]>
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 (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 long rdy = 0; | |
363 | unsigned char *src, *data; | |
364 | z_stream stream; | |
365 | int st; | |
366 | ||
367 | src = xmalloc(len); | |
368 | data = src; | |
369 | do { | |
370 | ssize_t n = pread(pack_fd, data + rdy, len - rdy, from + rdy); | |
371 | if (n < 0) | |
372 | die_errno("cannot pread pack file"); | |
373 | if (!n) | |
374 | die("premature end of pack file, %lu bytes missing", | |
375 | len - rdy); | |
376 | rdy += n; | |
377 | } while (rdy < len); | |
378 | data = xmalloc(obj->size); | |
379 | memset(&stream, 0, sizeof(stream)); | |
380 | stream.next_out = data; | |
381 | stream.avail_out = obj->size; | |
382 | stream.next_in = src; | |
383 | stream.avail_in = len; | |
384 | git_inflate_init(&stream); | |
385 | while ((st = git_inflate(&stream, Z_FINISH)) == Z_OK); | |
386 | git_inflate_end(&stream); | |
387 | if (st != Z_STREAM_END || stream.total_out != obj->size) | |
388 | die("serious inflate inconsistency"); | |
389 | free(src); | |
390 | return data; | |
391 | } | |
392 | ||
393 | static int find_delta(const union delta_base *base) | |
394 | { | |
395 | int first = 0, last = nr_deltas; | |
396 | ||
397 | while (first < last) { | |
398 | int next = (first + last) / 2; | |
399 | struct delta_entry *delta = &deltas[next]; | |
400 | int cmp; | |
401 | ||
402 | cmp = memcmp(base, &delta->base, UNION_BASE_SZ); | |
403 | if (!cmp) | |
404 | return next; | |
405 | if (cmp < 0) { | |
406 | last = next; | |
407 | continue; | |
408 | } | |
409 | first = next+1; | |
410 | } | |
411 | return -first-1; | |
412 | } | |
413 | ||
414 | static void find_delta_children(const union delta_base *base, | |
415 | int *first_index, int *last_index) | |
416 | { | |
417 | int first = find_delta(base); | |
418 | int last = first; | |
419 | int end = nr_deltas - 1; | |
420 | ||
421 | if (first < 0) { | |
422 | *first_index = 0; | |
423 | *last_index = -1; | |
424 | return; | |
425 | } | |
426 | while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ)) | |
427 | --first; | |
428 | while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ)) | |
429 | ++last; | |
430 | *first_index = first; | |
431 | *last_index = last; | |
432 | } | |
433 | ||
434 | static void sha1_object(const void *data, unsigned long size, | |
435 | enum object_type type, unsigned char *sha1) | |
436 | { | |
437 | hash_sha1_file(data, size, typename(type), sha1); | |
438 | if (has_sha1_file(sha1)) { | |
439 | void *has_data; | |
440 | enum object_type has_type; | |
441 | unsigned long has_size; | |
442 | has_data = read_sha1_file(sha1, &has_type, &has_size); | |
443 | if (!has_data) | |
444 | die("cannot read existing object %s", sha1_to_hex(sha1)); | |
445 | if (size != has_size || type != has_type || | |
446 | memcmp(data, has_data, size) != 0) | |
447 | die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1)); | |
448 | free(has_data); | |
449 | } | |
450 | if (strict) { | |
451 | if (type == OBJ_BLOB) { | |
452 | struct blob *blob = lookup_blob(sha1); | |
453 | if (blob) | |
454 | blob->object.flags |= FLAG_CHECKED; | |
455 | else | |
456 | die("invalid blob object %s", sha1_to_hex(sha1)); | |
457 | } else { | |
458 | struct object *obj; | |
459 | int eaten; | |
460 | void *buf = (void *) data; | |
461 | ||
462 | /* | |
463 | * we do not need to free the memory here, as the | |
464 | * buf is deleted by the caller. | |
465 | */ | |
466 | obj = parse_object_buffer(sha1, type, size, buf, &eaten); | |
467 | if (!obj) | |
468 | die("invalid %s", typename(type)); | |
469 | if (fsck_object(obj, 1, fsck_error_function)) | |
470 | die("Error in object"); | |
471 | if (fsck_walk(obj, mark_link, NULL)) | |
472 | die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1)); | |
473 | ||
474 | if (obj->type == OBJ_TREE) { | |
475 | struct tree *item = (struct tree *) obj; | |
476 | item->buffer = NULL; | |
477 | } | |
478 | if (obj->type == OBJ_COMMIT) { | |
479 | struct commit *commit = (struct commit *) obj; | |
480 | commit->buffer = NULL; | |
481 | } | |
482 | obj->flags |= FLAG_CHECKED; | |
483 | } | |
484 | } | |
485 | } | |
486 | ||
487 | static void *get_base_data(struct base_data *c) | |
488 | { | |
489 | if (!c->data) { | |
490 | struct object_entry *obj = c->obj; | |
491 | ||
492 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) { | |
493 | void *base = get_base_data(c->base); | |
494 | void *raw = get_data_from_pack(obj); | |
495 | c->data = patch_delta( | |
496 | base, c->base->size, | |
497 | raw, obj->size, | |
498 | &c->size); | |
499 | free(raw); | |
500 | if (!c->data) | |
501 | bad_object(obj->idx.offset, "failed to apply delta"); | |
502 | } else { | |
503 | c->data = get_data_from_pack(obj); | |
504 | c->size = obj->size; | |
505 | } | |
506 | ||
507 | base_cache_used += c->size; | |
508 | prune_base_data(c); | |
509 | } | |
510 | return c->data; | |
511 | } | |
512 | ||
513 | static void resolve_delta(struct object_entry *delta_obj, | |
514 | struct base_data *base, struct base_data *result) | |
515 | { | |
516 | void *base_data, *delta_data; | |
517 | ||
518 | delta_obj->real_type = base->obj->real_type; | |
519 | delta_data = get_data_from_pack(delta_obj); | |
520 | base_data = get_base_data(base); | |
521 | result->obj = delta_obj; | |
522 | result->data = patch_delta(base_data, base->size, | |
523 | delta_data, delta_obj->size, &result->size); | |
524 | free(delta_data); | |
525 | if (!result->data) | |
526 | bad_object(delta_obj->idx.offset, "failed to apply delta"); | |
527 | sha1_object(result->data, result->size, delta_obj->real_type, | |
528 | delta_obj->idx.sha1); | |
529 | nr_resolved_deltas++; | |
530 | } | |
531 | ||
532 | static void find_unresolved_deltas(struct base_data *base, | |
533 | struct base_data *prev_base) | |
534 | { | |
535 | int i, ref_first, ref_last, ofs_first, ofs_last; | |
536 | ||
537 | /* | |
538 | * This is a recursive function. Those brackets should help reducing | |
539 | * stack usage by limiting the scope of the delta_base union. | |
540 | */ | |
541 | { | |
542 | union delta_base base_spec; | |
543 | ||
544 | hashcpy(base_spec.sha1, base->obj->idx.sha1); | |
545 | find_delta_children(&base_spec, &ref_first, &ref_last); | |
546 | ||
547 | memset(&base_spec, 0, sizeof(base_spec)); | |
548 | base_spec.offset = base->obj->idx.offset; | |
549 | find_delta_children(&base_spec, &ofs_first, &ofs_last); | |
550 | } | |
551 | ||
552 | if (ref_last == -1 && ofs_last == -1) { | |
553 | free(base->data); | |
554 | return; | |
555 | } | |
556 | ||
557 | link_base_data(prev_base, base); | |
558 | ||
559 | for (i = ref_first; i <= ref_last; i++) { | |
560 | struct object_entry *child = objects + deltas[i].obj_no; | |
561 | if (child->real_type == OBJ_REF_DELTA) { | |
562 | struct base_data result; | |
563 | resolve_delta(child, base, &result); | |
564 | if (i == ref_last && ofs_last == -1) | |
565 | free_base_data(base); | |
566 | find_unresolved_deltas(&result, base); | |
567 | } | |
568 | } | |
569 | ||
570 | for (i = ofs_first; i <= ofs_last; i++) { | |
571 | struct object_entry *child = objects + deltas[i].obj_no; | |
572 | if (child->real_type == OBJ_OFS_DELTA) { | |
573 | struct base_data result; | |
574 | resolve_delta(child, base, &result); | |
575 | if (i == ofs_last) | |
576 | free_base_data(base); | |
577 | find_unresolved_deltas(&result, base); | |
578 | } | |
579 | } | |
580 | ||
581 | unlink_base_data(base); | |
582 | } | |
583 | ||
584 | static int compare_delta_entry(const void *a, const void *b) | |
585 | { | |
586 | const struct delta_entry *delta_a = a; | |
587 | const struct delta_entry *delta_b = b; | |
588 | return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ); | |
589 | } | |
590 | ||
591 | /* Parse all objects and return the pack content SHA1 hash */ | |
592 | static void parse_pack_objects(unsigned char *sha1) | |
593 | { | |
594 | int i; | |
595 | struct delta_entry *delta = deltas; | |
596 | struct stat st; | |
597 | ||
598 | /* | |
599 | * First pass: | |
600 | * - find locations of all objects; | |
601 | * - calculate SHA1 of all non-delta objects; | |
602 | * - remember base (SHA1 or offset) for all deltas. | |
603 | */ | |
604 | if (verbose) | |
605 | progress = start_progress( | |
606 | from_stdin ? "Receiving objects" : "Indexing objects", | |
607 | nr_objects); | |
608 | for (i = 0; i < nr_objects; i++) { | |
609 | struct object_entry *obj = &objects[i]; | |
610 | void *data = unpack_raw_entry(obj, &delta->base); | |
611 | obj->real_type = obj->type; | |
612 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) { | |
613 | nr_deltas++; | |
614 | delta->obj_no = i; | |
615 | delta++; | |
616 | } else | |
617 | sha1_object(data, obj->size, obj->type, obj->idx.sha1); | |
618 | free(data); | |
619 | display_progress(progress, i+1); | |
620 | } | |
621 | objects[i].idx.offset = consumed_bytes; | |
622 | stop_progress(&progress); | |
623 | ||
624 | /* Check pack integrity */ | |
625 | flush(); | |
626 | git_SHA1_Final(sha1, &input_ctx); | |
627 | if (hashcmp(fill(20), sha1)) | |
628 | die("pack is corrupted (SHA1 mismatch)"); | |
629 | use(20); | |
630 | ||
631 | /* If input_fd is a file, we should have reached its end now. */ | |
632 | if (fstat(input_fd, &st)) | |
633 | die_errno("cannot fstat packfile"); | |
634 | if (S_ISREG(st.st_mode) && | |
635 | lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size) | |
636 | die("pack has junk at the end"); | |
637 | ||
638 | if (!nr_deltas) | |
639 | return; | |
640 | ||
641 | /* Sort deltas by base SHA1/offset for fast searching */ | |
642 | qsort(deltas, nr_deltas, sizeof(struct delta_entry), | |
643 | compare_delta_entry); | |
644 | ||
645 | /* | |
646 | * Second pass: | |
647 | * - for all non-delta objects, look if it is used as a base for | |
648 | * deltas; | |
649 | * - if used as a base, uncompress the object and apply all deltas, | |
650 | * recursively checking if the resulting object is used as a base | |
651 | * for some more deltas. | |
652 | */ | |
653 | if (verbose) | |
654 | progress = start_progress("Resolving deltas", nr_deltas); | |
655 | for (i = 0; i < nr_objects; i++) { | |
656 | struct object_entry *obj = &objects[i]; | |
657 | struct base_data base_obj; | |
658 | ||
659 | if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) | |
660 | continue; | |
661 | base_obj.obj = obj; | |
662 | base_obj.data = NULL; | |
663 | find_unresolved_deltas(&base_obj, NULL); | |
664 | display_progress(progress, nr_resolved_deltas); | |
665 | } | |
666 | } | |
667 | ||
668 | static int write_compressed(struct sha1file *f, void *in, unsigned int size) | |
669 | { | |
670 | z_stream stream; | |
671 | unsigned long maxsize; | |
672 | void *out; | |
673 | ||
674 | memset(&stream, 0, sizeof(stream)); | |
675 | deflateInit(&stream, zlib_compression_level); | |
676 | maxsize = deflateBound(&stream, size); | |
677 | out = xmalloc(maxsize); | |
678 | ||
679 | /* Compress it */ | |
680 | stream.next_in = in; | |
681 | stream.avail_in = size; | |
682 | stream.next_out = out; | |
683 | stream.avail_out = maxsize; | |
684 | while (deflate(&stream, Z_FINISH) == Z_OK); | |
685 | deflateEnd(&stream); | |
686 | ||
687 | size = stream.total_out; | |
688 | sha1write(f, out, size); | |
689 | free(out); | |
690 | return size; | |
691 | } | |
692 | ||
693 | static struct object_entry *append_obj_to_pack(struct sha1file *f, | |
694 | const unsigned char *sha1, void *buf, | |
695 | unsigned long size, enum object_type type) | |
696 | { | |
697 | struct object_entry *obj = &objects[nr_objects++]; | |
698 | unsigned char header[10]; | |
699 | unsigned long s = size; | |
700 | int n = 0; | |
701 | unsigned char c = (type << 4) | (s & 15); | |
702 | s >>= 4; | |
703 | while (s) { | |
704 | header[n++] = c | 0x80; | |
705 | c = s & 0x7f; | |
706 | s >>= 7; | |
707 | } | |
708 | header[n++] = c; | |
709 | crc32_begin(f); | |
710 | sha1write(f, header, n); | |
711 | obj[0].size = size; | |
712 | obj[0].hdr_size = n; | |
713 | obj[0].type = type; | |
714 | obj[0].real_type = type; | |
715 | obj[1].idx.offset = obj[0].idx.offset + n; | |
716 | obj[1].idx.offset += write_compressed(f, buf, size); | |
717 | obj[0].idx.crc32 = crc32_end(f); | |
718 | sha1flush(f); | |
719 | hashcpy(obj->idx.sha1, sha1); | |
720 | return obj; | |
721 | } | |
722 | ||
723 | static int delta_pos_compare(const void *_a, const void *_b) | |
724 | { | |
725 | struct delta_entry *a = *(struct delta_entry **)_a; | |
726 | struct delta_entry *b = *(struct delta_entry **)_b; | |
727 | return a->obj_no - b->obj_no; | |
728 | } | |
729 | ||
730 | static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved) | |
731 | { | |
732 | struct delta_entry **sorted_by_pos; | |
733 | int i, n = 0; | |
734 | ||
735 | /* | |
736 | * Since many unresolved deltas may well be themselves base objects | |
737 | * for more unresolved deltas, we really want to include the | |
738 | * smallest number of base objects that would cover as much delta | |
739 | * as possible by picking the | |
740 | * trunc deltas first, allowing for other deltas to resolve without | |
741 | * additional base objects. Since most base objects are to be found | |
742 | * before deltas depending on them, a good heuristic is to start | |
743 | * resolving deltas in the same order as their position in the pack. | |
744 | */ | |
745 | sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos)); | |
746 | for (i = 0; i < nr_deltas; i++) { | |
747 | if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA) | |
748 | continue; | |
749 | sorted_by_pos[n++] = &deltas[i]; | |
750 | } | |
751 | qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare); | |
752 | ||
753 | for (i = 0; i < n; i++) { | |
754 | struct delta_entry *d = sorted_by_pos[i]; | |
755 | enum object_type type; | |
756 | struct base_data base_obj; | |
757 | ||
758 | if (objects[d->obj_no].real_type != OBJ_REF_DELTA) | |
759 | continue; | |
760 | base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size); | |
761 | if (!base_obj.data) | |
762 | continue; | |
763 | ||
764 | if (check_sha1_signature(d->base.sha1, base_obj.data, | |
765 | base_obj.size, typename(type))) | |
766 | die("local object %s is corrupt", sha1_to_hex(d->base.sha1)); | |
767 | base_obj.obj = append_obj_to_pack(f, d->base.sha1, | |
768 | base_obj.data, base_obj.size, type); | |
769 | find_unresolved_deltas(&base_obj, NULL); | |
770 | display_progress(progress, nr_resolved_deltas); | |
771 | } | |
772 | free(sorted_by_pos); | |
773 | } | |
774 | ||
775 | static void final(const char *final_pack_name, const char *curr_pack_name, | |
776 | const char *final_index_name, const char *curr_index_name, | |
777 | const char *keep_name, const char *keep_msg, | |
778 | unsigned char *sha1) | |
779 | { | |
780 | const char *report = "pack"; | |
781 | char name[PATH_MAX]; | |
782 | int err; | |
783 | ||
784 | if (!from_stdin) { | |
785 | close(input_fd); | |
786 | } else { | |
787 | fsync_or_die(output_fd, curr_pack_name); | |
788 | err = close(output_fd); | |
789 | if (err) | |
790 | die_errno("error while closing pack file"); | |
791 | } | |
792 | ||
793 | if (keep_msg) { | |
794 | int keep_fd, keep_msg_len = strlen(keep_msg); | |
795 | ||
796 | if (!keep_name) | |
797 | keep_fd = odb_pack_keep(name, sizeof(name), sha1); | |
798 | else | |
799 | keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600); | |
800 | ||
801 | if (keep_fd < 0) { | |
802 | if (errno != EEXIST) | |
803 | die_errno("cannot write keep file '%s'", | |
804 | keep_name); | |
805 | } else { | |
806 | if (keep_msg_len > 0) { | |
807 | write_or_die(keep_fd, keep_msg, keep_msg_len); | |
808 | write_or_die(keep_fd, "\n", 1); | |
809 | } | |
810 | if (close(keep_fd) != 0) | |
811 | die_errno("cannot close written keep file '%s'", | |
812 | keep_name); | |
813 | report = "keep"; | |
814 | } | |
815 | } | |
816 | ||
817 | if (final_pack_name != curr_pack_name) { | |
818 | if (!final_pack_name) { | |
819 | snprintf(name, sizeof(name), "%s/pack/pack-%s.pack", | |
820 | get_object_directory(), sha1_to_hex(sha1)); | |
821 | final_pack_name = name; | |
822 | } | |
823 | if (move_temp_to_file(curr_pack_name, final_pack_name)) | |
824 | die("cannot store pack file"); | |
825 | } else if (from_stdin) | |
826 | chmod(final_pack_name, 0444); | |
827 | ||
828 | if (final_index_name != curr_index_name) { | |
829 | if (!final_index_name) { | |
830 | snprintf(name, sizeof(name), "%s/pack/pack-%s.idx", | |
831 | get_object_directory(), sha1_to_hex(sha1)); | |
832 | final_index_name = name; | |
833 | } | |
834 | if (move_temp_to_file(curr_index_name, final_index_name)) | |
835 | die("cannot store index file"); | |
836 | } else | |
837 | chmod(final_index_name, 0444); | |
838 | ||
839 | if (!from_stdin) { | |
840 | printf("%s\n", sha1_to_hex(sha1)); | |
841 | } else { | |
842 | char buf[48]; | |
843 | int len = snprintf(buf, sizeof(buf), "%s\t%s\n", | |
844 | report, sha1_to_hex(sha1)); | |
845 | write_or_die(1, buf, len); | |
846 | ||
847 | /* | |
848 | * Let's just mimic git-unpack-objects here and write | |
849 | * the last part of the input buffer to stdout. | |
850 | */ | |
851 | while (input_len) { | |
852 | err = xwrite(1, input_buffer + input_offset, input_len); | |
853 | if (err <= 0) | |
854 | break; | |
855 | input_len -= err; | |
856 | input_offset += err; | |
857 | } | |
858 | } | |
859 | } | |
860 | ||
861 | static int git_index_pack_config(const char *k, const char *v, void *cb) | |
862 | { | |
863 | if (!strcmp(k, "pack.indexversion")) { | |
864 | pack_idx_default_version = git_config_int(k, v); | |
865 | if (pack_idx_default_version > 2) | |
866 | die("bad pack.indexversion=%"PRIu32, | |
867 | pack_idx_default_version); | |
868 | return 0; | |
869 | } | |
870 | return git_default_config(k, v, cb); | |
871 | } | |
872 | ||
873 | int cmd_index_pack(int argc, const char **argv, const char *prefix) | |
874 | { | |
875 | int i, fix_thin_pack = 0; | |
876 | const char *curr_pack, *curr_index; | |
877 | const char *index_name = NULL, *pack_name = NULL; | |
878 | const char *keep_name = NULL, *keep_msg = NULL; | |
879 | char *index_name_buf = NULL, *keep_name_buf = NULL; | |
880 | struct pack_idx_entry **idx_objects; | |
881 | unsigned char pack_sha1[20]; | |
882 | ||
883 | if (argc == 2 && !strcmp(argv[1], "-h")) | |
884 | usage(index_pack_usage); | |
885 | ||
886 | /* | |
887 | * We wish to read the repository's config file if any, and | |
888 | * for that it is necessary to call setup_git_directory_gently(). | |
889 | * However if the cwd was inside .git/objects/pack/ then we need | |
890 | * to go back there or all the pack name arguments will be wrong. | |
891 | * And in that case we cannot rely on any prefix returned by | |
892 | * setup_git_directory_gently() either. | |
893 | */ | |
894 | { | |
895 | char cwd[PATH_MAX+1]; | |
896 | int nongit; | |
897 | ||
898 | if (!getcwd(cwd, sizeof(cwd)-1)) | |
899 | die("Unable to get current working directory"); | |
900 | setup_git_directory_gently(&nongit); | |
901 | git_config(git_index_pack_config, NULL); | |
902 | if (chdir(cwd)) | |
903 | die("Cannot come back to cwd"); | |
904 | } | |
905 | ||
906 | for (i = 1; i < argc; i++) { | |
907 | const char *arg = argv[i]; | |
908 | ||
909 | if (*arg == '-') { | |
910 | if (!strcmp(arg, "--stdin")) { | |
911 | from_stdin = 1; | |
912 | } else if (!strcmp(arg, "--fix-thin")) { | |
913 | fix_thin_pack = 1; | |
914 | } else if (!strcmp(arg, "--strict")) { | |
915 | strict = 1; | |
916 | } else if (!strcmp(arg, "--keep")) { | |
917 | keep_msg = ""; | |
918 | } else if (!prefixcmp(arg, "--keep=")) { | |
919 | keep_msg = arg + 7; | |
920 | } else if (!prefixcmp(arg, "--pack_header=")) { | |
921 | struct pack_header *hdr; | |
922 | char *c; | |
923 | ||
924 | hdr = (struct pack_header *)input_buffer; | |
925 | hdr->hdr_signature = htonl(PACK_SIGNATURE); | |
926 | hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10)); | |
927 | if (*c != ',') | |
928 | die("bad %s", arg); | |
929 | hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10)); | |
930 | if (*c) | |
931 | die("bad %s", arg); | |
932 | input_len = sizeof(*hdr); | |
933 | } else if (!strcmp(arg, "-v")) { | |
934 | verbose = 1; | |
935 | } else if (!strcmp(arg, "-o")) { | |
936 | if (index_name || (i+1) >= argc) | |
937 | usage(index_pack_usage); | |
938 | index_name = argv[++i]; | |
939 | } else if (!prefixcmp(arg, "--index-version=")) { | |
940 | char *c; | |
941 | pack_idx_default_version = strtoul(arg + 16, &c, 10); | |
942 | if (pack_idx_default_version > 2) | |
943 | die("bad %s", arg); | |
944 | if (*c == ',') | |
945 | pack_idx_off32_limit = strtoul(c+1, &c, 0); | |
946 | if (*c || pack_idx_off32_limit & 0x80000000) | |
947 | die("bad %s", arg); | |
948 | } else | |
949 | usage(index_pack_usage); | |
950 | continue; | |
951 | } | |
952 | ||
953 | if (pack_name) | |
954 | usage(index_pack_usage); | |
955 | pack_name = arg; | |
956 | } | |
957 | ||
958 | if (!pack_name && !from_stdin) | |
959 | usage(index_pack_usage); | |
960 | if (fix_thin_pack && !from_stdin) | |
961 | die("--fix-thin cannot be used without --stdin"); | |
962 | if (!index_name && pack_name) { | |
963 | int len = strlen(pack_name); | |
964 | if (!has_extension(pack_name, ".pack")) | |
965 | die("packfile name '%s' does not end with '.pack'", | |
966 | pack_name); | |
967 | index_name_buf = xmalloc(len); | |
968 | memcpy(index_name_buf, pack_name, len - 5); | |
969 | strcpy(index_name_buf + len - 5, ".idx"); | |
970 | index_name = index_name_buf; | |
971 | } | |
972 | if (keep_msg && !keep_name && pack_name) { | |
973 | int len = strlen(pack_name); | |
974 | if (!has_extension(pack_name, ".pack")) | |
975 | die("packfile name '%s' does not end with '.pack'", | |
976 | pack_name); | |
977 | keep_name_buf = xmalloc(len); | |
978 | memcpy(keep_name_buf, pack_name, len - 5); | |
979 | strcpy(keep_name_buf + len - 5, ".keep"); | |
980 | keep_name = keep_name_buf; | |
981 | } | |
982 | ||
983 | curr_pack = open_pack_file(pack_name); | |
984 | parse_pack_header(); | |
985 | objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry)); | |
986 | deltas = xmalloc(nr_objects * sizeof(struct delta_entry)); | |
987 | parse_pack_objects(pack_sha1); | |
988 | if (nr_deltas == nr_resolved_deltas) { | |
989 | stop_progress(&progress); | |
990 | /* Flush remaining pack final 20-byte SHA1. */ | |
991 | flush(); | |
992 | } else { | |
993 | if (fix_thin_pack) { | |
994 | struct sha1file *f; | |
995 | unsigned char read_sha1[20], tail_sha1[20]; | |
996 | char msg[48]; | |
997 | int nr_unresolved = nr_deltas - nr_resolved_deltas; | |
998 | int nr_objects_initial = nr_objects; | |
999 | if (nr_unresolved <= 0) | |
1000 | die("confusion beyond insanity"); | |
1001 | objects = xrealloc(objects, | |
1002 | (nr_objects + nr_unresolved + 1) | |
1003 | * sizeof(*objects)); | |
1004 | f = sha1fd(output_fd, curr_pack); | |
1005 | fix_unresolved_deltas(f, nr_unresolved); | |
1006 | sprintf(msg, "completed with %d local objects", | |
1007 | nr_objects - nr_objects_initial); | |
1008 | stop_progress_msg(&progress, msg); | |
1009 | sha1close(f, tail_sha1, 0); | |
1010 | hashcpy(read_sha1, pack_sha1); | |
1011 | fixup_pack_header_footer(output_fd, pack_sha1, | |
1012 | curr_pack, nr_objects, | |
1013 | read_sha1, consumed_bytes-20); | |
1014 | if (hashcmp(read_sha1, tail_sha1) != 0) | |
1015 | die("Unexpected tail checksum for %s " | |
1016 | "(disk corruption?)", curr_pack); | |
1017 | } | |
1018 | if (nr_deltas != nr_resolved_deltas) | |
1019 | die("pack has %d unresolved deltas", | |
1020 | nr_deltas - nr_resolved_deltas); | |
1021 | } | |
1022 | free(deltas); | |
1023 | if (strict) | |
1024 | check_objects(); | |
1025 | ||
1026 | idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *)); | |
1027 | for (i = 0; i < nr_objects; i++) | |
1028 | idx_objects[i] = &objects[i].idx; | |
1029 | curr_index = write_idx_file(index_name, idx_objects, nr_objects, pack_sha1); | |
1030 | free(idx_objects); | |
1031 | ||
1032 | final(pack_name, curr_pack, | |
1033 | index_name, curr_index, | |
1034 | keep_name, keep_msg, | |
1035 | pack_sha1); | |
1036 | free(objects); | |
1037 | free(index_name_buf); | |
1038 | free(keep_name_buf); | |
1039 | if (pack_name == NULL) | |
1040 | free((void *) curr_pack); | |
1041 | if (index_name == NULL) | |
1042 | free((void *) curr_index); | |
1043 | ||
1044 | return 0; | |
1045 | } |