d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

* Add llt_clear_entry() callback which is able to do all lle

Browse Source 
cleanup including unlinking/freeing
* Relax locking in lltable_prefix_free_af/lltable_free
* Do not pass @llt to lle free callback: it is always NULL now.
* Unify arptimer/nd6_llinfo_timer: explicitly unlock lle avoiding
   unlock/lock sequinces
* Do not pass unlocked lle to nd6_ns_output(): add nd6_llinfo_get_holdsrc()
   to retrieve preferred source address from lle hold queue and pass it
   instead of lle.
* Finally, make nd6_create() create and return unlocked lle
* Separate defrtr handling code from nd6_free():
   use nd6_check_del_defrtr() to check if we need to keep entry instead of
    performing GC,
   use nd6_check_recalc_defrtr() to perform actual recalc on lle removal.
* Move isRouter handling from nd6_cache_lladdr() to separate
   nd6_check_router()
* Add initial code to maintain lle runtime flags in sync.
This commit is contained in:
Alexander V. Chernikov 2014-12-07 15:42:46 +00:00
parent a6e934e359
commit a743ccd468
10 changed files with 555 additions and 368 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
sys/net/if_llatbl.c
View File

@ -254,17 +254,17 @@ lltable_free(struct lltable *llt)
for (i = 0; i < LLTBL_HASHTBL_SIZE; i++) { for (i = 0; i < LLTBL_HASHTBL_SIZE; i++) {
LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) { LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) {
LLE_WLOCK(lle); LLE_WLOCK(lle);
llt->llt_stop_timers(lle);
LIST_INSERT_HEAD(&dchain, lle, lle_chain); LIST_INSERT_HEAD(&dchain, lle, lle_chain);
} }
} }
IF_AFDATA_RUN_WLOCK(llt->llt_ifp); IF_AFDATA_RUN_WLOCK(llt->llt_ifp);
llentries_unlink(&dchain); llentries_unlink(&dchain);
IF_AFDATA_RUN_WUNLOCK(llt->llt_ifp); IF_AFDATA_RUN_WUNLOCK(llt->llt_ifp);
LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next)
llentry_free(lle);
IF_AFDATA_CFG_WUNLOCK(llt->llt_ifp); IF_AFDATA_CFG_WUNLOCK(llt->llt_ifp);
LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next)
llt->llt_clear_entry(llt, lle);
free(llt, M_LLTABLE); free(llt, M_LLTABLE);
} }
@ -282,7 +282,6 @@ lltable_prefix_free_af(struct lltable *llt, const struct sockaddr *prefix,
LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) { LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) {
if (llt->llt_match_prefix(prefix, mask, flags, lle)) { if (llt->llt_match_prefix(prefix, mask, flags, lle)) {
LLE_WLOCK(lle); LLE_WLOCK(lle);
llt->llt_stop_timers(lle);
LIST_INSERT_HEAD(&dchain, lle, lle_chain); LIST_INSERT_HEAD(&dchain, lle, lle_chain);
} }
} }
@ -290,9 +289,10 @@ lltable_prefix_free_af(struct lltable *llt, const struct sockaddr *prefix,
IF_AFDATA_RUN_WLOCK(llt->llt_ifp); IF_AFDATA_RUN_WLOCK(llt->llt_ifp);
llentries_unlink(&dchain); llentries_unlink(&dchain);
IF_AFDATA_RUN_WUNLOCK(llt->llt_ifp); IF_AFDATA_RUN_WUNLOCK(llt->llt_ifp);
LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next)
llentry_free(lle);
IF_AFDATA_CFG_WUNLOCK(llt->llt_ifp); IF_AFDATA_CFG_WUNLOCK(llt->llt_ifp);
LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next)
llt->llt_clear_entry(llt, lle);
} }
#if 0 #if 0

8
sys/net/if_llatbl.h
View File

@ -68,7 +68,7 @@ struct llentry {
/* FIELDS PROTECTED BY LLE rwlock */ /* FIELDS PROTECTED BY LLE rwlock */
struct lltable *lle_tbl; struct lltable *lle_tbl;
struct llentries *lle_head; struct llentries *lle_head;
void (*lle_free)(struct lltable *, struct llentry *); void (*lle_free)(struct llentry *);
struct mbuf *la_hold; struct mbuf *la_hold;
int la_numheld; /* # of packets currently held */ int la_numheld; /* # of packets currently held */
time_t la_expire; time_t la_expire;
@ -117,7 +117,7 @@ struct llentry {
#define LLE_FREE_LOCKED(lle) do { \ #define LLE_FREE_LOCKED(lle) do { \
if ((lle)->lle_refcnt == 1) \ if ((lle)->lle_refcnt == 1) \
(lle)->lle_free((lle)->lle_tbl, (lle)); \ (lle)->lle_free((lle)); \
else { \ else { \
LLE_REMREF(lle); \ LLE_REMREF(lle); \
LLE_WUNLOCK(lle); \ LLE_WUNLOCK(lle); \
@ -158,7 +158,7 @@ typedef int (llt_dump_entry_t)(struct lltable *, struct llentry *,
typedef uint32_t (llt_hash_t)(const struct llentry *); typedef uint32_t (llt_hash_t)(const struct llentry *);
typedef int (llt_match_prefix_t)(const struct sockaddr *, typedef int (llt_match_prefix_t)(const struct sockaddr *,
const struct sockaddr *, u_int, struct llentry *); const struct sockaddr *, u_int, struct llentry *);
typedef void (llt_stop_timers_t)(struct llentry *lle); typedef void (llt_clear_entry_t)(struct lltable *, struct llentry *);
struct lltable { struct lltable {
SLIST_ENTRY(lltable) llt_link; SLIST_ENTRY(lltable) llt_link;
@ -172,7 +172,7 @@ struct lltable {
llt_dump_entry_t *llt_dump_entry; llt_dump_entry_t *llt_dump_entry;
llt_hash_t *llt_hash; llt_hash_t *llt_hash;
llt_match_prefix_t *llt_match_prefix; llt_match_prefix_t *llt_match_prefix;
llt_stop_timers_t *llt_stop_timers; llt_clear_entry_t *llt_clear_entry;
}; };
MALLOC_DECLARE(M_LLTABLE); MALLOC_DECLARE(M_LLTABLE);

125
sys/netinet/if_ether.c
View File

@ -182,43 +182,33 @@ static void
arptimer(void *arg) arptimer(void *arg)
{ {
struct llentry *lle = (struct llentry *)arg; struct llentry *lle = (struct llentry *)arg;
struct lltable *llt;
struct ifnet *ifp; struct ifnet *ifp;
size_t pkts_dropped; int evt;
uint16_t la_flags;
int state;
if (lle->la_flags & LLE_STATIC) { if (lle->la_flags & LLE_STATIC) {
/* TODO: ensure we won't get here */
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
return; return;
} }
if (lle->la_falgs & LLE_DELETED) { if (lle->la_flags & LLE_DELETED) {
/* XXX: Temporary */
/* We have been deleted. Drop callref and return */ /* We have been deleted. Drop callref and return */
if ((lle->la_flags & LLE_CALLOUTREF) != 0) { KASSERT((lle->la_flags & LLE_CALLOUTREF) != 0,
LLE_REMREF(lle); ("arptimer was called without callout reference"));
lle->la_flags &= ~LLE_CALLOUTREF;
}
pkts_dropped = llentry_free(lle); /* Assume the entry was already cleared */
ARPSTAT_ADD(dropped, pkts_dropped); lle->la_flags &= ~LLE_CALLOUTREF;
LLE_FREE_LOCKED(lle);
return; return;
} }
/* Unlink entry */ llt = lle->lle_tbl;
IF_AFDATA_RUN_WLOCK(ifp); ifp = llt->llt_ifp;
llentry_unlink(lle);
IF_AFDATA_RUN_WUNLOCK(ifp);
pkts_dropped = llentry_free(lle);
ARPSTAT_ADD(dropped, pkts_dropped);
la_flags = lle->la_flags;
state = (la_flags & LLE_DELETED) ? ARP_LLINFO_DELETED : lle->ln_state;
ifp = lle->lle_tbl->llt_ifp;
CURVNET_SET(ifp->if_vnet); CURVNET_SET(ifp->if_vnet);
switch (state) { switch (lle->ln_state) {
case ARP_LLINFO_REACHABLE: case ARP_LLINFO_REACHABLE:
/* /*
@ -234,6 +224,7 @@ arptimer(void *arg)
lle->ln_state = ARP_LLINFO_VERIFY; lle->ln_state = ARP_LLINFO_VERIFY;
callout_schedule(&lle->la_timer, hz * V_arpt_rexmit); callout_schedule(&lle->la_timer, hz * V_arpt_rexmit);
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
CURVNET_RESTORE();
return; return;
case ARP_LLINFO_VERIFY: case ARP_LLINFO_VERIFY:
if (lle->r_kick == 0 && lle->la_preempt > 0) { if (lle->r_kick == 0 && lle->la_preempt > 0) {
@ -245,62 +236,80 @@ arptimer(void *arg)
lle->r_kick = 1; lle->r_kick = 1;
callout_schedule(&lle->la_timer, hz * V_arpt_rexmit); callout_schedule(&lle->la_timer, hz * V_arpt_rexmit);
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
CURVNET_RESTORE();
return; return;
} }
/* Nothing happened. Reschedule if not too late */ /* Nothing happened. Reschedule if not too late */
if (lle->la_expire > time_uptime) { if (lle->la_expire > time_uptime) {
callout_schedule(&lle->la_timer, hz * V_arpt_rexmit); callout_schedule(&lle->la_timer, hz * V_arpt_rexmit);
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
CURVNET_RESTORE();
return; return;
} }
break; break;
case ARP_LLINFO_INCOMPLETE: case ARP_LLINFO_INCOMPLETE:
case ARP_LLINFO_DELETED:
break; break;
} }
if ((lle->la_flags & LLE_DELETED) == 0) { /* We have to delete entr */
int evt; if (lle->la_flags & LLE_VALID)
evt = LLENTRY_EXPIRED;
else
evt = LLENTRY_TIMEDOUT;
EVENTHANDLER_INVOKE(lle_event, lle, evt);
if (lle->la_flags & LLE_VALID) llt->llt_clear_entry(llt, lle);
evt = LLENTRY_EXPIRED;
else
evt = LLENTRY_TIMEDOUT;
EVENTHANDLER_INVOKE(lle_event, lle, evt);
}
callout_stop(&lle->la_timer);
/* XXX: LOR avoidance. We still have ref on lle. */
LLE_WUNLOCK(lle);
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(lle);
/*
* Note other thread could have removed given entry
* stopping callout and removing LLE reference.
*/
//llt->llt_stop_timers(lle);
if ((lle->la_flags & LLE_CALLOUTREF) != 0) {
LLE_REMREF(lle);
lle->la_flags &= ~LLE_CALLOUTREF;
}
/* Unlink entry */
IF_AFDATA_RUN_WLOCK(ifp);
llentry_unlink(lle);
IF_AFDATA_RUN_WUNLOCK(ifp);
pkts_dropped = llentry_free(lle);
ARPSTAT_ADD(dropped, pkts_dropped);
IF_AFDATA_CFG_WUNLOCK(ifp);
ARPSTAT_INC(timeouts); ARPSTAT_INC(timeouts);
CURVNET_RESTORE(); CURVNET_RESTORE();
} }
/*
* Calback for lltable.
*/
void
arp_lltable_clear_entry(struct lltable *llt, struct llentry *lle)
{
struct ifnet *ifp;
size_t pkts_dropped;
LLE_WLOCK_ASSERT(lle);
KASSERT(llt != NULL, ("lltable is NULL"));
/* Unlink entry from table if not already */
if ((lle->la_flags & LLE_LINKED) != 0) {
ifp = llt->llt_ifp;
/*
* Lock order needs to be maintained
*/
LLE_ADDREF(lle);
LLE_WUNLOCK(lle);
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(lle);
LLE_REMREF(lle);
IF_AFDATA_RUN_WLOCK(ifp);
llentry_unlink(lle);
IF_AFDATA_RUN_WUNLOCK(ifp);
IF_AFDATA_CFG_WUNLOCK(ifp);
}
/* cancel timer */
if (callout_stop(&lle->la_timer) != 0) {
if ((lle->la_flags & LLE_CALLOUTREF) != 0) {
LLE_REMREF(lle);
lle->la_flags &= ~LLE_CALLOUTREF;
}
}
/* Finally, free entry */
pkts_dropped = llentry_free(lle);
ARPSTAT_ADD(dropped, pkts_dropped);
}
/* /*
* Broadcast an ARP request. Caller specifies: * Broadcast an ARP request. Caller specifies:
* - arp header source ip address * - arp header source ip address

2
sys/netinet/if_ether.h
View File

@ -112,6 +112,7 @@ struct sockaddr_inarp {
extern u_char ether_ipmulticast_min[ETHER_ADDR_LEN]; extern u_char ether_ipmulticast_min[ETHER_ADDR_LEN];
extern u_char ether_ipmulticast_max[ETHER_ADDR_LEN]; extern u_char ether_ipmulticast_max[ETHER_ADDR_LEN];
struct lltable;
struct llentry; struct llentry;
struct ifaddr; struct ifaddr;
@ -124,6 +125,7 @@ void arprequest(struct ifnet *, const struct in_addr *,
void arp_ifinit(struct ifnet *, struct ifaddr *); void arp_ifinit(struct ifnet *, struct ifaddr *);
void arp_ifinit2(struct ifnet *, struct ifaddr *, u_char *); void arp_ifinit2(struct ifnet *, struct ifaddr *, u_char *);
void arp_ifscrub(struct ifnet *, uint32_t); void arp_ifscrub(struct ifnet *, uint32_t);
void arp_lltable_clear_entry(struct lltable *, struct llentry *);
#endif #endif
#endif #endif

18
sys/netinet/in.c
View File

@ -994,14 +994,15 @@ struct in_llentry {
}; };
/* /*
* Deletes an address from the address table. * Frees unlinked record.
* This function is called by the timer functions * This function is called by the timer functions
* such as arptimer() and nd6_llinfo_timer(), and * such as arptimer() and nd6_llinfo_timer(), and
* the caller does the locking. * the caller does the locking.
*/ */
static void static void
in_lltable_free(struct lltable *llt, struct llentry *lle) in_lltable_free(struct llentry *lle)
{ {
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
LLE_LOCK_DESTROY(lle); LLE_LOCK_DESTROY(lle);
free(lle, M_LLTABLE); free(lle, M_LLTABLE);
@ -1035,17 +1036,6 @@ in_lltable_new(const struct sockaddr *l3addr, u_int flags)
return (&lle->base); return (&lle->base);
} }
static void
in_lltable_stop_timers(struct llentry *lle)
{
LLE_WLOCK_ASSERT(lle);
if (callout_stop(&lle->la_timer)) {
LLE_REMREF(lle);
lle->la_flags &= ~LLE_CALLOUTREF;
}
}
#define IN_ARE_MASKED_ADDR_EQUAL(d, a, m) ( \ #define IN_ARE_MASKED_ADDR_EQUAL(d, a, m) ( \
(((ntohl((d)->sin_addr.s_addr) ^ (a)->sin_addr.s_addr) & (m)->sin_addr.s_addr)) == 0 ) (((ntohl((d)->sin_addr.s_addr) ^ (a)->sin_addr.s_addr) & (m)->sin_addr.s_addr)) == 0 )
@ -1314,7 +1304,7 @@ in_domifattach(struct ifnet *ifp)
llt->llt_delete = in_lltable_delete; llt->llt_delete = in_lltable_delete;
llt->llt_dump_entry = in_lltable_dump_entry; llt->llt_dump_entry = in_lltable_dump_entry;
llt->llt_hash = in_lltable_hash; llt->llt_hash = in_lltable_hash;
llt->llt_stop_timers = in_lltable_stop_timers; llt->llt_clear_entry = arp_lltable_clear_entry;
llt->llt_match_prefix = in_lltable_match_prefix; llt->llt_match_prefix = in_lltable_match_prefix;
} }
ii->ii_llt = llt; ii->ii_llt = llt;

38
sys/netinet/toecore.c
View File

@ -37,6 +37,8 @@ __FBSDID("$FreeBSD$");
#include <sys/mbuf.h> #include <sys/mbuf.h>
#include <sys/module.h> #include <sys/module.h>
#include <sys/types.h> #include <sys/types.h>
#include <sys/lock.h>
#include <sys/rmlock.h>
#include <sys/sockopt.h> #include <sys/sockopt.h>
#include <sys/sysctl.h> #include <sys/sysctl.h>
#include <sys/socket.h> #include <sys/socket.h>
@ -453,7 +455,7 @@ toe_route_redirect_event(void *arg __unused, struct nhopu_extended *nh0,
static int static int
toe_nd6_resolve(struct ifnet *ifp, struct sockaddr *sa, uint8_t *lladdr) toe_nd6_resolve(struct ifnet *ifp, struct sockaddr *sa, uint8_t *lladdr)
{ {
struct llentry *lle; struct llentry *lle, *lle_tmp;
struct sockaddr_in6 *sin6 = (void *)sa; struct sockaddr_in6 *sin6 = (void *)sa;
int rc, flags = 0; int rc, flags = 0;
@ -462,19 +464,39 @@ restart:
lle = lltable_lookup_lle(LLTABLE6(ifp), flags, sa); lle = lltable_lookup_lle(LLTABLE6(ifp), flags, sa);
IF_AFDATA_RUNLOCK(ifp); IF_AFDATA_RUNLOCK(ifp);
if (lle == NULL) { if (lle == NULL) {
IF_AFDATA_CFG_WLOCK(ifp);
lle = nd6_create(&sin6->sin6_addr, 0, ifp); lle = nd6_create(&sin6->sin6_addr, 0, ifp);
IF_AFDATA_CFG_WUNLOCK(ifp);
if (lle == NULL) if (lle == NULL)
return (ENOMEM); /* Couldn't create entry in cache. */ return (ENOMEM); /* Couldn't create entry in cache. */
lle->ln_state = ND6_LLINFO_INCOMPLETE; lle->ln_state = ND6_LLINFO_INCOMPLETE;
nd6_llinfo_settimer_locked(lle, IF_AFDATA_CFG_WLOCK(ifp);
(long)ND_IFINFO(ifp)->retrans * hz / 1000); LLE_WLOCK(lle);
LLE_WUNLOCK(lle); /* Check if the same record was addded */
lle_tmp = lltable_lookup_lle(LLTABLE6(ifp), LLE_EXCLUSIVE, sa);
if (lle_tmp == NULL) {
/*
* No entry has been found. Link new one.
*/
IF_AFDATA_RUN_WLOCK(ifp);
llentry_link(LLTABLE6(ifp), lle);
IF_AFDATA_RUN_WUNLOCK(ifp);
}
IF_AFDATA_CFG_WUNLOCK(ifp);
if (lle_tmp == NULL) {
/* Set up timer for our new lle */
nd6_llinfo_settimer_locked(lle,
(long)ND_IFINFO(ifp)->retrans * hz / 1000);
LLE_WUNLOCK(lle);
nd6_ns_output(ifp, NULL, &sin6->sin6_addr, NULL, 0); nd6_ns_output(ifp, NULL, &sin6->sin6_addr, NULL, 0);
return (EWOULDBLOCK); return (EWOULDBLOCK);
}
/* Existing lle has been found. Free new one */
LLE_FREE_LOCKED(lle);
lle = lle_tmp;
lle_tmp = NULL;
flags |= LLE_EXCLUSIVE;
} }
if (lle->ln_state == ND6_LLINFO_STALE) { if (lle->ln_state == ND6_LLINFO_STALE) {

30
sys/netinet6/in6.c
View File

@ -2052,14 +2052,12 @@ struct in6_llentry {
}; };
/* /*
* Deletes an address from the address table. * Frees already unlinked @lle.
* This function is called by the timer functions
* such as arptimer() and nd6_llinfo_timer(), and
* the caller does the locking.
*/ */
static void static void
in6_lltable_free(struct lltable *llt, struct llentry *lle) in6_lltable_free(struct llentry *lle)
{ {
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
LLE_LOCK_DESTROY(lle); LLE_LOCK_DESTROY(lle);
free(lle, M_LLTABLE); free(lle, M_LLTABLE);
@ -2087,17 +2085,6 @@ in6_lltable_new(const struct sockaddr *l3addr, u_int flags)
return (&lle->base); return (&lle->base);
} }
static void
in6_lltable_stop_timers(struct llentry *lle)
{
LLE_WLOCK_ASSERT(lle);
if (callout_stop(&lle->la_timer)) {
LLE_REMREF(lle);
lle->la_flags &= ~LLE_CALLOUTREF;
}
}
static int static int
in6_lltable_match_prefix(const struct sockaddr *prefix, in6_lltable_match_prefix(const struct sockaddr *prefix,
const struct sockaddr *mask, u_int flags, struct llentry *lle) const struct sockaddr *mask, u_int flags, struct llentry *lle)
@ -2219,19 +2206,13 @@ static struct llentry *
in6_lltable_create(struct lltable *llt, u_int flags, in6_lltable_create(struct lltable *llt, u_int flags,
const struct sockaddr *l3addr) const struct sockaddr *l3addr)
{ {
const struct sockaddr_in6 *sin6 = (const struct sockaddr_in6 *)l3addr;
struct ifnet *ifp = llt->llt_ifp; struct ifnet *ifp = llt->llt_ifp;
struct llentry *lle; struct llentry *lle;
KASSERT(l3addr->sa_family == AF_INET6, KASSERT(l3addr->sa_family == AF_INET6,
("sin_family %d", l3addr->sa_family)); ("sin_family %d", l3addr->sa_family));
lle = in6_lltable_find_dst(llt, &sin6->sin6_addr); IF_AFDATA_CFG_UNLOCK_ASSERT(ifp);
if (lle != NULL) {
LLE_WLOCK(lle);
return (lle);
}
/* /*
* A route that covers the given address must have * A route that covers the given address must have
@ -2248,7 +2229,6 @@ in6_lltable_create(struct lltable *llt, u_int flags,
return NULL; return NULL;
} }
lle->la_flags = flags; lle->la_flags = flags;
LLE_WLOCK(lle);
return (lle); return (lle);
} }
@ -2381,7 +2361,7 @@ in6_domifattach(struct ifnet *ifp)
ext->lltable->llt_delete = in6_lltable_delete; ext->lltable->llt_delete = in6_lltable_delete;
ext->lltable->llt_dump_entry = in6_lltable_dump_entry; ext->lltable->llt_dump_entry = in6_lltable_dump_entry;
ext->lltable->llt_hash = in6_lltable_hash; ext->lltable->llt_hash = in6_lltable_hash;
ext->lltable->llt_stop_timers = in6_lltable_stop_timers; ext->lltable->llt_clear_entry = nd6_lltable_clear_entry;
ext->lltable->llt_match_prefix = in6_lltable_match_prefix; ext->lltable->llt_match_prefix = in6_lltable_match_prefix;
} }

658
sys/netinet6/nd6.c
View File

@ -132,10 +132,14 @@ static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *,
static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
static void nd6_slowtimo(void *); static void nd6_slowtimo(void *);
static int regen_tmpaddr(struct in6_ifaddr *); static int regen_tmpaddr(struct in6_ifaddr *);
static struct llentry *nd6_free(struct llentry *, int); static void nd6_free(struct llentry *, int);
static void nd6_llinfo_timer(void *); static void nd6_llinfo_timer(void *);
static void clear_llinfo_pqueue(struct llentry *); static void clear_llinfo_pqueue(struct llentry *);
static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *); static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
static struct in6_addr *nd6_llinfo_get_holdsrc(struct llentry *,
struct in6_addr *);
static int nd6_check_del_defrtr(struct lltable *, struct llentry *);
static void nd6_check_recalc_defrtr(struct lltable *, struct llentry *);
static VNET_DEFINE(struct callout, nd6_slowtimo_ch); static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
#define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
@ -433,6 +437,10 @@ nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
ln->la_expire = 0; ln->la_expire = 0;
ln->ln_ntick = 0; ln->ln_ntick = 0;
canceled = callout_stop(&ln->ln_timer_ch); canceled = callout_stop(&ln->ln_timer_ch);
if (canceled != 0) {
ln->la_flags &= ~LLE_CALLOUTREF;
LLE_REMREF(ln);
}
} else { } else {
ln->la_expire = time_uptime + tick / hz; ln->la_expire = time_uptime + tick / hz;
LLE_ADDREF(ln); LLE_ADDREF(ln);
@ -445,11 +453,11 @@ nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
canceled = callout_reset(&ln->ln_timer_ch, tick, canceled = callout_reset(&ln->ln_timer_ch, tick,
nd6_llinfo_timer, ln); nd6_llinfo_timer, ln);
} }
if (canceled != 0)
LLE_REMREF(ln);
else
ln->la_flags |= LLE_CALLOUTREF;
} }
if (canceled)
LLE_REMREF(ln);
else
ln->la_flags |= LLE_CALLOUTREF;
} }
void void
@ -461,6 +469,34 @@ nd6_llinfo_settimer(struct llentry *ln, long tick)
LLE_WUNLOCK(ln); LLE_WUNLOCK(ln);
} }
/*
* Gets source address of the first packet in hold queue
* and stores it in @src.
* Returns pointer to @src (if hold queue is not empty) or NULL.
*
*/
static struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
{
struct ip6_hdr *phdr;
if (ln->la_hold == NULL)
return (NULL);
/*
* assuming every packet in la_hold has the same IP
* header
*/
phdr = mtod(ln->la_hold, struct ip6_hdr *);
/* XXX pullup? */
if (sizeof(*phdr) < ln->la_hold->m_len) {
*src = phdr->ip6_src;
return (src);
}
return (NULL);
}
static void static void
nd6_llinfo_timer(void *arg) nd6_llinfo_timer(void *arg)
{ {
@ -468,13 +504,28 @@ nd6_llinfo_timer(void *arg)
struct in6_addr *dst; struct in6_addr *dst;
struct ifnet *ifp; struct ifnet *ifp;
struct nd_ifinfo *ndi = NULL; struct nd_ifinfo *ndi = NULL;
struct in6_addr src, *psrc;
KASSERT(arg != NULL, ("%s: arg NULL", __func__)); KASSERT(arg != NULL, ("%s: arg NULL", __func__));
ln = (struct llentry *)arg; ln = (struct llentry *)arg;
LLE_WLOCK_ASSERT(ln); LLE_WLOCK_ASSERT(ln);
ifp = ln->lle_tbl->llt_ifp;
CURVNET_SET(ifp->if_vnet); if (ln->la_flags & LLE_STATIC) {
/* TODO: ensure we won't get here */
LLE_WUNLOCK(ln);
return;
}
if (ln->la_flags & LLE_DELETED) {
/* We have been deleted. Drop callref and return */
KASSERT((ln->la_flags & LLE_CALLOUTREF) != 0,
("nd6_llinfo_timer was called without callout reference"));
/* Assume the entry was already cleared */
ln->la_flags &= ~LLE_CALLOUTREF;
LLE_FREE_LOCKED(ln);
return;
}
if (ln->ln_ntick > 0) { if (ln->ln_ntick > 0) {
if (ln->ln_ntick > INT_MAX) { if (ln->ln_ntick > INT_MAX) {
@ -484,29 +535,27 @@ nd6_llinfo_timer(void *arg)
ln->ln_ntick = 0; ln->ln_ntick = 0;
nd6_llinfo_settimer_locked(ln, ln->ln_ntick); nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
} }
goto done; LLE_WUNLOCK(ln);
return;
} }
ifp = ln->lle_tbl->llt_ifp;
CURVNET_SET(ifp->if_vnet);
ndi = ND_IFINFO(ifp); ndi = ND_IFINFO(ifp);
dst = &L3_ADDR_SIN6(ln)->sin6_addr; dst = &L3_ADDR_SIN6(ln)->sin6_addr;
if (ln->la_flags & LLE_STATIC) {
goto done;
}
if (ln->la_flags & LLE_DELETED) {
(void)nd6_free(ln, 0);
ln = NULL;
goto done;
}
/*
* Each case statement needs to unlock @ln before break/return.
*/
switch (ln->ln_state) { switch (ln->ln_state) {
case ND6_LLINFO_INCOMPLETE: case ND6_LLINFO_INCOMPLETE:
if (ln->la_asked < V_nd6_mmaxtries) { if (ln->la_asked < V_nd6_mmaxtries) {
ln->la_asked++; ln->la_asked++;
nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
psrc = nd6_llinfo_get_holdsrc(ln, &src);
LLE_WUNLOCK(ln); LLE_WUNLOCK(ln);
nd6_ns_output(ifp, NULL, dst, ln, 0); nd6_ns_output(ifp, NULL, dst, psrc, 0);
LLE_WLOCK(ln);
} else { } else {
struct mbuf *m = ln->la_hold; struct mbuf *m = ln->la_hold;
if (m) { if (m) {
@ -522,7 +571,7 @@ nd6_llinfo_timer(void *arg)
clear_llinfo_pqueue(ln); clear_llinfo_pqueue(ln);
} }
EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
(void)nd6_free(ln, 0); nd6_free(ln, 0);
ln = NULL; ln = NULL;
if (m != NULL) if (m != NULL)
icmp6_error2(m, ICMP6_DST_UNREACH, icmp6_error2(m, ICMP6_DST_UNREACH,
@ -534,15 +583,17 @@ nd6_llinfo_timer(void *arg)
ln->ln_state = ND6_LLINFO_STALE; ln->ln_state = ND6_LLINFO_STALE;
nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
} }
LLE_WUNLOCK(ln);
break; break;
case ND6_LLINFO_STALE: case ND6_LLINFO_STALE:
/* Garbage Collection(RFC 2461 5.3) */ /* Garbage Collection(RFC 2461 5.3) */
if (!ND6_LLINFO_PERMANENT(ln)) { if (!ND6_LLINFO_PERMANENT(ln)) {
EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
(void)nd6_free(ln, 1); nd6_free(ln, 1);
ln = NULL; ln = NULL;
} }
LLE_WUNLOCK(ln);
break; break;
case ND6_LLINFO_DELAY: case ND6_LLINFO_DELAY:
@ -551,24 +602,25 @@ nd6_llinfo_timer(void *arg)
ln->la_asked = 1; ln->la_asked = 1;
ln->ln_state = ND6_LLINFO_PROBE; ln->ln_state = ND6_LLINFO_PROBE;
nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
psrc = nd6_llinfo_get_holdsrc(ln, &src);
LLE_WUNLOCK(ln); LLE_WUNLOCK(ln);
nd6_ns_output(ifp, dst, dst, ln, 0); nd6_ns_output(ifp, dst, dst, psrc, 0);
LLE_WLOCK(ln);
} else { } else {
ln->ln_state = ND6_LLINFO_STALE; /* XXX */ ln->ln_state = ND6_LLINFO_STALE; /* XXX */
nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
LLE_WUNLOCK(ln);
} }
break; break;
case ND6_LLINFO_PROBE: case ND6_LLINFO_PROBE:
if (ln->la_asked < V_nd6_umaxtries) { if (ln->la_asked < V_nd6_umaxtries) {
ln->la_asked++; ln->la_asked++;
nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
psrc = nd6_llinfo_get_holdsrc(ln, &src);
LLE_WUNLOCK(ln); LLE_WUNLOCK(ln);
nd6_ns_output(ifp, dst, dst, ln, 0); nd6_ns_output(ifp, dst, dst, psrc, 0);
LLE_WLOCK(ln);
} else { } else {
EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
(void)nd6_free(ln, 0); nd6_free(ln, 0);
ln = NULL; ln = NULL;
} }
break; break;
@ -576,9 +628,7 @@ nd6_llinfo_timer(void *arg)
panic("%s: paths in a dark night can be confusing: %d", panic("%s: paths in a dark night can be confusing: %d",
__func__, ln->ln_state); __func__, ln->ln_state);
} }
done:
if (ln != NULL)
LLE_FREE_LOCKED(ln);
CURVNET_RESTORE(); CURVNET_RESTORE();
} }
@ -866,9 +916,8 @@ nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
return (ln); return (ln);
} }
/* /*
* the caller acquires and releases the lock on the lltbls * Creates and returns new, unlinked and unlocked lle.
* Returns the llentry wlocked
*/ */
struct llentry * struct llentry *
nd6_create(struct in6_addr *addr6, int flags, struct ifnet *ifp) nd6_create(struct in6_addr *addr6, int flags, struct ifnet *ifp)
@ -881,15 +930,10 @@ nd6_create(struct in6_addr *addr6, int flags, struct ifnet *ifp)
sin6.sin6_family = AF_INET6; sin6.sin6_family = AF_INET6;
sin6.sin6_addr = *addr6; sin6.sin6_addr = *addr6;
IF_AFDATA_CFG_WLOCK_ASSERT(ifp); IF_AFDATA_CFG_UNLOCK_ASSERT(ifp);
ln = lltable_create_lle(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6); ln = lltable_create_lle(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
if (ln != NULL) { ln->ln_state = ND6_LLINFO_NOSTATE;
IF_AFDATA_RUN_WLOCK(ifp);
ln->ln_state = ND6_LLINFO_NOSTATE;
llentry_link(LLTABLE6(ifp), ln);
IF_AFDATA_RUN_WUNLOCK(ifp);
}
return (ln); return (ln);
} }
@ -1025,34 +1069,96 @@ nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
/* /*
* Free an nd6 llinfo entry. * Free an nd6 llinfo entry.
* Since the function would cause significant changes in the kernel, DO NOT * Internal function used by timer code.
* make it global, unless you have a strong reason for the change, and are sure
* that the change is safe.
*/ */
static struct llentry * static void
nd6_free(struct llentry *ln, int gc) nd6_free(struct llentry *ln, int gc)
{ {
struct llentry *next; struct lltable *llt;
struct nd_defrouter *dr;
struct ifnet *ifp;
LLE_WLOCK_ASSERT(ln); LLE_WLOCK_ASSERT(ln);
/* if ((ln->la_flags & LLE_DELETED) != 0) {
* we used to have pfctlinput(PRC_HOSTDEAD) here. /* Unlinked entry. Stop timer/callout. */
* even though it is not harmful, it was not really necessary. nd6_llinfo_settimer_locked(ln, -1);
*/ llentry_free(ln);
return;
}
llt = ln->lle_tbl;
/* Check if we can delete/unlink given entry */
if (gc != 0 && nd6_check_del_defrtr(llt, ln) == 0) {
LLE_WUNLOCK(ln);
return;
}
llt->llt_clear_entry(ln->lle_tbl, ln);
}
/*
* Calback for lltable.
*/
void
nd6_lltable_clear_entry(struct lltable *llt, struct llentry *ln)
{
struct ifnet *ifp;
LLE_WLOCK_ASSERT(ln);
KASSERT(llt != NULL, ("lltable is NULL"));
/* Unlink entry from table */
if ((ln->la_flags & LLE_LINKED) != 0) {
ifp = llt->llt_ifp;
/*
* Lock order needs to be maintained
*/
LLE_ADDREF(ln);
LLE_WUNLOCK(ln);
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(ln);
LLE_REMREF(ln);
IF_AFDATA_RUN_WLOCK(ifp);
llentry_unlink(ln);
IF_AFDATA_RUN_WUNLOCK(ifp);
IF_AFDATA_CFG_WUNLOCK(ifp);
}
/* cancel timer */ /* cancel timer */
nd6_llinfo_settimer_locked(ln, -1); nd6_llinfo_settimer_locked(ln, -1);
ifp = ln->lle_tbl->llt_ifp; /* Check if default router needs to be recalculated */
nd6_check_recalc_defrtr(llt, ln);
/* Finally, free entry */
llentry_free(ln);
}
/*
* Checks if we can delete given entry.
* Perfoms defrtr selection if needed.
*
* return non-zero value if lle can be deleted.
*/
static int
nd6_check_del_defrtr(struct lltable *llt, struct llentry *ln)
{
struct ifnet *ifp;
struct nd_defrouter *dr;
struct in6_addr dst;
ifp = llt->llt_ifp;
dst = L3_ADDR_SIN6(ln)->sin6_addr;
LLE_WLOCK_ASSERT(ln);
if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); dr = defrouter_lookup(&dst, ifp);
if (dr != NULL && dr->expire && if (dr != NULL && dr->expire &&
ln->ln_state == ND6_LLINFO_STALE && gc) { ln->ln_state == ND6_LLINFO_STALE) {
/* /*
* If the reason for the deletion is just garbage * If the reason for the deletion is just garbage
* collection, and the neighbor is an active default * collection, and the neighbor is an active default
@ -1072,11 +1178,27 @@ nd6_free(struct llentry *ln, int gc)
nd6_llinfo_settimer_locked(ln, nd6_llinfo_settimer_locked(ln,
(long)V_nd6_gctimer * hz); (long)V_nd6_gctimer * hz);
next = LIST_NEXT(ln, lle_next); return (0);
LLE_REMREF(ln);
LLE_WUNLOCK(ln);
return (next);
} }
}
return (1);
}
static void
nd6_check_recalc_defrtr(struct lltable *llt, struct llentry *ln)
{
struct ifnet *ifp;
struct nd_defrouter *dr;
struct in6_addr dst;
ifp = llt->llt_ifp;
dst = L3_ADDR_SIN6(ln)->sin6_addr;
LLE_WLOCK_ASSERT(ln);
if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
dr = defrouter_lookup(&dst, ifp);
if (dr) { if (dr) {
/* /*
@ -1103,6 +1225,7 @@ nd6_free(struct llentry *ln, int gc)
* defrouter_select() in the block further down for calls * defrouter_select() in the block further down for calls
* into nd6_lookup(). We still hold a ref. * into nd6_lookup(). We still hold a ref.
*/ */
LLE_ADDREF(ln);
LLE_WUNLOCK(ln); LLE_WUNLOCK(ln);
/* /*
@ -1110,7 +1233,7 @@ nd6_free(struct llentry *ln, int gc)
* is in the Default Router List. * is in the Default Router List.
* See a corresponding comment in nd6_na_input(). * See a corresponding comment in nd6_na_input().
*/ */
rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); rt6_flush(&dst, ifp);
} }
if (dr) { if (dr) {
@ -1128,44 +1251,11 @@ nd6_free(struct llentry *ln, int gc)
defrouter_select(); defrouter_select();
} }
if (ln->ln_router || dr) if (ln->ln_router || dr) {
LLE_WLOCK(ln); LLE_WLOCK(ln);
LLE_REMREF(ln);
}
} }
/*
* Before deleting the entry, remember the next entry as the
* return value. We need this because pfxlist_onlink_check() above
* might have freed other entries (particularly the old next entry) as
* a side effect (XXX).
*/
next = LIST_NEXT(ln, lle_next);
/*
* Save to unlock. We still hold an extra reference and will not
* free(9) in llentry_free() if someone else holds one as well.
*/
LLE_WUNLOCK(ln);
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(ln);
/*
* Note other thread could have removed given entry
* stopping callout and removing LLE reference.
*/
if ((ln->la_flags & LLE_CALLOUTREF) != 0) {
LLE_REMREF(ln);
ln->la_flags &= ~LLE_CALLOUTREF;
}
IF_AFDATA_RUN_WLOCK(ifp);
llentry_unlink(ln);
IF_AFDATA_RUN_WUNLOCK(ifp);
llentry_free(ln);
IF_AFDATA_CFG_WUNLOCK(ifp);
return (next);
} }
/* /*
@ -1564,6 +1654,79 @@ nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
return (error); return (error);
} }
static int
nd6_check_router(int type, int code, int is_new, int old_addr, int new_addr,
int ln_router)
{
/*
* ICMP6 type dependent behavior.
*
* NS: clear IsRouter if new entry
* RS: clear IsRouter
* RA: set IsRouter if there's lladdr
* redir: clear IsRouter if new entry
*
* RA case, (1):
* The spec says that we must set IsRouter in the following cases:
* - If lladdr exist, set IsRouter. This means (1-5).
* - If it is old entry (!newentry), set IsRouter. This means (7).
* So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
* A quetion arises for (1) case. (1) case has no lladdr in the
* neighbor cache, this is similar to (6).
* This case is rare but we figured that we MUST NOT set IsRouter.
*
* newentry olladdr lladdr llchange NS RS RA redir
* D R
* 0 n n -- (1) c ? s
* 0 y n -- (2) c s s
* 0 n y -- (3) c s s
* 0 y y n (4) c s s
* 0 y y y (5) c s s
* 1 -- n -- (6) c c c s
* 1 -- y -- (7) c c s c s
*
* (c=clear s=set)
*/
switch (type & 0xff) {
case ND_NEIGHBOR_SOLICIT:
/*
* New entry must have is_router flag cleared.
*/
if (is_new) /* (6-7) */
ln_router = 0;
break;
case ND_REDIRECT:
/*
* If the icmp is a redirect to a better router, always set the
* is_router flag. Otherwise, if the entry is newly created,
* clear the flag. [RFC 2461, sec 8.3]
*/
if (code == ND_REDIRECT_ROUTER)
ln_router = 1;
else if (is_new) /* (6-7) */
ln_router = 0;
break;
case ND_ROUTER_SOLICIT:
/*
* is_router flag must always be cleared.
*/
ln_router = 0;
break;
case ND_ROUTER_ADVERT:
/*
* Mark an entry with lladdr as a router.
*/
if ((!is_new && (old_addr || new_addr)) || /* (2-5) */
(is_new && new_addr)) { /* (7) */
ln_router = 1;
}
break;
}
return (ln_router);
}
/* /*
* Create neighbor cache entry and cache link-layer address, * Create neighbor cache entry and cache link-layer address,
* on reception of inbound ND6 packets. (RS/RA/NS/redirect) * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
@ -1584,12 +1747,13 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
int do_update; int do_update;
int olladdr; int olladdr;
int llchange; int llchange;
int flags;
int newstate = 0; int newstate = 0;
uint16_t router = 0; uint16_t router = 0;
struct sockaddr_in6 sin6; struct sockaddr_in6 sin6;
struct mbuf *chain = NULL; struct mbuf *chain = NULL;
int static_route = 0; int r_update;
int new_rvalid, old_rvalid;
struct llentry *ln_tmp;
IF_AFDATA_CFG_UNLOCK_ASSERT(ifp); IF_AFDATA_CFG_UNLOCK_ASSERT(ifp);
@ -1609,21 +1773,19 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
* Spec says nothing in sections for RA, RS and NA. There's small * Spec says nothing in sections for RA, RS and NA. There's small
* description on it in NS section (RFC 2461 7.2.3). * description on it in NS section (RFC 2461 7.2.3).
*/ */
flags = lladdr ? ND6_EXCLUSIVE : 0;
IF_AFDATA_CFG_RLOCK(ifp); IF_AFDATA_CFG_RLOCK(ifp);
ln = nd6_lookup(from, flags, ifp); ln = nd6_lookup(from, ND6_EXCLUSIVE, ifp);
IF_AFDATA_CFG_RUNLOCK(ifp); IF_AFDATA_CFG_RUNLOCK(ifp);
if (ln == NULL) { if (ln == NULL) {
flags |= ND6_EXCLUSIVE;
IF_AFDATA_CFG_WLOCK(ifp);
ln = nd6_create(from, 0, ifp); ln = nd6_create(from, 0, ifp);
IF_AFDATA_CFG_WUNLOCK(ifp); if (ln != NULL)
LLE_WLOCK(ln);
is_newentry = 1; is_newentry = 1;
} else { } else {
/* do nothing if static ndp is set */ /* do nothing if record is static */
if (ln->la_flags & LLE_STATIC) { if (ln->la_flags & LLE_STATIC) {
static_route = 1; LLE_WUNLOCK(ln);
goto done; return (NULL);
} }
is_newentry = 0; is_newentry = 0;
} }
@ -1637,26 +1799,6 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
} else } else
llchange = 0; llchange = 0;
/*
* newentry olladdr lladdr llchange (*=record)
* 0 n n -- (1)
* 0 y n -- (2)
* 0 n y -- (3) * STALE
* 0 y y n (4) *
* 0 y y y (5) * STALE
* 1 -- n -- (6) NOSTATE(= PASSIVE)
* 1 -- y -- (7) * STALE
*/
if (lladdr) { /* (3-5) and (7) */
/*
* Record source link-layer address
* XXX is it dependent to ifp->if_type?
*/
bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
ln->la_flags |= LLE_VALID;
EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
}
if (!is_newentry) { if (!is_newentry) {
if ((!olladdr && lladdr != NULL) || /* (3) */ if ((!olladdr && lladdr != NULL) || /* (3) */
@ -1673,6 +1815,108 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
newstate = ND6_LLINFO_STALE; newstate = ND6_LLINFO_STALE;
} }
/*
* newentry olladdr lladdr llchange (*=record)
* 0 n n -- (1)
* 0 y n -- (2)
* 0 n y -- (3) * STALE
* 0 y y n (4) *
* 0 y y y (5) * STALE
* 1 -- n -- (6) NOSTATE(= PASSIVE)
* 1 -- y -- (7) * STALE
*/
if (lladdr != NULL || do_update != 0) { /* (3-5) and (7) */
/*
* Record source link-layer address
* XXX is it dependent to ifp->if_type?
*
* We have to update either link-layer address
* or state. In most cases this require cfg/runtime locks
* to be held, so we have to do unlock/lock procedure to
* maintain proper lock order.
*/
LLE_ADDREF(ln);
LLE_WUNLOCK(ln);
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(ln);
/* Check if entry got deleted */
if (ln->la_flags & LLE_DELETED) {
IF_AFDATA_CFG_WUNLOCK(ifp);
LLE_FREE_LOCKED(ln);
return (NULL);
}
/* Check if similar entry has been added */
if (is_newentry != 0 &&
(ln_tmp = nd6_lookup(from, ND6_EXCLUSIVE, ifp)) != NULL) {
LLE_FREE_LOCKED(ln);
ln = ln_tmp;
ln_tmp = NULL;
is_newentry = 0;
}
/*
* Check if we really need to update runtime data:
* e.g. lladdr & r_flags
*/
r_update = 0;
/*
* no old lladdr, but new one exists. (3-5),(7)
*/
if (((ln->la_flags & LLE_VALID) == 0 && lladdr != NULL) ||
((ln->la_flags & LLE_VALID) != 0 &&
memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen))) {
r_update = 1;
}
/*
* state switch changing rlle_valid flag
*/
new_rvalid = 0;
if (newstate == ND6_LLINFO_REACHABLE ||
newstate == ND6_LLINFO_DELAY)
new_rvalid = 1;
old_rvalid = (ln->r_flags & RLLE_VALID) ? 1 : 0;
if (old_rvalid != new_rvalid)
r_update = 1;
/* linking new entry requires runtime lock */
if (is_newentry != 0)
r_update = 1;
if (r_update != 0) {
IF_AFDATA_RUN_WLOCK(ifp);
if (is_newentry != 0)
llentry_link(LLTABLE6(ifp), ln);
if (lladdr != NULL) {
bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
ln->la_flags |= LLE_VALID;
}
if (new_rvalid != 0)
ln->r_flags |= RLLE_VALID;
else
ln->r_flags &= ~RLLE_VALID;
IF_AFDATA_RUN_WUNLOCK(ifp);
}
IF_AFDATA_CFG_WUNLOCK(ifp);
LLE_REMREF(ln);
if (lladdr != NULL) {
/* We might still have to copy lladdr */
if (r_update == 0) {
bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
ln->la_flags |= LLE_VALID;
}
EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
}
}
if (do_update) { if (do_update) {
/* /*
* Update the state of the neighbor cache. * Update the state of the neighbor cache.
@ -1722,82 +1966,13 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
} }
} }
/* /* Check if we need to update router status */
* ICMP6 type dependent behavior. router = nd6_check_router(type, code, is_newentry, olladdr,
* lladdr != NULL ? 1 : 0, ln->ln_router);
* NS: clear IsRouter if new entry
* RS: clear IsRouter
* RA: set IsRouter if there's lladdr
* redir: clear IsRouter if new entry
*
* RA case, (1):
* The spec says that we must set IsRouter in the following cases:
* - If lladdr exist, set IsRouter. This means (1-5).
* - If it is old entry (!newentry), set IsRouter. This means (7).
* So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
* A quetion arises for (1) case. (1) case has no lladdr in the
* neighbor cache, this is similar to (6).
* This case is rare but we figured that we MUST NOT set IsRouter.
*
* newentry olladdr lladdr llchange NS RS RA redir
* D R
* 0 n n -- (1) c ? s
* 0 y n -- (2) c s s
* 0 n y -- (3) c s s
* 0 y y n (4) c s s
* 0 y y y (5) c s s
* 1 -- n -- (6) c c c s
* 1 -- y -- (7) c c s c s
*
* (c=clear s=set)
*/
switch (type & 0xff) {
case ND_NEIGHBOR_SOLICIT:
/*
* New entry must have is_router flag cleared.
*/
if (is_newentry) /* (6-7) */
ln->ln_router = 0;
break;
case ND_REDIRECT:
/*
* If the icmp is a redirect to a better router, always set the
* is_router flag. Otherwise, if the entry is newly created,
* clear the flag. [RFC 2461, sec 8.3]
*/
if (code == ND_REDIRECT_ROUTER)
ln->ln_router = 1;
else if (is_newentry) /* (6-7) */
ln->ln_router = 0;
break;
case ND_ROUTER_SOLICIT:
/*
* is_router flag must always be cleared.
*/
ln->ln_router = 0;
break;
case ND_ROUTER_ADVERT:
/*
* Mark an entry with lladdr as a router.
*/
if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
(is_newentry && lladdr)) { /* (7) */
ln->ln_router = 1;
}
break;
}
if (ln != NULL) { ln->ln_router = router;
static_route = (ln->la_flags & LLE_STATIC); LLE_WUNLOCK(ln);
router = ln->ln_router;
if (flags & ND6_EXCLUSIVE)
LLE_WUNLOCK(ln);
else
LLE_RUNLOCK(ln);
if (static_route)
ln = NULL;
}
if (chain) if (chain)
nd6_output_flush(ifp, ifp, chain, &sin6); nd6_output_flush(ifp, ifp, chain, &sin6);
@ -1824,16 +1999,6 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
defrouter_select(); defrouter_select();
} }
return (ln);
done:
if (ln != NULL) {
if (flags & ND6_EXCLUSIVE)
LLE_WUNLOCK(ln);
else
LLE_RUNLOCK(ln);
if (static_route)
ln = NULL;
}
return (ln); return (ln);
} }
@ -1975,6 +2140,8 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
int error = 0; int error = 0;
int has_lle = 0; int has_lle = 0;
int ip6len; int ip6len;
struct llentry *lle_tmp;
struct in6_addr src, *psrc;
#ifdef INVARIANTS #ifdef INVARIANTS
if (lle != NULL) { if (lle != NULL) {
@ -2016,9 +2183,35 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
* the condition below is not very efficient. But we believe * the condition below is not very efficient. But we believe
* it is tolerable, because this should be a rare case. * it is tolerable, because this should be a rare case.
*/ */
IF_AFDATA_CFG_WLOCK(ifp);
lle = nd6_create(&dst->sin6_addr, 0, ifp); lle = nd6_create(&dst->sin6_addr, 0, ifp);
IF_AFDATA_CFG_WUNLOCK(ifp); if (lle != NULL) {
IF_AFDATA_CFG_WLOCK(ifp);
LLE_WLOCK(lle);
/* Check if the same record was addded */
lle_tmp = nd6_lookup(&dst->sin6_addr,
ND6_EXCLUSIVE, ifp);
if (lle_tmp == NULL) {
/*
* No entry has been found.
* Link new one.
*/
IF_AFDATA_RUN_WLOCK(ifp);
llentry_link(LLTABLE6(ifp), lle);
IF_AFDATA_RUN_WUNLOCK(ifp);
}
IF_AFDATA_CFG_WUNLOCK(ifp);
if (lle_tmp != NULL) {
/*
* Existing lle has been found.
* Free new one.
*/
LLE_FREE_LOCKED(lle);
lle = lle_tmp;
lle_tmp = NULL;
}
}
} }
} }
if (lle == NULL) { if (lle == NULL) {
@ -2106,8 +2299,9 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
nd6_llinfo_settimer_locked(lle, nd6_llinfo_settimer_locked(lle,
(long)ND_IFINFO(ifp)->retrans * hz / 1000); (long)ND_IFINFO(ifp)->retrans * hz / 1000);
psrc = nd6_llinfo_get_holdsrc(lle, &src);
LLE_WUNLOCK(lle); LLE_WUNLOCK(lle);
nd6_ns_output(ifp, NULL, &dst->sin6_addr, lle, 0); nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
if (has_lle != 0) if (has_lle != 0)
LLE_WLOCK(lle); LLE_WLOCK(lle);
} else if (has_lle == 0) { } else if (has_lle == 0) {
@ -2269,29 +2463,35 @@ nd6_add_ifa_lle(struct in6_ifaddr *ia)
struct llentry *ln; struct llentry *ln;
ifp = ia->ia_ifa.ifa_ifp; ifp = ia->ia_ifa.ifa_ifp;
IF_AFDATA_CFG_WLOCK(ifp); ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
ln = lltable_create_lle(LLTABLE6(ifp), LLE_IFADDR, ln = lltable_create_lle(LLTABLE6(ifp), LLE_IFADDR,
(struct sockaddr *)&ia->ia_addr); (struct sockaddr *)&ia->ia_addr);
if (ln != NULL) { if (ln == NULL)
IF_AFDATA_RUN_WLOCK(ifp); return (ENOBUFS);
bcopy(IF_LLADDR(ifp), &ln->ll_addr, ifp->if_addrlen);
ln->la_flags |= (LLE_VALID | LLE_STATIC);
ln->r_flags |= RLLE_VALID;
ln->la_expire = 0; /* for IPv6 this means permanent */
ln->ln_state = ND6_LLINFO_REACHABLE;
llentry_link(LLTABLE6(ifp), ln);
IF_AFDATA_RUN_WUNLOCK(ifp);
}
ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; ln->la_flags |= (LLE_VALID | LLE_STATIC);
ln->r_flags |= RLLE_VALID;
ln->la_expire = 0; /* for IPv6 this means permanent */
ln->ln_state = ND6_LLINFO_REACHABLE;
IF_AFDATA_CFG_WLOCK(ifp);
/* Lock or new shiny lle */
LLE_WLOCK(ln);
lltable_delete_lle(LLTABLE6(ifp), LLE_IFADDR,
(struct sockaddr *)&ia->ia_addr);
bcopy(IF_LLADDR(ifp), &ln->ll_addr, ifp->if_addrlen);
/* Finally, link our lle to the list */
IF_AFDATA_RUN_WLOCK(ifp);
llentry_link(LLTABLE6(ifp), ln);
IF_AFDATA_RUN_WUNLOCK(ifp);
IF_AFDATA_CFG_WUNLOCK(ifp); IF_AFDATA_CFG_WUNLOCK(ifp);
if (ln != NULL) {
LLE_WUNLOCK(ln);
in6_newaddrmsg(ia, RTM_ADD);
return (0);
}
return (ENOBUFS); LLE_WUNLOCK(ln);
in6_newaddrmsg(ia, RTM_ADD);
return (0);
} }
/* /*

4
sys/netinet6/nd6.h
View File

@ -41,6 +41,7 @@
#include <sys/queue.h> #include <sys/queue.h>
#include <sys/callout.h> #include <sys/callout.h>
struct lltable;
struct llentry; struct llentry;
#define ND6_LLINFO_NOSTATE -2 #define ND6_LLINFO_NOSTATE -2
@ -419,6 +420,7 @@ int nd6_add_ifa_lle(struct in6_ifaddr *);
void nd6_rem_ifa_lle(struct in6_ifaddr *); void nd6_rem_ifa_lle(struct in6_ifaddr *);
int nd6_storelladdr(struct ifnet *, struct mbuf *, int nd6_storelladdr(struct ifnet *, struct mbuf *,
const struct sockaddr *, u_char *, struct llentry **); const struct sockaddr *, u_char *, struct llentry **);
void nd6_lltable_clear_entry(struct lltable *, struct llentry *);
/* nd6_nbr.c */ /* nd6_nbr.c */
void nd6_na_input(struct mbuf *, int, int); void nd6_na_input(struct mbuf *, int, int);
@ -426,7 +428,7 @@ void nd6_na_output(struct ifnet *, const struct in6_addr *,
const struct in6_addr *, u_long, int, struct sockaddr *); const struct in6_addr *, u_long, int, struct sockaddr *);
void nd6_ns_input(struct mbuf *, int, int); void nd6_ns_input(struct mbuf *, int, int);
void nd6_ns_output(struct ifnet *, const struct in6_addr *, void nd6_ns_output(struct ifnet *, const struct in6_addr *,
const struct in6_addr *, struct llentry *, int); const struct in6_addr *, const struct in6_addr *, int);
caddr_t nd6_ifptomac(struct ifnet *); caddr_t nd6_ifptomac(struct ifnet *);
void nd6_dad_start(struct ifaddr *, int); void nd6_dad_start(struct ifaddr *, int);
void nd6_dad_stop(struct ifaddr *); void nd6_dad_stop(struct ifaddr *);

28
sys/netinet6/nd6_nbr.c
View File

@ -381,7 +381,7 @@ nd6_ns_input(struct mbuf *m, int off, int icmp6len)
*/ */
void void
nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6, nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
const struct in6_addr *taddr6, struct llentry *ln, int dad) const struct in6_addr *taddr6, const struct in6_addr *csrc, int dad)
{ {
struct mbuf *m; struct mbuf *m;
struct m_tag *mtag; struct m_tag *mtag;
@ -462,29 +462,11 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
* - saddr6 belongs to the outgoing interface. * - saddr6 belongs to the outgoing interface.
* Otherwise, we perform the source address selection as usual. * Otherwise, we perform the source address selection as usual.
*/ */
struct in6_addr *hsrc;
hsrc = NULL; if (csrc != NULL)
if (ln != NULL) { ip6->ip6_src = *csrc;
LLE_RLOCK(ln); if (csrc && (ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
if (ln->la_hold != NULL) { (struct in6_addr *)csrc)) != NULL) {
struct ip6_hdr *hip6; /* hold ip6 */
/*
* assuming every packet in la_hold has the same IP
* header
*/
hip6 = mtod(ln->la_hold, struct ip6_hdr *);
/* XXX pullup? */
if (sizeof(*hip6) < ln->la_hold->m_len) {
ip6->ip6_src = hip6->ip6_src;
hsrc = &hip6->ip6_src;
}
}
LLE_RUNLOCK(ln);
}
if (hsrc && (ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
hsrc)) != NULL) {
/* ip6_src set already. */ /* ip6_src set already. */
ifa_free(ifa); ifa_free(ifa);
} else { } else {