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