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
41 #define RETRY_US_LO 5000
42 #define RETRY_US_HI 10000
43 #define RETRY_MAX 2000 /* >10 sec */
45 #define LINKCHANGE_INT (2 * HZ)
46 #define VF_TAKEOVER_INT (HZ / 10)
48 static unsigned int ring_size __ro_after_init = 128;
49 module_param(ring_size, uint, 0444);
50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
51 unsigned int netvsc_ring_bytes __ro_after_init;
53 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
54 NETIF_MSG_LINK | NETIF_MSG_IFUP |
55 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
58 static int debug = -1;
59 module_param(debug, int, 0444);
60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62 static LIST_HEAD(netvsc_dev_list);
64 static void netvsc_change_rx_flags(struct net_device *net, int change)
66 struct net_device_context *ndev_ctx = netdev_priv(net);
67 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
73 if (change & IFF_PROMISC) {
74 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
75 dev_set_promiscuity(vf_netdev, inc);
78 if (change & IFF_ALLMULTI) {
79 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
80 dev_set_allmulti(vf_netdev, inc);
84 static void netvsc_set_rx_mode(struct net_device *net)
86 struct net_device_context *ndev_ctx = netdev_priv(net);
87 struct net_device *vf_netdev;
88 struct netvsc_device *nvdev;
91 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
93 dev_uc_sync(vf_netdev, net);
94 dev_mc_sync(vf_netdev, net);
97 nvdev = rcu_dereference(ndev_ctx->nvdev);
99 rndis_filter_update(nvdev);
103 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
104 struct net_device *ndev)
106 nvscdev->tx_disable = false;
107 virt_wmb(); /* ensure queue wake up mechanism is on */
109 netif_tx_wake_all_queues(ndev);
112 static int netvsc_open(struct net_device *net)
114 struct net_device_context *ndev_ctx = netdev_priv(net);
115 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
116 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
117 struct rndis_device *rdev;
120 netif_carrier_off(net);
122 /* Open up the device */
123 ret = rndis_filter_open(nvdev);
125 netdev_err(net, "unable to open device (ret %d).\n", ret);
129 rdev = nvdev->extension;
130 if (!rdev->link_state) {
131 netif_carrier_on(net);
132 netvsc_tx_enable(nvdev, net);
136 /* Setting synthetic device up transparently sets
137 * slave as up. If open fails, then slave will be
138 * still be offline (and not used).
140 ret = dev_open(vf_netdev, NULL);
143 "unable to open slave: %s: %d\n",
144 vf_netdev->name, ret);
149 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
151 unsigned int retry = 0;
154 /* Ensure pending bytes in ring are read */
158 for (i = 0; i < nvdev->num_chn; i++) {
159 struct vmbus_channel *chn
160 = nvdev->chan_table[i].channel;
165 /* make sure receive not running now */
166 napi_synchronize(&nvdev->chan_table[i].napi);
168 aread = hv_get_bytes_to_read(&chn->inbound);
172 aread = hv_get_bytes_to_read(&chn->outbound);
180 if (++retry > RETRY_MAX)
183 usleep_range(RETRY_US_LO, RETRY_US_HI);
187 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
188 struct net_device *ndev)
191 nvscdev->tx_disable = true;
192 virt_wmb(); /* ensure txq will not wake up after stop */
195 netif_tx_disable(ndev);
198 static int netvsc_close(struct net_device *net)
200 struct net_device_context *net_device_ctx = netdev_priv(net);
201 struct net_device *vf_netdev
202 = rtnl_dereference(net_device_ctx->vf_netdev);
203 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
206 netvsc_tx_disable(nvdev, net);
208 /* No need to close rndis filter if it is removed already */
212 ret = rndis_filter_close(nvdev);
214 netdev_err(net, "unable to close device (ret %d).\n", ret);
218 ret = netvsc_wait_until_empty(nvdev);
220 netdev_err(net, "Ring buffer not empty after closing rndis\n");
223 dev_close(vf_netdev);
228 static inline void *init_ppi_data(struct rndis_message *msg,
229 u32 ppi_size, u32 pkt_type)
231 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
232 struct rndis_per_packet_info *ppi;
234 rndis_pkt->data_offset += ppi_size;
235 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
236 + rndis_pkt->per_pkt_info_len;
238 ppi->size = ppi_size;
239 ppi->type = pkt_type;
241 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
243 rndis_pkt->per_pkt_info_len += ppi_size;
248 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
249 * packets. We can use ethtool to change UDP hash level when necessary.
251 static inline u32 netvsc_get_hash(
253 const struct net_device_context *ndc)
255 struct flow_keys flow;
256 u32 hash, pkt_proto = 0;
257 static u32 hashrnd __read_mostly;
259 net_get_random_once(&hashrnd, sizeof(hashrnd));
261 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
264 switch (flow.basic.ip_proto) {
266 if (flow.basic.n_proto == htons(ETH_P_IP))
267 pkt_proto = HV_TCP4_L4HASH;
268 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
269 pkt_proto = HV_TCP6_L4HASH;
274 if (flow.basic.n_proto == htons(ETH_P_IP))
275 pkt_proto = HV_UDP4_L4HASH;
276 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
277 pkt_proto = HV_UDP6_L4HASH;
282 if (pkt_proto & ndc->l4_hash) {
283 return skb_get_hash(skb);
285 if (flow.basic.n_proto == htons(ETH_P_IP))
286 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
287 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
288 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
292 __skb_set_sw_hash(skb, hash, false);
298 static inline int netvsc_get_tx_queue(struct net_device *ndev,
299 struct sk_buff *skb, int old_idx)
301 const struct net_device_context *ndc = netdev_priv(ndev);
302 struct sock *sk = skb->sk;
305 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
306 (VRSS_SEND_TAB_SIZE - 1)];
308 /* If queue index changed record the new value */
309 if (q_idx != old_idx &&
310 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
311 sk_tx_queue_set(sk, q_idx);
317 * Select queue for transmit.
319 * If a valid queue has already been assigned, then use that.
320 * Otherwise compute tx queue based on hash and the send table.
322 * This is basically similar to default (netdev_pick_tx) with the added step
323 * of using the host send_table when no other queue has been assigned.
325 * TODO support XPS - but get_xps_queue not exported
327 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
329 int q_idx = sk_tx_queue_get(skb->sk);
331 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
332 /* If forwarding a packet, we use the recorded queue when
333 * available for better cache locality.
335 if (skb_rx_queue_recorded(skb))
336 q_idx = skb_get_rx_queue(skb);
338 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
344 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
345 struct net_device *sb_dev)
347 struct net_device_context *ndc = netdev_priv(ndev);
348 struct net_device *vf_netdev;
352 vf_netdev = rcu_dereference(ndc->vf_netdev);
354 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
356 if (vf_ops->ndo_select_queue)
357 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
359 txq = netdev_pick_tx(vf_netdev, skb, NULL);
361 /* Record the queue selected by VF so that it can be
362 * used for common case where VF has more queues than
363 * the synthetic device.
365 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
367 txq = netvsc_pick_tx(ndev, skb);
371 while (txq >= ndev->real_num_tx_queues)
372 txq -= ndev->real_num_tx_queues;
377 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
378 struct hv_page_buffer *pb)
382 hvpfn += offset >> HV_HYP_PAGE_SHIFT;
383 offset = offset & ~HV_HYP_PAGE_MASK;
388 bytes = HV_HYP_PAGE_SIZE - offset;
392 pb[j].offset = offset;
398 if (offset == HV_HYP_PAGE_SIZE && len) {
408 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
409 struct hv_netvsc_packet *packet,
410 struct hv_page_buffer *pb)
413 char *data = skb->data;
414 int frags = skb_shinfo(skb)->nr_frags;
417 /* The packet is laid out thus:
418 * 1. hdr: RNDIS header and PPI
420 * 3. skb fragment data
422 slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
423 offset_in_hvpage(hdr),
427 packet->rmsg_size = len;
428 packet->rmsg_pgcnt = slots_used;
430 slots_used += fill_pg_buf(virt_to_hvpfn(data),
431 offset_in_hvpage(data),
435 for (i = 0; i < frags; i++) {
436 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
438 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
446 static int count_skb_frag_slots(struct sk_buff *skb)
448 int i, frags = skb_shinfo(skb)->nr_frags;
451 for (i = 0; i < frags; i++) {
452 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
453 unsigned long size = skb_frag_size(frag);
454 unsigned long offset = skb_frag_off(frag);
456 /* Skip unused frames from start of page */
457 offset &= ~HV_HYP_PAGE_MASK;
458 pages += HVPFN_UP(offset + size);
463 static int netvsc_get_slots(struct sk_buff *skb)
465 char *data = skb->data;
466 unsigned int offset = offset_in_hvpage(data);
467 unsigned int len = skb_headlen(skb);
471 slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
472 frag_slots = count_skb_frag_slots(skb);
473 return slots + frag_slots;
476 static u32 net_checksum_info(struct sk_buff *skb)
478 if (skb->protocol == htons(ETH_P_IP)) {
479 struct iphdr *ip = ip_hdr(skb);
481 if (ip->protocol == IPPROTO_TCP)
482 return TRANSPORT_INFO_IPV4_TCP;
483 else if (ip->protocol == IPPROTO_UDP)
484 return TRANSPORT_INFO_IPV4_UDP;
486 struct ipv6hdr *ip6 = ipv6_hdr(skb);
488 if (ip6->nexthdr == IPPROTO_TCP)
489 return TRANSPORT_INFO_IPV6_TCP;
490 else if (ip6->nexthdr == IPPROTO_UDP)
491 return TRANSPORT_INFO_IPV6_UDP;
494 return TRANSPORT_INFO_NOT_IP;
497 /* Send skb on the slave VF device. */
498 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
501 struct net_device_context *ndev_ctx = netdev_priv(net);
502 unsigned int len = skb->len;
505 skb->dev = vf_netdev;
506 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
508 rc = dev_queue_xmit(skb);
509 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
510 struct netvsc_vf_pcpu_stats *pcpu_stats
511 = this_cpu_ptr(ndev_ctx->vf_stats);
513 u64_stats_update_begin(&pcpu_stats->syncp);
514 pcpu_stats->tx_packets++;
515 pcpu_stats->tx_bytes += len;
516 u64_stats_update_end(&pcpu_stats->syncp);
518 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
524 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
526 struct net_device_context *net_device_ctx = netdev_priv(net);
527 struct hv_netvsc_packet *packet = NULL;
529 unsigned int num_data_pgs;
530 struct rndis_message *rndis_msg;
531 struct net_device *vf_netdev;
534 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
536 /* If VF is present and up then redirect packets to it.
537 * Skip the VF if it is marked down or has no carrier.
538 * If netpoll is in uses, then VF can not be used either.
540 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
541 if (vf_netdev && netif_running(vf_netdev) &&
542 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
543 net_device_ctx->data_path_is_vf)
544 return netvsc_vf_xmit(net, vf_netdev, skb);
546 /* We will atmost need two pages to describe the rndis
547 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
548 * of pages in a single packet. If skb is scattered around
549 * more pages we try linearizing it.
552 num_data_pgs = netvsc_get_slots(skb) + 2;
554 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
555 ++net_device_ctx->eth_stats.tx_scattered;
557 if (skb_linearize(skb))
560 num_data_pgs = netvsc_get_slots(skb) + 2;
561 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
562 ++net_device_ctx->eth_stats.tx_too_big;
568 * Place the rndis header in the skb head room and
569 * the skb->cb will be used for hv_netvsc_packet
572 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
576 /* Use the skb control buffer for building up the packet */
577 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
578 sizeof_field(struct sk_buff, cb));
579 packet = (struct hv_netvsc_packet *)skb->cb;
581 packet->q_idx = skb_get_queue_mapping(skb);
583 packet->total_data_buflen = skb->len;
584 packet->total_bytes = skb->len;
585 packet->total_packets = 1;
587 rndis_msg = (struct rndis_message *)skb->head;
589 /* Add the rndis header */
590 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
591 rndis_msg->msg_len = packet->total_data_buflen;
593 rndis_msg->msg.pkt = (struct rndis_packet) {
594 .data_offset = sizeof(struct rndis_packet),
595 .data_len = packet->total_data_buflen,
596 .per_pkt_info_offset = sizeof(struct rndis_packet),
599 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
601 hash = skb_get_hash_raw(skb);
602 if (hash != 0 && net->real_num_tx_queues > 1) {
605 rndis_msg_size += NDIS_HASH_PPI_SIZE;
606 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
611 /* When using AF_PACKET we need to drop VLAN header from
612 * the frame and update the SKB to allow the HOST OS
613 * to transmit the 802.1Q packet
615 if (skb->protocol == htons(ETH_P_8021Q)) {
618 skb_reset_mac_header(skb);
619 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
620 if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
621 ++net_device_ctx->eth_stats.vlan_error;
625 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
626 /* Update the NDIS header pkt lengths */
627 packet->total_data_buflen -= VLAN_HLEN;
628 packet->total_bytes -= VLAN_HLEN;
629 rndis_msg->msg_len = packet->total_data_buflen;
630 rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
634 if (skb_vlan_tag_present(skb)) {
635 struct ndis_pkt_8021q_info *vlan;
637 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
638 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
642 vlan->vlanid = skb_vlan_tag_get_id(skb);
643 vlan->cfi = skb_vlan_tag_get_cfi(skb);
644 vlan->pri = skb_vlan_tag_get_prio(skb);
647 if (skb_is_gso(skb)) {
648 struct ndis_tcp_lso_info *lso_info;
650 rndis_msg_size += NDIS_LSO_PPI_SIZE;
651 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
652 TCP_LARGESEND_PKTINFO);
655 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
656 if (skb->protocol == htons(ETH_P_IP)) {
657 lso_info->lso_v2_transmit.ip_version =
658 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
659 ip_hdr(skb)->tot_len = 0;
660 ip_hdr(skb)->check = 0;
661 tcp_hdr(skb)->check =
662 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
663 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
665 lso_info->lso_v2_transmit.ip_version =
666 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
667 tcp_v6_gso_csum_prep(skb);
669 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
670 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
671 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
672 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
673 struct ndis_tcp_ip_checksum_info *csum_info;
675 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
676 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
677 TCPIP_CHKSUM_PKTINFO);
679 csum_info->value = 0;
680 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
682 if (skb->protocol == htons(ETH_P_IP)) {
683 csum_info->transmit.is_ipv4 = 1;
685 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
686 csum_info->transmit.tcp_checksum = 1;
688 csum_info->transmit.udp_checksum = 1;
690 csum_info->transmit.is_ipv6 = 1;
692 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
693 csum_info->transmit.tcp_checksum = 1;
695 csum_info->transmit.udp_checksum = 1;
698 /* Can't do offload of this type of checksum */
699 if (skb_checksum_help(skb))
704 /* Start filling in the page buffers with the rndis hdr */
705 rndis_msg->msg_len += rndis_msg_size;
706 packet->total_data_buflen = rndis_msg->msg_len;
707 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
710 /* timestamp packet in software */
711 skb_tx_timestamp(skb);
713 ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
714 if (likely(ret == 0))
717 if (ret == -EAGAIN) {
718 ++net_device_ctx->eth_stats.tx_busy;
719 return NETDEV_TX_BUSY;
723 ++net_device_ctx->eth_stats.tx_no_space;
726 dev_kfree_skb_any(skb);
727 net->stats.tx_dropped++;
732 ++net_device_ctx->eth_stats.tx_no_memory;
736 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
737 struct net_device *ndev)
739 return netvsc_xmit(skb, ndev, false);
743 * netvsc_linkstatus_callback - Link up/down notification
745 void netvsc_linkstatus_callback(struct net_device *net,
746 struct rndis_message *resp,
747 void *data, u32 data_buflen)
749 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
750 struct net_device_context *ndev_ctx = netdev_priv(net);
751 struct netvsc_reconfig *event;
754 /* Ensure the packet is big enough to access its fields */
755 if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
756 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
761 /* Copy the RNDIS indicate status into nvchan->recv_buf */
762 memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
764 /* Update the physical link speed when changing to another vSwitch */
765 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
768 /* Validate status_buf_offset and status_buflen.
770 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
771 * for the status buffer field in resp->msg_len; perform the validation
772 * using data_buflen (>= resp->msg_len).
774 if (indicate->status_buflen < sizeof(speed) ||
775 indicate->status_buf_offset < sizeof(*indicate) ||
776 data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
777 data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
778 < indicate->status_buflen) {
779 netdev_err(net, "invalid rndis_indicate_status packet\n");
783 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
784 ndev_ctx->speed = speed;
788 /* Handle these link change statuses below */
789 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
790 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
791 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
794 if (net->reg_state != NETREG_REGISTERED)
797 event = kzalloc(sizeof(*event), GFP_ATOMIC);
800 event->event = indicate->status;
802 spin_lock_irqsave(&ndev_ctx->lock, flags);
803 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
804 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
806 schedule_delayed_work(&ndev_ctx->dwork, 0);
809 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
813 skb->queue_mapping = skb_get_rx_queue(skb);
814 __skb_push(skb, ETH_HLEN);
816 rc = netvsc_xmit(skb, ndev, true);
818 if (dev_xmit_complete(rc))
821 dev_kfree_skb_any(skb);
822 ndev->stats.tx_dropped++;
825 static void netvsc_comp_ipcsum(struct sk_buff *skb)
827 struct iphdr *iph = (struct iphdr *)skb->data;
830 iph->check = ip_fast_csum(iph, iph->ihl);
833 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
834 struct netvsc_channel *nvchan,
835 struct xdp_buff *xdp)
837 struct napi_struct *napi = &nvchan->napi;
838 const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
839 const struct ndis_tcp_ip_checksum_info *csum_info =
840 &nvchan->rsc.csum_info;
841 const u32 *hash_info = &nvchan->rsc.hash_info;
842 u8 ppi_flags = nvchan->rsc.ppi_flags;
844 void *xbuf = xdp->data_hard_start;
848 unsigned int hdroom = xdp->data - xdp->data_hard_start;
849 unsigned int xlen = xdp->data_end - xdp->data;
850 unsigned int frag_size = xdp->frame_sz;
852 skb = build_skb(xbuf, frag_size);
855 __free_page(virt_to_page(xbuf));
859 skb_reserve(skb, hdroom);
861 skb->dev = napi->dev;
863 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
868 /* Copy to skb. This copy is needed here since the memory
869 * pointed by hv_netvsc_packet cannot be deallocated.
871 for (i = 0; i < nvchan->rsc.cnt; i++)
872 skb_put_data(skb, nvchan->rsc.data[i],
876 skb->protocol = eth_type_trans(skb, net);
878 /* skb is already created with CHECKSUM_NONE */
879 skb_checksum_none_assert(skb);
881 /* Incoming packets may have IP header checksum verified by the host.
882 * They may not have IP header checksum computed after coalescing.
883 * We compute it here if the flags are set, because on Linux, the IP
884 * checksum is always checked.
886 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
887 csum_info->receive.ip_checksum_succeeded &&
888 skb->protocol == htons(ETH_P_IP)) {
889 /* Check that there is enough space to hold the IP header. */
890 if (skb_headlen(skb) < sizeof(struct iphdr)) {
894 netvsc_comp_ipcsum(skb);
897 /* Do L4 checksum offload if enabled and present. */
898 if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
899 if (csum_info->receive.tcp_checksum_succeeded ||
900 csum_info->receive.udp_checksum_succeeded)
901 skb->ip_summed = CHECKSUM_UNNECESSARY;
904 if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
905 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
907 if (ppi_flags & NVSC_RSC_VLAN) {
908 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
909 (vlan->cfi ? VLAN_CFI_MASK : 0);
911 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
919 * netvsc_recv_callback - Callback when we receive a packet from the
920 * "wire" on the specified device.
922 int netvsc_recv_callback(struct net_device *net,
923 struct netvsc_device *net_device,
924 struct netvsc_channel *nvchan)
926 struct net_device_context *net_device_ctx = netdev_priv(net);
927 struct vmbus_channel *channel = nvchan->channel;
928 u16 q_idx = channel->offermsg.offer.sub_channel_index;
930 struct netvsc_stats *rx_stats = &nvchan->rx_stats;
934 if (net->reg_state != NETREG_REGISTERED)
935 return NVSP_STAT_FAIL;
937 act = netvsc_run_xdp(net, nvchan, &xdp);
939 if (act != XDP_PASS && act != XDP_TX) {
940 u64_stats_update_begin(&rx_stats->syncp);
941 rx_stats->xdp_drop++;
942 u64_stats_update_end(&rx_stats->syncp);
944 return NVSP_STAT_SUCCESS; /* consumed by XDP */
947 /* Allocate a skb - TODO direct I/O to pages? */
948 skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
950 if (unlikely(!skb)) {
951 ++net_device_ctx->eth_stats.rx_no_memory;
952 return NVSP_STAT_FAIL;
955 skb_record_rx_queue(skb, q_idx);
958 * Even if injecting the packet, record the statistics
959 * on the synthetic device because modifying the VF device
960 * statistics will not work correctly.
962 u64_stats_update_begin(&rx_stats->syncp);
964 rx_stats->bytes += nvchan->rsc.pktlen;
966 if (skb->pkt_type == PACKET_BROADCAST)
967 ++rx_stats->broadcast;
968 else if (skb->pkt_type == PACKET_MULTICAST)
969 ++rx_stats->multicast;
970 u64_stats_update_end(&rx_stats->syncp);
973 netvsc_xdp_xmit(skb, net);
974 return NVSP_STAT_SUCCESS;
977 napi_gro_receive(&nvchan->napi, skb);
978 return NVSP_STAT_SUCCESS;
981 static void netvsc_get_drvinfo(struct net_device *net,
982 struct ethtool_drvinfo *info)
984 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
985 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
988 static void netvsc_get_channels(struct net_device *net,
989 struct ethtool_channels *channel)
991 struct net_device_context *net_device_ctx = netdev_priv(net);
992 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
995 channel->max_combined = nvdev->max_chn;
996 channel->combined_count = nvdev->num_chn;
1000 /* Alloc struct netvsc_device_info, and initialize it from either existing
1001 * struct netvsc_device, or from default values.
1004 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
1006 struct netvsc_device_info *dev_info;
1007 struct bpf_prog *prog;
1009 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1017 dev_info->num_chn = nvdev->num_chn;
1018 dev_info->send_sections = nvdev->send_section_cnt;
1019 dev_info->send_section_size = nvdev->send_section_size;
1020 dev_info->recv_sections = nvdev->recv_section_cnt;
1021 dev_info->recv_section_size = nvdev->recv_section_size;
1023 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1024 NETVSC_HASH_KEYLEN);
1026 prog = netvsc_xdp_get(nvdev);
1029 dev_info->bprog = prog;
1032 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1033 dev_info->send_sections = NETVSC_DEFAULT_TX;
1034 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1035 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1036 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1042 /* Free struct netvsc_device_info */
1043 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1045 if (dev_info->bprog) {
1047 bpf_prog_put(dev_info->bprog);
1053 static int netvsc_detach(struct net_device *ndev,
1054 struct netvsc_device *nvdev)
1056 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1057 struct hv_device *hdev = ndev_ctx->device_ctx;
1060 /* Don't try continuing to try and setup sub channels */
1061 if (cancel_work_sync(&nvdev->subchan_work))
1064 netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1066 /* If device was up (receiving) then shutdown */
1067 if (netif_running(ndev)) {
1068 netvsc_tx_disable(nvdev, ndev);
1070 ret = rndis_filter_close(nvdev);
1073 "unable to close device (ret %d).\n", ret);
1077 ret = netvsc_wait_until_empty(nvdev);
1080 "Ring buffer not empty after closing rndis\n");
1085 netif_device_detach(ndev);
1087 rndis_filter_device_remove(hdev, nvdev);
1092 static int netvsc_attach(struct net_device *ndev,
1093 struct netvsc_device_info *dev_info)
1095 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1096 struct hv_device *hdev = ndev_ctx->device_ctx;
1097 struct netvsc_device *nvdev;
1098 struct rndis_device *rdev;
1099 struct bpf_prog *prog;
1102 nvdev = rndis_filter_device_add(hdev, dev_info);
1104 return PTR_ERR(nvdev);
1106 if (nvdev->num_chn > 1) {
1107 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1109 /* if unavailable, just proceed with one queue */
1116 prog = dev_info->bprog;
1119 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1126 /* In any case device is now ready */
1127 nvdev->tx_disable = false;
1128 netif_device_attach(ndev);
1130 /* Note: enable and attach happen when sub-channels setup */
1131 netif_carrier_off(ndev);
1133 if (netif_running(ndev)) {
1134 ret = rndis_filter_open(nvdev);
1138 rdev = nvdev->extension;
1139 if (!rdev->link_state)
1140 netif_carrier_on(ndev);
1146 netif_device_detach(ndev);
1149 rndis_filter_device_remove(hdev, nvdev);
1154 static int netvsc_set_channels(struct net_device *net,
1155 struct ethtool_channels *channels)
1157 struct net_device_context *net_device_ctx = netdev_priv(net);
1158 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1159 unsigned int orig, count = channels->combined_count;
1160 struct netvsc_device_info *device_info;
1163 /* We do not support separate count for rx, tx, or other */
1165 channels->rx_count || channels->tx_count || channels->other_count)
1168 if (!nvdev || nvdev->destroy)
1171 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1174 if (count > nvdev->max_chn)
1177 orig = nvdev->num_chn;
1179 device_info = netvsc_devinfo_get(nvdev);
1184 device_info->num_chn = count;
1186 ret = netvsc_detach(net, nvdev);
1190 ret = netvsc_attach(net, device_info);
1192 device_info->num_chn = orig;
1193 if (netvsc_attach(net, device_info))
1194 netdev_err(net, "restoring channel setting failed\n");
1198 netvsc_devinfo_put(device_info);
1202 static void netvsc_init_settings(struct net_device *dev)
1204 struct net_device_context *ndc = netdev_priv(dev);
1206 ndc->l4_hash = HV_DEFAULT_L4HASH;
1208 ndc->speed = SPEED_UNKNOWN;
1209 ndc->duplex = DUPLEX_FULL;
1211 dev->features = NETIF_F_LRO;
1214 static int netvsc_get_link_ksettings(struct net_device *dev,
1215 struct ethtool_link_ksettings *cmd)
1217 struct net_device_context *ndc = netdev_priv(dev);
1218 struct net_device *vf_netdev;
1220 vf_netdev = rtnl_dereference(ndc->vf_netdev);
1223 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1225 cmd->base.speed = ndc->speed;
1226 cmd->base.duplex = ndc->duplex;
1227 cmd->base.port = PORT_OTHER;
1232 static int netvsc_set_link_ksettings(struct net_device *dev,
1233 const struct ethtool_link_ksettings *cmd)
1235 struct net_device_context *ndc = netdev_priv(dev);
1236 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1239 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1242 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1246 return ethtool_virtdev_set_link_ksettings(dev, cmd,
1247 &ndc->speed, &ndc->duplex);
1250 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1252 struct net_device_context *ndevctx = netdev_priv(ndev);
1253 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1254 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1255 int orig_mtu = ndev->mtu;
1256 struct netvsc_device_info *device_info;
1259 if (!nvdev || nvdev->destroy)
1262 device_info = netvsc_devinfo_get(nvdev);
1267 /* Change MTU of underlying VF netdev first. */
1269 ret = dev_set_mtu(vf_netdev, mtu);
1274 ret = netvsc_detach(ndev, nvdev);
1280 ret = netvsc_attach(ndev, device_info);
1284 /* Attempt rollback to original MTU */
1285 ndev->mtu = orig_mtu;
1287 if (netvsc_attach(ndev, device_info))
1288 netdev_err(ndev, "restoring mtu failed\n");
1291 dev_set_mtu(vf_netdev, orig_mtu);
1294 netvsc_devinfo_put(device_info);
1298 static void netvsc_get_vf_stats(struct net_device *net,
1299 struct netvsc_vf_pcpu_stats *tot)
1301 struct net_device_context *ndev_ctx = netdev_priv(net);
1304 memset(tot, 0, sizeof(*tot));
1306 for_each_possible_cpu(i) {
1307 const struct netvsc_vf_pcpu_stats *stats
1308 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1309 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1313 start = u64_stats_fetch_begin_irq(&stats->syncp);
1314 rx_packets = stats->rx_packets;
1315 tx_packets = stats->tx_packets;
1316 rx_bytes = stats->rx_bytes;
1317 tx_bytes = stats->tx_bytes;
1318 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1320 tot->rx_packets += rx_packets;
1321 tot->tx_packets += tx_packets;
1322 tot->rx_bytes += rx_bytes;
1323 tot->tx_bytes += tx_bytes;
1324 tot->tx_dropped += stats->tx_dropped;
1328 static void netvsc_get_pcpu_stats(struct net_device *net,
1329 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1331 struct net_device_context *ndev_ctx = netdev_priv(net);
1332 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1335 /* fetch percpu stats of vf */
1336 for_each_possible_cpu(i) {
1337 const struct netvsc_vf_pcpu_stats *stats =
1338 per_cpu_ptr(ndev_ctx->vf_stats, i);
1339 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1343 start = u64_stats_fetch_begin_irq(&stats->syncp);
1344 this_tot->vf_rx_packets = stats->rx_packets;
1345 this_tot->vf_tx_packets = stats->tx_packets;
1346 this_tot->vf_rx_bytes = stats->rx_bytes;
1347 this_tot->vf_tx_bytes = stats->tx_bytes;
1348 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1349 this_tot->rx_packets = this_tot->vf_rx_packets;
1350 this_tot->tx_packets = this_tot->vf_tx_packets;
1351 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1352 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1355 /* fetch percpu stats of netvsc */
1356 for (i = 0; i < nvdev->num_chn; i++) {
1357 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1358 const struct netvsc_stats *stats;
1359 struct netvsc_ethtool_pcpu_stats *this_tot =
1360 &pcpu_tot[nvchan->channel->target_cpu];
1364 stats = &nvchan->tx_stats;
1366 start = u64_stats_fetch_begin_irq(&stats->syncp);
1367 packets = stats->packets;
1368 bytes = stats->bytes;
1369 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1371 this_tot->tx_bytes += bytes;
1372 this_tot->tx_packets += packets;
1374 stats = &nvchan->rx_stats;
1376 start = u64_stats_fetch_begin_irq(&stats->syncp);
1377 packets = stats->packets;
1378 bytes = stats->bytes;
1379 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1381 this_tot->rx_bytes += bytes;
1382 this_tot->rx_packets += packets;
1386 static void netvsc_get_stats64(struct net_device *net,
1387 struct rtnl_link_stats64 *t)
1389 struct net_device_context *ndev_ctx = netdev_priv(net);
1390 struct netvsc_device *nvdev;
1391 struct netvsc_vf_pcpu_stats vf_tot;
1396 nvdev = rcu_dereference(ndev_ctx->nvdev);
1400 netdev_stats_to_stats64(t, &net->stats);
1402 netvsc_get_vf_stats(net, &vf_tot);
1403 t->rx_packets += vf_tot.rx_packets;
1404 t->tx_packets += vf_tot.tx_packets;
1405 t->rx_bytes += vf_tot.rx_bytes;
1406 t->tx_bytes += vf_tot.tx_bytes;
1407 t->tx_dropped += vf_tot.tx_dropped;
1409 for (i = 0; i < nvdev->num_chn; i++) {
1410 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1411 const struct netvsc_stats *stats;
1412 u64 packets, bytes, multicast;
1415 stats = &nvchan->tx_stats;
1417 start = u64_stats_fetch_begin_irq(&stats->syncp);
1418 packets = stats->packets;
1419 bytes = stats->bytes;
1420 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1422 t->tx_bytes += bytes;
1423 t->tx_packets += packets;
1425 stats = &nvchan->rx_stats;
1427 start = u64_stats_fetch_begin_irq(&stats->syncp);
1428 packets = stats->packets;
1429 bytes = stats->bytes;
1430 multicast = stats->multicast + stats->broadcast;
1431 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1433 t->rx_bytes += bytes;
1434 t->rx_packets += packets;
1435 t->multicast += multicast;
1441 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1443 struct net_device_context *ndc = netdev_priv(ndev);
1444 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1445 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1446 struct sockaddr *addr = p;
1449 err = eth_prepare_mac_addr_change(ndev, p);
1457 err = dev_set_mac_address(vf_netdev, addr, NULL);
1462 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1464 eth_commit_mac_addr_change(ndev, p);
1465 } else if (vf_netdev) {
1466 /* rollback change on VF */
1467 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1468 dev_set_mac_address(vf_netdev, addr, NULL);
1474 static const struct {
1475 char name[ETH_GSTRING_LEN];
1477 } netvsc_stats[] = {
1478 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1479 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1480 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1481 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1482 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1483 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1484 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1485 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1486 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1487 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1488 { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1490 { "cpu%u_rx_packets",
1491 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1493 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1494 { "cpu%u_tx_packets",
1495 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1497 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1498 { "cpu%u_vf_rx_packets",
1499 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1500 { "cpu%u_vf_rx_bytes",
1501 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1502 { "cpu%u_vf_tx_packets",
1503 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1504 { "cpu%u_vf_tx_bytes",
1505 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1507 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1508 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1509 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1510 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1511 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1514 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1515 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1517 /* statistics per queue (rx/tx packets/bytes) */
1518 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1520 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1521 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1523 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1525 struct net_device_context *ndc = netdev_priv(dev);
1526 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1531 switch (string_set) {
1533 return NETVSC_GLOBAL_STATS_LEN
1534 + NETVSC_VF_STATS_LEN
1535 + NETVSC_QUEUE_STATS_LEN(nvdev)
1536 + NETVSC_PCPU_STATS_LEN;
1542 static void netvsc_get_ethtool_stats(struct net_device *dev,
1543 struct ethtool_stats *stats, u64 *data)
1545 struct net_device_context *ndc = netdev_priv(dev);
1546 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1547 const void *nds = &ndc->eth_stats;
1548 const struct netvsc_stats *qstats;
1549 struct netvsc_vf_pcpu_stats sum;
1550 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1559 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1560 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1562 netvsc_get_vf_stats(dev, &sum);
1563 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1564 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1566 for (j = 0; j < nvdev->num_chn; j++) {
1567 qstats = &nvdev->chan_table[j].tx_stats;
1570 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1571 packets = qstats->packets;
1572 bytes = qstats->bytes;
1573 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1574 data[i++] = packets;
1577 qstats = &nvdev->chan_table[j].rx_stats;
1579 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1580 packets = qstats->packets;
1581 bytes = qstats->bytes;
1582 xdp_drop = qstats->xdp_drop;
1583 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1584 data[i++] = packets;
1586 data[i++] = xdp_drop;
1589 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1590 sizeof(struct netvsc_ethtool_pcpu_stats),
1592 netvsc_get_pcpu_stats(dev, pcpu_sum);
1593 for_each_present_cpu(cpu) {
1594 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1596 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1597 data[i++] = *(u64 *)((void *)this_sum
1598 + pcpu_stats[j].offset);
1603 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1605 struct net_device_context *ndc = netdev_priv(dev);
1606 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1613 switch (stringset) {
1615 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1616 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1617 p += ETH_GSTRING_LEN;
1620 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1621 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1622 p += ETH_GSTRING_LEN;
1625 for (i = 0; i < nvdev->num_chn; i++) {
1626 sprintf(p, "tx_queue_%u_packets", i);
1627 p += ETH_GSTRING_LEN;
1628 sprintf(p, "tx_queue_%u_bytes", i);
1629 p += ETH_GSTRING_LEN;
1630 sprintf(p, "rx_queue_%u_packets", i);
1631 p += ETH_GSTRING_LEN;
1632 sprintf(p, "rx_queue_%u_bytes", i);
1633 p += ETH_GSTRING_LEN;
1634 sprintf(p, "rx_queue_%u_xdp_drop", i);
1635 p += ETH_GSTRING_LEN;
1638 for_each_present_cpu(cpu) {
1639 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1640 sprintf(p, pcpu_stats[i].name, cpu);
1641 p += ETH_GSTRING_LEN;
1650 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1651 struct ethtool_rxnfc *info)
1653 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1655 info->data = RXH_IP_SRC | RXH_IP_DST;
1657 switch (info->flow_type) {
1659 if (ndc->l4_hash & HV_TCP4_L4HASH)
1660 info->data |= l4_flag;
1665 if (ndc->l4_hash & HV_TCP6_L4HASH)
1666 info->data |= l4_flag;
1671 if (ndc->l4_hash & HV_UDP4_L4HASH)
1672 info->data |= l4_flag;
1677 if (ndc->l4_hash & HV_UDP6_L4HASH)
1678 info->data |= l4_flag;
1694 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1697 struct net_device_context *ndc = netdev_priv(dev);
1698 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1703 switch (info->cmd) {
1704 case ETHTOOL_GRXRINGS:
1705 info->data = nvdev->num_chn;
1709 return netvsc_get_rss_hash_opts(ndc, info);
1714 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1715 struct ethtool_rxnfc *info)
1717 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1718 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1719 switch (info->flow_type) {
1721 ndc->l4_hash |= HV_TCP4_L4HASH;
1725 ndc->l4_hash |= HV_TCP6_L4HASH;
1729 ndc->l4_hash |= HV_UDP4_L4HASH;
1733 ndc->l4_hash |= HV_UDP6_L4HASH;
1743 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1744 switch (info->flow_type) {
1746 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1750 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1754 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1758 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1772 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1774 struct net_device_context *ndc = netdev_priv(ndev);
1776 if (info->cmd == ETHTOOL_SRXFH)
1777 return netvsc_set_rss_hash_opts(ndc, info);
1782 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1784 return NETVSC_HASH_KEYLEN;
1787 static u32 netvsc_rss_indir_size(struct net_device *dev)
1792 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1795 struct net_device_context *ndc = netdev_priv(dev);
1796 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1797 struct rndis_device *rndis_dev;
1804 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1806 rndis_dev = ndev->extension;
1808 for (i = 0; i < ITAB_NUM; i++)
1809 indir[i] = ndc->rx_table[i];
1813 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1818 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1819 const u8 *key, const u8 hfunc)
1821 struct net_device_context *ndc = netdev_priv(dev);
1822 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1823 struct rndis_device *rndis_dev;
1829 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1832 rndis_dev = ndev->extension;
1834 for (i = 0; i < ITAB_NUM; i++)
1835 if (indir[i] >= ndev->num_chn)
1838 for (i = 0; i < ITAB_NUM; i++)
1839 ndc->rx_table[i] = indir[i];
1846 key = rndis_dev->rss_key;
1849 return rndis_filter_set_rss_param(rndis_dev, key);
1852 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1853 * It does have pre-allocated receive area which is divided into sections.
1855 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1856 struct ethtool_ringparam *ring)
1860 ring->rx_pending = nvdev->recv_section_cnt;
1861 ring->tx_pending = nvdev->send_section_cnt;
1863 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1864 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1866 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1868 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1869 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1870 / nvdev->send_section_size;
1873 static void netvsc_get_ringparam(struct net_device *ndev,
1874 struct ethtool_ringparam *ring)
1876 struct net_device_context *ndevctx = netdev_priv(ndev);
1877 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1882 __netvsc_get_ringparam(nvdev, ring);
1885 static int netvsc_set_ringparam(struct net_device *ndev,
1886 struct ethtool_ringparam *ring)
1888 struct net_device_context *ndevctx = netdev_priv(ndev);
1889 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1890 struct netvsc_device_info *device_info;
1891 struct ethtool_ringparam orig;
1895 if (!nvdev || nvdev->destroy)
1898 memset(&orig, 0, sizeof(orig));
1899 __netvsc_get_ringparam(nvdev, &orig);
1901 new_tx = clamp_t(u32, ring->tx_pending,
1902 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1903 new_rx = clamp_t(u32, ring->rx_pending,
1904 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1906 if (new_tx == orig.tx_pending &&
1907 new_rx == orig.rx_pending)
1908 return 0; /* no change */
1910 device_info = netvsc_devinfo_get(nvdev);
1915 device_info->send_sections = new_tx;
1916 device_info->recv_sections = new_rx;
1918 ret = netvsc_detach(ndev, nvdev);
1922 ret = netvsc_attach(ndev, device_info);
1924 device_info->send_sections = orig.tx_pending;
1925 device_info->recv_sections = orig.rx_pending;
1927 if (netvsc_attach(ndev, device_info))
1928 netdev_err(ndev, "restoring ringparam failed");
1932 netvsc_devinfo_put(device_info);
1936 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1937 netdev_features_t features)
1939 struct net_device_context *ndevctx = netdev_priv(ndev);
1940 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1942 if (!nvdev || nvdev->destroy)
1945 if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1946 features ^= NETIF_F_LRO;
1947 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1953 static int netvsc_set_features(struct net_device *ndev,
1954 netdev_features_t features)
1956 netdev_features_t change = features ^ ndev->features;
1957 struct net_device_context *ndevctx = netdev_priv(ndev);
1958 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1959 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1960 struct ndis_offload_params offloads;
1963 if (!nvdev || nvdev->destroy)
1966 if (!(change & NETIF_F_LRO))
1969 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1971 if (features & NETIF_F_LRO) {
1972 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1973 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1975 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1976 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1979 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1982 features ^= NETIF_F_LRO;
1983 ndev->features = features;
1990 vf_netdev->wanted_features = features;
1991 netdev_update_features(vf_netdev);
1996 static int netvsc_get_regs_len(struct net_device *netdev)
1998 return VRSS_SEND_TAB_SIZE * sizeof(u32);
2001 static void netvsc_get_regs(struct net_device *netdev,
2002 struct ethtool_regs *regs, void *p)
2004 struct net_device_context *ndc = netdev_priv(netdev);
2007 /* increase the version, if buffer format is changed. */
2010 memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
2013 static u32 netvsc_get_msglevel(struct net_device *ndev)
2015 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2017 return ndev_ctx->msg_enable;
2020 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2022 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2024 ndev_ctx->msg_enable = val;
2027 static const struct ethtool_ops ethtool_ops = {
2028 .get_drvinfo = netvsc_get_drvinfo,
2029 .get_regs_len = netvsc_get_regs_len,
2030 .get_regs = netvsc_get_regs,
2031 .get_msglevel = netvsc_get_msglevel,
2032 .set_msglevel = netvsc_set_msglevel,
2033 .get_link = ethtool_op_get_link,
2034 .get_ethtool_stats = netvsc_get_ethtool_stats,
2035 .get_sset_count = netvsc_get_sset_count,
2036 .get_strings = netvsc_get_strings,
2037 .get_channels = netvsc_get_channels,
2038 .set_channels = netvsc_set_channels,
2039 .get_ts_info = ethtool_op_get_ts_info,
2040 .get_rxnfc = netvsc_get_rxnfc,
2041 .set_rxnfc = netvsc_set_rxnfc,
2042 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2043 .get_rxfh_indir_size = netvsc_rss_indir_size,
2044 .get_rxfh = netvsc_get_rxfh,
2045 .set_rxfh = netvsc_set_rxfh,
2046 .get_link_ksettings = netvsc_get_link_ksettings,
2047 .set_link_ksettings = netvsc_set_link_ksettings,
2048 .get_ringparam = netvsc_get_ringparam,
2049 .set_ringparam = netvsc_set_ringparam,
2052 static const struct net_device_ops device_ops = {
2053 .ndo_open = netvsc_open,
2054 .ndo_stop = netvsc_close,
2055 .ndo_start_xmit = netvsc_start_xmit,
2056 .ndo_change_rx_flags = netvsc_change_rx_flags,
2057 .ndo_set_rx_mode = netvsc_set_rx_mode,
2058 .ndo_fix_features = netvsc_fix_features,
2059 .ndo_set_features = netvsc_set_features,
2060 .ndo_change_mtu = netvsc_change_mtu,
2061 .ndo_validate_addr = eth_validate_addr,
2062 .ndo_set_mac_address = netvsc_set_mac_addr,
2063 .ndo_select_queue = netvsc_select_queue,
2064 .ndo_get_stats64 = netvsc_get_stats64,
2065 .ndo_bpf = netvsc_bpf,
2069 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2070 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2071 * present send GARP packet to network peers with netif_notify_peers().
2073 static void netvsc_link_change(struct work_struct *w)
2075 struct net_device_context *ndev_ctx =
2076 container_of(w, struct net_device_context, dwork.work);
2077 struct hv_device *device_obj = ndev_ctx->device_ctx;
2078 struct net_device *net = hv_get_drvdata(device_obj);
2079 unsigned long flags, next_reconfig, delay;
2080 struct netvsc_reconfig *event = NULL;
2081 struct netvsc_device *net_device;
2082 struct rndis_device *rdev;
2083 bool reschedule = false;
2085 /* if changes are happening, comeback later */
2086 if (!rtnl_trylock()) {
2087 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2091 net_device = rtnl_dereference(ndev_ctx->nvdev);
2095 rdev = net_device->extension;
2097 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2098 if (time_is_after_jiffies(next_reconfig)) {
2099 /* link_watch only sends one notification with current state
2100 * per second, avoid doing reconfig more frequently. Handle
2103 delay = next_reconfig - jiffies;
2104 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2105 schedule_delayed_work(&ndev_ctx->dwork, delay);
2108 ndev_ctx->last_reconfig = jiffies;
2110 spin_lock_irqsave(&ndev_ctx->lock, flags);
2111 if (!list_empty(&ndev_ctx->reconfig_events)) {
2112 event = list_first_entry(&ndev_ctx->reconfig_events,
2113 struct netvsc_reconfig, list);
2114 list_del(&event->list);
2115 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2117 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2122 switch (event->event) {
2123 /* Only the following events are possible due to the check in
2124 * netvsc_linkstatus_callback()
2126 case RNDIS_STATUS_MEDIA_CONNECT:
2127 if (rdev->link_state) {
2128 rdev->link_state = false;
2129 netif_carrier_on(net);
2130 netvsc_tx_enable(net_device, net);
2132 __netdev_notify_peers(net);
2136 case RNDIS_STATUS_MEDIA_DISCONNECT:
2137 if (!rdev->link_state) {
2138 rdev->link_state = true;
2139 netif_carrier_off(net);
2140 netvsc_tx_disable(net_device, net);
2144 case RNDIS_STATUS_NETWORK_CHANGE:
2145 /* Only makes sense if carrier is present */
2146 if (!rdev->link_state) {
2147 rdev->link_state = true;
2148 netif_carrier_off(net);
2149 netvsc_tx_disable(net_device, net);
2150 event->event = RNDIS_STATUS_MEDIA_CONNECT;
2151 spin_lock_irqsave(&ndev_ctx->lock, flags);
2152 list_add(&event->list, &ndev_ctx->reconfig_events);
2153 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2161 /* link_watch only sends one notification with current state per
2162 * second, handle next reconfig event in 2 seconds.
2165 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2173 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2175 struct net_device_context *net_device_ctx;
2176 struct net_device *dev;
2178 dev = netdev_master_upper_dev_get(vf_netdev);
2179 if (!dev || dev->netdev_ops != &device_ops)
2180 return NULL; /* not a netvsc device */
2182 net_device_ctx = netdev_priv(dev);
2183 if (!rtnl_dereference(net_device_ctx->nvdev))
2184 return NULL; /* device is removed */
2189 /* Called when VF is injecting data into network stack.
2190 * Change the associated network device from VF to netvsc.
2191 * note: already called with rcu_read_lock
2193 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2195 struct sk_buff *skb = *pskb;
2196 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2197 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2198 struct netvsc_vf_pcpu_stats *pcpu_stats
2199 = this_cpu_ptr(ndev_ctx->vf_stats);
2201 skb = skb_share_check(skb, GFP_ATOMIC);
2203 return RX_HANDLER_CONSUMED;
2209 u64_stats_update_begin(&pcpu_stats->syncp);
2210 pcpu_stats->rx_packets++;
2211 pcpu_stats->rx_bytes += skb->len;
2212 u64_stats_update_end(&pcpu_stats->syncp);
2214 return RX_HANDLER_ANOTHER;
2217 static int netvsc_vf_join(struct net_device *vf_netdev,
2218 struct net_device *ndev)
2220 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2223 ret = netdev_rx_handler_register(vf_netdev,
2224 netvsc_vf_handle_frame, ndev);
2226 netdev_err(vf_netdev,
2227 "can not register netvsc VF receive handler (err = %d)\n",
2229 goto rx_handler_failed;
2232 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2235 netdev_err(vf_netdev,
2236 "can not set master device %s (err = %d)\n",
2238 goto upper_link_failed;
2241 /* set slave flag before open to prevent IPv6 addrconf */
2242 vf_netdev->flags |= IFF_SLAVE;
2244 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2246 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2248 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2252 netdev_rx_handler_unregister(vf_netdev);
2257 static void __netvsc_vf_setup(struct net_device *ndev,
2258 struct net_device *vf_netdev)
2262 /* Align MTU of VF with master */
2263 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2265 netdev_warn(vf_netdev,
2266 "unable to change mtu to %u\n", ndev->mtu);
2268 /* set multicast etc flags on VF */
2269 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2271 /* sync address list from ndev to VF */
2272 netif_addr_lock_bh(ndev);
2273 dev_uc_sync(vf_netdev, ndev);
2274 dev_mc_sync(vf_netdev, ndev);
2275 netif_addr_unlock_bh(ndev);
2277 if (netif_running(ndev)) {
2278 ret = dev_open(vf_netdev, NULL);
2280 netdev_warn(vf_netdev,
2281 "unable to open: %d\n", ret);
2285 /* Setup VF as slave of the synthetic device.
2286 * Runs in workqueue to avoid recursion in netlink callbacks.
2288 static void netvsc_vf_setup(struct work_struct *w)
2290 struct net_device_context *ndev_ctx
2291 = container_of(w, struct net_device_context, vf_takeover.work);
2292 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2293 struct net_device *vf_netdev;
2295 if (!rtnl_trylock()) {
2296 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2300 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2302 __netvsc_vf_setup(ndev, vf_netdev);
2307 /* Find netvsc by VF serial number.
2308 * The PCI hyperv controller records the serial number as the slot kobj name.
2310 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2312 struct device *parent = vf_netdev->dev.parent;
2313 struct net_device_context *ndev_ctx;
2314 struct pci_dev *pdev;
2317 if (!parent || !dev_is_pci(parent))
2318 return NULL; /* not a PCI device */
2320 pdev = to_pci_dev(parent);
2322 netdev_notice(vf_netdev, "no PCI slot information\n");
2326 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2327 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2328 pci_slot_name(pdev->slot));
2332 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2333 if (!ndev_ctx->vf_alloc)
2336 if (ndev_ctx->vf_serial == serial)
2337 return hv_get_drvdata(ndev_ctx->device_ctx);
2340 netdev_notice(vf_netdev,
2341 "no netdev found for vf serial:%u\n", serial);
2345 static int netvsc_register_vf(struct net_device *vf_netdev)
2347 struct net_device_context *net_device_ctx;
2348 struct netvsc_device *netvsc_dev;
2349 struct bpf_prog *prog;
2350 struct net_device *ndev;
2353 if (vf_netdev->addr_len != ETH_ALEN)
2356 ndev = get_netvsc_byslot(vf_netdev);
2360 net_device_ctx = netdev_priv(ndev);
2361 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2362 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2365 /* if synthetic interface is a different namespace,
2366 * then move the VF to that namespace; join will be
2367 * done again in that context.
2369 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2370 ret = dev_change_net_namespace(vf_netdev,
2371 dev_net(ndev), "eth%d");
2373 netdev_err(vf_netdev,
2374 "could not move to same namespace as %s: %d\n",
2377 netdev_info(vf_netdev,
2378 "VF moved to namespace with: %s\n",
2383 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2385 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2388 dev_hold(vf_netdev);
2389 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2391 vf_netdev->wanted_features = ndev->features;
2392 netdev_update_features(vf_netdev);
2394 prog = netvsc_xdp_get(netvsc_dev);
2395 netvsc_vf_setxdp(vf_netdev, prog);
2400 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2402 * Typically a UP or DOWN event is followed by a CHANGE event, so
2403 * net_device_ctx->data_path_is_vf is used to cache the current data path
2404 * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2407 * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2408 * interface, there is only the CHANGE event and no UP or DOWN event.
2410 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2412 struct net_device_context *net_device_ctx;
2413 struct netvsc_device *netvsc_dev;
2414 struct net_device *ndev;
2415 bool vf_is_up = false;
2417 if (event != NETDEV_GOING_DOWN)
2418 vf_is_up = netif_running(vf_netdev);
2420 ndev = get_netvsc_byref(vf_netdev);
2424 net_device_ctx = netdev_priv(ndev);
2425 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2429 if (net_device_ctx->data_path_is_vf == vf_is_up)
2432 netvsc_switch_datapath(ndev, vf_is_up);
2433 netdev_info(ndev, "Data path switched %s VF: %s\n",
2434 vf_is_up ? "to" : "from", vf_netdev->name);
2439 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2441 struct net_device *ndev;
2442 struct net_device_context *net_device_ctx;
2444 ndev = get_netvsc_byref(vf_netdev);
2448 net_device_ctx = netdev_priv(ndev);
2449 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2451 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2453 netvsc_vf_setxdp(vf_netdev, NULL);
2455 netdev_rx_handler_unregister(vf_netdev);
2456 netdev_upper_dev_unlink(vf_netdev, ndev);
2457 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2463 static int netvsc_probe(struct hv_device *dev,
2464 const struct hv_vmbus_device_id *dev_id)
2466 struct net_device *net = NULL;
2467 struct net_device_context *net_device_ctx;
2468 struct netvsc_device_info *device_info = NULL;
2469 struct netvsc_device *nvdev;
2472 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2477 netif_carrier_off(net);
2479 netvsc_init_settings(net);
2481 net_device_ctx = netdev_priv(net);
2482 net_device_ctx->device_ctx = dev;
2483 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2484 if (netif_msg_probe(net_device_ctx))
2485 netdev_dbg(net, "netvsc msg_enable: %d\n",
2486 net_device_ctx->msg_enable);
2488 hv_set_drvdata(dev, net);
2490 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2492 spin_lock_init(&net_device_ctx->lock);
2493 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2494 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2496 net_device_ctx->vf_stats
2497 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2498 if (!net_device_ctx->vf_stats)
2501 net->netdev_ops = &device_ops;
2502 net->ethtool_ops = ðtool_ops;
2503 SET_NETDEV_DEV(net, &dev->device);
2505 /* We always need headroom for rndis header */
2506 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2508 /* Initialize the number of queues to be 1, we may change it if more
2509 * channels are offered later.
2511 netif_set_real_num_tx_queues(net, 1);
2512 netif_set_real_num_rx_queues(net, 1);
2514 /* Notify the netvsc driver of the new device */
2515 device_info = netvsc_devinfo_get(NULL);
2519 goto devinfo_failed;
2522 nvdev = rndis_filter_device_add(dev, device_info);
2523 if (IS_ERR(nvdev)) {
2524 ret = PTR_ERR(nvdev);
2525 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2529 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2531 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2532 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2533 * all subchannels to show up, but that may not happen because
2534 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2535 * -> ... -> device_add() -> ... -> __device_attach() can't get
2536 * the device lock, so all the subchannels can't be processed --
2537 * finally netvsc_subchan_work() hangs forever.
2541 if (nvdev->num_chn > 1)
2542 schedule_work(&nvdev->subchan_work);
2544 /* hw_features computed in rndis_netdev_set_hwcaps() */
2545 net->features = net->hw_features |
2546 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2547 NETIF_F_HW_VLAN_CTAG_RX;
2548 net->vlan_features = net->features;
2550 netdev_lockdep_set_classes(net);
2552 /* MTU range: 68 - 1500 or 65521 */
2553 net->min_mtu = NETVSC_MTU_MIN;
2554 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2555 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2557 net->max_mtu = ETH_DATA_LEN;
2559 nvdev->tx_disable = false;
2561 ret = register_netdevice(net);
2563 pr_err("Unable to register netdev.\n");
2564 goto register_failed;
2567 list_add(&net_device_ctx->list, &netvsc_dev_list);
2570 netvsc_devinfo_put(device_info);
2575 rndis_filter_device_remove(dev, nvdev);
2577 netvsc_devinfo_put(device_info);
2579 free_percpu(net_device_ctx->vf_stats);
2581 hv_set_drvdata(dev, NULL);
2587 static int netvsc_remove(struct hv_device *dev)
2589 struct net_device_context *ndev_ctx;
2590 struct net_device *vf_netdev, *net;
2591 struct netvsc_device *nvdev;
2593 net = hv_get_drvdata(dev);
2595 dev_err(&dev->device, "No net device to remove\n");
2599 ndev_ctx = netdev_priv(net);
2601 cancel_delayed_work_sync(&ndev_ctx->dwork);
2604 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2606 cancel_work_sync(&nvdev->subchan_work);
2607 netvsc_xdp_set(net, NULL, NULL, nvdev);
2611 * Call to the vsc driver to let it know that the device is being
2612 * removed. Also blocks mtu and channel changes.
2614 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2616 netvsc_unregister_vf(vf_netdev);
2619 rndis_filter_device_remove(dev, nvdev);
2621 unregister_netdevice(net);
2622 list_del(&ndev_ctx->list);
2626 hv_set_drvdata(dev, NULL);
2628 free_percpu(ndev_ctx->vf_stats);
2633 static int netvsc_suspend(struct hv_device *dev)
2635 struct net_device_context *ndev_ctx;
2636 struct netvsc_device *nvdev;
2637 struct net_device *net;
2640 net = hv_get_drvdata(dev);
2642 ndev_ctx = netdev_priv(net);
2643 cancel_delayed_work_sync(&ndev_ctx->dwork);
2647 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2648 if (nvdev == NULL) {
2653 /* Save the current config info */
2654 ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2656 ret = netvsc_detach(net, nvdev);
2663 static int netvsc_resume(struct hv_device *dev)
2665 struct net_device *net = hv_get_drvdata(dev);
2666 struct net_device_context *net_device_ctx;
2667 struct netvsc_device_info *device_info;
2672 net_device_ctx = netdev_priv(net);
2674 /* Reset the data path to the netvsc NIC before re-opening the vmbus
2675 * channel. Later netvsc_netdev_event() will switch the data path to
2676 * the VF upon the UP or CHANGE event.
2678 net_device_ctx->data_path_is_vf = false;
2679 device_info = net_device_ctx->saved_netvsc_dev_info;
2681 ret = netvsc_attach(net, device_info);
2683 netvsc_devinfo_put(device_info);
2684 net_device_ctx->saved_netvsc_dev_info = NULL;
2690 static const struct hv_vmbus_device_id id_table[] = {
2696 MODULE_DEVICE_TABLE(vmbus, id_table);
2698 /* The one and only one */
2699 static struct hv_driver netvsc_drv = {
2700 .name = KBUILD_MODNAME,
2701 .id_table = id_table,
2702 .probe = netvsc_probe,
2703 .remove = netvsc_remove,
2704 .suspend = netvsc_suspend,
2705 .resume = netvsc_resume,
2707 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2712 * On Hyper-V, every VF interface is matched with a corresponding
2713 * synthetic interface. The synthetic interface is presented first
2714 * to the guest. When the corresponding VF instance is registered,
2715 * we will take care of switching the data path.
2717 static int netvsc_netdev_event(struct notifier_block *this,
2718 unsigned long event, void *ptr)
2720 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2722 /* Skip our own events */
2723 if (event_dev->netdev_ops == &device_ops)
2726 /* Avoid non-Ethernet type devices */
2727 if (event_dev->type != ARPHRD_ETHER)
2730 /* Avoid Vlan dev with same MAC registering as VF */
2731 if (is_vlan_dev(event_dev))
2734 /* Avoid Bonding master dev with same MAC registering as VF */
2735 if ((event_dev->priv_flags & IFF_BONDING) &&
2736 (event_dev->flags & IFF_MASTER))
2740 case NETDEV_REGISTER:
2741 return netvsc_register_vf(event_dev);
2742 case NETDEV_UNREGISTER:
2743 return netvsc_unregister_vf(event_dev);
2747 case NETDEV_GOING_DOWN:
2748 return netvsc_vf_changed(event_dev, event);
2754 static struct notifier_block netvsc_netdev_notifier = {
2755 .notifier_call = netvsc_netdev_event,
2758 static void __exit netvsc_drv_exit(void)
2760 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2761 vmbus_driver_unregister(&netvsc_drv);
2764 static int __init netvsc_drv_init(void)
2768 if (ring_size < RING_SIZE_MIN) {
2769 ring_size = RING_SIZE_MIN;
2770 pr_info("Increased ring_size to %u (min allowed)\n",
2773 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2775 ret = vmbus_driver_register(&netvsc_drv);
2779 register_netdevice_notifier(&netvsc_netdev_notifier);
2783 MODULE_LICENSE("GPL");
2784 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2786 module_init(netvsc_drv_init);
2787 module_exit(netvsc_drv_exit);