Re: [PATCH] analyzer: fix ICE casued by dup2 in sm-fd.cc[PR106551]

[David Malcolm <dmalcolm@redhat.com>]

On Tue, 2022-08-09 at 21:42 +0530, Immad Mir wrote:
> This patch fixes the ICE caused by valid_to_unchecked_state,
> at analyzer/sm-fd.cc by handling the m_start state in
> check_for_dup.
> 
> Tested lightly on x86_64.
> 
> gcc/analyzer/ChangeLog:
>         PR analyzer/106551
>         * sm-fd.cc (check_for_dup): handle the m_start
>         state when transitioning the state of LHS
>         of dup, dup2 and dup3 call.
> 
> gcc/testsuite/ChangeLog:
>         * gcc.dg/analyzer/fd-dup-1.c: New testcases.
> 
> Signed-off-by: Immad Mir <mirimmad@outlook.com>
> ---
>  gcc/analyzer/sm-fd.cc                    |  4 ++--
>  gcc/testsuite/gcc.dg/analyzer/fd-dup-1.c | 28
> +++++++++++++++++++++++-
>  2 files changed, 29 insertions(+), 3 deletions(-)
> 
> diff --git a/gcc/analyzer/sm-fd.cc b/gcc/analyzer/sm-fd.cc
> index 8bb76d72b05..c8b9930a7b6 100644
> --- a/gcc/analyzer/sm-fd.cc
> +++ b/gcc/analyzer/sm-fd.cc
> @@ -983,7 +983,7 @@ fd_state_machine::check_for_dup (sm_context
> *sm_ctxt, const supernode *node,
>      case DUP_1:
>        if (lhs)
>         {
> -         if (is_constant_fd_p (state_arg_1))
> +         if (is_constant_fd_p (state_arg_1) || state_arg_1 ==
> m_start)
>             sm_ctxt->set_next_state (stmt, lhs,
> m_unchecked_read_write);
>           else
>             sm_ctxt->set_next_state (stmt, lhs,
> @@ -1011,7 +1011,7 @@ fd_state_machine::check_for_dup (sm_context
> *sm_ctxt, const supernode *node,
>        file descriptor i.e the first argument.  */
>        if (lhs)
>         {
> -         if (is_constant_fd_p (state_arg_1))
> +         if (is_constant_fd_p (state_arg_1) || state_arg_1 ==
> m_start)
>             sm_ctxt->set_next_state (stmt, lhs,
> m_unchecked_read_write);
>           else
>             sm_ctxt->set_next_state (stmt, lhs,
> diff --git a/gcc/testsuite/gcc.dg/analyzer/fd-dup-1.c
> b/gcc/testsuite/gcc.dg/analyzer/fd-dup-1.c
> index eba2570568f..ed4d6de57db 100644
> --- a/gcc/testsuite/gcc.dg/analyzer/fd-dup-1.c
> +++ b/gcc/testsuite/gcc.dg/analyzer/fd-dup-1.c
> @@ -220,4 +220,30 @@ test_19 (const char *path, void *buf)
>          close (fd);
>      }
>      
> -}
> \ No newline at end of file
> +}
> +
> +void
> +test_20 ()
> +{
> +    int m;
> +    int fd = dup (m); /* { dg-warning "'dup' on possibly invalid
> file descriptor 'm'" } */
> +    close (fd);
> +}
> +
> +void
> +test_21 ()
> +{
> +    int m;
> +    int fd = dup2 (m, 1); /* { dg-warning "'dup2' on possibly
> invalid file descriptor 'm'" } */
> +    close (fd);
> +}
> +
> +void
> +test_22 (int flags)
> +{
> +    int m;
> +    int fd = dup3 (m, 1, flags); /* { dg-warning "'dup3' on possibly
> invalid file descriptor 'm'" } */
> +    close (fd);
> +}

Thanks for the updated patch.

The test cases looked suspicious to me - I was wondering why the
analyzer doesn't complain about the uninitialized values being passed
to the various dup functions as parameters.  So your test cases seem to
have uncovered a hidden pre-existing bug in the analyzer's
uninitialized value detection, which I've filed for myself to deal with
as PR analyzer/106573.

If you convert the "int m;" locals into an extern global, like in
comment #0 of bug 106551, does that still trigger the crash on the
unpatched sm-fd.cc?  If so, then that's greatly preferable as a
regression test, since otherwise I'll have to modify that test case
when I fix bug 106573.

Dave