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