cf585fcc5eee5e41aac066a08d6db080f33011b8
[git/git.git] / builtin / unpack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "config.h"
4 #include "object-store.h"
5 #include "object.h"
6 #include "delta.h"
7 #include "pack.h"
8 #include "blob.h"
9 #include "commit.h"
10 #include "tag.h"
11 #include "tree.h"
12 #include "tree-walk.h"
13 #include "progress.h"
14 #include "decorate.h"
15 #include "fsck.h"
16
17 static int dry_run, quiet, recover, has_errors, strict;
18 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict]";
19
20 /* We always read in 4kB chunks. */
21 static unsigned char buffer[4096];
22 static unsigned int offset, len;
23 static off_t consumed_bytes;
24 static off_t max_input_size;
25 static git_hash_ctx ctx;
26 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
27
28 /*
29 * When running under --strict mode, objects whose reachability are
30 * suspect are kept in core without getting written in the object
31 * store.
32 */
33 struct obj_buffer {
34 char *buffer;
35 unsigned long size;
36 };
37
38 static struct decoration obj_decorate;
39
40 static struct obj_buffer *lookup_object_buffer(struct object *base)
41 {
42 return lookup_decoration(&obj_decorate, base);
43 }
44
45 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
46 {
47 struct obj_buffer *obj;
48 obj = xcalloc(1, sizeof(struct obj_buffer));
49 obj->buffer = buffer;
50 obj->size = size;
51 if (add_decoration(&obj_decorate, object, obj))
52 die("object %s tried to add buffer twice!", oid_to_hex(&object->oid));
53 }
54
55 /*
56 * Make sure at least "min" bytes are available in the buffer, and
57 * return the pointer to the buffer.
58 */
59 static void *fill(int min)
60 {
61 if (min <= len)
62 return buffer + offset;
63 if (min > sizeof(buffer))
64 die("cannot fill %d bytes", min);
65 if (offset) {
66 the_hash_algo->update_fn(&ctx, buffer, offset);
67 memmove(buffer, buffer + offset, len);
68 offset = 0;
69 }
70 do {
71 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
72 if (ret <= 0) {
73 if (!ret)
74 die("early EOF");
75 die_errno("read error on input");
76 }
77 len += ret;
78 } while (len < min);
79 return buffer;
80 }
81
82 static void use(int bytes)
83 {
84 if (bytes > len)
85 die("used more bytes than were available");
86 len -= bytes;
87 offset += bytes;
88
89 /* make sure off_t is sufficiently large not to wrap */
90 if (signed_add_overflows(consumed_bytes, bytes))
91 die("pack too large for current definition of off_t");
92 consumed_bytes += bytes;
93 if (max_input_size && consumed_bytes > max_input_size)
94 die(_("pack exceeds maximum allowed size"));
95 }
96
97 static void *get_data(unsigned long size)
98 {
99 git_zstream stream;
100 void *buf = xmallocz(size);
101
102 memset(&stream, 0, sizeof(stream));
103
104 stream.next_out = buf;
105 stream.avail_out = size;
106 stream.next_in = fill(1);
107 stream.avail_in = len;
108 git_inflate_init(&stream);
109
110 for (;;) {
111 int ret = git_inflate(&stream, 0);
112 use(len - stream.avail_in);
113 if (stream.total_out == size && ret == Z_STREAM_END)
114 break;
115 if (ret != Z_OK) {
116 error("inflate returned %d", ret);
117 FREE_AND_NULL(buf);
118 if (!recover)
119 exit(1);
120 has_errors = 1;
121 break;
122 }
123 stream.next_in = fill(1);
124 stream.avail_in = len;
125 }
126 git_inflate_end(&stream);
127 return buf;
128 }
129
130 struct delta_info {
131 struct object_id base_oid;
132 unsigned nr;
133 off_t base_offset;
134 unsigned long size;
135 void *delta;
136 struct delta_info *next;
137 };
138
139 static struct delta_info *delta_list;
140
141 static void add_delta_to_list(unsigned nr, const struct object_id *base_oid,
142 off_t base_offset,
143 void *delta, unsigned long size)
144 {
145 struct delta_info *info = xmalloc(sizeof(*info));
146
147 oidcpy(&info->base_oid, base_oid);
148 info->base_offset = base_offset;
149 info->size = size;
150 info->delta = delta;
151 info->nr = nr;
152 info->next = delta_list;
153 delta_list = info;
154 }
155
156 struct obj_info {
157 off_t offset;
158 struct object_id oid;
159 struct object *obj;
160 };
161
162 /* Remember to update object flag allocation in object.h */
163 #define FLAG_OPEN (1u<<20)
164 #define FLAG_WRITTEN (1u<<21)
165
166 static struct obj_info *obj_list;
167 static unsigned nr_objects;
168
169 /*
170 * Called only from check_object() after it verified this object
171 * is Ok.
172 */
173 static void write_cached_object(struct object *obj, struct obj_buffer *obj_buf)
174 {
175 struct object_id oid;
176
177 if (write_object_file(obj_buf->buffer, obj_buf->size,
178 type_name(obj->type), &oid) < 0)
179 die("failed to write object %s", oid_to_hex(&obj->oid));
180 obj->flags |= FLAG_WRITTEN;
181 }
182
183 /*
184 * At the very end of the processing, write_rest() scans the objects
185 * that have reachability requirements and calls this function.
186 * Verify its reachability and validity recursively and write it out.
187 */
188 static int check_object(struct object *obj, int type, void *data, struct fsck_options *options)
189 {
190 struct obj_buffer *obj_buf;
191
192 if (!obj)
193 return 1;
194
195 if (obj->flags & FLAG_WRITTEN)
196 return 0;
197
198 if (type != OBJ_ANY && obj->type != type)
199 die("object type mismatch");
200
201 if (!(obj->flags & FLAG_OPEN)) {
202 unsigned long size;
203 int type = oid_object_info(the_repository, &obj->oid, &size);
204 if (type != obj->type || type <= 0)
205 die("object of unexpected type");
206 obj->flags |= FLAG_WRITTEN;
207 return 0;
208 }
209
210 obj_buf = lookup_object_buffer(obj);
211 if (!obj_buf)
212 die("Whoops! Cannot find object '%s'", oid_to_hex(&obj->oid));
213 if (fsck_object(obj, obj_buf->buffer, obj_buf->size, &fsck_options))
214 die("fsck error in packed object");
215 fsck_options.walk = check_object;
216 if (fsck_walk(obj, NULL, &fsck_options))
217 die("Error on reachable objects of %s", oid_to_hex(&obj->oid));
218 write_cached_object(obj, obj_buf);
219 return 0;
220 }
221
222 static void write_rest(void)
223 {
224 unsigned i;
225 for (i = 0; i < nr_objects; i++) {
226 if (obj_list[i].obj)
227 check_object(obj_list[i].obj, OBJ_ANY, NULL, NULL);
228 }
229 }
230
231 static void added_object(unsigned nr, enum object_type type,
232 void *data, unsigned long size);
233
234 /*
235 * Write out nr-th object from the list, now we know the contents
236 * of it. Under --strict, this buffers structured objects in-core,
237 * to be checked at the end.
238 */
239 static void write_object(unsigned nr, enum object_type type,
240 void *buf, unsigned long size)
241 {
242 if (!strict) {
243 if (write_object_file(buf, size, type_name(type),
244 &obj_list[nr].oid) < 0)
245 die("failed to write object");
246 added_object(nr, type, buf, size);
247 free(buf);
248 obj_list[nr].obj = NULL;
249 } else if (type == OBJ_BLOB) {
250 struct blob *blob;
251 if (write_object_file(buf, size, type_name(type),
252 &obj_list[nr].oid) < 0)
253 die("failed to write object");
254 added_object(nr, type, buf, size);
255 free(buf);
256
257 blob = lookup_blob(&obj_list[nr].oid);
258 if (blob)
259 blob->object.flags |= FLAG_WRITTEN;
260 else
261 die("invalid blob object");
262 obj_list[nr].obj = NULL;
263 } else {
264 struct object *obj;
265 int eaten;
266 hash_object_file(buf, size, type_name(type), &obj_list[nr].oid);
267 added_object(nr, type, buf, size);
268 obj = parse_object_buffer(&obj_list[nr].oid, type, size, buf,
269 &eaten);
270 if (!obj)
271 die("invalid %s", type_name(type));
272 add_object_buffer(obj, buf, size);
273 obj->flags |= FLAG_OPEN;
274 obj_list[nr].obj = obj;
275 }
276 }
277
278 static void resolve_delta(unsigned nr, enum object_type type,
279 void *base, unsigned long base_size,
280 void *delta, unsigned long delta_size)
281 {
282 void *result;
283 unsigned long result_size;
284
285 result = patch_delta(base, base_size,
286 delta, delta_size,
287 &result_size);
288 if (!result)
289 die("failed to apply delta");
290 free(delta);
291 write_object(nr, type, result, result_size);
292 }
293
294 /*
295 * We now know the contents of an object (which is nr-th in the pack);
296 * resolve all the deltified objects that are based on it.
297 */
298 static void added_object(unsigned nr, enum object_type type,
299 void *data, unsigned long size)
300 {
301 struct delta_info **p = &delta_list;
302 struct delta_info *info;
303
304 while ((info = *p) != NULL) {
305 if (!oidcmp(&info->base_oid, &obj_list[nr].oid) ||
306 info->base_offset == obj_list[nr].offset) {
307 *p = info->next;
308 p = &delta_list;
309 resolve_delta(info->nr, type, data, size,
310 info->delta, info->size);
311 free(info);
312 continue;
313 }
314 p = &info->next;
315 }
316 }
317
318 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
319 unsigned nr)
320 {
321 void *buf = get_data(size);
322
323 if (!dry_run && buf)
324 write_object(nr, type, buf, size);
325 else
326 free(buf);
327 }
328
329 static int resolve_against_held(unsigned nr, const struct object_id *base,
330 void *delta_data, unsigned long delta_size)
331 {
332 struct object *obj;
333 struct obj_buffer *obj_buffer;
334 obj = lookup_object(base->hash);
335 if (!obj)
336 return 0;
337 obj_buffer = lookup_object_buffer(obj);
338 if (!obj_buffer)
339 return 0;
340 resolve_delta(nr, obj->type, obj_buffer->buffer,
341 obj_buffer->size, delta_data, delta_size);
342 return 1;
343 }
344
345 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
346 unsigned nr)
347 {
348 void *delta_data, *base;
349 unsigned long base_size;
350 struct object_id base_oid;
351
352 if (type == OBJ_REF_DELTA) {
353 hashcpy(base_oid.hash, fill(the_hash_algo->rawsz));
354 use(the_hash_algo->rawsz);
355 delta_data = get_data(delta_size);
356 if (dry_run || !delta_data) {
357 free(delta_data);
358 return;
359 }
360 if (has_object_file(&base_oid))
361 ; /* Ok we have this one */
362 else if (resolve_against_held(nr, &base_oid,
363 delta_data, delta_size))
364 return; /* we are done */
365 else {
366 /* cannot resolve yet --- queue it */
367 oidclr(&obj_list[nr].oid);
368 add_delta_to_list(nr, &base_oid, 0, delta_data, delta_size);
369 return;
370 }
371 } else {
372 unsigned base_found = 0;
373 unsigned char *pack, c;
374 off_t base_offset;
375 unsigned lo, mid, hi;
376
377 pack = fill(1);
378 c = *pack;
379 use(1);
380 base_offset = c & 127;
381 while (c & 128) {
382 base_offset += 1;
383 if (!base_offset || MSB(base_offset, 7))
384 die("offset value overflow for delta base object");
385 pack = fill(1);
386 c = *pack;
387 use(1);
388 base_offset = (base_offset << 7) + (c & 127);
389 }
390 base_offset = obj_list[nr].offset - base_offset;
391 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
392 die("offset value out of bound for delta base object");
393
394 delta_data = get_data(delta_size);
395 if (dry_run || !delta_data) {
396 free(delta_data);
397 return;
398 }
399 lo = 0;
400 hi = nr;
401 while (lo < hi) {
402 mid = lo + (hi - lo) / 2;
403 if (base_offset < obj_list[mid].offset) {
404 hi = mid;
405 } else if (base_offset > obj_list[mid].offset) {
406 lo = mid + 1;
407 } else {
408 oidcpy(&base_oid, &obj_list[mid].oid);
409 base_found = !is_null_oid(&base_oid);
410 break;
411 }
412 }
413 if (!base_found) {
414 /*
415 * The delta base object is itself a delta that
416 * has not been resolved yet.
417 */
418 oidclr(&obj_list[nr].oid);
419 add_delta_to_list(nr, &null_oid, base_offset, delta_data, delta_size);
420 return;
421 }
422 }
423
424 if (resolve_against_held(nr, &base_oid, delta_data, delta_size))
425 return;
426
427 base = read_object_file(&base_oid, &type, &base_size);
428 if (!base) {
429 error("failed to read delta-pack base object %s",
430 oid_to_hex(&base_oid));
431 if (!recover)
432 exit(1);
433 has_errors = 1;
434 return;
435 }
436 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
437 free(base);
438 }
439
440 static void unpack_one(unsigned nr)
441 {
442 unsigned shift;
443 unsigned char *pack;
444 unsigned long size, c;
445 enum object_type type;
446
447 obj_list[nr].offset = consumed_bytes;
448
449 pack = fill(1);
450 c = *pack;
451 use(1);
452 type = (c >> 4) & 7;
453 size = (c & 15);
454 shift = 4;
455 while (c & 0x80) {
456 pack = fill(1);
457 c = *pack;
458 use(1);
459 size += (c & 0x7f) << shift;
460 shift += 7;
461 }
462
463 switch (type) {
464 case OBJ_COMMIT:
465 case OBJ_TREE:
466 case OBJ_BLOB:
467 case OBJ_TAG:
468 unpack_non_delta_entry(type, size, nr);
469 return;
470 case OBJ_REF_DELTA:
471 case OBJ_OFS_DELTA:
472 unpack_delta_entry(type, size, nr);
473 return;
474 default:
475 error("bad object type %d", type);
476 has_errors = 1;
477 if (recover)
478 return;
479 exit(1);
480 }
481 }
482
483 static void unpack_all(void)
484 {
485 int i;
486 struct progress *progress = NULL;
487 struct pack_header *hdr = fill(sizeof(struct pack_header));
488
489 nr_objects = ntohl(hdr->hdr_entries);
490
491 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
492 die("bad pack file");
493 if (!pack_version_ok(hdr->hdr_version))
494 die("unknown pack file version %"PRIu32,
495 ntohl(hdr->hdr_version));
496 use(sizeof(struct pack_header));
497
498 if (!quiet)
499 progress = start_progress(_("Unpacking objects"), nr_objects);
500 obj_list = xcalloc(nr_objects, sizeof(*obj_list));
501 for (i = 0; i < nr_objects; i++) {
502 unpack_one(i);
503 display_progress(progress, i + 1);
504 }
505 stop_progress(&progress);
506
507 if (delta_list)
508 die("unresolved deltas left after unpacking");
509 }
510
511 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
512 {
513 int i;
514 struct object_id oid;
515
516 check_replace_refs = 0;
517
518 git_config(git_default_config, NULL);
519
520 quiet = !isatty(2);
521
522 for (i = 1 ; i < argc; i++) {
523 const char *arg = argv[i];
524
525 if (*arg == '-') {
526 if (!strcmp(arg, "-n")) {
527 dry_run = 1;
528 continue;
529 }
530 if (!strcmp(arg, "-q")) {
531 quiet = 1;
532 continue;
533 }
534 if (!strcmp(arg, "-r")) {
535 recover = 1;
536 continue;
537 }
538 if (!strcmp(arg, "--strict")) {
539 strict = 1;
540 continue;
541 }
542 if (skip_prefix(arg, "--strict=", &arg)) {
543 strict = 1;
544 fsck_set_msg_types(&fsck_options, arg);
545 continue;
546 }
547 if (starts_with(arg, "--pack_header=")) {
548 struct pack_header *hdr;
549 char *c;
550
551 hdr = (struct pack_header *)buffer;
552 hdr->hdr_signature = htonl(PACK_SIGNATURE);
553 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
554 if (*c != ',')
555 die("bad %s", arg);
556 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
557 if (*c)
558 die("bad %s", arg);
559 len = sizeof(*hdr);
560 continue;
561 }
562 if (skip_prefix(arg, "--max-input-size=", &arg)) {
563 max_input_size = strtoumax(arg, NULL, 10);
564 continue;
565 }
566 usage(unpack_usage);
567 }
568
569 /* We don't take any non-flag arguments now.. Maybe some day */
570 usage(unpack_usage);
571 }
572 the_hash_algo->init_fn(&ctx);
573 unpack_all();
574 the_hash_algo->update_fn(&ctx, buffer, offset);
575 the_hash_algo->final_fn(oid.hash, &ctx);
576 if (strict) {
577 write_rest();
578 if (fsck_finish(&fsck_options))
579 die(_("fsck error in pack objects"));
580 }
581 if (hashcmp(fill(the_hash_algo->rawsz), oid.hash))
582 die("final sha1 did not match");
583 use(the_hash_algo->rawsz);
584
585 /* Write the last part of the buffer to stdout */
586 while (len) {
587 int ret = xwrite(1, buffer + offset, len);
588 if (ret <= 0)
589 break;
590 len -= ret;
591 offset += ret;
592 }
593
594 /* All done */
595 return has_errors;
596 }