2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)ip_output.c 8.3 (Berkeley) 1/21/94
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_ipsec.h"
39 #include "opt_kern_tls.h"
40 #include "opt_mbuf_stress_test.h"
41 #include "opt_mpath.h"
42 #include "opt_ratelimit.h"
43 #include "opt_route.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
52 #include <sys/malloc.h>
56 #include <sys/protosw.h>
57 #include <sys/rmlock.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/ucred.h>
65 #include <net/if_var.h>
66 #include <net/if_llatbl.h>
67 #include <net/netisr.h>
69 #include <net/route.h>
70 #include <net/route/nhop.h>
71 #include <net/rss_config.h>
74 #include <netinet/in.h>
75 #include <netinet/in_fib.h>
76 #include <netinet/in_kdtrace.h>
77 #include <netinet/in_systm.h>
78 #include <netinet/ip.h>
79 #include <netinet/in_fib.h>
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_rss.h>
82 #include <netinet/in_var.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/ip_options.h>
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
89 #if defined(SCTP) || defined(SCTP_SUPPORT)
90 #include <netinet/sctp.h>
91 #include <netinet/sctp_crc32.h>
94 #include <netipsec/ipsec_support.h>
96 #include <machine/in_cksum.h>
98 #include <security/mac/mac_framework.h>
100 #ifdef MBUF_STRESS_TEST
101 static int mbuf_frag_size = 0;
102 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
103 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
106 static void ip_mloopback(struct ifnet *, const struct mbuf *, int);
108 extern int in_mcast_loop;
109 extern struct protosw inetsw[];
112 ip_output_pfil(struct mbuf **mp, struct ifnet *ifp, int flags,
113 struct inpcb *inp, struct sockaddr_in *dst, int *fibnum, int *error)
115 struct m_tag *fwd_tag = NULL;
119 int pflags = PFIL_OUT;
121 if (flags & IP_FORWARDING)
125 ip = mtod(m, struct ip *);
127 /* Run through list of hooks for output packets. */
128 odst.s_addr = ip->ip_dst.s_addr;
129 switch (pfil_run_hooks(V_inet_pfil_head, mp, ifp, pflags, inp)) {
134 return 1; /* Finished */
139 ip = mtod(m, struct ip *);
141 /* See if destination IP address was changed by packet filter. */
142 if (odst.s_addr != ip->ip_dst.s_addr) {
143 m->m_flags |= M_SKIP_FIREWALL;
144 /* If destination is now ourself drop to ip_input(). */
145 if (in_localip(ip->ip_dst)) {
146 m->m_flags |= M_FASTFWD_OURS;
147 if (m->m_pkthdr.rcvif == NULL)
148 m->m_pkthdr.rcvif = V_loif;
149 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
150 m->m_pkthdr.csum_flags |=
151 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
152 m->m_pkthdr.csum_data = 0xffff;
154 m->m_pkthdr.csum_flags |=
155 CSUM_IP_CHECKED | CSUM_IP_VALID;
156 #if defined(SCTP) || defined(SCTP_SUPPORT)
157 if (m->m_pkthdr.csum_flags & CSUM_SCTP)
158 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
160 *error = netisr_queue(NETISR_IP, m);
161 return 1; /* Finished */
164 bzero(dst, sizeof(*dst));
165 dst->sin_family = AF_INET;
166 dst->sin_len = sizeof(*dst);
167 dst->sin_addr = ip->ip_dst;
169 return -1; /* Reloop */
171 /* See if fib was changed by packet filter. */
172 if ((*fibnum) != M_GETFIB(m)) {
173 m->m_flags |= M_SKIP_FIREWALL;
174 *fibnum = M_GETFIB(m);
175 return -1; /* Reloop for FIB change */
178 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
179 if (m->m_flags & M_FASTFWD_OURS) {
180 if (m->m_pkthdr.rcvif == NULL)
181 m->m_pkthdr.rcvif = V_loif;
182 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
183 m->m_pkthdr.csum_flags |=
184 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
185 m->m_pkthdr.csum_data = 0xffff;
187 #if defined(SCTP) || defined(SCTP_SUPPORT)
188 if (m->m_pkthdr.csum_flags & CSUM_SCTP)
189 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
191 m->m_pkthdr.csum_flags |=
192 CSUM_IP_CHECKED | CSUM_IP_VALID;
194 *error = netisr_queue(NETISR_IP, m);
195 return 1; /* Finished */
197 /* Or forward to some other address? */
198 if ((m->m_flags & M_IP_NEXTHOP) &&
199 ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
200 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
201 m->m_flags |= M_SKIP_FIREWALL;
202 m->m_flags &= ~M_IP_NEXTHOP;
203 m_tag_delete(m, fwd_tag);
205 return -1; /* Reloop for CHANGE of dst */
212 ip_output_send(struct inpcb *inp, struct ifnet *ifp, struct mbuf *m,
213 const struct sockaddr_in *gw, struct route *ro, bool stamp_tag)
216 struct ktls_session *tls = NULL;
218 struct m_snd_tag *mst;
221 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
226 * If this is an unencrypted TLS record, save a reference to
227 * the record. This local reference is used to call
228 * ktls_output_eagain after the mbuf has been freed (thus
229 * dropping the mbuf's reference) in if_output.
231 if (m->m_next != NULL && mbuf_has_tls_session(m->m_next)) {
232 tls = ktls_hold(m->m_next->m_epg_tls);
236 * If a TLS session doesn't have a valid tag, it must
237 * have had an earlier ifp mismatch, so drop this
245 * Always stamp tags that include NIC ktls.
251 if (inp != NULL && mst == NULL) {
252 if ((inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) != 0 ||
253 (inp->inp_snd_tag != NULL &&
254 inp->inp_snd_tag->ifp != ifp))
255 in_pcboutput_txrtlmt(inp, ifp, m);
257 if (inp->inp_snd_tag != NULL)
258 mst = inp->inp_snd_tag;
261 if (stamp_tag && mst != NULL) {
262 KASSERT(m->m_pkthdr.rcvif == NULL,
263 ("trying to add a send tag to a forwarded packet"));
264 if (mst->ifp != ifp) {
269 /* stamp send tag on mbuf */
270 m->m_pkthdr.snd_tag = m_snd_tag_ref(mst);
271 m->m_pkthdr.csum_flags |= CSUM_SND_TAG;
274 error = (*ifp->if_output)(ifp, m, (const struct sockaddr *)gw, ro);
277 /* Check for route change invalidating send tags. */
281 error = ktls_output_eagain(inp, tls);
287 in_pcboutput_eagain(inp);
292 /* rte<>ro_flags translation */
294 rt_update_ro_flags(struct route *ro)
296 int nh_flags = ro->ro_nh->nh_flags;
298 ro->ro_flags &= ~ (RT_REJECT|RT_BLACKHOLE|RT_HAS_GW);
300 ro->ro_flags |= (nh_flags & NHF_REJECT) ? RT_REJECT : 0;
301 ro->ro_flags |= (nh_flags & NHF_BLACKHOLE) ? RT_BLACKHOLE : 0;
302 ro->ro_flags |= (nh_flags & NHF_GATEWAY) ? RT_HAS_GW : 0;
306 * IP output. The packet in mbuf chain m contains a skeletal IP
307 * header (with len, off, ttl, proto, tos, src, dst).
308 * The mbuf chain containing the packet will be freed.
309 * The mbuf opt, if present, will not be freed.
310 * If route ro is present and has ro_rt initialized, route lookup would be
311 * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL,
312 * then result of route lookup is stored in ro->ro_rt.
314 * In the IP forwarding case, the packet will arrive with options already
315 * inserted, so must have a NULL opt pointer.
318 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
319 struct ip_moptions *imo, struct inpcb *inp)
321 struct rm_priotracker in_ifa_tracker;
323 struct ifnet *ifp = NULL; /* keep compiler happy */
325 int hlen = sizeof (struct ip);
328 struct sockaddr_in *dst, sin;
329 const struct sockaddr_in *gw;
330 struct in_ifaddr *ia = NULL;
333 uint16_t ip_len, ip_off;
335 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
336 int no_route_but_check_spd = 0;
343 INP_LOCK_ASSERT(inp);
344 M_SETFIB(m, inp->inp_inc.inc_fibnum);
345 if ((flags & IP_NODEFAULTFLOWID) == 0) {
346 m->m_pkthdr.flowid = inp->inp_flowid;
347 M_HASHTYPE_SET(m, inp->inp_flowtype);
350 m->m_pkthdr.numa_domain = inp->inp_numa_domain;
356 m = ip_insertoptions(m, opt, &len);
358 hlen = len; /* ip->ip_hl is updated above */
360 ip = mtod(m, struct ip *);
361 ip_len = ntohs(ip->ip_len);
362 ip_off = ntohs(ip->ip_off);
364 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
365 ip->ip_v = IPVERSION;
366 ip->ip_hl = hlen >> 2;
369 /* Header already set, fetch hlen from there */
370 hlen = ip->ip_hl << 2;
372 if ((flags & IP_FORWARDING) == 0)
373 IPSTAT_INC(ips_localout);
378 * gw is readonly but can point either to dst OR rt_gateway,
379 * therefore we need restore gw if we're redoing lookup.
381 fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m);
383 dst = (struct sockaddr_in *)&ro->ro_dst;
386 if (ro == NULL || ro->ro_nh == NULL) {
387 bzero(dst, sizeof(*dst));
388 dst->sin_family = AF_INET;
389 dst->sin_len = sizeof(*dst);
390 dst->sin_addr = ip->ip_dst;
395 * Validate route against routing table additions;
396 * a better/more specific route might have been added.
398 if (inp != NULL && ro != NULL && ro->ro_nh != NULL)
399 NH_VALIDATE(ro, &inp->inp_rt_cookie, fibnum);
401 * If there is a cached route,
402 * check that it is to the same destination
403 * and is still up. If not, free it and try again.
404 * The address family should also be checked in case of sharing the
406 * Also check whether routing cache needs invalidation.
408 if (ro != NULL && ro->ro_nh != NULL &&
409 ((!NH_IS_VALID(ro->ro_nh)) || dst->sin_family != AF_INET ||
410 dst->sin_addr.s_addr != ip->ip_dst.s_addr))
411 RO_INVALIDATE_CACHE(ro);
414 * If routing to interface only, short circuit routing lookup.
415 * The use of an all-ones broadcast address implies this; an
416 * interface is specified by the broadcast address of an interface,
417 * or the destination address of a ptp interface.
419 if (flags & IP_SENDONES) {
420 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst),
421 M_GETFIB(m)))) == NULL &&
422 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
423 M_GETFIB(m)))) == NULL) {
424 IPSTAT_INC(ips_noroute);
428 ip->ip_dst.s_addr = INADDR_BROADCAST;
429 dst->sin_addr = ip->ip_dst;
434 src = IA_SIN(ia)->sin_addr;
435 } else if (flags & IP_ROUTETOIF) {
436 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
437 M_GETFIB(m)))) == NULL &&
438 (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0,
439 M_GETFIB(m)))) == NULL) {
440 IPSTAT_INC(ips_noroute);
447 isbroadcast = ifp->if_flags & IFF_BROADCAST ?
448 in_ifaddr_broadcast(dst->sin_addr, ia) : 0;
449 src = IA_SIN(ia)->sin_addr;
450 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
451 imo != NULL && imo->imo_multicast_ifp != NULL) {
453 * Bypass the normal routing lookup for multicast
454 * packets if the interface is specified.
456 ifp = imo->imo_multicast_ifp;
458 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
459 isbroadcast = 0; /* fool gcc */
460 /* Interface may have no addresses. */
462 src = IA_SIN(ia)->sin_addr;
464 src.s_addr = INADDR_ANY;
465 } else if (ro != NULL) {
466 if (ro->ro_nh == NULL) {
468 * We want to do any cloning requested by the link
469 * layer, as this is probably required in all cases
470 * for correct operation (as it is for ARP).
474 flowid = ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr);
476 flowid = m->m_pkthdr.flowid;
478 ro->ro_nh = fib4_lookup(fibnum, dst->sin_addr, 0,
481 if (ro->ro_nh == NULL || (!NH_IS_VALID(ro->ro_nh))) {
482 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
484 * There is no route for this packet, but it is
485 * possible that a matching SPD entry exists.
487 no_route_but_check_spd = 1;
490 IPSTAT_INC(ips_noroute);
491 error = EHOSTUNREACH;
495 ia = ifatoia(ro->ro_nh->nh_ifa);
496 ifp = ro->ro_nh->nh_ifp;
497 counter_u64_add(ro->ro_nh->nh_pksent, 1);
498 rt_update_ro_flags(ro);
499 if (ro->ro_nh->nh_flags & NHF_GATEWAY)
500 gw = &ro->ro_nh->gw4_sa;
501 if (ro->ro_nh->nh_flags & NHF_HOST)
502 isbroadcast = (ro->ro_nh->nh_flags & NHF_BROADCAST);
503 else if (ifp->if_flags & IFF_BROADCAST)
504 isbroadcast = in_ifaddr_broadcast(gw->sin_addr, ia);
507 if (ro->ro_nh->nh_flags & NHF_HOST)
508 mtu = ro->ro_nh->nh_mtu;
511 src = IA_SIN(ia)->sin_addr;
513 struct nhop_object *nh;
515 nh = fib4_lookup(M_GETFIB(m), ip->ip_dst, 0, NHR_NONE,
518 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
520 * There is no route for this packet, but it is
521 * possible that a matching SPD entry exists.
523 no_route_but_check_spd = 1;
526 IPSTAT_INC(ips_noroute);
527 error = EHOSTUNREACH;
533 * We are rewriting here dst to be gw actually, contradicting
534 * comment at the beginning of the function. However, in this
535 * case we are always dealing with on stack dst.
536 * In case if pfil(9) sends us back to beginning of the
537 * function, the dst would be rewritten by ip_output_pfil().
540 if (nh->nh_flags & NHF_GATEWAY)
541 dst->sin_addr = nh->gw4_sa.sin_addr;
542 ia = ifatoia(nh->nh_ifa);
543 src = IA_SIN(ia)->sin_addr;
544 isbroadcast = (((nh->nh_flags & (NHF_HOST | NHF_BROADCAST)) ==
545 (NHF_HOST | NHF_BROADCAST)) ||
546 ((ifp->if_flags & IFF_BROADCAST) &&
547 in_ifaddr_broadcast(dst->sin_addr, ia)));
550 /* Catch a possible divide by zero later. */
551 KASSERT(mtu > 0, ("%s: mtu %d <= 0, ro=%p (nh_flags=0x%08x) ifp=%p",
553 (ro != NULL && ro->ro_nh != NULL) ? ro->ro_nh->nh_flags : 0, ifp));
555 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
556 m->m_flags |= M_MCAST;
558 * IP destination address is multicast. Make sure "gw"
559 * still points to the address in "ro". (It may have been
560 * changed to point to a gateway address, above.)
564 * See if the caller provided any multicast options
567 ip->ip_ttl = imo->imo_multicast_ttl;
568 if (imo->imo_multicast_vif != -1)
571 ip_mcast_src(imo->imo_multicast_vif) :
574 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
576 * Confirm that the outgoing interface supports multicast.
578 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
579 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
580 IPSTAT_INC(ips_noroute);
586 * If source address not specified yet, use address
587 * of outgoing interface.
589 if (ip->ip_src.s_addr == INADDR_ANY)
592 if ((imo == NULL && in_mcast_loop) ||
593 (imo && imo->imo_multicast_loop)) {
595 * Loop back multicast datagram if not expressly
596 * forbidden to do so, even if we are not a member
597 * of the group; ip_input() will filter it later,
598 * thus deferring a hash lookup and mutex acquisition
599 * at the expense of a cheap copy using m_copym().
601 ip_mloopback(ifp, m, hlen);
604 * If we are acting as a multicast router, perform
605 * multicast forwarding as if the packet had just
606 * arrived on the interface to which we are about
607 * to send. The multicast forwarding function
608 * recursively calls this function, using the
609 * IP_FORWARDING flag to prevent infinite recursion.
611 * Multicasts that are looped back by ip_mloopback(),
612 * above, will be forwarded by the ip_input() routine,
615 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) {
617 * If rsvp daemon is not running, do not
618 * set ip_moptions. This ensures that the packet
619 * is multicast and not just sent down one link
620 * as prescribed by rsvpd.
625 ip_mforward(ip, ifp, m, imo) != 0) {
633 * Multicasts with a time-to-live of zero may be looped-
634 * back, above, but must not be transmitted on a network.
635 * Also, multicasts addressed to the loopback interface
636 * are not sent -- the above call to ip_mloopback() will
637 * loop back a copy. ip_input() will drop the copy if
638 * this host does not belong to the destination group on
639 * the loopback interface.
641 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
650 * If the source address is not specified yet, use the address
651 * of the outoing interface.
653 if (ip->ip_src.s_addr == INADDR_ANY)
657 * Look for broadcast address and
658 * verify user is allowed to send
662 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
663 error = EADDRNOTAVAIL;
666 if ((flags & IP_ALLOWBROADCAST) == 0) {
670 /* don't allow broadcast messages to be fragmented */
675 m->m_flags |= M_BCAST;
677 m->m_flags &= ~M_BCAST;
681 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
682 if (IPSEC_ENABLED(ipv4)) {
683 if ((error = IPSEC_OUTPUT(ipv4, m, inp)) != 0) {
684 if (error == EINPROGRESS)
690 * Check if there was a route for this packet; return error if not.
692 if (no_route_but_check_spd) {
693 IPSTAT_INC(ips_noroute);
694 error = EHOSTUNREACH;
697 /* Update variables that are affected by ipsec4_output(). */
698 ip = mtod(m, struct ip *);
699 hlen = ip->ip_hl << 2;
702 /* Jump over all PFIL processing if hooks are not active. */
703 if (PFIL_HOOKED_OUT(V_inet_pfil_head)) {
704 switch (ip_output_pfil(&m, ifp, flags, inp, dst, &fibnum,
706 case 1: /* Finished */
709 case 0: /* Continue normally */
710 ip = mtod(m, struct ip *);
713 case -1: /* Need to try again */
714 /* Reset everything for a new round */
717 ro->ro_prepend = NULL;
720 ip = mtod(m, struct ip *);
725 /* IN_LOOPBACK must not appear on the wire - RFC1122. */
726 if (IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) ||
727 IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) {
728 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
729 IPSTAT_INC(ips_badaddr);
730 error = EADDRNOTAVAIL;
735 m->m_pkthdr.csum_flags |= CSUM_IP;
736 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
737 m = mb_unmapped_to_ext(m);
739 IPSTAT_INC(ips_odropped);
744 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
745 } else if ((ifp->if_capenable & IFCAP_NOMAP) == 0) {
746 m = mb_unmapped_to_ext(m);
748 IPSTAT_INC(ips_odropped);
753 #if defined(SCTP) || defined(SCTP_SUPPORT)
754 if (m->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
755 m = mb_unmapped_to_ext(m);
757 IPSTAT_INC(ips_odropped);
761 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
762 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
767 * If small enough for interface, or the interface will take
768 * care of the fragmentation for us, we can just send directly.
769 * Note that if_vxlan could have requested TSO even though the outer
770 * frame is UDP. It is correct to not fragment such datagrams and
771 * instead just pass them on to the driver.
774 (m->m_pkthdr.csum_flags & ifp->if_hwassist &
775 (CSUM_TSO | CSUM_INNER_TSO)) != 0) {
777 if (m->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
778 ip->ip_sum = in_cksum(m, hlen);
779 m->m_pkthdr.csum_flags &= ~CSUM_IP;
783 * Record statistics for this interface address.
784 * With CSUM_TSO the byte/packet count will be slightly
785 * incorrect because we count the IP+TCP headers only
786 * once instead of for every generated packet.
788 if (!(flags & IP_FORWARDING) && ia) {
789 if (m->m_pkthdr.csum_flags &
790 (CSUM_TSO | CSUM_INNER_TSO))
791 counter_u64_add(ia->ia_ifa.ifa_opackets,
792 m->m_pkthdr.len / m->m_pkthdr.tso_segsz);
794 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
796 counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len);
798 #ifdef MBUF_STRESS_TEST
799 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
800 m = m_fragment(m, M_NOWAIT, mbuf_frag_size);
803 * Reset layer specific mbuf flags
804 * to avoid confusing lower layers.
807 IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
808 error = ip_output_send(inp, ifp, m, gw, ro,
809 (flags & IP_NO_SND_TAG_RL) ? false : true);
813 /* Balk when DF bit is set or the interface didn't support TSO. */
814 if ((ip_off & IP_DF) ||
815 (m->m_pkthdr.csum_flags & (CSUM_TSO | CSUM_INNER_TSO))) {
817 IPSTAT_INC(ips_cantfrag);
822 * Too large for interface; fragment if possible. If successful,
823 * on return, m will point to a list of packets to be sent.
825 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist);
832 /* Record statistics for this interface address. */
834 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
835 counter_u64_add(ia->ia_ifa.ifa_obytes,
839 * Reset layer specific mbuf flags
840 * to avoid confusing upper layers.
844 IP_PROBE(send, NULL, NULL, mtod(m, struct ip *), ifp,
845 mtod(m, struct ip *), NULL);
846 error = ip_output_send(inp, ifp, m, gw, ro, true);
852 IPSTAT_INC(ips_fragmented);
862 * Create a chain of fragments which fit the given mtu. m_frag points to the
863 * mbuf to be fragmented; on return it points to the chain with the fragments.
864 * Return 0 if no error. If error, m_frag may contain a partially built
865 * chain of fragments that should be freed by the caller.
867 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
870 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
871 u_long if_hwassist_flags)
874 int hlen = ip->ip_hl << 2;
875 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */
877 struct mbuf *m0 = *m_frag; /* the original packet */
881 uint16_t ip_len, ip_off;
883 ip_len = ntohs(ip->ip_len);
884 ip_off = ntohs(ip->ip_off);
886 if (ip_off & IP_DF) { /* Fragmentation not allowed */
887 IPSTAT_INC(ips_cantfrag);
892 * Must be able to put at least 8 bytes per fragment.
898 * If the interface will not calculate checksums on
899 * fragmented packets, then do it here.
901 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
902 m0 = mb_unmapped_to_ext(m0);
905 IPSTAT_INC(ips_odropped);
908 in_delayed_cksum(m0);
909 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
911 #if defined(SCTP) || defined(SCTP_SUPPORT)
912 if (m0->m_pkthdr.csum_flags & CSUM_SCTP) {
913 m0 = mb_unmapped_to_ext(m0);
916 IPSTAT_INC(ips_odropped);
919 sctp_delayed_cksum(m0, hlen);
920 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
923 if (len > PAGE_SIZE) {
925 * Fragment large datagrams such that each segment
926 * contains a multiple of PAGE_SIZE amount of data,
927 * plus headers. This enables a receiver to perform
928 * page-flipping zero-copy optimizations.
930 * XXX When does this help given that sender and receiver
931 * could have different page sizes, and also mtu could
932 * be less than the receiver's page size ?
936 off = MIN(mtu, m0->m_pkthdr.len);
939 * firstlen (off - hlen) must be aligned on an
943 goto smart_frag_failure;
944 off = ((off - hlen) & ~7) + hlen;
945 newlen = (~PAGE_MASK) & mtu;
946 if ((newlen + sizeof (struct ip)) > mtu) {
947 /* we failed, go back the default */
958 firstlen = off - hlen;
959 mnext = &m0->m_nextpkt; /* pointer to next packet */
962 * Loop through length of segment after first fragment,
963 * make new header and copy data of each part and link onto chain.
964 * Here, m0 is the original packet, m is the fragment being created.
965 * The fragments are linked off the m_nextpkt of the original
966 * packet, which after processing serves as the first fragment.
968 for (nfrags = 1; off < ip_len; off += len, nfrags++) {
969 struct ip *mhip; /* ip header on the fragment */
971 int mhlen = sizeof (struct ip);
973 m = m_gethdr(M_NOWAIT, MT_DATA);
976 IPSTAT_INC(ips_odropped);
980 * Make sure the complete packet header gets copied
981 * from the originating mbuf to the newly created
982 * mbuf. This also ensures that existing firewall
983 * classification(s), VLAN tags and so on get copied
984 * to the resulting fragmented packet(s):
986 if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) {
989 IPSTAT_INC(ips_odropped);
993 * In the first mbuf, leave room for the link header, then
994 * copy the original IP header including options. The payload
995 * goes into an additional mbuf chain returned by m_copym().
997 m->m_data += max_linkhdr;
998 mhip = mtod(m, struct ip *);
1000 if (hlen > sizeof (struct ip)) {
1001 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1002 mhip->ip_v = IPVERSION;
1003 mhip->ip_hl = mhlen >> 2;
1006 /* XXX do we need to add ip_off below ? */
1007 mhip->ip_off = ((off - hlen) >> 3) + ip_off;
1008 if (off + len >= ip_len)
1011 mhip->ip_off |= IP_MF;
1012 mhip->ip_len = htons((u_short)(len + mhlen));
1013 m->m_next = m_copym(m0, off, len, M_NOWAIT);
1014 if (m->m_next == NULL) { /* copy failed */
1016 error = ENOBUFS; /* ??? */
1017 IPSTAT_INC(ips_odropped);
1020 m->m_pkthdr.len = mhlen + len;
1022 mac_netinet_fragment(m0, m);
1024 mhip->ip_off = htons(mhip->ip_off);
1026 if (m->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
1027 mhip->ip_sum = in_cksum(m, mhlen);
1028 m->m_pkthdr.csum_flags &= ~CSUM_IP;
1031 mnext = &m->m_nextpkt;
1033 IPSTAT_ADD(ips_ofragments, nfrags);
1036 * Update first fragment by trimming what's been copied out
1037 * and updating header.
1039 m_adj(m0, hlen + firstlen - ip_len);
1040 m0->m_pkthdr.len = hlen + firstlen;
1041 ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1042 ip->ip_off = htons(ip_off | IP_MF);
1044 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
1045 ip->ip_sum = in_cksum(m0, hlen);
1046 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
1055 in_delayed_cksum(struct mbuf *m)
1059 uint16_t cklen, csum, offset;
1061 ip = mtod(m, struct ip *);
1062 offset = ip->ip_hl << 2 ;
1064 if (m->m_pkthdr.csum_flags & CSUM_UDP) {
1065 /* if udp header is not in the first mbuf copy udplen */
1066 if (offset + sizeof(struct udphdr) > m->m_len) {
1067 m_copydata(m, offset + offsetof(struct udphdr,
1068 uh_ulen), sizeof(cklen), (caddr_t)&cklen);
1069 cklen = ntohs(cklen);
1071 uh = (struct udphdr *)mtodo(m, offset);
1072 cklen = ntohs(uh->uh_ulen);
1074 csum = in_cksum_skip(m, cklen + offset, offset);
1078 cklen = ntohs(ip->ip_len);
1079 csum = in_cksum_skip(m, cklen, offset);
1081 offset += m->m_pkthdr.csum_data; /* checksum offset */
1083 if (offset + sizeof(csum) > m->m_len)
1084 m_copyback(m, offset, sizeof(csum), (caddr_t)&csum);
1086 *(u_short *)mtodo(m, offset) = csum;
1090 * IP socket option processing.
1093 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1095 struct inpcb *inp = sotoinpcb(so);
1098 uint32_t rss_bucket;
1103 if (sopt->sopt_level != IPPROTO_IP) {
1106 if (sopt->sopt_level == SOL_SOCKET &&
1107 sopt->sopt_dir == SOPT_SET) {
1108 switch (sopt->sopt_name) {
1111 if ((so->so_options & SO_REUSEADDR) != 0)
1112 inp->inp_flags2 |= INP_REUSEADDR;
1114 inp->inp_flags2 &= ~INP_REUSEADDR;
1120 if ((so->so_options & SO_REUSEPORT) != 0)
1121 inp->inp_flags2 |= INP_REUSEPORT;
1123 inp->inp_flags2 &= ~INP_REUSEPORT;
1127 case SO_REUSEPORT_LB:
1129 if ((so->so_options & SO_REUSEPORT_LB) != 0)
1130 inp->inp_flags2 |= INP_REUSEPORT_LB;
1132 inp->inp_flags2 &= ~INP_REUSEPORT_LB;
1138 inp->inp_inc.inc_fibnum = so->so_fibnum;
1142 case SO_MAX_PACING_RATE:
1145 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
1159 switch (sopt->sopt_dir) {
1161 switch (sopt->sopt_name) {
1168 if (sopt->sopt_valsize > MLEN) {
1172 m = m_get(sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
1177 m->m_len = sopt->sopt_valsize;
1178 error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1185 error = ip_pcbopts(inp, sopt->sopt_name, m);
1191 if (sopt->sopt_td != NULL) {
1192 error = priv_check(sopt->sopt_td,
1193 PRIV_NETINET_BINDANY);
1200 case IP_RSS_LISTEN_BUCKET:
1206 case IP_RECVRETOPTS:
1207 case IP_ORIGDSTADDR:
1208 case IP_RECVDSTADDR:
1216 case IP_RECVRSSBUCKETID:
1218 error = sooptcopyin(sopt, &optval, sizeof optval,
1223 switch (sopt->sopt_name) {
1225 inp->inp_ip_tos = optval;
1229 inp->inp_ip_ttl = optval;
1233 if (optval >= 0 && optval <= MAXTTL)
1234 inp->inp_ip_minttl = optval;
1239 #define OPTSET(bit) do { \
1242 inp->inp_flags |= bit; \
1244 inp->inp_flags &= ~bit; \
1248 #define OPTSET2(bit, val) do { \
1251 inp->inp_flags2 |= bit; \
1253 inp->inp_flags2 &= ~bit; \
1258 OPTSET(INP_RECVOPTS);
1261 case IP_RECVRETOPTS:
1262 OPTSET(INP_RECVRETOPTS);
1265 case IP_RECVDSTADDR:
1266 OPTSET(INP_RECVDSTADDR);
1269 case IP_ORIGDSTADDR:
1270 OPTSET2(INP_ORIGDSTADDR, optval);
1274 OPTSET(INP_RECVTTL);
1282 OPTSET(INP_ONESBCAST);
1285 OPTSET(INP_DONTFRAG);
1288 OPTSET(INP_BINDANY);
1291 OPTSET(INP_RECVTOS);
1294 OPTSET2(INP_BINDMULTI, optval);
1297 OPTSET2(INP_RECVFLOWID, optval);
1300 case IP_RSS_LISTEN_BUCKET:
1301 if ((optval >= 0) &&
1302 (optval < rss_getnumbuckets())) {
1303 inp->inp_rss_listen_bucket = optval;
1304 OPTSET2(INP_RSS_BUCKET_SET, 1);
1309 case IP_RECVRSSBUCKETID:
1310 OPTSET2(INP_RECVRSSBUCKETID, optval);
1319 * Multicast socket options are processed by the in_mcast
1322 case IP_MULTICAST_IF:
1323 case IP_MULTICAST_VIF:
1324 case IP_MULTICAST_TTL:
1325 case IP_MULTICAST_LOOP:
1326 case IP_ADD_MEMBERSHIP:
1327 case IP_DROP_MEMBERSHIP:
1328 case IP_ADD_SOURCE_MEMBERSHIP:
1329 case IP_DROP_SOURCE_MEMBERSHIP:
1330 case IP_BLOCK_SOURCE:
1331 case IP_UNBLOCK_SOURCE:
1333 case MCAST_JOIN_GROUP:
1334 case MCAST_LEAVE_GROUP:
1335 case MCAST_JOIN_SOURCE_GROUP:
1336 case MCAST_LEAVE_SOURCE_GROUP:
1337 case MCAST_BLOCK_SOURCE:
1338 case MCAST_UNBLOCK_SOURCE:
1339 error = inp_setmoptions(inp, sopt);
1343 error = sooptcopyin(sopt, &optval, sizeof optval,
1350 case IP_PORTRANGE_DEFAULT:
1351 inp->inp_flags &= ~(INP_LOWPORT);
1352 inp->inp_flags &= ~(INP_HIGHPORT);
1355 case IP_PORTRANGE_HIGH:
1356 inp->inp_flags &= ~(INP_LOWPORT);
1357 inp->inp_flags |= INP_HIGHPORT;
1360 case IP_PORTRANGE_LOW:
1361 inp->inp_flags &= ~(INP_HIGHPORT);
1362 inp->inp_flags |= INP_LOWPORT;
1372 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1373 case IP_IPSEC_POLICY:
1374 if (IPSEC_ENABLED(ipv4)) {
1375 error = IPSEC_PCBCTL(ipv4, inp, sopt);
1382 error = ENOPROTOOPT;
1388 switch (sopt->sopt_name) {
1392 if (inp->inp_options) {
1393 struct mbuf *options;
1395 options = m_copym(inp->inp_options, 0,
1396 M_COPYALL, M_NOWAIT);
1398 if (options != NULL) {
1399 error = sooptcopyout(sopt,
1400 mtod(options, char *),
1407 sopt->sopt_valsize = 0;
1415 case IP_RECVRETOPTS:
1416 case IP_ORIGDSTADDR:
1417 case IP_RECVDSTADDR:
1430 case IP_RSSBUCKETID:
1431 case IP_RECVRSSBUCKETID:
1433 switch (sopt->sopt_name) {
1435 optval = inp->inp_ip_tos;
1439 optval = inp->inp_ip_ttl;
1443 optval = inp->inp_ip_minttl;
1446 #define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0)
1447 #define OPTBIT2(bit) (inp->inp_flags2 & bit ? 1 : 0)
1450 optval = OPTBIT(INP_RECVOPTS);
1453 case IP_RECVRETOPTS:
1454 optval = OPTBIT(INP_RECVRETOPTS);
1457 case IP_RECVDSTADDR:
1458 optval = OPTBIT(INP_RECVDSTADDR);
1461 case IP_ORIGDSTADDR:
1462 optval = OPTBIT2(INP_ORIGDSTADDR);
1466 optval = OPTBIT(INP_RECVTTL);
1470 optval = OPTBIT(INP_RECVIF);
1474 if (inp->inp_flags & INP_HIGHPORT)
1475 optval = IP_PORTRANGE_HIGH;
1476 else if (inp->inp_flags & INP_LOWPORT)
1477 optval = IP_PORTRANGE_LOW;
1483 optval = OPTBIT(INP_ONESBCAST);
1486 optval = OPTBIT(INP_DONTFRAG);
1489 optval = OPTBIT(INP_BINDANY);
1492 optval = OPTBIT(INP_RECVTOS);
1495 optval = inp->inp_flowid;
1498 optval = inp->inp_flowtype;
1501 optval = OPTBIT2(INP_RECVFLOWID);
1504 case IP_RSSBUCKETID:
1505 retval = rss_hash2bucket(inp->inp_flowid,
1509 optval = rss_bucket;
1513 case IP_RECVRSSBUCKETID:
1514 optval = OPTBIT2(INP_RECVRSSBUCKETID);
1518 optval = OPTBIT2(INP_BINDMULTI);
1521 error = sooptcopyout(sopt, &optval, sizeof optval);
1525 * Multicast socket options are processed by the in_mcast
1528 case IP_MULTICAST_IF:
1529 case IP_MULTICAST_VIF:
1530 case IP_MULTICAST_TTL:
1531 case IP_MULTICAST_LOOP:
1533 error = inp_getmoptions(inp, sopt);
1536 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1537 case IP_IPSEC_POLICY:
1538 if (IPSEC_ENABLED(ipv4)) {
1539 error = IPSEC_PCBCTL(ipv4, inp, sopt);
1546 error = ENOPROTOOPT;
1555 * Routine called from ip_output() to loop back a copy of an IP multicast
1556 * packet to the input queue of a specified interface. Note that this
1557 * calls the output routine of the loopback "driver", but with an interface
1558 * pointer that might NOT be a loopback interface -- evil, but easier than
1559 * replicating that code here.
1562 ip_mloopback(struct ifnet *ifp, const struct mbuf *m, int hlen)
1568 * Make a deep copy of the packet because we're going to
1569 * modify the pack in order to generate checksums.
1571 copym = m_dup(m, M_NOWAIT);
1572 if (copym != NULL && (!M_WRITABLE(copym) || copym->m_len < hlen))
1573 copym = m_pullup(copym, hlen);
1574 if (copym != NULL) {
1575 /* If needed, compute the checksum and mark it as valid. */
1576 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1577 in_delayed_cksum(copym);
1578 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1579 copym->m_pkthdr.csum_flags |=
1580 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1581 copym->m_pkthdr.csum_data = 0xffff;
1584 * We don't bother to fragment if the IP length is greater
1585 * than the interface's MTU. Can this possibly matter?
1587 ip = mtod(copym, struct ip *);
1589 ip->ip_sum = in_cksum(copym, hlen);
1590 if_simloop(ifp, copym, AF_INET, 0);