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_start_queue(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;
152 static void netvsc_xmit_completion(void *context)
154 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
155 struct sk_buff *skb = (struct sk_buff *)
156 (unsigned long)packet->completion.send.send_completion_tid;
161 dev_kfree_skb_any(skb);
164 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
165 struct hv_page_buffer *pb)
169 /* Deal with compund pages by ignoring unused part
172 page += (offset >> PAGE_SHIFT);
173 offset &= ~PAGE_MASK;
178 bytes = PAGE_SIZE - offset;
181 pb[j].pfn = page_to_pfn(page);
182 pb[j].offset = offset;
188 if (offset == PAGE_SIZE && len) {
198 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
199 struct hv_page_buffer *pb)
202 char *data = skb->data;
203 int frags = skb_shinfo(skb)->nr_frags;
206 /* The packet is laid out thus:
209 * 3. skb fragment data
212 slots_used += fill_pg_buf(virt_to_page(hdr),
214 len, &pb[slots_used]);
216 slots_used += fill_pg_buf(virt_to_page(data),
217 offset_in_page(data),
218 skb_headlen(skb), &pb[slots_used]);
220 for (i = 0; i < frags; i++) {
221 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
223 slots_used += fill_pg_buf(skb_frag_page(frag),
225 skb_frag_size(frag), &pb[slots_used]);
230 static int count_skb_frag_slots(struct sk_buff *skb)
232 int i, frags = skb_shinfo(skb)->nr_frags;
235 for (i = 0; i < frags; i++) {
236 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
237 unsigned long size = skb_frag_size(frag);
238 unsigned long offset = frag->page_offset;
240 /* Skip unused frames from start of page */
241 offset &= ~PAGE_MASK;
242 pages += PFN_UP(offset + size);
247 static int netvsc_get_slots(struct sk_buff *skb)
249 char *data = skb->data;
250 unsigned int offset = offset_in_page(data);
251 unsigned int len = skb_headlen(skb);
255 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
256 frag_slots = count_skb_frag_slots(skb);
257 return slots + frag_slots;
260 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
262 u32 ret_val = TRANSPORT_INFO_NOT_IP;
264 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
265 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
269 *trans_off = skb_transport_offset(skb);
271 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
272 struct iphdr *iphdr = ip_hdr(skb);
274 if (iphdr->protocol == IPPROTO_TCP)
275 ret_val = TRANSPORT_INFO_IPV4_TCP;
276 else if (iphdr->protocol == IPPROTO_UDP)
277 ret_val = TRANSPORT_INFO_IPV4_UDP;
279 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
280 ret_val = TRANSPORT_INFO_IPV6_TCP;
281 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
282 ret_val = TRANSPORT_INFO_IPV6_UDP;
289 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
291 struct net_device_context *net_device_ctx = netdev_priv(net);
292 struct hv_netvsc_packet *packet;
294 unsigned int num_data_pgs;
295 struct rndis_message *rndis_msg;
296 struct rndis_packet *rndis_pkt;
299 struct rndis_per_packet_info *ppi;
300 struct ndis_tcp_ip_checksum_info *csum_info;
301 struct ndis_tcp_lso_info *lso_info;
306 /* We will atmost need two pages to describe the rndis
307 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
308 * of pages in a single packet.
310 num_data_pgs = netvsc_get_slots(skb) + 2;
311 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
312 netdev_err(net, "Packet too big: %u\n", skb->len);
314 net->stats.tx_dropped++;
318 /* Allocate a netvsc packet based on # of frags. */
319 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
320 (num_data_pgs * sizeof(struct hv_page_buffer)) +
321 sizeof(struct rndis_message) +
322 NDIS_VLAN_PPI_SIZE, GFP_ATOMIC);
324 /* out of memory, drop packet */
325 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
328 net->stats.tx_dropped++;
332 packet->vlan_tci = skb->vlan_tci;
334 packet->is_data_pkt = true;
335 packet->total_data_buflen = skb->len;
337 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
338 sizeof(struct hv_netvsc_packet) +
339 (num_data_pgs * sizeof(struct hv_page_buffer)));
341 /* Set the completion routine */
342 packet->completion.send.send_completion = netvsc_xmit_completion;
343 packet->completion.send.send_completion_ctx = packet;
344 packet->completion.send.send_completion_tid = (unsigned long)skb;
346 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
348 /* Add the rndis header */
349 rndis_msg = packet->rndis_msg;
350 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
351 rndis_msg->msg_len = packet->total_data_buflen;
352 rndis_pkt = &rndis_msg->msg.pkt;
353 rndis_pkt->data_offset = sizeof(struct rndis_packet);
354 rndis_pkt->data_len = packet->total_data_buflen;
355 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
357 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
360 struct ndis_pkt_8021q_info *vlan;
362 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
363 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
365 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
367 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
368 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
372 net_trans_info = get_net_transport_info(skb, &hdr_offset);
373 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
377 * Setup the sendside checksum offload only if this is not a
383 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
384 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
385 TCPIP_CHKSUM_PKTINFO);
387 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
390 if (net_trans_info & (INFO_IPV4 << 16))
391 csum_info->transmit.is_ipv4 = 1;
393 csum_info->transmit.is_ipv6 = 1;
395 if (net_trans_info & INFO_TCP) {
396 csum_info->transmit.tcp_checksum = 1;
397 csum_info->transmit.tcp_header_offset = hdr_offset;
398 } else if (net_trans_info & INFO_UDP) {
399 csum_info->transmit.udp_checksum = 1;
404 rndis_msg_size += NDIS_LSO_PPI_SIZE;
405 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
406 TCP_LARGESEND_PKTINFO);
408 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
411 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
412 if (net_trans_info & (INFO_IPV4 << 16)) {
413 lso_info->lso_v2_transmit.ip_version =
414 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
415 ip_hdr(skb)->tot_len = 0;
416 ip_hdr(skb)->check = 0;
417 tcp_hdr(skb)->check =
418 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
419 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
421 lso_info->lso_v2_transmit.ip_version =
422 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
423 ipv6_hdr(skb)->payload_len = 0;
424 tcp_hdr(skb)->check =
425 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
426 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
428 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
429 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
432 /* Start filling in the page buffers with the rndis hdr */
433 rndis_msg->msg_len += rndis_msg_size;
434 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
435 skb, &packet->page_buf[0]);
437 ret = netvsc_send(net_device_ctx->device_ctx, packet);
440 net->stats.tx_bytes += skb->len;
441 net->stats.tx_packets++;
444 if (ret != -EAGAIN) {
445 dev_kfree_skb_any(skb);
446 net->stats.tx_dropped++;
450 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
454 * netvsc_linkstatus_callback - Link up/down notification
456 void netvsc_linkstatus_callback(struct hv_device *device_obj,
459 struct net_device *net;
460 struct net_device_context *ndev_ctx;
461 struct netvsc_device *net_device;
462 struct rndis_device *rdev;
464 net_device = hv_get_drvdata(device_obj);
465 rdev = net_device->extension;
467 rdev->link_state = status != 1;
469 net = net_device->ndev;
471 if (!net || net->reg_state != NETREG_REGISTERED)
474 ndev_ctx = netdev_priv(net);
476 schedule_delayed_work(&ndev_ctx->dwork, 0);
477 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
479 schedule_delayed_work(&ndev_ctx->dwork, 0);
484 * netvsc_recv_callback - Callback when we receive a packet from the
485 * "wire" on the specified device.
487 int netvsc_recv_callback(struct hv_device *device_obj,
488 struct hv_netvsc_packet *packet,
489 struct ndis_tcp_ip_checksum_info *csum_info)
491 struct net_device *net;
494 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
495 if (!net || net->reg_state != NETREG_REGISTERED) {
496 packet->status = NVSP_STAT_FAIL;
500 /* Allocate a skb - TODO direct I/O to pages? */
501 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
502 if (unlikely(!skb)) {
503 ++net->stats.rx_dropped;
504 packet->status = NVSP_STAT_FAIL;
509 * Copy to skb. This copy is needed here since the memory pointed by
510 * hv_netvsc_packet cannot be deallocated
512 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
513 packet->total_data_buflen);
515 skb->protocol = eth_type_trans(skb, net);
517 /* We only look at the IP checksum here.
518 * Should we be dropping the packet if checksum
519 * failed? How do we deal with other checksums - TCP/UDP?
521 if (csum_info->receive.ip_checksum_succeeded)
522 skb->ip_summed = CHECKSUM_UNNECESSARY;
524 skb->ip_summed = CHECKSUM_NONE;
527 if (packet->vlan_tci & VLAN_TAG_PRESENT)
528 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
531 net->stats.rx_packets++;
532 net->stats.rx_bytes += packet->total_data_buflen;
535 * Pass the skb back up. Network stack will deallocate the skb when it
544 static void netvsc_get_drvinfo(struct net_device *net,
545 struct ethtool_drvinfo *info)
547 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
548 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
551 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
553 struct net_device_context *ndevctx = netdev_priv(ndev);
554 struct hv_device *hdev = ndevctx->device_ctx;
555 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
556 struct netvsc_device_info device_info;
557 int limit = ETH_DATA_LEN;
559 if (nvdev == NULL || nvdev->destroy)
562 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
565 if (mtu < 68 || mtu > limit)
568 nvdev->start_remove = true;
569 cancel_work_sync(&ndevctx->work);
570 netif_tx_disable(ndev);
571 rndis_filter_device_remove(hdev);
575 ndevctx->device_ctx = hdev;
576 hv_set_drvdata(hdev, ndev);
577 device_info.ring_size = ring_size;
578 rndis_filter_device_add(hdev, &device_info);
579 netif_wake_queue(ndev);
585 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
587 struct net_device_context *ndevctx = netdev_priv(ndev);
588 struct hv_device *hdev = ndevctx->device_ctx;
589 struct sockaddr *addr = p;
590 char save_adr[ETH_ALEN];
591 unsigned char save_aatype;
594 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
595 save_aatype = ndev->addr_assign_type;
597 err = eth_mac_addr(ndev, p);
601 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
603 /* roll back to saved MAC */
604 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
605 ndev->addr_assign_type = save_aatype;
612 static const struct ethtool_ops ethtool_ops = {
613 .get_drvinfo = netvsc_get_drvinfo,
614 .get_link = ethtool_op_get_link,
617 static const struct net_device_ops device_ops = {
618 .ndo_open = netvsc_open,
619 .ndo_stop = netvsc_close,
620 .ndo_start_xmit = netvsc_start_xmit,
621 .ndo_set_rx_mode = netvsc_set_multicast_list,
622 .ndo_change_mtu = netvsc_change_mtu,
623 .ndo_validate_addr = eth_validate_addr,
624 .ndo_set_mac_address = netvsc_set_mac_addr,
628 * Send GARP packet to network peers after migrations.
629 * After Quick Migration, the network is not immediately operational in the
630 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
631 * another netif_notify_peers() into a delayed work, otherwise GARP packet
632 * will not be sent after quick migration, and cause network disconnection.
633 * Also, we update the carrier status here.
635 static void netvsc_link_change(struct work_struct *w)
637 struct net_device_context *ndev_ctx;
638 struct net_device *net;
639 struct netvsc_device *net_device;
640 struct rndis_device *rdev;
645 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
646 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
647 rdev = net_device->extension;
648 net = net_device->ndev;
650 if (rdev->link_state) {
651 netif_carrier_off(net);
654 netif_carrier_on(net);
661 netdev_notify_peers(net);
665 static int netvsc_probe(struct hv_device *dev,
666 const struct hv_vmbus_device_id *dev_id)
668 struct net_device *net = NULL;
669 struct net_device_context *net_device_ctx;
670 struct netvsc_device_info device_info;
673 net = alloc_etherdev(sizeof(struct net_device_context));
677 netif_carrier_off(net);
679 net_device_ctx = netdev_priv(net);
680 net_device_ctx->device_ctx = dev;
681 hv_set_drvdata(dev, net);
682 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
683 INIT_WORK(&net_device_ctx->work, do_set_multicast);
685 net->netdev_ops = &device_ops;
687 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
689 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
690 NETIF_F_IP_CSUM | NETIF_F_TSO;
692 SET_ETHTOOL_OPS(net, ðtool_ops);
693 SET_NETDEV_DEV(net, &dev->device);
695 /* Notify the netvsc driver of the new device */
696 device_info.ring_size = ring_size;
697 ret = rndis_filter_device_add(dev, &device_info);
699 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
701 hv_set_drvdata(dev, NULL);
704 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
706 ret = register_netdev(net);
708 pr_err("Unable to register netdev.\n");
709 rndis_filter_device_remove(dev);
712 schedule_delayed_work(&net_device_ctx->dwork, 0);
718 static int netvsc_remove(struct hv_device *dev)
720 struct net_device *net;
721 struct net_device_context *ndev_ctx;
722 struct netvsc_device *net_device;
724 net_device = hv_get_drvdata(dev);
725 net = net_device->ndev;
728 dev_err(&dev->device, "No net device to remove\n");
732 net_device->start_remove = true;
734 ndev_ctx = netdev_priv(net);
735 cancel_delayed_work_sync(&ndev_ctx->dwork);
736 cancel_work_sync(&ndev_ctx->work);
738 /* Stop outbound asap */
739 netif_tx_disable(net);
741 unregister_netdev(net);
744 * Call to the vsc driver to let it know that the device is being
747 rndis_filter_device_remove(dev);
753 static const struct hv_vmbus_device_id id_table[] = {
759 MODULE_DEVICE_TABLE(vmbus, id_table);
761 /* The one and only one */
762 static struct hv_driver netvsc_drv = {
763 .name = KBUILD_MODNAME,
764 .id_table = id_table,
765 .probe = netvsc_probe,
766 .remove = netvsc_remove,
769 static void __exit netvsc_drv_exit(void)
771 vmbus_driver_unregister(&netvsc_drv);
774 static int __init netvsc_drv_init(void)
776 if (ring_size < RING_SIZE_MIN) {
777 ring_size = RING_SIZE_MIN;
778 pr_info("Increased ring_size to %d (min allowed)\n",
781 return vmbus_driver_register(&netvsc_drv);
784 MODULE_LICENSE("GPL");
785 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
787 module_init(netvsc_drv_init);
788 module_exit(netvsc_drv_exit);