xref: /linux/net/sched/act_mpls.c (revision 908fc4c2) 
1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2019 Netronome Systems, Inc. */
3 
4 #include <linux/if_arp.h>
5 #include <linux/init.h>
6 #include <linux/kernel.h>
7 #include <linux/module.h>
8 #include <linux/mpls.h>
9 #include <linux/rtnetlink.h>
10 #include <linux/skbuff.h>
11 #include <linux/tc_act/tc_mpls.h>
12 #include <net/mpls.h>
13 #include <net/netlink.h>
14 #include <net/pkt_sched.h>
15 #include <net/pkt_cls.h>
16 #include <net/tc_act/tc_mpls.h>
17 
18 static unsigned int mpls_net_id;
19 static struct tc_action_ops act_mpls_ops;
20 
21 #define ACT_MPLS_TTL_DEFAULT	255
22 
23 static __be32 tcf_mpls_get_lse(struct mpls_shim_hdr *lse,
24 			       struct tcf_mpls_params *p, bool set_bos)
25 {
26 	u32 new_lse = 0;
27 
28 	if (lse)
29 		new_lse = be32_to_cpu(lse->label_stack_entry);
30 
31 	if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET) {
32 		new_lse &= ~MPLS_LS_LABEL_MASK;
33 		new_lse |= p->tcfm_label << MPLS_LS_LABEL_SHIFT;
34 	}
35 	if (p->tcfm_ttl) {
36 		new_lse &= ~MPLS_LS_TTL_MASK;
37 		new_lse |= p->tcfm_ttl << MPLS_LS_TTL_SHIFT;
38 	}
39 	if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET) {
40 		new_lse &= ~MPLS_LS_TC_MASK;
41 		new_lse |= p->tcfm_tc << MPLS_LS_TC_SHIFT;
42 	}
43 	if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET) {
44 		new_lse &= ~MPLS_LS_S_MASK;
45 		new_lse |= p->tcfm_bos << MPLS_LS_S_SHIFT;
46 	} else if (set_bos) {
47 		new_lse |= 1 << MPLS_LS_S_SHIFT;
48 	}
49 
50 	return cpu_to_be32(new_lse);
51 }
52 
53 static int tcf_mpls_act(struct sk_buff *skb, const struct tc_action *a,
54 			struct tcf_result *res)
55 {
56 	struct tcf_mpls *m = to_mpls(a);
57 	struct tcf_mpls_params *p;
58 	__be32 new_lse;
59 	int ret, mac_len;
60 
61 	tcf_lastuse_update(&m->tcf_tm);
62 	bstats_update(this_cpu_ptr(m->common.cpu_bstats), skb);
63 
64 	/* Ensure 'data' points at mac_header prior calling mpls manipulating
65 	 * functions.
66 	 */
67 	if (skb_at_tc_ingress(skb)) {
68 		skb_push_rcsum(skb, skb->mac_len);
69 		mac_len = skb->mac_len;
70 	} else {
71 		mac_len = skb_network_header(skb) - skb_mac_header(skb);
72 	}
73 
74 	ret = READ_ONCE(m->tcf_action);
75 
76 	p = rcu_dereference_bh(m->mpls_p);
77 
78 	switch (p->tcfm_action) {
79 	case TCA_MPLS_ACT_POP:
80 		if (skb_mpls_pop(skb, p->tcfm_proto, mac_len,
81 				 skb->dev && skb->dev->type == ARPHRD_ETHER))
82 			goto drop;
83 		break;
84 	case TCA_MPLS_ACT_PUSH:
85 		new_lse = tcf_mpls_get_lse(NULL, p, !eth_p_mpls(skb_protocol(skb, true)));
86 		if (skb_mpls_push(skb, new_lse, p->tcfm_proto, mac_len,
87 				  skb->dev && skb->dev->type == ARPHRD_ETHER))
88 			goto drop;
89 		break;
90 	case TCA_MPLS_ACT_MAC_PUSH:
91 		if (skb_vlan_tag_present(skb)) {
92 			if (__vlan_insert_inner_tag(skb, skb->vlan_proto,
93 						    skb_vlan_tag_get(skb),
94 						    ETH_HLEN) < 0)
95 				goto drop;
96 
97 			skb->protocol = skb->vlan_proto;
98 			__vlan_hwaccel_clear_tag(skb);
99 		}
100 
101 		new_lse = tcf_mpls_get_lse(NULL, p, mac_len ||
102 					   !eth_p_mpls(skb->protocol));
103 
104 		if (skb_mpls_push(skb, new_lse, p->tcfm_proto, 0, false))
105 			goto drop;
106 		break;
107 	case TCA_MPLS_ACT_MODIFY:
108 		if (!pskb_may_pull(skb,
109 				   skb_network_offset(skb) + MPLS_HLEN))
110 			goto drop;
111 		new_lse = tcf_mpls_get_lse(mpls_hdr(skb), p, false);
112 		if (skb_mpls_update_lse(skb, new_lse))
113 			goto drop;
114 		break;
115 	case TCA_MPLS_ACT_DEC_TTL:
116 		if (skb_mpls_dec_ttl(skb))
117 			goto drop;
118 		break;
119 	}
120 
121 	if (skb_at_tc_ingress(skb))
122 		skb_pull_rcsum(skb, skb->mac_len);
123 
124 	return ret;
125 
126 drop:
127 	qstats_drop_inc(this_cpu_ptr(m->common.cpu_qstats));
128 	return TC_ACT_SHOT;
129 }
130 
131 static int valid_label(const struct nlattr *attr,
132 		       struct netlink_ext_ack *extack)
133 {
134 	const u32 *label = nla_data(attr);
135 
136 	if (*label & ~MPLS_LABEL_MASK || *label == MPLS_LABEL_IMPLNULL) {
137 		NL_SET_ERR_MSG_MOD(extack, "MPLS label out of range");
138 		return -EINVAL;
139 	}
140 
141 	return 0;
142 }
143 
144 static const struct nla_policy mpls_policy[TCA_MPLS_MAX + 1] = {
145 	[TCA_MPLS_PARMS]	= NLA_POLICY_EXACT_LEN(sizeof(struct tc_mpls)),
146 	[TCA_MPLS_PROTO]	= { .type = NLA_U16 },
147 	[TCA_MPLS_LABEL]	= NLA_POLICY_VALIDATE_FN(NLA_U32, valid_label),
148 	[TCA_MPLS_TC]		= NLA_POLICY_RANGE(NLA_U8, 0, 7),
149 	[TCA_MPLS_TTL]		= NLA_POLICY_MIN(NLA_U8, 1),
150 	[TCA_MPLS_BOS]		= NLA_POLICY_RANGE(NLA_U8, 0, 1),
151 };
152 
153 static int tcf_mpls_init(struct net *net, struct nlattr *nla,
154 			 struct nlattr *est, struct tc_action **a,
155 			 struct tcf_proto *tp, u32 flags,
156 			 struct netlink_ext_ack *extack)
157 {
158 	struct tc_action_net *tn = net_generic(net, mpls_net_id);
159 	bool bind = flags & TCA_ACT_FLAGS_BIND;
160 	struct nlattr *tb[TCA_MPLS_MAX + 1];
161 	struct tcf_chain *goto_ch = NULL;
162 	struct tcf_mpls_params *p;
163 	struct tc_mpls *parm;
164 	bool exists = false;
165 	struct tcf_mpls *m;
166 	int ret = 0, err;
167 	u8 mpls_ttl = 0;
168 	u32 index;
169 
170 	if (!nla) {
171 		NL_SET_ERR_MSG_MOD(extack, "Missing netlink attributes");
172 		return -EINVAL;
173 	}
174 
175 	err = nla_parse_nested(tb, TCA_MPLS_MAX, nla, mpls_policy, extack);
176 	if (err < 0)
177 		return err;
178 
179 	if (!tb[TCA_MPLS_PARMS]) {
180 		NL_SET_ERR_MSG_MOD(extack, "No MPLS params");
181 		return -EINVAL;
182 	}
183 	parm = nla_data(tb[TCA_MPLS_PARMS]);
184 	index = parm->index;
185 
186 	/* Verify parameters against action type. */
187 	switch (parm->m_action) {
188 	case TCA_MPLS_ACT_POP:
189 		if (!tb[TCA_MPLS_PROTO]) {
190 			NL_SET_ERR_MSG_MOD(extack, "Protocol must be set for MPLS pop");
191 			return -EINVAL;
192 		}
193 		if (!eth_proto_is_802_3(nla_get_be16(tb[TCA_MPLS_PROTO]))) {
194 			NL_SET_ERR_MSG_MOD(extack, "Invalid protocol type for MPLS pop");
195 			return -EINVAL;
196 		}
197 		if (tb[TCA_MPLS_LABEL] || tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] ||
198 		    tb[TCA_MPLS_BOS]) {
199 			NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC or BOS cannot be used with MPLS pop");
200 			return -EINVAL;
201 		}
202 		break;
203 	case TCA_MPLS_ACT_DEC_TTL:
204 		if (tb[TCA_MPLS_PROTO] || tb[TCA_MPLS_LABEL] ||
205 		    tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || tb[TCA_MPLS_BOS]) {
206 			NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC, BOS or protocol cannot be used with MPLS dec_ttl");
207 			return -EINVAL;
208 		}
209 		break;
210 	case TCA_MPLS_ACT_PUSH:
211 	case TCA_MPLS_ACT_MAC_PUSH:
212 		if (!tb[TCA_MPLS_LABEL]) {
213 			NL_SET_ERR_MSG_MOD(extack, "Label is required for MPLS push");
214 			return -EINVAL;
215 		}
216 		if (tb[TCA_MPLS_PROTO] &&
217 		    !eth_p_mpls(nla_get_be16(tb[TCA_MPLS_PROTO]))) {
218 			NL_SET_ERR_MSG_MOD(extack, "Protocol must be an MPLS type for MPLS push");
219 			return -EPROTONOSUPPORT;
220 		}
221 		/* Push needs a TTL - if not specified, set a default value. */
222 		if (!tb[TCA_MPLS_TTL]) {
223 #if IS_ENABLED(CONFIG_MPLS)
224 			mpls_ttl = net->mpls.default_ttl ?
225 				   net->mpls.default_ttl : ACT_MPLS_TTL_DEFAULT;
226 #else
227 			mpls_ttl = ACT_MPLS_TTL_DEFAULT;
228 #endif
229 		}
230 		break;
231 	case TCA_MPLS_ACT_MODIFY:
232 		if (tb[TCA_MPLS_PROTO]) {
233 			NL_SET_ERR_MSG_MOD(extack, "Protocol cannot be used with MPLS modify");
234 			return -EINVAL;
235 		}
236 		break;
237 	default:
238 		NL_SET_ERR_MSG_MOD(extack, "Unknown MPLS action");
239 		return -EINVAL;
240 	}
241 
242 	err = tcf_idr_check_alloc(tn, &index, a, bind);
243 	if (err < 0)
244 		return err;
245 	exists = err;
246 	if (exists && bind)
247 		return 0;
248 
249 	if (!exists) {
250 		ret = tcf_idr_create(tn, index, est, a,
251 				     &act_mpls_ops, bind, true, flags);
252 		if (ret) {
253 			tcf_idr_cleanup(tn, index);
254 			return ret;
255 		}
256 
257 		ret = ACT_P_CREATED;
258 	} else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
259 		tcf_idr_release(*a, bind);
260 		return -EEXIST;
261 	}
262 
263 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
264 	if (err < 0)
265 		goto release_idr;
266 
267 	m = to_mpls(*a);
268 
269 	p = kzalloc(sizeof(*p), GFP_KERNEL);
270 	if (!p) {
271 		err = -ENOMEM;
272 		goto put_chain;
273 	}
274 
275 	p->tcfm_action = parm->m_action;
276 	p->tcfm_label = tb[TCA_MPLS_LABEL] ? nla_get_u32(tb[TCA_MPLS_LABEL]) :
277 					     ACT_MPLS_LABEL_NOT_SET;
278 	p->tcfm_tc = tb[TCA_MPLS_TC] ? nla_get_u8(tb[TCA_MPLS_TC]) :
279 				       ACT_MPLS_TC_NOT_SET;
280 	p->tcfm_ttl = tb[TCA_MPLS_TTL] ? nla_get_u8(tb[TCA_MPLS_TTL]) :
281 					 mpls_ttl;
282 	p->tcfm_bos = tb[TCA_MPLS_BOS] ? nla_get_u8(tb[TCA_MPLS_BOS]) :
283 					 ACT_MPLS_BOS_NOT_SET;
284 	p->tcfm_proto = tb[TCA_MPLS_PROTO] ? nla_get_be16(tb[TCA_MPLS_PROTO]) :
285 					     htons(ETH_P_MPLS_UC);
286 
287 	spin_lock_bh(&m->tcf_lock);
288 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
289 	p = rcu_replace_pointer(m->mpls_p, p, lockdep_is_held(&m->tcf_lock));
290 	spin_unlock_bh(&m->tcf_lock);
291 
292 	if (goto_ch)
293 		tcf_chain_put_by_act(goto_ch);
294 	if (p)
295 		kfree_rcu(p, rcu);
296 
297 	return ret;
298 put_chain:
299 	if (goto_ch)
300 		tcf_chain_put_by_act(goto_ch);
301 release_idr:
302 	tcf_idr_release(*a, bind);
303 	return err;
304 }
305 
306 static void tcf_mpls_cleanup(struct tc_action *a)
307 {
308 	struct tcf_mpls *m = to_mpls(a);
309 	struct tcf_mpls_params *p;
310 
311 	p = rcu_dereference_protected(m->mpls_p, 1);
312 	if (p)
313 		kfree_rcu(p, rcu);
314 }
315 
316 static int tcf_mpls_dump(struct sk_buff *skb, struct tc_action *a,
317 			 int bind, int ref)
318 {
319 	unsigned char *b = skb_tail_pointer(skb);
320 	struct tcf_mpls *m = to_mpls(a);
321 	struct tcf_mpls_params *p;
322 	struct tc_mpls opt = {
323 		.index    = m->tcf_index,
324 		.refcnt   = refcount_read(&m->tcf_refcnt) - ref,
325 		.bindcnt  = atomic_read(&m->tcf_bindcnt) - bind,
326 	};
327 	struct tcf_t t;
328 
329 	spin_lock_bh(&m->tcf_lock);
330 	opt.action = m->tcf_action;
331 	p = rcu_dereference_protected(m->mpls_p, lockdep_is_held(&m->tcf_lock));
332 	opt.m_action = p->tcfm_action;
333 
334 	if (nla_put(skb, TCA_MPLS_PARMS, sizeof(opt), &opt))
335 		goto nla_put_failure;
336 
337 	if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET &&
338 	    nla_put_u32(skb, TCA_MPLS_LABEL, p->tcfm_label))
339 		goto nla_put_failure;
340 
341 	if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET &&
342 	    nla_put_u8(skb, TCA_MPLS_TC, p->tcfm_tc))
343 		goto nla_put_failure;
344 
345 	if (p->tcfm_ttl && nla_put_u8(skb, TCA_MPLS_TTL, p->tcfm_ttl))
346 		goto nla_put_failure;
347 
348 	if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET &&
349 	    nla_put_u8(skb, TCA_MPLS_BOS, p->tcfm_bos))
350 		goto nla_put_failure;
351 
352 	if (nla_put_be16(skb, TCA_MPLS_PROTO, p->tcfm_proto))
353 		goto nla_put_failure;
354 
355 	tcf_tm_dump(&t, &m->tcf_tm);
356 
357 	if (nla_put_64bit(skb, TCA_MPLS_TM, sizeof(t), &t, TCA_MPLS_PAD))
358 		goto nla_put_failure;
359 
360 	spin_unlock_bh(&m->tcf_lock);
361 
362 	return skb->len;
363 
364 nla_put_failure:
365 	spin_unlock_bh(&m->tcf_lock);
366 	nlmsg_trim(skb, b);
367 	return -EMSGSIZE;
368 }
369 
370 static int tcf_mpls_walker(struct net *net, struct sk_buff *skb,
371 			   struct netlink_callback *cb, int type,
372 			   const struct tc_action_ops *ops,
373 			   struct netlink_ext_ack *extack)
374 {
375 	struct tc_action_net *tn = net_generic(net, mpls_net_id);
376 
377 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
378 }
379 
380 static int tcf_mpls_search(struct net *net, struct tc_action **a, u32 index)
381 {
382 	struct tc_action_net *tn = net_generic(net, mpls_net_id);
383 
384 	return tcf_idr_search(tn, a, index);
385 }
386 
387 static int tcf_mpls_offload_act_setup(struct tc_action *act, void *entry_data,
388 				      u32 *index_inc, bool bind,
389 				      struct netlink_ext_ack *extack)
390 {
391 	if (bind) {
392 		struct flow_action_entry *entry = entry_data;
393 
394 		switch (tcf_mpls_action(act)) {
395 		case TCA_MPLS_ACT_PUSH:
396 			entry->id = FLOW_ACTION_MPLS_PUSH;
397 			entry->mpls_push.proto = tcf_mpls_proto(act);
398 			entry->mpls_push.label = tcf_mpls_label(act);
399 			entry->mpls_push.tc = tcf_mpls_tc(act);
400 			entry->mpls_push.bos = tcf_mpls_bos(act);
401 			entry->mpls_push.ttl = tcf_mpls_ttl(act);
402 			break;
403 		case TCA_MPLS_ACT_POP:
404 			entry->id = FLOW_ACTION_MPLS_POP;
405 			entry->mpls_pop.proto = tcf_mpls_proto(act);
406 			break;
407 		case TCA_MPLS_ACT_MODIFY:
408 			entry->id = FLOW_ACTION_MPLS_MANGLE;
409 			entry->mpls_mangle.label = tcf_mpls_label(act);
410 			entry->mpls_mangle.tc = tcf_mpls_tc(act);
411 			entry->mpls_mangle.bos = tcf_mpls_bos(act);
412 			entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
413 			break;
414 		case TCA_MPLS_ACT_DEC_TTL:
415 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"dec_ttl\" option is used");
416 			return -EOPNOTSUPP;
417 		case TCA_MPLS_ACT_MAC_PUSH:
418 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"mac_push\" option is used");
419 			return -EOPNOTSUPP;
420 		default:
421 			NL_SET_ERR_MSG_MOD(extack, "Unsupported MPLS mode offload");
422 			return -EOPNOTSUPP;
423 		}
424 		*index_inc = 1;
425 	} else {
426 		struct flow_offload_action *fl_action = entry_data;
427 
428 		switch (tcf_mpls_action(act)) {
429 		case TCA_MPLS_ACT_PUSH:
430 			fl_action->id = FLOW_ACTION_MPLS_PUSH;
431 			break;
432 		case TCA_MPLS_ACT_POP:
433 			fl_action->id = FLOW_ACTION_MPLS_POP;
434 			break;
435 		case TCA_MPLS_ACT_MODIFY:
436 			fl_action->id = FLOW_ACTION_MPLS_MANGLE;
437 			break;
438 		default:
439 			return -EOPNOTSUPP;
440 		}
441 	}
442 
443 	return 0;
444 }
445 
446 static struct tc_action_ops act_mpls_ops = {
447 	.kind		=	"mpls",
448 	.id		=	TCA_ID_MPLS,
449 	.owner		=	THIS_MODULE,
450 	.act		=	tcf_mpls_act,
451 	.dump		=	tcf_mpls_dump,
452 	.init		=	tcf_mpls_init,
453 	.cleanup	=	tcf_mpls_cleanup,
454 	.walk		=	tcf_mpls_walker,
455 	.lookup		=	tcf_mpls_search,
456 	.offload_act_setup =	tcf_mpls_offload_act_setup,
457 	.size		=	sizeof(struct tcf_mpls),
458 };
459 
460 static __net_init int mpls_init_net(struct net *net)
461 {
462 	struct tc_action_net *tn = net_generic(net, mpls_net_id);
463 
464 	return tc_action_net_init(net, tn, &act_mpls_ops);
465 }
466 
467 static void __net_exit mpls_exit_net(struct list_head *net_list)
468 {
469 	tc_action_net_exit(net_list, mpls_net_id);
470 }
471 
472 static struct pernet_operations mpls_net_ops = {
473 	.init = mpls_init_net,
474 	.exit_batch = mpls_exit_net,
475 	.id   = &mpls_net_id,
476 	.size = sizeof(struct tc_action_net),
477 };
478 
479 static int __init mpls_init_module(void)
480 {
481 	return tcf_register_action(&act_mpls_ops, &mpls_net_ops);
482 }
483 
484 static void __exit mpls_cleanup_module(void)
485 {
486 	tcf_unregister_action(&act_mpls_ops, &mpls_net_ops);
487 }
488 
489 module_init(mpls_init_module);
490 module_exit(mpls_cleanup_module);
491 
492 MODULE_SOFTDEP("post: mpls_gso");
493 MODULE_AUTHOR("Netronome Systems <oss-drivers@netronome.com>");
494 MODULE_LICENSE("GPL");
495 MODULE_DESCRIPTION("MPLS manipulation actions");
496