xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 7c448f5aab8be6977d0860e608e7d497b495d28c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/msan.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/taskqueue.h>
50 #include <sys/smp.h>
51 #include <machine/smp.h>
52 
53 #include <vm/uma.h>
54 
55 #include <net/debugnet.h>
56 #include <net/ethernet.h>
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_arp.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_media.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <net/bpf.h>
67 
68 #include <netinet/in_systm.h>
69 #include <netinet/in.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_lro.h>
76 
77 #include <machine/bus.h>
78 #include <machine/resource.h>
79 #include <sys/bus.h>
80 #include <sys/rman.h>
81 
82 #include <dev/virtio/virtio.h>
83 #include <dev/virtio/virtqueue.h>
84 #include <dev/virtio/network/virtio_net.h>
85 #include <dev/virtio/network/if_vtnetvar.h>
86 #include "virtio_if.h"
87 
88 #if defined(INET) || defined(INET6)
89 #include <machine/in_cksum.h>
90 #endif
91 
92 #ifdef __NO_STRICT_ALIGNMENT
93 #define VTNET_ETHER_ALIGN 0
94 #else /* Strict alignment */
95 #define VTNET_ETHER_ALIGN ETHER_ALIGN
96 #endif
97 
98 static int	vtnet_modevent(module_t, int, void *);
99 
100 static int	vtnet_probe(device_t);
101 static int	vtnet_attach(device_t);
102 static int	vtnet_detach(device_t);
103 static int	vtnet_suspend(device_t);
104 static int	vtnet_resume(device_t);
105 static int	vtnet_shutdown(device_t);
106 static int	vtnet_attach_completed(device_t);
107 static int	vtnet_config_change(device_t);
108 
109 static int	vtnet_negotiate_features(struct vtnet_softc *);
110 static int	vtnet_setup_features(struct vtnet_softc *);
111 static int	vtnet_init_rxq(struct vtnet_softc *, int);
112 static int	vtnet_init_txq(struct vtnet_softc *, int);
113 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
114 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
115 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
116 static void	vtnet_free_rx_filters(struct vtnet_softc *);
117 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
118 static void	vtnet_alloc_interface(struct vtnet_softc *);
119 static int	vtnet_setup_interface(struct vtnet_softc *);
120 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
121 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
122 static int	vtnet_ioctl_multi(struct vtnet_softc *);
123 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
124 static int	vtnet_ioctl(if_t, u_long, caddr_t);
125 static uint64_t	vtnet_get_counter(if_t, ift_counter);
126 
127 static int	vtnet_rxq_populate(struct vtnet_rxq *);
128 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
129 static struct mbuf *
130 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
131 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
132 		    struct mbuf *, int);
133 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
134 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
135 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
136 #if defined(INET) || defined(INET6)
137 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
138 		     bool, int, struct virtio_net_hdr *);
139 static void	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
140 		    int);
141 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
142 		     struct virtio_net_hdr *);
143 #endif
144 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
145 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
146 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
147 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
148 		    struct virtio_net_hdr *);
149 static int	vtnet_rxq_eof(struct vtnet_rxq *);
150 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
151 static void	vtnet_rx_vq_intr(void *);
152 static void	vtnet_rxq_tq_intr(void *, int);
153 
154 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
155 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
156 static int	vtnet_txq_notify(struct vtnet_txq *);
157 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
158 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
159 		    int *, int *, int *);
160 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
161 		    int, struct virtio_net_hdr *);
162 static struct mbuf *
163 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
164 		    struct virtio_net_hdr *);
165 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
166 		    struct vtnet_tx_header *);
167 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
168 
169 /* Required for ALTQ */
170 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
171 static void	vtnet_start(if_t);
172 
173 /* Required for MQ */
174 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
175 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
176 static void	vtnet_txq_tq_deferred(void *, int);
177 static void	vtnet_qflush(if_t);
178 
179 
180 static void	vtnet_txq_start(struct vtnet_txq *);
181 static void	vtnet_txq_tq_intr(void *, int);
182 static int	vtnet_txq_eof(struct vtnet_txq *);
183 static void	vtnet_tx_vq_intr(void *);
184 static void	vtnet_tx_start_all(struct vtnet_softc *);
185 
186 static int	vtnet_watchdog(struct vtnet_txq *);
187 static void	vtnet_accum_stats(struct vtnet_softc *,
188 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
189 static void	vtnet_tick(void *);
190 
191 static void	vtnet_start_taskqueues(struct vtnet_softc *);
192 static void	vtnet_free_taskqueues(struct vtnet_softc *);
193 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
194 
195 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
196 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
197 static void	vtnet_stop(struct vtnet_softc *);
198 static int	vtnet_virtio_reinit(struct vtnet_softc *);
199 static void	vtnet_init_rx_filters(struct vtnet_softc *);
200 static int	vtnet_init_rx_queues(struct vtnet_softc *);
201 static int	vtnet_init_tx_queues(struct vtnet_softc *);
202 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
203 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
204 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
205 static int	vtnet_reinit(struct vtnet_softc *);
206 static void	vtnet_init_locked(struct vtnet_softc *, int);
207 static void	vtnet_init(void *);
208 
209 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
210 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
211 		    struct sglist *, int, int);
212 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
213 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
214 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
215 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
216 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
217 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
218 static void	vtnet_rx_filter(struct vtnet_softc *);
219 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
220 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
222 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
223 static void	vtnet_register_vlan(void *, if_t, uint16_t);
224 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
225 
226 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
227 static int	vtnet_is_link_up(struct vtnet_softc *);
228 static void	vtnet_update_link_status(struct vtnet_softc *);
229 static int	vtnet_ifmedia_upd(if_t);
230 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
231 static void	vtnet_get_macaddr(struct vtnet_softc *);
232 static void	vtnet_set_macaddr(struct vtnet_softc *);
233 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
234 static void	vtnet_vlan_tag_remove(struct mbuf *);
235 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
236 
237 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
238 		    struct sysctl_oid_list *, struct vtnet_rxq *);
239 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
240 		    struct sysctl_oid_list *, struct vtnet_txq *);
241 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
242 static void	vtnet_load_tunables(struct vtnet_softc *);
243 static void	vtnet_setup_sysctl(struct vtnet_softc *);
244 
245 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
246 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
247 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
248 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
249 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
250 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
251 static void	vtnet_enable_interrupts(struct vtnet_softc *);
252 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
253 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
254 static void	vtnet_disable_interrupts(struct vtnet_softc *);
255 
256 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
257 
258 DEBUGNET_DEFINE(vtnet);
259 
260 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
261 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
262 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
264 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
265 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
266 
267 /* Tunables. */
268 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
269     "VirtIO Net driver parameters");
270 
271 static int vtnet_csum_disable = 0;
272 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
273     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
274 
275 static int vtnet_fixup_needs_csum = 0;
276 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
277     &vtnet_fixup_needs_csum, 0,
278     "Calculate valid checksum for NEEDS_CSUM packets");
279 
280 static int vtnet_tso_disable = 0;
281 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
282     &vtnet_tso_disable, 0, "Disables TSO");
283 
284 static int vtnet_lro_disable = 0;
285 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
286     &vtnet_lro_disable, 0, "Disables hardware LRO");
287 
288 static int vtnet_mq_disable = 0;
289 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
290     &vtnet_mq_disable, 0, "Disables multiqueue support");
291 
292 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
293 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
294     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
295 
296 static int vtnet_tso_maxlen = IP_MAXPACKET;
297 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
298     &vtnet_tso_maxlen, 0, "TSO burst limit");
299 
300 static int vtnet_rx_process_limit = 1024;
301 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
302     &vtnet_rx_process_limit, 0,
303     "Number of RX segments processed in one pass");
304 
305 static int vtnet_lro_entry_count = 128;
306 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
307     &vtnet_lro_entry_count, 0, "Software LRO entry count");
308 
309 /* Enable sorted LRO, and the depth of the mbuf queue. */
310 static int vtnet_lro_mbufq_depth = 0;
311 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
312     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
313 
314 /* Deactivate ALTQ Support */
315 static int vtnet_altq_disable = 0;
316 SYSCTL_INT(_hw_vtnet, OID_AUTO, altq_disable, CTLFLAG_RDTUN,
317     &vtnet_altq_disable, 0, "Disables ALTQ Support");
318 
319 /*
320  * For the driver to be considered as having altq enabled,
321  * it must be compiled with an ALTQ capable kernel,
322  * and the tunable hw.vtnet.altq_disable must be zero
323  */
324 #define VTNET_ALTQ_ENABLED (VTNET_ALTQ_CAPABLE && (!vtnet_altq_disable))
325 
326 
327 static uma_zone_t vtnet_tx_header_zone;
328 
329 static struct virtio_feature_desc vtnet_feature_desc[] = {
330 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
331 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
332 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
333 	{ VIRTIO_NET_F_MAC,			"MAC"			},
334 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
335 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
336 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
337 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
338 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
339 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
340 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
341 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
342 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
343 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
344 	{ VIRTIO_NET_F_STATUS,			"Status"		},
345 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
346 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
347 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
348 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
349 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
350 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
351 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
352 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
353 
354 	{ 0, NULL }
355 };
356 
357 static device_method_t vtnet_methods[] = {
358 	/* Device methods. */
359 	DEVMETHOD(device_probe,			vtnet_probe),
360 	DEVMETHOD(device_attach,		vtnet_attach),
361 	DEVMETHOD(device_detach,		vtnet_detach),
362 	DEVMETHOD(device_suspend,		vtnet_suspend),
363 	DEVMETHOD(device_resume,		vtnet_resume),
364 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
365 
366 	/* VirtIO methods. */
367 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
368 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
369 
370 	DEVMETHOD_END
371 };
372 
373 #ifdef DEV_NETMAP
374 #include <dev/netmap/if_vtnet_netmap.h>
375 #endif
376 
377 static driver_t vtnet_driver = {
378     .name = "vtnet",
379     .methods = vtnet_methods,
380     .size = sizeof(struct vtnet_softc)
381 };
382 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
383 MODULE_VERSION(vtnet, 1);
384 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
385 #ifdef DEV_NETMAP
386 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
387 #endif
388 
389 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
390 
391 static int
vtnet_modevent(module_t mod __unused,int type,void * unused __unused)392 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
393 {
394 	int error = 0;
395 	static int loaded = 0;
396 
397 	switch (type) {
398 	case MOD_LOAD:
399 		if (loaded++ == 0) {
400 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
401 				sizeof(struct vtnet_tx_header),
402 				NULL, NULL, NULL, NULL, 0, 0);
403 #ifdef DEBUGNET
404 			/*
405 			 * We need to allocate from this zone in the transmit path, so ensure
406 			 * that we have at least one item per header available.
407 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
408 			 * buckets
409 			 */
410 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
411 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
412 #endif
413 		}
414 		break;
415 	case MOD_QUIESCE:
416 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
417 			error = EBUSY;
418 		break;
419 	case MOD_UNLOAD:
420 		if (--loaded == 0) {
421 			uma_zdestroy(vtnet_tx_header_zone);
422 			vtnet_tx_header_zone = NULL;
423 		}
424 		break;
425 	case MOD_SHUTDOWN:
426 		break;
427 	default:
428 		error = EOPNOTSUPP;
429 		break;
430 	}
431 
432 	return (error);
433 }
434 
435 static int
vtnet_probe(device_t dev)436 vtnet_probe(device_t dev)
437 {
438 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
439 }
440 
441 static int
vtnet_attach(device_t dev)442 vtnet_attach(device_t dev)
443 {
444 	struct vtnet_softc *sc;
445 	int error;
446 
447 	sc = device_get_softc(dev);
448 	sc->vtnet_dev = dev;
449 	virtio_set_feature_desc(dev, vtnet_feature_desc);
450 
451 	VTNET_CORE_LOCK_INIT(sc);
452 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
453 	vtnet_load_tunables(sc);
454 
455 	vtnet_alloc_interface(sc);
456 	vtnet_setup_sysctl(sc);
457 
458 	error = vtnet_setup_features(sc);
459 	if (error) {
460 		device_printf(dev, "cannot setup features\n");
461 		goto fail;
462 	}
463 
464 	error = vtnet_alloc_rx_filters(sc);
465 	if (error) {
466 		device_printf(dev, "cannot allocate Rx filters\n");
467 		goto fail;
468 	}
469 
470 	error = vtnet_alloc_rxtx_queues(sc);
471 	if (error) {
472 		device_printf(dev, "cannot allocate queues\n");
473 		goto fail;
474 	}
475 
476 	error = vtnet_alloc_virtqueues(sc);
477 	if (error) {
478 		device_printf(dev, "cannot allocate virtqueues\n");
479 		goto fail;
480 	}
481 
482 	error = vtnet_setup_interface(sc);
483 	if (error) {
484 		device_printf(dev, "cannot setup interface\n");
485 		goto fail;
486 	}
487 
488 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
489 	if (error) {
490 		device_printf(dev, "cannot setup interrupts\n");
491 		ether_ifdetach(sc->vtnet_ifp);
492 		goto fail;
493 	}
494 
495 #ifdef DEV_NETMAP
496 	vtnet_netmap_attach(sc);
497 #endif
498 	vtnet_start_taskqueues(sc);
499 
500 fail:
501 	if (error)
502 		vtnet_detach(dev);
503 
504 	return (error);
505 }
506 
507 static int
vtnet_detach(device_t dev)508 vtnet_detach(device_t dev)
509 {
510 	struct vtnet_softc *sc;
511 	if_t ifp;
512 
513 	sc = device_get_softc(dev);
514 	ifp = sc->vtnet_ifp;
515 
516 	if (device_is_attached(dev)) {
517 		VTNET_CORE_LOCK(sc);
518 		vtnet_stop(sc);
519 		VTNET_CORE_UNLOCK(sc);
520 
521 		callout_drain(&sc->vtnet_tick_ch);
522 		vtnet_drain_taskqueues(sc);
523 
524 		ether_ifdetach(ifp);
525 	}
526 
527 #ifdef DEV_NETMAP
528 	netmap_detach(ifp);
529 #endif
530 
531 	if (sc->vtnet_pfil != NULL) {
532 		pfil_head_unregister(sc->vtnet_pfil);
533 		sc->vtnet_pfil = NULL;
534 	}
535 
536 	vtnet_free_taskqueues(sc);
537 
538 	if (sc->vtnet_vlan_attach != NULL) {
539 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
540 		sc->vtnet_vlan_attach = NULL;
541 	}
542 	if (sc->vtnet_vlan_detach != NULL) {
543 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
544 		sc->vtnet_vlan_detach = NULL;
545 	}
546 
547 	ifmedia_removeall(&sc->vtnet_media);
548 
549 	if (ifp != NULL) {
550 		if_free(ifp);
551 		sc->vtnet_ifp = NULL;
552 	}
553 
554 	vtnet_free_rxtx_queues(sc);
555 	vtnet_free_rx_filters(sc);
556 
557 	if (sc->vtnet_ctrl_vq != NULL)
558 		vtnet_free_ctrl_vq(sc);
559 
560 	VTNET_CORE_LOCK_DESTROY(sc);
561 
562 	return (0);
563 }
564 
565 static int
vtnet_suspend(device_t dev)566 vtnet_suspend(device_t dev)
567 {
568 	struct vtnet_softc *sc;
569 
570 	sc = device_get_softc(dev);
571 
572 	VTNET_CORE_LOCK(sc);
573 	vtnet_stop(sc);
574 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
575 	VTNET_CORE_UNLOCK(sc);
576 
577 	return (0);
578 }
579 
580 static int
vtnet_resume(device_t dev)581 vtnet_resume(device_t dev)
582 {
583 	struct vtnet_softc *sc;
584 	if_t ifp;
585 
586 	sc = device_get_softc(dev);
587 	ifp = sc->vtnet_ifp;
588 
589 	VTNET_CORE_LOCK(sc);
590 	if (if_getflags(ifp) & IFF_UP)
591 		vtnet_init_locked(sc, 0);
592 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
593 	VTNET_CORE_UNLOCK(sc);
594 
595 	return (0);
596 }
597 
598 static int
vtnet_shutdown(device_t dev)599 vtnet_shutdown(device_t dev)
600 {
601 	/*
602 	 * Suspend already does all of what we need to
603 	 * do here; we just never expect to be resumed.
604 	 */
605 	return (vtnet_suspend(dev));
606 }
607 
608 static int
vtnet_attach_completed(device_t dev)609 vtnet_attach_completed(device_t dev)
610 {
611 	struct vtnet_softc *sc;
612 
613 	sc = device_get_softc(dev);
614 
615 	VTNET_CORE_LOCK(sc);
616 	vtnet_attached_set_macaddr(sc);
617 	VTNET_CORE_UNLOCK(sc);
618 
619 	return (0);
620 }
621 
622 static int
vtnet_config_change(device_t dev)623 vtnet_config_change(device_t dev)
624 {
625 	struct vtnet_softc *sc;
626 
627 	sc = device_get_softc(dev);
628 
629 	VTNET_CORE_LOCK(sc);
630 	vtnet_update_link_status(sc);
631 	if (sc->vtnet_link_active != 0)
632 		vtnet_tx_start_all(sc);
633 	VTNET_CORE_UNLOCK(sc);
634 
635 	return (0);
636 }
637 
638 static int
vtnet_negotiate_features(struct vtnet_softc * sc)639 vtnet_negotiate_features(struct vtnet_softc *sc)
640 {
641 	device_t dev;
642 	uint64_t features, negotiated_features;
643 	int no_csum;
644 
645 	dev = sc->vtnet_dev;
646 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
647 	    VTNET_LEGACY_FEATURES;
648 
649 	/*
650 	 * TSO and LRO are only available when their corresponding checksum
651 	 * offload feature is also negotiated.
652 	 */
653 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
654 	if (no_csum)
655 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
656 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
657 		features &= ~VTNET_TSO_FEATURES;
658 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
659 		features &= ~VTNET_LRO_FEATURES;
660 
661 	/* Deactivate MQ Feature flag, if driver has ALTQ enabled, or MQ is explicitly disabled */
662 	if (VTNET_ALTQ_ENABLED || vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
663 		features &= ~VIRTIO_NET_F_MQ;
664 
665 	negotiated_features = virtio_negotiate_features(dev, features);
666 
667 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
668 		uint16_t mtu;
669 
670 		mtu = virtio_read_dev_config_2(dev,
671 		    offsetof(struct virtio_net_config, mtu));
672 		if (mtu < VTNET_MIN_MTU) {
673 			device_printf(dev, "Invalid MTU value: %d. "
674 			    "MTU feature disabled.\n", mtu);
675 			features &= ~VIRTIO_NET_F_MTU;
676 			negotiated_features =
677 			    virtio_negotiate_features(dev, features);
678 		}
679 	}
680 
681 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
682 		uint16_t npairs;
683 
684 		npairs = virtio_read_dev_config_2(dev,
685 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
686 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
687 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
688 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
689 			    "%d. Multiqueue feature disabled.\n", npairs);
690 			features &= ~VIRTIO_NET_F_MQ;
691 			negotiated_features =
692 			    virtio_negotiate_features(dev, features);
693 		}
694 	}
695 
696 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
697 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
698 		/*
699 		 * LRO without mergeable buffers requires special care. This
700 		 * is not ideal because every receive buffer must be large
701 		 * enough to hold the maximum TCP packet, the Ethernet header,
702 		 * and the header. This requires up to 34 descriptors with
703 		 * MCLBYTES clusters. If we do not have indirect descriptors,
704 		 * LRO is disabled since the virtqueue will not contain very
705 		 * many receive buffers.
706 		 */
707 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
708 			device_printf(dev,
709 			    "Host LRO disabled since both mergeable buffers "
710 			    "and indirect descriptors were not negotiated\n");
711 			features &= ~VTNET_LRO_FEATURES;
712 			negotiated_features =
713 			    virtio_negotiate_features(dev, features);
714 		} else
715 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
716 	}
717 
718 	sc->vtnet_features = negotiated_features;
719 	sc->vtnet_negotiated_features = negotiated_features;
720 
721 	return (virtio_finalize_features(dev));
722 }
723 
724 static int
vtnet_setup_features(struct vtnet_softc * sc)725 vtnet_setup_features(struct vtnet_softc *sc)
726 {
727 	device_t dev;
728 	int error;
729 
730 	dev = sc->vtnet_dev;
731 
732 	error = vtnet_negotiate_features(sc);
733 	if (error)
734 		return (error);
735 
736 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
737 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
738 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
739 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
740 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
741 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
742 
743 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
744 		/* This feature should always be negotiated. */
745 		sc->vtnet_flags |= VTNET_FLAG_MAC;
746 	}
747 
748 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
749 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
750 		    offsetof(struct virtio_net_config, mtu));
751 	} else
752 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
753 
754 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
755 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
756 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
757 	} else if (vtnet_modern(sc)) {
758 		/* This is identical to the mergeable header. */
759 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
760 	} else
761 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
762 
763 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
764 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
765 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
766 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
767 	else
768 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
769 
770 	/*
771 	 * Favor "hardware" LRO if negotiated, but support software LRO as
772 	 * a fallback; there is usually little benefit (or worse) with both.
773 	 */
774 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
775 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
776 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
777 
778 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
779 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
780 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
781 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
782 	else
783 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
784 
785 	sc->vtnet_req_vq_pairs = 1;
786 	sc->vtnet_max_vq_pairs = 1;
787 
788 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
789 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
790 
791 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
792 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
793 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
794 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
795 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
796 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
797 
798 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
799 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
800 			    offsetof(struct virtio_net_config,
801 			    max_virtqueue_pairs));
802 		}
803 	}
804 
805 	if (sc->vtnet_max_vq_pairs > 1) {
806 		int req;
807 
808 		/*
809 		 * Limit the maximum number of requested queue pairs to the
810 		 * number of CPUs and the configured maximum.
811 		 */
812 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
813 		if (req < 0)
814 			req = 1;
815 		if (req == 0)
816 			req = mp_ncpus;
817 		if (req > sc->vtnet_max_vq_pairs)
818 			req = sc->vtnet_max_vq_pairs;
819 		if (req > mp_ncpus)
820 			req = mp_ncpus;
821 		if (req > 1) {
822 			sc->vtnet_req_vq_pairs = req;
823 			sc->vtnet_flags |= VTNET_FLAG_MQ;
824 		}
825 	}
826 
827 	return (0);
828 }
829 
830 static int
vtnet_init_rxq(struct vtnet_softc * sc,int id)831 vtnet_init_rxq(struct vtnet_softc *sc, int id)
832 {
833 	struct vtnet_rxq *rxq;
834 
835 	rxq = &sc->vtnet_rxqs[id];
836 
837 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
838 	    device_get_nameunit(sc->vtnet_dev), id);
839 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
840 
841 	rxq->vtnrx_sc = sc;
842 	rxq->vtnrx_id = id;
843 
844 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
845 	if (rxq->vtnrx_sg == NULL)
846 		return (ENOMEM);
847 
848 #if defined(INET) || defined(INET6)
849 	if (vtnet_software_lro(sc)) {
850 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
851 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
852 			return (ENOMEM);
853 	}
854 #endif
855 
856 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
857 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
858 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
859 
860 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
861 }
862 
863 static int
vtnet_init_txq(struct vtnet_softc * sc,int id)864 vtnet_init_txq(struct vtnet_softc *sc, int id)
865 {
866 	struct vtnet_txq *txq;
867 
868 	txq = &sc->vtnet_txqs[id];
869 
870 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
871 	    device_get_nameunit(sc->vtnet_dev), id);
872 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
873 
874 	txq->vtntx_sc = sc;
875 	txq->vtntx_id = id;
876 
877 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
878 	if (txq->vtntx_sg == NULL)
879 		return (ENOMEM);
880 
881 	if (!VTNET_ALTQ_ENABLED) {
882 		txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
883 		    M_NOWAIT, &txq->vtntx_mtx);
884 		if (txq->vtntx_br == NULL)
885 			return (ENOMEM);
886 
887 		TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
888 	}
889 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
890 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
891 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
892 	if (txq->vtntx_tq == NULL)
893 		return (ENOMEM);
894 
895 	return (0);
896 }
897 
898 static int
vtnet_alloc_rxtx_queues(struct vtnet_softc * sc)899 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
900 {
901 	int i, npairs, error;
902 
903 	npairs = sc->vtnet_max_vq_pairs;
904 
905 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
906 	    M_NOWAIT | M_ZERO);
907 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
908 	    M_NOWAIT | M_ZERO);
909 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
910 		return (ENOMEM);
911 
912 	for (i = 0; i < npairs; i++) {
913 		error = vtnet_init_rxq(sc, i);
914 		if (error)
915 			return (error);
916 		error = vtnet_init_txq(sc, i);
917 		if (error)
918 			return (error);
919 	}
920 
921 	vtnet_set_rx_process_limit(sc);
922 	vtnet_setup_queue_sysctl(sc);
923 
924 	return (0);
925 }
926 
927 static void
vtnet_destroy_rxq(struct vtnet_rxq * rxq)928 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
929 {
930 
931 	rxq->vtnrx_sc = NULL;
932 	rxq->vtnrx_id = -1;
933 
934 #if defined(INET) || defined(INET6)
935 	tcp_lro_free(&rxq->vtnrx_lro);
936 #endif
937 
938 	if (rxq->vtnrx_sg != NULL) {
939 		sglist_free(rxq->vtnrx_sg);
940 		rxq->vtnrx_sg = NULL;
941 	}
942 
943 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
944 		mtx_destroy(&rxq->vtnrx_mtx);
945 }
946 
947 static void
vtnet_destroy_txq(struct vtnet_txq * txq)948 vtnet_destroy_txq(struct vtnet_txq *txq)
949 {
950 
951 	txq->vtntx_sc = NULL;
952 	txq->vtntx_id = -1;
953 
954 	if (txq->vtntx_sg != NULL) {
955 		sglist_free(txq->vtntx_sg);
956 		txq->vtntx_sg = NULL;
957 	}
958 
959 	if (!VTNET_ALTQ_ENABLED) {
960 		if (txq->vtntx_br != NULL) {
961 			buf_ring_free(txq->vtntx_br, M_DEVBUF);
962 			txq->vtntx_br = NULL;
963 		}
964 	}
965 
966 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
967 		mtx_destroy(&txq->vtntx_mtx);
968 }
969 
970 static void
vtnet_free_rxtx_queues(struct vtnet_softc * sc)971 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
972 {
973 	int i;
974 
975 	if (sc->vtnet_rxqs != NULL) {
976 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
977 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
978 		free(sc->vtnet_rxqs, M_DEVBUF);
979 		sc->vtnet_rxqs = NULL;
980 	}
981 
982 	if (sc->vtnet_txqs != NULL) {
983 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
984 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
985 		free(sc->vtnet_txqs, M_DEVBUF);
986 		sc->vtnet_txqs = NULL;
987 	}
988 }
989 
990 static int
vtnet_alloc_rx_filters(struct vtnet_softc * sc)991 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
992 {
993 
994 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
995 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
996 		    M_DEVBUF, M_NOWAIT | M_ZERO);
997 		if (sc->vtnet_mac_filter == NULL)
998 			return (ENOMEM);
999 	}
1000 
1001 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1002 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
1003 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
1004 		if (sc->vtnet_vlan_filter == NULL)
1005 			return (ENOMEM);
1006 	}
1007 
1008 	return (0);
1009 }
1010 
1011 static void
vtnet_free_rx_filters(struct vtnet_softc * sc)1012 vtnet_free_rx_filters(struct vtnet_softc *sc)
1013 {
1014 
1015 	if (sc->vtnet_mac_filter != NULL) {
1016 		free(sc->vtnet_mac_filter, M_DEVBUF);
1017 		sc->vtnet_mac_filter = NULL;
1018 	}
1019 
1020 	if (sc->vtnet_vlan_filter != NULL) {
1021 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1022 		sc->vtnet_vlan_filter = NULL;
1023 	}
1024 }
1025 
1026 static int
vtnet_alloc_virtqueues(struct vtnet_softc * sc)1027 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1028 {
1029 	device_t dev;
1030 	struct vq_alloc_info *info;
1031 	struct vtnet_rxq *rxq;
1032 	struct vtnet_txq *txq;
1033 	int i, idx, nvqs, error;
1034 
1035 	dev = sc->vtnet_dev;
1036 
1037 	nvqs = sc->vtnet_max_vq_pairs * 2;
1038 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1039 		nvqs++;
1040 
1041 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1042 	if (info == NULL)
1043 		return (ENOMEM);
1044 
1045 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1046 		rxq = &sc->vtnet_rxqs[i];
1047 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1048 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1049 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1050 
1051 		txq = &sc->vtnet_txqs[i];
1052 		VQ_ALLOC_INFO_INIT(&info[idx + 1], sc->vtnet_tx_nsegs,
1053 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1054 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1055 	}
1056 
1057 	/* These queues will not be used so allocate the minimum resources. */
1058 	for (; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1059 		rxq = &sc->vtnet_rxqs[i];
1060 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1061 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1062 
1063 		txq = &sc->vtnet_txqs[i];
1064 		VQ_ALLOC_INFO_INIT(&info[idx + 1], 0, NULL, txq, &txq->vtntx_vq,
1065 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1066 	}
1067 
1068 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1069 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1070 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1071 	}
1072 
1073 	error = virtio_alloc_virtqueues(dev, nvqs, info);
1074 	free(info, M_TEMP);
1075 
1076 	return (error);
1077 }
1078 
1079 static void
vtnet_alloc_interface(struct vtnet_softc * sc)1080 vtnet_alloc_interface(struct vtnet_softc *sc)
1081 {
1082 	device_t dev;
1083 	if_t ifp;
1084 
1085 	dev = sc->vtnet_dev;
1086 
1087 	ifp = if_alloc(IFT_ETHER);
1088 	sc->vtnet_ifp = ifp;
1089 	if_setsoftc(ifp, sc);
1090 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1091 }
1092 
1093 static int
vtnet_setup_interface(struct vtnet_softc * sc)1094 vtnet_setup_interface(struct vtnet_softc *sc)
1095 {
1096 	device_t dev;
1097 	struct pfil_head_args pa;
1098 	if_t ifp;
1099 
1100 	dev = sc->vtnet_dev;
1101 	ifp = sc->vtnet_ifp;
1102 
1103 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1104 	if_setbaudrate(ifp, IF_Gbps(10));
1105 	if_setinitfn(ifp, vtnet_init);
1106 	if_setioctlfn(ifp, vtnet_ioctl);
1107 	if_setgetcounterfn(ifp, vtnet_get_counter);
1108 
1109 	if (!VTNET_ALTQ_ENABLED) {
1110 		if_settransmitfn(ifp, vtnet_txq_mq_start);
1111 		if_setqflushfn(ifp, vtnet_qflush);
1112 	} else {
1113 		struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1114 		if_setstartfn(ifp, vtnet_start);
1115 		if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1116 		if_setsendqready(ifp);
1117 	}
1118 
1119 	vtnet_get_macaddr(sc);
1120 
1121 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1122 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1123 
1124 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1125 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1126 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1127 
1128 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1129 		int gso;
1130 
1131 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1132 
1133 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1134 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1135 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1136 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1137 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1138 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1139 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1140 
1141 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1142 			int tso_maxlen;
1143 
1144 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1145 
1146 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1147 			    vtnet_tso_maxlen);
1148 			if_sethwtsomax(ifp, tso_maxlen -
1149 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1150 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1151 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1152 		}
1153 	}
1154 
1155 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1156 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1157 #ifdef notyet
1158 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1159 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1160 #endif
1161 
1162 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1163 		    vtnet_fixup_needs_csum) != 0)
1164 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1165 
1166 		/* Support either "hardware" or software LRO. */
1167 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1168 	}
1169 
1170 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1171 		/*
1172 		 * VirtIO does not support VLAN tagging, but we can fake
1173 		 * it by inserting and removing the 802.1Q header during
1174 		 * transmit and receive. We are then able to do checksum
1175 		 * offloading of VLAN frames.
1176 		 */
1177 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1178 	}
1179 
1180 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1181 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1182 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1183 	if_setcapabilitiesbit(ifp, IFCAP_HWSTATS, 0);
1184 
1185 	/*
1186 	 * Capabilities after here are not enabled by default.
1187 	 */
1188 	if_setcapenable(ifp, if_getcapabilities(ifp));
1189 
1190 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1191 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1192 
1193 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1194 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1195 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1196 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1197 	}
1198 
1199 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1200 
1201 	/* Tell the upper layer(s) we support long frames. */
1202 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1203 
1204 	DEBUGNET_SET(ifp, vtnet);
1205 
1206 	pa.pa_version = PFIL_VERSION;
1207 	pa.pa_flags = PFIL_IN;
1208 	pa.pa_type = PFIL_TYPE_ETHERNET;
1209 	pa.pa_headname = if_name(ifp);
1210 	sc->vtnet_pfil = pfil_head_register(&pa);
1211 
1212 	return (0);
1213 }
1214 
1215 static int
vtnet_rx_cluster_size(struct vtnet_softc * sc,int mtu)1216 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1217 {
1218 	int framesz;
1219 
1220 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1221 		return (MJUMPAGESIZE);
1222 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1223 		return (MCLBYTES);
1224 
1225 	/*
1226 	 * Try to scale the receive mbuf cluster size from the MTU. We
1227 	 * could also use the VQ size to influence the selected size,
1228 	 * but that would only matter for very small queues.
1229 	 */
1230 	if (vtnet_modern(sc)) {
1231 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1232 		framesz = sizeof(struct virtio_net_hdr_v1);
1233 	} else
1234 		framesz = sizeof(struct vtnet_rx_header);
1235 	framesz += sizeof(struct ether_vlan_header) + mtu;
1236 	/*
1237 	 * Account for the offsetting we'll do elsewhere so we allocate the
1238 	 * right size for the mtu.
1239 	 */
1240 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1241 		framesz += VTNET_ETHER_ALIGN;
1242 	}
1243 
1244 	if (framesz <= MCLBYTES)
1245 		return (MCLBYTES);
1246 	else if (framesz <= MJUMPAGESIZE)
1247 		return (MJUMPAGESIZE);
1248 	else if (framesz <= MJUM9BYTES)
1249 		return (MJUM9BYTES);
1250 
1251 	/* Sane default; avoid 16KB clusters. */
1252 	return (MCLBYTES);
1253 }
1254 
1255 static int
vtnet_ioctl_mtu(struct vtnet_softc * sc,u_int mtu)1256 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1257 {
1258 	if_t ifp;
1259 	int clustersz;
1260 
1261 	ifp = sc->vtnet_ifp;
1262 	VTNET_CORE_LOCK_ASSERT(sc);
1263 
1264 	if (if_getmtu(ifp) == mtu)
1265 		return (0);
1266 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1267 		return (EINVAL);
1268 
1269 	if_setmtu(ifp, mtu);
1270 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1271 
1272 	if (clustersz != sc->vtnet_rx_clustersz &&
1273 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1274 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1275 		vtnet_init_locked(sc, 0);
1276 	}
1277 
1278 	return (0);
1279 }
1280 
1281 static int
vtnet_ioctl_ifflags(struct vtnet_softc * sc)1282 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1283 {
1284 	if_t ifp;
1285 	int drv_running;
1286 
1287 	ifp = sc->vtnet_ifp;
1288 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1289 
1290 	VTNET_CORE_LOCK_ASSERT(sc);
1291 
1292 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1293 		if (drv_running)
1294 			vtnet_stop(sc);
1295 		goto out;
1296 	}
1297 
1298 	if (!drv_running) {
1299 		vtnet_init_locked(sc, 0);
1300 		goto out;
1301 	}
1302 
1303 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1304 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1305 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1306 			vtnet_rx_filter(sc);
1307 		else {
1308 			/*
1309 			 * We don't support filtering out multicast, so
1310 			 * ALLMULTI is always set.
1311 			 */
1312 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1313 			if_setflagbits(ifp, IFF_PROMISC, 0);
1314 		}
1315 	}
1316 
1317 out:
1318 	sc->vtnet_if_flags = if_getflags(ifp);
1319 	return (0);
1320 }
1321 
1322 static int
vtnet_ioctl_multi(struct vtnet_softc * sc)1323 vtnet_ioctl_multi(struct vtnet_softc *sc)
1324 {
1325 	if_t ifp;
1326 
1327 	ifp = sc->vtnet_ifp;
1328 
1329 	VTNET_CORE_LOCK_ASSERT(sc);
1330 
1331 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1332 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1333 		vtnet_rx_filter_mac(sc);
1334 
1335 	return (0);
1336 }
1337 
1338 static int
vtnet_ioctl_ifcap(struct vtnet_softc * sc,struct ifreq * ifr)1339 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1340 {
1341 	if_t ifp;
1342 	int mask, reinit, update;
1343 
1344 	ifp = sc->vtnet_ifp;
1345 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1346 	reinit = update = 0;
1347 
1348 	VTNET_CORE_LOCK_ASSERT(sc);
1349 
1350 	if (mask & IFCAP_TXCSUM)
1351 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1352 	if (mask & IFCAP_TXCSUM_IPV6)
1353 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1354 	if (mask & IFCAP_TSO4)
1355 		if_togglecapenable(ifp, IFCAP_TSO4);
1356 	if (mask & IFCAP_TSO6)
1357 		if_togglecapenable(ifp, IFCAP_TSO6);
1358 
1359 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1360 		/*
1361 		 * These Rx features require the negotiated features to
1362 		 * be updated. Avoid a full reinit if possible.
1363 		 */
1364 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1365 			update = 1;
1366 		else
1367 			reinit = 1;
1368 
1369 		/* BMV: Avoid needless renegotiation for just software LRO. */
1370 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1371 		    IFCAP_LRO && vtnet_software_lro(sc))
1372 			reinit = update = 0;
1373 
1374 		if (mask & IFCAP_RXCSUM)
1375 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1376 		if (mask & IFCAP_RXCSUM_IPV6)
1377 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1378 		if (mask & IFCAP_LRO)
1379 			if_togglecapenable(ifp, IFCAP_LRO);
1380 
1381 		/*
1382 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1383 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1384 		 * checksums.
1385 		 */
1386 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1387 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1388 #ifdef notyet
1389 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1390 #endif
1391 		} else
1392 			if_setcapenablebit(ifp, 0,
1393 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1394 	}
1395 
1396 	if (mask & IFCAP_VLAN_HWFILTER) {
1397 		/* These Rx features require renegotiation. */
1398 		reinit = 1;
1399 
1400 		if (mask & IFCAP_VLAN_HWFILTER)
1401 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1402 	}
1403 
1404 	if (mask & IFCAP_VLAN_HWTSO)
1405 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1406 	if (mask & IFCAP_VLAN_HWTAGGING)
1407 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1408 
1409 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1410 		if (reinit) {
1411 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1412 			vtnet_init_locked(sc, 0);
1413 		} else if (update)
1414 			vtnet_update_rx_offloads(sc);
1415 	}
1416 
1417 	return (0);
1418 }
1419 
1420 static int
vtnet_ioctl(if_t ifp,u_long cmd,caddr_t data)1421 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1422 {
1423 	struct vtnet_softc *sc;
1424 	struct ifreq *ifr;
1425 	int error;
1426 
1427 	sc = if_getsoftc(ifp);
1428 	ifr = (struct ifreq *) data;
1429 	error = 0;
1430 
1431 	switch (cmd) {
1432 	case SIOCSIFMTU:
1433 		VTNET_CORE_LOCK(sc);
1434 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1435 		VTNET_CORE_UNLOCK(sc);
1436 		break;
1437 
1438 	case SIOCSIFFLAGS:
1439 		VTNET_CORE_LOCK(sc);
1440 		error = vtnet_ioctl_ifflags(sc);
1441 		VTNET_CORE_UNLOCK(sc);
1442 		break;
1443 
1444 	case SIOCADDMULTI:
1445 	case SIOCDELMULTI:
1446 		VTNET_CORE_LOCK(sc);
1447 		error = vtnet_ioctl_multi(sc);
1448 		VTNET_CORE_UNLOCK(sc);
1449 		break;
1450 
1451 	case SIOCSIFMEDIA:
1452 	case SIOCGIFMEDIA:
1453 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1454 		break;
1455 
1456 	case SIOCSIFCAP:
1457 		VTNET_CORE_LOCK(sc);
1458 		error = vtnet_ioctl_ifcap(sc, ifr);
1459 		VTNET_CORE_UNLOCK(sc);
1460 		VLAN_CAPABILITIES(ifp);
1461 		break;
1462 
1463 	default:
1464 		error = ether_ioctl(ifp, cmd, data);
1465 		break;
1466 	}
1467 
1468 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1469 
1470 	return (error);
1471 }
1472 
1473 static int
vtnet_rxq_populate(struct vtnet_rxq * rxq)1474 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1475 {
1476 	struct virtqueue *vq;
1477 	int nbufs, error;
1478 
1479 #ifdef DEV_NETMAP
1480 	error = vtnet_netmap_rxq_populate(rxq);
1481 	if (error >= 0)
1482 		return (error);
1483 #endif  /* DEV_NETMAP */
1484 
1485 	vq = rxq->vtnrx_vq;
1486 	error = ENOSPC;
1487 
1488 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1489 		error = vtnet_rxq_new_buf(rxq);
1490 		if (error)
1491 			break;
1492 	}
1493 
1494 	if (nbufs > 0) {
1495 		virtqueue_notify(vq);
1496 		/*
1497 		 * EMSGSIZE signifies the virtqueue did not have enough
1498 		 * entries available to hold the last mbuf. This is not
1499 		 * an error.
1500 		 */
1501 		if (error == EMSGSIZE)
1502 			error = 0;
1503 	}
1504 
1505 	return (error);
1506 }
1507 
1508 static void
vtnet_rxq_free_mbufs(struct vtnet_rxq * rxq)1509 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1510 {
1511 	struct virtqueue *vq;
1512 	struct mbuf *m;
1513 	int last;
1514 #ifdef DEV_NETMAP
1515 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1516 							rxq->vtnrx_id, NR_RX);
1517 #else  /* !DEV_NETMAP */
1518 	void *kring = NULL;
1519 #endif /* !DEV_NETMAP */
1520 
1521 	vq = rxq->vtnrx_vq;
1522 	last = 0;
1523 
1524 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1525 		if (kring == NULL)
1526 			m_freem(m);
1527 	}
1528 
1529 	KASSERT(virtqueue_empty(vq),
1530 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1531 }
1532 
1533 static struct mbuf *
vtnet_rx_alloc_buf(struct vtnet_softc * sc,int nbufs,struct mbuf ** m_tailp)1534 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1535 {
1536 	struct mbuf *m_head, *m_tail, *m;
1537 	int i, size;
1538 
1539 	m_head = NULL;
1540 	size = sc->vtnet_rx_clustersz;
1541 
1542 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1543 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1544 
1545 	for (i = 0; i < nbufs; i++) {
1546 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1547 		if (m == NULL) {
1548 			sc->vtnet_stats.mbuf_alloc_failed++;
1549 			m_freem(m_head);
1550 			return (NULL);
1551 		}
1552 
1553 		m->m_len = size;
1554 		/*
1555 		 * Need to offset the mbuf if the header we're going to add
1556 		 * will misalign.
1557 		 */
1558 		if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1559 			m_adj(m, VTNET_ETHER_ALIGN);
1560 		}
1561 		if (m_head != NULL) {
1562 			m_tail->m_next = m;
1563 			m_tail = m;
1564 		} else
1565 			m_head = m_tail = m;
1566 	}
1567 
1568 	if (m_tailp != NULL)
1569 		*m_tailp = m_tail;
1570 
1571 	return (m_head);
1572 }
1573 
1574 /*
1575  * Slow path for when LRO without mergeable buffers is negotiated.
1576  */
1577 static int
vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq * rxq,struct mbuf * m0,int len0)1578 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1579     int len0)
1580 {
1581 	struct vtnet_softc *sc;
1582 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1583 	int len, clustersz, nreplace, error;
1584 
1585 	sc = rxq->vtnrx_sc;
1586 	clustersz = sc->vtnet_rx_clustersz;
1587 	/*
1588 	 * Need to offset the mbuf if the header we're going to add will
1589 	 * misalign, account for that here.
1590 	 */
1591 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1592 		clustersz -= VTNET_ETHER_ALIGN;
1593 
1594 	m_prev = NULL;
1595 	m_tail = NULL;
1596 	nreplace = 0;
1597 
1598 	m = m0;
1599 	len = len0;
1600 
1601 	/*
1602 	 * Since these mbuf chains are so large, avoid allocating a complete
1603 	 * replacement when the received frame did not consume the entire
1604 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1605 	 */
1606 	while (len > 0) {
1607 		if (m == NULL) {
1608 			sc->vtnet_stats.rx_frame_too_large++;
1609 			return (EMSGSIZE);
1610 		}
1611 
1612 		/*
1613 		 * Every mbuf should have the expected cluster size since that
1614 		 * is also used to allocate the replacements.
1615 		 */
1616 		KASSERT(m->m_len == clustersz,
1617 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1618 		    m->m_len, clustersz));
1619 
1620 		m->m_len = MIN(m->m_len, len);
1621 		len -= m->m_len;
1622 
1623 		m_prev = m;
1624 		m = m->m_next;
1625 		nreplace++;
1626 	}
1627 
1628 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1629 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1630 	    nreplace, sc->vtnet_rx_nmbufs));
1631 
1632 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1633 	if (m_new == NULL) {
1634 		m_prev->m_len = clustersz;
1635 		return (ENOBUFS);
1636 	}
1637 
1638 	/*
1639 	 * Move any unused mbufs from the received mbuf chain onto the
1640 	 * end of the replacement chain.
1641 	 */
1642 	if (m_prev->m_next != NULL) {
1643 		m_tail->m_next = m_prev->m_next;
1644 		m_prev->m_next = NULL;
1645 	}
1646 
1647 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1648 	if (error) {
1649 		/*
1650 		 * The replacement is suppose to be an copy of the one
1651 		 * dequeued so this is a very unexpected error.
1652 		 *
1653 		 * Restore the m0 chain to the original state if it was
1654 		 * modified so we can then discard it.
1655 		 */
1656 		if (m_tail->m_next != NULL) {
1657 			m_prev->m_next = m_tail->m_next;
1658 			m_tail->m_next = NULL;
1659 		}
1660 		m_prev->m_len = clustersz;
1661 		sc->vtnet_stats.rx_enq_replacement_failed++;
1662 		m_freem(m_new);
1663 	}
1664 
1665 	return (error);
1666 }
1667 
1668 static int
vtnet_rxq_replace_buf(struct vtnet_rxq * rxq,struct mbuf * m,int len)1669 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1670 {
1671 	struct vtnet_softc *sc;
1672 	struct mbuf *m_new;
1673 	int error;
1674 
1675 	sc = rxq->vtnrx_sc;
1676 
1677 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1678 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1679 
1680 	MPASS(m->m_next == NULL);
1681 	if (m->m_len < len)
1682 		return (EMSGSIZE);
1683 
1684 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1685 	if (m_new == NULL)
1686 		return (ENOBUFS);
1687 
1688 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1689 	if (error) {
1690 		sc->vtnet_stats.rx_enq_replacement_failed++;
1691 		m_freem(m_new);
1692 	} else
1693 		m->m_len = len;
1694 
1695 	return (error);
1696 }
1697 
1698 static int
vtnet_rxq_enqueue_buf(struct vtnet_rxq * rxq,struct mbuf * m)1699 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1700 {
1701 	struct vtnet_softc *sc;
1702 	struct sglist *sg;
1703 	int header_inlined, error;
1704 
1705 	sc = rxq->vtnrx_sc;
1706 	sg = rxq->vtnrx_sg;
1707 
1708 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1709 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1710 	VTNET_RXQ_LOCK_ASSERT(rxq);
1711 
1712 	sglist_reset(sg);
1713 	header_inlined = vtnet_modern(sc) ||
1714 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1715 
1716 	/*
1717 	 * Note: The mbuf has been already adjusted when we allocate it if we
1718 	 * have to do strict alignment.
1719 	 */
1720 	if (header_inlined)
1721 		error = sglist_append_mbuf(sg, m);
1722 	else {
1723 		struct vtnet_rx_header *rxhdr =
1724 		    mtod(m, struct vtnet_rx_header *);
1725 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1726 
1727 		/* Append the header and remaining mbuf data. */
1728 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1729 		if (error)
1730 			return (error);
1731 		error = sglist_append(sg, &rxhdr[1],
1732 		    m->m_len - sizeof(struct vtnet_rx_header));
1733 		if (error)
1734 			return (error);
1735 
1736 		if (m->m_next != NULL)
1737 			error = sglist_append_mbuf(sg, m->m_next);
1738 	}
1739 
1740 	if (error)
1741 		return (error);
1742 
1743 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1744 }
1745 
1746 static int
vtnet_rxq_new_buf(struct vtnet_rxq * rxq)1747 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1748 {
1749 	struct vtnet_softc *sc;
1750 	struct mbuf *m;
1751 	int error;
1752 
1753 	sc = rxq->vtnrx_sc;
1754 
1755 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1756 	if (m == NULL)
1757 		return (ENOBUFS);
1758 
1759 	error = vtnet_rxq_enqueue_buf(rxq, m);
1760 	if (error)
1761 		m_freem(m);
1762 
1763 	return (error);
1764 }
1765 
1766 #if defined(INET) || defined(INET6)
1767 static int
vtnet_rxq_csum_needs_csum(struct vtnet_rxq * rxq,struct mbuf * m,bool isipv6,int protocol,struct virtio_net_hdr * hdr)1768 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, bool isipv6,
1769     int protocol, struct virtio_net_hdr *hdr)
1770 {
1771 	struct vtnet_softc *sc;
1772 
1773 	/*
1774 	 * The packet is likely from another VM on the same host or from the
1775 	 * host that itself performed checksum offloading so Tx/Rx is basically
1776 	 * a memcpy and the checksum has little value so far.
1777 	 */
1778 
1779 	KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP,
1780 	    ("%s: unsupported IP protocol %d", __func__, protocol));
1781 
1782 	/*
1783 	 * If the user don't want us to fix it up here by computing the
1784 	 * checksum, just forward the order to compute the checksum by setting
1785 	 * the corresponding mbuf flag (e.g., CSUM_TCP).
1786 	 */
1787 	sc = rxq->vtnrx_sc;
1788 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1789 		switch (protocol) {
1790 		case IPPROTO_TCP:
1791 			m->m_pkthdr.csum_flags |=
1792 			    (isipv6 ? CSUM_TCP_IPV6 : CSUM_TCP);
1793 			break;
1794 		case IPPROTO_UDP:
1795 			m->m_pkthdr.csum_flags |=
1796 			    (isipv6 ? CSUM_UDP_IPV6 : CSUM_UDP);
1797 			break;
1798 		}
1799 		m->m_pkthdr.csum_data = hdr->csum_offset;
1800 		return (0);
1801 	}
1802 
1803 	/*
1804 	 * Compute the checksum in the driver so the packet will contain a
1805 	 * valid checksum. The checksum is at csum_offset from csum_start.
1806 	 */
1807 	int csum_off, csum_end;
1808 	uint16_t csum;
1809 
1810 	csum_off = hdr->csum_start + hdr->csum_offset;
1811 	csum_end = csum_off + sizeof(uint16_t);
1812 
1813 	/* Assume checksum will be in the first mbuf. */
1814 	if (m->m_len < csum_end || m->m_pkthdr.len < csum_end) {
1815 		sc->vtnet_stats.rx_csum_bad_offset++;
1816 		return (1);
1817 	}
1818 
1819 	/*
1820 	 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1821 	 * checksum and write it at the specified offset. We could
1822 	 * try to verify the packet: csum_start should probably
1823 	 * correspond to the start of the TCP/UDP header.
1824 	 *
1825 	 * BMV: Need to properly handle UDP with zero checksum. Is
1826 	 * the IPv4 header checksum implicitly validated?
1827 	 */
1828 	csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1829 	*(uint16_t *)(mtodo(m, csum_off)) = csum;
1830 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1831 	m->m_pkthdr.csum_data = 0xFFFF;
1832 
1833 	return (0);
1834 }
1835 
1836 static void
vtnet_rxq_csum_data_valid(struct vtnet_rxq * rxq,struct mbuf * m,int protocol)1837 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m, int protocol)
1838 {
1839 	KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP,
1840 	    ("%s: unsupported IP protocol %d", __func__, protocol));
1841 
1842 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1843 	m->m_pkthdr.csum_data = 0xFFFF;
1844 }
1845 
1846 static int
vtnet_rxq_csum(struct vtnet_rxq * rxq,struct mbuf * m,struct virtio_net_hdr * hdr)1847 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1848     struct virtio_net_hdr *hdr)
1849 {
1850 	const struct ether_header *eh;
1851 	struct vtnet_softc *sc;
1852 	int hoff, protocol;
1853 	uint16_t etype;
1854 	bool isipv6;
1855 
1856 	KASSERT(hdr->flags &
1857 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID),
1858 	    ("%s: missing checksum offloading flag %x", __func__, hdr->flags));
1859 
1860 	eh = mtod(m, const struct ether_header *);
1861 	etype = ntohs(eh->ether_type);
1862 	if (etype == ETHERTYPE_VLAN) {
1863 		/* TODO BMV: Handle QinQ. */
1864 		const struct ether_vlan_header *evh =
1865 		    mtod(m, const struct ether_vlan_header *);
1866 		etype = ntohs(evh->evl_proto);
1867 		hoff = sizeof(struct ether_vlan_header);
1868 	} else
1869 		hoff = sizeof(struct ether_header);
1870 
1871 	sc = rxq->vtnrx_sc;
1872 
1873 	/* Check whether ethernet type is IP or IPv6, and get protocol. */
1874 	switch (etype) {
1875 #if defined(INET)
1876 	case ETHERTYPE_IP:
1877 		if (__predict_false(m->m_len < hoff + sizeof(struct ip))) {
1878 			sc->vtnet_stats.rx_csum_inaccessible_ipproto++;
1879 			return (1);
1880 		} else {
1881 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1882 			protocol = ip->ip_p;
1883 		}
1884 		isipv6 = false;
1885 		break;
1886 #endif
1887 #if defined(INET6)
1888 	case ETHERTYPE_IPV6:
1889 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1890 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0) {
1891 			sc->vtnet_stats.rx_csum_inaccessible_ipproto++;
1892 			return (1);
1893 		}
1894 		isipv6 = true;
1895 		break;
1896 #endif
1897 	default:
1898 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1899 		return (1);
1900 	}
1901 
1902 	/* Check whether protocol is TCP or UDP. */
1903 	switch (protocol) {
1904 	case IPPROTO_TCP:
1905 	case IPPROTO_UDP:
1906 		break;
1907 	default:
1908 		/*
1909 		 * FreeBSD does not support checksum offloading of this
1910 		 * protocol here.
1911 		 */
1912 		sc->vtnet_stats.rx_csum_bad_ipproto++;
1913 		return (1);
1914 	}
1915 
1916 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1917 		return (vtnet_rxq_csum_needs_csum(rxq, m, isipv6, protocol,
1918 		    hdr));
1919 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1920 		vtnet_rxq_csum_data_valid(rxq, m, protocol);
1921 
1922 	return (0);
1923 }
1924 #endif
1925 
1926 static void
vtnet_rxq_discard_merged_bufs(struct vtnet_rxq * rxq,int nbufs)1927 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1928 {
1929 	struct mbuf *m;
1930 
1931 	while (--nbufs > 0) {
1932 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1933 		if (m == NULL)
1934 			break;
1935 		vtnet_rxq_discard_buf(rxq, m);
1936 	}
1937 }
1938 
1939 static void
vtnet_rxq_discard_buf(struct vtnet_rxq * rxq,struct mbuf * m)1940 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1941 {
1942 	int error __diagused;
1943 
1944 	/*
1945 	 * Requeue the discarded mbuf. This should always be successful
1946 	 * since it was just dequeued.
1947 	 */
1948 	error = vtnet_rxq_enqueue_buf(rxq, m);
1949 	KASSERT(error == 0,
1950 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1951 }
1952 
1953 static int
vtnet_rxq_merged_eof(struct vtnet_rxq * rxq,struct mbuf * m_head,int nbufs)1954 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1955 {
1956 	struct vtnet_softc *sc;
1957 	struct virtqueue *vq;
1958 	struct mbuf *m_tail;
1959 
1960 	sc = rxq->vtnrx_sc;
1961 	vq = rxq->vtnrx_vq;
1962 	m_tail = m_head;
1963 
1964 	while (--nbufs > 0) {
1965 		struct mbuf *m;
1966 		uint32_t len;
1967 
1968 		m = virtqueue_dequeue(vq, &len);
1969 		if (m == NULL) {
1970 			rxq->vtnrx_stats.vrxs_ierrors++;
1971 			goto fail;
1972 		}
1973 
1974 		if (vtnet_rxq_new_buf(rxq) != 0) {
1975 			rxq->vtnrx_stats.vrxs_iqdrops++;
1976 			vtnet_rxq_discard_buf(rxq, m);
1977 			if (nbufs > 1)
1978 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1979 			goto fail;
1980 		}
1981 
1982 		if (m->m_len < len)
1983 			len = m->m_len;
1984 
1985 		m->m_len = len;
1986 		m->m_flags &= ~M_PKTHDR;
1987 
1988 		m_head->m_pkthdr.len += len;
1989 		m_tail->m_next = m;
1990 		m_tail = m;
1991 	}
1992 
1993 	return (0);
1994 
1995 fail:
1996 	sc->vtnet_stats.rx_mergeable_failed++;
1997 	m_freem(m_head);
1998 
1999 	return (1);
2000 }
2001 
2002 #if defined(INET) || defined(INET6)
2003 static int
vtnet_lro_rx(struct vtnet_rxq * rxq,struct mbuf * m)2004 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
2005 {
2006 	struct lro_ctrl *lro;
2007 
2008 	lro = &rxq->vtnrx_lro;
2009 
2010 	if (lro->lro_mbuf_max != 0) {
2011 		tcp_lro_queue_mbuf(lro, m);
2012 		return (0);
2013 	}
2014 
2015 	return (tcp_lro_rx(lro, m, 0));
2016 }
2017 #endif
2018 
2019 static void
vtnet_rxq_input(struct vtnet_rxq * rxq,struct mbuf * m,struct virtio_net_hdr * hdr)2020 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2021     struct virtio_net_hdr *hdr)
2022 {
2023 	struct vtnet_softc *sc;
2024 	if_t ifp;
2025 
2026 	sc = rxq->vtnrx_sc;
2027 	ifp = sc->vtnet_ifp;
2028 
2029 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2030 		struct ether_header *eh = mtod(m, struct ether_header *);
2031 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2032 			vtnet_vlan_tag_remove(m);
2033 			/*
2034 			 * With the 802.1Q header removed, update the
2035 			 * checksum starting location accordingly.
2036 			 */
2037 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2038 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2039 		}
2040 	}
2041 
2042 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2043 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2044 
2045 	if (hdr->flags &
2046 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2047 #if defined(INET) || defined(INET6)
2048 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2049 			rxq->vtnrx_stats.vrxs_csum++;
2050 		else
2051 			rxq->vtnrx_stats.vrxs_csum_failed++;
2052 #else
2053 		sc->vtnet_stats.rx_csum_bad_ethtype++;
2054 		rxq->vtnrx_stats.vrxs_csum_failed++;
2055 #endif
2056 	}
2057 
2058 	if (hdr->gso_size != 0) {
2059 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2060 		case VIRTIO_NET_HDR_GSO_TCPV4:
2061 		case VIRTIO_NET_HDR_GSO_TCPV6:
2062 			m->m_pkthdr.lro_nsegs =
2063 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2064 			rxq->vtnrx_stats.vrxs_host_lro++;
2065 			break;
2066 		}
2067 	}
2068 
2069 	rxq->vtnrx_stats.vrxs_ipackets++;
2070 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2071 
2072 #if defined(INET) || defined(INET6)
2073 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2074 		if (vtnet_lro_rx(rxq, m) == 0)
2075 			return;
2076 	}
2077 #endif
2078 
2079 	if_input(ifp, m);
2080 }
2081 
2082 static int
vtnet_rxq_eof(struct vtnet_rxq * rxq)2083 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2084 {
2085 	struct virtio_net_hdr lhdr, *hdr;
2086 	struct vtnet_softc *sc;
2087 	if_t ifp;
2088 	struct virtqueue *vq;
2089 	int deq, count;
2090 
2091 	sc = rxq->vtnrx_sc;
2092 	vq = rxq->vtnrx_vq;
2093 	ifp = sc->vtnet_ifp;
2094 	deq = 0;
2095 	count = sc->vtnet_rx_process_limit;
2096 
2097 	VTNET_RXQ_LOCK_ASSERT(rxq);
2098 
2099 	CURVNET_SET(if_getvnet(ifp));
2100 	while (count-- > 0) {
2101 		struct mbuf *m;
2102 		uint32_t len, nbufs, adjsz;
2103 
2104 		m = virtqueue_dequeue(vq, &len);
2105 		if (m == NULL)
2106 			break;
2107 		deq++;
2108 
2109 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2110 			rxq->vtnrx_stats.vrxs_ierrors++;
2111 			vtnet_rxq_discard_buf(rxq, m);
2112 			continue;
2113 		}
2114 
2115 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2116 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2117 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2118 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2119 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2120 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2121 		} else if (vtnet_modern(sc)) {
2122 			nbufs = 1; /* num_buffers is always 1 */
2123 			adjsz = sizeof(struct virtio_net_hdr_v1);
2124 		} else {
2125 			nbufs = 1;
2126 			adjsz = sizeof(struct vtnet_rx_header);
2127 			/*
2128 			 * Account for our gap between the header and start of
2129 			 * data to keep the segments separated.
2130 			 */
2131 			len += VTNET_RX_HEADER_PAD;
2132 		}
2133 
2134 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2135 			rxq->vtnrx_stats.vrxs_iqdrops++;
2136 			vtnet_rxq_discard_buf(rxq, m);
2137 			if (nbufs > 1)
2138 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2139 			continue;
2140 		}
2141 
2142 		m->m_pkthdr.len = len;
2143 		m->m_pkthdr.rcvif = ifp;
2144 		m->m_pkthdr.csum_flags = 0;
2145 
2146 		if (nbufs > 1) {
2147 			/* Dequeue the rest of chain. */
2148 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2149 				continue;
2150 		}
2151 
2152 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2153 
2154 		/*
2155 		 * Save an endian swapped version of the header prior to it
2156 		 * being stripped. The header is always at the start of the
2157 		 * mbuf data. num_buffers was already saved (and not needed)
2158 		 * so use the standard header.
2159 		 */
2160 		hdr = mtod(m, struct virtio_net_hdr *);
2161 		lhdr.flags = hdr->flags;
2162 		lhdr.gso_type = hdr->gso_type;
2163 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2164 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2165 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2166 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2167 		m_adj(m, adjsz);
2168 
2169 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2170 			pfil_return_t pfil;
2171 
2172 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2173 			switch (pfil) {
2174 			case PFIL_DROPPED:
2175 			case PFIL_CONSUMED:
2176 				continue;
2177 			default:
2178 				KASSERT(pfil == PFIL_PASS,
2179 				    ("Filter returned %d!", pfil));
2180 			}
2181 		}
2182 
2183 		vtnet_rxq_input(rxq, m, &lhdr);
2184 	}
2185 
2186 	if (deq > 0) {
2187 #if defined(INET) || defined(INET6)
2188 		if (vtnet_software_lro(sc))
2189 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2190 #endif
2191 		virtqueue_notify(vq);
2192 	}
2193 	CURVNET_RESTORE();
2194 
2195 	return (count > 0 ? 0 : EAGAIN);
2196 }
2197 
2198 static void
vtnet_rx_vq_process(struct vtnet_rxq * rxq,int tries)2199 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2200 {
2201 	struct vtnet_softc *sc;
2202 	if_t ifp;
2203 	u_int more;
2204 #ifdef DEV_NETMAP
2205 	int nmirq;
2206 #endif /* DEV_NETMAP */
2207 
2208 	sc = rxq->vtnrx_sc;
2209 	ifp = sc->vtnet_ifp;
2210 
2211 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2212 		/*
2213 		 * Ignore this interrupt. Either this is a spurious interrupt
2214 		 * or multiqueue without per-VQ MSIX so every queue needs to
2215 		 * be polled (a brain dead configuration we could try harder
2216 		 * to avoid).
2217 		 */
2218 		vtnet_rxq_disable_intr(rxq);
2219 		return;
2220 	}
2221 
2222 	VTNET_RXQ_LOCK(rxq);
2223 
2224 #ifdef DEV_NETMAP
2225 	/*
2226 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2227 	 * This is not necessary to serialize the access to the RX vq, but
2228 	 * rather to avoid races that may happen if this interface is
2229 	 * attached to a VALE switch, which would cause received packets
2230 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2231 	 * busy when called from netmap_bwrap_intr_notify()).
2232 	 */
2233 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2234 	if (nmirq != NM_IRQ_PASS) {
2235 		VTNET_RXQ_UNLOCK(rxq);
2236 		if (nmirq == NM_IRQ_RESCHED) {
2237 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2238 		}
2239 		return;
2240 	}
2241 #endif /* DEV_NETMAP */
2242 
2243 again:
2244 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2245 		VTNET_RXQ_UNLOCK(rxq);
2246 		return;
2247 	}
2248 
2249 	more = vtnet_rxq_eof(rxq);
2250 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2251 		if (!more)
2252 			vtnet_rxq_disable_intr(rxq);
2253 		/*
2254 		 * This is an occasional condition or race (when !more),
2255 		 * so retry a few times before scheduling the taskqueue.
2256 		 */
2257 		if (tries-- > 0)
2258 			goto again;
2259 
2260 		rxq->vtnrx_stats.vrxs_rescheduled++;
2261 		VTNET_RXQ_UNLOCK(rxq);
2262 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2263 	} else
2264 		VTNET_RXQ_UNLOCK(rxq);
2265 }
2266 
2267 static void
vtnet_rx_vq_intr(void * xrxq)2268 vtnet_rx_vq_intr(void *xrxq)
2269 {
2270 	struct vtnet_rxq *rxq;
2271 
2272 	rxq = xrxq;
2273 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2274 }
2275 
2276 static void
vtnet_rxq_tq_intr(void * xrxq,int pending __unused)2277 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2278 {
2279 	struct vtnet_rxq *rxq;
2280 
2281 	rxq = xrxq;
2282 	vtnet_rx_vq_process(rxq, 0);
2283 }
2284 
2285 static int
vtnet_txq_intr_threshold(struct vtnet_txq * txq)2286 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2287 {
2288 	struct vtnet_softc *sc;
2289 	int threshold;
2290 
2291 	sc = txq->vtntx_sc;
2292 
2293 	/*
2294 	 * The Tx interrupt is disabled until the queue free count falls
2295 	 * below our threshold. Completed frames are drained from the Tx
2296 	 * virtqueue before transmitting new frames and in the watchdog
2297 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2298 	 * at the cost of not freeing mbufs as quickly as they otherwise
2299 	 * would be.
2300 	 */
2301 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2302 
2303 	/*
2304 	 * Without indirect descriptors, leave enough room for the most
2305 	 * segments we handle.
2306 	 */
2307 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2308 	    threshold < sc->vtnet_tx_nsegs)
2309 		threshold = sc->vtnet_tx_nsegs;
2310 
2311 	return (threshold);
2312 }
2313 
2314 static int
vtnet_txq_below_threshold(struct vtnet_txq * txq)2315 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2316 {
2317 	struct virtqueue *vq;
2318 
2319 	vq = txq->vtntx_vq;
2320 
2321 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2322 }
2323 
2324 static int
vtnet_txq_notify(struct vtnet_txq * txq)2325 vtnet_txq_notify(struct vtnet_txq *txq)
2326 {
2327 	struct virtqueue *vq;
2328 
2329 	vq = txq->vtntx_vq;
2330 
2331 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2332 	virtqueue_notify(vq);
2333 
2334 	if (vtnet_txq_enable_intr(txq) == 0)
2335 		return (0);
2336 
2337 	/*
2338 	 * Drain frames that were completed since last checked. If this
2339 	 * causes the queue to go above the threshold, the caller should
2340 	 * continue transmitting.
2341 	 */
2342 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2343 		virtqueue_disable_intr(vq);
2344 		return (1);
2345 	}
2346 
2347 	return (0);
2348 }
2349 
2350 static void
vtnet_txq_free_mbufs(struct vtnet_txq * txq)2351 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2352 {
2353 	struct virtqueue *vq;
2354 	struct vtnet_tx_header *txhdr;
2355 	int last;
2356 #ifdef DEV_NETMAP
2357 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2358 							txq->vtntx_id, NR_TX);
2359 #else  /* !DEV_NETMAP */
2360 	void *kring = NULL;
2361 #endif /* !DEV_NETMAP */
2362 
2363 	vq = txq->vtntx_vq;
2364 	last = 0;
2365 
2366 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2367 		if (kring == NULL) {
2368 			m_freem(txhdr->vth_mbuf);
2369 			uma_zfree(vtnet_tx_header_zone, txhdr);
2370 		}
2371 	}
2372 
2373 	KASSERT(virtqueue_empty(vq),
2374 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2375 }
2376 
2377 /*
2378  * BMV: This can go away once we finally have offsets in the mbuf header.
2379  */
2380 static int
vtnet_txq_offload_ctx(struct vtnet_txq * txq,struct mbuf * m,int * etype,int * proto,int * start)2381 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2382     int *proto, int *start)
2383 {
2384 	struct vtnet_softc *sc;
2385 	struct ether_vlan_header *evh;
2386 #if defined(INET) || defined(INET6)
2387 	int offset;
2388 #endif
2389 
2390 	sc = txq->vtntx_sc;
2391 
2392 	evh = mtod(m, struct ether_vlan_header *);
2393 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2394 		/* BMV: We should handle nested VLAN tags too. */
2395 		*etype = ntohs(evh->evl_proto);
2396 #if defined(INET) || defined(INET6)
2397 		offset = sizeof(struct ether_vlan_header);
2398 #endif
2399 	} else {
2400 		*etype = ntohs(evh->evl_encap_proto);
2401 #if defined(INET) || defined(INET6)
2402 		offset = sizeof(struct ether_header);
2403 #endif
2404 	}
2405 
2406 	switch (*etype) {
2407 #if defined(INET)
2408 	case ETHERTYPE_IP: {
2409 		struct ip *ip, iphdr;
2410 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2411 			m_copydata(m, offset, sizeof(struct ip),
2412 			    (caddr_t) &iphdr);
2413 			ip = &iphdr;
2414 		} else
2415 			ip = (struct ip *)(m->m_data + offset);
2416 		*proto = ip->ip_p;
2417 		*start = offset + (ip->ip_hl << 2);
2418 		break;
2419 	}
2420 #endif
2421 #if defined(INET6)
2422 	case ETHERTYPE_IPV6:
2423 		*proto = -1;
2424 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2425 		/* Assert the network stack sent us a valid packet. */
2426 		KASSERT(*start > offset,
2427 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2428 		    *start, offset, *proto));
2429 		break;
2430 #endif
2431 	default:
2432 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2433 		return (EINVAL);
2434 	}
2435 
2436 	return (0);
2437 }
2438 
2439 static int
vtnet_txq_offload_tso(struct vtnet_txq * txq,struct mbuf * m,int eth_type,int offset,struct virtio_net_hdr * hdr)2440 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2441     int offset, struct virtio_net_hdr *hdr)
2442 {
2443 	static struct timeval lastecn;
2444 	static int curecn;
2445 	struct vtnet_softc *sc;
2446 	struct tcphdr *tcp, tcphdr;
2447 
2448 	sc = txq->vtntx_sc;
2449 
2450 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2451 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2452 		tcp = &tcphdr;
2453 	} else
2454 		tcp = (struct tcphdr *)(m->m_data + offset);
2455 
2456 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2457 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2458 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2459 	    VIRTIO_NET_HDR_GSO_TCPV6;
2460 
2461 	if (__predict_false(tcp_get_flags(tcp) & TH_CWR)) {
2462 		/*
2463 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2464 		 * FreeBSD, ECN support is not on a per-interface basis,
2465 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2466 		 * knob. The default is off.
2467 		 */
2468 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2469 			if (ppsratecheck(&lastecn, &curecn, 1))
2470 				if_printf(sc->vtnet_ifp,
2471 				    "TSO with ECN not negotiated with host\n");
2472 			return (ENOTSUP);
2473 		}
2474 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2475 	}
2476 
2477 	txq->vtntx_stats.vtxs_tso++;
2478 
2479 	return (0);
2480 }
2481 
2482 static struct mbuf *
vtnet_txq_offload(struct vtnet_txq * txq,struct mbuf * m,struct virtio_net_hdr * hdr)2483 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2484     struct virtio_net_hdr *hdr)
2485 {
2486 	struct vtnet_softc *sc;
2487 	int flags, etype, csum_start, proto, error;
2488 
2489 	sc = txq->vtntx_sc;
2490 	flags = m->m_pkthdr.csum_flags;
2491 
2492 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2493 	if (error)
2494 		goto drop;
2495 
2496 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2497 		/* Sanity check the parsed mbuf matches the offload flags. */
2498 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2499 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2500 		    && etype != ETHERTYPE_IPV6))) {
2501 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2502 			goto drop;
2503 		}
2504 
2505 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2506 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2507 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2508 		txq->vtntx_stats.vtxs_csum++;
2509 	} else if ((flags & (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) &&
2510 	           (proto == IPPROTO_TCP || proto == IPPROTO_UDP) &&
2511 	           (m->m_pkthdr.csum_data == 0xFFFF)) {
2512 		hdr->flags |= VIRTIO_NET_HDR_F_DATA_VALID;
2513 	}
2514 
2515 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2516 		/*
2517 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2518 		 * VirtIO TSO reqires the checksum offloading above.
2519 		 */
2520 		if (__predict_false(proto != IPPROTO_TCP)) {
2521 			sc->vtnet_stats.tx_tso_not_tcp++;
2522 			goto drop;
2523 		} else if (__predict_false((hdr->flags &
2524 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2525 			sc->vtnet_stats.tx_tso_without_csum++;
2526 			goto drop;
2527 		}
2528 
2529 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2530 		if (error)
2531 			goto drop;
2532 	}
2533 
2534 	return (m);
2535 
2536 drop:
2537 	m_freem(m);
2538 	return (NULL);
2539 }
2540 
2541 static int
vtnet_txq_enqueue_buf(struct vtnet_txq * txq,struct mbuf ** m_head,struct vtnet_tx_header * txhdr)2542 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2543     struct vtnet_tx_header *txhdr)
2544 {
2545 	struct vtnet_softc *sc;
2546 	struct virtqueue *vq;
2547 	struct sglist *sg;
2548 	struct mbuf *m;
2549 	int error;
2550 
2551 	sc = txq->vtntx_sc;
2552 	vq = txq->vtntx_vq;
2553 	sg = txq->vtntx_sg;
2554 	m = *m_head;
2555 
2556 	sglist_reset(sg);
2557 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2558 	if (error != 0 || sg->sg_nseg != 1) {
2559 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2560 		    __func__, error, sg->sg_nseg));
2561 		goto fail;
2562 	}
2563 
2564 	error = sglist_append_mbuf(sg, m);
2565 	if (error) {
2566 		m = m_defrag(m, M_NOWAIT);
2567 		if (m == NULL) {
2568 			sc->vtnet_stats.tx_defrag_failed++;
2569 			goto fail;
2570 		}
2571 
2572 		*m_head = m;
2573 		sc->vtnet_stats.tx_defragged++;
2574 
2575 		error = sglist_append_mbuf(sg, m);
2576 		if (error)
2577 			goto fail;
2578 	}
2579 
2580 	txhdr->vth_mbuf = m;
2581 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2582 
2583 	return (error);
2584 
2585 fail:
2586 	m_freem(*m_head);
2587 	*m_head = NULL;
2588 
2589 	return (ENOBUFS);
2590 }
2591 
2592 static int
vtnet_txq_encap(struct vtnet_txq * txq,struct mbuf ** m_head,int flags)2593 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2594 {
2595 	struct vtnet_tx_header *txhdr;
2596 	struct virtio_net_hdr *hdr;
2597 	struct mbuf *m;
2598 	int error;
2599 
2600 	m = *m_head;
2601 	M_ASSERTPKTHDR(m);
2602 
2603 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2604 	if (txhdr == NULL) {
2605 		m_freem(m);
2606 		*m_head = NULL;
2607 		return (ENOMEM);
2608 	}
2609 
2610 	/*
2611 	 * Always use the non-mergeable header, regardless if mergable headers
2612 	 * were negotiated, because for transmit num_buffers is always zero.
2613 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2614 	 */
2615 	hdr = &txhdr->vth_uhdr.hdr;
2616 
2617 	if (m->m_flags & M_VLANTAG) {
2618 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2619 		if ((*m_head = m) == NULL) {
2620 			error = ENOBUFS;
2621 			goto fail;
2622 		}
2623 		m->m_flags &= ~M_VLANTAG;
2624 	}
2625 
2626 	if (m->m_pkthdr.csum_flags &
2627 	    (VTNET_CSUM_ALL_OFFLOAD | CSUM_DATA_VALID)) {
2628 		m = vtnet_txq_offload(txq, m, hdr);
2629 		if ((*m_head = m) == NULL) {
2630 			error = ENOBUFS;
2631 			goto fail;
2632 		}
2633 	}
2634 
2635 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2636 fail:
2637 	if (error)
2638 		uma_zfree(vtnet_tx_header_zone, txhdr);
2639 
2640 	return (error);
2641 }
2642 
2643 
2644 static void
vtnet_start_locked(struct vtnet_txq * txq,if_t ifp)2645 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2646 {
2647 	struct vtnet_softc *sc;
2648 	struct virtqueue *vq;
2649 	struct mbuf *m0;
2650 	int tries, enq;
2651 
2652 	sc = txq->vtntx_sc;
2653 	vq = txq->vtntx_vq;
2654 	tries = 0;
2655 
2656 	VTNET_TXQ_LOCK_ASSERT(txq);
2657 
2658 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2659 	    sc->vtnet_link_active == 0)
2660 		return;
2661 
2662 	vtnet_txq_eof(txq);
2663 
2664 again:
2665 	enq = 0;
2666 
2667 	while (!if_sendq_empty(ifp)) {
2668 		if (virtqueue_full(vq))
2669 			break;
2670 
2671 		m0 = if_dequeue(ifp);
2672 		if (m0 == NULL)
2673 			break;
2674 
2675 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2676 			if (m0 != NULL)
2677 				if_sendq_prepend(ifp, m0);
2678 			break;
2679 		}
2680 
2681 		enq++;
2682 		ETHER_BPF_MTAP(ifp, m0);
2683 	}
2684 
2685 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2686 		if (tries++ < VTNET_NOTIFY_RETRIES)
2687 			goto again;
2688 
2689 		txq->vtntx_stats.vtxs_rescheduled++;
2690 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2691 	}
2692 }
2693 
2694 static void
vtnet_start(if_t ifp)2695 vtnet_start(if_t ifp)
2696 {
2697 	struct vtnet_softc *sc;
2698 	struct vtnet_txq *txq;
2699 
2700 	sc = if_getsoftc(ifp);
2701 	txq = &sc->vtnet_txqs[0];
2702 
2703 	VTNET_TXQ_LOCK(txq);
2704 	vtnet_start_locked(txq, ifp);
2705 	VTNET_TXQ_UNLOCK(txq);
2706 }
2707 
2708 
2709 static int
vtnet_txq_mq_start_locked(struct vtnet_txq * txq,struct mbuf * m)2710 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2711 {
2712 	struct vtnet_softc *sc;
2713 	struct virtqueue *vq;
2714 	struct buf_ring *br;
2715 	if_t ifp;
2716 	int enq, tries, error;
2717 
2718 	sc = txq->vtntx_sc;
2719 	vq = txq->vtntx_vq;
2720 	br = txq->vtntx_br;
2721 	ifp = sc->vtnet_ifp;
2722 	tries = 0;
2723 	error = 0;
2724 
2725 	VTNET_TXQ_LOCK_ASSERT(txq);
2726 
2727 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2728 	    sc->vtnet_link_active == 0) {
2729 		if (m != NULL)
2730 			error = drbr_enqueue(ifp, br, m);
2731 		return (error);
2732 	}
2733 
2734 	if (m != NULL) {
2735 		error = drbr_enqueue(ifp, br, m);
2736 		if (error)
2737 			return (error);
2738 	}
2739 
2740 	vtnet_txq_eof(txq);
2741 
2742 again:
2743 	enq = 0;
2744 
2745 	while ((m = drbr_peek(ifp, br)) != NULL) {
2746 		if (virtqueue_full(vq)) {
2747 			drbr_putback(ifp, br, m);
2748 			break;
2749 		}
2750 
2751 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2752 			if (m != NULL)
2753 				drbr_putback(ifp, br, m);
2754 			else
2755 				drbr_advance(ifp, br);
2756 			break;
2757 		}
2758 		drbr_advance(ifp, br);
2759 
2760 		enq++;
2761 		ETHER_BPF_MTAP(ifp, m);
2762 	}
2763 
2764 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2765 		if (tries++ < VTNET_NOTIFY_RETRIES)
2766 			goto again;
2767 
2768 		txq->vtntx_stats.vtxs_rescheduled++;
2769 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2770 	}
2771 
2772 	return (0);
2773 }
2774 
2775 static int
vtnet_txq_mq_start(if_t ifp,struct mbuf * m)2776 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2777 {
2778 	struct vtnet_softc *sc;
2779 	struct vtnet_txq *txq;
2780 	int i, npairs, error;
2781 
2782 	sc = if_getsoftc(ifp);
2783 	npairs = sc->vtnet_act_vq_pairs;
2784 
2785 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2786 		i = m->m_pkthdr.flowid % npairs;
2787 	else
2788 		i = curcpu % npairs;
2789 
2790 	txq = &sc->vtnet_txqs[i];
2791 
2792 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2793 		error = vtnet_txq_mq_start_locked(txq, m);
2794 		VTNET_TXQ_UNLOCK(txq);
2795 	} else {
2796 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2797 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2798 	}
2799 
2800 	return (error);
2801 }
2802 
2803 static void
vtnet_txq_tq_deferred(void * xtxq,int pending __unused)2804 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2805 {
2806 	struct vtnet_softc *sc;
2807 	struct vtnet_txq *txq;
2808 
2809 	txq = xtxq;
2810 	sc = txq->vtntx_sc;
2811 
2812 	VTNET_TXQ_LOCK(txq);
2813 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2814 		vtnet_txq_mq_start_locked(txq, NULL);
2815 	VTNET_TXQ_UNLOCK(txq);
2816 }
2817 
2818 
2819 static void
vtnet_txq_start(struct vtnet_txq * txq)2820 vtnet_txq_start(struct vtnet_txq *txq)
2821 {
2822 	struct vtnet_softc *sc;
2823 	if_t ifp;
2824 
2825 	sc = txq->vtntx_sc;
2826 	ifp = sc->vtnet_ifp;
2827 
2828 	if (!VTNET_ALTQ_ENABLED) {
2829 		if (!drbr_empty(ifp, txq->vtntx_br))
2830 			vtnet_txq_mq_start_locked(txq, NULL);
2831 	} else {
2832 		if (!if_sendq_empty(ifp))
2833 			vtnet_start_locked(txq, ifp);
2834 
2835 	}
2836 }
2837 
2838 static void
vtnet_txq_tq_intr(void * xtxq,int pending __unused)2839 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2840 {
2841 	struct vtnet_softc *sc;
2842 	struct vtnet_txq *txq;
2843 	if_t ifp;
2844 
2845 	txq = xtxq;
2846 	sc = txq->vtntx_sc;
2847 	ifp = sc->vtnet_ifp;
2848 
2849 	VTNET_TXQ_LOCK(txq);
2850 
2851 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2852 		VTNET_TXQ_UNLOCK(txq);
2853 		return;
2854 	}
2855 
2856 	vtnet_txq_eof(txq);
2857 	vtnet_txq_start(txq);
2858 
2859 	VTNET_TXQ_UNLOCK(txq);
2860 }
2861 
2862 static int
vtnet_txq_eof(struct vtnet_txq * txq)2863 vtnet_txq_eof(struct vtnet_txq *txq)
2864 {
2865 	struct virtqueue *vq;
2866 	struct vtnet_tx_header *txhdr;
2867 	struct mbuf *m;
2868 	int deq;
2869 
2870 	vq = txq->vtntx_vq;
2871 	deq = 0;
2872 	VTNET_TXQ_LOCK_ASSERT(txq);
2873 
2874 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2875 		m = txhdr->vth_mbuf;
2876 		deq++;
2877 
2878 		txq->vtntx_stats.vtxs_opackets++;
2879 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2880 		if (m->m_flags & M_MCAST)
2881 			txq->vtntx_stats.vtxs_omcasts++;
2882 
2883 		m_freem(m);
2884 		uma_zfree(vtnet_tx_header_zone, txhdr);
2885 	}
2886 
2887 	if (virtqueue_empty(vq))
2888 		txq->vtntx_watchdog = 0;
2889 
2890 	return (deq);
2891 }
2892 
2893 static void
vtnet_tx_vq_intr(void * xtxq)2894 vtnet_tx_vq_intr(void *xtxq)
2895 {
2896 	struct vtnet_softc *sc;
2897 	struct vtnet_txq *txq;
2898 	if_t ifp;
2899 
2900 	txq = xtxq;
2901 	sc = txq->vtntx_sc;
2902 	ifp = sc->vtnet_ifp;
2903 
2904 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2905 		/*
2906 		 * Ignore this interrupt. Either this is a spurious interrupt
2907 		 * or multiqueue without per-VQ MSIX so every queue needs to
2908 		 * be polled (a brain dead configuration we could try harder
2909 		 * to avoid).
2910 		 */
2911 		vtnet_txq_disable_intr(txq);
2912 		return;
2913 	}
2914 
2915 #ifdef DEV_NETMAP
2916 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2917 		return;
2918 #endif /* DEV_NETMAP */
2919 
2920 	VTNET_TXQ_LOCK(txq);
2921 
2922 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2923 		VTNET_TXQ_UNLOCK(txq);
2924 		return;
2925 	}
2926 
2927 	vtnet_txq_eof(txq);
2928 	vtnet_txq_start(txq);
2929 
2930 	VTNET_TXQ_UNLOCK(txq);
2931 }
2932 
2933 static void
vtnet_tx_start_all(struct vtnet_softc * sc)2934 vtnet_tx_start_all(struct vtnet_softc *sc)
2935 {
2936 	struct vtnet_txq *txq;
2937 	int i;
2938 
2939 	VTNET_CORE_LOCK_ASSERT(sc);
2940 
2941 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2942 		txq = &sc->vtnet_txqs[i];
2943 
2944 		VTNET_TXQ_LOCK(txq);
2945 		vtnet_txq_start(txq);
2946 		VTNET_TXQ_UNLOCK(txq);
2947 	}
2948 }
2949 
2950 static void
vtnet_qflush(if_t ifp)2951 vtnet_qflush(if_t ifp)
2952 {
2953 	struct vtnet_softc *sc;
2954 	struct vtnet_txq *txq;
2955 	struct mbuf *m;
2956 	int i;
2957 
2958 	sc = if_getsoftc(ifp);
2959 
2960 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2961 		txq = &sc->vtnet_txqs[i];
2962 
2963 		VTNET_TXQ_LOCK(txq);
2964 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2965 			m_freem(m);
2966 		VTNET_TXQ_UNLOCK(txq);
2967 	}
2968 
2969 	if_qflush(ifp);
2970 }
2971 
2972 static int
vtnet_watchdog(struct vtnet_txq * txq)2973 vtnet_watchdog(struct vtnet_txq *txq)
2974 {
2975 	if_t ifp;
2976 
2977 	ifp = txq->vtntx_sc->vtnet_ifp;
2978 
2979 	VTNET_TXQ_LOCK(txq);
2980 	if (txq->vtntx_watchdog == 1) {
2981 		/*
2982 		 * Only drain completed frames if the watchdog is about to
2983 		 * expire. If any frames were drained, there may be enough
2984 		 * free descriptors now available to transmit queued frames.
2985 		 * In that case, the timer will immediately be decremented
2986 		 * below, but the timeout is generous enough that should not
2987 		 * be a problem.
2988 		 */
2989 		if (vtnet_txq_eof(txq) != 0)
2990 			vtnet_txq_start(txq);
2991 	}
2992 
2993 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2994 		VTNET_TXQ_UNLOCK(txq);
2995 		return (0);
2996 	}
2997 	VTNET_TXQ_UNLOCK(txq);
2998 
2999 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
3000 	return (1);
3001 }
3002 
3003 static void
vtnet_accum_stats(struct vtnet_softc * sc,struct vtnet_rxq_stats * rxacc,struct vtnet_txq_stats * txacc)3004 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
3005     struct vtnet_txq_stats *txacc)
3006 {
3007 
3008 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
3009 	bzero(txacc, sizeof(struct vtnet_txq_stats));
3010 
3011 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3012 		struct vtnet_rxq_stats *rxst;
3013 		struct vtnet_txq_stats *txst;
3014 
3015 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
3016 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
3017 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
3018 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
3019 		rxacc->vrxs_csum += rxst->vrxs_csum;
3020 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
3021 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3022 
3023 		txst = &sc->vtnet_txqs[i].vtntx_stats;
3024 		txacc->vtxs_opackets += txst->vtxs_opackets;
3025 		txacc->vtxs_obytes += txst->vtxs_obytes;
3026 		txacc->vtxs_csum += txst->vtxs_csum;
3027 		txacc->vtxs_tso += txst->vtxs_tso;
3028 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3029 	}
3030 }
3031 
3032 static uint64_t
vtnet_get_counter(if_t ifp,ift_counter cnt)3033 vtnet_get_counter(if_t ifp, ift_counter cnt)
3034 {
3035 	struct vtnet_softc *sc;
3036 	struct vtnet_rxq_stats rxaccum;
3037 	struct vtnet_txq_stats txaccum;
3038 
3039 	sc = if_getsoftc(ifp);
3040 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3041 
3042 	switch (cnt) {
3043 	case IFCOUNTER_IPACKETS:
3044 		return (rxaccum.vrxs_ipackets);
3045 	case IFCOUNTER_IQDROPS:
3046 		return (rxaccum.vrxs_iqdrops);
3047 	case IFCOUNTER_IERRORS:
3048 		return (rxaccum.vrxs_ierrors);
3049 	case IFCOUNTER_IBYTES:
3050 		return (rxaccum.vrxs_ibytes);
3051 	case IFCOUNTER_OPACKETS:
3052 		return (txaccum.vtxs_opackets);
3053 	case IFCOUNTER_OBYTES:
3054 		return (txaccum.vtxs_obytes);
3055 	case IFCOUNTER_OMCASTS:
3056 		return (txaccum.vtxs_omcasts);
3057 	default:
3058 		return (if_get_counter_default(ifp, cnt));
3059 	}
3060 }
3061 
3062 static void
vtnet_tick(void * xsc)3063 vtnet_tick(void *xsc)
3064 {
3065 	struct vtnet_softc *sc;
3066 	if_t ifp;
3067 	int i, timedout;
3068 
3069 	sc = xsc;
3070 	ifp = sc->vtnet_ifp;
3071 	timedout = 0;
3072 
3073 	VTNET_CORE_LOCK_ASSERT(sc);
3074 
3075 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3076 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3077 
3078 	if (timedout != 0) {
3079 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3080 		vtnet_init_locked(sc, 0);
3081 	} else
3082 		callout_schedule(&sc->vtnet_tick_ch, hz);
3083 }
3084 
3085 static void
vtnet_start_taskqueues(struct vtnet_softc * sc)3086 vtnet_start_taskqueues(struct vtnet_softc *sc)
3087 {
3088 	device_t dev;
3089 	struct vtnet_rxq *rxq;
3090 	struct vtnet_txq *txq;
3091 	int i, error;
3092 
3093 	dev = sc->vtnet_dev;
3094 
3095 	/*
3096 	 * Errors here are very difficult to recover from - we cannot
3097 	 * easily fail because, if this is during boot, we will hang
3098 	 * when freeing any successfully started taskqueues because
3099 	 * the scheduler isn't up yet.
3100 	 *
3101 	 * Most drivers just ignore the return value - it only fails
3102 	 * with ENOMEM so an error is not likely.
3103 	 */
3104 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3105 		rxq = &sc->vtnet_rxqs[i];
3106 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3107 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3108 		if (error) {
3109 			device_printf(dev, "failed to start rx taskq %d\n",
3110 			    rxq->vtnrx_id);
3111 		}
3112 
3113 		txq = &sc->vtnet_txqs[i];
3114 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3115 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3116 		if (error) {
3117 			device_printf(dev, "failed to start tx taskq %d\n",
3118 			    txq->vtntx_id);
3119 		}
3120 	}
3121 }
3122 
3123 static void
vtnet_free_taskqueues(struct vtnet_softc * sc)3124 vtnet_free_taskqueues(struct vtnet_softc *sc)
3125 {
3126 	struct vtnet_rxq *rxq;
3127 	struct vtnet_txq *txq;
3128 	int i;
3129 
3130 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3131 		rxq = &sc->vtnet_rxqs[i];
3132 		if (rxq->vtnrx_tq != NULL) {
3133 			taskqueue_free(rxq->vtnrx_tq);
3134 			rxq->vtnrx_tq = NULL;
3135 		}
3136 
3137 		txq = &sc->vtnet_txqs[i];
3138 		if (txq->vtntx_tq != NULL) {
3139 			taskqueue_free(txq->vtntx_tq);
3140 			txq->vtntx_tq = NULL;
3141 		}
3142 	}
3143 }
3144 
3145 static void
vtnet_drain_taskqueues(struct vtnet_softc * sc)3146 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3147 {
3148 	struct vtnet_rxq *rxq;
3149 	struct vtnet_txq *txq;
3150 	int i;
3151 
3152 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3153 		rxq = &sc->vtnet_rxqs[i];
3154 		if (rxq->vtnrx_tq != NULL)
3155 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3156 
3157 		txq = &sc->vtnet_txqs[i];
3158 		if (txq->vtntx_tq != NULL) {
3159 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3160 			if (!VTNET_ALTQ_ENABLED)
3161 				taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3162 		}
3163 	}
3164 }
3165 
3166 static void
vtnet_drain_rxtx_queues(struct vtnet_softc * sc)3167 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3168 {
3169 	struct vtnet_rxq *rxq;
3170 	struct vtnet_txq *txq;
3171 	int i;
3172 
3173 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3174 		rxq = &sc->vtnet_rxqs[i];
3175 		vtnet_rxq_free_mbufs(rxq);
3176 
3177 		txq = &sc->vtnet_txqs[i];
3178 		vtnet_txq_free_mbufs(txq);
3179 	}
3180 }
3181 
3182 static void
vtnet_stop_rendezvous(struct vtnet_softc * sc)3183 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3184 {
3185 	struct vtnet_rxq *rxq;
3186 	struct vtnet_txq *txq;
3187 	int i;
3188 
3189 	VTNET_CORE_LOCK_ASSERT(sc);
3190 
3191 	/*
3192 	 * Lock and unlock the per-queue mutex so we known the stop
3193 	 * state is visible. Doing only the active queues should be
3194 	 * sufficient, but it does not cost much extra to do all the
3195 	 * queues.
3196 	 */
3197 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3198 		rxq = &sc->vtnet_rxqs[i];
3199 		VTNET_RXQ_LOCK(rxq);
3200 		VTNET_RXQ_UNLOCK(rxq);
3201 
3202 		txq = &sc->vtnet_txqs[i];
3203 		VTNET_TXQ_LOCK(txq);
3204 		VTNET_TXQ_UNLOCK(txq);
3205 	}
3206 }
3207 
3208 static void
vtnet_stop(struct vtnet_softc * sc)3209 vtnet_stop(struct vtnet_softc *sc)
3210 {
3211 	device_t dev;
3212 	if_t ifp;
3213 
3214 	dev = sc->vtnet_dev;
3215 	ifp = sc->vtnet_ifp;
3216 
3217 	VTNET_CORE_LOCK_ASSERT(sc);
3218 
3219 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3220 	sc->vtnet_link_active = 0;
3221 	callout_stop(&sc->vtnet_tick_ch);
3222 
3223 	/* Only advisory. */
3224 	vtnet_disable_interrupts(sc);
3225 
3226 #ifdef DEV_NETMAP
3227 	/* Stop any pending txsync/rxsync and disable them. */
3228 	netmap_disable_all_rings(ifp);
3229 #endif /* DEV_NETMAP */
3230 
3231 	/*
3232 	 * Stop the host adapter. This resets it to the pre-initialized
3233 	 * state. It will not generate any interrupts until after it is
3234 	 * reinitialized.
3235 	 */
3236 	virtio_stop(dev);
3237 	vtnet_stop_rendezvous(sc);
3238 
3239 	vtnet_drain_rxtx_queues(sc);
3240 	sc->vtnet_act_vq_pairs = 1;
3241 }
3242 
3243 static int
vtnet_virtio_reinit(struct vtnet_softc * sc)3244 vtnet_virtio_reinit(struct vtnet_softc *sc)
3245 {
3246 	device_t dev;
3247 	if_t ifp;
3248 	uint64_t features;
3249 	int error;
3250 
3251 	dev = sc->vtnet_dev;
3252 	ifp = sc->vtnet_ifp;
3253 	features = sc->vtnet_negotiated_features;
3254 
3255 	/*
3256 	 * Re-negotiate with the host, removing any disabled receive
3257 	 * features. Transmit features are disabled only on our side
3258 	 * via if_capenable and if_hwassist.
3259 	 */
3260 
3261 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3262 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3263 
3264 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3265 		features &= ~VTNET_LRO_FEATURES;
3266 
3267 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3268 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3269 
3270 	error = virtio_reinit(dev, features);
3271 	if (error) {
3272 		device_printf(dev, "virtio reinit error %d\n", error);
3273 		return (error);
3274 	}
3275 
3276 	sc->vtnet_features = features;
3277 	virtio_reinit_complete(dev);
3278 
3279 	return (0);
3280 }
3281 
3282 static void
vtnet_init_rx_filters(struct vtnet_softc * sc)3283 vtnet_init_rx_filters(struct vtnet_softc *sc)
3284 {
3285 	if_t ifp;
3286 
3287 	ifp = sc->vtnet_ifp;
3288 
3289 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3290 		vtnet_rx_filter(sc);
3291 		vtnet_rx_filter_mac(sc);
3292 	}
3293 
3294 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3295 		vtnet_rx_filter_vlan(sc);
3296 }
3297 
3298 static int
vtnet_init_rx_queues(struct vtnet_softc * sc)3299 vtnet_init_rx_queues(struct vtnet_softc *sc)
3300 {
3301 	device_t dev;
3302 	if_t ifp;
3303 	struct vtnet_rxq *rxq;
3304 	int i, clustersz, error;
3305 
3306 	dev = sc->vtnet_dev;
3307 	ifp = sc->vtnet_ifp;
3308 
3309 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3310 	sc->vtnet_rx_clustersz = clustersz;
3311 
3312 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3313 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3314 		    VTNET_MAX_RX_SIZE, clustersz);
3315 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3316 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3317 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3318 	} else
3319 		sc->vtnet_rx_nmbufs = 1;
3320 
3321 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3322 		rxq = &sc->vtnet_rxqs[i];
3323 
3324 		/* Hold the lock to satisfy asserts. */
3325 		VTNET_RXQ_LOCK(rxq);
3326 		error = vtnet_rxq_populate(rxq);
3327 		VTNET_RXQ_UNLOCK(rxq);
3328 
3329 		if (error) {
3330 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3331 			return (error);
3332 		}
3333 	}
3334 
3335 	return (0);
3336 }
3337 
3338 static int
vtnet_init_tx_queues(struct vtnet_softc * sc)3339 vtnet_init_tx_queues(struct vtnet_softc *sc)
3340 {
3341 	struct vtnet_txq *txq;
3342 	int i;
3343 
3344 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3345 		txq = &sc->vtnet_txqs[i];
3346 		txq->vtntx_watchdog = 0;
3347 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3348 #ifdef DEV_NETMAP
3349 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3350 #endif /* DEV_NETMAP */
3351 	}
3352 
3353 	return (0);
3354 }
3355 
3356 static int
vtnet_init_rxtx_queues(struct vtnet_softc * sc)3357 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3358 {
3359 	int error;
3360 
3361 	error = vtnet_init_rx_queues(sc);
3362 	if (error)
3363 		return (error);
3364 
3365 	error = vtnet_init_tx_queues(sc);
3366 	if (error)
3367 		return (error);
3368 
3369 	return (0);
3370 }
3371 
3372 static void
vtnet_set_active_vq_pairs(struct vtnet_softc * sc)3373 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3374 {
3375 	device_t dev;
3376 	int npairs;
3377 
3378 	dev = sc->vtnet_dev;
3379 
3380 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3381 		sc->vtnet_act_vq_pairs = 1;
3382 		return;
3383 	}
3384 
3385 	npairs = sc->vtnet_req_vq_pairs;
3386 
3387 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3388 		device_printf(dev, "cannot set active queue pairs to %d, "
3389 		    "falling back to 1 queue pair\n", npairs);
3390 		npairs = 1;
3391 	}
3392 
3393 	sc->vtnet_act_vq_pairs = npairs;
3394 }
3395 
3396 static void
vtnet_update_rx_offloads(struct vtnet_softc * sc)3397 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3398 {
3399 	if_t ifp;
3400 	uint64_t features;
3401 	int error;
3402 
3403 	ifp = sc->vtnet_ifp;
3404 	features = sc->vtnet_features;
3405 
3406 	VTNET_CORE_LOCK_ASSERT(sc);
3407 
3408 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3409 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3410 			features |= VIRTIO_NET_F_GUEST_CSUM;
3411 		else
3412 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3413 	}
3414 
3415 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3416 		if (if_getcapenable(ifp) & IFCAP_LRO)
3417 			features |= VTNET_LRO_FEATURES;
3418 		else
3419 			features &= ~VTNET_LRO_FEATURES;
3420 	}
3421 
3422 	error = vtnet_ctrl_guest_offloads(sc,
3423 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3424 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3425 			VIRTIO_NET_F_GUEST_UFO));
3426 	if (error) {
3427 		device_printf(sc->vtnet_dev,
3428 		    "%s: cannot update Rx features\n", __func__);
3429 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3430 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3431 			vtnet_init_locked(sc, 0);
3432 		}
3433 	} else
3434 		sc->vtnet_features = features;
3435 }
3436 
3437 static int
vtnet_reinit(struct vtnet_softc * sc)3438 vtnet_reinit(struct vtnet_softc *sc)
3439 {
3440 	if_t ifp;
3441 	int error;
3442 
3443 	ifp = sc->vtnet_ifp;
3444 
3445 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3446 
3447 	error = vtnet_virtio_reinit(sc);
3448 	if (error)
3449 		return (error);
3450 
3451 	vtnet_set_macaddr(sc);
3452 	vtnet_set_active_vq_pairs(sc);
3453 
3454 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3455 		vtnet_init_rx_filters(sc);
3456 
3457 	if_sethwassist(ifp, 0);
3458 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3459 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3460 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3461 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3462 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3463 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3464 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3465 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3466 
3467 	error = vtnet_init_rxtx_queues(sc);
3468 	if (error)
3469 		return (error);
3470 
3471 	return (0);
3472 }
3473 
3474 static void
vtnet_init_locked(struct vtnet_softc * sc,int init_mode)3475 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3476 {
3477 	if_t ifp;
3478 
3479 	ifp = sc->vtnet_ifp;
3480 
3481 	VTNET_CORE_LOCK_ASSERT(sc);
3482 
3483 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3484 		return;
3485 
3486 	vtnet_stop(sc);
3487 
3488 #ifdef DEV_NETMAP
3489 	/* Once stopped we can update the netmap flags, if necessary. */
3490 	switch (init_mode) {
3491 	case VTNET_INIT_NETMAP_ENTER:
3492 		nm_set_native_flags(NA(ifp));
3493 		break;
3494 	case VTNET_INIT_NETMAP_EXIT:
3495 		nm_clear_native_flags(NA(ifp));
3496 		break;
3497 	}
3498 #endif /* DEV_NETMAP */
3499 
3500 	if (vtnet_reinit(sc) != 0) {
3501 		vtnet_stop(sc);
3502 		return;
3503 	}
3504 
3505 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3506 	vtnet_update_link_status(sc);
3507 	vtnet_enable_interrupts(sc);
3508 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3509 
3510 #ifdef DEV_NETMAP
3511 	/* Re-enable txsync/rxsync. */
3512 	netmap_enable_all_rings(ifp);
3513 #endif /* DEV_NETMAP */
3514 }
3515 
3516 static void
vtnet_init(void * xsc)3517 vtnet_init(void *xsc)
3518 {
3519 	struct vtnet_softc *sc;
3520 
3521 	sc = xsc;
3522 
3523 	VTNET_CORE_LOCK(sc);
3524 	vtnet_init_locked(sc, 0);
3525 	VTNET_CORE_UNLOCK(sc);
3526 }
3527 
3528 static void
vtnet_free_ctrl_vq(struct vtnet_softc * sc)3529 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3530 {
3531 
3532 	/*
3533 	 * The control virtqueue is only polled and therefore it should
3534 	 * already be empty.
3535 	 */
3536 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3537 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3538 }
3539 
3540 static void
vtnet_exec_ctrl_cmd(struct vtnet_softc * sc,void * cookie,struct sglist * sg,int readable,int writable)3541 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3542     struct sglist *sg, int readable, int writable)
3543 {
3544 	struct virtqueue *vq;
3545 
3546 	vq = sc->vtnet_ctrl_vq;
3547 
3548 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3549 	VTNET_CORE_LOCK_ASSERT(sc);
3550 
3551 	if (!virtqueue_empty(vq))
3552 		return;
3553 
3554 	/*
3555 	 * Poll for the response, but the command is likely completed before
3556 	 * returning from the notify.
3557 	 */
3558 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3559 		virtqueue_notify(vq);
3560 		virtqueue_poll(vq, NULL);
3561 	}
3562 }
3563 
3564 static int
vtnet_ctrl_mac_cmd(struct vtnet_softc * sc,uint8_t * hwaddr)3565 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3566 {
3567 	struct sglist_seg segs[3];
3568 	struct sglist sg;
3569 	struct {
3570 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3571 		uint8_t pad1;
3572 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3573 		uint8_t pad2;
3574 		uint8_t ack;
3575 	} s;
3576 	int error;
3577 
3578 	error = 0;
3579 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3580 
3581 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3582 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3583 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3584 	s.ack = VIRTIO_NET_ERR;
3585 
3586 	sglist_init(&sg, nitems(segs), segs);
3587 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3588 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3589 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3590 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3591 
3592 	if (error == 0)
3593 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3594 
3595 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3596 }
3597 
3598 static int
vtnet_ctrl_guest_offloads(struct vtnet_softc * sc,uint64_t offloads)3599 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3600 {
3601 	struct sglist_seg segs[3];
3602 	struct sglist sg;
3603 	struct {
3604 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3605 		uint8_t pad1;
3606 		uint64_t offloads __aligned(8);
3607 		uint8_t pad2;
3608 		uint8_t ack;
3609 	} s;
3610 	int error;
3611 
3612 	error = 0;
3613 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3614 
3615 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3616 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3617 	s.offloads = vtnet_gtoh64(sc, offloads);
3618 	s.ack = VIRTIO_NET_ERR;
3619 
3620 	sglist_init(&sg, nitems(segs), segs);
3621 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3622 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3623 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3624 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3625 
3626 	if (error == 0)
3627 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3628 
3629 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3630 }
3631 
3632 static int
vtnet_ctrl_mq_cmd(struct vtnet_softc * sc,uint16_t npairs)3633 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3634 {
3635 	struct sglist_seg segs[3];
3636 	struct sglist sg;
3637 	struct {
3638 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3639 		uint8_t pad1;
3640 		struct virtio_net_ctrl_mq mq __aligned(2);
3641 		uint8_t pad2;
3642 		uint8_t ack;
3643 	} s;
3644 	int error;
3645 
3646 	error = 0;
3647 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3648 
3649 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3650 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3651 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3652 	s.ack = VIRTIO_NET_ERR;
3653 
3654 	sglist_init(&sg, nitems(segs), segs);
3655 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3656 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3657 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3658 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3659 
3660 	if (error == 0)
3661 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3662 
3663 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3664 }
3665 
3666 static int
vtnet_ctrl_rx_cmd(struct vtnet_softc * sc,uint8_t cmd,bool on)3667 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3668 {
3669 	struct sglist_seg segs[3];
3670 	struct sglist sg;
3671 	struct {
3672 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3673 		uint8_t pad1;
3674 		uint8_t onoff;
3675 		uint8_t pad2;
3676 		uint8_t ack;
3677 	} s;
3678 	int error;
3679 
3680 	error = 0;
3681 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3682 
3683 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3684 	s.hdr.cmd = cmd;
3685 	s.onoff = on;
3686 	s.ack = VIRTIO_NET_ERR;
3687 
3688 	sglist_init(&sg, nitems(segs), segs);
3689 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3690 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3691 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3692 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3693 
3694 	if (error == 0)
3695 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3696 
3697 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3698 }
3699 
3700 static int
vtnet_set_promisc(struct vtnet_softc * sc,bool on)3701 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3702 {
3703 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3704 }
3705 
3706 static int
vtnet_set_allmulti(struct vtnet_softc * sc,bool on)3707 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3708 {
3709 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3710 }
3711 
3712 static void
vtnet_rx_filter(struct vtnet_softc * sc)3713 vtnet_rx_filter(struct vtnet_softc *sc)
3714 {
3715 	device_t dev;
3716 	if_t ifp;
3717 
3718 	dev = sc->vtnet_dev;
3719 	ifp = sc->vtnet_ifp;
3720 
3721 	VTNET_CORE_LOCK_ASSERT(sc);
3722 
3723 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3724 		device_printf(dev, "cannot %s promiscuous mode\n",
3725 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3726 	}
3727 
3728 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3729 		device_printf(dev, "cannot %s all-multicast mode\n",
3730 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3731 	}
3732 }
3733 
3734 static u_int
vtnet_copy_ifaddr(void * arg,struct sockaddr_dl * sdl,u_int ucnt)3735 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3736 {
3737 	struct vtnet_softc *sc = arg;
3738 
3739 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3740 		return (0);
3741 
3742 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3743 		bcopy(LLADDR(sdl),
3744 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3745 		    ETHER_ADDR_LEN);
3746 
3747 	return (1);
3748 }
3749 
3750 static u_int
vtnet_copy_maddr(void * arg,struct sockaddr_dl * sdl,u_int mcnt)3751 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3752 {
3753 	struct vtnet_mac_filter *filter = arg;
3754 
3755 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3756 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3757 		    ETHER_ADDR_LEN);
3758 
3759 	return (1);
3760 }
3761 
3762 static void
vtnet_rx_filter_mac(struct vtnet_softc * sc)3763 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3764 {
3765 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3766 	struct vtnet_mac_filter *filter;
3767 	struct sglist_seg segs[4];
3768 	struct sglist sg;
3769 	if_t ifp;
3770 	bool promisc, allmulti;
3771 	u_int ucnt, mcnt;
3772 	int error;
3773 	uint8_t ack;
3774 
3775 	ifp = sc->vtnet_ifp;
3776 	filter = sc->vtnet_mac_filter;
3777 	error = 0;
3778 
3779 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3780 	VTNET_CORE_LOCK_ASSERT(sc);
3781 
3782 	/* Unicast MAC addresses: */
3783 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3784 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3785 
3786 	if (promisc) {
3787 		ucnt = 0;
3788 		if_printf(ifp, "more than %d MAC addresses assigned, "
3789 		    "falling back to promiscuous mode\n",
3790 		    VTNET_MAX_MAC_ENTRIES);
3791 	}
3792 
3793 	/* Multicast MAC addresses: */
3794 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3795 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3796 
3797 	if (allmulti) {
3798 		mcnt = 0;
3799 		if_printf(ifp, "more than %d multicast MAC addresses "
3800 		    "assigned, falling back to all-multicast mode\n",
3801 		    VTNET_MAX_MAC_ENTRIES);
3802 	}
3803 
3804 	if (promisc && allmulti)
3805 		goto out;
3806 
3807 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3808 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3809 
3810 	hdr.class = VIRTIO_NET_CTRL_MAC;
3811 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3812 	ack = VIRTIO_NET_ERR;
3813 
3814 	sglist_init(&sg, nitems(segs), segs);
3815 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3816 	error |= sglist_append(&sg, &filter->vmf_unicast,
3817 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3818 	error |= sglist_append(&sg, &filter->vmf_multicast,
3819 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3820 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3821 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3822 
3823 	if (error == 0)
3824 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3825 	if (ack != VIRTIO_NET_OK)
3826 		if_printf(ifp, "error setting host MAC filter table\n");
3827 
3828 out:
3829 	if (promisc && vtnet_set_promisc(sc, true) != 0)
3830 		if_printf(ifp, "cannot enable promiscuous mode\n");
3831 	if (allmulti && vtnet_set_allmulti(sc, true) != 0)
3832 		if_printf(ifp, "cannot enable all-multicast mode\n");
3833 }
3834 
3835 static int
vtnet_exec_vlan_filter(struct vtnet_softc * sc,int add,uint16_t tag)3836 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3837 {
3838 	struct sglist_seg segs[3];
3839 	struct sglist sg;
3840 	struct {
3841 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3842 		uint8_t pad1;
3843 		uint16_t tag __aligned(2);
3844 		uint8_t pad2;
3845 		uint8_t ack;
3846 	} s;
3847 	int error;
3848 
3849 	error = 0;
3850 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3851 
3852 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3853 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3854 	s.tag = vtnet_gtoh16(sc, tag);
3855 	s.ack = VIRTIO_NET_ERR;
3856 
3857 	sglist_init(&sg, nitems(segs), segs);
3858 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3859 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3860 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3861 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3862 
3863 	if (error == 0)
3864 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3865 
3866 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3867 }
3868 
3869 static void
vtnet_rx_filter_vlan(struct vtnet_softc * sc)3870 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3871 {
3872 	int i, bit;
3873 	uint32_t w;
3874 	uint16_t tag;
3875 
3876 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3877 	VTNET_CORE_LOCK_ASSERT(sc);
3878 
3879 	/* Enable the filter for each configured VLAN. */
3880 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3881 		w = sc->vtnet_vlan_filter[i];
3882 
3883 		while ((bit = ffs(w) - 1) != -1) {
3884 			w &= ~(1 << bit);
3885 			tag = sizeof(w) * CHAR_BIT * i + bit;
3886 
3887 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3888 				device_printf(sc->vtnet_dev,
3889 				    "cannot enable VLAN %d filter\n", tag);
3890 			}
3891 		}
3892 	}
3893 }
3894 
3895 static void
vtnet_update_vlan_filter(struct vtnet_softc * sc,int add,uint16_t tag)3896 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3897 {
3898 	if_t ifp;
3899 	int idx, bit;
3900 
3901 	ifp = sc->vtnet_ifp;
3902 	idx = (tag >> 5) & 0x7F;
3903 	bit = tag & 0x1F;
3904 
3905 	if (tag == 0 || tag > 4095)
3906 		return;
3907 
3908 	VTNET_CORE_LOCK(sc);
3909 
3910 	if (add)
3911 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3912 	else
3913 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3914 
3915 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3916 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3917 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3918 		device_printf(sc->vtnet_dev,
3919 		    "cannot %s VLAN %d %s the host filter table\n",
3920 		    add ? "add" : "remove", tag, add ? "to" : "from");
3921 	}
3922 
3923 	VTNET_CORE_UNLOCK(sc);
3924 }
3925 
3926 static void
vtnet_register_vlan(void * arg,if_t ifp,uint16_t tag)3927 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3928 {
3929 
3930 	if (if_getsoftc(ifp) != arg)
3931 		return;
3932 
3933 	vtnet_update_vlan_filter(arg, 1, tag);
3934 }
3935 
3936 static void
vtnet_unregister_vlan(void * arg,if_t ifp,uint16_t tag)3937 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3938 {
3939 
3940 	if (if_getsoftc(ifp) != arg)
3941 		return;
3942 
3943 	vtnet_update_vlan_filter(arg, 0, tag);
3944 }
3945 
3946 static void
vtnet_update_speed_duplex(struct vtnet_softc * sc)3947 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3948 {
3949 	if_t ifp;
3950 	uint32_t speed;
3951 
3952 	ifp = sc->vtnet_ifp;
3953 
3954 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3955 		return;
3956 
3957 	/* BMV: Ignore duplex. */
3958 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3959 	    offsetof(struct virtio_net_config, speed));
3960 	if (speed != UINT32_MAX)
3961 		if_setbaudrate(ifp, IF_Mbps(speed));
3962 }
3963 
3964 static int
vtnet_is_link_up(struct vtnet_softc * sc)3965 vtnet_is_link_up(struct vtnet_softc *sc)
3966 {
3967 	uint16_t status;
3968 
3969 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3970 		return (1);
3971 
3972 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3973 	    offsetof(struct virtio_net_config, status));
3974 
3975 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3976 }
3977 
3978 static void
vtnet_update_link_status(struct vtnet_softc * sc)3979 vtnet_update_link_status(struct vtnet_softc *sc)
3980 {
3981 	if_t ifp;
3982 	int link;
3983 
3984 	ifp = sc->vtnet_ifp;
3985 	VTNET_CORE_LOCK_ASSERT(sc);
3986 	link = vtnet_is_link_up(sc);
3987 
3988 	/* Notify if the link status has changed. */
3989 	if (link != 0 && sc->vtnet_link_active == 0) {
3990 		vtnet_update_speed_duplex(sc);
3991 		sc->vtnet_link_active = 1;
3992 		if_link_state_change(ifp, LINK_STATE_UP);
3993 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3994 		sc->vtnet_link_active = 0;
3995 		if_link_state_change(ifp, LINK_STATE_DOWN);
3996 	}
3997 }
3998 
3999 static int
vtnet_ifmedia_upd(if_t ifp __unused)4000 vtnet_ifmedia_upd(if_t ifp __unused)
4001 {
4002 	return (EOPNOTSUPP);
4003 }
4004 
4005 static void
vtnet_ifmedia_sts(if_t ifp,struct ifmediareq * ifmr)4006 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
4007 {
4008 	struct vtnet_softc *sc;
4009 
4010 	sc = if_getsoftc(ifp);
4011 
4012 	ifmr->ifm_status = IFM_AVALID;
4013 	ifmr->ifm_active = IFM_ETHER;
4014 
4015 	VTNET_CORE_LOCK(sc);
4016 	if (vtnet_is_link_up(sc) != 0) {
4017 		ifmr->ifm_status |= IFM_ACTIVE;
4018 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
4019 	} else
4020 		ifmr->ifm_active |= IFM_NONE;
4021 	VTNET_CORE_UNLOCK(sc);
4022 }
4023 
4024 static void
vtnet_get_macaddr(struct vtnet_softc * sc)4025 vtnet_get_macaddr(struct vtnet_softc *sc)
4026 {
4027 
4028 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4029 		virtio_read_device_config_array(sc->vtnet_dev,
4030 		    offsetof(struct virtio_net_config, mac),
4031 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4032 	} else {
4033 		/* Generate a random locally administered unicast address. */
4034 		sc->vtnet_hwaddr[0] = 0xB2;
4035 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4036 	}
4037 }
4038 
4039 static void
vtnet_set_macaddr(struct vtnet_softc * sc)4040 vtnet_set_macaddr(struct vtnet_softc *sc)
4041 {
4042 	device_t dev;
4043 	int error;
4044 
4045 	dev = sc->vtnet_dev;
4046 
4047 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4048 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4049 		if (error)
4050 			device_printf(dev, "unable to set MAC address\n");
4051 		return;
4052 	}
4053 
4054 	/* MAC in config is read-only in modern VirtIO. */
4055 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4056 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4057 			virtio_write_dev_config_1(dev,
4058 			    offsetof(struct virtio_net_config, mac) + i,
4059 			    sc->vtnet_hwaddr[i]);
4060 		}
4061 	}
4062 }
4063 
4064 static void
vtnet_attached_set_macaddr(struct vtnet_softc * sc)4065 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4066 {
4067 
4068 	/* Assign MAC address if it was generated. */
4069 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4070 		vtnet_set_macaddr(sc);
4071 }
4072 
4073 static void
vtnet_vlan_tag_remove(struct mbuf * m)4074 vtnet_vlan_tag_remove(struct mbuf *m)
4075 {
4076 	struct ether_vlan_header *evh;
4077 
4078 	evh = mtod(m, struct ether_vlan_header *);
4079 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4080 	m->m_flags |= M_VLANTAG;
4081 
4082 	/* Strip the 802.1Q header. */
4083 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4084 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4085 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4086 }
4087 
4088 static void
vtnet_set_rx_process_limit(struct vtnet_softc * sc)4089 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4090 {
4091 	int limit;
4092 
4093 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4094 	    vtnet_rx_process_limit);
4095 	if (limit < 0)
4096 		limit = INT_MAX;
4097 	sc->vtnet_rx_process_limit = limit;
4098 }
4099 
4100 static void
vtnet_setup_rxq_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_rxq * rxq)4101 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4102     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4103 {
4104 	struct sysctl_oid *node;
4105 	struct sysctl_oid_list *list;
4106 	struct vtnet_rxq_stats *stats;
4107 	char namebuf[16];
4108 
4109 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4110 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4111 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4112 	list = SYSCTL_CHILDREN(node);
4113 
4114 	stats = &rxq->vtnrx_stats;
4115 
4116 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets",
4117 	    CTLFLAG_RD | CTLFLAG_STATS,
4118 	    &stats->vrxs_ipackets, "Receive packets");
4119 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes",
4120 	    CTLFLAG_RD | CTLFLAG_STATS,
4121 	    &stats->vrxs_ibytes, "Receive bytes");
4122 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops",
4123 	    CTLFLAG_RD | CTLFLAG_STATS,
4124 	    &stats->vrxs_iqdrops, "Receive drops");
4125 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors",
4126 	    CTLFLAG_RD | CTLFLAG_STATS,
4127 	    &stats->vrxs_ierrors, "Receive errors");
4128 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4129 	    CTLFLAG_RD | CTLFLAG_STATS,
4130 	    &stats->vrxs_csum, "Receive checksum offloaded");
4131 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed",
4132 	    CTLFLAG_RD | CTLFLAG_STATS,
4133 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4134 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro",
4135 	    CTLFLAG_RD | CTLFLAG_STATS,
4136 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4137 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4138 	    CTLFLAG_RD | CTLFLAG_STATS,
4139 	    &stats->vrxs_rescheduled,
4140 	    "Receive interrupt handler rescheduled");
4141 }
4142 
4143 static void
vtnet_setup_txq_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_txq * txq)4144 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4145     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4146 {
4147 	struct sysctl_oid *node;
4148 	struct sysctl_oid_list *list;
4149 	struct vtnet_txq_stats *stats;
4150 	char namebuf[16];
4151 
4152 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4153 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4154 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4155 	list = SYSCTL_CHILDREN(node);
4156 
4157 	stats = &txq->vtntx_stats;
4158 
4159 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets",
4160 	    CTLFLAG_RD | CTLFLAG_STATS,
4161 	    &stats->vtxs_opackets, "Transmit packets");
4162 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes",
4163 	    CTLFLAG_RD | CTLFLAG_STATS,
4164 	    &stats->vtxs_obytes, "Transmit bytes");
4165 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts",
4166 	    CTLFLAG_RD | CTLFLAG_STATS,
4167 	    &stats->vtxs_omcasts, "Transmit multicasts");
4168 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4169 	    CTLFLAG_RD | CTLFLAG_STATS,
4170 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4171 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso",
4172 	    CTLFLAG_RD | CTLFLAG_STATS,
4173 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4174 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4175 	    CTLFLAG_RD | CTLFLAG_STATS,
4176 	    &stats->vtxs_rescheduled,
4177 	    "Transmit interrupt handler rescheduled");
4178 }
4179 
4180 static void
vtnet_setup_queue_sysctl(struct vtnet_softc * sc)4181 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4182 {
4183 	device_t dev;
4184 	struct sysctl_ctx_list *ctx;
4185 	struct sysctl_oid *tree;
4186 	struct sysctl_oid_list *child;
4187 	int i;
4188 
4189 	dev = sc->vtnet_dev;
4190 	ctx = device_get_sysctl_ctx(dev);
4191 	tree = device_get_sysctl_tree(dev);
4192 	child = SYSCTL_CHILDREN(tree);
4193 
4194 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4195 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4196 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4197 	}
4198 }
4199 
4200 static int
vtnet_sysctl_rx_csum_failed(SYSCTL_HANDLER_ARGS)4201 vtnet_sysctl_rx_csum_failed(SYSCTL_HANDLER_ARGS)
4202 {
4203 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4204 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4205 	struct vtnet_rxq_stats *rxst;
4206 	int i;
4207 
4208 	stats->rx_csum_failed = 0;
4209 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4210 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4211 		stats->rx_csum_failed += rxst->vrxs_csum_failed;
4212 	}
4213 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_failed, req));
4214 }
4215 
4216 static int
vtnet_sysctl_rx_csum_offloaded(SYSCTL_HANDLER_ARGS)4217 vtnet_sysctl_rx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4218 {
4219 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4220 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4221 	struct vtnet_rxq_stats *rxst;
4222 	int i;
4223 
4224 	stats->rx_csum_offloaded = 0;
4225 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4226 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4227 		stats->rx_csum_offloaded += rxst->vrxs_csum;
4228 	}
4229 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_offloaded, req));
4230 }
4231 
4232 static int
vtnet_sysctl_rx_task_rescheduled(SYSCTL_HANDLER_ARGS)4233 vtnet_sysctl_rx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4234 {
4235 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4236 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4237 	struct vtnet_rxq_stats *rxst;
4238 	int i;
4239 
4240 	stats->rx_task_rescheduled = 0;
4241 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4242 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4243 		stats->rx_task_rescheduled += rxst->vrxs_rescheduled;
4244 	}
4245 	return (sysctl_handle_64(oidp, NULL, stats->rx_task_rescheduled, req));
4246 }
4247 
4248 static int
vtnet_sysctl_tx_csum_offloaded(SYSCTL_HANDLER_ARGS)4249 vtnet_sysctl_tx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4250 {
4251 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4252 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4253 	struct vtnet_txq_stats *txst;
4254 	int i;
4255 
4256 	stats->tx_csum_offloaded = 0;
4257 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4258 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4259 		stats->tx_csum_offloaded += txst->vtxs_csum;
4260 	}
4261 	return (sysctl_handle_64(oidp, NULL, stats->tx_csum_offloaded, req));
4262 }
4263 
4264 static int
vtnet_sysctl_tx_tso_offloaded(SYSCTL_HANDLER_ARGS)4265 vtnet_sysctl_tx_tso_offloaded(SYSCTL_HANDLER_ARGS)
4266 {
4267 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4268 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4269 	struct vtnet_txq_stats *txst;
4270 	int i;
4271 
4272 	stats->tx_tso_offloaded = 0;
4273 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4274 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4275 		stats->tx_tso_offloaded += txst->vtxs_tso;
4276 	}
4277 	return (sysctl_handle_64(oidp, NULL, stats->tx_tso_offloaded, req));
4278 }
4279 
4280 static int
vtnet_sysctl_tx_task_rescheduled(SYSCTL_HANDLER_ARGS)4281 vtnet_sysctl_tx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4282 {
4283 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4284 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4285 	struct vtnet_txq_stats *txst;
4286 	int i;
4287 
4288 	stats->tx_task_rescheduled = 0;
4289 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4290 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4291 		stats->tx_task_rescheduled += txst->vtxs_rescheduled;
4292 	}
4293 	return (sysctl_handle_64(oidp, NULL, stats->tx_task_rescheduled, req));
4294 }
4295 
4296 static void
vtnet_setup_stat_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_softc * sc)4297 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4298     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4299 {
4300 	struct vtnet_statistics *stats;
4301 	struct vtnet_rxq_stats rxaccum;
4302 	struct vtnet_txq_stats txaccum;
4303 
4304 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4305 
4306 	stats = &sc->vtnet_stats;
4307 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4308 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4309 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4310 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4311 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4312 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4313 
4314 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4315 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->mbuf_alloc_failed,
4316 	    "Mbuf cluster allocation failures");
4317 
4318 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4319 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_frame_too_large,
4320 	    "Received frame larger than the mbuf chain");
4321 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4322 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_enq_replacement_failed,
4323 	    "Enqueuing the replacement receive mbuf failed");
4324 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4325 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_mergeable_failed,
4326 	    "Mergeable buffers receive failures");
4327 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4328 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ethtype,
4329 	    "Received checksum offloaded buffer with unsupported "
4330 	    "Ethernet type");
4331 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4332 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ipproto,
4333 	    "Received checksum offloaded buffer with incorrect IP protocol");
4334 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4335 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_offset,
4336 	    "Received checksum offloaded buffer with incorrect offset");
4337 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_inaccessible_ipproto",
4338 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_inaccessible_ipproto,
4339 	    "Received checksum offloaded buffer with inaccessible IP protocol");
4340 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_failed",
4341 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4342 	    sc, 0, vtnet_sysctl_rx_csum_failed, "QU",
4343 	    "Received buffer checksum offload failed");
4344 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_offloaded",
4345 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4346 	    sc, 0, vtnet_sysctl_rx_csum_offloaded, "QU",
4347 	    "Received buffer checksum offload succeeded");
4348 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_task_rescheduled",
4349 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4350 	    sc, 0, vtnet_sysctl_rx_task_rescheduled, "QU",
4351 	    "Times the receive interrupt task rescheduled itself");
4352 
4353 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4354 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_unknown_ethtype,
4355 	    "Aborted transmit of checksum offloaded buffer with unknown "
4356 	    "Ethernet type");
4357 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4358 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_proto_mismatch,
4359 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4360 	    "protocols");
4361 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4362 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_not_tcp,
4363 	    "Aborted transmit of TSO buffer with non TCP protocol");
4364 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4365 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_without_csum,
4366 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4367 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4368 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defragged,
4369 	    "Transmit mbufs defragged");
4370 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4371 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defrag_failed,
4372 	    "Aborted transmit of buffer because defrag failed");
4373 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_csum_offloaded",
4374 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4375 	    sc, 0, vtnet_sysctl_tx_csum_offloaded, "QU",
4376 	    "Offloaded checksum of transmitted buffer");
4377 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_tso_offloaded",
4378 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4379 	    sc, 0, vtnet_sysctl_tx_tso_offloaded, "QU",
4380 	    "Segmentation offload of transmitted buffer");
4381 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_task_rescheduled",
4382 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4383 	    sc, 0, vtnet_sysctl_tx_task_rescheduled, "QU",
4384 	    "Times the transmit interrupt task rescheduled itself");
4385 }
4386 
4387 static void
vtnet_setup_sysctl(struct vtnet_softc * sc)4388 vtnet_setup_sysctl(struct vtnet_softc *sc)
4389 {
4390 	device_t dev;
4391 	struct sysctl_ctx_list *ctx;
4392 	struct sysctl_oid *tree;
4393 	struct sysctl_oid_list *child;
4394 
4395 	dev = sc->vtnet_dev;
4396 	ctx = device_get_sysctl_ctx(dev);
4397 	tree = device_get_sysctl_tree(dev);
4398 	child = SYSCTL_CHILDREN(tree);
4399 
4400 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4401 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4402 	    "Number of maximum supported virtqueue pairs");
4403 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4404 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4405 	    "Number of requested virtqueue pairs");
4406 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4407 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4408 	    "Number of active virtqueue pairs");
4409 
4410 	vtnet_setup_stat_sysctl(ctx, child, sc);
4411 }
4412 
4413 static void
vtnet_load_tunables(struct vtnet_softc * sc)4414 vtnet_load_tunables(struct vtnet_softc *sc)
4415 {
4416 
4417 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4418 	    "lro_entry_count", vtnet_lro_entry_count);
4419 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4420 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4421 
4422 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4423 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4424 }
4425 
4426 static int
vtnet_rxq_enable_intr(struct vtnet_rxq * rxq)4427 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4428 {
4429 
4430 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4431 }
4432 
4433 static void
vtnet_rxq_disable_intr(struct vtnet_rxq * rxq)4434 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4435 {
4436 
4437 	virtqueue_disable_intr(rxq->vtnrx_vq);
4438 }
4439 
4440 static int
vtnet_txq_enable_intr(struct vtnet_txq * txq)4441 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4442 {
4443 	struct virtqueue *vq;
4444 
4445 	vq = txq->vtntx_vq;
4446 
4447 	if (vtnet_txq_below_threshold(txq) != 0)
4448 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4449 
4450 	/*
4451 	 * The free count is above our threshold. Keep the Tx interrupt
4452 	 * disabled until the queue is fuller.
4453 	 */
4454 	return (0);
4455 }
4456 
4457 static void
vtnet_txq_disable_intr(struct vtnet_txq * txq)4458 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4459 {
4460 
4461 	virtqueue_disable_intr(txq->vtntx_vq);
4462 }
4463 
4464 static void
vtnet_enable_rx_interrupts(struct vtnet_softc * sc)4465 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4466 {
4467 	struct vtnet_rxq *rxq;
4468 	int i;
4469 
4470 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4471 		rxq = &sc->vtnet_rxqs[i];
4472 		if (vtnet_rxq_enable_intr(rxq) != 0)
4473 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4474 	}
4475 }
4476 
4477 static void
vtnet_enable_tx_interrupts(struct vtnet_softc * sc)4478 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4479 {
4480 	int i;
4481 
4482 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4483 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4484 }
4485 
4486 static void
vtnet_enable_interrupts(struct vtnet_softc * sc)4487 vtnet_enable_interrupts(struct vtnet_softc *sc)
4488 {
4489 
4490 	vtnet_enable_rx_interrupts(sc);
4491 	vtnet_enable_tx_interrupts(sc);
4492 }
4493 
4494 static void
vtnet_disable_rx_interrupts(struct vtnet_softc * sc)4495 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4496 {
4497 	int i;
4498 
4499 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4500 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4501 }
4502 
4503 static void
vtnet_disable_tx_interrupts(struct vtnet_softc * sc)4504 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4505 {
4506 	int i;
4507 
4508 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4509 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4510 }
4511 
4512 static void
vtnet_disable_interrupts(struct vtnet_softc * sc)4513 vtnet_disable_interrupts(struct vtnet_softc *sc)
4514 {
4515 
4516 	vtnet_disable_rx_interrupts(sc);
4517 	vtnet_disable_tx_interrupts(sc);
4518 }
4519 
4520 static int
vtnet_tunable_int(struct vtnet_softc * sc,const char * knob,int def)4521 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4522 {
4523 	char path[64];
4524 
4525 	snprintf(path, sizeof(path),
4526 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4527 	TUNABLE_INT_FETCH(path, &def);
4528 
4529 	return (def);
4530 }
4531 
4532 #ifdef DEBUGNET
4533 static void
vtnet_debugnet_init(if_t ifp,int * nrxr,int * ncl,int * clsize)4534 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4535 {
4536 	struct vtnet_softc *sc;
4537 
4538 	sc = if_getsoftc(ifp);
4539 
4540 	VTNET_CORE_LOCK(sc);
4541 	*nrxr = sc->vtnet_req_vq_pairs;
4542 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4543 	*clsize = sc->vtnet_rx_clustersz;
4544 	VTNET_CORE_UNLOCK(sc);
4545 }
4546 
4547 static void
vtnet_debugnet_event(if_t ifp __unused,enum debugnet_ev event)4548 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4549 {
4550 	struct vtnet_softc *sc;
4551 	static bool sw_lro_enabled = false;
4552 
4553 	/*
4554 	 * Disable software LRO, since it would require entering the network
4555 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4556 	 */
4557 	sc = if_getsoftc(ifp);
4558 	switch (event) {
4559 	case DEBUGNET_START:
4560 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4561 		if (sw_lro_enabled)
4562 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4563 		break;
4564 	case DEBUGNET_END:
4565 		if (sw_lro_enabled)
4566 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4567 		break;
4568 	}
4569 }
4570 
4571 static int
vtnet_debugnet_transmit(if_t ifp,struct mbuf * m)4572 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4573 {
4574 	struct vtnet_softc *sc;
4575 	struct vtnet_txq *txq;
4576 	int error;
4577 
4578 	sc = if_getsoftc(ifp);
4579 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4580 	    IFF_DRV_RUNNING)
4581 		return (EBUSY);
4582 
4583 	txq = &sc->vtnet_txqs[0];
4584 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4585 	if (error == 0)
4586 		(void)vtnet_txq_notify(txq);
4587 	return (error);
4588 }
4589 
4590 static int
vtnet_debugnet_poll(if_t ifp,int count)4591 vtnet_debugnet_poll(if_t ifp, int count)
4592 {
4593 	struct vtnet_softc *sc;
4594 	int i;
4595 
4596 	sc = if_getsoftc(ifp);
4597 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4598 	    IFF_DRV_RUNNING)
4599 		return (EBUSY);
4600 
4601 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4602 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4603 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4604 	return (0);
4605 }
4606 #endif /* DEBUGNET */
4607