]>
Commit | Line | Data |
---|---|---|
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 (signed_add_overflows(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 | int status; | |
270 | z_stream stream; | |
271 | void *buf = xmalloc(size); | |
272 | ||
273 | memset(&stream, 0, sizeof(stream)); | |
274 | git_inflate_init(&stream); | |
275 | stream.next_out = buf; | |
276 | stream.avail_out = size; | |
277 | ||
278 | do { | |
279 | stream.next_in = fill(1); | |
280 | stream.avail_in = input_len; | |
281 | status = git_inflate(&stream, 0); | |
282 | use(input_len - stream.avail_in); | |
283 | } while (status == Z_OK); | |
284 | if (stream.total_out != size || status != Z_STREAM_END) | |
285 | bad_object(offset, "inflate returned %d", status); | |
286 | git_inflate_end(&stream); | |
287 | return buf; | |
288 | } | |
289 | ||
290 | static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base) | |
291 | { | |
292 | unsigned char *p; | |
293 | unsigned long size, c; | |
294 | off_t base_offset; | |
295 | unsigned shift; | |
296 | void *data; | |
297 | ||
298 | obj->idx.offset = consumed_bytes; | |
299 | input_crc32 = crc32(0, Z_NULL, 0); | |
300 | ||
301 | p = fill(1); | |
302 | c = *p; | |
303 | use(1); | |
304 | obj->type = (c >> 4) & 7; | |
305 | size = (c & 15); | |
306 | shift = 4; | |
307 | while (c & 0x80) { | |
308 | p = fill(1); | |
309 | c = *p; | |
310 | use(1); | |
311 | size += (c & 0x7f) << shift; | |
312 | shift += 7; | |
313 | } | |
314 | obj->size = size; | |
315 | ||
316 | switch (obj->type) { | |
317 | case OBJ_REF_DELTA: | |
318 | hashcpy(delta_base->sha1, fill(20)); | |
319 | use(20); | |
320 | break; | |
321 | case OBJ_OFS_DELTA: | |
322 | memset(delta_base, 0, sizeof(*delta_base)); | |
323 | p = fill(1); | |
324 | c = *p; | |
325 | use(1); | |
326 | base_offset = c & 127; | |
327 | while (c & 128) { | |
328 | base_offset += 1; | |
329 | if (!base_offset || MSB(base_offset, 7)) | |
330 | bad_object(obj->idx.offset, "offset value overflow for delta base object"); | |
331 | p = fill(1); | |
332 | c = *p; | |
333 | use(1); | |
334 | base_offset = (base_offset << 7) + (c & 127); | |
335 | } | |
336 | delta_base->offset = obj->idx.offset - base_offset; | |
337 | if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset) | |
338 | bad_object(obj->idx.offset, "delta base offset is out of bound"); | |
339 | break; | |
340 | case OBJ_COMMIT: | |
341 | case OBJ_TREE: | |
342 | case OBJ_BLOB: | |
343 | case OBJ_TAG: | |
344 | break; | |
345 | default: | |
346 | bad_object(obj->idx.offset, "unknown object type %d", obj->type); | |
347 | } | |
348 | obj->hdr_size = consumed_bytes - obj->idx.offset; | |
349 | ||
350 | data = unpack_entry_data(obj->idx.offset, obj->size); | |
351 | obj->idx.crc32 = input_crc32; | |
352 | return data; | |
353 | } | |
354 | ||
355 | static void *get_data_from_pack(struct object_entry *obj) | |
356 | { | |
357 | off_t from = obj[0].idx.offset + obj[0].hdr_size; | |
358 | unsigned long len = obj[1].idx.offset - from; | |
359 | unsigned char *data, *inbuf; | |
360 | z_stream stream; | |
361 | int status; | |
362 | ||
363 | data = xmalloc(obj->size); | |
364 | inbuf = xmalloc((len < 64*1024) ? len : 64*1024); | |
365 | ||
366 | memset(&stream, 0, sizeof(stream)); | |
367 | git_inflate_init(&stream); | |
368 | stream.next_out = data; | |
369 | stream.avail_out = obj->size; | |
370 | ||
371 | do { | |
372 | ssize_t n = (len < 64*1024) ? len : 64*1024; | |
373 | n = pread(pack_fd, inbuf, n, from); | |
374 | if (n < 0) | |
375 | die_errno("cannot pread pack file"); | |
376 | if (!n) | |
377 | die("premature end of pack file, %lu bytes missing", len); | |
378 | from += n; | |
379 | len -= n; | |
380 | stream.next_in = inbuf; | |
381 | stream.avail_in = n; | |
382 | status = git_inflate(&stream, 0); | |
383 | } while (len && status == Z_OK && !stream.avail_in); | |
384 | ||
385 | /* This has been inflated OK when first encountered, so... */ | |
386 | if (status != Z_STREAM_END || stream.total_out != obj->size) | |
387 | die("serious inflate inconsistency"); | |
388 | ||
389 | git_inflate_end(&stream); | |
390 | free(inbuf); | |
391 | return data; | |
392 | } | |
393 | ||
394 | static int compare_delta_bases(const union delta_base *base1, | |
395 | const union delta_base *base2, | |
396 | enum object_type type1, | |
397 | enum object_type type2) | |
398 | { | |
399 | int cmp = type1 - type2; | |
400 | if (cmp) | |
401 | return cmp; | |
402 | return memcmp(base1, base2, UNION_BASE_SZ); | |
403 | } | |
404 | ||
405 | static int find_delta(const union delta_base *base, enum object_type type) | |
406 | { | |
407 | int first = 0, last = nr_deltas; | |
408 | ||
409 | while (first < last) { | |
410 | int next = (first + last) / 2; | |
411 | struct delta_entry *delta = &deltas[next]; | |
412 | int cmp; | |
413 | ||
414 | cmp = compare_delta_bases(base, &delta->base, | |
415 | type, objects[delta->obj_no].type); | |
416 | if (!cmp) | |
417 | return next; | |
418 | if (cmp < 0) { | |
419 | last = next; | |
420 | continue; | |
421 | } | |
422 | first = next+1; | |
423 | } | |
424 | return -first-1; | |
425 | } | |
426 | ||
427 | static void find_delta_children(const union delta_base *base, | |
428 | int *first_index, int *last_index, | |
429 | enum object_type type) | |
430 | { | |
431 | int first = find_delta(base, type); | |
432 | int last = first; | |
433 | int end = nr_deltas - 1; | |
434 | ||
435 | if (first < 0) { | |
436 | *first_index = 0; | |
437 | *last_index = -1; | |
438 | return; | |
439 | } | |
440 | while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ)) | |
441 | --first; | |
442 | while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ)) | |
443 | ++last; | |
444 | *first_index = first; | |
445 | *last_index = last; | |
446 | } | |
447 | ||
448 | static void sha1_object(const void *data, unsigned long size, | |
449 | enum object_type type, unsigned char *sha1) | |
450 | { | |
451 | hash_sha1_file(data, size, typename(type), sha1); | |
452 | if (has_sha1_file(sha1)) { | |
453 | void *has_data; | |
454 | enum object_type has_type; | |
455 | unsigned long has_size; | |
456 | has_data = read_sha1_file(sha1, &has_type, &has_size); | |
457 | if (!has_data) | |
458 | die("cannot read existing object %s", sha1_to_hex(sha1)); | |
459 | if (size != has_size || type != has_type || | |
460 | memcmp(data, has_data, size) != 0) | |
461 | die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1)); | |
462 | free(has_data); | |
463 | } | |
464 | if (strict) { | |
465 | if (type == OBJ_BLOB) { | |
466 | struct blob *blob = lookup_blob(sha1); | |
467 | if (blob) | |
468 | blob->object.flags |= FLAG_CHECKED; | |
469 | else | |
470 | die("invalid blob object %s", sha1_to_hex(sha1)); | |
471 | } else { | |
472 | struct object *obj; | |
473 | int eaten; | |
474 | void *buf = (void *) data; | |
475 | ||
476 | /* | |
477 | * we do not need to free the memory here, as the | |
478 | * buf is deleted by the caller. | |
479 | */ | |
480 | obj = parse_object_buffer(sha1, type, size, buf, &eaten); | |
481 | if (!obj) | |
482 | die("invalid %s", typename(type)); | |
483 | if (fsck_object(obj, 1, fsck_error_function)) | |
484 | die("Error in object"); | |
485 | if (fsck_walk(obj, mark_link, NULL)) | |
486 | die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1)); | |
487 | ||
488 | if (obj->type == OBJ_TREE) { | |
489 | struct tree *item = (struct tree *) obj; | |
490 | item->buffer = NULL; | |
491 | } | |
492 | if (obj->type == OBJ_COMMIT) { | |
493 | struct commit *commit = (struct commit *) obj; | |
494 | commit->buffer = NULL; | |
495 | } | |
496 | obj->flags |= FLAG_CHECKED; | |
497 | } | |
498 | } | |
499 | } | |
500 | ||
501 | static void *get_base_data(struct base_data *c) | |
502 | { | |
503 | if (!c->data) { | |
504 | struct object_entry *obj = c->obj; | |
505 | ||
506 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) { | |
507 | void *base = get_base_data(c->base); | |
508 | void *raw = get_data_from_pack(obj); | |
509 | c->data = patch_delta( | |
510 | base, c->base->size, | |
511 | raw, obj->size, | |
512 | &c->size); | |
513 | free(raw); | |
514 | if (!c->data) | |
515 | bad_object(obj->idx.offset, "failed to apply delta"); | |
516 | } else { | |
517 | c->data = get_data_from_pack(obj); | |
518 | c->size = obj->size; | |
519 | } | |
520 | ||
521 | base_cache_used += c->size; | |
522 | prune_base_data(c); | |
523 | } | |
524 | return c->data; | |
525 | } | |
526 | ||
527 | static void resolve_delta(struct object_entry *delta_obj, | |
528 | struct base_data *base, struct base_data *result) | |
529 | { | |
530 | void *base_data, *delta_data; | |
531 | ||
532 | delta_obj->real_type = base->obj->real_type; | |
533 | delta_data = get_data_from_pack(delta_obj); | |
534 | base_data = get_base_data(base); | |
535 | result->obj = delta_obj; | |
536 | result->data = patch_delta(base_data, base->size, | |
537 | delta_data, delta_obj->size, &result->size); | |
538 | free(delta_data); | |
539 | if (!result->data) | |
540 | bad_object(delta_obj->idx.offset, "failed to apply delta"); | |
541 | sha1_object(result->data, result->size, delta_obj->real_type, | |
542 | delta_obj->idx.sha1); | |
543 | nr_resolved_deltas++; | |
544 | } | |
545 | ||
546 | static void find_unresolved_deltas(struct base_data *base, | |
547 | struct base_data *prev_base) | |
548 | { | |
549 | int i, ref_first, ref_last, ofs_first, ofs_last; | |
550 | ||
551 | /* | |
552 | * This is a recursive function. Those brackets should help reducing | |
553 | * stack usage by limiting the scope of the delta_base union. | |
554 | */ | |
555 | { | |
556 | union delta_base base_spec; | |
557 | ||
558 | hashcpy(base_spec.sha1, base->obj->idx.sha1); | |
559 | find_delta_children(&base_spec, | |
560 | &ref_first, &ref_last, OBJ_REF_DELTA); | |
561 | ||
562 | memset(&base_spec, 0, sizeof(base_spec)); | |
563 | base_spec.offset = base->obj->idx.offset; | |
564 | find_delta_children(&base_spec, | |
565 | &ofs_first, &ofs_last, OBJ_OFS_DELTA); | |
566 | } | |
567 | ||
568 | if (ref_last == -1 && ofs_last == -1) { | |
569 | free(base->data); | |
570 | return; | |
571 | } | |
572 | ||
573 | link_base_data(prev_base, base); | |
574 | ||
575 | for (i = ref_first; i <= ref_last; i++) { | |
576 | struct object_entry *child = objects + deltas[i].obj_no; | |
577 | struct base_data result; | |
578 | ||
579 | assert(child->real_type == OBJ_REF_DELTA); | |
580 | resolve_delta(child, base, &result); | |
581 | if (i == ref_last && ofs_last == -1) | |
582 | free_base_data(base); | |
583 | find_unresolved_deltas(&result, base); | |
584 | } | |
585 | ||
586 | for (i = ofs_first; i <= ofs_last; i++) { | |
587 | struct object_entry *child = objects + deltas[i].obj_no; | |
588 | struct base_data result; | |
589 | ||
590 | assert(child->real_type == OBJ_OFS_DELTA); | |
591 | resolve_delta(child, base, &result); | |
592 | if (i == ofs_last) | |
593 | free_base_data(base); | |
594 | find_unresolved_deltas(&result, base); | |
595 | } | |
596 | ||
597 | unlink_base_data(base); | |
598 | } | |
599 | ||
600 | static int compare_delta_entry(const void *a, const void *b) | |
601 | { | |
602 | const struct delta_entry *delta_a = a; | |
603 | const struct delta_entry *delta_b = b; | |
604 | ||
605 | /* group by type (ref vs ofs) and then by value (sha-1 or offset) */ | |
606 | return compare_delta_bases(&delta_a->base, &delta_b->base, | |
607 | objects[delta_a->obj_no].type, | |
608 | objects[delta_b->obj_no].type); | |
609 | } | |
610 | ||
611 | /* Parse all objects and return the pack content SHA1 hash */ | |
612 | static void parse_pack_objects(unsigned char *sha1) | |
613 | { | |
614 | int i; | |
615 | struct delta_entry *delta = deltas; | |
616 | struct stat st; | |
617 | ||
618 | /* | |
619 | * First pass: | |
620 | * - find locations of all objects; | |
621 | * - calculate SHA1 of all non-delta objects; | |
622 | * - remember base (SHA1 or offset) for all deltas. | |
623 | */ | |
624 | if (verbose) | |
625 | progress = start_progress( | |
626 | from_stdin ? "Receiving objects" : "Indexing objects", | |
627 | nr_objects); | |
628 | for (i = 0; i < nr_objects; i++) { | |
629 | struct object_entry *obj = &objects[i]; | |
630 | void *data = unpack_raw_entry(obj, &delta->base); | |
631 | obj->real_type = obj->type; | |
632 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) { | |
633 | nr_deltas++; | |
634 | delta->obj_no = i; | |
635 | delta++; | |
636 | } else | |
637 | sha1_object(data, obj->size, obj->type, obj->idx.sha1); | |
638 | free(data); | |
639 | display_progress(progress, i+1); | |
640 | } | |
641 | objects[i].idx.offset = consumed_bytes; | |
642 | stop_progress(&progress); | |
643 | ||
644 | /* Check pack integrity */ | |
645 | flush(); | |
646 | git_SHA1_Final(sha1, &input_ctx); | |
647 | if (hashcmp(fill(20), sha1)) | |
648 | die("pack is corrupted (SHA1 mismatch)"); | |
649 | use(20); | |
650 | ||
651 | /* If input_fd is a file, we should have reached its end now. */ | |
652 | if (fstat(input_fd, &st)) | |
653 | die_errno("cannot fstat packfile"); | |
654 | if (S_ISREG(st.st_mode) && | |
655 | lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size) | |
656 | die("pack has junk at the end"); | |
657 | ||
658 | if (!nr_deltas) | |
659 | return; | |
660 | ||
661 | /* Sort deltas by base SHA1/offset for fast searching */ | |
662 | qsort(deltas, nr_deltas, sizeof(struct delta_entry), | |
663 | compare_delta_entry); | |
664 | ||
665 | /* | |
666 | * Second pass: | |
667 | * - for all non-delta objects, look if it is used as a base for | |
668 | * deltas; | |
669 | * - if used as a base, uncompress the object and apply all deltas, | |
670 | * recursively checking if the resulting object is used as a base | |
671 | * for some more deltas. | |
672 | */ | |
673 | if (verbose) | |
674 | progress = start_progress("Resolving deltas", nr_deltas); | |
675 | for (i = 0; i < nr_objects; i++) { | |
676 | struct object_entry *obj = &objects[i]; | |
677 | struct base_data base_obj; | |
678 | ||
679 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) | |
680 | continue; | |
681 | base_obj.obj = obj; | |
682 | base_obj.data = NULL; | |
683 | find_unresolved_deltas(&base_obj, NULL); | |
684 | display_progress(progress, nr_resolved_deltas); | |
685 | } | |
686 | } | |
687 | ||
688 | static int write_compressed(struct sha1file *f, void *in, unsigned int size) | |
689 | { | |
690 | z_stream stream; | |
691 | int status; | |
692 | unsigned char outbuf[4096]; | |
693 | ||
694 | memset(&stream, 0, sizeof(stream)); | |
695 | deflateInit(&stream, zlib_compression_level); | |
696 | stream.next_in = in; | |
697 | stream.avail_in = size; | |
698 | ||
699 | do { | |
700 | stream.next_out = outbuf; | |
701 | stream.avail_out = sizeof(outbuf); | |
702 | status = deflate(&stream, Z_FINISH); | |
703 | sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out); | |
704 | } while (status == Z_OK); | |
705 | ||
706 | if (status != Z_STREAM_END) | |
707 | die("unable to deflate appended object (%d)", status); | |
708 | size = stream.total_out; | |
709 | deflateEnd(&stream); | |
710 | return size; | |
711 | } | |
712 | ||
713 | static struct object_entry *append_obj_to_pack(struct sha1file *f, | |
714 | const unsigned char *sha1, void *buf, | |
715 | unsigned long size, enum object_type type) | |
716 | { | |
717 | struct object_entry *obj = &objects[nr_objects++]; | |
718 | unsigned char header[10]; | |
719 | unsigned long s = size; | |
720 | int n = 0; | |
721 | unsigned char c = (type << 4) | (s & 15); | |
722 | s >>= 4; | |
723 | while (s) { | |
724 | header[n++] = c | 0x80; | |
725 | c = s & 0x7f; | |
726 | s >>= 7; | |
727 | } | |
728 | header[n++] = c; | |
729 | crc32_begin(f); | |
730 | sha1write(f, header, n); | |
731 | obj[0].size = size; | |
732 | obj[0].hdr_size = n; | |
733 | obj[0].type = type; | |
734 | obj[0].real_type = type; | |
735 | obj[1].idx.offset = obj[0].idx.offset + n; | |
736 | obj[1].idx.offset += write_compressed(f, buf, size); | |
737 | obj[0].idx.crc32 = crc32_end(f); | |
738 | sha1flush(f); | |
739 | hashcpy(obj->idx.sha1, sha1); | |
740 | return obj; | |
741 | } | |
742 | ||
743 | static int delta_pos_compare(const void *_a, const void *_b) | |
744 | { | |
745 | struct delta_entry *a = *(struct delta_entry **)_a; | |
746 | struct delta_entry *b = *(struct delta_entry **)_b; | |
747 | return a->obj_no - b->obj_no; | |
748 | } | |
749 | ||
750 | static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved) | |
751 | { | |
752 | struct delta_entry **sorted_by_pos; | |
753 | int i, n = 0; | |
754 | ||
755 | /* | |
756 | * Since many unresolved deltas may well be themselves base objects | |
757 | * for more unresolved deltas, we really want to include the | |
758 | * smallest number of base objects that would cover as much delta | |
759 | * as possible by picking the | |
760 | * trunc deltas first, allowing for other deltas to resolve without | |
761 | * additional base objects. Since most base objects are to be found | |
762 | * before deltas depending on them, a good heuristic is to start | |
763 | * resolving deltas in the same order as their position in the pack. | |
764 | */ | |
765 | sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos)); | |
766 | for (i = 0; i < nr_deltas; i++) { | |
767 | if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA) | |
768 | continue; | |
769 | sorted_by_pos[n++] = &deltas[i]; | |
770 | } | |
771 | qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare); | |
772 | ||
773 | for (i = 0; i < n; i++) { | |
774 | struct delta_entry *d = sorted_by_pos[i]; | |
775 | enum object_type type; | |
776 | struct base_data base_obj; | |
777 | ||
778 | if (objects[d->obj_no].real_type != OBJ_REF_DELTA) | |
779 | continue; | |
780 | base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size); | |
781 | if (!base_obj.data) | |
782 | continue; | |
783 | ||
784 | if (check_sha1_signature(d->base.sha1, base_obj.data, | |
785 | base_obj.size, typename(type))) | |
786 | die("local object %s is corrupt", sha1_to_hex(d->base.sha1)); | |
787 | base_obj.obj = append_obj_to_pack(f, d->base.sha1, | |
788 | base_obj.data, base_obj.size, type); | |
789 | find_unresolved_deltas(&base_obj, NULL); | |
790 | display_progress(progress, nr_resolved_deltas); | |
791 | } | |
792 | free(sorted_by_pos); | |
793 | } | |
794 | ||
795 | static void final(const char *final_pack_name, const char *curr_pack_name, | |
796 | const char *final_index_name, const char *curr_index_name, | |
797 | const char *keep_name, const char *keep_msg, | |
798 | unsigned char *sha1) | |
799 | { | |
800 | const char *report = "pack"; | |
801 | char name[PATH_MAX]; | |
802 | int err; | |
803 | ||
804 | if (!from_stdin) { | |
805 | close(input_fd); | |
806 | } else { | |
807 | fsync_or_die(output_fd, curr_pack_name); | |
808 | err = close(output_fd); | |
809 | if (err) | |
810 | die_errno("error while closing pack file"); | |
811 | } | |
812 | ||
813 | if (keep_msg) { | |
814 | int keep_fd, keep_msg_len = strlen(keep_msg); | |
815 | ||
816 | if (!keep_name) | |
817 | keep_fd = odb_pack_keep(name, sizeof(name), sha1); | |
818 | else | |
819 | keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600); | |
820 | ||
821 | if (keep_fd < 0) { | |
822 | if (errno != EEXIST) | |
823 | die_errno("cannot write keep file '%s'", | |
824 | keep_name); | |
825 | } else { | |
826 | if (keep_msg_len > 0) { | |
827 | write_or_die(keep_fd, keep_msg, keep_msg_len); | |
828 | write_or_die(keep_fd, "\n", 1); | |
829 | } | |
830 | if (close(keep_fd) != 0) | |
831 | die_errno("cannot close written keep file '%s'", | |
832 | keep_name); | |
833 | report = "keep"; | |
834 | } | |
835 | } | |
836 | ||
837 | if (final_pack_name != curr_pack_name) { | |
838 | if (!final_pack_name) { | |
839 | snprintf(name, sizeof(name), "%s/pack/pack-%s.pack", | |
840 | get_object_directory(), sha1_to_hex(sha1)); | |
841 | final_pack_name = name; | |
842 | } | |
843 | if (move_temp_to_file(curr_pack_name, final_pack_name)) | |
844 | die("cannot store pack file"); | |
845 | } else if (from_stdin) | |
846 | chmod(final_pack_name, 0444); | |
847 | ||
848 | if (final_index_name != curr_index_name) { | |
849 | if (!final_index_name) { | |
850 | snprintf(name, sizeof(name), "%s/pack/pack-%s.idx", | |
851 | get_object_directory(), sha1_to_hex(sha1)); | |
852 | final_index_name = name; | |
853 | } | |
854 | if (move_temp_to_file(curr_index_name, final_index_name)) | |
855 | die("cannot store index file"); | |
856 | } else | |
857 | chmod(final_index_name, 0444); | |
858 | ||
859 | if (!from_stdin) { | |
860 | printf("%s\n", sha1_to_hex(sha1)); | |
861 | } else { | |
862 | char buf[48]; | |
863 | int len = snprintf(buf, sizeof(buf), "%s\t%s\n", | |
864 | report, sha1_to_hex(sha1)); | |
865 | write_or_die(1, buf, len); | |
866 | ||
867 | /* | |
868 | * Let's just mimic git-unpack-objects here and write | |
869 | * the last part of the input buffer to stdout. | |
870 | */ | |
871 | while (input_len) { | |
872 | err = xwrite(1, input_buffer + input_offset, input_len); | |
873 | if (err <= 0) | |
874 | break; | |
875 | input_len -= err; | |
876 | input_offset += err; | |
877 | } | |
878 | } | |
879 | } | |
880 | ||
881 | static int git_index_pack_config(const char *k, const char *v, void *cb) | |
882 | { | |
883 | struct pack_idx_option *opts = cb; | |
884 | ||
885 | if (!strcmp(k, "pack.indexversion")) { | |
886 | opts->version = git_config_int(k, v); | |
887 | if (opts->version > 2) | |
888 | die("bad pack.indexversion=%"PRIu32, opts->version); | |
889 | return 0; | |
890 | } | |
891 | return git_default_config(k, v, cb); | |
892 | } | |
893 | ||
894 | int cmd_index_pack(int argc, const char **argv, const char *prefix) | |
895 | { | |
896 | int i, fix_thin_pack = 0; | |
897 | const char *curr_pack, *curr_index; | |
898 | const char *index_name = NULL, *pack_name = NULL; | |
899 | const char *keep_name = NULL, *keep_msg = NULL; | |
900 | char *index_name_buf = NULL, *keep_name_buf = NULL; | |
901 | struct pack_idx_entry **idx_objects; | |
902 | struct pack_idx_option opts; | |
903 | unsigned char pack_sha1[20]; | |
904 | ||
905 | if (argc == 2 && !strcmp(argv[1], "-h")) | |
906 | usage(index_pack_usage); | |
907 | ||
908 | read_replace_refs = 0; | |
909 | ||
910 | reset_pack_idx_option(&opts); | |
911 | git_config(git_index_pack_config, &opts); | |
912 | if (prefix && chdir(prefix)) | |
913 | die("Cannot come back to cwd"); | |
914 | ||
915 | for (i = 1; i < argc; i++) { | |
916 | const char *arg = argv[i]; | |
917 | ||
918 | if (*arg == '-') { | |
919 | if (!strcmp(arg, "--stdin")) { | |
920 | from_stdin = 1; | |
921 | } else if (!strcmp(arg, "--fix-thin")) { | |
922 | fix_thin_pack = 1; | |
923 | } else if (!strcmp(arg, "--strict")) { | |
924 | strict = 1; | |
925 | } else if (!strcmp(arg, "--keep")) { | |
926 | keep_msg = ""; | |
927 | } else if (!prefixcmp(arg, "--keep=")) { | |
928 | keep_msg = arg + 7; | |
929 | } else if (!prefixcmp(arg, "--pack_header=")) { | |
930 | struct pack_header *hdr; | |
931 | char *c; | |
932 | ||
933 | hdr = (struct pack_header *)input_buffer; | |
934 | hdr->hdr_signature = htonl(PACK_SIGNATURE); | |
935 | hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10)); | |
936 | if (*c != ',') | |
937 | die("bad %s", arg); | |
938 | hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10)); | |
939 | if (*c) | |
940 | die("bad %s", arg); | |
941 | input_len = sizeof(*hdr); | |
942 | } else if (!strcmp(arg, "-v")) { | |
943 | verbose = 1; | |
944 | } else if (!strcmp(arg, "-o")) { | |
945 | if (index_name || (i+1) >= argc) | |
946 | usage(index_pack_usage); | |
947 | index_name = argv[++i]; | |
948 | } else if (!prefixcmp(arg, "--index-version=")) { | |
949 | char *c; | |
950 | opts.version = strtoul(arg + 16, &c, 10); | |
951 | if (opts.version > 2) | |
952 | die("bad %s", arg); | |
953 | if (*c == ',') | |
954 | opts.off32_limit = strtoul(c+1, &c, 0); | |
955 | if (*c || opts.off32_limit & 0x80000000) | |
956 | die("bad %s", arg); | |
957 | } else | |
958 | usage(index_pack_usage); | |
959 | continue; | |
960 | } | |
961 | ||
962 | if (pack_name) | |
963 | usage(index_pack_usage); | |
964 | pack_name = arg; | |
965 | } | |
966 | ||
967 | if (!pack_name && !from_stdin) | |
968 | usage(index_pack_usage); | |
969 | if (fix_thin_pack && !from_stdin) | |
970 | die("--fix-thin cannot be used without --stdin"); | |
971 | if (!index_name && pack_name) { | |
972 | int len = strlen(pack_name); | |
973 | if (!has_extension(pack_name, ".pack")) | |
974 | die("packfile name '%s' does not end with '.pack'", | |
975 | pack_name); | |
976 | index_name_buf = xmalloc(len); | |
977 | memcpy(index_name_buf, pack_name, len - 5); | |
978 | strcpy(index_name_buf + len - 5, ".idx"); | |
979 | index_name = index_name_buf; | |
980 | } | |
981 | if (keep_msg && !keep_name && pack_name) { | |
982 | int len = strlen(pack_name); | |
983 | if (!has_extension(pack_name, ".pack")) | |
984 | die("packfile name '%s' does not end with '.pack'", | |
985 | pack_name); | |
986 | keep_name_buf = xmalloc(len); | |
987 | memcpy(keep_name_buf, pack_name, len - 5); | |
988 | strcpy(keep_name_buf + len - 5, ".keep"); | |
989 | keep_name = keep_name_buf; | |
990 | } | |
991 | ||
992 | curr_pack = open_pack_file(pack_name); | |
993 | parse_pack_header(); | |
994 | objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry)); | |
995 | deltas = xmalloc(nr_objects * sizeof(struct delta_entry)); | |
996 | parse_pack_objects(pack_sha1); | |
997 | if (nr_deltas == nr_resolved_deltas) { | |
998 | stop_progress(&progress); | |
999 | /* Flush remaining pack final 20-byte SHA1. */ | |
1000 | flush(); | |
1001 | } else { | |
1002 | if (fix_thin_pack) { | |
1003 | struct sha1file *f; | |
1004 | unsigned char read_sha1[20], tail_sha1[20]; | |
1005 | char msg[48]; | |
1006 | int nr_unresolved = nr_deltas - nr_resolved_deltas; | |
1007 | int nr_objects_initial = nr_objects; | |
1008 | if (nr_unresolved <= 0) | |
1009 | die("confusion beyond insanity"); | |
1010 | objects = xrealloc(objects, | |
1011 | (nr_objects + nr_unresolved + 1) | |
1012 | * sizeof(*objects)); | |
1013 | f = sha1fd(output_fd, curr_pack); | |
1014 | fix_unresolved_deltas(f, nr_unresolved); | |
1015 | sprintf(msg, "completed with %d local objects", | |
1016 | nr_objects - nr_objects_initial); | |
1017 | stop_progress_msg(&progress, msg); | |
1018 | sha1close(f, tail_sha1, 0); | |
1019 | hashcpy(read_sha1, pack_sha1); | |
1020 | fixup_pack_header_footer(output_fd, pack_sha1, | |
1021 | curr_pack, nr_objects, | |
1022 | read_sha1, consumed_bytes-20); | |
1023 | if (hashcmp(read_sha1, tail_sha1) != 0) | |
1024 | die("Unexpected tail checksum for %s " | |
1025 | "(disk corruption?)", curr_pack); | |
1026 | } | |
1027 | if (nr_deltas != nr_resolved_deltas) | |
1028 | die("pack has %d unresolved deltas", | |
1029 | nr_deltas - nr_resolved_deltas); | |
1030 | } | |
1031 | free(deltas); | |
1032 | if (strict) | |
1033 | check_objects(); | |
1034 | ||
1035 | idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *)); | |
1036 | for (i = 0; i < nr_objects; i++) | |
1037 | idx_objects[i] = &objects[i].idx; | |
1038 | curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1); | |
1039 | free(idx_objects); | |
1040 | ||
1041 | final(pack_name, curr_pack, | |
1042 | index_name, curr_index, | |
1043 | keep_name, keep_msg, | |
1044 | pack_sha1); | |
1045 | free(objects); | |
1046 | free(index_name_buf); | |
1047 | free(keep_name_buf); | |
1048 | if (pack_name == NULL) | |
1049 | free((void *) curr_pack); | |
1050 | if (index_name == NULL) | |
1051 | free((void *) curr_index); | |
1052 | ||
1053 | return 0; | |
1054 | } |