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#include <net/tc_wrapper.h>
22
23#include <net/netfilter/nf_conntrack.h>
24#include <net/netfilter/nf_conntrack_core.h>
25#include <net/netfilter/nf_conntrack_ecache.h>
26#include <net/netfilter/nf_conntrack_zones.h>
27
28static struct tc_action_ops act_ctinfo_ops;
29
30static 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 = (((READ_ONCE(ct->mark) & cp->dscpmask) >> cp->dscpmaskshift) << 2) &
37 ~INET_ECN_MASK;
38
39 switch (proto) {
40 case NFPROTO_IPV4:
41 dscp = ipv4_get_dsfield(iph: ip_hdr(skb)) & ~INET_ECN_MASK;
42 if (dscp != newdscp) {
43 if (likely(!skb_try_make_writable(skb, wlen))) {
44 ipv4_change_dsfield(iph: ip_hdr(skb),
45 mask: INET_ECN_MASK,
46 value: newdscp);
47 atomic64_inc(v: &ca->stats_dscp_set);
48 } else {
49 atomic64_inc(v: &ca->stats_dscp_error);
50 }
51 }
52 break;
53 case NFPROTO_IPV6:
54 dscp = ipv6_get_dsfield(ipv6h: ipv6_hdr(skb)) & ~INET_ECN_MASK;
55 if (dscp != newdscp) {
56 if (likely(!skb_try_make_writable(skb, wlen))) {
57 ipv6_change_dsfield(ipv6h: ipv6_hdr(skb),
58 mask: INET_ECN_MASK,
59 value: newdscp);
60 atomic64_inc(v: &ca->stats_dscp_set);
61 } else {
62 atomic64_inc(v: &ca->stats_dscp_error);
63 }
64 }
65 break;
66 default:
67 break;
68 }
69}
70
71static 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 atomic64_inc(v: &ca->stats_cpmark_set);
76 skb->mark = READ_ONCE(ct->mark) & cp->cpmarkmask;
77}
78
79TC_INDIRECT_SCOPE int tcf_ctinfo_act(struct sk_buff *skb,
80 const struct tc_action *a,
81 struct tcf_result *res)
82{
83 const struct nf_conntrack_tuple_hash *thash = NULL;
84 struct tcf_ctinfo *ca = to_ctinfo(a);
85 struct nf_conntrack_tuple tuple;
86 struct nf_conntrack_zone zone;
87 enum ip_conntrack_info ctinfo;
88 struct tcf_ctinfo_params *cp;
89 struct nf_conn *ct;
90 int proto, wlen;
91
92 cp = rcu_dereference_bh(ca->params);
93
94 tcf_lastuse_update(tm: &ca->tcf_tm);
95 tcf_action_update_bstats(a: &ca->common, skb);
96
97 wlen = skb_network_offset(skb);
98 switch (skb_protocol(skb, skip_vlan: true)) {
99 case htons(ETH_P_IP):
100 wlen += sizeof(struct iphdr);
101 if (!pskb_may_pull(skb, len: 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, len: 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: &ctinfo);
118 if (!ct) { /* look harder, usually ingress */
119 if (!nf_ct_get_tuplepr(skb, nhoff: skb_network_offset(skb),
120 l3num: proto, net: cp->net, tuple: &tuple))
121 goto out;
122 zone.id = cp->zone;
123 zone.dir = NF_CT_DEFAULT_ZONE_DIR;
124
125 thash = nf_conntrack_find_get(net: cp->net, zone: &zone, tuple: &tuple);
126 if (!thash)
127 goto out;
128
129 ct = nf_ct_tuplehash_to_ctrack(hash: 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);
141out:
142 return cp->action;
143}
144
145static 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
154static 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, id: 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, policy: 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(nla: 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(nla: 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 =
199 nla_get_u32_default(nla: tb[TCA_CTINFO_PARMS_DSCP_STATEMASK],
200 defvalue: 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: &index, a, bind);
213 if (!err) {
214 ret = tcf_idr_create_from_flags(tn, index, est, a,
215 ops: &act_ctinfo_ops, bind, flags);
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 ACT_P_BOUND;
224 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
225 tcf_idr_release(a: *a, bind);
226 return -EEXIST;
227 }
228 } else {
229 return err;
230 }
231
232 err = tcf_action_check_ctrlact(action: actparm->action, tp, handle: &goto_ch, newchain: 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 = nla_get_u16_default(nla: tb[TCA_CTINFO_ZONE], defvalue: 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(nla: tb[TCA_CTINFO_PARMS_CPMARK_MASK]);
256 cp_new->mode |= CTINFO_MODE_CPMARK;
257 }
258
259 cp_new->action = actparm->action;
260
261 spin_lock_bh(lock: &ci->tcf_lock);
262 goto_ch = tcf_action_set_ctrlact(a: *a, action: actparm->action, newchain: goto_ch);
263 cp_new = rcu_replace_pointer(ci->params, cp_new,
264 lockdep_is_held(&ci->tcf_lock));
265 spin_unlock_bh(lock: &ci->tcf_lock);
266
267 if (goto_ch)
268 tcf_chain_put_by_act(chain: goto_ch);
269 if (cp_new)
270 kfree_rcu(cp_new, rcu);
271
272 return ret;
273
274put_chain:
275 if (goto_ch)
276 tcf_chain_put_by_act(chain: goto_ch);
277release_idr:
278 tcf_idr_release(a: *a, bind);
279 return err;
280}
281
282static int tcf_ctinfo_dump(struct sk_buff *skb, struct tc_action *a,
283 int bind, int ref)
284{
285 const struct tcf_ctinfo *ci = to_ctinfo(a);
286 unsigned char *b = skb_tail_pointer(skb);
287 const struct tcf_ctinfo_params *cp;
288 struct tc_ctinfo opt = {
289 .index = ci->tcf_index,
290 .refcnt = refcount_read(r: &ci->tcf_refcnt) - ref,
291 .bindcnt = atomic_read(v: &ci->tcf_bindcnt) - bind,
292 };
293 struct tcf_t t;
294
295 rcu_read_lock();
296 cp = rcu_dereference(ci->params);
297
298 tcf_tm_dump(dtm: &t, stm: &ci->tcf_tm);
299 if (nla_put_64bit(skb, attrtype: TCA_CTINFO_TM, attrlen: sizeof(t), data: &t, padattr: TCA_CTINFO_PAD))
300 goto nla_put_failure;
301
302 opt.action = cp->action;
303 if (nla_put(skb, attrtype: TCA_CTINFO_ACT, attrlen: sizeof(opt), data: &opt))
304 goto nla_put_failure;
305
306 if (nla_put_u16(skb, attrtype: TCA_CTINFO_ZONE, value: cp->zone))
307 goto nla_put_failure;
308
309 if (cp->mode & CTINFO_MODE_DSCP) {
310 if (nla_put_u32(skb, attrtype: TCA_CTINFO_PARMS_DSCP_MASK,
311 value: cp->dscpmask))
312 goto nla_put_failure;
313 if (nla_put_u32(skb, attrtype: TCA_CTINFO_PARMS_DSCP_STATEMASK,
314 value: cp->dscpstatemask))
315 goto nla_put_failure;
316 }
317
318 if (cp->mode & CTINFO_MODE_CPMARK) {
319 if (nla_put_u32(skb, attrtype: TCA_CTINFO_PARMS_CPMARK_MASK,
320 value: cp->cpmarkmask))
321 goto nla_put_failure;
322 }
323
324 if (nla_put_u64_64bit(skb, attrtype: TCA_CTINFO_STATS_DSCP_SET,
325 value: atomic64_read(v: &ci->stats_dscp_set),
326 padattr: TCA_CTINFO_PAD))
327 goto nla_put_failure;
328
329 if (nla_put_u64_64bit(skb, attrtype: TCA_CTINFO_STATS_DSCP_ERROR,
330 value: atomic64_read(v: &ci->stats_dscp_error),
331 padattr: TCA_CTINFO_PAD))
332 goto nla_put_failure;
333
334 if (nla_put_u64_64bit(skb, attrtype: TCA_CTINFO_STATS_CPMARK_SET,
335 value: atomic64_read(v: &ci->stats_cpmark_set),
336 padattr: TCA_CTINFO_PAD))
337 goto nla_put_failure;
338
339 rcu_read_unlock();
340 return skb->len;
341
342nla_put_failure:
343 rcu_read_unlock();
344 nlmsg_trim(skb, mark: b);
345 return -1;
346}
347
348static void tcf_ctinfo_cleanup(struct tc_action *a)
349{
350 struct tcf_ctinfo *ci = to_ctinfo(a);
351 struct tcf_ctinfo_params *cp;
352
353 cp = rcu_dereference_protected(ci->params, 1);
354 if (cp)
355 kfree_rcu(cp, rcu);
356}
357
358static struct tc_action_ops act_ctinfo_ops = {
359 .kind = "ctinfo",
360 .id = TCA_ID_CTINFO,
361 .owner = THIS_MODULE,
362 .act = tcf_ctinfo_act,
363 .dump = tcf_ctinfo_dump,
364 .init = tcf_ctinfo_init,
365 .cleanup= tcf_ctinfo_cleanup,
366 .size = sizeof(struct tcf_ctinfo),
367};
368MODULE_ALIAS_NET_ACT("ctinfo");
369
370static __net_init int ctinfo_init_net(struct net *net)
371{
372 struct tc_action_net *tn = net_generic(net, id: act_ctinfo_ops.net_id);
373
374 return tc_action_net_init(net, tn, ops: &act_ctinfo_ops);
375}
376
377static void __net_exit ctinfo_exit_net(struct list_head *net_list)
378{
379 tc_action_net_exit(net_list, id: act_ctinfo_ops.net_id);
380}
381
382static struct pernet_operations ctinfo_net_ops = {
383 .init = ctinfo_init_net,
384 .exit_batch = ctinfo_exit_net,
385 .id = &act_ctinfo_ops.net_id,
386 .size = sizeof(struct tc_action_net),
387};
388
389static int __init ctinfo_init_module(void)
390{
391 return tcf_register_action(a: &act_ctinfo_ops, ops: &ctinfo_net_ops);
392}
393
394static void __exit ctinfo_cleanup_module(void)
395{
396 tcf_unregister_action(a: &act_ctinfo_ops, ops: &ctinfo_net_ops);
397}
398
399module_init(ctinfo_init_module);
400module_exit(ctinfo_cleanup_module);
401MODULE_AUTHOR("Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>");
402MODULE_DESCRIPTION("Connection tracking mark actions");
403MODULE_LICENSE("GPL");
404

source code of linux/net/sched/act_ctinfo.c