| Message ID | 20220822071902.3419042-1-tcs_kernel@tencent.com (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | net/ieee802154: fix uninit value bug in dgram_sendmsg | expand |
Hello. On 22.08.22 09:19, Haimin Zhang wrote: > There is uninit value bug in dgram_sendmsg function in > net/ieee802154/socket.c when the length of valid data pointed by the > msg->msg_name isn't verified. > > This length is specified by msg->msg_namelen. Function > ieee802154_addr_from_sa is called by dgram_sendmsg, which use > msg->msg_name as struct sockaddr_ieee802154* and read it, that will > eventually lead to uninit value read. So we should check the length of > msg->msg_name is not less than sizeof(struct sockaddr_ieee802154) > before entering the ieee802154_addr_from_sa. > > Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com> This patch has been applied to the wpan tree and will be part of the next pull request to net. Thanks! Btw, I got a warning from the checkpatch script that your author and SOB email addresses do not match. Might be a good idea to fix this. If you are having trouble to send patches through the company mail server there is always the option to use the gmail address for sending the mail and an internal From: header in the patch to fix up the author. regards Stefan Schmidt
Hi, On Tue, Aug 23, 2022 at 5:42 AM Stefan Schmidt <stefan@datenfreihafen.org> wrote: > > Hello. > > On 22.08.22 09:19, Haimin Zhang wrote: > > There is uninit value bug in dgram_sendmsg function in > > net/ieee802154/socket.c when the length of valid data pointed by the > > msg->msg_name isn't verified. > > > > This length is specified by msg->msg_namelen. Function > > ieee802154_addr_from_sa is called by dgram_sendmsg, which use > > msg->msg_name as struct sockaddr_ieee802154* and read it, that will > > eventually lead to uninit value read. So we should check the length of > > msg->msg_name is not less than sizeof(struct sockaddr_ieee802154) > > before entering the ieee802154_addr_from_sa. > > > > Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com> > > > This patch has been applied to the wpan tree and will be > part of the next pull request to net. Thanks! For me this patch is buggy or at least it is questionable how to deal with the size of ieee802154_addr_sa here. There should be a helper to calculate the size which depends on the addr_type field. It is not required to send the last 6 bytes if addr_type is IEEE802154_ADDR_SHORT. Nitpick is that we should check in the beginning of that function. - Alex
Hi, cc mailing lists again. On Wed, Aug 24, 2022 at 7:55 AM zhang haiming <tcs.kernel@gmail.com> wrote: > > If msg_namelen is too small like 1, the addr_type field will be > unexpected. Meanwhile, check msg_namelen < sizeof(*daddr) is Then check if space for addr_type is available, if not -EINVAL. If addr_type available, evaluate it, if it's unknown -EINVAL, the minimum length differs here if it's known. > necessary and enough as dgram_bind and dgram_connect did. > you probably found similar issues. It is a nitpick and I see that the current behaviour is not correct here. - Alex
Hello Alex. On 23.08.22 14:22, Alexander Aring wrote: > Hi, > > On Tue, Aug 23, 2022 at 5:42 AM Stefan Schmidt > <stefan@datenfreihafen.org> wrote: >> >> Hello. >> >> On 22.08.22 09:19, Haimin Zhang wrote: >>> There is uninit value bug in dgram_sendmsg function in >>> net/ieee802154/socket.c when the length of valid data pointed by the >>> msg->msg_name isn't verified. >>> >>> This length is specified by msg->msg_namelen. Function >>> ieee802154_addr_from_sa is called by dgram_sendmsg, which use >>> msg->msg_name as struct sockaddr_ieee802154* and read it, that will >>> eventually lead to uninit value read. So we should check the length of >>> msg->msg_name is not less than sizeof(struct sockaddr_ieee802154) >>> before entering the ieee802154_addr_from_sa. >>> >>> Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com> >> >> >> This patch has been applied to the wpan tree and will be >> part of the next pull request to net. Thanks! > > For me this patch is buggy or at least it is questionable how to deal > with the size of ieee802154_addr_sa here. You are right. I completely missed this. Thanks for spotting! > There should be a helper to calculate the size which depends on the > addr_type field. It is not required to send the last 6 bytes if > addr_type is IEEE802154_ADDR_SHORT. > Nitpick is that we should check in the beginning of that function. Haimin, in ieee802154 we could have two different sizes for ieee802154_addr_sa depending on the addr_type. We have short and extended addresses. Could you please rework this patch to take this into account as Alex suggested? I reverted your original patch from my tree. regards Stefan Schmidt
Thanks to all. I have sent patch v2 to fix this. On Mon, Aug 29, 2022 at 5:08 PM Stefan Schmidt <stefan@datenfreihafen.org> wrote: > > > Hello Alex. > > On 23.08.22 14:22, Alexander Aring wrote: > > Hi, > > > > On Tue, Aug 23, 2022 at 5:42 AM Stefan Schmidt > > <stefan@datenfreihafen.org> wrote: > >> > >> Hello. > >> > >> On 22.08.22 09:19, Haimin Zhang wrote: > >>> There is uninit value bug in dgram_sendmsg function in > >>> net/ieee802154/socket.c when the length of valid data pointed by the > >>> msg->msg_name isn't verified. > >>> > >>> This length is specified by msg->msg_namelen. Function > >>> ieee802154_addr_from_sa is called by dgram_sendmsg, which use > >>> msg->msg_name as struct sockaddr_ieee802154* and read it, that will > >>> eventually lead to uninit value read. So we should check the length of > >>> msg->msg_name is not less than sizeof(struct sockaddr_ieee802154) > >>> before entering the ieee802154_addr_from_sa. > >>> > >>> Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com> > >> > >> > >> This patch has been applied to the wpan tree and will be > >> part of the next pull request to net. Thanks! > > > > For me this patch is buggy or at least it is questionable how to deal > > with the size of ieee802154_addr_sa here. > > You are right. I completely missed this. Thanks for spotting! > > > There should be a helper to calculate the size which depends on the > > addr_type field. It is not required to send the last 6 bytes if > > addr_type is IEEE802154_ADDR_SHORT. > > Nitpick is that we should check in the beginning of that function. > > Haimin, in ieee802154 we could have two different sizes for > ieee802154_addr_sa depending on the addr_type. We have short and > extended addresses. > > Could you please rework this patch to take this into account as Alex > suggested? > > I reverted your original patch from my tree. > > regards > Stefan Schmidt
diff --git a/net/ieee802154/socket.c b/net/ieee802154/socket.c index 718fb77bb..efbe08590 100644 --- a/net/ieee802154/socket.c +++ b/net/ieee802154/socket.c @@ -655,6 +655,10 @@ static int dgram_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) if (msg->msg_name) { DECLARE_SOCKADDR(struct sockaddr_ieee802154*, daddr, msg->msg_name); + if (msg->msg_namelen < sizeof(*daddr)) { + err = -EINVAL; + goto out_skb; + } ieee802154_addr_from_sa(&dst_addr, &daddr->addr); } else {
There is uninit value bug in dgram_sendmsg function in net/ieee802154/socket.c when the length of valid data pointed by the msg->msg_name isn't verified. This length is specified by msg->msg_namelen. Function ieee802154_addr_from_sa is called by dgram_sendmsg, which use msg->msg_name as struct sockaddr_ieee802154* and read it, that will eventually lead to uninit value read. So we should check the length of msg->msg_name is not less than sizeof(struct sockaddr_ieee802154) before entering the ieee802154_addr_from_sa. Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com> --- net/ieee802154/socket.c | 4 ++++ 1 file changed, 4 insertions(+)