Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 #include <linux/kernel.h>
0003 #include <linux/module.h>
0004 #include <linux/init.h>
0005 #include <linux/netlink.h>
0006 #include <linux/netfilter.h>
0007 #include <linux/workqueue.h>
0008 #include <linux/spinlock.h>
0009 #include <linux/netfilter/nf_conntrack_common.h>
0010 #include <linux/netfilter/nf_tables.h>
0011 #include <net/ip.h> /* for ipv4 options. */
0012 #include <net/netfilter/nf_tables.h>
0013 #include <net/netfilter/nf_tables_core.h>
0014 #include <net/netfilter/nf_conntrack_core.h>
0015 #include <net/netfilter/nf_conntrack_extend.h>
0016 #include <net/netfilter/nf_flow_table.h>
0017 
0018 struct nft_flow_offload {
0019     struct nft_flowtable    *flowtable;
0020 };
0021 
0022 static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
0023 {
0024     if (dst_xfrm(dst))
0025         return FLOW_OFFLOAD_XMIT_XFRM;
0026 
0027     return FLOW_OFFLOAD_XMIT_NEIGH;
0028 }
0029 
0030 static void nft_default_forward_path(struct nf_flow_route *route,
0031                      struct dst_entry *dst_cache,
0032                      enum ip_conntrack_dir dir)
0033 {
0034     route->tuple[!dir].in.ifindex   = dst_cache->dev->ifindex;
0035     route->tuple[dir].dst       = dst_cache;
0036     route->tuple[dir].xmit_type = nft_xmit_type(dst_cache);
0037 }
0038 
0039 static bool nft_is_valid_ether_device(const struct net_device *dev)
0040 {
0041     if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
0042         dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
0043         return false;
0044 
0045     return true;
0046 }
0047 
0048 static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
0049                      const struct dst_entry *dst_cache,
0050                      const struct nf_conn *ct,
0051                      enum ip_conntrack_dir dir, u8 *ha,
0052                      struct net_device_path_stack *stack)
0053 {
0054     const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
0055     struct net_device *dev = dst_cache->dev;
0056     struct neighbour *n;
0057     u8 nud_state;
0058 
0059     if (!nft_is_valid_ether_device(dev))
0060         goto out;
0061 
0062     n = dst_neigh_lookup(dst_cache, daddr);
0063     if (!n)
0064         return -1;
0065 
0066     read_lock_bh(&n->lock);
0067     nud_state = n->nud_state;
0068     ether_addr_copy(ha, n->ha);
0069     read_unlock_bh(&n->lock);
0070     neigh_release(n);
0071 
0072     if (!(nud_state & NUD_VALID))
0073         return -1;
0074 
0075 out:
0076     return dev_fill_forward_path(dev, ha, stack);
0077 }
0078 
0079 struct nft_forward_info {
0080     const struct net_device *indev;
0081     const struct net_device *outdev;
0082     const struct net_device *hw_outdev;
0083     struct id {
0084         __u16   id;
0085         __be16  proto;
0086     } encap[NF_FLOW_TABLE_ENCAP_MAX];
0087     u8 num_encaps;
0088     u8 ingress_vlans;
0089     u8 h_source[ETH_ALEN];
0090     u8 h_dest[ETH_ALEN];
0091     enum flow_offload_xmit_type xmit_type;
0092 };
0093 
0094 static void nft_dev_path_info(const struct net_device_path_stack *stack,
0095                   struct nft_forward_info *info,
0096                   unsigned char *ha, struct nf_flowtable *flowtable)
0097 {
0098     const struct net_device_path *path;
0099     int i;
0100 
0101     memcpy(info->h_dest, ha, ETH_ALEN);
0102 
0103     for (i = 0; i < stack->num_paths; i++) {
0104         path = &stack->path[i];
0105         switch (path->type) {
0106         case DEV_PATH_ETHERNET:
0107         case DEV_PATH_DSA:
0108         case DEV_PATH_VLAN:
0109         case DEV_PATH_PPPOE:
0110             info->indev = path->dev;
0111             if (is_zero_ether_addr(info->h_source))
0112                 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
0113 
0114             if (path->type == DEV_PATH_ETHERNET)
0115                 break;
0116             if (path->type == DEV_PATH_DSA) {
0117                 i = stack->num_paths;
0118                 break;
0119             }
0120 
0121             /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
0122             if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
0123                 info->indev = NULL;
0124                 break;
0125             }
0126             if (!info->outdev)
0127                 info->outdev = path->dev;
0128             info->encap[info->num_encaps].id = path->encap.id;
0129             info->encap[info->num_encaps].proto = path->encap.proto;
0130             info->num_encaps++;
0131             if (path->type == DEV_PATH_PPPOE)
0132                 memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
0133             break;
0134         case DEV_PATH_BRIDGE:
0135             if (is_zero_ether_addr(info->h_source))
0136                 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
0137 
0138             switch (path->bridge.vlan_mode) {
0139             case DEV_PATH_BR_VLAN_UNTAG_HW:
0140                 info->ingress_vlans |= BIT(info->num_encaps - 1);
0141                 break;
0142             case DEV_PATH_BR_VLAN_TAG:
0143                 info->encap[info->num_encaps].id = path->bridge.vlan_id;
0144                 info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
0145                 info->num_encaps++;
0146                 break;
0147             case DEV_PATH_BR_VLAN_UNTAG:
0148                 info->num_encaps--;
0149                 break;
0150             case DEV_PATH_BR_VLAN_KEEP:
0151                 break;
0152             }
0153             info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
0154             break;
0155         default:
0156             info->indev = NULL;
0157             break;
0158         }
0159     }
0160     if (!info->outdev)
0161         info->outdev = info->indev;
0162 
0163     info->hw_outdev = info->indev;
0164 
0165     if (nf_flowtable_hw_offload(flowtable) &&
0166         nft_is_valid_ether_device(info->indev))
0167         info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
0168 }
0169 
0170 static bool nft_flowtable_find_dev(const struct net_device *dev,
0171                    struct nft_flowtable *ft)
0172 {
0173     struct nft_hook *hook;
0174     bool found = false;
0175 
0176     list_for_each_entry_rcu(hook, &ft->hook_list, list) {
0177         if (hook->ops.dev != dev)
0178             continue;
0179 
0180         found = true;
0181         break;
0182     }
0183 
0184     return found;
0185 }
0186 
0187 static void nft_dev_forward_path(struct nf_flow_route *route,
0188                  const struct nf_conn *ct,
0189                  enum ip_conntrack_dir dir,
0190                  struct nft_flowtable *ft)
0191 {
0192     const struct dst_entry *dst = route->tuple[dir].dst;
0193     struct net_device_path_stack stack;
0194     struct nft_forward_info info = {};
0195     unsigned char ha[ETH_ALEN];
0196     int i;
0197 
0198     if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
0199         nft_dev_path_info(&stack, &info, ha, &ft->data);
0200 
0201     if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
0202         return;
0203 
0204     route->tuple[!dir].in.ifindex = info.indev->ifindex;
0205     for (i = 0; i < info.num_encaps; i++) {
0206         route->tuple[!dir].in.encap[i].id = info.encap[i].id;
0207         route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
0208     }
0209     route->tuple[!dir].in.num_encaps = info.num_encaps;
0210     route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
0211 
0212     if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
0213         memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
0214         memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
0215         route->tuple[dir].out.ifindex = info.outdev->ifindex;
0216         route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
0217         route->tuple[dir].xmit_type = info.xmit_type;
0218     }
0219 }
0220 
0221 static int nft_flow_route(const struct nft_pktinfo *pkt,
0222               const struct nf_conn *ct,
0223               struct nf_flow_route *route,
0224               enum ip_conntrack_dir dir,
0225               struct nft_flowtable *ft)
0226 {
0227     struct dst_entry *this_dst = skb_dst(pkt->skb);
0228     struct dst_entry *other_dst = NULL;
0229     struct flowi fl;
0230 
0231     memset(&fl, 0, sizeof(fl));
0232     switch (nft_pf(pkt)) {
0233     case NFPROTO_IPV4:
0234         fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
0235         fl.u.ip4.saddr = ct->tuplehash[!dir].tuple.src.u3.ip;
0236         fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
0237         fl.u.ip4.flowi4_iif = this_dst->dev->ifindex;
0238         fl.u.ip4.flowi4_tos = RT_TOS(ip_hdr(pkt->skb)->tos);
0239         fl.u.ip4.flowi4_mark = pkt->skb->mark;
0240         fl.u.ip4.flowi4_flags = FLOWI_FLAG_ANYSRC;
0241         break;
0242     case NFPROTO_IPV6:
0243         fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
0244         fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.src.u3.in6;
0245         fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
0246         fl.u.ip6.flowi6_iif = this_dst->dev->ifindex;
0247         fl.u.ip6.flowlabel = ip6_flowinfo(ipv6_hdr(pkt->skb));
0248         fl.u.ip6.flowi6_mark = pkt->skb->mark;
0249         fl.u.ip6.flowi6_flags = FLOWI_FLAG_ANYSRC;
0250         break;
0251     }
0252 
0253     nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
0254     if (!other_dst)
0255         return -ENOENT;
0256 
0257     nft_default_forward_path(route, this_dst, dir);
0258     nft_default_forward_path(route, other_dst, !dir);
0259 
0260     if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
0261         route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
0262         nft_dev_forward_path(route, ct, dir, ft);
0263         nft_dev_forward_path(route, ct, !dir, ft);
0264     }
0265 
0266     return 0;
0267 }
0268 
0269 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
0270 {
0271     if (skb_sec_path(skb))
0272         return true;
0273 
0274     if (family == NFPROTO_IPV4) {
0275         const struct ip_options *opt;
0276 
0277         opt = &(IPCB(skb)->opt);
0278 
0279         if (unlikely(opt->optlen))
0280             return true;
0281     }
0282 
0283     return false;
0284 }
0285 
0286 static void nft_flow_offload_eval(const struct nft_expr *expr,
0287                   struct nft_regs *regs,
0288                   const struct nft_pktinfo *pkt)
0289 {
0290     struct nft_flow_offload *priv = nft_expr_priv(expr);
0291     struct nf_flowtable *flowtable = &priv->flowtable->data;
0292     struct tcphdr _tcph, *tcph = NULL;
0293     struct nf_flow_route route = {};
0294     enum ip_conntrack_info ctinfo;
0295     struct flow_offload *flow;
0296     enum ip_conntrack_dir dir;
0297     struct nf_conn *ct;
0298     int ret;
0299 
0300     if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
0301         goto out;
0302 
0303     ct = nf_ct_get(pkt->skb, &ctinfo);
0304     if (!ct)
0305         goto out;
0306 
0307     switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
0308     case IPPROTO_TCP:
0309         tcph = skb_header_pointer(pkt->skb, nft_thoff(pkt),
0310                       sizeof(_tcph), &_tcph);
0311         if (unlikely(!tcph || tcph->fin || tcph->rst ||
0312                  !nf_conntrack_tcp_established(ct)))
0313             goto out;
0314         break;
0315     case IPPROTO_UDP:
0316         break;
0317 #ifdef CONFIG_NF_CT_PROTO_GRE
0318     case IPPROTO_GRE: {
0319         struct nf_conntrack_tuple *tuple;
0320 
0321         if (ct->status & IPS_NAT_MASK)
0322             goto out;
0323         tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
0324         /* No support for GRE v1 */
0325         if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
0326             goto out;
0327         break;
0328     }
0329 #endif
0330     default:
0331         goto out;
0332     }
0333 
0334     if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
0335         ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
0336         goto out;
0337 
0338     if (!nf_ct_is_confirmed(ct))
0339         goto out;
0340 
0341     if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
0342         goto out;
0343 
0344     dir = CTINFO2DIR(ctinfo);
0345     if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
0346         goto err_flow_route;
0347 
0348     flow = flow_offload_alloc(ct);
0349     if (!flow)
0350         goto err_flow_alloc;
0351 
0352     if (flow_offload_route_init(flow, &route) < 0)
0353         goto err_flow_add;
0354 
0355     if (tcph) {
0356         ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
0357         ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
0358     }
0359 
0360     ret = flow_offload_add(flowtable, flow);
0361     if (ret < 0)
0362         goto err_flow_add;
0363 
0364     dst_release(route.tuple[!dir].dst);
0365     return;
0366 
0367 err_flow_add:
0368     flow_offload_free(flow);
0369 err_flow_alloc:
0370     dst_release(route.tuple[!dir].dst);
0371 err_flow_route:
0372     clear_bit(IPS_OFFLOAD_BIT, &ct->status);
0373 out:
0374     regs->verdict.code = NFT_BREAK;
0375 }
0376 
0377 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
0378                      const struct nft_expr *expr,
0379                      const struct nft_data **data)
0380 {
0381     unsigned int hook_mask = (1 << NF_INET_FORWARD);
0382 
0383     return nft_chain_validate_hooks(ctx->chain, hook_mask);
0384 }
0385 
0386 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
0387     [NFTA_FLOW_TABLE_NAME]  = { .type = NLA_STRING,
0388                     .len = NFT_NAME_MAXLEN - 1 },
0389 };
0390 
0391 static int nft_flow_offload_init(const struct nft_ctx *ctx,
0392                  const struct nft_expr *expr,
0393                  const struct nlattr * const tb[])
0394 {
0395     struct nft_flow_offload *priv = nft_expr_priv(expr);
0396     u8 genmask = nft_genmask_next(ctx->net);
0397     struct nft_flowtable *flowtable;
0398 
0399     if (!tb[NFTA_FLOW_TABLE_NAME])
0400         return -EINVAL;
0401 
0402     flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
0403                      genmask);
0404     if (IS_ERR(flowtable))
0405         return PTR_ERR(flowtable);
0406 
0407     priv->flowtable = flowtable;
0408     flowtable->use++;
0409 
0410     return nf_ct_netns_get(ctx->net, ctx->family);
0411 }
0412 
0413 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
0414                     const struct nft_expr *expr,
0415                     enum nft_trans_phase phase)
0416 {
0417     struct nft_flow_offload *priv = nft_expr_priv(expr);
0418 
0419     nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
0420 }
0421 
0422 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
0423                       const struct nft_expr *expr)
0424 {
0425     struct nft_flow_offload *priv = nft_expr_priv(expr);
0426 
0427     priv->flowtable->use++;
0428 }
0429 
0430 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
0431                      const struct nft_expr *expr)
0432 {
0433     nf_ct_netns_put(ctx->net, ctx->family);
0434 }
0435 
0436 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
0437 {
0438     struct nft_flow_offload *priv = nft_expr_priv(expr);
0439 
0440     if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
0441         goto nla_put_failure;
0442 
0443     return 0;
0444 
0445 nla_put_failure:
0446     return -1;
0447 }
0448 
0449 static struct nft_expr_type nft_flow_offload_type;
0450 static const struct nft_expr_ops nft_flow_offload_ops = {
0451     .type       = &nft_flow_offload_type,
0452     .size       = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
0453     .eval       = nft_flow_offload_eval,
0454     .init       = nft_flow_offload_init,
0455     .activate   = nft_flow_offload_activate,
0456     .deactivate = nft_flow_offload_deactivate,
0457     .destroy    = nft_flow_offload_destroy,
0458     .validate   = nft_flow_offload_validate,
0459     .dump       = nft_flow_offload_dump,
0460     .reduce     = NFT_REDUCE_READONLY,
0461 };
0462 
0463 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
0464     .name       = "flow_offload",
0465     .ops        = &nft_flow_offload_ops,
0466     .policy     = nft_flow_offload_policy,
0467     .maxattr    = NFTA_FLOW_MAX,
0468     .owner      = THIS_MODULE,
0469 };
0470 
0471 static int flow_offload_netdev_event(struct notifier_block *this,
0472                      unsigned long event, void *ptr)
0473 {
0474     struct net_device *dev = netdev_notifier_info_to_dev(ptr);
0475 
0476     if (event != NETDEV_DOWN)
0477         return NOTIFY_DONE;
0478 
0479     nf_flow_table_cleanup(dev);
0480 
0481     return NOTIFY_DONE;
0482 }
0483 
0484 static struct notifier_block flow_offload_netdev_notifier = {
0485     .notifier_call  = flow_offload_netdev_event,
0486 };
0487 
0488 static int __init nft_flow_offload_module_init(void)
0489 {
0490     int err;
0491 
0492     err = register_netdevice_notifier(&flow_offload_netdev_notifier);
0493     if (err)
0494         goto err;
0495 
0496     err = nft_register_expr(&nft_flow_offload_type);
0497     if (err < 0)
0498         goto register_expr;
0499 
0500     return 0;
0501 
0502 register_expr:
0503     unregister_netdevice_notifier(&flow_offload_netdev_notifier);
0504 err:
0505     return err;
0506 }
0507 
0508 static void __exit nft_flow_offload_module_exit(void)
0509 {
0510     nft_unregister_expr(&nft_flow_offload_type);
0511     unregister_netdevice_notifier(&flow_offload_netdev_notifier);
0512 }
0513 
0514 module_init(nft_flow_offload_module_init);
0515 module_exit(nft_flow_offload_module_exit);
0516 
0517 MODULE_LICENSE("GPL");
0518 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
0519 MODULE_ALIAS_NFT_EXPR("flow_offload");
0520 MODULE_DESCRIPTION("nftables hardware flow offload module");
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!