Loading...
1#include <linux/types.h>
2#include <linux/skbuff.h>
3#include <linux/socket.h>
4#include <linux/sysctl.h>
5#include <linux/net.h>
6#include <linux/module.h>
7#include <linux/if_arp.h>
8#include <linux/ipv6.h>
9#include <linux/mpls.h>
10#include <linux/vmalloc.h>
11#include <net/ip.h>
12#include <net/dst.h>
13#include <net/sock.h>
14#include <net/arp.h>
15#include <net/ip_fib.h>
16#include <net/netevent.h>
17#include <net/netns/generic.h>
18#if IS_ENABLED(CONFIG_IPV6)
19#include <net/ipv6.h>
20#include <net/addrconf.h>
21#endif
22#include <net/nexthop.h>
23#include "internal.h"
24
25/* Maximum number of labels to look ahead at when selecting a path of
26 * a multipath route
27 */
28#define MAX_MP_SELECT_LABELS 4
29
30#define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
31
32static int zero = 0;
33static int label_limit = (1 << 20) - 1;
34
35static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
36 struct nlmsghdr *nlh, struct net *net, u32 portid,
37 unsigned int nlm_flags);
38
39static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
40{
41 struct mpls_route *rt = NULL;
42
43 if (index < net->mpls.platform_labels) {
44 struct mpls_route __rcu **platform_label =
45 rcu_dereference(net->mpls.platform_label);
46 rt = rcu_dereference(platform_label[index]);
47 }
48 return rt;
49}
50
51static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
52{
53 return rcu_dereference_rtnl(dev->mpls_ptr);
54}
55
56bool mpls_output_possible(const struct net_device *dev)
57{
58 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
59}
60EXPORT_SYMBOL_GPL(mpls_output_possible);
61
62static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
63{
64 u8 *nh0_via = PTR_ALIGN((u8 *)&rt->rt_nh[rt->rt_nhn], VIA_ALEN_ALIGN);
65 int nh_index = nh - rt->rt_nh;
66
67 return nh0_via + rt->rt_max_alen * nh_index;
68}
69
70static const u8 *mpls_nh_via(const struct mpls_route *rt,
71 const struct mpls_nh *nh)
72{
73 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
74}
75
76static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
77{
78 /* The size of the layer 2.5 labels to be added for this route */
79 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
80}
81
82unsigned int mpls_dev_mtu(const struct net_device *dev)
83{
84 /* The amount of data the layer 2 frame can hold */
85 return dev->mtu;
86}
87EXPORT_SYMBOL_GPL(mpls_dev_mtu);
88
89bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
90{
91 if (skb->len <= mtu)
92 return false;
93
94 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
95 return false;
96
97 return true;
98}
99EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
100
101static u32 mpls_multipath_hash(struct mpls_route *rt,
102 struct sk_buff *skb, bool bos)
103{
104 struct mpls_entry_decoded dec;
105 struct mpls_shim_hdr *hdr;
106 bool eli_seen = false;
107 int label_index;
108 u32 hash = 0;
109
110 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS && !bos;
111 label_index++) {
112 if (!pskb_may_pull(skb, sizeof(*hdr) * label_index))
113 break;
114
115 /* Read and decode the current label */
116 hdr = mpls_hdr(skb) + label_index;
117 dec = mpls_entry_decode(hdr);
118
119 /* RFC6790 - reserved labels MUST NOT be used as keys
120 * for the load-balancing function
121 */
122 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
123 hash = jhash_1word(dec.label, hash);
124
125 /* The entropy label follows the entropy label
126 * indicator, so this means that the entropy
127 * label was just added to the hash - no need to
128 * go any deeper either in the label stack or in the
129 * payload
130 */
131 if (eli_seen)
132 break;
133 } else if (dec.label == MPLS_LABEL_ENTROPY) {
134 eli_seen = true;
135 }
136
137 bos = dec.bos;
138 if (bos && pskb_may_pull(skb, sizeof(*hdr) * label_index +
139 sizeof(struct iphdr))) {
140 const struct iphdr *v4hdr;
141
142 v4hdr = (const struct iphdr *)(mpls_hdr(skb) +
143 label_index);
144 if (v4hdr->version == 4) {
145 hash = jhash_3words(ntohl(v4hdr->saddr),
146 ntohl(v4hdr->daddr),
147 v4hdr->protocol, hash);
148 } else if (v4hdr->version == 6 &&
149 pskb_may_pull(skb, sizeof(*hdr) * label_index +
150 sizeof(struct ipv6hdr))) {
151 const struct ipv6hdr *v6hdr;
152
153 v6hdr = (const struct ipv6hdr *)(mpls_hdr(skb) +
154 label_index);
155
156 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
157 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
158 hash = jhash_1word(v6hdr->nexthdr, hash);
159 }
160 }
161 }
162
163 return hash;
164}
165
166static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
167 struct sk_buff *skb, bool bos)
168{
169 int alive = ACCESS_ONCE(rt->rt_nhn_alive);
170 u32 hash = 0;
171 int nh_index = 0;
172 int n = 0;
173
174 /* No need to look further into packet if there's only
175 * one path
176 */
177 if (rt->rt_nhn == 1)
178 goto out;
179
180 if (alive <= 0)
181 return NULL;
182
183 hash = mpls_multipath_hash(rt, skb, bos);
184 nh_index = hash % alive;
185 if (alive == rt->rt_nhn)
186 goto out;
187 for_nexthops(rt) {
188 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
189 continue;
190 if (n == nh_index)
191 return nh;
192 n++;
193 } endfor_nexthops(rt);
194
195out:
196 return &rt->rt_nh[nh_index];
197}
198
199static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
200 struct mpls_entry_decoded dec)
201{
202 enum mpls_payload_type payload_type;
203 bool success = false;
204
205 /* The IPv4 code below accesses through the IPv4 header
206 * checksum, which is 12 bytes into the packet.
207 * The IPv6 code below accesses through the IPv6 hop limit
208 * which is 8 bytes into the packet.
209 *
210 * For all supported cases there should always be at least 12
211 * bytes of packet data present. The IPv4 header is 20 bytes
212 * without options and the IPv6 header is always 40 bytes
213 * long.
214 */
215 if (!pskb_may_pull(skb, 12))
216 return false;
217
218 payload_type = rt->rt_payload_type;
219 if (payload_type == MPT_UNSPEC)
220 payload_type = ip_hdr(skb)->version;
221
222 switch (payload_type) {
223 case MPT_IPV4: {
224 struct iphdr *hdr4 = ip_hdr(skb);
225 skb->protocol = htons(ETH_P_IP);
226 csum_replace2(&hdr4->check,
227 htons(hdr4->ttl << 8),
228 htons(dec.ttl << 8));
229 hdr4->ttl = dec.ttl;
230 success = true;
231 break;
232 }
233 case MPT_IPV6: {
234 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
235 skb->protocol = htons(ETH_P_IPV6);
236 hdr6->hop_limit = dec.ttl;
237 success = true;
238 break;
239 }
240 case MPT_UNSPEC:
241 break;
242 }
243
244 return success;
245}
246
247static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
248 struct packet_type *pt, struct net_device *orig_dev)
249{
250 struct net *net = dev_net(dev);
251 struct mpls_shim_hdr *hdr;
252 struct mpls_route *rt;
253 struct mpls_nh *nh;
254 struct mpls_entry_decoded dec;
255 struct net_device *out_dev;
256 struct mpls_dev *mdev;
257 unsigned int hh_len;
258 unsigned int new_header_size;
259 unsigned int mtu;
260 int err;
261
262 /* Careful this entire function runs inside of an rcu critical section */
263
264 mdev = mpls_dev_get(dev);
265 if (!mdev || !mdev->input_enabled)
266 goto drop;
267
268 if (skb->pkt_type != PACKET_HOST)
269 goto drop;
270
271 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
272 goto drop;
273
274 if (!pskb_may_pull(skb, sizeof(*hdr)))
275 goto drop;
276
277 /* Read and decode the label */
278 hdr = mpls_hdr(skb);
279 dec = mpls_entry_decode(hdr);
280
281 /* Pop the label */
282 skb_pull(skb, sizeof(*hdr));
283 skb_reset_network_header(skb);
284
285 skb_orphan(skb);
286
287 rt = mpls_route_input_rcu(net, dec.label);
288 if (!rt)
289 goto drop;
290
291 nh = mpls_select_multipath(rt, skb, dec.bos);
292 if (!nh)
293 goto drop;
294
295 /* Find the output device */
296 out_dev = rcu_dereference(nh->nh_dev);
297 if (!mpls_output_possible(out_dev))
298 goto drop;
299
300 if (skb_warn_if_lro(skb))
301 goto drop;
302
303 skb_forward_csum(skb);
304
305 /* Verify ttl is valid */
306 if (dec.ttl <= 1)
307 goto drop;
308 dec.ttl -= 1;
309
310 /* Verify the destination can hold the packet */
311 new_header_size = mpls_nh_header_size(nh);
312 mtu = mpls_dev_mtu(out_dev);
313 if (mpls_pkt_too_big(skb, mtu - new_header_size))
314 goto drop;
315
316 hh_len = LL_RESERVED_SPACE(out_dev);
317 if (!out_dev->header_ops)
318 hh_len = 0;
319
320 /* Ensure there is enough space for the headers in the skb */
321 if (skb_cow(skb, hh_len + new_header_size))
322 goto drop;
323
324 skb->dev = out_dev;
325 skb->protocol = htons(ETH_P_MPLS_UC);
326
327 if (unlikely(!new_header_size && dec.bos)) {
328 /* Penultimate hop popping */
329 if (!mpls_egress(rt, skb, dec))
330 goto drop;
331 } else {
332 bool bos;
333 int i;
334 skb_push(skb, new_header_size);
335 skb_reset_network_header(skb);
336 /* Push the new labels */
337 hdr = mpls_hdr(skb);
338 bos = dec.bos;
339 for (i = nh->nh_labels - 1; i >= 0; i--) {
340 hdr[i] = mpls_entry_encode(nh->nh_label[i],
341 dec.ttl, 0, bos);
342 bos = false;
343 }
344 }
345
346 /* If via wasn't specified then send out using device address */
347 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
348 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
349 out_dev->dev_addr, skb);
350 else
351 err = neigh_xmit(nh->nh_via_table, out_dev,
352 mpls_nh_via(rt, nh), skb);
353 if (err)
354 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
355 __func__, err);
356 return 0;
357
358drop:
359 kfree_skb(skb);
360 return NET_RX_DROP;
361}
362
363static struct packet_type mpls_packet_type __read_mostly = {
364 .type = cpu_to_be16(ETH_P_MPLS_UC),
365 .func = mpls_forward,
366};
367
368static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
369 [RTA_DST] = { .type = NLA_U32 },
370 [RTA_OIF] = { .type = NLA_U32 },
371};
372
373struct mpls_route_config {
374 u32 rc_protocol;
375 u32 rc_ifindex;
376 u8 rc_via_table;
377 u8 rc_via_alen;
378 u8 rc_via[MAX_VIA_ALEN];
379 u32 rc_label;
380 u8 rc_output_labels;
381 u32 rc_output_label[MAX_NEW_LABELS];
382 u32 rc_nlflags;
383 enum mpls_payload_type rc_payload_type;
384 struct nl_info rc_nlinfo;
385 struct rtnexthop *rc_mp;
386 int rc_mp_len;
387};
388
389static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
390{
391 u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
392 struct mpls_route *rt;
393
394 rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
395 VIA_ALEN_ALIGN) +
396 num_nh * max_alen_aligned,
397 GFP_KERNEL);
398 if (rt) {
399 rt->rt_nhn = num_nh;
400 rt->rt_nhn_alive = num_nh;
401 rt->rt_max_alen = max_alen_aligned;
402 }
403
404 return rt;
405}
406
407static void mpls_rt_free(struct mpls_route *rt)
408{
409 if (rt)
410 kfree_rcu(rt, rt_rcu);
411}
412
413static void mpls_notify_route(struct net *net, unsigned index,
414 struct mpls_route *old, struct mpls_route *new,
415 const struct nl_info *info)
416{
417 struct nlmsghdr *nlh = info ? info->nlh : NULL;
418 unsigned portid = info ? info->portid : 0;
419 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
420 struct mpls_route *rt = new ? new : old;
421 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
422 /* Ignore reserved labels for now */
423 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
424 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
425}
426
427static void mpls_route_update(struct net *net, unsigned index,
428 struct mpls_route *new,
429 const struct nl_info *info)
430{
431 struct mpls_route __rcu **platform_label;
432 struct mpls_route *rt;
433
434 ASSERT_RTNL();
435
436 platform_label = rtnl_dereference(net->mpls.platform_label);
437 rt = rtnl_dereference(platform_label[index]);
438 rcu_assign_pointer(platform_label[index], new);
439
440 mpls_notify_route(net, index, rt, new, info);
441
442 /* If we removed a route free it now */
443 mpls_rt_free(rt);
444}
445
446static unsigned find_free_label(struct net *net)
447{
448 struct mpls_route __rcu **platform_label;
449 size_t platform_labels;
450 unsigned index;
451
452 platform_label = rtnl_dereference(net->mpls.platform_label);
453 platform_labels = net->mpls.platform_labels;
454 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
455 index++) {
456 if (!rtnl_dereference(platform_label[index]))
457 return index;
458 }
459 return LABEL_NOT_SPECIFIED;
460}
461
462#if IS_ENABLED(CONFIG_INET)
463static struct net_device *inet_fib_lookup_dev(struct net *net,
464 const void *addr)
465{
466 struct net_device *dev;
467 struct rtable *rt;
468 struct in_addr daddr;
469
470 memcpy(&daddr, addr, sizeof(struct in_addr));
471 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
472 if (IS_ERR(rt))
473 return ERR_CAST(rt);
474
475 dev = rt->dst.dev;
476 dev_hold(dev);
477
478 ip_rt_put(rt);
479
480 return dev;
481}
482#else
483static struct net_device *inet_fib_lookup_dev(struct net *net,
484 const void *addr)
485{
486 return ERR_PTR(-EAFNOSUPPORT);
487}
488#endif
489
490#if IS_ENABLED(CONFIG_IPV6)
491static struct net_device *inet6_fib_lookup_dev(struct net *net,
492 const void *addr)
493{
494 struct net_device *dev;
495 struct dst_entry *dst;
496 struct flowi6 fl6;
497 int err;
498
499 if (!ipv6_stub)
500 return ERR_PTR(-EAFNOSUPPORT);
501
502 memset(&fl6, 0, sizeof(fl6));
503 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
504 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
505 if (err)
506 return ERR_PTR(err);
507
508 dev = dst->dev;
509 dev_hold(dev);
510 dst_release(dst);
511
512 return dev;
513}
514#else
515static struct net_device *inet6_fib_lookup_dev(struct net *net,
516 const void *addr)
517{
518 return ERR_PTR(-EAFNOSUPPORT);
519}
520#endif
521
522static struct net_device *find_outdev(struct net *net,
523 struct mpls_route *rt,
524 struct mpls_nh *nh, int oif)
525{
526 struct net_device *dev = NULL;
527
528 if (!oif) {
529 switch (nh->nh_via_table) {
530 case NEIGH_ARP_TABLE:
531 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
532 break;
533 case NEIGH_ND_TABLE:
534 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
535 break;
536 case NEIGH_LINK_TABLE:
537 break;
538 }
539 } else {
540 dev = dev_get_by_index(net, oif);
541 }
542
543 if (!dev)
544 return ERR_PTR(-ENODEV);
545
546 if (IS_ERR(dev))
547 return dev;
548
549 /* The caller is holding rtnl anyways, so release the dev reference */
550 dev_put(dev);
551
552 return dev;
553}
554
555static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
556 struct mpls_nh *nh, int oif)
557{
558 struct net_device *dev = NULL;
559 int err = -ENODEV;
560
561 dev = find_outdev(net, rt, nh, oif);
562 if (IS_ERR(dev)) {
563 err = PTR_ERR(dev);
564 dev = NULL;
565 goto errout;
566 }
567
568 /* Ensure this is a supported device */
569 err = -EINVAL;
570 if (!mpls_dev_get(dev))
571 goto errout;
572
573 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
574 (dev->addr_len != nh->nh_via_alen))
575 goto errout;
576
577 RCU_INIT_POINTER(nh->nh_dev, dev);
578
579 if (!(dev->flags & IFF_UP)) {
580 nh->nh_flags |= RTNH_F_DEAD;
581 } else {
582 unsigned int flags;
583
584 flags = dev_get_flags(dev);
585 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
586 nh->nh_flags |= RTNH_F_LINKDOWN;
587 }
588
589 return 0;
590
591errout:
592 return err;
593}
594
595static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
596 struct mpls_route *rt)
597{
598 struct net *net = cfg->rc_nlinfo.nl_net;
599 struct mpls_nh *nh = rt->rt_nh;
600 int err;
601 int i;
602
603 if (!nh)
604 return -ENOMEM;
605
606 err = -EINVAL;
607 /* Ensure only a supported number of labels are present */
608 if (cfg->rc_output_labels > MAX_NEW_LABELS)
609 goto errout;
610
611 nh->nh_labels = cfg->rc_output_labels;
612 for (i = 0; i < nh->nh_labels; i++)
613 nh->nh_label[i] = cfg->rc_output_label[i];
614
615 nh->nh_via_table = cfg->rc_via_table;
616 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
617 nh->nh_via_alen = cfg->rc_via_alen;
618
619 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
620 if (err)
621 goto errout;
622
623 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
624 rt->rt_nhn_alive--;
625
626 return 0;
627
628errout:
629 return err;
630}
631
632static int mpls_nh_build(struct net *net, struct mpls_route *rt,
633 struct mpls_nh *nh, int oif, struct nlattr *via,
634 struct nlattr *newdst)
635{
636 int err = -ENOMEM;
637
638 if (!nh)
639 goto errout;
640
641 if (newdst) {
642 err = nla_get_labels(newdst, MAX_NEW_LABELS,
643 &nh->nh_labels, nh->nh_label);
644 if (err)
645 goto errout;
646 }
647
648 if (via) {
649 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
650 __mpls_nh_via(rt, nh));
651 if (err)
652 goto errout;
653 } else {
654 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
655 }
656
657 err = mpls_nh_assign_dev(net, rt, nh, oif);
658 if (err)
659 goto errout;
660
661 return 0;
662
663errout:
664 return err;
665}
666
667static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
668 u8 cfg_via_alen, u8 *max_via_alen)
669{
670 int nhs = 0;
671 int remaining = len;
672
673 if (!rtnh) {
674 *max_via_alen = cfg_via_alen;
675 return 1;
676 }
677
678 *max_via_alen = 0;
679
680 while (rtnh_ok(rtnh, remaining)) {
681 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
682 int attrlen;
683
684 attrlen = rtnh_attrlen(rtnh);
685 nla = nla_find(attrs, attrlen, RTA_VIA);
686 if (nla && nla_len(nla) >=
687 offsetof(struct rtvia, rtvia_addr)) {
688 int via_alen = nla_len(nla) -
689 offsetof(struct rtvia, rtvia_addr);
690
691 if (via_alen <= MAX_VIA_ALEN)
692 *max_via_alen = max_t(u16, *max_via_alen,
693 via_alen);
694 }
695
696 nhs++;
697 rtnh = rtnh_next(rtnh, &remaining);
698 }
699
700 /* leftover implies invalid nexthop configuration, discard it */
701 return remaining > 0 ? 0 : nhs;
702}
703
704static int mpls_nh_build_multi(struct mpls_route_config *cfg,
705 struct mpls_route *rt)
706{
707 struct rtnexthop *rtnh = cfg->rc_mp;
708 struct nlattr *nla_via, *nla_newdst;
709 int remaining = cfg->rc_mp_len;
710 int nhs = 0;
711 int err = 0;
712
713 change_nexthops(rt) {
714 int attrlen;
715
716 nla_via = NULL;
717 nla_newdst = NULL;
718
719 err = -EINVAL;
720 if (!rtnh_ok(rtnh, remaining))
721 goto errout;
722
723 /* neither weighted multipath nor any flags
724 * are supported
725 */
726 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
727 goto errout;
728
729 attrlen = rtnh_attrlen(rtnh);
730 if (attrlen > 0) {
731 struct nlattr *attrs = rtnh_attrs(rtnh);
732
733 nla_via = nla_find(attrs, attrlen, RTA_VIA);
734 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
735 }
736
737 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
738 rtnh->rtnh_ifindex, nla_via, nla_newdst);
739 if (err)
740 goto errout;
741
742 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
743 rt->rt_nhn_alive--;
744
745 rtnh = rtnh_next(rtnh, &remaining);
746 nhs++;
747 } endfor_nexthops(rt);
748
749 rt->rt_nhn = nhs;
750
751 return 0;
752
753errout:
754 return err;
755}
756
757static int mpls_route_add(struct mpls_route_config *cfg)
758{
759 struct mpls_route __rcu **platform_label;
760 struct net *net = cfg->rc_nlinfo.nl_net;
761 struct mpls_route *rt, *old;
762 int err = -EINVAL;
763 u8 max_via_alen;
764 unsigned index;
765 int nhs;
766
767 index = cfg->rc_label;
768
769 /* If a label was not specified during insert pick one */
770 if ((index == LABEL_NOT_SPECIFIED) &&
771 (cfg->rc_nlflags & NLM_F_CREATE)) {
772 index = find_free_label(net);
773 }
774
775 /* Reserved labels may not be set */
776 if (index < MPLS_LABEL_FIRST_UNRESERVED)
777 goto errout;
778
779 /* The full 20 bit range may not be supported. */
780 if (index >= net->mpls.platform_labels)
781 goto errout;
782
783 /* Append makes no sense with mpls */
784 err = -EOPNOTSUPP;
785 if (cfg->rc_nlflags & NLM_F_APPEND)
786 goto errout;
787
788 err = -EEXIST;
789 platform_label = rtnl_dereference(net->mpls.platform_label);
790 old = rtnl_dereference(platform_label[index]);
791 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
792 goto errout;
793
794 err = -EEXIST;
795 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
796 goto errout;
797
798 err = -ENOENT;
799 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
800 goto errout;
801
802 err = -EINVAL;
803 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
804 cfg->rc_via_alen, &max_via_alen);
805 if (nhs == 0)
806 goto errout;
807
808 err = -ENOMEM;
809 rt = mpls_rt_alloc(nhs, max_via_alen);
810 if (!rt)
811 goto errout;
812
813 rt->rt_protocol = cfg->rc_protocol;
814 rt->rt_payload_type = cfg->rc_payload_type;
815
816 if (cfg->rc_mp)
817 err = mpls_nh_build_multi(cfg, rt);
818 else
819 err = mpls_nh_build_from_cfg(cfg, rt);
820 if (err)
821 goto freert;
822
823 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
824
825 return 0;
826
827freert:
828 mpls_rt_free(rt);
829errout:
830 return err;
831}
832
833static int mpls_route_del(struct mpls_route_config *cfg)
834{
835 struct net *net = cfg->rc_nlinfo.nl_net;
836 unsigned index;
837 int err = -EINVAL;
838
839 index = cfg->rc_label;
840
841 /* Reserved labels may not be removed */
842 if (index < MPLS_LABEL_FIRST_UNRESERVED)
843 goto errout;
844
845 /* The full 20 bit range may not be supported */
846 if (index >= net->mpls.platform_labels)
847 goto errout;
848
849 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
850
851 err = 0;
852errout:
853 return err;
854}
855
856#define MPLS_PERDEV_SYSCTL_OFFSET(field) \
857 (&((struct mpls_dev *)0)->field)
858
859static const struct ctl_table mpls_dev_table[] = {
860 {
861 .procname = "input",
862 .maxlen = sizeof(int),
863 .mode = 0644,
864 .proc_handler = proc_dointvec,
865 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
866 },
867 { }
868};
869
870static int mpls_dev_sysctl_register(struct net_device *dev,
871 struct mpls_dev *mdev)
872{
873 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
874 struct ctl_table *table;
875 int i;
876
877 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
878 if (!table)
879 goto out;
880
881 /* Table data contains only offsets relative to the base of
882 * the mdev at this point, so make them absolute.
883 */
884 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
885 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
886
887 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
888
889 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
890 if (!mdev->sysctl)
891 goto free;
892
893 return 0;
894
895free:
896 kfree(table);
897out:
898 return -ENOBUFS;
899}
900
901static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
902{
903 struct ctl_table *table;
904
905 table = mdev->sysctl->ctl_table_arg;
906 unregister_net_sysctl_table(mdev->sysctl);
907 kfree(table);
908}
909
910static struct mpls_dev *mpls_add_dev(struct net_device *dev)
911{
912 struct mpls_dev *mdev;
913 int err = -ENOMEM;
914
915 ASSERT_RTNL();
916
917 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
918 if (!mdev)
919 return ERR_PTR(err);
920
921 err = mpls_dev_sysctl_register(dev, mdev);
922 if (err)
923 goto free;
924
925 rcu_assign_pointer(dev->mpls_ptr, mdev);
926
927 return mdev;
928
929free:
930 kfree(mdev);
931 return ERR_PTR(err);
932}
933
934static void mpls_ifdown(struct net_device *dev, int event)
935{
936 struct mpls_route __rcu **platform_label;
937 struct net *net = dev_net(dev);
938 unsigned index;
939
940 platform_label = rtnl_dereference(net->mpls.platform_label);
941 for (index = 0; index < net->mpls.platform_labels; index++) {
942 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
943
944 if (!rt)
945 continue;
946
947 change_nexthops(rt) {
948 if (rtnl_dereference(nh->nh_dev) != dev)
949 continue;
950 switch (event) {
951 case NETDEV_DOWN:
952 case NETDEV_UNREGISTER:
953 nh->nh_flags |= RTNH_F_DEAD;
954 /* fall through */
955 case NETDEV_CHANGE:
956 nh->nh_flags |= RTNH_F_LINKDOWN;
957 ACCESS_ONCE(rt->rt_nhn_alive) = rt->rt_nhn_alive - 1;
958 break;
959 }
960 if (event == NETDEV_UNREGISTER)
961 RCU_INIT_POINTER(nh->nh_dev, NULL);
962 } endfor_nexthops(rt);
963 }
964
965
966 return;
967}
968
969static void mpls_ifup(struct net_device *dev, unsigned int nh_flags)
970{
971 struct mpls_route __rcu **platform_label;
972 struct net *net = dev_net(dev);
973 unsigned index;
974 int alive;
975
976 platform_label = rtnl_dereference(net->mpls.platform_label);
977 for (index = 0; index < net->mpls.platform_labels; index++) {
978 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
979
980 if (!rt)
981 continue;
982
983 alive = 0;
984 change_nexthops(rt) {
985 struct net_device *nh_dev =
986 rtnl_dereference(nh->nh_dev);
987
988 if (!(nh->nh_flags & nh_flags)) {
989 alive++;
990 continue;
991 }
992 if (nh_dev != dev)
993 continue;
994 alive++;
995 nh->nh_flags &= ~nh_flags;
996 } endfor_nexthops(rt);
997
998 ACCESS_ONCE(rt->rt_nhn_alive) = alive;
999 }
1000
1001 return;
1002}
1003
1004static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1005 void *ptr)
1006{
1007 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1008 struct mpls_dev *mdev;
1009 unsigned int flags;
1010
1011 if (event == NETDEV_REGISTER) {
1012 /* For now just support ethernet devices */
1013 if ((dev->type == ARPHRD_ETHER) ||
1014 (dev->type == ARPHRD_LOOPBACK)) {
1015 mdev = mpls_add_dev(dev);
1016 if (IS_ERR(mdev))
1017 return notifier_from_errno(PTR_ERR(mdev));
1018 }
1019 return NOTIFY_OK;
1020 }
1021
1022 mdev = mpls_dev_get(dev);
1023 if (!mdev)
1024 return NOTIFY_OK;
1025
1026 switch (event) {
1027 case NETDEV_DOWN:
1028 mpls_ifdown(dev, event);
1029 break;
1030 case NETDEV_UP:
1031 flags = dev_get_flags(dev);
1032 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1033 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1034 else
1035 mpls_ifup(dev, RTNH_F_DEAD);
1036 break;
1037 case NETDEV_CHANGE:
1038 flags = dev_get_flags(dev);
1039 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1040 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1041 else
1042 mpls_ifdown(dev, event);
1043 break;
1044 case NETDEV_UNREGISTER:
1045 mpls_ifdown(dev, event);
1046 mdev = mpls_dev_get(dev);
1047 if (mdev) {
1048 mpls_dev_sysctl_unregister(mdev);
1049 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1050 kfree_rcu(mdev, rcu);
1051 }
1052 break;
1053 case NETDEV_CHANGENAME:
1054 mdev = mpls_dev_get(dev);
1055 if (mdev) {
1056 int err;
1057
1058 mpls_dev_sysctl_unregister(mdev);
1059 err = mpls_dev_sysctl_register(dev, mdev);
1060 if (err)
1061 return notifier_from_errno(err);
1062 }
1063 break;
1064 }
1065 return NOTIFY_OK;
1066}
1067
1068static struct notifier_block mpls_dev_notifier = {
1069 .notifier_call = mpls_dev_notify,
1070};
1071
1072static int nla_put_via(struct sk_buff *skb,
1073 u8 table, const void *addr, int alen)
1074{
1075 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1076 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1077 };
1078 struct nlattr *nla;
1079 struct rtvia *via;
1080 int family = AF_UNSPEC;
1081
1082 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1083 if (!nla)
1084 return -EMSGSIZE;
1085
1086 if (table <= NEIGH_NR_TABLES)
1087 family = table_to_family[table];
1088
1089 via = nla_data(nla);
1090 via->rtvia_family = family;
1091 memcpy(via->rtvia_addr, addr, alen);
1092 return 0;
1093}
1094
1095int nla_put_labels(struct sk_buff *skb, int attrtype,
1096 u8 labels, const u32 label[])
1097{
1098 struct nlattr *nla;
1099 struct mpls_shim_hdr *nla_label;
1100 bool bos;
1101 int i;
1102 nla = nla_reserve(skb, attrtype, labels*4);
1103 if (!nla)
1104 return -EMSGSIZE;
1105
1106 nla_label = nla_data(nla);
1107 bos = true;
1108 for (i = labels - 1; i >= 0; i--) {
1109 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1110 bos = false;
1111 }
1112
1113 return 0;
1114}
1115EXPORT_SYMBOL_GPL(nla_put_labels);
1116
1117int nla_get_labels(const struct nlattr *nla,
1118 u32 max_labels, u8 *labels, u32 label[])
1119{
1120 unsigned len = nla_len(nla);
1121 unsigned nla_labels;
1122 struct mpls_shim_hdr *nla_label;
1123 bool bos;
1124 int i;
1125
1126 /* len needs to be an even multiple of 4 (the label size) */
1127 if (len & 3)
1128 return -EINVAL;
1129
1130 /* Limit the number of new labels allowed */
1131 nla_labels = len/4;
1132 if (nla_labels > max_labels)
1133 return -EINVAL;
1134
1135 nla_label = nla_data(nla);
1136 bos = true;
1137 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1138 struct mpls_entry_decoded dec;
1139 dec = mpls_entry_decode(nla_label + i);
1140
1141 /* Ensure the bottom of stack flag is properly set
1142 * and ttl and tc are both clear.
1143 */
1144 if ((dec.bos != bos) || dec.ttl || dec.tc)
1145 return -EINVAL;
1146
1147 switch (dec.label) {
1148 case MPLS_LABEL_IMPLNULL:
1149 /* RFC3032: This is a label that an LSR may
1150 * assign and distribute, but which never
1151 * actually appears in the encapsulation.
1152 */
1153 return -EINVAL;
1154 }
1155
1156 label[i] = dec.label;
1157 }
1158 *labels = nla_labels;
1159 return 0;
1160}
1161EXPORT_SYMBOL_GPL(nla_get_labels);
1162
1163int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1164 u8 *via_table, u8 via_addr[])
1165{
1166 struct rtvia *via = nla_data(nla);
1167 int err = -EINVAL;
1168 int alen;
1169
1170 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1171 goto errout;
1172 alen = nla_len(nla) -
1173 offsetof(struct rtvia, rtvia_addr);
1174 if (alen > MAX_VIA_ALEN)
1175 goto errout;
1176
1177 /* Validate the address family */
1178 switch (via->rtvia_family) {
1179 case AF_PACKET:
1180 *via_table = NEIGH_LINK_TABLE;
1181 break;
1182 case AF_INET:
1183 *via_table = NEIGH_ARP_TABLE;
1184 if (alen != 4)
1185 goto errout;
1186 break;
1187 case AF_INET6:
1188 *via_table = NEIGH_ND_TABLE;
1189 if (alen != 16)
1190 goto errout;
1191 break;
1192 default:
1193 /* Unsupported address family */
1194 goto errout;
1195 }
1196
1197 memcpy(via_addr, via->rtvia_addr, alen);
1198 *via_alen = alen;
1199 err = 0;
1200
1201errout:
1202 return err;
1203}
1204
1205static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1206 struct mpls_route_config *cfg)
1207{
1208 struct rtmsg *rtm;
1209 struct nlattr *tb[RTA_MAX+1];
1210 int index;
1211 int err;
1212
1213 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1214 if (err < 0)
1215 goto errout;
1216
1217 err = -EINVAL;
1218 rtm = nlmsg_data(nlh);
1219 memset(cfg, 0, sizeof(*cfg));
1220
1221 if (rtm->rtm_family != AF_MPLS)
1222 goto errout;
1223 if (rtm->rtm_dst_len != 20)
1224 goto errout;
1225 if (rtm->rtm_src_len != 0)
1226 goto errout;
1227 if (rtm->rtm_tos != 0)
1228 goto errout;
1229 if (rtm->rtm_table != RT_TABLE_MAIN)
1230 goto errout;
1231 /* Any value is acceptable for rtm_protocol */
1232
1233 /* As mpls uses destination specific addresses
1234 * (or source specific address in the case of multicast)
1235 * all addresses have universal scope.
1236 */
1237 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1238 goto errout;
1239 if (rtm->rtm_type != RTN_UNICAST)
1240 goto errout;
1241 if (rtm->rtm_flags != 0)
1242 goto errout;
1243
1244 cfg->rc_label = LABEL_NOT_SPECIFIED;
1245 cfg->rc_protocol = rtm->rtm_protocol;
1246 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1247 cfg->rc_nlflags = nlh->nlmsg_flags;
1248 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1249 cfg->rc_nlinfo.nlh = nlh;
1250 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1251
1252 for (index = 0; index <= RTA_MAX; index++) {
1253 struct nlattr *nla = tb[index];
1254 if (!nla)
1255 continue;
1256
1257 switch(index) {
1258 case RTA_OIF:
1259 cfg->rc_ifindex = nla_get_u32(nla);
1260 break;
1261 case RTA_NEWDST:
1262 if (nla_get_labels(nla, MAX_NEW_LABELS,
1263 &cfg->rc_output_labels,
1264 cfg->rc_output_label))
1265 goto errout;
1266 break;
1267 case RTA_DST:
1268 {
1269 u8 label_count;
1270 if (nla_get_labels(nla, 1, &label_count,
1271 &cfg->rc_label))
1272 goto errout;
1273
1274 /* Reserved labels may not be set */
1275 if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1276 goto errout;
1277
1278 break;
1279 }
1280 case RTA_VIA:
1281 {
1282 if (nla_get_via(nla, &cfg->rc_via_alen,
1283 &cfg->rc_via_table, cfg->rc_via))
1284 goto errout;
1285 break;
1286 }
1287 case RTA_MULTIPATH:
1288 {
1289 cfg->rc_mp = nla_data(nla);
1290 cfg->rc_mp_len = nla_len(nla);
1291 break;
1292 }
1293 default:
1294 /* Unsupported attribute */
1295 goto errout;
1296 }
1297 }
1298
1299 err = 0;
1300errout:
1301 return err;
1302}
1303
1304static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1305{
1306 struct mpls_route_config cfg;
1307 int err;
1308
1309 err = rtm_to_route_config(skb, nlh, &cfg);
1310 if (err < 0)
1311 return err;
1312
1313 return mpls_route_del(&cfg);
1314}
1315
1316
1317static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1318{
1319 struct mpls_route_config cfg;
1320 int err;
1321
1322 err = rtm_to_route_config(skb, nlh, &cfg);
1323 if (err < 0)
1324 return err;
1325
1326 return mpls_route_add(&cfg);
1327}
1328
1329static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1330 u32 label, struct mpls_route *rt, int flags)
1331{
1332 struct net_device *dev;
1333 struct nlmsghdr *nlh;
1334 struct rtmsg *rtm;
1335
1336 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1337 if (nlh == NULL)
1338 return -EMSGSIZE;
1339
1340 rtm = nlmsg_data(nlh);
1341 rtm->rtm_family = AF_MPLS;
1342 rtm->rtm_dst_len = 20;
1343 rtm->rtm_src_len = 0;
1344 rtm->rtm_tos = 0;
1345 rtm->rtm_table = RT_TABLE_MAIN;
1346 rtm->rtm_protocol = rt->rt_protocol;
1347 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1348 rtm->rtm_type = RTN_UNICAST;
1349 rtm->rtm_flags = 0;
1350
1351 if (nla_put_labels(skb, RTA_DST, 1, &label))
1352 goto nla_put_failure;
1353 if (rt->rt_nhn == 1) {
1354 const struct mpls_nh *nh = rt->rt_nh;
1355
1356 if (nh->nh_labels &&
1357 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1358 nh->nh_label))
1359 goto nla_put_failure;
1360 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1361 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1362 nh->nh_via_alen))
1363 goto nla_put_failure;
1364 dev = rtnl_dereference(nh->nh_dev);
1365 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1366 goto nla_put_failure;
1367 if (nh->nh_flags & RTNH_F_LINKDOWN)
1368 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1369 if (nh->nh_flags & RTNH_F_DEAD)
1370 rtm->rtm_flags |= RTNH_F_DEAD;
1371 } else {
1372 struct rtnexthop *rtnh;
1373 struct nlattr *mp;
1374 int dead = 0;
1375 int linkdown = 0;
1376
1377 mp = nla_nest_start(skb, RTA_MULTIPATH);
1378 if (!mp)
1379 goto nla_put_failure;
1380
1381 for_nexthops(rt) {
1382 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1383 if (!rtnh)
1384 goto nla_put_failure;
1385
1386 dev = rtnl_dereference(nh->nh_dev);
1387 if (dev)
1388 rtnh->rtnh_ifindex = dev->ifindex;
1389 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1390 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1391 linkdown++;
1392 }
1393 if (nh->nh_flags & RTNH_F_DEAD) {
1394 rtnh->rtnh_flags |= RTNH_F_DEAD;
1395 dead++;
1396 }
1397
1398 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1399 nh->nh_labels,
1400 nh->nh_label))
1401 goto nla_put_failure;
1402 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1403 nla_put_via(skb, nh->nh_via_table,
1404 mpls_nh_via(rt, nh),
1405 nh->nh_via_alen))
1406 goto nla_put_failure;
1407
1408 /* length of rtnetlink header + attributes */
1409 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1410 } endfor_nexthops(rt);
1411
1412 if (linkdown == rt->rt_nhn)
1413 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1414 if (dead == rt->rt_nhn)
1415 rtm->rtm_flags |= RTNH_F_DEAD;
1416
1417 nla_nest_end(skb, mp);
1418 }
1419
1420 nlmsg_end(skb, nlh);
1421 return 0;
1422
1423nla_put_failure:
1424 nlmsg_cancel(skb, nlh);
1425 return -EMSGSIZE;
1426}
1427
1428static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1429{
1430 struct net *net = sock_net(skb->sk);
1431 struct mpls_route __rcu **platform_label;
1432 size_t platform_labels;
1433 unsigned int index;
1434
1435 ASSERT_RTNL();
1436
1437 index = cb->args[0];
1438 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1439 index = MPLS_LABEL_FIRST_UNRESERVED;
1440
1441 platform_label = rtnl_dereference(net->mpls.platform_label);
1442 platform_labels = net->mpls.platform_labels;
1443 for (; index < platform_labels; index++) {
1444 struct mpls_route *rt;
1445 rt = rtnl_dereference(platform_label[index]);
1446 if (!rt)
1447 continue;
1448
1449 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1450 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1451 index, rt, NLM_F_MULTI) < 0)
1452 break;
1453 }
1454 cb->args[0] = index;
1455
1456 return skb->len;
1457}
1458
1459static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1460{
1461 size_t payload =
1462 NLMSG_ALIGN(sizeof(struct rtmsg))
1463 + nla_total_size(4); /* RTA_DST */
1464
1465 if (rt->rt_nhn == 1) {
1466 struct mpls_nh *nh = rt->rt_nh;
1467
1468 if (nh->nh_dev)
1469 payload += nla_total_size(4); /* RTA_OIF */
1470 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1471 payload += nla_total_size(2 + nh->nh_via_alen);
1472 if (nh->nh_labels) /* RTA_NEWDST */
1473 payload += nla_total_size(nh->nh_labels * 4);
1474 } else {
1475 /* each nexthop is packed in an attribute */
1476 size_t nhsize = 0;
1477
1478 for_nexthops(rt) {
1479 nhsize += nla_total_size(sizeof(struct rtnexthop));
1480 /* RTA_VIA */
1481 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1482 nhsize += nla_total_size(2 + nh->nh_via_alen);
1483 if (nh->nh_labels)
1484 nhsize += nla_total_size(nh->nh_labels * 4);
1485 } endfor_nexthops(rt);
1486 /* nested attribute */
1487 payload += nla_total_size(nhsize);
1488 }
1489
1490 return payload;
1491}
1492
1493static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1494 struct nlmsghdr *nlh, struct net *net, u32 portid,
1495 unsigned int nlm_flags)
1496{
1497 struct sk_buff *skb;
1498 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1499 int err = -ENOBUFS;
1500
1501 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1502 if (skb == NULL)
1503 goto errout;
1504
1505 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1506 if (err < 0) {
1507 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1508 WARN_ON(err == -EMSGSIZE);
1509 kfree_skb(skb);
1510 goto errout;
1511 }
1512 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1513
1514 return;
1515errout:
1516 if (err < 0)
1517 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1518}
1519
1520static int resize_platform_label_table(struct net *net, size_t limit)
1521{
1522 size_t size = sizeof(struct mpls_route *) * limit;
1523 size_t old_limit;
1524 size_t cp_size;
1525 struct mpls_route __rcu **labels = NULL, **old;
1526 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1527 unsigned index;
1528
1529 if (size) {
1530 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1531 if (!labels)
1532 labels = vzalloc(size);
1533
1534 if (!labels)
1535 goto nolabels;
1536 }
1537
1538 /* In case the predefined labels need to be populated */
1539 if (limit > MPLS_LABEL_IPV4NULL) {
1540 struct net_device *lo = net->loopback_dev;
1541 rt0 = mpls_rt_alloc(1, lo->addr_len);
1542 if (!rt0)
1543 goto nort0;
1544 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1545 rt0->rt_protocol = RTPROT_KERNEL;
1546 rt0->rt_payload_type = MPT_IPV4;
1547 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1548 rt0->rt_nh->nh_via_alen = lo->addr_len;
1549 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1550 lo->addr_len);
1551 }
1552 if (limit > MPLS_LABEL_IPV6NULL) {
1553 struct net_device *lo = net->loopback_dev;
1554 rt2 = mpls_rt_alloc(1, lo->addr_len);
1555 if (!rt2)
1556 goto nort2;
1557 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1558 rt2->rt_protocol = RTPROT_KERNEL;
1559 rt2->rt_payload_type = MPT_IPV6;
1560 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1561 rt2->rt_nh->nh_via_alen = lo->addr_len;
1562 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1563 lo->addr_len);
1564 }
1565
1566 rtnl_lock();
1567 /* Remember the original table */
1568 old = rtnl_dereference(net->mpls.platform_label);
1569 old_limit = net->mpls.platform_labels;
1570
1571 /* Free any labels beyond the new table */
1572 for (index = limit; index < old_limit; index++)
1573 mpls_route_update(net, index, NULL, NULL);
1574
1575 /* Copy over the old labels */
1576 cp_size = size;
1577 if (old_limit < limit)
1578 cp_size = old_limit * sizeof(struct mpls_route *);
1579
1580 memcpy(labels, old, cp_size);
1581
1582 /* If needed set the predefined labels */
1583 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1584 (limit > MPLS_LABEL_IPV6NULL)) {
1585 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1586 rt2 = NULL;
1587 }
1588
1589 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1590 (limit > MPLS_LABEL_IPV4NULL)) {
1591 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1592 rt0 = NULL;
1593 }
1594
1595 /* Update the global pointers */
1596 net->mpls.platform_labels = limit;
1597 rcu_assign_pointer(net->mpls.platform_label, labels);
1598
1599 rtnl_unlock();
1600
1601 mpls_rt_free(rt2);
1602 mpls_rt_free(rt0);
1603
1604 if (old) {
1605 synchronize_rcu();
1606 kvfree(old);
1607 }
1608 return 0;
1609
1610nort2:
1611 mpls_rt_free(rt0);
1612nort0:
1613 kvfree(labels);
1614nolabels:
1615 return -ENOMEM;
1616}
1617
1618static int mpls_platform_labels(struct ctl_table *table, int write,
1619 void __user *buffer, size_t *lenp, loff_t *ppos)
1620{
1621 struct net *net = table->data;
1622 int platform_labels = net->mpls.platform_labels;
1623 int ret;
1624 struct ctl_table tmp = {
1625 .procname = table->procname,
1626 .data = &platform_labels,
1627 .maxlen = sizeof(int),
1628 .mode = table->mode,
1629 .extra1 = &zero,
1630 .extra2 = &label_limit,
1631 };
1632
1633 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1634
1635 if (write && ret == 0)
1636 ret = resize_platform_label_table(net, platform_labels);
1637
1638 return ret;
1639}
1640
1641static const struct ctl_table mpls_table[] = {
1642 {
1643 .procname = "platform_labels",
1644 .data = NULL,
1645 .maxlen = sizeof(int),
1646 .mode = 0644,
1647 .proc_handler = mpls_platform_labels,
1648 },
1649 { }
1650};
1651
1652static int mpls_net_init(struct net *net)
1653{
1654 struct ctl_table *table;
1655
1656 net->mpls.platform_labels = 0;
1657 net->mpls.platform_label = NULL;
1658
1659 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1660 if (table == NULL)
1661 return -ENOMEM;
1662
1663 table[0].data = net;
1664 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1665 if (net->mpls.ctl == NULL) {
1666 kfree(table);
1667 return -ENOMEM;
1668 }
1669
1670 return 0;
1671}
1672
1673static void mpls_net_exit(struct net *net)
1674{
1675 struct mpls_route __rcu **platform_label;
1676 size_t platform_labels;
1677 struct ctl_table *table;
1678 unsigned int index;
1679
1680 table = net->mpls.ctl->ctl_table_arg;
1681 unregister_net_sysctl_table(net->mpls.ctl);
1682 kfree(table);
1683
1684 /* An rcu grace period has passed since there was a device in
1685 * the network namespace (and thus the last in flight packet)
1686 * left this network namespace. This is because
1687 * unregister_netdevice_many and netdev_run_todo has completed
1688 * for each network device that was in this network namespace.
1689 *
1690 * As such no additional rcu synchronization is necessary when
1691 * freeing the platform_label table.
1692 */
1693 rtnl_lock();
1694 platform_label = rtnl_dereference(net->mpls.platform_label);
1695 platform_labels = net->mpls.platform_labels;
1696 for (index = 0; index < platform_labels; index++) {
1697 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1698 RCU_INIT_POINTER(platform_label[index], NULL);
1699 mpls_rt_free(rt);
1700 }
1701 rtnl_unlock();
1702
1703 kvfree(platform_label);
1704}
1705
1706static struct pernet_operations mpls_net_ops = {
1707 .init = mpls_net_init,
1708 .exit = mpls_net_exit,
1709};
1710
1711static int __init mpls_init(void)
1712{
1713 int err;
1714
1715 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1716
1717 err = register_pernet_subsys(&mpls_net_ops);
1718 if (err)
1719 goto out;
1720
1721 err = register_netdevice_notifier(&mpls_dev_notifier);
1722 if (err)
1723 goto out_unregister_pernet;
1724
1725 dev_add_pack(&mpls_packet_type);
1726
1727 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1728 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1729 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1730 err = 0;
1731out:
1732 return err;
1733
1734out_unregister_pernet:
1735 unregister_pernet_subsys(&mpls_net_ops);
1736 goto out;
1737}
1738module_init(mpls_init);
1739
1740static void __exit mpls_exit(void)
1741{
1742 rtnl_unregister_all(PF_MPLS);
1743 dev_remove_pack(&mpls_packet_type);
1744 unregister_netdevice_notifier(&mpls_dev_notifier);
1745 unregister_pernet_subsys(&mpls_net_ops);
1746}
1747module_exit(mpls_exit);
1748
1749MODULE_DESCRIPTION("MultiProtocol Label Switching");
1750MODULE_LICENSE("GPL v2");
1751MODULE_ALIAS_NETPROTO(PF_MPLS);
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/types.h>
3#include <linux/skbuff.h>
4#include <linux/socket.h>
5#include <linux/sysctl.h>
6#include <linux/net.h>
7#include <linux/module.h>
8#include <linux/if_arp.h>
9#include <linux/ipv6.h>
10#include <linux/mpls.h>
11#include <linux/netconf.h>
12#include <linux/nospec.h>
13#include <linux/vmalloc.h>
14#include <linux/percpu.h>
15#include <net/ip.h>
16#include <net/dst.h>
17#include <net/sock.h>
18#include <net/arp.h>
19#include <net/ip_fib.h>
20#include <net/netevent.h>
21#include <net/ip_tunnels.h>
22#include <net/netns/generic.h>
23#if IS_ENABLED(CONFIG_IPV6)
24#include <net/ipv6.h>
25#endif
26#include <net/ipv6_stubs.h>
27#include <net/rtnh.h>
28#include "internal.h"
29
30/* max memory we will use for mpls_route */
31#define MAX_MPLS_ROUTE_MEM 4096
32
33/* Maximum number of labels to look ahead at when selecting a path of
34 * a multipath route
35 */
36#define MAX_MP_SELECT_LABELS 4
37
38#define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
39
40static int label_limit = (1 << 20) - 1;
41static int ttl_max = 255;
42
43#if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
44static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
45{
46 return sizeof(struct mpls_shim_hdr);
47}
48
49static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
50 .encap_hlen = ipgre_mpls_encap_hlen,
51};
52
53static int ipgre_tunnel_encap_add_mpls_ops(void)
54{
55 return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
56}
57
58static void ipgre_tunnel_encap_del_mpls_ops(void)
59{
60 ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
61}
62#else
63static int ipgre_tunnel_encap_add_mpls_ops(void)
64{
65 return 0;
66}
67
68static void ipgre_tunnel_encap_del_mpls_ops(void)
69{
70}
71#endif
72
73static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
74 struct nlmsghdr *nlh, struct net *net, u32 portid,
75 unsigned int nlm_flags);
76
77static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
78{
79 struct mpls_route *rt = NULL;
80
81 if (index < net->mpls.platform_labels) {
82 struct mpls_route __rcu **platform_label =
83 rcu_dereference(net->mpls.platform_label);
84 rt = rcu_dereference(platform_label[index]);
85 }
86 return rt;
87}
88
89bool mpls_output_possible(const struct net_device *dev)
90{
91 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
92}
93EXPORT_SYMBOL_GPL(mpls_output_possible);
94
95static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
96{
97 return (u8 *)nh + rt->rt_via_offset;
98}
99
100static const u8 *mpls_nh_via(const struct mpls_route *rt,
101 const struct mpls_nh *nh)
102{
103 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
104}
105
106static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
107{
108 /* The size of the layer 2.5 labels to be added for this route */
109 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
110}
111
112unsigned int mpls_dev_mtu(const struct net_device *dev)
113{
114 /* The amount of data the layer 2 frame can hold */
115 return dev->mtu;
116}
117EXPORT_SYMBOL_GPL(mpls_dev_mtu);
118
119bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
120{
121 if (skb->len <= mtu)
122 return false;
123
124 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
125 return false;
126
127 return true;
128}
129EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
130
131void mpls_stats_inc_outucastpkts(struct net_device *dev,
132 const struct sk_buff *skb)
133{
134 struct mpls_dev *mdev;
135
136 if (skb->protocol == htons(ETH_P_MPLS_UC)) {
137 mdev = mpls_dev_get(dev);
138 if (mdev)
139 MPLS_INC_STATS_LEN(mdev, skb->len,
140 tx_packets,
141 tx_bytes);
142 } else if (skb->protocol == htons(ETH_P_IP)) {
143 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
144#if IS_ENABLED(CONFIG_IPV6)
145 } else if (skb->protocol == htons(ETH_P_IPV6)) {
146 struct inet6_dev *in6dev = __in6_dev_get(dev);
147
148 if (in6dev)
149 IP6_UPD_PO_STATS(dev_net(dev), in6dev,
150 IPSTATS_MIB_OUT, skb->len);
151#endif
152 }
153}
154EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
155
156static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
157{
158 struct mpls_entry_decoded dec;
159 unsigned int mpls_hdr_len = 0;
160 struct mpls_shim_hdr *hdr;
161 bool eli_seen = false;
162 int label_index;
163 u32 hash = 0;
164
165 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
166 label_index++) {
167 mpls_hdr_len += sizeof(*hdr);
168 if (!pskb_may_pull(skb, mpls_hdr_len))
169 break;
170
171 /* Read and decode the current label */
172 hdr = mpls_hdr(skb) + label_index;
173 dec = mpls_entry_decode(hdr);
174
175 /* RFC6790 - reserved labels MUST NOT be used as keys
176 * for the load-balancing function
177 */
178 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
179 hash = jhash_1word(dec.label, hash);
180
181 /* The entropy label follows the entropy label
182 * indicator, so this means that the entropy
183 * label was just added to the hash - no need to
184 * go any deeper either in the label stack or in the
185 * payload
186 */
187 if (eli_seen)
188 break;
189 } else if (dec.label == MPLS_LABEL_ENTROPY) {
190 eli_seen = true;
191 }
192
193 if (!dec.bos)
194 continue;
195
196 /* found bottom label; does skb have room for a header? */
197 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
198 const struct iphdr *v4hdr;
199
200 v4hdr = (const struct iphdr *)(hdr + 1);
201 if (v4hdr->version == 4) {
202 hash = jhash_3words(ntohl(v4hdr->saddr),
203 ntohl(v4hdr->daddr),
204 v4hdr->protocol, hash);
205 } else if (v4hdr->version == 6 &&
206 pskb_may_pull(skb, mpls_hdr_len +
207 sizeof(struct ipv6hdr))) {
208 const struct ipv6hdr *v6hdr;
209
210 v6hdr = (const struct ipv6hdr *)(hdr + 1);
211 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
212 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
213 hash = jhash_1word(v6hdr->nexthdr, hash);
214 }
215 }
216
217 break;
218 }
219
220 return hash;
221}
222
223static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
224{
225 return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
226}
227
228/* number of alive nexthops (rt->rt_nhn_alive) and the flags for
229 * a next hop (nh->nh_flags) are modified by netdev event handlers.
230 * Since those fields can change at any moment, use READ_ONCE to
231 * access both.
232 */
233static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
234 struct sk_buff *skb)
235{
236 u32 hash = 0;
237 int nh_index = 0;
238 int n = 0;
239 u8 alive;
240
241 /* No need to look further into packet if there's only
242 * one path
243 */
244 if (rt->rt_nhn == 1)
245 return rt->rt_nh;
246
247 alive = READ_ONCE(rt->rt_nhn_alive);
248 if (alive == 0)
249 return NULL;
250
251 hash = mpls_multipath_hash(rt, skb);
252 nh_index = hash % alive;
253 if (alive == rt->rt_nhn)
254 goto out;
255 for_nexthops(rt) {
256 unsigned int nh_flags = READ_ONCE(nh->nh_flags);
257
258 if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
259 continue;
260 if (n == nh_index)
261 return nh;
262 n++;
263 } endfor_nexthops(rt);
264
265out:
266 return mpls_get_nexthop(rt, nh_index);
267}
268
269static bool mpls_egress(struct net *net, struct mpls_route *rt,
270 struct sk_buff *skb, struct mpls_entry_decoded dec)
271{
272 enum mpls_payload_type payload_type;
273 bool success = false;
274
275 /* The IPv4 code below accesses through the IPv4 header
276 * checksum, which is 12 bytes into the packet.
277 * The IPv6 code below accesses through the IPv6 hop limit
278 * which is 8 bytes into the packet.
279 *
280 * For all supported cases there should always be at least 12
281 * bytes of packet data present. The IPv4 header is 20 bytes
282 * without options and the IPv6 header is always 40 bytes
283 * long.
284 */
285 if (!pskb_may_pull(skb, 12))
286 return false;
287
288 payload_type = rt->rt_payload_type;
289 if (payload_type == MPT_UNSPEC)
290 payload_type = ip_hdr(skb)->version;
291
292 switch (payload_type) {
293 case MPT_IPV4: {
294 struct iphdr *hdr4 = ip_hdr(skb);
295 u8 new_ttl;
296 skb->protocol = htons(ETH_P_IP);
297
298 /* If propagating TTL, take the decremented TTL from
299 * the incoming MPLS header, otherwise decrement the
300 * TTL, but only if not 0 to avoid underflow.
301 */
302 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
303 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
304 net->mpls.ip_ttl_propagate))
305 new_ttl = dec.ttl;
306 else
307 new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
308
309 csum_replace2(&hdr4->check,
310 htons(hdr4->ttl << 8),
311 htons(new_ttl << 8));
312 hdr4->ttl = new_ttl;
313 success = true;
314 break;
315 }
316 case MPT_IPV6: {
317 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
318 skb->protocol = htons(ETH_P_IPV6);
319
320 /* If propagating TTL, take the decremented TTL from
321 * the incoming MPLS header, otherwise decrement the
322 * hop limit, but only if not 0 to avoid underflow.
323 */
324 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
325 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
326 net->mpls.ip_ttl_propagate))
327 hdr6->hop_limit = dec.ttl;
328 else if (hdr6->hop_limit)
329 hdr6->hop_limit = hdr6->hop_limit - 1;
330 success = true;
331 break;
332 }
333 case MPT_UNSPEC:
334 /* Should have decided which protocol it is by now */
335 break;
336 }
337
338 return success;
339}
340
341static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
342 struct packet_type *pt, struct net_device *orig_dev)
343{
344 struct net *net = dev_net(dev);
345 struct mpls_shim_hdr *hdr;
346 struct mpls_route *rt;
347 struct mpls_nh *nh;
348 struct mpls_entry_decoded dec;
349 struct net_device *out_dev;
350 struct mpls_dev *out_mdev;
351 struct mpls_dev *mdev;
352 unsigned int hh_len;
353 unsigned int new_header_size;
354 unsigned int mtu;
355 int err;
356
357 /* Careful this entire function runs inside of an rcu critical section */
358
359 mdev = mpls_dev_get(dev);
360 if (!mdev)
361 goto drop;
362
363 MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
364 rx_bytes);
365
366 if (!mdev->input_enabled) {
367 MPLS_INC_STATS(mdev, rx_dropped);
368 goto drop;
369 }
370
371 if (skb->pkt_type != PACKET_HOST)
372 goto err;
373
374 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
375 goto err;
376
377 if (!pskb_may_pull(skb, sizeof(*hdr)))
378 goto err;
379
380 /* Read and decode the label */
381 hdr = mpls_hdr(skb);
382 dec = mpls_entry_decode(hdr);
383
384 rt = mpls_route_input_rcu(net, dec.label);
385 if (!rt) {
386 MPLS_INC_STATS(mdev, rx_noroute);
387 goto drop;
388 }
389
390 nh = mpls_select_multipath(rt, skb);
391 if (!nh)
392 goto err;
393
394 /* Pop the label */
395 skb_pull(skb, sizeof(*hdr));
396 skb_reset_network_header(skb);
397
398 skb_orphan(skb);
399
400 if (skb_warn_if_lro(skb))
401 goto err;
402
403 skb_forward_csum(skb);
404
405 /* Verify ttl is valid */
406 if (dec.ttl <= 1)
407 goto err;
408 dec.ttl -= 1;
409
410 /* Find the output device */
411 out_dev = rcu_dereference(nh->nh_dev);
412 if (!mpls_output_possible(out_dev))
413 goto tx_err;
414
415 /* Verify the destination can hold the packet */
416 new_header_size = mpls_nh_header_size(nh);
417 mtu = mpls_dev_mtu(out_dev);
418 if (mpls_pkt_too_big(skb, mtu - new_header_size))
419 goto tx_err;
420
421 hh_len = LL_RESERVED_SPACE(out_dev);
422 if (!out_dev->header_ops)
423 hh_len = 0;
424
425 /* Ensure there is enough space for the headers in the skb */
426 if (skb_cow(skb, hh_len + new_header_size))
427 goto tx_err;
428
429 skb->dev = out_dev;
430 skb->protocol = htons(ETH_P_MPLS_UC);
431
432 if (unlikely(!new_header_size && dec.bos)) {
433 /* Penultimate hop popping */
434 if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
435 goto err;
436 } else {
437 bool bos;
438 int i;
439 skb_push(skb, new_header_size);
440 skb_reset_network_header(skb);
441 /* Push the new labels */
442 hdr = mpls_hdr(skb);
443 bos = dec.bos;
444 for (i = nh->nh_labels - 1; i >= 0; i--) {
445 hdr[i] = mpls_entry_encode(nh->nh_label[i],
446 dec.ttl, 0, bos);
447 bos = false;
448 }
449 }
450
451 mpls_stats_inc_outucastpkts(out_dev, skb);
452
453 /* If via wasn't specified then send out using device address */
454 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
455 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
456 out_dev->dev_addr, skb);
457 else
458 err = neigh_xmit(nh->nh_via_table, out_dev,
459 mpls_nh_via(rt, nh), skb);
460 if (err)
461 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
462 __func__, err);
463 return 0;
464
465tx_err:
466 out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
467 if (out_mdev)
468 MPLS_INC_STATS(out_mdev, tx_errors);
469 goto drop;
470err:
471 MPLS_INC_STATS(mdev, rx_errors);
472drop:
473 kfree_skb(skb);
474 return NET_RX_DROP;
475}
476
477static struct packet_type mpls_packet_type __read_mostly = {
478 .type = cpu_to_be16(ETH_P_MPLS_UC),
479 .func = mpls_forward,
480};
481
482static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
483 [RTA_DST] = { .type = NLA_U32 },
484 [RTA_OIF] = { .type = NLA_U32 },
485 [RTA_TTL_PROPAGATE] = { .type = NLA_U8 },
486};
487
488struct mpls_route_config {
489 u32 rc_protocol;
490 u32 rc_ifindex;
491 u8 rc_via_table;
492 u8 rc_via_alen;
493 u8 rc_via[MAX_VIA_ALEN];
494 u32 rc_label;
495 u8 rc_ttl_propagate;
496 u8 rc_output_labels;
497 u32 rc_output_label[MAX_NEW_LABELS];
498 u32 rc_nlflags;
499 enum mpls_payload_type rc_payload_type;
500 struct nl_info rc_nlinfo;
501 struct rtnexthop *rc_mp;
502 int rc_mp_len;
503};
504
505/* all nexthops within a route have the same size based on max
506 * number of labels and max via length for a hop
507 */
508static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
509{
510 u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
511 struct mpls_route *rt;
512 size_t size;
513
514 size = sizeof(*rt) + num_nh * nh_size;
515 if (size > MAX_MPLS_ROUTE_MEM)
516 return ERR_PTR(-EINVAL);
517
518 rt = kzalloc(size, GFP_KERNEL);
519 if (!rt)
520 return ERR_PTR(-ENOMEM);
521
522 rt->rt_nhn = num_nh;
523 rt->rt_nhn_alive = num_nh;
524 rt->rt_nh_size = nh_size;
525 rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
526
527 return rt;
528}
529
530static void mpls_rt_free(struct mpls_route *rt)
531{
532 if (rt)
533 kfree_rcu(rt, rt_rcu);
534}
535
536static void mpls_notify_route(struct net *net, unsigned index,
537 struct mpls_route *old, struct mpls_route *new,
538 const struct nl_info *info)
539{
540 struct nlmsghdr *nlh = info ? info->nlh : NULL;
541 unsigned portid = info ? info->portid : 0;
542 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
543 struct mpls_route *rt = new ? new : old;
544 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
545 /* Ignore reserved labels for now */
546 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
547 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
548}
549
550static void mpls_route_update(struct net *net, unsigned index,
551 struct mpls_route *new,
552 const struct nl_info *info)
553{
554 struct mpls_route __rcu **platform_label;
555 struct mpls_route *rt;
556
557 ASSERT_RTNL();
558
559 platform_label = rtnl_dereference(net->mpls.platform_label);
560 rt = rtnl_dereference(platform_label[index]);
561 rcu_assign_pointer(platform_label[index], new);
562
563 mpls_notify_route(net, index, rt, new, info);
564
565 /* If we removed a route free it now */
566 mpls_rt_free(rt);
567}
568
569static unsigned find_free_label(struct net *net)
570{
571 struct mpls_route __rcu **platform_label;
572 size_t platform_labels;
573 unsigned index;
574
575 platform_label = rtnl_dereference(net->mpls.platform_label);
576 platform_labels = net->mpls.platform_labels;
577 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
578 index++) {
579 if (!rtnl_dereference(platform_label[index]))
580 return index;
581 }
582 return LABEL_NOT_SPECIFIED;
583}
584
585#if IS_ENABLED(CONFIG_INET)
586static struct net_device *inet_fib_lookup_dev(struct net *net,
587 const void *addr)
588{
589 struct net_device *dev;
590 struct rtable *rt;
591 struct in_addr daddr;
592
593 memcpy(&daddr, addr, sizeof(struct in_addr));
594 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
595 if (IS_ERR(rt))
596 return ERR_CAST(rt);
597
598 dev = rt->dst.dev;
599 dev_hold(dev);
600
601 ip_rt_put(rt);
602
603 return dev;
604}
605#else
606static struct net_device *inet_fib_lookup_dev(struct net *net,
607 const void *addr)
608{
609 return ERR_PTR(-EAFNOSUPPORT);
610}
611#endif
612
613#if IS_ENABLED(CONFIG_IPV6)
614static struct net_device *inet6_fib_lookup_dev(struct net *net,
615 const void *addr)
616{
617 struct net_device *dev;
618 struct dst_entry *dst;
619 struct flowi6 fl6;
620
621 if (!ipv6_stub)
622 return ERR_PTR(-EAFNOSUPPORT);
623
624 memset(&fl6, 0, sizeof(fl6));
625 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
626 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
627 if (IS_ERR(dst))
628 return ERR_CAST(dst);
629
630 dev = dst->dev;
631 dev_hold(dev);
632 dst_release(dst);
633
634 return dev;
635}
636#else
637static struct net_device *inet6_fib_lookup_dev(struct net *net,
638 const void *addr)
639{
640 return ERR_PTR(-EAFNOSUPPORT);
641}
642#endif
643
644static struct net_device *find_outdev(struct net *net,
645 struct mpls_route *rt,
646 struct mpls_nh *nh, int oif)
647{
648 struct net_device *dev = NULL;
649
650 if (!oif) {
651 switch (nh->nh_via_table) {
652 case NEIGH_ARP_TABLE:
653 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
654 break;
655 case NEIGH_ND_TABLE:
656 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
657 break;
658 case NEIGH_LINK_TABLE:
659 break;
660 }
661 } else {
662 dev = dev_get_by_index(net, oif);
663 }
664
665 if (!dev)
666 return ERR_PTR(-ENODEV);
667
668 if (IS_ERR(dev))
669 return dev;
670
671 /* The caller is holding rtnl anyways, so release the dev reference */
672 dev_put(dev);
673
674 return dev;
675}
676
677static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
678 struct mpls_nh *nh, int oif)
679{
680 struct net_device *dev = NULL;
681 int err = -ENODEV;
682
683 dev = find_outdev(net, rt, nh, oif);
684 if (IS_ERR(dev)) {
685 err = PTR_ERR(dev);
686 dev = NULL;
687 goto errout;
688 }
689
690 /* Ensure this is a supported device */
691 err = -EINVAL;
692 if (!mpls_dev_get(dev))
693 goto errout;
694
695 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
696 (dev->addr_len != nh->nh_via_alen))
697 goto errout;
698
699 RCU_INIT_POINTER(nh->nh_dev, dev);
700
701 if (!(dev->flags & IFF_UP)) {
702 nh->nh_flags |= RTNH_F_DEAD;
703 } else {
704 unsigned int flags;
705
706 flags = dev_get_flags(dev);
707 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
708 nh->nh_flags |= RTNH_F_LINKDOWN;
709 }
710
711 return 0;
712
713errout:
714 return err;
715}
716
717static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
718 u8 via_addr[], struct netlink_ext_ack *extack)
719{
720 struct rtvia *via = nla_data(nla);
721 int err = -EINVAL;
722 int alen;
723
724 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
725 NL_SET_ERR_MSG_ATTR(extack, nla,
726 "Invalid attribute length for RTA_VIA");
727 goto errout;
728 }
729 alen = nla_len(nla) -
730 offsetof(struct rtvia, rtvia_addr);
731 if (alen > MAX_VIA_ALEN) {
732 NL_SET_ERR_MSG_ATTR(extack, nla,
733 "Invalid address length for RTA_VIA");
734 goto errout;
735 }
736
737 /* Validate the address family */
738 switch (via->rtvia_family) {
739 case AF_PACKET:
740 *via_table = NEIGH_LINK_TABLE;
741 break;
742 case AF_INET:
743 *via_table = NEIGH_ARP_TABLE;
744 if (alen != 4)
745 goto errout;
746 break;
747 case AF_INET6:
748 *via_table = NEIGH_ND_TABLE;
749 if (alen != 16)
750 goto errout;
751 break;
752 default:
753 /* Unsupported address family */
754 goto errout;
755 }
756
757 memcpy(via_addr, via->rtvia_addr, alen);
758 *via_alen = alen;
759 err = 0;
760
761errout:
762 return err;
763}
764
765static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
766 struct mpls_route *rt)
767{
768 struct net *net = cfg->rc_nlinfo.nl_net;
769 struct mpls_nh *nh = rt->rt_nh;
770 int err;
771 int i;
772
773 if (!nh)
774 return -ENOMEM;
775
776 nh->nh_labels = cfg->rc_output_labels;
777 for (i = 0; i < nh->nh_labels; i++)
778 nh->nh_label[i] = cfg->rc_output_label[i];
779
780 nh->nh_via_table = cfg->rc_via_table;
781 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
782 nh->nh_via_alen = cfg->rc_via_alen;
783
784 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
785 if (err)
786 goto errout;
787
788 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
789 rt->rt_nhn_alive--;
790
791 return 0;
792
793errout:
794 return err;
795}
796
797static int mpls_nh_build(struct net *net, struct mpls_route *rt,
798 struct mpls_nh *nh, int oif, struct nlattr *via,
799 struct nlattr *newdst, u8 max_labels,
800 struct netlink_ext_ack *extack)
801{
802 int err = -ENOMEM;
803
804 if (!nh)
805 goto errout;
806
807 if (newdst) {
808 err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
809 nh->nh_label, extack);
810 if (err)
811 goto errout;
812 }
813
814 if (via) {
815 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
816 __mpls_nh_via(rt, nh), extack);
817 if (err)
818 goto errout;
819 } else {
820 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
821 }
822
823 err = mpls_nh_assign_dev(net, rt, nh, oif);
824 if (err)
825 goto errout;
826
827 return 0;
828
829errout:
830 return err;
831}
832
833static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
834 u8 cfg_via_alen, u8 *max_via_alen,
835 u8 *max_labels)
836{
837 int remaining = len;
838 u8 nhs = 0;
839
840 *max_via_alen = 0;
841 *max_labels = 0;
842
843 while (rtnh_ok(rtnh, remaining)) {
844 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
845 int attrlen;
846 u8 n_labels = 0;
847
848 attrlen = rtnh_attrlen(rtnh);
849 nla = nla_find(attrs, attrlen, RTA_VIA);
850 if (nla && nla_len(nla) >=
851 offsetof(struct rtvia, rtvia_addr)) {
852 int via_alen = nla_len(nla) -
853 offsetof(struct rtvia, rtvia_addr);
854
855 if (via_alen <= MAX_VIA_ALEN)
856 *max_via_alen = max_t(u16, *max_via_alen,
857 via_alen);
858 }
859
860 nla = nla_find(attrs, attrlen, RTA_NEWDST);
861 if (nla &&
862 nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
863 NULL, NULL) != 0)
864 return 0;
865
866 *max_labels = max_t(u8, *max_labels, n_labels);
867
868 /* number of nexthops is tracked by a u8.
869 * Check for overflow.
870 */
871 if (nhs == 255)
872 return 0;
873 nhs++;
874
875 rtnh = rtnh_next(rtnh, &remaining);
876 }
877
878 /* leftover implies invalid nexthop configuration, discard it */
879 return remaining > 0 ? 0 : nhs;
880}
881
882static int mpls_nh_build_multi(struct mpls_route_config *cfg,
883 struct mpls_route *rt, u8 max_labels,
884 struct netlink_ext_ack *extack)
885{
886 struct rtnexthop *rtnh = cfg->rc_mp;
887 struct nlattr *nla_via, *nla_newdst;
888 int remaining = cfg->rc_mp_len;
889 int err = 0;
890 u8 nhs = 0;
891
892 change_nexthops(rt) {
893 int attrlen;
894
895 nla_via = NULL;
896 nla_newdst = NULL;
897
898 err = -EINVAL;
899 if (!rtnh_ok(rtnh, remaining))
900 goto errout;
901
902 /* neither weighted multipath nor any flags
903 * are supported
904 */
905 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
906 goto errout;
907
908 attrlen = rtnh_attrlen(rtnh);
909 if (attrlen > 0) {
910 struct nlattr *attrs = rtnh_attrs(rtnh);
911
912 nla_via = nla_find(attrs, attrlen, RTA_VIA);
913 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
914 }
915
916 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
917 rtnh->rtnh_ifindex, nla_via, nla_newdst,
918 max_labels, extack);
919 if (err)
920 goto errout;
921
922 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
923 rt->rt_nhn_alive--;
924
925 rtnh = rtnh_next(rtnh, &remaining);
926 nhs++;
927 } endfor_nexthops(rt);
928
929 rt->rt_nhn = nhs;
930
931 return 0;
932
933errout:
934 return err;
935}
936
937static bool mpls_label_ok(struct net *net, unsigned int *index,
938 struct netlink_ext_ack *extack)
939{
940 bool is_ok = true;
941
942 /* Reserved labels may not be set */
943 if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
944 NL_SET_ERR_MSG(extack,
945 "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
946 is_ok = false;
947 }
948
949 /* The full 20 bit range may not be supported. */
950 if (is_ok && *index >= net->mpls.platform_labels) {
951 NL_SET_ERR_MSG(extack,
952 "Label >= configured maximum in platform_labels");
953 is_ok = false;
954 }
955
956 *index = array_index_nospec(*index, net->mpls.platform_labels);
957 return is_ok;
958}
959
960static int mpls_route_add(struct mpls_route_config *cfg,
961 struct netlink_ext_ack *extack)
962{
963 struct mpls_route __rcu **platform_label;
964 struct net *net = cfg->rc_nlinfo.nl_net;
965 struct mpls_route *rt, *old;
966 int err = -EINVAL;
967 u8 max_via_alen;
968 unsigned index;
969 u8 max_labels;
970 u8 nhs;
971
972 index = cfg->rc_label;
973
974 /* If a label was not specified during insert pick one */
975 if ((index == LABEL_NOT_SPECIFIED) &&
976 (cfg->rc_nlflags & NLM_F_CREATE)) {
977 index = find_free_label(net);
978 }
979
980 if (!mpls_label_ok(net, &index, extack))
981 goto errout;
982
983 /* Append makes no sense with mpls */
984 err = -EOPNOTSUPP;
985 if (cfg->rc_nlflags & NLM_F_APPEND) {
986 NL_SET_ERR_MSG(extack, "MPLS does not support route append");
987 goto errout;
988 }
989
990 err = -EEXIST;
991 platform_label = rtnl_dereference(net->mpls.platform_label);
992 old = rtnl_dereference(platform_label[index]);
993 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
994 goto errout;
995
996 err = -EEXIST;
997 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
998 goto errout;
999
1000 err = -ENOENT;
1001 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1002 goto errout;
1003
1004 err = -EINVAL;
1005 if (cfg->rc_mp) {
1006 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1007 cfg->rc_via_alen, &max_via_alen,
1008 &max_labels);
1009 } else {
1010 max_via_alen = cfg->rc_via_alen;
1011 max_labels = cfg->rc_output_labels;
1012 nhs = 1;
1013 }
1014
1015 if (nhs == 0) {
1016 NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1017 goto errout;
1018 }
1019
1020 err = -ENOMEM;
1021 rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1022 if (IS_ERR(rt)) {
1023 err = PTR_ERR(rt);
1024 goto errout;
1025 }
1026
1027 rt->rt_protocol = cfg->rc_protocol;
1028 rt->rt_payload_type = cfg->rc_payload_type;
1029 rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1030
1031 if (cfg->rc_mp)
1032 err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1033 else
1034 err = mpls_nh_build_from_cfg(cfg, rt);
1035 if (err)
1036 goto freert;
1037
1038 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1039
1040 return 0;
1041
1042freert:
1043 mpls_rt_free(rt);
1044errout:
1045 return err;
1046}
1047
1048static int mpls_route_del(struct mpls_route_config *cfg,
1049 struct netlink_ext_ack *extack)
1050{
1051 struct net *net = cfg->rc_nlinfo.nl_net;
1052 unsigned index;
1053 int err = -EINVAL;
1054
1055 index = cfg->rc_label;
1056
1057 if (!mpls_label_ok(net, &index, extack))
1058 goto errout;
1059
1060 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1061
1062 err = 0;
1063errout:
1064 return err;
1065}
1066
1067static void mpls_get_stats(struct mpls_dev *mdev,
1068 struct mpls_link_stats *stats)
1069{
1070 struct mpls_pcpu_stats *p;
1071 int i;
1072
1073 memset(stats, 0, sizeof(*stats));
1074
1075 for_each_possible_cpu(i) {
1076 struct mpls_link_stats local;
1077 unsigned int start;
1078
1079 p = per_cpu_ptr(mdev->stats, i);
1080 do {
1081 start = u64_stats_fetch_begin(&p->syncp);
1082 local = p->stats;
1083 } while (u64_stats_fetch_retry(&p->syncp, start));
1084
1085 stats->rx_packets += local.rx_packets;
1086 stats->rx_bytes += local.rx_bytes;
1087 stats->tx_packets += local.tx_packets;
1088 stats->tx_bytes += local.tx_bytes;
1089 stats->rx_errors += local.rx_errors;
1090 stats->tx_errors += local.tx_errors;
1091 stats->rx_dropped += local.rx_dropped;
1092 stats->tx_dropped += local.tx_dropped;
1093 stats->rx_noroute += local.rx_noroute;
1094 }
1095}
1096
1097static int mpls_fill_stats_af(struct sk_buff *skb,
1098 const struct net_device *dev)
1099{
1100 struct mpls_link_stats *stats;
1101 struct mpls_dev *mdev;
1102 struct nlattr *nla;
1103
1104 mdev = mpls_dev_get(dev);
1105 if (!mdev)
1106 return -ENODATA;
1107
1108 nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1109 sizeof(struct mpls_link_stats),
1110 MPLS_STATS_UNSPEC);
1111 if (!nla)
1112 return -EMSGSIZE;
1113
1114 stats = nla_data(nla);
1115 mpls_get_stats(mdev, stats);
1116
1117 return 0;
1118}
1119
1120static size_t mpls_get_stats_af_size(const struct net_device *dev)
1121{
1122 struct mpls_dev *mdev;
1123
1124 mdev = mpls_dev_get(dev);
1125 if (!mdev)
1126 return 0;
1127
1128 return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1129}
1130
1131static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1132 u32 portid, u32 seq, int event,
1133 unsigned int flags, int type)
1134{
1135 struct nlmsghdr *nlh;
1136 struct netconfmsg *ncm;
1137 bool all = false;
1138
1139 nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1140 flags);
1141 if (!nlh)
1142 return -EMSGSIZE;
1143
1144 if (type == NETCONFA_ALL)
1145 all = true;
1146
1147 ncm = nlmsg_data(nlh);
1148 ncm->ncm_family = AF_MPLS;
1149
1150 if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1151 goto nla_put_failure;
1152
1153 if ((all || type == NETCONFA_INPUT) &&
1154 nla_put_s32(skb, NETCONFA_INPUT,
1155 mdev->input_enabled) < 0)
1156 goto nla_put_failure;
1157
1158 nlmsg_end(skb, nlh);
1159 return 0;
1160
1161nla_put_failure:
1162 nlmsg_cancel(skb, nlh);
1163 return -EMSGSIZE;
1164}
1165
1166static int mpls_netconf_msgsize_devconf(int type)
1167{
1168 int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1169 + nla_total_size(4); /* NETCONFA_IFINDEX */
1170 bool all = false;
1171
1172 if (type == NETCONFA_ALL)
1173 all = true;
1174
1175 if (all || type == NETCONFA_INPUT)
1176 size += nla_total_size(4);
1177
1178 return size;
1179}
1180
1181static void mpls_netconf_notify_devconf(struct net *net, int event,
1182 int type, struct mpls_dev *mdev)
1183{
1184 struct sk_buff *skb;
1185 int err = -ENOBUFS;
1186
1187 skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1188 if (!skb)
1189 goto errout;
1190
1191 err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1192 if (err < 0) {
1193 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1194 WARN_ON(err == -EMSGSIZE);
1195 kfree_skb(skb);
1196 goto errout;
1197 }
1198
1199 rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1200 return;
1201errout:
1202 if (err < 0)
1203 rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1204}
1205
1206static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1207 [NETCONFA_IFINDEX] = { .len = sizeof(int) },
1208};
1209
1210static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1211 const struct nlmsghdr *nlh,
1212 struct nlattr **tb,
1213 struct netlink_ext_ack *extack)
1214{
1215 int i, err;
1216
1217 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1218 NL_SET_ERR_MSG_MOD(extack,
1219 "Invalid header for netconf get request");
1220 return -EINVAL;
1221 }
1222
1223 if (!netlink_strict_get_check(skb))
1224 return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1225 tb, NETCONFA_MAX,
1226 devconf_mpls_policy, extack);
1227
1228 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1229 tb, NETCONFA_MAX,
1230 devconf_mpls_policy, extack);
1231 if (err)
1232 return err;
1233
1234 for (i = 0; i <= NETCONFA_MAX; i++) {
1235 if (!tb[i])
1236 continue;
1237
1238 switch (i) {
1239 case NETCONFA_IFINDEX:
1240 break;
1241 default:
1242 NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1243 return -EINVAL;
1244 }
1245 }
1246
1247 return 0;
1248}
1249
1250static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1251 struct nlmsghdr *nlh,
1252 struct netlink_ext_ack *extack)
1253{
1254 struct net *net = sock_net(in_skb->sk);
1255 struct nlattr *tb[NETCONFA_MAX + 1];
1256 struct net_device *dev;
1257 struct mpls_dev *mdev;
1258 struct sk_buff *skb;
1259 int ifindex;
1260 int err;
1261
1262 err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1263 if (err < 0)
1264 goto errout;
1265
1266 err = -EINVAL;
1267 if (!tb[NETCONFA_IFINDEX])
1268 goto errout;
1269
1270 ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1271 dev = __dev_get_by_index(net, ifindex);
1272 if (!dev)
1273 goto errout;
1274
1275 mdev = mpls_dev_get(dev);
1276 if (!mdev)
1277 goto errout;
1278
1279 err = -ENOBUFS;
1280 skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1281 if (!skb)
1282 goto errout;
1283
1284 err = mpls_netconf_fill_devconf(skb, mdev,
1285 NETLINK_CB(in_skb).portid,
1286 nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1287 NETCONFA_ALL);
1288 if (err < 0) {
1289 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1290 WARN_ON(err == -EMSGSIZE);
1291 kfree_skb(skb);
1292 goto errout;
1293 }
1294 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1295errout:
1296 return err;
1297}
1298
1299static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1300 struct netlink_callback *cb)
1301{
1302 const struct nlmsghdr *nlh = cb->nlh;
1303 struct net *net = sock_net(skb->sk);
1304 struct hlist_head *head;
1305 struct net_device *dev;
1306 struct mpls_dev *mdev;
1307 int idx, s_idx;
1308 int h, s_h;
1309
1310 if (cb->strict_check) {
1311 struct netlink_ext_ack *extack = cb->extack;
1312 struct netconfmsg *ncm;
1313
1314 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1315 NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1316 return -EINVAL;
1317 }
1318
1319 if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1320 NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1321 return -EINVAL;
1322 }
1323 }
1324
1325 s_h = cb->args[0];
1326 s_idx = idx = cb->args[1];
1327
1328 for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1329 idx = 0;
1330 head = &net->dev_index_head[h];
1331 rcu_read_lock();
1332 cb->seq = net->dev_base_seq;
1333 hlist_for_each_entry_rcu(dev, head, index_hlist) {
1334 if (idx < s_idx)
1335 goto cont;
1336 mdev = mpls_dev_get(dev);
1337 if (!mdev)
1338 goto cont;
1339 if (mpls_netconf_fill_devconf(skb, mdev,
1340 NETLINK_CB(cb->skb).portid,
1341 nlh->nlmsg_seq,
1342 RTM_NEWNETCONF,
1343 NLM_F_MULTI,
1344 NETCONFA_ALL) < 0) {
1345 rcu_read_unlock();
1346 goto done;
1347 }
1348 nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1349cont:
1350 idx++;
1351 }
1352 rcu_read_unlock();
1353 }
1354done:
1355 cb->args[0] = h;
1356 cb->args[1] = idx;
1357
1358 return skb->len;
1359}
1360
1361#define MPLS_PERDEV_SYSCTL_OFFSET(field) \
1362 (&((struct mpls_dev *)0)->field)
1363
1364static int mpls_conf_proc(struct ctl_table *ctl, int write,
1365 void *buffer, size_t *lenp, loff_t *ppos)
1366{
1367 int oval = *(int *)ctl->data;
1368 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1369
1370 if (write) {
1371 struct mpls_dev *mdev = ctl->extra1;
1372 int i = (int *)ctl->data - (int *)mdev;
1373 struct net *net = ctl->extra2;
1374 int val = *(int *)ctl->data;
1375
1376 if (i == offsetof(struct mpls_dev, input_enabled) &&
1377 val != oval) {
1378 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1379 NETCONFA_INPUT, mdev);
1380 }
1381 }
1382
1383 return ret;
1384}
1385
1386static const struct ctl_table mpls_dev_table[] = {
1387 {
1388 .procname = "input",
1389 .maxlen = sizeof(int),
1390 .mode = 0644,
1391 .proc_handler = mpls_conf_proc,
1392 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1393 },
1394 { }
1395};
1396
1397static int mpls_dev_sysctl_register(struct net_device *dev,
1398 struct mpls_dev *mdev)
1399{
1400 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1401 struct net *net = dev_net(dev);
1402 struct ctl_table *table;
1403 int i;
1404
1405 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1406 if (!table)
1407 goto out;
1408
1409 /* Table data contains only offsets relative to the base of
1410 * the mdev at this point, so make them absolute.
1411 */
1412 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1413 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1414 table[i].extra1 = mdev;
1415 table[i].extra2 = net;
1416 }
1417
1418 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1419
1420 mdev->sysctl = register_net_sysctl(net, path, table);
1421 if (!mdev->sysctl)
1422 goto free;
1423
1424 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1425 return 0;
1426
1427free:
1428 kfree(table);
1429out:
1430 return -ENOBUFS;
1431}
1432
1433static void mpls_dev_sysctl_unregister(struct net_device *dev,
1434 struct mpls_dev *mdev)
1435{
1436 struct net *net = dev_net(dev);
1437 struct ctl_table *table;
1438
1439 table = mdev->sysctl->ctl_table_arg;
1440 unregister_net_sysctl_table(mdev->sysctl);
1441 kfree(table);
1442
1443 mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1444}
1445
1446static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1447{
1448 struct mpls_dev *mdev;
1449 int err = -ENOMEM;
1450 int i;
1451
1452 ASSERT_RTNL();
1453
1454 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1455 if (!mdev)
1456 return ERR_PTR(err);
1457
1458 mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1459 if (!mdev->stats)
1460 goto free;
1461
1462 for_each_possible_cpu(i) {
1463 struct mpls_pcpu_stats *mpls_stats;
1464
1465 mpls_stats = per_cpu_ptr(mdev->stats, i);
1466 u64_stats_init(&mpls_stats->syncp);
1467 }
1468
1469 mdev->dev = dev;
1470
1471 err = mpls_dev_sysctl_register(dev, mdev);
1472 if (err)
1473 goto free;
1474
1475 rcu_assign_pointer(dev->mpls_ptr, mdev);
1476
1477 return mdev;
1478
1479free:
1480 free_percpu(mdev->stats);
1481 kfree(mdev);
1482 return ERR_PTR(err);
1483}
1484
1485static void mpls_dev_destroy_rcu(struct rcu_head *head)
1486{
1487 struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1488
1489 free_percpu(mdev->stats);
1490 kfree(mdev);
1491}
1492
1493static void mpls_ifdown(struct net_device *dev, int event)
1494{
1495 struct mpls_route __rcu **platform_label;
1496 struct net *net = dev_net(dev);
1497 u8 alive, deleted;
1498 unsigned index;
1499
1500 platform_label = rtnl_dereference(net->mpls.platform_label);
1501 for (index = 0; index < net->mpls.platform_labels; index++) {
1502 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1503
1504 if (!rt)
1505 continue;
1506
1507 alive = 0;
1508 deleted = 0;
1509 change_nexthops(rt) {
1510 unsigned int nh_flags = nh->nh_flags;
1511
1512 if (rtnl_dereference(nh->nh_dev) != dev)
1513 goto next;
1514
1515 switch (event) {
1516 case NETDEV_DOWN:
1517 case NETDEV_UNREGISTER:
1518 nh_flags |= RTNH_F_DEAD;
1519 fallthrough;
1520 case NETDEV_CHANGE:
1521 nh_flags |= RTNH_F_LINKDOWN;
1522 break;
1523 }
1524 if (event == NETDEV_UNREGISTER)
1525 RCU_INIT_POINTER(nh->nh_dev, NULL);
1526
1527 if (nh->nh_flags != nh_flags)
1528 WRITE_ONCE(nh->nh_flags, nh_flags);
1529next:
1530 if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1531 alive++;
1532 if (!rtnl_dereference(nh->nh_dev))
1533 deleted++;
1534 } endfor_nexthops(rt);
1535
1536 WRITE_ONCE(rt->rt_nhn_alive, alive);
1537
1538 /* if there are no more nexthops, delete the route */
1539 if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1540 mpls_route_update(net, index, NULL, NULL);
1541 }
1542}
1543
1544static void mpls_ifup(struct net_device *dev, unsigned int flags)
1545{
1546 struct mpls_route __rcu **platform_label;
1547 struct net *net = dev_net(dev);
1548 unsigned index;
1549 u8 alive;
1550
1551 platform_label = rtnl_dereference(net->mpls.platform_label);
1552 for (index = 0; index < net->mpls.platform_labels; index++) {
1553 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1554
1555 if (!rt)
1556 continue;
1557
1558 alive = 0;
1559 change_nexthops(rt) {
1560 unsigned int nh_flags = nh->nh_flags;
1561 struct net_device *nh_dev =
1562 rtnl_dereference(nh->nh_dev);
1563
1564 if (!(nh_flags & flags)) {
1565 alive++;
1566 continue;
1567 }
1568 if (nh_dev != dev)
1569 continue;
1570 alive++;
1571 nh_flags &= ~flags;
1572 WRITE_ONCE(nh->nh_flags, nh_flags);
1573 } endfor_nexthops(rt);
1574
1575 WRITE_ONCE(rt->rt_nhn_alive, alive);
1576 }
1577}
1578
1579static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1580 void *ptr)
1581{
1582 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1583 struct mpls_dev *mdev;
1584 unsigned int flags;
1585
1586 if (event == NETDEV_REGISTER) {
1587 mdev = mpls_add_dev(dev);
1588 if (IS_ERR(mdev))
1589 return notifier_from_errno(PTR_ERR(mdev));
1590
1591 return NOTIFY_OK;
1592 }
1593
1594 mdev = mpls_dev_get(dev);
1595 if (!mdev)
1596 return NOTIFY_OK;
1597
1598 switch (event) {
1599 case NETDEV_DOWN:
1600 mpls_ifdown(dev, event);
1601 break;
1602 case NETDEV_UP:
1603 flags = dev_get_flags(dev);
1604 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1605 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1606 else
1607 mpls_ifup(dev, RTNH_F_DEAD);
1608 break;
1609 case NETDEV_CHANGE:
1610 flags = dev_get_flags(dev);
1611 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1612 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1613 else
1614 mpls_ifdown(dev, event);
1615 break;
1616 case NETDEV_UNREGISTER:
1617 mpls_ifdown(dev, event);
1618 mdev = mpls_dev_get(dev);
1619 if (mdev) {
1620 mpls_dev_sysctl_unregister(dev, mdev);
1621 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1622 call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1623 }
1624 break;
1625 case NETDEV_CHANGENAME:
1626 mdev = mpls_dev_get(dev);
1627 if (mdev) {
1628 int err;
1629
1630 mpls_dev_sysctl_unregister(dev, mdev);
1631 err = mpls_dev_sysctl_register(dev, mdev);
1632 if (err)
1633 return notifier_from_errno(err);
1634 }
1635 break;
1636 }
1637 return NOTIFY_OK;
1638}
1639
1640static struct notifier_block mpls_dev_notifier = {
1641 .notifier_call = mpls_dev_notify,
1642};
1643
1644static int nla_put_via(struct sk_buff *skb,
1645 u8 table, const void *addr, int alen)
1646{
1647 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1648 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1649 };
1650 struct nlattr *nla;
1651 struct rtvia *via;
1652 int family = AF_UNSPEC;
1653
1654 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1655 if (!nla)
1656 return -EMSGSIZE;
1657
1658 if (table <= NEIGH_NR_TABLES)
1659 family = table_to_family[table];
1660
1661 via = nla_data(nla);
1662 via->rtvia_family = family;
1663 memcpy(via->rtvia_addr, addr, alen);
1664 return 0;
1665}
1666
1667int nla_put_labels(struct sk_buff *skb, int attrtype,
1668 u8 labels, const u32 label[])
1669{
1670 struct nlattr *nla;
1671 struct mpls_shim_hdr *nla_label;
1672 bool bos;
1673 int i;
1674 nla = nla_reserve(skb, attrtype, labels*4);
1675 if (!nla)
1676 return -EMSGSIZE;
1677
1678 nla_label = nla_data(nla);
1679 bos = true;
1680 for (i = labels - 1; i >= 0; i--) {
1681 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1682 bos = false;
1683 }
1684
1685 return 0;
1686}
1687EXPORT_SYMBOL_GPL(nla_put_labels);
1688
1689int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1690 u32 label[], struct netlink_ext_ack *extack)
1691{
1692 unsigned len = nla_len(nla);
1693 struct mpls_shim_hdr *nla_label;
1694 u8 nla_labels;
1695 bool bos;
1696 int i;
1697
1698 /* len needs to be an even multiple of 4 (the label size). Number
1699 * of labels is a u8 so check for overflow.
1700 */
1701 if (len & 3 || len / 4 > 255) {
1702 NL_SET_ERR_MSG_ATTR(extack, nla,
1703 "Invalid length for labels attribute");
1704 return -EINVAL;
1705 }
1706
1707 /* Limit the number of new labels allowed */
1708 nla_labels = len/4;
1709 if (nla_labels > max_labels) {
1710 NL_SET_ERR_MSG(extack, "Too many labels");
1711 return -EINVAL;
1712 }
1713
1714 /* when label == NULL, caller wants number of labels */
1715 if (!label)
1716 goto out;
1717
1718 nla_label = nla_data(nla);
1719 bos = true;
1720 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1721 struct mpls_entry_decoded dec;
1722 dec = mpls_entry_decode(nla_label + i);
1723
1724 /* Ensure the bottom of stack flag is properly set
1725 * and ttl and tc are both clear.
1726 */
1727 if (dec.ttl) {
1728 NL_SET_ERR_MSG_ATTR(extack, nla,
1729 "TTL in label must be 0");
1730 return -EINVAL;
1731 }
1732
1733 if (dec.tc) {
1734 NL_SET_ERR_MSG_ATTR(extack, nla,
1735 "Traffic class in label must be 0");
1736 return -EINVAL;
1737 }
1738
1739 if (dec.bos != bos) {
1740 NL_SET_BAD_ATTR(extack, nla);
1741 if (bos) {
1742 NL_SET_ERR_MSG(extack,
1743 "BOS bit must be set in first label");
1744 } else {
1745 NL_SET_ERR_MSG(extack,
1746 "BOS bit can only be set in first label");
1747 }
1748 return -EINVAL;
1749 }
1750
1751 switch (dec.label) {
1752 case MPLS_LABEL_IMPLNULL:
1753 /* RFC3032: This is a label that an LSR may
1754 * assign and distribute, but which never
1755 * actually appears in the encapsulation.
1756 */
1757 NL_SET_ERR_MSG_ATTR(extack, nla,
1758 "Implicit NULL Label (3) can not be used in encapsulation");
1759 return -EINVAL;
1760 }
1761
1762 label[i] = dec.label;
1763 }
1764out:
1765 *labels = nla_labels;
1766 return 0;
1767}
1768EXPORT_SYMBOL_GPL(nla_get_labels);
1769
1770static int rtm_to_route_config(struct sk_buff *skb,
1771 struct nlmsghdr *nlh,
1772 struct mpls_route_config *cfg,
1773 struct netlink_ext_ack *extack)
1774{
1775 struct rtmsg *rtm;
1776 struct nlattr *tb[RTA_MAX+1];
1777 int index;
1778 int err;
1779
1780 err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1781 rtm_mpls_policy, extack);
1782 if (err < 0)
1783 goto errout;
1784
1785 err = -EINVAL;
1786 rtm = nlmsg_data(nlh);
1787
1788 if (rtm->rtm_family != AF_MPLS) {
1789 NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1790 goto errout;
1791 }
1792 if (rtm->rtm_dst_len != 20) {
1793 NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1794 goto errout;
1795 }
1796 if (rtm->rtm_src_len != 0) {
1797 NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1798 goto errout;
1799 }
1800 if (rtm->rtm_tos != 0) {
1801 NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1802 goto errout;
1803 }
1804 if (rtm->rtm_table != RT_TABLE_MAIN) {
1805 NL_SET_ERR_MSG(extack,
1806 "MPLS only supports the main route table");
1807 goto errout;
1808 }
1809 /* Any value is acceptable for rtm_protocol */
1810
1811 /* As mpls uses destination specific addresses
1812 * (or source specific address in the case of multicast)
1813 * all addresses have universal scope.
1814 */
1815 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1816 NL_SET_ERR_MSG(extack,
1817 "Invalid route scope - MPLS only supports UNIVERSE");
1818 goto errout;
1819 }
1820 if (rtm->rtm_type != RTN_UNICAST) {
1821 NL_SET_ERR_MSG(extack,
1822 "Invalid route type - MPLS only supports UNICAST");
1823 goto errout;
1824 }
1825 if (rtm->rtm_flags != 0) {
1826 NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1827 goto errout;
1828 }
1829
1830 cfg->rc_label = LABEL_NOT_SPECIFIED;
1831 cfg->rc_protocol = rtm->rtm_protocol;
1832 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1833 cfg->rc_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
1834 cfg->rc_nlflags = nlh->nlmsg_flags;
1835 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1836 cfg->rc_nlinfo.nlh = nlh;
1837 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1838
1839 for (index = 0; index <= RTA_MAX; index++) {
1840 struct nlattr *nla = tb[index];
1841 if (!nla)
1842 continue;
1843
1844 switch (index) {
1845 case RTA_OIF:
1846 cfg->rc_ifindex = nla_get_u32(nla);
1847 break;
1848 case RTA_NEWDST:
1849 if (nla_get_labels(nla, MAX_NEW_LABELS,
1850 &cfg->rc_output_labels,
1851 cfg->rc_output_label, extack))
1852 goto errout;
1853 break;
1854 case RTA_DST:
1855 {
1856 u8 label_count;
1857 if (nla_get_labels(nla, 1, &label_count,
1858 &cfg->rc_label, extack))
1859 goto errout;
1860
1861 if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1862 &cfg->rc_label, extack))
1863 goto errout;
1864 break;
1865 }
1866 case RTA_GATEWAY:
1867 NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1868 goto errout;
1869 case RTA_VIA:
1870 {
1871 if (nla_get_via(nla, &cfg->rc_via_alen,
1872 &cfg->rc_via_table, cfg->rc_via,
1873 extack))
1874 goto errout;
1875 break;
1876 }
1877 case RTA_MULTIPATH:
1878 {
1879 cfg->rc_mp = nla_data(nla);
1880 cfg->rc_mp_len = nla_len(nla);
1881 break;
1882 }
1883 case RTA_TTL_PROPAGATE:
1884 {
1885 u8 ttl_propagate = nla_get_u8(nla);
1886
1887 if (ttl_propagate > 1) {
1888 NL_SET_ERR_MSG_ATTR(extack, nla,
1889 "RTA_TTL_PROPAGATE can only be 0 or 1");
1890 goto errout;
1891 }
1892 cfg->rc_ttl_propagate = ttl_propagate ?
1893 MPLS_TTL_PROP_ENABLED :
1894 MPLS_TTL_PROP_DISABLED;
1895 break;
1896 }
1897 default:
1898 NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1899 /* Unsupported attribute */
1900 goto errout;
1901 }
1902 }
1903
1904 err = 0;
1905errout:
1906 return err;
1907}
1908
1909static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1910 struct netlink_ext_ack *extack)
1911{
1912 struct mpls_route_config *cfg;
1913 int err;
1914
1915 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1916 if (!cfg)
1917 return -ENOMEM;
1918
1919 err = rtm_to_route_config(skb, nlh, cfg, extack);
1920 if (err < 0)
1921 goto out;
1922
1923 err = mpls_route_del(cfg, extack);
1924out:
1925 kfree(cfg);
1926
1927 return err;
1928}
1929
1930
1931static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1932 struct netlink_ext_ack *extack)
1933{
1934 struct mpls_route_config *cfg;
1935 int err;
1936
1937 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1938 if (!cfg)
1939 return -ENOMEM;
1940
1941 err = rtm_to_route_config(skb, nlh, cfg, extack);
1942 if (err < 0)
1943 goto out;
1944
1945 err = mpls_route_add(cfg, extack);
1946out:
1947 kfree(cfg);
1948
1949 return err;
1950}
1951
1952static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1953 u32 label, struct mpls_route *rt, int flags)
1954{
1955 struct net_device *dev;
1956 struct nlmsghdr *nlh;
1957 struct rtmsg *rtm;
1958
1959 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1960 if (nlh == NULL)
1961 return -EMSGSIZE;
1962
1963 rtm = nlmsg_data(nlh);
1964 rtm->rtm_family = AF_MPLS;
1965 rtm->rtm_dst_len = 20;
1966 rtm->rtm_src_len = 0;
1967 rtm->rtm_tos = 0;
1968 rtm->rtm_table = RT_TABLE_MAIN;
1969 rtm->rtm_protocol = rt->rt_protocol;
1970 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1971 rtm->rtm_type = RTN_UNICAST;
1972 rtm->rtm_flags = 0;
1973
1974 if (nla_put_labels(skb, RTA_DST, 1, &label))
1975 goto nla_put_failure;
1976
1977 if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1978 bool ttl_propagate =
1979 rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1980
1981 if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1982 ttl_propagate))
1983 goto nla_put_failure;
1984 }
1985 if (rt->rt_nhn == 1) {
1986 const struct mpls_nh *nh = rt->rt_nh;
1987
1988 if (nh->nh_labels &&
1989 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1990 nh->nh_label))
1991 goto nla_put_failure;
1992 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1993 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1994 nh->nh_via_alen))
1995 goto nla_put_failure;
1996 dev = rtnl_dereference(nh->nh_dev);
1997 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1998 goto nla_put_failure;
1999 if (nh->nh_flags & RTNH_F_LINKDOWN)
2000 rtm->rtm_flags |= RTNH_F_LINKDOWN;
2001 if (nh->nh_flags & RTNH_F_DEAD)
2002 rtm->rtm_flags |= RTNH_F_DEAD;
2003 } else {
2004 struct rtnexthop *rtnh;
2005 struct nlattr *mp;
2006 u8 linkdown = 0;
2007 u8 dead = 0;
2008
2009 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2010 if (!mp)
2011 goto nla_put_failure;
2012
2013 for_nexthops(rt) {
2014 dev = rtnl_dereference(nh->nh_dev);
2015 if (!dev)
2016 continue;
2017
2018 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2019 if (!rtnh)
2020 goto nla_put_failure;
2021
2022 rtnh->rtnh_ifindex = dev->ifindex;
2023 if (nh->nh_flags & RTNH_F_LINKDOWN) {
2024 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2025 linkdown++;
2026 }
2027 if (nh->nh_flags & RTNH_F_DEAD) {
2028 rtnh->rtnh_flags |= RTNH_F_DEAD;
2029 dead++;
2030 }
2031
2032 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2033 nh->nh_labels,
2034 nh->nh_label))
2035 goto nla_put_failure;
2036 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2037 nla_put_via(skb, nh->nh_via_table,
2038 mpls_nh_via(rt, nh),
2039 nh->nh_via_alen))
2040 goto nla_put_failure;
2041
2042 /* length of rtnetlink header + attributes */
2043 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2044 } endfor_nexthops(rt);
2045
2046 if (linkdown == rt->rt_nhn)
2047 rtm->rtm_flags |= RTNH_F_LINKDOWN;
2048 if (dead == rt->rt_nhn)
2049 rtm->rtm_flags |= RTNH_F_DEAD;
2050
2051 nla_nest_end(skb, mp);
2052 }
2053
2054 nlmsg_end(skb, nlh);
2055 return 0;
2056
2057nla_put_failure:
2058 nlmsg_cancel(skb, nlh);
2059 return -EMSGSIZE;
2060}
2061
2062#if IS_ENABLED(CONFIG_INET)
2063static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2064 struct fib_dump_filter *filter,
2065 struct netlink_callback *cb)
2066{
2067 return ip_valid_fib_dump_req(net, nlh, filter, cb);
2068}
2069#else
2070static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2071 struct fib_dump_filter *filter,
2072 struct netlink_callback *cb)
2073{
2074 struct netlink_ext_ack *extack = cb->extack;
2075 struct nlattr *tb[RTA_MAX + 1];
2076 struct rtmsg *rtm;
2077 int err, i;
2078
2079 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2080 NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2081 return -EINVAL;
2082 }
2083
2084 rtm = nlmsg_data(nlh);
2085 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
2086 rtm->rtm_table || rtm->rtm_scope || rtm->rtm_type ||
2087 rtm->rtm_flags) {
2088 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2089 return -EINVAL;
2090 }
2091
2092 if (rtm->rtm_protocol) {
2093 filter->protocol = rtm->rtm_protocol;
2094 filter->filter_set = 1;
2095 cb->answer_flags = NLM_F_DUMP_FILTERED;
2096 }
2097
2098 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2099 rtm_mpls_policy, extack);
2100 if (err < 0)
2101 return err;
2102
2103 for (i = 0; i <= RTA_MAX; ++i) {
2104 int ifindex;
2105
2106 if (i == RTA_OIF) {
2107 ifindex = nla_get_u32(tb[i]);
2108 filter->dev = __dev_get_by_index(net, ifindex);
2109 if (!filter->dev)
2110 return -ENODEV;
2111 filter->filter_set = 1;
2112 } else if (tb[i]) {
2113 NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2114 return -EINVAL;
2115 }
2116 }
2117
2118 return 0;
2119}
2120#endif
2121
2122static bool mpls_rt_uses_dev(struct mpls_route *rt,
2123 const struct net_device *dev)
2124{
2125 struct net_device *nh_dev;
2126
2127 if (rt->rt_nhn == 1) {
2128 struct mpls_nh *nh = rt->rt_nh;
2129
2130 nh_dev = rtnl_dereference(nh->nh_dev);
2131 if (dev == nh_dev)
2132 return true;
2133 } else {
2134 for_nexthops(rt) {
2135 nh_dev = rtnl_dereference(nh->nh_dev);
2136 if (nh_dev == dev)
2137 return true;
2138 } endfor_nexthops(rt);
2139 }
2140
2141 return false;
2142}
2143
2144static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2145{
2146 const struct nlmsghdr *nlh = cb->nlh;
2147 struct net *net = sock_net(skb->sk);
2148 struct mpls_route __rcu **platform_label;
2149 struct fib_dump_filter filter = {};
2150 unsigned int flags = NLM_F_MULTI;
2151 size_t platform_labels;
2152 unsigned int index;
2153
2154 ASSERT_RTNL();
2155
2156 if (cb->strict_check) {
2157 int err;
2158
2159 err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2160 if (err < 0)
2161 return err;
2162
2163 /* for MPLS, there is only 1 table with fixed type and flags.
2164 * If either are set in the filter then return nothing.
2165 */
2166 if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2167 (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2168 filter.flags)
2169 return skb->len;
2170 }
2171
2172 index = cb->args[0];
2173 if (index < MPLS_LABEL_FIRST_UNRESERVED)
2174 index = MPLS_LABEL_FIRST_UNRESERVED;
2175
2176 platform_label = rtnl_dereference(net->mpls.platform_label);
2177 platform_labels = net->mpls.platform_labels;
2178
2179 if (filter.filter_set)
2180 flags |= NLM_F_DUMP_FILTERED;
2181
2182 for (; index < platform_labels; index++) {
2183 struct mpls_route *rt;
2184
2185 rt = rtnl_dereference(platform_label[index]);
2186 if (!rt)
2187 continue;
2188
2189 if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2190 (filter.protocol && rt->rt_protocol != filter.protocol))
2191 continue;
2192
2193 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2194 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2195 index, rt, flags) < 0)
2196 break;
2197 }
2198 cb->args[0] = index;
2199
2200 return skb->len;
2201}
2202
2203static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2204{
2205 size_t payload =
2206 NLMSG_ALIGN(sizeof(struct rtmsg))
2207 + nla_total_size(4) /* RTA_DST */
2208 + nla_total_size(1); /* RTA_TTL_PROPAGATE */
2209
2210 if (rt->rt_nhn == 1) {
2211 struct mpls_nh *nh = rt->rt_nh;
2212
2213 if (nh->nh_dev)
2214 payload += nla_total_size(4); /* RTA_OIF */
2215 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2216 payload += nla_total_size(2 + nh->nh_via_alen);
2217 if (nh->nh_labels) /* RTA_NEWDST */
2218 payload += nla_total_size(nh->nh_labels * 4);
2219 } else {
2220 /* each nexthop is packed in an attribute */
2221 size_t nhsize = 0;
2222
2223 for_nexthops(rt) {
2224 if (!rtnl_dereference(nh->nh_dev))
2225 continue;
2226 nhsize += nla_total_size(sizeof(struct rtnexthop));
2227 /* RTA_VIA */
2228 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2229 nhsize += nla_total_size(2 + nh->nh_via_alen);
2230 if (nh->nh_labels)
2231 nhsize += nla_total_size(nh->nh_labels * 4);
2232 } endfor_nexthops(rt);
2233 /* nested attribute */
2234 payload += nla_total_size(nhsize);
2235 }
2236
2237 return payload;
2238}
2239
2240static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2241 struct nlmsghdr *nlh, struct net *net, u32 portid,
2242 unsigned int nlm_flags)
2243{
2244 struct sk_buff *skb;
2245 u32 seq = nlh ? nlh->nlmsg_seq : 0;
2246 int err = -ENOBUFS;
2247
2248 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2249 if (skb == NULL)
2250 goto errout;
2251
2252 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2253 if (err < 0) {
2254 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2255 WARN_ON(err == -EMSGSIZE);
2256 kfree_skb(skb);
2257 goto errout;
2258 }
2259 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2260
2261 return;
2262errout:
2263 if (err < 0)
2264 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2265}
2266
2267static int mpls_valid_getroute_req(struct sk_buff *skb,
2268 const struct nlmsghdr *nlh,
2269 struct nlattr **tb,
2270 struct netlink_ext_ack *extack)
2271{
2272 struct rtmsg *rtm;
2273 int i, err;
2274
2275 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2276 NL_SET_ERR_MSG_MOD(extack,
2277 "Invalid header for get route request");
2278 return -EINVAL;
2279 }
2280
2281 if (!netlink_strict_get_check(skb))
2282 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2283 rtm_mpls_policy, extack);
2284
2285 rtm = nlmsg_data(nlh);
2286 if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2287 rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2288 rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2289 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2290 return -EINVAL;
2291 }
2292 if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2293 NL_SET_ERR_MSG_MOD(extack,
2294 "Invalid flags for get route request");
2295 return -EINVAL;
2296 }
2297
2298 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2299 rtm_mpls_policy, extack);
2300 if (err)
2301 return err;
2302
2303 if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2304 NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2305 return -EINVAL;
2306 }
2307
2308 for (i = 0; i <= RTA_MAX; i++) {
2309 if (!tb[i])
2310 continue;
2311
2312 switch (i) {
2313 case RTA_DST:
2314 case RTA_NEWDST:
2315 break;
2316 default:
2317 NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2318 return -EINVAL;
2319 }
2320 }
2321
2322 return 0;
2323}
2324
2325static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2326 struct netlink_ext_ack *extack)
2327{
2328 struct net *net = sock_net(in_skb->sk);
2329 u32 portid = NETLINK_CB(in_skb).portid;
2330 u32 in_label = LABEL_NOT_SPECIFIED;
2331 struct nlattr *tb[RTA_MAX + 1];
2332 u32 labels[MAX_NEW_LABELS];
2333 struct mpls_shim_hdr *hdr;
2334 unsigned int hdr_size = 0;
2335 struct net_device *dev;
2336 struct mpls_route *rt;
2337 struct rtmsg *rtm, *r;
2338 struct nlmsghdr *nlh;
2339 struct sk_buff *skb;
2340 struct mpls_nh *nh;
2341 u8 n_labels;
2342 int err;
2343
2344 err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2345 if (err < 0)
2346 goto errout;
2347
2348 rtm = nlmsg_data(in_nlh);
2349
2350 if (tb[RTA_DST]) {
2351 u8 label_count;
2352
2353 if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2354 &in_label, extack)) {
2355 err = -EINVAL;
2356 goto errout;
2357 }
2358
2359 if (!mpls_label_ok(net, &in_label, extack)) {
2360 err = -EINVAL;
2361 goto errout;
2362 }
2363 }
2364
2365 rt = mpls_route_input_rcu(net, in_label);
2366 if (!rt) {
2367 err = -ENETUNREACH;
2368 goto errout;
2369 }
2370
2371 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2372 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2373 if (!skb) {
2374 err = -ENOBUFS;
2375 goto errout;
2376 }
2377
2378 err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2379 RTM_NEWROUTE, in_label, rt, 0);
2380 if (err < 0) {
2381 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2382 WARN_ON(err == -EMSGSIZE);
2383 goto errout_free;
2384 }
2385
2386 return rtnl_unicast(skb, net, portid);
2387 }
2388
2389 if (tb[RTA_NEWDST]) {
2390 if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2391 labels, extack) != 0) {
2392 err = -EINVAL;
2393 goto errout;
2394 }
2395
2396 hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2397 }
2398
2399 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2400 if (!skb) {
2401 err = -ENOBUFS;
2402 goto errout;
2403 }
2404
2405 skb->protocol = htons(ETH_P_MPLS_UC);
2406
2407 if (hdr_size) {
2408 bool bos;
2409 int i;
2410
2411 if (skb_cow(skb, hdr_size)) {
2412 err = -ENOBUFS;
2413 goto errout_free;
2414 }
2415
2416 skb_reserve(skb, hdr_size);
2417 skb_push(skb, hdr_size);
2418 skb_reset_network_header(skb);
2419
2420 /* Push new labels */
2421 hdr = mpls_hdr(skb);
2422 bos = true;
2423 for (i = n_labels - 1; i >= 0; i--) {
2424 hdr[i] = mpls_entry_encode(labels[i],
2425 1, 0, bos);
2426 bos = false;
2427 }
2428 }
2429
2430 nh = mpls_select_multipath(rt, skb);
2431 if (!nh) {
2432 err = -ENETUNREACH;
2433 goto errout_free;
2434 }
2435
2436 if (hdr_size) {
2437 skb_pull(skb, hdr_size);
2438 skb_reset_network_header(skb);
2439 }
2440
2441 nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2442 RTM_NEWROUTE, sizeof(*r), 0);
2443 if (!nlh) {
2444 err = -EMSGSIZE;
2445 goto errout_free;
2446 }
2447
2448 r = nlmsg_data(nlh);
2449 r->rtm_family = AF_MPLS;
2450 r->rtm_dst_len = 20;
2451 r->rtm_src_len = 0;
2452 r->rtm_table = RT_TABLE_MAIN;
2453 r->rtm_type = RTN_UNICAST;
2454 r->rtm_scope = RT_SCOPE_UNIVERSE;
2455 r->rtm_protocol = rt->rt_protocol;
2456 r->rtm_flags = 0;
2457
2458 if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2459 goto nla_put_failure;
2460
2461 if (nh->nh_labels &&
2462 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2463 nh->nh_label))
2464 goto nla_put_failure;
2465
2466 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2467 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2468 nh->nh_via_alen))
2469 goto nla_put_failure;
2470 dev = rtnl_dereference(nh->nh_dev);
2471 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2472 goto nla_put_failure;
2473
2474 nlmsg_end(skb, nlh);
2475
2476 err = rtnl_unicast(skb, net, portid);
2477errout:
2478 return err;
2479
2480nla_put_failure:
2481 nlmsg_cancel(skb, nlh);
2482 err = -EMSGSIZE;
2483errout_free:
2484 kfree_skb(skb);
2485 return err;
2486}
2487
2488static int resize_platform_label_table(struct net *net, size_t limit)
2489{
2490 size_t size = sizeof(struct mpls_route *) * limit;
2491 size_t old_limit;
2492 size_t cp_size;
2493 struct mpls_route __rcu **labels = NULL, **old;
2494 struct mpls_route *rt0 = NULL, *rt2 = NULL;
2495 unsigned index;
2496
2497 if (size) {
2498 labels = kvzalloc(size, GFP_KERNEL);
2499 if (!labels)
2500 goto nolabels;
2501 }
2502
2503 /* In case the predefined labels need to be populated */
2504 if (limit > MPLS_LABEL_IPV4NULL) {
2505 struct net_device *lo = net->loopback_dev;
2506 rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2507 if (IS_ERR(rt0))
2508 goto nort0;
2509 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2510 rt0->rt_protocol = RTPROT_KERNEL;
2511 rt0->rt_payload_type = MPT_IPV4;
2512 rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2513 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2514 rt0->rt_nh->nh_via_alen = lo->addr_len;
2515 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2516 lo->addr_len);
2517 }
2518 if (limit > MPLS_LABEL_IPV6NULL) {
2519 struct net_device *lo = net->loopback_dev;
2520 rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2521 if (IS_ERR(rt2))
2522 goto nort2;
2523 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2524 rt2->rt_protocol = RTPROT_KERNEL;
2525 rt2->rt_payload_type = MPT_IPV6;
2526 rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2527 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2528 rt2->rt_nh->nh_via_alen = lo->addr_len;
2529 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2530 lo->addr_len);
2531 }
2532
2533 rtnl_lock();
2534 /* Remember the original table */
2535 old = rtnl_dereference(net->mpls.platform_label);
2536 old_limit = net->mpls.platform_labels;
2537
2538 /* Free any labels beyond the new table */
2539 for (index = limit; index < old_limit; index++)
2540 mpls_route_update(net, index, NULL, NULL);
2541
2542 /* Copy over the old labels */
2543 cp_size = size;
2544 if (old_limit < limit)
2545 cp_size = old_limit * sizeof(struct mpls_route *);
2546
2547 memcpy(labels, old, cp_size);
2548
2549 /* If needed set the predefined labels */
2550 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2551 (limit > MPLS_LABEL_IPV6NULL)) {
2552 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2553 rt2 = NULL;
2554 }
2555
2556 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2557 (limit > MPLS_LABEL_IPV4NULL)) {
2558 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2559 rt0 = NULL;
2560 }
2561
2562 /* Update the global pointers */
2563 net->mpls.platform_labels = limit;
2564 rcu_assign_pointer(net->mpls.platform_label, labels);
2565
2566 rtnl_unlock();
2567
2568 mpls_rt_free(rt2);
2569 mpls_rt_free(rt0);
2570
2571 if (old) {
2572 synchronize_rcu();
2573 kvfree(old);
2574 }
2575 return 0;
2576
2577nort2:
2578 mpls_rt_free(rt0);
2579nort0:
2580 kvfree(labels);
2581nolabels:
2582 return -ENOMEM;
2583}
2584
2585static int mpls_platform_labels(struct ctl_table *table, int write,
2586 void *buffer, size_t *lenp, loff_t *ppos)
2587{
2588 struct net *net = table->data;
2589 int platform_labels = net->mpls.platform_labels;
2590 int ret;
2591 struct ctl_table tmp = {
2592 .procname = table->procname,
2593 .data = &platform_labels,
2594 .maxlen = sizeof(int),
2595 .mode = table->mode,
2596 .extra1 = SYSCTL_ZERO,
2597 .extra2 = &label_limit,
2598 };
2599
2600 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2601
2602 if (write && ret == 0)
2603 ret = resize_platform_label_table(net, platform_labels);
2604
2605 return ret;
2606}
2607
2608#define MPLS_NS_SYSCTL_OFFSET(field) \
2609 (&((struct net *)0)->field)
2610
2611static const struct ctl_table mpls_table[] = {
2612 {
2613 .procname = "platform_labels",
2614 .data = NULL,
2615 .maxlen = sizeof(int),
2616 .mode = 0644,
2617 .proc_handler = mpls_platform_labels,
2618 },
2619 {
2620 .procname = "ip_ttl_propagate",
2621 .data = MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2622 .maxlen = sizeof(int),
2623 .mode = 0644,
2624 .proc_handler = proc_dointvec_minmax,
2625 .extra1 = SYSCTL_ZERO,
2626 .extra2 = SYSCTL_ONE,
2627 },
2628 {
2629 .procname = "default_ttl",
2630 .data = MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2631 .maxlen = sizeof(int),
2632 .mode = 0644,
2633 .proc_handler = proc_dointvec_minmax,
2634 .extra1 = SYSCTL_ONE,
2635 .extra2 = &ttl_max,
2636 },
2637 { }
2638};
2639
2640static int mpls_net_init(struct net *net)
2641{
2642 struct ctl_table *table;
2643 int i;
2644
2645 net->mpls.platform_labels = 0;
2646 net->mpls.platform_label = NULL;
2647 net->mpls.ip_ttl_propagate = 1;
2648 net->mpls.default_ttl = 255;
2649
2650 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2651 if (table == NULL)
2652 return -ENOMEM;
2653
2654 /* Table data contains only offsets relative to the base of
2655 * the mdev at this point, so make them absolute.
2656 */
2657 for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2658 table[i].data = (char *)net + (uintptr_t)table[i].data;
2659
2660 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2661 if (net->mpls.ctl == NULL) {
2662 kfree(table);
2663 return -ENOMEM;
2664 }
2665
2666 return 0;
2667}
2668
2669static void mpls_net_exit(struct net *net)
2670{
2671 struct mpls_route __rcu **platform_label;
2672 size_t platform_labels;
2673 struct ctl_table *table;
2674 unsigned int index;
2675
2676 table = net->mpls.ctl->ctl_table_arg;
2677 unregister_net_sysctl_table(net->mpls.ctl);
2678 kfree(table);
2679
2680 /* An rcu grace period has passed since there was a device in
2681 * the network namespace (and thus the last in flight packet)
2682 * left this network namespace. This is because
2683 * unregister_netdevice_many and netdev_run_todo has completed
2684 * for each network device that was in this network namespace.
2685 *
2686 * As such no additional rcu synchronization is necessary when
2687 * freeing the platform_label table.
2688 */
2689 rtnl_lock();
2690 platform_label = rtnl_dereference(net->mpls.platform_label);
2691 platform_labels = net->mpls.platform_labels;
2692 for (index = 0; index < platform_labels; index++) {
2693 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2694 RCU_INIT_POINTER(platform_label[index], NULL);
2695 mpls_notify_route(net, index, rt, NULL, NULL);
2696 mpls_rt_free(rt);
2697 }
2698 rtnl_unlock();
2699
2700 kvfree(platform_label);
2701}
2702
2703static struct pernet_operations mpls_net_ops = {
2704 .init = mpls_net_init,
2705 .exit = mpls_net_exit,
2706};
2707
2708static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2709 .family = AF_MPLS,
2710 .fill_stats_af = mpls_fill_stats_af,
2711 .get_stats_af_size = mpls_get_stats_af_size,
2712};
2713
2714static int __init mpls_init(void)
2715{
2716 int err;
2717
2718 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2719
2720 err = register_pernet_subsys(&mpls_net_ops);
2721 if (err)
2722 goto out;
2723
2724 err = register_netdevice_notifier(&mpls_dev_notifier);
2725 if (err)
2726 goto out_unregister_pernet;
2727
2728 dev_add_pack(&mpls_packet_type);
2729
2730 rtnl_af_register(&mpls_af_ops);
2731
2732 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2733 mpls_rtm_newroute, NULL, 0);
2734 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2735 mpls_rtm_delroute, NULL, 0);
2736 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2737 mpls_getroute, mpls_dump_routes, 0);
2738 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2739 mpls_netconf_get_devconf,
2740 mpls_netconf_dump_devconf, 0);
2741 err = ipgre_tunnel_encap_add_mpls_ops();
2742 if (err)
2743 pr_err("Can't add mpls over gre tunnel ops\n");
2744
2745 err = 0;
2746out:
2747 return err;
2748
2749out_unregister_pernet:
2750 unregister_pernet_subsys(&mpls_net_ops);
2751 goto out;
2752}
2753module_init(mpls_init);
2754
2755static void __exit mpls_exit(void)
2756{
2757 rtnl_unregister_all(PF_MPLS);
2758 rtnl_af_unregister(&mpls_af_ops);
2759 dev_remove_pack(&mpls_packet_type);
2760 unregister_netdevice_notifier(&mpls_dev_notifier);
2761 unregister_pernet_subsys(&mpls_net_ops);
2762 ipgre_tunnel_encap_del_mpls_ops();
2763}
2764module_exit(mpls_exit);
2765
2766MODULE_DESCRIPTION("MultiProtocol Label Switching");
2767MODULE_LICENSE("GPL v2");
2768MODULE_ALIAS_NETPROTO(PF_MPLS);