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 struct net_device_context {
44 /* point back to our device context */
45 struct hv_device *device_ctx;
46 struct delayed_work dwork;
47 struct work_struct work;
50 #define RING_SIZE_MIN 64
51 static int ring_size = 128;
52 module_param(ring_size, int, S_IRUGO);
53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
55 static void do_set_multicast(struct work_struct *w)
57 struct net_device_context *ndevctx =
58 container_of(w, struct net_device_context, work);
59 struct netvsc_device *nvdev;
60 struct rndis_device *rdev;
62 nvdev = hv_get_drvdata(ndevctx->device_ctx);
63 if (nvdev == NULL || nvdev->ndev == NULL)
66 rdev = nvdev->extension;
70 if (nvdev->ndev->flags & IFF_PROMISC)
71 rndis_filter_set_packet_filter(rdev,
72 NDIS_PACKET_TYPE_PROMISCUOUS);
74 rndis_filter_set_packet_filter(rdev,
75 NDIS_PACKET_TYPE_BROADCAST |
76 NDIS_PACKET_TYPE_ALL_MULTICAST |
77 NDIS_PACKET_TYPE_DIRECTED);
80 static void netvsc_set_multicast_list(struct net_device *net)
82 struct net_device_context *net_device_ctx = netdev_priv(net);
84 schedule_work(&net_device_ctx->work);
87 static int netvsc_open(struct net_device *net)
89 struct net_device_context *net_device_ctx = netdev_priv(net);
90 struct hv_device *device_obj = net_device_ctx->device_ctx;
91 struct netvsc_device *nvdev;
92 struct rndis_device *rdev;
95 netif_carrier_off(net);
97 /* Open up the device */
98 ret = rndis_filter_open(device_obj);
100 netdev_err(net, "unable to open device (ret %d).\n", ret);
104 netif_tx_start_all_queues(net);
106 nvdev = hv_get_drvdata(device_obj);
107 rdev = nvdev->extension;
108 if (!rdev->link_state)
109 netif_carrier_on(net);
114 static int netvsc_close(struct net_device *net)
116 struct net_device_context *net_device_ctx = netdev_priv(net);
117 struct hv_device *device_obj = net_device_ctx->device_ctx;
120 netif_tx_disable(net);
122 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
123 cancel_work_sync(&net_device_ctx->work);
124 ret = rndis_filter_close(device_obj);
126 netdev_err(net, "unable to close device (ret %d).\n", ret);
131 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
134 struct rndis_packet *rndis_pkt;
135 struct rndis_per_packet_info *ppi;
137 rndis_pkt = &msg->msg.pkt;
138 rndis_pkt->data_offset += ppi_size;
140 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
141 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
143 ppi->size = ppi_size;
144 ppi->type = pkt_type;
145 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
147 rndis_pkt->per_pkt_info_len += ppi_size;
161 /* Toeplitz hash function
162 * data: network byte order
163 * return: host byte order
165 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
174 subk.ka = ntohl(*(u32 *)key);
176 for (i = 0; i < dlen; i++) {
177 subk.kb = key[k_next];
178 k_next = (k_next + 1) % klen;
179 dt = ((u8 *)data)[i];
180 for (j = 0; j < 8; j++) {
191 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
193 struct flow_keys flow;
196 if (!skb_flow_dissect(skb, &flow) ||
197 !(flow.n_proto == htons(ETH_P_IP) ||
198 flow.n_proto == htons(ETH_P_IPV6)))
201 if (flow.ip_proto == IPPROTO_TCP)
206 *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
211 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
212 void *accel_priv, select_queue_fallback_t fallback)
214 struct net_device_context *net_device_ctx = netdev_priv(ndev);
215 struct hv_device *hdev = net_device_ctx->device_ctx;
216 struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
220 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
223 if (netvsc_set_hash(&hash, skb)) {
224 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
225 ndev->real_num_tx_queues;
226 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
232 void netvsc_xmit_completion(void *context)
234 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
235 struct sk_buff *skb = (struct sk_buff *)
236 (unsigned long)packet->send_completion_tid;
238 if (!packet->part_of_skb)
242 dev_kfree_skb_any(skb);
245 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
246 struct hv_page_buffer *pb)
250 /* Deal with compund pages by ignoring unused part
253 page += (offset >> PAGE_SHIFT);
254 offset &= ~PAGE_MASK;
259 bytes = PAGE_SIZE - offset;
262 pb[j].pfn = page_to_pfn(page);
263 pb[j].offset = offset;
269 if (offset == PAGE_SIZE && len) {
279 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
280 struct hv_netvsc_packet *packet)
282 struct hv_page_buffer *pb = packet->page_buf;
284 char *data = skb->data;
285 int frags = skb_shinfo(skb)->nr_frags;
288 /* The packet is laid out thus:
289 * 1. hdr: RNDIS header and PPI
291 * 3. skb fragment data
294 slots_used += fill_pg_buf(virt_to_page(hdr),
296 len, &pb[slots_used]);
298 packet->rmsg_size = len;
299 packet->rmsg_pgcnt = slots_used;
301 slots_used += fill_pg_buf(virt_to_page(data),
302 offset_in_page(data),
303 skb_headlen(skb), &pb[slots_used]);
305 for (i = 0; i < frags; i++) {
306 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
308 slots_used += fill_pg_buf(skb_frag_page(frag),
310 skb_frag_size(frag), &pb[slots_used]);
315 static int count_skb_frag_slots(struct sk_buff *skb)
317 int i, frags = skb_shinfo(skb)->nr_frags;
320 for (i = 0; i < frags; i++) {
321 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
322 unsigned long size = skb_frag_size(frag);
323 unsigned long offset = frag->page_offset;
325 /* Skip unused frames from start of page */
326 offset &= ~PAGE_MASK;
327 pages += PFN_UP(offset + size);
332 static int netvsc_get_slots(struct sk_buff *skb)
334 char *data = skb->data;
335 unsigned int offset = offset_in_page(data);
336 unsigned int len = skb_headlen(skb);
340 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
341 frag_slots = count_skb_frag_slots(skb);
342 return slots + frag_slots;
345 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
347 u32 ret_val = TRANSPORT_INFO_NOT_IP;
349 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
350 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
354 *trans_off = skb_transport_offset(skb);
356 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
357 struct iphdr *iphdr = ip_hdr(skb);
359 if (iphdr->protocol == IPPROTO_TCP)
360 ret_val = TRANSPORT_INFO_IPV4_TCP;
361 else if (iphdr->protocol == IPPROTO_UDP)
362 ret_val = TRANSPORT_INFO_IPV4_UDP;
364 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
365 ret_val = TRANSPORT_INFO_IPV6_TCP;
366 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
367 ret_val = TRANSPORT_INFO_IPV6_UDP;
374 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
376 struct net_device_context *net_device_ctx = netdev_priv(net);
377 struct hv_netvsc_packet *packet = NULL;
379 unsigned int num_data_pgs;
380 struct rndis_message *rndis_msg;
381 struct rndis_packet *rndis_pkt;
385 struct rndis_per_packet_info *ppi;
386 struct ndis_tcp_ip_checksum_info *csum_info;
387 struct ndis_tcp_lso_info *lso_info;
394 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
397 /* We will atmost need two pages to describe the rndis
398 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
399 * of pages in a single packet. If skb is scattered around
400 * more pages we try linearizing it.
404 skb_length = skb->len;
405 head_room = skb_headroom(skb);
406 num_data_pgs = netvsc_get_slots(skb) + 2;
407 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
408 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
409 num_data_pgs, skb->len);
412 } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
413 if (skb_linearize(skb)) {
414 net_alert_ratelimited("failed to linearize skb\n");
422 pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
424 if (head_room < pkt_sz) {
425 packet = kmalloc(pkt_sz, GFP_ATOMIC);
427 /* out of memory, drop packet */
428 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
432 packet->part_of_skb = false;
434 /* Use the headroom for building up the packet */
435 packet = (struct hv_netvsc_packet *)skb->head;
436 packet->part_of_skb = true;
440 packet->xmit_more = skb->xmit_more;
442 packet->vlan_tci = skb->vlan_tci;
443 packet->page_buf = page_buf;
445 packet->q_idx = skb_get_queue_mapping(skb);
447 packet->is_data_pkt = true;
448 packet->total_data_buflen = skb->len;
450 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
451 sizeof(struct hv_netvsc_packet));
453 memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
455 /* Set the completion routine */
456 packet->send_completion = netvsc_xmit_completion;
457 packet->send_completion_ctx = packet;
458 packet->send_completion_tid = (unsigned long)skb;
460 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
462 /* Add the rndis header */
463 rndis_msg = packet->rndis_msg;
464 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
465 rndis_msg->msg_len = packet->total_data_buflen;
466 rndis_pkt = &rndis_msg->msg.pkt;
467 rndis_pkt->data_offset = sizeof(struct rndis_packet);
468 rndis_pkt->data_len = packet->total_data_buflen;
469 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
471 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
473 hash = skb_get_hash_raw(skb);
474 if (hash != 0 && net->real_num_tx_queues > 1) {
475 rndis_msg_size += NDIS_HASH_PPI_SIZE;
476 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
478 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
482 struct ndis_pkt_8021q_info *vlan;
484 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
485 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
487 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
489 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
490 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
494 net_trans_info = get_net_transport_info(skb, &hdr_offset);
495 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
499 * Setup the sendside checksum offload only if this is not a
505 if ((skb->ip_summed == CHECKSUM_NONE) ||
506 (skb->ip_summed == CHECKSUM_UNNECESSARY))
509 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
510 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
511 TCPIP_CHKSUM_PKTINFO);
513 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
516 if (net_trans_info & (INFO_IPV4 << 16))
517 csum_info->transmit.is_ipv4 = 1;
519 csum_info->transmit.is_ipv6 = 1;
521 if (net_trans_info & INFO_TCP) {
522 csum_info->transmit.tcp_checksum = 1;
523 csum_info->transmit.tcp_header_offset = hdr_offset;
524 } else if (net_trans_info & INFO_UDP) {
525 /* UDP checksum offload is not supported on ws2008r2.
526 * Furthermore, on ws2012 and ws2012r2, there are some
527 * issues with udp checksum offload from Linux guests.
528 * (these are host issues).
529 * For now compute the checksum here.
534 ret = skb_cow_head(skb, 0);
539 udp_len = ntohs(uh->len);
541 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
543 udp_len, IPPROTO_UDP,
544 csum_partial(uh, udp_len, 0));
546 uh->check = CSUM_MANGLED_0;
548 csum_info->transmit.udp_checksum = 0;
553 rndis_msg_size += NDIS_LSO_PPI_SIZE;
554 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
555 TCP_LARGESEND_PKTINFO);
557 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
560 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
561 if (net_trans_info & (INFO_IPV4 << 16)) {
562 lso_info->lso_v2_transmit.ip_version =
563 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
564 ip_hdr(skb)->tot_len = 0;
565 ip_hdr(skb)->check = 0;
566 tcp_hdr(skb)->check =
567 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
568 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
570 lso_info->lso_v2_transmit.ip_version =
571 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
572 ipv6_hdr(skb)->payload_len = 0;
573 tcp_hdr(skb)->check =
574 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
575 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
577 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
578 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
581 /* Start filling in the page buffers with the rndis hdr */
582 rndis_msg->msg_len += rndis_msg_size;
583 packet->total_data_buflen = rndis_msg->msg_len;
584 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
587 ret = netvsc_send(net_device_ctx->device_ctx, packet);
591 net->stats.tx_bytes += skb_length;
592 net->stats.tx_packets++;
594 if (packet && !packet->part_of_skb)
596 if (ret != -EAGAIN) {
597 dev_kfree_skb_any(skb);
598 net->stats.tx_dropped++;
602 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
606 * netvsc_linkstatus_callback - Link up/down notification
608 void netvsc_linkstatus_callback(struct hv_device *device_obj,
609 struct rndis_message *resp)
611 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
612 struct net_device *net;
613 struct net_device_context *ndev_ctx;
614 struct netvsc_device *net_device;
615 struct rndis_device *rdev;
617 net_device = hv_get_drvdata(device_obj);
618 rdev = net_device->extension;
620 switch (indicate->status) {
621 case RNDIS_STATUS_MEDIA_CONNECT:
622 rdev->link_state = false;
624 case RNDIS_STATUS_MEDIA_DISCONNECT:
625 rdev->link_state = true;
627 case RNDIS_STATUS_NETWORK_CHANGE:
628 rdev->link_change = true;
634 net = net_device->ndev;
636 if (!net || net->reg_state != NETREG_REGISTERED)
639 ndev_ctx = netdev_priv(net);
640 if (!rdev->link_state) {
641 schedule_delayed_work(&ndev_ctx->dwork, 0);
642 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
644 schedule_delayed_work(&ndev_ctx->dwork, 0);
649 * netvsc_recv_callback - Callback when we receive a packet from the
650 * "wire" on the specified device.
652 int netvsc_recv_callback(struct hv_device *device_obj,
653 struct hv_netvsc_packet *packet,
654 struct ndis_tcp_ip_checksum_info *csum_info)
656 struct net_device *net;
659 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
660 if (!net || net->reg_state != NETREG_REGISTERED) {
661 packet->status = NVSP_STAT_FAIL;
665 /* Allocate a skb - TODO direct I/O to pages? */
666 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
667 if (unlikely(!skb)) {
668 ++net->stats.rx_dropped;
669 packet->status = NVSP_STAT_FAIL;
674 * Copy to skb. This copy is needed here since the memory pointed by
675 * hv_netvsc_packet cannot be deallocated
677 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
678 packet->total_data_buflen);
680 skb->protocol = eth_type_trans(skb, net);
682 /* We only look at the IP checksum here.
683 * Should we be dropping the packet if checksum
684 * failed? How do we deal with other checksums - TCP/UDP?
686 if (csum_info->receive.ip_checksum_succeeded)
687 skb->ip_summed = CHECKSUM_UNNECESSARY;
689 skb->ip_summed = CHECKSUM_NONE;
692 if (packet->vlan_tci & VLAN_TAG_PRESENT)
693 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
696 skb_record_rx_queue(skb, packet->channel->
697 offermsg.offer.sub_channel_index);
699 net->stats.rx_packets++;
700 net->stats.rx_bytes += packet->total_data_buflen;
703 * Pass the skb back up. Network stack will deallocate the skb when it
712 static void netvsc_get_drvinfo(struct net_device *net,
713 struct ethtool_drvinfo *info)
715 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
716 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
719 static void netvsc_get_channels(struct net_device *net,
720 struct ethtool_channels *channel)
722 struct net_device_context *net_device_ctx = netdev_priv(net);
723 struct hv_device *dev = net_device_ctx->device_ctx;
724 struct netvsc_device *nvdev = hv_get_drvdata(dev);
727 channel->max_combined = nvdev->max_chn;
728 channel->combined_count = nvdev->num_chn;
732 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
734 struct net_device_context *ndevctx = netdev_priv(ndev);
735 struct hv_device *hdev = ndevctx->device_ctx;
736 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
737 struct netvsc_device_info device_info;
738 int limit = ETH_DATA_LEN;
740 if (nvdev == NULL || nvdev->destroy)
743 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
744 limit = NETVSC_MTU - ETH_HLEN;
746 /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
747 if (mtu < ETH_DATA_LEN || mtu > limit)
750 nvdev->start_remove = true;
751 cancel_work_sync(&ndevctx->work);
752 netif_tx_disable(ndev);
753 rndis_filter_device_remove(hdev);
757 ndevctx->device_ctx = hdev;
758 hv_set_drvdata(hdev, ndev);
759 device_info.ring_size = ring_size;
760 rndis_filter_device_add(hdev, &device_info);
761 netif_tx_wake_all_queues(ndev);
767 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
769 struct net_device_context *ndevctx = netdev_priv(ndev);
770 struct hv_device *hdev = ndevctx->device_ctx;
771 struct sockaddr *addr = p;
772 char save_adr[ETH_ALEN];
773 unsigned char save_aatype;
776 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
777 save_aatype = ndev->addr_assign_type;
779 err = eth_mac_addr(ndev, p);
783 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
785 /* roll back to saved MAC */
786 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
787 ndev->addr_assign_type = save_aatype;
793 #ifdef CONFIG_NET_POLL_CONTROLLER
794 static void netvsc_poll_controller(struct net_device *net)
796 /* As netvsc_start_xmit() works synchronous we don't have to
797 * trigger anything here.
802 static const struct ethtool_ops ethtool_ops = {
803 .get_drvinfo = netvsc_get_drvinfo,
804 .get_link = ethtool_op_get_link,
805 .get_channels = netvsc_get_channels,
808 static const struct net_device_ops device_ops = {
809 .ndo_open = netvsc_open,
810 .ndo_stop = netvsc_close,
811 .ndo_start_xmit = netvsc_start_xmit,
812 .ndo_set_rx_mode = netvsc_set_multicast_list,
813 .ndo_change_mtu = netvsc_change_mtu,
814 .ndo_validate_addr = eth_validate_addr,
815 .ndo_set_mac_address = netvsc_set_mac_addr,
816 .ndo_select_queue = netvsc_select_queue,
817 #ifdef CONFIG_NET_POLL_CONTROLLER
818 .ndo_poll_controller = netvsc_poll_controller,
823 * Send GARP packet to network peers after migrations.
824 * After Quick Migration, the network is not immediately operational in the
825 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
826 * another netif_notify_peers() into a delayed work, otherwise GARP packet
827 * will not be sent after quick migration, and cause network disconnection.
828 * Also, we update the carrier status here.
830 static void netvsc_link_change(struct work_struct *w)
832 struct net_device_context *ndev_ctx;
833 struct net_device *net;
834 struct netvsc_device *net_device;
835 struct rndis_device *rdev;
836 bool notify, refresh = false;
837 char *argv[] = { "/etc/init.d/network", "restart", NULL };
838 char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
842 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
843 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
844 rdev = net_device->extension;
845 net = net_device->ndev;
847 if (rdev->link_state) {
848 netif_carrier_off(net);
851 netif_carrier_on(net);
853 if (rdev->link_change) {
854 rdev->link_change = false;
862 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
865 netdev_notify_peers(net);
869 static int netvsc_probe(struct hv_device *dev,
870 const struct hv_vmbus_device_id *dev_id)
872 struct net_device *net = NULL;
873 struct net_device_context *net_device_ctx;
874 struct netvsc_device_info device_info;
875 struct netvsc_device *nvdev;
877 u32 max_needed_headroom;
879 net = alloc_etherdev_mq(sizeof(struct net_device_context),
884 max_needed_headroom = sizeof(struct hv_netvsc_packet) +
887 netif_carrier_off(net);
889 net_device_ctx = netdev_priv(net);
890 net_device_ctx->device_ctx = dev;
891 hv_set_drvdata(dev, net);
892 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
893 INIT_WORK(&net_device_ctx->work, do_set_multicast);
895 net->netdev_ops = &device_ops;
897 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
899 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
900 NETIF_F_IP_CSUM | NETIF_F_TSO;
902 net->ethtool_ops = ðtool_ops;
903 SET_NETDEV_DEV(net, &dev->device);
906 * Request additional head room in the skb.
907 * We will use this space to build the rndis
908 * heaser and other state we need to maintain.
910 net->needed_headroom = max_needed_headroom;
912 /* Notify the netvsc driver of the new device */
913 device_info.ring_size = ring_size;
914 ret = rndis_filter_device_add(dev, &device_info);
916 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
918 hv_set_drvdata(dev, NULL);
921 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
923 nvdev = hv_get_drvdata(dev);
924 netif_set_real_num_tx_queues(net, nvdev->num_chn);
925 netif_set_real_num_rx_queues(net, nvdev->num_chn);
927 ret = register_netdev(net);
929 pr_err("Unable to register netdev.\n");
930 rndis_filter_device_remove(dev);
933 schedule_delayed_work(&net_device_ctx->dwork, 0);
939 static int netvsc_remove(struct hv_device *dev)
941 struct net_device *net;
942 struct net_device_context *ndev_ctx;
943 struct netvsc_device *net_device;
945 net_device = hv_get_drvdata(dev);
946 net = net_device->ndev;
949 dev_err(&dev->device, "No net device to remove\n");
953 net_device->start_remove = true;
955 ndev_ctx = netdev_priv(net);
956 cancel_delayed_work_sync(&ndev_ctx->dwork);
957 cancel_work_sync(&ndev_ctx->work);
959 /* Stop outbound asap */
960 netif_tx_disable(net);
962 unregister_netdev(net);
965 * Call to the vsc driver to let it know that the device is being
968 rndis_filter_device_remove(dev);
974 static const struct hv_vmbus_device_id id_table[] = {
980 MODULE_DEVICE_TABLE(vmbus, id_table);
982 /* The one and only one */
983 static struct hv_driver netvsc_drv = {
984 .name = KBUILD_MODNAME,
985 .id_table = id_table,
986 .probe = netvsc_probe,
987 .remove = netvsc_remove,
990 static void __exit netvsc_drv_exit(void)
992 vmbus_driver_unregister(&netvsc_drv);
995 static int __init netvsc_drv_init(void)
997 if (ring_size < RING_SIZE_MIN) {
998 ring_size = RING_SIZE_MIN;
999 pr_info("Increased ring_size to %d (min allowed)\n",
1002 return vmbus_driver_register(&netvsc_drv);
1005 MODULE_LICENSE("GPL");
1006 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1008 module_init(netvsc_drv_init);
1009 module_exit(netvsc_drv_exit);