1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
12 #include <linux/netdevice.h>
13 #include <linux/slab.h>
14 #include <linux/ethtool.h>
15 #include <linux/etherdevice.h>
16 #include <linux/u64_stats_sync.h>
18 #include <net/rtnetlink.h>
22 #include <linux/veth.h>
23 #include <linux/module.h>
24 #include <linux/bpf.h>
25 #include <linux/filter.h>
26 #include <linux/ptr_ring.h>
27 #include <linux/bpf_trace.h>
28 #include <linux/net_tstamp.h>
30 #define DRV_NAME "veth"
31 #define DRV_VERSION "1.0"
33 #define VETH_XDP_FLAG BIT(0)
34 #define VETH_RING_SIZE 256
35 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
37 /* Separating two types of XDP xmit */
38 #define VETH_XDP_TX BIT(0)
39 #define VETH_XDP_REDIR BIT(1)
41 #define VETH_XDP_TX_BULK_SIZE 16
43 struct veth_rq_stats {
47 struct u64_stats_sync syncp;
51 struct napi_struct xdp_napi;
52 struct net_device *dev;
53 struct bpf_prog __rcu *xdp_prog;
54 struct xdp_mem_info xdp_mem;
55 struct veth_rq_stats stats;
56 bool rx_notify_masked;
57 struct ptr_ring xdp_ring;
58 struct xdp_rxq_info xdp_rxq;
62 struct net_device __rcu *peer;
64 struct bpf_prog *_xdp_prog;
66 unsigned int requested_headroom;
69 struct veth_xdp_tx_bq {
70 struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
78 struct veth_q_stat_desc {
79 char desc[ETH_GSTRING_LEN];
83 #define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m)
85 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
86 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
87 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
88 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
91 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
94 const char string[ETH_GSTRING_LEN];
95 } ethtool_stats_keys[] = {
99 static int veth_get_link_ksettings(struct net_device *dev,
100 struct ethtool_link_ksettings *cmd)
102 cmd->base.speed = SPEED_10000;
103 cmd->base.duplex = DUPLEX_FULL;
104 cmd->base.port = PORT_TP;
105 cmd->base.autoneg = AUTONEG_DISABLE;
109 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
111 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
112 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
115 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
117 char *p = (char *)buf;
122 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys));
123 p += sizeof(ethtool_stats_keys);
124 for (i = 0; i < dev->real_num_rx_queues; i++) {
125 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
126 snprintf(p, ETH_GSTRING_LEN,
128 i, veth_rq_stats_desc[j].desc);
129 p += ETH_GSTRING_LEN;
136 static int veth_get_sset_count(struct net_device *dev, int sset)
140 return ARRAY_SIZE(ethtool_stats_keys) +
141 VETH_RQ_STATS_LEN * dev->real_num_rx_queues;
147 static void veth_get_ethtool_stats(struct net_device *dev,
148 struct ethtool_stats *stats, u64 *data)
150 struct veth_priv *priv = netdev_priv(dev);
151 struct net_device *peer = rtnl_dereference(priv->peer);
154 data[0] = peer ? peer->ifindex : 0;
156 for (i = 0; i < dev->real_num_rx_queues; i++) {
157 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
158 const void *stats_base = (void *)rq_stats;
163 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
164 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
165 offset = veth_rq_stats_desc[j].offset;
166 data[idx + j] = *(u64 *)(stats_base + offset);
168 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
169 idx += VETH_RQ_STATS_LEN;
173 static const struct ethtool_ops veth_ethtool_ops = {
174 .get_drvinfo = veth_get_drvinfo,
175 .get_link = ethtool_op_get_link,
176 .get_strings = veth_get_strings,
177 .get_sset_count = veth_get_sset_count,
178 .get_ethtool_stats = veth_get_ethtool_stats,
179 .get_link_ksettings = veth_get_link_ksettings,
180 .get_ts_info = ethtool_op_get_ts_info,
183 /* general routines */
185 static bool veth_is_xdp_frame(void *ptr)
187 return (unsigned long)ptr & VETH_XDP_FLAG;
190 static void *veth_ptr_to_xdp(void *ptr)
192 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
195 static void *veth_xdp_to_ptr(void *ptr)
197 return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
200 static void veth_ptr_free(void *ptr)
202 if (veth_is_xdp_frame(ptr))
203 xdp_return_frame(veth_ptr_to_xdp(ptr));
208 static void __veth_xdp_flush(struct veth_rq *rq)
210 /* Write ptr_ring before reading rx_notify_masked */
212 if (!READ_ONCE(rq->rx_notify_masked) &&
213 napi_schedule_prep(&rq->xdp_napi)) {
214 WRITE_ONCE(rq->rx_notify_masked, true);
215 __napi_schedule(&rq->xdp_napi);
219 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
221 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
222 dev_kfree_skb_any(skb);
226 return NET_RX_SUCCESS;
229 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
230 struct veth_rq *rq, bool xdp)
232 return __dev_forward_skb(dev, skb) ?: xdp ?
233 veth_xdp_rx(rq, skb) :
237 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
239 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
240 struct veth_rq *rq = NULL;
241 struct net_device *rcv;
242 int length = skb->len;
243 bool rcv_xdp = false;
247 rcv = rcu_dereference(priv->peer);
248 if (unlikely(!rcv) || !pskb_may_pull(skb, ETH_HLEN)) {
253 rcv_priv = netdev_priv(rcv);
254 rxq = skb_get_queue_mapping(skb);
255 if (rxq < rcv->real_num_rx_queues) {
256 rq = &rcv_priv->rq[rxq];
257 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
260 skb_tx_timestamp(skb);
261 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
263 struct pcpu_lstats *stats = this_cpu_ptr(dev->lstats);
265 u64_stats_update_begin(&stats->syncp);
266 stats->bytes += length;
268 u64_stats_update_end(&stats->syncp);
272 atomic64_inc(&priv->dropped);
276 __veth_xdp_flush(rq);
283 static u64 veth_stats_tx(struct pcpu_lstats *result, struct net_device *dev)
285 struct veth_priv *priv = netdev_priv(dev);
290 for_each_possible_cpu(cpu) {
291 struct pcpu_lstats *stats = per_cpu_ptr(dev->lstats, cpu);
296 start = u64_stats_fetch_begin_irq(&stats->syncp);
297 packets = stats->packets;
298 bytes = stats->bytes;
299 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
300 result->packets += packets;
301 result->bytes += bytes;
303 return atomic64_read(&priv->dropped);
306 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
308 struct veth_priv *priv = netdev_priv(dev);
311 result->xdp_packets = 0;
312 result->xdp_bytes = 0;
313 result->xdp_drops = 0;
314 for (i = 0; i < dev->num_rx_queues; i++) {
315 struct veth_rq_stats *stats = &priv->rq[i].stats;
316 u64 packets, bytes, drops;
320 start = u64_stats_fetch_begin_irq(&stats->syncp);
321 packets = stats->xdp_packets;
322 bytes = stats->xdp_bytes;
323 drops = stats->xdp_drops;
324 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
325 result->xdp_packets += packets;
326 result->xdp_bytes += bytes;
327 result->xdp_drops += drops;
331 static void veth_get_stats64(struct net_device *dev,
332 struct rtnl_link_stats64 *tot)
334 struct veth_priv *priv = netdev_priv(dev);
335 struct net_device *peer;
336 struct veth_rq_stats rx;
337 struct pcpu_lstats tx;
339 tot->tx_dropped = veth_stats_tx(&tx, dev);
340 tot->tx_bytes = tx.bytes;
341 tot->tx_packets = tx.packets;
343 veth_stats_rx(&rx, dev);
344 tot->rx_dropped = rx.xdp_drops;
345 tot->rx_bytes = rx.xdp_bytes;
346 tot->rx_packets = rx.xdp_packets;
349 peer = rcu_dereference(priv->peer);
351 tot->rx_dropped += veth_stats_tx(&tx, peer);
352 tot->rx_bytes += tx.bytes;
353 tot->rx_packets += tx.packets;
355 veth_stats_rx(&rx, peer);
356 tot->tx_bytes += rx.xdp_bytes;
357 tot->tx_packets += rx.xdp_packets;
362 /* fake multicast ability */
363 static void veth_set_multicast_list(struct net_device *dev)
367 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
373 buflen = SKB_DATA_ALIGN(headroom + len) +
374 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
376 skb = build_skb(head, buflen);
380 skb_reserve(skb, headroom);
386 static int veth_select_rxq(struct net_device *dev)
388 return smp_processor_id() % dev->real_num_rx_queues;
391 static int veth_xdp_xmit(struct net_device *dev, int n,
392 struct xdp_frame **frames, u32 flags)
394 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
395 struct net_device *rcv;
396 int i, ret, drops = n;
397 unsigned int max_len;
400 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
405 rcv = rcu_dereference(priv->peer);
406 if (unlikely(!rcv)) {
411 rcv_priv = netdev_priv(rcv);
412 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
413 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
414 * side. This means an XDP program is loaded on the peer and the peer
417 if (!rcu_access_pointer(rq->xdp_prog)) {
423 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
425 spin_lock(&rq->xdp_ring.producer_lock);
426 for (i = 0; i < n; i++) {
427 struct xdp_frame *frame = frames[i];
428 void *ptr = veth_xdp_to_ptr(frame);
430 if (unlikely(frame->len > max_len ||
431 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
432 xdp_return_frame_rx_napi(frame);
436 spin_unlock(&rq->xdp_ring.producer_lock);
438 if (flags & XDP_XMIT_FLUSH)
439 __veth_xdp_flush(rq);
446 atomic64_add(drops, &priv->dropped);
451 static void veth_xdp_flush_bq(struct net_device *dev, struct veth_xdp_tx_bq *bq)
453 int sent, i, err = 0;
455 sent = veth_xdp_xmit(dev, bq->count, bq->q, 0);
459 for (i = 0; i < bq->count; i++)
460 xdp_return_frame(bq->q[i]);
462 trace_xdp_bulk_tx(dev, sent, bq->count - sent, err);
467 static void veth_xdp_flush(struct net_device *dev, struct veth_xdp_tx_bq *bq)
469 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
470 struct net_device *rcv;
474 veth_xdp_flush_bq(dev, bq);
475 rcv = rcu_dereference(priv->peer);
479 rcv_priv = netdev_priv(rcv);
480 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
481 /* xdp_ring is initialized on receive side? */
482 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
485 __veth_xdp_flush(rq);
490 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp,
491 struct veth_xdp_tx_bq *bq)
493 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
495 if (unlikely(!frame))
498 if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
499 veth_xdp_flush_bq(dev, bq);
501 bq->q[bq->count++] = frame;
506 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
507 struct xdp_frame *frame,
508 unsigned int *xdp_xmit,
509 struct veth_xdp_tx_bq *bq)
511 void *hard_start = frame->data - frame->headroom;
512 int len = frame->len, delta = 0;
513 struct xdp_frame orig_frame;
514 struct bpf_prog *xdp_prog;
515 unsigned int headroom;
518 /* bpf_xdp_adjust_head() assures BPF cannot access xdp_frame area */
519 hard_start -= sizeof(struct xdp_frame);
522 xdp_prog = rcu_dereference(rq->xdp_prog);
523 if (likely(xdp_prog)) {
527 xdp.data_hard_start = hard_start;
528 xdp.data = frame->data;
529 xdp.data_end = frame->data + frame->len;
530 xdp.data_meta = frame->data - frame->metasize;
531 xdp.rxq = &rq->xdp_rxq;
533 act = bpf_prog_run_xdp(xdp_prog, &xdp);
537 delta = frame->data - xdp.data;
538 len = xdp.data_end - xdp.data;
542 xdp.rxq->mem = frame->mem;
543 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
544 trace_xdp_exception(rq->dev, xdp_prog, act);
548 *xdp_xmit |= VETH_XDP_TX;
553 xdp.rxq->mem = frame->mem;
554 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
558 *xdp_xmit |= VETH_XDP_REDIR;
562 bpf_warn_invalid_xdp_action(act);
565 trace_xdp_exception(rq->dev, xdp_prog, act);
573 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
574 skb = veth_build_skb(hard_start, headroom, len, 0);
576 xdp_return_frame(frame);
580 xdp_release_frame(frame);
581 xdp_scrub_frame(frame);
582 skb->protocol = eth_type_trans(skb, rq->dev);
587 xdp_return_frame(frame);
592 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
593 unsigned int *xdp_xmit,
594 struct veth_xdp_tx_bq *bq)
596 u32 pktlen, headroom, act, metalen;
597 void *orig_data, *orig_data_end;
598 struct bpf_prog *xdp_prog;
599 int mac_len, delta, off;
605 xdp_prog = rcu_dereference(rq->xdp_prog);
606 if (unlikely(!xdp_prog)) {
611 mac_len = skb->data - skb_mac_header(skb);
612 pktlen = skb->len + mac_len;
613 headroom = skb_headroom(skb) - mac_len;
615 if (skb_shared(skb) || skb_head_is_locked(skb) ||
616 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
617 struct sk_buff *nskb;
622 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
623 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
624 if (size > PAGE_SIZE)
627 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
631 head = page_address(page);
632 start = head + VETH_XDP_HEADROOM;
633 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
634 page_frag_free(head);
638 nskb = veth_build_skb(head,
639 VETH_XDP_HEADROOM + mac_len, skb->len,
642 page_frag_free(head);
646 skb_copy_header(nskb, skb);
647 head_off = skb_headroom(nskb) - skb_headroom(skb);
648 skb_headers_offset_update(nskb, head_off);
653 xdp.data_hard_start = skb->head;
654 xdp.data = skb_mac_header(skb);
655 xdp.data_end = xdp.data + pktlen;
656 xdp.data_meta = xdp.data;
657 xdp.rxq = &rq->xdp_rxq;
658 orig_data = xdp.data;
659 orig_data_end = xdp.data_end;
661 act = bpf_prog_run_xdp(xdp_prog, &xdp);
667 get_page(virt_to_page(xdp.data));
669 xdp.rxq->mem = rq->xdp_mem;
670 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
671 trace_xdp_exception(rq->dev, xdp_prog, act);
674 *xdp_xmit |= VETH_XDP_TX;
678 get_page(virt_to_page(xdp.data));
680 xdp.rxq->mem = rq->xdp_mem;
681 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
683 *xdp_xmit |= VETH_XDP_REDIR;
687 bpf_warn_invalid_xdp_action(act);
690 trace_xdp_exception(rq->dev, xdp_prog, act);
697 delta = orig_data - xdp.data;
698 off = mac_len + delta;
700 __skb_push(skb, off);
702 __skb_pull(skb, -off);
703 skb->mac_header -= delta;
704 off = xdp.data_end - orig_data_end;
707 skb->protocol = eth_type_trans(skb, rq->dev);
709 metalen = xdp.data - xdp.data_meta;
711 skb_metadata_set(skb, metalen);
720 page_frag_free(xdp.data);
725 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit,
726 struct veth_xdp_tx_bq *bq)
728 int i, done = 0, drops = 0, bytes = 0;
730 for (i = 0; i < budget; i++) {
731 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
732 unsigned int xdp_xmit_one = 0;
738 if (veth_is_xdp_frame(ptr)) {
739 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
742 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one, bq);
746 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one, bq);
748 *xdp_xmit |= xdp_xmit_one;
751 napi_gro_receive(&rq->xdp_napi, skb);
752 else if (!xdp_xmit_one)
758 u64_stats_update_begin(&rq->stats.syncp);
759 rq->stats.xdp_packets += done;
760 rq->stats.xdp_bytes += bytes;
761 rq->stats.xdp_drops += drops;
762 u64_stats_update_end(&rq->stats.syncp);
767 static int veth_poll(struct napi_struct *napi, int budget)
770 container_of(napi, struct veth_rq, xdp_napi);
771 unsigned int xdp_xmit = 0;
772 struct veth_xdp_tx_bq bq;
777 xdp_set_return_frame_no_direct();
778 done = veth_xdp_rcv(rq, budget, &xdp_xmit, &bq);
780 if (done < budget && napi_complete_done(napi, done)) {
781 /* Write rx_notify_masked before reading ptr_ring */
782 smp_store_mb(rq->rx_notify_masked, false);
783 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
784 if (napi_schedule_prep(&rq->xdp_napi)) {
785 WRITE_ONCE(rq->rx_notify_masked, true);
786 __napi_schedule(&rq->xdp_napi);
791 if (xdp_xmit & VETH_XDP_TX)
792 veth_xdp_flush(rq->dev, &bq);
793 if (xdp_xmit & VETH_XDP_REDIR)
795 xdp_clear_return_frame_no_direct();
800 static int veth_napi_add(struct net_device *dev)
802 struct veth_priv *priv = netdev_priv(dev);
805 for (i = 0; i < dev->real_num_rx_queues; i++) {
806 struct veth_rq *rq = &priv->rq[i];
808 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
813 for (i = 0; i < dev->real_num_rx_queues; i++) {
814 struct veth_rq *rq = &priv->rq[i];
816 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
817 napi_enable(&rq->xdp_napi);
822 for (i--; i >= 0; i--)
823 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
828 static void veth_napi_del(struct net_device *dev)
830 struct veth_priv *priv = netdev_priv(dev);
833 for (i = 0; i < dev->real_num_rx_queues; i++) {
834 struct veth_rq *rq = &priv->rq[i];
836 napi_disable(&rq->xdp_napi);
837 napi_hash_del(&rq->xdp_napi);
841 for (i = 0; i < dev->real_num_rx_queues; i++) {
842 struct veth_rq *rq = &priv->rq[i];
844 netif_napi_del(&rq->xdp_napi);
845 rq->rx_notify_masked = false;
846 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
850 static int veth_enable_xdp(struct net_device *dev)
852 struct veth_priv *priv = netdev_priv(dev);
855 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
856 for (i = 0; i < dev->real_num_rx_queues; i++) {
857 struct veth_rq *rq = &priv->rq[i];
859 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
863 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
864 MEM_TYPE_PAGE_SHARED,
869 /* Save original mem info as it can be overwritten */
870 rq->xdp_mem = rq->xdp_rxq.mem;
873 err = veth_napi_add(dev);
878 for (i = 0; i < dev->real_num_rx_queues; i++)
879 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
883 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
885 for (i--; i >= 0; i--)
886 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
891 static void veth_disable_xdp(struct net_device *dev)
893 struct veth_priv *priv = netdev_priv(dev);
896 for (i = 0; i < dev->real_num_rx_queues; i++)
897 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
899 for (i = 0; i < dev->real_num_rx_queues; i++) {
900 struct veth_rq *rq = &priv->rq[i];
902 rq->xdp_rxq.mem = rq->xdp_mem;
903 xdp_rxq_info_unreg(&rq->xdp_rxq);
907 static int veth_open(struct net_device *dev)
909 struct veth_priv *priv = netdev_priv(dev);
910 struct net_device *peer = rtnl_dereference(priv->peer);
916 if (priv->_xdp_prog) {
917 err = veth_enable_xdp(dev);
922 if (peer->flags & IFF_UP) {
923 netif_carrier_on(dev);
924 netif_carrier_on(peer);
930 static int veth_close(struct net_device *dev)
932 struct veth_priv *priv = netdev_priv(dev);
933 struct net_device *peer = rtnl_dereference(priv->peer);
935 netif_carrier_off(dev);
937 netif_carrier_off(peer);
940 veth_disable_xdp(dev);
945 static int is_valid_veth_mtu(int mtu)
947 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
950 static int veth_alloc_queues(struct net_device *dev)
952 struct veth_priv *priv = netdev_priv(dev);
955 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
959 for (i = 0; i < dev->num_rx_queues; i++) {
960 priv->rq[i].dev = dev;
961 u64_stats_init(&priv->rq[i].stats.syncp);
967 static void veth_free_queues(struct net_device *dev)
969 struct veth_priv *priv = netdev_priv(dev);
974 static int veth_dev_init(struct net_device *dev)
978 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
982 err = veth_alloc_queues(dev);
984 free_percpu(dev->lstats);
991 static void veth_dev_free(struct net_device *dev)
993 veth_free_queues(dev);
994 free_percpu(dev->lstats);
997 #ifdef CONFIG_NET_POLL_CONTROLLER
998 static void veth_poll_controller(struct net_device *dev)
1000 /* veth only receives frames when its peer sends one
1001 * Since it has nothing to do with disabling irqs, we are guaranteed
1002 * never to have pending data when we poll for it so
1003 * there is nothing to do here.
1005 * We need this though so netpoll recognizes us as an interface that
1006 * supports polling, which enables bridge devices in virt setups to
1007 * still use netconsole
1010 #endif /* CONFIG_NET_POLL_CONTROLLER */
1012 static int veth_get_iflink(const struct net_device *dev)
1014 struct veth_priv *priv = netdev_priv(dev);
1015 struct net_device *peer;
1019 peer = rcu_dereference(priv->peer);
1020 iflink = peer ? peer->ifindex : 0;
1026 static netdev_features_t veth_fix_features(struct net_device *dev,
1027 netdev_features_t features)
1029 struct veth_priv *priv = netdev_priv(dev);
1030 struct net_device *peer;
1032 peer = rtnl_dereference(priv->peer);
1034 struct veth_priv *peer_priv = netdev_priv(peer);
1036 if (peer_priv->_xdp_prog)
1037 features &= ~NETIF_F_GSO_SOFTWARE;
1043 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1045 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1046 struct net_device *peer;
1052 peer = rcu_dereference(priv->peer);
1053 if (unlikely(!peer))
1056 peer_priv = netdev_priv(peer);
1057 priv->requested_headroom = new_hr;
1058 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1059 dev->needed_headroom = new_hr;
1060 peer->needed_headroom = new_hr;
1066 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1067 struct netlink_ext_ack *extack)
1069 struct veth_priv *priv = netdev_priv(dev);
1070 struct bpf_prog *old_prog;
1071 struct net_device *peer;
1072 unsigned int max_mtu;
1075 old_prog = priv->_xdp_prog;
1076 priv->_xdp_prog = prog;
1077 peer = rtnl_dereference(priv->peer);
1081 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1086 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1087 peer->hard_header_len -
1088 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1089 if (peer->mtu > max_mtu) {
1090 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1095 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1096 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1101 if (dev->flags & IFF_UP) {
1102 err = veth_enable_xdp(dev);
1104 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1110 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1111 peer->max_mtu = max_mtu;
1117 if (dev->flags & IFF_UP)
1118 veth_disable_xdp(dev);
1121 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1122 peer->max_mtu = ETH_MAX_MTU;
1125 bpf_prog_put(old_prog);
1128 if ((!!old_prog ^ !!prog) && peer)
1129 netdev_update_features(peer);
1133 priv->_xdp_prog = old_prog;
1138 static u32 veth_xdp_query(struct net_device *dev)
1140 struct veth_priv *priv = netdev_priv(dev);
1141 const struct bpf_prog *xdp_prog;
1143 xdp_prog = priv->_xdp_prog;
1145 return xdp_prog->aux->id;
1150 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1152 switch (xdp->command) {
1153 case XDP_SETUP_PROG:
1154 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1155 case XDP_QUERY_PROG:
1156 xdp->prog_id = veth_xdp_query(dev);
1163 static const struct net_device_ops veth_netdev_ops = {
1164 .ndo_init = veth_dev_init,
1165 .ndo_open = veth_open,
1166 .ndo_stop = veth_close,
1167 .ndo_start_xmit = veth_xmit,
1168 .ndo_get_stats64 = veth_get_stats64,
1169 .ndo_set_rx_mode = veth_set_multicast_list,
1170 .ndo_set_mac_address = eth_mac_addr,
1171 #ifdef CONFIG_NET_POLL_CONTROLLER
1172 .ndo_poll_controller = veth_poll_controller,
1174 .ndo_get_iflink = veth_get_iflink,
1175 .ndo_fix_features = veth_fix_features,
1176 .ndo_features_check = passthru_features_check,
1177 .ndo_set_rx_headroom = veth_set_rx_headroom,
1178 .ndo_bpf = veth_xdp,
1179 .ndo_xdp_xmit = veth_xdp_xmit,
1182 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1183 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1184 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1185 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1186 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1188 static void veth_setup(struct net_device *dev)
1192 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1193 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1194 dev->priv_flags |= IFF_NO_QUEUE;
1195 dev->priv_flags |= IFF_PHONY_HEADROOM;
1197 dev->netdev_ops = &veth_netdev_ops;
1198 dev->ethtool_ops = &veth_ethtool_ops;
1199 dev->features |= NETIF_F_LLTX;
1200 dev->features |= VETH_FEATURES;
1201 dev->vlan_features = dev->features &
1202 ~(NETIF_F_HW_VLAN_CTAG_TX |
1203 NETIF_F_HW_VLAN_STAG_TX |
1204 NETIF_F_HW_VLAN_CTAG_RX |
1205 NETIF_F_HW_VLAN_STAG_RX);
1206 dev->needs_free_netdev = true;
1207 dev->priv_destructor = veth_dev_free;
1208 dev->max_mtu = ETH_MAX_MTU;
1210 dev->hw_features = VETH_FEATURES;
1211 dev->hw_enc_features = VETH_FEATURES;
1212 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1219 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1220 struct netlink_ext_ack *extack)
1222 if (tb[IFLA_ADDRESS]) {
1223 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1225 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1226 return -EADDRNOTAVAIL;
1229 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1235 static struct rtnl_link_ops veth_link_ops;
1237 static int veth_newlink(struct net *src_net, struct net_device *dev,
1238 struct nlattr *tb[], struct nlattr *data[],
1239 struct netlink_ext_ack *extack)
1242 struct net_device *peer;
1243 struct veth_priv *priv;
1244 char ifname[IFNAMSIZ];
1245 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1246 unsigned char name_assign_type;
1247 struct ifinfomsg *ifmp;
1251 * create and register peer first
1253 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1254 struct nlattr *nla_peer;
1256 nla_peer = data[VETH_INFO_PEER];
1257 ifmp = nla_data(nla_peer);
1258 err = rtnl_nla_parse_ifla(peer_tb,
1259 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1260 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1265 err = veth_validate(peer_tb, NULL, extack);
1275 if (ifmp && tbp[IFLA_IFNAME]) {
1276 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1277 name_assign_type = NET_NAME_USER;
1279 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1280 name_assign_type = NET_NAME_ENUM;
1283 net = rtnl_link_get_net(src_net, tbp);
1285 return PTR_ERR(net);
1287 peer = rtnl_create_link(net, ifname, name_assign_type,
1288 &veth_link_ops, tbp, extack);
1291 return PTR_ERR(peer);
1294 if (!ifmp || !tbp[IFLA_ADDRESS])
1295 eth_hw_addr_random(peer);
1297 if (ifmp && (dev->ifindex != 0))
1298 peer->ifindex = ifmp->ifi_index;
1300 peer->gso_max_size = dev->gso_max_size;
1301 peer->gso_max_segs = dev->gso_max_segs;
1303 err = register_netdevice(peer);
1307 goto err_register_peer;
1309 netif_carrier_off(peer);
1311 err = rtnl_configure_link(peer, ifmp);
1313 goto err_configure_peer;
1318 * note, that since we've registered new device the dev's name
1319 * should be re-allocated
1322 if (tb[IFLA_ADDRESS] == NULL)
1323 eth_hw_addr_random(dev);
1325 if (tb[IFLA_IFNAME])
1326 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1328 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1330 err = register_netdevice(dev);
1332 goto err_register_dev;
1334 netif_carrier_off(dev);
1337 * tie the deviced together
1340 priv = netdev_priv(dev);
1341 rcu_assign_pointer(priv->peer, peer);
1343 priv = netdev_priv(peer);
1344 rcu_assign_pointer(priv->peer, dev);
1351 unregister_netdevice(peer);
1359 static void veth_dellink(struct net_device *dev, struct list_head *head)
1361 struct veth_priv *priv;
1362 struct net_device *peer;
1364 priv = netdev_priv(dev);
1365 peer = rtnl_dereference(priv->peer);
1367 /* Note : dellink() is called from default_device_exit_batch(),
1368 * before a rcu_synchronize() point. The devices are guaranteed
1369 * not being freed before one RCU grace period.
1371 RCU_INIT_POINTER(priv->peer, NULL);
1372 unregister_netdevice_queue(dev, head);
1375 priv = netdev_priv(peer);
1376 RCU_INIT_POINTER(priv->peer, NULL);
1377 unregister_netdevice_queue(peer, head);
1381 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1382 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1385 static struct net *veth_get_link_net(const struct net_device *dev)
1387 struct veth_priv *priv = netdev_priv(dev);
1388 struct net_device *peer = rtnl_dereference(priv->peer);
1390 return peer ? dev_net(peer) : dev_net(dev);
1393 static struct rtnl_link_ops veth_link_ops = {
1395 .priv_size = sizeof(struct veth_priv),
1396 .setup = veth_setup,
1397 .validate = veth_validate,
1398 .newlink = veth_newlink,
1399 .dellink = veth_dellink,
1400 .policy = veth_policy,
1401 .maxtype = VETH_INFO_MAX,
1402 .get_link_net = veth_get_link_net,
1409 static __init int veth_init(void)
1411 return rtnl_link_register(&veth_link_ops);
1414 static __exit void veth_exit(void)
1416 rtnl_link_unregister(&veth_link_ops);
1419 module_init(veth_init);
1420 module_exit(veth_exit);
1422 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1423 MODULE_LICENSE("GPL v2");
1424 MODULE_ALIAS_RTNL_LINK(DRV_NAME);