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 (!rq->rx_notify_masked) {
213 rq->rx_notify_masked = true;
214 napi_schedule(&rq->xdp_napi);
218 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
220 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
221 dev_kfree_skb_any(skb);
225 return NET_RX_SUCCESS;
228 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
229 struct veth_rq *rq, bool xdp)
231 return __dev_forward_skb(dev, skb) ?: xdp ?
232 veth_xdp_rx(rq, skb) :
236 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
238 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
239 struct veth_rq *rq = NULL;
240 struct net_device *rcv;
241 int length = skb->len;
242 bool rcv_xdp = false;
246 rcv = rcu_dereference(priv->peer);
247 if (unlikely(!rcv)) {
252 rcv_priv = netdev_priv(rcv);
253 rxq = skb_get_queue_mapping(skb);
254 if (rxq < rcv->real_num_rx_queues) {
255 rq = &rcv_priv->rq[rxq];
256 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
258 skb_record_rx_queue(skb, rxq);
261 skb_tx_timestamp(skb);
262 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
264 dev_lstats_add(dev, length);
267 atomic64_inc(&priv->dropped);
271 __veth_xdp_flush(rq);
278 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
280 struct veth_priv *priv = netdev_priv(dev);
282 dev_lstats_read(dev, packets, bytes);
283 return atomic64_read(&priv->dropped);
286 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
288 struct veth_priv *priv = netdev_priv(dev);
291 result->xdp_packets = 0;
292 result->xdp_bytes = 0;
293 result->xdp_drops = 0;
294 for (i = 0; i < dev->num_rx_queues; i++) {
295 struct veth_rq_stats *stats = &priv->rq[i].stats;
296 u64 packets, bytes, drops;
300 start = u64_stats_fetch_begin_irq(&stats->syncp);
301 packets = stats->xdp_packets;
302 bytes = stats->xdp_bytes;
303 drops = stats->xdp_drops;
304 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
305 result->xdp_packets += packets;
306 result->xdp_bytes += bytes;
307 result->xdp_drops += drops;
311 static void veth_get_stats64(struct net_device *dev,
312 struct rtnl_link_stats64 *tot)
314 struct veth_priv *priv = netdev_priv(dev);
315 struct net_device *peer;
316 struct veth_rq_stats rx;
319 tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
320 tot->tx_bytes = bytes;
321 tot->tx_packets = packets;
323 veth_stats_rx(&rx, dev);
324 tot->rx_dropped = rx.xdp_drops;
325 tot->rx_bytes = rx.xdp_bytes;
326 tot->rx_packets = rx.xdp_packets;
329 peer = rcu_dereference(priv->peer);
331 tot->rx_dropped += veth_stats_tx(peer, &packets, &bytes);
332 tot->rx_bytes += bytes;
333 tot->rx_packets += packets;
335 veth_stats_rx(&rx, peer);
336 tot->tx_bytes += rx.xdp_bytes;
337 tot->tx_packets += rx.xdp_packets;
342 /* fake multicast ability */
343 static void veth_set_multicast_list(struct net_device *dev)
347 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
353 buflen = SKB_DATA_ALIGN(headroom + len) +
354 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
356 skb = build_skb(head, buflen);
360 skb_reserve(skb, headroom);
366 static int veth_select_rxq(struct net_device *dev)
368 return smp_processor_id() % dev->real_num_rx_queues;
371 static int veth_xdp_xmit(struct net_device *dev, int n,
372 struct xdp_frame **frames, u32 flags)
374 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
375 struct net_device *rcv;
376 int i, ret, drops = n;
377 unsigned int max_len;
380 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
385 rcv = rcu_dereference(priv->peer);
386 if (unlikely(!rcv)) {
391 rcv_priv = netdev_priv(rcv);
392 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
393 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
394 * side. This means an XDP program is loaded on the peer and the peer
397 if (!rcu_access_pointer(rq->xdp_prog)) {
403 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
405 spin_lock(&rq->xdp_ring.producer_lock);
406 for (i = 0; i < n; i++) {
407 struct xdp_frame *frame = frames[i];
408 void *ptr = veth_xdp_to_ptr(frame);
410 if (unlikely(frame->len > max_len ||
411 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
412 xdp_return_frame_rx_napi(frame);
416 spin_unlock(&rq->xdp_ring.producer_lock);
418 if (flags & XDP_XMIT_FLUSH)
419 __veth_xdp_flush(rq);
426 atomic64_add(drops, &priv->dropped);
431 static void veth_xdp_flush_bq(struct net_device *dev, struct veth_xdp_tx_bq *bq)
433 int sent, i, err = 0;
435 sent = veth_xdp_xmit(dev, bq->count, bq->q, 0);
439 for (i = 0; i < bq->count; i++)
440 xdp_return_frame(bq->q[i]);
442 trace_xdp_bulk_tx(dev, sent, bq->count - sent, err);
447 static void veth_xdp_flush(struct net_device *dev, struct veth_xdp_tx_bq *bq)
449 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
450 struct net_device *rcv;
454 veth_xdp_flush_bq(dev, bq);
455 rcv = rcu_dereference(priv->peer);
459 rcv_priv = netdev_priv(rcv);
460 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
461 /* xdp_ring is initialized on receive side? */
462 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
465 __veth_xdp_flush(rq);
470 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp,
471 struct veth_xdp_tx_bq *bq)
473 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
475 if (unlikely(!frame))
478 if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
479 veth_xdp_flush_bq(dev, bq);
481 bq->q[bq->count++] = frame;
486 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
487 struct xdp_frame *frame,
488 unsigned int *xdp_xmit,
489 struct veth_xdp_tx_bq *bq)
491 void *hard_start = frame->data - frame->headroom;
492 void *head = hard_start - sizeof(struct xdp_frame);
493 int len = frame->len, delta = 0;
494 struct xdp_frame orig_frame;
495 struct bpf_prog *xdp_prog;
496 unsigned int headroom;
500 xdp_prog = rcu_dereference(rq->xdp_prog);
501 if (likely(xdp_prog)) {
505 xdp.data_hard_start = hard_start;
506 xdp.data = frame->data;
507 xdp.data_end = frame->data + frame->len;
508 xdp.data_meta = frame->data - frame->metasize;
509 xdp.rxq = &rq->xdp_rxq;
511 act = bpf_prog_run_xdp(xdp_prog, &xdp);
515 delta = frame->data - xdp.data;
516 len = xdp.data_end - xdp.data;
520 xdp.data_hard_start = head;
521 xdp.rxq->mem = frame->mem;
522 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
523 trace_xdp_exception(rq->dev, xdp_prog, act);
527 *xdp_xmit |= VETH_XDP_TX;
532 xdp.data_hard_start = head;
533 xdp.rxq->mem = frame->mem;
534 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
538 *xdp_xmit |= VETH_XDP_REDIR;
542 bpf_warn_invalid_xdp_action(act);
545 trace_xdp_exception(rq->dev, xdp_prog, act);
553 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
554 skb = veth_build_skb(head, headroom, len, 0);
556 xdp_return_frame(frame);
560 xdp_release_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,
574 struct veth_xdp_tx_bq *bq)
576 u32 pktlen, headroom, act, metalen;
577 void *orig_data, *orig_data_end;
578 struct bpf_prog *xdp_prog;
579 int mac_len, delta, off;
585 xdp_prog = rcu_dereference(rq->xdp_prog);
586 if (unlikely(!xdp_prog)) {
591 mac_len = skb->data - skb_mac_header(skb);
592 pktlen = skb->len + mac_len;
593 headroom = skb_headroom(skb) - mac_len;
595 if (skb_shared(skb) || skb_head_is_locked(skb) ||
596 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
597 struct sk_buff *nskb;
602 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
603 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
604 if (size > PAGE_SIZE)
607 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
611 head = page_address(page);
612 start = head + VETH_XDP_HEADROOM;
613 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
614 page_frag_free(head);
618 nskb = veth_build_skb(head,
619 VETH_XDP_HEADROOM + mac_len, skb->len,
622 page_frag_free(head);
626 skb_copy_header(nskb, skb);
627 head_off = skb_headroom(nskb) - skb_headroom(skb);
628 skb_headers_offset_update(nskb, head_off);
633 xdp.data_hard_start = skb->head;
634 xdp.data = skb_mac_header(skb);
635 xdp.data_end = xdp.data + pktlen;
636 xdp.data_meta = xdp.data;
637 xdp.rxq = &rq->xdp_rxq;
638 orig_data = xdp.data;
639 orig_data_end = xdp.data_end;
641 act = bpf_prog_run_xdp(xdp_prog, &xdp);
647 get_page(virt_to_page(xdp.data));
649 xdp.rxq->mem = rq->xdp_mem;
650 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
651 trace_xdp_exception(rq->dev, xdp_prog, act);
654 *xdp_xmit |= VETH_XDP_TX;
658 get_page(virt_to_page(xdp.data));
660 xdp.rxq->mem = rq->xdp_mem;
661 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
663 *xdp_xmit |= VETH_XDP_REDIR;
667 bpf_warn_invalid_xdp_action(act);
670 trace_xdp_exception(rq->dev, xdp_prog, act);
677 delta = orig_data - xdp.data;
678 off = mac_len + delta;
680 __skb_push(skb, off);
682 __skb_pull(skb, -off);
683 skb->mac_header -= delta;
684 off = xdp.data_end - orig_data_end;
687 skb->protocol = eth_type_trans(skb, rq->dev);
689 metalen = xdp.data - xdp.data_meta;
691 skb_metadata_set(skb, metalen);
700 page_frag_free(xdp.data);
705 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit,
706 struct veth_xdp_tx_bq *bq)
708 int i, done = 0, drops = 0, bytes = 0;
710 for (i = 0; i < budget; i++) {
711 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
712 unsigned int xdp_xmit_one = 0;
718 if (veth_is_xdp_frame(ptr)) {
719 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
722 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one, bq);
726 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one, bq);
728 *xdp_xmit |= xdp_xmit_one;
731 napi_gro_receive(&rq->xdp_napi, skb);
732 else if (!xdp_xmit_one)
738 u64_stats_update_begin(&rq->stats.syncp);
739 rq->stats.xdp_packets += done;
740 rq->stats.xdp_bytes += bytes;
741 rq->stats.xdp_drops += drops;
742 u64_stats_update_end(&rq->stats.syncp);
747 static int veth_poll(struct napi_struct *napi, int budget)
750 container_of(napi, struct veth_rq, xdp_napi);
751 unsigned int xdp_xmit = 0;
752 struct veth_xdp_tx_bq bq;
757 xdp_set_return_frame_no_direct();
758 done = veth_xdp_rcv(rq, budget, &xdp_xmit, &bq);
760 if (done < budget && napi_complete_done(napi, done)) {
761 /* Write rx_notify_masked before reading ptr_ring */
762 smp_store_mb(rq->rx_notify_masked, false);
763 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
764 rq->rx_notify_masked = true;
765 napi_schedule(&rq->xdp_napi);
769 if (xdp_xmit & VETH_XDP_TX)
770 veth_xdp_flush(rq->dev, &bq);
771 if (xdp_xmit & VETH_XDP_REDIR)
773 xdp_clear_return_frame_no_direct();
778 static int veth_napi_add(struct net_device *dev)
780 struct veth_priv *priv = netdev_priv(dev);
783 for (i = 0; i < dev->real_num_rx_queues; i++) {
784 struct veth_rq *rq = &priv->rq[i];
786 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
791 for (i = 0; i < dev->real_num_rx_queues; i++) {
792 struct veth_rq *rq = &priv->rq[i];
794 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
795 napi_enable(&rq->xdp_napi);
800 for (i--; i >= 0; i--)
801 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
806 static void veth_napi_del(struct net_device *dev)
808 struct veth_priv *priv = netdev_priv(dev);
811 for (i = 0; i < dev->real_num_rx_queues; i++) {
812 struct veth_rq *rq = &priv->rq[i];
814 napi_disable(&rq->xdp_napi);
815 napi_hash_del(&rq->xdp_napi);
819 for (i = 0; i < dev->real_num_rx_queues; i++) {
820 struct veth_rq *rq = &priv->rq[i];
822 netif_napi_del(&rq->xdp_napi);
823 rq->rx_notify_masked = false;
824 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
828 static int veth_enable_xdp(struct net_device *dev)
830 struct veth_priv *priv = netdev_priv(dev);
833 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
834 for (i = 0; i < dev->real_num_rx_queues; i++) {
835 struct veth_rq *rq = &priv->rq[i];
837 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
841 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
842 MEM_TYPE_PAGE_SHARED,
847 /* Save original mem info as it can be overwritten */
848 rq->xdp_mem = rq->xdp_rxq.mem;
851 err = veth_napi_add(dev);
856 for (i = 0; i < dev->real_num_rx_queues; i++)
857 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
861 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
863 for (i--; i >= 0; i--)
864 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
869 static void veth_disable_xdp(struct net_device *dev)
871 struct veth_priv *priv = netdev_priv(dev);
874 for (i = 0; i < dev->real_num_rx_queues; i++)
875 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
877 for (i = 0; i < dev->real_num_rx_queues; i++) {
878 struct veth_rq *rq = &priv->rq[i];
880 rq->xdp_rxq.mem = rq->xdp_mem;
881 xdp_rxq_info_unreg(&rq->xdp_rxq);
885 static int veth_open(struct net_device *dev)
887 struct veth_priv *priv = netdev_priv(dev);
888 struct net_device *peer = rtnl_dereference(priv->peer);
894 if (priv->_xdp_prog) {
895 err = veth_enable_xdp(dev);
900 if (peer->flags & IFF_UP) {
901 netif_carrier_on(dev);
902 netif_carrier_on(peer);
908 static int veth_close(struct net_device *dev)
910 struct veth_priv *priv = netdev_priv(dev);
911 struct net_device *peer = rtnl_dereference(priv->peer);
913 netif_carrier_off(dev);
915 netif_carrier_off(peer);
918 veth_disable_xdp(dev);
923 static int is_valid_veth_mtu(int mtu)
925 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
928 static int veth_alloc_queues(struct net_device *dev)
930 struct veth_priv *priv = netdev_priv(dev);
933 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
937 for (i = 0; i < dev->num_rx_queues; i++) {
938 priv->rq[i].dev = dev;
939 u64_stats_init(&priv->rq[i].stats.syncp);
945 static void veth_free_queues(struct net_device *dev)
947 struct veth_priv *priv = netdev_priv(dev);
952 static int veth_dev_init(struct net_device *dev)
956 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
960 err = veth_alloc_queues(dev);
962 free_percpu(dev->lstats);
969 static void veth_dev_free(struct net_device *dev)
971 veth_free_queues(dev);
972 free_percpu(dev->lstats);
975 #ifdef CONFIG_NET_POLL_CONTROLLER
976 static void veth_poll_controller(struct net_device *dev)
978 /* veth only receives frames when its peer sends one
979 * Since it has nothing to do with disabling irqs, we are guaranteed
980 * never to have pending data when we poll for it so
981 * there is nothing to do here.
983 * We need this though so netpoll recognizes us as an interface that
984 * supports polling, which enables bridge devices in virt setups to
985 * still use netconsole
988 #endif /* CONFIG_NET_POLL_CONTROLLER */
990 static int veth_get_iflink(const struct net_device *dev)
992 struct veth_priv *priv = netdev_priv(dev);
993 struct net_device *peer;
997 peer = rcu_dereference(priv->peer);
998 iflink = peer ? peer->ifindex : 0;
1004 static netdev_features_t veth_fix_features(struct net_device *dev,
1005 netdev_features_t features)
1007 struct veth_priv *priv = netdev_priv(dev);
1008 struct net_device *peer;
1010 peer = rtnl_dereference(priv->peer);
1012 struct veth_priv *peer_priv = netdev_priv(peer);
1014 if (peer_priv->_xdp_prog)
1015 features &= ~NETIF_F_GSO_SOFTWARE;
1021 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1023 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1024 struct net_device *peer;
1030 peer = rcu_dereference(priv->peer);
1031 if (unlikely(!peer))
1034 peer_priv = netdev_priv(peer);
1035 priv->requested_headroom = new_hr;
1036 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1037 dev->needed_headroom = new_hr;
1038 peer->needed_headroom = new_hr;
1044 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1045 struct netlink_ext_ack *extack)
1047 struct veth_priv *priv = netdev_priv(dev);
1048 struct bpf_prog *old_prog;
1049 struct net_device *peer;
1050 unsigned int max_mtu;
1053 old_prog = priv->_xdp_prog;
1054 priv->_xdp_prog = prog;
1055 peer = rtnl_dereference(priv->peer);
1059 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1064 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1065 peer->hard_header_len -
1066 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1067 if (peer->mtu > max_mtu) {
1068 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1073 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1074 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1079 if (dev->flags & IFF_UP) {
1080 err = veth_enable_xdp(dev);
1082 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1088 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1089 peer->max_mtu = max_mtu;
1095 if (dev->flags & IFF_UP)
1096 veth_disable_xdp(dev);
1099 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1100 peer->max_mtu = ETH_MAX_MTU;
1103 bpf_prog_put(old_prog);
1106 if ((!!old_prog ^ !!prog) && peer)
1107 netdev_update_features(peer);
1111 priv->_xdp_prog = old_prog;
1116 static u32 veth_xdp_query(struct net_device *dev)
1118 struct veth_priv *priv = netdev_priv(dev);
1119 const struct bpf_prog *xdp_prog;
1121 xdp_prog = priv->_xdp_prog;
1123 return xdp_prog->aux->id;
1128 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1130 switch (xdp->command) {
1131 case XDP_SETUP_PROG:
1132 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1133 case XDP_QUERY_PROG:
1134 xdp->prog_id = veth_xdp_query(dev);
1141 static const struct net_device_ops veth_netdev_ops = {
1142 .ndo_init = veth_dev_init,
1143 .ndo_open = veth_open,
1144 .ndo_stop = veth_close,
1145 .ndo_start_xmit = veth_xmit,
1146 .ndo_get_stats64 = veth_get_stats64,
1147 .ndo_set_rx_mode = veth_set_multicast_list,
1148 .ndo_set_mac_address = eth_mac_addr,
1149 #ifdef CONFIG_NET_POLL_CONTROLLER
1150 .ndo_poll_controller = veth_poll_controller,
1152 .ndo_get_iflink = veth_get_iflink,
1153 .ndo_fix_features = veth_fix_features,
1154 .ndo_features_check = passthru_features_check,
1155 .ndo_set_rx_headroom = veth_set_rx_headroom,
1156 .ndo_bpf = veth_xdp,
1157 .ndo_xdp_xmit = veth_xdp_xmit,
1160 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1161 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1162 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1163 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1164 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1166 static void veth_setup(struct net_device *dev)
1170 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1171 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1172 dev->priv_flags |= IFF_NO_QUEUE;
1173 dev->priv_flags |= IFF_PHONY_HEADROOM;
1175 dev->netdev_ops = &veth_netdev_ops;
1176 dev->ethtool_ops = &veth_ethtool_ops;
1177 dev->features |= NETIF_F_LLTX;
1178 dev->features |= VETH_FEATURES;
1179 dev->vlan_features = dev->features &
1180 ~(NETIF_F_HW_VLAN_CTAG_TX |
1181 NETIF_F_HW_VLAN_STAG_TX |
1182 NETIF_F_HW_VLAN_CTAG_RX |
1183 NETIF_F_HW_VLAN_STAG_RX);
1184 dev->needs_free_netdev = true;
1185 dev->priv_destructor = veth_dev_free;
1186 dev->max_mtu = ETH_MAX_MTU;
1188 dev->hw_features = VETH_FEATURES;
1189 dev->hw_enc_features = VETH_FEATURES;
1190 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1197 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1198 struct netlink_ext_ack *extack)
1200 if (tb[IFLA_ADDRESS]) {
1201 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1203 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1204 return -EADDRNOTAVAIL;
1207 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1213 static struct rtnl_link_ops veth_link_ops;
1215 static int veth_newlink(struct net *src_net, struct net_device *dev,
1216 struct nlattr *tb[], struct nlattr *data[],
1217 struct netlink_ext_ack *extack)
1220 struct net_device *peer;
1221 struct veth_priv *priv;
1222 char ifname[IFNAMSIZ];
1223 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1224 unsigned char name_assign_type;
1225 struct ifinfomsg *ifmp;
1229 * create and register peer first
1231 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1232 struct nlattr *nla_peer;
1234 nla_peer = data[VETH_INFO_PEER];
1235 ifmp = nla_data(nla_peer);
1236 err = rtnl_nla_parse_ifla(peer_tb,
1237 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1238 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1243 err = veth_validate(peer_tb, NULL, extack);
1253 if (ifmp && tbp[IFLA_IFNAME]) {
1254 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1255 name_assign_type = NET_NAME_USER;
1257 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1258 name_assign_type = NET_NAME_ENUM;
1261 net = rtnl_link_get_net(src_net, tbp);
1263 return PTR_ERR(net);
1265 peer = rtnl_create_link(net, ifname, name_assign_type,
1266 &veth_link_ops, tbp, extack);
1269 return PTR_ERR(peer);
1272 if (!ifmp || !tbp[IFLA_ADDRESS])
1273 eth_hw_addr_random(peer);
1275 if (ifmp && (dev->ifindex != 0))
1276 peer->ifindex = ifmp->ifi_index;
1278 peer->gso_max_size = dev->gso_max_size;
1279 peer->gso_max_segs = dev->gso_max_segs;
1281 err = register_netdevice(peer);
1285 goto err_register_peer;
1287 netif_carrier_off(peer);
1289 err = rtnl_configure_link(peer, ifmp);
1291 goto err_configure_peer;
1296 * note, that since we've registered new device the dev's name
1297 * should be re-allocated
1300 if (tb[IFLA_ADDRESS] == NULL)
1301 eth_hw_addr_random(dev);
1303 if (tb[IFLA_IFNAME])
1304 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1306 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1308 err = register_netdevice(dev);
1310 goto err_register_dev;
1312 netif_carrier_off(dev);
1315 * tie the deviced together
1318 priv = netdev_priv(dev);
1319 rcu_assign_pointer(priv->peer, peer);
1321 priv = netdev_priv(peer);
1322 rcu_assign_pointer(priv->peer, dev);
1329 unregister_netdevice(peer);
1337 static void veth_dellink(struct net_device *dev, struct list_head *head)
1339 struct veth_priv *priv;
1340 struct net_device *peer;
1342 priv = netdev_priv(dev);
1343 peer = rtnl_dereference(priv->peer);
1345 /* Note : dellink() is called from default_device_exit_batch(),
1346 * before a rcu_synchronize() point. The devices are guaranteed
1347 * not being freed before one RCU grace period.
1349 RCU_INIT_POINTER(priv->peer, NULL);
1350 unregister_netdevice_queue(dev, head);
1353 priv = netdev_priv(peer);
1354 RCU_INIT_POINTER(priv->peer, NULL);
1355 unregister_netdevice_queue(peer, head);
1359 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1360 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1363 static struct net *veth_get_link_net(const struct net_device *dev)
1365 struct veth_priv *priv = netdev_priv(dev);
1366 struct net_device *peer = rtnl_dereference(priv->peer);
1368 return peer ? dev_net(peer) : dev_net(dev);
1371 static struct rtnl_link_ops veth_link_ops = {
1373 .priv_size = sizeof(struct veth_priv),
1374 .setup = veth_setup,
1375 .validate = veth_validate,
1376 .newlink = veth_newlink,
1377 .dellink = veth_dellink,
1378 .policy = veth_policy,
1379 .maxtype = VETH_INFO_MAX,
1380 .get_link_net = veth_get_link_net,
1387 static __init int veth_init(void)
1389 return rtnl_link_register(&veth_link_ops);
1392 static __exit void veth_exit(void)
1394 rtnl_link_unregister(&veth_link_ops);
1397 module_init(veth_init);
1398 module_exit(veth_exit);
1400 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1401 MODULE_LICENSE("GPL v2");
1402 MODULE_ALIAS_RTNL_LINK(DRV_NAME);