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>
37 #include <net/route.h>
39 #include <net/pkt_sched.h>
41 #include "hyperv_net.h"
43 #define RING_SIZE_MIN 64
44 #define LINKCHANGE_INT (2 * HZ)
45 #define NETVSC_HW_FEATURES (NETIF_F_RXCSUM | \
50 static int ring_size = 128;
51 module_param(ring_size, int, S_IRUGO);
52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
54 static int max_num_vrss_chns = 8;
56 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
57 NETIF_MSG_LINK | NETIF_MSG_IFUP |
58 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
61 static int debug = -1;
62 module_param(debug, int, S_IRUGO);
63 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
65 static void do_set_multicast(struct work_struct *w)
67 struct net_device_context *ndevctx =
68 container_of(w, struct net_device_context, work);
69 struct hv_device *device_obj = ndevctx->device_ctx;
70 struct net_device *ndev = hv_get_drvdata(device_obj);
71 struct netvsc_device *nvdev = ndevctx->nvdev;
72 struct rndis_device *rdev;
77 rdev = nvdev->extension;
81 if (ndev->flags & IFF_PROMISC)
82 rndis_filter_set_packet_filter(rdev,
83 NDIS_PACKET_TYPE_PROMISCUOUS);
85 rndis_filter_set_packet_filter(rdev,
86 NDIS_PACKET_TYPE_BROADCAST |
87 NDIS_PACKET_TYPE_ALL_MULTICAST |
88 NDIS_PACKET_TYPE_DIRECTED);
91 static void netvsc_set_multicast_list(struct net_device *net)
93 struct net_device_context *net_device_ctx = netdev_priv(net);
95 schedule_work(&net_device_ctx->work);
98 static int netvsc_open(struct net_device *net)
100 struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
101 struct rndis_device *rdev;
104 netif_carrier_off(net);
106 /* Open up the device */
107 ret = rndis_filter_open(nvdev);
109 netdev_err(net, "unable to open device (ret %d).\n", ret);
113 netif_tx_wake_all_queues(net);
115 rdev = nvdev->extension;
116 if (!rdev->link_state)
117 netif_carrier_on(net);
122 static int netvsc_close(struct net_device *net)
124 struct net_device_context *net_device_ctx = netdev_priv(net);
125 struct netvsc_device *nvdev = net_device_ctx->nvdev;
127 u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
128 struct vmbus_channel *chn;
130 netif_tx_disable(net);
132 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
133 cancel_work_sync(&net_device_ctx->work);
134 ret = rndis_filter_close(nvdev);
136 netdev_err(net, "unable to close device (ret %d).\n", ret);
140 /* Ensure pending bytes in ring are read */
143 for (i = 0; i < nvdev->num_chn; i++) {
144 chn = nvdev->chn_table[i];
148 hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
154 hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
162 if (retry > retry_max || aread == 0)
172 netdev_err(net, "Ring buffer not empty after closing rndis\n");
179 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
182 struct rndis_packet *rndis_pkt;
183 struct rndis_per_packet_info *ppi;
185 rndis_pkt = &msg->msg.pkt;
186 rndis_pkt->data_offset += ppi_size;
188 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
189 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
191 ppi->size = ppi_size;
192 ppi->type = pkt_type;
193 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
195 rndis_pkt->per_pkt_info_len += ppi_size;
200 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
201 void *accel_priv, select_queue_fallback_t fallback)
203 struct net_device_context *net_device_ctx = netdev_priv(ndev);
204 struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
208 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
211 hash = skb_get_hash(skb);
212 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
213 ndev->real_num_tx_queues;
215 if (!nvsc_dev->chn_table[q_idx])
221 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
222 struct hv_page_buffer *pb)
226 /* Deal with compund pages by ignoring unused part
229 page += (offset >> PAGE_SHIFT);
230 offset &= ~PAGE_MASK;
235 bytes = PAGE_SIZE - offset;
238 pb[j].pfn = page_to_pfn(page);
239 pb[j].offset = offset;
245 if (offset == PAGE_SIZE && len) {
255 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
256 struct hv_netvsc_packet *packet,
257 struct hv_page_buffer **page_buf)
259 struct hv_page_buffer *pb = *page_buf;
261 char *data = skb->data;
262 int frags = skb_shinfo(skb)->nr_frags;
265 /* The packet is laid out thus:
266 * 1. hdr: RNDIS header and PPI
268 * 3. skb fragment data
271 slots_used += fill_pg_buf(virt_to_page(hdr),
273 len, &pb[slots_used]);
275 packet->rmsg_size = len;
276 packet->rmsg_pgcnt = slots_used;
278 slots_used += fill_pg_buf(virt_to_page(data),
279 offset_in_page(data),
280 skb_headlen(skb), &pb[slots_used]);
282 for (i = 0; i < frags; i++) {
283 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
285 slots_used += fill_pg_buf(skb_frag_page(frag),
287 skb_frag_size(frag), &pb[slots_used]);
292 static int count_skb_frag_slots(struct sk_buff *skb)
294 int i, frags = skb_shinfo(skb)->nr_frags;
297 for (i = 0; i < frags; i++) {
298 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
299 unsigned long size = skb_frag_size(frag);
300 unsigned long offset = frag->page_offset;
302 /* Skip unused frames from start of page */
303 offset &= ~PAGE_MASK;
304 pages += PFN_UP(offset + size);
309 static int netvsc_get_slots(struct sk_buff *skb)
311 char *data = skb->data;
312 unsigned int offset = offset_in_page(data);
313 unsigned int len = skb_headlen(skb);
317 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
318 frag_slots = count_skb_frag_slots(skb);
319 return slots + frag_slots;
322 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
324 u32 ret_val = TRANSPORT_INFO_NOT_IP;
326 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
327 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
331 *trans_off = skb_transport_offset(skb);
333 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
334 struct iphdr *iphdr = ip_hdr(skb);
336 if (iphdr->protocol == IPPROTO_TCP)
337 ret_val = TRANSPORT_INFO_IPV4_TCP;
338 else if (iphdr->protocol == IPPROTO_UDP)
339 ret_val = TRANSPORT_INFO_IPV4_UDP;
341 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
342 ret_val = TRANSPORT_INFO_IPV6_TCP;
343 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
344 ret_val = TRANSPORT_INFO_IPV6_UDP;
351 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
353 struct net_device_context *net_device_ctx = netdev_priv(net);
354 struct hv_netvsc_packet *packet = NULL;
356 unsigned int num_data_pgs;
357 struct rndis_message *rndis_msg;
358 struct rndis_packet *rndis_pkt;
360 struct rndis_per_packet_info *ppi;
361 struct ndis_tcp_ip_checksum_info *csum_info;
366 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
367 struct hv_page_buffer *pb = page_buf;
368 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
370 /* We will atmost need two pages to describe the rndis
371 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
372 * of pages in a single packet. If skb is scattered around
373 * more pages we try linearizing it.
376 skb_length = skb->len;
377 num_data_pgs = netvsc_get_slots(skb) + 2;
379 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
380 if (skb_linearize(skb)) {
381 net_alert_ratelimited("failed to linearize skb\n");
386 num_data_pgs = netvsc_get_slots(skb) + 2;
387 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
388 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
389 num_data_pgs, skb->len);
396 * Place the rndis header in the skb head room and
397 * the skb->cb will be used for hv_netvsc_packet
400 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
402 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
406 /* Use the skb control buffer for building up the packet */
407 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
408 FIELD_SIZEOF(struct sk_buff, cb));
409 packet = (struct hv_netvsc_packet *)skb->cb;
411 packet->q_idx = skb_get_queue_mapping(skb);
413 packet->total_data_buflen = skb->len;
415 rndis_msg = (struct rndis_message *)skb->head;
417 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
419 /* Add the rndis header */
420 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
421 rndis_msg->msg_len = packet->total_data_buflen;
422 rndis_pkt = &rndis_msg->msg.pkt;
423 rndis_pkt->data_offset = sizeof(struct rndis_packet);
424 rndis_pkt->data_len = packet->total_data_buflen;
425 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
427 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
429 hash = skb_get_hash_raw(skb);
430 if (hash != 0 && net->real_num_tx_queues > 1) {
431 rndis_msg_size += NDIS_HASH_PPI_SIZE;
432 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
434 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
437 if (skb_vlan_tag_present(skb)) {
438 struct ndis_pkt_8021q_info *vlan;
440 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
441 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
443 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
445 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
446 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
450 net_trans_info = get_net_transport_info(skb, &hdr_offset);
451 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
455 * Setup the sendside checksum offload only if this is not a
458 if (skb_is_gso(skb)) {
459 struct ndis_tcp_lso_info *lso_info;
461 rndis_msg_size += NDIS_LSO_PPI_SIZE;
462 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
463 TCP_LARGESEND_PKTINFO);
465 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
468 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
469 if (net_trans_info & (INFO_IPV4 << 16)) {
470 lso_info->lso_v2_transmit.ip_version =
471 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
472 ip_hdr(skb)->tot_len = 0;
473 ip_hdr(skb)->check = 0;
474 tcp_hdr(skb)->check =
475 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
476 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
478 lso_info->lso_v2_transmit.ip_version =
479 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
480 ipv6_hdr(skb)->payload_len = 0;
481 tcp_hdr(skb)->check =
482 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
483 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
485 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
486 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
490 if ((skb->ip_summed == CHECKSUM_NONE) ||
491 (skb->ip_summed == CHECKSUM_UNNECESSARY))
494 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
495 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
496 TCPIP_CHKSUM_PKTINFO);
498 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
501 if (net_trans_info & (INFO_IPV4 << 16))
502 csum_info->transmit.is_ipv4 = 1;
504 csum_info->transmit.is_ipv6 = 1;
506 if (net_trans_info & INFO_TCP) {
507 csum_info->transmit.tcp_checksum = 1;
508 csum_info->transmit.tcp_header_offset = hdr_offset;
509 } else if (net_trans_info & INFO_UDP) {
510 /* UDP checksum offload is not supported on ws2008r2.
511 * Furthermore, on ws2012 and ws2012r2, there are some
512 * issues with udp checksum offload from Linux guests.
513 * (these are host issues).
514 * For now compute the checksum here.
519 ret = skb_cow_head(skb, 0);
524 udp_len = ntohs(uh->len);
526 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
528 udp_len, IPPROTO_UDP,
529 csum_partial(uh, udp_len, 0));
531 uh->check = CSUM_MANGLED_0;
533 csum_info->transmit.udp_checksum = 0;
537 /* Start filling in the page buffers with the rndis hdr */
538 rndis_msg->msg_len += rndis_msg_size;
539 packet->total_data_buflen = rndis_msg->msg_len;
540 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
543 /* timestamp packet in software */
544 skb_tx_timestamp(skb);
545 ret = netvsc_send(net_device_ctx->device_ctx, packet,
546 rndis_msg, &pb, skb);
547 if (likely(ret == 0)) {
548 u64_stats_update_begin(&tx_stats->syncp);
550 tx_stats->bytes += skb_length;
551 u64_stats_update_end(&tx_stats->syncp);
555 return NETDEV_TX_BUSY;
558 dev_kfree_skb_any(skb);
559 net->stats.tx_dropped++;
565 * netvsc_linkstatus_callback - Link up/down notification
567 void netvsc_linkstatus_callback(struct hv_device *device_obj,
568 struct rndis_message *resp)
570 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
571 struct net_device *net;
572 struct net_device_context *ndev_ctx;
573 struct netvsc_reconfig *event;
576 net = hv_get_drvdata(device_obj);
581 ndev_ctx = netdev_priv(net);
583 /* Update the physical link speed when changing to another vSwitch */
584 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
587 speed = *(u32 *)((void *)indicate + indicate->
588 status_buf_offset) / 10000;
589 ndev_ctx->speed = speed;
593 /* Handle these link change statuses below */
594 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
595 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
596 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
599 if (net->reg_state != NETREG_REGISTERED)
602 event = kzalloc(sizeof(*event), GFP_ATOMIC);
605 event->event = indicate->status;
607 spin_lock_irqsave(&ndev_ctx->lock, flags);
608 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
609 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
611 schedule_delayed_work(&ndev_ctx->dwork, 0);
614 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
615 struct hv_netvsc_packet *packet,
616 struct ndis_tcp_ip_checksum_info *csum_info,
617 void *data, u16 vlan_tci)
621 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
626 * Copy to skb. This copy is needed here since the memory pointed by
627 * hv_netvsc_packet cannot be deallocated
629 memcpy(skb_put(skb, packet->total_data_buflen), data,
630 packet->total_data_buflen);
632 skb->protocol = eth_type_trans(skb, net);
634 /* We only look at the IP checksum here.
635 * Should we be dropping the packet if checksum
636 * failed? How do we deal with other checksums - TCP/UDP?
638 if (csum_info->receive.ip_checksum_succeeded)
639 skb->ip_summed = CHECKSUM_UNNECESSARY;
641 skb->ip_summed = CHECKSUM_NONE;
644 if (vlan_tci & VLAN_TAG_PRESENT)
645 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
652 * netvsc_recv_callback - Callback when we receive a packet from the
653 * "wire" on the specified device.
655 int netvsc_recv_callback(struct hv_device *device_obj,
656 struct hv_netvsc_packet *packet,
658 struct ndis_tcp_ip_checksum_info *csum_info,
659 struct vmbus_channel *channel,
662 struct net_device *net = hv_get_drvdata(device_obj);
663 struct net_device_context *net_device_ctx = netdev_priv(net);
665 struct sk_buff *vf_skb;
666 struct netvsc_stats *rx_stats;
667 u32 bytes_recvd = packet->total_data_buflen;
670 if (!net || net->reg_state != NETREG_REGISTERED)
671 return NVSP_STAT_FAIL;
673 if (READ_ONCE(net_device_ctx->vf_inject)) {
674 atomic_inc(&net_device_ctx->vf_use_cnt);
675 if (!READ_ONCE(net_device_ctx->vf_inject)) {
677 * We raced; just move on.
679 atomic_dec(&net_device_ctx->vf_use_cnt);
680 goto vf_injection_done;
684 * Inject this packet into the VF inerface.
685 * On Hyper-V, multicast and brodcast packets
686 * are only delivered on the synthetic interface
687 * (after subjecting these to policy filters on
688 * the host). Deliver these via the VF interface
691 vf_skb = netvsc_alloc_recv_skb(net_device_ctx->vf_netdev,
692 packet, csum_info, *data,
694 if (vf_skb != NULL) {
695 ++net_device_ctx->vf_netdev->stats.rx_packets;
696 net_device_ctx->vf_netdev->stats.rx_bytes +=
698 netif_receive_skb(vf_skb);
700 ++net->stats.rx_dropped;
701 ret = NVSP_STAT_FAIL;
703 atomic_dec(&net_device_ctx->vf_use_cnt);
708 rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
710 /* Allocate a skb - TODO direct I/O to pages? */
711 skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
712 if (unlikely(!skb)) {
713 ++net->stats.rx_dropped;
714 return NVSP_STAT_FAIL;
716 skb_record_rx_queue(skb, channel->
717 offermsg.offer.sub_channel_index);
719 u64_stats_update_begin(&rx_stats->syncp);
721 rx_stats->bytes += packet->total_data_buflen;
722 u64_stats_update_end(&rx_stats->syncp);
725 * Pass the skb back up. Network stack will deallocate the skb when it
734 static void netvsc_get_drvinfo(struct net_device *net,
735 struct ethtool_drvinfo *info)
737 struct net_device_context *net_device_ctx = netdev_priv(net);
738 struct hv_device *dev = net_device_ctx->device_ctx;
740 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
741 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
742 strlcpy(info->bus_info, vmbus_dev_name(dev), sizeof(info->bus_info));
745 static void netvsc_get_channels(struct net_device *net,
746 struct ethtool_channels *channel)
748 struct net_device_context *net_device_ctx = netdev_priv(net);
749 struct netvsc_device *nvdev = net_device_ctx->nvdev;
752 channel->max_combined = nvdev->max_chn;
753 channel->combined_count = nvdev->num_chn;
757 static int netvsc_set_channels(struct net_device *net,
758 struct ethtool_channels *channels)
760 struct net_device_context *net_device_ctx = netdev_priv(net);
761 struct hv_device *dev = net_device_ctx->device_ctx;
762 struct netvsc_device *nvdev = net_device_ctx->nvdev;
763 struct netvsc_device_info device_info;
767 bool recovering = false;
769 if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
772 num_chn = nvdev->num_chn;
773 max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
775 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
776 pr_info("vRSS unsupported before NVSP Version 5\n");
780 /* We do not support rx, tx, or other */
782 channels->rx_count ||
783 channels->tx_count ||
784 channels->other_count ||
785 (channels->combined_count < 1))
788 if (channels->combined_count > max_chn) {
789 pr_info("combined channels too high, using %d\n", max_chn);
790 channels->combined_count = max_chn;
793 ret = netvsc_close(net);
798 net_device_ctx->start_remove = true;
799 rndis_filter_device_remove(dev);
801 nvdev->num_chn = channels->combined_count;
803 memset(&device_info, 0, sizeof(device_info));
804 device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
805 device_info.ring_size = ring_size;
806 device_info.max_num_vrss_chns = max_num_vrss_chns;
808 ret = rndis_filter_device_add(dev, &device_info);
811 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
817 nvdev = net_device_ctx->nvdev;
819 ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
822 netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
828 ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
831 netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
839 net_device_ctx->start_remove = false;
840 /* We may have missed link change notifications */
841 schedule_delayed_work(&net_device_ctx->dwork, 0);
846 /* If the above failed, we attempt to recover through the same
847 * process but with the original number of channels.
849 netdev_err(net, "could not set channels, recovering\n");
851 channels->combined_count = num_chn;
855 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
857 struct ethtool_cmd diff1 = *cmd;
858 struct ethtool_cmd diff2 = {};
860 ethtool_cmd_speed_set(&diff1, 0);
862 /* advertising and cmd are usually set */
863 diff1.advertising = 0;
865 /* We set port to PORT_OTHER */
866 diff2.port = PORT_OTHER;
868 return !memcmp(&diff1, &diff2, sizeof(diff1));
871 static void netvsc_init_settings(struct net_device *dev)
873 struct net_device_context *ndc = netdev_priv(dev);
875 ndc->speed = SPEED_UNKNOWN;
876 ndc->duplex = DUPLEX_UNKNOWN;
879 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
881 struct net_device_context *ndc = netdev_priv(dev);
883 ethtool_cmd_speed_set(cmd, ndc->speed);
884 cmd->duplex = ndc->duplex;
885 cmd->port = PORT_OTHER;
890 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
892 struct net_device_context *ndc = netdev_priv(dev);
895 speed = ethtool_cmd_speed(cmd);
896 if (!ethtool_validate_speed(speed) ||
897 !ethtool_validate_duplex(cmd->duplex) ||
898 !netvsc_validate_ethtool_ss_cmd(cmd))
902 ndc->duplex = cmd->duplex;
907 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
909 struct net_device_context *ndevctx = netdev_priv(ndev);
910 struct netvsc_device *nvdev = ndevctx->nvdev;
911 struct hv_device *hdev = ndevctx->device_ctx;
912 struct netvsc_device_info device_info;
913 int limit = ETH_DATA_LEN;
917 if (ndevctx->start_remove || !nvdev || nvdev->destroy)
920 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
921 limit = NETVSC_MTU - ETH_HLEN;
923 if (mtu < NETVSC_MTU_MIN || mtu > limit)
926 ret = netvsc_close(ndev);
930 num_chn = nvdev->num_chn;
932 ndevctx->start_remove = true;
933 rndis_filter_device_remove(hdev);
937 memset(&device_info, 0, sizeof(device_info));
938 device_info.ring_size = ring_size;
939 device_info.num_chn = num_chn;
940 device_info.max_num_vrss_chns = max_num_vrss_chns;
941 rndis_filter_device_add(hdev, &device_info);
945 ndevctx->start_remove = false;
947 /* We may have missed link change notifications */
948 schedule_delayed_work(&ndevctx->dwork, 0);
953 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
954 struct rtnl_link_stats64 *t)
956 struct net_device_context *ndev_ctx = netdev_priv(net);
959 for_each_possible_cpu(cpu) {
960 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
962 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
964 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
968 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
969 tx_packets = tx_stats->packets;
970 tx_bytes = tx_stats->bytes;
971 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
974 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
975 rx_packets = rx_stats->packets;
976 rx_bytes = rx_stats->bytes;
977 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
979 t->tx_bytes += tx_bytes;
980 t->tx_packets += tx_packets;
981 t->rx_bytes += rx_bytes;
982 t->rx_packets += rx_packets;
985 t->tx_dropped = net->stats.tx_dropped;
986 t->tx_errors = net->stats.tx_dropped;
988 t->rx_dropped = net->stats.rx_dropped;
989 t->rx_errors = net->stats.rx_errors;
994 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
996 struct sockaddr *addr = p;
997 char save_adr[ETH_ALEN];
998 unsigned char save_aatype;
1001 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
1002 save_aatype = ndev->addr_assign_type;
1004 err = eth_mac_addr(ndev, p);
1008 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1010 /* roll back to saved MAC */
1011 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1012 ndev->addr_assign_type = save_aatype;
1018 #ifdef CONFIG_NET_POLL_CONTROLLER
1019 static void netvsc_poll_controller(struct net_device *net)
1021 /* As netvsc_start_xmit() works synchronous we don't have to
1022 * trigger anything here.
1027 static const struct ethtool_ops ethtool_ops = {
1028 .get_drvinfo = netvsc_get_drvinfo,
1029 .get_link = ethtool_op_get_link,
1030 .get_channels = netvsc_get_channels,
1031 .set_channels = netvsc_set_channels,
1032 .get_ts_info = ethtool_op_get_ts_info,
1033 .get_settings = netvsc_get_settings,
1034 .set_settings = netvsc_set_settings,
1037 static const struct net_device_ops device_ops = {
1038 .ndo_open = netvsc_open,
1039 .ndo_stop = netvsc_close,
1040 .ndo_start_xmit = netvsc_start_xmit,
1041 .ndo_set_rx_mode = netvsc_set_multicast_list,
1042 .ndo_change_mtu = netvsc_change_mtu,
1043 .ndo_validate_addr = eth_validate_addr,
1044 .ndo_set_mac_address = netvsc_set_mac_addr,
1045 .ndo_select_queue = netvsc_select_queue,
1046 .ndo_get_stats64 = netvsc_get_stats64,
1047 #ifdef CONFIG_NET_POLL_CONTROLLER
1048 .ndo_poll_controller = netvsc_poll_controller,
1053 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1054 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1055 * present send GARP packet to network peers with netif_notify_peers().
1057 static void netvsc_link_change(struct work_struct *w)
1059 struct net_device_context *ndev_ctx =
1060 container_of(w, struct net_device_context, dwork.work);
1061 struct hv_device *device_obj = ndev_ctx->device_ctx;
1062 struct net_device *net = hv_get_drvdata(device_obj);
1063 struct netvsc_device *net_device;
1064 struct rndis_device *rdev;
1065 struct netvsc_reconfig *event = NULL;
1066 bool notify = false, reschedule = false;
1067 unsigned long flags, next_reconfig, delay;
1070 if (ndev_ctx->start_remove)
1073 net_device = ndev_ctx->nvdev;
1074 rdev = net_device->extension;
1076 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1077 if (time_is_after_jiffies(next_reconfig)) {
1078 /* link_watch only sends one notification with current state
1079 * per second, avoid doing reconfig more frequently. Handle
1082 delay = next_reconfig - jiffies;
1083 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1084 schedule_delayed_work(&ndev_ctx->dwork, delay);
1087 ndev_ctx->last_reconfig = jiffies;
1089 spin_lock_irqsave(&ndev_ctx->lock, flags);
1090 if (!list_empty(&ndev_ctx->reconfig_events)) {
1091 event = list_first_entry(&ndev_ctx->reconfig_events,
1092 struct netvsc_reconfig, list);
1093 list_del(&event->list);
1094 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1096 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1101 switch (event->event) {
1102 /* Only the following events are possible due to the check in
1103 * netvsc_linkstatus_callback()
1105 case RNDIS_STATUS_MEDIA_CONNECT:
1106 if (rdev->link_state) {
1107 rdev->link_state = false;
1108 netif_carrier_on(net);
1109 netif_tx_wake_all_queues(net);
1115 case RNDIS_STATUS_MEDIA_DISCONNECT:
1116 if (!rdev->link_state) {
1117 rdev->link_state = true;
1118 netif_carrier_off(net);
1119 netif_tx_stop_all_queues(net);
1123 case RNDIS_STATUS_NETWORK_CHANGE:
1124 /* Only makes sense if carrier is present */
1125 if (!rdev->link_state) {
1126 rdev->link_state = true;
1127 netif_carrier_off(net);
1128 netif_tx_stop_all_queues(net);
1129 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1130 spin_lock_irqsave(&ndev_ctx->lock, flags);
1131 list_add(&event->list, &ndev_ctx->reconfig_events);
1132 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1141 netdev_notify_peers(net);
1143 /* link_watch only sends one notification with current state per
1144 * second, handle next reconfig event in 2 seconds.
1147 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1155 static void netvsc_free_netdev(struct net_device *netdev)
1157 struct net_device_context *net_device_ctx = netdev_priv(netdev);
1159 free_percpu(net_device_ctx->tx_stats);
1160 free_percpu(net_device_ctx->rx_stats);
1161 free_netdev(netdev);
1164 static struct net_device *get_netvsc_net_device(char *mac)
1166 struct net_device *dev, *found = NULL;
1170 for_each_netdev(&init_net, dev) {
1171 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1172 if (dev->netdev_ops != &device_ops)
1182 static int netvsc_register_vf(struct net_device *vf_netdev)
1184 struct net_device *ndev;
1185 struct net_device_context *net_device_ctx;
1186 struct netvsc_device *netvsc_dev;
1187 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1189 if (eth_ops == NULL || eth_ops == ðtool_ops)
1193 * We will use the MAC address to locate the synthetic interface to
1194 * associate with the VF interface. If we don't find a matching
1195 * synthetic interface, move on.
1197 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1201 net_device_ctx = netdev_priv(ndev);
1202 netvsc_dev = net_device_ctx->nvdev;
1203 if (!netvsc_dev || net_device_ctx->vf_netdev)
1206 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1208 * Take a reference on the module.
1210 try_module_get(THIS_MODULE);
1211 net_device_ctx->vf_netdev = vf_netdev;
1215 static void netvsc_inject_enable(struct net_device_context *net_device_ctx)
1217 net_device_ctx->vf_inject = true;
1220 static void netvsc_inject_disable(struct net_device_context *net_device_ctx)
1222 net_device_ctx->vf_inject = false;
1224 /* Wait for currently active users to drain out. */
1225 while (atomic_read(&net_device_ctx->vf_use_cnt) != 0)
1229 static int netvsc_vf_up(struct net_device *vf_netdev)
1231 struct net_device *ndev;
1232 struct netvsc_device *netvsc_dev;
1233 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1234 struct net_device_context *net_device_ctx;
1236 if (eth_ops == ðtool_ops)
1239 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1243 net_device_ctx = netdev_priv(ndev);
1244 netvsc_dev = net_device_ctx->nvdev;
1246 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1249 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1250 netvsc_inject_enable(net_device_ctx);
1253 * Open the device before switching data path.
1255 rndis_filter_open(netvsc_dev);
1258 * notify the host to switch the data path.
1260 netvsc_switch_datapath(ndev, true);
1261 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1263 netif_carrier_off(ndev);
1265 /* Now notify peers through VF device. */
1266 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1271 static int netvsc_vf_down(struct net_device *vf_netdev)
1273 struct net_device *ndev;
1274 struct netvsc_device *netvsc_dev;
1275 struct net_device_context *net_device_ctx;
1276 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1278 if (eth_ops == ðtool_ops)
1281 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1285 net_device_ctx = netdev_priv(ndev);
1286 netvsc_dev = net_device_ctx->nvdev;
1288 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1291 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1292 netvsc_inject_disable(net_device_ctx);
1293 netvsc_switch_datapath(ndev, false);
1294 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1295 rndis_filter_close(netvsc_dev);
1296 netif_carrier_on(ndev);
1298 /* Now notify peers through netvsc device. */
1299 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1304 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1306 struct net_device *ndev;
1307 struct netvsc_device *netvsc_dev;
1308 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1309 struct net_device_context *net_device_ctx;
1311 if (eth_ops == ðtool_ops)
1314 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1318 net_device_ctx = netdev_priv(ndev);
1319 netvsc_dev = net_device_ctx->nvdev;
1320 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1322 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1323 netvsc_inject_disable(net_device_ctx);
1324 net_device_ctx->vf_netdev = NULL;
1325 module_put(THIS_MODULE);
1329 static int netvsc_probe(struct hv_device *dev,
1330 const struct hv_vmbus_device_id *dev_id)
1332 struct net_device *net = NULL;
1333 struct net_device_context *net_device_ctx;
1334 struct netvsc_device_info device_info;
1335 struct netvsc_device *nvdev;
1338 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1343 netif_carrier_off(net);
1345 netvsc_init_settings(net);
1347 net_device_ctx = netdev_priv(net);
1348 net_device_ctx->device_ctx = dev;
1349 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1350 if (netif_msg_probe(net_device_ctx))
1351 netdev_dbg(net, "netvsc msg_enable: %d\n",
1352 net_device_ctx->msg_enable);
1354 net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1355 if (!net_device_ctx->tx_stats) {
1359 net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1360 if (!net_device_ctx->rx_stats) {
1361 free_percpu(net_device_ctx->tx_stats);
1366 hv_set_drvdata(dev, net);
1368 net_device_ctx->start_remove = false;
1370 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1371 INIT_WORK(&net_device_ctx->work, do_set_multicast);
1373 spin_lock_init(&net_device_ctx->lock);
1374 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1376 atomic_set(&net_device_ctx->vf_use_cnt, 0);
1377 net_device_ctx->vf_netdev = NULL;
1378 net_device_ctx->vf_inject = false;
1380 net->netdev_ops = &device_ops;
1382 net->hw_features = NETVSC_HW_FEATURES;
1383 net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1385 net->ethtool_ops = ðtool_ops;
1386 SET_NETDEV_DEV(net, &dev->device);
1388 /* We always need headroom for rndis header */
1389 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1391 /* Notify the netvsc driver of the new device */
1392 memset(&device_info, 0, sizeof(device_info));
1393 device_info.ring_size = ring_size;
1394 device_info.max_num_vrss_chns = max_num_vrss_chns;
1395 ret = rndis_filter_device_add(dev, &device_info);
1397 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1398 netvsc_free_netdev(net);
1399 hv_set_drvdata(dev, NULL);
1402 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1404 nvdev = net_device_ctx->nvdev;
1405 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1406 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1408 ret = register_netdev(net);
1410 pr_err("Unable to register netdev.\n");
1411 rndis_filter_device_remove(dev);
1412 netvsc_free_netdev(net);
1418 static int netvsc_remove(struct hv_device *dev)
1420 struct net_device *net;
1421 struct net_device_context *ndev_ctx;
1422 struct netvsc_device *net_device;
1424 net = hv_get_drvdata(dev);
1427 dev_err(&dev->device, "No net device to remove\n");
1431 ndev_ctx = netdev_priv(net);
1432 net_device = ndev_ctx->nvdev;
1434 /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1435 * removing the device.
1438 ndev_ctx->start_remove = true;
1441 cancel_delayed_work_sync(&ndev_ctx->dwork);
1442 cancel_work_sync(&ndev_ctx->work);
1444 /* Stop outbound asap */
1445 netif_tx_disable(net);
1447 unregister_netdev(net);
1450 * Call to the vsc driver to let it know that the device is being
1453 rndis_filter_device_remove(dev);
1455 hv_set_drvdata(dev, NULL);
1457 netvsc_free_netdev(net);
1461 static const struct hv_vmbus_device_id id_table[] = {
1467 MODULE_DEVICE_TABLE(vmbus, id_table);
1469 /* The one and only one */
1470 static struct hv_driver netvsc_drv = {
1471 .name = KBUILD_MODNAME,
1472 .id_table = id_table,
1473 .probe = netvsc_probe,
1474 .remove = netvsc_remove,
1478 * On Hyper-V, every VF interface is matched with a corresponding
1479 * synthetic interface. The synthetic interface is presented first
1480 * to the guest. When the corresponding VF instance is registered,
1481 * we will take care of switching the data path.
1483 static int netvsc_netdev_event(struct notifier_block *this,
1484 unsigned long event, void *ptr)
1486 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1488 /* Avoid Vlan dev with same MAC registering as VF */
1489 if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1492 /* Avoid Bonding master dev with same MAC registering as VF */
1493 if (event_dev->priv_flags & IFF_BONDING &&
1494 event_dev->flags & IFF_MASTER)
1498 case NETDEV_REGISTER:
1499 return netvsc_register_vf(event_dev);
1500 case NETDEV_UNREGISTER:
1501 return netvsc_unregister_vf(event_dev);
1503 return netvsc_vf_up(event_dev);
1505 return netvsc_vf_down(event_dev);
1511 static struct notifier_block netvsc_netdev_notifier = {
1512 .notifier_call = netvsc_netdev_event,
1515 static void __exit netvsc_drv_exit(void)
1517 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1518 vmbus_driver_unregister(&netvsc_drv);
1521 static int __init netvsc_drv_init(void)
1525 if (ring_size < RING_SIZE_MIN) {
1526 ring_size = RING_SIZE_MIN;
1527 pr_info("Increased ring_size to %d (min allowed)\n",
1530 ret = vmbus_driver_register(&netvsc_drv);
1535 register_netdevice_notifier(&netvsc_netdev_notifier);
1539 MODULE_LICENSE("GPL");
1540 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1542 module_init(netvsc_drv_init);
1543 module_exit(netvsc_drv_exit);