Merge branch 'sb/diff-color-move-more'
[git] / builtin / index-pack.c
1 #include "builtin.h"
2 #include "config.h"
3 #include "delta.h"
4 #include "pack.h"
5 #include "csum-file.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "progress.h"
11 #include "fsck.h"
12 #include "exec-cmd.h"
13 #include "streaming.h"
14 #include "thread-utils.h"
15 #include "packfile.h"
16 #include "object-store.h"
17
18 static const char index_pack_usage[] =
19 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
20
21 struct object_entry {
22         struct pack_idx_entry idx;
23         unsigned long size;
24         unsigned char hdr_size;
25         signed char type;
26         signed char real_type;
27 };
28
29 struct object_stat {
30         unsigned delta_depth;
31         int base_object_no;
32 };
33
34 struct base_data {
35         struct base_data *base;
36         struct base_data *child;
37         struct object_entry *obj;
38         void *data;
39         unsigned long size;
40         int ref_first, ref_last;
41         int ofs_first, ofs_last;
42 };
43
44 struct thread_local {
45 #ifndef NO_PTHREADS
46         pthread_t thread;
47 #endif
48         struct base_data *base_cache;
49         size_t base_cache_used;
50         int pack_fd;
51 };
52
53 /* Remember to update object flag allocation in object.h */
54 #define FLAG_LINK (1u<<20)
55 #define FLAG_CHECKED (1u<<21)
56
57 struct ofs_delta_entry {
58         off_t offset;
59         int obj_no;
60 };
61
62 struct ref_delta_entry {
63         struct object_id oid;
64         int obj_no;
65 };
66
67 static struct object_entry *objects;
68 static struct object_stat *obj_stat;
69 static struct ofs_delta_entry *ofs_deltas;
70 static struct ref_delta_entry *ref_deltas;
71 static struct thread_local nothread_data;
72 static int nr_objects;
73 static int nr_ofs_deltas;
74 static int nr_ref_deltas;
75 static int ref_deltas_alloc;
76 static int nr_resolved_deltas;
77 static int nr_threads;
78
79 static int from_stdin;
80 static int strict;
81 static int do_fsck_object;
82 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
83 static int verbose;
84 static int show_resolving_progress;
85 static int show_stat;
86 static int check_self_contained_and_connected;
87
88 static struct progress *progress;
89
90 /* We always read in 4kB chunks. */
91 static unsigned char input_buffer[4096];
92 static unsigned int input_offset, input_len;
93 static off_t consumed_bytes;
94 static off_t max_input_size;
95 static unsigned deepest_delta;
96 static git_hash_ctx input_ctx;
97 static uint32_t input_crc32;
98 static int input_fd, output_fd;
99 static const char *curr_pack;
100
101 #ifndef NO_PTHREADS
102
103 static struct thread_local *thread_data;
104 static int nr_dispatched;
105 static int threads_active;
106
107 static pthread_mutex_t read_mutex;
108 #define read_lock()             lock_mutex(&read_mutex)
109 #define read_unlock()           unlock_mutex(&read_mutex)
110
111 static pthread_mutex_t counter_mutex;
112 #define counter_lock()          lock_mutex(&counter_mutex)
113 #define counter_unlock()        unlock_mutex(&counter_mutex)
114
115 static pthread_mutex_t work_mutex;
116 #define work_lock()             lock_mutex(&work_mutex)
117 #define work_unlock()           unlock_mutex(&work_mutex)
118
119 static pthread_mutex_t deepest_delta_mutex;
120 #define deepest_delta_lock()    lock_mutex(&deepest_delta_mutex)
121 #define deepest_delta_unlock()  unlock_mutex(&deepest_delta_mutex)
122
123 static pthread_mutex_t type_cas_mutex;
124 #define type_cas_lock()         lock_mutex(&type_cas_mutex)
125 #define type_cas_unlock()       unlock_mutex(&type_cas_mutex)
126
127 static pthread_key_t key;
128
129 static inline void lock_mutex(pthread_mutex_t *mutex)
130 {
131         if (threads_active)
132                 pthread_mutex_lock(mutex);
133 }
134
135 static inline void unlock_mutex(pthread_mutex_t *mutex)
136 {
137         if (threads_active)
138                 pthread_mutex_unlock(mutex);
139 }
140
141 /*
142  * Mutex and conditional variable can't be statically-initialized on Windows.
143  */
144 static void init_thread(void)
145 {
146         int i;
147         init_recursive_mutex(&read_mutex);
148         pthread_mutex_init(&counter_mutex, NULL);
149         pthread_mutex_init(&work_mutex, NULL);
150         pthread_mutex_init(&type_cas_mutex, NULL);
151         if (show_stat)
152                 pthread_mutex_init(&deepest_delta_mutex, NULL);
153         pthread_key_create(&key, NULL);
154         thread_data = xcalloc(nr_threads, sizeof(*thread_data));
155         for (i = 0; i < nr_threads; i++) {
156                 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
157                 if (thread_data[i].pack_fd == -1)
158                         die_errno(_("unable to open %s"), curr_pack);
159         }
160
161         threads_active = 1;
162 }
163
164 static void cleanup_thread(void)
165 {
166         int i;
167         if (!threads_active)
168                 return;
169         threads_active = 0;
170         pthread_mutex_destroy(&read_mutex);
171         pthread_mutex_destroy(&counter_mutex);
172         pthread_mutex_destroy(&work_mutex);
173         pthread_mutex_destroy(&type_cas_mutex);
174         if (show_stat)
175                 pthread_mutex_destroy(&deepest_delta_mutex);
176         for (i = 0; i < nr_threads; i++)
177                 close(thread_data[i].pack_fd);
178         pthread_key_delete(key);
179         free(thread_data);
180 }
181
182 #else
183
184 #define read_lock()
185 #define read_unlock()
186
187 #define counter_lock()
188 #define counter_unlock()
189
190 #define work_lock()
191 #define work_unlock()
192
193 #define deepest_delta_lock()
194 #define deepest_delta_unlock()
195
196 #define type_cas_lock()
197 #define type_cas_unlock()
198
199 #endif
200
201
202 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
203 {
204         if (!obj)
205                 return -1;
206
207         if (type != OBJ_ANY && obj->type != type)
208                 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
209
210         obj->flags |= FLAG_LINK;
211         return 0;
212 }
213
214 /* The content of each linked object must have been checked
215    or it must be already present in the object database */
216 static unsigned check_object(struct object *obj)
217 {
218         if (!obj)
219                 return 0;
220
221         if (!(obj->flags & FLAG_LINK))
222                 return 0;
223
224         if (!(obj->flags & FLAG_CHECKED)) {
225                 unsigned long size;
226                 int type = oid_object_info(the_repository, &obj->oid, &size);
227                 if (type <= 0)
228                         die(_("did not receive expected object %s"),
229                               oid_to_hex(&obj->oid));
230                 if (type != obj->type)
231                         die(_("object %s: expected type %s, found %s"),
232                             oid_to_hex(&obj->oid),
233                             type_name(obj->type), type_name(type));
234                 obj->flags |= FLAG_CHECKED;
235                 return 1;
236         }
237
238         return 0;
239 }
240
241 static unsigned check_objects(void)
242 {
243         unsigned i, max, foreign_nr = 0;
244
245         max = get_max_object_index();
246         for (i = 0; i < max; i++)
247                 foreign_nr += check_object(get_indexed_object(i));
248         return foreign_nr;
249 }
250
251
252 /* Discard current buffer used content. */
253 static void flush(void)
254 {
255         if (input_offset) {
256                 if (output_fd >= 0)
257                         write_or_die(output_fd, input_buffer, input_offset);
258                 the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset);
259                 memmove(input_buffer, input_buffer + input_offset, input_len);
260                 input_offset = 0;
261         }
262 }
263
264 /*
265  * Make sure at least "min" bytes are available in the buffer, and
266  * return the pointer to the buffer.
267  */
268 static void *fill(int min)
269 {
270         if (min <= input_len)
271                 return input_buffer + input_offset;
272         if (min > sizeof(input_buffer))
273                 die(Q_("cannot fill %d byte",
274                        "cannot fill %d bytes",
275                        min),
276                     min);
277         flush();
278         do {
279                 ssize_t ret = xread(input_fd, input_buffer + input_len,
280                                 sizeof(input_buffer) - input_len);
281                 if (ret <= 0) {
282                         if (!ret)
283                                 die(_("early EOF"));
284                         die_errno(_("read error on input"));
285                 }
286                 input_len += ret;
287                 if (from_stdin)
288                         display_throughput(progress, consumed_bytes + input_len);
289         } while (input_len < min);
290         return input_buffer;
291 }
292
293 static void use(int bytes)
294 {
295         if (bytes > input_len)
296                 die(_("used more bytes than were available"));
297         input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
298         input_len -= bytes;
299         input_offset += bytes;
300
301         /* make sure off_t is sufficiently large not to wrap */
302         if (signed_add_overflows(consumed_bytes, bytes))
303                 die(_("pack too large for current definition of off_t"));
304         consumed_bytes += bytes;
305         if (max_input_size && consumed_bytes > max_input_size)
306                 die(_("pack exceeds maximum allowed size"));
307 }
308
309 static const char *open_pack_file(const char *pack_name)
310 {
311         if (from_stdin) {
312                 input_fd = 0;
313                 if (!pack_name) {
314                         struct strbuf tmp_file = STRBUF_INIT;
315                         output_fd = odb_mkstemp(&tmp_file,
316                                                 "pack/tmp_pack_XXXXXX");
317                         pack_name = strbuf_detach(&tmp_file, NULL);
318                 } else {
319                         output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
320                         if (output_fd < 0)
321                                 die_errno(_("unable to create '%s'"), pack_name);
322                 }
323                 nothread_data.pack_fd = output_fd;
324         } else {
325                 input_fd = open(pack_name, O_RDONLY);
326                 if (input_fd < 0)
327                         die_errno(_("cannot open packfile '%s'"), pack_name);
328                 output_fd = -1;
329                 nothread_data.pack_fd = input_fd;
330         }
331         the_hash_algo->init_fn(&input_ctx);
332         return pack_name;
333 }
334
335 static void parse_pack_header(void)
336 {
337         struct pack_header *hdr = fill(sizeof(struct pack_header));
338
339         /* Header consistency check */
340         if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
341                 die(_("pack signature mismatch"));
342         if (!pack_version_ok(hdr->hdr_version))
343                 die(_("pack version %"PRIu32" unsupported"),
344                         ntohl(hdr->hdr_version));
345
346         nr_objects = ntohl(hdr->hdr_entries);
347         use(sizeof(struct pack_header));
348 }
349
350 static NORETURN void bad_object(off_t offset, const char *format,
351                        ...) __attribute__((format (printf, 2, 3)));
352
353 static NORETURN void bad_object(off_t offset, const char *format, ...)
354 {
355         va_list params;
356         char buf[1024];
357
358         va_start(params, format);
359         vsnprintf(buf, sizeof(buf), format, params);
360         va_end(params);
361         die(_("pack has bad object at offset %"PRIuMAX": %s"),
362             (uintmax_t)offset, buf);
363 }
364
365 static inline struct thread_local *get_thread_data(void)
366 {
367 #ifndef NO_PTHREADS
368         if (threads_active)
369                 return pthread_getspecific(key);
370         assert(!threads_active &&
371                "This should only be reached when all threads are gone");
372 #endif
373         return &nothread_data;
374 }
375
376 #ifndef NO_PTHREADS
377 static void set_thread_data(struct thread_local *data)
378 {
379         if (threads_active)
380                 pthread_setspecific(key, data);
381 }
382 #endif
383
384 static struct base_data *alloc_base_data(void)
385 {
386         struct base_data *base = xcalloc(1, sizeof(struct base_data));
387         base->ref_last = -1;
388         base->ofs_last = -1;
389         return base;
390 }
391
392 static void free_base_data(struct base_data *c)
393 {
394         if (c->data) {
395                 FREE_AND_NULL(c->data);
396                 get_thread_data()->base_cache_used -= c->size;
397         }
398 }
399
400 static void prune_base_data(struct base_data *retain)
401 {
402         struct base_data *b;
403         struct thread_local *data = get_thread_data();
404         for (b = data->base_cache;
405              data->base_cache_used > delta_base_cache_limit && b;
406              b = b->child) {
407                 if (b->data && b != retain)
408                         free_base_data(b);
409         }
410 }
411
412 static void link_base_data(struct base_data *base, struct base_data *c)
413 {
414         if (base)
415                 base->child = c;
416         else
417                 get_thread_data()->base_cache = c;
418
419         c->base = base;
420         c->child = NULL;
421         if (c->data)
422                 get_thread_data()->base_cache_used += c->size;
423         prune_base_data(c);
424 }
425
426 static void unlink_base_data(struct base_data *c)
427 {
428         struct base_data *base = c->base;
429         if (base)
430                 base->child = NULL;
431         else
432                 get_thread_data()->base_cache = NULL;
433         free_base_data(c);
434 }
435
436 static int is_delta_type(enum object_type type)
437 {
438         return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
439 }
440
441 static void *unpack_entry_data(off_t offset, unsigned long size,
442                                enum object_type type, struct object_id *oid)
443 {
444         static char fixed_buf[8192];
445         int status;
446         git_zstream stream;
447         void *buf;
448         git_hash_ctx c;
449         char hdr[32];
450         int hdrlen;
451
452         if (!is_delta_type(type)) {
453                 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), size) + 1;
454                 the_hash_algo->init_fn(&c);
455                 the_hash_algo->update_fn(&c, hdr, hdrlen);
456         } else
457                 oid = NULL;
458         if (type == OBJ_BLOB && size > big_file_threshold)
459                 buf = fixed_buf;
460         else
461                 buf = xmallocz(size);
462
463         memset(&stream, 0, sizeof(stream));
464         git_inflate_init(&stream);
465         stream.next_out = buf;
466         stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
467
468         do {
469                 unsigned char *last_out = stream.next_out;
470                 stream.next_in = fill(1);
471                 stream.avail_in = input_len;
472                 status = git_inflate(&stream, 0);
473                 use(input_len - stream.avail_in);
474                 if (oid)
475                         the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out);
476                 if (buf == fixed_buf) {
477                         stream.next_out = buf;
478                         stream.avail_out = sizeof(fixed_buf);
479                 }
480         } while (status == Z_OK);
481         if (stream.total_out != size || status != Z_STREAM_END)
482                 bad_object(offset, _("inflate returned %d"), status);
483         git_inflate_end(&stream);
484         if (oid)
485                 the_hash_algo->final_fn(oid->hash, &c);
486         return buf == fixed_buf ? NULL : buf;
487 }
488
489 static void *unpack_raw_entry(struct object_entry *obj,
490                               off_t *ofs_offset,
491                               struct object_id *ref_oid,
492                               struct object_id *oid)
493 {
494         unsigned char *p;
495         unsigned long size, c;
496         off_t base_offset;
497         unsigned shift;
498         void *data;
499
500         obj->idx.offset = consumed_bytes;
501         input_crc32 = crc32(0, NULL, 0);
502
503         p = fill(1);
504         c = *p;
505         use(1);
506         obj->type = (c >> 4) & 7;
507         size = (c & 15);
508         shift = 4;
509         while (c & 0x80) {
510                 p = fill(1);
511                 c = *p;
512                 use(1);
513                 size += (c & 0x7f) << shift;
514                 shift += 7;
515         }
516         obj->size = size;
517
518         switch (obj->type) {
519         case OBJ_REF_DELTA:
520                 hashcpy(ref_oid->hash, fill(the_hash_algo->rawsz));
521                 use(the_hash_algo->rawsz);
522                 break;
523         case OBJ_OFS_DELTA:
524                 p = fill(1);
525                 c = *p;
526                 use(1);
527                 base_offset = c & 127;
528                 while (c & 128) {
529                         base_offset += 1;
530                         if (!base_offset || MSB(base_offset, 7))
531                                 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
532                         p = fill(1);
533                         c = *p;
534                         use(1);
535                         base_offset = (base_offset << 7) + (c & 127);
536                 }
537                 *ofs_offset = obj->idx.offset - base_offset;
538                 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
539                         bad_object(obj->idx.offset, _("delta base offset is out of bound"));
540                 break;
541         case OBJ_COMMIT:
542         case OBJ_TREE:
543         case OBJ_BLOB:
544         case OBJ_TAG:
545                 break;
546         default:
547                 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
548         }
549         obj->hdr_size = consumed_bytes - obj->idx.offset;
550
551         data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid);
552         obj->idx.crc32 = input_crc32;
553         return data;
554 }
555
556 static void *unpack_data(struct object_entry *obj,
557                          int (*consume)(const unsigned char *, unsigned long, void *),
558                          void *cb_data)
559 {
560         off_t from = obj[0].idx.offset + obj[0].hdr_size;
561         off_t len = obj[1].idx.offset - from;
562         unsigned char *data, *inbuf;
563         git_zstream stream;
564         int status;
565
566         data = xmallocz(consume ? 64*1024 : obj->size);
567         inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
568
569         memset(&stream, 0, sizeof(stream));
570         git_inflate_init(&stream);
571         stream.next_out = data;
572         stream.avail_out = consume ? 64*1024 : obj->size;
573
574         do {
575                 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
576                 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
577                 if (n < 0)
578                         die_errno(_("cannot pread pack file"));
579                 if (!n)
580                         die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
581                                "premature end of pack file, %"PRIuMAX" bytes missing",
582                                (unsigned int)len),
583                             (uintmax_t)len);
584                 from += n;
585                 len -= n;
586                 stream.next_in = inbuf;
587                 stream.avail_in = n;
588                 if (!consume)
589                         status = git_inflate(&stream, 0);
590                 else {
591                         do {
592                                 status = git_inflate(&stream, 0);
593                                 if (consume(data, stream.next_out - data, cb_data)) {
594                                         free(inbuf);
595                                         free(data);
596                                         return NULL;
597                                 }
598                                 stream.next_out = data;
599                                 stream.avail_out = 64*1024;
600                         } while (status == Z_OK && stream.avail_in);
601                 }
602         } while (len && status == Z_OK && !stream.avail_in);
603
604         /* This has been inflated OK when first encountered, so... */
605         if (status != Z_STREAM_END || stream.total_out != obj->size)
606                 die(_("serious inflate inconsistency"));
607
608         git_inflate_end(&stream);
609         free(inbuf);
610         if (consume) {
611                 FREE_AND_NULL(data);
612         }
613         return data;
614 }
615
616 static void *get_data_from_pack(struct object_entry *obj)
617 {
618         return unpack_data(obj, NULL, NULL);
619 }
620
621 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
622                                    enum object_type type1,
623                                    enum object_type type2)
624 {
625         int cmp = type1 - type2;
626         if (cmp)
627                 return cmp;
628         return offset1 < offset2 ? -1 :
629                offset1 > offset2 ?  1 :
630                0;
631 }
632
633 static int find_ofs_delta(const off_t offset, enum object_type type)
634 {
635         int first = 0, last = nr_ofs_deltas;
636
637         while (first < last) {
638                 int next = first + (last - first) / 2;
639                 struct ofs_delta_entry *delta = &ofs_deltas[next];
640                 int cmp;
641
642                 cmp = compare_ofs_delta_bases(offset, delta->offset,
643                                               type, objects[delta->obj_no].type);
644                 if (!cmp)
645                         return next;
646                 if (cmp < 0) {
647                         last = next;
648                         continue;
649                 }
650                 first = next+1;
651         }
652         return -first-1;
653 }
654
655 static void find_ofs_delta_children(off_t offset,
656                                     int *first_index, int *last_index,
657                                     enum object_type type)
658 {
659         int first = find_ofs_delta(offset, type);
660         int last = first;
661         int end = nr_ofs_deltas - 1;
662
663         if (first < 0) {
664                 *first_index = 0;
665                 *last_index = -1;
666                 return;
667         }
668         while (first > 0 && ofs_deltas[first - 1].offset == offset)
669                 --first;
670         while (last < end && ofs_deltas[last + 1].offset == offset)
671                 ++last;
672         *first_index = first;
673         *last_index = last;
674 }
675
676 static int compare_ref_delta_bases(const struct object_id *oid1,
677                                    const struct object_id *oid2,
678                                    enum object_type type1,
679                                    enum object_type type2)
680 {
681         int cmp = type1 - type2;
682         if (cmp)
683                 return cmp;
684         return oidcmp(oid1, oid2);
685 }
686
687 static int find_ref_delta(const struct object_id *oid, enum object_type type)
688 {
689         int first = 0, last = nr_ref_deltas;
690
691         while (first < last) {
692                 int next = first + (last - first) / 2;
693                 struct ref_delta_entry *delta = &ref_deltas[next];
694                 int cmp;
695
696                 cmp = compare_ref_delta_bases(oid, &delta->oid,
697                                               type, objects[delta->obj_no].type);
698                 if (!cmp)
699                         return next;
700                 if (cmp < 0) {
701                         last = next;
702                         continue;
703                 }
704                 first = next+1;
705         }
706         return -first-1;
707 }
708
709 static void find_ref_delta_children(const struct object_id *oid,
710                                     int *first_index, int *last_index,
711                                     enum object_type type)
712 {
713         int first = find_ref_delta(oid, type);
714         int last = first;
715         int end = nr_ref_deltas - 1;
716
717         if (first < 0) {
718                 *first_index = 0;
719                 *last_index = -1;
720                 return;
721         }
722         while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid))
723                 --first;
724         while (last < end && oideq(&ref_deltas[last + 1].oid, oid))
725                 ++last;
726         *first_index = first;
727         *last_index = last;
728 }
729
730 struct compare_data {
731         struct object_entry *entry;
732         struct git_istream *st;
733         unsigned char *buf;
734         unsigned long buf_size;
735 };
736
737 static int compare_objects(const unsigned char *buf, unsigned long size,
738                            void *cb_data)
739 {
740         struct compare_data *data = cb_data;
741
742         if (data->buf_size < size) {
743                 free(data->buf);
744                 data->buf = xmalloc(size);
745                 data->buf_size = size;
746         }
747
748         while (size) {
749                 ssize_t len = read_istream(data->st, data->buf, size);
750                 if (len == 0)
751                         die(_("SHA1 COLLISION FOUND WITH %s !"),
752                             oid_to_hex(&data->entry->idx.oid));
753                 if (len < 0)
754                         die(_("unable to read %s"),
755                             oid_to_hex(&data->entry->idx.oid));
756                 if (memcmp(buf, data->buf, len))
757                         die(_("SHA1 COLLISION FOUND WITH %s !"),
758                             oid_to_hex(&data->entry->idx.oid));
759                 size -= len;
760                 buf += len;
761         }
762         return 0;
763 }
764
765 static int check_collison(struct object_entry *entry)
766 {
767         struct compare_data data;
768         enum object_type type;
769         unsigned long size;
770
771         if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
772                 return -1;
773
774         memset(&data, 0, sizeof(data));
775         data.entry = entry;
776         data.st = open_istream(&entry->idx.oid, &type, &size, NULL);
777         if (!data.st)
778                 return -1;
779         if (size != entry->size || type != entry->type)
780                 die(_("SHA1 COLLISION FOUND WITH %s !"),
781                     oid_to_hex(&entry->idx.oid));
782         unpack_data(entry, compare_objects, &data);
783         close_istream(data.st);
784         free(data.buf);
785         return 0;
786 }
787
788 static void sha1_object(const void *data, struct object_entry *obj_entry,
789                         unsigned long size, enum object_type type,
790                         const struct object_id *oid)
791 {
792         void *new_data = NULL;
793         int collision_test_needed = 0;
794
795         assert(data || obj_entry);
796
797         if (startup_info->have_repository) {
798                 read_lock();
799                 collision_test_needed =
800                         has_sha1_file_with_flags(oid->hash, OBJECT_INFO_QUICK);
801                 read_unlock();
802         }
803
804         if (collision_test_needed && !data) {
805                 read_lock();
806                 if (!check_collison(obj_entry))
807                         collision_test_needed = 0;
808                 read_unlock();
809         }
810         if (collision_test_needed) {
811                 void *has_data;
812                 enum object_type has_type;
813                 unsigned long has_size;
814                 read_lock();
815                 has_type = oid_object_info(the_repository, oid, &has_size);
816                 if (has_type < 0)
817                         die(_("cannot read existing object info %s"), oid_to_hex(oid));
818                 if (has_type != type || has_size != size)
819                         die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
820                 has_data = read_object_file(oid, &has_type, &has_size);
821                 read_unlock();
822                 if (!data)
823                         data = new_data = get_data_from_pack(obj_entry);
824                 if (!has_data)
825                         die(_("cannot read existing object %s"), oid_to_hex(oid));
826                 if (size != has_size || type != has_type ||
827                     memcmp(data, has_data, size) != 0)
828                         die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid));
829                 free(has_data);
830         }
831
832         if (strict || do_fsck_object) {
833                 read_lock();
834                 if (type == OBJ_BLOB) {
835                         struct blob *blob = lookup_blob(the_repository, oid);
836                         if (blob)
837                                 blob->object.flags |= FLAG_CHECKED;
838                         else
839                                 die(_("invalid blob object %s"), oid_to_hex(oid));
840                         if (do_fsck_object &&
841                             fsck_object(&blob->object, (void *)data, size, &fsck_options))
842                                 die(_("fsck error in packed object"));
843                 } else {
844                         struct object *obj;
845                         int eaten;
846                         void *buf = (void *) data;
847
848                         assert(data && "data can only be NULL for large _blobs_");
849
850                         /*
851                          * we do not need to free the memory here, as the
852                          * buf is deleted by the caller.
853                          */
854                         obj = parse_object_buffer(the_repository, oid, type,
855                                                   size, buf,
856                                                   &eaten);
857                         if (!obj)
858                                 die(_("invalid %s"), type_name(type));
859                         if (do_fsck_object &&
860                             fsck_object(obj, buf, size, &fsck_options))
861                                 die(_("fsck error in packed object"));
862                         if (strict && fsck_walk(obj, NULL, &fsck_options))
863                                 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
864
865                         if (obj->type == OBJ_TREE) {
866                                 struct tree *item = (struct tree *) obj;
867                                 item->buffer = NULL;
868                                 obj->parsed = 0;
869                         }
870                         if (obj->type == OBJ_COMMIT) {
871                                 struct commit *commit = (struct commit *) obj;
872                                 if (detach_commit_buffer(commit, NULL) != data)
873                                         BUG("parse_object_buffer transmogrified our buffer");
874                         }
875                         obj->flags |= FLAG_CHECKED;
876                 }
877                 read_unlock();
878         }
879
880         free(new_data);
881 }
882
883 /*
884  * This function is part of find_unresolved_deltas(). There are two
885  * walkers going in the opposite ways.
886  *
887  * The first one in find_unresolved_deltas() traverses down from
888  * parent node to children, deflating nodes along the way. However,
889  * memory for deflated nodes is limited by delta_base_cache_limit, so
890  * at some point parent node's deflated content may be freed.
891  *
892  * The second walker is this function, which goes from current node up
893  * to top parent if necessary to deflate the node. In normal
894  * situation, its parent node would be already deflated, so it just
895  * needs to apply delta.
896  *
897  * In the worst case scenario, parent node is no longer deflated because
898  * we're running out of delta_base_cache_limit; we need to re-deflate
899  * parents, possibly up to the top base.
900  *
901  * All deflated objects here are subject to be freed if we exceed
902  * delta_base_cache_limit, just like in find_unresolved_deltas(), we
903  * just need to make sure the last node is not freed.
904  */
905 static void *get_base_data(struct base_data *c)
906 {
907         if (!c->data) {
908                 struct object_entry *obj = c->obj;
909                 struct base_data **delta = NULL;
910                 int delta_nr = 0, delta_alloc = 0;
911
912                 while (is_delta_type(c->obj->type) && !c->data) {
913                         ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
914                         delta[delta_nr++] = c;
915                         c = c->base;
916                 }
917                 if (!delta_nr) {
918                         c->data = get_data_from_pack(obj);
919                         c->size = obj->size;
920                         get_thread_data()->base_cache_used += c->size;
921                         prune_base_data(c);
922                 }
923                 for (; delta_nr > 0; delta_nr--) {
924                         void *base, *raw;
925                         c = delta[delta_nr - 1];
926                         obj = c->obj;
927                         base = get_base_data(c->base);
928                         raw = get_data_from_pack(obj);
929                         c->data = patch_delta(
930                                 base, c->base->size,
931                                 raw, obj->size,
932                                 &c->size);
933                         free(raw);
934                         if (!c->data)
935                                 bad_object(obj->idx.offset, _("failed to apply delta"));
936                         get_thread_data()->base_cache_used += c->size;
937                         prune_base_data(c);
938                 }
939                 free(delta);
940         }
941         return c->data;
942 }
943
944 static void resolve_delta(struct object_entry *delta_obj,
945                           struct base_data *base, struct base_data *result)
946 {
947         void *base_data, *delta_data;
948
949         if (show_stat) {
950                 int i = delta_obj - objects;
951                 int j = base->obj - objects;
952                 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
953                 deepest_delta_lock();
954                 if (deepest_delta < obj_stat[i].delta_depth)
955                         deepest_delta = obj_stat[i].delta_depth;
956                 deepest_delta_unlock();
957                 obj_stat[i].base_object_no = j;
958         }
959         delta_data = get_data_from_pack(delta_obj);
960         base_data = get_base_data(base);
961         result->obj = delta_obj;
962         result->data = patch_delta(base_data, base->size,
963                                    delta_data, delta_obj->size, &result->size);
964         free(delta_data);
965         if (!result->data)
966                 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
967         hash_object_file(result->data, result->size,
968                          type_name(delta_obj->real_type), &delta_obj->idx.oid);
969         sha1_object(result->data, NULL, result->size, delta_obj->real_type,
970                     &delta_obj->idx.oid);
971         counter_lock();
972         nr_resolved_deltas++;
973         counter_unlock();
974 }
975
976 /*
977  * Standard boolean compare-and-swap: atomically check whether "*type" is
978  * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
979  * and return false.
980  */
981 static int compare_and_swap_type(signed char *type,
982                                  enum object_type want,
983                                  enum object_type set)
984 {
985         enum object_type old;
986
987         type_cas_lock();
988         old = *type;
989         if (old == want)
990                 *type = set;
991         type_cas_unlock();
992
993         return old == want;
994 }
995
996 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
997                                                   struct base_data *prev_base)
998 {
999         if (base->ref_last == -1 && base->ofs_last == -1) {
1000                 find_ref_delta_children(&base->obj->idx.oid,
1001                                         &base->ref_first, &base->ref_last,
1002                                         OBJ_REF_DELTA);
1003
1004                 find_ofs_delta_children(base->obj->idx.offset,
1005                                         &base->ofs_first, &base->ofs_last,
1006                                         OBJ_OFS_DELTA);
1007
1008                 if (base->ref_last == -1 && base->ofs_last == -1) {
1009                         free(base->data);
1010                         return NULL;
1011                 }
1012
1013                 link_base_data(prev_base, base);
1014         }
1015
1016         if (base->ref_first <= base->ref_last) {
1017                 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1018                 struct base_data *result = alloc_base_data();
1019
1020                 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1021                                            base->obj->real_type))
1022                         BUG("child->real_type != OBJ_REF_DELTA");
1023
1024                 resolve_delta(child, base, result);
1025                 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1026                         free_base_data(base);
1027
1028                 base->ref_first++;
1029                 return result;
1030         }
1031
1032         if (base->ofs_first <= base->ofs_last) {
1033                 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1034                 struct base_data *result = alloc_base_data();
1035
1036                 assert(child->real_type == OBJ_OFS_DELTA);
1037                 child->real_type = base->obj->real_type;
1038                 resolve_delta(child, base, result);
1039                 if (base->ofs_first == base->ofs_last)
1040                         free_base_data(base);
1041
1042                 base->ofs_first++;
1043                 return result;
1044         }
1045
1046         unlink_base_data(base);
1047         return NULL;
1048 }
1049
1050 static void find_unresolved_deltas(struct base_data *base)
1051 {
1052         struct base_data *new_base, *prev_base = NULL;
1053         for (;;) {
1054                 new_base = find_unresolved_deltas_1(base, prev_base);
1055
1056                 if (new_base) {
1057                         prev_base = base;
1058                         base = new_base;
1059                 } else {
1060                         free(base);
1061                         base = prev_base;
1062                         if (!base)
1063                                 return;
1064                         prev_base = base->base;
1065                 }
1066         }
1067 }
1068
1069 static int compare_ofs_delta_entry(const void *a, const void *b)
1070 {
1071         const struct ofs_delta_entry *delta_a = a;
1072         const struct ofs_delta_entry *delta_b = b;
1073
1074         return delta_a->offset < delta_b->offset ? -1 :
1075                delta_a->offset > delta_b->offset ?  1 :
1076                0;
1077 }
1078
1079 static int compare_ref_delta_entry(const void *a, const void *b)
1080 {
1081         const struct ref_delta_entry *delta_a = a;
1082         const struct ref_delta_entry *delta_b = b;
1083
1084         return oidcmp(&delta_a->oid, &delta_b->oid);
1085 }
1086
1087 static void resolve_base(struct object_entry *obj)
1088 {
1089         struct base_data *base_obj = alloc_base_data();
1090         base_obj->obj = obj;
1091         base_obj->data = NULL;
1092         find_unresolved_deltas(base_obj);
1093 }
1094
1095 #ifndef NO_PTHREADS
1096 static void *threaded_second_pass(void *data)
1097 {
1098         set_thread_data(data);
1099         for (;;) {
1100                 int i;
1101                 counter_lock();
1102                 display_progress(progress, nr_resolved_deltas);
1103                 counter_unlock();
1104                 work_lock();
1105                 while (nr_dispatched < nr_objects &&
1106                        is_delta_type(objects[nr_dispatched].type))
1107                         nr_dispatched++;
1108                 if (nr_dispatched >= nr_objects) {
1109                         work_unlock();
1110                         break;
1111                 }
1112                 i = nr_dispatched++;
1113                 work_unlock();
1114
1115                 resolve_base(&objects[i]);
1116         }
1117         return NULL;
1118 }
1119 #endif
1120
1121 /*
1122  * First pass:
1123  * - find locations of all objects;
1124  * - calculate SHA1 of all non-delta objects;
1125  * - remember base (SHA1 or offset) for all deltas.
1126  */
1127 static void parse_pack_objects(unsigned char *hash)
1128 {
1129         int i, nr_delays = 0;
1130         struct ofs_delta_entry *ofs_delta = ofs_deltas;
1131         struct object_id ref_delta_oid;
1132         struct stat st;
1133
1134         if (verbose)
1135                 progress = start_progress(
1136                                 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1137                                 nr_objects);
1138         for (i = 0; i < nr_objects; i++) {
1139                 struct object_entry *obj = &objects[i];
1140                 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1141                                               &ref_delta_oid,
1142                                               &obj->idx.oid);
1143                 obj->real_type = obj->type;
1144                 if (obj->type == OBJ_OFS_DELTA) {
1145                         nr_ofs_deltas++;
1146                         ofs_delta->obj_no = i;
1147                         ofs_delta++;
1148                 } else if (obj->type == OBJ_REF_DELTA) {
1149                         ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1150                         oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid);
1151                         ref_deltas[nr_ref_deltas].obj_no = i;
1152                         nr_ref_deltas++;
1153                 } else if (!data) {
1154                         /* large blobs, check later */
1155                         obj->real_type = OBJ_BAD;
1156                         nr_delays++;
1157                 } else
1158                         sha1_object(data, NULL, obj->size, obj->type,
1159                                     &obj->idx.oid);
1160                 free(data);
1161                 display_progress(progress, i+1);
1162         }
1163         objects[i].idx.offset = consumed_bytes;
1164         stop_progress(&progress);
1165
1166         /* Check pack integrity */
1167         flush();
1168         the_hash_algo->final_fn(hash, &input_ctx);
1169         if (!hasheq(fill(the_hash_algo->rawsz), hash))
1170                 die(_("pack is corrupted (SHA1 mismatch)"));
1171         use(the_hash_algo->rawsz);
1172
1173         /* If input_fd is a file, we should have reached its end now. */
1174         if (fstat(input_fd, &st))
1175                 die_errno(_("cannot fstat packfile"));
1176         if (S_ISREG(st.st_mode) &&
1177                         lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1178                 die(_("pack has junk at the end"));
1179
1180         for (i = 0; i < nr_objects; i++) {
1181                 struct object_entry *obj = &objects[i];
1182                 if (obj->real_type != OBJ_BAD)
1183                         continue;
1184                 obj->real_type = obj->type;
1185                 sha1_object(NULL, obj, obj->size, obj->type,
1186                             &obj->idx.oid);
1187                 nr_delays--;
1188         }
1189         if (nr_delays)
1190                 die(_("confusion beyond insanity in parse_pack_objects()"));
1191 }
1192
1193 /*
1194  * Second pass:
1195  * - for all non-delta objects, look if it is used as a base for
1196  *   deltas;
1197  * - if used as a base, uncompress the object and apply all deltas,
1198  *   recursively checking if the resulting object is used as a base
1199  *   for some more deltas.
1200  */
1201 static void resolve_deltas(void)
1202 {
1203         int i;
1204
1205         if (!nr_ofs_deltas && !nr_ref_deltas)
1206                 return;
1207
1208         /* Sort deltas by base SHA1/offset for fast searching */
1209         QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry);
1210         QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry);
1211
1212         if (verbose || show_resolving_progress)
1213                 progress = start_progress(_("Resolving deltas"),
1214                                           nr_ref_deltas + nr_ofs_deltas);
1215
1216 #ifndef NO_PTHREADS
1217         nr_dispatched = 0;
1218         if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1219                 init_thread();
1220                 for (i = 0; i < nr_threads; i++) {
1221                         int ret = pthread_create(&thread_data[i].thread, NULL,
1222                                                  threaded_second_pass, thread_data + i);
1223                         if (ret)
1224                                 die(_("unable to create thread: %s"),
1225                                     strerror(ret));
1226                 }
1227                 for (i = 0; i < nr_threads; i++)
1228                         pthread_join(thread_data[i].thread, NULL);
1229                 cleanup_thread();
1230                 return;
1231         }
1232 #endif
1233
1234         for (i = 0; i < nr_objects; i++) {
1235                 struct object_entry *obj = &objects[i];
1236
1237                 if (is_delta_type(obj->type))
1238                         continue;
1239                 resolve_base(obj);
1240                 display_progress(progress, nr_resolved_deltas);
1241         }
1242 }
1243
1244 /*
1245  * Third pass:
1246  * - append objects to convert thin pack to full pack if required
1247  * - write the final pack hash
1248  */
1249 static void fix_unresolved_deltas(struct hashfile *f);
1250 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash)
1251 {
1252         if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1253                 stop_progress(&progress);
1254                 /* Flush remaining pack final hash. */
1255                 flush();
1256                 return;
1257         }
1258
1259         if (fix_thin_pack) {
1260                 struct hashfile *f;
1261                 unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ];
1262                 struct strbuf msg = STRBUF_INIT;
1263                 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1264                 int nr_objects_initial = nr_objects;
1265                 if (nr_unresolved <= 0)
1266                         die(_("confusion beyond insanity"));
1267                 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1268                 memset(objects + nr_objects + 1, 0,
1269                        nr_unresolved * sizeof(*objects));
1270                 f = hashfd(output_fd, curr_pack);
1271                 fix_unresolved_deltas(f);
1272                 strbuf_addf(&msg, Q_("completed with %d local object",
1273                                      "completed with %d local objects",
1274                                      nr_objects - nr_objects_initial),
1275                             nr_objects - nr_objects_initial);
1276                 stop_progress_msg(&progress, msg.buf);
1277                 strbuf_release(&msg);
1278                 finalize_hashfile(f, tail_hash, 0);
1279                 hashcpy(read_hash, pack_hash);
1280                 fixup_pack_header_footer(output_fd, pack_hash,
1281                                          curr_pack, nr_objects,
1282                                          read_hash, consumed_bytes-the_hash_algo->rawsz);
1283                 if (!hasheq(read_hash, tail_hash))
1284                         die(_("Unexpected tail checksum for %s "
1285                               "(disk corruption?)"), curr_pack);
1286         }
1287         if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1288                 die(Q_("pack has %d unresolved delta",
1289                        "pack has %d unresolved deltas",
1290                        nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1291                     nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1292 }
1293
1294 static int write_compressed(struct hashfile *f, void *in, unsigned int size)
1295 {
1296         git_zstream stream;
1297         int status;
1298         unsigned char outbuf[4096];
1299
1300         git_deflate_init(&stream, zlib_compression_level);
1301         stream.next_in = in;
1302         stream.avail_in = size;
1303
1304         do {
1305                 stream.next_out = outbuf;
1306                 stream.avail_out = sizeof(outbuf);
1307                 status = git_deflate(&stream, Z_FINISH);
1308                 hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out);
1309         } while (status == Z_OK);
1310
1311         if (status != Z_STREAM_END)
1312                 die(_("unable to deflate appended object (%d)"), status);
1313         size = stream.total_out;
1314         git_deflate_end(&stream);
1315         return size;
1316 }
1317
1318 static struct object_entry *append_obj_to_pack(struct hashfile *f,
1319                                const unsigned char *sha1, void *buf,
1320                                unsigned long size, enum object_type type)
1321 {
1322         struct object_entry *obj = &objects[nr_objects++];
1323         unsigned char header[10];
1324         unsigned long s = size;
1325         int n = 0;
1326         unsigned char c = (type << 4) | (s & 15);
1327         s >>= 4;
1328         while (s) {
1329                 header[n++] = c | 0x80;
1330                 c = s & 0x7f;
1331                 s >>= 7;
1332         }
1333         header[n++] = c;
1334         crc32_begin(f);
1335         hashwrite(f, header, n);
1336         obj[0].size = size;
1337         obj[0].hdr_size = n;
1338         obj[0].type = type;
1339         obj[0].real_type = type;
1340         obj[1].idx.offset = obj[0].idx.offset + n;
1341         obj[1].idx.offset += write_compressed(f, buf, size);
1342         obj[0].idx.crc32 = crc32_end(f);
1343         hashflush(f);
1344         hashcpy(obj->idx.oid.hash, sha1);
1345         return obj;
1346 }
1347
1348 static int delta_pos_compare(const void *_a, const void *_b)
1349 {
1350         struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1351         struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1352         return a->obj_no - b->obj_no;
1353 }
1354
1355 static void fix_unresolved_deltas(struct hashfile *f)
1356 {
1357         struct ref_delta_entry **sorted_by_pos;
1358         int i;
1359
1360         /*
1361          * Since many unresolved deltas may well be themselves base objects
1362          * for more unresolved deltas, we really want to include the
1363          * smallest number of base objects that would cover as much delta
1364          * as possible by picking the
1365          * trunc deltas first, allowing for other deltas to resolve without
1366          * additional base objects.  Since most base objects are to be found
1367          * before deltas depending on them, a good heuristic is to start
1368          * resolving deltas in the same order as their position in the pack.
1369          */
1370         ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1371         for (i = 0; i < nr_ref_deltas; i++)
1372                 sorted_by_pos[i] = &ref_deltas[i];
1373         QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
1374
1375         for (i = 0; i < nr_ref_deltas; i++) {
1376                 struct ref_delta_entry *d = sorted_by_pos[i];
1377                 enum object_type type;
1378                 struct base_data *base_obj = alloc_base_data();
1379
1380                 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1381                         continue;
1382                 base_obj->data = read_object_file(&d->oid, &type,
1383                                                   &base_obj->size);
1384                 if (!base_obj->data)
1385                         continue;
1386
1387                 if (check_object_signature(&d->oid, base_obj->data,
1388                                 base_obj->size, type_name(type)))
1389                         die(_("local object %s is corrupt"), oid_to_hex(&d->oid));
1390                 base_obj->obj = append_obj_to_pack(f, d->oid.hash,
1391                                         base_obj->data, base_obj->size, type);
1392                 find_unresolved_deltas(base_obj);
1393                 display_progress(progress, nr_resolved_deltas);
1394         }
1395         free(sorted_by_pos);
1396 }
1397
1398 static const char *derive_filename(const char *pack_name, const char *suffix,
1399                                    struct strbuf *buf)
1400 {
1401         size_t len;
1402         if (!strip_suffix(pack_name, ".pack", &len))
1403                 die(_("packfile name '%s' does not end with '.pack'"),
1404                     pack_name);
1405         strbuf_add(buf, pack_name, len);
1406         strbuf_addch(buf, '.');
1407         strbuf_addstr(buf, suffix);
1408         return buf->buf;
1409 }
1410
1411 static void write_special_file(const char *suffix, const char *msg,
1412                                const char *pack_name, const unsigned char *hash,
1413                                const char **report)
1414 {
1415         struct strbuf name_buf = STRBUF_INIT;
1416         const char *filename;
1417         int fd;
1418         int msg_len = strlen(msg);
1419
1420         if (pack_name)
1421                 filename = derive_filename(pack_name, suffix, &name_buf);
1422         else
1423                 filename = odb_pack_name(&name_buf, hash, suffix);
1424
1425         fd = odb_pack_keep(filename);
1426         if (fd < 0) {
1427                 if (errno != EEXIST)
1428                         die_errno(_("cannot write %s file '%s'"),
1429                                   suffix, filename);
1430         } else {
1431                 if (msg_len > 0) {
1432                         write_or_die(fd, msg, msg_len);
1433                         write_or_die(fd, "\n", 1);
1434                 }
1435                 if (close(fd) != 0)
1436                         die_errno(_("cannot close written %s file '%s'"),
1437                                   suffix, filename);
1438                 if (report)
1439                         *report = suffix;
1440         }
1441         strbuf_release(&name_buf);
1442 }
1443
1444 static void final(const char *final_pack_name, const char *curr_pack_name,
1445                   const char *final_index_name, const char *curr_index_name,
1446                   const char *keep_msg, const char *promisor_msg,
1447                   unsigned char *hash)
1448 {
1449         const char *report = "pack";
1450         struct strbuf pack_name = STRBUF_INIT;
1451         struct strbuf index_name = STRBUF_INIT;
1452         int err;
1453
1454         if (!from_stdin) {
1455                 close(input_fd);
1456         } else {
1457                 fsync_or_die(output_fd, curr_pack_name);
1458                 err = close(output_fd);
1459                 if (err)
1460                         die_errno(_("error while closing pack file"));
1461         }
1462
1463         if (keep_msg)
1464                 write_special_file("keep", keep_msg, final_pack_name, hash,
1465                                    &report);
1466         if (promisor_msg)
1467                 write_special_file("promisor", promisor_msg, final_pack_name,
1468                                    hash, NULL);
1469
1470         if (final_pack_name != curr_pack_name) {
1471                 if (!final_pack_name)
1472                         final_pack_name = odb_pack_name(&pack_name, hash, "pack");
1473                 if (finalize_object_file(curr_pack_name, final_pack_name))
1474                         die(_("cannot store pack file"));
1475         } else if (from_stdin)
1476                 chmod(final_pack_name, 0444);
1477
1478         if (final_index_name != curr_index_name) {
1479                 if (!final_index_name)
1480                         final_index_name = odb_pack_name(&index_name, hash, "idx");
1481                 if (finalize_object_file(curr_index_name, final_index_name))
1482                         die(_("cannot store index file"));
1483         } else
1484                 chmod(final_index_name, 0444);
1485
1486         if (do_fsck_object) {
1487                 struct packed_git *p;
1488                 p = add_packed_git(final_index_name, strlen(final_index_name), 0);
1489                 if (p)
1490                         install_packed_git(the_repository, p);
1491         }
1492
1493         if (!from_stdin) {
1494                 printf("%s\n", sha1_to_hex(hash));
1495         } else {
1496                 struct strbuf buf = STRBUF_INIT;
1497
1498                 strbuf_addf(&buf, "%s\t%s\n", report, sha1_to_hex(hash));
1499                 write_or_die(1, buf.buf, buf.len);
1500                 strbuf_release(&buf);
1501
1502                 /*
1503                  * Let's just mimic git-unpack-objects here and write
1504                  * the last part of the input buffer to stdout.
1505                  */
1506                 while (input_len) {
1507                         err = xwrite(1, input_buffer + input_offset, input_len);
1508                         if (err <= 0)
1509                                 break;
1510                         input_len -= err;
1511                         input_offset += err;
1512                 }
1513         }
1514
1515         strbuf_release(&index_name);
1516         strbuf_release(&pack_name);
1517 }
1518
1519 static int git_index_pack_config(const char *k, const char *v, void *cb)
1520 {
1521         struct pack_idx_option *opts = cb;
1522
1523         if (!strcmp(k, "pack.indexversion")) {
1524                 opts->version = git_config_int(k, v);
1525                 if (opts->version > 2)
1526                         die(_("bad pack.indexversion=%"PRIu32), opts->version);
1527                 return 0;
1528         }
1529         if (!strcmp(k, "pack.threads")) {
1530                 nr_threads = git_config_int(k, v);
1531                 if (nr_threads < 0)
1532                         die(_("invalid number of threads specified (%d)"),
1533                             nr_threads);
1534 #ifdef NO_PTHREADS
1535                 if (nr_threads != 1)
1536                         warning(_("no threads support, ignoring %s"), k);
1537                 nr_threads = 1;
1538 #endif
1539                 return 0;
1540         }
1541         return git_default_config(k, v, cb);
1542 }
1543
1544 static int cmp_uint32(const void *a_, const void *b_)
1545 {
1546         uint32_t a = *((uint32_t *)a_);
1547         uint32_t b = *((uint32_t *)b_);
1548
1549         return (a < b) ? -1 : (a != b);
1550 }
1551
1552 static void read_v2_anomalous_offsets(struct packed_git *p,
1553                                       struct pack_idx_option *opts)
1554 {
1555         const uint32_t *idx1, *idx2;
1556         uint32_t i;
1557         const uint32_t hashwords = the_hash_algo->rawsz / sizeof(uint32_t);
1558
1559         /* The address of the 4-byte offset table */
1560         idx1 = (((const uint32_t *)p->index_data)
1561                 + 2 /* 8-byte header */
1562                 + 256 /* fan out */
1563                 + hashwords * p->num_objects /* object ID table */
1564                 + p->num_objects /* CRC32 table */
1565                 );
1566
1567         /* The address of the 8-byte offset table */
1568         idx2 = idx1 + p->num_objects;
1569
1570         for (i = 0; i < p->num_objects; i++) {
1571                 uint32_t off = ntohl(idx1[i]);
1572                 if (!(off & 0x80000000))
1573                         continue;
1574                 off = off & 0x7fffffff;
1575                 check_pack_index_ptr(p, &idx2[off * 2]);
1576                 if (idx2[off * 2])
1577                         continue;
1578                 /*
1579                  * The real offset is ntohl(idx2[off * 2]) in high 4
1580                  * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1581                  * octets.  But idx2[off * 2] is Zero!!!
1582                  */
1583                 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1584                 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1585         }
1586
1587         QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32);
1588 }
1589
1590 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1591 {
1592         struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1593
1594         if (!p)
1595                 die(_("Cannot open existing pack file '%s'"), pack_name);
1596         if (open_pack_index(p))
1597                 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1598
1599         /* Read the attributes from the existing idx file */
1600         opts->version = p->index_version;
1601
1602         if (opts->version == 2)
1603                 read_v2_anomalous_offsets(p, opts);
1604
1605         /*
1606          * Get rid of the idx file as we do not need it anymore.
1607          * NEEDSWORK: extract this bit from free_pack_by_name() in
1608          * sha1-file.c, perhaps?  It shouldn't matter very much as we
1609          * know we haven't installed this pack (hence we never have
1610          * read anything from it).
1611          */
1612         close_pack_index(p);
1613         free(p);
1614 }
1615
1616 static void show_pack_info(int stat_only)
1617 {
1618         int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1619         unsigned long *chain_histogram = NULL;
1620
1621         if (deepest_delta)
1622                 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1623
1624         for (i = 0; i < nr_objects; i++) {
1625                 struct object_entry *obj = &objects[i];
1626
1627                 if (is_delta_type(obj->type))
1628                         chain_histogram[obj_stat[i].delta_depth - 1]++;
1629                 if (stat_only)
1630                         continue;
1631                 printf("%s %-6s %lu %lu %"PRIuMAX,
1632                        oid_to_hex(&obj->idx.oid),
1633                        type_name(obj->real_type), obj->size,
1634                        (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1635                        (uintmax_t)obj->idx.offset);
1636                 if (is_delta_type(obj->type)) {
1637                         struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1638                         printf(" %u %s", obj_stat[i].delta_depth,
1639                                oid_to_hex(&bobj->idx.oid));
1640                 }
1641                 putchar('\n');
1642         }
1643
1644         if (baseobjects)
1645                 printf_ln(Q_("non delta: %d object",
1646                              "non delta: %d objects",
1647                              baseobjects),
1648                           baseobjects);
1649         for (i = 0; i < deepest_delta; i++) {
1650                 if (!chain_histogram[i])
1651                         continue;
1652                 printf_ln(Q_("chain length = %d: %lu object",
1653                              "chain length = %d: %lu objects",
1654                              chain_histogram[i]),
1655                           i + 1,
1656                           chain_histogram[i]);
1657         }
1658 }
1659
1660 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1661 {
1662         int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1663         const char *curr_index;
1664         const char *index_name = NULL, *pack_name = NULL;
1665         const char *keep_msg = NULL;
1666         const char *promisor_msg = NULL;
1667         struct strbuf index_name_buf = STRBUF_INIT;
1668         struct pack_idx_entry **idx_objects;
1669         struct pack_idx_option opts;
1670         unsigned char pack_hash[GIT_MAX_RAWSZ];
1671         unsigned foreign_nr = 1;        /* zero is a "good" value, assume bad */
1672         int report_end_of_input = 0;
1673
1674         /*
1675          * index-pack never needs to fetch missing objects, since it only
1676          * accesses the repo to do hash collision checks
1677          */
1678         fetch_if_missing = 0;
1679
1680         if (argc == 2 && !strcmp(argv[1], "-h"))
1681                 usage(index_pack_usage);
1682
1683         read_replace_refs = 0;
1684         fsck_options.walk = mark_link;
1685
1686         reset_pack_idx_option(&opts);
1687         git_config(git_index_pack_config, &opts);
1688         if (prefix && chdir(prefix))
1689                 die(_("Cannot come back to cwd"));
1690
1691         for (i = 1; i < argc; i++) {
1692                 const char *arg = argv[i];
1693
1694                 if (*arg == '-') {
1695                         if (!strcmp(arg, "--stdin")) {
1696                                 from_stdin = 1;
1697                         } else if (!strcmp(arg, "--fix-thin")) {
1698                                 fix_thin_pack = 1;
1699                         } else if (skip_to_optional_arg(arg, "--strict", &arg)) {
1700                                 strict = 1;
1701                                 do_fsck_object = 1;
1702                                 fsck_set_msg_types(&fsck_options, arg);
1703                         } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1704                                 strict = 1;
1705                                 check_self_contained_and_connected = 1;
1706                         } else if (!strcmp(arg, "--fsck-objects")) {
1707                                 do_fsck_object = 1;
1708                         } else if (!strcmp(arg, "--verify")) {
1709                                 verify = 1;
1710                         } else if (!strcmp(arg, "--verify-stat")) {
1711                                 verify = 1;
1712                                 show_stat = 1;
1713                         } else if (!strcmp(arg, "--verify-stat-only")) {
1714                                 verify = 1;
1715                                 show_stat = 1;
1716                                 stat_only = 1;
1717                         } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
1718                                 ; /* nothing to do */
1719                         } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
1720                                 ; /* already parsed */
1721                         } else if (starts_with(arg, "--threads=")) {
1722                                 char *end;
1723                                 nr_threads = strtoul(arg+10, &end, 0);
1724                                 if (!arg[10] || *end || nr_threads < 0)
1725                                         usage(index_pack_usage);
1726 #ifdef NO_PTHREADS
1727                                 if (nr_threads != 1)
1728                                         warning(_("no threads support, "
1729                                                   "ignoring %s"), arg);
1730                                 nr_threads = 1;
1731 #endif
1732                         } else if (starts_with(arg, "--pack_header=")) {
1733                                 struct pack_header *hdr;
1734                                 char *c;
1735
1736                                 hdr = (struct pack_header *)input_buffer;
1737                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1738                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1739                                 if (*c != ',')
1740                                         die(_("bad %s"), arg);
1741                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1742                                 if (*c)
1743                                         die(_("bad %s"), arg);
1744                                 input_len = sizeof(*hdr);
1745                         } else if (!strcmp(arg, "-v")) {
1746                                 verbose = 1;
1747                         } else if (!strcmp(arg, "--show-resolving-progress")) {
1748                                 show_resolving_progress = 1;
1749                         } else if (!strcmp(arg, "--report-end-of-input")) {
1750                                 report_end_of_input = 1;
1751                         } else if (!strcmp(arg, "-o")) {
1752                                 if (index_name || (i+1) >= argc)
1753                                         usage(index_pack_usage);
1754                                 index_name = argv[++i];
1755                         } else if (starts_with(arg, "--index-version=")) {
1756                                 char *c;
1757                                 opts.version = strtoul(arg + 16, &c, 10);
1758                                 if (opts.version > 2)
1759                                         die(_("bad %s"), arg);
1760                                 if (*c == ',')
1761                                         opts.off32_limit = strtoul(c+1, &c, 0);
1762                                 if (*c || opts.off32_limit & 0x80000000)
1763                                         die(_("bad %s"), arg);
1764                         } else if (skip_prefix(arg, "--max-input-size=", &arg)) {
1765                                 max_input_size = strtoumax(arg, NULL, 10);
1766                         } else
1767                                 usage(index_pack_usage);
1768                         continue;
1769                 }
1770
1771                 if (pack_name)
1772                         usage(index_pack_usage);
1773                 pack_name = arg;
1774         }
1775
1776         if (!pack_name && !from_stdin)
1777                 usage(index_pack_usage);
1778         if (fix_thin_pack && !from_stdin)
1779                 die(_("--fix-thin cannot be used without --stdin"));
1780         if (from_stdin && !startup_info->have_repository)
1781                 die(_("--stdin requires a git repository"));
1782         if (!index_name && pack_name)
1783                 index_name = derive_filename(pack_name, "idx", &index_name_buf);
1784
1785         if (verify) {
1786                 if (!index_name)
1787                         die(_("--verify with no packfile name given"));
1788                 read_idx_option(&opts, index_name);
1789                 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1790         }
1791         if (strict)
1792                 opts.flags |= WRITE_IDX_STRICT;
1793
1794 #ifndef NO_PTHREADS
1795         if (!nr_threads) {
1796                 nr_threads = online_cpus();
1797                 /* An experiment showed that more threads does not mean faster */
1798                 if (nr_threads > 3)
1799                         nr_threads = 3;
1800         }
1801 #endif
1802
1803         curr_pack = open_pack_file(pack_name);
1804         parse_pack_header();
1805         objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1806         if (show_stat)
1807                 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1808         ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1809         parse_pack_objects(pack_hash);
1810         if (report_end_of_input)
1811                 write_in_full(2, "\0", 1);
1812         resolve_deltas();
1813         conclude_pack(fix_thin_pack, curr_pack, pack_hash);
1814         free(ofs_deltas);
1815         free(ref_deltas);
1816         if (strict)
1817                 foreign_nr = check_objects();
1818
1819         if (show_stat)
1820                 show_pack_info(stat_only);
1821
1822         ALLOC_ARRAY(idx_objects, nr_objects);
1823         for (i = 0; i < nr_objects; i++)
1824                 idx_objects[i] = &objects[i].idx;
1825         curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash);
1826         free(idx_objects);
1827
1828         if (!verify)
1829                 final(pack_name, curr_pack,
1830                       index_name, curr_index,
1831                       keep_msg, promisor_msg,
1832                       pack_hash);
1833         else
1834                 close(input_fd);
1835
1836         if (do_fsck_object && fsck_finish(&fsck_options))
1837                 die(_("fsck error in pack objects"));
1838
1839         free(objects);
1840         strbuf_release(&index_name_buf);
1841         if (pack_name == NULL)
1842                 free((void *) curr_pack);
1843         if (index_name == NULL)
1844                 free((void *) curr_index);
1845
1846         /*
1847          * Let the caller know this pack is not self contained
1848          */
1849         if (check_self_contained_and_connected && foreign_nr)
1850                 return 1;
1851
1852         return 0;
1853 }