Post 2.3 cyle (batch #5)
[git.git] / builtin / unpack-objects.c
blobac6667242c562bb7a215982754211a3e035aae8a
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "delta.h"
5 #include "pack.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "tree-walk.h"
11 #include "progress.h"
12 #include "decorate.h"
13 #include "fsck.h"
15 static int dry_run, quiet, recover, has_errors, strict;
16 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict] < pack-file";
18 /* We always read in 4kB chunks. */
19 static unsigned char buffer[4096];
20 static unsigned int offset, len;
21 static off_t consumed_bytes;
22 static git_SHA_CTX ctx;
25 * When running under --strict mode, objects whose reachability are
26 * suspect are kept in core without getting written in the object
27 * store.
29 struct obj_buffer {
30 char *buffer;
31 unsigned long size;
34 static struct decoration obj_decorate;
36 static struct obj_buffer *lookup_object_buffer(struct object *base)
38 return lookup_decoration(&obj_decorate, base);
41 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
43 struct obj_buffer *obj;
44 obj = xcalloc(1, sizeof(struct obj_buffer));
45 obj->buffer = buffer;
46 obj->size = size;
47 if (add_decoration(&obj_decorate, object, obj))
48 die("object %s tried to add buffer twice!", sha1_to_hex(object->sha1));
52 * Make sure at least "min" bytes are available in the buffer, and
53 * return the pointer to the buffer.
55 static void *fill(int min)
57 if (min <= len)
58 return buffer + offset;
59 if (min > sizeof(buffer))
60 die("cannot fill %d bytes", min);
61 if (offset) {
62 git_SHA1_Update(&ctx, buffer, offset);
63 memmove(buffer, buffer + offset, len);
64 offset = 0;
66 do {
67 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
68 if (ret <= 0) {
69 if (!ret)
70 die("early EOF");
71 die_errno("read error on input");
73 len += ret;
74 } while (len < min);
75 return buffer;
78 static void use(int bytes)
80 if (bytes > len)
81 die("used more bytes than were available");
82 len -= bytes;
83 offset += bytes;
85 /* make sure off_t is sufficiently large not to wrap */
86 if (signed_add_overflows(consumed_bytes, bytes))
87 die("pack too large for current definition of off_t");
88 consumed_bytes += bytes;
91 static void *get_data(unsigned long size)
93 git_zstream stream;
94 void *buf = xmallocz(size);
96 memset(&stream, 0, sizeof(stream));
98 stream.next_out = buf;
99 stream.avail_out = size;
100 stream.next_in = fill(1);
101 stream.avail_in = len;
102 git_inflate_init(&stream);
104 for (;;) {
105 int ret = git_inflate(&stream, 0);
106 use(len - stream.avail_in);
107 if (stream.total_out == size && ret == Z_STREAM_END)
108 break;
109 if (ret != Z_OK) {
110 error("inflate returned %d", ret);
111 free(buf);
112 buf = NULL;
113 if (!recover)
114 exit(1);
115 has_errors = 1;
116 break;
118 stream.next_in = fill(1);
119 stream.avail_in = len;
121 git_inflate_end(&stream);
122 return buf;
125 struct delta_info {
126 unsigned char base_sha1[20];
127 unsigned nr;
128 off_t base_offset;
129 unsigned long size;
130 void *delta;
131 struct delta_info *next;
134 static struct delta_info *delta_list;
136 static void add_delta_to_list(unsigned nr, unsigned const char *base_sha1,
137 off_t base_offset,
138 void *delta, unsigned long size)
140 struct delta_info *info = xmalloc(sizeof(*info));
142 hashcpy(info->base_sha1, base_sha1);
143 info->base_offset = base_offset;
144 info->size = size;
145 info->delta = delta;
146 info->nr = nr;
147 info->next = delta_list;
148 delta_list = info;
151 struct obj_info {
152 off_t offset;
153 unsigned char sha1[20];
154 struct object *obj;
157 #define FLAG_OPEN (1u<<20)
158 #define FLAG_WRITTEN (1u<<21)
160 static struct obj_info *obj_list;
161 static unsigned nr_objects;
164 * Called only from check_object() after it verified this object
165 * is Ok.
167 static void write_cached_object(struct object *obj, struct obj_buffer *obj_buf)
169 unsigned char sha1[20];
171 if (write_sha1_file(obj_buf->buffer, obj_buf->size, typename(obj->type), sha1) < 0)
172 die("failed to write object %s", sha1_to_hex(obj->sha1));
173 obj->flags |= FLAG_WRITTEN;
177 * At the very end of the processing, write_rest() scans the objects
178 * that have reachability requirements and calls this function.
179 * Verify its reachability and validity recursively and write it out.
181 static int check_object(struct object *obj, int type, void *data)
183 struct obj_buffer *obj_buf;
185 if (!obj)
186 return 1;
188 if (obj->flags & FLAG_WRITTEN)
189 return 0;
191 if (type != OBJ_ANY && obj->type != type)
192 die("object type mismatch");
194 if (!(obj->flags & FLAG_OPEN)) {
195 unsigned long size;
196 int type = sha1_object_info(obj->sha1, &size);
197 if (type != obj->type || type <= 0)
198 die("object of unexpected type");
199 obj->flags |= FLAG_WRITTEN;
200 return 0;
203 obj_buf = lookup_object_buffer(obj);
204 if (!obj_buf)
205 die("Whoops! Cannot find object '%s'", sha1_to_hex(obj->sha1));
206 if (fsck_object(obj, obj_buf->buffer, obj_buf->size, 1,
207 fsck_error_function))
208 die("Error in object");
209 if (fsck_walk(obj, check_object, NULL))
210 die("Error on reachable objects of %s", sha1_to_hex(obj->sha1));
211 write_cached_object(obj, obj_buf);
212 return 0;
215 static void write_rest(void)
217 unsigned i;
218 for (i = 0; i < nr_objects; i++) {
219 if (obj_list[i].obj)
220 check_object(obj_list[i].obj, OBJ_ANY, NULL);
224 static void added_object(unsigned nr, enum object_type type,
225 void *data, unsigned long size);
228 * Write out nr-th object from the list, now we know the contents
229 * of it. Under --strict, this buffers structured objects in-core,
230 * to be checked at the end.
232 static void write_object(unsigned nr, enum object_type type,
233 void *buf, unsigned long size)
235 if (!strict) {
236 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
237 die("failed to write object");
238 added_object(nr, type, buf, size);
239 free(buf);
240 obj_list[nr].obj = NULL;
241 } else if (type == OBJ_BLOB) {
242 struct blob *blob;
243 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
244 die("failed to write object");
245 added_object(nr, type, buf, size);
246 free(buf);
248 blob = lookup_blob(obj_list[nr].sha1);
249 if (blob)
250 blob->object.flags |= FLAG_WRITTEN;
251 else
252 die("invalid blob object");
253 obj_list[nr].obj = NULL;
254 } else {
255 struct object *obj;
256 int eaten;
257 hash_sha1_file(buf, size, typename(type), obj_list[nr].sha1);
258 added_object(nr, type, buf, size);
259 obj = parse_object_buffer(obj_list[nr].sha1, type, size, buf, &eaten);
260 if (!obj)
261 die("invalid %s", typename(type));
262 add_object_buffer(obj, buf, size);
263 obj->flags |= FLAG_OPEN;
264 obj_list[nr].obj = obj;
268 static void resolve_delta(unsigned nr, enum object_type type,
269 void *base, unsigned long base_size,
270 void *delta, unsigned long delta_size)
272 void *result;
273 unsigned long result_size;
275 result = patch_delta(base, base_size,
276 delta, delta_size,
277 &result_size);
278 if (!result)
279 die("failed to apply delta");
280 free(delta);
281 write_object(nr, type, result, result_size);
285 * We now know the contents of an object (which is nr-th in the pack);
286 * resolve all the deltified objects that are based on it.
288 static void added_object(unsigned nr, enum object_type type,
289 void *data, unsigned long size)
291 struct delta_info **p = &delta_list;
292 struct delta_info *info;
294 while ((info = *p) != NULL) {
295 if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
296 info->base_offset == obj_list[nr].offset) {
297 *p = info->next;
298 p = &delta_list;
299 resolve_delta(info->nr, type, data, size,
300 info->delta, info->size);
301 free(info);
302 continue;
304 p = &info->next;
308 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
309 unsigned nr)
311 void *buf = get_data(size);
313 if (!dry_run && buf)
314 write_object(nr, type, buf, size);
315 else
316 free(buf);
319 static int resolve_against_held(unsigned nr, const unsigned char *base,
320 void *delta_data, unsigned long delta_size)
322 struct object *obj;
323 struct obj_buffer *obj_buffer;
324 obj = lookup_object(base);
325 if (!obj)
326 return 0;
327 obj_buffer = lookup_object_buffer(obj);
328 if (!obj_buffer)
329 return 0;
330 resolve_delta(nr, obj->type, obj_buffer->buffer,
331 obj_buffer->size, delta_data, delta_size);
332 return 1;
335 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
336 unsigned nr)
338 void *delta_data, *base;
339 unsigned long base_size;
340 unsigned char base_sha1[20];
342 if (type == OBJ_REF_DELTA) {
343 hashcpy(base_sha1, fill(20));
344 use(20);
345 delta_data = get_data(delta_size);
346 if (dry_run || !delta_data) {
347 free(delta_data);
348 return;
350 if (has_sha1_file(base_sha1))
351 ; /* Ok we have this one */
352 else if (resolve_against_held(nr, base_sha1,
353 delta_data, delta_size))
354 return; /* we are done */
355 else {
356 /* cannot resolve yet --- queue it */
357 hashcpy(obj_list[nr].sha1, null_sha1);
358 add_delta_to_list(nr, base_sha1, 0, delta_data, delta_size);
359 return;
361 } else {
362 unsigned base_found = 0;
363 unsigned char *pack, c;
364 off_t base_offset;
365 unsigned lo, mid, hi;
367 pack = fill(1);
368 c = *pack;
369 use(1);
370 base_offset = c & 127;
371 while (c & 128) {
372 base_offset += 1;
373 if (!base_offset || MSB(base_offset, 7))
374 die("offset value overflow for delta base object");
375 pack = fill(1);
376 c = *pack;
377 use(1);
378 base_offset = (base_offset << 7) + (c & 127);
380 base_offset = obj_list[nr].offset - base_offset;
381 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
382 die("offset value out of bound for delta base object");
384 delta_data = get_data(delta_size);
385 if (dry_run || !delta_data) {
386 free(delta_data);
387 return;
389 lo = 0;
390 hi = nr;
391 while (lo < hi) {
392 mid = (lo + hi)/2;
393 if (base_offset < obj_list[mid].offset) {
394 hi = mid;
395 } else if (base_offset > obj_list[mid].offset) {
396 lo = mid + 1;
397 } else {
398 hashcpy(base_sha1, obj_list[mid].sha1);
399 base_found = !is_null_sha1(base_sha1);
400 break;
403 if (!base_found) {
405 * The delta base object is itself a delta that
406 * has not been resolved yet.
408 hashcpy(obj_list[nr].sha1, null_sha1);
409 add_delta_to_list(nr, null_sha1, base_offset, delta_data, delta_size);
410 return;
414 if (resolve_against_held(nr, base_sha1, delta_data, delta_size))
415 return;
417 base = read_sha1_file(base_sha1, &type, &base_size);
418 if (!base) {
419 error("failed to read delta-pack base object %s",
420 sha1_to_hex(base_sha1));
421 if (!recover)
422 exit(1);
423 has_errors = 1;
424 return;
426 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
427 free(base);
430 static void unpack_one(unsigned nr)
432 unsigned shift;
433 unsigned char *pack;
434 unsigned long size, c;
435 enum object_type type;
437 obj_list[nr].offset = consumed_bytes;
439 pack = fill(1);
440 c = *pack;
441 use(1);
442 type = (c >> 4) & 7;
443 size = (c & 15);
444 shift = 4;
445 while (c & 0x80) {
446 pack = fill(1);
447 c = *pack;
448 use(1);
449 size += (c & 0x7f) << shift;
450 shift += 7;
453 switch (type) {
454 case OBJ_COMMIT:
455 case OBJ_TREE:
456 case OBJ_BLOB:
457 case OBJ_TAG:
458 unpack_non_delta_entry(type, size, nr);
459 return;
460 case OBJ_REF_DELTA:
461 case OBJ_OFS_DELTA:
462 unpack_delta_entry(type, size, nr);
463 return;
464 default:
465 error("bad object type %d", type);
466 has_errors = 1;
467 if (recover)
468 return;
469 exit(1);
473 static void unpack_all(void)
475 int i;
476 struct progress *progress = NULL;
477 struct pack_header *hdr = fill(sizeof(struct pack_header));
479 nr_objects = ntohl(hdr->hdr_entries);
481 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
482 die("bad pack file");
483 if (!pack_version_ok(hdr->hdr_version))
484 die("unknown pack file version %"PRIu32,
485 ntohl(hdr->hdr_version));
486 use(sizeof(struct pack_header));
488 if (!quiet)
489 progress = start_progress(_("Unpacking objects"), nr_objects);
490 obj_list = xcalloc(nr_objects, sizeof(*obj_list));
491 for (i = 0; i < nr_objects; i++) {
492 unpack_one(i);
493 display_progress(progress, i + 1);
495 stop_progress(&progress);
497 if (delta_list)
498 die("unresolved deltas left after unpacking");
501 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
503 int i;
504 unsigned char sha1[20];
506 check_replace_refs = 0;
508 git_config(git_default_config, NULL);
510 quiet = !isatty(2);
512 for (i = 1 ; i < argc; i++) {
513 const char *arg = argv[i];
515 if (*arg == '-') {
516 if (!strcmp(arg, "-n")) {
517 dry_run = 1;
518 continue;
520 if (!strcmp(arg, "-q")) {
521 quiet = 1;
522 continue;
524 if (!strcmp(arg, "-r")) {
525 recover = 1;
526 continue;
528 if (!strcmp(arg, "--strict")) {
529 strict = 1;
530 continue;
532 if (starts_with(arg, "--pack_header=")) {
533 struct pack_header *hdr;
534 char *c;
536 hdr = (struct pack_header *)buffer;
537 hdr->hdr_signature = htonl(PACK_SIGNATURE);
538 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
539 if (*c != ',')
540 die("bad %s", arg);
541 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
542 if (*c)
543 die("bad %s", arg);
544 len = sizeof(*hdr);
545 continue;
547 usage(unpack_usage);
550 /* We don't take any non-flag arguments now.. Maybe some day */
551 usage(unpack_usage);
553 git_SHA1_Init(&ctx);
554 unpack_all();
555 git_SHA1_Update(&ctx, buffer, offset);
556 git_SHA1_Final(sha1, &ctx);
557 if (strict)
558 write_rest();
559 if (hashcmp(fill(20), sha1))
560 die("final sha1 did not match");
561 use(20);
563 /* Write the last part of the buffer to stdout */
564 while (len) {
565 int ret = xwrite(1, buffer + offset, len);
566 if (ret <= 0)
567 break;
568 len -= ret;
569 offset += ret;
572 /* All done */
573 return has_errors;