act_mpls.c

  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#include <net/tc_wrapper.h>
 18
 19static struct tc_action_ops act_mpls_ops;
 20
 21#define ACT_MPLS_TTL_DEFAULT	255
 22
 23static __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
 53TC_INDIRECT_SCOPE int tcf_mpls_act(struct sk_buff *skb,
 54				   const struct tc_action *a,
 55				   struct tcf_result *res)
 56{
 57	struct tcf_mpls *m = to_mpls(a);
 58	struct tcf_mpls_params *p;
 59	__be32 new_lse;
 60	int mac_len;
 61
 62	tcf_lastuse_update(&m->tcf_tm);
 63	bstats_update(this_cpu_ptr(m->common.cpu_bstats), skb);
 64
 65	/* Ensure 'data' points at mac_header prior calling mpls manipulating
 66	 * functions.
 67	 */
 68	if (skb_at_tc_ingress(skb)) {
 69		skb_push_rcsum(skb, skb->mac_len);
 70		mac_len = skb->mac_len;
 71	} else {
 72		mac_len = skb_network_offset(skb);
 73	}
 74
 75	p = rcu_dereference_bh(m->mpls_p);
 76
 77	switch (p->tcfm_action) {
 78	case TCA_MPLS_ACT_POP:
 79		if (skb_mpls_pop(skb, p->tcfm_proto, mac_len,
 80				 skb->dev && skb->dev->type == ARPHRD_ETHER))
 81			goto drop;
 82		break;
 83	case TCA_MPLS_ACT_PUSH:
 84		new_lse = tcf_mpls_get_lse(NULL, p, !eth_p_mpls(skb_protocol(skb, true)));
 85		if (skb_mpls_push(skb, new_lse, p->tcfm_proto, mac_len,
 86				  skb->dev && skb->dev->type == ARPHRD_ETHER))
 87			goto drop;
 88		break;
 89	case TCA_MPLS_ACT_MAC_PUSH:
 90		if (skb_vlan_tag_present(skb)) {
 91			if (__vlan_insert_inner_tag(skb, skb->vlan_proto,
 92						    skb_vlan_tag_get(skb),
 93						    ETH_HLEN) < 0)
 94				goto drop;
 95
 96			skb->protocol = skb->vlan_proto;
 97			__vlan_hwaccel_clear_tag(skb);
 98		}
 99
100		new_lse = tcf_mpls_get_lse(NULL, p, mac_len ||
101					   !eth_p_mpls(skb->protocol));
102
103		if (skb_mpls_push(skb, new_lse, p->tcfm_proto, 0, false))
104			goto drop;
105		break;
106	case TCA_MPLS_ACT_MODIFY:
107		if (!pskb_may_pull(skb,
108				   skb_network_offset(skb) + MPLS_HLEN))
109			goto drop;
110		new_lse = tcf_mpls_get_lse(mpls_hdr(skb), p, false);
111		if (skb_mpls_update_lse(skb, new_lse))
112			goto drop;
113		break;
114	case TCA_MPLS_ACT_DEC_TTL:
115		if (skb_mpls_dec_ttl(skb))
116			goto drop;
117		break;
118	}
119
120	if (skb_at_tc_ingress(skb))
121		skb_pull_rcsum(skb, skb->mac_len);
122
123	return p->action;
124
125drop:
126	qstats_drop_inc(this_cpu_ptr(m->common.cpu_qstats));
127	return TC_ACT_SHOT;
128}
129
130static int valid_label(const struct nlattr *attr,
131		       struct netlink_ext_ack *extack)
132{
133	const u32 *label = nla_data(attr);
134
135	if (nla_len(attr) != sizeof(*label)) {
136		NL_SET_ERR_MSG_MOD(extack, "Invalid MPLS label length");
137		return -EINVAL;
138	}
139
140	if (*label & ~MPLS_LABEL_MASK || *label == MPLS_LABEL_IMPLNULL) {
141		NL_SET_ERR_MSG_MOD(extack, "MPLS label out of range");
142		return -EINVAL;
143	}
144
145	return 0;
146}
147
148static const struct nla_policy mpls_policy[TCA_MPLS_MAX + 1] = {
149	[TCA_MPLS_PARMS]	= NLA_POLICY_EXACT_LEN(sizeof(struct tc_mpls)),
150	[TCA_MPLS_PROTO]	= { .type = NLA_U16 },
151	[TCA_MPLS_LABEL]	= NLA_POLICY_VALIDATE_FN(NLA_BINARY,
152							 valid_label),
153	[TCA_MPLS_TC]		= NLA_POLICY_RANGE(NLA_U8, 0, 7),
154	[TCA_MPLS_TTL]		= NLA_POLICY_MIN(NLA_U8, 1),
155	[TCA_MPLS_BOS]		= NLA_POLICY_RANGE(NLA_U8, 0, 1),
156};
157
158static int tcf_mpls_init(struct net *net, struct nlattr *nla,
159			 struct nlattr *est, struct tc_action **a,
160			 struct tcf_proto *tp, u32 flags,
161			 struct netlink_ext_ack *extack)
162{
163	struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id);
164	bool bind = flags & TCA_ACT_FLAGS_BIND;
165	struct nlattr *tb[TCA_MPLS_MAX + 1];
166	struct tcf_chain *goto_ch = NULL;
167	struct tcf_mpls_params *p;
168	struct tc_mpls *parm;
169	bool exists = false;
170	struct tcf_mpls *m;
171	int ret = 0, err;
172	u8 mpls_ttl = 0;
173	u32 index;
174
175	if (!nla) {
176		NL_SET_ERR_MSG_MOD(extack, "Missing netlink attributes");
177		return -EINVAL;
178	}
179
180	err = nla_parse_nested(tb, TCA_MPLS_MAX, nla, mpls_policy, extack);
181	if (err < 0)
182		return err;
183
184	if (!tb[TCA_MPLS_PARMS]) {
185		NL_SET_ERR_MSG_MOD(extack, "No MPLS params");
186		return -EINVAL;
187	}
188	parm = nla_data(tb[TCA_MPLS_PARMS]);
189	index = parm->index;
190
191	err = tcf_idr_check_alloc(tn, &index, a, bind);
192	if (err < 0)
193		return err;
194	exists = err;
195	if (exists && bind)
196		return ACT_P_BOUND;
197
198	if (!exists) {
199		ret = tcf_idr_create(tn, index, est, a, &act_mpls_ops, bind,
200				     true, flags);
201		if (ret) {
202			tcf_idr_cleanup(tn, index);
203			return ret;
204		}
205
206		ret = ACT_P_CREATED;
207	} else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
208		tcf_idr_release(*a, bind);
209		return -EEXIST;
210	}
211
212	/* Verify parameters against action type. */
213	switch (parm->m_action) {
214	case TCA_MPLS_ACT_POP:
215		if (!tb[TCA_MPLS_PROTO]) {
216			NL_SET_ERR_MSG_MOD(extack, "Protocol must be set for MPLS pop");
217			err = -EINVAL;
218			goto release_idr;
219		}
220		if (!eth_proto_is_802_3(nla_get_be16(tb[TCA_MPLS_PROTO]))) {
221			NL_SET_ERR_MSG_MOD(extack, "Invalid protocol type for MPLS pop");
222			err = -EINVAL;
223			goto release_idr;
224		}
225		if (tb[TCA_MPLS_LABEL] || tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] ||
226		    tb[TCA_MPLS_BOS]) {
227			NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC or BOS cannot be used with MPLS pop");
228			err = -EINVAL;
229			goto release_idr;
230		}
231		break;
232	case TCA_MPLS_ACT_DEC_TTL:
233		if (tb[TCA_MPLS_PROTO] || tb[TCA_MPLS_LABEL] ||
234		    tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || tb[TCA_MPLS_BOS]) {
235			NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC, BOS or protocol cannot be used with MPLS dec_ttl");
236			err = -EINVAL;
237			goto release_idr;
238		}
239		break;
240	case TCA_MPLS_ACT_PUSH:
241	case TCA_MPLS_ACT_MAC_PUSH:
242		if (!tb[TCA_MPLS_LABEL]) {
243			NL_SET_ERR_MSG_MOD(extack, "Label is required for MPLS push");
244			err = -EINVAL;
245			goto release_idr;
246		}
247		if (tb[TCA_MPLS_PROTO] &&
248		    !eth_p_mpls(nla_get_be16(tb[TCA_MPLS_PROTO]))) {
249			NL_SET_ERR_MSG_MOD(extack, "Protocol must be an MPLS type for MPLS push");
250			err = -EPROTONOSUPPORT;
251			goto release_idr;
252		}
253		/* Push needs a TTL - if not specified, set a default value. */
254		if (!tb[TCA_MPLS_TTL]) {
255#if IS_ENABLED(CONFIG_MPLS)
256			mpls_ttl = net->mpls.default_ttl ?
257				   net->mpls.default_ttl : ACT_MPLS_TTL_DEFAULT;
258#else
259			mpls_ttl = ACT_MPLS_TTL_DEFAULT;
260#endif
261		}
262		break;
263	case TCA_MPLS_ACT_MODIFY:
264		if (tb[TCA_MPLS_PROTO]) {
265			NL_SET_ERR_MSG_MOD(extack, "Protocol cannot be used with MPLS modify");
266			err = -EINVAL;
267			goto release_idr;
268		}
269		break;
270	default:
271		NL_SET_ERR_MSG_MOD(extack, "Unknown MPLS action");
272		err = -EINVAL;
273		goto release_idr;
274	}
275
276	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
277	if (err < 0)
278		goto release_idr;
279
280	m = to_mpls(*a);
281
282	p = kzalloc(sizeof(*p), GFP_KERNEL);
283	if (!p) {
284		err = -ENOMEM;
285		goto put_chain;
286	}
287
288	p->tcfm_action = parm->m_action;
289	p->tcfm_label = nla_get_u32_default(tb[TCA_MPLS_LABEL],
290					    ACT_MPLS_LABEL_NOT_SET);
291	p->tcfm_tc = nla_get_u8_default(tb[TCA_MPLS_TC], ACT_MPLS_TC_NOT_SET);
292	p->tcfm_ttl = nla_get_u8_default(tb[TCA_MPLS_TTL], mpls_ttl);
293	p->tcfm_bos = nla_get_u8_default(tb[TCA_MPLS_BOS],
294					 ACT_MPLS_BOS_NOT_SET);
295	p->tcfm_proto = nla_get_be16_default(tb[TCA_MPLS_PROTO],
296					     htons(ETH_P_MPLS_UC));
297	p->action = parm->action;
298
299	spin_lock_bh(&m->tcf_lock);
300	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
301	p = rcu_replace_pointer(m->mpls_p, p, lockdep_is_held(&m->tcf_lock));
302	spin_unlock_bh(&m->tcf_lock);
303
304	if (goto_ch)
305		tcf_chain_put_by_act(goto_ch);
306	if (p)
307		kfree_rcu(p, rcu);
308
309	return ret;
310put_chain:
311	if (goto_ch)
312		tcf_chain_put_by_act(goto_ch);
313release_idr:
314	tcf_idr_release(*a, bind);
315	return err;
316}
317
318static void tcf_mpls_cleanup(struct tc_action *a)
319{
320	struct tcf_mpls *m = to_mpls(a);
321	struct tcf_mpls_params *p;
322
323	p = rcu_dereference_protected(m->mpls_p, 1);
324	if (p)
325		kfree_rcu(p, rcu);
326}
327
328static int tcf_mpls_dump(struct sk_buff *skb, struct tc_action *a,
329			 int bind, int ref)
330{
331	unsigned char *b = skb_tail_pointer(skb);
332	const struct tcf_mpls *m = to_mpls(a);
333	const struct tcf_mpls_params *p;
334	struct tc_mpls opt = {
335		.index    = m->tcf_index,
336		.refcnt   = refcount_read(&m->tcf_refcnt) - ref,
337		.bindcnt  = atomic_read(&m->tcf_bindcnt) - bind,
338	};
339	struct tcf_t t;
340
341	rcu_read_lock();
342	p = rcu_dereference(m->mpls_p);
343	opt.m_action = p->tcfm_action;
344	opt.action = p->action;
345
346	if (nla_put(skb, TCA_MPLS_PARMS, sizeof(opt), &opt))
347		goto nla_put_failure;
348
349	if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET &&
350	    nla_put_u32(skb, TCA_MPLS_LABEL, p->tcfm_label))
351		goto nla_put_failure;
352
353	if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET &&
354	    nla_put_u8(skb, TCA_MPLS_TC, p->tcfm_tc))
355		goto nla_put_failure;
356
357	if (p->tcfm_ttl && nla_put_u8(skb, TCA_MPLS_TTL, p->tcfm_ttl))
358		goto nla_put_failure;
359
360	if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET &&
361	    nla_put_u8(skb, TCA_MPLS_BOS, p->tcfm_bos))
362		goto nla_put_failure;
363
364	if (nla_put_be16(skb, TCA_MPLS_PROTO, p->tcfm_proto))
365		goto nla_put_failure;
366
367	tcf_tm_dump(&t, &m->tcf_tm);
368
369	if (nla_put_64bit(skb, TCA_MPLS_TM, sizeof(t), &t, TCA_MPLS_PAD))
370		goto nla_put_failure;
371
372	rcu_read_unlock();
373
374	return skb->len;
375
376nla_put_failure:
377	rcu_read_unlock();
378	nlmsg_trim(skb, b);
379	return -EMSGSIZE;
380}
381
382static int tcf_mpls_offload_act_setup(struct tc_action *act, void *entry_data,
383				      u32 *index_inc, bool bind,
384				      struct netlink_ext_ack *extack)
385{
386	if (bind) {
387		struct flow_action_entry *entry = entry_data;
388
389		switch (tcf_mpls_action(act)) {
390		case TCA_MPLS_ACT_PUSH:
391			entry->id = FLOW_ACTION_MPLS_PUSH;
392			entry->mpls_push.proto = tcf_mpls_proto(act);
393			entry->mpls_push.label = tcf_mpls_label(act);
394			entry->mpls_push.tc = tcf_mpls_tc(act);
395			entry->mpls_push.bos = tcf_mpls_bos(act);
396			entry->mpls_push.ttl = tcf_mpls_ttl(act);
397			break;
398		case TCA_MPLS_ACT_POP:
399			entry->id = FLOW_ACTION_MPLS_POP;
400			entry->mpls_pop.proto = tcf_mpls_proto(act);
401			break;
402		case TCA_MPLS_ACT_MODIFY:
403			entry->id = FLOW_ACTION_MPLS_MANGLE;
404			entry->mpls_mangle.label = tcf_mpls_label(act);
405			entry->mpls_mangle.tc = tcf_mpls_tc(act);
406			entry->mpls_mangle.bos = tcf_mpls_bos(act);
407			entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
408			break;
409		case TCA_MPLS_ACT_DEC_TTL:
410			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"dec_ttl\" option is used");
411			return -EOPNOTSUPP;
412		case TCA_MPLS_ACT_MAC_PUSH:
413			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"mac_push\" option is used");
414			return -EOPNOTSUPP;
415		default:
416			NL_SET_ERR_MSG_MOD(extack, "Unsupported MPLS mode offload");
417			return -EOPNOTSUPP;
418		}
419		*index_inc = 1;
420	} else {
421		struct flow_offload_action *fl_action = entry_data;
422
423		switch (tcf_mpls_action(act)) {
424		case TCA_MPLS_ACT_PUSH:
425			fl_action->id = FLOW_ACTION_MPLS_PUSH;
426			break;
427		case TCA_MPLS_ACT_POP:
428			fl_action->id = FLOW_ACTION_MPLS_POP;
429			break;
430		case TCA_MPLS_ACT_MODIFY:
431			fl_action->id = FLOW_ACTION_MPLS_MANGLE;
432			break;
433		default:
434			return -EOPNOTSUPP;
435		}
436	}
437
438	return 0;
439}
440
441static struct tc_action_ops act_mpls_ops = {
442	.kind		=	"mpls",
443	.id		=	TCA_ID_MPLS,
444	.owner		=	THIS_MODULE,
445	.act		=	tcf_mpls_act,
446	.dump		=	tcf_mpls_dump,
447	.init		=	tcf_mpls_init,
448	.cleanup	=	tcf_mpls_cleanup,
449	.offload_act_setup =	tcf_mpls_offload_act_setup,
450	.size		=	sizeof(struct tcf_mpls),
451};
452MODULE_ALIAS_NET_ACT("mpls");
453
454static __net_init int mpls_init_net(struct net *net)
455{
456	struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id);
457
458	return tc_action_net_init(net, tn, &act_mpls_ops);
459}
460
461static void __net_exit mpls_exit_net(struct list_head *net_list)
462{
463	tc_action_net_exit(net_list, act_mpls_ops.net_id);
464}
465
466static struct pernet_operations mpls_net_ops = {
467	.init = mpls_init_net,
468	.exit_batch = mpls_exit_net,
469	.id   = &act_mpls_ops.net_id,
470	.size = sizeof(struct tc_action_net),
471};
472
473static int __init mpls_init_module(void)
474{
475	return tcf_register_action(&act_mpls_ops, &mpls_net_ops);
476}
477
478static void __exit mpls_cleanup_module(void)
479{
480	tcf_unregister_action(&act_mpls_ops, &mpls_net_ops);
481}
482
483module_init(mpls_init_module);
484module_exit(mpls_cleanup_module);
485
486MODULE_SOFTDEP("post: mpls_gso");
487MODULE_AUTHOR("Netronome Systems <oss-drivers@netronome.com>");
488MODULE_LICENSE("GPL");
489MODULE_DESCRIPTION("MPLS manipulation actions");