1 // SPDX-License-Identifier: GPL-2.0+ 2 /* net/sched/act_ctinfo.c netfilter ctinfo connmark actions 3 * 4 * Copyright (c) 2019 Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk> 5 */ 6 7 #include <linux/module.h> 8 #include <linux/init.h> 9 #include <linux/kernel.h> 10 #include <linux/skbuff.h> 11 #include <linux/rtnetlink.h> 12 #include <linux/pkt_cls.h> 13 #include <linux/ip.h> 14 #include <linux/ipv6.h> 15 #include <net/netlink.h> 16 #include <net/pkt_sched.h> 17 #include <net/act_api.h> 18 #include <net/pkt_cls.h> 19 #include <uapi/linux/tc_act/tc_ctinfo.h> 20 #include <net/tc_act/tc_ctinfo.h> 21 22 #include <net/netfilter/nf_conntrack.h> 23 #include <net/netfilter/nf_conntrack_core.h> 24 #include <net/netfilter/nf_conntrack_ecache.h> 25 #include <net/netfilter/nf_conntrack_zones.h> 26 27 static struct tc_action_ops act_ctinfo_ops; 28 static unsigned int ctinfo_net_id; 29 30 static void tcf_ctinfo_dscp_set(struct nf_conn *ct, struct tcf_ctinfo *ca, 31 struct tcf_ctinfo_params *cp, 32 struct sk_buff *skb, int wlen, int proto) 33 { 34 u8 dscp, newdscp; 35 36 newdscp = (((ct->mark & cp->dscpmask) >> cp->dscpmaskshift) << 2) & 37 ~INET_ECN_MASK; 38 39 switch (proto) { 40 case NFPROTO_IPV4: 41 dscp = ipv4_get_dsfield(ip_hdr(skb)) & ~INET_ECN_MASK; 42 if (dscp != newdscp) { 43 if (likely(!skb_try_make_writable(skb, wlen))) { 44 ipv4_change_dsfield(ip_hdr(skb), 45 INET_ECN_MASK, 46 newdscp); 47 ca->stats_dscp_set++; 48 } else { 49 ca->stats_dscp_error++; 50 } 51 } 52 break; 53 case NFPROTO_IPV6: 54 dscp = ipv6_get_dsfield(ipv6_hdr(skb)) & ~INET_ECN_MASK; 55 if (dscp != newdscp) { 56 if (likely(!skb_try_make_writable(skb, wlen))) { 57 ipv6_change_dsfield(ipv6_hdr(skb), 58 INET_ECN_MASK, 59 newdscp); 60 ca->stats_dscp_set++; 61 } else { 62 ca->stats_dscp_error++; 63 } 64 } 65 break; 66 default: 67 break; 68 } 69 } 70 71 static void tcf_ctinfo_cpmark_set(struct nf_conn *ct, struct tcf_ctinfo *ca, 72 struct tcf_ctinfo_params *cp, 73 struct sk_buff *skb) 74 { 75 ca->stats_cpmark_set++; 76 skb->mark = ct->mark & cp->cpmarkmask; 77 } 78 79 static int tcf_ctinfo_act(struct sk_buff *skb, const struct tc_action *a, 80 struct tcf_result *res) 81 { 82 const struct nf_conntrack_tuple_hash *thash = NULL; 83 struct tcf_ctinfo *ca = to_ctinfo(a); 84 struct nf_conntrack_tuple tuple; 85 struct nf_conntrack_zone zone; 86 enum ip_conntrack_info ctinfo; 87 struct tcf_ctinfo_params *cp; 88 struct nf_conn *ct; 89 int proto, wlen; 90 int action; 91 92 cp = rcu_dereference_bh(ca->params); 93 94 tcf_lastuse_update(&ca->tcf_tm); 95 bstats_update(&ca->tcf_bstats, skb); 96 action = READ_ONCE(ca->tcf_action); 97 98 wlen = skb_network_offset(skb); 99 switch (skb_protocol(skb, true)) { 100 case htons(ETH_P_IP): 101 wlen += sizeof(struct iphdr); 102 if (!pskb_may_pull(skb, wlen)) 103 goto out; 104 105 proto = NFPROTO_IPV4; 106 break; 107 case htons(ETH_P_IPV6): 108 wlen += sizeof(struct ipv6hdr); 109 if (!pskb_may_pull(skb, wlen)) 110 goto out; 111 112 proto = NFPROTO_IPV6; 113 break; 114 default: 115 goto out; 116 } 117 118 ct = nf_ct_get(skb, &ctinfo); 119 if (!ct) { /* look harder, usually ingress */ 120 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), 121 proto, cp->net, &tuple)) 122 goto out; 123 zone.id = cp->zone; 124 zone.dir = NF_CT_DEFAULT_ZONE_DIR; 125 126 thash = nf_conntrack_find_get(cp->net, &zone, &tuple); 127 if (!thash) 128 goto out; 129 130 ct = nf_ct_tuplehash_to_ctrack(thash); 131 } 132 133 if (cp->mode & CTINFO_MODE_DSCP) 134 if (!cp->dscpstatemask || (ct->mark & cp->dscpstatemask)) 135 tcf_ctinfo_dscp_set(ct, ca, cp, skb, wlen, proto); 136 137 if (cp->mode & CTINFO_MODE_CPMARK) 138 tcf_ctinfo_cpmark_set(ct, ca, cp, skb); 139 140 if (thash) 141 nf_ct_put(ct); 142 out: 143 return action; 144 } 145 146 static const struct nla_policy ctinfo_policy[TCA_CTINFO_MAX + 1] = { 147 [TCA_CTINFO_ACT] = 148 NLA_POLICY_EXACT_LEN(sizeof(struct tc_ctinfo)), 149 [TCA_CTINFO_ZONE] = { .type = NLA_U16 }, 150 [TCA_CTINFO_PARMS_DSCP_MASK] = { .type = NLA_U32 }, 151 [TCA_CTINFO_PARMS_DSCP_STATEMASK] = { .type = NLA_U32 }, 152 [TCA_CTINFO_PARMS_CPMARK_MASK] = { .type = NLA_U32 }, 153 }; 154 155 static int tcf_ctinfo_init(struct net *net, struct nlattr *nla, 156 struct nlattr *est, struct tc_action **a, 157 struct tcf_proto *tp, u32 flags, 158 struct netlink_ext_ack *extack) 159 { 160 struct tc_action_net *tn = net_generic(net, ctinfo_net_id); 161 bool bind = flags & TCA_ACT_FLAGS_BIND; 162 u32 dscpmask = 0, dscpstatemask, index; 163 struct nlattr *tb[TCA_CTINFO_MAX + 1]; 164 struct tcf_ctinfo_params *cp_new; 165 struct tcf_chain *goto_ch = NULL; 166 struct tc_ctinfo *actparm; 167 struct tcf_ctinfo *ci; 168 u8 dscpmaskshift; 169 int ret = 0, err; 170 171 if (!nla) { 172 NL_SET_ERR_MSG_MOD(extack, "ctinfo requires attributes to be passed"); 173 return -EINVAL; 174 } 175 176 err = nla_parse_nested(tb, TCA_CTINFO_MAX, nla, ctinfo_policy, extack); 177 if (err < 0) 178 return err; 179 180 if (!tb[TCA_CTINFO_ACT]) { 181 NL_SET_ERR_MSG_MOD(extack, 182 "Missing required TCA_CTINFO_ACT attribute"); 183 return -EINVAL; 184 } 185 actparm = nla_data(tb[TCA_CTINFO_ACT]); 186 187 /* do some basic validation here before dynamically allocating things */ 188 /* that we would otherwise have to clean up. */ 189 if (tb[TCA_CTINFO_PARMS_DSCP_MASK]) { 190 dscpmask = nla_get_u32(tb[TCA_CTINFO_PARMS_DSCP_MASK]); 191 /* need contiguous 6 bit mask */ 192 dscpmaskshift = dscpmask ? __ffs(dscpmask) : 0; 193 if ((~0 & (dscpmask >> dscpmaskshift)) != 0x3f) { 194 NL_SET_ERR_MSG_ATTR(extack, 195 tb[TCA_CTINFO_PARMS_DSCP_MASK], 196 "dscp mask must be 6 contiguous bits"); 197 return -EINVAL; 198 } 199 dscpstatemask = tb[TCA_CTINFO_PARMS_DSCP_STATEMASK] ? 200 nla_get_u32(tb[TCA_CTINFO_PARMS_DSCP_STATEMASK]) : 0; 201 /* mask & statemask must not overlap */ 202 if (dscpmask & dscpstatemask) { 203 NL_SET_ERR_MSG_ATTR(extack, 204 tb[TCA_CTINFO_PARMS_DSCP_STATEMASK], 205 "dscp statemask must not overlap dscp mask"); 206 return -EINVAL; 207 } 208 } 209 210 /* done the validation:now to the actual action allocation */ 211 index = actparm->index; 212 err = tcf_idr_check_alloc(tn, &index, a, bind); 213 if (!err) { 214 ret = tcf_idr_create(tn, index, est, a, 215 &act_ctinfo_ops, bind, false, 0); 216 if (ret) { 217 tcf_idr_cleanup(tn, index); 218 return ret; 219 } 220 ret = ACT_P_CREATED; 221 } else if (err > 0) { 222 if (bind) /* don't override defaults */ 223 return 0; 224 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 225 tcf_idr_release(*a, bind); 226 return -EEXIST; 227 } 228 } else { 229 return err; 230 } 231 232 err = tcf_action_check_ctrlact(actparm->action, tp, &goto_ch, extack); 233 if (err < 0) 234 goto release_idr; 235 236 ci = to_ctinfo(*a); 237 238 cp_new = kzalloc(sizeof(*cp_new), GFP_KERNEL); 239 if (unlikely(!cp_new)) { 240 err = -ENOMEM; 241 goto put_chain; 242 } 243 244 cp_new->net = net; 245 cp_new->zone = tb[TCA_CTINFO_ZONE] ? 246 nla_get_u16(tb[TCA_CTINFO_ZONE]) : 0; 247 if (dscpmask) { 248 cp_new->dscpmask = dscpmask; 249 cp_new->dscpmaskshift = dscpmaskshift; 250 cp_new->dscpstatemask = dscpstatemask; 251 cp_new->mode |= CTINFO_MODE_DSCP; 252 } 253 254 if (tb[TCA_CTINFO_PARMS_CPMARK_MASK]) { 255 cp_new->cpmarkmask = 256 nla_get_u32(tb[TCA_CTINFO_PARMS_CPMARK_MASK]); 257 cp_new->mode |= CTINFO_MODE_CPMARK; 258 } 259 260 spin_lock_bh(&ci->tcf_lock); 261 goto_ch = tcf_action_set_ctrlact(*a, actparm->action, goto_ch); 262 cp_new = rcu_replace_pointer(ci->params, cp_new, 263 lockdep_is_held(&ci->tcf_lock)); 264 spin_unlock_bh(&ci->tcf_lock); 265 266 if (goto_ch) 267 tcf_chain_put_by_act(goto_ch); 268 if (cp_new) 269 kfree_rcu(cp_new, rcu); 270 271 return ret; 272 273 put_chain: 274 if (goto_ch) 275 tcf_chain_put_by_act(goto_ch); 276 release_idr: 277 tcf_idr_release(*a, bind); 278 return err; 279 } 280 281 static int tcf_ctinfo_dump(struct sk_buff *skb, struct tc_action *a, 282 int bind, int ref) 283 { 284 struct tcf_ctinfo *ci = to_ctinfo(a); 285 struct tc_ctinfo opt = { 286 .index = ci->tcf_index, 287 .refcnt = refcount_read(&ci->tcf_refcnt) - ref, 288 .bindcnt = atomic_read(&ci->tcf_bindcnt) - bind, 289 }; 290 unsigned char *b = skb_tail_pointer(skb); 291 struct tcf_ctinfo_params *cp; 292 struct tcf_t t; 293 294 spin_lock_bh(&ci->tcf_lock); 295 cp = rcu_dereference_protected(ci->params, 296 lockdep_is_held(&ci->tcf_lock)); 297 298 tcf_tm_dump(&t, &ci->tcf_tm); 299 if (nla_put_64bit(skb, TCA_CTINFO_TM, sizeof(t), &t, TCA_CTINFO_PAD)) 300 goto nla_put_failure; 301 302 opt.action = ci->tcf_action; 303 if (nla_put(skb, TCA_CTINFO_ACT, sizeof(opt), &opt)) 304 goto nla_put_failure; 305 306 if (nla_put_u16(skb, TCA_CTINFO_ZONE, cp->zone)) 307 goto nla_put_failure; 308 309 if (cp->mode & CTINFO_MODE_DSCP) { 310 if (nla_put_u32(skb, TCA_CTINFO_PARMS_DSCP_MASK, 311 cp->dscpmask)) 312 goto nla_put_failure; 313 if (nla_put_u32(skb, TCA_CTINFO_PARMS_DSCP_STATEMASK, 314 cp->dscpstatemask)) 315 goto nla_put_failure; 316 } 317 318 if (cp->mode & CTINFO_MODE_CPMARK) { 319 if (nla_put_u32(skb, TCA_CTINFO_PARMS_CPMARK_MASK, 320 cp->cpmarkmask)) 321 goto nla_put_failure; 322 } 323 324 if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_DSCP_SET, 325 ci->stats_dscp_set, TCA_CTINFO_PAD)) 326 goto nla_put_failure; 327 328 if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_DSCP_ERROR, 329 ci->stats_dscp_error, TCA_CTINFO_PAD)) 330 goto nla_put_failure; 331 332 if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_CPMARK_SET, 333 ci->stats_cpmark_set, TCA_CTINFO_PAD)) 334 goto nla_put_failure; 335 336 spin_unlock_bh(&ci->tcf_lock); 337 return skb->len; 338 339 nla_put_failure: 340 spin_unlock_bh(&ci->tcf_lock); 341 nlmsg_trim(skb, b); 342 return -1; 343 } 344 345 static int tcf_ctinfo_walker(struct net *net, struct sk_buff *skb, 346 struct netlink_callback *cb, int type, 347 const struct tc_action_ops *ops, 348 struct netlink_ext_ack *extack) 349 { 350 struct tc_action_net *tn = net_generic(net, ctinfo_net_id); 351 352 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 353 } 354 355 static int tcf_ctinfo_search(struct net *net, struct tc_action **a, u32 index) 356 { 357 struct tc_action_net *tn = net_generic(net, ctinfo_net_id); 358 359 return tcf_idr_search(tn, a, index); 360 } 361 362 static void tcf_ctinfo_cleanup(struct tc_action *a) 363 { 364 struct tcf_ctinfo *ci = to_ctinfo(a); 365 struct tcf_ctinfo_params *cp; 366 367 cp = rcu_dereference_protected(ci->params, 1); 368 if (cp) 369 kfree_rcu(cp, rcu); 370 } 371 372 static struct tc_action_ops act_ctinfo_ops = { 373 .kind = "ctinfo", 374 .id = TCA_ID_CTINFO, 375 .owner = THIS_MODULE, 376 .act = tcf_ctinfo_act, 377 .dump = tcf_ctinfo_dump, 378 .init = tcf_ctinfo_init, 379 .cleanup= tcf_ctinfo_cleanup, 380 .walk = tcf_ctinfo_walker, 381 .lookup = tcf_ctinfo_search, 382 .size = sizeof(struct tcf_ctinfo), 383 }; 384 385 static __net_init int ctinfo_init_net(struct net *net) 386 { 387 struct tc_action_net *tn = net_generic(net, ctinfo_net_id); 388 389 return tc_action_net_init(net, tn, &act_ctinfo_ops); 390 } 391 392 static void __net_exit ctinfo_exit_net(struct list_head *net_list) 393 { 394 tc_action_net_exit(net_list, ctinfo_net_id); 395 } 396 397 static struct pernet_operations ctinfo_net_ops = { 398 .init = ctinfo_init_net, 399 .exit_batch = ctinfo_exit_net, 400 .id = &ctinfo_net_id, 401 .size = sizeof(struct tc_action_net), 402 }; 403 404 static int __init ctinfo_init_module(void) 405 { 406 return tcf_register_action(&act_ctinfo_ops, &ctinfo_net_ops); 407 } 408 409 static void __exit ctinfo_cleanup_module(void) 410 { 411 tcf_unregister_action(&act_ctinfo_ops, &ctinfo_net_ops); 412 } 413 414 module_init(ctinfo_init_module); 415 module_exit(ctinfo_cleanup_module); 416 MODULE_AUTHOR("Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>"); 417 MODULE_DESCRIPTION("Connection tracking mark actions"); 418 MODULE_LICENSE("GPL"); 419