2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * IPv4 Forwarding Information Base: semantics.
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #include <asm/uaccess.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
42 #include <net/ip_fib.h>
43 #include <net/netlink.h>
44 #include <net/nexthop.h>
45 #include <net/lwtunnel.h>
47 #include "fib_lookup.h"
49 static DEFINE_SPINLOCK(fib_info_lock);
50 static struct hlist_head *fib_info_hash;
51 static struct hlist_head *fib_info_laddrhash;
52 static unsigned int fib_info_hash_size;
53 static unsigned int fib_info_cnt;
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
60 u32 fib_multipath_secret __read_mostly;
62 #define for_nexthops(fi) { \
63 int nhsel; const struct fib_nh *nh; \
64 for (nhsel = 0, nh = (fi)->fib_nh; \
65 nhsel < (fi)->fib_nhs; \
68 #define change_nexthops(fi) { \
69 int nhsel; struct fib_nh *nexthop_nh; \
70 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
71 nhsel < (fi)->fib_nhs; \
72 nexthop_nh++, nhsel++)
74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
76 /* Hope, that gcc will optimize it to get rid of dummy loop */
78 #define for_nexthops(fi) { \
79 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
80 for (nhsel = 0; nhsel < 1; nhsel++)
82 #define change_nexthops(fi) { \
84 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
85 for (nhsel = 0; nhsel < 1; nhsel++)
87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
89 #define endfor_nexthops(fi) }
92 const struct fib_prop fib_props[RTN_MAX + 1] = {
95 .scope = RT_SCOPE_NOWHERE,
99 .scope = RT_SCOPE_UNIVERSE,
103 .scope = RT_SCOPE_HOST,
107 .scope = RT_SCOPE_LINK,
111 .scope = RT_SCOPE_LINK,
115 .scope = RT_SCOPE_UNIVERSE,
119 .scope = RT_SCOPE_UNIVERSE,
121 [RTN_UNREACHABLE] = {
122 .error = -EHOSTUNREACH,
123 .scope = RT_SCOPE_UNIVERSE,
127 .scope = RT_SCOPE_UNIVERSE,
131 .scope = RT_SCOPE_UNIVERSE,
135 .scope = RT_SCOPE_NOWHERE,
139 .scope = RT_SCOPE_NOWHERE,
143 static void rt_fibinfo_free(struct rtable __rcu **rtp)
145 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
150 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
151 * because we waited an RCU grace period before calling
152 * free_fib_info_rcu()
158 static void free_nh_exceptions(struct fib_nh *nh)
160 struct fnhe_hash_bucket *hash;
163 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
166 for (i = 0; i < FNHE_HASH_SIZE; i++) {
167 struct fib_nh_exception *fnhe;
169 fnhe = rcu_dereference_protected(hash[i].chain, 1);
171 struct fib_nh_exception *next;
173 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
175 rt_fibinfo_free(&fnhe->fnhe_rth_input);
176 rt_fibinfo_free(&fnhe->fnhe_rth_output);
186 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
193 for_each_possible_cpu(cpu) {
196 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
203 /* Release a nexthop info record */
204 static void free_fib_info_rcu(struct rcu_head *head)
206 struct fib_info *fi = container_of(head, struct fib_info, rcu);
207 struct dst_metrics *m;
209 change_nexthops(fi) {
210 if (nexthop_nh->nh_dev)
211 dev_put(nexthop_nh->nh_dev);
212 lwtstate_put(nexthop_nh->nh_lwtstate);
213 free_nh_exceptions(nexthop_nh);
214 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
215 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
216 } endfor_nexthops(fi);
219 if (m != &dst_default_metrics && atomic_dec_and_test(&m->refcnt))
224 void free_fib_info(struct fib_info *fi)
226 if (fi->fib_dead == 0) {
227 pr_warn("Freeing alive fib_info %p\n", fi);
231 #ifdef CONFIG_IP_ROUTE_CLASSID
232 change_nexthops(fi) {
233 if (nexthop_nh->nh_tclassid)
234 fi->fib_net->ipv4.fib_num_tclassid_users--;
235 } endfor_nexthops(fi);
237 call_rcu(&fi->rcu, free_fib_info_rcu);
240 void fib_release_info(struct fib_info *fi)
242 spin_lock_bh(&fib_info_lock);
243 if (fi && --fi->fib_treeref == 0) {
244 hlist_del(&fi->fib_hash);
246 hlist_del(&fi->fib_lhash);
247 change_nexthops(fi) {
248 if (!nexthop_nh->nh_dev)
250 hlist_del(&nexthop_nh->nh_hash);
251 } endfor_nexthops(fi)
255 spin_unlock_bh(&fib_info_lock);
258 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
260 const struct fib_nh *onh = ofi->fib_nh;
263 if (nh->nh_oif != onh->nh_oif ||
264 nh->nh_gw != onh->nh_gw ||
265 nh->nh_scope != onh->nh_scope ||
266 #ifdef CONFIG_IP_ROUTE_MULTIPATH
267 nh->nh_weight != onh->nh_weight ||
269 #ifdef CONFIG_IP_ROUTE_CLASSID
270 nh->nh_tclassid != onh->nh_tclassid ||
272 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
273 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
276 } endfor_nexthops(fi);
280 static inline unsigned int fib_devindex_hashfn(unsigned int val)
282 unsigned int mask = DEVINDEX_HASHSIZE - 1;
285 (val >> DEVINDEX_HASHBITS) ^
286 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
289 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
291 unsigned int mask = (fib_info_hash_size - 1);
292 unsigned int val = fi->fib_nhs;
294 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
295 val ^= (__force u32)fi->fib_prefsrc;
296 val ^= fi->fib_priority;
298 val ^= fib_devindex_hashfn(nh->nh_oif);
299 } endfor_nexthops(fi)
301 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
304 static struct fib_info *fib_find_info(const struct fib_info *nfi)
306 struct hlist_head *head;
310 hash = fib_info_hashfn(nfi);
311 head = &fib_info_hash[hash];
313 hlist_for_each_entry(fi, head, fib_hash) {
314 if (!net_eq(fi->fib_net, nfi->fib_net))
316 if (fi->fib_nhs != nfi->fib_nhs)
318 if (nfi->fib_protocol == fi->fib_protocol &&
319 nfi->fib_scope == fi->fib_scope &&
320 nfi->fib_prefsrc == fi->fib_prefsrc &&
321 nfi->fib_priority == fi->fib_priority &&
322 nfi->fib_type == fi->fib_type &&
323 memcmp(nfi->fib_metrics, fi->fib_metrics,
324 sizeof(u32) * RTAX_MAX) == 0 &&
325 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
326 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
333 /* Check, that the gateway is already configured.
334 * Used only by redirect accept routine.
336 int ip_fib_check_default(__be32 gw, struct net_device *dev)
338 struct hlist_head *head;
342 spin_lock(&fib_info_lock);
344 hash = fib_devindex_hashfn(dev->ifindex);
345 head = &fib_info_devhash[hash];
346 hlist_for_each_entry(nh, head, nh_hash) {
347 if (nh->nh_dev == dev &&
349 !(nh->nh_flags & RTNH_F_DEAD)) {
350 spin_unlock(&fib_info_lock);
355 spin_unlock(&fib_info_lock);
360 static inline size_t fib_nlmsg_size(struct fib_info *fi)
362 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
363 + nla_total_size(4) /* RTA_TABLE */
364 + nla_total_size(4) /* RTA_DST */
365 + nla_total_size(4) /* RTA_PRIORITY */
366 + nla_total_size(4) /* RTA_PREFSRC */
367 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
369 /* space for nested metrics */
370 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
373 size_t nh_encapsize = 0;
374 /* Also handles the special case fib_nhs == 1 */
376 /* each nexthop is packed in an attribute */
377 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
379 /* may contain flow and gateway attribute */
380 nhsize += 2 * nla_total_size(4);
382 /* grab encap info */
384 if (nh->nh_lwtstate) {
386 nh_encapsize += lwtunnel_get_encap_size(
389 nh_encapsize += nla_total_size(2);
391 } endfor_nexthops(fi);
393 /* all nexthops are packed in a nested attribute */
394 payload += nla_total_size((fi->fib_nhs * nhsize) +
402 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
403 int dst_len, u32 tb_id, const struct nl_info *info,
404 unsigned int nlm_flags)
407 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
410 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
414 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
415 fa->fa_type, key, dst_len,
416 fa->fa_tos, fa->fa_info, nlm_flags);
418 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
419 WARN_ON(err == -EMSGSIZE);
423 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
424 info->nlh, GFP_KERNEL);
428 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
431 static int fib_detect_death(struct fib_info *fi, int order,
432 struct fib_info **last_resort, int *last_idx,
436 int state = NUD_NONE;
438 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
440 state = n->nud_state;
445 if (state == NUD_REACHABLE)
447 if ((state & NUD_VALID) && order != dflt)
449 if ((state & NUD_VALID) ||
450 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
457 #ifdef CONFIG_IP_ROUTE_MULTIPATH
459 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
463 while (rtnh_ok(rtnh, remaining)) {
465 rtnh = rtnh_next(rtnh, &remaining);
468 /* leftover implies invalid nexthop configuration, discard it */
469 return remaining > 0 ? 0 : nhs;
472 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
473 int remaining, struct fib_config *cfg)
475 struct net *net = cfg->fc_nlinfo.nl_net;
478 change_nexthops(fi) {
481 if (!rtnh_ok(rtnh, remaining))
484 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
487 nexthop_nh->nh_flags =
488 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
489 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
490 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
492 attrlen = rtnh_attrlen(rtnh);
494 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
496 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
497 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
498 #ifdef CONFIG_IP_ROUTE_CLASSID
499 nla = nla_find(attrs, attrlen, RTA_FLOW);
500 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
501 if (nexthop_nh->nh_tclassid)
502 fi->fib_net->ipv4.fib_num_tclassid_users++;
504 nla = nla_find(attrs, attrlen, RTA_ENCAP);
506 struct lwtunnel_state *lwtstate;
507 struct net_device *dev = NULL;
508 struct nlattr *nla_entype;
510 nla_entype = nla_find(attrs, attrlen,
515 dev = __dev_get_by_index(net, cfg->fc_oif);
516 ret = lwtunnel_build_state(dev, nla_get_u16(
522 nexthop_nh->nh_lwtstate =
523 lwtstate_get(lwtstate);
527 rtnh = rtnh_next(rtnh, &remaining);
528 } endfor_nexthops(fi);
539 static void fib_rebalance(struct fib_info *fi)
543 struct in_device *in_dev;
550 if (nh->nh_flags & RTNH_F_DEAD)
553 in_dev = __in_dev_get_rtnl(nh->nh_dev);
556 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
557 nh->nh_flags & RTNH_F_LINKDOWN)
560 total += nh->nh_weight;
561 } endfor_nexthops(fi);
564 change_nexthops(fi) {
567 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
569 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
572 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
573 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
576 w += nexthop_nh->nh_weight;
577 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
581 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
582 } endfor_nexthops(fi);
584 net_get_random_once(&fib_multipath_secret,
585 sizeof(fib_multipath_secret));
588 static inline void fib_add_weight(struct fib_info *fi,
589 const struct fib_nh *nh)
591 fi->fib_weight += nh->nh_weight;
594 #else /* CONFIG_IP_ROUTE_MULTIPATH */
596 #define fib_rebalance(fi) do { } while (0)
597 #define fib_add_weight(fi, nh) do { } while (0)
599 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
601 static int fib_encap_match(struct net *net, u16 encap_type,
602 struct nlattr *encap,
603 int oif, const struct fib_nh *nh,
604 const struct fib_config *cfg)
606 struct lwtunnel_state *lwtstate;
607 struct net_device *dev = NULL;
610 if (encap_type == LWTUNNEL_ENCAP_NONE)
614 dev = __dev_get_by_index(net, oif);
615 ret = lwtunnel_build_state(dev, encap_type, encap,
616 AF_INET, cfg, &lwtstate);
618 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
619 lwtstate_free(lwtstate);
625 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
627 struct net *net = cfg->fc_nlinfo.nl_net;
628 #ifdef CONFIG_IP_ROUTE_MULTIPATH
629 struct rtnexthop *rtnh;
633 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
636 if (cfg->fc_oif || cfg->fc_gw) {
638 if (fib_encap_match(net, cfg->fc_encap_type,
639 cfg->fc_encap, cfg->fc_oif,
643 #ifdef CONFIG_IP_ROUTE_CLASSID
645 cfg->fc_flow != fi->fib_nh->nh_tclassid)
648 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
649 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
654 #ifdef CONFIG_IP_ROUTE_MULTIPATH
659 remaining = cfg->fc_mp_len;
664 if (!rtnh_ok(rtnh, remaining))
667 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
670 attrlen = rtnh_attrlen(rtnh);
672 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
674 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
675 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
677 #ifdef CONFIG_IP_ROUTE_CLASSID
678 nla = nla_find(attrs, attrlen, RTA_FLOW);
679 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
684 rtnh = rtnh_next(rtnh, &remaining);
685 } endfor_nexthops(fi);
695 * Semantics of nexthop is very messy by historical reasons.
696 * We have to take into account, that:
697 * a) gateway can be actually local interface address,
698 * so that gatewayed route is direct.
699 * b) gateway must be on-link address, possibly
700 * described not by an ifaddr, but also by a direct route.
701 * c) If both gateway and interface are specified, they should not
703 * d) If we use tunnel routes, gateway could be not on-link.
705 * Attempt to reconcile all of these (alas, self-contradictory) conditions
706 * results in pretty ugly and hairy code with obscure logic.
708 * I chose to generalized it instead, so that the size
709 * of code does not increase practically, but it becomes
711 * Every prefix is assigned a "scope" value: "host" is local address,
712 * "link" is direct route,
713 * [ ... "site" ... "interior" ... ]
714 * and "universe" is true gateway route with global meaning.
716 * Every prefix refers to a set of "nexthop"s (gw, oif),
717 * where gw must have narrower scope. This recursion stops
718 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
719 * which means that gw is forced to be on link.
721 * Code is still hairy, but now it is apparently logically
722 * consistent and very flexible. F.e. as by-product it allows
723 * to co-exists in peace independent exterior and interior
726 * Normally it looks as following.
728 * {universe prefix} -> (gw, oif) [scope link]
730 * |-> {link prefix} -> (gw, oif) [scope local]
732 * |-> {local prefix} (terminal node)
734 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
739 struct net_device *dev;
741 net = cfg->fc_nlinfo.nl_net;
743 struct fib_result res;
745 if (nh->nh_flags & RTNH_F_ONLINK) {
746 unsigned int addr_type;
748 if (cfg->fc_scope >= RT_SCOPE_LINK)
750 dev = __dev_get_by_index(net, nh->nh_oif);
753 if (!(dev->flags & IFF_UP))
755 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
756 if (addr_type != RTN_UNICAST)
758 if (!netif_carrier_ok(dev))
759 nh->nh_flags |= RTNH_F_LINKDOWN;
762 nh->nh_scope = RT_SCOPE_LINK;
767 struct fib_table *tbl = NULL;
768 struct flowi4 fl4 = {
770 .flowi4_scope = cfg->fc_scope + 1,
771 .flowi4_oif = nh->nh_oif,
772 .flowi4_iif = LOOPBACK_IFINDEX,
775 /* It is not necessary, but requires a bit of thinking */
776 if (fl4.flowi4_scope < RT_SCOPE_LINK)
777 fl4.flowi4_scope = RT_SCOPE_LINK;
779 if (cfg->fc_table && cfg->fc_table != RT_TABLE_MAIN)
780 tbl = fib_get_table(net, cfg->fc_table);
783 err = fib_table_lookup(tbl, &fl4, &res,
784 FIB_LOOKUP_IGNORE_LINKSTATE |
787 /* on error or if no table given do full lookup. This
788 * is needed for example when nexthops are in the local
789 * table rather than the given table
792 err = fib_lookup(net, &fl4, &res,
793 FIB_LOOKUP_IGNORE_LINKSTATE);
802 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
804 nh->nh_scope = res.scope;
805 nh->nh_oif = FIB_RES_OIF(res);
806 nh->nh_dev = dev = FIB_RES_DEV(res);
810 if (!netif_carrier_ok(dev))
811 nh->nh_flags |= RTNH_F_LINKDOWN;
812 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
814 struct in_device *in_dev;
816 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
821 in_dev = inetdev_by_index(net, nh->nh_oif);
825 if (!(in_dev->dev->flags & IFF_UP))
827 nh->nh_dev = in_dev->dev;
828 dev_hold(nh->nh_dev);
829 nh->nh_scope = RT_SCOPE_HOST;
830 if (!netif_carrier_ok(nh->nh_dev))
831 nh->nh_flags |= RTNH_F_LINKDOWN;
839 static inline unsigned int fib_laddr_hashfn(__be32 val)
841 unsigned int mask = (fib_info_hash_size - 1);
843 return ((__force u32)val ^
844 ((__force u32)val >> 7) ^
845 ((__force u32)val >> 14)) & mask;
848 static struct hlist_head *fib_info_hash_alloc(int bytes)
850 if (bytes <= PAGE_SIZE)
851 return kzalloc(bytes, GFP_KERNEL);
853 return (struct hlist_head *)
854 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
858 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
863 if (bytes <= PAGE_SIZE)
866 free_pages((unsigned long) hash, get_order(bytes));
869 static void fib_info_hash_move(struct hlist_head *new_info_hash,
870 struct hlist_head *new_laddrhash,
871 unsigned int new_size)
873 struct hlist_head *old_info_hash, *old_laddrhash;
874 unsigned int old_size = fib_info_hash_size;
875 unsigned int i, bytes;
877 spin_lock_bh(&fib_info_lock);
878 old_info_hash = fib_info_hash;
879 old_laddrhash = fib_info_laddrhash;
880 fib_info_hash_size = new_size;
882 for (i = 0; i < old_size; i++) {
883 struct hlist_head *head = &fib_info_hash[i];
884 struct hlist_node *n;
887 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
888 struct hlist_head *dest;
889 unsigned int new_hash;
891 new_hash = fib_info_hashfn(fi);
892 dest = &new_info_hash[new_hash];
893 hlist_add_head(&fi->fib_hash, dest);
896 fib_info_hash = new_info_hash;
898 for (i = 0; i < old_size; i++) {
899 struct hlist_head *lhead = &fib_info_laddrhash[i];
900 struct hlist_node *n;
903 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
904 struct hlist_head *ldest;
905 unsigned int new_hash;
907 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
908 ldest = &new_laddrhash[new_hash];
909 hlist_add_head(&fi->fib_lhash, ldest);
912 fib_info_laddrhash = new_laddrhash;
914 spin_unlock_bh(&fib_info_lock);
916 bytes = old_size * sizeof(struct hlist_head *);
917 fib_info_hash_free(old_info_hash, bytes);
918 fib_info_hash_free(old_laddrhash, bytes);
921 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
923 nh->nh_saddr = inet_select_addr(nh->nh_dev,
925 nh->nh_parent->fib_scope);
926 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
931 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
933 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
934 fib_prefsrc != cfg->fc_dst) {
935 u32 tb_id = cfg->fc_table;
938 if (tb_id == RT_TABLE_MAIN)
939 tb_id = RT_TABLE_LOCAL;
941 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
944 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
945 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
946 fib_prefsrc, RT_TABLE_LOCAL);
956 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
965 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
966 int type = nla_type(nla);
974 if (type == RTAX_CC_ALGO) {
975 char tmp[TCP_CA_NAME_MAX];
977 nla_strlcpy(tmp, nla, sizeof(tmp));
978 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
979 if (val == TCP_CA_UNSPEC)
982 if (nla_len(nla) != sizeof(u32))
984 val = nla_get_u32(nla);
986 if (type == RTAX_ADVMSS && val > 65535 - 40)
988 if (type == RTAX_MTU && val > 65535 - 15)
990 if (type == RTAX_HOPLIMIT && val > 255)
992 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
994 fi->fib_metrics->metrics[type - 1] = val;
998 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1003 struct fib_info *fib_create_info(struct fib_config *cfg)
1006 struct fib_info *fi = NULL;
1007 struct fib_info *ofi;
1009 struct net *net = cfg->fc_nlinfo.nl_net;
1011 if (cfg->fc_type > RTN_MAX)
1014 /* Fast check to catch the most weird cases */
1015 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1018 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1021 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1023 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1030 if (fib_info_cnt >= fib_info_hash_size) {
1031 unsigned int new_size = fib_info_hash_size << 1;
1032 struct hlist_head *new_info_hash;
1033 struct hlist_head *new_laddrhash;
1038 bytes = new_size * sizeof(struct hlist_head *);
1039 new_info_hash = fib_info_hash_alloc(bytes);
1040 new_laddrhash = fib_info_hash_alloc(bytes);
1041 if (!new_info_hash || !new_laddrhash) {
1042 fib_info_hash_free(new_info_hash, bytes);
1043 fib_info_hash_free(new_laddrhash, bytes);
1045 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1047 if (!fib_info_hash_size)
1051 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1055 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1056 if (unlikely(!fi->fib_metrics)) {
1058 return ERR_PTR(err);
1060 atomic_set(&fi->fib_metrics->refcnt, 1);
1062 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1066 fi->fib_protocol = cfg->fc_protocol;
1067 fi->fib_scope = cfg->fc_scope;
1068 fi->fib_flags = cfg->fc_flags;
1069 fi->fib_priority = cfg->fc_priority;
1070 fi->fib_prefsrc = cfg->fc_prefsrc;
1071 fi->fib_type = cfg->fc_type;
1072 fi->fib_tb_id = cfg->fc_table;
1075 change_nexthops(fi) {
1076 nexthop_nh->nh_parent = fi;
1077 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1078 if (!nexthop_nh->nh_pcpu_rth_output)
1080 } endfor_nexthops(fi)
1082 err = fib_convert_metrics(fi, cfg);
1087 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1088 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1091 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1093 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1095 #ifdef CONFIG_IP_ROUTE_CLASSID
1096 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1103 struct fib_nh *nh = fi->fib_nh;
1105 if (cfg->fc_encap) {
1106 struct lwtunnel_state *lwtstate;
1107 struct net_device *dev = NULL;
1109 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1112 dev = __dev_get_by_index(net, cfg->fc_oif);
1113 err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1114 cfg->fc_encap, AF_INET, cfg,
1119 nh->nh_lwtstate = lwtstate_get(lwtstate);
1121 nh->nh_oif = cfg->fc_oif;
1122 nh->nh_gw = cfg->fc_gw;
1123 nh->nh_flags = cfg->fc_flags;
1124 #ifdef CONFIG_IP_ROUTE_CLASSID
1125 nh->nh_tclassid = cfg->fc_flow;
1126 if (nh->nh_tclassid)
1127 fi->fib_net->ipv4.fib_num_tclassid_users++;
1129 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1134 if (fib_props[cfg->fc_type].error) {
1135 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1139 switch (cfg->fc_type) {
1151 if (cfg->fc_scope > RT_SCOPE_HOST)
1154 if (cfg->fc_scope == RT_SCOPE_HOST) {
1155 struct fib_nh *nh = fi->fib_nh;
1157 /* Local address is added. */
1158 if (nhs != 1 || nh->nh_gw)
1160 nh->nh_scope = RT_SCOPE_NOWHERE;
1161 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1168 change_nexthops(fi) {
1169 err = fib_check_nh(cfg, fi, nexthop_nh);
1172 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1174 } endfor_nexthops(fi)
1175 if (linkdown == fi->fib_nhs)
1176 fi->fib_flags |= RTNH_F_LINKDOWN;
1179 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1182 change_nexthops(fi) {
1183 fib_info_update_nh_saddr(net, nexthop_nh);
1184 fib_add_weight(fi, nexthop_nh);
1185 } endfor_nexthops(fi)
1190 ofi = fib_find_info(fi);
1199 atomic_inc(&fi->fib_clntref);
1200 spin_lock_bh(&fib_info_lock);
1201 hlist_add_head(&fi->fib_hash,
1202 &fib_info_hash[fib_info_hashfn(fi)]);
1203 if (fi->fib_prefsrc) {
1204 struct hlist_head *head;
1206 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1207 hlist_add_head(&fi->fib_lhash, head);
1209 change_nexthops(fi) {
1210 struct hlist_head *head;
1213 if (!nexthop_nh->nh_dev)
1215 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1216 head = &fib_info_devhash[hash];
1217 hlist_add_head(&nexthop_nh->nh_hash, head);
1218 } endfor_nexthops(fi)
1219 spin_unlock_bh(&fib_info_lock);
1231 return ERR_PTR(err);
1234 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1235 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1236 struct fib_info *fi, unsigned int flags)
1238 struct nlmsghdr *nlh;
1241 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1245 rtm = nlmsg_data(nlh);
1246 rtm->rtm_family = AF_INET;
1247 rtm->rtm_dst_len = dst_len;
1248 rtm->rtm_src_len = 0;
1251 rtm->rtm_table = tb_id;
1253 rtm->rtm_table = RT_TABLE_COMPAT;
1254 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1255 goto nla_put_failure;
1256 rtm->rtm_type = type;
1257 rtm->rtm_flags = fi->fib_flags;
1258 rtm->rtm_scope = fi->fib_scope;
1259 rtm->rtm_protocol = fi->fib_protocol;
1261 if (rtm->rtm_dst_len &&
1262 nla_put_in_addr(skb, RTA_DST, dst))
1263 goto nla_put_failure;
1264 if (fi->fib_priority &&
1265 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1266 goto nla_put_failure;
1267 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1268 goto nla_put_failure;
1270 if (fi->fib_prefsrc &&
1271 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1272 goto nla_put_failure;
1273 if (fi->fib_nhs == 1) {
1274 struct in_device *in_dev;
1276 if (fi->fib_nh->nh_gw &&
1277 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1278 goto nla_put_failure;
1279 if (fi->fib_nh->nh_oif &&
1280 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1281 goto nla_put_failure;
1282 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1283 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1285 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1286 rtm->rtm_flags |= RTNH_F_DEAD;
1288 #ifdef CONFIG_IP_ROUTE_CLASSID
1289 if (fi->fib_nh[0].nh_tclassid &&
1290 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1291 goto nla_put_failure;
1293 if (fi->fib_nh->nh_lwtstate &&
1294 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1295 goto nla_put_failure;
1297 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1298 if (fi->fib_nhs > 1) {
1299 struct rtnexthop *rtnh;
1302 mp = nla_nest_start(skb, RTA_MULTIPATH);
1304 goto nla_put_failure;
1307 struct in_device *in_dev;
1309 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1311 goto nla_put_failure;
1313 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1314 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1315 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1317 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1318 rtnh->rtnh_flags |= RTNH_F_DEAD;
1320 rtnh->rtnh_hops = nh->nh_weight - 1;
1321 rtnh->rtnh_ifindex = nh->nh_oif;
1324 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1325 goto nla_put_failure;
1326 #ifdef CONFIG_IP_ROUTE_CLASSID
1327 if (nh->nh_tclassid &&
1328 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1329 goto nla_put_failure;
1331 if (nh->nh_lwtstate &&
1332 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1333 goto nla_put_failure;
1335 /* length of rtnetlink header + attributes */
1336 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1337 } endfor_nexthops(fi);
1339 nla_nest_end(skb, mp);
1342 nlmsg_end(skb, nlh);
1346 nlmsg_cancel(skb, nlh);
1352 * - local address disappeared -> we must delete all the entries
1354 * - device went down -> we must shutdown all nexthops going via it.
1356 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1359 unsigned int hash = fib_laddr_hashfn(local);
1360 struct hlist_head *head = &fib_info_laddrhash[hash];
1361 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1362 struct net *net = dev_net(dev);
1363 struct fib_info *fi;
1365 if (!fib_info_laddrhash || local == 0)
1368 hlist_for_each_entry(fi, head, fib_lhash) {
1369 if (!net_eq(fi->fib_net, net) ||
1370 fi->fib_tb_id != tb_id)
1372 if (fi->fib_prefsrc == local) {
1373 fi->fib_flags |= RTNH_F_DEAD;
1380 /* Update the PMTU of exceptions when:
1381 * - the new MTU of the first hop becomes smaller than the PMTU
1382 * - the old MTU was the same as the PMTU, and it limited discovery of
1383 * larger MTUs on the path. With that limit raised, we can now
1384 * discover larger MTUs
1385 * A special case is locked exceptions, for which the PMTU is smaller
1386 * than the minimal accepted PMTU:
1387 * - if the new MTU is greater than the PMTU, don't make any change
1388 * - otherwise, unlock and set PMTU
1390 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig)
1392 struct fnhe_hash_bucket *bucket;
1395 bucket = rcu_dereference_protected(nh->nh_exceptions, 1);
1399 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1400 struct fib_nh_exception *fnhe;
1402 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1404 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1405 if (fnhe->fnhe_mtu_locked) {
1406 if (new <= fnhe->fnhe_pmtu) {
1407 fnhe->fnhe_pmtu = new;
1408 fnhe->fnhe_mtu_locked = false;
1410 } else if (new < fnhe->fnhe_pmtu ||
1411 orig == fnhe->fnhe_pmtu) {
1412 fnhe->fnhe_pmtu = new;
1418 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1420 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1421 struct hlist_head *head = &fib_info_devhash[hash];
1424 hlist_for_each_entry(nh, head, nh_hash) {
1425 if (nh->nh_dev == dev)
1426 nh_update_mtu(nh, dev->mtu, orig_mtu);
1430 /* Event force Flags Description
1431 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1432 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1433 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1434 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1436 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1439 int scope = RT_SCOPE_NOWHERE;
1440 struct fib_info *prev_fi = NULL;
1441 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1442 struct hlist_head *head = &fib_info_devhash[hash];
1448 hlist_for_each_entry(nh, head, nh_hash) {
1449 struct fib_info *fi = nh->nh_parent;
1452 BUG_ON(!fi->fib_nhs);
1453 if (nh->nh_dev != dev || fi == prev_fi)
1457 change_nexthops(fi) {
1458 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1460 else if (nexthop_nh->nh_dev == dev &&
1461 nexthop_nh->nh_scope != scope) {
1464 case NETDEV_UNREGISTER:
1465 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1468 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1473 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1474 if (event == NETDEV_UNREGISTER &&
1475 nexthop_nh->nh_dev == dev) {
1480 } endfor_nexthops(fi)
1481 if (dead == fi->fib_nhs) {
1484 case NETDEV_UNREGISTER:
1485 fi->fib_flags |= RTNH_F_DEAD;
1488 fi->fib_flags |= RTNH_F_LINKDOWN;
1500 /* Must be invoked inside of an RCU protected region. */
1501 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1503 struct fib_info *fi = NULL, *last_resort = NULL;
1504 struct hlist_head *fa_head = res->fa_head;
1505 struct fib_table *tb = res->table;
1506 u8 slen = 32 - res->prefixlen;
1507 int order = -1, last_idx = -1;
1508 struct fib_alias *fa, *fa1 = NULL;
1509 u32 last_prio = res->fi->fib_priority;
1512 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1513 struct fib_info *next_fi = fa->fa_info;
1515 if (fa->fa_slen != slen)
1517 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1519 if (fa->tb_id != tb->tb_id)
1521 if (next_fi->fib_priority > last_prio &&
1522 fa->fa_tos == last_tos) {
1527 if (next_fi->fib_flags & RTNH_F_DEAD)
1529 last_tos = fa->fa_tos;
1530 last_prio = next_fi->fib_priority;
1532 if (next_fi->fib_scope != res->scope ||
1533 fa->fa_type != RTN_UNICAST)
1535 if (!next_fi->fib_nh[0].nh_gw ||
1536 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1539 fib_alias_accessed(fa);
1542 if (next_fi != res->fi)
1545 } else if (!fib_detect_death(fi, order, &last_resort,
1546 &last_idx, fa1->fa_default)) {
1547 fib_result_assign(res, fi);
1548 fa1->fa_default = order;
1555 if (order <= 0 || !fi) {
1557 fa1->fa_default = -1;
1561 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1563 fib_result_assign(res, fi);
1564 fa1->fa_default = order;
1569 fib_result_assign(res, last_resort);
1570 fa1->fa_default = last_idx;
1576 * Dead device goes up. We wake up dead nexthops.
1577 * It takes sense only on multipath routes.
1579 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1581 struct fib_info *prev_fi;
1583 struct hlist_head *head;
1587 if (!(dev->flags & IFF_UP))
1590 if (nh_flags & RTNH_F_DEAD) {
1591 unsigned int flags = dev_get_flags(dev);
1593 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1594 nh_flags |= RTNH_F_LINKDOWN;
1598 hash = fib_devindex_hashfn(dev->ifindex);
1599 head = &fib_info_devhash[hash];
1602 hlist_for_each_entry(nh, head, nh_hash) {
1603 struct fib_info *fi = nh->nh_parent;
1606 BUG_ON(!fi->fib_nhs);
1607 if (nh->nh_dev != dev || fi == prev_fi)
1612 change_nexthops(fi) {
1613 if (!(nexthop_nh->nh_flags & nh_flags)) {
1617 if (!nexthop_nh->nh_dev ||
1618 !(nexthop_nh->nh_dev->flags & IFF_UP))
1620 if (nexthop_nh->nh_dev != dev ||
1621 !__in_dev_get_rtnl(dev))
1624 nexthop_nh->nh_flags &= ~nh_flags;
1625 } endfor_nexthops(fi)
1628 fi->fib_flags &= ~nh_flags;
1638 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1639 static bool fib_good_nh(const struct fib_nh *nh)
1641 int state = NUD_REACHABLE;
1643 if (nh->nh_scope == RT_SCOPE_LINK) {
1644 struct neighbour *n;
1648 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1649 (__force u32)nh->nh_gw);
1651 state = n->nud_state;
1653 rcu_read_unlock_bh();
1656 return !!(state & NUD_VALID);
1659 void fib_select_multipath(struct fib_result *res, int hash)
1661 struct fib_info *fi = res->fi;
1662 struct net *net = fi->fib_net;
1666 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1667 if (!fib_good_nh(nh))
1670 res->nh_sel = nhsel;
1675 if (hash > atomic_read(&nh->nh_upper_bound))
1678 res->nh_sel = nhsel;
1680 } endfor_nexthops(fi);
1684 void fib_select_path(struct net *net, struct fib_result *res,
1685 struct flowi4 *fl4, int mp_hash)
1689 oif_check = (fl4->flowi4_oif == 0 ||
1690 fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1692 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1693 if (res->fi->fib_nhs > 1 && oif_check) {
1695 mp_hash = get_hash_from_flowi4(fl4) >> 1;
1697 fib_select_multipath(res, mp_hash);
1701 if (!res->prefixlen &&
1702 res->table->tb_num_default > 1 &&
1703 res->type == RTN_UNICAST && oif_check)
1704 fib_select_default(fl4, res);
1707 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1709 EXPORT_SYMBOL_GPL(fib_select_path);