projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'integrity-v6.7' of git://git.kernel.org/pub/scm/linux/kernel/git/zohar...
[linux-block.git] / drivers / net / hyperv / netvsc_drv.c
CommitLineData
9952f691 1// SPDX-License-Identifier: GPL-2.0-only
fceaf24a 2/*
fceaf24a
HJ		 3 * Copyright (c) 2009, Microsoft Corporation.
4 *
fceaf24a 5 * Authors:
d0e94d17 6 * Haiyang Zhang <haiyangz@microsoft.com>
fceaf24a 7 * Hank Janssen <hjanssen@microsoft.com>
fceaf24a 8 */
eb335bc4
HJ		 9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
fceaf24a 11#include <linux/init.h>
9079ce69 12#include <linux/atomic.h>
cc69837f 13#include <linux/ethtool.h>
fceaf24a
HJ		 14#include <linux/module.h>
15#include <linux/highmem.h>
16#include <linux/device.h>
fceaf24a 17#include <linux/io.h>
fceaf24a
HJ		 18#include <linux/delay.h>
19#include <linux/netdevice.h>
20#include <linux/inetdevice.h>
21#include <linux/etherdevice.h>
b93c1b5a 22#include <linux/pci.h>
fceaf24a 23#include <linux/skbuff.h>
c802db11 24#include <linux/if_vlan.h>
fceaf24a 25#include <linux/in.h>
5a0e3ad6 26#include <linux/slab.h>
27f5aa92 27#include <linux/rtnetlink.h>
0c195567 28#include <linux/netpoll.h>
351e1581 29#include <linux/bpf.h>
27f5aa92 30
fceaf24a
HJ		 31#include <net/arp.h>
32#include <net/route.h>
33#include <net/sock.h>
34#include <net/pkt_sched.h>
8eb1b3c3
MK		 35#include <net/checksum.h>
36#include <net/ip6_checksum.h>
3f335ea2 37
5ca7252a 38#include "hyperv_net.h"
fceaf24a 39
7b2ee50c 40#define RING_SIZE_MIN 64
8b532797 41
27a70af3 42#define LINKCHANGE_INT (2 * HZ)
6123c668 43#define VF_TAKEOVER_INT (HZ / 10)
a50af86d 44
a7f99d0f 45static unsigned int ring_size __ro_after_init = 128;
d61e4038 46module_param(ring_size, uint, 0444);
450d7a4b 47MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
a7f99d0f 48unsigned int netvsc_ring_bytes __ro_after_init;
fceaf24a 49
3f300ff4
SX		 50static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
53 NETIF_MSG_TX_ERR;
54
55static int debug = -1;
d61e4038 56module_param(debug, int, 0444);
3f300ff4
SX		 57MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
58
7bf7bb37
SH		 59static LIST_HEAD(netvsc_dev_list);
60
bee9d41b 61static void netvsc_change_rx_flags(struct net_device *net, int change)
fceaf24a 62{
bee9d41b
SH		 63 struct net_device_context *ndev_ctx = netdev_priv(net);
64 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
65 int inc;
66
67 if (!vf_netdev)
68 return;
69
70 if (change & IFF_PROMISC) {
71 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
72 dev_set_promiscuity(vf_netdev, inc);
73 }
74
75 if (change & IFF_ALLMULTI) {
76 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
77 dev_set_allmulti(vf_netdev, inc);
78 }
79}
80
81static void netvsc_set_rx_mode(struct net_device *net)
82{
83 struct net_device_context *ndev_ctx = netdev_priv(net);
35a57b7f
SH		 84 struct net_device *vf_netdev;
85 struct netvsc_device *nvdev;
bee9d41b 86
35a57b7f
SH		 87 rcu_read_lock();
88 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
bee9d41b
SH		 89 if (vf_netdev) {
90 dev_uc_sync(vf_netdev, net);
91 dev_mc_sync(vf_netdev, net);
92 }
d426b2e3 93
35a57b7f
SH		 94 nvdev = rcu_dereference(ndev_ctx->nvdev);
95 if (nvdev)
96 rndis_filter_update(nvdev);
97 rcu_read_unlock();
fceaf24a
HJ		 98}
99
1b704c4a
HZ		 100static void netvsc_tx_enable(struct netvsc_device *nvscdev,
101 struct net_device *ndev)
102{
103 nvscdev->tx_disable = false;
104 virt_wmb(); /* ensure queue wake up mechanism is on */
105
106 netif_tx_wake_all_queues(ndev);
107}
108
fceaf24a
HJ		 109static int netvsc_open(struct net_device *net)
110{
53fa1a6f 111 struct net_device_context *ndev_ctx = netdev_priv(net);
0c195567 112 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
79e8cbe7 113 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
891de74d 114 struct rndis_device *rdev;
02fafbc6 115 int ret = 0;
fceaf24a 116
891de74d
HZ		 117 netif_carrier_off(net);
118
d515d0ff 119 /* Open up the device */
2f5fa6c8 120 ret = rndis_filter_open(nvdev);
d515d0ff
HZ		 121 if (ret != 0) {
122 netdev_err(net, "unable to open device (ret %d).\n", ret);
123 return ret;
fceaf24a
HJ		 124 }
125
891de74d 126 rdev = nvdev->extension;
52acf73b 127 if (!rdev->link_state) {
891de74d 128 netif_carrier_on(net);
1b704c4a 129 netvsc_tx_enable(nvdev, net);
52acf73b 130 }
891de74d 131
0c195567 132 if (vf_netdev) {
133 /* Setting synthetic device up transparently sets
134 * slave as up. If open fails, then slave will be
135 * still be offline (and not used).
136 */
00f54e68 137 ret = dev_open(vf_netdev, NULL);
0c195567 138 if (ret)
139 netdev_warn(net,
140 "unable to open slave: %s: %d\n",
141 vf_netdev->name, ret);
142 }
143 return 0;
fceaf24a
HJ		 144}
145
7b2ee50c 146static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
fceaf24a 147{
7b2ee50c
SH		 148 unsigned int retry = 0;
149 int i;
2de8530b
HZ		 150
151 /* Ensure pending bytes in ring are read */
7b2ee50c
SH		 152 for (;;) {
153 u32 aread = 0;
154
2de8530b 155 for (i = 0; i < nvdev->num_chn; i++) {
7b2ee50c
SH		 156 struct vmbus_channel *chn
157 = nvdev->chan_table[i].channel;
158
2de8530b
HZ		 159 if (!chn)
160 continue;
161
7b2ee50c
SH		 162 /* make sure receive not running now */
163 napi_synchronize(&nvdev->chan_table[i].napi);
164
40975962 165 aread = hv_get_bytes_to_read(&chn->inbound);
2de8530b
HZ		 166 if (aread)
167 break;
168
40975962 169 aread = hv_get_bytes_to_read(&chn->outbound);
2de8530b
HZ		 170 if (aread)
171 break;
172 }
173
7b2ee50c
SH		 174 if (aread == 0)
175 return 0;
2de8530b 176
7b2ee50c
SH		 177 if (++retry > RETRY_MAX)
178 return -ETIMEDOUT;
2de8530b 179
7b2ee50c 180 usleep_range(RETRY_US_LO, RETRY_US_HI);
2de8530b 181 }
7b2ee50c 182}
2de8530b 183
1b704c4a
HZ		 184static void netvsc_tx_disable(struct netvsc_device *nvscdev,
185 struct net_device *ndev)
186{
187 if (nvscdev) {
188 nvscdev->tx_disable = true;
189 virt_wmb(); /* ensure txq will not wake up after stop */
190 }
191
192 netif_tx_disable(ndev);
193}
194
7b2ee50c
SH		 195static int netvsc_close(struct net_device *net)
196{
197 struct net_device_context *net_device_ctx = netdev_priv(net);
198 struct net_device *vf_netdev
199 = rtnl_dereference(net_device_ctx->vf_netdev);
200 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
201 int ret;
202
1b704c4a 203 netvsc_tx_disable(nvdev, net);
7b2ee50c
SH		 204
205 /* No need to close rndis filter if it is removed already */
206 if (!nvdev)
207 return 0;
208
209 ret = rndis_filter_close(nvdev);
210 if (ret != 0) {
211 netdev_err(net, "unable to close device (ret %d).\n", ret);
212 return ret;
2de8530b 213 }
fceaf24a 214
7b2ee50c
SH		 215 ret = netvsc_wait_until_empty(nvdev);
216 if (ret)
217 netdev_err(net, "Ring buffer not empty after closing rndis\n");
218
0c195567 219 if (vf_netdev)
220 dev_close(vf_netdev);
221
fceaf24a
HJ		 222 return ret;
223}
224
f5a22550
SH		 225static inline void *init_ppi_data(struct rndis_message *msg,
226 u32 ppi_size, u32 pkt_type)
8a00251a 227{
f5a22550 228 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
8a00251a
KS		 229 struct rndis_per_packet_info *ppi;
230
8a00251a 231 rndis_pkt->data_offset += ppi_size;
f5a22550
SH		 232 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
233 + rndis_pkt->per_pkt_info_len;
8a00251a
KS		 234
235 ppi->size = ppi_size;
236 ppi->type = pkt_type;
e3a9667a 237 ppi->internal = 0;
8a00251a
KS		 238 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
239
240 rndis_pkt->per_pkt_info_len += ppi_size;
241
f5a22550 242 return ppi + 1;
8a00251a
KS		 243}
244
8db91f6a
HZ		 245static inline int netvsc_get_tx_queue(struct net_device *ndev,
246 struct sk_buff *skb, int old_idx)
247{
248 const struct net_device_context *ndc = netdev_priv(ndev);
249 struct sock *sk = skb->sk;
250 int q_idx;
251
39e91cfb
HZ		 252 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
253 (VRSS_SEND_TAB_SIZE - 1)];
8db91f6a
HZ		 254
255 /* If queue index changed record the new value */
256 if (q_idx != old_idx &&
257 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
258 sk_tx_queue_set(sk, q_idx);
259
260 return q_idx;
261}
262
d8e18ee0 263/*
264 * Select queue for transmit.
265 *
266 * If a valid queue has already been assigned, then use that.
267 * Otherwise compute tx queue based on hash and the send table.
268 *
a350ecce 269 * This is basically similar to default (netdev_pick_tx) with the added step
d8e18ee0 270 * of using the host send_table when no other queue has been assigned.
271 *
272 * TODO support XPS - but get_xps_queue not exported
273 */
0c195567 274static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
5b54dac8 275{
8db91f6a
HZ		 276 int q_idx = sk_tx_queue_get(skb->sk);
277
0c195567 278 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
8db91f6a
HZ		 279 /* If forwarding a packet, we use the recorded queue when
280 * available for better cache locality.
281 */
282 if (skb_rx_queue_recorded(skb))
283 q_idx = skb_get_rx_queue(skb);
284 else
285 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
d8e18ee0 286 }
5b54dac8
HZ		 287
288 return q_idx;
289}
290
0c195567 291static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
a350ecce 292 struct net_device *sb_dev)
0c195567 293{
294 struct net_device_context *ndc = netdev_priv(ndev);
295 struct net_device *vf_netdev;
296 u16 txq;
297
298 rcu_read_lock();
299 vf_netdev = rcu_dereference(ndc->vf_netdev);
300 if (vf_netdev) {
b3bf5666
SH		 301 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
302
303 if (vf_ops->ndo_select_queue)
a350ecce 304 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
b3bf5666 305 else
a350ecce 306 txq = netdev_pick_tx(vf_netdev, skb, NULL);
b3bf5666
SH		 307
308 /* Record the queue selected by VF so that it can be
309 * used for common case where VF has more queues than
310 * the synthetic device.
311 */
312 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
0c195567 313 } else {
314 txq = netvsc_pick_tx(ndev, skb);
315 }
316 rcu_read_unlock();
317
4d820543 318 while (txq >= ndev->real_num_tx_queues)
0c195567 319 txq -= ndev->real_num_tx_queues;
320
321 return txq;
322}
323
11d8620e 324static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
89bb42b1 325 struct hv_page_buffer *pb)
54a7357f
KS		 326{
327 int j = 0;
328
11d8620e
BF		 329 hvpfn += offset >> HV_HYP_PAGE_SHIFT;
330 offset = offset & ~HV_HYP_PAGE_MASK;
54a7357f
KS		 331
332 while (len > 0) {
333 unsigned long bytes;
334
11d8620e 335 bytes = HV_HYP_PAGE_SIZE - offset;
54a7357f
KS		 336 if (bytes > len)
337 bytes = len;
11d8620e 338 pb[j].pfn = hvpfn;
54a7357f
KS		 339 pb[j].offset = offset;
340 pb[j].len = bytes;
341
342 offset += bytes;
343 len -= bytes;
344
11d8620e
BF		 345 if (offset == HV_HYP_PAGE_SIZE && len) {
346 hvpfn++;
54a7357f
KS		 347 offset = 0;
348 j++;
349 }
350 }
351
352 return j + 1;
353}
354
8a00251a 355static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
a9f2e2d6 356 struct hv_netvsc_packet *packet,
02b6de01 357 struct hv_page_buffer *pb)
54a7357f
KS		 358{
359 u32 slots_used = 0;
360 char *data = skb->data;
361 int frags = skb_shinfo(skb)->nr_frags;
362 int i;
363
364 /* The packet is laid out thus:
aa0a34be 365 * 1. hdr: RNDIS header and PPI
54a7357f
KS		 366 * 2. skb linear data
367 * 3. skb fragment data
368 */
11d8620e
BF		 369 slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
370 offset_in_hvpage(hdr),
371 len,
372 &pb[slots_used]);
54a7357f 373
aa0a34be
HZ		 374 packet->rmsg_size = len;
375 packet->rmsg_pgcnt = slots_used;
376
11d8620e
BF		 377 slots_used += fill_pg_buf(virt_to_hvpfn(data),
378 offset_in_hvpage(data),
379 skb_headlen(skb),
380 &pb[slots_used]);
54a7357f
KS		 381
382 for (i = 0; i < frags; i++) {
383 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
384
11d8620e
BF		 385 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
386 skb_frag_off(frag),
387 skb_frag_size(frag),
388 &pb[slots_used]);
54a7357f 389 }
8a00251a 390 return slots_used;
54a7357f
KS		 391}
392
80d887db 393static int count_skb_frag_slots(struct sk_buff *skb)
394{
395 int i, frags = skb_shinfo(skb)->nr_frags;
396 int pages = 0;
397
398 for (i = 0; i < frags; i++) {
399 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
400 unsigned long size = skb_frag_size(frag);
b54c9d5b 401 unsigned long offset = skb_frag_off(frag);
80d887db 402
403 /* Skip unused frames from start of page */
11d8620e
BF		 404 offset &= ~HV_HYP_PAGE_MASK;
405 pages += HVPFN_UP(offset + size);
80d887db 406 }
407 return pages;
408}
409
410static int netvsc_get_slots(struct sk_buff *skb)
54a7357f 411{
80d887db 412 char *data = skb->data;
11d8620e 413 unsigned int offset = offset_in_hvpage(data);
80d887db 414 unsigned int len = skb_headlen(skb);
415 int slots;
416 int frag_slots;
417
11d8620e 418 slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
80d887db 419 frag_slots = count_skb_frag_slots(skb);
420 return slots + frag_slots;
54a7357f
KS		 421}
422
23312a3b 423static u32 net_checksum_info(struct sk_buff *skb)
08cd04bf 424{
23312a3b 425 if (skb->protocol == htons(ETH_P_IP)) {
426 struct iphdr *ip = ip_hdr(skb);
08cd04bf 427
23312a3b 428 if (ip->protocol == IPPROTO_TCP)
429 return TRANSPORT_INFO_IPV4_TCP;
430 else if (ip->protocol == IPPROTO_UDP)
431 return TRANSPORT_INFO_IPV4_UDP;
08cd04bf 432 } else {
23312a3b 433 struct ipv6hdr *ip6 = ipv6_hdr(skb);
434
435 if (ip6->nexthdr == IPPROTO_TCP)
436 return TRANSPORT_INFO_IPV6_TCP;
37b9dfa0 437 else if (ip6->nexthdr == IPPROTO_UDP)
23312a3b 438 return TRANSPORT_INFO_IPV6_UDP;
08cd04bf
KS		 439 }
440
23312a3b 441 return TRANSPORT_INFO_NOT_IP;
08cd04bf
KS		 442}
443
0c195567 444/* Send skb on the slave VF device. */
445static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
446 struct sk_buff *skb)
447{
448 struct net_device_context *ndev_ctx = netdev_priv(net);
449 unsigned int len = skb->len;
450 int rc;
451
452 skb->dev = vf_netdev;
c3d897e0 453 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
0c195567 454
455 rc = dev_queue_xmit(skb);
456 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
457 struct netvsc_vf_pcpu_stats *pcpu_stats
458 = this_cpu_ptr(ndev_ctx->vf_stats);
459
460 u64_stats_update_begin(&pcpu_stats->syncp);
461 pcpu_stats->tx_packets++;
462 pcpu_stats->tx_bytes += len;
463 u64_stats_update_end(&pcpu_stats->syncp);
464 } else {
465 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
466 }
467
468 return rc;
469}
470
351e1581 471static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
fceaf24a 472{
fceaf24a 473 struct net_device_context *net_device_ctx = netdev_priv(net);
981a1bd8 474 struct hv_netvsc_packet *packet = NULL;
02fafbc6 475 int ret;
8a00251a
KS		 476 unsigned int num_data_pgs;
477 struct rndis_message *rndis_msg;
0c195567 478 struct net_device *vf_netdev;
8a00251a 479 u32 rndis_msg_size;
307f0995 480 u32 hash;
02b6de01 481 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
fceaf24a 482
7c9864bb
SH		 483 /* If VF is present and up then redirect packets to it.
484 * Skip the VF if it is marked down or has no carrier.
485 * If netpoll is in uses, then VF can not be used either.
0c195567 486 */
487 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
488 if (vf_netdev && netif_running(vf_netdev) &&
69d25a6c
LL		 489 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
490 net_device_ctx->data_path_is_vf)
0c195567 491 return netvsc_vf_xmit(net, vf_netdev, skb);
492
80d887db 493 /* We will atmost need two pages to describe the rndis
494 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
e88f7e07
VK		 495 * of pages in a single packet. If skb is scattered around
496 * more pages we try linearizing it.
54a7357f 497 */
80d887db 498
499 num_data_pgs = netvsc_get_slots(skb) + 2;
500
0ab05141 501 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
4323b47c
SH		 502 ++net_device_ctx->eth_stats.tx_scattered;
503
504 if (skb_linearize(skb))
505 goto no_memory;
0ab05141 506
80d887db 507 num_data_pgs = netvsc_get_slots(skb) + 2;
0ab05141 508 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
4323b47c 509 ++net_device_ctx->eth_stats.tx_too_big;
0ab05141
SH		 510 goto drop;
511 }
54a7357f 512 }
fceaf24a 513
c0eb4540
KS		 514 /*
515 * Place the rndis header in the skb head room and
516 * the skb->cb will be used for hv_netvsc_packet
517 * structure.
518 */
519 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
4323b47c
SH		 520 if (ret)
521 goto no_memory;
522
c0eb4540
KS		 523 /* Use the skb control buffer for building up the packet */
524 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
c593642c 525 sizeof_field(struct sk_buff, cb));
c0eb4540 526 packet = (struct hv_netvsc_packet *)skb->cb;
fceaf24a 527
5b54dac8
HZ		 528 packet->q_idx = skb_get_queue_mapping(skb);
529
4d447c9a 530 packet->total_data_buflen = skb->len;
793e3955 531 packet->total_bytes = skb->len;
532 packet->total_packets = 1;
fceaf24a 533
c0eb4540 534 rndis_msg = (struct rndis_message *)skb->head;
b08cc791 535
8a00251a 536 /* Add the rndis header */
8a00251a
KS		 537 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
538 rndis_msg->msg_len = packet->total_data_buflen;
f5a22550
SH		 539
540 rndis_msg->msg.pkt = (struct rndis_packet) {
541 .data_offset = sizeof(struct rndis_packet),
542 .data_len = packet->total_data_buflen,
543 .per_pkt_info_offset = sizeof(struct rndis_packet),
544 };
8a00251a
KS		 545
546 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
547
307f0995
HZ		 548 hash = skb_get_hash_raw(skb);
549 if (hash != 0 && net->real_num_tx_queues > 1) {
f5a22550
SH		 550 u32 *hash_info;
551
307f0995 552 rndis_msg_size += NDIS_HASH_PPI_SIZE;
f5a22550
SH		 553 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
554 NBL_HASH_VALUE);
555 *hash_info = hash;
307f0995
HZ		 556 }
557
fdd8fac4
SK		 558 /* When using AF_PACKET we need to drop VLAN header from
559 * the frame and update the SKB to allow the HOST OS
560 * to transmit the 802.1Q packet
561 */
562 if (skb->protocol == htons(ETH_P_8021Q)) {
563 u16 vlan_tci;
564
565 skb_reset_mac_header(skb);
566 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
567 if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
568 ++net_device_ctx->eth_stats.vlan_error;
569 goto drop;
570 }
571
572 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
573 /* Update the NDIS header pkt lengths */
574 packet->total_data_buflen -= VLAN_HLEN;
575 packet->total_bytes -= VLAN_HLEN;
576 rndis_msg->msg_len = packet->total_data_buflen;
577 rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
578 }
579 }
580
0ab05141 581 if (skb_vlan_tag_present(skb)) {
8a00251a
KS		 582 struct ndis_pkt_8021q_info *vlan;
583
584 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
f5a22550
SH		 585 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
586 IEEE_8021Q_INFO);
00f5024e 587
f5a22550 588 vlan->value = 0;
98ba780e
MM		 589 vlan->vlanid = skb_vlan_tag_get_id(skb);
590 vlan->cfi = skb_vlan_tag_get_cfi(skb);
591 vlan->pri = skb_vlan_tag_get_prio(skb);
8a00251a
KS		 592 }
593
23312a3b 594 if (skb_is_gso(skb)) {
0ab05141
SH		 595 struct ndis_tcp_lso_info *lso_info;
596
597 rndis_msg_size += NDIS_LSO_PPI_SIZE;
f5a22550
SH		 598 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
599 TCP_LARGESEND_PKTINFO);
0ab05141 600
f5a22550 601 lso_info->value = 0;
0ab05141 602 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
23312a3b 603 if (skb->protocol == htons(ETH_P_IP)) {
0ab05141
SH		 604 lso_info->lso_v2_transmit.ip_version =
605 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
606 ip_hdr(skb)->tot_len = 0;
607 ip_hdr(skb)->check = 0;
608 tcp_hdr(skb)->check =
609 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
610 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
611 } else {
612 lso_info->lso_v2_transmit.ip_version =
613 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
1eb2c576 614 tcp_v6_gso_csum_prep(skb);
0ab05141 615 }
23312a3b 616 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
0ab05141 617 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
ad19bc8a 618 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
23312a3b 619 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
620 struct ndis_tcp_ip_checksum_info *csum_info;
621
ad19bc8a 622 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
f5a22550
SH		 623 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
624 TCPIP_CHKSUM_PKTINFO);
ad19bc8a 625
f5a22550 626 csum_info->value = 0;
23312a3b 627 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
628
629 if (skb->protocol == htons(ETH_P_IP)) {
ad19bc8a 630 csum_info->transmit.is_ipv4 = 1;
23312a3b 631
632 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
633 csum_info->transmit.tcp_checksum = 1;
634 else
635 csum_info->transmit.udp_checksum = 1;
636 } else {
ad19bc8a 637 csum_info->transmit.is_ipv6 = 1;
638
23312a3b 639 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
640 csum_info->transmit.tcp_checksum = 1;
641 else
642 csum_info->transmit.udp_checksum = 1;
643 }
ad19bc8a 644 } else {
23312a3b 645 /* Can't do offload of this type of checksum */
ad19bc8a 646 if (skb_checksum_help(skb))
647 goto drop;
648 }
08cd04bf
KS		 649 }
650
8a00251a
KS		 651 /* Start filling in the page buffers with the rndis hdr */
652 rndis_msg->msg_len += rndis_msg_size;
942396b0 653 packet->total_data_buflen = rndis_msg->msg_len;
8a00251a 654 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
02b6de01 655 skb, packet, pb);
8a00251a 656
76d13b56 657 /* timestamp packet in software */
658 skb_tx_timestamp(skb);
2a926f79 659
351e1581 660 ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
793e3955 661 if (likely(ret == 0))
0ab05141 662 return NETDEV_TX_OK;
4323b47c
SH		 663
664 if (ret == -EAGAIN) {
665 ++net_device_ctx->eth_stats.tx_busy;
0ab05141 666 return NETDEV_TX_BUSY;
4323b47c
SH		 667 }
668
669 if (ret == -ENOSPC)
670 ++net_device_ctx->eth_stats.tx_no_space;
0ab05141
SH		 671
672drop:
673 dev_kfree_skb_any(skb);
674 net->stats.tx_dropped++;
fceaf24a 675
0ab05141 676 return NETDEV_TX_OK;
4323b47c
SH		 677
678no_memory:
679 ++net_device_ctx->eth_stats.tx_no_memory;
680 goto drop;
fceaf24a 681}
89bb42b1 682
7fdc66de
NC		 683static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
684 struct net_device *ndev)
351e1581
HZ		 685{
686 return netvsc_xmit(skb, ndev, false);
687}
688
3e189519 689/*
02fafbc6
GKH		 690 * netvsc_linkstatus_callback - Link up/down notification
691 */
79cf1bae 692void netvsc_linkstatus_callback(struct net_device *net,
0ba35fe9 693 struct rndis_message *resp,
3946688e 694 void *data, u32 data_buflen)
fceaf24a 695{
3a494e71 696 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
79cf1bae 697 struct net_device_context *ndev_ctx = netdev_priv(net);
27a70af3
VK		 698 struct netvsc_reconfig *event;
699 unsigned long flags;
891de74d 700
44144185
AB		 701 /* Ensure the packet is big enough to access its fields */
702 if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
703 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
704 resp->msg_len);
705 return;
706 }
707
0ba35fe9
APM		 708 /* Copy the RNDIS indicate status into nvchan->recv_buf */
709 memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
710
7f5d5af0
HZ		 711 /* Update the physical link speed when changing to another vSwitch */
712 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
713 u32 speed;
714
3946688e
APM		 715 /* Validate status_buf_offset and status_buflen.
716 *
717 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
718 * for the status buffer field in resp->msg_len; perform the validation
719 * using data_buflen (>= resp->msg_len).
720 */
505e3f00
APM		 721 if (indicate->status_buflen < sizeof(speed) ||
722 indicate->status_buf_offset < sizeof(*indicate) ||
3946688e
APM		 723 data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
724 data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
505e3f00
APM		 725 < indicate->status_buflen) {
726 netdev_err(net, "invalid rndis_indicate_status packet\n");
727 return;
728 }
729
0ba35fe9 730 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
7f5d5af0
HZ		 731 ndev_ctx->speed = speed;
732 return;
733 }
734
735 /* Handle these link change statuses below */
27a70af3
VK		 736 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
737 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
738 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
3a494e71 739 return;
891de74d 740
7f5d5af0 741 if (net->reg_state != NETREG_REGISTERED)
fceaf24a 742 return;
fceaf24a 743
27a70af3
VK		 744 event = kzalloc(sizeof(*event), GFP_ATOMIC);
745 if (!event)
746 return;
747 event->event = indicate->status;
748
749 spin_lock_irqsave(&ndev_ctx->lock, flags);
750 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
751 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
752
753 schedule_delayed_work(&ndev_ctx->dwork, 0);
fceaf24a
HJ		 754}
755
78e0a006 756/* This function should only be called after skb_record_rx_queue() */
1cb9d3b6 757void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
351e1581
HZ		 758{
759 int rc;
760
761 skb->queue_mapping = skb_get_rx_queue(skb);
762 __skb_push(skb, ETH_HLEN);
763
764 rc = netvsc_xmit(skb, ndev, true);
765
766 if (dev_xmit_complete(rc))
767 return;
768
769 dev_kfree_skb_any(skb);
770 ndev->stats.tx_dropped++;
771}
772
bf48648d
HZ		 773static void netvsc_comp_ipcsum(struct sk_buff *skb)
774{
775 struct iphdr *iph = (struct iphdr *)skb->data;
776
777 iph->check = 0;
778 iph->check = ip_fast_csum(iph, iph->ihl);
779}
780
84bf9cef 781static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
351e1581
HZ		 782 struct netvsc_channel *nvchan,
783 struct xdp_buff *xdp)
fceaf24a 784{
c8e4eff4 785 struct napi_struct *napi = &nvchan->napi;
0ba35fe9 786 const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
c8e4eff4 787 const struct ndis_tcp_ip_checksum_info *csum_info =
0ba35fe9
APM		 788 &nvchan->rsc.csum_info;
789 const u32 *hash_info = &nvchan->rsc.hash_info;
790 u8 ppi_flags = nvchan->rsc.ppi_flags;
fceaf24a 791 struct sk_buff *skb;
351e1581 792 void *xbuf = xdp->data_hard_start;
c8e4eff4 793 int i;
fceaf24a 794
351e1581
HZ		 795 if (xbuf) {
796 unsigned int hdroom = xdp->data - xdp->data_hard_start;
797 unsigned int xlen = xdp->data_end - xdp->data;
7358877a 798 unsigned int frag_size = xdp->frame_sz;
fceaf24a 799
351e1581
HZ		 800 skb = build_skb(xbuf, frag_size);
801
802 if (!skb) {
803 __free_page(virt_to_page(xbuf));
804 return NULL;
805 }
806
807 skb_reserve(skb, hdroom);
808 skb_put(skb, xlen);
809 skb->dev = napi->dev;
810 } else {
811 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
812
813 if (!skb)
814 return NULL;
815
816 /* Copy to skb. This copy is needed here since the memory
817 * pointed by hv_netvsc_packet cannot be deallocated.
818 */
819 for (i = 0; i < nvchan->rsc.cnt; i++)
820 skb_put_data(skb, nvchan->rsc.data[i],
821 nvchan->rsc.len[i]);
822 }
fceaf24a
HJ		 823
824 skb->protocol = eth_type_trans(skb, net);
e52fed71
SH		 825
826 /* skb is already created with CHECKSUM_NONE */
827 skb_checksum_none_assert(skb);
828
bf48648d
HZ		 829 /* Incoming packets may have IP header checksum verified by the host.
830 * They may not have IP header checksum computed after coalescing.
831 * We compute it here if the flags are set, because on Linux, the IP
832 * checksum is always checked.
833 */
0ba35fe9 834 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
bf48648d 835 csum_info->receive.ip_checksum_succeeded &&
505e3f00
APM		 836 skb->protocol == htons(ETH_P_IP)) {
837 /* Check that there is enough space to hold the IP header. */
838 if (skb_headlen(skb) < sizeof(struct iphdr)) {
839 kfree_skb(skb);
840 return NULL;
841 }
bf48648d 842 netvsc_comp_ipcsum(skb);
505e3f00 843 }
bf48648d 844
df9f540c 845 /* Do L4 checksum offload if enabled and present. */
0ba35fe9 846 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
e52fed71
SH		 847 if (csum_info->receive.tcp_checksum_succeeded ||
848 csum_info->receive.udp_checksum_succeeded)
e3d605ed 849 skb->ip_summed = CHECKSUM_UNNECESSARY;
e3d605ed
KS		 850 }
851
0ba35fe9 852 if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
1fac7ca4
SH		 853 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
854
0ba35fe9 855 if (ppi_flags & NVSC_RSC_VLAN) {
98ba780e
MM		 856 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
857 (vlan->cfi ? VLAN_CFI_MASK : 0);
dc54a08c 858
93725cbd 859 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
760d1e36 860 vlan_tci);
dc54a08c 861 }
fceaf24a 862
84bf9cef
KS		 863 return skb;
864}
865
866/*
867 * netvsc_recv_callback - Callback when we receive a packet from the
868 * "wire" on the specified device.
869 */
dc54a08c 870int netvsc_recv_callback(struct net_device *net,
345ac089 871 struct netvsc_device *net_device,
c8e4eff4 872 struct netvsc_channel *nvchan)
84bf9cef 873{
3d541ac5 874 struct net_device_context *net_device_ctx = netdev_priv(net);
c8e4eff4 875 struct vmbus_channel *channel = nvchan->channel;
742fe54c 876 u16 q_idx = channel->offermsg.offer.sub_channel_index;
84bf9cef 877 struct sk_buff *skb;
1cb9d3b6 878 struct netvsc_stats_rx *rx_stats = &nvchan->rx_stats;
351e1581
HZ		 879 struct xdp_buff xdp;
880 u32 act;
84bf9cef 881
9cbcc428 882 if (net->reg_state != NETREG_REGISTERED)
84bf9cef
KS		 883 return NVSP_STAT_FAIL;
884
351e1581
HZ		 885 act = netvsc_run_xdp(net, nvchan, &xdp);
886
1cb9d3b6
HZ		 887 if (act == XDP_REDIRECT)
888 return NVSP_STAT_SUCCESS;
889
351e1581
HZ		 890 if (act != XDP_PASS && act != XDP_TX) {
891 u64_stats_update_begin(&rx_stats->syncp);
892 rx_stats->xdp_drop++;
893 u64_stats_update_end(&rx_stats->syncp);
894
895 return NVSP_STAT_SUCCESS; /* consumed by XDP */
896 }
897
84bf9cef 898 /* Allocate a skb - TODO direct I/O to pages? */
351e1581 899 skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
c8e4eff4 900
84bf9cef 901 if (unlikely(!skb)) {
f61a9d62 902 ++net_device_ctx->eth_stats.rx_no_memory;
84bf9cef
KS		 903 return NVSP_STAT_FAIL;
904 }
5b54dac8 905
0c195567 906 skb_record_rx_queue(skb, q_idx);
9cbcc428
SH		 907
908 /*
909 * Even if injecting the packet, record the statistics
910 * on the synthetic device because modifying the VF device
911 * statistics will not work correctly.
912 */
4b02b58b 913 u64_stats_update_begin(&rx_stats->syncp);
1cb9d3b6
HZ		 914 if (act == XDP_TX)
915 rx_stats->xdp_tx++;
916
7eafd9b4 917 rx_stats->packets++;
c8e4eff4 918 rx_stats->bytes += nvchan->rsc.pktlen;
f7ad75b7
SH		 919
920 if (skb->pkt_type == PACKET_BROADCAST)
921 ++rx_stats->broadcast;
922 else if (skb->pkt_type == PACKET_MULTICAST)
923 ++rx_stats->multicast;
4b02b58b 924 u64_stats_update_end(&rx_stats->syncp);
9495c282 925
351e1581
HZ		 926 if (act == XDP_TX) {
927 netvsc_xdp_xmit(skb, net);
928 return NVSP_STAT_SUCCESS;
929 }
930
742fe54c 931 napi_gro_receive(&nvchan->napi, skb);
5c71dadb 932 return NVSP_STAT_SUCCESS;
fceaf24a
HJ		 933}
934
f82f4ad7
SH		 935static void netvsc_get_drvinfo(struct net_device *net,
936 struct ethtool_drvinfo *info)
937{
fb3ceec1
WS		 938 strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
939 strscpy(info->fw_version, "N/A", sizeof(info->fw_version));
f82f4ad7
SH		 940}
941
59995370
AS		 942static void netvsc_get_channels(struct net_device *net,
943 struct ethtool_channels *channel)
944{
945 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 946 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
59995370
AS		 947
948 if (nvdev) {
949 channel->max_combined = nvdev->max_chn;
950 channel->combined_count = nvdev->num_chn;
951 }
952}
953
7c9f335a
HZ		 954/* Alloc struct netvsc_device_info, and initialize it from either existing
955 * struct netvsc_device, or from default values.
956 */
351e1581
HZ		 957static
958struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
7c9f335a
HZ		 959{
960 struct netvsc_device_info *dev_info;
351e1581 961 struct bpf_prog *prog;
7c9f335a
HZ		 962
963 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
964
965 if (!dev_info)
966 return NULL;
967
968 if (nvdev) {
351e1581
HZ		 969 ASSERT_RTNL();
970
7c9f335a
HZ		 971 dev_info->num_chn = nvdev->num_chn;
972 dev_info->send_sections = nvdev->send_section_cnt;
973 dev_info->send_section_size = nvdev->send_section_size;
974 dev_info->recv_sections = nvdev->recv_section_cnt;
975 dev_info->recv_section_size = nvdev->recv_section_size;
17d91256
HZ		 976
977 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
978 NETVSC_HASH_KEYLEN);
351e1581
HZ		 979
980 prog = netvsc_xdp_get(nvdev);
981 if (prog) {
982 bpf_prog_inc(prog);
983 dev_info->bprog = prog;
984 }
7c9f335a
HZ		 985 } else {
986 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
987 dev_info->send_sections = NETVSC_DEFAULT_TX;
988 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
989 dev_info->recv_sections = NETVSC_DEFAULT_RX;
990 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
991 }
992
993 return dev_info;
994}
995
351e1581
HZ		 996/* Free struct netvsc_device_info */
997static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
998{
999 if (dev_info->bprog) {
1000 ASSERT_RTNL();
1001 bpf_prog_put(dev_info->bprog);
1002 }
1003
1004 kfree(dev_info);
1005}
1006
7b2ee50c
SH		 1007static int netvsc_detach(struct net_device *ndev,
1008 struct netvsc_device *nvdev)
1009{
1010 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1011 struct hv_device *hdev = ndev_ctx->device_ctx;
1012 int ret;
1013
1014 /* Don't try continuing to try and setup sub channels */
1015 if (cancel_work_sync(&nvdev->subchan_work))
1016 nvdev->num_chn = 1;
1017
351e1581
HZ		 1018 netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1019
7b2ee50c
SH		 1020 /* If device was up (receiving) then shutdown */
1021 if (netif_running(ndev)) {
1b704c4a 1022 netvsc_tx_disable(nvdev, ndev);
7b2ee50c
SH		 1023
1024 ret = rndis_filter_close(nvdev);
1025 if (ret) {
1026 netdev_err(ndev,
1027 "unable to close device (ret %d).\n", ret);
1028 return ret;
1029 }
1030
1031 ret = netvsc_wait_until_empty(nvdev);
1032 if (ret) {
1033 netdev_err(ndev,
1034 "Ring buffer not empty after closing rndis\n");
1035 return ret;
1036 }
1037 }
1038
1039 netif_device_detach(ndev);
1040
1041 rndis_filter_device_remove(hdev, nvdev);
1042
1043 return 0;
1044}
1045
1046static int netvsc_attach(struct net_device *ndev,
1047 struct netvsc_device_info *dev_info)
1048{
1049 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1050 struct hv_device *hdev = ndev_ctx->device_ctx;
1051 struct netvsc_device *nvdev;
1052 struct rndis_device *rdev;
351e1581
HZ		 1053 struct bpf_prog *prog;
1054 int ret = 0;
7b2ee50c
SH		 1055
1056 nvdev = rndis_filter_device_add(hdev, dev_info);
1057 if (IS_ERR(nvdev))
1058 return PTR_ERR(nvdev);
1059
3ffe64f1 1060 if (nvdev->num_chn > 1) {
17d91256 1061 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
3ffe64f1
SH		 1062
1063 /* if unavailable, just proceed with one queue */
1064 if (ret) {
1065 nvdev->max_chn = 1;
1066 nvdev->num_chn = 1;
1067 }
1068 }
1069
351e1581
HZ		 1070 prog = dev_info->bprog;
1071 if (prog) {
184367dc 1072 bpf_prog_inc(prog);
351e1581 1073 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
184367dc
HZ		 1074 if (ret) {
1075 bpf_prog_put(prog);
351e1581 1076 goto err1;
184367dc 1077 }
351e1581
HZ		 1078 }
1079
3ffe64f1 1080 /* In any case device is now ready */
f6f13c12 1081 nvdev->tx_disable = false;
3ffe64f1 1082 netif_device_attach(ndev);
7b2ee50c 1083
3ffe64f1 1084 /* Note: enable and attach happen when sub-channels setup */
7b2ee50c
SH		 1085 netif_carrier_off(ndev);
1086
1087 if (netif_running(ndev)) {
1088 ret = rndis_filter_open(nvdev);
1089 if (ret)
351e1581 1090 goto err2;
7b2ee50c
SH		 1091
1092 rdev = nvdev->extension;
1093 if (!rdev->link_state)
1094 netif_carrier_on(ndev);
1095 }
1096
1097 return 0;
719b85c3 1098
351e1581 1099err2:
719b85c3
HZ		 1100 netif_device_detach(ndev);
1101
351e1581 1102err1:
719b85c3
HZ		 1103 rndis_filter_device_remove(hdev, nvdev);
1104
1105 return ret;
7b2ee50c
SH		 1106}
1107
b5960e6e
AS		 1108static int netvsc_set_channels(struct net_device *net,
1109 struct ethtool_channels *channels)
1110{
1111 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 1112 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
7ca45933 1113 unsigned int orig, count = channels->combined_count;
7c9f335a 1114 struct netvsc_device_info *device_info;
7b2ee50c 1115 int ret;
2b01888d 1116
1117 /* We do not support separate count for rx, tx, or other */
1118 if (count == 0 ||
1119 channels->rx_count || channels->tx_count || channels->other_count)
1120 return -EINVAL;
1121
a0be450e 1122 if (!nvdev || nvdev->destroy)
b5960e6e
AS		 1123 return -ENODEV;
1124
2b01888d 1125 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
b5960e6e 1126 return -EINVAL;
b5960e6e 1127
2b01888d 1128 if (count > nvdev->max_chn)
b5960e6e
AS		 1129 return -EINVAL;
1130
7ca45933 1131 orig = nvdev->num_chn;
b5960e6e 1132
7c9f335a
HZ		 1133 device_info = netvsc_devinfo_get(nvdev);
1134
1135 if (!device_info)
1136 return -ENOMEM;
1137
1138 device_info->num_chn = count;
8b532797 1139
7b2ee50c
SH		 1140 ret = netvsc_detach(net, nvdev);
1141 if (ret)
7c9f335a 1142 goto out;
7ca45933 1143
7c9f335a 1144 ret = netvsc_attach(net, device_info);
7b2ee50c 1145 if (ret) {
7c9f335a
HZ		 1146 device_info->num_chn = orig;
1147 if (netvsc_attach(net, device_info))
7b2ee50c 1148 netdev_err(net, "restoring channel setting failed\n");
7ca45933 1149 }
b5960e6e 1150
7c9f335a 1151out:
351e1581 1152 netvsc_devinfo_put(device_info);
b5960e6e 1153 return ret;
b5960e6e
AS		 1154}
1155
49eb9389 1156static void netvsc_init_settings(struct net_device *dev)
1157{
1158 struct net_device_context *ndc = netdev_priv(dev);
1159
486e3981 1160 ndc->l4_hash = HV_DEFAULT_L4HASH;
4823eb2f 1161
49eb9389 1162 ndc->speed = SPEED_UNKNOWN;
f3c9d40e 1163 ndc->duplex = DUPLEX_FULL;
d6792a5a
HZ		 1164
1165 dev->features = NETIF_F_LRO;
49eb9389 1166}
1167
5e8456fd
PR		 1168static int netvsc_get_link_ksettings(struct net_device *dev,
1169 struct ethtool_link_ksettings *cmd)
49eb9389 1170{
1171 struct net_device_context *ndc = netdev_priv(dev);
9aedc6e2
CF		 1172 struct net_device *vf_netdev;
1173
1174 vf_netdev = rtnl_dereference(ndc->vf_netdev);
1175
1176 if (vf_netdev)
1177 return __ethtool_get_link_ksettings(vf_netdev, cmd);
49eb9389 1178
5e8456fd
PR		 1179 cmd->base.speed = ndc->speed;
1180 cmd->base.duplex = ndc->duplex;
1181 cmd->base.port = PORT_OTHER;
49eb9389 1182
1183 return 0;
1184}
1185
5e8456fd
PR		 1186static int netvsc_set_link_ksettings(struct net_device *dev,
1187 const struct ethtool_link_ksettings *cmd)
49eb9389 1188{
1189 struct net_device_context *ndc = netdev_priv(dev);
9aedc6e2 1190 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
49eb9389 1191
9aedc6e2
CF		 1192 if (vf_netdev) {
1193 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1194 return -EOPNOTSUPP;
49eb9389 1195
9aedc6e2
CF		 1196 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1197 cmd);
1198 }
49eb9389 1199
9aedc6e2
CF		 1200 return ethtool_virtdev_set_link_ksettings(dev, cmd,
1201 &ndc->speed, &ndc->duplex);
49eb9389 1202}
1203
4d447c9a
HZ		 1204static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1205{
1206 struct net_device_context *ndevctx = netdev_priv(ndev);
0c195567 1207 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
545a8e79 1208 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
9749fed5 1209 int orig_mtu = ndev->mtu;
7c9f335a 1210 struct netvsc_device_info *device_info;
9749fed5 1211 int ret = 0;
4d447c9a 1212
a0be450e 1213 if (!nvdev || nvdev->destroy)
4d447c9a
HZ		 1214 return -ENODEV;
1215
7c9f335a
HZ		 1216 device_info = netvsc_devinfo_get(nvdev);
1217
1218 if (!device_info)
1219 return -ENOMEM;
1220
0c195567 1221 /* Change MTU of underlying VF netdev first. */
1222 if (vf_netdev) {
1223 ret = dev_set_mtu(vf_netdev, mtu);
1224 if (ret)
7c9f335a 1225 goto out;
0c195567 1226 }
1227
7b2ee50c
SH		 1228 ret = netvsc_detach(ndev, nvdev);
1229 if (ret)
1230 goto rollback_vf;
4d447c9a
HZ		 1231
1232 ndev->mtu = mtu;
1233
7c9f335a
HZ		 1234 ret = netvsc_attach(ndev, device_info);
1235 if (!ret)
1236 goto out;
ea383bf1 1237
7b2ee50c
SH		 1238 /* Attempt rollback to original MTU */
1239 ndev->mtu = orig_mtu;
163891d7 1240
7c9f335a 1241 if (netvsc_attach(ndev, device_info))
7b2ee50c
SH		 1242 netdev_err(ndev, "restoring mtu failed\n");
1243rollback_vf:
1244 if (vf_netdev)
1245 dev_set_mtu(vf_netdev, orig_mtu);
1bdcec8a 1246
7c9f335a 1247out:
351e1581 1248 netvsc_devinfo_put(device_info);
9749fed5 1249 return ret;
4d447c9a
HZ		 1250}
1251
0c195567 1252static void netvsc_get_vf_stats(struct net_device *net,
1253 struct netvsc_vf_pcpu_stats *tot)
1254{
1255 struct net_device_context *ndev_ctx = netdev_priv(net);
1256 int i;
1257
1258 memset(tot, 0, sizeof(*tot));
1259
1260 for_each_possible_cpu(i) {
1261 const struct netvsc_vf_pcpu_stats *stats
1262 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1263 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1264 unsigned int start;
1265
1266 do {
068c38ad 1267 start = u64_stats_fetch_begin(&stats->syncp);
0c195567 1268 rx_packets = stats->rx_packets;
1269 tx_packets = stats->tx_packets;
1270 rx_bytes = stats->rx_bytes;
1271 tx_bytes = stats->tx_bytes;
068c38ad 1272 } while (u64_stats_fetch_retry(&stats->syncp, start));
0c195567 1273
1274 tot->rx_packets += rx_packets;
1275 tot->tx_packets += tx_packets;
1276 tot->rx_bytes += rx_bytes;
1277 tot->tx_bytes += tx_bytes;
1278 tot->tx_dropped += stats->tx_dropped;
1279 }
1280}
1281
6ae74671
YR		 1282static void netvsc_get_pcpu_stats(struct net_device *net,
1283 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1284{
1285 struct net_device_context *ndev_ctx = netdev_priv(net);
1286 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1287 int i;
1288
1289 /* fetch percpu stats of vf */
1290 for_each_possible_cpu(i) {
1291 const struct netvsc_vf_pcpu_stats *stats =
1292 per_cpu_ptr(ndev_ctx->vf_stats, i);
1293 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1294 unsigned int start;
1295
1296 do {
068c38ad 1297 start = u64_stats_fetch_begin(&stats->syncp);
6ae74671
YR		 1298 this_tot->vf_rx_packets = stats->rx_packets;
1299 this_tot->vf_tx_packets = stats->tx_packets;
1300 this_tot->vf_rx_bytes = stats->rx_bytes;
1301 this_tot->vf_tx_bytes = stats->tx_bytes;
068c38ad 1302 } while (u64_stats_fetch_retry(&stats->syncp, start));
6ae74671
YR		 1303 this_tot->rx_packets = this_tot->vf_rx_packets;
1304 this_tot->tx_packets = this_tot->vf_tx_packets;
1305 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1306 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1307 }
1308
1309 /* fetch percpu stats of netvsc */
1310 for (i = 0; i < nvdev->num_chn; i++) {
1311 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1cb9d3b6
HZ		 1312 const struct netvsc_stats_tx *tx_stats;
1313 const struct netvsc_stats_rx *rx_stats;
6ae74671
YR		 1314 struct netvsc_ethtool_pcpu_stats *this_tot =
1315 &pcpu_tot[nvchan->channel->target_cpu];
1316 u64 packets, bytes;
1317 unsigned int start;
1318
1cb9d3b6 1319 tx_stats = &nvchan->tx_stats;
6ae74671 1320 do {
068c38ad 1321 start = u64_stats_fetch_begin(&tx_stats->syncp);
1cb9d3b6
HZ		 1322 packets = tx_stats->packets;
1323 bytes = tx_stats->bytes;
068c38ad 1324 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
6ae74671
YR		 1325
1326 this_tot->tx_bytes += bytes;
1327 this_tot->tx_packets += packets;
1328
1cb9d3b6 1329 rx_stats = &nvchan->rx_stats;
6ae74671 1330 do {
068c38ad 1331 start = u64_stats_fetch_begin(&rx_stats->syncp);
1cb9d3b6
HZ		 1332 packets = rx_stats->packets;
1333 bytes = rx_stats->bytes;
068c38ad 1334 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
6ae74671
YR		 1335
1336 this_tot->rx_bytes += bytes;
1337 this_tot->rx_packets += packets;
1338 }
1339}
1340
bc1f4470 1341static void netvsc_get_stats64(struct net_device *net,
1342 struct rtnl_link_stats64 *t)
7eafd9b4 1343{
1344 struct net_device_context *ndev_ctx = netdev_priv(net);
6d0d779d 1345 struct netvsc_device *nvdev;
0c195567 1346 struct netvsc_vf_pcpu_stats vf_tot;
89bb42b1 1347 int i;
6c80f3fc 1348
6d0d779d
DC		 1349 rcu_read_lock();
1350
1351 nvdev = rcu_dereference(ndev_ctx->nvdev);
6c80f3fc 1352 if (!nvdev)
6d0d779d 1353 goto out;
6c80f3fc 1354
0c195567 1355 netdev_stats_to_stats64(t, &net->stats);
1356
1357 netvsc_get_vf_stats(net, &vf_tot);
1358 t->rx_packets += vf_tot.rx_packets;
1359 t->tx_packets += vf_tot.tx_packets;
1360 t->rx_bytes += vf_tot.rx_bytes;
1361 t->tx_bytes += vf_tot.tx_bytes;
1362 t->tx_dropped += vf_tot.tx_dropped;
1363
6c80f3fc
SX		 1364 for (i = 0; i < nvdev->num_chn; i++) {
1365 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1cb9d3b6
HZ		 1366 const struct netvsc_stats_tx *tx_stats;
1367 const struct netvsc_stats_rx *rx_stats;
6c80f3fc 1368 u64 packets, bytes, multicast;
7eafd9b4 1369 unsigned int start;
1370
1cb9d3b6 1371 tx_stats = &nvchan->tx_stats;
7eafd9b4 1372 do {
068c38ad 1373 start = u64_stats_fetch_begin(&tx_stats->syncp);
1cb9d3b6
HZ		 1374 packets = tx_stats->packets;
1375 bytes = tx_stats->bytes;
068c38ad 1376 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
6c80f3fc
SX		 1377
1378 t->tx_bytes += bytes;
1379 t->tx_packets += packets;
7eafd9b4 1380
1cb9d3b6 1381 rx_stats = &nvchan->rx_stats;
7eafd9b4 1382 do {
068c38ad 1383 start = u64_stats_fetch_begin(&rx_stats->syncp);
1cb9d3b6
HZ		 1384 packets = rx_stats->packets;
1385 bytes = rx_stats->bytes;
1386 multicast = rx_stats->multicast + rx_stats->broadcast;
068c38ad 1387 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
6c80f3fc
SX		 1388
1389 t->rx_bytes += bytes;
1390 t->rx_packets += packets;
1391 t->multicast += multicast;
7eafd9b4 1392 }
6d0d779d
DC		 1393out:
1394 rcu_read_unlock();
7eafd9b4 1395}
1ce09e89
HZ		 1396
1397static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1398{
867047c4 1399 struct net_device_context *ndc = netdev_priv(ndev);
16ba3266 1400 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
867047c4 1401 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1ce09e89 1402 struct sockaddr *addr = p;
1ce09e89
HZ		 1403 int err;
1404
16ba3266 1405 err = eth_prepare_mac_addr_change(ndev, p);
1406 if (err)
1ce09e89
HZ		 1407 return err;
1408
867047c4 1409 if (!nvdev)
1410 return -ENODEV;
1411
16ba3266 1412 if (vf_netdev) {
3a37a963 1413 err = dev_set_mac_address(vf_netdev, addr, NULL);
16ba3266 1414 if (err)
1415 return err;
1416 }
1417
867047c4 1418 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
16ba3266 1419 if (!err) {
1420 eth_commit_mac_addr_change(ndev, p);
1421 } else if (vf_netdev) {
1422 /* rollback change on VF */
1423 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
3a37a963 1424 dev_set_mac_address(vf_netdev, addr, NULL);
1ce09e89
HZ		 1425 }
1426
1427 return err;
1428}
1429
4323b47c
SH		 1430static const struct {
1431 char name[ETH_GSTRING_LEN];
1432 u16 offset;
1433} netvsc_stats[] = {
1434 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
f61a9d62 1435 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
4323b47c
SH		 1436 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1437 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1438 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
cad5c197 1439 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1440 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
f61a9d62 1441 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
09af87d1
SX		 1442 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1443 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
fdd8fac4 1444 { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
6ae74671
YR		 1445}, pcpu_stats[] = {
1446 { "cpu%u_rx_packets",
1447 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1448 { "cpu%u_rx_bytes",
1449 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1450 { "cpu%u_tx_packets",
1451 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1452 { "cpu%u_tx_bytes",
1453 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1454 { "cpu%u_vf_rx_packets",
1455 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1456 { "cpu%u_vf_rx_bytes",
1457 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1458 { "cpu%u_vf_tx_packets",
1459 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1460 { "cpu%u_vf_tx_bytes",
1461 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
0c195567 1462}, vf_stats[] = {
1463 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1464 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1465 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1466 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1467 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
4323b47c
SH		 1468};
1469
6c80f3fc 1470#define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
0c195567 1471#define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
6c80f3fc 1472
6ae74671
YR		 1473/* statistics per queue (rx/tx packets/bytes) */
1474#define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1475
1cb9d3b6
HZ		 1476/* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1477#define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
6c80f3fc 1478
4323b47c
SH		 1479static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1480{
6c80f3fc 1481 struct net_device_context *ndc = netdev_priv(dev);
fbd4c7e7 1482 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1483
1484 if (!nvdev)
1485 return -ENODEV;
6c80f3fc 1486
4323b47c
SH		 1487 switch (string_set) {
1488 case ETH_SS_STATS:
0c195567 1489 return NETVSC_GLOBAL_STATS_LEN
1490 + NETVSC_VF_STATS_LEN
6ae74671
YR		 1491 + NETVSC_QUEUE_STATS_LEN(nvdev)
1492 + NETVSC_PCPU_STATS_LEN;
4323b47c
SH		 1493 default:
1494 return -EINVAL;
1495 }
1496}
1497
1498static void netvsc_get_ethtool_stats(struct net_device *dev,
1499 struct ethtool_stats *stats, u64 *data)
1500{
1501 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1502 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
4323b47c 1503 const void *nds = &ndc->eth_stats;
1cb9d3b6
HZ		 1504 const struct netvsc_stats_tx *tx_stats;
1505 const struct netvsc_stats_rx *rx_stats;
0c195567 1506 struct netvsc_vf_pcpu_stats sum;
6ae74671 1507 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
6c80f3fc
SX		 1508 unsigned int start;
1509 u64 packets, bytes;
351e1581 1510 u64 xdp_drop;
1cb9d3b6
HZ		 1511 u64 xdp_redirect;
1512 u64 xdp_tx;
1513 u64 xdp_xmit;
6ae74671 1514 int i, j, cpu;
4323b47c 1515
545a8e79 1516 if (!nvdev)
1517 return;
1518
6c80f3fc 1519 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
4323b47c 1520 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
6c80f3fc 1521
0c195567 1522 netvsc_get_vf_stats(dev, &sum);
1523 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1524 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1525
6c80f3fc 1526 for (j = 0; j < nvdev->num_chn; j++) {
1cb9d3b6 1527 tx_stats = &nvdev->chan_table[j].tx_stats;
6c80f3fc
SX		 1528
1529 do {
068c38ad 1530 start = u64_stats_fetch_begin(&tx_stats->syncp);
1cb9d3b6
HZ		 1531 packets = tx_stats->packets;
1532 bytes = tx_stats->bytes;
1533 xdp_xmit = tx_stats->xdp_xmit;
068c38ad 1534 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
6c80f3fc
SX		 1535 data[i++] = packets;
1536 data[i++] = bytes;
1cb9d3b6 1537 data[i++] = xdp_xmit;
6c80f3fc 1538
1cb9d3b6 1539 rx_stats = &nvdev->chan_table[j].rx_stats;
6c80f3fc 1540 do {
068c38ad 1541 start = u64_stats_fetch_begin(&rx_stats->syncp);
1cb9d3b6
HZ		 1542 packets = rx_stats->packets;
1543 bytes = rx_stats->bytes;
1544 xdp_drop = rx_stats->xdp_drop;
1545 xdp_redirect = rx_stats->xdp_redirect;
1546 xdp_tx = rx_stats->xdp_tx;
068c38ad 1547 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
6c80f3fc
SX		 1548 data[i++] = packets;
1549 data[i++] = bytes;
351e1581 1550 data[i++] = xdp_drop;
1cb9d3b6
HZ		 1551 data[i++] = xdp_redirect;
1552 data[i++] = xdp_tx;
6c80f3fc 1553 }
6ae74671
YR		 1554
1555 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1556 sizeof(struct netvsc_ethtool_pcpu_stats),
1557 GFP_KERNEL);
886e44c9
JJ		 1558 if (!pcpu_sum)
1559 return;
1560
6ae74671
YR		 1561 netvsc_get_pcpu_stats(dev, pcpu_sum);
1562 for_each_present_cpu(cpu) {
1563 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1564
1565 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1566 data[i++] = *(u64 *)((void *)this_sum
1567 + pcpu_stats[j].offset);
1568 }
1569 kvfree(pcpu_sum);
4323b47c
SH		 1570}
1571
1572static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1573{
6c80f3fc 1574 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1575 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
6c80f3fc 1576 u8 *p = data;
6ae74671 1577 int i, cpu;
4323b47c 1578
545a8e79 1579 if (!nvdev)
1580 return;
1581
4323b47c
SH		 1582 switch (stringset) {
1583 case ETH_SS_STATS:
3ae0ed37
AD		 1584 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1585 ethtool_sprintf(&p, netvsc_stats[i].name);
0c195567 1586
3ae0ed37
AD		 1587 for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1588 ethtool_sprintf(&p, vf_stats[i].name);
6c80f3fc 1589
6c80f3fc 1590 for (i = 0; i < nvdev->num_chn; i++) {
3ae0ed37
AD		 1591 ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1592 ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1cb9d3b6 1593 ethtool_sprintf(&p, "tx_queue_%u_xdp_xmit", i);
3ae0ed37
AD		 1594 ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1595 ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1596 ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1cb9d3b6
HZ		 1597 ethtool_sprintf(&p, "rx_queue_%u_xdp_redirect", i);
1598 ethtool_sprintf(&p, "rx_queue_%u_xdp_tx", i);
6c80f3fc
SX		 1599 }
1600
6ae74671 1601 for_each_present_cpu(cpu) {
3ae0ed37
AD		 1602 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1603 ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
6ae74671
YR		 1604 }
1605
4323b47c
SH		 1606 break;
1607 }
1608}
1609
b5a5dc8d 1610static int
4823eb2f
HZ		 1611netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1612 struct ethtool_rxnfc *info)
b5a5dc8d 1613{
486e3981
HZ		 1614 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1615
b5a5dc8d 1616 info->data = RXH_IP_SRC | RXH_IP_DST;
1617
1618 switch (info->flow_type) {
1619 case TCP_V4_FLOW:
0518ec4f
HZ		 1620 if (ndc->l4_hash & HV_TCP4_L4HASH)
1621 info->data |= l4_flag;
1622
1623 break;
1624
b5a5dc8d 1625 case TCP_V6_FLOW:
0518ec4f
HZ		 1626 if (ndc->l4_hash & HV_TCP6_L4HASH)
1627 info->data |= l4_flag;
1628
4823eb2f
HZ		 1629 break;
1630
b5a5dc8d 1631 case UDP_V4_FLOW:
486e3981
HZ		 1632 if (ndc->l4_hash & HV_UDP4_L4HASH)
1633 info->data |= l4_flag;
4823eb2f
HZ		 1634
1635 break;
1636
b5a5dc8d 1637 case UDP_V6_FLOW:
486e3981
HZ		 1638 if (ndc->l4_hash & HV_UDP6_L4HASH)
1639 info->data |= l4_flag;
4823eb2f
HZ		 1640
1641 break;
1642
b5a5dc8d 1643 case IPV4_FLOW:
1644 case IPV6_FLOW:
1645 break;
1646 default:
1647 info->data = 0;
1648 break;
1649 }
1650
1651 return 0;
1652}
1653
b448f4e8 1654static int
1655netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1656 u32 *rules)
1657{
1658 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1659 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1660
1661 if (!nvdev)
1662 return -ENODEV;
b448f4e8 1663
1664 switch (info->cmd) {
1665 case ETHTOOL_GRXRINGS:
1666 info->data = nvdev->num_chn;
1667 return 0;
b5a5dc8d 1668
1669 case ETHTOOL_GRXFH:
4823eb2f 1670 return netvsc_get_rss_hash_opts(ndc, info);
b448f4e8 1671 }
1672 return -EOPNOTSUPP;
1673}
1674
4823eb2f
HZ		 1675static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1676 struct ethtool_rxnfc *info)
1677{
1678 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1679 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
486e3981 1680 switch (info->flow_type) {
0518ec4f
HZ		 1681 case TCP_V4_FLOW:
1682 ndc->l4_hash |= HV_TCP4_L4HASH;
1683 break;
1684
1685 case TCP_V6_FLOW:
1686 ndc->l4_hash |= HV_TCP6_L4HASH;
1687 break;
1688
486e3981
HZ		 1689 case UDP_V4_FLOW:
1690 ndc->l4_hash |= HV_UDP4_L4HASH;
1691 break;
1692
1693 case UDP_V6_FLOW:
1694 ndc->l4_hash |= HV_UDP6_L4HASH;
1695 break;
1696
1697 default:
4823eb2f 1698 return -EOPNOTSUPP;
486e3981 1699 }
4823eb2f
HZ		 1700
1701 return 0;
1702 }
1703
1704 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
486e3981 1705 switch (info->flow_type) {
0518ec4f
HZ		 1706 case TCP_V4_FLOW:
1707 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1708 break;
1709
1710 case TCP_V6_FLOW:
1711 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1712 break;
1713
486e3981
HZ		 1714 case UDP_V4_FLOW:
1715 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1716 break;
1717
1718 case UDP_V6_FLOW:
1719 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1720 break;
1721
1722 default:
4823eb2f 1723 return -EOPNOTSUPP;
486e3981 1724 }
4823eb2f
HZ		 1725
1726 return 0;
1727 }
1728
1729 return -EOPNOTSUPP;
1730}
1731
1732static int
1733netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1734{
1735 struct net_device_context *ndc = netdev_priv(ndev);
1736
1737 if (info->cmd == ETHTOOL_SRXFH)
1738 return netvsc_set_rss_hash_opts(ndc, info);
1739
1740 return -EOPNOTSUPP;
1741}
1742
962f3fee 1743static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1744{
1745 return NETVSC_HASH_KEYLEN;
1746}
1747
1748static u32 netvsc_rss_indir_size(struct net_device *dev)
1749{
4cab498f
SG		 1750 struct net_device_context *ndc = netdev_priv(dev);
1751
1752 return ndc->rx_table_sz;
962f3fee 1753}
1754
1755static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1756 u8 *hfunc)
1757{
1758 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1759 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1760 struct rndis_device *rndis_dev;
ff4a4419 1761 int i;
962f3fee 1762
545a8e79 1763 if (!ndev)
1764 return -ENODEV;
1765
962f3fee 1766 if (hfunc)
1767 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1768
eb996edb 1769 rndis_dev = ndev->extension;
ff4a4419 1770 if (indir) {
4cab498f 1771 for (i = 0; i < ndc->rx_table_sz; i++)
b0689faa 1772 indir[i] = ndc->rx_table[i];
ff4a4419 1773 }
1774
962f3fee 1775 if (key)
1776 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1777
1778 return 0;
1779}
1780
1781static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1782 const u8 *key, const u8 hfunc)
1783{
1784 struct net_device_context *ndc = netdev_priv(dev);
545a8e79 1785 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1786 struct rndis_device *rndis_dev;
ff4a4419 1787 int i;
962f3fee 1788
545a8e79 1789 if (!ndev)
1790 return -ENODEV;
1791
962f3fee 1792 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1793 return -EOPNOTSUPP;
1794
eb996edb 1795 rndis_dev = ndev->extension;
ff4a4419 1796 if (indir) {
4cab498f 1797 for (i = 0; i < ndc->rx_table_sz; i++)
db3cd7af 1798 if (indir[i] >= ndev->num_chn)
ff4a4419 1799 return -EINVAL;
1800
4cab498f 1801 for (i = 0; i < ndc->rx_table_sz; i++)
b0689faa 1802 ndc->rx_table[i] = indir[i];
ff4a4419 1803 }
1804
1805 if (!key) {
1806 if (!indir)
1807 return 0;
1808
1809 key = rndis_dev->rss_key;
1810 }
962f3fee 1811
715e2ec5 1812 return rndis_filter_set_rss_param(rndis_dev, key);
962f3fee 1813}
1814
8b532797 1815/* Hyper-V RNDIS protocol does not have ring in the HW sense.
1816 * It does have pre-allocated receive area which is divided into sections.
1817 */
1818static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1819 struct ethtool_ringparam *ring)
1820{
1821 u32 max_buf_size;
1822
1823 ring->rx_pending = nvdev->recv_section_cnt;
1824 ring->tx_pending = nvdev->send_section_cnt;
1825
1826 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1827 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1828 else
1829 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1830
1831 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1832 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1833 / nvdev->send_section_size;
1834}
1835
1836static void netvsc_get_ringparam(struct net_device *ndev,
74624944
HC		 1837 struct ethtool_ringparam *ring,
1838 struct kernel_ethtool_ringparam *kernel_ring,
1839 struct netlink_ext_ack *extack)
8b532797 1840{
1841 struct net_device_context *ndevctx = netdev_priv(ndev);
1842 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1843
1844 if (!nvdev)
1845 return;
1846
1847 __netvsc_get_ringparam(nvdev, ring);
1848}
1849
1850static int netvsc_set_ringparam(struct net_device *ndev,
74624944
HC		 1851 struct ethtool_ringparam *ring,
1852 struct kernel_ethtool_ringparam *kernel_ring,
1853 struct netlink_ext_ack *extack)
8b532797 1854{
1855 struct net_device_context *ndevctx = netdev_priv(ndev);
1856 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
7c9f335a 1857 struct netvsc_device_info *device_info;
8b532797 1858 struct ethtool_ringparam orig;
1859 u32 new_tx, new_rx;
8b532797 1860 int ret = 0;
1861
1862 if (!nvdev || nvdev->destroy)
1863 return -ENODEV;
1864
1865 memset(&orig, 0, sizeof(orig));
1866 __netvsc_get_ringparam(nvdev, &orig);
1867
1868 new_tx = clamp_t(u32, ring->tx_pending,
1869 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1870 new_rx = clamp_t(u32, ring->rx_pending,
1871 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1872
1873 if (new_tx == orig.tx_pending &&
1874 new_rx == orig.rx_pending)
1875 return 0; /* no change */
1876
7c9f335a
HZ		 1877 device_info = netvsc_devinfo_get(nvdev);
1878
1879 if (!device_info)
1880 return -ENOMEM;
1881
1882 device_info->send_sections = new_tx;
1883 device_info->recv_sections = new_rx;
8b532797 1884
7b2ee50c
SH		 1885 ret = netvsc_detach(ndev, nvdev);
1886 if (ret)
7c9f335a 1887 goto out;
8b532797 1888
7c9f335a 1889 ret = netvsc_attach(ndev, device_info);
7b2ee50c 1890 if (ret) {
7c9f335a
HZ		 1891 device_info->send_sections = orig.tx_pending;
1892 device_info->recv_sections = orig.rx_pending;
8b532797 1893
7c9f335a 1894 if (netvsc_attach(ndev, device_info))
7b2ee50c
SH		 1895 netdev_err(ndev, "restoring ringparam failed");
1896 }
8b532797 1897
7c9f335a 1898out:
351e1581 1899 netvsc_devinfo_put(device_info);
8b532797 1900 return ret;
1901}
1902
351e1581
HZ		 1903static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1904 netdev_features_t features)
1905{
1906 struct net_device_context *ndevctx = netdev_priv(ndev);
1907 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1908
1909 if (!nvdev || nvdev->destroy)
1910 return features;
1911
1912 if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1913 features ^= NETIF_F_LRO;
1914 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1915 }
1916
1917 return features;
1918}
1919
d6792a5a
HZ		 1920static int netvsc_set_features(struct net_device *ndev,
1921 netdev_features_t features)
1922{
1923 netdev_features_t change = features ^ ndev->features;
1924 struct net_device_context *ndevctx = netdev_priv(ndev);
1925 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
68622d07 1926 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
d6792a5a 1927 struct ndis_offload_params offloads;
68622d07 1928 int ret = 0;
d6792a5a
HZ		 1929
1930 if (!nvdev || nvdev->destroy)
1931 return -ENODEV;
1932
1933 if (!(change & NETIF_F_LRO))
68622d07 1934 goto syncvf;
d6792a5a
HZ		 1935
1936 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1937
1938 if (features & NETIF_F_LRO) {
1939 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1940 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1941 } else {
1942 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1943 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1944 }
1945
68622d07
HZ		 1946 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1947
c4509a5a 1948 if (ret) {
68622d07 1949 features ^= NETIF_F_LRO;
c4509a5a
HZ		 1950 ndev->features = features;
1951 }
68622d07
HZ		 1952
1953syncvf:
1954 if (!vf_netdev)
1955 return ret;
1956
1957 vf_netdev->wanted_features = features;
1958 netdev_update_features(vf_netdev);
1959
1960 return ret;
d6792a5a
HZ		 1961}
1962
4a062d66
CS		 1963static int netvsc_get_regs_len(struct net_device *netdev)
1964{
1965 return VRSS_SEND_TAB_SIZE * sizeof(u32);
1966}
1967
1968static void netvsc_get_regs(struct net_device *netdev,
1969 struct ethtool_regs *regs, void *p)
1970{
1971 struct net_device_context *ndc = netdev_priv(netdev);
1972 u32 *regs_buff = p;
1973
1974 /* increase the version, if buffer format is changed. */
1975 regs->version = 1;
1976
1977 memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1978}
1979
273de02a
HZ		 1980static u32 netvsc_get_msglevel(struct net_device *ndev)
1981{
1982 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1983
1984 return ndev_ctx->msg_enable;
1985}
1986
1987static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1988{
1989 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1990
1991 ndev_ctx->msg_enable = val;
1992}
1993
f82f4ad7
SH		 1994static const struct ethtool_ops ethtool_ops = {
1995 .get_drvinfo = netvsc_get_drvinfo,
4a062d66
CS		 1996 .get_regs_len = netvsc_get_regs_len,
1997 .get_regs = netvsc_get_regs,
273de02a
HZ		 1998 .get_msglevel = netvsc_get_msglevel,
1999 .set_msglevel = netvsc_set_msglevel,
f82f4ad7 2000 .get_link = ethtool_op_get_link,
4323b47c
SH		 2001 .get_ethtool_stats = netvsc_get_ethtool_stats,
2002 .get_sset_count = netvsc_get_sset_count,
2003 .get_strings = netvsc_get_strings,
59995370 2004 .get_channels = netvsc_get_channels,
b5960e6e 2005 .set_channels = netvsc_set_channels,
76d13b56 2006 .get_ts_info = ethtool_op_get_ts_info,
b448f4e8 2007 .get_rxnfc = netvsc_get_rxnfc,
4823eb2f 2008 .set_rxnfc = netvsc_set_rxnfc,
962f3fee 2009 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2010 .get_rxfh_indir_size = netvsc_rss_indir_size,
2011 .get_rxfh = netvsc_get_rxfh,
2012 .set_rxfh = netvsc_set_rxfh,
5e8456fd
PR		 2013 .get_link_ksettings = netvsc_get_link_ksettings,
2014 .set_link_ksettings = netvsc_set_link_ksettings,
8b532797 2015 .get_ringparam = netvsc_get_ringparam,
2016 .set_ringparam = netvsc_set_ringparam,
f82f4ad7
SH		 2017};
2018
df2fff28
GKH		 2019static const struct net_device_ops device_ops = {
2020 .ndo_open = netvsc_open,
2021 .ndo_stop = netvsc_close,
2022 .ndo_start_xmit = netvsc_start_xmit,
bee9d41b
SH		 2023 .ndo_change_rx_flags = netvsc_change_rx_flags,
2024 .ndo_set_rx_mode = netvsc_set_rx_mode,
351e1581 2025 .ndo_fix_features = netvsc_fix_features,
d6792a5a 2026 .ndo_set_features = netvsc_set_features,
4d447c9a 2027 .ndo_change_mtu = netvsc_change_mtu,
b681b588 2028 .ndo_validate_addr = eth_validate_addr,
1ce09e89 2029 .ndo_set_mac_address = netvsc_set_mac_addr,
5b54dac8 2030 .ndo_select_queue = netvsc_select_queue,
7eafd9b4 2031 .ndo_get_stats64 = netvsc_get_stats64,
351e1581 2032 .ndo_bpf = netvsc_bpf,
1cb9d3b6 2033 .ndo_xdp_xmit = netvsc_ndoxdp_xmit,
df2fff28
GKH		 2034};
2035
c996edcf 2036/*
27a70af3
VK		 2037 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2038 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2039 * present send GARP packet to network peers with netif_notify_peers().
c996edcf 2040 */
891de74d 2041static void netvsc_link_change(struct work_struct *w)
c996edcf 2042{
0a1275ca
VK		 2043 struct net_device_context *ndev_ctx =
2044 container_of(w, struct net_device_context, dwork.work);
2045 struct hv_device *device_obj = ndev_ctx->device_ctx;
2046 struct net_device *net = hv_get_drvdata(device_obj);
935d8a0a
LP		 2047 unsigned long flags, next_reconfig, delay;
2048 struct netvsc_reconfig *event = NULL;
2ddd5e5f 2049 struct netvsc_device *net_device;
891de74d 2050 struct rndis_device *rdev;
935d8a0a 2051 bool reschedule = false;
c996edcf 2052
9b4e946c 2053 /* if changes are happening, comeback later */
2054 if (!rtnl_trylock()) {
2055 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2056 return;
2057 }
2058
a0be450e 2059 net_device = rtnl_dereference(ndev_ctx->nvdev);
2060 if (!net_device)
1bdcec8a
VK		 2061 goto out_unlock;
2062
891de74d 2063 rdev = net_device->extension;
891de74d 2064
27a70af3
VK		 2065 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2066 if (time_is_after_jiffies(next_reconfig)) {
2067 /* link_watch only sends one notification with current state
2068 * per second, avoid doing reconfig more frequently. Handle
2069 * wrap around.
2070 */
2071 delay = next_reconfig - jiffies;
2072 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2073 schedule_delayed_work(&ndev_ctx->dwork, delay);
1bdcec8a 2074 goto out_unlock;
27a70af3
VK		 2075 }
2076 ndev_ctx->last_reconfig = jiffies;
2077
2078 spin_lock_irqsave(&ndev_ctx->lock, flags);
2079 if (!list_empty(&ndev_ctx->reconfig_events)) {
2080 event = list_first_entry(&ndev_ctx->reconfig_events,
2081 struct netvsc_reconfig, list);
2082 list_del(&event->list);
2083 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2084 }
2085 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2086
2087 if (!event)
1bdcec8a 2088 goto out_unlock;
27a70af3
VK		 2089
2090 switch (event->event) {
2091 /* Only the following events are possible due to the check in
2092 * netvsc_linkstatus_callback()
2093 */
2094 case RNDIS_STATUS_MEDIA_CONNECT:
2095 if (rdev->link_state) {
2096 rdev->link_state = false;
0c195567 2097 netif_carrier_on(net);
1b704c4a 2098 netvsc_tx_enable(net_device, net);
27a70af3 2099 } else {
935d8a0a 2100 __netdev_notify_peers(net);
27a70af3
VK		 2101 }
2102 kfree(event);
2103 break;
2104 case RNDIS_STATUS_MEDIA_DISCONNECT:
2105 if (!rdev->link_state) {
2106 rdev->link_state = true;
2107 netif_carrier_off(net);
1b704c4a 2108 netvsc_tx_disable(net_device, net);
27a70af3
VK		 2109 }
2110 kfree(event);
2111 break;
2112 case RNDIS_STATUS_NETWORK_CHANGE:
2113 /* Only makes sense if carrier is present */
2114 if (!rdev->link_state) {
2115 rdev->link_state = true;
2116 netif_carrier_off(net);
1b704c4a 2117 netvsc_tx_disable(net_device, net);
27a70af3
VK		 2118 event->event = RNDIS_STATUS_MEDIA_CONNECT;
2119 spin_lock_irqsave(&ndev_ctx->lock, flags);
15cfd407 2120 list_add(&event->list, &ndev_ctx->reconfig_events);
27a70af3
VK		 2121 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2122 reschedule = true;
3a494e71 2123 }
27a70af3 2124 break;
891de74d
HZ		 2125 }
2126
2127 rtnl_unlock();
2128
27a70af3
VK		 2129 /* link_watch only sends one notification with current state per
2130 * second, handle next reconfig event in 2 seconds.
2131 */
2132 if (reschedule)
2133 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1bdcec8a
VK		 2134
2135 return;
2136
2137out_unlock:
2138 rtnl_unlock();
c996edcf
HZ		 2139}
2140
8cde8f0c
SH		 2141static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2142{
7bf7bb37 2143 struct net_device_context *net_device_ctx;
8cde8f0c
SH		 2144 struct net_device *dev;
2145
7bf7bb37
SH		 2146 dev = netdev_master_upper_dev_get(vf_netdev);
2147 if (!dev || dev->netdev_ops != &device_ops)
2148 return NULL; /* not a netvsc device */
8cde8f0c 2149
7bf7bb37
SH		 2150 net_device_ctx = netdev_priv(dev);
2151 if (!rtnl_dereference(net_device_ctx->nvdev))
2152 return NULL; /* device is removed */
8cde8f0c 2153
7bf7bb37 2154 return dev;
8cde8f0c
SH		 2155}
2156
0c195567 2157/* Called when VF is injecting data into network stack.
2158 * Change the associated network device from VF to netvsc.
2159 * note: already called with rcu_read_lock
2160 */
2161static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2162{
2163 struct sk_buff *skb = *pskb;
2164 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2165 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2166 struct netvsc_vf_pcpu_stats *pcpu_stats
2167 = this_cpu_ptr(ndev_ctx->vf_stats);
2168
996ed047
SH		 2169 skb = skb_share_check(skb, GFP_ATOMIC);
2170 if (unlikely(!skb))
2171 return RX_HANDLER_CONSUMED;
2172
2173 *pskb = skb;
2174
0c195567 2175 skb->dev = ndev;
2176
2177 u64_stats_update_begin(&pcpu_stats->syncp);
2178 pcpu_stats->rx_packets++;
2179 pcpu_stats->rx_bytes += skb->len;
2180 u64_stats_update_end(&pcpu_stats->syncp);
2181
2182 return RX_HANDLER_ANOTHER;
2183}
2184
8cde8f0c
SH		 2185static int netvsc_vf_join(struct net_device *vf_netdev,
2186 struct net_device *ndev)
2187{
2188 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2189 int ret;
2190
2191 ret = netdev_rx_handler_register(vf_netdev,
2192 netvsc_vf_handle_frame, ndev);
2193 if (ret != 0) {
2194 netdev_err(vf_netdev,
2195 "can not register netvsc VF receive handler (err = %d)\n",
2196 ret);
2197 goto rx_handler_failed;
2198 }
2199
2200 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2201 NULL, NULL, NULL);
2202 if (ret != 0) {
2203 netdev_err(vf_netdev,
2204 "can not set master device %s (err = %d)\n",
2205 ndev->name, ret);
2206 goto upper_link_failed;
2207 }
2208
2209 /* set slave flag before open to prevent IPv6 addrconf */
2210 vf_netdev->flags |= IFF_SLAVE;
2211
2212 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2213
2214 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2215
2216 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2217 return 0;
2218
2219upper_link_failed:
2220 netdev_rx_handler_unregister(vf_netdev);
2221rx_handler_failed:
2222 return ret;
2223}
2224
0c195567 2225static void __netvsc_vf_setup(struct net_device *ndev,
2226 struct net_device *vf_netdev)
2227{
2228 int ret;
2229
0c195567 2230 /* Align MTU of VF with master */
2231 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2232 if (ret)
2233 netdev_warn(vf_netdev,
2234 "unable to change mtu to %u\n", ndev->mtu);
2235
bee9d41b 2236 /* set multicast etc flags on VF */
567c5e13 2237 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
b0dee791
SH		 2238
2239 /* sync address list from ndev to VF */
2240 netif_addr_lock_bh(ndev);
bee9d41b
SH		 2241 dev_uc_sync(vf_netdev, ndev);
2242 dev_mc_sync(vf_netdev, ndev);
b0dee791 2243 netif_addr_unlock_bh(ndev);
bee9d41b 2244
0c195567 2245 if (netif_running(ndev)) {
00f54e68 2246 ret = dev_open(vf_netdev, NULL);
0c195567 2247 if (ret)
2248 netdev_warn(vf_netdev,
2249 "unable to open: %d\n", ret);
2250 }
2251}
2252
2253/* Setup VF as slave of the synthetic device.
2254 * Runs in workqueue to avoid recursion in netlink callbacks.
2255 */
2256static void netvsc_vf_setup(struct work_struct *w)
2257{
2258 struct net_device_context *ndev_ctx
6123c668 2259 = container_of(w, struct net_device_context, vf_takeover.work);
0c195567 2260 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2261 struct net_device *vf_netdev;
2262
fb84af8a 2263 if (!rtnl_trylock()) {
6123c668 2264 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
fb84af8a 2265 return;
2266 }
2267
0c195567 2268 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2269 if (vf_netdev)
2270 __netvsc_vf_setup(ndev, vf_netdev);
2271
2272 rtnl_unlock();
2273}
2274
00547955
HZ		 2275/* Find netvsc by VF serial number.
2276 * The PCI hyperv controller records the serial number as the slot kobj name.
00d7ddba
SH		 2277 */
2278static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2279{
2280 struct device *parent = vf_netdev->dev.parent;
2281 struct net_device_context *ndev_ctx;
64ff412a 2282 struct net_device *ndev;
00d7ddba 2283 struct pci_dev *pdev;
00547955 2284 u32 serial;
00d7ddba
SH		 2285
2286 if (!parent || !dev_is_pci(parent))
2287 return NULL; /* not a PCI device */
2288
2289 pdev = to_pci_dev(parent);
2290 if (!pdev->slot) {
2291 netdev_notice(vf_netdev, "no PCI slot information\n");
2292 return NULL;
2293 }
2294
00547955
HZ		 2295 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2296 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2297 pci_slot_name(pdev->slot));
2298 return NULL;
2299 }
2300
00d7ddba
SH		 2301 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2302 if (!ndev_ctx->vf_alloc)
2303 continue;
2304
64ff412a
DC		 2305 if (ndev_ctx->vf_serial != serial)
2306 continue;
2307
2308 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2309 if (ndev->addr_len != vf_netdev->addr_len ||
2310 memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2311 ndev->addr_len) != 0)
2312 continue;
2313
2314 return ndev;
2315
00d7ddba
SH		 2316 }
2317
365e1ece
GK		 2318 /* Fallback path to check synthetic vf with
2319 * help of mac addr
2320 */
2321 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2322 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2323 if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr)) {
2324 netdev_notice(vf_netdev,
2325 "falling back to mac addr based matching\n");
2326 return ndev;
2327 }
2328 }
2329
00d7ddba 2330 netdev_notice(vf_netdev,
00547955 2331 "no netdev found for vf serial:%u\n", serial);
00d7ddba
SH		 2332 return NULL;
2333}
2334
8cde8f0c 2335static int netvsc_register_vf(struct net_device *vf_netdev)
84bf9cef 2336{
0a1275ca 2337 struct net_device_context *net_device_ctx;
84bf9cef 2338 struct netvsc_device *netvsc_dev;
351e1581 2339 struct bpf_prog *prog;
00d7ddba 2340 struct net_device *ndev;
c0a41b88 2341 int ret;
84bf9cef 2342
8cde8f0c
SH		 2343 if (vf_netdev->addr_len != ETH_ALEN)
2344 return NOTIFY_DONE;
2345
00d7ddba 2346 ndev = get_netvsc_byslot(vf_netdev);
8cde8f0c
SH		 2347 if (!ndev)
2348 return NOTIFY_DONE;
2349
0a1275ca 2350 net_device_ctx = netdev_priv(ndev);
545a8e79 2351 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
f207c10d 2352 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
8cde8f0c 2353 return NOTIFY_DONE;
0c195567 2354
52d3b494 2355 /* if synthetic interface is a different namespace,
c0a41b88
SH		 2356 * then move the VF to that namespace; join will be
2357 * done again in that context.
2358 */
2359 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2360 ret = dev_change_net_namespace(vf_netdev,
0854fa82 2361 dev_net(ndev), "eth%d");
c0a41b88
SH		 2362 if (ret)
2363 netdev_err(vf_netdev,
2364 "could not move to same namespace as %s: %d\n",
2365 ndev->name, ret);
2366 else
2367 netdev_info(vf_netdev,
2368 "VF moved to namespace with: %s\n",
2369 ndev->name);
8cde8f0c 2370 return NOTIFY_DONE;
c0a41b88 2371 }
1ff78076 2372
8cde8f0c 2373 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
0c195567 2374
c0a41b88
SH		 2375 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2376 return NOTIFY_DONE;
2377
07d0f000 2378 dev_hold(vf_netdev);
8cde8f0c 2379 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
68622d07 2380
536ba2e0
HZ		 2381 if (ndev->needed_headroom < vf_netdev->needed_headroom)
2382 ndev->needed_headroom = vf_netdev->needed_headroom;
2383
68622d07
HZ		 2384 vf_netdev->wanted_features = ndev->features;
2385 netdev_update_features(vf_netdev);
2386
351e1581
HZ		 2387 prog = netvsc_xdp_get(netvsc_dev);
2388 netvsc_vf_setxdp(vf_netdev, prog);
2389
8cde8f0c 2390 return NOTIFY_OK;
84bf9cef
KS		 2391}
2392
da26658c
DC		 2393/* Change the data path when VF UP/DOWN/CHANGE are detected.
2394 *
2395 * Typically a UP or DOWN event is followed by a CHANGE event, so
2396 * net_device_ctx->data_path_is_vf is used to cache the current data path
2397 * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2398 * message.
2399 *
2400 * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2401 * interface, there is only the CHANGE event and no UP or DOWN event.
2402 */
34b06a2e 2403static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
84bf9cef 2404{
7b83f520 2405 struct net_device_context *net_device_ctx;
84bf9cef 2406 struct netvsc_device *netvsc_dev;
8cde8f0c 2407 struct net_device *ndev;
34b06a2e 2408 bool vf_is_up = false;
d0922bf7 2409 int ret;
34b06a2e
LL		 2410
2411 if (event != NETDEV_GOING_DOWN)
2412 vf_is_up = netif_running(vf_netdev);
fb84af8a 2413
8cde8f0c
SH		 2414 ndev = get_netvsc_byref(vf_netdev);
2415 if (!ndev)
2416 return NOTIFY_DONE;
2417
7b83f520 2418 net_device_ctx = netdev_priv(ndev);
2419 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
0c195567 2420 if (!netvsc_dev)
8cde8f0c 2421 return NOTIFY_DONE;
84bf9cef 2422
da26658c
DC		 2423 if (net_device_ctx->data_path_is_vf == vf_is_up)
2424 return NOTIFY_OK;
da26658c 2425
365e1ece
GK		 2426 if (vf_is_up && !net_device_ctx->vf_alloc) {
2427 netdev_info(ndev, "Waiting for the VF association from host\n");
2428 wait_for_completion(&net_device_ctx->vf_add);
2429 }
2430
d0922bf7
HZ		 2431 ret = netvsc_switch_datapath(ndev, vf_is_up);
2432
2433 if (ret) {
2434 netdev_err(ndev,
2435 "Data path failed to switch %s VF: %s, err: %d\n",
2436 vf_is_up ? "to" : "from", vf_netdev->name, ret);
2437 return NOTIFY_DONE;
2438 } else {
2439 netdev_info(ndev, "Data path switched %s VF: %s\n",
2440 vf_is_up ? "to" : "from", vf_netdev->name);
2441 }
84bf9cef 2442
8cde8f0c 2443 return NOTIFY_OK;
84bf9cef
KS		 2444}
2445
8cde8f0c 2446static int netvsc_unregister_vf(struct net_device *vf_netdev)
84bf9cef 2447{
8cde8f0c 2448 struct net_device *ndev;
0a1275ca 2449 struct net_device_context *net_device_ctx;
84bf9cef 2450
8cde8f0c
SH		 2451 ndev = get_netvsc_byref(vf_netdev);
2452 if (!ndev)
2453 return NOTIFY_DONE;
1ff78076
SS		 2454
2455 net_device_ctx = netdev_priv(ndev);
8cde8f0c 2456 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1ff78076 2457
0a1275ca 2458 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
f207c10d 2459
351e1581
HZ		 2460 netvsc_vf_setxdp(vf_netdev, NULL);
2461
365e1ece 2462 reinit_completion(&net_device_ctx->vf_add);
8cde8f0c
SH		 2463 netdev_rx_handler_unregister(vf_netdev);
2464 netdev_upper_dev_unlink(vf_netdev, ndev);
f207c10d 2465 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
07d0f000 2466 dev_put(vf_netdev);
ec158f77 2467
536ba2e0
HZ		 2468 ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2469
8cde8f0c 2470 return NOTIFY_OK;
84bf9cef
KS		 2471}
2472
84946899
S		 2473static int netvsc_probe(struct hv_device *dev,
2474 const struct hv_vmbus_device_id *dev_id)
df2fff28 2475{
df2fff28
GKH		 2476 struct net_device *net = NULL;
2477 struct net_device_context *net_device_ctx;
7c9f335a 2478 struct netvsc_device_info *device_info = NULL;
5b54dac8 2479 struct netvsc_device *nvdev;
0c195567 2480 int ret = -ENOMEM;
df2fff28 2481
5b54dac8 2482 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2b01888d 2483 VRSS_CHANNEL_MAX);
df2fff28 2484 if (!net)
0c195567 2485 goto no_net;
df2fff28 2486
1b07da51
HZ		 2487 netif_carrier_off(net);
2488
b37879e6
HZ		 2489 netvsc_init_settings(net);
2490
df2fff28 2491 net_device_ctx = netdev_priv(net);
9efd21e1 2492 net_device_ctx->device_ctx = dev;
3f300ff4
SX		 2493 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2494 if (netif_msg_probe(net_device_ctx))
2495 netdev_dbg(net, "netvsc msg_enable: %d\n",
2496 net_device_ctx->msg_enable);
2497
2ddd5e5f 2498 hv_set_drvdata(dev, net);
f580aec4 2499
891de74d 2500 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
df2fff28 2501
365e1ece 2502 init_completion(&net_device_ctx->vf_add);
27a70af3
VK		 2503 spin_lock_init(&net_device_ctx->lock);
2504 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
6123c668 2505 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
0c195567 2506
2507 net_device_ctx->vf_stats
2508 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2509 if (!net_device_ctx->vf_stats)
2510 goto no_stats;
27a70af3 2511
df2fff28 2512 net->netdev_ops = &device_ops;
7ad24ea4 2513 net->ethtool_ops = &ethtool_ops;
9efd21e1 2514 SET_NETDEV_DEV(net, &dev->device);
846da38d 2515 dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
df2fff28 2516
14a03cf8
VK		 2517 /* We always need headroom for rndis header */
2518 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2519
6450f8f2
HZ		 2520 /* Initialize the number of queues to be 1, we may change it if more
2521 * channels are offered later.
2522 */
2523 netif_set_real_num_tx_queues(net, 1);
2524 netif_set_real_num_rx_queues(net, 1);
2525
692e084e 2526 /* Notify the netvsc driver of the new device */
7c9f335a
HZ		 2527 device_info = netvsc_devinfo_get(NULL);
2528
2529 if (!device_info) {
2530 ret = -ENOMEM;
2531 goto devinfo_failed;
2532 }
2533
2534 nvdev = rndis_filter_device_add(dev, device_info);
9749fed5 2535 if (IS_ERR(nvdev)) {
2536 ret = PTR_ERR(nvdev);
692e084e 2537 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
0c195567 2538 goto rndis_failed;
df2fff28 2539 }
0c195567 2540
2f23e5ce 2541 eth_hw_addr_set(net, device_info->mac_adr);
692e084e 2542
e04e7a7b
DC		 2543 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2544 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2545 * all subchannels to show up, but that may not happen because
2546 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2547 * -> ... -> device_add() -> ... -> __device_attach() can't get
2548 * the device lock, so all the subchannels can't be processed --
52d3b494 2549 * finally netvsc_subchan_work() hangs forever.
e04e7a7b
DC		 2550 */
2551 rtnl_lock();
2552
3ffe64f1
SH		 2553 if (nvdev->num_chn > 1)
2554 schedule_work(&nvdev->subchan_work);
2555
aefd80e8 2556 /* hw_features computed in rndis_netdev_set_hwcaps() */
23312a3b 2557 net->features = net->hw_features |
b441f795
HZ		 2558 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2559 NETIF_F_HW_VLAN_CTAG_RX;
23312a3b 2560 net->vlan_features = net->features;
2561
1a33e10e
CW		 2562 netdev_lockdep_set_classes(net);
2563
450bdf5b
LB		 2564 net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2565 NETDEV_XDP_ACT_NDO_XMIT;
66c0e13a 2566
d0c2c997
JW		 2567 /* MTU range: 68 - 1500 or 65521 */
2568 net->min_mtu = NETVSC_MTU_MIN;
2569 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2570 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2571 else
2572 net->max_mtu = ETH_DATA_LEN;
2573
f6f13c12
HZ		 2574 nvdev->tx_disable = false;
2575
7bf7bb37 2576 ret = register_netdevice(net);
a68f9614
HZ		 2577 if (ret != 0) {
2578 pr_err("Unable to register netdev.\n");
0c195567 2579 goto register_failed;
a68f9614
HZ		 2580 }
2581
7bf7bb37
SH		 2582 list_add(&net_device_ctx->list, &netvsc_dev_list);
2583 rtnl_unlock();
7c9f335a 2584
351e1581 2585 netvsc_devinfo_put(device_info);
7bf7bb37 2586 return 0;
0c195567 2587
2588register_failed:
7bf7bb37 2589 rtnl_unlock();
0c195567 2590 rndis_filter_device_remove(dev, nvdev);
2591rndis_failed:
351e1581 2592 netvsc_devinfo_put(device_info);
7c9f335a 2593devinfo_failed:
0c195567 2594 free_percpu(net_device_ctx->vf_stats);
2595no_stats:
2596 hv_set_drvdata(dev, NULL);
2597 free_netdev(net);
2598no_net:
2599 return ret;
df2fff28
GKH		 2600}
2601
96ec2939 2602static void netvsc_remove(struct hv_device *dev)
df2fff28 2603{
122a5f64 2604 struct net_device_context *ndev_ctx;
7b2ee50c
SH		 2605 struct net_device *vf_netdev, *net;
2606 struct netvsc_device *nvdev;
2ddd5e5f 2607
3d541ac5 2608 net = hv_get_drvdata(dev);
df2fff28 2609 if (net == NULL) {
415b023a 2610 dev_err(&dev->device, "No net device to remove\n");
96ec2939 2611 return;
df2fff28
GKH		 2612 }
2613
122a5f64 2614 ndev_ctx = netdev_priv(net);
3d541ac5 2615
122a5f64
HZ		 2616 cancel_delayed_work_sync(&ndev_ctx->dwork);
2617
018349d7
SH		 2618 rtnl_lock();
2619 nvdev = rtnl_dereference(ndev_ctx->nvdev);
351e1581 2620 if (nvdev) {
7b2ee50c 2621 cancel_work_sync(&nvdev->subchan_work);
351e1581
HZ		 2622 netvsc_xdp_set(net, NULL, NULL, nvdev);
2623 }
7b2ee50c 2624
df2fff28
GKH		 2625 /*
2626 * Call to the vsc driver to let it know that the device is being
a0be450e 2627 * removed. Also blocks mtu and channel changes.
df2fff28 2628 */
ec158f77
SH		 2629 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2630 if (vf_netdev)
8cde8f0c 2631 netvsc_unregister_vf(vf_netdev);
ec158f77 2632
7b2ee50c
SH		 2633 if (nvdev)
2634 rndis_filter_device_remove(dev, nvdev);
2635
8195b139 2636 unregister_netdevice(net);
7bf7bb37 2637 list_del(&ndev_ctx->list);
8195b139 2638
a0be450e 2639 rtnl_unlock();
2640
3d541ac5
VK		 2641 hv_set_drvdata(dev, NULL);
2642
0c195567 2643 free_percpu(ndev_ctx->vf_stats);
6c80f3fc 2644 free_netdev(net);
df2fff28
GKH		 2645}
2646
0efeea5f
DC		 2647static int netvsc_suspend(struct hv_device *dev)
2648{
2649 struct net_device_context *ndev_ctx;
0efeea5f 2650 struct netvsc_device *nvdev;
19162fd4 2651 struct net_device *net;
0efeea5f
DC		 2652 int ret;
2653
2654 net = hv_get_drvdata(dev);
2655
2656 ndev_ctx = netdev_priv(net);
2657 cancel_delayed_work_sync(&ndev_ctx->dwork);
2658
2659 rtnl_lock();
2660
2661 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2662 if (nvdev == NULL) {
2663 ret = -ENODEV;
2664 goto out;
2665 }
2666
0efeea5f
DC		 2667 /* Save the current config info */
2668 ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
eb4c0788
YL		 2669 if (!ndev_ctx->saved_netvsc_dev_info) {
2670 ret = -ENOMEM;
2671 goto out;
2672 }
0efeea5f
DC		 2673 ret = netvsc_detach(net, nvdev);
2674out:
2675 rtnl_unlock();
2676
2677 return ret;
2678}
2679
2680static int netvsc_resume(struct hv_device *dev)
2681{
2682 struct net_device *net = hv_get_drvdata(dev);
2683 struct net_device_context *net_device_ctx;
2684 struct netvsc_device_info *device_info;
2685 int ret;
2686
2687 rtnl_lock();
2688
2689 net_device_ctx = netdev_priv(net);
da26658c
DC		 2690
2691 /* Reset the data path to the netvsc NIC before re-opening the vmbus
2692 * channel. Later netvsc_netdev_event() will switch the data path to
2693 * the VF upon the UP or CHANGE event.
2694 */
2695 net_device_ctx->data_path_is_vf = false;
0efeea5f
DC		 2696 device_info = net_device_ctx->saved_netvsc_dev_info;
2697
2698 ret = netvsc_attach(net, device_info);
2699
351e1581 2700 netvsc_devinfo_put(device_info);
0efeea5f
DC		 2701 net_device_ctx->saved_netvsc_dev_info = NULL;
2702
351e1581
HZ		 2703 rtnl_unlock();
2704
0efeea5f
DC		 2705 return ret;
2706}
345c4cc3 2707static const struct hv_vmbus_device_id id_table[] = {
c45cf2d4 2708 /* Network guid */
8f505944 2709 { HV_NIC_GUID, },
c45cf2d4 2710 { },
345c4cc3
S		 2711};
2712
2713MODULE_DEVICE_TABLE(vmbus, id_table);
2714
f1542a66 2715/* The one and only one */
fde0ef9b 2716static struct hv_driver netvsc_drv = {
d31b20fc 2717 .name = KBUILD_MODNAME,
345c4cc3 2718 .id_table = id_table,
fde0ef9b
S		 2719 .probe = netvsc_probe,
2720 .remove = netvsc_remove,
0efeea5f
DC		 2721 .suspend = netvsc_suspend,
2722 .resume = netvsc_resume,
af0a5646 2723 .driver = {
9a33629b 2724 .probe_type = PROBE_FORCE_SYNCHRONOUS,
af0a5646 2725 },
d4890970 2726};
f1542a66 2727
8cde8f0c
SH		 2728/*
2729 * On Hyper-V, every VF interface is matched with a corresponding
2730 * synthetic interface. The synthetic interface is presented first
2731 * to the guest. When the corresponding VF instance is registered,
2732 * we will take care of switching the data path.
2733 */
2734static int netvsc_netdev_event(struct notifier_block *this,
2735 unsigned long event, void *ptr)
2736{
2737 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2738
2739 /* Skip our own events */
2740 if (event_dev->netdev_ops == &device_ops)
2741 return NOTIFY_DONE;
2742
2743 /* Avoid non-Ethernet type devices */
2744 if (event_dev->type != ARPHRD_ETHER)
2745 return NOTIFY_DONE;
2746
2747 /* Avoid Vlan dev with same MAC registering as VF */
2748 if (is_vlan_dev(event_dev))
2749 return NOTIFY_DONE;
2750
2751 /* Avoid Bonding master dev with same MAC registering as VF */
c60882a4 2752 if (netif_is_bond_master(event_dev))
8cde8f0c
SH		 2753 return NOTIFY_DONE;
2754
2755 switch (event) {
2756 case NETDEV_REGISTER:
2757 return netvsc_register_vf(event_dev);
2758 case NETDEV_UNREGISTER:
2759 return netvsc_unregister_vf(event_dev);
2760 case NETDEV_UP:
2761 case NETDEV_DOWN:
de214e52 2762 case NETDEV_CHANGE:
34b06a2e
LL		 2763 case NETDEV_GOING_DOWN:
2764 return netvsc_vf_changed(event_dev, event);
8cde8f0c
SH		 2765 default:
2766 return NOTIFY_DONE;
2767 }
2768}
2769
2770static struct notifier_block netvsc_netdev_notifier = {
2771 .notifier_call = netvsc_netdev_event,
2772};
2773
a9869c94 2774static void __exit netvsc_drv_exit(void)
fceaf24a 2775{
8cde8f0c 2776 unregister_netdevice_notifier(&netvsc_netdev_notifier);
768fa219 2777 vmbus_driver_unregister(&netvsc_drv);
fceaf24a
HJ		 2778}
2779
1fde28cf 2780static int __init netvsc_drv_init(void)
df2fff28 2781{
84bf9cef
KS		 2782 int ret;
2783
fa85a6c2
HZ		 2784 if (ring_size < RING_SIZE_MIN) {
2785 ring_size = RING_SIZE_MIN;
a7f99d0f 2786 pr_info("Increased ring_size to %u (min allowed)\n",
fa85a6c2
HZ		 2787 ring_size);
2788 }
a7f99d0f 2789 netvsc_ring_bytes = ring_size * PAGE_SIZE;
84bf9cef 2790
a7f99d0f 2791 ret = vmbus_driver_register(&netvsc_drv);
84bf9cef
KS		 2792 if (ret)
2793 return ret;
2794
8cde8f0c 2795 register_netdevice_notifier(&netvsc_netdev_notifier);
84bf9cef 2796 return 0;
df2fff28
GKH		 2797}
2798
26c14cc1 2799MODULE_LICENSE("GPL");
7880fc54 2800MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
fceaf24a 2801
1fde28cf 2802module_init(netvsc_drv_init);
a9869c94 2803module_exit(netvsc_drv_exit);
Linux 6.x block layer and io_uring tree(s)