gitweb: Prepare for splitting gitweb
[git/jnareb-git/bp-gitweb.git] / builtin / unpack-objects.c
blobf63973c9143c4109a3d1e40df9ec66a32498fe00
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 z_stream stream;
94 void *buf = xmalloc(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\n", 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)
169 unsigned char sha1[20];
170 struct obj_buffer *obj_buf = lookup_object_buffer(obj);
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 if (!obj)
184 return 1;
186 if (obj->flags & FLAG_WRITTEN)
187 return 0;
189 if (type != OBJ_ANY && obj->type != type)
190 die("object type mismatch");
192 if (!(obj->flags & FLAG_OPEN)) {
193 unsigned long size;
194 int type = sha1_object_info(obj->sha1, &size);
195 if (type != obj->type || type <= 0)
196 die("object of unexpected type");
197 obj->flags |= FLAG_WRITTEN;
198 return 0;
201 if (fsck_object(obj, 1, fsck_error_function))
202 die("Error in object");
203 if (fsck_walk(obj, check_object, NULL))
204 die("Error on reachable objects of %s", sha1_to_hex(obj->sha1));
205 write_cached_object(obj);
206 return 0;
209 static void write_rest(void)
211 unsigned i;
212 for (i = 0; i < nr_objects; i++) {
213 if (obj_list[i].obj)
214 check_object(obj_list[i].obj, OBJ_ANY, NULL);
218 static void added_object(unsigned nr, enum object_type type,
219 void *data, unsigned long size);
222 * Write out nr-th object from the list, now we know the contents
223 * of it. Under --strict, this buffers structured objects in-core,
224 * to be checked at the end.
226 static void write_object(unsigned nr, enum object_type type,
227 void *buf, unsigned long size)
229 if (!strict) {
230 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
231 die("failed to write object");
232 added_object(nr, type, buf, size);
233 free(buf);
234 obj_list[nr].obj = NULL;
235 } else if (type == OBJ_BLOB) {
236 struct blob *blob;
237 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
238 die("failed to write object");
239 added_object(nr, type, buf, size);
240 free(buf);
242 blob = lookup_blob(obj_list[nr].sha1);
243 if (blob)
244 blob->object.flags |= FLAG_WRITTEN;
245 else
246 die("invalid blob object");
247 obj_list[nr].obj = NULL;
248 } else {
249 struct object *obj;
250 int eaten;
251 hash_sha1_file(buf, size, typename(type), obj_list[nr].sha1);
252 added_object(nr, type, buf, size);
253 obj = parse_object_buffer(obj_list[nr].sha1, type, size, buf, &eaten);
254 if (!obj)
255 die("invalid %s", typename(type));
256 add_object_buffer(obj, buf, size);
257 obj->flags |= FLAG_OPEN;
258 obj_list[nr].obj = obj;
262 static void resolve_delta(unsigned nr, enum object_type type,
263 void *base, unsigned long base_size,
264 void *delta, unsigned long delta_size)
266 void *result;
267 unsigned long result_size;
269 result = patch_delta(base, base_size,
270 delta, delta_size,
271 &result_size);
272 if (!result)
273 die("failed to apply delta");
274 free(delta);
275 write_object(nr, type, result, result_size);
279 * We now know the contents of an object (which is nr-th in the pack);
280 * resolve all the deltified objects that are based on it.
282 static void added_object(unsigned nr, enum object_type type,
283 void *data, unsigned long size)
285 struct delta_info **p = &delta_list;
286 struct delta_info *info;
288 while ((info = *p) != NULL) {
289 if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
290 info->base_offset == obj_list[nr].offset) {
291 *p = info->next;
292 p = &delta_list;
293 resolve_delta(info->nr, type, data, size,
294 info->delta, info->size);
295 free(info);
296 continue;
298 p = &info->next;
302 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
303 unsigned nr)
305 void *buf = get_data(size);
307 if (!dry_run && buf)
308 write_object(nr, type, buf, size);
309 else
310 free(buf);
313 static int resolve_against_held(unsigned nr, const unsigned char *base,
314 void *delta_data, unsigned long delta_size)
316 struct object *obj;
317 struct obj_buffer *obj_buffer;
318 obj = lookup_object(base);
319 if (!obj)
320 return 0;
321 obj_buffer = lookup_object_buffer(obj);
322 if (!obj_buffer)
323 return 0;
324 resolve_delta(nr, obj->type, obj_buffer->buffer,
325 obj_buffer->size, delta_data, delta_size);
326 return 1;
329 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
330 unsigned nr)
332 void *delta_data, *base;
333 unsigned long base_size;
334 unsigned char base_sha1[20];
336 if (type == OBJ_REF_DELTA) {
337 hashcpy(base_sha1, fill(20));
338 use(20);
339 delta_data = get_data(delta_size);
340 if (dry_run || !delta_data) {
341 free(delta_data);
342 return;
344 if (has_sha1_file(base_sha1))
345 ; /* Ok we have this one */
346 else if (resolve_against_held(nr, base_sha1,
347 delta_data, delta_size))
348 return; /* we are done */
349 else {
350 /* cannot resolve yet --- queue it */
351 hashcpy(obj_list[nr].sha1, null_sha1);
352 add_delta_to_list(nr, base_sha1, 0, delta_data, delta_size);
353 return;
355 } else {
356 unsigned base_found = 0;
357 unsigned char *pack, c;
358 off_t base_offset;
359 unsigned lo, mid, hi;
361 pack = fill(1);
362 c = *pack;
363 use(1);
364 base_offset = c & 127;
365 while (c & 128) {
366 base_offset += 1;
367 if (!base_offset || MSB(base_offset, 7))
368 die("offset value overflow for delta base object");
369 pack = fill(1);
370 c = *pack;
371 use(1);
372 base_offset = (base_offset << 7) + (c & 127);
374 base_offset = obj_list[nr].offset - base_offset;
375 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
376 die("offset value out of bound for delta base object");
378 delta_data = get_data(delta_size);
379 if (dry_run || !delta_data) {
380 free(delta_data);
381 return;
383 lo = 0;
384 hi = nr;
385 while (lo < hi) {
386 mid = (lo + hi)/2;
387 if (base_offset < obj_list[mid].offset) {
388 hi = mid;
389 } else if (base_offset > obj_list[mid].offset) {
390 lo = mid + 1;
391 } else {
392 hashcpy(base_sha1, obj_list[mid].sha1);
393 base_found = !is_null_sha1(base_sha1);
394 break;
397 if (!base_found) {
399 * The delta base object is itself a delta that
400 * has not been resolved yet.
402 hashcpy(obj_list[nr].sha1, null_sha1);
403 add_delta_to_list(nr, null_sha1, base_offset, delta_data, delta_size);
404 return;
408 if (resolve_against_held(nr, base_sha1, delta_data, delta_size))
409 return;
411 base = read_sha1_file(base_sha1, &type, &base_size);
412 if (!base) {
413 error("failed to read delta-pack base object %s",
414 sha1_to_hex(base_sha1));
415 if (!recover)
416 exit(1);
417 has_errors = 1;
418 return;
420 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
421 free(base);
424 static void unpack_one(unsigned nr)
426 unsigned shift;
427 unsigned char *pack;
428 unsigned long size, c;
429 enum object_type type;
431 obj_list[nr].offset = consumed_bytes;
433 pack = fill(1);
434 c = *pack;
435 use(1);
436 type = (c >> 4) & 7;
437 size = (c & 15);
438 shift = 4;
439 while (c & 0x80) {
440 pack = fill(1);
441 c = *pack;
442 use(1);
443 size += (c & 0x7f) << shift;
444 shift += 7;
447 switch (type) {
448 case OBJ_COMMIT:
449 case OBJ_TREE:
450 case OBJ_BLOB:
451 case OBJ_TAG:
452 unpack_non_delta_entry(type, size, nr);
453 return;
454 case OBJ_REF_DELTA:
455 case OBJ_OFS_DELTA:
456 unpack_delta_entry(type, size, nr);
457 return;
458 default:
459 error("bad object type %d", type);
460 has_errors = 1;
461 if (recover)
462 return;
463 exit(1);
467 static void unpack_all(void)
469 int i;
470 struct progress *progress = NULL;
471 struct pack_header *hdr = fill(sizeof(struct pack_header));
473 nr_objects = ntohl(hdr->hdr_entries);
475 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
476 die("bad pack file");
477 if (!pack_version_ok(hdr->hdr_version))
478 die("unknown pack file version %"PRIu32,
479 ntohl(hdr->hdr_version));
480 use(sizeof(struct pack_header));
482 if (!quiet)
483 progress = start_progress("Unpacking objects", nr_objects);
484 obj_list = xcalloc(nr_objects, sizeof(*obj_list));
485 for (i = 0; i < nr_objects; i++) {
486 unpack_one(i);
487 display_progress(progress, i + 1);
489 stop_progress(&progress);
491 if (delta_list)
492 die("unresolved deltas left after unpacking");
495 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
497 int i;
498 unsigned char sha1[20];
500 read_replace_refs = 0;
502 git_config(git_default_config, NULL);
504 quiet = !isatty(2);
506 for (i = 1 ; i < argc; i++) {
507 const char *arg = argv[i];
509 if (*arg == '-') {
510 if (!strcmp(arg, "-n")) {
511 dry_run = 1;
512 continue;
514 if (!strcmp(arg, "-q")) {
515 quiet = 1;
516 continue;
518 if (!strcmp(arg, "-r")) {
519 recover = 1;
520 continue;
522 if (!strcmp(arg, "--strict")) {
523 strict = 1;
524 continue;
526 if (!prefixcmp(arg, "--pack_header=")) {
527 struct pack_header *hdr;
528 char *c;
530 hdr = (struct pack_header *)buffer;
531 hdr->hdr_signature = htonl(PACK_SIGNATURE);
532 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
533 if (*c != ',')
534 die("bad %s", arg);
535 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
536 if (*c)
537 die("bad %s", arg);
538 len = sizeof(*hdr);
539 continue;
541 usage(unpack_usage);
544 /* We don't take any non-flag arguments now.. Maybe some day */
545 usage(unpack_usage);
547 git_SHA1_Init(&ctx);
548 unpack_all();
549 git_SHA1_Update(&ctx, buffer, offset);
550 git_SHA1_Final(sha1, &ctx);
551 if (strict)
552 write_rest();
553 if (hashcmp(fill(20), sha1))
554 die("final sha1 did not match");
555 use(20);
557 /* Write the last part of the buffer to stdout */
558 while (len) {
559 int ret = xwrite(1, buffer + offset, len);
560 if (ret <= 0)
561 break;
562 len -= ret;
563 offset += ret;
566 /* All done */
567 return has_errors;