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 u32 size, len, max_head_size, off;
733 struct sk_buff *nskb;
737 /* We need a private copy of the skb and data buffers since
738 * the ebpf program can modify it. We segment the original skb
739 * into order-0 pages without linearize it.
741 * Make sure we have enough space for linear and paged area
743 max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE -
745 if (skb->len > PAGE_SIZE * MAX_SKB_FRAGS + max_head_size)
748 /* Allocate skb head */
749 page = page_pool_dev_alloc_pages(rq->page_pool);
753 nskb = napi_build_skb(page_address(page), PAGE_SIZE);
755 page_pool_put_full_page(rq->page_pool, page, true);
759 skb_reserve(nskb, VETH_XDP_HEADROOM);
760 skb_copy_header(nskb, skb);
761 skb_mark_for_recycle(nskb);
763 size = min_t(u32, skb->len, max_head_size);
764 if (skb_copy_bits(skb, 0, nskb->data, size)) {
770 head_off = skb_headroom(nskb) - skb_headroom(skb);
771 skb_headers_offset_update(nskb, head_off);
773 /* Allocate paged area of new skb */
775 len = skb->len - off;
777 for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
778 page = page_pool_dev_alloc_pages(rq->page_pool);
784 size = min_t(u32, len, PAGE_SIZE);
785 skb_add_rx_frag(nskb, i, page, 0, size, PAGE_SIZE);
786 if (skb_copy_bits(skb, off, page_address(page),
800 /* SKB "head" area always have tailroom for skb_shared_info */
801 frame_sz = skb_end_pointer(skb) - skb->head;
802 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
803 xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
804 xdp_prepare_buff(xdp, skb->head, skb_headroom(skb),
805 skb_headlen(skb), true);
807 if (skb_is_nonlinear(skb)) {
808 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
809 xdp_buff_set_frags_flag(xdp);
811 xdp_buff_clear_frags_flag(xdp);
823 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
825 struct veth_xdp_tx_bq *bq,
826 struct veth_stats *stats)
828 void *orig_data, *orig_data_end;
829 struct bpf_prog *xdp_prog;
830 struct veth_xdp_buff vxbuf;
831 struct xdp_buff *xdp = &vxbuf.xdp;
835 skb_prepare_for_gro(skb);
838 xdp_prog = rcu_dereference(rq->xdp_prog);
839 if (unlikely(!xdp_prog)) {
844 __skb_push(skb, skb->data - skb_mac_header(skb));
845 if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
849 orig_data = xdp->data;
850 orig_data_end = xdp->data_end;
852 act = bpf_prog_run_xdp(xdp_prog, xdp);
860 xdp->rxq->mem = rq->xdp_mem;
861 if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
862 trace_xdp_exception(rq->dev, xdp_prog, act);
872 xdp->rxq->mem = rq->xdp_mem;
873 if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
877 stats->xdp_redirect++;
881 bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
884 trace_xdp_exception(rq->dev, xdp_prog, act);
892 /* check if bpf_xdp_adjust_head was used */
893 off = orig_data - xdp->data;
895 __skb_push(skb, off);
897 __skb_pull(skb, -off);
899 skb_reset_mac_header(skb);
901 /* check if bpf_xdp_adjust_tail was used */
902 off = xdp->data_end - orig_data_end;
904 __skb_put(skb, off); /* positive on grow, negative on shrink */
906 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
907 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
909 if (xdp_buff_has_frags(xdp))
910 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
914 skb->protocol = eth_type_trans(skb, rq->dev);
916 metalen = xdp->data - xdp->data_meta;
918 skb_metadata_set(skb, metalen);
929 xdp_return_buff(xdp);
934 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
935 struct veth_xdp_tx_bq *bq,
936 struct veth_stats *stats)
938 int i, done = 0, n_xdpf = 0;
939 void *xdpf[VETH_XDP_BATCH];
941 for (i = 0; i < budget; i++) {
942 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
947 if (veth_is_xdp_frame(ptr)) {
949 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
951 stats->xdp_bytes += xdp_get_frame_len(frame);
952 frame = veth_xdp_rcv_one(rq, frame, bq, stats);
955 xdpf[n_xdpf++] = frame;
956 if (n_xdpf == VETH_XDP_BATCH) {
957 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
964 struct sk_buff *skb = ptr;
966 stats->xdp_bytes += skb->len;
967 skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
969 if (skb_shared(skb) || skb_unclone(skb, GFP_ATOMIC))
970 netif_receive_skb(skb);
972 napi_gro_receive(&rq->xdp_napi, skb);
979 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
981 u64_stats_update_begin(&rq->stats.syncp);
982 rq->stats.vs.xdp_redirect += stats->xdp_redirect;
983 rq->stats.vs.xdp_bytes += stats->xdp_bytes;
984 rq->stats.vs.xdp_drops += stats->xdp_drops;
985 rq->stats.vs.rx_drops += stats->rx_drops;
986 rq->stats.vs.xdp_packets += done;
987 u64_stats_update_end(&rq->stats.syncp);
992 static int veth_poll(struct napi_struct *napi, int budget)
995 container_of(napi, struct veth_rq, xdp_napi);
996 struct veth_stats stats = {};
997 struct veth_xdp_tx_bq bq;
1002 xdp_set_return_frame_no_direct();
1003 done = veth_xdp_rcv(rq, budget, &bq, &stats);
1005 if (stats.xdp_redirect > 0)
1008 if (done < budget && napi_complete_done(napi, done)) {
1009 /* Write rx_notify_masked before reading ptr_ring */
1010 smp_store_mb(rq->rx_notify_masked, false);
1011 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
1012 if (napi_schedule_prep(&rq->xdp_napi)) {
1013 WRITE_ONCE(rq->rx_notify_masked, true);
1014 __napi_schedule(&rq->xdp_napi);
1019 if (stats.xdp_tx > 0)
1020 veth_xdp_flush(rq, &bq);
1021 xdp_clear_return_frame_no_direct();
1026 static int veth_create_page_pool(struct veth_rq *rq)
1028 struct page_pool_params pp_params = {
1030 .pool_size = VETH_RING_SIZE,
1031 .nid = NUMA_NO_NODE,
1032 .dev = &rq->dev->dev,
1035 rq->page_pool = page_pool_create(&pp_params);
1036 if (IS_ERR(rq->page_pool)) {
1037 int err = PTR_ERR(rq->page_pool);
1039 rq->page_pool = NULL;
1046 static int __veth_napi_enable_range(struct net_device *dev, int start, int end)
1048 struct veth_priv *priv = netdev_priv(dev);
1051 for (i = start; i < end; i++) {
1052 err = veth_create_page_pool(&priv->rq[i]);
1057 for (i = start; i < end; i++) {
1058 struct veth_rq *rq = &priv->rq[i];
1060 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
1065 for (i = start; i < end; i++) {
1066 struct veth_rq *rq = &priv->rq[i];
1068 napi_enable(&rq->xdp_napi);
1069 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1075 for (i--; i >= start; i--)
1076 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
1079 for (i--; i >= start; i--) {
1080 page_pool_destroy(priv->rq[i].page_pool);
1081 priv->rq[i].page_pool = NULL;
1087 static int __veth_napi_enable(struct net_device *dev)
1089 return __veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1092 static void veth_napi_del_range(struct net_device *dev, int start, int end)
1094 struct veth_priv *priv = netdev_priv(dev);
1097 for (i = start; i < end; i++) {
1098 struct veth_rq *rq = &priv->rq[i];
1100 rcu_assign_pointer(priv->rq[i].napi, NULL);
1101 napi_disable(&rq->xdp_napi);
1102 __netif_napi_del(&rq->xdp_napi);
1106 for (i = start; i < end; i++) {
1107 struct veth_rq *rq = &priv->rq[i];
1109 rq->rx_notify_masked = false;
1110 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
1113 for (i = start; i < end; i++) {
1114 page_pool_destroy(priv->rq[i].page_pool);
1115 priv->rq[i].page_pool = NULL;
1119 static void veth_napi_del(struct net_device *dev)
1121 veth_napi_del_range(dev, 0, dev->real_num_rx_queues);
1124 static bool veth_gro_requested(const struct net_device *dev)
1126 return !!(dev->wanted_features & NETIF_F_GRO);
1129 static int veth_enable_xdp_range(struct net_device *dev, int start, int end,
1130 bool napi_already_on)
1132 struct veth_priv *priv = netdev_priv(dev);
1135 for (i = start; i < end; i++) {
1136 struct veth_rq *rq = &priv->rq[i];
1138 if (!napi_already_on)
1139 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1140 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
1144 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1145 MEM_TYPE_PAGE_SHARED,
1150 /* Save original mem info as it can be overwritten */
1151 rq->xdp_mem = rq->xdp_rxq.mem;
1156 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1158 for (i--; i >= start; i--) {
1159 struct veth_rq *rq = &priv->rq[i];
1161 xdp_rxq_info_unreg(&rq->xdp_rxq);
1162 if (!napi_already_on)
1163 netif_napi_del(&rq->xdp_napi);
1169 static void veth_disable_xdp_range(struct net_device *dev, int start, int end,
1172 struct veth_priv *priv = netdev_priv(dev);
1175 for (i = start; i < end; i++) {
1176 struct veth_rq *rq = &priv->rq[i];
1178 rq->xdp_rxq.mem = rq->xdp_mem;
1179 xdp_rxq_info_unreg(&rq->xdp_rxq);
1182 netif_napi_del(&rq->xdp_napi);
1186 static int veth_enable_xdp(struct net_device *dev)
1188 bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
1189 struct veth_priv *priv = netdev_priv(dev);
1192 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
1193 err = veth_enable_xdp_range(dev, 0, dev->real_num_rx_queues, napi_already_on);
1197 if (!napi_already_on) {
1198 err = __veth_napi_enable(dev);
1200 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, true);
1204 if (!veth_gro_requested(dev)) {
1205 /* user-space did not require GRO, but adding XDP
1206 * is supposed to get GRO working
1208 dev->features |= NETIF_F_GRO;
1209 netdev_features_change(dev);
1214 for (i = 0; i < dev->real_num_rx_queues; i++) {
1215 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1216 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1222 static void veth_disable_xdp(struct net_device *dev)
1224 struct veth_priv *priv = netdev_priv(dev);
1227 for (i = 0; i < dev->real_num_rx_queues; i++)
1228 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1230 if (!netif_running(dev) || !veth_gro_requested(dev)) {
1233 /* if user-space did not require GRO, since adding XDP
1234 * enabled it, clear it now
1236 if (!veth_gro_requested(dev) && netif_running(dev)) {
1237 dev->features &= ~NETIF_F_GRO;
1238 netdev_features_change(dev);
1242 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, false);
1245 static int veth_napi_enable_range(struct net_device *dev, int start, int end)
1247 struct veth_priv *priv = netdev_priv(dev);
1250 for (i = start; i < end; i++) {
1251 struct veth_rq *rq = &priv->rq[i];
1253 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1256 err = __veth_napi_enable_range(dev, start, end);
1258 for (i = start; i < end; i++) {
1259 struct veth_rq *rq = &priv->rq[i];
1261 netif_napi_del(&rq->xdp_napi);
1268 static int veth_napi_enable(struct net_device *dev)
1270 return veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1273 static void veth_disable_range_safe(struct net_device *dev, int start, int end)
1275 struct veth_priv *priv = netdev_priv(dev);
1280 if (priv->_xdp_prog) {
1281 veth_napi_del_range(dev, start, end);
1282 veth_disable_xdp_range(dev, start, end, false);
1283 } else if (veth_gro_requested(dev)) {
1284 veth_napi_del_range(dev, start, end);
1288 static int veth_enable_range_safe(struct net_device *dev, int start, int end)
1290 struct veth_priv *priv = netdev_priv(dev);
1296 if (priv->_xdp_prog) {
1297 /* these channels are freshly initialized, napi is not on there even
1298 * when GRO is requeste
1300 err = veth_enable_xdp_range(dev, start, end, false);
1304 err = __veth_napi_enable_range(dev, start, end);
1306 /* on error always delete the newly added napis */
1307 veth_disable_xdp_range(dev, start, end, true);
1310 } else if (veth_gro_requested(dev)) {
1311 return veth_napi_enable_range(dev, start, end);
1316 static void veth_set_xdp_features(struct net_device *dev)
1318 struct veth_priv *priv = netdev_priv(dev);
1319 struct net_device *peer;
1321 peer = rtnl_dereference(priv->peer);
1322 if (peer && peer->real_num_tx_queues <= dev->real_num_rx_queues) {
1323 struct veth_priv *priv_peer = netdev_priv(peer);
1324 xdp_features_t val = NETDEV_XDP_ACT_BASIC |
1325 NETDEV_XDP_ACT_REDIRECT |
1326 NETDEV_XDP_ACT_RX_SG;
1328 if (priv_peer->_xdp_prog || veth_gro_requested(peer))
1329 val |= NETDEV_XDP_ACT_NDO_XMIT |
1330 NETDEV_XDP_ACT_NDO_XMIT_SG;
1331 xdp_set_features_flag(dev, val);
1333 xdp_clear_features_flag(dev);
1337 static int veth_set_channels(struct net_device *dev,
1338 struct ethtool_channels *ch)
1340 struct veth_priv *priv = netdev_priv(dev);
1341 unsigned int old_rx_count, new_rx_count;
1342 struct veth_priv *peer_priv;
1343 struct net_device *peer;
1346 /* sanity check. Upper bounds are already enforced by the caller */
1347 if (!ch->rx_count || !ch->tx_count)
1350 /* avoid braking XDP, if that is enabled */
1351 peer = rtnl_dereference(priv->peer);
1352 peer_priv = peer ? netdev_priv(peer) : NULL;
1353 if (priv->_xdp_prog && peer && ch->rx_count < peer->real_num_tx_queues)
1356 if (peer && peer_priv && peer_priv->_xdp_prog && ch->tx_count > peer->real_num_rx_queues)
1359 old_rx_count = dev->real_num_rx_queues;
1360 new_rx_count = ch->rx_count;
1361 if (netif_running(dev)) {
1362 /* turn device off */
1363 netif_carrier_off(dev);
1365 netif_carrier_off(peer);
1367 /* try to allocate new resurces, as needed*/
1368 err = veth_enable_range_safe(dev, old_rx_count, new_rx_count);
1373 err = netif_set_real_num_rx_queues(dev, ch->rx_count);
1377 err = netif_set_real_num_tx_queues(dev, ch->tx_count);
1379 int err2 = netif_set_real_num_rx_queues(dev, old_rx_count);
1381 /* this error condition could happen only if rx and tx change
1382 * in opposite directions (e.g. tx nr raises, rx nr decreases)
1383 * and we can't do anything to fully restore the original
1387 pr_warn("Can't restore rx queues config %d -> %d %d",
1388 new_rx_count, old_rx_count, err2);
1394 if (netif_running(dev)) {
1395 /* note that we need to swap the arguments WRT the enable part
1396 * to identify the range we have to disable
1398 veth_disable_range_safe(dev, new_rx_count, old_rx_count);
1399 netif_carrier_on(dev);
1401 netif_carrier_on(peer);
1404 /* update XDP supported features */
1405 veth_set_xdp_features(dev);
1407 veth_set_xdp_features(peer);
1412 new_rx_count = old_rx_count;
1413 old_rx_count = ch->rx_count;
1417 static int veth_open(struct net_device *dev)
1419 struct veth_priv *priv = netdev_priv(dev);
1420 struct net_device *peer = rtnl_dereference(priv->peer);
1426 if (priv->_xdp_prog) {
1427 err = veth_enable_xdp(dev);
1430 } else if (veth_gro_requested(dev)) {
1431 err = veth_napi_enable(dev);
1436 if (peer->flags & IFF_UP) {
1437 netif_carrier_on(dev);
1438 netif_carrier_on(peer);
1441 veth_set_xdp_features(dev);
1446 static int veth_close(struct net_device *dev)
1448 struct veth_priv *priv = netdev_priv(dev);
1449 struct net_device *peer = rtnl_dereference(priv->peer);
1451 netif_carrier_off(dev);
1453 netif_carrier_off(peer);
1455 if (priv->_xdp_prog)
1456 veth_disable_xdp(dev);
1457 else if (veth_gro_requested(dev))
1463 static int is_valid_veth_mtu(int mtu)
1465 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1468 static int veth_alloc_queues(struct net_device *dev)
1470 struct veth_priv *priv = netdev_priv(dev);
1473 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL_ACCOUNT);
1477 for (i = 0; i < dev->num_rx_queues; i++) {
1478 priv->rq[i].dev = dev;
1479 u64_stats_init(&priv->rq[i].stats.syncp);
1485 static void veth_free_queues(struct net_device *dev)
1487 struct veth_priv *priv = netdev_priv(dev);
1492 static int veth_dev_init(struct net_device *dev)
1494 return veth_alloc_queues(dev);
1497 static void veth_dev_free(struct net_device *dev)
1499 veth_free_queues(dev);
1502 #ifdef CONFIG_NET_POLL_CONTROLLER
1503 static void veth_poll_controller(struct net_device *dev)
1505 /* veth only receives frames when its peer sends one
1506 * Since it has nothing to do with disabling irqs, we are guaranteed
1507 * never to have pending data when we poll for it so
1508 * there is nothing to do here.
1510 * We need this though so netpoll recognizes us as an interface that
1511 * supports polling, which enables bridge devices in virt setups to
1512 * still use netconsole
1515 #endif /* CONFIG_NET_POLL_CONTROLLER */
1517 static int veth_get_iflink(const struct net_device *dev)
1519 struct veth_priv *priv = netdev_priv(dev);
1520 struct net_device *peer;
1524 peer = rcu_dereference(priv->peer);
1525 iflink = peer ? peer->ifindex : 0;
1531 static netdev_features_t veth_fix_features(struct net_device *dev,
1532 netdev_features_t features)
1534 struct veth_priv *priv = netdev_priv(dev);
1535 struct net_device *peer;
1537 peer = rtnl_dereference(priv->peer);
1539 struct veth_priv *peer_priv = netdev_priv(peer);
1541 if (peer_priv->_xdp_prog)
1542 features &= ~NETIF_F_GSO_SOFTWARE;
1544 if (priv->_xdp_prog)
1545 features |= NETIF_F_GRO;
1550 static int veth_set_features(struct net_device *dev,
1551 netdev_features_t features)
1553 netdev_features_t changed = features ^ dev->features;
1554 struct veth_priv *priv = netdev_priv(dev);
1555 struct net_device *peer;
1558 if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1561 peer = rtnl_dereference(priv->peer);
1562 if (features & NETIF_F_GRO) {
1563 err = veth_napi_enable(dev);
1568 xdp_features_set_redirect_target(peer, true);
1571 xdp_features_clear_redirect_target(peer);
1577 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1579 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1580 struct net_device *peer;
1586 peer = rcu_dereference(priv->peer);
1587 if (unlikely(!peer))
1590 peer_priv = netdev_priv(peer);
1591 priv->requested_headroom = new_hr;
1592 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1593 dev->needed_headroom = new_hr;
1594 peer->needed_headroom = new_hr;
1600 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1601 struct netlink_ext_ack *extack)
1603 struct veth_priv *priv = netdev_priv(dev);
1604 struct bpf_prog *old_prog;
1605 struct net_device *peer;
1606 unsigned int max_mtu;
1609 old_prog = priv->_xdp_prog;
1610 priv->_xdp_prog = prog;
1611 peer = rtnl_dereference(priv->peer);
1615 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1620 max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
1621 peer->hard_header_len;
1622 /* Allow increasing the max_mtu if the program supports
1625 if (prog->aux->xdp_has_frags)
1626 max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
1628 if (peer->mtu > max_mtu) {
1629 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1634 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1635 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1640 if (dev->flags & IFF_UP) {
1641 err = veth_enable_xdp(dev);
1643 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1649 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1650 peer->max_mtu = max_mtu;
1653 xdp_features_set_redirect_target(peer, true);
1658 if (peer && !veth_gro_requested(dev))
1659 xdp_features_clear_redirect_target(peer);
1661 if (dev->flags & IFF_UP)
1662 veth_disable_xdp(dev);
1665 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1666 peer->max_mtu = ETH_MAX_MTU;
1669 bpf_prog_put(old_prog);
1672 if ((!!old_prog ^ !!prog) && peer)
1673 netdev_update_features(peer);
1677 priv->_xdp_prog = old_prog;
1682 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1684 switch (xdp->command) {
1685 case XDP_SETUP_PROG:
1686 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1692 static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
1694 struct veth_xdp_buff *_ctx = (void *)ctx;
1699 *timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
1703 static int veth_xdp_rx_hash(const struct xdp_md *ctx, u32 *hash,
1704 enum xdp_rss_hash_type *rss_type)
1706 struct veth_xdp_buff *_ctx = (void *)ctx;
1707 struct sk_buff *skb = _ctx->skb;
1712 *hash = skb_get_hash(skb);
1713 *rss_type = skb->l4_hash ? XDP_RSS_TYPE_L4_ANY : XDP_RSS_TYPE_NONE;
1718 static const struct net_device_ops veth_netdev_ops = {
1719 .ndo_init = veth_dev_init,
1720 .ndo_open = veth_open,
1721 .ndo_stop = veth_close,
1722 .ndo_start_xmit = veth_xmit,
1723 .ndo_get_stats64 = veth_get_stats64,
1724 .ndo_set_rx_mode = veth_set_multicast_list,
1725 .ndo_set_mac_address = eth_mac_addr,
1726 #ifdef CONFIG_NET_POLL_CONTROLLER
1727 .ndo_poll_controller = veth_poll_controller,
1729 .ndo_get_iflink = veth_get_iflink,
1730 .ndo_fix_features = veth_fix_features,
1731 .ndo_set_features = veth_set_features,
1732 .ndo_features_check = passthru_features_check,
1733 .ndo_set_rx_headroom = veth_set_rx_headroom,
1734 .ndo_bpf = veth_xdp,
1735 .ndo_xdp_xmit = veth_ndo_xdp_xmit,
1736 .ndo_get_peer_dev = veth_peer_dev,
1739 static const struct xdp_metadata_ops veth_xdp_metadata_ops = {
1740 .xmo_rx_timestamp = veth_xdp_rx_timestamp,
1741 .xmo_rx_hash = veth_xdp_rx_hash,
1744 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1745 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1746 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1747 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1748 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1750 static void veth_setup(struct net_device *dev)
1754 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1755 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1756 dev->priv_flags |= IFF_NO_QUEUE;
1757 dev->priv_flags |= IFF_PHONY_HEADROOM;
1759 dev->netdev_ops = &veth_netdev_ops;
1760 dev->xdp_metadata_ops = &veth_xdp_metadata_ops;
1761 dev->ethtool_ops = &veth_ethtool_ops;
1762 dev->features |= NETIF_F_LLTX;
1763 dev->features |= VETH_FEATURES;
1764 dev->vlan_features = dev->features &
1765 ~(NETIF_F_HW_VLAN_CTAG_TX |
1766 NETIF_F_HW_VLAN_STAG_TX |
1767 NETIF_F_HW_VLAN_CTAG_RX |
1768 NETIF_F_HW_VLAN_STAG_RX);
1769 dev->needs_free_netdev = true;
1770 dev->priv_destructor = veth_dev_free;
1771 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1772 dev->max_mtu = ETH_MAX_MTU;
1774 dev->hw_features = VETH_FEATURES;
1775 dev->hw_enc_features = VETH_FEATURES;
1776 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1777 netif_set_tso_max_size(dev, GSO_MAX_SIZE);
1784 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1785 struct netlink_ext_ack *extack)
1787 if (tb[IFLA_ADDRESS]) {
1788 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1790 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1791 return -EADDRNOTAVAIL;
1794 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1800 static struct rtnl_link_ops veth_link_ops;
1802 static void veth_disable_gro(struct net_device *dev)
1804 dev->features &= ~NETIF_F_GRO;
1805 dev->wanted_features &= ~NETIF_F_GRO;
1806 netdev_update_features(dev);
1809 static int veth_init_queues(struct net_device *dev, struct nlattr *tb[])
1813 if (!tb[IFLA_NUM_TX_QUEUES] && dev->num_tx_queues > 1) {
1814 err = netif_set_real_num_tx_queues(dev, 1);
1818 if (!tb[IFLA_NUM_RX_QUEUES] && dev->num_rx_queues > 1) {
1819 err = netif_set_real_num_rx_queues(dev, 1);
1826 static int veth_newlink(struct net *src_net, struct net_device *dev,
1827 struct nlattr *tb[], struct nlattr *data[],
1828 struct netlink_ext_ack *extack)
1831 struct net_device *peer;
1832 struct veth_priv *priv;
1833 char ifname[IFNAMSIZ];
1834 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1835 unsigned char name_assign_type;
1836 struct ifinfomsg *ifmp;
1840 * create and register peer first
1842 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1843 struct nlattr *nla_peer;
1845 nla_peer = data[VETH_INFO_PEER];
1846 ifmp = nla_data(nla_peer);
1847 err = rtnl_nla_parse_ifinfomsg(peer_tb, nla_peer, extack);
1851 err = veth_validate(peer_tb, NULL, extack);
1861 if (ifmp && tbp[IFLA_IFNAME]) {
1862 nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1863 name_assign_type = NET_NAME_USER;
1865 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1866 name_assign_type = NET_NAME_ENUM;
1869 net = rtnl_link_get_net(src_net, tbp);
1871 return PTR_ERR(net);
1873 peer = rtnl_create_link(net, ifname, name_assign_type,
1874 &veth_link_ops, tbp, extack);
1877 return PTR_ERR(peer);
1880 if (!ifmp || !tbp[IFLA_ADDRESS])
1881 eth_hw_addr_random(peer);
1883 if (ifmp && (dev->ifindex != 0))
1884 peer->ifindex = ifmp->ifi_index;
1886 netif_inherit_tso_max(peer, dev);
1888 err = register_netdevice(peer);
1892 goto err_register_peer;
1894 /* keep GRO disabled by default to be consistent with the established
1897 veth_disable_gro(peer);
1898 netif_carrier_off(peer);
1900 err = rtnl_configure_link(peer, ifmp, 0, NULL);
1902 goto err_configure_peer;
1907 * note, that since we've registered new device the dev's name
1908 * should be re-allocated
1911 if (tb[IFLA_ADDRESS] == NULL)
1912 eth_hw_addr_random(dev);
1914 if (tb[IFLA_IFNAME])
1915 nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1917 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1919 err = register_netdevice(dev);
1921 goto err_register_dev;
1923 netif_carrier_off(dev);
1926 * tie the deviced together
1929 priv = netdev_priv(dev);
1930 rcu_assign_pointer(priv->peer, peer);
1931 err = veth_init_queues(dev, tb);
1935 priv = netdev_priv(peer);
1936 rcu_assign_pointer(priv->peer, dev);
1937 err = veth_init_queues(peer, tb);
1941 veth_disable_gro(dev);
1942 /* update XDP supported features */
1943 veth_set_xdp_features(dev);
1944 veth_set_xdp_features(peer);
1949 unregister_netdevice(dev);
1953 unregister_netdevice(peer);
1961 static void veth_dellink(struct net_device *dev, struct list_head *head)
1963 struct veth_priv *priv;
1964 struct net_device *peer;
1966 priv = netdev_priv(dev);
1967 peer = rtnl_dereference(priv->peer);
1969 /* Note : dellink() is called from default_device_exit_batch(),
1970 * before a rcu_synchronize() point. The devices are guaranteed
1971 * not being freed before one RCU grace period.
1973 RCU_INIT_POINTER(priv->peer, NULL);
1974 unregister_netdevice_queue(dev, head);
1977 priv = netdev_priv(peer);
1978 RCU_INIT_POINTER(priv->peer, NULL);
1979 unregister_netdevice_queue(peer, head);
1983 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1984 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1987 static struct net *veth_get_link_net(const struct net_device *dev)
1989 struct veth_priv *priv = netdev_priv(dev);
1990 struct net_device *peer = rtnl_dereference(priv->peer);
1992 return peer ? dev_net(peer) : dev_net(dev);
1995 static unsigned int veth_get_num_queues(void)
1997 /* enforce the same queue limit as rtnl_create_link */
1998 int queues = num_possible_cpus();
2005 static struct rtnl_link_ops veth_link_ops = {
2007 .priv_size = sizeof(struct veth_priv),
2008 .setup = veth_setup,
2009 .validate = veth_validate,
2010 .newlink = veth_newlink,
2011 .dellink = veth_dellink,
2012 .policy = veth_policy,
2013 .maxtype = VETH_INFO_MAX,
2014 .get_link_net = veth_get_link_net,
2015 .get_num_tx_queues = veth_get_num_queues,
2016 .get_num_rx_queues = veth_get_num_queues,
2023 static __init int veth_init(void)
2025 return rtnl_link_register(&veth_link_ops);
2028 static __exit void veth_exit(void)
2030 rtnl_link_unregister(&veth_link_ops);
2033 module_init(veth_init);
2034 module_exit(veth_exit);
2036 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
2037 MODULE_LICENSE("GPL v2");
2038 MODULE_ALIAS_RTNL_LINK(DRV_NAME);