/*
- * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* @(#)ip_output.c 8.3 (Berkeley) 1/21/94
* $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.16 2001/07/19 06:37:26 kris Exp $
*/
+/*
+ * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
+ * support for mandatory and extensible security protections. This notice
+ * is included in support of clause 2.2 (b) of the Apple Public License,
+ * Version 2.0.
+ */
#define _IP_VHL
#include <kern/locks.h>
#include <sys/sysctl.h>
+#include <machine/endian.h>
+
#include <net/if.h>
+#include <net/if_dl.h>
#include <net/route.h>
#include <netinet/in.h>
#include <netinet/kpi_ipfilter_var.h>
+#if CONFIG_MACF_NET
+#include <security/mac_framework.h>
+#endif
+
#include "faith.h"
#include <net/dlil.h>
#include <sys/kdebug.h>
+#include <libkern/OSAtomic.h>
#define DBG_LAYER_BEG NETDBG_CODE(DBG_NETIP, 1)
#define DBG_LAYER_END NETDBG_CODE(DBG_NETIP, 3)
#include <netinet/ip_dummynet.h>
#endif
+#if PF
+#include <net/pfvar.h>
+#endif /* PF */
+
#if IPFIREWALL_FORWARD_DEBUG
#define print_ip(a) printf("%ld.%ld.%ld.%ld",(ntohl(a.s_addr)>>24)&0xFF,\
(ntohl(a.s_addr)>>16)&0xFF,\
(ntohl(a.s_addr))&0xFF);
#endif
-#if IPSEC
-extern lck_mtx_t *sadb_mutex;
-#endif
u_short ip_id;
static int ip_pcbopts(int, struct mbuf **, struct mbuf *);
static int ip_setmoptions(struct sockopt *, struct ip_moptions **);
+static void ip_out_cksum_stats(int, u_int32_t);
+static struct ifaddr *in_selectsrcif(struct ip *, struct route *, unsigned int);
+static void ip_bindif(struct inpcb *, unsigned int);
+
int ip_createmoptions(struct ip_moptions **imop);
int ip_addmembership(struct ip_moptions *imo, struct ip_mreq *mreq);
int ip_dropmembership(struct ip_moptions *imo, struct ip_mreq *mreq);
int ip_optcopy(struct ip *, struct ip *);
-extern int (*fr_checkp)(struct ip *, int, struct ifnet *, int, struct mbuf **);
-#ifdef __APPLE__
-extern struct mbuf* m_dup(register struct mbuf *m, int how);
-#endif
+void in_delayed_cksum_offset(struct mbuf *, int );
+void in_cksum_offset(struct mbuf* , size_t );
-extern int apple_hwcksum_tx;
-extern u_long route_generation;
+extern int (*fr_checkp)(struct ip *, int, struct ifnet *, int, struct mbuf **);
extern struct protosw inetsw[];
static int ip_maxchainsent = 0;
SYSCTL_INT(_net_inet_ip, OID_AUTO, maxchainsent, CTLFLAG_RW,
&ip_maxchainsent, 0, "use dlil_output_list");
+#if DEBUG
+static int forge_ce = 0;
+SYSCTL_INT(_net_inet_ip, OID_AUTO, forge_ce, CTLFLAG_RW,
+ &forge_ce, 0, "Forge ECN CE");
+#endif /* DEBUG */
+
+static int ip_select_srcif_debug = 0;
+SYSCTL_INT(_net_inet_ip, OID_AUTO, select_srcif_debug, CTLFLAG_RW,
+ &ip_select_srcif_debug, 0, "log source interface selection debug info");
+
/*
* IP output. The packet in mbuf chain m contains a skeletal IP
* header (with len, off, ttl, proto, tos, src, dst).
struct mbuf *opt,
struct route *ro,
int flags,
- struct ip_moptions *imo)
+ struct ip_moptions *imo,
+ struct ip_out_args *ipoa)
{
int error;
- error = ip_output_list(m0, 0, opt, ro, flags, imo);
+ error = ip_output_list(m0, 0, opt, ro, flags, imo, ipoa);
return error;
}
+/*
+ * Returns: 0 Success
+ * ENOMEM
+ * EADDRNOTAVAIL
+ * ENETUNREACH
+ * EHOSTUNREACH
+ * EACCES
+ * EMSGSIZE
+ * ENOBUFS
+ * ipsec4_getpolicybyaddr:??? [IPSEC 4th argument, contents modified]
+ * ipsec4_getpolicybysock:??? [IPSEC 4th argument, contents modified]
+ * key_spdacquire:??? [IPSEC]
+ * ipsec4_output:??? [IPSEC]
+ * <fr_checkp>:??? [firewall]
+ * ip_dn_io_ptr:??? [dummynet]
+ * dlil_output:??? [DLIL]
+ * dlil_output_list:??? [DLIL]
+ *
+ * Notes: The ipsec4_getpolicyby{addr|sock} function error returns are
+ * only used as the error return from this function where one of
+ * these functions fails to return a policy.
+ */
int
ip_output_list(
struct mbuf *m0,
struct mbuf *opt,
struct route *ro,
int flags,
- struct ip_moptions *imo)
+ struct ip_moptions *imo,
+ struct ip_out_args *ipoa
+ )
{
- struct ip *ip, *mhip;
+ struct ip *ip;
struct ifnet *ifp = NULL;
- struct mbuf *m = m0;
+ struct mbuf *m = m0, **mppn = NULL;
int hlen = sizeof (struct ip);
- int len, off, error = 0;
+ int len = 0, off, error = 0;
struct sockaddr_in *dst = NULL;
- struct in_ifaddr *ia = NULL;
+ struct in_ifaddr *ia = NULL, *src_ia = NULL;
int isbroadcast, sw_csum;
struct in_addr pkt_dst;
#if IPSEC
#if IPFIREWALL_FORWARD
int fwd_rewrite_src = 0;
#endif
+#if IPFIREWALL
struct ip_fw_args args;
+#endif
int didfilter = 0;
ipfilter_t inject_filter_ref = 0;
struct m_tag *tag;
- struct route dn_route;
+ struct route saved_route;
+ struct ip_out_args saved_ipoa;
struct mbuf * packetlist;
- int pktcnt = 0;
-
- lck_mtx_lock(ip_mutex);
+ int pktcnt = 0, tso = 0;
+ unsigned int ifscope;
+ boolean_t select_srcif;
KERNEL_DEBUG(DBG_FNC_IP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0);
packetlist = m0;
+#if IPFIREWALL
+ args.next_hop = NULL;
args.eh = NULL;
args.rule = NULL;
- args.next_hop = NULL;
args.divert_rule = 0; /* divert cookie */
-
+ args.ipoa = NULL;
+
+ if (SLIST_EMPTY(&m0->m_pkthdr.tags))
+ goto ipfw_tags_done;
+
/* Grab info from mtags prepended to the chain */
#if DUMMYNET
- if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_DUMMYNET, NULL)) != NULL) {
+ if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID,
+ KERNEL_TAG_TYPE_DUMMYNET, NULL)) != NULL) {
struct dn_pkt_tag *dn_tag;
-
+
dn_tag = (struct dn_pkt_tag *)(tag+1);
args.rule = dn_tag->rule;
opt = NULL;
- dn_route = dn_tag->ro;
- ro = &dn_route;
-
+ saved_route = dn_tag->ro;
+ ro = &saved_route;
+
imo = NULL;
dst = dn_tag->dn_dst;
ifp = dn_tag->ifp;
flags = dn_tag->flags;
-
+ saved_ipoa = dn_tag->ipoa;
+ ipoa = &saved_ipoa;
+
m_tag_delete(m0, tag);
}
#endif /* DUMMYNET */
- if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_DIVERT, NULL)) != NULL) {
+#if IPDIVERT
+ if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID,
+ KERNEL_TAG_TYPE_DIVERT, NULL)) != NULL) {
struct divert_tag *div_tag;
-
+
div_tag = (struct divert_tag *)(tag+1);
args.divert_rule = div_tag->cookie;
m_tag_delete(m0, tag);
}
- if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_IPFORWARD, NULL)) != NULL) {
+#endif /* IPDIVERT */
+
+ if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID,
+ KERNEL_TAG_TYPE_IPFORWARD, NULL)) != NULL) {
struct ip_fwd_tag *ipfwd_tag;
-
+
ipfwd_tag = (struct ip_fwd_tag *)(tag+1);
args.next_hop = ipfwd_tag->next_hop;
-
+
m_tag_delete(m0, tag);
}
+ipfw_tags_done:
+#endif /* IPFIREWALL */
m = m0;
-
+
#if DIAGNOSTIC
if ( !m || (m->m_flags & M_PKTHDR) != 0)
panic("ip_output no HDR");
mtod(m, struct ip *)->ip_p);
#endif
+ /*
+ * At present the IP_OUTARGS flag implies a request for IP to
+ * perform source interface selection. In the forwarding case,
+ * only the ifscope value is used, as source interface selection
+ * doesn't take place.
+ */
+ if (ip_doscopedroute && (flags & IP_OUTARGS)) {
+ select_srcif = !(flags & IP_FORWARDING);
+ ifscope = ipoa->ipoa_ifscope;
+ } else {
+ select_srcif = FALSE;
+ ifscope = IFSCOPE_NONE;
+ }
+
+#if IPFIREWALL
if (args.rule != NULL) { /* dummynet already saw us */
- ip = mtod(m, struct ip *);
- hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
- if (ro->ro_rt != NULL)
- ia = (struct in_ifaddr *)ro->ro_rt->rt_ifa;
+ ip = mtod(m, struct ip *);
+ hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
+ if (ro->ro_rt != NULL) {
+ RT_LOCK_SPIN(ro->ro_rt);
+ ia = (struct in_ifaddr *)ro->ro_rt->rt_ifa;
if (ia)
ifaref(&ia->ia_ifa);
+ RT_UNLOCK(ro->ro_rt);
+ }
#if IPSEC
- if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0) {
- so = ipsec_getsocket(m);
- (void)ipsec_setsocket(m, NULL);
- }
+ if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0) {
+ so = ipsec_getsocket(m);
+ (void)ipsec_setsocket(m, NULL);
+ }
#endif
- goto sendit;
+ goto sendit;
}
+#endif /* IPFIREWALL */
#if IPSEC
if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0) {
* No need to proccess packet twice if we've
* already seen it
*/
- inject_filter_ref = ipf_get_inject_filter(m);
+ if (!SLIST_EMPTY(&m->m_pkthdr.tags))
+ inject_filter_ref = ipf_get_inject_filter(m);
+ else
+ inject_filter_ref = 0;
if (opt) {
m = ip_insertoptions(m, opt, &len);
hlen = len;
}
ip = mtod(m, struct ip *);
+#if IPFIREWALL
pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
+#else
+ pkt_dst = ip->ip_dst;
+#endif
/*
* Fill in IP header.
#else
ip->ip_id = htons(ip_id++);
#endif
- ipstat.ips_localout++;
+ OSAddAtomic(1, &ipstat.ips_localout);
} else {
hlen = IP_VHL_HL(ip->ip_vhl) << 2;
}
+
+#if DEBUG
+ /* For debugging, we let the stack forge congestion */
+ if (forge_ce != 0 &&
+ ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT1 ||
+ (ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT0)) {
+ ip->ip_tos = (ip->ip_tos & ~IPTOS_ECN_MASK) | IPTOS_ECN_CE;
+ forge_ce--;
+ }
+#endif /* DEBUG */
KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr,
ip->ip_src.s_addr, ip->ip_p, ip->ip_off, ip->ip_len);
-
+
dst = (struct sockaddr_in *)&ro->ro_dst;
/*
* cache with IPv6.
*/
- {
- if (ro->ro_rt && (ro->ro_rt->generation_id != route_generation) &&
- ((flags & (IP_ROUTETOIF | IP_FORWARDING)) == 0) && (ip->ip_src.s_addr != INADDR_ANY) &&
- (ifa_foraddr(ip->ip_src.s_addr) == 0)) {
- error = EADDRNOTAVAIL;
- goto bad;
+ if (ro->ro_rt != NULL) {
+ if (ro->ro_rt->generation_id != route_generation &&
+ ((flags & (IP_ROUTETOIF | IP_FORWARDING)) == 0) &&
+ (ip->ip_src.s_addr != INADDR_ANY)) {
+ src_ia = ifa_foraddr(ip->ip_src.s_addr);
+ if (src_ia == NULL) {
+ error = EADDRNOTAVAIL;
+ goto bad;
+ }
+ ifafree(&src_ia->ia_ifa);
}
+ /*
+ * Test rt_flags without holding rt_lock for performance
+ * reasons; if the route is down it will hopefully be
+ * caught by the layer below (since it uses this route
+ * as a hint) or during the next transmit.
+ */
+ if ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
+ dst->sin_family != AF_INET ||
+ dst->sin_addr.s_addr != pkt_dst.s_addr) {
+ rtfree(ro->ro_rt);
+ ro->ro_rt = NULL;
+ }
+ /*
+ * If we're doing source interface selection, we may not
+ * want to use this route; only synch up the generation
+ * count otherwise.
+ */
+ if (!select_srcif && ro->ro_rt != NULL &&
+ ro->ro_rt->generation_id != route_generation)
+ ro->ro_rt->generation_id = route_generation;
}
- if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
- dst->sin_family != AF_INET ||
- dst->sin_addr.s_addr != pkt_dst.s_addr)) {
- rtfree(ro->ro_rt);
- ro->ro_rt = (struct rtentry *)0;
- }
- if (ro->ro_rt == 0) {
+ if (ro->ro_rt == NULL) {
bzero(dst, sizeof(*dst));
dst->sin_family = AF_INET;
dst->sin_len = sizeof(*dst);
ifafree(&ia->ia_ifa);
if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0) {
if ((ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
- ipstat.ips_noroute++;
+ OSAddAtomic(1, &ipstat.ips_noroute);
error = ENETUNREACH;
goto bad;
}
ifp = ia->ia_ifp;
ip->ip_ttl = 1;
isbroadcast = in_broadcast(dst->sin_addr, ifp);
+ } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
+ imo != NULL && imo->imo_multicast_ifp != NULL) {
+ /*
+ * Bypass the normal routing lookup for multicast
+ * packets if the interface is specified.
+ */
+ ifp = imo->imo_multicast_ifp;
+ isbroadcast = 0;
+ if (ia != NULL)
+ ifafree(&ia->ia_ifa);
+
+ /* Macro takes reference on ia */
+ IFP_TO_IA(ifp, ia);
} else {
+ boolean_t cloneok = FALSE;
+ /*
+ * Perform source interface selection; the source IP address
+ * must belong to one of the addresses of the interface used
+ * by the route. For performance reasons, do this only if
+ * there is no route, or if the routing table has changed,
+ * or if we haven't done source interface selection on this
+ * route (for this PCB instance) before.
+ */
+ if (select_srcif && ip->ip_src.s_addr != INADDR_ANY &&
+ (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP) ||
+ ro->ro_rt->generation_id != route_generation ||
+ !(ro->ro_flags & ROF_SRCIF_SELECTED))) {
+ struct ifaddr *ifa;
+
+ /* Find the source interface */
+ ifa = in_selectsrcif(ip, ro, ifscope);
+
+ /*
+ * If the source address is spoofed (in the case
+ * of IP_RAWOUTPUT), or if this is destined for
+ * local/loopback, just let it go out using the
+ * interface of the route. Otherwise, there's no
+ * interface having such an address, so bail out.
+ */
+ if (ifa == NULL && !(flags & IP_RAWOUTPUT) &&
+ ifscope != lo_ifp->if_index) {
+ error = EADDRNOTAVAIL;
+ goto bad;
+ }
+
+ /*
+ * If the caller didn't explicitly specify the scope,
+ * pick it up from the source interface. If the cached
+ * route was wrong and was blown away as part of source
+ * interface selection, don't mask out RTF_PRCLONING
+ * since that route may have been allocated by the ULP,
+ * unless the IP header was created by the caller or
+ * the destination is IPv4 LLA. The check for the
+ * latter is needed because IPv4 LLAs are never scoped
+ * in the current implementation, and we don't want to
+ * replace the resolved IPv4 LLA route with one whose
+ * gateway points to that of the default gateway on
+ * the primary interface of the system.
+ */
+ if (ifa != NULL) {
+ if (ifscope == IFSCOPE_NONE)
+ ifscope = ifa->ifa_ifp->if_index;
+ ifafree(ifa);
+ cloneok = (!(flags & IP_RAWOUTPUT) &&
+ !(IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))));
+ }
+ }
+
/*
* If this is the case, we probably don't want to allocate
* a protocol-cloned route since we didn't get one from the
* the link layer, as this is probably required in all cases
* for correct operation (as it is for ARP).
*/
- if (ro->ro_rt == 0)
- rtalloc_ign(ro, RTF_PRCLONING);
- if (ro->ro_rt == 0) {
- ipstat.ips_noroute++;
+ if (ro->ro_rt == NULL) {
+ unsigned long ign = RTF_PRCLONING;
+ /*
+ * We make an exception here: if the destination
+ * address is INADDR_BROADCAST, allocate a protocol-
+ * cloned host route so that we end up with a route
+ * marked with the RTF_BROADCAST flag. Otherwise,
+ * we would end up referring to the default route,
+ * instead of creating a cloned host route entry.
+ * That would introduce inconsistencies between ULPs
+ * that allocate a route and those that don't. The
+ * RTF_BROADCAST route is important since we'd want
+ * to send out undirected IP broadcast packets using
+ * link-level broadcast address. Another exception
+ * is for ULP-created routes that got blown away by
+ * source interface selection (see above).
+ *
+ * These exceptions will no longer be necessary when
+ * the RTF_PRCLONING scheme is no longer present.
+ */
+ if (cloneok || dst->sin_addr.s_addr == INADDR_BROADCAST)
+ ign &= ~RTF_PRCLONING;
+
+ /*
+ * Loosen the route lookup criteria if the ifscope
+ * corresponds to the loopback interface; this is
+ * needed to support Application Layer Gateways
+ * listening on loopback, in conjunction with packet
+ * filter redirection rules. The final source IP
+ * address will be rewritten by the packet filter
+ * prior to the RFC1122 loopback check below.
+ */
+ if (ifscope == lo_ifp->if_index)
+ rtalloc_ign(ro, ign);
+ else
+ rtalloc_scoped_ign(ro, ign, ifscope);
+ }
+
+ if (ro->ro_rt == NULL) {
+ OSAddAtomic(1, &ipstat.ips_noroute);
error = EHOSTUNREACH;
goto bad;
}
+
if (ia)
ifafree(&ia->ia_ifa);
+ RT_LOCK_SPIN(ro->ro_rt);
ia = ifatoia(ro->ro_rt->rt_ifa);
if (ia)
ifaref(&ia->ia_ifa);
ro->ro_rt->rt_use++;
if (ro->ro_rt->rt_flags & RTF_GATEWAY)
dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
- if (ro->ro_rt->rt_flags & RTF_HOST)
+ if (ro->ro_rt->rt_flags & RTF_HOST) {
isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
- else
+ } else {
+ /* Become a regular mutex */
+ RT_CONVERT_LOCK(ro->ro_rt);
isbroadcast = in_broadcast(dst->sin_addr, ifp);
+ }
+ RT_UNLOCK(ro->ro_rt);
}
+
if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
struct in_multi *inm;
if (imo->imo_multicast_ifp != NULL) {
ifp = imo->imo_multicast_ifp;
}
+#if MROUTING
if (imo->imo_multicast_vif != -1 &&
((flags & IP_RAWOUTPUT) == 0 || ip->ip_src.s_addr == INADDR_ANY))
ip->ip_src.s_addr =
ip_mcast_src(imo->imo_multicast_vif);
+#endif /* MROUTING */
} else
if ((flags & IP_RAWOUTPUT) == 0) ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
/*
*/
if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
if ((ifp->if_flags & IFF_MULTICAST) == 0) {
- ipstat.ips_noroute++;
+ OSAddAtomic(1, &ipstat.ips_noroute);
error = ENETUNREACH;
goto bad;
}
* of outgoing interface.
*/
if (ip->ip_src.s_addr == INADDR_ANY) {
- register struct in_ifaddr *ia1;
-
+ struct in_ifaddr *ia1;
+ lck_rw_lock_shared(in_ifaddr_rwlock);
TAILQ_FOREACH(ia1, &in_ifaddrhead, ia_link)
if (ia1->ia_ifp == ifp) {
ip->ip_src = IA_SIN(ia1)->sin_addr;
-
break;
}
+ lck_rw_done(in_ifaddr_rwlock);
if (ip->ip_src.s_addr == INADDR_ANY) {
error = ENETUNREACH;
goto bad;
ipf_pktopts.ippo_mcast_loop = imo->imo_multicast_loop;
}
- lck_mtx_unlock(ip_mutex);
ipf_ref();
/* 4135317 - always pass network byte order to filter */
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
-
+#endif
+
TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
if (seen == 0) {
if ((struct ipfilter *)inject_filter_ref == filter)
}
if (result != 0) {
ipf_unref();
- lck_mtx_lock(ip_mutex);
goto bad;
}
}
}
/* set back to host byte order */
+ ip = mtod(m, struct ip *);
+
+#if BYTE_ORDER != BIG_ENDIAN
NTOHS(ip->ip_len);
NTOHS(ip->ip_off);
-
- lck_mtx_lock(ip_mutex);
+#endif
+
ipf_unref();
didfilter = 1;
}
ip_mloopback(ifp, m, dst, hlen);
}
+#if MROUTING
else {
/*
* If we are acting as a multicast router, perform
imo = NULL;
if (ip_mforward(ip, ifp, m, imo) != 0) {
m_freem(m);
- lck_mtx_unlock(ip_mutex);
goto done;
}
}
}
+#endif /* MROUTING */
/*
* Multicasts with a time-to-live of zero may be looped-
*/
if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
m_freem(m);
- lck_mtx_unlock(ip_mutex);
goto done;
}
}
sendit:
+#if PF
+ /* Invoke outbound packet filter */
+ if (pf_af_hook(ifp, mppn, &m, AF_INET, FALSE) != 0) {
+ if (packetlist == m0) {
+ packetlist = m;
+ mppn = NULL;
+ }
+ if (m != NULL) {
+ m0 = m;
+ /* Next packet in the chain */
+ goto loopit;
+ } else if (packetlist != NULL) {
+ /* No more packet; send down the chain */
+ goto sendchain;
+ }
+ /* Nothing left; we're done */
+ goto done;
+ }
+ m0 = m;
+ ip = mtod(m, struct ip *);
+ pkt_dst = ip->ip_dst;
+ hlen = IP_VHL_HL(ip->ip_vhl) << 2;
+#endif /* PF */
/*
* Force IP TTL to 255 following draft-ietf-zeroconf-ipv4-linklocal.txt
*/
}
}
-injectit:
if (!didfilter && !TAILQ_EMPTY(&ipv4_filters)) {
struct ipfilter *filter;
int seen = (inject_filter_ref == 0);
-
- lck_mtx_unlock(ip_mutex);
+
+ /* Check that a TSO frame isn't passed to a filter.
+ * This could happen if a filter is inserted while
+ * TCP is sending the TSO packet.
+ */
+ if (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4) {
+ error = EMSGSIZE;
+ goto bad;
+ }
+
ipf_ref();
/* 4135317 - always pass network byte order to filter */
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
-
+#endif
+
TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
if (seen == 0) {
if ((struct ipfilter *)inject_filter_ref == filter)
}
if (result != 0) {
ipf_unref();
- lck_mtx_lock(ip_mutex);
goto bad;
}
}
}
/* set back to host byte order */
+ ip = mtod(m, struct ip *);
+
+#if BYTE_ORDER != BIG_ENDIAN
NTOHS(ip->ip_len);
NTOHS(ip->ip_off);
-
+#endif
+
ipf_unref();
- lck_mtx_lock(ip_mutex);
}
#if IPSEC
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_START, 0,0,0,0,0);
- lck_mtx_lock(sadb_mutex);
/* get SP for this packet */
if (so == NULL)
sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
if (sp == NULL) {
- ipsecstat.out_inval++;
+ IPSEC_STAT_INCREMENT(ipsecstat.out_inval);
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
- lck_mtx_unlock(sadb_mutex);
goto bad;
}
/* check policy */
switch (sp->policy) {
case IPSEC_POLICY_DISCARD:
+ case IPSEC_POLICY_GENERATE:
/*
* This packet is just discarded.
*/
- ipsecstat.out_polvio++;
+ IPSEC_STAT_INCREMENT(ipsecstat.out_polvio);
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 1,0,0,0,0);
- lck_mtx_unlock(sadb_mutex);
goto bad;
case IPSEC_POLICY_BYPASS:
case IPSEC_POLICY_NONE:
/* no need to do IPsec. */
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 2,0,0,0,0);
- lck_mtx_unlock(sadb_mutex);
goto skip_ipsec;
case IPSEC_POLICY_IPSEC:
/* acquire a policy */
error = key_spdacquire(sp);
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 3,0,0,0,0);
- lck_mtx_unlock(sadb_mutex);
goto bad;
}
break;
m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
}
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
+#endif
- lck_mtx_unlock(ip_mutex);
error = ipsec4_output(&state, sp, flags);
- lck_mtx_unlock(sadb_mutex);
- lck_mtx_lock(ip_mutex);
m0 = m = state.m;
hlen = ip->ip_hl << 2;
#endif
/* Check that there wasn't a route change and src is still valid */
-
- if (ro->ro_rt && ro->ro_rt->generation_id != route_generation) {
- if (ifa_foraddr(ip->ip_src.s_addr) == 0 && ((flags & (IP_ROUTETOIF | IP_FORWARDING)) == 0)) {
- error = EADDRNOTAVAIL;
- KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 5,0,0,0,0);
+ if (ro->ro_rt != NULL && ro->ro_rt->generation_id != route_generation) {
+ if ((src_ia = ifa_foraddr(ip->ip_src.s_addr)) == NULL &&
+ ((flags & (IP_ROUTETOIF | IP_FORWARDING)) == 0)) {
+ error = EADDRNOTAVAIL;
+ KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
+ 5,0,0,0,0);
goto bad;
}
rtfree(ro->ro_rt);
ro->ro_rt = NULL;
+ if (src_ia != NULL)
+ ifafree(&src_ia->ia_ifa);
}
if (ro->ro_rt == NULL) {
if ((flags & IP_ROUTETOIF) == 0) {
- printf("ip_output: "
- "can't update route after IPsec processing\n");
- error = EHOSTUNREACH; /*XXX*/
- KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 6,0,0,0,0);
+ printf("ip_output: can't update route after "
+ "IPsec processing\n");
+ error = EHOSTUNREACH; /*XXX*/
+ KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
+ 6,0,0,0,0);
goto bad;
}
} else {
if (ia)
ifafree(&ia->ia_ifa);
+ RT_LOCK_SPIN(ro->ro_rt);
ia = ifatoia(ro->ro_rt->rt_ifa);
if (ia)
ifaref(&ia->ia_ifa);
ifp = ro->ro_rt->rt_ifp;
+ RT_UNLOCK(ro->ro_rt);
}
/* make it flipped, again. */
+
+#if BYTE_ORDER != BIG_ENDIAN
NTOHS(ip->ip_len);
NTOHS(ip->ip_off);
+#endif
+
KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END, 7,0xff,0xff,0xff,0xff);
/* Pass to filters again */
if (!TAILQ_EMPTY(&ipv4_filters)) {
struct ipfilter *filter;
- lck_mtx_unlock(ip_mutex);
+ /* Check that a TSO frame isn't passed to a filter.
+ * This could happen if a filter is inserted while
+ * TCP is sending the TSO packet.
+ */
+ if (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4) {
+ error = EMSGSIZE;
+ goto bad;
+ }
+
ipf_ref();
/* 4135317 - always pass network byte order to filter */
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
-
+#endif
+
TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
if (filter->ipf_filter.ipf_output) {
errno_t result;
}
if (result != 0) {
ipf_unref();
- lck_mtx_lock(ip_mutex);
goto bad;
}
}
}
/* set back to host byte order */
+ ip = mtod(m, struct ip *);
+
+#if BYTE_ORDER != BIG_ENDIAN
NTOHS(ip->ip_len);
NTOHS(ip->ip_off);
-
+#endif
+
ipf_unref();
- lck_mtx_lock(ip_mutex);
}
skip_ipsec:
#endif /*IPSEC*/
+#if IPFIREWALL
/*
* IpHack's section.
* - Xlate: translate packet's addr/port (NAT).
struct mbuf *m1 = m;
if ((error = (*fr_checkp)(ip, hlen, ifp, 1, &m1)) || !m1) {
- lck_mtx_unlock(ip_mutex);
goto done;
}
ip = mtod(m0 = m = m1, struct ip *);
args.m = m;
args.next_hop = dst;
args.oif = ifp;
- lck_mtx_unlock(ip_mutex);
off = ip_fw_chk_ptr(&args);
m = args.m;
dst = args.next_hop;
goto done ;
}
ip = mtod(m, struct ip *);
+
if (off == 0 && dst == old) {/* common case */
- lck_mtx_lock(ip_mutex);
goto pass ;
}
#if DUMMYNET
if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
- /*
- * pass the pkt to dummynet. Need to include
- * pipe number, m, ifp, ro, dst because these are
- * not recomputed in the next pass.
- * All other parameters have been already used and
- * so they are not needed anymore.
- * XXX note: if the ifp or ro entry are deleted
- * while a pkt is in dummynet, we are in trouble!
- */
- args.ro = ro;
- args.dst = dst;
- args.flags = flags;
-
- error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
- &args);
- goto done;
+ /*
+ * pass the pkt to dummynet. Need to include
+ * pipe number, m, ifp, ro, dst because these are
+ * not recomputed in the next pass.
+ * All other parameters have been already used and
+ * so they are not needed anymore.
+ * XXX note: if the ifp or ro entry are deleted
+ * while a pkt is in dummynet, we are in trouble!
+ */
+ args.ro = ro;
+ args.dst = dst;
+ args.flags = flags;
+ if (flags & IP_OUTARGS)
+ args.ipoa = ipoa;
+
+ error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
+ &args);
+ goto done;
}
#endif /* DUMMYNET */
- lck_mtx_lock(ip_mutex);
#if IPDIVERT
if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
struct mbuf *clone = NULL;
}
/* Restore packet header fields to original values */
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
+#endif
/* Deliver packet to divert input routine */
divert_packet(m, 0, off & 0xffff, args.divert_rule);
ip = mtod(m, struct ip *);
goto pass;
}
- lck_mtx_unlock(ip_mutex);
goto done;
}
#endif
* as the packet runs through ip_input() as
* it is done through a ISR.
*/
+ lck_rw_lock_shared(in_ifaddr_rwlock);
TAILQ_FOREACH(ia_fw, &in_ifaddrhead, ia_link) {
/*
* If the addr to forward to is one
dst->sin_addr.s_addr)
break;
}
- if (ia) {
+ lck_rw_done(in_ifaddr_rwlock);
+ if (ia_fw) {
/* tell ip_input "dont filter" */
struct m_tag *fwd_tag;
struct ip_fwd_tag *ipfwd_tag;
- fwd_tag = m_tag_alloc(KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_IPFORWARD,
- sizeof(struct sockaddr_in), M_NOWAIT);
+ fwd_tag = m_tag_alloc(KERNEL_MODULE_TAG_ID,
+ KERNEL_TAG_TYPE_IPFORWARD,
+ sizeof (*ipfwd_tag), M_NOWAIT);
if (fwd_tag == NULL) {
error = ENOBUFS;
goto bad;
m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
ip->ip_sum = in_cksum(m, hlen);
}
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
-
- lck_mtx_unlock(ip_mutex);
+#endif
/* we need to call dlil_output to run filters
* and resync to avoid recursion loops.
*/
bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
- ro_fwd->ro_rt = 0;
+ ro_fwd->ro_rt = NULL;
rtalloc_ign(ro_fwd, RTF_PRCLONING);
- if (ro_fwd->ro_rt == 0) {
- ipstat.ips_noroute++;
+ if (ro_fwd->ro_rt == NULL) {
+ OSAddAtomic(1, &ipstat.ips_noroute);
error = EHOSTUNREACH;
goto bad;
}
+ RT_LOCK_SPIN(ro_fwd->ro_rt);
ia_fw = ifatoia(ro_fwd->ro_rt->rt_ifa);
+ if (ia_fw != NULL)
+ ifaref(&ia_fw->ia_ifa);
ifp = ro_fwd->ro_rt->rt_ifp;
ro_fwd->ro_rt->rt_use++;
if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
dst = (struct sockaddr_in *)ro_fwd->ro_rt->rt_gateway;
- if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
+ if (ro_fwd->ro_rt->rt_flags & RTF_HOST) {
isbroadcast =
(ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
- else
+ } else {
+ /* Become a regular mutex */
+ RT_CONVERT_LOCK(ro_fwd->ro_rt);
isbroadcast = in_broadcast(dst->sin_addr, ifp);
+ }
+ RT_UNLOCK(ro_fwd->ro_rt);
rtfree(ro->ro_rt);
ro->ro_rt = ro_fwd->ro_rt;
dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
* which would have been gotten from the-then
* interface, do it again, from the new one.
*/
- if (fwd_rewrite_src)
- ip->ip_src = IA_SIN(ia_fw)->sin_addr;
+ if (ia_fw != NULL) {
+ if (fwd_rewrite_src)
+ ip->ip_src = IA_SIN(ia_fw)->sin_addr;
+ ifafree(&ia_fw->ia_ifa);
+ }
goto pass ;
}
#endif /* IPFIREWALL_FORWARD */
*/
m_freem(m);
error = EACCES; /* not sure this is the right error msg */
- lck_mtx_unlock(ip_mutex);
goto done;
}
+#endif /* IPFIREWALL */
pass:
#if __APPLE__
if ((ifp->if_flags & IFF_LOOPBACK) == 0 &&
((ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
(ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)) {
- ipstat.ips_badaddr++;
+ OSAddAtomic(1, &ipstat.ips_badaddr);
m_freem(m);
/*
* Do not simply drop the packet just like a firewall -- we want the
* loopback as the source address.
*/
error = ENETUNREACH;
- lck_mtx_unlock(ip_mutex);
goto done;
}
#endif
m->m_pkthdr.csum_flags |= CSUM_IP;
+ tso = (ifp->if_hwassist & IFNET_TSO_IPV4) && (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4);
+
sw_csum = m->m_pkthdr.csum_flags
& ~IF_HWASSIST_CSUM_FLAGS(ifp->if_hwassist);
/* let the software handle any UDP or TCP checksums */
sw_csum |= (CSUM_DELAY_DATA & m->m_pkthdr.csum_flags);
}
+ } else if (apple_hwcksum_tx == 0) {
+ sw_csum |= (CSUM_DELAY_DATA | CSUM_DELAY_IP) &
+ m->m_pkthdr.csum_flags;
}
if (sw_csum & CSUM_DELAY_DATA) {
sw_csum &= ~CSUM_DELAY_DATA;
m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
}
-
- m->m_pkthdr.csum_flags &= IF_HWASSIST_CSUM_FLAGS(ifp->if_hwassist);
+
+ if (apple_hwcksum_tx != 0) {
+ m->m_pkthdr.csum_flags &=
+ IF_HWASSIST_CSUM_FLAGS(ifp->if_hwassist);
+ } else {
+ m->m_pkthdr.csum_flags = 0;
+ }
/*
* If small enough for interface, or the interface will take
* care of the fragmentation for us, can just send directly.
*/
- if ((u_short)ip->ip_len <= ifp->if_mtu ||
+ if ((u_short)ip->ip_len <= ifp->if_mtu || tso ||
ifp->if_hwassist & CSUM_FRAGMENT) {
+ if (tso)
+ m->m_pkthdr.csum_flags |= CSUM_TSO_IPV4;
+
+
+#if BYTE_ORDER != BIG_ENDIAN
HTONS(ip->ip_len);
HTONS(ip->ip_off);
+#endif
+
ip->ip_sum = 0;
if (sw_csum & CSUM_DELAY_IP) {
ip->ip_sum = in_cksum(m, hlen);
ipsec_delaux(m);
#endif
if (packetchain == 0) {
- lck_mtx_unlock(ip_mutex);
- error = dlil_output(ifp, PF_INET, m, (void *) ro->ro_rt,
- (struct sockaddr *)dst, 0);
- goto done;
+ error = ifnet_output(ifp, PF_INET, m, ro->ro_rt,
+ (struct sockaddr *)dst);
+ goto done;
}
else { /* packet chaining allows us to reuse the route for all packets */
+ mppn = &m->m_nextpkt;
m = m->m_nextpkt;
if (m == NULL) {
+#if PF
+sendchain:
+#endif /* PF */
if (pktcnt > ip_maxchainsent)
ip_maxchainsent = pktcnt;
//send
- lck_mtx_unlock(ip_mutex);
- error = dlil_output_list(ifp, PF_INET, packetlist, (void *) ro->ro_rt,
- (struct sockaddr *)dst, 0);
+ error = ifnet_output(ifp, PF_INET, packetlist,
+ ro->ro_rt, (struct sockaddr *)dst);
pktcnt = 0;
goto done;
* Too large for interface; fragment if possible.
* Must be able to put at least 8 bytes per fragment.
*/
- if (ip->ip_off & IP_DF) {
+
+ if (ip->ip_off & IP_DF || (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4)) {
error = EMSGSIZE;
/*
* This case can happen if the user changed the MTU
+ *
* of an interface after enabling IP on it. Because
* most netifs don't keep track of routes pointing to
* them, there is no way for one to update all its
* routes when the MTU is changed.
*/
+ RT_LOCK_SPIN(ro->ro_rt);
if (ro->ro_rt && (ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST))
&& !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU)
&& (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
}
- ipstat.ips_cantfrag++;
+ RT_UNLOCK(ro->ro_rt);
+ OSAddAtomic(1, &ipstat.ips_cantfrag);
goto bad;
}
- len = (ifp->if_mtu - hlen) &~ 7;
- if (len < 8) {
- error = EMSGSIZE;
+
+ error = ip_fragment(m, ifp, ifp->if_mtu, sw_csum);
+ if (error != 0) {
+ m0 = m = NULL;
goto bad;
}
+ KERNEL_DEBUG(DBG_LAYER_END, ip->ip_dst.s_addr,
+ ip->ip_src.s_addr, ip->ip_p, ip->ip_off, ip->ip_len);
+
+ for (m = m0; m; m = m0) {
+ m0 = m->m_nextpkt;
+ m->m_nextpkt = 0;
+#if IPSEC
+ /* clean ipsec history once it goes out of the node */
+ if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0)
+ ipsec_delaux(m);
+#endif
+ if (error == 0) {
+#ifndef __APPLE__
+ /* Record statistics for this interface address. */
+ if (ia != NULL) {
+ ia->ia_ifa.if_opackets++;
+ ia->ia_ifa.if_obytes += m->m_pkthdr.len;
+ }
+#endif
+ if ((packetchain != 0) && (pktcnt > 0))
+ panic("ip_output: mix of packet in packetlist is wrong=%p", packetlist);
+ error = ifnet_output(ifp, PF_INET, m, ro->ro_rt,
+ (struct sockaddr *)dst);
+ } else
+ m_freem(m);
+ }
+
+ if (error == 0)
+ OSAddAtomic(1, &ipstat.ips_fragmented);
+
+done:
+ if (ia) {
+ ifafree(&ia->ia_ifa);
+ ia = NULL;
+ }
+#if IPSEC
+ if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0) {
+ if (ro == &iproute && ro->ro_rt) {
+ rtfree(ro->ro_rt);
+ ro->ro_rt = NULL;
+ }
+ if (sp != NULL) {
+ KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
+ printf("DP ip_output call free SP:%x\n", sp));
+ key_freesp(sp, KEY_SADB_UNLOCKED);
+ }
+ }
+#endif /* IPSEC */
+
+ KERNEL_DEBUG(DBG_FNC_IP_OUTPUT | DBG_FUNC_END, error,0,0,0,0);
+ return (error);
+bad:
+ m_freem(m0);
+ goto done;
+}
+
+int
+ip_fragment(struct mbuf *m, struct ifnet *ifp, unsigned long mtu, int sw_csum)
+{
+ struct ip *ip, *mhip;
+ int len, hlen, mhlen, firstlen, off, error = 0;
+ struct mbuf **mnext = &m->m_nextpkt, *m0;
+ int nfrags = 1;
+
+ ip = mtod(m, struct ip *);
+#ifdef _IP_VHL
+ hlen = IP_VHL_HL(ip->ip_vhl) << 2;
+#else
+ hlen = ip->ip_hl << 2;
+#endif
+
+ firstlen = len = (mtu - hlen) &~ 7;
+ if (len < 8) {
+ m_freem(m);
+ return (EMSGSIZE);
+ }
+
/*
* if the interface will not calculate checksums on
* fragmented packets, then do it here.
if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
(ifp->if_hwassist & CSUM_IP_FRAGS) == 0) {
in_delayed_cksum(m);
- if (m == NULL) {
- lck_mtx_unlock(ip_mutex);
- return(ENOMEM);
- }
m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
}
-
- {
- int mhlen, firstlen = len;
- struct mbuf **mnext = &m->m_nextpkt;
- int nfrags = 1;
-
/*
* Loop through length of segment after first fragment,
* make new header and copy data of each part and link onto chain.
m0 = m;
mhlen = sizeof (struct ip);
for (off = hlen + len; off < (u_short)ip->ip_len; off += len) {
- MGETHDR(m, M_DONTWAIT, MT_HEADER);
+ MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */
if (m == 0) {
error = ENOBUFS;
- ipstat.ips_odropped++;
+ OSAddAtomic(1, &ipstat.ips_odropped);
goto sendorfree;
}
m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
if (m->m_next == 0) {
(void) m_free(m);
error = ENOBUFS; /* ??? */
- ipstat.ips_odropped++;
+ OSAddAtomic(1, &ipstat.ips_odropped);
goto sendorfree;
}
m->m_pkthdr.len = mhlen + len;
m->m_pkthdr.rcvif = 0;
m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
m->m_pkthdr.socket_id = m0->m_pkthdr.socket_id;
- HTONS(mhip->ip_off);
- mhip->ip_sum = 0;
- if (sw_csum & CSUM_DELAY_IP) {
- mhip->ip_sum = in_cksum(m, mhlen);
- }
- *mnext = m;
- mnext = &m->m_nextpkt;
- nfrags++;
- }
- ipstat.ips_ofragments += nfrags;
-
- /* set first/last markers for fragment chain */
- m->m_flags |= M_LASTFRAG;
- m0->m_flags |= M_FIRSTFRAG | M_FRAG;
- m0->m_pkthdr.csum_data = nfrags;
-
- /*
- * Update first fragment by trimming what's been copied out
- * and updating header, then send each fragment (in order).
- */
- m = m0;
- m_adj(m, hlen + firstlen - (u_short)ip->ip_len);
- m->m_pkthdr.len = hlen + firstlen;
- ip->ip_len = htons((u_short)m->m_pkthdr.len);
- ip->ip_off |= IP_MF;
- HTONS(ip->ip_off);
- ip->ip_sum = 0;
- if (sw_csum & CSUM_DELAY_IP) {
- ip->ip_sum = in_cksum(m, hlen);
- }
-sendorfree:
-
- KERNEL_DEBUG(DBG_LAYER_END, ip->ip_dst.s_addr,
- ip->ip_src.s_addr, ip->ip_p, ip->ip_off, ip->ip_len);
-
- lck_mtx_unlock(ip_mutex);
- for (m = m0; m; m = m0) {
- m0 = m->m_nextpkt;
- m->m_nextpkt = 0;
-#if IPSEC
- /* clean ipsec history once it goes out of the node */
- if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0)
- ipsec_delaux(m);
+#if CONFIG_MACF_NET
+ mac_netinet_fragment(m0, m);
#endif
- if (error == 0) {
-#ifndef __APPLE__
- /* Record statistics for this interface address. */
- if (ia != NULL) {
- ia->ia_ifa.if_opackets++;
- ia->ia_ifa.if_obytes += m->m_pkthdr.len;
- }
-#endif
- if ((packetchain != 0) && (pktcnt > 0))
- panic("ip_output: mix of packet in packetlist is wrong=%x", packetlist);
- error = dlil_output(ifp, PF_INET, m, (void *) ro->ro_rt,
- (struct sockaddr *)dst, 0);
- } else
- m_freem(m);
- }
- if (error == 0)
- ipstat.ips_fragmented++;
- }
-done:
- if (ia) {
- ifafree(&ia->ia_ifa);
- ia = NULL;
- }
-#if IPSEC
- if (ipsec_bypass == 0 && (flags & IP_NOIPSEC) == 0) {
- if (ro == &iproute && ro->ro_rt) {
- rtfree(ro->ro_rt);
- ro->ro_rt = NULL;
- }
- if (sp != NULL) {
- KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
- printf("DP ip_output call free SP:%x\n", sp));
- lck_mtx_lock(sadb_mutex);
- key_freesp(sp);
- lck_mtx_unlock(sadb_mutex);
+#if BYTE_ORDER != BIG_ENDIAN
+ HTONS(mhip->ip_off);
+#endif
+
+ mhip->ip_sum = 0;
+ if (sw_csum & CSUM_DELAY_IP) {
+ mhip->ip_sum = in_cksum(m, mhlen);
+ }
+ *mnext = m;
+ mnext = &m->m_nextpkt;
+ nfrags++;
}
+ OSAddAtomic(nfrags, &ipstat.ips_ofragments);
+
+ /* set first/last markers for fragment chain */
+ m->m_flags |= M_LASTFRAG;
+ m0->m_flags |= M_FIRSTFRAG | M_FRAG;
+ m0->m_pkthdr.csum_data = nfrags;
+
+ /*
+ * Update first fragment by trimming what's been copied out
+ * and updating header, then send each fragment (in order).
+ */
+ m = m0;
+ m_adj(m, hlen + firstlen - (u_short)ip->ip_len);
+ m->m_pkthdr.len = hlen + firstlen;
+ ip->ip_len = htons((u_short)m->m_pkthdr.len);
+ ip->ip_off |= IP_MF;
+
+#if BYTE_ORDER != BIG_ENDIAN
+ HTONS(ip->ip_off);
+#endif
+
+ ip->ip_sum = 0;
+ if (sw_csum & CSUM_DELAY_IP) {
+ ip->ip_sum = in_cksum(m, hlen);
}
-#endif /* IPSEC */
+sendorfree:
+ if (error)
+ m_freem_list(m0);
- KERNEL_DEBUG(DBG_FNC_IP_OUTPUT | DBG_FUNC_END, error,0,0,0,0);
return (error);
-bad:
- m_freem(m0);
- lck_mtx_unlock(ip_mutex);
- goto done;
+}
+
+static void
+ip_out_cksum_stats(int proto, u_int32_t len)
+{
+ switch (proto) {
+ case IPPROTO_TCP:
+ tcp_out_cksum_stats(len);
+ break;
+ case IPPROTO_UDP:
+ udp_out_cksum_stats(len);
+ break;
+ default:
+ /* keep only TCP or UDP stats for now */
+ break;
+ }
}
void
ip_offset -= m->m_len;
m = m->m_next;
if (m == NULL) {
- printf("in_delayed_cksum_offset failed - ip_offset wasn't in the packet\n");
+ printf("in_delayed_cksum_withoffset failed - ip_offset wasn't in the packet\n");
return;
}
}
printf("delayed m_pullup, m->len: %d off: %d\n",
m->m_len, ip_offset);
#endif
- m_copydata(m, ip_offset, sizeof(struct ip), buf);
+ m_copydata(m, ip_offset, sizeof(struct ip), (caddr_t) buf);
ip = (struct ip *)buf;
} else {
csum = in_cksum_skip(m, ip_len, offset);
+ /* Update stats */
+ ip_out_cksum_stats(ip->ip_p, ip_len - offset);
+
if (m0->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
csum = 0xffff;
offset += m0->m_pkthdr.csum_data & 0xFFFF; /* checksum offset */
if (ip_offset + sizeof(struct ip) > m->m_len) {
#if DEBUG
- printf("in_cksum_offset - delayed m_pullup, m->len: %d off: %d\n",
+ printf("in_cksum_offset - delayed m_pullup, m->len: %d off: %lu\n",
m->m_len, ip_offset);
#endif
- m_copydata(m, ip_offset, sizeof(struct ip), buf);
+ m_copydata(m, ip_offset, sizeof(struct ip), (caddr_t) buf);
ip = (struct ip *)buf;
ip->ip_sum = 0;
if (ntohs(ip->ip_len) != (m->m_pkthdr.len - ip_offset)) {
ip->ip_len = SWAP16(ip->ip_len);
printf("in_cksum_offset: ip_len %d (%d) "
- "doesn't match actual length %d\n",
+ "doesn't match actual length %lu\n",
ip->ip_len, SWAP16(ip->ip_len),
(m->m_pkthdr.len - ip_offset));
return;
char tmp[2];
#if DEBUG
- printf("in_cksum_offset m_copyback, m->len: %d off: %d p: %d\n",
+ printf("in_cksum_offset m_copyback, m->len: %u off: %lu p: %d\n",
m->m_len, ip_offset + offsetof(struct ip, ip_sum), ip->ip_p);
#endif
*(u_short *)tmp = ip->ip_sum;
if (p->ipopt_dst.s_addr)
ip->ip_dst = p->ipopt_dst;
if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
- MGETHDR(n, M_DONTWAIT, MT_HEADER);
+ MGETHDR(n, M_DONTWAIT, MT_HEADER); /* MAC-OK */
if (n == 0)
return (m);
n->m_pkthdr.rcvif = 0;
+#if CONFIG_MACF_NET
+ mac_mbuf_label_copy(m, n);
+#endif
n->m_pkthdr.len = m->m_pkthdr.len + optlen;
m->m_len -= sizeof(struct ip);
m->m_data += sizeof(struct ip);
error = EMSGSIZE;
break;
}
- MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_HEADER);
+ MGET(m, sopt->sopt_p != kernproc ? M_WAIT : M_DONTWAIT,
+ MT_HEADER);
if (m == 0) {
error = ENOBUFS;
break;
break;
#undef OPTSET
+#if CONFIG_FORCE_OUT_IFP
+ /*
+ * Apple private interface, similar to IP_BOUND_IF, except
+ * that the parameter is a NULL-terminated string containing
+ * the name of the network interface; an emptry string means
+ * unbind. Applications are encouraged to use IP_BOUND_IF
+ * instead, as that is the current "official" API.
+ */
+ case IP_FORCE_OUT_IFP: {
+ char ifname[IFNAMSIZ];
+ unsigned int ifscope;
+
+ /* This option is settable only for IPv4 */
+ if (!(inp->inp_vflag & INP_IPV4)) {
+ error = EINVAL;
+ break;
+ }
+
+ /* Verify interface name parameter is sane */
+ if (sopt->sopt_valsize > sizeof(ifname)) {
+ error = EINVAL;
+ break;
+ }
+
+ /* Copy the interface name */
+ if (sopt->sopt_valsize != 0) {
+ error = sooptcopyin(sopt, ifname,
+ sizeof (ifname), sopt->sopt_valsize);
+ if (error)
+ break;
+ }
+
+ if (sopt->sopt_valsize == 0 || ifname[0] == NULL) {
+ /* Unbind this socket from any interface */
+ ifscope = IFSCOPE_NONE;
+ } else {
+ ifnet_t ifp;
+
+ /* Verify name is NULL terminated */
+ if (ifname[sopt->sopt_valsize - 1] != NULL) {
+ error = EINVAL;
+ break;
+ }
+
+ /* Bail out if given bogus interface name */
+ if (ifnet_find_by_name(ifname, &ifp) != 0) {
+ error = ENXIO;
+ break;
+ }
+
+ /* Bind this socket to this interface */
+ ifscope = ifp->if_index;
+
+ /*
+ * Won't actually free; since we don't release
+ * this later, we should do it now.
+ */
+ ifnet_release(ifp);
+ }
+ ip_bindif(inp, ifscope);
+ }
+ break;
+#endif
case IP_MULTICAST_IF:
case IP_MULTICAST_VIF:
case IP_MULTICAST_TTL:
break;
if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
break;
- priv = (sopt->sopt_p != NULL &&
- proc_suser(sopt->sopt_p) != 0) ? 0 : 1;
+ priv = (proc_suser(sopt->sopt_p) == 0);
if (m) {
req = mtod(m, caddr_t);
len = m->m_len;
}
optname = sopt->sopt_name;
- lck_mtx_lock(sadb_mutex);
error = ipsec4_set_policy(inp, optname, req, len, priv);
- lck_mtx_unlock(sadb_mutex);
m_freem(m);
break;
}
#endif /*IPSEC*/
+#if TRAFFIC_MGT
+ case IP_TRAFFIC_MGT_BACKGROUND:
+ {
+ unsigned background = 0;
+ error = sooptcopyin(sopt, &background, sizeof(background), sizeof(background));
+ if (error)
+ break;
+
+ if (background) {
+ socket_set_traffic_mgt_flags(so,
+ TRAFFIC_MGT_SO_BACKGROUND |
+ TRAFFIC_MGT_SO_BG_REGULATE);
+ } else {
+ socket_clear_traffic_mgt_flags(so,
+ TRAFFIC_MGT_SO_BACKGROUND |
+ TRAFFIC_MGT_SO_BG_REGULATE);
+ }
+
+ break;
+ }
+#endif /* TRAFFIC_MGT */
+
+ /*
+ * On a multihomed system, scoped routing can be used to
+ * restrict the source interface used for sending packets.
+ * The socket option IP_BOUND_IF binds a particular AF_INET
+ * socket to an interface such that data sent on the socket
+ * is restricted to that interface. This is unlike the
+ * SO_DONTROUTE option where the routing table is bypassed;
+ * therefore it allows for a greater flexibility and control
+ * over the system behavior, and does not place any restriction
+ * on the destination address type (e.g. unicast, multicast,
+ * or broadcast if applicable) or whether or not the host is
+ * directly reachable. Note that in the multicast transmit
+ * case, IP_MULTICAST_IF takes precedence over IP_BOUND_IF,
+ * since the former practically bypasses the routing table;
+ * in this case, IP_BOUND_IF sets the default interface used
+ * for sending multicast packets in the absence of an explicit
+ * transmit interface set via IP_MULTICAST_IF.
+ */
+ case IP_BOUND_IF:
+ /* This option is settable only for IPv4 */
+ if (!(inp->inp_vflag & INP_IPV4)) {
+ error = EINVAL;
+ break;
+ }
+
+ error = sooptcopyin(sopt, &optval, sizeof (optval),
+ sizeof (optval));
+
+ if (error)
+ break;
+
+ ip_bindif(inp, optval);
+ break;
+
default:
error = ENOPROTOOPT;
break;
req = mtod(m, caddr_t);
len = m->m_len;
}
- lck_mtx_lock(sadb_mutex);
error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
- lck_mtx_unlock(sadb_mutex);
if (error == 0)
error = soopt_mcopyout(sopt, m); /* XXX */
if (error == 0)
}
#endif /*IPSEC*/
+#if TRAFFIC_MGT
+ case IP_TRAFFIC_MGT_BACKGROUND:
+ {
+ unsigned background = so->so_traffic_mgt_flags;
+ return (sooptcopyout(sopt, &background, sizeof(background)));
+ break;
+ }
+#endif /* TRAFFIC_MGT */
+
+ case IP_BOUND_IF:
+ if (inp->inp_flags & INP_BOUND_IF)
+ optval = inp->inp_boundif;
+ error = sooptcopyout(sopt, &optval, sizeof (optval));
+ break;
+
default:
error = ENOPROTOOPT;
break;
* with destination address if source routed.
*/
static int
-ip_pcbopts(optname, pcbopt, m)
- int optname;
- struct mbuf **pcbopt;
- register struct mbuf *m;
+ip_pcbopts(
+ __unused int optname,
+ struct mbuf **pcbopt,
+ register struct mbuf *m)
{
register int cnt, optlen;
register u_char *cp;
struct ip_moptions **imop;
{
int error = 0;
- int i;
struct in_addr addr;
struct ip_mreq mreq;
struct ifnet *ifp = NULL;
switch (sopt->sopt_name) {
/* store an index number for the vif you wanna use in the send */
- case IP_MULTICAST_VIF:
- if (legal_vif_num == 0) {
- error = EOPNOTSUPP;
- break;
- }
- error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
- if (error)
- break;
- if (!legal_vif_num(i) && (i != -1)) {
- error = EINVAL;
+#if MROUTING
+ case IP_MULTICAST_VIF:
+ {
+ int i;
+ if (legal_vif_num == 0) {
+ error = EOPNOTSUPP;
+ break;
+ }
+ error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
+ if (error)
+ break;
+ if (!legal_vif_num(i) && (i != -1)) {
+ error = EINVAL;
+ break;
+ }
+ imo->imo_multicast_vif = i;
break;
}
- imo->imo_multicast_vif = i;
- break;
+#endif /* MROUTING */
case IP_MULTICAST_IF:
/*
* If all options have default values, no need to keep the mbuf.
*/
if (imo->imo_multicast_ifp == NULL &&
- imo->imo_multicast_vif == -1 &&
+ imo->imo_multicast_vif == (u_int32_t)-1 &&
imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
imo->imo_num_memberships == 0) {
struct ifnet *ifp = NULL;
int error = 0;
int i;
-
+
+ bzero((caddr_t)&ro, sizeof(ro));
+
if (!IN_MULTICAST(ntohl(mreq->imr_multiaddr.s_addr))) {
error = EINVAL;
- return error;
+ goto done;
}
/*
* If no interface address was provided, use the interface of
* the route to the given multicast address.
*/
if (mreq->imr_interface.s_addr == INADDR_ANY) {
- bzero((caddr_t)&ro, sizeof(ro));
dst = (struct sockaddr_in *)&ro.ro_dst;
dst->sin_len = sizeof(*dst);
dst->sin_family = AF_INET;
dst->sin_addr = mreq->imr_multiaddr;
- rtalloc(&ro);
+ rtalloc_ign(&ro, 0);
if (ro.ro_rt != NULL) {
ifp = ro.ro_rt->rt_ifp;
- rtfree(ro.ro_rt);
- }
- else {
+ } else {
/* If there's no default route, try using loopback */
- mreq->imr_interface.s_addr = INADDR_LOOPBACK;
+ mreq->imr_interface.s_addr = htonl(INADDR_LOOPBACK);
}
}
-
+
if (ifp == NULL) {
ifp = ip_multicast_if(&mreq->imr_interface, NULL);
}
*/
if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
error = EADDRNOTAVAIL;
- return error;
+ goto done;
}
/*
* See if the membership already exists or if all the
}
if (i < imo->imo_num_memberships) {
error = EADDRINUSE;
- return error;
+ goto done;
}
if (i == IP_MAX_MEMBERSHIPS) {
error = ETOOMANYREFS;
- return error;
+ goto done;
}
/*
* Everything looks good; add a new record to the multicast
if ((imo->imo_membership[i] =
in_addmulti(&mreq->imr_multiaddr, ifp)) == NULL) {
error = ENOBUFS;
- return error;
+ goto done;
}
++imo->imo_num_memberships;
-
+
+done:
+ if (ro.ro_rt != NULL)
+ rtfree(ro.ro_rt);
+
return error;
}
error = 0;
switch (sopt->sopt_name) {
+#if MROUTING
case IP_MULTICAST_VIF:
if (imo != NULL)
optval = imo->imo_multicast_vif;
optval = -1;
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
+#endif /* MROUTING */
case IP_MULTICAST_IF:
if (imo == NULL || imo->imo_multicast_ifp == NULL)
IFP_TO_IA(imo->imo_multicast_ifp, ia);
addr.s_addr = (ia == NULL) ? INADDR_ANY
: IA_SIN(ia)->sin_addr.s_addr;
+ if (ia != NULL)
+ ifafree(&ia->ia_ifa);
}
error = sooptcopyout(sopt, &addr, sizeof addr);
break;
{
register struct ip *ip;
struct mbuf *copym;
+ int sw_csum = (apple_hwcksum_tx == 0);
copym = m_copy(m, 0, M_COPYALL);
if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
copym = m_pullup(copym, hlen);
- if (copym != NULL) {
- /*
- * We don't bother to fragment if the IP length is greater
- * than the interface's MTU. Can this possibly matter?
- */
- ip = mtod(copym, struct ip *);
- HTONS(ip->ip_len);
- HTONS(ip->ip_off);
- ip->ip_sum = 0;
- ip->ip_sum = in_cksum(copym, hlen);
- /*
- * NB:
- * It's not clear whether there are any lingering
- * reentrancy problems in other areas which might
- * be exposed by using ip_input directly (in
- * particular, everything which modifies the packet
- * in-place). Yet another option is using the
- * protosw directly to deliver the looped back
- * packet. For the moment, we'll err on the side
- * of safety by using if_simloop().
- */
+
+ if (copym == NULL)
+ return;
+
+ /*
+ * We don't bother to fragment if the IP length is greater
+ * than the interface's MTU. Can this possibly matter?
+ */
+ ip = mtod(copym, struct ip *);
+
+#if BYTE_ORDER != BIG_ENDIAN
+ HTONS(ip->ip_len);
+ HTONS(ip->ip_off);
+#endif
+
+ ip->ip_sum = 0;
+ ip->ip_sum = in_cksum(copym, hlen);
+ /*
+ * NB:
+ * It's not clear whether there are any lingering
+ * reentrancy problems in other areas which might
+ * be exposed by using ip_input directly (in
+ * particular, everything which modifies the packet
+ * in-place). Yet another option is using the
+ * protosw directly to deliver the looped back
+ * packet. For the moment, we'll err on the side
+ * of safety by using if_simloop().
+ */
#if 1 /* XXX */
- if (dst->sin_family != AF_INET) {
- printf("ip_mloopback: bad address family %d\n",
- dst->sin_family);
- dst->sin_family = AF_INET;
- }
+ if (dst->sin_family != AF_INET) {
+ printf("ip_mloopback: bad address family %d\n",
+ dst->sin_family);
+ dst->sin_family = AF_INET;
+ }
#endif
-
/*
- * Mark checksum as valid or calculate checksum for loopback.
- *
- * This is done this way because we have to embed the ifp of
- * the interface we will send the original copy of the packet
- * out on in the mbuf. ip_input will check if_hwassist of the
- * embedded ifp and ignore all csum_flags if if_hwassist is 0.
- * The UDP checksum has not been calculated yet.
- */
- if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
- if (IF_HWASSIST_CSUM_FLAGS(ifp->if_hwassist)) {
- copym->m_pkthdr.csum_flags |=
- CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
- CSUM_IP_CHECKED | CSUM_IP_VALID;
- copym->m_pkthdr.csum_data = 0xffff;
- } else {
- NTOHS(ip->ip_len);
- in_delayed_cksum(copym);
- HTONS(ip->ip_len);
- }
+ * Mark checksum as valid or calculate checksum for loopback.
+ *
+ * This is done this way because we have to embed the ifp of
+ * the interface we will send the original copy of the packet
+ * out on in the mbuf. ip_input will check if_hwassist of the
+ * embedded ifp and ignore all csum_flags if if_hwassist is 0.
+ * The UDP checksum has not been calculated yet.
+ */
+ if (sw_csum || (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
+ if (!sw_csum && IF_HWASSIST_CSUM_FLAGS(ifp->if_hwassist)) {
+ copym->m_pkthdr.csum_flags |=
+ CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
+ CSUM_IP_CHECKED | CSUM_IP_VALID;
+ copym->m_pkthdr.csum_data = 0xffff;
+ } else {
+
+#if BYTE_ORDER != BIG_ENDIAN
+ NTOHS(ip->ip_len);
+#endif
+
+ in_delayed_cksum(copym);
+
+#if BYTE_ORDER != BIG_ENDIAN
+ HTONS(ip->ip_len);
+#endif
+
+ }
}
+ /*
+ * TedW:
+ * We need to send all loopback traffic down to dlil in case
+ * a filter has tapped-in.
+ */
+
+ /*
+ * Stuff the 'real' ifp into the pkthdr, to be used in matching
+ * in ip_input(); we need the loopback ifp/dl_tag passed as args
+ * to make the loopback driver compliant with the data link
+ * requirements.
+ */
+ if (lo_ifp) {
+ copym->m_pkthdr.rcvif = ifp;
+ dlil_output(lo_ifp, PF_INET, copym, 0,
+ (struct sockaddr *) dst, 0);
+ } else {
+ printf("Warning: ip_output call to dlil_find_dltag failed!\n");
+ m_freem(copym);
+ }
+}
+
+/*
+ * Given a source IP address (and route, if available), determine the best
+ * interface to send the packet from. Checking for (and updating) the
+ * ROF_SRCIF_SELECTED flag in the pcb-supplied route placeholder is done
+ * without any locks based on the assumption that ip_output() is single-
+ * threaded per-pcb, i.e. for any given pcb there can only be one thread
+ * performing output at the IP layer.
+ */
+static struct ifaddr *
+in_selectsrcif(struct ip *ip, struct route *ro, unsigned int ifscope)
+{
+ struct ifaddr *ifa = NULL;
+ struct in_addr src = ip->ip_src;
+ struct in_addr dst = ip->ip_dst;
+ struct ifnet *rt_ifp;
+ char s_src[MAX_IPv4_STR_LEN], s_dst[MAX_IPv4_STR_LEN];
+
+ if (ip_select_srcif_debug) {
+ (void) inet_ntop(AF_INET, &src.s_addr, s_src, sizeof (s_src));
+ (void) inet_ntop(AF_INET, &dst.s_addr, s_dst, sizeof (s_dst));
+ }
+
+ if (ro->ro_rt != NULL)
+ RT_LOCK(ro->ro_rt);
+
+ rt_ifp = (ro->ro_rt != NULL) ? ro->ro_rt->rt_ifp : NULL;
+
+ /*
+ * Given the source IP address, find a suitable source interface
+ * to use for transmission; if the caller has specified a scope,
+ * optimize the search by looking at the addresses only for that
+ * interface. This is still suboptimal, however, as we need to
+ * traverse the per-interface list.
+ */
+ if (ifscope != IFSCOPE_NONE || ro->ro_rt != NULL) {
+ unsigned int scope = ifscope;
/*
- * TedW:
- * We need to send all loopback traffic down to dlil in case
- * a filter has tapped-in.
+ * If no scope is specified and the route is stale (pointing
+ * to a defunct interface) use the current primary interface;
+ * this happens when switching between interfaces configured
+ * with the same IP address. Otherwise pick up the scope
+ * information from the route; the ULP may have looked up a
+ * correct route and we just need to verify it here and mark
+ * it with the ROF_SRCIF_SELECTED flag below.
*/
+ if (scope == IFSCOPE_NONE) {
+ scope = rt_ifp->if_index;
+ if (scope != get_primary_ifscope() &&
+ ro->ro_rt->generation_id != route_generation)
+ scope = get_primary_ifscope();
+ }
+
+ ifa = (struct ifaddr *)ifa_foraddr_scoped(src.s_addr, scope);
+
+ if (ifa == NULL && ip->ip_p != IPPROTO_UDP &&
+ ip->ip_p != IPPROTO_TCP && ipforwarding) {
+ /*
+ * If forwarding is enabled, and if the packet isn't
+ * TCP or UDP, check if the source address belongs
+ * to one of our own interfaces; if so, demote the
+ * interface scope and do a route lookup right below.
+ */
+ ifa = (struct ifaddr *)ifa_foraddr(src.s_addr);
+ if (ifa != NULL) {
+ ifafree(ifa);
+ ifa = NULL;
+ ifscope = IFSCOPE_NONE;
+ }
+ }
+
+ if (ip_select_srcif_debug && ifa != NULL) {
+ if (ro->ro_rt != NULL) {
+ printf("%s->%s ifscope %d->%d ifa_if %s%d "
+ "ro_if %s%d\n", s_src, s_dst, ifscope,
+ scope, ifa->ifa_ifp->if_name,
+ ifa->ifa_ifp->if_unit, rt_ifp->if_name,
+ rt_ifp->if_unit);
+ } else {
+ printf("%s->%s ifscope %d->%d ifa_if %s%d\n",
+ s_src, s_dst, ifscope, scope,
+ ifa->ifa_ifp->if_name,
+ ifa->ifa_ifp->if_unit);
+ }
+ }
+ }
+
+ /*
+ * Slow path; search for an interface having the corresponding source
+ * IP address if the scope was not specified by the caller, and:
+ *
+ * 1) There currently isn't any route, or,
+ * 2) The interface used by the route does not own that source
+ * IP address; in this case, the route will get blown away
+ * and we'll do a more specific scoped search using the newly
+ * found interface.
+ */
+ if (ifa == NULL && ifscope == IFSCOPE_NONE) {
+ ifa = (struct ifaddr *)ifa_foraddr(src.s_addr);
/*
- * Stuff the 'real' ifp into the pkthdr, to be used in matching
- * in ip_input(); we need the loopback ifp/dl_tag passed as args
- * to make the loopback driver compliant with the data link
- * requirements.
+ * If we have the IP address, but not the route, we don't
+ * really know whether or not it belongs to the correct
+ * interface (it could be shared across multiple interfaces.)
+ * The only way to find out is to do a route lookup.
*/
- if (lo_ifp) {
- copym->m_pkthdr.rcvif = ifp;
- dlil_output(lo_ifp, PF_INET, copym, 0, (struct sockaddr *) dst, 0);
- } else {
- printf("Warning: ip_output call to dlil_find_dltag failed!\n");
- m_freem(copym);
+ if (ifa != NULL && ro->ro_rt == NULL) {
+ struct rtentry *rt;
+ struct sockaddr_in sin;
+ struct ifaddr *oifa = NULL;
+
+ bzero(&sin, sizeof (sin));
+ sin.sin_family = AF_INET;
+ sin.sin_len = sizeof (sin);
+ sin.sin_addr = dst;
+
+ lck_mtx_lock(rnh_lock);
+ if ((rt = rt_lookup(TRUE, (struct sockaddr *)&sin, NULL,
+ rt_tables[AF_INET], IFSCOPE_NONE)) != NULL) {
+ RT_LOCK(rt);
+ /*
+ * If the route uses a different interface,
+ * use that one instead. The IP address of
+ * the ifaddr that we pick up here is not
+ * relevant.
+ */
+ if (ifa->ifa_ifp != rt->rt_ifp) {
+ oifa = ifa;
+ ifa = rt->rt_ifa;
+ ifaref(ifa);
+ RT_UNLOCK(rt);
+ } else {
+ RT_UNLOCK(rt);
+ }
+ rtfree_locked(rt);
+ }
+ lck_mtx_unlock(rnh_lock);
+
+ if (oifa != NULL) {
+ struct ifaddr *iifa;
+
+ /*
+ * See if the interface pointed to by the
+ * route is configured with the source IP
+ * address of the packet.
+ */
+ iifa = (struct ifaddr *)ifa_foraddr_scoped(
+ src.s_addr, ifa->ifa_ifp->if_index);
+
+ if (iifa != NULL) {
+ /*
+ * Found it; drop the original one
+ * as well as the route interface
+ * address, and use this instead.
+ */
+ ifafree(oifa);
+ ifafree(ifa);
+ ifa = iifa;
+ } else if (!ipforwarding ||
+ (rt->rt_flags & RTF_GATEWAY)) {
+ /*
+ * This interface doesn't have that
+ * source IP address; drop the route
+ * interface address and just use the
+ * original one, and let the caller
+ * do a scoped route lookup.
+ */
+ ifafree(ifa);
+ ifa = oifa;
+ } else {
+ /*
+ * Forwarding is enabled and the source
+ * address belongs to one of our own
+ * interfaces which isn't the outgoing
+ * interface, and we have a route, and
+ * the destination is on a network that
+ * is directly attached (onlink); drop
+ * the original one and use the route
+ * interface address instead.
+ */
+ ifafree(oifa);
+ }
+ }
+ } else if (ifa != NULL && ro->ro_rt != NULL &&
+ !(ro->ro_rt->rt_flags & RTF_GATEWAY) &&
+ ifa->ifa_ifp != ro->ro_rt->rt_ifp && ipforwarding) {
+ /*
+ * Forwarding is enabled and the source address belongs
+ * to one of our own interfaces which isn't the same
+ * as the interface used by the known route; drop the
+ * original one and use the route interface address.
+ */
+ ifafree(ifa);
+ ifa = ro->ro_rt->rt_ifa;
+ ifaref(ifa);
+ }
+
+ if (ip_select_srcif_debug && ifa != NULL) {
+ printf("%s->%s ifscope %d ifa_if %s%d\n",
+ s_src, s_dst, ifscope, ifa->ifa_ifp->if_name,
+ ifa->ifa_ifp->if_unit);
+ }
+ }
+
+ if (ro->ro_rt != NULL)
+ RT_LOCK_ASSERT_HELD(ro->ro_rt);
+ /*
+ * If there is a non-loopback route with the wrong interface, or if
+ * there is no interface configured with such an address, blow it
+ * away. Except for local/loopback, we look for one with a matching
+ * interface scope/index.
+ */
+ if (ro->ro_rt != NULL &&
+ (ifa == NULL || (ifa->ifa_ifp != rt_ifp && rt_ifp != lo_ifp) ||
+ !(ro->ro_rt->rt_flags & RTF_UP))) {
+ if (ip_select_srcif_debug) {
+ if (ifa != NULL) {
+ printf("%s->%s ifscope %d ro_if %s%d != "
+ "ifa_if %s%d (cached route cleared)\n",
+ s_src, s_dst, ifscope, rt_ifp->if_name,
+ rt_ifp->if_unit, ifa->ifa_ifp->if_name,
+ ifa->ifa_ifp->if_unit);
+ } else {
+ printf("%s->%s ifscope %d ro_if %s%d "
+ "(no ifa_if found)\n",
+ s_src, s_dst, ifscope, rt_ifp->if_name,
+ rt_ifp->if_unit);
+ }
+ }
+
+ RT_UNLOCK(ro->ro_rt);
+ rtfree(ro->ro_rt);
+ ro->ro_rt = NULL;
+ ro->ro_flags &= ~ROF_SRCIF_SELECTED;
+
+ /*
+ * If the destination is IPv4 LLA and the route's interface
+ * doesn't match the source interface, then the source IP
+ * address is wrong; it most likely belongs to the primary
+ * interface associated with the IPv4 LL subnet. Drop the
+ * packet rather than letting it go out and return an error
+ * to the ULP. This actually applies not only to IPv4 LL
+ * but other shared subnets; for now we explicitly test only
+ * for the former case and save the latter for future.
+ */
+ if (IN_LINKLOCAL(ntohl(dst.s_addr)) &&
+ !IN_LINKLOCAL(ntohl(src.s_addr)) && ifa != NULL) {
+ ifafree(ifa);
+ ifa = NULL;
}
+ }
+
+ if (ip_select_srcif_debug && ifa == NULL) {
+ printf("%s->%s ifscope %d (neither ro_if/ifa_if found)\n",
+ s_src, s_dst, ifscope);
+ }
+
+ /*
+ * If there is a route, mark it accordingly. If there isn't one,
+ * we'll get here again during the next transmit (possibly with a
+ * route) and the flag will get set at that point. For IPv4 LLA
+ * destination, mark it only if the route has been fully resolved;
+ * otherwise we want to come back here again when the route points
+ * to the interface over which the ARP reply arrives on.
+ */
+ if (ro->ro_rt != NULL && (!IN_LINKLOCAL(ntohl(dst.s_addr)) ||
+ (ro->ro_rt->rt_gateway->sa_family == AF_LINK &&
+ SDL(ro->ro_rt->rt_gateway)->sdl_alen != 0))) {
+ ro->ro_flags |= ROF_SRCIF_SELECTED;
+ ro->ro_rt->generation_id = route_generation;
+ }
+
+ if (ro->ro_rt != NULL)
+ RT_UNLOCK(ro->ro_rt);
+
+ return (ifa);
+}
+
+/*
+ * Handler for setting IP_FORCE_OUT_IFP or IP_BOUND_IF socket option.
+ */
+static void
+ip_bindif(struct inpcb *inp, unsigned int ifscope)
+{
+ /*
+ * A zero interface scope value indicates an "unbind".
+ * Otherwise, take in whatever value the app desires;
+ * the app may already know the scope (or force itself
+ * to such a scope) ahead of time before the interface
+ * gets attached. It doesn't matter either way; any
+ * route lookup from this point on will require an
+ * exact match for the embedded interface scope.
+ */
+ inp->inp_boundif = ifscope;
+ if (inp->inp_boundif == IFSCOPE_NONE)
+ inp->inp_flags &= ~INP_BOUND_IF;
+ else
+ inp->inp_flags |= INP_BOUND_IF;
-/* if_simloop(ifp, copym, (struct sockaddr *)dst, 0);*/
+ /* Blow away any cached route in the PCB */
+ if (inp->inp_route.ro_rt != NULL) {
+ rtfree(inp->inp_route.ro_rt);
+ inp->inp_route.ro_rt = NULL;
}
}
projects / apple / xnu.git / blobdiff