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 #define VETH_XDP_TX_BULK_SIZE 16
49 u64 peer_tq_xdp_xmit_err;
52 struct veth_rq_stats {
54 struct u64_stats_sync syncp;
58 struct napi_struct xdp_napi;
59 struct net_device *dev;
60 struct bpf_prog __rcu *xdp_prog;
61 struct xdp_mem_info xdp_mem;
62 struct veth_rq_stats stats;
63 bool rx_notify_masked;
64 struct ptr_ring xdp_ring;
65 struct xdp_rxq_info xdp_rxq;
69 struct net_device __rcu *peer;
71 struct bpf_prog *_xdp_prog;
73 unsigned int requested_headroom;
76 struct veth_xdp_tx_bq {
77 struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
85 struct veth_q_stat_desc {
86 char desc[ETH_GSTRING_LEN];
90 #define VETH_RQ_STAT(m) offsetof(struct veth_stats, m)
92 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
93 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
94 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
95 { "drops", VETH_RQ_STAT(rx_drops) },
96 { "xdp_redirect", VETH_RQ_STAT(xdp_redirect) },
97 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
98 { "xdp_tx", VETH_RQ_STAT(xdp_tx) },
99 { "xdp_tx_errors", VETH_RQ_STAT(xdp_tx_err) },
102 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
104 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
105 { "xdp_xmit", VETH_RQ_STAT(peer_tq_xdp_xmit) },
106 { "xdp_xmit_errors", VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
109 #define VETH_TQ_STATS_LEN ARRAY_SIZE(veth_tq_stats_desc)
112 const char string[ETH_GSTRING_LEN];
113 } ethtool_stats_keys[] = {
117 static int veth_get_link_ksettings(struct net_device *dev,
118 struct ethtool_link_ksettings *cmd)
120 cmd->base.speed = SPEED_10000;
121 cmd->base.duplex = DUPLEX_FULL;
122 cmd->base.port = PORT_TP;
123 cmd->base.autoneg = AUTONEG_DISABLE;
127 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
129 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
130 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
133 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
135 char *p = (char *)buf;
140 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys));
141 p += sizeof(ethtool_stats_keys);
142 for (i = 0; i < dev->real_num_rx_queues; i++) {
143 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
144 snprintf(p, ETH_GSTRING_LEN,
146 i, veth_rq_stats_desc[j].desc);
147 p += ETH_GSTRING_LEN;
150 for (i = 0; i < dev->real_num_tx_queues; i++) {
151 for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
152 snprintf(p, ETH_GSTRING_LEN,
154 i, veth_tq_stats_desc[j].desc);
155 p += ETH_GSTRING_LEN;
162 static int veth_get_sset_count(struct net_device *dev, int sset)
166 return ARRAY_SIZE(ethtool_stats_keys) +
167 VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
168 VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
174 static void veth_get_ethtool_stats(struct net_device *dev,
175 struct ethtool_stats *stats, u64 *data)
177 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
178 struct net_device *peer = rtnl_dereference(priv->peer);
181 data[0] = peer ? peer->ifindex : 0;
183 for (i = 0; i < dev->real_num_rx_queues; i++) {
184 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
185 const void *stats_base = (void *)&rq_stats->vs;
190 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
191 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
192 offset = veth_rq_stats_desc[j].offset;
193 data[idx + j] = *(u64 *)(stats_base + offset);
195 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
196 idx += VETH_RQ_STATS_LEN;
202 rcv_priv = netdev_priv(peer);
203 for (i = 0; i < peer->real_num_rx_queues; i++) {
204 const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
205 const void *base = (void *)&rq_stats->vs;
206 unsigned int start, tx_idx = idx;
209 tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
211 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
212 for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
213 offset = veth_tq_stats_desc[j].offset;
214 data[tx_idx + j] += *(u64 *)(base + offset);
216 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
220 static const struct ethtool_ops veth_ethtool_ops = {
221 .get_drvinfo = veth_get_drvinfo,
222 .get_link = ethtool_op_get_link,
223 .get_strings = veth_get_strings,
224 .get_sset_count = veth_get_sset_count,
225 .get_ethtool_stats = veth_get_ethtool_stats,
226 .get_link_ksettings = veth_get_link_ksettings,
227 .get_ts_info = ethtool_op_get_ts_info,
230 /* general routines */
232 static bool veth_is_xdp_frame(void *ptr)
234 return (unsigned long)ptr & VETH_XDP_FLAG;
237 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
239 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
242 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
244 return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
247 static void veth_ptr_free(void *ptr)
249 if (veth_is_xdp_frame(ptr))
250 xdp_return_frame(veth_ptr_to_xdp(ptr));
255 static void __veth_xdp_flush(struct veth_rq *rq)
257 /* Write ptr_ring before reading rx_notify_masked */
259 if (!READ_ONCE(rq->rx_notify_masked) &&
260 napi_schedule_prep(&rq->xdp_napi)) {
261 WRITE_ONCE(rq->rx_notify_masked, true);
262 __napi_schedule(&rq->xdp_napi);
266 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
268 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
269 dev_kfree_skb_any(skb);
273 return NET_RX_SUCCESS;
276 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
277 struct veth_rq *rq, bool xdp)
279 return __dev_forward_skb(dev, skb) ?: xdp ?
280 veth_xdp_rx(rq, skb) :
284 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
286 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
287 struct veth_rq *rq = NULL;
288 struct net_device *rcv;
289 int length = skb->len;
290 bool rcv_xdp = false;
294 rcv = rcu_dereference(priv->peer);
295 if (unlikely(!rcv) || !pskb_may_pull(skb, ETH_HLEN)) {
300 rcv_priv = netdev_priv(rcv);
301 rxq = skb_get_queue_mapping(skb);
302 if (rxq < rcv->real_num_rx_queues) {
303 rq = &rcv_priv->rq[rxq];
304 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
307 skb_tx_timestamp(skb);
308 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
310 dev_lstats_add(dev, length);
313 atomic64_inc(&priv->dropped);
317 __veth_xdp_flush(rq);
324 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
326 struct veth_priv *priv = netdev_priv(dev);
328 dev_lstats_read(dev, packets, bytes);
329 return atomic64_read(&priv->dropped);
332 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
334 struct veth_priv *priv = netdev_priv(dev);
337 result->peer_tq_xdp_xmit_err = 0;
338 result->xdp_packets = 0;
339 result->xdp_tx_err = 0;
340 result->xdp_bytes = 0;
341 result->rx_drops = 0;
342 for (i = 0; i < dev->num_rx_queues; i++) {
343 u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
344 struct veth_rq_stats *stats = &priv->rq[i].stats;
348 start = u64_stats_fetch_begin_irq(&stats->syncp);
349 peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
350 xdp_tx_err = stats->vs.xdp_tx_err;
351 packets = stats->vs.xdp_packets;
352 bytes = stats->vs.xdp_bytes;
353 drops = stats->vs.rx_drops;
354 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
355 result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
356 result->xdp_tx_err += xdp_tx_err;
357 result->xdp_packets += packets;
358 result->xdp_bytes += bytes;
359 result->rx_drops += drops;
363 static void veth_get_stats64(struct net_device *dev,
364 struct rtnl_link_stats64 *tot)
366 struct veth_priv *priv = netdev_priv(dev);
367 struct net_device *peer;
368 struct veth_stats rx;
371 tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
372 tot->tx_bytes = bytes;
373 tot->tx_packets = packets;
375 veth_stats_rx(&rx, dev);
376 tot->tx_dropped += rx.xdp_tx_err;
377 tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
378 tot->rx_bytes = rx.xdp_bytes;
379 tot->rx_packets = rx.xdp_packets;
382 peer = rcu_dereference(priv->peer);
384 veth_stats_tx(peer, &packets, &bytes);
385 tot->rx_bytes += bytes;
386 tot->rx_packets += packets;
388 veth_stats_rx(&rx, peer);
389 tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
390 tot->rx_dropped += rx.xdp_tx_err;
391 tot->tx_bytes += rx.xdp_bytes;
392 tot->tx_packets += rx.xdp_packets;
397 /* fake multicast ability */
398 static void veth_set_multicast_list(struct net_device *dev)
402 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
407 skb = build_skb(head, buflen);
411 skb_reserve(skb, headroom);
417 static int veth_select_rxq(struct net_device *dev)
419 return smp_processor_id() % dev->real_num_rx_queues;
422 static struct net_device *veth_peer_dev(struct net_device *dev)
424 struct veth_priv *priv = netdev_priv(dev);
426 /* Callers must be under RCU read side. */
427 return rcu_dereference(priv->peer);
430 static int veth_xdp_xmit(struct net_device *dev, int n,
431 struct xdp_frame **frames,
432 u32 flags, bool ndo_xmit)
434 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
435 int i, ret = -ENXIO, drops = 0;
436 struct net_device *rcv;
437 unsigned int max_len;
440 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
444 rcv = rcu_dereference(priv->peer);
448 rcv_priv = netdev_priv(rcv);
449 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
450 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
451 * side. This means an XDP program is loaded on the peer and the peer
454 if (!rcu_access_pointer(rq->xdp_prog))
457 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
459 spin_lock(&rq->xdp_ring.producer_lock);
460 for (i = 0; i < n; i++) {
461 struct xdp_frame *frame = frames[i];
462 void *ptr = veth_xdp_to_ptr(frame);
464 if (unlikely(frame->len > max_len ||
465 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
466 xdp_return_frame_rx_napi(frame);
470 spin_unlock(&rq->xdp_ring.producer_lock);
472 if (flags & XDP_XMIT_FLUSH)
473 __veth_xdp_flush(rq);
477 u64_stats_update_begin(&rq->stats.syncp);
478 rq->stats.vs.peer_tq_xdp_xmit += n - drops;
479 rq->stats.vs.peer_tq_xdp_xmit_err += drops;
480 u64_stats_update_end(&rq->stats.syncp);
489 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
490 struct xdp_frame **frames, u32 flags)
494 err = veth_xdp_xmit(dev, n, frames, flags, true);
496 struct veth_priv *priv = netdev_priv(dev);
498 atomic64_add(n, &priv->dropped);
504 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
506 int sent, i, err = 0;
508 sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
512 for (i = 0; i < bq->count; i++)
513 xdp_return_frame(bq->q[i]);
515 trace_xdp_bulk_tx(rq->dev, sent, bq->count - sent, err);
517 u64_stats_update_begin(&rq->stats.syncp);
518 rq->stats.vs.xdp_tx += sent;
519 rq->stats.vs.xdp_tx_err += bq->count - sent;
520 u64_stats_update_end(&rq->stats.syncp);
525 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
527 struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
528 struct net_device *rcv;
529 struct veth_rq *rcv_rq;
532 veth_xdp_flush_bq(rq, bq);
533 rcv = rcu_dereference(priv->peer);
537 rcv_priv = netdev_priv(rcv);
538 rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
539 /* xdp_ring is initialized on receive side? */
540 if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
543 __veth_xdp_flush(rcv_rq);
548 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
549 struct veth_xdp_tx_bq *bq)
551 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
553 if (unlikely(!frame))
556 if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
557 veth_xdp_flush_bq(rq, bq);
559 bq->q[bq->count++] = frame;
564 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
565 struct xdp_frame *frame,
566 struct veth_xdp_tx_bq *bq,
567 struct veth_stats *stats)
569 void *hard_start = frame->data - frame->headroom;
570 int len = frame->len, delta = 0;
571 struct xdp_frame orig_frame;
572 struct bpf_prog *xdp_prog;
573 unsigned int headroom;
576 /* bpf_xdp_adjust_head() assures BPF cannot access xdp_frame area */
577 hard_start -= sizeof(struct xdp_frame);
580 xdp_prog = rcu_dereference(rq->xdp_prog);
581 if (likely(xdp_prog)) {
585 xdp_convert_frame_to_buff(frame, &xdp);
586 xdp.rxq = &rq->xdp_rxq;
588 act = bpf_prog_run_xdp(xdp_prog, &xdp);
592 delta = frame->data - xdp.data;
593 len = xdp.data_end - xdp.data;
597 xdp.rxq->mem = frame->mem;
598 if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
599 trace_xdp_exception(rq->dev, xdp_prog, act);
609 xdp.rxq->mem = frame->mem;
610 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
615 stats->xdp_redirect++;
619 bpf_warn_invalid_xdp_action(act);
622 trace_xdp_exception(rq->dev, xdp_prog, act);
631 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
632 skb = veth_build_skb(hard_start, headroom, len, frame->frame_sz);
634 xdp_return_frame(frame);
639 xdp_release_frame(frame);
640 xdp_scrub_frame(frame);
641 skb->protocol = eth_type_trans(skb, rq->dev);
646 xdp_return_frame(frame);
651 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
653 struct veth_xdp_tx_bq *bq,
654 struct veth_stats *stats)
656 u32 pktlen, headroom, act, metalen;
657 void *orig_data, *orig_data_end;
658 struct bpf_prog *xdp_prog;
659 int mac_len, delta, off;
665 xdp_prog = rcu_dereference(rq->xdp_prog);
666 if (unlikely(!xdp_prog)) {
671 mac_len = skb->data - skb_mac_header(skb);
672 pktlen = skb->len + mac_len;
673 headroom = skb_headroom(skb) - mac_len;
675 if (skb_shared(skb) || skb_head_is_locked(skb) ||
676 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
677 struct sk_buff *nskb;
682 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
683 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
684 if (size > PAGE_SIZE)
687 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
691 head = page_address(page);
692 start = head + VETH_XDP_HEADROOM;
693 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
694 page_frag_free(head);
698 nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
699 skb->len, PAGE_SIZE);
701 page_frag_free(head);
705 skb_copy_header(nskb, skb);
706 head_off = skb_headroom(nskb) - skb_headroom(skb);
707 skb_headers_offset_update(nskb, head_off);
712 xdp.data_hard_start = skb->head;
713 xdp.data = skb_mac_header(skb);
714 xdp.data_end = xdp.data + pktlen;
715 xdp.data_meta = xdp.data;
716 xdp.rxq = &rq->xdp_rxq;
718 /* SKB "head" area always have tailroom for skb_shared_info */
719 xdp.frame_sz = (void *)skb_end_pointer(skb) - xdp.data_hard_start;
720 xdp.frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
722 orig_data = xdp.data;
723 orig_data_end = xdp.data_end;
725 act = bpf_prog_run_xdp(xdp_prog, &xdp);
731 get_page(virt_to_page(xdp.data));
733 xdp.rxq->mem = rq->xdp_mem;
734 if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
735 trace_xdp_exception(rq->dev, xdp_prog, act);
743 get_page(virt_to_page(xdp.data));
745 xdp.rxq->mem = rq->xdp_mem;
746 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
750 stats->xdp_redirect++;
754 bpf_warn_invalid_xdp_action(act);
757 trace_xdp_exception(rq->dev, xdp_prog, act);
765 /* check if bpf_xdp_adjust_head was used */
766 delta = orig_data - xdp.data;
767 off = mac_len + delta;
769 __skb_push(skb, off);
771 __skb_pull(skb, -off);
772 skb->mac_header -= delta;
774 /* check if bpf_xdp_adjust_tail was used */
775 off = xdp.data_end - orig_data_end;
777 __skb_put(skb, off); /* positive on grow, negative on shrink */
778 skb->protocol = eth_type_trans(skb, rq->dev);
780 metalen = xdp.data - xdp.data_meta;
782 skb_metadata_set(skb, metalen);
793 page_frag_free(xdp.data);
798 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
799 struct veth_xdp_tx_bq *bq,
800 struct veth_stats *stats)
804 for (i = 0; i < budget; i++) {
805 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
811 if (veth_is_xdp_frame(ptr)) {
812 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
814 stats->xdp_bytes += frame->len;
815 skb = veth_xdp_rcv_one(rq, frame, bq, stats);
818 stats->xdp_bytes += skb->len;
819 skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
823 napi_gro_receive(&rq->xdp_napi, skb);
828 u64_stats_update_begin(&rq->stats.syncp);
829 rq->stats.vs.xdp_redirect += stats->xdp_redirect;
830 rq->stats.vs.xdp_bytes += stats->xdp_bytes;
831 rq->stats.vs.xdp_drops += stats->xdp_drops;
832 rq->stats.vs.rx_drops += stats->rx_drops;
833 rq->stats.vs.xdp_packets += done;
834 u64_stats_update_end(&rq->stats.syncp);
839 static int veth_poll(struct napi_struct *napi, int budget)
842 container_of(napi, struct veth_rq, xdp_napi);
843 struct veth_stats stats = {};
844 struct veth_xdp_tx_bq bq;
849 xdp_set_return_frame_no_direct();
850 done = veth_xdp_rcv(rq, budget, &bq, &stats);
852 if (done < budget && napi_complete_done(napi, done)) {
853 /* Write rx_notify_masked before reading ptr_ring */
854 smp_store_mb(rq->rx_notify_masked, false);
855 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
856 if (napi_schedule_prep(&rq->xdp_napi)) {
857 WRITE_ONCE(rq->rx_notify_masked, true);
858 __napi_schedule(&rq->xdp_napi);
863 if (stats.xdp_tx > 0)
864 veth_xdp_flush(rq, &bq);
865 if (stats.xdp_redirect > 0)
867 xdp_clear_return_frame_no_direct();
872 static int veth_napi_add(struct net_device *dev)
874 struct veth_priv *priv = netdev_priv(dev);
877 for (i = 0; i < dev->real_num_rx_queues; i++) {
878 struct veth_rq *rq = &priv->rq[i];
880 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
885 for (i = 0; i < dev->real_num_rx_queues; i++) {
886 struct veth_rq *rq = &priv->rq[i];
888 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
889 napi_enable(&rq->xdp_napi);
894 for (i--; i >= 0; i--)
895 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
900 static void veth_napi_del(struct net_device *dev)
902 struct veth_priv *priv = netdev_priv(dev);
905 for (i = 0; i < dev->real_num_rx_queues; i++) {
906 struct veth_rq *rq = &priv->rq[i];
908 napi_disable(&rq->xdp_napi);
909 __netif_napi_del(&rq->xdp_napi);
913 for (i = 0; i < dev->real_num_rx_queues; i++) {
914 struct veth_rq *rq = &priv->rq[i];
916 rq->rx_notify_masked = false;
917 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
921 static int veth_enable_xdp(struct net_device *dev)
923 struct veth_priv *priv = netdev_priv(dev);
926 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
927 for (i = 0; i < dev->real_num_rx_queues; i++) {
928 struct veth_rq *rq = &priv->rq[i];
930 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
934 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
935 MEM_TYPE_PAGE_SHARED,
940 /* Save original mem info as it can be overwritten */
941 rq->xdp_mem = rq->xdp_rxq.mem;
944 err = veth_napi_add(dev);
949 for (i = 0; i < dev->real_num_rx_queues; i++)
950 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
954 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
956 for (i--; i >= 0; i--)
957 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
962 static void veth_disable_xdp(struct net_device *dev)
964 struct veth_priv *priv = netdev_priv(dev);
967 for (i = 0; i < dev->real_num_rx_queues; i++)
968 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
970 for (i = 0; i < dev->real_num_rx_queues; i++) {
971 struct veth_rq *rq = &priv->rq[i];
973 rq->xdp_rxq.mem = rq->xdp_mem;
974 xdp_rxq_info_unreg(&rq->xdp_rxq);
978 static int veth_open(struct net_device *dev)
980 struct veth_priv *priv = netdev_priv(dev);
981 struct net_device *peer = rtnl_dereference(priv->peer);
987 if (priv->_xdp_prog) {
988 err = veth_enable_xdp(dev);
993 if (peer->flags & IFF_UP) {
994 netif_carrier_on(dev);
995 netif_carrier_on(peer);
1001 static int veth_close(struct net_device *dev)
1003 struct veth_priv *priv = netdev_priv(dev);
1004 struct net_device *peer = rtnl_dereference(priv->peer);
1006 netif_carrier_off(dev);
1008 netif_carrier_off(peer);
1010 if (priv->_xdp_prog)
1011 veth_disable_xdp(dev);
1016 static int is_valid_veth_mtu(int mtu)
1018 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1021 static int veth_alloc_queues(struct net_device *dev)
1023 struct veth_priv *priv = netdev_priv(dev);
1026 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1030 for (i = 0; i < dev->num_rx_queues; i++) {
1031 priv->rq[i].dev = dev;
1032 u64_stats_init(&priv->rq[i].stats.syncp);
1038 static void veth_free_queues(struct net_device *dev)
1040 struct veth_priv *priv = netdev_priv(dev);
1045 static int veth_dev_init(struct net_device *dev)
1049 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1053 err = veth_alloc_queues(dev);
1055 free_percpu(dev->lstats);
1062 static void veth_dev_free(struct net_device *dev)
1064 veth_free_queues(dev);
1065 free_percpu(dev->lstats);
1068 #ifdef CONFIG_NET_POLL_CONTROLLER
1069 static void veth_poll_controller(struct net_device *dev)
1071 /* veth only receives frames when its peer sends one
1072 * Since it has nothing to do with disabling irqs, we are guaranteed
1073 * never to have pending data when we poll for it so
1074 * there is nothing to do here.
1076 * We need this though so netpoll recognizes us as an interface that
1077 * supports polling, which enables bridge devices in virt setups to
1078 * still use netconsole
1081 #endif /* CONFIG_NET_POLL_CONTROLLER */
1083 static int veth_get_iflink(const struct net_device *dev)
1085 struct veth_priv *priv = netdev_priv(dev);
1086 struct net_device *peer;
1090 peer = rcu_dereference(priv->peer);
1091 iflink = peer ? peer->ifindex : 0;
1097 static netdev_features_t veth_fix_features(struct net_device *dev,
1098 netdev_features_t features)
1100 struct veth_priv *priv = netdev_priv(dev);
1101 struct net_device *peer;
1103 peer = rtnl_dereference(priv->peer);
1105 struct veth_priv *peer_priv = netdev_priv(peer);
1107 if (peer_priv->_xdp_prog)
1108 features &= ~NETIF_F_GSO_SOFTWARE;
1114 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1116 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1117 struct net_device *peer;
1123 peer = rcu_dereference(priv->peer);
1124 if (unlikely(!peer))
1127 peer_priv = netdev_priv(peer);
1128 priv->requested_headroom = new_hr;
1129 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1130 dev->needed_headroom = new_hr;
1131 peer->needed_headroom = new_hr;
1137 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1138 struct netlink_ext_ack *extack)
1140 struct veth_priv *priv = netdev_priv(dev);
1141 struct bpf_prog *old_prog;
1142 struct net_device *peer;
1143 unsigned int max_mtu;
1146 old_prog = priv->_xdp_prog;
1147 priv->_xdp_prog = prog;
1148 peer = rtnl_dereference(priv->peer);
1152 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1157 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1158 peer->hard_header_len -
1159 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1160 if (peer->mtu > max_mtu) {
1161 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1166 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1167 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1172 if (dev->flags & IFF_UP) {
1173 err = veth_enable_xdp(dev);
1175 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1181 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1182 peer->max_mtu = max_mtu;
1188 if (dev->flags & IFF_UP)
1189 veth_disable_xdp(dev);
1192 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1193 peer->max_mtu = ETH_MAX_MTU;
1196 bpf_prog_put(old_prog);
1199 if ((!!old_prog ^ !!prog) && peer)
1200 netdev_update_features(peer);
1204 priv->_xdp_prog = old_prog;
1209 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1211 switch (xdp->command) {
1212 case XDP_SETUP_PROG:
1213 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1219 static const struct net_device_ops veth_netdev_ops = {
1220 .ndo_init = veth_dev_init,
1221 .ndo_open = veth_open,
1222 .ndo_stop = veth_close,
1223 .ndo_start_xmit = veth_xmit,
1224 .ndo_get_stats64 = veth_get_stats64,
1225 .ndo_set_rx_mode = veth_set_multicast_list,
1226 .ndo_set_mac_address = eth_mac_addr,
1227 #ifdef CONFIG_NET_POLL_CONTROLLER
1228 .ndo_poll_controller = veth_poll_controller,
1230 .ndo_get_iflink = veth_get_iflink,
1231 .ndo_fix_features = veth_fix_features,
1232 .ndo_features_check = passthru_features_check,
1233 .ndo_set_rx_headroom = veth_set_rx_headroom,
1234 .ndo_bpf = veth_xdp,
1235 .ndo_xdp_xmit = veth_ndo_xdp_xmit,
1236 .ndo_get_peer_dev = veth_peer_dev,
1239 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1240 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1241 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1242 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1243 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1245 static void veth_setup(struct net_device *dev)
1249 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1250 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1251 dev->priv_flags |= IFF_NO_QUEUE;
1252 dev->priv_flags |= IFF_PHONY_HEADROOM;
1254 dev->netdev_ops = &veth_netdev_ops;
1255 dev->ethtool_ops = &veth_ethtool_ops;
1256 dev->features |= NETIF_F_LLTX;
1257 dev->features |= VETH_FEATURES;
1258 dev->vlan_features = dev->features &
1259 ~(NETIF_F_HW_VLAN_CTAG_TX |
1260 NETIF_F_HW_VLAN_STAG_TX |
1261 NETIF_F_HW_VLAN_CTAG_RX |
1262 NETIF_F_HW_VLAN_STAG_RX);
1263 dev->needs_free_netdev = true;
1264 dev->priv_destructor = veth_dev_free;
1265 dev->max_mtu = ETH_MAX_MTU;
1267 dev->hw_features = VETH_FEATURES;
1268 dev->hw_enc_features = VETH_FEATURES;
1269 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1276 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1277 struct netlink_ext_ack *extack)
1279 if (tb[IFLA_ADDRESS]) {
1280 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1282 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1283 return -EADDRNOTAVAIL;
1286 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1292 static struct rtnl_link_ops veth_link_ops;
1294 static int veth_newlink(struct net *src_net, struct net_device *dev,
1295 struct nlattr *tb[], struct nlattr *data[],
1296 struct netlink_ext_ack *extack)
1299 struct net_device *peer;
1300 struct veth_priv *priv;
1301 char ifname[IFNAMSIZ];
1302 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1303 unsigned char name_assign_type;
1304 struct ifinfomsg *ifmp;
1308 * create and register peer first
1310 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1311 struct nlattr *nla_peer;
1313 nla_peer = data[VETH_INFO_PEER];
1314 ifmp = nla_data(nla_peer);
1315 err = rtnl_nla_parse_ifla(peer_tb,
1316 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1317 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1322 err = veth_validate(peer_tb, NULL, extack);
1332 if (ifmp && tbp[IFLA_IFNAME]) {
1333 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1334 name_assign_type = NET_NAME_USER;
1336 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1337 name_assign_type = NET_NAME_ENUM;
1340 net = rtnl_link_get_net(src_net, tbp);
1342 return PTR_ERR(net);
1344 peer = rtnl_create_link(net, ifname, name_assign_type,
1345 &veth_link_ops, tbp, extack);
1348 return PTR_ERR(peer);
1351 if (!ifmp || !tbp[IFLA_ADDRESS])
1352 eth_hw_addr_random(peer);
1354 if (ifmp && (dev->ifindex != 0))
1355 peer->ifindex = ifmp->ifi_index;
1357 peer->gso_max_size = dev->gso_max_size;
1358 peer->gso_max_segs = dev->gso_max_segs;
1360 err = register_netdevice(peer);
1364 goto err_register_peer;
1366 netif_carrier_off(peer);
1368 err = rtnl_configure_link(peer, ifmp);
1370 goto err_configure_peer;
1375 * note, that since we've registered new device the dev's name
1376 * should be re-allocated
1379 if (tb[IFLA_ADDRESS] == NULL)
1380 eth_hw_addr_random(dev);
1382 if (tb[IFLA_IFNAME])
1383 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1385 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1387 err = register_netdevice(dev);
1389 goto err_register_dev;
1391 netif_carrier_off(dev);
1394 * tie the deviced together
1397 priv = netdev_priv(dev);
1398 rcu_assign_pointer(priv->peer, peer);
1400 priv = netdev_priv(peer);
1401 rcu_assign_pointer(priv->peer, dev);
1408 unregister_netdevice(peer);
1416 static void veth_dellink(struct net_device *dev, struct list_head *head)
1418 struct veth_priv *priv;
1419 struct net_device *peer;
1421 priv = netdev_priv(dev);
1422 peer = rtnl_dereference(priv->peer);
1424 /* Note : dellink() is called from default_device_exit_batch(),
1425 * before a rcu_synchronize() point. The devices are guaranteed
1426 * not being freed before one RCU grace period.
1428 RCU_INIT_POINTER(priv->peer, NULL);
1429 unregister_netdevice_queue(dev, head);
1432 priv = netdev_priv(peer);
1433 RCU_INIT_POINTER(priv->peer, NULL);
1434 unregister_netdevice_queue(peer, head);
1438 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1439 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1442 static struct net *veth_get_link_net(const struct net_device *dev)
1444 struct veth_priv *priv = netdev_priv(dev);
1445 struct net_device *peer = rtnl_dereference(priv->peer);
1447 return peer ? dev_net(peer) : dev_net(dev);
1450 static struct rtnl_link_ops veth_link_ops = {
1452 .priv_size = sizeof(struct veth_priv),
1453 .setup = veth_setup,
1454 .validate = veth_validate,
1455 .newlink = veth_newlink,
1456 .dellink = veth_dellink,
1457 .policy = veth_policy,
1458 .maxtype = VETH_INFO_MAX,
1459 .get_link_net = veth_get_link_net,
1466 static __init int veth_init(void)
1468 return rtnl_link_register(&veth_link_ops);
1471 static __exit void veth_exit(void)
1473 rtnl_link_unregister(&veth_link_ops);
1476 module_init(veth_init);
1477 module_exit(veth_exit);
1479 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1480 MODULE_LICENSE("GPL v2");
1481 MODULE_ALIAS_RTNL_LINK(DRV_NAME);