| 1 | #include <linux/kernel.h> |
| 2 | #include <linux/init.h> |
| 3 | #include <linux/module.h> |
| 4 | #include <linux/netfilter.h> |
| 5 | #include <linux/rhashtable.h> |
| 6 | #include <linux/netdevice.h> |
| 7 | #include <linux/tc_act/tc_csum.h> |
| 8 | #include <net/flow_offload.h> |
| 9 | #include <net/netfilter/nf_flow_table.h> |
| 10 | #include <net/netfilter/nf_tables.h> |
| 11 | #include <net/netfilter/nf_conntrack.h> |
| 12 | #include <net/netfilter/nf_conntrack_acct.h> |
| 13 | #include <net/netfilter/nf_conntrack_core.h> |
| 14 | #include <net/netfilter/nf_conntrack_tuple.h> |
| 15 | |
| 16 | static struct workqueue_struct *nf_flow_offload_add_wq; |
| 17 | static struct workqueue_struct *nf_flow_offload_del_wq; |
| 18 | static struct workqueue_struct *nf_flow_offload_stats_wq; |
| 19 | |
| 20 | struct flow_offload_work { |
| 21 | struct list_head list; |
| 22 | enum flow_cls_command cmd; |
| 23 | struct nf_flowtable *flowtable; |
| 24 | struct flow_offload *flow; |
| 25 | struct work_struct work; |
| 26 | }; |
| 27 | |
| 28 | #define NF_FLOW_DISSECTOR(__match, __type, __field) \ |
| 29 | (__match)->dissector.offset[__type] = \ |
| 30 | offsetof(struct nf_flow_key, __field) |
| 31 | |
| 32 | static void nf_flow_rule_lwt_match(struct nf_flow_match *match, |
| 33 | struct ip_tunnel_info *tun_info) |
| 34 | { |
| 35 | struct nf_flow_key *mask = &match->mask; |
| 36 | struct nf_flow_key *key = &match->key; |
| 37 | unsigned long long enc_keys; |
| 38 | |
| 39 | if (!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX)) |
| 40 | return; |
| 41 | |
| 42 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_ENC_CONTROL, enc_control); |
| 43 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_ENC_KEYID, enc_key_id); |
| 44 | key->enc_key_id.keyid = tunnel_id_to_key32(tun_id: tun_info->key.tun_id); |
| 45 | mask->enc_key_id.keyid = 0xffffffff; |
| 46 | enc_keys = BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) | |
| 47 | BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL); |
| 48 | |
| 49 | if (ip_tunnel_info_af(tun_info) == AF_INET) { |
| 50 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS, |
| 51 | enc_ipv4); |
| 52 | key->enc_ipv4.src = tun_info->key.u.ipv4.dst; |
| 53 | key->enc_ipv4.dst = tun_info->key.u.ipv4.src; |
| 54 | if (key->enc_ipv4.src) |
| 55 | mask->enc_ipv4.src = 0xffffffff; |
| 56 | if (key->enc_ipv4.dst) |
| 57 | mask->enc_ipv4.dst = 0xffffffff; |
| 58 | enc_keys |= BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS); |
| 59 | key->enc_control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; |
| 60 | } else { |
| 61 | memcpy(&key->enc_ipv6.src, &tun_info->key.u.ipv6.dst, |
| 62 | sizeof(struct in6_addr)); |
| 63 | memcpy(&key->enc_ipv6.dst, &tun_info->key.u.ipv6.src, |
| 64 | sizeof(struct in6_addr)); |
| 65 | if (memcmp(p: &key->enc_ipv6.src, q: &in6addr_any, |
| 66 | size: sizeof(struct in6_addr))) |
| 67 | memset(&mask->enc_ipv6.src, 0xff, |
| 68 | sizeof(struct in6_addr)); |
| 69 | if (memcmp(p: &key->enc_ipv6.dst, q: &in6addr_any, |
| 70 | size: sizeof(struct in6_addr))) |
| 71 | memset(&mask->enc_ipv6.dst, 0xff, |
| 72 | sizeof(struct in6_addr)); |
| 73 | enc_keys |= BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS); |
| 74 | key->enc_control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; |
| 75 | } |
| 76 | |
| 77 | match->dissector.used_keys |= enc_keys; |
| 78 | } |
| 79 | |
| 80 | static void nf_flow_rule_vlan_match(struct flow_dissector_key_vlan *key, |
| 81 | struct flow_dissector_key_vlan *mask, |
| 82 | u16 vlan_id, __be16 proto) |
| 83 | { |
| 84 | key->vlan_id = vlan_id; |
| 85 | mask->vlan_id = VLAN_VID_MASK; |
| 86 | key->vlan_tpid = proto; |
| 87 | mask->vlan_tpid = 0xffff; |
| 88 | } |
| 89 | |
| 90 | static int nf_flow_rule_match(struct nf_flow_match *match, |
| 91 | const struct flow_offload_tuple *tuple, |
| 92 | struct dst_entry *other_dst) |
| 93 | { |
| 94 | struct nf_flow_key *mask = &match->mask; |
| 95 | struct nf_flow_key *key = &match->key; |
| 96 | struct ip_tunnel_info *tun_info; |
| 97 | bool vlan_encap = false; |
| 98 | |
| 99 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_META, meta); |
| 100 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_CONTROL, control); |
| 101 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_BASIC, basic); |
| 102 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4); |
| 103 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6); |
| 104 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_TCP, tcp); |
| 105 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_PORTS, tp); |
| 106 | |
| 107 | if (other_dst && other_dst->lwtstate) { |
| 108 | tun_info = lwt_tun_info(lwtstate: other_dst->lwtstate); |
| 109 | nf_flow_rule_lwt_match(match, tun_info); |
| 110 | } |
| 111 | |
| 112 | if (tuple->xmit_type == FLOW_OFFLOAD_XMIT_TC) |
| 113 | key->meta.ingress_ifindex = tuple->tc.iifidx; |
| 114 | else |
| 115 | key->meta.ingress_ifindex = tuple->iifidx; |
| 116 | |
| 117 | mask->meta.ingress_ifindex = 0xffffffff; |
| 118 | |
| 119 | if (tuple->encap_num > 0 && !(tuple->in_vlan_ingress & BIT(0)) && |
| 120 | tuple->encap[0].proto == htons(ETH_P_8021Q)) { |
| 121 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_VLAN, vlan); |
| 122 | nf_flow_rule_vlan_match(key: &key->vlan, mask: &mask->vlan, |
| 123 | vlan_id: tuple->encap[0].id, |
| 124 | proto: tuple->encap[0].proto); |
| 125 | vlan_encap = true; |
| 126 | } |
| 127 | |
| 128 | if (tuple->encap_num > 1 && !(tuple->in_vlan_ingress & BIT(1)) && |
| 129 | tuple->encap[1].proto == htons(ETH_P_8021Q)) { |
| 130 | if (vlan_encap) { |
| 131 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_CVLAN, |
| 132 | cvlan); |
| 133 | nf_flow_rule_vlan_match(key: &key->cvlan, mask: &mask->cvlan, |
| 134 | vlan_id: tuple->encap[1].id, |
| 135 | proto: tuple->encap[1].proto); |
| 136 | } else { |
| 137 | NF_FLOW_DISSECTOR(match, FLOW_DISSECTOR_KEY_VLAN, |
| 138 | vlan); |
| 139 | nf_flow_rule_vlan_match(key: &key->vlan, mask: &mask->vlan, |
| 140 | vlan_id: tuple->encap[1].id, |
| 141 | proto: tuple->encap[1].proto); |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | switch (tuple->l3proto) { |
| 146 | case AF_INET: |
| 147 | key->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; |
| 148 | key->basic.n_proto = htons(ETH_P_IP); |
| 149 | key->ipv4.src = tuple->src_v4.s_addr; |
| 150 | mask->ipv4.src = 0xffffffff; |
| 151 | key->ipv4.dst = tuple->dst_v4.s_addr; |
| 152 | mask->ipv4.dst = 0xffffffff; |
| 153 | break; |
| 154 | case AF_INET6: |
| 155 | key->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; |
| 156 | key->basic.n_proto = htons(ETH_P_IPV6); |
| 157 | key->ipv6.src = tuple->src_v6; |
| 158 | memset(&mask->ipv6.src, 0xff, sizeof(mask->ipv6.src)); |
| 159 | key->ipv6.dst = tuple->dst_v6; |
| 160 | memset(&mask->ipv6.dst, 0xff, sizeof(mask->ipv6.dst)); |
| 161 | break; |
| 162 | default: |
| 163 | return -EOPNOTSUPP; |
| 164 | } |
| 165 | mask->control.addr_type = 0xffff; |
| 166 | match->dissector.used_keys |= BIT_ULL(key->control.addr_type); |
| 167 | mask->basic.n_proto = 0xffff; |
| 168 | |
| 169 | switch (tuple->l4proto) { |
| 170 | case IPPROTO_TCP: |
| 171 | key->tcp.flags = 0; |
| 172 | mask->tcp.flags = cpu_to_be16(be32_to_cpu(TCP_FLAG_RST | TCP_FLAG_FIN) >> 16); |
| 173 | match->dissector.used_keys |= BIT_ULL(FLOW_DISSECTOR_KEY_TCP); |
| 174 | break; |
| 175 | case IPPROTO_UDP: |
| 176 | case IPPROTO_GRE: |
| 177 | break; |
| 178 | default: |
| 179 | return -EOPNOTSUPP; |
| 180 | } |
| 181 | |
| 182 | key->basic.ip_proto = tuple->l4proto; |
| 183 | mask->basic.ip_proto = 0xff; |
| 184 | |
| 185 | match->dissector.used_keys |= BIT_ULL(FLOW_DISSECTOR_KEY_META) | |
| 186 | BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) | |
| 187 | BIT_ULL(FLOW_DISSECTOR_KEY_BASIC); |
| 188 | |
| 189 | switch (tuple->l4proto) { |
| 190 | case IPPROTO_TCP: |
| 191 | case IPPROTO_UDP: |
| 192 | key->tp.src = tuple->src_port; |
| 193 | mask->tp.src = 0xffff; |
| 194 | key->tp.dst = tuple->dst_port; |
| 195 | mask->tp.dst = 0xffff; |
| 196 | |
| 197 | match->dissector.used_keys |= BIT_ULL(FLOW_DISSECTOR_KEY_PORTS); |
| 198 | break; |
| 199 | } |
| 200 | |
| 201 | return 0; |
| 202 | } |
| 203 | |
| 204 | static void flow_offload_mangle(struct flow_action_entry *entry, |
| 205 | enum flow_action_mangle_base htype, u32 offset, |
| 206 | const __be32 *value, const __be32 *mask) |
| 207 | { |
| 208 | entry->id = FLOW_ACTION_MANGLE; |
| 209 | entry->mangle.htype = htype; |
| 210 | entry->mangle.offset = offset; |
| 211 | memcpy(&entry->mangle.mask, mask, sizeof(u32)); |
| 212 | memcpy(&entry->mangle.val, value, sizeof(u32)); |
| 213 | } |
| 214 | |
| 215 | static inline struct flow_action_entry * |
| 216 | flow_action_entry_next(struct nf_flow_rule *flow_rule) |
| 217 | { |
| 218 | int i = flow_rule->rule->action.num_entries++; |
| 219 | |
| 220 | return &flow_rule->rule->action.entries[i]; |
| 221 | } |
| 222 | |
| 223 | static int flow_offload_eth_src(struct net *net, |
| 224 | const struct flow_offload *flow, |
| 225 | enum flow_offload_tuple_dir dir, |
| 226 | struct nf_flow_rule *flow_rule) |
| 227 | { |
| 228 | struct flow_action_entry *entry0 = flow_action_entry_next(flow_rule); |
| 229 | struct flow_action_entry *entry1 = flow_action_entry_next(flow_rule); |
| 230 | const struct flow_offload_tuple *other_tuple, *this_tuple; |
| 231 | struct net_device *dev = NULL; |
| 232 | const unsigned char *addr; |
| 233 | u32 mask, val; |
| 234 | u16 val16; |
| 235 | |
| 236 | this_tuple = &flow->tuplehash[dir].tuple; |
| 237 | |
| 238 | switch (this_tuple->xmit_type) { |
| 239 | case FLOW_OFFLOAD_XMIT_DIRECT: |
| 240 | addr = this_tuple->out.h_source; |
| 241 | break; |
| 242 | case FLOW_OFFLOAD_XMIT_NEIGH: |
| 243 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 244 | dev = dev_get_by_index(net, ifindex: other_tuple->iifidx); |
| 245 | if (!dev) |
| 246 | return -ENOENT; |
| 247 | |
| 248 | addr = dev->dev_addr; |
| 249 | break; |
| 250 | default: |
| 251 | return -EOPNOTSUPP; |
| 252 | } |
| 253 | |
| 254 | mask = ~0xffff0000; |
| 255 | memcpy(&val16, addr, 2); |
| 256 | val = val16 << 16; |
| 257 | flow_offload_mangle(entry: entry0, htype: FLOW_ACT_MANGLE_HDR_TYPE_ETH, offset: 4, |
| 258 | value: &val, mask: &mask); |
| 259 | |
| 260 | mask = ~0xffffffff; |
| 261 | memcpy(&val, addr + 2, 4); |
| 262 | flow_offload_mangle(entry: entry1, htype: FLOW_ACT_MANGLE_HDR_TYPE_ETH, offset: 8, |
| 263 | value: &val, mask: &mask); |
| 264 | |
| 265 | dev_put(dev); |
| 266 | |
| 267 | return 0; |
| 268 | } |
| 269 | |
| 270 | static int flow_offload_eth_dst(struct net *net, |
| 271 | const struct flow_offload *flow, |
| 272 | enum flow_offload_tuple_dir dir, |
| 273 | struct nf_flow_rule *flow_rule) |
| 274 | { |
| 275 | struct flow_action_entry *entry0 = flow_action_entry_next(flow_rule); |
| 276 | struct flow_action_entry *entry1 = flow_action_entry_next(flow_rule); |
| 277 | const struct flow_offload_tuple *other_tuple, *this_tuple; |
| 278 | const struct dst_entry *dst_cache; |
| 279 | unsigned char ha[ETH_ALEN]; |
| 280 | struct neighbour *n; |
| 281 | const void *daddr; |
| 282 | u32 mask, val; |
| 283 | u8 nud_state; |
| 284 | u16 val16; |
| 285 | |
| 286 | this_tuple = &flow->tuplehash[dir].tuple; |
| 287 | |
| 288 | switch (this_tuple->xmit_type) { |
| 289 | case FLOW_OFFLOAD_XMIT_DIRECT: |
| 290 | ether_addr_copy(dst: ha, src: this_tuple->out.h_dest); |
| 291 | break; |
| 292 | case FLOW_OFFLOAD_XMIT_NEIGH: |
| 293 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 294 | daddr = &other_tuple->src_v4; |
| 295 | dst_cache = this_tuple->dst_cache; |
| 296 | n = dst_neigh_lookup(dst: dst_cache, daddr); |
| 297 | if (!n) |
| 298 | return -ENOENT; |
| 299 | |
| 300 | read_lock_bh(&n->lock); |
| 301 | nud_state = n->nud_state; |
| 302 | ether_addr_copy(dst: ha, src: n->ha); |
| 303 | read_unlock_bh(&n->lock); |
| 304 | neigh_release(neigh: n); |
| 305 | |
| 306 | if (!(nud_state & NUD_VALID)) |
| 307 | return -ENOENT; |
| 308 | break; |
| 309 | default: |
| 310 | return -EOPNOTSUPP; |
| 311 | } |
| 312 | |
| 313 | mask = ~0xffffffff; |
| 314 | memcpy(&val, ha, 4); |
| 315 | flow_offload_mangle(entry: entry0, htype: FLOW_ACT_MANGLE_HDR_TYPE_ETH, offset: 0, |
| 316 | value: &val, mask: &mask); |
| 317 | |
| 318 | mask = ~0x0000ffff; |
| 319 | memcpy(&val16, ha + 4, 2); |
| 320 | val = val16; |
| 321 | flow_offload_mangle(entry: entry1, htype: FLOW_ACT_MANGLE_HDR_TYPE_ETH, offset: 4, |
| 322 | value: &val, mask: &mask); |
| 323 | |
| 324 | return 0; |
| 325 | } |
| 326 | |
| 327 | static void flow_offload_ipv4_snat(struct net *net, |
| 328 | const struct flow_offload *flow, |
| 329 | enum flow_offload_tuple_dir dir, |
| 330 | struct nf_flow_rule *flow_rule) |
| 331 | { |
| 332 | struct flow_action_entry *entry = flow_action_entry_next(flow_rule); |
| 333 | u32 mask = ~htonl(0xffffffff); |
| 334 | __be32 addr; |
| 335 | u32 offset; |
| 336 | |
| 337 | switch (dir) { |
| 338 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 339 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v4.s_addr; |
| 340 | offset = offsetof(struct iphdr, saddr); |
| 341 | break; |
| 342 | case FLOW_OFFLOAD_DIR_REPLY: |
| 343 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v4.s_addr; |
| 344 | offset = offsetof(struct iphdr, daddr); |
| 345 | break; |
| 346 | default: |
| 347 | return; |
| 348 | } |
| 349 | |
| 350 | flow_offload_mangle(entry, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP4, offset, |
| 351 | value: &addr, mask: &mask); |
| 352 | } |
| 353 | |
| 354 | static void flow_offload_ipv4_dnat(struct net *net, |
| 355 | const struct flow_offload *flow, |
| 356 | enum flow_offload_tuple_dir dir, |
| 357 | struct nf_flow_rule *flow_rule) |
| 358 | { |
| 359 | struct flow_action_entry *entry = flow_action_entry_next(flow_rule); |
| 360 | u32 mask = ~htonl(0xffffffff); |
| 361 | __be32 addr; |
| 362 | u32 offset; |
| 363 | |
| 364 | switch (dir) { |
| 365 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 366 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v4.s_addr; |
| 367 | offset = offsetof(struct iphdr, daddr); |
| 368 | break; |
| 369 | case FLOW_OFFLOAD_DIR_REPLY: |
| 370 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v4.s_addr; |
| 371 | offset = offsetof(struct iphdr, saddr); |
| 372 | break; |
| 373 | default: |
| 374 | return; |
| 375 | } |
| 376 | |
| 377 | flow_offload_mangle(entry, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP4, offset, |
| 378 | value: &addr, mask: &mask); |
| 379 | } |
| 380 | |
| 381 | static void flow_offload_ipv6_mangle(struct nf_flow_rule *flow_rule, |
| 382 | unsigned int offset, |
| 383 | const __be32 *addr, const __be32 *mask) |
| 384 | { |
| 385 | struct flow_action_entry *entry; |
| 386 | int i; |
| 387 | |
| 388 | for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) { |
| 389 | entry = flow_action_entry_next(flow_rule); |
| 390 | flow_offload_mangle(entry, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP6, |
| 391 | offset: offset + i * sizeof(u32), value: &addr[i], mask); |
| 392 | } |
| 393 | } |
| 394 | |
| 395 | static void flow_offload_ipv6_snat(struct net *net, |
| 396 | const struct flow_offload *flow, |
| 397 | enum flow_offload_tuple_dir dir, |
| 398 | struct nf_flow_rule *flow_rule) |
| 399 | { |
| 400 | u32 mask = ~htonl(0xffffffff); |
| 401 | const __be32 *addr; |
| 402 | u32 offset; |
| 403 | |
| 404 | switch (dir) { |
| 405 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 406 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6.s6_addr32; |
| 407 | offset = offsetof(struct ipv6hdr, saddr); |
| 408 | break; |
| 409 | case FLOW_OFFLOAD_DIR_REPLY: |
| 410 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6.s6_addr32; |
| 411 | offset = offsetof(struct ipv6hdr, daddr); |
| 412 | break; |
| 413 | default: |
| 414 | return; |
| 415 | } |
| 416 | |
| 417 | flow_offload_ipv6_mangle(flow_rule, offset, addr, mask: &mask); |
| 418 | } |
| 419 | |
| 420 | static void flow_offload_ipv6_dnat(struct net *net, |
| 421 | const struct flow_offload *flow, |
| 422 | enum flow_offload_tuple_dir dir, |
| 423 | struct nf_flow_rule *flow_rule) |
| 424 | { |
| 425 | u32 mask = ~htonl(0xffffffff); |
| 426 | const __be32 *addr; |
| 427 | u32 offset; |
| 428 | |
| 429 | switch (dir) { |
| 430 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 431 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6.s6_addr32; |
| 432 | offset = offsetof(struct ipv6hdr, daddr); |
| 433 | break; |
| 434 | case FLOW_OFFLOAD_DIR_REPLY: |
| 435 | addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6.s6_addr32; |
| 436 | offset = offsetof(struct ipv6hdr, saddr); |
| 437 | break; |
| 438 | default: |
| 439 | return; |
| 440 | } |
| 441 | |
| 442 | flow_offload_ipv6_mangle(flow_rule, offset, addr, mask: &mask); |
| 443 | } |
| 444 | |
| 445 | static int flow_offload_l4proto(const struct flow_offload *flow) |
| 446 | { |
| 447 | u8 protonum = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.l4proto; |
| 448 | u8 type = 0; |
| 449 | |
| 450 | switch (protonum) { |
| 451 | case IPPROTO_TCP: |
| 452 | type = FLOW_ACT_MANGLE_HDR_TYPE_TCP; |
| 453 | break; |
| 454 | case IPPROTO_UDP: |
| 455 | type = FLOW_ACT_MANGLE_HDR_TYPE_UDP; |
| 456 | break; |
| 457 | default: |
| 458 | break; |
| 459 | } |
| 460 | |
| 461 | return type; |
| 462 | } |
| 463 | |
| 464 | static void flow_offload_port_snat(struct net *net, |
| 465 | const struct flow_offload *flow, |
| 466 | enum flow_offload_tuple_dir dir, |
| 467 | struct nf_flow_rule *flow_rule) |
| 468 | { |
| 469 | struct flow_action_entry *entry = flow_action_entry_next(flow_rule); |
| 470 | u32 mask, port; |
| 471 | u32 offset; |
| 472 | |
| 473 | switch (dir) { |
| 474 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 475 | port = ntohs(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port); |
| 476 | offset = 0; /* offsetof(struct tcphdr, source); */ |
| 477 | port = htonl(port << 16); |
| 478 | mask = ~htonl(0xffff0000); |
| 479 | break; |
| 480 | case FLOW_OFFLOAD_DIR_REPLY: |
| 481 | port = ntohs(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port); |
| 482 | offset = 0; /* offsetof(struct tcphdr, dest); */ |
| 483 | port = htonl(port); |
| 484 | mask = ~htonl(0xffff); |
| 485 | break; |
| 486 | default: |
| 487 | return; |
| 488 | } |
| 489 | |
| 490 | flow_offload_mangle(entry, htype: flow_offload_l4proto(flow), offset, |
| 491 | value: &port, mask: &mask); |
| 492 | } |
| 493 | |
| 494 | static void flow_offload_port_dnat(struct net *net, |
| 495 | const struct flow_offload *flow, |
| 496 | enum flow_offload_tuple_dir dir, |
| 497 | struct nf_flow_rule *flow_rule) |
| 498 | { |
| 499 | struct flow_action_entry *entry = flow_action_entry_next(flow_rule); |
| 500 | u32 mask, port; |
| 501 | u32 offset; |
| 502 | |
| 503 | switch (dir) { |
| 504 | case FLOW_OFFLOAD_DIR_ORIGINAL: |
| 505 | port = ntohs(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port); |
| 506 | offset = 0; /* offsetof(struct tcphdr, dest); */ |
| 507 | port = htonl(port); |
| 508 | mask = ~htonl(0xffff); |
| 509 | break; |
| 510 | case FLOW_OFFLOAD_DIR_REPLY: |
| 511 | port = ntohs(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port); |
| 512 | offset = 0; /* offsetof(struct tcphdr, source); */ |
| 513 | port = htonl(port << 16); |
| 514 | mask = ~htonl(0xffff0000); |
| 515 | break; |
| 516 | default: |
| 517 | return; |
| 518 | } |
| 519 | |
| 520 | flow_offload_mangle(entry, htype: flow_offload_l4proto(flow), offset, |
| 521 | value: &port, mask: &mask); |
| 522 | } |
| 523 | |
| 524 | static void flow_offload_ipv4_checksum(struct net *net, |
| 525 | const struct flow_offload *flow, |
| 526 | struct nf_flow_rule *flow_rule) |
| 527 | { |
| 528 | u8 protonum = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.l4proto; |
| 529 | struct flow_action_entry *entry = flow_action_entry_next(flow_rule); |
| 530 | |
| 531 | entry->id = FLOW_ACTION_CSUM; |
| 532 | entry->csum_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR; |
| 533 | |
| 534 | switch (protonum) { |
| 535 | case IPPROTO_TCP: |
| 536 | entry->csum_flags |= TCA_CSUM_UPDATE_FLAG_TCP; |
| 537 | break; |
| 538 | case IPPROTO_UDP: |
| 539 | entry->csum_flags |= TCA_CSUM_UPDATE_FLAG_UDP; |
| 540 | break; |
| 541 | } |
| 542 | } |
| 543 | |
| 544 | static void flow_offload_redirect(struct net *net, |
| 545 | const struct flow_offload *flow, |
| 546 | enum flow_offload_tuple_dir dir, |
| 547 | struct nf_flow_rule *flow_rule) |
| 548 | { |
| 549 | const struct flow_offload_tuple *this_tuple, *other_tuple; |
| 550 | struct flow_action_entry *entry; |
| 551 | struct net_device *dev; |
| 552 | int ifindex; |
| 553 | |
| 554 | this_tuple = &flow->tuplehash[dir].tuple; |
| 555 | switch (this_tuple->xmit_type) { |
| 556 | case FLOW_OFFLOAD_XMIT_DIRECT: |
| 557 | this_tuple = &flow->tuplehash[dir].tuple; |
| 558 | ifindex = this_tuple->out.ifidx; |
| 559 | break; |
| 560 | case FLOW_OFFLOAD_XMIT_NEIGH: |
| 561 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 562 | ifindex = other_tuple->iifidx; |
| 563 | break; |
| 564 | default: |
| 565 | return; |
| 566 | } |
| 567 | |
| 568 | dev = dev_get_by_index(net, ifindex); |
| 569 | if (!dev) |
| 570 | return; |
| 571 | |
| 572 | entry = flow_action_entry_next(flow_rule); |
| 573 | entry->id = FLOW_ACTION_REDIRECT; |
| 574 | entry->dev = dev; |
| 575 | } |
| 576 | |
| 577 | static void flow_offload_encap_tunnel(const struct flow_offload *flow, |
| 578 | enum flow_offload_tuple_dir dir, |
| 579 | struct nf_flow_rule *flow_rule) |
| 580 | { |
| 581 | const struct flow_offload_tuple *this_tuple; |
| 582 | struct flow_action_entry *entry; |
| 583 | struct dst_entry *dst; |
| 584 | |
| 585 | this_tuple = &flow->tuplehash[dir].tuple; |
| 586 | if (this_tuple->xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) |
| 587 | return; |
| 588 | |
| 589 | dst = this_tuple->dst_cache; |
| 590 | if (dst && dst->lwtstate) { |
| 591 | struct ip_tunnel_info *tun_info; |
| 592 | |
| 593 | tun_info = lwt_tun_info(lwtstate: dst->lwtstate); |
| 594 | if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX)) { |
| 595 | entry = flow_action_entry_next(flow_rule); |
| 596 | entry->id = FLOW_ACTION_TUNNEL_ENCAP; |
| 597 | entry->tunnel = tun_info; |
| 598 | } |
| 599 | } |
| 600 | } |
| 601 | |
| 602 | static void flow_offload_decap_tunnel(const struct flow_offload *flow, |
| 603 | enum flow_offload_tuple_dir dir, |
| 604 | struct nf_flow_rule *flow_rule) |
| 605 | { |
| 606 | const struct flow_offload_tuple *other_tuple; |
| 607 | struct flow_action_entry *entry; |
| 608 | struct dst_entry *dst; |
| 609 | |
| 610 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 611 | if (other_tuple->xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) |
| 612 | return; |
| 613 | |
| 614 | dst = other_tuple->dst_cache; |
| 615 | if (dst && dst->lwtstate) { |
| 616 | struct ip_tunnel_info *tun_info; |
| 617 | |
| 618 | tun_info = lwt_tun_info(lwtstate: dst->lwtstate); |
| 619 | if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX)) { |
| 620 | entry = flow_action_entry_next(flow_rule); |
| 621 | entry->id = FLOW_ACTION_TUNNEL_DECAP; |
| 622 | } |
| 623 | } |
| 624 | } |
| 625 | |
| 626 | static int |
| 627 | nf_flow_rule_route_common(struct net *net, const struct flow_offload *flow, |
| 628 | enum flow_offload_tuple_dir dir, |
| 629 | struct nf_flow_rule *flow_rule) |
| 630 | { |
| 631 | const struct flow_offload_tuple *other_tuple; |
| 632 | const struct flow_offload_tuple *tuple; |
| 633 | int i; |
| 634 | |
| 635 | flow_offload_decap_tunnel(flow, dir, flow_rule); |
| 636 | flow_offload_encap_tunnel(flow, dir, flow_rule); |
| 637 | |
| 638 | if (flow_offload_eth_src(net, flow, dir, flow_rule) < 0 || |
| 639 | flow_offload_eth_dst(net, flow, dir, flow_rule) < 0) |
| 640 | return -1; |
| 641 | |
| 642 | tuple = &flow->tuplehash[dir].tuple; |
| 643 | |
| 644 | for (i = 0; i < tuple->encap_num; i++) { |
| 645 | struct flow_action_entry *entry; |
| 646 | |
| 647 | if (tuple->in_vlan_ingress & BIT(i)) |
| 648 | continue; |
| 649 | |
| 650 | if (tuple->encap[i].proto == htons(ETH_P_8021Q)) { |
| 651 | entry = flow_action_entry_next(flow_rule); |
| 652 | entry->id = FLOW_ACTION_VLAN_POP; |
| 653 | } |
| 654 | } |
| 655 | |
| 656 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 657 | |
| 658 | for (i = 0; i < other_tuple->encap_num; i++) { |
| 659 | struct flow_action_entry *entry; |
| 660 | |
| 661 | if (other_tuple->in_vlan_ingress & BIT(i)) |
| 662 | continue; |
| 663 | |
| 664 | entry = flow_action_entry_next(flow_rule); |
| 665 | |
| 666 | switch (other_tuple->encap[i].proto) { |
| 667 | case htons(ETH_P_PPP_SES): |
| 668 | entry->id = FLOW_ACTION_PPPOE_PUSH; |
| 669 | entry->pppoe.sid = other_tuple->encap[i].id; |
| 670 | break; |
| 671 | case htons(ETH_P_8021Q): |
| 672 | entry->id = FLOW_ACTION_VLAN_PUSH; |
| 673 | entry->vlan.vid = other_tuple->encap[i].id; |
| 674 | entry->vlan.proto = other_tuple->encap[i].proto; |
| 675 | break; |
| 676 | } |
| 677 | } |
| 678 | |
| 679 | return 0; |
| 680 | } |
| 681 | |
| 682 | int nf_flow_rule_route_ipv4(struct net *net, struct flow_offload *flow, |
| 683 | enum flow_offload_tuple_dir dir, |
| 684 | struct nf_flow_rule *flow_rule) |
| 685 | { |
| 686 | if (nf_flow_rule_route_common(net, flow, dir, flow_rule) < 0) |
| 687 | return -1; |
| 688 | |
| 689 | if (test_bit(NF_FLOW_SNAT, &flow->flags)) { |
| 690 | flow_offload_ipv4_snat(net, flow, dir, flow_rule); |
| 691 | flow_offload_port_snat(net, flow, dir, flow_rule); |
| 692 | } |
| 693 | if (test_bit(NF_FLOW_DNAT, &flow->flags)) { |
| 694 | flow_offload_ipv4_dnat(net, flow, dir, flow_rule); |
| 695 | flow_offload_port_dnat(net, flow, dir, flow_rule); |
| 696 | } |
| 697 | if (test_bit(NF_FLOW_SNAT, &flow->flags) || |
| 698 | test_bit(NF_FLOW_DNAT, &flow->flags)) |
| 699 | flow_offload_ipv4_checksum(net, flow, flow_rule); |
| 700 | |
| 701 | flow_offload_redirect(net, flow, dir, flow_rule); |
| 702 | |
| 703 | return 0; |
| 704 | } |
| 705 | EXPORT_SYMBOL_GPL(nf_flow_rule_route_ipv4); |
| 706 | |
| 707 | int nf_flow_rule_route_ipv6(struct net *net, struct flow_offload *flow, |
| 708 | enum flow_offload_tuple_dir dir, |
| 709 | struct nf_flow_rule *flow_rule) |
| 710 | { |
| 711 | if (nf_flow_rule_route_common(net, flow, dir, flow_rule) < 0) |
| 712 | return -1; |
| 713 | |
| 714 | if (test_bit(NF_FLOW_SNAT, &flow->flags)) { |
| 715 | flow_offload_ipv6_snat(net, flow, dir, flow_rule); |
| 716 | flow_offload_port_snat(net, flow, dir, flow_rule); |
| 717 | } |
| 718 | if (test_bit(NF_FLOW_DNAT, &flow->flags)) { |
| 719 | flow_offload_ipv6_dnat(net, flow, dir, flow_rule); |
| 720 | flow_offload_port_dnat(net, flow, dir, flow_rule); |
| 721 | } |
| 722 | |
| 723 | flow_offload_redirect(net, flow, dir, flow_rule); |
| 724 | |
| 725 | return 0; |
| 726 | } |
| 727 | EXPORT_SYMBOL_GPL(nf_flow_rule_route_ipv6); |
| 728 | |
| 729 | #define NF_FLOW_RULE_ACTION_MAX 16 |
| 730 | |
| 731 | static struct nf_flow_rule * |
| 732 | nf_flow_offload_rule_alloc(struct net *net, |
| 733 | const struct flow_offload_work *offload, |
| 734 | enum flow_offload_tuple_dir dir) |
| 735 | { |
| 736 | const struct nf_flowtable *flowtable = offload->flowtable; |
| 737 | const struct flow_offload_tuple *tuple, *other_tuple; |
| 738 | struct flow_offload *flow = offload->flow; |
| 739 | struct dst_entry *other_dst = NULL; |
| 740 | struct nf_flow_rule *flow_rule; |
| 741 | int err = -ENOMEM; |
| 742 | |
| 743 | flow_rule = kzalloc(sizeof(*flow_rule), GFP_KERNEL); |
| 744 | if (!flow_rule) |
| 745 | goto err_flow; |
| 746 | |
| 747 | flow_rule->rule = flow_rule_alloc(NF_FLOW_RULE_ACTION_MAX); |
| 748 | if (!flow_rule->rule) |
| 749 | goto err_flow_rule; |
| 750 | |
| 751 | flow_rule->rule->match.dissector = &flow_rule->match.dissector; |
| 752 | flow_rule->rule->match.mask = &flow_rule->match.mask; |
| 753 | flow_rule->rule->match.key = &flow_rule->match.key; |
| 754 | |
| 755 | tuple = &flow->tuplehash[dir].tuple; |
| 756 | other_tuple = &flow->tuplehash[!dir].tuple; |
| 757 | if (other_tuple->xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) |
| 758 | other_dst = other_tuple->dst_cache; |
| 759 | |
| 760 | err = nf_flow_rule_match(match: &flow_rule->match, tuple, other_dst); |
| 761 | if (err < 0) |
| 762 | goto err_flow_match; |
| 763 | |
| 764 | flow_rule->rule->action.num_entries = 0; |
| 765 | if (flowtable->type->action(net, flow, dir, flow_rule) < 0) |
| 766 | goto err_flow_match; |
| 767 | |
| 768 | return flow_rule; |
| 769 | |
| 770 | err_flow_match: |
| 771 | kfree(objp: flow_rule->rule); |
| 772 | err_flow_rule: |
| 773 | kfree(objp: flow_rule); |
| 774 | err_flow: |
| 775 | return NULL; |
| 776 | } |
| 777 | |
| 778 | static void __nf_flow_offload_destroy(struct nf_flow_rule *flow_rule) |
| 779 | { |
| 780 | struct flow_action_entry *entry; |
| 781 | int i; |
| 782 | |
| 783 | for (i = 0; i < flow_rule->rule->action.num_entries; i++) { |
| 784 | entry = &flow_rule->rule->action.entries[i]; |
| 785 | if (entry->id != FLOW_ACTION_REDIRECT) |
| 786 | continue; |
| 787 | |
| 788 | dev_put(dev: entry->dev); |
| 789 | } |
| 790 | kfree(objp: flow_rule->rule); |
| 791 | kfree(objp: flow_rule); |
| 792 | } |
| 793 | |
| 794 | static void nf_flow_offload_destroy(struct nf_flow_rule *flow_rule[]) |
| 795 | { |
| 796 | int i; |
| 797 | |
| 798 | for (i = 0; i < FLOW_OFFLOAD_DIR_MAX; i++) |
| 799 | __nf_flow_offload_destroy(flow_rule: flow_rule[i]); |
| 800 | } |
| 801 | |
| 802 | static int nf_flow_offload_alloc(const struct flow_offload_work *offload, |
| 803 | struct nf_flow_rule *flow_rule[]) |
| 804 | { |
| 805 | struct net *net = read_pnet(pnet: &offload->flowtable->net); |
| 806 | |
| 807 | flow_rule[0] = nf_flow_offload_rule_alloc(net, offload, |
| 808 | dir: FLOW_OFFLOAD_DIR_ORIGINAL); |
| 809 | if (!flow_rule[0]) |
| 810 | return -ENOMEM; |
| 811 | |
| 812 | flow_rule[1] = nf_flow_offload_rule_alloc(net, offload, |
| 813 | dir: FLOW_OFFLOAD_DIR_REPLY); |
| 814 | if (!flow_rule[1]) { |
| 815 | __nf_flow_offload_destroy(flow_rule: flow_rule[0]); |
| 816 | return -ENOMEM; |
| 817 | } |
| 818 | |
| 819 | return 0; |
| 820 | } |
| 821 | |
| 822 | static void nf_flow_offload_init(struct flow_cls_offload *cls_flow, |
| 823 | __be16 proto, int priority, |
| 824 | enum flow_cls_command cmd, |
| 825 | const struct flow_offload_tuple *tuple, |
| 826 | struct netlink_ext_ack *extack) |
| 827 | { |
| 828 | cls_flow->common.protocol = proto; |
| 829 | cls_flow->common.prio = priority; |
| 830 | cls_flow->common.extack = extack; |
| 831 | cls_flow->command = cmd; |
| 832 | cls_flow->cookie = (unsigned long)tuple; |
| 833 | } |
| 834 | |
| 835 | static int nf_flow_offload_tuple(struct nf_flowtable *flowtable, |
| 836 | struct flow_offload *flow, |
| 837 | struct nf_flow_rule *flow_rule, |
| 838 | enum flow_offload_tuple_dir dir, |
| 839 | int priority, int cmd, |
| 840 | struct flow_stats *stats, |
| 841 | struct list_head *block_cb_list) |
| 842 | { |
| 843 | struct flow_cls_offload cls_flow = {}; |
| 844 | struct netlink_ext_ack extack = {}; |
| 845 | struct flow_block_cb *block_cb; |
| 846 | __be16 proto = ETH_P_ALL; |
| 847 | int err, i = 0; |
| 848 | |
| 849 | nf_flow_offload_init(cls_flow: &cls_flow, proto, priority, cmd, |
| 850 | tuple: &flow->tuplehash[dir].tuple, extack: &extack); |
| 851 | if (cmd == FLOW_CLS_REPLACE) |
| 852 | cls_flow.rule = flow_rule->rule; |
| 853 | |
| 854 | down_read(sem: &flowtable->flow_block_lock); |
| 855 | list_for_each_entry(block_cb, block_cb_list, list) { |
| 856 | err = block_cb->cb(TC_SETUP_CLSFLOWER, &cls_flow, |
| 857 | block_cb->cb_priv); |
| 858 | if (err < 0) |
| 859 | continue; |
| 860 | |
| 861 | i++; |
| 862 | } |
| 863 | up_read(sem: &flowtable->flow_block_lock); |
| 864 | |
| 865 | if (cmd == FLOW_CLS_STATS) |
| 866 | memcpy(stats, &cls_flow.stats, sizeof(*stats)); |
| 867 | |
| 868 | return i; |
| 869 | } |
| 870 | |
| 871 | static int flow_offload_tuple_add(struct flow_offload_work *offload, |
| 872 | struct nf_flow_rule *flow_rule, |
| 873 | enum flow_offload_tuple_dir dir) |
| 874 | { |
| 875 | return nf_flow_offload_tuple(flowtable: offload->flowtable, flow: offload->flow, |
| 876 | flow_rule, dir, |
| 877 | priority: offload->flowtable->priority, |
| 878 | cmd: FLOW_CLS_REPLACE, NULL, |
| 879 | block_cb_list: &offload->flowtable->flow_block.cb_list); |
| 880 | } |
| 881 | |
| 882 | static void flow_offload_tuple_del(struct flow_offload_work *offload, |
| 883 | enum flow_offload_tuple_dir dir) |
| 884 | { |
| 885 | nf_flow_offload_tuple(flowtable: offload->flowtable, flow: offload->flow, NULL, dir, |
| 886 | priority: offload->flowtable->priority, |
| 887 | cmd: FLOW_CLS_DESTROY, NULL, |
| 888 | block_cb_list: &offload->flowtable->flow_block.cb_list); |
| 889 | } |
| 890 | |
| 891 | static int flow_offload_rule_add(struct flow_offload_work *offload, |
| 892 | struct nf_flow_rule *flow_rule[]) |
| 893 | { |
| 894 | int ok_count = 0; |
| 895 | |
| 896 | ok_count += flow_offload_tuple_add(offload, flow_rule: flow_rule[0], |
| 897 | dir: FLOW_OFFLOAD_DIR_ORIGINAL); |
| 898 | if (test_bit(NF_FLOW_HW_BIDIRECTIONAL, &offload->flow->flags)) |
| 899 | ok_count += flow_offload_tuple_add(offload, flow_rule: flow_rule[1], |
| 900 | dir: FLOW_OFFLOAD_DIR_REPLY); |
| 901 | if (ok_count == 0) |
| 902 | return -ENOENT; |
| 903 | |
| 904 | return 0; |
| 905 | } |
| 906 | |
| 907 | static void flow_offload_work_add(struct flow_offload_work *offload) |
| 908 | { |
| 909 | struct nf_flow_rule *flow_rule[FLOW_OFFLOAD_DIR_MAX]; |
| 910 | int err; |
| 911 | |
| 912 | err = nf_flow_offload_alloc(offload, flow_rule); |
| 913 | if (err < 0) |
| 914 | return; |
| 915 | |
| 916 | err = flow_offload_rule_add(offload, flow_rule); |
| 917 | if (err < 0) |
| 918 | goto out; |
| 919 | |
| 920 | set_bit(nr: IPS_HW_OFFLOAD_BIT, addr: &offload->flow->ct->status); |
| 921 | |
| 922 | out: |
| 923 | nf_flow_offload_destroy(flow_rule); |
| 924 | } |
| 925 | |
| 926 | static void flow_offload_work_del(struct flow_offload_work *offload) |
| 927 | { |
| 928 | clear_bit(nr: IPS_HW_OFFLOAD_BIT, addr: &offload->flow->ct->status); |
| 929 | flow_offload_tuple_del(offload, dir: FLOW_OFFLOAD_DIR_ORIGINAL); |
| 930 | if (test_bit(NF_FLOW_HW_BIDIRECTIONAL, &offload->flow->flags)) |
| 931 | flow_offload_tuple_del(offload, dir: FLOW_OFFLOAD_DIR_REPLY); |
| 932 | set_bit(nr: NF_FLOW_HW_DEAD, addr: &offload->flow->flags); |
| 933 | } |
| 934 | |
| 935 | static void flow_offload_tuple_stats(struct flow_offload_work *offload, |
| 936 | enum flow_offload_tuple_dir dir, |
| 937 | struct flow_stats *stats) |
| 938 | { |
| 939 | nf_flow_offload_tuple(flowtable: offload->flowtable, flow: offload->flow, NULL, dir, |
| 940 | priority: offload->flowtable->priority, |
| 941 | cmd: FLOW_CLS_STATS, stats, |
| 942 | block_cb_list: &offload->flowtable->flow_block.cb_list); |
| 943 | } |
| 944 | |
| 945 | static void flow_offload_work_stats(struct flow_offload_work *offload) |
| 946 | { |
| 947 | struct flow_stats stats[FLOW_OFFLOAD_DIR_MAX] = {}; |
| 948 | u64 lastused; |
| 949 | |
| 950 | flow_offload_tuple_stats(offload, dir: FLOW_OFFLOAD_DIR_ORIGINAL, stats: &stats[0]); |
| 951 | if (test_bit(NF_FLOW_HW_BIDIRECTIONAL, &offload->flow->flags)) |
| 952 | flow_offload_tuple_stats(offload, dir: FLOW_OFFLOAD_DIR_REPLY, |
| 953 | stats: &stats[1]); |
| 954 | |
| 955 | lastused = max_t(u64, stats[0].lastused, stats[1].lastused); |
| 956 | offload->flow->timeout = max_t(u64, offload->flow->timeout, |
| 957 | lastused + flow_offload_get_timeout(offload->flow)); |
| 958 | |
| 959 | if (offload->flowtable->flags & NF_FLOWTABLE_COUNTER) { |
| 960 | if (stats[0].pkts) |
| 961 | nf_ct_acct_add(ct: offload->flow->ct, |
| 962 | dir: FLOW_OFFLOAD_DIR_ORIGINAL, |
| 963 | packets: stats[0].pkts, bytes: stats[0].bytes); |
| 964 | if (stats[1].pkts) |
| 965 | nf_ct_acct_add(ct: offload->flow->ct, |
| 966 | dir: FLOW_OFFLOAD_DIR_REPLY, |
| 967 | packets: stats[1].pkts, bytes: stats[1].bytes); |
| 968 | } |
| 969 | } |
| 970 | |
| 971 | static void flow_offload_work_handler(struct work_struct *work) |
| 972 | { |
| 973 | struct flow_offload_work *offload; |
| 974 | struct net *net; |
| 975 | |
| 976 | offload = container_of(work, struct flow_offload_work, work); |
| 977 | net = read_pnet(pnet: &offload->flowtable->net); |
| 978 | switch (offload->cmd) { |
| 979 | case FLOW_CLS_REPLACE: |
| 980 | flow_offload_work_add(offload); |
| 981 | NF_FLOW_TABLE_STAT_DEC_ATOMIC(net, count_wq_add); |
| 982 | break; |
| 983 | case FLOW_CLS_DESTROY: |
| 984 | flow_offload_work_del(offload); |
| 985 | NF_FLOW_TABLE_STAT_DEC_ATOMIC(net, count_wq_del); |
| 986 | break; |
| 987 | case FLOW_CLS_STATS: |
| 988 | flow_offload_work_stats(offload); |
| 989 | NF_FLOW_TABLE_STAT_DEC_ATOMIC(net, count_wq_stats); |
| 990 | break; |
| 991 | default: |
| 992 | WARN_ON_ONCE(1); |
| 993 | } |
| 994 | |
| 995 | clear_bit(nr: NF_FLOW_HW_PENDING, addr: &offload->flow->flags); |
| 996 | kfree(objp: offload); |
| 997 | } |
| 998 | |
| 999 | static void flow_offload_queue_work(struct flow_offload_work *offload) |
| 1000 | { |
| 1001 | struct net *net = read_pnet(pnet: &offload->flowtable->net); |
| 1002 | |
| 1003 | if (offload->cmd == FLOW_CLS_REPLACE) { |
| 1004 | NF_FLOW_TABLE_STAT_INC_ATOMIC(net, count_wq_add); |
| 1005 | queue_work(wq: nf_flow_offload_add_wq, work: &offload->work); |
| 1006 | } else if (offload->cmd == FLOW_CLS_DESTROY) { |
| 1007 | NF_FLOW_TABLE_STAT_INC_ATOMIC(net, count_wq_del); |
| 1008 | queue_work(wq: nf_flow_offload_del_wq, work: &offload->work); |
| 1009 | } else { |
| 1010 | NF_FLOW_TABLE_STAT_INC_ATOMIC(net, count_wq_stats); |
| 1011 | queue_work(wq: nf_flow_offload_stats_wq, work: &offload->work); |
| 1012 | } |
| 1013 | } |
| 1014 | |
| 1015 | static struct flow_offload_work * |
| 1016 | nf_flow_offload_work_alloc(struct nf_flowtable *flowtable, |
| 1017 | struct flow_offload *flow, unsigned int cmd) |
| 1018 | { |
| 1019 | struct flow_offload_work *offload; |
| 1020 | |
| 1021 | if (test_and_set_bit(nr: NF_FLOW_HW_PENDING, addr: &flow->flags)) |
| 1022 | return NULL; |
| 1023 | |
| 1024 | offload = kmalloc(sizeof(struct flow_offload_work), GFP_ATOMIC); |
| 1025 | if (!offload) { |
| 1026 | clear_bit(nr: NF_FLOW_HW_PENDING, addr: &flow->flags); |
| 1027 | return NULL; |
| 1028 | } |
| 1029 | |
| 1030 | offload->cmd = cmd; |
| 1031 | offload->flow = flow; |
| 1032 | offload->flowtable = flowtable; |
| 1033 | INIT_WORK(&offload->work, flow_offload_work_handler); |
| 1034 | |
| 1035 | return offload; |
| 1036 | } |
| 1037 | |
| 1038 | |
| 1039 | void nf_flow_offload_add(struct nf_flowtable *flowtable, |
| 1040 | struct flow_offload *flow) |
| 1041 | { |
| 1042 | struct flow_offload_work *offload; |
| 1043 | |
| 1044 | offload = nf_flow_offload_work_alloc(flowtable, flow, cmd: FLOW_CLS_REPLACE); |
| 1045 | if (!offload) |
| 1046 | return; |
| 1047 | |
| 1048 | flow_offload_queue_work(offload); |
| 1049 | } |
| 1050 | |
| 1051 | void nf_flow_offload_del(struct nf_flowtable *flowtable, |
| 1052 | struct flow_offload *flow) |
| 1053 | { |
| 1054 | struct flow_offload_work *offload; |
| 1055 | |
| 1056 | offload = nf_flow_offload_work_alloc(flowtable, flow, cmd: FLOW_CLS_DESTROY); |
| 1057 | if (!offload) |
| 1058 | return; |
| 1059 | |
| 1060 | set_bit(nr: NF_FLOW_HW_DYING, addr: &flow->flags); |
| 1061 | flow_offload_queue_work(offload); |
| 1062 | } |
| 1063 | |
| 1064 | void nf_flow_offload_stats(struct nf_flowtable *flowtable, |
| 1065 | struct flow_offload *flow) |
| 1066 | { |
| 1067 | struct flow_offload_work *offload; |
| 1068 | __s32 delta; |
| 1069 | |
| 1070 | delta = nf_flow_timeout_delta(timeout: flow->timeout); |
| 1071 | if ((delta >= (9 * flow_offload_get_timeout(flow)) / 10)) |
| 1072 | return; |
| 1073 | |
| 1074 | offload = nf_flow_offload_work_alloc(flowtable, flow, cmd: FLOW_CLS_STATS); |
| 1075 | if (!offload) |
| 1076 | return; |
| 1077 | |
| 1078 | flow_offload_queue_work(offload); |
| 1079 | } |
| 1080 | |
| 1081 | void nf_flow_table_offload_flush_cleanup(struct nf_flowtable *flowtable) |
| 1082 | { |
| 1083 | if (nf_flowtable_hw_offload(flowtable)) { |
| 1084 | flush_workqueue(nf_flow_offload_del_wq); |
| 1085 | nf_flow_table_gc_run(flow_table: flowtable); |
| 1086 | } |
| 1087 | } |
| 1088 | |
| 1089 | void nf_flow_table_offload_flush(struct nf_flowtable *flowtable) |
| 1090 | { |
| 1091 | if (nf_flowtable_hw_offload(flowtable)) { |
| 1092 | flush_workqueue(nf_flow_offload_add_wq); |
| 1093 | flush_workqueue(nf_flow_offload_del_wq); |
| 1094 | flush_workqueue(nf_flow_offload_stats_wq); |
| 1095 | } |
| 1096 | } |
| 1097 | |
| 1098 | static int nf_flow_table_block_setup(struct nf_flowtable *flowtable, |
| 1099 | struct flow_block_offload *bo, |
| 1100 | enum flow_block_command cmd) |
| 1101 | { |
| 1102 | struct flow_block_cb *block_cb, *next; |
| 1103 | int err = 0; |
| 1104 | |
| 1105 | down_write(sem: &flowtable->flow_block_lock); |
| 1106 | switch (cmd) { |
| 1107 | case FLOW_BLOCK_BIND: |
| 1108 | list_splice(list: &bo->cb_list, head: &flowtable->flow_block.cb_list); |
| 1109 | break; |
| 1110 | case FLOW_BLOCK_UNBIND: |
| 1111 | list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) { |
| 1112 | list_del(entry: &block_cb->list); |
| 1113 | flow_block_cb_free(block_cb); |
| 1114 | } |
| 1115 | break; |
| 1116 | default: |
| 1117 | WARN_ON_ONCE(1); |
| 1118 | err = -EOPNOTSUPP; |
| 1119 | } |
| 1120 | up_write(sem: &flowtable->flow_block_lock); |
| 1121 | |
| 1122 | return err; |
| 1123 | } |
| 1124 | |
| 1125 | static void nf_flow_table_block_offload_init(struct flow_block_offload *bo, |
| 1126 | struct net *net, |
| 1127 | enum flow_block_command cmd, |
| 1128 | struct nf_flowtable *flowtable, |
| 1129 | struct netlink_ext_ack *extack) |
| 1130 | { |
| 1131 | memset(bo, 0, sizeof(*bo)); |
| 1132 | bo->net = net; |
| 1133 | bo->block = &flowtable->flow_block; |
| 1134 | bo->command = cmd; |
| 1135 | bo->binder_type = FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS; |
| 1136 | bo->extack = extack; |
| 1137 | bo->cb_list_head = &flowtable->flow_block.cb_list; |
| 1138 | INIT_LIST_HEAD(list: &bo->cb_list); |
| 1139 | } |
| 1140 | |
| 1141 | static void nf_flow_table_indr_cleanup(struct flow_block_cb *block_cb) |
| 1142 | { |
| 1143 | struct nf_flowtable *flowtable = block_cb->indr.data; |
| 1144 | struct net_device *dev = block_cb->indr.dev; |
| 1145 | |
| 1146 | nf_flow_table_gc_cleanup(flowtable, dev); |
| 1147 | down_write(sem: &flowtable->flow_block_lock); |
| 1148 | list_del(entry: &block_cb->list); |
| 1149 | list_del(entry: &block_cb->driver_list); |
| 1150 | flow_block_cb_free(block_cb); |
| 1151 | up_write(sem: &flowtable->flow_block_lock); |
| 1152 | } |
| 1153 | |
| 1154 | static int nf_flow_table_indr_offload_cmd(struct flow_block_offload *bo, |
| 1155 | struct nf_flowtable *flowtable, |
| 1156 | struct net_device *dev, |
| 1157 | enum flow_block_command cmd, |
| 1158 | struct netlink_ext_ack *extack) |
| 1159 | { |
| 1160 | nf_flow_table_block_offload_init(bo, net: dev_net(dev), cmd, flowtable, |
| 1161 | extack); |
| 1162 | |
| 1163 | return flow_indr_dev_setup_offload(dev, NULL, type: TC_SETUP_FT, data: flowtable, bo, |
| 1164 | cleanup: nf_flow_table_indr_cleanup); |
| 1165 | } |
| 1166 | |
| 1167 | static int nf_flow_table_offload_cmd(struct flow_block_offload *bo, |
| 1168 | struct nf_flowtable *flowtable, |
| 1169 | struct net_device *dev, |
| 1170 | enum flow_block_command cmd, |
| 1171 | struct netlink_ext_ack *extack) |
| 1172 | { |
| 1173 | int err; |
| 1174 | |
| 1175 | nf_flow_table_block_offload_init(bo, net: dev_net(dev), cmd, flowtable, |
| 1176 | extack); |
| 1177 | down_write(sem: &flowtable->flow_block_lock); |
| 1178 | err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_FT, bo); |
| 1179 | up_write(sem: &flowtable->flow_block_lock); |
| 1180 | if (err < 0) |
| 1181 | return err; |
| 1182 | |
| 1183 | return 0; |
| 1184 | } |
| 1185 | |
| 1186 | int nf_flow_table_offload_setup(struct nf_flowtable *flowtable, |
| 1187 | struct net_device *dev, |
| 1188 | enum flow_block_command cmd) |
| 1189 | { |
| 1190 | struct netlink_ext_ack extack = {}; |
| 1191 | struct flow_block_offload bo; |
| 1192 | int err; |
| 1193 | |
| 1194 | if (!nf_flowtable_hw_offload(flowtable)) |
| 1195 | return nf_flow_offload_xdp_setup(flowtable, dev, cmd); |
| 1196 | |
| 1197 | if (dev->netdev_ops->ndo_setup_tc) |
| 1198 | err = nf_flow_table_offload_cmd(bo: &bo, flowtable, dev, cmd, |
| 1199 | extack: &extack); |
| 1200 | else |
| 1201 | err = nf_flow_table_indr_offload_cmd(bo: &bo, flowtable, dev, cmd, |
| 1202 | extack: &extack); |
| 1203 | if (err < 0) |
| 1204 | return err; |
| 1205 | |
| 1206 | return nf_flow_table_block_setup(flowtable, bo: &bo, cmd); |
| 1207 | } |
| 1208 | EXPORT_SYMBOL_GPL(nf_flow_table_offload_setup); |
| 1209 | |
| 1210 | int nf_flow_table_offload_init(void) |
| 1211 | { |
| 1212 | nf_flow_offload_add_wq = alloc_workqueue("nf_ft_offload_add" , |
| 1213 | WQ_UNBOUND | WQ_SYSFS, 0); |
| 1214 | if (!nf_flow_offload_add_wq) |
| 1215 | return -ENOMEM; |
| 1216 | |
| 1217 | nf_flow_offload_del_wq = alloc_workqueue("nf_ft_offload_del" , |
| 1218 | WQ_UNBOUND | WQ_SYSFS, 0); |
| 1219 | if (!nf_flow_offload_del_wq) |
| 1220 | goto err_del_wq; |
| 1221 | |
| 1222 | nf_flow_offload_stats_wq = alloc_workqueue("nf_ft_offload_stats" , |
| 1223 | WQ_UNBOUND | WQ_SYSFS, 0); |
| 1224 | if (!nf_flow_offload_stats_wq) |
| 1225 | goto err_stats_wq; |
| 1226 | |
| 1227 | return 0; |
| 1228 | |
| 1229 | err_stats_wq: |
| 1230 | destroy_workqueue(wq: nf_flow_offload_del_wq); |
| 1231 | err_del_wq: |
| 1232 | destroy_workqueue(wq: nf_flow_offload_add_wq); |
| 1233 | return -ENOMEM; |
| 1234 | } |
| 1235 | |
| 1236 | void nf_flow_table_offload_exit(void) |
| 1237 | { |
| 1238 | destroy_workqueue(wq: nf_flow_offload_add_wq); |
| 1239 | destroy_workqueue(wq: nf_flow_offload_del_wq); |
| 1240 | destroy_workqueue(wq: nf_flow_offload_stats_wq); |
| 1241 | } |
| 1242 | |