The second batch
[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 "url.h"
13 #include "utf8.h"
14 #include "decorate.h"
15 #include "oidset.h"
16 #include "packfile.h"
17 #include "submodule-config.h"
18 #include "config.h"
19 #include "credential.h"
20 #include "help.h"
21
22 #define STR(x) #x
23 #define MSG_ID(id, msg_type) { STR(id), NULL, NULL, FSCK_##msg_type },
24 static struct {
25         const char *id_string;
26         const char *downcased;
27         const char *camelcased;
28         enum fsck_msg_type msg_type;
29 } msg_id_info[FSCK_MSG_MAX + 1] = {
30         FOREACH_FSCK_MSG_ID(MSG_ID)
31         { NULL, NULL, NULL, -1 }
32 };
33 #undef MSG_ID
34 #undef STR
35
36 static void prepare_msg_ids(void)
37 {
38         int i;
39
40         if (msg_id_info[0].downcased)
41                 return;
42
43         /* convert id_string to lower case, without underscores. */
44         for (i = 0; i < FSCK_MSG_MAX; i++) {
45                 const char *p = msg_id_info[i].id_string;
46                 int len = strlen(p);
47                 char *q = xmalloc(len);
48
49                 msg_id_info[i].downcased = q;
50                 while (*p)
51                         if (*p == '_')
52                                 p++;
53                         else
54                                 *(q)++ = tolower(*(p)++);
55                 *q = '\0';
56
57                 p = msg_id_info[i].id_string;
58                 q = xmalloc(len);
59                 msg_id_info[i].camelcased = q;
60                 while (*p) {
61                         if (*p == '_') {
62                                 p++;
63                                 if (*p)
64                                         *q++ = *p++;
65                         } else {
66                                 *q++ = tolower(*p++);
67                         }
68                 }
69                 *q = '\0';
70         }
71 }
72
73 static int parse_msg_id(const char *text)
74 {
75         int i;
76
77         prepare_msg_ids();
78
79         for (i = 0; i < FSCK_MSG_MAX; i++)
80                 if (!strcmp(text, msg_id_info[i].downcased))
81                         return i;
82
83         return -1;
84 }
85
86 void list_config_fsck_msg_ids(struct string_list *list, const char *prefix)
87 {
88         int i;
89
90         prepare_msg_ids();
91
92         for (i = 0; i < FSCK_MSG_MAX; i++)
93                 list_config_item(list, prefix, msg_id_info[i].camelcased);
94 }
95
96 static enum fsck_msg_type fsck_msg_type(enum fsck_msg_id msg_id,
97         struct fsck_options *options)
98 {
99         assert(msg_id >= 0 && msg_id < FSCK_MSG_MAX);
100
101         if (!options->msg_type) {
102                 enum fsck_msg_type msg_type = msg_id_info[msg_id].msg_type;
103
104                 if (options->strict && msg_type == FSCK_WARN)
105                         msg_type = FSCK_ERROR;
106                 return msg_type;
107         }
108
109         return options->msg_type[msg_id];
110 }
111
112 static enum fsck_msg_type parse_msg_type(const char *str)
113 {
114         if (!strcmp(str, "error"))
115                 return FSCK_ERROR;
116         else if (!strcmp(str, "warn"))
117                 return FSCK_WARN;
118         else if (!strcmp(str, "ignore"))
119                 return FSCK_IGNORE;
120         else
121                 die("Unknown fsck message type: '%s'", str);
122 }
123
124 int is_valid_msg_type(const char *msg_id, const char *msg_type)
125 {
126         if (parse_msg_id(msg_id) < 0)
127                 return 0;
128         parse_msg_type(msg_type);
129         return 1;
130 }
131
132 void fsck_set_msg_type_from_ids(struct fsck_options *options,
133                                 enum fsck_msg_id msg_id,
134                                 enum fsck_msg_type msg_type)
135 {
136         if (!options->msg_type) {
137                 int i;
138                 enum fsck_msg_type *severity;
139                 ALLOC_ARRAY(severity, FSCK_MSG_MAX);
140                 for (i = 0; i < FSCK_MSG_MAX; i++)
141                         severity[i] = fsck_msg_type(i, options);
142                 options->msg_type = severity;
143         }
144
145         options->msg_type[msg_id] = msg_type;
146 }
147
148 void fsck_set_msg_type(struct fsck_options *options,
149                        const char *msg_id_str, const char *msg_type_str)
150 {
151         int msg_id = parse_msg_id(msg_id_str);
152         enum fsck_msg_type msg_type = parse_msg_type(msg_type_str);
153
154         if (msg_id < 0)
155                 die("Unhandled message id: %s", msg_id_str);
156
157         if (msg_type != FSCK_ERROR && msg_id_info[msg_id].msg_type == FSCK_FATAL)
158                 die("Cannot demote %s to %s", msg_id_str, msg_type_str);
159
160         fsck_set_msg_type_from_ids(options, msg_id, msg_type);
161 }
162
163 void fsck_set_msg_types(struct fsck_options *options, const char *values)
164 {
165         char *buf = xstrdup(values), *to_free = buf;
166         int done = 0;
167
168         while (!done) {
169                 int len = strcspn(buf, " ,|"), equal;
170
171                 done = !buf[len];
172                 if (!len) {
173                         buf++;
174                         continue;
175                 }
176                 buf[len] = '\0';
177
178                 for (equal = 0;
179                      equal < len && buf[equal] != '=' && buf[equal] != ':';
180                      equal++)
181                         buf[equal] = tolower(buf[equal]);
182                 buf[equal] = '\0';
183
184                 if (!strcmp(buf, "skiplist")) {
185                         if (equal == len)
186                                 die("skiplist requires a path");
187                         oidset_parse_file(&options->skiplist, buf + equal + 1);
188                         buf += len + 1;
189                         continue;
190                 }
191
192                 if (equal == len)
193                         die("Missing '=': '%s'", buf);
194
195                 fsck_set_msg_type(options, buf, buf + equal + 1);
196                 buf += len + 1;
197         }
198         free(to_free);
199 }
200
201 static int object_on_skiplist(struct fsck_options *opts,
202                               const struct object_id *oid)
203 {
204         return opts && oid && oidset_contains(&opts->skiplist, oid);
205 }
206
207 __attribute__((format (printf, 5, 6)))
208 static int report(struct fsck_options *options,
209                   const struct object_id *oid, enum object_type object_type,
210                   enum fsck_msg_id msg_id, const char *fmt, ...)
211 {
212         va_list ap;
213         struct strbuf sb = STRBUF_INIT;
214         enum fsck_msg_type msg_type = fsck_msg_type(msg_id, options);
215         int result;
216
217         if (msg_type == FSCK_IGNORE)
218                 return 0;
219
220         if (object_on_skiplist(options, oid))
221                 return 0;
222
223         if (msg_type == FSCK_FATAL)
224                 msg_type = FSCK_ERROR;
225         else if (msg_type == FSCK_INFO)
226                 msg_type = FSCK_WARN;
227
228         prepare_msg_ids();
229         strbuf_addf(&sb, "%s: ", msg_id_info[msg_id].camelcased);
230
231         va_start(ap, fmt);
232         strbuf_vaddf(&sb, fmt, ap);
233         result = options->error_func(options, oid, object_type,
234                                      msg_type, msg_id, sb.buf);
235         strbuf_release(&sb);
236         va_end(ap);
237
238         return result;
239 }
240
241 void fsck_enable_object_names(struct fsck_options *options)
242 {
243         if (!options->object_names)
244                 options->object_names = kh_init_oid_map();
245 }
246
247 const char *fsck_get_object_name(struct fsck_options *options,
248                                  const struct object_id *oid)
249 {
250         khiter_t pos;
251         if (!options->object_names)
252                 return NULL;
253         pos = kh_get_oid_map(options->object_names, *oid);
254         if (pos >= kh_end(options->object_names))
255                 return NULL;
256         return kh_value(options->object_names, pos);
257 }
258
259 void fsck_put_object_name(struct fsck_options *options,
260                           const struct object_id *oid,
261                           const char *fmt, ...)
262 {
263         va_list ap;
264         struct strbuf buf = STRBUF_INIT;
265         khiter_t pos;
266         int hashret;
267
268         if (!options->object_names)
269                 return;
270
271         pos = kh_put_oid_map(options->object_names, *oid, &hashret);
272         if (!hashret)
273                 return;
274         va_start(ap, fmt);
275         strbuf_vaddf(&buf, fmt, ap);
276         kh_value(options->object_names, pos) = strbuf_detach(&buf, NULL);
277         va_end(ap);
278 }
279
280 const char *fsck_describe_object(struct fsck_options *options,
281                                  const struct object_id *oid)
282 {
283         static struct strbuf bufs[] = {
284                 STRBUF_INIT, STRBUF_INIT, STRBUF_INIT, STRBUF_INIT
285         };
286         static int b = 0;
287         struct strbuf *buf;
288         const char *name = fsck_get_object_name(options, oid);
289
290         buf = bufs + b;
291         b = (b + 1) % ARRAY_SIZE(bufs);
292         strbuf_reset(buf);
293         strbuf_addstr(buf, oid_to_hex(oid));
294         if (name)
295                 strbuf_addf(buf, " (%s)", name);
296
297         return buf->buf;
298 }
299
300 static int fsck_walk_tree(struct tree *tree, void *data, struct fsck_options *options)
301 {
302         struct tree_desc desc;
303         struct name_entry entry;
304         int res = 0;
305         const char *name;
306
307         if (parse_tree(tree))
308                 return -1;
309
310         name = fsck_get_object_name(options, &tree->object.oid);
311         if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
312                 return -1;
313         while (tree_entry_gently(&desc, &entry)) {
314                 struct object *obj;
315                 int result;
316
317                 if (S_ISGITLINK(entry.mode))
318                         continue;
319
320                 if (S_ISDIR(entry.mode)) {
321                         obj = (struct object *)lookup_tree(the_repository, &entry.oid);
322                         if (name && obj)
323                                 fsck_put_object_name(options, &entry.oid, "%s%s/",
324                                                      name, entry.path);
325                         result = options->walk(obj, OBJ_TREE, data, options);
326                 }
327                 else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode)) {
328                         obj = (struct object *)lookup_blob(the_repository, &entry.oid);
329                         if (name && obj)
330                                 fsck_put_object_name(options, &entry.oid, "%s%s",
331                                                      name, entry.path);
332                         result = options->walk(obj, OBJ_BLOB, data, options);
333                 }
334                 else {
335                         result = error("in tree %s: entry %s has bad mode %.6o",
336                                        fsck_describe_object(options, &tree->object.oid),
337                                        entry.path, entry.mode);
338                 }
339                 if (result < 0)
340                         return result;
341                 if (!res)
342                         res = result;
343         }
344         return res;
345 }
346
347 static int fsck_walk_commit(struct commit *commit, void *data, struct fsck_options *options)
348 {
349         int counter = 0, generation = 0, name_prefix_len = 0;
350         struct commit_list *parents;
351         int res;
352         int result;
353         const char *name;
354
355         if (parse_commit(commit))
356                 return -1;
357
358         name = fsck_get_object_name(options, &commit->object.oid);
359         if (name)
360                 fsck_put_object_name(options, get_commit_tree_oid(commit),
361                                      "%s:", name);
362
363         result = options->walk((struct object *)get_commit_tree(commit),
364                                OBJ_TREE, data, options);
365         if (result < 0)
366                 return result;
367         res = result;
368
369         parents = commit->parents;
370         if (name && parents) {
371                 int len = strlen(name), power;
372
373                 if (len && name[len - 1] == '^') {
374                         generation = 1;
375                         name_prefix_len = len - 1;
376                 }
377                 else { /* parse ~<generation> suffix */
378                         for (generation = 0, power = 1;
379                              len && isdigit(name[len - 1]);
380                              power *= 10)
381                                 generation += power * (name[--len] - '0');
382                         if (power > 1 && len && name[len - 1] == '~')
383                                 name_prefix_len = len - 1;
384                         else {
385                                 /* Maybe a non-first parent, e.g. HEAD^2 */
386                                 generation = 0;
387                                 name_prefix_len = len;
388                         }
389                 }
390         }
391
392         while (parents) {
393                 if (name) {
394                         struct object_id *oid = &parents->item->object.oid;
395
396                         if (counter++)
397                                 fsck_put_object_name(options, oid, "%s^%d",
398                                                      name, counter);
399                         else if (generation > 0)
400                                 fsck_put_object_name(options, oid, "%.*s~%d",
401                                                      name_prefix_len, name,
402                                                      generation + 1);
403                         else
404                                 fsck_put_object_name(options, oid, "%s^", name);
405                 }
406                 result = options->walk((struct object *)parents->item, OBJ_COMMIT, data, options);
407                 if (result < 0)
408                         return result;
409                 if (!res)
410                         res = result;
411                 parents = parents->next;
412         }
413         return res;
414 }
415
416 static int fsck_walk_tag(struct tag *tag, void *data, struct fsck_options *options)
417 {
418         const char *name = fsck_get_object_name(options, &tag->object.oid);
419
420         if (parse_tag(tag))
421                 return -1;
422         if (name)
423                 fsck_put_object_name(options, &tag->tagged->oid, "%s", name);
424         return options->walk(tag->tagged, OBJ_ANY, data, options);
425 }
426
427 int fsck_walk(struct object *obj, void *data, struct fsck_options *options)
428 {
429         if (!obj)
430                 return -1;
431
432         if (obj->type == OBJ_NONE)
433                 parse_object(the_repository, &obj->oid);
434
435         switch (obj->type) {
436         case OBJ_BLOB:
437                 return 0;
438         case OBJ_TREE:
439                 return fsck_walk_tree((struct tree *)obj, data, options);
440         case OBJ_COMMIT:
441                 return fsck_walk_commit((struct commit *)obj, data, options);
442         case OBJ_TAG:
443                 return fsck_walk_tag((struct tag *)obj, data, options);
444         default:
445                 error("Unknown object type for %s",
446                       fsck_describe_object(options, &obj->oid));
447                 return -1;
448         }
449 }
450
451 struct name_stack {
452         const char **names;
453         size_t nr, alloc;
454 };
455
456 static void name_stack_push(struct name_stack *stack, const char *name)
457 {
458         ALLOC_GROW(stack->names, stack->nr + 1, stack->alloc);
459         stack->names[stack->nr++] = name;
460 }
461
462 static const char *name_stack_pop(struct name_stack *stack)
463 {
464         return stack->nr ? stack->names[--stack->nr] : NULL;
465 }
466
467 static void name_stack_clear(struct name_stack *stack)
468 {
469         FREE_AND_NULL(stack->names);
470         stack->nr = stack->alloc = 0;
471 }
472
473 /*
474  * The entries in a tree are ordered in the _path_ order,
475  * which means that a directory entry is ordered by adding
476  * a slash to the end of it.
477  *
478  * So a directory called "a" is ordered _after_ a file
479  * called "a.c", because "a/" sorts after "a.c".
480  */
481 #define TREE_UNORDERED (-1)
482 #define TREE_HAS_DUPS  (-2)
483
484 static int is_less_than_slash(unsigned char c)
485 {
486         return '\0' < c && c < '/';
487 }
488
489 static int verify_ordered(unsigned mode1, const char *name1,
490                           unsigned mode2, const char *name2,
491                           struct name_stack *candidates)
492 {
493         int len1 = strlen(name1);
494         int len2 = strlen(name2);
495         int len = len1 < len2 ? len1 : len2;
496         unsigned char c1, c2;
497         int cmp;
498
499         cmp = memcmp(name1, name2, len);
500         if (cmp < 0)
501                 return 0;
502         if (cmp > 0)
503                 return TREE_UNORDERED;
504
505         /*
506          * Ok, the first <len> characters are the same.
507          * Now we need to order the next one, but turn
508          * a '\0' into a '/' for a directory entry.
509          */
510         c1 = name1[len];
511         c2 = name2[len];
512         if (!c1 && !c2)
513                 /*
514                  * git-write-tree used to write out a nonsense tree that has
515                  * entries with the same name, one blob and one tree.  Make
516                  * sure we do not have duplicate entries.
517                  */
518                 return TREE_HAS_DUPS;
519         if (!c1 && S_ISDIR(mode1))
520                 c1 = '/';
521         if (!c2 && S_ISDIR(mode2))
522                 c2 = '/';
523
524         /*
525          * There can be non-consecutive duplicates due to the implicitly
526          * added slash, e.g.:
527          *
528          *   foo
529          *   foo.bar
530          *   foo.bar.baz
531          *   foo.bar/
532          *   foo/
533          *
534          * Record non-directory candidates (like "foo" and "foo.bar" in
535          * the example) on a stack and check directory candidates (like
536          * foo/" and "foo.bar/") against that stack.
537          */
538         if (!c1 && is_less_than_slash(c2)) {
539                 name_stack_push(candidates, name1);
540         } else if (c2 == '/' && is_less_than_slash(c1)) {
541                 for (;;) {
542                         const char *p;
543                         const char *f_name = name_stack_pop(candidates);
544
545                         if (!f_name)
546                                 break;
547                         if (!skip_prefix(name2, f_name, &p))
548                                 continue;
549                         if (!*p)
550                                 return TREE_HAS_DUPS;
551                         if (is_less_than_slash(*p)) {
552                                 name_stack_push(candidates, f_name);
553                                 break;
554                         }
555                 }
556         }
557
558         return c1 < c2 ? 0 : TREE_UNORDERED;
559 }
560
561 static int fsck_tree(const struct object_id *tree_oid,
562                      const char *buffer, unsigned long size,
563                      struct fsck_options *options)
564 {
565         int retval = 0;
566         int has_null_sha1 = 0;
567         int has_full_path = 0;
568         int has_empty_name = 0;
569         int has_dot = 0;
570         int has_dotdot = 0;
571         int has_dotgit = 0;
572         int has_zero_pad = 0;
573         int has_bad_modes = 0;
574         int has_dup_entries = 0;
575         int not_properly_sorted = 0;
576         struct tree_desc desc;
577         unsigned o_mode;
578         const char *o_name;
579         struct name_stack df_dup_candidates = { NULL };
580
581         if (init_tree_desc_gently(&desc, buffer, size)) {
582                 retval += report(options, tree_oid, OBJ_TREE,
583                                  FSCK_MSG_BAD_TREE,
584                                  "cannot be parsed as a tree");
585                 return retval;
586         }
587
588         o_mode = 0;
589         o_name = NULL;
590
591         while (desc.size) {
592                 unsigned short mode;
593                 const char *name, *backslash;
594                 const struct object_id *entry_oid;
595
596                 entry_oid = tree_entry_extract(&desc, &name, &mode);
597
598                 has_null_sha1 |= is_null_oid(entry_oid);
599                 has_full_path |= !!strchr(name, '/');
600                 has_empty_name |= !*name;
601                 has_dot |= !strcmp(name, ".");
602                 has_dotdot |= !strcmp(name, "..");
603                 has_dotgit |= is_hfs_dotgit(name) || is_ntfs_dotgit(name);
604                 has_zero_pad |= *(char *)desc.buffer == '0';
605
606                 if (is_hfs_dotgitmodules(name) || is_ntfs_dotgitmodules(name)) {
607                         if (!S_ISLNK(mode))
608                                 oidset_insert(&options->gitmodules_found,
609                                               entry_oid);
610                         else
611                                 retval += report(options,
612                                                  tree_oid, OBJ_TREE,
613                                                  FSCK_MSG_GITMODULES_SYMLINK,
614                                                  ".gitmodules is a symbolic link");
615                 }
616
617                 if (S_ISLNK(mode)) {
618                         if (is_hfs_dotgitignore(name) ||
619                             is_ntfs_dotgitignore(name))
620                                 retval += report(options, tree_oid, OBJ_TREE,
621                                                  FSCK_MSG_GITIGNORE_SYMLINK,
622                                                  ".gitignore is a symlink");
623                         if (is_hfs_dotgitattributes(name) ||
624                             is_ntfs_dotgitattributes(name))
625                                 retval += report(options, tree_oid, OBJ_TREE,
626                                                  FSCK_MSG_GITATTRIBUTES_SYMLINK,
627                                                  ".gitattributes is a symlink");
628                         if (is_hfs_dotmailmap(name) ||
629                             is_ntfs_dotmailmap(name))
630                                 retval += report(options, tree_oid, OBJ_TREE,
631                                                  FSCK_MSG_MAILMAP_SYMLINK,
632                                                  ".mailmap is a symlink");
633                 }
634
635                 if ((backslash = strchr(name, '\\'))) {
636                         while (backslash) {
637                                 backslash++;
638                                 has_dotgit |= is_ntfs_dotgit(backslash);
639                                 if (is_ntfs_dotgitmodules(backslash)) {
640                                         if (!S_ISLNK(mode))
641                                                 oidset_insert(&options->gitmodules_found,
642                                                               entry_oid);
643                                         else
644                                                 retval += report(options, tree_oid, OBJ_TREE,
645                                                                  FSCK_MSG_GITMODULES_SYMLINK,
646                                                                  ".gitmodules is a symbolic link");
647                                 }
648                                 backslash = strchr(backslash, '\\');
649                         }
650                 }
651
652                 if (update_tree_entry_gently(&desc)) {
653                         retval += report(options, tree_oid, OBJ_TREE,
654                                          FSCK_MSG_BAD_TREE,
655                                          "cannot be parsed as a tree");
656                         break;
657                 }
658
659                 switch (mode) {
660                 /*
661                  * Standard modes..
662                  */
663                 case S_IFREG | 0755:
664                 case S_IFREG | 0644:
665                 case S_IFLNK:
666                 case S_IFDIR:
667                 case S_IFGITLINK:
668                         break;
669                 /*
670                  * This is nonstandard, but we had a few of these
671                  * early on when we honored the full set of mode
672                  * bits..
673                  */
674                 case S_IFREG | 0664:
675                         if (!options->strict)
676                                 break;
677                         /* fallthrough */
678                 default:
679                         has_bad_modes = 1;
680                 }
681
682                 if (o_name) {
683                         switch (verify_ordered(o_mode, o_name, mode, name,
684                                                &df_dup_candidates)) {
685                         case TREE_UNORDERED:
686                                 not_properly_sorted = 1;
687                                 break;
688                         case TREE_HAS_DUPS:
689                                 has_dup_entries = 1;
690                                 break;
691                         default:
692                                 break;
693                         }
694                 }
695
696                 o_mode = mode;
697                 o_name = name;
698         }
699
700         name_stack_clear(&df_dup_candidates);
701
702         if (has_null_sha1)
703                 retval += report(options, tree_oid, OBJ_TREE,
704                                  FSCK_MSG_NULL_SHA1,
705                                  "contains entries pointing to null sha1");
706         if (has_full_path)
707                 retval += report(options, tree_oid, OBJ_TREE,
708                                  FSCK_MSG_FULL_PATHNAME,
709                                  "contains full pathnames");
710         if (has_empty_name)
711                 retval += report(options, tree_oid, OBJ_TREE,
712                                  FSCK_MSG_EMPTY_NAME,
713                                  "contains empty pathname");
714         if (has_dot)
715                 retval += report(options, tree_oid, OBJ_TREE,
716                                  FSCK_MSG_HAS_DOT,
717                                  "contains '.'");
718         if (has_dotdot)
719                 retval += report(options, tree_oid, OBJ_TREE,
720                                  FSCK_MSG_HAS_DOTDOT,
721                                  "contains '..'");
722         if (has_dotgit)
723                 retval += report(options, tree_oid, OBJ_TREE,
724                                  FSCK_MSG_HAS_DOTGIT,
725                                  "contains '.git'");
726         if (has_zero_pad)
727                 retval += report(options, tree_oid, OBJ_TREE,
728                                  FSCK_MSG_ZERO_PADDED_FILEMODE,
729                                  "contains zero-padded file modes");
730         if (has_bad_modes)
731                 retval += report(options, tree_oid, OBJ_TREE,
732                                  FSCK_MSG_BAD_FILEMODE,
733                                  "contains bad file modes");
734         if (has_dup_entries)
735                 retval += report(options, tree_oid, OBJ_TREE,
736                                  FSCK_MSG_DUPLICATE_ENTRIES,
737                                  "contains duplicate file entries");
738         if (not_properly_sorted)
739                 retval += report(options, tree_oid, OBJ_TREE,
740                                  FSCK_MSG_TREE_NOT_SORTED,
741                                  "not properly sorted");
742         return retval;
743 }
744
745 static int verify_headers(const void *data, unsigned long size,
746                           const struct object_id *oid, enum object_type type,
747                           struct fsck_options *options)
748 {
749         const char *buffer = (const char *)data;
750         unsigned long i;
751
752         for (i = 0; i < size; i++) {
753                 switch (buffer[i]) {
754                 case '\0':
755                         return report(options, oid, type,
756                                 FSCK_MSG_NUL_IN_HEADER,
757                                 "unterminated header: NUL at offset %ld", i);
758                 case '\n':
759                         if (i + 1 < size && buffer[i + 1] == '\n')
760                                 return 0;
761                 }
762         }
763
764         /*
765          * We did not find double-LF that separates the header
766          * and the body.  Not having a body is not a crime but
767          * we do want to see the terminating LF for the last header
768          * line.
769          */
770         if (size && buffer[size - 1] == '\n')
771                 return 0;
772
773         return report(options, oid, type,
774                 FSCK_MSG_UNTERMINATED_HEADER, "unterminated header");
775 }
776
777 static int fsck_ident(const char **ident,
778                       const struct object_id *oid, enum object_type type,
779                       struct fsck_options *options)
780 {
781         const char *p = *ident;
782         char *end;
783
784         *ident = strchrnul(*ident, '\n');
785         if (**ident == '\n')
786                 (*ident)++;
787
788         if (*p == '<')
789                 return report(options, oid, type, FSCK_MSG_MISSING_NAME_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
790         p += strcspn(p, "<>\n");
791         if (*p == '>')
792                 return report(options, oid, type, FSCK_MSG_BAD_NAME, "invalid author/committer line - bad name");
793         if (*p != '<')
794                 return report(options, oid, type, FSCK_MSG_MISSING_EMAIL, "invalid author/committer line - missing email");
795         if (p[-1] != ' ')
796                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
797         p++;
798         p += strcspn(p, "<>\n");
799         if (*p != '>')
800                 return report(options, oid, type, FSCK_MSG_BAD_EMAIL, "invalid author/committer line - bad email");
801         p++;
802         if (*p != ' ')
803                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_DATE, "invalid author/committer line - missing space before date");
804         p++;
805         if (*p == '0' && p[1] != ' ')
806                 return report(options, oid, type, FSCK_MSG_ZERO_PADDED_DATE, "invalid author/committer line - zero-padded date");
807         if (date_overflows(parse_timestamp(p, &end, 10)))
808                 return report(options, oid, type, FSCK_MSG_BAD_DATE_OVERFLOW, "invalid author/committer line - date causes integer overflow");
809         if ((end == p || *end != ' '))
810                 return report(options, oid, type, FSCK_MSG_BAD_DATE, "invalid author/committer line - bad date");
811         p = end + 1;
812         if ((*p != '+' && *p != '-') ||
813             !isdigit(p[1]) ||
814             !isdigit(p[2]) ||
815             !isdigit(p[3]) ||
816             !isdigit(p[4]) ||
817             (p[5] != '\n'))
818                 return report(options, oid, type, FSCK_MSG_BAD_TIMEZONE, "invalid author/committer line - bad time zone");
819         p += 6;
820         return 0;
821 }
822
823 static int fsck_commit(const struct object_id *oid,
824                        const char *buffer, unsigned long size,
825                        struct fsck_options *options)
826 {
827         struct object_id tree_oid, parent_oid;
828         unsigned author_count;
829         int err;
830         const char *buffer_begin = buffer;
831         const char *p;
832
833         if (verify_headers(buffer, size, oid, OBJ_COMMIT, options))
834                 return -1;
835
836         if (!skip_prefix(buffer, "tree ", &buffer))
837                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_TREE, "invalid format - expected 'tree' line");
838         if (parse_oid_hex(buffer, &tree_oid, &p) || *p != '\n') {
839                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_TREE_SHA1, "invalid 'tree' line format - bad sha1");
840                 if (err)
841                         return err;
842         }
843         buffer = p + 1;
844         while (skip_prefix(buffer, "parent ", &buffer)) {
845                 if (parse_oid_hex(buffer, &parent_oid, &p) || *p != '\n') {
846                         err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_PARENT_SHA1, "invalid 'parent' line format - bad sha1");
847                         if (err)
848                                 return err;
849                 }
850                 buffer = p + 1;
851         }
852         author_count = 0;
853         while (skip_prefix(buffer, "author ", &buffer)) {
854                 author_count++;
855                 err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
856                 if (err)
857                         return err;
858         }
859         if (author_count < 1)
860                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_AUTHOR, "invalid format - expected 'author' line");
861         else if (author_count > 1)
862                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MULTIPLE_AUTHORS, "invalid format - multiple 'author' lines");
863         if (err)
864                 return err;
865         if (!skip_prefix(buffer, "committer ", &buffer))
866                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_COMMITTER, "invalid format - expected 'committer' line");
867         err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
868         if (err)
869                 return err;
870         if (memchr(buffer_begin, '\0', size)) {
871                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_NUL_IN_COMMIT,
872                              "NUL byte in the commit object body");
873                 if (err)
874                         return err;
875         }
876         return 0;
877 }
878
879 static int fsck_tag(const struct object_id *oid, const char *buffer,
880                     unsigned long size, struct fsck_options *options)
881 {
882         struct object_id tagged_oid;
883         int tagged_type;
884         return fsck_tag_standalone(oid, buffer, size, options, &tagged_oid,
885                                    &tagged_type);
886 }
887
888 int fsck_tag_standalone(const struct object_id *oid, const char *buffer,
889                         unsigned long size, struct fsck_options *options,
890                         struct object_id *tagged_oid,
891                         int *tagged_type)
892 {
893         int ret = 0;
894         char *eol;
895         struct strbuf sb = STRBUF_INIT;
896         const char *p;
897
898         ret = verify_headers(buffer, size, oid, OBJ_TAG, options);
899         if (ret)
900                 goto done;
901
902         if (!skip_prefix(buffer, "object ", &buffer)) {
903                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_OBJECT, "invalid format - expected 'object' line");
904                 goto done;
905         }
906         if (parse_oid_hex(buffer, tagged_oid, &p) || *p != '\n') {
907                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_OBJECT_SHA1, "invalid 'object' line format - bad sha1");
908                 if (ret)
909                         goto done;
910         }
911         buffer = p + 1;
912
913         if (!skip_prefix(buffer, "type ", &buffer)) {
914                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE_ENTRY, "invalid format - expected 'type' line");
915                 goto done;
916         }
917         eol = strchr(buffer, '\n');
918         if (!eol) {
919                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE, "invalid format - unexpected end after 'type' line");
920                 goto done;
921         }
922         *tagged_type = type_from_string_gently(buffer, eol - buffer, 1);
923         if (*tagged_type < 0)
924                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_TYPE, "invalid 'type' value");
925         if (ret)
926                 goto done;
927         buffer = eol + 1;
928
929         if (!skip_prefix(buffer, "tag ", &buffer)) {
930                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG_ENTRY, "invalid format - expected 'tag' line");
931                 goto done;
932         }
933         eol = strchr(buffer, '\n');
934         if (!eol) {
935                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG, "invalid format - unexpected end after 'type' line");
936                 goto done;
937         }
938         strbuf_addf(&sb, "refs/tags/%.*s", (int)(eol - buffer), buffer);
939         if (check_refname_format(sb.buf, 0)) {
940                 ret = report(options, oid, OBJ_TAG,
941                              FSCK_MSG_BAD_TAG_NAME,
942                              "invalid 'tag' name: %.*s",
943                              (int)(eol - buffer), buffer);
944                 if (ret)
945                         goto done;
946         }
947         buffer = eol + 1;
948
949         if (!skip_prefix(buffer, "tagger ", &buffer)) {
950                 /* early tags do not contain 'tagger' lines; warn only */
951                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAGGER_ENTRY, "invalid format - expected 'tagger' line");
952                 if (ret)
953                         goto done;
954         }
955         else
956                 ret = fsck_ident(&buffer, oid, OBJ_TAG, options);
957         if (!*buffer)
958                 goto done;
959
960         if (!starts_with(buffer, "\n")) {
961                 /*
962                  * The verify_headers() check will allow
963                  * e.g. "[...]tagger <tagger>\nsome
964                  * garbage\n\nmessage" to pass, thinking "some
965                  * garbage" could be a custom header. E.g. "mktag"
966                  * doesn't want any unknown headers.
967                  */
968                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_EXTRA_HEADER_ENTRY, "invalid format - extra header(s) after 'tagger'");
969                 if (ret)
970                         goto done;
971         }
972
973 done:
974         strbuf_release(&sb);
975         return ret;
976 }
977
978 /*
979  * Like builtin/submodule--helper.c's starts_with_dot_slash, but without
980  * relying on the platform-dependent is_dir_sep helper.
981  *
982  * This is for use in checking whether a submodule URL is interpreted as
983  * relative to the current directory on any platform, since \ is a
984  * directory separator on Windows but not on other platforms.
985  */
986 static int starts_with_dot_slash(const char *str)
987 {
988         return str[0] == '.' && (str[1] == '/' || str[1] == '\\');
989 }
990
991 /*
992  * Like starts_with_dot_slash, this is a variant of submodule--helper's
993  * helper of the same name with the twist that it accepts backslash as a
994  * directory separator even on non-Windows platforms.
995  */
996 static int starts_with_dot_dot_slash(const char *str)
997 {
998         return str[0] == '.' && starts_with_dot_slash(str + 1);
999 }
1000
1001 static int submodule_url_is_relative(const char *url)
1002 {
1003         return starts_with_dot_slash(url) || starts_with_dot_dot_slash(url);
1004 }
1005
1006 /*
1007  * Count directory components that a relative submodule URL should chop
1008  * from the remote_url it is to be resolved against.
1009  *
1010  * In other words, this counts "../" components at the start of a
1011  * submodule URL.
1012  *
1013  * Returns the number of directory components to chop and writes a
1014  * pointer to the next character of url after all leading "./" and
1015  * "../" components to out.
1016  */
1017 static int count_leading_dotdots(const char *url, const char **out)
1018 {
1019         int result = 0;
1020         while (1) {
1021                 if (starts_with_dot_dot_slash(url)) {
1022                         result++;
1023                         url += strlen("../");
1024                         continue;
1025                 }
1026                 if (starts_with_dot_slash(url)) {
1027                         url += strlen("./");
1028                         continue;
1029                 }
1030                 *out = url;
1031                 return result;
1032         }
1033 }
1034 /*
1035  * Check whether a transport is implemented by git-remote-curl.
1036  *
1037  * If it is, returns 1 and writes the URL that would be passed to
1038  * git-remote-curl to the "out" parameter.
1039  *
1040  * Otherwise, returns 0 and leaves "out" untouched.
1041  *
1042  * Examples:
1043  *   http::https://example.com/repo.git -> 1, https://example.com/repo.git
1044  *   https://example.com/repo.git -> 1, https://example.com/repo.git
1045  *   git://example.com/repo.git -> 0
1046  *
1047  * This is for use in checking for previously exploitable bugs that
1048  * required a submodule URL to be passed to git-remote-curl.
1049  */
1050 static int url_to_curl_url(const char *url, const char **out)
1051 {
1052         /*
1053          * We don't need to check for case-aliases, "http.exe", and so
1054          * on because in the default configuration, is_transport_allowed
1055          * prevents URLs with those schemes from being cloned
1056          * automatically.
1057          */
1058         if (skip_prefix(url, "http::", out) ||
1059             skip_prefix(url, "https::", out) ||
1060             skip_prefix(url, "ftp::", out) ||
1061             skip_prefix(url, "ftps::", out))
1062                 return 1;
1063         if (starts_with(url, "http://") ||
1064             starts_with(url, "https://") ||
1065             starts_with(url, "ftp://") ||
1066             starts_with(url, "ftps://")) {
1067                 *out = url;
1068                 return 1;
1069         }
1070         return 0;
1071 }
1072
1073 static int check_submodule_url(const char *url)
1074 {
1075         const char *curl_url;
1076
1077         if (looks_like_command_line_option(url))
1078                 return -1;
1079
1080         if (submodule_url_is_relative(url) || starts_with(url, "git://")) {
1081                 char *decoded;
1082                 const char *next;
1083                 int has_nl;
1084
1085                 /*
1086                  * This could be appended to an http URL and url-decoded;
1087                  * check for malicious characters.
1088                  */
1089                 decoded = url_decode(url);
1090                 has_nl = !!strchr(decoded, '\n');
1091
1092                 free(decoded);
1093                 if (has_nl)
1094                         return -1;
1095
1096                 /*
1097                  * URLs which escape their root via "../" can overwrite
1098                  * the host field and previous components, resolving to
1099                  * URLs like https::example.com/submodule.git and
1100                  * https:///example.com/submodule.git that were
1101                  * susceptible to CVE-2020-11008.
1102                  */
1103                 if (count_leading_dotdots(url, &next) > 0 &&
1104                     (*next == ':' || *next == '/'))
1105                         return -1;
1106         }
1107
1108         else if (url_to_curl_url(url, &curl_url)) {
1109                 struct credential c = CREDENTIAL_INIT;
1110                 int ret = 0;
1111                 if (credential_from_url_gently(&c, curl_url, 1) ||
1112                     !*c.host)
1113                         ret = -1;
1114                 credential_clear(&c);
1115                 return ret;
1116         }
1117
1118         return 0;
1119 }
1120
1121 struct fsck_gitmodules_data {
1122         const struct object_id *oid;
1123         struct fsck_options *options;
1124         int ret;
1125 };
1126
1127 static int fsck_gitmodules_fn(const char *var, const char *value, void *vdata)
1128 {
1129         struct fsck_gitmodules_data *data = vdata;
1130         const char *subsection, *key;
1131         size_t subsection_len;
1132         char *name;
1133
1134         if (parse_config_key(var, "submodule", &subsection, &subsection_len, &key) < 0 ||
1135             !subsection)
1136                 return 0;
1137
1138         name = xmemdupz(subsection, subsection_len);
1139         if (check_submodule_name(name) < 0)
1140                 data->ret |= report(data->options,
1141                                     data->oid, OBJ_BLOB,
1142                                     FSCK_MSG_GITMODULES_NAME,
1143                                     "disallowed submodule name: %s",
1144                                     name);
1145         if (!strcmp(key, "url") && value &&
1146             check_submodule_url(value) < 0)
1147                 data->ret |= report(data->options,
1148                                     data->oid, OBJ_BLOB,
1149                                     FSCK_MSG_GITMODULES_URL,
1150                                     "disallowed submodule url: %s",
1151                                     value);
1152         if (!strcmp(key, "path") && value &&
1153             looks_like_command_line_option(value))
1154                 data->ret |= report(data->options,
1155                                     data->oid, OBJ_BLOB,
1156                                     FSCK_MSG_GITMODULES_PATH,
1157                                     "disallowed submodule path: %s",
1158                                     value);
1159         if (!strcmp(key, "update") && value &&
1160             parse_submodule_update_type(value) == SM_UPDATE_COMMAND)
1161                 data->ret |= report(data->options, data->oid, OBJ_BLOB,
1162                                     FSCK_MSG_GITMODULES_UPDATE,
1163                                     "disallowed submodule update setting: %s",
1164                                     value);
1165         free(name);
1166
1167         return 0;
1168 }
1169
1170 static int fsck_blob(const struct object_id *oid, const char *buf,
1171                      unsigned long size, struct fsck_options *options)
1172 {
1173         struct fsck_gitmodules_data data;
1174         struct config_options config_opts = { 0 };
1175
1176         if (!oidset_contains(&options->gitmodules_found, oid))
1177                 return 0;
1178         oidset_insert(&options->gitmodules_done, oid);
1179
1180         if (object_on_skiplist(options, oid))
1181                 return 0;
1182
1183         if (!buf) {
1184                 /*
1185                  * A missing buffer here is a sign that the caller found the
1186                  * blob too gigantic to load into memory. Let's just consider
1187                  * that an error.
1188                  */
1189                 return report(options, oid, OBJ_BLOB,
1190                               FSCK_MSG_GITMODULES_LARGE,
1191                               ".gitmodules too large to parse");
1192         }
1193
1194         data.oid = oid;
1195         data.options = options;
1196         data.ret = 0;
1197         config_opts.error_action = CONFIG_ERROR_SILENT;
1198         if (git_config_from_mem(fsck_gitmodules_fn, CONFIG_ORIGIN_BLOB,
1199                                 ".gitmodules", buf, size, &data, &config_opts))
1200                 data.ret |= report(options, oid, OBJ_BLOB,
1201                                    FSCK_MSG_GITMODULES_PARSE,
1202                                    "could not parse gitmodules blob");
1203
1204         return data.ret;
1205 }
1206
1207 int fsck_object(struct object *obj, void *data, unsigned long size,
1208         struct fsck_options *options)
1209 {
1210         if (!obj)
1211                 return report(options, NULL, OBJ_NONE, FSCK_MSG_BAD_OBJECT_SHA1, "no valid object to fsck");
1212
1213         if (obj->type == OBJ_BLOB)
1214                 return fsck_blob(&obj->oid, data, size, options);
1215         if (obj->type == OBJ_TREE)
1216                 return fsck_tree(&obj->oid, data, size, options);
1217         if (obj->type == OBJ_COMMIT)
1218                 return fsck_commit(&obj->oid, data, size, options);
1219         if (obj->type == OBJ_TAG)
1220                 return fsck_tag(&obj->oid, data, size, options);
1221
1222         return report(options, &obj->oid, obj->type,
1223                       FSCK_MSG_UNKNOWN_TYPE,
1224                       "unknown type '%d' (internal fsck error)",
1225                       obj->type);
1226 }
1227
1228 int fsck_error_function(struct fsck_options *o,
1229                         const struct object_id *oid,
1230                         enum object_type object_type,
1231                         enum fsck_msg_type msg_type,
1232                         enum fsck_msg_id msg_id,
1233                         const char *message)
1234 {
1235         if (msg_type == FSCK_WARN) {
1236                 warning("object %s: %s", fsck_describe_object(o, oid), message);
1237                 return 0;
1238         }
1239         error("object %s: %s", fsck_describe_object(o, oid), message);
1240         return 1;
1241 }
1242
1243 int fsck_finish(struct fsck_options *options)
1244 {
1245         int ret = 0;
1246         struct oidset_iter iter;
1247         const struct object_id *oid;
1248
1249         oidset_iter_init(&options->gitmodules_found, &iter);
1250         while ((oid = oidset_iter_next(&iter))) {
1251                 enum object_type type;
1252                 unsigned long size;
1253                 char *buf;
1254
1255                 if (oidset_contains(&options->gitmodules_done, oid))
1256                         continue;
1257
1258                 buf = read_object_file(oid, &type, &size);
1259                 if (!buf) {
1260                         if (is_promisor_object(oid))
1261                                 continue;
1262                         ret |= report(options,
1263                                       oid, OBJ_BLOB,
1264                                       FSCK_MSG_GITMODULES_MISSING,
1265                                       "unable to read .gitmodules blob");
1266                         continue;
1267                 }
1268
1269                 if (type == OBJ_BLOB)
1270                         ret |= fsck_blob(oid, buf, size, options);
1271                 else
1272                         ret |= report(options,
1273                                       oid, type,
1274                                       FSCK_MSG_GITMODULES_BLOB,
1275                                       "non-blob found at .gitmodules");
1276                 free(buf);
1277         }
1278
1279
1280         oidset_clear(&options->gitmodules_found);
1281         oidset_clear(&options->gitmodules_done);
1282         return ret;
1283 }
1284
1285 int git_fsck_config(const char *var, const char *value, void *cb)
1286 {
1287         struct fsck_options *options = cb;
1288         if (strcmp(var, "fsck.skiplist") == 0) {
1289                 const char *path;
1290                 struct strbuf sb = STRBUF_INIT;
1291
1292                 if (git_config_pathname(&path, var, value))
1293                         return 1;
1294                 strbuf_addf(&sb, "skiplist=%s", path);
1295                 free((char *)path);
1296                 fsck_set_msg_types(options, sb.buf);
1297                 strbuf_release(&sb);
1298                 return 0;
1299         }
1300
1301         if (skip_prefix(var, "fsck.", &var)) {
1302                 fsck_set_msg_type(options, var, value);
1303                 return 0;
1304         }
1305
1306         return git_default_config(var, value, cb);
1307 }
1308
1309 /*
1310  * Custom error callbacks that are used in more than one place.
1311  */
1312
1313 int fsck_error_cb_print_missing_gitmodules(struct fsck_options *o,
1314                                            const struct object_id *oid,
1315                                            enum object_type object_type,
1316                                            enum fsck_msg_type msg_type,
1317                                            enum fsck_msg_id msg_id,
1318                                            const char *message)
1319 {
1320         if (msg_id == FSCK_MSG_GITMODULES_MISSING) {
1321                 puts(oid_to_hex(oid));
1322                 return 0;
1323         }
1324         return fsck_error_function(o, oid, object_type, msg_type, msg_id, message);
1325 }