Merge branch 'jk/test-annoyances'
[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 "streaming.h"
13 #include "thread-utils.h"
14
15 static const char index_pack_usage[] =
16 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
17
18 struct object_entry {
19         struct pack_idx_entry idx;
20         unsigned long size;
21         unsigned int hdr_size;
22         enum object_type type;
23         enum object_type real_type;
24         unsigned delta_depth;
25         int base_object_no;
26 };
27
28 union delta_base {
29         unsigned char sha1[20];
30         off_t offset;
31 };
32
33 struct base_data {
34         struct base_data *base;
35         struct base_data *child;
36         struct object_entry *obj;
37         void *data;
38         unsigned long size;
39         int ref_first, ref_last;
40         int ofs_first, ofs_last;
41 };
42
43 struct thread_local {
44 #ifndef NO_PTHREADS
45         pthread_t thread;
46 #endif
47         struct base_data *base_cache;
48         size_t base_cache_used;
49         int pack_fd;
50 };
51
52 /*
53  * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
54  * to memcmp() only the first 20 bytes.
55  */
56 #define UNION_BASE_SZ   20
57
58 #define FLAG_LINK (1u<<20)
59 #define FLAG_CHECKED (1u<<21)
60
61 struct delta_entry {
62         union delta_base base;
63         int obj_no;
64 };
65
66 static struct object_entry *objects;
67 static struct delta_entry *deltas;
68 static struct thread_local nothread_data;
69 static int nr_objects;
70 static int nr_deltas;
71 static int nr_resolved_deltas;
72 static int nr_threads;
73
74 static int from_stdin;
75 static int strict;
76 static int do_fsck_object;
77 static int verbose;
78 static int show_stat;
79 static int check_self_contained_and_connected;
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;
91 static const char *curr_pack;
92
93 #ifndef NO_PTHREADS
94
95 static struct thread_local *thread_data;
96 static int nr_dispatched;
97 static int threads_active;
98
99 static pthread_mutex_t read_mutex;
100 #define read_lock()             lock_mutex(&read_mutex)
101 #define read_unlock()           unlock_mutex(&read_mutex)
102
103 static pthread_mutex_t counter_mutex;
104 #define counter_lock()          lock_mutex(&counter_mutex)
105 #define counter_unlock()        unlock_mutex(&counter_mutex)
106
107 static pthread_mutex_t work_mutex;
108 #define work_lock()             lock_mutex(&work_mutex)
109 #define work_unlock()           unlock_mutex(&work_mutex)
110
111 static pthread_mutex_t deepest_delta_mutex;
112 #define deepest_delta_lock()    lock_mutex(&deepest_delta_mutex)
113 #define deepest_delta_unlock()  unlock_mutex(&deepest_delta_mutex)
114
115 static pthread_mutex_t type_cas_mutex;
116 #define type_cas_lock()         lock_mutex(&type_cas_mutex)
117 #define type_cas_unlock()       unlock_mutex(&type_cas_mutex)
118
119 static pthread_key_t key;
120
121 static inline void lock_mutex(pthread_mutex_t *mutex)
122 {
123         if (threads_active)
124                 pthread_mutex_lock(mutex);
125 }
126
127 static inline void unlock_mutex(pthread_mutex_t *mutex)
128 {
129         if (threads_active)
130                 pthread_mutex_unlock(mutex);
131 }
132
133 /*
134  * Mutex and conditional variable can't be statically-initialized on Windows.
135  */
136 static void init_thread(void)
137 {
138         int i;
139         init_recursive_mutex(&read_mutex);
140         pthread_mutex_init(&counter_mutex, NULL);
141         pthread_mutex_init(&work_mutex, NULL);
142         pthread_mutex_init(&type_cas_mutex, NULL);
143         if (show_stat)
144                 pthread_mutex_init(&deepest_delta_mutex, NULL);
145         pthread_key_create(&key, NULL);
146         thread_data = xcalloc(nr_threads, sizeof(*thread_data));
147         for (i = 0; i < nr_threads; i++) {
148                 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
149                 if (thread_data[i].pack_fd == -1)
150                         die_errno(_("unable to open %s"), curr_pack);
151         }
152
153         threads_active = 1;
154 }
155
156 static void cleanup_thread(void)
157 {
158         int i;
159         if (!threads_active)
160                 return;
161         threads_active = 0;
162         pthread_mutex_destroy(&read_mutex);
163         pthread_mutex_destroy(&counter_mutex);
164         pthread_mutex_destroy(&work_mutex);
165         pthread_mutex_destroy(&type_cas_mutex);
166         if (show_stat)
167                 pthread_mutex_destroy(&deepest_delta_mutex);
168         for (i = 0; i < nr_threads; i++)
169                 close(thread_data[i].pack_fd);
170         pthread_key_delete(key);
171         free(thread_data);
172 }
173
174 #else
175
176 #define read_lock()
177 #define read_unlock()
178
179 #define counter_lock()
180 #define counter_unlock()
181
182 #define work_lock()
183 #define work_unlock()
184
185 #define deepest_delta_lock()
186 #define deepest_delta_unlock()
187
188 #define type_cas_lock()
189 #define type_cas_unlock()
190
191 #endif
192
193
194 static int mark_link(struct object *obj, int type, void *data)
195 {
196         if (!obj)
197                 return -1;
198
199         if (type != OBJ_ANY && obj->type != type)
200                 die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
201
202         obj->flags |= FLAG_LINK;
203         return 0;
204 }
205
206 /* The content of each linked object must have been checked
207    or it must be already present in the object database */
208 static unsigned check_object(struct object *obj)
209 {
210         if (!obj)
211                 return 0;
212
213         if (!(obj->flags & FLAG_LINK))
214                 return 0;
215
216         if (!(obj->flags & FLAG_CHECKED)) {
217                 unsigned long size;
218                 int type = sha1_object_info(obj->sha1, &size);
219                 if (type <= 0)
220                         die(_("did not receive expected object %s"),
221                               sha1_to_hex(obj->sha1));
222                 if (type != obj->type)
223                         die(_("object %s: expected type %s, found %s"),
224                             sha1_to_hex(obj->sha1),
225                             typename(obj->type), typename(type));
226                 obj->flags |= FLAG_CHECKED;
227                 return 1;
228         }
229
230         return 0;
231 }
232
233 static unsigned check_objects(void)
234 {
235         unsigned i, max, foreign_nr = 0;
236
237         max = get_max_object_index();
238         for (i = 0; i < max; i++)
239                 foreign_nr += check_object(get_indexed_object(i));
240         return foreign_nr;
241 }
242
243
244 /* Discard current buffer used content. */
245 static void flush(void)
246 {
247         if (input_offset) {
248                 if (output_fd >= 0)
249                         write_or_die(output_fd, input_buffer, input_offset);
250                 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
251                 memmove(input_buffer, input_buffer + input_offset, input_len);
252                 input_offset = 0;
253         }
254 }
255
256 /*
257  * Make sure at least "min" bytes are available in the buffer, and
258  * return the pointer to the buffer.
259  */
260 static void *fill(int min)
261 {
262         if (min <= input_len)
263                 return input_buffer + input_offset;
264         if (min > sizeof(input_buffer))
265                 die(Q_("cannot fill %d byte",
266                        "cannot fill %d bytes",
267                        min),
268                     min);
269         flush();
270         do {
271                 ssize_t ret = xread(input_fd, input_buffer + input_len,
272                                 sizeof(input_buffer) - input_len);
273                 if (ret <= 0) {
274                         if (!ret)
275                                 die(_("early EOF"));
276                         die_errno(_("read error on input"));
277                 }
278                 input_len += ret;
279                 if (from_stdin)
280                         display_throughput(progress, consumed_bytes + input_len);
281         } while (input_len < min);
282         return input_buffer;
283 }
284
285 static void use(int bytes)
286 {
287         if (bytes > input_len)
288                 die(_("used more bytes than were available"));
289         input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
290         input_len -= bytes;
291         input_offset += bytes;
292
293         /* make sure off_t is sufficiently large not to wrap */
294         if (signed_add_overflows(consumed_bytes, bytes))
295                 die(_("pack too large for current definition of off_t"));
296         consumed_bytes += bytes;
297 }
298
299 static const char *open_pack_file(const char *pack_name)
300 {
301         if (from_stdin) {
302                 input_fd = 0;
303                 if (!pack_name) {
304                         static char tmp_file[PATH_MAX];
305                         output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
306                                                 "pack/tmp_pack_XXXXXX");
307                         pack_name = xstrdup(tmp_file);
308                 } else
309                         output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
310                 if (output_fd < 0)
311                         die_errno(_("unable to create '%s'"), pack_name);
312                 nothread_data.pack_fd = output_fd;
313         } else {
314                 input_fd = open(pack_name, O_RDONLY);
315                 if (input_fd < 0)
316                         die_errno(_("cannot open packfile '%s'"), pack_name);
317                 output_fd = -1;
318                 nothread_data.pack_fd = input_fd;
319         }
320         git_SHA1_Init(&input_ctx);
321         return pack_name;
322 }
323
324 static void parse_pack_header(void)
325 {
326         struct pack_header *hdr = fill(sizeof(struct pack_header));
327
328         /* Header consistency check */
329         if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
330                 die(_("pack signature mismatch"));
331         if (!pack_version_ok(hdr->hdr_version))
332                 die(_("pack version %"PRIu32" unsupported"),
333                         ntohl(hdr->hdr_version));
334
335         nr_objects = ntohl(hdr->hdr_entries);
336         use(sizeof(struct pack_header));
337 }
338
339 static NORETURN void bad_object(unsigned long offset, const char *format,
340                        ...) __attribute__((format (printf, 2, 3)));
341
342 static NORETURN void bad_object(unsigned long offset, const char *format, ...)
343 {
344         va_list params;
345         char buf[1024];
346
347         va_start(params, format);
348         vsnprintf(buf, sizeof(buf), format, params);
349         va_end(params);
350         die(_("pack has bad object at offset %lu: %s"), offset, buf);
351 }
352
353 static inline struct thread_local *get_thread_data(void)
354 {
355 #ifndef NO_PTHREADS
356         if (threads_active)
357                 return pthread_getspecific(key);
358         assert(!threads_active &&
359                "This should only be reached when all threads are gone");
360 #endif
361         return &nothread_data;
362 }
363
364 #ifndef NO_PTHREADS
365 static void set_thread_data(struct thread_local *data)
366 {
367         if (threads_active)
368                 pthread_setspecific(key, data);
369 }
370 #endif
371
372 static struct base_data *alloc_base_data(void)
373 {
374         struct base_data *base = xcalloc(1, sizeof(struct base_data));
375         base->ref_last = -1;
376         base->ofs_last = -1;
377         return base;
378 }
379
380 static void free_base_data(struct base_data *c)
381 {
382         if (c->data) {
383                 free(c->data);
384                 c->data = NULL;
385                 get_thread_data()->base_cache_used -= c->size;
386         }
387 }
388
389 static void prune_base_data(struct base_data *retain)
390 {
391         struct base_data *b;
392         struct thread_local *data = get_thread_data();
393         for (b = data->base_cache;
394              data->base_cache_used > delta_base_cache_limit && b;
395              b = b->child) {
396                 if (b->data && b != retain)
397                         free_base_data(b);
398         }
399 }
400
401 static void link_base_data(struct base_data *base, struct base_data *c)
402 {
403         if (base)
404                 base->child = c;
405         else
406                 get_thread_data()->base_cache = c;
407
408         c->base = base;
409         c->child = NULL;
410         if (c->data)
411                 get_thread_data()->base_cache_used += c->size;
412         prune_base_data(c);
413 }
414
415 static void unlink_base_data(struct base_data *c)
416 {
417         struct base_data *base = c->base;
418         if (base)
419                 base->child = NULL;
420         else
421                 get_thread_data()->base_cache = NULL;
422         free_base_data(c);
423 }
424
425 static int is_delta_type(enum object_type type)
426 {
427         return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
428 }
429
430 static void *unpack_entry_data(unsigned long offset, unsigned long size,
431                                enum object_type type, unsigned char *sha1)
432 {
433         static char fixed_buf[8192];
434         int status;
435         git_zstream stream;
436         void *buf;
437         git_SHA_CTX c;
438         char hdr[32];
439         int hdrlen;
440
441         if (!is_delta_type(type)) {
442                 hdrlen = sprintf(hdr, "%s %lu", typename(type), size) + 1;
443                 git_SHA1_Init(&c);
444                 git_SHA1_Update(&c, hdr, hdrlen);
445         } else
446                 sha1 = NULL;
447         if (type == OBJ_BLOB && size > big_file_threshold)
448                 buf = fixed_buf;
449         else
450                 buf = xmallocz(size);
451
452         memset(&stream, 0, sizeof(stream));
453         git_inflate_init(&stream);
454         stream.next_out = buf;
455         stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
456
457         do {
458                 unsigned char *last_out = stream.next_out;
459                 stream.next_in = fill(1);
460                 stream.avail_in = input_len;
461                 status = git_inflate(&stream, 0);
462                 use(input_len - stream.avail_in);
463                 if (sha1)
464                         git_SHA1_Update(&c, last_out, stream.next_out - last_out);
465                 if (buf == fixed_buf) {
466                         stream.next_out = buf;
467                         stream.avail_out = sizeof(fixed_buf);
468                 }
469         } while (status == Z_OK);
470         if (stream.total_out != size || status != Z_STREAM_END)
471                 bad_object(offset, _("inflate returned %d"), status);
472         git_inflate_end(&stream);
473         if (sha1)
474                 git_SHA1_Final(sha1, &c);
475         return buf == fixed_buf ? NULL : buf;
476 }
477
478 static void *unpack_raw_entry(struct object_entry *obj,
479                               union delta_base *delta_base,
480                               unsigned char *sha1)
481 {
482         unsigned char *p;
483         unsigned long size, c;
484         off_t base_offset;
485         unsigned shift;
486         void *data;
487
488         obj->idx.offset = consumed_bytes;
489         input_crc32 = crc32(0, NULL, 0);
490
491         p = fill(1);
492         c = *p;
493         use(1);
494         obj->type = (c >> 4) & 7;
495         size = (c & 15);
496         shift = 4;
497         while (c & 0x80) {
498                 p = fill(1);
499                 c = *p;
500                 use(1);
501                 size += (c & 0x7f) << shift;
502                 shift += 7;
503         }
504         obj->size = size;
505
506         switch (obj->type) {
507         case OBJ_REF_DELTA:
508                 hashcpy(delta_base->sha1, fill(20));
509                 use(20);
510                 break;
511         case OBJ_OFS_DELTA:
512                 memset(delta_base, 0, sizeof(*delta_base));
513                 p = fill(1);
514                 c = *p;
515                 use(1);
516                 base_offset = c & 127;
517                 while (c & 128) {
518                         base_offset += 1;
519                         if (!base_offset || MSB(base_offset, 7))
520                                 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
521                         p = fill(1);
522                         c = *p;
523                         use(1);
524                         base_offset = (base_offset << 7) + (c & 127);
525                 }
526                 delta_base->offset = obj->idx.offset - base_offset;
527                 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
528                         bad_object(obj->idx.offset, _("delta base offset is out of bound"));
529                 break;
530         case OBJ_COMMIT:
531         case OBJ_TREE:
532         case OBJ_BLOB:
533         case OBJ_TAG:
534                 break;
535         default:
536                 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
537         }
538         obj->hdr_size = consumed_bytes - obj->idx.offset;
539
540         data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
541         obj->idx.crc32 = input_crc32;
542         return data;
543 }
544
545 static void *unpack_data(struct object_entry *obj,
546                          int (*consume)(const unsigned char *, unsigned long, void *),
547                          void *cb_data)
548 {
549         off_t from = obj[0].idx.offset + obj[0].hdr_size;
550         unsigned long len = obj[1].idx.offset - from;
551         unsigned char *data, *inbuf;
552         git_zstream stream;
553         int status;
554
555         data = xmallocz(consume ? 64*1024 : obj->size);
556         inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
557
558         memset(&stream, 0, sizeof(stream));
559         git_inflate_init(&stream);
560         stream.next_out = data;
561         stream.avail_out = consume ? 64*1024 : obj->size;
562
563         do {
564                 ssize_t n = (len < 64*1024) ? len : 64*1024;
565                 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
566                 if (n < 0)
567                         die_errno(_("cannot pread pack file"));
568                 if (!n)
569                         die(Q_("premature end of pack file, %lu byte missing",
570                                "premature end of pack file, %lu bytes missing",
571                                len),
572                             len);
573                 from += n;
574                 len -= n;
575                 stream.next_in = inbuf;
576                 stream.avail_in = n;
577                 if (!consume)
578                         status = git_inflate(&stream, 0);
579                 else {
580                         do {
581                                 status = git_inflate(&stream, 0);
582                                 if (consume(data, stream.next_out - data, cb_data)) {
583                                         free(inbuf);
584                                         free(data);
585                                         return NULL;
586                                 }
587                                 stream.next_out = data;
588                                 stream.avail_out = 64*1024;
589                         } while (status == Z_OK && stream.avail_in);
590                 }
591         } while (len && status == Z_OK && !stream.avail_in);
592
593         /* This has been inflated OK when first encountered, so... */
594         if (status != Z_STREAM_END || stream.total_out != obj->size)
595                 die(_("serious inflate inconsistency"));
596
597         git_inflate_end(&stream);
598         free(inbuf);
599         if (consume) {
600                 free(data);
601                 data = NULL;
602         }
603         return data;
604 }
605
606 static void *get_data_from_pack(struct object_entry *obj)
607 {
608         return unpack_data(obj, NULL, NULL);
609 }
610
611 static int compare_delta_bases(const union delta_base *base1,
612                                const union delta_base *base2,
613                                enum object_type type1,
614                                enum object_type type2)
615 {
616         int cmp = type1 - type2;
617         if (cmp)
618                 return cmp;
619         return memcmp(base1, base2, UNION_BASE_SZ);
620 }
621
622 static int find_delta(const union delta_base *base, enum object_type type)
623 {
624         int first = 0, last = nr_deltas;
625
626         while (first < last) {
627                 int next = (first + last) / 2;
628                 struct delta_entry *delta = &deltas[next];
629                 int cmp;
630
631                 cmp = compare_delta_bases(base, &delta->base,
632                                           type, objects[delta->obj_no].type);
633                 if (!cmp)
634                         return next;
635                 if (cmp < 0) {
636                         last = next;
637                         continue;
638                 }
639                 first = next+1;
640         }
641         return -first-1;
642 }
643
644 static void find_delta_children(const union delta_base *base,
645                                 int *first_index, int *last_index,
646                                 enum object_type type)
647 {
648         int first = find_delta(base, type);
649         int last = first;
650         int end = nr_deltas - 1;
651
652         if (first < 0) {
653                 *first_index = 0;
654                 *last_index = -1;
655                 return;
656         }
657         while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
658                 --first;
659         while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
660                 ++last;
661         *first_index = first;
662         *last_index = last;
663 }
664
665 struct compare_data {
666         struct object_entry *entry;
667         struct git_istream *st;
668         unsigned char *buf;
669         unsigned long buf_size;
670 };
671
672 static int compare_objects(const unsigned char *buf, unsigned long size,
673                            void *cb_data)
674 {
675         struct compare_data *data = cb_data;
676
677         if (data->buf_size < size) {
678                 free(data->buf);
679                 data->buf = xmalloc(size);
680                 data->buf_size = size;
681         }
682
683         while (size) {
684                 ssize_t len = read_istream(data->st, data->buf, size);
685                 if (len == 0)
686                         die(_("SHA1 COLLISION FOUND WITH %s !"),
687                             sha1_to_hex(data->entry->idx.sha1));
688                 if (len < 0)
689                         die(_("unable to read %s"),
690                             sha1_to_hex(data->entry->idx.sha1));
691                 if (memcmp(buf, data->buf, len))
692                         die(_("SHA1 COLLISION FOUND WITH %s !"),
693                             sha1_to_hex(data->entry->idx.sha1));
694                 size -= len;
695                 buf += len;
696         }
697         return 0;
698 }
699
700 static int check_collison(struct object_entry *entry)
701 {
702         struct compare_data data;
703         enum object_type type;
704         unsigned long size;
705
706         if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
707                 return -1;
708
709         memset(&data, 0, sizeof(data));
710         data.entry = entry;
711         data.st = open_istream(entry->idx.sha1, &type, &size, NULL);
712         if (!data.st)
713                 return -1;
714         if (size != entry->size || type != entry->type)
715                 die(_("SHA1 COLLISION FOUND WITH %s !"),
716                     sha1_to_hex(entry->idx.sha1));
717         unpack_data(entry, compare_objects, &data);
718         close_istream(data.st);
719         free(data.buf);
720         return 0;
721 }
722
723 static void sha1_object(const void *data, struct object_entry *obj_entry,
724                         unsigned long size, enum object_type type,
725                         const unsigned char *sha1)
726 {
727         void *new_data = NULL;
728         int collision_test_needed;
729
730         assert(data || obj_entry);
731
732         read_lock();
733         collision_test_needed = has_sha1_file(sha1);
734         read_unlock();
735
736         if (collision_test_needed && !data) {
737                 read_lock();
738                 if (!check_collison(obj_entry))
739                         collision_test_needed = 0;
740                 read_unlock();
741         }
742         if (collision_test_needed) {
743                 void *has_data;
744                 enum object_type has_type;
745                 unsigned long has_size;
746                 read_lock();
747                 has_type = sha1_object_info(sha1, &has_size);
748                 if (has_type != type || has_size != size)
749                         die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
750                 has_data = read_sha1_file(sha1, &has_type, &has_size);
751                 read_unlock();
752                 if (!data)
753                         data = new_data = get_data_from_pack(obj_entry);
754                 if (!has_data)
755                         die(_("cannot read existing object %s"), sha1_to_hex(sha1));
756                 if (size != has_size || type != has_type ||
757                     memcmp(data, has_data, size) != 0)
758                         die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
759                 free(has_data);
760         }
761
762         if (strict) {
763                 read_lock();
764                 if (type == OBJ_BLOB) {
765                         struct blob *blob = lookup_blob(sha1);
766                         if (blob)
767                                 blob->object.flags |= FLAG_CHECKED;
768                         else
769                                 die(_("invalid blob object %s"), sha1_to_hex(sha1));
770                 } else {
771                         struct object *obj;
772                         int eaten;
773                         void *buf = (void *) data;
774
775                         assert(data && "data can only be NULL for large _blobs_");
776
777                         /*
778                          * we do not need to free the memory here, as the
779                          * buf is deleted by the caller.
780                          */
781                         obj = parse_object_buffer(sha1, type, size, buf, &eaten);
782                         if (!obj)
783                                 die(_("invalid %s"), typename(type));
784                         if (do_fsck_object &&
785                             fsck_object(obj, buf, size, 1,
786                                     fsck_error_function))
787                                 die(_("Error in object"));
788                         if (fsck_walk(obj, mark_link, NULL))
789                                 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
790
791                         if (obj->type == OBJ_TREE) {
792                                 struct tree *item = (struct tree *) obj;
793                                 item->buffer = NULL;
794                                 obj->parsed = 0;
795                         }
796                         if (obj->type == OBJ_COMMIT) {
797                                 struct commit *commit = (struct commit *) obj;
798                                 if (detach_commit_buffer(commit, NULL) != data)
799                                         die("BUG: parse_object_buffer transmogrified our buffer");
800                         }
801                         obj->flags |= FLAG_CHECKED;
802                 }
803                 read_unlock();
804         }
805
806         free(new_data);
807 }
808
809 /*
810  * This function is part of find_unresolved_deltas(). There are two
811  * walkers going in the opposite ways.
812  *
813  * The first one in find_unresolved_deltas() traverses down from
814  * parent node to children, deflating nodes along the way. However,
815  * memory for deflated nodes is limited by delta_base_cache_limit, so
816  * at some point parent node's deflated content may be freed.
817  *
818  * The second walker is this function, which goes from current node up
819  * to top parent if necessary to deflate the node. In normal
820  * situation, its parent node would be already deflated, so it just
821  * needs to apply delta.
822  *
823  * In the worst case scenario, parent node is no longer deflated because
824  * we're running out of delta_base_cache_limit; we need to re-deflate
825  * parents, possibly up to the top base.
826  *
827  * All deflated objects here are subject to be freed if we exceed
828  * delta_base_cache_limit, just like in find_unresolved_deltas(), we
829  * just need to make sure the last node is not freed.
830  */
831 static void *get_base_data(struct base_data *c)
832 {
833         if (!c->data) {
834                 struct object_entry *obj = c->obj;
835                 struct base_data **delta = NULL;
836                 int delta_nr = 0, delta_alloc = 0;
837
838                 while (is_delta_type(c->obj->type) && !c->data) {
839                         ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
840                         delta[delta_nr++] = c;
841                         c = c->base;
842                 }
843                 if (!delta_nr) {
844                         c->data = get_data_from_pack(obj);
845                         c->size = obj->size;
846                         get_thread_data()->base_cache_used += c->size;
847                         prune_base_data(c);
848                 }
849                 for (; delta_nr > 0; delta_nr--) {
850                         void *base, *raw;
851                         c = delta[delta_nr - 1];
852                         obj = c->obj;
853                         base = get_base_data(c->base);
854                         raw = get_data_from_pack(obj);
855                         c->data = patch_delta(
856                                 base, c->base->size,
857                                 raw, obj->size,
858                                 &c->size);
859                         free(raw);
860                         if (!c->data)
861                                 bad_object(obj->idx.offset, _("failed to apply delta"));
862                         get_thread_data()->base_cache_used += c->size;
863                         prune_base_data(c);
864                 }
865                 free(delta);
866         }
867         return c->data;
868 }
869
870 static void resolve_delta(struct object_entry *delta_obj,
871                           struct base_data *base, struct base_data *result)
872 {
873         void *base_data, *delta_data;
874
875         if (show_stat) {
876                 delta_obj->delta_depth = base->obj->delta_depth + 1;
877                 deepest_delta_lock();
878                 if (deepest_delta < delta_obj->delta_depth)
879                         deepest_delta = delta_obj->delta_depth;
880                 deepest_delta_unlock();
881         }
882         delta_obj->base_object_no = base->obj - objects;
883         delta_data = get_data_from_pack(delta_obj);
884         base_data = get_base_data(base);
885         result->obj = delta_obj;
886         result->data = patch_delta(base_data, base->size,
887                                    delta_data, delta_obj->size, &result->size);
888         free(delta_data);
889         if (!result->data)
890                 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
891         hash_sha1_file(result->data, result->size,
892                        typename(delta_obj->real_type), delta_obj->idx.sha1);
893         sha1_object(result->data, NULL, result->size, delta_obj->real_type,
894                     delta_obj->idx.sha1);
895         counter_lock();
896         nr_resolved_deltas++;
897         counter_unlock();
898 }
899
900 /*
901  * Standard boolean compare-and-swap: atomically check whether "*type" is
902  * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
903  * and return false.
904  */
905 static int compare_and_swap_type(enum object_type *type,
906                                  enum object_type want,
907                                  enum object_type set)
908 {
909         enum object_type old;
910
911         type_cas_lock();
912         old = *type;
913         if (old == want)
914                 *type = set;
915         type_cas_unlock();
916
917         return old == want;
918 }
919
920 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
921                                                   struct base_data *prev_base)
922 {
923         if (base->ref_last == -1 && base->ofs_last == -1) {
924                 union delta_base base_spec;
925
926                 hashcpy(base_spec.sha1, base->obj->idx.sha1);
927                 find_delta_children(&base_spec,
928                                     &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
929
930                 memset(&base_spec, 0, sizeof(base_spec));
931                 base_spec.offset = base->obj->idx.offset;
932                 find_delta_children(&base_spec,
933                                     &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
934
935                 if (base->ref_last == -1 && base->ofs_last == -1) {
936                         free(base->data);
937                         return NULL;
938                 }
939
940                 link_base_data(prev_base, base);
941         }
942
943         if (base->ref_first <= base->ref_last) {
944                 struct object_entry *child = objects + deltas[base->ref_first].obj_no;
945                 struct base_data *result = alloc_base_data();
946
947                 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
948                                            base->obj->real_type))
949                         die("BUG: child->real_type != OBJ_REF_DELTA");
950
951                 resolve_delta(child, base, result);
952                 if (base->ref_first == base->ref_last && base->ofs_last == -1)
953                         free_base_data(base);
954
955                 base->ref_first++;
956                 return result;
957         }
958
959         if (base->ofs_first <= base->ofs_last) {
960                 struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
961                 struct base_data *result = alloc_base_data();
962
963                 assert(child->real_type == OBJ_OFS_DELTA);
964                 child->real_type = base->obj->real_type;
965                 resolve_delta(child, base, result);
966                 if (base->ofs_first == base->ofs_last)
967                         free_base_data(base);
968
969                 base->ofs_first++;
970                 return result;
971         }
972
973         unlink_base_data(base);
974         return NULL;
975 }
976
977 static void find_unresolved_deltas(struct base_data *base)
978 {
979         struct base_data *new_base, *prev_base = NULL;
980         for (;;) {
981                 new_base = find_unresolved_deltas_1(base, prev_base);
982
983                 if (new_base) {
984                         prev_base = base;
985                         base = new_base;
986                 } else {
987                         free(base);
988                         base = prev_base;
989                         if (!base)
990                                 return;
991                         prev_base = base->base;
992                 }
993         }
994 }
995
996 static int compare_delta_entry(const void *a, const void *b)
997 {
998         const struct delta_entry *delta_a = a;
999         const struct delta_entry *delta_b = b;
1000
1001         /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
1002         return compare_delta_bases(&delta_a->base, &delta_b->base,
1003                                    objects[delta_a->obj_no].type,
1004                                    objects[delta_b->obj_no].type);
1005 }
1006
1007 static void resolve_base(struct object_entry *obj)
1008 {
1009         struct base_data *base_obj = alloc_base_data();
1010         base_obj->obj = obj;
1011         base_obj->data = NULL;
1012         find_unresolved_deltas(base_obj);
1013 }
1014
1015 #ifndef NO_PTHREADS
1016 static void *threaded_second_pass(void *data)
1017 {
1018         set_thread_data(data);
1019         for (;;) {
1020                 int i;
1021                 counter_lock();
1022                 display_progress(progress, nr_resolved_deltas);
1023                 counter_unlock();
1024                 work_lock();
1025                 while (nr_dispatched < nr_objects &&
1026                        is_delta_type(objects[nr_dispatched].type))
1027                         nr_dispatched++;
1028                 if (nr_dispatched >= nr_objects) {
1029                         work_unlock();
1030                         break;
1031                 }
1032                 i = nr_dispatched++;
1033                 work_unlock();
1034
1035                 resolve_base(&objects[i]);
1036         }
1037         return NULL;
1038 }
1039 #endif
1040
1041 /*
1042  * First pass:
1043  * - find locations of all objects;
1044  * - calculate SHA1 of all non-delta objects;
1045  * - remember base (SHA1 or offset) for all deltas.
1046  */
1047 static void parse_pack_objects(unsigned char *sha1)
1048 {
1049         int i, nr_delays = 0;
1050         struct delta_entry *delta = deltas;
1051         struct stat st;
1052
1053         if (verbose)
1054                 progress = start_progress(
1055                                 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1056                                 nr_objects);
1057         for (i = 0; i < nr_objects; i++) {
1058                 struct object_entry *obj = &objects[i];
1059                 void *data = unpack_raw_entry(obj, &delta->base, obj->idx.sha1);
1060                 obj->real_type = obj->type;
1061                 if (is_delta_type(obj->type)) {
1062                         nr_deltas++;
1063                         delta->obj_no = i;
1064                         delta++;
1065                 } else if (!data) {
1066                         /* large blobs, check later */
1067                         obj->real_type = OBJ_BAD;
1068                         nr_delays++;
1069                 } else
1070                         sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
1071                 free(data);
1072                 display_progress(progress, i+1);
1073         }
1074         objects[i].idx.offset = consumed_bytes;
1075         stop_progress(&progress);
1076
1077         /* Check pack integrity */
1078         flush();
1079         git_SHA1_Final(sha1, &input_ctx);
1080         if (hashcmp(fill(20), sha1))
1081                 die(_("pack is corrupted (SHA1 mismatch)"));
1082         use(20);
1083
1084         /* If input_fd is a file, we should have reached its end now. */
1085         if (fstat(input_fd, &st))
1086                 die_errno(_("cannot fstat packfile"));
1087         if (S_ISREG(st.st_mode) &&
1088                         lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1089                 die(_("pack has junk at the end"));
1090
1091         for (i = 0; i < nr_objects; i++) {
1092                 struct object_entry *obj = &objects[i];
1093                 if (obj->real_type != OBJ_BAD)
1094                         continue;
1095                 obj->real_type = obj->type;
1096                 sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
1097                 nr_delays--;
1098         }
1099         if (nr_delays)
1100                 die(_("confusion beyond insanity in parse_pack_objects()"));
1101 }
1102
1103 /*
1104  * Second pass:
1105  * - for all non-delta objects, look if it is used as a base for
1106  *   deltas;
1107  * - if used as a base, uncompress the object and apply all deltas,
1108  *   recursively checking if the resulting object is used as a base
1109  *   for some more deltas.
1110  */
1111 static void resolve_deltas(void)
1112 {
1113         int i;
1114
1115         if (!nr_deltas)
1116                 return;
1117
1118         /* Sort deltas by base SHA1/offset for fast searching */
1119         qsort(deltas, nr_deltas, sizeof(struct delta_entry),
1120               compare_delta_entry);
1121
1122         if (verbose)
1123                 progress = start_progress(_("Resolving deltas"), nr_deltas);
1124
1125 #ifndef NO_PTHREADS
1126         nr_dispatched = 0;
1127         if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1128                 init_thread();
1129                 for (i = 0; i < nr_threads; i++) {
1130                         int ret = pthread_create(&thread_data[i].thread, NULL,
1131                                                  threaded_second_pass, thread_data + i);
1132                         if (ret)
1133                                 die(_("unable to create thread: %s"),
1134                                     strerror(ret));
1135                 }
1136                 for (i = 0; i < nr_threads; i++)
1137                         pthread_join(thread_data[i].thread, NULL);
1138                 cleanup_thread();
1139                 return;
1140         }
1141 #endif
1142
1143         for (i = 0; i < nr_objects; i++) {
1144                 struct object_entry *obj = &objects[i];
1145
1146                 if (is_delta_type(obj->type))
1147                         continue;
1148                 resolve_base(obj);
1149                 display_progress(progress, nr_resolved_deltas);
1150         }
1151 }
1152
1153 /*
1154  * Third pass:
1155  * - append objects to convert thin pack to full pack if required
1156  * - write the final 20-byte SHA-1
1157  */
1158 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
1159 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1160 {
1161         if (nr_deltas == nr_resolved_deltas) {
1162                 stop_progress(&progress);
1163                 /* Flush remaining pack final 20-byte SHA1. */
1164                 flush();
1165                 return;
1166         }
1167
1168         if (fix_thin_pack) {
1169                 struct sha1file *f;
1170                 unsigned char read_sha1[20], tail_sha1[20];
1171                 struct strbuf msg = STRBUF_INIT;
1172                 int nr_unresolved = nr_deltas - nr_resolved_deltas;
1173                 int nr_objects_initial = nr_objects;
1174                 if (nr_unresolved <= 0)
1175                         die(_("confusion beyond insanity"));
1176                 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1177                 memset(objects + nr_objects + 1, 0,
1178                        nr_unresolved * sizeof(*objects));
1179                 f = sha1fd(output_fd, curr_pack);
1180                 fix_unresolved_deltas(f, nr_unresolved);
1181                 strbuf_addf(&msg, _("completed with %d local objects"),
1182                             nr_objects - nr_objects_initial);
1183                 stop_progress_msg(&progress, msg.buf);
1184                 strbuf_release(&msg);
1185                 sha1close(f, tail_sha1, 0);
1186                 hashcpy(read_sha1, pack_sha1);
1187                 fixup_pack_header_footer(output_fd, pack_sha1,
1188                                          curr_pack, nr_objects,
1189                                          read_sha1, consumed_bytes-20);
1190                 if (hashcmp(read_sha1, tail_sha1) != 0)
1191                         die(_("Unexpected tail checksum for %s "
1192                               "(disk corruption?)"), curr_pack);
1193         }
1194         if (nr_deltas != nr_resolved_deltas)
1195                 die(Q_("pack has %d unresolved delta",
1196                        "pack has %d unresolved deltas",
1197                        nr_deltas - nr_resolved_deltas),
1198                     nr_deltas - nr_resolved_deltas);
1199 }
1200
1201 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1202 {
1203         git_zstream stream;
1204         int status;
1205         unsigned char outbuf[4096];
1206
1207         git_deflate_init(&stream, zlib_compression_level);
1208         stream.next_in = in;
1209         stream.avail_in = size;
1210
1211         do {
1212                 stream.next_out = outbuf;
1213                 stream.avail_out = sizeof(outbuf);
1214                 status = git_deflate(&stream, Z_FINISH);
1215                 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1216         } while (status == Z_OK);
1217
1218         if (status != Z_STREAM_END)
1219                 die(_("unable to deflate appended object (%d)"), status);
1220         size = stream.total_out;
1221         git_deflate_end(&stream);
1222         return size;
1223 }
1224
1225 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1226                                const unsigned char *sha1, void *buf,
1227                                unsigned long size, enum object_type type)
1228 {
1229         struct object_entry *obj = &objects[nr_objects++];
1230         unsigned char header[10];
1231         unsigned long s = size;
1232         int n = 0;
1233         unsigned char c = (type << 4) | (s & 15);
1234         s >>= 4;
1235         while (s) {
1236                 header[n++] = c | 0x80;
1237                 c = s & 0x7f;
1238                 s >>= 7;
1239         }
1240         header[n++] = c;
1241         crc32_begin(f);
1242         sha1write(f, header, n);
1243         obj[0].size = size;
1244         obj[0].hdr_size = n;
1245         obj[0].type = type;
1246         obj[0].real_type = type;
1247         obj[1].idx.offset = obj[0].idx.offset + n;
1248         obj[1].idx.offset += write_compressed(f, buf, size);
1249         obj[0].idx.crc32 = crc32_end(f);
1250         sha1flush(f);
1251         hashcpy(obj->idx.sha1, sha1);
1252         return obj;
1253 }
1254
1255 static int delta_pos_compare(const void *_a, const void *_b)
1256 {
1257         struct delta_entry *a = *(struct delta_entry **)_a;
1258         struct delta_entry *b = *(struct delta_entry **)_b;
1259         return a->obj_no - b->obj_no;
1260 }
1261
1262 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
1263 {
1264         struct delta_entry **sorted_by_pos;
1265         int i, n = 0;
1266
1267         /*
1268          * Since many unresolved deltas may well be themselves base objects
1269          * for more unresolved deltas, we really want to include the
1270          * smallest number of base objects that would cover as much delta
1271          * as possible by picking the
1272          * trunc deltas first, allowing for other deltas to resolve without
1273          * additional base objects.  Since most base objects are to be found
1274          * before deltas depending on them, a good heuristic is to start
1275          * resolving deltas in the same order as their position in the pack.
1276          */
1277         sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
1278         for (i = 0; i < nr_deltas; i++) {
1279                 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1280                         continue;
1281                 sorted_by_pos[n++] = &deltas[i];
1282         }
1283         qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1284
1285         for (i = 0; i < n; i++) {
1286                 struct delta_entry *d = sorted_by_pos[i];
1287                 enum object_type type;
1288                 struct base_data *base_obj = alloc_base_data();
1289
1290                 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1291                         continue;
1292                 base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1293                 if (!base_obj->data)
1294                         continue;
1295
1296                 if (check_sha1_signature(d->base.sha1, base_obj->data,
1297                                 base_obj->size, typename(type)))
1298                         die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
1299                 base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1300                                         base_obj->data, base_obj->size, type);
1301                 find_unresolved_deltas(base_obj);
1302                 display_progress(progress, nr_resolved_deltas);
1303         }
1304         free(sorted_by_pos);
1305 }
1306
1307 static void final(const char *final_pack_name, const char *curr_pack_name,
1308                   const char *final_index_name, const char *curr_index_name,
1309                   const char *keep_name, const char *keep_msg,
1310                   unsigned char *sha1)
1311 {
1312         const char *report = "pack";
1313         char name[PATH_MAX];
1314         int err;
1315
1316         if (!from_stdin) {
1317                 close(input_fd);
1318         } else {
1319                 fsync_or_die(output_fd, curr_pack_name);
1320                 err = close(output_fd);
1321                 if (err)
1322                         die_errno(_("error while closing pack file"));
1323         }
1324
1325         if (keep_msg) {
1326                 int keep_fd, keep_msg_len = strlen(keep_msg);
1327
1328                 if (!keep_name)
1329                         keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1330                 else
1331                         keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1332
1333                 if (keep_fd < 0) {
1334                         if (errno != EEXIST)
1335                                 die_errno(_("cannot write keep file '%s'"),
1336                                           keep_name ? keep_name : name);
1337                 } else {
1338                         if (keep_msg_len > 0) {
1339                                 write_or_die(keep_fd, keep_msg, keep_msg_len);
1340                                 write_or_die(keep_fd, "\n", 1);
1341                         }
1342                         if (close(keep_fd) != 0)
1343                                 die_errno(_("cannot close written keep file '%s'"),
1344                                           keep_name ? keep_name : name);
1345                         report = "keep";
1346                 }
1347         }
1348
1349         if (final_pack_name != curr_pack_name) {
1350                 if (!final_pack_name) {
1351                         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1352                                  get_object_directory(), sha1_to_hex(sha1));
1353                         final_pack_name = name;
1354                 }
1355                 if (move_temp_to_file(curr_pack_name, final_pack_name))
1356                         die(_("cannot store pack file"));
1357         } else if (from_stdin)
1358                 chmod(final_pack_name, 0444);
1359
1360         if (final_index_name != curr_index_name) {
1361                 if (!final_index_name) {
1362                         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1363                                  get_object_directory(), sha1_to_hex(sha1));
1364                         final_index_name = name;
1365                 }
1366                 if (move_temp_to_file(curr_index_name, final_index_name))
1367                         die(_("cannot store index file"));
1368         } else
1369                 chmod(final_index_name, 0444);
1370
1371         if (!from_stdin) {
1372                 printf("%s\n", sha1_to_hex(sha1));
1373         } else {
1374                 char buf[48];
1375                 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1376                                    report, sha1_to_hex(sha1));
1377                 write_or_die(1, buf, len);
1378
1379                 /*
1380                  * Let's just mimic git-unpack-objects here and write
1381                  * the last part of the input buffer to stdout.
1382                  */
1383                 while (input_len) {
1384                         err = xwrite(1, input_buffer + input_offset, input_len);
1385                         if (err <= 0)
1386                                 break;
1387                         input_len -= err;
1388                         input_offset += err;
1389                 }
1390         }
1391 }
1392
1393 static int git_index_pack_config(const char *k, const char *v, void *cb)
1394 {
1395         struct pack_idx_option *opts = cb;
1396
1397         if (!strcmp(k, "pack.indexversion")) {
1398                 opts->version = git_config_int(k, v);
1399                 if (opts->version > 2)
1400                         die(_("bad pack.indexversion=%"PRIu32), opts->version);
1401                 return 0;
1402         }
1403         if (!strcmp(k, "pack.threads")) {
1404                 nr_threads = git_config_int(k, v);
1405                 if (nr_threads < 0)
1406                         die(_("invalid number of threads specified (%d)"),
1407                             nr_threads);
1408 #ifdef NO_PTHREADS
1409                 if (nr_threads != 1)
1410                         warning(_("no threads support, ignoring %s"), k);
1411                 nr_threads = 1;
1412 #endif
1413                 return 0;
1414         }
1415         return git_default_config(k, v, cb);
1416 }
1417
1418 static int cmp_uint32(const void *a_, const void *b_)
1419 {
1420         uint32_t a = *((uint32_t *)a_);
1421         uint32_t b = *((uint32_t *)b_);
1422
1423         return (a < b) ? -1 : (a != b);
1424 }
1425
1426 static void read_v2_anomalous_offsets(struct packed_git *p,
1427                                       struct pack_idx_option *opts)
1428 {
1429         const uint32_t *idx1, *idx2;
1430         uint32_t i;
1431
1432         /* The address of the 4-byte offset table */
1433         idx1 = (((const uint32_t *)p->index_data)
1434                 + 2 /* 8-byte header */
1435                 + 256 /* fan out */
1436                 + 5 * p->num_objects /* 20-byte SHA-1 table */
1437                 + p->num_objects /* CRC32 table */
1438                 );
1439
1440         /* The address of the 8-byte offset table */
1441         idx2 = idx1 + p->num_objects;
1442
1443         for (i = 0; i < p->num_objects; i++) {
1444                 uint32_t off = ntohl(idx1[i]);
1445                 if (!(off & 0x80000000))
1446                         continue;
1447                 off = off & 0x7fffffff;
1448                 if (idx2[off * 2])
1449                         continue;
1450                 /*
1451                  * The real offset is ntohl(idx2[off * 2]) in high 4
1452                  * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1453                  * octets.  But idx2[off * 2] is Zero!!!
1454                  */
1455                 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1456                 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1457         }
1458
1459         if (1 < opts->anomaly_nr)
1460                 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1461 }
1462
1463 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1464 {
1465         struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1466
1467         if (!p)
1468                 die(_("Cannot open existing pack file '%s'"), pack_name);
1469         if (open_pack_index(p))
1470                 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1471
1472         /* Read the attributes from the existing idx file */
1473         opts->version = p->index_version;
1474
1475         if (opts->version == 2)
1476                 read_v2_anomalous_offsets(p, opts);
1477
1478         /*
1479          * Get rid of the idx file as we do not need it anymore.
1480          * NEEDSWORK: extract this bit from free_pack_by_name() in
1481          * sha1_file.c, perhaps?  It shouldn't matter very much as we
1482          * know we haven't installed this pack (hence we never have
1483          * read anything from it).
1484          */
1485         close_pack_index(p);
1486         free(p);
1487 }
1488
1489 static void show_pack_info(int stat_only)
1490 {
1491         int i, baseobjects = nr_objects - nr_deltas;
1492         unsigned long *chain_histogram = NULL;
1493
1494         if (deepest_delta)
1495                 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1496
1497         for (i = 0; i < nr_objects; i++) {
1498                 struct object_entry *obj = &objects[i];
1499
1500                 if (is_delta_type(obj->type))
1501                         chain_histogram[obj->delta_depth - 1]++;
1502                 if (stat_only)
1503                         continue;
1504                 printf("%s %-6s %lu %lu %"PRIuMAX,
1505                        sha1_to_hex(obj->idx.sha1),
1506                        typename(obj->real_type), obj->size,
1507                        (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1508                        (uintmax_t)obj->idx.offset);
1509                 if (is_delta_type(obj->type)) {
1510                         struct object_entry *bobj = &objects[obj->base_object_no];
1511                         printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1512                 }
1513                 putchar('\n');
1514         }
1515
1516         if (baseobjects)
1517                 printf_ln(Q_("non delta: %d object",
1518                              "non delta: %d objects",
1519                              baseobjects),
1520                           baseobjects);
1521         for (i = 0; i < deepest_delta; i++) {
1522                 if (!chain_histogram[i])
1523                         continue;
1524                 printf_ln(Q_("chain length = %d: %lu object",
1525                              "chain length = %d: %lu objects",
1526                              chain_histogram[i]),
1527                           i + 1,
1528                           chain_histogram[i]);
1529         }
1530 }
1531
1532 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1533 {
1534         int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1535         const char *curr_index;
1536         const char *index_name = NULL, *pack_name = NULL;
1537         const char *keep_name = NULL, *keep_msg = NULL;
1538         struct strbuf index_name_buf = STRBUF_INIT,
1539                       keep_name_buf = STRBUF_INIT;
1540         struct pack_idx_entry **idx_objects;
1541         struct pack_idx_option opts;
1542         unsigned char pack_sha1[20];
1543         unsigned foreign_nr = 1;        /* zero is a "good" value, assume bad */
1544
1545         if (argc == 2 && !strcmp(argv[1], "-h"))
1546                 usage(index_pack_usage);
1547
1548         check_replace_refs = 0;
1549
1550         reset_pack_idx_option(&opts);
1551         git_config(git_index_pack_config, &opts);
1552         if (prefix && chdir(prefix))
1553                 die(_("Cannot come back to cwd"));
1554
1555         for (i = 1; i < argc; i++) {
1556                 const char *arg = argv[i];
1557
1558                 if (*arg == '-') {
1559                         if (!strcmp(arg, "--stdin")) {
1560                                 from_stdin = 1;
1561                         } else if (!strcmp(arg, "--fix-thin")) {
1562                                 fix_thin_pack = 1;
1563                         } else if (!strcmp(arg, "--strict")) {
1564                                 strict = 1;
1565                                 do_fsck_object = 1;
1566                         } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1567                                 strict = 1;
1568                                 check_self_contained_and_connected = 1;
1569                         } else if (!strcmp(arg, "--verify")) {
1570                                 verify = 1;
1571                         } else if (!strcmp(arg, "--verify-stat")) {
1572                                 verify = 1;
1573                                 show_stat = 1;
1574                         } else if (!strcmp(arg, "--verify-stat-only")) {
1575                                 verify = 1;
1576                                 show_stat = 1;
1577                                 stat_only = 1;
1578                         } else if (!strcmp(arg, "--keep")) {
1579                                 keep_msg = "";
1580                         } else if (starts_with(arg, "--keep=")) {
1581                                 keep_msg = arg + 7;
1582                         } else if (starts_with(arg, "--threads=")) {
1583                                 char *end;
1584                                 nr_threads = strtoul(arg+10, &end, 0);
1585                                 if (!arg[10] || *end || nr_threads < 0)
1586                                         usage(index_pack_usage);
1587 #ifdef NO_PTHREADS
1588                                 if (nr_threads != 1)
1589                                         warning(_("no threads support, "
1590                                                   "ignoring %s"), arg);
1591                                 nr_threads = 1;
1592 #endif
1593                         } else if (starts_with(arg, "--pack_header=")) {
1594                                 struct pack_header *hdr;
1595                                 char *c;
1596
1597                                 hdr = (struct pack_header *)input_buffer;
1598                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1599                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1600                                 if (*c != ',')
1601                                         die(_("bad %s"), arg);
1602                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1603                                 if (*c)
1604                                         die(_("bad %s"), arg);
1605                                 input_len = sizeof(*hdr);
1606                         } else if (!strcmp(arg, "-v")) {
1607                                 verbose = 1;
1608                         } else if (!strcmp(arg, "-o")) {
1609                                 if (index_name || (i+1) >= argc)
1610                                         usage(index_pack_usage);
1611                                 index_name = argv[++i];
1612                         } else if (starts_with(arg, "--index-version=")) {
1613                                 char *c;
1614                                 opts.version = strtoul(arg + 16, &c, 10);
1615                                 if (opts.version > 2)
1616                                         die(_("bad %s"), arg);
1617                                 if (*c == ',')
1618                                         opts.off32_limit = strtoul(c+1, &c, 0);
1619                                 if (*c || opts.off32_limit & 0x80000000)
1620                                         die(_("bad %s"), arg);
1621                         } else
1622                                 usage(index_pack_usage);
1623                         continue;
1624                 }
1625
1626                 if (pack_name)
1627                         usage(index_pack_usage);
1628                 pack_name = arg;
1629         }
1630
1631         if (!pack_name && !from_stdin)
1632                 usage(index_pack_usage);
1633         if (fix_thin_pack && !from_stdin)
1634                 die(_("--fix-thin cannot be used without --stdin"));
1635         if (!index_name && pack_name) {
1636                 size_t len;
1637                 if (!strip_suffix(pack_name, ".pack", &len))
1638                         die(_("packfile name '%s' does not end with '.pack'"),
1639                             pack_name);
1640                 strbuf_add(&index_name_buf, pack_name, len);
1641                 strbuf_addstr(&index_name_buf, ".idx");
1642                 index_name = index_name_buf.buf;
1643         }
1644         if (keep_msg && !keep_name && pack_name) {
1645                 size_t len;
1646                 if (!strip_suffix(pack_name, ".pack", &len))
1647                         die(_("packfile name '%s' does not end with '.pack'"),
1648                             pack_name);
1649                 strbuf_add(&keep_name_buf, pack_name, len);
1650                 strbuf_addstr(&keep_name_buf, ".idx");
1651                 keep_name = keep_name_buf.buf;
1652         }
1653         if (verify) {
1654                 if (!index_name)
1655                         die(_("--verify with no packfile name given"));
1656                 read_idx_option(&opts, index_name);
1657                 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1658         }
1659         if (strict)
1660                 opts.flags |= WRITE_IDX_STRICT;
1661
1662 #ifndef NO_PTHREADS
1663         if (!nr_threads) {
1664                 nr_threads = online_cpus();
1665                 /* An experiment showed that more threads does not mean faster */
1666                 if (nr_threads > 3)
1667                         nr_threads = 3;
1668         }
1669 #endif
1670
1671         curr_pack = open_pack_file(pack_name);
1672         parse_pack_header();
1673         objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1674         deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1675         parse_pack_objects(pack_sha1);
1676         resolve_deltas();
1677         conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1678         free(deltas);
1679         if (strict)
1680                 foreign_nr = check_objects();
1681
1682         if (show_stat)
1683                 show_pack_info(stat_only);
1684
1685         idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1686         for (i = 0; i < nr_objects; i++)
1687                 idx_objects[i] = &objects[i].idx;
1688         curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1689         free(idx_objects);
1690
1691         if (!verify)
1692                 final(pack_name, curr_pack,
1693                       index_name, curr_index,
1694                       keep_name, keep_msg,
1695                       pack_sha1);
1696         else
1697                 close(input_fd);
1698         free(objects);
1699         strbuf_release(&index_name_buf);
1700         strbuf_release(&keep_name_buf);
1701         if (pack_name == NULL)
1702                 free((void *) curr_pack);
1703         if (index_name == NULL)
1704                 free((void *) curr_index);
1705
1706         /*
1707          * Let the caller know this pack is not self contained
1708          */
1709         if (check_self_contained_and_connected && foreign_nr)
1710                 return 1;
1711
1712         return 0;
1713 }