hash: provide per-algorithm null OIDs
[git.git] / builtin / unpack-objects.c
blob4a9466295ba10de5752cca92fa08b1a28461777b
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"
17 static int dry_run, quiet, recover, has_errors, strict;
18 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict]";
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 static struct progress *progress;
30 * When running under --strict mode, objects whose reachability are
31 * suspect are kept in core without getting written in the object
32 * store.
34 struct obj_buffer {
35 char *buffer;
36 unsigned long size;
39 static struct decoration obj_decorate;
41 static struct obj_buffer *lookup_object_buffer(struct object *base)
43 return lookup_decoration(&obj_decorate, base);
46 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
48 struct obj_buffer *obj;
49 CALLOC_ARRAY(obj, 1);
50 obj->buffer = buffer;
51 obj->size = size;
52 if (add_decoration(&obj_decorate, object, obj))
53 die("object %s tried to add buffer twice!", oid_to_hex(&object->oid));
57 * Make sure at least "min" bytes are available in the buffer, and
58 * return the pointer to the buffer.
60 static void *fill(int min)
62 if (min <= len)
63 return buffer + offset;
64 if (min > sizeof(buffer))
65 die("cannot fill %d bytes", min);
66 if (offset) {
67 the_hash_algo->update_fn(&ctx, buffer, offset);
68 memmove(buffer, buffer + offset, len);
69 offset = 0;
71 do {
72 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
73 if (ret <= 0) {
74 if (!ret)
75 die("early EOF");
76 die_errno("read error on input");
78 len += ret;
79 } while (len < min);
80 return buffer;
83 static void use(int bytes)
85 if (bytes > len)
86 die("used more bytes than were available");
87 len -= bytes;
88 offset += bytes;
90 /* make sure off_t is sufficiently large not to wrap */
91 if (signed_add_overflows(consumed_bytes, bytes))
92 die("pack too large for current definition of off_t");
93 consumed_bytes += bytes;
94 if (max_input_size && consumed_bytes > max_input_size)
95 die(_("pack exceeds maximum allowed size"));
96 display_throughput(progress, consumed_bytes);
99 static void *get_data(unsigned long size)
101 git_zstream stream;
102 void *buf = xmallocz(size);
104 memset(&stream, 0, sizeof(stream));
106 stream.next_out = buf;
107 stream.avail_out = size;
108 stream.next_in = fill(1);
109 stream.avail_in = len;
110 git_inflate_init(&stream);
112 for (;;) {
113 int ret = git_inflate(&stream, 0);
114 use(len - stream.avail_in);
115 if (stream.total_out == size && ret == Z_STREAM_END)
116 break;
117 if (ret != Z_OK) {
118 error("inflate returned %d", ret);
119 FREE_AND_NULL(buf);
120 if (!recover)
121 exit(1);
122 has_errors = 1;
123 break;
125 stream.next_in = fill(1);
126 stream.avail_in = len;
128 git_inflate_end(&stream);
129 return buf;
132 struct delta_info {
133 struct object_id base_oid;
134 unsigned nr;
135 off_t base_offset;
136 unsigned long size;
137 void *delta;
138 struct delta_info *next;
141 static struct delta_info *delta_list;
143 static void add_delta_to_list(unsigned nr, const struct object_id *base_oid,
144 off_t base_offset,
145 void *delta, unsigned long size)
147 struct delta_info *info = xmalloc(sizeof(*info));
149 oidcpy(&info->base_oid, base_oid);
150 info->base_offset = base_offset;
151 info->size = size;
152 info->delta = delta;
153 info->nr = nr;
154 info->next = delta_list;
155 delta_list = info;
158 struct obj_info {
159 off_t offset;
160 struct object_id oid;
161 struct object *obj;
164 /* Remember to update object flag allocation in object.h */
165 #define FLAG_OPEN (1u<<20)
166 #define FLAG_WRITTEN (1u<<21)
168 static struct obj_info *obj_list;
169 static unsigned nr_objects;
172 * Called only from check_object() after it verified this object
173 * is Ok.
175 static void write_cached_object(struct object *obj, struct obj_buffer *obj_buf)
177 struct object_id oid;
179 if (write_object_file(obj_buf->buffer, obj_buf->size,
180 type_name(obj->type), &oid) < 0)
181 die("failed to write object %s", oid_to_hex(&obj->oid));
182 obj->flags |= FLAG_WRITTEN;
186 * At the very end of the processing, write_rest() scans the objects
187 * that have reachability requirements and calls this function.
188 * Verify its reachability and validity recursively and write it out.
190 static int check_object(struct object *obj, enum object_type type,
191 void *data, struct fsck_options *options)
193 struct obj_buffer *obj_buf;
195 if (!obj)
196 return 1;
198 if (obj->flags & FLAG_WRITTEN)
199 return 0;
201 if (type != OBJ_ANY && obj->type != type)
202 die("object type mismatch");
204 if (!(obj->flags & FLAG_OPEN)) {
205 unsigned long size;
206 int type = oid_object_info(the_repository, &obj->oid, &size);
207 if (type != obj->type || type <= 0)
208 die("object of unexpected type");
209 obj->flags |= FLAG_WRITTEN;
210 return 0;
213 obj_buf = lookup_object_buffer(obj);
214 if (!obj_buf)
215 die("Whoops! Cannot find object '%s'", oid_to_hex(&obj->oid));
216 if (fsck_object(obj, obj_buf->buffer, obj_buf->size, &fsck_options))
217 die("fsck error in packed object");
218 fsck_options.walk = check_object;
219 if (fsck_walk(obj, NULL, &fsck_options))
220 die("Error on reachable objects of %s", oid_to_hex(&obj->oid));
221 write_cached_object(obj, obj_buf);
222 return 0;
225 static void write_rest(void)
227 unsigned i;
228 for (i = 0; i < nr_objects; i++) {
229 if (obj_list[i].obj)
230 check_object(obj_list[i].obj, OBJ_ANY, NULL, NULL);
234 static void added_object(unsigned nr, enum object_type type,
235 void *data, unsigned long size);
238 * Write out nr-th object from the list, now we know the contents
239 * of it. Under --strict, this buffers structured objects in-core,
240 * to be checked at the end.
242 static void write_object(unsigned nr, enum object_type type,
243 void *buf, unsigned long size)
245 if (!strict) {
246 if (write_object_file(buf, size, type_name(type),
247 &obj_list[nr].oid) < 0)
248 die("failed to write object");
249 added_object(nr, type, buf, size);
250 free(buf);
251 obj_list[nr].obj = NULL;
252 } else if (type == OBJ_BLOB) {
253 struct blob *blob;
254 if (write_object_file(buf, size, type_name(type),
255 &obj_list[nr].oid) < 0)
256 die("failed to write object");
257 added_object(nr, type, buf, size);
258 free(buf);
260 blob = lookup_blob(the_repository, &obj_list[nr].oid);
261 if (blob)
262 blob->object.flags |= FLAG_WRITTEN;
263 else
264 die("invalid blob object");
265 obj_list[nr].obj = NULL;
266 } else {
267 struct object *obj;
268 int eaten;
269 hash_object_file(the_hash_algo, buf, size, type_name(type),
270 &obj_list[nr].oid);
271 added_object(nr, type, buf, size);
272 obj = parse_object_buffer(the_repository, &obj_list[nr].oid,
273 type, size, buf,
274 &eaten);
275 if (!obj)
276 die("invalid %s", type_name(type));
277 add_object_buffer(obj, buf, size);
278 obj->flags |= FLAG_OPEN;
279 obj_list[nr].obj = obj;
283 static void resolve_delta(unsigned nr, enum object_type type,
284 void *base, unsigned long base_size,
285 void *delta, unsigned long delta_size)
287 void *result;
288 unsigned long result_size;
290 result = patch_delta(base, base_size,
291 delta, delta_size,
292 &result_size);
293 if (!result)
294 die("failed to apply delta");
295 free(delta);
296 write_object(nr, type, result, result_size);
300 * We now know the contents of an object (which is nr-th in the pack);
301 * resolve all the deltified objects that are based on it.
303 static void added_object(unsigned nr, enum object_type type,
304 void *data, unsigned long size)
306 struct delta_info **p = &delta_list;
307 struct delta_info *info;
309 while ((info = *p) != NULL) {
310 if (oideq(&info->base_oid, &obj_list[nr].oid) ||
311 info->base_offset == obj_list[nr].offset) {
312 *p = info->next;
313 p = &delta_list;
314 resolve_delta(info->nr, type, data, size,
315 info->delta, info->size);
316 free(info);
317 continue;
319 p = &info->next;
323 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
324 unsigned nr)
326 void *buf = get_data(size);
328 if (!dry_run && buf)
329 write_object(nr, type, buf, size);
330 else
331 free(buf);
334 static int resolve_against_held(unsigned nr, const struct object_id *base,
335 void *delta_data, unsigned long delta_size)
337 struct object *obj;
338 struct obj_buffer *obj_buffer;
339 obj = lookup_object(the_repository, base);
340 if (!obj)
341 return 0;
342 obj_buffer = lookup_object_buffer(obj);
343 if (!obj_buffer)
344 return 0;
345 resolve_delta(nr, obj->type, obj_buffer->buffer,
346 obj_buffer->size, delta_data, delta_size);
347 return 1;
350 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
351 unsigned nr)
353 void *delta_data, *base;
354 unsigned long base_size;
355 struct object_id base_oid;
357 if (type == OBJ_REF_DELTA) {
358 oidread(&base_oid, fill(the_hash_algo->rawsz));
359 use(the_hash_algo->rawsz);
360 delta_data = get_data(delta_size);
361 if (dry_run || !delta_data) {
362 free(delta_data);
363 return;
365 if (has_object_file(&base_oid))
366 ; /* Ok we have this one */
367 else if (resolve_against_held(nr, &base_oid,
368 delta_data, delta_size))
369 return; /* we are done */
370 else {
371 /* cannot resolve yet --- queue it */
372 oidclr(&obj_list[nr].oid);
373 add_delta_to_list(nr, &base_oid, 0, delta_data, delta_size);
374 return;
376 } else {
377 unsigned base_found = 0;
378 unsigned char *pack, c;
379 off_t base_offset;
380 unsigned lo, mid, hi;
382 pack = fill(1);
383 c = *pack;
384 use(1);
385 base_offset = c & 127;
386 while (c & 128) {
387 base_offset += 1;
388 if (!base_offset || MSB(base_offset, 7))
389 die("offset value overflow for delta base object");
390 pack = fill(1);
391 c = *pack;
392 use(1);
393 base_offset = (base_offset << 7) + (c & 127);
395 base_offset = obj_list[nr].offset - base_offset;
396 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
397 die("offset value out of bound for delta base object");
399 delta_data = get_data(delta_size);
400 if (dry_run || !delta_data) {
401 free(delta_data);
402 return;
404 lo = 0;
405 hi = nr;
406 while (lo < hi) {
407 mid = lo + (hi - lo) / 2;
408 if (base_offset < obj_list[mid].offset) {
409 hi = mid;
410 } else if (base_offset > obj_list[mid].offset) {
411 lo = mid + 1;
412 } else {
413 oidcpy(&base_oid, &obj_list[mid].oid);
414 base_found = !is_null_oid(&base_oid);
415 break;
418 if (!base_found) {
420 * The delta base object is itself a delta that
421 * has not been resolved yet.
423 oidclr(&obj_list[nr].oid);
424 add_delta_to_list(nr, null_oid(), base_offset,
425 delta_data, delta_size);
426 return;
430 if (resolve_against_held(nr, &base_oid, delta_data, delta_size))
431 return;
433 base = read_object_file(&base_oid, &type, &base_size);
434 if (!base) {
435 error("failed to read delta-pack base object %s",
436 oid_to_hex(&base_oid));
437 if (!recover)
438 exit(1);
439 has_errors = 1;
440 return;
442 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
443 free(base);
446 static void unpack_one(unsigned nr)
448 unsigned shift;
449 unsigned char *pack;
450 unsigned long size, c;
451 enum object_type type;
453 obj_list[nr].offset = consumed_bytes;
455 pack = fill(1);
456 c = *pack;
457 use(1);
458 type = (c >> 4) & 7;
459 size = (c & 15);
460 shift = 4;
461 while (c & 0x80) {
462 pack = fill(1);
463 c = *pack;
464 use(1);
465 size += (c & 0x7f) << shift;
466 shift += 7;
469 switch (type) {
470 case OBJ_COMMIT:
471 case OBJ_TREE:
472 case OBJ_BLOB:
473 case OBJ_TAG:
474 unpack_non_delta_entry(type, size, nr);
475 return;
476 case OBJ_REF_DELTA:
477 case OBJ_OFS_DELTA:
478 unpack_delta_entry(type, size, nr);
479 return;
480 default:
481 error("bad object type %d", type);
482 has_errors = 1;
483 if (recover)
484 return;
485 exit(1);
489 static void unpack_all(void)
491 int i;
492 struct pack_header *hdr = fill(sizeof(struct pack_header));
494 nr_objects = ntohl(hdr->hdr_entries);
496 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
497 die("bad pack file");
498 if (!pack_version_ok(hdr->hdr_version))
499 die("unknown pack file version %"PRIu32,
500 ntohl(hdr->hdr_version));
501 use(sizeof(struct pack_header));
503 if (!quiet)
504 progress = start_progress(_("Unpacking objects"), nr_objects);
505 CALLOC_ARRAY(obj_list, nr_objects);
506 for (i = 0; i < nr_objects; i++) {
507 unpack_one(i);
508 display_progress(progress, i + 1);
510 stop_progress(&progress);
512 if (delta_list)
513 die("unresolved deltas left after unpacking");
516 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
518 int i;
519 struct object_id oid;
521 read_replace_refs = 0;
523 git_config(git_default_config, NULL);
525 quiet = !isatty(2);
527 for (i = 1 ; i < argc; i++) {
528 const char *arg = argv[i];
530 if (*arg == '-') {
531 if (!strcmp(arg, "-n")) {
532 dry_run = 1;
533 continue;
535 if (!strcmp(arg, "-q")) {
536 quiet = 1;
537 continue;
539 if (!strcmp(arg, "-r")) {
540 recover = 1;
541 continue;
543 if (!strcmp(arg, "--strict")) {
544 strict = 1;
545 continue;
547 if (skip_prefix(arg, "--strict=", &arg)) {
548 strict = 1;
549 fsck_set_msg_types(&fsck_options, arg);
550 continue;
552 if (starts_with(arg, "--pack_header=")) {
553 struct pack_header *hdr;
554 char *c;
556 hdr = (struct pack_header *)buffer;
557 hdr->hdr_signature = htonl(PACK_SIGNATURE);
558 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
559 if (*c != ',')
560 die("bad %s", arg);
561 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
562 if (*c)
563 die("bad %s", arg);
564 len = sizeof(*hdr);
565 continue;
567 if (skip_prefix(arg, "--max-input-size=", &arg)) {
568 max_input_size = strtoumax(arg, NULL, 10);
569 continue;
571 usage(unpack_usage);
574 /* We don't take any non-flag arguments now.. Maybe some day */
575 usage(unpack_usage);
577 the_hash_algo->init_fn(&ctx);
578 unpack_all();
579 the_hash_algo->update_fn(&ctx, buffer, offset);
580 the_hash_algo->final_oid_fn(&oid, &ctx);
581 if (strict) {
582 write_rest();
583 if (fsck_finish(&fsck_options))
584 die(_("fsck error in pack objects"));
586 if (!hasheq(fill(the_hash_algo->rawsz), oid.hash))
587 die("final sha1 did not match");
588 use(the_hash_algo->rawsz);
590 /* Write the last part of the buffer to stdout */
591 while (len) {
592 int ret = xwrite(1, buffer + offset, len);
593 if (ret <= 0)
594 break;
595 len -= ret;
596 offset += ret;
599 /* All done */
600 return has_errors;