Merge branch 'js/ref-namespaces'
[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
13 static const char index_pack_usage[] =
14 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
15
16 struct object_entry {
17         struct pack_idx_entry idx;
18         unsigned long size;
19         unsigned int hdr_size;
20         enum object_type type;
21         enum object_type real_type;
22         unsigned delta_depth;
23         int base_object_no;
24 };
25
26 union delta_base {
27         unsigned char sha1[20];
28         off_t offset;
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 };
38
39 /*
40  * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
41  * to memcmp() only the first 20 bytes.
42  */
43 #define UNION_BASE_SZ   20
44
45 #define FLAG_LINK (1u<<20)
46 #define FLAG_CHECKED (1u<<21)
47
48 struct delta_entry {
49         union delta_base base;
50         int obj_no;
51 };
52
53 static struct object_entry *objects;
54 static struct delta_entry *deltas;
55 static struct base_data *base_cache;
56 static size_t base_cache_used;
57 static int nr_objects;
58 static int nr_deltas;
59 static int nr_resolved_deltas;
60
61 static int from_stdin;
62 static int strict;
63 static int verbose;
64
65 static struct progress *progress;
66
67 /* We always read in 4kB chunks. */
68 static unsigned char input_buffer[4096];
69 static unsigned int input_offset, input_len;
70 static off_t consumed_bytes;
71 static unsigned deepest_delta;
72 static git_SHA_CTX input_ctx;
73 static uint32_t input_crc32;
74 static int input_fd, output_fd, pack_fd;
75
76 static int mark_link(struct object *obj, int type, void *data)
77 {
78         if (!obj)
79                 return -1;
80
81         if (type != OBJ_ANY && obj->type != type)
82                 die("object type mismatch at %s", sha1_to_hex(obj->sha1));
83
84         obj->flags |= FLAG_LINK;
85         return 0;
86 }
87
88 /* The content of each linked object must have been checked
89    or it must be already present in the object database */
90 static void check_object(struct object *obj)
91 {
92         if (!obj)
93                 return;
94
95         if (!(obj->flags & FLAG_LINK))
96                 return;
97
98         if (!(obj->flags & FLAG_CHECKED)) {
99                 unsigned long size;
100                 int type = sha1_object_info(obj->sha1, &size);
101                 if (type != obj->type || type <= 0)
102                         die("object of unexpected type");
103                 obj->flags |= FLAG_CHECKED;
104                 return;
105         }
106 }
107
108 static void check_objects(void)
109 {
110         unsigned i, max;
111
112         max = get_max_object_index();
113         for (i = 0; i < max; i++)
114                 check_object(get_indexed_object(i));
115 }
116
117
118 /* Discard current buffer used content. */
119 static void flush(void)
120 {
121         if (input_offset) {
122                 if (output_fd >= 0)
123                         write_or_die(output_fd, input_buffer, input_offset);
124                 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
125                 memmove(input_buffer, input_buffer + input_offset, input_len);
126                 input_offset = 0;
127         }
128 }
129
130 /*
131  * Make sure at least "min" bytes are available in the buffer, and
132  * return the pointer to the buffer.
133  */
134 static void *fill(int min)
135 {
136         if (min <= input_len)
137                 return input_buffer + input_offset;
138         if (min > sizeof(input_buffer))
139                 die("cannot fill %d bytes", min);
140         flush();
141         do {
142                 ssize_t ret = xread(input_fd, input_buffer + input_len,
143                                 sizeof(input_buffer) - input_len);
144                 if (ret <= 0) {
145                         if (!ret)
146                                 die("early EOF");
147                         die_errno("read error on input");
148                 }
149                 input_len += ret;
150                 if (from_stdin)
151                         display_throughput(progress, consumed_bytes + input_len);
152         } while (input_len < min);
153         return input_buffer;
154 }
155
156 static void use(int bytes)
157 {
158         if (bytes > input_len)
159                 die("used more bytes than were available");
160         input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
161         input_len -= bytes;
162         input_offset += bytes;
163
164         /* make sure off_t is sufficiently large not to wrap */
165         if (signed_add_overflows(consumed_bytes, bytes))
166                 die("pack too large for current definition of off_t");
167         consumed_bytes += bytes;
168 }
169
170 static const char *open_pack_file(const char *pack_name)
171 {
172         if (from_stdin) {
173                 input_fd = 0;
174                 if (!pack_name) {
175                         static char tmpfile[PATH_MAX];
176                         output_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
177                                                 "pack/tmp_pack_XXXXXX");
178                         pack_name = xstrdup(tmpfile);
179                 } else
180                         output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
181                 if (output_fd < 0)
182                         die_errno("unable to create '%s'", pack_name);
183                 pack_fd = output_fd;
184         } else {
185                 input_fd = open(pack_name, O_RDONLY);
186                 if (input_fd < 0)
187                         die_errno("cannot open packfile '%s'", pack_name);
188                 output_fd = -1;
189                 pack_fd = input_fd;
190         }
191         git_SHA1_Init(&input_ctx);
192         return pack_name;
193 }
194
195 static void parse_pack_header(void)
196 {
197         struct pack_header *hdr = fill(sizeof(struct pack_header));
198
199         /* Header consistency check */
200         if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
201                 die("pack signature mismatch");
202         if (!pack_version_ok(hdr->hdr_version))
203                 die("pack version %"PRIu32" unsupported",
204                         ntohl(hdr->hdr_version));
205
206         nr_objects = ntohl(hdr->hdr_entries);
207         use(sizeof(struct pack_header));
208 }
209
210 static NORETURN void bad_object(unsigned long offset, const char *format,
211                        ...) __attribute__((format (printf, 2, 3)));
212
213 static NORETURN void bad_object(unsigned long offset, const char *format, ...)
214 {
215         va_list params;
216         char buf[1024];
217
218         va_start(params, format);
219         vsnprintf(buf, sizeof(buf), format, params);
220         va_end(params);
221         die("pack has bad object at offset %lu: %s", offset, buf);
222 }
223
224 static void free_base_data(struct base_data *c)
225 {
226         if (c->data) {
227                 free(c->data);
228                 c->data = NULL;
229                 base_cache_used -= c->size;
230         }
231 }
232
233 static void prune_base_data(struct base_data *retain)
234 {
235         struct base_data *b;
236         for (b = base_cache;
237              base_cache_used > delta_base_cache_limit && b;
238              b = b->child) {
239                 if (b->data && b != retain)
240                         free_base_data(b);
241         }
242 }
243
244 static void link_base_data(struct base_data *base, struct base_data *c)
245 {
246         if (base)
247                 base->child = c;
248         else
249                 base_cache = c;
250
251         c->base = base;
252         c->child = NULL;
253         if (c->data)
254                 base_cache_used += c->size;
255         prune_base_data(c);
256 }
257
258 static void unlink_base_data(struct base_data *c)
259 {
260         struct base_data *base = c->base;
261         if (base)
262                 base->child = NULL;
263         else
264                 base_cache = NULL;
265         free_base_data(c);
266 }
267
268 static void *unpack_entry_data(unsigned long offset, unsigned long size)
269 {
270         int status;
271         git_zstream stream;
272         void *buf = xmalloc(size);
273
274         memset(&stream, 0, sizeof(stream));
275         git_inflate_init(&stream);
276         stream.next_out = buf;
277         stream.avail_out = size;
278
279         do {
280                 stream.next_in = fill(1);
281                 stream.avail_in = input_len;
282                 status = git_inflate(&stream, 0);
283                 use(input_len - stream.avail_in);
284         } while (status == Z_OK);
285         if (stream.total_out != size || status != Z_STREAM_END)
286                 bad_object(offset, "inflate returned %d", status);
287         git_inflate_end(&stream);
288         return buf;
289 }
290
291 static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
292 {
293         unsigned char *p;
294         unsigned long size, c;
295         off_t base_offset;
296         unsigned shift;
297         void *data;
298
299         obj->idx.offset = consumed_bytes;
300         input_crc32 = crc32(0, NULL, 0);
301
302         p = fill(1);
303         c = *p;
304         use(1);
305         obj->type = (c >> 4) & 7;
306         size = (c & 15);
307         shift = 4;
308         while (c & 0x80) {
309                 p = fill(1);
310                 c = *p;
311                 use(1);
312                 size += (c & 0x7f) << shift;
313                 shift += 7;
314         }
315         obj->size = size;
316
317         switch (obj->type) {
318         case OBJ_REF_DELTA:
319                 hashcpy(delta_base->sha1, fill(20));
320                 use(20);
321                 break;
322         case OBJ_OFS_DELTA:
323                 memset(delta_base, 0, sizeof(*delta_base));
324                 p = fill(1);
325                 c = *p;
326                 use(1);
327                 base_offset = c & 127;
328                 while (c & 128) {
329                         base_offset += 1;
330                         if (!base_offset || MSB(base_offset, 7))
331                                 bad_object(obj->idx.offset, "offset value overflow for delta base object");
332                         p = fill(1);
333                         c = *p;
334                         use(1);
335                         base_offset = (base_offset << 7) + (c & 127);
336                 }
337                 delta_base->offset = obj->idx.offset - base_offset;
338                 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
339                         bad_object(obj->idx.offset, "delta base offset is out of bound");
340                 break;
341         case OBJ_COMMIT:
342         case OBJ_TREE:
343         case OBJ_BLOB:
344         case OBJ_TAG:
345                 break;
346         default:
347                 bad_object(obj->idx.offset, "unknown object type %d", obj->type);
348         }
349         obj->hdr_size = consumed_bytes - obj->idx.offset;
350
351         data = unpack_entry_data(obj->idx.offset, obj->size);
352         obj->idx.crc32 = input_crc32;
353         return data;
354 }
355
356 static void *get_data_from_pack(struct object_entry *obj)
357 {
358         off_t from = obj[0].idx.offset + obj[0].hdr_size;
359         unsigned long len = obj[1].idx.offset - from;
360         unsigned char *data, *inbuf;
361         git_zstream stream;
362         int status;
363
364         data = xmalloc(obj->size);
365         inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
366
367         memset(&stream, 0, sizeof(stream));
368         git_inflate_init(&stream);
369         stream.next_out = data;
370         stream.avail_out = obj->size;
371
372         do {
373                 ssize_t n = (len < 64*1024) ? len : 64*1024;
374                 n = pread(pack_fd, inbuf, n, from);
375                 if (n < 0)
376                         die_errno("cannot pread pack file");
377                 if (!n)
378                         die("premature end of pack file, %lu bytes missing", len);
379                 from += n;
380                 len -= n;
381                 stream.next_in = inbuf;
382                 stream.avail_in = n;
383                 status = git_inflate(&stream, 0);
384         } while (len && status == Z_OK && !stream.avail_in);
385
386         /* This has been inflated OK when first encountered, so... */
387         if (status != Z_STREAM_END || stream.total_out != obj->size)
388                 die("serious inflate inconsistency");
389
390         git_inflate_end(&stream);
391         free(inbuf);
392         return data;
393 }
394
395 static int compare_delta_bases(const union delta_base *base1,
396                                const union delta_base *base2,
397                                enum object_type type1,
398                                enum object_type type2)
399 {
400         int cmp = type1 - type2;
401         if (cmp)
402                 return cmp;
403         return memcmp(base1, base2, UNION_BASE_SZ);
404 }
405
406 static int find_delta(const union delta_base *base, enum object_type type)
407 {
408         int first = 0, last = nr_deltas;
409
410         while (first < last) {
411                 int next = (first + last) / 2;
412                 struct delta_entry *delta = &deltas[next];
413                 int cmp;
414
415                 cmp = compare_delta_bases(base, &delta->base,
416                                           type, objects[delta->obj_no].type);
417                 if (!cmp)
418                         return next;
419                 if (cmp < 0) {
420                         last = next;
421                         continue;
422                 }
423                 first = next+1;
424         }
425         return -first-1;
426 }
427
428 static void find_delta_children(const union delta_base *base,
429                                 int *first_index, int *last_index,
430                                 enum object_type type)
431 {
432         int first = find_delta(base, type);
433         int last = first;
434         int end = nr_deltas - 1;
435
436         if (first < 0) {
437                 *first_index = 0;
438                 *last_index = -1;
439                 return;
440         }
441         while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
442                 --first;
443         while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
444                 ++last;
445         *first_index = first;
446         *last_index = last;
447 }
448
449 static void sha1_object(const void *data, unsigned long size,
450                         enum object_type type, unsigned char *sha1)
451 {
452         hash_sha1_file(data, size, typename(type), sha1);
453         if (has_sha1_file(sha1)) {
454                 void *has_data;
455                 enum object_type has_type;
456                 unsigned long has_size;
457                 has_data = read_sha1_file(sha1, &has_type, &has_size);
458                 if (!has_data)
459                         die("cannot read existing object %s", sha1_to_hex(sha1));
460                 if (size != has_size || type != has_type ||
461                     memcmp(data, has_data, size) != 0)
462                         die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
463                 free(has_data);
464         }
465         if (strict) {
466                 if (type == OBJ_BLOB) {
467                         struct blob *blob = lookup_blob(sha1);
468                         if (blob)
469                                 blob->object.flags |= FLAG_CHECKED;
470                         else
471                                 die("invalid blob object %s", sha1_to_hex(sha1));
472                 } else {
473                         struct object *obj;
474                         int eaten;
475                         void *buf = (void *) data;
476
477                         /*
478                          * we do not need to free the memory here, as the
479                          * buf is deleted by the caller.
480                          */
481                         obj = parse_object_buffer(sha1, type, size, buf, &eaten);
482                         if (!obj)
483                                 die("invalid %s", typename(type));
484                         if (fsck_object(obj, 1, fsck_error_function))
485                                 die("Error in object");
486                         if (fsck_walk(obj, mark_link, NULL))
487                                 die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1));
488
489                         if (obj->type == OBJ_TREE) {
490                                 struct tree *item = (struct tree *) obj;
491                                 item->buffer = NULL;
492                         }
493                         if (obj->type == OBJ_COMMIT) {
494                                 struct commit *commit = (struct commit *) obj;
495                                 commit->buffer = NULL;
496                         }
497                         obj->flags |= FLAG_CHECKED;
498                 }
499         }
500 }
501
502 static int is_delta_type(enum object_type type)
503 {
504         return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
505 }
506
507 static void *get_base_data(struct base_data *c)
508 {
509         if (!c->data) {
510                 struct object_entry *obj = c->obj;
511
512                 if (is_delta_type(obj->type)) {
513                         void *base = get_base_data(c->base);
514                         void *raw = get_data_from_pack(obj);
515                         c->data = patch_delta(
516                                 base, c->base->size,
517                                 raw, obj->size,
518                                 &c->size);
519                         free(raw);
520                         if (!c->data)
521                                 bad_object(obj->idx.offset, "failed to apply delta");
522                 } else {
523                         c->data = get_data_from_pack(obj);
524                         c->size = obj->size;
525                 }
526
527                 base_cache_used += c->size;
528                 prune_base_data(c);
529         }
530         return c->data;
531 }
532
533 static void resolve_delta(struct object_entry *delta_obj,
534                           struct base_data *base, struct base_data *result)
535 {
536         void *base_data, *delta_data;
537
538         delta_obj->real_type = base->obj->real_type;
539         delta_obj->delta_depth = base->obj->delta_depth + 1;
540         if (deepest_delta < delta_obj->delta_depth)
541                 deepest_delta = delta_obj->delta_depth;
542         delta_obj->base_object_no = base->obj - objects;
543         delta_data = get_data_from_pack(delta_obj);
544         base_data = get_base_data(base);
545         result->obj = delta_obj;
546         result->data = patch_delta(base_data, base->size,
547                                    delta_data, delta_obj->size, &result->size);
548         free(delta_data);
549         if (!result->data)
550                 bad_object(delta_obj->idx.offset, "failed to apply delta");
551         sha1_object(result->data, result->size, delta_obj->real_type,
552                     delta_obj->idx.sha1);
553         nr_resolved_deltas++;
554 }
555
556 static void find_unresolved_deltas(struct base_data *base,
557                                    struct base_data *prev_base)
558 {
559         int i, ref_first, ref_last, ofs_first, ofs_last;
560
561         /*
562          * This is a recursive function. Those brackets should help reducing
563          * stack usage by limiting the scope of the delta_base union.
564          */
565         {
566                 union delta_base base_spec;
567
568                 hashcpy(base_spec.sha1, base->obj->idx.sha1);
569                 find_delta_children(&base_spec,
570                                     &ref_first, &ref_last, OBJ_REF_DELTA);
571
572                 memset(&base_spec, 0, sizeof(base_spec));
573                 base_spec.offset = base->obj->idx.offset;
574                 find_delta_children(&base_spec,
575                                     &ofs_first, &ofs_last, OBJ_OFS_DELTA);
576         }
577
578         if (ref_last == -1 && ofs_last == -1) {
579                 free(base->data);
580                 return;
581         }
582
583         link_base_data(prev_base, base);
584
585         for (i = ref_first; i <= ref_last; i++) {
586                 struct object_entry *child = objects + deltas[i].obj_no;
587                 struct base_data result;
588
589                 assert(child->real_type == OBJ_REF_DELTA);
590                 resolve_delta(child, base, &result);
591                 if (i == ref_last && ofs_last == -1)
592                         free_base_data(base);
593                 find_unresolved_deltas(&result, base);
594         }
595
596         for (i = ofs_first; i <= ofs_last; i++) {
597                 struct object_entry *child = objects + deltas[i].obj_no;
598                 struct base_data result;
599
600                 assert(child->real_type == OBJ_OFS_DELTA);
601                 resolve_delta(child, base, &result);
602                 if (i == ofs_last)
603                         free_base_data(base);
604                 find_unresolved_deltas(&result, base);
605         }
606
607         unlink_base_data(base);
608 }
609
610 static int compare_delta_entry(const void *a, const void *b)
611 {
612         const struct delta_entry *delta_a = a;
613         const struct delta_entry *delta_b = b;
614
615         /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
616         return compare_delta_bases(&delta_a->base, &delta_b->base,
617                                    objects[delta_a->obj_no].type,
618                                    objects[delta_b->obj_no].type);
619 }
620
621 /* Parse all objects and return the pack content SHA1 hash */
622 static void parse_pack_objects(unsigned char *sha1)
623 {
624         int i;
625         struct delta_entry *delta = deltas;
626         struct stat st;
627
628         /*
629          * First pass:
630          * - find locations of all objects;
631          * - calculate SHA1 of all non-delta objects;
632          * - remember base (SHA1 or offset) for all deltas.
633          */
634         if (verbose)
635                 progress = start_progress(
636                                 from_stdin ? "Receiving objects" : "Indexing objects",
637                                 nr_objects);
638         for (i = 0; i < nr_objects; i++) {
639                 struct object_entry *obj = &objects[i];
640                 void *data = unpack_raw_entry(obj, &delta->base);
641                 obj->real_type = obj->type;
642                 if (is_delta_type(obj->type)) {
643                         nr_deltas++;
644                         delta->obj_no = i;
645                         delta++;
646                 } else
647                         sha1_object(data, obj->size, obj->type, obj->idx.sha1);
648                 free(data);
649                 display_progress(progress, i+1);
650         }
651         objects[i].idx.offset = consumed_bytes;
652         stop_progress(&progress);
653
654         /* Check pack integrity */
655         flush();
656         git_SHA1_Final(sha1, &input_ctx);
657         if (hashcmp(fill(20), sha1))
658                 die("pack is corrupted (SHA1 mismatch)");
659         use(20);
660
661         /* If input_fd is a file, we should have reached its end now. */
662         if (fstat(input_fd, &st))
663                 die_errno("cannot fstat packfile");
664         if (S_ISREG(st.st_mode) &&
665                         lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
666                 die("pack has junk at the end");
667
668         if (!nr_deltas)
669                 return;
670
671         /* Sort deltas by base SHA1/offset for fast searching */
672         qsort(deltas, nr_deltas, sizeof(struct delta_entry),
673               compare_delta_entry);
674
675         /*
676          * Second pass:
677          * - for all non-delta objects, look if it is used as a base for
678          *   deltas;
679          * - if used as a base, uncompress the object and apply all deltas,
680          *   recursively checking if the resulting object is used as a base
681          *   for some more deltas.
682          */
683         if (verbose)
684                 progress = start_progress("Resolving deltas", nr_deltas);
685         for (i = 0; i < nr_objects; i++) {
686                 struct object_entry *obj = &objects[i];
687                 struct base_data base_obj;
688
689                 if (is_delta_type(obj->type))
690                         continue;
691                 base_obj.obj = obj;
692                 base_obj.data = NULL;
693                 find_unresolved_deltas(&base_obj, NULL);
694                 display_progress(progress, nr_resolved_deltas);
695         }
696 }
697
698 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
699 {
700         git_zstream stream;
701         int status;
702         unsigned char outbuf[4096];
703
704         memset(&stream, 0, sizeof(stream));
705         git_deflate_init(&stream, zlib_compression_level);
706         stream.next_in = in;
707         stream.avail_in = size;
708
709         do {
710                 stream.next_out = outbuf;
711                 stream.avail_out = sizeof(outbuf);
712                 status = git_deflate(&stream, Z_FINISH);
713                 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
714         } while (status == Z_OK);
715
716         if (status != Z_STREAM_END)
717                 die("unable to deflate appended object (%d)", status);
718         size = stream.total_out;
719         git_deflate_end(&stream);
720         return size;
721 }
722
723 static struct object_entry *append_obj_to_pack(struct sha1file *f,
724                                const unsigned char *sha1, void *buf,
725                                unsigned long size, enum object_type type)
726 {
727         struct object_entry *obj = &objects[nr_objects++];
728         unsigned char header[10];
729         unsigned long s = size;
730         int n = 0;
731         unsigned char c = (type << 4) | (s & 15);
732         s >>= 4;
733         while (s) {
734                 header[n++] = c | 0x80;
735                 c = s & 0x7f;
736                 s >>= 7;
737         }
738         header[n++] = c;
739         crc32_begin(f);
740         sha1write(f, header, n);
741         obj[0].size = size;
742         obj[0].hdr_size = n;
743         obj[0].type = type;
744         obj[0].real_type = type;
745         obj[1].idx.offset = obj[0].idx.offset + n;
746         obj[1].idx.offset += write_compressed(f, buf, size);
747         obj[0].idx.crc32 = crc32_end(f);
748         sha1flush(f);
749         hashcpy(obj->idx.sha1, sha1);
750         return obj;
751 }
752
753 static int delta_pos_compare(const void *_a, const void *_b)
754 {
755         struct delta_entry *a = *(struct delta_entry **)_a;
756         struct delta_entry *b = *(struct delta_entry **)_b;
757         return a->obj_no - b->obj_no;
758 }
759
760 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
761 {
762         struct delta_entry **sorted_by_pos;
763         int i, n = 0;
764
765         /*
766          * Since many unresolved deltas may well be themselves base objects
767          * for more unresolved deltas, we really want to include the
768          * smallest number of base objects that would cover as much delta
769          * as possible by picking the
770          * trunc deltas first, allowing for other deltas to resolve without
771          * additional base objects.  Since most base objects are to be found
772          * before deltas depending on them, a good heuristic is to start
773          * resolving deltas in the same order as their position in the pack.
774          */
775         sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
776         for (i = 0; i < nr_deltas; i++) {
777                 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
778                         continue;
779                 sorted_by_pos[n++] = &deltas[i];
780         }
781         qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
782
783         for (i = 0; i < n; i++) {
784                 struct delta_entry *d = sorted_by_pos[i];
785                 enum object_type type;
786                 struct base_data base_obj;
787
788                 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
789                         continue;
790                 base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size);
791                 if (!base_obj.data)
792                         continue;
793
794                 if (check_sha1_signature(d->base.sha1, base_obj.data,
795                                 base_obj.size, typename(type)))
796                         die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
797                 base_obj.obj = append_obj_to_pack(f, d->base.sha1,
798                                         base_obj.data, base_obj.size, type);
799                 find_unresolved_deltas(&base_obj, NULL);
800                 display_progress(progress, nr_resolved_deltas);
801         }
802         free(sorted_by_pos);
803 }
804
805 static void final(const char *final_pack_name, const char *curr_pack_name,
806                   const char *final_index_name, const char *curr_index_name,
807                   const char *keep_name, const char *keep_msg,
808                   unsigned char *sha1)
809 {
810         const char *report = "pack";
811         char name[PATH_MAX];
812         int err;
813
814         if (!from_stdin) {
815                 close(input_fd);
816         } else {
817                 fsync_or_die(output_fd, curr_pack_name);
818                 err = close(output_fd);
819                 if (err)
820                         die_errno("error while closing pack file");
821         }
822
823         if (keep_msg) {
824                 int keep_fd, keep_msg_len = strlen(keep_msg);
825
826                 if (!keep_name)
827                         keep_fd = odb_pack_keep(name, sizeof(name), sha1);
828                 else
829                         keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
830
831                 if (keep_fd < 0) {
832                         if (errno != EEXIST)
833                                 die_errno("cannot write keep file '%s'",
834                                           keep_name);
835                 } else {
836                         if (keep_msg_len > 0) {
837                                 write_or_die(keep_fd, keep_msg, keep_msg_len);
838                                 write_or_die(keep_fd, "\n", 1);
839                         }
840                         if (close(keep_fd) != 0)
841                                 die_errno("cannot close written keep file '%s'",
842                                     keep_name);
843                         report = "keep";
844                 }
845         }
846
847         if (final_pack_name != curr_pack_name) {
848                 if (!final_pack_name) {
849                         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
850                                  get_object_directory(), sha1_to_hex(sha1));
851                         final_pack_name = name;
852                 }
853                 if (move_temp_to_file(curr_pack_name, final_pack_name))
854                         die("cannot store pack file");
855         } else if (from_stdin)
856                 chmod(final_pack_name, 0444);
857
858         if (final_index_name != curr_index_name) {
859                 if (!final_index_name) {
860                         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
861                                  get_object_directory(), sha1_to_hex(sha1));
862                         final_index_name = name;
863                 }
864                 if (move_temp_to_file(curr_index_name, final_index_name))
865                         die("cannot store index file");
866         } else
867                 chmod(final_index_name, 0444);
868
869         if (!from_stdin) {
870                 printf("%s\n", sha1_to_hex(sha1));
871         } else {
872                 char buf[48];
873                 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
874                                    report, sha1_to_hex(sha1));
875                 write_or_die(1, buf, len);
876
877                 /*
878                  * Let's just mimic git-unpack-objects here and write
879                  * the last part of the input buffer to stdout.
880                  */
881                 while (input_len) {
882                         err = xwrite(1, input_buffer + input_offset, input_len);
883                         if (err <= 0)
884                                 break;
885                         input_len -= err;
886                         input_offset += err;
887                 }
888         }
889 }
890
891 static int git_index_pack_config(const char *k, const char *v, void *cb)
892 {
893         struct pack_idx_option *opts = cb;
894
895         if (!strcmp(k, "pack.indexversion")) {
896                 opts->version = git_config_int(k, v);
897                 if (opts->version > 2)
898                         die("bad pack.indexversion=%"PRIu32, opts->version);
899                 return 0;
900         }
901         return git_default_config(k, v, cb);
902 }
903
904 static int cmp_uint32(const void *a_, const void *b_)
905 {
906         uint32_t a = *((uint32_t *)a_);
907         uint32_t b = *((uint32_t *)b_);
908
909         return (a < b) ? -1 : (a != b);
910 }
911
912 static void read_v2_anomalous_offsets(struct packed_git *p,
913                                       struct pack_idx_option *opts)
914 {
915         const uint32_t *idx1, *idx2;
916         uint32_t i;
917
918         /* The address of the 4-byte offset table */
919         idx1 = (((const uint32_t *)p->index_data)
920                 + 2 /* 8-byte header */
921                 + 256 /* fan out */
922                 + 5 * p->num_objects /* 20-byte SHA-1 table */
923                 + p->num_objects /* CRC32 table */
924                 );
925
926         /* The address of the 8-byte offset table */
927         idx2 = idx1 + p->num_objects;
928
929         for (i = 0; i < p->num_objects; i++) {
930                 uint32_t off = ntohl(idx1[i]);
931                 if (!(off & 0x80000000))
932                         continue;
933                 off = off & 0x7fffffff;
934                 if (idx2[off * 2])
935                         continue;
936                 /*
937                  * The real offset is ntohl(idx2[off * 2]) in high 4
938                  * octets, and ntohl(idx2[off * 2 + 1]) in low 4
939                  * octets.  But idx2[off * 2] is Zero!!!
940                  */
941                 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
942                 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
943         }
944
945         if (1 < opts->anomaly_nr)
946                 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
947 }
948
949 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
950 {
951         struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
952
953         if (!p)
954                 die("Cannot open existing pack file '%s'", pack_name);
955         if (open_pack_index(p))
956                 die("Cannot open existing pack idx file for '%s'", pack_name);
957
958         /* Read the attributes from the existing idx file */
959         opts->version = p->index_version;
960
961         if (opts->version == 2)
962                 read_v2_anomalous_offsets(p, opts);
963
964         /*
965          * Get rid of the idx file as we do not need it anymore.
966          * NEEDSWORK: extract this bit from free_pack_by_name() in
967          * sha1_file.c, perhaps?  It shouldn't matter very much as we
968          * know we haven't installed this pack (hence we never have
969          * read anything from it).
970          */
971         close_pack_index(p);
972         free(p);
973 }
974
975 static void show_pack_info(int stat_only)
976 {
977         int i, baseobjects = nr_objects - nr_deltas;
978         unsigned long *chain_histogram = NULL;
979
980         if (deepest_delta)
981                 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
982
983         for (i = 0; i < nr_objects; i++) {
984                 struct object_entry *obj = &objects[i];
985
986                 if (is_delta_type(obj->type))
987                         chain_histogram[obj->delta_depth - 1]++;
988                 if (stat_only)
989                         continue;
990                 printf("%s %-6s %lu %lu %"PRIuMAX,
991                        sha1_to_hex(obj->idx.sha1),
992                        typename(obj->real_type), obj->size,
993                        (unsigned long)(obj[1].idx.offset - obj->idx.offset),
994                        (uintmax_t)obj->idx.offset);
995                 if (is_delta_type(obj->type)) {
996                         struct object_entry *bobj = &objects[obj->base_object_no];
997                         printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
998                 }
999                 putchar('\n');
1000         }
1001
1002         if (baseobjects)
1003                 printf("non delta: %d object%s\n",
1004                        baseobjects, baseobjects > 1 ? "s" : "");
1005         for (i = 0; i < deepest_delta; i++) {
1006                 if (!chain_histogram[i])
1007                         continue;
1008                 printf("chain length = %d: %lu object%s\n",
1009                        i + 1,
1010                        chain_histogram[i],
1011                        chain_histogram[i] > 1 ? "s" : "");
1012         }
1013 }
1014
1015 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1016 {
1017         int i, fix_thin_pack = 0, verify = 0, stat_only = 0, stat = 0;
1018         const char *curr_pack, *curr_index;
1019         const char *index_name = NULL, *pack_name = NULL;
1020         const char *keep_name = NULL, *keep_msg = NULL;
1021         char *index_name_buf = NULL, *keep_name_buf = NULL;
1022         struct pack_idx_entry **idx_objects;
1023         struct pack_idx_option opts;
1024         unsigned char pack_sha1[20];
1025
1026         if (argc == 2 && !strcmp(argv[1], "-h"))
1027                 usage(index_pack_usage);
1028
1029         read_replace_refs = 0;
1030
1031         reset_pack_idx_option(&opts);
1032         git_config(git_index_pack_config, &opts);
1033         if (prefix && chdir(prefix))
1034                 die("Cannot come back to cwd");
1035
1036         for (i = 1; i < argc; i++) {
1037                 const char *arg = argv[i];
1038
1039                 if (*arg == '-') {
1040                         if (!strcmp(arg, "--stdin")) {
1041                                 from_stdin = 1;
1042                         } else if (!strcmp(arg, "--fix-thin")) {
1043                                 fix_thin_pack = 1;
1044                         } else if (!strcmp(arg, "--strict")) {
1045                                 strict = 1;
1046                         } else if (!strcmp(arg, "--verify")) {
1047                                 verify = 1;
1048                         } else if (!strcmp(arg, "--verify-stat")) {
1049                                 verify = 1;
1050                                 stat = 1;
1051                         } else if (!strcmp(arg, "--verify-stat-only")) {
1052                                 verify = 1;
1053                                 stat = 1;
1054                                 stat_only = 1;
1055                         } else if (!strcmp(arg, "--keep")) {
1056                                 keep_msg = "";
1057                         } else if (!prefixcmp(arg, "--keep=")) {
1058                                 keep_msg = arg + 7;
1059                         } else if (!prefixcmp(arg, "--pack_header=")) {
1060                                 struct pack_header *hdr;
1061                                 char *c;
1062
1063                                 hdr = (struct pack_header *)input_buffer;
1064                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1065                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1066                                 if (*c != ',')
1067                                         die("bad %s", arg);
1068                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1069                                 if (*c)
1070                                         die("bad %s", arg);
1071                                 input_len = sizeof(*hdr);
1072                         } else if (!strcmp(arg, "-v")) {
1073                                 verbose = 1;
1074                         } else if (!strcmp(arg, "-o")) {
1075                                 if (index_name || (i+1) >= argc)
1076                                         usage(index_pack_usage);
1077                                 index_name = argv[++i];
1078                         } else if (!prefixcmp(arg, "--index-version=")) {
1079                                 char *c;
1080                                 opts.version = strtoul(arg + 16, &c, 10);
1081                                 if (opts.version > 2)
1082                                         die("bad %s", arg);
1083                                 if (*c == ',')
1084                                         opts.off32_limit = strtoul(c+1, &c, 0);
1085                                 if (*c || opts.off32_limit & 0x80000000)
1086                                         die("bad %s", arg);
1087                         } else
1088                                 usage(index_pack_usage);
1089                         continue;
1090                 }
1091
1092                 if (pack_name)
1093                         usage(index_pack_usage);
1094                 pack_name = arg;
1095         }
1096
1097         if (!pack_name && !from_stdin)
1098                 usage(index_pack_usage);
1099         if (fix_thin_pack && !from_stdin)
1100                 die("--fix-thin cannot be used without --stdin");
1101         if (!index_name && pack_name) {
1102                 int len = strlen(pack_name);
1103                 if (!has_extension(pack_name, ".pack"))
1104                         die("packfile name '%s' does not end with '.pack'",
1105                             pack_name);
1106                 index_name_buf = xmalloc(len);
1107                 memcpy(index_name_buf, pack_name, len - 5);
1108                 strcpy(index_name_buf + len - 5, ".idx");
1109                 index_name = index_name_buf;
1110         }
1111         if (keep_msg && !keep_name && pack_name) {
1112                 int len = strlen(pack_name);
1113                 if (!has_extension(pack_name, ".pack"))
1114                         die("packfile name '%s' does not end with '.pack'",
1115                             pack_name);
1116                 keep_name_buf = xmalloc(len);
1117                 memcpy(keep_name_buf, pack_name, len - 5);
1118                 strcpy(keep_name_buf + len - 5, ".keep");
1119                 keep_name = keep_name_buf;
1120         }
1121         if (verify) {
1122                 if (!index_name)
1123                         die("--verify with no packfile name given");
1124                 read_idx_option(&opts, index_name);
1125                 opts.flags |= WRITE_IDX_VERIFY;
1126         }
1127
1128         curr_pack = open_pack_file(pack_name);
1129         parse_pack_header();
1130         objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1131         deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1132         parse_pack_objects(pack_sha1);
1133         if (nr_deltas == nr_resolved_deltas) {
1134                 stop_progress(&progress);
1135                 /* Flush remaining pack final 20-byte SHA1. */
1136                 flush();
1137         } else {
1138                 if (fix_thin_pack) {
1139                         struct sha1file *f;
1140                         unsigned char read_sha1[20], tail_sha1[20];
1141                         char msg[48];
1142                         int nr_unresolved = nr_deltas - nr_resolved_deltas;
1143                         int nr_objects_initial = nr_objects;
1144                         if (nr_unresolved <= 0)
1145                                 die("confusion beyond insanity");
1146                         objects = xrealloc(objects,
1147                                            (nr_objects + nr_unresolved + 1)
1148                                            * sizeof(*objects));
1149                         f = sha1fd(output_fd, curr_pack);
1150                         fix_unresolved_deltas(f, nr_unresolved);
1151                         sprintf(msg, "completed with %d local objects",
1152                                 nr_objects - nr_objects_initial);
1153                         stop_progress_msg(&progress, msg);
1154                         sha1close(f, tail_sha1, 0);
1155                         hashcpy(read_sha1, pack_sha1);
1156                         fixup_pack_header_footer(output_fd, pack_sha1,
1157                                                  curr_pack, nr_objects,
1158                                                  read_sha1, consumed_bytes-20);
1159                         if (hashcmp(read_sha1, tail_sha1) != 0)
1160                                 die("Unexpected tail checksum for %s "
1161                                     "(disk corruption?)", curr_pack);
1162                 }
1163                 if (nr_deltas != nr_resolved_deltas)
1164                         die("pack has %d unresolved deltas",
1165                             nr_deltas - nr_resolved_deltas);
1166         }
1167         free(deltas);
1168         if (strict)
1169                 check_objects();
1170
1171         if (stat)
1172                 show_pack_info(stat_only);
1173
1174         idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1175         for (i = 0; i < nr_objects; i++)
1176                 idx_objects[i] = &objects[i].idx;
1177         curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1178         free(idx_objects);
1179
1180         if (!verify)
1181                 final(pack_name, curr_pack,
1182                       index_name, curr_index,
1183                       keep_name, keep_msg,
1184                       pack_sha1);
1185         else
1186                 close(input_fd);
1187         free(objects);
1188         free(index_name_buf);
1189         free(keep_name_buf);
1190         if (pack_name == NULL)
1191                 free((void *) curr_pack);
1192         if (index_name == NULL)
1193                 free((void *) curr_index);
1194
1195         return 0;
1196 }