2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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>
41 __FBSDID("$FreeBSD$");
45 #include "opt_inet6.h"
47 #include <sys/param.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>
62 #include <sys/types.h>
63 #include <sys/buf_ring.h>
65 #include <sys/interrupt.h>
68 #include <net/ethernet.h>
70 #include <net/if_var.h>
71 #include <net/if_clone.h>
72 #include <net/if_media.h>
73 #include <net/if_var.h>
74 #include <net/if_types.h>
75 #include <net/netisr.h>
77 #include <net/rss_config.h>
79 #include <netinet/in_rss.h>
82 #include <netinet6/in6_rss.h>
87 static int epair_clone_match(struct if_clone *, const char *);
88 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
89 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
91 static const char epairname[] = "epair";
92 #define RXRSIZE 4096 /* Probably overkill by 4-8x. */
94 static MALLOC_DEFINE(M_EPAIR, epairname,
95 "Pair of virtual cross-over connected Ethernet-like interfaces");
97 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
98 #define V_epair_cloner VNET(epair_cloner)
100 static unsigned int next_index = 0;
101 #define EPAIR_LOCK_INIT() mtx_init(&epair_n_index_mtx, "epairidx", \
103 #define EPAIR_LOCK_DESTROY() mtx_destroy(&epair_n_index_mtx)
104 #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx)
105 #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx)
107 #define BIT_QUEUE_TASK 0
108 #define BIT_MBUF_QUEUED 1
113 struct buf_ring *rxring[2];
114 volatile int ridx; /* 0 || 1 */
115 volatile long state; /* taskqueue coordination */
117 struct epair_softc *sc;
120 static struct mtx epair_n_index_mtx;
122 struct ifnet *ifp; /* This ifp. */
123 struct ifnet *oifp; /* other ifp of pair. */
125 struct epair_queue *queues;
126 struct ifmedia media; /* Media config (fake). */
127 STAILQ_ENTRY(epair_softc) entry;
130 struct epair_tasks_t {
132 struct taskqueue *tq[MAXCPU];
135 static struct epair_tasks_t epair_tasks;
138 epair_clear_mbuf(struct mbuf *m)
140 /* Remove any CSUM_SND_TAG as ether_input will barf. */
141 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
142 m_snd_tag_rele(m->m_pkthdr.snd_tag);
143 m->m_pkthdr.snd_tag = NULL;
144 m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
147 m_tag_delete_nonpersistent(m);
151 epair_if_input(struct epair_softc *sc, struct epair_queue *q, int ridx)
157 CURVNET_SET(ifp->if_vnet);
158 while (! buf_ring_empty(q->rxring[ridx])) {
159 m = buf_ring_dequeue_mc(q->rxring[ridx]);
163 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
164 (*ifp->if_input)(ifp, m);
171 epair_tx_start_deferred(void *arg, int pending)
173 struct epair_queue *q = (struct epair_queue *)arg;
174 struct epair_softc *sc = q->sc;
178 ridx = atomic_load_int(&q->ridx);
180 nidx = (ridx == 0) ? 1 : 0;
181 } while (!atomic_fcmpset_int(&q->ridx, &ridx, nidx));
182 epair_if_input(sc, q, ridx);
184 atomic_clear_long(&q->state, (1 << BIT_QUEUE_TASK));
185 if (atomic_testandclear_long(&q->state, BIT_MBUF_QUEUED))
186 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
192 epair_menq(struct mbuf *m, struct epair_softc *osc)
194 struct ifnet *ifp, *oifp;
198 struct epair_queue *q = NULL;
201 struct ether_header *eh;
205 * I know this looks weird. We pass the "other sc" as we need that one
206 * and can get both ifps from it as well.
213 if_setrcvif(m, oifp);
214 M_SETFIB(m, oifp->if_fib);
216 /* Save values as once the mbuf is queued, it's not ours anymore. */
217 len = m->m_pkthdr.len;
220 MPASS(m->m_nextpkt == NULL);
221 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
224 ret = rss_m2bucket(m, &bucket);
226 /* Actually hash the packet. */
227 eh = mtod(m, struct ether_header *);
229 switch (ntohs(eh->ether_type)) {
232 rss_soft_m2cpuid_v4(m, 0, &bucket);
237 rss_soft_m2cpuid_v6(m, 0, &bucket);
245 bucket %= osc->num_queues;
249 q = &osc->queues[bucket];
251 atomic_set_long(&q->state, (1 << BIT_MBUF_QUEUED));
252 ridx = atomic_load_int(&q->ridx);
253 ret = buf_ring_enqueue(q->rxring[ridx], m);
256 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
261 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
263 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
264 * but as we bypass all this we have to duplicate
265 * the logic another time.
267 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
268 if (mflags & (M_BCAST|M_MCAST))
269 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
270 /* Someone else received the packet. */
271 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
273 if (!atomic_testandset_long(&q->state, BIT_QUEUE_TASK))
274 taskqueue_enqueue(epair_tasks.tq[bucket], &q->tx_task);
280 epair_start(struct ifnet *ifp)
283 struct epair_softc *sc;
287 * We get packets here from ether_output via if_handoff()
288 * and need to put them into the input queue of the oifp
289 * and will put the packet into the receive-queue (rxq) of the
290 * other interface (oifp) of our pair.
296 IFQ_DEQUEUE(&ifp->if_snd, m);
302 /* In case either interface is not usable drop the packet. */
303 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
304 (ifp->if_flags & IFF_UP) == 0 ||
305 (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
306 (oifp->if_flags & IFF_UP) == 0) {
311 (void) epair_menq(m, sc);
316 epair_transmit(struct ifnet *ifp, struct mbuf *m)
318 struct epair_softc *sc;
328 * We are not going to use the interface en/dequeue mechanism
329 * on the TX side. We are called from ether_output_frame()
330 * and will put the packet into the receive-queue (rxq) of the
331 * other interface (oifp) of our pair.
333 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
335 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
338 if ((ifp->if_flags & IFF_UP) == 0) {
340 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
347 * In case the outgoing interface is not usable,
352 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
353 (oifp->if_flags & IFF_UP) == 0) {
354 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
358 len = m->m_pkthdr.len;
362 /* Support ALTQ via the classic if_start() path. */
363 IF_LOCK(&ifp->if_snd);
364 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
365 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
367 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
368 IF_UNLOCK(&ifp->if_snd);
370 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
371 if (mflags & (M_BCAST|M_MCAST))
372 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
377 IF_UNLOCK(&ifp->if_snd);
380 error = epair_menq(m, oifp->if_softc);
385 epair_media_change(struct ifnet *ifp __unused)
393 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
396 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
397 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
401 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
403 struct epair_softc *sc;
407 ifr = (struct ifreq *)data;
418 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
422 /* We basically allow all kinds of MTUs. */
423 ifp->if_mtu = ifr->ifr_mtu;
428 /* Let the common ethernet handler process this. */
429 error = ether_ioctl(ifp, cmd, data);
437 epair_init(void *dummy __unused)
442 * Interface cloning functions.
443 * We use our private ones so that we can create/destroy our secondary
444 * device along with the primary one.
447 epair_clone_match(struct if_clone *ifc, const char *name)
452 * Our base name is epair.
453 * Our interfaces will be named epair<n>[ab].
454 * So accept anything of the following list:
457 * but not the epair<n>[ab] versions.
459 if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
462 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
463 if (*cp < '0' || *cp > '9')
471 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
474 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
477 /* Copy epairNa etheraddr and change the last byte. */
478 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
480 ether_ifattach(ifp, eaddr);
482 if_clone_addif(ifc, ifp);
486 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
488 struct epair_softc *sca, *scb;
491 int error, unit, wildcard;
495 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
497 /* Try to see if a special unit was requested. */
498 error = ifc_name2unit(name, &unit);
501 wildcard = (unit < 0);
503 error = ifc_alloc_unit(ifc, &unit);
508 * If no unit had been given, we need to adjust the ifName.
509 * Also make sure there is space for our extra [ab] suffix.
511 for (dp = name; *dp != '\0'; dp++);
513 error = snprintf(dp, len - (dp - name), "%d", unit);
514 if (error > len - (dp - name) - 1) {
515 /* ifName too long. */
516 ifc_free_unit(ifc, unit);
521 if (len - (dp - name) - 1 < 1) {
522 /* No space left for our [ab] suffix. */
523 ifc_free_unit(ifc, unit);
527 /* Must not change dp so we can replace 'a' by 'b' later. */
530 /* Check if 'a' and 'b' interfaces already exist. */
531 if (ifunit(name) != NULL)
534 if (ifunit(name) != NULL)
537 /* Allocate memory for both [ab] interfaces */
538 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
539 sca->ifp = if_alloc(IFT_ETHER);
540 sca->num_queues = epair_tasks.tasks;
541 if (sca->ifp == NULL) {
543 ifc_free_unit(ifc, unit);
546 sca->queues = mallocarray(sca->num_queues, sizeof(struct epair_queue),
548 for (int i = 0; i < sca->num_queues; i++) {
549 struct epair_queue *q = &sca->queues[i];
551 q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
552 q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
556 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
559 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
560 scb->ifp = if_alloc(IFT_ETHER);
561 scb->num_queues = epair_tasks.tasks;
562 if (scb->ifp == NULL) {
566 ifc_free_unit(ifc, unit);
569 scb->queues = mallocarray(scb->num_queues, sizeof(struct epair_queue),
571 for (int i = 0; i < scb->num_queues; i++) {
572 struct epair_queue *q = &scb->queues[i];
574 q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
575 q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
579 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
583 * Cross-reference the interfaces so we will be able to free both.
585 sca->oifp = scb->ifp;
586 scb->oifp = sca->ifp;
590 /* Get an approximate distribution. */
591 hash = next_index % mp_ncpus;
597 /* Initialise pseudo media types. */
598 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
599 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
600 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
601 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
602 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
603 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
605 /* Finish initialization of interface <n>a. */
608 strlcpy(ifp->if_xname, name, IFNAMSIZ);
609 ifp->if_dname = epairname;
610 ifp->if_dunit = unit;
611 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
612 ifp->if_flags |= IFF_KNOWSEPOCH;
613 ifp->if_capabilities = IFCAP_VLAN_MTU;
614 ifp->if_capenable = IFCAP_VLAN_MTU;
615 ifp->if_start = epair_start;
616 ifp->if_ioctl = epair_ioctl;
617 ifp->if_init = epair_init;
618 if_setsendqlen(ifp, ifqmaxlen);
619 if_setsendqready(ifp);
622 * Calculate the etheraddr hashing the hostid and the
623 * interface index. The result would be hopefully unique.
624 * Note that the "a" component of an epair instance may get moved
625 * to a different VNET after creation. In that case its index
626 * will be freed and the index can get reused by new epair instance.
627 * Make sure we do not create same etheraddr again.
629 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
631 arc4rand(&hostid, sizeof(hostid), 0);
634 if (ifp->if_index > next_index)
635 next_index = ifp->if_index;
639 key[0] = (uint32_t)next_index;
641 key[1] = (uint32_t)(hostid & 0xffffffff);
642 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
643 hash = jenkins_hash32(key, 3, 0);
646 memcpy(&eaddr[1], &hash, 4);
648 ether_ifattach(ifp, eaddr);
649 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
650 ifp->if_transmit = epair_transmit;
652 /* Swap the name and finish initialization of interface <n>b. */
657 strlcpy(ifp->if_xname, name, IFNAMSIZ);
658 ifp->if_dname = epairname;
659 ifp->if_dunit = unit;
660 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
661 ifp->if_flags |= IFF_KNOWSEPOCH;
662 ifp->if_capabilities = IFCAP_VLAN_MTU;
663 ifp->if_capenable = IFCAP_VLAN_MTU;
664 ifp->if_start = epair_start;
665 ifp->if_ioctl = epair_ioctl;
666 ifp->if_init = epair_init;
667 if_setsendqlen(ifp, ifqmaxlen);
668 if_setsendqready(ifp);
669 /* We need to play some tricks here for the second interface. */
670 strlcpy(name, epairname, len);
672 /* Correctly set the name for the cloner list. */
673 strlcpy(name, scb->ifp->if_xname, len);
674 epair_clone_add(ifc, scb);
676 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
677 ifp->if_transmit = epair_transmit;
680 * Restore name to <n>a as the ifp for this will go into the
681 * cloner list for the initial call.
683 strlcpy(name, sca->ifp->if_xname, len);
685 /* Tell the world, that we are ready to rock. */
686 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
687 if_link_state_change(sca->ifp, LINK_STATE_UP);
688 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
689 if_link_state_change(scb->ifp, LINK_STATE_UP);
695 epair_drain_rings(struct epair_softc *sc)
700 for (ridx = 0; ridx < 2; ridx++) {
701 for (int i = 0; i < sc->num_queues; i++) {
702 struct epair_queue *q = &sc->queues[i];
704 m = buf_ring_dequeue_sc(q->rxring[ridx]);
714 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
717 struct epair_softc *sca, *scb;
721 * In case we called into if_clone_destroyif() ourselves
722 * again to remove the second interface, the softc will be
723 * NULL. In that case so not do anything but return success.
725 if (ifp->if_softc == NULL)
728 unit = ifp->if_dunit;
731 scb = oifp->if_softc;
733 /* Frist get the interfaces down and detached. */
734 if_link_state_change(ifp, LINK_STATE_DOWN);
735 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
736 if_link_state_change(oifp, LINK_STATE_DOWN);
737 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
740 ether_ifdetach(oifp);
742 /* Third free any queued packets and all the resources. */
743 CURVNET_SET_QUIET(oifp->if_vnet);
744 epair_drain_rings(scb);
745 oifp->if_softc = NULL;
746 error = if_clone_destroyif(ifc, oifp);
748 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
751 ifmedia_removeall(&scb->media);
752 for (int i = 0; i < scb->num_queues; i++) {
753 struct epair_queue *q = &scb->queues[i];
754 buf_ring_free(q->rxring[0], M_EPAIR);
755 buf_ring_free(q->rxring[1], M_EPAIR);
757 free(scb->queues, M_EPAIR);
761 epair_drain_rings(sca);
763 ifmedia_removeall(&sca->media);
764 for (int i = 0; i < sca->num_queues; i++) {
765 struct epair_queue *q = &sca->queues[i];
766 buf_ring_free(q->rxring[0], M_EPAIR);
767 buf_ring_free(q->rxring[1], M_EPAIR);
769 free(sca->queues, M_EPAIR);
772 /* Last free the cloner unit. */
773 ifc_free_unit(ifc, unit);
779 vnet_epair_init(const void *unused __unused)
782 V_epair_cloner = if_clone_advanced(epairname, 0,
783 epair_clone_match, epair_clone_create, epair_clone_destroy);
785 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
786 vnet_epair_init, NULL);
789 vnet_epair_uninit(const void *unused __unused)
792 if_clone_detach(V_epair_cloner);
794 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
795 vnet_epair_uninit, NULL);
801 epair_tasks.tasks = 0;
809 /* Pin to this CPU so we get appropriate NUMA allocations. */
810 thread_lock(curthread);
811 sched_bind(curthread, cpu);
812 thread_unlock(curthread);
814 snprintf(name, sizeof(name), "epair_task_%d", cpu);
816 epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK,
817 taskqueue_thread_enqueue,
818 &epair_tasks.tq[cpu]);
819 CPU_SETOF(cpu, &cpu_mask);
820 taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET,
821 &cpu_mask, "%s", name);
825 thread_lock(curthread);
826 sched_unbind(curthread);
827 thread_unlock(curthread);
829 snprintf(name, sizeof(name), "epair_task");
831 epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK,
832 taskqueue_thread_enqueue,
834 taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name);
836 epair_tasks.tasks = 1;
846 for (int i = 0; i < epair_tasks.tasks; i++) {
847 taskqueue_drain_all(epair_tasks.tq[i]);
848 taskqueue_free(epair_tasks.tq[i]);
853 epair_modevent(module_t mod, int type, void *data)
860 ret = epair_mod_init();
864 printf("%s: %s initialized.\n", __func__, epairname);
868 EPAIR_LOCK_DESTROY();
870 printf("%s: %s unloaded.\n", __func__, epairname);
878 static moduledata_t epair_mod = {
884 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
885 MODULE_VERSION(if_epair, 3);