Hello,
On Sat, 16 Mar 2019, Jacky Hu wrote:
> ipip packets are blocked in some public cloud environments, this patch
> allows gue encapsulation with the tunneling method, which would make
> tunneling working in those environments.
>
> Signed-off-by: Jacky Hu <hengqing.hu@xxxxxxxxx>
> ---
> include/net/ip_vs.h | 5 ++
> include/uapi/linux/ip_vs.h | 17 ++++++
> net/netfilter/ipvs/ip_vs_ctl.c | 20 ++++++-
> net/netfilter/ipvs/ip_vs_xmit.c | 93 +++++++++++++++++++++++++++++++--
> 4 files changed, 130 insertions(+), 5 deletions(-)
>
> diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
> index a0d2e0bb9a94..56c1770b00fe 100644
> --- a/include/net/ip_vs.h
> +++ b/include/net/ip_vs.h
> @@ -603,6 +603,9 @@ struct ip_vs_dest_user_kern {
>
> /* Address family of addr */
> u16 af;
> +
> + u16 tun_type; /* tunnel type */
> + __be16 tun_port; /* tunnel port */
> };
>
>
> @@ -663,6 +666,8 @@ struct ip_vs_dest {
> atomic_t conn_flags; /* flags to copy to conn */
> atomic_t weight; /* server weight */
> atomic_t last_weight; /* server latest weight */
> + atomic_t tun_type; /* tunnel type */
> + atomic_t tun_port; /* tunnel port */
>
> refcount_t refcnt; /* reference counter */
> struct ip_vs_stats stats; /* statistics */
> diff --git a/include/uapi/linux/ip_vs.h b/include/uapi/linux/ip_vs.h
> index 1c916b2f89dc..b43297691337 100644
> --- a/include/uapi/linux/ip_vs.h
> +++ b/include/uapi/linux/ip_vs.h
> @@ -124,6 +124,13 @@
>
> #define IP_VS_PEDATA_MAXLEN 255
>
> +/* Tunnel types */
> +enum {
> + IP_VS_CONN_F_TUNNEL_TYPE_IPIP = 0, /* IPIP */
> + IP_VS_CONN_F_TUNNEL_TYPE_GUE, /* GUE */
> + IP_VS_CONN_F_TUNNEL_TYPE_MAX,
> +};
> +
> /*
> * The struct ip_vs_service_user and struct ip_vs_dest_user are
> * used to set IPVS rules through setsockopt.
> @@ -155,6 +162,9 @@ struct ip_vs_dest_user {
> /* thresholds for active connections */
> __u32 u_threshold; /* upper threshold */
> __u32 l_threshold; /* lower threshold */
> +
> + __u16 tun_type; /* tunnel type */
> + __be16 tun_port; /* tunnel port */
The struct ip_vs_dest_user can not be changed anymore, it is
for the old setsockopt interface, for the new netlink interface we
extend only struct ip_vs_dest_user_kern which is not visible to
user space.
> };
>
>
> @@ -220,6 +230,9 @@ struct ip_vs_dest_entry {
> __u32 u_threshold; /* upper threshold */
> __u32 l_threshold; /* lower threshold */
>
> + __u16 tun_type; /* tunnel type */
> + __be16 tun_port; /* tunnel port */
> +
Same for struct ip_vs_dest_entry, we can not change it
> __u32 activeconns; /* active connections */
> __u32 inactconns; /* inactive connections */
> __u32 persistconns; /* persistent connections */
> @@ -392,6 +405,10 @@ enum {
>
> IPVS_DEST_ATTR_STATS64, /* nested attribute for dest stats */
>
> + IPVS_DEST_ATTR_TUN_TYPE, /* tunnel type */
> +
> + IPVS_DEST_ATTR_TUN_PORT, /* tunnel port */
> +
> __IPVS_DEST_ATTR_MAX,
> };
>
> diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
> index 432141f04af3..48509b03a5ea 100644
> --- a/net/netfilter/ipvs/ip_vs_ctl.c
> +++ b/net/netfilter/ipvs/ip_vs_ctl.c
> @@ -830,6 +830,10 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
> conn_flags = udest->conn_flags & IP_VS_CONN_F_DEST_MASK;
> conn_flags |= IP_VS_CONN_F_INACTIVE;
>
> + /* set the tunnel info */
> + atomic_set(&dest->tun_type, udest->tun_type);
> + atomic_set(&dest->tun_port, udest->tun_port);
> +
> /* set the IP_VS_CONN_F_NOOUTPUT flag if not masquerading/NAT */
> if ((conn_flags & IP_VS_CONN_F_FWD_MASK) != IP_VS_CONN_F_MASQ) {
> conn_flags |= IP_VS_CONN_F_NOOUTPUT;
> @@ -2869,6 +2873,8 @@ static const struct nla_policy ip_vs_dest_policy[IPVS_DEST_ATTR_MAX + 1] = {
> [IPVS_DEST_ATTR_PERSIST_CONNS] = { .type = NLA_U32 },
> [IPVS_DEST_ATTR_STATS] = { .type = NLA_NESTED },
> [IPVS_DEST_ATTR_ADDR_FAMILY] = { .type = NLA_U16 },
> + [IPVS_DEST_ATTR_TUN_TYPE] = { .type = NLA_U8 },
> + [IPVS_DEST_ATTR_TUN_PORT] = { .type = NLA_U16 },
> };
>
> static int ip_vs_genl_fill_stats(struct sk_buff *skb, int container_type,
> @@ -3172,6 +3178,10 @@ static int ip_vs_genl_fill_dest(struct sk_buff *skb, struct ip_vs_dest *dest)
> IP_VS_CONN_F_FWD_MASK)) ||
> nla_put_u32(skb, IPVS_DEST_ATTR_WEIGHT,
> atomic_read(&dest->weight)) ||
> + nla_put_u8(skb, IPVS_DEST_ATTR_TUN_TYPE,
> + atomic_read(&dest->tun_type)) ||
> + nla_put_u16(skb, IPVS_DEST_ATTR_TUN_PORT,
> + atomic_read(&dest->tun_port)) ||
nla_put_be16 for port
> nla_put_u32(skb, IPVS_DEST_ATTR_U_THRESH, dest->u_threshold) ||
> nla_put_u32(skb, IPVS_DEST_ATTR_L_THRESH, dest->l_threshold) ||
> nla_put_u32(skb, IPVS_DEST_ATTR_ACTIVE_CONNS,
> @@ -3294,12 +3304,14 @@ static int ip_vs_genl_parse_dest(struct ip_vs_dest_user_kern *udest,
> /* If a full entry was requested, check for the additional fields */
> if (full_entry) {
> struct nlattr *nla_fwd, *nla_weight, *nla_u_thresh,
> - *nla_l_thresh;
> + *nla_l_thresh, *nla_tun_type, *nla_tun_port;
>
> nla_fwd = attrs[IPVS_DEST_ATTR_FWD_METHOD];
> nla_weight = attrs[IPVS_DEST_ATTR_WEIGHT];
> nla_u_thresh = attrs[IPVS_DEST_ATTR_U_THRESH];
> nla_l_thresh = attrs[IPVS_DEST_ATTR_L_THRESH];
> + nla_tun_type = attrs[IPVS_DEST_ATTR_TUN_TYPE];
> + nla_tun_port = attrs[IPVS_DEST_ATTR_TUN_PORT];
>
> if (!(nla_fwd && nla_weight && nla_u_thresh && nla_l_thresh))
> return -EINVAL;
> @@ -3309,6 +3321,12 @@ static int ip_vs_genl_parse_dest(struct ip_vs_dest_user_kern *udest,
> udest->weight = nla_get_u32(nla_weight);
> udest->u_threshold = nla_get_u32(nla_u_thresh);
> udest->l_threshold = nla_get_u32(nla_l_thresh);
> +
> + if (nla_tun_type)
> + udest->tun_type = nla_get_u8(nla_tun_type);
> +
> + if (nla_tun_port)
> + udest->tun_port = nla_get_u16(nla_tun_port);
nla_get_be16 for port
> }
>
> return 0;
> diff --git a/net/netfilter/ipvs/ip_vs_xmit.c b/net/netfilter/ipvs/ip_vs_xmit.c
> index 473cce2a5231..e748c5605b04 100644
> --- a/net/netfilter/ipvs/ip_vs_xmit.c
> +++ b/net/netfilter/ipvs/ip_vs_xmit.c
> @@ -32,6 +32,7 @@
> #include <linux/slab.h>
> #include <linux/tcp.h> /* for tcphdr */
> #include <net/ip.h>
> +#include <net/gue.h>
> #include <net/tcp.h> /* for csum_tcpudp_magic */
> #include <net/udp.h>
> #include <net/icmp.h> /* for icmp_send */
> @@ -989,6 +990,42 @@ static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
> }
> }
>
> +static int
> +__build_gue_header(struct net *net, struct sk_buff *skb,
> + struct ip_vs_conn *cp, __u8 *next_protocol)
> +{
> + __be16 dport;
> + __be16 sport;
> + struct udphdr *udph; /* Our new UDP header */
> + struct guehdr *gueh; /* Our new GUE header */
> +
> + skb_push(skb, sizeof(struct guehdr));
> +
> + gueh = (struct guehdr *)skb->data;
> +
> + gueh->control = 0;
> + gueh->version = 0;
> + gueh->hlen = 0;
> + gueh->flags = 0;
> + gueh->proto_ctype = *next_protocol;
> +
> + skb_push(skb, sizeof(struct udphdr));
> + skb_reset_transport_header(skb);
> +
> + udph = udp_hdr(skb);
> +
> + dport = htons(atomic_read(&cp->dest->tun_port));
> + sport = udp_flow_src_port(net, skb, 0, 0, false);
udp_flow_src_port should be called before pushing new headers,
eg. move it above:
__be16 sport = udp_flow_src_port(net, skb, 0, 0, false);
> + udph->dest = dport;
> + udph->source = sport;
> + udph->len = htons(skb->len);
> + udph->check = 0;
For udph->check there can be an additional patch/feature:
- we can add options for checksums, for example:
--tun-nocsum: uh->check = 0, SKB_GSO_UDP_TUNNEL
No checksum
default
--tun-csum:
Checksum
SKB_GSO_UDP_TUNNEL_CSUM
udp_set_csum/udp6_set_csum as in fou_build_udp() and
fou6_build_udp()
--tun-remcsum:
Remote Checksum Offload
SKB_GSO_UDP_TUNNEL_CSUM | SKB_GSO_TUNNEL_REMCSUM
udp_set_csum/udp6_set_csum
like net/ipv4/fou.c:__gue_build_header() REMCSUM handling
But for now check = 0 should be fine.
> +
> + *next_protocol = IPPROTO_UDP;
> +
> + return 0;
> +}
> +
> /*
> * IP Tunneling transmitter
> *
> @@ -1025,6 +1062,7 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
> struct iphdr *iph; /* Our new IP header */
> unsigned int max_headroom; /* The extra header space needed */
> int ret, local;
> + int tun_type, gso_type;
>
> EnterFunction(10);
>
> @@ -1033,7 +1071,7 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
> IP_VS_RT_MODE_NON_LOCAL |
> IP_VS_RT_MODE_CONNECT |
> IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
> - if (local < 0)
> + if (!cp->dest || local < 0)
May be this cp->dest check can be moved to __ip_vs_get_out_rt()
and __ip_vs_get_out_rt_v6() because we need to account the added header
when calculating mtu and also because it should not stop the traffic when
local = true:
mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
add:
if (!dest)
goto err_put;
switch (atomic_read(&dest->tun_type)) {
case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
mtu -= sizeof(struct udphdr) + sizeof(struct guehdr);
break;
default:
break;
}
> goto tx_error;
> if (local)
> return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
> @@ -1046,6 +1084,14 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
> */
> max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
>
> + tun_type = atomic_read(&cp->dest->tun_type);
> +
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + max_headroom += sizeof(struct udphdr) + sizeof(struct guehdr);
> + break;
> + }
> +
> /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
> dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
> skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
> @@ -1054,11 +1100,26 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
> if (IS_ERR(skb))
> goto tx_error;
>
> - if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET, cp->af)))
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + gso_type = SKB_GSO_UDP_TUNNEL;
> + break;
> + default:
> + gso_type = __tun_gso_type_mask(AF_INET, cp->af);
> + break;
> + }
> +
> + if (iptunnel_handle_offloads(skb, gso_type))
> goto tx_error;
>
> skb->transport_header = skb->network_header;
>
We should call skb_set_inner_ipproto(skb, next_protocol) to
properly segment the skb in skb_udp_tunnel_segment(), after
iptunnel_handle_offloads and before it is changed by __build_gue_header
from IPPROTO_IPIP/IPPROTO_IPV6 to UDP.
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + __build_gue_header(dev_net(tdev), skb, cp, &next_protocol);
Use net instead of dev_net(tdev)
> + break;
> + }
> +
> skb_push(skb, sizeof(struct iphdr));
> skb_reset_network_header(skb);
> memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
> @@ -1112,6 +1173,7 @@ ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
> struct ipv6hdr *iph; /* Our new IP header */
> unsigned int max_headroom; /* The extra header space needed */
> int ret, local;
> + int tun_type, gso_type;
>
> EnterFunction(10);
>
> @@ -1121,7 +1183,7 @@ ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
> IP_VS_RT_MODE_LOCAL |
> IP_VS_RT_MODE_NON_LOCAL |
> IP_VS_RT_MODE_TUNNEL);
> - if (local < 0)
> + if (!cp->dest || local < 0)
Above dest check will be moved before the mtu check
> goto tx_error;
> if (local)
> return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
> @@ -1134,17 +1196,40 @@ ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
> */
> max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
>
> + tun_type = atomic_read(&cp->dest->tun_type);
> +
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + max_headroom += sizeof(struct udphdr) + sizeof(struct guehdr);
> + break;
> + }
> +
> skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
> &next_protocol, &payload_len,
> &dsfield, &ttl, NULL);
> if (IS_ERR(skb))
> goto tx_error;
>
> - if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET6, cp->af)))
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + gso_type = SKB_GSO_UDP_TUNNEL;
> + break;
> + default:
> + gso_type = __tun_gso_type_mask(AF_INET, cp->af);
For IPv6 use AF_INET6
> + break;
> + }
> +
> + if (iptunnel_handle_offloads(skb, gso_type))
> goto tx_error;
>
> skb->transport_header = skb->network_header;
>
skb_set_inner_ipproto(skb, next_protocol);
> + switch (tun_type) {
> + case IP_VS_CONN_F_TUNNEL_TYPE_GUE:
> + __build_gue_header(dev_net(tdev), skb, cp, &next_protocol);
Use net instead of dev_net(tdev)
> + break;
> + }
> +
> skb_push(skb, sizeof(struct ipv6hdr));
> skb_reset_network_header(skb);
> memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
> --
Also:
- just for correctness we can set udest->tun_type to
IP_VS_CONN_F_TUNNEL_TYPE_IPIP in ip_vs_copy_udest_compat()
- __build_gue_header: can be named ipvs_gue_encap because we add 2 headers
- ip_vs_add_dest() and ip_vs_edit_dest() can return EINVAL when
GUE comes with port 0.
As for ipvsadm v2 patch, I'll comment it again when we stabilize
the kernel patch but for now few comments:
- there are missing switch-case breaks
- for FMT_TUN_INFO in print_title it would be enough to show only this:
-> RemoteAddress:Port Forward TunnelInfo":
- in print_service_entry() better to use func to print
tunnames[e->tun_type] in case the kernel returns something ipvsadm
does not support yet
- libipvs/ip_vs.h: struct ip_vs_dest_kern: can not be changed
- libipvs/ip_vs.h: struct ip_vs_dest_entry_kern: can not be changed
Regards
--
Julian Anastasov <ja@xxxxxx>
