1
2
3
4
5
6
7
8
9
10
11
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>
17
18#include <net/rtnetlink.h>
19#include <net/dst.h>
20#include <net/xfrm.h>
21#include <net/xdp.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
30#define DRV_NAME "veth"
31#define DRV_VERSION "1.0"
32
33#define VETH_XDP_FLAG BIT(0)
34#define VETH_RING_SIZE 256
35#define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
36
37
38#define VETH_XDP_TX BIT(0)
39#define VETH_XDP_REDIR BIT(1)
40
41#define VETH_XDP_TX_BULK_SIZE 16
42
43struct veth_rq_stats {
44 u64 xdp_packets;
45 u64 xdp_bytes;
46 u64 xdp_drops;
47 struct u64_stats_sync syncp;
48};
49
50struct veth_rq {
51 struct napi_struct xdp_napi;
52 struct net_device *dev;
53 struct bpf_prog __rcu *xdp_prog;
54 struct xdp_mem_info xdp_mem;
55 struct veth_rq_stats stats;
56 bool rx_notify_masked;
57 struct ptr_ring xdp_ring;
58 struct xdp_rxq_info xdp_rxq;
59};
60
61struct veth_priv {
62 struct net_device __rcu *peer;
63 atomic64_t dropped;
64 struct bpf_prog *_xdp_prog;
65 struct veth_rq *rq;
66 unsigned int requested_headroom;
67};
68
69struct veth_xdp_tx_bq {
70 struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
71 unsigned int count;
72};
73
74
75
76
77
78struct veth_q_stat_desc {
79 char desc[ETH_GSTRING_LEN];
80 size_t offset;
81};
82
83#define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m)
84
85static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
86 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
87 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
88 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
89};
90
91#define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
92
93static struct {
94 const char string[ETH_GSTRING_LEN];
95} ethtool_stats_keys[] = {
96 { "peer_ifindex" },
97};
98
99static int veth_get_link_ksettings(struct net_device *dev,
100 struct ethtool_link_ksettings *cmd)
101{
102 cmd->base.speed = SPEED_10000;
103 cmd->base.duplex = DUPLEX_FULL;
104 cmd->base.port = PORT_TP;
105 cmd->base.autoneg = AUTONEG_DISABLE;
106 return 0;
107}
108
109static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
110{
111 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
112 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
113}
114
115static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
116{
117 char *p = (char *)buf;
118 int i, j;
119
120 switch(stringset) {
121 case ETH_SS_STATS:
122 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys));
123 p += sizeof(ethtool_stats_keys);
124 for (i = 0; i < dev->real_num_rx_queues; i++) {
125 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
126 snprintf(p, ETH_GSTRING_LEN,
127 "rx_queue_%u_%.11s",
128 i, veth_rq_stats_desc[j].desc);
129 p += ETH_GSTRING_LEN;
130 }
131 }
132 break;
133 }
134}
135
136static int veth_get_sset_count(struct net_device *dev, int sset)
137{
138 switch (sset) {
139 case ETH_SS_STATS:
140 return ARRAY_SIZE(ethtool_stats_keys) +
141 VETH_RQ_STATS_LEN * dev->real_num_rx_queues;
142 default:
143 return -EOPNOTSUPP;
144 }
145}
146
147static void veth_get_ethtool_stats(struct net_device *dev,
148 struct ethtool_stats *stats, u64 *data)
149{
150 struct veth_priv *priv = netdev_priv(dev);
151 struct net_device *peer = rtnl_dereference(priv->peer);
152 int i, j, idx;
153
154 data[0] = peer ? peer->ifindex : 0;
155 idx = 1;
156 for (i = 0; i < dev->real_num_rx_queues; i++) {
157 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
158 const void *stats_base = (void *)rq_stats;
159 unsigned int start;
160 size_t offset;
161
162 do {
163 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
164 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
165 offset = veth_rq_stats_desc[j].offset;
166 data[idx + j] = *(u64 *)(stats_base + offset);
167 }
168 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
169 idx += VETH_RQ_STATS_LEN;
170 }
171}
172
173static const struct ethtool_ops veth_ethtool_ops = {
174 .get_drvinfo = veth_get_drvinfo,
175 .get_link = ethtool_op_get_link,
176 .get_strings = veth_get_strings,
177 .get_sset_count = veth_get_sset_count,
178 .get_ethtool_stats = veth_get_ethtool_stats,
179 .get_link_ksettings = veth_get_link_ksettings,
180 .get_ts_info = ethtool_op_get_ts_info,
181};
182
183
184
185static bool veth_is_xdp_frame(void *ptr)
186{
187 return (unsigned long)ptr & VETH_XDP_FLAG;
188}
189
190static void *veth_ptr_to_xdp(void *ptr)
191{
192 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
193}
194
195static void *veth_xdp_to_ptr(void *ptr)
196{
197 return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
198}
199
200static void veth_ptr_free(void *ptr)
201{
202 if (veth_is_xdp_frame(ptr))
203 xdp_return_frame(veth_ptr_to_xdp(ptr));
204 else
205 kfree_skb(ptr);
206}
207
208static void __veth_xdp_flush(struct veth_rq *rq)
209{
210
211 smp_mb();
212 if (!rq->rx_notify_masked) {
213 rq->rx_notify_masked = true;
214 napi_schedule(&rq->xdp_napi);
215 }
216}
217
218static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
219{
220 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
221 dev_kfree_skb_any(skb);
222 return NET_RX_DROP;
223 }
224
225 return NET_RX_SUCCESS;
226}
227
228static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
229 struct veth_rq *rq, bool xdp)
230{
231 return __dev_forward_skb(dev, skb) ?: xdp ?
232 veth_xdp_rx(rq, skb) :
233 netif_rx(skb);
234}
235
236static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
237{
238 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
239 struct veth_rq *rq = NULL;
240 struct net_device *rcv;
241 int length = skb->len;
242 bool rcv_xdp = false;
243 int rxq;
244
245 rcu_read_lock();
246 rcv = rcu_dereference(priv->peer);
247 if (unlikely(!rcv)) {
248 kfree_skb(skb);
249 goto drop;
250 }
251
252 rcv_priv = netdev_priv(rcv);
253 rxq = skb_get_queue_mapping(skb);
254 if (rxq < rcv->real_num_rx_queues) {
255 rq = &rcv_priv->rq[rxq];
256 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
257 if (rcv_xdp)
258 skb_record_rx_queue(skb, rxq);
259 }
260
261 skb_tx_timestamp(skb);
262 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
263 if (!rcv_xdp)
264 dev_lstats_add(dev, length);
265 } else {
266drop:
267 atomic64_inc(&priv->dropped);
268 }
269
270 if (rcv_xdp)
271 __veth_xdp_flush(rq);
272
273 rcu_read_unlock();
274
275 return NETDEV_TX_OK;
276}
277
278static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
279{
280 struct veth_priv *priv = netdev_priv(dev);
281
282 dev_lstats_read(dev, packets, bytes);
283 return atomic64_read(&priv->dropped);
284}
285
286static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
287{
288 struct veth_priv *priv = netdev_priv(dev);
289 int i;
290
291 result->xdp_packets = 0;
292 result->xdp_bytes = 0;
293 result->xdp_drops = 0;
294 for (i = 0; i < dev->num_rx_queues; i++) {
295 struct veth_rq_stats *stats = &priv->rq[i].stats;
296 u64 packets, bytes, drops;
297 unsigned int start;
298
299 do {
300 start = u64_stats_fetch_begin_irq(&stats->syncp);
301 packets = stats->xdp_packets;
302 bytes = stats->xdp_bytes;
303 drops = stats->xdp_drops;
304 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
305 result->xdp_packets += packets;
306 result->xdp_bytes += bytes;
307 result->xdp_drops += drops;
308 }
309}
310
311static void veth_get_stats64(struct net_device *dev,
312 struct rtnl_link_stats64 *tot)
313{
314 struct veth_priv *priv = netdev_priv(dev);
315 struct net_device *peer;
316 struct veth_rq_stats rx;
317 u64 packets, bytes;
318
319 tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
320 tot->tx_bytes = bytes;
321 tot->tx_packets = packets;
322
323 veth_stats_rx(&rx, dev);
324 tot->rx_dropped = rx.xdp_drops;
325 tot->rx_bytes = rx.xdp_bytes;
326 tot->rx_packets = rx.xdp_packets;
327
328 rcu_read_lock();
329 peer = rcu_dereference(priv->peer);
330 if (peer) {
331 veth_stats_tx(peer, &packets, &bytes);
332 tot->rx_bytes += bytes;
333 tot->rx_packets += packets;
334
335 veth_stats_rx(&rx, peer);
336 tot->tx_bytes += rx.xdp_bytes;
337 tot->tx_packets += rx.xdp_packets;
338 }
339 rcu_read_unlock();
340}
341
342
343static void veth_set_multicast_list(struct net_device *dev)
344{
345}
346
347static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
348 int buflen)
349{
350 struct sk_buff *skb;
351
352 if (!buflen) {
353 buflen = SKB_DATA_ALIGN(headroom + len) +
354 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
355 }
356 skb = build_skb(head, buflen);
357 if (!skb)
358 return NULL;
359
360 skb_reserve(skb, headroom);
361 skb_put(skb, len);
362
363 return skb;
364}
365
366static int veth_select_rxq(struct net_device *dev)
367{
368 return smp_processor_id() % dev->real_num_rx_queues;
369}
370
371static int veth_xdp_xmit(struct net_device *dev, int n,
372 struct xdp_frame **frames, u32 flags)
373{
374 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
375 struct net_device *rcv;
376 int i, ret, drops = n;
377 unsigned int max_len;
378 struct veth_rq *rq;
379
380 rcu_read_lock();
381 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
382 ret = -EINVAL;
383 goto drop;
384 }
385
386 rcv = rcu_dereference(priv->peer);
387 if (unlikely(!rcv)) {
388 ret = -ENXIO;
389 goto drop;
390 }
391
392 rcv_priv = netdev_priv(rcv);
393 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
394
395
396
397
398 if (!rcu_access_pointer(rq->xdp_prog)) {
399 ret = -ENXIO;
400 goto drop;
401 }
402
403 drops = 0;
404 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
405
406 spin_lock(&rq->xdp_ring.producer_lock);
407 for (i = 0; i < n; i++) {
408 struct xdp_frame *frame = frames[i];
409 void *ptr = veth_xdp_to_ptr(frame);
410
411 if (unlikely(frame->len > max_len ||
412 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
413 xdp_return_frame_rx_napi(frame);
414 drops++;
415 }
416 }
417 spin_unlock(&rq->xdp_ring.producer_lock);
418
419 if (flags & XDP_XMIT_FLUSH)
420 __veth_xdp_flush(rq);
421
422 if (likely(!drops)) {
423 rcu_read_unlock();
424 return n;
425 }
426
427 ret = n - drops;
428drop:
429 rcu_read_unlock();
430 atomic64_add(drops, &priv->dropped);
431
432 return ret;
433}
434
435static void veth_xdp_flush_bq(struct net_device *dev, struct veth_xdp_tx_bq *bq)
436{
437 int sent, i, err = 0;
438
439 sent = veth_xdp_xmit(dev, bq->count, bq->q, 0);
440 if (sent < 0) {
441 err = sent;
442 sent = 0;
443 for (i = 0; i < bq->count; i++)
444 xdp_return_frame(bq->q[i]);
445 }
446 trace_xdp_bulk_tx(dev, sent, bq->count - sent, err);
447
448 bq->count = 0;
449}
450
451static void veth_xdp_flush(struct net_device *dev, struct veth_xdp_tx_bq *bq)
452{
453 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
454 struct net_device *rcv;
455 struct veth_rq *rq;
456
457 rcu_read_lock();
458 veth_xdp_flush_bq(dev, bq);
459 rcv = rcu_dereference(priv->peer);
460 if (unlikely(!rcv))
461 goto out;
462
463 rcv_priv = netdev_priv(rcv);
464 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
465
466 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
467 goto out;
468
469 __veth_xdp_flush(rq);
470out:
471 rcu_read_unlock();
472}
473
474static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp,
475 struct veth_xdp_tx_bq *bq)
476{
477 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
478
479 if (unlikely(!frame))
480 return -EOVERFLOW;
481
482 if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
483 veth_xdp_flush_bq(dev, bq);
484
485 bq->q[bq->count++] = frame;
486
487 return 0;
488}
489
490static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
491 struct xdp_frame *frame,
492 unsigned int *xdp_xmit,
493 struct veth_xdp_tx_bq *bq)
494{
495 void *hard_start = frame->data - frame->headroom;
496 void *head = hard_start - sizeof(struct xdp_frame);
497 int len = frame->len, delta = 0;
498 struct xdp_frame orig_frame;
499 struct bpf_prog *xdp_prog;
500 unsigned int headroom;
501 struct sk_buff *skb;
502
503 rcu_read_lock();
504 xdp_prog = rcu_dereference(rq->xdp_prog);
505 if (likely(xdp_prog)) {
506 struct xdp_buff xdp;
507 u32 act;
508
509 xdp.data_hard_start = hard_start;
510 xdp.data = frame->data;
511 xdp.data_end = frame->data + frame->len;
512 xdp.data_meta = frame->data - frame->metasize;
513 xdp.rxq = &rq->xdp_rxq;
514
515 act = bpf_prog_run_xdp(xdp_prog, &xdp);
516
517 switch (act) {
518 case XDP_PASS:
519 delta = frame->data - xdp.data;
520 len = xdp.data_end - xdp.data;
521 break;
522 case XDP_TX:
523 orig_frame = *frame;
524 xdp.data_hard_start = head;
525 xdp.rxq->mem = frame->mem;
526 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
527 trace_xdp_exception(rq->dev, xdp_prog, act);
528 frame = &orig_frame;
529 goto err_xdp;
530 }
531 *xdp_xmit |= VETH_XDP_TX;
532 rcu_read_unlock();
533 goto xdp_xmit;
534 case XDP_REDIRECT:
535 orig_frame = *frame;
536 xdp.data_hard_start = head;
537 xdp.rxq->mem = frame->mem;
538 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
539 frame = &orig_frame;
540 goto err_xdp;
541 }
542 *xdp_xmit |= VETH_XDP_REDIR;
543 rcu_read_unlock();
544 goto xdp_xmit;
545 default:
546 bpf_warn_invalid_xdp_action(act);
547
548 case XDP_ABORTED:
549 trace_xdp_exception(rq->dev, xdp_prog, act);
550
551 case XDP_DROP:
552 goto err_xdp;
553 }
554 }
555 rcu_read_unlock();
556
557 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
558 skb = veth_build_skb(head, headroom, len, 0);
559 if (!skb) {
560 xdp_return_frame(frame);
561 goto err;
562 }
563
564 xdp_release_frame(frame);
565 xdp_scrub_frame(frame);
566 skb->protocol = eth_type_trans(skb, rq->dev);
567err:
568 return skb;
569err_xdp:
570 rcu_read_unlock();
571 xdp_return_frame(frame);
572xdp_xmit:
573 return NULL;
574}
575
576static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
577 unsigned int *xdp_xmit,
578 struct veth_xdp_tx_bq *bq)
579{
580 u32 pktlen, headroom, act, metalen;
581 void *orig_data, *orig_data_end;
582 struct bpf_prog *xdp_prog;
583 int mac_len, delta, off;
584 struct xdp_buff xdp;
585
586 skb_orphan(skb);
587
588 rcu_read_lock();
589 xdp_prog = rcu_dereference(rq->xdp_prog);
590 if (unlikely(!xdp_prog)) {
591 rcu_read_unlock();
592 goto out;
593 }
594
595 mac_len = skb->data - skb_mac_header(skb);
596 pktlen = skb->len + mac_len;
597 headroom = skb_headroom(skb) - mac_len;
598
599 if (skb_shared(skb) || skb_head_is_locked(skb) ||
600 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
601 struct sk_buff *nskb;
602 int size, head_off;
603 void *head, *start;
604 struct page *page;
605
606 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
607 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
608 if (size > PAGE_SIZE)
609 goto drop;
610
611 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
612 if (!page)
613 goto drop;
614
615 head = page_address(page);
616 start = head + VETH_XDP_HEADROOM;
617 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
618 page_frag_free(head);
619 goto drop;
620 }
621
622 nskb = veth_build_skb(head,
623 VETH_XDP_HEADROOM + mac_len, skb->len,
624 PAGE_SIZE);
625 if (!nskb) {
626 page_frag_free(head);
627 goto drop;
628 }
629
630 skb_copy_header(nskb, skb);
631 head_off = skb_headroom(nskb) - skb_headroom(skb);
632 skb_headers_offset_update(nskb, head_off);
633 consume_skb(skb);
634 skb = nskb;
635 }
636
637 xdp.data_hard_start = skb->head;
638 xdp.data = skb_mac_header(skb);
639 xdp.data_end = xdp.data + pktlen;
640 xdp.data_meta = xdp.data;
641 xdp.rxq = &rq->xdp_rxq;
642 orig_data = xdp.data;
643 orig_data_end = xdp.data_end;
644
645 act = bpf_prog_run_xdp(xdp_prog, &xdp);
646
647 switch (act) {
648 case XDP_PASS:
649 break;
650 case XDP_TX:
651 get_page(virt_to_page(xdp.data));
652 consume_skb(skb);
653 xdp.rxq->mem = rq->xdp_mem;
654 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
655 trace_xdp_exception(rq->dev, xdp_prog, act);
656 goto err_xdp;
657 }
658 *xdp_xmit |= VETH_XDP_TX;
659 rcu_read_unlock();
660 goto xdp_xmit;
661 case XDP_REDIRECT:
662 get_page(virt_to_page(xdp.data));
663 consume_skb(skb);
664 xdp.rxq->mem = rq->xdp_mem;
665 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
666 goto err_xdp;
667 *xdp_xmit |= VETH_XDP_REDIR;
668 rcu_read_unlock();
669 goto xdp_xmit;
670 default:
671 bpf_warn_invalid_xdp_action(act);
672
673 case XDP_ABORTED:
674 trace_xdp_exception(rq->dev, xdp_prog, act);
675
676 case XDP_DROP:
677 goto drop;
678 }
679 rcu_read_unlock();
680
681 delta = orig_data - xdp.data;
682 off = mac_len + delta;
683 if (off > 0)
684 __skb_push(skb, off);
685 else if (off < 0)
686 __skb_pull(skb, -off);
687 skb->mac_header -= delta;
688 off = xdp.data_end - orig_data_end;
689 if (off != 0)
690 __skb_put(skb, off);
691 skb->protocol = eth_type_trans(skb, rq->dev);
692
693 metalen = xdp.data - xdp.data_meta;
694 if (metalen)
695 skb_metadata_set(skb, metalen);
696out:
697 return skb;
698drop:
699 rcu_read_unlock();
700 kfree_skb(skb);
701 return NULL;
702err_xdp:
703 rcu_read_unlock();
704 page_frag_free(xdp.data);
705xdp_xmit:
706 return NULL;
707}
708
709static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit,
710 struct veth_xdp_tx_bq *bq)
711{
712 int i, done = 0, drops = 0, bytes = 0;
713
714 for (i = 0; i < budget; i++) {
715 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
716 unsigned int xdp_xmit_one = 0;
717 struct sk_buff *skb;
718
719 if (!ptr)
720 break;
721
722 if (veth_is_xdp_frame(ptr)) {
723 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
724
725 bytes += frame->len;
726 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one, bq);
727 } else {
728 skb = ptr;
729 bytes += skb->len;
730 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one, bq);
731 }
732 *xdp_xmit |= xdp_xmit_one;
733
734 if (skb)
735 napi_gro_receive(&rq->xdp_napi, skb);
736 else if (!xdp_xmit_one)
737 drops++;
738
739 done++;
740 }
741
742 u64_stats_update_begin(&rq->stats.syncp);
743 rq->stats.xdp_packets += done;
744 rq->stats.xdp_bytes += bytes;
745 rq->stats.xdp_drops += drops;
746 u64_stats_update_end(&rq->stats.syncp);
747
748 return done;
749}
750
751static int veth_poll(struct napi_struct *napi, int budget)
752{
753 struct veth_rq *rq =
754 container_of(napi, struct veth_rq, xdp_napi);
755 unsigned int xdp_xmit = 0;
756 struct veth_xdp_tx_bq bq;
757 int done;
758
759 bq.count = 0;
760
761 xdp_set_return_frame_no_direct();
762 done = veth_xdp_rcv(rq, budget, &xdp_xmit, &bq);
763
764 if (done < budget && napi_complete_done(napi, done)) {
765
766 smp_store_mb(rq->rx_notify_masked, false);
767 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
768 rq->rx_notify_masked = true;
769 napi_schedule(&rq->xdp_napi);
770 }
771 }
772
773 if (xdp_xmit & VETH_XDP_TX)
774 veth_xdp_flush(rq->dev, &bq);
775 if (xdp_xmit & VETH_XDP_REDIR)
776 xdp_do_flush();
777 xdp_clear_return_frame_no_direct();
778
779 return done;
780}
781
782static int veth_napi_add(struct net_device *dev)
783{
784 struct veth_priv *priv = netdev_priv(dev);
785 int err, i;
786
787 for (i = 0; i < dev->real_num_rx_queues; i++) {
788 struct veth_rq *rq = &priv->rq[i];
789
790 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
791 if (err)
792 goto err_xdp_ring;
793 }
794
795 for (i = 0; i < dev->real_num_rx_queues; i++) {
796 struct veth_rq *rq = &priv->rq[i];
797
798 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
799 napi_enable(&rq->xdp_napi);
800 }
801
802 return 0;
803err_xdp_ring:
804 for (i--; i >= 0; i--)
805 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
806
807 return err;
808}
809
810static void veth_napi_del(struct net_device *dev)
811{
812 struct veth_priv *priv = netdev_priv(dev);
813 int i;
814
815 for (i = 0; i < dev->real_num_rx_queues; i++) {
816 struct veth_rq *rq = &priv->rq[i];
817
818 napi_disable(&rq->xdp_napi);
819 napi_hash_del(&rq->xdp_napi);
820 }
821 synchronize_net();
822
823 for (i = 0; i < dev->real_num_rx_queues; i++) {
824 struct veth_rq *rq = &priv->rq[i];
825
826 netif_napi_del(&rq->xdp_napi);
827 rq->rx_notify_masked = false;
828 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
829 }
830}
831
832static int veth_enable_xdp(struct net_device *dev)
833{
834 struct veth_priv *priv = netdev_priv(dev);
835 int err, i;
836
837 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
838 for (i = 0; i < dev->real_num_rx_queues; i++) {
839 struct veth_rq *rq = &priv->rq[i];
840
841 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
842 if (err < 0)
843 goto err_rxq_reg;
844
845 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
846 MEM_TYPE_PAGE_SHARED,
847 NULL);
848 if (err < 0)
849 goto err_reg_mem;
850
851
852 rq->xdp_mem = rq->xdp_rxq.mem;
853 }
854
855 err = veth_napi_add(dev);
856 if (err)
857 goto err_rxq_reg;
858 }
859
860 for (i = 0; i < dev->real_num_rx_queues; i++)
861 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
862
863 return 0;
864err_reg_mem:
865 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
866err_rxq_reg:
867 for (i--; i >= 0; i--)
868 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
869
870 return err;
871}
872
873static void veth_disable_xdp(struct net_device *dev)
874{
875 struct veth_priv *priv = netdev_priv(dev);
876 int i;
877
878 for (i = 0; i < dev->real_num_rx_queues; i++)
879 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
880 veth_napi_del(dev);
881 for (i = 0; i < dev->real_num_rx_queues; i++) {
882 struct veth_rq *rq = &priv->rq[i];
883
884 rq->xdp_rxq.mem = rq->xdp_mem;
885 xdp_rxq_info_unreg(&rq->xdp_rxq);
886 }
887}
888
889static int veth_open(struct net_device *dev)
890{
891 struct veth_priv *priv = netdev_priv(dev);
892 struct net_device *peer = rtnl_dereference(priv->peer);
893 int err;
894
895 if (!peer)
896 return -ENOTCONN;
897
898 if (priv->_xdp_prog) {
899 err = veth_enable_xdp(dev);
900 if (err)
901 return err;
902 }
903
904 if (peer->flags & IFF_UP) {
905 netif_carrier_on(dev);
906 netif_carrier_on(peer);
907 }
908
909 return 0;
910}
911
912static int veth_close(struct net_device *dev)
913{
914 struct veth_priv *priv = netdev_priv(dev);
915 struct net_device *peer = rtnl_dereference(priv->peer);
916
917 netif_carrier_off(dev);
918 if (peer)
919 netif_carrier_off(peer);
920
921 if (priv->_xdp_prog)
922 veth_disable_xdp(dev);
923
924 return 0;
925}
926
927static int is_valid_veth_mtu(int mtu)
928{
929 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
930}
931
932static int veth_alloc_queues(struct net_device *dev)
933{
934 struct veth_priv *priv = netdev_priv(dev);
935 int i;
936
937 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
938 if (!priv->rq)
939 return -ENOMEM;
940
941 for (i = 0; i < dev->num_rx_queues; i++) {
942 priv->rq[i].dev = dev;
943 u64_stats_init(&priv->rq[i].stats.syncp);
944 }
945
946 return 0;
947}
948
949static void veth_free_queues(struct net_device *dev)
950{
951 struct veth_priv *priv = netdev_priv(dev);
952
953 kfree(priv->rq);
954}
955
956static int veth_dev_init(struct net_device *dev)
957{
958 int err;
959
960 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
961 if (!dev->lstats)
962 return -ENOMEM;
963
964 err = veth_alloc_queues(dev);
965 if (err) {
966 free_percpu(dev->lstats);
967 return err;
968 }
969
970 return 0;
971}
972
973static void veth_dev_free(struct net_device *dev)
974{
975 veth_free_queues(dev);
976 free_percpu(dev->lstats);
977}
978
979#ifdef CONFIG_NET_POLL_CONTROLLER
980static void veth_poll_controller(struct net_device *dev)
981{
982
983
984
985
986
987
988
989
990
991}
992#endif
993
994static int veth_get_iflink(const struct net_device *dev)
995{
996 struct veth_priv *priv = netdev_priv(dev);
997 struct net_device *peer;
998 int iflink;
999
1000 rcu_read_lock();
1001 peer = rcu_dereference(priv->peer);
1002 iflink = peer ? peer->ifindex : 0;
1003 rcu_read_unlock();
1004
1005 return iflink;
1006}
1007
1008static netdev_features_t veth_fix_features(struct net_device *dev,
1009 netdev_features_t features)
1010{
1011 struct veth_priv *priv = netdev_priv(dev);
1012 struct net_device *peer;
1013
1014 peer = rtnl_dereference(priv->peer);
1015 if (peer) {
1016 struct veth_priv *peer_priv = netdev_priv(peer);
1017
1018 if (peer_priv->_xdp_prog)
1019 features &= ~NETIF_F_GSO_SOFTWARE;
1020 }
1021
1022 return features;
1023}
1024
1025static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1026{
1027 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1028 struct net_device *peer;
1029
1030 if (new_hr < 0)
1031 new_hr = 0;
1032
1033 rcu_read_lock();
1034 peer = rcu_dereference(priv->peer);
1035 if (unlikely(!peer))
1036 goto out;
1037
1038 peer_priv = netdev_priv(peer);
1039 priv->requested_headroom = new_hr;
1040 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1041 dev->needed_headroom = new_hr;
1042 peer->needed_headroom = new_hr;
1043
1044out:
1045 rcu_read_unlock();
1046}
1047
1048static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1049 struct netlink_ext_ack *extack)
1050{
1051 struct veth_priv *priv = netdev_priv(dev);
1052 struct bpf_prog *old_prog;
1053 struct net_device *peer;
1054 unsigned int max_mtu;
1055 int err;
1056
1057 old_prog = priv->_xdp_prog;
1058 priv->_xdp_prog = prog;
1059 peer = rtnl_dereference(priv->peer);
1060
1061 if (prog) {
1062 if (!peer) {
1063 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1064 err = -ENOTCONN;
1065 goto err;
1066 }
1067
1068 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1069 peer->hard_header_len -
1070 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1071 if (peer->mtu > max_mtu) {
1072 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1073 err = -ERANGE;
1074 goto err;
1075 }
1076
1077 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1078 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1079 err = -ENOSPC;
1080 goto err;
1081 }
1082
1083 if (dev->flags & IFF_UP) {
1084 err = veth_enable_xdp(dev);
1085 if (err) {
1086 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1087 goto err;
1088 }
1089 }
1090
1091 if (!old_prog) {
1092 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1093 peer->max_mtu = max_mtu;
1094 }
1095 }
1096
1097 if (old_prog) {
1098 if (!prog) {
1099 if (dev->flags & IFF_UP)
1100 veth_disable_xdp(dev);
1101
1102 if (peer) {
1103 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1104 peer->max_mtu = ETH_MAX_MTU;
1105 }
1106 }
1107 bpf_prog_put(old_prog);
1108 }
1109
1110 if ((!!old_prog ^ !!prog) && peer)
1111 netdev_update_features(peer);
1112
1113 return 0;
1114err:
1115 priv->_xdp_prog = old_prog;
1116
1117 return err;
1118}
1119
1120static u32 veth_xdp_query(struct net_device *dev)
1121{
1122 struct veth_priv *priv = netdev_priv(dev);
1123 const struct bpf_prog *xdp_prog;
1124
1125 xdp_prog = priv->_xdp_prog;
1126 if (xdp_prog)
1127 return xdp_prog->aux->id;
1128
1129 return 0;
1130}
1131
1132static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1133{
1134 switch (xdp->command) {
1135 case XDP_SETUP_PROG:
1136 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1137 case XDP_QUERY_PROG:
1138 xdp->prog_id = veth_xdp_query(dev);
1139 return 0;
1140 default:
1141 return -EINVAL;
1142 }
1143}
1144
1145static const struct net_device_ops veth_netdev_ops = {
1146 .ndo_init = veth_dev_init,
1147 .ndo_open = veth_open,
1148 .ndo_stop = veth_close,
1149 .ndo_start_xmit = veth_xmit,
1150 .ndo_get_stats64 = veth_get_stats64,
1151 .ndo_set_rx_mode = veth_set_multicast_list,
1152 .ndo_set_mac_address = eth_mac_addr,
1153#ifdef CONFIG_NET_POLL_CONTROLLER
1154 .ndo_poll_controller = veth_poll_controller,
1155#endif
1156 .ndo_get_iflink = veth_get_iflink,
1157 .ndo_fix_features = veth_fix_features,
1158 .ndo_features_check = passthru_features_check,
1159 .ndo_set_rx_headroom = veth_set_rx_headroom,
1160 .ndo_bpf = veth_xdp,
1161 .ndo_xdp_xmit = veth_xdp_xmit,
1162};
1163
1164#define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1165 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1166 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1167 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1168 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1169
1170static void veth_setup(struct net_device *dev)
1171{
1172 ether_setup(dev);
1173
1174 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1175 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1176 dev->priv_flags |= IFF_NO_QUEUE;
1177 dev->priv_flags |= IFF_PHONY_HEADROOM;
1178
1179 dev->netdev_ops = &veth_netdev_ops;
1180 dev->ethtool_ops = &veth_ethtool_ops;
1181 dev->features |= NETIF_F_LLTX;
1182 dev->features |= VETH_FEATURES;
1183 dev->vlan_features = dev->features &
1184 ~(NETIF_F_HW_VLAN_CTAG_TX |
1185 NETIF_F_HW_VLAN_STAG_TX |
1186 NETIF_F_HW_VLAN_CTAG_RX |
1187 NETIF_F_HW_VLAN_STAG_RX);
1188 dev->needs_free_netdev = true;
1189 dev->priv_destructor = veth_dev_free;
1190 dev->max_mtu = ETH_MAX_MTU;
1191
1192 dev->hw_features = VETH_FEATURES;
1193 dev->hw_enc_features = VETH_FEATURES;
1194 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1195}
1196
1197
1198
1199
1200
1201static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1202 struct netlink_ext_ack *extack)
1203{
1204 if (tb[IFLA_ADDRESS]) {
1205 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1206 return -EINVAL;
1207 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1208 return -EADDRNOTAVAIL;
1209 }
1210 if (tb[IFLA_MTU]) {
1211 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1212 return -EINVAL;
1213 }
1214 return 0;
1215}
1216
1217static struct rtnl_link_ops veth_link_ops;
1218
1219static int veth_newlink(struct net *src_net, struct net_device *dev,
1220 struct nlattr *tb[], struct nlattr *data[],
1221 struct netlink_ext_ack *extack)
1222{
1223 int err;
1224 struct net_device *peer;
1225 struct veth_priv *priv;
1226 char ifname[IFNAMSIZ];
1227 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1228 unsigned char name_assign_type;
1229 struct ifinfomsg *ifmp;
1230 struct net *net;
1231
1232
1233
1234
1235 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1236 struct nlattr *nla_peer;
1237
1238 nla_peer = data[VETH_INFO_PEER];
1239 ifmp = nla_data(nla_peer);
1240 err = rtnl_nla_parse_ifla(peer_tb,
1241 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1242 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1243 NULL);
1244 if (err < 0)
1245 return err;
1246
1247 err = veth_validate(peer_tb, NULL, extack);
1248 if (err < 0)
1249 return err;
1250
1251 tbp = peer_tb;
1252 } else {
1253 ifmp = NULL;
1254 tbp = tb;
1255 }
1256
1257 if (ifmp && tbp[IFLA_IFNAME]) {
1258 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1259 name_assign_type = NET_NAME_USER;
1260 } else {
1261 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1262 name_assign_type = NET_NAME_ENUM;
1263 }
1264
1265 net = rtnl_link_get_net(src_net, tbp);
1266 if (IS_ERR(net))
1267 return PTR_ERR(net);
1268
1269 peer = rtnl_create_link(net, ifname, name_assign_type,
1270 &veth_link_ops, tbp, extack);
1271 if (IS_ERR(peer)) {
1272 put_net(net);
1273 return PTR_ERR(peer);
1274 }
1275
1276 if (!ifmp || !tbp[IFLA_ADDRESS])
1277 eth_hw_addr_random(peer);
1278
1279 if (ifmp && (dev->ifindex != 0))
1280 peer->ifindex = ifmp->ifi_index;
1281
1282 peer->gso_max_size = dev->gso_max_size;
1283 peer->gso_max_segs = dev->gso_max_segs;
1284
1285 err = register_netdevice(peer);
1286 put_net(net);
1287 net = NULL;
1288 if (err < 0)
1289 goto err_register_peer;
1290
1291 netif_carrier_off(peer);
1292
1293 err = rtnl_configure_link(peer, ifmp);
1294 if (err < 0)
1295 goto err_configure_peer;
1296
1297
1298
1299
1300
1301
1302
1303
1304 if (tb[IFLA_ADDRESS] == NULL)
1305 eth_hw_addr_random(dev);
1306
1307 if (tb[IFLA_IFNAME])
1308 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1309 else
1310 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1311
1312 err = register_netdevice(dev);
1313 if (err < 0)
1314 goto err_register_dev;
1315
1316 netif_carrier_off(dev);
1317
1318
1319
1320
1321
1322 priv = netdev_priv(dev);
1323 rcu_assign_pointer(priv->peer, peer);
1324
1325 priv = netdev_priv(peer);
1326 rcu_assign_pointer(priv->peer, dev);
1327
1328 return 0;
1329
1330err_register_dev:
1331
1332err_configure_peer:
1333 unregister_netdevice(peer);
1334 return err;
1335
1336err_register_peer:
1337 free_netdev(peer);
1338 return err;
1339}
1340
1341static void veth_dellink(struct net_device *dev, struct list_head *head)
1342{
1343 struct veth_priv *priv;
1344 struct net_device *peer;
1345
1346 priv = netdev_priv(dev);
1347 peer = rtnl_dereference(priv->peer);
1348
1349
1350
1351
1352
1353 RCU_INIT_POINTER(priv->peer, NULL);
1354 unregister_netdevice_queue(dev, head);
1355
1356 if (peer) {
1357 priv = netdev_priv(peer);
1358 RCU_INIT_POINTER(priv->peer, NULL);
1359 unregister_netdevice_queue(peer, head);
1360 }
1361}
1362
1363static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1364 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1365};
1366
1367static struct net *veth_get_link_net(const struct net_device *dev)
1368{
1369 struct veth_priv *priv = netdev_priv(dev);
1370 struct net_device *peer = rtnl_dereference(priv->peer);
1371
1372 return peer ? dev_net(peer) : dev_net(dev);
1373}
1374
1375static struct rtnl_link_ops veth_link_ops = {
1376 .kind = DRV_NAME,
1377 .priv_size = sizeof(struct veth_priv),
1378 .setup = veth_setup,
1379 .validate = veth_validate,
1380 .newlink = veth_newlink,
1381 .dellink = veth_dellink,
1382 .policy = veth_policy,
1383 .maxtype = VETH_INFO_MAX,
1384 .get_link_net = veth_get_link_net,
1385};
1386
1387
1388
1389
1390
1391static __init int veth_init(void)
1392{
1393 return rtnl_link_register(&veth_link_ops);
1394}
1395
1396static __exit void veth_exit(void)
1397{
1398 rtnl_link_unregister(&veth_link_ops);
1399}
1400
1401module_init(veth_init);
1402module_exit(veth_exit);
1403
1404MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1405MODULE_LICENSE("GPL v2");
1406MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1407
