1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2009, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/init.h>
12 #include <linux/atomic.h>
13 #include <linux/ethtool.h>
14 #include <linux/module.h>
15 #include <linux/highmem.h>
16 #include <linux/device.h>
18 #include <linux/delay.h>
19 #include <linux/netdevice.h>
20 #include <linux/inetdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/pci.h>
23 #include <linux/skbuff.h>
24 #include <linux/if_vlan.h>
26 #include <linux/slab.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/netpoll.h>
29 #include <linux/bpf.h>
32 #include <net/route.h>
34 #include <net/pkt_sched.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
38 #include "hyperv_net.h"
40 #define RING_SIZE_MIN 64
42 #define LINKCHANGE_INT (2 * HZ)
43 #define VF_TAKEOVER_INT (HZ / 10)
45 /* Macros to define the context of vf registration */
46 #define VF_REG_IN_PROBE 1
47 #define VF_REG_IN_NOTIFIER 2
49 static unsigned int ring_size __ro_after_init = 128;
50 module_param(ring_size, uint, 0444);
51 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of 4K pages)");
52 unsigned int netvsc_ring_bytes __ro_after_init;
54 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
55 NETIF_MSG_LINK | NETIF_MSG_IFUP |
56 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
59 static int debug = -1;
60 module_param(debug, int, 0444);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
63 static LIST_HEAD(netvsc_dev_list);
65 static void netvsc_change_rx_flags(struct net_device *net, int change)
67 struct net_device_context *ndev_ctx = netdev_priv(net);
68 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
74 if (change & IFF_PROMISC) {
75 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
76 dev_set_promiscuity(vf_netdev, inc);
79 if (change & IFF_ALLMULTI) {
80 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
81 dev_set_allmulti(vf_netdev, inc);
85 static void netvsc_set_rx_mode(struct net_device *net)
87 struct net_device_context *ndev_ctx = netdev_priv(net);
88 struct net_device *vf_netdev;
89 struct netvsc_device *nvdev;
92 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
94 dev_uc_sync(vf_netdev, net);
95 dev_mc_sync(vf_netdev, net);
98 nvdev = rcu_dereference(ndev_ctx->nvdev);
100 rndis_filter_update(nvdev);
104 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
105 struct net_device *ndev)
107 nvscdev->tx_disable = false;
108 virt_wmb(); /* ensure queue wake up mechanism is on */
110 netif_tx_wake_all_queues(ndev);
113 static int netvsc_open(struct net_device *net)
115 struct net_device_context *ndev_ctx = netdev_priv(net);
116 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
117 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
118 struct rndis_device *rdev;
121 netif_carrier_off(net);
123 /* Open up the device */
124 ret = rndis_filter_open(nvdev);
126 netdev_err(net, "unable to open device (ret %d).\n", ret);
130 rdev = nvdev->extension;
131 if (!rdev->link_state) {
132 netif_carrier_on(net);
133 netvsc_tx_enable(nvdev, net);
137 /* Setting synthetic device up transparently sets
138 * slave as up. If open fails, then slave will be
139 * still be offline (and not used).
141 ret = dev_open(vf_netdev, NULL);
144 "unable to open slave: %s: %d\n",
145 vf_netdev->name, ret);
150 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
152 unsigned int retry = 0;
155 /* Ensure pending bytes in ring are read */
159 for (i = 0; i < nvdev->num_chn; i++) {
160 struct vmbus_channel *chn
161 = nvdev->chan_table[i].channel;
166 /* make sure receive not running now */
167 napi_synchronize(&nvdev->chan_table[i].napi);
169 aread = hv_get_bytes_to_read(&chn->inbound);
173 aread = hv_get_bytes_to_read(&chn->outbound);
181 if (++retry > RETRY_MAX)
184 usleep_range(RETRY_US_LO, RETRY_US_HI);
188 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
189 struct net_device *ndev)
192 nvscdev->tx_disable = true;
193 virt_wmb(); /* ensure txq will not wake up after stop */
196 netif_tx_disable(ndev);
199 static int netvsc_close(struct net_device *net)
201 struct net_device_context *net_device_ctx = netdev_priv(net);
202 struct net_device *vf_netdev
203 = rtnl_dereference(net_device_ctx->vf_netdev);
204 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
207 netvsc_tx_disable(nvdev, net);
209 /* No need to close rndis filter if it is removed already */
213 ret = rndis_filter_close(nvdev);
215 netdev_err(net, "unable to close device (ret %d).\n", ret);
219 ret = netvsc_wait_until_empty(nvdev);
221 netdev_err(net, "Ring buffer not empty after closing rndis\n");
224 dev_close(vf_netdev);
229 static inline void *init_ppi_data(struct rndis_message *msg,
230 u32 ppi_size, u32 pkt_type)
232 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
233 struct rndis_per_packet_info *ppi;
235 rndis_pkt->data_offset += ppi_size;
236 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
237 + rndis_pkt->per_pkt_info_len;
239 ppi->size = ppi_size;
240 ppi->type = pkt_type;
242 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
244 rndis_pkt->per_pkt_info_len += ppi_size;
249 static inline int netvsc_get_tx_queue(struct net_device *ndev,
250 struct sk_buff *skb, int old_idx)
252 const struct net_device_context *ndc = netdev_priv(ndev);
253 struct sock *sk = skb->sk;
256 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
257 (VRSS_SEND_TAB_SIZE - 1)];
259 /* If queue index changed record the new value */
260 if (q_idx != old_idx &&
261 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
262 sk_tx_queue_set(sk, q_idx);
268 * Select queue for transmit.
270 * If a valid queue has already been assigned, then use that.
271 * Otherwise compute tx queue based on hash and the send table.
273 * This is basically similar to default (netdev_pick_tx) with the added step
274 * of using the host send_table when no other queue has been assigned.
276 * TODO support XPS - but get_xps_queue not exported
278 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
280 int q_idx = sk_tx_queue_get(skb->sk);
282 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
283 /* If forwarding a packet, we use the recorded queue when
284 * available for better cache locality.
286 if (skb_rx_queue_recorded(skb))
287 q_idx = skb_get_rx_queue(skb);
289 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
295 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
296 struct net_device *sb_dev)
298 struct net_device_context *ndc = netdev_priv(ndev);
299 struct net_device *vf_netdev;
303 vf_netdev = rcu_dereference(ndc->vf_netdev);
305 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
307 if (vf_ops->ndo_select_queue)
308 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
310 txq = netdev_pick_tx(vf_netdev, skb, NULL);
312 /* Record the queue selected by VF so that it can be
313 * used for common case where VF has more queues than
314 * the synthetic device.
316 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
318 txq = netvsc_pick_tx(ndev, skb);
322 while (txq >= ndev->real_num_tx_queues)
323 txq -= ndev->real_num_tx_queues;
328 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
329 struct hv_page_buffer *pb)
333 hvpfn += offset >> HV_HYP_PAGE_SHIFT;
334 offset = offset & ~HV_HYP_PAGE_MASK;
339 bytes = HV_HYP_PAGE_SIZE - offset;
343 pb[j].offset = offset;
349 if (offset == HV_HYP_PAGE_SIZE && len) {
359 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
360 struct hv_netvsc_packet *packet,
361 struct hv_page_buffer *pb)
364 char *data = skb->data;
365 int frags = skb_shinfo(skb)->nr_frags;
368 /* The packet is laid out thus:
369 * 1. hdr: RNDIS header and PPI
371 * 3. skb fragment data
373 slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
374 offset_in_hvpage(hdr),
378 packet->rmsg_size = len;
379 packet->rmsg_pgcnt = slots_used;
381 slots_used += fill_pg_buf(virt_to_hvpfn(data),
382 offset_in_hvpage(data),
386 for (i = 0; i < frags; i++) {
387 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
389 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
397 static int count_skb_frag_slots(struct sk_buff *skb)
399 int i, frags = skb_shinfo(skb)->nr_frags;
402 for (i = 0; i < frags; i++) {
403 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
404 unsigned long size = skb_frag_size(frag);
405 unsigned long offset = skb_frag_off(frag);
407 /* Skip unused frames from start of page */
408 offset &= ~HV_HYP_PAGE_MASK;
409 pages += HVPFN_UP(offset + size);
414 static int netvsc_get_slots(struct sk_buff *skb)
416 char *data = skb->data;
417 unsigned int offset = offset_in_hvpage(data);
418 unsigned int len = skb_headlen(skb);
422 slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
423 frag_slots = count_skb_frag_slots(skb);
424 return slots + frag_slots;
427 static u32 net_checksum_info(struct sk_buff *skb)
429 if (skb->protocol == htons(ETH_P_IP)) {
430 struct iphdr *ip = ip_hdr(skb);
432 if (ip->protocol == IPPROTO_TCP)
433 return TRANSPORT_INFO_IPV4_TCP;
434 else if (ip->protocol == IPPROTO_UDP)
435 return TRANSPORT_INFO_IPV4_UDP;
437 struct ipv6hdr *ip6 = ipv6_hdr(skb);
439 if (ip6->nexthdr == IPPROTO_TCP)
440 return TRANSPORT_INFO_IPV6_TCP;
441 else if (ip6->nexthdr == IPPROTO_UDP)
442 return TRANSPORT_INFO_IPV6_UDP;
445 return TRANSPORT_INFO_NOT_IP;
448 /* Send skb on the slave VF device. */
449 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
452 struct net_device_context *ndev_ctx = netdev_priv(net);
453 unsigned int len = skb->len;
456 skb->dev = vf_netdev;
457 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
459 rc = dev_queue_xmit(skb);
460 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
461 struct netvsc_vf_pcpu_stats *pcpu_stats
462 = this_cpu_ptr(ndev_ctx->vf_stats);
464 u64_stats_update_begin(&pcpu_stats->syncp);
465 pcpu_stats->tx_packets++;
466 pcpu_stats->tx_bytes += len;
467 u64_stats_update_end(&pcpu_stats->syncp);
469 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
475 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
477 struct net_device_context *net_device_ctx = netdev_priv(net);
478 struct hv_netvsc_packet *packet = NULL;
480 unsigned int num_data_pgs;
481 struct rndis_message *rndis_msg;
482 struct net_device *vf_netdev;
485 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
487 /* If VF is present and up then redirect packets to it.
488 * Skip the VF if it is marked down or has no carrier.
489 * If netpoll is in uses, then VF can not be used either.
491 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
492 if (vf_netdev && netif_running(vf_netdev) &&
493 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
494 net_device_ctx->data_path_is_vf)
495 return netvsc_vf_xmit(net, vf_netdev, skb);
497 /* We will atmost need two pages to describe the rndis
498 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
499 * of pages in a single packet. If skb is scattered around
500 * more pages we try linearizing it.
503 num_data_pgs = netvsc_get_slots(skb) + 2;
505 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
506 ++net_device_ctx->eth_stats.tx_scattered;
508 if (skb_linearize(skb))
511 num_data_pgs = netvsc_get_slots(skb) + 2;
512 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
513 ++net_device_ctx->eth_stats.tx_too_big;
519 * Place the rndis header in the skb head room and
520 * the skb->cb will be used for hv_netvsc_packet
523 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
527 /* Use the skb control buffer for building up the packet */
528 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
529 sizeof_field(struct sk_buff, cb));
530 packet = (struct hv_netvsc_packet *)skb->cb;
532 packet->q_idx = skb_get_queue_mapping(skb);
534 packet->total_data_buflen = skb->len;
535 packet->total_bytes = skb->len;
536 packet->total_packets = 1;
538 rndis_msg = (struct rndis_message *)skb->head;
540 /* Add the rndis header */
541 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
542 rndis_msg->msg_len = packet->total_data_buflen;
544 rndis_msg->msg.pkt = (struct rndis_packet) {
545 .data_offset = sizeof(struct rndis_packet),
546 .data_len = packet->total_data_buflen,
547 .per_pkt_info_offset = sizeof(struct rndis_packet),
550 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
552 hash = skb_get_hash_raw(skb);
553 if (hash != 0 && net->real_num_tx_queues > 1) {
556 rndis_msg_size += NDIS_HASH_PPI_SIZE;
557 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
562 /* When using AF_PACKET we need to drop VLAN header from
563 * the frame and update the SKB to allow the HOST OS
564 * to transmit the 802.1Q packet
566 if (skb->protocol == htons(ETH_P_8021Q)) {
569 skb_reset_mac_header(skb);
570 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
571 if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
572 ++net_device_ctx->eth_stats.vlan_error;
576 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
577 /* Update the NDIS header pkt lengths */
578 packet->total_data_buflen -= VLAN_HLEN;
579 packet->total_bytes -= VLAN_HLEN;
580 rndis_msg->msg_len = packet->total_data_buflen;
581 rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
585 if (skb_vlan_tag_present(skb)) {
586 struct ndis_pkt_8021q_info *vlan;
588 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
589 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
593 vlan->vlanid = skb_vlan_tag_get_id(skb);
594 vlan->cfi = skb_vlan_tag_get_cfi(skb);
595 vlan->pri = skb_vlan_tag_get_prio(skb);
598 if (skb_is_gso(skb)) {
599 struct ndis_tcp_lso_info *lso_info;
601 rndis_msg_size += NDIS_LSO_PPI_SIZE;
602 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
603 TCP_LARGESEND_PKTINFO);
606 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
607 if (skb->protocol == htons(ETH_P_IP)) {
608 lso_info->lso_v2_transmit.ip_version =
609 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
610 ip_hdr(skb)->tot_len = 0;
611 ip_hdr(skb)->check = 0;
612 tcp_hdr(skb)->check =
613 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
614 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
616 lso_info->lso_v2_transmit.ip_version =
617 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
618 tcp_v6_gso_csum_prep(skb);
620 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
621 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
622 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
623 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
624 struct ndis_tcp_ip_checksum_info *csum_info;
626 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
627 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
628 TCPIP_CHKSUM_PKTINFO);
630 csum_info->value = 0;
631 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
633 if (skb->protocol == htons(ETH_P_IP)) {
634 csum_info->transmit.is_ipv4 = 1;
636 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
637 csum_info->transmit.tcp_checksum = 1;
639 csum_info->transmit.udp_checksum = 1;
641 csum_info->transmit.is_ipv6 = 1;
643 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
644 csum_info->transmit.tcp_checksum = 1;
646 csum_info->transmit.udp_checksum = 1;
649 /* Can't do offload of this type of checksum */
650 if (skb_checksum_help(skb))
655 /* Start filling in the page buffers with the rndis hdr */
656 rndis_msg->msg_len += rndis_msg_size;
657 packet->total_data_buflen = rndis_msg->msg_len;
658 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
661 /* timestamp packet in software */
662 skb_tx_timestamp(skb);
664 ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
665 if (likely(ret == 0))
668 if (ret == -EAGAIN) {
669 ++net_device_ctx->eth_stats.tx_busy;
670 return NETDEV_TX_BUSY;
674 ++net_device_ctx->eth_stats.tx_no_space;
677 dev_kfree_skb_any(skb);
678 net->stats.tx_dropped++;
683 ++net_device_ctx->eth_stats.tx_no_memory;
687 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
688 struct net_device *ndev)
690 return netvsc_xmit(skb, ndev, false);
694 * netvsc_linkstatus_callback - Link up/down notification
696 void netvsc_linkstatus_callback(struct net_device *net,
697 struct rndis_message *resp,
698 void *data, u32 data_buflen)
700 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
701 struct net_device_context *ndev_ctx = netdev_priv(net);
702 struct netvsc_reconfig *event;
705 /* Ensure the packet is big enough to access its fields */
706 if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
707 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
712 /* Copy the RNDIS indicate status into nvchan->recv_buf */
713 memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
715 /* Update the physical link speed when changing to another vSwitch */
716 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
719 /* Validate status_buf_offset and status_buflen.
721 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
722 * for the status buffer field in resp->msg_len; perform the validation
723 * using data_buflen (>= resp->msg_len).
725 if (indicate->status_buflen < sizeof(speed) ||
726 indicate->status_buf_offset < sizeof(*indicate) ||
727 data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
728 data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
729 < indicate->status_buflen) {
730 netdev_err(net, "invalid rndis_indicate_status packet\n");
734 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
735 ndev_ctx->speed = speed;
739 /* Handle these link change statuses below */
740 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
741 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
742 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
745 if (net->reg_state != NETREG_REGISTERED)
748 event = kzalloc(sizeof(*event), GFP_ATOMIC);
751 event->event = indicate->status;
753 spin_lock_irqsave(&ndev_ctx->lock, flags);
754 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
755 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
757 schedule_delayed_work(&ndev_ctx->dwork, 0);
760 /* This function should only be called after skb_record_rx_queue() */
761 void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
765 skb->queue_mapping = skb_get_rx_queue(skb);
766 __skb_push(skb, ETH_HLEN);
768 rc = netvsc_xmit(skb, ndev, true);
770 if (dev_xmit_complete(rc))
773 dev_kfree_skb_any(skb);
774 ndev->stats.tx_dropped++;
777 static void netvsc_comp_ipcsum(struct sk_buff *skb)
779 struct iphdr *iph = (struct iphdr *)skb->data;
782 iph->check = ip_fast_csum(iph, iph->ihl);
785 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
786 struct netvsc_channel *nvchan,
787 struct xdp_buff *xdp)
789 struct napi_struct *napi = &nvchan->napi;
790 const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
791 const struct ndis_tcp_ip_checksum_info *csum_info =
792 &nvchan->rsc.csum_info;
793 const u32 *hash_info = &nvchan->rsc.hash_info;
794 u8 ppi_flags = nvchan->rsc.ppi_flags;
796 void *xbuf = xdp->data_hard_start;
800 unsigned int hdroom = xdp->data - xdp->data_hard_start;
801 unsigned int xlen = xdp->data_end - xdp->data;
802 unsigned int frag_size = xdp->frame_sz;
804 skb = build_skb(xbuf, frag_size);
807 __free_page(virt_to_page(xbuf));
811 skb_reserve(skb, hdroom);
813 skb->dev = napi->dev;
815 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
820 /* Copy to skb. This copy is needed here since the memory
821 * pointed by hv_netvsc_packet cannot be deallocated.
823 for (i = 0; i < nvchan->rsc.cnt; i++)
824 skb_put_data(skb, nvchan->rsc.data[i],
828 skb->protocol = eth_type_trans(skb, net);
830 /* skb is already created with CHECKSUM_NONE */
831 skb_checksum_none_assert(skb);
833 /* Incoming packets may have IP header checksum verified by the host.
834 * They may not have IP header checksum computed after coalescing.
835 * We compute it here if the flags are set, because on Linux, the IP
836 * checksum is always checked.
838 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
839 csum_info->receive.ip_checksum_succeeded &&
840 skb->protocol == htons(ETH_P_IP)) {
841 /* Check that there is enough space to hold the IP header. */
842 if (skb_headlen(skb) < sizeof(struct iphdr)) {
846 netvsc_comp_ipcsum(skb);
849 /* Do L4 checksum offload if enabled and present. */
850 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
851 if (csum_info->receive.tcp_checksum_succeeded ||
852 csum_info->receive.udp_checksum_succeeded)
853 skb->ip_summed = CHECKSUM_UNNECESSARY;
856 if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
857 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
859 if (ppi_flags & NVSC_RSC_VLAN) {
860 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
861 (vlan->cfi ? VLAN_CFI_MASK : 0);
863 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
871 * netvsc_recv_callback - Callback when we receive a packet from the
872 * "wire" on the specified device.
874 int netvsc_recv_callback(struct net_device *net,
875 struct netvsc_device *net_device,
876 struct netvsc_channel *nvchan)
878 struct net_device_context *net_device_ctx = netdev_priv(net);
879 struct vmbus_channel *channel = nvchan->channel;
880 u16 q_idx = channel->offermsg.offer.sub_channel_index;
882 struct netvsc_stats_rx *rx_stats = &nvchan->rx_stats;
886 if (net->reg_state != NETREG_REGISTERED)
887 return NVSP_STAT_FAIL;
889 act = netvsc_run_xdp(net, nvchan, &xdp);
891 if (act == XDP_REDIRECT)
892 return NVSP_STAT_SUCCESS;
894 if (act != XDP_PASS && act != XDP_TX) {
895 u64_stats_update_begin(&rx_stats->syncp);
896 rx_stats->xdp_drop++;
897 u64_stats_update_end(&rx_stats->syncp);
899 return NVSP_STAT_SUCCESS; /* consumed by XDP */
902 /* Allocate a skb - TODO direct I/O to pages? */
903 skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
905 if (unlikely(!skb)) {
906 ++net_device_ctx->eth_stats.rx_no_memory;
907 return NVSP_STAT_FAIL;
910 skb_record_rx_queue(skb, q_idx);
913 * Even if injecting the packet, record the statistics
914 * on the synthetic device because modifying the VF device
915 * statistics will not work correctly.
917 u64_stats_update_begin(&rx_stats->syncp);
922 rx_stats->bytes += nvchan->rsc.pktlen;
924 if (skb->pkt_type == PACKET_BROADCAST)
925 ++rx_stats->broadcast;
926 else if (skb->pkt_type == PACKET_MULTICAST)
927 ++rx_stats->multicast;
928 u64_stats_update_end(&rx_stats->syncp);
931 netvsc_xdp_xmit(skb, net);
932 return NVSP_STAT_SUCCESS;
935 napi_gro_receive(&nvchan->napi, skb);
936 return NVSP_STAT_SUCCESS;
939 static void netvsc_get_drvinfo(struct net_device *net,
940 struct ethtool_drvinfo *info)
942 strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
943 strscpy(info->fw_version, "N/A", sizeof(info->fw_version));
946 static void netvsc_get_channels(struct net_device *net,
947 struct ethtool_channels *channel)
949 struct net_device_context *net_device_ctx = netdev_priv(net);
950 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
953 channel->max_combined = nvdev->max_chn;
954 channel->combined_count = nvdev->num_chn;
958 /* Alloc struct netvsc_device_info, and initialize it from either existing
959 * struct netvsc_device, or from default values.
962 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
964 struct netvsc_device_info *dev_info;
965 struct bpf_prog *prog;
967 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
975 dev_info->num_chn = nvdev->num_chn;
976 dev_info->send_sections = nvdev->send_section_cnt;
977 dev_info->send_section_size = nvdev->send_section_size;
978 dev_info->recv_sections = nvdev->recv_section_cnt;
979 dev_info->recv_section_size = nvdev->recv_section_size;
981 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
984 prog = netvsc_xdp_get(nvdev);
987 dev_info->bprog = prog;
990 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
991 dev_info->send_sections = NETVSC_DEFAULT_TX;
992 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
993 dev_info->recv_sections = NETVSC_DEFAULT_RX;
994 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1000 /* Free struct netvsc_device_info */
1001 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1003 if (dev_info->bprog) {
1005 bpf_prog_put(dev_info->bprog);
1011 static int netvsc_detach(struct net_device *ndev,
1012 struct netvsc_device *nvdev)
1014 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1015 struct hv_device *hdev = ndev_ctx->device_ctx;
1018 /* Don't try continuing to try and setup sub channels */
1019 if (cancel_work_sync(&nvdev->subchan_work))
1022 netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1024 /* If device was up (receiving) then shutdown */
1025 if (netif_running(ndev)) {
1026 netvsc_tx_disable(nvdev, ndev);
1028 ret = rndis_filter_close(nvdev);
1031 "unable to close device (ret %d).\n", ret);
1035 ret = netvsc_wait_until_empty(nvdev);
1038 "Ring buffer not empty after closing rndis\n");
1043 netif_device_detach(ndev);
1045 rndis_filter_device_remove(hdev, nvdev);
1050 static int netvsc_attach(struct net_device *ndev,
1051 struct netvsc_device_info *dev_info)
1053 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1054 struct hv_device *hdev = ndev_ctx->device_ctx;
1055 struct netvsc_device *nvdev;
1056 struct rndis_device *rdev;
1057 struct bpf_prog *prog;
1060 nvdev = rndis_filter_device_add(hdev, dev_info);
1062 return PTR_ERR(nvdev);
1064 if (nvdev->num_chn > 1) {
1065 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1067 /* if unavailable, just proceed with one queue */
1074 prog = dev_info->bprog;
1077 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1084 /* In any case device is now ready */
1085 nvdev->tx_disable = false;
1086 netif_device_attach(ndev);
1088 /* Note: enable and attach happen when sub-channels setup */
1089 netif_carrier_off(ndev);
1091 if (netif_running(ndev)) {
1092 ret = rndis_filter_open(nvdev);
1096 rdev = nvdev->extension;
1097 if (!rdev->link_state)
1098 netif_carrier_on(ndev);
1104 netif_device_detach(ndev);
1107 rndis_filter_device_remove(hdev, nvdev);
1112 static int netvsc_set_channels(struct net_device *net,
1113 struct ethtool_channels *channels)
1115 struct net_device_context *net_device_ctx = netdev_priv(net);
1116 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1117 unsigned int orig, count = channels->combined_count;
1118 struct netvsc_device_info *device_info;
1121 /* We do not support separate count for rx, tx, or other */
1123 channels->rx_count || channels->tx_count || channels->other_count)
1126 if (!nvdev || nvdev->destroy)
1129 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1132 if (count > nvdev->max_chn)
1135 orig = nvdev->num_chn;
1137 device_info = netvsc_devinfo_get(nvdev);
1142 device_info->num_chn = count;
1144 ret = netvsc_detach(net, nvdev);
1148 ret = netvsc_attach(net, device_info);
1150 device_info->num_chn = orig;
1151 if (netvsc_attach(net, device_info))
1152 netdev_err(net, "restoring channel setting failed\n");
1156 netvsc_devinfo_put(device_info);
1160 static void netvsc_init_settings(struct net_device *dev)
1162 struct net_device_context *ndc = netdev_priv(dev);
1164 ndc->l4_hash = HV_DEFAULT_L4HASH;
1166 ndc->speed = SPEED_UNKNOWN;
1167 ndc->duplex = DUPLEX_FULL;
1169 dev->features = NETIF_F_LRO;
1172 static int netvsc_get_link_ksettings(struct net_device *dev,
1173 struct ethtool_link_ksettings *cmd)
1175 struct net_device_context *ndc = netdev_priv(dev);
1176 struct net_device *vf_netdev;
1178 vf_netdev = rtnl_dereference(ndc->vf_netdev);
1181 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1183 cmd->base.speed = ndc->speed;
1184 cmd->base.duplex = ndc->duplex;
1185 cmd->base.port = PORT_OTHER;
1190 static int netvsc_set_link_ksettings(struct net_device *dev,
1191 const struct ethtool_link_ksettings *cmd)
1193 struct net_device_context *ndc = netdev_priv(dev);
1194 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1197 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1200 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1204 return ethtool_virtdev_set_link_ksettings(dev, cmd,
1205 &ndc->speed, &ndc->duplex);
1208 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1210 struct net_device_context *ndevctx = netdev_priv(ndev);
1211 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1212 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1213 int orig_mtu = ndev->mtu;
1214 struct netvsc_device_info *device_info;
1217 if (!nvdev || nvdev->destroy)
1220 device_info = netvsc_devinfo_get(nvdev);
1225 /* Change MTU of underlying VF netdev first. */
1227 ret = dev_set_mtu(vf_netdev, mtu);
1232 ret = netvsc_detach(ndev, nvdev);
1238 ret = netvsc_attach(ndev, device_info);
1242 /* Attempt rollback to original MTU */
1243 ndev->mtu = orig_mtu;
1245 if (netvsc_attach(ndev, device_info))
1246 netdev_err(ndev, "restoring mtu failed\n");
1249 dev_set_mtu(vf_netdev, orig_mtu);
1252 netvsc_devinfo_put(device_info);
1256 static void netvsc_get_vf_stats(struct net_device *net,
1257 struct netvsc_vf_pcpu_stats *tot)
1259 struct net_device_context *ndev_ctx = netdev_priv(net);
1262 memset(tot, 0, sizeof(*tot));
1264 for_each_possible_cpu(i) {
1265 const struct netvsc_vf_pcpu_stats *stats
1266 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1267 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1271 start = u64_stats_fetch_begin(&stats->syncp);
1272 rx_packets = stats->rx_packets;
1273 tx_packets = stats->tx_packets;
1274 rx_bytes = stats->rx_bytes;
1275 tx_bytes = stats->tx_bytes;
1276 } while (u64_stats_fetch_retry(&stats->syncp, start));
1278 tot->rx_packets += rx_packets;
1279 tot->tx_packets += tx_packets;
1280 tot->rx_bytes += rx_bytes;
1281 tot->tx_bytes += tx_bytes;
1282 tot->tx_dropped += stats->tx_dropped;
1286 static void netvsc_get_pcpu_stats(struct net_device *net,
1287 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1289 struct net_device_context *ndev_ctx = netdev_priv(net);
1290 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1293 /* fetch percpu stats of vf */
1294 for_each_possible_cpu(i) {
1295 const struct netvsc_vf_pcpu_stats *stats =
1296 per_cpu_ptr(ndev_ctx->vf_stats, i);
1297 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1301 start = u64_stats_fetch_begin(&stats->syncp);
1302 this_tot->vf_rx_packets = stats->rx_packets;
1303 this_tot->vf_tx_packets = stats->tx_packets;
1304 this_tot->vf_rx_bytes = stats->rx_bytes;
1305 this_tot->vf_tx_bytes = stats->tx_bytes;
1306 } while (u64_stats_fetch_retry(&stats->syncp, start));
1307 this_tot->rx_packets = this_tot->vf_rx_packets;
1308 this_tot->tx_packets = this_tot->vf_tx_packets;
1309 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1310 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1313 /* fetch percpu stats of netvsc */
1314 for (i = 0; i < nvdev->num_chn; i++) {
1315 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1316 const struct netvsc_stats_tx *tx_stats;
1317 const struct netvsc_stats_rx *rx_stats;
1318 struct netvsc_ethtool_pcpu_stats *this_tot =
1319 &pcpu_tot[nvchan->channel->target_cpu];
1323 tx_stats = &nvchan->tx_stats;
1325 start = u64_stats_fetch_begin(&tx_stats->syncp);
1326 packets = tx_stats->packets;
1327 bytes = tx_stats->bytes;
1328 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1330 this_tot->tx_bytes += bytes;
1331 this_tot->tx_packets += packets;
1333 rx_stats = &nvchan->rx_stats;
1335 start = u64_stats_fetch_begin(&rx_stats->syncp);
1336 packets = rx_stats->packets;
1337 bytes = rx_stats->bytes;
1338 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1340 this_tot->rx_bytes += bytes;
1341 this_tot->rx_packets += packets;
1345 static void netvsc_get_stats64(struct net_device *net,
1346 struct rtnl_link_stats64 *t)
1348 struct net_device_context *ndev_ctx = netdev_priv(net);
1349 struct netvsc_device *nvdev;
1350 struct netvsc_vf_pcpu_stats vf_tot;
1355 nvdev = rcu_dereference(ndev_ctx->nvdev);
1359 netdev_stats_to_stats64(t, &net->stats);
1361 netvsc_get_vf_stats(net, &vf_tot);
1362 t->rx_packets += vf_tot.rx_packets;
1363 t->tx_packets += vf_tot.tx_packets;
1364 t->rx_bytes += vf_tot.rx_bytes;
1365 t->tx_bytes += vf_tot.tx_bytes;
1366 t->tx_dropped += vf_tot.tx_dropped;
1368 for (i = 0; i < nvdev->num_chn; i++) {
1369 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1370 const struct netvsc_stats_tx *tx_stats;
1371 const struct netvsc_stats_rx *rx_stats;
1372 u64 packets, bytes, multicast;
1375 tx_stats = &nvchan->tx_stats;
1377 start = u64_stats_fetch_begin(&tx_stats->syncp);
1378 packets = tx_stats->packets;
1379 bytes = tx_stats->bytes;
1380 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1382 t->tx_bytes += bytes;
1383 t->tx_packets += packets;
1385 rx_stats = &nvchan->rx_stats;
1387 start = u64_stats_fetch_begin(&rx_stats->syncp);
1388 packets = rx_stats->packets;
1389 bytes = rx_stats->bytes;
1390 multicast = rx_stats->multicast + rx_stats->broadcast;
1391 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1393 t->rx_bytes += bytes;
1394 t->rx_packets += packets;
1395 t->multicast += multicast;
1401 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1403 struct net_device_context *ndc = netdev_priv(ndev);
1404 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1405 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1406 struct sockaddr *addr = p;
1409 err = eth_prepare_mac_addr_change(ndev, p);
1417 err = dev_set_mac_address(vf_netdev, addr, NULL);
1422 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1424 eth_commit_mac_addr_change(ndev, p);
1425 } else if (vf_netdev) {
1426 /* rollback change on VF */
1427 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1428 dev_set_mac_address(vf_netdev, addr, NULL);
1434 static const struct {
1435 char name[ETH_GSTRING_LEN];
1437 } netvsc_stats[] = {
1438 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1439 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1440 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1441 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1442 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1443 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1444 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1445 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1446 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1447 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1448 { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1450 { "cpu%u_rx_packets",
1451 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1453 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1454 { "cpu%u_tx_packets",
1455 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1457 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1458 { "cpu%u_vf_rx_packets",
1459 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1460 { "cpu%u_vf_rx_bytes",
1461 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1462 { "cpu%u_vf_tx_packets",
1463 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1464 { "cpu%u_vf_tx_bytes",
1465 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1467 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1468 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1469 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1470 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1471 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1474 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1475 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1477 /* statistics per queue (rx/tx packets/bytes) */
1478 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1480 /* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1481 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
1483 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1485 struct net_device_context *ndc = netdev_priv(dev);
1486 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1491 switch (string_set) {
1493 return NETVSC_GLOBAL_STATS_LEN
1494 + NETVSC_VF_STATS_LEN
1495 + NETVSC_QUEUE_STATS_LEN(nvdev)
1496 + NETVSC_PCPU_STATS_LEN;
1502 static void netvsc_get_ethtool_stats(struct net_device *dev,
1503 struct ethtool_stats *stats, u64 *data)
1505 struct net_device_context *ndc = netdev_priv(dev);
1506 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1507 const void *nds = &ndc->eth_stats;
1508 const struct netvsc_stats_tx *tx_stats;
1509 const struct netvsc_stats_rx *rx_stats;
1510 struct netvsc_vf_pcpu_stats sum;
1511 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1523 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1524 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1526 netvsc_get_vf_stats(dev, &sum);
1527 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1528 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1530 for (j = 0; j < nvdev->num_chn; j++) {
1531 tx_stats = &nvdev->chan_table[j].tx_stats;
1534 start = u64_stats_fetch_begin(&tx_stats->syncp);
1535 packets = tx_stats->packets;
1536 bytes = tx_stats->bytes;
1537 xdp_xmit = tx_stats->xdp_xmit;
1538 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1539 data[i++] = packets;
1541 data[i++] = xdp_xmit;
1543 rx_stats = &nvdev->chan_table[j].rx_stats;
1545 start = u64_stats_fetch_begin(&rx_stats->syncp);
1546 packets = rx_stats->packets;
1547 bytes = rx_stats->bytes;
1548 xdp_drop = rx_stats->xdp_drop;
1549 xdp_redirect = rx_stats->xdp_redirect;
1550 xdp_tx = rx_stats->xdp_tx;
1551 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1552 data[i++] = packets;
1554 data[i++] = xdp_drop;
1555 data[i++] = xdp_redirect;
1559 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1560 sizeof(struct netvsc_ethtool_pcpu_stats),
1565 netvsc_get_pcpu_stats(dev, pcpu_sum);
1566 for_each_present_cpu(cpu) {
1567 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1569 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1570 data[i++] = *(u64 *)((void *)this_sum
1571 + pcpu_stats[j].offset);
1576 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1578 struct net_device_context *ndc = netdev_priv(dev);
1579 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1586 switch (stringset) {
1588 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1589 ethtool_puts(&p, netvsc_stats[i].name);
1591 for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1592 ethtool_puts(&p, vf_stats[i].name);
1594 for (i = 0; i < nvdev->num_chn; i++) {
1595 ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1596 ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1597 ethtool_sprintf(&p, "tx_queue_%u_xdp_xmit", i);
1598 ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1599 ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1600 ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1601 ethtool_sprintf(&p, "rx_queue_%u_xdp_redirect", i);
1602 ethtool_sprintf(&p, "rx_queue_%u_xdp_tx", i);
1605 for_each_present_cpu(cpu) {
1606 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1607 ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1615 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1616 struct ethtool_rxnfc *info)
1618 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1620 info->data = RXH_IP_SRC | RXH_IP_DST;
1622 switch (info->flow_type) {
1624 if (ndc->l4_hash & HV_TCP4_L4HASH)
1625 info->data |= l4_flag;
1630 if (ndc->l4_hash & HV_TCP6_L4HASH)
1631 info->data |= l4_flag;
1636 if (ndc->l4_hash & HV_UDP4_L4HASH)
1637 info->data |= l4_flag;
1642 if (ndc->l4_hash & HV_UDP6_L4HASH)
1643 info->data |= l4_flag;
1659 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1662 struct net_device_context *ndc = netdev_priv(dev);
1663 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1668 switch (info->cmd) {
1669 case ETHTOOL_GRXRINGS:
1670 info->data = nvdev->num_chn;
1674 return netvsc_get_rss_hash_opts(ndc, info);
1679 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1680 struct ethtool_rxnfc *info)
1682 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1683 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1684 switch (info->flow_type) {
1686 ndc->l4_hash |= HV_TCP4_L4HASH;
1690 ndc->l4_hash |= HV_TCP6_L4HASH;
1694 ndc->l4_hash |= HV_UDP4_L4HASH;
1698 ndc->l4_hash |= HV_UDP6_L4HASH;
1708 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1709 switch (info->flow_type) {
1711 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1715 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1719 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1723 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1737 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1739 struct net_device_context *ndc = netdev_priv(ndev);
1741 if (info->cmd == ETHTOOL_SRXFH)
1742 return netvsc_set_rss_hash_opts(ndc, info);
1747 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1749 return NETVSC_HASH_KEYLEN;
1752 static u32 netvsc_rss_indir_size(struct net_device *dev)
1754 struct net_device_context *ndc = netdev_priv(dev);
1756 return ndc->rx_table_sz;
1759 static int netvsc_get_rxfh(struct net_device *dev,
1760 struct ethtool_rxfh_param *rxfh)
1762 struct net_device_context *ndc = netdev_priv(dev);
1763 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1764 struct rndis_device *rndis_dev;
1770 rxfh->hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1772 rndis_dev = ndev->extension;
1774 for (i = 0; i < ndc->rx_table_sz; i++)
1775 rxfh->indir[i] = ndc->rx_table[i];
1779 memcpy(rxfh->key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1784 static int netvsc_set_rxfh(struct net_device *dev,
1785 struct ethtool_rxfh_param *rxfh,
1786 struct netlink_ext_ack *extack)
1788 struct net_device_context *ndc = netdev_priv(dev);
1789 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1790 struct rndis_device *rndis_dev;
1791 u8 *key = rxfh->key;
1797 if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
1798 rxfh->hfunc != ETH_RSS_HASH_TOP)
1801 rndis_dev = ndev->extension;
1803 for (i = 0; i < ndc->rx_table_sz; i++)
1804 if (rxfh->indir[i] >= ndev->num_chn)
1807 for (i = 0; i < ndc->rx_table_sz; i++)
1808 ndc->rx_table[i] = rxfh->indir[i];
1815 key = rndis_dev->rss_key;
1818 return rndis_filter_set_rss_param(rndis_dev, key);
1821 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1822 * It does have pre-allocated receive area which is divided into sections.
1824 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1825 struct ethtool_ringparam *ring)
1829 ring->rx_pending = nvdev->recv_section_cnt;
1830 ring->tx_pending = nvdev->send_section_cnt;
1832 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1833 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1835 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1837 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1838 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1839 / nvdev->send_section_size;
1842 static void netvsc_get_ringparam(struct net_device *ndev,
1843 struct ethtool_ringparam *ring,
1844 struct kernel_ethtool_ringparam *kernel_ring,
1845 struct netlink_ext_ack *extack)
1847 struct net_device_context *ndevctx = netdev_priv(ndev);
1848 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1853 __netvsc_get_ringparam(nvdev, ring);
1856 static int netvsc_set_ringparam(struct net_device *ndev,
1857 struct ethtool_ringparam *ring,
1858 struct kernel_ethtool_ringparam *kernel_ring,
1859 struct netlink_ext_ack *extack)
1861 struct net_device_context *ndevctx = netdev_priv(ndev);
1862 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1863 struct netvsc_device_info *device_info;
1864 struct ethtool_ringparam orig;
1868 if (!nvdev || nvdev->destroy)
1871 memset(&orig, 0, sizeof(orig));
1872 __netvsc_get_ringparam(nvdev, &orig);
1874 new_tx = clamp_t(u32, ring->tx_pending,
1875 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1876 new_rx = clamp_t(u32, ring->rx_pending,
1877 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1879 if (new_tx == orig.tx_pending &&
1880 new_rx == orig.rx_pending)
1881 return 0; /* no change */
1883 device_info = netvsc_devinfo_get(nvdev);
1888 device_info->send_sections = new_tx;
1889 device_info->recv_sections = new_rx;
1891 ret = netvsc_detach(ndev, nvdev);
1895 ret = netvsc_attach(ndev, device_info);
1897 device_info->send_sections = orig.tx_pending;
1898 device_info->recv_sections = orig.rx_pending;
1900 if (netvsc_attach(ndev, device_info))
1901 netdev_err(ndev, "restoring ringparam failed");
1905 netvsc_devinfo_put(device_info);
1909 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1910 netdev_features_t features)
1912 struct net_device_context *ndevctx = netdev_priv(ndev);
1913 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1915 if (!nvdev || nvdev->destroy)
1918 if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1919 features ^= NETIF_F_LRO;
1920 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1926 static int netvsc_set_features(struct net_device *ndev,
1927 netdev_features_t features)
1929 netdev_features_t change = features ^ ndev->features;
1930 struct net_device_context *ndevctx = netdev_priv(ndev);
1931 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1932 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1933 struct ndis_offload_params offloads;
1936 if (!nvdev || nvdev->destroy)
1939 if (!(change & NETIF_F_LRO))
1942 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1944 if (features & NETIF_F_LRO) {
1945 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1946 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1948 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1949 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1952 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1955 features ^= NETIF_F_LRO;
1956 ndev->features = features;
1963 vf_netdev->wanted_features = features;
1964 netdev_update_features(vf_netdev);
1969 static int netvsc_get_regs_len(struct net_device *netdev)
1971 return VRSS_SEND_TAB_SIZE * sizeof(u32);
1974 static void netvsc_get_regs(struct net_device *netdev,
1975 struct ethtool_regs *regs, void *p)
1977 struct net_device_context *ndc = netdev_priv(netdev);
1980 /* increase the version, if buffer format is changed. */
1983 memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1986 static u32 netvsc_get_msglevel(struct net_device *ndev)
1988 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1990 return ndev_ctx->msg_enable;
1993 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1995 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1997 ndev_ctx->msg_enable = val;
2000 static const struct ethtool_ops ethtool_ops = {
2001 .get_drvinfo = netvsc_get_drvinfo,
2002 .get_regs_len = netvsc_get_regs_len,
2003 .get_regs = netvsc_get_regs,
2004 .get_msglevel = netvsc_get_msglevel,
2005 .set_msglevel = netvsc_set_msglevel,
2006 .get_link = ethtool_op_get_link,
2007 .get_ethtool_stats = netvsc_get_ethtool_stats,
2008 .get_sset_count = netvsc_get_sset_count,
2009 .get_strings = netvsc_get_strings,
2010 .get_channels = netvsc_get_channels,
2011 .set_channels = netvsc_set_channels,
2012 .get_ts_info = ethtool_op_get_ts_info,
2013 .get_rxnfc = netvsc_get_rxnfc,
2014 .set_rxnfc = netvsc_set_rxnfc,
2015 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2016 .get_rxfh_indir_size = netvsc_rss_indir_size,
2017 .get_rxfh = netvsc_get_rxfh,
2018 .set_rxfh = netvsc_set_rxfh,
2019 .get_link_ksettings = netvsc_get_link_ksettings,
2020 .set_link_ksettings = netvsc_set_link_ksettings,
2021 .get_ringparam = netvsc_get_ringparam,
2022 .set_ringparam = netvsc_set_ringparam,
2025 static const struct net_device_ops device_ops = {
2026 .ndo_open = netvsc_open,
2027 .ndo_stop = netvsc_close,
2028 .ndo_start_xmit = netvsc_start_xmit,
2029 .ndo_change_rx_flags = netvsc_change_rx_flags,
2030 .ndo_set_rx_mode = netvsc_set_rx_mode,
2031 .ndo_fix_features = netvsc_fix_features,
2032 .ndo_set_features = netvsc_set_features,
2033 .ndo_change_mtu = netvsc_change_mtu,
2034 .ndo_validate_addr = eth_validate_addr,
2035 .ndo_set_mac_address = netvsc_set_mac_addr,
2036 .ndo_select_queue = netvsc_select_queue,
2037 .ndo_get_stats64 = netvsc_get_stats64,
2038 .ndo_bpf = netvsc_bpf,
2039 .ndo_xdp_xmit = netvsc_ndoxdp_xmit,
2043 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2044 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2045 * present send GARP packet to network peers with netif_notify_peers().
2047 static void netvsc_link_change(struct work_struct *w)
2049 struct net_device_context *ndev_ctx =
2050 container_of(w, struct net_device_context, dwork.work);
2051 struct hv_device *device_obj = ndev_ctx->device_ctx;
2052 struct net_device *net = hv_get_drvdata(device_obj);
2053 unsigned long flags, next_reconfig, delay;
2054 struct netvsc_reconfig *event = NULL;
2055 struct netvsc_device *net_device;
2056 struct rndis_device *rdev;
2057 bool reschedule = false;
2059 /* if changes are happening, comeback later */
2060 if (!rtnl_trylock()) {
2061 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2065 net_device = rtnl_dereference(ndev_ctx->nvdev);
2069 rdev = net_device->extension;
2071 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2072 if (time_is_after_jiffies(next_reconfig)) {
2073 /* link_watch only sends one notification with current state
2074 * per second, avoid doing reconfig more frequently. Handle
2077 delay = next_reconfig - jiffies;
2078 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2079 schedule_delayed_work(&ndev_ctx->dwork, delay);
2082 ndev_ctx->last_reconfig = jiffies;
2084 spin_lock_irqsave(&ndev_ctx->lock, flags);
2085 if (!list_empty(&ndev_ctx->reconfig_events)) {
2086 event = list_first_entry(&ndev_ctx->reconfig_events,
2087 struct netvsc_reconfig, list);
2088 list_del(&event->list);
2089 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2091 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2096 switch (event->event) {
2097 /* Only the following events are possible due to the check in
2098 * netvsc_linkstatus_callback()
2100 case RNDIS_STATUS_MEDIA_CONNECT:
2101 if (rdev->link_state) {
2102 rdev->link_state = false;
2103 netif_carrier_on(net);
2104 netvsc_tx_enable(net_device, net);
2106 __netdev_notify_peers(net);
2110 case RNDIS_STATUS_MEDIA_DISCONNECT:
2111 if (!rdev->link_state) {
2112 rdev->link_state = true;
2113 netif_carrier_off(net);
2114 netvsc_tx_disable(net_device, net);
2118 case RNDIS_STATUS_NETWORK_CHANGE:
2119 /* Only makes sense if carrier is present */
2120 if (!rdev->link_state) {
2121 rdev->link_state = true;
2122 netif_carrier_off(net);
2123 netvsc_tx_disable(net_device, net);
2124 event->event = RNDIS_STATUS_MEDIA_CONNECT;
2125 spin_lock_irqsave(&ndev_ctx->lock, flags);
2126 list_add(&event->list, &ndev_ctx->reconfig_events);
2127 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2135 /* link_watch only sends one notification with current state per
2136 * second, handle next reconfig event in 2 seconds.
2139 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2147 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2149 struct net_device_context *net_device_ctx;
2150 struct net_device *dev;
2152 dev = netdev_master_upper_dev_get(vf_netdev);
2153 if (!dev || dev->netdev_ops != &device_ops)
2154 return NULL; /* not a netvsc device */
2156 net_device_ctx = netdev_priv(dev);
2157 if (!rtnl_dereference(net_device_ctx->nvdev))
2158 return NULL; /* device is removed */
2163 /* Called when VF is injecting data into network stack.
2164 * Change the associated network device from VF to netvsc.
2165 * note: already called with rcu_read_lock
2167 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2169 struct sk_buff *skb = *pskb;
2170 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2171 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2172 struct netvsc_vf_pcpu_stats *pcpu_stats
2173 = this_cpu_ptr(ndev_ctx->vf_stats);
2175 skb = skb_share_check(skb, GFP_ATOMIC);
2177 return RX_HANDLER_CONSUMED;
2183 u64_stats_update_begin(&pcpu_stats->syncp);
2184 pcpu_stats->rx_packets++;
2185 pcpu_stats->rx_bytes += skb->len;
2186 u64_stats_update_end(&pcpu_stats->syncp);
2188 return RX_HANDLER_ANOTHER;
2191 static int netvsc_vf_join(struct net_device *vf_netdev,
2192 struct net_device *ndev, int context)
2194 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2197 ret = netdev_rx_handler_register(vf_netdev,
2198 netvsc_vf_handle_frame, ndev);
2200 netdev_err(vf_netdev,
2201 "can not register netvsc VF receive handler (err = %d)\n",
2203 goto rx_handler_failed;
2206 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2209 netdev_err(vf_netdev,
2210 "can not set master device %s (err = %d)\n",
2212 goto upper_link_failed;
2215 /* If this registration is called from probe context vf_takeover
2216 * is taken care of later in probe itself.
2218 if (context == VF_REG_IN_NOTIFIER)
2219 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2221 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2223 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2227 netdev_rx_handler_unregister(vf_netdev);
2232 static void __netvsc_vf_setup(struct net_device *ndev,
2233 struct net_device *vf_netdev)
2237 /* Align MTU of VF with master */
2238 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2240 netdev_warn(vf_netdev,
2241 "unable to change mtu to %u\n", ndev->mtu);
2243 /* set multicast etc flags on VF */
2244 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2246 /* sync address list from ndev to VF */
2247 netif_addr_lock_bh(ndev);
2248 dev_uc_sync(vf_netdev, ndev);
2249 dev_mc_sync(vf_netdev, ndev);
2250 netif_addr_unlock_bh(ndev);
2252 if (netif_running(ndev)) {
2253 ret = dev_open(vf_netdev, NULL);
2255 netdev_warn(vf_netdev,
2256 "unable to open: %d\n", ret);
2260 /* Setup VF as slave of the synthetic device.
2261 * Runs in workqueue to avoid recursion in netlink callbacks.
2263 static void netvsc_vf_setup(struct work_struct *w)
2265 struct net_device_context *ndev_ctx
2266 = container_of(w, struct net_device_context, vf_takeover.work);
2267 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2268 struct net_device *vf_netdev;
2270 if (!rtnl_trylock()) {
2271 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2275 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2277 __netvsc_vf_setup(ndev, vf_netdev);
2282 /* Find netvsc by VF serial number.
2283 * The PCI hyperv controller records the serial number as the slot kobj name.
2285 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2287 struct device *parent = vf_netdev->dev.parent;
2288 struct net_device_context *ndev_ctx;
2289 struct net_device *ndev;
2290 struct pci_dev *pdev;
2293 if (!parent || !dev_is_pci(parent))
2294 return NULL; /* not a PCI device */
2296 pdev = to_pci_dev(parent);
2298 netdev_notice(vf_netdev, "no PCI slot information\n");
2302 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2303 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2304 pci_slot_name(pdev->slot));
2308 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2309 if (!ndev_ctx->vf_alloc)
2312 if (ndev_ctx->vf_serial != serial)
2315 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2316 if (ndev->addr_len != vf_netdev->addr_len ||
2317 memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2318 ndev->addr_len) != 0)
2325 /* Fallback path to check synthetic vf with help of mac addr.
2326 * Because this function can be called before vf_netdev is
2327 * initialized (NETDEV_POST_INIT) when its perm_addr has not been copied
2328 * from dev_addr, also try to match to its dev_addr.
2329 * Note: On Hyper-V and Azure, it's not possible to set a MAC address
2330 * on a VF that matches to the MAC of a unrelated NETVSC device.
2332 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2333 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2334 if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr) ||
2335 ether_addr_equal(vf_netdev->dev_addr, ndev->perm_addr))
2339 netdev_notice(vf_netdev,
2340 "no netdev found for vf serial:%u\n", serial);
2344 static int netvsc_prepare_bonding(struct net_device *vf_netdev)
2346 struct net_device *ndev;
2348 ndev = get_netvsc_byslot(vf_netdev);
2352 /* set slave flag before open to prevent IPv6 addrconf */
2353 vf_netdev->flags |= IFF_SLAVE;
2357 static int netvsc_register_vf(struct net_device *vf_netdev, int context)
2359 struct net_device_context *net_device_ctx;
2360 struct netvsc_device *netvsc_dev;
2361 struct bpf_prog *prog;
2362 struct net_device *ndev;
2365 if (vf_netdev->addr_len != ETH_ALEN)
2368 ndev = get_netvsc_byslot(vf_netdev);
2372 net_device_ctx = netdev_priv(ndev);
2373 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2374 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2377 /* if synthetic interface is a different namespace,
2378 * then move the VF to that namespace; join will be
2379 * done again in that context.
2381 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2382 ret = dev_change_net_namespace(vf_netdev,
2383 dev_net(ndev), "eth%d");
2385 netdev_err(vf_netdev,
2386 "could not move to same namespace as %s: %d\n",
2389 netdev_info(vf_netdev,
2390 "VF moved to namespace with: %s\n",
2395 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2397 if (netvsc_vf_join(vf_netdev, ndev, context) != 0)
2400 dev_hold(vf_netdev);
2401 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2403 if (ndev->needed_headroom < vf_netdev->needed_headroom)
2404 ndev->needed_headroom = vf_netdev->needed_headroom;
2406 vf_netdev->wanted_features = ndev->features;
2407 netdev_update_features(vf_netdev);
2409 prog = netvsc_xdp_get(netvsc_dev);
2410 netvsc_vf_setxdp(vf_netdev, prog);
2415 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2417 * Typically a UP or DOWN event is followed by a CHANGE event, so
2418 * net_device_ctx->data_path_is_vf is used to cache the current data path
2419 * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2422 * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2423 * interface, there is only the CHANGE event and no UP or DOWN event.
2425 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2427 struct net_device_context *net_device_ctx;
2428 struct netvsc_device *netvsc_dev;
2429 struct net_device *ndev;
2430 bool vf_is_up = false;
2433 if (event != NETDEV_GOING_DOWN)
2434 vf_is_up = netif_running(vf_netdev);
2436 ndev = get_netvsc_byref(vf_netdev);
2440 net_device_ctx = netdev_priv(ndev);
2441 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2445 if (net_device_ctx->data_path_is_vf == vf_is_up)
2448 if (vf_is_up && !net_device_ctx->vf_alloc) {
2449 netdev_info(ndev, "Waiting for the VF association from host\n");
2450 wait_for_completion(&net_device_ctx->vf_add);
2453 ret = netvsc_switch_datapath(ndev, vf_is_up);
2457 "Data path failed to switch %s VF: %s, err: %d\n",
2458 vf_is_up ? "to" : "from", vf_netdev->name, ret);
2461 netdev_info(ndev, "Data path switched %s VF: %s\n",
2462 vf_is_up ? "to" : "from", vf_netdev->name);
2468 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2470 struct net_device *ndev;
2471 struct net_device_context *net_device_ctx;
2473 ndev = get_netvsc_byref(vf_netdev);
2477 net_device_ctx = netdev_priv(ndev);
2478 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2480 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2482 netvsc_vf_setxdp(vf_netdev, NULL);
2484 reinit_completion(&net_device_ctx->vf_add);
2485 netdev_rx_handler_unregister(vf_netdev);
2486 netdev_upper_dev_unlink(vf_netdev, ndev);
2487 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2490 ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2495 static int check_dev_is_matching_vf(struct net_device *event_ndev)
2497 /* Skip NetVSC interfaces */
2498 if (event_ndev->netdev_ops == &device_ops)
2501 /* Avoid non-Ethernet type devices */
2502 if (event_ndev->type != ARPHRD_ETHER)
2505 /* Avoid Vlan dev with same MAC registering as VF */
2506 if (is_vlan_dev(event_ndev))
2509 /* Avoid Bonding master dev with same MAC registering as VF */
2510 if (netif_is_bond_master(event_ndev))
2516 static int netvsc_probe(struct hv_device *dev,
2517 const struct hv_vmbus_device_id *dev_id)
2519 struct net_device *net = NULL, *vf_netdev;
2520 struct net_device_context *net_device_ctx;
2521 struct netvsc_device_info *device_info = NULL;
2522 struct netvsc_device *nvdev;
2525 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2530 netif_carrier_off(net);
2532 netvsc_init_settings(net);
2534 net_device_ctx = netdev_priv(net);
2535 net_device_ctx->device_ctx = dev;
2536 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2537 if (netif_msg_probe(net_device_ctx))
2538 netdev_dbg(net, "netvsc msg_enable: %d\n",
2539 net_device_ctx->msg_enable);
2541 hv_set_drvdata(dev, net);
2543 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2545 init_completion(&net_device_ctx->vf_add);
2546 spin_lock_init(&net_device_ctx->lock);
2547 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2548 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2550 net_device_ctx->vf_stats
2551 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2552 if (!net_device_ctx->vf_stats)
2555 net->netdev_ops = &device_ops;
2556 net->ethtool_ops = ðtool_ops;
2557 SET_NETDEV_DEV(net, &dev->device);
2558 dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
2560 /* We always need headroom for rndis header */
2561 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2563 /* Initialize the number of queues to be 1, we may change it if more
2564 * channels are offered later.
2566 netif_set_real_num_tx_queues(net, 1);
2567 netif_set_real_num_rx_queues(net, 1);
2569 /* Notify the netvsc driver of the new device */
2570 device_info = netvsc_devinfo_get(NULL);
2574 goto devinfo_failed;
2577 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2578 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2579 * all subchannels to show up, but that may not happen because
2580 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2581 * -> ... -> device_add() -> ... -> __device_attach() can't get
2582 * the device lock, so all the subchannels can't be processed --
2583 * finally netvsc_subchan_work() hangs forever.
2585 * The rtnl lock also needs to be held before rndis_filter_device_add()
2586 * which advertises nvsp_2_vsc_capability / sriov bit, and triggers
2587 * VF NIC offering and registering. If VF NIC finished register_netdev()
2588 * earlier it may cause name based config failure.
2592 nvdev = rndis_filter_device_add(dev, device_info);
2593 if (IS_ERR(nvdev)) {
2594 ret = PTR_ERR(nvdev);
2595 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2599 eth_hw_addr_set(net, device_info->mac_adr);
2601 if (nvdev->num_chn > 1)
2602 schedule_work(&nvdev->subchan_work);
2604 /* hw_features computed in rndis_netdev_set_hwcaps() */
2605 net->features = net->hw_features |
2606 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2607 NETIF_F_HW_VLAN_CTAG_RX;
2608 net->vlan_features = net->features;
2610 netdev_lockdep_set_classes(net);
2612 net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2613 NETDEV_XDP_ACT_NDO_XMIT;
2615 /* MTU range: 68 - 1500 or 65521 */
2616 net->min_mtu = NETVSC_MTU_MIN;
2617 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2618 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2620 net->max_mtu = ETH_DATA_LEN;
2622 nvdev->tx_disable = false;
2624 ret = register_netdevice(net);
2626 pr_err("Unable to register netdev.\n");
2627 goto register_failed;
2630 list_add(&net_device_ctx->list, &netvsc_dev_list);
2632 /* When the hv_netvsc driver is unloaded and reloaded, the
2633 * NET_DEVICE_REGISTER for the vf device is replayed before probe
2634 * is complete. This is because register_netdevice_notifier() gets
2635 * registered before vmbus_driver_register() so that callback func
2636 * is set before probe and we don't miss events like NETDEV_POST_INIT
2637 * So, in this section we try to register the matching vf device that
2638 * is present as a netdevice, knowing that its register call is not
2639 * processed in the netvsc_netdev_notifier(as probing is progress and
2640 * get_netvsc_byslot fails).
2642 for_each_netdev(dev_net(net), vf_netdev) {
2643 ret = check_dev_is_matching_vf(vf_netdev);
2647 if (net != get_netvsc_byslot(vf_netdev))
2650 netvsc_prepare_bonding(vf_netdev);
2651 netvsc_register_vf(vf_netdev, VF_REG_IN_PROBE);
2652 __netvsc_vf_setup(net, vf_netdev);
2657 netvsc_devinfo_put(device_info);
2661 rndis_filter_device_remove(dev, nvdev);
2664 netvsc_devinfo_put(device_info);
2666 free_percpu(net_device_ctx->vf_stats);
2668 hv_set_drvdata(dev, NULL);
2674 static void netvsc_remove(struct hv_device *dev)
2676 struct net_device_context *ndev_ctx;
2677 struct net_device *vf_netdev, *net;
2678 struct netvsc_device *nvdev;
2680 net = hv_get_drvdata(dev);
2682 dev_err(&dev->device, "No net device to remove\n");
2686 ndev_ctx = netdev_priv(net);
2688 cancel_delayed_work_sync(&ndev_ctx->dwork);
2691 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2693 cancel_work_sync(&nvdev->subchan_work);
2694 netvsc_xdp_set(net, NULL, NULL, nvdev);
2698 * Call to the vsc driver to let it know that the device is being
2699 * removed. Also blocks mtu and channel changes.
2701 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2703 netvsc_unregister_vf(vf_netdev);
2706 rndis_filter_device_remove(dev, nvdev);
2708 unregister_netdevice(net);
2709 list_del(&ndev_ctx->list);
2713 hv_set_drvdata(dev, NULL);
2715 free_percpu(ndev_ctx->vf_stats);
2719 static int netvsc_suspend(struct hv_device *dev)
2721 struct net_device_context *ndev_ctx;
2722 struct netvsc_device *nvdev;
2723 struct net_device *net;
2726 net = hv_get_drvdata(dev);
2728 ndev_ctx = netdev_priv(net);
2729 cancel_delayed_work_sync(&ndev_ctx->dwork);
2733 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2734 if (nvdev == NULL) {
2739 /* Save the current config info */
2740 ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2741 if (!ndev_ctx->saved_netvsc_dev_info) {
2745 ret = netvsc_detach(net, nvdev);
2752 static int netvsc_resume(struct hv_device *dev)
2754 struct net_device *net = hv_get_drvdata(dev);
2755 struct net_device_context *net_device_ctx;
2756 struct netvsc_device_info *device_info;
2761 net_device_ctx = netdev_priv(net);
2763 /* Reset the data path to the netvsc NIC before re-opening the vmbus
2764 * channel. Later netvsc_netdev_event() will switch the data path to
2765 * the VF upon the UP or CHANGE event.
2767 net_device_ctx->data_path_is_vf = false;
2768 device_info = net_device_ctx->saved_netvsc_dev_info;
2770 ret = netvsc_attach(net, device_info);
2772 netvsc_devinfo_put(device_info);
2773 net_device_ctx->saved_netvsc_dev_info = NULL;
2779 static const struct hv_vmbus_device_id id_table[] = {
2785 MODULE_DEVICE_TABLE(vmbus, id_table);
2787 /* The one and only one */
2788 static struct hv_driver netvsc_drv = {
2789 .name = KBUILD_MODNAME,
2790 .id_table = id_table,
2791 .probe = netvsc_probe,
2792 .remove = netvsc_remove,
2793 .suspend = netvsc_suspend,
2794 .resume = netvsc_resume,
2796 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2801 * On Hyper-V, every VF interface is matched with a corresponding
2802 * synthetic interface. The synthetic interface is presented first
2803 * to the guest. When the corresponding VF instance is registered,
2804 * we will take care of switching the data path.
2806 static int netvsc_netdev_event(struct notifier_block *this,
2807 unsigned long event, void *ptr)
2809 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2812 ret = check_dev_is_matching_vf(event_dev);
2817 case NETDEV_POST_INIT:
2818 return netvsc_prepare_bonding(event_dev);
2819 case NETDEV_REGISTER:
2820 return netvsc_register_vf(event_dev, VF_REG_IN_NOTIFIER);
2821 case NETDEV_UNREGISTER:
2822 return netvsc_unregister_vf(event_dev);
2826 case NETDEV_GOING_DOWN:
2827 return netvsc_vf_changed(event_dev, event);
2833 static struct notifier_block netvsc_netdev_notifier = {
2834 .notifier_call = netvsc_netdev_event,
2837 static void __exit netvsc_drv_exit(void)
2839 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2840 vmbus_driver_unregister(&netvsc_drv);
2843 static int __init netvsc_drv_init(void)
2847 if (ring_size < RING_SIZE_MIN) {
2848 ring_size = RING_SIZE_MIN;
2849 pr_info("Increased ring_size to %u (min allowed)\n",
2852 netvsc_ring_bytes = VMBUS_RING_SIZE(ring_size * 4096);
2854 register_netdevice_notifier(&netvsc_netdev_notifier);
2856 ret = vmbus_driver_register(&netvsc_drv);
2863 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2867 MODULE_LICENSE("GPL");
2868 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2870 module_init(netvsc_drv_init);
2871 module_exit(netvsc_drv_exit);