroy(&read_mutex);
167 pthread_mutex_destroy(&counter_mutex);
168 pthread_mutex_destroy(&work_mutex);
169 pthread_mutex_destroy(&type_cas_mutex);
170 if (show_stat)
171 pthread_mutex_destroy(&deepest_delta_mutex);
172 for (i = 0; i < nr_threads; i++)
173 close(thread_data[i].pack_fd);
174 pthread_key_delete(key);
175 free(thread_data);
176}
177static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
179{
180 if (!obj)
181 return -1;
182 if (type != OBJ_ANY && obj->type != type)
184 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
185 obj->flags |= FLAG_LINK;
187 return 0;
188}
189/* The content of each linked object must have been checked
191 or it must be already present in the object database */
192static unsigned check_object(struct object *obj)
193{
194 if (!obj)
195 return 0;
196 if (!(obj->flags & FLAG_LINK))
198 return 0;
199 if (!(obj->flags & FLAG_CHECKED)) {
201 unsigned long size;
202 int type = oid_object_info(the_repository, &obj->oid, &size);
203 if (type <= 0)
204 die(_("did not receive expected object %s"),
205 oid_to_hex(&obj->oid));
206 if (type != obj->type)
207 die(_("object %s: expected type %s, found %s"),
208 oid_to_hex(&obj->oid),
209 type_name(obj->type), type_name(type));
210 obj->flags |= FLAG_CHECKED;
211 return 1;
212 }
213 return 0;
215}
216static unsigned check_objects(void)
218{
219 unsigned i, max, foreign_nr = 0;
220 max = get_max_object_index();
222 for (i = 0; i < max; i++)
223 foreign_nr += check_object(get_indexed_object(i));
224 return foreign_nr;
225}
226/* Discard current buffer used content. */
229static void flush(void)
230{
231 if (input_offset) {
232 if (output_fd >= 0)
233 write_or_die(output_fd, input_buffer, input_offset);
234 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
235 memmove(input_buffer, input_buffer + input_offset, input_len);
236 input_offset = 0;
237 }
238}
239/*
241 * Make sure at least "min" bytes are available in the buffer, and
242 * return the pointer to the buffer.
243 */
244static void *fill(int min)
245{
246 if (min <= input_len)
247 return input_buffer + input_offset;
248 if (min > sizeof(input_buffer))
249 die(Q_("cannot fill %d byte",
250 "cannot fill %d bytes",
251 min),
252 min);
253 flush();
254 do {
255 ssize_t ret = xread(input_fd, input_buffer + input_len,
256 sizeof(input_buffer) - input_len);
257 if (ret <= 0) {
258 if (!ret)
259 die(_("early EOF"));
260 die_errno(_("read error on input"));
261 }
262 input_len += ret;
263 if (from_stdin)
264 display_throughput(progress, consumed_bytes + input_len);
265 } while (input_len < min);
266 return input_buffer;
267}
268static void use(int bytes)
270{
271 if (bytes > input_len)
272 die(_("used more bytes than were available"));
273 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
274 input_len -= bytes;
275 input_offset += bytes;
276 /* make sure off_t is sufficiently large not to wrap */
278 if (signed_add_overflows(consumed_bytes, bytes))
279 die(_("pack too large for current definition of off_t"));
280 consumed_bytes += bytes;
281 if (max_input_size && consumed_bytes > max_input_size)
282 die(_("pack exceeds maximum allowed size"));
283}
284static const char *open_pack_file(const char *pack_name)
286{
287 if (from_stdin) {
288 input_fd = 0;
289 if (!pack_name) {
290 struct strbuf tmp_file = STRBUF_INIT;
291 output_fd = odb_mkstemp(&tmp_file,
292 "pack/tmp_pack_XXXXXX");
293 pack_name = strbuf_detach(&tmp_file, NULL);
294 } else {
295 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
296 if (output_fd < 0)
297 die_errno(_("unable to create '%s'"), pack_name);
298 }
299 nothread_data.pack_fd = output_fd;
300 } else {
301 input_fd = open(pack_name, O_RDONLY);
302 if (input_fd < 0)
303 die_errno(_("cannot open packfile '%s'"), pack_name);
304 output_fd = -1;
305 nothread_data.pack_fd = input_fd;
306 }
307 the_hash_algo->init_fn(&input_ctx);
308 return pack_name;
309}
310static void parse_pack_header(void)
312{
313 struct pack_header *hdr = fill(sizeof(struct pack_header));
314 /* Header consistency check */
316 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
317 die(_("pack signature mismatch"));
318 if (!pack_version_ok(hdr->hdr_version))
319 die(_("pack version %"PRIu32" unsupported"),
320 ntohl(hdr->hdr_version));
321 nr_objects = ntohl(hdr->hdr_entries);
323 use(sizeof(struct pack_header));
324}
325static NORETURN void bad_object(off_t offset, const char *format,
327 ...) __attribute__((format (printf, 2, 3)));
328static NORETURN void bad_object(off_t offset, const char *format, ...)
330{
331 va_list params;
332 char buf[1024];
333 va_start(params, format);
335 vsnprintf(buf, sizeof(buf), format, params);
336 va_end(params);
337 die(_("pack has bad object at offset %"PRIuMAX": %s"),
338 (uintmax_t)offset, buf);
339}
340static inline struct thread_local *get_thread_data(void)
342{
343 if (HAVE_THREADS) {
344 if (threads_active)
345 return pthread_getspecific(key);
346 assert(!threads_active &&
347 "This should only be reached when all threads are gone");
348 }
349 return ¬hread_data;
350}
351static void set_thread_data(struct thread_local *data)
353{
354 if (threads_active)
355 pthread_setspecific(key, data);
356}
357static struct base_data *alloc_base_data(void)
359{
360 struct base_data *base = xcalloc(1, sizeof(struct base_data));
361 base->ref_last = -1;
362 base->ofs_last = -1;
363 return base;
364}
365static void free_base_data(struct base_data *c)
367{
368 if (c->data) {
369 FREE_AND_NULL(c->data);
370 get_thread_data()->base_cache_used -= c->size;
371 }
372}
373static void prune_base_data(struct base_data *retain)
375{
376 struct base_data *b;
377 struct thread_local *data = get_thread_data();
378 for (b = data->base_cache;
379 data->base_cache_used > delta_base_cache_limit && b;
380 b = b->child) {
381 if (b->data && b != retain)
382 free_base_data(b);
383 }
384}
385static void link_base_data(struct base_data *base, struct base_data *c)
387{
388 if (base)
389 base->child = c;
390 else
391 get_thread_data()->base_cache = c;
392 c->base = base;
394 c->child = NULL;
395 if (c->data)
396 get_thread_data()->base_cache_used += c->size;
397 prune_base_data(c);
398}
399static void unlink_base_data(struct base_data *c)
401{
402 struct base_data *base = c->base;
403 if (base)
404 base->child = NULL;
405 else
406 get_thread_data()->base_cache = NULL;
407 free_base_data(c);
408}
409static int is_delta_type(enum object_type type)
411{
412 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
413}
414static void *unpack_entry_data(off_t offset, unsigned long size,
416 enum object_type type, struct object_id *oid)
417{
418 static char fixed_buf[8192];
419 int status;
420 git_zstream stream;
421 void *buf;
422 git_hash_ctx c;
423 char hdr[32];
424 int hdrlen;
425 if (!is_delta_type(type)) {
427 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), size) + 1;
428 the_hash_algo->init_fn(&c);
429 the_hash_algo->update_fn(&c, hdr, hdrlen);
430 } else
431 oid = NULL;
432 if (type == OBJ_BLOB && size > big_file_threshold)
433 buf = fixed_buf;
434 else
435 buf = xmallocz(size);
436 memset(&stream, 0, sizeof(stream));
438 git_inflate_init(&stream);
439 stream.next_out = buf;
440 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
441 do {
443 unsigned char *last_out = stream.next_out;
444 stream.next_in = fill(1);
445 stream.avail_in = input_len;
446 status = git_inflate(&stream, 0);
447 use(input_len - stream.avail_in);
448 if (oid)
449 the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
450 if (buf == fixed_buf) {
451 stream.next_out = buf;
452 stream.avail_out = sizeof(fixed_buf);
453 }
454 } while (status == Z_OK);
455 if (stream.total_out != size || status != Z_STREAM_END)
456 bad_object(offset, _("inflate returned %d"), status);
457 git_inflate_end(&stream);
458 if (oid)
459 the_hash_algo->final_fn(oid->hash, &c);
460 return buf == fixed_buf ? NULL : buf;
461}
462static void *unpack_raw_entry(struct object_entry *obj,
464 off_t *ofs_offset,
465 struct object_id *ref_oid,
466 struct object_id *oid)
467{
468 unsigned char *p;
469 unsigned long size, c;
470 off_t base_offset;
471 unsigned shift;
472 void *data;
473 obj->idx.offset = consumed_bytes;
475 input_crc32 = crc32(0, NULL, 0);
476 p = fill(1);
478 c = *p;
479 use(1);
480 obj->type = (c >> 4) & 7;
481 size = (c & 15);
482 shift = 4;
483 while (c & 0x80) {
484 p = fill(1);
485 c = *p;
486 use(1);
487 size += (c & 0x7f) << shift;
488 shift += 7;
489 }
490 obj->size = size;
491 switch (obj->type) {
493 case OBJ_REF_DELTA:
494 hashcpy(ref_oid->hash, fill(the_hash_algo->rawsz));
495 use(the_hash_algo->rawsz);
496 break;
497 case OBJ_OFS_DELTA:
498 p = fill(1);
499 c = *p;
500 use(1);
501 base_offset = c & 127;
502 while (c & 128) {
503 base_offset += 1;
504 if (!base_offset || MSB(base_offset, 7))
505 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
506 p = fill(1);
507 c = *p;
508 use(1);
509 base_offset = (base_offset << 7) + (c & 127);
510 }
511 *ofs_offset = obj->idx.offset - base_offset;
512 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
513 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
514 break;
515 case OBJ_COMMIT:
516 case OBJ_TREE:
517 case OBJ_BLOB:
518 case OBJ_TAG:
519 break;
520 default:
521 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
522 }
523 obj->hdr_size = consumed_bytes - obj->idx.offset;
524 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
526 obj->idx.crc32 = input_crc32;
527 return data;
528}
529static void *unpack_data(struct object_entry *obj,
531 int (*consume)(const unsigned char *, unsigned long, void *),
532 void *cb_data)
533{
534 off_t from = obj[0].idx.offset + obj[0].hdr_size;
535 off_t len = obj[1].idx.offset - from;
536 unsigned char *data, *inbuf;
537 git_zstream stream;
538 int status;
539 data = xmallocz(consume ? 64*1024 : obj->size);
541 inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
542 memset(&stream, 0, sizeof(stream));
544 git_inflate_init(&stream);
545 stream.next_out = data;
546 stream.avail_out = consume ? 64*1024 : obj->size;
547 do {
549 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
550 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
551 if (n < 0)
552 die_errno(_("cannot pread pack file"));
553 if (!n)
554 die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
555 "premature end of pack file, %"PRIuMAX" bytes missing",
556 (unsigned int)len),
557 (uintmax_t)len);
558 from += n;
559 len -= n;
560 stream.next_in = inbuf;
561 stream.avail_in = n;
562 if (!consume)
563 status = git_inflate(&stream, 0);
564 else {
565 do {
566 status = git_inflate(&stream, 0);
567 if (consume(data, stream.next_out - data, cb_data)) {
568 free(inbuf);
569 free(data);
570 return NULL;
571 }
572 stream.next_out = data;
573 stream.avail_out = 64*1024;
574 } while (status == Z_OK && stream.avail_in);
575 }
576 } while (len && status == Z_OK && !stream.avail_in);
577 /* This has been inflated OK when first encountered, so... */
579 if (status != Z_STREAM_END || stream.total_out != obj->size)
580 die(_("serious inflate inconsistency"));
581 git_inflate_end(&stream);
583 free(inbuf);
584 if (consume) {
585 FREE_AND_NULL(data);
586 }
587 return data;
588}
589static void *get_data_from_pack(struct object_entry *obj)
591{
592 return unpack_data(obj, NULL, NULL);
593}
594static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
596 enum object_type type1,
597 enum object_type type2)
598{
599 int cmp = type1 - type2;
600 if (cmp)
601 return cmp;
602 return offset1 < offset2 ? -1 :
603 offset1 > offset2 ? 1 :
604 0;
605}
606static int find_ofs_delta(const off_t offset, enum object_type type)
608{
609 int first = 0, last = nr_ofs_deltas;
610 while (first < last) {
612 int next = first + (last - first) / 2;
613 struct ofs_delta_entry *delta = &ofs_deltas[next];
614 int cmp;
615 cmp = compare_ofs_delta_bases(offset, delta->offset,
617 type, objects[delta->obj_no].type);
618 if (!cmp)
619 return next;
620 if (cmp < 0) {
621 last = next;
622 continue;
623 }
624 first = next+1;
625 }
626 return -first-1;
627}
628static void find_ofs_delta_children(off_t offset,
630 int *first_index, int *last_index,
631 enum object_type type)
632{
633 int first = find_ofs_delta(offset, type);
634 int last = first;
635 int end = nr_ofs_deltas - 1;
636 if (first < 0) {
638 *first_index = 0;
639 *last_index = -1;
640 return;
641 }
642 while (first > 0 && ofs_deltas[first - 1].offset == offset)
643 --first;
644 while (last < end && ofs_deltas[last + 1].offset == offset)
645 ++last;
646 *first_index = first;
647 *last_index = last;
648}
649static int compare_ref_delta_bases(const struct object_id *oid1,
651 const struct object_id *oid2,
652 enum object_type type1,
653 enum object_type type2)
654{
655 int cmp = type1 - type2;
656 if (cmp)
657 return cmp;
658 return oidcmp(oid1, oid2);
659}
660static int find_ref_delta(const struct object_id *oid, enum object_type type)
662{
663 int first = 0, last = nr_ref_deltas;
664 while (first < last) {
666 int next = first + (last - first) / 2;
667 struct ref_delta_entry *delta = &ref_deltas[next];
668 int cmp;
669 cmp = compare_ref_delta_bases(oid, &delta->oid,
671 type, objects[delta->obj_no].type);
672 if (!cmp)
673 return next;
674 if (cmp < 0) {
675 last = next;
676 continue;
677 }
678 first = next+1;
679 }
680 return -first-1;
681}
682static void find_ref_delta_children(const struct object_id *oid,
684 int *first_index, int *last_index,
685 enum object_type type)
686{
687 int first = find_ref_delta(oid, type);
688 int last = first;
689 int end = nr_ref_deltas - 1;
690 if (first < 0) {
692 *first_index = 0;
693 *last_index = -1;
694 return;
695 }
696 while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
697 --first;
698 while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
699 ++last;
700 *first_index = first;
701 *last_index = last;
702}
703struct compare_data {
705 struct object_entry *entry;
706 struct git_istream *st;
707 unsigned char *buf;
708 unsigned long buf_size;
709};
710static int compare_objects(const unsigned char *buf, unsigned long size,
712 void *cb_data)
713{
714 struct compare_data *data = cb_data;
715 if (data->buf_size < size) {
717 free(data->buf);
718 data->buf = xmalloc(size);
719 data->buf_size = size;
720 }
721 while (size) {
723 ssize_t len = read_istream(data->st, data->buf, size);
724 if (len == 0)
725 die(_("SHA1 COLLISION FOUND WITH %s !"),
726 oid_to_hex(&data->entry->idx.oid));
727 if (len < 0)
728 die(_("unable to read %s"),
729 oid_to_hex(&data->entry->idx.oid));
730 if (memcmp(buf, data->buf, len))
731 die(_("SHA1 COLLISION FOUND WITH %s !"),
732 oid_to_hex(&data->entry->idx.oid));
733 size -= len;
734 buf += len;
735 }
736 return 0;
737}
738static int check_collison(struct object_entry *entry)
740{
741 struct compare_data data;
742 enum object_type type;
743 unsigned long size;
744 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
746 return -1;
747 memset(&data, 0, sizeof(data));
749 data.entry = entry;
750 data.st = open_istream(&entry->idx.oid, &type, &size, NULL);
751 if (!data.st)
752 return -1;
753 if (size != entry->size || type != entry->type)
754 die(_("SHA1 COLLISION FOUND WITH %s !"),
755 oid_to_hex(&entry->idx.oid));
756 unpack_data(entry, compare_objects, &data);
757 close_istream(data.st);
758 free(data.buf);
759 return 0;
760}
761static void sha1_object(const void *data, struct object_entry *obj_entry,
763 unsigned long size, enum object_type type,
764 const struct object_id *oid)
765{
766 void *new_data = NULL;
767 int collision_test_needed = 0;
768 assert(data || obj_entry);
770 if (startup_info->have_repository) {
772 read_lock();
773 collision_test_needed =
774 has_sha1_file_with_flags(oid->hash, OBJECT_INFO_QUICK);
775 read_unlock();
776 }
777 if (collision_test_needed && !data) {
779 read_lock();
780 if (!check_collison(obj_entry))
781 collision_test_needed = 0;
782 read_unlock();
783 }
784 if (collision_test_needed) {
785 void *has_data;
786 enum object_type has_type;
787 unsigned long has_size;
788 read_lock();
789 has_type = oid_object_info(the_repository, oid, &has_size);
790 if (has_type < 0)
791 die(_("cannot read existing object info %s"), oid_to_hex(oid));
792 if (has_type != type || has_size != size)
793 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
794 has_data = read_object_file(oid, &has_type, &has_size);
795 read_unlock();
796 if (!data)
797 data = new_data = get_data_from_pack(obj_entry);
798 if (!has_data)
799 die(_("cannot read existing object %s"), oid_to_hex(oid));
800 if (size != has_size || type != has_type ||
801 memcmp(data, has_data, size) != 0)
802 die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
803 free(has_data);
804 }
805 if (strict || do_fsck_object) {
807 read_lock();
808 if (type == OBJ_BLOB) {
809 struct blob *blob = lookup_blob(the_repository, oid);
810 if (blob)
811 blob->object.flags |= FLAG_CHECKED;
812 else
813 die(_("invalid blob object %s"), oid_to_hex(oid));
814 if (do_fsck_object &&
815 fsck_object(&blob->object, (void *)data, size, &fsck_options))
816 die(_("fsck error in packed object"));
817 } else {
818 struct object *obj;
819 int eaten;
820 void *buf = (void *) data;
821 assert(data && "data can only be NULL for large _blobs_");
823 /*
825 * we do not need to free the memory here, as the
826 * buf is deleted by the caller.
827 */
828 obj = parse_object_buffer(the_repository, oid, type,
829 size, buf,
830 &eaten);
831 if (!obj)
832 die(_("invalid %s"), type_name(type));
833 if (do_fsck_object &&
834 fsck_object(obj, buf, size, &fsck_options))
835 die(_("fsck error in packed object"));
836 if (strict && fsck_walk(obj, NULL, &fsck_options))
837 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
838 if (obj->type == OBJ_TREE) {
840 struct tree *item = (struct tree *) obj;
841 item->buffer = NULL;
842 obj->parsed = 0;
843 }
844 if (obj->type == OBJ_COMMIT) {
845 struct commit *commit = (struct commit *) obj;
846 if (detach_commit_buffer(commit, NULL) != data)
847 BUG("parse_object_buffer transmogrified our buffer");
848 }
849 obj->flags |= FLAG_CHECKED;
850 }
851 read_unlock();
852 }
853 free(new_data);
855}
856/*
858 * This function is part of find_unresolved_deltas(). There are two
859 * walkers going in the opposite ways.
860 *
861 * The first one in find_unresolved_deltas() traverses down from
862 * parent node to children, deflating nodes along the way. However,
863 * memory for deflated nodes is limited by delta_base_cache_limit, so
864 * at some point parent node's deflated content may be freed.
865 *
866 * The second walker is this function, which goes from current node up
867 * to top parent if necessary to deflate the node. In normal
868 * situation, its parent node would be already deflated, so it just
869 * needs to apply delta.
870 *
871 * In the worst case scenario, parent node is no longer deflated because
872 * we're running out of delta_base_cache_limit; we need to re-deflate
873 * parents, possibly up to the top base.
874 *
875 * All deflated objects here are subject to be freed if we exceed
876 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
877 * just need to make sure the last node is not freed.
878 */
879static void *get_base_data(struct base_data *c)
880{
881 if (!c->data) {
882 struct object_entry *obj = c->obj;
883 struct base_data **delta = NULL;
884 int delta_nr = 0, delta_alloc = 0;
885 while (is_delta_type(c->obj->type) && !c->data) {
887 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
888 delta[delta_nr++] = c;
889 c = c->base;
890 }
891 if (!delta_nr) {
892 c->data = get_data_from_pack(obj);
893 c->size = obj->size;
894 get_thread_data()->base_cache_used += c->size;
895 prune_base_data(c);
896 }
897 for (; delta_nr > 0; delta_nr--) {
898 void *base, *raw;
899 c = delta[delta_nr - 1];
900 obj = c->obj;
901 base = get_base_data(c->base);
902 raw = get_data_from_pack(obj);
903 c->data = patch_delta(
904 base, c->base->size,
905 raw, obj->size,
906 &c->size);
907 free(raw);
908 if (!c->data)
909 bad_object(obj->idx.offset, _("failed to apply delta"));
910 get_thread_data()->base_cache_used += c->size;
911 prune_base_data(c);
912 }
913 free(delta);
914 }
915 return c->data;
916}
917static void resolve_delta(struct object_entry *delta_obj,
919 struct base_data *base, struct base_data *result)
920{
921 void *base_data, *delta_data;
922 if (show_stat) {
924 int i = delta_obj - objects;
925 int j = base->obj - objects;
926 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
927 deepest_delta_lock();
928 if (deepest_delta < obj_stat[i].delta_depth)
929 deepest_delta = obj_stat[i].delta_depth;
930 deepest_delta_unlock();
931 obj_stat[i].base_object_no = j;
932 }
933 delta_data = get_data_from_pack(delta_obj);
934 base_data = get_base_data(base);
935 result->obj = delta_obj;
936 result->data = patch_delta(base_data, base->size,
937 delta_data, delta_obj->size, &result->size);
938 free(delta_data);
939 if (!result->data)
940 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
941 hash_object_file(result->data, result->size,
942 type_name(delta_obj->real_type), &delta_obj->idx.oid);
943 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
944 &delta_obj->idx.oid);
945 counter_lock();
946 nr_resolved_deltas++;
947 counter_unlock();
948}
949/*
951 * Standard boolean compare-and-swap: atomically check whether "*type" is
952 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
953 * and return false.
954 */
955static int compare_and_swap_type(signed char *type,
956 enum object_type want,
957 enum object_type set)
958{
959 enum object_type old;
960 type_cas_lock();
962 old = *type;
963 if (old == want)
964 *type = set;
965 type_cas_unlock();
966 return old == want;
968}
969static struct base_data *find_unresolved_deltas_1(struct base_data *base,
971 struct base_data *prev_base)
972{
973 if (base->ref_last == -1 && base->ofs_last == -1) {
974 find_ref_delta_children(&base->obj->idx.oid,
975 &base->ref_first, &base->ref_last,
976 OBJ_REF_DELTA);
977 find_ofs_delta_children(base->obj->idx.offset,
979 &base->ofs_first, &base->ofs_last,
980 OBJ_OFS_DELTA);
981 if (base->ref_last == -1 && base->ofs_last == -1) {
983 free(base->data);
984 return NULL;
985 }
986 link_base_data(prev_base, base);
988 }
989 if (base->ref_first <= base->ref_last) {
991 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
992 struct base_data *result = alloc_base_data();
993 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
995 base->obj->real_type))
996 BUG("child->real_type != OBJ_REF_DELTA");
997 resolve_delta(child, base, result);
999 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1000 free_base_data(base);
1001 base->ref_first++;
1003 return result;
1004 }
1005 if (base->ofs_first <= base->ofs_last) {
1007 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1008 struct base_data *result = alloc_base_data();
1009 assert(child->real_type == OBJ_OFS_DELTA);
1011 child->real_type = base->obj->real_type;
1012 resolve_delta(child, base, result);
1013 if (base->ofs_first == base->ofs_last)
1014 free_base_data(base);
1015 base->ofs_first++;
1017 return result;
1018 }
1019 unlink_base_data(base);
1021 return NULL;
1022}
1023static void find_unresolved_deltas(struct base_data *base)
1025{
1026 struct base_data *new_base, *prev_base = NULL;
1027 for (;;) {
1028 new_base = find_unresolved_deltas_1(base, prev_base);
1029 if (new_base) {
1031 prev_base = base;
1032 base = new_base;
1033 } else {
1034 free(base);
1035 base = prev_base;
1036 if (!base)
1037 return;
1038 prev_base = base->base;
1039 }
1040 }
1041}
1042static int compare_ofs_delta_entry(const void *a, const void *b)
1044{
1045 const struct ofs_delta_entry *delta_a = a;
1046 const struct ofs_delta_entry *delta_b = b;
1047 return delta_a->offset < delta_b->offset ? -1 :
1049 delta_a->offset > delta_b->offset ? 1 :
1050 0;
1051}
1052static int compare_ref_delta_entry(const void *a, const void *b)
1054{
1055 const struct ref_delta_entry *delta_a = a;
1056 const struct ref_delta_entry *delta_b = b;
1057 return oidcmp(&delta_a->oid, &delta_b->oid);
1059}
1060static void resolve_base(struct object_entry *obj)
1062{
1063 struct base_data *base_obj = alloc_base_data();
1064 base_obj->obj = obj;
1065 base_obj->data = NULL;
1066 find_unresolved_deltas(base_obj);
1067}
1068static void *threaded_second_pass(void *data)
1070{
1071 set_thread_data(data);
1072 for (;;) {
1073 int i;
1074 counter_lock();
1075 display_progress(progress, nr_resolved_deltas);
1076 counter_unlock();
1077 work_lock();
1078 while (nr_dispatched < nr_objects &&
1079 is_delta_type(objects[nr_dispatched].type))
1080 nr_dispatched++;
1081 if (nr_dispatched >= nr_objects) {
1082 work_unlock();
1083 break;
1084 }
1085 i = nr_dispatched++;
1086 work_unlock();
1087 resolve_base(&objects[i]);
1089 }
1090 return NULL;
1091}
1092/*
1094 * First pass:
1095 * - find locations of all objects;
1096 * - calculate SHA1 of all non-delta objects;
1097 * - remember base (SHA1 or offset) for all deltas.
1098 */
1099static void parse_pack_objects(unsigned char *hash)
1100{
1101 int i, nr_delays = 0;
1102 struct ofs_delta_entry *ofs_delta = ofs_deltas;
1103 struct object_id ref_delta_oid;
1104 struct stat st;
1105 if (verbose)
1107 progress = start_progress(
1108 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1109 nr_objects);
1110 for (i = 0; i < nr_objects; i++) {
1111 struct object_entry *obj = &objects[i];
1112 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1113 &ref_delta_oid,
1114 &obj->idx.oid);
1115 obj->real_type = obj->type;
1116 if (obj->type == OBJ_OFS_DELTA) {
1117 nr_ofs_deltas++;
1118 ofs_delta->obj_no = i;
1119 ofs_delta++;
1120 } else if (obj->type == OBJ_REF_DELTA) {
1121 ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1122 oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1123 ref_deltas[nr_ref_deltas].obj_no = i;
1124 nr_ref_deltas++;
1125 } else if (!data) {
1126 /* large blobs, check later */
1127 obj->real_type = OBJ_BAD;
1128 nr_delays++;
1129 } else
1130 sha1_object(data, NULL, obj->size, obj->type,
1131 &obj->idx.oid);
1132 free(data);
1133 display_progress(progress, i+1);
1134 }
1135 objects[i].idx.offset = consumed_bytes;
1136 stop_progress(&progress);
1137 /* Check pack integrity */
1139 flush();
1140 the_hash_algo->final_fn(hash, &input_ctx);
1141 if (!hasheq(fill(the_hash_algo->rawsz), hash))
1142 die(_("pack is corrupted (SHA1 mismatch)"));
1143 use(the_hash_algo->rawsz);
1144 /* If input_fd is a file, we should have reached its end now. */
1146 if (fstat(input_fd, &st))
1147 die_errno(_("cannot fstat packfile"));
1148 if (S_ISREG(st.st_mode) &&
1149 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1150 die(_("pack has junk at the end"));
1151 for (i = 0; i < nr_objects; i++) {
1153 struct object_entry *obj = &objects[i];
1154 if (obj->real_type != OBJ_BAD)
1155 continue;
1156 obj->real_type = obj->type;
1157 sha1_object(NULL, obj, obj->size, obj->type,
1158 &obj->idx.oid);
1159 nr_delays--;
1160 }
1161 if (nr_delays)
1162 die(_("confusion beyond insanity in parse_pack_objects()"));
1163}
1164/*
1166 * Second pass:
1167 * - for all non-delta objects, look if it is used as a base for
1168 * deltas;
1169 * - if used as a base, uncompress the object and apply all deltas,
1170 * recursively checking if the resulting object is used as a base
1171 * for some more deltas.
1172 */
1173static void resolve_deltas(void)
1174{
1175 int i;
1176 if (!nr_ofs_deltas && !nr_ref_deltas)
1178 return;
1179 /* Sort deltas by base SHA1/offset for fast searching */
1181 QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1182 QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1183 if (verbose || show_resolving_progress)
1185 progress = start_progress(_("Resolving deltas"),
1186 nr_ref_deltas + nr_ofs_deltas);
1187 nr_dispatched = 0;
1189 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1190 init_thread();
1191 for (i = 0; i < nr_threads; i++) {
1192 int ret = pthread_create(&thread_data[i].thread, NULL,
1193 threaded_second_pass, thread_data + i);
1194 if (ret)
1195 die(_("unable to create thread: %s"),
1196 strerror(ret));
1197 }
1198 for (i = 0; i < nr_threads; i++)
1199 pthread_join(thread_data[i].thread, NULL);
1200 cleanup_thread();
1201 return;
1202 }
1203 for (i = 0; i < nr_objects; i++) {
1205 struct object_entry *obj = &objects[i];
1206 if (is_delta_type(obj->type))
1208 continue;
1209 resolve_base(obj);
1210 display_progress(progress, nr_resolved_deltas);
1211 }
1212}
1213/*
1215 * Third pass:
1216 * - append objects to convert thin pack to full pack if required
1217 * - write the final pack hash
1218 */
1219static void fix_unresolved_deltas(struct hashfile *f);
1220static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1221{
1222 if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1223 stop_progress(&progress);
1224 /* Flush remaining pack final hash. */
1225 flush();
1226 return;
1227 }
1228 if (fix_thin_pack) {
1230 struct hashfile *f;
1231 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1232 struct strbuf msg = STRBUF_INIT;
1233 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1234 int nr_objects_initial = nr_objects;
1235 if (nr_unresolved <= 0)
1236 die(_("confusion beyond insanity"));
1237 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1238 memset(objects + nr_objects + 1, 0,
1239 nr_unresolved * sizeof(*objects));
1240 f = hashfd(output_fd, curr_pack);
1241 fix_unresolved_deltas(f);
1242 strbuf_addf(&msg, Q_("completed with %d local object",
1243 "completed with %d local objects",
1244 nr_objects - nr_objects_initial),
1245 nr_objects - nr_objects_initial);
1246 stop_progress_msg(&progress, msg.buf);
1247 strbuf_release(&msg);
1248 finalize_hashfile(f, tail_hash, 0);
1249 hashcpy(read_hash, pack_hash);
1250 fixup_pack_header_footer(output_fd, pack_hash,
1251 curr_pack, nr_objects,
1252 read_hash, consumed_bytes-the_hash_algo->rawsz);
1253 if (!hasheq(read_hash, tail_hash))
1254 die(_("Unexpected tail checksum for %s "
1255 "(disk corruption?)"), curr_pack);
1256 }
1257 if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1258 die(Q_("pack has %d unresolved delta",
1259 "pack has %d unresolved deltas",
1260 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1261 nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1262}
1263static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1265{
1266 git_zstream stream;
1267 int status;
1268 unsigned char outbuf[4096];
1269 git_deflate_init(&stream, zlib_compression_level);
1271 stream.next_in = in;
1272 stream.avail_in = size;
1273 do {
1275 stream.next_out = outbuf;
1276 stream.avail_out = sizeof(outbuf);
1277 status = git_deflate(&stream, Z_FINISH);
1278 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1279 } while (status == Z_OK);
1280 if (status != Z_STREAM_END)
1282 die(_("unable to deflate appended object (%d)"), status);
1283 size = stream.total_out;
1284 git_deflate_end(&stream);
1285 return size;
1286}
1287static struct object_entry *append_obj_to_pack(struct hashfile *f,
1289 const unsigned char *sha1, void *buf,
1290 unsigned long size, enum object_type type)
1291{
1292 struct object_entry *obj = &objects[nr_objects++];
1293 unsigned char header[10];
1294 unsigned long s = size;
1295 int n = 0;
1296 unsigned char c = (type << 4) | (s & 15);
1297 s >>= 4;
1298 while (s) {
1299 header[n++] = c | 0x80;
1300 c = s & 0x7f;
1301 s >>= 7;
1302 }
1303 header[n++] = c;
1304 crc32_begin(f);
1305 hashwrite(f, header, n);
1306 obj[0].size = size;
1307 obj[0].hdr_size = n;
1308 obj[0].type = type;
1309 obj[0].real_type = type;
1310 obj[1].idx.offset = obj[0].idx.offset + n;
1311 obj[1].idx.offset += write_compressed(f, buf, size);
1312 obj[0].idx.crc32 = crc32_end(f);
1313 hashflush(f);
1314 hashcpy(obj->idx.oid.hash, sha1);
1315 return obj;
1316}
1317static int delta_pos_compare(const void *_a, const void *_b)
1319{
1320 struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1321 struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1322 return a->obj_no - b->obj_no;
1323}
1324static void fix_unresolved_deltas(struct hashfile *f)
1326{
1327 struct ref_delta_entry **sorted_by_pos;
1328 int i;
1329 /*
1331 * Since many unresolved deltas may well be themselves base objects
1332 * for more unresolved deltas, we really want to include the
1333 * smallest number of base objects that would cover as much delta
1334 * as possible by picking the
1335 * trunc deltas first, allowing for other deltas to resolve without
1336 * additional base objects. Since most base objects are to be found
1337 * before deltas depending on them, a good heuristic is to start
1338 * resolving deltas in the same order as their position in the pack.
1339 */
1340 ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1341 for (i = 0; i < nr_ref_deltas; i++)
1342 sorted_by_pos[i] = &ref_deltas[i];
1343 QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1344 for (i = 0; i < nr_ref_deltas; i++) {
1346 struct ref_delta_entry *d = sorted_by_pos[i];
1347 enum object_type type;
1348 struct base_data *base_obj = alloc_base_data();
1349 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1351 continue;
1352 base_obj->data = read_object_file(&d->oid, &type,
1353 &base_obj->size);
1354 if (!base_obj->data)
1355 continue;
1356 if (check_object_signature(&d->oid, base_obj->data,
1358 base_obj->size, type_name(type)))
1359 die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1360 base_obj->obj = append_obj_to_pack(f, d->oid.hash,
1361 base_obj->data, base_obj->size, type);
1362 find_unresolved_deltas(base_obj);
1363 display_progress(progress, nr_resolved_deltas);
1364 }
1365 free(sorted_by_pos);
1366}
1367static const char *derive_filename(const char *pack_name, const char *suffix,
1369 struct strbuf *buf)
1370{
1371 size_t len;
1372 if (!strip_suffix(pack_name, ".pack", &len))
1373 die(_("packfile name '%s' does not end with '.pack'"),
1374 pack_name);
1375 strbuf_add(buf, pack_name, len);
1376 strbuf_addch(buf, '.');
1377 strbuf_addstr(buf, suffix);
1378 return buf->buf;
1379}
1380static void write_special_file(const char *suffix, const char *msg,
1382 const char *pack_name, const unsigned char *hash,
1383 const char **report)
1384{
1385 struct strbuf name_buf = STRBUF_INIT;
1386 const char *filename;
1387 int fd;
1388 int msg_len = strlen(msg);
1389 if (pack_name)
1391 filename = derive_filename(pack_name, suffix, &name_buf);
1392 else
1393 filename = odb_pack_name(&name_buf, hash, suffix);
1394 fd = odb_pack_keep(filename);
1396 if (fd < 0) {
1397 if (errno != EEXIST)
1398 die_errno(_("cannot write %s file '%s'"),
1399 suffix, filename);
1400 } else {
1401 if (msg_len > 0) {
1402 write_or_die(fd, msg, msg_len);
1403 write_or_die(fd, "\n", 1);
1404 }
1405 if (close(fd) != 0)
1406 die_errno(_("cannot close written %s file '%s'"),
1407 suffix, filename);
1408 if (report)
1409 *report = suffix;
1410 }
1411 strbuf_release(&name_buf);
1412}
1413static void final(const char *final_pack_name, const char *curr_pack_name,
1415 const char *final_index_name, const char *curr_index_name,
1416 const char *keep_msg, const char *promisor_msg,
1417 unsigned char *hash)
1418{
1419 const char *report = "pack";
1420 struct strbuf pack_name = STRBUF_INIT;
1421 struct strbuf index_name = STRBUF_INIT;
1422 int err;
1423 if (!from_stdin) {
1425 close(input_fd);
1426 } else {
1427 fsync_or_die(output_fd, curr_pack_name);
1428 err = close(output_fd);
1429 if (err)
1430 die_errno(_("error while closing pack file"));
1431 }
1432 if (keep_msg)
1434 write_special_file("keep", keep_msg, final_pack_name, hash,
1435 &report);
1436 if (promisor_msg)
1437 write_special_file("promisor", promisor_msg, final_pack_name,
1438 hash, NULL);
1439 if (final_pack_name != curr_pack_name) {
1441 if (!final_pack_name)
1442 final_pack_name = odb_pack_name(&pack_name, hash, "pack");
1443 if (finalize_object_file(curr_pack_name, final_pack_name))
1444 die(_("cannot store pack file"));
1445 } else if (from_stdin)
1446 chmod(final_pack_name, 0444);
1447 if (final_index_name != curr_index_name) {
1449 if (!final_index_name)
1450 final_index_name = odb_pack_name(&index_name, hash, "idx");
1451 if (finalize_object_file(curr_index_name, final_index_name))
1452 die(_("cannot store index file"));
1453 } else
1454 chmod(final_index_name, 0444);
1455 if (do_fsck_object) {
1457 struct packed_git *p;
1458 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1459 if (p)
1460 install_packed_git(the_repository, p);
1461 }
1462 if (!from_stdin) {
1464 printf("%s\n", sha1_to_hex(hash));
1465 } else {
1466 struct strbuf buf = STRBUF_INIT;
1467 strbuf_addf(&buf, "%s\t%s\n", report, sha1_to_hex(hash));
1469 write_or_die(1, buf.buf, buf.len);
1470 strbuf_release(&buf);
1471 /*
1473 * Let's just mimic git-unpack-objects here and write
1474 * the last part of the input buffer to stdout.
1475 */
1476 while (input_len) {
1477 err = xwrite(1, input_buffer + input_offset, input_len);
1478 if (err <= 0)
1479 break;
1480 input_len -= err;
1481 input_offset += err;
1482 }
1483 }
1484 strbuf_release(&index_name);
1486 strbuf_release(&pack_name);
1487}
1488static int git_index_pack_config(const char *k, const char *v, void *cb)
1490{
1491 struct pack_idx_option *opts = cb;
1492 if (!strcmp(k, "pack.indexversion")) {
1494 opts->version = git_config_int(k, v);
1495 if (opts->version > 2)
1496 die(_("bad pack.indexversion=%"PRIu32), opts->version);
1497 return 0;
1498 }
1499 if (!strcmp(k, "pack.threads")) {
1500 nr_threads = git_config_int(k, v);
1501 if (nr_threads < 0)
1502 die(_("invalid number of threads specified (%d)"),
1503 nr_threads);
1504 if (!HAVE_THREADS && nr_threads != 1) {
1505 warning(_("no threads support, ignoring %s"), k);
1506 nr_threads = 1;
1507 }
1508 return 0;
1509 }
1510 return git_default_config(k, v, cb);
1511}
1512static int cmp_uint32(const void *a_, const void *b_)
1514{
1515 uint32_t a = *((uint32_t *)a_);
1516 uint32_t b = *((uint32_t *)b_);
1517 return (a < b) ? -1 : (a != b);
1519}
1520static void read_v2_anomalous_offsets(struct packed_git *p,
1522 struct pack_idx_option *opts)
1523{
1524 const uint32_t *idx1, *idx2;
1525 uint32_t i;
1526 const uint32_t hashwords = the_hash_algo->rawsz / sizeof(uint32_t);
1527 /* The address of the 4-byte offset table */
1529 idx1 = (((const uint32_t *)p->index_data)
1530 + 2 /* 8-byte header */
1531 + 256 /* fan out */
1532 + hashwords * p->num_objects /* object ID table */
1533 + p->num_objects /* CRC32 table */
1534 );
1535 /* The address of the 8-byte offset table */
1537 idx2 = idx1 + p->num_objects;
1538 for (i = 0; i < p->num_objects; i++) {
1540 uint32_t off = ntohl(idx1[i]);
1541 if (!(off & 0x80000000))
1542 continue;
1543 off = off & 0x7fffffff;
1544 check_pack_index_ptr(p, &idx2[off * 2]);
1545 if (idx2[off * 2])
1546 continue;
1547 /*
1548 * The real offset is ntohl(idx2[off * 2]) in high 4
1549 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1550 * octets. But idx2[off * 2] is Zero!!!
1551 */
1552 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1553 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1554 }
1555 QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1557}
1558static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1560{
1561 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1562 if (!p)
1564 die(_("Cannot open existing pack file '%s'"), pack_name);
1565 if (open_pack_index(p))
1566 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1567 /* Read the attributes from the existing idx file */
1569 opts->version = p->index_version;
1570 if (opts->version == 2)
1572 read_v2_anomalous_offsets(p, opts);
1573 /*
1575 * Get rid of the idx file as we do not need it anymore.
1576 * NEEDSWORK: extract this bit from free_pack_by_name() in
1577 * sha1-file.c, perhaps? It shouldn't matter very much as we
1578 * know we haven't installed this pack (hence we never have
1579 * read anything from it).
1580 */
1581 close_pack_index(p);
1582 free(p);
1583}
1584static void show_pack_info(int stat_only)
1586{
1587 int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1588 unsigned long *chain_histogram = NULL;
1589 if (deepest_delta)
1591 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1592 for (i = 0; i < nr_objects; i++) {
1594 struct object_entry *obj = &objects[i];
1595 if (is_delta_type(obj->type))
1597 chain_histogram[obj_stat[i].delta_depth - 1]++;
1598 if (stat_only)
1599 continue;
1600 printf("%s %-6s %lu %lu %"PRIuMAX,
1601 oid_to_hex(&obj->idx.oid),
1602 type_name(obj->real_type), obj->size,
1603 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1604 (uintmax_t)obj->idx.offset);
1605 if (is_delta_type(obj->type)) {
1606 struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1607 printf(" %u %s", obj_stat[i].delta_depth,
1608 oid_to_hex(&bobj->idx.oid));
1609 }
1610 putchar('\n');
1611 }
1612 if (baseobjects)
1614 printf_ln(Q_("non delta: %d object",
1615 "non delta: %d objects",
1616 baseobjects),
1617 baseobjects);
1618 for (i = 0; i < deepest_delta; i++) {
1619 if (!chain_histogram[i])
1620 continue;
1621 printf_ln(Q_("chain length = %d: %lu object",
1622 "chain length = %d: %lu objects",
1623 chain_histogram[i]),
1624 i + 1,
1625 chain_histogram[i]);
1626 }
1627}
1628int cmd_index_pack(int argc, const char **argv, const char *prefix)
1630{
1631 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1632 const char *curr_index;
1633 const char *index_name = NULL, *pack_name = NULL;
1634 const char *keep_msg = NULL;
1635 const char *promisor_msg = NULL;
1636 struct strbuf index_name_buf = STRBUF_INIT;
1637 struct pack_idx_entry **idx_objects;
1638 struct pack_idx_option opts;
1639 unsigned char pack_hash[GIT_MAX_RAWSZ];
1640 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
1641 int report_end_of_input = 0;
1642 /*
1644 * index-pack never needs to fetch missing objects, since it only
1645 * accesses the repo to do hash collision checks
1646 */
1647 fetch_if_missing = 0;
1648 if (argc == 2 && !strcmp(argv[1], "-h"))
1650 usage(index_pack_usage);
1651 read_replace_refs = 0;
1653 fsck_options.walk = mark_link;
1654 reset_pack_idx_option(&opts);
1656 git_config(git_index_pack_config, &opts);
1657 if (prefix && chdir(prefix))
1658 die(_("Cannot come back to cwd"));
1659 for (i = 1; i < argc; i++) {
1661 const char *arg = argv[i];
1662 if (*arg == '-') {
1664 if (!strcmp(arg, "--stdin")) {
1665 from_stdin = 1;
1666 } else if (!strcmp(arg, "--fix-thin")) {
1667 fix_thin_pack = 1;
1668 } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1669 strict = 1;
1670 do_fsck_object = 1;
1671 fsck_set_msg_types(&fsck_options, arg);
1672 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1673 strict = 1;
1674 check_self_contained_and_connected = 1;
1675 } else if (!strcmp(arg, "--fsck-objects")) {
1676 do_fsck_object = 1;
1677 } else if (!strcmp(arg, "--verify")) {
1678 verify = 1;
1679 } else if (!strcmp(arg, "--verify-stat")) {
1680 verify = 1;
1681 show_stat = 1;
1682 } else if (!strcmp(arg, "--verify-stat-only")) {
1683 verify = 1;
1684 show_stat = 1;
1685 stat_only = 1;
1686 } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1687 ; /* nothing to do */
1688 } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1689 ; /* already parsed */
1690 } else if (starts_with(arg, "--threads=")) {
1691 char *end;
1692 nr_threads = strtoul(arg+10, &end, 0);
1693 if (!arg[10] || *end || nr_threads < 0)
1694 usage(index_pack_usage);
1695 if (!HAVE_THREADS && nr_threads != 1) {
1696 warning(_("no threads support, ignoring %s"), arg);
1697 nr_threads = 1;
1698 }
1699 } else if (starts_with(arg, "--pack_header=")) {
1700 struct pack_header *hdr;
1701 char *c;
1702 hdr = (struct pack_header *)input_buffer;
1704 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1705 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1706 if (*c != ',')
1707 die(_("bad %s"), arg);
1708 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1709 if (*c)
1710 die(_("bad %s"), arg);
1711 input_len = sizeof(*hdr);
1712 } else if (!strcmp(arg, "-v")) {
1713 verbose = 1;
1714 } else if (!strcmp(arg, "--show-resolving-progress")) {
1715 show_resolving_progress = 1;
1716 } else if (!strcmp(arg, "--report-end-of-input")) {
1717 report_end_of_input = 1;
1718 } else if (!strcmp(arg, "-o")) {
1719 if (index_name || (i+1) >= argc)
1720 usage(index_pack_usage);
1721 index_name = argv[++i];
1722 } else if (starts_with(arg, "--index-version=")) {
1723 char *c;
1724 opts.version = strtoul(arg + 16, &c, 10);
1725 if (opts.version > 2)
1726 die(_("bad %s"), arg);
1727 if (*c == ',')
1728 opts.off32_limit = strtoul(c+1, &c, 0);
1729 if (*c || opts.off32_limit & 0x80000000)
1730 die(_("bad %s"), arg);
1731 } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1732 max_input_size = strtoumax(arg, NULL, 10);
1733 } else
1734 usage(index_pack_usage);
1735 continue;
1736 }
1737 if (pack_name)
1739 usage(index_pack_usage);
1740 pack_name = arg;
1741 }
1742 if (!pack_name && !from_stdin)
1744 usage(index_pack_usage);
1745 if (fix_thin_pack && !from_stdin)
1746 die(_("--fix-thin cannot be used without --stdin"));
1747 if (from_stdin && !startup_info->have_repository)
1748 die(_("--stdin requires a git repository"));
1749 if (!index_name && pack_name)
1750 index_name = derive_filename(pack_name, "idx", &index_name_buf);
1751 if (verify) {
1753 if (!index_name)
1754 die(_("--verify with no packfile name given"));
1755 read_idx_option(&opts, index_name);
1756 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1757 }
1758 if (strict)
1759 opts.flags |= WRITE_IDX_STRICT;
1760 if (HAVE_THREADS && !nr_threads) {
1762 nr_threads = online_cpus();
1763 /* An experiment showed that more threads does not mean faster */
1764 if (nr_threads > 3)
1765 nr_threads = 3;
1766 }
1767 curr_pack = open_pack_file(pack_name);
1769 parse_pack_header();
1770 objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1771 if (show_stat)
1772 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1773 ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1774 parse_pack_objects(pack_hash);
1775 if (report_end_of_input)
1776 write_in_full(2, "\0", 1);
1777 resolve_deltas();
1778 conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1779 free(ofs_deltas);
1780 free(ref_deltas);
1781 if (strict)
1782 foreign_nr = check_objects();
1783 if (show_stat)
1785 show_pack_info(stat_only);
1786 ALLOC_ARRAY(idx_objects, nr_objects);
1788 for (i = 0; i < nr_objects; i++)
1789 idx_objects[i] = &objects[i].idx;
1790 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1791 free(idx_objects);
1792 if (!verify)
1794 final(pack_name, curr_pack,
1795 index_name, curr_index,
1796 keep_msg, promisor_msg,
1797 pack_hash);
1798 else
1799 close(input_fd);
1800 if (do_fsck_object && fsck_finish(&fsck_options))
1802 die(_("fsck error in pack objects"));
1803 free(objects);
1805 strbuf_release(&index_name_buf);
1806 if (pack_name == NULL)
1807 free((void *) curr_pack);
1808 if (index_name == NULL)
1809 free((void *) curr_index);
1810 /*
1812 * Let the caller know this pack is not self contained
1813 */
1814 if (check_self_contained_and_connected && foreign_nr)
1815 return 1;
1816 return 0;
1818}