]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/net/hyperv/netvsc_drv.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
netvsc: cleanup datapath switch
[mirror_ubuntu-bionic-kernel.git] / drivers / net / hyperv / netvsc_drv.c
CommitLineData
fceaf24a 1/*
fceaf24a
HJ		 2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
adf8d3ff 14 * this program; if not, see <http://www.gnu.org/licenses/>.
fceaf24a
HJ		 15 *
16 * Authors:
d0e94d17 17 * Haiyang Zhang <haiyangz@microsoft.com>
fceaf24a 18 * Hank Janssen <hjanssen@microsoft.com>
fceaf24a 19 */
eb335bc4
HJ		 20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
fceaf24a 22#include <linux/init.h>
9079ce69 23#include <linux/atomic.h>
fceaf24a
HJ		 24#include <linux/module.h>
25#include <linux/highmem.h>
26#include <linux/device.h>
fceaf24a 27#include <linux/io.h>
fceaf24a
HJ		 28#include <linux/delay.h>
29#include <linux/netdevice.h>
30#include <linux/inetdevice.h>
31#include <linux/etherdevice.h>
32#include <linux/skbuff.h>
c802db11 33#include <linux/if_vlan.h>
fceaf24a 34#include <linux/in.h>
5a0e3ad6 35#include <linux/slab.h>
27f5aa92 36#include <linux/rtnetlink.h>
0c195567 37#include <linux/netpoll.h>
27f5aa92 38
fceaf24a
HJ		 39#include <net/arp.h>
40#include <net/route.h>
41#include <net/sock.h>
42#include <net/pkt_sched.h>
8eb1b3c3
MK		 43#include <net/checksum.h>
44#include <net/ip6_checksum.h>
3f335ea2 45
5ca7252a 46#include "hyperv_net.h"
fceaf24a 47
8b532797 48#define RING_SIZE_MIN 64
49#define NETVSC_MIN_TX_SECTIONS 10
50#define NETVSC_DEFAULT_TX 192 /* ~1M */
51#define NETVSC_MIN_RX_SECTIONS 10 /* ~64K */
52#define NETVSC_DEFAULT_RX 2048 /* ~4M */
53
27a70af3 54#define LINKCHANGE_INT (2 * HZ)
6123c668 55#define VF_TAKEOVER_INT (HZ / 10)
a50af86d 56
99c8da0f 57static int ring_size = 128;
450d7a4b
SH		 58module_param(ring_size, int, S_IRUGO);
59MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
fceaf24a 60
3f300ff4
SX		 61static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
62 NETIF_MSG_LINK | NETIF_MSG_IFUP |
63 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
64 NETIF_MSG_TX_ERR;
65
66static int debug = -1;
67module_param(debug, int, S_IRUGO);
68MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
69
4e9bfefa 70static void netvsc_set_multicast_list(struct net_device *net)
fceaf24a 71{
792df872 72 struct net_device_context *net_device_ctx = netdev_priv(net);
4f19c0d8 73 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
d426b2e3 74
4f19c0d8 75 rndis_filter_update(nvdev);
fceaf24a
HJ		 76}
77
fceaf24a
HJ		 78static int netvsc_open(struct net_device *net)
79{
53fa1a6f 80 struct net_device_context *ndev_ctx = netdev_priv(net);
0c195567 81 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
79e8cbe7 82 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
891de74d 83 struct rndis_device *rdev;
02fafbc6 84 int ret = 0;
fceaf24a 85
891de74d
HZ		 86 netif_carrier_off(net);
87
d515d0ff 88 /* Open up the device */
2f5fa6c8 89 ret = rndis_filter_open(nvdev);
d515d0ff
HZ		 90 if (ret != 0) {
91 netdev_err(net, "unable to open device (ret %d).\n", ret);
92 return ret;
fceaf24a
HJ		 93 }
94
2de8530b 95 netif_tx_wake_all_queues(net);
d515d0ff 96
891de74d 97 rdev = nvdev->extension;
0c195567 98
99 if (!rdev->link_state)
891de74d
HZ		 100 netif_carrier_on(net);
101
0c195567 102 if (vf_netdev) {
103 /* Setting synthetic device up transparently sets
104 * slave as up. If open fails, then slave will be
105 * still be offline (and not used).
106 */
107 ret = dev_open(vf_netdev);
108 if (ret)
109 netdev_warn(net,
110 "unable to open slave: %s: %d\n",
111 vf_netdev->name, ret);
112 }
113 return 0;
fceaf24a
HJ		 114}
115
fceaf24a
HJ		 116static int netvsc_close(struct net_device *net)
117{
fceaf24a 118 struct net_device_context *net_device_ctx = netdev_priv(net);
0c195567 119 struct net_device *vf_netdev
120 = rtnl_dereference(net_device_ctx->vf_netdev);
545a8e79 121 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
c6f71c41 122 int ret = 0;
40975962 123 u32 aread, i, msec = 10, retry = 0, retry_max = 20;
2de8530b 124 struct vmbus_channel *chn;
fceaf24a 125
0a282538 126 netif_tx_disable(net);
fceaf24a 127
c6f71c41
HZ		 128 /* No need to close rndis filter if it is removed already */
129 if (!nvdev)
130 goto out;
131
2f5fa6c8 132 ret = rndis_filter_close(nvdev);
2de8530b 133 if (ret != 0) {
eb335bc4 134 netdev_err(net, "unable to close device (ret %d).\n", ret);
2de8530b
HZ		 135 return ret;
136 }
137
138 /* Ensure pending bytes in ring are read */
139 while (true) {
140 aread = 0;
141 for (i = 0; i < nvdev->num_chn; i++) {
b8b835a8 142 chn = nvdev->chan_table[i].channel;
2de8530b
HZ		 143 if (!chn)
144 continue;
145
40975962 146 aread = hv_get_bytes_to_read(&chn->inbound);
2de8530b
HZ		 147 if (aread)
148 break;
149
40975962 150 aread = hv_get_bytes_to_read(&chn->outbound);
2de8530b
HZ		 151 if (aread)
152 break;
153 }
154
155 retry++;
156 if (retry > retry_max || aread == 0)
157 break;
158
159 msleep(msec);
160
161 if (msec < 1000)
162 msec *= 2;
163 }
164
165 if (aread) {
166 netdev_err(net, "Ring buffer not empty after closing rndis\n");
167 ret = -ETIMEDOUT;
168 }
fceaf24a 169
c6f71c41 170out:
0c195567 171 if (vf_netdev)
172 dev_close(vf_netdev);
173
fceaf24a
HJ		 174 return ret;
175}
176
8a00251a 177static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
89bb42b1 178 int pkt_type)
8a00251a
KS		 179{
180 struct rndis_packet *rndis_pkt;
181 struct rndis_per_packet_info *ppi;
182
183 rndis_pkt = &msg->msg.pkt;
184 rndis_pkt->data_offset += ppi_size;
185
186 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
187 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
188
189 ppi->size = ppi_size;
190 ppi->type = pkt_type;
191 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
192
193 rndis_pkt->per_pkt_info_len += ppi_size;
194
195 return ppi;
196}
197
4823eb2f
HZ		 198/* Azure hosts don't support non-TCP port numbers in hashing for fragmented
199 * packets. We can use ethtool to change UDP hash level when necessary.
f72860af 200 */
4823eb2f
HZ		 201static inline u32 netvsc_get_hash(
202 struct sk_buff *skb,
203 const struct net_device_context *ndc)
f72860af
HZ		 204{
205 struct flow_keys flow;
206 u32 hash;
207 static u32 hashrnd __read_mostly;
208
209 net_get_random_once(&hashrnd, sizeof(hashrnd));
210
211 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
212 return 0;
213
4823eb2f
HZ		 214 if (flow.basic.ip_proto == IPPROTO_TCP ||
215 (flow.basic.ip_proto == IPPROTO_UDP &&
216 ((flow.basic.n_proto == htons(ETH_P_IP) && ndc->udp4_l4_hash) ||
217 (flow.basic.n_proto == htons(ETH_P_IPV6) &&
218 ndc->udp6_l4_hash)))) {
f72860af
HZ		 219 return skb_get_hash(skb);
220 } else {
221 if (flow.basic.n_proto == htons(ETH_P_IP))
222 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
223 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
224 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
225 else
226 hash = 0;
227
228 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
229 }
230
231 return hash;
232}
233
8db91f6a
HZ		 234static inline int netvsc_get_tx_queue(struct net_device *ndev,
235 struct sk_buff *skb, int old_idx)
236{
237 const struct net_device_context *ndc = netdev_priv(ndev);
238 struct sock *sk = skb->sk;
239 int q_idx;
240
4823eb2f 241 q_idx = ndc->tx_send_table[netvsc_get_hash(skb, ndc) &
8db91f6a
HZ		 242 (VRSS_SEND_TAB_SIZE - 1)];
243
244 /* If queue index changed record the new value */
245 if (q_idx != old_idx &&
246 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
247 sk_tx_queue_set(sk, q_idx);
248
249 return q_idx;
250}
251
d8e18ee0 252/*
253 * Select queue for transmit.
254 *
255 * If a valid queue has already been assigned, then use that.
256 * Otherwise compute tx queue based on hash and the send table.
257 *
258 * This is basically similar to default (__netdev_pick_tx) with the added step
259 * of using the host send_table when no other queue has been assigned.
260 *
261 * TODO support XPS - but get_xps_queue not exported
262 */
0c195567 263static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
5b54dac8 264{
8db91f6a
HZ		 265 int q_idx = sk_tx_queue_get(skb->sk);
266
0c195567 267 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
8db91f6a
HZ		 268 /* If forwarding a packet, we use the recorded queue when
269 * available for better cache locality.
270 */
271 if (skb_rx_queue_recorded(skb))
272 q_idx = skb_get_rx_queue(skb);
273 else
274 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
d8e18ee0 275 }
5b54dac8
HZ		 276
277 return q_idx;
278}
279
0c195567 280static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
281 void *accel_priv,
282 select_queue_fallback_t fallback)
283{
284 struct net_device_context *ndc = netdev_priv(ndev);
285 struct net_device *vf_netdev;
286 u16 txq;
287
288 rcu_read_lock();
289 vf_netdev = rcu_dereference(ndc->vf_netdev);
290 if (vf_netdev) {
291 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
292 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
293 } else {
294 txq = netvsc_pick_tx(ndev, skb);
295 }
296 rcu_read_unlock();
297
298 while (unlikely(txq >= ndev->real_num_tx_queues))
299 txq -= ndev->real_num_tx_queues;
300
301 return txq;
302}
303
54a7357f 304static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
89bb42b1 305 struct hv_page_buffer *pb)
54a7357f
KS		 306{
307 int j = 0;
308
309 /* Deal with compund pages by ignoring unused part
310 * of the page.
311 */
312 page += (offset >> PAGE_SHIFT);
313 offset &= ~PAGE_MASK;
314
315 while (len > 0) {
316 unsigned long bytes;
317
318 bytes = PAGE_SIZE - offset;
319 if (bytes > len)
320 bytes = len;
321 pb[j].pfn = page_to_pfn(page);
322 pb[j].offset = offset;
323 pb[j].len = bytes;
324
325 offset += bytes;
326 len -= bytes;
327
328 if (offset == PAGE_SIZE && len) {
329 page++;
330 offset = 0;
331 j++;
332 }
333 }
334
335 return j + 1;
336}
337
8a00251a 338static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
a9f2e2d6 339 struct hv_netvsc_packet *packet,
02b6de01 340 struct hv_page_buffer *pb)
54a7357f
KS		 341{
342 u32 slots_used = 0;
343 char *data = skb->data;
344 int frags = skb_shinfo(skb)->nr_frags;
345 int i;
346
347 /* The packet is laid out thus:
aa0a34be 348 * 1. hdr: RNDIS header and PPI
54a7357f
KS		 349 * 2. skb linear data
350 * 3. skb fragment data
351 */
ea5a32c0 352 slots_used += fill_pg_buf(virt_to_page(hdr),
353 offset_in_page(hdr),
354 len, &pb[slots_used]);
54a7357f 355
aa0a34be
HZ		 356 packet->rmsg_size = len;
357 packet->rmsg_pgcnt = slots_used;
358
54a7357f
KS		 359 slots_used += fill_pg_buf(virt_to_page(data),
360 offset_in_page(data),
361 skb_headlen(skb), &pb[slots_used]);
362
363 for (i = 0; i < frags; i++) {
364 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
365
366 slots_used += fill_pg_buf(skb_frag_page(frag),
367 frag->page_offset,
368 skb_frag_size(frag), &pb[slots_used]);
369 }
8a00251a 370 return slots_used;
54a7357f
KS		 371}
372
80d887db 373static int count_skb_frag_slots(struct sk_buff *skb)
374{
375 int i, frags = skb_shinfo(skb)->nr_frags;
376 int pages = 0;
377
378 for (i = 0; i < frags; i++) {
379 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
380 unsigned long size = skb_frag_size(frag);
381 unsigned long offset = frag->page_offset;
382
383 /* Skip unused frames from start of page */
384 offset &= ~PAGE_MASK;
385 pages += PFN_UP(offset + size);
386 }
387 return pages;
388}
389
390static int netvsc_get_slots(struct sk_buff *skb)
54a7357f 391{
80d887db 392 char *data = skb->data;
393 unsigned int offset = offset_in_page(data);
394 unsigned int len = skb_headlen(skb);
395 int slots;
396 int frag_slots;
397
398 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
399 frag_slots = count_skb_frag_slots(skb);
400 return slots + frag_slots;
54a7357f
KS		 401}
402
23312a3b 403static u32 net_checksum_info(struct sk_buff *skb)
08cd04bf 404{
23312a3b 405 if (skb->protocol == htons(ETH_P_IP)) {
406 struct iphdr *ip = ip_hdr(skb);
08cd04bf 407
23312a3b 408 if (ip->protocol == IPPROTO_TCP)
409 return TRANSPORT_INFO_IPV4_TCP;
410 else if (ip->protocol == IPPROTO_UDP)
411 return TRANSPORT_INFO_IPV4_UDP;
08cd04bf 412 } else {
23312a3b 413 struct ipv6hdr *ip6 = ipv6_hdr(skb);
414
415 if (ip6->nexthdr == IPPROTO_TCP)
416 return TRANSPORT_INFO_IPV6_TCP;
37b9dfa0 417 else if (ip6->nexthdr == IPPROTO_UDP)
23312a3b 418 return TRANSPORT_INFO_IPV6_UDP;
08cd04bf
KS		 419 }
420
23312a3b 421 return TRANSPORT_INFO_NOT_IP;
08cd04bf
KS		 422}
423
0c195567 424/* Send skb on the slave VF device. */
425static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
426 struct sk_buff *skb)
427{
428 struct net_device_context *ndev_ctx = netdev_priv(net);
429 unsigned int len = skb->len;
430 int rc;
431
432 skb->dev = vf_netdev;
433 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
434
435 rc = dev_queue_xmit(skb);
436 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
437 struct netvsc_vf_pcpu_stats *pcpu_stats
438 = this_cpu_ptr(ndev_ctx->vf_stats);
439
440 u64_stats_update_begin(&pcpu_stats->syncp);
441 pcpu_stats->tx_packets++;
442 pcpu_stats->tx_bytes += len;
443 u64_stats_update_end(&pcpu_stats->syncp);
444 } else {
445 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
446 }
447
448 return rc;
449}
450
02fafbc6 451static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
fceaf24a 452{
fceaf24a 453 struct net_device_context *net_device_ctx = netdev_priv(net);
981a1bd8 454 struct hv_netvsc_packet *packet = NULL;
02fafbc6 455 int ret;
8a00251a
KS		 456 unsigned int num_data_pgs;
457 struct rndis_message *rndis_msg;
458 struct rndis_packet *rndis_pkt;
0c195567 459 struct net_device *vf_netdev;
8a00251a 460 u32 rndis_msg_size;
8a00251a 461 struct rndis_per_packet_info *ppi;
307f0995 462 u32 hash;
02b6de01 463 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
fceaf24a 464
0c195567 465 /* if VF is present and up then redirect packets
466 * already called with rcu_read_lock_bh
467 */
468 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
469 if (vf_netdev && netif_running(vf_netdev) &&
470 !netpoll_tx_running(net))
471 return netvsc_vf_xmit(net, vf_netdev, skb);
472
80d887db 473 /* We will atmost need two pages to describe the rndis
474 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
e88f7e07
VK		 475 * of pages in a single packet. If skb is scattered around
476 * more pages we try linearizing it.
54a7357f 477 */
80d887db 478
479 num_data_pgs = netvsc_get_slots(skb) + 2;
480
0ab05141 481 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
4323b47c
SH		 482 ++net_device_ctx->eth_stats.tx_scattered;
483
484 if (skb_linearize(skb))
485 goto no_memory;
0ab05141 486
80d887db 487 num_data_pgs = netvsc_get_slots(skb) + 2;
0ab05141 488 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
4323b47c 489 ++net_device_ctx->eth_stats.tx_too_big;
0ab05141
SH		 490 goto drop;
491 }
54a7357f 492 }
fceaf24a 493
c0eb4540
KS		 494 /*
495 * Place the rndis header in the skb head room and
496 * the skb->cb will be used for hv_netvsc_packet
497 * structure.
498 */
499 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
4323b47c
SH		 500 if (ret)
501 goto no_memory;
502
c0eb4540
KS		 503 /* Use the skb control buffer for building up the packet */
504 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
505 FIELD_SIZEOF(struct sk_buff, cb));
506 packet = (struct hv_netvsc_packet *)skb->cb;
fceaf24a 507
5b54dac8
HZ		 508 packet->q_idx = skb_get_queue_mapping(skb);
509
4d447c9a 510 packet->total_data_buflen = skb->len;
793e3955 511 packet->total_bytes = skb->len;
512 packet->total_packets = 1;
fceaf24a 513
c0eb4540 514 rndis_msg = (struct rndis_message *)skb->head;
b08cc791 515
24476760 516 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
fceaf24a 517
8a00251a 518 /* Add the rndis header */
8a00251a
KS		 519 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
520 rndis_msg->msg_len = packet->total_data_buflen;
521 rndis_pkt = &rndis_msg->msg.pkt;
522 rndis_pkt->data_offset = sizeof(struct rndis_packet);
523 rndis_pkt->data_len = packet->total_data_buflen;
524 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
525
526 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
527
307f0995
HZ		 528 hash = skb_get_hash_raw(skb);
529 if (hash != 0 && net->real_num_tx_queues > 1) {
530 rndis_msg_size += NDIS_HASH_PPI_SIZE;
531 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
532 NBL_HASH_VALUE);
533 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
534 }
535
0ab05141 536 if (skb_vlan_tag_present(skb)) {
8a00251a
KS		 537 struct ndis_pkt_8021q_info *vlan;
538
539 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
540 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
00f5024e 541 IEEE_8021Q_INFO);
542
543 vlan = (void *)ppi + ppi->ppi_offset;
760d1e36
KS		 544 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
545 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
8a00251a
KS		 546 VLAN_PRIO_SHIFT;
547 }
548
23312a3b 549 if (skb_is_gso(skb)) {
0ab05141
SH		 550 struct ndis_tcp_lso_info *lso_info;
551
552 rndis_msg_size += NDIS_LSO_PPI_SIZE;
553 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
554 TCP_LARGESEND_PKTINFO);
555
00f5024e 556 lso_info = (void *)ppi + ppi->ppi_offset;
0ab05141
SH		 557
558 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
23312a3b 559 if (skb->protocol == htons(ETH_P_IP)) {
0ab05141
SH		 560 lso_info->lso_v2_transmit.ip_version =
561 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
562 ip_hdr(skb)->tot_len = 0;
563 ip_hdr(skb)->check = 0;
564 tcp_hdr(skb)->check =
565 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
566 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
567 } else {
568 lso_info->lso_v2_transmit.ip_version =
569 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
570 ipv6_hdr(skb)->payload_len = 0;
571 tcp_hdr(skb)->check =
572 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
573 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
574 }
23312a3b 575 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
0ab05141 576 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
ad19bc8a 577 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
23312a3b 578 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
579 struct ndis_tcp_ip_checksum_info *csum_info;
580
ad19bc8a 581 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
582 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
583 TCPIP_CHKSUM_PKTINFO);
584
585 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
586 ppi->ppi_offset);
587
23312a3b 588 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
589
590 if (skb->protocol == htons(ETH_P_IP)) {
ad19bc8a 591 csum_info->transmit.is_ipv4 = 1;
23312a3b 592
593 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
594 csum_info->transmit.tcp_checksum = 1;
595 else
596 csum_info->transmit.udp_checksum = 1;
597 } else {
ad19bc8a 598 csum_info->transmit.is_ipv6 = 1;
599
23312a3b 600 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
601 csum_info->transmit.tcp_checksum = 1;
602 else
603 csum_info->transmit.udp_checksum = 1;
604 }
ad19bc8a 605 } else {
23312a3b 606 /* Can't do offload of this type of checksum */
ad19bc8a 607 if (skb_checksum_help(skb))
608 goto drop;
609 }
08cd04bf
KS		 610 }
611
8a00251a
KS		 612 /* Start filling in the page buffers with the rndis hdr */
613 rndis_msg->msg_len += rndis_msg_size;
942396b0 614 packet->total_data_buflen = rndis_msg->msg_len;
8a00251a 615 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
02b6de01 616 skb, packet, pb);
8a00251a 617
76d13b56 618 /* timestamp packet in software */
619 skb_tx_timestamp(skb);
2a926f79 620
02b6de01 621 ret = netvsc_send(net_device_ctx, packet, rndis_msg, pb, skb);
793e3955 622 if (likely(ret == 0))
0ab05141 623 return NETDEV_TX_OK;
4323b47c
SH		 624
625 if (ret == -EAGAIN) {
626 ++net_device_ctx->eth_stats.tx_busy;
0ab05141 627 return NETDEV_TX_BUSY;
4323b47c
SH		 628 }
629
630 if (ret == -ENOSPC)
631 ++net_device_ctx->eth_stats.tx_no_space;
0ab05141
SH		 632
633drop:
634 dev_kfree_skb_any(skb);
635 net->stats.tx_dropped++;
fceaf24a 636
0ab05141 637 return NETDEV_TX_OK;
4323b47c
SH		 638
639no_memory:
640 ++net_device_ctx->eth_stats.tx_no_memory;
641 goto drop;
fceaf24a 642}
89bb42b1 643
3e189519 644/*
02fafbc6
GKH		 645 * netvsc_linkstatus_callback - Link up/down notification
646 */
90ef117a 647void netvsc_linkstatus_callback(struct hv_device *device_obj,
3a494e71 648 struct rndis_message *resp)
fceaf24a 649{
3a494e71 650 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
2ddd5e5f 651 struct net_device *net;
c996edcf 652 struct net_device_context *ndev_ctx;
27a70af3
VK		 653 struct netvsc_reconfig *event;
654 unsigned long flags;
891de74d 655
7f5d5af0
HZ		 656 net = hv_get_drvdata(device_obj);
657
658 if (!net)
659 return;
660
661 ndev_ctx = netdev_priv(net);
662
663 /* Update the physical link speed when changing to another vSwitch */
664 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
665 u32 speed;
666
89bb42b1 667 speed = *(u32 *)((void *)indicate
668 + indicate->status_buf_offset) / 10000;
7f5d5af0
HZ		 669 ndev_ctx->speed = speed;
670 return;
671 }
672
673 /* Handle these link change statuses below */
27a70af3
VK		 674 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
675 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
676 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
3a494e71 677 return;
891de74d 678
7f5d5af0 679 if (net->reg_state != NETREG_REGISTERED)
fceaf24a 680 return;
fceaf24a 681
27a70af3
VK		 682 event = kzalloc(sizeof(*event), GFP_ATOMIC);
683 if (!event)
684 return;
685 event->event = indicate->status;
686
687 spin_lock_irqsave(&ndev_ctx->lock, flags);
688 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
689 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
690
691 schedule_delayed_work(&ndev_ctx->dwork, 0);
fceaf24a
HJ		 692}
693
84bf9cef 694static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
e91e7dd7 695 struct napi_struct *napi,
dc54a08c 696 const struct ndis_tcp_ip_checksum_info *csum_info,
697 const struct ndis_pkt_8021q_info *vlan,
698 void *data, u32 buflen)
fceaf24a 699{
fceaf24a 700 struct sk_buff *skb;
fceaf24a 701
e91e7dd7 702 skb = napi_alloc_skb(napi, buflen);
84bf9cef
KS		 703 if (!skb)
704 return skb;
fceaf24a 705
02fafbc6
GKH		 706 /*
707 * Copy to skb. This copy is needed here since the memory pointed by
708 * hv_netvsc_packet cannot be deallocated
709 */
59ae1d12 710 skb_put_data(skb, data, buflen);
fceaf24a
HJ		 711
712 skb->protocol = eth_type_trans(skb, net);
e52fed71
SH		 713
714 /* skb is already created with CHECKSUM_NONE */
715 skb_checksum_none_assert(skb);
716
717 /*
718 * In Linux, the IP checksum is always checked.
719 * Do L4 checksum offload if enabled and present.
720 */
721 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
722 if (csum_info->receive.tcp_checksum_succeeded ||
723 csum_info->receive.udp_checksum_succeeded)
e3d605ed 724 skb->ip_summed = CHECKSUM_UNNECESSARY;
e3d605ed
KS		 725 }
726
dc54a08c 727 if (vlan) {
728 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
729
93725cbd 730 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
760d1e36 731 vlan_tci);
dc54a08c 732 }
fceaf24a 733
84bf9cef
KS		 734 return skb;
735}
736
737/*
738 * netvsc_recv_callback - Callback when we receive a packet from the
739 * "wire" on the specified device.
740 */
dc54a08c 741int netvsc_recv_callback(struct net_device *net,
742 struct vmbus_channel *channel,
743 void *data, u32 len,
744 const struct ndis_tcp_ip_checksum_info *csum_info,
745 const struct ndis_pkt_8021q_info *vlan)
84bf9cef 746{
3d541ac5 747 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 748 struct netvsc_device *net_device;
742fe54c 749 u16 q_idx = channel->offermsg.offer.sub_channel_index;
545a8e79 750 struct netvsc_channel *nvchan;
84bf9cef 751 struct sk_buff *skb;
84bf9cef 752 struct netvsc_stats *rx_stats;
84bf9cef 753
9cbcc428 754 if (net->reg_state != NETREG_REGISTERED)
84bf9cef
KS		 755 return NVSP_STAT_FAIL;
756
0719e72c 757 rcu_read_lock();
545a8e79 758 net_device = rcu_dereference(net_device_ctx->nvdev);
759 if (unlikely(!net_device))
760 goto drop;
761
762 nvchan = &net_device->chan_table[q_idx];
84bf9cef
KS		 763
764 /* Allocate a skb - TODO direct I/O to pages? */
e91e7dd7 765 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
766 csum_info, vlan, data, len);
84bf9cef 767 if (unlikely(!skb)) {
545a8e79 768drop:
84bf9cef 769 ++net->stats.rx_dropped;
0719e72c 770 rcu_read_unlock();
84bf9cef
KS		 771 return NVSP_STAT_FAIL;
772 }
5b54dac8 773
0c195567 774 skb_record_rx_queue(skb, q_idx);
9cbcc428
SH		 775
776 /*
777 * Even if injecting the packet, record the statistics
778 * on the synthetic device because modifying the VF device
779 * statistics will not work correctly.
780 */
742fe54c 781 rx_stats = &nvchan->rx_stats;
4b02b58b 782 u64_stats_update_begin(&rx_stats->syncp);
7eafd9b4 783 rx_stats->packets++;
dc54a08c 784 rx_stats->bytes += len;
f7ad75b7
SH		 785
786 if (skb->pkt_type == PACKET_BROADCAST)
787 ++rx_stats->broadcast;
788 else if (skb->pkt_type == PACKET_MULTICAST)
789 ++rx_stats->multicast;
4b02b58b 790 u64_stats_update_end(&rx_stats->syncp);
9495c282 791
742fe54c 792 napi_gro_receive(&nvchan->napi, skb);
0719e72c 793 rcu_read_unlock();
fceaf24a 794
fceaf24a
HJ		 795 return 0;
796}
797
f82f4ad7
SH		 798static void netvsc_get_drvinfo(struct net_device *net,
799 struct ethtool_drvinfo *info)
800{
7826d43f 801 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
7826d43f 802 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
f82f4ad7
SH		 803}
804
59995370
AS		 805static void netvsc_get_channels(struct net_device *net,
806 struct ethtool_channels *channel)
807{
808 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 809 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
59995370
AS		 810
811 if (nvdev) {
812 channel->max_combined = nvdev->max_chn;
813 channel->combined_count = nvdev->num_chn;
814 }
815}
816
b5960e6e
AS		 817static int netvsc_set_channels(struct net_device *net,
818 struct ethtool_channels *channels)
819{
820 struct net_device_context *net_device_ctx = netdev_priv(net);
821 struct hv_device *dev = net_device_ctx->device_ctx;
545a8e79 822 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
7ca45933 823 unsigned int orig, count = channels->combined_count;
824 struct netvsc_device_info device_info;
ea383bf1 825 bool was_opened;
7ca45933 826 int ret = 0;
2b01888d 827
828 /* We do not support separate count for rx, tx, or other */
829 if (count == 0 ||
830 channels->rx_count || channels->tx_count || channels->other_count)
831 return -EINVAL;
832
b92b7d33 833 if (count > net->num_tx_queues || count > VRSS_CHANNEL_MAX)
2b01888d 834 return -EINVAL;
b5960e6e 835
a0be450e 836 if (!nvdev || nvdev->destroy)
b5960e6e
AS		 837 return -ENODEV;
838
2b01888d 839 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
b5960e6e 840 return -EINVAL;
b5960e6e 841
2b01888d 842 if (count > nvdev->max_chn)
b5960e6e
AS		 843 return -EINVAL;
844
7ca45933 845 orig = nvdev->num_chn;
ea383bf1 846 was_opened = rndis_filter_opened(nvdev);
847 if (was_opened)
848 rndis_filter_close(nvdev);
b5960e6e 849
7ca45933 850 memset(&device_info, 0, sizeof(device_info));
851 device_info.num_chn = count;
852 device_info.ring_size = ring_size;
8b532797 853 device_info.send_sections = nvdev->send_section_cnt;
854 device_info.recv_sections = nvdev->recv_section_cnt;
855
856 rndis_filter_device_remove(dev, nvdev);
7ca45933 857
858 nvdev = rndis_filter_device_add(dev, &device_info);
859 if (!IS_ERR(nvdev)) {
860 netif_set_real_num_tx_queues(net, nvdev->num_chn);
861 netif_set_real_num_rx_queues(net, nvdev->num_chn);
7ca45933 862 } else {
d6aac1f2 863 ret = PTR_ERR(nvdev);
7ca45933 864 device_info.num_chn = orig;
68d715f6 865 nvdev = rndis_filter_device_add(dev, &device_info);
866
867 if (IS_ERR(nvdev)) {
868 netdev_err(net, "restoring channel setting failed: %ld\n",
869 PTR_ERR(nvdev));
870 return ret;
871 }
7ca45933 872 }
b5960e6e 873
ea383bf1 874 if (was_opened)
875 rndis_filter_open(nvdev);
163891d7 876
1bdcec8a 877 /* We may have missed link change notifications */
1b01994a 878 net_device_ctx->last_reconfig = 0;
1bdcec8a 879 schedule_delayed_work(&net_device_ctx->dwork, 0);
b5960e6e
AS		 880
881 return ret;
b5960e6e
AS		 882}
883
5e8456fd
PR		 884static bool
885netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
49eb9389 886{
5e8456fd
PR		 887 struct ethtool_link_ksettings diff1 = *cmd;
888 struct ethtool_link_ksettings diff2 = {};
49eb9389 889
5e8456fd
PR		 890 diff1.base.speed = 0;
891 diff1.base.duplex = 0;
49eb9389 892 /* advertising and cmd are usually set */
5e8456fd
PR		 893 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
894 diff1.base.cmd = 0;
49eb9389 895 /* We set port to PORT_OTHER */
5e8456fd 896 diff2.base.port = PORT_OTHER;
49eb9389 897
898 return !memcmp(&diff1, &diff2, sizeof(diff1));
899}
900
901static void netvsc_init_settings(struct net_device *dev)
902{
903 struct net_device_context *ndc = netdev_priv(dev);
904
4823eb2f
HZ		 905 ndc->udp4_l4_hash = true;
906 ndc->udp6_l4_hash = true;
907
49eb9389 908 ndc->speed = SPEED_UNKNOWN;
f3c9d40e 909 ndc->duplex = DUPLEX_FULL;
49eb9389 910}
911
5e8456fd
PR		 912static int netvsc_get_link_ksettings(struct net_device *dev,
913 struct ethtool_link_ksettings *cmd)
49eb9389 914{
915 struct net_device_context *ndc = netdev_priv(dev);
916
5e8456fd
PR		 917 cmd->base.speed = ndc->speed;
918 cmd->base.duplex = ndc->duplex;
919 cmd->base.port = PORT_OTHER;
49eb9389 920
921 return 0;
922}
923
5e8456fd
PR		 924static int netvsc_set_link_ksettings(struct net_device *dev,
925 const struct ethtool_link_ksettings *cmd)
49eb9389 926{
927 struct net_device_context *ndc = netdev_priv(dev);
928 u32 speed;
929
5e8456fd 930 speed = cmd->base.speed;
49eb9389 931 if (!ethtool_validate_speed(speed) ||
5e8456fd 932 !ethtool_validate_duplex(cmd->base.duplex) ||
49eb9389 933 !netvsc_validate_ethtool_ss_cmd(cmd))
934 return -EINVAL;
935
936 ndc->speed = speed;
5e8456fd 937 ndc->duplex = cmd->base.duplex;
49eb9389 938
939 return 0;
940}
941
4d447c9a
HZ		 942static int netvsc_change_mtu(struct net_device *ndev, int mtu)
943{
944 struct net_device_context *ndevctx = netdev_priv(ndev);
0c195567 945 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
545a8e79 946 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
3d541ac5 947 struct hv_device *hdev = ndevctx->device_ctx;
9749fed5 948 int orig_mtu = ndev->mtu;
4d447c9a 949 struct netvsc_device_info device_info;
ea383bf1 950 bool was_opened;
9749fed5 951 int ret = 0;
4d447c9a 952
a0be450e 953 if (!nvdev || nvdev->destroy)
4d447c9a
HZ		 954 return -ENODEV;
955
0c195567 956 /* Change MTU of underlying VF netdev first. */
957 if (vf_netdev) {
958 ret = dev_set_mtu(vf_netdev, mtu);
959 if (ret)
960 return ret;
961 }
962
ea383bf1 963 netif_device_detach(ndev);
964 was_opened = rndis_filter_opened(nvdev);
965 if (was_opened)
966 rndis_filter_close(nvdev);
2de8530b 967
152669bd
DC		 968 memset(&device_info, 0, sizeof(device_info));
969 device_info.ring_size = ring_size;
970 device_info.num_chn = nvdev->num_chn;
8b532797 971 device_info.send_sections = nvdev->send_section_cnt;
972 device_info.recv_sections = nvdev->recv_section_cnt;
152669bd 973
2289f0aa 974 rndis_filter_device_remove(hdev, nvdev);
4d447c9a
HZ		 975
976 ndev->mtu = mtu;
977
9749fed5 978 nvdev = rndis_filter_device_add(hdev, &device_info);
979 if (IS_ERR(nvdev)) {
980 ret = PTR_ERR(nvdev);
981
982 /* Attempt rollback to original MTU */
983 ndev->mtu = orig_mtu;
68d715f6 984 nvdev = rndis_filter_device_add(hdev, &device_info);
0c195567 985
986 if (vf_netdev)
987 dev_set_mtu(vf_netdev, orig_mtu);
68d715f6 988
989 if (IS_ERR(nvdev)) {
990 netdev_err(ndev, "restoring mtu failed: %ld\n",
991 PTR_ERR(nvdev));
992 return ret;
993 }
9749fed5 994 }
4d447c9a 995
ea383bf1 996 if (was_opened)
997 rndis_filter_open(nvdev);
998
999 netif_device_attach(ndev);
163891d7 1000
1bdcec8a
VK		 1001 /* We may have missed link change notifications */
1002 schedule_delayed_work(&ndevctx->dwork, 0);
1003
9749fed5 1004 return ret;
4d447c9a
HZ		 1005}
1006
0c195567 1007static void netvsc_get_vf_stats(struct net_device *net,
1008 struct netvsc_vf_pcpu_stats *tot)
1009{
1010 struct net_device_context *ndev_ctx = netdev_priv(net);
1011 int i;
1012
1013 memset(tot, 0, sizeof(*tot));
1014
1015 for_each_possible_cpu(i) {
1016 const struct netvsc_vf_pcpu_stats *stats
1017 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1018 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1019 unsigned int start;
1020
1021 do {
1022 start = u64_stats_fetch_begin_irq(&stats->syncp);
1023 rx_packets = stats->rx_packets;
1024 tx_packets = stats->tx_packets;
1025 rx_bytes = stats->rx_bytes;
1026 tx_bytes = stats->tx_bytes;
1027 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1028
1029 tot->rx_packets += rx_packets;
1030 tot->tx_packets += tx_packets;
1031 tot->rx_bytes += rx_bytes;
1032 tot->tx_bytes += tx_bytes;
1033 tot->tx_dropped += stats->tx_dropped;
1034 }
1035}
1036
bc1f4470 1037static void netvsc_get_stats64(struct net_device *net,
1038 struct rtnl_link_stats64 *t)
7eafd9b4 1039{
1040 struct net_device_context *ndev_ctx = netdev_priv(net);
776e726b 1041 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
0c195567 1042 struct netvsc_vf_pcpu_stats vf_tot;
89bb42b1 1043 int i;
6c80f3fc
SX		 1044
1045 if (!nvdev)
1046 return;
1047
0c195567 1048 netdev_stats_to_stats64(t, &net->stats);
1049
1050 netvsc_get_vf_stats(net, &vf_tot);
1051 t->rx_packets += vf_tot.rx_packets;
1052 t->tx_packets += vf_tot.tx_packets;
1053 t->rx_bytes += vf_tot.rx_bytes;
1054 t->tx_bytes += vf_tot.tx_bytes;
1055 t->tx_dropped += vf_tot.tx_dropped;
1056
6c80f3fc
SX		 1057 for (i = 0; i < nvdev->num_chn; i++) {
1058 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1059 const struct netvsc_stats *stats;
1060 u64 packets, bytes, multicast;
7eafd9b4 1061 unsigned int start;
1062
6c80f3fc 1063 stats = &nvchan->tx_stats;
7eafd9b4 1064 do {
6c80f3fc
SX		 1065 start = u64_stats_fetch_begin_irq(&stats->syncp);
1066 packets = stats->packets;
1067 bytes = stats->bytes;
1068 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1069
1070 t->tx_bytes += bytes;
1071 t->tx_packets += packets;
7eafd9b4 1072
6c80f3fc 1073 stats = &nvchan->rx_stats;
7eafd9b4 1074 do {
6c80f3fc
SX		 1075 start = u64_stats_fetch_begin_irq(&stats->syncp);
1076 packets = stats->packets;
1077 bytes = stats->bytes;
1078 multicast = stats->multicast + stats->broadcast;
1079 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1080
1081 t->rx_bytes += bytes;
1082 t->rx_packets += packets;
1083 t->multicast += multicast;
7eafd9b4 1084 }
7eafd9b4 1085}
1ce09e89
HZ		 1086
1087static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1088{
867047c4 1089 struct net_device_context *ndc = netdev_priv(ndev);
16ba3266 1090 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
867047c4 1091 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1ce09e89 1092 struct sockaddr *addr = p;
1ce09e89
HZ		 1093 int err;
1094
16ba3266 1095 err = eth_prepare_mac_addr_change(ndev, p);
1096 if (err)
1ce09e89
HZ		 1097 return err;
1098
867047c4 1099 if (!nvdev)
1100 return -ENODEV;
1101
16ba3266 1102 if (vf_netdev) {
1103 err = dev_set_mac_address(vf_netdev, addr);
1104 if (err)
1105 return err;
1106 }
1107
867047c4 1108 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
16ba3266 1109 if (!err) {
1110 eth_commit_mac_addr_change(ndev, p);
1111 } else if (vf_netdev) {
1112 /* rollback change on VF */
1113 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1114 dev_set_mac_address(vf_netdev, addr);
1ce09e89
HZ		 1115 }
1116
1117 return err;
1118}
1119
4323b47c
SH		 1120static const struct {
1121 char name[ETH_GSTRING_LEN];
1122 u16 offset;
1123} netvsc_stats[] = {
1124 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1125 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1126 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1127 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1128 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
cad5c197 1129 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1130 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
0c195567 1131}, vf_stats[] = {
1132 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1133 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1134 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1135 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1136 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
4323b47c
SH		 1137};
1138
6c80f3fc 1139#define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
0c195567 1140#define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
6c80f3fc
SX		 1141
1142/* 4 statistics per queue (rx/tx packets/bytes) */
1143#define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1144
4323b47c
SH		 1145static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1146{
6c80f3fc 1147 struct net_device_context *ndc = netdev_priv(dev);
fbd4c7e7 1148 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1149
1150 if (!nvdev)
1151 return -ENODEV;
6c80f3fc 1152
4323b47c
SH		 1153 switch (string_set) {
1154 case ETH_SS_STATS:
0c195567 1155 return NETVSC_GLOBAL_STATS_LEN
1156 + NETVSC_VF_STATS_LEN
1157 + NETVSC_QUEUE_STATS_LEN(nvdev);
4323b47c
SH		 1158 default:
1159 return -EINVAL;
1160 }
1161}
1162
1163static void netvsc_get_ethtool_stats(struct net_device *dev,
1164 struct ethtool_stats *stats, u64 *data)
1165{
1166 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1167 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
4323b47c 1168 const void *nds = &ndc->eth_stats;
6c80f3fc 1169 const struct netvsc_stats *qstats;
0c195567 1170 struct netvsc_vf_pcpu_stats sum;
6c80f3fc
SX		 1171 unsigned int start;
1172 u64 packets, bytes;
1173 int i, j;
4323b47c 1174
545a8e79 1175 if (!nvdev)
1176 return;
1177
6c80f3fc 1178 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
4323b47c 1179 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
6c80f3fc 1180
0c195567 1181 netvsc_get_vf_stats(dev, &sum);
1182 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1183 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1184
6c80f3fc
SX		 1185 for (j = 0; j < nvdev->num_chn; j++) {
1186 qstats = &nvdev->chan_table[j].tx_stats;
1187
1188 do {
1189 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1190 packets = qstats->packets;
1191 bytes = qstats->bytes;
1192 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1193 data[i++] = packets;
1194 data[i++] = bytes;
1195
1196 qstats = &nvdev->chan_table[j].rx_stats;
1197 do {
1198 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1199 packets = qstats->packets;
1200 bytes = qstats->bytes;
1201 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1202 data[i++] = packets;
1203 data[i++] = bytes;
1204 }
4323b47c
SH		 1205}
1206
1207static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1208{
6c80f3fc 1209 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1210 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
6c80f3fc 1211 u8 *p = data;
4323b47c
SH		 1212 int i;
1213
545a8e79 1214 if (!nvdev)
1215 return;
1216
4323b47c
SH		 1217 switch (stringset) {
1218 case ETH_SS_STATS:
0c195567 1219 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1220 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1221 p += ETH_GSTRING_LEN;
1222 }
1223
1224 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1225 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1226 p += ETH_GSTRING_LEN;
1227 }
6c80f3fc 1228
6c80f3fc
SX		 1229 for (i = 0; i < nvdev->num_chn; i++) {
1230 sprintf(p, "tx_queue_%u_packets", i);
1231 p += ETH_GSTRING_LEN;
1232 sprintf(p, "tx_queue_%u_bytes", i);
1233 p += ETH_GSTRING_LEN;
1234 sprintf(p, "rx_queue_%u_packets", i);
1235 p += ETH_GSTRING_LEN;
1236 sprintf(p, "rx_queue_%u_bytes", i);
1237 p += ETH_GSTRING_LEN;
1238 }
1239
4323b47c
SH		 1240 break;
1241 }
1242}
1243
b5a5dc8d 1244static int
4823eb2f
HZ		 1245netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1246 struct ethtool_rxnfc *info)
b5a5dc8d 1247{
1248 info->data = RXH_IP_SRC | RXH_IP_DST;
1249
1250 switch (info->flow_type) {
1251 case TCP_V4_FLOW:
1252 case TCP_V6_FLOW:
1253 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
4823eb2f
HZ		 1254 break;
1255
b5a5dc8d 1256 case UDP_V4_FLOW:
4823eb2f
HZ		 1257 if (ndc->udp4_l4_hash)
1258 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1259
1260 break;
1261
b5a5dc8d 1262 case UDP_V6_FLOW:
4823eb2f
HZ		 1263 if (ndc->udp6_l4_hash)
1264 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1265
1266 break;
1267
b5a5dc8d 1268 case IPV4_FLOW:
1269 case IPV6_FLOW:
1270 break;
1271 default:
1272 info->data = 0;
1273 break;
1274 }
1275
1276 return 0;
1277}
1278
b448f4e8 1279static int
1280netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1281 u32 *rules)
1282{
1283 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1284 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1285
1286 if (!nvdev)
1287 return -ENODEV;
b448f4e8 1288
1289 switch (info->cmd) {
1290 case ETHTOOL_GRXRINGS:
1291 info->data = nvdev->num_chn;
1292 return 0;
b5a5dc8d 1293
1294 case ETHTOOL_GRXFH:
4823eb2f 1295 return netvsc_get_rss_hash_opts(ndc, info);
b448f4e8 1296 }
1297 return -EOPNOTSUPP;
1298}
1299
4823eb2f
HZ		 1300static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1301 struct ethtool_rxnfc *info)
1302{
1303 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1304 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1305 if (info->flow_type == UDP_V4_FLOW)
1306 ndc->udp4_l4_hash = true;
1307 else if (info->flow_type == UDP_V6_FLOW)
1308 ndc->udp6_l4_hash = true;
1309 else
1310 return -EOPNOTSUPP;
1311
1312 return 0;
1313 }
1314
1315 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1316 if (info->flow_type == UDP_V4_FLOW)
1317 ndc->udp4_l4_hash = false;
1318 else if (info->flow_type == UDP_V6_FLOW)
1319 ndc->udp6_l4_hash = false;
1320 else
1321 return -EOPNOTSUPP;
1322
1323 return 0;
1324 }
1325
1326 return -EOPNOTSUPP;
1327}
1328
1329static int
1330netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1331{
1332 struct net_device_context *ndc = netdev_priv(ndev);
1333
1334 if (info->cmd == ETHTOOL_SRXFH)
1335 return netvsc_set_rss_hash_opts(ndc, info);
1336
1337 return -EOPNOTSUPP;
1338}
1339
316158fe 1340#ifdef CONFIG_NET_POLL_CONTROLLER
a5ecd439 1341static void netvsc_poll_controller(struct net_device *dev)
316158fe 1342{
a5ecd439 1343 struct net_device_context *ndc = netdev_priv(dev);
1344 struct netvsc_device *ndev;
1345 int i;
1346
1347 rcu_read_lock();
1348 ndev = rcu_dereference(ndc->nvdev);
1349 if (ndev) {
1350 for (i = 0; i < ndev->num_chn; i++) {
1351 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1352
1353 napi_schedule(&nvchan->napi);
1354 }
1355 }
1356 rcu_read_unlock();
316158fe
RW		 1357}
1358#endif
1ce09e89 1359
962f3fee 1360static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1361{
1362 return NETVSC_HASH_KEYLEN;
1363}
1364
1365static u32 netvsc_rss_indir_size(struct net_device *dev)
1366{
ff4a4419 1367 return ITAB_NUM;
962f3fee 1368}
1369
1370static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1371 u8 *hfunc)
1372{
1373 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1374 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1375 struct rndis_device *rndis_dev;
ff4a4419 1376 int i;
962f3fee 1377
545a8e79 1378 if (!ndev)
1379 return -ENODEV;
1380
962f3fee 1381 if (hfunc)
1382 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1383
eb996edb 1384 rndis_dev = ndev->extension;
ff4a4419 1385 if (indir) {
1386 for (i = 0; i < ITAB_NUM; i++)
1387 indir[i] = rndis_dev->ind_table[i];
1388 }
1389
962f3fee 1390 if (key)
1391 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1392
1393 return 0;
1394}
1395
1396static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1397 const u8 *key, const u8 hfunc)
1398{
1399 struct net_device_context *ndc = netdev_priv(dev);
545a8e79 1400 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1401 struct rndis_device *rndis_dev;
ff4a4419 1402 int i;
962f3fee 1403
545a8e79 1404 if (!ndev)
1405 return -ENODEV;
1406
962f3fee 1407 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1408 return -EOPNOTSUPP;
1409
eb996edb 1410 rndis_dev = ndev->extension;
ff4a4419 1411 if (indir) {
1412 for (i = 0; i < ITAB_NUM; i++)
b92b7d33 1413 if (indir[i] >= VRSS_CHANNEL_MAX)
ff4a4419 1414 return -EINVAL;
1415
1416 for (i = 0; i < ITAB_NUM; i++)
1417 rndis_dev->ind_table[i] = indir[i];
1418 }
1419
1420 if (!key) {
1421 if (!indir)
1422 return 0;
1423
1424 key = rndis_dev->rss_key;
1425 }
962f3fee 1426
1427 return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1428}
1429
8b532797 1430/* Hyper-V RNDIS protocol does not have ring in the HW sense.
1431 * It does have pre-allocated receive area which is divided into sections.
1432 */
1433static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1434 struct ethtool_ringparam *ring)
1435{
1436 u32 max_buf_size;
1437
1438 ring->rx_pending = nvdev->recv_section_cnt;
1439 ring->tx_pending = nvdev->send_section_cnt;
1440
1441 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1442 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1443 else
1444 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1445
1446 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1447 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1448 / nvdev->send_section_size;
1449}
1450
1451static void netvsc_get_ringparam(struct net_device *ndev,
1452 struct ethtool_ringparam *ring)
1453{
1454 struct net_device_context *ndevctx = netdev_priv(ndev);
1455 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1456
1457 if (!nvdev)
1458 return;
1459
1460 __netvsc_get_ringparam(nvdev, ring);
1461}
1462
1463static int netvsc_set_ringparam(struct net_device *ndev,
1464 struct ethtool_ringparam *ring)
1465{
1466 struct net_device_context *ndevctx = netdev_priv(ndev);
1467 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1468 struct hv_device *hdev = ndevctx->device_ctx;
1469 struct netvsc_device_info device_info;
1470 struct ethtool_ringparam orig;
1471 u32 new_tx, new_rx;
1472 bool was_opened;
1473 int ret = 0;
1474
1475 if (!nvdev || nvdev->destroy)
1476 return -ENODEV;
1477
1478 memset(&orig, 0, sizeof(orig));
1479 __netvsc_get_ringparam(nvdev, &orig);
1480
1481 new_tx = clamp_t(u32, ring->tx_pending,
1482 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1483 new_rx = clamp_t(u32, ring->rx_pending,
1484 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1485
1486 if (new_tx == orig.tx_pending &&
1487 new_rx == orig.rx_pending)
1488 return 0; /* no change */
1489
1490 memset(&device_info, 0, sizeof(device_info));
1491 device_info.num_chn = nvdev->num_chn;
1492 device_info.ring_size = ring_size;
1493 device_info.send_sections = new_tx;
1494 device_info.recv_sections = new_rx;
1495
1496 netif_device_detach(ndev);
1497 was_opened = rndis_filter_opened(nvdev);
1498 if (was_opened)
1499 rndis_filter_close(nvdev);
1500
1501 rndis_filter_device_remove(hdev, nvdev);
1502
1503 nvdev = rndis_filter_device_add(hdev, &device_info);
1504 if (IS_ERR(nvdev)) {
1505 ret = PTR_ERR(nvdev);
1506
1507 device_info.send_sections = orig.tx_pending;
1508 device_info.recv_sections = orig.rx_pending;
1509 nvdev = rndis_filter_device_add(hdev, &device_info);
1510 if (IS_ERR(nvdev)) {
1511 netdev_err(ndev, "restoring ringparam failed: %ld\n",
1512 PTR_ERR(nvdev));
1513 return ret;
1514 }
1515 }
1516
1517 if (was_opened)
1518 rndis_filter_open(nvdev);
1519 netif_device_attach(ndev);
1520
1521 /* We may have missed link change notifications */
1522 ndevctx->last_reconfig = 0;
1523 schedule_delayed_work(&ndevctx->dwork, 0);
1524
1525 return ret;
1526}
1527
f82f4ad7
SH		 1528static const struct ethtool_ops ethtool_ops = {
1529 .get_drvinfo = netvsc_get_drvinfo,
f82f4ad7 1530 .get_link = ethtool_op_get_link,
4323b47c
SH		 1531 .get_ethtool_stats = netvsc_get_ethtool_stats,
1532 .get_sset_count = netvsc_get_sset_count,
1533 .get_strings = netvsc_get_strings,
59995370 1534 .get_channels = netvsc_get_channels,
b5960e6e 1535 .set_channels = netvsc_set_channels,
76d13b56 1536 .get_ts_info = ethtool_op_get_ts_info,
b448f4e8 1537 .get_rxnfc = netvsc_get_rxnfc,
4823eb2f 1538 .set_rxnfc = netvsc_set_rxnfc,
962f3fee 1539 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1540 .get_rxfh_indir_size = netvsc_rss_indir_size,
1541 .get_rxfh = netvsc_get_rxfh,
1542 .set_rxfh = netvsc_set_rxfh,
5e8456fd
PR		 1543 .get_link_ksettings = netvsc_get_link_ksettings,
1544 .set_link_ksettings = netvsc_set_link_ksettings,
8b532797 1545 .get_ringparam = netvsc_get_ringparam,
1546 .set_ringparam = netvsc_set_ringparam,
f82f4ad7
SH		 1547};
1548
df2fff28
GKH		 1549static const struct net_device_ops device_ops = {
1550 .ndo_open = netvsc_open,
1551 .ndo_stop = netvsc_close,
1552 .ndo_start_xmit = netvsc_start_xmit,
afc4b13d 1553 .ndo_set_rx_mode = netvsc_set_multicast_list,
4d447c9a 1554 .ndo_change_mtu = netvsc_change_mtu,
b681b588 1555 .ndo_validate_addr = eth_validate_addr,
1ce09e89 1556 .ndo_set_mac_address = netvsc_set_mac_addr,
5b54dac8 1557 .ndo_select_queue = netvsc_select_queue,
7eafd9b4 1558 .ndo_get_stats64 = netvsc_get_stats64,
316158fe
RW		 1559#ifdef CONFIG_NET_POLL_CONTROLLER
1560 .ndo_poll_controller = netvsc_poll_controller,
1561#endif
df2fff28
GKH		 1562};
1563
c996edcf 1564/*
27a70af3
VK		 1565 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1566 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1567 * present send GARP packet to network peers with netif_notify_peers().
c996edcf 1568 */
891de74d 1569static void netvsc_link_change(struct work_struct *w)
c996edcf 1570{
0a1275ca
VK		 1571 struct net_device_context *ndev_ctx =
1572 container_of(w, struct net_device_context, dwork.work);
1573 struct hv_device *device_obj = ndev_ctx->device_ctx;
1574 struct net_device *net = hv_get_drvdata(device_obj);
2ddd5e5f 1575 struct netvsc_device *net_device;
891de74d 1576 struct rndis_device *rdev;
27a70af3
VK		 1577 struct netvsc_reconfig *event = NULL;
1578 bool notify = false, reschedule = false;
1579 unsigned long flags, next_reconfig, delay;
c996edcf 1580
9b4e946c 1581 /* if changes are happening, comeback later */
1582 if (!rtnl_trylock()) {
1583 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1584 return;
1585 }
1586
a0be450e 1587 net_device = rtnl_dereference(ndev_ctx->nvdev);
1588 if (!net_device)
1bdcec8a
VK		 1589 goto out_unlock;
1590
891de74d 1591 rdev = net_device->extension;
891de74d 1592
27a70af3
VK		 1593 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1594 if (time_is_after_jiffies(next_reconfig)) {
1595 /* link_watch only sends one notification with current state
1596 * per second, avoid doing reconfig more frequently. Handle
1597 * wrap around.
1598 */
1599 delay = next_reconfig - jiffies;
1600 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1601 schedule_delayed_work(&ndev_ctx->dwork, delay);
1bdcec8a 1602 goto out_unlock;
27a70af3
VK		 1603 }
1604 ndev_ctx->last_reconfig = jiffies;
1605
1606 spin_lock_irqsave(&ndev_ctx->lock, flags);
1607 if (!list_empty(&ndev_ctx->reconfig_events)) {
1608 event = list_first_entry(&ndev_ctx->reconfig_events,
1609 struct netvsc_reconfig, list);
1610 list_del(&event->list);
1611 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1612 }
1613 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1614
1615 if (!event)
1bdcec8a 1616 goto out_unlock;
27a70af3
VK		 1617
1618 switch (event->event) {
1619 /* Only the following events are possible due to the check in
1620 * netvsc_linkstatus_callback()
1621 */
1622 case RNDIS_STATUS_MEDIA_CONNECT:
1623 if (rdev->link_state) {
1624 rdev->link_state = false;
0c195567 1625 netif_carrier_on(net);
27a70af3
VK		 1626 netif_tx_wake_all_queues(net);
1627 } else {
1628 notify = true;
1629 }
1630 kfree(event);
1631 break;
1632 case RNDIS_STATUS_MEDIA_DISCONNECT:
1633 if (!rdev->link_state) {
1634 rdev->link_state = true;
1635 netif_carrier_off(net);
1636 netif_tx_stop_all_queues(net);
1637 }
1638 kfree(event);
1639 break;
1640 case RNDIS_STATUS_NETWORK_CHANGE:
1641 /* Only makes sense if carrier is present */
1642 if (!rdev->link_state) {
1643 rdev->link_state = true;
1644 netif_carrier_off(net);
1645 netif_tx_stop_all_queues(net);
1646 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1647 spin_lock_irqsave(&ndev_ctx->lock, flags);
15cfd407 1648 list_add(&event->list, &ndev_ctx->reconfig_events);
27a70af3
VK		 1649 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1650 reschedule = true;
3a494e71 1651 }
27a70af3 1652 break;
891de74d
HZ		 1653 }
1654
1655 rtnl_unlock();
1656
1657 if (notify)
1658 netdev_notify_peers(net);
27a70af3
VK		 1659
1660 /* link_watch only sends one notification with current state per
1661 * second, handle next reconfig event in 2 seconds.
1662 */
1663 if (reschedule)
1664 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1bdcec8a
VK		 1665
1666 return;
1667
1668out_unlock:
1669 rtnl_unlock();
c996edcf
HZ		 1670}
1671
e8ff40d4 1672static struct net_device *get_netvsc_bymac(const u8 *mac)
84bf9cef 1673{
e8ff40d4 1674 struct net_device *dev;
84bf9cef 1675
8737caaf 1676 ASSERT_RTNL();
84bf9cef
KS		 1677
1678 for_each_netdev(&init_net, dev) {
e8ff40d4
SH		 1679 if (dev->netdev_ops != &device_ops)
1680 continue; /* not a netvsc device */
1681
1682 if (ether_addr_equal(mac, dev->perm_addr))
1683 return dev;
1684 }
1685
1686 return NULL;
1687}
1688
f207c10d 1689static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
e8ff40d4
SH		 1690{
1691 struct net_device *dev;
1692
1693 ASSERT_RTNL();
1694
1695 for_each_netdev(&init_net, dev) {
1696 struct net_device_context *net_device_ctx;
1697
1698 if (dev->netdev_ops != &device_ops)
1699 continue; /* not a netvsc device */
1700
1701 net_device_ctx = netdev_priv(dev);
79e8cbe7 1702 if (!rtnl_dereference(net_device_ctx->nvdev))
e8ff40d4
SH		 1703 continue; /* device is removed */
1704
f207c10d 1705 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
e8ff40d4 1706 return dev; /* a match */
84bf9cef 1707 }
84bf9cef 1708
e8ff40d4 1709 return NULL;
84bf9cef
KS		 1710}
1711
0c195567 1712/* Called when VF is injecting data into network stack.
1713 * Change the associated network device from VF to netvsc.
1714 * note: already called with rcu_read_lock
1715 */
1716static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1717{
1718 struct sk_buff *skb = *pskb;
1719 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1720 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1721 struct netvsc_vf_pcpu_stats *pcpu_stats
1722 = this_cpu_ptr(ndev_ctx->vf_stats);
1723
1724 skb->dev = ndev;
1725
1726 u64_stats_update_begin(&pcpu_stats->syncp);
1727 pcpu_stats->rx_packets++;
1728 pcpu_stats->rx_bytes += skb->len;
1729 u64_stats_update_end(&pcpu_stats->syncp);
1730
1731 return RX_HANDLER_ANOTHER;
1732}
1733
1734static int netvsc_vf_join(struct net_device *vf_netdev,
1735 struct net_device *ndev)
1736{
1737 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1738 int ret;
1739
1740 ret = netdev_rx_handler_register(vf_netdev,
1741 netvsc_vf_handle_frame, ndev);
1742 if (ret != 0) {
1743 netdev_err(vf_netdev,
1744 "can not register netvsc VF receive handler (err = %d)\n",
1745 ret);
1746 goto rx_handler_failed;
1747 }
1748
1749 ret = netdev_upper_dev_link(vf_netdev, ndev);
1750 if (ret != 0) {
1751 netdev_err(vf_netdev,
1752 "can not set master device %s (err = %d)\n",
1753 ndev->name, ret);
1754 goto upper_link_failed;
1755 }
1756
1757 /* set slave flag before open to prevent IPv6 addrconf */
1758 vf_netdev->flags |= IFF_SLAVE;
1759
6123c668 1760 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1761
1762 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
0c195567 1763
1764 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1765 return 0;
1766
1767upper_link_failed:
1768 netdev_rx_handler_unregister(vf_netdev);
1769rx_handler_failed:
1770 return ret;
1771}
1772
1773static void __netvsc_vf_setup(struct net_device *ndev,
1774 struct net_device *vf_netdev)
1775{
1776 int ret;
1777
0c195567 1778 /* Align MTU of VF with master */
1779 ret = dev_set_mtu(vf_netdev, ndev->mtu);
1780 if (ret)
1781 netdev_warn(vf_netdev,
1782 "unable to change mtu to %u\n", ndev->mtu);
1783
1784 if (netif_running(ndev)) {
1785 ret = dev_open(vf_netdev);
1786 if (ret)
1787 netdev_warn(vf_netdev,
1788 "unable to open: %d\n", ret);
1789 }
1790}
1791
1792/* Setup VF as slave of the synthetic device.
1793 * Runs in workqueue to avoid recursion in netlink callbacks.
1794 */
1795static void netvsc_vf_setup(struct work_struct *w)
1796{
1797 struct net_device_context *ndev_ctx
6123c668 1798 = container_of(w, struct net_device_context, vf_takeover.work);
0c195567 1799 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1800 struct net_device *vf_netdev;
1801
fb84af8a 1802 if (!rtnl_trylock()) {
6123c668 1803 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
fb84af8a 1804 return;
1805 }
1806
0c195567 1807 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1808 if (vf_netdev)
1809 __netvsc_vf_setup(ndev, vf_netdev);
1810
1811 rtnl_unlock();
1812}
1813
84bf9cef
KS		 1814static int netvsc_register_vf(struct net_device *vf_netdev)
1815{
0a1275ca
VK		 1816 struct net_device *ndev;
1817 struct net_device_context *net_device_ctx;
84bf9cef 1818 struct netvsc_device *netvsc_dev;
84bf9cef 1819
e8ff40d4
SH		 1820 if (vf_netdev->addr_len != ETH_ALEN)
1821 return NOTIFY_DONE;
1822
84bf9cef
KS		 1823 /*
1824 * We will use the MAC address to locate the synthetic interface to
1825 * associate with the VF interface. If we don't find a matching
1826 * synthetic interface, move on.
1827 */
e8ff40d4 1828 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
0a1275ca
VK		 1829 if (!ndev)
1830 return NOTIFY_DONE;
1831
1832 net_device_ctx = netdev_priv(ndev);
545a8e79 1833 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
f207c10d 1834 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
84bf9cef
KS		 1835 return NOTIFY_DONE;
1836
0c195567 1837 if (netvsc_vf_join(vf_netdev, ndev) != 0)
1838 return NOTIFY_DONE;
1839
0a1275ca 1840 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
0c195567 1841
1842 /* Prevent this module from being unloaded while VF is registered */
84bf9cef 1843 try_module_get(THIS_MODULE);
07d0f000
SH		 1844
1845 dev_hold(vf_netdev);
f207c10d 1846 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
84bf9cef
KS		 1847 return NOTIFY_OK;
1848}
1849
9a0c48df
SH		 1850/* VF up/down change detected, schedule to change data path */
1851static int netvsc_vf_changed(struct net_device *vf_netdev)
84bf9cef 1852{
7b83f520 1853 struct net_device_context *net_device_ctx;
84bf9cef 1854 struct netvsc_device *netvsc_dev;
7b83f520 1855 struct net_device *ndev;
9a0c48df 1856 bool vf_is_up = netif_running(vf_netdev);
fb84af8a 1857
7b83f520 1858 ndev = get_netvsc_byref(vf_netdev);
1859 if (!ndev)
1860 return NOTIFY_DONE;
0c195567 1861
7b83f520 1862 net_device_ctx = netdev_priv(ndev);
1863 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
0c195567 1864 if (!netvsc_dev)
7b83f520 1865 return NOTIFY_DONE;
84bf9cef 1866
9a0c48df
SH		 1867 netvsc_switch_datapath(ndev, vf_is_up);
1868 netdev_info(ndev, "Data path switched %s VF: %s\n",
1869 vf_is_up ? "to" : "from", vf_netdev->name);
84bf9cef
KS		 1870
1871 return NOTIFY_OK;
1872}
1873
84bf9cef
KS		 1874static int netvsc_unregister_vf(struct net_device *vf_netdev)
1875{
0a1275ca 1876 struct net_device *ndev;
0a1275ca 1877 struct net_device_context *net_device_ctx;
84bf9cef 1878
e8ff40d4 1879 ndev = get_netvsc_byref(vf_netdev);
0a1275ca
VK		 1880 if (!ndev)
1881 return NOTIFY_DONE;
1882
1883 net_device_ctx = netdev_priv(ndev);
6123c668 1884 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
e8ff40d4 1885
0a1275ca 1886 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
f207c10d 1887
0c195567 1888 netdev_upper_dev_unlink(vf_netdev, ndev);
f207c10d 1889 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
07d0f000 1890 dev_put(vf_netdev);
84bf9cef
KS		 1891 module_put(THIS_MODULE);
1892 return NOTIFY_OK;
1893}
1894
84946899
S		 1895static int netvsc_probe(struct hv_device *dev,
1896 const struct hv_vmbus_device_id *dev_id)
df2fff28 1897{
df2fff28
GKH		 1898 struct net_device *net = NULL;
1899 struct net_device_context *net_device_ctx;
1900 struct netvsc_device_info device_info;
5b54dac8 1901 struct netvsc_device *nvdev;
0c195567 1902 int ret = -ENOMEM;
df2fff28 1903
5b54dac8 1904 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2b01888d 1905 VRSS_CHANNEL_MAX);
df2fff28 1906 if (!net)
0c195567 1907 goto no_net;
df2fff28 1908
1b07da51
HZ		 1909 netif_carrier_off(net);
1910
b37879e6
HZ		 1911 netvsc_init_settings(net);
1912
df2fff28 1913 net_device_ctx = netdev_priv(net);
9efd21e1 1914 net_device_ctx->device_ctx = dev;
3f300ff4
SX		 1915 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1916 if (netif_msg_probe(net_device_ctx))
1917 netdev_dbg(net, "netvsc msg_enable: %d\n",
1918 net_device_ctx->msg_enable);
1919
2ddd5e5f 1920 hv_set_drvdata(dev, net);
f580aec4 1921
891de74d 1922 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
df2fff28 1923
27a70af3
VK		 1924 spin_lock_init(&net_device_ctx->lock);
1925 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
6123c668 1926 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
0c195567 1927
1928 net_device_ctx->vf_stats
1929 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
1930 if (!net_device_ctx->vf_stats)
1931 goto no_stats;
27a70af3 1932
df2fff28 1933 net->netdev_ops = &device_ops;
7ad24ea4 1934 net->ethtool_ops = &ethtool_ops;
9efd21e1 1935 SET_NETDEV_DEV(net, &dev->device);
df2fff28 1936
14a03cf8
VK		 1937 /* We always need headroom for rndis header */
1938 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1939
692e084e 1940 /* Notify the netvsc driver of the new device */
8ebdcc52 1941 memset(&device_info, 0, sizeof(device_info));
692e084e 1942 device_info.ring_size = ring_size;
3071ada4 1943 device_info.num_chn = VRSS_CHANNEL_DEFAULT;
8b532797 1944 device_info.send_sections = NETVSC_DEFAULT_TX;
1945 device_info.recv_sections = NETVSC_DEFAULT_RX;
9749fed5 1946
1947 nvdev = rndis_filter_device_add(dev, &device_info);
1948 if (IS_ERR(nvdev)) {
1949 ret = PTR_ERR(nvdev);
692e084e 1950 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
0c195567 1951 goto rndis_failed;
df2fff28 1952 }
0c195567 1953
692e084e
HZ		 1954 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1955
23312a3b 1956 /* hw_features computed in rndis_filter_device_add */
1957 net->features = net->hw_features |
1958 NETIF_F_HIGHDMA | NETIF_F_SG |
1959 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1960 net->vlan_features = net->features;
1961
5b54dac8
HZ		 1962 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1963 netif_set_real_num_rx_queues(net, nvdev->num_chn);
5b54dac8 1964
9749fed5 1965 netdev_lockdep_set_classes(net);
1966
d0c2c997
JW		 1967 /* MTU range: 68 - 1500 or 65521 */
1968 net->min_mtu = NETVSC_MTU_MIN;
1969 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1970 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1971 else
1972 net->max_mtu = ETH_DATA_LEN;
1973
a68f9614
HZ		 1974 ret = register_netdev(net);
1975 if (ret != 0) {
1976 pr_err("Unable to register netdev.\n");
0c195567 1977 goto register_failed;
a68f9614
HZ		 1978 }
1979
df2fff28 1980 return ret;
0c195567 1981
1982register_failed:
1983 rndis_filter_device_remove(dev, nvdev);
1984rndis_failed:
1985 free_percpu(net_device_ctx->vf_stats);
1986no_stats:
1987 hv_set_drvdata(dev, NULL);
1988 free_netdev(net);
1989no_net:
1990 return ret;
df2fff28
GKH		 1991}
1992
415b023a 1993static int netvsc_remove(struct hv_device *dev)
df2fff28 1994{
2ddd5e5f 1995 struct net_device *net;
122a5f64 1996 struct net_device_context *ndev_ctx;
2ddd5e5f 1997
3d541ac5 1998 net = hv_get_drvdata(dev);
df2fff28 1999
df2fff28 2000 if (net == NULL) {
415b023a 2001 dev_err(&dev->device, "No net device to remove\n");
df2fff28
GKH		 2002 return 0;
2003 }
2004
122a5f64 2005 ndev_ctx = netdev_priv(net);
3d541ac5 2006
a0be450e 2007 netif_device_detach(net);
f580aec4 2008
122a5f64
HZ		 2009 cancel_delayed_work_sync(&ndev_ctx->dwork);
2010
df2fff28
GKH		 2011 /*
2012 * Call to the vsc driver to let it know that the device is being
a0be450e 2013 * removed. Also blocks mtu and channel changes.
df2fff28 2014 */
a0be450e 2015 rtnl_lock();
79e8cbe7 2016 rndis_filter_device_remove(dev,
2017 rtnl_dereference(ndev_ctx->nvdev));
a0be450e 2018 rtnl_unlock();
2019
2020 unregister_netdev(net);
df2fff28 2021
3d541ac5
VK		 2022 hv_set_drvdata(dev, NULL);
2023
0c195567 2024 free_percpu(ndev_ctx->vf_stats);
6c80f3fc 2025 free_netdev(net);
df06bcff 2026 return 0;
df2fff28
GKH		 2027}
2028
345c4cc3 2029static const struct hv_vmbus_device_id id_table[] = {
c45cf2d4 2030 /* Network guid */
8f505944 2031 { HV_NIC_GUID, },
c45cf2d4 2032 { },
345c4cc3
S		 2033};
2034
2035MODULE_DEVICE_TABLE(vmbus, id_table);
2036
f1542a66 2037/* The one and only one */
fde0ef9b 2038static struct hv_driver netvsc_drv = {
d31b20fc 2039 .name = KBUILD_MODNAME,
345c4cc3 2040 .id_table = id_table,
fde0ef9b
S		 2041 .probe = netvsc_probe,
2042 .remove = netvsc_remove,
d4890970 2043};
f1542a66 2044
84bf9cef
KS		 2045/*
2046 * On Hyper-V, every VF interface is matched with a corresponding
2047 * synthetic interface. The synthetic interface is presented first
2048 * to the guest. When the corresponding VF instance is registered,
2049 * we will take care of switching the data path.
2050 */
2051static int netvsc_netdev_event(struct notifier_block *this,
2052 unsigned long event, void *ptr)
2053{
2054 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2055
ee837a13
SH		 2056 /* Skip our own events */
2057 if (event_dev->netdev_ops == &device_ops)
2058 return NOTIFY_DONE;
2059
2060 /* Avoid non-Ethernet type devices */
2061 if (event_dev->type != ARPHRD_ETHER)
2062 return NOTIFY_DONE;
2063
0dbff144 2064 /* Avoid Vlan dev with same MAC registering as VF */
d0d7b10b 2065 if (is_vlan_dev(event_dev))
0dbff144
VK		 2066 return NOTIFY_DONE;
2067
2068 /* Avoid Bonding master dev with same MAC registering as VF */
ee837a13
SH		 2069 if ((event_dev->priv_flags & IFF_BONDING) &&
2070 (event_dev->flags & IFF_MASTER))
cb2911fe
HZ		 2071 return NOTIFY_DONE;
2072
84bf9cef
KS		 2073 switch (event) {
2074 case NETDEV_REGISTER:
2075 return netvsc_register_vf(event_dev);
2076 case NETDEV_UNREGISTER:
2077 return netvsc_unregister_vf(event_dev);
2078 case NETDEV_UP:
84bf9cef 2079 case NETDEV_DOWN:
9a0c48df 2080 return netvsc_vf_changed(event_dev);
84bf9cef
KS		 2081 default:
2082 return NOTIFY_DONE;
2083 }
2084}
2085
2086static struct notifier_block netvsc_netdev_notifier = {
2087 .notifier_call = netvsc_netdev_event,
2088};
2089
a9869c94 2090static void __exit netvsc_drv_exit(void)
fceaf24a 2091{
84bf9cef 2092 unregister_netdevice_notifier(&netvsc_netdev_notifier);
768fa219 2093 vmbus_driver_unregister(&netvsc_drv);
fceaf24a
HJ		 2094}
2095
1fde28cf 2096static int __init netvsc_drv_init(void)
df2fff28 2097{
84bf9cef
KS		 2098 int ret;
2099
fa85a6c2
HZ		 2100 if (ring_size < RING_SIZE_MIN) {
2101 ring_size = RING_SIZE_MIN;
2102 pr_info("Increased ring_size to %d (min allowed)\n",
2103 ring_size);
2104 }
84bf9cef
KS		 2105 ret = vmbus_driver_register(&netvsc_drv);
2106
2107 if (ret)
2108 return ret;
2109
2110 register_netdevice_notifier(&netvsc_netdev_notifier);
2111 return 0;
df2fff28
GKH		 2112}
2113
26c14cc1 2114MODULE_LICENSE("GPL");
7880fc54 2115MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
fceaf24a 2116
1fde28cf 2117module_init(netvsc_drv_init);
a9869c94 2118module_exit(netvsc_drv_exit);
ubuntu-bionic-kernel mirror