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>
29 #include <net/page_pool/helpers.h>
31 #define DRV_NAME "veth"
32 #define DRV_VERSION "1.0"
34 #define VETH_XDP_FLAG BIT(0)
35 #define VETH_RING_SIZE 256
36 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
38 #define VETH_XDP_TX_BULK_SIZE 16
39 #define VETH_XDP_BATCH 16
51 u64 peer_tq_xdp_xmit_err;
54 struct veth_rq_stats {
56 struct u64_stats_sync syncp;
60 struct napi_struct xdp_napi;
61 struct napi_struct __rcu *napi; /* points to xdp_napi when the latter is initialized */
62 struct net_device *dev;
63 struct bpf_prog __rcu *xdp_prog;
64 struct xdp_mem_info xdp_mem;
65 struct veth_rq_stats stats;
66 bool rx_notify_masked;
67 struct ptr_ring xdp_ring;
68 struct xdp_rxq_info xdp_rxq;
69 struct page_pool *page_pool;
73 struct net_device __rcu *peer;
75 struct bpf_prog *_xdp_prog;
77 unsigned int requested_headroom;
80 struct veth_xdp_tx_bq {
81 struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
89 struct veth_q_stat_desc {
90 char desc[ETH_GSTRING_LEN];
94 #define VETH_RQ_STAT(m) offsetof(struct veth_stats, m)
96 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
97 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
98 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
99 { "drops", VETH_RQ_STAT(rx_drops) },
100 { "xdp_redirect", VETH_RQ_STAT(xdp_redirect) },
101 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
102 { "xdp_tx", VETH_RQ_STAT(xdp_tx) },
103 { "xdp_tx_errors", VETH_RQ_STAT(xdp_tx_err) },
106 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
108 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
109 { "xdp_xmit", VETH_RQ_STAT(peer_tq_xdp_xmit) },
110 { "xdp_xmit_errors", VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
113 #define VETH_TQ_STATS_LEN ARRAY_SIZE(veth_tq_stats_desc)
116 const char string[ETH_GSTRING_LEN];
117 } ethtool_stats_keys[] = {
121 struct veth_xdp_buff {
126 static int veth_get_link_ksettings(struct net_device *dev,
127 struct ethtool_link_ksettings *cmd)
129 cmd->base.speed = SPEED_10000;
130 cmd->base.duplex = DUPLEX_FULL;
131 cmd->base.port = PORT_TP;
132 cmd->base.autoneg = AUTONEG_DISABLE;
136 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
138 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
139 strscpy(info->version, DRV_VERSION, sizeof(info->version));
142 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
149 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys));
150 p += sizeof(ethtool_stats_keys);
151 for (i = 0; i < dev->real_num_rx_queues; i++)
152 for (j = 0; j < VETH_RQ_STATS_LEN; j++)
153 ethtool_sprintf(&p, "rx_queue_%u_%.18s",
154 i, veth_rq_stats_desc[j].desc);
156 for (i = 0; i < dev->real_num_tx_queues; i++)
157 for (j = 0; j < VETH_TQ_STATS_LEN; j++)
158 ethtool_sprintf(&p, "tx_queue_%u_%.18s",
159 i, veth_tq_stats_desc[j].desc);
161 page_pool_ethtool_stats_get_strings(p);
166 static int veth_get_sset_count(struct net_device *dev, int sset)
170 return ARRAY_SIZE(ethtool_stats_keys) +
171 VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
172 VETH_TQ_STATS_LEN * dev->real_num_tx_queues +
173 page_pool_ethtool_stats_get_count();
179 static void veth_get_page_pool_stats(struct net_device *dev, u64 *data)
181 #ifdef CONFIG_PAGE_POOL_STATS
182 struct veth_priv *priv = netdev_priv(dev);
183 struct page_pool_stats pp_stats = {};
186 for (i = 0; i < dev->real_num_rx_queues; i++) {
187 if (!priv->rq[i].page_pool)
189 page_pool_get_stats(priv->rq[i].page_pool, &pp_stats);
191 page_pool_ethtool_stats_get(data, &pp_stats);
192 #endif /* CONFIG_PAGE_POOL_STATS */
195 static void veth_get_ethtool_stats(struct net_device *dev,
196 struct ethtool_stats *stats, u64 *data)
198 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
199 struct net_device *peer = rtnl_dereference(priv->peer);
200 int i, j, idx, pp_idx;
202 data[0] = peer ? peer->ifindex : 0;
204 for (i = 0; i < dev->real_num_rx_queues; i++) {
205 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
206 const void *stats_base = (void *)&rq_stats->vs;
211 start = u64_stats_fetch_begin(&rq_stats->syncp);
212 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
213 offset = veth_rq_stats_desc[j].offset;
214 data[idx + j] = *(u64 *)(stats_base + offset);
216 } while (u64_stats_fetch_retry(&rq_stats->syncp, start));
217 idx += VETH_RQ_STATS_LEN;
222 goto page_pool_stats;
224 rcv_priv = netdev_priv(peer);
225 for (i = 0; i < peer->real_num_rx_queues; i++) {
226 const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
227 const void *base = (void *)&rq_stats->vs;
228 unsigned int start, tx_idx = idx;
231 tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
233 start = u64_stats_fetch_begin(&rq_stats->syncp);
234 for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
235 offset = veth_tq_stats_desc[j].offset;
236 data[tx_idx + j] += *(u64 *)(base + offset);
238 } while (u64_stats_fetch_retry(&rq_stats->syncp, start));
240 pp_idx = idx + dev->real_num_tx_queues * VETH_TQ_STATS_LEN;
243 veth_get_page_pool_stats(dev, &data[pp_idx]);
246 static void veth_get_channels(struct net_device *dev,
247 struct ethtool_channels *channels)
249 channels->tx_count = dev->real_num_tx_queues;
250 channels->rx_count = dev->real_num_rx_queues;
251 channels->max_tx = dev->num_tx_queues;
252 channels->max_rx = dev->num_rx_queues;
255 static int veth_set_channels(struct net_device *dev,
256 struct ethtool_channels *ch);
258 static const struct ethtool_ops veth_ethtool_ops = {
259 .get_drvinfo = veth_get_drvinfo,
260 .get_link = ethtool_op_get_link,
261 .get_strings = veth_get_strings,
262 .get_sset_count = veth_get_sset_count,
263 .get_ethtool_stats = veth_get_ethtool_stats,
264 .get_link_ksettings = veth_get_link_ksettings,
265 .get_ts_info = ethtool_op_get_ts_info,
266 .get_channels = veth_get_channels,
267 .set_channels = veth_set_channels,
270 /* general routines */
272 static bool veth_is_xdp_frame(void *ptr)
274 return (unsigned long)ptr & VETH_XDP_FLAG;
277 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
279 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
282 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
284 return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
287 static void veth_ptr_free(void *ptr)
289 if (veth_is_xdp_frame(ptr))
290 xdp_return_frame(veth_ptr_to_xdp(ptr));
295 static void __veth_xdp_flush(struct veth_rq *rq)
297 /* Write ptr_ring before reading rx_notify_masked */
299 if (!READ_ONCE(rq->rx_notify_masked) &&
300 napi_schedule_prep(&rq->xdp_napi)) {
301 WRITE_ONCE(rq->rx_notify_masked, true);
302 __napi_schedule(&rq->xdp_napi);
306 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
308 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
309 dev_kfree_skb_any(skb);
313 return NET_RX_SUCCESS;
316 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
317 struct veth_rq *rq, bool xdp)
319 return __dev_forward_skb(dev, skb) ?: xdp ?
320 veth_xdp_rx(rq, skb) :
324 /* return true if the specified skb has chances of GRO aggregation
325 * Don't strive for accuracy, but try to avoid GRO overhead in the most
327 * When XDP is enabled, all traffic is considered eligible, as the xmit
328 * device has TSO off.
329 * When TSO is enabled on the xmit device, we are likely interested only
330 * in UDP aggregation, explicitly check for that if the skb is suspected
331 * - the sock_wfree destructor is used by UDP, ICMP and XDP sockets -
332 * to belong to locally generated UDP traffic.
334 static bool veth_skb_is_eligible_for_gro(const struct net_device *dev,
335 const struct net_device *rcv,
336 const struct sk_buff *skb)
338 return !(dev->features & NETIF_F_ALL_TSO) ||
339 (skb->destructor == sock_wfree &&
340 rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
343 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
345 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
346 struct veth_rq *rq = NULL;
347 int ret = NETDEV_TX_OK;
348 struct net_device *rcv;
349 int length = skb->len;
350 bool use_napi = false;
354 rcv = rcu_dereference(priv->peer);
355 if (unlikely(!rcv) || !pskb_may_pull(skb, ETH_HLEN)) {
360 rcv_priv = netdev_priv(rcv);
361 rxq = skb_get_queue_mapping(skb);
362 if (rxq < rcv->real_num_rx_queues) {
363 rq = &rcv_priv->rq[rxq];
365 /* The napi pointer is available when an XDP program is
366 * attached or when GRO is enabled
367 * Don't bother with napi/GRO if the skb can't be aggregated
369 use_napi = rcu_access_pointer(rq->napi) &&
370 veth_skb_is_eligible_for_gro(dev, rcv, skb);
373 skb_tx_timestamp(skb);
374 if (likely(veth_forward_skb(rcv, skb, rq, use_napi) == NET_RX_SUCCESS)) {
376 dev_sw_netstats_tx_add(dev, 1, length);
378 __veth_xdp_flush(rq);
381 atomic64_inc(&priv->dropped);
390 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
392 struct veth_priv *priv = netdev_priv(dev);
395 result->peer_tq_xdp_xmit_err = 0;
396 result->xdp_packets = 0;
397 result->xdp_tx_err = 0;
398 result->xdp_bytes = 0;
399 result->rx_drops = 0;
400 for (i = 0; i < dev->num_rx_queues; i++) {
401 u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
402 struct veth_rq_stats *stats = &priv->rq[i].stats;
406 start = u64_stats_fetch_begin(&stats->syncp);
407 peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
408 xdp_tx_err = stats->vs.xdp_tx_err;
409 packets = stats->vs.xdp_packets;
410 bytes = stats->vs.xdp_bytes;
411 drops = stats->vs.rx_drops;
412 } while (u64_stats_fetch_retry(&stats->syncp, start));
413 result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
414 result->xdp_tx_err += xdp_tx_err;
415 result->xdp_packets += packets;
416 result->xdp_bytes += bytes;
417 result->rx_drops += drops;
421 static void veth_get_stats64(struct net_device *dev,
422 struct rtnl_link_stats64 *tot)
424 struct veth_priv *priv = netdev_priv(dev);
425 struct net_device *peer;
426 struct veth_stats rx;
428 tot->tx_dropped = atomic64_read(&priv->dropped);
429 dev_fetch_sw_netstats(tot, dev->tstats);
431 veth_stats_rx(&rx, dev);
432 tot->tx_dropped += rx.xdp_tx_err;
433 tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
434 tot->rx_bytes += rx.xdp_bytes;
435 tot->rx_packets += rx.xdp_packets;
438 peer = rcu_dereference(priv->peer);
440 struct rtnl_link_stats64 tot_peer = {};
442 dev_fetch_sw_netstats(&tot_peer, peer->tstats);
443 tot->rx_bytes += tot_peer.tx_bytes;
444 tot->rx_packets += tot_peer.tx_packets;
446 veth_stats_rx(&rx, peer);
447 tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
448 tot->rx_dropped += rx.xdp_tx_err;
449 tot->tx_bytes += rx.xdp_bytes;
450 tot->tx_packets += rx.xdp_packets;
455 /* fake multicast ability */
456 static void veth_set_multicast_list(struct net_device *dev)
460 static int veth_select_rxq(struct net_device *dev)
462 return smp_processor_id() % dev->real_num_rx_queues;
465 static struct net_device *veth_peer_dev(struct net_device *dev)
467 struct veth_priv *priv = netdev_priv(dev);
469 /* Callers must be under RCU read side. */
470 return rcu_dereference(priv->peer);
473 static int veth_xdp_xmit(struct net_device *dev, int n,
474 struct xdp_frame **frames,
475 u32 flags, bool ndo_xmit)
477 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
478 int i, ret = -ENXIO, nxmit = 0;
479 struct net_device *rcv;
480 unsigned int max_len;
483 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
487 rcv = rcu_dereference(priv->peer);
491 rcv_priv = netdev_priv(rcv);
492 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
493 /* The napi pointer is set if NAPI is enabled, which ensures that
494 * xdp_ring is initialized on receive side and the peer device is up.
496 if (!rcu_access_pointer(rq->napi))
499 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
501 spin_lock(&rq->xdp_ring.producer_lock);
502 for (i = 0; i < n; i++) {
503 struct xdp_frame *frame = frames[i];
504 void *ptr = veth_xdp_to_ptr(frame);
506 if (unlikely(xdp_get_frame_len(frame) > max_len ||
507 __ptr_ring_produce(&rq->xdp_ring, ptr)))
511 spin_unlock(&rq->xdp_ring.producer_lock);
513 if (flags & XDP_XMIT_FLUSH)
514 __veth_xdp_flush(rq);
518 u64_stats_update_begin(&rq->stats.syncp);
519 rq->stats.vs.peer_tq_xdp_xmit += nxmit;
520 rq->stats.vs.peer_tq_xdp_xmit_err += n - nxmit;
521 u64_stats_update_end(&rq->stats.syncp);
530 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
531 struct xdp_frame **frames, u32 flags)
535 err = veth_xdp_xmit(dev, n, frames, flags, true);
537 struct veth_priv *priv = netdev_priv(dev);
539 atomic64_add(n, &priv->dropped);
545 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
547 int sent, i, err = 0, drops;
549 sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
555 for (i = sent; unlikely(i < bq->count); i++)
556 xdp_return_frame(bq->q[i]);
558 drops = bq->count - sent;
559 trace_xdp_bulk_tx(rq->dev, sent, drops, err);
561 u64_stats_update_begin(&rq->stats.syncp);
562 rq->stats.vs.xdp_tx += sent;
563 rq->stats.vs.xdp_tx_err += drops;
564 u64_stats_update_end(&rq->stats.syncp);
569 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
571 struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
572 struct net_device *rcv;
573 struct veth_rq *rcv_rq;
576 veth_xdp_flush_bq(rq, bq);
577 rcv = rcu_dereference(priv->peer);
581 rcv_priv = netdev_priv(rcv);
582 rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
583 /* xdp_ring is initialized on receive side? */
584 if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
587 __veth_xdp_flush(rcv_rq);
592 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
593 struct veth_xdp_tx_bq *bq)
595 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
597 if (unlikely(!frame))
600 if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
601 veth_xdp_flush_bq(rq, bq);
603 bq->q[bq->count++] = frame;
608 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
609 struct xdp_frame *frame,
610 struct veth_xdp_tx_bq *bq,
611 struct veth_stats *stats)
613 struct xdp_frame orig_frame;
614 struct bpf_prog *xdp_prog;
617 xdp_prog = rcu_dereference(rq->xdp_prog);
618 if (likely(xdp_prog)) {
619 struct veth_xdp_buff vxbuf;
620 struct xdp_buff *xdp = &vxbuf.xdp;
623 xdp_convert_frame_to_buff(frame, xdp);
624 xdp->rxq = &rq->xdp_rxq;
627 act = bpf_prog_run_xdp(xdp_prog, xdp);
631 if (xdp_update_frame_from_buff(xdp, frame))
636 xdp->rxq->mem = frame->mem;
637 if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
638 trace_xdp_exception(rq->dev, xdp_prog, act);
648 xdp->rxq->mem = frame->mem;
649 if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
654 stats->xdp_redirect++;
658 bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
661 trace_xdp_exception(rq->dev, xdp_prog, act);
673 xdp_return_frame(frame);
678 /* frames array contains VETH_XDP_BATCH at most */
679 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
680 int n_xdpf, struct veth_xdp_tx_bq *bq,
681 struct veth_stats *stats)
683 void *skbs[VETH_XDP_BATCH];
686 if (xdp_alloc_skb_bulk(skbs, n_xdpf,
687 GFP_ATOMIC | __GFP_ZERO) < 0) {
688 for (i = 0; i < n_xdpf; i++)
689 xdp_return_frame(frames[i]);
690 stats->rx_drops += n_xdpf;
695 for (i = 0; i < n_xdpf; i++) {
696 struct sk_buff *skb = skbs[i];
698 skb = __xdp_build_skb_from_frame(frames[i], skb,
701 xdp_return_frame(frames[i]);
705 napi_gro_receive(&rq->xdp_napi, skb);
709 static void veth_xdp_get(struct xdp_buff *xdp)
711 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
714 get_page(virt_to_page(xdp->data));
715 if (likely(!xdp_buff_has_frags(xdp)))
718 for (i = 0; i < sinfo->nr_frags; i++)
719 __skb_frag_ref(&sinfo->frags[i]);
722 static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
723 struct xdp_buff *xdp,
724 struct sk_buff **pskb)
726 struct sk_buff *skb = *pskb;
729 if (skb_shared(skb) || skb_head_is_locked(skb) ||
730 skb_shinfo(skb)->nr_frags ||
731 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
732 if (skb_pp_cow_data(rq->page_pool, pskb, XDP_PACKET_HEADROOM))
738 /* SKB "head" area always have tailroom for skb_shared_info */
739 frame_sz = skb_end_pointer(skb) - skb->head;
740 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
741 xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
742 xdp_prepare_buff(xdp, skb->head, skb_headroom(skb),
743 skb_headlen(skb), true);
745 if (skb_is_nonlinear(skb)) {
746 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
747 xdp_buff_set_frags_flag(xdp);
749 xdp_buff_clear_frags_flag(xdp);
761 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
763 struct veth_xdp_tx_bq *bq,
764 struct veth_stats *stats)
766 void *orig_data, *orig_data_end;
767 struct bpf_prog *xdp_prog;
768 struct veth_xdp_buff vxbuf;
769 struct xdp_buff *xdp = &vxbuf.xdp;
773 skb_prepare_for_gro(skb);
776 xdp_prog = rcu_dereference(rq->xdp_prog);
777 if (unlikely(!xdp_prog)) {
782 __skb_push(skb, skb->data - skb_mac_header(skb));
783 if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
787 orig_data = xdp->data;
788 orig_data_end = xdp->data_end;
790 act = bpf_prog_run_xdp(xdp_prog, xdp);
798 xdp->rxq->mem = rq->xdp_mem;
799 if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
800 trace_xdp_exception(rq->dev, xdp_prog, act);
810 xdp->rxq->mem = rq->xdp_mem;
811 if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
815 stats->xdp_redirect++;
819 bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
822 trace_xdp_exception(rq->dev, xdp_prog, act);
830 /* check if bpf_xdp_adjust_head was used */
831 off = orig_data - xdp->data;
833 __skb_push(skb, off);
835 __skb_pull(skb, -off);
837 skb_reset_mac_header(skb);
839 /* check if bpf_xdp_adjust_tail was used */
840 off = xdp->data_end - orig_data_end;
842 __skb_put(skb, off); /* positive on grow, negative on shrink */
844 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
845 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
847 if (xdp_buff_has_frags(xdp))
848 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
852 skb->protocol = eth_type_trans(skb, rq->dev);
854 metalen = xdp->data - xdp->data_meta;
856 skb_metadata_set(skb, metalen);
867 xdp_return_buff(xdp);
872 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
873 struct veth_xdp_tx_bq *bq,
874 struct veth_stats *stats)
876 int i, done = 0, n_xdpf = 0;
877 void *xdpf[VETH_XDP_BATCH];
879 for (i = 0; i < budget; i++) {
880 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
885 if (veth_is_xdp_frame(ptr)) {
887 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
889 stats->xdp_bytes += xdp_get_frame_len(frame);
890 frame = veth_xdp_rcv_one(rq, frame, bq, stats);
893 xdpf[n_xdpf++] = frame;
894 if (n_xdpf == VETH_XDP_BATCH) {
895 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
902 struct sk_buff *skb = ptr;
904 stats->xdp_bytes += skb->len;
905 skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
907 if (skb_shared(skb) || skb_unclone(skb, GFP_ATOMIC))
908 netif_receive_skb(skb);
910 napi_gro_receive(&rq->xdp_napi, skb);
917 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
919 u64_stats_update_begin(&rq->stats.syncp);
920 rq->stats.vs.xdp_redirect += stats->xdp_redirect;
921 rq->stats.vs.xdp_bytes += stats->xdp_bytes;
922 rq->stats.vs.xdp_drops += stats->xdp_drops;
923 rq->stats.vs.rx_drops += stats->rx_drops;
924 rq->stats.vs.xdp_packets += done;
925 u64_stats_update_end(&rq->stats.syncp);
930 static int veth_poll(struct napi_struct *napi, int budget)
933 container_of(napi, struct veth_rq, xdp_napi);
934 struct veth_stats stats = {};
935 struct veth_xdp_tx_bq bq;
940 xdp_set_return_frame_no_direct();
941 done = veth_xdp_rcv(rq, budget, &bq, &stats);
943 if (stats.xdp_redirect > 0)
946 if (done < budget && napi_complete_done(napi, done)) {
947 /* Write rx_notify_masked before reading ptr_ring */
948 smp_store_mb(rq->rx_notify_masked, false);
949 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
950 if (napi_schedule_prep(&rq->xdp_napi)) {
951 WRITE_ONCE(rq->rx_notify_masked, true);
952 __napi_schedule(&rq->xdp_napi);
957 if (stats.xdp_tx > 0)
958 veth_xdp_flush(rq, &bq);
959 xdp_clear_return_frame_no_direct();
964 static int veth_create_page_pool(struct veth_rq *rq)
966 struct page_pool_params pp_params = {
968 .pool_size = VETH_RING_SIZE,
970 .dev = &rq->dev->dev,
973 rq->page_pool = page_pool_create(&pp_params);
974 if (IS_ERR(rq->page_pool)) {
975 int err = PTR_ERR(rq->page_pool);
977 rq->page_pool = NULL;
984 static int __veth_napi_enable_range(struct net_device *dev, int start, int end)
986 struct veth_priv *priv = netdev_priv(dev);
989 for (i = start; i < end; i++) {
990 err = veth_create_page_pool(&priv->rq[i]);
995 for (i = start; i < end; i++) {
996 struct veth_rq *rq = &priv->rq[i];
998 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
1003 for (i = start; i < end; i++) {
1004 struct veth_rq *rq = &priv->rq[i];
1006 napi_enable(&rq->xdp_napi);
1007 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1013 for (i--; i >= start; i--)
1014 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
1017 for (i--; i >= start; i--) {
1018 page_pool_destroy(priv->rq[i].page_pool);
1019 priv->rq[i].page_pool = NULL;
1025 static int __veth_napi_enable(struct net_device *dev)
1027 return __veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1030 static void veth_napi_del_range(struct net_device *dev, int start, int end)
1032 struct veth_priv *priv = netdev_priv(dev);
1035 for (i = start; i < end; i++) {
1036 struct veth_rq *rq = &priv->rq[i];
1038 rcu_assign_pointer(priv->rq[i].napi, NULL);
1039 napi_disable(&rq->xdp_napi);
1040 __netif_napi_del(&rq->xdp_napi);
1044 for (i = start; i < end; i++) {
1045 struct veth_rq *rq = &priv->rq[i];
1047 rq->rx_notify_masked = false;
1048 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
1051 for (i = start; i < end; i++) {
1052 page_pool_destroy(priv->rq[i].page_pool);
1053 priv->rq[i].page_pool = NULL;
1057 static void veth_napi_del(struct net_device *dev)
1059 veth_napi_del_range(dev, 0, dev->real_num_rx_queues);
1062 static bool veth_gro_requested(const struct net_device *dev)
1064 return !!(dev->wanted_features & NETIF_F_GRO);
1067 static int veth_enable_xdp_range(struct net_device *dev, int start, int end,
1068 bool napi_already_on)
1070 struct veth_priv *priv = netdev_priv(dev);
1073 for (i = start; i < end; i++) {
1074 struct veth_rq *rq = &priv->rq[i];
1076 if (!napi_already_on)
1077 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1078 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
1082 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1083 MEM_TYPE_PAGE_SHARED,
1088 /* Save original mem info as it can be overwritten */
1089 rq->xdp_mem = rq->xdp_rxq.mem;
1094 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1096 for (i--; i >= start; i--) {
1097 struct veth_rq *rq = &priv->rq[i];
1099 xdp_rxq_info_unreg(&rq->xdp_rxq);
1100 if (!napi_already_on)
1101 netif_napi_del(&rq->xdp_napi);
1107 static void veth_disable_xdp_range(struct net_device *dev, int start, int end,
1110 struct veth_priv *priv = netdev_priv(dev);
1113 for (i = start; i < end; i++) {
1114 struct veth_rq *rq = &priv->rq[i];
1116 rq->xdp_rxq.mem = rq->xdp_mem;
1117 xdp_rxq_info_unreg(&rq->xdp_rxq);
1120 netif_napi_del(&rq->xdp_napi);
1124 static int veth_enable_xdp(struct net_device *dev)
1126 bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
1127 struct veth_priv *priv = netdev_priv(dev);
1130 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
1131 err = veth_enable_xdp_range(dev, 0, dev->real_num_rx_queues, napi_already_on);
1135 if (!napi_already_on) {
1136 err = __veth_napi_enable(dev);
1138 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, true);
1144 for (i = 0; i < dev->real_num_rx_queues; i++) {
1145 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1146 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1152 static void veth_disable_xdp(struct net_device *dev)
1154 struct veth_priv *priv = netdev_priv(dev);
1157 for (i = 0; i < dev->real_num_rx_queues; i++)
1158 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1160 if (!netif_running(dev) || !veth_gro_requested(dev))
1163 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, false);
1166 static int veth_napi_enable_range(struct net_device *dev, int start, int end)
1168 struct veth_priv *priv = netdev_priv(dev);
1171 for (i = start; i < end; i++) {
1172 struct veth_rq *rq = &priv->rq[i];
1174 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1177 err = __veth_napi_enable_range(dev, start, end);
1179 for (i = start; i < end; i++) {
1180 struct veth_rq *rq = &priv->rq[i];
1182 netif_napi_del(&rq->xdp_napi);
1189 static int veth_napi_enable(struct net_device *dev)
1191 return veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1194 static void veth_disable_range_safe(struct net_device *dev, int start, int end)
1196 struct veth_priv *priv = netdev_priv(dev);
1201 if (priv->_xdp_prog) {
1202 veth_napi_del_range(dev, start, end);
1203 veth_disable_xdp_range(dev, start, end, false);
1204 } else if (veth_gro_requested(dev)) {
1205 veth_napi_del_range(dev, start, end);
1209 static int veth_enable_range_safe(struct net_device *dev, int start, int end)
1211 struct veth_priv *priv = netdev_priv(dev);
1217 if (priv->_xdp_prog) {
1218 /* these channels are freshly initialized, napi is not on there even
1219 * when GRO is requeste
1221 err = veth_enable_xdp_range(dev, start, end, false);
1225 err = __veth_napi_enable_range(dev, start, end);
1227 /* on error always delete the newly added napis */
1228 veth_disable_xdp_range(dev, start, end, true);
1231 } else if (veth_gro_requested(dev)) {
1232 return veth_napi_enable_range(dev, start, end);
1237 static void veth_set_xdp_features(struct net_device *dev)
1239 struct veth_priv *priv = netdev_priv(dev);
1240 struct net_device *peer;
1242 peer = rtnl_dereference(priv->peer);
1243 if (peer && peer->real_num_tx_queues <= dev->real_num_rx_queues) {
1244 struct veth_priv *priv_peer = netdev_priv(peer);
1245 xdp_features_t val = NETDEV_XDP_ACT_BASIC |
1246 NETDEV_XDP_ACT_REDIRECT |
1247 NETDEV_XDP_ACT_RX_SG;
1249 if (priv_peer->_xdp_prog || veth_gro_requested(peer))
1250 val |= NETDEV_XDP_ACT_NDO_XMIT |
1251 NETDEV_XDP_ACT_NDO_XMIT_SG;
1252 xdp_set_features_flag(dev, val);
1254 xdp_clear_features_flag(dev);
1258 static int veth_set_channels(struct net_device *dev,
1259 struct ethtool_channels *ch)
1261 struct veth_priv *priv = netdev_priv(dev);
1262 unsigned int old_rx_count, new_rx_count;
1263 struct veth_priv *peer_priv;
1264 struct net_device *peer;
1267 /* sanity check. Upper bounds are already enforced by the caller */
1268 if (!ch->rx_count || !ch->tx_count)
1271 /* avoid braking XDP, if that is enabled */
1272 peer = rtnl_dereference(priv->peer);
1273 peer_priv = peer ? netdev_priv(peer) : NULL;
1274 if (priv->_xdp_prog && peer && ch->rx_count < peer->real_num_tx_queues)
1277 if (peer && peer_priv && peer_priv->_xdp_prog && ch->tx_count > peer->real_num_rx_queues)
1280 old_rx_count = dev->real_num_rx_queues;
1281 new_rx_count = ch->rx_count;
1282 if (netif_running(dev)) {
1283 /* turn device off */
1284 netif_carrier_off(dev);
1286 netif_carrier_off(peer);
1288 /* try to allocate new resurces, as needed*/
1289 err = veth_enable_range_safe(dev, old_rx_count, new_rx_count);
1294 err = netif_set_real_num_rx_queues(dev, ch->rx_count);
1298 err = netif_set_real_num_tx_queues(dev, ch->tx_count);
1300 int err2 = netif_set_real_num_rx_queues(dev, old_rx_count);
1302 /* this error condition could happen only if rx and tx change
1303 * in opposite directions (e.g. tx nr raises, rx nr decreases)
1304 * and we can't do anything to fully restore the original
1308 pr_warn("Can't restore rx queues config %d -> %d %d",
1309 new_rx_count, old_rx_count, err2);
1315 if (netif_running(dev)) {
1316 /* note that we need to swap the arguments WRT the enable part
1317 * to identify the range we have to disable
1319 veth_disable_range_safe(dev, new_rx_count, old_rx_count);
1320 netif_carrier_on(dev);
1322 netif_carrier_on(peer);
1325 /* update XDP supported features */
1326 veth_set_xdp_features(dev);
1328 veth_set_xdp_features(peer);
1333 new_rx_count = old_rx_count;
1334 old_rx_count = ch->rx_count;
1338 static int veth_open(struct net_device *dev)
1340 struct veth_priv *priv = netdev_priv(dev);
1341 struct net_device *peer = rtnl_dereference(priv->peer);
1347 if (priv->_xdp_prog) {
1348 err = veth_enable_xdp(dev);
1351 } else if (veth_gro_requested(dev)) {
1352 err = veth_napi_enable(dev);
1357 if (peer->flags & IFF_UP) {
1358 netif_carrier_on(dev);
1359 netif_carrier_on(peer);
1362 veth_set_xdp_features(dev);
1367 static int veth_close(struct net_device *dev)
1369 struct veth_priv *priv = netdev_priv(dev);
1370 struct net_device *peer = rtnl_dereference(priv->peer);
1372 netif_carrier_off(dev);
1374 netif_carrier_off(peer);
1376 if (priv->_xdp_prog)
1377 veth_disable_xdp(dev);
1378 else if (veth_gro_requested(dev))
1384 static int is_valid_veth_mtu(int mtu)
1386 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1389 static int veth_alloc_queues(struct net_device *dev)
1391 struct veth_priv *priv = netdev_priv(dev);
1394 priv->rq = kvcalloc(dev->num_rx_queues, sizeof(*priv->rq),
1395 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
1399 for (i = 0; i < dev->num_rx_queues; i++) {
1400 priv->rq[i].dev = dev;
1401 u64_stats_init(&priv->rq[i].stats.syncp);
1407 static void veth_free_queues(struct net_device *dev)
1409 struct veth_priv *priv = netdev_priv(dev);
1414 static int veth_dev_init(struct net_device *dev)
1416 netdev_lockdep_set_classes(dev);
1417 return veth_alloc_queues(dev);
1420 static void veth_dev_free(struct net_device *dev)
1422 veth_free_queues(dev);
1425 #ifdef CONFIG_NET_POLL_CONTROLLER
1426 static void veth_poll_controller(struct net_device *dev)
1428 /* veth only receives frames when its peer sends one
1429 * Since it has nothing to do with disabling irqs, we are guaranteed
1430 * never to have pending data when we poll for it so
1431 * there is nothing to do here.
1433 * We need this though so netpoll recognizes us as an interface that
1434 * supports polling, which enables bridge devices in virt setups to
1435 * still use netconsole
1438 #endif /* CONFIG_NET_POLL_CONTROLLER */
1440 static int veth_get_iflink(const struct net_device *dev)
1442 struct veth_priv *priv = netdev_priv(dev);
1443 struct net_device *peer;
1447 peer = rcu_dereference(priv->peer);
1448 iflink = peer ? READ_ONCE(peer->ifindex) : 0;
1454 static netdev_features_t veth_fix_features(struct net_device *dev,
1455 netdev_features_t features)
1457 struct veth_priv *priv = netdev_priv(dev);
1458 struct net_device *peer;
1460 peer = rtnl_dereference(priv->peer);
1462 struct veth_priv *peer_priv = netdev_priv(peer);
1464 if (peer_priv->_xdp_prog)
1465 features &= ~NETIF_F_GSO_SOFTWARE;
1467 if (priv->_xdp_prog)
1468 features |= NETIF_F_GRO;
1473 static int veth_set_features(struct net_device *dev,
1474 netdev_features_t features)
1476 netdev_features_t changed = features ^ dev->features;
1477 struct veth_priv *priv = netdev_priv(dev);
1478 struct net_device *peer;
1481 if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1484 peer = rtnl_dereference(priv->peer);
1485 if (features & NETIF_F_GRO) {
1486 err = veth_napi_enable(dev);
1491 xdp_features_set_redirect_target(peer, true);
1494 xdp_features_clear_redirect_target(peer);
1500 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1502 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1503 struct net_device *peer;
1509 peer = rcu_dereference(priv->peer);
1510 if (unlikely(!peer))
1513 peer_priv = netdev_priv(peer);
1514 priv->requested_headroom = new_hr;
1515 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1516 dev->needed_headroom = new_hr;
1517 peer->needed_headroom = new_hr;
1523 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1524 struct netlink_ext_ack *extack)
1526 struct veth_priv *priv = netdev_priv(dev);
1527 struct bpf_prog *old_prog;
1528 struct net_device *peer;
1529 unsigned int max_mtu;
1532 old_prog = priv->_xdp_prog;
1533 priv->_xdp_prog = prog;
1534 peer = rtnl_dereference(priv->peer);
1538 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1543 max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
1544 peer->hard_header_len;
1545 /* Allow increasing the max_mtu if the program supports
1548 if (prog->aux->xdp_has_frags)
1549 max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
1551 if (peer->mtu > max_mtu) {
1552 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1557 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1558 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1563 if (dev->flags & IFF_UP) {
1564 err = veth_enable_xdp(dev);
1566 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1572 if (!veth_gro_requested(dev)) {
1573 /* user-space did not require GRO, but adding
1574 * XDP is supposed to get GRO working
1576 dev->features |= NETIF_F_GRO;
1577 netdev_features_change(dev);
1580 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1581 peer->max_mtu = max_mtu;
1584 xdp_features_set_redirect_target(peer, true);
1589 if (peer && !veth_gro_requested(dev))
1590 xdp_features_clear_redirect_target(peer);
1592 if (dev->flags & IFF_UP)
1593 veth_disable_xdp(dev);
1595 /* if user-space did not require GRO, since adding XDP
1596 * enabled it, clear it now
1598 if (!veth_gro_requested(dev)) {
1599 dev->features &= ~NETIF_F_GRO;
1600 netdev_features_change(dev);
1604 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1605 peer->max_mtu = ETH_MAX_MTU;
1608 bpf_prog_put(old_prog);
1611 if ((!!old_prog ^ !!prog) && peer)
1612 netdev_update_features(peer);
1616 priv->_xdp_prog = old_prog;
1621 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1623 switch (xdp->command) {
1624 case XDP_SETUP_PROG:
1625 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1631 static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
1633 struct veth_xdp_buff *_ctx = (void *)ctx;
1638 *timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
1642 static int veth_xdp_rx_hash(const struct xdp_md *ctx, u32 *hash,
1643 enum xdp_rss_hash_type *rss_type)
1645 struct veth_xdp_buff *_ctx = (void *)ctx;
1646 struct sk_buff *skb = _ctx->skb;
1651 *hash = skb_get_hash(skb);
1652 *rss_type = skb->l4_hash ? XDP_RSS_TYPE_L4_ANY : XDP_RSS_TYPE_NONE;
1657 static int veth_xdp_rx_vlan_tag(const struct xdp_md *ctx, __be16 *vlan_proto,
1660 const struct veth_xdp_buff *_ctx = (void *)ctx;
1661 const struct sk_buff *skb = _ctx->skb;
1667 err = __vlan_hwaccel_get_tag(skb, vlan_tci);
1671 *vlan_proto = skb->vlan_proto;
1675 static const struct net_device_ops veth_netdev_ops = {
1676 .ndo_init = veth_dev_init,
1677 .ndo_open = veth_open,
1678 .ndo_stop = veth_close,
1679 .ndo_start_xmit = veth_xmit,
1680 .ndo_get_stats64 = veth_get_stats64,
1681 .ndo_set_rx_mode = veth_set_multicast_list,
1682 .ndo_set_mac_address = eth_mac_addr,
1683 #ifdef CONFIG_NET_POLL_CONTROLLER
1684 .ndo_poll_controller = veth_poll_controller,
1686 .ndo_get_iflink = veth_get_iflink,
1687 .ndo_fix_features = veth_fix_features,
1688 .ndo_set_features = veth_set_features,
1689 .ndo_features_check = passthru_features_check,
1690 .ndo_set_rx_headroom = veth_set_rx_headroom,
1691 .ndo_bpf = veth_xdp,
1692 .ndo_xdp_xmit = veth_ndo_xdp_xmit,
1693 .ndo_get_peer_dev = veth_peer_dev,
1696 static const struct xdp_metadata_ops veth_xdp_metadata_ops = {
1697 .xmo_rx_timestamp = veth_xdp_rx_timestamp,
1698 .xmo_rx_hash = veth_xdp_rx_hash,
1699 .xmo_rx_vlan_tag = veth_xdp_rx_vlan_tag,
1702 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1703 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1704 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1705 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1706 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1708 static void veth_setup(struct net_device *dev)
1712 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1713 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1714 dev->priv_flags |= IFF_NO_QUEUE;
1715 dev->priv_flags |= IFF_PHONY_HEADROOM;
1717 dev->netdev_ops = &veth_netdev_ops;
1718 dev->xdp_metadata_ops = &veth_xdp_metadata_ops;
1719 dev->ethtool_ops = &veth_ethtool_ops;
1720 dev->features |= NETIF_F_LLTX;
1721 dev->features |= VETH_FEATURES;
1722 dev->vlan_features = dev->features &
1723 ~(NETIF_F_HW_VLAN_CTAG_TX |
1724 NETIF_F_HW_VLAN_STAG_TX |
1725 NETIF_F_HW_VLAN_CTAG_RX |
1726 NETIF_F_HW_VLAN_STAG_RX);
1727 dev->needs_free_netdev = true;
1728 dev->priv_destructor = veth_dev_free;
1729 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1730 dev->max_mtu = ETH_MAX_MTU;
1732 dev->hw_features = VETH_FEATURES;
1733 dev->hw_enc_features = VETH_FEATURES;
1734 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1735 netif_set_tso_max_size(dev, GSO_MAX_SIZE);
1742 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1743 struct netlink_ext_ack *extack)
1745 if (tb[IFLA_ADDRESS]) {
1746 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1748 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1749 return -EADDRNOTAVAIL;
1752 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1758 static struct rtnl_link_ops veth_link_ops;
1760 static void veth_disable_gro(struct net_device *dev)
1762 dev->features &= ~NETIF_F_GRO;
1763 dev->wanted_features &= ~NETIF_F_GRO;
1764 netdev_update_features(dev);
1767 static int veth_init_queues(struct net_device *dev, struct nlattr *tb[])
1771 if (!tb[IFLA_NUM_TX_QUEUES] && dev->num_tx_queues > 1) {
1772 err = netif_set_real_num_tx_queues(dev, 1);
1776 if (!tb[IFLA_NUM_RX_QUEUES] && dev->num_rx_queues > 1) {
1777 err = netif_set_real_num_rx_queues(dev, 1);
1784 static int veth_newlink(struct net *src_net, struct net_device *dev,
1785 struct nlattr *tb[], struct nlattr *data[],
1786 struct netlink_ext_ack *extack)
1789 struct net_device *peer;
1790 struct veth_priv *priv;
1791 char ifname[IFNAMSIZ];
1792 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1793 unsigned char name_assign_type;
1794 struct ifinfomsg *ifmp;
1798 * create and register peer first
1800 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1801 struct nlattr *nla_peer;
1803 nla_peer = data[VETH_INFO_PEER];
1804 ifmp = nla_data(nla_peer);
1805 err = rtnl_nla_parse_ifinfomsg(peer_tb, nla_peer, extack);
1809 err = veth_validate(peer_tb, NULL, extack);
1819 if (ifmp && tbp[IFLA_IFNAME]) {
1820 nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1821 name_assign_type = NET_NAME_USER;
1823 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1824 name_assign_type = NET_NAME_ENUM;
1827 net = rtnl_link_get_net(src_net, tbp);
1829 return PTR_ERR(net);
1831 peer = rtnl_create_link(net, ifname, name_assign_type,
1832 &veth_link_ops, tbp, extack);
1835 return PTR_ERR(peer);
1838 if (!ifmp || !tbp[IFLA_ADDRESS])
1839 eth_hw_addr_random(peer);
1841 if (ifmp && (dev->ifindex != 0))
1842 peer->ifindex = ifmp->ifi_index;
1844 netif_inherit_tso_max(peer, dev);
1846 err = register_netdevice(peer);
1850 goto err_register_peer;
1852 /* keep GRO disabled by default to be consistent with the established
1855 veth_disable_gro(peer);
1856 netif_carrier_off(peer);
1858 err = rtnl_configure_link(peer, ifmp, 0, NULL);
1860 goto err_configure_peer;
1865 * note, that since we've registered new device the dev's name
1866 * should be re-allocated
1869 if (tb[IFLA_ADDRESS] == NULL)
1870 eth_hw_addr_random(dev);
1872 if (tb[IFLA_IFNAME])
1873 nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1875 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1877 err = register_netdevice(dev);
1879 goto err_register_dev;
1881 netif_carrier_off(dev);
1884 * tie the deviced together
1887 priv = netdev_priv(dev);
1888 rcu_assign_pointer(priv->peer, peer);
1889 err = veth_init_queues(dev, tb);
1893 priv = netdev_priv(peer);
1894 rcu_assign_pointer(priv->peer, dev);
1895 err = veth_init_queues(peer, tb);
1899 veth_disable_gro(dev);
1900 /* update XDP supported features */
1901 veth_set_xdp_features(dev);
1902 veth_set_xdp_features(peer);
1907 unregister_netdevice(dev);
1911 unregister_netdevice(peer);
1919 static void veth_dellink(struct net_device *dev, struct list_head *head)
1921 struct veth_priv *priv;
1922 struct net_device *peer;
1924 priv = netdev_priv(dev);
1925 peer = rtnl_dereference(priv->peer);
1927 /* Note : dellink() is called from default_device_exit_batch(),
1928 * before a rcu_synchronize() point. The devices are guaranteed
1929 * not being freed before one RCU grace period.
1931 RCU_INIT_POINTER(priv->peer, NULL);
1932 unregister_netdevice_queue(dev, head);
1935 priv = netdev_priv(peer);
1936 RCU_INIT_POINTER(priv->peer, NULL);
1937 unregister_netdevice_queue(peer, head);
1941 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1942 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1945 static struct net *veth_get_link_net(const struct net_device *dev)
1947 struct veth_priv *priv = netdev_priv(dev);
1948 struct net_device *peer = rtnl_dereference(priv->peer);
1950 return peer ? dev_net(peer) : dev_net(dev);
1953 static unsigned int veth_get_num_queues(void)
1955 /* enforce the same queue limit as rtnl_create_link */
1956 int queues = num_possible_cpus();
1963 static struct rtnl_link_ops veth_link_ops = {
1965 .priv_size = sizeof(struct veth_priv),
1966 .setup = veth_setup,
1967 .validate = veth_validate,
1968 .newlink = veth_newlink,
1969 .dellink = veth_dellink,
1970 .policy = veth_policy,
1971 .maxtype = VETH_INFO_MAX,
1972 .get_link_net = veth_get_link_net,
1973 .get_num_tx_queues = veth_get_num_queues,
1974 .get_num_rx_queues = veth_get_num_queues,
1981 static __init int veth_init(void)
1983 return rtnl_link_register(&veth_link_ops);
1986 static __exit void veth_exit(void)
1988 rtnl_link_unregister(&veth_link_ops);
1991 module_init(veth_init);
1992 module_exit(veth_exit);
1994 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1995 MODULE_LICENSE("GPL v2");
1996 MODULE_ALIAS_RTNL_LINK(DRV_NAME);