2 * Copyright (c) 2009, Microsoft Corporation.
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.
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
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/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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>
35 #include <linux/slab.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/netpoll.h>
40 #include <net/route.h>
42 #include <net/pkt_sched.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
46 #include "hyperv_net.h"
48 #define RING_SIZE_MIN 64
49 #define RETRY_US_LO 5000
50 #define RETRY_US_HI 10000
51 #define RETRY_MAX 2000 /* >10 sec */
53 #define LINKCHANGE_INT (2 * HZ)
54 #define VF_TAKEOVER_INT (HZ / 10)
56 static unsigned int ring_size __ro_after_init = 128;
57 module_param(ring_size, uint, 0444);
58 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
59 unsigned int netvsc_ring_bytes __ro_after_init;
61 static 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 |
66 static int debug = -1;
67 module_param(debug, int, 0444);
68 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
70 static LIST_HEAD(netvsc_dev_list);
72 static void netvsc_change_rx_flags(struct net_device *net, int change)
74 struct net_device_context *ndev_ctx = netdev_priv(net);
75 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
81 if (change & IFF_PROMISC) {
82 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
83 dev_set_promiscuity(vf_netdev, inc);
86 if (change & IFF_ALLMULTI) {
87 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
88 dev_set_allmulti(vf_netdev, inc);
92 static void netvsc_set_rx_mode(struct net_device *net)
94 struct net_device_context *ndev_ctx = netdev_priv(net);
95 struct net_device *vf_netdev;
96 struct netvsc_device *nvdev;
99 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
101 dev_uc_sync(vf_netdev, net);
102 dev_mc_sync(vf_netdev, net);
105 nvdev = rcu_dereference(ndev_ctx->nvdev);
107 rndis_filter_update(nvdev);
111 static int netvsc_open(struct net_device *net)
113 struct net_device_context *ndev_ctx = netdev_priv(net);
114 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
115 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
116 struct rndis_device *rdev;
119 netif_carrier_off(net);
121 /* Open up the device */
122 ret = rndis_filter_open(nvdev);
124 netdev_err(net, "unable to open device (ret %d).\n", ret);
128 rdev = nvdev->extension;
129 if (!rdev->link_state) {
130 netif_carrier_on(net);
131 netif_tx_wake_all_queues(net);
135 /* Setting synthetic device up transparently sets
136 * slave as up. If open fails, then slave will be
137 * still be offline (and not used).
139 ret = dev_open(vf_netdev);
142 "unable to open slave: %s: %d\n",
143 vf_netdev->name, ret);
148 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
150 unsigned int retry = 0;
153 /* Ensure pending bytes in ring are read */
157 for (i = 0; i < nvdev->num_chn; i++) {
158 struct vmbus_channel *chn
159 = nvdev->chan_table[i].channel;
164 /* make sure receive not running now */
165 napi_synchronize(&nvdev->chan_table[i].napi);
167 aread = hv_get_bytes_to_read(&chn->inbound);
171 aread = hv_get_bytes_to_read(&chn->outbound);
179 if (++retry > RETRY_MAX)
182 usleep_range(RETRY_US_LO, RETRY_US_HI);
186 static int netvsc_close(struct net_device *net)
188 struct net_device_context *net_device_ctx = netdev_priv(net);
189 struct net_device *vf_netdev
190 = rtnl_dereference(net_device_ctx->vf_netdev);
191 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
194 netif_tx_disable(net);
196 /* No need to close rndis filter if it is removed already */
200 ret = rndis_filter_close(nvdev);
202 netdev_err(net, "unable to close device (ret %d).\n", ret);
206 ret = netvsc_wait_until_empty(nvdev);
208 netdev_err(net, "Ring buffer not empty after closing rndis\n");
211 dev_close(vf_netdev);
216 static inline void *init_ppi_data(struct rndis_message *msg,
217 u32 ppi_size, u32 pkt_type)
219 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
220 struct rndis_per_packet_info *ppi;
222 rndis_pkt->data_offset += ppi_size;
223 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
224 + rndis_pkt->per_pkt_info_len;
226 ppi->size = ppi_size;
227 ppi->type = pkt_type;
228 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
230 rndis_pkt->per_pkt_info_len += ppi_size;
235 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
236 * packets. We can use ethtool to change UDP hash level when necessary.
238 static inline u32 netvsc_get_hash(
240 const struct net_device_context *ndc)
242 struct flow_keys flow;
243 u32 hash, pkt_proto = 0;
244 static u32 hashrnd __read_mostly;
246 net_get_random_once(&hashrnd, sizeof(hashrnd));
248 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
251 switch (flow.basic.ip_proto) {
253 if (flow.basic.n_proto == htons(ETH_P_IP))
254 pkt_proto = HV_TCP4_L4HASH;
255 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
256 pkt_proto = HV_TCP6_L4HASH;
261 if (flow.basic.n_proto == htons(ETH_P_IP))
262 pkt_proto = HV_UDP4_L4HASH;
263 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
264 pkt_proto = HV_UDP6_L4HASH;
269 if (pkt_proto & ndc->l4_hash) {
270 return skb_get_hash(skb);
272 if (flow.basic.n_proto == htons(ETH_P_IP))
273 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
274 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
275 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
279 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
285 static inline int netvsc_get_tx_queue(struct net_device *ndev,
286 struct sk_buff *skb, int old_idx)
288 const struct net_device_context *ndc = netdev_priv(ndev);
289 struct sock *sk = skb->sk;
292 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
293 (VRSS_SEND_TAB_SIZE - 1)];
295 /* If queue index changed record the new value */
296 if (q_idx != old_idx &&
297 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
298 sk_tx_queue_set(sk, q_idx);
304 * Select queue for transmit.
306 * If a valid queue has already been assigned, then use that.
307 * Otherwise compute tx queue based on hash and the send table.
309 * This is basically similar to default (__netdev_pick_tx) with the added step
310 * of using the host send_table when no other queue has been assigned.
312 * TODO support XPS - but get_xps_queue not exported
314 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
316 int q_idx = sk_tx_queue_get(skb->sk);
318 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
319 /* If forwarding a packet, we use the recorded queue when
320 * available for better cache locality.
322 if (skb_rx_queue_recorded(skb))
323 q_idx = skb_get_rx_queue(skb);
325 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
331 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
333 select_queue_fallback_t fallback)
335 struct net_device_context *ndc = netdev_priv(ndev);
336 struct net_device *vf_netdev;
340 vf_netdev = rcu_dereference(ndc->vf_netdev);
342 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
344 if (vf_ops->ndo_select_queue)
345 txq = vf_ops->ndo_select_queue(vf_netdev, skb,
346 accel_priv, fallback);
348 txq = fallback(vf_netdev, skb);
350 /* Record the queue selected by VF so that it can be
351 * used for common case where VF has more queues than
352 * the synthetic device.
354 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
356 txq = netvsc_pick_tx(ndev, skb);
360 while (unlikely(txq >= ndev->real_num_tx_queues))
361 txq -= ndev->real_num_tx_queues;
366 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
367 struct hv_page_buffer *pb)
371 /* Deal with compund pages by ignoring unused part
374 page += (offset >> PAGE_SHIFT);
375 offset &= ~PAGE_MASK;
380 bytes = PAGE_SIZE - offset;
383 pb[j].pfn = page_to_pfn(page);
384 pb[j].offset = offset;
390 if (offset == PAGE_SIZE && len) {
400 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
401 struct hv_netvsc_packet *packet,
402 struct hv_page_buffer *pb)
405 char *data = skb->data;
406 int frags = skb_shinfo(skb)->nr_frags;
409 /* The packet is laid out thus:
410 * 1. hdr: RNDIS header and PPI
412 * 3. skb fragment data
414 slots_used += fill_pg_buf(virt_to_page(hdr),
416 len, &pb[slots_used]);
418 packet->rmsg_size = len;
419 packet->rmsg_pgcnt = slots_used;
421 slots_used += fill_pg_buf(virt_to_page(data),
422 offset_in_page(data),
423 skb_headlen(skb), &pb[slots_used]);
425 for (i = 0; i < frags; i++) {
426 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
428 slots_used += fill_pg_buf(skb_frag_page(frag),
430 skb_frag_size(frag), &pb[slots_used]);
435 static int count_skb_frag_slots(struct sk_buff *skb)
437 int i, frags = skb_shinfo(skb)->nr_frags;
440 for (i = 0; i < frags; i++) {
441 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
442 unsigned long size = skb_frag_size(frag);
443 unsigned long offset = frag->page_offset;
445 /* Skip unused frames from start of page */
446 offset &= ~PAGE_MASK;
447 pages += PFN_UP(offset + size);
452 static int netvsc_get_slots(struct sk_buff *skb)
454 char *data = skb->data;
455 unsigned int offset = offset_in_page(data);
456 unsigned int len = skb_headlen(skb);
460 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
461 frag_slots = count_skb_frag_slots(skb);
462 return slots + frag_slots;
465 static u32 net_checksum_info(struct sk_buff *skb)
467 if (skb->protocol == htons(ETH_P_IP)) {
468 struct iphdr *ip = ip_hdr(skb);
470 if (ip->protocol == IPPROTO_TCP)
471 return TRANSPORT_INFO_IPV4_TCP;
472 else if (ip->protocol == IPPROTO_UDP)
473 return TRANSPORT_INFO_IPV4_UDP;
475 struct ipv6hdr *ip6 = ipv6_hdr(skb);
477 if (ip6->nexthdr == IPPROTO_TCP)
478 return TRANSPORT_INFO_IPV6_TCP;
479 else if (ip6->nexthdr == IPPROTO_UDP)
480 return TRANSPORT_INFO_IPV6_UDP;
483 return TRANSPORT_INFO_NOT_IP;
486 /* Send skb on the slave VF device. */
487 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
490 struct net_device_context *ndev_ctx = netdev_priv(net);
491 unsigned int len = skb->len;
494 skb->dev = vf_netdev;
495 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
497 rc = dev_queue_xmit(skb);
498 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
499 struct netvsc_vf_pcpu_stats *pcpu_stats
500 = this_cpu_ptr(ndev_ctx->vf_stats);
502 u64_stats_update_begin(&pcpu_stats->syncp);
503 pcpu_stats->tx_packets++;
504 pcpu_stats->tx_bytes += len;
505 u64_stats_update_end(&pcpu_stats->syncp);
507 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
513 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
515 struct net_device_context *net_device_ctx = netdev_priv(net);
516 struct hv_netvsc_packet *packet = NULL;
518 unsigned int num_data_pgs;
519 struct rndis_message *rndis_msg;
520 struct net_device *vf_netdev;
523 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
525 /* if VF is present and up then redirect packets
526 * already called with rcu_read_lock_bh
528 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
529 if (vf_netdev && netif_running(vf_netdev) &&
530 !netpoll_tx_running(net))
531 return netvsc_vf_xmit(net, vf_netdev, skb);
533 /* We will atmost need two pages to describe the rndis
534 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
535 * of pages in a single packet. If skb is scattered around
536 * more pages we try linearizing it.
539 num_data_pgs = netvsc_get_slots(skb) + 2;
541 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
542 ++net_device_ctx->eth_stats.tx_scattered;
544 if (skb_linearize(skb))
547 num_data_pgs = netvsc_get_slots(skb) + 2;
548 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
549 ++net_device_ctx->eth_stats.tx_too_big;
555 * Place the rndis header in the skb head room and
556 * the skb->cb will be used for hv_netvsc_packet
559 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
563 /* Use the skb control buffer for building up the packet */
564 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
565 FIELD_SIZEOF(struct sk_buff, cb));
566 packet = (struct hv_netvsc_packet *)skb->cb;
568 packet->q_idx = skb_get_queue_mapping(skb);
570 packet->total_data_buflen = skb->len;
571 packet->total_bytes = skb->len;
572 packet->total_packets = 1;
574 rndis_msg = (struct rndis_message *)skb->head;
576 /* Add the rndis header */
577 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
578 rndis_msg->msg_len = packet->total_data_buflen;
580 rndis_msg->msg.pkt = (struct rndis_packet) {
581 .data_offset = sizeof(struct rndis_packet),
582 .data_len = packet->total_data_buflen,
583 .per_pkt_info_offset = sizeof(struct rndis_packet),
586 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
588 hash = skb_get_hash_raw(skb);
589 if (hash != 0 && net->real_num_tx_queues > 1) {
592 rndis_msg_size += NDIS_HASH_PPI_SIZE;
593 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
598 if (skb_vlan_tag_present(skb)) {
599 struct ndis_pkt_8021q_info *vlan;
601 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
602 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
606 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
607 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
611 if (skb_is_gso(skb)) {
612 struct ndis_tcp_lso_info *lso_info;
614 rndis_msg_size += NDIS_LSO_PPI_SIZE;
615 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
616 TCP_LARGESEND_PKTINFO);
619 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
620 if (skb->protocol == htons(ETH_P_IP)) {
621 lso_info->lso_v2_transmit.ip_version =
622 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
623 ip_hdr(skb)->tot_len = 0;
624 ip_hdr(skb)->check = 0;
625 tcp_hdr(skb)->check =
626 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
627 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
629 lso_info->lso_v2_transmit.ip_version =
630 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
631 ipv6_hdr(skb)->payload_len = 0;
632 tcp_hdr(skb)->check =
633 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
634 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
636 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
637 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
638 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
639 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
640 struct ndis_tcp_ip_checksum_info *csum_info;
642 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
643 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
644 TCPIP_CHKSUM_PKTINFO);
646 csum_info->value = 0;
647 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
649 if (skb->protocol == htons(ETH_P_IP)) {
650 csum_info->transmit.is_ipv4 = 1;
652 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
653 csum_info->transmit.tcp_checksum = 1;
655 csum_info->transmit.udp_checksum = 1;
657 csum_info->transmit.is_ipv6 = 1;
659 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
660 csum_info->transmit.tcp_checksum = 1;
662 csum_info->transmit.udp_checksum = 1;
665 /* Can't do offload of this type of checksum */
666 if (skb_checksum_help(skb))
671 /* Start filling in the page buffers with the rndis hdr */
672 rndis_msg->msg_len += rndis_msg_size;
673 packet->total_data_buflen = rndis_msg->msg_len;
674 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
677 /* timestamp packet in software */
678 skb_tx_timestamp(skb);
680 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
681 if (likely(ret == 0))
684 if (ret == -EAGAIN) {
685 ++net_device_ctx->eth_stats.tx_busy;
686 return NETDEV_TX_BUSY;
690 ++net_device_ctx->eth_stats.tx_no_space;
693 dev_kfree_skb_any(skb);
694 net->stats.tx_dropped++;
699 ++net_device_ctx->eth_stats.tx_no_memory;
704 * netvsc_linkstatus_callback - Link up/down notification
706 void netvsc_linkstatus_callback(struct net_device *net,
707 struct rndis_message *resp)
709 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
710 struct net_device_context *ndev_ctx = netdev_priv(net);
711 struct netvsc_reconfig *event;
714 /* Update the physical link speed when changing to another vSwitch */
715 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
718 speed = *(u32 *)((void *)indicate
719 + indicate->status_buf_offset) / 10000;
720 ndev_ctx->speed = speed;
724 /* Handle these link change statuses below */
725 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
726 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
727 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
730 if (net->reg_state != NETREG_REGISTERED)
733 event = kzalloc(sizeof(*event), GFP_ATOMIC);
736 event->event = indicate->status;
738 spin_lock_irqsave(&ndev_ctx->lock, flags);
739 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
740 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
742 schedule_delayed_work(&ndev_ctx->dwork, 0);
745 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
746 struct napi_struct *napi,
747 const struct ndis_tcp_ip_checksum_info *csum_info,
748 const struct ndis_pkt_8021q_info *vlan,
749 void *data, u32 buflen)
753 skb = napi_alloc_skb(napi, buflen);
758 * Copy to skb. This copy is needed here since the memory pointed by
759 * hv_netvsc_packet cannot be deallocated
761 skb_put_data(skb, data, buflen);
763 skb->protocol = eth_type_trans(skb, net);
765 /* skb is already created with CHECKSUM_NONE */
766 skb_checksum_none_assert(skb);
769 * In Linux, the IP checksum is always checked.
770 * Do L4 checksum offload if enabled and present.
772 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
773 if (csum_info->receive.tcp_checksum_succeeded ||
774 csum_info->receive.udp_checksum_succeeded)
775 skb->ip_summed = CHECKSUM_UNNECESSARY;
779 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
781 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
789 * netvsc_recv_callback - Callback when we receive a packet from the
790 * "wire" on the specified device.
792 int netvsc_recv_callback(struct net_device *net,
793 struct netvsc_device *net_device,
794 struct vmbus_channel *channel,
796 const struct ndis_tcp_ip_checksum_info *csum_info,
797 const struct ndis_pkt_8021q_info *vlan)
799 struct net_device_context *net_device_ctx = netdev_priv(net);
800 u16 q_idx = channel->offermsg.offer.sub_channel_index;
801 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
803 struct netvsc_stats *rx_stats;
805 if (net->reg_state != NETREG_REGISTERED)
806 return NVSP_STAT_FAIL;
808 /* Allocate a skb - TODO direct I/O to pages? */
809 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
810 csum_info, vlan, data, len);
811 if (unlikely(!skb)) {
812 ++net_device_ctx->eth_stats.rx_no_memory;
814 return NVSP_STAT_FAIL;
817 skb_record_rx_queue(skb, q_idx);
820 * Even if injecting the packet, record the statistics
821 * on the synthetic device because modifying the VF device
822 * statistics will not work correctly.
824 rx_stats = &nvchan->rx_stats;
825 u64_stats_update_begin(&rx_stats->syncp);
827 rx_stats->bytes += len;
829 if (skb->pkt_type == PACKET_BROADCAST)
830 ++rx_stats->broadcast;
831 else if (skb->pkt_type == PACKET_MULTICAST)
832 ++rx_stats->multicast;
833 u64_stats_update_end(&rx_stats->syncp);
835 napi_gro_receive(&nvchan->napi, skb);
836 return NVSP_STAT_SUCCESS;
839 static void netvsc_get_drvinfo(struct net_device *net,
840 struct ethtool_drvinfo *info)
842 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
843 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
846 static void netvsc_get_channels(struct net_device *net,
847 struct ethtool_channels *channel)
849 struct net_device_context *net_device_ctx = netdev_priv(net);
850 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
853 channel->max_combined = nvdev->max_chn;
854 channel->combined_count = nvdev->num_chn;
858 static int netvsc_detach(struct net_device *ndev,
859 struct netvsc_device *nvdev)
861 struct net_device_context *ndev_ctx = netdev_priv(ndev);
862 struct hv_device *hdev = ndev_ctx->device_ctx;
865 /* Don't try continuing to try and setup sub channels */
866 if (cancel_work_sync(&nvdev->subchan_work))
869 /* If device was up (receiving) then shutdown */
870 if (netif_running(ndev)) {
871 netif_tx_disable(ndev);
873 ret = rndis_filter_close(nvdev);
876 "unable to close device (ret %d).\n", ret);
880 ret = netvsc_wait_until_empty(nvdev);
883 "Ring buffer not empty after closing rndis\n");
888 netif_device_detach(ndev);
890 rndis_filter_device_remove(hdev, nvdev);
895 static int netvsc_attach(struct net_device *ndev,
896 struct netvsc_device_info *dev_info)
898 struct net_device_context *ndev_ctx = netdev_priv(ndev);
899 struct hv_device *hdev = ndev_ctx->device_ctx;
900 struct netvsc_device *nvdev;
901 struct rndis_device *rdev;
904 nvdev = rndis_filter_device_add(hdev, dev_info);
906 return PTR_ERR(nvdev);
908 /* Note: enable and attach happen when sub-channels setup */
910 netif_carrier_off(ndev);
912 if (netif_running(ndev)) {
913 ret = rndis_filter_open(nvdev);
917 rdev = nvdev->extension;
918 if (!rdev->link_state)
919 netif_carrier_on(ndev);
925 static int netvsc_set_channels(struct net_device *net,
926 struct ethtool_channels *channels)
928 struct net_device_context *net_device_ctx = netdev_priv(net);
929 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
930 unsigned int orig, count = channels->combined_count;
931 struct netvsc_device_info device_info;
934 /* We do not support separate count for rx, tx, or other */
936 channels->rx_count || channels->tx_count || channels->other_count)
939 if (!nvdev || nvdev->destroy)
942 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
945 if (count > nvdev->max_chn)
948 orig = nvdev->num_chn;
950 memset(&device_info, 0, sizeof(device_info));
951 device_info.num_chn = count;
952 device_info.send_sections = nvdev->send_section_cnt;
953 device_info.send_section_size = nvdev->send_section_size;
954 device_info.recv_sections = nvdev->recv_section_cnt;
955 device_info.recv_section_size = nvdev->recv_section_size;
957 ret = netvsc_detach(net, nvdev);
961 ret = netvsc_attach(net, &device_info);
963 device_info.num_chn = orig;
964 if (netvsc_attach(net, &device_info))
965 netdev_err(net, "restoring channel setting failed\n");
972 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
974 struct ethtool_link_ksettings diff1 = *cmd;
975 struct ethtool_link_ksettings diff2 = {};
977 diff1.base.speed = 0;
978 diff1.base.duplex = 0;
979 /* advertising and cmd are usually set */
980 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
982 /* We set port to PORT_OTHER */
983 diff2.base.port = PORT_OTHER;
985 return !memcmp(&diff1, &diff2, sizeof(diff1));
988 static void netvsc_init_settings(struct net_device *dev)
990 struct net_device_context *ndc = netdev_priv(dev);
992 ndc->l4_hash = HV_DEFAULT_L4HASH;
994 ndc->speed = SPEED_UNKNOWN;
995 ndc->duplex = DUPLEX_FULL;
998 static int netvsc_get_link_ksettings(struct net_device *dev,
999 struct ethtool_link_ksettings *cmd)
1001 struct net_device_context *ndc = netdev_priv(dev);
1003 cmd->base.speed = ndc->speed;
1004 cmd->base.duplex = ndc->duplex;
1005 cmd->base.port = PORT_OTHER;
1010 static int netvsc_set_link_ksettings(struct net_device *dev,
1011 const struct ethtool_link_ksettings *cmd)
1013 struct net_device_context *ndc = netdev_priv(dev);
1016 speed = cmd->base.speed;
1017 if (!ethtool_validate_speed(speed) ||
1018 !ethtool_validate_duplex(cmd->base.duplex) ||
1019 !netvsc_validate_ethtool_ss_cmd(cmd))
1023 ndc->duplex = cmd->base.duplex;
1028 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1030 struct net_device_context *ndevctx = netdev_priv(ndev);
1031 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1032 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1033 int orig_mtu = ndev->mtu;
1034 struct netvsc_device_info device_info;
1037 if (!nvdev || nvdev->destroy)
1040 /* Change MTU of underlying VF netdev first. */
1042 ret = dev_set_mtu(vf_netdev, mtu);
1047 memset(&device_info, 0, sizeof(device_info));
1048 device_info.num_chn = nvdev->num_chn;
1049 device_info.send_sections = nvdev->send_section_cnt;
1050 device_info.send_section_size = nvdev->send_section_size;
1051 device_info.recv_sections = nvdev->recv_section_cnt;
1052 device_info.recv_section_size = nvdev->recv_section_size;
1054 ret = netvsc_detach(ndev, nvdev);
1060 ret = netvsc_attach(ndev, &device_info);
1067 /* Attempt rollback to original MTU */
1068 ndev->mtu = orig_mtu;
1070 if (netvsc_attach(ndev, &device_info))
1071 netdev_err(ndev, "restoring mtu failed\n");
1074 dev_set_mtu(vf_netdev, orig_mtu);
1079 static void netvsc_get_vf_stats(struct net_device *net,
1080 struct netvsc_vf_pcpu_stats *tot)
1082 struct net_device_context *ndev_ctx = netdev_priv(net);
1085 memset(tot, 0, sizeof(*tot));
1087 for_each_possible_cpu(i) {
1088 const struct netvsc_vf_pcpu_stats *stats
1089 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1090 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1094 start = u64_stats_fetch_begin_irq(&stats->syncp);
1095 rx_packets = stats->rx_packets;
1096 tx_packets = stats->tx_packets;
1097 rx_bytes = stats->rx_bytes;
1098 tx_bytes = stats->tx_bytes;
1099 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1101 tot->rx_packets += rx_packets;
1102 tot->tx_packets += tx_packets;
1103 tot->rx_bytes += rx_bytes;
1104 tot->tx_bytes += tx_bytes;
1105 tot->tx_dropped += stats->tx_dropped;
1109 static void netvsc_get_stats64(struct net_device *net,
1110 struct rtnl_link_stats64 *t)
1112 struct net_device_context *ndev_ctx = netdev_priv(net);
1113 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1114 struct netvsc_vf_pcpu_stats vf_tot;
1120 netdev_stats_to_stats64(t, &net->stats);
1122 netvsc_get_vf_stats(net, &vf_tot);
1123 t->rx_packets += vf_tot.rx_packets;
1124 t->tx_packets += vf_tot.tx_packets;
1125 t->rx_bytes += vf_tot.rx_bytes;
1126 t->tx_bytes += vf_tot.tx_bytes;
1127 t->tx_dropped += vf_tot.tx_dropped;
1129 for (i = 0; i < nvdev->num_chn; i++) {
1130 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1131 const struct netvsc_stats *stats;
1132 u64 packets, bytes, multicast;
1135 stats = &nvchan->tx_stats;
1137 start = u64_stats_fetch_begin_irq(&stats->syncp);
1138 packets = stats->packets;
1139 bytes = stats->bytes;
1140 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1142 t->tx_bytes += bytes;
1143 t->tx_packets += packets;
1145 stats = &nvchan->rx_stats;
1147 start = u64_stats_fetch_begin_irq(&stats->syncp);
1148 packets = stats->packets;
1149 bytes = stats->bytes;
1150 multicast = stats->multicast + stats->broadcast;
1151 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1153 t->rx_bytes += bytes;
1154 t->rx_packets += packets;
1155 t->multicast += multicast;
1159 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1161 struct net_device_context *ndc = netdev_priv(ndev);
1162 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1163 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1164 struct sockaddr *addr = p;
1167 err = eth_prepare_mac_addr_change(ndev, p);
1175 err = dev_set_mac_address(vf_netdev, addr);
1180 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1182 eth_commit_mac_addr_change(ndev, p);
1183 } else if (vf_netdev) {
1184 /* rollback change on VF */
1185 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1186 dev_set_mac_address(vf_netdev, addr);
1192 static const struct {
1193 char name[ETH_GSTRING_LEN];
1195 } netvsc_stats[] = {
1196 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1197 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1198 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1199 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1200 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1201 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1202 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1203 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1204 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1205 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1207 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1208 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1209 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1210 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1211 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1214 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1215 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1217 /* 4 statistics per queue (rx/tx packets/bytes) */
1218 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1220 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1222 struct net_device_context *ndc = netdev_priv(dev);
1223 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1228 switch (string_set) {
1230 return NETVSC_GLOBAL_STATS_LEN
1231 + NETVSC_VF_STATS_LEN
1232 + NETVSC_QUEUE_STATS_LEN(nvdev);
1238 static void netvsc_get_ethtool_stats(struct net_device *dev,
1239 struct ethtool_stats *stats, u64 *data)
1241 struct net_device_context *ndc = netdev_priv(dev);
1242 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1243 const void *nds = &ndc->eth_stats;
1244 const struct netvsc_stats *qstats;
1245 struct netvsc_vf_pcpu_stats sum;
1253 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1254 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1256 netvsc_get_vf_stats(dev, &sum);
1257 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1258 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1260 for (j = 0; j < nvdev->num_chn; j++) {
1261 qstats = &nvdev->chan_table[j].tx_stats;
1264 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1265 packets = qstats->packets;
1266 bytes = qstats->bytes;
1267 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1268 data[i++] = packets;
1271 qstats = &nvdev->chan_table[j].rx_stats;
1273 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1274 packets = qstats->packets;
1275 bytes = qstats->bytes;
1276 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1277 data[i++] = packets;
1282 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1284 struct net_device_context *ndc = netdev_priv(dev);
1285 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1292 switch (stringset) {
1294 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1295 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1296 p += ETH_GSTRING_LEN;
1299 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1300 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1301 p += ETH_GSTRING_LEN;
1304 for (i = 0; i < nvdev->num_chn; i++) {
1305 sprintf(p, "tx_queue_%u_packets", i);
1306 p += ETH_GSTRING_LEN;
1307 sprintf(p, "tx_queue_%u_bytes", i);
1308 p += ETH_GSTRING_LEN;
1309 sprintf(p, "rx_queue_%u_packets", i);
1310 p += ETH_GSTRING_LEN;
1311 sprintf(p, "rx_queue_%u_bytes", i);
1312 p += ETH_GSTRING_LEN;
1320 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1321 struct ethtool_rxnfc *info)
1323 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1325 info->data = RXH_IP_SRC | RXH_IP_DST;
1327 switch (info->flow_type) {
1329 if (ndc->l4_hash & HV_TCP4_L4HASH)
1330 info->data |= l4_flag;
1335 if (ndc->l4_hash & HV_TCP6_L4HASH)
1336 info->data |= l4_flag;
1341 if (ndc->l4_hash & HV_UDP4_L4HASH)
1342 info->data |= l4_flag;
1347 if (ndc->l4_hash & HV_UDP6_L4HASH)
1348 info->data |= l4_flag;
1364 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1367 struct net_device_context *ndc = netdev_priv(dev);
1368 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1373 switch (info->cmd) {
1374 case ETHTOOL_GRXRINGS:
1375 info->data = nvdev->num_chn;
1379 return netvsc_get_rss_hash_opts(ndc, info);
1384 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1385 struct ethtool_rxnfc *info)
1387 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1388 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1389 switch (info->flow_type) {
1391 ndc->l4_hash |= HV_TCP4_L4HASH;
1395 ndc->l4_hash |= HV_TCP6_L4HASH;
1399 ndc->l4_hash |= HV_UDP4_L4HASH;
1403 ndc->l4_hash |= HV_UDP6_L4HASH;
1413 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1414 switch (info->flow_type) {
1416 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1420 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1424 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1428 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1442 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1444 struct net_device_context *ndc = netdev_priv(ndev);
1446 if (info->cmd == ETHTOOL_SRXFH)
1447 return netvsc_set_rss_hash_opts(ndc, info);
1452 #ifdef CONFIG_NET_POLL_CONTROLLER
1453 static void netvsc_poll_controller(struct net_device *dev)
1455 struct net_device_context *ndc = netdev_priv(dev);
1456 struct netvsc_device *ndev;
1460 ndev = rcu_dereference(ndc->nvdev);
1462 for (i = 0; i < ndev->num_chn; i++) {
1463 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1465 napi_schedule(&nvchan->napi);
1472 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1474 return NETVSC_HASH_KEYLEN;
1477 static u32 netvsc_rss_indir_size(struct net_device *dev)
1482 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1485 struct net_device_context *ndc = netdev_priv(dev);
1486 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1487 struct rndis_device *rndis_dev;
1494 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1496 rndis_dev = ndev->extension;
1498 for (i = 0; i < ITAB_NUM; i++)
1499 indir[i] = rndis_dev->rx_table[i];
1503 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1508 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1509 const u8 *key, const u8 hfunc)
1511 struct net_device_context *ndc = netdev_priv(dev);
1512 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1513 struct rndis_device *rndis_dev;
1519 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1522 rndis_dev = ndev->extension;
1524 for (i = 0; i < ITAB_NUM; i++)
1525 if (indir[i] >= ndev->num_chn)
1528 for (i = 0; i < ITAB_NUM; i++)
1529 rndis_dev->rx_table[i] = indir[i];
1536 key = rndis_dev->rss_key;
1539 return rndis_filter_set_rss_param(rndis_dev, key);
1542 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1543 * It does have pre-allocated receive area which is divided into sections.
1545 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1546 struct ethtool_ringparam *ring)
1550 ring->rx_pending = nvdev->recv_section_cnt;
1551 ring->tx_pending = nvdev->send_section_cnt;
1553 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1554 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1556 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1558 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1559 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1560 / nvdev->send_section_size;
1563 static void netvsc_get_ringparam(struct net_device *ndev,
1564 struct ethtool_ringparam *ring)
1566 struct net_device_context *ndevctx = netdev_priv(ndev);
1567 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1572 __netvsc_get_ringparam(nvdev, ring);
1575 static int netvsc_set_ringparam(struct net_device *ndev,
1576 struct ethtool_ringparam *ring)
1578 struct net_device_context *ndevctx = netdev_priv(ndev);
1579 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1580 struct netvsc_device_info device_info;
1581 struct ethtool_ringparam orig;
1585 if (!nvdev || nvdev->destroy)
1588 memset(&orig, 0, sizeof(orig));
1589 __netvsc_get_ringparam(nvdev, &orig);
1591 new_tx = clamp_t(u32, ring->tx_pending,
1592 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1593 new_rx = clamp_t(u32, ring->rx_pending,
1594 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1596 if (new_tx == orig.tx_pending &&
1597 new_rx == orig.rx_pending)
1598 return 0; /* no change */
1600 memset(&device_info, 0, sizeof(device_info));
1601 device_info.num_chn = nvdev->num_chn;
1602 device_info.send_sections = new_tx;
1603 device_info.send_section_size = nvdev->send_section_size;
1604 device_info.recv_sections = new_rx;
1605 device_info.recv_section_size = nvdev->recv_section_size;
1607 ret = netvsc_detach(ndev, nvdev);
1611 ret = netvsc_attach(ndev, &device_info);
1613 device_info.send_sections = orig.tx_pending;
1614 device_info.recv_sections = orig.rx_pending;
1616 if (netvsc_attach(ndev, &device_info))
1617 netdev_err(ndev, "restoring ringparam failed");
1623 static u32 netvsc_get_msglevel(struct net_device *ndev)
1625 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1627 return ndev_ctx->msg_enable;
1630 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1632 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1634 ndev_ctx->msg_enable = val;
1637 static const struct ethtool_ops ethtool_ops = {
1638 .get_drvinfo = netvsc_get_drvinfo,
1639 .get_msglevel = netvsc_get_msglevel,
1640 .set_msglevel = netvsc_set_msglevel,
1641 .get_link = ethtool_op_get_link,
1642 .get_ethtool_stats = netvsc_get_ethtool_stats,
1643 .get_sset_count = netvsc_get_sset_count,
1644 .get_strings = netvsc_get_strings,
1645 .get_channels = netvsc_get_channels,
1646 .set_channels = netvsc_set_channels,
1647 .get_ts_info = ethtool_op_get_ts_info,
1648 .get_rxnfc = netvsc_get_rxnfc,
1649 .set_rxnfc = netvsc_set_rxnfc,
1650 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1651 .get_rxfh_indir_size = netvsc_rss_indir_size,
1652 .get_rxfh = netvsc_get_rxfh,
1653 .set_rxfh = netvsc_set_rxfh,
1654 .get_link_ksettings = netvsc_get_link_ksettings,
1655 .set_link_ksettings = netvsc_set_link_ksettings,
1656 .get_ringparam = netvsc_get_ringparam,
1657 .set_ringparam = netvsc_set_ringparam,
1660 static const struct net_device_ops device_ops = {
1661 .ndo_open = netvsc_open,
1662 .ndo_stop = netvsc_close,
1663 .ndo_start_xmit = netvsc_start_xmit,
1664 .ndo_change_rx_flags = netvsc_change_rx_flags,
1665 .ndo_set_rx_mode = netvsc_set_rx_mode,
1666 .ndo_change_mtu = netvsc_change_mtu,
1667 .ndo_validate_addr = eth_validate_addr,
1668 .ndo_set_mac_address = netvsc_set_mac_addr,
1669 .ndo_select_queue = netvsc_select_queue,
1670 .ndo_get_stats64 = netvsc_get_stats64,
1671 #ifdef CONFIG_NET_POLL_CONTROLLER
1672 .ndo_poll_controller = netvsc_poll_controller,
1677 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1678 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1679 * present send GARP packet to network peers with netif_notify_peers().
1681 static void netvsc_link_change(struct work_struct *w)
1683 struct net_device_context *ndev_ctx =
1684 container_of(w, struct net_device_context, dwork.work);
1685 struct hv_device *device_obj = ndev_ctx->device_ctx;
1686 struct net_device *net = hv_get_drvdata(device_obj);
1687 struct netvsc_device *net_device;
1688 struct rndis_device *rdev;
1689 struct netvsc_reconfig *event = NULL;
1690 bool notify = false, reschedule = false;
1691 unsigned long flags, next_reconfig, delay;
1693 /* if changes are happening, comeback later */
1694 if (!rtnl_trylock()) {
1695 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1699 net_device = rtnl_dereference(ndev_ctx->nvdev);
1703 rdev = net_device->extension;
1705 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1706 if (time_is_after_jiffies(next_reconfig)) {
1707 /* link_watch only sends one notification with current state
1708 * per second, avoid doing reconfig more frequently. Handle
1711 delay = next_reconfig - jiffies;
1712 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1713 schedule_delayed_work(&ndev_ctx->dwork, delay);
1716 ndev_ctx->last_reconfig = jiffies;
1718 spin_lock_irqsave(&ndev_ctx->lock, flags);
1719 if (!list_empty(&ndev_ctx->reconfig_events)) {
1720 event = list_first_entry(&ndev_ctx->reconfig_events,
1721 struct netvsc_reconfig, list);
1722 list_del(&event->list);
1723 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1725 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1730 switch (event->event) {
1731 /* Only the following events are possible due to the check in
1732 * netvsc_linkstatus_callback()
1734 case RNDIS_STATUS_MEDIA_CONNECT:
1735 if (rdev->link_state) {
1736 rdev->link_state = false;
1737 netif_carrier_on(net);
1738 netif_tx_wake_all_queues(net);
1744 case RNDIS_STATUS_MEDIA_DISCONNECT:
1745 if (!rdev->link_state) {
1746 rdev->link_state = true;
1747 netif_carrier_off(net);
1748 netif_tx_stop_all_queues(net);
1752 case RNDIS_STATUS_NETWORK_CHANGE:
1753 /* Only makes sense if carrier is present */
1754 if (!rdev->link_state) {
1755 rdev->link_state = true;
1756 netif_carrier_off(net);
1757 netif_tx_stop_all_queues(net);
1758 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1759 spin_lock_irqsave(&ndev_ctx->lock, flags);
1760 list_add(&event->list, &ndev_ctx->reconfig_events);
1761 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1770 netdev_notify_peers(net);
1772 /* link_watch only sends one notification with current state per
1773 * second, handle next reconfig event in 2 seconds.
1776 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1784 static struct net_device *get_netvsc_bymac(const u8 *mac)
1786 struct net_device_context *ndev_ctx;
1788 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
1789 struct net_device *dev = hv_get_drvdata(ndev_ctx->device_ctx);
1791 if (ether_addr_equal(mac, dev->perm_addr))
1798 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1800 struct net_device_context *net_device_ctx;
1801 struct net_device *dev;
1803 dev = netdev_master_upper_dev_get(vf_netdev);
1804 if (!dev || dev->netdev_ops != &device_ops)
1805 return NULL; /* not a netvsc device */
1807 net_device_ctx = netdev_priv(dev);
1808 if (!rtnl_dereference(net_device_ctx->nvdev))
1809 return NULL; /* device is removed */
1814 /* Called when VF is injecting data into network stack.
1815 * Change the associated network device from VF to netvsc.
1816 * note: already called with rcu_read_lock
1818 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1820 struct sk_buff *skb = *pskb;
1821 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1822 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1823 struct netvsc_vf_pcpu_stats *pcpu_stats
1824 = this_cpu_ptr(ndev_ctx->vf_stats);
1828 u64_stats_update_begin(&pcpu_stats->syncp);
1829 pcpu_stats->rx_packets++;
1830 pcpu_stats->rx_bytes += skb->len;
1831 u64_stats_update_end(&pcpu_stats->syncp);
1833 return RX_HANDLER_ANOTHER;
1836 static int netvsc_vf_join(struct net_device *vf_netdev,
1837 struct net_device *ndev)
1839 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1842 ret = netdev_rx_handler_register(vf_netdev,
1843 netvsc_vf_handle_frame, ndev);
1845 netdev_err(vf_netdev,
1846 "can not register netvsc VF receive handler (err = %d)\n",
1848 goto rx_handler_failed;
1851 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
1854 netdev_err(vf_netdev,
1855 "can not set master device %s (err = %d)\n",
1857 goto upper_link_failed;
1860 /* set slave flag before open to prevent IPv6 addrconf */
1861 vf_netdev->flags |= IFF_SLAVE;
1863 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1865 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1867 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1871 netdev_rx_handler_unregister(vf_netdev);
1876 static void __netvsc_vf_setup(struct net_device *ndev,
1877 struct net_device *vf_netdev)
1881 /* Align MTU of VF with master */
1882 ret = dev_set_mtu(vf_netdev, ndev->mtu);
1884 netdev_warn(vf_netdev,
1885 "unable to change mtu to %u\n", ndev->mtu);
1887 /* set multicast etc flags on VF */
1888 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
1890 /* sync address list from ndev to VF */
1891 netif_addr_lock_bh(ndev);
1892 dev_uc_sync(vf_netdev, ndev);
1893 dev_mc_sync(vf_netdev, ndev);
1894 netif_addr_unlock_bh(ndev);
1896 if (netif_running(ndev)) {
1897 ret = dev_open(vf_netdev);
1899 netdev_warn(vf_netdev,
1900 "unable to open: %d\n", ret);
1904 /* Setup VF as slave of the synthetic device.
1905 * Runs in workqueue to avoid recursion in netlink callbacks.
1907 static void netvsc_vf_setup(struct work_struct *w)
1909 struct net_device_context *ndev_ctx
1910 = container_of(w, struct net_device_context, vf_takeover.work);
1911 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1912 struct net_device *vf_netdev;
1914 if (!rtnl_trylock()) {
1915 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
1919 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1921 __netvsc_vf_setup(ndev, vf_netdev);
1926 static int netvsc_register_vf(struct net_device *vf_netdev)
1928 struct net_device *ndev;
1929 struct net_device_context *net_device_ctx;
1930 struct netvsc_device *netvsc_dev;
1933 if (vf_netdev->addr_len != ETH_ALEN)
1937 * We will use the MAC address to locate the synthetic interface to
1938 * associate with the VF interface. If we don't find a matching
1939 * synthetic interface, move on.
1941 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1945 net_device_ctx = netdev_priv(ndev);
1946 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1947 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1950 /* if syntihetic interface is a different namespace,
1951 * then move the VF to that namespace; join will be
1952 * done again in that context.
1954 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
1955 ret = dev_change_net_namespace(vf_netdev,
1956 dev_net(ndev), "eth%d");
1958 netdev_err(vf_netdev,
1959 "could not move to same namespace as %s: %d\n",
1962 netdev_info(vf_netdev,
1963 "VF moved to namespace with: %s\n",
1968 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1970 if (netvsc_vf_join(vf_netdev, ndev) != 0)
1973 dev_hold(vf_netdev);
1974 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1978 /* VF up/down change detected, schedule to change data path */
1979 static int netvsc_vf_changed(struct net_device *vf_netdev)
1981 struct net_device_context *net_device_ctx;
1982 struct netvsc_device *netvsc_dev;
1983 struct net_device *ndev;
1984 bool vf_is_up = netif_running(vf_netdev);
1986 ndev = get_netvsc_byref(vf_netdev);
1990 net_device_ctx = netdev_priv(ndev);
1991 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1995 netvsc_switch_datapath(ndev, vf_is_up);
1996 netdev_info(ndev, "Data path switched %s VF: %s\n",
1997 vf_is_up ? "to" : "from", vf_netdev->name);
2002 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2004 struct net_device *ndev;
2005 struct net_device_context *net_device_ctx;
2007 ndev = get_netvsc_byref(vf_netdev);
2011 net_device_ctx = netdev_priv(ndev);
2012 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2014 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2016 netdev_rx_handler_unregister(vf_netdev);
2017 netdev_upper_dev_unlink(vf_netdev, ndev);
2018 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2024 static int netvsc_probe(struct hv_device *dev,
2025 const struct hv_vmbus_device_id *dev_id)
2027 struct net_device *net = NULL;
2028 struct net_device_context *net_device_ctx;
2029 struct netvsc_device_info device_info;
2030 struct netvsc_device *nvdev;
2033 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2038 netif_carrier_off(net);
2040 netvsc_init_settings(net);
2042 net_device_ctx = netdev_priv(net);
2043 net_device_ctx->device_ctx = dev;
2044 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2045 if (netif_msg_probe(net_device_ctx))
2046 netdev_dbg(net, "netvsc msg_enable: %d\n",
2047 net_device_ctx->msg_enable);
2049 hv_set_drvdata(dev, net);
2051 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2053 spin_lock_init(&net_device_ctx->lock);
2054 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2055 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2057 net_device_ctx->vf_stats
2058 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2059 if (!net_device_ctx->vf_stats)
2062 net->netdev_ops = &device_ops;
2063 net->ethtool_ops = ðtool_ops;
2064 SET_NETDEV_DEV(net, &dev->device);
2066 /* We always need headroom for rndis header */
2067 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2069 /* Initialize the number of queues to be 1, we may change it if more
2070 * channels are offered later.
2072 netif_set_real_num_tx_queues(net, 1);
2073 netif_set_real_num_rx_queues(net, 1);
2075 /* Notify the netvsc driver of the new device */
2076 memset(&device_info, 0, sizeof(device_info));
2077 device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2078 device_info.send_sections = NETVSC_DEFAULT_TX;
2079 device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2080 device_info.recv_sections = NETVSC_DEFAULT_RX;
2081 device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2083 nvdev = rndis_filter_device_add(dev, &device_info);
2084 if (IS_ERR(nvdev)) {
2085 ret = PTR_ERR(nvdev);
2086 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2090 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2092 /* hw_features computed in rndis_netdev_set_hwcaps() */
2093 net->features = net->hw_features |
2094 NETIF_F_HIGHDMA | NETIF_F_SG |
2095 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2096 net->vlan_features = net->features;
2098 netdev_lockdep_set_classes(net);
2100 /* MTU range: 68 - 1500 or 65521 */
2101 net->min_mtu = NETVSC_MTU_MIN;
2102 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2103 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2105 net->max_mtu = ETH_DATA_LEN;
2108 ret = register_netdevice(net);
2110 pr_err("Unable to register netdev.\n");
2111 goto register_failed;
2114 list_add(&net_device_ctx->list, &netvsc_dev_list);
2120 rndis_filter_device_remove(dev, nvdev);
2122 free_percpu(net_device_ctx->vf_stats);
2124 hv_set_drvdata(dev, NULL);
2130 static int netvsc_remove(struct hv_device *dev)
2132 struct net_device_context *ndev_ctx;
2133 struct net_device *vf_netdev, *net;
2134 struct netvsc_device *nvdev;
2136 net = hv_get_drvdata(dev);
2138 dev_err(&dev->device, "No net device to remove\n");
2142 ndev_ctx = netdev_priv(net);
2144 cancel_delayed_work_sync(&ndev_ctx->dwork);
2147 nvdev = rcu_dereference(ndev_ctx->nvdev);
2150 cancel_work_sync(&nvdev->subchan_work);
2153 * Call to the vsc driver to let it know that the device is being
2154 * removed. Also blocks mtu and channel changes.
2157 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2159 netvsc_unregister_vf(vf_netdev);
2162 rndis_filter_device_remove(dev, nvdev);
2164 unregister_netdevice(net);
2165 list_del(&ndev_ctx->list);
2170 hv_set_drvdata(dev, NULL);
2172 free_percpu(ndev_ctx->vf_stats);
2177 static const struct hv_vmbus_device_id id_table[] = {
2183 MODULE_DEVICE_TABLE(vmbus, id_table);
2185 /* The one and only one */
2186 static struct hv_driver netvsc_drv = {
2187 .name = KBUILD_MODNAME,
2188 .id_table = id_table,
2189 .probe = netvsc_probe,
2190 .remove = netvsc_remove,
2194 * On Hyper-V, every VF interface is matched with a corresponding
2195 * synthetic interface. The synthetic interface is presented first
2196 * to the guest. When the corresponding VF instance is registered,
2197 * we will take care of switching the data path.
2199 static int netvsc_netdev_event(struct notifier_block *this,
2200 unsigned long event, void *ptr)
2202 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2204 /* Skip our own events */
2205 if (event_dev->netdev_ops == &device_ops)
2208 /* Avoid non-Ethernet type devices */
2209 if (event_dev->type != ARPHRD_ETHER)
2212 /* Avoid Vlan dev with same MAC registering as VF */
2213 if (is_vlan_dev(event_dev))
2216 /* Avoid Bonding master dev with same MAC registering as VF */
2217 if ((event_dev->priv_flags & IFF_BONDING) &&
2218 (event_dev->flags & IFF_MASTER))
2222 case NETDEV_REGISTER:
2223 return netvsc_register_vf(event_dev);
2224 case NETDEV_UNREGISTER:
2225 return netvsc_unregister_vf(event_dev);
2228 return netvsc_vf_changed(event_dev);
2234 static struct notifier_block netvsc_netdev_notifier = {
2235 .notifier_call = netvsc_netdev_event,
2238 static void __exit netvsc_drv_exit(void)
2240 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2241 vmbus_driver_unregister(&netvsc_drv);
2244 static int __init netvsc_drv_init(void)
2248 if (ring_size < RING_SIZE_MIN) {
2249 ring_size = RING_SIZE_MIN;
2250 pr_info("Increased ring_size to %u (min allowed)\n",
2253 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2255 ret = vmbus_driver_register(&netvsc_drv);
2259 register_netdevice_notifier(&netvsc_netdev_notifier);
2263 MODULE_LICENSE("GPL");
2264 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2266 module_init(netvsc_drv_init);
2267 module_exit(netvsc_drv_exit);