2 * Copyright (c) 2011-2012 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
30 * Link-layer Reachability Record
32 * Each interface maintains a red-black tree which contains records related
33 * to the on-link nodes which we are interested in communicating with. Each
34 * record gets allocated and inserted into the tree in the following manner:
35 * upon processing an ARP announcement or reply from a known node (i.e. there
36 * exists a ARP route entry for the node), and if a link-layer reachability
37 * record for the node doesn't yet exist; and, upon processing a ND6 RS/RA/
38 * NS/NA/redirect from a node, and if a link-layer reachability record for the
39 * node doesn't yet exist.
41 * Each newly created record is then referred to by the resolver route entry;
42 * if a record already exists, its reference count gets increased for the new
43 * resolver entry which now refers to it. A record gets removed from the tree
44 * and freed once its reference counts drops to zero, i.e. when there is no
45 * more resolver entry referring to it.
47 * A record contains the link-layer protocol (e.g. Ethertype IP/IPv6), the
48 * HW address of the sender, the "last heard from" timestamp (lr_lastrcvd) and
49 * the number of references made to it (lr_reqcnt). Because the key for each
50 * record in the red-black tree consists of the link-layer protocol, therefore
51 * the namespace for the records is partitioned based on the type of link-layer
52 * protocol, i.e. an Ethertype IP link-layer record is only referred to by one
53 * or more ARP entries; an Ethernet IPv6 link-layer record is only referred to
54 * by one or more ND6 entries. Therefore, lr_reqcnt represents the number of
55 * resolver entry references to the record for the same protocol family.
57 * Upon receiving packets from the network, the protocol's input callback
58 * (e.g. ether_inet{6}_input) informs the corresponding resolver (ARP/ND6)
59 * about the (link-layer) origin of the packet. This results in searching
60 * for a matching record in the red-black tree for the interface where the
61 * packet arrived on. If there's no match, no further processing takes place.
62 * Otherwise, the lr_lastrcvd timestamp of the record is updated.
64 * When an IP/IPv6 packet is transmitted to the resolver (i.e. the destination
65 * is on-link), ARP/ND6 records the "last spoken to" timestamp in the route
66 * entry ({la,ln}_lastused).
68 * The reachability of the on-link node is determined by the following logic,
69 * upon sending a packet thru the resolver:
71 * a) If the record is only used by exactly one resolver entry (lr_reqcnt
72 * is 1), i.e. the target host does not have IP/IPv6 aliases that we know
73 * of, check if lr_lastrcvd is "recent." If so, simply send the packet;
74 * otherwise, re-resolve the target node.
76 * b) If the record is shared by multiple resolver entries (lr_reqcnt is
77 * greater than 1), i.e. the target host has more than one IP/IPv6 aliases
78 * on the same network interface, we can't rely on lr_lastrcvd alone, as
79 * one of the IP/IPv6 aliases could have been silently moved to another
80 * node for which we don't have a link-layer record. If lr_lastrcvd is
81 * not "recent", we re-resolve the target node. Otherwise, we perform
82 * an additional check against {la,ln}_lastused to see whether it is also
83 * "recent", relative to lr_lastrcvd. If so, simply send the packet;
84 * otherwise, re-resolve the target node.
86 * The value for "recent" is configurable by adjusting the basetime value for
87 * net.link.ether.inet.arp_llreach_base or net.inet6.icmp6.nd6_llreach_base.
88 * The default basetime value is 30 seconds, and the actual expiration time
89 * is calculated by multiplying the basetime value with some random factor,
90 * which results in a number between 15 to 45 seconds. Setting the basetime
91 * value to 0 effectively disables this feature for the corresponding resolver.
95 * The above logic is based upon the following assumptions:
97 * i) Network traffics are mostly bi-directional, i.e. the act of sending
98 * packets to an on-link node would most likely cause us to receive
99 * packets from that node.
101 * ii) If the on-link node's IP/IPv6 address silently moves to another
102 * on-link node for which we are not aware of, non-unicast packets
103 * from the old node would trigger the record's lr_lastrcvd to be
106 * We can mitigate the above by having the resolver check its {la,ln}_lastused
107 * timestamp at all times, i.e. not only when lr_reqcnt is greater than 1; but
108 * we currently optimize for the common cases.
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/kernel.h>
114 #include <sys/malloc.h>
115 #include <sys/tree.h>
116 #include <sys/sysctl.h>
117 #include <sys/mcache.h>
118 #include <sys/protosw.h>
120 #include <net/if_dl.h>
122 #include <net/if_var.h>
123 #include <net/if_llreach.h>
124 #include <net/dlil.h>
125 #include <net/kpi_interface.h>
126 #include <net/route.h>
128 #include <kern/assert.h>
129 #include <kern/locks.h>
130 #include <kern/zalloc.h>
133 #include <netinet6/in6_var.h>
134 #include <netinet6/nd6.h>
137 static unsigned int iflr_size
; /* size of if_llreach */
138 static struct zone
*iflr_zone
; /* zone for if_llreach */
140 #define IFLR_ZONE_MAX 128 /* maximum elements in zone */
141 #define IFLR_ZONE_NAME "if_llreach" /* zone name */
143 static struct if_llreach
*iflr_alloc(int);
144 static void iflr_free(struct if_llreach
*);
145 static __inline
int iflr_cmp(const struct if_llreach
*,
146 const struct if_llreach
*);
147 static __inline
int iflr_reachable(struct if_llreach
*, int, u_int64_t
);
148 static int sysctl_llreach_ifinfo SYSCTL_HANDLER_ARGS
;
150 /* The following is protected by if_llreach_lock */
151 RB_GENERATE_PREV(ll_reach_tree
, if_llreach
, lr_link
, iflr_cmp
);
153 SYSCTL_DECL(_net_link_generic_system
);
155 SYSCTL_NODE(_net_link_generic_system
, OID_AUTO
, llreach_info
,
156 CTLFLAG_RD
| CTLFLAG_LOCKED
, sysctl_llreach_ifinfo
,
157 "Per-interface tree of source link-layer reachability records");
160 * Link-layer reachability is based off node constants in RFC4861.
163 #define LL_COMPUTE_RTIME(x) ND_COMPUTE_RTIME(x)
165 #define LL_MIN_RANDOM_FACTOR 512 /* 1024 * 0.5 */
166 #define LL_MAX_RANDOM_FACTOR 1536 /* 1024 * 1.5 */
167 #define LL_COMPUTE_RTIME(x) \
168 (((LL_MIN_RANDOM_FACTOR * (x >> 10)) + (random() & \
169 ((LL_MAX_RANDOM_FACTOR - LL_MIN_RANDOM_FACTOR) * (x >> 10)))) / 1000)
173 ifnet_llreach_init(void)
175 iflr_size
= sizeof (struct if_llreach
);
176 iflr_zone
= zinit(iflr_size
,
177 IFLR_ZONE_MAX
* iflr_size
, 0, IFLR_ZONE_NAME
);
178 if (iflr_zone
== NULL
) {
179 panic("%s: failed allocating %s", __func__
, IFLR_ZONE_NAME
);
182 zone_change(iflr_zone
, Z_EXPAND
, TRUE
);
183 zone_change(iflr_zone
, Z_CALLERACCT
, FALSE
);
187 ifnet_llreach_ifattach(struct ifnet
*ifp
, boolean_t reuse
)
189 lck_rw_lock_exclusive(&ifp
->if_llreach_lock
);
190 /* Initialize link-layer source tree (if not already) */
192 RB_INIT(&ifp
->if_ll_srcs
);
193 lck_rw_done(&ifp
->if_llreach_lock
);
197 ifnet_llreach_ifdetach(struct ifnet
*ifp
)
201 * Nothing to do for now; the link-layer source tree might
202 * contain entries at this point, that are still referred
203 * to by route entries pointing to this ifp.
208 * Link-layer source tree comparison function.
210 * An ordered predicate is necessary; bcmp() is not documented to return
211 * an indication of order, memcmp() is, and is an ISO C99 requirement.
214 iflr_cmp(const struct if_llreach
*a
, const struct if_llreach
*b
)
216 return (memcmp(&a
->lr_key
, &b
->lr_key
, sizeof (a
->lr_key
)));
220 iflr_reachable(struct if_llreach
*lr
, int cmp_delta
, u_int64_t tval
)
225 now
= net_uptime(); /* current approx. uptime */
227 * No need for lr_lock; atomically read the last rcvd uptime.
229 expire
= lr
->lr_lastrcvd
+ lr
->lr_reachable
;
231 * If we haven't heard back from the local host for over
232 * lr_reachable seconds, consider that the host is no
236 return (expire
>= now
);
238 * If the caller supplied a reference time, consider the
239 * host is reachable if the record hasn't expired (see above)
240 * and if the reference time is within the past lr_reachable
243 return ((expire
>= now
) && (now
- tval
) < lr
->lr_reachable
);
247 ifnet_llreach_reachable(struct if_llreach
*lr
)
250 * Check whether the cache is too old to be trusted.
252 return (iflr_reachable(lr
, 0, 0));
256 ifnet_llreach_reachable_delta(struct if_llreach
*lr
, u_int64_t tval
)
259 * Check whether the cache is too old to be trusted.
261 return (iflr_reachable(lr
, 1, tval
));
265 ifnet_llreach_set_reachable(struct ifnet
*ifp
, u_int16_t llproto
, void *addr
,
268 struct if_llreach find
, *lr
;
270 VERIFY(alen
== IF_LLREACH_MAXLEN
); /* for now */
272 find
.lr_key
.proto
= llproto
;
273 bcopy(addr
, &find
.lr_key
.addr
, IF_LLREACH_MAXLEN
);
275 lck_rw_lock_shared(&ifp
->if_llreach_lock
);
276 lr
= RB_FIND(ll_reach_tree
, &ifp
->if_ll_srcs
, &find
);
278 lck_rw_done(&ifp
->if_llreach_lock
);
282 * No need for lr_lock; atomically update the last rcvd uptime.
284 lr
->lr_lastrcvd
= net_uptime();
285 lck_rw_done(&ifp
->if_llreach_lock
);
289 ifnet_llreach_alloc(struct ifnet
*ifp
, u_int16_t llproto
, void *addr
,
290 unsigned int alen
, u_int64_t llreach_base
)
292 struct if_llreach find
, *lr
;
295 if (llreach_base
== 0)
298 VERIFY(alen
== IF_LLREACH_MAXLEN
); /* for now */
300 find
.lr_key
.proto
= llproto
;
301 bcopy(addr
, &find
.lr_key
.addr
, IF_LLREACH_MAXLEN
);
303 lck_rw_lock_shared(&ifp
->if_llreach_lock
);
304 lr
= RB_FIND(ll_reach_tree
, &ifp
->if_ll_srcs
, &find
);
308 VERIFY(lr
->lr_reqcnt
>= 1);
310 VERIFY(lr
->lr_reqcnt
!= 0);
311 IFLR_ADDREF_LOCKED(lr
); /* for caller */
312 lr
->lr_lastrcvd
= net_uptime(); /* current approx. uptime */
314 lck_rw_done(&ifp
->if_llreach_lock
);
318 if (!lck_rw_lock_shared_to_exclusive(&ifp
->if_llreach_lock
))
319 lck_rw_lock_exclusive(&ifp
->if_llreach_lock
);
321 lck_rw_assert(&ifp
->if_llreach_lock
, LCK_RW_ASSERT_EXCLUSIVE
);
323 /* in case things have changed while becoming writer */
324 lr
= RB_FIND(ll_reach_tree
, &ifp
->if_ll_srcs
, &find
);
328 lr
= iflr_alloc(M_WAITOK
);
330 lck_rw_done(&ifp
->if_llreach_lock
);
335 VERIFY(lr
->lr_reqcnt
== 1);
336 IFLR_ADDREF_LOCKED(lr
); /* for RB tree */
337 IFLR_ADDREF_LOCKED(lr
); /* for caller */
338 lr
->lr_lastrcvd
= net_uptime(); /* current approx. uptime */
339 lr
->lr_baseup
= lr
->lr_lastrcvd
; /* base uptime */
341 lr
->lr_basecal
= now
.tv_sec
; /* base calendar time */
342 lr
->lr_basereachable
= llreach_base
;
343 lr
->lr_reachable
= LL_COMPUTE_RTIME(lr
->lr_basereachable
* 1000);
344 lr
->lr_debug
|= IFD_ATTACHED
;
346 lr
->lr_key
.proto
= llproto
;
347 bcopy(addr
, &lr
->lr_key
.addr
, IF_LLREACH_MAXLEN
);
348 lr
->lr_rssi
= IFNET_RSSI_UNKNOWN
;
349 lr
->lr_lqm
= IFNET_LQM_THRESH_UNKNOWN
;
350 lr
->lr_npm
= IFNET_NPM_THRESH_UNKNOWN
;
351 RB_INSERT(ll_reach_tree
, &ifp
->if_ll_srcs
, lr
);
353 lck_rw_done(&ifp
->if_llreach_lock
);
359 ifnet_llreach_free(struct if_llreach
*lr
)
363 /* no need to lock here; lr_ifp never changes */
366 lck_rw_lock_exclusive(&ifp
->if_llreach_lock
);
368 if (lr
->lr_reqcnt
== 0) {
369 panic("%s: lr=%p negative reqcnt", __func__
, lr
);
373 if (lr
->lr_reqcnt
> 0) {
375 lck_rw_done(&ifp
->if_llreach_lock
);
376 IFLR_REMREF(lr
); /* for caller */
379 if (!(lr
->lr_debug
& IFD_ATTACHED
)) {
380 panic("%s: Attempt to detach an unattached llreach lr=%p",
384 lr
->lr_debug
&= ~IFD_ATTACHED
;
385 RB_REMOVE(ll_reach_tree
, &ifp
->if_ll_srcs
, lr
);
387 lck_rw_done(&ifp
->if_llreach_lock
);
389 IFLR_REMREF(lr
); /* for RB tree */
390 IFLR_REMREF(lr
); /* for caller */
394 ifnet_llreach_up2calexp(struct if_llreach
*lr
, u_int64_t uptime
)
396 u_int64_t calendar
= 0;
402 getmicrotime(&cnow
); /* current calendar time */
403 unow
= net_uptime(); /* current approx. uptime */
405 * Take into account possible calendar time changes;
406 * adjust base calendar value if necessary, i.e.
407 * the calendar skew should equate to the uptime skew.
409 lr
->lr_basecal
+= (cnow
.tv_sec
- lr
->lr_basecal
) -
410 (unow
- lr
->lr_baseup
);
412 calendar
= lr
->lr_basecal
+ lr
->lr_reachable
+
413 (uptime
- lr
->lr_baseup
);
420 ifnet_llreach_up2upexp(struct if_llreach
*lr
, u_int64_t uptime
)
422 return (lr
->lr_reachable
+ uptime
);
426 ifnet_llreach_get_defrouter(struct ifnet
*ifp
, int af
,
427 struct ifnet_llreach_info
*iflri
)
429 struct radix_node_head
*rnh
;
430 struct sockaddr_storage dst_ss
, mask_ss
;
434 VERIFY(ifp
!= NULL
&& iflri
!= NULL
&&
435 (af
== AF_INET
|| af
== AF_INET6
));
437 bzero(iflri
, sizeof (*iflri
));
439 if ((rnh
= rt_tables
[af
]) == NULL
)
442 bzero(&dst_ss
, sizeof (dst_ss
));
443 bzero(&mask_ss
, sizeof (mask_ss
));
444 dst_ss
.ss_family
= af
;
445 dst_ss
.ss_len
= (af
== AF_INET
) ? sizeof (struct sockaddr_in
) :
446 sizeof (struct sockaddr_in6
);
448 lck_mtx_lock(rnh_lock
);
449 rt
= rt_lookup(TRUE
, SA(&dst_ss
), SA(&mask_ss
), rnh
, ifp
->if_index
);
451 struct rtentry
*gwrt
;
454 if ((rt
->rt_flags
& RTF_GATEWAY
) &&
455 (gwrt
= rt
->rt_gwroute
) != NULL
&&
456 rt_key(rt
)->sa_family
== rt_key(gwrt
)->sa_family
&&
457 (gwrt
->rt_flags
& RTF_UP
)) {
460 if (gwrt
->rt_llinfo_get_iflri
!= NULL
) {
461 (*gwrt
->rt_llinfo_get_iflri
)(gwrt
, iflri
);
470 lck_mtx_unlock(rnh_lock
);
475 static struct if_llreach
*
478 struct if_llreach
*lr
;
480 lr
= (how
== M_WAITOK
) ? zalloc(iflr_zone
) : zalloc_noblock(iflr_zone
);
482 bzero(lr
, iflr_size
);
483 lck_mtx_init(&lr
->lr_lock
, ifnet_lock_group
, ifnet_lock_attr
);
484 lr
->lr_debug
|= IFD_ALLOC
;
490 iflr_free(struct if_llreach
*lr
)
493 if (lr
->lr_debug
& IFD_ATTACHED
) {
494 panic("%s: attached lr=%p is being freed", __func__
, lr
);
496 } else if (!(lr
->lr_debug
& IFD_ALLOC
)) {
497 panic("%s: lr %p cannot be freed", __func__
, lr
);
499 } else if (lr
->lr_refcnt
!= 0) {
500 panic("%s: non-zero refcount lr=%p", __func__
, lr
);
502 } else if (lr
->lr_reqcnt
!= 0) {
503 panic("%s: non-zero reqcnt lr=%p", __func__
, lr
);
506 lr
->lr_debug
&= ~IFD_ALLOC
;
509 lck_mtx_destroy(&lr
->lr_lock
, ifnet_lock_group
);
510 zfree(iflr_zone
, lr
);
514 iflr_addref(struct if_llreach
*lr
, int locked
)
519 IFLR_LOCK_ASSERT_HELD(lr
);
521 if (++lr
->lr_refcnt
== 0) {
522 panic("%s: lr=%p wraparound refcnt", __func__
, lr
);
530 iflr_remref(struct if_llreach
*lr
)
533 if (lr
->lr_refcnt
== 0) {
534 panic("%s: lr=%p negative refcnt", __func__
, lr
);
538 if (lr
->lr_refcnt
> 0) {
544 iflr_free(lr
); /* deallocate it */
548 ifnet_lr2ri(struct if_llreach
*lr
, struct rt_reach_info
*ri
)
550 struct if_llreach_info lri
;
552 IFLR_LOCK_ASSERT_HELD(lr
);
554 bzero(ri
, sizeof (*ri
));
555 ifnet_lr2lri(lr
, &lri
);
556 ri
->ri_refcnt
= lri
.lri_refcnt
;
557 ri
->ri_probes
= lri
.lri_probes
;
558 ri
->ri_rcv_expire
= lri
.lri_expire
;
559 ri
->ri_rssi
= lri
.lri_rssi
;
560 ri
->ri_lqm
= lri
.lri_lqm
;
561 ri
->ri_npm
= lri
.lri_npm
;
565 ifnet_lr2iflri(struct if_llreach
*lr
, struct ifnet_llreach_info
*iflri
)
567 IFLR_LOCK_ASSERT_HELD(lr
);
569 bzero(iflri
, sizeof (*iflri
));
571 * Note here we return request count, not actual memory refcnt.
573 iflri
->iflri_refcnt
= lr
->lr_reqcnt
;
574 iflri
->iflri_probes
= lr
->lr_probes
;
575 iflri
->iflri_rcv_expire
= ifnet_llreach_up2upexp(lr
, lr
->lr_lastrcvd
);
576 iflri
->iflri_curtime
= net_uptime();
577 switch (lr
->lr_key
.proto
) {
579 iflri
->iflri_netproto
= PF_INET
;
582 iflri
->iflri_netproto
= PF_INET6
;
586 * This shouldn't be possible for the time being,
587 * since link-layer reachability records are only
588 * kept for ARP and ND6.
590 iflri
->iflri_netproto
= PF_UNSPEC
;
593 bcopy(&lr
->lr_key
.addr
, &iflri
->iflri_addr
, IF_LLREACH_MAXLEN
);
594 iflri
->iflri_rssi
= lr
->lr_rssi
;
595 iflri
->iflri_lqm
= lr
->lr_lqm
;
596 iflri
->iflri_npm
= lr
->lr_npm
;
600 ifnet_lr2lri(struct if_llreach
*lr
, struct if_llreach_info
*lri
)
602 IFLR_LOCK_ASSERT_HELD(lr
);
604 bzero(lri
, sizeof (*lri
));
606 * Note here we return request count, not actual memory refcnt.
608 lri
->lri_refcnt
= lr
->lr_reqcnt
;
609 lri
->lri_ifindex
= lr
->lr_ifp
->if_index
;
610 lri
->lri_probes
= lr
->lr_probes
;
611 lri
->lri_expire
= ifnet_llreach_up2calexp(lr
, lr
->lr_lastrcvd
);
612 lri
->lri_proto
= lr
->lr_key
.proto
;
613 bcopy(&lr
->lr_key
.addr
, &lri
->lri_addr
, IF_LLREACH_MAXLEN
);
614 lri
->lri_rssi
= lr
->lr_rssi
;
615 lri
->lri_lqm
= lr
->lr_lqm
;
616 lri
->lri_npm
= lr
->lr_npm
;
620 sysctl_llreach_ifinfo SYSCTL_HANDLER_ARGS
623 int *name
, retval
= 0;
624 unsigned int namelen
;
626 struct if_llreach
*lr
;
627 struct if_llreach_info lri
;
631 namelen
= (unsigned int)arg2
;
633 if (req
->newptr
!= USER_ADDR_NULL
)
640 ifnet_head_lock_shared();
641 if (ifindex
<= 0 || ifindex
> (u_int
)if_index
) {
642 printf("%s: ifindex %u out of range\n", __func__
, ifindex
);
647 ifp
= ifindex2ifnet
[ifindex
];
650 printf("%s: no ifp for ifindex %u\n", __func__
, ifindex
);
654 lck_rw_lock_shared(&ifp
->if_llreach_lock
);
655 RB_FOREACH(lr
, ll_reach_tree
, &ifp
->if_ll_srcs
) {
656 /* Export to if_llreach_info structure */
658 ifnet_lr2lri(lr
, &lri
);
661 if ((retval
= SYSCTL_OUT(req
, &lri
, sizeof (lri
))) != 0)
664 lck_rw_done(&ifp
->if_llreach_lock
);