1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11 #include <linux/init.h>
12 #include <linux/atomic.h>
13 #include <linux/ethtool.h>
14 #include <linux/module.h>
15 #include <linux/highmem.h>
16 #include <linux/device.h>
17 #include <linux/io.h>
18 #include <linux/delay.h>
19 #include <linux/netdevice.h>
20 #include <linux/inetdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/pci.h>
23 #include <linux/skbuff.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/slab.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/netpoll.h>
29 #include <linux/bpf.h>
30
31 #include <net/arp.h>
32 #include <net/route.h>
33 #include <net/sock.h>
34 #include <net/pkt_sched.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37
38 #include "hyperv_net.h"
39
40 #define RING_SIZE_MIN 64
41
42 #define LINKCHANGE_INT (2 * HZ)
43 #define VF_TAKEOVER_INT (HZ / 10)
44
45 static unsigned int ring_size __ro_after_init = 128;
46 module_param(ring_size, uint, 0444);
47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
48 unsigned int netvsc_ring_bytes __ro_after_init;
49
50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
53 NETIF_MSG_TX_ERR;
54
55 static int debug = -1;
56 module_param(debug, int, 0444);
57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
58
59 static LIST_HEAD(netvsc_dev_list);
60
netvsc_change_rx_flags(struct net_device * net,int change)61 static void netvsc_change_rx_flags(struct net_device *net, int change)
62 {
63 struct net_device_context *ndev_ctx = netdev_priv(net);
64 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
65 int inc;
66
67 if (!vf_netdev)
68 return;
69
70 if (change & IFF_PROMISC) {
71 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
72 dev_set_promiscuity(vf_netdev, inc);
73 }
74
75 if (change & IFF_ALLMULTI) {
76 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
77 dev_set_allmulti(vf_netdev, inc);
78 }
79 }
80
netvsc_set_rx_mode(struct net_device * net)81 static void netvsc_set_rx_mode(struct net_device *net)
82 {
83 struct net_device_context *ndev_ctx = netdev_priv(net);
84 struct net_device *vf_netdev;
85 struct netvsc_device *nvdev;
86
87 rcu_read_lock();
88 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
89 if (vf_netdev) {
90 dev_uc_sync(vf_netdev, net);
91 dev_mc_sync(vf_netdev, net);
92 }
93
94 nvdev = rcu_dereference(ndev_ctx->nvdev);
95 if (nvdev)
96 rndis_filter_update(nvdev);
97 rcu_read_unlock();
98 }
99
netvsc_tx_enable(struct netvsc_device * nvscdev,struct net_device * ndev)100 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
101 struct net_device *ndev)
102 {
103 nvscdev->tx_disable = false;
104 virt_wmb(); /* ensure queue wake up mechanism is on */
105
106 netif_tx_wake_all_queues(ndev);
107 }
108
netvsc_open(struct net_device * net)109 static int netvsc_open(struct net_device *net)
110 {
111 struct net_device_context *ndev_ctx = netdev_priv(net);
112 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
113 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
114 struct rndis_device *rdev;
115 int ret = 0;
116
117 netif_carrier_off(net);
118
119 /* Open up the device */
120 ret = rndis_filter_open(nvdev);
121 if (ret != 0) {
122 netdev_err(net, "unable to open device (ret %d).\n", ret);
123 return ret;
124 }
125
126 rdev = nvdev->extension;
127 if (!rdev->link_state) {
128 netif_carrier_on(net);
129 netvsc_tx_enable(nvdev, net);
130 }
131
132 if (vf_netdev) {
133 /* Setting synthetic device up transparently sets
134 * slave as up. If open fails, then slave will be
135 * still be offline (and not used).
136 */
137 ret = dev_open(vf_netdev, NULL);
138 if (ret)
139 netdev_warn(net,
140 "unable to open slave: %s: %d\n",
141 vf_netdev->name, ret);
142 }
143 return 0;
144 }
145
netvsc_wait_until_empty(struct netvsc_device * nvdev)146 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
147 {
148 unsigned int retry = 0;
149 int i;
150
151 /* Ensure pending bytes in ring are read */
152 for (;;) {
153 u32 aread = 0;
154
155 for (i = 0; i < nvdev->num_chn; i++) {
156 struct vmbus_channel *chn
157 = nvdev->chan_table[i].channel;
158
159 if (!chn)
160 continue;
161
162 /* make sure receive not running now */
163 napi_synchronize(&nvdev->chan_table[i].napi);
164
165 aread = hv_get_bytes_to_read(&chn->inbound);
166 if (aread)
167 break;
168
169 aread = hv_get_bytes_to_read(&chn->outbound);
170 if (aread)
171 break;
172 }
173
174 if (aread == 0)
175 return 0;
176
177 if (++retry > RETRY_MAX)
178 return -ETIMEDOUT;
179
180 usleep_range(RETRY_US_LO, RETRY_US_HI);
181 }
182 }
183
netvsc_tx_disable(struct netvsc_device * nvscdev,struct net_device * ndev)184 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
185 struct net_device *ndev)
186 {
187 if (nvscdev) {
188 nvscdev->tx_disable = true;
189 virt_wmb(); /* ensure txq will not wake up after stop */
190 }
191
192 netif_tx_disable(ndev);
193 }
194
netvsc_close(struct net_device * net)195 static int netvsc_close(struct net_device *net)
196 {
197 struct net_device_context *net_device_ctx = netdev_priv(net);
198 struct net_device *vf_netdev
199 = rtnl_dereference(net_device_ctx->vf_netdev);
200 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
201 int ret;
202
203 netvsc_tx_disable(nvdev, net);
204
205 /* No need to close rndis filter if it is removed already */
206 if (!nvdev)
207 return 0;
208
209 ret = rndis_filter_close(nvdev);
210 if (ret != 0) {
211 netdev_err(net, "unable to close device (ret %d).\n", ret);
212 return ret;
213 }
214
215 ret = netvsc_wait_until_empty(nvdev);
216 if (ret)
217 netdev_err(net, "Ring buffer not empty after closing rndis\n");
218
219 if (vf_netdev)
220 dev_close(vf_netdev);
221
222 return ret;
223 }
224
init_ppi_data(struct rndis_message * msg,u32 ppi_size,u32 pkt_type)225 static inline void *init_ppi_data(struct rndis_message *msg,
226 u32 ppi_size, u32 pkt_type)
227 {
228 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
229 struct rndis_per_packet_info *ppi;
230
231 rndis_pkt->data_offset += ppi_size;
232 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
233 + rndis_pkt->per_pkt_info_len;
234
235 ppi->size = ppi_size;
236 ppi->type = pkt_type;
237 ppi->internal = 0;
238 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
239
240 rndis_pkt->per_pkt_info_len += ppi_size;
241
242 return ppi + 1;
243 }
244
245 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
246 * packets. We can use ethtool to change UDP hash level when necessary.
247 */
netvsc_get_hash(struct sk_buff * skb,const struct net_device_context * ndc)248 static inline u32 netvsc_get_hash(
249 struct sk_buff *skb,
250 const struct net_device_context *ndc)
251 {
252 struct flow_keys flow;
253 u32 hash, pkt_proto = 0;
254 static u32 hashrnd __read_mostly;
255
256 net_get_random_once(&hashrnd, sizeof(hashrnd));
257
258 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
259 return 0;
260
261 switch (flow.basic.ip_proto) {
262 case IPPROTO_TCP:
263 if (flow.basic.n_proto == htons(ETH_P_IP))
264 pkt_proto = HV_TCP4_L4HASH;
265 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
266 pkt_proto = HV_TCP6_L4HASH;
267
268 break;
269
270 case IPPROTO_UDP:
271 if (flow.basic.n_proto == htons(ETH_P_IP))
272 pkt_proto = HV_UDP4_L4HASH;
273 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
274 pkt_proto = HV_UDP6_L4HASH;
275
276 break;
277 }
278
279 if (pkt_proto & ndc->l4_hash) {
280 return skb_get_hash(skb);
281 } else {
282 if (flow.basic.n_proto == htons(ETH_P_IP))
283 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
284 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
285 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
286 else
287 return 0;
288
289 __skb_set_sw_hash(skb, hash, false);
290 }
291
292 return hash;
293 }
294
netvsc_get_tx_queue(struct net_device * ndev,struct sk_buff * skb,int old_idx)295 static inline int netvsc_get_tx_queue(struct net_device *ndev,
296 struct sk_buff *skb, int old_idx)
297 {
298 const struct net_device_context *ndc = netdev_priv(ndev);
299 struct sock *sk = skb->sk;
300 int q_idx;
301
302 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
303 (VRSS_SEND_TAB_SIZE - 1)];
304
305 /* If queue index changed record the new value */
306 if (q_idx != old_idx &&
307 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
308 sk_tx_queue_set(sk, q_idx);
309
310 return q_idx;
311 }
312
313 /*
314 * Select queue for transmit.
315 *
316 * If a valid queue has already been assigned, then use that.
317 * Otherwise compute tx queue based on hash and the send table.
318 *
319 * This is basically similar to default (netdev_pick_tx) with the added step
320 * of using the host send_table when no other queue has been assigned.
321 *
322 * TODO support XPS - but get_xps_queue not exported
323 */
netvsc_pick_tx(struct net_device * ndev,struct sk_buff * skb)324 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
325 {
326 int q_idx = sk_tx_queue_get(skb->sk);
327
328 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
329 /* If forwarding a packet, we use the recorded queue when
330 * available for better cache locality.
331 */
332 if (skb_rx_queue_recorded(skb))
333 q_idx = skb_get_rx_queue(skb);
334 else
335 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
336 }
337
338 return q_idx;
339 }
340
netvsc_select_queue(struct net_device * ndev,struct sk_buff * skb,struct net_device * sb_dev)341 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
342 struct net_device *sb_dev)
343 {
344 struct net_device_context *ndc = netdev_priv(ndev);
345 struct net_device *vf_netdev;
346 u16 txq;
347
348 rcu_read_lock();
349 vf_netdev = rcu_dereference(ndc->vf_netdev);
350 if (vf_netdev) {
351 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
352
353 if (vf_ops->ndo_select_queue)
354 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
355 else
356 txq = netdev_pick_tx(vf_netdev, skb, NULL);
357
358 /* Record the queue selected by VF so that it can be
359 * used for common case where VF has more queues than
360 * the synthetic device.
361 */
362 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
363 } else {
364 txq = netvsc_pick_tx(ndev, skb);
365 }
366 rcu_read_unlock();
367
368 while (txq >= ndev->real_num_tx_queues)
369 txq -= ndev->real_num_tx_queues;
370
371 return txq;
372 }
373
fill_pg_buf(unsigned long hvpfn,u32 offset,u32 len,struct hv_page_buffer * pb)374 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
375 struct hv_page_buffer *pb)
376 {
377 int j = 0;
378
379 hvpfn += offset >> HV_HYP_PAGE_SHIFT;
380 offset = offset & ~HV_HYP_PAGE_MASK;
381
382 while (len > 0) {
383 unsigned long bytes;
384
385 bytes = HV_HYP_PAGE_SIZE - offset;
386 if (bytes > len)
387 bytes = len;
388 pb[j].pfn = hvpfn;
389 pb[j].offset = offset;
390 pb[j].len = bytes;
391
392 offset += bytes;
393 len -= bytes;
394
395 if (offset == HV_HYP_PAGE_SIZE && len) {
396 hvpfn++;
397 offset = 0;
398 j++;
399 }
400 }
401
402 return j + 1;
403 }
404
init_page_array(void * hdr,u32 len,struct sk_buff * skb,struct hv_netvsc_packet * packet,struct hv_page_buffer * pb)405 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
406 struct hv_netvsc_packet *packet,
407 struct hv_page_buffer *pb)
408 {
409 u32 slots_used = 0;
410 char *data = skb->data;
411 int frags = skb_shinfo(skb)->nr_frags;
412 int i;
413
414 /* The packet is laid out thus:
415 * 1. hdr: RNDIS header and PPI
416 * 2. skb linear data
417 * 3. skb fragment data
418 */
419 slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
420 offset_in_hvpage(hdr),
421 len,
422 &pb[slots_used]);
423
424 packet->rmsg_size = len;
425 packet->rmsg_pgcnt = slots_used;
426
427 slots_used += fill_pg_buf(virt_to_hvpfn(data),
428 offset_in_hvpage(data),
429 skb_headlen(skb),
430 &pb[slots_used]);
431
432 for (i = 0; i < frags; i++) {
433 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
434
435 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
436 skb_frag_off(frag),
437 skb_frag_size(frag),
438 &pb[slots_used]);
439 }
440 return slots_used;
441 }
442
count_skb_frag_slots(struct sk_buff * skb)443 static int count_skb_frag_slots(struct sk_buff *skb)
444 {
445 int i, frags = skb_shinfo(skb)->nr_frags;
446 int pages = 0;
447
448 for (i = 0; i < frags; i++) {
449 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
450 unsigned long size = skb_frag_size(frag);
451 unsigned long offset = skb_frag_off(frag);
452
453 /* Skip unused frames from start of page */
454 offset &= ~HV_HYP_PAGE_MASK;
455 pages += HVPFN_UP(offset + size);
456 }
457 return pages;
458 }
459
netvsc_get_slots(struct sk_buff * skb)460 static int netvsc_get_slots(struct sk_buff *skb)
461 {
462 char *data = skb->data;
463 unsigned int offset = offset_in_hvpage(data);
464 unsigned int len = skb_headlen(skb);
465 int slots;
466 int frag_slots;
467
468 slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
469 frag_slots = count_skb_frag_slots(skb);
470 return slots + frag_slots;
471 }
472
net_checksum_info(struct sk_buff * skb)473 static u32 net_checksum_info(struct sk_buff *skb)
474 {
475 if (skb->protocol == htons(ETH_P_IP)) {
476 struct iphdr *ip = ip_hdr(skb);
477
478 if (ip->protocol == IPPROTO_TCP)
479 return TRANSPORT_INFO_IPV4_TCP;
480 else if (ip->protocol == IPPROTO_UDP)
481 return TRANSPORT_INFO_IPV4_UDP;
482 } else {
483 struct ipv6hdr *ip6 = ipv6_hdr(skb);
484
485 if (ip6->nexthdr == IPPROTO_TCP)
486 return TRANSPORT_INFO_IPV6_TCP;
487 else if (ip6->nexthdr == IPPROTO_UDP)
488 return TRANSPORT_INFO_IPV6_UDP;
489 }
490
491 return TRANSPORT_INFO_NOT_IP;
492 }
493
494 /* Send skb on the slave VF device. */
netvsc_vf_xmit(struct net_device * net,struct net_device * vf_netdev,struct sk_buff * skb)495 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
496 struct sk_buff *skb)
497 {
498 struct net_device_context *ndev_ctx = netdev_priv(net);
499 unsigned int len = skb->len;
500 int rc;
501
502 skb->dev = vf_netdev;
503 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
504
505 rc = dev_queue_xmit(skb);
506 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
507 struct netvsc_vf_pcpu_stats *pcpu_stats
508 = this_cpu_ptr(ndev_ctx->vf_stats);
509
510 u64_stats_update_begin(&pcpu_stats->syncp);
511 pcpu_stats->tx_packets++;
512 pcpu_stats->tx_bytes += len;
513 u64_stats_update_end(&pcpu_stats->syncp);
514 } else {
515 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
516 }
517
518 return rc;
519 }
520
netvsc_xmit(struct sk_buff * skb,struct net_device * net,bool xdp_tx)521 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
522 {
523 struct net_device_context *net_device_ctx = netdev_priv(net);
524 struct hv_netvsc_packet *packet = NULL;
525 int ret;
526 unsigned int num_data_pgs;
527 struct rndis_message *rndis_msg;
528 struct net_device *vf_netdev;
529 u32 rndis_msg_size;
530 u32 hash;
531 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
532
533 /* If VF is present and up then redirect packets to it.
534 * Skip the VF if it is marked down or has no carrier.
535 * If netpoll is in uses, then VF can not be used either.
536 */
537 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
538 if (vf_netdev && netif_running(vf_netdev) &&
539 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
540 net_device_ctx->data_path_is_vf)
541 return netvsc_vf_xmit(net, vf_netdev, skb);
542
543 /* We will atmost need two pages to describe the rndis
544 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
545 * of pages in a single packet. If skb is scattered around
546 * more pages we try linearizing it.
547 */
548
549 num_data_pgs = netvsc_get_slots(skb) + 2;
550
551 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
552 ++net_device_ctx->eth_stats.tx_scattered;
553
554 if (skb_linearize(skb))
555 goto no_memory;
556
557 num_data_pgs = netvsc_get_slots(skb) + 2;
558 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
559 ++net_device_ctx->eth_stats.tx_too_big;
560 goto drop;
561 }
562 }
563
564 /*
565 * Place the rndis header in the skb head room and
566 * the skb->cb will be used for hv_netvsc_packet
567 * structure.
568 */
569 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
570 if (ret)
571 goto no_memory;
572
573 /* Use the skb control buffer for building up the packet */
574 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
575 sizeof_field(struct sk_buff, cb));
576 packet = (struct hv_netvsc_packet *)skb->cb;
577
578 packet->q_idx = skb_get_queue_mapping(skb);
579
580 packet->total_data_buflen = skb->len;
581 packet->total_bytes = skb->len;
582 packet->total_packets = 1;
583
584 rndis_msg = (struct rndis_message *)skb->head;
585
586 /* Add the rndis header */
587 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
588 rndis_msg->msg_len = packet->total_data_buflen;
589
590 rndis_msg->msg.pkt = (struct rndis_packet) {
591 .data_offset = sizeof(struct rndis_packet),
592 .data_len = packet->total_data_buflen,
593 .per_pkt_info_offset = sizeof(struct rndis_packet),
594 };
595
596 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
597
598 hash = skb_get_hash_raw(skb);
599 if (hash != 0 && net->real_num_tx_queues > 1) {
600 u32 *hash_info;
601
602 rndis_msg_size += NDIS_HASH_PPI_SIZE;
603 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
604 NBL_HASH_VALUE);
605 *hash_info = hash;
606 }
607
608 /* When using AF_PACKET we need to drop VLAN header from
609 * the frame and update the SKB to allow the HOST OS
610 * to transmit the 802.1Q packet
611 */
612 if (skb->protocol == htons(ETH_P_8021Q)) {
613 u16 vlan_tci;
614
615 skb_reset_mac_header(skb);
616 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
617 if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
618 ++net_device_ctx->eth_stats.vlan_error;
619 goto drop;
620 }
621
622 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
623 /* Update the NDIS header pkt lengths */
624 packet->total_data_buflen -= VLAN_HLEN;
625 packet->total_bytes -= VLAN_HLEN;
626 rndis_msg->msg_len = packet->total_data_buflen;
627 rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
628 }
629 }
630
631 if (skb_vlan_tag_present(skb)) {
632 struct ndis_pkt_8021q_info *vlan;
633
634 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
635 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
636 IEEE_8021Q_INFO);
637
638 vlan->value = 0;
639 vlan->vlanid = skb_vlan_tag_get_id(skb);
640 vlan->cfi = skb_vlan_tag_get_cfi(skb);
641 vlan->pri = skb_vlan_tag_get_prio(skb);
642 }
643
644 if (skb_is_gso(skb)) {
645 struct ndis_tcp_lso_info *lso_info;
646
647 rndis_msg_size += NDIS_LSO_PPI_SIZE;
648 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
649 TCP_LARGESEND_PKTINFO);
650
651 lso_info->value = 0;
652 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
653 if (skb->protocol == htons(ETH_P_IP)) {
654 lso_info->lso_v2_transmit.ip_version =
655 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
656 ip_hdr(skb)->tot_len = 0;
657 ip_hdr(skb)->check = 0;
658 tcp_hdr(skb)->check =
659 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
660 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
661 } else {
662 lso_info->lso_v2_transmit.ip_version =
663 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
664 tcp_v6_gso_csum_prep(skb);
665 }
666 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
667 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
668 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
669 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
670 struct ndis_tcp_ip_checksum_info *csum_info;
671
672 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
673 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
674 TCPIP_CHKSUM_PKTINFO);
675
676 csum_info->value = 0;
677 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
678
679 if (skb->protocol == htons(ETH_P_IP)) {
680 csum_info->transmit.is_ipv4 = 1;
681
682 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
683 csum_info->transmit.tcp_checksum = 1;
684 else
685 csum_info->transmit.udp_checksum = 1;
686 } else {
687 csum_info->transmit.is_ipv6 = 1;
688
689 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
690 csum_info->transmit.tcp_checksum = 1;
691 else
692 csum_info->transmit.udp_checksum = 1;
693 }
694 } else {
695 /* Can't do offload of this type of checksum */
696 if (skb_checksum_help(skb))
697 goto drop;
698 }
699 }
700
701 /* Start filling in the page buffers with the rndis hdr */
702 rndis_msg->msg_len += rndis_msg_size;
703 packet->total_data_buflen = rndis_msg->msg_len;
704 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
705 skb, packet, pb);
706
707 /* timestamp packet in software */
708 skb_tx_timestamp(skb);
709
710 ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
711 if (likely(ret == 0))
712 return NETDEV_TX_OK;
713
714 if (ret == -EAGAIN) {
715 ++net_device_ctx->eth_stats.tx_busy;
716 return NETDEV_TX_BUSY;
717 }
718
719 if (ret == -ENOSPC)
720 ++net_device_ctx->eth_stats.tx_no_space;
721
722 drop:
723 dev_kfree_skb_any(skb);
724 net->stats.tx_dropped++;
725
726 return NETDEV_TX_OK;
727
728 no_memory:
729 ++net_device_ctx->eth_stats.tx_no_memory;
730 goto drop;
731 }
732
netvsc_start_xmit(struct sk_buff * skb,struct net_device * ndev)733 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
734 struct net_device *ndev)
735 {
736 return netvsc_xmit(skb, ndev, false);
737 }
738
739 /*
740 * netvsc_linkstatus_callback - Link up/down notification
741 */
netvsc_linkstatus_callback(struct net_device * net,struct rndis_message * resp,void * data,u32 data_buflen)742 void netvsc_linkstatus_callback(struct net_device *net,
743 struct rndis_message *resp,
744 void *data, u32 data_buflen)
745 {
746 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
747 struct net_device_context *ndev_ctx = netdev_priv(net);
748 struct netvsc_reconfig *event;
749 unsigned long flags;
750
751 /* Ensure the packet is big enough to access its fields */
752 if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
753 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
754 resp->msg_len);
755 return;
756 }
757
758 /* Copy the RNDIS indicate status into nvchan->recv_buf */
759 memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
760
761 /* Update the physical link speed when changing to another vSwitch */
762 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
763 u32 speed;
764
765 /* Validate status_buf_offset and status_buflen.
766 *
767 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
768 * for the status buffer field in resp->msg_len; perform the validation
769 * using data_buflen (>= resp->msg_len).
770 */
771 if (indicate->status_buflen < sizeof(speed) ||
772 indicate->status_buf_offset < sizeof(*indicate) ||
773 data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
774 data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
775 < indicate->status_buflen) {
776 netdev_err(net, "invalid rndis_indicate_status packet\n");
777 return;
778 }
779
780 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
781 ndev_ctx->speed = speed;
782 return;
783 }
784
785 /* Handle these link change statuses below */
786 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
787 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
788 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
789 return;
790
791 if (net->reg_state != NETREG_REGISTERED)
792 return;
793
794 event = kzalloc(sizeof(*event), GFP_ATOMIC);
795 if (!event)
796 return;
797 event->event = indicate->status;
798
799 spin_lock_irqsave(&ndev_ctx->lock, flags);
800 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
801 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
802
803 schedule_delayed_work(&ndev_ctx->dwork, 0);
804 }
805
806 /* This function should only be called after skb_record_rx_queue() */
netvsc_xdp_xmit(struct sk_buff * skb,struct net_device * ndev)807 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
808 {
809 int rc;
810
811 skb->queue_mapping = skb_get_rx_queue(skb);
812 __skb_push(skb, ETH_HLEN);
813
814 rc = netvsc_xmit(skb, ndev, true);
815
816 if (dev_xmit_complete(rc))
817 return;
818
819 dev_kfree_skb_any(skb);
820 ndev->stats.tx_dropped++;
821 }
822
netvsc_comp_ipcsum(struct sk_buff * skb)823 static void netvsc_comp_ipcsum(struct sk_buff *skb)
824 {
825 struct iphdr *iph = (struct iphdr *)skb->data;
826
827 iph->check = 0;
828 iph->check = ip_fast_csum(iph, iph->ihl);
829 }
830
netvsc_alloc_recv_skb(struct net_device * net,struct netvsc_channel * nvchan,struct xdp_buff * xdp)831 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
832 struct netvsc_channel *nvchan,
833 struct xdp_buff *xdp)
834 {
835 struct napi_struct *napi = &nvchan->napi;
836 const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
837 const struct ndis_tcp_ip_checksum_info *csum_info =
838 &nvchan->rsc.csum_info;
839 const u32 *hash_info = &nvchan->rsc.hash_info;
840 u8 ppi_flags = nvchan->rsc.ppi_flags;
841 struct sk_buff *skb;
842 void *xbuf = xdp->data_hard_start;
843 int i;
844
845 if (xbuf) {
846 unsigned int hdroom = xdp->data - xdp->data_hard_start;
847 unsigned int xlen = xdp->data_end - xdp->data;
848 unsigned int frag_size = xdp->frame_sz;
849
850 skb = build_skb(xbuf, frag_size);
851
852 if (!skb) {
853 __free_page(virt_to_page(xbuf));
854 return NULL;
855 }
856
857 skb_reserve(skb, hdroom);
858 skb_put(skb, xlen);
859 skb->dev = napi->dev;
860 } else {
861 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
862
863 if (!skb)
864 return NULL;
865
866 /* Copy to skb. This copy is needed here since the memory
867 * pointed by hv_netvsc_packet cannot be deallocated.
868 */
869 for (i = 0; i < nvchan->rsc.cnt; i++)
870 skb_put_data(skb, nvchan->rsc.data[i],
871 nvchan->rsc.len[i]);
872 }
873
874 skb->protocol = eth_type_trans(skb, net);
875
876 /* skb is already created with CHECKSUM_NONE */
877 skb_checksum_none_assert(skb);
878
879 /* Incoming packets may have IP header checksum verified by the host.
880 * They may not have IP header checksum computed after coalescing.
881 * We compute it here if the flags are set, because on Linux, the IP
882 * checksum is always checked.
883 */
884 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
885 csum_info->receive.ip_checksum_succeeded &&
886 skb->protocol == htons(ETH_P_IP)) {
887 /* Check that there is enough space to hold the IP header. */
888 if (skb_headlen(skb) < sizeof(struct iphdr)) {
889 kfree_skb(skb);
890 return NULL;
891 }
892 netvsc_comp_ipcsum(skb);
893 }
894
895 /* Do L4 checksum offload if enabled and present. */
896 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
897 if (csum_info->receive.tcp_checksum_succeeded ||
898 csum_info->receive.udp_checksum_succeeded)
899 skb->ip_summed = CHECKSUM_UNNECESSARY;
900 }
901
902 if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
903 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
904
905 if (ppi_flags & NVSC_RSC_VLAN) {
906 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
907 (vlan->cfi ? VLAN_CFI_MASK : 0);
908
909 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
910 vlan_tci);
911 }
912
913 return skb;
914 }
915
916 /*
917 * netvsc_recv_callback - Callback when we receive a packet from the
918 * "wire" on the specified device.
919 */
netvsc_recv_callback(struct net_device * net,struct netvsc_device * net_device,struct netvsc_channel * nvchan)920 int netvsc_recv_callback(struct net_device *net,
921 struct netvsc_device *net_device,
922 struct netvsc_channel *nvchan)
923 {
924 struct net_device_context *net_device_ctx = netdev_priv(net);
925 struct vmbus_channel *channel = nvchan->channel;
926 u16 q_idx = channel->offermsg.offer.sub_channel_index;
927 struct sk_buff *skb;
928 struct netvsc_stats *rx_stats = &nvchan->rx_stats;
929 struct xdp_buff xdp;
930 u32 act;
931
932 if (net->reg_state != NETREG_REGISTERED)
933 return NVSP_STAT_FAIL;
934
935 act = netvsc_run_xdp(net, nvchan, &xdp);
936
937 if (act != XDP_PASS && act != XDP_TX) {
938 u64_stats_update_begin(&rx_stats->syncp);
939 rx_stats->xdp_drop++;
940 u64_stats_update_end(&rx_stats->syncp);
941
942 return NVSP_STAT_SUCCESS; /* consumed by XDP */
943 }
944
945 /* Allocate a skb - TODO direct I/O to pages? */
946 skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
947
948 if (unlikely(!skb)) {
949 ++net_device_ctx->eth_stats.rx_no_memory;
950 return NVSP_STAT_FAIL;
951 }
952
953 skb_record_rx_queue(skb, q_idx);
954
955 /*
956 * Even if injecting the packet, record the statistics
957 * on the synthetic device because modifying the VF device
958 * statistics will not work correctly.
959 */
960 u64_stats_update_begin(&rx_stats->syncp);
961 rx_stats->packets++;
962 rx_stats->bytes += nvchan->rsc.pktlen;
963
964 if (skb->pkt_type == PACKET_BROADCAST)
965 ++rx_stats->broadcast;
966 else if (skb->pkt_type == PACKET_MULTICAST)
967 ++rx_stats->multicast;
968 u64_stats_update_end(&rx_stats->syncp);
969
970 if (act == XDP_TX) {
971 netvsc_xdp_xmit(skb, net);
972 return NVSP_STAT_SUCCESS;
973 }
974
975 napi_gro_receive(&nvchan->napi, skb);
976 return NVSP_STAT_SUCCESS;
977 }
978
netvsc_get_drvinfo(struct net_device * net,struct ethtool_drvinfo * info)979 static void netvsc_get_drvinfo(struct net_device *net,
980 struct ethtool_drvinfo *info)
981 {
982 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
983 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
984 }
985
netvsc_get_channels(struct net_device * net,struct ethtool_channels * channel)986 static void netvsc_get_channels(struct net_device *net,
987 struct ethtool_channels *channel)
988 {
989 struct net_device_context *net_device_ctx = netdev_priv(net);
990 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
991
992 if (nvdev) {
993 channel->max_combined = nvdev->max_chn;
994 channel->combined_count = nvdev->num_chn;
995 }
996 }
997
998 /* Alloc struct netvsc_device_info, and initialize it from either existing
999 * struct netvsc_device, or from default values.
1000 */
1001 static
netvsc_devinfo_get(struct netvsc_device * nvdev)1002 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
1003 {
1004 struct netvsc_device_info *dev_info;
1005 struct bpf_prog *prog;
1006
1007 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1008
1009 if (!dev_info)
1010 return NULL;
1011
1012 if (nvdev) {
1013 ASSERT_RTNL();
1014
1015 dev_info->num_chn = nvdev->num_chn;
1016 dev_info->send_sections = nvdev->send_section_cnt;
1017 dev_info->send_section_size = nvdev->send_section_size;
1018 dev_info->recv_sections = nvdev->recv_section_cnt;
1019 dev_info->recv_section_size = nvdev->recv_section_size;
1020
1021 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1022 NETVSC_HASH_KEYLEN);
1023
1024 prog = netvsc_xdp_get(nvdev);
1025 if (prog) {
1026 bpf_prog_inc(prog);
1027 dev_info->bprog = prog;
1028 }
1029 } else {
1030 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1031 dev_info->send_sections = NETVSC_DEFAULT_TX;
1032 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1033 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1034 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1035 }
1036
1037 return dev_info;
1038 }
1039
1040 /* Free struct netvsc_device_info */
netvsc_devinfo_put(struct netvsc_device_info * dev_info)1041 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1042 {
1043 if (dev_info->bprog) {
1044 ASSERT_RTNL();
1045 bpf_prog_put(dev_info->bprog);
1046 }
1047
1048 kfree(dev_info);
1049 }
1050
netvsc_detach(struct net_device * ndev,struct netvsc_device * nvdev)1051 static int netvsc_detach(struct net_device *ndev,
1052 struct netvsc_device *nvdev)
1053 {
1054 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1055 struct hv_device *hdev = ndev_ctx->device_ctx;
1056 int ret;
1057
1058 /* Don't try continuing to try and setup sub channels */
1059 if (cancel_work_sync(&nvdev->subchan_work))
1060 nvdev->num_chn = 1;
1061
1062 netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1063
1064 /* If device was up (receiving) then shutdown */
1065 if (netif_running(ndev)) {
1066 netvsc_tx_disable(nvdev, ndev);
1067
1068 ret = rndis_filter_close(nvdev);
1069 if (ret) {
1070 netdev_err(ndev,
1071 "unable to close device (ret %d).\n", ret);
1072 return ret;
1073 }
1074
1075 ret = netvsc_wait_until_empty(nvdev);
1076 if (ret) {
1077 netdev_err(ndev,
1078 "Ring buffer not empty after closing rndis\n");
1079 return ret;
1080 }
1081 }
1082
1083 netif_device_detach(ndev);
1084
1085 rndis_filter_device_remove(hdev, nvdev);
1086
1087 return 0;
1088 }
1089
netvsc_attach(struct net_device * ndev,struct netvsc_device_info * dev_info)1090 static int netvsc_attach(struct net_device *ndev,
1091 struct netvsc_device_info *dev_info)
1092 {
1093 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1094 struct hv_device *hdev = ndev_ctx->device_ctx;
1095 struct netvsc_device *nvdev;
1096 struct rndis_device *rdev;
1097 struct bpf_prog *prog;
1098 int ret = 0;
1099
1100 nvdev = rndis_filter_device_add(hdev, dev_info);
1101 if (IS_ERR(nvdev))
1102 return PTR_ERR(nvdev);
1103
1104 if (nvdev->num_chn > 1) {
1105 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1106
1107 /* if unavailable, just proceed with one queue */
1108 if (ret) {
1109 nvdev->max_chn = 1;
1110 nvdev->num_chn = 1;
1111 }
1112 }
1113
1114 prog = dev_info->bprog;
1115 if (prog) {
1116 bpf_prog_inc(prog);
1117 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1118 if (ret) {
1119 bpf_prog_put(prog);
1120 goto err1;
1121 }
1122 }
1123
1124 /* In any case device is now ready */
1125 nvdev->tx_disable = false;
1126 netif_device_attach(ndev);
1127
1128 /* Note: enable and attach happen when sub-channels setup */
1129 netif_carrier_off(ndev);
1130
1131 if (netif_running(ndev)) {
1132 ret = rndis_filter_open(nvdev);
1133 if (ret)
1134 goto err2;
1135
1136 rdev = nvdev->extension;
1137 if (!rdev->link_state)
1138 netif_carrier_on(ndev);
1139 }
1140
1141 return 0;
1142
1143 err2:
1144 netif_device_detach(ndev);
1145
1146 err1:
1147 rndis_filter_device_remove(hdev, nvdev);
1148
1149 return ret;
1150 }
1151
netvsc_set_channels(struct net_device * net,struct ethtool_channels * channels)1152 static int netvsc_set_channels(struct net_device *net,
1153 struct ethtool_channels *channels)
1154 {
1155 struct net_device_context *net_device_ctx = netdev_priv(net);
1156 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1157 unsigned int orig, count = channels->combined_count;
1158 struct netvsc_device_info *device_info;
1159 int ret;
1160
1161 /* We do not support separate count for rx, tx, or other */
1162 if (count == 0 ||
1163 channels->rx_count || channels->tx_count || channels->other_count)
1164 return -EINVAL;
1165
1166 if (!nvdev || nvdev->destroy)
1167 return -ENODEV;
1168
1169 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1170 return -EINVAL;
1171
1172 if (count > nvdev->max_chn)
1173 return -EINVAL;
1174
1175 orig = nvdev->num_chn;
1176
1177 device_info = netvsc_devinfo_get(nvdev);
1178
1179 if (!device_info)
1180 return -ENOMEM;
1181
1182 device_info->num_chn = count;
1183
1184 ret = netvsc_detach(net, nvdev);
1185 if (ret)
1186 goto out;
1187
1188 ret = netvsc_attach(net, device_info);
1189 if (ret) {
1190 device_info->num_chn = orig;
1191 if (netvsc_attach(net, device_info))
1192 netdev_err(net, "restoring channel setting failed\n");
1193 }
1194
1195 out:
1196 netvsc_devinfo_put(device_info);
1197 return ret;
1198 }
1199
netvsc_init_settings(struct net_device * dev)1200 static void netvsc_init_settings(struct net_device *dev)
1201 {
1202 struct net_device_context *ndc = netdev_priv(dev);
1203
1204 ndc->l4_hash = HV_DEFAULT_L4HASH;
1205
1206 ndc->speed = SPEED_UNKNOWN;
1207 ndc->duplex = DUPLEX_FULL;
1208
1209 dev->features = NETIF_F_LRO;
1210 }
1211
netvsc_get_link_ksettings(struct net_device * dev,struct ethtool_link_ksettings * cmd)1212 static int netvsc_get_link_ksettings(struct net_device *dev,
1213 struct ethtool_link_ksettings *cmd)
1214 {
1215 struct net_device_context *ndc = netdev_priv(dev);
1216 struct net_device *vf_netdev;
1217
1218 vf_netdev = rtnl_dereference(ndc->vf_netdev);
1219
1220 if (vf_netdev)
1221 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1222
1223 cmd->base.speed = ndc->speed;
1224 cmd->base.duplex = ndc->duplex;
1225 cmd->base.port = PORT_OTHER;
1226
1227 return 0;
1228 }
1229
netvsc_set_link_ksettings(struct net_device * dev,const struct ethtool_link_ksettings * cmd)1230 static int netvsc_set_link_ksettings(struct net_device *dev,
1231 const struct ethtool_link_ksettings *cmd)
1232 {
1233 struct net_device_context *ndc = netdev_priv(dev);
1234 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1235
1236 if (vf_netdev) {
1237 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1238 return -EOPNOTSUPP;
1239
1240 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1241 cmd);
1242 }
1243
1244 return ethtool_virtdev_set_link_ksettings(dev, cmd,
1245 &ndc->speed, &ndc->duplex);
1246 }
1247
netvsc_change_mtu(struct net_device * ndev,int mtu)1248 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1249 {
1250 struct net_device_context *ndevctx = netdev_priv(ndev);
1251 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1252 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1253 int orig_mtu = ndev->mtu;
1254 struct netvsc_device_info *device_info;
1255 int ret = 0;
1256
1257 if (!nvdev || nvdev->destroy)
1258 return -ENODEV;
1259
1260 device_info = netvsc_devinfo_get(nvdev);
1261
1262 if (!device_info)
1263 return -ENOMEM;
1264
1265 /* Change MTU of underlying VF netdev first. */
1266 if (vf_netdev) {
1267 ret = dev_set_mtu(vf_netdev, mtu);
1268 if (ret)
1269 goto out;
1270 }
1271
1272 ret = netvsc_detach(ndev, nvdev);
1273 if (ret)
1274 goto rollback_vf;
1275
1276 ndev->mtu = mtu;
1277
1278 ret = netvsc_attach(ndev, device_info);
1279 if (!ret)
1280 goto out;
1281
1282 /* Attempt rollback to original MTU */
1283 ndev->mtu = orig_mtu;
1284
1285 if (netvsc_attach(ndev, device_info))
1286 netdev_err(ndev, "restoring mtu failed\n");
1287 rollback_vf:
1288 if (vf_netdev)
1289 dev_set_mtu(vf_netdev, orig_mtu);
1290
1291 out:
1292 netvsc_devinfo_put(device_info);
1293 return ret;
1294 }
1295
netvsc_get_vf_stats(struct net_device * net,struct netvsc_vf_pcpu_stats * tot)1296 static void netvsc_get_vf_stats(struct net_device *net,
1297 struct netvsc_vf_pcpu_stats *tot)
1298 {
1299 struct net_device_context *ndev_ctx = netdev_priv(net);
1300 int i;
1301
1302 memset(tot, 0, sizeof(*tot));
1303
1304 for_each_possible_cpu(i) {
1305 const struct netvsc_vf_pcpu_stats *stats
1306 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1307 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1308 unsigned int start;
1309
1310 do {
1311 start = u64_stats_fetch_begin_irq(&stats->syncp);
1312 rx_packets = stats->rx_packets;
1313 tx_packets = stats->tx_packets;
1314 rx_bytes = stats->rx_bytes;
1315 tx_bytes = stats->tx_bytes;
1316 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1317
1318 tot->rx_packets += rx_packets;
1319 tot->tx_packets += tx_packets;
1320 tot->rx_bytes += rx_bytes;
1321 tot->tx_bytes += tx_bytes;
1322 tot->tx_dropped += stats->tx_dropped;
1323 }
1324 }
1325
netvsc_get_pcpu_stats(struct net_device * net,struct netvsc_ethtool_pcpu_stats * pcpu_tot)1326 static void netvsc_get_pcpu_stats(struct net_device *net,
1327 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1328 {
1329 struct net_device_context *ndev_ctx = netdev_priv(net);
1330 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1331 int i;
1332
1333 /* fetch percpu stats of vf */
1334 for_each_possible_cpu(i) {
1335 const struct netvsc_vf_pcpu_stats *stats =
1336 per_cpu_ptr(ndev_ctx->vf_stats, i);
1337 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1338 unsigned int start;
1339
1340 do {
1341 start = u64_stats_fetch_begin_irq(&stats->syncp);
1342 this_tot->vf_rx_packets = stats->rx_packets;
1343 this_tot->vf_tx_packets = stats->tx_packets;
1344 this_tot->vf_rx_bytes = stats->rx_bytes;
1345 this_tot->vf_tx_bytes = stats->tx_bytes;
1346 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1347 this_tot->rx_packets = this_tot->vf_rx_packets;
1348 this_tot->tx_packets = this_tot->vf_tx_packets;
1349 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1350 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1351 }
1352
1353 /* fetch percpu stats of netvsc */
1354 for (i = 0; i < nvdev->num_chn; i++) {
1355 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1356 const struct netvsc_stats *stats;
1357 struct netvsc_ethtool_pcpu_stats *this_tot =
1358 &pcpu_tot[nvchan->channel->target_cpu];
1359 u64 packets, bytes;
1360 unsigned int start;
1361
1362 stats = &nvchan->tx_stats;
1363 do {
1364 start = u64_stats_fetch_begin_irq(&stats->syncp);
1365 packets = stats->packets;
1366 bytes = stats->bytes;
1367 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1368
1369 this_tot->tx_bytes += bytes;
1370 this_tot->tx_packets += packets;
1371
1372 stats = &nvchan->rx_stats;
1373 do {
1374 start = u64_stats_fetch_begin_irq(&stats->syncp);
1375 packets = stats->packets;
1376 bytes = stats->bytes;
1377 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1378
1379 this_tot->rx_bytes += bytes;
1380 this_tot->rx_packets += packets;
1381 }
1382 }
1383
netvsc_get_stats64(struct net_device * net,struct rtnl_link_stats64 * t)1384 static void netvsc_get_stats64(struct net_device *net,
1385 struct rtnl_link_stats64 *t)
1386 {
1387 struct net_device_context *ndev_ctx = netdev_priv(net);
1388 struct netvsc_device *nvdev;
1389 struct netvsc_vf_pcpu_stats vf_tot;
1390 int i;
1391
1392 rcu_read_lock();
1393
1394 nvdev = rcu_dereference(ndev_ctx->nvdev);
1395 if (!nvdev)
1396 goto out;
1397
1398 netdev_stats_to_stats64(t, &net->stats);
1399
1400 netvsc_get_vf_stats(net, &vf_tot);
1401 t->rx_packets += vf_tot.rx_packets;
1402 t->tx_packets += vf_tot.tx_packets;
1403 t->rx_bytes += vf_tot.rx_bytes;
1404 t->tx_bytes += vf_tot.tx_bytes;
1405 t->tx_dropped += vf_tot.tx_dropped;
1406
1407 for (i = 0; i < nvdev->num_chn; i++) {
1408 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1409 const struct netvsc_stats *stats;
1410 u64 packets, bytes, multicast;
1411 unsigned int start;
1412
1413 stats = &nvchan->tx_stats;
1414 do {
1415 start = u64_stats_fetch_begin_irq(&stats->syncp);
1416 packets = stats->packets;
1417 bytes = stats->bytes;
1418 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1419
1420 t->tx_bytes += bytes;
1421 t->tx_packets += packets;
1422
1423 stats = &nvchan->rx_stats;
1424 do {
1425 start = u64_stats_fetch_begin_irq(&stats->syncp);
1426 packets = stats->packets;
1427 bytes = stats->bytes;
1428 multicast = stats->multicast + stats->broadcast;
1429 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1430
1431 t->rx_bytes += bytes;
1432 t->rx_packets += packets;
1433 t->multicast += multicast;
1434 }
1435 out:
1436 rcu_read_unlock();
1437 }
1438
netvsc_set_mac_addr(struct net_device * ndev,void * p)1439 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1440 {
1441 struct net_device_context *ndc = netdev_priv(ndev);
1442 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1443 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1444 struct sockaddr *addr = p;
1445 int err;
1446
1447 err = eth_prepare_mac_addr_change(ndev, p);
1448 if (err)
1449 return err;
1450
1451 if (!nvdev)
1452 return -ENODEV;
1453
1454 if (vf_netdev) {
1455 err = dev_set_mac_address(vf_netdev, addr, NULL);
1456 if (err)
1457 return err;
1458 }
1459
1460 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1461 if (!err) {
1462 eth_commit_mac_addr_change(ndev, p);
1463 } else if (vf_netdev) {
1464 /* rollback change on VF */
1465 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1466 dev_set_mac_address(vf_netdev, addr, NULL);
1467 }
1468
1469 return err;
1470 }
1471
1472 static const struct {
1473 char name[ETH_GSTRING_LEN];
1474 u16 offset;
1475 } netvsc_stats[] = {
1476 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1477 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1478 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1479 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1480 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1481 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1482 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1483 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1484 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1485 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1486 { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1487 }, pcpu_stats[] = {
1488 { "cpu%u_rx_packets",
1489 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1490 { "cpu%u_rx_bytes",
1491 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1492 { "cpu%u_tx_packets",
1493 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1494 { "cpu%u_tx_bytes",
1495 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1496 { "cpu%u_vf_rx_packets",
1497 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1498 { "cpu%u_vf_rx_bytes",
1499 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1500 { "cpu%u_vf_tx_packets",
1501 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1502 { "cpu%u_vf_tx_bytes",
1503 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1504 }, vf_stats[] = {
1505 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1506 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1507 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1508 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1509 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1510 };
1511
1512 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1513 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1514
1515 /* statistics per queue (rx/tx packets/bytes) */
1516 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1517
1518 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1519 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1520
netvsc_get_sset_count(struct net_device * dev,int string_set)1521 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1522 {
1523 struct net_device_context *ndc = netdev_priv(dev);
1524 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1525
1526 if (!nvdev)
1527 return -ENODEV;
1528
1529 switch (string_set) {
1530 case ETH_SS_STATS:
1531 return NETVSC_GLOBAL_STATS_LEN
1532 + NETVSC_VF_STATS_LEN
1533 + NETVSC_QUEUE_STATS_LEN(nvdev)
1534 + NETVSC_PCPU_STATS_LEN;
1535 default:
1536 return -EINVAL;
1537 }
1538 }
1539
netvsc_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * stats,u64 * data)1540 static void netvsc_get_ethtool_stats(struct net_device *dev,
1541 struct ethtool_stats *stats, u64 *data)
1542 {
1543 struct net_device_context *ndc = netdev_priv(dev);
1544 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1545 const void *nds = &ndc->eth_stats;
1546 const struct netvsc_stats *qstats;
1547 struct netvsc_vf_pcpu_stats sum;
1548 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1549 unsigned int start;
1550 u64 packets, bytes;
1551 u64 xdp_drop;
1552 int i, j, cpu;
1553
1554 if (!nvdev)
1555 return;
1556
1557 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1558 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1559
1560 netvsc_get_vf_stats(dev, &sum);
1561 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1562 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1563
1564 for (j = 0; j < nvdev->num_chn; j++) {
1565 qstats = &nvdev->chan_table[j].tx_stats;
1566
1567 do {
1568 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1569 packets = qstats->packets;
1570 bytes = qstats->bytes;
1571 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1572 data[i++] = packets;
1573 data[i++] = bytes;
1574
1575 qstats = &nvdev->chan_table[j].rx_stats;
1576 do {
1577 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1578 packets = qstats->packets;
1579 bytes = qstats->bytes;
1580 xdp_drop = qstats->xdp_drop;
1581 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1582 data[i++] = packets;
1583 data[i++] = bytes;
1584 data[i++] = xdp_drop;
1585 }
1586
1587 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1588 sizeof(struct netvsc_ethtool_pcpu_stats),
1589 GFP_KERNEL);
1590 netvsc_get_pcpu_stats(dev, pcpu_sum);
1591 for_each_present_cpu(cpu) {
1592 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1593
1594 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1595 data[i++] = *(u64 *)((void *)this_sum
1596 + pcpu_stats[j].offset);
1597 }
1598 kvfree(pcpu_sum);
1599 }
1600
netvsc_get_strings(struct net_device * dev,u32 stringset,u8 * data)1601 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1602 {
1603 struct net_device_context *ndc = netdev_priv(dev);
1604 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1605 u8 *p = data;
1606 int i, cpu;
1607
1608 if (!nvdev)
1609 return;
1610
1611 switch (stringset) {
1612 case ETH_SS_STATS:
1613 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1614 ethtool_sprintf(&p, netvsc_stats[i].name);
1615
1616 for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1617 ethtool_sprintf(&p, vf_stats[i].name);
1618
1619 for (i = 0; i < nvdev->num_chn; i++) {
1620 ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1621 ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1622 ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1623 ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1624 ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1625 }
1626
1627 for_each_present_cpu(cpu) {
1628 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1629 ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1630 }
1631
1632 break;
1633 }
1634 }
1635
1636 static int
netvsc_get_rss_hash_opts(struct net_device_context * ndc,struct ethtool_rxnfc * info)1637 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1638 struct ethtool_rxnfc *info)
1639 {
1640 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1641
1642 info->data = RXH_IP_SRC | RXH_IP_DST;
1643
1644 switch (info->flow_type) {
1645 case TCP_V4_FLOW:
1646 if (ndc->l4_hash & HV_TCP4_L4HASH)
1647 info->data |= l4_flag;
1648
1649 break;
1650
1651 case TCP_V6_FLOW:
1652 if (ndc->l4_hash & HV_TCP6_L4HASH)
1653 info->data |= l4_flag;
1654
1655 break;
1656
1657 case UDP_V4_FLOW:
1658 if (ndc->l4_hash & HV_UDP4_L4HASH)
1659 info->data |= l4_flag;
1660
1661 break;
1662
1663 case UDP_V6_FLOW:
1664 if (ndc->l4_hash & HV_UDP6_L4HASH)
1665 info->data |= l4_flag;
1666
1667 break;
1668
1669 case IPV4_FLOW:
1670 case IPV6_FLOW:
1671 break;
1672 default:
1673 info->data = 0;
1674 break;
1675 }
1676
1677 return 0;
1678 }
1679
1680 static int
netvsc_get_rxnfc(struct net_device * dev,struct ethtool_rxnfc * info,u32 * rules)1681 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1682 u32 *rules)
1683 {
1684 struct net_device_context *ndc = netdev_priv(dev);
1685 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1686
1687 if (!nvdev)
1688 return -ENODEV;
1689
1690 switch (info->cmd) {
1691 case ETHTOOL_GRXRINGS:
1692 info->data = nvdev->num_chn;
1693 return 0;
1694
1695 case ETHTOOL_GRXFH:
1696 return netvsc_get_rss_hash_opts(ndc, info);
1697 }
1698 return -EOPNOTSUPP;
1699 }
1700
netvsc_set_rss_hash_opts(struct net_device_context * ndc,struct ethtool_rxnfc * info)1701 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1702 struct ethtool_rxnfc *info)
1703 {
1704 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1705 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1706 switch (info->flow_type) {
1707 case TCP_V4_FLOW:
1708 ndc->l4_hash |= HV_TCP4_L4HASH;
1709 break;
1710
1711 case TCP_V6_FLOW:
1712 ndc->l4_hash |= HV_TCP6_L4HASH;
1713 break;
1714
1715 case UDP_V4_FLOW:
1716 ndc->l4_hash |= HV_UDP4_L4HASH;
1717 break;
1718
1719 case UDP_V6_FLOW:
1720 ndc->l4_hash |= HV_UDP6_L4HASH;
1721 break;
1722
1723 default:
1724 return -EOPNOTSUPP;
1725 }
1726
1727 return 0;
1728 }
1729
1730 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1731 switch (info->flow_type) {
1732 case TCP_V4_FLOW:
1733 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1734 break;
1735
1736 case TCP_V6_FLOW:
1737 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1738 break;
1739
1740 case UDP_V4_FLOW:
1741 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1742 break;
1743
1744 case UDP_V6_FLOW:
1745 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1746 break;
1747
1748 default:
1749 return -EOPNOTSUPP;
1750 }
1751
1752 return 0;
1753 }
1754
1755 return -EOPNOTSUPP;
1756 }
1757
1758 static int
netvsc_set_rxnfc(struct net_device * ndev,struct ethtool_rxnfc * info)1759 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1760 {
1761 struct net_device_context *ndc = netdev_priv(ndev);
1762
1763 if (info->cmd == ETHTOOL_SRXFH)
1764 return netvsc_set_rss_hash_opts(ndc, info);
1765
1766 return -EOPNOTSUPP;
1767 }
1768
netvsc_get_rxfh_key_size(struct net_device * dev)1769 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1770 {
1771 return NETVSC_HASH_KEYLEN;
1772 }
1773
netvsc_rss_indir_size(struct net_device * dev)1774 static u32 netvsc_rss_indir_size(struct net_device *dev)
1775 {
1776 return ITAB_NUM;
1777 }
1778
netvsc_get_rxfh(struct net_device * dev,u32 * indir,u8 * key,u8 * hfunc)1779 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1780 u8 *hfunc)
1781 {
1782 struct net_device_context *ndc = netdev_priv(dev);
1783 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1784 struct rndis_device *rndis_dev;
1785 int i;
1786
1787 if (!ndev)
1788 return -ENODEV;
1789
1790 if (hfunc)
1791 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1792
1793 rndis_dev = ndev->extension;
1794 if (indir) {
1795 for (i = 0; i < ITAB_NUM; i++)
1796 indir[i] = ndc->rx_table[i];
1797 }
1798
1799 if (key)
1800 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1801
1802 return 0;
1803 }
1804
netvsc_set_rxfh(struct net_device * dev,const u32 * indir,const u8 * key,const u8 hfunc)1805 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1806 const u8 *key, const u8 hfunc)
1807 {
1808 struct net_device_context *ndc = netdev_priv(dev);
1809 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1810 struct rndis_device *rndis_dev;
1811 int i;
1812
1813 if (!ndev)
1814 return -ENODEV;
1815
1816 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1817 return -EOPNOTSUPP;
1818
1819 rndis_dev = ndev->extension;
1820 if (indir) {
1821 for (i = 0; i < ITAB_NUM; i++)
1822 if (indir[i] >= ndev->num_chn)
1823 return -EINVAL;
1824
1825 for (i = 0; i < ITAB_NUM; i++)
1826 ndc->rx_table[i] = indir[i];
1827 }
1828
1829 if (!key) {
1830 if (!indir)
1831 return 0;
1832
1833 key = rndis_dev->rss_key;
1834 }
1835
1836 return rndis_filter_set_rss_param(rndis_dev, key);
1837 }
1838
1839 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1840 * It does have pre-allocated receive area which is divided into sections.
1841 */
__netvsc_get_ringparam(struct netvsc_device * nvdev,struct ethtool_ringparam * ring)1842 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1843 struct ethtool_ringparam *ring)
1844 {
1845 u32 max_buf_size;
1846
1847 ring->rx_pending = nvdev->recv_section_cnt;
1848 ring->tx_pending = nvdev->send_section_cnt;
1849
1850 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1851 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1852 else
1853 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1854
1855 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1856 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1857 / nvdev->send_section_size;
1858 }
1859
netvsc_get_ringparam(struct net_device * ndev,struct ethtool_ringparam * ring)1860 static void netvsc_get_ringparam(struct net_device *ndev,
1861 struct ethtool_ringparam *ring)
1862 {
1863 struct net_device_context *ndevctx = netdev_priv(ndev);
1864 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1865
1866 if (!nvdev)
1867 return;
1868
1869 __netvsc_get_ringparam(nvdev, ring);
1870 }
1871
netvsc_set_ringparam(struct net_device * ndev,struct ethtool_ringparam * ring)1872 static int netvsc_set_ringparam(struct net_device *ndev,
1873 struct ethtool_ringparam *ring)
1874 {
1875 struct net_device_context *ndevctx = netdev_priv(ndev);
1876 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1877 struct netvsc_device_info *device_info;
1878 struct ethtool_ringparam orig;
1879 u32 new_tx, new_rx;
1880 int ret = 0;
1881
1882 if (!nvdev || nvdev->destroy)
1883 return -ENODEV;
1884
1885 memset(&orig, 0, sizeof(orig));
1886 __netvsc_get_ringparam(nvdev, &orig);
1887
1888 new_tx = clamp_t(u32, ring->tx_pending,
1889 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1890 new_rx = clamp_t(u32, ring->rx_pending,
1891 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1892
1893 if (new_tx == orig.tx_pending &&
1894 new_rx == orig.rx_pending)
1895 return 0; /* no change */
1896
1897 device_info = netvsc_devinfo_get(nvdev);
1898
1899 if (!device_info)
1900 return -ENOMEM;
1901
1902 device_info->send_sections = new_tx;
1903 device_info->recv_sections = new_rx;
1904
1905 ret = netvsc_detach(ndev, nvdev);
1906 if (ret)
1907 goto out;
1908
1909 ret = netvsc_attach(ndev, device_info);
1910 if (ret) {
1911 device_info->send_sections = orig.tx_pending;
1912 device_info->recv_sections = orig.rx_pending;
1913
1914 if (netvsc_attach(ndev, device_info))
1915 netdev_err(ndev, "restoring ringparam failed");
1916 }
1917
1918 out:
1919 netvsc_devinfo_put(device_info);
1920 return ret;
1921 }
1922
netvsc_fix_features(struct net_device * ndev,netdev_features_t features)1923 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1924 netdev_features_t features)
1925 {
1926 struct net_device_context *ndevctx = netdev_priv(ndev);
1927 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1928
1929 if (!nvdev || nvdev->destroy)
1930 return features;
1931
1932 if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1933 features ^= NETIF_F_LRO;
1934 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1935 }
1936
1937 return features;
1938 }
1939
netvsc_set_features(struct net_device * ndev,netdev_features_t features)1940 static int netvsc_set_features(struct net_device *ndev,
1941 netdev_features_t features)
1942 {
1943 netdev_features_t change = features ^ ndev->features;
1944 struct net_device_context *ndevctx = netdev_priv(ndev);
1945 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1946 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1947 struct ndis_offload_params offloads;
1948 int ret = 0;
1949
1950 if (!nvdev || nvdev->destroy)
1951 return -ENODEV;
1952
1953 if (!(change & NETIF_F_LRO))
1954 goto syncvf;
1955
1956 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1957
1958 if (features & NETIF_F_LRO) {
1959 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1960 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1961 } else {
1962 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1963 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1964 }
1965
1966 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1967
1968 if (ret) {
1969 features ^= NETIF_F_LRO;
1970 ndev->features = features;
1971 }
1972
1973 syncvf:
1974 if (!vf_netdev)
1975 return ret;
1976
1977 vf_netdev->wanted_features = features;
1978 netdev_update_features(vf_netdev);
1979
1980 return ret;
1981 }
1982
netvsc_get_regs_len(struct net_device * netdev)1983 static int netvsc_get_regs_len(struct net_device *netdev)
1984 {
1985 return VRSS_SEND_TAB_SIZE * sizeof(u32);
1986 }
1987
netvsc_get_regs(struct net_device * netdev,struct ethtool_regs * regs,void * p)1988 static void netvsc_get_regs(struct net_device *netdev,
1989 struct ethtool_regs *regs, void *p)
1990 {
1991 struct net_device_context *ndc = netdev_priv(netdev);
1992 u32 *regs_buff = p;
1993
1994 /* increase the version, if buffer format is changed. */
1995 regs->version = 1;
1996
1997 memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1998 }
1999
netvsc_get_msglevel(struct net_device * ndev)2000 static u32 netvsc_get_msglevel(struct net_device *ndev)
2001 {
2002 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2003
2004 return ndev_ctx->msg_enable;
2005 }
2006
netvsc_set_msglevel(struct net_device * ndev,u32 val)2007 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2008 {
2009 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2010
2011 ndev_ctx->msg_enable = val;
2012 }
2013
2014 static const struct ethtool_ops ethtool_ops = {
2015 .get_drvinfo = netvsc_get_drvinfo,
2016 .get_regs_len = netvsc_get_regs_len,
2017 .get_regs = netvsc_get_regs,
2018 .get_msglevel = netvsc_get_msglevel,
2019 .set_msglevel = netvsc_set_msglevel,
2020 .get_link = ethtool_op_get_link,
2021 .get_ethtool_stats = netvsc_get_ethtool_stats,
2022 .get_sset_count = netvsc_get_sset_count,
2023 .get_strings = netvsc_get_strings,
2024 .get_channels = netvsc_get_channels,
2025 .set_channels = netvsc_set_channels,
2026 .get_ts_info = ethtool_op_get_ts_info,
2027 .get_rxnfc = netvsc_get_rxnfc,
2028 .set_rxnfc = netvsc_set_rxnfc,
2029 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2030 .get_rxfh_indir_size = netvsc_rss_indir_size,
2031 .get_rxfh = netvsc_get_rxfh,
2032 .set_rxfh = netvsc_set_rxfh,
2033 .get_link_ksettings = netvsc_get_link_ksettings,
2034 .set_link_ksettings = netvsc_set_link_ksettings,
2035 .get_ringparam = netvsc_get_ringparam,
2036 .set_ringparam = netvsc_set_ringparam,
2037 };
2038
2039 static const struct net_device_ops device_ops = {
2040 .ndo_open = netvsc_open,
2041 .ndo_stop = netvsc_close,
2042 .ndo_start_xmit = netvsc_start_xmit,
2043 .ndo_change_rx_flags = netvsc_change_rx_flags,
2044 .ndo_set_rx_mode = netvsc_set_rx_mode,
2045 .ndo_fix_features = netvsc_fix_features,
2046 .ndo_set_features = netvsc_set_features,
2047 .ndo_change_mtu = netvsc_change_mtu,
2048 .ndo_validate_addr = eth_validate_addr,
2049 .ndo_set_mac_address = netvsc_set_mac_addr,
2050 .ndo_select_queue = netvsc_select_queue,
2051 .ndo_get_stats64 = netvsc_get_stats64,
2052 .ndo_bpf = netvsc_bpf,
2053 };
2054
2055 /*
2056 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2057 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2058 * present send GARP packet to network peers with netif_notify_peers().
2059 */
netvsc_link_change(struct work_struct * w)2060 static void netvsc_link_change(struct work_struct *w)
2061 {
2062 struct net_device_context *ndev_ctx =
2063 container_of(w, struct net_device_context, dwork.work);
2064 struct hv_device *device_obj = ndev_ctx->device_ctx;
2065 struct net_device *net = hv_get_drvdata(device_obj);
2066 unsigned long flags, next_reconfig, delay;
2067 struct netvsc_reconfig *event = NULL;
2068 struct netvsc_device *net_device;
2069 struct rndis_device *rdev;
2070 bool reschedule = false;
2071
2072 /* if changes are happening, comeback later */
2073 if (!rtnl_trylock()) {
2074 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2075 return;
2076 }
2077
2078 net_device = rtnl_dereference(ndev_ctx->nvdev);
2079 if (!net_device)
2080 goto out_unlock;
2081
2082 rdev = net_device->extension;
2083
2084 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2085 if (time_is_after_jiffies(next_reconfig)) {
2086 /* link_watch only sends one notification with current state
2087 * per second, avoid doing reconfig more frequently. Handle
2088 * wrap around.
2089 */
2090 delay = next_reconfig - jiffies;
2091 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2092 schedule_delayed_work(&ndev_ctx->dwork, delay);
2093 goto out_unlock;
2094 }
2095 ndev_ctx->last_reconfig = jiffies;
2096
2097 spin_lock_irqsave(&ndev_ctx->lock, flags);
2098 if (!list_empty(&ndev_ctx->reconfig_events)) {
2099 event = list_first_entry(&ndev_ctx->reconfig_events,
2100 struct netvsc_reconfig, list);
2101 list_del(&event->list);
2102 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2103 }
2104 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2105
2106 if (!event)
2107 goto out_unlock;
2108
2109 switch (event->event) {
2110 /* Only the following events are possible due to the check in
2111 * netvsc_linkstatus_callback()
2112 */
2113 case RNDIS_STATUS_MEDIA_CONNECT:
2114 if (rdev->link_state) {
2115 rdev->link_state = false;
2116 netif_carrier_on(net);
2117 netvsc_tx_enable(net_device, net);
2118 } else {
2119 __netdev_notify_peers(net);
2120 }
2121 kfree(event);
2122 break;
2123 case RNDIS_STATUS_MEDIA_DISCONNECT:
2124 if (!rdev->link_state) {
2125 rdev->link_state = true;
2126 netif_carrier_off(net);
2127 netvsc_tx_disable(net_device, net);
2128 }
2129 kfree(event);
2130 break;
2131 case RNDIS_STATUS_NETWORK_CHANGE:
2132 /* Only makes sense if carrier is present */
2133 if (!rdev->link_state) {
2134 rdev->link_state = true;
2135 netif_carrier_off(net);
2136 netvsc_tx_disable(net_device, net);
2137 event->event = RNDIS_STATUS_MEDIA_CONNECT;
2138 spin_lock_irqsave(&ndev_ctx->lock, flags);
2139 list_add(&event->list, &ndev_ctx->reconfig_events);
2140 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2141 reschedule = true;
2142 }
2143 break;
2144 }
2145
2146 rtnl_unlock();
2147
2148 /* link_watch only sends one notification with current state per
2149 * second, handle next reconfig event in 2 seconds.
2150 */
2151 if (reschedule)
2152 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2153
2154 return;
2155
2156 out_unlock:
2157 rtnl_unlock();
2158 }
2159
get_netvsc_byref(struct net_device * vf_netdev)2160 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2161 {
2162 struct net_device_context *net_device_ctx;
2163 struct net_device *dev;
2164
2165 dev = netdev_master_upper_dev_get(vf_netdev);
2166 if (!dev || dev->netdev_ops != &device_ops)
2167 return NULL; /* not a netvsc device */
2168
2169 net_device_ctx = netdev_priv(dev);
2170 if (!rtnl_dereference(net_device_ctx->nvdev))
2171 return NULL; /* device is removed */
2172
2173 return dev;
2174 }
2175
2176 /* Called when VF is injecting data into network stack.
2177 * Change the associated network device from VF to netvsc.
2178 * note: already called with rcu_read_lock
2179 */
netvsc_vf_handle_frame(struct sk_buff ** pskb)2180 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2181 {
2182 struct sk_buff *skb = *pskb;
2183 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2184 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2185 struct netvsc_vf_pcpu_stats *pcpu_stats
2186 = this_cpu_ptr(ndev_ctx->vf_stats);
2187
2188 skb = skb_share_check(skb, GFP_ATOMIC);
2189 if (unlikely(!skb))
2190 return RX_HANDLER_CONSUMED;
2191
2192 *pskb = skb;
2193
2194 skb->dev = ndev;
2195
2196 u64_stats_update_begin(&pcpu_stats->syncp);
2197 pcpu_stats->rx_packets++;
2198 pcpu_stats->rx_bytes += skb->len;
2199 u64_stats_update_end(&pcpu_stats->syncp);
2200
2201 return RX_HANDLER_ANOTHER;
2202 }
2203
netvsc_vf_join(struct net_device * vf_netdev,struct net_device * ndev)2204 static int netvsc_vf_join(struct net_device *vf_netdev,
2205 struct net_device *ndev)
2206 {
2207 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2208 int ret;
2209
2210 ret = netdev_rx_handler_register(vf_netdev,
2211 netvsc_vf_handle_frame, ndev);
2212 if (ret != 0) {
2213 netdev_err(vf_netdev,
2214 "can not register netvsc VF receive handler (err = %d)\n",
2215 ret);
2216 goto rx_handler_failed;
2217 }
2218
2219 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2220 NULL, NULL, NULL);
2221 if (ret != 0) {
2222 netdev_err(vf_netdev,
2223 "can not set master device %s (err = %d)\n",
2224 ndev->name, ret);
2225 goto upper_link_failed;
2226 }
2227
2228 /* set slave flag before open to prevent IPv6 addrconf */
2229 vf_netdev->flags |= IFF_SLAVE;
2230
2231 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2232
2233 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2234
2235 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2236 return 0;
2237
2238 upper_link_failed:
2239 netdev_rx_handler_unregister(vf_netdev);
2240 rx_handler_failed:
2241 return ret;
2242 }
2243
__netvsc_vf_setup(struct net_device * ndev,struct net_device * vf_netdev)2244 static void __netvsc_vf_setup(struct net_device *ndev,
2245 struct net_device *vf_netdev)
2246 {
2247 int ret;
2248
2249 /* Align MTU of VF with master */
2250 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2251 if (ret)
2252 netdev_warn(vf_netdev,
2253 "unable to change mtu to %u\n", ndev->mtu);
2254
2255 /* set multicast etc flags on VF */
2256 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2257
2258 /* sync address list from ndev to VF */
2259 netif_addr_lock_bh(ndev);
2260 dev_uc_sync(vf_netdev, ndev);
2261 dev_mc_sync(vf_netdev, ndev);
2262 netif_addr_unlock_bh(ndev);
2263
2264 if (netif_running(ndev)) {
2265 ret = dev_open(vf_netdev, NULL);
2266 if (ret)
2267 netdev_warn(vf_netdev,
2268 "unable to open: %d\n", ret);
2269 }
2270 }
2271
2272 /* Setup VF as slave of the synthetic device.
2273 * Runs in workqueue to avoid recursion in netlink callbacks.
2274 */
netvsc_vf_setup(struct work_struct * w)2275 static void netvsc_vf_setup(struct work_struct *w)
2276 {
2277 struct net_device_context *ndev_ctx
2278 = container_of(w, struct net_device_context, vf_takeover.work);
2279 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2280 struct net_device *vf_netdev;
2281
2282 if (!rtnl_trylock()) {
2283 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2284 return;
2285 }
2286
2287 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2288 if (vf_netdev)
2289 __netvsc_vf_setup(ndev, vf_netdev);
2290
2291 rtnl_unlock();
2292 }
2293
2294 /* Find netvsc by VF serial number.
2295 * The PCI hyperv controller records the serial number as the slot kobj name.
2296 */
get_netvsc_byslot(const struct net_device * vf_netdev)2297 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2298 {
2299 struct device *parent = vf_netdev->dev.parent;
2300 struct net_device_context *ndev_ctx;
2301 struct net_device *ndev;
2302 struct pci_dev *pdev;
2303 u32 serial;
2304
2305 if (!parent || !dev_is_pci(parent))
2306 return NULL; /* not a PCI device */
2307
2308 pdev = to_pci_dev(parent);
2309 if (!pdev->slot) {
2310 netdev_notice(vf_netdev, "no PCI slot information\n");
2311 return NULL;
2312 }
2313
2314 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2315 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2316 pci_slot_name(pdev->slot));
2317 return NULL;
2318 }
2319
2320 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2321 if (!ndev_ctx->vf_alloc)
2322 continue;
2323
2324 if (ndev_ctx->vf_serial != serial)
2325 continue;
2326
2327 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2328 if (ndev->addr_len != vf_netdev->addr_len ||
2329 memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2330 ndev->addr_len) != 0)
2331 continue;
2332
2333 return ndev;
2334
2335 }
2336
2337 netdev_notice(vf_netdev,
2338 "no netdev found for vf serial:%u\n", serial);
2339 return NULL;
2340 }
2341
netvsc_register_vf(struct net_device * vf_netdev)2342 static int netvsc_register_vf(struct net_device *vf_netdev)
2343 {
2344 struct net_device_context *net_device_ctx;
2345 struct netvsc_device *netvsc_dev;
2346 struct bpf_prog *prog;
2347 struct net_device *ndev;
2348 int ret;
2349
2350 if (vf_netdev->addr_len != ETH_ALEN)
2351 return NOTIFY_DONE;
2352
2353 ndev = get_netvsc_byslot(vf_netdev);
2354 if (!ndev)
2355 return NOTIFY_DONE;
2356
2357 net_device_ctx = netdev_priv(ndev);
2358 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2359 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2360 return NOTIFY_DONE;
2361
2362 /* if synthetic interface is a different namespace,
2363 * then move the VF to that namespace; join will be
2364 * done again in that context.
2365 */
2366 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2367 ret = dev_change_net_namespace(vf_netdev,
2368 dev_net(ndev), "eth%d");
2369 if (ret)
2370 netdev_err(vf_netdev,
2371 "could not move to same namespace as %s: %d\n",
2372 ndev->name, ret);
2373 else
2374 netdev_info(vf_netdev,
2375 "VF moved to namespace with: %s\n",
2376 ndev->name);
2377 return NOTIFY_DONE;
2378 }
2379
2380 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2381
2382 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2383 return NOTIFY_DONE;
2384
2385 dev_hold(vf_netdev);
2386 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2387
2388 if (ndev->needed_headroom < vf_netdev->needed_headroom)
2389 ndev->needed_headroom = vf_netdev->needed_headroom;
2390
2391 vf_netdev->wanted_features = ndev->features;
2392 netdev_update_features(vf_netdev);
2393
2394 prog = netvsc_xdp_get(netvsc_dev);
2395 netvsc_vf_setxdp(vf_netdev, prog);
2396
2397 return NOTIFY_OK;
2398 }
2399
2400 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2401 *
2402 * Typically a UP or DOWN event is followed by a CHANGE event, so
2403 * net_device_ctx->data_path_is_vf is used to cache the current data path
2404 * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2405 * message.
2406 *
2407 * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2408 * interface, there is only the CHANGE event and no UP or DOWN event.
2409 */
netvsc_vf_changed(struct net_device * vf_netdev,unsigned long event)2410 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2411 {
2412 struct net_device_context *net_device_ctx;
2413 struct netvsc_device *netvsc_dev;
2414 struct net_device *ndev;
2415 bool vf_is_up = false;
2416 int ret;
2417
2418 if (event != NETDEV_GOING_DOWN)
2419 vf_is_up = netif_running(vf_netdev);
2420
2421 ndev = get_netvsc_byref(vf_netdev);
2422 if (!ndev)
2423 return NOTIFY_DONE;
2424
2425 net_device_ctx = netdev_priv(ndev);
2426 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2427 if (!netvsc_dev)
2428 return NOTIFY_DONE;
2429
2430 if (net_device_ctx->data_path_is_vf == vf_is_up)
2431 return NOTIFY_OK;
2432
2433 ret = netvsc_switch_datapath(ndev, vf_is_up);
2434
2435 if (ret) {
2436 netdev_err(ndev,
2437 "Data path failed to switch %s VF: %s, err: %d\n",
2438 vf_is_up ? "to" : "from", vf_netdev->name, ret);
2439 return NOTIFY_DONE;
2440 } else {
2441 netdev_info(ndev, "Data path switched %s VF: %s\n",
2442 vf_is_up ? "to" : "from", vf_netdev->name);
2443 }
2444
2445 return NOTIFY_OK;
2446 }
2447
netvsc_unregister_vf(struct net_device * vf_netdev)2448 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2449 {
2450 struct net_device *ndev;
2451 struct net_device_context *net_device_ctx;
2452
2453 ndev = get_netvsc_byref(vf_netdev);
2454 if (!ndev)
2455 return NOTIFY_DONE;
2456
2457 net_device_ctx = netdev_priv(ndev);
2458 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2459
2460 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2461
2462 netvsc_vf_setxdp(vf_netdev, NULL);
2463
2464 netdev_rx_handler_unregister(vf_netdev);
2465 netdev_upper_dev_unlink(vf_netdev, ndev);
2466 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2467 dev_put(vf_netdev);
2468
2469 ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2470
2471 return NOTIFY_OK;
2472 }
2473
netvsc_probe(struct hv_device * dev,const struct hv_vmbus_device_id * dev_id)2474 static int netvsc_probe(struct hv_device *dev,
2475 const struct hv_vmbus_device_id *dev_id)
2476 {
2477 struct net_device *net = NULL;
2478 struct net_device_context *net_device_ctx;
2479 struct netvsc_device_info *device_info = NULL;
2480 struct netvsc_device *nvdev;
2481 int ret = -ENOMEM;
2482
2483 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2484 VRSS_CHANNEL_MAX);
2485 if (!net)
2486 goto no_net;
2487
2488 netif_carrier_off(net);
2489
2490 netvsc_init_settings(net);
2491
2492 net_device_ctx = netdev_priv(net);
2493 net_device_ctx->device_ctx = dev;
2494 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2495 if (netif_msg_probe(net_device_ctx))
2496 netdev_dbg(net, "netvsc msg_enable: %d\n",
2497 net_device_ctx->msg_enable);
2498
2499 hv_set_drvdata(dev, net);
2500
2501 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2502
2503 spin_lock_init(&net_device_ctx->lock);
2504 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2505 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2506
2507 net_device_ctx->vf_stats
2508 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2509 if (!net_device_ctx->vf_stats)
2510 goto no_stats;
2511
2512 net->netdev_ops = &device_ops;
2513 net->ethtool_ops = ðtool_ops;
2514 SET_NETDEV_DEV(net, &dev->device);
2515
2516 /* We always need headroom for rndis header */
2517 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2518
2519 /* Initialize the number of queues to be 1, we may change it if more
2520 * channels are offered later.
2521 */
2522 netif_set_real_num_tx_queues(net, 1);
2523 netif_set_real_num_rx_queues(net, 1);
2524
2525 /* Notify the netvsc driver of the new device */
2526 device_info = netvsc_devinfo_get(NULL);
2527
2528 if (!device_info) {
2529 ret = -ENOMEM;
2530 goto devinfo_failed;
2531 }
2532
2533 nvdev = rndis_filter_device_add(dev, device_info);
2534 if (IS_ERR(nvdev)) {
2535 ret = PTR_ERR(nvdev);
2536 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2537 goto rndis_failed;
2538 }
2539
2540 eth_hw_addr_set(net, device_info->mac_adr);
2541
2542 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2543 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2544 * all subchannels to show up, but that may not happen because
2545 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2546 * -> ... -> device_add() -> ... -> __device_attach() can't get
2547 * the device lock, so all the subchannels can't be processed --
2548 * finally netvsc_subchan_work() hangs forever.
2549 */
2550 rtnl_lock();
2551
2552 if (nvdev->num_chn > 1)
2553 schedule_work(&nvdev->subchan_work);
2554
2555 /* hw_features computed in rndis_netdev_set_hwcaps() */
2556 net->features = net->hw_features |
2557 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2558 NETIF_F_HW_VLAN_CTAG_RX;
2559 net->vlan_features = net->features;
2560
2561 netdev_lockdep_set_classes(net);
2562
2563 /* MTU range: 68 - 1500 or 65521 */
2564 net->min_mtu = NETVSC_MTU_MIN;
2565 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2566 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2567 else
2568 net->max_mtu = ETH_DATA_LEN;
2569
2570 nvdev->tx_disable = false;
2571
2572 ret = register_netdevice(net);
2573 if (ret != 0) {
2574 pr_err("Unable to register netdev.\n");
2575 goto register_failed;
2576 }
2577
2578 list_add(&net_device_ctx->list, &netvsc_dev_list);
2579 rtnl_unlock();
2580
2581 netvsc_devinfo_put(device_info);
2582 return 0;
2583
2584 register_failed:
2585 rtnl_unlock();
2586 rndis_filter_device_remove(dev, nvdev);
2587 rndis_failed:
2588 netvsc_devinfo_put(device_info);
2589 devinfo_failed:
2590 free_percpu(net_device_ctx->vf_stats);
2591 no_stats:
2592 hv_set_drvdata(dev, NULL);
2593 free_netdev(net);
2594 no_net:
2595 return ret;
2596 }
2597
netvsc_remove(struct hv_device * dev)2598 static int netvsc_remove(struct hv_device *dev)
2599 {
2600 struct net_device_context *ndev_ctx;
2601 struct net_device *vf_netdev, *net;
2602 struct netvsc_device *nvdev;
2603
2604 net = hv_get_drvdata(dev);
2605 if (net == NULL) {
2606 dev_err(&dev->device, "No net device to remove\n");
2607 return 0;
2608 }
2609
2610 ndev_ctx = netdev_priv(net);
2611
2612 cancel_delayed_work_sync(&ndev_ctx->dwork);
2613
2614 rtnl_lock();
2615 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2616 if (nvdev) {
2617 cancel_work_sync(&nvdev->subchan_work);
2618 netvsc_xdp_set(net, NULL, NULL, nvdev);
2619 }
2620
2621 /*
2622 * Call to the vsc driver to let it know that the device is being
2623 * removed. Also blocks mtu and channel changes.
2624 */
2625 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2626 if (vf_netdev)
2627 netvsc_unregister_vf(vf_netdev);
2628
2629 if (nvdev)
2630 rndis_filter_device_remove(dev, nvdev);
2631
2632 unregister_netdevice(net);
2633 list_del(&ndev_ctx->list);
2634
2635 rtnl_unlock();
2636
2637 hv_set_drvdata(dev, NULL);
2638
2639 free_percpu(ndev_ctx->vf_stats);
2640 free_netdev(net);
2641 return 0;
2642 }
2643
netvsc_suspend(struct hv_device * dev)2644 static int netvsc_suspend(struct hv_device *dev)
2645 {
2646 struct net_device_context *ndev_ctx;
2647 struct netvsc_device *nvdev;
2648 struct net_device *net;
2649 int ret;
2650
2651 net = hv_get_drvdata(dev);
2652
2653 ndev_ctx = netdev_priv(net);
2654 cancel_delayed_work_sync(&ndev_ctx->dwork);
2655
2656 rtnl_lock();
2657
2658 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2659 if (nvdev == NULL) {
2660 ret = -ENODEV;
2661 goto out;
2662 }
2663
2664 /* Save the current config info */
2665 ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2666
2667 ret = netvsc_detach(net, nvdev);
2668 out:
2669 rtnl_unlock();
2670
2671 return ret;
2672 }
2673
netvsc_resume(struct hv_device * dev)2674 static int netvsc_resume(struct hv_device *dev)
2675 {
2676 struct net_device *net = hv_get_drvdata(dev);
2677 struct net_device_context *net_device_ctx;
2678 struct netvsc_device_info *device_info;
2679 int ret;
2680
2681 rtnl_lock();
2682
2683 net_device_ctx = netdev_priv(net);
2684
2685 /* Reset the data path to the netvsc NIC before re-opening the vmbus
2686 * channel. Later netvsc_netdev_event() will switch the data path to
2687 * the VF upon the UP or CHANGE event.
2688 */
2689 net_device_ctx->data_path_is_vf = false;
2690 device_info = net_device_ctx->saved_netvsc_dev_info;
2691
2692 ret = netvsc_attach(net, device_info);
2693
2694 netvsc_devinfo_put(device_info);
2695 net_device_ctx->saved_netvsc_dev_info = NULL;
2696
2697 rtnl_unlock();
2698
2699 return ret;
2700 }
2701 static const struct hv_vmbus_device_id id_table[] = {
2702 /* Network guid */
2703 { HV_NIC_GUID, },
2704 { },
2705 };
2706
2707 MODULE_DEVICE_TABLE(vmbus, id_table);
2708
2709 /* The one and only one */
2710 static struct hv_driver netvsc_drv = {
2711 .name = KBUILD_MODNAME,
2712 .id_table = id_table,
2713 .probe = netvsc_probe,
2714 .remove = netvsc_remove,
2715 .suspend = netvsc_suspend,
2716 .resume = netvsc_resume,
2717 .driver = {
2718 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2719 },
2720 };
2721
2722 /*
2723 * On Hyper-V, every VF interface is matched with a corresponding
2724 * synthetic interface. The synthetic interface is presented first
2725 * to the guest. When the corresponding VF instance is registered,
2726 * we will take care of switching the data path.
2727 */
netvsc_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)2728 static int netvsc_netdev_event(struct notifier_block *this,
2729 unsigned long event, void *ptr)
2730 {
2731 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2732
2733 /* Skip our own events */
2734 if (event_dev->netdev_ops == &device_ops)
2735 return NOTIFY_DONE;
2736
2737 /* Avoid non-Ethernet type devices */
2738 if (event_dev->type != ARPHRD_ETHER)
2739 return NOTIFY_DONE;
2740
2741 /* Avoid Vlan dev with same MAC registering as VF */
2742 if (is_vlan_dev(event_dev))
2743 return NOTIFY_DONE;
2744
2745 /* Avoid Bonding master dev with same MAC registering as VF */
2746 if (netif_is_bond_master(event_dev))
2747 return NOTIFY_DONE;
2748
2749 switch (event) {
2750 case NETDEV_REGISTER:
2751 return netvsc_register_vf(event_dev);
2752 case NETDEV_UNREGISTER:
2753 return netvsc_unregister_vf(event_dev);
2754 case NETDEV_UP:
2755 case NETDEV_DOWN:
2756 case NETDEV_CHANGE:
2757 case NETDEV_GOING_DOWN:
2758 return netvsc_vf_changed(event_dev, event);
2759 default:
2760 return NOTIFY_DONE;
2761 }
2762 }
2763
2764 static struct notifier_block netvsc_netdev_notifier = {
2765 .notifier_call = netvsc_netdev_event,
2766 };
2767
netvsc_drv_exit(void)2768 static void __exit netvsc_drv_exit(void)
2769 {
2770 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2771 vmbus_driver_unregister(&netvsc_drv);
2772 }
2773
netvsc_drv_init(void)2774 static int __init netvsc_drv_init(void)
2775 {
2776 int ret;
2777
2778 if (ring_size < RING_SIZE_MIN) {
2779 ring_size = RING_SIZE_MIN;
2780 pr_info("Increased ring_size to %u (min allowed)\n",
2781 ring_size);
2782 }
2783 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2784
2785 ret = vmbus_driver_register(&netvsc_drv);
2786 if (ret)
2787 return ret;
2788
2789 register_netdevice_notifier(&netvsc_netdev_notifier);
2790 return 0;
2791 }
2792
2793 MODULE_LICENSE("GPL");
2794 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2795
2796 module_init(netvsc_drv_init);
2797 module_exit(netvsc_drv_exit);
2798