Merge branch 'js/msgfmt-on-windows' of ../git-gui into js/git-gui-msgfmt-on-windows
[git] / builtin / unpack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "config.h"
4 #include "object.h"
5 #include "delta.h"
6 #include "pack.h"
7 #include "blob.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "tree.h"
11 #include "tree-walk.h"
12 #include "progress.h"
13 #include "decorate.h"
14 #include "fsck.h"
15
16 static int dry_run, quiet, recover, has_errors, strict;
17 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict]";
18
19 /* We always read in 4kB chunks. */
20 static unsigned char buffer[4096];
21 static unsigned int offset, len;
22 static off_t consumed_bytes;
23 static off_t max_input_size;
24 static git_SHA_CTX ctx;
25 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
26
27 /*
28  * When running under --strict mode, objects whose reachability are
29  * suspect are kept in core without getting written in the object
30  * store.
31  */
32 struct obj_buffer {
33         char *buffer;
34         unsigned long size;
35 };
36
37 static struct decoration obj_decorate;
38
39 static struct obj_buffer *lookup_object_buffer(struct object *base)
40 {
41         return lookup_decoration(&obj_decorate, base);
42 }
43
44 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
45 {
46         struct obj_buffer *obj;
47         obj = xcalloc(1, sizeof(struct obj_buffer));
48         obj->buffer = buffer;
49         obj->size = size;
50         if (add_decoration(&obj_decorate, object, obj))
51                 die("object %s tried to add buffer twice!", oid_to_hex(&object->oid));
52 }
53
54 /*
55  * Make sure at least "min" bytes are available in the buffer, and
56  * return the pointer to the buffer.
57  */
58 static void *fill(int min)
59 {
60         if (min <= len)
61                 return buffer + offset;
62         if (min > sizeof(buffer))
63                 die("cannot fill %d bytes", min);
64         if (offset) {
65                 git_SHA1_Update(&ctx, buffer, offset);
66                 memmove(buffer, buffer + offset, len);
67                 offset = 0;
68         }
69         do {
70                 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
71                 if (ret <= 0) {
72                         if (!ret)
73                                 die("early EOF");
74                         die_errno("read error on input");
75                 }
76                 len += ret;
77         } while (len < min);
78         return buffer;
79 }
80
81 static void use(int bytes)
82 {
83         if (bytes > len)
84                 die("used more bytes than were available");
85         len -= bytes;
86         offset += bytes;
87
88         /* make sure off_t is sufficiently large not to wrap */
89         if (signed_add_overflows(consumed_bytes, bytes))
90                 die("pack too large for current definition of off_t");
91         consumed_bytes += bytes;
92         if (max_input_size && consumed_bytes > max_input_size)
93                 die(_("pack exceeds maximum allowed size"));
94 }
95
96 static void *get_data(unsigned long size)
97 {
98         git_zstream stream;
99         void *buf = xmallocz(size);
100
101         memset(&stream, 0, sizeof(stream));
102
103         stream.next_out = buf;
104         stream.avail_out = size;
105         stream.next_in = fill(1);
106         stream.avail_in = len;
107         git_inflate_init(&stream);
108
109         for (;;) {
110                 int ret = git_inflate(&stream, 0);
111                 use(len - stream.avail_in);
112                 if (stream.total_out == size && ret == Z_STREAM_END)
113                         break;
114                 if (ret != Z_OK) {
115                         error("inflate returned %d", ret);
116                         FREE_AND_NULL(buf);
117                         if (!recover)
118                                 exit(1);
119                         has_errors = 1;
120                         break;
121                 }
122                 stream.next_in = fill(1);
123                 stream.avail_in = len;
124         }
125         git_inflate_end(&stream);
126         return buf;
127 }
128
129 struct delta_info {
130         struct object_id base_oid;
131         unsigned nr;
132         off_t base_offset;
133         unsigned long size;
134         void *delta;
135         struct delta_info *next;
136 };
137
138 static struct delta_info *delta_list;
139
140 static void add_delta_to_list(unsigned nr, const struct object_id *base_oid,
141                               off_t base_offset,
142                               void *delta, unsigned long size)
143 {
144         struct delta_info *info = xmalloc(sizeof(*info));
145
146         oidcpy(&info->base_oid, base_oid);
147         info->base_offset = base_offset;
148         info->size = size;
149         info->delta = delta;
150         info->nr = nr;
151         info->next = delta_list;
152         delta_list = info;
153 }
154
155 struct obj_info {
156         off_t offset;
157         struct object_id oid;
158         struct object *obj;
159 };
160
161 #define FLAG_OPEN (1u<<20)
162 #define FLAG_WRITTEN (1u<<21)
163
164 static struct obj_info *obj_list;
165 static unsigned nr_objects;
166
167 /*
168  * Called only from check_object() after it verified this object
169  * is Ok.
170  */
171 static void write_cached_object(struct object *obj, struct obj_buffer *obj_buf)
172 {
173         struct object_id oid;
174
175         if (write_sha1_file(obj_buf->buffer, obj_buf->size, typename(obj->type), oid.hash) < 0)
176                 die("failed to write object %s", oid_to_hex(&obj->oid));
177         obj->flags |= FLAG_WRITTEN;
178 }
179
180 /*
181  * At the very end of the processing, write_rest() scans the objects
182  * that have reachability requirements and calls this function.
183  * Verify its reachability and validity recursively and write it out.
184  */
185 static int check_object(struct object *obj, int type, void *data, struct fsck_options *options)
186 {
187         struct obj_buffer *obj_buf;
188
189         if (!obj)
190                 return 1;
191
192         if (obj->flags & FLAG_WRITTEN)
193                 return 0;
194
195         if (type != OBJ_ANY && obj->type != type)
196                 die("object type mismatch");
197
198         if (!(obj->flags & FLAG_OPEN)) {
199                 unsigned long size;
200                 int type = sha1_object_info(obj->oid.hash, &size);
201                 if (type != obj->type || type <= 0)
202                         die("object of unexpected type");
203                 obj->flags |= FLAG_WRITTEN;
204                 return 0;
205         }
206
207         obj_buf = lookup_object_buffer(obj);
208         if (!obj_buf)
209                 die("Whoops! Cannot find object '%s'", oid_to_hex(&obj->oid));
210         if (fsck_object(obj, obj_buf->buffer, obj_buf->size, &fsck_options))
211                 die("Error in object");
212         fsck_options.walk = check_object;
213         if (fsck_walk(obj, NULL, &fsck_options))
214                 die("Error on reachable objects of %s", oid_to_hex(&obj->oid));
215         write_cached_object(obj, obj_buf);
216         return 0;
217 }
218
219 static void write_rest(void)
220 {
221         unsigned i;
222         for (i = 0; i < nr_objects; i++) {
223                 if (obj_list[i].obj)
224                         check_object(obj_list[i].obj, OBJ_ANY, NULL, NULL);
225         }
226 }
227
228 static void added_object(unsigned nr, enum object_type type,
229                          void *data, unsigned long size);
230
231 /*
232  * Write out nr-th object from the list, now we know the contents
233  * of it.  Under --strict, this buffers structured objects in-core,
234  * to be checked at the end.
235  */
236 static void write_object(unsigned nr, enum object_type type,
237                          void *buf, unsigned long size)
238 {
239         if (!strict) {
240                 if (write_sha1_file(buf, size, typename(type), obj_list[nr].oid.hash) < 0)
241                         die("failed to write object");
242                 added_object(nr, type, buf, size);
243                 free(buf);
244                 obj_list[nr].obj = NULL;
245         } else if (type == OBJ_BLOB) {
246                 struct blob *blob;
247                 if (write_sha1_file(buf, size, typename(type), obj_list[nr].oid.hash) < 0)
248                         die("failed to write object");
249                 added_object(nr, type, buf, size);
250                 free(buf);
251
252                 blob = lookup_blob(&obj_list[nr].oid);
253                 if (blob)
254                         blob->object.flags |= FLAG_WRITTEN;
255                 else
256                         die("invalid blob object");
257                 obj_list[nr].obj = NULL;
258         } else {
259                 struct object *obj;
260                 int eaten;
261                 hash_sha1_file(buf, size, typename(type), obj_list[nr].oid.hash);
262                 added_object(nr, type, buf, size);
263                 obj = parse_object_buffer(&obj_list[nr].oid, type, size, buf,
264                                           &eaten);
265                 if (!obj)
266                         die("invalid %s", typename(type));
267                 add_object_buffer(obj, buf, size);
268                 obj->flags |= FLAG_OPEN;
269                 obj_list[nr].obj = obj;
270         }
271 }
272
273 static void resolve_delta(unsigned nr, enum object_type type,
274                           void *base, unsigned long base_size,
275                           void *delta, unsigned long delta_size)
276 {
277         void *result;
278         unsigned long result_size;
279
280         result = patch_delta(base, base_size,
281                              delta, delta_size,
282                              &result_size);
283         if (!result)
284                 die("failed to apply delta");
285         free(delta);
286         write_object(nr, type, result, result_size);
287 }
288
289 /*
290  * We now know the contents of an object (which is nr-th in the pack);
291  * resolve all the deltified objects that are based on it.
292  */
293 static void added_object(unsigned nr, enum object_type type,
294                          void *data, unsigned long size)
295 {
296         struct delta_info **p = &delta_list;
297         struct delta_info *info;
298
299         while ((info = *p) != NULL) {
300                 if (!oidcmp(&info->base_oid, &obj_list[nr].oid) ||
301                     info->base_offset == obj_list[nr].offset) {
302                         *p = info->next;
303                         p = &delta_list;
304                         resolve_delta(info->nr, type, data, size,
305                                       info->delta, info->size);
306                         free(info);
307                         continue;
308                 }
309                 p = &info->next;
310         }
311 }
312
313 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
314                                    unsigned nr)
315 {
316         void *buf = get_data(size);
317
318         if (!dry_run && buf)
319                 write_object(nr, type, buf, size);
320         else
321                 free(buf);
322 }
323
324 static int resolve_against_held(unsigned nr, const struct object_id *base,
325                                 void *delta_data, unsigned long delta_size)
326 {
327         struct object *obj;
328         struct obj_buffer *obj_buffer;
329         obj = lookup_object(base->hash);
330         if (!obj)
331                 return 0;
332         obj_buffer = lookup_object_buffer(obj);
333         if (!obj_buffer)
334                 return 0;
335         resolve_delta(nr, obj->type, obj_buffer->buffer,
336                       obj_buffer->size, delta_data, delta_size);
337         return 1;
338 }
339
340 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
341                                unsigned nr)
342 {
343         void *delta_data, *base;
344         unsigned long base_size;
345         struct object_id base_oid;
346
347         if (type == OBJ_REF_DELTA) {
348                 hashcpy(base_oid.hash, fill(GIT_SHA1_RAWSZ));
349                 use(GIT_SHA1_RAWSZ);
350                 delta_data = get_data(delta_size);
351                 if (dry_run || !delta_data) {
352                         free(delta_data);
353                         return;
354                 }
355                 if (has_object_file(&base_oid))
356                         ; /* Ok we have this one */
357                 else if (resolve_against_held(nr, &base_oid,
358                                               delta_data, delta_size))
359                         return; /* we are done */
360                 else {
361                         /* cannot resolve yet --- queue it */
362                         oidclr(&obj_list[nr].oid);
363                         add_delta_to_list(nr, &base_oid, 0, delta_data, delta_size);
364                         return;
365                 }
366         } else {
367                 unsigned base_found = 0;
368                 unsigned char *pack, c;
369                 off_t base_offset;
370                 unsigned lo, mid, hi;
371
372                 pack = fill(1);
373                 c = *pack;
374                 use(1);
375                 base_offset = c & 127;
376                 while (c & 128) {
377                         base_offset += 1;
378                         if (!base_offset || MSB(base_offset, 7))
379                                 die("offset value overflow for delta base object");
380                         pack = fill(1);
381                         c = *pack;
382                         use(1);
383                         base_offset = (base_offset << 7) + (c & 127);
384                 }
385                 base_offset = obj_list[nr].offset - base_offset;
386                 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
387                         die("offset value out of bound for delta base object");
388
389                 delta_data = get_data(delta_size);
390                 if (dry_run || !delta_data) {
391                         free(delta_data);
392                         return;
393                 }
394                 lo = 0;
395                 hi = nr;
396                 while (lo < hi) {
397                         mid = (lo + hi)/2;
398                         if (base_offset < obj_list[mid].offset) {
399                                 hi = mid;
400                         } else if (base_offset > obj_list[mid].offset) {
401                                 lo = mid + 1;
402                         } else {
403                                 oidcpy(&base_oid, &obj_list[mid].oid);
404                                 base_found = !is_null_oid(&base_oid);
405                                 break;
406                         }
407                 }
408                 if (!base_found) {
409                         /*
410                          * The delta base object is itself a delta that
411                          * has not been resolved yet.
412                          */
413                         oidclr(&obj_list[nr].oid);
414                         add_delta_to_list(nr, &null_oid, base_offset, delta_data, delta_size);
415                         return;
416                 }
417         }
418
419         if (resolve_against_held(nr, &base_oid, delta_data, delta_size))
420                 return;
421
422         base = read_sha1_file(base_oid.hash, &type, &base_size);
423         if (!base) {
424                 error("failed to read delta-pack base object %s",
425                       oid_to_hex(&base_oid));
426                 if (!recover)
427                         exit(1);
428                 has_errors = 1;
429                 return;
430         }
431         resolve_delta(nr, type, base, base_size, delta_data, delta_size);
432         free(base);
433 }
434
435 static void unpack_one(unsigned nr)
436 {
437         unsigned shift;
438         unsigned char *pack;
439         unsigned long size, c;
440         enum object_type type;
441
442         obj_list[nr].offset = consumed_bytes;
443
444         pack = fill(1);
445         c = *pack;
446         use(1);
447         type = (c >> 4) & 7;
448         size = (c & 15);
449         shift = 4;
450         while (c & 0x80) {
451                 pack = fill(1);
452                 c = *pack;
453                 use(1);
454                 size += (c & 0x7f) << shift;
455                 shift += 7;
456         }
457
458         switch (type) {
459         case OBJ_COMMIT:
460         case OBJ_TREE:
461         case OBJ_BLOB:
462         case OBJ_TAG:
463                 unpack_non_delta_entry(type, size, nr);
464                 return;
465         case OBJ_REF_DELTA:
466         case OBJ_OFS_DELTA:
467                 unpack_delta_entry(type, size, nr);
468                 return;
469         default:
470                 error("bad object type %d", type);
471                 has_errors = 1;
472                 if (recover)
473                         return;
474                 exit(1);
475         }
476 }
477
478 static void unpack_all(void)
479 {
480         int i;
481         struct progress *progress = NULL;
482         struct pack_header *hdr = fill(sizeof(struct pack_header));
483
484         nr_objects = ntohl(hdr->hdr_entries);
485
486         if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
487                 die("bad pack file");
488         if (!pack_version_ok(hdr->hdr_version))
489                 die("unknown pack file version %"PRIu32,
490                         ntohl(hdr->hdr_version));
491         use(sizeof(struct pack_header));
492
493         if (!quiet)
494                 progress = start_progress(_("Unpacking objects"), nr_objects);
495         obj_list = xcalloc(nr_objects, sizeof(*obj_list));
496         for (i = 0; i < nr_objects; i++) {
497                 unpack_one(i);
498                 display_progress(progress, i + 1);
499         }
500         stop_progress(&progress);
501
502         if (delta_list)
503                 die("unresolved deltas left after unpacking");
504 }
505
506 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
507 {
508         int i;
509         struct object_id oid;
510
511         check_replace_refs = 0;
512
513         git_config(git_default_config, NULL);
514
515         quiet = !isatty(2);
516
517         for (i = 1 ; i < argc; i++) {
518                 const char *arg = argv[i];
519
520                 if (*arg == '-') {
521                         if (!strcmp(arg, "-n")) {
522                                 dry_run = 1;
523                                 continue;
524                         }
525                         if (!strcmp(arg, "-q")) {
526                                 quiet = 1;
527                                 continue;
528                         }
529                         if (!strcmp(arg, "-r")) {
530                                 recover = 1;
531                                 continue;
532                         }
533                         if (!strcmp(arg, "--strict")) {
534                                 strict = 1;
535                                 continue;
536                         }
537                         if (skip_prefix(arg, "--strict=", &arg)) {
538                                 strict = 1;
539                                 fsck_set_msg_types(&fsck_options, arg);
540                                 continue;
541                         }
542                         if (starts_with(arg, "--pack_header=")) {
543                                 struct pack_header *hdr;
544                                 char *c;
545
546                                 hdr = (struct pack_header *)buffer;
547                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
548                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
549                                 if (*c != ',')
550                                         die("bad %s", arg);
551                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
552                                 if (*c)
553                                         die("bad %s", arg);
554                                 len = sizeof(*hdr);
555                                 continue;
556                         }
557                         if (skip_prefix(arg, "--max-input-size=", &arg)) {
558                                 max_input_size = strtoumax(arg, NULL, 10);
559                                 continue;
560                         }
561                         usage(unpack_usage);
562                 }
563
564                 /* We don't take any non-flag arguments now.. Maybe some day */
565                 usage(unpack_usage);
566         }
567         git_SHA1_Init(&ctx);
568         unpack_all();
569         git_SHA1_Update(&ctx, buffer, offset);
570         git_SHA1_Final(oid.hash, &ctx);
571         if (strict)
572                 write_rest();
573         if (hashcmp(fill(GIT_SHA1_RAWSZ), oid.hash))
574                 die("final sha1 did not match");
575         use(GIT_SHA1_RAWSZ);
576
577         /* Write the last part of the buffer to stdout */
578         while (len) {
579                 int ret = xwrite(1, buffer + offset, len);
580                 if (ret <= 0)
581                         break;
582                 len -= ret;
583                 offset += ret;
584         }
585
586         /* All done */
587         return has_errors;
588 }