[RFC PATCH 1/3] ipv4: Run a reverse sk_lookup on sendmsg.
In order to check if egress traffic should be allowed through, we run a
reverse socket lookup (i.e. normal socket lookup with the src/dst
addresses and ports reversed) to check if the corresponding ingress
traffic is allowed in. Thus, if there's a sk_lookup reverse call
returns a socket that matches the egress socket, we also let the egress
traffic through - following the principle of, allowing return traffic to
proceed if ingress traffic is allowed in. The reverse lookup is only
performed in case an sk_lookup ebpf program is attached and the source
address and/or port for the return traffic have been modified.
The src address and port can be modified by using ancilliary messages.
Up until now, it was possible to specify a different source address to
sendmsg by providing it in an IP_PKTINFO anciliarry message, but there's
no way to change the source port. This patch also extends the ancilliary
messages supported by sendmsg to support the IP_ORIGDSTADDR ancilliary
message, reusing the same cmsg and struct used in recvmsg - which
already supports specifying a port.
Suggested-by: Jakub Sitnicki
Signed-off-by: Tiago Lam
---
include/net/ip.h | 1 +
net/ipv4/ip_sockglue.c | 11 +++
net/ipv4/udp.c | 33 -
3 files changed, 44 insertions(+), 1 deletion(-)
diff --git a/include/net/ip.h b/include/net/ip.h
index c5606cadb1a5..e5753abd7247 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -75,6 +75,7 @@ static inline unsigned int ip_hdrlen(const struct sk_buff
*skb)
struct ipcm_cookie {
struct sockcm_cookiesockc;
__be32 addr;
+ __be16 port;
int oif;
struct ip_options_rcu *opt;
__u8protocol;
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index cf377377b52d..6e55bd25b5f7 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -297,6 +297,17 @@ int ip_cmsg_send(struct sock *sk, struct msghdr *msg,
struct ipcm_cookie *ipc,
ipc->addr = info->ipi_spec_dst.s_addr;
break;
}
+ case IP_ORIGDSTADDR:
+ {
+ struct sockaddr_in *dst_addr;
+
+ if (cmsg->cmsg_len != CMSG_LEN(sizeof(struct
sockaddr_in)))
+ return -EINVAL;
+ dst_addr = (struct sockaddr_in *)CMSG_DATA(cmsg);
+ ipc->port = dst_addr->sin_port;
+ ipc->addr = dst_addr->sin_addr.s_addr;
+ break;
+ }
case IP_TTL:
if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
return -EINVAL;
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 49c622e743e8..b9dc0a88b0c6 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1060,6 +1060,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg,
size_t len)
DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
struct flowi4 fl4_stack;
struct flowi4 *fl4;
+ __u8 flow_flags = inet_sk_flowi_flags(sk);
int ulen = len;
struct ipcm_cookie ipc;
struct rtable *rt = NULL;
@@ -1179,6 +1180,37 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg,
size_t len)
}
}
+ /* If we're egressing with a different source address and/or port, we
+* perform a reverse socket lookup. The rationale behind this is that
we
+* can allow return UDP traffic that has ingressed through sk_lookup to
+* also egress correctly. In case this the reverse lookup fails.
+*
+* The lookup is performed if either source address and/or port
changed, and
+* neither is "0".
+*/
+ if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
+ !connected &&
+ (ipc.port && ipc.addr) &&
+ (inet->inet_saddr != ipc.addr || inet->inet_sport != ipc.port)) {
+ struct sock *sk_egress;
+
+ bpf_sk_lookup_run_v4(sock_net(sk), IPPROTO_UDP,
+daddr, dport, ipc.addr, ntohs(ipc.port),
1, &sk_egress);
+ if (IS_ERR_OR_NULL(sk_egress) ||
+ atomic64_read(&sk_egress->sk_cookie) !=
atomic64_read(&sk->sk_cookie)) {
+ net_info_ratelimited("No reverse socket lookup match
for local addr %pI4:%d remote addr %pI4:%d\n",
+&ipc.addr, ntohs(ipc.port),
&daddr, ntohs(dport));
+ } else {
+ /* Override the source port to use with the one we got
in cmsg,
+* and tell routing to let us use a non-local address.
Otherwise
+
[RFC PATCH 0/3] Allow sk_lookup UDP return traffic to egress.
Currently, sk_lookup allows an ebpf program to run on the ingress socket lookup path, and accept traffic not only on a range of addresses, but also on a range of ports. At Cloudflare we use sk_lookup for two main cases: 1. Sharing a single port between multiple services - i.e. two services (or more) use disjoint IP ranges but share the same port; 2. Receiving traffic on all ports - i.e. a service which accepts traffic on specific IP ranges but any port [1]. However, one main challenge we face while using sk_lookup for these use cases is how to source return UDP traffic: - On point 1. above, sometimes this range of addresses are not local (i.e. there's no local routes for these in the server), which means we need IP_TRANSPARENT set to be able to egress traffic from addresses we've received traffic on (or simply IP_FREEBIND in the case of IPv6); - And on point 2. above, allowing traffic to a range of ports means a service could get traffic on multiple ports, but currently there's no way to set the source UDP port egress traffic should be sourced from - it's possible to receive the original destination port using the IP_ORIGDSTADDR ancilliary message in recvmsg, but not set it in sendmsg. Both of these limitations can be worked around, but in a sub-optimal way. Using IP_TRANSPARENT, for instance, requires special privileges. And while one could use UDP connected sockets to send return traffic, creating a connected socket for each different address a UDP traffic is received on does have performance implications. Given sk_lookup allows services to accept traffic on a range of addresses or ports, it seems sensible to also allow return traffic to proceed through as well, without needing extra configurations / set ups. This patch set allows to do exactly this by performing a reverse socket lookup on the egress path - where it looks to see if the egress socket matches a socket in the attached sk_lookup ebpf program for the traffic that's being sent. If it does, traffic is allowed to proceed. The downsides to this is that this runs on the egress hot path, although this work tries to minimise its impact by only performing the reverse socket lookup when necessary. Further performance measurements are to be taken, but we're reaching out early for feedback to see what the technical concerns are and if we can address them. [1] https://blog.cloudflare.com/how-we-built-spectrum/ Suggested-by: Jakub Sitnicki Signed-off-by: Tiago Lam --- Tiago Lam (3): ipv4: Run a reverse sk_lookup on sendmsg. ipv6: Run a reverse sk_lookup on sendmsg. bpf: Add sk_lookup test to use ORIGDSTADDR cmsg. include/net/ip.h | 1 + net/ipv4/ip_sockglue.c | 11 net/ipv4/udp.c | 33 +- net/ipv6/datagram.c| 76 ++ net/ipv6/udp.c | 8 ++- tools/testing/selftests/bpf/prog_tests/sk_lookup.c | 70 +--- 6 files changed, 174 insertions(+), 25 deletions(-) --- base-commit: da3ea35007d0af457a0afc87e84fddaebc4e0b63 change-id: 20240909-reverse-sk-lookup-f7bf36292bc4 Best regards, -- Tiago Lam
[RFC PATCH 3/3] bpf: Add sk_lookup test to use ORIGDSTADDR cmsg.
This patch reuses the framework already in place for sk_lookup, allowing
it now to send a reply from the server fd directly, instead of having to
create a socket bound to the original destination address and reply from
there. It does this by passing the source address and port from where to
reply from in a IP_ORIGDSTADDR ancilliary message passed in sendmsg.
Signed-off-by: Tiago Lam
---
tools/testing/selftests/bpf/prog_tests/sk_lookup.c | 70 +++---
1 file changed, 48 insertions(+), 22 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
index ae87c00867ba..b99aff2e3976 100644
--- a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
+++ b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
@@ -75,6 +75,7 @@ struct test {
struct inet_addr listen_at;
enum server accept_on;
bool reuseport_has_conns; /* Add a connected socket to reuseport group
*/
+ bool dont_bind_reply; /* Don't bind, send direct from server fd. */
};
struct cb_opts {
@@ -363,7 +364,7 @@ static void v4_to_v6(struct sockaddr_storage *ss)
memset(&v6->sin6_addr.s6_addr[0], 0, 10);
}
-static int udp_recv_send(int server_fd)
+static int udp_recv_send(int server_fd, bool dont_bind_reply)
{
char cmsg_buf[CMSG_SPACE(sizeof(struct sockaddr_storage))];
struct sockaddr_storage _src_addr = { 0 };
@@ -415,26 +416,41 @@ static int udp_recv_send(int server_fd)
v4_to_v6(dst_addr);
}
- /* Reply from original destination address. */
- fd = start_server_addr(SOCK_DGRAM, dst_addr, sizeof(*dst_addr), NULL);
- if (!ASSERT_OK_FD(fd, "start_server_addr")) {
- log_err("failed to create tx socket");
- return -1;
- }
+ if (dont_bind_reply) {
+ /* Reply directly from server fd, specifying the source address
and/or
+* port in struct msghdr.
+*/
+ n = sendmsg(server_fd, &msg, 0);
+ if (CHECK(n <= 0, "sendmsg", "failed\n")) {
+ log_err("failed to send echo reply");
+ return -1;
+ }
+ } else {
+ /* Reply from original destination address. */
+ fd = socket(dst_addr->ss_family, SOCK_DGRAM, 0);
+ if (CHECK(fd < 0, "socket", "failed\n")) {
+ log_err("failed to create tx socket");
+ return -1;
+ }
- msg.msg_control = NULL;
- msg.msg_controllen = 0;
- n = sendmsg(fd, &msg, 0);
- if (CHECK(n <= 0, "sendmsg", "failed\n")) {
- log_err("failed to send echo reply");
- ret = -1;
- goto out;
- }
+ ret = bind(fd, (struct sockaddr *)dst_addr, sizeof(*dst_addr));
+ if (CHECK(ret, "bind", "failed\n")) {
+ log_err("failed to bind tx socket");
+ close(fd);
+ return ret;
+ }
- ret = 0;
-out:
- close(fd);
- return ret;
+ msg.msg_control = NULL;
+ msg.msg_controllen = 0;
+ n = sendmsg(fd, &msg, 0);
+ if (CHECK(n <= 0, "sendmsg", "failed\n")) {
+ log_err("failed to send echo reply");
+ close(fd);
+ return -1;
+ }
+
+ close(fd);
+ }
}
static int tcp_echo_test(int client_fd, int server_fd)
@@ -454,14 +470,14 @@ static int tcp_echo_test(int client_fd, int server_fd)
return 0;
}
-static int udp_echo_test(int client_fd, int server_fd)
+static int udp_echo_test(int client_fd, int server_fd, bool dont_bind_reply)
{
int err;
err = send_byte(client_fd);
if (err)
return -1;
- err = udp_recv_send(server_fd);
+ err = udp_recv_send(server_fd, dont_bind_reply);
if (err)
return -1;
err = recv_byte(client_fd);
@@ -653,7 +669,7 @@ static void run_lookup_prog(const struct test *t)
if (t->sotype == SOCK_STREAM)
tcp_echo_test(client_fd, server_fds[t->accept_on]);
else
- udp_echo_test(client_fd, server_fds[t->accept_on]);
+ udp_echo_test(client_fd, server_fds[t->accept_on],
t->dont_bind_reply);
close(client_fd);
close:
@@ -775,6 +791,16 @@ static void test_redirect_lookup(struct test_sk_lookup
*skel)
.listen_at = { INT_IP4, INT_PORT },
.accept_on = SERVER_B,
},
+ {
+ .desc
[RFC PATCH 2/3] ipv6: Run a reverse sk_lookup on sendmsg.
This follows the same rationale provided for the ipv4 counterpart, where
it now runs a reverse socket lookup when source addresses and/or ports
are changed, on sendmsg, to check whether egress traffic should be
allowed to go through or not.
As with ipv4, the ipv6 sendmsg path is also extended here to support the
IPV6_ORIGDSTADDR ancilliary message to be able to specify a source
address/port.
Suggested-by: Jakub Sitnicki
Signed-off-by: Tiago Lam
---
net/ipv6/datagram.c | 76 +
net/ipv6/udp.c | 8 --
2 files changed, 82 insertions(+), 2 deletions(-)
diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
index fff78496803d..4214dda1c320 100644
--- a/net/ipv6/datagram.c
+++ b/net/ipv6/datagram.c
@@ -756,6 +756,27 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct msghdr
*msg,
}
EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
+static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock *sk,
+struct in6_addr *saddr, __be16 sport)
+{
+ if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
+ (saddr && sport) &&
+ (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
inet_sk(sk)->inet_sport != sport)) {
+ struct sock *sk_egress;
+
+ bpf_sk_lookup_run_v6(sock_net(sk), IPPROTO_UDP, &fl6->daddr,
fl6->fl6_dport,
+saddr, ntohs(sport), 0, &sk_egress);
+ if (!IS_ERR_OR_NULL(sk_egress) &&
+ atomic64_read(&sk_egress->sk_cookie) ==
atomic64_read(&sk->sk_cookie))
+ return true;
+
+ net_info_ratelimited("No reverse socket lookup match for local
addr %pI6:%d remote addr %pI6:%d\n",
+&saddr, ntohs(sport), &fl6->daddr,
ntohs(fl6->fl6_dport));
+ }
+
+ return false;
+}
+
int ip6_datagram_send_ctl(struct net *net, struct sock *sk,
struct msghdr *msg, struct flowi6 *fl6,
struct ipcm6_cookie *ipc6)
@@ -844,7 +865,62 @@ int ip6_datagram_send_ctl(struct net *net, struct sock *sk,
break;
}
+ case IPV6_ORIGDSTADDR:
+ {
+ struct sockaddr_in6 *sockaddr_in;
+ struct net_device *dev = NULL;
+
+ if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct
sockaddr_in6))) {
+ err = -EINVAL;
+ goto exit_f;
+ }
+
+ sockaddr_in = (struct sockaddr_in6 *)CMSG_DATA(cmsg);
+
+ addr_type = __ipv6_addr_type(&sockaddr_in->sin6_addr);
+
+ if (addr_type & IPV6_ADDR_LINKLOCAL)
+ return -EINVAL;
+
+ /* If we're egressing with a different source address
and/or port, we
+* perform a reverse socket lookup. The rationale
behind this is that we
+* can allow return UDP traffic that has ingressed
through sk_lookup to
+* also egress correctly. In case the reverse lookup
fails, we
+* continue with the normal path.
+*
+* The lookup is performed if either source address
and/or port changed, and
+* neither is "0".
+*/
+ if (reverse_sk_lookup(fl6, sk, &sockaddr_in->sin6_addr,
+ sockaddr_in->sin6_port)) {
+ /* Override the source port and address to use
with the one we
+* got in cmsg and bail early.
+*/
+ fl6->saddr = sockaddr_in->sin6_addr;
+ fl6->fl6_sport = sockaddr_in->sin6_port;
+ break;
+ }
+ if (addr_type != IPV6_ADDR_ANY) {
+ int strict = __ipv6_addr_src_scope(addr_type)
<= IPV6_ADDR_SCOPE_LINKLOCAL;
+
+ if (!ipv6_can_nonlocal_bind(net, inet_sk(sk)) &&
+ !ipv6_chk_addr_and_flags(net,
+
&sockaddr_in->sin6_addr,
+dev, !strict, 0,
+IFA_F_TENTATIVE) &&
+ !ipv6_chk_acast_addr_src(net, dev,
+
&sockaddr_i
Re: [RFC PATCH 2/3] ipv6: Run a reverse sk_lookup on sendmsg.
On Tue, Sep 24, 2024 at 04:58:19PM -0700, Martin KaFai Lau wrote:
> On 9/17/24 6:15 PM, Tiago Lam wrote:
> > On Fri, Sep 13, 2024 at 11:24:09AM -0700, Martin KaFai Lau wrote:
> > > On 9/13/24 2:39 AM, Tiago Lam wrote:
> > > > This follows the same rationale provided for the ipv4 counterpart, where
> > > > it now runs a reverse socket lookup when source addresses and/or ports
> > > > are changed, on sendmsg, to check whether egress traffic should be
> > > > allowed to go through or not.
> > > >
> > > > As with ipv4, the ipv6 sendmsg path is also extended here to support the
> > > > IPV6_ORIGDSTADDR ancilliary message to be able to specify a source
> > > > address/port.
> > > >
> > > > Suggested-by: Jakub Sitnicki
> > > > Signed-off-by: Tiago Lam
> > > > ---
> > > >net/ipv6/datagram.c | 76
> > > > +
> > > >net/ipv6/udp.c | 8 --
> > > >2 files changed, 82 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
> > > > index fff78496803d..4214dda1c320 100644
> > > > --- a/net/ipv6/datagram.c
> > > > +++ b/net/ipv6/datagram.c
> > > > @@ -756,6 +756,27 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct
> > > > msghdr *msg,
> > > >}
> > > >EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
> > > > +static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock
> > > > *sk,
> > > > +struct in6_addr *saddr, __be16
> > > > sport)
> > > > +{
> > > > + if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
> > > > + (saddr && sport) &&
> > > > + (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
> > > > inet_sk(sk)->inet_sport != sport)) {
> > > > + struct sock *sk_egress;
> > > > +
> > > > + bpf_sk_lookup_run_v6(sock_net(sk), IPPROTO_UDP,
> > > > &fl6->daddr, fl6->fl6_dport,
> > > > +saddr, ntohs(sport), 0,
> > > > &sk_egress);
> > >
> > > iirc, in the ingress path, the sk could also be selected by a tc bpf prog
> > > doing bpf_sk_assign. Then this re-run on sk_lookup may give an incorrect
> > > result?
> > >
> >
> > If it does give the incorrect result, we still fallback to the normal
> > egress path.
> >
> > > In general, is it necessary to rerun any bpf prog if the user space has
> > > specified the IP[v6]_ORIGDSTADDR.
> > >
> >
> > More generally, wouldn't that also be the case if someone calls
> > bpf_sk_assign() in both TC and sk_lookup on ingress? It can lead to some
> > interference between the two.
> >
> > It seems like the interesting cases are:
> > 1. Calling bpf_sk_assign() on both TC and sk_lookup ingress: if this
> > happens sk_lookup on egress should match the correct socket when doing
> > the reverse lookup;
> > 2. Calling bpf_sk_assign() only on ingress TC: in this case it will
> > depend if an sk_lookup program is attached or not:
> >a. If not, there's no reverse lookup on egress either;
> >b. But if yes, although the reverse sk_lookup here won't match the
> >initial socket assigned at ingress TC, the packets will still fallback
> >to the normal egress path;
> >
> > You're right in that case 2b above will continue with the same
> > restrictions as before.
>
> imo, all these cases you described above is a good signal that neither the
> TC nor the BPF_PROG_TYPE_SK_LOOKUP program type is the right bpf prog to run
> here _if_ a bpf prog was indeed useful here.
>
> I only followed some of the other discussion in v1 and v2. For now, I still
> don't see running a bpf prog is useful here to process the
> IP[V6]_ORIGDSTADDR. Jakub Sitnicki and I had discussed a similar point
> during the LPC.
>
> If a bpf prog was indeed needed to process a cmsg, this should work closer
> to what Jakub Sitnicki had proposed for getting the meta data during LPC
> (but I believe the verdict there is also that a bpf prog is not needed). It
> should be a bpf prog that can work in a more generic way to process any BPF
> specific cmsg and can do other operations in the future using kfunc (e.g.
> route lookup or something). Saying yes/no to a particular local IP and port
> could be one of things that the bpf prog can do when processing the cmsg.
Thanks for the feeback here, Martin. And apologies for the delay in
respoding to this.
I think you do have a point, and after syncing up some more with Jakub
about your discussion during the LPC, the argument that applications can
already bind to a specific address + port to send their traffic from
makes sense to me. However, I think we could introduce a new cmsg in
sendmsg to allow apps to set the source port to egress from, extending
what they can already do with IP_PKTINFO, i.e. setting the source IP.
We'd need to take care with priviledged and reserved ports, but this
would avoid applications having to do an extra bind. Do you have any
thoughts on this?
Tiago.
Re: [RFC PATCH 3/3] bpf: Add sk_lookup test to use ORIGDSTADDR cmsg.
On Fri, Sep 13, 2024 at 08:10:24PM +0800, Philo Lu wrote:
> Hi Tiago,
>
> On 2024/9/13 17:39, Tiago Lam wrote:
> > This patch reuses the framework already in place for sk_lookup, allowing
> > it now to send a reply from the server fd directly, instead of having to
> > create a socket bound to the original destination address and reply from
> > there. It does this by passing the source address and port from where to
> > reply from in a IP_ORIGDSTADDR ancilliary message passed in sendmsg.
> >
> > Signed-off-by: Tiago Lam
> > ---
> > tools/testing/selftests/bpf/prog_tests/sk_lookup.c | 70
> > +++---
> > 1 file changed, 48 insertions(+), 22 deletions(-)
> >
> > diff --git a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
> > b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
> > index ae87c00867ba..b99aff2e3976 100644
> > --- a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
> > +++ b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
> > @@ -75,6 +75,7 @@ struct test {
> > struct inet_addr listen_at;
> > enum server accept_on;
> > bool reuseport_has_conns; /* Add a connected socket to reuseport group
> > */
> > + bool dont_bind_reply; /* Don't bind, send direct from server fd. */
> > };
> > struct cb_opts {
> > @@ -363,7 +364,7 @@ static void v4_to_v6(struct sockaddr_storage *ss)
> > memset(&v6->sin6_addr.s6_addr[0], 0, 10);
> > }
> > -static int udp_recv_send(int server_fd)
> > +static int udp_recv_send(int server_fd, bool dont_bind_reply)
> > {
> > char cmsg_buf[CMSG_SPACE(sizeof(struct sockaddr_storage))];
> > struct sockaddr_storage _src_addr = { 0 };
> > @@ -415,26 +416,41 @@ static int udp_recv_send(int server_fd)
> > v4_to_v6(dst_addr);
> > }
> > - /* Reply from original destination address. */
> > - fd = start_server_addr(SOCK_DGRAM, dst_addr, sizeof(*dst_addr), NULL);
> > - if (!ASSERT_OK_FD(fd, "start_server_addr")) {
> > - log_err("failed to create tx socket");
> > - return -1;
> > - }
> > + if (dont_bind_reply) {
> > + /* Reply directly from server fd, specifying the source address
> > and/or
> > +* port in struct msghdr.
> > +*/
> > + n = sendmsg(server_fd, &msg, 0);
> > + if (CHECK(n <= 0, "sendmsg", "failed\n")) {
> > + log_err("failed to send echo reply");
> > + return -1;
> > + }
> > + } else {
> > + /* Reply from original destination address. */
> > + fd = socket(dst_addr->ss_family, SOCK_DGRAM, 0);
> > + if (CHECK(fd < 0, "socket", "failed\n")) {
> > + log_err("failed to create tx socket");
> > + return -1;
> > + }
> Just curious, why not use start_server_addr() here like before?
>
Thanks, you're right. Initially I rebased this on commit
e3d332aaf898ed755b29c8cdf59be2cfba1cac4b (v6.6.31), which didn't have
start_server_addr(), and used bind() instead. I must have messed up the
rebased.
I've fixed this and included your other suggestion in the patch below
and will fold it into the next revision.
diff --git a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
index b99aff2e3976..0628a67681c5 100644
--- a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
+++ b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
@@ -374,7 +374,7 @@ static int udp_recv_send(int server_fd, bool
dont_bind_reply)
struct iovec iov = { 0 };
struct cmsghdr *cm;
char buf[1];
- int ret, fd;
+ int fd;
ssize_t n;
iov.iov_base = buf;
@@ -427,19 +427,12 @@ static int udp_recv_send(int server_fd, bool
dont_bind_reply)
}
} else {
/* Reply from original destination address. */
- fd = socket(dst_addr->ss_family, SOCK_DGRAM, 0);
- if (CHECK(fd < 0, "socket", "failed\n")) {
+ fd = start_server_addr(SOCK_DGRAM, dst_addr, sizeof(*dst_addr),
NULL);
+ if (!ASSERT_OK_FD(fd, "start_server_addr")) {
log_err("failed to create tx socket");
return -1;
}
- ret = bind(fd, (struct sockaddr *)dst_addr, sizeof(*dst_addr));
- if (CHECK(ret, "bind", "failed\n")) {
- log_err("failed to bind tx socket");
- close(fd);
- return ret;
- }
-
msg.msg_control = NULL;
msg.msg_controllen = 0;
n = sendmsg(fd, &msg, 0);
@@ -451,6 +444,8 @@ static int udp_recv_send(int server_fd, bool
dont_bind_reply)
close(fd);
}
+
+ return 0;
}
static int tcp_echo_test(int client_fd, int server_fd)
Re: [RFC PATCH 2/3] ipv6: Run a reverse sk_lookup on sendmsg.
On Sat, Sep 14, 2024 at 01:40:25PM +0200, Eric Dumazet wrote:
> On Fri, Sep 13, 2024 at 11:39 AM Tiago Lam wrote:
> >
> > This follows the same rationale provided for the ipv4 counterpart, where
> > it now runs a reverse socket lookup when source addresses and/or ports
> > are changed, on sendmsg, to check whether egress traffic should be
> > allowed to go through or not.
> >
> > As with ipv4, the ipv6 sendmsg path is also extended here to support the
> > IPV6_ORIGDSTADDR ancilliary message to be able to specify a source
> > address/port.
> >
> > Suggested-by: Jakub Sitnicki
> > Signed-off-by: Tiago Lam
> > ---
> > net/ipv6/datagram.c | 76
> > +
> > net/ipv6/udp.c | 8 --
> > 2 files changed, 82 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
> > index fff78496803d..4214dda1c320 100644
> > --- a/net/ipv6/datagram.c
> > +++ b/net/ipv6/datagram.c
> > @@ -756,6 +756,27 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct
> > msghdr *msg,
> > }
> > EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
> >
> > +static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock *sk,
> > +struct in6_addr *saddr, __be16 sport)
> > +{
> > + if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
> > + (saddr && sport) &&
> > + (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
> > inet_sk(sk)->inet_sport != sport)) {
> > + struct sock *sk_egress;
> > +
> > + bpf_sk_lookup_run_v6(sock_net(sk), IPPROTO_UDP,
> > &fl6->daddr, fl6->fl6_dport,
> > +saddr, ntohs(sport), 0, &sk_egress);
> > + if (!IS_ERR_OR_NULL(sk_egress) &&
> > + atomic64_read(&sk_egress->sk_cookie) ==
> > atomic64_read(&sk->sk_cookie))
>
> I do not understand this.
>
> 1) sk_cookie is not always initialized. It is done on demand, when/if
> __sock_gen_cookie() was called.
>
> 2) if sk1 and sk2 share the same sk_cookie, then sk1 == sk2 ???
>
> So why not simply testing sk_egress == sk ?
>
Oh, yes, you're right. I'll include this in my next revision, thanks.
Re: [RFC PATCH 2/3] ipv6: Run a reverse sk_lookup on sendmsg.
On Sat, Sep 14, 2024 at 09:59:50AM +0100, Simon Horman wrote:
> On Fri, Sep 13, 2024 at 10:39:20AM +0100, Tiago Lam wrote:
> > This follows the same rationale provided for the ipv4 counterpart, where
> > it now runs a reverse socket lookup when source addresses and/or ports
> > are changed, on sendmsg, to check whether egress traffic should be
> > allowed to go through or not.
> >
> > As with ipv4, the ipv6 sendmsg path is also extended here to support the
> > IPV6_ORIGDSTADDR ancilliary message to be able to specify a source
>
> Hi Tiago Lam,
>
> Some minor nits from my side.
>
> ancilliary -> ancillary
>
> Likewise in patch 3/3.
> Flagged by checkpatch.pl --codespell
>
> > address/port.
> >
> > Suggested-by: Jakub Sitnicki
> > Signed-off-by: Tiago Lam
> > ---
> > net/ipv6/datagram.c | 76
> > +
> > net/ipv6/udp.c | 8 --
> > 2 files changed, 82 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
> > index fff78496803d..4214dda1c320 100644
> > --- a/net/ipv6/datagram.c
> > +++ b/net/ipv6/datagram.c
> > @@ -756,6 +756,27 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct
> > msghdr *msg,
> > }
> > EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
> >
> > +static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock *sk,
> > +struct in6_addr *saddr, __be16 sport)
> > +{
> > + if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
> > + (saddr && sport) &&
> > + (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
> > inet_sk(sk)->inet_sport != sport)) {
>
> Please consider, where it can trivially be achieved, limiting Networking
> code to 80 columns wide.
>
> Checkpatch can be run with a flag to check for this.
>
Thanks for the hints here, I've addressed these and will include the
changes into the next revision. I use b4 which takes care of some of
this checks, but I'll make sure I change my settings to use
`--max-line-length=80` as well.
Tiago.
Re: [RFC PATCH 2/3] ipv6: Run a reverse sk_lookup on sendmsg.
On Fri, Sep 13, 2024 at 11:24:09AM -0700, Martin KaFai Lau wrote:
> On 9/13/24 2:39 AM, Tiago Lam wrote:
> > This follows the same rationale provided for the ipv4 counterpart, where
> > it now runs a reverse socket lookup when source addresses and/or ports
> > are changed, on sendmsg, to check whether egress traffic should be
> > allowed to go through or not.
> >
> > As with ipv4, the ipv6 sendmsg path is also extended here to support the
> > IPV6_ORIGDSTADDR ancilliary message to be able to specify a source
> > address/port.
> >
> > Suggested-by: Jakub Sitnicki
> > Signed-off-by: Tiago Lam
> > ---
> > net/ipv6/datagram.c | 76
> > +
> > net/ipv6/udp.c | 8 --
> > 2 files changed, 82 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
> > index fff78496803d..4214dda1c320 100644
> > --- a/net/ipv6/datagram.c
> > +++ b/net/ipv6/datagram.c
> > @@ -756,6 +756,27 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct
> > msghdr *msg,
> > }
> > EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
> > +static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock *sk,
> > +struct in6_addr *saddr, __be16 sport)
> > +{
> > + if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
> > + (saddr && sport) &&
> > + (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
> > inet_sk(sk)->inet_sport != sport)) {
> > + struct sock *sk_egress;
> > +
> > + bpf_sk_lookup_run_v6(sock_net(sk), IPPROTO_UDP, &fl6->daddr,
> > fl6->fl6_dport,
> > +saddr, ntohs(sport), 0, &sk_egress);
>
> iirc, in the ingress path, the sk could also be selected by a tc bpf prog
> doing bpf_sk_assign. Then this re-run on sk_lookup may give an incorrect
> result?
>
If it does give the incorrect result, we still fallback to the normal
egress path.
> In general, is it necessary to rerun any bpf prog if the user space has
> specified the IP[v6]_ORIGDSTADDR.
>
More generally, wouldn't that also be the case if someone calls
bpf_sk_assign() in both TC and sk_lookup on ingress? It can lead to some
interference between the two.
It seems like the interesting cases are:
1. Calling bpf_sk_assign() on both TC and sk_lookup ingress: if this
happens sk_lookup on egress should match the correct socket when doing
the reverse lookup;
2. Calling bpf_sk_assign() only on ingress TC: in this case it will
depend if an sk_lookup program is attached or not:
a. If not, there's no reverse lookup on egress either;
b. But if yes, although the reverse sk_lookup here won't match the
initial socket assigned at ingress TC, the packets will still fallback
to the normal egress path;
You're right in that case 2b above will continue with the same
restrictions as before.
Tiago.
Re: [RFC PATCH 1/3] ipv4: Run a reverse sk_lookup on sendmsg.
On Wed, Sep 18, 2024 at 08:45:23AM -0400, Willem de Bruijn wrote: > Tiago Lam wrote: > > In order to check if egress traffic should be allowed through, we run a > > reverse socket lookup (i.e. normal socket lookup with the src/dst > > addresses and ports reversed) to check if the corresponding ingress > > traffic is allowed in. > > The subject and this description makes it sound that the change always > runs a reverse sk_lookup on sendmsg. > > It also focuses on the mechanism, rather than the purpose. > > The feature here adds IP_ORIGDSTADDR as a way to respond from a > user configured address. With the sk_lookup limited to this new > special case, as a safety to allow it. > > If I read this correctly, I suggest rewording the cover letter and > commit to make this intent and behavior more explicit. > I think that makes sense, given this is really about two things: 1. Allowing users to use IP_ORIGDSTADDR to set src address and/or port on sendmsg(); 2. When they do, allow for that return traffic to exit without any extra configuration, thus limiting how users can take advantage of this new functionality. I've made a few changes which hopefully makes that clearer in v2, which I'm sending shortly. Thanks for these suggestions! Tiago.
[RFC PATCH v2 0/3] Allow sk_lookup UDP return traffic to egress when setting src port/address.
Currently, sk_lookup allows an ebpf program to run on the ingress socket lookup path, and accept traffic not only on a range of addresses, but also on a range of ports. At Cloudflare we use sk_lookup for two main cases: 1. Sharing a single port between multiple services - i.e. two services (or more) use disjoint IP ranges but share the same port; 2. Receiving traffic on all ports - i.e. a service which accepts traffic on specific IP ranges but any port [1]. However, one main challenge we face while using sk_lookup for these use cases is how to source return UDP traffic: - On point 1. above, sometimes this range of addresses are not local (i.e. there's no local routes for these in the server), which means we need IP_TRANSPARENT set to be able to egress traffic from addresses we've received traffic on (or simply IP_FREEBIND in the case of IPv6); - And on point 2. above, allowing traffic to a range of ports means a service could get traffic on multiple ports, but currently there's no way to set the source UDP port egress traffic should be sourced from - it's possible to receive the original destination port using the IP_ORIGDSTADDR ancilliary message in recvmsg, but not set it in sendmsg. Both of these limitations can be worked around, but in a sub-optimal way. Using IP_TRANSPARENT, for instance, requires special privileges. And while one could use UDP connected sockets to send return traffic, creating a connected socket for each different address a UDP traffic is received on does have performance implications. Given sk_lookup allows services to accept traffic on a range of addresses or ports, it seems sensible to also allow return traffic to proceed through as well, without needing extra configurations / set ups. This patch sets out to fix both of this issues by: 1. Allowing users to set the src address/port egress traffic should be sent from, when calling sendmsg(); 2. Validating that this egress traffic comes from a socket that matches an ingress socket in sk_lookup. - If it does, traffic is allowed to proceed; - Otherwise it falls back to the regular egress path. The downsides to this is that this runs on the egress hot path, although this work tries to minimise its impact by only performing the reverse socket lookup when necessary (i.e. only when the src address/port are modified). Further performance measurements are to be taken, but we're reaching out early for feedback to see what the technical concerns are and if we can address them. [1] https://blog.cloudflare.com/how-we-built-spectrum/ Suggested-by: Jakub Sitnicki Signed-off-by: Tiago Lam --- Changes in v2: - Amended commit messages and cover letter to make the intent and implementation clearer (Willem de Bruijn); - Fixed socket comparison by not using socket cookies and comparing them directly (Eric Dumazet); - Fixed misspellings and checkpatch.pl warnings on line lengths (Simon Horman); - Fixed usage of start_server_addr() and gcc compilation (Philo Lu); - Link to v1: https://lore.kernel.org/r/20240913-reverse-sk-lookup-v1-0-e721ea003...@cloudflare.com --- Tiago Lam (3): ipv4: Support setting src port in sendmsg(). ipv6: Support setting src port in sendmsg(). bpf: Add sk_lookup test to use ORIGDSTADDR cmsg. include/net/ip.h | 1 + net/ipv4/ip_sockglue.c | 11 +++ net/ipv4/udp.c | 35 +- net/ipv6/datagram.c| 79 ++ net/ipv6/udp.c | 8 ++- tools/testing/selftests/bpf/prog_tests/sk_lookup.c | 67 -- 6 files changed, 176 insertions(+), 25 deletions(-) --- base-commit: 6562a89739bbefddb5495c09aaab67c1c3756f36 change-id: 20240909-reverse-sk-lookup-f7bf36292bc4 Best regards, -- Tiago Lam
[RFC PATCH v2 1/3] ipv4: Support setting src port in sendmsg().
sendmsg() doesn't currently allow users to set the src port from which
egress traffic should be sent from. This is possible if a user wants to
configure the src address from which egress traffic should be sent from
- with the IP_PKTINFO ancillary message, a user is currently able to
specify a source address to egress from when calling sendmsg().
However, this still requires the user to set the IP_TRANSPARENT flag
using setsockopt(), which happens to require special privileges in the
case of IPv4.
To support users setting the src port for egress traffic when using
sendmsg(), this patch extends the ancillary messages supported by
sendmsg() to support the IP_ORIGDSTADDR ancillary message, reusing the
same cmsg and struct used in recvmsg() - which already supports
specifying a port.
Additionally, to avoid having to have special configurations, such as
IP_TRANSPARENT, this patch allows egress traffic that's been configured
using (the newly added) IP_ORIGDSTADDR to proceed if there's an ingress
socket lookup (sk_lookup) that matches that traffic - by performing a
reserve sk_lookup. Thus, if the sk_lookup reverse call returns a socket
that matches the egress socket, we also let the egress traffic through -
following the principle of, allowing return traffic to proceed if
ingress traffic is allowed in. In case no match is found in the reverse
sk_lookup, traffic falls back to the regular egress path.
This reverse lookup is only performed in case an sk_lookup ebpf program
is attached and the source address and/or port for the return traffic
have been modified using the (newly added) IP_ORIGDSTADDR in sendmsg.
Suggested-by: Jakub Sitnicki
Signed-off-by: Tiago Lam
---
include/net/ip.h | 1 +
net/ipv4/ip_sockglue.c | 11 +++
net/ipv4/udp.c | 35 ++-
3 files changed, 46 insertions(+), 1 deletion(-)
diff --git a/include/net/ip.h b/include/net/ip.h
index c5606cadb1a5..e5753abd7247 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -75,6 +75,7 @@ static inline unsigned int ip_hdrlen(const struct sk_buff
*skb)
struct ipcm_cookie {
struct sockcm_cookiesockc;
__be32 addr;
+ __be16 port;
int oif;
struct ip_options_rcu *opt;
__u8protocol;
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index cf377377b52d..6e55bd25b5f7 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -297,6 +297,17 @@ int ip_cmsg_send(struct sock *sk, struct msghdr *msg,
struct ipcm_cookie *ipc,
ipc->addr = info->ipi_spec_dst.s_addr;
break;
}
+ case IP_ORIGDSTADDR:
+ {
+ struct sockaddr_in *dst_addr;
+
+ if (cmsg->cmsg_len != CMSG_LEN(sizeof(struct
sockaddr_in)))
+ return -EINVAL;
+ dst_addr = (struct sockaddr_in *)CMSG_DATA(cmsg);
+ ipc->port = dst_addr->sin_port;
+ ipc->addr = dst_addr->sin_addr.s_addr;
+ break;
+ }
case IP_TTL:
if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
return -EINVAL;
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 49c622e743e8..208cee40c0ec 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1060,6 +1060,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg,
size_t len)
DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
struct flowi4 fl4_stack;
struct flowi4 *fl4;
+ __u8 flow_flags = inet_sk_flowi_flags(sk);
int ulen = len;
struct ipcm_cookie ipc;
struct rtable *rt = NULL;
@@ -1179,6 +1180,39 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg,
size_t len)
}
}
+ /* If we're egressing with a different source address and/or port, we
+* perform a reverse socket lookup. The rationale behind this is that
+* we can allow return UDP traffic that has ingressed through sk_lookup
+* to also egress correctly. In case this the reverse lookup fails.
+*
+* The lookup is performed if either source address and/or port
+* changed, and neither is "0".
+*/
+ if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
+ !connected &&
+ (ipc.port && ipc.addr) &&
+ (inet->inet_saddr != ipc.addr || inet->inet_sport != ipc.port)) {
+ struct sock *sk_egress;
+
+ bpf_sk_lookup_run_v4(sock_net(sk), IPPROTO_UDP,
+daddr, dport, ipc.addr, ntohs(ipc.port),
+1, &sk_egress);
+ if (IS_ERR_OR_NULL(sk_egr
[RFC PATCH v2 3/3] bpf: Add sk_lookup test to use ORIGDSTADDR cmsg.
This patch reuses the framework already in place for sk_lookup, allowing
it now to send a reply from the server fd directly, instead of having to
create a socket bound to the original destination address and reply from
there. It does this by passing the source address and port from where to
reply from in a IP_ORIGDSTADDR ancillary message passed in sendmsg.
Signed-off-by: Tiago Lam
---
tools/testing/selftests/bpf/prog_tests/sk_lookup.c | 67 +++---
1 file changed, 45 insertions(+), 22 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
index ae87c00867ba..df780624c16c 100644
--- a/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
+++ b/tools/testing/selftests/bpf/prog_tests/sk_lookup.c
@@ -75,6 +75,7 @@ struct test {
struct inet_addr listen_at;
enum server accept_on;
bool reuseport_has_conns; /* Add a connected socket to reuseport group
*/
+ bool dont_bind_reply; /* Don't bind, send direct from server fd. */
};
struct cb_opts {
@@ -363,7 +364,7 @@ static void v4_to_v6(struct sockaddr_storage *ss)
memset(&v6->sin6_addr.s6_addr[0], 0, 10);
}
-static int udp_recv_send(int server_fd)
+static int udp_recv_send(int server_fd, bool dont_bind_reply)
{
char cmsg_buf[CMSG_SPACE(sizeof(struct sockaddr_storage))];
struct sockaddr_storage _src_addr = { 0 };
@@ -373,7 +374,7 @@ static int udp_recv_send(int server_fd)
struct iovec iov = { 0 };
struct cmsghdr *cm;
char buf[1];
- int ret, fd;
+ int fd;
ssize_t n;
iov.iov_base = buf;
@@ -415,26 +416,37 @@ static int udp_recv_send(int server_fd)
v4_to_v6(dst_addr);
}
- /* Reply from original destination address. */
- fd = start_server_addr(SOCK_DGRAM, dst_addr, sizeof(*dst_addr), NULL);
- if (!ASSERT_OK_FD(fd, "start_server_addr")) {
- log_err("failed to create tx socket");
- return -1;
- }
+ if (dont_bind_reply) {
+ /* Reply directly from server fd, specifying the source address
+* and/or port in struct msghdr.
+*/
+ n = sendmsg(server_fd, &msg, 0);
+ if (CHECK(n <= 0, "sendmsg", "failed\n")) {
+ log_err("failed to send echo reply");
+ return -1;
+ }
+ } else {
+ /* Reply from original destination address. */
+ fd = start_server_addr(SOCK_DGRAM, dst_addr, sizeof(*dst_addr),
+ NULL);
+ if (!ASSERT_OK_FD(fd, "start_server_addr")) {
+ log_err("failed to create tx socket");
+ return -1;
+ }
- msg.msg_control = NULL;
- msg.msg_controllen = 0;
- n = sendmsg(fd, &msg, 0);
- if (CHECK(n <= 0, "sendmsg", "failed\n")) {
- log_err("failed to send echo reply");
- ret = -1;
- goto out;
+ msg.msg_control = NULL;
+ msg.msg_controllen = 0;
+ n = sendmsg(fd, &msg, 0);
+ if (CHECK(n <= 0, "sendmsg", "failed\n")) {
+ log_err("failed to send echo reply");
+ close(fd);
+ return -1;
+ }
+
+ close(fd);
}
- ret = 0;
-out:
- close(fd);
- return ret;
+ return 0;
}
static int tcp_echo_test(int client_fd, int server_fd)
@@ -454,14 +466,14 @@ static int tcp_echo_test(int client_fd, int server_fd)
return 0;
}
-static int udp_echo_test(int client_fd, int server_fd)
+static int udp_echo_test(int client_fd, int server_fd, bool dont_bind_reply)
{
int err;
err = send_byte(client_fd);
if (err)
return -1;
- err = udp_recv_send(server_fd);
+ err = udp_recv_send(server_fd, dont_bind_reply);
if (err)
return -1;
err = recv_byte(client_fd);
@@ -653,7 +665,8 @@ static void run_lookup_prog(const struct test *t)
if (t->sotype == SOCK_STREAM)
tcp_echo_test(client_fd, server_fds[t->accept_on]);
else
- udp_echo_test(client_fd, server_fds[t->accept_on]);
+ udp_echo_test(client_fd, server_fds[t->accept_on],
+ t->dont_bind_reply);
close(client_fd);
close:
@@ -775,6 +788,16 @@ static void test_redirect_lookup(struct test_sk_lookup
*skel)
.listen_at = { INT_IP4, INT_PORT },
.accept_on = SERVER_B,
},
+ {
+ .desc = &
[RFC PATCH v2 2/3] ipv6: Support setting src port in sendmsg().
This follows the same rationale provided for the ipv4 counterpart, where
the sendmsg() path is also extended here to support the IPV6_ORIGDSTADDR
ancillary message to be able to specify a source address/port. This
allows users to configure the source address and/or port egress traffic
should be sent from.
To limit its usage, a reverse socket lookup is performed to check if the
configured egress source address and/or port have any ingress sk_lookup
match. If it does, traffic is allowed to proceed, otherwise it falls
back to the regular egress path.
Suggested-by: Jakub Sitnicki
Signed-off-by: Tiago Lam
---
net/ipv6/datagram.c | 79 +
net/ipv6/udp.c | 8 --
2 files changed, 85 insertions(+), 2 deletions(-)
diff --git a/net/ipv6/datagram.c b/net/ipv6/datagram.c
index fff78496803d..369c64a478ec 100644
--- a/net/ipv6/datagram.c
+++ b/net/ipv6/datagram.c
@@ -756,6 +756,29 @@ void ip6_datagram_recv_ctl(struct sock *sk, struct msghdr
*msg,
}
EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
+static inline bool reverse_sk_lookup(struct flowi6 *fl6, struct sock *sk,
+struct in6_addr *saddr, __be16 sport)
+{
+ if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
+ (saddr && sport) &&
+ (ipv6_addr_cmp(&sk->sk_v6_rcv_saddr, saddr) ||
+ inet_sk(sk)->inet_sport != sport)) {
+ struct sock *sk_egress;
+
+ bpf_sk_lookup_run_v6(sock_net(sk), IPPROTO_UDP, &fl6->daddr,
+fl6->fl6_dport, saddr, ntohs(sport), 0,
+&sk_egress);
+ if (!IS_ERR_OR_NULL(sk_egress) && sk_egress == sk)
+ return true;
+
+ net_info_ratelimited("No reverse socket lookup match for local
addr %pI6:%d remote addr %pI6:%d\n",
+&saddr, ntohs(sport), &fl6->daddr,
+ntohs(fl6->fl6_dport));
+ }
+
+ return false;
+}
+
int ip6_datagram_send_ctl(struct net *net, struct sock *sk,
struct msghdr *msg, struct flowi6 *fl6,
struct ipcm6_cookie *ipc6)
@@ -844,7 +867,63 @@ int ip6_datagram_send_ctl(struct net *net, struct sock *sk,
break;
}
+ case IPV6_ORIGDSTADDR:
+ {
+ struct sockaddr_in6 *sockaddr_in;
+ struct net_device *dev = NULL;
+
+ if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct
sockaddr_in6))) {
+ err = -EINVAL;
+ goto exit_f;
+ }
+
+ sockaddr_in = (struct sockaddr_in6 *)CMSG_DATA(cmsg);
+
+ addr_type = __ipv6_addr_type(&sockaddr_in->sin6_addr);
+
+ if (addr_type & IPV6_ADDR_LINKLOCAL)
+ return -EINVAL;
+
+ /* If we're egressing with a different source address
+* and/or port, we perform a reverse socket lookup. The
+* rationale behind this is that we can allow return
+* UDP traffic that has ingressed through sk_lookup to
+* also egress correctly. In case the reverse lookup
+* fails, we continue with the normal path.
+*
+* The lookup is performed if either source address
+* and/or port changed, and neither is "0".
+*/
+ if (reverse_sk_lookup(fl6, sk, &sockaddr_in->sin6_addr,
+ sockaddr_in->sin6_port)) {
+ /* Override the source port and address to use
+* with the one we got in cmsg and bail early.
+*/
+ fl6->saddr = sockaddr_in->sin6_addr;
+ fl6->fl6_sport = sockaddr_in->sin6_port;
+ break;
+ }
+ if (addr_type != IPV6_ADDR_ANY) {
+ int strict = __ipv6_addr_src_scope(addr_type)
<= IPV6_ADDR_SCOPE_LINKLOCAL;
+
+ if (!ipv6_can_nonlocal_bind(net, inet_sk(sk)) &&
+ !ipv6_chk_addr_and_flags(net,
+
&sockaddr_in->sin6_addr,
+dev, !strict, 0,
+IFA_F_TENTATIVE) &am
