projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile
[deliverable/linux.git] / net / mpls / af_mpls.c
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
32 static int zero = 0;
33 static int label_limit = (1 << 20) - 1;
34
35 static 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
39 static 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
51 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
52 {
53 return rcu_dereference_rtnl(dev->mpls_ptr);
54 }
55
56 bool mpls_output_possible(const struct net_device *dev)
57 {
58 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
59 }
60 EXPORT_SYMBOL_GPL(mpls_output_possible);
61
62 static 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
70 static 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
76 static 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
82 unsigned 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 }
87 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
88
89 bool 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 }
99 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
100
101 static 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
166 static 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
195 out:
196 return &rt->rt_nh[nh_index];
197 }
198
199 static 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
247 static 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
358 drop:
359 kfree_skb(skb);
360 return NET_RX_DROP;
361 }
362
363 static struct packet_type mpls_packet_type __read_mostly = {
364 .type = cpu_to_be16(ETH_P_MPLS_UC),
365 .func = mpls_forward,
366 };
367
368 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
369 [RTA_DST] = { .type = NLA_U32 },
370 [RTA_OIF] = { .type = NLA_U32 },
371 };
372
373 struct 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
389 static 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
407 static void mpls_rt_free(struct mpls_route *rt)
408 {
409 if (rt)
410 kfree_rcu(rt, rt_rcu);
411 }
412
413 static 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
427 static 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
446 static 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)
463 static 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
483 static 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)
491 static 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
515 static struct net_device *inet6_fib_lookup_dev(struct net *net,
516 const void *addr)
517 {
518 return ERR_PTR(-EAFNOSUPPORT);
519 }
520 #endif
521
522 static 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 /* The caller is holding rtnl anyways, so release the dev reference */
547 dev_put(dev);
548
549 return dev;
550 }
551
552 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
553 struct mpls_nh *nh, int oif)
554 {
555 struct net_device *dev = NULL;
556 int err = -ENODEV;
557
558 dev = find_outdev(net, rt, nh, oif);
559 if (IS_ERR(dev)) {
560 err = PTR_ERR(dev);
561 dev = NULL;
562 goto errout;
563 }
564
565 /* Ensure this is a supported device */
566 err = -EINVAL;
567 if (!mpls_dev_get(dev))
568 goto errout;
569
570 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
571 (dev->addr_len != nh->nh_via_alen))
572 goto errout;
573
574 RCU_INIT_POINTER(nh->nh_dev, dev);
575
576 if (!(dev->flags & IFF_UP)) {
577 nh->nh_flags |= RTNH_F_DEAD;
578 } else {
579 unsigned int flags;
580
581 flags = dev_get_flags(dev);
582 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
583 nh->nh_flags |= RTNH_F_LINKDOWN;
584 }
585
586 return 0;
587
588 errout:
589 return err;
590 }
591
592 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
593 struct mpls_route *rt)
594 {
595 struct net *net = cfg->rc_nlinfo.nl_net;
596 struct mpls_nh *nh = rt->rt_nh;
597 int err;
598 int i;
599
600 if (!nh)
601 return -ENOMEM;
602
603 err = -EINVAL;
604 /* Ensure only a supported number of labels are present */
605 if (cfg->rc_output_labels > MAX_NEW_LABELS)
606 goto errout;
607
608 nh->nh_labels = cfg->rc_output_labels;
609 for (i = 0; i < nh->nh_labels; i++)
610 nh->nh_label[i] = cfg->rc_output_label[i];
611
612 nh->nh_via_table = cfg->rc_via_table;
613 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
614 nh->nh_via_alen = cfg->rc_via_alen;
615
616 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
617 if (err)
618 goto errout;
619
620 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
621 rt->rt_nhn_alive--;
622
623 return 0;
624
625 errout:
626 return err;
627 }
628
629 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
630 struct mpls_nh *nh, int oif, struct nlattr *via,
631 struct nlattr *newdst)
632 {
633 int err = -ENOMEM;
634
635 if (!nh)
636 goto errout;
637
638 if (newdst) {
639 err = nla_get_labels(newdst, MAX_NEW_LABELS,
640 &nh->nh_labels, nh->nh_label);
641 if (err)
642 goto errout;
643 }
644
645 if (via) {
646 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
647 __mpls_nh_via(rt, nh));
648 if (err)
649 goto errout;
650 } else {
651 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
652 }
653
654 err = mpls_nh_assign_dev(net, rt, nh, oif);
655 if (err)
656 goto errout;
657
658 return 0;
659
660 errout:
661 return err;
662 }
663
664 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
665 u8 cfg_via_alen, u8 *max_via_alen)
666 {
667 int nhs = 0;
668 int remaining = len;
669
670 if (!rtnh) {
671 *max_via_alen = cfg_via_alen;
672 return 1;
673 }
674
675 *max_via_alen = 0;
676
677 while (rtnh_ok(rtnh, remaining)) {
678 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
679 int attrlen;
680
681 attrlen = rtnh_attrlen(rtnh);
682 nla = nla_find(attrs, attrlen, RTA_VIA);
683 if (nla && nla_len(nla) >=
684 offsetof(struct rtvia, rtvia_addr)) {
685 int via_alen = nla_len(nla) -
686 offsetof(struct rtvia, rtvia_addr);
687
688 if (via_alen <= MAX_VIA_ALEN)
689 *max_via_alen = max_t(u16, *max_via_alen,
690 via_alen);
691 }
692
693 nhs++;
694 rtnh = rtnh_next(rtnh, &remaining);
695 }
696
697 /* leftover implies invalid nexthop configuration, discard it */
698 return remaining > 0 ? 0 : nhs;
699 }
700
701 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
702 struct mpls_route *rt)
703 {
704 struct rtnexthop *rtnh = cfg->rc_mp;
705 struct nlattr *nla_via, *nla_newdst;
706 int remaining = cfg->rc_mp_len;
707 int nhs = 0;
708 int err = 0;
709
710 change_nexthops(rt) {
711 int attrlen;
712
713 nla_via = NULL;
714 nla_newdst = NULL;
715
716 err = -EINVAL;
717 if (!rtnh_ok(rtnh, remaining))
718 goto errout;
719
720 /* neither weighted multipath nor any flags
721 * are supported
722 */
723 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
724 goto errout;
725
726 attrlen = rtnh_attrlen(rtnh);
727 if (attrlen > 0) {
728 struct nlattr *attrs = rtnh_attrs(rtnh);
729
730 nla_via = nla_find(attrs, attrlen, RTA_VIA);
731 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
732 }
733
734 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
735 rtnh->rtnh_ifindex, nla_via, nla_newdst);
736 if (err)
737 goto errout;
738
739 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
740 rt->rt_nhn_alive--;
741
742 rtnh = rtnh_next(rtnh, &remaining);
743 nhs++;
744 } endfor_nexthops(rt);
745
746 rt->rt_nhn = nhs;
747
748 return 0;
749
750 errout:
751 return err;
752 }
753
754 static int mpls_route_add(struct mpls_route_config *cfg)
755 {
756 struct mpls_route __rcu **platform_label;
757 struct net *net = cfg->rc_nlinfo.nl_net;
758 struct mpls_route *rt, *old;
759 int err = -EINVAL;
760 u8 max_via_alen;
761 unsigned index;
762 int nhs;
763
764 index = cfg->rc_label;
765
766 /* If a label was not specified during insert pick one */
767 if ((index == LABEL_NOT_SPECIFIED) &&
768 (cfg->rc_nlflags & NLM_F_CREATE)) {
769 index = find_free_label(net);
770 }
771
772 /* Reserved labels may not be set */
773 if (index < MPLS_LABEL_FIRST_UNRESERVED)
774 goto errout;
775
776 /* The full 20 bit range may not be supported. */
777 if (index >= net->mpls.platform_labels)
778 goto errout;
779
780 /* Append makes no sense with mpls */
781 err = -EOPNOTSUPP;
782 if (cfg->rc_nlflags & NLM_F_APPEND)
783 goto errout;
784
785 err = -EEXIST;
786 platform_label = rtnl_dereference(net->mpls.platform_label);
787 old = rtnl_dereference(platform_label[index]);
788 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
789 goto errout;
790
791 err = -EEXIST;
792 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
793 goto errout;
794
795 err = -ENOENT;
796 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
797 goto errout;
798
799 err = -EINVAL;
800 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
801 cfg->rc_via_alen, &max_via_alen);
802 if (nhs == 0)
803 goto errout;
804
805 err = -ENOMEM;
806 rt = mpls_rt_alloc(nhs, max_via_alen);
807 if (!rt)
808 goto errout;
809
810 rt->rt_protocol = cfg->rc_protocol;
811 rt->rt_payload_type = cfg->rc_payload_type;
812
813 if (cfg->rc_mp)
814 err = mpls_nh_build_multi(cfg, rt);
815 else
816 err = mpls_nh_build_from_cfg(cfg, rt);
817 if (err)
818 goto freert;
819
820 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
821
822 return 0;
823
824 freert:
825 mpls_rt_free(rt);
826 errout:
827 return err;
828 }
829
830 static int mpls_route_del(struct mpls_route_config *cfg)
831 {
832 struct net *net = cfg->rc_nlinfo.nl_net;
833 unsigned index;
834 int err = -EINVAL;
835
836 index = cfg->rc_label;
837
838 /* Reserved labels may not be removed */
839 if (index < MPLS_LABEL_FIRST_UNRESERVED)
840 goto errout;
841
842 /* The full 20 bit range may not be supported */
843 if (index >= net->mpls.platform_labels)
844 goto errout;
845
846 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
847
848 err = 0;
849 errout:
850 return err;
851 }
852
853 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
854 (&((struct mpls_dev *)0)->field)
855
856 static const struct ctl_table mpls_dev_table[] = {
857 {
858 .procname = "input",
859 .maxlen = sizeof(int),
860 .mode = 0644,
861 .proc_handler = proc_dointvec,
862 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
863 },
864 { }
865 };
866
867 static int mpls_dev_sysctl_register(struct net_device *dev,
868 struct mpls_dev *mdev)
869 {
870 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
871 struct ctl_table *table;
872 int i;
873
874 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
875 if (!table)
876 goto out;
877
878 /* Table data contains only offsets relative to the base of
879 * the mdev at this point, so make them absolute.
880 */
881 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
882 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
883
884 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
885
886 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
887 if (!mdev->sysctl)
888 goto free;
889
890 return 0;
891
892 free:
893 kfree(table);
894 out:
895 return -ENOBUFS;
896 }
897
898 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
899 {
900 struct ctl_table *table;
901
902 table = mdev->sysctl->ctl_table_arg;
903 unregister_net_sysctl_table(mdev->sysctl);
904 kfree(table);
905 }
906
907 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
908 {
909 struct mpls_dev *mdev;
910 int err = -ENOMEM;
911
912 ASSERT_RTNL();
913
914 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
915 if (!mdev)
916 return ERR_PTR(err);
917
918 err = mpls_dev_sysctl_register(dev, mdev);
919 if (err)
920 goto free;
921
922 rcu_assign_pointer(dev->mpls_ptr, mdev);
923
924 return mdev;
925
926 free:
927 kfree(mdev);
928 return ERR_PTR(err);
929 }
930
931 static void mpls_ifdown(struct net_device *dev, int event)
932 {
933 struct mpls_route __rcu **platform_label;
934 struct net *net = dev_net(dev);
935 unsigned index;
936
937 platform_label = rtnl_dereference(net->mpls.platform_label);
938 for (index = 0; index < net->mpls.platform_labels; index++) {
939 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
940
941 if (!rt)
942 continue;
943
944 change_nexthops(rt) {
945 if (rtnl_dereference(nh->nh_dev) != dev)
946 continue;
947 switch (event) {
948 case NETDEV_DOWN:
949 case NETDEV_UNREGISTER:
950 nh->nh_flags |= RTNH_F_DEAD;
951 /* fall through */
952 case NETDEV_CHANGE:
953 nh->nh_flags |= RTNH_F_LINKDOWN;
954 ACCESS_ONCE(rt->rt_nhn_alive) = rt->rt_nhn_alive - 1;
955 break;
956 }
957 if (event == NETDEV_UNREGISTER)
958 RCU_INIT_POINTER(nh->nh_dev, NULL);
959 } endfor_nexthops(rt);
960 }
961
962
963 return;
964 }
965
966 static void mpls_ifup(struct net_device *dev, unsigned int nh_flags)
967 {
968 struct mpls_route __rcu **platform_label;
969 struct net *net = dev_net(dev);
970 unsigned index;
971 int alive;
972
973 platform_label = rtnl_dereference(net->mpls.platform_label);
974 for (index = 0; index < net->mpls.platform_labels; index++) {
975 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
976
977 if (!rt)
978 continue;
979
980 alive = 0;
981 change_nexthops(rt) {
982 struct net_device *nh_dev =
983 rtnl_dereference(nh->nh_dev);
984
985 if (!(nh->nh_flags & nh_flags)) {
986 alive++;
987 continue;
988 }
989 if (nh_dev != dev)
990 continue;
991 alive++;
992 nh->nh_flags &= ~nh_flags;
993 } endfor_nexthops(rt);
994
995 ACCESS_ONCE(rt->rt_nhn_alive) = alive;
996 }
997
998 return;
999 }
1000
1001 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1002 void *ptr)
1003 {
1004 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1005 struct mpls_dev *mdev;
1006 unsigned int flags;
1007
1008 if (event == NETDEV_REGISTER) {
1009 /* For now just support ethernet devices */
1010 if ((dev->type == ARPHRD_ETHER) ||
1011 (dev->type == ARPHRD_LOOPBACK)) {
1012 mdev = mpls_add_dev(dev);
1013 if (IS_ERR(mdev))
1014 return notifier_from_errno(PTR_ERR(mdev));
1015 }
1016 return NOTIFY_OK;
1017 }
1018
1019 mdev = mpls_dev_get(dev);
1020 if (!mdev)
1021 return NOTIFY_OK;
1022
1023 switch (event) {
1024 case NETDEV_DOWN:
1025 mpls_ifdown(dev, event);
1026 break;
1027 case NETDEV_UP:
1028 flags = dev_get_flags(dev);
1029 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1030 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1031 else
1032 mpls_ifup(dev, RTNH_F_DEAD);
1033 break;
1034 case NETDEV_CHANGE:
1035 flags = dev_get_flags(dev);
1036 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1037 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1038 else
1039 mpls_ifdown(dev, event);
1040 break;
1041 case NETDEV_UNREGISTER:
1042 mpls_ifdown(dev, event);
1043 mdev = mpls_dev_get(dev);
1044 if (mdev) {
1045 mpls_dev_sysctl_unregister(mdev);
1046 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1047 kfree_rcu(mdev, rcu);
1048 }
1049 break;
1050 case NETDEV_CHANGENAME:
1051 mdev = mpls_dev_get(dev);
1052 if (mdev) {
1053 int err;
1054
1055 mpls_dev_sysctl_unregister(mdev);
1056 err = mpls_dev_sysctl_register(dev, mdev);
1057 if (err)
1058 return notifier_from_errno(err);
1059 }
1060 break;
1061 }
1062 return NOTIFY_OK;
1063 }
1064
1065 static struct notifier_block mpls_dev_notifier = {
1066 .notifier_call = mpls_dev_notify,
1067 };
1068
1069 static int nla_put_via(struct sk_buff *skb,
1070 u8 table, const void *addr, int alen)
1071 {
1072 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1073 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1074 };
1075 struct nlattr *nla;
1076 struct rtvia *via;
1077 int family = AF_UNSPEC;
1078
1079 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1080 if (!nla)
1081 return -EMSGSIZE;
1082
1083 if (table <= NEIGH_NR_TABLES)
1084 family = table_to_family[table];
1085
1086 via = nla_data(nla);
1087 via->rtvia_family = family;
1088 memcpy(via->rtvia_addr, addr, alen);
1089 return 0;
1090 }
1091
1092 int nla_put_labels(struct sk_buff *skb, int attrtype,
1093 u8 labels, const u32 label[])
1094 {
1095 struct nlattr *nla;
1096 struct mpls_shim_hdr *nla_label;
1097 bool bos;
1098 int i;
1099 nla = nla_reserve(skb, attrtype, labels*4);
1100 if (!nla)
1101 return -EMSGSIZE;
1102
1103 nla_label = nla_data(nla);
1104 bos = true;
1105 for (i = labels - 1; i >= 0; i--) {
1106 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1107 bos = false;
1108 }
1109
1110 return 0;
1111 }
1112 EXPORT_SYMBOL_GPL(nla_put_labels);
1113
1114 int nla_get_labels(const struct nlattr *nla,
1115 u32 max_labels, u8 *labels, u32 label[])
1116 {
1117 unsigned len = nla_len(nla);
1118 unsigned nla_labels;
1119 struct mpls_shim_hdr *nla_label;
1120 bool bos;
1121 int i;
1122
1123 /* len needs to be an even multiple of 4 (the label size) */
1124 if (len & 3)
1125 return -EINVAL;
1126
1127 /* Limit the number of new labels allowed */
1128 nla_labels = len/4;
1129 if (nla_labels > max_labels)
1130 return -EINVAL;
1131
1132 nla_label = nla_data(nla);
1133 bos = true;
1134 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1135 struct mpls_entry_decoded dec;
1136 dec = mpls_entry_decode(nla_label + i);
1137
1138 /* Ensure the bottom of stack flag is properly set
1139 * and ttl and tc are both clear.
1140 */
1141 if ((dec.bos != bos) || dec.ttl || dec.tc)
1142 return -EINVAL;
1143
1144 switch (dec.label) {
1145 case MPLS_LABEL_IMPLNULL:
1146 /* RFC3032: This is a label that an LSR may
1147 * assign and distribute, but which never
1148 * actually appears in the encapsulation.
1149 */
1150 return -EINVAL;
1151 }
1152
1153 label[i] = dec.label;
1154 }
1155 *labels = nla_labels;
1156 return 0;
1157 }
1158 EXPORT_SYMBOL_GPL(nla_get_labels);
1159
1160 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1161 u8 *via_table, u8 via_addr[])
1162 {
1163 struct rtvia *via = nla_data(nla);
1164 int err = -EINVAL;
1165 int alen;
1166
1167 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1168 goto errout;
1169 alen = nla_len(nla) -
1170 offsetof(struct rtvia, rtvia_addr);
1171 if (alen > MAX_VIA_ALEN)
1172 goto errout;
1173
1174 /* Validate the address family */
1175 switch (via->rtvia_family) {
1176 case AF_PACKET:
1177 *via_table = NEIGH_LINK_TABLE;
1178 break;
1179 case AF_INET:
1180 *via_table = NEIGH_ARP_TABLE;
1181 if (alen != 4)
1182 goto errout;
1183 break;
1184 case AF_INET6:
1185 *via_table = NEIGH_ND_TABLE;
1186 if (alen != 16)
1187 goto errout;
1188 break;
1189 default:
1190 /* Unsupported address family */
1191 goto errout;
1192 }
1193
1194 memcpy(via_addr, via->rtvia_addr, alen);
1195 *via_alen = alen;
1196 err = 0;
1197
1198 errout:
1199 return err;
1200 }
1201
1202 static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1203 struct mpls_route_config *cfg)
1204 {
1205 struct rtmsg *rtm;
1206 struct nlattr *tb[RTA_MAX+1];
1207 int index;
1208 int err;
1209
1210 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1211 if (err < 0)
1212 goto errout;
1213
1214 err = -EINVAL;
1215 rtm = nlmsg_data(nlh);
1216 memset(cfg, 0, sizeof(*cfg));
1217
1218 if (rtm->rtm_family != AF_MPLS)
1219 goto errout;
1220 if (rtm->rtm_dst_len != 20)
1221 goto errout;
1222 if (rtm->rtm_src_len != 0)
1223 goto errout;
1224 if (rtm->rtm_tos != 0)
1225 goto errout;
1226 if (rtm->rtm_table != RT_TABLE_MAIN)
1227 goto errout;
1228 /* Any value is acceptable for rtm_protocol */
1229
1230 /* As mpls uses destination specific addresses
1231 * (or source specific address in the case of multicast)
1232 * all addresses have universal scope.
1233 */
1234 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1235 goto errout;
1236 if (rtm->rtm_type != RTN_UNICAST)
1237 goto errout;
1238 if (rtm->rtm_flags != 0)
1239 goto errout;
1240
1241 cfg->rc_label = LABEL_NOT_SPECIFIED;
1242 cfg->rc_protocol = rtm->rtm_protocol;
1243 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1244 cfg->rc_nlflags = nlh->nlmsg_flags;
1245 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1246 cfg->rc_nlinfo.nlh = nlh;
1247 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1248
1249 for (index = 0; index <= RTA_MAX; index++) {
1250 struct nlattr *nla = tb[index];
1251 if (!nla)
1252 continue;
1253
1254 switch(index) {
1255 case RTA_OIF:
1256 cfg->rc_ifindex = nla_get_u32(nla);
1257 break;
1258 case RTA_NEWDST:
1259 if (nla_get_labels(nla, MAX_NEW_LABELS,
1260 &cfg->rc_output_labels,
1261 cfg->rc_output_label))
1262 goto errout;
1263 break;
1264 case RTA_DST:
1265 {
1266 u8 label_count;
1267 if (nla_get_labels(nla, 1, &label_count,
1268 &cfg->rc_label))
1269 goto errout;
1270
1271 /* Reserved labels may not be set */
1272 if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1273 goto errout;
1274
1275 break;
1276 }
1277 case RTA_VIA:
1278 {
1279 if (nla_get_via(nla, &cfg->rc_via_alen,
1280 &cfg->rc_via_table, cfg->rc_via))
1281 goto errout;
1282 break;
1283 }
1284 case RTA_MULTIPATH:
1285 {
1286 cfg->rc_mp = nla_data(nla);
1287 cfg->rc_mp_len = nla_len(nla);
1288 break;
1289 }
1290 default:
1291 /* Unsupported attribute */
1292 goto errout;
1293 }
1294 }
1295
1296 err = 0;
1297 errout:
1298 return err;
1299 }
1300
1301 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1302 {
1303 struct mpls_route_config cfg;
1304 int err;
1305
1306 err = rtm_to_route_config(skb, nlh, &cfg);
1307 if (err < 0)
1308 return err;
1309
1310 return mpls_route_del(&cfg);
1311 }
1312
1313
1314 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1315 {
1316 struct mpls_route_config cfg;
1317 int err;
1318
1319 err = rtm_to_route_config(skb, nlh, &cfg);
1320 if (err < 0)
1321 return err;
1322
1323 return mpls_route_add(&cfg);
1324 }
1325
1326 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1327 u32 label, struct mpls_route *rt, int flags)
1328 {
1329 struct net_device *dev;
1330 struct nlmsghdr *nlh;
1331 struct rtmsg *rtm;
1332
1333 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1334 if (nlh == NULL)
1335 return -EMSGSIZE;
1336
1337 rtm = nlmsg_data(nlh);
1338 rtm->rtm_family = AF_MPLS;
1339 rtm->rtm_dst_len = 20;
1340 rtm->rtm_src_len = 0;
1341 rtm->rtm_tos = 0;
1342 rtm->rtm_table = RT_TABLE_MAIN;
1343 rtm->rtm_protocol = rt->rt_protocol;
1344 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1345 rtm->rtm_type = RTN_UNICAST;
1346 rtm->rtm_flags = 0;
1347
1348 if (nla_put_labels(skb, RTA_DST, 1, &label))
1349 goto nla_put_failure;
1350 if (rt->rt_nhn == 1) {
1351 const struct mpls_nh *nh = rt->rt_nh;
1352
1353 if (nh->nh_labels &&
1354 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1355 nh->nh_label))
1356 goto nla_put_failure;
1357 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1358 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1359 nh->nh_via_alen))
1360 goto nla_put_failure;
1361 dev = rtnl_dereference(nh->nh_dev);
1362 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1363 goto nla_put_failure;
1364 if (nh->nh_flags & RTNH_F_LINKDOWN)
1365 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1366 if (nh->nh_flags & RTNH_F_DEAD)
1367 rtm->rtm_flags |= RTNH_F_DEAD;
1368 } else {
1369 struct rtnexthop *rtnh;
1370 struct nlattr *mp;
1371 int dead = 0;
1372 int linkdown = 0;
1373
1374 mp = nla_nest_start(skb, RTA_MULTIPATH);
1375 if (!mp)
1376 goto nla_put_failure;
1377
1378 for_nexthops(rt) {
1379 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1380 if (!rtnh)
1381 goto nla_put_failure;
1382
1383 dev = rtnl_dereference(nh->nh_dev);
1384 if (dev)
1385 rtnh->rtnh_ifindex = dev->ifindex;
1386 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1387 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1388 linkdown++;
1389 }
1390 if (nh->nh_flags & RTNH_F_DEAD) {
1391 rtnh->rtnh_flags |= RTNH_F_DEAD;
1392 dead++;
1393 }
1394
1395 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1396 nh->nh_labels,
1397 nh->nh_label))
1398 goto nla_put_failure;
1399 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1400 nla_put_via(skb, nh->nh_via_table,
1401 mpls_nh_via(rt, nh),
1402 nh->nh_via_alen))
1403 goto nla_put_failure;
1404
1405 /* length of rtnetlink header + attributes */
1406 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1407 } endfor_nexthops(rt);
1408
1409 if (linkdown == rt->rt_nhn)
1410 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1411 if (dead == rt->rt_nhn)
1412 rtm->rtm_flags |= RTNH_F_DEAD;
1413
1414 nla_nest_end(skb, mp);
1415 }
1416
1417 nlmsg_end(skb, nlh);
1418 return 0;
1419
1420 nla_put_failure:
1421 nlmsg_cancel(skb, nlh);
1422 return -EMSGSIZE;
1423 }
1424
1425 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1426 {
1427 struct net *net = sock_net(skb->sk);
1428 struct mpls_route __rcu **platform_label;
1429 size_t platform_labels;
1430 unsigned int index;
1431
1432 ASSERT_RTNL();
1433
1434 index = cb->args[0];
1435 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1436 index = MPLS_LABEL_FIRST_UNRESERVED;
1437
1438 platform_label = rtnl_dereference(net->mpls.platform_label);
1439 platform_labels = net->mpls.platform_labels;
1440 for (; index < platform_labels; index++) {
1441 struct mpls_route *rt;
1442 rt = rtnl_dereference(platform_label[index]);
1443 if (!rt)
1444 continue;
1445
1446 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1447 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1448 index, rt, NLM_F_MULTI) < 0)
1449 break;
1450 }
1451 cb->args[0] = index;
1452
1453 return skb->len;
1454 }
1455
1456 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1457 {
1458 size_t payload =
1459 NLMSG_ALIGN(sizeof(struct rtmsg))
1460 + nla_total_size(4); /* RTA_DST */
1461
1462 if (rt->rt_nhn == 1) {
1463 struct mpls_nh *nh = rt->rt_nh;
1464
1465 if (nh->nh_dev)
1466 payload += nla_total_size(4); /* RTA_OIF */
1467 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1468 payload += nla_total_size(2 + nh->nh_via_alen);
1469 if (nh->nh_labels) /* RTA_NEWDST */
1470 payload += nla_total_size(nh->nh_labels * 4);
1471 } else {
1472 /* each nexthop is packed in an attribute */
1473 size_t nhsize = 0;
1474
1475 for_nexthops(rt) {
1476 nhsize += nla_total_size(sizeof(struct rtnexthop));
1477 /* RTA_VIA */
1478 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1479 nhsize += nla_total_size(2 + nh->nh_via_alen);
1480 if (nh->nh_labels)
1481 nhsize += nla_total_size(nh->nh_labels * 4);
1482 } endfor_nexthops(rt);
1483 /* nested attribute */
1484 payload += nla_total_size(nhsize);
1485 }
1486
1487 return payload;
1488 }
1489
1490 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1491 struct nlmsghdr *nlh, struct net *net, u32 portid,
1492 unsigned int nlm_flags)
1493 {
1494 struct sk_buff *skb;
1495 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1496 int err = -ENOBUFS;
1497
1498 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1499 if (skb == NULL)
1500 goto errout;
1501
1502 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1503 if (err < 0) {
1504 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1505 WARN_ON(err == -EMSGSIZE);
1506 kfree_skb(skb);
1507 goto errout;
1508 }
1509 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1510
1511 return;
1512 errout:
1513 if (err < 0)
1514 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1515 }
1516
1517 static int resize_platform_label_table(struct net *net, size_t limit)
1518 {
1519 size_t size = sizeof(struct mpls_route *) * limit;
1520 size_t old_limit;
1521 size_t cp_size;
1522 struct mpls_route __rcu **labels = NULL, **old;
1523 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1524 unsigned index;
1525
1526 if (size) {
1527 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1528 if (!labels)
1529 labels = vzalloc(size);
1530
1531 if (!labels)
1532 goto nolabels;
1533 }
1534
1535 /* In case the predefined labels need to be populated */
1536 if (limit > MPLS_LABEL_IPV4NULL) {
1537 struct net_device *lo = net->loopback_dev;
1538 rt0 = mpls_rt_alloc(1, lo->addr_len);
1539 if (!rt0)
1540 goto nort0;
1541 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1542 rt0->rt_protocol = RTPROT_KERNEL;
1543 rt0->rt_payload_type = MPT_IPV4;
1544 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1545 rt0->rt_nh->nh_via_alen = lo->addr_len;
1546 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1547 lo->addr_len);
1548 }
1549 if (limit > MPLS_LABEL_IPV6NULL) {
1550 struct net_device *lo = net->loopback_dev;
1551 rt2 = mpls_rt_alloc(1, lo->addr_len);
1552 if (!rt2)
1553 goto nort2;
1554 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1555 rt2->rt_protocol = RTPROT_KERNEL;
1556 rt2->rt_payload_type = MPT_IPV6;
1557 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1558 rt2->rt_nh->nh_via_alen = lo->addr_len;
1559 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1560 lo->addr_len);
1561 }
1562
1563 rtnl_lock();
1564 /* Remember the original table */
1565 old = rtnl_dereference(net->mpls.platform_label);
1566 old_limit = net->mpls.platform_labels;
1567
1568 /* Free any labels beyond the new table */
1569 for (index = limit; index < old_limit; index++)
1570 mpls_route_update(net, index, NULL, NULL);
1571
1572 /* Copy over the old labels */
1573 cp_size = size;
1574 if (old_limit < limit)
1575 cp_size = old_limit * sizeof(struct mpls_route *);
1576
1577 memcpy(labels, old, cp_size);
1578
1579 /* If needed set the predefined labels */
1580 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1581 (limit > MPLS_LABEL_IPV6NULL)) {
1582 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1583 rt2 = NULL;
1584 }
1585
1586 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1587 (limit > MPLS_LABEL_IPV4NULL)) {
1588 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1589 rt0 = NULL;
1590 }
1591
1592 /* Update the global pointers */
1593 net->mpls.platform_labels = limit;
1594 rcu_assign_pointer(net->mpls.platform_label, labels);
1595
1596 rtnl_unlock();
1597
1598 mpls_rt_free(rt2);
1599 mpls_rt_free(rt0);
1600
1601 if (old) {
1602 synchronize_rcu();
1603 kvfree(old);
1604 }
1605 return 0;
1606
1607 nort2:
1608 mpls_rt_free(rt0);
1609 nort0:
1610 kvfree(labels);
1611 nolabels:
1612 return -ENOMEM;
1613 }
1614
1615 static int mpls_platform_labels(struct ctl_table *table, int write,
1616 void __user *buffer, size_t *lenp, loff_t *ppos)
1617 {
1618 struct net *net = table->data;
1619 int platform_labels = net->mpls.platform_labels;
1620 int ret;
1621 struct ctl_table tmp = {
1622 .procname = table->procname,
1623 .data = &platform_labels,
1624 .maxlen = sizeof(int),
1625 .mode = table->mode,
1626 .extra1 = &zero,
1627 .extra2 = &label_limit,
1628 };
1629
1630 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1631
1632 if (write && ret == 0)
1633 ret = resize_platform_label_table(net, platform_labels);
1634
1635 return ret;
1636 }
1637
1638 static const struct ctl_table mpls_table[] = {
1639 {
1640 .procname = "platform_labels",
1641 .data = NULL,
1642 .maxlen = sizeof(int),
1643 .mode = 0644,
1644 .proc_handler = mpls_platform_labels,
1645 },
1646 { }
1647 };
1648
1649 static int mpls_net_init(struct net *net)
1650 {
1651 struct ctl_table *table;
1652
1653 net->mpls.platform_labels = 0;
1654 net->mpls.platform_label = NULL;
1655
1656 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1657 if (table == NULL)
1658 return -ENOMEM;
1659
1660 table[0].data = net;
1661 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1662 if (net->mpls.ctl == NULL) {
1663 kfree(table);
1664 return -ENOMEM;
1665 }
1666
1667 return 0;
1668 }
1669
1670 static void mpls_net_exit(struct net *net)
1671 {
1672 struct mpls_route __rcu **platform_label;
1673 size_t platform_labels;
1674 struct ctl_table *table;
1675 unsigned int index;
1676
1677 table = net->mpls.ctl->ctl_table_arg;
1678 unregister_net_sysctl_table(net->mpls.ctl);
1679 kfree(table);
1680
1681 /* An rcu grace period has passed since there was a device in
1682 * the network namespace (and thus the last in flight packet)
1683 * left this network namespace. This is because
1684 * unregister_netdevice_many and netdev_run_todo has completed
1685 * for each network device that was in this network namespace.
1686 *
1687 * As such no additional rcu synchronization is necessary when
1688 * freeing the platform_label table.
1689 */
1690 rtnl_lock();
1691 platform_label = rtnl_dereference(net->mpls.platform_label);
1692 platform_labels = net->mpls.platform_labels;
1693 for (index = 0; index < platform_labels; index++) {
1694 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1695 RCU_INIT_POINTER(platform_label[index], NULL);
1696 mpls_rt_free(rt);
1697 }
1698 rtnl_unlock();
1699
1700 kvfree(platform_label);
1701 }
1702
1703 static struct pernet_operations mpls_net_ops = {
1704 .init = mpls_net_init,
1705 .exit = mpls_net_exit,
1706 };
1707
1708 static int __init mpls_init(void)
1709 {
1710 int err;
1711
1712 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1713
1714 err = register_pernet_subsys(&mpls_net_ops);
1715 if (err)
1716 goto out;
1717
1718 err = register_netdevice_notifier(&mpls_dev_notifier);
1719 if (err)
1720 goto out_unregister_pernet;
1721
1722 dev_add_pack(&mpls_packet_type);
1723
1724 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1725 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1726 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1727 err = 0;
1728 out:
1729 return err;
1730
1731 out_unregister_pernet:
1732 unregister_pernet_subsys(&mpls_net_ops);
1733 goto out;
1734 }
1735 module_init(mpls_init);
1736
1737 static void __exit mpls_exit(void)
1738 {
1739 rtnl_unregister_all(PF_MPLS);
1740 dev_remove_pack(&mpls_packet_type);
1741 unregister_netdevice_notifier(&mpls_dev_notifier);
1742 unregister_pernet_subsys(&mpls_net_ops);
1743 }
1744 module_exit(mpls_exit);
1745
1746 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1747 MODULE_LICENSE("GPL v2");
1748 MODULE_ALIAS_NETPROTO(PF_MPLS);
Linux kernel - Deliverables
This page took 0.103094 seconds and 6 git commands to generate.