Merge branch 'nd/pack-ofs-4gb-limit'
[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 char hdr_size;
22         signed char type;
23         signed char real_type;
24 };
25
26 struct object_stat {
27         unsigned delta_depth;
28         int base_object_no;
29 };
30
31 struct base_data {
32         struct base_data *base;
33         struct base_data *child;
34         struct object_entry *obj;
35         void *data;
36         unsigned long size;
37         int ref_first, ref_last;
38         int ofs_first, ofs_last;
39 };
40
41 struct thread_local {
42 #ifndef NO_PTHREADS
43         pthread_t thread;
44 #endif
45         struct base_data *base_cache;
46         size_t base_cache_used;
47         int pack_fd;
48 };
49
50 #define FLAG_LINK (1u<<20)
51 #define FLAG_CHECKED (1u<<21)
52
53 struct ofs_delta_entry {
54         off_t offset;
55         int obj_no;
56 };
57
58 struct ref_delta_entry {
59         unsigned char sha1[20];
60         int obj_no;
61 };
62
63 static struct object_entry *objects;
64 static struct object_stat *obj_stat;
65 static struct ofs_delta_entry *ofs_deltas;
66 static struct ref_delta_entry *ref_deltas;
67 static struct thread_local nothread_data;
68 static int nr_objects;
69 static int nr_ofs_deltas;
70 static int nr_ref_deltas;
71 static int ref_deltas_alloc;
72 static int nr_resolved_deltas;
73 static int nr_threads;
74
75 static int from_stdin;
76 static int strict;
77 static int do_fsck_object;
78 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
79 static int verbose;
80 static int show_stat;
81 static int check_self_contained_and_connected;
82
83 static struct progress *progress;
84
85 /* We always read in 4kB chunks. */
86 static unsigned char input_buffer[4096];
87 static unsigned int input_offset, input_len;
88 static off_t consumed_bytes;
89 static unsigned deepest_delta;
90 static git_SHA_CTX input_ctx;
91 static uint32_t input_crc32;
92 static int input_fd, output_fd;
93 static const char *curr_pack;
94
95 #ifndef NO_PTHREADS
96
97 static struct thread_local *thread_data;
98 static int nr_dispatched;
99 static int threads_active;
100
101 static pthread_mutex_t read_mutex;
102 #define read_lock()             lock_mutex(&read_mutex)
103 #define read_unlock()           unlock_mutex(&read_mutex)
104
105 static pthread_mutex_t counter_mutex;
106 #define counter_lock()          lock_mutex(&counter_mutex)
107 #define counter_unlock()        unlock_mutex(&counter_mutex)
108
109 static pthread_mutex_t work_mutex;
110 #define work_lock()             lock_mutex(&work_mutex)
111 #define work_unlock()           unlock_mutex(&work_mutex)
112
113 static pthread_mutex_t deepest_delta_mutex;
114 #define deepest_delta_lock()    lock_mutex(&deepest_delta_mutex)
115 #define deepest_delta_unlock()  unlock_mutex(&deepest_delta_mutex)
116
117 static pthread_mutex_t type_cas_mutex;
118 #define type_cas_lock()         lock_mutex(&type_cas_mutex)
119 #define type_cas_unlock()       unlock_mutex(&type_cas_mutex)
120
121 static pthread_key_t key;
122
123 static inline void lock_mutex(pthread_mutex_t *mutex)
124 {
125         if (threads_active)
126                 pthread_mutex_lock(mutex);
127 }
128
129 static inline void unlock_mutex(pthread_mutex_t *mutex)
130 {
131         if (threads_active)
132                 pthread_mutex_unlock(mutex);
133 }
134
135 /*
136  * Mutex and conditional variable can't be statically-initialized on Windows.
137  */
138 static void init_thread(void)
139 {
140         int i;
141         init_recursive_mutex(&read_mutex);
142         pthread_mutex_init(&counter_mutex, NULL);
143         pthread_mutex_init(&work_mutex, NULL);
144         pthread_mutex_init(&type_cas_mutex, NULL);
145         if (show_stat)
146                 pthread_mutex_init(&deepest_delta_mutex, NULL);
147         pthread_key_create(&key, NULL);
148         thread_data = xcalloc(nr_threads, sizeof(*thread_data));
149         for (i = 0; i < nr_threads; i++) {
150                 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
151                 if (thread_data[i].pack_fd == -1)
152                         die_errno(_("unable to open %s"), curr_pack);
153         }
154
155         threads_active = 1;
156 }
157
158 static void cleanup_thread(void)
159 {
160         int i;
161         if (!threads_active)
162                 return;
163         threads_active = 0;
164         pthread_mutex_destroy(&read_mutex);
165         pthread_mutex_destroy(&counter_mutex);
166         pthread_mutex_destroy(&work_mutex);
167         pthread_mutex_destroy(&type_cas_mutex);
168         if (show_stat)
169                 pthread_mutex_destroy(&deepest_delta_mutex);
170         for (i = 0; i < nr_threads; i++)
171                 close(thread_data[i].pack_fd);
172         pthread_key_delete(key);
173         free(thread_data);
174 }
175
176 #else
177
178 #define read_lock()
179 #define read_unlock()
180
181 #define counter_lock()
182 #define counter_unlock()
183
184 #define work_lock()
185 #define work_unlock()
186
187 #define deepest_delta_lock()
188 #define deepest_delta_unlock()
189
190 #define type_cas_lock()
191 #define type_cas_unlock()
192
193 #endif
194
195
196 static int mark_link(struct object *obj, int type, void *data, struct fsck_options *options)
197 {
198         if (!obj)
199                 return -1;
200
201         if (type != OBJ_ANY && obj->type != type)
202                 die(_("object type mismatch at %s"), oid_to_hex(&obj->oid));
203
204         obj->flags |= FLAG_LINK;
205         return 0;
206 }
207
208 /* The content of each linked object must have been checked
209    or it must be already present in the object database */
210 static unsigned check_object(struct object *obj)
211 {
212         if (!obj)
213                 return 0;
214
215         if (!(obj->flags & FLAG_LINK))
216                 return 0;
217
218         if (!(obj->flags & FLAG_CHECKED)) {
219                 unsigned long size;
220                 int type = sha1_object_info(obj->oid.hash, &size);
221                 if (type <= 0)
222                         die(_("did not receive expected object %s"),
223                               oid_to_hex(&obj->oid));
224                 if (type != obj->type)
225                         die(_("object %s: expected type %s, found %s"),
226                             oid_to_hex(&obj->oid),
227                             typename(obj->type), typename(type));
228                 obj->flags |= FLAG_CHECKED;
229                 return 1;
230         }
231
232         return 0;
233 }
234
235 static unsigned check_objects(void)
236 {
237         unsigned i, max, foreign_nr = 0;
238
239         max = get_max_object_index();
240         for (i = 0; i < max; i++)
241                 foreign_nr += check_object(get_indexed_object(i));
242         return foreign_nr;
243 }
244
245
246 /* Discard current buffer used content. */
247 static void flush(void)
248 {
249         if (input_offset) {
250                 if (output_fd >= 0)
251                         write_or_die(output_fd, input_buffer, input_offset);
252                 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
253                 memmove(input_buffer, input_buffer + input_offset, input_len);
254                 input_offset = 0;
255         }
256 }
257
258 /*
259  * Make sure at least "min" bytes are available in the buffer, and
260  * return the pointer to the buffer.
261  */
262 static void *fill(int min)
263 {
264         if (min <= input_len)
265                 return input_buffer + input_offset;
266         if (min > sizeof(input_buffer))
267                 die(Q_("cannot fill %d byte",
268                        "cannot fill %d bytes",
269                        min),
270                     min);
271         flush();
272         do {
273                 ssize_t ret = xread(input_fd, input_buffer + input_len,
274                                 sizeof(input_buffer) - input_len);
275                 if (ret <= 0) {
276                         if (!ret)
277                                 die(_("early EOF"));
278                         die_errno(_("read error on input"));
279                 }
280                 input_len += ret;
281                 if (from_stdin)
282                         display_throughput(progress, consumed_bytes + input_len);
283         } while (input_len < min);
284         return input_buffer;
285 }
286
287 static void use(int bytes)
288 {
289         if (bytes > input_len)
290                 die(_("used more bytes than were available"));
291         input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
292         input_len -= bytes;
293         input_offset += bytes;
294
295         /* make sure off_t is sufficiently large not to wrap */
296         if (signed_add_overflows(consumed_bytes, bytes))
297                 die(_("pack too large for current definition of off_t"));
298         consumed_bytes += bytes;
299 }
300
301 static const char *open_pack_file(const char *pack_name)
302 {
303         if (from_stdin) {
304                 input_fd = 0;
305                 if (!pack_name) {
306                         static char tmp_file[PATH_MAX];
307                         output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
308                                                 "pack/tmp_pack_XXXXXX");
309                         pack_name = xstrdup(tmp_file);
310                 } else
311                         output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
312                 if (output_fd < 0)
313                         die_errno(_("unable to create '%s'"), pack_name);
314                 nothread_data.pack_fd = output_fd;
315         } else {
316                 input_fd = open(pack_name, O_RDONLY);
317                 if (input_fd < 0)
318                         die_errno(_("cannot open packfile '%s'"), pack_name);
319                 output_fd = -1;
320                 nothread_data.pack_fd = input_fd;
321         }
322         git_SHA1_Init(&input_ctx);
323         return pack_name;
324 }
325
326 static void parse_pack_header(void)
327 {
328         struct pack_header *hdr = fill(sizeof(struct pack_header));
329
330         /* Header consistency check */
331         if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
332                 die(_("pack signature mismatch"));
333         if (!pack_version_ok(hdr->hdr_version))
334                 die(_("pack version %"PRIu32" unsupported"),
335                         ntohl(hdr->hdr_version));
336
337         nr_objects = ntohl(hdr->hdr_entries);
338         use(sizeof(struct pack_header));
339 }
340
341 static NORETURN void bad_object(off_t offset, const char *format,
342                        ...) __attribute__((format (printf, 2, 3)));
343
344 static NORETURN void bad_object(off_t offset, const char *format, ...)
345 {
346         va_list params;
347         char buf[1024];
348
349         va_start(params, format);
350         vsnprintf(buf, sizeof(buf), format, params);
351         va_end(params);
352         die(_("pack has bad object at offset %"PRIuMAX": %s"),
353             (uintmax_t)offset, buf);
354 }
355
356 static inline struct thread_local *get_thread_data(void)
357 {
358 #ifndef NO_PTHREADS
359         if (threads_active)
360                 return pthread_getspecific(key);
361         assert(!threads_active &&
362                "This should only be reached when all threads are gone");
363 #endif
364         return &nothread_data;
365 }
366
367 #ifndef NO_PTHREADS
368 static void set_thread_data(struct thread_local *data)
369 {
370         if (threads_active)
371                 pthread_setspecific(key, data);
372 }
373 #endif
374
375 static struct base_data *alloc_base_data(void)
376 {
377         struct base_data *base = xcalloc(1, sizeof(struct base_data));
378         base->ref_last = -1;
379         base->ofs_last = -1;
380         return base;
381 }
382
383 static void free_base_data(struct base_data *c)
384 {
385         if (c->data) {
386                 free(c->data);
387                 c->data = NULL;
388                 get_thread_data()->base_cache_used -= c->size;
389         }
390 }
391
392 static void prune_base_data(struct base_data *retain)
393 {
394         struct base_data *b;
395         struct thread_local *data = get_thread_data();
396         for (b = data->base_cache;
397              data->base_cache_used > delta_base_cache_limit && b;
398              b = b->child) {
399                 if (b->data && b != retain)
400                         free_base_data(b);
401         }
402 }
403
404 static void link_base_data(struct base_data *base, struct base_data *c)
405 {
406         if (base)
407                 base->child = c;
408         else
409                 get_thread_data()->base_cache = c;
410
411         c->base = base;
412         c->child = NULL;
413         if (c->data)
414                 get_thread_data()->base_cache_used += c->size;
415         prune_base_data(c);
416 }
417
418 static void unlink_base_data(struct base_data *c)
419 {
420         struct base_data *base = c->base;
421         if (base)
422                 base->child = NULL;
423         else
424                 get_thread_data()->base_cache = NULL;
425         free_base_data(c);
426 }
427
428 static int is_delta_type(enum object_type type)
429 {
430         return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
431 }
432
433 static void *unpack_entry_data(off_t offset, unsigned long size,
434                                enum object_type type, unsigned char *sha1)
435 {
436         static char fixed_buf[8192];
437         int status;
438         git_zstream stream;
439         void *buf;
440         git_SHA_CTX c;
441         char hdr[32];
442         int hdrlen;
443
444         if (!is_delta_type(type)) {
445                 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), size) + 1;
446                 git_SHA1_Init(&c);
447                 git_SHA1_Update(&c, hdr, hdrlen);
448         } else
449                 sha1 = NULL;
450         if (type == OBJ_BLOB && size > big_file_threshold)
451                 buf = fixed_buf;
452         else
453                 buf = xmallocz(size);
454
455         memset(&stream, 0, sizeof(stream));
456         git_inflate_init(&stream);
457         stream.next_out = buf;
458         stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
459
460         do {
461                 unsigned char *last_out = stream.next_out;
462                 stream.next_in = fill(1);
463                 stream.avail_in = input_len;
464                 status = git_inflate(&stream, 0);
465                 use(input_len - stream.avail_in);
466                 if (sha1)
467                         git_SHA1_Update(&c, last_out, stream.next_out - last_out);
468                 if (buf == fixed_buf) {
469                         stream.next_out = buf;
470                         stream.avail_out = sizeof(fixed_buf);
471                 }
472         } while (status == Z_OK);
473         if (stream.total_out != size || status != Z_STREAM_END)
474                 bad_object(offset, _("inflate returned %d"), status);
475         git_inflate_end(&stream);
476         if (sha1)
477                 git_SHA1_Final(sha1, &c);
478         return buf == fixed_buf ? NULL : buf;
479 }
480
481 static void *unpack_raw_entry(struct object_entry *obj,
482                               off_t *ofs_offset,
483                               unsigned char *ref_sha1,
484                               unsigned char *sha1)
485 {
486         unsigned char *p;
487         unsigned long size, c;
488         off_t base_offset;
489         unsigned shift;
490         void *data;
491
492         obj->idx.offset = consumed_bytes;
493         input_crc32 = crc32(0, NULL, 0);
494
495         p = fill(1);
496         c = *p;
497         use(1);
498         obj->type = (c >> 4) & 7;
499         size = (c & 15);
500         shift = 4;
501         while (c & 0x80) {
502                 p = fill(1);
503                 c = *p;
504                 use(1);
505                 size += (c & 0x7f) << shift;
506                 shift += 7;
507         }
508         obj->size = size;
509
510         switch (obj->type) {
511         case OBJ_REF_DELTA:
512                 hashcpy(ref_sha1, fill(20));
513                 use(20);
514                 break;
515         case OBJ_OFS_DELTA:
516                 p = fill(1);
517                 c = *p;
518                 use(1);
519                 base_offset = c & 127;
520                 while (c & 128) {
521                         base_offset += 1;
522                         if (!base_offset || MSB(base_offset, 7))
523                                 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
524                         p = fill(1);
525                         c = *p;
526                         use(1);
527                         base_offset = (base_offset << 7) + (c & 127);
528                 }
529                 *ofs_offset = obj->idx.offset - base_offset;
530                 if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset)
531                         bad_object(obj->idx.offset, _("delta base offset is out of bound"));
532                 break;
533         case OBJ_COMMIT:
534         case OBJ_TREE:
535         case OBJ_BLOB:
536         case OBJ_TAG:
537                 break;
538         default:
539                 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
540         }
541         obj->hdr_size = consumed_bytes - obj->idx.offset;
542
543         data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
544         obj->idx.crc32 = input_crc32;
545         return data;
546 }
547
548 static void *unpack_data(struct object_entry *obj,
549                          int (*consume)(const unsigned char *, unsigned long, void *),
550                          void *cb_data)
551 {
552         off_t from = obj[0].idx.offset + obj[0].hdr_size;
553         off_t len = obj[1].idx.offset - from;
554         unsigned char *data, *inbuf;
555         git_zstream stream;
556         int status;
557
558         data = xmallocz(consume ? 64*1024 : obj->size);
559         inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024);
560
561         memset(&stream, 0, sizeof(stream));
562         git_inflate_init(&stream);
563         stream.next_out = data;
564         stream.avail_out = consume ? 64*1024 : obj->size;
565
566         do {
567                 ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024;
568                 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
569                 if (n < 0)
570                         die_errno(_("cannot pread pack file"));
571                 if (!n)
572                         die(Q_("premature end of pack file, %"PRIuMAX" byte missing",
573                                "premature end of pack file, %"PRIuMAX" bytes missing",
574                                (unsigned int)len),
575                             (uintmax_t)len);
576                 from += n;
577                 len -= n;
578                 stream.next_in = inbuf;
579                 stream.avail_in = n;
580                 if (!consume)
581                         status = git_inflate(&stream, 0);
582                 else {
583                         do {
584                                 status = git_inflate(&stream, 0);
585                                 if (consume(data, stream.next_out - data, cb_data)) {
586                                         free(inbuf);
587                                         free(data);
588                                         return NULL;
589                                 }
590                                 stream.next_out = data;
591                                 stream.avail_out = 64*1024;
592                         } while (status == Z_OK && stream.avail_in);
593                 }
594         } while (len && status == Z_OK && !stream.avail_in);
595
596         /* This has been inflated OK when first encountered, so... */
597         if (status != Z_STREAM_END || stream.total_out != obj->size)
598                 die(_("serious inflate inconsistency"));
599
600         git_inflate_end(&stream);
601         free(inbuf);
602         if (consume) {
603                 free(data);
604                 data = NULL;
605         }
606         return data;
607 }
608
609 static void *get_data_from_pack(struct object_entry *obj)
610 {
611         return unpack_data(obj, NULL, NULL);
612 }
613
614 static int compare_ofs_delta_bases(off_t offset1, off_t offset2,
615                                    enum object_type type1,
616                                    enum object_type type2)
617 {
618         int cmp = type1 - type2;
619         if (cmp)
620                 return cmp;
621         return offset1 < offset2 ? -1 :
622                offset1 > offset2 ?  1 :
623                0;
624 }
625
626 static int find_ofs_delta(const off_t offset, enum object_type type)
627 {
628         int first = 0, last = nr_ofs_deltas;
629
630         while (first < last) {
631                 int next = (first + last) / 2;
632                 struct ofs_delta_entry *delta = &ofs_deltas[next];
633                 int cmp;
634
635                 cmp = compare_ofs_delta_bases(offset, delta->offset,
636                                               type, objects[delta->obj_no].type);
637                 if (!cmp)
638                         return next;
639                 if (cmp < 0) {
640                         last = next;
641                         continue;
642                 }
643                 first = next+1;
644         }
645         return -first-1;
646 }
647
648 static void find_ofs_delta_children(off_t offset,
649                                     int *first_index, int *last_index,
650                                     enum object_type type)
651 {
652         int first = find_ofs_delta(offset, type);
653         int last = first;
654         int end = nr_ofs_deltas - 1;
655
656         if (first < 0) {
657                 *first_index = 0;
658                 *last_index = -1;
659                 return;
660         }
661         while (first > 0 && ofs_deltas[first - 1].offset == offset)
662                 --first;
663         while (last < end && ofs_deltas[last + 1].offset == offset)
664                 ++last;
665         *first_index = first;
666         *last_index = last;
667 }
668
669 static int compare_ref_delta_bases(const unsigned char *sha1,
670                                    const unsigned char *sha2,
671                                    enum object_type type1,
672                                    enum object_type type2)
673 {
674         int cmp = type1 - type2;
675         if (cmp)
676                 return cmp;
677         return hashcmp(sha1, sha2);
678 }
679
680 static int find_ref_delta(const unsigned char *sha1, enum object_type type)
681 {
682         int first = 0, last = nr_ref_deltas;
683
684         while (first < last) {
685                 int next = (first + last) / 2;
686                 struct ref_delta_entry *delta = &ref_deltas[next];
687                 int cmp;
688
689                 cmp = compare_ref_delta_bases(sha1, delta->sha1,
690                                               type, objects[delta->obj_no].type);
691                 if (!cmp)
692                         return next;
693                 if (cmp < 0) {
694                         last = next;
695                         continue;
696                 }
697                 first = next+1;
698         }
699         return -first-1;
700 }
701
702 static void find_ref_delta_children(const unsigned char *sha1,
703                                     int *first_index, int *last_index,
704                                     enum object_type type)
705 {
706         int first = find_ref_delta(sha1, type);
707         int last = first;
708         int end = nr_ref_deltas - 1;
709
710         if (first < 0) {
711                 *first_index = 0;
712                 *last_index = -1;
713                 return;
714         }
715         while (first > 0 && !hashcmp(ref_deltas[first - 1].sha1, sha1))
716                 --first;
717         while (last < end && !hashcmp(ref_deltas[last + 1].sha1, sha1))
718                 ++last;
719         *first_index = first;
720         *last_index = last;
721 }
722
723 struct compare_data {
724         struct object_entry *entry;
725         struct git_istream *st;
726         unsigned char *buf;
727         unsigned long buf_size;
728 };
729
730 static int compare_objects(const unsigned char *buf, unsigned long size,
731                            void *cb_data)
732 {
733         struct compare_data *data = cb_data;
734
735         if (data->buf_size < size) {
736                 free(data->buf);
737                 data->buf = xmalloc(size);
738                 data->buf_size = size;
739         }
740
741         while (size) {
742                 ssize_t len = read_istream(data->st, data->buf, size);
743                 if (len == 0)
744                         die(_("SHA1 COLLISION FOUND WITH %s !"),
745                             sha1_to_hex(data->entry->idx.sha1));
746                 if (len < 0)
747                         die(_("unable to read %s"),
748                             sha1_to_hex(data->entry->idx.sha1));
749                 if (memcmp(buf, data->buf, len))
750                         die(_("SHA1 COLLISION FOUND WITH %s !"),
751                             sha1_to_hex(data->entry->idx.sha1));
752                 size -= len;
753                 buf += len;
754         }
755         return 0;
756 }
757
758 static int check_collison(struct object_entry *entry)
759 {
760         struct compare_data data;
761         enum object_type type;
762         unsigned long size;
763
764         if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
765                 return -1;
766
767         memset(&data, 0, sizeof(data));
768         data.entry = entry;
769         data.st = open_istream(entry->idx.sha1, &type, &size, NULL);
770         if (!data.st)
771                 return -1;
772         if (size != entry->size || type != entry->type)
773                 die(_("SHA1 COLLISION FOUND WITH %s !"),
774                     sha1_to_hex(entry->idx.sha1));
775         unpack_data(entry, compare_objects, &data);
776         close_istream(data.st);
777         free(data.buf);
778         return 0;
779 }
780
781 static void sha1_object(const void *data, struct object_entry *obj_entry,
782                         unsigned long size, enum object_type type,
783                         const unsigned char *sha1)
784 {
785         void *new_data = NULL;
786         int collision_test_needed;
787
788         assert(data || obj_entry);
789
790         read_lock();
791         collision_test_needed = has_sha1_file_with_flags(sha1, HAS_SHA1_QUICK);
792         read_unlock();
793
794         if (collision_test_needed && !data) {
795                 read_lock();
796                 if (!check_collison(obj_entry))
797                         collision_test_needed = 0;
798                 read_unlock();
799         }
800         if (collision_test_needed) {
801                 void *has_data;
802                 enum object_type has_type;
803                 unsigned long has_size;
804                 read_lock();
805                 has_type = sha1_object_info(sha1, &has_size);
806                 if (has_type != type || has_size != size)
807                         die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
808                 has_data = read_sha1_file(sha1, &has_type, &has_size);
809                 read_unlock();
810                 if (!data)
811                         data = new_data = get_data_from_pack(obj_entry);
812                 if (!has_data)
813                         die(_("cannot read existing object %s"), sha1_to_hex(sha1));
814                 if (size != has_size || type != has_type ||
815                     memcmp(data, has_data, size) != 0)
816                         die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
817                 free(has_data);
818         }
819
820         if (strict) {
821                 read_lock();
822                 if (type == OBJ_BLOB) {
823                         struct blob *blob = lookup_blob(sha1);
824                         if (blob)
825                                 blob->object.flags |= FLAG_CHECKED;
826                         else
827                                 die(_("invalid blob object %s"), sha1_to_hex(sha1));
828                 } else {
829                         struct object *obj;
830                         int eaten;
831                         void *buf = (void *) data;
832
833                         assert(data && "data can only be NULL for large _blobs_");
834
835                         /*
836                          * we do not need to free the memory here, as the
837                          * buf is deleted by the caller.
838                          */
839                         obj = parse_object_buffer(sha1, type, size, buf, &eaten);
840                         if (!obj)
841                                 die(_("invalid %s"), typename(type));
842                         if (do_fsck_object &&
843                             fsck_object(obj, buf, size, &fsck_options))
844                                 die(_("Error in object"));
845                         if (fsck_walk(obj, NULL, &fsck_options))
846                                 die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid));
847
848                         if (obj->type == OBJ_TREE) {
849                                 struct tree *item = (struct tree *) obj;
850                                 item->buffer = NULL;
851                                 obj->parsed = 0;
852                         }
853                         if (obj->type == OBJ_COMMIT) {
854                                 struct commit *commit = (struct commit *) obj;
855                                 if (detach_commit_buffer(commit, NULL) != data)
856                                         die("BUG: parse_object_buffer transmogrified our buffer");
857                         }
858                         obj->flags |= FLAG_CHECKED;
859                 }
860                 read_unlock();
861         }
862
863         free(new_data);
864 }
865
866 /*
867  * This function is part of find_unresolved_deltas(). There are two
868  * walkers going in the opposite ways.
869  *
870  * The first one in find_unresolved_deltas() traverses down from
871  * parent node to children, deflating nodes along the way. However,
872  * memory for deflated nodes is limited by delta_base_cache_limit, so
873  * at some point parent node's deflated content may be freed.
874  *
875  * The second walker is this function, which goes from current node up
876  * to top parent if necessary to deflate the node. In normal
877  * situation, its parent node would be already deflated, so it just
878  * needs to apply delta.
879  *
880  * In the worst case scenario, parent node is no longer deflated because
881  * we're running out of delta_base_cache_limit; we need to re-deflate
882  * parents, possibly up to the top base.
883  *
884  * All deflated objects here are subject to be freed if we exceed
885  * delta_base_cache_limit, just like in find_unresolved_deltas(), we
886  * just need to make sure the last node is not freed.
887  */
888 static void *get_base_data(struct base_data *c)
889 {
890         if (!c->data) {
891                 struct object_entry *obj = c->obj;
892                 struct base_data **delta = NULL;
893                 int delta_nr = 0, delta_alloc = 0;
894
895                 while (is_delta_type(c->obj->type) && !c->data) {
896                         ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
897                         delta[delta_nr++] = c;
898                         c = c->base;
899                 }
900                 if (!delta_nr) {
901                         c->data = get_data_from_pack(obj);
902                         c->size = obj->size;
903                         get_thread_data()->base_cache_used += c->size;
904                         prune_base_data(c);
905                 }
906                 for (; delta_nr > 0; delta_nr--) {
907                         void *base, *raw;
908                         c = delta[delta_nr - 1];
909                         obj = c->obj;
910                         base = get_base_data(c->base);
911                         raw = get_data_from_pack(obj);
912                         c->data = patch_delta(
913                                 base, c->base->size,
914                                 raw, obj->size,
915                                 &c->size);
916                         free(raw);
917                         if (!c->data)
918                                 bad_object(obj->idx.offset, _("failed to apply delta"));
919                         get_thread_data()->base_cache_used += c->size;
920                         prune_base_data(c);
921                 }
922                 free(delta);
923         }
924         return c->data;
925 }
926
927 static void resolve_delta(struct object_entry *delta_obj,
928                           struct base_data *base, struct base_data *result)
929 {
930         void *base_data, *delta_data;
931
932         if (show_stat) {
933                 int i = delta_obj - objects;
934                 int j = base->obj - objects;
935                 obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1;
936                 deepest_delta_lock();
937                 if (deepest_delta < obj_stat[i].delta_depth)
938                         deepest_delta = obj_stat[i].delta_depth;
939                 deepest_delta_unlock();
940                 obj_stat[i].base_object_no = j;
941         }
942         delta_data = get_data_from_pack(delta_obj);
943         base_data = get_base_data(base);
944         result->obj = delta_obj;
945         result->data = patch_delta(base_data, base->size,
946                                    delta_data, delta_obj->size, &result->size);
947         free(delta_data);
948         if (!result->data)
949                 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
950         hash_sha1_file(result->data, result->size,
951                        typename(delta_obj->real_type), delta_obj->idx.sha1);
952         sha1_object(result->data, NULL, result->size, delta_obj->real_type,
953                     delta_obj->idx.sha1);
954         counter_lock();
955         nr_resolved_deltas++;
956         counter_unlock();
957 }
958
959 /*
960  * Standard boolean compare-and-swap: atomically check whether "*type" is
961  * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
962  * and return false.
963  */
964 static int compare_and_swap_type(signed char *type,
965                                  enum object_type want,
966                                  enum object_type set)
967 {
968         enum object_type old;
969
970         type_cas_lock();
971         old = *type;
972         if (old == want)
973                 *type = set;
974         type_cas_unlock();
975
976         return old == want;
977 }
978
979 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
980                                                   struct base_data *prev_base)
981 {
982         if (base->ref_last == -1 && base->ofs_last == -1) {
983                 find_ref_delta_children(base->obj->idx.sha1,
984                                         &base->ref_first, &base->ref_last,
985                                         OBJ_REF_DELTA);
986
987                 find_ofs_delta_children(base->obj->idx.offset,
988                                         &base->ofs_first, &base->ofs_last,
989                                         OBJ_OFS_DELTA);
990
991                 if (base->ref_last == -1 && base->ofs_last == -1) {
992                         free(base->data);
993                         return NULL;
994                 }
995
996                 link_base_data(prev_base, base);
997         }
998
999         if (base->ref_first <= base->ref_last) {
1000                 struct object_entry *child = objects + ref_deltas[base->ref_first].obj_no;
1001                 struct base_data *result = alloc_base_data();
1002
1003                 if (!compare_and_swap_type(&child->real_type, OBJ_REF_DELTA,
1004                                            base->obj->real_type))
1005                         die("BUG: child->real_type != OBJ_REF_DELTA");
1006
1007                 resolve_delta(child, base, result);
1008                 if (base->ref_first == base->ref_last && base->ofs_last == -1)
1009                         free_base_data(base);
1010
1011                 base->ref_first++;
1012                 return result;
1013         }
1014
1015         if (base->ofs_first <= base->ofs_last) {
1016                 struct object_entry *child = objects + ofs_deltas[base->ofs_first].obj_no;
1017                 struct base_data *result = alloc_base_data();
1018
1019                 assert(child->real_type == OBJ_OFS_DELTA);
1020                 child->real_type = base->obj->real_type;
1021                 resolve_delta(child, base, result);
1022                 if (base->ofs_first == base->ofs_last)
1023                         free_base_data(base);
1024
1025                 base->ofs_first++;
1026                 return result;
1027         }
1028
1029         unlink_base_data(base);
1030         return NULL;
1031 }
1032
1033 static void find_unresolved_deltas(struct base_data *base)
1034 {
1035         struct base_data *new_base, *prev_base = NULL;
1036         for (;;) {
1037                 new_base = find_unresolved_deltas_1(base, prev_base);
1038
1039                 if (new_base) {
1040                         prev_base = base;
1041                         base = new_base;
1042                 } else {
1043                         free(base);
1044                         base = prev_base;
1045                         if (!base)
1046                                 return;
1047                         prev_base = base->base;
1048                 }
1049         }
1050 }
1051
1052 static int compare_ofs_delta_entry(const void *a, const void *b)
1053 {
1054         const struct ofs_delta_entry *delta_a = a;
1055         const struct ofs_delta_entry *delta_b = b;
1056
1057         return delta_a->offset < delta_b->offset ? -1 :
1058                delta_a->offset > delta_b->offset ?  1 :
1059                0;
1060 }
1061
1062 static int compare_ref_delta_entry(const void *a, const void *b)
1063 {
1064         const struct ref_delta_entry *delta_a = a;
1065         const struct ref_delta_entry *delta_b = b;
1066
1067         return hashcmp(delta_a->sha1, delta_b->sha1);
1068 }
1069
1070 static void resolve_base(struct object_entry *obj)
1071 {
1072         struct base_data *base_obj = alloc_base_data();
1073         base_obj->obj = obj;
1074         base_obj->data = NULL;
1075         find_unresolved_deltas(base_obj);
1076 }
1077
1078 #ifndef NO_PTHREADS
1079 static void *threaded_second_pass(void *data)
1080 {
1081         set_thread_data(data);
1082         for (;;) {
1083                 int i;
1084                 counter_lock();
1085                 display_progress(progress, nr_resolved_deltas);
1086                 counter_unlock();
1087                 work_lock();
1088                 while (nr_dispatched < nr_objects &&
1089                        is_delta_type(objects[nr_dispatched].type))
1090                         nr_dispatched++;
1091                 if (nr_dispatched >= nr_objects) {
1092                         work_unlock();
1093                         break;
1094                 }
1095                 i = nr_dispatched++;
1096                 work_unlock();
1097
1098                 resolve_base(&objects[i]);
1099         }
1100         return NULL;
1101 }
1102 #endif
1103
1104 /*
1105  * First pass:
1106  * - find locations of all objects;
1107  * - calculate SHA1 of all non-delta objects;
1108  * - remember base (SHA1 or offset) for all deltas.
1109  */
1110 static void parse_pack_objects(unsigned char *sha1)
1111 {
1112         int i, nr_delays = 0;
1113         struct ofs_delta_entry *ofs_delta = ofs_deltas;
1114         unsigned char ref_delta_sha1[20];
1115         struct stat st;
1116
1117         if (verbose)
1118                 progress = start_progress(
1119                                 from_stdin ? _("Receiving objects") : _("Indexing objects"),
1120                                 nr_objects);
1121         for (i = 0; i < nr_objects; i++) {
1122                 struct object_entry *obj = &objects[i];
1123                 void *data = unpack_raw_entry(obj, &ofs_delta->offset,
1124                                               ref_delta_sha1, obj->idx.sha1);
1125                 obj->real_type = obj->type;
1126                 if (obj->type == OBJ_OFS_DELTA) {
1127                         nr_ofs_deltas++;
1128                         ofs_delta->obj_no = i;
1129                         ofs_delta++;
1130                 } else if (obj->type == OBJ_REF_DELTA) {
1131                         ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc);
1132                         hashcpy(ref_deltas[nr_ref_deltas].sha1, ref_delta_sha1);
1133                         ref_deltas[nr_ref_deltas].obj_no = i;
1134                         nr_ref_deltas++;
1135                 } else if (!data) {
1136                         /* large blobs, check later */
1137                         obj->real_type = OBJ_BAD;
1138                         nr_delays++;
1139                 } else
1140                         sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
1141                 free(data);
1142                 display_progress(progress, i+1);
1143         }
1144         objects[i].idx.offset = consumed_bytes;
1145         stop_progress(&progress);
1146
1147         /* Check pack integrity */
1148         flush();
1149         git_SHA1_Final(sha1, &input_ctx);
1150         if (hashcmp(fill(20), sha1))
1151                 die(_("pack is corrupted (SHA1 mismatch)"));
1152         use(20);
1153
1154         /* If input_fd is a file, we should have reached its end now. */
1155         if (fstat(input_fd, &st))
1156                 die_errno(_("cannot fstat packfile"));
1157         if (S_ISREG(st.st_mode) &&
1158                         lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1159                 die(_("pack has junk at the end"));
1160
1161         for (i = 0; i < nr_objects; i++) {
1162                 struct object_entry *obj = &objects[i];
1163                 if (obj->real_type != OBJ_BAD)
1164                         continue;
1165                 obj->real_type = obj->type;
1166                 sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
1167                 nr_delays--;
1168         }
1169         if (nr_delays)
1170                 die(_("confusion beyond insanity in parse_pack_objects()"));
1171 }
1172
1173 /*
1174  * Second pass:
1175  * - for all non-delta objects, look if it is used as a base for
1176  *   deltas;
1177  * - if used as a base, uncompress the object and apply all deltas,
1178  *   recursively checking if the resulting object is used as a base
1179  *   for some more deltas.
1180  */
1181 static void resolve_deltas(void)
1182 {
1183         int i;
1184
1185         if (!nr_ofs_deltas && !nr_ref_deltas)
1186                 return;
1187
1188         /* Sort deltas by base SHA1/offset for fast searching */
1189         qsort(ofs_deltas, nr_ofs_deltas, sizeof(struct ofs_delta_entry),
1190               compare_ofs_delta_entry);
1191         qsort(ref_deltas, nr_ref_deltas, sizeof(struct ref_delta_entry),
1192               compare_ref_delta_entry);
1193
1194         if (verbose)
1195                 progress = start_progress(_("Resolving deltas"),
1196                                           nr_ref_deltas + nr_ofs_deltas);
1197
1198 #ifndef NO_PTHREADS
1199         nr_dispatched = 0;
1200         if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1201                 init_thread();
1202                 for (i = 0; i < nr_threads; i++) {
1203                         int ret = pthread_create(&thread_data[i].thread, NULL,
1204                                                  threaded_second_pass, thread_data + i);
1205                         if (ret)
1206                                 die(_("unable to create thread: %s"),
1207                                     strerror(ret));
1208                 }
1209                 for (i = 0; i < nr_threads; i++)
1210                         pthread_join(thread_data[i].thread, NULL);
1211                 cleanup_thread();
1212                 return;
1213         }
1214 #endif
1215
1216         for (i = 0; i < nr_objects; i++) {
1217                 struct object_entry *obj = &objects[i];
1218
1219                 if (is_delta_type(obj->type))
1220                         continue;
1221                 resolve_base(obj);
1222                 display_progress(progress, nr_resolved_deltas);
1223         }
1224 }
1225
1226 /*
1227  * Third pass:
1228  * - append objects to convert thin pack to full pack if required
1229  * - write the final 20-byte SHA-1
1230  */
1231 static void fix_unresolved_deltas(struct sha1file *f);
1232 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1233 {
1234         if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) {
1235                 stop_progress(&progress);
1236                 /* Flush remaining pack final 20-byte SHA1. */
1237                 flush();
1238                 return;
1239         }
1240
1241         if (fix_thin_pack) {
1242                 struct sha1file *f;
1243                 unsigned char read_sha1[20], tail_sha1[20];
1244                 struct strbuf msg = STRBUF_INIT;
1245                 int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas;
1246                 int nr_objects_initial = nr_objects;
1247                 if (nr_unresolved <= 0)
1248                         die(_("confusion beyond insanity"));
1249                 REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1);
1250                 memset(objects + nr_objects + 1, 0,
1251                        nr_unresolved * sizeof(*objects));
1252                 f = sha1fd(output_fd, curr_pack);
1253                 fix_unresolved_deltas(f);
1254                 strbuf_addf(&msg, Q_("completed with %d local object",
1255                                      "completed with %d local objects",
1256                                      nr_objects - nr_objects_initial),
1257                             nr_objects - nr_objects_initial);
1258                 stop_progress_msg(&progress, msg.buf);
1259                 strbuf_release(&msg);
1260                 sha1close(f, tail_sha1, 0);
1261                 hashcpy(read_sha1, pack_sha1);
1262                 fixup_pack_header_footer(output_fd, pack_sha1,
1263                                          curr_pack, nr_objects,
1264                                          read_sha1, consumed_bytes-20);
1265                 if (hashcmp(read_sha1, tail_sha1) != 0)
1266                         die(_("Unexpected tail checksum for %s "
1267                               "(disk corruption?)"), curr_pack);
1268         }
1269         if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas)
1270                 die(Q_("pack has %d unresolved delta",
1271                        "pack has %d unresolved deltas",
1272                        nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas),
1273                     nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas);
1274 }
1275
1276 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1277 {
1278         git_zstream stream;
1279         int status;
1280         unsigned char outbuf[4096];
1281
1282         git_deflate_init(&stream, zlib_compression_level);
1283         stream.next_in = in;
1284         stream.avail_in = size;
1285
1286         do {
1287                 stream.next_out = outbuf;
1288                 stream.avail_out = sizeof(outbuf);
1289                 status = git_deflate(&stream, Z_FINISH);
1290                 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1291         } while (status == Z_OK);
1292
1293         if (status != Z_STREAM_END)
1294                 die(_("unable to deflate appended object (%d)"), status);
1295         size = stream.total_out;
1296         git_deflate_end(&stream);
1297         return size;
1298 }
1299
1300 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1301                                const unsigned char *sha1, void *buf,
1302                                unsigned long size, enum object_type type)
1303 {
1304         struct object_entry *obj = &objects[nr_objects++];
1305         unsigned char header[10];
1306         unsigned long s = size;
1307         int n = 0;
1308         unsigned char c = (type << 4) | (s & 15);
1309         s >>= 4;
1310         while (s) {
1311                 header[n++] = c | 0x80;
1312                 c = s & 0x7f;
1313                 s >>= 7;
1314         }
1315         header[n++] = c;
1316         crc32_begin(f);
1317         sha1write(f, header, n);
1318         obj[0].size = size;
1319         obj[0].hdr_size = n;
1320         obj[0].type = type;
1321         obj[0].real_type = type;
1322         obj[1].idx.offset = obj[0].idx.offset + n;
1323         obj[1].idx.offset += write_compressed(f, buf, size);
1324         obj[0].idx.crc32 = crc32_end(f);
1325         sha1flush(f);
1326         hashcpy(obj->idx.sha1, sha1);
1327         return obj;
1328 }
1329
1330 static int delta_pos_compare(const void *_a, const void *_b)
1331 {
1332         struct ref_delta_entry *a = *(struct ref_delta_entry **)_a;
1333         struct ref_delta_entry *b = *(struct ref_delta_entry **)_b;
1334         return a->obj_no - b->obj_no;
1335 }
1336
1337 static void fix_unresolved_deltas(struct sha1file *f)
1338 {
1339         struct ref_delta_entry **sorted_by_pos;
1340         int i;
1341
1342         /*
1343          * Since many unresolved deltas may well be themselves base objects
1344          * for more unresolved deltas, we really want to include the
1345          * smallest number of base objects that would cover as much delta
1346          * as possible by picking the
1347          * trunc deltas first, allowing for other deltas to resolve without
1348          * additional base objects.  Since most base objects are to be found
1349          * before deltas depending on them, a good heuristic is to start
1350          * resolving deltas in the same order as their position in the pack.
1351          */
1352         ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas);
1353         for (i = 0; i < nr_ref_deltas; i++)
1354                 sorted_by_pos[i] = &ref_deltas[i];
1355         qsort(sorted_by_pos, nr_ref_deltas, sizeof(*sorted_by_pos), delta_pos_compare);
1356
1357         for (i = 0; i < nr_ref_deltas; i++) {
1358                 struct ref_delta_entry *d = sorted_by_pos[i];
1359                 enum object_type type;
1360                 struct base_data *base_obj = alloc_base_data();
1361
1362                 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1363                         continue;
1364                 base_obj->data = read_sha1_file(d->sha1, &type, &base_obj->size);
1365                 if (!base_obj->data)
1366                         continue;
1367
1368                 if (check_sha1_signature(d->sha1, base_obj->data,
1369                                 base_obj->size, typename(type)))
1370                         die(_("local object %s is corrupt"), sha1_to_hex(d->sha1));
1371                 base_obj->obj = append_obj_to_pack(f, d->sha1,
1372                                         base_obj->data, base_obj->size, type);
1373                 find_unresolved_deltas(base_obj);
1374                 display_progress(progress, nr_resolved_deltas);
1375         }
1376         free(sorted_by_pos);
1377 }
1378
1379 static void final(const char *final_pack_name, const char *curr_pack_name,
1380                   const char *final_index_name, const char *curr_index_name,
1381                   const char *keep_name, const char *keep_msg,
1382                   unsigned char *sha1)
1383 {
1384         const char *report = "pack";
1385         char name[PATH_MAX];
1386         int err;
1387
1388         if (!from_stdin) {
1389                 close(input_fd);
1390         } else {
1391                 fsync_or_die(output_fd, curr_pack_name);
1392                 err = close(output_fd);
1393                 if (err)
1394                         die_errno(_("error while closing pack file"));
1395         }
1396
1397         if (keep_msg) {
1398                 int keep_fd, keep_msg_len = strlen(keep_msg);
1399
1400                 if (!keep_name)
1401                         keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1402                 else
1403                         keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1404
1405                 if (keep_fd < 0) {
1406                         if (errno != EEXIST)
1407                                 die_errno(_("cannot write keep file '%s'"),
1408                                           keep_name ? keep_name : name);
1409                 } else {
1410                         if (keep_msg_len > 0) {
1411                                 write_or_die(keep_fd, keep_msg, keep_msg_len);
1412                                 write_or_die(keep_fd, "\n", 1);
1413                         }
1414                         if (close(keep_fd) != 0)
1415                                 die_errno(_("cannot close written keep file '%s'"),
1416                                           keep_name ? keep_name : name);
1417                         report = "keep";
1418                 }
1419         }
1420
1421         if (final_pack_name != curr_pack_name) {
1422                 if (!final_pack_name) {
1423                         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1424                                  get_object_directory(), sha1_to_hex(sha1));
1425                         final_pack_name = name;
1426                 }
1427                 if (finalize_object_file(curr_pack_name, final_pack_name))
1428                         die(_("cannot store pack file"));
1429         } else if (from_stdin)
1430                 chmod(final_pack_name, 0444);
1431
1432         if (final_index_name != curr_index_name) {
1433                 if (!final_index_name) {
1434                         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1435                                  get_object_directory(), sha1_to_hex(sha1));
1436                         final_index_name = name;
1437                 }
1438                 if (finalize_object_file(curr_index_name, final_index_name))
1439                         die(_("cannot store index file"));
1440         } else
1441                 chmod(final_index_name, 0444);
1442
1443         if (!from_stdin) {
1444                 printf("%s\n", sha1_to_hex(sha1));
1445         } else {
1446                 char buf[48];
1447                 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1448                                    report, sha1_to_hex(sha1));
1449                 write_or_die(1, buf, len);
1450
1451                 /*
1452                  * Let's just mimic git-unpack-objects here and write
1453                  * the last part of the input buffer to stdout.
1454                  */
1455                 while (input_len) {
1456                         err = xwrite(1, input_buffer + input_offset, input_len);
1457                         if (err <= 0)
1458                                 break;
1459                         input_len -= err;
1460                         input_offset += err;
1461                 }
1462         }
1463 }
1464
1465 static int git_index_pack_config(const char *k, const char *v, void *cb)
1466 {
1467         struct pack_idx_option *opts = cb;
1468
1469         if (!strcmp(k, "pack.indexversion")) {
1470                 opts->version = git_config_int(k, v);
1471                 if (opts->version > 2)
1472                         die(_("bad pack.indexversion=%"PRIu32), opts->version);
1473                 return 0;
1474         }
1475         if (!strcmp(k, "pack.threads")) {
1476                 nr_threads = git_config_int(k, v);
1477                 if (nr_threads < 0)
1478                         die(_("invalid number of threads specified (%d)"),
1479                             nr_threads);
1480 #ifdef NO_PTHREADS
1481                 if (nr_threads != 1)
1482                         warning(_("no threads support, ignoring %s"), k);
1483                 nr_threads = 1;
1484 #endif
1485                 return 0;
1486         }
1487         return git_default_config(k, v, cb);
1488 }
1489
1490 static int cmp_uint32(const void *a_, const void *b_)
1491 {
1492         uint32_t a = *((uint32_t *)a_);
1493         uint32_t b = *((uint32_t *)b_);
1494
1495         return (a < b) ? -1 : (a != b);
1496 }
1497
1498 static void read_v2_anomalous_offsets(struct packed_git *p,
1499                                       struct pack_idx_option *opts)
1500 {
1501         const uint32_t *idx1, *idx2;
1502         uint32_t i;
1503
1504         /* The address of the 4-byte offset table */
1505         idx1 = (((const uint32_t *)p->index_data)
1506                 + 2 /* 8-byte header */
1507                 + 256 /* fan out */
1508                 + 5 * p->num_objects /* 20-byte SHA-1 table */
1509                 + p->num_objects /* CRC32 table */
1510                 );
1511
1512         /* The address of the 8-byte offset table */
1513         idx2 = idx1 + p->num_objects;
1514
1515         for (i = 0; i < p->num_objects; i++) {
1516                 uint32_t off = ntohl(idx1[i]);
1517                 if (!(off & 0x80000000))
1518                         continue;
1519                 off = off & 0x7fffffff;
1520                 check_pack_index_ptr(p, &idx2[off * 2]);
1521                 if (idx2[off * 2])
1522                         continue;
1523                 /*
1524                  * The real offset is ntohl(idx2[off * 2]) in high 4
1525                  * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1526                  * octets.  But idx2[off * 2] is Zero!!!
1527                  */
1528                 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1529                 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1530         }
1531
1532         if (1 < opts->anomaly_nr)
1533                 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1534 }
1535
1536 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1537 {
1538         struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1539
1540         if (!p)
1541                 die(_("Cannot open existing pack file '%s'"), pack_name);
1542         if (open_pack_index(p))
1543                 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1544
1545         /* Read the attributes from the existing idx file */
1546         opts->version = p->index_version;
1547
1548         if (opts->version == 2)
1549                 read_v2_anomalous_offsets(p, opts);
1550
1551         /*
1552          * Get rid of the idx file as we do not need it anymore.
1553          * NEEDSWORK: extract this bit from free_pack_by_name() in
1554          * sha1_file.c, perhaps?  It shouldn't matter very much as we
1555          * know we haven't installed this pack (hence we never have
1556          * read anything from it).
1557          */
1558         close_pack_index(p);
1559         free(p);
1560 }
1561
1562 static void show_pack_info(int stat_only)
1563 {
1564         int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas;
1565         unsigned long *chain_histogram = NULL;
1566
1567         if (deepest_delta)
1568                 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1569
1570         for (i = 0; i < nr_objects; i++) {
1571                 struct object_entry *obj = &objects[i];
1572
1573                 if (is_delta_type(obj->type))
1574                         chain_histogram[obj_stat[i].delta_depth - 1]++;
1575                 if (stat_only)
1576                         continue;
1577                 printf("%s %-6s %lu %lu %"PRIuMAX,
1578                        sha1_to_hex(obj->idx.sha1),
1579                        typename(obj->real_type), obj->size,
1580                        (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1581                        (uintmax_t)obj->idx.offset);
1582                 if (is_delta_type(obj->type)) {
1583                         struct object_entry *bobj = &objects[obj_stat[i].base_object_no];
1584                         printf(" %u %s", obj_stat[i].delta_depth, sha1_to_hex(bobj->idx.sha1));
1585                 }
1586                 putchar('\n');
1587         }
1588
1589         if (baseobjects)
1590                 printf_ln(Q_("non delta: %d object",
1591                              "non delta: %d objects",
1592                              baseobjects),
1593                           baseobjects);
1594         for (i = 0; i < deepest_delta; i++) {
1595                 if (!chain_histogram[i])
1596                         continue;
1597                 printf_ln(Q_("chain length = %d: %lu object",
1598                              "chain length = %d: %lu objects",
1599                              chain_histogram[i]),
1600                           i + 1,
1601                           chain_histogram[i]);
1602         }
1603 }
1604
1605 static const char *derive_filename(const char *pack_name, const char *suffix,
1606                                    struct strbuf *buf)
1607 {
1608         size_t len;
1609         if (!strip_suffix(pack_name, ".pack", &len))
1610                 die(_("packfile name '%s' does not end with '.pack'"),
1611                     pack_name);
1612         strbuf_add(buf, pack_name, len);
1613         strbuf_addstr(buf, suffix);
1614         return buf->buf;
1615 }
1616
1617 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1618 {
1619         int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
1620         const char *curr_index;
1621         const char *index_name = NULL, *pack_name = NULL;
1622         const char *keep_name = NULL, *keep_msg = NULL;
1623         struct strbuf index_name_buf = STRBUF_INIT,
1624                       keep_name_buf = STRBUF_INIT;
1625         struct pack_idx_entry **idx_objects;
1626         struct pack_idx_option opts;
1627         unsigned char pack_sha1[20];
1628         unsigned foreign_nr = 1;        /* zero is a "good" value, assume bad */
1629
1630         if (argc == 2 && !strcmp(argv[1], "-h"))
1631                 usage(index_pack_usage);
1632
1633         check_replace_refs = 0;
1634         fsck_options.walk = mark_link;
1635
1636         reset_pack_idx_option(&opts);
1637         git_config(git_index_pack_config, &opts);
1638         if (prefix && chdir(prefix))
1639                 die(_("Cannot come back to cwd"));
1640
1641         for (i = 1; i < argc; i++) {
1642                 const char *arg = argv[i];
1643
1644                 if (*arg == '-') {
1645                         if (!strcmp(arg, "--stdin")) {
1646                                 from_stdin = 1;
1647                         } else if (!strcmp(arg, "--fix-thin")) {
1648                                 fix_thin_pack = 1;
1649                         } else if (!strcmp(arg, "--strict")) {
1650                                 strict = 1;
1651                                 do_fsck_object = 1;
1652                         } else if (skip_prefix(arg, "--strict=", &arg)) {
1653                                 strict = 1;
1654                                 do_fsck_object = 1;
1655                                 fsck_set_msg_types(&fsck_options, arg);
1656                         } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1657                                 strict = 1;
1658                                 check_self_contained_and_connected = 1;
1659                         } else if (!strcmp(arg, "--verify")) {
1660                                 verify = 1;
1661                         } else if (!strcmp(arg, "--verify-stat")) {
1662                                 verify = 1;
1663                                 show_stat = 1;
1664                         } else if (!strcmp(arg, "--verify-stat-only")) {
1665                                 verify = 1;
1666                                 show_stat = 1;
1667                                 stat_only = 1;
1668                         } else if (!strcmp(arg, "--keep")) {
1669                                 keep_msg = "";
1670                         } else if (starts_with(arg, "--keep=")) {
1671                                 keep_msg = arg + 7;
1672                         } else if (starts_with(arg, "--threads=")) {
1673                                 char *end;
1674                                 nr_threads = strtoul(arg+10, &end, 0);
1675                                 if (!arg[10] || *end || nr_threads < 0)
1676                                         usage(index_pack_usage);
1677 #ifdef NO_PTHREADS
1678                                 if (nr_threads != 1)
1679                                         warning(_("no threads support, "
1680                                                   "ignoring %s"), arg);
1681                                 nr_threads = 1;
1682 #endif
1683                         } else if (starts_with(arg, "--pack_header=")) {
1684                                 struct pack_header *hdr;
1685                                 char *c;
1686
1687                                 hdr = (struct pack_header *)input_buffer;
1688                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1689                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1690                                 if (*c != ',')
1691                                         die(_("bad %s"), arg);
1692                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1693                                 if (*c)
1694                                         die(_("bad %s"), arg);
1695                                 input_len = sizeof(*hdr);
1696                         } else if (!strcmp(arg, "-v")) {
1697                                 verbose = 1;
1698                         } else if (!strcmp(arg, "-o")) {
1699                                 if (index_name || (i+1) >= argc)
1700                                         usage(index_pack_usage);
1701                                 index_name = argv[++i];
1702                         } else if (starts_with(arg, "--index-version=")) {
1703                                 char *c;
1704                                 opts.version = strtoul(arg + 16, &c, 10);
1705                                 if (opts.version > 2)
1706                                         die(_("bad %s"), arg);
1707                                 if (*c == ',')
1708                                         opts.off32_limit = strtoul(c+1, &c, 0);
1709                                 if (*c || opts.off32_limit & 0x80000000)
1710                                         die(_("bad %s"), arg);
1711                         } else
1712                                 usage(index_pack_usage);
1713                         continue;
1714                 }
1715
1716                 if (pack_name)
1717                         usage(index_pack_usage);
1718                 pack_name = arg;
1719         }
1720
1721         if (!pack_name && !from_stdin)
1722                 usage(index_pack_usage);
1723         if (fix_thin_pack && !from_stdin)
1724                 die(_("--fix-thin cannot be used without --stdin"));
1725         if (!index_name && pack_name)
1726                 index_name = derive_filename(pack_name, ".idx", &index_name_buf);
1727         if (keep_msg && !keep_name && pack_name)
1728                 keep_name = derive_filename(pack_name, ".keep", &keep_name_buf);
1729
1730         if (verify) {
1731                 if (!index_name)
1732                         die(_("--verify with no packfile name given"));
1733                 read_idx_option(&opts, index_name);
1734                 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1735         }
1736         if (strict)
1737                 opts.flags |= WRITE_IDX_STRICT;
1738
1739 #ifndef NO_PTHREADS
1740         if (!nr_threads) {
1741                 nr_threads = online_cpus();
1742                 /* An experiment showed that more threads does not mean faster */
1743                 if (nr_threads > 3)
1744                         nr_threads = 3;
1745         }
1746 #endif
1747
1748         curr_pack = open_pack_file(pack_name);
1749         parse_pack_header();
1750         objects = xcalloc(st_add(nr_objects, 1), sizeof(struct object_entry));
1751         if (show_stat)
1752                 obj_stat = xcalloc(st_add(nr_objects, 1), sizeof(struct object_stat));
1753         ofs_deltas = xcalloc(nr_objects, sizeof(struct ofs_delta_entry));
1754         parse_pack_objects(pack_sha1);
1755         resolve_deltas();
1756         conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1757         free(ofs_deltas);
1758         free(ref_deltas);
1759         if (strict)
1760                 foreign_nr = check_objects();
1761
1762         if (show_stat)
1763                 show_pack_info(stat_only);
1764
1765         ALLOC_ARRAY(idx_objects, nr_objects);
1766         for (i = 0; i < nr_objects; i++)
1767                 idx_objects[i] = &objects[i].idx;
1768         curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1769         free(idx_objects);
1770
1771         if (!verify)
1772                 final(pack_name, curr_pack,
1773                       index_name, curr_index,
1774                       keep_name, keep_msg,
1775                       pack_sha1);
1776         else
1777                 close(input_fd);
1778         free(objects);
1779         strbuf_release(&index_name_buf);
1780         strbuf_release(&keep_name_buf);
1781         if (pack_name == NULL)
1782                 free((void *) curr_pack);
1783         if (index_name == NULL)
1784                 free((void *) curr_index);
1785
1786         /*
1787          * Let the caller know this pack is not self contained
1788          */
1789         if (check_self_contained_and_connected && foreign_nr)
1790                 return 1;
1791
1792         return 0;
1793 }