127e713ddec575ee15e4ff052711ef2121799a64
[git/git.git] / builtin-index-pack.c
1 #include "cache.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> }] [--strict] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";
15
16 struct object_entry
17 {
18 struct pack_idx_entry idx;
19 unsigned long size;
20 unsigned int hdr_size;
21 enum object_type type;
22 enum object_type real_type;
23 };
24
25 union delta_base {
26 unsigned char sha1[20];
27 off_t offset;
28 };
29
30 struct base_data {
31 struct base_data *base;
32 struct base_data *child;
33 struct object_entry *obj;
34 void *data;
35 unsigned long size;
36 };
37
38 /*
39 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
40 * to memcmp() only the first 20 bytes.
41 */
42 #define UNION_BASE_SZ 20
43
44 #define FLAG_LINK (1u<<20)
45 #define FLAG_CHECKED (1u<<21)
46
47 struct delta_entry
48 {
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 git_SHA_CTX input_ctx;
72 static uint32_t input_crc32;
73 static int input_fd, output_fd, pack_fd;
74
75 static int mark_link(struct object *obj, int type, void *data)
76 {
77 if (!obj)
78 return -1;
79
80 if (type != OBJ_ANY && obj->type != type)
81 die("object type mismatch at %s", sha1_to_hex(obj->sha1));
82
83 obj->flags |= FLAG_LINK;
84 return 0;
85 }
86
87 /* The content of each linked object must have been checked
88 or it must be already present in the object database */
89 static void check_object(struct object *obj)
90 {
91 if (!obj)
92 return;
93
94 if (!(obj->flags & FLAG_LINK))
95 return;
96
97 if (!(obj->flags & FLAG_CHECKED)) {
98 unsigned long size;
99 int type = sha1_object_info(obj->sha1, &size);
100 if (type != obj->type || type <= 0)
101 die("object of unexpected type");
102 obj->flags |= FLAG_CHECKED;
103 return;
104 }
105 }
106
107 static void check_objects(void)
108 {
109 unsigned i, max;
110
111 max = get_max_object_index();
112 for (i = 0; i < max; i++)
113 check_object(get_indexed_object(i));
114 }
115
116
117 /* Discard current buffer used content. */
118 static void flush(void)
119 {
120 if (input_offset) {
121 if (output_fd >= 0)
122 write_or_die(output_fd, input_buffer, input_offset);
123 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
124 memmove(input_buffer, input_buffer + input_offset, input_len);
125 input_offset = 0;
126 }
127 }
128
129 /*
130 * Make sure at least "min" bytes are available in the buffer, and
131 * return the pointer to the buffer.
132 */
133 static void *fill(int min)
134 {
135 if (min <= input_len)
136 return input_buffer + input_offset;
137 if (min > sizeof(input_buffer))
138 die("cannot fill %d bytes", min);
139 flush();
140 do {
141 ssize_t ret = xread(input_fd, input_buffer + input_len,
142 sizeof(input_buffer) - input_len);
143 if (ret <= 0) {
144 if (!ret)
145 die("early EOF");
146 die_errno("read error on input");
147 }
148 input_len += ret;
149 if (from_stdin)
150 display_throughput(progress, consumed_bytes + input_len);
151 } while (input_len < min);
152 return input_buffer;
153 }
154
155 static void use(int bytes)
156 {
157 if (bytes > input_len)
158 die("used more bytes than were available");
159 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
160 input_len -= bytes;
161 input_offset += bytes;
162
163 /* make sure off_t is sufficiently large not to wrap */
164 if (consumed_bytes > consumed_bytes + bytes)
165 die("pack too large for current definition of off_t");
166 consumed_bytes += bytes;
167 }
168
169 static const char *open_pack_file(const char *pack_name)
170 {
171 if (from_stdin) {
172 input_fd = 0;
173 if (!pack_name) {
174 static char tmpfile[PATH_MAX];
175 output_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
176 "pack/tmp_pack_XXXXXX");
177 pack_name = xstrdup(tmpfile);
178 } else
179 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
180 if (output_fd < 0)
181 die_errno("unable to create '%s'", pack_name);
182 pack_fd = output_fd;
183 } else {
184 input_fd = open(pack_name, O_RDONLY);
185 if (input_fd < 0)
186 die_errno("cannot open packfile '%s'", pack_name);
187 output_fd = -1;
188 pack_fd = input_fd;
189 }
190 git_SHA1_Init(&input_ctx);
191 return pack_name;
192 }
193
194 static void parse_pack_header(void)
195 {
196 struct pack_header *hdr = fill(sizeof(struct pack_header));
197
198 /* Header consistency check */
199 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
200 die("pack signature mismatch");
201 if (!pack_version_ok(hdr->hdr_version))
202 die("pack version %"PRIu32" unsupported",
203 ntohl(hdr->hdr_version));
204
205 nr_objects = ntohl(hdr->hdr_entries);
206 use(sizeof(struct pack_header));
207 }
208
209 static NORETURN void bad_object(unsigned long offset, const char *format,
210 ...) __attribute__((format (printf, 2, 3)));
211
212 static void bad_object(unsigned long offset, const char *format, ...)
213 {
214 va_list params;
215 char buf[1024];
216
217 va_start(params, format);
218 vsnprintf(buf, sizeof(buf), format, params);
219 va_end(params);
220 die("pack has bad object at offset %lu: %s", offset, buf);
221 }
222
223 static void free_base_data(struct base_data *c)
224 {
225 if (c->data) {
226 free(c->data);
227 c->data = NULL;
228 base_cache_used -= c->size;
229 }
230 }
231
232 static void prune_base_data(struct base_data *retain)
233 {
234 struct base_data *b;
235 for (b = base_cache;
236 base_cache_used > delta_base_cache_limit && b;
237 b = b->child) {
238 if (b->data && b != retain)
239 free_base_data(b);
240 }
241 }
242
243 static void link_base_data(struct base_data *base, struct base_data *c)
244 {
245 if (base)
246 base->child = c;
247 else
248 base_cache = c;
249
250 c->base = base;
251 c->child = NULL;
252 if (c->data)
253 base_cache_used += c->size;
254 prune_base_data(c);
255 }
256
257 static void unlink_base_data(struct base_data *c)
258 {
259 struct base_data *base = c->base;
260 if (base)
261 base->child = NULL;
262 else
263 base_cache = NULL;
264 free_base_data(c);
265 }
266
267 static void *unpack_entry_data(unsigned long offset, unsigned long size)
268 {
269 z_stream stream;
270 void *buf = xmalloc(size);
271
272 memset(&stream, 0, sizeof(stream));
273 stream.next_out = buf;
274 stream.avail_out = size;
275 stream.next_in = fill(1);
276 stream.avail_in = input_len;
277 git_inflate_init(&stream);
278
279 for (;;) {
280 int ret = git_inflate(&stream, 0);
281 use(input_len - stream.avail_in);
282 if (stream.total_out == size && ret == Z_STREAM_END)
283 break;
284 if (ret != Z_OK)
285 bad_object(offset, "inflate returned %d", ret);
286 stream.next_in = fill(1);
287 stream.avail_in = input_len;
288 }
289 git_inflate_end(&stream);
290 return buf;
291 }
292
293 static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
294 {
295 unsigned char *p;
296 unsigned long size, c;
297 off_t base_offset;
298 unsigned shift;
299 void *data;
300
301 obj->idx.offset = consumed_bytes;
302 input_crc32 = crc32(0, Z_NULL, 0);
303
304 p = fill(1);
305 c = *p;
306 use(1);
307 obj->type = (c >> 4) & 7;
308 size = (c & 15);
309 shift = 4;
310 while (c & 0x80) {
311 p = fill(1);
312 c = *p;
313 use(1);
314 size += (c & 0x7f) << shift;
315 shift += 7;
316 }
317 obj->size = size;
318
319 switch (obj->type) {
320 case OBJ_REF_DELTA:
321 hashcpy(delta_base->sha1, fill(20));
322 use(20);
323 break;
324 case OBJ_OFS_DELTA:
325 memset(delta_base, 0, sizeof(*delta_base));
326 p = fill(1);
327 c = *p;
328 use(1);
329 base_offset = c & 127;
330 while (c & 128) {
331 base_offset += 1;
332 if (!base_offset || MSB(base_offset, 7))
333 bad_object(obj->idx.offset, "offset value overflow for delta base object");
334 p = fill(1);
335 c = *p;
336 use(1);
337 base_offset = (base_offset << 7) + (c & 127);
338 }
339 delta_base->offset = obj->idx.offset - base_offset;
340 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
341 bad_object(obj->idx.offset, "delta base offset is out of bound");
342 break;
343 case OBJ_COMMIT:
344 case OBJ_TREE:
345 case OBJ_BLOB:
346 case OBJ_TAG:
347 break;
348 default:
349 bad_object(obj->idx.offset, "unknown object type %d", obj->type);
350 }
351 obj->hdr_size = consumed_bytes - obj->idx.offset;
352
353 data = unpack_entry_data(obj->idx.offset, obj->size);
354 obj->idx.crc32 = input_crc32;
355 return data;
356 }
357
358 static void *get_data_from_pack(struct object_entry *obj)
359 {
360 off_t from = obj[0].idx.offset + obj[0].hdr_size;
361 unsigned long len = obj[1].idx.offset - from;
362 unsigned char *data, *inbuf;
363 z_stream stream;
364 int status;
365
366 data = xmalloc(obj->size);
367 inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
368
369 memset(&stream, 0, sizeof(stream));
370 git_inflate_init(&stream);
371 stream.next_out = data;
372 stream.avail_out = obj->size;
373
374 do {
375 ssize_t n = (len < 64*1024) ? len : 64*1024;
376 n = pread(pack_fd, inbuf, n, from);
377 if (n < 0)
378 die_errno("cannot pread pack file");
379 if (!n)
380 die("premature end of pack file, %lu bytes missing", len);
381 from += n;
382 len -= n;
383 stream.next_in = inbuf;
384 stream.avail_in = n;
385 status = git_inflate(&stream, 0);
386 } while (len && status == Z_OK && !stream.avail_in);
387
388 /* This has been inflated OK when first encountered, so... */
389 if (status != Z_STREAM_END || stream.total_out != obj->size)
390 die("serious inflate inconsistency");
391
392 git_inflate_end(&stream);
393 free(inbuf);
394 return data;
395 }
396
397 static int find_delta(const union delta_base *base)
398 {
399 int first = 0, last = nr_deltas;
400
401 while (first < last) {
402 int next = (first + last) / 2;
403 struct delta_entry *delta = &deltas[next];
404 int cmp;
405
406 cmp = memcmp(base, &delta->base, UNION_BASE_SZ);
407 if (!cmp)
408 return next;
409 if (cmp < 0) {
410 last = next;
411 continue;
412 }
413 first = next+1;
414 }
415 return -first-1;
416 }
417
418 static void find_delta_children(const union delta_base *base,
419 int *first_index, int *last_index)
420 {
421 int first = find_delta(base);
422 int last = first;
423 int end = nr_deltas - 1;
424
425 if (first < 0) {
426 *first_index = 0;
427 *last_index = -1;
428 return;
429 }
430 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
431 --first;
432 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
433 ++last;
434 *first_index = first;
435 *last_index = last;
436 }
437
438 static void sha1_object(const void *data, unsigned long size,
439 enum object_type type, unsigned char *sha1)
440 {
441 hash_sha1_file(data, size, typename(type), sha1);
442 if (has_sha1_file(sha1)) {
443 void *has_data;
444 enum object_type has_type;
445 unsigned long has_size;
446 has_data = read_sha1_file(sha1, &has_type, &has_size);
447 if (!has_data)
448 die("cannot read existing object %s", sha1_to_hex(sha1));
449 if (size != has_size || type != has_type ||
450 memcmp(data, has_data, size) != 0)
451 die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
452 free(has_data);
453 }
454 if (strict) {
455 if (type == OBJ_BLOB) {
456 struct blob *blob = lookup_blob(sha1);
457 if (blob)
458 blob->object.flags |= FLAG_CHECKED;
459 else
460 die("invalid blob object %s", sha1_to_hex(sha1));
461 } else {
462 struct object *obj;
463 int eaten;
464 void *buf = (void *) data;
465
466 /*
467 * we do not need to free the memory here, as the
468 * buf is deleted by the caller.
469 */
470 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
471 if (!obj)
472 die("invalid %s", typename(type));
473 if (fsck_object(obj, 1, fsck_error_function))
474 die("Error in object");
475 if (fsck_walk(obj, mark_link, NULL))
476 die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1));
477
478 if (obj->type == OBJ_TREE) {
479 struct tree *item = (struct tree *) obj;
480 item->buffer = NULL;
481 }
482 if (obj->type == OBJ_COMMIT) {
483 struct commit *commit = (struct commit *) obj;
484 commit->buffer = NULL;
485 }
486 obj->flags |= FLAG_CHECKED;
487 }
488 }
489 }
490
491 static void *get_base_data(struct base_data *c)
492 {
493 if (!c->data) {
494 struct object_entry *obj = c->obj;
495
496 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
497 void *base = get_base_data(c->base);
498 void *raw = get_data_from_pack(obj);
499 c->data = patch_delta(
500 base, c->base->size,
501 raw, obj->size,
502 &c->size);
503 free(raw);
504 if (!c->data)
505 bad_object(obj->idx.offset, "failed to apply delta");
506 } else {
507 c->data = get_data_from_pack(obj);
508 c->size = obj->size;
509 }
510
511 base_cache_used += c->size;
512 prune_base_data(c);
513 }
514 return c->data;
515 }
516
517 static void resolve_delta(struct object_entry *delta_obj,
518 struct base_data *base, struct base_data *result)
519 {
520 void *base_data, *delta_data;
521
522 delta_obj->real_type = base->obj->real_type;
523 delta_data = get_data_from_pack(delta_obj);
524 base_data = get_base_data(base);
525 result->obj = delta_obj;
526 result->data = patch_delta(base_data, base->size,
527 delta_data, delta_obj->size, &result->size);
528 free(delta_data);
529 if (!result->data)
530 bad_object(delta_obj->idx.offset, "failed to apply delta");
531 sha1_object(result->data, result->size, delta_obj->real_type,
532 delta_obj->idx.sha1);
533 nr_resolved_deltas++;
534 }
535
536 static void find_unresolved_deltas(struct base_data *base,
537 struct base_data *prev_base)
538 {
539 int i, ref_first, ref_last, ofs_first, ofs_last;
540
541 /*
542 * This is a recursive function. Those brackets should help reducing
543 * stack usage by limiting the scope of the delta_base union.
544 */
545 {
546 union delta_base base_spec;
547
548 hashcpy(base_spec.sha1, base->obj->idx.sha1);
549 find_delta_children(&base_spec, &ref_first, &ref_last);
550
551 memset(&base_spec, 0, sizeof(base_spec));
552 base_spec.offset = base->obj->idx.offset;
553 find_delta_children(&base_spec, &ofs_first, &ofs_last);
554 }
555
556 if (ref_last == -1 && ofs_last == -1) {
557 free(base->data);
558 return;
559 }
560
561 link_base_data(prev_base, base);
562
563 for (i = ref_first; i <= ref_last; i++) {
564 struct object_entry *child = objects + deltas[i].obj_no;
565 if (child->real_type == OBJ_REF_DELTA) {
566 struct base_data result;
567 resolve_delta(child, base, &result);
568 if (i == ref_last && ofs_last == -1)
569 free_base_data(base);
570 find_unresolved_deltas(&result, base);
571 }
572 }
573
574 for (i = ofs_first; i <= ofs_last; i++) {
575 struct object_entry *child = objects + deltas[i].obj_no;
576 if (child->real_type == OBJ_OFS_DELTA) {
577 struct base_data result;
578 resolve_delta(child, base, &result);
579 if (i == ofs_last)
580 free_base_data(base);
581 find_unresolved_deltas(&result, base);
582 }
583 }
584
585 unlink_base_data(base);
586 }
587
588 static int compare_delta_entry(const void *a, const void *b)
589 {
590 const struct delta_entry *delta_a = a;
591 const struct delta_entry *delta_b = b;
592 return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ);
593 }
594
595 /* Parse all objects and return the pack content SHA1 hash */
596 static void parse_pack_objects(unsigned char *sha1)
597 {
598 int i;
599 struct delta_entry *delta = deltas;
600 struct stat st;
601
602 /*
603 * First pass:
604 * - find locations of all objects;
605 * - calculate SHA1 of all non-delta objects;
606 * - remember base (SHA1 or offset) for all deltas.
607 */
608 if (verbose)
609 progress = start_progress(
610 from_stdin ? "Receiving objects" : "Indexing objects",
611 nr_objects);
612 for (i = 0; i < nr_objects; i++) {
613 struct object_entry *obj = &objects[i];
614 void *data = unpack_raw_entry(obj, &delta->base);
615 obj->real_type = obj->type;
616 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
617 nr_deltas++;
618 delta->obj_no = i;
619 delta++;
620 } else
621 sha1_object(data, obj->size, obj->type, obj->idx.sha1);
622 free(data);
623 display_progress(progress, i+1);
624 }
625 objects[i].idx.offset = consumed_bytes;
626 stop_progress(&progress);
627
628 /* Check pack integrity */
629 flush();
630 git_SHA1_Final(sha1, &input_ctx);
631 if (hashcmp(fill(20), sha1))
632 die("pack is corrupted (SHA1 mismatch)");
633 use(20);
634
635 /* If input_fd is a file, we should have reached its end now. */
636 if (fstat(input_fd, &st))
637 die_errno("cannot fstat packfile");
638 if (S_ISREG(st.st_mode) &&
639 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
640 die("pack has junk at the end");
641
642 if (!nr_deltas)
643 return;
644
645 /* Sort deltas by base SHA1/offset for fast searching */
646 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
647 compare_delta_entry);
648
649 /*
650 * Second pass:
651 * - for all non-delta objects, look if it is used as a base for
652 * deltas;
653 * - if used as a base, uncompress the object and apply all deltas,
654 * recursively checking if the resulting object is used as a base
655 * for some more deltas.
656 */
657 if (verbose)
658 progress = start_progress("Resolving deltas", nr_deltas);
659 for (i = 0; i < nr_objects; i++) {
660 struct object_entry *obj = &objects[i];
661 struct base_data base_obj;
662
663 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA)
664 continue;
665 base_obj.obj = obj;
666 base_obj.data = NULL;
667 find_unresolved_deltas(&base_obj, NULL);
668 display_progress(progress, nr_resolved_deltas);
669 }
670 }
671
672 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
673 {
674 z_stream stream;
675 unsigned long maxsize;
676 void *out;
677
678 memset(&stream, 0, sizeof(stream));
679 deflateInit(&stream, zlib_compression_level);
680 maxsize = deflateBound(&stream, size);
681 out = xmalloc(maxsize);
682
683 /* Compress it */
684 stream.next_in = in;
685 stream.avail_in = size;
686 stream.next_out = out;
687 stream.avail_out = maxsize;
688 while (deflate(&stream, Z_FINISH) == Z_OK);
689 deflateEnd(&stream);
690
691 size = stream.total_out;
692 sha1write(f, out, size);
693 free(out);
694 return size;
695 }
696
697 static struct object_entry *append_obj_to_pack(struct sha1file *f,
698 const unsigned char *sha1, void *buf,
699 unsigned long size, enum object_type type)
700 {
701 struct object_entry *obj = &objects[nr_objects++];
702 unsigned char header[10];
703 unsigned long s = size;
704 int n = 0;
705 unsigned char c = (type << 4) | (s & 15);
706 s >>= 4;
707 while (s) {
708 header[n++] = c | 0x80;
709 c = s & 0x7f;
710 s >>= 7;
711 }
712 header[n++] = c;
713 crc32_begin(f);
714 sha1write(f, header, n);
715 obj[0].size = size;
716 obj[0].hdr_size = n;
717 obj[0].type = type;
718 obj[0].real_type = type;
719 obj[1].idx.offset = obj[0].idx.offset + n;
720 obj[1].idx.offset += write_compressed(f, buf, size);
721 obj[0].idx.crc32 = crc32_end(f);
722 sha1flush(f);
723 hashcpy(obj->idx.sha1, sha1);
724 return obj;
725 }
726
727 static int delta_pos_compare(const void *_a, const void *_b)
728 {
729 struct delta_entry *a = *(struct delta_entry **)_a;
730 struct delta_entry *b = *(struct delta_entry **)_b;
731 return a->obj_no - b->obj_no;
732 }
733
734 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
735 {
736 struct delta_entry **sorted_by_pos;
737 int i, n = 0;
738
739 /*
740 * Since many unresolved deltas may well be themselves base objects
741 * for more unresolved deltas, we really want to include the
742 * smallest number of base objects that would cover as much delta
743 * as possible by picking the
744 * trunc deltas first, allowing for other deltas to resolve without
745 * additional base objects. Since most base objects are to be found
746 * before deltas depending on them, a good heuristic is to start
747 * resolving deltas in the same order as their position in the pack.
748 */
749 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
750 for (i = 0; i < nr_deltas; i++) {
751 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
752 continue;
753 sorted_by_pos[n++] = &deltas[i];
754 }
755 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
756
757 for (i = 0; i < n; i++) {
758 struct delta_entry *d = sorted_by_pos[i];
759 enum object_type type;
760 struct base_data base_obj;
761
762 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
763 continue;
764 base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size);
765 if (!base_obj.data)
766 continue;
767
768 if (check_sha1_signature(d->base.sha1, base_obj.data,
769 base_obj.size, typename(type)))
770 die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
771 base_obj.obj = append_obj_to_pack(f, d->base.sha1,
772 base_obj.data, base_obj.size, type);
773 find_unresolved_deltas(&base_obj, NULL);
774 display_progress(progress, nr_resolved_deltas);
775 }
776 free(sorted_by_pos);
777 }
778
779 static void final(const char *final_pack_name, const char *curr_pack_name,
780 const char *final_index_name, const char *curr_index_name,
781 const char *keep_name, const char *keep_msg,
782 unsigned char *sha1)
783 {
784 const char *report = "pack";
785 char name[PATH_MAX];
786 int err;
787
788 if (!from_stdin) {
789 close(input_fd);
790 } else {
791 fsync_or_die(output_fd, curr_pack_name);
792 err = close(output_fd);
793 if (err)
794 die_errno("error while closing pack file");
795 }
796
797 if (keep_msg) {
798 int keep_fd, keep_msg_len = strlen(keep_msg);
799
800 if (!keep_name)
801 keep_fd = odb_pack_keep(name, sizeof(name), sha1);
802 else
803 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
804
805 if (keep_fd < 0) {
806 if (errno != EEXIST)
807 die_errno("cannot write keep file '%s'",
808 keep_name);
809 } else {
810 if (keep_msg_len > 0) {
811 write_or_die(keep_fd, keep_msg, keep_msg_len);
812 write_or_die(keep_fd, "\n", 1);
813 }
814 if (close(keep_fd) != 0)
815 die_errno("cannot close written keep file '%s'",
816 keep_name);
817 report = "keep";
818 }
819 }
820
821 if (final_pack_name != curr_pack_name) {
822 if (!final_pack_name) {
823 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
824 get_object_directory(), sha1_to_hex(sha1));
825 final_pack_name = name;
826 }
827 if (move_temp_to_file(curr_pack_name, final_pack_name))
828 die("cannot store pack file");
829 } else if (from_stdin)
830 chmod(final_pack_name, 0444);
831
832 if (final_index_name != curr_index_name) {
833 if (!final_index_name) {
834 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
835 get_object_directory(), sha1_to_hex(sha1));
836 final_index_name = name;
837 }
838 if (move_temp_to_file(curr_index_name, final_index_name))
839 die("cannot store index file");
840 } else
841 chmod(final_index_name, 0444);
842
843 if (!from_stdin) {
844 printf("%s\n", sha1_to_hex(sha1));
845 } else {
846 char buf[48];
847 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
848 report, sha1_to_hex(sha1));
849 write_or_die(1, buf, len);
850
851 /*
852 * Let's just mimic git-unpack-objects here and write
853 * the last part of the input buffer to stdout.
854 */
855 while (input_len) {
856 err = xwrite(1, input_buffer + input_offset, input_len);
857 if (err <= 0)
858 break;
859 input_len -= err;
860 input_offset += err;
861 }
862 }
863 }
864
865 static int git_index_pack_config(const char *k, const char *v, void *cb)
866 {
867 if (!strcmp(k, "pack.indexversion")) {
868 pack_idx_default_version = git_config_int(k, v);
869 if (pack_idx_default_version > 2)
870 die("bad pack.indexversion=%"PRIu32,
871 pack_idx_default_version);
872 return 0;
873 }
874 return git_default_config(k, v, cb);
875 }
876
877 int cmd_index_pack(int argc, const char **argv, const char *prefix)
878 {
879 int i, fix_thin_pack = 0;
880 const char *curr_pack, *curr_index;
881 const char *index_name = NULL, *pack_name = NULL;
882 const char *keep_name = NULL, *keep_msg = NULL;
883 char *index_name_buf = NULL, *keep_name_buf = NULL;
884 struct pack_idx_entry **idx_objects;
885 unsigned char pack_sha1[20];
886
887 if (argc == 2 && !strcmp(argv[1], "-h"))
888 usage(index_pack_usage);
889
890 /*
891 * We wish to read the repository's config file if any, and
892 * for that it is necessary to call setup_git_directory_gently().
893 * However if the cwd was inside .git/objects/pack/ then we need
894 * to go back there or all the pack name arguments will be wrong.
895 * And in that case we cannot rely on any prefix returned by
896 * setup_git_directory_gently() either.
897 */
898 {
899 char cwd[PATH_MAX+1];
900 int nongit;
901
902 if (!getcwd(cwd, sizeof(cwd)-1))
903 die("Unable to get current working directory");
904 setup_git_directory_gently(&nongit);
905 git_config(git_index_pack_config, NULL);
906 if (chdir(cwd))
907 die("Cannot come back to cwd");
908 }
909
910 for (i = 1; i < argc; i++) {
911 const char *arg = argv[i];
912
913 if (*arg == '-') {
914 if (!strcmp(arg, "--stdin")) {
915 from_stdin = 1;
916 } else if (!strcmp(arg, "--fix-thin")) {
917 fix_thin_pack = 1;
918 } else if (!strcmp(arg, "--strict")) {
919 strict = 1;
920 } else if (!strcmp(arg, "--keep")) {
921 keep_msg = "";
922 } else if (!prefixcmp(arg, "--keep=")) {
923 keep_msg = arg + 7;
924 } else if (!prefixcmp(arg, "--pack_header=")) {
925 struct pack_header *hdr;
926 char *c;
927
928 hdr = (struct pack_header *)input_buffer;
929 hdr->hdr_signature = htonl(PACK_SIGNATURE);
930 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
931 if (*c != ',')
932 die("bad %s", arg);
933 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
934 if (*c)
935 die("bad %s", arg);
936 input_len = sizeof(*hdr);
937 } else if (!strcmp(arg, "-v")) {
938 verbose = 1;
939 } else if (!strcmp(arg, "-o")) {
940 if (index_name || (i+1) >= argc)
941 usage(index_pack_usage);
942 index_name = argv[++i];
943 } else if (!prefixcmp(arg, "--index-version=")) {
944 char *c;
945 pack_idx_default_version = strtoul(arg + 16, &c, 10);
946 if (pack_idx_default_version > 2)
947 die("bad %s", arg);
948 if (*c == ',')
949 pack_idx_off32_limit = strtoul(c+1, &c, 0);
950 if (*c || pack_idx_off32_limit & 0x80000000)
951 die("bad %s", arg);
952 } else
953 usage(index_pack_usage);
954 continue;
955 }
956
957 if (pack_name)
958 usage(index_pack_usage);
959 pack_name = arg;
960 }
961
962 if (!pack_name && !from_stdin)
963 usage(index_pack_usage);
964 if (fix_thin_pack && !from_stdin)
965 die("--fix-thin cannot be used without --stdin");
966 if (!index_name && pack_name) {
967 int len = strlen(pack_name);
968 if (!has_extension(pack_name, ".pack"))
969 die("packfile name '%s' does not end with '.pack'",
970 pack_name);
971 index_name_buf = xmalloc(len);
972 memcpy(index_name_buf, pack_name, len - 5);
973 strcpy(index_name_buf + len - 5, ".idx");
974 index_name = index_name_buf;
975 }
976 if (keep_msg && !keep_name && pack_name) {
977 int len = strlen(pack_name);
978 if (!has_extension(pack_name, ".pack"))
979 die("packfile name '%s' does not end with '.pack'",
980 pack_name);
981 keep_name_buf = xmalloc(len);
982 memcpy(keep_name_buf, pack_name, len - 5);
983 strcpy(keep_name_buf + len - 5, ".keep");
984 keep_name = keep_name_buf;
985 }
986
987 curr_pack = open_pack_file(pack_name);
988 parse_pack_header();
989 objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry));
990 deltas = xmalloc(nr_objects * sizeof(struct delta_entry));
991 parse_pack_objects(pack_sha1);
992 if (nr_deltas == nr_resolved_deltas) {
993 stop_progress(&progress);
994 /* Flush remaining pack final 20-byte SHA1. */
995 flush();
996 } else {
997 if (fix_thin_pack) {
998 struct sha1file *f;
999 unsigned char read_sha1[20], tail_sha1[20];
1000 char msg[48];
1001 int nr_unresolved = nr_deltas - nr_resolved_deltas;
1002 int nr_objects_initial = nr_objects;
1003 if (nr_unresolved <= 0)
1004 die("confusion beyond insanity");
1005 objects = xrealloc(objects,
1006 (nr_objects + nr_unresolved + 1)
1007 * sizeof(*objects));
1008 f = sha1fd(output_fd, curr_pack);
1009 fix_unresolved_deltas(f, nr_unresolved);
1010 sprintf(msg, "completed with %d local objects",
1011 nr_objects - nr_objects_initial);
1012 stop_progress_msg(&progress, msg);
1013 sha1close(f, tail_sha1, 0);
1014 hashcpy(read_sha1, pack_sha1);
1015 fixup_pack_header_footer(output_fd, pack_sha1,
1016 curr_pack, nr_objects,
1017 read_sha1, consumed_bytes-20);
1018 if (hashcmp(read_sha1, tail_sha1) != 0)
1019 die("Unexpected tail checksum for %s "
1020 "(disk corruption?)", curr_pack);
1021 }
1022 if (nr_deltas != nr_resolved_deltas)
1023 die("pack has %d unresolved deltas",
1024 nr_deltas - nr_resolved_deltas);
1025 }
1026 free(deltas);
1027 if (strict)
1028 check_objects();
1029
1030 idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1031 for (i = 0; i < nr_objects; i++)
1032 idx_objects[i] = &objects[i].idx;
1033 curr_index = write_idx_file(index_name, idx_objects, nr_objects, pack_sha1);
1034 free(idx_objects);
1035
1036 final(pack_name, curr_pack,
1037 index_name, curr_index,
1038 keep_name, keep_msg,
1039 pack_sha1);
1040 free(objects);
1041 free(index_name_buf);
1042 free(keep_name_buf);
1043 if (pack_name == NULL)
1044 free((void *) curr_pack);
1045 if (index_name == NULL)
1046 free((void *) curr_index);
1047
1048 return 0;
1049 }