multi-pack-index: add 'verify' verb
[git] / fsck.c
1 #include "cache.h"
2 #include "object-store.h"
3 #include "repository.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "tree.h"
7 #include "tree-walk.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "fsck.h"
11 #include "refs.h"
12 #include "utf8.h"
13 #include "sha1-array.h"
14 #include "decorate.h"
15 #include "oidset.h"
16 #include "packfile.h"
17 #include "submodule-config.h"
18 #include "config.h"
19 #include "help.h"
20
21 static struct oidset gitmodules_found = OIDSET_INIT;
22 static struct oidset gitmodules_done = OIDSET_INIT;
23
24 #define FSCK_FATAL -1
25 #define FSCK_INFO -2
26
27 #define FOREACH_MSG_ID(FUNC) \
28         /* fatal errors */ \
29         FUNC(NUL_IN_HEADER, FATAL) \
30         FUNC(UNTERMINATED_HEADER, FATAL) \
31         /* errors */ \
32         FUNC(BAD_DATE, ERROR) \
33         FUNC(BAD_DATE_OVERFLOW, ERROR) \
34         FUNC(BAD_EMAIL, ERROR) \
35         FUNC(BAD_NAME, ERROR) \
36         FUNC(BAD_OBJECT_SHA1, ERROR) \
37         FUNC(BAD_PARENT_SHA1, ERROR) \
38         FUNC(BAD_TAG_OBJECT, ERROR) \
39         FUNC(BAD_TIMEZONE, ERROR) \
40         FUNC(BAD_TREE, ERROR) \
41         FUNC(BAD_TREE_SHA1, ERROR) \
42         FUNC(BAD_TYPE, ERROR) \
43         FUNC(DUPLICATE_ENTRIES, ERROR) \
44         FUNC(MISSING_AUTHOR, ERROR) \
45         FUNC(MISSING_COMMITTER, ERROR) \
46         FUNC(MISSING_EMAIL, ERROR) \
47         FUNC(MISSING_GRAFT, ERROR) \
48         FUNC(MISSING_NAME_BEFORE_EMAIL, ERROR) \
49         FUNC(MISSING_OBJECT, ERROR) \
50         FUNC(MISSING_PARENT, ERROR) \
51         FUNC(MISSING_SPACE_BEFORE_DATE, ERROR) \
52         FUNC(MISSING_SPACE_BEFORE_EMAIL, ERROR) \
53         FUNC(MISSING_TAG, ERROR) \
54         FUNC(MISSING_TAG_ENTRY, ERROR) \
55         FUNC(MISSING_TAG_OBJECT, ERROR) \
56         FUNC(MISSING_TREE, ERROR) \
57         FUNC(MISSING_TREE_OBJECT, ERROR) \
58         FUNC(MISSING_TYPE, ERROR) \
59         FUNC(MISSING_TYPE_ENTRY, ERROR) \
60         FUNC(MULTIPLE_AUTHORS, ERROR) \
61         FUNC(TAG_OBJECT_NOT_TAG, ERROR) \
62         FUNC(TREE_NOT_SORTED, ERROR) \
63         FUNC(UNKNOWN_TYPE, ERROR) \
64         FUNC(ZERO_PADDED_DATE, ERROR) \
65         FUNC(GITMODULES_MISSING, ERROR) \
66         FUNC(GITMODULES_BLOB, ERROR) \
67         FUNC(GITMODULES_LARGE, ERROR) \
68         FUNC(GITMODULES_NAME, ERROR) \
69         FUNC(GITMODULES_SYMLINK, ERROR) \
70         /* warnings */ \
71         FUNC(BAD_FILEMODE, WARN) \
72         FUNC(EMPTY_NAME, WARN) \
73         FUNC(FULL_PATHNAME, WARN) \
74         FUNC(HAS_DOT, WARN) \
75         FUNC(HAS_DOTDOT, WARN) \
76         FUNC(HAS_DOTGIT, WARN) \
77         FUNC(NULL_SHA1, WARN) \
78         FUNC(ZERO_PADDED_FILEMODE, WARN) \
79         FUNC(NUL_IN_COMMIT, WARN) \
80         /* infos (reported as warnings, but ignored by default) */ \
81         FUNC(GITMODULES_PARSE, INFO) \
82         FUNC(BAD_TAG_NAME, INFO) \
83         FUNC(MISSING_TAGGER_ENTRY, INFO)
84
85 #define MSG_ID(id, msg_type) FSCK_MSG_##id,
86 enum fsck_msg_id {
87         FOREACH_MSG_ID(MSG_ID)
88         FSCK_MSG_MAX
89 };
90 #undef MSG_ID
91
92 #define STR(x) #x
93 #define MSG_ID(id, msg_type) { STR(id), NULL, NULL, FSCK_##msg_type },
94 static struct {
95         const char *id_string;
96         const char *downcased;
97         const char *camelcased;
98         int msg_type;
99 } msg_id_info[FSCK_MSG_MAX + 1] = {
100         FOREACH_MSG_ID(MSG_ID)
101         { NULL, NULL, NULL, -1 }
102 };
103 #undef MSG_ID
104
105 static void prepare_msg_ids(void)
106 {
107         int i;
108
109         if (msg_id_info[0].downcased)
110                 return;
111
112         /* convert id_string to lower case, without underscores. */
113         for (i = 0; i < FSCK_MSG_MAX; i++) {
114                 const char *p = msg_id_info[i].id_string;
115                 int len = strlen(p);
116                 char *q = xmalloc(len);
117
118                 msg_id_info[i].downcased = q;
119                 while (*p)
120                         if (*p == '_')
121                                 p++;
122                         else
123                                 *(q)++ = tolower(*(p)++);
124                 *q = '\0';
125
126                 p = msg_id_info[i].id_string;
127                 q = xmalloc(len);
128                 msg_id_info[i].camelcased = q;
129                 while (*p) {
130                         if (*p == '_') {
131                                 p++;
132                                 if (*p)
133                                         *q++ = *p++;
134                         } else {
135                                 *q++ = tolower(*p++);
136                         }
137                 }
138                 *q = '\0';
139         }
140 }
141
142 static int parse_msg_id(const char *text)
143 {
144         int i;
145
146         prepare_msg_ids();
147
148         for (i = 0; i < FSCK_MSG_MAX; i++)
149                 if (!strcmp(text, msg_id_info[i].downcased))
150                         return i;
151
152         return -1;
153 }
154
155 void list_config_fsck_msg_ids(struct string_list *list, const char *prefix)
156 {
157         int i;
158
159         prepare_msg_ids();
160
161         for (i = 0; i < FSCK_MSG_MAX; i++)
162                 list_config_item(list, prefix, msg_id_info[i].camelcased);
163 }
164
165 static int fsck_msg_type(enum fsck_msg_id msg_id,
166         struct fsck_options *options)
167 {
168         int msg_type;
169
170         assert(msg_id >= 0 && msg_id < FSCK_MSG_MAX);
171
172         if (options->msg_type)
173                 msg_type = options->msg_type[msg_id];
174         else {
175                 msg_type = msg_id_info[msg_id].msg_type;
176                 if (options->strict && msg_type == FSCK_WARN)
177                         msg_type = FSCK_ERROR;
178         }
179
180         return msg_type;
181 }
182
183 static void init_skiplist(struct fsck_options *options, const char *path)
184 {
185         static struct oid_array skiplist = OID_ARRAY_INIT;
186         int sorted, fd;
187         char buffer[GIT_MAX_HEXSZ + 1];
188         struct object_id oid;
189
190         if (options->skiplist)
191                 sorted = options->skiplist->sorted;
192         else {
193                 sorted = 1;
194                 options->skiplist = &skiplist;
195         }
196
197         fd = open(path, O_RDONLY);
198         if (fd < 0)
199                 die("Could not open skip list: %s", path);
200         for (;;) {
201                 const char *p;
202                 int result = read_in_full(fd, buffer, sizeof(buffer));
203                 if (result < 0)
204                         die_errno("Could not read '%s'", path);
205                 if (!result)
206                         break;
207                 if (parse_oid_hex(buffer, &oid, &p) || *p != '\n')
208                         die("Invalid SHA-1: %s", buffer);
209                 oid_array_append(&skiplist, &oid);
210                 if (sorted && skiplist.nr > 1 &&
211                                 oidcmp(&skiplist.oid[skiplist.nr - 2],
212                                        &oid) > 0)
213                         sorted = 0;
214         }
215         close(fd);
216
217         if (sorted)
218                 skiplist.sorted = 1;
219 }
220
221 static int parse_msg_type(const char *str)
222 {
223         if (!strcmp(str, "error"))
224                 return FSCK_ERROR;
225         else if (!strcmp(str, "warn"))
226                 return FSCK_WARN;
227         else if (!strcmp(str, "ignore"))
228                 return FSCK_IGNORE;
229         else
230                 die("Unknown fsck message type: '%s'", str);
231 }
232
233 int is_valid_msg_type(const char *msg_id, const char *msg_type)
234 {
235         if (parse_msg_id(msg_id) < 0)
236                 return 0;
237         parse_msg_type(msg_type);
238         return 1;
239 }
240
241 void fsck_set_msg_type(struct fsck_options *options,
242                 const char *msg_id, const char *msg_type)
243 {
244         int id = parse_msg_id(msg_id), type;
245
246         if (id < 0)
247                 die("Unhandled message id: %s", msg_id);
248         type = parse_msg_type(msg_type);
249
250         if (type != FSCK_ERROR && msg_id_info[id].msg_type == FSCK_FATAL)
251                 die("Cannot demote %s to %s", msg_id, msg_type);
252
253         if (!options->msg_type) {
254                 int i;
255                 int *msg_type;
256                 ALLOC_ARRAY(msg_type, FSCK_MSG_MAX);
257                 for (i = 0; i < FSCK_MSG_MAX; i++)
258                         msg_type[i] = fsck_msg_type(i, options);
259                 options->msg_type = msg_type;
260         }
261
262         options->msg_type[id] = type;
263 }
264
265 void fsck_set_msg_types(struct fsck_options *options, const char *values)
266 {
267         char *buf = xstrdup(values), *to_free = buf;
268         int done = 0;
269
270         while (!done) {
271                 int len = strcspn(buf, " ,|"), equal;
272
273                 done = !buf[len];
274                 if (!len) {
275                         buf++;
276                         continue;
277                 }
278                 buf[len] = '\0';
279
280                 for (equal = 0;
281                      equal < len && buf[equal] != '=' && buf[equal] != ':';
282                      equal++)
283                         buf[equal] = tolower(buf[equal]);
284                 buf[equal] = '\0';
285
286                 if (!strcmp(buf, "skiplist")) {
287                         if (equal == len)
288                                 die("skiplist requires a path");
289                         init_skiplist(options, buf + equal + 1);
290                         buf += len + 1;
291                         continue;
292                 }
293
294                 if (equal == len)
295                         die("Missing '=': '%s'", buf);
296
297                 fsck_set_msg_type(options, buf, buf + equal + 1);
298                 buf += len + 1;
299         }
300         free(to_free);
301 }
302
303 static void append_msg_id(struct strbuf *sb, const char *msg_id)
304 {
305         for (;;) {
306                 char c = *(msg_id)++;
307
308                 if (!c)
309                         break;
310                 if (c != '_')
311                         strbuf_addch(sb, tolower(c));
312                 else {
313                         assert(*msg_id);
314                         strbuf_addch(sb, *(msg_id)++);
315                 }
316         }
317
318         strbuf_addstr(sb, ": ");
319 }
320
321 static int object_on_skiplist(struct fsck_options *opts, struct object *obj)
322 {
323         if (opts && opts->skiplist && obj)
324                 return oid_array_lookup(opts->skiplist, &obj->oid) >= 0;
325         return 0;
326 }
327
328 __attribute__((format (printf, 4, 5)))
329 static int report(struct fsck_options *options, struct object *object,
330         enum fsck_msg_id id, const char *fmt, ...)
331 {
332         va_list ap;
333         struct strbuf sb = STRBUF_INIT;
334         int msg_type = fsck_msg_type(id, options), result;
335
336         if (msg_type == FSCK_IGNORE)
337                 return 0;
338
339         if (object_on_skiplist(options, object))
340                 return 0;
341
342         if (msg_type == FSCK_FATAL)
343                 msg_type = FSCK_ERROR;
344         else if (msg_type == FSCK_INFO)
345                 msg_type = FSCK_WARN;
346
347         append_msg_id(&sb, msg_id_info[id].id_string);
348
349         va_start(ap, fmt);
350         strbuf_vaddf(&sb, fmt, ap);
351         result = options->error_func(options, object, msg_type, sb.buf);
352         strbuf_release(&sb);
353         va_end(ap);
354
355         return result;
356 }
357
358 static char *get_object_name(struct fsck_options *options, struct object *obj)
359 {
360         if (!options->object_names)
361                 return NULL;
362         return lookup_decoration(options->object_names, obj);
363 }
364
365 static void put_object_name(struct fsck_options *options, struct object *obj,
366         const char *fmt, ...)
367 {
368         va_list ap;
369         struct strbuf buf = STRBUF_INIT;
370         char *existing;
371
372         if (!options->object_names)
373                 return;
374         existing = lookup_decoration(options->object_names, obj);
375         if (existing)
376                 return;
377         va_start(ap, fmt);
378         strbuf_vaddf(&buf, fmt, ap);
379         add_decoration(options->object_names, obj, strbuf_detach(&buf, NULL));
380         va_end(ap);
381 }
382
383 static const char *describe_object(struct fsck_options *o, struct object *obj)
384 {
385         static struct strbuf buf = STRBUF_INIT;
386         char *name;
387
388         strbuf_reset(&buf);
389         strbuf_addstr(&buf, oid_to_hex(&obj->oid));
390         if (o->object_names && (name = lookup_decoration(o->object_names, obj)))
391                 strbuf_addf(&buf, " (%s)", name);
392
393         return buf.buf;
394 }
395
396 static int fsck_walk_tree(struct tree *tree, void *data, struct fsck_options *options)
397 {
398         struct tree_desc desc;
399         struct name_entry entry;
400         int res = 0;
401         const char *name;
402
403         if (parse_tree(tree))
404                 return -1;
405
406         name = get_object_name(options, &tree->object);
407         if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
408                 return -1;
409         while (tree_entry_gently(&desc, &entry)) {
410                 struct object *obj;
411                 int result;
412
413                 if (S_ISGITLINK(entry.mode))
414                         continue;
415
416                 if (S_ISDIR(entry.mode)) {
417                         obj = (struct object *)lookup_tree(the_repository, entry.oid);
418                         if (name && obj)
419                                 put_object_name(options, obj, "%s%s/", name,
420                                         entry.path);
421                         result = options->walk(obj, OBJ_TREE, data, options);
422                 }
423                 else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode)) {
424                         obj = (struct object *)lookup_blob(the_repository, entry.oid);
425                         if (name && obj)
426                                 put_object_name(options, obj, "%s%s", name,
427                                         entry.path);
428                         result = options->walk(obj, OBJ_BLOB, data, options);
429                 }
430                 else {
431                         result = error("in tree %s: entry %s has bad mode %.6o",
432                                         describe_object(options, &tree->object), entry.path, entry.mode);
433                 }
434                 if (result < 0)
435                         return result;
436                 if (!res)
437                         res = result;
438         }
439         return res;
440 }
441
442 static int fsck_walk_commit(struct commit *commit, void *data, struct fsck_options *options)
443 {
444         int counter = 0, generation = 0, name_prefix_len = 0;
445         struct commit_list *parents;
446         int res;
447         int result;
448         const char *name;
449
450         if (parse_commit(commit))
451                 return -1;
452
453         name = get_object_name(options, &commit->object);
454         if (name)
455                 put_object_name(options, &get_commit_tree(commit)->object,
456                                 "%s:", name);
457
458         result = options->walk((struct object *)get_commit_tree(commit),
459                                OBJ_TREE, data, options);
460         if (result < 0)
461                 return result;
462         res = result;
463
464         parents = commit->parents;
465         if (name && parents) {
466                 int len = strlen(name), power;
467
468                 if (len && name[len - 1] == '^') {
469                         generation = 1;
470                         name_prefix_len = len - 1;
471                 }
472                 else { /* parse ~<generation> suffix */
473                         for (generation = 0, power = 1;
474                              len && isdigit(name[len - 1]);
475                              power *= 10)
476                                 generation += power * (name[--len] - '0');
477                         if (power > 1 && len && name[len - 1] == '~')
478                                 name_prefix_len = len - 1;
479                 }
480         }
481
482         while (parents) {
483                 if (name) {
484                         struct object *obj = &parents->item->object;
485
486                         if (++counter > 1)
487                                 put_object_name(options, obj, "%s^%d",
488                                         name, counter);
489                         else if (generation > 0)
490                                 put_object_name(options, obj, "%.*s~%d",
491                                         name_prefix_len, name, generation + 1);
492                         else
493                                 put_object_name(options, obj, "%s^", name);
494                 }
495                 result = options->walk((struct object *)parents->item, OBJ_COMMIT, data, options);
496                 if (result < 0)
497                         return result;
498                 if (!res)
499                         res = result;
500                 parents = parents->next;
501         }
502         return res;
503 }
504
505 static int fsck_walk_tag(struct tag *tag, void *data, struct fsck_options *options)
506 {
507         char *name = get_object_name(options, &tag->object);
508
509         if (parse_tag(tag))
510                 return -1;
511         if (name)
512                 put_object_name(options, tag->tagged, "%s", name);
513         return options->walk(tag->tagged, OBJ_ANY, data, options);
514 }
515
516 int fsck_walk(struct object *obj, void *data, struct fsck_options *options)
517 {
518         if (!obj)
519                 return -1;
520
521         if (obj->type == OBJ_NONE)
522                 parse_object(the_repository, &obj->oid);
523
524         switch (obj->type) {
525         case OBJ_BLOB:
526                 return 0;
527         case OBJ_TREE:
528                 return fsck_walk_tree((struct tree *)obj, data, options);
529         case OBJ_COMMIT:
530                 return fsck_walk_commit((struct commit *)obj, data, options);
531         case OBJ_TAG:
532                 return fsck_walk_tag((struct tag *)obj, data, options);
533         default:
534                 error("Unknown object type for %s", describe_object(options, obj));
535                 return -1;
536         }
537 }
538
539 /*
540  * The entries in a tree are ordered in the _path_ order,
541  * which means that a directory entry is ordered by adding
542  * a slash to the end of it.
543  *
544  * So a directory called "a" is ordered _after_ a file
545  * called "a.c", because "a/" sorts after "a.c".
546  */
547 #define TREE_UNORDERED (-1)
548 #define TREE_HAS_DUPS  (-2)
549
550 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
551 {
552         int len1 = strlen(name1);
553         int len2 = strlen(name2);
554         int len = len1 < len2 ? len1 : len2;
555         unsigned char c1, c2;
556         int cmp;
557
558         cmp = memcmp(name1, name2, len);
559         if (cmp < 0)
560                 return 0;
561         if (cmp > 0)
562                 return TREE_UNORDERED;
563
564         /*
565          * Ok, the first <len> characters are the same.
566          * Now we need to order the next one, but turn
567          * a '\0' into a '/' for a directory entry.
568          */
569         c1 = name1[len];
570         c2 = name2[len];
571         if (!c1 && !c2)
572                 /*
573                  * git-write-tree used to write out a nonsense tree that has
574                  * entries with the same name, one blob and one tree.  Make
575                  * sure we do not have duplicate entries.
576                  */
577                 return TREE_HAS_DUPS;
578         if (!c1 && S_ISDIR(mode1))
579                 c1 = '/';
580         if (!c2 && S_ISDIR(mode2))
581                 c2 = '/';
582         return c1 < c2 ? 0 : TREE_UNORDERED;
583 }
584
585 static int fsck_tree(struct tree *item, struct fsck_options *options)
586 {
587         int retval = 0;
588         int has_null_sha1 = 0;
589         int has_full_path = 0;
590         int has_empty_name = 0;
591         int has_dot = 0;
592         int has_dotdot = 0;
593         int has_dotgit = 0;
594         int has_zero_pad = 0;
595         int has_bad_modes = 0;
596         int has_dup_entries = 0;
597         int not_properly_sorted = 0;
598         struct tree_desc desc;
599         unsigned o_mode;
600         const char *o_name;
601
602         if (init_tree_desc_gently(&desc, item->buffer, item->size)) {
603                 retval += report(options, &item->object, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
604                 return retval;
605         }
606
607         o_mode = 0;
608         o_name = NULL;
609
610         while (desc.size) {
611                 unsigned mode;
612                 const char *name;
613                 const struct object_id *oid;
614
615                 oid = tree_entry_extract(&desc, &name, &mode);
616
617                 has_null_sha1 |= is_null_oid(oid);
618                 has_full_path |= !!strchr(name, '/');
619                 has_empty_name |= !*name;
620                 has_dot |= !strcmp(name, ".");
621                 has_dotdot |= !strcmp(name, "..");
622                 has_dotgit |= is_hfs_dotgit(name) || is_ntfs_dotgit(name);
623                 has_zero_pad |= *(char *)desc.buffer == '0';
624
625                 if (is_hfs_dotgitmodules(name) || is_ntfs_dotgitmodules(name)) {
626                         if (!S_ISLNK(mode))
627                                 oidset_insert(&gitmodules_found, oid);
628                         else
629                                 retval += report(options, &item->object,
630                                                  FSCK_MSG_GITMODULES_SYMLINK,
631                                                  ".gitmodules is a symbolic link");
632                 }
633
634                 if (update_tree_entry_gently(&desc)) {
635                         retval += report(options, &item->object, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
636                         break;
637                 }
638
639                 switch (mode) {
640                 /*
641                  * Standard modes..
642                  */
643                 case S_IFREG | 0755:
644                 case S_IFREG | 0644:
645                 case S_IFLNK:
646                 case S_IFDIR:
647                 case S_IFGITLINK:
648                         break;
649                 /*
650                  * This is nonstandard, but we had a few of these
651                  * early on when we honored the full set of mode
652                  * bits..
653                  */
654                 case S_IFREG | 0664:
655                         if (!options->strict)
656                                 break;
657                         /* fallthrough */
658                 default:
659                         has_bad_modes = 1;
660                 }
661
662                 if (o_name) {
663                         switch (verify_ordered(o_mode, o_name, mode, name)) {
664                         case TREE_UNORDERED:
665                                 not_properly_sorted = 1;
666                                 break;
667                         case TREE_HAS_DUPS:
668                                 has_dup_entries = 1;
669                                 break;
670                         default:
671                                 break;
672                         }
673                 }
674
675                 o_mode = mode;
676                 o_name = name;
677         }
678
679         if (has_null_sha1)
680                 retval += report(options, &item->object, FSCK_MSG_NULL_SHA1, "contains entries pointing to null sha1");
681         if (has_full_path)
682                 retval += report(options, &item->object, FSCK_MSG_FULL_PATHNAME, "contains full pathnames");
683         if (has_empty_name)
684                 retval += report(options, &item->object, FSCK_MSG_EMPTY_NAME, "contains empty pathname");
685         if (has_dot)
686                 retval += report(options, &item->object, FSCK_MSG_HAS_DOT, "contains '.'");
687         if (has_dotdot)
688                 retval += report(options, &item->object, FSCK_MSG_HAS_DOTDOT, "contains '..'");
689         if (has_dotgit)
690                 retval += report(options, &item->object, FSCK_MSG_HAS_DOTGIT, "contains '.git'");
691         if (has_zero_pad)
692                 retval += report(options, &item->object, FSCK_MSG_ZERO_PADDED_FILEMODE, "contains zero-padded file modes");
693         if (has_bad_modes)
694                 retval += report(options, &item->object, FSCK_MSG_BAD_FILEMODE, "contains bad file modes");
695         if (has_dup_entries)
696                 retval += report(options, &item->object, FSCK_MSG_DUPLICATE_ENTRIES, "contains duplicate file entries");
697         if (not_properly_sorted)
698                 retval += report(options, &item->object, FSCK_MSG_TREE_NOT_SORTED, "not properly sorted");
699         return retval;
700 }
701
702 static int verify_headers(const void *data, unsigned long size,
703                           struct object *obj, struct fsck_options *options)
704 {
705         const char *buffer = (const char *)data;
706         unsigned long i;
707
708         for (i = 0; i < size; i++) {
709                 switch (buffer[i]) {
710                 case '\0':
711                         return report(options, obj,
712                                 FSCK_MSG_NUL_IN_HEADER,
713                                 "unterminated header: NUL at offset %ld", i);
714                 case '\n':
715                         if (i + 1 < size && buffer[i + 1] == '\n')
716                                 return 0;
717                 }
718         }
719
720         /*
721          * We did not find double-LF that separates the header
722          * and the body.  Not having a body is not a crime but
723          * we do want to see the terminating LF for the last header
724          * line.
725          */
726         if (size && buffer[size - 1] == '\n')
727                 return 0;
728
729         return report(options, obj,
730                 FSCK_MSG_UNTERMINATED_HEADER, "unterminated header");
731 }
732
733 static int fsck_ident(const char **ident, struct object *obj, struct fsck_options *options)
734 {
735         const char *p = *ident;
736         char *end;
737
738         *ident = strchrnul(*ident, '\n');
739         if (**ident == '\n')
740                 (*ident)++;
741
742         if (*p == '<')
743                 return report(options, obj, FSCK_MSG_MISSING_NAME_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
744         p += strcspn(p, "<>\n");
745         if (*p == '>')
746                 return report(options, obj, FSCK_MSG_BAD_NAME, "invalid author/committer line - bad name");
747         if (*p != '<')
748                 return report(options, obj, FSCK_MSG_MISSING_EMAIL, "invalid author/committer line - missing email");
749         if (p[-1] != ' ')
750                 return report(options, obj, FSCK_MSG_MISSING_SPACE_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
751         p++;
752         p += strcspn(p, "<>\n");
753         if (*p != '>')
754                 return report(options, obj, FSCK_MSG_BAD_EMAIL, "invalid author/committer line - bad email");
755         p++;
756         if (*p != ' ')
757                 return report(options, obj, FSCK_MSG_MISSING_SPACE_BEFORE_DATE, "invalid author/committer line - missing space before date");
758         p++;
759         if (*p == '0' && p[1] != ' ')
760                 return report(options, obj, FSCK_MSG_ZERO_PADDED_DATE, "invalid author/committer line - zero-padded date");
761         if (date_overflows(parse_timestamp(p, &end, 10)))
762                 return report(options, obj, FSCK_MSG_BAD_DATE_OVERFLOW, "invalid author/committer line - date causes integer overflow");
763         if ((end == p || *end != ' '))
764                 return report(options, obj, FSCK_MSG_BAD_DATE, "invalid author/committer line - bad date");
765         p = end + 1;
766         if ((*p != '+' && *p != '-') ||
767             !isdigit(p[1]) ||
768             !isdigit(p[2]) ||
769             !isdigit(p[3]) ||
770             !isdigit(p[4]) ||
771             (p[5] != '\n'))
772                 return report(options, obj, FSCK_MSG_BAD_TIMEZONE, "invalid author/committer line - bad time zone");
773         p += 6;
774         return 0;
775 }
776
777 static int fsck_commit_buffer(struct commit *commit, const char *buffer,
778         unsigned long size, struct fsck_options *options)
779 {
780         struct object_id tree_oid, oid;
781         struct commit_graft *graft;
782         unsigned parent_count, parent_line_count = 0, author_count;
783         int err;
784         const char *buffer_begin = buffer;
785         const char *p;
786
787         if (verify_headers(buffer, size, &commit->object, options))
788                 return -1;
789
790         if (!skip_prefix(buffer, "tree ", &buffer))
791                 return report(options, &commit->object, FSCK_MSG_MISSING_TREE, "invalid format - expected 'tree' line");
792         if (parse_oid_hex(buffer, &tree_oid, &p) || *p != '\n') {
793                 err = report(options, &commit->object, FSCK_MSG_BAD_TREE_SHA1, "invalid 'tree' line format - bad sha1");
794                 if (err)
795                         return err;
796         }
797         buffer = p + 1;
798         while (skip_prefix(buffer, "parent ", &buffer)) {
799                 if (parse_oid_hex(buffer, &oid, &p) || *p != '\n') {
800                         err = report(options, &commit->object, FSCK_MSG_BAD_PARENT_SHA1, "invalid 'parent' line format - bad sha1");
801                         if (err)
802                                 return err;
803                 }
804                 buffer = p + 1;
805                 parent_line_count++;
806         }
807         graft = lookup_commit_graft(the_repository, &commit->object.oid);
808         parent_count = commit_list_count(commit->parents);
809         if (graft) {
810                 if (graft->nr_parent == -1 && !parent_count)
811                         ; /* shallow commit */
812                 else if (graft->nr_parent != parent_count) {
813                         err = report(options, &commit->object, FSCK_MSG_MISSING_GRAFT, "graft objects missing");
814                         if (err)
815                                 return err;
816                 }
817         } else {
818                 if (parent_count != parent_line_count) {
819                         err = report(options, &commit->object, FSCK_MSG_MISSING_PARENT, "parent objects missing");
820                         if (err)
821                                 return err;
822                 }
823         }
824         author_count = 0;
825         while (skip_prefix(buffer, "author ", &buffer)) {
826                 author_count++;
827                 err = fsck_ident(&buffer, &commit->object, options);
828                 if (err)
829                         return err;
830         }
831         if (author_count < 1)
832                 err = report(options, &commit->object, FSCK_MSG_MISSING_AUTHOR, "invalid format - expected 'author' line");
833         else if (author_count > 1)
834                 err = report(options, &commit->object, FSCK_MSG_MULTIPLE_AUTHORS, "invalid format - multiple 'author' lines");
835         if (err)
836                 return err;
837         if (!skip_prefix(buffer, "committer ", &buffer))
838                 return report(options, &commit->object, FSCK_MSG_MISSING_COMMITTER, "invalid format - expected 'committer' line");
839         err = fsck_ident(&buffer, &commit->object, options);
840         if (err)
841                 return err;
842         if (!get_commit_tree(commit)) {
843                 err = report(options, &commit->object, FSCK_MSG_BAD_TREE, "could not load commit's tree %s", oid_to_hex(&tree_oid));
844                 if (err)
845                         return err;
846         }
847         if (memchr(buffer_begin, '\0', size)) {
848                 err = report(options, &commit->object, FSCK_MSG_NUL_IN_COMMIT,
849                              "NUL byte in the commit object body");
850                 if (err)
851                         return err;
852         }
853         return 0;
854 }
855
856 static int fsck_commit(struct commit *commit, const char *data,
857         unsigned long size, struct fsck_options *options)
858 {
859         const char *buffer = data ?  data : get_commit_buffer(commit, &size);
860         int ret = fsck_commit_buffer(commit, buffer, size, options);
861         if (!data)
862                 unuse_commit_buffer(commit, buffer);
863         return ret;
864 }
865
866 static int fsck_tag_buffer(struct tag *tag, const char *data,
867         unsigned long size, struct fsck_options *options)
868 {
869         struct object_id oid;
870         int ret = 0;
871         const char *buffer;
872         char *to_free = NULL, *eol;
873         struct strbuf sb = STRBUF_INIT;
874         const char *p;
875
876         if (data)
877                 buffer = data;
878         else {
879                 enum object_type type;
880
881                 buffer = to_free =
882                         read_object_file(&tag->object.oid, &type, &size);
883                 if (!buffer)
884                         return report(options, &tag->object,
885                                 FSCK_MSG_MISSING_TAG_OBJECT,
886                                 "cannot read tag object");
887
888                 if (type != OBJ_TAG) {
889                         ret = report(options, &tag->object,
890                                 FSCK_MSG_TAG_OBJECT_NOT_TAG,
891                                 "expected tag got %s",
892                             type_name(type));
893                         goto done;
894                 }
895         }
896
897         ret = verify_headers(buffer, size, &tag->object, options);
898         if (ret)
899                 goto done;
900
901         if (!skip_prefix(buffer, "object ", &buffer)) {
902                 ret = report(options, &tag->object, FSCK_MSG_MISSING_OBJECT, "invalid format - expected 'object' line");
903                 goto done;
904         }
905         if (parse_oid_hex(buffer, &oid, &p) || *p != '\n') {
906                 ret = report(options, &tag->object, FSCK_MSG_BAD_OBJECT_SHA1, "invalid 'object' line format - bad sha1");
907                 if (ret)
908                         goto done;
909         }
910         buffer = p + 1;
911
912         if (!skip_prefix(buffer, "type ", &buffer)) {
913                 ret = report(options, &tag->object, FSCK_MSG_MISSING_TYPE_ENTRY, "invalid format - expected 'type' line");
914                 goto done;
915         }
916         eol = strchr(buffer, '\n');
917         if (!eol) {
918                 ret = report(options, &tag->object, FSCK_MSG_MISSING_TYPE, "invalid format - unexpected end after 'type' line");
919                 goto done;
920         }
921         if (type_from_string_gently(buffer, eol - buffer, 1) < 0)
922                 ret = report(options, &tag->object, FSCK_MSG_BAD_TYPE, "invalid 'type' value");
923         if (ret)
924                 goto done;
925         buffer = eol + 1;
926
927         if (!skip_prefix(buffer, "tag ", &buffer)) {
928                 ret = report(options, &tag->object, FSCK_MSG_MISSING_TAG_ENTRY, "invalid format - expected 'tag' line");
929                 goto done;
930         }
931         eol = strchr(buffer, '\n');
932         if (!eol) {
933                 ret = report(options, &tag->object, FSCK_MSG_MISSING_TAG, "invalid format - unexpected end after 'type' line");
934                 goto done;
935         }
936         strbuf_addf(&sb, "refs/tags/%.*s", (int)(eol - buffer), buffer);
937         if (check_refname_format(sb.buf, 0)) {
938                 ret = report(options, &tag->object, FSCK_MSG_BAD_TAG_NAME,
939                            "invalid 'tag' name: %.*s",
940                            (int)(eol - buffer), buffer);
941                 if (ret)
942                         goto done;
943         }
944         buffer = eol + 1;
945
946         if (!skip_prefix(buffer, "tagger ", &buffer)) {
947                 /* early tags do not contain 'tagger' lines; warn only */
948                 ret = report(options, &tag->object, FSCK_MSG_MISSING_TAGGER_ENTRY, "invalid format - expected 'tagger' line");
949                 if (ret)
950                         goto done;
951         }
952         else
953                 ret = fsck_ident(&buffer, &tag->object, options);
954
955 done:
956         strbuf_release(&sb);
957         free(to_free);
958         return ret;
959 }
960
961 static int fsck_tag(struct tag *tag, const char *data,
962         unsigned long size, struct fsck_options *options)
963 {
964         struct object *tagged = tag->tagged;
965
966         if (!tagged)
967                 return report(options, &tag->object, FSCK_MSG_BAD_TAG_OBJECT, "could not load tagged object");
968
969         return fsck_tag_buffer(tag, data, size, options);
970 }
971
972 struct fsck_gitmodules_data {
973         struct object *obj;
974         struct fsck_options *options;
975         int ret;
976 };
977
978 static int fsck_gitmodules_fn(const char *var, const char *value, void *vdata)
979 {
980         struct fsck_gitmodules_data *data = vdata;
981         const char *subsection, *key;
982         int subsection_len;
983         char *name;
984
985         if (parse_config_key(var, "submodule", &subsection, &subsection_len, &key) < 0 ||
986             !subsection)
987                 return 0;
988
989         name = xmemdupz(subsection, subsection_len);
990         if (check_submodule_name(name) < 0)
991                 data->ret |= report(data->options, data->obj,
992                                     FSCK_MSG_GITMODULES_NAME,
993                                     "disallowed submodule name: %s",
994                                     name);
995         free(name);
996
997         return 0;
998 }
999
1000 static int fsck_blob(struct blob *blob, const char *buf,
1001                      unsigned long size, struct fsck_options *options)
1002 {
1003         struct fsck_gitmodules_data data;
1004         struct config_options config_opts = { 0 };
1005
1006         if (!oidset_contains(&gitmodules_found, &blob->object.oid))
1007                 return 0;
1008         oidset_insert(&gitmodules_done, &blob->object.oid);
1009
1010         if (object_on_skiplist(options, &blob->object))
1011                 return 0;
1012
1013         if (!buf) {
1014                 /*
1015                  * A missing buffer here is a sign that the caller found the
1016                  * blob too gigantic to load into memory. Let's just consider
1017                  * that an error.
1018                  */
1019                 return report(options, &blob->object,
1020                               FSCK_MSG_GITMODULES_LARGE,
1021                               ".gitmodules too large to parse");
1022         }
1023
1024         data.obj = &blob->object;
1025         data.options = options;
1026         data.ret = 0;
1027         config_opts.error_action = CONFIG_ERROR_SILENT;
1028         if (git_config_from_mem(fsck_gitmodules_fn, CONFIG_ORIGIN_BLOB,
1029                                 ".gitmodules", buf, size, &data, &config_opts))
1030                 data.ret |= report(options, &blob->object,
1031                                    FSCK_MSG_GITMODULES_PARSE,
1032                                    "could not parse gitmodules blob");
1033
1034         return data.ret;
1035 }
1036
1037 int fsck_object(struct object *obj, void *data, unsigned long size,
1038         struct fsck_options *options)
1039 {
1040         if (!obj)
1041                 return report(options, obj, FSCK_MSG_BAD_OBJECT_SHA1, "no valid object to fsck");
1042
1043         if (obj->type == OBJ_BLOB)
1044                 return fsck_blob((struct blob *)obj, data, size, options);
1045         if (obj->type == OBJ_TREE)
1046                 return fsck_tree((struct tree *) obj, options);
1047         if (obj->type == OBJ_COMMIT)
1048                 return fsck_commit((struct commit *) obj, (const char *) data,
1049                         size, options);
1050         if (obj->type == OBJ_TAG)
1051                 return fsck_tag((struct tag *) obj, (const char *) data,
1052                         size, options);
1053
1054         return report(options, obj, FSCK_MSG_UNKNOWN_TYPE, "unknown type '%d' (internal fsck error)",
1055                           obj->type);
1056 }
1057
1058 int fsck_error_function(struct fsck_options *o,
1059         struct object *obj, int msg_type, const char *message)
1060 {
1061         if (msg_type == FSCK_WARN) {
1062                 warning("object %s: %s", describe_object(o, obj), message);
1063                 return 0;
1064         }
1065         error("object %s: %s", describe_object(o, obj), message);
1066         return 1;
1067 }
1068
1069 int fsck_finish(struct fsck_options *options)
1070 {
1071         int ret = 0;
1072         struct oidset_iter iter;
1073         const struct object_id *oid;
1074
1075         oidset_iter_init(&gitmodules_found, &iter);
1076         while ((oid = oidset_iter_next(&iter))) {
1077                 struct blob *blob;
1078                 enum object_type type;
1079                 unsigned long size;
1080                 char *buf;
1081
1082                 if (oidset_contains(&gitmodules_done, oid))
1083                         continue;
1084
1085                 blob = lookup_blob(the_repository, oid);
1086                 if (!blob) {
1087                         struct object *obj = lookup_unknown_object(oid->hash);
1088                         ret |= report(options, obj,
1089                                       FSCK_MSG_GITMODULES_BLOB,
1090                                       "non-blob found at .gitmodules");
1091                         continue;
1092                 }
1093
1094                 buf = read_object_file(oid, &type, &size);
1095                 if (!buf) {
1096                         if (is_promisor_object(&blob->object.oid))
1097                                 continue;
1098                         ret |= report(options, &blob->object,
1099                                       FSCK_MSG_GITMODULES_MISSING,
1100                                       "unable to read .gitmodules blob");
1101                         continue;
1102                 }
1103
1104                 if (type == OBJ_BLOB)
1105                         ret |= fsck_blob(blob, buf, size, options);
1106                 else
1107                         ret |= report(options, &blob->object,
1108                                       FSCK_MSG_GITMODULES_BLOB,
1109                                       "non-blob found at .gitmodules");
1110                 free(buf);
1111         }
1112
1113
1114         oidset_clear(&gitmodules_found);
1115         oidset_clear(&gitmodules_done);
1116         return ret;
1117 }