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