Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[muen/linux.git] / drivers / net / hyperv / netvsc_drv.c
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, see <http://www.gnu.org/licenses/>.
15  *
16  * Authors:
17  *   Haiyang Zhang <haiyangz@microsoft.com>
18  *   Hank Janssen  <hjanssen@microsoft.com>
19  */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
27 #include <linux/io.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/slab.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/netpoll.h>
38
39 #include <net/arp.h>
40 #include <net/route.h>
41 #include <net/sock.h>
42 #include <net/pkt_sched.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <net/failover.h>
46
47 #include "hyperv_net.h"
48
49 #define RING_SIZE_MIN   64
50 #define RETRY_US_LO     5000
51 #define RETRY_US_HI     10000
52 #define RETRY_MAX       2000    /* >10 sec */
53
54 #define LINKCHANGE_INT (2 * HZ)
55 #define VF_TAKEOVER_INT (HZ / 10)
56
57 static unsigned int ring_size __ro_after_init = 128;
58 module_param(ring_size, uint, 0444);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
60 unsigned int netvsc_ring_bytes __ro_after_init;
61
62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
63                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
64                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
65                                 NETIF_MSG_TX_ERR;
66
67 static int debug = -1;
68 module_param(debug, int, 0444);
69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
70
71 static void netvsc_change_rx_flags(struct net_device *net, int change)
72 {
73         struct net_device_context *ndev_ctx = netdev_priv(net);
74         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
75         int inc;
76
77         if (!vf_netdev)
78                 return;
79
80         if (change & IFF_PROMISC) {
81                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
82                 dev_set_promiscuity(vf_netdev, inc);
83         }
84
85         if (change & IFF_ALLMULTI) {
86                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
87                 dev_set_allmulti(vf_netdev, inc);
88         }
89 }
90
91 static void netvsc_set_rx_mode(struct net_device *net)
92 {
93         struct net_device_context *ndev_ctx = netdev_priv(net);
94         struct net_device *vf_netdev;
95         struct netvsc_device *nvdev;
96
97         rcu_read_lock();
98         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
99         if (vf_netdev) {
100                 dev_uc_sync(vf_netdev, net);
101                 dev_mc_sync(vf_netdev, net);
102         }
103
104         nvdev = rcu_dereference(ndev_ctx->nvdev);
105         if (nvdev)
106                 rndis_filter_update(nvdev);
107         rcu_read_unlock();
108 }
109
110 static int netvsc_open(struct net_device *net)
111 {
112         struct net_device_context *ndev_ctx = netdev_priv(net);
113         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
114         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
115         struct rndis_device *rdev;
116         int ret = 0;
117
118         netif_carrier_off(net);
119
120         /* Open up the device */
121         ret = rndis_filter_open(nvdev);
122         if (ret != 0) {
123                 netdev_err(net, "unable to open device (ret %d).\n", ret);
124                 return ret;
125         }
126
127         rdev = nvdev->extension;
128         if (!rdev->link_state) {
129                 netif_carrier_on(net);
130                 netif_tx_wake_all_queues(net);
131         }
132
133         if (vf_netdev) {
134                 /* Setting synthetic device up transparently sets
135                  * slave as up. If open fails, then slave will be
136                  * still be offline (and not used).
137                  */
138                 ret = dev_open(vf_netdev);
139                 if (ret)
140                         netdev_warn(net,
141                                     "unable to open slave: %s: %d\n",
142                                     vf_netdev->name, ret);
143         }
144         return 0;
145 }
146
147 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
148 {
149         unsigned int retry = 0;
150         int i;
151
152         /* Ensure pending bytes in ring are read */
153         for (;;) {
154                 u32 aread = 0;
155
156                 for (i = 0; i < nvdev->num_chn; i++) {
157                         struct vmbus_channel *chn
158                                 = nvdev->chan_table[i].channel;
159
160                         if (!chn)
161                                 continue;
162
163                         /* make sure receive not running now */
164                         napi_synchronize(&nvdev->chan_table[i].napi);
165
166                         aread = hv_get_bytes_to_read(&chn->inbound);
167                         if (aread)
168                                 break;
169
170                         aread = hv_get_bytes_to_read(&chn->outbound);
171                         if (aread)
172                                 break;
173                 }
174
175                 if (aread == 0)
176                         return 0;
177
178                 if (++retry > RETRY_MAX)
179                         return -ETIMEDOUT;
180
181                 usleep_range(RETRY_US_LO, RETRY_US_HI);
182         }
183 }
184
185 static int netvsc_close(struct net_device *net)
186 {
187         struct net_device_context *net_device_ctx = netdev_priv(net);
188         struct net_device *vf_netdev
189                 = rtnl_dereference(net_device_ctx->vf_netdev);
190         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
191         int ret;
192
193         netif_tx_disable(net);
194
195         /* No need to close rndis filter if it is removed already */
196         if (!nvdev)
197                 return 0;
198
199         ret = rndis_filter_close(nvdev);
200         if (ret != 0) {
201                 netdev_err(net, "unable to close device (ret %d).\n", ret);
202                 return ret;
203         }
204
205         ret = netvsc_wait_until_empty(nvdev);
206         if (ret)
207                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
208
209         if (vf_netdev)
210                 dev_close(vf_netdev);
211
212         return ret;
213 }
214
215 static inline void *init_ppi_data(struct rndis_message *msg,
216                                   u32 ppi_size, u32 pkt_type)
217 {
218         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
219         struct rndis_per_packet_info *ppi;
220
221         rndis_pkt->data_offset += ppi_size;
222         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
223                 + rndis_pkt->per_pkt_info_len;
224
225         ppi->size = ppi_size;
226         ppi->type = pkt_type;
227         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
228
229         rndis_pkt->per_pkt_info_len += ppi_size;
230
231         return ppi + 1;
232 }
233
234 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
235  * packets. We can use ethtool to change UDP hash level when necessary.
236  */
237 static inline u32 netvsc_get_hash(
238         struct sk_buff *skb,
239         const struct net_device_context *ndc)
240 {
241         struct flow_keys flow;
242         u32 hash, pkt_proto = 0;
243         static u32 hashrnd __read_mostly;
244
245         net_get_random_once(&hashrnd, sizeof(hashrnd));
246
247         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
248                 return 0;
249
250         switch (flow.basic.ip_proto) {
251         case IPPROTO_TCP:
252                 if (flow.basic.n_proto == htons(ETH_P_IP))
253                         pkt_proto = HV_TCP4_L4HASH;
254                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
255                         pkt_proto = HV_TCP6_L4HASH;
256
257                 break;
258
259         case IPPROTO_UDP:
260                 if (flow.basic.n_proto == htons(ETH_P_IP))
261                         pkt_proto = HV_UDP4_L4HASH;
262                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
263                         pkt_proto = HV_UDP6_L4HASH;
264
265                 break;
266         }
267
268         if (pkt_proto & ndc->l4_hash) {
269                 return skb_get_hash(skb);
270         } else {
271                 if (flow.basic.n_proto == htons(ETH_P_IP))
272                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
273                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
274                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
275                 else
276                         hash = 0;
277
278                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
279         }
280
281         return hash;
282 }
283
284 static inline int netvsc_get_tx_queue(struct net_device *ndev,
285                                       struct sk_buff *skb, int old_idx)
286 {
287         const struct net_device_context *ndc = netdev_priv(ndev);
288         struct sock *sk = skb->sk;
289         int q_idx;
290
291         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
292                               (VRSS_SEND_TAB_SIZE - 1)];
293
294         /* If queue index changed record the new value */
295         if (q_idx != old_idx &&
296             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
297                 sk_tx_queue_set(sk, q_idx);
298
299         return q_idx;
300 }
301
302 /*
303  * Select queue for transmit.
304  *
305  * If a valid queue has already been assigned, then use that.
306  * Otherwise compute tx queue based on hash and the send table.
307  *
308  * This is basically similar to default (__netdev_pick_tx) with the added step
309  * of using the host send_table when no other queue has been assigned.
310  *
311  * TODO support XPS - but get_xps_queue not exported
312  */
313 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
314 {
315         int q_idx = sk_tx_queue_get(skb->sk);
316
317         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
318                 /* If forwarding a packet, we use the recorded queue when
319                  * available for better cache locality.
320                  */
321                 if (skb_rx_queue_recorded(skb))
322                         q_idx = skb_get_rx_queue(skb);
323                 else
324                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
325         }
326
327         return q_idx;
328 }
329
330 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
331                                void *accel_priv,
332                                select_queue_fallback_t fallback)
333 {
334         struct net_device_context *ndc = netdev_priv(ndev);
335         struct net_device *vf_netdev;
336         u16 txq;
337
338         rcu_read_lock();
339         vf_netdev = rcu_dereference(ndc->vf_netdev);
340         if (vf_netdev) {
341                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
342
343                 if (vf_ops->ndo_select_queue)
344                         txq = vf_ops->ndo_select_queue(vf_netdev, skb,
345                                                        accel_priv, fallback);
346                 else
347                         txq = fallback(vf_netdev, skb);
348
349                 /* Record the queue selected by VF so that it can be
350                  * used for common case where VF has more queues than
351                  * the synthetic device.
352                  */
353                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
354         } else {
355                 txq = netvsc_pick_tx(ndev, skb);
356         }
357         rcu_read_unlock();
358
359         while (unlikely(txq >= ndev->real_num_tx_queues))
360                 txq -= ndev->real_num_tx_queues;
361
362         return txq;
363 }
364
365 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
366                        struct hv_page_buffer *pb)
367 {
368         int j = 0;
369
370         /* Deal with compund pages by ignoring unused part
371          * of the page.
372          */
373         page += (offset >> PAGE_SHIFT);
374         offset &= ~PAGE_MASK;
375
376         while (len > 0) {
377                 unsigned long bytes;
378
379                 bytes = PAGE_SIZE - offset;
380                 if (bytes > len)
381                         bytes = len;
382                 pb[j].pfn = page_to_pfn(page);
383                 pb[j].offset = offset;
384                 pb[j].len = bytes;
385
386                 offset += bytes;
387                 len -= bytes;
388
389                 if (offset == PAGE_SIZE && len) {
390                         page++;
391                         offset = 0;
392                         j++;
393                 }
394         }
395
396         return j + 1;
397 }
398
399 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
400                            struct hv_netvsc_packet *packet,
401                            struct hv_page_buffer *pb)
402 {
403         u32 slots_used = 0;
404         char *data = skb->data;
405         int frags = skb_shinfo(skb)->nr_frags;
406         int i;
407
408         /* The packet is laid out thus:
409          * 1. hdr: RNDIS header and PPI
410          * 2. skb linear data
411          * 3. skb fragment data
412          */
413         slots_used += fill_pg_buf(virt_to_page(hdr),
414                                   offset_in_page(hdr),
415                                   len, &pb[slots_used]);
416
417         packet->rmsg_size = len;
418         packet->rmsg_pgcnt = slots_used;
419
420         slots_used += fill_pg_buf(virt_to_page(data),
421                                 offset_in_page(data),
422                                 skb_headlen(skb), &pb[slots_used]);
423
424         for (i = 0; i < frags; i++) {
425                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
426
427                 slots_used += fill_pg_buf(skb_frag_page(frag),
428                                         frag->page_offset,
429                                         skb_frag_size(frag), &pb[slots_used]);
430         }
431         return slots_used;
432 }
433
434 static int count_skb_frag_slots(struct sk_buff *skb)
435 {
436         int i, frags = skb_shinfo(skb)->nr_frags;
437         int pages = 0;
438
439         for (i = 0; i < frags; i++) {
440                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
441                 unsigned long size = skb_frag_size(frag);
442                 unsigned long offset = frag->page_offset;
443
444                 /* Skip unused frames from start of page */
445                 offset &= ~PAGE_MASK;
446                 pages += PFN_UP(offset + size);
447         }
448         return pages;
449 }
450
451 static int netvsc_get_slots(struct sk_buff *skb)
452 {
453         char *data = skb->data;
454         unsigned int offset = offset_in_page(data);
455         unsigned int len = skb_headlen(skb);
456         int slots;
457         int frag_slots;
458
459         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
460         frag_slots = count_skb_frag_slots(skb);
461         return slots + frag_slots;
462 }
463
464 static u32 net_checksum_info(struct sk_buff *skb)
465 {
466         if (skb->protocol == htons(ETH_P_IP)) {
467                 struct iphdr *ip = ip_hdr(skb);
468
469                 if (ip->protocol == IPPROTO_TCP)
470                         return TRANSPORT_INFO_IPV4_TCP;
471                 else if (ip->protocol == IPPROTO_UDP)
472                         return TRANSPORT_INFO_IPV4_UDP;
473         } else {
474                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
475
476                 if (ip6->nexthdr == IPPROTO_TCP)
477                         return TRANSPORT_INFO_IPV6_TCP;
478                 else if (ip6->nexthdr == IPPROTO_UDP)
479                         return TRANSPORT_INFO_IPV6_UDP;
480         }
481
482         return TRANSPORT_INFO_NOT_IP;
483 }
484
485 /* Send skb on the slave VF device. */
486 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
487                           struct sk_buff *skb)
488 {
489         struct net_device_context *ndev_ctx = netdev_priv(net);
490         unsigned int len = skb->len;
491         int rc;
492
493         skb->dev = vf_netdev;
494         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
495
496         rc = dev_queue_xmit(skb);
497         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
498                 struct netvsc_vf_pcpu_stats *pcpu_stats
499                         = this_cpu_ptr(ndev_ctx->vf_stats);
500
501                 u64_stats_update_begin(&pcpu_stats->syncp);
502                 pcpu_stats->tx_packets++;
503                 pcpu_stats->tx_bytes += len;
504                 u64_stats_update_end(&pcpu_stats->syncp);
505         } else {
506                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
507         }
508
509         return rc;
510 }
511
512 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
513 {
514         struct net_device_context *net_device_ctx = netdev_priv(net);
515         struct hv_netvsc_packet *packet = NULL;
516         int ret;
517         unsigned int num_data_pgs;
518         struct rndis_message *rndis_msg;
519         struct net_device *vf_netdev;
520         u32 rndis_msg_size;
521         u32 hash;
522         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
523
524         /* if VF is present and up then redirect packets
525          * already called with rcu_read_lock_bh
526          */
527         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
528         if (vf_netdev && netif_running(vf_netdev) &&
529             !netpoll_tx_running(net))
530                 return netvsc_vf_xmit(net, vf_netdev, skb);
531
532         /* We will atmost need two pages to describe the rndis
533          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
534          * of pages in a single packet. If skb is scattered around
535          * more pages we try linearizing it.
536          */
537
538         num_data_pgs = netvsc_get_slots(skb) + 2;
539
540         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
541                 ++net_device_ctx->eth_stats.tx_scattered;
542
543                 if (skb_linearize(skb))
544                         goto no_memory;
545
546                 num_data_pgs = netvsc_get_slots(skb) + 2;
547                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
548                         ++net_device_ctx->eth_stats.tx_too_big;
549                         goto drop;
550                 }
551         }
552
553         /*
554          * Place the rndis header in the skb head room and
555          * the skb->cb will be used for hv_netvsc_packet
556          * structure.
557          */
558         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
559         if (ret)
560                 goto no_memory;
561
562         /* Use the skb control buffer for building up the packet */
563         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
564                         FIELD_SIZEOF(struct sk_buff, cb));
565         packet = (struct hv_netvsc_packet *)skb->cb;
566
567         packet->q_idx = skb_get_queue_mapping(skb);
568
569         packet->total_data_buflen = skb->len;
570         packet->total_bytes = skb->len;
571         packet->total_packets = 1;
572
573         rndis_msg = (struct rndis_message *)skb->head;
574
575         /* Add the rndis header */
576         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
577         rndis_msg->msg_len = packet->total_data_buflen;
578
579         rndis_msg->msg.pkt = (struct rndis_packet) {
580                 .data_offset = sizeof(struct rndis_packet),
581                 .data_len = packet->total_data_buflen,
582                 .per_pkt_info_offset = sizeof(struct rndis_packet),
583         };
584
585         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
586
587         hash = skb_get_hash_raw(skb);
588         if (hash != 0 && net->real_num_tx_queues > 1) {
589                 u32 *hash_info;
590
591                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
592                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
593                                           NBL_HASH_VALUE);
594                 *hash_info = hash;
595         }
596
597         if (skb_vlan_tag_present(skb)) {
598                 struct ndis_pkt_8021q_info *vlan;
599
600                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
601                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
602                                      IEEE_8021Q_INFO);
603
604                 vlan->value = 0;
605                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
606                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
607                                 VLAN_PRIO_SHIFT;
608         }
609
610         if (skb_is_gso(skb)) {
611                 struct ndis_tcp_lso_info *lso_info;
612
613                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
614                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
615                                          TCP_LARGESEND_PKTINFO);
616
617                 lso_info->value = 0;
618                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
619                 if (skb->protocol == htons(ETH_P_IP)) {
620                         lso_info->lso_v2_transmit.ip_version =
621                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
622                         ip_hdr(skb)->tot_len = 0;
623                         ip_hdr(skb)->check = 0;
624                         tcp_hdr(skb)->check =
625                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
626                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
627                 } else {
628                         lso_info->lso_v2_transmit.ip_version =
629                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
630                         ipv6_hdr(skb)->payload_len = 0;
631                         tcp_hdr(skb)->check =
632                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
633                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
634                 }
635                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
636                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
637         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
638                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
639                         struct ndis_tcp_ip_checksum_info *csum_info;
640
641                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
642                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
643                                                   TCPIP_CHKSUM_PKTINFO);
644
645                         csum_info->value = 0;
646                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
647
648                         if (skb->protocol == htons(ETH_P_IP)) {
649                                 csum_info->transmit.is_ipv4 = 1;
650
651                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
652                                         csum_info->transmit.tcp_checksum = 1;
653                                 else
654                                         csum_info->transmit.udp_checksum = 1;
655                         } else {
656                                 csum_info->transmit.is_ipv6 = 1;
657
658                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
659                                         csum_info->transmit.tcp_checksum = 1;
660                                 else
661                                         csum_info->transmit.udp_checksum = 1;
662                         }
663                 } else {
664                         /* Can't do offload of this type of checksum */
665                         if (skb_checksum_help(skb))
666                                 goto drop;
667                 }
668         }
669
670         /* Start filling in the page buffers with the rndis hdr */
671         rndis_msg->msg_len += rndis_msg_size;
672         packet->total_data_buflen = rndis_msg->msg_len;
673         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
674                                                skb, packet, pb);
675
676         /* timestamp packet in software */
677         skb_tx_timestamp(skb);
678
679         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
680         if (likely(ret == 0))
681                 return NETDEV_TX_OK;
682
683         if (ret == -EAGAIN) {
684                 ++net_device_ctx->eth_stats.tx_busy;
685                 return NETDEV_TX_BUSY;
686         }
687
688         if (ret == -ENOSPC)
689                 ++net_device_ctx->eth_stats.tx_no_space;
690
691 drop:
692         dev_kfree_skb_any(skb);
693         net->stats.tx_dropped++;
694
695         return NETDEV_TX_OK;
696
697 no_memory:
698         ++net_device_ctx->eth_stats.tx_no_memory;
699         goto drop;
700 }
701
702 /*
703  * netvsc_linkstatus_callback - Link up/down notification
704  */
705 void netvsc_linkstatus_callback(struct net_device *net,
706                                 struct rndis_message *resp)
707 {
708         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
709         struct net_device_context *ndev_ctx = netdev_priv(net);
710         struct netvsc_reconfig *event;
711         unsigned long flags;
712
713         /* Update the physical link speed when changing to another vSwitch */
714         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
715                 u32 speed;
716
717                 speed = *(u32 *)((void *)indicate
718                                  + indicate->status_buf_offset) / 10000;
719                 ndev_ctx->speed = speed;
720                 return;
721         }
722
723         /* Handle these link change statuses below */
724         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
725             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
726             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
727                 return;
728
729         if (net->reg_state != NETREG_REGISTERED)
730                 return;
731
732         event = kzalloc(sizeof(*event), GFP_ATOMIC);
733         if (!event)
734                 return;
735         event->event = indicate->status;
736
737         spin_lock_irqsave(&ndev_ctx->lock, flags);
738         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
739         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
740
741         schedule_delayed_work(&ndev_ctx->dwork, 0);
742 }
743
744 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
745                                              struct napi_struct *napi,
746                                              const struct ndis_tcp_ip_checksum_info *csum_info,
747                                              const struct ndis_pkt_8021q_info *vlan,
748                                              void *data, u32 buflen)
749 {
750         struct sk_buff *skb;
751
752         skb = napi_alloc_skb(napi, buflen);
753         if (!skb)
754                 return skb;
755
756         /*
757          * Copy to skb. This copy is needed here since the memory pointed by
758          * hv_netvsc_packet cannot be deallocated
759          */
760         skb_put_data(skb, data, buflen);
761
762         skb->protocol = eth_type_trans(skb, net);
763
764         /* skb is already created with CHECKSUM_NONE */
765         skb_checksum_none_assert(skb);
766
767         /*
768          * In Linux, the IP checksum is always checked.
769          * Do L4 checksum offload if enabled and present.
770          */
771         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
772                 if (csum_info->receive.tcp_checksum_succeeded ||
773                     csum_info->receive.udp_checksum_succeeded)
774                         skb->ip_summed = CHECKSUM_UNNECESSARY;
775         }
776
777         if (vlan) {
778                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
779
780                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
781                                        vlan_tci);
782         }
783
784         return skb;
785 }
786
787 /*
788  * netvsc_recv_callback -  Callback when we receive a packet from the
789  * "wire" on the specified device.
790  */
791 int netvsc_recv_callback(struct net_device *net,
792                          struct netvsc_device *net_device,
793                          struct vmbus_channel *channel,
794                          void  *data, u32 len,
795                          const struct ndis_tcp_ip_checksum_info *csum_info,
796                          const struct ndis_pkt_8021q_info *vlan)
797 {
798         struct net_device_context *net_device_ctx = netdev_priv(net);
799         u16 q_idx = channel->offermsg.offer.sub_channel_index;
800         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
801         struct sk_buff *skb;
802         struct netvsc_stats *rx_stats;
803
804         if (net->reg_state != NETREG_REGISTERED)
805                 return NVSP_STAT_FAIL;
806
807         /* Allocate a skb - TODO direct I/O to pages? */
808         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
809                                     csum_info, vlan, data, len);
810         if (unlikely(!skb)) {
811                 ++net_device_ctx->eth_stats.rx_no_memory;
812                 rcu_read_unlock();
813                 return NVSP_STAT_FAIL;
814         }
815
816         skb_record_rx_queue(skb, q_idx);
817
818         /*
819          * Even if injecting the packet, record the statistics
820          * on the synthetic device because modifying the VF device
821          * statistics will not work correctly.
822          */
823         rx_stats = &nvchan->rx_stats;
824         u64_stats_update_begin(&rx_stats->syncp);
825         rx_stats->packets++;
826         rx_stats->bytes += len;
827
828         if (skb->pkt_type == PACKET_BROADCAST)
829                 ++rx_stats->broadcast;
830         else if (skb->pkt_type == PACKET_MULTICAST)
831                 ++rx_stats->multicast;
832         u64_stats_update_end(&rx_stats->syncp);
833
834         napi_gro_receive(&nvchan->napi, skb);
835         return NVSP_STAT_SUCCESS;
836 }
837
838 static void netvsc_get_drvinfo(struct net_device *net,
839                                struct ethtool_drvinfo *info)
840 {
841         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
842         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
843 }
844
845 static void netvsc_get_channels(struct net_device *net,
846                                 struct ethtool_channels *channel)
847 {
848         struct net_device_context *net_device_ctx = netdev_priv(net);
849         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
850
851         if (nvdev) {
852                 channel->max_combined   = nvdev->max_chn;
853                 channel->combined_count = nvdev->num_chn;
854         }
855 }
856
857 static int netvsc_detach(struct net_device *ndev,
858                          struct netvsc_device *nvdev)
859 {
860         struct net_device_context *ndev_ctx = netdev_priv(ndev);
861         struct hv_device *hdev = ndev_ctx->device_ctx;
862         int ret;
863
864         /* Don't try continuing to try and setup sub channels */
865         if (cancel_work_sync(&nvdev->subchan_work))
866                 nvdev->num_chn = 1;
867
868         /* If device was up (receiving) then shutdown */
869         if (netif_running(ndev)) {
870                 netif_tx_disable(ndev);
871
872                 ret = rndis_filter_close(nvdev);
873                 if (ret) {
874                         netdev_err(ndev,
875                                    "unable to close device (ret %d).\n", ret);
876                         return ret;
877                 }
878
879                 ret = netvsc_wait_until_empty(nvdev);
880                 if (ret) {
881                         netdev_err(ndev,
882                                    "Ring buffer not empty after closing rndis\n");
883                         return ret;
884                 }
885         }
886
887         netif_device_detach(ndev);
888
889         rndis_filter_device_remove(hdev, nvdev);
890
891         return 0;
892 }
893
894 static int netvsc_attach(struct net_device *ndev,
895                          struct netvsc_device_info *dev_info)
896 {
897         struct net_device_context *ndev_ctx = netdev_priv(ndev);
898         struct hv_device *hdev = ndev_ctx->device_ctx;
899         struct netvsc_device *nvdev;
900         struct rndis_device *rdev;
901         int ret;
902
903         nvdev = rndis_filter_device_add(hdev, dev_info);
904         if (IS_ERR(nvdev))
905                 return PTR_ERR(nvdev);
906
907         /* Note: enable and attach happen when sub-channels setup */
908
909         netif_carrier_off(ndev);
910
911         if (netif_running(ndev)) {
912                 ret = rndis_filter_open(nvdev);
913                 if (ret)
914                         return ret;
915
916                 rdev = nvdev->extension;
917                 if (!rdev->link_state)
918                         netif_carrier_on(ndev);
919         }
920
921         return 0;
922 }
923
924 static int netvsc_set_channels(struct net_device *net,
925                                struct ethtool_channels *channels)
926 {
927         struct net_device_context *net_device_ctx = netdev_priv(net);
928         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
929         unsigned int orig, count = channels->combined_count;
930         struct netvsc_device_info device_info;
931         int ret;
932
933         /* We do not support separate count for rx, tx, or other */
934         if (count == 0 ||
935             channels->rx_count || channels->tx_count || channels->other_count)
936                 return -EINVAL;
937
938         if (!nvdev || nvdev->destroy)
939                 return -ENODEV;
940
941         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
942                 return -EINVAL;
943
944         if (count > nvdev->max_chn)
945                 return -EINVAL;
946
947         orig = nvdev->num_chn;
948
949         memset(&device_info, 0, sizeof(device_info));
950         device_info.num_chn = count;
951         device_info.send_sections = nvdev->send_section_cnt;
952         device_info.send_section_size = nvdev->send_section_size;
953         device_info.recv_sections = nvdev->recv_section_cnt;
954         device_info.recv_section_size = nvdev->recv_section_size;
955
956         ret = netvsc_detach(net, nvdev);
957         if (ret)
958                 return ret;
959
960         ret = netvsc_attach(net, &device_info);
961         if (ret) {
962                 device_info.num_chn = orig;
963                 if (netvsc_attach(net, &device_info))
964                         netdev_err(net, "restoring channel setting failed\n");
965         }
966
967         return ret;
968 }
969
970 static bool
971 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
972 {
973         struct ethtool_link_ksettings diff1 = *cmd;
974         struct ethtool_link_ksettings diff2 = {};
975
976         diff1.base.speed = 0;
977         diff1.base.duplex = 0;
978         /* advertising and cmd are usually set */
979         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
980         diff1.base.cmd = 0;
981         /* We set port to PORT_OTHER */
982         diff2.base.port = PORT_OTHER;
983
984         return !memcmp(&diff1, &diff2, sizeof(diff1));
985 }
986
987 static void netvsc_init_settings(struct net_device *dev)
988 {
989         struct net_device_context *ndc = netdev_priv(dev);
990
991         ndc->l4_hash = HV_DEFAULT_L4HASH;
992
993         ndc->speed = SPEED_UNKNOWN;
994         ndc->duplex = DUPLEX_FULL;
995 }
996
997 static int netvsc_get_link_ksettings(struct net_device *dev,
998                                      struct ethtool_link_ksettings *cmd)
999 {
1000         struct net_device_context *ndc = netdev_priv(dev);
1001
1002         cmd->base.speed = ndc->speed;
1003         cmd->base.duplex = ndc->duplex;
1004         cmd->base.port = PORT_OTHER;
1005
1006         return 0;
1007 }
1008
1009 static int netvsc_set_link_ksettings(struct net_device *dev,
1010                                      const struct ethtool_link_ksettings *cmd)
1011 {
1012         struct net_device_context *ndc = netdev_priv(dev);
1013         u32 speed;
1014
1015         speed = cmd->base.speed;
1016         if (!ethtool_validate_speed(speed) ||
1017             !ethtool_validate_duplex(cmd->base.duplex) ||
1018             !netvsc_validate_ethtool_ss_cmd(cmd))
1019                 return -EINVAL;
1020
1021         ndc->speed = speed;
1022         ndc->duplex = cmd->base.duplex;
1023
1024         return 0;
1025 }
1026
1027 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1028 {
1029         struct net_device_context *ndevctx = netdev_priv(ndev);
1030         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1031         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1032         int orig_mtu = ndev->mtu;
1033         struct netvsc_device_info device_info;
1034         int ret = 0;
1035
1036         if (!nvdev || nvdev->destroy)
1037                 return -ENODEV;
1038
1039         /* Change MTU of underlying VF netdev first. */
1040         if (vf_netdev) {
1041                 ret = dev_set_mtu(vf_netdev, mtu);
1042                 if (ret)
1043                         return ret;
1044         }
1045
1046         memset(&device_info, 0, sizeof(device_info));
1047         device_info.num_chn = nvdev->num_chn;
1048         device_info.send_sections = nvdev->send_section_cnt;
1049         device_info.send_section_size = nvdev->send_section_size;
1050         device_info.recv_sections = nvdev->recv_section_cnt;
1051         device_info.recv_section_size = nvdev->recv_section_size;
1052
1053         ret = netvsc_detach(ndev, nvdev);
1054         if (ret)
1055                 goto rollback_vf;
1056
1057         ndev->mtu = mtu;
1058
1059         ret = netvsc_attach(ndev, &device_info);
1060         if (ret)
1061                 goto rollback;
1062
1063         return 0;
1064
1065 rollback:
1066         /* Attempt rollback to original MTU */
1067         ndev->mtu = orig_mtu;
1068
1069         if (netvsc_attach(ndev, &device_info))
1070                 netdev_err(ndev, "restoring mtu failed\n");
1071 rollback_vf:
1072         if (vf_netdev)
1073                 dev_set_mtu(vf_netdev, orig_mtu);
1074
1075         return ret;
1076 }
1077
1078 static void netvsc_get_vf_stats(struct net_device *net,
1079                                 struct netvsc_vf_pcpu_stats *tot)
1080 {
1081         struct net_device_context *ndev_ctx = netdev_priv(net);
1082         int i;
1083
1084         memset(tot, 0, sizeof(*tot));
1085
1086         for_each_possible_cpu(i) {
1087                 const struct netvsc_vf_pcpu_stats *stats
1088                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1089                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1090                 unsigned int start;
1091
1092                 do {
1093                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1094                         rx_packets = stats->rx_packets;
1095                         tx_packets = stats->tx_packets;
1096                         rx_bytes = stats->rx_bytes;
1097                         tx_bytes = stats->tx_bytes;
1098                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1099
1100                 tot->rx_packets += rx_packets;
1101                 tot->tx_packets += tx_packets;
1102                 tot->rx_bytes   += rx_bytes;
1103                 tot->tx_bytes   += tx_bytes;
1104                 tot->tx_dropped += stats->tx_dropped;
1105         }
1106 }
1107
1108 static void netvsc_get_stats64(struct net_device *net,
1109                                struct rtnl_link_stats64 *t)
1110 {
1111         struct net_device_context *ndev_ctx = netdev_priv(net);
1112         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1113         struct netvsc_vf_pcpu_stats vf_tot;
1114         int i;
1115
1116         if (!nvdev)
1117                 return;
1118
1119         netdev_stats_to_stats64(t, &net->stats);
1120
1121         netvsc_get_vf_stats(net, &vf_tot);
1122         t->rx_packets += vf_tot.rx_packets;
1123         t->tx_packets += vf_tot.tx_packets;
1124         t->rx_bytes   += vf_tot.rx_bytes;
1125         t->tx_bytes   += vf_tot.tx_bytes;
1126         t->tx_dropped += vf_tot.tx_dropped;
1127
1128         for (i = 0; i < nvdev->num_chn; i++) {
1129                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1130                 const struct netvsc_stats *stats;
1131                 u64 packets, bytes, multicast;
1132                 unsigned int start;
1133
1134                 stats = &nvchan->tx_stats;
1135                 do {
1136                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1137                         packets = stats->packets;
1138                         bytes = stats->bytes;
1139                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1140
1141                 t->tx_bytes     += bytes;
1142                 t->tx_packets   += packets;
1143
1144                 stats = &nvchan->rx_stats;
1145                 do {
1146                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1147                         packets = stats->packets;
1148                         bytes = stats->bytes;
1149                         multicast = stats->multicast + stats->broadcast;
1150                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1151
1152                 t->rx_bytes     += bytes;
1153                 t->rx_packets   += packets;
1154                 t->multicast    += multicast;
1155         }
1156 }
1157
1158 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1159 {
1160         struct net_device_context *ndc = netdev_priv(ndev);
1161         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1162         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1163         struct sockaddr *addr = p;
1164         int err;
1165
1166         err = eth_prepare_mac_addr_change(ndev, p);
1167         if (err)
1168                 return err;
1169
1170         if (!nvdev)
1171                 return -ENODEV;
1172
1173         if (vf_netdev) {
1174                 err = dev_set_mac_address(vf_netdev, addr);
1175                 if (err)
1176                         return err;
1177         }
1178
1179         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1180         if (!err) {
1181                 eth_commit_mac_addr_change(ndev, p);
1182         } else if (vf_netdev) {
1183                 /* rollback change on VF */
1184                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1185                 dev_set_mac_address(vf_netdev, addr);
1186         }
1187
1188         return err;
1189 }
1190
1191 static const struct {
1192         char name[ETH_GSTRING_LEN];
1193         u16 offset;
1194 } netvsc_stats[] = {
1195         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1196         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1197         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1198         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1199         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1200         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1201         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1202         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1203         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1204         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1205 }, vf_stats[] = {
1206         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1207         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1208         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1209         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1210         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1211 };
1212
1213 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1214 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1215
1216 /* 4 statistics per queue (rx/tx packets/bytes) */
1217 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1218
1219 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1220 {
1221         struct net_device_context *ndc = netdev_priv(dev);
1222         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1223
1224         if (!nvdev)
1225                 return -ENODEV;
1226
1227         switch (string_set) {
1228         case ETH_SS_STATS:
1229                 return NETVSC_GLOBAL_STATS_LEN
1230                         + NETVSC_VF_STATS_LEN
1231                         + NETVSC_QUEUE_STATS_LEN(nvdev);
1232         default:
1233                 return -EINVAL;
1234         }
1235 }
1236
1237 static void netvsc_get_ethtool_stats(struct net_device *dev,
1238                                      struct ethtool_stats *stats, u64 *data)
1239 {
1240         struct net_device_context *ndc = netdev_priv(dev);
1241         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1242         const void *nds = &ndc->eth_stats;
1243         const struct netvsc_stats *qstats;
1244         struct netvsc_vf_pcpu_stats sum;
1245         unsigned int start;
1246         u64 packets, bytes;
1247         int i, j;
1248
1249         if (!nvdev)
1250                 return;
1251
1252         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1253                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1254
1255         netvsc_get_vf_stats(dev, &sum);
1256         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1257                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1258
1259         for (j = 0; j < nvdev->num_chn; j++) {
1260                 qstats = &nvdev->chan_table[j].tx_stats;
1261
1262                 do {
1263                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1264                         packets = qstats->packets;
1265                         bytes = qstats->bytes;
1266                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1267                 data[i++] = packets;
1268                 data[i++] = bytes;
1269
1270                 qstats = &nvdev->chan_table[j].rx_stats;
1271                 do {
1272                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1273                         packets = qstats->packets;
1274                         bytes = qstats->bytes;
1275                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1276                 data[i++] = packets;
1277                 data[i++] = bytes;
1278         }
1279 }
1280
1281 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1282 {
1283         struct net_device_context *ndc = netdev_priv(dev);
1284         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1285         u8 *p = data;
1286         int i;
1287
1288         if (!nvdev)
1289                 return;
1290
1291         switch (stringset) {
1292         case ETH_SS_STATS:
1293                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1294                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1295                         p += ETH_GSTRING_LEN;
1296                 }
1297
1298                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1299                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1300                         p += ETH_GSTRING_LEN;
1301                 }
1302
1303                 for (i = 0; i < nvdev->num_chn; i++) {
1304                         sprintf(p, "tx_queue_%u_packets", i);
1305                         p += ETH_GSTRING_LEN;
1306                         sprintf(p, "tx_queue_%u_bytes", i);
1307                         p += ETH_GSTRING_LEN;
1308                         sprintf(p, "rx_queue_%u_packets", i);
1309                         p += ETH_GSTRING_LEN;
1310                         sprintf(p, "rx_queue_%u_bytes", i);
1311                         p += ETH_GSTRING_LEN;
1312                 }
1313
1314                 break;
1315         }
1316 }
1317
1318 static int
1319 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1320                          struct ethtool_rxnfc *info)
1321 {
1322         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1323
1324         info->data = RXH_IP_SRC | RXH_IP_DST;
1325
1326         switch (info->flow_type) {
1327         case TCP_V4_FLOW:
1328                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1329                         info->data |= l4_flag;
1330
1331                 break;
1332
1333         case TCP_V6_FLOW:
1334                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1335                         info->data |= l4_flag;
1336
1337                 break;
1338
1339         case UDP_V4_FLOW:
1340                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1341                         info->data |= l4_flag;
1342
1343                 break;
1344
1345         case UDP_V6_FLOW:
1346                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1347                         info->data |= l4_flag;
1348
1349                 break;
1350
1351         case IPV4_FLOW:
1352         case IPV6_FLOW:
1353                 break;
1354         default:
1355                 info->data = 0;
1356                 break;
1357         }
1358
1359         return 0;
1360 }
1361
1362 static int
1363 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1364                  u32 *rules)
1365 {
1366         struct net_device_context *ndc = netdev_priv(dev);
1367         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1368
1369         if (!nvdev)
1370                 return -ENODEV;
1371
1372         switch (info->cmd) {
1373         case ETHTOOL_GRXRINGS:
1374                 info->data = nvdev->num_chn;
1375                 return 0;
1376
1377         case ETHTOOL_GRXFH:
1378                 return netvsc_get_rss_hash_opts(ndc, info);
1379         }
1380         return -EOPNOTSUPP;
1381 }
1382
1383 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1384                                     struct ethtool_rxnfc *info)
1385 {
1386         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1387                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1388                 switch (info->flow_type) {
1389                 case TCP_V4_FLOW:
1390                         ndc->l4_hash |= HV_TCP4_L4HASH;
1391                         break;
1392
1393                 case TCP_V6_FLOW:
1394                         ndc->l4_hash |= HV_TCP6_L4HASH;
1395                         break;
1396
1397                 case UDP_V4_FLOW:
1398                         ndc->l4_hash |= HV_UDP4_L4HASH;
1399                         break;
1400
1401                 case UDP_V6_FLOW:
1402                         ndc->l4_hash |= HV_UDP6_L4HASH;
1403                         break;
1404
1405                 default:
1406                         return -EOPNOTSUPP;
1407                 }
1408
1409                 return 0;
1410         }
1411
1412         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1413                 switch (info->flow_type) {
1414                 case TCP_V4_FLOW:
1415                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1416                         break;
1417
1418                 case TCP_V6_FLOW:
1419                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1420                         break;
1421
1422                 case UDP_V4_FLOW:
1423                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1424                         break;
1425
1426                 case UDP_V6_FLOW:
1427                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1428                         break;
1429
1430                 default:
1431                         return -EOPNOTSUPP;
1432                 }
1433
1434                 return 0;
1435         }
1436
1437         return -EOPNOTSUPP;
1438 }
1439
1440 static int
1441 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1442 {
1443         struct net_device_context *ndc = netdev_priv(ndev);
1444
1445         if (info->cmd == ETHTOOL_SRXFH)
1446                 return netvsc_set_rss_hash_opts(ndc, info);
1447
1448         return -EOPNOTSUPP;
1449 }
1450
1451 #ifdef CONFIG_NET_POLL_CONTROLLER
1452 static void netvsc_poll_controller(struct net_device *dev)
1453 {
1454         struct net_device_context *ndc = netdev_priv(dev);
1455         struct netvsc_device *ndev;
1456         int i;
1457
1458         rcu_read_lock();
1459         ndev = rcu_dereference(ndc->nvdev);
1460         if (ndev) {
1461                 for (i = 0; i < ndev->num_chn; i++) {
1462                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1463
1464                         napi_schedule(&nvchan->napi);
1465                 }
1466         }
1467         rcu_read_unlock();
1468 }
1469 #endif
1470
1471 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1472 {
1473         return NETVSC_HASH_KEYLEN;
1474 }
1475
1476 static u32 netvsc_rss_indir_size(struct net_device *dev)
1477 {
1478         return ITAB_NUM;
1479 }
1480
1481 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1482                            u8 *hfunc)
1483 {
1484         struct net_device_context *ndc = netdev_priv(dev);
1485         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1486         struct rndis_device *rndis_dev;
1487         int i;
1488
1489         if (!ndev)
1490                 return -ENODEV;
1491
1492         if (hfunc)
1493                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1494
1495         rndis_dev = ndev->extension;
1496         if (indir) {
1497                 for (i = 0; i < ITAB_NUM; i++)
1498                         indir[i] = rndis_dev->rx_table[i];
1499         }
1500
1501         if (key)
1502                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1503
1504         return 0;
1505 }
1506
1507 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1508                            const u8 *key, const u8 hfunc)
1509 {
1510         struct net_device_context *ndc = netdev_priv(dev);
1511         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1512         struct rndis_device *rndis_dev;
1513         int i;
1514
1515         if (!ndev)
1516                 return -ENODEV;
1517
1518         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1519                 return -EOPNOTSUPP;
1520
1521         rndis_dev = ndev->extension;
1522         if (indir) {
1523                 for (i = 0; i < ITAB_NUM; i++)
1524                         if (indir[i] >= ndev->num_chn)
1525                                 return -EINVAL;
1526
1527                 for (i = 0; i < ITAB_NUM; i++)
1528                         rndis_dev->rx_table[i] = indir[i];
1529         }
1530
1531         if (!key) {
1532                 if (!indir)
1533                         return 0;
1534
1535                 key = rndis_dev->rss_key;
1536         }
1537
1538         return rndis_filter_set_rss_param(rndis_dev, key);
1539 }
1540
1541 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1542  * It does have pre-allocated receive area which is divided into sections.
1543  */
1544 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1545                                    struct ethtool_ringparam *ring)
1546 {
1547         u32 max_buf_size;
1548
1549         ring->rx_pending = nvdev->recv_section_cnt;
1550         ring->tx_pending = nvdev->send_section_cnt;
1551
1552         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1553                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1554         else
1555                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1556
1557         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1558         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1559                 / nvdev->send_section_size;
1560 }
1561
1562 static void netvsc_get_ringparam(struct net_device *ndev,
1563                                  struct ethtool_ringparam *ring)
1564 {
1565         struct net_device_context *ndevctx = netdev_priv(ndev);
1566         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1567
1568         if (!nvdev)
1569                 return;
1570
1571         __netvsc_get_ringparam(nvdev, ring);
1572 }
1573
1574 static int netvsc_set_ringparam(struct net_device *ndev,
1575                                 struct ethtool_ringparam *ring)
1576 {
1577         struct net_device_context *ndevctx = netdev_priv(ndev);
1578         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1579         struct netvsc_device_info device_info;
1580         struct ethtool_ringparam orig;
1581         u32 new_tx, new_rx;
1582         int ret = 0;
1583
1584         if (!nvdev || nvdev->destroy)
1585                 return -ENODEV;
1586
1587         memset(&orig, 0, sizeof(orig));
1588         __netvsc_get_ringparam(nvdev, &orig);
1589
1590         new_tx = clamp_t(u32, ring->tx_pending,
1591                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1592         new_rx = clamp_t(u32, ring->rx_pending,
1593                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1594
1595         if (new_tx == orig.tx_pending &&
1596             new_rx == orig.rx_pending)
1597                 return 0;        /* no change */
1598
1599         memset(&device_info, 0, sizeof(device_info));
1600         device_info.num_chn = nvdev->num_chn;
1601         device_info.send_sections = new_tx;
1602         device_info.send_section_size = nvdev->send_section_size;
1603         device_info.recv_sections = new_rx;
1604         device_info.recv_section_size = nvdev->recv_section_size;
1605
1606         ret = netvsc_detach(ndev, nvdev);
1607         if (ret)
1608                 return ret;
1609
1610         ret = netvsc_attach(ndev, &device_info);
1611         if (ret) {
1612                 device_info.send_sections = orig.tx_pending;
1613                 device_info.recv_sections = orig.rx_pending;
1614
1615                 if (netvsc_attach(ndev, &device_info))
1616                         netdev_err(ndev, "restoring ringparam failed");
1617         }
1618
1619         return ret;
1620 }
1621
1622 static u32 netvsc_get_msglevel(struct net_device *ndev)
1623 {
1624         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1625
1626         return ndev_ctx->msg_enable;
1627 }
1628
1629 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1630 {
1631         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1632
1633         ndev_ctx->msg_enable = val;
1634 }
1635
1636 static const struct ethtool_ops ethtool_ops = {
1637         .get_drvinfo    = netvsc_get_drvinfo,
1638         .get_msglevel   = netvsc_get_msglevel,
1639         .set_msglevel   = netvsc_set_msglevel,
1640         .get_link       = ethtool_op_get_link,
1641         .get_ethtool_stats = netvsc_get_ethtool_stats,
1642         .get_sset_count = netvsc_get_sset_count,
1643         .get_strings    = netvsc_get_strings,
1644         .get_channels   = netvsc_get_channels,
1645         .set_channels   = netvsc_set_channels,
1646         .get_ts_info    = ethtool_op_get_ts_info,
1647         .get_rxnfc      = netvsc_get_rxnfc,
1648         .set_rxnfc      = netvsc_set_rxnfc,
1649         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1650         .get_rxfh_indir_size = netvsc_rss_indir_size,
1651         .get_rxfh       = netvsc_get_rxfh,
1652         .set_rxfh       = netvsc_set_rxfh,
1653         .get_link_ksettings = netvsc_get_link_ksettings,
1654         .set_link_ksettings = netvsc_set_link_ksettings,
1655         .get_ringparam  = netvsc_get_ringparam,
1656         .set_ringparam  = netvsc_set_ringparam,
1657 };
1658
1659 static const struct net_device_ops device_ops = {
1660         .ndo_open =                     netvsc_open,
1661         .ndo_stop =                     netvsc_close,
1662         .ndo_start_xmit =               netvsc_start_xmit,
1663         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1664         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1665         .ndo_change_mtu =               netvsc_change_mtu,
1666         .ndo_validate_addr =            eth_validate_addr,
1667         .ndo_set_mac_address =          netvsc_set_mac_addr,
1668         .ndo_select_queue =             netvsc_select_queue,
1669         .ndo_get_stats64 =              netvsc_get_stats64,
1670 #ifdef CONFIG_NET_POLL_CONTROLLER
1671         .ndo_poll_controller =          netvsc_poll_controller,
1672 #endif
1673 };
1674
1675 /*
1676  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1677  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1678  * present send GARP packet to network peers with netif_notify_peers().
1679  */
1680 static void netvsc_link_change(struct work_struct *w)
1681 {
1682         struct net_device_context *ndev_ctx =
1683                 container_of(w, struct net_device_context, dwork.work);
1684         struct hv_device *device_obj = ndev_ctx->device_ctx;
1685         struct net_device *net = hv_get_drvdata(device_obj);
1686         struct netvsc_device *net_device;
1687         struct rndis_device *rdev;
1688         struct netvsc_reconfig *event = NULL;
1689         bool notify = false, reschedule = false;
1690         unsigned long flags, next_reconfig, delay;
1691
1692         /* if changes are happening, comeback later */
1693         if (!rtnl_trylock()) {
1694                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1695                 return;
1696         }
1697
1698         net_device = rtnl_dereference(ndev_ctx->nvdev);
1699         if (!net_device)
1700                 goto out_unlock;
1701
1702         rdev = net_device->extension;
1703
1704         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1705         if (time_is_after_jiffies(next_reconfig)) {
1706                 /* link_watch only sends one notification with current state
1707                  * per second, avoid doing reconfig more frequently. Handle
1708                  * wrap around.
1709                  */
1710                 delay = next_reconfig - jiffies;
1711                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1712                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1713                 goto out_unlock;
1714         }
1715         ndev_ctx->last_reconfig = jiffies;
1716
1717         spin_lock_irqsave(&ndev_ctx->lock, flags);
1718         if (!list_empty(&ndev_ctx->reconfig_events)) {
1719                 event = list_first_entry(&ndev_ctx->reconfig_events,
1720                                          struct netvsc_reconfig, list);
1721                 list_del(&event->list);
1722                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1723         }
1724         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1725
1726         if (!event)
1727                 goto out_unlock;
1728
1729         switch (event->event) {
1730                 /* Only the following events are possible due to the check in
1731                  * netvsc_linkstatus_callback()
1732                  */
1733         case RNDIS_STATUS_MEDIA_CONNECT:
1734                 if (rdev->link_state) {
1735                         rdev->link_state = false;
1736                         netif_carrier_on(net);
1737                         netif_tx_wake_all_queues(net);
1738                 } else {
1739                         notify = true;
1740                 }
1741                 kfree(event);
1742                 break;
1743         case RNDIS_STATUS_MEDIA_DISCONNECT:
1744                 if (!rdev->link_state) {
1745                         rdev->link_state = true;
1746                         netif_carrier_off(net);
1747                         netif_tx_stop_all_queues(net);
1748                 }
1749                 kfree(event);
1750                 break;
1751         case RNDIS_STATUS_NETWORK_CHANGE:
1752                 /* Only makes sense if carrier is present */
1753                 if (!rdev->link_state) {
1754                         rdev->link_state = true;
1755                         netif_carrier_off(net);
1756                         netif_tx_stop_all_queues(net);
1757                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1758                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1759                         list_add(&event->list, &ndev_ctx->reconfig_events);
1760                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1761                         reschedule = true;
1762                 }
1763                 break;
1764         }
1765
1766         rtnl_unlock();
1767
1768         if (notify)
1769                 netdev_notify_peers(net);
1770
1771         /* link_watch only sends one notification with current state per
1772          * second, handle next reconfig event in 2 seconds.
1773          */
1774         if (reschedule)
1775                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1776
1777         return;
1778
1779 out_unlock:
1780         rtnl_unlock();
1781 }
1782
1783 /* Called when VF is injecting data into network stack.
1784  * Change the associated network device from VF to netvsc.
1785  * note: already called with rcu_read_lock
1786  */
1787 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1788 {
1789         struct sk_buff *skb = *pskb;
1790         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1791         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1792         struct netvsc_vf_pcpu_stats *pcpu_stats
1793                  = this_cpu_ptr(ndev_ctx->vf_stats);
1794
1795         skb->dev = ndev;
1796
1797         u64_stats_update_begin(&pcpu_stats->syncp);
1798         pcpu_stats->rx_packets++;
1799         pcpu_stats->rx_bytes += skb->len;
1800         u64_stats_update_end(&pcpu_stats->syncp);
1801
1802         return RX_HANDLER_ANOTHER;
1803 }
1804
1805 static void __netvsc_vf_setup(struct net_device *ndev,
1806                               struct net_device *vf_netdev)
1807 {
1808         int ret;
1809
1810         /* Align MTU of VF with master */
1811         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1812         if (ret)
1813                 netdev_warn(vf_netdev,
1814                             "unable to change mtu to %u\n", ndev->mtu);
1815
1816         /* set multicast etc flags on VF */
1817         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
1818
1819         /* sync address list from ndev to VF */
1820         netif_addr_lock_bh(ndev);
1821         dev_uc_sync(vf_netdev, ndev);
1822         dev_mc_sync(vf_netdev, ndev);
1823         netif_addr_unlock_bh(ndev);
1824
1825         if (netif_running(ndev)) {
1826                 ret = dev_open(vf_netdev);
1827                 if (ret)
1828                         netdev_warn(vf_netdev,
1829                                     "unable to open: %d\n", ret);
1830         }
1831 }
1832
1833 /* Setup VF as slave of the synthetic device.
1834  * Runs in workqueue to avoid recursion in netlink callbacks.
1835  */
1836 static void netvsc_vf_setup(struct work_struct *w)
1837 {
1838         struct net_device_context *ndev_ctx
1839                 = container_of(w, struct net_device_context, vf_takeover.work);
1840         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1841         struct net_device *vf_netdev;
1842
1843         if (!rtnl_trylock()) {
1844                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
1845                 return;
1846         }
1847
1848         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1849         if (vf_netdev)
1850                 __netvsc_vf_setup(ndev, vf_netdev);
1851
1852         rtnl_unlock();
1853 }
1854
1855 static int netvsc_pre_register_vf(struct net_device *vf_netdev,
1856                                   struct net_device *ndev)
1857 {
1858         struct net_device_context *net_device_ctx;
1859         struct netvsc_device *netvsc_dev;
1860
1861         net_device_ctx = netdev_priv(ndev);
1862         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1863         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1864                 return -ENODEV;
1865
1866         return 0;
1867 }
1868
1869 static int netvsc_register_vf(struct net_device *vf_netdev,
1870                               struct net_device *ndev)
1871 {
1872         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1873
1874         /* set slave flag before open to prevent IPv6 addrconf */
1875         vf_netdev->flags |= IFF_SLAVE;
1876
1877         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1878
1879         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1880
1881         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1882
1883         dev_hold(vf_netdev);
1884         rcu_assign_pointer(ndev_ctx->vf_netdev, vf_netdev);
1885
1886         return 0;
1887 }
1888
1889 /* VF up/down change detected, schedule to change data path */
1890 static int netvsc_vf_changed(struct net_device *vf_netdev,
1891                              struct net_device *ndev)
1892 {
1893         struct net_device_context *net_device_ctx;
1894         struct netvsc_device *netvsc_dev;
1895         bool vf_is_up = netif_running(vf_netdev);
1896
1897         net_device_ctx = netdev_priv(ndev);
1898         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1899         if (!netvsc_dev)
1900                 return -ENODEV;
1901
1902         netvsc_switch_datapath(ndev, vf_is_up);
1903         netdev_info(ndev, "Data path switched %s VF: %s\n",
1904                     vf_is_up ? "to" : "from", vf_netdev->name);
1905
1906         return 0;
1907 }
1908
1909 static int netvsc_pre_unregister_vf(struct net_device *vf_netdev,
1910                                     struct net_device *ndev)
1911 {
1912         struct net_device_context *net_device_ctx;
1913
1914         net_device_ctx = netdev_priv(ndev);
1915         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1916
1917         return 0;
1918 }
1919
1920 static int netvsc_unregister_vf(struct net_device *vf_netdev,
1921                                 struct net_device *ndev)
1922 {
1923         struct net_device_context *net_device_ctx;
1924
1925         net_device_ctx = netdev_priv(ndev);
1926
1927         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1928
1929         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1930         dev_put(vf_netdev);
1931
1932         return 0;
1933 }
1934
1935 static struct failover_ops netvsc_failover_ops = {
1936         .slave_pre_register     = netvsc_pre_register_vf,
1937         .slave_register         = netvsc_register_vf,
1938         .slave_pre_unregister   = netvsc_pre_unregister_vf,
1939         .slave_unregister       = netvsc_unregister_vf,
1940         .slave_link_change      = netvsc_vf_changed,
1941         .slave_handle_frame     = netvsc_vf_handle_frame,
1942 };
1943
1944 static int netvsc_probe(struct hv_device *dev,
1945                         const struct hv_vmbus_device_id *dev_id)
1946 {
1947         struct net_device *net = NULL;
1948         struct net_device_context *net_device_ctx;
1949         struct netvsc_device_info device_info;
1950         struct netvsc_device *nvdev;
1951         int ret = -ENOMEM;
1952
1953         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1954                                 VRSS_CHANNEL_MAX);
1955         if (!net)
1956                 goto no_net;
1957
1958         netif_carrier_off(net);
1959
1960         netvsc_init_settings(net);
1961
1962         net_device_ctx = netdev_priv(net);
1963         net_device_ctx->device_ctx = dev;
1964         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1965         if (netif_msg_probe(net_device_ctx))
1966                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1967                            net_device_ctx->msg_enable);
1968
1969         hv_set_drvdata(dev, net);
1970
1971         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1972
1973         spin_lock_init(&net_device_ctx->lock);
1974         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1975         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
1976
1977         net_device_ctx->vf_stats
1978                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
1979         if (!net_device_ctx->vf_stats)
1980                 goto no_stats;
1981
1982         net->netdev_ops = &device_ops;
1983         net->ethtool_ops = &ethtool_ops;
1984         SET_NETDEV_DEV(net, &dev->device);
1985
1986         /* We always need headroom for rndis header */
1987         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1988
1989         /* Initialize the number of queues to be 1, we may change it if more
1990          * channels are offered later.
1991          */
1992         netif_set_real_num_tx_queues(net, 1);
1993         netif_set_real_num_rx_queues(net, 1);
1994
1995         /* Notify the netvsc driver of the new device */
1996         memset(&device_info, 0, sizeof(device_info));
1997         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1998         device_info.send_sections = NETVSC_DEFAULT_TX;
1999         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2000         device_info.recv_sections = NETVSC_DEFAULT_RX;
2001         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2002
2003         nvdev = rndis_filter_device_add(dev, &device_info);
2004         if (IS_ERR(nvdev)) {
2005                 ret = PTR_ERR(nvdev);
2006                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2007                 goto rndis_failed;
2008         }
2009
2010         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2011
2012         /* hw_features computed in rndis_netdev_set_hwcaps() */
2013         net->features = net->hw_features |
2014                 NETIF_F_HIGHDMA | NETIF_F_SG |
2015                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2016         net->vlan_features = net->features;
2017
2018         netdev_lockdep_set_classes(net);
2019
2020         /* MTU range: 68 - 1500 or 65521 */
2021         net->min_mtu = NETVSC_MTU_MIN;
2022         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2023                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2024         else
2025                 net->max_mtu = ETH_DATA_LEN;
2026
2027         ret = register_netdev(net);
2028         if (ret != 0) {
2029                 pr_err("Unable to register netdev.\n");
2030                 goto register_failed;
2031         }
2032
2033         net_device_ctx->failover = failover_register(net, &netvsc_failover_ops);
2034         if (IS_ERR(net_device_ctx->failover)) {
2035                 ret = PTR_ERR(net_device_ctx->failover);
2036                 goto err_failover;
2037         }
2038
2039         return ret;
2040
2041 err_failover:
2042         unregister_netdev(net);
2043 register_failed:
2044         rndis_filter_device_remove(dev, nvdev);
2045 rndis_failed:
2046         free_percpu(net_device_ctx->vf_stats);
2047 no_stats:
2048         hv_set_drvdata(dev, NULL);
2049         free_netdev(net);
2050 no_net:
2051         return ret;
2052 }
2053
2054 static int netvsc_remove(struct hv_device *dev)
2055 {
2056         struct net_device_context *ndev_ctx;
2057         struct net_device *vf_netdev, *net;
2058         struct netvsc_device *nvdev;
2059
2060         net = hv_get_drvdata(dev);
2061         if (net == NULL) {
2062                 dev_err(&dev->device, "No net device to remove\n");
2063                 return 0;
2064         }
2065
2066         ndev_ctx = netdev_priv(net);
2067
2068         cancel_delayed_work_sync(&ndev_ctx->dwork);
2069
2070         rcu_read_lock();
2071         nvdev = rcu_dereference(ndev_ctx->nvdev);
2072
2073         if  (nvdev)
2074                 cancel_work_sync(&nvdev->subchan_work);
2075
2076         /*
2077          * Call to the vsc driver to let it know that the device is being
2078          * removed. Also blocks mtu and channel changes.
2079          */
2080         rtnl_lock();
2081         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2082         if (vf_netdev)
2083                 failover_slave_unregister(vf_netdev);
2084
2085         if (nvdev)
2086                 rndis_filter_device_remove(dev, nvdev);
2087
2088         unregister_netdevice(net);
2089
2090         failover_unregister(ndev_ctx->failover);
2091
2092         rtnl_unlock();
2093         rcu_read_unlock();
2094
2095         hv_set_drvdata(dev, NULL);
2096
2097         free_percpu(ndev_ctx->vf_stats);
2098         free_netdev(net);
2099         return 0;
2100 }
2101
2102 static const struct hv_vmbus_device_id id_table[] = {
2103         /* Network guid */
2104         { HV_NIC_GUID, },
2105         { },
2106 };
2107
2108 MODULE_DEVICE_TABLE(vmbus, id_table);
2109
2110 /* The one and only one */
2111 static struct  hv_driver netvsc_drv = {
2112         .name = KBUILD_MODNAME,
2113         .id_table = id_table,
2114         .probe = netvsc_probe,
2115         .remove = netvsc_remove,
2116 };
2117
2118 static void __exit netvsc_drv_exit(void)
2119 {
2120         vmbus_driver_unregister(&netvsc_drv);
2121 }
2122
2123 static int __init netvsc_drv_init(void)
2124 {
2125         int ret;
2126
2127         if (ring_size < RING_SIZE_MIN) {
2128                 ring_size = RING_SIZE_MIN;
2129                 pr_info("Increased ring_size to %u (min allowed)\n",
2130                         ring_size);
2131         }
2132         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2133
2134         ret = vmbus_driver_register(&netvsc_drv);
2135         if (ret)
2136                 return ret;
2137
2138         return 0;
2139 }
2140
2141 MODULE_LICENSE("GPL");
2142 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2143
2144 module_init(netvsc_drv_init);
2145 module_exit(netvsc_drv_exit);