fsck.c: call parse_msg_type() early in fsck_set_msg_type()
[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 static struct oidset gitmodules_found = OIDSET_INIT;
23 static struct oidset gitmodules_done = OIDSET_INIT;
24
25 #define FOREACH_MSG_ID(FUNC) \
26         /* fatal errors */ \
27         FUNC(NUL_IN_HEADER, FATAL) \
28         FUNC(UNTERMINATED_HEADER, FATAL) \
29         /* errors */ \
30         FUNC(BAD_DATE, ERROR) \
31         FUNC(BAD_DATE_OVERFLOW, ERROR) \
32         FUNC(BAD_EMAIL, ERROR) \
33         FUNC(BAD_NAME, ERROR) \
34         FUNC(BAD_OBJECT_SHA1, ERROR) \
35         FUNC(BAD_PARENT_SHA1, ERROR) \
36         FUNC(BAD_TAG_OBJECT, ERROR) \
37         FUNC(BAD_TIMEZONE, ERROR) \
38         FUNC(BAD_TREE, ERROR) \
39         FUNC(BAD_TREE_SHA1, ERROR) \
40         FUNC(BAD_TYPE, ERROR) \
41         FUNC(DUPLICATE_ENTRIES, ERROR) \
42         FUNC(MISSING_AUTHOR, ERROR) \
43         FUNC(MISSING_COMMITTER, ERROR) \
44         FUNC(MISSING_EMAIL, ERROR) \
45         FUNC(MISSING_NAME_BEFORE_EMAIL, ERROR) \
46         FUNC(MISSING_OBJECT, ERROR) \
47         FUNC(MISSING_SPACE_BEFORE_DATE, ERROR) \
48         FUNC(MISSING_SPACE_BEFORE_EMAIL, ERROR) \
49         FUNC(MISSING_TAG, ERROR) \
50         FUNC(MISSING_TAG_ENTRY, ERROR) \
51         FUNC(MISSING_TREE, ERROR) \
52         FUNC(MISSING_TREE_OBJECT, ERROR) \
53         FUNC(MISSING_TYPE, ERROR) \
54         FUNC(MISSING_TYPE_ENTRY, ERROR) \
55         FUNC(MULTIPLE_AUTHORS, ERROR) \
56         FUNC(TREE_NOT_SORTED, ERROR) \
57         FUNC(UNKNOWN_TYPE, ERROR) \
58         FUNC(ZERO_PADDED_DATE, ERROR) \
59         FUNC(GITMODULES_MISSING, ERROR) \
60         FUNC(GITMODULES_BLOB, ERROR) \
61         FUNC(GITMODULES_LARGE, ERROR) \
62         FUNC(GITMODULES_NAME, ERROR) \
63         FUNC(GITMODULES_SYMLINK, ERROR) \
64         FUNC(GITMODULES_URL, ERROR) \
65         FUNC(GITMODULES_PATH, ERROR) \
66         FUNC(GITMODULES_UPDATE, ERROR) \
67         /* warnings */ \
68         FUNC(BAD_FILEMODE, WARN) \
69         FUNC(EMPTY_NAME, WARN) \
70         FUNC(FULL_PATHNAME, WARN) \
71         FUNC(HAS_DOT, WARN) \
72         FUNC(HAS_DOTDOT, WARN) \
73         FUNC(HAS_DOTGIT, WARN) \
74         FUNC(NULL_SHA1, WARN) \
75         FUNC(ZERO_PADDED_FILEMODE, WARN) \
76         FUNC(NUL_IN_COMMIT, WARN) \
77         /* infos (reported as warnings, but ignored by default) */ \
78         FUNC(GITMODULES_PARSE, INFO) \
79         FUNC(BAD_TAG_NAME, INFO) \
80         FUNC(MISSING_TAGGER_ENTRY, INFO) \
81         /* ignored (elevated when requested) */ \
82         FUNC(EXTRA_HEADER_ENTRY, IGNORE)
83
84 #define MSG_ID(id, msg_type) FSCK_MSG_##id,
85 enum fsck_msg_id {
86         FOREACH_MSG_ID(MSG_ID)
87         FSCK_MSG_MAX
88 };
89 #undef MSG_ID
90
91 #define STR(x) #x
92 #define MSG_ID(id, msg_type) { STR(id), NULL, NULL, FSCK_##msg_type },
93 static struct {
94         const char *id_string;
95         const char *downcased;
96         const char *camelcased;
97         enum fsck_msg_type msg_type;
98 } msg_id_info[FSCK_MSG_MAX + 1] = {
99         FOREACH_MSG_ID(MSG_ID)
100         { NULL, NULL, NULL, -1 }
101 };
102 #undef MSG_ID
103
104 static void prepare_msg_ids(void)
105 {
106         int i;
107
108         if (msg_id_info[0].downcased)
109                 return;
110
111         /* convert id_string to lower case, without underscores. */
112         for (i = 0; i < FSCK_MSG_MAX; i++) {
113                 const char *p = msg_id_info[i].id_string;
114                 int len = strlen(p);
115                 char *q = xmalloc(len);
116
117                 msg_id_info[i].downcased = q;
118                 while (*p)
119                         if (*p == '_')
120                                 p++;
121                         else
122                                 *(q)++ = tolower(*(p)++);
123                 *q = '\0';
124
125                 p = msg_id_info[i].id_string;
126                 q = xmalloc(len);
127                 msg_id_info[i].camelcased = q;
128                 while (*p) {
129                         if (*p == '_') {
130                                 p++;
131                                 if (*p)
132                                         *q++ = *p++;
133                         } else {
134                                 *q++ = tolower(*p++);
135                         }
136                 }
137                 *q = '\0';
138         }
139 }
140
141 static int parse_msg_id(const char *text)
142 {
143         int i;
144
145         prepare_msg_ids();
146
147         for (i = 0; i < FSCK_MSG_MAX; i++)
148                 if (!strcmp(text, msg_id_info[i].downcased))
149                         return i;
150
151         return -1;
152 }
153
154 void list_config_fsck_msg_ids(struct string_list *list, const char *prefix)
155 {
156         int i;
157
158         prepare_msg_ids();
159
160         for (i = 0; i < FSCK_MSG_MAX; i++)
161                 list_config_item(list, prefix, msg_id_info[i].camelcased);
162 }
163
164 static enum fsck_msg_type fsck_msg_type(enum fsck_msg_id msg_id,
165         struct fsck_options *options)
166 {
167         assert(msg_id >= 0 && msg_id < FSCK_MSG_MAX);
168
169         if (!options->msg_type) {
170                 enum fsck_msg_type msg_type = msg_id_info[msg_id].msg_type;
171
172                 if (options->strict && msg_type == FSCK_WARN)
173                         msg_type = FSCK_ERROR;
174                 return msg_type;
175         }
176
177         return options->msg_type[msg_id];
178 }
179
180 static enum fsck_msg_type parse_msg_type(const char *str)
181 {
182         if (!strcmp(str, "error"))
183                 return FSCK_ERROR;
184         else if (!strcmp(str, "warn"))
185                 return FSCK_WARN;
186         else if (!strcmp(str, "ignore"))
187                 return FSCK_IGNORE;
188         else
189                 die("Unknown fsck message type: '%s'", str);
190 }
191
192 int is_valid_msg_type(const char *msg_id, const char *msg_type)
193 {
194         if (parse_msg_id(msg_id) < 0)
195                 return 0;
196         parse_msg_type(msg_type);
197         return 1;
198 }
199
200 void fsck_set_msg_type(struct fsck_options *options,
201                        const char *msg_id_str, const char *msg_type_str)
202 {
203         int msg_id = parse_msg_id(msg_id_str);
204         enum fsck_msg_type msg_type = parse_msg_type(msg_type_str);
205
206         if (msg_id < 0)
207                 die("Unhandled message id: %s", msg_id_str);
208
209         if (msg_type != FSCK_ERROR && msg_id_info[msg_id].msg_type == FSCK_FATAL)
210                 die("Cannot demote %s to %s", msg_id_str, msg_type_str);
211
212         if (!options->msg_type) {
213                 int i;
214                 enum fsck_msg_type *severity;
215                 ALLOC_ARRAY(severity, FSCK_MSG_MAX);
216                 for (i = 0; i < FSCK_MSG_MAX; i++)
217                         severity[i] = fsck_msg_type(i, options);
218                 options->msg_type = severity;
219         }
220
221         options->msg_type[msg_id] = msg_type;
222 }
223
224 void fsck_set_msg_types(struct fsck_options *options, const char *values)
225 {
226         char *buf = xstrdup(values), *to_free = buf;
227         int done = 0;
228
229         while (!done) {
230                 int len = strcspn(buf, " ,|"), equal;
231
232                 done = !buf[len];
233                 if (!len) {
234                         buf++;
235                         continue;
236                 }
237                 buf[len] = '\0';
238
239                 for (equal = 0;
240                      equal < len && buf[equal] != '=' && buf[equal] != ':';
241                      equal++)
242                         buf[equal] = tolower(buf[equal]);
243                 buf[equal] = '\0';
244
245                 if (!strcmp(buf, "skiplist")) {
246                         if (equal == len)
247                                 die("skiplist requires a path");
248                         oidset_parse_file(&options->skiplist, buf + equal + 1);
249                         buf += len + 1;
250                         continue;
251                 }
252
253                 if (equal == len)
254                         die("Missing '=': '%s'", buf);
255
256                 fsck_set_msg_type(options, buf, buf + equal + 1);
257                 buf += len + 1;
258         }
259         free(to_free);
260 }
261
262 static int object_on_skiplist(struct fsck_options *opts,
263                               const struct object_id *oid)
264 {
265         return opts && oid && oidset_contains(&opts->skiplist, oid);
266 }
267
268 __attribute__((format (printf, 5, 6)))
269 static int report(struct fsck_options *options,
270                   const struct object_id *oid, enum object_type object_type,
271                   enum fsck_msg_id msg_id, const char *fmt, ...)
272 {
273         va_list ap;
274         struct strbuf sb = STRBUF_INIT;
275         enum fsck_msg_type msg_type = fsck_msg_type(msg_id, options);
276         int result;
277
278         if (msg_type == FSCK_IGNORE)
279                 return 0;
280
281         if (object_on_skiplist(options, oid))
282                 return 0;
283
284         if (msg_type == FSCK_FATAL)
285                 msg_type = FSCK_ERROR;
286         else if (msg_type == FSCK_INFO)
287                 msg_type = FSCK_WARN;
288
289         prepare_msg_ids();
290         strbuf_addf(&sb, "%s: ", msg_id_info[msg_id].camelcased);
291
292         va_start(ap, fmt);
293         strbuf_vaddf(&sb, fmt, ap);
294         result = options->error_func(options, oid, object_type,
295                                      msg_type, sb.buf);
296         strbuf_release(&sb);
297         va_end(ap);
298
299         return result;
300 }
301
302 void fsck_enable_object_names(struct fsck_options *options)
303 {
304         if (!options->object_names)
305                 options->object_names = kh_init_oid_map();
306 }
307
308 const char *fsck_get_object_name(struct fsck_options *options,
309                                  const struct object_id *oid)
310 {
311         khiter_t pos;
312         if (!options->object_names)
313                 return NULL;
314         pos = kh_get_oid_map(options->object_names, *oid);
315         if (pos >= kh_end(options->object_names))
316                 return NULL;
317         return kh_value(options->object_names, pos);
318 }
319
320 void fsck_put_object_name(struct fsck_options *options,
321                           const struct object_id *oid,
322                           const char *fmt, ...)
323 {
324         va_list ap;
325         struct strbuf buf = STRBUF_INIT;
326         khiter_t pos;
327         int hashret;
328
329         if (!options->object_names)
330                 return;
331
332         pos = kh_put_oid_map(options->object_names, *oid, &hashret);
333         if (!hashret)
334                 return;
335         va_start(ap, fmt);
336         strbuf_vaddf(&buf, fmt, ap);
337         kh_value(options->object_names, pos) = strbuf_detach(&buf, NULL);
338         va_end(ap);
339 }
340
341 const char *fsck_describe_object(struct fsck_options *options,
342                                  const struct object_id *oid)
343 {
344         static struct strbuf bufs[] = {
345                 STRBUF_INIT, STRBUF_INIT, STRBUF_INIT, STRBUF_INIT
346         };
347         static int b = 0;
348         struct strbuf *buf;
349         const char *name = fsck_get_object_name(options, oid);
350
351         buf = bufs + b;
352         b = (b + 1) % ARRAY_SIZE(bufs);
353         strbuf_reset(buf);
354         strbuf_addstr(buf, oid_to_hex(oid));
355         if (name)
356                 strbuf_addf(buf, " (%s)", name);
357
358         return buf->buf;
359 }
360
361 static int fsck_walk_tree(struct tree *tree, void *data, struct fsck_options *options)
362 {
363         struct tree_desc desc;
364         struct name_entry entry;
365         int res = 0;
366         const char *name;
367
368         if (parse_tree(tree))
369                 return -1;
370
371         name = fsck_get_object_name(options, &tree->object.oid);
372         if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
373                 return -1;
374         while (tree_entry_gently(&desc, &entry)) {
375                 struct object *obj;
376                 int result;
377
378                 if (S_ISGITLINK(entry.mode))
379                         continue;
380
381                 if (S_ISDIR(entry.mode)) {
382                         obj = (struct object *)lookup_tree(the_repository, &entry.oid);
383                         if (name && obj)
384                                 fsck_put_object_name(options, &entry.oid, "%s%s/",
385                                                      name, entry.path);
386                         result = options->walk(obj, OBJ_TREE, data, options);
387                 }
388                 else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode)) {
389                         obj = (struct object *)lookup_blob(the_repository, &entry.oid);
390                         if (name && obj)
391                                 fsck_put_object_name(options, &entry.oid, "%s%s",
392                                                      name, entry.path);
393                         result = options->walk(obj, OBJ_BLOB, data, options);
394                 }
395                 else {
396                         result = error("in tree %s: entry %s has bad mode %.6o",
397                                        fsck_describe_object(options, &tree->object.oid),
398                                        entry.path, entry.mode);
399                 }
400                 if (result < 0)
401                         return result;
402                 if (!res)
403                         res = result;
404         }
405         return res;
406 }
407
408 static int fsck_walk_commit(struct commit *commit, void *data, struct fsck_options *options)
409 {
410         int counter = 0, generation = 0, name_prefix_len = 0;
411         struct commit_list *parents;
412         int res;
413         int result;
414         const char *name;
415
416         if (parse_commit(commit))
417                 return -1;
418
419         name = fsck_get_object_name(options, &commit->object.oid);
420         if (name)
421                 fsck_put_object_name(options, get_commit_tree_oid(commit),
422                                      "%s:", name);
423
424         result = options->walk((struct object *)get_commit_tree(commit),
425                                OBJ_TREE, data, options);
426         if (result < 0)
427                 return result;
428         res = result;
429
430         parents = commit->parents;
431         if (name && parents) {
432                 int len = strlen(name), power;
433
434                 if (len && name[len - 1] == '^') {
435                         generation = 1;
436                         name_prefix_len = len - 1;
437                 }
438                 else { /* parse ~<generation> suffix */
439                         for (generation = 0, power = 1;
440                              len && isdigit(name[len - 1]);
441                              power *= 10)
442                                 generation += power * (name[--len] - '0');
443                         if (power > 1 && len && name[len - 1] == '~')
444                                 name_prefix_len = len - 1;
445                         else {
446                                 /* Maybe a non-first parent, e.g. HEAD^2 */
447                                 generation = 0;
448                                 name_prefix_len = len;
449                         }
450                 }
451         }
452
453         while (parents) {
454                 if (name) {
455                         struct object_id *oid = &parents->item->object.oid;
456
457                         if (counter++)
458                                 fsck_put_object_name(options, oid, "%s^%d",
459                                                      name, counter);
460                         else if (generation > 0)
461                                 fsck_put_object_name(options, oid, "%.*s~%d",
462                                                      name_prefix_len, name,
463                                                      generation + 1);
464                         else
465                                 fsck_put_object_name(options, oid, "%s^", name);
466                 }
467                 result = options->walk((struct object *)parents->item, OBJ_COMMIT, data, options);
468                 if (result < 0)
469                         return result;
470                 if (!res)
471                         res = result;
472                 parents = parents->next;
473         }
474         return res;
475 }
476
477 static int fsck_walk_tag(struct tag *tag, void *data, struct fsck_options *options)
478 {
479         const char *name = fsck_get_object_name(options, &tag->object.oid);
480
481         if (parse_tag(tag))
482                 return -1;
483         if (name)
484                 fsck_put_object_name(options, &tag->tagged->oid, "%s", name);
485         return options->walk(tag->tagged, OBJ_ANY, data, options);
486 }
487
488 int fsck_walk(struct object *obj, void *data, struct fsck_options *options)
489 {
490         if (!obj)
491                 return -1;
492
493         if (obj->type == OBJ_NONE)
494                 parse_object(the_repository, &obj->oid);
495
496         switch (obj->type) {
497         case OBJ_BLOB:
498                 return 0;
499         case OBJ_TREE:
500                 return fsck_walk_tree((struct tree *)obj, data, options);
501         case OBJ_COMMIT:
502                 return fsck_walk_commit((struct commit *)obj, data, options);
503         case OBJ_TAG:
504                 return fsck_walk_tag((struct tag *)obj, data, options);
505         default:
506                 error("Unknown object type for %s",
507                       fsck_describe_object(options, &obj->oid));
508                 return -1;
509         }
510 }
511
512 struct name_stack {
513         const char **names;
514         size_t nr, alloc;
515 };
516
517 static void name_stack_push(struct name_stack *stack, const char *name)
518 {
519         ALLOC_GROW(stack->names, stack->nr + 1, stack->alloc);
520         stack->names[stack->nr++] = name;
521 }
522
523 static const char *name_stack_pop(struct name_stack *stack)
524 {
525         return stack->nr ? stack->names[--stack->nr] : NULL;
526 }
527
528 static void name_stack_clear(struct name_stack *stack)
529 {
530         FREE_AND_NULL(stack->names);
531         stack->nr = stack->alloc = 0;
532 }
533
534 /*
535  * The entries in a tree are ordered in the _path_ order,
536  * which means that a directory entry is ordered by adding
537  * a slash to the end of it.
538  *
539  * So a directory called "a" is ordered _after_ a file
540  * called "a.c", because "a/" sorts after "a.c".
541  */
542 #define TREE_UNORDERED (-1)
543 #define TREE_HAS_DUPS  (-2)
544
545 static int is_less_than_slash(unsigned char c)
546 {
547         return '\0' < c && c < '/';
548 }
549
550 static int verify_ordered(unsigned mode1, const char *name1,
551                           unsigned mode2, const char *name2,
552                           struct name_stack *candidates)
553 {
554         int len1 = strlen(name1);
555         int len2 = strlen(name2);
556         int len = len1 < len2 ? len1 : len2;
557         unsigned char c1, c2;
558         int cmp;
559
560         cmp = memcmp(name1, name2, len);
561         if (cmp < 0)
562                 return 0;
563         if (cmp > 0)
564                 return TREE_UNORDERED;
565
566         /*
567          * Ok, the first <len> characters are the same.
568          * Now we need to order the next one, but turn
569          * a '\0' into a '/' for a directory entry.
570          */
571         c1 = name1[len];
572         c2 = name2[len];
573         if (!c1 && !c2)
574                 /*
575                  * git-write-tree used to write out a nonsense tree that has
576                  * entries with the same name, one blob and one tree.  Make
577                  * sure we do not have duplicate entries.
578                  */
579                 return TREE_HAS_DUPS;
580         if (!c1 && S_ISDIR(mode1))
581                 c1 = '/';
582         if (!c2 && S_ISDIR(mode2))
583                 c2 = '/';
584
585         /*
586          * There can be non-consecutive duplicates due to the implicitly
587          * added slash, e.g.:
588          *
589          *   foo
590          *   foo.bar
591          *   foo.bar.baz
592          *   foo.bar/
593          *   foo/
594          *
595          * Record non-directory candidates (like "foo" and "foo.bar" in
596          * the example) on a stack and check directory candidates (like
597          * foo/" and "foo.bar/") against that stack.
598          */
599         if (!c1 && is_less_than_slash(c2)) {
600                 name_stack_push(candidates, name1);
601         } else if (c2 == '/' && is_less_than_slash(c1)) {
602                 for (;;) {
603                         const char *p;
604                         const char *f_name = name_stack_pop(candidates);
605
606                         if (!f_name)
607                                 break;
608                         if (!skip_prefix(name2, f_name, &p))
609                                 continue;
610                         if (!*p)
611                                 return TREE_HAS_DUPS;
612                         if (is_less_than_slash(*p)) {
613                                 name_stack_push(candidates, f_name);
614                                 break;
615                         }
616                 }
617         }
618
619         return c1 < c2 ? 0 : TREE_UNORDERED;
620 }
621
622 static int fsck_tree(const struct object_id *oid,
623                      const char *buffer, unsigned long size,
624                      struct fsck_options *options)
625 {
626         int retval = 0;
627         int has_null_sha1 = 0;
628         int has_full_path = 0;
629         int has_empty_name = 0;
630         int has_dot = 0;
631         int has_dotdot = 0;
632         int has_dotgit = 0;
633         int has_zero_pad = 0;
634         int has_bad_modes = 0;
635         int has_dup_entries = 0;
636         int not_properly_sorted = 0;
637         struct tree_desc desc;
638         unsigned o_mode;
639         const char *o_name;
640         struct name_stack df_dup_candidates = { NULL };
641
642         if (init_tree_desc_gently(&desc, buffer, size)) {
643                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
644                 return retval;
645         }
646
647         o_mode = 0;
648         o_name = NULL;
649
650         while (desc.size) {
651                 unsigned short mode;
652                 const char *name, *backslash;
653                 const struct object_id *oid;
654
655                 oid = tree_entry_extract(&desc, &name, &mode);
656
657                 has_null_sha1 |= is_null_oid(oid);
658                 has_full_path |= !!strchr(name, '/');
659                 has_empty_name |= !*name;
660                 has_dot |= !strcmp(name, ".");
661                 has_dotdot |= !strcmp(name, "..");
662                 has_dotgit |= is_hfs_dotgit(name) || is_ntfs_dotgit(name);
663                 has_zero_pad |= *(char *)desc.buffer == '0';
664
665                 if (is_hfs_dotgitmodules(name) || is_ntfs_dotgitmodules(name)) {
666                         if (!S_ISLNK(mode))
667                                 oidset_insert(&gitmodules_found, oid);
668                         else
669                                 retval += report(options,
670                                                  oid, OBJ_TREE,
671                                                  FSCK_MSG_GITMODULES_SYMLINK,
672                                                  ".gitmodules is a symbolic link");
673                 }
674
675                 if ((backslash = strchr(name, '\\'))) {
676                         while (backslash) {
677                                 backslash++;
678                                 has_dotgit |= is_ntfs_dotgit(backslash);
679                                 if (is_ntfs_dotgitmodules(backslash)) {
680                                         if (!S_ISLNK(mode))
681                                                 oidset_insert(&gitmodules_found, oid);
682                                         else
683                                                 retval += report(options, oid, OBJ_TREE,
684                                                                  FSCK_MSG_GITMODULES_SYMLINK,
685                                                                  ".gitmodules is a symbolic link");
686                                 }
687                                 backslash = strchr(backslash, '\\');
688                         }
689                 }
690
691                 if (update_tree_entry_gently(&desc)) {
692                         retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
693                         break;
694                 }
695
696                 switch (mode) {
697                 /*
698                  * Standard modes..
699                  */
700                 case S_IFREG | 0755:
701                 case S_IFREG | 0644:
702                 case S_IFLNK:
703                 case S_IFDIR:
704                 case S_IFGITLINK:
705                         break;
706                 /*
707                  * This is nonstandard, but we had a few of these
708                  * early on when we honored the full set of mode
709                  * bits..
710                  */
711                 case S_IFREG | 0664:
712                         if (!options->strict)
713                                 break;
714                         /* fallthrough */
715                 default:
716                         has_bad_modes = 1;
717                 }
718
719                 if (o_name) {
720                         switch (verify_ordered(o_mode, o_name, mode, name,
721                                                &df_dup_candidates)) {
722                         case TREE_UNORDERED:
723                                 not_properly_sorted = 1;
724                                 break;
725                         case TREE_HAS_DUPS:
726                                 has_dup_entries = 1;
727                                 break;
728                         default:
729                                 break;
730                         }
731                 }
732
733                 o_mode = mode;
734                 o_name = name;
735         }
736
737         name_stack_clear(&df_dup_candidates);
738
739         if (has_null_sha1)
740                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_NULL_SHA1, "contains entries pointing to null sha1");
741         if (has_full_path)
742                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_FULL_PATHNAME, "contains full pathnames");
743         if (has_empty_name)
744                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_EMPTY_NAME, "contains empty pathname");
745         if (has_dot)
746                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOT, "contains '.'");
747         if (has_dotdot)
748                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOTDOT, "contains '..'");
749         if (has_dotgit)
750                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOTGIT, "contains '.git'");
751         if (has_zero_pad)
752                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_ZERO_PADDED_FILEMODE, "contains zero-padded file modes");
753         if (has_bad_modes)
754                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_FILEMODE, "contains bad file modes");
755         if (has_dup_entries)
756                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_DUPLICATE_ENTRIES, "contains duplicate file entries");
757         if (not_properly_sorted)
758                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_TREE_NOT_SORTED, "not properly sorted");
759         return retval;
760 }
761
762 static int verify_headers(const void *data, unsigned long size,
763                           const struct object_id *oid, enum object_type type,
764                           struct fsck_options *options)
765 {
766         const char *buffer = (const char *)data;
767         unsigned long i;
768
769         for (i = 0; i < size; i++) {
770                 switch (buffer[i]) {
771                 case '\0':
772                         return report(options, oid, type,
773                                 FSCK_MSG_NUL_IN_HEADER,
774                                 "unterminated header: NUL at offset %ld", i);
775                 case '\n':
776                         if (i + 1 < size && buffer[i + 1] == '\n')
777                                 return 0;
778                 }
779         }
780
781         /*
782          * We did not find double-LF that separates the header
783          * and the body.  Not having a body is not a crime but
784          * we do want to see the terminating LF for the last header
785          * line.
786          */
787         if (size && buffer[size - 1] == '\n')
788                 return 0;
789
790         return report(options, oid, type,
791                 FSCK_MSG_UNTERMINATED_HEADER, "unterminated header");
792 }
793
794 static int fsck_ident(const char **ident,
795                       const struct object_id *oid, enum object_type type,
796                       struct fsck_options *options)
797 {
798         const char *p = *ident;
799         char *end;
800
801         *ident = strchrnul(*ident, '\n');
802         if (**ident == '\n')
803                 (*ident)++;
804
805         if (*p == '<')
806                 return report(options, oid, type, FSCK_MSG_MISSING_NAME_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
807         p += strcspn(p, "<>\n");
808         if (*p == '>')
809                 return report(options, oid, type, FSCK_MSG_BAD_NAME, "invalid author/committer line - bad name");
810         if (*p != '<')
811                 return report(options, oid, type, FSCK_MSG_MISSING_EMAIL, "invalid author/committer line - missing email");
812         if (p[-1] != ' ')
813                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
814         p++;
815         p += strcspn(p, "<>\n");
816         if (*p != '>')
817                 return report(options, oid, type, FSCK_MSG_BAD_EMAIL, "invalid author/committer line - bad email");
818         p++;
819         if (*p != ' ')
820                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_DATE, "invalid author/committer line - missing space before date");
821         p++;
822         if (*p == '0' && p[1] != ' ')
823                 return report(options, oid, type, FSCK_MSG_ZERO_PADDED_DATE, "invalid author/committer line - zero-padded date");
824         if (date_overflows(parse_timestamp(p, &end, 10)))
825                 return report(options, oid, type, FSCK_MSG_BAD_DATE_OVERFLOW, "invalid author/committer line - date causes integer overflow");
826         if ((end == p || *end != ' '))
827                 return report(options, oid, type, FSCK_MSG_BAD_DATE, "invalid author/committer line - bad date");
828         p = end + 1;
829         if ((*p != '+' && *p != '-') ||
830             !isdigit(p[1]) ||
831             !isdigit(p[2]) ||
832             !isdigit(p[3]) ||
833             !isdigit(p[4]) ||
834             (p[5] != '\n'))
835                 return report(options, oid, type, FSCK_MSG_BAD_TIMEZONE, "invalid author/committer line - bad time zone");
836         p += 6;
837         return 0;
838 }
839
840 static int fsck_commit(const struct object_id *oid,
841                        const char *buffer, unsigned long size,
842                        struct fsck_options *options)
843 {
844         struct object_id tree_oid, parent_oid;
845         unsigned author_count;
846         int err;
847         const char *buffer_begin = buffer;
848         const char *p;
849
850         if (verify_headers(buffer, size, oid, OBJ_COMMIT, options))
851                 return -1;
852
853         if (!skip_prefix(buffer, "tree ", &buffer))
854                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_TREE, "invalid format - expected 'tree' line");
855         if (parse_oid_hex(buffer, &tree_oid, &p) || *p != '\n') {
856                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_TREE_SHA1, "invalid 'tree' line format - bad sha1");
857                 if (err)
858                         return err;
859         }
860         buffer = p + 1;
861         while (skip_prefix(buffer, "parent ", &buffer)) {
862                 if (parse_oid_hex(buffer, &parent_oid, &p) || *p != '\n') {
863                         err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_PARENT_SHA1, "invalid 'parent' line format - bad sha1");
864                         if (err)
865                                 return err;
866                 }
867                 buffer = p + 1;
868         }
869         author_count = 0;
870         while (skip_prefix(buffer, "author ", &buffer)) {
871                 author_count++;
872                 err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
873                 if (err)
874                         return err;
875         }
876         if (author_count < 1)
877                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_AUTHOR, "invalid format - expected 'author' line");
878         else if (author_count > 1)
879                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MULTIPLE_AUTHORS, "invalid format - multiple 'author' lines");
880         if (err)
881                 return err;
882         if (!skip_prefix(buffer, "committer ", &buffer))
883                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_COMMITTER, "invalid format - expected 'committer' line");
884         err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
885         if (err)
886                 return err;
887         if (memchr(buffer_begin, '\0', size)) {
888                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_NUL_IN_COMMIT,
889                              "NUL byte in the commit object body");
890                 if (err)
891                         return err;
892         }
893         return 0;
894 }
895
896 static int fsck_tag(const struct object_id *oid, const char *buffer,
897                     unsigned long size, struct fsck_options *options)
898 {
899         struct object_id tagged_oid;
900         int tagged_type;
901         return fsck_tag_standalone(oid, buffer, size, options, &tagged_oid,
902                                    &tagged_type);
903 }
904
905 int fsck_tag_standalone(const struct object_id *oid, const char *buffer,
906                         unsigned long size, struct fsck_options *options,
907                         struct object_id *tagged_oid,
908                         int *tagged_type)
909 {
910         int ret = 0;
911         char *eol;
912         struct strbuf sb = STRBUF_INIT;
913         const char *p;
914
915         ret = verify_headers(buffer, size, oid, OBJ_TAG, options);
916         if (ret)
917                 goto done;
918
919         if (!skip_prefix(buffer, "object ", &buffer)) {
920                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_OBJECT, "invalid format - expected 'object' line");
921                 goto done;
922         }
923         if (parse_oid_hex(buffer, tagged_oid, &p) || *p != '\n') {
924                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_OBJECT_SHA1, "invalid 'object' line format - bad sha1");
925                 if (ret)
926                         goto done;
927         }
928         buffer = p + 1;
929
930         if (!skip_prefix(buffer, "type ", &buffer)) {
931                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE_ENTRY, "invalid format - expected 'type' line");
932                 goto done;
933         }
934         eol = strchr(buffer, '\n');
935         if (!eol) {
936                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE, "invalid format - unexpected end after 'type' line");
937                 goto done;
938         }
939         *tagged_type = type_from_string_gently(buffer, eol - buffer, 1);
940         if (*tagged_type < 0)
941                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_TYPE, "invalid 'type' value");
942         if (ret)
943                 goto done;
944         buffer = eol + 1;
945
946         if (!skip_prefix(buffer, "tag ", &buffer)) {
947                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG_ENTRY, "invalid format - expected 'tag' line");
948                 goto done;
949         }
950         eol = strchr(buffer, '\n');
951         if (!eol) {
952                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG, "invalid format - unexpected end after 'type' line");
953                 goto done;
954         }
955         strbuf_addf(&sb, "refs/tags/%.*s", (int)(eol - buffer), buffer);
956         if (check_refname_format(sb.buf, 0)) {
957                 ret = report(options, oid, OBJ_TAG,
958                              FSCK_MSG_BAD_TAG_NAME,
959                              "invalid 'tag' name: %.*s",
960                              (int)(eol - buffer), buffer);
961                 if (ret)
962                         goto done;
963         }
964         buffer = eol + 1;
965
966         if (!skip_prefix(buffer, "tagger ", &buffer)) {
967                 /* early tags do not contain 'tagger' lines; warn only */
968                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAGGER_ENTRY, "invalid format - expected 'tagger' line");
969                 if (ret)
970                         goto done;
971         }
972         else
973                 ret = fsck_ident(&buffer, oid, OBJ_TAG, options);
974         if (!*buffer)
975                 goto done;
976
977         if (!starts_with(buffer, "\n")) {
978                 /*
979                  * The verify_headers() check will allow
980                  * e.g. "[...]tagger <tagger>\nsome
981                  * garbage\n\nmessage" to pass, thinking "some
982                  * garbage" could be a custom header. E.g. "mktag"
983                  * doesn't want any unknown headers.
984                  */
985                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_EXTRA_HEADER_ENTRY, "invalid format - extra header(s) after 'tagger'");
986                 if (ret)
987                         goto done;
988         }
989
990 done:
991         strbuf_release(&sb);
992         return ret;
993 }
994
995 /*
996  * Like builtin/submodule--helper.c's starts_with_dot_slash, but without
997  * relying on the platform-dependent is_dir_sep helper.
998  *
999  * This is for use in checking whether a submodule URL is interpreted as
1000  * relative to the current directory on any platform, since \ is a
1001  * directory separator on Windows but not on other platforms.
1002  */
1003 static int starts_with_dot_slash(const char *str)
1004 {
1005         return str[0] == '.' && (str[1] == '/' || str[1] == '\\');
1006 }
1007
1008 /*
1009  * Like starts_with_dot_slash, this is a variant of submodule--helper's
1010  * helper of the same name with the twist that it accepts backslash as a
1011  * directory separator even on non-Windows platforms.
1012  */
1013 static int starts_with_dot_dot_slash(const char *str)
1014 {
1015         return str[0] == '.' && starts_with_dot_slash(str + 1);
1016 }
1017
1018 static int submodule_url_is_relative(const char *url)
1019 {
1020         return starts_with_dot_slash(url) || starts_with_dot_dot_slash(url);
1021 }
1022
1023 /*
1024  * Count directory components that a relative submodule URL should chop
1025  * from the remote_url it is to be resolved against.
1026  *
1027  * In other words, this counts "../" components at the start of a
1028  * submodule URL.
1029  *
1030  * Returns the number of directory components to chop and writes a
1031  * pointer to the next character of url after all leading "./" and
1032  * "../" components to out.
1033  */
1034 static int count_leading_dotdots(const char *url, const char **out)
1035 {
1036         int result = 0;
1037         while (1) {
1038                 if (starts_with_dot_dot_slash(url)) {
1039                         result++;
1040                         url += strlen("../");
1041                         continue;
1042                 }
1043                 if (starts_with_dot_slash(url)) {
1044                         url += strlen("./");
1045                         continue;
1046                 }
1047                 *out = url;
1048                 return result;
1049         }
1050 }
1051 /*
1052  * Check whether a transport is implemented by git-remote-curl.
1053  *
1054  * If it is, returns 1 and writes the URL that would be passed to
1055  * git-remote-curl to the "out" parameter.
1056  *
1057  * Otherwise, returns 0 and leaves "out" untouched.
1058  *
1059  * Examples:
1060  *   http::https://example.com/repo.git -> 1, https://example.com/repo.git
1061  *   https://example.com/repo.git -> 1, https://example.com/repo.git
1062  *   git://example.com/repo.git -> 0
1063  *
1064  * This is for use in checking for previously exploitable bugs that
1065  * required a submodule URL to be passed to git-remote-curl.
1066  */
1067 static int url_to_curl_url(const char *url, const char **out)
1068 {
1069         /*
1070          * We don't need to check for case-aliases, "http.exe", and so
1071          * on because in the default configuration, is_transport_allowed
1072          * prevents URLs with those schemes from being cloned
1073          * automatically.
1074          */
1075         if (skip_prefix(url, "http::", out) ||
1076             skip_prefix(url, "https::", out) ||
1077             skip_prefix(url, "ftp::", out) ||
1078             skip_prefix(url, "ftps::", out))
1079                 return 1;
1080         if (starts_with(url, "http://") ||
1081             starts_with(url, "https://") ||
1082             starts_with(url, "ftp://") ||
1083             starts_with(url, "ftps://")) {
1084                 *out = url;
1085                 return 1;
1086         }
1087         return 0;
1088 }
1089
1090 static int check_submodule_url(const char *url)
1091 {
1092         const char *curl_url;
1093
1094         if (looks_like_command_line_option(url))
1095                 return -1;
1096
1097         if (submodule_url_is_relative(url) || starts_with(url, "git://")) {
1098                 char *decoded;
1099                 const char *next;
1100                 int has_nl;
1101
1102                 /*
1103                  * This could be appended to an http URL and url-decoded;
1104                  * check for malicious characters.
1105                  */
1106                 decoded = url_decode(url);
1107                 has_nl = !!strchr(decoded, '\n');
1108
1109                 free(decoded);
1110                 if (has_nl)
1111                         return -1;
1112
1113                 /*
1114                  * URLs which escape their root via "../" can overwrite
1115                  * the host field and previous components, resolving to
1116                  * URLs like https::example.com/submodule.git and
1117                  * https:///example.com/submodule.git that were
1118                  * susceptible to CVE-2020-11008.
1119                  */
1120                 if (count_leading_dotdots(url, &next) > 0 &&
1121                     (*next == ':' || *next == '/'))
1122                         return -1;
1123         }
1124
1125         else if (url_to_curl_url(url, &curl_url)) {
1126                 struct credential c = CREDENTIAL_INIT;
1127                 int ret = 0;
1128                 if (credential_from_url_gently(&c, curl_url, 1) ||
1129                     !*c.host)
1130                         ret = -1;
1131                 credential_clear(&c);
1132                 return ret;
1133         }
1134
1135         return 0;
1136 }
1137
1138 struct fsck_gitmodules_data {
1139         const struct object_id *oid;
1140         struct fsck_options *options;
1141         int ret;
1142 };
1143
1144 static int fsck_gitmodules_fn(const char *var, const char *value, void *vdata)
1145 {
1146         struct fsck_gitmodules_data *data = vdata;
1147         const char *subsection, *key;
1148         size_t subsection_len;
1149         char *name;
1150
1151         if (parse_config_key(var, "submodule", &subsection, &subsection_len, &key) < 0 ||
1152             !subsection)
1153                 return 0;
1154
1155         name = xmemdupz(subsection, subsection_len);
1156         if (check_submodule_name(name) < 0)
1157                 data->ret |= report(data->options,
1158                                     data->oid, OBJ_BLOB,
1159                                     FSCK_MSG_GITMODULES_NAME,
1160                                     "disallowed submodule name: %s",
1161                                     name);
1162         if (!strcmp(key, "url") && value &&
1163             check_submodule_url(value) < 0)
1164                 data->ret |= report(data->options,
1165                                     data->oid, OBJ_BLOB,
1166                                     FSCK_MSG_GITMODULES_URL,
1167                                     "disallowed submodule url: %s",
1168                                     value);
1169         if (!strcmp(key, "path") && value &&
1170             looks_like_command_line_option(value))
1171                 data->ret |= report(data->options,
1172                                     data->oid, OBJ_BLOB,
1173                                     FSCK_MSG_GITMODULES_PATH,
1174                                     "disallowed submodule path: %s",
1175                                     value);
1176         if (!strcmp(key, "update") && value &&
1177             parse_submodule_update_type(value) == SM_UPDATE_COMMAND)
1178                 data->ret |= report(data->options, data->oid, OBJ_BLOB,
1179                                     FSCK_MSG_GITMODULES_UPDATE,
1180                                     "disallowed submodule update setting: %s",
1181                                     value);
1182         free(name);
1183
1184         return 0;
1185 }
1186
1187 static int fsck_blob(const struct object_id *oid, const char *buf,
1188                      unsigned long size, struct fsck_options *options)
1189 {
1190         struct fsck_gitmodules_data data;
1191         struct config_options config_opts = { 0 };
1192
1193         if (!oidset_contains(&gitmodules_found, oid))
1194                 return 0;
1195         oidset_insert(&gitmodules_done, oid);
1196
1197         if (object_on_skiplist(options, oid))
1198                 return 0;
1199
1200         if (!buf) {
1201                 /*
1202                  * A missing buffer here is a sign that the caller found the
1203                  * blob too gigantic to load into memory. Let's just consider
1204                  * that an error.
1205                  */
1206                 return report(options, oid, OBJ_BLOB,
1207                               FSCK_MSG_GITMODULES_LARGE,
1208                               ".gitmodules too large to parse");
1209         }
1210
1211         data.oid = oid;
1212         data.options = options;
1213         data.ret = 0;
1214         config_opts.error_action = CONFIG_ERROR_SILENT;
1215         if (git_config_from_mem(fsck_gitmodules_fn, CONFIG_ORIGIN_BLOB,
1216                                 ".gitmodules", buf, size, &data, &config_opts))
1217                 data.ret |= report(options, oid, OBJ_BLOB,
1218                                    FSCK_MSG_GITMODULES_PARSE,
1219                                    "could not parse gitmodules blob");
1220
1221         return data.ret;
1222 }
1223
1224 int fsck_object(struct object *obj, void *data, unsigned long size,
1225         struct fsck_options *options)
1226 {
1227         if (!obj)
1228                 return report(options, NULL, OBJ_NONE, FSCK_MSG_BAD_OBJECT_SHA1, "no valid object to fsck");
1229
1230         if (obj->type == OBJ_BLOB)
1231                 return fsck_blob(&obj->oid, data, size, options);
1232         if (obj->type == OBJ_TREE)
1233                 return fsck_tree(&obj->oid, data, size, options);
1234         if (obj->type == OBJ_COMMIT)
1235                 return fsck_commit(&obj->oid, data, size, options);
1236         if (obj->type == OBJ_TAG)
1237                 return fsck_tag(&obj->oid, data, size, options);
1238
1239         return report(options, &obj->oid, obj->type,
1240                       FSCK_MSG_UNKNOWN_TYPE,
1241                       "unknown type '%d' (internal fsck error)",
1242                       obj->type);
1243 }
1244
1245 int fsck_error_function(struct fsck_options *o,
1246                         const struct object_id *oid,
1247                         enum object_type object_type,
1248                         enum fsck_msg_type msg_type, const char *message)
1249 {
1250         if (msg_type == FSCK_WARN) {
1251                 warning("object %s: %s", fsck_describe_object(o, oid), message);
1252                 return 0;
1253         }
1254         error("object %s: %s", fsck_describe_object(o, oid), message);
1255         return 1;
1256 }
1257
1258 void register_found_gitmodules(const struct object_id *oid)
1259 {
1260         oidset_insert(&gitmodules_found, oid);
1261 }
1262
1263 int fsck_finish(struct fsck_options *options)
1264 {
1265         int ret = 0;
1266         struct oidset_iter iter;
1267         const struct object_id *oid;
1268
1269         oidset_iter_init(&gitmodules_found, &iter);
1270         while ((oid = oidset_iter_next(&iter))) {
1271                 enum object_type type;
1272                 unsigned long size;
1273                 char *buf;
1274
1275                 if (oidset_contains(&gitmodules_done, oid))
1276                         continue;
1277
1278                 buf = read_object_file(oid, &type, &size);
1279                 if (!buf) {
1280                         if (is_promisor_object(oid))
1281                                 continue;
1282                         ret |= report(options,
1283                                       oid, OBJ_BLOB,
1284                                       FSCK_MSG_GITMODULES_MISSING,
1285                                       "unable to read .gitmodules blob");
1286                         continue;
1287                 }
1288
1289                 if (type == OBJ_BLOB)
1290                         ret |= fsck_blob(oid, buf, size, options);
1291                 else
1292                         ret |= report(options,
1293                                       oid, type,
1294                                       FSCK_MSG_GITMODULES_BLOB,
1295                                       "non-blob found at .gitmodules");
1296                 free(buf);
1297         }
1298
1299
1300         oidset_clear(&gitmodules_found);
1301         oidset_clear(&gitmodules_done);
1302         return ret;
1303 }
1304
1305 int git_fsck_config(const char *var, const char *value, void *cb)
1306 {
1307         struct fsck_options *options = cb;
1308         if (strcmp(var, "fsck.skiplist") == 0) {
1309                 const char *path;
1310                 struct strbuf sb = STRBUF_INIT;
1311
1312                 if (git_config_pathname(&path, var, value))
1313                         return 1;
1314                 strbuf_addf(&sb, "skiplist=%s", path);
1315                 free((char *)path);
1316                 fsck_set_msg_types(options, sb.buf);
1317                 strbuf_release(&sb);
1318                 return 0;
1319         }
1320
1321         if (skip_prefix(var, "fsck.", &var)) {
1322                 fsck_set_msg_type(options, var, value);
1323                 return 0;
1324         }
1325
1326         return git_default_config(var, value, cb);
1327 }