fsck.h: move FSCK_{FATAL,INFO,ERROR,WARN,IGNORE} into an enum
[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;
205
206         if (msg_id < 0)
207                 die("Unhandled message id: %s", msg_id_str);
208         msg_type = parse_msg_type(msg_type_str);
209
210         if (msg_type != FSCK_ERROR && msg_id_info[msg_id].msg_type == FSCK_FATAL)
211                 die("Cannot demote %s to %s", msg_id_str, msg_type_str);
212
213         if (!options->msg_type) {
214                 int i;
215                 enum fsck_msg_type *severity;
216                 ALLOC_ARRAY(severity, FSCK_MSG_MAX);
217                 for (i = 0; i < FSCK_MSG_MAX; i++)
218                         severity[i] = fsck_msg_type(i, options);
219                 options->msg_type = severity;
220         }
221
222         options->msg_type[msg_id] = msg_type;
223 }
224
225 void fsck_set_msg_types(struct fsck_options *options, const char *values)
226 {
227         char *buf = xstrdup(values), *to_free = buf;
228         int done = 0;
229
230         while (!done) {
231                 int len = strcspn(buf, " ,|"), equal;
232
233                 done = !buf[len];
234                 if (!len) {
235                         buf++;
236                         continue;
237                 }
238                 buf[len] = '\0';
239
240                 for (equal = 0;
241                      equal < len && buf[equal] != '=' && buf[equal] != ':';
242                      equal++)
243                         buf[equal] = tolower(buf[equal]);
244                 buf[equal] = '\0';
245
246                 if (!strcmp(buf, "skiplist")) {
247                         if (equal == len)
248                                 die("skiplist requires a path");
249                         oidset_parse_file(&options->skiplist, buf + equal + 1);
250                         buf += len + 1;
251                         continue;
252                 }
253
254                 if (equal == len)
255                         die("Missing '=': '%s'", buf);
256
257                 fsck_set_msg_type(options, buf, buf + equal + 1);
258                 buf += len + 1;
259         }
260         free(to_free);
261 }
262
263 static int object_on_skiplist(struct fsck_options *opts,
264                               const struct object_id *oid)
265 {
266         return opts && oid && oidset_contains(&opts->skiplist, oid);
267 }
268
269 __attribute__((format (printf, 5, 6)))
270 static int report(struct fsck_options *options,
271                   const struct object_id *oid, enum object_type object_type,
272                   enum fsck_msg_id msg_id, const char *fmt, ...)
273 {
274         va_list ap;
275         struct strbuf sb = STRBUF_INIT;
276         enum fsck_msg_type msg_type = fsck_msg_type(msg_id, options);
277         int result;
278
279         if (msg_type == FSCK_IGNORE)
280                 return 0;
281
282         if (object_on_skiplist(options, oid))
283                 return 0;
284
285         if (msg_type == FSCK_FATAL)
286                 msg_type = FSCK_ERROR;
287         else if (msg_type == FSCK_INFO)
288                 msg_type = FSCK_WARN;
289
290         prepare_msg_ids();
291         strbuf_addf(&sb, "%s: ", msg_id_info[msg_id].camelcased);
292
293         va_start(ap, fmt);
294         strbuf_vaddf(&sb, fmt, ap);
295         result = options->error_func(options, oid, object_type,
296                                      msg_type, sb.buf);
297         strbuf_release(&sb);
298         va_end(ap);
299
300         return result;
301 }
302
303 void fsck_enable_object_names(struct fsck_options *options)
304 {
305         if (!options->object_names)
306                 options->object_names = kh_init_oid_map();
307 }
308
309 const char *fsck_get_object_name(struct fsck_options *options,
310                                  const struct object_id *oid)
311 {
312         khiter_t pos;
313         if (!options->object_names)
314                 return NULL;
315         pos = kh_get_oid_map(options->object_names, *oid);
316         if (pos >= kh_end(options->object_names))
317                 return NULL;
318         return kh_value(options->object_names, pos);
319 }
320
321 void fsck_put_object_name(struct fsck_options *options,
322                           const struct object_id *oid,
323                           const char *fmt, ...)
324 {
325         va_list ap;
326         struct strbuf buf = STRBUF_INIT;
327         khiter_t pos;
328         int hashret;
329
330         if (!options->object_names)
331                 return;
332
333         pos = kh_put_oid_map(options->object_names, *oid, &hashret);
334         if (!hashret)
335                 return;
336         va_start(ap, fmt);
337         strbuf_vaddf(&buf, fmt, ap);
338         kh_value(options->object_names, pos) = strbuf_detach(&buf, NULL);
339         va_end(ap);
340 }
341
342 const char *fsck_describe_object(struct fsck_options *options,
343                                  const struct object_id *oid)
344 {
345         static struct strbuf bufs[] = {
346                 STRBUF_INIT, STRBUF_INIT, STRBUF_INIT, STRBUF_INIT
347         };
348         static int b = 0;
349         struct strbuf *buf;
350         const char *name = fsck_get_object_name(options, oid);
351
352         buf = bufs + b;
353         b = (b + 1) % ARRAY_SIZE(bufs);
354         strbuf_reset(buf);
355         strbuf_addstr(buf, oid_to_hex(oid));
356         if (name)
357                 strbuf_addf(buf, " (%s)", name);
358
359         return buf->buf;
360 }
361
362 static int fsck_walk_tree(struct tree *tree, void *data, struct fsck_options *options)
363 {
364         struct tree_desc desc;
365         struct name_entry entry;
366         int res = 0;
367         const char *name;
368
369         if (parse_tree(tree))
370                 return -1;
371
372         name = fsck_get_object_name(options, &tree->object.oid);
373         if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
374                 return -1;
375         while (tree_entry_gently(&desc, &entry)) {
376                 struct object *obj;
377                 int result;
378
379                 if (S_ISGITLINK(entry.mode))
380                         continue;
381
382                 if (S_ISDIR(entry.mode)) {
383                         obj = (struct object *)lookup_tree(the_repository, &entry.oid);
384                         if (name && obj)
385                                 fsck_put_object_name(options, &entry.oid, "%s%s/",
386                                                      name, entry.path);
387                         result = options->walk(obj, OBJ_TREE, data, options);
388                 }
389                 else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode)) {
390                         obj = (struct object *)lookup_blob(the_repository, &entry.oid);
391                         if (name && obj)
392                                 fsck_put_object_name(options, &entry.oid, "%s%s",
393                                                      name, entry.path);
394                         result = options->walk(obj, OBJ_BLOB, data, options);
395                 }
396                 else {
397                         result = error("in tree %s: entry %s has bad mode %.6o",
398                                        fsck_describe_object(options, &tree->object.oid),
399                                        entry.path, entry.mode);
400                 }
401                 if (result < 0)
402                         return result;
403                 if (!res)
404                         res = result;
405         }
406         return res;
407 }
408
409 static int fsck_walk_commit(struct commit *commit, void *data, struct fsck_options *options)
410 {
411         int counter = 0, generation = 0, name_prefix_len = 0;
412         struct commit_list *parents;
413         int res;
414         int result;
415         const char *name;
416
417         if (parse_commit(commit))
418                 return -1;
419
420         name = fsck_get_object_name(options, &commit->object.oid);
421         if (name)
422                 fsck_put_object_name(options, get_commit_tree_oid(commit),
423                                      "%s:", name);
424
425         result = options->walk((struct object *)get_commit_tree(commit),
426                                OBJ_TREE, data, options);
427         if (result < 0)
428                 return result;
429         res = result;
430
431         parents = commit->parents;
432         if (name && parents) {
433                 int len = strlen(name), power;
434
435                 if (len && name[len - 1] == '^') {
436                         generation = 1;
437                         name_prefix_len = len - 1;
438                 }
439                 else { /* parse ~<generation> suffix */
440                         for (generation = 0, power = 1;
441                              len && isdigit(name[len - 1]);
442                              power *= 10)
443                                 generation += power * (name[--len] - '0');
444                         if (power > 1 && len && name[len - 1] == '~')
445                                 name_prefix_len = len - 1;
446                         else {
447                                 /* Maybe a non-first parent, e.g. HEAD^2 */
448                                 generation = 0;
449                                 name_prefix_len = len;
450                         }
451                 }
452         }
453
454         while (parents) {
455                 if (name) {
456                         struct object_id *oid = &parents->item->object.oid;
457
458                         if (counter++)
459                                 fsck_put_object_name(options, oid, "%s^%d",
460                                                      name, counter);
461                         else if (generation > 0)
462                                 fsck_put_object_name(options, oid, "%.*s~%d",
463                                                      name_prefix_len, name,
464                                                      generation + 1);
465                         else
466                                 fsck_put_object_name(options, oid, "%s^", name);
467                 }
468                 result = options->walk((struct object *)parents->item, OBJ_COMMIT, data, options);
469                 if (result < 0)
470                         return result;
471                 if (!res)
472                         res = result;
473                 parents = parents->next;
474         }
475         return res;
476 }
477
478 static int fsck_walk_tag(struct tag *tag, void *data, struct fsck_options *options)
479 {
480         const char *name = fsck_get_object_name(options, &tag->object.oid);
481
482         if (parse_tag(tag))
483                 return -1;
484         if (name)
485                 fsck_put_object_name(options, &tag->tagged->oid, "%s", name);
486         return options->walk(tag->tagged, OBJ_ANY, data, options);
487 }
488
489 int fsck_walk(struct object *obj, void *data, struct fsck_options *options)
490 {
491         if (!obj)
492                 return -1;
493
494         if (obj->type == OBJ_NONE)
495                 parse_object(the_repository, &obj->oid);
496
497         switch (obj->type) {
498         case OBJ_BLOB:
499                 return 0;
500         case OBJ_TREE:
501                 return fsck_walk_tree((struct tree *)obj, data, options);
502         case OBJ_COMMIT:
503                 return fsck_walk_commit((struct commit *)obj, data, options);
504         case OBJ_TAG:
505                 return fsck_walk_tag((struct tag *)obj, data, options);
506         default:
507                 error("Unknown object type for %s",
508                       fsck_describe_object(options, &obj->oid));
509                 return -1;
510         }
511 }
512
513 struct name_stack {
514         const char **names;
515         size_t nr, alloc;
516 };
517
518 static void name_stack_push(struct name_stack *stack, const char *name)
519 {
520         ALLOC_GROW(stack->names, stack->nr + 1, stack->alloc);
521         stack->names[stack->nr++] = name;
522 }
523
524 static const char *name_stack_pop(struct name_stack *stack)
525 {
526         return stack->nr ? stack->names[--stack->nr] : NULL;
527 }
528
529 static void name_stack_clear(struct name_stack *stack)
530 {
531         FREE_AND_NULL(stack->names);
532         stack->nr = stack->alloc = 0;
533 }
534
535 /*
536  * The entries in a tree are ordered in the _path_ order,
537  * which means that a directory entry is ordered by adding
538  * a slash to the end of it.
539  *
540  * So a directory called "a" is ordered _after_ a file
541  * called "a.c", because "a/" sorts after "a.c".
542  */
543 #define TREE_UNORDERED (-1)
544 #define TREE_HAS_DUPS  (-2)
545
546 static int is_less_than_slash(unsigned char c)
547 {
548         return '\0' < c && c < '/';
549 }
550
551 static int verify_ordered(unsigned mode1, const char *name1,
552                           unsigned mode2, const char *name2,
553                           struct name_stack *candidates)
554 {
555         int len1 = strlen(name1);
556         int len2 = strlen(name2);
557         int len = len1 < len2 ? len1 : len2;
558         unsigned char c1, c2;
559         int cmp;
560
561         cmp = memcmp(name1, name2, len);
562         if (cmp < 0)
563                 return 0;
564         if (cmp > 0)
565                 return TREE_UNORDERED;
566
567         /*
568          * Ok, the first <len> characters are the same.
569          * Now we need to order the next one, but turn
570          * a '\0' into a '/' for a directory entry.
571          */
572         c1 = name1[len];
573         c2 = name2[len];
574         if (!c1 && !c2)
575                 /*
576                  * git-write-tree used to write out a nonsense tree that has
577                  * entries with the same name, one blob and one tree.  Make
578                  * sure we do not have duplicate entries.
579                  */
580                 return TREE_HAS_DUPS;
581         if (!c1 && S_ISDIR(mode1))
582                 c1 = '/';
583         if (!c2 && S_ISDIR(mode2))
584                 c2 = '/';
585
586         /*
587          * There can be non-consecutive duplicates due to the implicitly
588          * added slash, e.g.:
589          *
590          *   foo
591          *   foo.bar
592          *   foo.bar.baz
593          *   foo.bar/
594          *   foo/
595          *
596          * Record non-directory candidates (like "foo" and "foo.bar" in
597          * the example) on a stack and check directory candidates (like
598          * foo/" and "foo.bar/") against that stack.
599          */
600         if (!c1 && is_less_than_slash(c2)) {
601                 name_stack_push(candidates, name1);
602         } else if (c2 == '/' && is_less_than_slash(c1)) {
603                 for (;;) {
604                         const char *p;
605                         const char *f_name = name_stack_pop(candidates);
606
607                         if (!f_name)
608                                 break;
609                         if (!skip_prefix(name2, f_name, &p))
610                                 continue;
611                         if (!*p)
612                                 return TREE_HAS_DUPS;
613                         if (is_less_than_slash(*p)) {
614                                 name_stack_push(candidates, f_name);
615                                 break;
616                         }
617                 }
618         }
619
620         return c1 < c2 ? 0 : TREE_UNORDERED;
621 }
622
623 static int fsck_tree(const struct object_id *oid,
624                      const char *buffer, unsigned long size,
625                      struct fsck_options *options)
626 {
627         int retval = 0;
628         int has_null_sha1 = 0;
629         int has_full_path = 0;
630         int has_empty_name = 0;
631         int has_dot = 0;
632         int has_dotdot = 0;
633         int has_dotgit = 0;
634         int has_zero_pad = 0;
635         int has_bad_modes = 0;
636         int has_dup_entries = 0;
637         int not_properly_sorted = 0;
638         struct tree_desc desc;
639         unsigned o_mode;
640         const char *o_name;
641         struct name_stack df_dup_candidates = { NULL };
642
643         if (init_tree_desc_gently(&desc, buffer, size)) {
644                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
645                 return retval;
646         }
647
648         o_mode = 0;
649         o_name = NULL;
650
651         while (desc.size) {
652                 unsigned short mode;
653                 const char *name, *backslash;
654                 const struct object_id *oid;
655
656                 oid = tree_entry_extract(&desc, &name, &mode);
657
658                 has_null_sha1 |= is_null_oid(oid);
659                 has_full_path |= !!strchr(name, '/');
660                 has_empty_name |= !*name;
661                 has_dot |= !strcmp(name, ".");
662                 has_dotdot |= !strcmp(name, "..");
663                 has_dotgit |= is_hfs_dotgit(name) || is_ntfs_dotgit(name);
664                 has_zero_pad |= *(char *)desc.buffer == '0';
665
666                 if (is_hfs_dotgitmodules(name) || is_ntfs_dotgitmodules(name)) {
667                         if (!S_ISLNK(mode))
668                                 oidset_insert(&gitmodules_found, oid);
669                         else
670                                 retval += report(options,
671                                                  oid, OBJ_TREE,
672                                                  FSCK_MSG_GITMODULES_SYMLINK,
673                                                  ".gitmodules is a symbolic link");
674                 }
675
676                 if ((backslash = strchr(name, '\\'))) {
677                         while (backslash) {
678                                 backslash++;
679                                 has_dotgit |= is_ntfs_dotgit(backslash);
680                                 if (is_ntfs_dotgitmodules(backslash)) {
681                                         if (!S_ISLNK(mode))
682                                                 oidset_insert(&gitmodules_found, oid);
683                                         else
684                                                 retval += report(options, oid, OBJ_TREE,
685                                                                  FSCK_MSG_GITMODULES_SYMLINK,
686                                                                  ".gitmodules is a symbolic link");
687                                 }
688                                 backslash = strchr(backslash, '\\');
689                         }
690                 }
691
692                 if (update_tree_entry_gently(&desc)) {
693                         retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_TREE, "cannot be parsed as a tree");
694                         break;
695                 }
696
697                 switch (mode) {
698                 /*
699                  * Standard modes..
700                  */
701                 case S_IFREG | 0755:
702                 case S_IFREG | 0644:
703                 case S_IFLNK:
704                 case S_IFDIR:
705                 case S_IFGITLINK:
706                         break;
707                 /*
708                  * This is nonstandard, but we had a few of these
709                  * early on when we honored the full set of mode
710                  * bits..
711                  */
712                 case S_IFREG | 0664:
713                         if (!options->strict)
714                                 break;
715                         /* fallthrough */
716                 default:
717                         has_bad_modes = 1;
718                 }
719
720                 if (o_name) {
721                         switch (verify_ordered(o_mode, o_name, mode, name,
722                                                &df_dup_candidates)) {
723                         case TREE_UNORDERED:
724                                 not_properly_sorted = 1;
725                                 break;
726                         case TREE_HAS_DUPS:
727                                 has_dup_entries = 1;
728                                 break;
729                         default:
730                                 break;
731                         }
732                 }
733
734                 o_mode = mode;
735                 o_name = name;
736         }
737
738         name_stack_clear(&df_dup_candidates);
739
740         if (has_null_sha1)
741                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_NULL_SHA1, "contains entries pointing to null sha1");
742         if (has_full_path)
743                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_FULL_PATHNAME, "contains full pathnames");
744         if (has_empty_name)
745                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_EMPTY_NAME, "contains empty pathname");
746         if (has_dot)
747                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOT, "contains '.'");
748         if (has_dotdot)
749                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOTDOT, "contains '..'");
750         if (has_dotgit)
751                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_HAS_DOTGIT, "contains '.git'");
752         if (has_zero_pad)
753                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_ZERO_PADDED_FILEMODE, "contains zero-padded file modes");
754         if (has_bad_modes)
755                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_BAD_FILEMODE, "contains bad file modes");
756         if (has_dup_entries)
757                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_DUPLICATE_ENTRIES, "contains duplicate file entries");
758         if (not_properly_sorted)
759                 retval += report(options, oid, OBJ_TREE, FSCK_MSG_TREE_NOT_SORTED, "not properly sorted");
760         return retval;
761 }
762
763 static int verify_headers(const void *data, unsigned long size,
764                           const struct object_id *oid, enum object_type type,
765                           struct fsck_options *options)
766 {
767         const char *buffer = (const char *)data;
768         unsigned long i;
769
770         for (i = 0; i < size; i++) {
771                 switch (buffer[i]) {
772                 case '\0':
773                         return report(options, oid, type,
774                                 FSCK_MSG_NUL_IN_HEADER,
775                                 "unterminated header: NUL at offset %ld", i);
776                 case '\n':
777                         if (i + 1 < size && buffer[i + 1] == '\n')
778                                 return 0;
779                 }
780         }
781
782         /*
783          * We did not find double-LF that separates the header
784          * and the body.  Not having a body is not a crime but
785          * we do want to see the terminating LF for the last header
786          * line.
787          */
788         if (size && buffer[size - 1] == '\n')
789                 return 0;
790
791         return report(options, oid, type,
792                 FSCK_MSG_UNTERMINATED_HEADER, "unterminated header");
793 }
794
795 static int fsck_ident(const char **ident,
796                       const struct object_id *oid, enum object_type type,
797                       struct fsck_options *options)
798 {
799         const char *p = *ident;
800         char *end;
801
802         *ident = strchrnul(*ident, '\n');
803         if (**ident == '\n')
804                 (*ident)++;
805
806         if (*p == '<')
807                 return report(options, oid, type, FSCK_MSG_MISSING_NAME_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
808         p += strcspn(p, "<>\n");
809         if (*p == '>')
810                 return report(options, oid, type, FSCK_MSG_BAD_NAME, "invalid author/committer line - bad name");
811         if (*p != '<')
812                 return report(options, oid, type, FSCK_MSG_MISSING_EMAIL, "invalid author/committer line - missing email");
813         if (p[-1] != ' ')
814                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_EMAIL, "invalid author/committer line - missing space before email");
815         p++;
816         p += strcspn(p, "<>\n");
817         if (*p != '>')
818                 return report(options, oid, type, FSCK_MSG_BAD_EMAIL, "invalid author/committer line - bad email");
819         p++;
820         if (*p != ' ')
821                 return report(options, oid, type, FSCK_MSG_MISSING_SPACE_BEFORE_DATE, "invalid author/committer line - missing space before date");
822         p++;
823         if (*p == '0' && p[1] != ' ')
824                 return report(options, oid, type, FSCK_MSG_ZERO_PADDED_DATE, "invalid author/committer line - zero-padded date");
825         if (date_overflows(parse_timestamp(p, &end, 10)))
826                 return report(options, oid, type, FSCK_MSG_BAD_DATE_OVERFLOW, "invalid author/committer line - date causes integer overflow");
827         if ((end == p || *end != ' '))
828                 return report(options, oid, type, FSCK_MSG_BAD_DATE, "invalid author/committer line - bad date");
829         p = end + 1;
830         if ((*p != '+' && *p != '-') ||
831             !isdigit(p[1]) ||
832             !isdigit(p[2]) ||
833             !isdigit(p[3]) ||
834             !isdigit(p[4]) ||
835             (p[5] != '\n'))
836                 return report(options, oid, type, FSCK_MSG_BAD_TIMEZONE, "invalid author/committer line - bad time zone");
837         p += 6;
838         return 0;
839 }
840
841 static int fsck_commit(const struct object_id *oid,
842                        const char *buffer, unsigned long size,
843                        struct fsck_options *options)
844 {
845         struct object_id tree_oid, parent_oid;
846         unsigned author_count;
847         int err;
848         const char *buffer_begin = buffer;
849         const char *p;
850
851         if (verify_headers(buffer, size, oid, OBJ_COMMIT, options))
852                 return -1;
853
854         if (!skip_prefix(buffer, "tree ", &buffer))
855                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_TREE, "invalid format - expected 'tree' line");
856         if (parse_oid_hex(buffer, &tree_oid, &p) || *p != '\n') {
857                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_TREE_SHA1, "invalid 'tree' line format - bad sha1");
858                 if (err)
859                         return err;
860         }
861         buffer = p + 1;
862         while (skip_prefix(buffer, "parent ", &buffer)) {
863                 if (parse_oid_hex(buffer, &parent_oid, &p) || *p != '\n') {
864                         err = report(options, oid, OBJ_COMMIT, FSCK_MSG_BAD_PARENT_SHA1, "invalid 'parent' line format - bad sha1");
865                         if (err)
866                                 return err;
867                 }
868                 buffer = p + 1;
869         }
870         author_count = 0;
871         while (skip_prefix(buffer, "author ", &buffer)) {
872                 author_count++;
873                 err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
874                 if (err)
875                         return err;
876         }
877         if (author_count < 1)
878                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_AUTHOR, "invalid format - expected 'author' line");
879         else if (author_count > 1)
880                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_MULTIPLE_AUTHORS, "invalid format - multiple 'author' lines");
881         if (err)
882                 return err;
883         if (!skip_prefix(buffer, "committer ", &buffer))
884                 return report(options, oid, OBJ_COMMIT, FSCK_MSG_MISSING_COMMITTER, "invalid format - expected 'committer' line");
885         err = fsck_ident(&buffer, oid, OBJ_COMMIT, options);
886         if (err)
887                 return err;
888         if (memchr(buffer_begin, '\0', size)) {
889                 err = report(options, oid, OBJ_COMMIT, FSCK_MSG_NUL_IN_COMMIT,
890                              "NUL byte in the commit object body");
891                 if (err)
892                         return err;
893         }
894         return 0;
895 }
896
897 static int fsck_tag(const struct object_id *oid, const char *buffer,
898                     unsigned long size, struct fsck_options *options)
899 {
900         struct object_id tagged_oid;
901         int tagged_type;
902         return fsck_tag_standalone(oid, buffer, size, options, &tagged_oid,
903                                    &tagged_type);
904 }
905
906 int fsck_tag_standalone(const struct object_id *oid, const char *buffer,
907                         unsigned long size, struct fsck_options *options,
908                         struct object_id *tagged_oid,
909                         int *tagged_type)
910 {
911         int ret = 0;
912         char *eol;
913         struct strbuf sb = STRBUF_INIT;
914         const char *p;
915
916         ret = verify_headers(buffer, size, oid, OBJ_TAG, options);
917         if (ret)
918                 goto done;
919
920         if (!skip_prefix(buffer, "object ", &buffer)) {
921                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_OBJECT, "invalid format - expected 'object' line");
922                 goto done;
923         }
924         if (parse_oid_hex(buffer, tagged_oid, &p) || *p != '\n') {
925                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_OBJECT_SHA1, "invalid 'object' line format - bad sha1");
926                 if (ret)
927                         goto done;
928         }
929         buffer = p + 1;
930
931         if (!skip_prefix(buffer, "type ", &buffer)) {
932                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE_ENTRY, "invalid format - expected 'type' line");
933                 goto done;
934         }
935         eol = strchr(buffer, '\n');
936         if (!eol) {
937                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TYPE, "invalid format - unexpected end after 'type' line");
938                 goto done;
939         }
940         *tagged_type = type_from_string_gently(buffer, eol - buffer, 1);
941         if (*tagged_type < 0)
942                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_BAD_TYPE, "invalid 'type' value");
943         if (ret)
944                 goto done;
945         buffer = eol + 1;
946
947         if (!skip_prefix(buffer, "tag ", &buffer)) {
948                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG_ENTRY, "invalid format - expected 'tag' line");
949                 goto done;
950         }
951         eol = strchr(buffer, '\n');
952         if (!eol) {
953                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAG, "invalid format - unexpected end after 'type' line");
954                 goto done;
955         }
956         strbuf_addf(&sb, "refs/tags/%.*s", (int)(eol - buffer), buffer);
957         if (check_refname_format(sb.buf, 0)) {
958                 ret = report(options, oid, OBJ_TAG,
959                              FSCK_MSG_BAD_TAG_NAME,
960                              "invalid 'tag' name: %.*s",
961                              (int)(eol - buffer), buffer);
962                 if (ret)
963                         goto done;
964         }
965         buffer = eol + 1;
966
967         if (!skip_prefix(buffer, "tagger ", &buffer)) {
968                 /* early tags do not contain 'tagger' lines; warn only */
969                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_MISSING_TAGGER_ENTRY, "invalid format - expected 'tagger' line");
970                 if (ret)
971                         goto done;
972         }
973         else
974                 ret = fsck_ident(&buffer, oid, OBJ_TAG, options);
975         if (!*buffer)
976                 goto done;
977
978         if (!starts_with(buffer, "\n")) {
979                 /*
980                  * The verify_headers() check will allow
981                  * e.g. "[...]tagger <tagger>\nsome
982                  * garbage\n\nmessage" to pass, thinking "some
983                  * garbage" could be a custom header. E.g. "mktag"
984                  * doesn't want any unknown headers.
985                  */
986                 ret = report(options, oid, OBJ_TAG, FSCK_MSG_EXTRA_HEADER_ENTRY, "invalid format - extra header(s) after 'tagger'");
987                 if (ret)
988                         goto done;
989         }
990
991 done:
992         strbuf_release(&sb);
993         return ret;
994 }
995
996 /*
997  * Like builtin/submodule--helper.c's starts_with_dot_slash, but without
998  * relying on the platform-dependent is_dir_sep helper.
999  *
1000  * This is for use in checking whether a submodule URL is interpreted as
1001  * relative to the current directory on any platform, since \ is a
1002  * directory separator on Windows but not on other platforms.
1003  */
1004 static int starts_with_dot_slash(const char *str)
1005 {
1006         return str[0] == '.' && (str[1] == '/' || str[1] == '\\');
1007 }
1008
1009 /*
1010  * Like starts_with_dot_slash, this is a variant of submodule--helper's
1011  * helper of the same name with the twist that it accepts backslash as a
1012  * directory separator even on non-Windows platforms.
1013  */
1014 static int starts_with_dot_dot_slash(const char *str)
1015 {
1016         return str[0] == '.' && starts_with_dot_slash(str + 1);
1017 }
1018
1019 static int submodule_url_is_relative(const char *url)
1020 {
1021         return starts_with_dot_slash(url) || starts_with_dot_dot_slash(url);
1022 }
1023
1024 /*
1025  * Count directory components that a relative submodule URL should chop
1026  * from the remote_url it is to be resolved against.
1027  *
1028  * In other words, this counts "../" components at the start of a
1029  * submodule URL.
1030  *
1031  * Returns the number of directory components to chop and writes a
1032  * pointer to the next character of url after all leading "./" and
1033  * "../" components to out.
1034  */
1035 static int count_leading_dotdots(const char *url, const char **out)
1036 {
1037         int result = 0;
1038         while (1) {
1039                 if (starts_with_dot_dot_slash(url)) {
1040                         result++;
1041                         url += strlen("../");
1042                         continue;
1043                 }
1044                 if (starts_with_dot_slash(url)) {
1045                         url += strlen("./");
1046                         continue;
1047                 }
1048                 *out = url;
1049                 return result;
1050         }
1051 }
1052 /*
1053  * Check whether a transport is implemented by git-remote-curl.
1054  *
1055  * If it is, returns 1 and writes the URL that would be passed to
1056  * git-remote-curl to the "out" parameter.
1057  *
1058  * Otherwise, returns 0 and leaves "out" untouched.
1059  *
1060  * Examples:
1061  *   http::https://example.com/repo.git -> 1, https://example.com/repo.git
1062  *   https://example.com/repo.git -> 1, https://example.com/repo.git
1063  *   git://example.com/repo.git -> 0
1064  *
1065  * This is for use in checking for previously exploitable bugs that
1066  * required a submodule URL to be passed to git-remote-curl.
1067  */
1068 static int url_to_curl_url(const char *url, const char **out)
1069 {
1070         /*
1071          * We don't need to check for case-aliases, "http.exe", and so
1072          * on because in the default configuration, is_transport_allowed
1073          * prevents URLs with those schemes from being cloned
1074          * automatically.
1075          */
1076         if (skip_prefix(url, "http::", out) ||
1077             skip_prefix(url, "https::", out) ||
1078             skip_prefix(url, "ftp::", out) ||
1079             skip_prefix(url, "ftps::", out))
1080                 return 1;
1081         if (starts_with(url, "http://") ||
1082             starts_with(url, "https://") ||
1083             starts_with(url, "ftp://") ||
1084             starts_with(url, "ftps://")) {
1085                 *out = url;
1086                 return 1;
1087         }
1088         return 0;
1089 }
1090
1091 static int check_submodule_url(const char *url)
1092 {
1093         const char *curl_url;
1094
1095         if (looks_like_command_line_option(url))
1096                 return -1;
1097
1098         if (submodule_url_is_relative(url) || starts_with(url, "git://")) {
1099                 char *decoded;
1100                 const char *next;
1101                 int has_nl;
1102
1103                 /*
1104                  * This could be appended to an http URL and url-decoded;
1105                  * check for malicious characters.
1106                  */
1107                 decoded = url_decode(url);
1108                 has_nl = !!strchr(decoded, '\n');
1109
1110                 free(decoded);
1111                 if (has_nl)
1112                         return -1;
1113
1114                 /*
1115                  * URLs which escape their root via "../" can overwrite
1116                  * the host field and previous components, resolving to
1117                  * URLs like https::example.com/submodule.git and
1118                  * https:///example.com/submodule.git that were
1119                  * susceptible to CVE-2020-11008.
1120                  */
1121                 if (count_leading_dotdots(url, &next) > 0 &&
1122                     (*next == ':' || *next == '/'))
1123                         return -1;
1124         }
1125
1126         else if (url_to_curl_url(url, &curl_url)) {
1127                 struct credential c = CREDENTIAL_INIT;
1128                 int ret = 0;
1129                 if (credential_from_url_gently(&c, curl_url, 1) ||
1130                     !*c.host)
1131                         ret = -1;
1132                 credential_clear(&c);
1133                 return ret;
1134         }
1135
1136         return 0;
1137 }
1138
1139 struct fsck_gitmodules_data {
1140         const struct object_id *oid;
1141         struct fsck_options *options;
1142         int ret;
1143 };
1144
1145 static int fsck_gitmodules_fn(const char *var, const char *value, void *vdata)
1146 {
1147         struct fsck_gitmodules_data *data = vdata;
1148         const char *subsection, *key;
1149         size_t subsection_len;
1150         char *name;
1151
1152         if (parse_config_key(var, "submodule", &subsection, &subsection_len, &key) < 0 ||
1153             !subsection)
1154                 return 0;
1155
1156         name = xmemdupz(subsection, subsection_len);
1157         if (check_submodule_name(name) < 0)
1158                 data->ret |= report(data->options,
1159                                     data->oid, OBJ_BLOB,
1160                                     FSCK_MSG_GITMODULES_NAME,
1161                                     "disallowed submodule name: %s",
1162                                     name);
1163         if (!strcmp(key, "url") && value &&
1164             check_submodule_url(value) < 0)
1165                 data->ret |= report(data->options,
1166                                     data->oid, OBJ_BLOB,
1167                                     FSCK_MSG_GITMODULES_URL,
1168                                     "disallowed submodule url: %s",
1169                                     value);
1170         if (!strcmp(key, "path") && value &&
1171             looks_like_command_line_option(value))
1172                 data->ret |= report(data->options,
1173                                     data->oid, OBJ_BLOB,
1174                                     FSCK_MSG_GITMODULES_PATH,
1175                                     "disallowed submodule path: %s",
1176                                     value);
1177         if (!strcmp(key, "update") && value &&
1178             parse_submodule_update_type(value) == SM_UPDATE_COMMAND)
1179                 data->ret |= report(data->options, data->oid, OBJ_BLOB,
1180                                     FSCK_MSG_GITMODULES_UPDATE,
1181                                     "disallowed submodule update setting: %s",
1182                                     value);
1183         free(name);
1184
1185         return 0;
1186 }
1187
1188 static int fsck_blob(const struct object_id *oid, const char *buf,
1189                      unsigned long size, struct fsck_options *options)
1190 {
1191         struct fsck_gitmodules_data data;
1192         struct config_options config_opts = { 0 };
1193
1194         if (!oidset_contains(&gitmodules_found, oid))
1195                 return 0;
1196         oidset_insert(&gitmodules_done, oid);
1197
1198         if (object_on_skiplist(options, oid))
1199                 return 0;
1200
1201         if (!buf) {
1202                 /*
1203                  * A missing buffer here is a sign that the caller found the
1204                  * blob too gigantic to load into memory. Let's just consider
1205                  * that an error.
1206                  */
1207                 return report(options, oid, OBJ_BLOB,
1208                               FSCK_MSG_GITMODULES_LARGE,
1209                               ".gitmodules too large to parse");
1210         }
1211
1212         data.oid = oid;
1213         data.options = options;
1214         data.ret = 0;
1215         config_opts.error_action = CONFIG_ERROR_SILENT;
1216         if (git_config_from_mem(fsck_gitmodules_fn, CONFIG_ORIGIN_BLOB,
1217                                 ".gitmodules", buf, size, &data, &config_opts))
1218                 data.ret |= report(options, oid, OBJ_BLOB,
1219                                    FSCK_MSG_GITMODULES_PARSE,
1220                                    "could not parse gitmodules blob");
1221
1222         return data.ret;
1223 }
1224
1225 int fsck_object(struct object *obj, void *data, unsigned long size,
1226         struct fsck_options *options)
1227 {
1228         if (!obj)
1229                 return report(options, NULL, OBJ_NONE, FSCK_MSG_BAD_OBJECT_SHA1, "no valid object to fsck");
1230
1231         if (obj->type == OBJ_BLOB)
1232                 return fsck_blob(&obj->oid, data, size, options);
1233         if (obj->type == OBJ_TREE)
1234                 return fsck_tree(&obj->oid, data, size, options);
1235         if (obj->type == OBJ_COMMIT)
1236                 return fsck_commit(&obj->oid, data, size, options);
1237         if (obj->type == OBJ_TAG)
1238                 return fsck_tag(&obj->oid, data, size, options);
1239
1240         return report(options, &obj->oid, obj->type,
1241                       FSCK_MSG_UNKNOWN_TYPE,
1242                       "unknown type '%d' (internal fsck error)",
1243                       obj->type);
1244 }
1245
1246 int fsck_error_function(struct fsck_options *o,
1247                         const struct object_id *oid,
1248                         enum object_type object_type,
1249                         enum fsck_msg_type msg_type, const char *message)
1250 {
1251         if (msg_type == FSCK_WARN) {
1252                 warning("object %s: %s", fsck_describe_object(o, oid), message);
1253                 return 0;
1254         }
1255         error("object %s: %s", fsck_describe_object(o, oid), message);
1256         return 1;
1257 }
1258
1259 void register_found_gitmodules(const struct object_id *oid)
1260 {
1261         oidset_insert(&gitmodules_found, oid);
1262 }
1263
1264 int fsck_finish(struct fsck_options *options)
1265 {
1266         int ret = 0;
1267         struct oidset_iter iter;
1268         const struct object_id *oid;
1269
1270         oidset_iter_init(&gitmodules_found, &iter);
1271         while ((oid = oidset_iter_next(&iter))) {
1272                 enum object_type type;
1273                 unsigned long size;
1274                 char *buf;
1275
1276                 if (oidset_contains(&gitmodules_done, oid))
1277                         continue;
1278
1279                 buf = read_object_file(oid, &type, &size);
1280                 if (!buf) {
1281                         if (is_promisor_object(oid))
1282                                 continue;
1283                         ret |= report(options,
1284                                       oid, OBJ_BLOB,
1285                                       FSCK_MSG_GITMODULES_MISSING,
1286                                       "unable to read .gitmodules blob");
1287                         continue;
1288                 }
1289
1290                 if (type == OBJ_BLOB)
1291                         ret |= fsck_blob(oid, buf, size, options);
1292                 else
1293                         ret |= report(options,
1294                                       oid, type,
1295                                       FSCK_MSG_GITMODULES_BLOB,
1296                                       "non-blob found at .gitmodules");
1297                 free(buf);
1298         }
1299
1300
1301         oidset_clear(&gitmodules_found);
1302         oidset_clear(&gitmodules_done);
1303         return ret;
1304 }
1305
1306 int git_fsck_config(const char *var, const char *value, void *cb)
1307 {
1308         struct fsck_options *options = cb;
1309         if (strcmp(var, "fsck.skiplist") == 0) {
1310                 const char *path;
1311                 struct strbuf sb = STRBUF_INIT;
1312
1313                 if (git_config_pathname(&path, var, value))
1314                         return 1;
1315                 strbuf_addf(&sb, "skiplist=%s", path);
1316                 free((char *)path);
1317                 fsck_set_msg_types(options, sb.buf);
1318                 strbuf_release(&sb);
1319                 return 0;
1320         }
1321
1322         if (skip_prefix(var, "fsck.", &var)) {
1323                 fsck_set_msg_type(options, var, value);
1324                 return 0;
1325         }
1326
1327         return git_default_config(var, value, cb);
1328 }