?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
docs: fix Co-Developed-by docs
[linux/fpc-iii.git] / drivers / net / veth.c
blobf412ea1cef18bbf215ac6ae46d257b39a8a95c74
1 /*
2 * drivers/net/veth.c
3 *
4 * Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
5 *
6 * Author: Pavel Emelianov <xemul@openvz.org>
7 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
8 *
9 */
10
11 #include <linux/netdevice.h>
12 #include <linux/slab.h>
13 #include <linux/ethtool.h>
14 #include <linux/etherdevice.h>
15 #include <linux/u64_stats_sync.h>
16
17 #include <net/rtnetlink.h>
18 #include <net/dst.h>
19 #include <net/xfrm.h>
20 #include <net/xdp.h>
21 #include <linux/veth.h>
22 #include <linux/module.h>
23 #include <linux/bpf.h>
24 #include <linux/filter.h>
25 #include <linux/ptr_ring.h>
26 #include <linux/bpf_trace.h>
27 #include <linux/net_tstamp.h>
28
29 #define DRV_NAME "veth"
30 #define DRV_VERSION "1.0"
31
32 #define VETH_XDP_FLAG BIT(0)
33 #define VETH_RING_SIZE 256
34 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
35
36 /* Separating two types of XDP xmit */
37 #define VETH_XDP_TX BIT(0)
38 #define VETH_XDP_REDIR BIT(1)
39
40 struct veth_rq_stats {
41 u64 xdp_packets;
42 u64 xdp_bytes;
43 u64 xdp_drops;
44 struct u64_stats_sync syncp;
45 };
46
47 struct veth_rq {
48 struct napi_struct xdp_napi;
49 struct net_device *dev;
50 struct bpf_prog __rcu *xdp_prog;
51 struct xdp_mem_info xdp_mem;
52 struct veth_rq_stats stats;
53 bool rx_notify_masked;
54 struct ptr_ring xdp_ring;
55 struct xdp_rxq_info xdp_rxq;
56 };
57
58 struct veth_priv {
59 struct net_device __rcu *peer;
60 atomic64_t dropped;
61 struct bpf_prog *_xdp_prog;
62 struct veth_rq *rq;
63 unsigned int requested_headroom;
64 };
65
66 /*
67 * ethtool interface
68 */
69
70 struct veth_q_stat_desc {
71 char desc[ETH_GSTRING_LEN];
72 size_t offset;
73 };
74
75 #define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m)
76
77 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
78 { "xdp_packets", VETH_RQ_STAT(xdp_packets) },
79 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) },
80 { "xdp_drops", VETH_RQ_STAT(xdp_drops) },
81 };
82
83 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc)
84
85 static struct {
86 const char string[ETH_GSTRING_LEN];
87 } ethtool_stats_keys[] = {
88 { "peer_ifindex" },
89 };
90
91 static int veth_get_link_ksettings(struct net_device *dev,
92 struct ethtool_link_ksettings *cmd)
93 {
94 cmd->base.speed = SPEED_10000;
95 cmd->base.duplex = DUPLEX_FULL;
96 cmd->base.port = PORT_TP;
97 cmd->base.autoneg = AUTONEG_DISABLE;
98 return 0;
99 }
100
101 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
102 {
103 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
104 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
105 }
106
107 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
108 {
109 char *p = (char *)buf;
110 int i, j;
111
112 switch(stringset) {
113 case ETH_SS_STATS:
114 memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
115 p += sizeof(ethtool_stats_keys);
116 for (i = 0; i < dev->real_num_rx_queues; i++) {
117 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
118 snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s",
119 i, veth_rq_stats_desc[j].desc);
120 p += ETH_GSTRING_LEN;
121 }
122 }
123 break;
124 }
125 }
126
127 static int veth_get_sset_count(struct net_device *dev, int sset)
128 {
129 switch (sset) {
130 case ETH_SS_STATS:
131 return ARRAY_SIZE(ethtool_stats_keys) +
132 VETH_RQ_STATS_LEN * dev->real_num_rx_queues;
133 default:
134 return -EOPNOTSUPP;
135 }
136 }
137
138 static void veth_get_ethtool_stats(struct net_device *dev,
139 struct ethtool_stats *stats, u64 *data)
140 {
141 struct veth_priv *priv = netdev_priv(dev);
142 struct net_device *peer = rtnl_dereference(priv->peer);
143 int i, j, idx;
144
145 data[0] = peer ? peer->ifindex : 0;
146 idx = 1;
147 for (i = 0; i < dev->real_num_rx_queues; i++) {
148 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
149 const void *stats_base = (void *)rq_stats;
150 unsigned int start;
151 size_t offset;
152
153 do {
154 start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
155 for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
156 offset = veth_rq_stats_desc[j].offset;
157 data[idx + j] = *(u64 *)(stats_base + offset);
158 }
159 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
160 idx += VETH_RQ_STATS_LEN;
161 }
162 }
163
164 static int veth_get_ts_info(struct net_device *dev,
165 struct ethtool_ts_info *info)
166 {
167 info->so_timestamping =
168 SOF_TIMESTAMPING_TX_SOFTWARE |
169 SOF_TIMESTAMPING_RX_SOFTWARE |
170 SOF_TIMESTAMPING_SOFTWARE;
171 info->phc_index = -1;
172
173 return 0;
174 }
175
176 static const struct ethtool_ops veth_ethtool_ops = {
177 .get_drvinfo = veth_get_drvinfo,
178 .get_link = ethtool_op_get_link,
179 .get_strings = veth_get_strings,
180 .get_sset_count = veth_get_sset_count,
181 .get_ethtool_stats = veth_get_ethtool_stats,
182 .get_link_ksettings = veth_get_link_ksettings,
183 .get_ts_info = veth_get_ts_info,
184 };
185
186 /* general routines */
187
188 static bool veth_is_xdp_frame(void *ptr)
189 {
190 return (unsigned long)ptr & VETH_XDP_FLAG;
191 }
192
193 static void *veth_ptr_to_xdp(void *ptr)
194 {
195 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
196 }
197
198 static void *veth_xdp_to_ptr(void *ptr)
199 {
200 return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
201 }
202
203 static void veth_ptr_free(void *ptr)
204 {
205 if (veth_is_xdp_frame(ptr))
206 xdp_return_frame(veth_ptr_to_xdp(ptr));
207 else
208 kfree_skb(ptr);
209 }
210
211 static void __veth_xdp_flush(struct veth_rq *rq)
212 {
213 /* Write ptr_ring before reading rx_notify_masked */
214 smp_mb();
215 if (!rq->rx_notify_masked) {
216 rq->rx_notify_masked = true;
217 napi_schedule(&rq->xdp_napi);
218 }
219 }
220
221 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
222 {
223 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
224 dev_kfree_skb_any(skb);
225 return NET_RX_DROP;
226 }
227
228 return NET_RX_SUCCESS;
229 }
230
231 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
232 struct veth_rq *rq, bool xdp)
233 {
234 return __dev_forward_skb(dev, skb) ?: xdp ?
235 veth_xdp_rx(rq, skb) :
236 netif_rx(skb);
237 }
238
239 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
240 {
241 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
242 struct veth_rq *rq = NULL;
243 struct net_device *rcv;
244 int length = skb->len;
245 bool rcv_xdp = false;
246 int rxq;
247
248 rcu_read_lock();
249 rcv = rcu_dereference(priv->peer);
250 if (unlikely(!rcv)) {
251 kfree_skb(skb);
252 goto drop;
253 }
254
255 rcv_priv = netdev_priv(rcv);
256 rxq = skb_get_queue_mapping(skb);
257 if (rxq < rcv->real_num_rx_queues) {
258 rq = &rcv_priv->rq[rxq];
259 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
260 if (rcv_xdp)
261 skb_record_rx_queue(skb, rxq);
262 }
263
264 skb_tx_timestamp(skb);
265 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
266 if (!rcv_xdp) {
267 struct pcpu_lstats *stats = this_cpu_ptr(dev->lstats);
268
269 u64_stats_update_begin(&stats->syncp);
270 stats->bytes += length;
271 stats->packets++;
272 u64_stats_update_end(&stats->syncp);
273 }
274 } else {
275 drop:
276 atomic64_inc(&priv->dropped);
277 }
278
279 if (rcv_xdp)
280 __veth_xdp_flush(rq);
281
282 rcu_read_unlock();
283
284 return NETDEV_TX_OK;
285 }
286
287 static u64 veth_stats_tx(struct pcpu_lstats *result, struct net_device *dev)
288 {
289 struct veth_priv *priv = netdev_priv(dev);
290 int cpu;
291
292 result->packets = 0;
293 result->bytes = 0;
294 for_each_possible_cpu(cpu) {
295 struct pcpu_lstats *stats = per_cpu_ptr(dev->lstats, cpu);
296 u64 packets, bytes;
297 unsigned int start;
298
299 do {
300 start = u64_stats_fetch_begin_irq(&stats->syncp);
301 packets = stats->packets;
302 bytes = stats->bytes;
303 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
304 result->packets += packets;
305 result->bytes += bytes;
306 }
307 return atomic64_read(&priv->dropped);
308 }
309
310 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
311 {
312 struct veth_priv *priv = netdev_priv(dev);
313 int i;
314
315 result->xdp_packets = 0;
316 result->xdp_bytes = 0;
317 result->xdp_drops = 0;
318 for (i = 0; i < dev->num_rx_queues; i++) {
319 struct veth_rq_stats *stats = &priv->rq[i].stats;
320 u64 packets, bytes, drops;
321 unsigned int start;
322
323 do {
324 start = u64_stats_fetch_begin_irq(&stats->syncp);
325 packets = stats->xdp_packets;
326 bytes = stats->xdp_bytes;
327 drops = stats->xdp_drops;
328 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
329 result->xdp_packets += packets;
330 result->xdp_bytes += bytes;
331 result->xdp_drops += drops;
332 }
333 }
334
335 static void veth_get_stats64(struct net_device *dev,
336 struct rtnl_link_stats64 *tot)
337 {
338 struct veth_priv *priv = netdev_priv(dev);
339 struct net_device *peer;
340 struct veth_rq_stats rx;
341 struct pcpu_lstats tx;
342
343 tot->tx_dropped = veth_stats_tx(&tx, dev);
344 tot->tx_bytes = tx.bytes;
345 tot->tx_packets = tx.packets;
346
347 veth_stats_rx(&rx, dev);
348 tot->rx_dropped = rx.xdp_drops;
349 tot->rx_bytes = rx.xdp_bytes;
350 tot->rx_packets = rx.xdp_packets;
351
352 rcu_read_lock();
353 peer = rcu_dereference(priv->peer);
354 if (peer) {
355 tot->rx_dropped += veth_stats_tx(&tx, peer);
356 tot->rx_bytes += tx.bytes;
357 tot->rx_packets += tx.packets;
358
359 veth_stats_rx(&rx, peer);
360 tot->tx_bytes += rx.xdp_bytes;
361 tot->tx_packets += rx.xdp_packets;
362 }
363 rcu_read_unlock();
364 }
365
366 /* fake multicast ability */
367 static void veth_set_multicast_list(struct net_device *dev)
368 {
369 }
370
371 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
372 int buflen)
373 {
374 struct sk_buff *skb;
375
376 if (!buflen) {
377 buflen = SKB_DATA_ALIGN(headroom + len) +
378 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
379 }
380 skb = build_skb(head, buflen);
381 if (!skb)
382 return NULL;
383
384 skb_reserve(skb, headroom);
385 skb_put(skb, len);
386
387 return skb;
388 }
389
390 static int veth_select_rxq(struct net_device *dev)
391 {
392 return smp_processor_id() % dev->real_num_rx_queues;
393 }
394
395 static int veth_xdp_xmit(struct net_device *dev, int n,
396 struct xdp_frame **frames, u32 flags)
397 {
398 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
399 struct net_device *rcv;
400 int i, ret, drops = n;
401 unsigned int max_len;
402 struct veth_rq *rq;
403
404 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
405 ret = -EINVAL;
406 goto drop;
407 }
408
409 rcv = rcu_dereference(priv->peer);
410 if (unlikely(!rcv)) {
411 ret = -ENXIO;
412 goto drop;
413 }
414
415 rcv_priv = netdev_priv(rcv);
416 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
417 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
418 * side. This means an XDP program is loaded on the peer and the peer
419 * device is up.
420 */
421 if (!rcu_access_pointer(rq->xdp_prog)) {
422 ret = -ENXIO;
423 goto drop;
424 }
425
426 drops = 0;
427 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
428
429 spin_lock(&rq->xdp_ring.producer_lock);
430 for (i = 0; i < n; i++) {
431 struct xdp_frame *frame = frames[i];
432 void *ptr = veth_xdp_to_ptr(frame);
433
434 if (unlikely(frame->len > max_len ||
435 __ptr_ring_produce(&rq->xdp_ring, ptr))) {
436 xdp_return_frame_rx_napi(frame);
437 drops++;
438 }
439 }
440 spin_unlock(&rq->xdp_ring.producer_lock);
441
442 if (flags & XDP_XMIT_FLUSH)
443 __veth_xdp_flush(rq);
444
445 if (likely(!drops))
446 return n;
447
448 ret = n - drops;
449 drop:
450 atomic64_add(drops, &priv->dropped);
451
452 return ret;
453 }
454
455 static void veth_xdp_flush(struct net_device *dev)
456 {
457 struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
458 struct net_device *rcv;
459 struct veth_rq *rq;
460
461 rcu_read_lock();
462 rcv = rcu_dereference(priv->peer);
463 if (unlikely(!rcv))
464 goto out;
465
466 rcv_priv = netdev_priv(rcv);
467 rq = &rcv_priv->rq[veth_select_rxq(rcv)];
468 /* xdp_ring is initialized on receive side? */
469 if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
470 goto out;
471
472 __veth_xdp_flush(rq);
473 out:
474 rcu_read_unlock();
475 }
476
477 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
478 {
479 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
480
481 if (unlikely(!frame))
482 return -EOVERFLOW;
483
484 return veth_xdp_xmit(dev, 1, &frame, 0);
485 }
486
487 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
488 struct xdp_frame *frame,
489 unsigned int *xdp_xmit)
490 {
491 void *hard_start = frame->data - frame->headroom;
492 void *head = hard_start - sizeof(struct xdp_frame);
493 int len = frame->len, delta = 0;
494 struct xdp_frame orig_frame;
495 struct bpf_prog *xdp_prog;
496 unsigned int headroom;
497 struct sk_buff *skb;
498
499 rcu_read_lock();
500 xdp_prog = rcu_dereference(rq->xdp_prog);
501 if (likely(xdp_prog)) {
502 struct xdp_buff xdp;
503 u32 act;
504
505 xdp.data_hard_start = hard_start;
506 xdp.data = frame->data;
507 xdp.data_end = frame->data + frame->len;
508 xdp.data_meta = frame->data - frame->metasize;
509 xdp.rxq = &rq->xdp_rxq;
510
511 act = bpf_prog_run_xdp(xdp_prog, &xdp);
512
513 switch (act) {
514 case XDP_PASS:
515 delta = frame->data - xdp.data;
516 len = xdp.data_end - xdp.data;
517 break;
518 case XDP_TX:
519 orig_frame = *frame;
520 xdp.data_hard_start = head;
521 xdp.rxq->mem = frame->mem;
522 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
523 trace_xdp_exception(rq->dev, xdp_prog, act);
524 frame = &orig_frame;
525 goto err_xdp;
526 }
527 *xdp_xmit |= VETH_XDP_TX;
528 rcu_read_unlock();
529 goto xdp_xmit;
530 case XDP_REDIRECT:
531 orig_frame = *frame;
532 xdp.data_hard_start = head;
533 xdp.rxq->mem = frame->mem;
534 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
535 frame = &orig_frame;
536 goto err_xdp;
537 }
538 *xdp_xmit |= VETH_XDP_REDIR;
539 rcu_read_unlock();
540 goto xdp_xmit;
541 default:
542 bpf_warn_invalid_xdp_action(act);
543 case XDP_ABORTED:
544 trace_xdp_exception(rq->dev, xdp_prog, act);
545 case XDP_DROP:
546 goto err_xdp;
547 }
548 }
549 rcu_read_unlock();
550
551 headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
552 skb = veth_build_skb(head, headroom, len, 0);
553 if (!skb) {
554 xdp_return_frame(frame);
555 goto err;
556 }
557
558 xdp_scrub_frame(frame);
559 skb->protocol = eth_type_trans(skb, rq->dev);
560 err:
561 return skb;
562 err_xdp:
563 rcu_read_unlock();
564 xdp_return_frame(frame);
565 xdp_xmit:
566 return NULL;
567 }
568
569 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
570 unsigned int *xdp_xmit)
571 {
572 u32 pktlen, headroom, act, metalen;
573 void *orig_data, *orig_data_end;
574 struct bpf_prog *xdp_prog;
575 int mac_len, delta, off;
576 struct xdp_buff xdp;
577
578 skb_orphan(skb);
579
580 rcu_read_lock();
581 xdp_prog = rcu_dereference(rq->xdp_prog);
582 if (unlikely(!xdp_prog)) {
583 rcu_read_unlock();
584 goto out;
585 }
586
587 mac_len = skb->data - skb_mac_header(skb);
588 pktlen = skb->len + mac_len;
589 headroom = skb_headroom(skb) - mac_len;
590
591 if (skb_shared(skb) || skb_head_is_locked(skb) ||
592 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
593 struct sk_buff *nskb;
594 int size, head_off;
595 void *head, *start;
596 struct page *page;
597
598 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
599 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
600 if (size > PAGE_SIZE)
601 goto drop;
602
603 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
604 if (!page)
605 goto drop;
606
607 head = page_address(page);
608 start = head + VETH_XDP_HEADROOM;
609 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
610 page_frag_free(head);
611 goto drop;
612 }
613
614 nskb = veth_build_skb(head,
615 VETH_XDP_HEADROOM + mac_len, skb->len,
616 PAGE_SIZE);
617 if (!nskb) {
618 page_frag_free(head);
619 goto drop;
620 }
621
622 skb_copy_header(nskb, skb);
623 head_off = skb_headroom(nskb) - skb_headroom(skb);
624 skb_headers_offset_update(nskb, head_off);
625 consume_skb(skb);
626 skb = nskb;
627 }
628
629 xdp.data_hard_start = skb->head;
630 xdp.data = skb_mac_header(skb);
631 xdp.data_end = xdp.data + pktlen;
632 xdp.data_meta = xdp.data;
633 xdp.rxq = &rq->xdp_rxq;
634 orig_data = xdp.data;
635 orig_data_end = xdp.data_end;
636
637 act = bpf_prog_run_xdp(xdp_prog, &xdp);
638
639 switch (act) {
640 case XDP_PASS:
641 break;
642 case XDP_TX:
643 get_page(virt_to_page(xdp.data));
644 consume_skb(skb);
645 xdp.rxq->mem = rq->xdp_mem;
646 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) {
647 trace_xdp_exception(rq->dev, xdp_prog, act);
648 goto err_xdp;
649 }
650 *xdp_xmit |= VETH_XDP_TX;
651 rcu_read_unlock();
652 goto xdp_xmit;
653 case XDP_REDIRECT:
654 get_page(virt_to_page(xdp.data));
655 consume_skb(skb);
656 xdp.rxq->mem = rq->xdp_mem;
657 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
658 goto err_xdp;
659 *xdp_xmit |= VETH_XDP_REDIR;
660 rcu_read_unlock();
661 goto xdp_xmit;
662 default:
663 bpf_warn_invalid_xdp_action(act);
664 case XDP_ABORTED:
665 trace_xdp_exception(rq->dev, xdp_prog, act);
666 case XDP_DROP:
667 goto drop;
668 }
669 rcu_read_unlock();
670
671 delta = orig_data - xdp.data;
672 off = mac_len + delta;
673 if (off > 0)
674 __skb_push(skb, off);
675 else if (off < 0)
676 __skb_pull(skb, -off);
677 skb->mac_header -= delta;
678 off = xdp.data_end - orig_data_end;
679 if (off != 0)
680 __skb_put(skb, off);
681 skb->protocol = eth_type_trans(skb, rq->dev);
682
683 metalen = xdp.data - xdp.data_meta;
684 if (metalen)
685 skb_metadata_set(skb, metalen);
686 out:
687 return skb;
688 drop:
689 rcu_read_unlock();
690 kfree_skb(skb);
691 return NULL;
692 err_xdp:
693 rcu_read_unlock();
694 page_frag_free(xdp.data);
695 xdp_xmit:
696 return NULL;
697 }
698
699 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit)
700 {
701 int i, done = 0, drops = 0, bytes = 0;
702
703 for (i = 0; i < budget; i++) {
704 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
705 unsigned int xdp_xmit_one = 0;
706 struct sk_buff *skb;
707
708 if (!ptr)
709 break;
710
711 if (veth_is_xdp_frame(ptr)) {
712 struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
713
714 bytes += frame->len;
715 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one);
716 } else {
717 skb = ptr;
718 bytes += skb->len;
719 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one);
720 }
721 *xdp_xmit |= xdp_xmit_one;
722
723 if (skb)
724 napi_gro_receive(&rq->xdp_napi, skb);
725 else if (!xdp_xmit_one)
726 drops++;
727
728 done++;
729 }
730
731 u64_stats_update_begin(&rq->stats.syncp);
732 rq->stats.xdp_packets += done;
733 rq->stats.xdp_bytes += bytes;
734 rq->stats.xdp_drops += drops;
735 u64_stats_update_end(&rq->stats.syncp);
736
737 return done;
738 }
739
740 static int veth_poll(struct napi_struct *napi, int budget)
741 {
742 struct veth_rq *rq =
743 container_of(napi, struct veth_rq, xdp_napi);
744 unsigned int xdp_xmit = 0;
745 int done;
746
747 xdp_set_return_frame_no_direct();
748 done = veth_xdp_rcv(rq, budget, &xdp_xmit);
749
750 if (done < budget && napi_complete_done(napi, done)) {
751 /* Write rx_notify_masked before reading ptr_ring */
752 smp_store_mb(rq->rx_notify_masked, false);
753 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
754 rq->rx_notify_masked = true;
755 napi_schedule(&rq->xdp_napi);
756 }
757 }
758
759 if (xdp_xmit & VETH_XDP_TX)
760 veth_xdp_flush(rq->dev);
761 if (xdp_xmit & VETH_XDP_REDIR)
762 xdp_do_flush_map();
763 xdp_clear_return_frame_no_direct();
764
765 return done;
766 }
767
768 static int veth_napi_add(struct net_device *dev)
769 {
770 struct veth_priv *priv = netdev_priv(dev);
771 int err, i;
772
773 for (i = 0; i < dev->real_num_rx_queues; i++) {
774 struct veth_rq *rq = &priv->rq[i];
775
776 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
777 if (err)
778 goto err_xdp_ring;
779 }
780
781 for (i = 0; i < dev->real_num_rx_queues; i++) {
782 struct veth_rq *rq = &priv->rq[i];
783
784 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
785 napi_enable(&rq->xdp_napi);
786 }
787
788 return 0;
789 err_xdp_ring:
790 for (i--; i >= 0; i--)
791 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
792
793 return err;
794 }
795
796 static void veth_napi_del(struct net_device *dev)
797 {
798 struct veth_priv *priv = netdev_priv(dev);
799 int i;
800
801 for (i = 0; i < dev->real_num_rx_queues; i++) {
802 struct veth_rq *rq = &priv->rq[i];
803
804 napi_disable(&rq->xdp_napi);
805 napi_hash_del(&rq->xdp_napi);
806 }
807 synchronize_net();
808
809 for (i = 0; i < dev->real_num_rx_queues; i++) {
810 struct veth_rq *rq = &priv->rq[i];
811
812 netif_napi_del(&rq->xdp_napi);
813 rq->rx_notify_masked = false;
814 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
815 }
816 }
817
818 static int veth_enable_xdp(struct net_device *dev)
819 {
820 struct veth_priv *priv = netdev_priv(dev);
821 int err, i;
822
823 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
824 for (i = 0; i < dev->real_num_rx_queues; i++) {
825 struct veth_rq *rq = &priv->rq[i];
826
827 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
828 if (err < 0)
829 goto err_rxq_reg;
830
831 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
832 MEM_TYPE_PAGE_SHARED,
833 NULL);
834 if (err < 0)
835 goto err_reg_mem;
836
837 /* Save original mem info as it can be overwritten */
838 rq->xdp_mem = rq->xdp_rxq.mem;
839 }
840
841 err = veth_napi_add(dev);
842 if (err)
843 goto err_rxq_reg;
844 }
845
846 for (i = 0; i < dev->real_num_rx_queues; i++)
847 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
848
849 return 0;
850 err_reg_mem:
851 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
852 err_rxq_reg:
853 for (i--; i >= 0; i--)
854 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
855
856 return err;
857 }
858
859 static void veth_disable_xdp(struct net_device *dev)
860 {
861 struct veth_priv *priv = netdev_priv(dev);
862 int i;
863
864 for (i = 0; i < dev->real_num_rx_queues; i++)
865 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
866 veth_napi_del(dev);
867 for (i = 0; i < dev->real_num_rx_queues; i++) {
868 struct veth_rq *rq = &priv->rq[i];
869
870 rq->xdp_rxq.mem = rq->xdp_mem;
871 xdp_rxq_info_unreg(&rq->xdp_rxq);
872 }
873 }
874
875 static int veth_open(struct net_device *dev)
876 {
877 struct veth_priv *priv = netdev_priv(dev);
878 struct net_device *peer = rtnl_dereference(priv->peer);
879 int err;
880
881 if (!peer)
882 return -ENOTCONN;
883
884 if (priv->_xdp_prog) {
885 err = veth_enable_xdp(dev);
886 if (err)
887 return err;
888 }
889
890 if (peer->flags & IFF_UP) {
891 netif_carrier_on(dev);
892 netif_carrier_on(peer);
893 }
894
895 return 0;
896 }
897
898 static int veth_close(struct net_device *dev)
899 {
900 struct veth_priv *priv = netdev_priv(dev);
901 struct net_device *peer = rtnl_dereference(priv->peer);
902
903 netif_carrier_off(dev);
904 if (peer)
905 netif_carrier_off(peer);
906
907 if (priv->_xdp_prog)
908 veth_disable_xdp(dev);
909
910 return 0;
911 }
912
913 static int is_valid_veth_mtu(int mtu)
914 {
915 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
916 }
917
918 static int veth_alloc_queues(struct net_device *dev)
919 {
920 struct veth_priv *priv = netdev_priv(dev);
921 int i;
922
923 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
924 if (!priv->rq)
925 return -ENOMEM;
926
927 for (i = 0; i < dev->num_rx_queues; i++) {
928 priv->rq[i].dev = dev;
929 u64_stats_init(&priv->rq[i].stats.syncp);
930 }
931
932 return 0;
933 }
934
935 static void veth_free_queues(struct net_device *dev)
936 {
937 struct veth_priv *priv = netdev_priv(dev);
938
939 kfree(priv->rq);
940 }
941
942 static int veth_dev_init(struct net_device *dev)
943 {
944 int err;
945
946 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
947 if (!dev->lstats)
948 return -ENOMEM;
949
950 err = veth_alloc_queues(dev);
951 if (err) {
952 free_percpu(dev->lstats);
953 return err;
954 }
955
956 return 0;
957 }
958
959 static void veth_dev_free(struct net_device *dev)
960 {
961 veth_free_queues(dev);
962 free_percpu(dev->lstats);
963 }
964
965 #ifdef CONFIG_NET_POLL_CONTROLLER
966 static void veth_poll_controller(struct net_device *dev)
967 {
968 /* veth only receives frames when its peer sends one
969 * Since it has nothing to do with disabling irqs, we are guaranteed
970 * never to have pending data when we poll for it so
971 * there is nothing to do here.
972 *
973 * We need this though so netpoll recognizes us as an interface that
974 * supports polling, which enables bridge devices in virt setups to
975 * still use netconsole
976 */
977 }
978 #endif /* CONFIG_NET_POLL_CONTROLLER */
979
980 static int veth_get_iflink(const struct net_device *dev)
981 {
982 struct veth_priv *priv = netdev_priv(dev);
983 struct net_device *peer;
984 int iflink;
985
986 rcu_read_lock();
987 peer = rcu_dereference(priv->peer);
988 iflink = peer ? peer->ifindex : 0;
989 rcu_read_unlock();
990
991 return iflink;
992 }
993
994 static netdev_features_t veth_fix_features(struct net_device *dev,
995 netdev_features_t features)
996 {
997 struct veth_priv *priv = netdev_priv(dev);
998 struct net_device *peer;
999
1000 peer = rtnl_dereference(priv->peer);
1001 if (peer) {
1002 struct veth_priv *peer_priv = netdev_priv(peer);
1003
1004 if (peer_priv->_xdp_prog)
1005 features &= ~NETIF_F_GSO_SOFTWARE;
1006 }
1007
1008 return features;
1009 }
1010
1011 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1012 {
1013 struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1014 struct net_device *peer;
1015
1016 if (new_hr < 0)
1017 new_hr = 0;
1018
1019 rcu_read_lock();
1020 peer = rcu_dereference(priv->peer);
1021 if (unlikely(!peer))
1022 goto out;
1023
1024 peer_priv = netdev_priv(peer);
1025 priv->requested_headroom = new_hr;
1026 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1027 dev->needed_headroom = new_hr;
1028 peer->needed_headroom = new_hr;
1029
1030 out:
1031 rcu_read_unlock();
1032 }
1033
1034 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1035 struct netlink_ext_ack *extack)
1036 {
1037 struct veth_priv *priv = netdev_priv(dev);
1038 struct bpf_prog *old_prog;
1039 struct net_device *peer;
1040 unsigned int max_mtu;
1041 int err;
1042
1043 old_prog = priv->_xdp_prog;
1044 priv->_xdp_prog = prog;
1045 peer = rtnl_dereference(priv->peer);
1046
1047 if (prog) {
1048 if (!peer) {
1049 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1050 err = -ENOTCONN;
1051 goto err;
1052 }
1053
1054 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1055 peer->hard_header_len -
1056 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1057 if (peer->mtu > max_mtu) {
1058 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1059 err = -ERANGE;
1060 goto err;
1061 }
1062
1063 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1064 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1065 err = -ENOSPC;
1066 goto err;
1067 }
1068
1069 if (dev->flags & IFF_UP) {
1070 err = veth_enable_xdp(dev);
1071 if (err) {
1072 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1073 goto err;
1074 }
1075 }
1076
1077 if (!old_prog) {
1078 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1079 peer->max_mtu = max_mtu;
1080 }
1081 }
1082
1083 if (old_prog) {
1084 if (!prog) {
1085 if (dev->flags & IFF_UP)
1086 veth_disable_xdp(dev);
1087
1088 if (peer) {
1089 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1090 peer->max_mtu = ETH_MAX_MTU;
1091 }
1092 }
1093 bpf_prog_put(old_prog);
1094 }
1095
1096 if ((!!old_prog ^ !!prog) && peer)
1097 netdev_update_features(peer);
1098
1099 return 0;
1100 err:
1101 priv->_xdp_prog = old_prog;
1102
1103 return err;
1104 }
1105
1106 static u32 veth_xdp_query(struct net_device *dev)
1107 {
1108 struct veth_priv *priv = netdev_priv(dev);
1109 const struct bpf_prog *xdp_prog;
1110
1111 xdp_prog = priv->_xdp_prog;
1112 if (xdp_prog)
1113 return xdp_prog->aux->id;
1114
1115 return 0;
1116 }
1117
1118 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1119 {
1120 switch (xdp->command) {
1121 case XDP_SETUP_PROG:
1122 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1123 case XDP_QUERY_PROG:
1124 xdp->prog_id = veth_xdp_query(dev);
1125 return 0;
1126 default:
1127 return -EINVAL;
1128 }
1129 }
1130
1131 static const struct net_device_ops veth_netdev_ops = {
1132 .ndo_init = veth_dev_init,
1133 .ndo_open = veth_open,
1134 .ndo_stop = veth_close,
1135 .ndo_start_xmit = veth_xmit,
1136 .ndo_get_stats64 = veth_get_stats64,
1137 .ndo_set_rx_mode = veth_set_multicast_list,
1138 .ndo_set_mac_address = eth_mac_addr,
1139 #ifdef CONFIG_NET_POLL_CONTROLLER
1140 .ndo_poll_controller = veth_poll_controller,
1141 #endif
1142 .ndo_get_iflink = veth_get_iflink,
1143 .ndo_fix_features = veth_fix_features,
1144 .ndo_features_check = passthru_features_check,
1145 .ndo_set_rx_headroom = veth_set_rx_headroom,
1146 .ndo_bpf = veth_xdp,
1147 .ndo_xdp_xmit = veth_xdp_xmit,
1148 };
1149
1150 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1151 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1152 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1153 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1154 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1155
1156 static void veth_setup(struct net_device *dev)
1157 {
1158 ether_setup(dev);
1159
1160 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1161 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1162 dev->priv_flags |= IFF_NO_QUEUE;
1163 dev->priv_flags |= IFF_PHONY_HEADROOM;
1164
1165 dev->netdev_ops = &veth_netdev_ops;
1166 dev->ethtool_ops = &veth_ethtool_ops;
1167 dev->features |= NETIF_F_LLTX;
1168 dev->features |= VETH_FEATURES;
1169 dev->vlan_features = dev->features &
1170 ~(NETIF_F_HW_VLAN_CTAG_TX |
1171 NETIF_F_HW_VLAN_STAG_TX |
1172 NETIF_F_HW_VLAN_CTAG_RX |
1173 NETIF_F_HW_VLAN_STAG_RX);
1174 dev->needs_free_netdev = true;
1175 dev->priv_destructor = veth_dev_free;
1176 dev->max_mtu = ETH_MAX_MTU;
1177
1178 dev->hw_features = VETH_FEATURES;
1179 dev->hw_enc_features = VETH_FEATURES;
1180 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1181 }
1182
1183 /*
1184 * netlink interface
1185 */
1186
1187 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1188 struct netlink_ext_ack *extack)
1189 {
1190 if (tb[IFLA_ADDRESS]) {
1191 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1192 return -EINVAL;
1193 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1194 return -EADDRNOTAVAIL;
1195 }
1196 if (tb[IFLA_MTU]) {
1197 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1198 return -EINVAL;
1199 }
1200 return 0;
1201 }
1202
1203 static struct rtnl_link_ops veth_link_ops;
1204
1205 static int veth_newlink(struct net *src_net, struct net_device *dev,
1206 struct nlattr *tb[], struct nlattr *data[],
1207 struct netlink_ext_ack *extack)
1208 {
1209 int err;
1210 struct net_device *peer;
1211 struct veth_priv *priv;
1212 char ifname[IFNAMSIZ];
1213 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1214 unsigned char name_assign_type;
1215 struct ifinfomsg *ifmp;
1216 struct net *net;
1217
1218 /*
1219 * create and register peer first
1220 */
1221 if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1222 struct nlattr *nla_peer;
1223
1224 nla_peer = data[VETH_INFO_PEER];
1225 ifmp = nla_data(nla_peer);
1226 err = rtnl_nla_parse_ifla(peer_tb,
1227 nla_data(nla_peer) + sizeof(struct ifinfomsg),
1228 nla_len(nla_peer) - sizeof(struct ifinfomsg),
1229 NULL);
1230 if (err < 0)
1231 return err;
1232
1233 err = veth_validate(peer_tb, NULL, extack);
1234 if (err < 0)
1235 return err;
1236
1237 tbp = peer_tb;
1238 } else {
1239 ifmp = NULL;
1240 tbp = tb;
1241 }
1242
1243 if (ifmp && tbp[IFLA_IFNAME]) {
1244 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1245 name_assign_type = NET_NAME_USER;
1246 } else {
1247 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1248 name_assign_type = NET_NAME_ENUM;
1249 }
1250
1251 net = rtnl_link_get_net(src_net, tbp);
1252 if (IS_ERR(net))
1253 return PTR_ERR(net);
1254
1255 peer = rtnl_create_link(net, ifname, name_assign_type,
1256 &veth_link_ops, tbp, extack);
1257 if (IS_ERR(peer)) {
1258 put_net(net);
1259 return PTR_ERR(peer);
1260 }
1261
1262 if (!ifmp || !tbp[IFLA_ADDRESS])
1263 eth_hw_addr_random(peer);
1264
1265 if (ifmp && (dev->ifindex != 0))
1266 peer->ifindex = ifmp->ifi_index;
1267
1268 peer->gso_max_size = dev->gso_max_size;
1269 peer->gso_max_segs = dev->gso_max_segs;
1270
1271 err = register_netdevice(peer);
1272 put_net(net);
1273 net = NULL;
1274 if (err < 0)
1275 goto err_register_peer;
1276
1277 netif_carrier_off(peer);
1278
1279 err = rtnl_configure_link(peer, ifmp);
1280 if (err < 0)
1281 goto err_configure_peer;
1282
1283 /*
1284 * register dev last
1285 *
1286 * note, that since we've registered new device the dev's name
1287 * should be re-allocated
1288 */
1289
1290 if (tb[IFLA_ADDRESS] == NULL)
1291 eth_hw_addr_random(dev);
1292
1293 if (tb[IFLA_IFNAME])
1294 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1295 else
1296 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1297
1298 err = register_netdevice(dev);
1299 if (err < 0)
1300 goto err_register_dev;
1301
1302 netif_carrier_off(dev);
1303
1304 /*
1305 * tie the deviced together
1306 */
1307
1308 priv = netdev_priv(dev);
1309 rcu_assign_pointer(priv->peer, peer);
1310
1311 priv = netdev_priv(peer);
1312 rcu_assign_pointer(priv->peer, dev);
1313
1314 return 0;
1315
1316 err_register_dev:
1317 /* nothing to do */
1318 err_configure_peer:
1319 unregister_netdevice(peer);
1320 return err;
1321
1322 err_register_peer:
1323 free_netdev(peer);
1324 return err;
1325 }
1326
1327 static void veth_dellink(struct net_device *dev, struct list_head *head)
1328 {
1329 struct veth_priv *priv;
1330 struct net_device *peer;
1331
1332 priv = netdev_priv(dev);
1333 peer = rtnl_dereference(priv->peer);
1334
1335 /* Note : dellink() is called from default_device_exit_batch(),
1336 * before a rcu_synchronize() point. The devices are guaranteed
1337 * not being freed before one RCU grace period.
1338 */
1339 RCU_INIT_POINTER(priv->peer, NULL);
1340 unregister_netdevice_queue(dev, head);
1341
1342 if (peer) {
1343 priv = netdev_priv(peer);
1344 RCU_INIT_POINTER(priv->peer, NULL);
1345 unregister_netdevice_queue(peer, head);
1346 }
1347 }
1348
1349 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1350 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) },
1351 };
1352
1353 static struct net *veth_get_link_net(const struct net_device *dev)
1354 {
1355 struct veth_priv *priv = netdev_priv(dev);
1356 struct net_device *peer = rtnl_dereference(priv->peer);
1357
1358 return peer ? dev_net(peer) : dev_net(dev);
1359 }
1360
1361 static struct rtnl_link_ops veth_link_ops = {
1362 .kind = DRV_NAME,
1363 .priv_size = sizeof(struct veth_priv),
1364 .setup = veth_setup,
1365 .validate = veth_validate,
1366 .newlink = veth_newlink,
1367 .dellink = veth_dellink,
1368 .policy = veth_policy,
1369 .maxtype = VETH_INFO_MAX,
1370 .get_link_net = veth_get_link_net,
1371 };
1372
1373 /*
1374 * init/fini
1375 */
1376
1377 static __init int veth_init(void)
1378 {
1379 return rtnl_link_register(&veth_link_ops);
1380 }
1381
1382 static __exit void veth_exit(void)
1383 {
1384 rtnl_link_unregister(&veth_link_ops);
1385 }
1386
1387 module_init(veth_init);
1388 module_exit(veth_exit);
1389
1390 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1391 MODULE_LICENSE("GPL v2");
1392 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
Linux with added FPC-III support