On 7/3/23 12:24, Magnus Karlsson wrote: > On Mon, 3 Jul 2023 at 12:13, Ilya Maximets <i.maximets@xxxxxxx> wrote: >> >> On 7/3/23 12:06, Ilya Maximets wrote: >>> On 7/3/23 11:48, Magnus Karlsson wrote: >>>> On Fri, 30 Jun 2023 at 16:58, Ilya Maximets <i.maximets@xxxxxxx> wrote: >>>>> >>>>> Initial creation of an AF_XDP socket requires CAP_NET_RAW capability. >>>>> A privileged process might create the socket and pass it to a >>>>> non-privileged process for later use. However, that process will be >>>>> able to bind the socket to any network interface. Even though it will >>>>> not be able to receive any traffic without modification of the BPF map, >>>>> the situation is not ideal. >>>>> >>>>> Sockets already have a mechanism that can be used to restrict what >>>>> interface they can be attached to. That is SO_BINDTODEVICE. >>>>> >>>>> To change the binding the process will need CAP_NET_RAW. >>>>> >>>>> Make xsk_bind() honor the SO_BINDTODEVICE in order to allow safer >>>>> workflow when non-privileged process is using AF_XDP. >>>> >>>> Rebinding an AF_XDP socket is not allowed today. Any such attempt will >>>> return an error from bind. So if I understand the purpose of >>>> SO_BINDTODEVICE correctly, you could say that this option is always >>>> set for an AF_XDP socket and it is not possible to toggle it. The only >>>> way to "rebind" an AF_XDP socket is to close it and open a new one. >>>> This was a conscious design decision from day one as it would be very >>>> hard to support this, especially in zero-copy mode. >>> >>> Hi, Magnus. >>> >>> The purpose of this patch is not to allow re-binding. The use case is >>> following: >>> >>> 1. First process creates a bare socket with socket(AF_XDP, ...). >>> 2. First process loads the XSK program to the interface. >>> 3. First process adds the socket fd to a BPF map. >>> 4. First process sends socket fd to a second process. >>> 5. Second process allocates UMEM. >>> 6. Second process binds socket to the interface. >> >> 7. Second process sends/receives the traffic. :) >> >>> >>> The idea is that the first process will call SO_BINDTODEVICE before >>> sending socket fd to a second process, so the second process is limited >>> in to which interface it can bind the socket. >>> >>> Does that make sense? > > Thanks for explaining this to me. Yes, that makes sense and seems > useful. Could you please send a v2 and include the flow (1-7) above in > your commit message? Would be good to add one step with the setsockopt > SO_BINDTODEVICE before step #4 just to be clear. With those changes > please feel free to include my ack: > > Acked-by: Magnus Karlsson <magnus.karlsson@xxxxxxxxx> Thanks! I'll update the commit message with the steps above to make it more clear. I was planning to send a non-RFC version of this patch once the tree is open (in a week). Or are the rules for bpf-next different? > > Thank you! > >>> This workflow allows the second process to have no capabilities >>> as long as it has sufficient RLIMIT_MEMLOCK. >> >> Note that steps 1-7 are working just fine today. i.e. the umem >> registration, bind, ring mapping and traffic send/receive do not >> require any extra capabilities. >> >> We may restrict the bind() call to require CAP_NET_RAW and then >> allow it for sockets that had SO_BINDTODEVICE as an alternative. >> But restriction will break the current uAPI. >> >>> >>> Best regards, Ilya Maximets. >>> >>>> >>>>> Signed-off-by: Ilya Maximets <i.maximets@xxxxxxx> >>>>> --- >>>>> >>>>> Posting as an RFC for now to probably get some feedback. >>>>> Will re-post once the tree is open. >>>>> >>>>> Documentation/networking/af_xdp.rst | 9 +++++++++ >>>>> net/xdp/xsk.c | 6 ++++++ >>>>> 2 files changed, 15 insertions(+) >>>>> >>>>> diff --git a/Documentation/networking/af_xdp.rst b/Documentation/networking/af_xdp.rst >>>>> index 247c6c4127e9..1cc35de336a4 100644 >>>>> --- a/Documentation/networking/af_xdp.rst >>>>> +++ b/Documentation/networking/af_xdp.rst >>>>> @@ -433,6 +433,15 @@ start N bytes into the buffer leaving the first N bytes for the >>>>> application to use. The final option is the flags field, but it will >>>>> be dealt with in separate sections for each UMEM flag. >>>>> >>>>> +SO_BINDTODEVICE setsockopt >>>>> +-------------------------- >>>>> + >>>>> +This is a generic SOL_SOCKET option that can be used to tie AF_XDP >>>>> +socket to a particular network interface. It is useful when a socket >>>>> +is created by a privileged process and passed to a non-privileged one. >>>>> +Once the option is set, kernel will refuse attempts to bind that socket >>>>> +to a different interface. Updating the value requires CAP_NET_RAW. >>>>> + >>>>> XDP_STATISTICS getsockopt >>>>> ------------------------- >>>>> >>>>> diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c >>>>> index 5a8c0dd250af..386ff641db0f 100644 >>>>> --- a/net/xdp/xsk.c >>>>> +++ b/net/xdp/xsk.c >>>>> @@ -886,6 +886,7 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len) >>>>> struct sock *sk = sock->sk; >>>>> struct xdp_sock *xs = xdp_sk(sk); >>>>> struct net_device *dev; >>>>> + int bound_dev_if; >>>>> u32 flags, qid; >>>>> int err = 0; >>>>> >>>>> @@ -899,6 +900,11 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len) >>>>> XDP_USE_NEED_WAKEUP)) >>>>> return -EINVAL; >>>>> >>>>> + bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); >>>>> + >>>>> + if (bound_dev_if && bound_dev_if != sxdp->sxdp_ifindex) >>>>> + return -EINVAL; >>>>> + >>>>> rtnl_lock(); >>>>> mutex_lock(&xs->mutex); >>>>> if (xs->state != XSK_READY) { >>>>> -- >>>>> 2.40.1 >>>>> >>>>> >>> >>