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