]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/net/hyperv/netvsc_drv.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
[mirror_ubuntu-jammy-kernel.git] / drivers / net / hyperv / netvsc_drv.c
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/module.h>
14 #include <linux/highmem.h>
15 #include <linux/device.h>
16 #include <linux/io.h>
17 #include <linux/delay.h>
18 #include <linux/netdevice.h>
19 #include <linux/inetdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/pci.h>
22 #include <linux/skbuff.h>
23 #include <linux/if_vlan.h>
24 #include <linux/in.h>
25 #include <linux/slab.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/netpoll.h>
28
29 #include <net/arp.h>
30 #include <net/route.h>
31 #include <net/sock.h>
32 #include <net/pkt_sched.h>
33 #include <net/checksum.h>
34 #include <net/ip6_checksum.h>
35
36 #include "hyperv_net.h"
37
38 #define RING_SIZE_MIN 64
39 #define RETRY_US_LO 5000
40 #define RETRY_US_HI 10000
41 #define RETRY_MAX 2000 /* >10 sec */
42
43 #define LINKCHANGE_INT (2 * HZ)
44 #define VF_TAKEOVER_INT (HZ / 10)
45
46 static unsigned int ring_size __ro_after_init = 128;
47 module_param(ring_size, uint, 0444);
48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
49 unsigned int netvsc_ring_bytes __ro_after_init;
50
51 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
52 NETIF_MSG_LINK | NETIF_MSG_IFUP |
53 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
54 NETIF_MSG_TX_ERR;
55
56 static int debug = -1;
57 module_param(debug, int, 0444);
58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59
60 static LIST_HEAD(netvsc_dev_list);
61
62 static void netvsc_change_rx_flags(struct net_device *net, int change)
63 {
64 struct net_device_context *ndev_ctx = netdev_priv(net);
65 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
66 int inc;
67
68 if (!vf_netdev)
69 return;
70
71 if (change & IFF_PROMISC) {
72 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
73 dev_set_promiscuity(vf_netdev, inc);
74 }
75
76 if (change & IFF_ALLMULTI) {
77 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
78 dev_set_allmulti(vf_netdev, inc);
79 }
80 }
81
82 static void netvsc_set_rx_mode(struct net_device *net)
83 {
84 struct net_device_context *ndev_ctx = netdev_priv(net);
85 struct net_device *vf_netdev;
86 struct netvsc_device *nvdev;
87
88 rcu_read_lock();
89 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
90 if (vf_netdev) {
91 dev_uc_sync(vf_netdev, net);
92 dev_mc_sync(vf_netdev, net);
93 }
94
95 nvdev = rcu_dereference(ndev_ctx->nvdev);
96 if (nvdev)
97 rndis_filter_update(nvdev);
98 rcu_read_unlock();
99 }
100
101 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
102 struct net_device *ndev)
103 {
104 nvscdev->tx_disable = false;
105 virt_wmb(); /* ensure queue wake up mechanism is on */
106
107 netif_tx_wake_all_queues(ndev);
108 }
109
110 static int netvsc_open(struct net_device *net)
111 {
112 struct net_device_context *ndev_ctx = netdev_priv(net);
113 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
114 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
115 struct rndis_device *rdev;
116 int ret = 0;
117
118 netif_carrier_off(net);
119
120 /* Open up the device */
121 ret = rndis_filter_open(nvdev);
122 if (ret != 0) {
123 netdev_err(net, "unable to open device (ret %d).\n", ret);
124 return ret;
125 }
126
127 rdev = nvdev->extension;
128 if (!rdev->link_state) {
129 netif_carrier_on(net);
130 netvsc_tx_enable(nvdev, net);
131 }
132
133 if (vf_netdev) {
134 /* Setting synthetic device up transparently sets
135 * slave as up. If open fails, then slave will be
136 * still be offline (and not used).
137 */
138 ret = dev_open(vf_netdev, NULL);
139 if (ret)
140 netdev_warn(net,
141 "unable to open slave: %s: %d\n",
142 vf_netdev->name, ret);
143 }
144 return 0;
145 }
146
147 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
148 {
149 unsigned int retry = 0;
150 int i;
151
152 /* Ensure pending bytes in ring are read */
153 for (;;) {
154 u32 aread = 0;
155
156 for (i = 0; i < nvdev->num_chn; i++) {
157 struct vmbus_channel *chn
158 = nvdev->chan_table[i].channel;
159
160 if (!chn)
161 continue;
162
163 /* make sure receive not running now */
164 napi_synchronize(&nvdev->chan_table[i].napi);
165
166 aread = hv_get_bytes_to_read(&chn->inbound);
167 if (aread)
168 break;
169
170 aread = hv_get_bytes_to_read(&chn->outbound);
171 if (aread)
172 break;
173 }
174
175 if (aread == 0)
176 return 0;
177
178 if (++retry > RETRY_MAX)
179 return -ETIMEDOUT;
180
181 usleep_range(RETRY_US_LO, RETRY_US_HI);
182 }
183 }
184
185 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
186 struct net_device *ndev)
187 {
188 if (nvscdev) {
189 nvscdev->tx_disable = true;
190 virt_wmb(); /* ensure txq will not wake up after stop */
191 }
192
193 netif_tx_disable(ndev);
194 }
195
196 static int netvsc_close(struct net_device *net)
197 {
198 struct net_device_context *net_device_ctx = netdev_priv(net);
199 struct net_device *vf_netdev
200 = rtnl_dereference(net_device_ctx->vf_netdev);
201 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
202 int ret;
203
204 netvsc_tx_disable(nvdev, net);
205
206 /* No need to close rndis filter if it is removed already */
207 if (!nvdev)
208 return 0;
209
210 ret = rndis_filter_close(nvdev);
211 if (ret != 0) {
212 netdev_err(net, "unable to close device (ret %d).\n", ret);
213 return ret;
214 }
215
216 ret = netvsc_wait_until_empty(nvdev);
217 if (ret)
218 netdev_err(net, "Ring buffer not empty after closing rndis\n");
219
220 if (vf_netdev)
221 dev_close(vf_netdev);
222
223 return ret;
224 }
225
226 static inline void *init_ppi_data(struct rndis_message *msg,
227 u32 ppi_size, u32 pkt_type)
228 {
229 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
230 struct rndis_per_packet_info *ppi;
231
232 rndis_pkt->data_offset += ppi_size;
233 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
234 + rndis_pkt->per_pkt_info_len;
235
236 ppi->size = ppi_size;
237 ppi->type = pkt_type;
238 ppi->internal = 0;
239 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
240
241 rndis_pkt->per_pkt_info_len += ppi_size;
242
243 return ppi + 1;
244 }
245
246 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
247 * packets. We can use ethtool to change UDP hash level when necessary.
248 */
249 static inline u32 netvsc_get_hash(
250 struct sk_buff *skb,
251 const struct net_device_context *ndc)
252 {
253 struct flow_keys flow;
254 u32 hash, pkt_proto = 0;
255 static u32 hashrnd __read_mostly;
256
257 net_get_random_once(&hashrnd, sizeof(hashrnd));
258
259 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
260 return 0;
261
262 switch (flow.basic.ip_proto) {
263 case IPPROTO_TCP:
264 if (flow.basic.n_proto == htons(ETH_P_IP))
265 pkt_proto = HV_TCP4_L4HASH;
266 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
267 pkt_proto = HV_TCP6_L4HASH;
268
269 break;
270
271 case IPPROTO_UDP:
272 if (flow.basic.n_proto == htons(ETH_P_IP))
273 pkt_proto = HV_UDP4_L4HASH;
274 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
275 pkt_proto = HV_UDP6_L4HASH;
276
277 break;
278 }
279
280 if (pkt_proto & ndc->l4_hash) {
281 return skb_get_hash(skb);
282 } else {
283 if (flow.basic.n_proto == htons(ETH_P_IP))
284 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
285 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
286 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
287 else
288 return 0;
289
290 __skb_set_sw_hash(skb, hash, false);
291 }
292
293 return hash;
294 }
295
296 static inline int netvsc_get_tx_queue(struct net_device *ndev,
297 struct sk_buff *skb, int old_idx)
298 {
299 const struct net_device_context *ndc = netdev_priv(ndev);
300 struct sock *sk = skb->sk;
301 int q_idx;
302
303 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
304 (VRSS_SEND_TAB_SIZE - 1)];
305
306 /* If queue index changed record the new value */
307 if (q_idx != old_idx &&
308 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
309 sk_tx_queue_set(sk, q_idx);
310
311 return q_idx;
312 }
313
314 /*
315 * Select queue for transmit.
316 *
317 * If a valid queue has already been assigned, then use that.
318 * Otherwise compute tx queue based on hash and the send table.
319 *
320 * This is basically similar to default (netdev_pick_tx) with the added step
321 * of using the host send_table when no other queue has been assigned.
322 *
323 * TODO support XPS - but get_xps_queue not exported
324 */
325 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
326 {
327 int q_idx = sk_tx_queue_get(skb->sk);
328
329 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
330 /* If forwarding a packet, we use the recorded queue when
331 * available for better cache locality.
332 */
333 if (skb_rx_queue_recorded(skb))
334 q_idx = skb_get_rx_queue(skb);
335 else
336 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
337 }
338
339 return q_idx;
340 }
341
342 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
343 struct net_device *sb_dev)
344 {
345 struct net_device_context *ndc = netdev_priv(ndev);
346 struct net_device *vf_netdev;
347 u16 txq;
348
349 rcu_read_lock();
350 vf_netdev = rcu_dereference(ndc->vf_netdev);
351 if (vf_netdev) {
352 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
353
354 if (vf_ops->ndo_select_queue)
355 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
356 else
357 txq = netdev_pick_tx(vf_netdev, skb, NULL);
358
359 /* Record the queue selected by VF so that it can be
360 * used for common case where VF has more queues than
361 * the synthetic device.
362 */
363 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
364 } else {
365 txq = netvsc_pick_tx(ndev, skb);
366 }
367 rcu_read_unlock();
368
369 while (unlikely(txq >= ndev->real_num_tx_queues))
370 txq -= ndev->real_num_tx_queues;
371
372 return txq;
373 }
374
375 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
376 struct hv_page_buffer *pb)
377 {
378 int j = 0;
379
380 /* Deal with compound pages by ignoring unused part
381 * of the page.
382 */
383 page += (offset >> PAGE_SHIFT);
384 offset &= ~PAGE_MASK;
385
386 while (len > 0) {
387 unsigned long bytes;
388
389 bytes = PAGE_SIZE - offset;
390 if (bytes > len)
391 bytes = len;
392 pb[j].pfn = page_to_pfn(page);
393 pb[j].offset = offset;
394 pb[j].len = bytes;
395
396 offset += bytes;
397 len -= bytes;
398
399 if (offset == PAGE_SIZE && len) {
400 page++;
401 offset = 0;
402 j++;
403 }
404 }
405
406 return j + 1;
407 }
408
409 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
410 struct hv_netvsc_packet *packet,
411 struct hv_page_buffer *pb)
412 {
413 u32 slots_used = 0;
414 char *data = skb->data;
415 int frags = skb_shinfo(skb)->nr_frags;
416 int i;
417
418 /* The packet is laid out thus:
419 * 1. hdr: RNDIS header and PPI
420 * 2. skb linear data
421 * 3. skb fragment data
422 */
423 slots_used += fill_pg_buf(virt_to_page(hdr),
424 offset_in_page(hdr),
425 len, &pb[slots_used]);
426
427 packet->rmsg_size = len;
428 packet->rmsg_pgcnt = slots_used;
429
430 slots_used += fill_pg_buf(virt_to_page(data),
431 offset_in_page(data),
432 skb_headlen(skb), &pb[slots_used]);
433
434 for (i = 0; i < frags; i++) {
435 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
436
437 slots_used += fill_pg_buf(skb_frag_page(frag),
438 skb_frag_off(frag),
439 skb_frag_size(frag), &pb[slots_used]);
440 }
441 return slots_used;
442 }
443
444 static int count_skb_frag_slots(struct sk_buff *skb)
445 {
446 int i, frags = skb_shinfo(skb)->nr_frags;
447 int pages = 0;
448
449 for (i = 0; i < frags; i++) {
450 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
451 unsigned long size = skb_frag_size(frag);
452 unsigned long offset = skb_frag_off(frag);
453
454 /* Skip unused frames from start of page */
455 offset &= ~PAGE_MASK;
456 pages += PFN_UP(offset + size);
457 }
458 return pages;
459 }
460
461 static int netvsc_get_slots(struct sk_buff *skb)
462 {
463 char *data = skb->data;
464 unsigned int offset = offset_in_page(data);
465 unsigned int len = skb_headlen(skb);
466 int slots;
467 int frag_slots;
468
469 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
470 frag_slots = count_skb_frag_slots(skb);
471 return slots + frag_slots;
472 }
473
474 static u32 net_checksum_info(struct sk_buff *skb)
475 {
476 if (skb->protocol == htons(ETH_P_IP)) {
477 struct iphdr *ip = ip_hdr(skb);
478
479 if (ip->protocol == IPPROTO_TCP)
480 return TRANSPORT_INFO_IPV4_TCP;
481 else if (ip->protocol == IPPROTO_UDP)
482 return TRANSPORT_INFO_IPV4_UDP;
483 } else {
484 struct ipv6hdr *ip6 = ipv6_hdr(skb);
485
486 if (ip6->nexthdr == IPPROTO_TCP)
487 return TRANSPORT_INFO_IPV6_TCP;
488 else if (ip6->nexthdr == IPPROTO_UDP)
489 return TRANSPORT_INFO_IPV6_UDP;
490 }
491
492 return TRANSPORT_INFO_NOT_IP;
493 }
494
495 /* Send skb on the slave VF device. */
496 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
497 struct sk_buff *skb)
498 {
499 struct net_device_context *ndev_ctx = netdev_priv(net);
500 unsigned int len = skb->len;
501 int rc;
502
503 skb->dev = vf_netdev;
504 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
505
506 rc = dev_queue_xmit(skb);
507 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
508 struct netvsc_vf_pcpu_stats *pcpu_stats
509 = this_cpu_ptr(ndev_ctx->vf_stats);
510
511 u64_stats_update_begin(&pcpu_stats->syncp);
512 pcpu_stats->tx_packets++;
513 pcpu_stats->tx_bytes += len;
514 u64_stats_update_end(&pcpu_stats->syncp);
515 } else {
516 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
517 }
518
519 return rc;
520 }
521
522 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
523 {
524 struct net_device_context *net_device_ctx = netdev_priv(net);
525 struct hv_netvsc_packet *packet = NULL;
526 int ret;
527 unsigned int num_data_pgs;
528 struct rndis_message *rndis_msg;
529 struct net_device *vf_netdev;
530 u32 rndis_msg_size;
531 u32 hash;
532 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
533
534 /* if VF is present and up then redirect packets
535 * already called with rcu_read_lock_bh
536 */
537 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
538 if (vf_netdev && netif_running(vf_netdev) &&
539 !netpoll_tx_running(net))
540 return netvsc_vf_xmit(net, vf_netdev, skb);
541
542 /* We will atmost need two pages to describe the rndis
543 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
544 * of pages in a single packet. If skb is scattered around
545 * more pages we try linearizing it.
546 */
547
548 num_data_pgs = netvsc_get_slots(skb) + 2;
549
550 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
551 ++net_device_ctx->eth_stats.tx_scattered;
552
553 if (skb_linearize(skb))
554 goto no_memory;
555
556 num_data_pgs = netvsc_get_slots(skb) + 2;
557 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
558 ++net_device_ctx->eth_stats.tx_too_big;
559 goto drop;
560 }
561 }
562
563 /*
564 * Place the rndis header in the skb head room and
565 * the skb->cb will be used for hv_netvsc_packet
566 * structure.
567 */
568 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
569 if (ret)
570 goto no_memory;
571
572 /* Use the skb control buffer for building up the packet */
573 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
574 FIELD_SIZEOF(struct sk_buff, cb));
575 packet = (struct hv_netvsc_packet *)skb->cb;
576
577 packet->q_idx = skb_get_queue_mapping(skb);
578
579 packet->total_data_buflen = skb->len;
580 packet->total_bytes = skb->len;
581 packet->total_packets = 1;
582
583 rndis_msg = (struct rndis_message *)skb->head;
584
585 /* Add the rndis header */
586 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
587 rndis_msg->msg_len = packet->total_data_buflen;
588
589 rndis_msg->msg.pkt = (struct rndis_packet) {
590 .data_offset = sizeof(struct rndis_packet),
591 .data_len = packet->total_data_buflen,
592 .per_pkt_info_offset = sizeof(struct rndis_packet),
593 };
594
595 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
596
597 hash = skb_get_hash_raw(skb);
598 if (hash != 0 && net->real_num_tx_queues > 1) {
599 u32 *hash_info;
600
601 rndis_msg_size += NDIS_HASH_PPI_SIZE;
602 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
603 NBL_HASH_VALUE);
604 *hash_info = hash;
605 }
606
607 if (skb_vlan_tag_present(skb)) {
608 struct ndis_pkt_8021q_info *vlan;
609
610 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
611 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
612 IEEE_8021Q_INFO);
613
614 vlan->value = 0;
615 vlan->vlanid = skb_vlan_tag_get_id(skb);
616 vlan->cfi = skb_vlan_tag_get_cfi(skb);
617 vlan->pri = skb_vlan_tag_get_prio(skb);
618 }
619
620 if (skb_is_gso(skb)) {
621 struct ndis_tcp_lso_info *lso_info;
622
623 rndis_msg_size += NDIS_LSO_PPI_SIZE;
624 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
625 TCP_LARGESEND_PKTINFO);
626
627 lso_info->value = 0;
628 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
629 if (skb->protocol == htons(ETH_P_IP)) {
630 lso_info->lso_v2_transmit.ip_version =
631 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
632 ip_hdr(skb)->tot_len = 0;
633 ip_hdr(skb)->check = 0;
634 tcp_hdr(skb)->check =
635 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
636 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
637 } else {
638 lso_info->lso_v2_transmit.ip_version =
639 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
640 ipv6_hdr(skb)->payload_len = 0;
641 tcp_hdr(skb)->check =
642 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
643 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
644 }
645 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
646 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
647 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
648 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
649 struct ndis_tcp_ip_checksum_info *csum_info;
650
651 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
652 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
653 TCPIP_CHKSUM_PKTINFO);
654
655 csum_info->value = 0;
656 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
657
658 if (skb->protocol == htons(ETH_P_IP)) {
659 csum_info->transmit.is_ipv4 = 1;
660
661 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
662 csum_info->transmit.tcp_checksum = 1;
663 else
664 csum_info->transmit.udp_checksum = 1;
665 } else {
666 csum_info->transmit.is_ipv6 = 1;
667
668 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
669 csum_info->transmit.tcp_checksum = 1;
670 else
671 csum_info->transmit.udp_checksum = 1;
672 }
673 } else {
674 /* Can't do offload of this type of checksum */
675 if (skb_checksum_help(skb))
676 goto drop;
677 }
678 }
679
680 /* Start filling in the page buffers with the rndis hdr */
681 rndis_msg->msg_len += rndis_msg_size;
682 packet->total_data_buflen = rndis_msg->msg_len;
683 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
684 skb, packet, pb);
685
686 /* timestamp packet in software */
687 skb_tx_timestamp(skb);
688
689 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
690 if (likely(ret == 0))
691 return NETDEV_TX_OK;
692
693 if (ret == -EAGAIN) {
694 ++net_device_ctx->eth_stats.tx_busy;
695 return NETDEV_TX_BUSY;
696 }
697
698 if (ret == -ENOSPC)
699 ++net_device_ctx->eth_stats.tx_no_space;
700
701 drop:
702 dev_kfree_skb_any(skb);
703 net->stats.tx_dropped++;
704
705 return NETDEV_TX_OK;
706
707 no_memory:
708 ++net_device_ctx->eth_stats.tx_no_memory;
709 goto drop;
710 }
711
712 /*
713 * netvsc_linkstatus_callback - Link up/down notification
714 */
715 void netvsc_linkstatus_callback(struct net_device *net,
716 struct rndis_message *resp)
717 {
718 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
719 struct net_device_context *ndev_ctx = netdev_priv(net);
720 struct netvsc_reconfig *event;
721 unsigned long flags;
722
723 /* Update the physical link speed when changing to another vSwitch */
724 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
725 u32 speed;
726
727 speed = *(u32 *)((void *)indicate
728 + indicate->status_buf_offset) / 10000;
729 ndev_ctx->speed = speed;
730 return;
731 }
732
733 /* Handle these link change statuses below */
734 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
735 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
736 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
737 return;
738
739 if (net->reg_state != NETREG_REGISTERED)
740 return;
741
742 event = kzalloc(sizeof(*event), GFP_ATOMIC);
743 if (!event)
744 return;
745 event->event = indicate->status;
746
747 spin_lock_irqsave(&ndev_ctx->lock, flags);
748 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
749 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
750
751 schedule_delayed_work(&ndev_ctx->dwork, 0);
752 }
753
754 static void netvsc_comp_ipcsum(struct sk_buff *skb)
755 {
756 struct iphdr *iph = (struct iphdr *)skb->data;
757
758 iph->check = 0;
759 iph->check = ip_fast_csum(iph, iph->ihl);
760 }
761
762 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
763 struct netvsc_channel *nvchan)
764 {
765 struct napi_struct *napi = &nvchan->napi;
766 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
767 const struct ndis_tcp_ip_checksum_info *csum_info =
768 nvchan->rsc.csum_info;
769 const u32 *hash_info = nvchan->rsc.hash_info;
770 struct sk_buff *skb;
771 int i;
772
773 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
774 if (!skb)
775 return skb;
776
777 /*
778 * Copy to skb. This copy is needed here since the memory pointed by
779 * hv_netvsc_packet cannot be deallocated
780 */
781 for (i = 0; i < nvchan->rsc.cnt; i++)
782 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
783
784 skb->protocol = eth_type_trans(skb, net);
785
786 /* skb is already created with CHECKSUM_NONE */
787 skb_checksum_none_assert(skb);
788
789 /* Incoming packets may have IP header checksum verified by the host.
790 * They may not have IP header checksum computed after coalescing.
791 * We compute it here if the flags are set, because on Linux, the IP
792 * checksum is always checked.
793 */
794 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
795 csum_info->receive.ip_checksum_succeeded &&
796 skb->protocol == htons(ETH_P_IP))
797 netvsc_comp_ipcsum(skb);
798
799 /* Do L4 checksum offload if enabled and present. */
800 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
801 if (csum_info->receive.tcp_checksum_succeeded ||
802 csum_info->receive.udp_checksum_succeeded)
803 skb->ip_summed = CHECKSUM_UNNECESSARY;
804 }
805
806 if (hash_info && (net->features & NETIF_F_RXHASH))
807 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
808
809 if (vlan) {
810 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
811 (vlan->cfi ? VLAN_CFI_MASK : 0);
812
813 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
814 vlan_tci);
815 }
816
817 return skb;
818 }
819
820 /*
821 * netvsc_recv_callback - Callback when we receive a packet from the
822 * "wire" on the specified device.
823 */
824 int netvsc_recv_callback(struct net_device *net,
825 struct netvsc_device *net_device,
826 struct netvsc_channel *nvchan)
827 {
828 struct net_device_context *net_device_ctx = netdev_priv(net);
829 struct vmbus_channel *channel = nvchan->channel;
830 u16 q_idx = channel->offermsg.offer.sub_channel_index;
831 struct sk_buff *skb;
832 struct netvsc_stats *rx_stats;
833
834 if (net->reg_state != NETREG_REGISTERED)
835 return NVSP_STAT_FAIL;
836
837 /* Allocate a skb - TODO direct I/O to pages? */
838 skb = netvsc_alloc_recv_skb(net, nvchan);
839
840 if (unlikely(!skb)) {
841 ++net_device_ctx->eth_stats.rx_no_memory;
842 return NVSP_STAT_FAIL;
843 }
844
845 skb_record_rx_queue(skb, q_idx);
846
847 /*
848 * Even if injecting the packet, record the statistics
849 * on the synthetic device because modifying the VF device
850 * statistics will not work correctly.
851 */
852 rx_stats = &nvchan->rx_stats;
853 u64_stats_update_begin(&rx_stats->syncp);
854 rx_stats->packets++;
855 rx_stats->bytes += nvchan->rsc.pktlen;
856
857 if (skb->pkt_type == PACKET_BROADCAST)
858 ++rx_stats->broadcast;
859 else if (skb->pkt_type == PACKET_MULTICAST)
860 ++rx_stats->multicast;
861 u64_stats_update_end(&rx_stats->syncp);
862
863 napi_gro_receive(&nvchan->napi, skb);
864 return NVSP_STAT_SUCCESS;
865 }
866
867 static void netvsc_get_drvinfo(struct net_device *net,
868 struct ethtool_drvinfo *info)
869 {
870 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
871 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
872 }
873
874 static void netvsc_get_channels(struct net_device *net,
875 struct ethtool_channels *channel)
876 {
877 struct net_device_context *net_device_ctx = netdev_priv(net);
878 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
879
880 if (nvdev) {
881 channel->max_combined = nvdev->max_chn;
882 channel->combined_count = nvdev->num_chn;
883 }
884 }
885
886 /* Alloc struct netvsc_device_info, and initialize it from either existing
887 * struct netvsc_device, or from default values.
888 */
889 static struct netvsc_device_info *netvsc_devinfo_get
890 (struct netvsc_device *nvdev)
891 {
892 struct netvsc_device_info *dev_info;
893
894 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
895
896 if (!dev_info)
897 return NULL;
898
899 if (nvdev) {
900 dev_info->num_chn = nvdev->num_chn;
901 dev_info->send_sections = nvdev->send_section_cnt;
902 dev_info->send_section_size = nvdev->send_section_size;
903 dev_info->recv_sections = nvdev->recv_section_cnt;
904 dev_info->recv_section_size = nvdev->recv_section_size;
905
906 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
907 NETVSC_HASH_KEYLEN);
908 } else {
909 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
910 dev_info->send_sections = NETVSC_DEFAULT_TX;
911 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
912 dev_info->recv_sections = NETVSC_DEFAULT_RX;
913 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
914 }
915
916 return dev_info;
917 }
918
919 static int netvsc_detach(struct net_device *ndev,
920 struct netvsc_device *nvdev)
921 {
922 struct net_device_context *ndev_ctx = netdev_priv(ndev);
923 struct hv_device *hdev = ndev_ctx->device_ctx;
924 int ret;
925
926 /* Don't try continuing to try and setup sub channels */
927 if (cancel_work_sync(&nvdev->subchan_work))
928 nvdev->num_chn = 1;
929
930 /* If device was up (receiving) then shutdown */
931 if (netif_running(ndev)) {
932 netvsc_tx_disable(nvdev, ndev);
933
934 ret = rndis_filter_close(nvdev);
935 if (ret) {
936 netdev_err(ndev,
937 "unable to close device (ret %d).\n", ret);
938 return ret;
939 }
940
941 ret = netvsc_wait_until_empty(nvdev);
942 if (ret) {
943 netdev_err(ndev,
944 "Ring buffer not empty after closing rndis\n");
945 return ret;
946 }
947 }
948
949 netif_device_detach(ndev);
950
951 rndis_filter_device_remove(hdev, nvdev);
952
953 return 0;
954 }
955
956 static int netvsc_attach(struct net_device *ndev,
957 struct netvsc_device_info *dev_info)
958 {
959 struct net_device_context *ndev_ctx = netdev_priv(ndev);
960 struct hv_device *hdev = ndev_ctx->device_ctx;
961 struct netvsc_device *nvdev;
962 struct rndis_device *rdev;
963 int ret;
964
965 nvdev = rndis_filter_device_add(hdev, dev_info);
966 if (IS_ERR(nvdev))
967 return PTR_ERR(nvdev);
968
969 if (nvdev->num_chn > 1) {
970 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
971
972 /* if unavailable, just proceed with one queue */
973 if (ret) {
974 nvdev->max_chn = 1;
975 nvdev->num_chn = 1;
976 }
977 }
978
979 /* In any case device is now ready */
980 netif_device_attach(ndev);
981
982 /* Note: enable and attach happen when sub-channels setup */
983 netif_carrier_off(ndev);
984
985 if (netif_running(ndev)) {
986 ret = rndis_filter_open(nvdev);
987 if (ret)
988 goto err;
989
990 rdev = nvdev->extension;
991 if (!rdev->link_state)
992 netif_carrier_on(ndev);
993 }
994
995 return 0;
996
997 err:
998 netif_device_detach(ndev);
999
1000 rndis_filter_device_remove(hdev, nvdev);
1001
1002 return ret;
1003 }
1004
1005 static int netvsc_set_channels(struct net_device *net,
1006 struct ethtool_channels *channels)
1007 {
1008 struct net_device_context *net_device_ctx = netdev_priv(net);
1009 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1010 unsigned int orig, count = channels->combined_count;
1011 struct netvsc_device_info *device_info;
1012 int ret;
1013
1014 /* We do not support separate count for rx, tx, or other */
1015 if (count == 0 ||
1016 channels->rx_count || channels->tx_count || channels->other_count)
1017 return -EINVAL;
1018
1019 if (!nvdev || nvdev->destroy)
1020 return -ENODEV;
1021
1022 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1023 return -EINVAL;
1024
1025 if (count > nvdev->max_chn)
1026 return -EINVAL;
1027
1028 orig = nvdev->num_chn;
1029
1030 device_info = netvsc_devinfo_get(nvdev);
1031
1032 if (!device_info)
1033 return -ENOMEM;
1034
1035 device_info->num_chn = count;
1036
1037 ret = netvsc_detach(net, nvdev);
1038 if (ret)
1039 goto out;
1040
1041 ret = netvsc_attach(net, device_info);
1042 if (ret) {
1043 device_info->num_chn = orig;
1044 if (netvsc_attach(net, device_info))
1045 netdev_err(net, "restoring channel setting failed\n");
1046 }
1047
1048 out:
1049 kfree(device_info);
1050 return ret;
1051 }
1052
1053 static bool
1054 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1055 {
1056 struct ethtool_link_ksettings diff1 = *cmd;
1057 struct ethtool_link_ksettings diff2 = {};
1058
1059 diff1.base.speed = 0;
1060 diff1.base.duplex = 0;
1061 /* advertising and cmd are usually set */
1062 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1063 diff1.base.cmd = 0;
1064 /* We set port to PORT_OTHER */
1065 diff2.base.port = PORT_OTHER;
1066
1067 return !memcmp(&diff1, &diff2, sizeof(diff1));
1068 }
1069
1070 static void netvsc_init_settings(struct net_device *dev)
1071 {
1072 struct net_device_context *ndc = netdev_priv(dev);
1073
1074 ndc->l4_hash = HV_DEFAULT_L4HASH;
1075
1076 ndc->speed = SPEED_UNKNOWN;
1077 ndc->duplex = DUPLEX_FULL;
1078
1079 dev->features = NETIF_F_LRO;
1080 }
1081
1082 static int netvsc_get_link_ksettings(struct net_device *dev,
1083 struct ethtool_link_ksettings *cmd)
1084 {
1085 struct net_device_context *ndc = netdev_priv(dev);
1086
1087 cmd->base.speed = ndc->speed;
1088 cmd->base.duplex = ndc->duplex;
1089 cmd->base.port = PORT_OTHER;
1090
1091 return 0;
1092 }
1093
1094 static int netvsc_set_link_ksettings(struct net_device *dev,
1095 const struct ethtool_link_ksettings *cmd)
1096 {
1097 struct net_device_context *ndc = netdev_priv(dev);
1098 u32 speed;
1099
1100 speed = cmd->base.speed;
1101 if (!ethtool_validate_speed(speed) ||
1102 !ethtool_validate_duplex(cmd->base.duplex) ||
1103 !netvsc_validate_ethtool_ss_cmd(cmd))
1104 return -EINVAL;
1105
1106 ndc->speed = speed;
1107 ndc->duplex = cmd->base.duplex;
1108
1109 return 0;
1110 }
1111
1112 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1113 {
1114 struct net_device_context *ndevctx = netdev_priv(ndev);
1115 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1116 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1117 int orig_mtu = ndev->mtu;
1118 struct netvsc_device_info *device_info;
1119 int ret = 0;
1120
1121 if (!nvdev || nvdev->destroy)
1122 return -ENODEV;
1123
1124 device_info = netvsc_devinfo_get(nvdev);
1125
1126 if (!device_info)
1127 return -ENOMEM;
1128
1129 /* Change MTU of underlying VF netdev first. */
1130 if (vf_netdev) {
1131 ret = dev_set_mtu(vf_netdev, mtu);
1132 if (ret)
1133 goto out;
1134 }
1135
1136 ret = netvsc_detach(ndev, nvdev);
1137 if (ret)
1138 goto rollback_vf;
1139
1140 ndev->mtu = mtu;
1141
1142 ret = netvsc_attach(ndev, device_info);
1143 if (!ret)
1144 goto out;
1145
1146 /* Attempt rollback to original MTU */
1147 ndev->mtu = orig_mtu;
1148
1149 if (netvsc_attach(ndev, device_info))
1150 netdev_err(ndev, "restoring mtu failed\n");
1151 rollback_vf:
1152 if (vf_netdev)
1153 dev_set_mtu(vf_netdev, orig_mtu);
1154
1155 out:
1156 kfree(device_info);
1157 return ret;
1158 }
1159
1160 static void netvsc_get_vf_stats(struct net_device *net,
1161 struct netvsc_vf_pcpu_stats *tot)
1162 {
1163 struct net_device_context *ndev_ctx = netdev_priv(net);
1164 int i;
1165
1166 memset(tot, 0, sizeof(*tot));
1167
1168 for_each_possible_cpu(i) {
1169 const struct netvsc_vf_pcpu_stats *stats
1170 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1171 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1172 unsigned int start;
1173
1174 do {
1175 start = u64_stats_fetch_begin_irq(&stats->syncp);
1176 rx_packets = stats->rx_packets;
1177 tx_packets = stats->tx_packets;
1178 rx_bytes = stats->rx_bytes;
1179 tx_bytes = stats->tx_bytes;
1180 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1181
1182 tot->rx_packets += rx_packets;
1183 tot->tx_packets += tx_packets;
1184 tot->rx_bytes += rx_bytes;
1185 tot->tx_bytes += tx_bytes;
1186 tot->tx_dropped += stats->tx_dropped;
1187 }
1188 }
1189
1190 static void netvsc_get_pcpu_stats(struct net_device *net,
1191 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1192 {
1193 struct net_device_context *ndev_ctx = netdev_priv(net);
1194 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1195 int i;
1196
1197 /* fetch percpu stats of vf */
1198 for_each_possible_cpu(i) {
1199 const struct netvsc_vf_pcpu_stats *stats =
1200 per_cpu_ptr(ndev_ctx->vf_stats, i);
1201 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1202 unsigned int start;
1203
1204 do {
1205 start = u64_stats_fetch_begin_irq(&stats->syncp);
1206 this_tot->vf_rx_packets = stats->rx_packets;
1207 this_tot->vf_tx_packets = stats->tx_packets;
1208 this_tot->vf_rx_bytes = stats->rx_bytes;
1209 this_tot->vf_tx_bytes = stats->tx_bytes;
1210 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1211 this_tot->rx_packets = this_tot->vf_rx_packets;
1212 this_tot->tx_packets = this_tot->vf_tx_packets;
1213 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1214 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1215 }
1216
1217 /* fetch percpu stats of netvsc */
1218 for (i = 0; i < nvdev->num_chn; i++) {
1219 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1220 const struct netvsc_stats *stats;
1221 struct netvsc_ethtool_pcpu_stats *this_tot =
1222 &pcpu_tot[nvchan->channel->target_cpu];
1223 u64 packets, bytes;
1224 unsigned int start;
1225
1226 stats = &nvchan->tx_stats;
1227 do {
1228 start = u64_stats_fetch_begin_irq(&stats->syncp);
1229 packets = stats->packets;
1230 bytes = stats->bytes;
1231 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1232
1233 this_tot->tx_bytes += bytes;
1234 this_tot->tx_packets += packets;
1235
1236 stats = &nvchan->rx_stats;
1237 do {
1238 start = u64_stats_fetch_begin_irq(&stats->syncp);
1239 packets = stats->packets;
1240 bytes = stats->bytes;
1241 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1242
1243 this_tot->rx_bytes += bytes;
1244 this_tot->rx_packets += packets;
1245 }
1246 }
1247
1248 static void netvsc_get_stats64(struct net_device *net,
1249 struct rtnl_link_stats64 *t)
1250 {
1251 struct net_device_context *ndev_ctx = netdev_priv(net);
1252 struct netvsc_device *nvdev;
1253 struct netvsc_vf_pcpu_stats vf_tot;
1254 int i;
1255
1256 rcu_read_lock();
1257
1258 nvdev = rcu_dereference(ndev_ctx->nvdev);
1259 if (!nvdev)
1260 goto out;
1261
1262 netdev_stats_to_stats64(t, &net->stats);
1263
1264 netvsc_get_vf_stats(net, &vf_tot);
1265 t->rx_packets += vf_tot.rx_packets;
1266 t->tx_packets += vf_tot.tx_packets;
1267 t->rx_bytes += vf_tot.rx_bytes;
1268 t->tx_bytes += vf_tot.tx_bytes;
1269 t->tx_dropped += vf_tot.tx_dropped;
1270
1271 for (i = 0; i < nvdev->num_chn; i++) {
1272 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1273 const struct netvsc_stats *stats;
1274 u64 packets, bytes, multicast;
1275 unsigned int start;
1276
1277 stats = &nvchan->tx_stats;
1278 do {
1279 start = u64_stats_fetch_begin_irq(&stats->syncp);
1280 packets = stats->packets;
1281 bytes = stats->bytes;
1282 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1283
1284 t->tx_bytes += bytes;
1285 t->tx_packets += packets;
1286
1287 stats = &nvchan->rx_stats;
1288 do {
1289 start = u64_stats_fetch_begin_irq(&stats->syncp);
1290 packets = stats->packets;
1291 bytes = stats->bytes;
1292 multicast = stats->multicast + stats->broadcast;
1293 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1294
1295 t->rx_bytes += bytes;
1296 t->rx_packets += packets;
1297 t->multicast += multicast;
1298 }
1299 out:
1300 rcu_read_unlock();
1301 }
1302
1303 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1304 {
1305 struct net_device_context *ndc = netdev_priv(ndev);
1306 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1307 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1308 struct sockaddr *addr = p;
1309 int err;
1310
1311 err = eth_prepare_mac_addr_change(ndev, p);
1312 if (err)
1313 return err;
1314
1315 if (!nvdev)
1316 return -ENODEV;
1317
1318 if (vf_netdev) {
1319 err = dev_set_mac_address(vf_netdev, addr, NULL);
1320 if (err)
1321 return err;
1322 }
1323
1324 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1325 if (!err) {
1326 eth_commit_mac_addr_change(ndev, p);
1327 } else if (vf_netdev) {
1328 /* rollback change on VF */
1329 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1330 dev_set_mac_address(vf_netdev, addr, NULL);
1331 }
1332
1333 return err;
1334 }
1335
1336 static const struct {
1337 char name[ETH_GSTRING_LEN];
1338 u16 offset;
1339 } netvsc_stats[] = {
1340 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1341 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1342 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1343 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1344 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1345 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1346 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1347 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1348 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1349 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1350 }, pcpu_stats[] = {
1351 { "cpu%u_rx_packets",
1352 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1353 { "cpu%u_rx_bytes",
1354 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1355 { "cpu%u_tx_packets",
1356 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1357 { "cpu%u_tx_bytes",
1358 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1359 { "cpu%u_vf_rx_packets",
1360 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1361 { "cpu%u_vf_rx_bytes",
1362 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1363 { "cpu%u_vf_tx_packets",
1364 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1365 { "cpu%u_vf_tx_bytes",
1366 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1367 }, vf_stats[] = {
1368 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1369 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1370 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1371 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1372 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1373 };
1374
1375 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1376 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1377
1378 /* statistics per queue (rx/tx packets/bytes) */
1379 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1380
1381 /* 4 statistics per queue (rx/tx packets/bytes) */
1382 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1383
1384 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1385 {
1386 struct net_device_context *ndc = netdev_priv(dev);
1387 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1388
1389 if (!nvdev)
1390 return -ENODEV;
1391
1392 switch (string_set) {
1393 case ETH_SS_STATS:
1394 return NETVSC_GLOBAL_STATS_LEN
1395 + NETVSC_VF_STATS_LEN
1396 + NETVSC_QUEUE_STATS_LEN(nvdev)
1397 + NETVSC_PCPU_STATS_LEN;
1398 default:
1399 return -EINVAL;
1400 }
1401 }
1402
1403 static void netvsc_get_ethtool_stats(struct net_device *dev,
1404 struct ethtool_stats *stats, u64 *data)
1405 {
1406 struct net_device_context *ndc = netdev_priv(dev);
1407 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1408 const void *nds = &ndc->eth_stats;
1409 const struct netvsc_stats *qstats;
1410 struct netvsc_vf_pcpu_stats sum;
1411 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1412 unsigned int start;
1413 u64 packets, bytes;
1414 int i, j, cpu;
1415
1416 if (!nvdev)
1417 return;
1418
1419 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1420 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1421
1422 netvsc_get_vf_stats(dev, &sum);
1423 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1424 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1425
1426 for (j = 0; j < nvdev->num_chn; j++) {
1427 qstats = &nvdev->chan_table[j].tx_stats;
1428
1429 do {
1430 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1431 packets = qstats->packets;
1432 bytes = qstats->bytes;
1433 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1434 data[i++] = packets;
1435 data[i++] = bytes;
1436
1437 qstats = &nvdev->chan_table[j].rx_stats;
1438 do {
1439 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1440 packets = qstats->packets;
1441 bytes = qstats->bytes;
1442 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1443 data[i++] = packets;
1444 data[i++] = bytes;
1445 }
1446
1447 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1448 sizeof(struct netvsc_ethtool_pcpu_stats),
1449 GFP_KERNEL);
1450 netvsc_get_pcpu_stats(dev, pcpu_sum);
1451 for_each_present_cpu(cpu) {
1452 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1453
1454 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1455 data[i++] = *(u64 *)((void *)this_sum
1456 + pcpu_stats[j].offset);
1457 }
1458 kvfree(pcpu_sum);
1459 }
1460
1461 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1462 {
1463 struct net_device_context *ndc = netdev_priv(dev);
1464 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1465 u8 *p = data;
1466 int i, cpu;
1467
1468 if (!nvdev)
1469 return;
1470
1471 switch (stringset) {
1472 case ETH_SS_STATS:
1473 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1474 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1475 p += ETH_GSTRING_LEN;
1476 }
1477
1478 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1479 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1480 p += ETH_GSTRING_LEN;
1481 }
1482
1483 for (i = 0; i < nvdev->num_chn; i++) {
1484 sprintf(p, "tx_queue_%u_packets", i);
1485 p += ETH_GSTRING_LEN;
1486 sprintf(p, "tx_queue_%u_bytes", i);
1487 p += ETH_GSTRING_LEN;
1488 sprintf(p, "rx_queue_%u_packets", i);
1489 p += ETH_GSTRING_LEN;
1490 sprintf(p, "rx_queue_%u_bytes", i);
1491 p += ETH_GSTRING_LEN;
1492 }
1493
1494 for_each_present_cpu(cpu) {
1495 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1496 sprintf(p, pcpu_stats[i].name, cpu);
1497 p += ETH_GSTRING_LEN;
1498 }
1499 }
1500
1501 break;
1502 }
1503 }
1504
1505 static int
1506 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1507 struct ethtool_rxnfc *info)
1508 {
1509 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1510
1511 info->data = RXH_IP_SRC | RXH_IP_DST;
1512
1513 switch (info->flow_type) {
1514 case TCP_V4_FLOW:
1515 if (ndc->l4_hash & HV_TCP4_L4HASH)
1516 info->data |= l4_flag;
1517
1518 break;
1519
1520 case TCP_V6_FLOW:
1521 if (ndc->l4_hash & HV_TCP6_L4HASH)
1522 info->data |= l4_flag;
1523
1524 break;
1525
1526 case UDP_V4_FLOW:
1527 if (ndc->l4_hash & HV_UDP4_L4HASH)
1528 info->data |= l4_flag;
1529
1530 break;
1531
1532 case UDP_V6_FLOW:
1533 if (ndc->l4_hash & HV_UDP6_L4HASH)
1534 info->data |= l4_flag;
1535
1536 break;
1537
1538 case IPV4_FLOW:
1539 case IPV6_FLOW:
1540 break;
1541 default:
1542 info->data = 0;
1543 break;
1544 }
1545
1546 return 0;
1547 }
1548
1549 static int
1550 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1551 u32 *rules)
1552 {
1553 struct net_device_context *ndc = netdev_priv(dev);
1554 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1555
1556 if (!nvdev)
1557 return -ENODEV;
1558
1559 switch (info->cmd) {
1560 case ETHTOOL_GRXRINGS:
1561 info->data = nvdev->num_chn;
1562 return 0;
1563
1564 case ETHTOOL_GRXFH:
1565 return netvsc_get_rss_hash_opts(ndc, info);
1566 }
1567 return -EOPNOTSUPP;
1568 }
1569
1570 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1571 struct ethtool_rxnfc *info)
1572 {
1573 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1574 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1575 switch (info->flow_type) {
1576 case TCP_V4_FLOW:
1577 ndc->l4_hash |= HV_TCP4_L4HASH;
1578 break;
1579
1580 case TCP_V6_FLOW:
1581 ndc->l4_hash |= HV_TCP6_L4HASH;
1582 break;
1583
1584 case UDP_V4_FLOW:
1585 ndc->l4_hash |= HV_UDP4_L4HASH;
1586 break;
1587
1588 case UDP_V6_FLOW:
1589 ndc->l4_hash |= HV_UDP6_L4HASH;
1590 break;
1591
1592 default:
1593 return -EOPNOTSUPP;
1594 }
1595
1596 return 0;
1597 }
1598
1599 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1600 switch (info->flow_type) {
1601 case TCP_V4_FLOW:
1602 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1603 break;
1604
1605 case TCP_V6_FLOW:
1606 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1607 break;
1608
1609 case UDP_V4_FLOW:
1610 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1611 break;
1612
1613 case UDP_V6_FLOW:
1614 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1615 break;
1616
1617 default:
1618 return -EOPNOTSUPP;
1619 }
1620
1621 return 0;
1622 }
1623
1624 return -EOPNOTSUPP;
1625 }
1626
1627 static int
1628 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1629 {
1630 struct net_device_context *ndc = netdev_priv(ndev);
1631
1632 if (info->cmd == ETHTOOL_SRXFH)
1633 return netvsc_set_rss_hash_opts(ndc, info);
1634
1635 return -EOPNOTSUPP;
1636 }
1637
1638 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1639 {
1640 return NETVSC_HASH_KEYLEN;
1641 }
1642
1643 static u32 netvsc_rss_indir_size(struct net_device *dev)
1644 {
1645 return ITAB_NUM;
1646 }
1647
1648 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1649 u8 *hfunc)
1650 {
1651 struct net_device_context *ndc = netdev_priv(dev);
1652 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1653 struct rndis_device *rndis_dev;
1654 int i;
1655
1656 if (!ndev)
1657 return -ENODEV;
1658
1659 if (hfunc)
1660 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1661
1662 rndis_dev = ndev->extension;
1663 if (indir) {
1664 for (i = 0; i < ITAB_NUM; i++)
1665 indir[i] = rndis_dev->rx_table[i];
1666 }
1667
1668 if (key)
1669 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1670
1671 return 0;
1672 }
1673
1674 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1675 const u8 *key, const 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 != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1686 return -EOPNOTSUPP;
1687
1688 rndis_dev = ndev->extension;
1689 if (indir) {
1690 for (i = 0; i < ITAB_NUM; i++)
1691 if (indir[i] >= ndev->num_chn)
1692 return -EINVAL;
1693
1694 for (i = 0; i < ITAB_NUM; i++)
1695 rndis_dev->rx_table[i] = indir[i];
1696 }
1697
1698 if (!key) {
1699 if (!indir)
1700 return 0;
1701
1702 key = rndis_dev->rss_key;
1703 }
1704
1705 return rndis_filter_set_rss_param(rndis_dev, key);
1706 }
1707
1708 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1709 * It does have pre-allocated receive area which is divided into sections.
1710 */
1711 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1712 struct ethtool_ringparam *ring)
1713 {
1714 u32 max_buf_size;
1715
1716 ring->rx_pending = nvdev->recv_section_cnt;
1717 ring->tx_pending = nvdev->send_section_cnt;
1718
1719 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1720 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1721 else
1722 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1723
1724 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1725 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1726 / nvdev->send_section_size;
1727 }
1728
1729 static void netvsc_get_ringparam(struct net_device *ndev,
1730 struct ethtool_ringparam *ring)
1731 {
1732 struct net_device_context *ndevctx = netdev_priv(ndev);
1733 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1734
1735 if (!nvdev)
1736 return;
1737
1738 __netvsc_get_ringparam(nvdev, ring);
1739 }
1740
1741 static int netvsc_set_ringparam(struct net_device *ndev,
1742 struct ethtool_ringparam *ring)
1743 {
1744 struct net_device_context *ndevctx = netdev_priv(ndev);
1745 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1746 struct netvsc_device_info *device_info;
1747 struct ethtool_ringparam orig;
1748 u32 new_tx, new_rx;
1749 int ret = 0;
1750
1751 if (!nvdev || nvdev->destroy)
1752 return -ENODEV;
1753
1754 memset(&orig, 0, sizeof(orig));
1755 __netvsc_get_ringparam(nvdev, &orig);
1756
1757 new_tx = clamp_t(u32, ring->tx_pending,
1758 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1759 new_rx = clamp_t(u32, ring->rx_pending,
1760 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1761
1762 if (new_tx == orig.tx_pending &&
1763 new_rx == orig.rx_pending)
1764 return 0; /* no change */
1765
1766 device_info = netvsc_devinfo_get(nvdev);
1767
1768 if (!device_info)
1769 return -ENOMEM;
1770
1771 device_info->send_sections = new_tx;
1772 device_info->recv_sections = new_rx;
1773
1774 ret = netvsc_detach(ndev, nvdev);
1775 if (ret)
1776 goto out;
1777
1778 ret = netvsc_attach(ndev, device_info);
1779 if (ret) {
1780 device_info->send_sections = orig.tx_pending;
1781 device_info->recv_sections = orig.rx_pending;
1782
1783 if (netvsc_attach(ndev, device_info))
1784 netdev_err(ndev, "restoring ringparam failed");
1785 }
1786
1787 out:
1788 kfree(device_info);
1789 return ret;
1790 }
1791
1792 static int netvsc_set_features(struct net_device *ndev,
1793 netdev_features_t features)
1794 {
1795 netdev_features_t change = features ^ ndev->features;
1796 struct net_device_context *ndevctx = netdev_priv(ndev);
1797 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1798 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1799 struct ndis_offload_params offloads;
1800 int ret = 0;
1801
1802 if (!nvdev || nvdev->destroy)
1803 return -ENODEV;
1804
1805 if (!(change & NETIF_F_LRO))
1806 goto syncvf;
1807
1808 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1809
1810 if (features & NETIF_F_LRO) {
1811 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1812 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1813 } else {
1814 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1815 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1816 }
1817
1818 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1819
1820 if (ret) {
1821 features ^= NETIF_F_LRO;
1822 ndev->features = features;
1823 }
1824
1825 syncvf:
1826 if (!vf_netdev)
1827 return ret;
1828
1829 vf_netdev->wanted_features = features;
1830 netdev_update_features(vf_netdev);
1831
1832 return ret;
1833 }
1834
1835 static u32 netvsc_get_msglevel(struct net_device *ndev)
1836 {
1837 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1838
1839 return ndev_ctx->msg_enable;
1840 }
1841
1842 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1843 {
1844 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1845
1846 ndev_ctx->msg_enable = val;
1847 }
1848
1849 static const struct ethtool_ops ethtool_ops = {
1850 .get_drvinfo = netvsc_get_drvinfo,
1851 .get_msglevel = netvsc_get_msglevel,
1852 .set_msglevel = netvsc_set_msglevel,
1853 .get_link = ethtool_op_get_link,
1854 .get_ethtool_stats = netvsc_get_ethtool_stats,
1855 .get_sset_count = netvsc_get_sset_count,
1856 .get_strings = netvsc_get_strings,
1857 .get_channels = netvsc_get_channels,
1858 .set_channels = netvsc_set_channels,
1859 .get_ts_info = ethtool_op_get_ts_info,
1860 .get_rxnfc = netvsc_get_rxnfc,
1861 .set_rxnfc = netvsc_set_rxnfc,
1862 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1863 .get_rxfh_indir_size = netvsc_rss_indir_size,
1864 .get_rxfh = netvsc_get_rxfh,
1865 .set_rxfh = netvsc_set_rxfh,
1866 .get_link_ksettings = netvsc_get_link_ksettings,
1867 .set_link_ksettings = netvsc_set_link_ksettings,
1868 .get_ringparam = netvsc_get_ringparam,
1869 .set_ringparam = netvsc_set_ringparam,
1870 };
1871
1872 static const struct net_device_ops device_ops = {
1873 .ndo_open = netvsc_open,
1874 .ndo_stop = netvsc_close,
1875 .ndo_start_xmit = netvsc_start_xmit,
1876 .ndo_change_rx_flags = netvsc_change_rx_flags,
1877 .ndo_set_rx_mode = netvsc_set_rx_mode,
1878 .ndo_set_features = netvsc_set_features,
1879 .ndo_change_mtu = netvsc_change_mtu,
1880 .ndo_validate_addr = eth_validate_addr,
1881 .ndo_set_mac_address = netvsc_set_mac_addr,
1882 .ndo_select_queue = netvsc_select_queue,
1883 .ndo_get_stats64 = netvsc_get_stats64,
1884 };
1885
1886 /*
1887 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1888 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1889 * present send GARP packet to network peers with netif_notify_peers().
1890 */
1891 static void netvsc_link_change(struct work_struct *w)
1892 {
1893 struct net_device_context *ndev_ctx =
1894 container_of(w, struct net_device_context, dwork.work);
1895 struct hv_device *device_obj = ndev_ctx->device_ctx;
1896 struct net_device *net = hv_get_drvdata(device_obj);
1897 struct netvsc_device *net_device;
1898 struct rndis_device *rdev;
1899 struct netvsc_reconfig *event = NULL;
1900 bool notify = false, reschedule = false;
1901 unsigned long flags, next_reconfig, delay;
1902
1903 /* if changes are happening, comeback later */
1904 if (!rtnl_trylock()) {
1905 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1906 return;
1907 }
1908
1909 net_device = rtnl_dereference(ndev_ctx->nvdev);
1910 if (!net_device)
1911 goto out_unlock;
1912
1913 rdev = net_device->extension;
1914
1915 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1916 if (time_is_after_jiffies(next_reconfig)) {
1917 /* link_watch only sends one notification with current state
1918 * per second, avoid doing reconfig more frequently. Handle
1919 * wrap around.
1920 */
1921 delay = next_reconfig - jiffies;
1922 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1923 schedule_delayed_work(&ndev_ctx->dwork, delay);
1924 goto out_unlock;
1925 }
1926 ndev_ctx->last_reconfig = jiffies;
1927
1928 spin_lock_irqsave(&ndev_ctx->lock, flags);
1929 if (!list_empty(&ndev_ctx->reconfig_events)) {
1930 event = list_first_entry(&ndev_ctx->reconfig_events,
1931 struct netvsc_reconfig, list);
1932 list_del(&event->list);
1933 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1934 }
1935 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1936
1937 if (!event)
1938 goto out_unlock;
1939
1940 switch (event->event) {
1941 /* Only the following events are possible due to the check in
1942 * netvsc_linkstatus_callback()
1943 */
1944 case RNDIS_STATUS_MEDIA_CONNECT:
1945 if (rdev->link_state) {
1946 rdev->link_state = false;
1947 netif_carrier_on(net);
1948 netvsc_tx_enable(net_device, net);
1949 } else {
1950 notify = true;
1951 }
1952 kfree(event);
1953 break;
1954 case RNDIS_STATUS_MEDIA_DISCONNECT:
1955 if (!rdev->link_state) {
1956 rdev->link_state = true;
1957 netif_carrier_off(net);
1958 netvsc_tx_disable(net_device, net);
1959 }
1960 kfree(event);
1961 break;
1962 case RNDIS_STATUS_NETWORK_CHANGE:
1963 /* Only makes sense if carrier is present */
1964 if (!rdev->link_state) {
1965 rdev->link_state = true;
1966 netif_carrier_off(net);
1967 netvsc_tx_disable(net_device, net);
1968 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1969 spin_lock_irqsave(&ndev_ctx->lock, flags);
1970 list_add(&event->list, &ndev_ctx->reconfig_events);
1971 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1972 reschedule = true;
1973 }
1974 break;
1975 }
1976
1977 rtnl_unlock();
1978
1979 if (notify)
1980 netdev_notify_peers(net);
1981
1982 /* link_watch only sends one notification with current state per
1983 * second, handle next reconfig event in 2 seconds.
1984 */
1985 if (reschedule)
1986 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1987
1988 return;
1989
1990 out_unlock:
1991 rtnl_unlock();
1992 }
1993
1994 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1995 {
1996 struct net_device_context *net_device_ctx;
1997 struct net_device *dev;
1998
1999 dev = netdev_master_upper_dev_get(vf_netdev);
2000 if (!dev || dev->netdev_ops != &device_ops)
2001 return NULL; /* not a netvsc device */
2002
2003 net_device_ctx = netdev_priv(dev);
2004 if (!rtnl_dereference(net_device_ctx->nvdev))
2005 return NULL; /* device is removed */
2006
2007 return dev;
2008 }
2009
2010 /* Called when VF is injecting data into network stack.
2011 * Change the associated network device from VF to netvsc.
2012 * note: already called with rcu_read_lock
2013 */
2014 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2015 {
2016 struct sk_buff *skb = *pskb;
2017 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2018 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2019 struct netvsc_vf_pcpu_stats *pcpu_stats
2020 = this_cpu_ptr(ndev_ctx->vf_stats);
2021
2022 skb = skb_share_check(skb, GFP_ATOMIC);
2023 if (unlikely(!skb))
2024 return RX_HANDLER_CONSUMED;
2025
2026 *pskb = skb;
2027
2028 skb->dev = ndev;
2029
2030 u64_stats_update_begin(&pcpu_stats->syncp);
2031 pcpu_stats->rx_packets++;
2032 pcpu_stats->rx_bytes += skb->len;
2033 u64_stats_update_end(&pcpu_stats->syncp);
2034
2035 return RX_HANDLER_ANOTHER;
2036 }
2037
2038 static int netvsc_vf_join(struct net_device *vf_netdev,
2039 struct net_device *ndev)
2040 {
2041 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2042 int ret;
2043
2044 ret = netdev_rx_handler_register(vf_netdev,
2045 netvsc_vf_handle_frame, ndev);
2046 if (ret != 0) {
2047 netdev_err(vf_netdev,
2048 "can not register netvsc VF receive handler (err = %d)\n",
2049 ret);
2050 goto rx_handler_failed;
2051 }
2052
2053 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2054 NULL, NULL, NULL);
2055 if (ret != 0) {
2056 netdev_err(vf_netdev,
2057 "can not set master device %s (err = %d)\n",
2058 ndev->name, ret);
2059 goto upper_link_failed;
2060 }
2061
2062 /* set slave flag before open to prevent IPv6 addrconf */
2063 vf_netdev->flags |= IFF_SLAVE;
2064
2065 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2066
2067 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2068
2069 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2070 return 0;
2071
2072 upper_link_failed:
2073 netdev_rx_handler_unregister(vf_netdev);
2074 rx_handler_failed:
2075 return ret;
2076 }
2077
2078 static void __netvsc_vf_setup(struct net_device *ndev,
2079 struct net_device *vf_netdev)
2080 {
2081 int ret;
2082
2083 /* Align MTU of VF with master */
2084 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2085 if (ret)
2086 netdev_warn(vf_netdev,
2087 "unable to change mtu to %u\n", ndev->mtu);
2088
2089 /* set multicast etc flags on VF */
2090 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2091
2092 /* sync address list from ndev to VF */
2093 netif_addr_lock_bh(ndev);
2094 dev_uc_sync(vf_netdev, ndev);
2095 dev_mc_sync(vf_netdev, ndev);
2096 netif_addr_unlock_bh(ndev);
2097
2098 if (netif_running(ndev)) {
2099 ret = dev_open(vf_netdev, NULL);
2100 if (ret)
2101 netdev_warn(vf_netdev,
2102 "unable to open: %d\n", ret);
2103 }
2104 }
2105
2106 /* Setup VF as slave of the synthetic device.
2107 * Runs in workqueue to avoid recursion in netlink callbacks.
2108 */
2109 static void netvsc_vf_setup(struct work_struct *w)
2110 {
2111 struct net_device_context *ndev_ctx
2112 = container_of(w, struct net_device_context, vf_takeover.work);
2113 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2114 struct net_device *vf_netdev;
2115
2116 if (!rtnl_trylock()) {
2117 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2118 return;
2119 }
2120
2121 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2122 if (vf_netdev)
2123 __netvsc_vf_setup(ndev, vf_netdev);
2124
2125 rtnl_unlock();
2126 }
2127
2128 /* Find netvsc by VF serial number.
2129 * The PCI hyperv controller records the serial number as the slot kobj name.
2130 */
2131 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2132 {
2133 struct device *parent = vf_netdev->dev.parent;
2134 struct net_device_context *ndev_ctx;
2135 struct pci_dev *pdev;
2136 u32 serial;
2137
2138 if (!parent || !dev_is_pci(parent))
2139 return NULL; /* not a PCI device */
2140
2141 pdev = to_pci_dev(parent);
2142 if (!pdev->slot) {
2143 netdev_notice(vf_netdev, "no PCI slot information\n");
2144 return NULL;
2145 }
2146
2147 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2148 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2149 pci_slot_name(pdev->slot));
2150 return NULL;
2151 }
2152
2153 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2154 if (!ndev_ctx->vf_alloc)
2155 continue;
2156
2157 if (ndev_ctx->vf_serial == serial)
2158 return hv_get_drvdata(ndev_ctx->device_ctx);
2159 }
2160
2161 netdev_notice(vf_netdev,
2162 "no netdev found for vf serial:%u\n", serial);
2163 return NULL;
2164 }
2165
2166 static int netvsc_register_vf(struct net_device *vf_netdev)
2167 {
2168 struct net_device_context *net_device_ctx;
2169 struct netvsc_device *netvsc_dev;
2170 struct net_device *ndev;
2171 int ret;
2172
2173 if (vf_netdev->addr_len != ETH_ALEN)
2174 return NOTIFY_DONE;
2175
2176 ndev = get_netvsc_byslot(vf_netdev);
2177 if (!ndev)
2178 return NOTIFY_DONE;
2179
2180 net_device_ctx = netdev_priv(ndev);
2181 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2182 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2183 return NOTIFY_DONE;
2184
2185 /* if synthetic interface is a different namespace,
2186 * then move the VF to that namespace; join will be
2187 * done again in that context.
2188 */
2189 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2190 ret = dev_change_net_namespace(vf_netdev,
2191 dev_net(ndev), "eth%d");
2192 if (ret)
2193 netdev_err(vf_netdev,
2194 "could not move to same namespace as %s: %d\n",
2195 ndev->name, ret);
2196 else
2197 netdev_info(vf_netdev,
2198 "VF moved to namespace with: %s\n",
2199 ndev->name);
2200 return NOTIFY_DONE;
2201 }
2202
2203 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2204
2205 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2206 return NOTIFY_DONE;
2207
2208 dev_hold(vf_netdev);
2209 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2210
2211 vf_netdev->wanted_features = ndev->features;
2212 netdev_update_features(vf_netdev);
2213
2214 return NOTIFY_OK;
2215 }
2216
2217 /* VF up/down change detected, schedule to change data path */
2218 static int netvsc_vf_changed(struct net_device *vf_netdev)
2219 {
2220 struct net_device_context *net_device_ctx;
2221 struct netvsc_device *netvsc_dev;
2222 struct net_device *ndev;
2223 bool vf_is_up = netif_running(vf_netdev);
2224
2225 ndev = get_netvsc_byref(vf_netdev);
2226 if (!ndev)
2227 return NOTIFY_DONE;
2228
2229 net_device_ctx = netdev_priv(ndev);
2230 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2231 if (!netvsc_dev)
2232 return NOTIFY_DONE;
2233
2234 netvsc_switch_datapath(ndev, vf_is_up);
2235 netdev_info(ndev, "Data path switched %s VF: %s\n",
2236 vf_is_up ? "to" : "from", vf_netdev->name);
2237
2238 return NOTIFY_OK;
2239 }
2240
2241 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2242 {
2243 struct net_device *ndev;
2244 struct net_device_context *net_device_ctx;
2245
2246 ndev = get_netvsc_byref(vf_netdev);
2247 if (!ndev)
2248 return NOTIFY_DONE;
2249
2250 net_device_ctx = netdev_priv(ndev);
2251 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2252
2253 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2254
2255 netdev_rx_handler_unregister(vf_netdev);
2256 netdev_upper_dev_unlink(vf_netdev, ndev);
2257 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2258 dev_put(vf_netdev);
2259
2260 return NOTIFY_OK;
2261 }
2262
2263 static int netvsc_probe(struct hv_device *dev,
2264 const struct hv_vmbus_device_id *dev_id)
2265 {
2266 struct net_device *net = NULL;
2267 struct net_device_context *net_device_ctx;
2268 struct netvsc_device_info *device_info = NULL;
2269 struct netvsc_device *nvdev;
2270 int ret = -ENOMEM;
2271
2272 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2273 VRSS_CHANNEL_MAX);
2274 if (!net)
2275 goto no_net;
2276
2277 netif_carrier_off(net);
2278
2279 netvsc_init_settings(net);
2280
2281 net_device_ctx = netdev_priv(net);
2282 net_device_ctx->device_ctx = dev;
2283 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2284 if (netif_msg_probe(net_device_ctx))
2285 netdev_dbg(net, "netvsc msg_enable: %d\n",
2286 net_device_ctx->msg_enable);
2287
2288 hv_set_drvdata(dev, net);
2289
2290 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2291
2292 spin_lock_init(&net_device_ctx->lock);
2293 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2294 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2295
2296 net_device_ctx->vf_stats
2297 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2298 if (!net_device_ctx->vf_stats)
2299 goto no_stats;
2300
2301 net->netdev_ops = &device_ops;
2302 net->ethtool_ops = &ethtool_ops;
2303 SET_NETDEV_DEV(net, &dev->device);
2304
2305 /* We always need headroom for rndis header */
2306 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2307
2308 /* Initialize the number of queues to be 1, we may change it if more
2309 * channels are offered later.
2310 */
2311 netif_set_real_num_tx_queues(net, 1);
2312 netif_set_real_num_rx_queues(net, 1);
2313
2314 /* Notify the netvsc driver of the new device */
2315 device_info = netvsc_devinfo_get(NULL);
2316
2317 if (!device_info) {
2318 ret = -ENOMEM;
2319 goto devinfo_failed;
2320 }
2321
2322 nvdev = rndis_filter_device_add(dev, device_info);
2323 if (IS_ERR(nvdev)) {
2324 ret = PTR_ERR(nvdev);
2325 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2326 goto rndis_failed;
2327 }
2328
2329 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2330
2331 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2332 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2333 * all subchannels to show up, but that may not happen because
2334 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2335 * -> ... -> device_add() -> ... -> __device_attach() can't get
2336 * the device lock, so all the subchannels can't be processed --
2337 * finally netvsc_subchan_work() hangs forever.
2338 */
2339 rtnl_lock();
2340
2341 if (nvdev->num_chn > 1)
2342 schedule_work(&nvdev->subchan_work);
2343
2344 /* hw_features computed in rndis_netdev_set_hwcaps() */
2345 net->features = net->hw_features |
2346 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2347 NETIF_F_HW_VLAN_CTAG_RX;
2348 net->vlan_features = net->features;
2349
2350 /* MTU range: 68 - 1500 or 65521 */
2351 net->min_mtu = NETVSC_MTU_MIN;
2352 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2353 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2354 else
2355 net->max_mtu = ETH_DATA_LEN;
2356
2357 ret = register_netdevice(net);
2358 if (ret != 0) {
2359 pr_err("Unable to register netdev.\n");
2360 goto register_failed;
2361 }
2362
2363 list_add(&net_device_ctx->list, &netvsc_dev_list);
2364 rtnl_unlock();
2365
2366 kfree(device_info);
2367 return 0;
2368
2369 register_failed:
2370 rtnl_unlock();
2371 rndis_filter_device_remove(dev, nvdev);
2372 rndis_failed:
2373 kfree(device_info);
2374 devinfo_failed:
2375 free_percpu(net_device_ctx->vf_stats);
2376 no_stats:
2377 hv_set_drvdata(dev, NULL);
2378 free_netdev(net);
2379 no_net:
2380 return ret;
2381 }
2382
2383 static int netvsc_remove(struct hv_device *dev)
2384 {
2385 struct net_device_context *ndev_ctx;
2386 struct net_device *vf_netdev, *net;
2387 struct netvsc_device *nvdev;
2388
2389 net = hv_get_drvdata(dev);
2390 if (net == NULL) {
2391 dev_err(&dev->device, "No net device to remove\n");
2392 return 0;
2393 }
2394
2395 ndev_ctx = netdev_priv(net);
2396
2397 cancel_delayed_work_sync(&ndev_ctx->dwork);
2398
2399 rtnl_lock();
2400 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2401 if (nvdev)
2402 cancel_work_sync(&nvdev->subchan_work);
2403
2404 /*
2405 * Call to the vsc driver to let it know that the device is being
2406 * removed. Also blocks mtu and channel changes.
2407 */
2408 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2409 if (vf_netdev)
2410 netvsc_unregister_vf(vf_netdev);
2411
2412 if (nvdev)
2413 rndis_filter_device_remove(dev, nvdev);
2414
2415 unregister_netdevice(net);
2416 list_del(&ndev_ctx->list);
2417
2418 rtnl_unlock();
2419
2420 hv_set_drvdata(dev, NULL);
2421
2422 free_percpu(ndev_ctx->vf_stats);
2423 free_netdev(net);
2424 return 0;
2425 }
2426
2427 static const struct hv_vmbus_device_id id_table[] = {
2428 /* Network guid */
2429 { HV_NIC_GUID, },
2430 { },
2431 };
2432
2433 MODULE_DEVICE_TABLE(vmbus, id_table);
2434
2435 /* The one and only one */
2436 static struct hv_driver netvsc_drv = {
2437 .name = KBUILD_MODNAME,
2438 .id_table = id_table,
2439 .probe = netvsc_probe,
2440 .remove = netvsc_remove,
2441 .driver = {
2442 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2443 },
2444 };
2445
2446 /*
2447 * On Hyper-V, every VF interface is matched with a corresponding
2448 * synthetic interface. The synthetic interface is presented first
2449 * to the guest. When the corresponding VF instance is registered,
2450 * we will take care of switching the data path.
2451 */
2452 static int netvsc_netdev_event(struct notifier_block *this,
2453 unsigned long event, void *ptr)
2454 {
2455 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2456
2457 /* Skip our own events */
2458 if (event_dev->netdev_ops == &device_ops)
2459 return NOTIFY_DONE;
2460
2461 /* Avoid non-Ethernet type devices */
2462 if (event_dev->type != ARPHRD_ETHER)
2463 return NOTIFY_DONE;
2464
2465 /* Avoid Vlan dev with same MAC registering as VF */
2466 if (is_vlan_dev(event_dev))
2467 return NOTIFY_DONE;
2468
2469 /* Avoid Bonding master dev with same MAC registering as VF */
2470 if ((event_dev->priv_flags & IFF_BONDING) &&
2471 (event_dev->flags & IFF_MASTER))
2472 return NOTIFY_DONE;
2473
2474 switch (event) {
2475 case NETDEV_REGISTER:
2476 return netvsc_register_vf(event_dev);
2477 case NETDEV_UNREGISTER:
2478 return netvsc_unregister_vf(event_dev);
2479 case NETDEV_UP:
2480 case NETDEV_DOWN:
2481 return netvsc_vf_changed(event_dev);
2482 default:
2483 return NOTIFY_DONE;
2484 }
2485 }
2486
2487 static struct notifier_block netvsc_netdev_notifier = {
2488 .notifier_call = netvsc_netdev_event,
2489 };
2490
2491 static void __exit netvsc_drv_exit(void)
2492 {
2493 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2494 vmbus_driver_unregister(&netvsc_drv);
2495 }
2496
2497 static int __init netvsc_drv_init(void)
2498 {
2499 int ret;
2500
2501 if (ring_size < RING_SIZE_MIN) {
2502 ring_size = RING_SIZE_MIN;
2503 pr_info("Increased ring_size to %u (min allowed)\n",
2504 ring_size);
2505 }
2506 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2507
2508 ret = vmbus_driver_register(&netvsc_drv);
2509 if (ret)
2510 return ret;
2511
2512 register_netdevice_notifier(&netvsc_netdev_notifier);
2513 return 0;
2514 }
2515
2516 MODULE_LICENSE("GPL");
2517 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2518
2519 module_init(netvsc_drv_init);
2520 module_exit(netvsc_drv_exit);
Ubuntu Jammy Linux kernel mirror