2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
69 #include <sys/queue.h>
74 #include <net/if_arp.h>
75 #include <net/ethernet.h>
76 #include <net/if_dl.h>
77 #include <net/if_media.h>
81 #include <net/if_var.h>
82 #include <net/if_types.h>
83 #include <net/if_vlan_var.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/in.h>
87 #include <netinet/ip.h>
88 #include <netinet/if_ether.h>
89 #include <netinet/tcp.h>
90 #include <netinet/udp.h>
91 #include <netinet/ip6.h>
94 #include <vm/vm_param.h>
95 #include <vm/vm_kern.h>
98 #include <machine/bus.h>
99 #include <machine/resource.h>
100 #include <machine/frame.h>
101 #include <machine/vmparam.h>
104 #include <sys/rman.h>
105 #include <sys/mutex.h>
106 #include <sys/errno.h>
107 #include <sys/types.h>
108 #include <machine/atomic.h>
110 #include <machine/intr_machdep.h>
112 #include <machine/in_cksum.h>
114 #include <dev/hyperv/include/hyperv.h>
115 #include "hv_net_vsc.h"
116 #include "hv_rndis.h"
117 #include "hv_rndis_filter.h"
120 /* Short for Hyper-V network interface */
121 #define NETVSC_DEVNAME "hn"
124 * It looks like offset 0 of buf is reserved to hold the softc pointer.
125 * The sc pointer evidently not needed, and is not presently populated.
126 * The packet offset is where the netvsc_packet starts in the buffer.
128 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
129 #define HV_NV_PACKET_OFFSET_IN_BUF 16
132 * A unified flag for all outbound check sum flags is useful,
133 * and it helps avoiding unnecessary check sum calculation in
134 * network forwarding scenario.
136 #define HV_CSUM_FOR_OUTBOUND \
137 (CSUM_IP|CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP|CSUM_IP_TSO| \
138 CSUM_IP_ISCSI|CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP| \
139 CSUM_IP6_TSO|CSUM_IP6_ISCSI)
145 struct hv_netvsc_driver_context {
150 * Be aware that this sleepable mutex will exhibit WITNESS errors when
151 * certain TCP and ARP code paths are taken. This appears to be a
152 * well-known condition, as all other drivers checked use a sleeping
153 * mutex to protect their transmit paths.
154 * Also Be aware that mutexes do not play well with semaphores, and there
155 * is a conflicting semaphore in a certain channel code path.
157 #define NV_LOCK_INIT(_sc, _name) \
158 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
159 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
160 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
161 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
162 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
169 int hv_promisc_mode = 0; /* normal mode by default */
171 /* The one and only one */
172 static struct hv_netvsc_driver_context g_netvsc_drv;
176 * Forward declarations
178 static void hn_stop(hn_softc_t *sc);
179 static void hn_ifinit_locked(hn_softc_t *sc);
180 static void hn_ifinit(void *xsc);
181 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
182 static int hn_start_locked(struct ifnet *ifp);
183 static void hn_start(struct ifnet *ifp);
186 * NetVsc get message transport protocol type
188 static uint32_t get_transport_proto_type(struct mbuf *m_head)
190 uint32_t ret_val = TRANSPORT_TYPE_NOT_IP;
191 uint16_t ether_type = 0;
193 struct ether_vlan_header *eh;
201 eh = mtod(m_head, struct ether_vlan_header*);
202 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
203 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
204 ether_type = eh->evl_proto;
206 ether_len = ETHER_HDR_LEN;
207 ether_type = eh->evl_encap_proto;
210 switch (ntohs(ether_type)) {
213 ip6 = (struct ip6_hdr *)(m_head->m_data + ether_len);
215 if (IPPROTO_TCP == ip6->ip6_nxt) {
216 ret_val = TRANSPORT_TYPE_IPV6_TCP;
217 } else if (IPPROTO_UDP == ip6->ip6_nxt) {
218 ret_val = TRANSPORT_TYPE_IPV6_UDP;
224 iph = (struct ip *)(m_head->m_data + ether_len);
226 if (IPPROTO_TCP == iph->ip_p) {
227 ret_val = TRANSPORT_TYPE_IPV4_TCP;
228 } else if (IPPROTO_UDP == iph->ip_p) {
229 ret_val = TRANSPORT_TYPE_IPV4_UDP;
234 ret_val = TRANSPORT_TYPE_NOT_IP;
242 * NetVsc driver initialization
243 * Note: Filter init is no longer required
246 netvsc_drv_init(void)
252 * NetVsc global initialization entry point
258 printf("Netvsc initializing... ");
261 * XXXKYS: cleanup initialization
263 if (!cold && !g_netvsc_drv.drv_inited) {
264 g_netvsc_drv.drv_inited = 1;
268 } else if (bootverbose)
269 printf("Already initialized!\n");
272 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
273 static const hv_guid g_net_vsc_device_type = {
274 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
275 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
279 * Standard probe entry point.
283 netvsc_probe(device_t dev)
287 p = vmbus_get_type(dev);
288 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
289 device_set_desc(dev, "Synthetic Network Interface");
291 printf("Netvsc probe... DONE \n");
293 return (BUS_PROBE_DEFAULT);
300 * Standard attach entry point.
302 * Called when the driver is loaded. It allocates needed resources,
303 * and initializes the "hardware" and software.
306 netvsc_attach(device_t dev)
308 struct hv_device *device_ctx = vmbus_get_devctx(dev);
309 netvsc_device_info device_info;
311 int unit = device_get_unit(dev);
317 sc = device_get_softc(dev);
322 bzero(sc, sizeof(hn_softc_t));
326 NV_LOCK_INIT(sc, "NetVSCLock");
328 sc->hn_dev_obj = device_ctx;
330 ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
333 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
334 ifp->if_dunit = unit;
335 ifp->if_dname = NETVSC_DEVNAME;
337 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
338 ifp->if_ioctl = hn_ioctl;
339 ifp->if_start = hn_start;
340 ifp->if_init = hn_ifinit;
341 /* needed by hv_rf_on_device_add() code */
342 ifp->if_mtu = ETHERMTU;
343 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
344 ifp->if_snd.ifq_drv_maxlen = 511;
345 IFQ_SET_READY(&ifp->if_snd);
348 * Tell upper layers that we support full VLAN capability.
350 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
351 ifp->if_capabilities |=
352 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO;
354 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO;
356 * Only enable UDP checksum offloading when it is on 2012R2 or
357 * later. UDP checksum offloading doesn't work on earlier
360 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
361 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_TSO;
363 ifp->if_hwassist = CSUM_TCP | CSUM_TSO;
365 ret = hv_rf_on_device_add(device_ctx, &device_info);
371 if (device_info.link_state == 0) {
375 ether_ifattach(ifp, device_info.mac_addr);
381 * Standard detach entry point
384 netvsc_detach(device_t dev)
386 struct hv_device *hv_device = vmbus_get_devctx(dev);
389 printf("netvsc_detach\n");
392 * XXXKYS: Need to clean up all our
393 * driver state; this is the driver
398 * XXXKYS: Need to stop outgoing traffic and unregister
402 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
408 * Standard shutdown entry point
411 netvsc_shutdown(device_t dev)
417 * Send completion processing
419 * Note: It looks like offset 0 of buf is reserved to hold the softc
420 * pointer. The sc pointer is not currently needed in this function, and
421 * it is not presently populated by the TX function.
424 netvsc_xmit_completion(void *context)
426 netvsc_packet *packet = (netvsc_packet *)context;
430 mb = (struct mbuf *)(uintptr_t)packet->compl.send.send_completion_tid;
431 buf = ((uint8_t *)packet) - HV_NV_PACKET_OFFSET_IN_BUF;
441 * Start a transmit of one or more packets
444 hn_start_locked(struct ifnet *ifp)
446 hn_softc_t *sc = ifp->if_softc;
447 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
448 netvsc_dev *net_dev = sc->net_dev;
449 device_t dev = device_ctx->device;
451 netvsc_packet *packet;
452 struct mbuf *m_head, *m;
453 struct mbuf *mc_head = NULL;
454 struct ether_vlan_header *eh;
455 rndis_msg *rndis_mesg;
456 rndis_packet *rndis_pkt;
457 rndis_per_packet_info *rppi;
458 ndis_8021q_info *rppi_vlan_info;
459 rndis_tcp_ip_csum_info *csum_info;
460 rndis_tcp_tso_info *tso_info;
467 uint32_t rndis_msg_size = 0;
468 uint32_t trans_proto_type;
469 uint32_t send_buf_section_idx =
470 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
472 while (!IFQ_DRV_IS_EMPTY(&sc->hn_ifp->if_snd)) {
473 IFQ_DRV_DEQUEUE(&sc->hn_ifp->if_snd, m_head);
474 if (m_head == NULL) {
481 /* Walk the mbuf list computing total length and num frags */
482 for (m = m_head; m != NULL; m = m->m_next) {
490 * Reserve the number of pages requested. Currently,
491 * one page is reserved for the message in the RNDIS
494 num_frags += HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
496 /* If exceeds # page_buffers in netvsc_packet */
497 if (num_frags > NETVSC_PACKET_MAXPAGE) {
498 device_printf(dev, "exceed max page buffers,%d,%d\n",
499 num_frags, NETVSC_PACKET_MAXPAGE);
501 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
506 * Allocate a buffer with space for a netvsc packet plus a
507 * number of reserved areas. First comes a (currently 16
508 * bytes, currently unused) reserved data area. Second is
509 * the netvsc_packet. Third is an area reserved for an
510 * rndis_filter_packet struct. Fourth (optional) is a
511 * rndis_per_packet_info struct.
512 * Changed malloc to M_NOWAIT to avoid sleep under spin lock.
513 * No longer reserving extra space for page buffers, as they
514 * are already part of the netvsc_packet.
516 buf = malloc(HV_NV_PACKET_OFFSET_IN_BUF +
517 sizeof(netvsc_packet) +
519 RNDIS_VLAN_PPI_SIZE +
522 M_NETVSC, M_ZERO | M_NOWAIT);
524 device_printf(dev, "hn:malloc packet failed\n");
526 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
530 packet = (netvsc_packet *)(buf + HV_NV_PACKET_OFFSET_IN_BUF);
531 *(vm_offset_t *)buf = HV_NV_SC_PTR_OFFSET_IN_BUF;
533 packet->is_data_pkt = TRUE;
535 /* Set up the rndis header */
536 packet->page_buf_count = num_frags;
538 /* Initialize it from the mbuf */
539 packet->tot_data_buf_len = len;
542 * extension points to the area reserved for the
543 * rndis_filter_packet, which is placed just after
544 * the netvsc_packet (and rppi struct, if present;
545 * length is updated later).
547 packet->rndis_mesg = packet + 1;
548 rndis_mesg = (rndis_msg *)packet->rndis_mesg;
549 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
551 rndis_pkt = &rndis_mesg->msg.packet;
552 rndis_pkt->data_offset = sizeof(rndis_packet);
553 rndis_pkt->data_length = packet->tot_data_buf_len;
554 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
556 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
559 * If the Hyper-V infrastructure needs to embed a VLAN tag,
560 * initialize netvsc_packet and rppi struct values as needed.
562 if (m_head->m_flags & M_VLANTAG) {
564 * set up some additional fields so the Hyper-V infrastructure will stuff the VLAN tag
567 packet->vlan_tci = m_head->m_pkthdr.ether_vtag;
569 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
571 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
574 /* VLAN info immediately follows rppi struct */
575 rppi_vlan_info = (ndis_8021q_info *)((char*)rppi +
576 rppi->per_packet_info_offset);
577 /* FreeBSD does not support CFI or priority */
578 rppi_vlan_info->u1.s1.vlan_id =
579 packet->vlan_tci & 0xfff;
582 /* Only check the flags for outbound and ignore the ones for inbound */
583 if (0 == (m_head->m_pkthdr.csum_flags & HV_CSUM_FOR_OUTBOUND)) {
587 eh = mtod(m_head, struct ether_vlan_header*);
588 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
589 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
591 ether_len = ETHER_HDR_LEN;
594 trans_proto_type = get_transport_proto_type(m_head);
595 if (TRANSPORT_TYPE_NOT_IP == trans_proto_type) {
600 * TSO packet needless to setup the send side checksum
603 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
607 /* setup checksum offload */
608 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
609 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
611 csum_info = (rndis_tcp_ip_csum_info *)((char*)rppi +
612 rppi->per_packet_info_offset);
614 if (trans_proto_type & (TYPE_IPV4 << 16)) {
615 csum_info->xmit.is_ipv4 = 1;
617 csum_info->xmit.is_ipv6 = 1;
620 if (trans_proto_type & TYPE_TCP) {
621 csum_info->xmit.tcp_csum = 1;
622 csum_info->xmit.tcp_header_offset = 0;
623 } else if (trans_proto_type & TYPE_UDP) {
624 csum_info->xmit.udp_csum = 1;
630 /* setup TCP segmentation offload */
631 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
632 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
633 tcp_large_send_info);
635 tso_info = (rndis_tcp_tso_info *)((char *)rppi +
636 rppi->per_packet_info_offset);
637 tso_info->lso_v2_xmit.type =
638 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
641 if (trans_proto_type & (TYPE_IPV4 << 16)) {
643 (struct ip *)(m_head->m_data + ether_len);
644 unsigned long iph_len = ip->ip_hl << 2;
646 (struct tcphdr *)((caddr_t)ip + iph_len);
648 tso_info->lso_v2_xmit.ip_version =
649 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
653 th->th_sum = in_pseudo(ip->ip_src.s_addr,
658 #if defined(INET6) && defined(INET)
663 struct ip6_hdr *ip6 =
664 (struct ip6_hdr *)(m_head->m_data + ether_len);
665 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
667 tso_info->lso_v2_xmit.ip_version =
668 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
670 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
673 tso_info->lso_v2_xmit.tcp_header_offset = 0;
674 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
677 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
678 packet->tot_data_buf_len = rndis_mesg->msg_len;
680 /* send packet with send buffer */
681 if (packet->tot_data_buf_len < net_dev->send_section_size) {
682 send_buf_section_idx =
683 hv_nv_get_next_send_section(net_dev);
684 if (send_buf_section_idx !=
685 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
686 char *dest = ((char *)net_dev->send_buf +
687 send_buf_section_idx *
688 net_dev->send_section_size);
690 memcpy(dest, rndis_mesg, rndis_msg_size);
691 dest += rndis_msg_size;
692 for (m = m_head; m != NULL; m = m->m_next) {
695 (void *)mtod(m, vm_offset_t),
701 packet->send_buf_section_idx =
702 send_buf_section_idx;
703 packet->send_buf_section_size =
704 packet->tot_data_buf_len;
705 packet->page_buf_count = 0;
710 /* send packet with page buffer */
711 packet->page_buffers[0].pfn =
712 atop(hv_get_phys_addr(rndis_mesg));
713 packet->page_buffers[0].offset =
714 (unsigned long)rndis_mesg & PAGE_MASK;
715 packet->page_buffers[0].length = rndis_msg_size;
718 * Fill the page buffers with mbuf info starting at index
719 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
721 i = HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
722 for (m = m_head; m != NULL; m = m->m_next) {
725 vtophys(mtod(m, vm_offset_t));
726 packet->page_buffers[i].pfn =
728 packet->page_buffers[i].offset =
729 paddr & (PAGE_SIZE - 1);
730 packet->page_buffers[i].length = m->m_len;
735 packet->send_buf_section_idx =
736 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
737 packet->send_buf_section_size = 0;
742 * If bpf, copy the mbuf chain. This is less expensive than
743 * it appears; the mbuf clusters are not copied, only their
744 * reference counts are incremented.
745 * Needed to avoid a race condition where the completion
746 * callback is invoked, freeing the mbuf chain, before the
747 * bpf_mtap code has a chance to run.
750 mc_head = m_copypacket(m_head, M_NOWAIT);
753 /* Set the completion routine */
754 packet->compl.send.on_send_completion = netvsc_xmit_completion;
755 packet->compl.send.send_completion_context = packet;
756 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)m_head;
758 /* Removed critical_enter(), does not appear necessary */
759 ret = hv_nv_on_send(device_ctx, packet);
761 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
762 /* if bpf && mc_head, call bpf_mtap code */
764 ETHER_BPF_MTAP(ifp, mc_head);
772 IF_PREPEND(&ifp->if_snd, m_head);
773 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
776 * Null the mbuf pointer so the completion function
777 * does not free the mbuf chain. We just pushed the
778 * mbuf chain back on the if_snd queue.
780 packet->compl.send.send_completion_tid = 0;
783 * Release the resources since we will not get any
786 netvsc_xmit_completion(packet);
787 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
790 /* if bpf && mc_head, free the mbuf chain copy */
800 * Link up/down notification
803 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
805 hn_softc_t *sc = device_get_softc(device_obj->device);
819 * Append the specified data to the indicated mbuf chain,
820 * Extend the mbuf chain if the new data does not fit in
823 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
824 * There should be an equivalent in the kernel mbuf code,
825 * but there does not appear to be one yet.
827 * Differs from m_append() in that additional mbufs are
828 * allocated with cluster size MJUMPAGESIZE, and filled
831 * Return 1 if able to complete the job; otherwise 0.
834 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
837 int remainder, space;
839 for (m = m0; m->m_next != NULL; m = m->m_next)
842 space = M_TRAILINGSPACE(m);
845 * Copy into available space.
847 if (space > remainder)
849 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
854 while (remainder > 0) {
856 * Allocate a new mbuf; could check space
857 * and allocate a cluster instead.
859 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
862 n->m_len = min(MJUMPAGESIZE, remainder);
863 bcopy(cp, mtod(n, caddr_t), n->m_len);
865 remainder -= n->m_len;
869 if (m0->m_flags & M_PKTHDR)
870 m0->m_pkthdr.len += len - remainder;
872 return (remainder == 0);
877 * Called when we receive a data packet from the "wire" on the
880 * Note: This is no longer used as a callback
883 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
884 rndis_tcp_ip_csum_info *csum_info)
886 hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
889 device_t dev = device_ctx->device;
893 return (0); /* TODO: KYS how can this be! */
898 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
903 * Bail out if packet contains more data than configured MTU.
905 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
910 * Get an mbuf with a cluster. For packets 2K or less,
911 * get a standard 2K cluster. For anything larger, get a
912 * 4K cluster. Any buffers larger than 4K can cause problems
913 * if looped around to the Hyper-V TX channel, so avoid them.
917 if (packet->tot_data_buf_len > MCLBYTES) {
922 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
925 device_printf(dev, "alloc mbuf failed.\n");
929 hv_m_append(m_new, packet->tot_data_buf_len,
932 m_new->m_pkthdr.rcvif = ifp;
934 /* receive side checksum offload */
935 m_new->m_pkthdr.csum_flags = 0;
936 if (NULL != csum_info) {
937 /* IP csum offload */
938 if (csum_info->receive.ip_csum_succeeded) {
939 m_new->m_pkthdr.csum_flags |=
940 (CSUM_IP_CHECKED | CSUM_IP_VALID);
943 /* TCP csum offload */
944 if (csum_info->receive.tcp_csum_succeeded) {
945 m_new->m_pkthdr.csum_flags |=
946 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
947 m_new->m_pkthdr.csum_data = 0xffff;
951 if ((packet->vlan_tci != 0) &&
952 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
953 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
954 m_new->m_flags |= M_VLANTAG;
958 * Note: Moved RX completion back to hv_nv_on_receive() so all
959 * messages (not just data messages) will trigger a response.
962 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
964 /* We're not holding the lock here, so don't release it */
965 (*ifp->if_input)(ifp, m_new);
971 * Rules for using sc->temp_unusable:
972 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
973 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
974 * sc->temp_unusable set, must release NV_LOCK() and exit
975 * 3. to retain exclusive control of the interface,
976 * sc->temp_unusable must be set by code before releasing NV_LOCK()
977 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
978 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
982 * Standard ioctl entry point. Called when the user wants to configure
986 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
988 hn_softc_t *sc = ifp->if_softc;
989 struct ifreq *ifr = (struct ifreq *)data;
991 struct ifaddr *ifa = (struct ifaddr *)data;
993 netvsc_device_info device_info;
994 struct hv_device *hn_dev;
1002 if (ifa->ifa_addr->sa_family == AF_INET) {
1003 ifp->if_flags |= IFF_UP;
1004 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1006 arp_ifinit(ifp, ifa);
1009 error = ether_ioctl(ifp, cmd, data);
1012 hn_dev = vmbus_get_devctx(sc->hn_dev);
1014 /* Check MTU value change */
1015 if (ifp->if_mtu == ifr->ifr_mtu)
1018 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1023 /* Obtain and record requested MTU */
1024 ifp->if_mtu = ifr->ifr_mtu;
1028 if (!sc->temp_unusable) {
1029 sc->temp_unusable = TRUE;
1033 if (retry_cnt > 0) {
1037 } while (retry_cnt > 0);
1039 if (retry_cnt == 0) {
1044 /* We must remove and add back the device to cause the new
1045 * MTU to take effect. This includes tearing down, but not
1046 * deleting the channel, then bringing it back up.
1048 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1051 sc->temp_unusable = FALSE;
1055 error = hv_rf_on_device_add(hn_dev, &device_info);
1058 sc->temp_unusable = FALSE;
1063 hn_ifinit_locked(sc);
1066 sc->temp_unusable = FALSE;
1072 if (!sc->temp_unusable) {
1073 sc->temp_unusable = TRUE;
1077 if (retry_cnt > 0) {
1081 } while (retry_cnt > 0);
1083 if (retry_cnt == 0) {
1088 if (ifp->if_flags & IFF_UP) {
1090 * If only the state of the PROMISC flag changed,
1091 * then just use the 'set promisc mode' command
1092 * instead of reinitializing the entire NIC. Doing
1093 * a full re-init means reloading the firmware and
1094 * waiting for it to start up, which may take a
1098 /* Fixme: Promiscuous mode? */
1099 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1100 ifp->if_flags & IFF_PROMISC &&
1101 !(sc->hn_if_flags & IFF_PROMISC)) {
1102 /* do something here for Hyper-V */
1103 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1104 !(ifp->if_flags & IFF_PROMISC) &&
1105 sc->hn_if_flags & IFF_PROMISC) {
1106 /* do something here for Hyper-V */
1109 hn_ifinit_locked(sc);
1111 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1116 sc->temp_unusable = FALSE;
1118 sc->hn_if_flags = ifp->if_flags;
1122 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1123 if (mask & IFCAP_TXCSUM) {
1124 if (IFCAP_TXCSUM & ifp->if_capenable) {
1125 ifp->if_capenable &= ~IFCAP_TXCSUM;
1126 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
1128 ifp->if_capenable |= IFCAP_TXCSUM;
1130 * Only enable UDP checksum offloading on
1131 * Windows Server 2012R2 or later releases.
1133 if (hv_vmbus_protocal_version >=
1134 HV_VMBUS_VERSION_WIN8_1) {
1136 (CSUM_TCP | CSUM_UDP);
1138 ifp->if_hwassist |= CSUM_TCP;
1143 if (mask & IFCAP_RXCSUM) {
1144 if (IFCAP_RXCSUM & ifp->if_capenable) {
1145 ifp->if_capenable &= ~IFCAP_RXCSUM;
1147 ifp->if_capenable |= IFCAP_RXCSUM;
1151 if (mask & IFCAP_TSO4) {
1152 ifp->if_capenable ^= IFCAP_TSO4;
1153 ifp->if_hwassist ^= CSUM_IP_TSO;
1156 if (mask & IFCAP_TSO6) {
1157 ifp->if_capenable ^= IFCAP_TSO6;
1158 ifp->if_hwassist ^= CSUM_IP6_TSO;
1166 /* Fixme: Multicast mode? */
1167 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1169 netvsc_setmulti(sc);
1180 error = ether_ioctl(ifp, cmd, data);
1191 hn_stop(hn_softc_t *sc)
1195 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1200 printf(" Closing Device ...\n");
1202 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1203 if_link_state_change(ifp, LINK_STATE_DOWN);
1204 sc->hn_initdone = 0;
1206 ret = hv_rf_on_close(device_ctx);
1210 * FreeBSD transmit entry point
1213 hn_start(struct ifnet *ifp)
1219 if (sc->temp_unusable) {
1223 hn_start_locked(ifp);
1231 hn_ifinit_locked(hn_softc_t *sc)
1234 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1239 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1243 hv_promisc_mode = 1;
1245 ret = hv_rf_on_open(device_ctx);
1249 sc->hn_initdone = 1;
1251 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1252 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1253 if_link_state_change(ifp, LINK_STATE_UP);
1260 hn_ifinit(void *xsc)
1262 hn_softc_t *sc = xsc;
1265 if (sc->temp_unusable) {
1269 sc->temp_unusable = TRUE;
1272 hn_ifinit_locked(sc);
1275 sc->temp_unusable = FALSE;
1284 hn_watchdog(struct ifnet *ifp)
1289 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1290 hn_ifinit(sc); /*???*/
1291 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1295 static device_method_t netvsc_methods[] = {
1296 /* Device interface */
1297 DEVMETHOD(device_probe, netvsc_probe),
1298 DEVMETHOD(device_attach, netvsc_attach),
1299 DEVMETHOD(device_detach, netvsc_detach),
1300 DEVMETHOD(device_shutdown, netvsc_shutdown),
1305 static driver_t netvsc_driver = {
1311 static devclass_t netvsc_devclass;
1313 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
1314 MODULE_VERSION(hn, 1);
1315 MODULE_DEPEND(hn, vmbus, 1, 1, 1);
1316 SYSINIT(netvsc_initx, SI_SUB_KTHREAD_IDLE, SI_ORDER_MIDDLE + 1, netvsc_init,