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>
72 #include <sys/sysctl.h>
75 #include <net/if_arp.h>
76 #include <net/ethernet.h>
77 #include <net/if_dl.h>
78 #include <net/if_media.h>
82 #include <net/if_var.h>
83 #include <net/if_types.h>
84 #include <net/if_vlan_var.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/in.h>
88 #include <netinet/ip.h>
89 #include <netinet/if_ether.h>
90 #include <netinet/tcp.h>
91 #include <netinet/udp.h>
92 #include <netinet/ip6.h>
95 #include <vm/vm_param.h>
96 #include <vm/vm_kern.h>
99 #include <machine/bus.h>
100 #include <machine/resource.h>
101 #include <machine/frame.h>
102 #include <machine/vmparam.h>
105 #include <sys/rman.h>
106 #include <sys/mutex.h>
107 #include <sys/errno.h>
108 #include <sys/types.h>
109 #include <machine/atomic.h>
111 #include <machine/intr_machdep.h>
113 #include <machine/in_cksum.h>
115 #include <dev/hyperv/include/hyperv.h>
116 #include "hv_net_vsc.h"
117 #include "hv_rndis.h"
118 #include "hv_rndis_filter.h"
121 /* Short for Hyper-V network interface */
122 #define NETVSC_DEVNAME "hn"
125 * It looks like offset 0 of buf is reserved to hold the softc pointer.
126 * The sc pointer evidently not needed, and is not presently populated.
127 * The packet offset is where the netvsc_packet starts in the buffer.
129 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
130 #define HV_NV_PACKET_OFFSET_IN_BUF 16
133 * A unified flag for all outbound check sum flags is useful,
134 * and it helps avoiding unnecessary check sum calculation in
135 * network forwarding scenario.
137 #define HV_CSUM_FOR_OUTBOUND \
138 (CSUM_IP|CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP|CSUM_IP_TSO| \
139 CSUM_IP_ISCSI|CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP| \
140 CSUM_IP6_TSO|CSUM_IP6_ISCSI)
142 /* XXX move to netinet/tcp_lro.h */
143 #define HN_LRO_HIWAT_MAX 65535
144 #define HN_LRO_HIWAT_DEF HN_LRO_HIWAT_MAX
145 /* YYY 2*MTU is a bit rough, but should be good enough. */
146 #define HN_LRO_HIWAT_MTULIM(ifp) (2 * (ifp)->if_mtu)
147 #define HN_LRO_HIWAT_ISVALID(sc, hiwat) \
148 ((hiwat) >= HN_LRO_HIWAT_MTULIM((sc)->hn_ifp) || \
149 (hiwat) <= HN_LRO_HIWAT_MAX)
152 * Be aware that this sleepable mutex will exhibit WITNESS errors when
153 * certain TCP and ARP code paths are taken. This appears to be a
154 * well-known condition, as all other drivers checked use a sleeping
155 * mutex to protect their transmit paths.
156 * Also Be aware that mutexes do not play well with semaphores, and there
157 * is a conflicting semaphore in a certain channel code path.
159 #define NV_LOCK_INIT(_sc, _name) \
160 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
161 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
162 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
163 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
164 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
171 int hv_promisc_mode = 0; /* normal mode by default */
173 /* Trust tcp segements verification on host side. */
174 static int hn_trust_hosttcp = 0;
175 TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp);
178 * Forward declarations
180 static void hn_stop(hn_softc_t *sc);
181 static void hn_ifinit_locked(hn_softc_t *sc);
182 static void hn_ifinit(void *xsc);
183 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
184 static int hn_start_locked(struct ifnet *ifp);
185 static void hn_start(struct ifnet *ifp);
186 static int hn_ifmedia_upd(struct ifnet *ifp);
187 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
189 static int hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS);
191 static int hn_check_iplen(const struct mbuf *, int);
194 hn_set_lro_hiwat(struct hn_softc *sc, int hiwat)
196 sc->hn_lro_hiwat = hiwat;
198 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
203 * NetVsc get message transport protocol type
205 static uint32_t get_transport_proto_type(struct mbuf *m_head)
207 uint32_t ret_val = TRANSPORT_TYPE_NOT_IP;
208 uint16_t ether_type = 0;
210 struct ether_vlan_header *eh;
218 eh = mtod(m_head, struct ether_vlan_header*);
219 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
220 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
221 ether_type = eh->evl_proto;
223 ether_len = ETHER_HDR_LEN;
224 ether_type = eh->evl_encap_proto;
227 switch (ntohs(ether_type)) {
230 ip6 = (struct ip6_hdr *)(m_head->m_data + ether_len);
232 if (IPPROTO_TCP == ip6->ip6_nxt) {
233 ret_val = TRANSPORT_TYPE_IPV6_TCP;
234 } else if (IPPROTO_UDP == ip6->ip6_nxt) {
235 ret_val = TRANSPORT_TYPE_IPV6_UDP;
241 iph = (struct ip *)(m_head->m_data + ether_len);
243 if (IPPROTO_TCP == iph->ip_p) {
244 ret_val = TRANSPORT_TYPE_IPV4_TCP;
245 } else if (IPPROTO_UDP == iph->ip_p) {
246 ret_val = TRANSPORT_TYPE_IPV4_UDP;
251 ret_val = TRANSPORT_TYPE_NOT_IP;
259 hn_ifmedia_upd(struct ifnet *ifp __unused)
266 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
268 struct hn_softc *sc = ifp->if_softc;
270 ifmr->ifm_status = IFM_AVALID;
271 ifmr->ifm_active = IFM_ETHER;
273 if (!sc->hn_carrier) {
274 ifmr->ifm_active |= IFM_NONE;
277 ifmr->ifm_status |= IFM_ACTIVE;
278 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
281 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
282 static const hv_guid g_net_vsc_device_type = {
283 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
284 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
288 * Standard probe entry point.
292 netvsc_probe(device_t dev)
296 p = vmbus_get_type(dev);
297 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
298 device_set_desc(dev, "Synthetic Network Interface");
300 printf("Netvsc probe... DONE \n");
302 return (BUS_PROBE_DEFAULT);
309 * Standard attach entry point.
311 * Called when the driver is loaded. It allocates needed resources,
312 * and initializes the "hardware" and software.
315 netvsc_attach(device_t dev)
317 struct hv_device *device_ctx = vmbus_get_devctx(dev);
318 netvsc_device_info device_info;
320 int unit = device_get_unit(dev);
322 struct sysctl_oid_list *child;
323 struct sysctl_ctx_list *ctx;
326 sc = device_get_softc(dev);
331 bzero(sc, sizeof(hn_softc_t));
334 sc->hn_lro_hiwat = HN_LRO_HIWAT_DEF;
335 sc->hn_trust_hosttcp = hn_trust_hosttcp;
337 NV_LOCK_INIT(sc, "NetVSCLock");
339 sc->hn_dev_obj = device_ctx;
341 ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
344 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
345 ifp->if_dunit = unit;
346 ifp->if_dname = NETVSC_DEVNAME;
348 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
349 ifp->if_ioctl = hn_ioctl;
350 ifp->if_start = hn_start;
351 ifp->if_init = hn_ifinit;
352 /* needed by hv_rf_on_device_add() code */
353 ifp->if_mtu = ETHERMTU;
354 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
355 ifp->if_snd.ifq_drv_maxlen = 511;
356 IFQ_SET_READY(&ifp->if_snd);
358 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
359 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
360 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
361 /* XXX ifmedia_set really should do this for us */
362 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
365 * Tell upper layers that we support full VLAN capability.
367 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
368 ifp->if_capabilities |=
369 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
372 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
375 * Only enable UDP checksum offloading when it is on 2012R2 or
376 * later. UDP checksum offloading doesn't work on earlier
379 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
380 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_TSO;
382 ifp->if_hwassist = CSUM_TCP | CSUM_TSO;
384 ret = hv_rf_on_device_add(device_ctx, &device_info);
390 if (device_info.link_state == 0) {
394 #if defined(INET) || defined(INET6)
395 tcp_lro_init(&sc->hn_lro);
396 /* Driver private LRO settings */
397 sc->hn_lro.ifp = ifp;
399 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
401 #endif /* INET || INET6 */
403 ether_ifattach(ifp, device_info.mac_addr);
405 ctx = device_get_sysctl_ctx(dev);
406 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
408 SYSCTL_ADD_U64(ctx, child, OID_AUTO, "lro_queued",
409 CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
410 SYSCTL_ADD_U64(ctx, child, OID_AUTO, "lro_flushed",
411 CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
412 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
413 CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
415 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_hiwat",
416 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_hiwat_sysctl,
417 "I", "LRO high watermark");
419 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "trust_hosttcp",
420 CTLFLAG_RW, &sc->hn_trust_hosttcp, 0,
421 "Trust tcp segement verification on host side, "
422 "when csum info is missing");
423 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
424 CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
425 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
426 CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
427 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
428 CTLFLAG_RW, &sc->hn_csum_trusted,
429 "# of TCP segements that we trust host's csum verification");
430 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
431 CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
434 struct sysctl_ctx_list *dc_ctx;
435 struct sysctl_oid_list *dc_child;
439 * Add sysctl nodes for devclass
441 dc = device_get_devclass(dev);
442 dc_ctx = devclass_get_sysctl_ctx(dc);
443 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
445 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
446 CTLFLAG_RD, &hn_trust_hosttcp, 0,
447 "Trust tcp segement verification on host side, "
448 "when csum info is missing (global setting)");
455 * Standard detach entry point
458 netvsc_detach(device_t dev)
460 struct hn_softc *sc = device_get_softc(dev);
461 struct hv_device *hv_device = vmbus_get_devctx(dev);
464 printf("netvsc_detach\n");
467 * XXXKYS: Need to clean up all our
468 * driver state; this is the driver
473 * XXXKYS: Need to stop outgoing traffic and unregister
477 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
479 ifmedia_removeall(&sc->hn_media);
480 #if defined(INET) || defined(INET6)
481 tcp_lro_free(&sc->hn_lro);
488 * Standard shutdown entry point
491 netvsc_shutdown(device_t dev)
497 * Send completion processing
499 * Note: It looks like offset 0 of buf is reserved to hold the softc
500 * pointer. The sc pointer is not currently needed in this function, and
501 * it is not presently populated by the TX function.
504 netvsc_xmit_completion(void *context)
506 netvsc_packet *packet = (netvsc_packet *)context;
510 mb = (struct mbuf *)(uintptr_t)packet->compl.send.send_completion_tid;
511 buf = ((uint8_t *)packet) - HV_NV_PACKET_OFFSET_IN_BUF;
521 * Start a transmit of one or more packets
524 hn_start_locked(struct ifnet *ifp)
526 hn_softc_t *sc = ifp->if_softc;
527 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
528 netvsc_dev *net_dev = sc->net_dev;
529 device_t dev = device_ctx->device;
531 netvsc_packet *packet;
532 struct mbuf *m_head, *m;
533 struct mbuf *mc_head = NULL;
534 struct ether_vlan_header *eh;
535 rndis_msg *rndis_mesg;
536 rndis_packet *rndis_pkt;
537 rndis_per_packet_info *rppi;
538 ndis_8021q_info *rppi_vlan_info;
539 rndis_tcp_ip_csum_info *csum_info;
540 rndis_tcp_tso_info *tso_info;
547 uint32_t rndis_msg_size = 0;
548 uint32_t trans_proto_type;
549 uint32_t send_buf_section_idx =
550 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
552 while (!IFQ_DRV_IS_EMPTY(&sc->hn_ifp->if_snd)) {
553 IFQ_DRV_DEQUEUE(&sc->hn_ifp->if_snd, m_head);
554 if (m_head == NULL) {
561 /* Walk the mbuf list computing total length and num frags */
562 for (m = m_head; m != NULL; m = m->m_next) {
570 * Reserve the number of pages requested. Currently,
571 * one page is reserved for the message in the RNDIS
574 num_frags += HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
576 /* If exceeds # page_buffers in netvsc_packet */
577 if (num_frags > NETVSC_PACKET_MAXPAGE) {
578 device_printf(dev, "exceed max page buffers,%d,%d\n",
579 num_frags, NETVSC_PACKET_MAXPAGE);
581 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
586 * Allocate a buffer with space for a netvsc packet plus a
587 * number of reserved areas. First comes a (currently 16
588 * bytes, currently unused) reserved data area. Second is
589 * the netvsc_packet. Third is an area reserved for an
590 * rndis_filter_packet struct. Fourth (optional) is a
591 * rndis_per_packet_info struct.
592 * Changed malloc to M_NOWAIT to avoid sleep under spin lock.
593 * No longer reserving extra space for page buffers, as they
594 * are already part of the netvsc_packet.
596 buf = malloc(HV_NV_PACKET_OFFSET_IN_BUF +
597 sizeof(netvsc_packet) +
599 RNDIS_VLAN_PPI_SIZE +
602 M_NETVSC, M_ZERO | M_NOWAIT);
604 device_printf(dev, "hn:malloc packet failed\n");
606 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
610 packet = (netvsc_packet *)(buf + HV_NV_PACKET_OFFSET_IN_BUF);
611 *(vm_offset_t *)buf = HV_NV_SC_PTR_OFFSET_IN_BUF;
613 packet->is_data_pkt = TRUE;
615 /* Set up the rndis header */
616 packet->page_buf_count = num_frags;
618 /* Initialize it from the mbuf */
619 packet->tot_data_buf_len = len;
622 * extension points to the area reserved for the
623 * rndis_filter_packet, which is placed just after
624 * the netvsc_packet (and rppi struct, if present;
625 * length is updated later).
627 packet->rndis_mesg = packet + 1;
628 rndis_mesg = (rndis_msg *)packet->rndis_mesg;
629 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
631 rndis_pkt = &rndis_mesg->msg.packet;
632 rndis_pkt->data_offset = sizeof(rndis_packet);
633 rndis_pkt->data_length = packet->tot_data_buf_len;
634 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
636 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
639 * If the Hyper-V infrastructure needs to embed a VLAN tag,
640 * initialize netvsc_packet and rppi struct values as needed.
642 if (m_head->m_flags & M_VLANTAG) {
644 * set up some additional fields so the Hyper-V infrastructure will stuff the VLAN tag
647 packet->vlan_tci = m_head->m_pkthdr.ether_vtag;
649 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
651 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
654 /* VLAN info immediately follows rppi struct */
655 rppi_vlan_info = (ndis_8021q_info *)((char*)rppi +
656 rppi->per_packet_info_offset);
657 /* FreeBSD does not support CFI or priority */
658 rppi_vlan_info->u1.s1.vlan_id =
659 packet->vlan_tci & 0xfff;
662 /* Only check the flags for outbound and ignore the ones for inbound */
663 if (0 == (m_head->m_pkthdr.csum_flags & HV_CSUM_FOR_OUTBOUND)) {
667 eh = mtod(m_head, struct ether_vlan_header*);
668 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
669 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
671 ether_len = ETHER_HDR_LEN;
674 trans_proto_type = get_transport_proto_type(m_head);
675 if (TRANSPORT_TYPE_NOT_IP == trans_proto_type) {
680 * TSO packet needless to setup the send side checksum
683 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
687 /* setup checksum offload */
688 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
689 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
691 csum_info = (rndis_tcp_ip_csum_info *)((char*)rppi +
692 rppi->per_packet_info_offset);
694 if (trans_proto_type & (TYPE_IPV4 << 16)) {
695 csum_info->xmit.is_ipv4 = 1;
697 csum_info->xmit.is_ipv6 = 1;
700 if (trans_proto_type & TYPE_TCP) {
701 csum_info->xmit.tcp_csum = 1;
702 csum_info->xmit.tcp_header_offset = 0;
703 } else if (trans_proto_type & TYPE_UDP) {
704 csum_info->xmit.udp_csum = 1;
710 /* setup TCP segmentation offload */
711 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
712 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
713 tcp_large_send_info);
715 tso_info = (rndis_tcp_tso_info *)((char *)rppi +
716 rppi->per_packet_info_offset);
717 tso_info->lso_v2_xmit.type =
718 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
721 if (trans_proto_type & (TYPE_IPV4 << 16)) {
723 (struct ip *)(m_head->m_data + ether_len);
724 unsigned long iph_len = ip->ip_hl << 2;
726 (struct tcphdr *)((caddr_t)ip + iph_len);
728 tso_info->lso_v2_xmit.ip_version =
729 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
733 th->th_sum = in_pseudo(ip->ip_src.s_addr,
738 #if defined(INET6) && defined(INET)
743 struct ip6_hdr *ip6 =
744 (struct ip6_hdr *)(m_head->m_data + ether_len);
745 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
747 tso_info->lso_v2_xmit.ip_version =
748 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
750 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
753 tso_info->lso_v2_xmit.tcp_header_offset = 0;
754 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
757 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
758 packet->tot_data_buf_len = rndis_mesg->msg_len;
760 /* send packet with send buffer */
761 if (packet->tot_data_buf_len < net_dev->send_section_size) {
762 send_buf_section_idx =
763 hv_nv_get_next_send_section(net_dev);
764 if (send_buf_section_idx !=
765 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
766 char *dest = ((char *)net_dev->send_buf +
767 send_buf_section_idx *
768 net_dev->send_section_size);
770 memcpy(dest, rndis_mesg, rndis_msg_size);
771 dest += rndis_msg_size;
772 for (m = m_head; m != NULL; m = m->m_next) {
775 (void *)mtod(m, vm_offset_t),
781 packet->send_buf_section_idx =
782 send_buf_section_idx;
783 packet->send_buf_section_size =
784 packet->tot_data_buf_len;
785 packet->page_buf_count = 0;
790 /* send packet with page buffer */
791 packet->page_buffers[0].pfn =
792 atop(hv_get_phys_addr(rndis_mesg));
793 packet->page_buffers[0].offset =
794 (unsigned long)rndis_mesg & PAGE_MASK;
795 packet->page_buffers[0].length = rndis_msg_size;
798 * Fill the page buffers with mbuf info starting at index
799 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
801 i = HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
802 for (m = m_head; m != NULL; m = m->m_next) {
805 vtophys(mtod(m, vm_offset_t));
806 packet->page_buffers[i].pfn =
808 packet->page_buffers[i].offset =
809 paddr & (PAGE_SIZE - 1);
810 packet->page_buffers[i].length = m->m_len;
815 packet->send_buf_section_idx =
816 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
817 packet->send_buf_section_size = 0;
822 * If bpf, copy the mbuf chain. This is less expensive than
823 * it appears; the mbuf clusters are not copied, only their
824 * reference counts are incremented.
825 * Needed to avoid a race condition where the completion
826 * callback is invoked, freeing the mbuf chain, before the
827 * bpf_mtap code has a chance to run.
830 mc_head = m_copypacket(m_head, M_NOWAIT);
833 /* Set the completion routine */
834 packet->compl.send.on_send_completion = netvsc_xmit_completion;
835 packet->compl.send.send_completion_context = packet;
836 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)m_head;
838 /* Removed critical_enter(), does not appear necessary */
839 ret = hv_nv_on_send(device_ctx, packet);
841 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
842 /* if bpf && mc_head, call bpf_mtap code */
844 ETHER_BPF_MTAP(ifp, mc_head);
852 IF_PREPEND(&ifp->if_snd, m_head);
853 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
856 * Null the mbuf pointer so the completion function
857 * does not free the mbuf chain. We just pushed the
858 * mbuf chain back on the if_snd queue.
860 packet->compl.send.send_completion_tid = 0;
863 * Release the resources since we will not get any
866 netvsc_xmit_completion(packet);
867 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
870 /* if bpf && mc_head, free the mbuf chain copy */
880 * Link up/down notification
883 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
885 hn_softc_t *sc = device_get_softc(device_obj->device);
899 * Append the specified data to the indicated mbuf chain,
900 * Extend the mbuf chain if the new data does not fit in
903 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
904 * There should be an equivalent in the kernel mbuf code,
905 * but there does not appear to be one yet.
907 * Differs from m_append() in that additional mbufs are
908 * allocated with cluster size MJUMPAGESIZE, and filled
911 * Return 1 if able to complete the job; otherwise 0.
914 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
917 int remainder, space;
919 for (m = m0; m->m_next != NULL; m = m->m_next)
922 space = M_TRAILINGSPACE(m);
925 * Copy into available space.
927 if (space > remainder)
929 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
934 while (remainder > 0) {
936 * Allocate a new mbuf; could check space
937 * and allocate a cluster instead.
939 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
942 n->m_len = min(MJUMPAGESIZE, remainder);
943 bcopy(cp, mtod(n, caddr_t), n->m_len);
945 remainder -= n->m_len;
949 if (m0->m_flags & M_PKTHDR)
950 m0->m_pkthdr.len += len - remainder;
952 return (remainder == 0);
957 * Called when we receive a data packet from the "wire" on the
960 * Note: This is no longer used as a callback
963 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
964 rndis_tcp_ip_csum_info *csum_info)
966 hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
969 device_t dev = device_ctx->device;
970 int size, do_lro = 0;
973 return (0); /* TODO: KYS how can this be! */
978 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
983 * Bail out if packet contains more data than configured MTU.
985 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
987 } else if (packet->tot_data_buf_len <= MHLEN) {
988 m_new = m_gethdr(M_NOWAIT, MT_DATA);
991 memcpy(mtod(m_new, void *), packet->data,
992 packet->tot_data_buf_len);
993 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
997 * Get an mbuf with a cluster. For packets 2K or less,
998 * get a standard 2K cluster. For anything larger, get a
999 * 4K cluster. Any buffers larger than 4K can cause problems
1000 * if looped around to the Hyper-V TX channel, so avoid them.
1003 if (packet->tot_data_buf_len > MCLBYTES) {
1005 size = MJUMPAGESIZE;
1008 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1009 if (m_new == NULL) {
1010 device_printf(dev, "alloc mbuf failed.\n");
1014 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1016 m_new->m_pkthdr.rcvif = ifp;
1018 /* receive side checksum offload */
1019 if (NULL != csum_info) {
1020 /* IP csum offload */
1021 if (csum_info->receive.ip_csum_succeeded) {
1022 m_new->m_pkthdr.csum_flags |=
1023 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1027 /* TCP csum offload */
1028 if (csum_info->receive.tcp_csum_succeeded) {
1029 m_new->m_pkthdr.csum_flags |=
1030 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1031 m_new->m_pkthdr.csum_data = 0xffff;
1035 if (csum_info->receive.ip_csum_succeeded &&
1036 csum_info->receive.tcp_csum_succeeded)
1039 const struct ether_header *eh;
1044 if (m_new->m_len < hoff)
1046 eh = mtod(m_new, struct ether_header *);
1047 etype = ntohs(eh->ether_type);
1048 if (etype == ETHERTYPE_VLAN) {
1049 const struct ether_vlan_header *evl;
1051 hoff = sizeof(*evl);
1052 if (m_new->m_len < hoff)
1054 evl = mtod(m_new, struct ether_vlan_header *);
1055 etype = ntohs(evl->evl_proto);
1058 if (etype == ETHERTYPE_IP) {
1061 pr = hn_check_iplen(m_new, hoff);
1062 if (pr == IPPROTO_TCP) {
1063 if (sc->hn_trust_hosttcp) {
1064 sc->hn_csum_trusted++;
1065 m_new->m_pkthdr.csum_flags |=
1066 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1067 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1068 m_new->m_pkthdr.csum_data = 0xffff;
1070 /* Rely on SW csum verification though... */
1076 if ((packet->vlan_tci != 0) &&
1077 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1078 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1079 m_new->m_flags |= M_VLANTAG;
1083 * Note: Moved RX completion back to hv_nv_on_receive() so all
1084 * messages (not just data messages) will trigger a response.
1087 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1089 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1090 #if defined(INET) || defined(INET6)
1091 struct lro_ctrl *lro = &sc->hn_lro;
1095 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1103 /* We're not holding the lock here, so don't release it */
1104 (*ifp->if_input)(ifp, m_new);
1110 netvsc_recv_rollup(struct hv_device *device_ctx)
1112 #if defined(INET) || defined(INET6)
1113 hn_softc_t *sc = device_get_softc(device_ctx->device);
1114 struct lro_ctrl *lro = &sc->hn_lro;
1115 struct lro_entry *queued;
1117 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
1118 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1119 tcp_lro_flush(lro, queued);
1125 * Rules for using sc->temp_unusable:
1126 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1127 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1128 * sc->temp_unusable set, must release NV_LOCK() and exit
1129 * 3. to retain exclusive control of the interface,
1130 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1131 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1132 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1136 * Standard ioctl entry point. Called when the user wants to configure
1140 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1142 hn_softc_t *sc = ifp->if_softc;
1143 struct ifreq *ifr = (struct ifreq *)data;
1145 struct ifaddr *ifa = (struct ifaddr *)data;
1147 netvsc_device_info device_info;
1148 struct hv_device *hn_dev;
1149 int mask, error = 0;
1150 int retry_cnt = 500;
1156 if (ifa->ifa_addr->sa_family == AF_INET) {
1157 ifp->if_flags |= IFF_UP;
1158 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1160 arp_ifinit(ifp, ifa);
1163 error = ether_ioctl(ifp, cmd, data);
1166 hn_dev = vmbus_get_devctx(sc->hn_dev);
1168 /* Check MTU value change */
1169 if (ifp->if_mtu == ifr->ifr_mtu)
1172 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1177 /* Obtain and record requested MTU */
1178 ifp->if_mtu = ifr->ifr_mtu;
1180 * Make sure that LRO high watermark is still valid,
1181 * after MTU change (the 2*MTU limit).
1183 if (!HN_LRO_HIWAT_ISVALID(sc, sc->hn_lro_hiwat))
1184 hn_set_lro_hiwat(sc, HN_LRO_HIWAT_MTULIM(ifp));
1188 if (!sc->temp_unusable) {
1189 sc->temp_unusable = TRUE;
1193 if (retry_cnt > 0) {
1197 } while (retry_cnt > 0);
1199 if (retry_cnt == 0) {
1204 /* We must remove and add back the device to cause the new
1205 * MTU to take effect. This includes tearing down, but not
1206 * deleting the channel, then bringing it back up.
1208 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1211 sc->temp_unusable = FALSE;
1215 error = hv_rf_on_device_add(hn_dev, &device_info);
1218 sc->temp_unusable = FALSE;
1223 hn_ifinit_locked(sc);
1226 sc->temp_unusable = FALSE;
1232 if (!sc->temp_unusable) {
1233 sc->temp_unusable = TRUE;
1237 if (retry_cnt > 0) {
1241 } while (retry_cnt > 0);
1243 if (retry_cnt == 0) {
1248 if (ifp->if_flags & IFF_UP) {
1250 * If only the state of the PROMISC flag changed,
1251 * then just use the 'set promisc mode' command
1252 * instead of reinitializing the entire NIC. Doing
1253 * a full re-init means reloading the firmware and
1254 * waiting for it to start up, which may take a
1258 /* Fixme: Promiscuous mode? */
1259 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1260 ifp->if_flags & IFF_PROMISC &&
1261 !(sc->hn_if_flags & IFF_PROMISC)) {
1262 /* do something here for Hyper-V */
1263 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1264 !(ifp->if_flags & IFF_PROMISC) &&
1265 sc->hn_if_flags & IFF_PROMISC) {
1266 /* do something here for Hyper-V */
1269 hn_ifinit_locked(sc);
1271 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1276 sc->temp_unusable = FALSE;
1278 sc->hn_if_flags = ifp->if_flags;
1282 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1283 if (mask & IFCAP_TXCSUM) {
1284 if (IFCAP_TXCSUM & ifp->if_capenable) {
1285 ifp->if_capenable &= ~IFCAP_TXCSUM;
1286 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
1288 ifp->if_capenable |= IFCAP_TXCSUM;
1290 * Only enable UDP checksum offloading on
1291 * Windows Server 2012R2 or later releases.
1293 if (hv_vmbus_protocal_version >=
1294 HV_VMBUS_VERSION_WIN8_1) {
1296 (CSUM_TCP | CSUM_UDP);
1298 ifp->if_hwassist |= CSUM_TCP;
1303 if (mask & IFCAP_RXCSUM) {
1304 if (IFCAP_RXCSUM & ifp->if_capenable) {
1305 ifp->if_capenable &= ~IFCAP_RXCSUM;
1307 ifp->if_capenable |= IFCAP_RXCSUM;
1310 if (mask & IFCAP_LRO)
1311 ifp->if_capenable ^= IFCAP_LRO;
1313 if (mask & IFCAP_TSO4) {
1314 ifp->if_capenable ^= IFCAP_TSO4;
1315 ifp->if_hwassist ^= CSUM_IP_TSO;
1318 if (mask & IFCAP_TSO6) {
1319 ifp->if_capenable ^= IFCAP_TSO6;
1320 ifp->if_hwassist ^= CSUM_IP6_TSO;
1328 /* Fixme: Multicast mode? */
1329 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1331 netvsc_setmulti(sc);
1340 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1343 error = ether_ioctl(ifp, cmd, data);
1354 hn_stop(hn_softc_t *sc)
1358 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1363 printf(" Closing Device ...\n");
1365 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1366 if_link_state_change(ifp, LINK_STATE_DOWN);
1367 sc->hn_initdone = 0;
1369 ret = hv_rf_on_close(device_ctx);
1373 * FreeBSD transmit entry point
1376 hn_start(struct ifnet *ifp)
1382 if (sc->temp_unusable) {
1386 hn_start_locked(ifp);
1394 hn_ifinit_locked(hn_softc_t *sc)
1397 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1402 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1406 hv_promisc_mode = 1;
1408 ret = hv_rf_on_open(device_ctx);
1412 sc->hn_initdone = 1;
1414 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1415 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1416 if_link_state_change(ifp, LINK_STATE_UP);
1423 hn_ifinit(void *xsc)
1425 hn_softc_t *sc = xsc;
1428 if (sc->temp_unusable) {
1432 sc->temp_unusable = TRUE;
1435 hn_ifinit_locked(sc);
1438 sc->temp_unusable = FALSE;
1447 hn_watchdog(struct ifnet *ifp)
1452 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1453 hn_ifinit(sc); /*???*/
1454 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1460 hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS)
1462 struct hn_softc *sc = arg1;
1465 hiwat = sc->hn_lro_hiwat;
1466 error = sysctl_handle_int(oidp, &hiwat, 0, req);
1467 if (error || req->newptr == NULL)
1470 if (!HN_LRO_HIWAT_ISVALID(sc, hiwat))
1473 if (sc->hn_lro_hiwat != hiwat)
1474 hn_set_lro_hiwat(sc, hiwat);
1477 #endif /* HN_LRO_HIWAT */
1480 hn_check_iplen(const struct mbuf *m, int hoff)
1482 const struct ip *ip;
1483 int len, iphlen, iplen;
1484 const struct tcphdr *th;
1485 int thoff; /* TCP data offset */
1487 len = hoff + sizeof(struct ip);
1489 /* The packet must be at least the size of an IP header. */
1490 if (m->m_pkthdr.len < len)
1491 return IPPROTO_DONE;
1493 /* The fixed IP header must reside completely in the first mbuf. */
1495 return IPPROTO_DONE;
1497 ip = mtodo(m, hoff);
1499 /* Bound check the packet's stated IP header length. */
1500 iphlen = ip->ip_hl << 2;
1501 if (iphlen < sizeof(struct ip)) /* minimum header length */
1502 return IPPROTO_DONE;
1504 /* The full IP header must reside completely in the one mbuf. */
1505 if (m->m_len < hoff + iphlen)
1506 return IPPROTO_DONE;
1508 iplen = ntohs(ip->ip_len);
1511 * Check that the amount of data in the buffers is as
1512 * at least much as the IP header would have us expect.
1514 if (m->m_pkthdr.len < hoff + iplen)
1515 return IPPROTO_DONE;
1518 * Ignore IP fragments.
1520 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1521 return IPPROTO_DONE;
1524 * The TCP/IP or UDP/IP header must be entirely contained within
1525 * the first fragment of a packet.
1529 if (iplen < iphlen + sizeof(struct tcphdr))
1530 return IPPROTO_DONE;
1531 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1532 return IPPROTO_DONE;
1533 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1534 thoff = th->th_off << 2;
1535 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1536 return IPPROTO_DONE;
1537 if (m->m_len < hoff + iphlen + thoff)
1538 return IPPROTO_DONE;
1541 if (iplen < iphlen + sizeof(struct udphdr))
1542 return IPPROTO_DONE;
1543 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1544 return IPPROTO_DONE;
1548 return IPPROTO_DONE;
1554 static device_method_t netvsc_methods[] = {
1555 /* Device interface */
1556 DEVMETHOD(device_probe, netvsc_probe),
1557 DEVMETHOD(device_attach, netvsc_attach),
1558 DEVMETHOD(device_detach, netvsc_detach),
1559 DEVMETHOD(device_shutdown, netvsc_shutdown),
1564 static driver_t netvsc_driver = {
1570 static devclass_t netvsc_devclass;
1572 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
1573 MODULE_VERSION(hn, 1);
1574 MODULE_DEPEND(hn, vmbus, 1, 1, 1);