1 // SPDX-License-Identifier: GPL-2.0-only 2 #include <linux/kernel.h> 3 #include <linux/module.h> 4 #include <linux/init.h> 5 #include <linux/netlink.h> 6 #include <linux/netfilter.h> 7 #include <linux/workqueue.h> 8 #include <linux/spinlock.h> 9 #include <linux/netfilter/nf_conntrack_common.h> 10 #include <linux/netfilter/nf_tables.h> 11 #include <net/ip.h> /* for ipv4 options. */ 12 #include <net/netfilter/nf_tables.h> 13 #include <net/netfilter/nf_tables_core.h> 14 #include <net/netfilter/nf_conntrack_core.h> 15 #include <net/netfilter/nf_conntrack_extend.h> 16 #include <net/netfilter/nf_flow_table.h> 17 18 struct nft_flow_offload { 19 struct nft_flowtable *flowtable; 20 }; 21 22 static int nft_flow_route(const struct nft_pktinfo *pkt, 23 const struct nf_conn *ct, 24 struct nf_flow_route *route, 25 enum ip_conntrack_dir dir) 26 { 27 struct dst_entry *this_dst = skb_dst(pkt->skb); 28 struct dst_entry *other_dst = NULL; 29 struct flowi fl; 30 31 memset(&fl, 0, sizeof(fl)); 32 switch (nft_pf(pkt)) { 33 case NFPROTO_IPV4: 34 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip; 35 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex; 36 break; 37 case NFPROTO_IPV6: 38 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6; 39 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex; 40 break; 41 } 42 43 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt)); 44 if (!other_dst) 45 return -ENOENT; 46 47 route->tuple[dir].dst = this_dst; 48 route->tuple[!dir].dst = other_dst; 49 50 return 0; 51 } 52 53 static bool nft_flow_offload_skip(struct sk_buff *skb, int family) 54 { 55 if (skb_sec_path(skb)) 56 return true; 57 58 if (family == NFPROTO_IPV4) { 59 const struct ip_options *opt; 60 61 opt = &(IPCB(skb)->opt); 62 63 if (unlikely(opt->optlen)) 64 return true; 65 } 66 67 return false; 68 } 69 70 static void nft_flow_offload_eval(const struct nft_expr *expr, 71 struct nft_regs *regs, 72 const struct nft_pktinfo *pkt) 73 { 74 struct nft_flow_offload *priv = nft_expr_priv(expr); 75 struct nf_flowtable *flowtable = &priv->flowtable->data; 76 struct tcphdr _tcph, *tcph = NULL; 77 enum ip_conntrack_info ctinfo; 78 struct nf_flow_route route; 79 struct flow_offload *flow; 80 enum ip_conntrack_dir dir; 81 struct nf_conn *ct; 82 int ret; 83 84 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt))) 85 goto out; 86 87 ct = nf_ct_get(pkt->skb, &ctinfo); 88 if (!ct) 89 goto out; 90 91 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) { 92 case IPPROTO_TCP: 93 tcph = skb_header_pointer(pkt->skb, pkt->xt.thoff, 94 sizeof(_tcph), &_tcph); 95 if (unlikely(!tcph || tcph->fin || tcph->rst)) 96 goto out; 97 break; 98 case IPPROTO_UDP: 99 break; 100 default: 101 goto out; 102 } 103 104 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 105 ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH)) 106 goto out; 107 108 if (!nf_ct_is_confirmed(ct)) 109 goto out; 110 111 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 112 goto out; 113 114 dir = CTINFO2DIR(ctinfo); 115 if (nft_flow_route(pkt, ct, &route, dir) < 0) 116 goto err_flow_route; 117 118 flow = flow_offload_alloc(ct); 119 if (!flow) 120 goto err_flow_alloc; 121 122 if (flow_offload_route_init(flow, &route) < 0) 123 goto err_flow_add; 124 125 if (tcph) { 126 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 127 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 128 } 129 130 ret = flow_offload_add(flowtable, flow); 131 if (ret < 0) 132 goto err_flow_add; 133 134 dst_release(route.tuple[!dir].dst); 135 return; 136 137 err_flow_add: 138 flow_offload_free(flow); 139 err_flow_alloc: 140 dst_release(route.tuple[!dir].dst); 141 err_flow_route: 142 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 143 out: 144 regs->verdict.code = NFT_BREAK; 145 } 146 147 static int nft_flow_offload_validate(const struct nft_ctx *ctx, 148 const struct nft_expr *expr, 149 const struct nft_data **data) 150 { 151 unsigned int hook_mask = (1 << NF_INET_FORWARD); 152 153 return nft_chain_validate_hooks(ctx->chain, hook_mask); 154 } 155 156 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = { 157 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING, 158 .len = NFT_NAME_MAXLEN - 1 }, 159 }; 160 161 static int nft_flow_offload_init(const struct nft_ctx *ctx, 162 const struct nft_expr *expr, 163 const struct nlattr * const tb[]) 164 { 165 struct nft_flow_offload *priv = nft_expr_priv(expr); 166 u8 genmask = nft_genmask_next(ctx->net); 167 struct nft_flowtable *flowtable; 168 169 if (!tb[NFTA_FLOW_TABLE_NAME]) 170 return -EINVAL; 171 172 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME], 173 genmask); 174 if (IS_ERR(flowtable)) 175 return PTR_ERR(flowtable); 176 177 priv->flowtable = flowtable; 178 flowtable->use++; 179 180 return nf_ct_netns_get(ctx->net, ctx->family); 181 } 182 183 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx, 184 const struct nft_expr *expr, 185 enum nft_trans_phase phase) 186 { 187 struct nft_flow_offload *priv = nft_expr_priv(expr); 188 189 nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase); 190 } 191 192 static void nft_flow_offload_activate(const struct nft_ctx *ctx, 193 const struct nft_expr *expr) 194 { 195 struct nft_flow_offload *priv = nft_expr_priv(expr); 196 197 priv->flowtable->use++; 198 } 199 200 static void nft_flow_offload_destroy(const struct nft_ctx *ctx, 201 const struct nft_expr *expr) 202 { 203 nf_ct_netns_put(ctx->net, ctx->family); 204 } 205 206 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr) 207 { 208 struct nft_flow_offload *priv = nft_expr_priv(expr); 209 210 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name)) 211 goto nla_put_failure; 212 213 return 0; 214 215 nla_put_failure: 216 return -1; 217 } 218 219 static struct nft_expr_type nft_flow_offload_type; 220 static const struct nft_expr_ops nft_flow_offload_ops = { 221 .type = &nft_flow_offload_type, 222 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)), 223 .eval = nft_flow_offload_eval, 224 .init = nft_flow_offload_init, 225 .activate = nft_flow_offload_activate, 226 .deactivate = nft_flow_offload_deactivate, 227 .destroy = nft_flow_offload_destroy, 228 .validate = nft_flow_offload_validate, 229 .dump = nft_flow_offload_dump, 230 }; 231 232 static struct nft_expr_type nft_flow_offload_type __read_mostly = { 233 .name = "flow_offload", 234 .ops = &nft_flow_offload_ops, 235 .policy = nft_flow_offload_policy, 236 .maxattr = NFTA_FLOW_MAX, 237 .owner = THIS_MODULE, 238 }; 239 240 static int flow_offload_netdev_event(struct notifier_block *this, 241 unsigned long event, void *ptr) 242 { 243 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 244 245 if (event != NETDEV_DOWN) 246 return NOTIFY_DONE; 247 248 nf_flow_table_cleanup(dev); 249 250 return NOTIFY_DONE; 251 } 252 253 static struct notifier_block flow_offload_netdev_notifier = { 254 .notifier_call = flow_offload_netdev_event, 255 }; 256 257 static int __init nft_flow_offload_module_init(void) 258 { 259 int err; 260 261 err = register_netdevice_notifier(&flow_offload_netdev_notifier); 262 if (err) 263 goto err; 264 265 err = nft_register_expr(&nft_flow_offload_type); 266 if (err < 0) 267 goto register_expr; 268 269 return 0; 270 271 register_expr: 272 unregister_netdevice_notifier(&flow_offload_netdev_notifier); 273 err: 274 return err; 275 } 276 277 static void __exit nft_flow_offload_module_exit(void) 278 { 279 nft_unregister_expr(&nft_flow_offload_type); 280 unregister_netdevice_notifier(&flow_offload_netdev_notifier); 281 } 282 283 module_init(nft_flow_offload_module_init); 284 module_exit(nft_flow_offload_module_exit); 285 286 MODULE_LICENSE("GPL"); 287 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>"); 288 MODULE_ALIAS_NFT_EXPR("flow_offload"); 289 MODULE_DESCRIPTION("nftables hardware flow offload module"); 290