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>
27 #include <linux/net_tstamp.h>
29 #define DRV_NAME "veth"
30 #define DRV_VERSION "1.0"
32 #define VETH_XDP_FLAG BIT(0)
33 #define VETH_RING_SIZE 256
34 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
36 /* Separating two types of XDP xmit */
37 #define VETH_XDP_TX BIT(0)
38 #define VETH_XDP_REDIR BIT(1)
40 struct veth_rq_stats {
44 struct u64_stats_sync syncp;
48 struct napi_struct xdp_napi;
49 struct net_device *dev;
50 struct bpf_prog __rcu *xdp_prog;
51 struct xdp_mem_info xdp_mem;
52 struct veth_rq_stats stats;
53 bool rx_notify_masked;
54 struct ptr_ring xdp_ring;
55 struct xdp_rxq_info xdp_rxq;
59 struct net_device __rcu *peer;
61 struct bpf_prog *_xdp_prog;
63 unsigned int requested_headroom;
70 struct veth_q_stat_desc {
71 char desc[ETH_GSTRING_LEN];
75 #define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m)
77 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
78 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
79 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
80 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
83 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
86 const char string[ETH_GSTRING_LEN];
87 } ethtool_stats_keys[] = {
91 static int veth_get_link_ksettings(struct net_device *dev,
92 struct ethtool_link_ksettings *cmd)
94 cmd->base.speed = SPEED_10000;
95 cmd->base.duplex = DUPLEX_FULL;
96 cmd->base.port = PORT_TP;
97 cmd->base.autoneg = AUTONEG_DISABLE;
101 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
103 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
104 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
107 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
109 char *p = (char *)buf;
114 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys));
115 p += sizeof(ethtool_stats_keys);
116 for (i = 0; i < dev->real_num_rx_queues; i++) {
117 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
118 snprintf(p, ETH_GSTRING_LEN,
120 i, veth_rq_stats_desc[j].desc);
121 p += ETH_GSTRING_LEN;
128 static int veth_get_sset_count(struct net_device *dev, int sset)
132 return ARRAY_SIZE(ethtool_stats_keys) +
133 VETH_RQ_STATS_LEN * dev->real_num_rx_queues;
139 static void veth_get_ethtool_stats(struct net_device *dev,
140 struct ethtool_stats *stats, u64 *data)
142 struct veth_priv *priv = netdev_priv(dev);
143 struct net_device *peer = rtnl_dereference(priv->peer);
146 data[0] = peer ? peer->ifindex : 0;
148 for (i = 0; i < dev->real_num_rx_queues; i++) {
149 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
150 const void *stats_base = (void *)rq_stats;
155 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
156 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
157 offset = veth_rq_stats_desc[j].offset;
158 data[idx + j] = *(u64 *)(stats_base + offset);
160 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
161 idx += VETH_RQ_STATS_LEN;
165 static int veth_get_ts_info(struct net_device *dev,
166 struct ethtool_ts_info *info)
168 info->so_timestamping =
169 SOF_TIMESTAMPING_TX_SOFTWARE |
170 SOF_TIMESTAMPING_RX_SOFTWARE |
171 SOF_TIMESTAMPING_SOFTWARE;
172 info->phc_index = -1;
177 static const struct ethtool_ops veth_ethtool_ops = {
178 .get_drvinfo = veth_get_drvinfo,
179 .get_link = ethtool_op_get_link,
180 .get_strings = veth_get_strings,
181 .get_sset_count = veth_get_sset_count,
182 .get_ethtool_stats = veth_get_ethtool_stats,
183 .get_link_ksettings = veth_get_link_ksettings,
184 .get_ts_info = veth_get_ts_info,
187 /* general routines */
189 static bool veth_is_xdp_frame(void *ptr)
191 return (unsigned long)ptr & VETH_XDP_FLAG;
194 static void *veth_ptr_to_xdp(void *ptr)
196 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
199 static void *veth_xdp_to_ptr(void *ptr)
201 return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
204 static void veth_ptr_free(void *ptr)
206 if (veth_is_xdp_frame(ptr))
207 xdp_return_frame(veth_ptr_to_xdp(ptr));
212 static void __veth_xdp_flush(struct veth_rq *rq)
214 /* Write ptr_ring before reading rx_notify_masked */
216 if (!rq->rx_notify_masked) {
217 rq->rx_notify_masked = true;
218 napi_schedule(&rq->xdp_napi);
222 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
224 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
225 dev_kfree_skb_any(skb);
229 return NET_RX_SUCCESS;
232 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
233 struct veth_rq *rq, bool xdp)
235 return __dev_forward_skb(dev, skb) ?: xdp ?
236 veth_xdp_rx(rq, skb) :
240 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
242 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
243 struct veth_rq *rq = NULL;
244 struct net_device *rcv;
245 int length = skb->len;
246 bool rcv_xdp = false;
250 rcv = rcu_dereference(priv->peer);
251 if (unlikely(!rcv)) {
256 rcv_priv = netdev_priv(rcv);
257 rxq = skb_get_queue_mapping(skb);
258 if (rxq < rcv->real_num_rx_queues) {
259 rq = &rcv_priv->rq[rxq];
260 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
262 skb_record_rx_queue(skb, rxq);
265 skb_tx_timestamp(skb);
266 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
268 struct pcpu_lstats *stats = this_cpu_ptr(dev->lstats);
270 u64_stats_update_begin(&stats->syncp);
271 stats->bytes += length;
273 u64_stats_update_end(&stats->syncp);
277 atomic64_inc(&priv->dropped);
281 __veth_xdp_flush(rq);
288 static u64 veth_stats_tx(struct pcpu_lstats *result, struct net_device *dev)
290 struct veth_priv *priv = netdev_priv(dev);
295 for_each_possible_cpu(cpu) {
296 struct pcpu_lstats *stats = per_cpu_ptr(dev->lstats, cpu);
301 start = u64_stats_fetch_begin_irq(&stats->syncp);
302 packets = stats->packets;
303 bytes = stats->bytes;
304 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
305 result->packets += packets;
306 result->bytes += bytes;
308 return atomic64_read(&priv->dropped);
311 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
313 struct veth_priv *priv = netdev_priv(dev);
316 result->xdp_packets = 0;
317 result->xdp_bytes = 0;
318 result->xdp_drops = 0;
319 for (i = 0; i < dev->num_rx_queues; i++) {
320 struct veth_rq_stats *stats = &priv->rq[i].stats;
321 u64 packets, bytes, drops;
325 start = u64_stats_fetch_begin_irq(&stats->syncp);
326 packets = stats->xdp_packets;
327 bytes = stats->xdp_bytes;
328 drops = stats->xdp_drops;
329 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
330 result->xdp_packets += packets;
331 result->xdp_bytes += bytes;
332 result->xdp_drops += drops;
336 static void veth_get_stats64(struct net_device *dev,
337 struct rtnl_link_stats64 *tot)
339 struct veth_priv *priv = netdev_priv(dev);
340 struct net_device *peer;
341 struct veth_rq_stats rx;
342 struct pcpu_lstats tx;
344 tot->tx_dropped = veth_stats_tx(&tx, dev);
345 tot->tx_bytes = tx.bytes;
346 tot->tx_packets = tx.packets;
348 veth_stats_rx(&rx, dev);
349 tot->rx_dropped = rx.xdp_drops;
350 tot->rx_bytes = rx.xdp_bytes;
351 tot->rx_packets = rx.xdp_packets;
354 peer = rcu_dereference(priv->peer);
356 tot->rx_dropped += veth_stats_tx(&tx, peer);
357 tot->rx_bytes += tx.bytes;
358 tot->rx_packets += tx.packets;
360 veth_stats_rx(&rx, peer);
361 tot->tx_bytes += rx.xdp_bytes;
362 tot->tx_packets += rx.xdp_packets;
367 /* fake multicast ability */
368 static void veth_set_multicast_list(struct net_device *dev)
372 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
378 buflen = SKB_DATA_ALIGN(headroom + len) +
379 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
381 skb = build_skb(head, buflen);
385 skb_reserve(skb, headroom);
391 static int veth_select_rxq(struct net_device *dev)
393 return smp_processor_id() % dev->real_num_rx_queues;
396 static int veth_xdp_xmit(struct net_device *dev, int n,
397 struct xdp_frame **frames, u32 flags)
399 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
400 struct net_device *rcv;
401 int i, ret, drops = n;
402 unsigned int max_len;
405 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
410 rcv = rcu_dereference(priv->peer);
411 if (unlikely(!rcv)) {
416 rcv_priv = netdev_priv(rcv);
417 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
418 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
419 * side. This means an XDP program is loaded on the peer and the peer
422 if (!rcu_access_pointer(rq->xdp_prog)) {
428 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
430 spin_lock(&rq->xdp_ring.producer_lock);
431 for (i = 0; i < n; i++) {
432 struct xdp_frame *frame = frames[i];
433 void *ptr = veth_xdp_to_ptr(frame);
435 if (unlikely(frame->len > max_len ||
436 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
437 xdp_return_frame_rx_napi(frame);
441 spin_unlock(&rq->xdp_ring.producer_lock);
443 if (flags & XDP_XMIT_FLUSH)
444 __veth_xdp_flush(rq);
451 atomic64_add(drops, &priv->dropped);
456 static void veth_xdp_flush(struct net_device *dev)
458 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
459 struct net_device *rcv;
463 rcv = rcu_dereference(priv->peer);
467 rcv_priv = netdev_priv(rcv);
468 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
469 /* xdp_ring is initialized on receive side? */
470 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
473 __veth_xdp_flush(rq);
478 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
480 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
482 if (unlikely(!frame))
485 return veth_xdp_xmit(dev, 1, &frame, 0);
488 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
489 struct xdp_frame *frame,
490 unsigned int *xdp_xmit)
492 void *hard_start = frame->data - frame->headroom;
493 void *head = hard_start - sizeof(struct xdp_frame);
494 int len = frame->len, delta = 0;
495 struct xdp_frame orig_frame;
496 struct bpf_prog *xdp_prog;
497 unsigned int headroom;
501 xdp_prog = rcu_dereference(rq->xdp_prog);
502 if (likely(xdp_prog)) {
506 xdp.data_hard_start = hard_start;
507 xdp.data = frame->data;
508 xdp.data_end = frame->data + frame->len;
509 xdp.data_meta = frame->data - frame->metasize;
510 xdp.rxq = &rq->xdp_rxq;
512 act = bpf_prog_run_xdp(xdp_prog, &xdp);
516 delta = frame->data - xdp.data;
517 len = xdp.data_end - xdp.data;
521 xdp.data_hard_start = head;
522 xdp.rxq->mem = frame->mem;
523 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
524 trace_xdp_exception(rq->dev, xdp_prog, act);
528 *xdp_xmit |= VETH_XDP_TX;
533 xdp.data_hard_start = head;
534 xdp.rxq->mem = frame->mem;
535 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
539 *xdp_xmit |= VETH_XDP_REDIR;
543 bpf_warn_invalid_xdp_action(act);
546 trace_xdp_exception(rq->dev, xdp_prog, act);
554 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
555 skb = veth_build_skb(head, headroom, len, 0);
557 xdp_return_frame(frame);
561 xdp_scrub_frame(frame);
562 skb->protocol = eth_type_trans(skb, rq->dev);
567 xdp_return_frame(frame);
572 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
573 unsigned int *xdp_xmit)
575 u32 pktlen, headroom, act, metalen;
576 void *orig_data, *orig_data_end;
577 struct bpf_prog *xdp_prog;
578 int mac_len, delta, off;
584 xdp_prog = rcu_dereference(rq->xdp_prog);
585 if (unlikely(!xdp_prog)) {
590 mac_len = skb->data - skb_mac_header(skb);
591 pktlen = skb->len + mac_len;
592 headroom = skb_headroom(skb) - mac_len;
594 if (skb_shared(skb) || skb_head_is_locked(skb) ||
595 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
596 struct sk_buff *nskb;
601 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
602 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
603 if (size > PAGE_SIZE)
606 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
610 head = page_address(page);
611 start = head + VETH_XDP_HEADROOM;
612 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
613 page_frag_free(head);
617 nskb = veth_build_skb(head,
618 VETH_XDP_HEADROOM + mac_len, skb->len,
621 page_frag_free(head);
625 skb_copy_header(nskb, skb);
626 head_off = skb_headroom(nskb) - skb_headroom(skb);
627 skb_headers_offset_update(nskb, head_off);
632 xdp.data_hard_start = skb->head;
633 xdp.data = skb_mac_header(skb);
634 xdp.data_end = xdp.data + pktlen;
635 xdp.data_meta = xdp.data;
636 xdp.rxq = &rq->xdp_rxq;
637 orig_data = xdp.data;
638 orig_data_end = xdp.data_end;
640 act = bpf_prog_run_xdp(xdp_prog, &xdp);
646 get_page(virt_to_page(xdp.data));
648 xdp.rxq->mem = rq->xdp_mem;
649 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
650 trace_xdp_exception(rq->dev, xdp_prog, act);
653 *xdp_xmit |= VETH_XDP_TX;
657 get_page(virt_to_page(xdp.data));
659 xdp.rxq->mem = rq->xdp_mem;
660 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
662 *xdp_xmit |= VETH_XDP_REDIR;
666 bpf_warn_invalid_xdp_action(act);
669 trace_xdp_exception(rq->dev, xdp_prog, act);
676 delta = orig_data - xdp.data;
677 off = mac_len + delta;
679 __skb_push(skb, off);
681 __skb_pull(skb, -off);
682 skb->mac_header -= delta;
683 off = xdp.data_end - orig_data_end;
686 skb->protocol = eth_type_trans(skb, rq->dev);
688 metalen = xdp.data - xdp.data_meta;
690 skb_metadata_set(skb, metalen);
699 page_frag_free(xdp.data);
704 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit)
706 int i, done = 0, drops = 0, bytes = 0;
708 for (i = 0; i < budget; i++) {
709 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
710 unsigned int xdp_xmit_one = 0;
716 if (veth_is_xdp_frame(ptr)) {
717 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
720 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one);
724 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one);
726 *xdp_xmit |= xdp_xmit_one;
729 napi_gro_receive(&rq->xdp_napi, skb);
730 else if (!xdp_xmit_one)
736 u64_stats_update_begin(&rq->stats.syncp);
737 rq->stats.xdp_packets += done;
738 rq->stats.xdp_bytes += bytes;
739 rq->stats.xdp_drops += drops;
740 u64_stats_update_end(&rq->stats.syncp);
745 static int veth_poll(struct napi_struct *napi, int budget)
748 container_of(napi, struct veth_rq, xdp_napi);
749 unsigned int xdp_xmit = 0;
752 xdp_set_return_frame_no_direct();
753 done = veth_xdp_rcv(rq, budget, &xdp_xmit);
755 if (done < budget && napi_complete_done(napi, done)) {
756 /* Write rx_notify_masked before reading ptr_ring */
757 smp_store_mb(rq->rx_notify_masked, false);
758 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
759 rq->rx_notify_masked = true;
760 napi_schedule(&rq->xdp_napi);
764 if (xdp_xmit & VETH_XDP_TX)
765 veth_xdp_flush(rq->dev);
766 if (xdp_xmit & VETH_XDP_REDIR)
768 xdp_clear_return_frame_no_direct();
773 static int veth_napi_add(struct net_device *dev)
775 struct veth_priv *priv = netdev_priv(dev);
778 for (i = 0; i < dev->real_num_rx_queues; i++) {
779 struct veth_rq *rq = &priv->rq[i];
781 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
786 for (i = 0; i < dev->real_num_rx_queues; i++) {
787 struct veth_rq *rq = &priv->rq[i];
789 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
790 napi_enable(&rq->xdp_napi);
795 for (i--; i >= 0; i--)
796 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
801 static void veth_napi_del(struct net_device *dev)
803 struct veth_priv *priv = netdev_priv(dev);
806 for (i = 0; i < dev->real_num_rx_queues; i++) {
807 struct veth_rq *rq = &priv->rq[i];
809 napi_disable(&rq->xdp_napi);
810 napi_hash_del(&rq->xdp_napi);
814 for (i = 0; i < dev->real_num_rx_queues; i++) {
815 struct veth_rq *rq = &priv->rq[i];
817 netif_napi_del(&rq->xdp_napi);
818 rq->rx_notify_masked = false;
819 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
823 static int veth_enable_xdp(struct net_device *dev)
825 struct veth_priv *priv = netdev_priv(dev);
828 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
829 for (i = 0; i < dev->real_num_rx_queues; i++) {
830 struct veth_rq *rq = &priv->rq[i];
832 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
836 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
837 MEM_TYPE_PAGE_SHARED,
842 /* Save original mem info as it can be overwritten */
843 rq->xdp_mem = rq->xdp_rxq.mem;
846 err = veth_napi_add(dev);
851 for (i = 0; i < dev->real_num_rx_queues; i++)
852 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
856 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
858 for (i--; i >= 0; i--)
859 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
864 static void veth_disable_xdp(struct net_device *dev)
866 struct veth_priv *priv = netdev_priv(dev);
869 for (i = 0; i < dev->real_num_rx_queues; i++)
870 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
872 for (i = 0; i < dev->real_num_rx_queues; i++) {
873 struct veth_rq *rq = &priv->rq[i];
875 rq->xdp_rxq.mem = rq->xdp_mem;
876 xdp_rxq_info_unreg(&rq->xdp_rxq);
880 static int veth_open(struct net_device *dev)
882 struct veth_priv *priv = netdev_priv(dev);
883 struct net_device *peer = rtnl_dereference(priv->peer);
889 if (priv->_xdp_prog) {
890 err = veth_enable_xdp(dev);
895 if (peer->flags & IFF_UP) {
896 netif_carrier_on(dev);
897 netif_carrier_on(peer);
903 static int veth_close(struct net_device *dev)
905 struct veth_priv *priv = netdev_priv(dev);
906 struct net_device *peer = rtnl_dereference(priv->peer);
908 netif_carrier_off(dev);
910 netif_carrier_off(peer);
913 veth_disable_xdp(dev);
918 static int is_valid_veth_mtu(int mtu)
920 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
923 static int veth_alloc_queues(struct net_device *dev)
925 struct veth_priv *priv = netdev_priv(dev);
928 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
932 for (i = 0; i < dev->num_rx_queues; i++) {
933 priv->rq[i].dev = dev;
934 u64_stats_init(&priv->rq[i].stats.syncp);
940 static void veth_free_queues(struct net_device *dev)
942 struct veth_priv *priv = netdev_priv(dev);
947 static int veth_dev_init(struct net_device *dev)
951 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
955 err = veth_alloc_queues(dev);
957 free_percpu(dev->lstats);
964 static void veth_dev_free(struct net_device *dev)
966 veth_free_queues(dev);
967 free_percpu(dev->lstats);
970 #ifdef CONFIG_NET_POLL_CONTROLLER
971 static void veth_poll_controller(struct net_device *dev)
973 /* veth only receives frames when its peer sends one
974 * Since it has nothing to do with disabling irqs, we are guaranteed
975 * never to have pending data when we poll for it so
976 * there is nothing to do here.
978 * We need this though so netpoll recognizes us as an interface that
979 * supports polling, which enables bridge devices in virt setups to
980 * still use netconsole
983 #endif /* CONFIG_NET_POLL_CONTROLLER */
985 static int veth_get_iflink(const struct net_device *dev)
987 struct veth_priv *priv = netdev_priv(dev);
988 struct net_device *peer;
992 peer = rcu_dereference(priv->peer);
993 iflink = peer ? peer->ifindex : 0;
999 static netdev_features_t veth_fix_features(struct net_device *dev,
1000 netdev_features_t features)
1002 struct veth_priv *priv = netdev_priv(dev);
1003 struct net_device *peer;
1005 peer = rtnl_dereference(priv->peer);
1007 struct veth_priv *peer_priv = netdev_priv(peer);
1009 if (peer_priv->_xdp_prog)
1010 features &= ~NETIF_F_GSO_SOFTWARE;
1016 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1018 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1019 struct net_device *peer;
1025 peer = rcu_dereference(priv->peer);
1026 if (unlikely(!peer))
1029 peer_priv = netdev_priv(peer);
1030 priv->requested_headroom = new_hr;
1031 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1032 dev->needed_headroom = new_hr;
1033 peer->needed_headroom = new_hr;
1039 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1040 struct netlink_ext_ack *extack)
1042 struct veth_priv *priv = netdev_priv(dev);
1043 struct bpf_prog *old_prog;
1044 struct net_device *peer;
1045 unsigned int max_mtu;
1048 old_prog = priv->_xdp_prog;
1049 priv->_xdp_prog = prog;
1050 peer = rtnl_dereference(priv->peer);
1054 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1059 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1060 peer->hard_header_len -
1061 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1062 if (peer->mtu > max_mtu) {
1063 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1068 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1069 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1074 if (dev->flags & IFF_UP) {
1075 err = veth_enable_xdp(dev);
1077 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1083 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1084 peer->max_mtu = max_mtu;
1090 if (dev->flags & IFF_UP)
1091 veth_disable_xdp(dev);
1094 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1095 peer->max_mtu = ETH_MAX_MTU;
1098 bpf_prog_put(old_prog);
1101 if ((!!old_prog ^ !!prog) && peer)
1102 netdev_update_features(peer);
1106 priv->_xdp_prog = old_prog;
1111 static u32 veth_xdp_query(struct net_device *dev)
1113 struct veth_priv *priv = netdev_priv(dev);
1114 const struct bpf_prog *xdp_prog;
1116 xdp_prog = priv->_xdp_prog;
1118 return xdp_prog->aux->id;
1123 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1125 switch (xdp->command) {
1126 case XDP_SETUP_PROG:
1127 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1128 case XDP_QUERY_PROG:
1129 xdp->prog_id = veth_xdp_query(dev);
1136 static const struct net_device_ops veth_netdev_ops = {
1137 .ndo_init = veth_dev_init,
1138 .ndo_open = veth_open,
1139 .ndo_stop = veth_close,
1140 .ndo_start_xmit = veth_xmit,
1141 .ndo_get_stats64 = veth_get_stats64,
1142 .ndo_set_rx_mode = veth_set_multicast_list,
1143 .ndo_set_mac_address = eth_mac_addr,
1144 #ifdef CONFIG_NET_POLL_CONTROLLER
1145 .ndo_poll_controller = veth_poll_controller,
1147 .ndo_get_iflink = veth_get_iflink,
1148 .ndo_fix_features = veth_fix_features,
1149 .ndo_features_check = passthru_features_check,
1150 .ndo_set_rx_headroom = veth_set_rx_headroom,
1151 .ndo_bpf = veth_xdp,
1152 .ndo_xdp_xmit = veth_xdp_xmit,
1155 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1156 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1157 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1158 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1159 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1161 static void veth_setup(struct net_device *dev)
1165 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1166 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1167 dev->priv_flags |= IFF_NO_QUEUE;
1168 dev->priv_flags |= IFF_PHONY_HEADROOM;
1170 dev->netdev_ops = &veth_netdev_ops;
1171 dev->ethtool_ops = &veth_ethtool_ops;
1172 dev->features |= NETIF_F_LLTX;
1173 dev->features |= VETH_FEATURES;
1174 dev->vlan_features = dev->features &
1175 ~(NETIF_F_HW_VLAN_CTAG_TX |
1176 NETIF_F_HW_VLAN_STAG_TX |
1177 NETIF_F_HW_VLAN_CTAG_RX |
1178 NETIF_F_HW_VLAN_STAG_RX);
1179 dev->needs_free_netdev = true;
1180 dev->priv_destructor = veth_dev_free;
1181 dev->max_mtu = ETH_MAX_MTU;
1183 dev->hw_features = VETH_FEATURES;
1184 dev->hw_enc_features = VETH_FEATURES;
1185 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1192 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1193 struct netlink_ext_ack *extack)
1195 if (tb[IFLA_ADDRESS]) {
1196 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1198 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1199 return -EADDRNOTAVAIL;
1202 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1208 static struct rtnl_link_ops veth_link_ops;
1210 static int veth_newlink(struct net *src_net, struct net_device *dev,
1211 struct nlattr *tb[], struct nlattr *data[],
1212 struct netlink_ext_ack *extack)
1215 struct net_device *peer;
1216 struct veth_priv *priv;
1217 char ifname[IFNAMSIZ];
1218 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1219 unsigned char name_assign_type;
1220 struct ifinfomsg *ifmp;
1224 * create and register peer first
1226 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1227 struct nlattr *nla_peer;
1229 nla_peer = data[VETH_INFO_PEER];
1230 ifmp = nla_data(nla_peer);
1231 err = rtnl_nla_parse_ifla(peer_tb,
1232 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1233 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1238 err = veth_validate(peer_tb, NULL, extack);
1248 if (ifmp && tbp[IFLA_IFNAME]) {
1249 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1250 name_assign_type = NET_NAME_USER;
1252 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1253 name_assign_type = NET_NAME_ENUM;
1256 net = rtnl_link_get_net(src_net, tbp);
1258 return PTR_ERR(net);
1260 peer = rtnl_create_link(net, ifname, name_assign_type,
1261 &veth_link_ops, tbp, extack);
1264 return PTR_ERR(peer);
1267 if (!ifmp || !tbp[IFLA_ADDRESS])
1268 eth_hw_addr_random(peer);
1270 if (ifmp && (dev->ifindex != 0))
1271 peer->ifindex = ifmp->ifi_index;
1273 peer->gso_max_size = dev->gso_max_size;
1274 peer->gso_max_segs = dev->gso_max_segs;
1276 err = register_netdevice(peer);
1280 goto err_register_peer;
1282 netif_carrier_off(peer);
1284 err = rtnl_configure_link(peer, ifmp);
1286 goto err_configure_peer;
1291 * note, that since we've registered new device the dev's name
1292 * should be re-allocated
1295 if (tb[IFLA_ADDRESS] == NULL)
1296 eth_hw_addr_random(dev);
1298 if (tb[IFLA_IFNAME])
1299 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1301 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1303 err = register_netdevice(dev);
1305 goto err_register_dev;
1307 netif_carrier_off(dev);
1310 * tie the deviced together
1313 priv = netdev_priv(dev);
1314 rcu_assign_pointer(priv->peer, peer);
1316 priv = netdev_priv(peer);
1317 rcu_assign_pointer(priv->peer, dev);
1324 unregister_netdevice(peer);
1332 static void veth_dellink(struct net_device *dev, struct list_head *head)
1334 struct veth_priv *priv;
1335 struct net_device *peer;
1337 priv = netdev_priv(dev);
1338 peer = rtnl_dereference(priv->peer);
1340 /* Note : dellink() is called from default_device_exit_batch(),
1341 * before a rcu_synchronize() point. The devices are guaranteed
1342 * not being freed before one RCU grace period.
1344 RCU_INIT_POINTER(priv->peer, NULL);
1345 unregister_netdevice_queue(dev, head);
1348 priv = netdev_priv(peer);
1349 RCU_INIT_POINTER(priv->peer, NULL);
1350 unregister_netdevice_queue(peer, head);
1354 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1355 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1358 static struct net *veth_get_link_net(const struct net_device *dev)
1360 struct veth_priv *priv = netdev_priv(dev);
1361 struct net_device *peer = rtnl_dereference(priv->peer);
1363 return peer ? dev_net(peer) : dev_net(dev);
1366 static struct rtnl_link_ops veth_link_ops = {
1368 .priv_size = sizeof(struct veth_priv),
1369 .setup = veth_setup,
1370 .validate = veth_validate,
1371 .newlink = veth_newlink,
1372 .dellink = veth_dellink,
1373 .policy = veth_policy,
1374 .maxtype = VETH_INFO_MAX,
1375 .get_link_net = veth_get_link_net,
1382 static __init int veth_init(void)
1384 return rtnl_link_register(&veth_link_ops);
1387 static __exit void veth_exit(void)
1389 rtnl_link_unregister(&veth_link_ops);
1392 module_init(veth_init);
1393 module_exit(veth_exit);
1395 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1396 MODULE_LICENSE("GPL v2");
1397 MODULE_ALIAS_RTNL_LINK(DRV_NAME);