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 static 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;
237 u32 index = packet->send_buf_index;
241 if (skb && (index == NETVSC_INVALID_INDEX))
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_page_buffer *pb)
283 char *data = skb->data;
284 int frags = skb_shinfo(skb)->nr_frags;
287 /* The packet is laid out thus:
290 * 3. skb fragment data
293 slots_used += fill_pg_buf(virt_to_page(hdr),
295 len, &pb[slots_used]);
297 slots_used += fill_pg_buf(virt_to_page(data),
298 offset_in_page(data),
299 skb_headlen(skb), &pb[slots_used]);
301 for (i = 0; i < frags; i++) {
302 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
304 slots_used += fill_pg_buf(skb_frag_page(frag),
306 skb_frag_size(frag), &pb[slots_used]);
311 static int count_skb_frag_slots(struct sk_buff *skb)
313 int i, frags = skb_shinfo(skb)->nr_frags;
316 for (i = 0; i < frags; i++) {
317 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
318 unsigned long size = skb_frag_size(frag);
319 unsigned long offset = frag->page_offset;
321 /* Skip unused frames from start of page */
322 offset &= ~PAGE_MASK;
323 pages += PFN_UP(offset + size);
328 static int netvsc_get_slots(struct sk_buff *skb)
330 char *data = skb->data;
331 unsigned int offset = offset_in_page(data);
332 unsigned int len = skb_headlen(skb);
336 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
337 frag_slots = count_skb_frag_slots(skb);
338 return slots + frag_slots;
341 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
343 u32 ret_val = TRANSPORT_INFO_NOT_IP;
345 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
346 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
350 *trans_off = skb_transport_offset(skb);
352 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
353 struct iphdr *iphdr = ip_hdr(skb);
355 if (iphdr->protocol == IPPROTO_TCP)
356 ret_val = TRANSPORT_INFO_IPV4_TCP;
357 else if (iphdr->protocol == IPPROTO_UDP)
358 ret_val = TRANSPORT_INFO_IPV4_UDP;
360 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
361 ret_val = TRANSPORT_INFO_IPV6_TCP;
362 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
363 ret_val = TRANSPORT_INFO_IPV6_UDP;
370 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
372 struct net_device_context *net_device_ctx = netdev_priv(net);
373 struct hv_netvsc_packet *packet;
375 unsigned int num_data_pgs;
376 struct rndis_message *rndis_msg;
377 struct rndis_packet *rndis_pkt;
380 struct rndis_per_packet_info *ppi;
381 struct ndis_tcp_ip_checksum_info *csum_info;
382 struct ndis_tcp_lso_info *lso_info;
386 u32 skb_length = skb->len;
389 /* We will atmost need two pages to describe the rndis
390 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
391 * of pages in a single packet.
393 num_data_pgs = netvsc_get_slots(skb) + 2;
394 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
395 netdev_err(net, "Packet too big: %u\n", skb->len);
397 net->stats.tx_dropped++;
401 /* Allocate a netvsc packet based on # of frags. */
402 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
403 (num_data_pgs * sizeof(struct hv_page_buffer)) +
404 sizeof(struct rndis_message) +
405 NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
406 NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE, GFP_ATOMIC);
408 /* out of memory, drop packet */
409 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
412 net->stats.tx_dropped++;
416 packet->vlan_tci = skb->vlan_tci;
418 packet->q_idx = skb_get_queue_mapping(skb);
420 packet->is_data_pkt = true;
421 packet->total_data_buflen = skb->len;
423 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
424 sizeof(struct hv_netvsc_packet) +
425 (num_data_pgs * sizeof(struct hv_page_buffer)));
427 /* Set the completion routine */
428 packet->send_completion = netvsc_xmit_completion;
429 packet->send_completion_ctx = packet;
430 packet->send_completion_tid = (unsigned long)skb;
432 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
434 /* Add the rndis header */
435 rndis_msg = packet->rndis_msg;
436 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
437 rndis_msg->msg_len = packet->total_data_buflen;
438 rndis_pkt = &rndis_msg->msg.pkt;
439 rndis_pkt->data_offset = sizeof(struct rndis_packet);
440 rndis_pkt->data_len = packet->total_data_buflen;
441 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
443 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
445 hash = skb_get_hash_raw(skb);
446 if (hash != 0 && net->real_num_tx_queues > 1) {
447 rndis_msg_size += NDIS_HASH_PPI_SIZE;
448 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
450 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
454 struct ndis_pkt_8021q_info *vlan;
456 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
457 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
459 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
461 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
462 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
466 net_trans_info = get_net_transport_info(skb, &hdr_offset);
467 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
471 * Setup the sendside checksum offload only if this is not a
477 if ((skb->ip_summed == CHECKSUM_NONE) ||
478 (skb->ip_summed == CHECKSUM_UNNECESSARY))
481 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
482 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
483 TCPIP_CHKSUM_PKTINFO);
485 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
488 if (net_trans_info & (INFO_IPV4 << 16))
489 csum_info->transmit.is_ipv4 = 1;
491 csum_info->transmit.is_ipv6 = 1;
493 if (net_trans_info & INFO_TCP) {
494 csum_info->transmit.tcp_checksum = 1;
495 csum_info->transmit.tcp_header_offset = hdr_offset;
496 } else if (net_trans_info & INFO_UDP) {
497 /* UDP checksum offload is not supported on ws2008r2.
498 * Furthermore, on ws2012 and ws2012r2, there are some
499 * issues with udp checksum offload from Linux guests.
500 * (these are host issues).
501 * For now compute the checksum here.
506 ret = skb_cow_head(skb, 0);
511 udp_len = ntohs(uh->len);
513 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
515 udp_len, IPPROTO_UDP,
516 csum_partial(uh, udp_len, 0));
518 uh->check = CSUM_MANGLED_0;
520 csum_info->transmit.udp_checksum = 0;
525 rndis_msg_size += NDIS_LSO_PPI_SIZE;
526 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
527 TCP_LARGESEND_PKTINFO);
529 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
532 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
533 if (net_trans_info & (INFO_IPV4 << 16)) {
534 lso_info->lso_v2_transmit.ip_version =
535 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
536 ip_hdr(skb)->tot_len = 0;
537 ip_hdr(skb)->check = 0;
538 tcp_hdr(skb)->check =
539 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
540 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
542 lso_info->lso_v2_transmit.ip_version =
543 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
544 ipv6_hdr(skb)->payload_len = 0;
545 tcp_hdr(skb)->check =
546 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
547 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
549 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
550 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
553 /* Start filling in the page buffers with the rndis hdr */
554 rndis_msg->msg_len += rndis_msg_size;
555 packet->total_data_buflen = rndis_msg->msg_len;
556 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
557 skb, &packet->page_buf[0]);
559 ret = netvsc_send(net_device_ctx->device_ctx, packet);
563 net->stats.tx_bytes += skb_length;
564 net->stats.tx_packets++;
567 if (ret != -EAGAIN) {
568 dev_kfree_skb_any(skb);
569 net->stats.tx_dropped++;
573 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
577 * netvsc_linkstatus_callback - Link up/down notification
579 void netvsc_linkstatus_callback(struct hv_device *device_obj,
580 struct rndis_message *resp)
582 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
583 struct net_device *net;
584 struct net_device_context *ndev_ctx;
585 struct netvsc_device *net_device;
586 struct rndis_device *rdev;
588 net_device = hv_get_drvdata(device_obj);
589 rdev = net_device->extension;
591 switch (indicate->status) {
592 case RNDIS_STATUS_MEDIA_CONNECT:
593 rdev->link_state = false;
595 case RNDIS_STATUS_MEDIA_DISCONNECT:
596 rdev->link_state = true;
598 case RNDIS_STATUS_NETWORK_CHANGE:
599 rdev->link_change = true;
605 net = net_device->ndev;
607 if (!net || net->reg_state != NETREG_REGISTERED)
610 ndev_ctx = netdev_priv(net);
611 if (!rdev->link_state) {
612 schedule_delayed_work(&ndev_ctx->dwork, 0);
613 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
615 schedule_delayed_work(&ndev_ctx->dwork, 0);
620 * netvsc_recv_callback - Callback when we receive a packet from the
621 * "wire" on the specified device.
623 int netvsc_recv_callback(struct hv_device *device_obj,
624 struct hv_netvsc_packet *packet,
625 struct ndis_tcp_ip_checksum_info *csum_info)
627 struct net_device *net;
630 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
631 if (!net || net->reg_state != NETREG_REGISTERED) {
632 packet->status = NVSP_STAT_FAIL;
636 /* Allocate a skb - TODO direct I/O to pages? */
637 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
638 if (unlikely(!skb)) {
639 ++net->stats.rx_dropped;
640 packet->status = NVSP_STAT_FAIL;
645 * Copy to skb. This copy is needed here since the memory pointed by
646 * hv_netvsc_packet cannot be deallocated
648 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
649 packet->total_data_buflen);
651 skb->protocol = eth_type_trans(skb, net);
653 /* We only look at the IP checksum here.
654 * Should we be dropping the packet if checksum
655 * failed? How do we deal with other checksums - TCP/UDP?
657 if (csum_info->receive.ip_checksum_succeeded)
658 skb->ip_summed = CHECKSUM_UNNECESSARY;
660 skb->ip_summed = CHECKSUM_NONE;
663 if (packet->vlan_tci & VLAN_TAG_PRESENT)
664 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
667 skb_record_rx_queue(skb, packet->channel->
668 offermsg.offer.sub_channel_index);
670 net->stats.rx_packets++;
671 net->stats.rx_bytes += packet->total_data_buflen;
674 * Pass the skb back up. Network stack will deallocate the skb when it
683 static void netvsc_get_drvinfo(struct net_device *net,
684 struct ethtool_drvinfo *info)
686 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
687 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
690 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
692 struct net_device_context *ndevctx = netdev_priv(ndev);
693 struct hv_device *hdev = ndevctx->device_ctx;
694 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
695 struct netvsc_device_info device_info;
696 int limit = ETH_DATA_LEN;
698 if (nvdev == NULL || nvdev->destroy)
701 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
702 limit = NETVSC_MTU - ETH_HLEN;
704 /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
705 if (mtu < ETH_DATA_LEN || mtu > limit)
708 nvdev->start_remove = true;
709 cancel_work_sync(&ndevctx->work);
710 netif_tx_disable(ndev);
711 rndis_filter_device_remove(hdev);
715 ndevctx->device_ctx = hdev;
716 hv_set_drvdata(hdev, ndev);
717 device_info.ring_size = ring_size;
718 rndis_filter_device_add(hdev, &device_info);
719 netif_tx_wake_all_queues(ndev);
725 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
727 struct net_device_context *ndevctx = netdev_priv(ndev);
728 struct hv_device *hdev = ndevctx->device_ctx;
729 struct sockaddr *addr = p;
730 char save_adr[ETH_ALEN];
731 unsigned char save_aatype;
734 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
735 save_aatype = ndev->addr_assign_type;
737 err = eth_mac_addr(ndev, p);
741 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
743 /* roll back to saved MAC */
744 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
745 ndev->addr_assign_type = save_aatype;
751 #ifdef CONFIG_NET_POLL_CONTROLLER
752 static void netvsc_poll_controller(struct net_device *net)
754 /* As netvsc_start_xmit() works synchronous we don't have to
755 * trigger anything here.
760 static const struct ethtool_ops ethtool_ops = {
761 .get_drvinfo = netvsc_get_drvinfo,
762 .get_link = ethtool_op_get_link,
765 static const struct net_device_ops device_ops = {
766 .ndo_open = netvsc_open,
767 .ndo_stop = netvsc_close,
768 .ndo_start_xmit = netvsc_start_xmit,
769 .ndo_set_rx_mode = netvsc_set_multicast_list,
770 .ndo_change_mtu = netvsc_change_mtu,
771 .ndo_validate_addr = eth_validate_addr,
772 .ndo_set_mac_address = netvsc_set_mac_addr,
773 .ndo_select_queue = netvsc_select_queue,
774 #ifdef CONFIG_NET_POLL_CONTROLLER
775 .ndo_poll_controller = netvsc_poll_controller,
780 * Send GARP packet to network peers after migrations.
781 * After Quick Migration, the network is not immediately operational in the
782 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
783 * another netif_notify_peers() into a delayed work, otherwise GARP packet
784 * will not be sent after quick migration, and cause network disconnection.
785 * Also, we update the carrier status here.
787 static void netvsc_link_change(struct work_struct *w)
789 struct net_device_context *ndev_ctx;
790 struct net_device *net;
791 struct netvsc_device *net_device;
792 struct rndis_device *rdev;
793 bool notify, refresh = false;
794 char *argv[] = { "/etc/init.d/network", "restart", NULL };
795 char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
799 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
800 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
801 rdev = net_device->extension;
802 net = net_device->ndev;
804 if (rdev->link_state) {
805 netif_carrier_off(net);
808 netif_carrier_on(net);
810 if (rdev->link_change) {
811 rdev->link_change = false;
819 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
822 netdev_notify_peers(net);
826 static int netvsc_probe(struct hv_device *dev,
827 const struct hv_vmbus_device_id *dev_id)
829 struct net_device *net = NULL;
830 struct net_device_context *net_device_ctx;
831 struct netvsc_device_info device_info;
832 struct netvsc_device *nvdev;
835 net = alloc_etherdev_mq(sizeof(struct net_device_context),
840 netif_carrier_off(net);
842 net_device_ctx = netdev_priv(net);
843 net_device_ctx->device_ctx = dev;
844 hv_set_drvdata(dev, net);
845 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
846 INIT_WORK(&net_device_ctx->work, do_set_multicast);
848 net->netdev_ops = &device_ops;
850 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
852 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
853 NETIF_F_IP_CSUM | NETIF_F_TSO;
855 net->ethtool_ops = ðtool_ops;
856 SET_NETDEV_DEV(net, &dev->device);
858 /* Notify the netvsc driver of the new device */
859 device_info.ring_size = ring_size;
860 ret = rndis_filter_device_add(dev, &device_info);
862 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
864 hv_set_drvdata(dev, NULL);
867 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
869 nvdev = hv_get_drvdata(dev);
870 netif_set_real_num_tx_queues(net, nvdev->num_chn);
871 netif_set_real_num_rx_queues(net, nvdev->num_chn);
873 ret = register_netdev(net);
875 pr_err("Unable to register netdev.\n");
876 rndis_filter_device_remove(dev);
879 schedule_delayed_work(&net_device_ctx->dwork, 0);
885 static int netvsc_remove(struct hv_device *dev)
887 struct net_device *net;
888 struct net_device_context *ndev_ctx;
889 struct netvsc_device *net_device;
891 net_device = hv_get_drvdata(dev);
892 net = net_device->ndev;
895 dev_err(&dev->device, "No net device to remove\n");
899 net_device->start_remove = true;
901 ndev_ctx = netdev_priv(net);
902 cancel_delayed_work_sync(&ndev_ctx->dwork);
903 cancel_work_sync(&ndev_ctx->work);
905 /* Stop outbound asap */
906 netif_tx_disable(net);
908 unregister_netdev(net);
911 * Call to the vsc driver to let it know that the device is being
914 rndis_filter_device_remove(dev);
920 static const struct hv_vmbus_device_id id_table[] = {
926 MODULE_DEVICE_TABLE(vmbus, id_table);
928 /* The one and only one */
929 static struct hv_driver netvsc_drv = {
930 .name = KBUILD_MODNAME,
931 .id_table = id_table,
932 .probe = netvsc_probe,
933 .remove = netvsc_remove,
936 static void __exit netvsc_drv_exit(void)
938 vmbus_driver_unregister(&netvsc_drv);
941 static int __init netvsc_drv_init(void)
943 if (ring_size < RING_SIZE_MIN) {
944 ring_size = RING_SIZE_MIN;
945 pr_info("Increased ring_size to %d (min allowed)\n",
948 return vmbus_driver_register(&netvsc_drv);
951 MODULE_LICENSE("GPL");
952 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
954 module_init(netvsc_drv_init);
955 module_exit(netvsc_drv_exit);