Loading...
Note: File does not exist in v3.15.
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_validate_mtu(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, struct sk_buff *skb)
102{
103 struct mpls_entry_decoded dec;
104 unsigned int mpls_hdr_len = 0;
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;
111 label_index++) {
112 mpls_hdr_len += sizeof(*hdr);
113 if (!pskb_may_pull(skb, mpls_hdr_len))
114 break;
115
116 /* Read and decode the current label */
117 hdr = mpls_hdr(skb) + label_index;
118 dec = mpls_entry_decode(hdr);
119
120 /* RFC6790 - reserved labels MUST NOT be used as keys
121 * for the load-balancing function
122 */
123 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
124 hash = jhash_1word(dec.label, hash);
125
126 /* The entropy label follows the entropy label
127 * indicator, so this means that the entropy
128 * label was just added to the hash - no need to
129 * go any deeper either in the label stack or in the
130 * payload
131 */
132 if (eli_seen)
133 break;
134 } else if (dec.label == MPLS_LABEL_ENTROPY) {
135 eli_seen = true;
136 }
137
138 if (!dec.bos)
139 continue;
140
141 /* found bottom label; does skb have room for a header? */
142 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
143 const struct iphdr *v4hdr;
144
145 v4hdr = (const struct iphdr *)(hdr + 1);
146 if (v4hdr->version == 4) {
147 hash = jhash_3words(ntohl(v4hdr->saddr),
148 ntohl(v4hdr->daddr),
149 v4hdr->protocol, hash);
150 } else if (v4hdr->version == 6 &&
151 pskb_may_pull(skb, mpls_hdr_len +
152 sizeof(struct ipv6hdr))) {
153 const struct ipv6hdr *v6hdr;
154
155 v6hdr = (const struct ipv6hdr *)(hdr + 1);
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 break;
163 }
164
165 return hash;
166}
167
168static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
169 struct sk_buff *skb)
170{
171 int alive = ACCESS_ONCE(rt->rt_nhn_alive);
172 u32 hash = 0;
173 int nh_index = 0;
174 int n = 0;
175
176 /* No need to look further into packet if there's only
177 * one path
178 */
179 if (rt->rt_nhn == 1)
180 goto out;
181
182 if (alive <= 0)
183 return NULL;
184
185 hash = mpls_multipath_hash(rt, skb);
186 nh_index = hash % alive;
187 if (alive == rt->rt_nhn)
188 goto out;
189 for_nexthops(rt) {
190 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
191 continue;
192 if (n == nh_index)
193 return nh;
194 n++;
195 } endfor_nexthops(rt);
196
197out:
198 return &rt->rt_nh[nh_index];
199}
200
201static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
202 struct mpls_entry_decoded dec)
203{
204 enum mpls_payload_type payload_type;
205 bool success = false;
206
207 /* The IPv4 code below accesses through the IPv4 header
208 * checksum, which is 12 bytes into the packet.
209 * The IPv6 code below accesses through the IPv6 hop limit
210 * which is 8 bytes into the packet.
211 *
212 * For all supported cases there should always be at least 12
213 * bytes of packet data present. The IPv4 header is 20 bytes
214 * without options and the IPv6 header is always 40 bytes
215 * long.
216 */
217 if (!pskb_may_pull(skb, 12))
218 return false;
219
220 payload_type = rt->rt_payload_type;
221 if (payload_type == MPT_UNSPEC)
222 payload_type = ip_hdr(skb)->version;
223
224 switch (payload_type) {
225 case MPT_IPV4: {
226 struct iphdr *hdr4 = ip_hdr(skb);
227 skb->protocol = htons(ETH_P_IP);
228 csum_replace2(&hdr4->check,
229 htons(hdr4->ttl << 8),
230 htons(dec.ttl << 8));
231 hdr4->ttl = dec.ttl;
232 success = true;
233 break;
234 }
235 case MPT_IPV6: {
236 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
237 skb->protocol = htons(ETH_P_IPV6);
238 hdr6->hop_limit = dec.ttl;
239 success = true;
240 break;
241 }
242 case MPT_UNSPEC:
243 break;
244 }
245
246 return success;
247}
248
249static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
250 struct packet_type *pt, struct net_device *orig_dev)
251{
252 struct net *net = dev_net(dev);
253 struct mpls_shim_hdr *hdr;
254 struct mpls_route *rt;
255 struct mpls_nh *nh;
256 struct mpls_entry_decoded dec;
257 struct net_device *out_dev;
258 struct mpls_dev *mdev;
259 unsigned int hh_len;
260 unsigned int new_header_size;
261 unsigned int mtu;
262 int err;
263
264 /* Careful this entire function runs inside of an rcu critical section */
265
266 mdev = mpls_dev_get(dev);
267 if (!mdev || !mdev->input_enabled)
268 goto drop;
269
270 if (skb->pkt_type != PACKET_HOST)
271 goto drop;
272
273 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
274 goto drop;
275
276 if (!pskb_may_pull(skb, sizeof(*hdr)))
277 goto drop;
278
279 /* Read and decode the label */
280 hdr = mpls_hdr(skb);
281 dec = mpls_entry_decode(hdr);
282
283 rt = mpls_route_input_rcu(net, dec.label);
284 if (!rt)
285 goto drop;
286
287 nh = mpls_select_multipath(rt, skb);
288 if (!nh)
289 goto drop;
290
291 /* Find the output device */
292 out_dev = rcu_dereference(nh->nh_dev);
293 if (!mpls_output_possible(out_dev))
294 goto drop;
295
296 /* Pop the label */
297 skb_pull(skb, sizeof(*hdr));
298 skb_reset_network_header(skb);
299
300 skb_orphan(skb);
301
302 if (skb_warn_if_lro(skb))
303 goto drop;
304
305 skb_forward_csum(skb);
306
307 /* Verify ttl is valid */
308 if (dec.ttl <= 1)
309 goto drop;
310 dec.ttl -= 1;
311
312 /* Verify the destination can hold the packet */
313 new_header_size = mpls_nh_header_size(nh);
314 mtu = mpls_dev_mtu(out_dev);
315 if (mpls_pkt_too_big(skb, mtu - new_header_size))
316 goto drop;
317
318 hh_len = LL_RESERVED_SPACE(out_dev);
319 if (!out_dev->header_ops)
320 hh_len = 0;
321
322 /* Ensure there is enough space for the headers in the skb */
323 if (skb_cow(skb, hh_len + new_header_size))
324 goto drop;
325
326 skb->dev = out_dev;
327 skb->protocol = htons(ETH_P_MPLS_UC);
328
329 if (unlikely(!new_header_size && dec.bos)) {
330 /* Penultimate hop popping */
331 if (!mpls_egress(rt, skb, dec))
332 goto drop;
333 } else {
334 bool bos;
335 int i;
336 skb_push(skb, new_header_size);
337 skb_reset_network_header(skb);
338 /* Push the new labels */
339 hdr = mpls_hdr(skb);
340 bos = dec.bos;
341 for (i = nh->nh_labels - 1; i >= 0; i--) {
342 hdr[i] = mpls_entry_encode(nh->nh_label[i],
343 dec.ttl, 0, bos);
344 bos = false;
345 }
346 }
347
348 /* If via wasn't specified then send out using device address */
349 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
350 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
351 out_dev->dev_addr, skb);
352 else
353 err = neigh_xmit(nh->nh_via_table, out_dev,
354 mpls_nh_via(rt, nh), skb);
355 if (err)
356 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
357 __func__, err);
358 return 0;
359
360drop:
361 kfree_skb(skb);
362 return NET_RX_DROP;
363}
364
365static struct packet_type mpls_packet_type __read_mostly = {
366 .type = cpu_to_be16(ETH_P_MPLS_UC),
367 .func = mpls_forward,
368};
369
370static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
371 [RTA_DST] = { .type = NLA_U32 },
372 [RTA_OIF] = { .type = NLA_U32 },
373};
374
375struct mpls_route_config {
376 u32 rc_protocol;
377 u32 rc_ifindex;
378 u8 rc_via_table;
379 u8 rc_via_alen;
380 u8 rc_via[MAX_VIA_ALEN];
381 u32 rc_label;
382 u8 rc_output_labels;
383 u32 rc_output_label[MAX_NEW_LABELS];
384 u32 rc_nlflags;
385 enum mpls_payload_type rc_payload_type;
386 struct nl_info rc_nlinfo;
387 struct rtnexthop *rc_mp;
388 int rc_mp_len;
389};
390
391static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
392{
393 u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
394 struct mpls_route *rt;
395
396 rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
397 VIA_ALEN_ALIGN) +
398 num_nh * max_alen_aligned,
399 GFP_KERNEL);
400 if (rt) {
401 rt->rt_nhn = num_nh;
402 rt->rt_nhn_alive = num_nh;
403 rt->rt_max_alen = max_alen_aligned;
404 }
405
406 return rt;
407}
408
409static void mpls_rt_free(struct mpls_route *rt)
410{
411 if (rt)
412 kfree_rcu(rt, rt_rcu);
413}
414
415static void mpls_notify_route(struct net *net, unsigned index,
416 struct mpls_route *old, struct mpls_route *new,
417 const struct nl_info *info)
418{
419 struct nlmsghdr *nlh = info ? info->nlh : NULL;
420 unsigned portid = info ? info->portid : 0;
421 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
422 struct mpls_route *rt = new ? new : old;
423 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
424 /* Ignore reserved labels for now */
425 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
426 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
427}
428
429static void mpls_route_update(struct net *net, unsigned index,
430 struct mpls_route *new,
431 const struct nl_info *info)
432{
433 struct mpls_route __rcu **platform_label;
434 struct mpls_route *rt;
435
436 ASSERT_RTNL();
437
438 platform_label = rtnl_dereference(net->mpls.platform_label);
439 rt = rtnl_dereference(platform_label[index]);
440 rcu_assign_pointer(platform_label[index], new);
441
442 mpls_notify_route(net, index, rt, new, info);
443
444 /* If we removed a route free it now */
445 mpls_rt_free(rt);
446}
447
448static unsigned find_free_label(struct net *net)
449{
450 struct mpls_route __rcu **platform_label;
451 size_t platform_labels;
452 unsigned index;
453
454 platform_label = rtnl_dereference(net->mpls.platform_label);
455 platform_labels = net->mpls.platform_labels;
456 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
457 index++) {
458 if (!rtnl_dereference(platform_label[index]))
459 return index;
460 }
461 return LABEL_NOT_SPECIFIED;
462}
463
464#if IS_ENABLED(CONFIG_INET)
465static struct net_device *inet_fib_lookup_dev(struct net *net,
466 const void *addr)
467{
468 struct net_device *dev;
469 struct rtable *rt;
470 struct in_addr daddr;
471
472 memcpy(&daddr, addr, sizeof(struct in_addr));
473 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
474 if (IS_ERR(rt))
475 return ERR_CAST(rt);
476
477 dev = rt->dst.dev;
478 dev_hold(dev);
479
480 ip_rt_put(rt);
481
482 return dev;
483}
484#else
485static struct net_device *inet_fib_lookup_dev(struct net *net,
486 const void *addr)
487{
488 return ERR_PTR(-EAFNOSUPPORT);
489}
490#endif
491
492#if IS_ENABLED(CONFIG_IPV6)
493static struct net_device *inet6_fib_lookup_dev(struct net *net,
494 const void *addr)
495{
496 struct net_device *dev;
497 struct dst_entry *dst;
498 struct flowi6 fl6;
499 int err;
500
501 if (!ipv6_stub)
502 return ERR_PTR(-EAFNOSUPPORT);
503
504 memset(&fl6, 0, sizeof(fl6));
505 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
506 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
507 if (err)
508 return ERR_PTR(err);
509
510 dev = dst->dev;
511 dev_hold(dev);
512 dst_release(dst);
513
514 return dev;
515}
516#else
517static struct net_device *inet6_fib_lookup_dev(struct net *net,
518 const void *addr)
519{
520 return ERR_PTR(-EAFNOSUPPORT);
521}
522#endif
523
524static struct net_device *find_outdev(struct net *net,
525 struct mpls_route *rt,
526 struct mpls_nh *nh, int oif)
527{
528 struct net_device *dev = NULL;
529
530 if (!oif) {
531 switch (nh->nh_via_table) {
532 case NEIGH_ARP_TABLE:
533 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
534 break;
535 case NEIGH_ND_TABLE:
536 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
537 break;
538 case NEIGH_LINK_TABLE:
539 break;
540 }
541 } else {
542 dev = dev_get_by_index(net, oif);
543 }
544
545 if (!dev)
546 return ERR_PTR(-ENODEV);
547
548 if (IS_ERR(dev))
549 return dev;
550
551 /* The caller is holding rtnl anyways, so release the dev reference */
552 dev_put(dev);
553
554 return dev;
555}
556
557static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
558 struct mpls_nh *nh, int oif)
559{
560 struct net_device *dev = NULL;
561 int err = -ENODEV;
562
563 dev = find_outdev(net, rt, nh, oif);
564 if (IS_ERR(dev)) {
565 err = PTR_ERR(dev);
566 dev = NULL;
567 goto errout;
568 }
569
570 /* Ensure this is a supported device */
571 err = -EINVAL;
572 if (!mpls_dev_get(dev))
573 goto errout;
574
575 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
576 (dev->addr_len != nh->nh_via_alen))
577 goto errout;
578
579 RCU_INIT_POINTER(nh->nh_dev, dev);
580
581 if (!(dev->flags & IFF_UP)) {
582 nh->nh_flags |= RTNH_F_DEAD;
583 } else {
584 unsigned int flags;
585
586 flags = dev_get_flags(dev);
587 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
588 nh->nh_flags |= RTNH_F_LINKDOWN;
589 }
590
591 return 0;
592
593errout:
594 return err;
595}
596
597static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
598 struct mpls_route *rt)
599{
600 struct net *net = cfg->rc_nlinfo.nl_net;
601 struct mpls_nh *nh = rt->rt_nh;
602 int err;
603 int i;
604
605 if (!nh)
606 return -ENOMEM;
607
608 err = -EINVAL;
609 /* Ensure only a supported number of labels are present */
610 if (cfg->rc_output_labels > MAX_NEW_LABELS)
611 goto errout;
612
613 nh->nh_labels = cfg->rc_output_labels;
614 for (i = 0; i < nh->nh_labels; i++)
615 nh->nh_label[i] = cfg->rc_output_label[i];
616
617 nh->nh_via_table = cfg->rc_via_table;
618 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
619 nh->nh_via_alen = cfg->rc_via_alen;
620
621 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
622 if (err)
623 goto errout;
624
625 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
626 rt->rt_nhn_alive--;
627
628 return 0;
629
630errout:
631 return err;
632}
633
634static int mpls_nh_build(struct net *net, struct mpls_route *rt,
635 struct mpls_nh *nh, int oif, struct nlattr *via,
636 struct nlattr *newdst)
637{
638 int err = -ENOMEM;
639
640 if (!nh)
641 goto errout;
642
643 if (newdst) {
644 err = nla_get_labels(newdst, MAX_NEW_LABELS,
645 &nh->nh_labels, nh->nh_label);
646 if (err)
647 goto errout;
648 }
649
650 if (via) {
651 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
652 __mpls_nh_via(rt, nh));
653 if (err)
654 goto errout;
655 } else {
656 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
657 }
658
659 err = mpls_nh_assign_dev(net, rt, nh, oif);
660 if (err)
661 goto errout;
662
663 return 0;
664
665errout:
666 return err;
667}
668
669static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
670 u8 cfg_via_alen, u8 *max_via_alen)
671{
672 int nhs = 0;
673 int remaining = len;
674
675 if (!rtnh) {
676 *max_via_alen = cfg_via_alen;
677 return 1;
678 }
679
680 *max_via_alen = 0;
681
682 while (rtnh_ok(rtnh, remaining)) {
683 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
684 int attrlen;
685
686 attrlen = rtnh_attrlen(rtnh);
687 nla = nla_find(attrs, attrlen, RTA_VIA);
688 if (nla && nla_len(nla) >=
689 offsetof(struct rtvia, rtvia_addr)) {
690 int via_alen = nla_len(nla) -
691 offsetof(struct rtvia, rtvia_addr);
692
693 if (via_alen <= MAX_VIA_ALEN)
694 *max_via_alen = max_t(u16, *max_via_alen,
695 via_alen);
696 }
697
698 nhs++;
699 rtnh = rtnh_next(rtnh, &remaining);
700 }
701
702 /* leftover implies invalid nexthop configuration, discard it */
703 return remaining > 0 ? 0 : nhs;
704}
705
706static int mpls_nh_build_multi(struct mpls_route_config *cfg,
707 struct mpls_route *rt)
708{
709 struct rtnexthop *rtnh = cfg->rc_mp;
710 struct nlattr *nla_via, *nla_newdst;
711 int remaining = cfg->rc_mp_len;
712 int nhs = 0;
713 int err = 0;
714
715 change_nexthops(rt) {
716 int attrlen;
717
718 nla_via = NULL;
719 nla_newdst = NULL;
720
721 err = -EINVAL;
722 if (!rtnh_ok(rtnh, remaining))
723 goto errout;
724
725 /* neither weighted multipath nor any flags
726 * are supported
727 */
728 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
729 goto errout;
730
731 attrlen = rtnh_attrlen(rtnh);
732 if (attrlen > 0) {
733 struct nlattr *attrs = rtnh_attrs(rtnh);
734
735 nla_via = nla_find(attrs, attrlen, RTA_VIA);
736 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
737 }
738
739 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
740 rtnh->rtnh_ifindex, nla_via, nla_newdst);
741 if (err)
742 goto errout;
743
744 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
745 rt->rt_nhn_alive--;
746
747 rtnh = rtnh_next(rtnh, &remaining);
748 nhs++;
749 } endfor_nexthops(rt);
750
751 rt->rt_nhn = nhs;
752
753 return 0;
754
755errout:
756 return err;
757}
758
759static int mpls_route_add(struct mpls_route_config *cfg)
760{
761 struct mpls_route __rcu **platform_label;
762 struct net *net = cfg->rc_nlinfo.nl_net;
763 struct mpls_route *rt, *old;
764 int err = -EINVAL;
765 u8 max_via_alen;
766 unsigned index;
767 int nhs;
768
769 index = cfg->rc_label;
770
771 /* If a label was not specified during insert pick one */
772 if ((index == LABEL_NOT_SPECIFIED) &&
773 (cfg->rc_nlflags & NLM_F_CREATE)) {
774 index = find_free_label(net);
775 }
776
777 /* Reserved labels may not be set */
778 if (index < MPLS_LABEL_FIRST_UNRESERVED)
779 goto errout;
780
781 /* The full 20 bit range may not be supported. */
782 if (index >= net->mpls.platform_labels)
783 goto errout;
784
785 /* Append makes no sense with mpls */
786 err = -EOPNOTSUPP;
787 if (cfg->rc_nlflags & NLM_F_APPEND)
788 goto errout;
789
790 err = -EEXIST;
791 platform_label = rtnl_dereference(net->mpls.platform_label);
792 old = rtnl_dereference(platform_label[index]);
793 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
794 goto errout;
795
796 err = -EEXIST;
797 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
798 goto errout;
799
800 err = -ENOENT;
801 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
802 goto errout;
803
804 err = -EINVAL;
805 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
806 cfg->rc_via_alen, &max_via_alen);
807 if (nhs == 0)
808 goto errout;
809
810 err = -ENOMEM;
811 rt = mpls_rt_alloc(nhs, max_via_alen);
812 if (!rt)
813 goto errout;
814
815 rt->rt_protocol = cfg->rc_protocol;
816 rt->rt_payload_type = cfg->rc_payload_type;
817
818 if (cfg->rc_mp)
819 err = mpls_nh_build_multi(cfg, rt);
820 else
821 err = mpls_nh_build_from_cfg(cfg, rt);
822 if (err)
823 goto freert;
824
825 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
826
827 return 0;
828
829freert:
830 mpls_rt_free(rt);
831errout:
832 return err;
833}
834
835static int mpls_route_del(struct mpls_route_config *cfg)
836{
837 struct net *net = cfg->rc_nlinfo.nl_net;
838 unsigned index;
839 int err = -EINVAL;
840
841 index = cfg->rc_label;
842
843 /* Reserved labels may not be removed */
844 if (index < MPLS_LABEL_FIRST_UNRESERVED)
845 goto errout;
846
847 /* The full 20 bit range may not be supported */
848 if (index >= net->mpls.platform_labels)
849 goto errout;
850
851 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
852
853 err = 0;
854errout:
855 return err;
856}
857
858#define MPLS_PERDEV_SYSCTL_OFFSET(field) \
859 (&((struct mpls_dev *)0)->field)
860
861static const struct ctl_table mpls_dev_table[] = {
862 {
863 .procname = "input",
864 .maxlen = sizeof(int),
865 .mode = 0644,
866 .proc_handler = proc_dointvec,
867 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
868 },
869 { }
870};
871
872static int mpls_dev_sysctl_register(struct net_device *dev,
873 struct mpls_dev *mdev)
874{
875 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
876 struct ctl_table *table;
877 int i;
878
879 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
880 if (!table)
881 goto out;
882
883 /* Table data contains only offsets relative to the base of
884 * the mdev at this point, so make them absolute.
885 */
886 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
887 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
888
889 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
890
891 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
892 if (!mdev->sysctl)
893 goto free;
894
895 return 0;
896
897free:
898 kfree(table);
899out:
900 return -ENOBUFS;
901}
902
903static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
904{
905 struct ctl_table *table;
906
907 table = mdev->sysctl->ctl_table_arg;
908 unregister_net_sysctl_table(mdev->sysctl);
909 kfree(table);
910}
911
912static struct mpls_dev *mpls_add_dev(struct net_device *dev)
913{
914 struct mpls_dev *mdev;
915 int err = -ENOMEM;
916
917 ASSERT_RTNL();
918
919 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
920 if (!mdev)
921 return ERR_PTR(err);
922
923 err = mpls_dev_sysctl_register(dev, mdev);
924 if (err)
925 goto free;
926
927 rcu_assign_pointer(dev->mpls_ptr, mdev);
928
929 return mdev;
930
931free:
932 kfree(mdev);
933 return ERR_PTR(err);
934}
935
936static void mpls_ifdown(struct net_device *dev, int event)
937{
938 struct mpls_route __rcu **platform_label;
939 struct net *net = dev_net(dev);
940 unsigned index;
941
942 platform_label = rtnl_dereference(net->mpls.platform_label);
943 for (index = 0; index < net->mpls.platform_labels; index++) {
944 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
945
946 if (!rt)
947 continue;
948
949 change_nexthops(rt) {
950 if (rtnl_dereference(nh->nh_dev) != dev)
951 continue;
952 switch (event) {
953 case NETDEV_DOWN:
954 case NETDEV_UNREGISTER:
955 nh->nh_flags |= RTNH_F_DEAD;
956 /* fall through */
957 case NETDEV_CHANGE:
958 nh->nh_flags |= RTNH_F_LINKDOWN;
959 if (event != NETDEV_UNREGISTER)
960 ACCESS_ONCE(rt->rt_nhn_alive) = rt->rt_nhn_alive - 1;
961 break;
962 }
963 if (event == NETDEV_UNREGISTER)
964 RCU_INIT_POINTER(nh->nh_dev, NULL);
965 } endfor_nexthops(rt);
966 }
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
1002static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1003 void *ptr)
1004{
1005 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1006 struct mpls_dev *mdev;
1007 unsigned int flags;
1008
1009 if (event == NETDEV_REGISTER) {
1010 /* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1011 if (dev->type == ARPHRD_ETHER ||
1012 dev->type == ARPHRD_LOOPBACK ||
1013 dev->type == ARPHRD_IPGRE ||
1014 dev->type == ARPHRD_SIT ||
1015 dev->type == ARPHRD_TUNNEL) {
1016 mdev = mpls_add_dev(dev);
1017 if (IS_ERR(mdev))
1018 return notifier_from_errno(PTR_ERR(mdev));
1019 }
1020 return NOTIFY_OK;
1021 }
1022
1023 mdev = mpls_dev_get(dev);
1024 if (!mdev)
1025 return NOTIFY_OK;
1026
1027 switch (event) {
1028 case NETDEV_DOWN:
1029 mpls_ifdown(dev, event);
1030 break;
1031 case NETDEV_UP:
1032 flags = dev_get_flags(dev);
1033 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1034 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1035 else
1036 mpls_ifup(dev, RTNH_F_DEAD);
1037 break;
1038 case NETDEV_CHANGE:
1039 flags = dev_get_flags(dev);
1040 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1041 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1042 else
1043 mpls_ifdown(dev, event);
1044 break;
1045 case NETDEV_UNREGISTER:
1046 mpls_ifdown(dev, event);
1047 mdev = mpls_dev_get(dev);
1048 if (mdev) {
1049 mpls_dev_sysctl_unregister(mdev);
1050 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1051 kfree_rcu(mdev, rcu);
1052 }
1053 break;
1054 case NETDEV_CHANGENAME:
1055 mdev = mpls_dev_get(dev);
1056 if (mdev) {
1057 int err;
1058
1059 mpls_dev_sysctl_unregister(mdev);
1060 err = mpls_dev_sysctl_register(dev, mdev);
1061 if (err)
1062 return notifier_from_errno(err);
1063 }
1064 break;
1065 }
1066 return NOTIFY_OK;
1067}
1068
1069static struct notifier_block mpls_dev_notifier = {
1070 .notifier_call = mpls_dev_notify,
1071};
1072
1073static int nla_put_via(struct sk_buff *skb,
1074 u8 table, const void *addr, int alen)
1075{
1076 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1077 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1078 };
1079 struct nlattr *nla;
1080 struct rtvia *via;
1081 int family = AF_UNSPEC;
1082
1083 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1084 if (!nla)
1085 return -EMSGSIZE;
1086
1087 if (table <= NEIGH_NR_TABLES)
1088 family = table_to_family[table];
1089
1090 via = nla_data(nla);
1091 via->rtvia_family = family;
1092 memcpy(via->rtvia_addr, addr, alen);
1093 return 0;
1094}
1095
1096int nla_put_labels(struct sk_buff *skb, int attrtype,
1097 u8 labels, const u32 label[])
1098{
1099 struct nlattr *nla;
1100 struct mpls_shim_hdr *nla_label;
1101 bool bos;
1102 int i;
1103 nla = nla_reserve(skb, attrtype, labels*4);
1104 if (!nla)
1105 return -EMSGSIZE;
1106
1107 nla_label = nla_data(nla);
1108 bos = true;
1109 for (i = labels - 1; i >= 0; i--) {
1110 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1111 bos = false;
1112 }
1113
1114 return 0;
1115}
1116EXPORT_SYMBOL_GPL(nla_put_labels);
1117
1118int nla_get_labels(const struct nlattr *nla,
1119 u32 max_labels, u8 *labels, u32 label[])
1120{
1121 unsigned len = nla_len(nla);
1122 unsigned nla_labels;
1123 struct mpls_shim_hdr *nla_label;
1124 bool bos;
1125 int i;
1126
1127 /* len needs to be an even multiple of 4 (the label size) */
1128 if (len & 3)
1129 return -EINVAL;
1130
1131 /* Limit the number of new labels allowed */
1132 nla_labels = len/4;
1133 if (nla_labels > max_labels)
1134 return -EINVAL;
1135
1136 nla_label = nla_data(nla);
1137 bos = true;
1138 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1139 struct mpls_entry_decoded dec;
1140 dec = mpls_entry_decode(nla_label + i);
1141
1142 /* Ensure the bottom of stack flag is properly set
1143 * and ttl and tc are both clear.
1144 */
1145 if ((dec.bos != bos) || dec.ttl || dec.tc)
1146 return -EINVAL;
1147
1148 switch (dec.label) {
1149 case MPLS_LABEL_IMPLNULL:
1150 /* RFC3032: This is a label that an LSR may
1151 * assign and distribute, but which never
1152 * actually appears in the encapsulation.
1153 */
1154 return -EINVAL;
1155 }
1156
1157 label[i] = dec.label;
1158 }
1159 *labels = nla_labels;
1160 return 0;
1161}
1162EXPORT_SYMBOL_GPL(nla_get_labels);
1163
1164int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1165 u8 *via_table, u8 via_addr[])
1166{
1167 struct rtvia *via = nla_data(nla);
1168 int err = -EINVAL;
1169 int alen;
1170
1171 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1172 goto errout;
1173 alen = nla_len(nla) -
1174 offsetof(struct rtvia, rtvia_addr);
1175 if (alen > MAX_VIA_ALEN)
1176 goto errout;
1177
1178 /* Validate the address family */
1179 switch (via->rtvia_family) {
1180 case AF_PACKET:
1181 *via_table = NEIGH_LINK_TABLE;
1182 break;
1183 case AF_INET:
1184 *via_table = NEIGH_ARP_TABLE;
1185 if (alen != 4)
1186 goto errout;
1187 break;
1188 case AF_INET6:
1189 *via_table = NEIGH_ND_TABLE;
1190 if (alen != 16)
1191 goto errout;
1192 break;
1193 default:
1194 /* Unsupported address family */
1195 goto errout;
1196 }
1197
1198 memcpy(via_addr, via->rtvia_addr, alen);
1199 *via_alen = alen;
1200 err = 0;
1201
1202errout:
1203 return err;
1204}
1205
1206static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1207 struct mpls_route_config *cfg)
1208{
1209 struct rtmsg *rtm;
1210 struct nlattr *tb[RTA_MAX+1];
1211 int index;
1212 int err;
1213
1214 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1215 if (err < 0)
1216 goto errout;
1217
1218 err = -EINVAL;
1219 rtm = nlmsg_data(nlh);
1220 memset(cfg, 0, sizeof(*cfg));
1221
1222 if (rtm->rtm_family != AF_MPLS)
1223 goto errout;
1224 if (rtm->rtm_dst_len != 20)
1225 goto errout;
1226 if (rtm->rtm_src_len != 0)
1227 goto errout;
1228 if (rtm->rtm_tos != 0)
1229 goto errout;
1230 if (rtm->rtm_table != RT_TABLE_MAIN)
1231 goto errout;
1232 /* Any value is acceptable for rtm_protocol */
1233
1234 /* As mpls uses destination specific addresses
1235 * (or source specific address in the case of multicast)
1236 * all addresses have universal scope.
1237 */
1238 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1239 goto errout;
1240 if (rtm->rtm_type != RTN_UNICAST)
1241 goto errout;
1242 if (rtm->rtm_flags != 0)
1243 goto errout;
1244
1245 cfg->rc_label = LABEL_NOT_SPECIFIED;
1246 cfg->rc_protocol = rtm->rtm_protocol;
1247 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1248 cfg->rc_nlflags = nlh->nlmsg_flags;
1249 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1250 cfg->rc_nlinfo.nlh = nlh;
1251 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1252
1253 for (index = 0; index <= RTA_MAX; index++) {
1254 struct nlattr *nla = tb[index];
1255 if (!nla)
1256 continue;
1257
1258 switch (index) {
1259 case RTA_OIF:
1260 cfg->rc_ifindex = nla_get_u32(nla);
1261 break;
1262 case RTA_NEWDST:
1263 if (nla_get_labels(nla, MAX_NEW_LABELS,
1264 &cfg->rc_output_labels,
1265 cfg->rc_output_label))
1266 goto errout;
1267 break;
1268 case RTA_DST:
1269 {
1270 u8 label_count;
1271 if (nla_get_labels(nla, 1, &label_count,
1272 &cfg->rc_label))
1273 goto errout;
1274
1275 /* Reserved labels may not be set */
1276 if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1277 goto errout;
1278
1279 break;
1280 }
1281 case RTA_VIA:
1282 {
1283 if (nla_get_via(nla, &cfg->rc_via_alen,
1284 &cfg->rc_via_table, cfg->rc_via))
1285 goto errout;
1286 break;
1287 }
1288 case RTA_MULTIPATH:
1289 {
1290 cfg->rc_mp = nla_data(nla);
1291 cfg->rc_mp_len = nla_len(nla);
1292 break;
1293 }
1294 default:
1295 /* Unsupported attribute */
1296 goto errout;
1297 }
1298 }
1299
1300 err = 0;
1301errout:
1302 return err;
1303}
1304
1305static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1306{
1307 struct mpls_route_config cfg;
1308 int err;
1309
1310 err = rtm_to_route_config(skb, nlh, &cfg);
1311 if (err < 0)
1312 return err;
1313
1314 return mpls_route_del(&cfg);
1315}
1316
1317
1318static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1319{
1320 struct mpls_route_config cfg;
1321 int err;
1322
1323 err = rtm_to_route_config(skb, nlh, &cfg);
1324 if (err < 0)
1325 return err;
1326
1327 return mpls_route_add(&cfg);
1328}
1329
1330static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1331 u32 label, struct mpls_route *rt, int flags)
1332{
1333 struct net_device *dev;
1334 struct nlmsghdr *nlh;
1335 struct rtmsg *rtm;
1336
1337 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1338 if (nlh == NULL)
1339 return -EMSGSIZE;
1340
1341 rtm = nlmsg_data(nlh);
1342 rtm->rtm_family = AF_MPLS;
1343 rtm->rtm_dst_len = 20;
1344 rtm->rtm_src_len = 0;
1345 rtm->rtm_tos = 0;
1346 rtm->rtm_table = RT_TABLE_MAIN;
1347 rtm->rtm_protocol = rt->rt_protocol;
1348 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1349 rtm->rtm_type = RTN_UNICAST;
1350 rtm->rtm_flags = 0;
1351
1352 if (nla_put_labels(skb, RTA_DST, 1, &label))
1353 goto nla_put_failure;
1354 if (rt->rt_nhn == 1) {
1355 const struct mpls_nh *nh = rt->rt_nh;
1356
1357 if (nh->nh_labels &&
1358 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1359 nh->nh_label))
1360 goto nla_put_failure;
1361 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1362 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1363 nh->nh_via_alen))
1364 goto nla_put_failure;
1365 dev = rtnl_dereference(nh->nh_dev);
1366 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1367 goto nla_put_failure;
1368 if (nh->nh_flags & RTNH_F_LINKDOWN)
1369 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1370 if (nh->nh_flags & RTNH_F_DEAD)
1371 rtm->rtm_flags |= RTNH_F_DEAD;
1372 } else {
1373 struct rtnexthop *rtnh;
1374 struct nlattr *mp;
1375 int dead = 0;
1376 int linkdown = 0;
1377
1378 mp = nla_nest_start(skb, RTA_MULTIPATH);
1379 if (!mp)
1380 goto nla_put_failure;
1381
1382 for_nexthops(rt) {
1383 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1384 if (!rtnh)
1385 goto nla_put_failure;
1386
1387 dev = rtnl_dereference(nh->nh_dev);
1388 if (dev)
1389 rtnh->rtnh_ifindex = dev->ifindex;
1390 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1391 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1392 linkdown++;
1393 }
1394 if (nh->nh_flags & RTNH_F_DEAD) {
1395 rtnh->rtnh_flags |= RTNH_F_DEAD;
1396 dead++;
1397 }
1398
1399 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1400 nh->nh_labels,
1401 nh->nh_label))
1402 goto nla_put_failure;
1403 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1404 nla_put_via(skb, nh->nh_via_table,
1405 mpls_nh_via(rt, nh),
1406 nh->nh_via_alen))
1407 goto nla_put_failure;
1408
1409 /* length of rtnetlink header + attributes */
1410 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1411 } endfor_nexthops(rt);
1412
1413 if (linkdown == rt->rt_nhn)
1414 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1415 if (dead == rt->rt_nhn)
1416 rtm->rtm_flags |= RTNH_F_DEAD;
1417
1418 nla_nest_end(skb, mp);
1419 }
1420
1421 nlmsg_end(skb, nlh);
1422 return 0;
1423
1424nla_put_failure:
1425 nlmsg_cancel(skb, nlh);
1426 return -EMSGSIZE;
1427}
1428
1429static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1430{
1431 struct net *net = sock_net(skb->sk);
1432 struct mpls_route __rcu **platform_label;
1433 size_t platform_labels;
1434 unsigned int index;
1435
1436 ASSERT_RTNL();
1437
1438 index = cb->args[0];
1439 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1440 index = MPLS_LABEL_FIRST_UNRESERVED;
1441
1442 platform_label = rtnl_dereference(net->mpls.platform_label);
1443 platform_labels = net->mpls.platform_labels;
1444 for (; index < platform_labels; index++) {
1445 struct mpls_route *rt;
1446 rt = rtnl_dereference(platform_label[index]);
1447 if (!rt)
1448 continue;
1449
1450 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1451 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1452 index, rt, NLM_F_MULTI) < 0)
1453 break;
1454 }
1455 cb->args[0] = index;
1456
1457 return skb->len;
1458}
1459
1460static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1461{
1462 size_t payload =
1463 NLMSG_ALIGN(sizeof(struct rtmsg))
1464 + nla_total_size(4); /* RTA_DST */
1465
1466 if (rt->rt_nhn == 1) {
1467 struct mpls_nh *nh = rt->rt_nh;
1468
1469 if (nh->nh_dev)
1470 payload += nla_total_size(4); /* RTA_OIF */
1471 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1472 payload += nla_total_size(2 + nh->nh_via_alen);
1473 if (nh->nh_labels) /* RTA_NEWDST */
1474 payload += nla_total_size(nh->nh_labels * 4);
1475 } else {
1476 /* each nexthop is packed in an attribute */
1477 size_t nhsize = 0;
1478
1479 for_nexthops(rt) {
1480 nhsize += nla_total_size(sizeof(struct rtnexthop));
1481 /* RTA_VIA */
1482 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1483 nhsize += nla_total_size(2 + nh->nh_via_alen);
1484 if (nh->nh_labels)
1485 nhsize += nla_total_size(nh->nh_labels * 4);
1486 } endfor_nexthops(rt);
1487 /* nested attribute */
1488 payload += nla_total_size(nhsize);
1489 }
1490
1491 return payload;
1492}
1493
1494static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1495 struct nlmsghdr *nlh, struct net *net, u32 portid,
1496 unsigned int nlm_flags)
1497{
1498 struct sk_buff *skb;
1499 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1500 int err = -ENOBUFS;
1501
1502 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1503 if (skb == NULL)
1504 goto errout;
1505
1506 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1507 if (err < 0) {
1508 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1509 WARN_ON(err == -EMSGSIZE);
1510 kfree_skb(skb);
1511 goto errout;
1512 }
1513 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1514
1515 return;
1516errout:
1517 if (err < 0)
1518 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1519}
1520
1521static int resize_platform_label_table(struct net *net, size_t limit)
1522{
1523 size_t size = sizeof(struct mpls_route *) * limit;
1524 size_t old_limit;
1525 size_t cp_size;
1526 struct mpls_route __rcu **labels = NULL, **old;
1527 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1528 unsigned index;
1529
1530 if (size) {
1531 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1532 if (!labels)
1533 labels = vzalloc(size);
1534
1535 if (!labels)
1536 goto nolabels;
1537 }
1538
1539 /* In case the predefined labels need to be populated */
1540 if (limit > MPLS_LABEL_IPV4NULL) {
1541 struct net_device *lo = net->loopback_dev;
1542 rt0 = mpls_rt_alloc(1, lo->addr_len);
1543 if (!rt0)
1544 goto nort0;
1545 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1546 rt0->rt_protocol = RTPROT_KERNEL;
1547 rt0->rt_payload_type = MPT_IPV4;
1548 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1549 rt0->rt_nh->nh_via_alen = lo->addr_len;
1550 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1551 lo->addr_len);
1552 }
1553 if (limit > MPLS_LABEL_IPV6NULL) {
1554 struct net_device *lo = net->loopback_dev;
1555 rt2 = mpls_rt_alloc(1, lo->addr_len);
1556 if (!rt2)
1557 goto nort2;
1558 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1559 rt2->rt_protocol = RTPROT_KERNEL;
1560 rt2->rt_payload_type = MPT_IPV6;
1561 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1562 rt2->rt_nh->nh_via_alen = lo->addr_len;
1563 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1564 lo->addr_len);
1565 }
1566
1567 rtnl_lock();
1568 /* Remember the original table */
1569 old = rtnl_dereference(net->mpls.platform_label);
1570 old_limit = net->mpls.platform_labels;
1571
1572 /* Free any labels beyond the new table */
1573 for (index = limit; index < old_limit; index++)
1574 mpls_route_update(net, index, NULL, NULL);
1575
1576 /* Copy over the old labels */
1577 cp_size = size;
1578 if (old_limit < limit)
1579 cp_size = old_limit * sizeof(struct mpls_route *);
1580
1581 memcpy(labels, old, cp_size);
1582
1583 /* If needed set the predefined labels */
1584 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1585 (limit > MPLS_LABEL_IPV6NULL)) {
1586 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1587 rt2 = NULL;
1588 }
1589
1590 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1591 (limit > MPLS_LABEL_IPV4NULL)) {
1592 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1593 rt0 = NULL;
1594 }
1595
1596 /* Update the global pointers */
1597 net->mpls.platform_labels = limit;
1598 rcu_assign_pointer(net->mpls.platform_label, labels);
1599
1600 rtnl_unlock();
1601
1602 mpls_rt_free(rt2);
1603 mpls_rt_free(rt0);
1604
1605 if (old) {
1606 synchronize_rcu();
1607 kvfree(old);
1608 }
1609 return 0;
1610
1611nort2:
1612 mpls_rt_free(rt0);
1613nort0:
1614 kvfree(labels);
1615nolabels:
1616 return -ENOMEM;
1617}
1618
1619static int mpls_platform_labels(struct ctl_table *table, int write,
1620 void __user *buffer, size_t *lenp, loff_t *ppos)
1621{
1622 struct net *net = table->data;
1623 int platform_labels = net->mpls.platform_labels;
1624 int ret;
1625 struct ctl_table tmp = {
1626 .procname = table->procname,
1627 .data = &platform_labels,
1628 .maxlen = sizeof(int),
1629 .mode = table->mode,
1630 .extra1 = &zero,
1631 .extra2 = &label_limit,
1632 };
1633
1634 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1635
1636 if (write && ret == 0)
1637 ret = resize_platform_label_table(net, platform_labels);
1638
1639 return ret;
1640}
1641
1642static const struct ctl_table mpls_table[] = {
1643 {
1644 .procname = "platform_labels",
1645 .data = NULL,
1646 .maxlen = sizeof(int),
1647 .mode = 0644,
1648 .proc_handler = mpls_platform_labels,
1649 },
1650 { }
1651};
1652
1653static int mpls_net_init(struct net *net)
1654{
1655 struct ctl_table *table;
1656
1657 net->mpls.platform_labels = 0;
1658 net->mpls.platform_label = NULL;
1659
1660 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1661 if (table == NULL)
1662 return -ENOMEM;
1663
1664 table[0].data = net;
1665 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1666 if (net->mpls.ctl == NULL) {
1667 kfree(table);
1668 return -ENOMEM;
1669 }
1670
1671 return 0;
1672}
1673
1674static void mpls_net_exit(struct net *net)
1675{
1676 struct mpls_route __rcu **platform_label;
1677 size_t platform_labels;
1678 struct ctl_table *table;
1679 unsigned int index;
1680
1681 table = net->mpls.ctl->ctl_table_arg;
1682 unregister_net_sysctl_table(net->mpls.ctl);
1683 kfree(table);
1684
1685 /* An rcu grace period has passed since there was a device in
1686 * the network namespace (and thus the last in flight packet)
1687 * left this network namespace. This is because
1688 * unregister_netdevice_many and netdev_run_todo has completed
1689 * for each network device that was in this network namespace.
1690 *
1691 * As such no additional rcu synchronization is necessary when
1692 * freeing the platform_label table.
1693 */
1694 rtnl_lock();
1695 platform_label = rtnl_dereference(net->mpls.platform_label);
1696 platform_labels = net->mpls.platform_labels;
1697 for (index = 0; index < platform_labels; index++) {
1698 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1699 RCU_INIT_POINTER(platform_label[index], NULL);
1700 mpls_notify_route(net, index, rt, NULL, NULL);
1701 mpls_rt_free(rt);
1702 }
1703 rtnl_unlock();
1704
1705 kvfree(platform_label);
1706}
1707
1708static struct pernet_operations mpls_net_ops = {
1709 .init = mpls_net_init,
1710 .exit = mpls_net_exit,
1711};
1712
1713static int __init mpls_init(void)
1714{
1715 int err;
1716
1717 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1718
1719 err = register_pernet_subsys(&mpls_net_ops);
1720 if (err)
1721 goto out;
1722
1723 err = register_netdevice_notifier(&mpls_dev_notifier);
1724 if (err)
1725 goto out_unregister_pernet;
1726
1727 dev_add_pack(&mpls_packet_type);
1728
1729 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1730 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1731 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1732 err = 0;
1733out:
1734 return err;
1735
1736out_unregister_pernet:
1737 unregister_pernet_subsys(&mpls_net_ops);
1738 goto out;
1739}
1740module_init(mpls_init);
1741
1742static void __exit mpls_exit(void)
1743{
1744 rtnl_unregister_all(PF_MPLS);
1745 dev_remove_pack(&mpls_packet_type);
1746 unregister_netdevice_notifier(&mpls_dev_notifier);
1747 unregister_pernet_subsys(&mpls_net_ops);
1748}
1749module_exit(mpls_exit);
1750
1751MODULE_DESCRIPTION("MultiProtocol Label Switching");
1752MODULE_LICENSE("GPL v2");
1753MODULE_ALIAS_NETPROTO(PF_MPLS);