xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 7cd2dcf07629713e5a3d60472cfe4701b705a167)
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 	taskqueue_start_threads(&sc->vtnet_tq, 1, PI_NET, "%s taskq",
443 	    device_get_nameunit(dev));
444 
445 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
446 	if (error) {
447 		device_printf(dev, "cannot setup virtqueue interrupts\n");
448 		taskqueue_free(sc->vtnet_tq);
449 		sc->vtnet_tq = NULL;
450 		ether_ifdetach(ifp);
451 		goto fail;
452 	}
453 
454 	/*
455 	 * Device defaults to promiscuous mode for backwards
456 	 * compatibility. Turn it off if possible.
457 	 */
458 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
459 		VTNET_LOCK(sc);
460 		if (vtnet_set_promisc(sc, 0) != 0) {
461 			ifp->if_flags |= IFF_PROMISC;
462 			device_printf(dev,
463 			    "cannot disable promiscuous mode\n");
464 		}
465 		VTNET_UNLOCK(sc);
466 	} else
467 		ifp->if_flags |= IFF_PROMISC;
468 
469 fail:
470 	if (error)
471 		vtnet_detach(dev);
472 
473 	return (error);
474 }
475 
476 static int
477 vtnet_detach(device_t dev)
478 {
479 	struct vtnet_softc *sc;
480 	struct ifnet *ifp;
481 
482 	sc = device_get_softc(dev);
483 	ifp = sc->vtnet_ifp;
484 
485 	KASSERT(mtx_initialized(VTNET_MTX(sc)),
486 	    ("vtnet mutex not initialized"));
487 
488 #ifdef DEVICE_POLLING
489 	if (ifp != NULL && ifp->if_capenable & IFCAP_POLLING)
490 		ether_poll_deregister(ifp);
491 #endif
492 
493 	if (device_is_attached(dev)) {
494 		VTNET_LOCK(sc);
495 		vtnet_stop(sc);
496 		VTNET_UNLOCK(sc);
497 
498 		callout_drain(&sc->vtnet_tick_ch);
499 		taskqueue_drain(taskqueue_fast, &sc->vtnet_cfgchg_task);
500 
501 		ether_ifdetach(ifp);
502 	}
503 
504 	if (sc->vtnet_tq != NULL) {
505 		taskqueue_drain(sc->vtnet_tq, &sc->vtnet_rx_intr_task);
506 		taskqueue_drain(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
507 		taskqueue_free(sc->vtnet_tq);
508 		sc->vtnet_tq = NULL;
509 	}
510 
511 	if (sc->vtnet_vlan_attach != NULL) {
512 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
513 		sc->vtnet_vlan_attach = NULL;
514 	}
515 	if (sc->vtnet_vlan_detach != NULL) {
516 		EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
517 		sc->vtnet_vlan_detach = NULL;
518 	}
519 
520 	if (sc->vtnet_mac_filter != NULL) {
521 		free(sc->vtnet_mac_filter, M_DEVBUF);
522 		sc->vtnet_mac_filter = NULL;
523 	}
524 
525 	if (ifp != NULL) {
526 		if_free(ifp);
527 		sc->vtnet_ifp = NULL;
528 	}
529 
530 	if (sc->vtnet_rx_vq != NULL)
531 		vtnet_free_rx_mbufs(sc);
532 	if (sc->vtnet_tx_vq != NULL)
533 		vtnet_free_tx_mbufs(sc);
534 	if (sc->vtnet_ctrl_vq != NULL)
535 		vtnet_free_ctrl_vq(sc);
536 
537 	ifmedia_removeall(&sc->vtnet_media);
538 	VTNET_LOCK_DESTROY(sc);
539 
540 	return (0);
541 }
542 
543 static int
544 vtnet_suspend(device_t dev)
545 {
546 	struct vtnet_softc *sc;
547 
548 	sc = device_get_softc(dev);
549 
550 	VTNET_LOCK(sc);
551 	vtnet_stop(sc);
552 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
553 	VTNET_UNLOCK(sc);
554 
555 	return (0);
556 }
557 
558 static int
559 vtnet_resume(device_t dev)
560 {
561 	struct vtnet_softc *sc;
562 	struct ifnet *ifp;
563 
564 	sc = device_get_softc(dev);
565 	ifp = sc->vtnet_ifp;
566 
567 	VTNET_LOCK(sc);
568 	if (ifp->if_flags & IFF_UP)
569 		vtnet_init_locked(sc);
570 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
571 	VTNET_UNLOCK(sc);
572 
573 	return (0);
574 }
575 
576 static int
577 vtnet_shutdown(device_t dev)
578 {
579 
580 	/*
581 	 * Suspend already does all of what we need to
582 	 * do here; we just never expect to be resumed.
583 	 */
584 	return (vtnet_suspend(dev));
585 }
586 
587 static int
588 vtnet_config_change(device_t dev)
589 {
590 	struct vtnet_softc *sc;
591 
592 	sc = device_get_softc(dev);
593 
594 	taskqueue_enqueue_fast(taskqueue_fast, &sc->vtnet_cfgchg_task);
595 
596 	return (1);
597 }
598 
599 static void
600 vtnet_negotiate_features(struct vtnet_softc *sc)
601 {
602 	device_t dev;
603 	uint64_t mask, features;
604 
605 	dev = sc->vtnet_dev;
606 	mask = 0;
607 
608 	if (vtnet_csum_disable)
609 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
610 
611 	/*
612 	 * TSO and LRO are only available when their corresponding
613 	 * checksum offload feature is also negotiated.
614 	 */
615 
616 	if (vtnet_csum_disable || vtnet_tso_disable)
617 		mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
618 		    VIRTIO_NET_F_HOST_ECN;
619 
620 	if (vtnet_csum_disable || vtnet_lro_disable)
621 		mask |= VTNET_LRO_FEATURES;
622 
623 	features = VTNET_FEATURES & ~mask;
624 #ifdef VTNET_TX_INTR_MODERATION
625 	features |= VIRTIO_F_NOTIFY_ON_EMPTY;
626 #endif
627 	sc->vtnet_features = virtio_negotiate_features(dev, features);
628 
629 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0 &&
630 	    virtio_with_feature(dev, VTNET_LRO_FEATURES)) {
631 		/*
632 		 * LRO without mergeable buffers requires special care. This
633 		 * is not ideal because every receive buffer must be large
634 		 * enough to hold the maximum TCP packet, the Ethernet header,
635 		 * and the vtnet_rx_header. This requires up to 34 descriptors
636 		 * when using MCLBYTES clusters. If we do not have indirect
637 		 * descriptors, LRO is disabled since the virtqueue will not
638 		 * be able to contain very many receive buffers.
639 		 */
640 		if (virtio_with_feature(dev,
641 		    VIRTIO_RING_F_INDIRECT_DESC) == 0) {
642 			device_printf(dev,
643 			    "LRO disabled due to lack of both mergeable "
644 			    "buffers and indirect descriptors\n");
645 
646 			sc->vtnet_features = virtio_negotiate_features(dev,
647 			    features & ~VTNET_LRO_FEATURES);
648 		} else
649 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
650 	}
651 }
652 
653 static int
654 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
655 {
656 	device_t dev;
657 	struct vq_alloc_info vq_info[3];
658 	int nvqs, rxsegs;
659 
660 	dev = sc->vtnet_dev;
661 	nvqs = 2;
662 
663 	/*
664 	 * Indirect descriptors are not needed for the Rx
665 	 * virtqueue when mergeable buffers are negotiated.
666 	 * The header is placed inline with the data, not
667 	 * in a separate descriptor, and mbuf clusters are
668 	 * always physically contiguous.
669 	 */
670 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
671 		rxsegs = sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ?
672 		    VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
673 	} else
674 		rxsegs = 0;
675 
676 	VQ_ALLOC_INFO_INIT(&vq_info[0], rxsegs,
677 	    vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
678 	    "%s receive", device_get_nameunit(dev));
679 
680 	VQ_ALLOC_INFO_INIT(&vq_info[1], VTNET_MAX_TX_SEGS,
681 	    vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
682 	    "%s transmit", device_get_nameunit(dev));
683 
684 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
685 		nvqs++;
686 
687 		VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
688 		    &sc->vtnet_ctrl_vq, "%s control",
689 		    device_get_nameunit(dev));
690 	}
691 
692 	return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
693 }
694 
695 static void
696 vtnet_get_hwaddr(struct vtnet_softc *sc)
697 {
698 	device_t dev;
699 
700 	dev = sc->vtnet_dev;
701 
702 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
703 		virtio_read_device_config(dev,
704 		    offsetof(struct virtio_net_config, mac),
705 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
706 	} else {
707 		/* Generate random locally administered unicast address. */
708 		sc->vtnet_hwaddr[0] = 0xB2;
709 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
710 
711 		vtnet_set_hwaddr(sc);
712 	}
713 }
714 
715 static void
716 vtnet_set_hwaddr(struct vtnet_softc *sc)
717 {
718 	device_t dev;
719 
720 	dev = sc->vtnet_dev;
721 
722 	virtio_write_device_config(dev,
723 	    offsetof(struct virtio_net_config, mac),
724 	    sc->vtnet_hwaddr, ETHER_ADDR_LEN);
725 }
726 
727 static int
728 vtnet_is_link_up(struct vtnet_softc *sc)
729 {
730 	device_t dev;
731 	struct ifnet *ifp;
732 	uint16_t status;
733 
734 	dev = sc->vtnet_dev;
735 	ifp = sc->vtnet_ifp;
736 
737 	VTNET_LOCK_ASSERT(sc);
738 
739 	if ((ifp->if_capenable & IFCAP_LINKSTATE) == 0)
740 		return (1);
741 
742 	status = virtio_read_dev_config_2(dev,
743 	    offsetof(struct virtio_net_config, status));
744 
745 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
746 }
747 
748 static void
749 vtnet_update_link_status(struct vtnet_softc *sc)
750 {
751 	struct ifnet *ifp;
752 	int link;
753 
754 	ifp = sc->vtnet_ifp;
755 
756 	link = vtnet_is_link_up(sc);
757 
758 	if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
759 		sc->vtnet_flags |= VTNET_FLAG_LINK;
760 		if_link_state_change(ifp, LINK_STATE_UP);
761 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
762 			vtnet_start_locked(ifp);
763 	} else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
764 		sc->vtnet_flags &= ~VTNET_FLAG_LINK;
765 		if_link_state_change(ifp, LINK_STATE_DOWN);
766 	}
767 }
768 
769 static void
770 vtnet_watchdog(struct vtnet_softc *sc)
771 {
772 	struct ifnet *ifp;
773 
774 	ifp = sc->vtnet_ifp;
775 
776 #ifdef VTNET_TX_INTR_MODERATION
777 	vtnet_txeof(sc);
778 #endif
779 
780 	if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
781 		return;
782 
783 	if_printf(ifp, "watchdog timeout -- resetting\n");
784 #ifdef VTNET_DEBUG
785 	virtqueue_dump(sc->vtnet_tx_vq);
786 #endif
787 	ifp->if_oerrors++;
788 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
789 	vtnet_init_locked(sc);
790 }
791 
792 static void
793 vtnet_config_change_task(void *arg, int pending)
794 {
795 	struct vtnet_softc *sc;
796 
797 	sc = arg;
798 
799 	VTNET_LOCK(sc);
800 	vtnet_update_link_status(sc);
801 	VTNET_UNLOCK(sc);
802 }
803 
804 static int
805 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
806 {
807 	struct vtnet_softc *sc;
808 	struct ifreq *ifr;
809 	int reinit, mask, error;
810 
811 	sc = ifp->if_softc;
812 	ifr = (struct ifreq *) data;
813 	reinit = 0;
814 	error = 0;
815 
816 	switch (cmd) {
817 	case SIOCSIFMTU:
818 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
819 			error = EINVAL;
820 		else if (ifp->if_mtu != ifr->ifr_mtu) {
821 			VTNET_LOCK(sc);
822 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
823 			VTNET_UNLOCK(sc);
824 		}
825 		break;
826 
827 	case SIOCSIFFLAGS:
828 		VTNET_LOCK(sc);
829 		if ((ifp->if_flags & IFF_UP) == 0) {
830 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
831 				vtnet_stop(sc);
832 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
833 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
834 			    (IFF_PROMISC | IFF_ALLMULTI)) {
835 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
836 					vtnet_rx_filter(sc);
837 				else
838 					error = ENOTSUP;
839 			}
840 		} else
841 			vtnet_init_locked(sc);
842 
843 		if (error == 0)
844 			sc->vtnet_if_flags = ifp->if_flags;
845 		VTNET_UNLOCK(sc);
846 		break;
847 
848 	case SIOCADDMULTI:
849 	case SIOCDELMULTI:
850 		VTNET_LOCK(sc);
851 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
852 		    (ifp->if_drv_flags & IFF_DRV_RUNNING))
853 			vtnet_rx_filter_mac(sc);
854 		VTNET_UNLOCK(sc);
855 		break;
856 
857 	case SIOCSIFMEDIA:
858 	case SIOCGIFMEDIA:
859 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
860 		break;
861 
862 	case SIOCSIFCAP:
863 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
864 
865 #ifdef DEVICE_POLLING
866 		if (mask & IFCAP_POLLING) {
867 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
868 				error = ether_poll_register(vtnet_poll, ifp);
869 				if (error)
870 					break;
871 
872 				VTNET_LOCK(sc);
873 				vtnet_disable_rx_intr(sc);
874 				vtnet_disable_tx_intr(sc);
875 				ifp->if_capenable |= IFCAP_POLLING;
876 				VTNET_UNLOCK(sc);
877 			} else {
878 				error = ether_poll_deregister(ifp);
879 
880 				/* Enable interrupts even in error case. */
881 				VTNET_LOCK(sc);
882 				vtnet_enable_tx_intr(sc);
883 				vtnet_enable_rx_intr(sc);
884 				ifp->if_capenable &= ~IFCAP_POLLING;
885 				VTNET_UNLOCK(sc);
886 			}
887 		}
888 #endif
889 		VTNET_LOCK(sc);
890 
891 		if (mask & IFCAP_TXCSUM) {
892 			ifp->if_capenable ^= IFCAP_TXCSUM;
893 			if (ifp->if_capenable & IFCAP_TXCSUM)
894 				ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
895 			else
896 				ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
897 		}
898 
899 		if (mask & IFCAP_TSO4) {
900 			ifp->if_capenable ^= IFCAP_TSO4;
901 			if (ifp->if_capenable & IFCAP_TSO4)
902 				ifp->if_hwassist |= CSUM_TSO;
903 			else
904 				ifp->if_hwassist &= ~CSUM_TSO;
905 		}
906 
907 		if (mask & IFCAP_RXCSUM) {
908 			ifp->if_capenable ^= IFCAP_RXCSUM;
909 			reinit = 1;
910 		}
911 
912 		if (mask & IFCAP_LRO) {
913 			ifp->if_capenable ^= IFCAP_LRO;
914 			reinit = 1;
915 		}
916 
917 		if (mask & IFCAP_VLAN_HWFILTER) {
918 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
919 			reinit = 1;
920 		}
921 
922 		if (mask & IFCAP_VLAN_HWTSO)
923 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
924 
925 		if (mask & IFCAP_VLAN_HWTAGGING)
926 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
927 
928 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
929 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
930 			vtnet_init_locked(sc);
931 		}
932 		VLAN_CAPABILITIES(ifp);
933 
934 		VTNET_UNLOCK(sc);
935 		break;
936 
937 	default:
938 		error = ether_ioctl(ifp, cmd, data);
939 		break;
940 	}
941 
942 	VTNET_LOCK_ASSERT_NOTOWNED(sc);
943 
944 	return (error);
945 }
946 
947 static int
948 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
949 {
950 	struct ifnet *ifp;
951 	int new_frame_size, clsize;
952 
953 	ifp = sc->vtnet_ifp;
954 
955 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
956 		new_frame_size = sizeof(struct vtnet_rx_header) +
957 		    sizeof(struct ether_vlan_header) + new_mtu;
958 
959 		if (new_frame_size > MJUM9BYTES)
960 			return (EINVAL);
961 
962 		if (new_frame_size <= MCLBYTES)
963 			clsize = MCLBYTES;
964 		else
965 			clsize = MJUM9BYTES;
966 	} else {
967 		new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
968 		    sizeof(struct ether_vlan_header) + new_mtu;
969 
970 		if (new_frame_size <= MCLBYTES)
971 			clsize = MCLBYTES;
972 		else
973 			clsize = MJUMPAGESIZE;
974 	}
975 
976 	sc->vtnet_rx_mbuf_size = clsize;
977 	sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
978 	KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
979 	    ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
980 
981 	ifp->if_mtu = new_mtu;
982 
983 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
984 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
985 		vtnet_init_locked(sc);
986 	}
987 
988 	return (0);
989 }
990 
991 static int
992 vtnet_init_rx_vq(struct vtnet_softc *sc)
993 {
994 	struct virtqueue *vq;
995 	int nbufs, error;
996 
997 	vq = sc->vtnet_rx_vq;
998 	nbufs = 0;
999 	error = ENOSPC;
1000 
1001 	while (!virtqueue_full(vq)) {
1002 		if ((error = vtnet_newbuf(sc)) != 0)
1003 			break;
1004 		nbufs++;
1005 	}
1006 
1007 	if (nbufs > 0) {
1008 		virtqueue_notify(vq);
1009 
1010 		/*
1011 		 * EMSGSIZE signifies the virtqueue did not have enough
1012 		 * entries available to hold the last mbuf. This is not
1013 		 * an error. We should not get ENOSPC since we check if
1014 		 * the virtqueue is full before attempting to add a
1015 		 * buffer.
1016 		 */
1017 		if (error == EMSGSIZE)
1018 			error = 0;
1019 	}
1020 
1021 	return (error);
1022 }
1023 
1024 static void
1025 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
1026 {
1027 	struct virtqueue *vq;
1028 	struct mbuf *m;
1029 	int last;
1030 
1031 	vq = sc->vtnet_rx_vq;
1032 	last = 0;
1033 
1034 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1035 		m_freem(m);
1036 
1037 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
1038 }
1039 
1040 static void
1041 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1042 {
1043 	struct virtqueue *vq;
1044 	struct vtnet_tx_header *txhdr;
1045 	int last;
1046 
1047 	vq = sc->vtnet_tx_vq;
1048 	last = 0;
1049 
1050 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1051 		m_freem(txhdr->vth_mbuf);
1052 		uma_zfree(vtnet_tx_header_zone, txhdr);
1053 	}
1054 
1055 	KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1056 }
1057 
1058 static void
1059 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1060 {
1061 
1062 	/*
1063 	 * The control virtqueue is only polled, therefore
1064 	 * it should already be empty.
1065 	 */
1066 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1067 	    ("Ctrl Vq not empty"));
1068 }
1069 
1070 #ifdef DEVICE_POLLING
1071 static int
1072 vtnet_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1073 {
1074 	struct vtnet_softc *sc;
1075 	int rx_done;
1076 
1077 	sc = ifp->if_softc;
1078 	rx_done = 0;
1079 
1080 	VTNET_LOCK(sc);
1081 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1082 		if (cmd == POLL_AND_CHECK_STATUS)
1083 			vtnet_update_link_status(sc);
1084 
1085 		if (virtqueue_nused(sc->vtnet_rx_vq) > 0)
1086 			vtnet_rxeof(sc, count, &rx_done);
1087 
1088 		vtnet_txeof(sc);
1089 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1090 			vtnet_start_locked(ifp);
1091 	}
1092 	VTNET_UNLOCK(sc);
1093 
1094 	return (rx_done);
1095 }
1096 #endif /* DEVICE_POLLING */
1097 
1098 static struct mbuf *
1099 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1100 {
1101 	struct mbuf *m_head, *m_tail, *m;
1102 	int i, clsize;
1103 
1104 	clsize = sc->vtnet_rx_mbuf_size;
1105 
1106 	m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, clsize);
1107 	if (m_head == NULL)
1108 		goto fail;
1109 
1110 	m_head->m_len = clsize;
1111 	m_tail = m_head;
1112 
1113 	if (nbufs > 1) {
1114 		KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1115 		    ("chained Rx mbuf requested without LRO_NOMRG"));
1116 
1117 		for (i = 1; i < nbufs; i++) {
1118 			m = m_getjcl(M_DONTWAIT, MT_DATA, 0, clsize);
1119 			if (m == NULL)
1120 				goto fail;
1121 
1122 			m->m_len = clsize;
1123 			m_tail->m_next = m;
1124 			m_tail = m;
1125 		}
1126 	}
1127 
1128 	if (m_tailp != NULL)
1129 		*m_tailp = m_tail;
1130 
1131 	return (m_head);
1132 
1133 fail:
1134 	sc->vtnet_stats.mbuf_alloc_failed++;
1135 	m_freem(m_head);
1136 
1137 	return (NULL);
1138 }
1139 
1140 static int
1141 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1142 {
1143 	struct mbuf *m, *m_prev;
1144 	struct mbuf *m_new, *m_tail;
1145 	int len, clsize, nreplace, error;
1146 
1147 	m = m0;
1148 	m_prev = NULL;
1149 	len = len0;
1150 
1151 	m_tail = NULL;
1152 	clsize = sc->vtnet_rx_mbuf_size;
1153 	nreplace = 0;
1154 
1155 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1156 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1157 
1158 	/*
1159 	 * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1160 	 * allocating an entire chain for each received frame. When
1161 	 * the received frame's length is less than that of the chain,
1162 	 * the unused mbufs are reassigned to the new chain.
1163 	 */
1164 	while (len > 0) {
1165 		/*
1166 		 * Something is seriously wrong if we received
1167 		 * a frame larger than the mbuf chain. Drop it.
1168 		 */
1169 		if (m == NULL) {
1170 			sc->vtnet_stats.rx_frame_too_large++;
1171 			return (EMSGSIZE);
1172 		}
1173 
1174 		KASSERT(m->m_len == clsize,
1175 		    ("mbuf length not expected cluster size: %d",
1176 		    m->m_len));
1177 
1178 		m->m_len = MIN(m->m_len, len);
1179 		len -= m->m_len;
1180 
1181 		m_prev = m;
1182 		m = m->m_next;
1183 		nreplace++;
1184 	}
1185 
1186 	KASSERT(m_prev != NULL, ("m_prev == NULL"));
1187 	KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1188 	    ("too many replacement mbufs: %d/%d", nreplace,
1189 	    sc->vtnet_rx_mbuf_count));
1190 
1191 	m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1192 	if (m_new == NULL) {
1193 		m_prev->m_len = clsize;
1194 		return (ENOBUFS);
1195 	}
1196 
1197 	/*
1198 	 * Move unused mbufs, if any, from the original chain
1199 	 * onto the end of the new chain.
1200 	 */
1201 	if (m_prev->m_next != NULL) {
1202 		m_tail->m_next = m_prev->m_next;
1203 		m_prev->m_next = NULL;
1204 	}
1205 
1206 	error = vtnet_enqueue_rxbuf(sc, m_new);
1207 	if (error) {
1208 		/*
1209 		 * BAD! We could not enqueue the replacement mbuf chain. We
1210 		 * must restore the m0 chain to the original state if it was
1211 		 * modified so we can subsequently discard it.
1212 		 *
1213 		 * NOTE: The replacement is suppose to be an identical copy
1214 		 * to the one just dequeued so this is an unexpected error.
1215 		 */
1216 		sc->vtnet_stats.rx_enq_replacement_failed++;
1217 
1218 		if (m_tail->m_next != NULL) {
1219 			m_prev->m_next = m_tail->m_next;
1220 			m_tail->m_next = NULL;
1221 		}
1222 
1223 		m_prev->m_len = clsize;
1224 		m_freem(m_new);
1225 	}
1226 
1227 	return (error);
1228 }
1229 
1230 static int
1231 vtnet_newbuf(struct vtnet_softc *sc)
1232 {
1233 	struct mbuf *m;
1234 	int error;
1235 
1236 	m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1237 	if (m == NULL)
1238 		return (ENOBUFS);
1239 
1240 	error = vtnet_enqueue_rxbuf(sc, m);
1241 	if (error)
1242 		m_freem(m);
1243 
1244 	return (error);
1245 }
1246 
1247 static void
1248 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1249 {
1250 	struct virtqueue *vq;
1251 	struct mbuf *m;
1252 
1253 	vq = sc->vtnet_rx_vq;
1254 
1255 	while (--nbufs > 0) {
1256 		if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1257 			break;
1258 		vtnet_discard_rxbuf(sc, m);
1259 	}
1260 }
1261 
1262 static void
1263 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1264 {
1265 	int error;
1266 
1267 	/*
1268 	 * Requeue the discarded mbuf. This should always be
1269 	 * successful since it was just dequeued.
1270 	 */
1271 	error = vtnet_enqueue_rxbuf(sc, m);
1272 	KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1273 }
1274 
1275 static int
1276 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1277 {
1278 	struct sglist sg;
1279 	struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1280 	struct vtnet_rx_header *rxhdr;
1281 	struct virtio_net_hdr *hdr;
1282 	uint8_t *mdata;
1283 	int offset, error;
1284 
1285 	VTNET_LOCK_ASSERT(sc);
1286 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ||
1287 	    m->m_next == NULL, ("chained Rx mbuf without LRO_NOMRG"));
1288 
1289 	sglist_init(&sg, VTNET_MAX_RX_SEGS, segs);
1290 
1291 	mdata = mtod(m, uint8_t *);
1292 	offset = 0;
1293 
1294 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1295 		rxhdr = (struct vtnet_rx_header *) mdata;
1296 		hdr = &rxhdr->vrh_hdr;
1297 		offset += sizeof(struct vtnet_rx_header);
1298 
1299 		error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1300 		KASSERT(error == 0, ("cannot add header to sglist"));
1301 	}
1302 
1303 	error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1304 	if (error)
1305 		return (error);
1306 
1307 	if (m->m_next != NULL) {
1308 		error = sglist_append_mbuf(&sg, m->m_next);
1309 		if (error)
1310 			return (error);
1311 	}
1312 
1313 	return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1314 }
1315 
1316 static void
1317 vtnet_vlan_tag_remove(struct mbuf *m)
1318 {
1319 	struct ether_vlan_header *evl;
1320 
1321 	evl = mtod(m, struct ether_vlan_header *);
1322 
1323 	m->m_pkthdr.ether_vtag = ntohs(evl->evl_tag);
1324 	m->m_flags |= M_VLANTAG;
1325 
1326 	/* Strip the 802.1Q header. */
1327 	bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1328 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
1329 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
1330 }
1331 
1332 #ifdef notyet
1333 static int
1334 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1335     struct virtio_net_hdr *hdr)
1336 {
1337 	struct ether_header *eh;
1338 	struct ether_vlan_header *evh;
1339 	struct ip *ip;
1340 	struct ip6_hdr *ip6;
1341 	struct udphdr *udp;
1342 	int ip_offset, csum_start, csum_offset, hlen;
1343 	uint16_t eth_type;
1344 	uint8_t ip_proto;
1345 
1346 	/*
1347 	 * Convert the VirtIO checksum interface to FreeBSD's interface.
1348 	 * The host only provides us with the offset at which to start
1349 	 * checksumming, and the offset from that to place the completed
1350 	 * checksum. While this maps well with how Linux does checksums,
1351 	 * for FreeBSD, we must parse the received packet in order to set
1352 	 * the appropriate CSUM_* flags.
1353 	 */
1354 
1355 	/*
1356 	 * Every mbuf added to the receive virtqueue is always at least
1357 	 * MCLBYTES big, so assume something is amiss if the first mbuf
1358 	 * does not contain both the Ethernet and protocol headers.
1359 	 */
1360 	ip_offset = sizeof(struct ether_header);
1361 	if (m->m_len < ip_offset)
1362 		return (1);
1363 
1364 	eh = mtod(m, struct ether_header *);
1365 	eth_type = ntohs(eh->ether_type);
1366 	if (eth_type == ETHERTYPE_VLAN) {
1367 		ip_offset = sizeof(struct ether_vlan_header);
1368 		if (m->m_len < ip_offset)
1369 			return (1);
1370 		evh = mtod(m, struct ether_vlan_header *);
1371 		eth_type = ntohs(evh->evl_proto);
1372 	}
1373 
1374 	switch (eth_type) {
1375 	case ETHERTYPE_IP:
1376 		if (m->m_len < ip_offset + sizeof(struct ip))
1377 			return (1);
1378 
1379 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1380 		 /* Sanity check the IP header. */
1381 		if (ip->ip_v != IPVERSION)
1382 			return (1);
1383 		hlen = ip->ip_hl << 2;
1384 		if (hlen < sizeof(struct ip))
1385 			return (1);
1386 		if (ntohs(ip->ip_len) < hlen)
1387 			return (1);
1388 		if (ntohs(ip->ip_len) != (m->m_pkthdr.len - ip_offset))
1389 			return (1);
1390 
1391 		ip_proto = ip->ip_p;
1392 		csum_start = ip_offset + hlen;
1393 		break;
1394 
1395 	case ETHERTYPE_IPV6:
1396 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr))
1397 			return (1);
1398 
1399 		/*
1400 		 * XXX FreeBSD does not handle any IPv6 checksum offloading
1401 		 * at the moment.
1402 		 */
1403 
1404 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1405 		/* XXX Assume no extension headers are present. */
1406 		ip_proto = ip6->ip6_nxt;
1407 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1408 		break;
1409 
1410 	default:
1411 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1412 		return (1);
1413 	}
1414 
1415 	/* Assume checksum begins right after the IP header. */
1416 	if (hdr->csum_start != csum_start) {
1417 		sc->vtnet_stats.rx_csum_bad_start++;
1418 		return (1);
1419 	}
1420 
1421 	switch (ip_proto) {
1422 	case IPPROTO_TCP:
1423 		csum_offset = offsetof(struct tcphdr, th_sum);
1424 		break;
1425 
1426 	case IPPROTO_UDP:
1427 		csum_offset = offsetof(struct udphdr, uh_sum);
1428 		break;
1429 
1430 	case IPPROTO_SCTP:
1431 		csum_offset = offsetof(struct sctphdr, checksum);
1432 		break;
1433 
1434 	default:
1435 		sc->vtnet_stats.rx_csum_bad_ipproto++;
1436 		return (1);
1437 	}
1438 
1439 	if (hdr->csum_offset != csum_offset) {
1440 		sc->vtnet_stats.rx_csum_bad_offset++;
1441 		return (1);
1442 	}
1443 
1444 	/*
1445 	 * The IP header checksum is almost certainly valid but I'm
1446 	 * uncertain if that is guaranteed.
1447 	 *
1448 	 * m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
1449 	 */
1450 
1451 	switch (ip_proto) {
1452 	case IPPROTO_UDP:
1453 		if (m->m_len < csum_start + sizeof(struct udphdr))
1454 			return (1);
1455 
1456 		udp = (struct udphdr *)(mtod(m, uint8_t *) + csum_start);
1457 		if (udp->uh_sum == 0)
1458 			return (0);
1459 
1460 		/* FALLTHROUGH */
1461 
1462 	case IPPROTO_TCP:
1463 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1464 		m->m_pkthdr.csum_data = 0xFFFF;
1465 		break;
1466 
1467 	case IPPROTO_SCTP:
1468 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1469 		break;
1470 	}
1471 
1472 	sc->vtnet_stats.rx_csum_offloaded++;
1473 
1474 	return (0);
1475 }
1476 #endif
1477 
1478 /*
1479  * Alternative method of doing receive checksum offloading. Rather
1480  * than parsing the received frame down to the IP header, use the
1481  * csum_offset to determine which CSUM_* flags are appropriate. We
1482  * can get by with doing this only because the checksum offsets are
1483  * unique for the things we care about.
1484  */
1485 static int
1486 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1487     struct virtio_net_hdr *hdr)
1488 {
1489 	struct ether_header *eh;
1490 	struct ether_vlan_header *evh;
1491 	struct udphdr *udp;
1492 	int csum_len;
1493 	uint16_t eth_type;
1494 
1495 	csum_len = hdr->csum_start + hdr->csum_offset;
1496 
1497 	if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1498 		return (1);
1499 	if (m->m_len < csum_len)
1500 		return (1);
1501 
1502 	eh = mtod(m, struct ether_header *);
1503 	eth_type = ntohs(eh->ether_type);
1504 	if (eth_type == ETHERTYPE_VLAN) {
1505 		evh = mtod(m, struct ether_vlan_header *);
1506 		eth_type = ntohs(evh->evl_proto);
1507 	}
1508 
1509 	if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1510 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1511 		return (1);
1512 	}
1513 
1514 	/* Use the offset to determine the appropriate CSUM_* flags. */
1515 	switch (hdr->csum_offset) {
1516 	case offsetof(struct udphdr, uh_sum):
1517 		if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1518 			return (1);
1519 		udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1520 		if (udp->uh_sum == 0)
1521 			return (0);
1522 
1523 		/* FALLTHROUGH */
1524 
1525 	case offsetof(struct tcphdr, th_sum):
1526 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1527 		m->m_pkthdr.csum_data = 0xFFFF;
1528 		break;
1529 
1530 	case offsetof(struct sctphdr, checksum):
1531 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1532 		break;
1533 
1534 	default:
1535 		sc->vtnet_stats.rx_csum_bad_offset++;
1536 		return (1);
1537 	}
1538 
1539 	sc->vtnet_stats.rx_csum_offloaded++;
1540 
1541 	return (0);
1542 }
1543 
1544 static int
1545 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1546 {
1547 	struct ifnet *ifp;
1548 	struct virtqueue *vq;
1549 	struct mbuf *m, *m_tail;
1550 	int len;
1551 
1552 	ifp = sc->vtnet_ifp;
1553 	vq = sc->vtnet_rx_vq;
1554 	m_tail = m_head;
1555 
1556 	while (--nbufs > 0) {
1557 		m = virtqueue_dequeue(vq, &len);
1558 		if (m == NULL) {
1559 			ifp->if_ierrors++;
1560 			goto fail;
1561 		}
1562 
1563 		if (vtnet_newbuf(sc) != 0) {
1564 			ifp->if_iqdrops++;
1565 			vtnet_discard_rxbuf(sc, m);
1566 			if (nbufs > 1)
1567 				vtnet_discard_merged_rxbuf(sc, nbufs);
1568 			goto fail;
1569 		}
1570 
1571 		if (m->m_len < len)
1572 			len = m->m_len;
1573 
1574 		m->m_len = len;
1575 		m->m_flags &= ~M_PKTHDR;
1576 
1577 		m_head->m_pkthdr.len += len;
1578 		m_tail->m_next = m;
1579 		m_tail = m;
1580 	}
1581 
1582 	return (0);
1583 
1584 fail:
1585 	sc->vtnet_stats.rx_mergeable_failed++;
1586 	m_freem(m_head);
1587 
1588 	return (1);
1589 }
1590 
1591 static int
1592 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1593 {
1594 	struct virtio_net_hdr lhdr;
1595 	struct ifnet *ifp;
1596 	struct virtqueue *vq;
1597 	struct mbuf *m;
1598 	struct ether_header *eh;
1599 	struct virtio_net_hdr *hdr;
1600 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1601 	int len, deq, nbufs, adjsz, rx_npkts;
1602 
1603 	ifp = sc->vtnet_ifp;
1604 	vq = sc->vtnet_rx_vq;
1605 	hdr = &lhdr;
1606 	deq = 0;
1607 	rx_npkts = 0;
1608 
1609 	VTNET_LOCK_ASSERT(sc);
1610 
1611 	while (--count >= 0) {
1612 		m = virtqueue_dequeue(vq, &len);
1613 		if (m == NULL)
1614 			break;
1615 		deq++;
1616 
1617 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1618 			ifp->if_ierrors++;
1619 			vtnet_discard_rxbuf(sc, m);
1620 			continue;
1621 		}
1622 
1623 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1624 			nbufs = 1;
1625 			adjsz = sizeof(struct vtnet_rx_header);
1626 			/*
1627 			 * Account for our pad between the header and
1628 			 * the actual start of the frame.
1629 			 */
1630 			len += VTNET_RX_HEADER_PAD;
1631 		} else {
1632 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1633 			nbufs = mhdr->num_buffers;
1634 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1635 		}
1636 
1637 		if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1638 			ifp->if_iqdrops++;
1639 			vtnet_discard_rxbuf(sc, m);
1640 			if (nbufs > 1)
1641 				vtnet_discard_merged_rxbuf(sc, nbufs);
1642 			continue;
1643 		}
1644 
1645 		m->m_pkthdr.len = len;
1646 		m->m_pkthdr.rcvif = ifp;
1647 		m->m_pkthdr.csum_flags = 0;
1648 
1649 		if (nbufs > 1) {
1650 			if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1651 				continue;
1652 		}
1653 
1654 		ifp->if_ipackets++;
1655 
1656 		/*
1657 		 * Save copy of header before we strip it. For both mergeable
1658 		 * and non-mergeable, the VirtIO header is placed first in the
1659 		 * mbuf's data. We no longer need num_buffers, so always use a
1660 		 * virtio_net_hdr.
1661 		 */
1662 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1663 		m_adj(m, adjsz);
1664 
1665 		if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1666 			eh = mtod(m, struct ether_header *);
1667 			if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1668 				vtnet_vlan_tag_remove(m);
1669 
1670 				/*
1671 				 * With the 802.1Q header removed, update the
1672 				 * checksum starting location accordingly.
1673 				 */
1674 				if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1675 					hdr->csum_start -=
1676 					    ETHER_VLAN_ENCAP_LEN;
1677 			}
1678 		}
1679 
1680 		if (ifp->if_capenable & IFCAP_RXCSUM &&
1681 		    hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1682 			if (vtnet_rx_csum(sc, m, hdr) != 0)
1683 				sc->vtnet_stats.rx_csum_failed++;
1684 		}
1685 
1686 		VTNET_UNLOCK(sc);
1687 		rx_npkts++;
1688 		(*ifp->if_input)(ifp, m);
1689 		VTNET_LOCK(sc);
1690 
1691 		/*
1692 		 * The interface may have been stopped while we were
1693 		 * passing the packet up the network stack.
1694 		 */
1695 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1696 			break;
1697 	}
1698 
1699 	if (deq > 0)
1700 		virtqueue_notify(vq);
1701 
1702 	if (rx_npktsp != NULL)
1703 		*rx_npktsp = rx_npkts;
1704 
1705 	return (count > 0 ? 0 : EAGAIN);
1706 }
1707 
1708 static void
1709 vtnet_rx_intr_task(void *arg, int pending)
1710 {
1711 	struct vtnet_softc *sc;
1712 	struct ifnet *ifp;
1713 	int more;
1714 
1715 	sc = arg;
1716 	ifp = sc->vtnet_ifp;
1717 
1718 	VTNET_LOCK(sc);
1719 
1720 #ifdef DEVICE_POLLING
1721 	if (ifp->if_capenable & IFCAP_POLLING) {
1722 		VTNET_UNLOCK(sc);
1723 		return;
1724 	}
1725 #endif
1726 
1727 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1728 		vtnet_enable_rx_intr(sc);
1729 		VTNET_UNLOCK(sc);
1730 		return;
1731 	}
1732 
1733 	more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1734 	if (!more && vtnet_enable_rx_intr(sc) != 0) {
1735 		vtnet_disable_rx_intr(sc);
1736 		more = 1;
1737 	}
1738 
1739 	VTNET_UNLOCK(sc);
1740 
1741 	if (more) {
1742 		sc->vtnet_stats.rx_task_rescheduled++;
1743 		taskqueue_enqueue_fast(sc->vtnet_tq,
1744 		    &sc->vtnet_rx_intr_task);
1745 	}
1746 }
1747 
1748 static int
1749 vtnet_rx_vq_intr(void *xsc)
1750 {
1751 	struct vtnet_softc *sc;
1752 
1753 	sc = xsc;
1754 
1755 	vtnet_disable_rx_intr(sc);
1756 	taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_rx_intr_task);
1757 
1758 	return (1);
1759 }
1760 
1761 static void
1762 vtnet_txeof(struct vtnet_softc *sc)
1763 {
1764 	struct virtqueue *vq;
1765 	struct ifnet *ifp;
1766 	struct vtnet_tx_header *txhdr;
1767 	int deq;
1768 
1769 	vq = sc->vtnet_tx_vq;
1770 	ifp = sc->vtnet_ifp;
1771 	deq = 0;
1772 
1773 	VTNET_LOCK_ASSERT(sc);
1774 
1775 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1776 		deq++;
1777 		ifp->if_opackets++;
1778 		m_freem(txhdr->vth_mbuf);
1779 		uma_zfree(vtnet_tx_header_zone, txhdr);
1780 	}
1781 
1782 	if (deq > 0) {
1783 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1784 		if (virtqueue_empty(vq))
1785 			sc->vtnet_watchdog_timer = 0;
1786 	}
1787 }
1788 
1789 static struct mbuf *
1790 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1791     struct virtio_net_hdr *hdr)
1792 {
1793 	struct ifnet *ifp;
1794 	struct ether_header *eh;
1795 	struct ether_vlan_header *evh;
1796 	struct ip *ip;
1797 	struct ip6_hdr *ip6;
1798 	struct tcphdr *tcp;
1799 	int ip_offset;
1800 	uint16_t eth_type, csum_start;
1801 	uint8_t ip_proto, gso_type;
1802 
1803 	ifp = sc->vtnet_ifp;
1804 	M_ASSERTPKTHDR(m);
1805 
1806 	ip_offset = sizeof(struct ether_header);
1807 	if (m->m_len < ip_offset) {
1808 		if ((m = m_pullup(m, ip_offset)) == NULL)
1809 			return (NULL);
1810 	}
1811 
1812 	eh = mtod(m, struct ether_header *);
1813 	eth_type = ntohs(eh->ether_type);
1814 	if (eth_type == ETHERTYPE_VLAN) {
1815 		ip_offset = sizeof(struct ether_vlan_header);
1816 		if (m->m_len < ip_offset) {
1817 			if ((m = m_pullup(m, ip_offset)) == NULL)
1818 				return (NULL);
1819 		}
1820 		evh = mtod(m, struct ether_vlan_header *);
1821 		eth_type = ntohs(evh->evl_proto);
1822 	}
1823 
1824 	switch (eth_type) {
1825 	case ETHERTYPE_IP:
1826 		if (m->m_len < ip_offset + sizeof(struct ip)) {
1827 			m = m_pullup(m, ip_offset + sizeof(struct ip));
1828 			if (m == NULL)
1829 				return (NULL);
1830 		}
1831 
1832 		ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1833 		ip_proto = ip->ip_p;
1834 		csum_start = ip_offset + (ip->ip_hl << 2);
1835 		gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1836 		break;
1837 
1838 	case ETHERTYPE_IPV6:
1839 		if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1840 			m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1841 			if (m == NULL)
1842 				return (NULL);
1843 		}
1844 
1845 		ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1846 		/*
1847 		 * XXX Assume no extension headers are present. Presently,
1848 		 * this will always be true in the case of TSO, and FreeBSD
1849 		 * does not perform checksum offloading of IPv6 yet.
1850 		 */
1851 		ip_proto = ip6->ip6_nxt;
1852 		csum_start = ip_offset + sizeof(struct ip6_hdr);
1853 		gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1854 		break;
1855 
1856 	default:
1857 		return (m);
1858 	}
1859 
1860 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1861 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1862 		hdr->csum_start = csum_start;
1863 		hdr->csum_offset = m->m_pkthdr.csum_data;
1864 
1865 		sc->vtnet_stats.tx_csum_offloaded++;
1866 	}
1867 
1868 	if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1869 		if (ip_proto != IPPROTO_TCP)
1870 			return (m);
1871 
1872 		if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1873 			m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1874 			if (m == NULL)
1875 				return (NULL);
1876 		}
1877 
1878 		tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1879 		hdr->gso_type = gso_type;
1880 		hdr->hdr_len = csum_start + (tcp->th_off << 2);
1881 		hdr->gso_size = m->m_pkthdr.tso_segsz;
1882 
1883 		if (tcp->th_flags & TH_CWR) {
1884 			/*
1885 			 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1886 			 * ECN support is only configurable globally with the
1887 			 * net.inet.tcp.ecn.enable sysctl knob.
1888 			 */
1889 			if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1890 				if_printf(ifp, "TSO with ECN not supported "
1891 				    "by host\n");
1892 				m_freem(m);
1893 				return (NULL);
1894 			}
1895 
1896 			hdr->flags |= VIRTIO_NET_HDR_GSO_ECN;
1897 		}
1898 
1899 		sc->vtnet_stats.tx_tso_offloaded++;
1900 	}
1901 
1902 	return (m);
1903 }
1904 
1905 static int
1906 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1907     struct vtnet_tx_header *txhdr)
1908 {
1909 	struct sglist sg;
1910 	struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1911 	struct virtqueue *vq;
1912 	struct mbuf *m;
1913 	int collapsed, error;
1914 
1915 	vq = sc->vtnet_tx_vq;
1916 	m = *m_head;
1917 	collapsed = 0;
1918 
1919 	sglist_init(&sg, VTNET_MAX_TX_SEGS, segs);
1920 	error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1921 	KASSERT(error == 0 && sg.sg_nseg == 1,
1922 	    ("cannot add header to sglist"));
1923 
1924 again:
1925 	error = sglist_append_mbuf(&sg, m);
1926 	if (error) {
1927 		if (collapsed)
1928 			goto fail;
1929 
1930 		m = m_collapse(m, M_DONTWAIT, VTNET_MAX_TX_SEGS - 1);
1931 		if (m == NULL)
1932 			goto fail;
1933 
1934 		*m_head = m;
1935 		collapsed = 1;
1936 		goto again;
1937 	}
1938 
1939 	txhdr->vth_mbuf = m;
1940 
1941 	return (virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0));
1942 
1943 fail:
1944 	m_freem(*m_head);
1945 	*m_head = NULL;
1946 
1947 	return (ENOBUFS);
1948 }
1949 
1950 static int
1951 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1952 {
1953 	struct vtnet_tx_header *txhdr;
1954 	struct virtio_net_hdr *hdr;
1955 	struct mbuf *m;
1956 	int error;
1957 
1958 	m = *m_head;
1959 
1960 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
1961 	if (txhdr == NULL) {
1962 		*m_head = NULL;
1963 		m_freem(m);
1964 		return (ENOMEM);
1965 	}
1966 
1967 	/*
1968 	 * Always use the non-mergeable header to simplify things. When
1969 	 * the mergeable feature is negotiated, the num_buffers field
1970 	 * must be set to zero. We use vtnet_hdr_size later to enqueue
1971 	 * the correct header size to the host.
1972 	 */
1973 	hdr = &txhdr->vth_uhdr.hdr;
1974 
1975 	if (m->m_flags & M_VLANTAG) {
1976 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1977 		if ((*m_head = m) == NULL) {
1978 			error = ENOBUFS;
1979 			goto fail;
1980 		}
1981 		m->m_flags &= ~M_VLANTAG;
1982 	}
1983 
1984 	if (m->m_pkthdr.csum_flags != 0) {
1985 		m = vtnet_tx_offload(sc, m, hdr);
1986 		if ((*m_head = m) == NULL) {
1987 			error = ENOBUFS;
1988 			goto fail;
1989 		}
1990 	}
1991 
1992 	error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1993 fail:
1994 	if (error)
1995 		uma_zfree(vtnet_tx_header_zone, txhdr);
1996 
1997 	return (error);
1998 }
1999 
2000 static void
2001 vtnet_start(struct ifnet *ifp)
2002 {
2003 	struct vtnet_softc *sc;
2004 
2005 	sc = ifp->if_softc;
2006 
2007 	VTNET_LOCK(sc);
2008 	vtnet_start_locked(ifp);
2009 	VTNET_UNLOCK(sc);
2010 }
2011 
2012 static void
2013 vtnet_start_locked(struct ifnet *ifp)
2014 {
2015 	struct vtnet_softc *sc;
2016 	struct virtqueue *vq;
2017 	struct mbuf *m0;
2018 	int enq;
2019 
2020 	sc = ifp->if_softc;
2021 	vq = sc->vtnet_tx_vq;
2022 	enq = 0;
2023 
2024 	VTNET_LOCK_ASSERT(sc);
2025 
2026 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2027 	    IFF_DRV_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
2028 		return;
2029 
2030 #ifdef VTNET_TX_INTR_MODERATION
2031 	if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
2032 		vtnet_txeof(sc);
2033 #endif
2034 
2035 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2036 		if (virtqueue_full(vq)) {
2037 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2038 			break;
2039 		}
2040 
2041 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2042 		if (m0 == NULL)
2043 			break;
2044 
2045 		if (vtnet_encap(sc, &m0) != 0) {
2046 			if (m0 == NULL)
2047 				break;
2048 			IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2049 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2050 			break;
2051 		}
2052 
2053 		enq++;
2054 		ETHER_BPF_MTAP(ifp, m0);
2055 	}
2056 
2057 	if (enq > 0) {
2058 		virtqueue_notify(vq);
2059 		sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
2060 	}
2061 }
2062 
2063 static void
2064 vtnet_tick(void *xsc)
2065 {
2066 	struct vtnet_softc *sc;
2067 
2068 	sc = xsc;
2069 
2070 	VTNET_LOCK_ASSERT(sc);
2071 #ifdef VTNET_DEBUG
2072 	virtqueue_dump(sc->vtnet_rx_vq);
2073 	virtqueue_dump(sc->vtnet_tx_vq);
2074 #endif
2075 
2076 	vtnet_watchdog(sc);
2077 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2078 }
2079 
2080 static void
2081 vtnet_tx_intr_task(void *arg, int pending)
2082 {
2083 	struct vtnet_softc *sc;
2084 	struct ifnet *ifp;
2085 
2086 	sc = arg;
2087 	ifp = sc->vtnet_ifp;
2088 
2089 	VTNET_LOCK(sc);
2090 
2091 #ifdef DEVICE_POLLING
2092 	if (ifp->if_capenable & IFCAP_POLLING) {
2093 		VTNET_UNLOCK(sc);
2094 		return;
2095 	}
2096 #endif
2097 
2098 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2099 		vtnet_enable_tx_intr(sc);
2100 		VTNET_UNLOCK(sc);
2101 		return;
2102 	}
2103 
2104 	vtnet_txeof(sc);
2105 
2106 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2107 		vtnet_start_locked(ifp);
2108 
2109 	if (vtnet_enable_tx_intr(sc) != 0) {
2110 		vtnet_disable_tx_intr(sc);
2111 		sc->vtnet_stats.tx_task_rescheduled++;
2112 		VTNET_UNLOCK(sc);
2113 		taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
2114 		return;
2115 	}
2116 
2117 	VTNET_UNLOCK(sc);
2118 }
2119 
2120 static int
2121 vtnet_tx_vq_intr(void *xsc)
2122 {
2123 	struct vtnet_softc *sc;
2124 
2125 	sc = xsc;
2126 
2127 	vtnet_disable_tx_intr(sc);
2128 	taskqueue_enqueue_fast(sc->vtnet_tq, &sc->vtnet_tx_intr_task);
2129 
2130 	return (1);
2131 }
2132 
2133 static void
2134 vtnet_stop(struct vtnet_softc *sc)
2135 {
2136 	device_t dev;
2137 	struct ifnet *ifp;
2138 
2139 	dev = sc->vtnet_dev;
2140 	ifp = sc->vtnet_ifp;
2141 
2142 	VTNET_LOCK_ASSERT(sc);
2143 
2144 	sc->vtnet_watchdog_timer = 0;
2145 	callout_stop(&sc->vtnet_tick_ch);
2146 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2147 
2148 	vtnet_disable_rx_intr(sc);
2149 	vtnet_disable_tx_intr(sc);
2150 
2151 	/*
2152 	 * Stop the host VirtIO adapter. Note this will reset the host
2153 	 * adapter's state back to the pre-initialized state, so in
2154 	 * order to make the device usable again, we must drive it
2155 	 * through virtio_reinit() and virtio_reinit_complete().
2156 	 */
2157 	virtio_stop(dev);
2158 
2159 	sc->vtnet_flags &= ~VTNET_FLAG_LINK;
2160 
2161 	vtnet_free_rx_mbufs(sc);
2162 	vtnet_free_tx_mbufs(sc);
2163 }
2164 
2165 static int
2166 vtnet_reinit(struct vtnet_softc *sc)
2167 {
2168 	struct ifnet *ifp;
2169 	uint64_t features;
2170 
2171 	ifp = sc->vtnet_ifp;
2172 	features = sc->vtnet_features;
2173 
2174 	/*
2175 	 * Re-negotiate with the host, removing any disabled receive
2176 	 * features. Transmit features are disabled only on our side
2177 	 * via if_capenable and if_hwassist.
2178 	 */
2179 
2180 	if (ifp->if_capabilities & IFCAP_RXCSUM) {
2181 		if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2182 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2183 	}
2184 
2185 	if (ifp->if_capabilities & IFCAP_LRO) {
2186 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2187 			features &= ~VTNET_LRO_FEATURES;
2188 	}
2189 
2190 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2191 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2192 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2193 	}
2194 
2195 	return (virtio_reinit(sc->vtnet_dev, features));
2196 }
2197 
2198 static void
2199 vtnet_init_locked(struct vtnet_softc *sc)
2200 {
2201 	device_t dev;
2202 	struct ifnet *ifp;
2203 	int error;
2204 
2205 	dev = sc->vtnet_dev;
2206 	ifp = sc->vtnet_ifp;
2207 
2208 	VTNET_LOCK_ASSERT(sc);
2209 
2210 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2211 		return;
2212 
2213 	/* Stop host's adapter, cancel any pending I/O. */
2214 	vtnet_stop(sc);
2215 
2216 	/* Reinitialize the host device. */
2217 	error = vtnet_reinit(sc);
2218 	if (error) {
2219 		device_printf(dev,
2220 		    "reinitialization failed, stopping device...\n");
2221 		vtnet_stop(sc);
2222 		return;
2223 	}
2224 
2225 	/* Update host with assigned MAC address. */
2226 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2227 	vtnet_set_hwaddr(sc);
2228 
2229 	ifp->if_hwassist = 0;
2230 	if (ifp->if_capenable & IFCAP_TXCSUM)
2231 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2232 	if (ifp->if_capenable & IFCAP_TSO4)
2233 		ifp->if_hwassist |= CSUM_TSO;
2234 
2235 	error = vtnet_init_rx_vq(sc);
2236 	if (error) {
2237 		device_printf(dev,
2238 		    "cannot allocate mbufs for Rx virtqueue\n");
2239 		vtnet_stop(sc);
2240 		return;
2241 	}
2242 
2243 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2244 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2245 			/* Restore promiscuous and all-multicast modes. */
2246 			vtnet_rx_filter(sc);
2247 
2248 			/* Restore filtered MAC addresses. */
2249 			vtnet_rx_filter_mac(sc);
2250 		}
2251 
2252 		/* Restore VLAN filters. */
2253 		if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2254 			vtnet_rx_filter_vlan(sc);
2255 	}
2256 
2257 #ifdef DEVICE_POLLING
2258 	if (ifp->if_capenable & IFCAP_POLLING) {
2259 		vtnet_disable_rx_intr(sc);
2260 		vtnet_disable_tx_intr(sc);
2261 	} else
2262 #endif
2263 	{
2264 		vtnet_enable_rx_intr(sc);
2265 		vtnet_enable_tx_intr(sc);
2266 	}
2267 
2268 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2269 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2270 
2271 	virtio_reinit_complete(dev);
2272 
2273 	vtnet_update_link_status(sc);
2274 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2275 }
2276 
2277 static void
2278 vtnet_init(void *xsc)
2279 {
2280 	struct vtnet_softc *sc;
2281 
2282 	sc = xsc;
2283 
2284 	VTNET_LOCK(sc);
2285 	vtnet_init_locked(sc);
2286 	VTNET_UNLOCK(sc);
2287 }
2288 
2289 static void
2290 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2291     struct sglist *sg, int readable, int writable)
2292 {
2293 	struct virtqueue *vq;
2294 	void *c;
2295 
2296 	vq = sc->vtnet_ctrl_vq;
2297 
2298 	VTNET_LOCK_ASSERT(sc);
2299 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2300 	    ("no control virtqueue"));
2301 	KASSERT(virtqueue_empty(vq),
2302 	    ("control command already enqueued"));
2303 
2304 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2305 		return;
2306 
2307 	virtqueue_notify(vq);
2308 
2309 	/*
2310 	 * Poll until the command is complete. Previously, we would
2311 	 * sleep until the control virtqueue interrupt handler woke
2312 	 * us up, but dropping the VTNET_MTX leads to serialization
2313 	 * difficulties.
2314 	 *
2315 	 * Furthermore, it appears QEMU/KVM only allocates three MSIX
2316 	 * vectors. Two of those vectors are needed for the Rx and Tx
2317 	 * virtqueues. We do not support sharing both a Vq and config
2318 	 * changed notification on the same MSIX vector.
2319 	 */
2320 	c = virtqueue_poll(vq, NULL);
2321 	KASSERT(c == cookie, ("unexpected control command response"));
2322 }
2323 
2324 static void
2325 vtnet_rx_filter(struct vtnet_softc *sc)
2326 {
2327 	device_t dev;
2328 	struct ifnet *ifp;
2329 
2330 	dev = sc->vtnet_dev;
2331 	ifp = sc->vtnet_ifp;
2332 
2333 	VTNET_LOCK_ASSERT(sc);
2334 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2335 	    ("CTRL_RX feature not negotiated"));
2336 
2337 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2338 		device_printf(dev, "cannot %s promiscuous mode\n",
2339 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
2340 
2341 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2342 		device_printf(dev, "cannot %s all-multicast mode\n",
2343 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
2344 }
2345 
2346 static int
2347 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2348 {
2349 	struct virtio_net_ctrl_hdr hdr;
2350 	struct sglist_seg segs[3];
2351 	struct sglist sg;
2352 	uint8_t onoff, ack;
2353 	int error;
2354 
2355 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
2356 		return (ENOTSUP);
2357 
2358 	error = 0;
2359 
2360 	hdr.class = VIRTIO_NET_CTRL_RX;
2361 	hdr.cmd = cmd;
2362 	onoff = !!on;
2363 	ack = VIRTIO_NET_ERR;
2364 
2365 	sglist_init(&sg, 3, segs);
2366 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2367 	error |= sglist_append(&sg, &onoff, sizeof(uint8_t));
2368 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2369 	KASSERT(error == 0 && sg.sg_nseg == 3,
2370 	    ("error adding Rx filter message to sglist"));
2371 
2372 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2373 
2374 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2375 }
2376 
2377 static int
2378 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2379 {
2380 
2381 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2382 }
2383 
2384 static int
2385 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2386 {
2387 
2388 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2389 }
2390 
2391 static void
2392 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2393 {
2394 	struct virtio_net_ctrl_hdr hdr;
2395 	struct vtnet_mac_filter *filter;
2396 	struct sglist_seg segs[4];
2397 	struct sglist sg;
2398 	struct ifnet *ifp;
2399 	struct ifaddr *ifa;
2400 	struct ifmultiaddr *ifma;
2401 	int ucnt, mcnt, promisc, allmulti, error;
2402 	uint8_t ack;
2403 
2404 	ifp = sc->vtnet_ifp;
2405 	filter = sc->vtnet_mac_filter;
2406 	ucnt = 0;
2407 	mcnt = 0;
2408 	promisc = 0;
2409 	allmulti = 0;
2410 	error = 0;
2411 
2412 	VTNET_LOCK_ASSERT(sc);
2413 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2414 	    ("CTRL_RX feature not negotiated"));
2415 
2416 	/* Unicast MAC addresses: */
2417 	if_addr_rlock(ifp);
2418 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2419 		if (ifa->ifa_addr->sa_family != AF_LINK)
2420 			continue;
2421 		else if (ucnt == VTNET_MAX_MAC_ENTRIES)
2422 			break;
2423 
2424 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2425 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2426 		ucnt++;
2427 	}
2428 	if_addr_runlock(ifp);
2429 
2430 	if (ucnt >= VTNET_MAX_MAC_ENTRIES) {
2431 		promisc = 1;
2432 		filter->vmf_unicast.nentries = 0;
2433 
2434 		if_printf(ifp, "more than %d MAC addresses assigned, "
2435 		    "falling back to promiscuous mode\n",
2436 		    VTNET_MAX_MAC_ENTRIES);
2437 	} else
2438 		filter->vmf_unicast.nentries = ucnt;
2439 
2440 	/* Multicast MAC addresses: */
2441 	if_maddr_rlock(ifp);
2442 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2443 		if (ifma->ifma_addr->sa_family != AF_LINK)
2444 			continue;
2445 		else if (mcnt == VTNET_MAX_MAC_ENTRIES)
2446 			break;
2447 
2448 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2449 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2450 		mcnt++;
2451 	}
2452 	if_maddr_runlock(ifp);
2453 
2454 	if (mcnt >= VTNET_MAX_MAC_ENTRIES) {
2455 		allmulti = 1;
2456 		filter->vmf_multicast.nentries = 0;
2457 
2458 		if_printf(ifp, "more than %d multicast MAC addresses "
2459 		    "assigned, falling back to all-multicast mode\n",
2460 		    VTNET_MAX_MAC_ENTRIES);
2461 	} else
2462 		filter->vmf_multicast.nentries = mcnt;
2463 
2464 	if (promisc && allmulti)
2465 		goto out;
2466 
2467 	hdr.class = VIRTIO_NET_CTRL_MAC;
2468 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2469 	ack = VIRTIO_NET_ERR;
2470 
2471 	sglist_init(&sg, 4, segs);
2472 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2473 	error |= sglist_append(&sg, &filter->vmf_unicast,
2474 	    sizeof(struct vtnet_mac_table));
2475 	error |= sglist_append(&sg, &filter->vmf_multicast,
2476 	    sizeof(struct vtnet_mac_table));
2477 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2478 	KASSERT(error == 0 && sg.sg_nseg == 4,
2479 	    ("error adding MAC filtering message to sglist"));
2480 
2481 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2482 
2483 	if (ack != VIRTIO_NET_OK)
2484 		if_printf(ifp, "error setting host MAC filter table\n");
2485 
2486 out:
2487 	if (promisc)
2488 		if (vtnet_set_promisc(sc, 1) != 0)
2489 			if_printf(ifp, "cannot enable promiscuous mode\n");
2490 	if (allmulti)
2491 		if (vtnet_set_allmulti(sc, 1) != 0)
2492 			if_printf(ifp, "cannot enable all-multicast mode\n");
2493 }
2494 
2495 static int
2496 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2497 {
2498 	struct virtio_net_ctrl_hdr hdr;
2499 	struct sglist_seg segs[3];
2500 	struct sglist sg;
2501 	uint8_t ack;
2502 	int error;
2503 
2504 	hdr.class = VIRTIO_NET_CTRL_VLAN;
2505 	hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2506 	ack = VIRTIO_NET_ERR;
2507 	error = 0;
2508 
2509 	sglist_init(&sg, 3, segs);
2510 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2511 	error |= sglist_append(&sg, &tag, sizeof(uint16_t));
2512 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2513 	KASSERT(error == 0 && sg.sg_nseg == 3,
2514 	    ("error adding VLAN control message to sglist"));
2515 
2516 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2517 
2518 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
2519 }
2520 
2521 static void
2522 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2523 {
2524 	device_t dev;
2525 	uint32_t w, mask;
2526 	uint16_t tag;
2527 	int i, nvlans, error;
2528 
2529 	VTNET_LOCK_ASSERT(sc);
2530 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2531 	    ("VLAN_FILTER feature not negotiated"));
2532 
2533 	dev = sc->vtnet_dev;
2534 	nvlans = sc->vtnet_nvlans;
2535 	error = 0;
2536 
2537 	/* Enable filtering for each configured VLAN. */
2538 	for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2539 		w = sc->vtnet_vlan_shadow[i];
2540 		for (mask = 1, tag = i * 32; w != 0; mask <<= 1, tag++) {
2541 			if ((w & mask) != 0) {
2542 				w &= ~mask;
2543 				nvlans--;
2544 				if (vtnet_exec_vlan_filter(sc, 1, tag) != 0)
2545 					error++;
2546 			}
2547 		}
2548 	}
2549 
2550 	KASSERT(nvlans == 0, ("VLAN count incorrect"));
2551 	if (error)
2552 		device_printf(dev, "cannot restore VLAN filter table\n");
2553 }
2554 
2555 static void
2556 vtnet_set_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2557 {
2558 	struct ifnet *ifp;
2559 	int idx, bit;
2560 
2561 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2562 	    ("VLAN_FILTER feature not negotiated"));
2563 
2564 	if ((tag == 0) || (tag > 4095))
2565 		return;
2566 
2567 	ifp = sc->vtnet_ifp;
2568 	idx = (tag >> 5) & 0x7F;
2569 	bit = tag & 0x1F;
2570 
2571 	VTNET_LOCK(sc);
2572 
2573 	/* Update shadow VLAN table. */
2574 	if (add) {
2575 		sc->vtnet_nvlans++;
2576 		sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2577 	} else {
2578 		sc->vtnet_nvlans--;
2579 		sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2580 	}
2581 
2582 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2583 		if (vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2584 			device_printf(sc->vtnet_dev,
2585 			    "cannot %s VLAN %d %s the host filter table\n",
2586 			    add ? "add" : "remove", tag,
2587 			    add ? "to" : "from");
2588 		}
2589 	}
2590 
2591 	VTNET_UNLOCK(sc);
2592 }
2593 
2594 static void
2595 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2596 {
2597 
2598 	if (ifp->if_softc != arg)
2599 		return;
2600 
2601 	vtnet_set_vlan_filter(arg, 1, tag);
2602 }
2603 
2604 static void
2605 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2606 {
2607 
2608 	if (ifp->if_softc != arg)
2609 		return;
2610 
2611 	vtnet_set_vlan_filter(arg, 0, tag);
2612 }
2613 
2614 static int
2615 vtnet_ifmedia_upd(struct ifnet *ifp)
2616 {
2617 	struct vtnet_softc *sc;
2618 	struct ifmedia *ifm;
2619 
2620 	sc = ifp->if_softc;
2621 	ifm = &sc->vtnet_media;
2622 
2623 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2624 		return (EINVAL);
2625 
2626 	return (0);
2627 }
2628 
2629 static void
2630 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2631 {
2632 	struct vtnet_softc *sc;
2633 
2634 	sc = ifp->if_softc;
2635 
2636 	ifmr->ifm_status = IFM_AVALID;
2637 	ifmr->ifm_active = IFM_ETHER;
2638 
2639 	VTNET_LOCK(sc);
2640 	if (vtnet_is_link_up(sc) != 0) {
2641 		ifmr->ifm_status |= IFM_ACTIVE;
2642 		ifmr->ifm_active |= VTNET_MEDIATYPE;
2643 	} else
2644 		ifmr->ifm_active |= IFM_NONE;
2645 	VTNET_UNLOCK(sc);
2646 }
2647 
2648 static void
2649 vtnet_add_statistics(struct vtnet_softc *sc)
2650 {
2651 	device_t dev;
2652 	struct vtnet_statistics *stats;
2653         struct sysctl_ctx_list *ctx;
2654 	struct sysctl_oid *tree;
2655 	struct sysctl_oid_list *child;
2656 
2657 	dev = sc->vtnet_dev;
2658 	stats = &sc->vtnet_stats;
2659 	ctx = device_get_sysctl_ctx(dev);
2660 	tree = device_get_sysctl_tree(dev);
2661 	child = SYSCTL_CHILDREN(tree);
2662 
2663 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2664 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
2665 	    "Mbuf cluster allocation failures");
2666 
2667 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_frame_too_large",
2668 	    CTLFLAG_RD, &stats->rx_frame_too_large,
2669 	    "Received frame larger than the mbuf chain");
2670 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2671 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
2672 	    "Enqueuing the replacement receive mbuf failed");
2673 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_mergeable_failed",
2674 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
2675 	    "Mergeable buffers receive failures");
2676 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2677 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
2678 	    "Received checksum offloaded buffer with unsupported "
2679 	    "Ethernet type");
2680 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_start",
2681 	    CTLFLAG_RD, &stats->rx_csum_bad_start,
2682 	    "Received checksum offloaded buffer with incorrect start offset");
2683 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2684 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
2685 	    "Received checksum offloaded buffer with incorrect IP protocol");
2686 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2687 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
2688 	    "Received checksum offloaded buffer with incorrect offset");
2689 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_failed",
2690 	    CTLFLAG_RD, &stats->rx_csum_failed,
2691 	    "Received buffer checksum offload failed");
2692 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_offloaded",
2693 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
2694 	    "Received buffer checksum offload succeeded");
2695 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_task_rescheduled",
2696 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
2697 	    "Times the receive interrupt task rescheduled itself");
2698 
2699 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_offloaded",
2700 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
2701 	    "Offloaded checksum of transmitted buffer");
2702 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_offloaded",
2703 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
2704 	    "Segmentation offload of transmitted buffer");
2705 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2706 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
2707 	    "Aborted transmit of checksum offloaded buffer with unknown "
2708 	    "Ethernet type");
2709 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2710 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
2711 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
2712 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_task_rescheduled",
2713 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
2714 	    "Times the transmit interrupt task rescheduled itself");
2715 }
2716 
2717 static int
2718 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2719 {
2720 
2721 	return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2722 }
2723 
2724 static void
2725 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2726 {
2727 
2728 	virtqueue_disable_intr(sc->vtnet_rx_vq);
2729 }
2730 
2731 static int
2732 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2733 {
2734 
2735 #ifdef VTNET_TX_INTR_MODERATION
2736 	return (0);
2737 #else
2738 	return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2739 #endif
2740 }
2741 
2742 static void
2743 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2744 {
2745 
2746 	virtqueue_disable_intr(sc->vtnet_tx_vq);
2747 }
2748