2 * Copyright (c) 2020 Mellanox Technologies. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include "opt_inet6.h"
30 #include <sys/cdefs.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/devctl.h>
34 #include <sys/eventhandler.h>
35 #include <sys/kernel.h>
37 #include <sys/module.h>
38 #include <sys/socket.h>
39 #include <sys/sysctl.h>
45 #include <net/ethernet.h>
46 #include <net/infiniband.h>
48 #include <net/if_var.h>
49 #include <net/if_private.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 #include <net/if_lagg.h>
53 #include <net/if_llatbl.h>
54 #include <net/if_types.h>
55 #include <net/netisr.h>
56 #include <net/route.h>
57 #include <netinet/if_ether.h>
58 #include <netinet/in.h>
59 #include <netinet/ip6.h>
60 #include <netinet6/in6_var.h>
61 #include <netinet6/nd6.h>
63 #include <security/mac/mac_framework.h>
65 /* if_lagg(4) support */
66 struct mbuf *(*lagg_input_infiniband_p)(struct ifnet *, struct mbuf *);
70 infiniband_ipv4_multicast_map(uint32_t addr,
71 const uint8_t *broadcast, uint8_t *buf)
76 scope = broadcast[5] & 0xF;
83 buf[5] = 0x10 | scope;
86 buf[8] = broadcast[8];
87 buf[9] = broadcast[9];
94 buf[16] = (addr >> 24) & 0xff;
95 buf[17] = (addr >> 16) & 0xff;
96 buf[18] = (addr >> 8) & 0xff;
97 buf[19] = addr & 0xff;
103 infiniband_ipv6_multicast_map(const struct in6_addr *addr,
104 const uint8_t *broadcast, uint8_t *buf)
108 scope = broadcast[5] & 0xF;
115 buf[5] = 0x10 | scope;
118 buf[8] = broadcast[8];
119 buf[9] = broadcast[9];
120 memcpy(&buf[10], &addr->s6_addr[6], 10);
125 * This is for clients that have an infiniband_header in the mbuf.
128 infiniband_bpf_mtap(struct ifnet *ifp, struct mbuf *mb)
130 struct infiniband_header *ibh;
131 struct ether_header eh;
133 if (!bpf_peers_present(ifp->if_bpf))
137 if (mb->m_len < sizeof(*ibh))
140 ibh = mtod(mb, struct infiniband_header *);
141 eh.ether_type = ibh->ib_protocol;
142 memset(eh.ether_shost, 0, ETHER_ADDR_LEN);
143 memcpy(eh.ether_dhost, ibh->ib_hwaddr + 4, ETHER_ADDR_LEN);
144 mb->m_data += sizeof(*ibh);
145 mb->m_len -= sizeof(*ibh);
146 mb->m_pkthdr.len -= sizeof(*ibh);
147 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
148 mb->m_data -= sizeof(*ibh);
149 mb->m_len += sizeof(*ibh);
150 mb->m_pkthdr.len += sizeof(*ibh);
154 update_mbuf_csumflags(struct mbuf *src, struct mbuf *dst)
158 if (src->m_pkthdr.csum_flags & CSUM_IP)
159 csum_flags |= (CSUM_IP_CHECKED|CSUM_IP_VALID);
160 if (src->m_pkthdr.csum_flags & CSUM_DELAY_DATA)
161 csum_flags |= (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
162 if (src->m_pkthdr.csum_flags & CSUM_SCTP)
163 csum_flags |= CSUM_SCTP_VALID;
164 dst->m_pkthdr.csum_flags |= csum_flags;
165 if (csum_flags & CSUM_DATA_VALID)
166 dst->m_pkthdr.csum_data = 0xffff;
170 * Handle link-layer encapsulation requests.
173 infiniband_requestencap(struct ifnet *ifp, struct if_encap_req *req)
175 struct infiniband_header *ih;
178 const uint8_t *lladdr;
180 if (req->rtype != IFENCAP_LL)
183 if (req->bufsize < INFINIBAND_HDR_LEN)
186 ih = (struct infiniband_header *)req->buf;
187 lladdr = req->lladdr;
190 switch (req->family) {
192 etype = htons(ETHERTYPE_IP);
195 etype = htons(ETHERTYPE_IPV6);
198 ah = (struct arphdr *)req->hdata;
199 ah->ar_hrd = htons(ARPHRD_INFINIBAND);
201 switch (ntohs(ah->ar_op)) {
202 case ARPOP_REVREQUEST:
204 etype = htons(ETHERTYPE_REVARP);
209 etype = htons(ETHERTYPE_ARP);
213 if (req->flags & IFENCAP_FLAG_BROADCAST)
214 lladdr = ifp->if_broadcastaddr;
217 return (EAFNOSUPPORT);
220 ih->ib_protocol = etype;
222 memcpy(ih->ib_hwaddr, lladdr, INFINIBAND_ADDR_LEN);
223 req->bufsize = sizeof(struct infiniband_header);
229 infiniband_resolve_addr(struct ifnet *ifp, struct mbuf *m,
230 const struct sockaddr *dst, struct route *ro, uint8_t *phdr,
231 uint32_t *pflags, struct llentry **plle)
233 #if defined(INET) || defined(INET6)
234 struct infiniband_header *ih = (struct infiniband_header *)phdr;
236 uint32_t lleflags = 0;
242 switch (dst->sa_family) {
245 if ((m->m_flags & (M_BCAST | M_MCAST)) == 0) {
246 error = arpresolve(ifp, 0, m, dst, phdr, &lleflags, plle);
248 if (m->m_flags & M_BCAST) {
249 memcpy(ih->ib_hwaddr, ifp->if_broadcastaddr,
250 INFINIBAND_ADDR_LEN);
252 infiniband_ipv4_multicast_map(
253 ((const struct sockaddr_in *)dst)->sin_addr.s_addr,
254 ifp->if_broadcastaddr, ih->ib_hwaddr);
256 ih->ib_protocol = htons(ETHERTYPE_IP);
263 if ((m->m_flags & M_MCAST) == 0) {
264 int af = RO_GET_FAMILY(ro, dst);
265 error = nd6_resolve(ifp, LLE_SF(af, 0), m, dst, phdr,
268 infiniband_ipv6_multicast_map(
269 &((const struct sockaddr_in6 *)dst)->sin6_addr,
270 ifp->if_broadcastaddr, ih->ib_hwaddr);
271 ih->ib_protocol = htons(ETHERTYPE_IPV6);
277 if_printf(ifp, "can't handle af%d\n", dst->sa_family);
280 return (EAFNOSUPPORT);
283 if (error == EHOSTDOWN) {
284 if (ro != NULL && (ro->ro_flags & RT_HAS_GW) != 0)
285 error = EHOSTUNREACH;
291 *pflags = RT_MAY_LOOP;
292 if (lleflags & LLE_IFADDR)
299 * Infiniband output routine.
302 infiniband_output(struct ifnet *ifp, struct mbuf *m,
303 const struct sockaddr *dst, struct route *ro)
305 uint8_t linkhdr[INFINIBAND_HDR_LEN];
307 struct llentry *lle = NULL;
308 struct infiniband_header *ih;
310 int hlen; /* link layer header length */
320 /* XXX BPF uses ro_prepend */
321 if (ro->ro_prepend != NULL) {
322 phdr = ro->ro_prepend;
324 } else if (!(m->m_flags & (M_BCAST | M_MCAST))) {
325 if ((ro->ro_flags & RT_LLE_CACHE) != 0) {
328 (lle->la_flags & LLE_VALID) == 0) {
330 lle = NULL; /* redundant */
334 /* if we lookup, keep cache */
338 * Notify LLE code that
342 llentry_provide_feedback(lle);
345 phdr = lle->r_linkdata;
346 hlen = lle->r_hdrlen;
347 pflags = lle->r_flags;
353 error = mac_ifnet_check_transmit(ifp, m);
359 if (ifp->if_flags & IFF_MONITOR) {
363 if (!((ifp->if_flags & IFF_UP) &&
364 (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
370 /* No prepend data supplied. Try to calculate ourselves. */
372 hlen = INFINIBAND_HDR_LEN;
373 error = infiniband_resolve_addr(ifp, m, dst, ro, phdr, &pflags,
374 addref ? &lle : NULL);
375 if (addref && lle != NULL)
378 return (error == EWOULDBLOCK ? 0 : error);
381 if ((pflags & RT_L2_ME) != 0) {
382 update_mbuf_csumflags(m, m);
383 return (if_simloop(ifp, m, RO_GET_FAMILY(ro, dst), 0));
387 * Add local infiniband header. If no space in first mbuf,
390 M_PREPEND(m, INFINIBAND_HDR_LEN, M_NOWAIT);
395 if ((pflags & RT_HAS_HEADER) == 0) {
396 ih = mtod(m, struct infiniband_header *);
397 memcpy(ih, phdr, hlen);
401 * Queue message on interface, update output statistics if
402 * successful, and start output if interface not yet active.
404 return (ifp->if_transmit(ifp, m));
412 * Process a received Infiniband packet.
415 infiniband_input(struct ifnet *ifp, struct mbuf *m)
417 struct infiniband_header *ibh;
418 struct epoch_tracker et;
422 needs_epoch = (ifp->if_flags & IFF_NEEDSEPOCH);
425 * This temporary code is here to prevent epoch unaware and unmarked
426 * drivers to panic the system. Once all drivers are taken care of,
427 * the whole INVARIANTS block should go away.
429 if (!needs_epoch && !in_epoch(net_epoch_preempt)) {
430 static bool printedonce;
435 if_printf(ifp, "called %s w/o net epoch! "
436 "PLEASE file a bug report.", __func__);
444 CURVNET_SET_QUIET(ifp->if_vnet);
445 if (__predict_false(needs_epoch))
448 if ((ifp->if_flags & IFF_UP) == 0) {
449 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
454 ibh = mtod(m, struct infiniband_header *);
457 * Reset layer specific mbuf flags to avoid confusing upper
460 m->m_flags &= ~M_VLANTAG;
463 if (INFINIBAND_IS_MULTICAST(ibh->ib_hwaddr)) {
464 if (memcmp(ibh->ib_hwaddr, ifp->if_broadcastaddr,
465 ifp->if_addrlen) == 0)
466 m->m_flags |= M_BCAST;
468 m->m_flags |= M_MCAST;
469 if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1);
472 /* Let BPF have it before we strip the header. */
473 infiniband_bpf_mtap(ifp, m);
475 /* Allow monitor mode to claim this frame, after stats are updated. */
476 if (ifp->if_flags & IFF_MONITOR) {
481 /* Direct packet to correct FIB based on interface config. */
482 M_SETFIB(m, ifp->if_fib);
484 /* Handle input from a lagg<N> port */
485 if (ifp->if_type == IFT_INFINIBANDLAG) {
486 KASSERT(lagg_input_infiniband_p != NULL,
487 ("%s: if_lagg not loaded!", __func__));
488 m = (*lagg_input_infiniband_p)(ifp, m);
489 if (__predict_false(m == NULL))
491 ifp = m->m_pkthdr.rcvif;
495 * Dispatch frame to upper layer.
497 switch (ibh->ib_protocol) {
499 case htons(ETHERTYPE_IP):
503 case htons(ETHERTYPE_ARP):
504 if (ifp->if_flags & IFF_NOARP) {
505 /* Discard packet if ARP is disabled on interface */
513 case htons(ETHERTYPE_IPV6):
518 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
523 /* Strip off the Infiniband header. */
524 m_adj(m, INFINIBAND_HDR_LEN);
528 * Tag the mbuf with an appropriate MAC label before any other
529 * consumers can get to it.
531 mac_ifnet_create_mbuf(ifp, m);
533 /* Allow monitor mode to claim this frame, after stats are updated. */
534 netisr_dispatch(isr, m);
536 if (__predict_false(needs_epoch))
542 infiniband_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
545 struct sockaddr_dl *sdl;
547 struct sockaddr_in *sin;
550 struct sockaddr_in6 *sin6;
554 switch (sa->sa_family) {
557 * No mapping needed. Just check that it's a valid MC address.
559 sdl = (struct sockaddr_dl *)sa;
560 e_addr = LLADDR(sdl);
561 if (!INFINIBAND_IS_MULTICAST(e_addr))
562 return (EADDRNOTAVAIL);
568 sin = (struct sockaddr_in *)sa;
569 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
570 return (EADDRNOTAVAIL);
571 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
572 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
573 e_addr = LLADDR(sdl);
574 infiniband_ipv4_multicast_map(
575 sin->sin_addr.s_addr, ifp->if_broadcastaddr, e_addr);
576 *llsa = (struct sockaddr *)sdl;
581 sin6 = (struct sockaddr_in6 *)sa;
583 * An IP6 address of 0 means listen to all of the
584 * multicast address used for IP6. This has no meaning
587 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
588 return (EADDRNOTAVAIL);
589 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
590 return (EADDRNOTAVAIL);
591 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
592 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
593 e_addr = LLADDR(sdl);
594 infiniband_ipv6_multicast_map(
595 &sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
596 *llsa = (struct sockaddr *)sdl;
600 return (EAFNOSUPPORT);
605 infiniband_ifattach(struct ifnet *ifp, const uint8_t *lla, const uint8_t *llb)
607 struct sockaddr_dl *sdl;
611 ifp->if_addrlen = INFINIBAND_ADDR_LEN;
612 ifp->if_hdrlen = INFINIBAND_HDR_LEN;
613 ifp->if_mtu = INFINIBAND_MTU;
615 ifp->if_output = infiniband_output;
616 ifp->if_input = infiniband_input;
617 ifp->if_resolvemulti = infiniband_resolvemulti;
618 ifp->if_requestencap = infiniband_requestencap;
620 if (ifp->if_baudrate == 0)
621 ifp->if_baudrate = IF_Gbps(10); /* default value */
623 ifp->if_broadcastaddr = llb;
626 KASSERT(ifa != NULL, ("%s: no lladdr!\n", __func__));
627 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
628 sdl->sdl_type = IFT_INFINIBAND;
629 sdl->sdl_alen = ifp->if_addrlen;
632 memcpy(LLADDR(sdl), lla, ifp->if_addrlen);
634 if (ifp->if_hw_addr != NULL)
635 memcpy(ifp->if_hw_addr, lla, ifp->if_addrlen);
640 /* Attach ethernet compatible network device */
641 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN);
643 /* Announce Infiniband MAC address if non-zero. */
644 for (i = 0; i < ifp->if_addrlen; i++)
647 if (i != ifp->if_addrlen)
648 if_printf(ifp, "Infiniband address: %20D\n", lla, ":");
650 /* Add necessary bits are setup; announce it now. */
651 EVENTHANDLER_INVOKE(infiniband_ifattach_event, ifp);
653 if (IS_DEFAULT_VNET(curvnet))
654 devctl_notify("INFINIBAND", ifp->if_xname, "IFATTACH", NULL);
658 * Perform common duties while detaching an Infiniband interface
661 infiniband_ifdetach(struct ifnet *ifp)
668 infiniband_modevent(module_t mod, int type, void *data)
679 static moduledata_t infiniband_mod = {
680 .name = "if_infiniband",
681 .evhand = &infiniband_modevent,
684 DECLARE_MODULE(if_infiniband, infiniband_mod, SI_SUB_INIT_IF, SI_ORDER_ANY);
685 MODULE_VERSION(if_infiniband, 1);