Merge branch 'jc/latin-1'
[git] / builtin / unpack-objects.c
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"
14
15 static int dry_run, quiet, recover, has_errors, strict;
16 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict]";
17
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 off_t max_input_size;
23 static git_SHA_CTX ctx;
24 static struct fsck_options fsck_options = FSCK_OPTIONS_STRICT;
25
26 /*
27  * When running under --strict mode, objects whose reachability are
28  * suspect are kept in core without getting written in the object
29  * store.
30  */
31 struct obj_buffer {
32         char *buffer;
33         unsigned long size;
34 };
35
36 static struct decoration obj_decorate;
37
38 static struct obj_buffer *lookup_object_buffer(struct object *base)
39 {
40         return lookup_decoration(&obj_decorate, base);
41 }
42
43 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
44 {
45         struct obj_buffer *obj;
46         obj = xcalloc(1, sizeof(struct obj_buffer));
47         obj->buffer = buffer;
48         obj->size = size;
49         if (add_decoration(&obj_decorate, object, obj))
50                 die("object %s tried to add buffer twice!", oid_to_hex(&object->oid));
51 }
52
53 /*
54  * Make sure at least "min" bytes are available in the buffer, and
55  * return the pointer to the buffer.
56  */
57 static void *fill(int min)
58 {
59         if (min <= len)
60                 return buffer + offset;
61         if (min > sizeof(buffer))
62                 die("cannot fill %d bytes", min);
63         if (offset) {
64                 git_SHA1_Update(&ctx, buffer, offset);
65                 memmove(buffer, buffer + offset, len);
66                 offset = 0;
67         }
68         do {
69                 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
70                 if (ret <= 0) {
71                         if (!ret)
72                                 die("early EOF");
73                         die_errno("read error on input");
74                 }
75                 len += ret;
76         } while (len < min);
77         return buffer;
78 }
79
80 static void use(int bytes)
81 {
82         if (bytes > len)
83                 die("used more bytes than were available");
84         len -= bytes;
85         offset += bytes;
86
87         /* make sure off_t is sufficiently large not to wrap */
88         if (signed_add_overflows(consumed_bytes, bytes))
89                 die("pack too large for current definition of off_t");
90         consumed_bytes += bytes;
91         if (max_input_size && consumed_bytes > max_input_size)
92                 die(_("pack exceeds maximum allowed size"));
93 }
94
95 static void *get_data(unsigned long size)
96 {
97         git_zstream stream;
98         void *buf = xmallocz(size);
99
100         memset(&stream, 0, sizeof(stream));
101
102         stream.next_out = buf;
103         stream.avail_out = size;
104         stream.next_in = fill(1);
105         stream.avail_in = len;
106         git_inflate_init(&stream);
107
108         for (;;) {
109                 int ret = git_inflate(&stream, 0);
110                 use(len - stream.avail_in);
111                 if (stream.total_out == size && ret == Z_STREAM_END)
112                         break;
113                 if (ret != Z_OK) {
114                         error("inflate returned %d", ret);
115                         free(buf);
116                         buf = NULL;
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         unsigned char base_sha1[20];
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, unsigned const char *base_sha1,
141                               off_t base_offset,
142                               void *delta, unsigned long size)
143 {
144         struct delta_info *info = xmalloc(sizeof(*info));
145
146         hashcpy(info->base_sha1, base_sha1);
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         unsigned char sha1[20];
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         unsigned char sha1[20];
174
175         if (write_sha1_file(obj_buf->buffer, obj_buf->size, typename(obj->type), sha1) < 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].sha1) < 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].sha1) < 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].sha1);
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].sha1);
262                 added_object(nr, type, buf, size);
263                 obj = parse_object_buffer(obj_list[nr].sha1, type, size, buf, &eaten);
264                 if (!obj)
265                         die("invalid %s", typename(type));
266                 add_object_buffer(obj, buf, size);
267                 obj->flags |= FLAG_OPEN;
268                 obj_list[nr].obj = obj;
269         }
270 }
271
272 static void resolve_delta(unsigned nr, enum object_type type,
273                           void *base, unsigned long base_size,
274                           void *delta, unsigned long delta_size)
275 {
276         void *result;
277         unsigned long result_size;
278
279         result = patch_delta(base, base_size,
280                              delta, delta_size,
281                              &result_size);
282         if (!result)
283                 die("failed to apply delta");
284         free(delta);
285         write_object(nr, type, result, result_size);
286 }
287
288 /*
289  * We now know the contents of an object (which is nr-th in the pack);
290  * resolve all the deltified objects that are based on it.
291  */
292 static void added_object(unsigned nr, enum object_type type,
293                          void *data, unsigned long size)
294 {
295         struct delta_info **p = &delta_list;
296         struct delta_info *info;
297
298         while ((info = *p) != NULL) {
299                 if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
300                     info->base_offset == obj_list[nr].offset) {
301                         *p = info->next;
302                         p = &delta_list;
303                         resolve_delta(info->nr, type, data, size,
304                                       info->delta, info->size);
305                         free(info);
306                         continue;
307                 }
308                 p = &info->next;
309         }
310 }
311
312 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
313                                    unsigned nr)
314 {
315         void *buf = get_data(size);
316
317         if (!dry_run && buf)
318                 write_object(nr, type, buf, size);
319         else
320                 free(buf);
321 }
322
323 static int resolve_against_held(unsigned nr, const unsigned char *base,
324                                 void *delta_data, unsigned long delta_size)
325 {
326         struct object *obj;
327         struct obj_buffer *obj_buffer;
328         obj = lookup_object(base);
329         if (!obj)
330                 return 0;
331         obj_buffer = lookup_object_buffer(obj);
332         if (!obj_buffer)
333                 return 0;
334         resolve_delta(nr, obj->type, obj_buffer->buffer,
335                       obj_buffer->size, delta_data, delta_size);
336         return 1;
337 }
338
339 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
340                                unsigned nr)
341 {
342         void *delta_data, *base;
343         unsigned long base_size;
344         unsigned char base_sha1[20];
345
346         if (type == OBJ_REF_DELTA) {
347                 hashcpy(base_sha1, fill(20));
348                 use(20);
349                 delta_data = get_data(delta_size);
350                 if (dry_run || !delta_data) {
351                         free(delta_data);
352                         return;
353                 }
354                 if (has_sha1_file(base_sha1))
355                         ; /* Ok we have this one */
356                 else if (resolve_against_held(nr, base_sha1,
357                                               delta_data, delta_size))
358                         return; /* we are done */
359                 else {
360                         /* cannot resolve yet --- queue it */
361                         hashclr(obj_list[nr].sha1);
362                         add_delta_to_list(nr, base_sha1, 0, delta_data, delta_size);
363                         return;
364                 }
365         } else {
366                 unsigned base_found = 0;
367                 unsigned char *pack, c;
368                 off_t base_offset;
369                 unsigned lo, mid, hi;
370
371                 pack = fill(1);
372                 c = *pack;
373                 use(1);
374                 base_offset = c & 127;
375                 while (c & 128) {
376                         base_offset += 1;
377                         if (!base_offset || MSB(base_offset, 7))
378                                 die("offset value overflow for delta base object");
379                         pack = fill(1);
380                         c = *pack;
381                         use(1);
382                         base_offset = (base_offset << 7) + (c & 127);
383                 }
384                 base_offset = obj_list[nr].offset - base_offset;
385                 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
386                         die("offset value out of bound for delta base object");
387
388                 delta_data = get_data(delta_size);
389                 if (dry_run || !delta_data) {
390                         free(delta_data);
391                         return;
392                 }
393                 lo = 0;
394                 hi = nr;
395                 while (lo < hi) {
396                         mid = (lo + hi)/2;
397                         if (base_offset < obj_list[mid].offset) {
398                                 hi = mid;
399                         } else if (base_offset > obj_list[mid].offset) {
400                                 lo = mid + 1;
401                         } else {
402                                 hashcpy(base_sha1, obj_list[mid].sha1);
403                                 base_found = !is_null_sha1(base_sha1);
404                                 break;
405                         }
406                 }
407                 if (!base_found) {
408                         /*
409                          * The delta base object is itself a delta that
410                          * has not been resolved yet.
411                          */
412                         hashclr(obj_list[nr].sha1);
413                         add_delta_to_list(nr, null_sha1, base_offset, delta_data, delta_size);
414                         return;
415                 }
416         }
417
418         if (resolve_against_held(nr, base_sha1, delta_data, delta_size))
419                 return;
420
421         base = read_sha1_file(base_sha1, &type, &base_size);
422         if (!base) {
423                 error("failed to read delta-pack base object %s",
424                       sha1_to_hex(base_sha1));
425                 if (!recover)
426                         exit(1);
427                 has_errors = 1;
428                 return;
429         }
430         resolve_delta(nr, type, base, base_size, delta_data, delta_size);
431         free(base);
432 }
433
434 static void unpack_one(unsigned nr)
435 {
436         unsigned shift;
437         unsigned char *pack;
438         unsigned long size, c;
439         enum object_type type;
440
441         obj_list[nr].offset = consumed_bytes;
442
443         pack = fill(1);
444         c = *pack;
445         use(1);
446         type = (c >> 4) & 7;
447         size = (c & 15);
448         shift = 4;
449         while (c & 0x80) {
450                 pack = fill(1);
451                 c = *pack;
452                 use(1);
453                 size += (c & 0x7f) << shift;
454                 shift += 7;
455         }
456
457         switch (type) {
458         case OBJ_COMMIT:
459         case OBJ_TREE:
460         case OBJ_BLOB:
461         case OBJ_TAG:
462                 unpack_non_delta_entry(type, size, nr);
463                 return;
464         case OBJ_REF_DELTA:
465         case OBJ_OFS_DELTA:
466                 unpack_delta_entry(type, size, nr);
467                 return;
468         default:
469                 error("bad object type %d", type);
470                 has_errors = 1;
471                 if (recover)
472                         return;
473                 exit(1);
474         }
475 }
476
477 static void unpack_all(void)
478 {
479         int i;
480         struct progress *progress = NULL;
481         struct pack_header *hdr = fill(sizeof(struct pack_header));
482
483         nr_objects = ntohl(hdr->hdr_entries);
484
485         if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
486                 die("bad pack file");
487         if (!pack_version_ok(hdr->hdr_version))
488                 die("unknown pack file version %"PRIu32,
489                         ntohl(hdr->hdr_version));
490         use(sizeof(struct pack_header));
491
492         if (!quiet)
493                 progress = start_progress(_("Unpacking objects"), nr_objects);
494         obj_list = xcalloc(nr_objects, sizeof(*obj_list));
495         for (i = 0; i < nr_objects; i++) {
496                 unpack_one(i);
497                 display_progress(progress, i + 1);
498         }
499         stop_progress(&progress);
500
501         if (delta_list)
502                 die("unresolved deltas left after unpacking");
503 }
504
505 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
506 {
507         int i;
508         unsigned char sha1[20];
509
510         check_replace_refs = 0;
511
512         git_config(git_default_config, NULL);
513
514         quiet = !isatty(2);
515
516         for (i = 1 ; i < argc; i++) {
517                 const char *arg = argv[i];
518
519                 if (*arg == '-') {
520                         if (!strcmp(arg, "-n")) {
521                                 dry_run = 1;
522                                 continue;
523                         }
524                         if (!strcmp(arg, "-q")) {
525                                 quiet = 1;
526                                 continue;
527                         }
528                         if (!strcmp(arg, "-r")) {
529                                 recover = 1;
530                                 continue;
531                         }
532                         if (!strcmp(arg, "--strict")) {
533                                 strict = 1;
534                                 continue;
535                         }
536                         if (skip_prefix(arg, "--strict=", &arg)) {
537                                 strict = 1;
538                                 fsck_set_msg_types(&fsck_options, arg);
539                                 continue;
540                         }
541                         if (starts_with(arg, "--pack_header=")) {
542                                 struct pack_header *hdr;
543                                 char *c;
544
545                                 hdr = (struct pack_header *)buffer;
546                                 hdr->hdr_signature = htonl(PACK_SIGNATURE);
547                                 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
548                                 if (*c != ',')
549                                         die("bad %s", arg);
550                                 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
551                                 if (*c)
552                                         die("bad %s", arg);
553                                 len = sizeof(*hdr);
554                                 continue;
555                         }
556                         if (skip_prefix(arg, "--max-input-size=", &arg)) {
557                                 max_input_size = strtoumax(arg, NULL, 10);
558                                 continue;
559                         }
560                         usage(unpack_usage);
561                 }
562
563                 /* We don't take any non-flag arguments now.. Maybe some day */
564                 usage(unpack_usage);
565         }
566         git_SHA1_Init(&ctx);
567         unpack_all();
568         git_SHA1_Update(&ctx, buffer, offset);
569         git_SHA1_Final(sha1, &ctx);
570         if (strict)
571                 write_rest();
572         if (hashcmp(fill(20), sha1))
573                 die("final sha1 did not match");
574         use(20);
575
576         /* Write the last part of the buffer to stdout */
577         while (len) {
578                 int ret = xwrite(1, buffer + offset, len);
579                 if (ret <= 0)
580                         break;
581                 len -= ret;
582                 offset += ret;
583         }
584
585         /* All done */
586         return has_errors;
587 }