15 * along with this program; if not, see http://www.gnu.org/copyleft/gpl.txt
16 */
17
18 /*
19 * There are a couple checks that try to see if a variable
20 * comes from the user. It would be better to unify them
21 * into one place. Also it we should follow the data down
22 * the call paths. Hence this file.
23 */
24
25 #include "smatch.h"
26 #include "smatch_slist.h"
27 #include "smatch_extra.h"
28
29 static int my_id;
30 static int my_call_id;
31
32 STATE(called);
33 static bool func_gets_user_data;
34
35 static const char * kstr_funcs[] = {
36 "kstrtoull", "kstrtoll", "kstrtoul", "kstrtol", "kstrtouint",
37 "kstrtoint", "kstrtou64", "kstrtos64", "kstrtou32", "kstrtos32",
38 "kstrtou16", "kstrtos16", "kstrtou8", "kstrtos8", "kstrtoull_from_user"
39 "kstrtoll_from_user", "kstrtoul_from_user", "kstrtol_from_user",
40 "kstrtouint_from_user", "kstrtoint_from_user", "kstrtou16_from_user",
41 "kstrtos16_from_user", "kstrtou8_from_user", "kstrtos8_from_user",
42 "kstrtou64_from_user", "kstrtos64_from_user", "kstrtou32_from_user",
43 "kstrtos32_from_user",
44 };
45
46 static const char *returns_user_data[] = {
47 "simple_strtol", "simple_strtoll", "simple_strtoul", "simple_strtoull",
48 "kvm_register_read", "nlmsg_data", "nla_data", "memdup_user",
49 "kmap_atomic", "skb_network_header",
50 };
51
52 static void set_points_to_user_data(struct expression *expr);
53
54 static struct stree *start_states;
55 static struct stree_stack *saved_stack;
56 static void save_start_states(struct statement *stmt)
57 {
58 start_states = clone_stree(__get_cur_stree());
59 }
60
61 static void free_start_states(void)
62 {
63 free_stree(&start_states);
64 }
65
66 static void match_save_states(struct expression *expr)
67 {
68 push_stree(&saved_stack, start_states);
69 start_states = NULL;
70 }
71
72 static void match_restore_states(struct expression *expr)
73 {
74 free_stree(&start_states);
75 start_states = pop_stree(&saved_stack);
76 }
77
78 static struct smatch_state *empty_state(struct sm_state *sm)
79 {
80 return alloc_estate_empty();
81 }
82
83 static void pre_merge_hook(struct sm_state *sm)
84 {
85 struct smatch_state *user;
86 struct smatch_state *extra;
87 struct range_list *rl;
88 sval_t dummy;
89 sval_t sval_100;
90
91 sval_100.value = 100;
92 sval_100.type = &int_ctype;
93
94 user = get_state(my_id, sm->name, sm->sym);
95 if (!user)
96 return;
97 if (!__in_function_def && !estate_rl(sm->state)) {
98 /*
99 * If the one side is capped and the other side is empty then
100 * let's just mark it as not-user data because the information
101 * isn't going to be useful. How this looks is:
102 *
103 * if (user_var > trusted)
104 * user_var = trusted; <-- empty state
105 * else
106 * <-- capped
107 *
108 * The problem is that sometimes things are capped to a literal
109 * and we'd like to keep the state in that case... Ugh. I've
110 * added a check which assumes that everything less than 100 is
111 * probably capped against a literal.
112 *
113 */
114 if (is_capped_var_sym(sm->name, sm->sym) &&
115 sval_cmp(estate_max(user), sval_100) > 0)
116 set_state(my_id, sm->name, sm->sym, alloc_estate_empty());
117 return;
118 }
119 extra = get_state(SMATCH_EXTRA, sm->name, sm->sym);
120 if (!extra || !estate_rl(extra))
121 return;
122 rl = rl_intersection(estate_rl(user), estate_rl(extra));
123 if (rl_to_sval(rl, &dummy))
124 rl = NULL;
125 set_state(my_id, sm->name, sm->sym, alloc_estate_rl(clone_rl(rl)));
126 }
127
128 static void extra_nomod_hook(const char *name, struct symbol *sym, struct expression *expr, struct smatch_state *state)
129 {
130 struct smatch_state *user;
131 struct range_list *rl;
132
133 user = get_state(my_id, name, sym);
134 if (!user)
135 return;
136 rl = rl_intersection(estate_rl(user), estate_rl(state));
137 if (rl_equiv(rl, estate_rl(user)))
138 return;
139 set_state(my_id, name, sym, alloc_estate_rl(rl));
140 }
141
142 static void tag_inner_struct_members(struct expression *expr, struct symbol *member)
143 {
144 struct expression *edge_member;
145 struct symbol *base = get_real_base_type(member);
146 struct symbol *tmp;
147
148 if (member->ident)
149 expr = member_expression(expr, '.', member->ident);
150
151 FOR_EACH_PTR(base->symbol_list, tmp) {
152 struct symbol *type;
153
154 type = get_real_base_type(tmp);
155 if (!type)
156 continue;
157
158 if (type->type == SYM_UNION || type->type == SYM_STRUCT) {
159 tag_inner_struct_members(expr, tmp);
160 continue;
161 }
330 if (!expr->member)
331 return 0;
332 if (strcmp(expr->member->name, "data") != 0)
333 return 0;
334
335 sym = expr_to_sym(expr->deref);
336 if (!sym)
337 return 0;
338 sym = get_real_base_type(sym);
339 if (!sym || sym->type != SYM_PTR)
340 return 0;
341 sym = get_real_base_type(sym);
342 if (!sym || sym->type != SYM_STRUCT || !sym->ident)
343 return 0;
344 if (strcmp(sym->ident->name, "sk_buff") != 0)
345 return 0;
346
347 return 1;
348 }
349
350 static int get_rl_from_function(struct expression *expr, struct range_list **rl)
351 {
352 int i;
353
354 if (expr->type != EXPR_CALL || expr->fn->type != EXPR_SYMBOL ||
355 !expr->fn->symbol_name || !expr->fn->symbol_name->name)
356 return 0;
357
358 for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
359 if (strcmp(expr->fn->symbol_name->name, returns_user_data[i]) == 0) {
360 *rl = alloc_whole_rl(get_type(expr));
361 return 1;
362 }
363 }
364 return 0;
365 }
366
367 int points_to_user_data(struct expression *expr)
368 {
369 struct smatch_state *state;
370 struct range_list *rl;
371 char buf[256];
372 struct symbol *sym;
373 char *name;
374 int ret = 0;
375
376 expr = strip_expr(expr);
377 if (!expr)
378 return 0;
379 if (is_skb_data(expr))
380 return 1;
381 if (get_rl_from_function(expr, &rl))
382 return 1;
383
384 if (expr->type == EXPR_BINOP && expr->op == '+') {
385 if (points_to_user_data(expr->left))
386 return 1;
387 if (points_to_user_data(expr->right))
388 return 1;
389 return 0;
390 }
391
392 name = expr_to_var_sym(expr, &sym);
393 if (!name || !sym)
394 goto free;
395 snprintf(buf, sizeof(buf), "*%s", name);
396 state = get_state(my_id, buf, sym);
397 if (state && estate_rl(state))
398 ret = 1;
399 free:
400 free_string(name);
401 return ret;
402 }
403
404 static void set_points_to_user_data(struct expression *expr)
405 {
406 char *name;
407 struct symbol *sym;
408 char buf[256];
409
410 name = expr_to_var_sym(expr, &sym);
411 if (!name || !sym)
412 goto free;
413 snprintf(buf, sizeof(buf), "*%s", name);
414 set_state(my_id, buf, sym, alloc_estate_whole(&llong_ctype));
415 free:
416 free_string(name);
417 }
418
419 static int comes_from_skb_data(struct expression *expr)
420 {
421 expr = strip_expr(expr);
422 if (!expr || expr->type != EXPR_PREOP || expr->op != '*')
423 return 0;
424
425 expr = strip_expr(expr->unop);
426 if (!expr)
427 return 0;
428 if (expr->type == EXPR_BINOP && expr->op == '+')
429 expr = strip_expr(expr->left);
430
431 return is_skb_data(expr);
432 }
433
434 static int handle_struct_assignment(struct expression *expr)
472
473 static int handle_get_user(struct expression *expr)
474 {
475 char *name;
476 int ret = 0;
477
478 name = get_macro_name(expr->pos);
479 if (!name || strcmp(name, "get_user") != 0)
480 return 0;
481
482 name = expr_to_var(expr->right);
483 if (!name || strcmp(name, "__val_gu") != 0)
484 goto free;
485 set_state_expr(my_id, expr->left, alloc_estate_whole(get_type(expr->left)));
486 ret = 1;
487 free:
488 free_string(name);
489 return ret;
490 }
491
492 static void match_assign(struct expression *expr)
493 {
494 struct range_list *rl;
495
496 if (is_fake_call(expr->right))
497 goto clear_old_state;
498 if (handle_get_user(expr))
499 return;
500 if (points_to_user_data(expr->right))
501 set_points_to_user_data(expr->left);
502 if (handle_struct_assignment(expr))
503 return;
504
505 if (!get_user_rl(expr->right, &rl))
506 goto clear_old_state;
507
508 rl = cast_rl(get_type(expr->left), rl);
509 set_state_expr(my_id, expr->left, alloc_estate_rl(rl));
510
511 return;
512
513 clear_old_state:
514 if (get_state_expr(my_id, expr->left))
515 set_state_expr(my_id, expr->left, alloc_estate_empty());
516 }
517
518 static void handle_eq_noteq(struct expression *expr)
519 {
520 struct smatch_state *left_orig, *right_orig;
521
522 left_orig = get_state_expr(my_id, expr->left);
523 right_orig = get_state_expr(my_id, expr->right);
524
525 if (!left_orig && !right_orig)
526 return;
527 if (left_orig && right_orig)
528 return;
529
530 if (left_orig) {
531 set_true_false_states_expr(my_id, expr->left,
532 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
533 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
534 } else {
535 set_true_false_states_expr(my_id, expr->right,
536 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
537 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
538 }
539 }
540
541 static void handle_unsigned_lt_gt(struct expression *expr)
542 {
543 struct symbol *type;
544 struct range_list *left;
545 struct range_list *right;
546 struct range_list *non_negative;
547 sval_t min, minus_one;
548
549 /*
550 * conditions are mostly handled by smatch_extra.c. The special case
551 * here is that say you have if (user_int < unknown_u32) {
552 * In Smatch extra we say that, We have no idea what value
553 * unknown_u32 is so the only thin we can say for sure is that
554 * user_int is not -1 (UINT_MAX). But in check_user_data2.c we should
555 * assume that unless unknown_u32 is user data, it's probably less than
556 * INT_MAX.
557 *
558 */
559
560 type = get_type(expr);
561 if (!type_unsigned(type))
562 return;
563
564 /*
565 * Assume if (user < trusted) { ... because I am lazy and because this
566 * is the correct way to write code.
567 */
568 if (!get_user_rl(expr->left, &left))
569 return;
570 if (get_user_rl(expr->right, &right))
571 return;
572
573 if (!sval_is_negative(rl_min(left)))
574 return;
575 min = rl_min(left);
576 minus_one.type = rl_type(left);
577 minus_one.value = -1;
578 non_negative = remove_range(left, min, minus_one);
579
580 switch (expr->op) {
581 case '<':
582 case SPECIAL_UNSIGNED_LT:
583 case SPECIAL_LTE:
584 case SPECIAL_UNSIGNED_LTE:
585 set_true_false_states_expr(my_id, expr->left,
586 alloc_estate_rl(non_negative), NULL);
587 break;
588 case '>':
589 case SPECIAL_UNSIGNED_GT:
590 case SPECIAL_GTE:
591 case SPECIAL_UNSIGNED_GTE:
592 set_true_false_states_expr(my_id, expr->left,
593 NULL, alloc_estate_rl(non_negative));
594 break;
595 }
596 }
597
598 static void match_condition(struct expression *expr)
599 {
600 if (expr->type != EXPR_COMPARE)
601 return;
602
603 if (expr->op == SPECIAL_EQUAL ||
604 expr->op == SPECIAL_NOTEQUAL) {
605 handle_eq_noteq(expr);
606 return;
607 }
608
609 handle_unsigned_lt_gt(expr);
610 }
611
612 static void match_user_assign_function(const char *fn, struct expression *expr, void *unused)
613 {
614 tag_as_user_data(expr->left);
615 set_points_to_user_data(expr->left);
616 }
617
618 static void match_returns_user_rl(const char *fn, struct expression *expr, void *unused)
619 {
620 func_gets_user_data = true;
621 }
622
623 static int get_user_macro_rl(struct expression *expr, struct range_list **rl)
624 {
625 struct expression *parent;
626 char *macro;
627
628 if (!expr)
629 return 0;
637 while (parent && parent->type != EXPR_BINOP)
638 parent = expr_get_parent_expr(parent);
639 if (parent && parent->type == EXPR_BINOP) {
640 char *parent_macro = get_macro_name(parent->pos);
641
642 if (parent_macro && strcmp(macro, parent_macro) == 0)
643 return 0;
644 }
645
646 if (strcmp(macro, "ntohl") == 0) {
647 *rl = alloc_whole_rl(&uint_ctype);
648 return 1;
649 }
650 if (strcmp(macro, "ntohs") == 0) {
651 *rl = alloc_whole_rl(&ushort_ctype);
652 return 1;
653 }
654 return 0;
655 }
656
657 struct db_info {
658 struct range_list *rl;
659 struct expression *call;
660 };
661 static int returned_rl_callback(void *_info, int argc, char **argv, char **azColName)
662 {
663 struct db_info *db_info = _info;
664 struct range_list *rl;
665 char *return_ranges = argv[0];
666 char *user_ranges = argv[1];
667 struct expression *arg;
668 int comparison;
669
670 if (argc != 2)
671 return 0;
672
673 call_results_to_rl(db_info->call, get_type(db_info->call), user_ranges, &rl);
674 if (str_to_comparison_arg(return_ranges, db_info->call, &comparison, &arg) &&
675 comparison == SPECIAL_EQUAL) {
676 struct range_list *orig_rl;
677
678 if (!get_user_rl(arg, &orig_rl))
679 return 0;
680 rl = rl_intersection(rl, orig_rl);
681 if (!rl)
682 return 0;
683 }
684 db_info->rl = rl_union(db_info->rl, rl);
685
686 return 0;
687 }
688
689 static int has_user_data(struct symbol *sym)
690 {
691 struct sm_state *tmp;
692
693 FOR_EACH_MY_SM(my_id, __get_cur_stree(), tmp) {
694 if (tmp->sym == sym)
695 return 1;
696 } END_FOR_EACH_SM(tmp);
697 return 0;
698 }
699
700 static int we_pass_user_data(struct expression *call)
701 {
702 struct expression *arg;
703 struct symbol *sym;
704
705 FOR_EACH_PTR(call->args, arg) {
706 sym = expr_to_sym(arg);
707 if (!sym)
708 continue;
709 if (has_user_data(sym))
710 return 1;
711 } END_FOR_EACH_PTR(arg);
712
713 return 0;
714 }
715
716 static int db_returned_user_rl(struct expression *call, struct range_list **rl)
717 {
718 struct db_info db_info = {};
719
720 /* for function pointers assume everything is used */
721 if (call->fn->type != EXPR_SYMBOL)
722 return 0;
723 if (is_fake_call(call))
724 return 0;
725
726 db_info.call = call;
727 run_sql(&returned_rl_callback, &db_info,
728 "select return, value from return_states where %s and type = %d and parameter = -1 and key = '$';",
729 get_static_filter(call->fn->symbol), USER_DATA3_SET);
730 if (db_info.rl) {
731 func_gets_user_data = true;
732 *rl = db_info.rl;
733 return 1;
734 }
735
736 run_sql(&returned_rl_callback, &db_info,
737 "select return, value from return_states where %s and type = %d and parameter = -1 and key = '$';",
738 get_static_filter(call->fn->symbol), USER_DATA3);
739 if (db_info.rl) {
740 if (!we_pass_user_data(call))
741 return 0;
742 *rl = db_info.rl;
743 return 1;
744 }
745
746 return 0;
747 }
748
749 struct stree *get_user_stree(void)
750 {
751 return get_all_states_stree(my_id);
752 }
753
754 static int user_data_flag;
755 static int no_user_data_flag;
756 static struct range_list *var_user_rl(struct expression *expr)
757 {
758 struct smatch_state *state;
759 struct range_list *rl;
760 struct range_list *absolute_rl;
761
762 if (expr->type == EXPR_BINOP && expr->op == '%') {
763 struct range_list *left, *right;
764
765 if (!get_user_rl(expr->right, &right))
766 return NULL;
767 get_absolute_rl(expr->left, &left);
768 rl = rl_binop(left, '%', right);
769 goto found;
770 }
771
772 if (!option_spammy && expr->type == EXPR_BINOP && expr->op == '/') {
773 struct range_list *left = NULL;
774 struct range_list *right = NULL;
775 struct range_list *abs_right;
776
777 /*
778 * The specific bug I'm dealing with is:
779 *
780 * foo = capped_user / unknown;
781 *
782 * Instead of just saying foo is now entirely user_rl we should
783 * probably say instead that it is not at all user data.
784 *
785 */
786
787 get_user_rl(expr->left, &left);
788 get_user_rl(expr->right, &right);
789 get_absolute_rl(expr->right, &abs_right);
790
791 if (left && !right) {
792 rl = rl_binop(left, '/', abs_right);
822
823 if (!get_state_expr(my_id, array)) {
824 no_user_data_flag = 1;
825 return NULL;
826 }
827 }
828
829 if (expr->type == EXPR_PREOP && expr->op == '*' &&
830 is_user_rl(expr->unop)) {
831 rl = var_to_absolute_rl(expr);
832 goto found;
833 }
834
835 return NULL;
836 found:
837 user_data_flag = 1;
838 absolute_rl = var_to_absolute_rl(expr);
839 return clone_rl(rl_intersection(rl, absolute_rl));
840 }
841
842 int get_user_rl(struct expression *expr, struct range_list **rl)
843 {
844 user_data_flag = 0;
845 no_user_data_flag = 0;
846 custom_get_absolute_rl(expr, &var_user_rl, rl);
847 if (!user_data_flag || no_user_data_flag)
848 *rl = NULL;
849
850 return !!*rl;
851 }
852
853 int get_user_rl_spammy(struct expression *expr, struct range_list **rl)
854 {
855 int ret;
856
857 option_spammy++;
858 ret = get_user_rl(expr, rl);
859 option_spammy--;
860
861 return ret;
862 }
863
864 int is_user_rl(struct expression *expr)
865 {
866 struct range_list *tmp;
867
868 return get_user_rl_spammy(expr, &tmp);
869 }
870
871 int get_user_rl_var_sym(const char *name, struct symbol *sym, struct range_list **rl)
872 {
873 struct smatch_state *state;
874
875 state = get_state(my_id, name, sym);
876 if (state && estate_rl(state)) {
877 *rl = estate_rl(state);
878 return 1;
879 }
880 return 0;
881 }
882
883 static void match_call_info(struct expression *expr)
884 {
885 struct range_list *rl;
886 struct expression *arg;
887 struct symbol *type;
888 int i = 0;
889
890 i = -1;
891 FOR_EACH_PTR(expr->args, arg) {
892 i++;
893 type = get_arg_type(expr->fn, i);
894
895 if (!get_user_rl(arg, &rl))
896 continue;
897
898 rl = cast_rl(type, rl);
899 sql_insert_caller_info(expr, USER_DATA3, i, "$", show_rl(rl));
900 } END_FOR_EACH_PTR(arg);
901 }
902
903 static int is_struct_ptr(struct symbol *sym)
904 {
905 struct symbol *type;
906
907 if (!sym)
908 return 0;
909 type = get_real_base_type(sym);
910 if (!type || type->type != SYM_PTR)
911 return 0;
912 type = get_real_base_type(type);
913 if (!type || type->type != SYM_STRUCT)
914 return 0;
915 return 1;
916 }
917
918 static void struct_member_callback(struct expression *call, int param, char *printed_name, struct sm_state *sm)
919 {
920 struct smatch_state *state;
921 struct range_list *rl;
922 struct symbol *type;
923
924 /*
925 * Smatch uses a hack where if we get an unsigned long we say it's
926 * both user data and it points to user data. But if we pass it to a
927 * function which takes an int, then it's just user data. There's not
928 * enough bytes for it to be a pointer.
929 *
930 */
931 type = get_arg_type(call->fn, param);
932 if (type && type_bits(type) < type_bits(&ptr_ctype))
933 return;
934
935 if (strcmp(sm->state->name, "") == 0)
936 return;
937
938 if (strcmp(printed_name, "*$") == 0 &&
939 is_struct_ptr(sm->sym))
940 return;
941
942 state = get_state(SMATCH_EXTRA, sm->name, sm->sym);
943 if (!state || !estate_rl(state))
944 rl = estate_rl(sm->state);
945 else
946 rl = rl_intersection(estate_rl(sm->state), estate_rl(state));
947
948 sql_insert_caller_info(call, USER_DATA3, param, printed_name, show_rl(rl));
949 }
950
951 static void set_param_user_data(const char *name, struct symbol *sym, char *key, char *value)
952 {
953 struct range_list *rl = NULL;
954 struct smatch_state *state;
955 struct symbol *type;
956 char fullname[256];
957
958 if (strcmp(key, "*$") == 0)
959 snprintf(fullname, sizeof(fullname), "*%s", name);
960 else if (strncmp(key, "$", 1) == 0)
961 snprintf(fullname, 256, "%s%s", name, key + 1);
962 else
963 return;
964
965 type = get_member_type_from_key(symbol_expression(sym), key);
966
967 /* if the caller passes a void pointer with user data */
968 if (strcmp(key, "*$") == 0 && type && type != &void_ctype) {
969 struct expression *expr = symbol_expression(sym);
970
971 tag_as_user_data(expr);
972 set_points_to_user_data(expr);
973 return;
974 }
975 str_to_rl(type, value, &rl);
976 state = alloc_estate_rl(rl);
977 set_state(my_id, fullname, sym, state);
978 }
979
980 static void set_called(const char *name, struct symbol *sym, char *key, char *value)
981 {
982 set_state(my_call_id, "this_function", NULL, &called);
983 }
984
985 static void match_syscall_definition(struct symbol *sym)
986 {
987 struct symbol *arg;
988 char *macro;
989 char *name;
990 int is_syscall = 0;
991
992 macro = get_macro_name(sym->pos);
993 if (macro &&
994 (strncmp("SYSCALL_DEFINE", macro, strlen("SYSCALL_DEFINE")) == 0 ||
995 strncmp("COMPAT_SYSCALL_DEFINE", macro, strlen("COMPAT_SYSCALL_DEFINE")) == 0))
996 is_syscall = 1;
997
998 name = get_function();
999 if (!option_no_db && get_state(my_call_id, "this_function", NULL) != &called) {
1000 if (name && strncmp(name, "sys_", 4) == 0)
1001 is_syscall = 1;
1002 }
1003
1004 if (name && strncmp(name, "compat_sys_", 11) == 0)
1005 is_syscall = 1;
1006
1007 if (!is_syscall)
1008 return;
1009
1010 FOR_EACH_PTR(sym->ctype.base_type->arguments, arg) {
1011 set_state(my_id, arg->ident->name, arg, alloc_estate_whole(get_real_base_type(arg)));
1012 } END_FOR_EACH_PTR(arg);
1013 }
1014
1015 static void set_to_user_data(struct expression *expr, char *key, char *value)
1016 {
1017 char *name;
1018 struct symbol *sym;
1019 struct symbol *type;
1020 struct range_list *rl = NULL;
1021
1022 type = get_member_type_from_key(expr, key);
1023 name = get_variable_from_key(expr, key, &sym);
1024 if (!name || !sym)
1025 goto free;
1026
1027 call_results_to_rl(expr, type, value, &rl);
1028
1029 set_state(my_id, name, sym, alloc_estate_rl(rl));
1030 free:
1031 free_string(name);
1032
1033 }
1034
1035 static void returns_param_user_data(struct expression *expr, int param, char *key, char *value)
1036 {
1037 struct expression *arg;
1038 struct expression *call;
1039
1040 call = expr;
1041 while (call->type == EXPR_ASSIGNMENT)
1042 call = strip_expr(call->right);
1043 if (call->type != EXPR_CALL)
1044 return;
1045
1046 if (!we_pass_user_data(call))
1047 return;
1048
1049 if (param == -1) {
1050 if (expr->type != EXPR_ASSIGNMENT)
1051 return;
1052 set_to_user_data(expr->left, key, value);
1053 return;
1054 }
1055
1056 arg = get_argument_from_call_expr(call->args, param);
1057 if (!arg)
1058 return;
1059 set_to_user_data(arg, key, value);
1060 }
1061
1062 static void returns_param_user_data_set(struct expression *expr, int param, char *key, char *value)
1063 {
1064 struct expression *arg;
1065
1066 func_gets_user_data = true;
1067
1068 if (param == -1) {
1069 if (expr->type != EXPR_ASSIGNMENT)
1070 return;
1071 if (strcmp(key, "*$") == 0) {
1072 set_points_to_user_data(expr->left);
1073 tag_as_user_data(expr->left);
1074 } else {
1075 set_to_user_data(expr->left, key, value);
1076 }
1077 return;
1078 }
1079
1080 while (expr->type == EXPR_ASSIGNMENT)
1081 expr = strip_expr(expr->right);
1082 if (expr->type != EXPR_CALL)
1083 return;
1084
1085 arg = get_argument_from_call_expr(expr->args, param);
1086 if (!arg)
1087 return;
1088 set_to_user_data(arg, key, value);
1089 }
1090
1091 static int has_empty_state(struct sm_state *sm)
1092 {
1093 struct sm_state *tmp;
1094
1095 FOR_EACH_PTR(sm->possible, tmp) {
1096 if (!estate_rl(tmp->state))
1097 return 1;
1098 } END_FOR_EACH_PTR(tmp);
1099
1100 return 0;
1101 }
1102
1103 static void param_set_to_user_data(int return_id, char *return_ranges, struct expression *expr)
1104 {
1105 struct sm_state *sm;
1106 struct smatch_state *start_state;
1107 struct range_list *rl;
1108 int param;
1109 char *return_str;
1110 const char *param_name;
1111 struct symbol *ret_sym;
1112 bool return_found = false;
1113
1114 expr = strip_expr(expr);
1115 return_str = expr_to_str(expr);
1116 ret_sym = expr_to_sym(expr);
1117
1118 FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1119 if (has_empty_state(sm))
1120 continue;
1121
1122 param = get_param_num_from_sym(sm->sym);
1123 if (param < 0)
1124 continue;
1125
1126 /* The logic here was that if we were passed in a user data then
1127 * we don't record that. It's like the difference between
1128 * param_filter and param_set. When I think about it, I'm not
1129 * sure it actually works. It's probably harmless because we
1130 * checked earlier that we're not returning a parameter...
1131 * Let's mark this as a TODO.
1132 */
1133 start_state = get_state_stree(start_states, my_id, sm->name, sm->sym);
1134 if (start_state && rl_equiv(estate_rl(sm->state), estate_rl(start_state)))
1135 continue;
1136
1137 param_name = get_param_name(sm);
1138 if (!param_name)
1139 continue;
1140 if (strcmp(param_name, "$") == 0) /* The -1 param is handled after the loop */
1141 continue;
1142
1143 sql_insert_return_states(return_id, return_ranges,
1144 func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1145 param, param_name, show_rl(estate_rl(sm->state)));
1146 } END_FOR_EACH_SM(sm);
1147
1148 if (points_to_user_data(expr)) {
1149 sql_insert_return_states(return_id, return_ranges,
1150 (is_skb_data(expr) || !func_gets_user_data) ?
1151 USER_DATA3_SET : USER_DATA3,
1152 -1, "*$", "");
1153 goto free_string;
1154 }
1155
1156
1157 FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1158 if (!ret_sym)
1159 break;
1160 if (ret_sym != sm->sym)
1161 continue;
1162
1163 param_name = state_name_to_param_name(sm->name, return_str);
1164 if (!param_name)
1165 continue;
1166 if (strcmp(param_name, "$") == 0)
1167 return_found = true;
1168 sql_insert_return_states(return_id, return_ranges,
1169 func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1170 -1, param_name, show_rl(estate_rl(sm->state)));
1171 } END_FOR_EACH_SM(sm);
1172
1173
1174 if (!return_found && get_user_rl(expr, &rl)) {
1175 sql_insert_return_states(return_id, return_ranges,
1176 func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1177 -1, "$", show_rl(rl));
1178 goto free_string;
1179 }
1180
1181 free_string:
1182 free_string(return_str);
1183 }
1184
1185 static struct int_stack *gets_data_stack;
1186 static void match_function_def(struct symbol *sym)
1187 {
1188 func_gets_user_data = false;
1189 }
1190
1191 static void match_inline_start(struct expression *expr)
1192 {
1193 push_int(&gets_data_stack, func_gets_user_data);
1194 }
1195
1196 static void match_inline_end(struct expression *expr)
1197 {
1198 func_gets_user_data = pop_int(&gets_data_stack);
1199 }
1200
1201 void register_kernel_user_data2(int id)
1202 {
1203 int i;
1204
1205 my_id = id;
1206
1207 if (option_project != PROJ_KERNEL)
1208 return;
1209
1210 add_hook(&match_function_def, FUNC_DEF_HOOK);
1211 add_hook(&match_inline_start, INLINE_FN_START);
1212 add_hook(&match_inline_end, INLINE_FN_END);
1213
1214 add_hook(&save_start_states, AFTER_DEF_HOOK);
1215 add_hook(&free_start_states, AFTER_FUNC_HOOK);
1216 add_hook(&match_save_states, INLINE_FN_START);
1217 add_hook(&match_restore_states, INLINE_FN_END);
1218
1219 add_unmatched_state_hook(my_id, &empty_state);
1220 add_extra_nomod_hook(&extra_nomod_hook);
1221 add_pre_merge_hook(my_id, &pre_merge_hook);
1222 add_merge_hook(my_id, &merge_estates);
1223
1224 add_function_hook("copy_from_user", &match_user_copy, INT_PTR(0));
1225 add_function_hook("__copy_from_user", &match_user_copy, INT_PTR(0));
1226 add_function_hook("memcpy_fromiovec", &match_user_copy, INT_PTR(0));
1227 for (i = 0; i < ARRAY_SIZE(kstr_funcs); i++)
1228 add_function_hook(kstr_funcs[i], &match_user_copy, INT_PTR(2));
1229 add_function_hook("usb_control_msg", &match_user_copy, INT_PTR(6));
1230
1231 for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
1232 add_function_assign_hook(returns_user_data[i], &match_user_assign_function, NULL);
1233 add_function_hook(returns_user_data[i], &match_returns_user_rl, NULL);
1234 }
1235
1236 add_function_hook("sscanf", &match_sscanf, NULL);
1237
1238 add_hook(&match_syscall_definition, AFTER_DEF_HOOK);
1239
1240 add_hook(&match_assign, ASSIGNMENT_HOOK);
1241 add_hook(&match_condition, CONDITION_HOOK);
1242
1243 add_hook(&match_call_info, FUNCTION_CALL_HOOK);
1244 add_member_info_callback(my_id, struct_member_callback);
1245 select_caller_info_hook(set_param_user_data, USER_DATA3);
1246 select_return_states_hook(USER_DATA3, &returns_param_user_data);
1247 select_return_states_hook(USER_DATA3_SET, &returns_param_user_data_set);
1248 add_split_return_callback(¶m_set_to_user_data);
1249 }
1250
1251 void register_kernel_user_data3(int id)
1252 {
1253 my_call_id = id;
1254
1255 if (option_project != PROJ_KERNEL)
1256 return;
1257 select_caller_info_hook(set_called, INTERNAL);
1258 }
1259
|
15 * along with this program; if not, see http://www.gnu.org/copyleft/gpl.txt
16 */
17
18 /*
19 * There are a couple checks that try to see if a variable
20 * comes from the user. It would be better to unify them
21 * into one place. Also it we should follow the data down
22 * the call paths. Hence this file.
23 */
24
25 #include "smatch.h"
26 #include "smatch_slist.h"
27 #include "smatch_extra.h"
28
29 static int my_id;
30 static int my_call_id;
31
32 STATE(called);
33 static bool func_gets_user_data;
34
35 static const char *kstr_funcs[] = {
36 "kstrtoull", "kstrtoll", "kstrtoul", "kstrtol", "kstrtouint",
37 "kstrtoint", "kstrtou64", "kstrtos64", "kstrtou32", "kstrtos32",
38 "kstrtou16", "kstrtos16", "kstrtou8", "kstrtos8", "kstrtoull_from_user"
39 "kstrtoll_from_user", "kstrtoul_from_user", "kstrtol_from_user",
40 "kstrtouint_from_user", "kstrtoint_from_user", "kstrtou16_from_user",
41 "kstrtos16_from_user", "kstrtou8_from_user", "kstrtos8_from_user",
42 "kstrtou64_from_user", "kstrtos64_from_user", "kstrtou32_from_user",
43 "kstrtos32_from_user",
44 };
45
46 static const char *returns_user_data[] = {
47 "simple_strtol", "simple_strtoll", "simple_strtoul", "simple_strtoull",
48 "kvm_register_read",
49 };
50
51 static const char *returns_pointer_to_user_data[] = {
52 "nlmsg_data", "nla_data", "memdup_user", "kmap_atomic", "skb_network_header",
53 };
54
55 static void set_points_to_user_data(struct expression *expr);
56
57 static struct stree *start_states;
58 static struct stree_stack *saved_stack;
59 static void save_start_states(struct statement *stmt)
60 {
61 start_states = clone_stree(__get_cur_stree());
62 }
63
64 static void free_start_states(void)
65 {
66 free_stree(&start_states);
67 }
68
69 static void match_save_states(struct expression *expr)
70 {
71 push_stree(&saved_stack, start_states);
72 start_states = NULL;
73 }
74
75 static void match_restore_states(struct expression *expr)
76 {
77 free_stree(&start_states);
78 start_states = pop_stree(&saved_stack);
79 }
80
81 static struct smatch_state *empty_state(struct sm_state *sm)
82 {
83 return alloc_estate_empty();
84 }
85
86 static void pre_merge_hook(struct sm_state *sm)
87 {
88 struct smatch_state *user;
89 struct smatch_state *extra;
90 struct smatch_state *state;
91 struct range_list *rl;
92 sval_t dummy;
93 sval_t sval_100;
94
95 sval_100.value = 100;
96 sval_100.type = &int_ctype;
97
98 user = __get_state(my_id, sm->name, sm->sym);
99 if (!user || !estate_rl(user))
100 return;
101 extra = __get_state(SMATCH_EXTRA, sm->name, sm->sym);
102 if (!extra)
103 return;
104 rl = rl_intersection(estate_rl(user), estate_rl(extra));
105 if (rl_to_sval(rl, &dummy))
106 rl = NULL;
107 state = alloc_estate_rl(clone_rl(rl));
108 if (estate_capped(user) || is_capped_var_sym(sm->name, sm->sym))
109 estate_set_capped(state);
110 set_state(my_id, sm->name, sm->sym, state);
111 }
112
113 static void extra_nomod_hook(const char *name, struct symbol *sym, struct expression *expr, struct smatch_state *state)
114 {
115 struct smatch_state *user, *new;
116 struct range_list *rl;
117
118 user = __get_state(my_id, name, sym);
119 if (!user)
120 return;
121 rl = rl_intersection(estate_rl(user), estate_rl(state));
122 if (rl_equiv(rl, estate_rl(user)))
123 return;
124 new = alloc_estate_rl(rl);
125 if (estate_capped(user))
126 estate_set_capped(new);
127 set_state(my_id, name, sym, new);
128 }
129
130 static bool binop_capped(struct expression *expr)
131 {
132 struct range_list *left_rl;
133 int comparison;
134
135 if (expr->op == '-' && get_user_rl(expr->left, &left_rl)) {
136 if (user_rl_capped(expr->left))
137 return true;
138 comparison = get_comparison(expr->left, expr->right);
139 if (comparison && show_special(comparison)[0] == '>')
140 return true;
141 return false;
142 }
143
144 if (expr->op == '&' || expr->op == '%') {
145 if (is_capped(expr->left) || is_capped(expr->right))
146 return true;
147 if (user_rl_capped(expr->left) || user_rl_capped(expr->right))
148 return true;
149 return false;
150 }
151
152 if (user_rl_capped(expr->left) &&
153 user_rl_capped(expr->right))
154 return true;
155 return false;
156 }
157
158 bool user_rl_capped(struct expression *expr)
159 {
160 struct smatch_state *state;
161 struct range_list *rl;
162 sval_t sval;
163
164 expr = strip_expr(expr);
165 if (!expr)
166 return false;
167 if (get_value(expr, &sval))
168 return true;
169 if (expr->type == EXPR_BINOP)
170 return binop_capped(expr);
171 if ((expr->type == EXPR_PREOP || expr->type == EXPR_POSTOP) &&
172 (expr->op == SPECIAL_INCREMENT || expr->op == SPECIAL_DECREMENT))
173 return user_rl_capped(expr->unop);
174 state = get_state_expr(my_id, expr);
175 if (state)
176 return estate_capped(state);
177
178 if (get_user_rl(expr, &rl))
179 return false; /* uncapped user data */
180
181 return true; /* not actually user data */
182 }
183
184 static void tag_inner_struct_members(struct expression *expr, struct symbol *member)
185 {
186 struct expression *edge_member;
187 struct symbol *base = get_real_base_type(member);
188 struct symbol *tmp;
189
190 if (member->ident)
191 expr = member_expression(expr, '.', member->ident);
192
193 FOR_EACH_PTR(base->symbol_list, tmp) {
194 struct symbol *type;
195
196 type = get_real_base_type(tmp);
197 if (!type)
198 continue;
199
200 if (type->type == SYM_UNION || type->type == SYM_STRUCT) {
201 tag_inner_struct_members(expr, tmp);
202 continue;
203 }
372 if (!expr->member)
373 return 0;
374 if (strcmp(expr->member->name, "data") != 0)
375 return 0;
376
377 sym = expr_to_sym(expr->deref);
378 if (!sym)
379 return 0;
380 sym = get_real_base_type(sym);
381 if (!sym || sym->type != SYM_PTR)
382 return 0;
383 sym = get_real_base_type(sym);
384 if (!sym || sym->type != SYM_STRUCT || !sym->ident)
385 return 0;
386 if (strcmp(sym->ident->name, "sk_buff") != 0)
387 return 0;
388
389 return 1;
390 }
391
392 static bool is_points_to_user_data_fn(struct expression *expr)
393 {
394 int i;
395
396 expr = strip_expr(expr);
397 if (expr->type != EXPR_CALL || expr->fn->type != EXPR_SYMBOL ||
398 !expr->fn->symbol)
399 return false;
400 expr = expr->fn;
401 for (i = 0; i < ARRAY_SIZE(returns_pointer_to_user_data); i++) {
402 if (sym_name_is(returns_pointer_to_user_data[i], expr))
403 return true;
404 }
405 return false;
406 }
407
408 static int get_rl_from_function(struct expression *expr, struct range_list **rl)
409 {
410 int i;
411
412 if (expr->type != EXPR_CALL || expr->fn->type != EXPR_SYMBOL ||
413 !expr->fn->symbol_name || !expr->fn->symbol_name->name)
414 return 0;
415
416 for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
417 if (strcmp(expr->fn->symbol_name->name, returns_user_data[i]) == 0) {
418 *rl = alloc_whole_rl(get_type(expr));
419 return 1;
420 }
421 }
422 return 0;
423 }
424
425 int points_to_user_data(struct expression *expr)
426 {
427 struct smatch_state *state;
428 struct range_list *rl;
429 char buf[256];
430 struct symbol *sym;
431 char *name;
432 int ret = 0;
433
434 expr = strip_expr(expr);
435 if (!expr)
436 return 0;
437 if (is_skb_data(expr))
438 return 1;
439 if (is_points_to_user_data_fn(expr))
440 return 1;
441 if (get_rl_from_function(expr, &rl))
442 return 1;
443
444 if (expr->type == EXPR_BINOP && expr->op == '+') {
445 if (points_to_user_data(expr->left))
446 return 1;
447 if (points_to_user_data(expr->right))
448 return 1;
449 return 0;
450 }
451
452 name = expr_to_var_sym(expr, &sym);
453 if (!name || !sym)
454 goto free;
455 snprintf(buf, sizeof(buf), "*%s", name);
456 state = __get_state(my_id, buf, sym);
457 if (state && estate_rl(state))
458 ret = 1;
459 free:
460 free_string(name);
461 return ret;
462 }
463
464 static void set_points_to_user_data(struct expression *expr)
465 {
466 char *name;
467 struct symbol *sym;
468 char buf[256];
469 struct symbol *type;
470
471 name = expr_to_var_sym(expr, &sym);
472 if (!name || !sym)
473 goto free;
474 snprintf(buf, sizeof(buf), "*%s", name);
475 type = get_type(expr);
476 if (type && type->type == SYM_PTR)
477 type = get_real_base_type(type);
478 if (!type || type->type != SYM_BASETYPE)
479 type = &llong_ctype;
480 set_state(my_id, buf, sym, alloc_estate_whole(type));
481 free:
482 free_string(name);
483 }
484
485 static int comes_from_skb_data(struct expression *expr)
486 {
487 expr = strip_expr(expr);
488 if (!expr || expr->type != EXPR_PREOP || expr->op != '*')
489 return 0;
490
491 expr = strip_expr(expr->unop);
492 if (!expr)
493 return 0;
494 if (expr->type == EXPR_BINOP && expr->op == '+')
495 expr = strip_expr(expr->left);
496
497 return is_skb_data(expr);
498 }
499
500 static int handle_struct_assignment(struct expression *expr)
538
539 static int handle_get_user(struct expression *expr)
540 {
541 char *name;
542 int ret = 0;
543
544 name = get_macro_name(expr->pos);
545 if (!name || strcmp(name, "get_user") != 0)
546 return 0;
547
548 name = expr_to_var(expr->right);
549 if (!name || strcmp(name, "__val_gu") != 0)
550 goto free;
551 set_state_expr(my_id, expr->left, alloc_estate_whole(get_type(expr->left)));
552 ret = 1;
553 free:
554 free_string(name);
555 return ret;
556 }
557
558 static bool handle_op_assign(struct expression *expr)
559 {
560 struct expression *binop_expr;
561 struct smatch_state *state;
562 struct range_list *rl;
563
564 switch (expr->op) {
565 case SPECIAL_ADD_ASSIGN:
566 case SPECIAL_SUB_ASSIGN:
567 case SPECIAL_AND_ASSIGN:
568 case SPECIAL_MOD_ASSIGN:
569 case SPECIAL_SHL_ASSIGN:
570 case SPECIAL_SHR_ASSIGN:
571 case SPECIAL_OR_ASSIGN:
572 case SPECIAL_XOR_ASSIGN:
573 case SPECIAL_MUL_ASSIGN:
574 case SPECIAL_DIV_ASSIGN:
575 binop_expr = binop_expression(expr->left,
576 op_remove_assign(expr->op),
577 expr->right);
578 if (!get_user_rl(binop_expr, &rl))
579 return true;
580
581 rl = cast_rl(get_type(expr->left), rl);
582 state = alloc_estate_rl(rl);
583 if (user_rl_capped(binop_expr))
584 estate_set_capped(state);
585 set_state_expr(my_id, expr->left, state);
586 return true;
587 }
588 return false;
589 }
590
591 static void match_assign(struct expression *expr)
592 {
593 struct range_list *rl;
594 static struct expression *handled;
595 struct smatch_state *state;
596 struct expression *faked;
597
598 faked = get_faked_expression();
599 if (faked && faked == handled)
600 return;
601 if (is_fake_call(expr->right))
602 goto clear_old_state;
603 if (handle_get_user(expr))
604 return;
605 if (points_to_user_data(expr->right)) {
606 handled = expr;
607 set_points_to_user_data(expr->left);
608 }
609 if (handle_struct_assignment(expr))
610 return;
611
612 if (handle_op_assign(expr))
613 return;
614 if (expr->op != '=')
615 goto clear_old_state;
616
617 /* Handled by DB code */
618 if (expr->right->type == EXPR_CALL || __in_fake_parameter_assign)
619 return;
620
621 if (!get_user_rl(expr->right, &rl))
622 goto clear_old_state;
623
624 rl = cast_rl(get_type(expr->left), rl);
625 state = alloc_estate_rl(rl);
626 if (user_rl_capped(expr->right))
627 estate_set_capped(state);
628 set_state_expr(my_id, expr->left, state);
629
630 return;
631
632 clear_old_state:
633 if (get_state_expr(my_id, expr->left))
634 set_state_expr(my_id, expr->left, alloc_estate_empty());
635 }
636
637 static void handle_eq_noteq(struct expression *expr)
638 {
639 struct smatch_state *left_orig, *right_orig;
640
641 left_orig = get_state_expr(my_id, expr->left);
642 right_orig = get_state_expr(my_id, expr->right);
643
644 if (!left_orig && !right_orig)
645 return;
646 if (left_orig && right_orig)
647 return;
648
649 if (left_orig) {
650 set_true_false_states_expr(my_id, expr->left,
651 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
652 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
653 } else {
654 set_true_false_states_expr(my_id, expr->right,
655 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
656 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
657 }
658 }
659
660 static struct range_list *strip_negatives(struct range_list *rl)
661 {
662 sval_t min = rl_min(rl);
663 sval_t minus_one;
664 sval_t over;
665 sval_t max = sval_type_max(rl_type(rl));
666
667 minus_one.type = rl_type(rl);
668 minus_one.value = INT_MAX + 1ULL;
669 over.type = rl_type(rl);
670 over.value = -1;
671
672 if (!rl)
673 return NULL;
674
675 if (type_unsigned(rl_type(rl)) && type_bits(rl_type(rl)) > 31)
676 return remove_range(rl, over, max);
677
678 return remove_range(rl, min, minus_one);
679 }
680
681 static void handle_compare(struct expression *expr)
682 {
683 struct expression *left, *right;
684 struct range_list *left_rl = NULL;
685 struct range_list *right_rl = NULL;
686 struct range_list *user_rl;
687 struct smatch_state *capped_state;
688 struct smatch_state *left_true = NULL;
689 struct smatch_state *left_false = NULL;
690 struct smatch_state *right_true = NULL;
691 struct smatch_state *right_false = NULL;
692 struct symbol *type;
693 sval_t sval;
694
695 left = strip_expr(expr->left);
696 right = strip_expr(expr->right);
697
698 while (left->type == EXPR_ASSIGNMENT)
699 left = strip_expr(left->left);
700
701 /*
702 * Conditions are mostly handled by smatch_extra.c, but there are some
703 * times where the exact values are not known so we can't do that.
704 *
705 * Normally, we might consider using smatch_capped.c to supliment smatch
706 * extra but that doesn't work when we merge unknown uncapped kernel
707 * data with unknown capped user data. The result is uncapped user
708 * data. We need to keep it separate and say that the user data is
709 * capped. In the past, I would have marked this as just regular
710 * kernel data (not user data) but we can't do that these days because
711 * we need to track user data for Spectre.
712 *
713 * The other situation which we have to handle is when we do have an
714 * int and we compare against an unknown unsigned kernel variable. In
715 * that situation we assume that the kernel data is less than INT_MAX.
716 * Otherwise then we get all sorts of array underflow false positives.
717 *
718 */
719
720 /* Handled in smatch_extra.c */
721 if (get_implied_value(left, &sval) ||
722 get_implied_value(right, &sval))
723 return;
724
725 get_user_rl(left, &left_rl);
726 get_user_rl(right, &right_rl);
727
728 /* nothing to do */
729 if (!left_rl && !right_rl)
730 return;
731 /* if both sides are user data that's not a good limit */
732 if (left_rl && right_rl)
733 return;
734
735 if (left_rl)
736 user_rl = left_rl;
737 else
738 user_rl = right_rl;
739
740 type = get_type(expr);
741 if (type_unsigned(type))
742 user_rl = strip_negatives(user_rl);
743 capped_state = alloc_estate_rl(user_rl);
744 estate_set_capped(capped_state);
745
746 switch (expr->op) {
747 case '<':
748 case SPECIAL_UNSIGNED_LT:
749 case SPECIAL_LTE:
750 case SPECIAL_UNSIGNED_LTE:
751 if (left_rl)
752 left_true = capped_state;
753 else
754 right_false = capped_state;
755 break;
756 case '>':
757 case SPECIAL_UNSIGNED_GT:
758 case SPECIAL_GTE:
759 case SPECIAL_UNSIGNED_GTE:
760 if (left_rl)
761 left_false = capped_state;
762 else
763 right_true = capped_state;
764 break;
765 }
766
767 set_true_false_states_expr(my_id, left, left_true, left_false);
768 set_true_false_states_expr(my_id, right, right_true, right_false);
769 }
770
771 static void match_condition(struct expression *expr)
772 {
773 if (expr->type != EXPR_COMPARE)
774 return;
775
776 if (expr->op == SPECIAL_EQUAL ||
777 expr->op == SPECIAL_NOTEQUAL) {
778 handle_eq_noteq(expr);
779 return;
780 }
781
782 handle_compare(expr);
783 }
784
785 static void match_user_assign_function(const char *fn, struct expression *expr, void *unused)
786 {
787 tag_as_user_data(expr->left);
788 set_points_to_user_data(expr->left);
789 }
790
791 static void match_returns_user_rl(const char *fn, struct expression *expr, void *unused)
792 {
793 func_gets_user_data = true;
794 }
795
796 static int get_user_macro_rl(struct expression *expr, struct range_list **rl)
797 {
798 struct expression *parent;
799 char *macro;
800
801 if (!expr)
802 return 0;
810 while (parent && parent->type != EXPR_BINOP)
811 parent = expr_get_parent_expr(parent);
812 if (parent && parent->type == EXPR_BINOP) {
813 char *parent_macro = get_macro_name(parent->pos);
814
815 if (parent_macro && strcmp(macro, parent_macro) == 0)
816 return 0;
817 }
818
819 if (strcmp(macro, "ntohl") == 0) {
820 *rl = alloc_whole_rl(&uint_ctype);
821 return 1;
822 }
823 if (strcmp(macro, "ntohs") == 0) {
824 *rl = alloc_whole_rl(&ushort_ctype);
825 return 1;
826 }
827 return 0;
828 }
829
830 static int has_user_data(struct symbol *sym)
831 {
832 struct sm_state *tmp;
833
834 FOR_EACH_MY_SM(my_id, __get_cur_stree(), tmp) {
835 if (tmp->sym == sym)
836 return 1;
837 } END_FOR_EACH_SM(tmp);
838 return 0;
839 }
840
841 static int we_pass_user_data(struct expression *call)
842 {
843 struct expression *arg;
844 struct symbol *sym;
845
846 FOR_EACH_PTR(call->args, arg) {
847 sym = expr_to_sym(arg);
848 if (!sym)
849 continue;
850 if (has_user_data(sym))
851 return 1;
852 } END_FOR_EACH_PTR(arg);
853
854 return 0;
855 }
856
857 static int db_returned_user_rl(struct expression *call, struct range_list **rl)
858 {
859 struct smatch_state *state;
860 char buf[48];
861
862 if (is_fake_call(call))
863 return 0;
864 snprintf(buf, sizeof(buf), "return %p", call);
865 state = get_state(my_id, buf, NULL);
866 if (!state || !estate_rl(state))
867 return 0;
868 *rl = estate_rl(state);
869 return 1;
870 }
871
872 struct stree *get_user_stree(void)
873 {
874 return get_all_states_stree(my_id);
875 }
876
877 static int user_data_flag;
878 static int no_user_data_flag;
879 struct range_list *var_user_rl(struct expression *expr)
880 {
881 struct smatch_state *state;
882 struct range_list *rl;
883 struct range_list *absolute_rl;
884
885 if (expr->type == EXPR_PREOP && expr->op == '&') {
886 no_user_data_flag = 1;
887 return NULL;
888 }
889
890 if (expr->type == EXPR_BINOP && expr->op == '%') {
891 struct range_list *left, *right;
892
893 if (!get_user_rl(expr->right, &right))
894 return NULL;
895 get_absolute_rl(expr->left, &left);
896 rl = rl_binop(left, '%', right);
897 goto found;
898 }
899
900 if (expr->type == EXPR_BINOP && expr->op == '/') {
901 struct range_list *left = NULL;
902 struct range_list *right = NULL;
903 struct range_list *abs_right;
904
905 /*
906 * The specific bug I'm dealing with is:
907 *
908 * foo = capped_user / unknown;
909 *
910 * Instead of just saying foo is now entirely user_rl we should
911 * probably say instead that it is not at all user data.
912 *
913 */
914
915 get_user_rl(expr->left, &left);
916 get_user_rl(expr->right, &right);
917 get_absolute_rl(expr->right, &abs_right);
918
919 if (left && !right) {
920 rl = rl_binop(left, '/', abs_right);
950
951 if (!get_state_expr(my_id, array)) {
952 no_user_data_flag = 1;
953 return NULL;
954 }
955 }
956
957 if (expr->type == EXPR_PREOP && expr->op == '*' &&
958 is_user_rl(expr->unop)) {
959 rl = var_to_absolute_rl(expr);
960 goto found;
961 }
962
963 return NULL;
964 found:
965 user_data_flag = 1;
966 absolute_rl = var_to_absolute_rl(expr);
967 return clone_rl(rl_intersection(rl, absolute_rl));
968 }
969
970 static bool is_ptr_subtract(struct expression *expr)
971 {
972 expr = strip_expr(expr);
973 if (!expr)
974 return false;
975 if (expr->type == EXPR_BINOP && expr->op == '-' &&
976 type_is_ptr(get_type(expr->left))) {
977 return true;
978 }
979 return false;
980 }
981
982 int get_user_rl(struct expression *expr, struct range_list **rl)
983 {
984 if (is_ptr_subtract(expr))
985 return 0;
986
987 user_data_flag = 0;
988 no_user_data_flag = 0;
989 custom_get_absolute_rl(expr, &var_user_rl, rl);
990 if (!user_data_flag || no_user_data_flag)
991 *rl = NULL;
992
993 return !!*rl;
994 }
995
996 int is_user_rl(struct expression *expr)
997 {
998 struct range_list *tmp;
999
1000 return !!get_user_rl(expr, &tmp);
1001 }
1002
1003 int get_user_rl_var_sym(const char *name, struct symbol *sym, struct range_list **rl)
1004 {
1005 struct smatch_state *state;
1006
1007 state = get_state(my_id, name, sym);
1008 if (state && estate_rl(state)) {
1009 *rl = estate_rl(state);
1010 return 1;
1011 }
1012 return 0;
1013 }
1014
1015 static char *get_user_rl_str(struct expression *expr, struct symbol *type)
1016 {
1017 struct range_list *rl;
1018 static char buf[64];
1019
1020 if (!get_user_rl(expr, &rl))
1021 return NULL;
1022 rl = cast_rl(type, rl);
1023 snprintf(buf, sizeof(buf), "%s%s",
1024 show_rl(rl), user_rl_capped(expr) ? "[c]" : "");
1025 return buf;
1026 }
1027
1028 static void match_call_info(struct expression *expr)
1029 {
1030 struct expression *arg;
1031 struct symbol *type;
1032 char *str;
1033 int i;
1034
1035 i = -1;
1036 FOR_EACH_PTR(expr->args, arg) {
1037 i++;
1038 type = get_arg_type(expr->fn, i);
1039 str = get_user_rl_str(arg, type);
1040 if (!str)
1041 continue;
1042
1043 sql_insert_caller_info(expr, USER_DATA, i, "$", str);
1044 } END_FOR_EACH_PTR(arg);
1045 }
1046
1047 static int is_struct_ptr(struct symbol *sym)
1048 {
1049 struct symbol *type;
1050
1051 if (!sym)
1052 return 0;
1053 type = get_real_base_type(sym);
1054 if (!type || type->type != SYM_PTR)
1055 return 0;
1056 type = get_real_base_type(type);
1057 if (!type || type->type != SYM_STRUCT)
1058 return 0;
1059 return 1;
1060 }
1061
1062 static void struct_member_callback(struct expression *call, int param, char *printed_name, struct sm_state *sm)
1063 {
1064 struct smatch_state *state;
1065 struct range_list *rl;
1066 struct symbol *type;
1067 char buf[64];
1068
1069 /*
1070 * Smatch uses a hack where if we get an unsigned long we say it's
1071 * both user data and it points to user data. But if we pass it to a
1072 * function which takes an int, then it's just user data. There's not
1073 * enough bytes for it to be a pointer.
1074 *
1075 */
1076 type = get_arg_type(call->fn, param);
1077 if (type && type_bits(type) < type_bits(&ptr_ctype))
1078 return;
1079
1080 if (strcmp(sm->state->name, "") == 0)
1081 return;
1082
1083 if (strcmp(printed_name, "*$") == 0 &&
1084 is_struct_ptr(sm->sym))
1085 return;
1086
1087 state = __get_state(SMATCH_EXTRA, sm->name, sm->sym);
1088 if (!state || !estate_rl(state))
1089 rl = estate_rl(sm->state);
1090 else
1091 rl = rl_intersection(estate_rl(sm->state), estate_rl(state));
1092
1093 if (!rl)
1094 return;
1095
1096 snprintf(buf, sizeof(buf), "%s%s", show_rl(rl),
1097 estate_capped(sm->state) ? "[c]" : "");
1098 sql_insert_caller_info(call, USER_DATA, param, printed_name, buf);
1099 }
1100
1101 static void db_param_set(struct expression *expr, int param, char *key, char *value)
1102 {
1103 struct expression *arg;
1104 char *name;
1105 struct symbol *sym;
1106 struct smatch_state *state;
1107
1108 while (expr->type == EXPR_ASSIGNMENT)
1109 expr = strip_expr(expr->right);
1110 if (expr->type != EXPR_CALL)
1111 return;
1112
1113 arg = get_argument_from_call_expr(expr->args, param);
1114 if (!arg)
1115 return;
1116 name = get_variable_from_key(arg, key, &sym);
1117 if (!name || !sym)
1118 goto free;
1119
1120 state = get_state(my_id, name, sym);
1121 if (!state)
1122 goto free;
1123
1124 set_state(my_id, name, sym, alloc_estate_empty());
1125 free:
1126 free_string(name);
1127 }
1128
1129 static bool param_data_capped(const char *value)
1130 {
1131 if (strstr(value, ",c") || strstr(value, "[c"))
1132 return true;
1133 return false;
1134 }
1135
1136 static void set_param_user_data(const char *name, struct symbol *sym, char *key, char *value)
1137 {
1138 struct range_list *rl = NULL;
1139 struct smatch_state *state;
1140 struct expression *expr;
1141 struct symbol *type;
1142 char fullname[256];
1143 char *key_orig = key;
1144 bool add_star = false;
1145
1146 if (strcmp(key, "**$") == 0) {
1147 snprintf(fullname, sizeof(fullname), "**%s", name);
1148 } else {
1149 if (key[0] == '*') {
1150 add_star = true;
1151 key++;
1152 }
1153
1154 snprintf(fullname, 256, "%s%s%s", add_star ? "*" : "", name, key + 1);
1155 }
1156
1157 expr = symbol_expression(sym);
1158 type = get_member_type_from_key(expr, key_orig);
1159
1160 /*
1161 * Say this function takes a struct ponter but the caller passes
1162 * this_function(skb->data). We have two options, we could pass *$
1163 * as user data or we could pass foo->bar, foo->baz as user data.
1164 * The second option is easier to implement so we do that.
1165 *
1166 */
1167 if (strcmp(key_orig, "*$") == 0) {
1168 struct symbol *tmp = type;
1169
1170 while (tmp && tmp->type == SYM_PTR)
1171 tmp = get_real_base_type(tmp);
1172
1173 if (tmp && (tmp->type == SYM_STRUCT || tmp->type == SYM_UNION)) {
1174 tag_as_user_data(symbol_expression(sym));
1175 return;
1176 }
1177 }
1178
1179 str_to_rl(type, value, &rl);
1180 state = alloc_estate_rl(rl);
1181 if (param_data_capped(value) || is_capped(expr))
1182 estate_set_capped(state);
1183 set_state(my_id, fullname, sym, state);
1184 }
1185
1186 static void set_called(const char *name, struct symbol *sym, char *key, char *value)
1187 {
1188 set_state(my_call_id, "this_function", NULL, &called);
1189 }
1190
1191 static void match_syscall_definition(struct symbol *sym)
1192 {
1193 struct symbol *arg;
1194 char *macro;
1195 char *name;
1196 int is_syscall = 0;
1197
1198 macro = get_macro_name(sym->pos);
1199 if (macro &&
1200 (strncmp("SYSCALL_DEFINE", macro, strlen("SYSCALL_DEFINE")) == 0 ||
1201 strncmp("COMPAT_SYSCALL_DEFINE", macro, strlen("COMPAT_SYSCALL_DEFINE")) == 0))
1202 is_syscall = 1;
1203
1204 name = get_function();
1205 if (!option_no_db && get_state(my_call_id, "this_function", NULL) != &called) {
1206 if (name && strncmp(name, "sys_", 4) == 0)
1207 is_syscall = 1;
1208 }
1209
1210 if (name && strncmp(name, "compat_sys_", 11) == 0)
1211 is_syscall = 1;
1212
1213 if (!is_syscall)
1214 return;
1215
1216 FOR_EACH_PTR(sym->ctype.base_type->arguments, arg) {
1217 set_state(my_id, arg->ident->name, arg, alloc_estate_whole(get_real_base_type(arg)));
1218 } END_FOR_EACH_PTR(arg);
1219 }
1220
1221 static void store_user_data_return(struct expression *expr, char *key, char *value)
1222 {
1223 struct range_list *rl;
1224 struct symbol *type;
1225 char buf[48];
1226
1227 if (strcmp(key, "$") != 0)
1228 return;
1229
1230 type = get_type(expr);
1231 snprintf(buf, sizeof(buf), "return %p", expr);
1232 call_results_to_rl(expr, type, value, &rl);
1233
1234 set_state(my_id, buf, NULL, alloc_estate_rl(rl));
1235 }
1236
1237 static void set_to_user_data(struct expression *expr, char *key, char *value)
1238 {
1239 struct smatch_state *state;
1240 char *name;
1241 struct symbol *sym;
1242 struct symbol *type;
1243 struct range_list *rl = NULL;
1244
1245 type = get_member_type_from_key(expr, key);
1246 name = get_variable_from_key(expr, key, &sym);
1247 if (!name || !sym)
1248 goto free;
1249
1250 call_results_to_rl(expr, type, value, &rl);
1251
1252 state = alloc_estate_rl(rl);
1253 if (param_data_capped(value))
1254 estate_set_capped(state);
1255 set_state(my_id, name, sym, state);
1256 free:
1257 free_string(name);
1258 }
1259
1260 static void returns_param_user_data(struct expression *expr, int param, char *key, char *value)
1261 {
1262 struct expression *arg;
1263 struct expression *call;
1264
1265 call = expr;
1266 while (call->type == EXPR_ASSIGNMENT)
1267 call = strip_expr(call->right);
1268 if (call->type != EXPR_CALL)
1269 return;
1270
1271 if (!we_pass_user_data(call))
1272 return;
1273
1274 if (param == -1) {
1275 if (expr->type != EXPR_ASSIGNMENT) {
1276 store_user_data_return(expr, key, value);
1277 return;
1278 }
1279 set_to_user_data(expr->left, key, value);
1280 return;
1281 }
1282
1283 arg = get_argument_from_call_expr(call->args, param);
1284 if (!arg)
1285 return;
1286 set_to_user_data(arg, key, value);
1287 }
1288
1289 static void returns_param_user_data_set(struct expression *expr, int param, char *key, char *value)
1290 {
1291 struct expression *arg;
1292
1293 func_gets_user_data = true;
1294
1295 if (param == -1) {
1296 if (expr->type != EXPR_ASSIGNMENT) {
1297 store_user_data_return(expr, key, value);
1298 return;
1299 }
1300 if (strcmp(key, "*$") == 0) {
1301 set_points_to_user_data(expr->left);
1302 tag_as_user_data(expr->left);
1303 } else {
1304 set_to_user_data(expr->left, key, value);
1305 }
1306 return;
1307 }
1308
1309 while (expr->type == EXPR_ASSIGNMENT)
1310 expr = strip_expr(expr->right);
1311 if (expr->type != EXPR_CALL)
1312 return;
1313
1314 arg = get_argument_from_call_expr(expr->args, param);
1315 if (!arg)
1316 return;
1317 set_to_user_data(arg, key, value);
1318 }
1319
1320 static void param_set_to_user_data(int return_id, char *return_ranges, struct expression *expr)
1321 {
1322 struct sm_state *sm;
1323 struct smatch_state *start_state;
1324 struct range_list *rl;
1325 int param;
1326 char *return_str;
1327 const char *param_name;
1328 struct symbol *ret_sym;
1329 bool return_found = false;
1330 bool pointed_at_found = false;
1331 char buf[64];
1332
1333 expr = strip_expr(expr);
1334 return_str = expr_to_str(expr);
1335 ret_sym = expr_to_sym(expr);
1336
1337 FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1338 param = get_param_num_from_sym(sm->sym);
1339 if (param < 0)
1340 continue;
1341
1342 if (!param_was_set_var_sym(sm->name, sm->sym))
1343 continue;
1344
1345 /* The logic here was that if we were passed in a user data then
1346 * we don't record that. It's like the difference between
1347 * param_filter and param_set. When I think about it, I'm not
1348 * sure it actually works. It's probably harmless because we
1349 * checked earlier that we're not returning a parameter...
1350 * Let's mark this as a TODO.
1351 */
1352 start_state = get_state_stree(start_states, my_id, sm->name, sm->sym);
1353 if (start_state && rl_equiv(estate_rl(sm->state), estate_rl(start_state)))
1354 continue;
1355
1356 param_name = get_param_name(sm);
1357 if (!param_name)
1358 continue;
1359 if (strcmp(param_name, "$") == 0) /* The -1 param is handled after the loop */
1360 continue;
1361
1362 snprintf(buf, sizeof(buf), "%s%s",
1363 show_rl(estate_rl(sm->state)),
1364 estate_capped(sm->state) ? "[c]" : "");
1365 sql_insert_return_states(return_id, return_ranges,
1366 func_gets_user_data ? USER_DATA_SET : USER_DATA,
1367 param, param_name, buf);
1368 } END_FOR_EACH_SM(sm);
1369
1370 /* This if for "return foo;" where "foo->bar" is user data. */
1371 FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1372 if (!ret_sym)
1373 break;
1374 if (ret_sym != sm->sym)
1375 continue;
1376
1377 param_name = state_name_to_param_name(sm->name, return_str);
1378 if (!param_name)
1379 continue;
1380 if (strcmp(param_name, "$") == 0)
1381 return_found = true;
1382 if (strcmp(param_name, "*$") == 0)
1383 pointed_at_found = true;
1384 snprintf(buf, sizeof(buf), "%s%s",
1385 show_rl(estate_rl(sm->state)),
1386 estate_capped(sm->state) ? "[c]" : "");
1387 sql_insert_return_states(return_id, return_ranges,
1388 func_gets_user_data ? USER_DATA_SET : USER_DATA,
1389 -1, param_name, buf);
1390 } END_FOR_EACH_SM(sm);
1391
1392 /* This if for "return ntohl(foo);" */
1393 if (!return_found && get_user_rl(expr, &rl)) {
1394 snprintf(buf, sizeof(buf), "%s%s",
1395 show_rl(rl), user_rl_capped(expr) ? "[c]" : "");
1396 sql_insert_return_states(return_id, return_ranges,
1397 func_gets_user_data ? USER_DATA_SET : USER_DATA,
1398 -1, "$", buf);
1399 }
1400
1401 /*
1402 * This is to handle things like return skb->data where we don't set a
1403 * state for that.
1404 */
1405 if (!pointed_at_found && points_to_user_data(expr)) {
1406 sql_insert_return_states(return_id, return_ranges,
1407 (is_skb_data(expr) || func_gets_user_data) ?
1408 USER_DATA_SET : USER_DATA,
1409 -1, "*$", "s64min-s64max");
1410 }
1411
1412 free_string(return_str);
1413 }
1414
1415 static void returns_param_capped(struct expression *expr, int param, char *key, char *value)
1416 {
1417 struct smatch_state *state, *new;
1418 struct symbol *sym;
1419 char *name;
1420
1421 name = return_state_to_var_sym(expr, param, key, &sym);
1422 if (!name || !sym)
1423 goto free;
1424
1425 state = get_state(my_id, name, sym);
1426 if (!state || estate_capped(state))
1427 goto free;
1428
1429 new = clone_estate(state);
1430 estate_set_capped(new);
1431
1432 set_state(my_id, name, sym, new);
1433 free:
1434 free_string(name);
1435 }
1436
1437 static struct int_stack *gets_data_stack;
1438 static void match_function_def(struct symbol *sym)
1439 {
1440 func_gets_user_data = false;
1441 }
1442
1443 static void match_inline_start(struct expression *expr)
1444 {
1445 push_int(&gets_data_stack, func_gets_user_data);
1446 }
1447
1448 static void match_inline_end(struct expression *expr)
1449 {
1450 func_gets_user_data = pop_int(&gets_data_stack);
1451 }
1452
1453 void register_kernel_user_data(int id)
1454 {
1455 int i;
1456
1457 my_id = id;
1458
1459 if (option_project != PROJ_KERNEL)
1460 return;
1461
1462 set_dynamic_states(my_id);
1463
1464 add_hook(&match_function_def, FUNC_DEF_HOOK);
1465 add_hook(&match_inline_start, INLINE_FN_START);
1466 add_hook(&match_inline_end, INLINE_FN_END);
1467
1468 add_hook(&save_start_states, AFTER_DEF_HOOK);
1469 add_hook(&free_start_states, AFTER_FUNC_HOOK);
1470 add_hook(&match_save_states, INLINE_FN_START);
1471 add_hook(&match_restore_states, INLINE_FN_END);
1472
1473 add_unmatched_state_hook(my_id, &empty_state);
1474 add_extra_nomod_hook(&extra_nomod_hook);
1475 add_pre_merge_hook(my_id, &pre_merge_hook);
1476 add_merge_hook(my_id, &merge_estates);
1477
1478 add_function_hook("copy_from_user", &match_user_copy, INT_PTR(0));
1479 add_function_hook("__copy_from_user", &match_user_copy, INT_PTR(0));
1480 add_function_hook("memcpy_fromiovec", &match_user_copy, INT_PTR(0));
1481 for (i = 0; i < ARRAY_SIZE(kstr_funcs); i++)
1482 add_function_hook(kstr_funcs[i], &match_user_copy, INT_PTR(2));
1483 add_function_hook("usb_control_msg", &match_user_copy, INT_PTR(6));
1484
1485 for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
1486 add_function_assign_hook(returns_user_data[i], &match_user_assign_function, NULL);
1487 add_function_hook(returns_user_data[i], &match_returns_user_rl, NULL);
1488 }
1489
1490 add_function_hook("sscanf", &match_sscanf, NULL);
1491
1492 add_hook(&match_syscall_definition, AFTER_DEF_HOOK);
1493
1494 add_hook(&match_assign, ASSIGNMENT_HOOK);
1495 select_return_states_hook(PARAM_SET, &db_param_set);
1496 add_hook(&match_condition, CONDITION_HOOK);
1497
1498 add_hook(&match_call_info, FUNCTION_CALL_HOOK);
1499 add_member_info_callback(my_id, struct_member_callback);
1500 select_caller_info_hook(set_param_user_data, USER_DATA);
1501 select_return_states_hook(USER_DATA, &returns_param_user_data);
1502 select_return_states_hook(USER_DATA_SET, &returns_param_user_data_set);
1503 select_return_states_hook(CAPPED_DATA, &returns_param_capped);
1504 add_split_return_callback(¶m_set_to_user_data);
1505 }
1506
1507 void register_kernel_user_data2(int id)
1508 {
1509 my_call_id = id;
1510
1511 if (option_project != PROJ_KERNEL)
1512 return;
1513 select_caller_info_hook(set_called, INTERNAL);
1514 }
|