xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@daemoninthecloset.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/taskqueue.h>
46 #include <sys/random.h>
47 #include <sys/sglist.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 
51 #include <vm/uma.h>
52 
53 #include <net/ethernet.h>
54 #include <net/if.h>
55 #include <net/if_arp.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/if_vlan_var.h>
60 
61 #include <net/bpf.h>
62 
63 #include <netinet/in_systm.h>
64 #include <netinet/in.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip6.h>
67 #include <netinet/udp.h>
68 #include <netinet/tcp.h>
69 #include <netinet/sctp.h>
70 
71 #include <machine/bus.h>
72 #include <machine/resource.h>
73 #include <sys/bus.h>
74 #include <sys/rman.h>
75 
76 #include <dev/virtio/virtio.h>
77 #include <dev/virtio/virtqueue.h>
78 #include <dev/virtio/network/virtio_net.h>
79 #include <dev/virtio/network/if_vtnetvar.h>
80 
81 #include "virtio_if.h"
82 
83 static int	vtnet_modevent(module_t, int, void *);
84 
85 static int	vtnet_probe(device_t);
86 static int	vtnet_attach(device_t);
87 static int	vtnet_detach(device_t);
88 static int	vtnet_suspend(device_t);
89 static int	vtnet_resume(device_t);
90 static int	vtnet_shutdown(device_t);
91 static int	vtnet_config_change(device_t);
92 
93 static void	vtnet_negotiate_features(struct vtnet_softc *);
94 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
95 static void	vtnet_get_hwaddr(struct vtnet_softc *);
96 static void	vtnet_set_hwaddr(struct vtnet_softc *);
97 static int	vtnet_is_link_up(struct vtnet_softc *);
98 static void	vtnet_update_link_status(struct vtnet_softc *);
99 static void	vtnet_watchdog(struct vtnet_softc *);
100 static void	vtnet_config_change_task(void *, int);
101 static int	vtnet_change_mtu(struct vtnet_softc *, int);
102 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
103 
104 static int	vtnet_init_rx_vq(struct vtnet_softc *);
105 static void	vtnet_free_rx_mbufs(struct vtnet_softc *);
106 static void	vtnet_free_tx_mbufs(struct vtnet_softc *);
107 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
108 
109 #ifdef DEVICE_POLLING
110 static poll_handler_t vtnet_poll;
111 #endif
112 
113 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int,
114 		    struct mbuf **);
115 static int	vtnet_replace_rxbuf(struct vtnet_softc *,
116 		    struct mbuf *, int);
117 static int	vtnet_newbuf(struct vtnet_softc *);
118 static void	vtnet_discard_merged_rxbuf(struct vtnet_softc *, int);
119 static void	vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *);
120 static int	vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *);
121 static void	vtnet_vlan_tag_remove(struct mbuf *);
122 static int	vtnet_rx_csum(struct vtnet_softc *, struct mbuf *,
123 		    struct virtio_net_hdr *);
124 static int	vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int);
125 static int	vtnet_rxeof(struct vtnet_softc *, int, int *);
126 static void	vtnet_rx_intr_task(void *, int);
127 static int	vtnet_rx_vq_intr(void *);
128 
129 static void	vtnet_txeof(struct vtnet_softc *);
130 static struct mbuf * vtnet_tx_offload(struct vtnet_softc *, struct mbuf *,
131 		    struct virtio_net_hdr *);
132 static int	vtnet_enqueue_txbuf(struct vtnet_softc *, struct mbuf **,
133 		    struct vtnet_tx_header *);
134 static int	vtnet_encap(struct vtnet_softc *, struct mbuf **);
135 static void	vtnet_start_locked(struct ifnet *);
136 static void	vtnet_start(struct ifnet *);
137 static void	vtnet_tick(void *);
138 static void	vtnet_tx_intr_task(void *, int);
139 static int	vtnet_tx_vq_intr(void *);
140 
141 static void	vtnet_stop(struct vtnet_softc *);
142 static int	vtnet_reinit(struct vtnet_softc *);
143 static void	vtnet_init_locked(struct vtnet_softc *);
144 static void	vtnet_init(void *);
145 
146 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
147 		    struct sglist *, int, int);
148 
149 static void	vtnet_rx_filter(struct vtnet_softc *sc);
150 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
151 static int	vtnet_set_promisc(struct vtnet_softc *, int);
152 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
153 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
154 
155 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
156 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
157 static void	vtnet_set_vlan_filter(struct vtnet_softc *, int, uint16_t);
158 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
159 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
160 
161 static int	vtnet_ifmedia_upd(struct ifnet *);
162 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
163 
164 static void	vtnet_add_statistics(struct vtnet_softc *);
165 
166 static int	vtnet_enable_rx_intr(struct vtnet_softc *);
167 static int	vtnet_enable_tx_intr(struct vtnet_softc *);
168 static void	vtnet_disable_rx_intr(struct vtnet_softc *);
169 static void	vtnet_disable_tx_intr(struct vtnet_softc *);
170 
171 /* Tunables. */
172 static int vtnet_csum_disable = 0;
173 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
174 static int vtnet_tso_disable = 0;
175 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
176 static int vtnet_lro_disable = 0;
177 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
178 
179 /*
180  * Reducing the number of transmit completed interrupts can
181  * improve performance. To do so, the define below keeps the
182  * Tx vq interrupt disabled and adds calls to vtnet_txeof()
183  * in the start and watchdog paths. The price to pay for this
184  * is the m_free'ing of transmitted mbufs may be delayed until
185  * the watchdog fires.
186  */
187 #define VTNET_TX_INTR_MODERATION
188 
189 static uma_zone_t vtnet_tx_header_zone;
190 
191 static struct virtio_feature_desc vtnet_feature_desc[] = {
192 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
193 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
194 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
195 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
196 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
197 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
198 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
199 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
200 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
201 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
202 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
203 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
204 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
205 	{ VIRTIO_NET_F_STATUS,		"Status"	},
206 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
207 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
208 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
209 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
210 
211 	{ 0, NULL }
212 };
213 
214 static device_method_t vtnet_methods[] = {
215 	/* Device methods. */
216 	DEVMETHOD(device_probe,		vtnet_probe),
217 	DEVMETHOD(device_attach,	vtnet_attach),
218 	DEVMETHOD(device_detach,	vtnet_detach),
219 	DEVMETHOD(device_suspend,	vtnet_suspend),
220 	DEVMETHOD(device_resume,	vtnet_resume),
221 	DEVMETHOD(device_shutdown,	vtnet_shutdown),
222 
223 	/* VirtIO methods. */
224 	DEVMETHOD(virtio_config_change, vtnet_config_change),
225 
226 	DEVMETHOD_END
227 };
228 
229 static driver_t vtnet_driver = {
230 	"vtnet",
231 	vtnet_methods,
232 	sizeof(struct vtnet_softc)
233 };
234 static devclass_t vtnet_devclass;
235 
236 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
237     vtnet_modevent, 0);
238 MODULE_VERSION(vtnet, 1);
239 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
240 
241 static int
242 vtnet_modevent(module_t mod, int type, void *unused)
243 {
244 	int error;
245 
246 	error = 0;
247 
248 	switch (type) {
249 	case MOD_LOAD:
250 		vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
251 		    sizeof(struct vtnet_tx_header),
252 		    NULL, NULL, NULL, NULL, 0, 0);
253 		break;
254 	case MOD_QUIESCE:
255 	case MOD_UNLOAD:
256 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
257 			error = EBUSY;
258 		else if (type == MOD_UNLOAD) {
259 			uma_zdestroy(vtnet_tx_header_zone);
260 			vtnet_tx_header_zone = NULL;
261 		}
262 		break;
263 	case MOD_SHUTDOWN:
264 		break;
265 	default:
266 		error = EOPNOTSUPP;
267 		break;
268 	}
269 
270 	return (error);
271 }
272 
273 static int
274 vtnet_probe(device_t dev)
275 {
276 
277 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
278 		return (ENXIO);
279 
280 	device_set_desc(dev, "VirtIO Networking Adapter");
281 
282 	return (BUS_PROBE_DEFAULT);
283 }
284 
285 static int
286 vtnet_attach(device_t dev)
287 {
288 	struct vtnet_softc *sc;
289 	struct ifnet *ifp;
290 	int tx_size, error;
291 
292 	sc = device_get_softc(dev);
293 	sc->vtnet_dev = dev;
294 
295 	VTNET_LOCK_INIT(sc);
296 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_MTX(sc), 0);
297 
298 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
299 	    vtnet_ifmedia_sts);
300 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
301 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
302 
303 	vtnet_add_statistics(sc);
304 
305 	virtio_set_feature_desc(dev, vtnet_feature_desc);
306 	vtnet_negotiate_features(sc);
307 
308 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
309 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
310 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
311 	} else
312 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
313 
314 	sc->vtnet_rx_mbuf_size = MCLBYTES;
315 	sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
316 
317 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
318 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
319 
320 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) {
321 			sc->vtnet_mac_filter = malloc(
322 			    sizeof(struct vtnet_mac_filter), M_DEVBUF,
323 			    M_NOWAIT | M_ZERO);
324 			if (sc->vtnet_mac_filter == NULL) {
325 				device_printf(dev,
326 				    "cannot allocate mac filter table\n");
327 				error = ENOMEM;
328 				goto fail;
329 			}
330 
331 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
332 		}
333 
334 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
335 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
336 	}
337 
338 	vtnet_get_hwaddr(sc);
339 
340 	error = vtnet_alloc_virtqueues(sc);
341 	if (error) {
342 		device_printf(dev, "cannot allocate virtqueues\n");
343 		goto fail;
344 	}
345 
346 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
347 	if (ifp == NULL) {
348 		device_printf(dev, "cannot allocate ifnet structure\n");
349 		error = ENOSPC;
350 		goto fail;
351 	}
352 
353 	ifp->if_softc = sc;
354 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
355 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
356 	ifp->if_init = vtnet_init;
357 	ifp->if_start = vtnet_start;
358 	ifp->if_ioctl = vtnet_ioctl;
359 
360 	sc->vtnet_rx_size = virtqueue_size(sc->vtnet_rx_vq);
361 	sc->vtnet_rx_process_limit = sc->vtnet_rx_size;
362 
363 	tx_size = virtqueue_size(sc->vtnet_tx_vq);
364 	sc->vtnet_tx_size = tx_size;
365 	IFQ_SET_MAXLEN(&ifp->if_snd, tx_size - 1);
366 	ifp->if_snd.ifq_drv_maxlen = tx_size - 1;
367 	IFQ_SET_READY(&ifp->if_snd);
368 
369 	ether_ifattach(ifp, sc->vtnet_hwaddr);
370 
371 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
372 		ifp->if_capabilities |= IFCAP_LINKSTATE;
373 
374 	/* Tell the upper layer(s) we support long frames. */
375 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
376 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
377 
378 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
379 		ifp->if_capabilities |= IFCAP_TXCSUM;
380 
381 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
382 			ifp->if_capabilities |= IFCAP_TSO4;
383 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
384 			ifp->if_capabilities |= IFCAP_TSO6;
385 		if (ifp->if_capabilities & IFCAP_TSO)
386 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
387 
388 		if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
389 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
390 	}
391 
392 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
393 		ifp->if_capabilities |= IFCAP_RXCSUM;
394 
395 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
396 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
397 			ifp->if_capabilities |= IFCAP_LRO;
398 	}
399 
400 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
401 		/*
402 		 * VirtIO does not support VLAN tagging, but we can fake
403 		 * it by inserting and removing the 802.1Q header during
404 		 * transmit and receive. We are then able to do checksum
405 		 * offloading of VLAN frames.
406 		 */
407 		ifp->if_capabilities |=
408 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
409 	}
410 
411 	ifp->if_capenable = ifp->if_capabilities;
412 
413 	/*
414 	 * Capabilities after here are not enabled by default.
415 	 */
416 
417 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
418 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
419 
420 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
421 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
422 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
423 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
424 	}
425 
426 #ifdef DEVICE_POLLING
427 	ifp->if_capabilities |= IFCAP_POLLING;
428 #endif
429 
430 	TASK_INIT(&sc->vtnet_rx_intr_task, 0, vtnet_rx_intr_task, sc);
431 	TASK_INIT(&sc->vtnet_tx_intr_task, 0, vtnet_tx_intr_task, sc);
432 	TASK_INIT(&sc->vtnet_cfgchg_task, 0, vtnet_config_change_task, sc);
433 
434 	sc->vtnet_tq = taskqueue_create_fast("vtnet_taskq", M_NOWAIT,
435 	    taskqueue_thread_enqueue, &sc->vtnet_tq);
436 	if (sc->vtnet_tq == NULL) {
437 		error = ENOMEM;
438 		device_printf(dev, "cannot allocate taskqueue\n");
439 		ether_ifdetach(ifp);
440 		goto fail;
441 	}
442 
443 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
444 	if (error) {
445 		device_printf(dev, "cannot setup virtqueue interrupts\n");
446 		ether_ifdetach(ifp);
447 		goto fail;
448 	}
449 
450 	taskqueue_start_threads(&sc->vtnet_tq, 1, PI_NET, "%s taskq",
451 	    device_get_nameunit(dev));
452 
453 	/*
454 	 * Device defaults to promiscuous mode for backwards
455 	 * compatibility. Turn it off if possible.
456 	 */
457 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
458 		VTNET_LOCK(sc);
459 		if (vtnet_set_promisc(sc, 0) != 0) {
460 			ifp->if_flags |= IFF_PROMISC;
461 			device_printf(dev,
462 			    "cannot disable promiscuous mode\n");
463 		}
464 		VTNET_UNLOCK(sc);
465 	} else
466 		ifp->if_flags |= IFF_PROMISC;
467 
468 fail:
469 	if (error)
470 		vtnet_detach(dev);
471 
472 	return (error);
473 }
474 
475 static int
476 vtnet_detach(device_t dev)
477 {
478 	struct vtnet_softc *sc;
479 	struct ifnet *ifp;
480 
481 	sc = device_get_softc(dev);
482 	ifp = sc->vtnet_ifp;
483 
484 	KASSERT(mtx_initialized(VTNET_MTX(sc)),
485 	    ("vtnet mutex not initialized"));
486 
487 #ifdef DEVICE_POLLING
488 	if (ifp != NULL && ifp->if_capenable & IFCAP_POLLING)
489 		ether_poll_deregister(ifp);
490 #endif
491 
492 	if (device_is_attached(dev)) {
493 		VTNET_LOCK(sc);
494 		vtnet_stop(sc);
495 		VTNET_UNLOCK(sc);
496 
497 		callout_drain(&sc->vtnet_tick_ch);
498 		taskqueue_drain(taskqueue_fast, &sc->vtnet_cfgchg_task);
499 
500 		ether_ifdetach(ifp);
501 	}
502 
503 	if (sc->vtnet_tq != NULL) {
504 		taskqueue_drain(sc->vtnet_tq, &sc->vtnet_rx_intr_task);
505 		taskqueue_drain(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
506 		taskqueue_free(sc->vtnet_tq);
507 		sc->vtnet_tq = NULL;
508 	}
509 
510 	if (sc->vtnet_vlan_attach != NULL) {
511 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
512 		sc->vtnet_vlan_attach = NULL;
513 	}
514 	if (sc->vtnet_vlan_detach != NULL) {
515 		EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
516 		sc->vtnet_vlan_detach = NULL;
517 	}
518 
519 	if (sc->vtnet_mac_filter != NULL) {
520 		free(sc->vtnet_mac_filter, M_DEVBUF);
521 		sc->vtnet_mac_filter = NULL;
522 	}
523 
524 	if (ifp != NULL) {
525 		if_free(ifp);
526 		sc->vtnet_ifp = NULL;
527 	}
528 
529 	if (sc->vtnet_rx_vq != NULL)
530 		vtnet_free_rx_mbufs(sc);
531 	if (sc->vtnet_tx_vq != NULL)
532 		vtnet_free_tx_mbufs(sc);
533 	if (sc->vtnet_ctrl_vq != NULL)
534 		vtnet_free_ctrl_vq(sc);
535 
536 	ifmedia_removeall(&sc->vtnet_media);
537 	VTNET_LOCK_DESTROY(sc);
538 
539 	return (0);
540 }
541 
542 static int
543 vtnet_suspend(device_t dev)
544 {
545 	struct vtnet_softc *sc;
546 
547 	sc = device_get_softc(dev);
548 
549 	VTNET_LOCK(sc);
550 	vtnet_stop(sc);
551 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
552 	VTNET_UNLOCK(sc);
553 
554 	return (0);
555 }
556 
557 static int
558 vtnet_resume(device_t dev)
559 {
560 	struct vtnet_softc *sc;
561 	struct ifnet *ifp;
562 
563 	sc = device_get_softc(dev);
564 	ifp = sc->vtnet_ifp;
565 
566 	VTNET_LOCK(sc);
567 	if (ifp->if_flags & IFF_UP)
568 		vtnet_init_locked(sc);
569 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
570 	VTNET_UNLOCK(sc);
571 
572 	return (0);
573 }
574 
575 static int
576 vtnet_shutdown(device_t dev)
577 {
578 
579 	/*
580 	 * Suspend already does all of what we need to
581 	 * do here; we just never expect to be resumed.
582 	 */
583 	return (vtnet_suspend(dev));
584 }
585 
586 static int
587 vtnet_config_change(device_t dev)
588 {
589 	struct vtnet_softc *sc;
590 
591 	sc = device_get_softc(dev);
592 
593 	taskqueue_enqueue_fast(taskqueue_fast, &sc->vtnet_cfgchg_task);
594 
595 	return (1);
596 }
597 
598 static void
599 vtnet_negotiate_features(struct vtnet_softc *sc)
600 {
601 	device_t dev;
602 	uint64_t mask, features;
603 
604 	dev = sc->vtnet_dev;
605 	mask = 0;
606 
607 	if (vtnet_csum_disable)
608 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
609 
610 	/*
611 	 * TSO and LRO are only available when their corresponding
612 	 * checksum offload feature is also negotiated.
613 	 */
614 
615 	if (vtnet_csum_disable || vtnet_tso_disable)
616 		mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
617 		    VIRTIO_NET_F_HOST_ECN;
618 
619 	if (vtnet_csum_disable || vtnet_lro_disable)
620 		mask |= VTNET_LRO_FEATURES;
621 
622 	features = VTNET_FEATURES & ~mask;
623 #ifdef VTNET_TX_INTR_MODERATION
624 	features |= VIRTIO_F_NOTIFY_ON_EMPTY;
625 #endif
626 	sc->vtnet_features = virtio_negotiate_features(dev, features);
627 
628 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0 &&
629 	    virtio_with_feature(dev, VTNET_LRO_FEATURES)) {
630 		/*
631 		 * LRO without mergeable buffers requires special care. This
632 		 * is not ideal because every receive buffer must be large
633 		 * enough to hold the maximum TCP packet, the Ethernet header,
634 		 * and the vtnet_rx_header. This requires up to 34 descriptors
635 		 * when using MCLBYTES clusters. If we do not have indirect
636 		 * descriptors, LRO is disabled since the virtqueue will not
637 		 * be able to contain very many receive buffers.
638 		 */
639 		if (virtio_with_feature(dev,
640 		    VIRTIO_RING_F_INDIRECT_DESC) == 0) {
641 			device_printf(dev,
642 			    "LRO disabled due to lack of both mergeable "
643 			    "buffers and indirect descriptors\n");
644 
645 			sc->vtnet_features = virtio_negotiate_features(dev,
646 			    features & ~VTNET_LRO_FEATURES);
647 		} else
648 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
649 	}
650 }
651 
652 static int
653 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
654 {
655 	device_t dev;
656 	struct vq_alloc_info vq_info[3];
657 	int nvqs, rxsegs;
658 
659 	dev = sc->vtnet_dev;
660 	nvqs = 2;
661 
662 	/*
663 	 * Indirect descriptors are not needed for the Rx
664 	 * virtqueue when mergeable buffers are negotiated.
665 	 * The header is placed inline with the data, not
666 	 * in a separate descriptor, and mbuf clusters are
667 	 * always physically contiguous.
668 	 */
669 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
670 		rxsegs = sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ?
671 		    VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
672 	} else
673 		rxsegs = 0;
674 
675 	VQ_ALLOC_INFO_INIT(&vq_info[0], rxsegs,
676 	    vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
677 	    "%s receive", device_get_nameunit(dev));
678 
679 	VQ_ALLOC_INFO_INIT(&vq_info[1], VTNET_MAX_TX_SEGS,
680 	    vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
681 	    "%s transmit", device_get_nameunit(dev));
682 
683 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
684 		nvqs++;
685 
686 		VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
687 		    &sc->vtnet_ctrl_vq, "%s control",
688 		    device_get_nameunit(dev));
689 	}
690 
691 	return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
692 }
693 
694 static void
695 vtnet_get_hwaddr(struct vtnet_softc *sc)
696 {
697 	device_t dev;
698 
699 	dev = sc->vtnet_dev;
700 
701 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
702 		virtio_read_device_config(dev,
703 		    offsetof(struct virtio_net_config, mac),
704 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
705 	} else {
706 		/* Generate random locally administered unicast address. */
707 		sc->vtnet_hwaddr[0] = 0xB2;
708 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
709 
710 		vtnet_set_hwaddr(sc);
711 	}
712 }
713 
714 static void
715 vtnet_set_hwaddr(struct vtnet_softc *sc)
716 {
717 	device_t dev;
718 
719 	dev = sc->vtnet_dev;
720 
721 	virtio_write_device_config(dev,
722 	    offsetof(struct virtio_net_config, mac),
723 	    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
724 }
725 
726 static int
727 vtnet_is_link_up(struct vtnet_softc *sc)
728 {
729 	device_t dev;
730 	struct ifnet *ifp;
731 	uint16_t status;
732 
733 	dev = sc->vtnet_dev;
734 	ifp = sc->vtnet_ifp;
735 
736 	VTNET_LOCK_ASSERT(sc);
737 
738 	if ((ifp->if_capenable & IFCAP_LINKSTATE) == 0)
739 		return (1);
740 
741 	status = virtio_read_dev_config_2(dev,
742 	    offsetof(struct virtio_net_config, status));
743 
744 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
745 }
746 
747 static void
748 vtnet_update_link_status(struct vtnet_softc *sc)
749 {
750 	struct ifnet *ifp;
751 	int link;
752 
753 	ifp = sc->vtnet_ifp;
754 
755 	link = vtnet_is_link_up(sc);
756 
757 	if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
758 		sc->vtnet_flags |= VTNET_FLAG_LINK;
759 		if_link_state_change(ifp, LINK_STATE_UP);
760 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
761 			vtnet_start_locked(ifp);
762 	} else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
763 		sc->vtnet_flags &= ~VTNET_FLAG_LINK;
764 		if_link_state_change(ifp, LINK_STATE_DOWN);
765 	}
766 }
767 
768 static void
769 vtnet_watchdog(struct vtnet_softc *sc)
770 {
771 	struct ifnet *ifp;
772 
773 	ifp = sc->vtnet_ifp;
774 
775 #ifdef VTNET_TX_INTR_MODERATION
776 	vtnet_txeof(sc);
777 #endif
778 
779 	if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
780 		return;
781 
782 	if_printf(ifp, "watchdog timeout -- resetting\n");
783 #ifdef VTNET_DEBUG
784 	virtqueue_dump(sc->vtnet_tx_vq);
785 #endif
786 	ifp->if_oerrors++;
787 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
788 	vtnet_init_locked(sc);
789 }
790 
791 static void
792 vtnet_config_change_task(void *arg, int pending)
793 {
794 	struct vtnet_softc *sc;
795 
796 	sc = arg;
797 
798 	VTNET_LOCK(sc);
799 	vtnet_update_link_status(sc);
800 	VTNET_UNLOCK(sc);
801 }
802 
803 static int
804 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
805 {
806 	struct vtnet_softc *sc;
807 	struct ifreq *ifr;
808 	int reinit, mask, error;
809 
810 	sc = ifp->if_softc;
811 	ifr = (struct ifreq *) data;
812 	reinit = 0;
813 	error = 0;
814 
815 	switch (cmd) {
816 	case SIOCSIFMTU:
817 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
818 			error = EINVAL;
819 		else if (ifp->if_mtu != ifr->ifr_mtu) {
820 			VTNET_LOCK(sc);
821 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
822 			VTNET_UNLOCK(sc);
823 		}
824 		break;
825 
826 	case SIOCSIFFLAGS:
827 		VTNET_LOCK(sc);
828 		if ((ifp->if_flags & IFF_UP) == 0) {
829 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
830 				vtnet_stop(sc);
831 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
832 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
833 			    (IFF_PROMISC | IFF_ALLMULTI)) {
834 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
835 					vtnet_rx_filter(sc);
836 				else
837 					error = ENOTSUP;
838 			}
839 		} else
840 			vtnet_init_locked(sc);
841 
842 		if (error == 0)
843 			sc->vtnet_if_flags = ifp->if_flags;
844 		VTNET_UNLOCK(sc);
845 		break;
846 
847 	case SIOCADDMULTI:
848 	case SIOCDELMULTI:
849 		VTNET_LOCK(sc);
850 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
851 		    (ifp->if_drv_flags & IFF_DRV_RUNNING))
852 			vtnet_rx_filter_mac(sc);
853 		VTNET_UNLOCK(sc);
854 		break;
855 
856 	case SIOCSIFMEDIA:
857 	case SIOCGIFMEDIA:
858 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
859 		break;
860 
861 	case SIOCSIFCAP:
862 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
863 
864 #ifdef DEVICE_POLLING
865 		if (mask & IFCAP_POLLING) {
866 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
867 				error = ether_poll_register(vtnet_poll, ifp);
868 				if (error)
869 					break;
870 
871 				VTNET_LOCK(sc);
872 				vtnet_disable_rx_intr(sc);
873 				vtnet_disable_tx_intr(sc);
874 				ifp->if_capenable |= IFCAP_POLLING;
875 				VTNET_UNLOCK(sc);
876 			} else {
877 				error = ether_poll_deregister(ifp);
878 
879 				/* Enable interrupts even in error case. */
880 				VTNET_LOCK(sc);
881 				vtnet_enable_tx_intr(sc);
882 				vtnet_enable_rx_intr(sc);
883 				ifp->if_capenable &= ~IFCAP_POLLING;
884 				VTNET_UNLOCK(sc);
885 			}
886 		}
887 #endif
888 		VTNET_LOCK(sc);
889 
890 		if (mask & IFCAP_TXCSUM) {
891 			ifp->if_capenable ^= IFCAP_TXCSUM;
892 			if (ifp->if_capenable & IFCAP_TXCSUM)
893 				ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
894 			else
895 				ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
896 		}
897 
898 		if (mask & IFCAP_TSO4) {
899 			ifp->if_capenable ^= IFCAP_TSO4;
900 			if (ifp->if_capenable & IFCAP_TSO4)
901 				ifp->if_hwassist |= CSUM_TSO;
902 			else
903 				ifp->if_hwassist &= ~CSUM_TSO;
904 		}
905 
906 		if (mask & IFCAP_RXCSUM) {
907 			ifp->if_capenable ^= IFCAP_RXCSUM;
908 			reinit = 1;
909 		}
910 
911 		if (mask & IFCAP_LRO) {
912 			ifp->if_capenable ^= IFCAP_LRO;
913 			reinit = 1;
914 		}
915 
916 		if (mask & IFCAP_VLAN_HWFILTER) {
917 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
918 			reinit = 1;
919 		}
920 
921 		if (mask & IFCAP_VLAN_HWTSO)
922 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
923 
924 		if (mask & IFCAP_VLAN_HWTAGGING)
925 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
926 
927 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
928 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
929 			vtnet_init_locked(sc);
930 		}
931 		VLAN_CAPABILITIES(ifp);
932 
933 		VTNET_UNLOCK(sc);
934 		break;
935 
936 	default:
937 		error = ether_ioctl(ifp, cmd, data);
938 		break;
939 	}
940 
941 	VTNET_LOCK_ASSERT_NOTOWNED(sc);
942 
943 	return (error);
944 }
945 
946 static int
947 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
948 {
949 	struct ifnet *ifp;
950 	int new_frame_size, clsize;
951 
952 	ifp = sc->vtnet_ifp;
953 
954 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
955 		new_frame_size = sizeof(struct vtnet_rx_header) +
956 		    sizeof(struct ether_vlan_header) + new_mtu;
957 
958 		if (new_frame_size > MJUM9BYTES)
959 			return (EINVAL);
960 
961 		if (new_frame_size <= MCLBYTES)
962 			clsize = MCLBYTES;
963 		else
964 			clsize = MJUM9BYTES;
965 	} else {
966 		new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
967 		    sizeof(struct ether_vlan_header) + new_mtu;
968 
969 		if (new_frame_size <= MCLBYTES)
970 			clsize = MCLBYTES;
971 		else
972 			clsize = MJUMPAGESIZE;
973 	}
974 
975 	sc->vtnet_rx_mbuf_size = clsize;
976 	sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
977 	KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
978 	    ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
979 
980 	ifp->if_mtu = new_mtu;
981 
982 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
983 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
984 		vtnet_init_locked(sc);
985 	}
986 
987 	return (0);
988 }
989 
990 static int
991 vtnet_init_rx_vq(struct vtnet_softc *sc)
992 {
993 	struct virtqueue *vq;
994 	int nbufs, error;
995 
996 	vq = sc->vtnet_rx_vq;
997 	nbufs = 0;
998 	error = ENOSPC;
999 
1000 	while (!virtqueue_full(vq)) {
1001 		if ((error = vtnet_newbuf(sc)) != 0)
1002 			break;
1003 		nbufs++;
1004 	}
1005 
1006 	if (nbufs > 0) {
1007 		virtqueue_notify(vq);
1008 
1009 		/*
1010 		 * EMSGSIZE signifies the virtqueue did not have enough
1011 		 * entries available to hold the last mbuf. This is not
1012 		 * an error. We should not get ENOSPC since we check if
1013 		 * the virtqueue is full before attempting to add a
1014 		 * buffer.
1015 		 */
1016 		if (error == EMSGSIZE)
1017 			error = 0;
1018 	}
1019 
1020 	return (error);
1021 }
1022 
1023 static void
1024 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
1025 {
1026 	struct virtqueue *vq;
1027 	struct mbuf *m;
1028 	int last;
1029 
1030 	vq = sc->vtnet_rx_vq;
1031 	last = 0;
1032 
1033 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1034 		m_freem(m);
1035 
1036 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
1037 }
1038 
1039 static void
1040 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1041 {
1042 	struct virtqueue *vq;
1043 	struct vtnet_tx_header *txhdr;
1044 	int last;
1045 
1046 	vq = sc->vtnet_tx_vq;
1047 	last = 0;
1048 
1049 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1050 		m_freem(txhdr->vth_mbuf);
1051 		uma_zfree(vtnet_tx_header_zone, txhdr);
1052 	}
1053 
1054 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1055 }
1056 
1057 static void
1058 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1059 {
1060 
1061 	/*
1062 	 * The control virtqueue is only polled, therefore
1063 	 * it should already be empty.
1064 	 */
1065 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1066 	    ("Ctrl Vq not empty"));
1067 }
1068 
1069 #ifdef DEVICE_POLLING
1070 static int
1071 vtnet_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1072 {
1073 	struct vtnet_softc *sc;
1074 	int rx_done;
1075 
1076 	sc = ifp->if_softc;
1077 	rx_done = 0;
1078 
1079 	VTNET_LOCK(sc);
1080 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1081 		if (cmd == POLL_AND_CHECK_STATUS)
1082 			vtnet_update_link_status(sc);
1083 
1084 		if (virtqueue_nused(sc->vtnet_rx_vq) > 0)
1085 			vtnet_rxeof(sc, count, &rx_done);
1086 
1087 		vtnet_txeof(sc);
1088 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1089 			vtnet_start_locked(ifp);
1090 	}
1091 	VTNET_UNLOCK(sc);
1092 
1093 	return (rx_done);
1094 }
1095 #endif /* DEVICE_POLLING */
1096 
1097 static struct mbuf *
1098 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1099 {
1100 	struct mbuf *m_head, *m_tail, *m;
1101 	int i, clsize;
1102 
1103 	clsize = sc->vtnet_rx_mbuf_size;
1104 
1105 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1106 	if (m_head == NULL)
1107 		goto fail;
1108 
1109 	m_head->m_len = clsize;
1110 	m_tail = m_head;
1111 
1112 	if (nbufs > 1) {
1113 		KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1114 		    ("chained Rx mbuf requested without LRO_NOMRG"));
1115 
1116 		for (i = 1; i < nbufs; i++) {
1117 			m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1118 			if (m == NULL)
1119 				goto fail;
1120 
1121 			m->m_len = clsize;
1122 			m_tail->m_next = m;
1123 			m_tail = m;
1124 		}
1125 	}
1126 
1127 	if (m_tailp != NULL)
1128 		*m_tailp = m_tail;
1129 
1130 	return (m_head);
1131 
1132 fail:
1133 	sc->vtnet_stats.mbuf_alloc_failed++;
1134 	m_freem(m_head);
1135 
1136 	return (NULL);
1137 }
1138 
1139 static int
1140 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1141 {
1142 	struct mbuf *m, *m_prev;
1143 	struct mbuf *m_new, *m_tail;
1144 	int len, clsize, nreplace, error;
1145 
1146 	m = m0;
1147 	m_prev = NULL;
1148 	len = len0;
1149 
1150 	m_tail = NULL;
1151 	clsize = sc->vtnet_rx_mbuf_size;
1152 	nreplace = 0;
1153 
1154 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1155 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1156 
1157 	/*
1158 	 * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1159 	 * allocating an entire chain for each received frame. When
1160 	 * the received frame's length is less than that of the chain,
1161 	 * the unused mbufs are reassigned to the new chain.
1162 	 */
1163 	while (len > 0) {
1164 		/*
1165 		 * Something is seriously wrong if we received
1166 		 * a frame larger than the mbuf chain. Drop it.
1167 		 */
1168 		if (m == NULL) {
1169 			sc->vtnet_stats.rx_frame_too_large++;
1170 			return (EMSGSIZE);
1171 		}
1172 
1173 		KASSERT(m->m_len == clsize,
1174 		    ("mbuf length not expected cluster size: %d",
1175 		    m->m_len));
1176 
1177 		m->m_len = MIN(m->m_len, len);
1178 		len -= m->m_len;
1179 
1180 		m_prev = m;
1181 		m = m->m_next;
1182 		nreplace++;
1183 	}
1184 
1185 	KASSERT(m_prev != NULL, ("m_prev == NULL"));
1186 	KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1187 	    ("too many replacement mbufs: %d/%d", nreplace,
1188 	    sc->vtnet_rx_mbuf_count));
1189 
1190 	m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1191 	if (m_new == NULL) {
1192 		m_prev->m_len = clsize;
1193 		return (ENOBUFS);
1194 	}
1195 
1196 	/*
1197 	 * Move unused mbufs, if any, from the original chain
1198 	 * onto the end of the new chain.
1199 	 */
1200 	if (m_prev->m_next != NULL) {
1201 		m_tail->m_next = m_prev->m_next;
1202 		m_prev->m_next = NULL;
1203 	}
1204 
1205 	error = vtnet_enqueue_rxbuf(sc, m_new);
1206 	if (error) {
1207 		/*
1208 		 * BAD! We could not enqueue the replacement mbuf chain. We
1209 		 * must restore the m0 chain to the original state if it was
1210 		 * modified so we can subsequently discard it.
1211 		 *
1212 		 * NOTE: The replacement is suppose to be an identical copy
1213 		 * to the one just dequeued so this is an unexpected error.
1214 		 */
1215 		sc->vtnet_stats.rx_enq_replacement_failed++;
1216 
1217 		if (m_tail->m_next != NULL) {
1218 			m_prev->m_next = m_tail->m_next;
1219 			m_tail->m_next = NULL;
1220 		}
1221 
1222 		m_prev->m_len = clsize;
1223 		m_freem(m_new);
1224 	}
1225 
1226 	return (error);
1227 }
1228 
1229 static int
1230 vtnet_newbuf(struct vtnet_softc *sc)
1231 {
1232 	struct mbuf *m;
1233 	int error;
1234 
1235 	m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1236 	if (m == NULL)
1237 		return (ENOBUFS);
1238 
1239 	error = vtnet_enqueue_rxbuf(sc, m);
1240 	if (error)
1241 		m_freem(m);
1242 
1243 	return (error);
1244 }
1245 
1246 static void
1247 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1248 {
1249 	struct virtqueue *vq;
1250 	struct mbuf *m;
1251 
1252 	vq = sc->vtnet_rx_vq;
1253 
1254 	while (--nbufs > 0) {
1255 		if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1256 			break;
1257 		vtnet_discard_rxbuf(sc, m);
1258 	}
1259 }
1260 
1261 static void
1262 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1263 {
1264 	int error;
1265 
1266 	/*
1267 	 * Requeue the discarded mbuf. This should always be
1268 	 * successful since it was just dequeued.
1269 	 */
1270 	error = vtnet_enqueue_rxbuf(sc, m);
1271 	KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1272 }
1273 
1274 static int
1275 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1276 {
1277 	struct sglist sg;
1278 	struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1279 	struct vtnet_rx_header *rxhdr;
1280 	struct virtio_net_hdr *hdr;
1281 	uint8_t *mdata;
1282 	int offset, error;
1283 
1284 	VTNET_LOCK_ASSERT(sc);
1285 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1286 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1287 
1288 	sglist_init(&sg, VTNET_MAX_RX_SEGS, segs);
1289 
1290 	mdata = mtod(m, uint8_t *);
1291 	offset = 0;
1292 
1293 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1294 		rxhdr = (struct vtnet_rx_header *) mdata;
1295 		hdr = &rxhdr->vrh_hdr;
1296 		offset += sizeof(struct vtnet_rx_header);
1297 
1298 		error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1299 		KASSERT(error == 0, ("cannot add header to sglist"));
1300 	}
1301 
1302 	error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1303 	if (error)
1304 		return (error);
1305 
1306 	if (m->m_next != NULL) {
1307 		error = sglist_append_mbuf(&sg, m->m_next);
1308 		if (error)
1309 			return (error);
1310 	}
1311 
1312 	return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1313 }
1314 
1315 static void
1316 vtnet_vlan_tag_remove(struct mbuf *m)
1317 {
1318 	struct ether_vlan_header *evl;
1319 
1320 	evl = mtod(m, struct ether_vlan_header *);
1321 
1322 	m->m_pkthdr.ether_vtag = ntohs(evl->evl_tag);
1323 	m->m_flags |= M_VLANTAG;
1324 
1325 	/* Strip the 802.1Q header. */
1326 	bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1327 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
1328 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
1329 }
1330 
1331 #ifdef notyet
1332 static int
1333 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1334     struct virtio_net_hdr *hdr)
1335 {
1336 	struct ether_header *eh;
1337 	struct ether_vlan_header *evh;
1338 	struct ip *ip;
1339 	struct ip6_hdr *ip6;
1340 	struct udphdr *udp;
1341 	int ip_offset, csum_start, csum_offset, hlen;
1342 	uint16_t eth_type;
1343 	uint8_t ip_proto;
1344 
1345 	/*
1346 	 * Convert the VirtIO checksum interface to FreeBSD's interface.
1347 	 * The host only provides us with the offset at which to start
1348 	 * checksumming, and the offset from that to place the completed
1349 	 * checksum. While this maps well with how Linux does checksums,
1350 	 * for FreeBSD, we must parse the received packet in order to set
1351 	 * the appropriate CSUM_* flags.
1352 	 */
1353 
1354 	/*
1355 	 * Every mbuf added to the receive virtqueue is always at least
1356 	 * MCLBYTES big, so assume something is amiss if the first mbuf
1357 	 * does not contain both the Ethernet and protocol headers.
1358 	 */
1359 	ip_offset = sizeof(struct ether_header);
1360 	if (m->m_len < ip_offset)
1361 		return (1);
1362 
1363 	eh = mtod(m, struct ether_header *);
1364 	eth_type = ntohs(eh->ether_type);
1365 	if (eth_type == ETHERTYPE_VLAN) {
1366 		ip_offset = sizeof(struct ether_vlan_header);
1367 		if (m->m_len < ip_offset)
1368 			return (1);
1369 		evh = mtod(m, struct ether_vlan_header *);
1370 		eth_type = ntohs(evh->evl_proto);
1371 	}
1372 
1373 	switch (eth_type) {
1374 	case ETHERTYPE_IP:
1375 		if (m->m_len < ip_offset + sizeof(struct ip))
1376 			return (1);
1377 
1378 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1379 		 /* Sanity check the IP header. */
1380 		if (ip->ip_v != IPVERSION)
1381 			return (1);
1382 		hlen = ip->ip_hl << 2;
1383 		if (hlen < sizeof(struct ip))
1384 			return (1);
1385 		if (ntohs(ip->ip_len) < hlen)
1386 			return (1);
1387 		if (ntohs(ip->ip_len) != (m->m_pkthdr.len - ip_offset))
1388 			return (1);
1389 
1390 		ip_proto = ip->ip_p;
1391 		csum_start = ip_offset + hlen;
1392 		break;
1393 
1394 	case ETHERTYPE_IPV6:
1395 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr))
1396 			return (1);
1397 
1398 		/*
1399 		 * XXX FreeBSD does not handle any IPv6 checksum offloading
1400 		 * at the moment.
1401 		 */
1402 
1403 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1404 		/* XXX Assume no extension headers are present. */
1405 		ip_proto = ip6->ip6_nxt;
1406 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1407 		break;
1408 
1409 	default:
1410 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1411 		return (1);
1412 	}
1413 
1414 	/* Assume checksum begins right after the IP header. */
1415 	if (hdr->csum_start != csum_start) {
1416 		sc->vtnet_stats.rx_csum_bad_start++;
1417 		return (1);
1418 	}
1419 
1420 	switch (ip_proto) {
1421 	case IPPROTO_TCP:
1422 		csum_offset = offsetof(struct tcphdr, th_sum);
1423 		break;
1424 
1425 	case IPPROTO_UDP:
1426 		csum_offset = offsetof(struct udphdr, uh_sum);
1427 		break;
1428 
1429 	case IPPROTO_SCTP:
1430 		csum_offset = offsetof(struct sctphdr, checksum);
1431 		break;
1432 
1433 	default:
1434 		sc->vtnet_stats.rx_csum_bad_ipproto++;
1435 		return (1);
1436 	}
1437 
1438 	if (hdr->csum_offset != csum_offset) {
1439 		sc->vtnet_stats.rx_csum_bad_offset++;
1440 		return (1);
1441 	}
1442 
1443 	/*
1444 	 * The IP header checksum is almost certainly valid but I'm
1445 	 * uncertain if that is guaranteed.
1446 	 *
1447 	 * m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
1448 	 */
1449 
1450 	switch (ip_proto) {
1451 	case IPPROTO_UDP:
1452 		if (m->m_len < csum_start + sizeof(struct udphdr))
1453 			return (1);
1454 
1455 		udp = (struct udphdr *)(mtod(m, uint8_t *) + csum_start);
1456 		if (udp->uh_sum == 0)
1457 			return (0);
1458 
1459 		/* FALLTHROUGH */
1460 
1461 	case IPPROTO_TCP:
1462 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1463 		m->m_pkthdr.csum_data = 0xFFFF;
1464 		break;
1465 
1466 	case IPPROTO_SCTP:
1467 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1468 		break;
1469 	}
1470 
1471 	sc->vtnet_stats.rx_csum_offloaded++;
1472 
1473 	return (0);
1474 }
1475 #endif
1476 
1477 /*
1478  * Alternative method of doing receive checksum offloading. Rather
1479  * than parsing the received frame down to the IP header, use the
1480  * csum_offset to determine which CSUM_* flags are appropriate. We
1481  * can get by with doing this only because the checksum offsets are
1482  * unique for the things we care about.
1483  */
1484 static int
1485 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1486     struct virtio_net_hdr *hdr)
1487 {
1488 	struct ether_header *eh;
1489 	struct ether_vlan_header *evh;
1490 	struct udphdr *udp;
1491 	int csum_len;
1492 	uint16_t eth_type;
1493 
1494 	csum_len = hdr->csum_start + hdr->csum_offset;
1495 
1496 	if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1497 		return (1);
1498 	if (m->m_len < csum_len)
1499 		return (1);
1500 
1501 	eh = mtod(m, struct ether_header *);
1502 	eth_type = ntohs(eh->ether_type);
1503 	if (eth_type == ETHERTYPE_VLAN) {
1504 		evh = mtod(m, struct ether_vlan_header *);
1505 		eth_type = ntohs(evh->evl_proto);
1506 	}
1507 
1508 	if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1509 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1510 		return (1);
1511 	}
1512 
1513 	/* Use the offset to determine the appropriate CSUM_* flags. */
1514 	switch (hdr->csum_offset) {
1515 	case offsetof(struct udphdr, uh_sum):
1516 		if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1517 			return (1);
1518 		udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1519 		if (udp->uh_sum == 0)
1520 			return (0);
1521 
1522 		/* FALLTHROUGH */
1523 
1524 	case offsetof(struct tcphdr, th_sum):
1525 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1526 		m->m_pkthdr.csum_data = 0xFFFF;
1527 		break;
1528 
1529 	case offsetof(struct sctphdr, checksum):
1530 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1531 		break;
1532 
1533 	default:
1534 		sc->vtnet_stats.rx_csum_bad_offset++;
1535 		return (1);
1536 	}
1537 
1538 	sc->vtnet_stats.rx_csum_offloaded++;
1539 
1540 	return (0);
1541 }
1542 
1543 static int
1544 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1545 {
1546 	struct ifnet *ifp;
1547 	struct virtqueue *vq;
1548 	struct mbuf *m, *m_tail;
1549 	int len;
1550 
1551 	ifp = sc->vtnet_ifp;
1552 	vq = sc->vtnet_rx_vq;
1553 	m_tail = m_head;
1554 
1555 	while (--nbufs > 0) {
1556 		m = virtqueue_dequeue(vq, &len);
1557 		if (m == NULL) {
1558 			ifp->if_ierrors++;
1559 			goto fail;
1560 		}
1561 
1562 		if (vtnet_newbuf(sc) != 0) {
1563 			ifp->if_iqdrops++;
1564 			vtnet_discard_rxbuf(sc, m);
1565 			if (nbufs > 1)
1566 				vtnet_discard_merged_rxbuf(sc, nbufs);
1567 			goto fail;
1568 		}
1569 
1570 		if (m->m_len < len)
1571 			len = m->m_len;
1572 
1573 		m->m_len = len;
1574 		m->m_flags &= ~M_PKTHDR;
1575 
1576 		m_head->m_pkthdr.len += len;
1577 		m_tail->m_next = m;
1578 		m_tail = m;
1579 	}
1580 
1581 	return (0);
1582 
1583 fail:
1584 	sc->vtnet_stats.rx_mergeable_failed++;
1585 	m_freem(m_head);
1586 
1587 	return (1);
1588 }
1589 
1590 static int
1591 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1592 {
1593 	struct virtio_net_hdr lhdr;
1594 	struct ifnet *ifp;
1595 	struct virtqueue *vq;
1596 	struct mbuf *m;
1597 	struct ether_header *eh;
1598 	struct virtio_net_hdr *hdr;
1599 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1600 	int len, deq, nbufs, adjsz, rx_npkts;
1601 
1602 	ifp = sc->vtnet_ifp;
1603 	vq = sc->vtnet_rx_vq;
1604 	hdr = &lhdr;
1605 	deq = 0;
1606 	rx_npkts = 0;
1607 
1608 	VTNET_LOCK_ASSERT(sc);
1609 
1610 	while (--count >= 0) {
1611 		m = virtqueue_dequeue(vq, &len);
1612 		if (m == NULL)
1613 			break;
1614 		deq++;
1615 
1616 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1617 			ifp->if_ierrors++;
1618 			vtnet_discard_rxbuf(sc, m);
1619 			continue;
1620 		}
1621 
1622 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1623 			nbufs = 1;
1624 			adjsz = sizeof(struct vtnet_rx_header);
1625 			/*
1626 			 * Account for our pad between the header and
1627 			 * the actual start of the frame.
1628 			 */
1629 			len += VTNET_RX_HEADER_PAD;
1630 		} else {
1631 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1632 			nbufs = mhdr->num_buffers;
1633 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1634 		}
1635 
1636 		if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1637 			ifp->if_iqdrops++;
1638 			vtnet_discard_rxbuf(sc, m);
1639 			if (nbufs > 1)
1640 				vtnet_discard_merged_rxbuf(sc, nbufs);
1641 			continue;
1642 		}
1643 
1644 		m->m_pkthdr.len = len;
1645 		m->m_pkthdr.rcvif = ifp;
1646 		m->m_pkthdr.csum_flags = 0;
1647 
1648 		if (nbufs > 1) {
1649 			if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1650 				continue;
1651 		}
1652 
1653 		ifp->if_ipackets++;
1654 
1655 		/*
1656 		 * Save copy of header before we strip it. For both mergeable
1657 		 * and non-mergeable, the VirtIO header is placed first in the
1658 		 * mbuf's data. We no longer need num_buffers, so always use a
1659 		 * virtio_net_hdr.
1660 		 */
1661 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1662 		m_adj(m, adjsz);
1663 
1664 		if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1665 			eh = mtod(m, struct ether_header *);
1666 			if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1667 				vtnet_vlan_tag_remove(m);
1668 
1669 				/*
1670 				 * With the 802.1Q header removed, update the
1671 				 * checksum starting location accordingly.
1672 				 */
1673 				if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1674 					hdr->csum_start -=
1675 					    ETHER_VLAN_ENCAP_LEN;
1676 			}
1677 		}
1678 
1679 		if (ifp->if_capenable & IFCAP_RXCSUM &&
1680 		    hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1681 			if (vtnet_rx_csum(sc, m, hdr) != 0)
1682 				sc->vtnet_stats.rx_csum_failed++;
1683 		}
1684 
1685 		VTNET_UNLOCK(sc);
1686 		rx_npkts++;
1687 		(*ifp->if_input)(ifp, m);
1688 		VTNET_LOCK(sc);
1689 
1690 		/*
1691 		 * The interface may have been stopped while we were
1692 		 * passing the packet up the network stack.
1693 		 */
1694 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1695 			break;
1696 	}
1697 
1698 	if (deq > 0)
1699 		virtqueue_notify(vq);
1700 
1701 	if (rx_npktsp != NULL)
1702 		*rx_npktsp = rx_npkts;
1703 
1704 	return (count > 0 ? 0 : EAGAIN);
1705 }
1706 
1707 static void
1708 vtnet_rx_intr_task(void *arg, int pending)
1709 {
1710 	struct vtnet_softc *sc;
1711 	struct ifnet *ifp;
1712 	int more;
1713 
1714 	sc = arg;
1715 	ifp = sc->vtnet_ifp;
1716 
1717 	VTNET_LOCK(sc);
1718 
1719 #ifdef DEVICE_POLLING
1720 	if (ifp->if_capenable & IFCAP_POLLING) {
1721 		VTNET_UNLOCK(sc);
1722 		return;
1723 	}
1724 #endif
1725 
1726 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1727 		vtnet_enable_rx_intr(sc);
1728 		VTNET_UNLOCK(sc);
1729 		return;
1730 	}
1731 
1732 	more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1733 	if (!more && vtnet_enable_rx_intr(sc) != 0) {
1734 		vtnet_disable_rx_intr(sc);
1735 		more = 1;
1736 	}
1737 
1738 	VTNET_UNLOCK(sc);
1739 
1740 	if (more) {
1741 		sc->vtnet_stats.rx_task_rescheduled++;
1742 		taskqueue_enqueue_fast(sc->vtnet_tq,
1743 		    &sc->vtnet_rx_intr_task);
1744 	}
1745 }
1746 
1747 static int
1748 vtnet_rx_vq_intr(void *xsc)
1749 {
1750 	struct vtnet_softc *sc;
1751 
1752 	sc = xsc;
1753 
1754 	vtnet_disable_rx_intr(sc);
1755 	taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_rx_intr_task);
1756 
1757 	return (1);
1758 }
1759 
1760 static void
1761 vtnet_txeof(struct vtnet_softc *sc)
1762 {
1763 	struct virtqueue *vq;
1764 	struct ifnet *ifp;
1765 	struct vtnet_tx_header *txhdr;
1766 	int deq;
1767 
1768 	vq = sc->vtnet_tx_vq;
1769 	ifp = sc->vtnet_ifp;
1770 	deq = 0;
1771 
1772 	VTNET_LOCK_ASSERT(sc);
1773 
1774 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1775 		deq++;
1776 		ifp->if_opackets++;
1777 		m_freem(txhdr->vth_mbuf);
1778 		uma_zfree(vtnet_tx_header_zone, txhdr);
1779 	}
1780 
1781 	if (deq > 0) {
1782 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1783 		if (virtqueue_empty(vq))
1784 			sc->vtnet_watchdog_timer = 0;
1785 	}
1786 }
1787 
1788 static struct mbuf *
1789 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1790     struct virtio_net_hdr *hdr)
1791 {
1792 	struct ifnet *ifp;
1793 	struct ether_header *eh;
1794 	struct ether_vlan_header *evh;
1795 	struct ip *ip;
1796 	struct ip6_hdr *ip6;
1797 	struct tcphdr *tcp;
1798 	int ip_offset;
1799 	uint16_t eth_type, csum_start;
1800 	uint8_t ip_proto, gso_type;
1801 
1802 	ifp = sc->vtnet_ifp;
1803 	M_ASSERTPKTHDR(m);
1804 
1805 	ip_offset = sizeof(struct ether_header);
1806 	if (m->m_len < ip_offset) {
1807 		if ((m = m_pullup(m, ip_offset)) == NULL)
1808 			return (NULL);
1809 	}
1810 
1811 	eh = mtod(m, struct ether_header *);
1812 	eth_type = ntohs(eh->ether_type);
1813 	if (eth_type == ETHERTYPE_VLAN) {
1814 		ip_offset = sizeof(struct ether_vlan_header);
1815 		if (m->m_len < ip_offset) {
1816 			if ((m = m_pullup(m, ip_offset)) == NULL)
1817 				return (NULL);
1818 		}
1819 		evh = mtod(m, struct ether_vlan_header *);
1820 		eth_type = ntohs(evh->evl_proto);
1821 	}
1822 
1823 	switch (eth_type) {
1824 	case ETHERTYPE_IP:
1825 		if (m->m_len < ip_offset + sizeof(struct ip)) {
1826 			m = m_pullup(m, ip_offset + sizeof(struct ip));
1827 			if (m == NULL)
1828 				return (NULL);
1829 		}
1830 
1831 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1832 		ip_proto = ip->ip_p;
1833 		csum_start = ip_offset + (ip->ip_hl << 2);
1834 		gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1835 		break;
1836 
1837 	case ETHERTYPE_IPV6:
1838 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1839 			m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1840 			if (m == NULL)
1841 				return (NULL);
1842 		}
1843 
1844 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1845 		/*
1846 		 * XXX Assume no extension headers are present. Presently,
1847 		 * this will always be true in the case of TSO, and FreeBSD
1848 		 * does not perform checksum offloading of IPv6 yet.
1849 		 */
1850 		ip_proto = ip6->ip6_nxt;
1851 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1852 		gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1853 		break;
1854 
1855 	default:
1856 		return (m);
1857 	}
1858 
1859 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1860 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1861 		hdr->csum_start = csum_start;
1862 		hdr->csum_offset = m->m_pkthdr.csum_data;
1863 
1864 		sc->vtnet_stats.tx_csum_offloaded++;
1865 	}
1866 
1867 	if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1868 		if (ip_proto != IPPROTO_TCP)
1869 			return (m);
1870 
1871 		if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1872 			m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1873 			if (m == NULL)
1874 				return (NULL);
1875 		}
1876 
1877 		tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1878 		hdr->gso_type = gso_type;
1879 		hdr->hdr_len = csum_start + (tcp->th_off << 2);
1880 		hdr->gso_size = m->m_pkthdr.tso_segsz;
1881 
1882 		if (tcp->th_flags & TH_CWR) {
1883 			/*
1884 			 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1885 			 * ECN support is only configurable globally with the
1886 			 * net.inet.tcp.ecn.enable sysctl knob.
1887 			 */
1888 			if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1889 				if_printf(ifp, "TSO with ECN not supported "
1890 				    "by host\n");
1891 				m_freem(m);
1892 				return (NULL);
1893 			}
1894 
1895 			hdr->flags |= VIRTIO_NET_HDR_GSO_ECN;
1896 		}
1897 
1898 		sc->vtnet_stats.tx_tso_offloaded++;
1899 	}
1900 
1901 	return (m);
1902 }
1903 
1904 static int
1905 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1906     struct vtnet_tx_header *txhdr)
1907 {
1908 	struct sglist sg;
1909 	struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1910 	struct virtqueue *vq;
1911 	struct mbuf *m;
1912 	int collapsed, error;
1913 
1914 	vq = sc->vtnet_tx_vq;
1915 	m = *m_head;
1916 	collapsed = 0;
1917 
1918 	sglist_init(&sg, VTNET_MAX_TX_SEGS, segs);
1919 	error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1920 	KASSERT(error == 0 && sg.sg_nseg == 1,
1921 	    ("cannot add header to sglist"));
1922 
1923 again:
1924 	error = sglist_append_mbuf(&sg, m);
1925 	if (error) {
1926 		if (collapsed)
1927 			goto fail;
1928 
1929 		m = m_collapse(m, M_NOWAIT, VTNET_MAX_TX_SEGS - 1);
1930 		if (m == NULL)
1931 			goto fail;
1932 
1933 		*m_head = m;
1934 		collapsed = 1;
1935 		goto again;
1936 	}
1937 
1938 	txhdr->vth_mbuf = m;
1939 
1940 	return (virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0));
1941 
1942 fail:
1943 	m_freem(*m_head);
1944 	*m_head = NULL;
1945 
1946 	return (ENOBUFS);
1947 }
1948 
1949 static int
1950 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1951 {
1952 	struct vtnet_tx_header *txhdr;
1953 	struct virtio_net_hdr *hdr;
1954 	struct mbuf *m;
1955 	int error;
1956 
1957 	m = *m_head;
1958 
1959 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
1960 	if (txhdr == NULL) {
1961 		*m_head = NULL;
1962 		m_freem(m);
1963 		return (ENOMEM);
1964 	}
1965 
1966 	/*
1967 	 * Always use the non-mergeable header to simplify things. When
1968 	 * the mergeable feature is negotiated, the num_buffers field
1969 	 * must be set to zero. We use vtnet_hdr_size later to enqueue
1970 	 * the correct header size to the host.
1971 	 */
1972 	hdr = &txhdr->vth_uhdr.hdr;
1973 
1974 	if (m->m_flags & M_VLANTAG) {
1975 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1976 		if ((*m_head = m) == NULL) {
1977 			error = ENOBUFS;
1978 			goto fail;
1979 		}
1980 		m->m_flags &= ~M_VLANTAG;
1981 	}
1982 
1983 	if (m->m_pkthdr.csum_flags != 0) {
1984 		m = vtnet_tx_offload(sc, m, hdr);
1985 		if ((*m_head = m) == NULL) {
1986 			error = ENOBUFS;
1987 			goto fail;
1988 		}
1989 	}
1990 
1991 	error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1992 fail:
1993 	if (error)
1994 		uma_zfree(vtnet_tx_header_zone, txhdr);
1995 
1996 	return (error);
1997 }
1998 
1999 static void
2000 vtnet_start(struct ifnet *ifp)
2001 {
2002 	struct vtnet_softc *sc;
2003 
2004 	sc = ifp->if_softc;
2005 
2006 	VTNET_LOCK(sc);
2007 	vtnet_start_locked(ifp);
2008 	VTNET_UNLOCK(sc);
2009 }
2010 
2011 static void
2012 vtnet_start_locked(struct ifnet *ifp)
2013 {
2014 	struct vtnet_softc *sc;
2015 	struct virtqueue *vq;
2016 	struct mbuf *m0;
2017 	int enq;
2018 
2019 	sc = ifp->if_softc;
2020 	vq = sc->vtnet_tx_vq;
2021 	enq = 0;
2022 
2023 	VTNET_LOCK_ASSERT(sc);
2024 
2025 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2026 	    IFF_DRV_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
2027 		return;
2028 
2029 #ifdef VTNET_TX_INTR_MODERATION
2030 	if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
2031 		vtnet_txeof(sc);
2032 #endif
2033 
2034 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2035 		if (virtqueue_full(vq)) {
2036 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2037 			break;
2038 		}
2039 
2040 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2041 		if (m0 == NULL)
2042 			break;
2043 
2044 		if (vtnet_encap(sc, &m0) != 0) {
2045 			if (m0 == NULL)
2046 				break;
2047 			IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2048 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2049 			break;
2050 		}
2051 
2052 		enq++;
2053 		ETHER_BPF_MTAP(ifp, m0);
2054 	}
2055 
2056 	if (enq > 0) {
2057 		virtqueue_notify(vq);
2058 		sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
2059 	}
2060 }
2061 
2062 static void
2063 vtnet_tick(void *xsc)
2064 {
2065 	struct vtnet_softc *sc;
2066 
2067 	sc = xsc;
2068 
2069 	VTNET_LOCK_ASSERT(sc);
2070 #ifdef VTNET_DEBUG
2071 	virtqueue_dump(sc->vtnet_rx_vq);
2072 	virtqueue_dump(sc->vtnet_tx_vq);
2073 #endif
2074 
2075 	vtnet_watchdog(sc);
2076 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2077 }
2078 
2079 static void
2080 vtnet_tx_intr_task(void *arg, int pending)
2081 {
2082 	struct vtnet_softc *sc;
2083 	struct ifnet *ifp;
2084 
2085 	sc = arg;
2086 	ifp = sc->vtnet_ifp;
2087 
2088 	VTNET_LOCK(sc);
2089 
2090 #ifdef DEVICE_POLLING
2091 	if (ifp->if_capenable & IFCAP_POLLING) {
2092 		VTNET_UNLOCK(sc);
2093 		return;
2094 	}
2095 #endif
2096 
2097 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2098 		vtnet_enable_tx_intr(sc);
2099 		VTNET_UNLOCK(sc);
2100 		return;
2101 	}
2102 
2103 	vtnet_txeof(sc);
2104 
2105 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2106 		vtnet_start_locked(ifp);
2107 
2108 	if (vtnet_enable_tx_intr(sc) != 0) {
2109 		vtnet_disable_tx_intr(sc);
2110 		sc->vtnet_stats.tx_task_rescheduled++;
2111 		VTNET_UNLOCK(sc);
2112 		taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
2113 		return;
2114 	}
2115 
2116 	VTNET_UNLOCK(sc);
2117 }
2118 
2119 static int
2120 vtnet_tx_vq_intr(void *xsc)
2121 {
2122 	struct vtnet_softc *sc;
2123 
2124 	sc = xsc;
2125 
2126 	vtnet_disable_tx_intr(sc);
2127 	taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
2128 
2129 	return (1);
2130 }
2131 
2132 static void
2133 vtnet_stop(struct vtnet_softc *sc)
2134 {
2135 	device_t dev;
2136 	struct ifnet *ifp;
2137 
2138 	dev = sc->vtnet_dev;
2139 	ifp = sc->vtnet_ifp;
2140 
2141 	VTNET_LOCK_ASSERT(sc);
2142 
2143 	sc->vtnet_watchdog_timer = 0;
2144 	callout_stop(&sc->vtnet_tick_ch);
2145 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2146 
2147 	vtnet_disable_rx_intr(sc);
2148 	vtnet_disable_tx_intr(sc);
2149 
2150 	/*
2151 	 * Stop the host VirtIO adapter. Note this will reset the host
2152 	 * adapter's state back to the pre-initialized state, so in
2153 	 * order to make the device usable again, we must drive it
2154 	 * through virtio_reinit() and virtio_reinit_complete().
2155 	 */
2156 	virtio_stop(dev);
2157 
2158 	sc->vtnet_flags &= ~VTNET_FLAG_LINK;
2159 
2160 	vtnet_free_rx_mbufs(sc);
2161 	vtnet_free_tx_mbufs(sc);
2162 }
2163 
2164 static int
2165 vtnet_reinit(struct vtnet_softc *sc)
2166 {
2167 	struct ifnet *ifp;
2168 	uint64_t features;
2169 
2170 	ifp = sc->vtnet_ifp;
2171 	features = sc->vtnet_features;
2172 
2173 	/*
2174 	 * Re-negotiate with the host, removing any disabled receive
2175 	 * features. Transmit features are disabled only on our side
2176 	 * via if_capenable and if_hwassist.
2177 	 */
2178 
2179 	if (ifp->if_capabilities & IFCAP_RXCSUM) {
2180 		if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2181 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2182 	}
2183 
2184 	if (ifp->if_capabilities & IFCAP_LRO) {
2185 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2186 			features &= ~VTNET_LRO_FEATURES;
2187 	}
2188 
2189 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2190 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2191 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2192 	}
2193 
2194 	return (virtio_reinit(sc->vtnet_dev, features));
2195 }
2196 
2197 static void
2198 vtnet_init_locked(struct vtnet_softc *sc)
2199 {
2200 	device_t dev;
2201 	struct ifnet *ifp;
2202 	int error;
2203 
2204 	dev = sc->vtnet_dev;
2205 	ifp = sc->vtnet_ifp;
2206 
2207 	VTNET_LOCK_ASSERT(sc);
2208 
2209 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2210 		return;
2211 
2212 	/* Stop host's adapter, cancel any pending I/O. */
2213 	vtnet_stop(sc);
2214 
2215 	/* Reinitialize the host device. */
2216 	error = vtnet_reinit(sc);
2217 	if (error) {
2218 		device_printf(dev,
2219 		    "reinitialization failed, stopping device...\n");
2220 		vtnet_stop(sc);
2221 		return;
2222 	}
2223 
2224 	/* Update host with assigned MAC address. */
2225 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2226 	vtnet_set_hwaddr(sc);
2227 
2228 	ifp->if_hwassist = 0;
2229 	if (ifp->if_capenable & IFCAP_TXCSUM)
2230 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2231 	if (ifp->if_capenable & IFCAP_TSO4)
2232 		ifp->if_hwassist |= CSUM_TSO;
2233 
2234 	error = vtnet_init_rx_vq(sc);
2235 	if (error) {
2236 		device_printf(dev,
2237 		    "cannot allocate mbufs for Rx virtqueue\n");
2238 		vtnet_stop(sc);
2239 		return;
2240 	}
2241 
2242 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2243 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2244 			/* Restore promiscuous and all-multicast modes. */
2245 			vtnet_rx_filter(sc);
2246 
2247 			/* Restore filtered MAC addresses. */
2248 			vtnet_rx_filter_mac(sc);
2249 		}
2250 
2251 		/* Restore VLAN filters. */
2252 		if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2253 			vtnet_rx_filter_vlan(sc);
2254 	}
2255 
2256 #ifdef DEVICE_POLLING
2257 	if (ifp->if_capenable & IFCAP_POLLING) {
2258 		vtnet_disable_rx_intr(sc);
2259 		vtnet_disable_tx_intr(sc);
2260 	} else
2261 #endif
2262 	{
2263 		vtnet_enable_rx_intr(sc);
2264 		vtnet_enable_tx_intr(sc);
2265 	}
2266 
2267 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2268 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2269 
2270 	virtio_reinit_complete(dev);
2271 
2272 	vtnet_update_link_status(sc);
2273 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2274 }
2275 
2276 static void
2277 vtnet_init(void *xsc)
2278 {
2279 	struct vtnet_softc *sc;
2280 
2281 	sc = xsc;
2282 
2283 	VTNET_LOCK(sc);
2284 	vtnet_init_locked(sc);
2285 	VTNET_UNLOCK(sc);
2286 }
2287 
2288 static void
2289 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2290     struct sglist *sg, int readable, int writable)
2291 {
2292 	struct virtqueue *vq;
2293 	void *c;
2294 
2295 	vq = sc->vtnet_ctrl_vq;
2296 
2297 	VTNET_LOCK_ASSERT(sc);
2298 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2299 	    ("no control virtqueue"));
2300 	KASSERT(virtqueue_empty(vq),
2301 	    ("control command already enqueued"));
2302 
2303 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2304 		return;
2305 
2306 	virtqueue_notify(vq);
2307 
2308 	/*
2309 	 * Poll until the command is complete. Previously, we would
2310 	 * sleep until the control virtqueue interrupt handler woke
2311 	 * us up, but dropping the VTNET_MTX leads to serialization
2312 	 * difficulties.
2313 	 *
2314 	 * Furthermore, it appears QEMU/KVM only allocates three MSIX
2315 	 * vectors. Two of those vectors are needed for the Rx and Tx
2316 	 * virtqueues. We do not support sharing both a Vq and config
2317 	 * changed notification on the same MSIX vector.
2318 	 */
2319 	c = virtqueue_poll(vq, NULL);
2320 	KASSERT(c == cookie, ("unexpected control command response"));
2321 }
2322 
2323 static void
2324 vtnet_rx_filter(struct vtnet_softc *sc)
2325 {
2326 	device_t dev;
2327 	struct ifnet *ifp;
2328 
2329 	dev = sc->vtnet_dev;
2330 	ifp = sc->vtnet_ifp;
2331 
2332 	VTNET_LOCK_ASSERT(sc);
2333 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2334 	    ("CTRL_RX feature not negotiated"));
2335 
2336 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2337 		device_printf(dev, "cannot %s promiscuous mode\n",
2338 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
2339 
2340 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2341 		device_printf(dev, "cannot %s all-multicast mode\n",
2342 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
2343 }
2344 
2345 static int
2346 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2347 {
2348 	struct virtio_net_ctrl_hdr hdr;
2349 	struct sglist_seg segs[3];
2350 	struct sglist sg;
2351 	uint8_t onoff, ack;
2352 	int error;
2353 
2354 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
2355 		return (ENOTSUP);
2356 
2357 	error = 0;
2358 
2359 	hdr.class = VIRTIO_NET_CTRL_RX;
2360 	hdr.cmd = cmd;
2361 	onoff = !!on;
2362 	ack = VIRTIO_NET_ERR;
2363 
2364 	sglist_init(&sg, 3, segs);
2365 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2366 	error |= sglist_append(&sg, &onoff, sizeof(uint8_t));
2367 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2368 	KASSERT(error == 0 && sg.sg_nseg == 3,
2369 	    ("error adding Rx filter message to sglist"));
2370 
2371 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2372 
2373 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2374 }
2375 
2376 static int
2377 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2378 {
2379 
2380 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2381 }
2382 
2383 static int
2384 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2385 {
2386 
2387 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2388 }
2389 
2390 static void
2391 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2392 {
2393 	struct virtio_net_ctrl_hdr hdr;
2394 	struct vtnet_mac_filter *filter;
2395 	struct sglist_seg segs[4];
2396 	struct sglist sg;
2397 	struct ifnet *ifp;
2398 	struct ifaddr *ifa;
2399 	struct ifmultiaddr *ifma;
2400 	int ucnt, mcnt, promisc, allmulti, error;
2401 	uint8_t ack;
2402 
2403 	ifp = sc->vtnet_ifp;
2404 	filter = sc->vtnet_mac_filter;
2405 	ucnt = 0;
2406 	mcnt = 0;
2407 	promisc = 0;
2408 	allmulti = 0;
2409 	error = 0;
2410 
2411 	VTNET_LOCK_ASSERT(sc);
2412 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2413 	    ("CTRL_RX feature not negotiated"));
2414 
2415 	/* Unicast MAC addresses: */
2416 	if_addr_rlock(ifp);
2417 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2418 		if (ifa->ifa_addr->sa_family != AF_LINK)
2419 			continue;
2420 		else if (ucnt == VTNET_MAX_MAC_ENTRIES)
2421 			break;
2422 
2423 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2424 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2425 		ucnt++;
2426 	}
2427 	if_addr_runlock(ifp);
2428 
2429 	if (ucnt >= VTNET_MAX_MAC_ENTRIES) {
2430 		promisc = 1;
2431 		filter->vmf_unicast.nentries = 0;
2432 
2433 		if_printf(ifp, "more than %d MAC addresses assigned, "
2434 		    "falling back to promiscuous mode\n",
2435 		    VTNET_MAX_MAC_ENTRIES);
2436 	} else
2437 		filter->vmf_unicast.nentries = ucnt;
2438 
2439 	/* Multicast MAC addresses: */
2440 	if_maddr_rlock(ifp);
2441 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2442 		if (ifma->ifma_addr->sa_family != AF_LINK)
2443 			continue;
2444 		else if (mcnt == VTNET_MAX_MAC_ENTRIES)
2445 			break;
2446 
2447 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2448 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2449 		mcnt++;
2450 	}
2451 	if_maddr_runlock(ifp);
2452 
2453 	if (mcnt >= VTNET_MAX_MAC_ENTRIES) {
2454 		allmulti = 1;
2455 		filter->vmf_multicast.nentries = 0;
2456 
2457 		if_printf(ifp, "more than %d multicast MAC addresses "
2458 		    "assigned, falling back to all-multicast mode\n",
2459 		    VTNET_MAX_MAC_ENTRIES);
2460 	} else
2461 		filter->vmf_multicast.nentries = mcnt;
2462 
2463 	if (promisc && allmulti)
2464 		goto out;
2465 
2466 	hdr.class = VIRTIO_NET_CTRL_MAC;
2467 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2468 	ack = VIRTIO_NET_ERR;
2469 
2470 	sglist_init(&sg, 4, segs);
2471 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2472 	error |= sglist_append(&sg, &filter->vmf_unicast,
2473 	    sizeof(struct vtnet_mac_table));
2474 	error |= sglist_append(&sg, &filter->vmf_multicast,
2475 	    sizeof(struct vtnet_mac_table));
2476 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2477 	KASSERT(error == 0 && sg.sg_nseg == 4,
2478 	    ("error adding MAC filtering message to sglist"));
2479 
2480 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2481 
2482 	if (ack != VIRTIO_NET_OK)
2483 		if_printf(ifp, "error setting host MAC filter table\n");
2484 
2485 out:
2486 	if (promisc)
2487 		if (vtnet_set_promisc(sc, 1) != 0)
2488 			if_printf(ifp, "cannot enable promiscuous mode\n");
2489 	if (allmulti)
2490 		if (vtnet_set_allmulti(sc, 1) != 0)
2491 			if_printf(ifp, "cannot enable all-multicast mode\n");
2492 }
2493 
2494 static int
2495 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2496 {
2497 	struct virtio_net_ctrl_hdr hdr;
2498 	struct sglist_seg segs[3];
2499 	struct sglist sg;
2500 	uint8_t ack;
2501 	int error;
2502 
2503 	hdr.class = VIRTIO_NET_CTRL_VLAN;
2504 	hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2505 	ack = VIRTIO_NET_ERR;
2506 	error = 0;
2507 
2508 	sglist_init(&sg, 3, segs);
2509 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2510 	error |= sglist_append(&sg, &tag, sizeof(uint16_t));
2511 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2512 	KASSERT(error == 0 && sg.sg_nseg == 3,
2513 	    ("error adding VLAN control message to sglist"));
2514 
2515 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2516 
2517 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2518 }
2519 
2520 static void
2521 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2522 {
2523 	device_t dev;
2524 	uint32_t w, mask;
2525 	uint16_t tag;
2526 	int i, nvlans, error;
2527 
2528 	VTNET_LOCK_ASSERT(sc);
2529 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2530 	    ("VLAN_FILTER feature not negotiated"));
2531 
2532 	dev = sc->vtnet_dev;
2533 	nvlans = sc->vtnet_nvlans;
2534 	error = 0;
2535 
2536 	/* Enable filtering for each configured VLAN. */
2537 	for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2538 		w = sc->vtnet_vlan_shadow[i];
2539 		for (mask = 1, tag = i * 32; w != 0; mask <<= 1, tag++) {
2540 			if ((w & mask) != 0) {
2541 				w &= ~mask;
2542 				nvlans--;
2543 				if (vtnet_exec_vlan_filter(sc, 1, tag) != 0)
2544 					error++;
2545 			}
2546 		}
2547 	}
2548 
2549 	KASSERT(nvlans == 0, ("VLAN count incorrect"));
2550 	if (error)
2551 		device_printf(dev, "cannot restore VLAN filter table\n");
2552 }
2553 
2554 static void
2555 vtnet_set_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2556 {
2557 	struct ifnet *ifp;
2558 	int idx, bit;
2559 
2560 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2561 	    ("VLAN_FILTER feature not negotiated"));
2562 
2563 	if ((tag == 0) || (tag > 4095))
2564 		return;
2565 
2566 	ifp = sc->vtnet_ifp;
2567 	idx = (tag >> 5) & 0x7F;
2568 	bit = tag & 0x1F;
2569 
2570 	VTNET_LOCK(sc);
2571 
2572 	/* Update shadow VLAN table. */
2573 	if (add) {
2574 		sc->vtnet_nvlans++;
2575 		sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2576 	} else {
2577 		sc->vtnet_nvlans--;
2578 		sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2579 	}
2580 
2581 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2582 		if (vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2583 			device_printf(sc->vtnet_dev,
2584 			    "cannot %s VLAN %d %s the host filter table\n",
2585 			    add ? "add" : "remove", tag,
2586 			    add ? "to" : "from");
2587 		}
2588 	}
2589 
2590 	VTNET_UNLOCK(sc);
2591 }
2592 
2593 static void
2594 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2595 {
2596 
2597 	if (ifp->if_softc != arg)
2598 		return;
2599 
2600 	vtnet_set_vlan_filter(arg, 1, tag);
2601 }
2602 
2603 static void
2604 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2605 {
2606 
2607 	if (ifp->if_softc != arg)
2608 		return;
2609 
2610 	vtnet_set_vlan_filter(arg, 0, tag);
2611 }
2612 
2613 static int
2614 vtnet_ifmedia_upd(struct ifnet *ifp)
2615 {
2616 	struct vtnet_softc *sc;
2617 	struct ifmedia *ifm;
2618 
2619 	sc = ifp->if_softc;
2620 	ifm = &sc->vtnet_media;
2621 
2622 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2623 		return (EINVAL);
2624 
2625 	return (0);
2626 }
2627 
2628 static void
2629 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2630 {
2631 	struct vtnet_softc *sc;
2632 
2633 	sc = ifp->if_softc;
2634 
2635 	ifmr->ifm_status = IFM_AVALID;
2636 	ifmr->ifm_active = IFM_ETHER;
2637 
2638 	VTNET_LOCK(sc);
2639 	if (vtnet_is_link_up(sc) != 0) {
2640 		ifmr->ifm_status |= IFM_ACTIVE;
2641 		ifmr->ifm_active |= VTNET_MEDIATYPE;
2642 	} else
2643 		ifmr->ifm_active |= IFM_NONE;
2644 	VTNET_UNLOCK(sc);
2645 }
2646 
2647 static void
2648 vtnet_add_statistics(struct vtnet_softc *sc)
2649 {
2650 	device_t dev;
2651 	struct vtnet_statistics *stats;
2652         struct sysctl_ctx_list *ctx;
2653 	struct sysctl_oid *tree;
2654 	struct sysctl_oid_list *child;
2655 
2656 	dev = sc->vtnet_dev;
2657 	stats = &sc->vtnet_stats;
2658 	ctx = device_get_sysctl_ctx(dev);
2659 	tree = device_get_sysctl_tree(dev);
2660 	child = SYSCTL_CHILDREN(tree);
2661 
2662 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2663 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
2664 	    "Mbuf cluster allocation failures");
2665 
2666 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_frame_too_large",
2667 	    CTLFLAG_RD, &stats->rx_frame_too_large,
2668 	    "Received frame larger than the mbuf chain");
2669 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2670 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
2671 	    "Enqueuing the replacement receive mbuf failed");
2672 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_mergeable_failed",
2673 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
2674 	    "Mergeable buffers receive failures");
2675 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2676 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
2677 	    "Received checksum offloaded buffer with unsupported "
2678 	    "Ethernet type");
2679 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_start",
2680 	    CTLFLAG_RD, &stats->rx_csum_bad_start,
2681 	    "Received checksum offloaded buffer with incorrect start offset");
2682 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2683 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
2684 	    "Received checksum offloaded buffer with incorrect IP protocol");
2685 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2686 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
2687 	    "Received checksum offloaded buffer with incorrect offset");
2688 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_failed",
2689 	    CTLFLAG_RD, &stats->rx_csum_failed,
2690 	    "Received buffer checksum offload failed");
2691 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_offloaded",
2692 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
2693 	    "Received buffer checksum offload succeeded");
2694 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_task_rescheduled",
2695 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
2696 	    "Times the receive interrupt task rescheduled itself");
2697 
2698 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_offloaded",
2699 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
2700 	    "Offloaded checksum of transmitted buffer");
2701 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_offloaded",
2702 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
2703 	    "Segmentation offload of transmitted buffer");
2704 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2705 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
2706 	    "Aborted transmit of checksum offloaded buffer with unknown "
2707 	    "Ethernet type");
2708 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2709 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
2710 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
2711 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_task_rescheduled",
2712 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
2713 	    "Times the transmit interrupt task rescheduled itself");
2714 }
2715 
2716 static int
2717 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2718 {
2719 
2720 	return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2721 }
2722 
2723 static void
2724 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2725 {
2726 
2727 	virtqueue_disable_intr(sc->vtnet_rx_vq);
2728 }
2729 
2730 static int
2731 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2732 {
2733 
2734 #ifdef VTNET_TX_INTR_MODERATION
2735 	return (0);
2736 #else
2737 	return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2738 #endif
2739 }
2740 
2741 static void
2742 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2743 {
2744 
2745 	virtqueue_disable_intr(sc->vtnet_tx_vq);
2746 }
2747