Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 8 Dec 2020 16:28:44 +0300
From:   Dan Carpenter <dan.carpenter@oracle.com>
To:     Jakub Kicinski <kuba@kernel.org>,
        Zhang Changzhong <zhangchangzhong@huawei.com>
Cc:     linux-sparse@vger.kernel.org, linux-kernel@vger.kernel.org,
        edwin.peer@broadcom.com
Subject: Re: sparse annotation for error types?
Message-ID: <20201208132844.GC2767@kadam>
References: <20201205143250.2378b9f9@kicinski-fedora-pc1c0hjn.DHCP.thefacebook.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201205143250.2378b9f9@kicinski-fedora-pc1c0hjn.DHCP.thefacebook.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

Hi Zhang,

Are you using Coccinelle to detect these bugs?

On Sat, Dec 05, 2020 at 02:32:50PM -0800, Jakub Kicinski wrote:
> Hi!
> 
> Recently we've been getting a steady stream of patches from Changzhong
> to fix missing assignment to error variables before jumping to error
> cases.

I've mucked about with this a little in Smatch trying to work out some
heuristics to use.  I added a warning for a NULL return followed by a
goto.  Then on Friday I added a warning for a _dev_err() print followed
by a goto.  But neither of those rules catches the bug fixed by commit
4de377b65903 ("net: marvell: prestera: Fix error return code in
prestera_port_create()"), where the error was invalid data.

	if (idx >= size)
		goto free_whatever;

I'm going to print a warning if the function ends in a cleanup block
that can only be reached by gotos.  We'll see how that works tomorrow.

static void match_return(struct statement *stmt)
{
        struct sm_state *sm, *tmp;
        sval_t sval;
        char *name;
        bool is_last;

	// Only complain if the function returns a variable
        if (!stmt->ret_value || stmt->ret_value->type != EXPR_SYMBOL)
                return;

	// The function returns an int
        if (cur_func_return_type() != &int_ctype)
                return;

	// It's only reachable via a goto
        if (get_state(my_id, "path", NULL) != &label)
                return;

	// It returns a negative error code
        sm = get_extra_sm_state(stmt->ret_value);
        if (!sm || !estate_rl(sm->state) ||
            !sval_is_negative(rl_min(estate_rl(sm->state))))
                return;

        FOR_EACH_PTR(sm->possible, tmp) {
		// There is at least one path where "ret" is zero
                if (estate_get_single_value(tmp->state, &sval) &&
                    sval.value == 0)
                        goto warn;
        } END_FOR_EACH_PTR(tmp);

        return;
warn:
	// It's the last statement of a function
        is_last = is_last_stmt(stmt);

        name = expr_to_str(stmt->ret_value);
        sm_warning("missing error code '%s' rl='%s' is_last=%d", name, sm->state->name, is_last);
        free_string(name);
}

regards,
dan carpenter