index-pack: use streaming interface on large blobs (most of the time)
[git] / builtin / index-pack.c
1 #include "builtin.h"
2 #include "delta.h"
3 #include "pack.h"
4 #include "csum-file.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "progress.h"
10 #include "fsck.h"
11 #include "exec_cmd.h"
12 #include "thread-utils.h"
13
14 static const char index_pack_usage[] =
15 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
16
17 struct object_entry {
18         struct pack_idx_entry idx;
19         unsigned long size;
20         unsigned int hdr_size;
21         enum object_type type;
22         enum object_type real_type;
23         unsigned delta_depth;
24         int base_object_no;
25 };
26
27 union delta_base {
28         unsigned char sha1[20];
29         off_t offset;
30 };
31
32 struct base_data {
33         struct base_data *base;
34         struct base_data *child;
35         struct object_entry *obj;
36         void *data;
37         unsigned long size;
38         int ref_first, ref_last;
39         int ofs_first, ofs_last;
40 };
41
42 #if !defined(NO_PTHREADS) && defined(NO_PREAD)
43 /* NO_PREAD uses compat/pread.c, which is not thread-safe. Disable threading. */
44 #define NO_PTHREADS
45 #endif
46
47 struct thread_local {
48 #ifndef NO_PTHREADS
49         pthread_t thread;
50 #endif
51         struct base_data *base_cache;
52         size_t base_cache_used;
53 };
54
55 /*
56  * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
57  * to memcmp() only the first 20 bytes.
58  */
59 #define UNION_BASE_SZ   20
60
61 #define FLAG_LINK (1u<<20)
62 #define FLAG_CHECKED (1u<<21)
63
64 struct delta_entry {
65         union delta_base base;
66         int obj_no;
67 };
68
69 static struct object_entry *objects;
70 static struct delta_entry *deltas;
71 static struct thread_local nothread_data;
72 static int nr_objects;
73 static int nr_deltas;
74 static int nr_resolved_deltas;
75 static int nr_threads;
76
77 static int from_stdin;
78 static int strict;
79 static int verbose;
80
81 static struct progress *progress;
82
83 /* We always read in 4kB chunks. */
84 static unsigned char input_buffer[4096];
85 static unsigned int input_offset, input_len;
86 static off_t consumed_bytes;
87 static unsigned deepest_delta;
88 static git_SHA_CTX input_ctx;
89 static uint32_t input_crc32;
90 static int input_fd, output_fd, pack_fd;
91
92 #ifndef NO_PTHREADS
93
94 static struct thread_local *thread_data;
95 static int nr_dispatched;
96 static int threads_active;
97
98 static pthread_mutex_t read_mutex;
99 #define read_lock()             lock_mutex(&read_mutex)
100 #define read_unlock()           unlock_mutex(&read_mutex)
101
102 static pthread_mutex_t counter_mutex;
103 #define counter_lock()          lock_mutex(&counter_mutex)
104 #define counter_unlock()        unlock_mutex(&counter_mutex)
105
106 static pthread_mutex_t work_mutex;
107 #define work_lock()             lock_mutex(&work_mutex)
108 #define work_unlock()           unlock_mutex(&work_mutex)
109
110 static pthread_key_t key;
111
112 static inline void lock_mutex(pthread_mutex_t *mutex)
113 {
114         if (threads_active)
115                 pthread_mutex_lock(mutex);
116 }
117
118 static inline void unlock_mutex(pthread_mutex_t *mutex)
119 {
120         if (threads_active)
121                 pthread_mutex_unlock(mutex);
122 }
123
124 /*
125  * Mutex and conditional variable can't be statically-initialized on Windows.
126  */
127 static void init_thread(void)
128 {
129         init_recursive_mutex(&read_mutex);
130         pthread_mutex_init(&counter_mutex, NULL);
131         pthread_mutex_init(&work_mutex, NULL);
132         pthread_key_create(&key, NULL);
133         thread_data = xcalloc(nr_threads, sizeof(*thread_data));
134         threads_active = 1;
135 }
136
137 static void cleanup_thread(void)
138 {
139         if (!threads_active)
140                 return;
141         threads_active = 0;
142         pthread_mutex_destroy(&read_mutex);
143         pthread_mutex_destroy(&counter_mutex);
144         pthread_mutex_destroy(&work_mutex);
145         pthread_key_delete(key);
146         free(thread_data);
147 }
148
149 #else
150
151 #define read_lock()
152 #define read_unlock()
153
154 #define counter_lock()
155 #define counter_unlock()
156
157 #define work_lock()
158 #define work_unlock()
159
160 #endif
161
162
163 static int mark_link(struct object *obj, int type, void *data)
164 {
165         if (!obj)
166                 return -1;
167
168         if (type != OBJ_ANY && obj->type != type)
169                 die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
170
171         obj->flags |= FLAG_LINK;
172         return 0;
173 }
174
175 /* The content of each linked object must have been checked
176    or it must be already present in the object database */
177 static void check_object(struct object *obj)
178 {
179         if (!obj)
180                 return;
181
182         if (!(obj->flags & FLAG_LINK))
183                 return;
184
185         if (!(obj->flags & FLAG_CHECKED)) {
186                 unsigned long size;
187                 int type = sha1_object_info(obj->sha1, &size);
188                 if (type != obj->type || type <= 0)
189                         die(_("object of unexpected type"));
190                 obj->flags |= FLAG_CHECKED;
191                 return;
192         }
193 }
194
195 static void check_objects(void)
196 {
197         unsigned i, max;
198
199         max = get_max_object_index();
200         for (i = 0; i < max; i++)
201                 check_object(get_indexed_object(i));
202 }
203
204
205 /* Discard current buffer used content. */
206 static void flush(void)
207 {
208         if (input_offset) {
209                 if (output_fd >= 0)
210                         write_or_die(output_fd, input_buffer, input_offset);
211                 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
212                 memmove(input_buffer, input_buffer + input_offset, input_len);
213                 input_offset = 0;
214         }
215 }
216
217 /*
218  * Make sure at least "min" bytes are available in the buffer, and
219  * return the pointer to the buffer.
220  */
221 static void *fill(int min)
222 {
223         if (min <= input_len)
224                 return input_buffer + input_offset;
225         if (min > sizeof(input_buffer))
226                 die(Q_("cannot fill %d byte",
227                        "cannot fill %d bytes",
228                        min),
229                     min);
230         flush();
231         do {
232                 ssize_t ret = xread(input_fd, input_buffer + input_len,
233                                 sizeof(input_buffer) - input_len);
234                 if (ret <= 0) {
235                         if (!ret)
236                                 die(_("early EOF"));
237                         die_errno(_("read error on input"));
238                 }
239                 input_len += ret;
240                 if (from_stdin)
241                         display_throughput(progress, consumed_bytes + input_len);
242         } while (input_len < min);
243         return input_buffer;
244 }
245
246 static void use(int bytes)
247 {
248         if (bytes > input_len)
249                 die(_("used more bytes than were available"));
250         input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
251         input_len -= bytes;
252         input_offset += bytes;
253
254         /* make sure off_t is sufficiently large not to wrap */
255         if (signed_add_overflows(consumed_bytes, bytes))
256                 die(_("pack too large for current definition of off_t"));
257         consumed_bytes += bytes;
258 }
259
260 static const char *open_pack_file(const char *pack_name)
261 {
262         if (from_stdin) {
263                 input_fd = 0;
264                 if (!pack_name) {
265                         static char tmp_file[PATH_MAX];
266                         output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
267                                                 "pack/tmp_pack_XXXXXX");
268                         pack_name = xstrdup(tmp_file);
269                 } else
270                         output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
271                 if (output_fd < 0)
272                         die_errno(_("unable to create '%s'"), pack_name);
273                 pack_fd = output_fd;
274         } else {
275                 input_fd = open(pack_name, O_RDONLY);
276                 if (input_fd < 0)
277                         die_errno(_("cannot open packfile '%s'"), pack_name);
278                 output_fd = -1;
279                 pack_fd = input_fd;
280         }
281         git_SHA1_Init(&input_ctx);
282         return pack_name;
283 }
284
285 static void parse_pack_header(void)
286 {
287         struct pack_header *hdr = fill(sizeof(struct pack_header));
288
289         /* Header consistency check */
290         if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
291                 die(_("pack signature mismatch"));
292         if (!pack_version_ok(hdr->hdr_version))
293                 die("pack version %"PRIu32" unsupported",
294                         ntohl(hdr->hdr_version));
295
296         nr_objects = ntohl(hdr->hdr_entries);
297         use(sizeof(struct pack_header));
298 }
299
300 static NORETURN void bad_object(unsigned long offset, const char *format,
301                        ...) __attribute__((format (printf, 2, 3)));
302
303 static NORETURN void bad_object(unsigned long offset, const char *format, ...)
304 {
305         va_list params;
306         char buf[1024];
307
308         va_start(params, format);
309         vsnprintf(buf, sizeof(buf), format, params);
310         va_end(params);
311         die(_("pack has bad object at offset %lu: %s"), offset, buf);
312 }
313
314 static inline struct thread_local *get_thread_data(void)
315 {
316 #ifndef NO_PTHREADS
317         if (threads_active)
318                 return pthread_getspecific(key);
319         assert(!threads_active &&
320                "This should only be reached when all threads are gone");
321 #endif
322         return &nothread_data;
323 }
324
325 #ifndef NO_PTHREADS
326 static void set_thread_data(struct thread_local *data)
327 {
328         if (threads_active)
329                 pthread_setspecific(key, data);
330 }
331 #endif
332
333 static struct base_data *alloc_base_data(void)
334 {
335         struct base_data *base = xmalloc(sizeof(struct base_data));
336         memset(base, 0, sizeof(*base));
337         base->ref_last = -1;
338         base->ofs_last = -1;
339         return base;
340 }
341
342 static void free_base_data(struct base_data *c)
343 {
344         if (c->data) {
345                 free(c->data);
346                 c->data = NULL;
347                 get_thread_data()->base_cache_used -= c->size;
348         }
349 }
350
351 static void prune_base_data(struct base_data *retain)
352 {
353         struct base_data *b;
354         struct thread_local *data = get_thread_data();
355         for (b = data->base_cache;
356              data->base_cache_used > delta_base_cache_limit && b;
357              b = b->child) {
358                 if (b->data && b != retain)
359                         free_base_data(b);
360         }
361 }
362
363 static void link_base_data(struct base_data *base, struct base_data *c)
364 {
365         if (base)
366                 base->child = c;
367         else
368                 get_thread_data()->base_cache = c;
369
370         c->base = base;
371         c->child = NULL;
372         if (c->data)
373                 get_thread_data()->base_cache_used += c->size;
374         prune_base_data(c);
375 }
376
377 static void unlink_base_data(struct base_data *c)
378 {
379         struct base_data *base = c->base;
380         if (base)
381                 base->child = NULL;
382         else
383                 get_thread_data()->base_cache = NULL;
384         free_base_data(c);
385 }
386
387 static int is_delta_type(enum object_type type)
388 {
389         return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
390 }
391
392 static void *unpack_entry_data(unsigned long offset, unsigned long size,
393                                enum object_type type, unsigned char *sha1)
394 {
395         static char fixed_buf[8192];
396         int status;
397         git_zstream stream;
398         void *buf;
399         git_SHA_CTX c;
400         char hdr[32];
401         int hdrlen;
402
403         if (!is_delta_type(type)) {
404                 hdrlen = sprintf(hdr, "%s %lu", typename(type), size) + 1;
405                 git_SHA1_Init(&c);
406                 git_SHA1_Update(&c, hdr, hdrlen);
407         } else
408                 sha1 = NULL;
409         if (type == OBJ_BLOB && size > big_file_threshold)
410                 buf = fixed_buf;
411         else
412                 buf = xmalloc(size);
413
414         memset(&stream, 0, sizeof(stream));
415         git_inflate_init(&stream);
416         stream.next_out = buf;
417         stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
418
419         do {
420                 unsigned char *last_out = stream.next_out;
421                 stream.next_in = fill(1);
422                 stream.avail_in = input_len;
423                 status = git_inflate(&stream, 0);
424                 use(input_len - stream.avail_in);
425                 if (sha1)
426                         git_SHA1_Update(&c, last_out, stream.next_out - last_out);
427                 if (buf == fixed_buf) {
428                         stream.next_out = buf;
429                         stream.avail_out = sizeof(fixed_buf);
430                 }
431         } while (status == Z_OK);
432         if (stream.total_out != size || status != Z_STREAM_END)
433                 bad_object(offset, _("inflate returned %d"), status);
434         git_inflate_end(&stream);
435         if (sha1)
436                 git_SHA1_Final(sha1, &c);
437         return buf == fixed_buf ? NULL : buf;
438 }
439
440 static void *unpack_raw_entry(struct object_entry *obj,
441                               union delta_base *delta_base,
442                               unsigned char *sha1)
443 {
444         unsigned char *p;
445         unsigned long size, c;
446         off_t base_offset;
447         unsigned shift;
448         void *data;
449
450         obj->idx.offset = consumed_bytes;
451         input_crc32 = crc32(0, NULL, 0);
452
453         p = fill(1);
454         c = *p;
455         use(1);
456         obj->type = (c >> 4) & 7;
457         size = (c & 15);
458         shift = 4;
459         while (c & 0x80) {
460                 p = fill(1);
461                 c = *p;
462                 use(1);
463                 size += (c & 0x7f) << shift;
464                 shift += 7;
465         }
466         obj->size = size;
467
468         switch (obj->type) {
469         case OBJ_REF_DELTA:
470                 hashcpy(delta_base->sha1, fill(20));
471                 use(20);
472                 break;
473         case OBJ_OFS_DELTA:
474                 memset(delta_base, 0, sizeof(*delta_base));
475                 p = fill(1);
476                 c = *p;
477                 use(1);
478                 base_offset = c & 127;
479                 while (c & 128) {
480                         base_offset += 1;
481                         if (!base_offset || MSB(base_offset, 7))
482                                 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
483                         p = fill(1);
484                         c = *p;
485                         use(1);
486                         base_offset = (base_offset << 7) + (c & 127);
487                 }
488                 delta_base->offset = obj->idx.offset - base_offset;
489                 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
490                         bad_object(obj->idx.offset, _("delta base offset is out of bound"));
491                 break;
492         case OBJ_COMMIT:
493         case OBJ_TREE:
494         case OBJ_BLOB:
495         case OBJ_TAG:
496                 break;
497         default:
498                 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
499         }
500         obj->hdr_size = consumed_bytes - obj->idx.offset;
501
502         data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
503         obj->idx.crc32 = input_crc32;
504         return data;
505 }
506
507 static void *get_data_from_pack(struct object_entry *obj)
508 {
509         off_t from = obj[0].idx.offset + obj[0].hdr_size;
510         unsigned long len = obj[1].idx.offset - from;
511         unsigned char *data, *inbuf;
512         git_zstream stream;
513         int status;
514
515         data = xmalloc(obj->size);
516         inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
517
518         memset(&stream, 0, sizeof(stream));
519         git_inflate_init(&stream);
520         stream.next_out = data;
521         stream.avail_out = obj->size;
522
523         do {
524                 ssize_t n = (len < 64*1024) ? len : 64*1024;
525                 n = pread(pack_fd, inbuf, n, from);
526                 if (n < 0)
527                         die_errno(_("cannot pread pack file"));
528                 if (!n)
529                         die(Q_("premature end of pack file, %lu byte missing",
530                                "premature end of pack file, %lu bytes missing",
531                                len),
532                             len);
533                 from += n;
534                 len -= n;
535                 stream.next_in = inbuf;
536                 stream.avail_in = n;
537                 status = git_inflate(&stream, 0);
538         } while (len && status == Z_OK && !stream.avail_in);
539
540         /* This has been inflated OK when first encountered, so... */
541         if (status != Z_STREAM_END || stream.total_out != obj->size)
542                 die(_("serious inflate inconsistency"));
543
544         git_inflate_end(&stream);
545         free(inbuf);
546         return data;
547 }
548
549 static int compare_delta_bases(const union delta_base *base1,
550                                const union delta_base *base2,
551                                enum object_type type1,
552                                enum object_type type2)
553 {
554         int cmp = type1 - type2;
555         if (cmp)
556                 return cmp;
557         return memcmp(base1, base2, UNION_BASE_SZ);
558 }
559
560 static int find_delta(const union delta_base *base, enum object_type type)
561 {
562         int first = 0, last = nr_deltas;
563
564         while (first < last) {
565                 int next = (first + last) / 2;
566                 struct delta_entry *delta = &deltas[next];
567                 int cmp;
568
569                 cmp = compare_delta_bases(base, &delta->base,
570                                           type, objects[delta->obj_no].type);
571                 if (!cmp)
572                         return next;
573                 if (cmp < 0) {
574                         last = next;
575                         continue;
576                 }
577                 first = next+1;
578         }
579         return -first-1;
580 }
581
582 static void find_delta_children(const union delta_base *base,
583                                 int *first_index, int *last_index,
584                                 enum object_type type)
585 {
586         int first = find_delta(base, type);
587         int last = first;
588         int end = nr_deltas - 1;
589
590         if (first < 0) {
591                 *first_index = 0;
592                 *last_index = -1;
593                 return;
594         }
595         while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
596                 --first;
597         while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
598                 ++last;
599         *first_index = first;
600         *last_index = last;
601 }
602
603 static void sha1_object(const void *data, struct object_entry *obj_entry,
604                         unsigned long size, enum object_type type,
605                         const unsigned char *sha1)
606 {
607         void *new_data = NULL;
608
609         assert(data || obj_entry);
610
611         read_lock();
612         if (has_sha1_file(sha1)) {
613                 void *has_data;
614                 enum object_type has_type;
615                 unsigned long has_size;
616                 if (!data)
617                         data = new_data = get_data_from_pack(obj_entry);
618                 has_data = read_sha1_file(sha1, &has_type, &has_size);
619                 read_unlock();
620                 if (!has_data)
621                         die(_("cannot read existing object %s"), sha1_to_hex(sha1));
622                 if (size != has_size || type != has_type ||
623                     memcmp(data, has_data, size) != 0)
624                         die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
625                 free(has_data);
626         } else
627                 read_unlock();
628
629         if (strict) {
630                 read_lock();
631                 if (type == OBJ_BLOB) {
632                         struct blob *blob = lookup_blob(sha1);
633                         if (blob)
634                                 blob->object.flags |= FLAG_CHECKED;
635                         else
636                                 die(_("invalid blob object %s"), sha1_to_hex(sha1));
637                 } else {
638                         struct object *obj;
639                         int eaten;
640                         void *buf = (void *) data;
641
642                         if (!buf)
643                                 buf = new_data = get_data_from_pack(obj_entry);
644
645                         /*
646                          * we do not need to free the memory here, as the
647                          * buf is deleted by the caller.
648                          */
649                         obj = parse_object_buffer(sha1, type, size, buf, &eaten);
650                         if (!obj)
651                                 die(_("invalid %s"), typename(type));
652                         if (fsck_object(obj, 1, fsck_error_function))
653                                 die(_("Error in object"));
654                         if (fsck_walk(obj, mark_link, NULL))
655                                 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
656
657                         if (obj->type == OBJ_TREE) {
658                                 struct tree *item = (struct tree *) obj;
659                                 item->buffer = NULL;
660                         }
661                         if (obj->type == OBJ_COMMIT) {
662                                 struct commit *commit = (struct commit *) obj;
663                                 commit->buffer = NULL;
664                         }
665                         obj->flags |= FLAG_CHECKED;
666                 }
667                 read_unlock();
668         }
669
670         free(new_data);
671 }
672
673 /*
674  * This function is part of find_unresolved_deltas(). There are two
675  * walkers going in the opposite ways.
676  *
677  * The first one in find_unresolved_deltas() traverses down from
678  * parent node to children, deflating nodes along the way. However,
679  * memory for deflated nodes is limited by delta_base_cache_limit, so
680  * at some point parent node's deflated content may be freed.
681  *
682  * The second walker is this function, which goes from current node up
683  * to top parent if necessary to deflate the node. In normal
684  * situation, its parent node would be already deflated, so it just
685  * needs to apply delta.
686  *
687  * In the worst case scenario, parent node is no longer deflated because
688  * we're running out of delta_base_cache_limit; we need to re-deflate
689  * parents, possibly up to the top base.
690  *
691  * All deflated objects here are subject to be freed if we exceed
692  * delta_base_cache_limit, just like in find_unresolved_deltas(), we
693  * just need to make sure the last node is not freed.
694  */
695 static void *get_base_data(struct base_data *c)
696 {
697         if (!c->data) {
698                 struct object_entry *obj = c->obj;
699                 struct base_data **delta = NULL;
700                 int delta_nr = 0, delta_alloc = 0;
701
702                 while (is_delta_type(c->obj->type) && !c->data) {
703                         ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
704                         delta[delta_nr++] = c;
705                         c = c->base;
706                 }
707                 if (!delta_nr) {
708                         c->data = get_data_from_pack(obj);
709                         c->size = obj->size;
710                         get_thread_data()->base_cache_used += c->size;
711                         prune_base_data(c);
712                 }
713                 for (; delta_nr > 0; delta_nr--) {
714                         void *base, *raw;
715                         c = delta[delta_nr - 1];
716                         obj = c->obj;
717                         base = get_base_data(c->base);
718                         raw = get_data_from_pack(obj);
719                         c->data = patch_delta(
720                                 base, c->base->size,
721                                 raw, obj->size,
722                                 &c->size);
723                         free(raw);
724                         if (!c->data)
725                                 bad_object(obj->idx.offset, _("failed to apply delta"));
726                         get_thread_data()->base_cache_used += c->size;
727                         prune_base_data(c);
728                 }
729                 free(delta);
730         }
731         return c->data;
732 }
733
734 static void resolve_delta(struct object_entry *delta_obj,
735                           struct base_data *base, struct base_data *result)
736 {
737         void *base_data, *delta_data;
738
739         delta_obj->real_type = base->obj->real_type;
740         delta_obj->delta_depth = base->obj->delta_depth + 1;
741         if (deepest_delta < delta_obj->delta_depth)
742                 deepest_delta = delta_obj->delta_depth;
743         delta_obj->base_object_no = base->obj - objects;
744         delta_data = get_data_from_pack(delta_obj);
745         base_data = get_base_data(base);
746         result->obj = delta_obj;
747         result->data = patch_delta(base_data, base->size,
748                                    delta_data, delta_obj->size, &result->size);
749         free(delta_data);
750         if (!result->data)
751                 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
752         hash_sha1_file(result->data, result->size,
753                        typename(delta_obj->real_type), delta_obj->idx.sha1);
754         sha1_object(result->data, NULL, result->size, delta_obj->real_type,
755                     delta_obj->idx.sha1);
756         counter_lock();
757         nr_resolved_deltas++;
758         counter_unlock();
759 }
760
761 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
762                                                   struct base_data *prev_base)
763 {
764         if (base->ref_last == -1 && base->ofs_last == -1) {
765                 union delta_base base_spec;
766
767                 hashcpy(base_spec.sha1, base->obj->idx.sha1);
768                 find_delta_children(&base_spec,
769                                     &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
770
771                 memset(&base_spec, 0, sizeof(base_spec));
772                 base_spec.offset = base->obj->idx.offset;
773                 find_delta_children(&base_spec,
774                                     &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
775
776                 if (base->ref_last == -1 && base->ofs_last == -1) {
777                         free(base->data);
778                         return NULL;
779                 }
780
781                 link_base_data(prev_base, base);
782         }
783
784         if (base->ref_first <= base->ref_last) {
785                 struct object_entry *child = objects + deltas[base->ref_first].obj_no;
786                 struct base_data *result = alloc_base_data();
787
788                 assert(child->real_type == OBJ_REF_DELTA);
789                 resolve_delta(child, base, result);
790                 if (base->ref_first == base->ref_last && base->ofs_last == -1)
791                         free_base_data(base);
792
793                 base->ref_first++;
794                 return result;
795         }
796
797         if (base->ofs_first <= base->ofs_last) {
798                 struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
799                 struct base_data *result = alloc_base_data();
800
801                 assert(child->real_type == OBJ_OFS_DELTA);
802                 resolve_delta(child, base, result);
803                 if (base->ofs_first == base->ofs_last)
804                         free_base_data(base);
805
806                 base->ofs_first++;
807                 return result;
808         }
809
810         unlink_base_data(base);
811         return NULL;
812 }
813
814 static void find_unresolved_deltas(struct base_data *base)
815 {
816         struct base_data *new_base, *prev_base = NULL;
817         for (;;) {
818                 new_base = find_unresolved_deltas_1(base, prev_base);
819
820                 if (new_base) {
821                         prev_base = base;
822                         base = new_base;
823                 } else {
824                         free(base);
825                         base = prev_base;
826                         if (!base)
827                                 return;
828                         prev_base = base->base;
829                 }
830         }
831 }
832
833 static int compare_delta_entry(const void *a, const void *b)
834 {
835         const struct delta_entry *delta_a = a;
836         const struct delta_entry *delta_b = b;
837
838         /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
839         return compare_delta_bases(&delta_a->base, &delta_b->base,
840                                    objects[delta_a->obj_no].type,
841                                    objects[delta_b->obj_no].type);
842 }
843
844 static void resolve_base(struct object_entry *obj)
845 {
846         struct base_data *base_obj = alloc_base_data();
847         base_obj->obj = obj;
848         base_obj->data = NULL;
849         find_unresolved_deltas(base_obj);
850 }
851
852 #ifndef NO_PTHREADS
853 static void *threaded_second_pass(void *data)
854 {
855         set_thread_data(data);
856         for (;;) {
857                 int i;
858                 work_lock();
859                 display_progress(progress, nr_resolved_deltas);
860                 while (nr_dispatched < nr_objects &&
861                        is_delta_type(objects[nr_dispatched].type))
862                         nr_dispatched++;
863                 if (nr_dispatched >= nr_objects) {
864                         work_unlock();
865                         break;
866                 }
867                 i = nr_dispatched++;
868                 work_unlock();
869
870                 resolve_base(&objects[i]);
871         }
872         return NULL;
873 }
874 #endif
875
876 /*
877  * First pass:
878  * - find locations of all objects;
879  * - calculate SHA1 of all non-delta objects;
880  * - remember base (SHA1 or offset) for all deltas.
881  */
882 static void parse_pack_objects(unsigned char *sha1)
883 {
884         int i, nr_delays = 0;
885         struct delta_entry *delta = deltas;
886         struct stat st;
887
888         if (verbose)
889                 progress = start_progress(
890                                 from_stdin ? _("Receiving objects") : _("Indexing objects"),
891                                 nr_objects);
892         for (i = 0; i < nr_objects; i++) {
893                 struct object_entry *obj = &objects[i];
894                 void *data = unpack_raw_entry(obj, &delta->base, obj->idx.sha1);
895                 obj->real_type = obj->type;
896                 if (is_delta_type(obj->type)) {
897                         nr_deltas++;
898                         delta->obj_no = i;
899                         delta++;
900                 } else if (!data) {
901                         /* large blobs, check later */
902                         obj->real_type = OBJ_BAD;
903                         nr_delays++;
904                 } else
905                         sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
906                 free(data);
907                 display_progress(progress, i+1);
908         }
909         objects[i].idx.offset = consumed_bytes;
910         stop_progress(&progress);
911
912         /* Check pack integrity */
913         flush();
914         git_SHA1_Final(sha1, &input_ctx);
915         if (hashcmp(fill(20), sha1))
916                 die(_("pack is corrupted (SHA1 mismatch)"));
917         use(20);
918
919         /* If input_fd is a file, we should have reached its end now. */
920         if (fstat(input_fd, &st))
921                 die_errno(_("cannot fstat packfile"));
922         if (S_ISREG(st.st_mode) &&
923                         lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
924                 die(_("pack has junk at the end"));
925
926         for (i = 0; i < nr_objects; i++) {
927                 struct object_entry *obj = &objects[i];
928                 if (obj->real_type != OBJ_BAD)
929                         continue;
930                 obj->real_type = obj->type;
931                 sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
932                 nr_delays--;
933         }
934         if (nr_delays)
935                 die(_("confusion beyond insanity in parse_pack_objects()"));
936 }
937
938 /*
939  * Second pass:
940  * - for all non-delta objects, look if it is used as a base for
941  *   deltas;
942  * - if used as a base, uncompress the object and apply all deltas,
943  *   recursively checking if the resulting object is used as a base
944  *   for some more deltas.
945  */
946 static void resolve_deltas(void)
947 {
948         int i;
949
950         if (!nr_deltas)
951                 return;
952
953         /* Sort deltas by base SHA1/offset for fast searching */
954         qsort(deltas, nr_deltas, sizeof(struct delta_entry),
955               compare_delta_entry);
956
957         if (verbose)
958                 progress = start_progress(_("Resolving deltas"), nr_deltas);
959
960 #ifndef NO_PTHREADS
961         nr_dispatched = 0;
962         if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
963                 init_thread();
964                 for (i = 0; i < nr_threads; i++) {
965                         int ret = pthread_create(&thread_data[i].thread, NULL,
966                                                  threaded_second_pass, thread_data + i);
967                         if (ret)
968                                 die("unable to create thread: %s", strerror(ret));
969                 }
970                 for (i = 0; i < nr_threads; i++)
971                         pthread_join(thread_data[i].thread, NULL);
972                 cleanup_thread();
973                 return;
974         }
975 #endif
976
977         for (i = 0; i < nr_objects; i++) {
978                 struct object_entry *obj = &objects[i];
979
980                 if (is_delta_type(obj->type))
981                         continue;
982                 resolve_base(obj);
983                 display_progress(progress, nr_resolved_deltas);
984         }
985 }
986
987 /*
988  * Third pass:
989  * - append objects to convert thin pack to full pack if required
990  * - write the final 20-byte SHA-1
991  */
992 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
993 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
994 {
995         if (nr_deltas == nr_resolved_deltas) {
996                 stop_progress(&progress);
997                 /* Flush remaining pack final 20-byte SHA1. */
998                 flush();
999                 return;
1000         }
1001
1002         if (fix_thin_pack) {
1003                 struct sha1file *f;
1004                 unsigned char read_sha1[20], tail_sha1[20];
1005                 char msg[48];
1006                 int nr_unresolved = nr_deltas - nr_resolved_deltas;
1007                 int nr_objects_initial = nr_objects;
1008                 if (nr_unresolved <= 0)
1009                         die(_("confusion beyond insanity"));
1010                 objects = xrealloc(objects,
1011                                    (nr_objects + nr_unresolved + 1)
1012                                    * sizeof(*objects));
1013                 f = sha1fd(output_fd, curr_pack);
1014                 fix_unresolved_deltas(f, nr_unresolved);
1015                 sprintf(msg, "completed with %d local objects",
1016                         nr_objects - nr_objects_initial);
1017                 stop_progress_msg(&progress, msg);
1018                 sha1close(f, tail_sha1, 0);
1019                 hashcpy(read_sha1, pack_sha1);
1020                 fixup_pack_header_footer(output_fd, pack_sha1,
1021                                          curr_pack, nr_objects,
1022                                          read_sha1, consumed_bytes-20);
1023                 if (hashcmp(read_sha1, tail_sha1) != 0)
1024                         die("Unexpected tail checksum for %s "
1025                             "(disk corruption?)", curr_pack);
1026         }
1027         if (nr_deltas != nr_resolved_deltas)
1028                 die(Q_("pack has %d unresolved delta",
1029                        "pack has %d unresolved deltas",
1030                        nr_deltas - nr_resolved_deltas),
1031                     nr_deltas - nr_resolved_deltas);
1032 }
1033
1034 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1035 {
1036         git_zstream stream;
1037         int status;
1038         unsigned char outbuf[4096];
1039
1040         memset(&stream, 0, sizeof(stream));
1041         git_deflate_init(&stream, zlib_compression_level);
1042         stream.next_in = in;
1043         stream.avail_in = size;
1044
1045         do {
1046                 stream.next_out = outbuf;
1047                 stream.avail_out = sizeof(outbuf);
1048                 status = git_deflate(&stream, Z_FINISH);
1049                 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1050         } while (status == Z_OK);
1051
1052         if (status != Z_STREAM_END)
1053                 die(_("unable to deflate appended object (%d)"), status);
1054         size = stream.total_out;
1055         git_deflate_end(&stream);
1056         return size;
1057 }
1058
1059 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1060                                const unsigned char *sha1, void *buf,
1061                                unsigned long size, enum object_type type)
1062 {
1063         struct object_entry *obj = &objects[nr_objects++];
1064         unsigned char header[10];
1065         unsigned long s = size;
1066         int n = 0;
1067         unsigned char c = (type << 4) | (s & 15);
1068         s >>= 4;
1069         while (s) {
1070                 header[n++] = c | 0x80;
1071                 c = s & 0x7f;
1072                 s >>= 7;
1073         }
1074         header[n++] = c;
1075         crc32_begin(f);
1076         sha1write(f, header, n);
1077         obj[0].size = size;
1078         obj[0].hdr_size = n;
1079         obj[0].type = type;
1080         obj[0].real_type = type;
1081         obj[1].idx.offset = obj[0].idx.offset + n;
1082         obj[1].idx.offset += write_compressed(f, buf, size);
1083         obj[0].idx.crc32 = crc32_end(f);
1084         sha1flush(f);
1085         hashcpy(obj->idx.sha1, sha1);
1086         return obj;
1087 }
1088
1089 static int delta_pos_compare(const void *_a, const void *_b)
1090 {
1091         struct delta_entry *a = *(struct delta_entry **)_a;
1092         struct delta_entry *b = *(struct delta_entry **)_b;
1093         return a->obj_no - b->obj_no;
1094 }
1095
1096 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
1097 {
1098         struct delta_entry **sorted_by_pos;
1099         int i, n = 0;
1100
1101         /*
1102          * Since many unresolved deltas may well be themselves base objects
1103          * for more unresolved deltas, we really want to include the
1104          * smallest number of base objects that would cover as much delta
1105          * as possible by picking the
1106          * trunc deltas first, allowing for other deltas to resolve without
1107          * additional base objects.  Since most base objects are to be found
1108          * before deltas depending on them, a good heuristic is to start
1109          * resolving deltas in the same order as their position in the pack.
1110          */
1111         sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
1112         for (i = 0; i < nr_deltas; i++) {
1113                 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1114                         continue;
1115                 sorted_by_pos[n++] = &deltas[i];
1116         }
1117         qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1118
1119         for (i = 0; i < n; i++) {
1120                 struct delta_entry *d = sorted_by_pos[i];
1121                 enum object_type type;
1122                 struct base_data *base_obj = alloc_base_data();
1123
1124                 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1125                         continue;
1126                 base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1127                 if (!base_obj->data)
1128                         continue;
1129
1130                 if (check_sha1_signature(d->base.sha1, base_obj->data,
1131                                 base_obj->size, typename(type)))
1132                         die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
1133                 base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1134                                         base_obj->data, base_obj->size, type);
1135                 find_unresolved_deltas(base_obj);
1136                 display_progress(progress, nr_resolved_deltas);
1137         }
1138         free(sorted_by_pos);
1139 }
1140
1141 static void final(const char *final_pack_name, const char *curr_pack_name,
1142                   const char *final_index_name, const char *curr_index_name,
1143                   const char *keep_name, const char *keep_msg,
1144                   unsigned char *sha1)
1145 {
1146         const char *report = "pack";
1147         char name[PATH_MAX];
1148         int err;
1149
1150         if (!from_stdin) {
1151                 close(input_fd);
1152         } else {
1153                 fsync_or_die(output_fd, curr_pack_name);
1154                 err = close(output_fd);
1155                 if (err)
1156                         die_errno(_("error while closing pack file"));
1157         }
1158
1159         if (keep_msg) {
1160                 int keep_fd, keep_msg_len = strlen(keep_msg);
1161
1162                 if (!keep_name)
1163                         keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1164                 else
1165                         keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1166
1167                 if (keep_fd < 0) {
1168                         if (errno != EEXIST)
1169                                 die_errno(_("cannot write keep file '%s'"),
1170                                           keep_name);
1171                 } else {
1172                         if (keep_msg_len > 0) {
1173                                 write_or_die(keep_fd, keep_msg, keep_msg_len);
1174                                 write_or_die(keep_fd, "\n", 1);
1175                         }
1176                         if (close(keep_fd) != 0)
1177                                 die_errno(_("cannot close written keep file '%s'"),
1178                                     keep_name);
1179                         report = "keep";
1180                 }
1181         }
1182
1183         if (final_pack_name != curr_pack_name) {
1184                 if (!final_pack_name) {
1185                         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1186                                  get_object_directory(), sha1_to_hex(sha1));
1187                         final_pack_name = name;
1188                 }
1189                 if (move_temp_to_file(curr_pack_name, final_pack_name))
1190                         die(_("cannot store pack file"));
1191         } else if (from_stdin)
1192                 chmod(final_pack_name, 0444);
1193
1194         if (final_index_name != curr_index_name) {
1195                 if (!final_index_name) {
1196                         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1197                                  get_object_directory(), sha1_to_hex(sha1));
1198                         final_index_name = name;
1199                 }
1200                 if (move_temp_to_file(curr_index_name, final_index_name))
1201                         die(_("cannot store index file"));
1202         } else
1203                 chmod(final_index_name, 0444);
1204
1205         if (!from_stdin) {
1206                 printf("%s\n", sha1_to_hex(sha1));
1207         } else {
1208                 char buf[48];
1209                 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1210                                    report, sha1_to_hex(sha1));
1211                 write_or_die(1, buf, len);
1212
1213                 /*
1214                  * Let's just mimic git-unpack-objects here and write
1215                  * the last part of the input buffer to stdout.
1216                  */
1217                 while (input_len) {
1218                         err = xwrite(1, input_buffer + input_offset, input_len);
1219                         if (err <= 0)
1220                                 break;
1221                         input_len -= err;
1222                         input_offset += err;
1223                 }
1224         }
1225 }
1226
1227 static int git_index_pack_config(const char *k, const char *v, void *cb)
1228 {
1229         struct pack_idx_option *opts = cb;
1230
1231         if (!strcmp(k, "pack.indexversion")) {
1232                 opts->version = git_config_int(k, v);
1233                 if (opts->version > 2)
1234                         die("bad pack.indexversion=%"PRIu32, opts->version);
1235                 return 0;
1236         }
1237         if (!strcmp(k, "pack.threads")) {
1238                 nr_threads = git_config_int(k, v);
1239                 if (nr_threads < 0)
1240                         die("invalid number of threads specified (%d)",
1241                             nr_threads);
1242 #ifdef NO_PTHREADS
1243                 if (nr_threads != 1)
1244                         warning("no threads support, ignoring %s", k);
1245                 nr_threads = 1;
1246 #endif
1247                 return 0;
1248         }
1249         return git_default_config(k, v, cb);
1250 }
1251
1252 static int cmp_uint32(const void *a_, const void *b_)
1253 {
1254         uint32_t a = *((uint32_t *)a_);
1255         uint32_t b = *((uint32_t *)b_);
1256
1257         return (a < b) ? -1 : (a != b);
1258 }
1259
1260 static void read_v2_anomalous_offsets(struct packed_git *p,
1261                                       struct pack_idx_option *opts)
1262 {
1263         const uint32_t *idx1, *idx2;
1264         uint32_t i;
1265
1266         /* The address of the 4-byte offset table */
1267         idx1 = (((const uint32_t *)p->index_data)
1268                 + 2 /* 8-byte header */
1269                 + 256 /* fan out */
1270                 + 5 * p->num_objects /* 20-byte SHA-1 table */
1271                 + p->num_objects /* CRC32 table */
1272                 );
1273
1274         /* The address of the 8-byte offset table */
1275         idx2 = idx1 + p->num_objects;
1276
1277         for (i = 0; i < p->num_objects; i++) {
1278                 uint32_t off = ntohl(idx1[i]);
1279                 if (!(off & 0x80000000))
1280                         continue;
1281                 off = off & 0x7fffffff;
1282                 if (idx2[off * 2])
1283                         continue;
1284                 /*
1285                  * The real offset is ntohl(idx2[off * 2]) in high 4
1286                  * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1287                  * octets.  But idx2[off * 2] is Zero!!!
1288                  */
1289                 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1290                 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1291         }
1292
1293         if (1 < opts->anomaly_nr)
1294                 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1295 }
1296
1297 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1298 {
1299         struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1300
1301         if (!p)
1302                 die(_("Cannot open existing pack file '%s'"), pack_name);
1303         if (open_pack_index(p))
1304                 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1305
1306         /* Read the attributes from the existing idx file */
1307         opts->version = p->index_version;
1308
1309         if (opts->version == 2)
1310                 read_v2_anomalous_offsets(p, opts);
1311
1312         /*
1313          * Get rid of the idx file as we do not need it anymore.
1314          * NEEDSWORK: extract this bit from free_pack_by_name() in
1315          * sha1_file.c, perhaps?  It shouldn't matter very much as we
1316          * know we haven't installed this pack (hence we never have
1317          * read anything from it).
1318          */
1319         close_pack_index(p);
1320         free(p);
1321 }
1322
1323 static void show_pack_info(int stat_only)
1324 {
1325         int i, baseobjects = nr_objects - nr_deltas;
1326         unsigned long *chain_histogram = NULL;
1327
1328         if (deepest_delta)
1329                 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1330
1331         for (i = 0; i < nr_objects; i++) {
1332                 struct object_entry *obj = &objects[i];
1333
1334                 if (is_delta_type(obj->type))
1335                         chain_histogram[obj->delta_depth - 1]++;
1336                 if (stat_only)
1337                         continue;
1338                 printf("%s %-6s %lu %lu %"PRIuMAX,
1339                        sha1_to_hex(obj->idx.sha1),
1340                        typename(obj->real_type), obj->size,
1341                        (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1342                        (uintmax_t)obj->idx.offset);
1343                 if (is_delta_type(obj->type)) {
1344                         struct object_entry *bobj = &objects[obj->base_object_no];
1345                         printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1346                 }
1347                 putchar('\n');
1348         }
1349
1350         if (baseobjects)
1351                 printf_ln(Q_("non delta: %d object",
1352                              "non delta: %d objects",
1353                              baseobjects),
1354                           baseobjects);
1355         for (i = 0; i < deepest_delta; i++) {
1356                 if (!chain_histogram[i])
1357                         continue;
1358                 printf_ln(Q_("chain length = %d: %lu object",
1359                              "chain length = %d: %lu objects",
1360                              chain_histogram[i]),
1361                           i + 1,
1362                           chain_histogram[i]);
1363         }
1364 }
1365
1366 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1367 {
1368         int i, fix_thin_pack = 0, verify = 0, stat_only = 0, stat = 0;
1369         const char *curr_pack, *curr_index;
1370         const char *index_name = NULL, *pack_name = NULL;
1371         const char *keep_name = NULL, *keep_msg = NULL;
1372         char *index_name_buf = NULL, *keep_name_buf = NULL;
1373         struct pack_idx_entry **idx_objects;
1374         struct pack_idx_option opts;
1375         unsigned char pack_sha1[20];
1376
1377         if (argc == 2 && !strcmp(argv[1], "-h"))
1378                 usage(index_pack_usage);
1379
1380         read_replace_refs = 0;
1381
1382         reset_pack_idx_option(&opts);
1383         git_config(git_index_pack_config, &opts);
1384         if (prefix && chdir(prefix))
1385                 die(_("Cannot come back to cwd"));
1386
1387         for (i = 1; i < argc; i++) {
1388                 const char *arg = argv[i];
1389
1390                 if (*arg == '-') {
1391                         if (!strcmp(arg, "--stdin")) {
1392                                 from_stdin = 1;
1393                         } else if (!strcmp(arg, "--fix-thin")) {
1394                                 fix_thin_pack = 1;
1395                         } else if (!strcmp(arg, "--strict")) {
1396                                 strict = 1;
1397                         } else if (!strcmp(arg, "--verify")) {
1398                                 verify = 1;
1399                         } else if (!strcmp(arg, "--verify-stat")) {
1400                                 verify = 1;
1401                                 stat = 1;
1402                         } else if (!strcmp(arg, "--verify-stat-only")) {
1403                                 verify = 1;
1404                                 stat = 1;
1405                                 stat_only = 1;
1406                         } else if (!strcmp(arg, "--keep")) {
1407                                 keep_msg = "";
1408                         } else if (!prefixcmp(arg, "--keep=")) {
1409                                 keep_msg = arg + 7;
1410                         } else if (!prefixcmp(arg, "--threads=")) {
1411                                 char *end;
1412                                 nr_threads = strtoul(arg+10, &end, 0);
1413                                 if (!arg[10] || *end || nr_threads < 0)
1414                                         usage(index_pack_usage);
1415 #ifdef NO_PTHREADS
1416                                 if (nr_threads != 1)
1417                                         warning("no threads support, "
1418                                                 "ignoring %s", arg);
1419                                 nr_threads = 1;
1420 #endif
1421                         } else if (!prefixcmp(arg, "--pack_header=")) {
1422                                 struct pack_header *hdr;
1423                                 char *c;
1424
1425                                 hdr = (struct pack_header *)input_buffer;
1426                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1427                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1428                                 if (*c != ',')
1429                                         die(_("bad %s"), arg);
1430                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1431                                 if (*c)
1432                                         die(_("bad %s"), arg);
1433                                 input_len = sizeof(*hdr);
1434                         } else if (!strcmp(arg, "-v")) {
1435                                 verbose = 1;
1436                         } else if (!strcmp(arg, "-o")) {
1437                                 if (index_name || (i+1) >= argc)
1438                                         usage(index_pack_usage);
1439                                 index_name = argv[++i];
1440                         } else if (!prefixcmp(arg, "--index-version=")) {
1441                                 char *c;
1442                                 opts.version = strtoul(arg + 16, &c, 10);
1443                                 if (opts.version > 2)
1444                                         die(_("bad %s"), arg);
1445                                 if (*c == ',')
1446                                         opts.off32_limit = strtoul(c+1, &c, 0);
1447                                 if (*c || opts.off32_limit & 0x80000000)
1448                                         die(_("bad %s"), arg);
1449                         } else
1450                                 usage(index_pack_usage);
1451                         continue;
1452                 }
1453
1454                 if (pack_name)
1455                         usage(index_pack_usage);
1456                 pack_name = arg;
1457         }
1458
1459         if (!pack_name && !from_stdin)
1460                 usage(index_pack_usage);
1461         if (fix_thin_pack && !from_stdin)
1462                 die(_("--fix-thin cannot be used without --stdin"));
1463         if (!index_name && pack_name) {
1464                 int len = strlen(pack_name);
1465                 if (!has_extension(pack_name, ".pack"))
1466                         die(_("packfile name '%s' does not end with '.pack'"),
1467                             pack_name);
1468                 index_name_buf = xmalloc(len);
1469                 memcpy(index_name_buf, pack_name, len - 5);
1470                 strcpy(index_name_buf + len - 5, ".idx");
1471                 index_name = index_name_buf;
1472         }
1473         if (keep_msg && !keep_name && pack_name) {
1474                 int len = strlen(pack_name);
1475                 if (!has_extension(pack_name, ".pack"))
1476                         die(_("packfile name '%s' does not end with '.pack'"),
1477                             pack_name);
1478                 keep_name_buf = xmalloc(len);
1479                 memcpy(keep_name_buf, pack_name, len - 5);
1480                 strcpy(keep_name_buf + len - 5, ".keep");
1481                 keep_name = keep_name_buf;
1482         }
1483         if (verify) {
1484                 if (!index_name)
1485                         die(_("--verify with no packfile name given"));
1486                 read_idx_option(&opts, index_name);
1487                 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1488         }
1489         if (strict)
1490                 opts.flags |= WRITE_IDX_STRICT;
1491
1492 #ifndef NO_PTHREADS
1493         if (!nr_threads) {
1494                 nr_threads = online_cpus();
1495                 /* An experiment showed that more threads does not mean faster */
1496                 if (nr_threads > 3)
1497                         nr_threads = 3;
1498         }
1499 #endif
1500
1501         curr_pack = open_pack_file(pack_name);
1502         parse_pack_header();
1503         objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1504         deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1505         parse_pack_objects(pack_sha1);
1506         resolve_deltas();
1507         conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1508         free(deltas);
1509         if (strict)
1510                 check_objects();
1511
1512         if (stat)
1513                 show_pack_info(stat_only);
1514
1515         idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1516         for (i = 0; i < nr_objects; i++)
1517                 idx_objects[i] = &objects[i].idx;
1518         curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1519         free(idx_objects);
1520
1521         if (!verify)
1522                 final(pack_name, curr_pack,
1523                       index_name, curr_index,
1524                       keep_name, keep_msg,
1525                       pack_sha1);
1526         else
1527                 close(input_fd);
1528         free(objects);
1529         free(index_name_buf);
1530         free(keep_name_buf);
1531         if (pack_name == NULL)
1532                 free((void *) curr_pack);
1533         if (index_name == NULL)
1534                 free((void *) curr_index);
1535
1536         return 0;
1537 }