?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / net / hyperv / netvsc_drv.c
blobbdb55db4523b15e7989314d105f33640dde37e2c
1 /*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
19 */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
27 #include <linux/io.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/pci.h>
33 #include <linux/skbuff.h>
34 #include <linux/if_vlan.h>
35 #include <linux/in.h>
36 #include <linux/slab.h>
37 #include <linux/rtnetlink.h>
38 #include <linux/netpoll.h>
39
40 #include <net/arp.h>
41 #include <net/route.h>
42 #include <net/sock.h>
43 #include <net/pkt_sched.h>
44 #include <net/checksum.h>
45 #include <net/ip6_checksum.h>
46
47 #include "hyperv_net.h"
48
49 #define RING_SIZE_MIN 64
50 #define RETRY_US_LO 5000
51 #define RETRY_US_HI 10000
52 #define RETRY_MAX 2000 /* >10 sec */
53
54 #define LINKCHANGE_INT (2 * HZ)
55 #define VF_TAKEOVER_INT (HZ / 10)
56
57 static unsigned int ring_size __ro_after_init = 128;
58 module_param(ring_size, uint, 0444);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
60 unsigned int netvsc_ring_bytes __ro_after_init;
61
62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
63 NETIF_MSG_LINK | NETIF_MSG_IFUP |
64 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
65 NETIF_MSG_TX_ERR;
66
67 static int debug = -1;
68 module_param(debug, int, 0444);
69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
70
71 static LIST_HEAD(netvsc_dev_list);
72
73 static void netvsc_change_rx_flags(struct net_device *net, int change)
74 {
75 struct net_device_context *ndev_ctx = netdev_priv(net);
76 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
77 int inc;
78
79 if (!vf_netdev)
80 return;
81
82 if (change & IFF_PROMISC) {
83 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
84 dev_set_promiscuity(vf_netdev, inc);
85 }
86
87 if (change & IFF_ALLMULTI) {
88 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
89 dev_set_allmulti(vf_netdev, inc);
90 }
91 }
92
93 static void netvsc_set_rx_mode(struct net_device *net)
94 {
95 struct net_device_context *ndev_ctx = netdev_priv(net);
96 struct net_device *vf_netdev;
97 struct netvsc_device *nvdev;
98
99 rcu_read_lock();
100 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
101 if (vf_netdev) {
102 dev_uc_sync(vf_netdev, net);
103 dev_mc_sync(vf_netdev, net);
104 }
105
106 nvdev = rcu_dereference(ndev_ctx->nvdev);
107 if (nvdev)
108 rndis_filter_update(nvdev);
109 rcu_read_unlock();
110 }
111
112 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
113 struct net_device *ndev)
114 {
115 nvscdev->tx_disable = false;
116 virt_wmb(); /* ensure queue wake up mechanism is on */
117
118 netif_tx_wake_all_queues(ndev);
119 }
120
121 static int netvsc_open(struct net_device *net)
122 {
123 struct net_device_context *ndev_ctx = netdev_priv(net);
124 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
125 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
126 struct rndis_device *rdev;
127 int ret = 0;
128
129 netif_carrier_off(net);
130
131 /* Open up the device */
132 ret = rndis_filter_open(nvdev);
133 if (ret != 0) {
134 netdev_err(net, "unable to open device (ret %d).\n", ret);
135 return ret;
136 }
137
138 rdev = nvdev->extension;
139 if (!rdev->link_state) {
140 netif_carrier_on(net);
141 netvsc_tx_enable(nvdev, net);
142 }
143
144 if (vf_netdev) {
145 /* Setting synthetic device up transparently sets
146 * slave as up. If open fails, then slave will be
147 * still be offline (and not used).
148 */
149 ret = dev_open(vf_netdev);
150 if (ret)
151 netdev_warn(net,
152 "unable to open slave: %s: %d\n",
153 vf_netdev->name, ret);
154 }
155 return 0;
156 }
157
158 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
159 {
160 unsigned int retry = 0;
161 int i;
162
163 /* Ensure pending bytes in ring are read */
164 for (;;) {
165 u32 aread = 0;
166
167 for (i = 0; i < nvdev->num_chn; i++) {
168 struct vmbus_channel *chn
169 = nvdev->chan_table[i].channel;
170
171 if (!chn)
172 continue;
173
174 /* make sure receive not running now */
175 napi_synchronize(&nvdev->chan_table[i].napi);
176
177 aread = hv_get_bytes_to_read(&chn->inbound);
178 if (aread)
179 break;
180
181 aread = hv_get_bytes_to_read(&chn->outbound);
182 if (aread)
183 break;
184 }
185
186 if (aread == 0)
187 return 0;
188
189 if (++retry > RETRY_MAX)
190 return -ETIMEDOUT;
191
192 usleep_range(RETRY_US_LO, RETRY_US_HI);
193 }
194 }
195
196 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
197 struct net_device *ndev)
198 {
199 if (nvscdev) {
200 nvscdev->tx_disable = true;
201 virt_wmb(); /* ensure txq will not wake up after stop */
202 }
203
204 netif_tx_disable(ndev);
205 }
206
207 static int netvsc_close(struct net_device *net)
208 {
209 struct net_device_context *net_device_ctx = netdev_priv(net);
210 struct net_device *vf_netdev
211 = rtnl_dereference(net_device_ctx->vf_netdev);
212 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
213 int ret;
214
215 netvsc_tx_disable(nvdev, net);
216
217 /* No need to close rndis filter if it is removed already */
218 if (!nvdev)
219 return 0;
220
221 ret = rndis_filter_close(nvdev);
222 if (ret != 0) {
223 netdev_err(net, "unable to close device (ret %d).\n", ret);
224 return ret;
225 }
226
227 ret = netvsc_wait_until_empty(nvdev);
228 if (ret)
229 netdev_err(net, "Ring buffer not empty after closing rndis\n");
230
231 if (vf_netdev)
232 dev_close(vf_netdev);
233
234 return ret;
235 }
236
237 static inline void *init_ppi_data(struct rndis_message *msg,
238 u32 ppi_size, u32 pkt_type)
239 {
240 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
241 struct rndis_per_packet_info *ppi;
242
243 rndis_pkt->data_offset += ppi_size;
244 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
245 + rndis_pkt->per_pkt_info_len;
246
247 ppi->size = ppi_size;
248 ppi->type = pkt_type;
249 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
250
251 rndis_pkt->per_pkt_info_len += ppi_size;
252
253 return ppi + 1;
254 }
255
256 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
257 * packets. We can use ethtool to change UDP hash level when necessary.
258 */
259 static inline u32 netvsc_get_hash(
260 struct sk_buff *skb,
261 const struct net_device_context *ndc)
262 {
263 struct flow_keys flow;
264 u32 hash, pkt_proto = 0;
265 static u32 hashrnd __read_mostly;
266
267 net_get_random_once(&hashrnd, sizeof(hashrnd));
268
269 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
270 return 0;
271
272 switch (flow.basic.ip_proto) {
273 case IPPROTO_TCP:
274 if (flow.basic.n_proto == htons(ETH_P_IP))
275 pkt_proto = HV_TCP4_L4HASH;
276 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
277 pkt_proto = HV_TCP6_L4HASH;
278
279 break;
280
281 case IPPROTO_UDP:
282 if (flow.basic.n_proto == htons(ETH_P_IP))
283 pkt_proto = HV_UDP4_L4HASH;
284 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
285 pkt_proto = HV_UDP6_L4HASH;
286
287 break;
288 }
289
290 if (pkt_proto & ndc->l4_hash) {
291 return skb_get_hash(skb);
292 } else {
293 if (flow.basic.n_proto == htons(ETH_P_IP))
294 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
295 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
296 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
297 else
298 return 0;
299
300 __skb_set_sw_hash(skb, hash, false);
301 }
302
303 return hash;
304 }
305
306 static inline int netvsc_get_tx_queue(struct net_device *ndev,
307 struct sk_buff *skb, int old_idx)
308 {
309 const struct net_device_context *ndc = netdev_priv(ndev);
310 struct sock *sk = skb->sk;
311 int q_idx;
312
313 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
314 (VRSS_SEND_TAB_SIZE - 1)];
315
316 /* If queue index changed record the new value */
317 if (q_idx != old_idx &&
318 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
319 sk_tx_queue_set(sk, q_idx);
320
321 return q_idx;
322 }
323
324 /*
325 * Select queue for transmit.
326 *
327 * If a valid queue has already been assigned, then use that.
328 * Otherwise compute tx queue based on hash and the send table.
329 *
330 * This is basically similar to default (__netdev_pick_tx) with the added step
331 * of using the host send_table when no other queue has been assigned.
332 *
333 * TODO support XPS - but get_xps_queue not exported
334 */
335 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
336 {
337 int q_idx = sk_tx_queue_get(skb->sk);
338
339 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
340 /* If forwarding a packet, we use the recorded queue when
341 * available for better cache locality.
342 */
343 if (skb_rx_queue_recorded(skb))
344 q_idx = skb_get_rx_queue(skb);
345 else
346 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
347 }
348
349 return q_idx;
350 }
351
352 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
353 struct net_device *sb_dev,
354 select_queue_fallback_t fallback)
355 {
356 struct net_device_context *ndc = netdev_priv(ndev);
357 struct net_device *vf_netdev;
358 u16 txq;
359
360 rcu_read_lock();
361 vf_netdev = rcu_dereference(ndc->vf_netdev);
362 if (vf_netdev) {
363 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
364
365 if (vf_ops->ndo_select_queue)
366 txq = vf_ops->ndo_select_queue(vf_netdev, skb,
367 sb_dev, fallback);
368 else
369 txq = fallback(vf_netdev, skb, NULL);
370
371 /* Record the queue selected by VF so that it can be
372 * used for common case where VF has more queues than
373 * the synthetic device.
374 */
375 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
376 } else {
377 txq = netvsc_pick_tx(ndev, skb);
378 }
379 rcu_read_unlock();
380
381 while (unlikely(txq >= ndev->real_num_tx_queues))
382 txq -= ndev->real_num_tx_queues;
383
384 return txq;
385 }
386
387 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
388 struct hv_page_buffer *pb)
389 {
390 int j = 0;
391
392 /* Deal with compund pages by ignoring unused part
393 * of the page.
394 */
395 page += (offset >> PAGE_SHIFT);
396 offset &= ~PAGE_MASK;
397
398 while (len > 0) {
399 unsigned long bytes;
400
401 bytes = PAGE_SIZE - offset;
402 if (bytes > len)
403 bytes = len;
404 pb[j].pfn = page_to_pfn(page);
405 pb[j].offset = offset;
406 pb[j].len = bytes;
407
408 offset += bytes;
409 len -= bytes;
410
411 if (offset == PAGE_SIZE && len) {
412 page++;
413 offset = 0;
414 j++;
415 }
416 }
417
418 return j + 1;
419 }
420
421 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
422 struct hv_netvsc_packet *packet,
423 struct hv_page_buffer *pb)
424 {
425 u32 slots_used = 0;
426 char *data = skb->data;
427 int frags = skb_shinfo(skb)->nr_frags;
428 int i;
429
430 /* The packet is laid out thus:
431 * 1. hdr: RNDIS header and PPI
432 * 2. skb linear data
433 * 3. skb fragment data
434 */
435 slots_used += fill_pg_buf(virt_to_page(hdr),
436 offset_in_page(hdr),
437 len, &pb[slots_used]);
438
439 packet->rmsg_size = len;
440 packet->rmsg_pgcnt = slots_used;
441
442 slots_used += fill_pg_buf(virt_to_page(data),
443 offset_in_page(data),
444 skb_headlen(skb), &pb[slots_used]);
445
446 for (i = 0; i < frags; i++) {
447 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
448
449 slots_used += fill_pg_buf(skb_frag_page(frag),
450 frag->page_offset,
451 skb_frag_size(frag), &pb[slots_used]);
452 }
453 return slots_used;
454 }
455
456 static int count_skb_frag_slots(struct sk_buff *skb)
457 {
458 int i, frags = skb_shinfo(skb)->nr_frags;
459 int pages = 0;
460
461 for (i = 0; i < frags; i++) {
462 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
463 unsigned long size = skb_frag_size(frag);
464 unsigned long offset = frag->page_offset;
465
466 /* Skip unused frames from start of page */
467 offset &= ~PAGE_MASK;
468 pages += PFN_UP(offset + size);
469 }
470 return pages;
471 }
472
473 static int netvsc_get_slots(struct sk_buff *skb)
474 {
475 char *data = skb->data;
476 unsigned int offset = offset_in_page(data);
477 unsigned int len = skb_headlen(skb);
478 int slots;
479 int frag_slots;
480
481 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
482 frag_slots = count_skb_frag_slots(skb);
483 return slots + frag_slots;
484 }
485
486 static u32 net_checksum_info(struct sk_buff *skb)
487 {
488 if (skb->protocol == htons(ETH_P_IP)) {
489 struct iphdr *ip = ip_hdr(skb);
490
491 if (ip->protocol == IPPROTO_TCP)
492 return TRANSPORT_INFO_IPV4_TCP;
493 else if (ip->protocol == IPPROTO_UDP)
494 return TRANSPORT_INFO_IPV4_UDP;
495 } else {
496 struct ipv6hdr *ip6 = ipv6_hdr(skb);
497
498 if (ip6->nexthdr == IPPROTO_TCP)
499 return TRANSPORT_INFO_IPV6_TCP;
500 else if (ip6->nexthdr == IPPROTO_UDP)
501 return TRANSPORT_INFO_IPV6_UDP;
502 }
503
504 return TRANSPORT_INFO_NOT_IP;
505 }
506
507 /* Send skb on the slave VF device. */
508 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
509 struct sk_buff *skb)
510 {
511 struct net_device_context *ndev_ctx = netdev_priv(net);
512 unsigned int len = skb->len;
513 int rc;
514
515 skb->dev = vf_netdev;
516 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
517
518 rc = dev_queue_xmit(skb);
519 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
520 struct netvsc_vf_pcpu_stats *pcpu_stats
521 = this_cpu_ptr(ndev_ctx->vf_stats);
522
523 u64_stats_update_begin(&pcpu_stats->syncp);
524 pcpu_stats->tx_packets++;
525 pcpu_stats->tx_bytes += len;
526 u64_stats_update_end(&pcpu_stats->syncp);
527 } else {
528 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
529 }
530
531 return rc;
532 }
533
534 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
535 {
536 struct net_device_context *net_device_ctx = netdev_priv(net);
537 struct hv_netvsc_packet *packet = NULL;
538 int ret;
539 unsigned int num_data_pgs;
540 struct rndis_message *rndis_msg;
541 struct net_device *vf_netdev;
542 u32 rndis_msg_size;
543 u32 hash;
544 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
545
546 /* if VF is present and up then redirect packets
547 * already called with rcu_read_lock_bh
548 */
549 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
550 if (vf_netdev && netif_running(vf_netdev) &&
551 !netpoll_tx_running(net))
552 return netvsc_vf_xmit(net, vf_netdev, skb);
553
554 /* We will atmost need two pages to describe the rndis
555 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
556 * of pages in a single packet. If skb is scattered around
557 * more pages we try linearizing it.
558 */
559
560 num_data_pgs = netvsc_get_slots(skb) + 2;
561
562 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
563 ++net_device_ctx->eth_stats.tx_scattered;
564
565 if (skb_linearize(skb))
566 goto no_memory;
567
568 num_data_pgs = netvsc_get_slots(skb) + 2;
569 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
570 ++net_device_ctx->eth_stats.tx_too_big;
571 goto drop;
572 }
573 }
574
575 /*
576 * Place the rndis header in the skb head room and
577 * the skb->cb will be used for hv_netvsc_packet
578 * structure.
579 */
580 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
581 if (ret)
582 goto no_memory;
583
584 /* Use the skb control buffer for building up the packet */
585 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
586 FIELD_SIZEOF(struct sk_buff, cb));
587 packet = (struct hv_netvsc_packet *)skb->cb;
588
589 packet->q_idx = skb_get_queue_mapping(skb);
590
591 packet->total_data_buflen = skb->len;
592 packet->total_bytes = skb->len;
593 packet->total_packets = 1;
594
595 rndis_msg = (struct rndis_message *)skb->head;
596
597 /* Add the rndis header */
598 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
599 rndis_msg->msg_len = packet->total_data_buflen;
600
601 rndis_msg->msg.pkt = (struct rndis_packet) {
602 .data_offset = sizeof(struct rndis_packet),
603 .data_len = packet->total_data_buflen,
604 .per_pkt_info_offset = sizeof(struct rndis_packet),
605 };
606
607 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
608
609 hash = skb_get_hash_raw(skb);
610 if (hash != 0 && net->real_num_tx_queues > 1) {
611 u32 *hash_info;
612
613 rndis_msg_size += NDIS_HASH_PPI_SIZE;
614 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
615 NBL_HASH_VALUE);
616 *hash_info = hash;
617 }
618
619 if (skb_vlan_tag_present(skb)) {
620 struct ndis_pkt_8021q_info *vlan;
621
622 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
623 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
624 IEEE_8021Q_INFO);
625
626 vlan->value = 0;
627 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
628 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
629 VLAN_PRIO_SHIFT;
630 }
631
632 if (skb_is_gso(skb)) {
633 struct ndis_tcp_lso_info *lso_info;
634
635 rndis_msg_size += NDIS_LSO_PPI_SIZE;
636 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
637 TCP_LARGESEND_PKTINFO);
638
639 lso_info->value = 0;
640 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
641 if (skb->protocol == htons(ETH_P_IP)) {
642 lso_info->lso_v2_transmit.ip_version =
643 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
644 ip_hdr(skb)->tot_len = 0;
645 ip_hdr(skb)->check = 0;
646 tcp_hdr(skb)->check =
647 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
648 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
649 } else {
650 lso_info->lso_v2_transmit.ip_version =
651 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
652 ipv6_hdr(skb)->payload_len = 0;
653 tcp_hdr(skb)->check =
654 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
655 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
656 }
657 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
658 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
659 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
660 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
661 struct ndis_tcp_ip_checksum_info *csum_info;
662
663 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
664 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
665 TCPIP_CHKSUM_PKTINFO);
666
667 csum_info->value = 0;
668 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
669
670 if (skb->protocol == htons(ETH_P_IP)) {
671 csum_info->transmit.is_ipv4 = 1;
672
673 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
674 csum_info->transmit.tcp_checksum = 1;
675 else
676 csum_info->transmit.udp_checksum = 1;
677 } else {
678 csum_info->transmit.is_ipv6 = 1;
679
680 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
681 csum_info->transmit.tcp_checksum = 1;
682 else
683 csum_info->transmit.udp_checksum = 1;
684 }
685 } else {
686 /* Can't do offload of this type of checksum */
687 if (skb_checksum_help(skb))
688 goto drop;
689 }
690 }
691
692 /* Start filling in the page buffers with the rndis hdr */
693 rndis_msg->msg_len += rndis_msg_size;
694 packet->total_data_buflen = rndis_msg->msg_len;
695 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
696 skb, packet, pb);
697
698 /* timestamp packet in software */
699 skb_tx_timestamp(skb);
700
701 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
702 if (likely(ret == 0))
703 return NETDEV_TX_OK;
704
705 if (ret == -EAGAIN) {
706 ++net_device_ctx->eth_stats.tx_busy;
707 return NETDEV_TX_BUSY;
708 }
709
710 if (ret == -ENOSPC)
711 ++net_device_ctx->eth_stats.tx_no_space;
712
713 drop:
714 dev_kfree_skb_any(skb);
715 net->stats.tx_dropped++;
716
717 return NETDEV_TX_OK;
718
719 no_memory:
720 ++net_device_ctx->eth_stats.tx_no_memory;
721 goto drop;
722 }
723
724 /*
725 * netvsc_linkstatus_callback - Link up/down notification
726 */
727 void netvsc_linkstatus_callback(struct net_device *net,
728 struct rndis_message *resp)
729 {
730 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
731 struct net_device_context *ndev_ctx = netdev_priv(net);
732 struct netvsc_reconfig *event;
733 unsigned long flags;
734
735 /* Update the physical link speed when changing to another vSwitch */
736 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
737 u32 speed;
738
739 speed = *(u32 *)((void *)indicate
740 + indicate->status_buf_offset) / 10000;
741 ndev_ctx->speed = speed;
742 return;
743 }
744
745 /* Handle these link change statuses below */
746 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
747 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
748 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
749 return;
750
751 if (net->reg_state != NETREG_REGISTERED)
752 return;
753
754 event = kzalloc(sizeof(*event), GFP_ATOMIC);
755 if (!event)
756 return;
757 event->event = indicate->status;
758
759 spin_lock_irqsave(&ndev_ctx->lock, flags);
760 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
761 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
762
763 schedule_delayed_work(&ndev_ctx->dwork, 0);
764 }
765
766 static void netvsc_comp_ipcsum(struct sk_buff *skb)
767 {
768 struct iphdr *iph = (struct iphdr *)skb->data;
769
770 iph->check = 0;
771 iph->check = ip_fast_csum(iph, iph->ihl);
772 }
773
774 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
775 struct napi_struct *napi,
776 const struct ndis_tcp_ip_checksum_info *csum_info,
777 const struct ndis_pkt_8021q_info *vlan,
778 void *data, u32 buflen)
779 {
780 struct sk_buff *skb;
781
782 skb = napi_alloc_skb(napi, buflen);
783 if (!skb)
784 return skb;
785
786 /*
787 * Copy to skb. This copy is needed here since the memory pointed by
788 * hv_netvsc_packet cannot be deallocated
789 */
790 skb_put_data(skb, data, buflen);
791
792 skb->protocol = eth_type_trans(skb, net);
793
794 /* skb is already created with CHECKSUM_NONE */
795 skb_checksum_none_assert(skb);
796
797 /* Incoming packets may have IP header checksum verified by the host.
798 * They may not have IP header checksum computed after coalescing.
799 * We compute it here if the flags are set, because on Linux, the IP
800 * checksum is always checked.
801 */
802 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
803 csum_info->receive.ip_checksum_succeeded &&
804 skb->protocol == htons(ETH_P_IP))
805 netvsc_comp_ipcsum(skb);
806
807 /* Do L4 checksum offload if enabled and present. */
808 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
809 if (csum_info->receive.tcp_checksum_succeeded ||
810 csum_info->receive.udp_checksum_succeeded)
811 skb->ip_summed = CHECKSUM_UNNECESSARY;
812 }
813
814 if (vlan) {
815 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
816
817 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
818 vlan_tci);
819 }
820
821 return skb;
822 }
823
824 /*
825 * netvsc_recv_callback - Callback when we receive a packet from the
826 * "wire" on the specified device.
827 */
828 int netvsc_recv_callback(struct net_device *net,
829 struct netvsc_device *net_device,
830 struct vmbus_channel *channel,
831 void *data, u32 len,
832 const struct ndis_tcp_ip_checksum_info *csum_info,
833 const struct ndis_pkt_8021q_info *vlan)
834 {
835 struct net_device_context *net_device_ctx = netdev_priv(net);
836 u16 q_idx = channel->offermsg.offer.sub_channel_index;
837 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
838 struct sk_buff *skb;
839 struct netvsc_stats *rx_stats;
840
841 if (net->reg_state != NETREG_REGISTERED)
842 return NVSP_STAT_FAIL;
843
844 /* Allocate a skb - TODO direct I/O to pages? */
845 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
846 csum_info, vlan, data, len);
847 if (unlikely(!skb)) {
848 ++net_device_ctx->eth_stats.rx_no_memory;
849 return NVSP_STAT_FAIL;
850 }
851
852 skb_record_rx_queue(skb, q_idx);
853
854 /*
855 * Even if injecting the packet, record the statistics
856 * on the synthetic device because modifying the VF device
857 * statistics will not work correctly.
858 */
859 rx_stats = &nvchan->rx_stats;
860 u64_stats_update_begin(&rx_stats->syncp);
861 rx_stats->packets++;
862 rx_stats->bytes += len;
863
864 if (skb->pkt_type == PACKET_BROADCAST)
865 ++rx_stats->broadcast;
866 else if (skb->pkt_type == PACKET_MULTICAST)
867 ++rx_stats->multicast;
868 u64_stats_update_end(&rx_stats->syncp);
869
870 napi_gro_receive(&nvchan->napi, skb);
871 return NVSP_STAT_SUCCESS;
872 }
873
874 static void netvsc_get_drvinfo(struct net_device *net,
875 struct ethtool_drvinfo *info)
876 {
877 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
878 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
879 }
880
881 static void netvsc_get_channels(struct net_device *net,
882 struct ethtool_channels *channel)
883 {
884 struct net_device_context *net_device_ctx = netdev_priv(net);
885 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
886
887 if (nvdev) {
888 channel->max_combined = nvdev->max_chn;
889 channel->combined_count = nvdev->num_chn;
890 }
891 }
892
893 /* Alloc struct netvsc_device_info, and initialize it from either existing
894 * struct netvsc_device, or from default values.
895 */
896 static struct netvsc_device_info *netvsc_devinfo_get
897 (struct netvsc_device *nvdev)
898 {
899 struct netvsc_device_info *dev_info;
900
901 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
902
903 if (!dev_info)
904 return NULL;
905
906 if (nvdev) {
907 dev_info->num_chn = nvdev->num_chn;
908 dev_info->send_sections = nvdev->send_section_cnt;
909 dev_info->send_section_size = nvdev->send_section_size;
910 dev_info->recv_sections = nvdev->recv_section_cnt;
911 dev_info->recv_section_size = nvdev->recv_section_size;
912
913 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
914 NETVSC_HASH_KEYLEN);
915 } else {
916 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
917 dev_info->send_sections = NETVSC_DEFAULT_TX;
918 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
919 dev_info->recv_sections = NETVSC_DEFAULT_RX;
920 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
921 }
922
923 return dev_info;
924 }
925
926 static int netvsc_detach(struct net_device *ndev,
927 struct netvsc_device *nvdev)
928 {
929 struct net_device_context *ndev_ctx = netdev_priv(ndev);
930 struct hv_device *hdev = ndev_ctx->device_ctx;
931 int ret;
932
933 /* Don't try continuing to try and setup sub channels */
934 if (cancel_work_sync(&nvdev->subchan_work))
935 nvdev->num_chn = 1;
936
937 /* If device was up (receiving) then shutdown */
938 if (netif_running(ndev)) {
939 netvsc_tx_disable(nvdev, ndev);
940
941 ret = rndis_filter_close(nvdev);
942 if (ret) {
943 netdev_err(ndev,
944 "unable to close device (ret %d).\n", ret);
945 return ret;
946 }
947
948 ret = netvsc_wait_until_empty(nvdev);
949 if (ret) {
950 netdev_err(ndev,
951 "Ring buffer not empty after closing rndis\n");
952 return ret;
953 }
954 }
955
956 netif_device_detach(ndev);
957
958 rndis_filter_device_remove(hdev, nvdev);
959
960 return 0;
961 }
962
963 static int netvsc_attach(struct net_device *ndev,
964 struct netvsc_device_info *dev_info)
965 {
966 struct net_device_context *ndev_ctx = netdev_priv(ndev);
967 struct hv_device *hdev = ndev_ctx->device_ctx;
968 struct netvsc_device *nvdev;
969 struct rndis_device *rdev;
970 int ret;
971
972 nvdev = rndis_filter_device_add(hdev, dev_info);
973 if (IS_ERR(nvdev))
974 return PTR_ERR(nvdev);
975
976 if (nvdev->num_chn > 1) {
977 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
978
979 /* if unavailable, just proceed with one queue */
980 if (ret) {
981 nvdev->max_chn = 1;
982 nvdev->num_chn = 1;
983 }
984 }
985
986 /* In any case device is now ready */
987 nvdev->tx_disable = false;
988 netif_device_attach(ndev);
989
990 /* Note: enable and attach happen when sub-channels setup */
991 netif_carrier_off(ndev);
992
993 if (netif_running(ndev)) {
994 ret = rndis_filter_open(nvdev);
995 if (ret)
996 goto err;
997
998 rdev = nvdev->extension;
999 if (!rdev->link_state)
1000 netif_carrier_on(ndev);
1001 }
1002
1003 return 0;
1004
1005 err:
1006 netif_device_detach(ndev);
1007
1008 rndis_filter_device_remove(hdev, nvdev);
1009
1010 return ret;
1011 }
1012
1013 static int netvsc_set_channels(struct net_device *net,
1014 struct ethtool_channels *channels)
1015 {
1016 struct net_device_context *net_device_ctx = netdev_priv(net);
1017 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1018 unsigned int orig, count = channels->combined_count;
1019 struct netvsc_device_info *device_info;
1020 int ret;
1021
1022 /* We do not support separate count for rx, tx, or other */
1023 if (count == 0 ||
1024 channels->rx_count || channels->tx_count || channels->other_count)
1025 return -EINVAL;
1026
1027 if (!nvdev || nvdev->destroy)
1028 return -ENODEV;
1029
1030 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1031 return -EINVAL;
1032
1033 if (count > nvdev->max_chn)
1034 return -EINVAL;
1035
1036 orig = nvdev->num_chn;
1037
1038 device_info = netvsc_devinfo_get(nvdev);
1039
1040 if (!device_info)
1041 return -ENOMEM;
1042
1043 device_info->num_chn = count;
1044
1045 ret = netvsc_detach(net, nvdev);
1046 if (ret)
1047 goto out;
1048
1049 ret = netvsc_attach(net, device_info);
1050 if (ret) {
1051 device_info->num_chn = orig;
1052 if (netvsc_attach(net, device_info))
1053 netdev_err(net, "restoring channel setting failed\n");
1054 }
1055
1056 out:
1057 kfree(device_info);
1058 return ret;
1059 }
1060
1061 static bool
1062 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1063 {
1064 struct ethtool_link_ksettings diff1 = *cmd;
1065 struct ethtool_link_ksettings diff2 = {};
1066
1067 diff1.base.speed = 0;
1068 diff1.base.duplex = 0;
1069 /* advertising and cmd are usually set */
1070 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1071 diff1.base.cmd = 0;
1072 /* We set port to PORT_OTHER */
1073 diff2.base.port = PORT_OTHER;
1074
1075 return !memcmp(&diff1, &diff2, sizeof(diff1));
1076 }
1077
1078 static void netvsc_init_settings(struct net_device *dev)
1079 {
1080 struct net_device_context *ndc = netdev_priv(dev);
1081
1082 ndc->l4_hash = HV_DEFAULT_L4HASH;
1083
1084 ndc->speed = SPEED_UNKNOWN;
1085 ndc->duplex = DUPLEX_FULL;
1086 }
1087
1088 static int netvsc_get_link_ksettings(struct net_device *dev,
1089 struct ethtool_link_ksettings *cmd)
1090 {
1091 struct net_device_context *ndc = netdev_priv(dev);
1092
1093 cmd->base.speed = ndc->speed;
1094 cmd->base.duplex = ndc->duplex;
1095 cmd->base.port = PORT_OTHER;
1096
1097 return 0;
1098 }
1099
1100 static int netvsc_set_link_ksettings(struct net_device *dev,
1101 const struct ethtool_link_ksettings *cmd)
1102 {
1103 struct net_device_context *ndc = netdev_priv(dev);
1104 u32 speed;
1105
1106 speed = cmd->base.speed;
1107 if (!ethtool_validate_speed(speed) ||
1108 !ethtool_validate_duplex(cmd->base.duplex) ||
1109 !netvsc_validate_ethtool_ss_cmd(cmd))
1110 return -EINVAL;
1111
1112 ndc->speed = speed;
1113 ndc->duplex = cmd->base.duplex;
1114
1115 return 0;
1116 }
1117
1118 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1119 {
1120 struct net_device_context *ndevctx = netdev_priv(ndev);
1121 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1122 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1123 int orig_mtu = ndev->mtu;
1124 struct netvsc_device_info *device_info;
1125 int ret = 0;
1126
1127 if (!nvdev || nvdev->destroy)
1128 return -ENODEV;
1129
1130 device_info = netvsc_devinfo_get(nvdev);
1131
1132 if (!device_info)
1133 return -ENOMEM;
1134
1135 /* Change MTU of underlying VF netdev first. */
1136 if (vf_netdev) {
1137 ret = dev_set_mtu(vf_netdev, mtu);
1138 if (ret)
1139 goto out;
1140 }
1141
1142 ret = netvsc_detach(ndev, nvdev);
1143 if (ret)
1144 goto rollback_vf;
1145
1146 ndev->mtu = mtu;
1147
1148 ret = netvsc_attach(ndev, device_info);
1149 if (!ret)
1150 goto out;
1151
1152 /* Attempt rollback to original MTU */
1153 ndev->mtu = orig_mtu;
1154
1155 if (netvsc_attach(ndev, device_info))
1156 netdev_err(ndev, "restoring mtu failed\n");
1157 rollback_vf:
1158 if (vf_netdev)
1159 dev_set_mtu(vf_netdev, orig_mtu);
1160
1161 out:
1162 kfree(device_info);
1163 return ret;
1164 }
1165
1166 static void netvsc_get_vf_stats(struct net_device *net,
1167 struct netvsc_vf_pcpu_stats *tot)
1168 {
1169 struct net_device_context *ndev_ctx = netdev_priv(net);
1170 int i;
1171
1172 memset(tot, 0, sizeof(*tot));
1173
1174 for_each_possible_cpu(i) {
1175 const struct netvsc_vf_pcpu_stats *stats
1176 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1177 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1178 unsigned int start;
1179
1180 do {
1181 start = u64_stats_fetch_begin_irq(&stats->syncp);
1182 rx_packets = stats->rx_packets;
1183 tx_packets = stats->tx_packets;
1184 rx_bytes = stats->rx_bytes;
1185 tx_bytes = stats->tx_bytes;
1186 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1187
1188 tot->rx_packets += rx_packets;
1189 tot->tx_packets += tx_packets;
1190 tot->rx_bytes += rx_bytes;
1191 tot->tx_bytes += tx_bytes;
1192 tot->tx_dropped += stats->tx_dropped;
1193 }
1194 }
1195
1196 static void netvsc_get_pcpu_stats(struct net_device *net,
1197 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1198 {
1199 struct net_device_context *ndev_ctx = netdev_priv(net);
1200 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1201 int i;
1202
1203 /* fetch percpu stats of vf */
1204 for_each_possible_cpu(i) {
1205 const struct netvsc_vf_pcpu_stats *stats =
1206 per_cpu_ptr(ndev_ctx->vf_stats, i);
1207 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1208 unsigned int start;
1209
1210 do {
1211 start = u64_stats_fetch_begin_irq(&stats->syncp);
1212 this_tot->vf_rx_packets = stats->rx_packets;
1213 this_tot->vf_tx_packets = stats->tx_packets;
1214 this_tot->vf_rx_bytes = stats->rx_bytes;
1215 this_tot->vf_tx_bytes = stats->tx_bytes;
1216 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1217 this_tot->rx_packets = this_tot->vf_rx_packets;
1218 this_tot->tx_packets = this_tot->vf_tx_packets;
1219 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1220 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1221 }
1222
1223 /* fetch percpu stats of netvsc */
1224 for (i = 0; i < nvdev->num_chn; i++) {
1225 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1226 const struct netvsc_stats *stats;
1227 struct netvsc_ethtool_pcpu_stats *this_tot =
1228 &pcpu_tot[nvchan->channel->target_cpu];
1229 u64 packets, bytes;
1230 unsigned int start;
1231
1232 stats = &nvchan->tx_stats;
1233 do {
1234 start = u64_stats_fetch_begin_irq(&stats->syncp);
1235 packets = stats->packets;
1236 bytes = stats->bytes;
1237 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1238
1239 this_tot->tx_bytes += bytes;
1240 this_tot->tx_packets += packets;
1241
1242 stats = &nvchan->rx_stats;
1243 do {
1244 start = u64_stats_fetch_begin_irq(&stats->syncp);
1245 packets = stats->packets;
1246 bytes = stats->bytes;
1247 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1248
1249 this_tot->rx_bytes += bytes;
1250 this_tot->rx_packets += packets;
1251 }
1252 }
1253
1254 static void netvsc_get_stats64(struct net_device *net,
1255 struct rtnl_link_stats64 *t)
1256 {
1257 struct net_device_context *ndev_ctx = netdev_priv(net);
1258 struct netvsc_device *nvdev;
1259 struct netvsc_vf_pcpu_stats vf_tot;
1260 int i;
1261
1262 rcu_read_lock();
1263
1264 nvdev = rcu_dereference(ndev_ctx->nvdev);
1265 if (!nvdev)
1266 goto out;
1267
1268 netdev_stats_to_stats64(t, &net->stats);
1269
1270 netvsc_get_vf_stats(net, &vf_tot);
1271 t->rx_packets += vf_tot.rx_packets;
1272 t->tx_packets += vf_tot.tx_packets;
1273 t->rx_bytes += vf_tot.rx_bytes;
1274 t->tx_bytes += vf_tot.tx_bytes;
1275 t->tx_dropped += vf_tot.tx_dropped;
1276
1277 for (i = 0; i < nvdev->num_chn; i++) {
1278 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1279 const struct netvsc_stats *stats;
1280 u64 packets, bytes, multicast;
1281 unsigned int start;
1282
1283 stats = &nvchan->tx_stats;
1284 do {
1285 start = u64_stats_fetch_begin_irq(&stats->syncp);
1286 packets = stats->packets;
1287 bytes = stats->bytes;
1288 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1289
1290 t->tx_bytes += bytes;
1291 t->tx_packets += packets;
1292
1293 stats = &nvchan->rx_stats;
1294 do {
1295 start = u64_stats_fetch_begin_irq(&stats->syncp);
1296 packets = stats->packets;
1297 bytes = stats->bytes;
1298 multicast = stats->multicast + stats->broadcast;
1299 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1300
1301 t->rx_bytes += bytes;
1302 t->rx_packets += packets;
1303 t->multicast += multicast;
1304 }
1305 out:
1306 rcu_read_unlock();
1307 }
1308
1309 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1310 {
1311 struct net_device_context *ndc = netdev_priv(ndev);
1312 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1313 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1314 struct sockaddr *addr = p;
1315 int err;
1316
1317 err = eth_prepare_mac_addr_change(ndev, p);
1318 if (err)
1319 return err;
1320
1321 if (!nvdev)
1322 return -ENODEV;
1323
1324 if (vf_netdev) {
1325 err = dev_set_mac_address(vf_netdev, addr);
1326 if (err)
1327 return err;
1328 }
1329
1330 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1331 if (!err) {
1332 eth_commit_mac_addr_change(ndev, p);
1333 } else if (vf_netdev) {
1334 /* rollback change on VF */
1335 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1336 dev_set_mac_address(vf_netdev, addr);
1337 }
1338
1339 return err;
1340 }
1341
1342 static const struct {
1343 char name[ETH_GSTRING_LEN];
1344 u16 offset;
1345 } netvsc_stats[] = {
1346 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1347 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1348 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1349 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1350 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1351 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1352 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1353 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1354 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1355 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1356 }, pcpu_stats[] = {
1357 { "cpu%u_rx_packets",
1358 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1359 { "cpu%u_rx_bytes",
1360 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1361 { "cpu%u_tx_packets",
1362 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1363 { "cpu%u_tx_bytes",
1364 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1365 { "cpu%u_vf_rx_packets",
1366 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1367 { "cpu%u_vf_rx_bytes",
1368 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1369 { "cpu%u_vf_tx_packets",
1370 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1371 { "cpu%u_vf_tx_bytes",
1372 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1373 }, vf_stats[] = {
1374 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1375 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1376 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1377 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1378 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1379 };
1380
1381 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1382 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1383
1384 /* statistics per queue (rx/tx packets/bytes) */
1385 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1386
1387 /* 4 statistics per queue (rx/tx packets/bytes) */
1388 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1389
1390 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1391 {
1392 struct net_device_context *ndc = netdev_priv(dev);
1393 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1394
1395 if (!nvdev)
1396 return -ENODEV;
1397
1398 switch (string_set) {
1399 case ETH_SS_STATS:
1400 return NETVSC_GLOBAL_STATS_LEN
1401 + NETVSC_VF_STATS_LEN
1402 + NETVSC_QUEUE_STATS_LEN(nvdev)
1403 + NETVSC_PCPU_STATS_LEN;
1404 default:
1405 return -EINVAL;
1406 }
1407 }
1408
1409 static void netvsc_get_ethtool_stats(struct net_device *dev,
1410 struct ethtool_stats *stats, u64 *data)
1411 {
1412 struct net_device_context *ndc = netdev_priv(dev);
1413 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1414 const void *nds = &ndc->eth_stats;
1415 const struct netvsc_stats *qstats;
1416 struct netvsc_vf_pcpu_stats sum;
1417 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1418 unsigned int start;
1419 u64 packets, bytes;
1420 int i, j, cpu;
1421
1422 if (!nvdev)
1423 return;
1424
1425 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1426 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1427
1428 netvsc_get_vf_stats(dev, &sum);
1429 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1430 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1431
1432 for (j = 0; j < nvdev->num_chn; j++) {
1433 qstats = &nvdev->chan_table[j].tx_stats;
1434
1435 do {
1436 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1437 packets = qstats->packets;
1438 bytes = qstats->bytes;
1439 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1440 data[i++] = packets;
1441 data[i++] = bytes;
1442
1443 qstats = &nvdev->chan_table[j].rx_stats;
1444 do {
1445 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1446 packets = qstats->packets;
1447 bytes = qstats->bytes;
1448 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1449 data[i++] = packets;
1450 data[i++] = bytes;
1451 }
1452
1453 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1454 sizeof(struct netvsc_ethtool_pcpu_stats),
1455 GFP_KERNEL);
1456 netvsc_get_pcpu_stats(dev, pcpu_sum);
1457 for_each_present_cpu(cpu) {
1458 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1459
1460 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1461 data[i++] = *(u64 *)((void *)this_sum
1462 + pcpu_stats[j].offset);
1463 }
1464 kvfree(pcpu_sum);
1465 }
1466
1467 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1468 {
1469 struct net_device_context *ndc = netdev_priv(dev);
1470 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1471 u8 *p = data;
1472 int i, cpu;
1473
1474 if (!nvdev)
1475 return;
1476
1477 switch (stringset) {
1478 case ETH_SS_STATS:
1479 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1480 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1481 p += ETH_GSTRING_LEN;
1482 }
1483
1484 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1485 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1486 p += ETH_GSTRING_LEN;
1487 }
1488
1489 for (i = 0; i < nvdev->num_chn; i++) {
1490 sprintf(p, "tx_queue_%u_packets", i);
1491 p += ETH_GSTRING_LEN;
1492 sprintf(p, "tx_queue_%u_bytes", i);
1493 p += ETH_GSTRING_LEN;
1494 sprintf(p, "rx_queue_%u_packets", i);
1495 p += ETH_GSTRING_LEN;
1496 sprintf(p, "rx_queue_%u_bytes", i);
1497 p += ETH_GSTRING_LEN;
1498 }
1499
1500 for_each_present_cpu(cpu) {
1501 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1502 sprintf(p, pcpu_stats[i].name, cpu);
1503 p += ETH_GSTRING_LEN;
1504 }
1505 }
1506
1507 break;
1508 }
1509 }
1510
1511 static int
1512 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1513 struct ethtool_rxnfc *info)
1514 {
1515 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1516
1517 info->data = RXH_IP_SRC | RXH_IP_DST;
1518
1519 switch (info->flow_type) {
1520 case TCP_V4_FLOW:
1521 if (ndc->l4_hash & HV_TCP4_L4HASH)
1522 info->data |= l4_flag;
1523
1524 break;
1525
1526 case TCP_V6_FLOW:
1527 if (ndc->l4_hash & HV_TCP6_L4HASH)
1528 info->data |= l4_flag;
1529
1530 break;
1531
1532 case UDP_V4_FLOW:
1533 if (ndc->l4_hash & HV_UDP4_L4HASH)
1534 info->data |= l4_flag;
1535
1536 break;
1537
1538 case UDP_V6_FLOW:
1539 if (ndc->l4_hash & HV_UDP6_L4HASH)
1540 info->data |= l4_flag;
1541
1542 break;
1543
1544 case IPV4_FLOW:
1545 case IPV6_FLOW:
1546 break;
1547 default:
1548 info->data = 0;
1549 break;
1550 }
1551
1552 return 0;
1553 }
1554
1555 static int
1556 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1557 u32 *rules)
1558 {
1559 struct net_device_context *ndc = netdev_priv(dev);
1560 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1561
1562 if (!nvdev)
1563 return -ENODEV;
1564
1565 switch (info->cmd) {
1566 case ETHTOOL_GRXRINGS:
1567 info->data = nvdev->num_chn;
1568 return 0;
1569
1570 case ETHTOOL_GRXFH:
1571 return netvsc_get_rss_hash_opts(ndc, info);
1572 }
1573 return -EOPNOTSUPP;
1574 }
1575
1576 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1577 struct ethtool_rxnfc *info)
1578 {
1579 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1580 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1581 switch (info->flow_type) {
1582 case TCP_V4_FLOW:
1583 ndc->l4_hash |= HV_TCP4_L4HASH;
1584 break;
1585
1586 case TCP_V6_FLOW:
1587 ndc->l4_hash |= HV_TCP6_L4HASH;
1588 break;
1589
1590 case UDP_V4_FLOW:
1591 ndc->l4_hash |= HV_UDP4_L4HASH;
1592 break;
1593
1594 case UDP_V6_FLOW:
1595 ndc->l4_hash |= HV_UDP6_L4HASH;
1596 break;
1597
1598 default:
1599 return -EOPNOTSUPP;
1600 }
1601
1602 return 0;
1603 }
1604
1605 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1606 switch (info->flow_type) {
1607 case TCP_V4_FLOW:
1608 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1609 break;
1610
1611 case TCP_V6_FLOW:
1612 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1613 break;
1614
1615 case UDP_V4_FLOW:
1616 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1617 break;
1618
1619 case UDP_V6_FLOW:
1620 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1621 break;
1622
1623 default:
1624 return -EOPNOTSUPP;
1625 }
1626
1627 return 0;
1628 }
1629
1630 return -EOPNOTSUPP;
1631 }
1632
1633 static int
1634 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1635 {
1636 struct net_device_context *ndc = netdev_priv(ndev);
1637
1638 if (info->cmd == ETHTOOL_SRXFH)
1639 return netvsc_set_rss_hash_opts(ndc, info);
1640
1641 return -EOPNOTSUPP;
1642 }
1643
1644 #ifdef CONFIG_NET_POLL_CONTROLLER
1645 static void netvsc_poll_controller(struct net_device *dev)
1646 {
1647 struct net_device_context *ndc = netdev_priv(dev);
1648 struct netvsc_device *ndev;
1649 int i;
1650
1651 rcu_read_lock();
1652 ndev = rcu_dereference(ndc->nvdev);
1653 if (ndev) {
1654 for (i = 0; i < ndev->num_chn; i++) {
1655 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1656
1657 napi_schedule(&nvchan->napi);
1658 }
1659 }
1660 rcu_read_unlock();
1661 }
1662 #endif
1663
1664 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1665 {
1666 return NETVSC_HASH_KEYLEN;
1667 }
1668
1669 static u32 netvsc_rss_indir_size(struct net_device *dev)
1670 {
1671 return ITAB_NUM;
1672 }
1673
1674 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1675 u8 *hfunc)
1676 {
1677 struct net_device_context *ndc = netdev_priv(dev);
1678 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1679 struct rndis_device *rndis_dev;
1680 int i;
1681
1682 if (!ndev)
1683 return -ENODEV;
1684
1685 if (hfunc)
1686 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1687
1688 rndis_dev = ndev->extension;
1689 if (indir) {
1690 for (i = 0; i < ITAB_NUM; i++)
1691 indir[i] = ndc->rx_table[i];
1692 }
1693
1694 if (key)
1695 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1696
1697 return 0;
1698 }
1699
1700 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1701 const u8 *key, const u8 hfunc)
1702 {
1703 struct net_device_context *ndc = netdev_priv(dev);
1704 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1705 struct rndis_device *rndis_dev;
1706 int i;
1707
1708 if (!ndev)
1709 return -ENODEV;
1710
1711 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1712 return -EOPNOTSUPP;
1713
1714 rndis_dev = ndev->extension;
1715 if (indir) {
1716 for (i = 0; i < ITAB_NUM; i++)
1717 if (indir[i] >= ndev->num_chn)
1718 return -EINVAL;
1719
1720 for (i = 0; i < ITAB_NUM; i++)
1721 ndc->rx_table[i] = indir[i];
1722 }
1723
1724 if (!key) {
1725 if (!indir)
1726 return 0;
1727
1728 key = rndis_dev->rss_key;
1729 }
1730
1731 return rndis_filter_set_rss_param(rndis_dev, key);
1732 }
1733
1734 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1735 * It does have pre-allocated receive area which is divided into sections.
1736 */
1737 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1738 struct ethtool_ringparam *ring)
1739 {
1740 u32 max_buf_size;
1741
1742 ring->rx_pending = nvdev->recv_section_cnt;
1743 ring->tx_pending = nvdev->send_section_cnt;
1744
1745 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1746 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1747 else
1748 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1749
1750 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1751 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1752 / nvdev->send_section_size;
1753 }
1754
1755 static void netvsc_get_ringparam(struct net_device *ndev,
1756 struct ethtool_ringparam *ring)
1757 {
1758 struct net_device_context *ndevctx = netdev_priv(ndev);
1759 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1760
1761 if (!nvdev)
1762 return;
1763
1764 __netvsc_get_ringparam(nvdev, ring);
1765 }
1766
1767 static int netvsc_set_ringparam(struct net_device *ndev,
1768 struct ethtool_ringparam *ring)
1769 {
1770 struct net_device_context *ndevctx = netdev_priv(ndev);
1771 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1772 struct netvsc_device_info *device_info;
1773 struct ethtool_ringparam orig;
1774 u32 new_tx, new_rx;
1775 int ret = 0;
1776
1777 if (!nvdev || nvdev->destroy)
1778 return -ENODEV;
1779
1780 memset(&orig, 0, sizeof(orig));
1781 __netvsc_get_ringparam(nvdev, &orig);
1782
1783 new_tx = clamp_t(u32, ring->tx_pending,
1784 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1785 new_rx = clamp_t(u32, ring->rx_pending,
1786 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1787
1788 if (new_tx == orig.tx_pending &&
1789 new_rx == orig.rx_pending)
1790 return 0; /* no change */
1791
1792 device_info = netvsc_devinfo_get(nvdev);
1793
1794 if (!device_info)
1795 return -ENOMEM;
1796
1797 device_info->send_sections = new_tx;
1798 device_info->recv_sections = new_rx;
1799
1800 ret = netvsc_detach(ndev, nvdev);
1801 if (ret)
1802 goto out;
1803
1804 ret = netvsc_attach(ndev, device_info);
1805 if (ret) {
1806 device_info->send_sections = orig.tx_pending;
1807 device_info->recv_sections = orig.rx_pending;
1808
1809 if (netvsc_attach(ndev, device_info))
1810 netdev_err(ndev, "restoring ringparam failed");
1811 }
1812
1813 out:
1814 kfree(device_info);
1815 return ret;
1816 }
1817
1818 static u32 netvsc_get_msglevel(struct net_device *ndev)
1819 {
1820 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1821
1822 return ndev_ctx->msg_enable;
1823 }
1824
1825 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1826 {
1827 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1828
1829 ndev_ctx->msg_enable = val;
1830 }
1831
1832 static const struct ethtool_ops ethtool_ops = {
1833 .get_drvinfo = netvsc_get_drvinfo,
1834 .get_msglevel = netvsc_get_msglevel,
1835 .set_msglevel = netvsc_set_msglevel,
1836 .get_link = ethtool_op_get_link,
1837 .get_ethtool_stats = netvsc_get_ethtool_stats,
1838 .get_sset_count = netvsc_get_sset_count,
1839 .get_strings = netvsc_get_strings,
1840 .get_channels = netvsc_get_channels,
1841 .set_channels = netvsc_set_channels,
1842 .get_ts_info = ethtool_op_get_ts_info,
1843 .get_rxnfc = netvsc_get_rxnfc,
1844 .set_rxnfc = netvsc_set_rxnfc,
1845 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1846 .get_rxfh_indir_size = netvsc_rss_indir_size,
1847 .get_rxfh = netvsc_get_rxfh,
1848 .set_rxfh = netvsc_set_rxfh,
1849 .get_link_ksettings = netvsc_get_link_ksettings,
1850 .set_link_ksettings = netvsc_set_link_ksettings,
1851 .get_ringparam = netvsc_get_ringparam,
1852 .set_ringparam = netvsc_set_ringparam,
1853 };
1854
1855 static const struct net_device_ops device_ops = {
1856 .ndo_open = netvsc_open,
1857 .ndo_stop = netvsc_close,
1858 .ndo_start_xmit = netvsc_start_xmit,
1859 .ndo_change_rx_flags = netvsc_change_rx_flags,
1860 .ndo_set_rx_mode = netvsc_set_rx_mode,
1861 .ndo_change_mtu = netvsc_change_mtu,
1862 .ndo_validate_addr = eth_validate_addr,
1863 .ndo_set_mac_address = netvsc_set_mac_addr,
1864 .ndo_select_queue = netvsc_select_queue,
1865 .ndo_get_stats64 = netvsc_get_stats64,
1866 #ifdef CONFIG_NET_POLL_CONTROLLER
1867 .ndo_poll_controller = netvsc_poll_controller,
1868 #endif
1869 };
1870
1871 /*
1872 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1873 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1874 * present send GARP packet to network peers with netif_notify_peers().
1875 */
1876 static void netvsc_link_change(struct work_struct *w)
1877 {
1878 struct net_device_context *ndev_ctx =
1879 container_of(w, struct net_device_context, dwork.work);
1880 struct hv_device *device_obj = ndev_ctx->device_ctx;
1881 struct net_device *net = hv_get_drvdata(device_obj);
1882 struct netvsc_device *net_device;
1883 struct rndis_device *rdev;
1884 struct netvsc_reconfig *event = NULL;
1885 bool notify = false, reschedule = false;
1886 unsigned long flags, next_reconfig, delay;
1887
1888 /* if changes are happening, comeback later */
1889 if (!rtnl_trylock()) {
1890 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1891 return;
1892 }
1893
1894 net_device = rtnl_dereference(ndev_ctx->nvdev);
1895 if (!net_device)
1896 goto out_unlock;
1897
1898 rdev = net_device->extension;
1899
1900 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1901 if (time_is_after_jiffies(next_reconfig)) {
1902 /* link_watch only sends one notification with current state
1903 * per second, avoid doing reconfig more frequently. Handle
1904 * wrap around.
1905 */
1906 delay = next_reconfig - jiffies;
1907 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1908 schedule_delayed_work(&ndev_ctx->dwork, delay);
1909 goto out_unlock;
1910 }
1911 ndev_ctx->last_reconfig = jiffies;
1912
1913 spin_lock_irqsave(&ndev_ctx->lock, flags);
1914 if (!list_empty(&ndev_ctx->reconfig_events)) {
1915 event = list_first_entry(&ndev_ctx->reconfig_events,
1916 struct netvsc_reconfig, list);
1917 list_del(&event->list);
1918 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1919 }
1920 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1921
1922 if (!event)
1923 goto out_unlock;
1924
1925 switch (event->event) {
1926 /* Only the following events are possible due to the check in
1927 * netvsc_linkstatus_callback()
1928 */
1929 case RNDIS_STATUS_MEDIA_CONNECT:
1930 if (rdev->link_state) {
1931 rdev->link_state = false;
1932 netif_carrier_on(net);
1933 netvsc_tx_enable(net_device, net);
1934 } else {
1935 notify = true;
1936 }
1937 kfree(event);
1938 break;
1939 case RNDIS_STATUS_MEDIA_DISCONNECT:
1940 if (!rdev->link_state) {
1941 rdev->link_state = true;
1942 netif_carrier_off(net);
1943 netvsc_tx_disable(net_device, net);
1944 }
1945 kfree(event);
1946 break;
1947 case RNDIS_STATUS_NETWORK_CHANGE:
1948 /* Only makes sense if carrier is present */
1949 if (!rdev->link_state) {
1950 rdev->link_state = true;
1951 netif_carrier_off(net);
1952 netvsc_tx_disable(net_device, net);
1953 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1954 spin_lock_irqsave(&ndev_ctx->lock, flags);
1955 list_add(&event->list, &ndev_ctx->reconfig_events);
1956 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1957 reschedule = true;
1958 }
1959 break;
1960 }
1961
1962 rtnl_unlock();
1963
1964 if (notify)
1965 netdev_notify_peers(net);
1966
1967 /* link_watch only sends one notification with current state per
1968 * second, handle next reconfig event in 2 seconds.
1969 */
1970 if (reschedule)
1971 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1972
1973 return;
1974
1975 out_unlock:
1976 rtnl_unlock();
1977 }
1978
1979 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1980 {
1981 struct net_device_context *net_device_ctx;
1982 struct net_device *dev;
1983
1984 dev = netdev_master_upper_dev_get(vf_netdev);
1985 if (!dev || dev->netdev_ops != &device_ops)
1986 return NULL; /* not a netvsc device */
1987
1988 net_device_ctx = netdev_priv(dev);
1989 if (!rtnl_dereference(net_device_ctx->nvdev))
1990 return NULL; /* device is removed */
1991
1992 return dev;
1993 }
1994
1995 /* Called when VF is injecting data into network stack.
1996 * Change the associated network device from VF to netvsc.
1997 * note: already called with rcu_read_lock
1998 */
1999 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2000 {
2001 struct sk_buff *skb = *pskb;
2002 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2003 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2004 struct netvsc_vf_pcpu_stats *pcpu_stats
2005 = this_cpu_ptr(ndev_ctx->vf_stats);
2006
2007 skb = skb_share_check(skb, GFP_ATOMIC);
2008 if (unlikely(!skb))
2009 return RX_HANDLER_CONSUMED;
2010
2011 *pskb = skb;
2012
2013 skb->dev = ndev;
2014
2015 u64_stats_update_begin(&pcpu_stats->syncp);
2016 pcpu_stats->rx_packets++;
2017 pcpu_stats->rx_bytes += skb->len;
2018 u64_stats_update_end(&pcpu_stats->syncp);
2019
2020 return RX_HANDLER_ANOTHER;
2021 }
2022
2023 static int netvsc_vf_join(struct net_device *vf_netdev,
2024 struct net_device *ndev)
2025 {
2026 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2027 int ret;
2028
2029 ret = netdev_rx_handler_register(vf_netdev,
2030 netvsc_vf_handle_frame, ndev);
2031 if (ret != 0) {
2032 netdev_err(vf_netdev,
2033 "can not register netvsc VF receive handler (err = %d)\n",
2034 ret);
2035 goto rx_handler_failed;
2036 }
2037
2038 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2039 NULL, NULL, NULL);
2040 if (ret != 0) {
2041 netdev_err(vf_netdev,
2042 "can not set master device %s (err = %d)\n",
2043 ndev->name, ret);
2044 goto upper_link_failed;
2045 }
2046
2047 /* set slave flag before open to prevent IPv6 addrconf */
2048 vf_netdev->flags |= IFF_SLAVE;
2049
2050 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2051
2052 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2053
2054 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2055 return 0;
2056
2057 upper_link_failed:
2058 netdev_rx_handler_unregister(vf_netdev);
2059 rx_handler_failed:
2060 return ret;
2061 }
2062
2063 static void __netvsc_vf_setup(struct net_device *ndev,
2064 struct net_device *vf_netdev)
2065 {
2066 int ret;
2067
2068 /* Align MTU of VF with master */
2069 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2070 if (ret)
2071 netdev_warn(vf_netdev,
2072 "unable to change mtu to %u\n", ndev->mtu);
2073
2074 /* set multicast etc flags on VF */
2075 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
2076
2077 /* sync address list from ndev to VF */
2078 netif_addr_lock_bh(ndev);
2079 dev_uc_sync(vf_netdev, ndev);
2080 dev_mc_sync(vf_netdev, ndev);
2081 netif_addr_unlock_bh(ndev);
2082
2083 if (netif_running(ndev)) {
2084 ret = dev_open(vf_netdev);
2085 if (ret)
2086 netdev_warn(vf_netdev,
2087 "unable to open: %d\n", ret);
2088 }
2089 }
2090
2091 /* Setup VF as slave of the synthetic device.
2092 * Runs in workqueue to avoid recursion in netlink callbacks.
2093 */
2094 static void netvsc_vf_setup(struct work_struct *w)
2095 {
2096 struct net_device_context *ndev_ctx
2097 = container_of(w, struct net_device_context, vf_takeover.work);
2098 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2099 struct net_device *vf_netdev;
2100
2101 if (!rtnl_trylock()) {
2102 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2103 return;
2104 }
2105
2106 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2107 if (vf_netdev)
2108 __netvsc_vf_setup(ndev, vf_netdev);
2109
2110 rtnl_unlock();
2111 }
2112
2113 /* Find netvsc by VF serial number.
2114 * The PCI hyperv controller records the serial number as the slot kobj name.
2115 */
2116 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2117 {
2118 struct device *parent = vf_netdev->dev.parent;
2119 struct net_device_context *ndev_ctx;
2120 struct pci_dev *pdev;
2121 u32 serial;
2122
2123 if (!parent || !dev_is_pci(parent))
2124 return NULL; /* not a PCI device */
2125
2126 pdev = to_pci_dev(parent);
2127 if (!pdev->slot) {
2128 netdev_notice(vf_netdev, "no PCI slot information\n");
2129 return NULL;
2130 }
2131
2132 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2133 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2134 pci_slot_name(pdev->slot));
2135 return NULL;
2136 }
2137
2138 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2139 if (!ndev_ctx->vf_alloc)
2140 continue;
2141
2142 if (ndev_ctx->vf_serial == serial)
2143 return hv_get_drvdata(ndev_ctx->device_ctx);
2144 }
2145
2146 netdev_notice(vf_netdev,
2147 "no netdev found for vf serial:%u\n", serial);
2148 return NULL;
2149 }
2150
2151 static int netvsc_register_vf(struct net_device *vf_netdev)
2152 {
2153 struct net_device_context *net_device_ctx;
2154 struct netvsc_device *netvsc_dev;
2155 struct net_device *ndev;
2156 int ret;
2157
2158 if (vf_netdev->addr_len != ETH_ALEN)
2159 return NOTIFY_DONE;
2160
2161 ndev = get_netvsc_byslot(vf_netdev);
2162 if (!ndev)
2163 return NOTIFY_DONE;
2164
2165 net_device_ctx = netdev_priv(ndev);
2166 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2167 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2168 return NOTIFY_DONE;
2169
2170 /* if syntihetic interface is a different namespace,
2171 * then move the VF to that namespace; join will be
2172 * done again in that context.
2173 */
2174 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2175 ret = dev_change_net_namespace(vf_netdev,
2176 dev_net(ndev), "eth%d");
2177 if (ret)
2178 netdev_err(vf_netdev,
2179 "could not move to same namespace as %s: %d\n",
2180 ndev->name, ret);
2181 else
2182 netdev_info(vf_netdev,
2183 "VF moved to namespace with: %s\n",
2184 ndev->name);
2185 return NOTIFY_DONE;
2186 }
2187
2188 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2189
2190 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2191 return NOTIFY_DONE;
2192
2193 dev_hold(vf_netdev);
2194 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2195 return NOTIFY_OK;
2196 }
2197
2198 /* VF up/down change detected, schedule to change data path */
2199 static int netvsc_vf_changed(struct net_device *vf_netdev)
2200 {
2201 struct net_device_context *net_device_ctx;
2202 struct netvsc_device *netvsc_dev;
2203 struct net_device *ndev;
2204 bool vf_is_up = netif_running(vf_netdev);
2205
2206 ndev = get_netvsc_byref(vf_netdev);
2207 if (!ndev)
2208 return NOTIFY_DONE;
2209
2210 net_device_ctx = netdev_priv(ndev);
2211 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2212 if (!netvsc_dev)
2213 return NOTIFY_DONE;
2214
2215 netvsc_switch_datapath(ndev, vf_is_up);
2216 netdev_info(ndev, "Data path switched %s VF: %s\n",
2217 vf_is_up ? "to" : "from", vf_netdev->name);
2218
2219 return NOTIFY_OK;
2220 }
2221
2222 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2223 {
2224 struct net_device *ndev;
2225 struct net_device_context *net_device_ctx;
2226
2227 ndev = get_netvsc_byref(vf_netdev);
2228 if (!ndev)
2229 return NOTIFY_DONE;
2230
2231 net_device_ctx = netdev_priv(ndev);
2232 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2233
2234 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2235
2236 netdev_rx_handler_unregister(vf_netdev);
2237 netdev_upper_dev_unlink(vf_netdev, ndev);
2238 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2239 dev_put(vf_netdev);
2240
2241 return NOTIFY_OK;
2242 }
2243
2244 static int netvsc_probe(struct hv_device *dev,
2245 const struct hv_vmbus_device_id *dev_id)
2246 {
2247 struct net_device *net = NULL;
2248 struct net_device_context *net_device_ctx;
2249 struct netvsc_device_info *device_info = NULL;
2250 struct netvsc_device *nvdev;
2251 int ret = -ENOMEM;
2252
2253 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2254 VRSS_CHANNEL_MAX);
2255 if (!net)
2256 goto no_net;
2257
2258 netif_carrier_off(net);
2259
2260 netvsc_init_settings(net);
2261
2262 net_device_ctx = netdev_priv(net);
2263 net_device_ctx->device_ctx = dev;
2264 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2265 if (netif_msg_probe(net_device_ctx))
2266 netdev_dbg(net, "netvsc msg_enable: %d\n",
2267 net_device_ctx->msg_enable);
2268
2269 hv_set_drvdata(dev, net);
2270
2271 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2272
2273 spin_lock_init(&net_device_ctx->lock);
2274 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2275 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2276
2277 net_device_ctx->vf_stats
2278 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2279 if (!net_device_ctx->vf_stats)
2280 goto no_stats;
2281
2282 net->netdev_ops = &device_ops;
2283 net->ethtool_ops = &ethtool_ops;
2284 SET_NETDEV_DEV(net, &dev->device);
2285
2286 /* We always need headroom for rndis header */
2287 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2288
2289 /* Initialize the number of queues to be 1, we may change it if more
2290 * channels are offered later.
2291 */
2292 netif_set_real_num_tx_queues(net, 1);
2293 netif_set_real_num_rx_queues(net, 1);
2294
2295 /* Notify the netvsc driver of the new device */
2296 device_info = netvsc_devinfo_get(NULL);
2297
2298 if (!device_info) {
2299 ret = -ENOMEM;
2300 goto devinfo_failed;
2301 }
2302
2303 nvdev = rndis_filter_device_add(dev, device_info);
2304 if (IS_ERR(nvdev)) {
2305 ret = PTR_ERR(nvdev);
2306 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2307 goto rndis_failed;
2308 }
2309
2310 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2311
2312 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2313 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2314 * all subchannels to show up, but that may not happen because
2315 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2316 * -> ... -> device_add() -> ... -> __device_attach() can't get
2317 * the device lock, so all the subchannels can't be processed --
2318 * finally netvsc_subchan_work() hangs for ever.
2319 */
2320 rtnl_lock();
2321
2322 if (nvdev->num_chn > 1)
2323 schedule_work(&nvdev->subchan_work);
2324
2325 /* hw_features computed in rndis_netdev_set_hwcaps() */
2326 net->features = net->hw_features |
2327 NETIF_F_HIGHDMA | NETIF_F_SG |
2328 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2329 net->vlan_features = net->features;
2330
2331 netdev_lockdep_set_classes(net);
2332
2333 /* MTU range: 68 - 1500 or 65521 */
2334 net->min_mtu = NETVSC_MTU_MIN;
2335 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2336 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2337 else
2338 net->max_mtu = ETH_DATA_LEN;
2339
2340 nvdev->tx_disable = false;
2341
2342 ret = register_netdevice(net);
2343 if (ret != 0) {
2344 pr_err("Unable to register netdev.\n");
2345 goto register_failed;
2346 }
2347
2348 list_add(&net_device_ctx->list, &netvsc_dev_list);
2349 rtnl_unlock();
2350
2351 kfree(device_info);
2352 return 0;
2353
2354 register_failed:
2355 rtnl_unlock();
2356 rndis_filter_device_remove(dev, nvdev);
2357 rndis_failed:
2358 kfree(device_info);
2359 devinfo_failed:
2360 free_percpu(net_device_ctx->vf_stats);
2361 no_stats:
2362 hv_set_drvdata(dev, NULL);
2363 free_netdev(net);
2364 no_net:
2365 return ret;
2366 }
2367
2368 static int netvsc_remove(struct hv_device *dev)
2369 {
2370 struct net_device_context *ndev_ctx;
2371 struct net_device *vf_netdev, *net;
2372 struct netvsc_device *nvdev;
2373
2374 net = hv_get_drvdata(dev);
2375 if (net == NULL) {
2376 dev_err(&dev->device, "No net device to remove\n");
2377 return 0;
2378 }
2379
2380 ndev_ctx = netdev_priv(net);
2381
2382 cancel_delayed_work_sync(&ndev_ctx->dwork);
2383
2384 rtnl_lock();
2385 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2386 if (nvdev)
2387 cancel_work_sync(&nvdev->subchan_work);
2388
2389 /*
2390 * Call to the vsc driver to let it know that the device is being
2391 * removed. Also blocks mtu and channel changes.
2392 */
2393 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2394 if (vf_netdev)
2395 netvsc_unregister_vf(vf_netdev);
2396
2397 if (nvdev)
2398 rndis_filter_device_remove(dev, nvdev);
2399
2400 unregister_netdevice(net);
2401 list_del(&ndev_ctx->list);
2402
2403 rtnl_unlock();
2404
2405 hv_set_drvdata(dev, NULL);
2406
2407 free_percpu(ndev_ctx->vf_stats);
2408 free_netdev(net);
2409 return 0;
2410 }
2411
2412 static const struct hv_vmbus_device_id id_table[] = {
2413 /* Network guid */
2414 { HV_NIC_GUID, },
2415 { },
2416 };
2417
2418 MODULE_DEVICE_TABLE(vmbus, id_table);
2419
2420 /* The one and only one */
2421 static struct hv_driver netvsc_drv = {
2422 .name = KBUILD_MODNAME,
2423 .id_table = id_table,
2424 .probe = netvsc_probe,
2425 .remove = netvsc_remove,
2426 .driver = {
2427 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2428 },
2429 };
2430
2431 /*
2432 * On Hyper-V, every VF interface is matched with a corresponding
2433 * synthetic interface. The synthetic interface is presented first
2434 * to the guest. When the corresponding VF instance is registered,
2435 * we will take care of switching the data path.
2436 */
2437 static int netvsc_netdev_event(struct notifier_block *this,
2438 unsigned long event, void *ptr)
2439 {
2440 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2441
2442 /* Skip our own events */
2443 if (event_dev->netdev_ops == &device_ops)
2444 return NOTIFY_DONE;
2445
2446 /* Avoid non-Ethernet type devices */
2447 if (event_dev->type != ARPHRD_ETHER)
2448 return NOTIFY_DONE;
2449
2450 /* Avoid Vlan dev with same MAC registering as VF */
2451 if (is_vlan_dev(event_dev))
2452 return NOTIFY_DONE;
2453
2454 /* Avoid Bonding master dev with same MAC registering as VF */
2455 if ((event_dev->priv_flags & IFF_BONDING) &&
2456 (event_dev->flags & IFF_MASTER))
2457 return NOTIFY_DONE;
2458
2459 switch (event) {
2460 case NETDEV_REGISTER:
2461 return netvsc_register_vf(event_dev);
2462 case NETDEV_UNREGISTER:
2463 return netvsc_unregister_vf(event_dev);
2464 case NETDEV_UP:
2465 case NETDEV_DOWN:
2466 return netvsc_vf_changed(event_dev);
2467 default:
2468 return NOTIFY_DONE;
2469 }
2470 }
2471
2472 static struct notifier_block netvsc_netdev_notifier = {
2473 .notifier_call = netvsc_netdev_event,
2474 };
2475
2476 static void __exit netvsc_drv_exit(void)
2477 {
2478 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2479 vmbus_driver_unregister(&netvsc_drv);
2480 }
2481
2482 static int __init netvsc_drv_init(void)
2483 {
2484 int ret;
2485
2486 if (ring_size < RING_SIZE_MIN) {
2487 ring_size = RING_SIZE_MIN;
2488 pr_info("Increased ring_size to %u (min allowed)\n",
2489 ring_size);
2490 }
2491 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2492
2493 ret = vmbus_driver_register(&netvsc_drv);
2494 if (ret)
2495 return ret;
2496
2497 register_netdevice_notifier(&netvsc_netdev_notifier);
2498 return 0;
2499 }
2500
2501 MODULE_LICENSE("GPL");
2502 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2503
2504 module_init(netvsc_drv_init);
2505 module_exit(netvsc_drv_exit);
Linux with added FPC-III support