On Fri, 2010-12-31 at 15:51 +0100, Jesper Juhl wrote:
> On Fri, 31 Dec 2010, Malcolm Priestley wrote:
>
> > On Fri, 2010-12-31 at 00:11 +0100, Jesper Juhl wrote:
> > > Hi,
> > >
> > > We may leak the storage allocated to 'state' in
> > > drivers/media/dvb/frontends/ix2505v.c::ix2505v_attach() on error.
> > > This patch makes sure we free the allocated memory in the failure case.
> > >
> > >
> > > Signed-off-by: Jesper Juhl <j...@chaosbits.net>
> > > ---
> > > ix2505v.c | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > Compile tested only.
> > >
> > > diff --git a/drivers/media/dvb/frontends/ix2505v.c
> > > b/drivers/media/dvb/frontends/ix2505v.c
> > > index 55f2eba..fcb173d 100644
> > > --- a/drivers/media/dvb/frontends/ix2505v.c
> > > +++ b/drivers/media/dvb/frontends/ix2505v.c
> > > @@ -293,6 +293,7 @@ struct dvb_frontend *ix2505v_attach(struct
> > > dvb_frontend *fe,
> > > ret = ix2505v_read_status_reg(state);
> > >
> > > if (ret & 0x80) {
> > > + kfree(state);
> > > deb_i2c("%s: No IX2505V found\n", __func__);
> > > goto error;
> > > }
> > >
> > Memory is freed in...
> >
> > error:
> > ix2505v_release(fe);
> > return NULL;
> >
> > via...
> >
> > static int ix2505v_release(struct dvb_frontend *fe)
> > {
> > struct ix2505v_state *state = fe->tuner_priv;
> >
> > fe->tuner_priv = NULL;
> > kfree(state);
> >
> > return 0;
> > }
> >
>
> Except that 'state' has not been assigned to fe->tuner_priv at this
> point, so ix2505v_release() cannot free the memory that was just
> allocated with kzalloc().
>
>
> state is a local variable:
> struct ix2505v_state *state = NULL;
> ...
>
> we allocate memory and assign it to 'state' here:
> state = kzalloc(sizeof(struct ix2505v_state), GFP_KERNEL);
> if (NULL == state)
> return NULL;
>
> state->config = config;
> state->i2c = i2c;
>
> here 'state' is used, but not in a way that saves it anywhere:
> if (state->config->tuner_write_only) {
> if (fe->ops.i2c_gate_ctrl)
> fe->ops.i2c_gate_ctrl(fe, 1);
>
> this function call involves 'state' but it does not save it anywhere
> either:
> ret = ix2505v_read_status_reg(state);
>
> if (ret & 0x80) {
> deb_i2c("%s: No IX2505V found\n", __func__);
> so when we jump to error here 'state' still exists only as the local
> variable, it has not been assigned to anything else.
> goto error;
> }
> ...
> error:
> there is no way this function call can free 'state' on this path since
> it has not been assigned to fe->tuner_priv.
> ix2505v_release(fe);
> The local variable state goes out of scope here and leaks the memory it
> points to:
> return NULL;
> }
>
> Am I missing something?
Oh, Sorry, I see it now.
Now there is two options.
Either;
1) Move fe->tuner_priv = state to below line 287, so it can be released
by ix2505v_release and fe->tuner_priv returned to NULL;
2) or not calling ix2505v_release changing line 314 to kfree(state).
fe->tuner_priv will remain NULL through out.
Currently, tuner_write_only is not implemented in the dvb-usb-lmedm04
driver, as sometimes it returned unpredictable results, and wrongly
failed to attach the tuner. Although, I will test it again.
Regards
Malcolm
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
