2005-12-24 03:12:47

by Lee Revell

Subject: kernel/auditsc.c bug

gcc 4.0.2 complains that state is (not "may be", is) used initialized on
line 607:

605 list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_USER], list) {
606 if (audit_filter_user_rules(cb, &e->rule, &state)) {
607 if (state == AUDIT_DISABLED)
608 ret = 0;
609 break;
610 }

AFAICT state will always have been initialized if
audit_filter_user_rules() returns nonzero:

590 switch (rule->action) {
591 case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
592 case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT; break;
593 case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
594 }
595 return 1;

Is GCC correct that this is a bug (no default case in the switch
statement)?

Lee

2005-12-24 03:28:18

by Dmitry Torokhov

Subject: Re: kernel/auditsc.c bug

On Friday 23 December 2005 22:17, Lee Revell wrote:
> gcc 4.0.2 complains that state is (not "may be", is) used initialized on
> line 607:
>
> 605 list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_USER], list) {
> 606 if (audit_filter_user_rules(cb, &e->rule, &state)) {
> 607 if (state == AUDIT_DISABLED)
> 608 ret = 0;
> 609 break;
> 610 }
>
> AFAICT state will always have been initialized if
> audit_filter_user_rules() returns nonzero:
>
> 590 switch (rule->action) {
> 591 case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
> 592 case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT; break;
> 593 case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
> 594 }
> 595 return 1;
>
> Is GCC correct that this is a bug (no default case in the switch
> statement)?
>

Well, rule actions are #defines, how can a compiler know that the switch
covers all possible values? If they were enums OTOH...

--
Dmitry