4 * Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
6 * Author: Pavel Emelianov <xemul@openvz.org>
7 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
11 #include <linux/netdevice.h>
12 #include <linux/slab.h>
13 #include <linux/ethtool.h>
14 #include <linux/etherdevice.h>
15 #include <linux/u64_stats_sync.h>
17 #include <net/rtnetlink.h>
21 #include <linux/veth.h>
22 #include <linux/module.h>
23 #include <linux/bpf.h>
24 #include <linux/filter.h>
25 #include <linux/ptr_ring.h>
26 #include <linux/bpf_trace.h>
28 #define DRV_NAME "veth"
29 #define DRV_VERSION "1.0"
31 #define VETH_XDP_FLAG BIT(0)
32 #define VETH_RING_SIZE 256
33 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
35 /* Separating two types of XDP xmit */
36 #define VETH_XDP_TX BIT(0)
37 #define VETH_XDP_REDIR BIT(1)
42 struct u64_stats_sync syncp;
46 struct napi_struct xdp_napi;
47 struct net_device *dev;
48 struct bpf_prog __rcu *xdp_prog;
49 struct xdp_mem_info xdp_mem;
50 bool rx_notify_masked;
51 struct ptr_ring xdp_ring;
52 struct xdp_rxq_info xdp_rxq;
56 struct net_device __rcu *peer;
58 struct bpf_prog *_xdp_prog;
60 unsigned int requested_headroom;
68 const char string[ETH_GSTRING_LEN];
69 } ethtool_stats_keys[] = {
73 static int veth_get_link_ksettings(struct net_device *dev,
74 struct ethtool_link_ksettings *cmd)
76 cmd->base.speed = SPEED_10000;
77 cmd->base.duplex = DUPLEX_FULL;
78 cmd->base.port = PORT_TP;
79 cmd->base.autoneg = AUTONEG_DISABLE;
83 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
85 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
86 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
89 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
93 memcpy(buf, ðtool_stats_keys, sizeof(ethtool_stats_keys));
98 static int veth_get_sset_count(struct net_device *dev, int sset)
102 return ARRAY_SIZE(ethtool_stats_keys);
108 static void veth_get_ethtool_stats(struct net_device *dev,
109 struct ethtool_stats *stats, u64 *data)
111 struct veth_priv *priv = netdev_priv(dev);
112 struct net_device *peer = rtnl_dereference(priv->peer);
114 data[0] = peer ? peer->ifindex : 0;
117 static const struct ethtool_ops veth_ethtool_ops = {
118 .get_drvinfo = veth_get_drvinfo,
119 .get_link = ethtool_op_get_link,
120 .get_strings = veth_get_strings,
121 .get_sset_count = veth_get_sset_count,
122 .get_ethtool_stats = veth_get_ethtool_stats,
123 .get_link_ksettings = veth_get_link_ksettings,
126 /* general routines */
128 static bool veth_is_xdp_frame(void *ptr)
130 return (unsigned long)ptr & VETH_XDP_FLAG;
133 static void *veth_ptr_to_xdp(void *ptr)
135 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
138 static void *veth_xdp_to_ptr(void *ptr)
140 return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
143 static void veth_ptr_free(void *ptr)
145 if (veth_is_xdp_frame(ptr))
146 xdp_return_frame(veth_ptr_to_xdp(ptr));
151 static void __veth_xdp_flush(struct veth_rq *rq)
153 /* Write ptr_ring before reading rx_notify_masked */
155 if (!rq->rx_notify_masked) {
156 rq->rx_notify_masked = true;
157 napi_schedule(&rq->xdp_napi);
161 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
163 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
164 dev_kfree_skb_any(skb);
168 return NET_RX_SUCCESS;
171 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
172 struct veth_rq *rq, bool xdp)
174 return __dev_forward_skb(dev, skb) ?: xdp ?
175 veth_xdp_rx(rq, skb) :
179 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
181 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
182 struct veth_rq *rq = NULL;
183 struct net_device *rcv;
184 int length = skb->len;
185 bool rcv_xdp = false;
189 rcv = rcu_dereference(priv->peer);
190 if (unlikely(!rcv)) {
195 rcv_priv = netdev_priv(rcv);
196 rxq = skb_get_queue_mapping(skb);
197 if (rxq < rcv->real_num_rx_queues) {
198 rq = &rcv_priv->rq[rxq];
199 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
200 skb_record_rx_queue(skb, rxq);
203 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
204 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
206 u64_stats_update_begin(&stats->syncp);
207 stats->bytes += length;
209 u64_stats_update_end(&stats->syncp);
212 atomic64_inc(&priv->dropped);
216 __veth_xdp_flush(rq);
223 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
225 struct veth_priv *priv = netdev_priv(dev);
230 for_each_possible_cpu(cpu) {
231 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
236 start = u64_stats_fetch_begin_irq(&stats->syncp);
237 packets = stats->packets;
238 bytes = stats->bytes;
239 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
240 result->packets += packets;
241 result->bytes += bytes;
243 return atomic64_read(&priv->dropped);
246 static void veth_get_stats64(struct net_device *dev,
247 struct rtnl_link_stats64 *tot)
249 struct veth_priv *priv = netdev_priv(dev);
250 struct net_device *peer;
251 struct pcpu_vstats one;
253 tot->tx_dropped = veth_stats_one(&one, dev);
254 tot->tx_bytes = one.bytes;
255 tot->tx_packets = one.packets;
258 peer = rcu_dereference(priv->peer);
260 tot->rx_dropped = veth_stats_one(&one, peer);
261 tot->rx_bytes = one.bytes;
262 tot->rx_packets = one.packets;
267 /* fake multicast ability */
268 static void veth_set_multicast_list(struct net_device *dev)
272 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
278 buflen = SKB_DATA_ALIGN(headroom + len) +
279 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
281 skb = build_skb(head, buflen);
285 skb_reserve(skb, headroom);
291 static int veth_select_rxq(struct net_device *dev)
293 return smp_processor_id() % dev->real_num_rx_queues;
296 static int veth_xdp_xmit(struct net_device *dev, int n,
297 struct xdp_frame **frames, u32 flags)
299 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
300 struct net_device *rcv;
301 unsigned int max_len;
305 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
308 rcv = rcu_dereference(priv->peer);
312 rcv_priv = netdev_priv(rcv);
313 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
314 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
315 * side. This means an XDP program is loaded on the peer and the peer
318 if (!rcu_access_pointer(rq->xdp_prog))
321 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
323 spin_lock(&rq->xdp_ring.producer_lock);
324 for (i = 0; i < n; i++) {
325 struct xdp_frame *frame = frames[i];
326 void *ptr = veth_xdp_to_ptr(frame);
328 if (unlikely(frame->len > max_len ||
329 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
330 xdp_return_frame_rx_napi(frame);
334 spin_unlock(&rq->xdp_ring.producer_lock);
336 if (flags & XDP_XMIT_FLUSH)
337 __veth_xdp_flush(rq);
342 static void veth_xdp_flush(struct net_device *dev)
344 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
345 struct net_device *rcv;
349 rcv = rcu_dereference(priv->peer);
353 rcv_priv = netdev_priv(rcv);
354 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
355 /* xdp_ring is initialized on receive side? */
356 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
359 __veth_xdp_flush(rq);
364 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
366 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
368 if (unlikely(!frame))
371 return veth_xdp_xmit(dev, 1, &frame, 0);
374 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
375 struct xdp_frame *frame,
376 unsigned int *xdp_xmit)
378 void *hard_start = frame->data - frame->headroom;
379 int len = frame->len, delta = 0;
380 struct xdp_frame orig_frame;
381 struct bpf_prog *xdp_prog;
382 unsigned int headroom;
385 /* bpf_xdp_adjust_head() assures BPF cannot access xdp_frame area */
386 hard_start -= sizeof(struct xdp_frame);
389 xdp_prog = rcu_dereference(rq->xdp_prog);
390 if (likely(xdp_prog)) {
394 xdp.data_hard_start = hard_start;
395 xdp.data = frame->data;
396 xdp.data_end = frame->data + frame->len;
397 xdp.data_meta = frame->data - frame->metasize;
398 xdp.rxq = &rq->xdp_rxq;
400 act = bpf_prog_run_xdp(xdp_prog, &xdp);
404 delta = frame->data - xdp.data;
405 len = xdp.data_end - xdp.data;
409 xdp.rxq->mem = frame->mem;
410 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
411 trace_xdp_exception(rq->dev, xdp_prog, act);
415 *xdp_xmit |= VETH_XDP_TX;
420 xdp.rxq->mem = frame->mem;
421 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
425 *xdp_xmit |= VETH_XDP_REDIR;
429 bpf_warn_invalid_xdp_action(act);
431 trace_xdp_exception(rq->dev, xdp_prog, act);
438 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
439 skb = veth_build_skb(hard_start, headroom, len, 0);
441 xdp_return_frame(frame);
445 xdp_scrub_frame(frame);
446 skb->protocol = eth_type_trans(skb, rq->dev);
451 xdp_return_frame(frame);
456 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
457 unsigned int *xdp_xmit)
459 u32 pktlen, headroom, act, metalen;
460 void *orig_data, *orig_data_end;
461 struct bpf_prog *xdp_prog;
462 int mac_len, delta, off;
468 xdp_prog = rcu_dereference(rq->xdp_prog);
469 if (unlikely(!xdp_prog)) {
474 mac_len = skb->data - skb_mac_header(skb);
475 pktlen = skb->len + mac_len;
476 headroom = skb_headroom(skb) - mac_len;
478 if (skb_shared(skb) || skb_head_is_locked(skb) ||
479 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
480 struct sk_buff *nskb;
485 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
486 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
487 if (size > PAGE_SIZE)
490 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
494 head = page_address(page);
495 start = head + VETH_XDP_HEADROOM;
496 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
497 page_frag_free(head);
501 nskb = veth_build_skb(head,
502 VETH_XDP_HEADROOM + mac_len, skb->len,
505 page_frag_free(head);
509 skb_copy_header(nskb, skb);
510 head_off = skb_headroom(nskb) - skb_headroom(skb);
511 skb_headers_offset_update(nskb, head_off);
516 xdp.data_hard_start = skb->head;
517 xdp.data = skb_mac_header(skb);
518 xdp.data_end = xdp.data + pktlen;
519 xdp.data_meta = xdp.data;
520 xdp.rxq = &rq->xdp_rxq;
521 orig_data = xdp.data;
522 orig_data_end = xdp.data_end;
524 act = bpf_prog_run_xdp(xdp_prog, &xdp);
530 get_page(virt_to_page(xdp.data));
532 xdp.rxq->mem = rq->xdp_mem;
533 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
534 trace_xdp_exception(rq->dev, xdp_prog, act);
537 *xdp_xmit |= VETH_XDP_TX;
541 get_page(virt_to_page(xdp.data));
543 xdp.rxq->mem = rq->xdp_mem;
544 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
546 *xdp_xmit |= VETH_XDP_REDIR;
550 bpf_warn_invalid_xdp_action(act);
552 trace_xdp_exception(rq->dev, xdp_prog, act);
558 delta = orig_data - xdp.data;
559 off = mac_len + delta;
561 __skb_push(skb, off);
563 __skb_pull(skb, -off);
564 skb->mac_header -= delta;
565 off = xdp.data_end - orig_data_end;
568 skb->protocol = eth_type_trans(skb, rq->dev);
570 metalen = xdp.data - xdp.data_meta;
572 skb_metadata_set(skb, metalen);
581 page_frag_free(xdp.data);
586 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit)
590 for (i = 0; i < budget; i++) {
591 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
597 if (veth_is_xdp_frame(ptr)) {
598 skb = veth_xdp_rcv_one(rq, veth_ptr_to_xdp(ptr),
601 skb = veth_xdp_rcv_skb(rq, ptr, xdp_xmit);
605 napi_gro_receive(&rq->xdp_napi, skb);
613 static int veth_poll(struct napi_struct *napi, int budget)
616 container_of(napi, struct veth_rq, xdp_napi);
617 unsigned int xdp_xmit = 0;
620 xdp_set_return_frame_no_direct();
621 done = veth_xdp_rcv(rq, budget, &xdp_xmit);
623 if (done < budget && napi_complete_done(napi, done)) {
624 /* Write rx_notify_masked before reading ptr_ring */
625 smp_store_mb(rq->rx_notify_masked, false);
626 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
627 rq->rx_notify_masked = true;
628 napi_schedule(&rq->xdp_napi);
632 if (xdp_xmit & VETH_XDP_TX)
633 veth_xdp_flush(rq->dev);
634 if (xdp_xmit & VETH_XDP_REDIR)
636 xdp_clear_return_frame_no_direct();
641 static int veth_napi_add(struct net_device *dev)
643 struct veth_priv *priv = netdev_priv(dev);
646 for (i = 0; i < dev->real_num_rx_queues; i++) {
647 struct veth_rq *rq = &priv->rq[i];
649 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
654 for (i = 0; i < dev->real_num_rx_queues; i++) {
655 struct veth_rq *rq = &priv->rq[i];
657 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
658 napi_enable(&rq->xdp_napi);
663 for (i--; i >= 0; i--)
664 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
669 static void veth_napi_del(struct net_device *dev)
671 struct veth_priv *priv = netdev_priv(dev);
674 for (i = 0; i < dev->real_num_rx_queues; i++) {
675 struct veth_rq *rq = &priv->rq[i];
677 napi_disable(&rq->xdp_napi);
678 napi_hash_del(&rq->xdp_napi);
682 for (i = 0; i < dev->real_num_rx_queues; i++) {
683 struct veth_rq *rq = &priv->rq[i];
685 netif_napi_del(&rq->xdp_napi);
686 rq->rx_notify_masked = false;
687 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
691 static int veth_enable_xdp(struct net_device *dev)
693 struct veth_priv *priv = netdev_priv(dev);
696 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
697 for (i = 0; i < dev->real_num_rx_queues; i++) {
698 struct veth_rq *rq = &priv->rq[i];
700 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
704 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
705 MEM_TYPE_PAGE_SHARED,
710 /* Save original mem info as it can be overwritten */
711 rq->xdp_mem = rq->xdp_rxq.mem;
714 err = veth_napi_add(dev);
719 for (i = 0; i < dev->real_num_rx_queues; i++)
720 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
724 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
726 for (i--; i >= 0; i--)
727 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
732 static void veth_disable_xdp(struct net_device *dev)
734 struct veth_priv *priv = netdev_priv(dev);
737 for (i = 0; i < dev->real_num_rx_queues; i++)
738 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
740 for (i = 0; i < dev->real_num_rx_queues; i++) {
741 struct veth_rq *rq = &priv->rq[i];
743 rq->xdp_rxq.mem = rq->xdp_mem;
744 xdp_rxq_info_unreg(&rq->xdp_rxq);
748 static int veth_open(struct net_device *dev)
750 struct veth_priv *priv = netdev_priv(dev);
751 struct net_device *peer = rtnl_dereference(priv->peer);
757 if (priv->_xdp_prog) {
758 err = veth_enable_xdp(dev);
763 if (peer->flags & IFF_UP) {
764 netif_carrier_on(dev);
765 netif_carrier_on(peer);
771 static int veth_close(struct net_device *dev)
773 struct veth_priv *priv = netdev_priv(dev);
774 struct net_device *peer = rtnl_dereference(priv->peer);
776 netif_carrier_off(dev);
778 netif_carrier_off(peer);
781 veth_disable_xdp(dev);
786 static int is_valid_veth_mtu(int mtu)
788 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
791 static int veth_alloc_queues(struct net_device *dev)
793 struct veth_priv *priv = netdev_priv(dev);
796 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
800 for (i = 0; i < dev->num_rx_queues; i++)
801 priv->rq[i].dev = dev;
806 static void veth_free_queues(struct net_device *dev)
808 struct veth_priv *priv = netdev_priv(dev);
813 static int veth_dev_init(struct net_device *dev)
817 dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
821 err = veth_alloc_queues(dev);
823 free_percpu(dev->vstats);
830 static void veth_dev_free(struct net_device *dev)
832 veth_free_queues(dev);
833 free_percpu(dev->vstats);
836 #ifdef CONFIG_NET_POLL_CONTROLLER
837 static void veth_poll_controller(struct net_device *dev)
839 /* veth only receives frames when its peer sends one
840 * Since it has nothing to do with disabling irqs, we are guaranteed
841 * never to have pending data when we poll for it so
842 * there is nothing to do here.
844 * We need this though so netpoll recognizes us as an interface that
845 * supports polling, which enables bridge devices in virt setups to
846 * still use netconsole
849 #endif /* CONFIG_NET_POLL_CONTROLLER */
851 static int veth_get_iflink(const struct net_device *dev)
853 struct veth_priv *priv = netdev_priv(dev);
854 struct net_device *peer;
858 peer = rcu_dereference(priv->peer);
859 iflink = peer ? peer->ifindex : 0;
865 static netdev_features_t veth_fix_features(struct net_device *dev,
866 netdev_features_t features)
868 struct veth_priv *priv = netdev_priv(dev);
869 struct net_device *peer;
871 peer = rtnl_dereference(priv->peer);
873 struct veth_priv *peer_priv = netdev_priv(peer);
875 if (peer_priv->_xdp_prog)
876 features &= ~NETIF_F_GSO_SOFTWARE;
882 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
884 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
885 struct net_device *peer;
891 peer = rcu_dereference(priv->peer);
895 peer_priv = netdev_priv(peer);
896 priv->requested_headroom = new_hr;
897 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
898 dev->needed_headroom = new_hr;
899 peer->needed_headroom = new_hr;
905 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
906 struct netlink_ext_ack *extack)
908 struct veth_priv *priv = netdev_priv(dev);
909 struct bpf_prog *old_prog;
910 struct net_device *peer;
911 unsigned int max_mtu;
914 old_prog = priv->_xdp_prog;
915 priv->_xdp_prog = prog;
916 peer = rtnl_dereference(priv->peer);
920 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
925 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
926 peer->hard_header_len -
927 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
928 if (peer->mtu > max_mtu) {
929 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
934 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
935 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
940 if (dev->flags & IFF_UP) {
941 err = veth_enable_xdp(dev);
943 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
949 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
950 peer->max_mtu = max_mtu;
956 if (dev->flags & IFF_UP)
957 veth_disable_xdp(dev);
960 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
961 peer->max_mtu = ETH_MAX_MTU;
964 bpf_prog_put(old_prog);
967 if ((!!old_prog ^ !!prog) && peer)
968 netdev_update_features(peer);
972 priv->_xdp_prog = old_prog;
977 static u32 veth_xdp_query(struct net_device *dev)
979 struct veth_priv *priv = netdev_priv(dev);
980 const struct bpf_prog *xdp_prog;
982 xdp_prog = priv->_xdp_prog;
984 return xdp_prog->aux->id;
989 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
991 switch (xdp->command) {
993 return veth_xdp_set(dev, xdp->prog, xdp->extack);
995 xdp->prog_id = veth_xdp_query(dev);
1002 static const struct net_device_ops veth_netdev_ops = {
1003 .ndo_init = veth_dev_init,
1004 .ndo_open = veth_open,
1005 .ndo_stop = veth_close,
1006 .ndo_start_xmit = veth_xmit,
1007 .ndo_get_stats64 = veth_get_stats64,
1008 .ndo_set_rx_mode = veth_set_multicast_list,
1009 .ndo_set_mac_address = eth_mac_addr,
1010 #ifdef CONFIG_NET_POLL_CONTROLLER
1011 .ndo_poll_controller = veth_poll_controller,
1013 .ndo_get_iflink = veth_get_iflink,
1014 .ndo_fix_features = veth_fix_features,
1015 .ndo_features_check = passthru_features_check,
1016 .ndo_set_rx_headroom = veth_set_rx_headroom,
1017 .ndo_bpf = veth_xdp,
1018 .ndo_xdp_xmit = veth_xdp_xmit,
1021 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1022 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1023 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1024 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1025 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1027 static void veth_setup(struct net_device *dev)
1031 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1032 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1033 dev->priv_flags |= IFF_NO_QUEUE;
1034 dev->priv_flags |= IFF_PHONY_HEADROOM;
1036 dev->netdev_ops = &veth_netdev_ops;
1037 dev->ethtool_ops = &veth_ethtool_ops;
1038 dev->features |= NETIF_F_LLTX;
1039 dev->features |= VETH_FEATURES;
1040 dev->vlan_features = dev->features &
1041 ~(NETIF_F_HW_VLAN_CTAG_TX |
1042 NETIF_F_HW_VLAN_STAG_TX |
1043 NETIF_F_HW_VLAN_CTAG_RX |
1044 NETIF_F_HW_VLAN_STAG_RX);
1045 dev->needs_free_netdev = true;
1046 dev->priv_destructor = veth_dev_free;
1047 dev->max_mtu = ETH_MAX_MTU;
1049 dev->hw_features = VETH_FEATURES;
1050 dev->hw_enc_features = VETH_FEATURES;
1051 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1058 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1059 struct netlink_ext_ack *extack)
1061 if (tb[IFLA_ADDRESS]) {
1062 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1064 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1065 return -EADDRNOTAVAIL;
1068 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1074 static struct rtnl_link_ops veth_link_ops;
1076 static int veth_newlink(struct net *src_net, struct net_device *dev,
1077 struct nlattr *tb[], struct nlattr *data[],
1078 struct netlink_ext_ack *extack)
1081 struct net_device *peer;
1082 struct veth_priv *priv;
1083 char ifname[IFNAMSIZ];
1084 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1085 unsigned char name_assign_type;
1086 struct ifinfomsg *ifmp;
1090 * create and register peer first
1092 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1093 struct nlattr *nla_peer;
1095 nla_peer = data[VETH_INFO_PEER];
1096 ifmp = nla_data(nla_peer);
1097 err = rtnl_nla_parse_ifla(peer_tb,
1098 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1099 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1104 err = veth_validate(peer_tb, NULL, extack);
1114 if (ifmp && tbp[IFLA_IFNAME]) {
1115 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1116 name_assign_type = NET_NAME_USER;
1118 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1119 name_assign_type = NET_NAME_ENUM;
1122 net = rtnl_link_get_net(src_net, tbp);
1124 return PTR_ERR(net);
1126 peer = rtnl_create_link(net, ifname, name_assign_type,
1127 &veth_link_ops, tbp);
1130 return PTR_ERR(peer);
1133 if (!ifmp || !tbp[IFLA_ADDRESS])
1134 eth_hw_addr_random(peer);
1136 if (ifmp && (dev->ifindex != 0))
1137 peer->ifindex = ifmp->ifi_index;
1139 peer->gso_max_size = dev->gso_max_size;
1140 peer->gso_max_segs = dev->gso_max_segs;
1142 err = register_netdevice(peer);
1146 goto err_register_peer;
1148 netif_carrier_off(peer);
1150 err = rtnl_configure_link(peer, ifmp);
1152 goto err_configure_peer;
1157 * note, that since we've registered new device the dev's name
1158 * should be re-allocated
1161 if (tb[IFLA_ADDRESS] == NULL)
1162 eth_hw_addr_random(dev);
1164 if (tb[IFLA_IFNAME])
1165 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1167 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1169 err = register_netdevice(dev);
1171 goto err_register_dev;
1173 netif_carrier_off(dev);
1176 * tie the deviced together
1179 priv = netdev_priv(dev);
1180 rcu_assign_pointer(priv->peer, peer);
1182 priv = netdev_priv(peer);
1183 rcu_assign_pointer(priv->peer, dev);
1190 unregister_netdevice(peer);
1198 static void veth_dellink(struct net_device *dev, struct list_head *head)
1200 struct veth_priv *priv;
1201 struct net_device *peer;
1203 priv = netdev_priv(dev);
1204 peer = rtnl_dereference(priv->peer);
1206 /* Note : dellink() is called from default_device_exit_batch(),
1207 * before a rcu_synchronize() point. The devices are guaranteed
1208 * not being freed before one RCU grace period.
1210 RCU_INIT_POINTER(priv->peer, NULL);
1211 unregister_netdevice_queue(dev, head);
1214 priv = netdev_priv(peer);
1215 RCU_INIT_POINTER(priv->peer, NULL);
1216 unregister_netdevice_queue(peer, head);
1220 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1221 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1224 static struct net *veth_get_link_net(const struct net_device *dev)
1226 struct veth_priv *priv = netdev_priv(dev);
1227 struct net_device *peer = rtnl_dereference(priv->peer);
1229 return peer ? dev_net(peer) : dev_net(dev);
1232 static struct rtnl_link_ops veth_link_ops = {
1234 .priv_size = sizeof(struct veth_priv),
1235 .setup = veth_setup,
1236 .validate = veth_validate,
1237 .newlink = veth_newlink,
1238 .dellink = veth_dellink,
1239 .policy = veth_policy,
1240 .maxtype = VETH_INFO_MAX,
1241 .get_link_net = veth_get_link_net,
1248 static __init int veth_init(void)
1250 return rtnl_link_register(&veth_link_ops);
1253 static __exit void veth_exit(void)
1255 rtnl_link_unregister(&veth_link_ops);
1258 module_init(veth_init);
1259 module_exit(veth_exit);
1261 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1262 MODULE_LICENSE("GPL v2");
1263 MODULE_ALIAS_RTNL_LINK(DRV_NAME);