media: ddbridge: fix teardown/deregistration order in ddb_input_detach()

Brought to attention by Matthias Schwarzott <zzam@gentoo.org> by fixing
possible use-after-free faults in some demod drivers:

In ddb_input_detach(), the i2c_client is unregistered and removed before
dvb frontends are unregistered and detached. While no use-after-free issue
was observed so far, there is another issue with this:

dvb->attached keeps track of the state of the input/output registration,
and the i2c_client unregistration takes place only if everything was
successful (dvb->attached == 0x31). If for some reason an error occurred
during the frontend setup, that value stays at 0x20. In the following
error handling and cleanup, ddb_input_detach() will skip down to that
state, leaving the i2c_client registered, causing refcount issues.

Fix this by moving the i2c_client deregistration down to case 0x20.

Cc: Matthias Schwarzott <zzam@gentoo.org>
Signed-off-by: Daniel Scheller <d.scheller@gmx.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
This commit is contained in:
Daniel Scheller 2017-08-23 12:09:59 -04:00 committed by Mauro Carvalho Chehab
parent 6b852620fa
commit 1502efd2d5

View file

@ -1255,11 +1255,6 @@ static void dvb_input_detach(struct ddb_input *input)
switch (dvb->attached) {
case 0x31:
client = dvb->i2c_client[0];
if (client) {
module_put(client->dev.driver->owner);
i2c_unregister_device(client);
}
if (dvb->fe2)
dvb_unregister_frontend(dvb->fe2);
if (dvb->fe)
@ -1273,6 +1268,12 @@ static void dvb_input_detach(struct ddb_input *input)
dvb->fe = dvb->fe2 = NULL;
/* fallthrough */
case 0x20:
client = dvb->i2c_client[0];
if (client) {
module_put(client->dev.driver->owner);
i2c_unregister_device(client);
}
dvb_net_release(&dvb->dvbnet);
/* fallthrough */
case 0x12: