2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2008 The FreeBSD Foundation
5 * Copyright (c) 2009-2021 Bjoern A. Zeeb <bz@FreeBSD.org>
7 * This software was developed by CK Software GmbH under sponsorship
8 * from the FreeBSD Foundation.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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
33 * A pair of virtual back-to-back connected ethernet like interfaces
34 * (``two interfaces with a virtual cross-over cable'').
36 * This is mostly intended to be used to provide connectivity between
37 * different virtual network stack instances.
40 #include <sys/cdefs.h>
43 #include "opt_inet6.h"
45 #include <sys/param.h>
48 #include <sys/interrupt.h>
50 #include <sys/kernel.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
54 #include <sys/module.h>
56 #include <sys/queue.h>
57 #include <sys/sched.h>
59 #include <sys/socket.h>
60 #include <sys/sockio.h>
61 #include <sys/taskqueue.h>
64 #include <net/ethernet.h>
66 #include <net/if_var.h>
67 #include <net/if_clone.h>
68 #include <net/if_media.h>
69 #include <net/if_var.h>
70 #include <net/if_types.h>
71 #include <net/netisr.h>
73 #include <net/rss_config.h>
75 #include <netinet/in_rss.h>
78 #include <netinet6/in6_rss.h>
83 static int epair_clone_match(struct if_clone *, const char *);
84 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
85 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
87 static const char epairname[] = "epair";
88 #define RXRSIZE 4096 /* Probably overkill by 4-8x. */
90 static MALLOC_DEFINE(M_EPAIR, epairname,
91 "Pair of virtual cross-over connected Ethernet-like interfaces");
93 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
94 #define V_epair_cloner VNET(epair_cloner)
96 static unsigned int next_index = 0;
97 #define EPAIR_LOCK_INIT() mtx_init(&epair_n_index_mtx, "epairidx", \
99 #define EPAIR_LOCK_DESTROY() mtx_destroy(&epair_n_index_mtx)
100 #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx)
101 #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx)
114 struct epair_softc *sc;
117 static struct mtx epair_n_index_mtx;
119 struct ifnet *ifp; /* This ifp. */
120 struct ifnet *oifp; /* other ifp of pair. */
122 struct epair_queue *queues;
123 struct ifmedia media; /* Media config (fake). */
124 STAILQ_ENTRY(epair_softc) entry;
127 struct epair_tasks_t {
129 struct taskqueue *tq[MAXCPU];
132 static struct epair_tasks_t epair_tasks;
135 epair_clear_mbuf(struct mbuf *m)
139 /* Remove any CSUM_SND_TAG as ether_input will barf. */
140 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
141 m_snd_tag_rele(m->m_pkthdr.snd_tag);
142 m->m_pkthdr.snd_tag = NULL;
143 m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
146 /* Clear vlan information. */
147 m->m_flags &= ~M_VLANTAG;
148 m->m_pkthdr.ether_vtag = 0;
150 m_tag_delete_nonpersistent(m);
154 epair_tx_start_deferred(void *arg, int pending)
156 struct epair_queue *q = (struct epair_queue *)arg;
164 CURVNET_SET(ifp->if_vnet);
167 m = mbufq_flush(&q->q);
168 q->state = EPAIR_QUEUE_RUNNING;
172 n = STAILQ_NEXT(m, m_stailqpkt);
179 * Avoid flushing the queue more than once per task. We can otherwise
180 * end up starving ourselves in a multi-epair routing configuration.
183 if (mbufq_len(&q->q) > 0) {
185 q->state = EPAIR_QUEUE_WAKING;
188 q->state = EPAIR_QUEUE_IDLE;
193 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
199 static struct epair_queue *
200 epair_select_queue(struct epair_softc *sc, struct mbuf *m)
204 struct ether_header *eh;
207 ret = rss_m2bucket(m, &bucket);
209 /* Actually hash the packet. */
210 eh = mtod(m, struct ether_header *);
212 switch (ntohs(eh->ether_type)) {
215 rss_soft_m2cpuid_v4(m, 0, &bucket);
220 rss_soft_m2cpuid_v6(m, 0, &bucket);
228 bucket %= sc->num_queues;
232 return (&sc->queues[bucket]);
236 epair_prepare_mbuf(struct mbuf *m, struct ifnet *src_ifp)
240 if_setrcvif(m, src_ifp);
241 M_SETFIB(m, src_ifp->if_fib);
243 MPASS(m->m_nextpkt == NULL);
244 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
248 epair_menq(struct mbuf *m, struct epair_softc *osc)
250 struct epair_queue *q;
251 struct ifnet *ifp, *oifp;
256 * I know this looks weird. We pass the "other sc" as we need that one
257 * and can get both ifps from it as well.
262 epair_prepare_mbuf(m, oifp);
264 /* Save values as once the mbuf is queued, it's not ours anymore. */
265 len = m->m_pkthdr.len;
266 mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
268 q = epair_select_queue(osc, m);
271 if (q->state == EPAIR_QUEUE_IDLE) {
272 q->state = EPAIR_QUEUE_WAKING;
273 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
275 error = mbufq_enqueue(&q->q, m);
280 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
282 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
283 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
285 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
286 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
291 epair_start(struct ifnet *ifp)
294 struct epair_softc *sc;
298 * We get packets here from ether_output via if_handoff()
299 * and need to put them into the input queue of the oifp
300 * and will put the packet into the receive-queue (rxq) of the
301 * other interface (oifp) of our pair.
307 IFQ_DEQUEUE(&ifp->if_snd, m);
313 /* In case either interface is not usable drop the packet. */
314 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
315 (ifp->if_flags & IFF_UP) == 0 ||
316 (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
317 (oifp->if_flags & IFF_UP) == 0) {
322 (void) epair_menq(m, sc);
327 epair_transmit(struct ifnet *ifp, struct mbuf *m)
329 struct epair_softc *sc;
341 * We are not going to use the interface en/dequeue mechanism
342 * on the TX side. We are called from ether_output_frame()
343 * and will put the packet into the receive-queue (rxq) of the
344 * other interface (oifp) of our pair.
346 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
348 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
351 if ((ifp->if_flags & IFF_UP) == 0) {
353 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
360 * In case the outgoing interface is not usable,
365 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
366 (oifp->if_flags & IFF_UP) == 0) {
367 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
373 len = m->m_pkthdr.len;
374 mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
377 /* Support ALTQ via the classic if_start() path. */
378 IF_LOCK(&ifp->if_snd);
379 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
380 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
382 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
383 IF_UNLOCK(&ifp->if_snd);
385 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
387 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
392 IF_UNLOCK(&ifp->if_snd);
395 epair_menq(m, oifp->if_softc);
400 epair_media_change(struct ifnet *ifp __unused)
408 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
411 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
412 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
416 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
418 struct epair_softc *sc;
422 ifr = (struct ifreq *)data;
433 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
437 /* We basically allow all kinds of MTUs. */
438 ifp->if_mtu = ifr->ifr_mtu;
443 /* Let the common ethernet handler process this. */
444 error = ether_ioctl(ifp, cmd, data);
452 epair_init(void *dummy __unused)
457 * Interface cloning functions.
458 * We use our private ones so that we can create/destroy our secondary
459 * device along with the primary one.
462 epair_clone_match(struct if_clone *ifc, const char *name)
467 * Our base name is epair.
468 * Our interfaces will be named epair<n>[ab].
469 * So accept anything of the following list:
472 * but not the epair<n>[ab] versions.
474 if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
477 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
478 if (*cp < '0' || *cp > '9')
486 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
489 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
492 /* Copy epairNa etheraddr and change the last byte. */
493 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
495 ether_ifattach(ifp, eaddr);
497 if_clone_addif(ifc, ifp);
501 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
503 struct epair_softc *sca, *scb;
506 int error, unit, wildcard;
510 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
512 /* Try to see if a special unit was requested. */
513 error = ifc_name2unit(name, &unit);
516 wildcard = (unit < 0);
518 error = ifc_alloc_unit(ifc, &unit);
523 * If no unit had been given, we need to adjust the ifName.
524 * Also make sure there is space for our extra [ab] suffix.
526 for (dp = name; *dp != '\0'; dp++);
528 error = snprintf(dp, len - (dp - name), "%d", unit);
529 if (error > len - (dp - name) - 1) {
530 /* ifName too long. */
531 ifc_free_unit(ifc, unit);
536 if (len - (dp - name) - 1 < 1) {
537 /* No space left for our [ab] suffix. */
538 ifc_free_unit(ifc, unit);
542 /* Must not change dp so we can replace 'a' by 'b' later. */
545 /* Check if 'a' and 'b' interfaces already exist. */
546 if (ifunit(name) != NULL)
549 if (ifunit(name) != NULL)
552 /* Allocate memory for both [ab] interfaces */
553 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
554 sca->ifp = if_alloc(IFT_ETHER);
555 sca->num_queues = epair_tasks.tasks;
556 if (sca->ifp == NULL) {
558 ifc_free_unit(ifc, unit);
561 sca->queues = mallocarray(sca->num_queues, sizeof(struct epair_queue),
563 for (int i = 0; i < sca->num_queues; i++) {
564 struct epair_queue *q = &sca->queues[i];
566 q->state = EPAIR_QUEUE_IDLE;
567 mtx_init(&q->mtx, "epaiq", NULL, MTX_DEF | MTX_NEW);
568 mbufq_init(&q->q, RXRSIZE);
570 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
573 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
574 scb->ifp = if_alloc(IFT_ETHER);
575 scb->num_queues = epair_tasks.tasks;
576 if (scb->ifp == NULL) {
580 ifc_free_unit(ifc, unit);
583 scb->queues = mallocarray(scb->num_queues, sizeof(struct epair_queue),
585 for (int i = 0; i < scb->num_queues; i++) {
586 struct epair_queue *q = &scb->queues[i];
588 q->state = EPAIR_QUEUE_IDLE;
589 mtx_init(&q->mtx, "epaiq", NULL, MTX_DEF | MTX_NEW);
590 mbufq_init(&q->q, RXRSIZE);
592 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
596 * Cross-reference the interfaces so we will be able to free both.
598 sca->oifp = scb->ifp;
599 scb->oifp = sca->ifp;
603 /* Get an approximate distribution. */
604 hash = next_index % mp_ncpus;
610 /* Initialise pseudo media types. */
611 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
612 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
613 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
614 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
615 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
616 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
618 /* Finish initialization of interface <n>a. */
621 strlcpy(ifp->if_xname, name, IFNAMSIZ);
622 ifp->if_dname = epairname;
623 ifp->if_dunit = unit;
624 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
625 ifp->if_flags |= IFF_KNOWSEPOCH;
626 ifp->if_capabilities = IFCAP_VLAN_MTU;
627 ifp->if_capenable = IFCAP_VLAN_MTU;
628 ifp->if_start = epair_start;
629 ifp->if_ioctl = epair_ioctl;
630 ifp->if_init = epair_init;
631 if_setsendqlen(ifp, ifqmaxlen);
632 if_setsendqready(ifp);
635 * Calculate the etheraddr hashing the hostid and the
636 * interface index. The result would be hopefully unique.
637 * Note that the "a" component of an epair instance may get moved
638 * to a different VNET after creation. In that case its index
639 * will be freed and the index can get reused by new epair instance.
640 * Make sure we do not create same etheraddr again.
642 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
644 arc4rand(&hostid, sizeof(hostid), 0);
647 if (ifp->if_index > next_index)
648 next_index = ifp->if_index;
652 key[0] = (uint32_t)next_index;
654 key[1] = (uint32_t)(hostid & 0xffffffff);
655 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
656 hash = jenkins_hash32(key, 3, 0);
659 memcpy(&eaddr[1], &hash, 4);
661 ether_ifattach(ifp, eaddr);
662 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
663 ifp->if_transmit = epair_transmit;
665 /* Swap the name and finish initialization of interface <n>b. */
670 strlcpy(ifp->if_xname, name, IFNAMSIZ);
671 ifp->if_dname = epairname;
672 ifp->if_dunit = unit;
673 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
674 ifp->if_flags |= IFF_KNOWSEPOCH;
675 ifp->if_capabilities = IFCAP_VLAN_MTU;
676 ifp->if_capenable = IFCAP_VLAN_MTU;
677 ifp->if_start = epair_start;
678 ifp->if_ioctl = epair_ioctl;
679 ifp->if_init = epair_init;
680 if_setsendqlen(ifp, ifqmaxlen);
681 if_setsendqready(ifp);
682 /* We need to play some tricks here for the second interface. */
683 strlcpy(name, epairname, len);
685 /* Correctly set the name for the cloner list. */
686 strlcpy(name, scb->ifp->if_xname, len);
687 epair_clone_add(ifc, scb);
689 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
690 ifp->if_transmit = epair_transmit;
693 * Restore name to <n>a as the ifp for this will go into the
694 * cloner list for the initial call.
696 strlcpy(name, sca->ifp->if_xname, len);
698 /* Tell the world, that we are ready to rock. */
699 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
700 if_link_state_change(sca->ifp, LINK_STATE_UP);
701 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
702 if_link_state_change(scb->ifp, LINK_STATE_UP);
708 epair_drain_rings(struct epair_softc *sc)
710 for (int i = 0; i < sc->num_queues; i++) {
711 struct epair_queue *q;
716 m = mbufq_flush(&q->q);
719 for (; m != NULL; m = n) {
727 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
730 struct epair_softc *sca, *scb;
734 * In case we called into if_clone_destroyif() ourselves
735 * again to remove the second interface, the softc will be
736 * NULL. In that case so not do anything but return success.
738 if (ifp->if_softc == NULL)
741 unit = ifp->if_dunit;
744 scb = oifp->if_softc;
746 /* Frist get the interfaces down and detached. */
747 if_link_state_change(ifp, LINK_STATE_DOWN);
748 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
749 if_link_state_change(oifp, LINK_STATE_DOWN);
750 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
753 ether_ifdetach(oifp);
755 /* Third free any queued packets and all the resources. */
756 CURVNET_SET_QUIET(oifp->if_vnet);
757 epair_drain_rings(scb);
758 oifp->if_softc = NULL;
759 error = if_clone_destroyif(ifc, oifp);
761 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
764 ifmedia_removeall(&scb->media);
765 for (int i = 0; i < scb->num_queues; i++) {
766 struct epair_queue *q = &scb->queues[i];
767 mtx_destroy(&q->mtx);
769 free(scb->queues, M_EPAIR);
773 epair_drain_rings(sca);
775 ifmedia_removeall(&sca->media);
776 for (int i = 0; i < sca->num_queues; i++) {
777 struct epair_queue *q = &sca->queues[i];
778 mtx_destroy(&q->mtx);
780 free(sca->queues, M_EPAIR);
783 /* Last free the cloner unit. */
784 ifc_free_unit(ifc, unit);
790 vnet_epair_init(const void *unused __unused)
793 V_epair_cloner = if_clone_advanced(epairname, 0,
794 epair_clone_match, epair_clone_create, epair_clone_destroy);
796 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
797 vnet_epair_init, NULL);
800 vnet_epair_uninit(const void *unused __unused)
803 if_clone_detach(V_epair_cloner);
805 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
806 vnet_epair_uninit, NULL);
812 epair_tasks.tasks = 0;
820 /* Pin to this CPU so we get appropriate NUMA allocations. */
821 thread_lock(curthread);
822 sched_bind(curthread, cpu);
823 thread_unlock(curthread);
825 snprintf(name, sizeof(name), "epair_task_%d", cpu);
827 epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK,
828 taskqueue_thread_enqueue,
829 &epair_tasks.tq[cpu]);
830 CPU_SETOF(cpu, &cpu_mask);
831 taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET,
832 &cpu_mask, "%s", name);
836 thread_lock(curthread);
837 sched_unbind(curthread);
838 thread_unlock(curthread);
840 snprintf(name, sizeof(name), "epair_task");
842 epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK,
843 taskqueue_thread_enqueue,
845 taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name);
847 epair_tasks.tasks = 1;
854 epair_mod_cleanup(void)
857 for (int i = 0; i < epair_tasks.tasks; i++) {
858 taskqueue_drain_all(epair_tasks.tq[i]);
859 taskqueue_free(epair_tasks.tq[i]);
864 epair_modevent(module_t mod, int type, void *data)
871 ret = epair_mod_init();
875 printf("%s: %s initialized.\n", __func__, epairname);
879 EPAIR_LOCK_DESTROY();
881 printf("%s: %s unloaded.\n", __func__, epairname);
889 static moduledata_t epair_mod = {
895 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
896 MODULE_VERSION(if_epair, 3);