xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision f0157ce528a128e2abb181a5c766033a2ce49a5f)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /* Driver for VirtIO network devices. */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #ifdef HAVE_KERNEL_OPTION_HEADERS
33 #include "opt_device_polling.h"
34 #endif
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/sockio.h>
40 #include <sys/mbuf.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_arp.h>
55 #include <net/if_dl.h>
56 #include <net/if_types.h>
57 #include <net/if_media.h>
58 #include <net/if_vlan_var.h>
59 
60 #include <net/bpf.h>
61 
62 #include <netinet/in_systm.h>
63 #include <netinet/in.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip6.h>
66 #include <netinet/udp.h>
67 #include <netinet/tcp.h>
68 #include <netinet/sctp.h>
69 
70 #include <machine/bus.h>
71 #include <machine/resource.h>
72 #include <sys/bus.h>
73 #include <sys/rman.h>
74 
75 #include <dev/virtio/virtio.h>
76 #include <dev/virtio/virtqueue.h>
77 #include <dev/virtio/network/virtio_net.h>
78 #include <dev/virtio/network/if_vtnetvar.h>
79 
80 #include "virtio_if.h"
81 
82 static int	vtnet_modevent(module_t, int, void *);
83 
84 static int	vtnet_probe(device_t);
85 static int	vtnet_attach(device_t);
86 static int	vtnet_detach(device_t);
87 static int	vtnet_suspend(device_t);
88 static int	vtnet_resume(device_t);
89 static int	vtnet_shutdown(device_t);
90 static int	vtnet_config_change(device_t);
91 
92 static void	vtnet_negotiate_features(struct vtnet_softc *);
93 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
94 static void	vtnet_get_hwaddr(struct vtnet_softc *);
95 static void	vtnet_set_hwaddr(struct vtnet_softc *);
96 static int	vtnet_is_link_up(struct vtnet_softc *);
97 static void	vtnet_update_link_status(struct vtnet_softc *);
98 static void	vtnet_watchdog(struct vtnet_softc *);
99 static int	vtnet_change_mtu(struct vtnet_softc *, int);
100 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
101 
102 static int	vtnet_init_rx_vq(struct vtnet_softc *);
103 static void	vtnet_free_rx_mbufs(struct vtnet_softc *);
104 static void	vtnet_free_tx_mbufs(struct vtnet_softc *);
105 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
106 
107 #ifdef DEVICE_POLLING
108 static poll_handler_t vtnet_poll;
109 #endif
110 
111 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int,
112 		    struct mbuf **);
113 static int	vtnet_replace_rxbuf(struct vtnet_softc *,
114 		    struct mbuf *, int);
115 static int	vtnet_newbuf(struct vtnet_softc *);
116 static void	vtnet_discard_merged_rxbuf(struct vtnet_softc *, int);
117 static void	vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *);
118 static int	vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *);
119 static void	vtnet_vlan_tag_remove(struct mbuf *);
120 static int	vtnet_rx_csum(struct vtnet_softc *, struct mbuf *,
121 		    struct virtio_net_hdr *);
122 static int	vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int);
123 static int	vtnet_rxeof(struct vtnet_softc *, int, int *);
124 static void	vtnet_rx_vq_intr(void *);
125 
126 static void	vtnet_txeof(struct vtnet_softc *);
127 static struct mbuf * vtnet_tx_offload(struct vtnet_softc *, struct mbuf *,
128 		    struct virtio_net_hdr *);
129 static int	vtnet_enqueue_txbuf(struct vtnet_softc *, struct mbuf **,
130 		    struct vtnet_tx_header *);
131 static int	vtnet_encap(struct vtnet_softc *, struct mbuf **);
132 static void	vtnet_start_locked(struct ifnet *);
133 static void	vtnet_start(struct ifnet *);
134 static void	vtnet_tick(void *);
135 static void	vtnet_tx_vq_intr(void *);
136 
137 static void	vtnet_stop(struct vtnet_softc *);
138 static int	vtnet_reinit(struct vtnet_softc *);
139 static void	vtnet_init_locked(struct vtnet_softc *);
140 static void	vtnet_init(void *);
141 
142 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
143 		    struct sglist *, int, int);
144 
145 static void	vtnet_rx_filter(struct vtnet_softc *sc);
146 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
147 static int	vtnet_set_promisc(struct vtnet_softc *, int);
148 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
149 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
150 
151 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
152 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
153 static void	vtnet_set_vlan_filter(struct vtnet_softc *, int, uint16_t);
154 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
155 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
156 
157 static int	vtnet_ifmedia_upd(struct ifnet *);
158 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
159 
160 static void	vtnet_add_statistics(struct vtnet_softc *);
161 
162 static int	vtnet_enable_rx_intr(struct vtnet_softc *);
163 static int	vtnet_enable_tx_intr(struct vtnet_softc *);
164 static void	vtnet_disable_rx_intr(struct vtnet_softc *);
165 static void	vtnet_disable_tx_intr(struct vtnet_softc *);
166 
167 /* Tunables. */
168 static int vtnet_csum_disable = 0;
169 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
170 static int vtnet_tso_disable = 0;
171 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
172 static int vtnet_lro_disable = 0;
173 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
174 
175 /*
176  * Reducing the number of transmit completed interrupts can
177  * improve performance. To do so, the define below keeps the
178  * Tx vq interrupt disabled and adds calls to vtnet_txeof()
179  * in the start and watchdog paths. The price to pay for this
180  * is the m_free'ing of transmitted mbufs may be delayed until
181  * the watchdog fires.
182  */
183 #define VTNET_TX_INTR_MODERATION
184 
185 static uma_zone_t vtnet_tx_header_zone;
186 
187 static struct virtio_feature_desc vtnet_feature_desc[] = {
188 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
189 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
190 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
191 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
192 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
193 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
194 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
195 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
196 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
197 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
198 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
199 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
200 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
201 	{ VIRTIO_NET_F_STATUS,		"Status"	},
202 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
203 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
204 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
205 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
206 
207 	{ 0, NULL }
208 };
209 
210 static device_method_t vtnet_methods[] = {
211 	/* Device methods. */
212 	DEVMETHOD(device_probe,		vtnet_probe),
213 	DEVMETHOD(device_attach,	vtnet_attach),
214 	DEVMETHOD(device_detach,	vtnet_detach),
215 	DEVMETHOD(device_suspend,	vtnet_suspend),
216 	DEVMETHOD(device_resume,	vtnet_resume),
217 	DEVMETHOD(device_shutdown,	vtnet_shutdown),
218 
219 	/* VirtIO methods. */
220 	DEVMETHOD(virtio_config_change, vtnet_config_change),
221 
222 	DEVMETHOD_END
223 };
224 
225 static driver_t vtnet_driver = {
226 	"vtnet",
227 	vtnet_methods,
228 	sizeof(struct vtnet_softc)
229 };
230 static devclass_t vtnet_devclass;
231 
232 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
233     vtnet_modevent, 0);
234 MODULE_VERSION(vtnet, 1);
235 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
236 
237 static int
238 vtnet_modevent(module_t mod, int type, void *unused)
239 {
240 	int error;
241 
242 	error = 0;
243 
244 	switch (type) {
245 	case MOD_LOAD:
246 		vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
247 		    sizeof(struct vtnet_tx_header),
248 		    NULL, NULL, NULL, NULL, 0, 0);
249 		break;
250 	case MOD_QUIESCE:
251 	case MOD_UNLOAD:
252 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
253 			error = EBUSY;
254 		else if (type == MOD_UNLOAD) {
255 			uma_zdestroy(vtnet_tx_header_zone);
256 			vtnet_tx_header_zone = NULL;
257 		}
258 		break;
259 	case MOD_SHUTDOWN:
260 		break;
261 	default:
262 		error = EOPNOTSUPP;
263 		break;
264 	}
265 
266 	return (error);
267 }
268 
269 static int
270 vtnet_probe(device_t dev)
271 {
272 
273 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
274 		return (ENXIO);
275 
276 	device_set_desc(dev, "VirtIO Networking Adapter");
277 
278 	return (BUS_PROBE_DEFAULT);
279 }
280 
281 static int
282 vtnet_attach(device_t dev)
283 {
284 	struct vtnet_softc *sc;
285 	struct ifnet *ifp;
286 	int tx_size, error;
287 
288 	sc = device_get_softc(dev);
289 	sc->vtnet_dev = dev;
290 
291 	VTNET_LOCK_INIT(sc);
292 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_MTX(sc), 0);
293 
294 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
295 	    vtnet_ifmedia_sts);
296 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
297 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
298 
299 	vtnet_add_statistics(sc);
300 
301 	virtio_set_feature_desc(dev, vtnet_feature_desc);
302 	vtnet_negotiate_features(sc);
303 
304 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
305 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
306 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
307 	} else
308 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
309 
310 	sc->vtnet_rx_mbuf_size = MCLBYTES;
311 	sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
312 
313 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
314 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
315 
316 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) {
317 			sc->vtnet_mac_filter = malloc(
318 			    sizeof(struct vtnet_mac_filter), M_DEVBUF,
319 			    M_NOWAIT | M_ZERO);
320 			if (sc->vtnet_mac_filter == NULL) {
321 				device_printf(dev,
322 				    "cannot allocate mac filter table\n");
323 				error = ENOMEM;
324 				goto fail;
325 			}
326 
327 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
328 		}
329 
330 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
331 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
332 	}
333 
334 	vtnet_get_hwaddr(sc);
335 
336 	error = vtnet_alloc_virtqueues(sc);
337 	if (error) {
338 		device_printf(dev, "cannot allocate virtqueues\n");
339 		goto fail;
340 	}
341 
342 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
343 	if (ifp == NULL) {
344 		device_printf(dev, "cannot allocate ifnet structure\n");
345 		error = ENOSPC;
346 		goto fail;
347 	}
348 
349 	ifp->if_softc = sc;
350 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
351 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
352 	ifp->if_init = vtnet_init;
353 	ifp->if_start = vtnet_start;
354 	ifp->if_ioctl = vtnet_ioctl;
355 
356 	sc->vtnet_rx_size = virtqueue_size(sc->vtnet_rx_vq);
357 	sc->vtnet_rx_process_limit = sc->vtnet_rx_size;
358 
359 	tx_size = virtqueue_size(sc->vtnet_tx_vq);
360 	sc->vtnet_tx_size = tx_size;
361 	IFQ_SET_MAXLEN(&ifp->if_snd, tx_size - 1);
362 	ifp->if_snd.ifq_drv_maxlen = tx_size - 1;
363 	IFQ_SET_READY(&ifp->if_snd);
364 
365 	ether_ifattach(ifp, sc->vtnet_hwaddr);
366 
367 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
368 		ifp->if_capabilities |= IFCAP_LINKSTATE;
369 
370 	/* Tell the upper layer(s) we support long frames. */
371 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
372 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
373 
374 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
375 		ifp->if_capabilities |= IFCAP_TXCSUM;
376 
377 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
378 			ifp->if_capabilities |= IFCAP_TSO4;
379 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
380 			ifp->if_capabilities |= IFCAP_TSO6;
381 		if (ifp->if_capabilities & IFCAP_TSO)
382 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
383 
384 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
385 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
386 	}
387 
388 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
389 		ifp->if_capabilities |= IFCAP_RXCSUM;
390 
391 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
392 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
393 			ifp->if_capabilities |= IFCAP_LRO;
394 	}
395 
396 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
397 		/*
398 		 * VirtIO does not support VLAN tagging, but we can fake
399 		 * it by inserting and removing the 802.1Q header during
400 		 * transmit and receive. We are then able to do checksum
401 		 * offloading of VLAN frames.
402 		 */
403 		ifp->if_capabilities |=
404 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
405 	}
406 
407 	ifp->if_capenable = ifp->if_capabilities;
408 
409 	/*
410 	 * Capabilities after here are not enabled by default.
411 	 */
412 
413 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
414 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
415 
416 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
417 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
418 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
419 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
420 	}
421 
422 #ifdef DEVICE_POLLING
423 	ifp->if_capabilities |= IFCAP_POLLING;
424 #endif
425 
426 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
427 	if (error) {
428 		device_printf(dev, "cannot setup virtqueue interrupts\n");
429 		ether_ifdetach(ifp);
430 		goto fail;
431 	}
432 
433 	/*
434 	 * Device defaults to promiscuous mode for backwards
435 	 * compatibility. Turn it off if possible.
436 	 */
437 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
438 		VTNET_LOCK(sc);
439 		if (vtnet_set_promisc(sc, 0) != 0) {
440 			ifp->if_flags |= IFF_PROMISC;
441 			device_printf(dev,
442 			    "cannot disable promiscuous mode\n");
443 		}
444 		VTNET_UNLOCK(sc);
445 	} else
446 		ifp->if_flags |= IFF_PROMISC;
447 
448 fail:
449 	if (error)
450 		vtnet_detach(dev);
451 
452 	return (error);
453 }
454 
455 static int
456 vtnet_detach(device_t dev)
457 {
458 	struct vtnet_softc *sc;
459 	struct ifnet *ifp;
460 
461 	sc = device_get_softc(dev);
462 	ifp = sc->vtnet_ifp;
463 
464 	KASSERT(mtx_initialized(VTNET_MTX(sc)),
465 	    ("vtnet mutex not initialized"));
466 
467 #ifdef DEVICE_POLLING
468 	if (ifp != NULL && ifp->if_capenable & IFCAP_POLLING)
469 		ether_poll_deregister(ifp);
470 #endif
471 
472 	if (device_is_attached(dev)) {
473 		VTNET_LOCK(sc);
474 		vtnet_stop(sc);
475 		VTNET_UNLOCK(sc);
476 
477 		callout_drain(&sc->vtnet_tick_ch);
478 
479 		ether_ifdetach(ifp);
480 	}
481 
482 	if (sc->vtnet_vlan_attach != NULL) {
483 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
484 		sc->vtnet_vlan_attach = NULL;
485 	}
486 	if (sc->vtnet_vlan_detach != NULL) {
487 		EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
488 		sc->vtnet_vlan_detach = NULL;
489 	}
490 
491 	if (sc->vtnet_mac_filter != NULL) {
492 		free(sc->vtnet_mac_filter, M_DEVBUF);
493 		sc->vtnet_mac_filter = NULL;
494 	}
495 
496 	if (ifp != NULL) {
497 		if_free(ifp);
498 		sc->vtnet_ifp = NULL;
499 	}
500 
501 	if (sc->vtnet_rx_vq != NULL)
502 		vtnet_free_rx_mbufs(sc);
503 	if (sc->vtnet_tx_vq != NULL)
504 		vtnet_free_tx_mbufs(sc);
505 	if (sc->vtnet_ctrl_vq != NULL)
506 		vtnet_free_ctrl_vq(sc);
507 
508 	ifmedia_removeall(&sc->vtnet_media);
509 	VTNET_LOCK_DESTROY(sc);
510 
511 	return (0);
512 }
513 
514 static int
515 vtnet_suspend(device_t dev)
516 {
517 	struct vtnet_softc *sc;
518 
519 	sc = device_get_softc(dev);
520 
521 	VTNET_LOCK(sc);
522 	vtnet_stop(sc);
523 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
524 	VTNET_UNLOCK(sc);
525 
526 	return (0);
527 }
528 
529 static int
530 vtnet_resume(device_t dev)
531 {
532 	struct vtnet_softc *sc;
533 	struct ifnet *ifp;
534 
535 	sc = device_get_softc(dev);
536 	ifp = sc->vtnet_ifp;
537 
538 	VTNET_LOCK(sc);
539 	if (ifp->if_flags & IFF_UP)
540 		vtnet_init_locked(sc);
541 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
542 	VTNET_UNLOCK(sc);
543 
544 	return (0);
545 }
546 
547 static int
548 vtnet_shutdown(device_t dev)
549 {
550 
551 	/*
552 	 * Suspend already does all of what we need to
553 	 * do here; we just never expect to be resumed.
554 	 */
555 	return (vtnet_suspend(dev));
556 }
557 
558 static int
559 vtnet_config_change(device_t dev)
560 {
561 	struct vtnet_softc *sc;
562 
563 	sc = device_get_softc(dev);
564 
565 	VTNET_LOCK(sc);
566 	vtnet_update_link_status(sc);
567 	VTNET_UNLOCK(sc);
568 
569 	return (0);
570 }
571 
572 static void
573 vtnet_negotiate_features(struct vtnet_softc *sc)
574 {
575 	device_t dev;
576 	uint64_t mask, features;
577 
578 	dev = sc->vtnet_dev;
579 	mask = 0;
580 
581 	if (vtnet_csum_disable)
582 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
583 
584 	/*
585 	 * TSO and LRO are only available when their corresponding
586 	 * checksum offload feature is also negotiated.
587 	 */
588 
589 	if (vtnet_csum_disable || vtnet_tso_disable)
590 		mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
591 		    VIRTIO_NET_F_HOST_ECN;
592 
593 	if (vtnet_csum_disable || vtnet_lro_disable)
594 		mask |= VTNET_LRO_FEATURES;
595 
596 	features = VTNET_FEATURES & ~mask;
597 #ifdef VTNET_TX_INTR_MODERATION
598 	features |= VIRTIO_F_NOTIFY_ON_EMPTY;
599 #endif
600 	sc->vtnet_features = virtio_negotiate_features(dev, features);
601 
602 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0 &&
603 	    virtio_with_feature(dev, VTNET_LRO_FEATURES)) {
604 		/*
605 		 * LRO without mergeable buffers requires special care. This
606 		 * is not ideal because every receive buffer must be large
607 		 * enough to hold the maximum TCP packet, the Ethernet header,
608 		 * and the vtnet_rx_header. This requires up to 34 descriptors
609 		 * when using MCLBYTES clusters. If we do not have indirect
610 		 * descriptors, LRO is disabled since the virtqueue will not
611 		 * be able to contain very many receive buffers.
612 		 */
613 		if (virtio_with_feature(dev,
614 		    VIRTIO_RING_F_INDIRECT_DESC) == 0) {
615 			device_printf(dev,
616 			    "LRO disabled due to lack of both mergeable "
617 			    "buffers and indirect descriptors\n");
618 
619 			sc->vtnet_features = virtio_negotiate_features(dev,
620 			    features & ~VTNET_LRO_FEATURES);
621 		} else
622 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
623 	}
624 }
625 
626 static int
627 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
628 {
629 	device_t dev;
630 	struct vq_alloc_info vq_info[3];
631 	int nvqs, rxsegs;
632 
633 	dev = sc->vtnet_dev;
634 	nvqs = 2;
635 
636 	/*
637 	 * Indirect descriptors are not needed for the Rx
638 	 * virtqueue when mergeable buffers are negotiated.
639 	 * The header is placed inline with the data, not
640 	 * in a separate descriptor, and mbuf clusters are
641 	 * always physically contiguous.
642 	 */
643 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
644 		rxsegs = sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ?
645 		    VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
646 	} else
647 		rxsegs = 0;
648 
649 	VQ_ALLOC_INFO_INIT(&vq_info[0], rxsegs,
650 	    vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
651 	    "%s receive", device_get_nameunit(dev));
652 
653 	VQ_ALLOC_INFO_INIT(&vq_info[1], VTNET_MAX_TX_SEGS,
654 	    vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
655 	    "%s transmit", device_get_nameunit(dev));
656 
657 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
658 		nvqs++;
659 
660 		VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
661 		    &sc->vtnet_ctrl_vq, "%s control",
662 		    device_get_nameunit(dev));
663 	}
664 
665 	return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
666 }
667 
668 static void
669 vtnet_get_hwaddr(struct vtnet_softc *sc)
670 {
671 	device_t dev;
672 
673 	dev = sc->vtnet_dev;
674 
675 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
676 		virtio_read_device_config(dev,
677 		    offsetof(struct virtio_net_config, mac),
678 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
679 	} else {
680 		/* Generate random locally administered unicast address. */
681 		sc->vtnet_hwaddr[0] = 0xB2;
682 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
683 
684 		vtnet_set_hwaddr(sc);
685 	}
686 }
687 
688 static void
689 vtnet_set_hwaddr(struct vtnet_softc *sc)
690 {
691 	device_t dev;
692 
693 	dev = sc->vtnet_dev;
694 
695 	virtio_write_device_config(dev,
696 	    offsetof(struct virtio_net_config, mac),
697 	    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
698 }
699 
700 static int
701 vtnet_is_link_up(struct vtnet_softc *sc)
702 {
703 	device_t dev;
704 	struct ifnet *ifp;
705 	uint16_t status;
706 
707 	dev = sc->vtnet_dev;
708 	ifp = sc->vtnet_ifp;
709 
710 	VTNET_LOCK_ASSERT(sc);
711 
712 	if ((ifp->if_capenable & IFCAP_LINKSTATE) == 0)
713 		return (1);
714 
715 	status = virtio_read_dev_config_2(dev,
716 	    offsetof(struct virtio_net_config, status));
717 
718 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
719 }
720 
721 static void
722 vtnet_update_link_status(struct vtnet_softc *sc)
723 {
724 	struct ifnet *ifp;
725 	int link;
726 
727 	ifp = sc->vtnet_ifp;
728 
729 	link = vtnet_is_link_up(sc);
730 
731 	if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
732 		sc->vtnet_flags |= VTNET_FLAG_LINK;
733 		if_link_state_change(ifp, LINK_STATE_UP);
734 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
735 			vtnet_start_locked(ifp);
736 	} else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
737 		sc->vtnet_flags &= ~VTNET_FLAG_LINK;
738 		if_link_state_change(ifp, LINK_STATE_DOWN);
739 	}
740 }
741 
742 static void
743 vtnet_watchdog(struct vtnet_softc *sc)
744 {
745 	struct ifnet *ifp;
746 
747 	ifp = sc->vtnet_ifp;
748 
749 #ifdef VTNET_TX_INTR_MODERATION
750 	vtnet_txeof(sc);
751 #endif
752 
753 	if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
754 		return;
755 
756 	if_printf(ifp, "watchdog timeout -- resetting\n");
757 #ifdef VTNET_DEBUG
758 	virtqueue_dump(sc->vtnet_tx_vq);
759 #endif
760 	ifp->if_oerrors++;
761 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
762 	vtnet_init_locked(sc);
763 }
764 
765 static int
766 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
767 {
768 	struct vtnet_softc *sc;
769 	struct ifreq *ifr;
770 	int reinit, mask, error;
771 
772 	sc = ifp->if_softc;
773 	ifr = (struct ifreq *) data;
774 	reinit = 0;
775 	error = 0;
776 
777 	switch (cmd) {
778 	case SIOCSIFMTU:
779 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
780 			error = EINVAL;
781 		else if (ifp->if_mtu != ifr->ifr_mtu) {
782 			VTNET_LOCK(sc);
783 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
784 			VTNET_UNLOCK(sc);
785 		}
786 		break;
787 
788 	case SIOCSIFFLAGS:
789 		VTNET_LOCK(sc);
790 		if ((ifp->if_flags & IFF_UP) == 0) {
791 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
792 				vtnet_stop(sc);
793 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
794 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
795 			    (IFF_PROMISC | IFF_ALLMULTI)) {
796 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
797 					vtnet_rx_filter(sc);
798 				else
799 					error = ENOTSUP;
800 			}
801 		} else
802 			vtnet_init_locked(sc);
803 
804 		if (error == 0)
805 			sc->vtnet_if_flags = ifp->if_flags;
806 		VTNET_UNLOCK(sc);
807 		break;
808 
809 	case SIOCADDMULTI:
810 	case SIOCDELMULTI:
811 		VTNET_LOCK(sc);
812 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
813 		    (ifp->if_drv_flags & IFF_DRV_RUNNING))
814 			vtnet_rx_filter_mac(sc);
815 		VTNET_UNLOCK(sc);
816 		break;
817 
818 	case SIOCSIFMEDIA:
819 	case SIOCGIFMEDIA:
820 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
821 		break;
822 
823 	case SIOCSIFCAP:
824 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
825 
826 #ifdef DEVICE_POLLING
827 		if (mask & IFCAP_POLLING) {
828 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
829 				error = ether_poll_register(vtnet_poll, ifp);
830 				if (error)
831 					break;
832 
833 				VTNET_LOCK(sc);
834 				vtnet_disable_rx_intr(sc);
835 				vtnet_disable_tx_intr(sc);
836 				ifp->if_capenable |= IFCAP_POLLING;
837 				VTNET_UNLOCK(sc);
838 			} else {
839 				error = ether_poll_deregister(ifp);
840 
841 				/* Enable interrupts even in error case. */
842 				VTNET_LOCK(sc);
843 				vtnet_enable_tx_intr(sc);
844 				vtnet_enable_rx_intr(sc);
845 				ifp->if_capenable &= ~IFCAP_POLLING;
846 				VTNET_UNLOCK(sc);
847 			}
848 		}
849 #endif
850 		VTNET_LOCK(sc);
851 
852 		if (mask & IFCAP_TXCSUM) {
853 			ifp->if_capenable ^= IFCAP_TXCSUM;
854 			if (ifp->if_capenable & IFCAP_TXCSUM)
855 				ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
856 			else
857 				ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
858 		}
859 
860 		if (mask & IFCAP_TSO4) {
861 			ifp->if_capenable ^= IFCAP_TSO4;
862 			if (ifp->if_capenable & IFCAP_TSO4)
863 				ifp->if_hwassist |= CSUM_TSO;
864 			else
865 				ifp->if_hwassist &= ~CSUM_TSO;
866 		}
867 
868 		if (mask & IFCAP_RXCSUM) {
869 			ifp->if_capenable ^= IFCAP_RXCSUM;
870 			reinit = 1;
871 		}
872 
873 		if (mask & IFCAP_LRO) {
874 			ifp->if_capenable ^= IFCAP_LRO;
875 			reinit = 1;
876 		}
877 
878 		if (mask & IFCAP_VLAN_HWFILTER) {
879 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
880 			reinit = 1;
881 		}
882 
883 		if (mask & IFCAP_VLAN_HWTSO)
884 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
885 
886 		if (mask & IFCAP_VLAN_HWTAGGING)
887 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
888 
889 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
890 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
891 			vtnet_init_locked(sc);
892 		}
893 		VLAN_CAPABILITIES(ifp);
894 
895 		VTNET_UNLOCK(sc);
896 		break;
897 
898 	default:
899 		error = ether_ioctl(ifp, cmd, data);
900 		break;
901 	}
902 
903 	VTNET_LOCK_ASSERT_NOTOWNED(sc);
904 
905 	return (error);
906 }
907 
908 static int
909 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
910 {
911 	struct ifnet *ifp;
912 	int new_frame_size, clsize;
913 
914 	ifp = sc->vtnet_ifp;
915 
916 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
917 		new_frame_size = sizeof(struct vtnet_rx_header) +
918 		    sizeof(struct ether_vlan_header) + new_mtu;
919 
920 		if (new_frame_size > MJUM9BYTES)
921 			return (EINVAL);
922 
923 		if (new_frame_size <= MCLBYTES)
924 			clsize = MCLBYTES;
925 		else
926 			clsize = MJUM9BYTES;
927 	} else {
928 		new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
929 		    sizeof(struct ether_vlan_header) + new_mtu;
930 
931 		if (new_frame_size <= MCLBYTES)
932 			clsize = MCLBYTES;
933 		else
934 			clsize = MJUMPAGESIZE;
935 	}
936 
937 	sc->vtnet_rx_mbuf_size = clsize;
938 	sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
939 	KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
940 	    ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
941 
942 	ifp->if_mtu = new_mtu;
943 
944 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
945 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
946 		vtnet_init_locked(sc);
947 	}
948 
949 	return (0);
950 }
951 
952 static int
953 vtnet_init_rx_vq(struct vtnet_softc *sc)
954 {
955 	struct virtqueue *vq;
956 	int nbufs, error;
957 
958 	vq = sc->vtnet_rx_vq;
959 	nbufs = 0;
960 	error = ENOSPC;
961 
962 	while (!virtqueue_full(vq)) {
963 		if ((error = vtnet_newbuf(sc)) != 0)
964 			break;
965 		nbufs++;
966 	}
967 
968 	if (nbufs > 0) {
969 		virtqueue_notify(vq);
970 
971 		/*
972 		 * EMSGSIZE signifies the virtqueue did not have enough
973 		 * entries available to hold the last mbuf. This is not
974 		 * an error. We should not get ENOSPC since we check if
975 		 * the virtqueue is full before attempting to add a
976 		 * buffer.
977 		 */
978 		if (error == EMSGSIZE)
979 			error = 0;
980 	}
981 
982 	return (error);
983 }
984 
985 static void
986 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
987 {
988 	struct virtqueue *vq;
989 	struct mbuf *m;
990 	int last;
991 
992 	vq = sc->vtnet_rx_vq;
993 	last = 0;
994 
995 	while ((m = virtqueue_drain(vq, &last)) != NULL)
996 		m_freem(m);
997 
998 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
999 }
1000 
1001 static void
1002 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1003 {
1004 	struct virtqueue *vq;
1005 	struct vtnet_tx_header *txhdr;
1006 	int last;
1007 
1008 	vq = sc->vtnet_tx_vq;
1009 	last = 0;
1010 
1011 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1012 		m_freem(txhdr->vth_mbuf);
1013 		uma_zfree(vtnet_tx_header_zone, txhdr);
1014 	}
1015 
1016 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1017 }
1018 
1019 static void
1020 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1021 {
1022 
1023 	/*
1024 	 * The control virtqueue is only polled, therefore
1025 	 * it should already be empty.
1026 	 */
1027 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1028 	    ("Ctrl Vq not empty"));
1029 }
1030 
1031 #ifdef DEVICE_POLLING
1032 static int
1033 vtnet_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1034 {
1035 	struct vtnet_softc *sc;
1036 	int rx_done;
1037 
1038 	sc = ifp->if_softc;
1039 	rx_done = 0;
1040 
1041 	VTNET_LOCK(sc);
1042 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1043 		if (cmd == POLL_AND_CHECK_STATUS)
1044 			vtnet_update_link_status(sc);
1045 
1046 		if (virtqueue_nused(sc->vtnet_rx_vq) > 0)
1047 			vtnet_rxeof(sc, count, &rx_done);
1048 
1049 		vtnet_txeof(sc);
1050 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1051 			vtnet_start_locked(ifp);
1052 	}
1053 	VTNET_UNLOCK(sc);
1054 
1055 	return (rx_done);
1056 }
1057 #endif /* DEVICE_POLLING */
1058 
1059 static struct mbuf *
1060 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1061 {
1062 	struct mbuf *m_head, *m_tail, *m;
1063 	int i, clsize;
1064 
1065 	clsize = sc->vtnet_rx_mbuf_size;
1066 
1067 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1068 	if (m_head == NULL)
1069 		goto fail;
1070 
1071 	m_head->m_len = clsize;
1072 	m_tail = m_head;
1073 
1074 	if (nbufs > 1) {
1075 		KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1076 		    ("chained Rx mbuf requested without LRO_NOMRG"));
1077 
1078 		for (i = 1; i < nbufs; i++) {
1079 			m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1080 			if (m == NULL)
1081 				goto fail;
1082 
1083 			m->m_len = clsize;
1084 			m_tail->m_next = m;
1085 			m_tail = m;
1086 		}
1087 	}
1088 
1089 	if (m_tailp != NULL)
1090 		*m_tailp = m_tail;
1091 
1092 	return (m_head);
1093 
1094 fail:
1095 	sc->vtnet_stats.mbuf_alloc_failed++;
1096 	m_freem(m_head);
1097 
1098 	return (NULL);
1099 }
1100 
1101 static int
1102 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1103 {
1104 	struct mbuf *m, *m_prev;
1105 	struct mbuf *m_new, *m_tail;
1106 	int len, clsize, nreplace, error;
1107 
1108 	m = m0;
1109 	m_prev = NULL;
1110 	len = len0;
1111 
1112 	m_tail = NULL;
1113 	clsize = sc->vtnet_rx_mbuf_size;
1114 	nreplace = 0;
1115 
1116 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1117 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1118 
1119 	/*
1120 	 * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1121 	 * allocating an entire chain for each received frame. When
1122 	 * the received frame's length is less than that of the chain,
1123 	 * the unused mbufs are reassigned to the new chain.
1124 	 */
1125 	while (len > 0) {
1126 		/*
1127 		 * Something is seriously wrong if we received
1128 		 * a frame larger than the mbuf chain. Drop it.
1129 		 */
1130 		if (m == NULL) {
1131 			sc->vtnet_stats.rx_frame_too_large++;
1132 			return (EMSGSIZE);
1133 		}
1134 
1135 		KASSERT(m->m_len == clsize,
1136 		    ("mbuf length not expected cluster size: %d",
1137 		    m->m_len));
1138 
1139 		m->m_len = MIN(m->m_len, len);
1140 		len -= m->m_len;
1141 
1142 		m_prev = m;
1143 		m = m->m_next;
1144 		nreplace++;
1145 	}
1146 
1147 	KASSERT(m_prev != NULL, ("m_prev == NULL"));
1148 	KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1149 	    ("too many replacement mbufs: %d/%d", nreplace,
1150 	    sc->vtnet_rx_mbuf_count));
1151 
1152 	m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1153 	if (m_new == NULL) {
1154 		m_prev->m_len = clsize;
1155 		return (ENOBUFS);
1156 	}
1157 
1158 	/*
1159 	 * Move unused mbufs, if any, from the original chain
1160 	 * onto the end of the new chain.
1161 	 */
1162 	if (m_prev->m_next != NULL) {
1163 		m_tail->m_next = m_prev->m_next;
1164 		m_prev->m_next = NULL;
1165 	}
1166 
1167 	error = vtnet_enqueue_rxbuf(sc, m_new);
1168 	if (error) {
1169 		/*
1170 		 * BAD! We could not enqueue the replacement mbuf chain. We
1171 		 * must restore the m0 chain to the original state if it was
1172 		 * modified so we can subsequently discard it.
1173 		 *
1174 		 * NOTE: The replacement is suppose to be an identical copy
1175 		 * to the one just dequeued so this is an unexpected error.
1176 		 */
1177 		sc->vtnet_stats.rx_enq_replacement_failed++;
1178 
1179 		if (m_tail->m_next != NULL) {
1180 			m_prev->m_next = m_tail->m_next;
1181 			m_tail->m_next = NULL;
1182 		}
1183 
1184 		m_prev->m_len = clsize;
1185 		m_freem(m_new);
1186 	}
1187 
1188 	return (error);
1189 }
1190 
1191 static int
1192 vtnet_newbuf(struct vtnet_softc *sc)
1193 {
1194 	struct mbuf *m;
1195 	int error;
1196 
1197 	m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1198 	if (m == NULL)
1199 		return (ENOBUFS);
1200 
1201 	error = vtnet_enqueue_rxbuf(sc, m);
1202 	if (error)
1203 		m_freem(m);
1204 
1205 	return (error);
1206 }
1207 
1208 static void
1209 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1210 {
1211 	struct virtqueue *vq;
1212 	struct mbuf *m;
1213 
1214 	vq = sc->vtnet_rx_vq;
1215 
1216 	while (--nbufs > 0) {
1217 		if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1218 			break;
1219 		vtnet_discard_rxbuf(sc, m);
1220 	}
1221 }
1222 
1223 static void
1224 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1225 {
1226 	int error;
1227 
1228 	/*
1229 	 * Requeue the discarded mbuf. This should always be
1230 	 * successful since it was just dequeued.
1231 	 */
1232 	error = vtnet_enqueue_rxbuf(sc, m);
1233 	KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1234 }
1235 
1236 static int
1237 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1238 {
1239 	struct sglist sg;
1240 	struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1241 	struct vtnet_rx_header *rxhdr;
1242 	struct virtio_net_hdr *hdr;
1243 	uint8_t *mdata;
1244 	int offset, error;
1245 
1246 	VTNET_LOCK_ASSERT(sc);
1247 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1248 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1249 
1250 	sglist_init(&sg, VTNET_MAX_RX_SEGS, segs);
1251 
1252 	mdata = mtod(m, uint8_t *);
1253 	offset = 0;
1254 
1255 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1256 		rxhdr = (struct vtnet_rx_header *) mdata;
1257 		hdr = &rxhdr->vrh_hdr;
1258 		offset += sizeof(struct vtnet_rx_header);
1259 
1260 		error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1261 		KASSERT(error == 0, ("cannot add header to sglist"));
1262 	}
1263 
1264 	error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1265 	if (error)
1266 		return (error);
1267 
1268 	if (m->m_next != NULL) {
1269 		error = sglist_append_mbuf(&sg, m->m_next);
1270 		if (error)
1271 			return (error);
1272 	}
1273 
1274 	return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1275 }
1276 
1277 static void
1278 vtnet_vlan_tag_remove(struct mbuf *m)
1279 {
1280 	struct ether_vlan_header *evl;
1281 
1282 	evl = mtod(m, struct ether_vlan_header *);
1283 
1284 	m->m_pkthdr.ether_vtag = ntohs(evl->evl_tag);
1285 	m->m_flags |= M_VLANTAG;
1286 
1287 	/* Strip the 802.1Q header. */
1288 	bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1289 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
1290 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
1291 }
1292 
1293 #ifdef notyet
1294 static int
1295 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1296     struct virtio_net_hdr *hdr)
1297 {
1298 	struct ether_header *eh;
1299 	struct ether_vlan_header *evh;
1300 	struct ip *ip;
1301 	struct ip6_hdr *ip6;
1302 	struct udphdr *udp;
1303 	int ip_offset, csum_start, csum_offset, hlen;
1304 	uint16_t eth_type;
1305 	uint8_t ip_proto;
1306 
1307 	/*
1308 	 * Convert the VirtIO checksum interface to FreeBSD's interface.
1309 	 * The host only provides us with the offset at which to start
1310 	 * checksumming, and the offset from that to place the completed
1311 	 * checksum. While this maps well with how Linux does checksums,
1312 	 * for FreeBSD, we must parse the received packet in order to set
1313 	 * the appropriate CSUM_* flags.
1314 	 */
1315 
1316 	/*
1317 	 * Every mbuf added to the receive virtqueue is always at least
1318 	 * MCLBYTES big, so assume something is amiss if the first mbuf
1319 	 * does not contain both the Ethernet and protocol headers.
1320 	 */
1321 	ip_offset = sizeof(struct ether_header);
1322 	if (m->m_len < ip_offset)
1323 		return (1);
1324 
1325 	eh = mtod(m, struct ether_header *);
1326 	eth_type = ntohs(eh->ether_type);
1327 	if (eth_type == ETHERTYPE_VLAN) {
1328 		ip_offset = sizeof(struct ether_vlan_header);
1329 		if (m->m_len < ip_offset)
1330 			return (1);
1331 		evh = mtod(m, struct ether_vlan_header *);
1332 		eth_type = ntohs(evh->evl_proto);
1333 	}
1334 
1335 	switch (eth_type) {
1336 	case ETHERTYPE_IP:
1337 		if (m->m_len < ip_offset + sizeof(struct ip))
1338 			return (1);
1339 
1340 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1341 		 /* Sanity check the IP header. */
1342 		if (ip->ip_v != IPVERSION)
1343 			return (1);
1344 		hlen = ip->ip_hl << 2;
1345 		if (hlen < sizeof(struct ip))
1346 			return (1);
1347 		if (ntohs(ip->ip_len) < hlen)
1348 			return (1);
1349 		if (ntohs(ip->ip_len) != (m->m_pkthdr.len - ip_offset))
1350 			return (1);
1351 
1352 		ip_proto = ip->ip_p;
1353 		csum_start = ip_offset + hlen;
1354 		break;
1355 
1356 	case ETHERTYPE_IPV6:
1357 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr))
1358 			return (1);
1359 
1360 		/*
1361 		 * XXX FreeBSD does not handle any IPv6 checksum offloading
1362 		 * at the moment.
1363 		 */
1364 
1365 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1366 		/* XXX Assume no extension headers are present. */
1367 		ip_proto = ip6->ip6_nxt;
1368 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1369 		break;
1370 
1371 	default:
1372 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1373 		return (1);
1374 	}
1375 
1376 	/* Assume checksum begins right after the IP header. */
1377 	if (hdr->csum_start != csum_start) {
1378 		sc->vtnet_stats.rx_csum_bad_start++;
1379 		return (1);
1380 	}
1381 
1382 	switch (ip_proto) {
1383 	case IPPROTO_TCP:
1384 		csum_offset = offsetof(struct tcphdr, th_sum);
1385 		break;
1386 
1387 	case IPPROTO_UDP:
1388 		csum_offset = offsetof(struct udphdr, uh_sum);
1389 		break;
1390 
1391 	case IPPROTO_SCTP:
1392 		csum_offset = offsetof(struct sctphdr, checksum);
1393 		break;
1394 
1395 	default:
1396 		sc->vtnet_stats.rx_csum_bad_ipproto++;
1397 		return (1);
1398 	}
1399 
1400 	if (hdr->csum_offset != csum_offset) {
1401 		sc->vtnet_stats.rx_csum_bad_offset++;
1402 		return (1);
1403 	}
1404 
1405 	/*
1406 	 * The IP header checksum is almost certainly valid but I'm
1407 	 * uncertain if that is guaranteed.
1408 	 *
1409 	 * m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
1410 	 */
1411 
1412 	switch (ip_proto) {
1413 	case IPPROTO_UDP:
1414 		if (m->m_len < csum_start + sizeof(struct udphdr))
1415 			return (1);
1416 
1417 		udp = (struct udphdr *)(mtod(m, uint8_t *) + csum_start);
1418 		if (udp->uh_sum == 0)
1419 			return (0);
1420 
1421 		/* FALLTHROUGH */
1422 
1423 	case IPPROTO_TCP:
1424 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1425 		m->m_pkthdr.csum_data = 0xFFFF;
1426 		break;
1427 
1428 	case IPPROTO_SCTP:
1429 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1430 		break;
1431 	}
1432 
1433 	sc->vtnet_stats.rx_csum_offloaded++;
1434 
1435 	return (0);
1436 }
1437 #endif
1438 
1439 /*
1440  * Alternative method of doing receive checksum offloading. Rather
1441  * than parsing the received frame down to the IP header, use the
1442  * csum_offset to determine which CSUM_* flags are appropriate. We
1443  * can get by with doing this only because the checksum offsets are
1444  * unique for the things we care about.
1445  */
1446 static int
1447 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1448     struct virtio_net_hdr *hdr)
1449 {
1450 	struct ether_header *eh;
1451 	struct ether_vlan_header *evh;
1452 	struct udphdr *udp;
1453 	int csum_len;
1454 	uint16_t eth_type;
1455 
1456 	csum_len = hdr->csum_start + hdr->csum_offset;
1457 
1458 	if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1459 		return (1);
1460 	if (m->m_len < csum_len)
1461 		return (1);
1462 
1463 	eh = mtod(m, struct ether_header *);
1464 	eth_type = ntohs(eh->ether_type);
1465 	if (eth_type == ETHERTYPE_VLAN) {
1466 		evh = mtod(m, struct ether_vlan_header *);
1467 		eth_type = ntohs(evh->evl_proto);
1468 	}
1469 
1470 	if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1471 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1472 		return (1);
1473 	}
1474 
1475 	/* Use the offset to determine the appropriate CSUM_* flags. */
1476 	switch (hdr->csum_offset) {
1477 	case offsetof(struct udphdr, uh_sum):
1478 		if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1479 			return (1);
1480 		udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1481 		if (udp->uh_sum == 0)
1482 			return (0);
1483 
1484 		/* FALLTHROUGH */
1485 
1486 	case offsetof(struct tcphdr, th_sum):
1487 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1488 		m->m_pkthdr.csum_data = 0xFFFF;
1489 		break;
1490 
1491 	case offsetof(struct sctphdr, checksum):
1492 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1493 		break;
1494 
1495 	default:
1496 		sc->vtnet_stats.rx_csum_bad_offset++;
1497 		return (1);
1498 	}
1499 
1500 	sc->vtnet_stats.rx_csum_offloaded++;
1501 
1502 	return (0);
1503 }
1504 
1505 static int
1506 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1507 {
1508 	struct ifnet *ifp;
1509 	struct virtqueue *vq;
1510 	struct mbuf *m, *m_tail;
1511 	int len;
1512 
1513 	ifp = sc->vtnet_ifp;
1514 	vq = sc->vtnet_rx_vq;
1515 	m_tail = m_head;
1516 
1517 	while (--nbufs > 0) {
1518 		m = virtqueue_dequeue(vq, &len);
1519 		if (m == NULL) {
1520 			ifp->if_ierrors++;
1521 			goto fail;
1522 		}
1523 
1524 		if (vtnet_newbuf(sc) != 0) {
1525 			ifp->if_iqdrops++;
1526 			vtnet_discard_rxbuf(sc, m);
1527 			if (nbufs > 1)
1528 				vtnet_discard_merged_rxbuf(sc, nbufs);
1529 			goto fail;
1530 		}
1531 
1532 		if (m->m_len < len)
1533 			len = m->m_len;
1534 
1535 		m->m_len = len;
1536 		m->m_flags &= ~M_PKTHDR;
1537 
1538 		m_head->m_pkthdr.len += len;
1539 		m_tail->m_next = m;
1540 		m_tail = m;
1541 	}
1542 
1543 	return (0);
1544 
1545 fail:
1546 	sc->vtnet_stats.rx_mergeable_failed++;
1547 	m_freem(m_head);
1548 
1549 	return (1);
1550 }
1551 
1552 static int
1553 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1554 {
1555 	struct virtio_net_hdr lhdr;
1556 	struct ifnet *ifp;
1557 	struct virtqueue *vq;
1558 	struct mbuf *m;
1559 	struct ether_header *eh;
1560 	struct virtio_net_hdr *hdr;
1561 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1562 	int len, deq, nbufs, adjsz, rx_npkts;
1563 
1564 	ifp = sc->vtnet_ifp;
1565 	vq = sc->vtnet_rx_vq;
1566 	hdr = &lhdr;
1567 	deq = 0;
1568 	rx_npkts = 0;
1569 
1570 	VTNET_LOCK_ASSERT(sc);
1571 
1572 	while (--count >= 0) {
1573 		m = virtqueue_dequeue(vq, &len);
1574 		if (m == NULL)
1575 			break;
1576 		deq++;
1577 
1578 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1579 			ifp->if_ierrors++;
1580 			vtnet_discard_rxbuf(sc, m);
1581 			continue;
1582 		}
1583 
1584 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1585 			nbufs = 1;
1586 			adjsz = sizeof(struct vtnet_rx_header);
1587 			/*
1588 			 * Account for our pad between the header and
1589 			 * the actual start of the frame.
1590 			 */
1591 			len += VTNET_RX_HEADER_PAD;
1592 		} else {
1593 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1594 			nbufs = mhdr->num_buffers;
1595 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1596 		}
1597 
1598 		if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1599 			ifp->if_iqdrops++;
1600 			vtnet_discard_rxbuf(sc, m);
1601 			if (nbufs > 1)
1602 				vtnet_discard_merged_rxbuf(sc, nbufs);
1603 			continue;
1604 		}
1605 
1606 		m->m_pkthdr.len = len;
1607 		m->m_pkthdr.rcvif = ifp;
1608 		m->m_pkthdr.csum_flags = 0;
1609 
1610 		if (nbufs > 1) {
1611 			if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1612 				continue;
1613 		}
1614 
1615 		ifp->if_ipackets++;
1616 
1617 		/*
1618 		 * Save copy of header before we strip it. For both mergeable
1619 		 * and non-mergeable, the VirtIO header is placed first in the
1620 		 * mbuf's data. We no longer need num_buffers, so always use a
1621 		 * virtio_net_hdr.
1622 		 */
1623 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1624 		m_adj(m, adjsz);
1625 
1626 		if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1627 			eh = mtod(m, struct ether_header *);
1628 			if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1629 				vtnet_vlan_tag_remove(m);
1630 
1631 				/*
1632 				 * With the 802.1Q header removed, update the
1633 				 * checksum starting location accordingly.
1634 				 */
1635 				if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1636 					hdr->csum_start -=
1637 					    ETHER_VLAN_ENCAP_LEN;
1638 			}
1639 		}
1640 
1641 		if (ifp->if_capenable & IFCAP_RXCSUM &&
1642 		    hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1643 			if (vtnet_rx_csum(sc, m, hdr) != 0)
1644 				sc->vtnet_stats.rx_csum_failed++;
1645 		}
1646 
1647 		VTNET_UNLOCK(sc);
1648 		rx_npkts++;
1649 		(*ifp->if_input)(ifp, m);
1650 		VTNET_LOCK(sc);
1651 
1652 		/*
1653 		 * The interface may have been stopped while we were
1654 		 * passing the packet up the network stack.
1655 		 */
1656 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1657 			break;
1658 	}
1659 
1660 	if (deq > 0)
1661 		virtqueue_notify(vq);
1662 
1663 	if (rx_npktsp != NULL)
1664 		*rx_npktsp = rx_npkts;
1665 
1666 	return (count > 0 ? 0 : EAGAIN);
1667 }
1668 
1669 static void
1670 vtnet_rx_vq_intr(void *xsc)
1671 {
1672 	struct vtnet_softc *sc;
1673 	struct ifnet *ifp;
1674 	int more;
1675 
1676 	sc = xsc;
1677 	ifp = sc->vtnet_ifp;
1678 
1679 again:
1680 	VTNET_LOCK(sc);
1681 
1682 #ifdef DEVICE_POLLING
1683 	if (ifp->if_capenable & IFCAP_POLLING) {
1684 		VTNET_UNLOCK(sc);
1685 		return;
1686 	}
1687 #endif
1688 
1689 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1690 		vtnet_enable_rx_intr(sc);
1691 		VTNET_UNLOCK(sc);
1692 		return;
1693 	}
1694 
1695 	more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1696 	if (more || vtnet_enable_rx_intr(sc) != 0) {
1697 		if (!more)
1698 			vtnet_disable_rx_intr(sc);
1699 		sc->vtnet_stats.rx_task_rescheduled++;
1700 		VTNET_UNLOCK(sc);
1701 		goto again;
1702 	}
1703 
1704 	VTNET_UNLOCK(sc);
1705 }
1706 
1707 static void
1708 vtnet_txeof(struct vtnet_softc *sc)
1709 {
1710 	struct virtqueue *vq;
1711 	struct ifnet *ifp;
1712 	struct vtnet_tx_header *txhdr;
1713 	int deq;
1714 
1715 	vq = sc->vtnet_tx_vq;
1716 	ifp = sc->vtnet_ifp;
1717 	deq = 0;
1718 
1719 	VTNET_LOCK_ASSERT(sc);
1720 
1721 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1722 		deq++;
1723 		ifp->if_opackets++;
1724 		m_freem(txhdr->vth_mbuf);
1725 		uma_zfree(vtnet_tx_header_zone, txhdr);
1726 	}
1727 
1728 	if (deq > 0) {
1729 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1730 		if (virtqueue_empty(vq))
1731 			sc->vtnet_watchdog_timer = 0;
1732 	}
1733 }
1734 
1735 static struct mbuf *
1736 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1737     struct virtio_net_hdr *hdr)
1738 {
1739 	struct ifnet *ifp;
1740 	struct ether_header *eh;
1741 	struct ether_vlan_header *evh;
1742 	struct ip *ip;
1743 	struct ip6_hdr *ip6;
1744 	struct tcphdr *tcp;
1745 	int ip_offset;
1746 	uint16_t eth_type, csum_start;
1747 	uint8_t ip_proto, gso_type;
1748 
1749 	ifp = sc->vtnet_ifp;
1750 
1751 	ip_offset = sizeof(struct ether_header);
1752 	if (m->m_len < ip_offset) {
1753 		if ((m = m_pullup(m, ip_offset)) == NULL)
1754 			return (NULL);
1755 	}
1756 
1757 	eh = mtod(m, struct ether_header *);
1758 	eth_type = ntohs(eh->ether_type);
1759 	if (eth_type == ETHERTYPE_VLAN) {
1760 		ip_offset = sizeof(struct ether_vlan_header);
1761 		if (m->m_len < ip_offset) {
1762 			if ((m = m_pullup(m, ip_offset)) == NULL)
1763 				return (NULL);
1764 		}
1765 		evh = mtod(m, struct ether_vlan_header *);
1766 		eth_type = ntohs(evh->evl_proto);
1767 	}
1768 
1769 	switch (eth_type) {
1770 	case ETHERTYPE_IP:
1771 		if (m->m_len < ip_offset + sizeof(struct ip)) {
1772 			m = m_pullup(m, ip_offset + sizeof(struct ip));
1773 			if (m == NULL)
1774 				return (NULL);
1775 		}
1776 
1777 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1778 		ip_proto = ip->ip_p;
1779 		csum_start = ip_offset + (ip->ip_hl << 2);
1780 		gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1781 		break;
1782 
1783 	case ETHERTYPE_IPV6:
1784 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1785 			m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1786 			if (m == NULL)
1787 				return (NULL);
1788 		}
1789 
1790 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1791 		/*
1792 		 * XXX Assume no extension headers are present. Presently,
1793 		 * this will always be true in the case of TSO, and FreeBSD
1794 		 * does not perform checksum offloading of IPv6 yet.
1795 		 */
1796 		ip_proto = ip6->ip6_nxt;
1797 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1798 		gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1799 		break;
1800 
1801 	default:
1802 		return (m);
1803 	}
1804 
1805 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1806 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1807 		hdr->csum_start = csum_start;
1808 		hdr->csum_offset = m->m_pkthdr.csum_data;
1809 
1810 		sc->vtnet_stats.tx_csum_offloaded++;
1811 	}
1812 
1813 	if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1814 		if (ip_proto != IPPROTO_TCP)
1815 			return (m);
1816 
1817 		if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1818 			m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1819 			if (m == NULL)
1820 				return (NULL);
1821 		}
1822 
1823 		tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1824 		hdr->gso_type = gso_type;
1825 		hdr->hdr_len = csum_start + (tcp->th_off << 2);
1826 		hdr->gso_size = m->m_pkthdr.tso_segsz;
1827 
1828 		if (tcp->th_flags & TH_CWR) {
1829 			/*
1830 			 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1831 			 * ECN support is only configurable globally with the
1832 			 * net.inet.tcp.ecn.enable sysctl knob.
1833 			 */
1834 			if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1835 				if_printf(ifp, "TSO with ECN not supported "
1836 				    "by host\n");
1837 				m_freem(m);
1838 				return (NULL);
1839 			}
1840 
1841 			hdr->flags |= VIRTIO_NET_HDR_GSO_ECN;
1842 		}
1843 
1844 		sc->vtnet_stats.tx_tso_offloaded++;
1845 	}
1846 
1847 	return (m);
1848 }
1849 
1850 static int
1851 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1852     struct vtnet_tx_header *txhdr)
1853 {
1854 	struct sglist sg;
1855 	struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1856 	struct virtqueue *vq;
1857 	struct mbuf *m;
1858 	int collapsed, error;
1859 
1860 	vq = sc->vtnet_tx_vq;
1861 	m = *m_head;
1862 	collapsed = 0;
1863 
1864 	sglist_init(&sg, VTNET_MAX_TX_SEGS, segs);
1865 	error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1866 	KASSERT(error == 0 && sg.sg_nseg == 1,
1867 	    ("%s: cannot add header to sglist error %d", __func__, error));
1868 
1869 again:
1870 	error = sglist_append_mbuf(&sg, m);
1871 	if (error) {
1872 		if (collapsed)
1873 			goto fail;
1874 
1875 		m = m_collapse(m, M_NOWAIT, VTNET_MAX_TX_SEGS - 1);
1876 		if (m == NULL)
1877 			goto fail;
1878 
1879 		*m_head = m;
1880 		collapsed = 1;
1881 		goto again;
1882 	}
1883 
1884 	txhdr->vth_mbuf = m;
1885 
1886 	return (virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0));
1887 
1888 fail:
1889 	m_freem(*m_head);
1890 	*m_head = NULL;
1891 
1892 	return (ENOBUFS);
1893 }
1894 
1895 static int
1896 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1897 {
1898 	struct vtnet_tx_header *txhdr;
1899 	struct virtio_net_hdr *hdr;
1900 	struct mbuf *m;
1901 	int error;
1902 
1903 	m = *m_head;
1904 	M_ASSERTPKTHDR(m);
1905 
1906 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
1907 	if (txhdr == NULL) {
1908 		*m_head = NULL;
1909 		m_freem(m);
1910 		return (ENOMEM);
1911 	}
1912 
1913 	/*
1914 	 * Always use the non-mergeable header to simplify things. When
1915 	 * the mergeable feature is negotiated, the num_buffers field
1916 	 * must be set to zero. We use vtnet_hdr_size later to enqueue
1917 	 * the correct header size to the host.
1918 	 */
1919 	hdr = &txhdr->vth_uhdr.hdr;
1920 
1921 	if (m->m_flags & M_VLANTAG) {
1922 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1923 		if ((*m_head = m) == NULL) {
1924 			error = ENOBUFS;
1925 			goto fail;
1926 		}
1927 		m->m_flags &= ~M_VLANTAG;
1928 	}
1929 
1930 	if (m->m_pkthdr.csum_flags != 0) {
1931 		m = vtnet_tx_offload(sc, m, hdr);
1932 		if ((*m_head = m) == NULL) {
1933 			error = ENOBUFS;
1934 			goto fail;
1935 		}
1936 	}
1937 
1938 	error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1939 fail:
1940 	if (error)
1941 		uma_zfree(vtnet_tx_header_zone, txhdr);
1942 
1943 	return (error);
1944 }
1945 
1946 static void
1947 vtnet_start(struct ifnet *ifp)
1948 {
1949 	struct vtnet_softc *sc;
1950 
1951 	sc = ifp->if_softc;
1952 
1953 	VTNET_LOCK(sc);
1954 	vtnet_start_locked(ifp);
1955 	VTNET_UNLOCK(sc);
1956 }
1957 
1958 static void
1959 vtnet_start_locked(struct ifnet *ifp)
1960 {
1961 	struct vtnet_softc *sc;
1962 	struct virtqueue *vq;
1963 	struct mbuf *m0;
1964 	int enq;
1965 
1966 	sc = ifp->if_softc;
1967 	vq = sc->vtnet_tx_vq;
1968 	enq = 0;
1969 
1970 	VTNET_LOCK_ASSERT(sc);
1971 
1972 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1973 	    IFF_DRV_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
1974 		return;
1975 
1976 #ifdef VTNET_TX_INTR_MODERATION
1977 	if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
1978 		vtnet_txeof(sc);
1979 #endif
1980 
1981 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1982 		if (virtqueue_full(vq)) {
1983 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1984 			break;
1985 		}
1986 
1987 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
1988 		if (m0 == NULL)
1989 			break;
1990 
1991 		if (vtnet_encap(sc, &m0) != 0) {
1992 			if (m0 == NULL)
1993 				break;
1994 			IFQ_DRV_PREPEND(&ifp->if_snd, m0);
1995 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1996 			break;
1997 		}
1998 
1999 		enq++;
2000 		ETHER_BPF_MTAP(ifp, m0);
2001 	}
2002 
2003 	if (enq > 0) {
2004 		virtqueue_notify(vq);
2005 		sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
2006 	}
2007 }
2008 
2009 static void
2010 vtnet_tick(void *xsc)
2011 {
2012 	struct vtnet_softc *sc;
2013 
2014 	sc = xsc;
2015 
2016 	VTNET_LOCK_ASSERT(sc);
2017 #ifdef VTNET_DEBUG
2018 	virtqueue_dump(sc->vtnet_rx_vq);
2019 	virtqueue_dump(sc->vtnet_tx_vq);
2020 #endif
2021 
2022 	vtnet_watchdog(sc);
2023 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2024 }
2025 
2026 static void
2027 vtnet_tx_vq_intr(void *xsc)
2028 {
2029 	struct vtnet_softc *sc;
2030 	struct ifnet *ifp;
2031 
2032 	sc = xsc;
2033 	ifp = sc->vtnet_ifp;
2034 
2035 again:
2036 	VTNET_LOCK(sc);
2037 
2038 #ifdef DEVICE_POLLING
2039 	if (ifp->if_capenable & IFCAP_POLLING) {
2040 		VTNET_UNLOCK(sc);
2041 		return;
2042 	}
2043 #endif
2044 
2045 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2046 		vtnet_enable_tx_intr(sc);
2047 		VTNET_UNLOCK(sc);
2048 		return;
2049 	}
2050 
2051 	vtnet_txeof(sc);
2052 
2053 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2054 		vtnet_start_locked(ifp);
2055 
2056 	if (vtnet_enable_tx_intr(sc) != 0) {
2057 		vtnet_disable_tx_intr(sc);
2058 		sc->vtnet_stats.tx_task_rescheduled++;
2059 		VTNET_UNLOCK(sc);
2060 		goto again;
2061 	}
2062 
2063 	VTNET_UNLOCK(sc);
2064 }
2065 
2066 static void
2067 vtnet_stop(struct vtnet_softc *sc)
2068 {
2069 	device_t dev;
2070 	struct ifnet *ifp;
2071 
2072 	dev = sc->vtnet_dev;
2073 	ifp = sc->vtnet_ifp;
2074 
2075 	VTNET_LOCK_ASSERT(sc);
2076 
2077 	sc->vtnet_watchdog_timer = 0;
2078 	callout_stop(&sc->vtnet_tick_ch);
2079 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2080 
2081 	vtnet_disable_rx_intr(sc);
2082 	vtnet_disable_tx_intr(sc);
2083 
2084 	/*
2085 	 * Stop the host VirtIO adapter. Note this will reset the host
2086 	 * adapter's state back to the pre-initialized state, so in
2087 	 * order to make the device usable again, we must drive it
2088 	 * through virtio_reinit() and virtio_reinit_complete().
2089 	 */
2090 	virtio_stop(dev);
2091 
2092 	sc->vtnet_flags &= ~VTNET_FLAG_LINK;
2093 
2094 	vtnet_free_rx_mbufs(sc);
2095 	vtnet_free_tx_mbufs(sc);
2096 }
2097 
2098 static int
2099 vtnet_reinit(struct vtnet_softc *sc)
2100 {
2101 	struct ifnet *ifp;
2102 	uint64_t features;
2103 
2104 	ifp = sc->vtnet_ifp;
2105 	features = sc->vtnet_features;
2106 
2107 	/*
2108 	 * Re-negotiate with the host, removing any disabled receive
2109 	 * features. Transmit features are disabled only on our side
2110 	 * via if_capenable and if_hwassist.
2111 	 */
2112 
2113 	if (ifp->if_capabilities & IFCAP_RXCSUM) {
2114 		if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2115 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2116 	}
2117 
2118 	if (ifp->if_capabilities & IFCAP_LRO) {
2119 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2120 			features &= ~VTNET_LRO_FEATURES;
2121 	}
2122 
2123 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2124 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2125 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2126 	}
2127 
2128 	return (virtio_reinit(sc->vtnet_dev, features));
2129 }
2130 
2131 static void
2132 vtnet_init_locked(struct vtnet_softc *sc)
2133 {
2134 	device_t dev;
2135 	struct ifnet *ifp;
2136 	int error;
2137 
2138 	dev = sc->vtnet_dev;
2139 	ifp = sc->vtnet_ifp;
2140 
2141 	VTNET_LOCK_ASSERT(sc);
2142 
2143 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2144 		return;
2145 
2146 	/* Stop host's adapter, cancel any pending I/O. */
2147 	vtnet_stop(sc);
2148 
2149 	/* Reinitialize the host device. */
2150 	error = vtnet_reinit(sc);
2151 	if (error) {
2152 		device_printf(dev,
2153 		    "reinitialization failed, stopping device...\n");
2154 		vtnet_stop(sc);
2155 		return;
2156 	}
2157 
2158 	/* Update host with assigned MAC address. */
2159 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2160 	vtnet_set_hwaddr(sc);
2161 
2162 	ifp->if_hwassist = 0;
2163 	if (ifp->if_capenable & IFCAP_TXCSUM)
2164 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2165 	if (ifp->if_capenable & IFCAP_TSO4)
2166 		ifp->if_hwassist |= CSUM_TSO;
2167 
2168 	error = vtnet_init_rx_vq(sc);
2169 	if (error) {
2170 		device_printf(dev,
2171 		    "cannot allocate mbufs for Rx virtqueue\n");
2172 		vtnet_stop(sc);
2173 		return;
2174 	}
2175 
2176 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2177 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2178 			/* Restore promiscuous and all-multicast modes. */
2179 			vtnet_rx_filter(sc);
2180 
2181 			/* Restore filtered MAC addresses. */
2182 			vtnet_rx_filter_mac(sc);
2183 		}
2184 
2185 		/* Restore VLAN filters. */
2186 		if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2187 			vtnet_rx_filter_vlan(sc);
2188 	}
2189 
2190 #ifdef DEVICE_POLLING
2191 	if (ifp->if_capenable & IFCAP_POLLING) {
2192 		vtnet_disable_rx_intr(sc);
2193 		vtnet_disable_tx_intr(sc);
2194 	} else
2195 #endif
2196 	{
2197 		vtnet_enable_rx_intr(sc);
2198 		vtnet_enable_tx_intr(sc);
2199 	}
2200 
2201 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2202 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2203 
2204 	virtio_reinit_complete(dev);
2205 
2206 	vtnet_update_link_status(sc);
2207 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2208 }
2209 
2210 static void
2211 vtnet_init(void *xsc)
2212 {
2213 	struct vtnet_softc *sc;
2214 
2215 	sc = xsc;
2216 
2217 	VTNET_LOCK(sc);
2218 	vtnet_init_locked(sc);
2219 	VTNET_UNLOCK(sc);
2220 }
2221 
2222 static void
2223 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2224     struct sglist *sg, int readable, int writable)
2225 {
2226 	struct virtqueue *vq;
2227 	void *c;
2228 
2229 	vq = sc->vtnet_ctrl_vq;
2230 
2231 	VTNET_LOCK_ASSERT(sc);
2232 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2233 	    ("no control virtqueue"));
2234 	KASSERT(virtqueue_empty(vq),
2235 	    ("control command already enqueued"));
2236 
2237 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2238 		return;
2239 
2240 	virtqueue_notify(vq);
2241 
2242 	/*
2243 	 * Poll until the command is complete. Previously, we would
2244 	 * sleep until the control virtqueue interrupt handler woke
2245 	 * us up, but dropping the VTNET_MTX leads to serialization
2246 	 * difficulties.
2247 	 *
2248 	 * Furthermore, it appears QEMU/KVM only allocates three MSIX
2249 	 * vectors. Two of those vectors are needed for the Rx and Tx
2250 	 * virtqueues. We do not support sharing both a Vq and config
2251 	 * changed notification on the same MSIX vector.
2252 	 */
2253 	c = virtqueue_poll(vq, NULL);
2254 	KASSERT(c == cookie, ("unexpected control command response"));
2255 }
2256 
2257 static void
2258 vtnet_rx_filter(struct vtnet_softc *sc)
2259 {
2260 	device_t dev;
2261 	struct ifnet *ifp;
2262 
2263 	dev = sc->vtnet_dev;
2264 	ifp = sc->vtnet_ifp;
2265 
2266 	VTNET_LOCK_ASSERT(sc);
2267 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2268 	    ("CTRL_RX feature not negotiated"));
2269 
2270 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2271 		device_printf(dev, "cannot %s promiscuous mode\n",
2272 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
2273 
2274 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2275 		device_printf(dev, "cannot %s all-multicast mode\n",
2276 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
2277 }
2278 
2279 static int
2280 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2281 {
2282 	struct virtio_net_ctrl_hdr hdr;
2283 	struct sglist_seg segs[3];
2284 	struct sglist sg;
2285 	uint8_t onoff, ack;
2286 	int error;
2287 
2288 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
2289 		return (ENOTSUP);
2290 
2291 	error = 0;
2292 
2293 	hdr.class = VIRTIO_NET_CTRL_RX;
2294 	hdr.cmd = cmd;
2295 	onoff = !!on;
2296 	ack = VIRTIO_NET_ERR;
2297 
2298 	sglist_init(&sg, 3, segs);
2299 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2300 	error |= sglist_append(&sg, &onoff, sizeof(uint8_t));
2301 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2302 	KASSERT(error == 0 && sg.sg_nseg == 3,
2303 	    ("error adding Rx filter message to sglist"));
2304 
2305 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2306 
2307 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2308 }
2309 
2310 static int
2311 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2312 {
2313 
2314 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2315 }
2316 
2317 static int
2318 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2319 {
2320 
2321 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2322 }
2323 
2324 static void
2325 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2326 {
2327 	struct virtio_net_ctrl_hdr hdr;
2328 	struct vtnet_mac_filter *filter;
2329 	struct sglist_seg segs[4];
2330 	struct sglist sg;
2331 	struct ifnet *ifp;
2332 	struct ifaddr *ifa;
2333 	struct ifmultiaddr *ifma;
2334 	int ucnt, mcnt, promisc, allmulti, error;
2335 	uint8_t ack;
2336 
2337 	ifp = sc->vtnet_ifp;
2338 	filter = sc->vtnet_mac_filter;
2339 	ucnt = 0;
2340 	mcnt = 0;
2341 	promisc = 0;
2342 	allmulti = 0;
2343 	error = 0;
2344 
2345 	VTNET_LOCK_ASSERT(sc);
2346 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2347 	    ("CTRL_RX feature not negotiated"));
2348 
2349 	/* Unicast MAC addresses: */
2350 	if_addr_rlock(ifp);
2351 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2352 		if (ifa->ifa_addr->sa_family != AF_LINK)
2353 			continue;
2354 		else if (ucnt == VTNET_MAX_MAC_ENTRIES)
2355 			break;
2356 
2357 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2358 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2359 		ucnt++;
2360 	}
2361 	if_addr_runlock(ifp);
2362 
2363 	if (ucnt >= VTNET_MAX_MAC_ENTRIES) {
2364 		promisc = 1;
2365 		filter->vmf_unicast.nentries = 0;
2366 
2367 		if_printf(ifp, "more than %d MAC addresses assigned, "
2368 		    "falling back to promiscuous mode\n",
2369 		    VTNET_MAX_MAC_ENTRIES);
2370 	} else
2371 		filter->vmf_unicast.nentries = ucnt;
2372 
2373 	/* Multicast MAC addresses: */
2374 	if_maddr_rlock(ifp);
2375 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2376 		if (ifma->ifma_addr->sa_family != AF_LINK)
2377 			continue;
2378 		else if (mcnt == VTNET_MAX_MAC_ENTRIES)
2379 			break;
2380 
2381 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2382 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2383 		mcnt++;
2384 	}
2385 	if_maddr_runlock(ifp);
2386 
2387 	if (mcnt >= VTNET_MAX_MAC_ENTRIES) {
2388 		allmulti = 1;
2389 		filter->vmf_multicast.nentries = 0;
2390 
2391 		if_printf(ifp, "more than %d multicast MAC addresses "
2392 		    "assigned, falling back to all-multicast mode\n",
2393 		    VTNET_MAX_MAC_ENTRIES);
2394 	} else
2395 		filter->vmf_multicast.nentries = mcnt;
2396 
2397 	if (promisc && allmulti)
2398 		goto out;
2399 
2400 	hdr.class = VIRTIO_NET_CTRL_MAC;
2401 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2402 	ack = VIRTIO_NET_ERR;
2403 
2404 	sglist_init(&sg, 4, segs);
2405 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2406 	error |= sglist_append(&sg, &filter->vmf_unicast,
2407 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
2408 	error |= sglist_append(&sg, &filter->vmf_multicast,
2409 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
2410 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2411 	KASSERT(error == 0 && sg.sg_nseg == 4,
2412 	    ("error adding MAC filtering message to sglist"));
2413 
2414 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2415 
2416 	if (ack != VIRTIO_NET_OK)
2417 		if_printf(ifp, "error setting host MAC filter table\n");
2418 
2419 out:
2420 	if (promisc)
2421 		if (vtnet_set_promisc(sc, 1) != 0)
2422 			if_printf(ifp, "cannot enable promiscuous mode\n");
2423 	if (allmulti)
2424 		if (vtnet_set_allmulti(sc, 1) != 0)
2425 			if_printf(ifp, "cannot enable all-multicast mode\n");
2426 }
2427 
2428 static int
2429 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2430 {
2431 	struct virtio_net_ctrl_hdr hdr;
2432 	struct sglist_seg segs[3];
2433 	struct sglist sg;
2434 	uint8_t ack;
2435 	int error;
2436 
2437 	hdr.class = VIRTIO_NET_CTRL_VLAN;
2438 	hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2439 	ack = VIRTIO_NET_ERR;
2440 	error = 0;
2441 
2442 	sglist_init(&sg, 3, segs);
2443 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2444 	error |= sglist_append(&sg, &tag, sizeof(uint16_t));
2445 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2446 	KASSERT(error == 0 && sg.sg_nseg == 3,
2447 	    ("error adding VLAN control message to sglist"));
2448 
2449 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2450 
2451 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2452 }
2453 
2454 static void
2455 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2456 {
2457 	device_t dev;
2458 	uint32_t w, mask;
2459 	uint16_t tag;
2460 	int i, nvlans, error;
2461 
2462 	VTNET_LOCK_ASSERT(sc);
2463 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2464 	    ("VLAN_FILTER feature not negotiated"));
2465 
2466 	dev = sc->vtnet_dev;
2467 	nvlans = sc->vtnet_nvlans;
2468 	error = 0;
2469 
2470 	/* Enable filtering for each configured VLAN. */
2471 	for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2472 		w = sc->vtnet_vlan_shadow[i];
2473 		for (mask = 1, tag = i * 32; w != 0; mask <<= 1, tag++) {
2474 			if ((w & mask) != 0) {
2475 				w &= ~mask;
2476 				nvlans--;
2477 				if (vtnet_exec_vlan_filter(sc, 1, tag) != 0)
2478 					error++;
2479 			}
2480 		}
2481 	}
2482 
2483 	KASSERT(nvlans == 0, ("VLAN count incorrect"));
2484 	if (error)
2485 		device_printf(dev, "cannot restore VLAN filter table\n");
2486 }
2487 
2488 static void
2489 vtnet_set_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2490 {
2491 	struct ifnet *ifp;
2492 	int idx, bit;
2493 
2494 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2495 	    ("VLAN_FILTER feature not negotiated"));
2496 
2497 	if ((tag == 0) || (tag > 4095))
2498 		return;
2499 
2500 	ifp = sc->vtnet_ifp;
2501 	idx = (tag >> 5) & 0x7F;
2502 	bit = tag & 0x1F;
2503 
2504 	VTNET_LOCK(sc);
2505 
2506 	/* Update shadow VLAN table. */
2507 	if (add) {
2508 		sc->vtnet_nvlans++;
2509 		sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2510 	} else {
2511 		sc->vtnet_nvlans--;
2512 		sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2513 	}
2514 
2515 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2516 		if (vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2517 			device_printf(sc->vtnet_dev,
2518 			    "cannot %s VLAN %d %s the host filter table\n",
2519 			    add ? "add" : "remove", tag,
2520 			    add ? "to" : "from");
2521 		}
2522 	}
2523 
2524 	VTNET_UNLOCK(sc);
2525 }
2526 
2527 static void
2528 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2529 {
2530 
2531 	if (ifp->if_softc != arg)
2532 		return;
2533 
2534 	vtnet_set_vlan_filter(arg, 1, tag);
2535 }
2536 
2537 static void
2538 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2539 {
2540 
2541 	if (ifp->if_softc != arg)
2542 		return;
2543 
2544 	vtnet_set_vlan_filter(arg, 0, tag);
2545 }
2546 
2547 static int
2548 vtnet_ifmedia_upd(struct ifnet *ifp)
2549 {
2550 	struct vtnet_softc *sc;
2551 	struct ifmedia *ifm;
2552 
2553 	sc = ifp->if_softc;
2554 	ifm = &sc->vtnet_media;
2555 
2556 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2557 		return (EINVAL);
2558 
2559 	return (0);
2560 }
2561 
2562 static void
2563 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2564 {
2565 	struct vtnet_softc *sc;
2566 
2567 	sc = ifp->if_softc;
2568 
2569 	ifmr->ifm_status = IFM_AVALID;
2570 	ifmr->ifm_active = IFM_ETHER;
2571 
2572 	VTNET_LOCK(sc);
2573 	if (vtnet_is_link_up(sc) != 0) {
2574 		ifmr->ifm_status |= IFM_ACTIVE;
2575 		ifmr->ifm_active |= VTNET_MEDIATYPE;
2576 	} else
2577 		ifmr->ifm_active |= IFM_NONE;
2578 	VTNET_UNLOCK(sc);
2579 }
2580 
2581 static void
2582 vtnet_add_statistics(struct vtnet_softc *sc)
2583 {
2584 	device_t dev;
2585 	struct vtnet_statistics *stats;
2586         struct sysctl_ctx_list *ctx;
2587 	struct sysctl_oid *tree;
2588 	struct sysctl_oid_list *child;
2589 
2590 	dev = sc->vtnet_dev;
2591 	stats = &sc->vtnet_stats;
2592 	ctx = device_get_sysctl_ctx(dev);
2593 	tree = device_get_sysctl_tree(dev);
2594 	child = SYSCTL_CHILDREN(tree);
2595 
2596 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2597 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
2598 	    "Mbuf cluster allocation failures");
2599 
2600 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_frame_too_large",
2601 	    CTLFLAG_RD, &stats->rx_frame_too_large,
2602 	    "Received frame larger than the mbuf chain");
2603 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2604 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
2605 	    "Enqueuing the replacement receive mbuf failed");
2606 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_mergeable_failed",
2607 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
2608 	    "Mergeable buffers receive failures");
2609 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2610 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
2611 	    "Received checksum offloaded buffer with unsupported "
2612 	    "Ethernet type");
2613 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_start",
2614 	    CTLFLAG_RD, &stats->rx_csum_bad_start,
2615 	    "Received checksum offloaded buffer with incorrect start offset");
2616 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2617 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
2618 	    "Received checksum offloaded buffer with incorrect IP protocol");
2619 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2620 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
2621 	    "Received checksum offloaded buffer with incorrect offset");
2622 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_failed",
2623 	    CTLFLAG_RD, &stats->rx_csum_failed,
2624 	    "Received buffer checksum offload failed");
2625 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_offloaded",
2626 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
2627 	    "Received buffer checksum offload succeeded");
2628 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_task_rescheduled",
2629 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
2630 	    "Times the receive interrupt task rescheduled itself");
2631 
2632 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_offloaded",
2633 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
2634 	    "Offloaded checksum of transmitted buffer");
2635 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_offloaded",
2636 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
2637 	    "Segmentation offload of transmitted buffer");
2638 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2639 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
2640 	    "Aborted transmit of checksum offloaded buffer with unknown "
2641 	    "Ethernet type");
2642 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2643 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
2644 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
2645 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_task_rescheduled",
2646 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
2647 	    "Times the transmit interrupt task rescheduled itself");
2648 }
2649 
2650 static int
2651 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2652 {
2653 
2654 	return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2655 }
2656 
2657 static void
2658 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2659 {
2660 
2661 	virtqueue_disable_intr(sc->vtnet_rx_vq);
2662 }
2663 
2664 static int
2665 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2666 {
2667 
2668 #ifdef VTNET_TX_INTR_MODERATION
2669 	return (0);
2670 #else
2671 	return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2672 #endif
2673 }
2674 
2675 static void
2676 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2677 {
2678 
2679 	virtqueue_disable_intr(sc->vtnet_tx_vq);
2680 }
2681