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"
44 #define RING_SIZE_MIN 64
45 static int ring_size = 128;
46 module_param(ring_size, int, S_IRUGO);
47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
49 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
50 NETIF_MSG_LINK | NETIF_MSG_IFUP |
51 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
54 static int debug = -1;
55 module_param(debug, int, S_IRUGO);
56 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
58 static void do_set_multicast(struct work_struct *w)
60 struct net_device_context *ndevctx =
61 container_of(w, struct net_device_context, work);
62 struct netvsc_device *nvdev;
63 struct rndis_device *rdev;
65 nvdev = hv_get_drvdata(ndevctx->device_ctx);
66 if (nvdev == NULL || nvdev->ndev == NULL)
69 rdev = nvdev->extension;
73 if (nvdev->ndev->flags & IFF_PROMISC)
74 rndis_filter_set_packet_filter(rdev,
75 NDIS_PACKET_TYPE_PROMISCUOUS);
77 rndis_filter_set_packet_filter(rdev,
78 NDIS_PACKET_TYPE_BROADCAST |
79 NDIS_PACKET_TYPE_ALL_MULTICAST |
80 NDIS_PACKET_TYPE_DIRECTED);
83 static void netvsc_set_multicast_list(struct net_device *net)
85 struct net_device_context *net_device_ctx = netdev_priv(net);
87 schedule_work(&net_device_ctx->work);
90 static int netvsc_open(struct net_device *net)
92 struct net_device_context *net_device_ctx = netdev_priv(net);
93 struct hv_device *device_obj = net_device_ctx->device_ctx;
94 struct netvsc_device *nvdev;
95 struct rndis_device *rdev;
98 netif_carrier_off(net);
100 /* Open up the device */
101 ret = rndis_filter_open(device_obj);
103 netdev_err(net, "unable to open device (ret %d).\n", ret);
107 netif_tx_start_all_queues(net);
109 nvdev = hv_get_drvdata(device_obj);
110 rdev = nvdev->extension;
111 if (!rdev->link_state)
112 netif_carrier_on(net);
117 static int netvsc_close(struct net_device *net)
119 struct net_device_context *net_device_ctx = netdev_priv(net);
120 struct hv_device *device_obj = net_device_ctx->device_ctx;
123 netif_tx_disable(net);
125 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
126 cancel_work_sync(&net_device_ctx->work);
127 ret = rndis_filter_close(device_obj);
129 netdev_err(net, "unable to close device (ret %d).\n", ret);
134 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
137 struct rndis_packet *rndis_pkt;
138 struct rndis_per_packet_info *ppi;
140 rndis_pkt = &msg->msg.pkt;
141 rndis_pkt->data_offset += ppi_size;
143 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
144 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
146 ppi->size = ppi_size;
147 ppi->type = pkt_type;
148 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
150 rndis_pkt->per_pkt_info_len += ppi_size;
164 /* Toeplitz hash function
165 * data: network byte order
166 * return: host byte order
168 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
177 subk.ka = ntohl(*(u32 *)key);
179 for (i = 0; i < dlen; i++) {
180 subk.kb = key[k_next];
181 k_next = (k_next + 1) % klen;
182 dt = ((u8 *)data)[i];
183 for (j = 0; j < 8; j++) {
194 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
196 struct flow_keys flow;
199 if (!skb_flow_dissect(skb, &flow) ||
200 !(flow.n_proto == htons(ETH_P_IP) ||
201 flow.n_proto == htons(ETH_P_IPV6)))
204 if (flow.ip_proto == IPPROTO_TCP)
209 *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
214 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
215 void *accel_priv, select_queue_fallback_t fallback)
217 struct net_device_context *net_device_ctx = netdev_priv(ndev);
218 struct hv_device *hdev = net_device_ctx->device_ctx;
219 struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
223 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
226 if (netvsc_set_hash(&hash, skb)) {
227 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
228 ndev->real_num_tx_queues;
229 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
235 void netvsc_xmit_completion(void *context)
237 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
238 struct sk_buff *skb = (struct sk_buff *)
239 (unsigned long)packet->send_completion_tid;
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;
393 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
396 /* We will atmost need two pages to describe the rndis
397 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
398 * of pages in a single packet. If skb is scattered around
399 * more pages we try linearizing it.
403 skb_length = skb->len;
404 num_data_pgs = netvsc_get_slots(skb) + 2;
405 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
406 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
407 num_data_pgs, skb->len);
410 } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
411 if (skb_linearize(skb)) {
412 net_alert_ratelimited("failed to linearize skb\n");
420 pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
422 ret = skb_cow_head(skb, pkt_sz);
424 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
428 /* Use the headroom for building up the packet */
429 packet = (struct hv_netvsc_packet *)skb->head;
432 packet->xmit_more = skb->xmit_more;
434 packet->vlan_tci = skb->vlan_tci;
435 packet->page_buf = page_buf;
437 packet->q_idx = skb_get_queue_mapping(skb);
439 packet->is_data_pkt = true;
440 packet->total_data_buflen = skb->len;
442 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
443 sizeof(struct hv_netvsc_packet));
445 memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
447 /* Set the completion routine */
448 packet->send_completion = netvsc_xmit_completion;
449 packet->send_completion_ctx = packet;
450 packet->send_completion_tid = (unsigned long)skb;
452 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
454 /* Add the rndis header */
455 rndis_msg = packet->rndis_msg;
456 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
457 rndis_msg->msg_len = packet->total_data_buflen;
458 rndis_pkt = &rndis_msg->msg.pkt;
459 rndis_pkt->data_offset = sizeof(struct rndis_packet);
460 rndis_pkt->data_len = packet->total_data_buflen;
461 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
463 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
465 hash = skb_get_hash_raw(skb);
466 if (hash != 0 && net->real_num_tx_queues > 1) {
467 rndis_msg_size += NDIS_HASH_PPI_SIZE;
468 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
470 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
474 struct ndis_pkt_8021q_info *vlan;
476 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
477 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
479 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
481 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
482 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
486 net_trans_info = get_net_transport_info(skb, &hdr_offset);
487 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
491 * Setup the sendside checksum offload only if this is not a
497 if ((skb->ip_summed == CHECKSUM_NONE) ||
498 (skb->ip_summed == CHECKSUM_UNNECESSARY))
501 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
502 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
503 TCPIP_CHKSUM_PKTINFO);
505 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
508 if (net_trans_info & (INFO_IPV4 << 16))
509 csum_info->transmit.is_ipv4 = 1;
511 csum_info->transmit.is_ipv6 = 1;
513 if (net_trans_info & INFO_TCP) {
514 csum_info->transmit.tcp_checksum = 1;
515 csum_info->transmit.tcp_header_offset = hdr_offset;
516 } else if (net_trans_info & INFO_UDP) {
517 /* UDP checksum offload is not supported on ws2008r2.
518 * Furthermore, on ws2012 and ws2012r2, there are some
519 * issues with udp checksum offload from Linux guests.
520 * (these are host issues).
521 * For now compute the checksum here.
526 ret = skb_cow_head(skb, 0);
531 udp_len = ntohs(uh->len);
533 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
535 udp_len, IPPROTO_UDP,
536 csum_partial(uh, udp_len, 0));
538 uh->check = CSUM_MANGLED_0;
540 csum_info->transmit.udp_checksum = 0;
545 rndis_msg_size += NDIS_LSO_PPI_SIZE;
546 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
547 TCP_LARGESEND_PKTINFO);
549 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
552 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
553 if (net_trans_info & (INFO_IPV4 << 16)) {
554 lso_info->lso_v2_transmit.ip_version =
555 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
556 ip_hdr(skb)->tot_len = 0;
557 ip_hdr(skb)->check = 0;
558 tcp_hdr(skb)->check =
559 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
560 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
562 lso_info->lso_v2_transmit.ip_version =
563 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
564 ipv6_hdr(skb)->payload_len = 0;
565 tcp_hdr(skb)->check =
566 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
567 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
569 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
570 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
573 /* Start filling in the page buffers with the rndis hdr */
574 rndis_msg->msg_len += rndis_msg_size;
575 packet->total_data_buflen = rndis_msg->msg_len;
576 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
579 ret = netvsc_send(net_device_ctx->device_ctx, packet);
583 net->stats.tx_bytes += skb_length;
584 net->stats.tx_packets++;
586 if (ret != -EAGAIN) {
587 dev_kfree_skb_any(skb);
588 net->stats.tx_dropped++;
592 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
596 * netvsc_linkstatus_callback - Link up/down notification
598 void netvsc_linkstatus_callback(struct hv_device *device_obj,
599 struct rndis_message *resp)
601 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
602 struct net_device *net;
603 struct net_device_context *ndev_ctx;
604 struct netvsc_device *net_device;
605 struct rndis_device *rdev;
607 net_device = hv_get_drvdata(device_obj);
608 rdev = net_device->extension;
610 switch (indicate->status) {
611 case RNDIS_STATUS_MEDIA_CONNECT:
612 rdev->link_state = false;
614 case RNDIS_STATUS_MEDIA_DISCONNECT:
615 rdev->link_state = true;
617 case RNDIS_STATUS_NETWORK_CHANGE:
618 rdev->link_change = true;
624 net = net_device->ndev;
626 if (!net || net->reg_state != NETREG_REGISTERED)
629 ndev_ctx = netdev_priv(net);
630 if (!rdev->link_state) {
631 schedule_delayed_work(&ndev_ctx->dwork, 0);
632 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
634 schedule_delayed_work(&ndev_ctx->dwork, 0);
639 * netvsc_recv_callback - Callback when we receive a packet from the
640 * "wire" on the specified device.
642 int netvsc_recv_callback(struct hv_device *device_obj,
643 struct hv_netvsc_packet *packet,
644 struct ndis_tcp_ip_checksum_info *csum_info)
646 struct net_device *net;
649 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
650 if (!net || net->reg_state != NETREG_REGISTERED) {
651 packet->status = NVSP_STAT_FAIL;
655 /* Allocate a skb - TODO direct I/O to pages? */
656 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
657 if (unlikely(!skb)) {
658 ++net->stats.rx_dropped;
659 packet->status = NVSP_STAT_FAIL;
664 * Copy to skb. This copy is needed here since the memory pointed by
665 * hv_netvsc_packet cannot be deallocated
667 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
668 packet->total_data_buflen);
670 skb->protocol = eth_type_trans(skb, net);
672 /* We only look at the IP checksum here.
673 * Should we be dropping the packet if checksum
674 * failed? How do we deal with other checksums - TCP/UDP?
676 if (csum_info->receive.ip_checksum_succeeded)
677 skb->ip_summed = CHECKSUM_UNNECESSARY;
679 skb->ip_summed = CHECKSUM_NONE;
682 if (packet->vlan_tci & VLAN_TAG_PRESENT)
683 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
686 skb_record_rx_queue(skb, packet->channel->
687 offermsg.offer.sub_channel_index);
689 net->stats.rx_packets++;
690 net->stats.rx_bytes += packet->total_data_buflen;
693 * Pass the skb back up. Network stack will deallocate the skb when it
702 static void netvsc_get_drvinfo(struct net_device *net,
703 struct ethtool_drvinfo *info)
705 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
706 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
709 static void netvsc_get_channels(struct net_device *net,
710 struct ethtool_channels *channel)
712 struct net_device_context *net_device_ctx = netdev_priv(net);
713 struct hv_device *dev = net_device_ctx->device_ctx;
714 struct netvsc_device *nvdev = hv_get_drvdata(dev);
717 channel->max_combined = nvdev->max_chn;
718 channel->combined_count = nvdev->num_chn;
722 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
724 struct net_device_context *ndevctx = netdev_priv(ndev);
725 struct hv_device *hdev = ndevctx->device_ctx;
726 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
727 struct netvsc_device_info device_info;
728 int limit = ETH_DATA_LEN;
730 if (nvdev == NULL || nvdev->destroy)
733 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
734 limit = NETVSC_MTU - ETH_HLEN;
736 /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
737 if (mtu < ETH_DATA_LEN || mtu > limit)
740 nvdev->start_remove = true;
741 cancel_work_sync(&ndevctx->work);
742 netif_tx_disable(ndev);
743 rndis_filter_device_remove(hdev);
747 ndevctx->device_ctx = hdev;
748 hv_set_drvdata(hdev, ndev);
749 device_info.ring_size = ring_size;
750 rndis_filter_device_add(hdev, &device_info);
751 netif_tx_wake_all_queues(ndev);
757 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
759 struct net_device_context *ndevctx = netdev_priv(ndev);
760 struct hv_device *hdev = ndevctx->device_ctx;
761 struct sockaddr *addr = p;
762 char save_adr[ETH_ALEN];
763 unsigned char save_aatype;
766 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
767 save_aatype = ndev->addr_assign_type;
769 err = eth_mac_addr(ndev, p);
773 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
775 /* roll back to saved MAC */
776 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
777 ndev->addr_assign_type = save_aatype;
783 #ifdef CONFIG_NET_POLL_CONTROLLER
784 static void netvsc_poll_controller(struct net_device *net)
786 /* As netvsc_start_xmit() works synchronous we don't have to
787 * trigger anything here.
792 static const struct ethtool_ops ethtool_ops = {
793 .get_drvinfo = netvsc_get_drvinfo,
794 .get_link = ethtool_op_get_link,
795 .get_channels = netvsc_get_channels,
798 static const struct net_device_ops device_ops = {
799 .ndo_open = netvsc_open,
800 .ndo_stop = netvsc_close,
801 .ndo_start_xmit = netvsc_start_xmit,
802 .ndo_set_rx_mode = netvsc_set_multicast_list,
803 .ndo_change_mtu = netvsc_change_mtu,
804 .ndo_validate_addr = eth_validate_addr,
805 .ndo_set_mac_address = netvsc_set_mac_addr,
806 .ndo_select_queue = netvsc_select_queue,
807 #ifdef CONFIG_NET_POLL_CONTROLLER
808 .ndo_poll_controller = netvsc_poll_controller,
813 * Send GARP packet to network peers after migrations.
814 * After Quick Migration, the network is not immediately operational in the
815 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
816 * another netif_notify_peers() into a delayed work, otherwise GARP packet
817 * will not be sent after quick migration, and cause network disconnection.
818 * Also, we update the carrier status here.
820 static void netvsc_link_change(struct work_struct *w)
822 struct net_device_context *ndev_ctx;
823 struct net_device *net;
824 struct netvsc_device *net_device;
825 struct rndis_device *rdev;
826 bool notify, refresh = false;
827 char *argv[] = { "/etc/init.d/network", "restart", NULL };
828 char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
832 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
833 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
834 rdev = net_device->extension;
835 net = net_device->ndev;
837 if (rdev->link_state) {
838 netif_carrier_off(net);
841 netif_carrier_on(net);
843 if (rdev->link_change) {
844 rdev->link_change = false;
852 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
855 netdev_notify_peers(net);
859 static int netvsc_probe(struct hv_device *dev,
860 const struct hv_vmbus_device_id *dev_id)
862 struct net_device *net = NULL;
863 struct net_device_context *net_device_ctx;
864 struct netvsc_device_info device_info;
865 struct netvsc_device *nvdev;
867 u32 max_needed_headroom;
869 net = alloc_etherdev_mq(sizeof(struct net_device_context),
874 max_needed_headroom = sizeof(struct hv_netvsc_packet) +
877 netif_carrier_off(net);
879 net_device_ctx = netdev_priv(net);
880 net_device_ctx->device_ctx = dev;
881 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
882 if (netif_msg_probe(net_device_ctx))
883 netdev_dbg(net, "netvsc msg_enable: %d\n",
884 net_device_ctx->msg_enable);
886 hv_set_drvdata(dev, net);
887 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
888 INIT_WORK(&net_device_ctx->work, do_set_multicast);
890 net->netdev_ops = &device_ops;
892 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
894 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
895 NETIF_F_IP_CSUM | NETIF_F_TSO;
897 net->ethtool_ops = ðtool_ops;
898 SET_NETDEV_DEV(net, &dev->device);
901 * Request additional head room in the skb.
902 * We will use this space to build the rndis
903 * heaser and other state we need to maintain.
905 net->needed_headroom = max_needed_headroom;
907 /* Notify the netvsc driver of the new device */
908 device_info.ring_size = ring_size;
909 ret = rndis_filter_device_add(dev, &device_info);
911 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
913 hv_set_drvdata(dev, NULL);
916 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
918 nvdev = hv_get_drvdata(dev);
919 netif_set_real_num_tx_queues(net, nvdev->num_chn);
920 netif_set_real_num_rx_queues(net, nvdev->num_chn);
922 ret = register_netdev(net);
924 pr_err("Unable to register netdev.\n");
925 rndis_filter_device_remove(dev);
928 schedule_delayed_work(&net_device_ctx->dwork, 0);
934 static int netvsc_remove(struct hv_device *dev)
936 struct net_device *net;
937 struct net_device_context *ndev_ctx;
938 struct netvsc_device *net_device;
940 net_device = hv_get_drvdata(dev);
941 net = net_device->ndev;
944 dev_err(&dev->device, "No net device to remove\n");
948 net_device->start_remove = true;
950 ndev_ctx = netdev_priv(net);
951 cancel_delayed_work_sync(&ndev_ctx->dwork);
952 cancel_work_sync(&ndev_ctx->work);
954 /* Stop outbound asap */
955 netif_tx_disable(net);
957 unregister_netdev(net);
960 * Call to the vsc driver to let it know that the device is being
963 rndis_filter_device_remove(dev);
969 static const struct hv_vmbus_device_id id_table[] = {
975 MODULE_DEVICE_TABLE(vmbus, id_table);
977 /* The one and only one */
978 static struct hv_driver netvsc_drv = {
979 .name = KBUILD_MODNAME,
980 .id_table = id_table,
981 .probe = netvsc_probe,
982 .remove = netvsc_remove,
985 static void __exit netvsc_drv_exit(void)
987 vmbus_driver_unregister(&netvsc_drv);
990 static int __init netvsc_drv_init(void)
992 if (ring_size < RING_SIZE_MIN) {
993 ring_size = RING_SIZE_MIN;
994 pr_info("Increased ring_size to %d (min allowed)\n",
997 return vmbus_driver_register(&netvsc_drv);
1000 MODULE_LICENSE("GPL");
1001 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1003 module_init(netvsc_drv_init);
1004 module_exit(netvsc_drv_exit);