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/param.h>
31 #include <sys/systm.h>
32 #include <sys/devctl.h>
33 #include <sys/eventhandler.h>
34 #include <sys/kernel.h>
36 #include <sys/module.h>
37 #include <sys/socket.h>
38 #include <sys/sysctl.h>
44 #include <net/ethernet.h>
45 #include <net/infiniband.h>
47 #include <net/if_var.h>
48 #include <net/if_private.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_lagg.h>
52 #include <net/if_llatbl.h>
53 #include <net/if_types.h>
54 #include <net/netisr.h>
55 #include <net/route.h>
56 #include <netinet/if_ether.h>
57 #include <netinet/in.h>
58 #include <netinet/ip6.h>
59 #include <netinet6/in6_var.h>
60 #include <netinet6/nd6.h>
62 #include <security/mac/mac_framework.h>
64 /* if_lagg(4) support */
65 struct mbuf *(*lagg_input_infiniband_p)(struct ifnet *, struct mbuf *);
69 infiniband_ipv4_multicast_map(uint32_t addr,
70 const uint8_t *broadcast, uint8_t *buf)
75 scope = broadcast[5] & 0xF;
82 buf[5] = 0x10 | scope;
85 buf[8] = broadcast[8];
86 buf[9] = broadcast[9];
93 buf[16] = (addr >> 24) & 0xff;
94 buf[17] = (addr >> 16) & 0xff;
95 buf[18] = (addr >> 8) & 0xff;
96 buf[19] = addr & 0xff;
102 infiniband_ipv6_multicast_map(const struct in6_addr *addr,
103 const uint8_t *broadcast, uint8_t *buf)
107 scope = broadcast[5] & 0xF;
114 buf[5] = 0x10 | scope;
117 buf[8] = broadcast[8];
118 buf[9] = broadcast[9];
119 memcpy(&buf[10], &addr->s6_addr[6], 10);
124 * This is for clients that have an infiniband_header in the mbuf.
127 infiniband_bpf_mtap(struct ifnet *ifp, struct mbuf *mb)
129 struct infiniband_header *ibh;
130 struct ether_header eh;
132 if (!bpf_peers_present(ifp->if_bpf))
136 if (mb->m_len < sizeof(*ibh))
139 ibh = mtod(mb, struct infiniband_header *);
140 eh.ether_type = ibh->ib_protocol;
141 memset(eh.ether_shost, 0, ETHER_ADDR_LEN);
142 memcpy(eh.ether_dhost, ibh->ib_hwaddr + 4, ETHER_ADDR_LEN);
143 mb->m_data += sizeof(*ibh);
144 mb->m_len -= sizeof(*ibh);
145 mb->m_pkthdr.len -= sizeof(*ibh);
146 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
147 mb->m_data -= sizeof(*ibh);
148 mb->m_len += sizeof(*ibh);
149 mb->m_pkthdr.len += sizeof(*ibh);
153 update_mbuf_csumflags(struct mbuf *src, struct mbuf *dst)
157 if (src->m_pkthdr.csum_flags & CSUM_IP)
158 csum_flags |= (CSUM_IP_CHECKED|CSUM_IP_VALID);
159 if (src->m_pkthdr.csum_flags & CSUM_DELAY_DATA)
160 csum_flags |= (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
161 if (src->m_pkthdr.csum_flags & CSUM_SCTP)
162 csum_flags |= CSUM_SCTP_VALID;
163 dst->m_pkthdr.csum_flags |= csum_flags;
164 if (csum_flags & CSUM_DATA_VALID)
165 dst->m_pkthdr.csum_data = 0xffff;
169 * Handle link-layer encapsulation requests.
172 infiniband_requestencap(struct ifnet *ifp, struct if_encap_req *req)
174 struct infiniband_header *ih;
177 const uint8_t *lladdr;
179 if (req->rtype != IFENCAP_LL)
182 if (req->bufsize < INFINIBAND_HDR_LEN)
185 ih = (struct infiniband_header *)req->buf;
186 lladdr = req->lladdr;
189 switch (req->family) {
191 etype = htons(ETHERTYPE_IP);
194 etype = htons(ETHERTYPE_IPV6);
197 ah = (struct arphdr *)req->hdata;
198 ah->ar_hrd = htons(ARPHRD_INFINIBAND);
200 switch (ntohs(ah->ar_op)) {
201 case ARPOP_REVREQUEST:
203 etype = htons(ETHERTYPE_REVARP);
208 etype = htons(ETHERTYPE_ARP);
212 if (req->flags & IFENCAP_FLAG_BROADCAST)
213 lladdr = ifp->if_broadcastaddr;
216 return (EAFNOSUPPORT);
219 ih->ib_protocol = etype;
221 memcpy(ih->ib_hwaddr, lladdr, INFINIBAND_ADDR_LEN);
222 req->bufsize = sizeof(struct infiniband_header);
228 infiniband_resolve_addr(struct ifnet *ifp, struct mbuf *m,
229 const struct sockaddr *dst, struct route *ro, uint8_t *phdr,
230 uint32_t *pflags, struct llentry **plle)
232 #if defined(INET) || defined(INET6)
233 struct infiniband_header *ih = (struct infiniband_header *)phdr;
235 uint32_t lleflags = 0;
241 switch (dst->sa_family) {
244 if ((m->m_flags & (M_BCAST | M_MCAST)) == 0) {
245 error = arpresolve(ifp, 0, m, dst, phdr, &lleflags, plle);
247 if (m->m_flags & M_BCAST) {
248 memcpy(ih->ib_hwaddr, ifp->if_broadcastaddr,
249 INFINIBAND_ADDR_LEN);
251 infiniband_ipv4_multicast_map(
252 ((const struct sockaddr_in *)dst)->sin_addr.s_addr,
253 ifp->if_broadcastaddr, ih->ib_hwaddr);
255 ih->ib_protocol = htons(ETHERTYPE_IP);
262 if ((m->m_flags & M_MCAST) == 0) {
263 int af = RO_GET_FAMILY(ro, dst);
264 error = nd6_resolve(ifp, LLE_SF(af, 0), m, dst, phdr,
267 infiniband_ipv6_multicast_map(
268 &((const struct sockaddr_in6 *)dst)->sin6_addr,
269 ifp->if_broadcastaddr, ih->ib_hwaddr);
270 ih->ib_protocol = htons(ETHERTYPE_IPV6);
276 if_printf(ifp, "can't handle af%d\n", dst->sa_family);
279 return (EAFNOSUPPORT);
282 if (error == EHOSTDOWN) {
283 if (ro != NULL && (ro->ro_flags & RT_HAS_GW) != 0)
284 error = EHOSTUNREACH;
290 *pflags = RT_MAY_LOOP;
291 if (lleflags & LLE_IFADDR)
298 * Infiniband output routine.
301 infiniband_output(struct ifnet *ifp, struct mbuf *m,
302 const struct sockaddr *dst, struct route *ro)
304 uint8_t linkhdr[INFINIBAND_HDR_LEN];
306 struct llentry *lle = NULL;
307 struct infiniband_header *ih;
309 int hlen; /* link layer header length */
319 /* XXX BPF uses ro_prepend */
320 if (ro->ro_prepend != NULL) {
321 phdr = ro->ro_prepend;
323 } else if (!(m->m_flags & (M_BCAST | M_MCAST))) {
324 if ((ro->ro_flags & RT_LLE_CACHE) != 0) {
327 (lle->la_flags & LLE_VALID) == 0) {
329 lle = NULL; /* redundant */
333 /* if we lookup, keep cache */
337 * Notify LLE code that
341 llentry_provide_feedback(lle);
344 phdr = lle->r_linkdata;
345 hlen = lle->r_hdrlen;
346 pflags = lle->r_flags;
352 error = mac_ifnet_check_transmit(ifp, m);
358 if (ifp->if_flags & IFF_MONITOR) {
362 if (!((ifp->if_flags & IFF_UP) &&
363 (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
369 /* No prepend data supplied. Try to calculate ourselves. */
371 hlen = INFINIBAND_HDR_LEN;
372 error = infiniband_resolve_addr(ifp, m, dst, ro, phdr, &pflags,
373 addref ? &lle : NULL);
374 if (addref && lle != NULL)
377 return (error == EWOULDBLOCK ? 0 : error);
380 if ((pflags & RT_L2_ME) != 0) {
381 update_mbuf_csumflags(m, m);
382 return (if_simloop(ifp, m, RO_GET_FAMILY(ro, dst), 0));
386 * Add local infiniband header. If no space in first mbuf,
389 M_PREPEND(m, INFINIBAND_HDR_LEN, M_NOWAIT);
394 if ((pflags & RT_HAS_HEADER) == 0) {
395 ih = mtod(m, struct infiniband_header *);
396 memcpy(ih, phdr, hlen);
400 * Queue message on interface, update output statistics if
401 * successful, and start output if interface not yet active.
403 return (ifp->if_transmit(ifp, m));
411 * Process a received Infiniband packet.
414 infiniband_input(struct ifnet *ifp, struct mbuf *m)
416 struct infiniband_header *ibh;
417 struct epoch_tracker et;
421 needs_epoch = (ifp->if_flags & IFF_NEEDSEPOCH);
424 * This temporary code is here to prevent epoch unaware and unmarked
425 * drivers to panic the system. Once all drivers are taken care of,
426 * the whole INVARIANTS block should go away.
428 if (!needs_epoch && !in_epoch(net_epoch_preempt)) {
429 static bool printedonce;
434 if_printf(ifp, "called %s w/o net epoch! "
435 "PLEASE file a bug report.", __func__);
443 CURVNET_SET_QUIET(ifp->if_vnet);
444 if (__predict_false(needs_epoch))
447 if ((ifp->if_flags & IFF_UP) == 0) {
448 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
453 ibh = mtod(m, struct infiniband_header *);
456 * Reset layer specific mbuf flags to avoid confusing upper
459 m->m_flags &= ~M_VLANTAG;
462 if (INFINIBAND_IS_MULTICAST(ibh->ib_hwaddr)) {
463 if (memcmp(ibh->ib_hwaddr, ifp->if_broadcastaddr,
464 ifp->if_addrlen) == 0)
465 m->m_flags |= M_BCAST;
467 m->m_flags |= M_MCAST;
468 if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1);
471 /* Let BPF have it before we strip the header. */
472 infiniband_bpf_mtap(ifp, m);
474 /* Allow monitor mode to claim this frame, after stats are updated. */
475 if (ifp->if_flags & IFF_MONITOR) {
480 /* Direct packet to correct FIB based on interface config. */
481 M_SETFIB(m, ifp->if_fib);
483 /* Handle input from a lagg<N> port */
484 if (ifp->if_type == IFT_INFINIBANDLAG) {
485 KASSERT(lagg_input_infiniband_p != NULL,
486 ("%s: if_lagg not loaded!", __func__));
487 m = (*lagg_input_infiniband_p)(ifp, m);
488 if (__predict_false(m == NULL))
490 ifp = m->m_pkthdr.rcvif;
494 * Dispatch frame to upper layer.
496 switch (ibh->ib_protocol) {
498 case htons(ETHERTYPE_IP):
502 case htons(ETHERTYPE_ARP):
503 if (ifp->if_flags & IFF_NOARP) {
504 /* Discard packet if ARP is disabled on interface */
512 case htons(ETHERTYPE_IPV6):
517 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
522 /* Strip off the Infiniband header. */
523 m_adj(m, INFINIBAND_HDR_LEN);
527 * Tag the mbuf with an appropriate MAC label before any other
528 * consumers can get to it.
530 mac_ifnet_create_mbuf(ifp, m);
532 /* Allow monitor mode to claim this frame, after stats are updated. */
533 netisr_dispatch(isr, m);
535 if (__predict_false(needs_epoch))
541 infiniband_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
544 struct sockaddr_dl *sdl;
546 struct sockaddr_in *sin;
549 struct sockaddr_in6 *sin6;
553 switch (sa->sa_family) {
556 * No mapping needed. Just check that it's a valid MC address.
558 sdl = (struct sockaddr_dl *)sa;
559 e_addr = LLADDR(sdl);
560 if (!INFINIBAND_IS_MULTICAST(e_addr))
561 return (EADDRNOTAVAIL);
567 sin = (struct sockaddr_in *)sa;
568 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
569 return (EADDRNOTAVAIL);
570 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
571 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
572 e_addr = LLADDR(sdl);
573 infiniband_ipv4_multicast_map(
574 sin->sin_addr.s_addr, ifp->if_broadcastaddr, e_addr);
575 *llsa = (struct sockaddr *)sdl;
580 sin6 = (struct sockaddr_in6 *)sa;
582 * An IP6 address of 0 means listen to all of the
583 * multicast address used for IP6. This has no meaning
586 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
587 return (EADDRNOTAVAIL);
588 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
589 return (EADDRNOTAVAIL);
590 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
591 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
592 e_addr = LLADDR(sdl);
593 infiniband_ipv6_multicast_map(
594 &sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
595 *llsa = (struct sockaddr *)sdl;
599 return (EAFNOSUPPORT);
604 infiniband_ifattach(struct ifnet *ifp, const uint8_t *lla, const uint8_t *llb)
606 struct sockaddr_dl *sdl;
610 ifp->if_addrlen = INFINIBAND_ADDR_LEN;
611 ifp->if_hdrlen = INFINIBAND_HDR_LEN;
612 ifp->if_mtu = INFINIBAND_MTU;
614 ifp->if_output = infiniband_output;
615 ifp->if_input = infiniband_input;
616 ifp->if_resolvemulti = infiniband_resolvemulti;
617 ifp->if_requestencap = infiniband_requestencap;
619 if (ifp->if_baudrate == 0)
620 ifp->if_baudrate = IF_Gbps(10); /* default value */
622 ifp->if_broadcastaddr = llb;
625 KASSERT(ifa != NULL, ("%s: no lladdr!\n", __func__));
626 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
627 sdl->sdl_type = IFT_INFINIBAND;
628 sdl->sdl_alen = ifp->if_addrlen;
631 memcpy(LLADDR(sdl), lla, ifp->if_addrlen);
633 if (ifp->if_hw_addr != NULL)
634 memcpy(ifp->if_hw_addr, lla, ifp->if_addrlen);
639 /* Attach ethernet compatible network device */
640 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN);
642 /* Announce Infiniband MAC address if non-zero. */
643 for (i = 0; i < ifp->if_addrlen; i++)
646 if (i != ifp->if_addrlen)
647 if_printf(ifp, "Infiniband address: %20D\n", lla, ":");
649 /* Add necessary bits are setup; announce it now. */
650 EVENTHANDLER_INVOKE(infiniband_ifattach_event, ifp);
652 if (IS_DEFAULT_VNET(curvnet))
653 devctl_notify("INFINIBAND", ifp->if_xname, "IFATTACH", NULL);
657 * Perform common duties while detaching an Infiniband interface
660 infiniband_ifdetach(struct ifnet *ifp)
667 infiniband_modevent(module_t mod, int type, void *data)
678 static moduledata_t infiniband_mod = {
679 .name = "if_infiniband",
680 .evhand = &infiniband_modevent,
683 DECLARE_MODULE(if_infiniband, infiniband_mod, SI_SUB_INIT_IF, SI_ORDER_ANY);
684 MODULE_VERSION(if_infiniband, 1);