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"
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/devctl.h>
35 #include <sys/eventhandler.h>
36 #include <sys/kernel.h>
38 #include <sys/module.h>
39 #include <sys/socket.h>
40 #include <sys/sysctl.h>
43 #include <net/ethernet.h>
44 #include <net/infiniband.h>
46 #include <net/if_var.h>
47 #include <net/if_dl.h>
48 #include <net/if_media.h>
49 #include <net/if_lagg.h>
50 #include <net/if_llatbl.h>
51 #include <net/if_types.h>
52 #include <net/netisr.h>
53 #include <net/route.h>
54 #include <netinet/if_ether.h>
55 #include <netinet/in.h>
56 #include <netinet/ip6.h>
57 #include <netinet6/in6_var.h>
58 #include <netinet6/nd6.h>
60 #include <security/mac/mac_framework.h>
62 /* if_lagg(4) support */
63 struct mbuf *(*lagg_input_infiniband_p)(struct ifnet *, struct mbuf *);
67 infiniband_ipv4_multicast_map(uint32_t addr,
68 const uint8_t *broadcast, uint8_t *buf)
73 scope = broadcast[5] & 0xF;
80 buf[5] = 0x10 | scope;
83 buf[8] = broadcast[8];
84 buf[9] = broadcast[9];
91 buf[16] = (addr >> 24) & 0xff;
92 buf[17] = (addr >> 16) & 0xff;
93 buf[18] = (addr >> 8) & 0xff;
94 buf[19] = addr & 0xff;
100 infiniband_ipv6_multicast_map(const struct in6_addr *addr,
101 const uint8_t *broadcast, uint8_t *buf)
105 scope = broadcast[5] & 0xF;
112 buf[5] = 0x10 | scope;
115 buf[8] = broadcast[8];
116 buf[9] = broadcast[9];
117 memcpy(&buf[10], &addr->s6_addr[6], 10);
122 * This is for clients that have an infiniband_header in the mbuf.
125 infiniband_bpf_mtap(struct ifnet *ifp, struct mbuf *mb)
127 struct infiniband_header *ibh;
128 struct ether_header eh;
130 if (mb->m_len < sizeof(*ibh))
133 ibh = mtod(mb, struct infiniband_header *);
134 eh.ether_type = ibh->ib_protocol;
135 memset(eh.ether_shost, 0, ETHER_ADDR_LEN);
136 memcpy(eh.ether_dhost, ibh->ib_hwaddr + 4, ETHER_ADDR_LEN);
137 mb->m_data += sizeof(*ibh);
138 mb->m_len -= sizeof(*ibh);
139 mb->m_pkthdr.len -= sizeof(*ibh);
140 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
141 mb->m_data -= sizeof(*ibh);
142 mb->m_len += sizeof(*ibh);
143 mb->m_pkthdr.len += sizeof(*ibh);
147 update_mbuf_csumflags(struct mbuf *src, struct mbuf *dst)
151 if (src->m_pkthdr.csum_flags & CSUM_IP)
152 csum_flags |= (CSUM_IP_CHECKED|CSUM_IP_VALID);
153 if (src->m_pkthdr.csum_flags & CSUM_DELAY_DATA)
154 csum_flags |= (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
155 if (src->m_pkthdr.csum_flags & CSUM_SCTP)
156 csum_flags |= CSUM_SCTP_VALID;
157 dst->m_pkthdr.csum_flags |= csum_flags;
158 if (csum_flags & CSUM_DATA_VALID)
159 dst->m_pkthdr.csum_data = 0xffff;
163 * Handle link-layer encapsulation requests.
166 infiniband_requestencap(struct ifnet *ifp, struct if_encap_req *req)
168 struct infiniband_header *ih;
171 const uint8_t *lladdr;
173 if (req->rtype != IFENCAP_LL)
176 if (req->bufsize < INFINIBAND_HDR_LEN)
179 ih = (struct infiniband_header *)req->buf;
180 lladdr = req->lladdr;
183 switch (req->family) {
185 etype = htons(ETHERTYPE_IP);
188 etype = htons(ETHERTYPE_IPV6);
191 ah = (struct arphdr *)req->hdata;
192 ah->ar_hrd = htons(ARPHRD_INFINIBAND);
194 switch (ntohs(ah->ar_op)) {
195 case ARPOP_REVREQUEST:
197 etype = htons(ETHERTYPE_REVARP);
202 etype = htons(ETHERTYPE_ARP);
206 if (req->flags & IFENCAP_FLAG_BROADCAST)
207 lladdr = ifp->if_broadcastaddr;
210 return (EAFNOSUPPORT);
213 ih->ib_protocol = etype;
215 memcpy(ih->ib_hwaddr, lladdr, INFINIBAND_ADDR_LEN);
216 req->bufsize = sizeof(struct infiniband_header);
222 infiniband_resolve_addr(struct ifnet *ifp, struct mbuf *m,
223 const struct sockaddr *dst, struct route *ro, uint8_t *phdr,
224 uint32_t *pflags, struct llentry **plle)
226 struct infiniband_header *ih;
227 uint32_t lleflags = 0;
232 ih = (struct infiniband_header *)phdr;
234 switch (dst->sa_family) {
237 if ((m->m_flags & (M_BCAST | M_MCAST)) == 0) {
238 error = arpresolve(ifp, 0, m, dst, phdr, &lleflags, plle);
240 if (m->m_flags & M_BCAST) {
241 memcpy(ih->ib_hwaddr, ifp->if_broadcastaddr,
242 INFINIBAND_ADDR_LEN);
244 infiniband_ipv4_multicast_map(
245 ((const struct sockaddr_in *)dst)->sin_addr.s_addr,
246 ifp->if_broadcastaddr, ih->ib_hwaddr);
248 ih->ib_protocol = htons(ETHERTYPE_IP);
255 if ((m->m_flags & M_MCAST) == 0) {
256 error = nd6_resolve(ifp, 0, m, dst, phdr, &lleflags, plle);
258 infiniband_ipv6_multicast_map(
259 &((const struct sockaddr_in6 *)dst)->sin6_addr,
260 ifp->if_broadcastaddr, ih->ib_hwaddr);
261 ih->ib_protocol = htons(ETHERTYPE_IPV6);
267 if_printf(ifp, "can't handle af%d\n", dst->sa_family);
270 return (EAFNOSUPPORT);
273 if (error == EHOSTDOWN) {
274 if (ro != NULL && (ro->ro_flags & RT_HAS_GW) != 0)
275 error = EHOSTUNREACH;
281 *pflags = RT_MAY_LOOP;
282 if (lleflags & LLE_IFADDR)
289 * Infiniband output routine.
292 infiniband_output(struct ifnet *ifp, struct mbuf *m,
293 const struct sockaddr *dst, struct route *ro)
295 uint8_t linkhdr[INFINIBAND_HDR_LEN];
297 struct llentry *lle = NULL;
298 struct infiniband_header *ih;
300 int hlen; /* link layer header length */
310 /* XXX BPF uses ro_prepend */
311 if (ro->ro_prepend != NULL) {
312 phdr = ro->ro_prepend;
314 } else if (!(m->m_flags & (M_BCAST | M_MCAST))) {
315 if ((ro->ro_flags & RT_LLE_CACHE) != 0) {
318 (lle->la_flags & LLE_VALID) == 0) {
320 lle = NULL; /* redundant */
324 /* if we lookup, keep cache */
328 * Notify LLE code that
332 llentry_mark_used(lle);
335 phdr = lle->r_linkdata;
336 hlen = lle->r_hdrlen;
337 pflags = lle->r_flags;
343 error = mac_ifnet_check_transmit(ifp, m);
349 if (ifp->if_flags & IFF_MONITOR) {
353 if (!((ifp->if_flags & IFF_UP) &&
354 (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
360 /* No prepend data supplied. Try to calculate ourselves. */
362 hlen = INFINIBAND_HDR_LEN;
363 error = infiniband_resolve_addr(ifp, m, dst, ro, phdr, &pflags,
364 addref ? &lle : NULL);
365 if (addref && lle != NULL)
368 return (error == EWOULDBLOCK ? 0 : error);
371 if ((pflags & RT_L2_ME) != 0) {
372 update_mbuf_csumflags(m, m);
373 return (if_simloop(ifp, m, dst->sa_family, 0));
377 * Add local infiniband header. If no space in first mbuf,
380 M_PREPEND(m, INFINIBAND_HDR_LEN, M_NOWAIT);
385 if ((pflags & RT_HAS_HEADER) == 0) {
386 ih = mtod(m, struct infiniband_header *);
387 memcpy(ih, phdr, hlen);
391 * Queue message on interface, update output statistics if
392 * successful, and start output if interface not yet active.
394 return (ifp->if_transmit(ifp, m));
402 * Process a received Infiniband packet.
405 infiniband_input(struct ifnet *ifp, struct mbuf *m)
407 struct infiniband_header *ibh;
408 struct epoch_tracker et;
411 CURVNET_SET_QUIET(ifp->if_vnet);
413 if ((ifp->if_flags & IFF_UP) == 0) {
414 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
419 ibh = mtod(m, struct infiniband_header *);
422 * Reset layer specific mbuf flags to avoid confusing upper
425 m->m_flags &= ~M_VLANTAG;
428 if (INFINIBAND_IS_MULTICAST(ibh->ib_hwaddr)) {
429 if (memcmp(ibh->ib_hwaddr, ifp->if_broadcastaddr,
430 ifp->if_addrlen) == 0)
431 m->m_flags |= M_BCAST;
433 m->m_flags |= M_MCAST;
434 if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1);
437 /* Let BPF have it before we strip the header. */
438 INFINIBAND_BPF_MTAP(ifp, m);
440 /* Allow monitor mode to claim this frame, after stats are updated. */
441 if (ifp->if_flags & IFF_MONITOR) {
446 /* Direct packet to correct FIB based on interface config. */
447 M_SETFIB(m, ifp->if_fib);
449 /* Handle input from a lagg<N> port */
450 if (ifp->if_type == IFT_INFINIBANDLAG) {
451 KASSERT(lagg_input_infiniband_p != NULL,
452 ("%s: if_lagg not loaded!", __func__));
453 m = (*lagg_input_infiniband_p)(ifp, m);
454 if (__predict_false(m == NULL))
456 ifp = m->m_pkthdr.rcvif;
460 * Dispatch frame to upper layer.
462 switch (ibh->ib_protocol) {
464 case htons(ETHERTYPE_IP):
468 case htons(ETHERTYPE_ARP):
469 if (ifp->if_flags & IFF_NOARP) {
470 /* Discard packet if ARP is disabled on interface */
478 case htons(ETHERTYPE_IPV6):
483 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
488 /* Strip off the Infiniband header. */
489 m_adj(m, INFINIBAND_HDR_LEN);
493 * Tag the mbuf with an appropriate MAC label before any other
494 * consumers can get to it.
496 mac_ifnet_create_mbuf(ifp, m);
498 /* Allow monitor mode to claim this frame, after stats are updated. */
500 netisr_dispatch(isr, m);
507 infiniband_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
510 struct sockaddr_dl *sdl;
512 struct sockaddr_in *sin;
515 struct sockaddr_in6 *sin6;
519 switch (sa->sa_family) {
522 * No mapping needed. Just check that it's a valid MC address.
524 sdl = (struct sockaddr_dl *)sa;
525 e_addr = LLADDR(sdl);
526 if (!INFINIBAND_IS_MULTICAST(e_addr))
527 return (EADDRNOTAVAIL);
533 sin = (struct sockaddr_in *)sa;
534 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
535 return (EADDRNOTAVAIL);
536 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
537 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
538 e_addr = LLADDR(sdl);
539 infiniband_ipv4_multicast_map(
540 sin->sin_addr.s_addr, ifp->if_broadcastaddr, e_addr);
541 *llsa = (struct sockaddr *)sdl;
546 sin6 = (struct sockaddr_in6 *)sa;
548 * An IP6 address of 0 means listen to all of the
549 * multicast address used for IP6. This has no meaning
552 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
553 return (EADDRNOTAVAIL);
554 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
555 return (EADDRNOTAVAIL);
556 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
557 sdl->sdl_alen = INFINIBAND_ADDR_LEN;
558 e_addr = LLADDR(sdl);
559 infiniband_ipv6_multicast_map(
560 &sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
561 *llsa = (struct sockaddr *)sdl;
565 return (EAFNOSUPPORT);
570 infiniband_ifattach(struct ifnet *ifp, const uint8_t *lla, const uint8_t *llb)
572 struct sockaddr_dl *sdl;
576 ifp->if_addrlen = INFINIBAND_ADDR_LEN;
577 ifp->if_hdrlen = INFINIBAND_HDR_LEN;
578 ifp->if_mtu = INFINIBAND_MTU;
580 ifp->if_output = infiniband_output;
581 ifp->if_input = infiniband_input;
582 ifp->if_resolvemulti = infiniband_resolvemulti;
583 ifp->if_requestencap = infiniband_requestencap;
585 if (ifp->if_baudrate == 0)
586 ifp->if_baudrate = IF_Gbps(10); /* default value */
588 ifp->if_broadcastaddr = llb;
591 KASSERT(ifa != NULL, ("%s: no lladdr!\n", __func__));
592 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
593 sdl->sdl_type = IFT_INFINIBAND;
594 sdl->sdl_alen = ifp->if_addrlen;
597 memcpy(LLADDR(sdl), lla, ifp->if_addrlen);
599 if (ifp->if_hw_addr != NULL)
600 memcpy(ifp->if_hw_addr, lla, ifp->if_addrlen);
605 /* Attach ethernet compatible network device */
606 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN);
608 /* Announce Infiniband MAC address if non-zero. */
609 for (i = 0; i < ifp->if_addrlen; i++)
612 if (i != ifp->if_addrlen)
613 if_printf(ifp, "Infiniband address: %20D\n", lla, ":");
615 /* Add necessary bits are setup; announce it now. */
616 EVENTHANDLER_INVOKE(infiniband_ifattach_event, ifp);
618 if (IS_DEFAULT_VNET(curvnet))
619 devctl_notify("INFINIBAND", ifp->if_xname, "IFATTACH", NULL);
623 * Perform common duties while detaching an Infiniband interface
626 infiniband_ifdetach(struct ifnet *ifp)
633 infiniband_modevent(module_t mod, int type, void *data)
644 static moduledata_t infiniband_mod = {
645 .name = "if_infiniband",
646 .evhand = &infiniband_modevent,
649 DECLARE_MODULE(if_infiniband, infiniband_mod, SI_SUB_INIT_IF, SI_ORDER_ANY);
650 MODULE_VERSION(if_infiniband, 1);