Merge branch 'en/rebase-merge-on-sequencer'
[git.git] / builtin / unpack-objects.c
blob80478808b3dcc7f98e5989140de1451783454c01
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;
29 * When running under --strict mode, objects whose reachability are
30 * suspect are kept in core without getting written in the object
31 * store.
33 struct obj_buffer {
34 char *buffer;
35 unsigned long size;
38 static struct decoration obj_decorate;
40 static struct obj_buffer *lookup_object_buffer(struct object *base)
42 return lookup_decoration(&obj_decorate, base);
45 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
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));
56 * Make sure at least "min" bytes are available in the buffer, and
57 * return the pointer to the buffer.
59 static void *fill(int min)
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;
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");
77 len += ret;
78 } while (len < min);
79 return buffer;
82 static void use(int bytes)
84 if (bytes > len)
85 die("used more bytes than were available");
86 len -= bytes;
87 offset += bytes;
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"));
97 static void *get_data(unsigned long size)
99 git_zstream stream;
100 void *buf = xmallocz(size);
102 memset(&stream, 0, sizeof(stream));
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);
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;
123 stream.next_in = fill(1);
124 stream.avail_in = len;
126 git_inflate_end(&stream);
127 return buf;
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;
139 static struct delta_info *delta_list;
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)
145 struct delta_info *info = xmalloc(sizeof(*info));
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;
156 struct obj_info {
157 off_t offset;
158 struct object_id oid;
159 struct object *obj;
162 /* Remember to update object flag allocation in object.h */
163 #define FLAG_OPEN (1u<<20)
164 #define FLAG_WRITTEN (1u<<21)
166 static struct obj_info *obj_list;
167 static unsigned nr_objects;
170 * Called only from check_object() after it verified this object
171 * is Ok.
173 static void write_cached_object(struct object *obj, struct obj_buffer *obj_buf)
175 struct object_id oid;
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;
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.
188 static int check_object(struct object *obj, int type, void *data, struct fsck_options *options)
190 struct obj_buffer *obj_buf;
192 if (!obj)
193 return 1;
195 if (obj->flags & FLAG_WRITTEN)
196 return 0;
198 if (type != OBJ_ANY && obj->type != type)
199 die("object type mismatch");
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;
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;
222 static void write_rest(void)
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);
231 static void added_object(unsigned nr, enum object_type type,
232 void *data, unsigned long size);
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.
239 static void write_object(unsigned nr, enum object_type type,
240 void *buf, unsigned long size)
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);
257 blob = lookup_blob(the_repository, &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(the_repository, &obj_list[nr].oid,
269 type, size, buf,
270 &eaten);
271 if (!obj)
272 die("invalid %s", type_name(type));
273 add_object_buffer(obj, buf, size);
274 obj->flags |= FLAG_OPEN;
275 obj_list[nr].obj = obj;
279 static void resolve_delta(unsigned nr, enum object_type type,
280 void *base, unsigned long base_size,
281 void *delta, unsigned long delta_size)
283 void *result;
284 unsigned long result_size;
286 result = patch_delta(base, base_size,
287 delta, delta_size,
288 &result_size);
289 if (!result)
290 die("failed to apply delta");
291 free(delta);
292 write_object(nr, type, result, result_size);
296 * We now know the contents of an object (which is nr-th in the pack);
297 * resolve all the deltified objects that are based on it.
299 static void added_object(unsigned nr, enum object_type type,
300 void *data, unsigned long size)
302 struct delta_info **p = &delta_list;
303 struct delta_info *info;
305 while ((info = *p) != NULL) {
306 if (oideq(&info->base_oid, &obj_list[nr].oid) ||
307 info->base_offset == obj_list[nr].offset) {
308 *p = info->next;
309 p = &delta_list;
310 resolve_delta(info->nr, type, data, size,
311 info->delta, info->size);
312 free(info);
313 continue;
315 p = &info->next;
319 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
320 unsigned nr)
322 void *buf = get_data(size);
324 if (!dry_run && buf)
325 write_object(nr, type, buf, size);
326 else
327 free(buf);
330 static int resolve_against_held(unsigned nr, const struct object_id *base,
331 void *delta_data, unsigned long delta_size)
333 struct object *obj;
334 struct obj_buffer *obj_buffer;
335 obj = lookup_object(the_repository, base->hash);
336 if (!obj)
337 return 0;
338 obj_buffer = lookup_object_buffer(obj);
339 if (!obj_buffer)
340 return 0;
341 resolve_delta(nr, obj->type, obj_buffer->buffer,
342 obj_buffer->size, delta_data, delta_size);
343 return 1;
346 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
347 unsigned nr)
349 void *delta_data, *base;
350 unsigned long base_size;
351 struct object_id base_oid;
353 if (type == OBJ_REF_DELTA) {
354 hashcpy(base_oid.hash, fill(the_hash_algo->rawsz));
355 use(the_hash_algo->rawsz);
356 delta_data = get_data(delta_size);
357 if (dry_run || !delta_data) {
358 free(delta_data);
359 return;
361 if (has_object_file(&base_oid))
362 ; /* Ok we have this one */
363 else if (resolve_against_held(nr, &base_oid,
364 delta_data, delta_size))
365 return; /* we are done */
366 else {
367 /* cannot resolve yet --- queue it */
368 oidclr(&obj_list[nr].oid);
369 add_delta_to_list(nr, &base_oid, 0, delta_data, delta_size);
370 return;
372 } else {
373 unsigned base_found = 0;
374 unsigned char *pack, c;
375 off_t base_offset;
376 unsigned lo, mid, hi;
378 pack = fill(1);
379 c = *pack;
380 use(1);
381 base_offset = c & 127;
382 while (c & 128) {
383 base_offset += 1;
384 if (!base_offset || MSB(base_offset, 7))
385 die("offset value overflow for delta base object");
386 pack = fill(1);
387 c = *pack;
388 use(1);
389 base_offset = (base_offset << 7) + (c & 127);
391 base_offset = obj_list[nr].offset - base_offset;
392 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
393 die("offset value out of bound for delta base object");
395 delta_data = get_data(delta_size);
396 if (dry_run || !delta_data) {
397 free(delta_data);
398 return;
400 lo = 0;
401 hi = nr;
402 while (lo < hi) {
403 mid = lo + (hi - lo) / 2;
404 if (base_offset < obj_list[mid].offset) {
405 hi = mid;
406 } else if (base_offset > obj_list[mid].offset) {
407 lo = mid + 1;
408 } else {
409 oidcpy(&base_oid, &obj_list[mid].oid);
410 base_found = !is_null_oid(&base_oid);
411 break;
414 if (!base_found) {
416 * The delta base object is itself a delta that
417 * has not been resolved yet.
419 oidclr(&obj_list[nr].oid);
420 add_delta_to_list(nr, &null_oid, base_offset, delta_data, delta_size);
421 return;
425 if (resolve_against_held(nr, &base_oid, delta_data, delta_size))
426 return;
428 base = read_object_file(&base_oid, &type, &base_size);
429 if (!base) {
430 error("failed to read delta-pack base object %s",
431 oid_to_hex(&base_oid));
432 if (!recover)
433 exit(1);
434 has_errors = 1;
435 return;
437 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
438 free(base);
441 static void unpack_one(unsigned nr)
443 unsigned shift;
444 unsigned char *pack;
445 unsigned long size, c;
446 enum object_type type;
448 obj_list[nr].offset = consumed_bytes;
450 pack = fill(1);
451 c = *pack;
452 use(1);
453 type = (c >> 4) & 7;
454 size = (c & 15);
455 shift = 4;
456 while (c & 0x80) {
457 pack = fill(1);
458 c = *pack;
459 use(1);
460 size += (c & 0x7f) << shift;
461 shift += 7;
464 switch (type) {
465 case OBJ_COMMIT:
466 case OBJ_TREE:
467 case OBJ_BLOB:
468 case OBJ_TAG:
469 unpack_non_delta_entry(type, size, nr);
470 return;
471 case OBJ_REF_DELTA:
472 case OBJ_OFS_DELTA:
473 unpack_delta_entry(type, size, nr);
474 return;
475 default:
476 error("bad object type %d", type);
477 has_errors = 1;
478 if (recover)
479 return;
480 exit(1);
484 static void unpack_all(void)
486 int i;
487 struct progress *progress = NULL;
488 struct pack_header *hdr = fill(sizeof(struct pack_header));
490 nr_objects = ntohl(hdr->hdr_entries);
492 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
493 die("bad pack file");
494 if (!pack_version_ok(hdr->hdr_version))
495 die("unknown pack file version %"PRIu32,
496 ntohl(hdr->hdr_version));
497 use(sizeof(struct pack_header));
499 if (!quiet)
500 progress = start_progress(_("Unpacking objects"), nr_objects);
501 obj_list = xcalloc(nr_objects, sizeof(*obj_list));
502 for (i = 0; i < nr_objects; i++) {
503 unpack_one(i);
504 display_progress(progress, i + 1);
506 stop_progress(&progress);
508 if (delta_list)
509 die("unresolved deltas left after unpacking");
512 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
514 int i;
515 struct object_id oid;
517 read_replace_refs = 0;
519 git_config(git_default_config, NULL);
521 quiet = !isatty(2);
523 for (i = 1 ; i < argc; i++) {
524 const char *arg = argv[i];
526 if (*arg == '-') {
527 if (!strcmp(arg, "-n")) {
528 dry_run = 1;
529 continue;
531 if (!strcmp(arg, "-q")) {
532 quiet = 1;
533 continue;
535 if (!strcmp(arg, "-r")) {
536 recover = 1;
537 continue;
539 if (!strcmp(arg, "--strict")) {
540 strict = 1;
541 continue;
543 if (skip_prefix(arg, "--strict=", &arg)) {
544 strict = 1;
545 fsck_set_msg_types(&fsck_options, arg);
546 continue;
548 if (starts_with(arg, "--pack_header=")) {
549 struct pack_header *hdr;
550 char *c;
552 hdr = (struct pack_header *)buffer;
553 hdr->hdr_signature = htonl(PACK_SIGNATURE);
554 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
555 if (*c != ',')
556 die("bad %s", arg);
557 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
558 if (*c)
559 die("bad %s", arg);
560 len = sizeof(*hdr);
561 continue;
563 if (skip_prefix(arg, "--max-input-size=", &arg)) {
564 max_input_size = strtoumax(arg, NULL, 10);
565 continue;
567 usage(unpack_usage);
570 /* We don't take any non-flag arguments now.. Maybe some day */
571 usage(unpack_usage);
573 the_hash_algo->init_fn(&ctx);
574 unpack_all();
575 the_hash_algo->update_fn(&ctx, buffer, offset);
576 the_hash_algo->final_fn(oid.hash, &ctx);
577 if (strict) {
578 write_rest();
579 if (fsck_finish(&fsck_options))
580 die(_("fsck error in pack objects"));
582 if (!hasheq(fill(the_hash_algo->rawsz), oid.hash))
583 die("final sha1 did not match");
584 use(the_hash_algo->rawsz);
586 /* Write the last part of the buffer to stdout */
587 while (len) {
588 int ret = xwrite(1, buffer + offset, len);
589 if (ret <= 0)
590 break;
591 len -= ret;
592 offset += ret;
595 /* All done */
596 return has_errors;