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 <sys/param.h>
32 #include <sys/eventhandler.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sockio.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/module.h>
39 #include <sys/msan.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49
50 #include <vm/uma.h>
51
52 #include <net/debugnet.h>
53 #include <net/ethernet.h>
54 #include <net/pfil.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_arp.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_media.h>
61 #include <net/if_vlan_var.h>
62
63 #include <net/bpf.h>
64
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/tcp.h>
72 #include <netinet/tcp_lro.h>
73
74 #include <machine/bus.h>
75 #include <machine/resource.h>
76 #include <sys/bus.h>
77 #include <sys/rman.h>
78
79 #include <dev/virtio/virtio.h>
80 #include <dev/virtio/virtqueue.h>
81 #include <dev/virtio/network/virtio_net.h>
82 #include <dev/virtio/network/if_vtnetvar.h>
83 #include "virtio_if.h"
84
85 #include "opt_inet.h"
86 #include "opt_inet6.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 static int vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
137 uint16_t, int, struct virtio_net_hdr *);
138 static int vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
139 uint16_t, int, struct virtio_net_hdr *);
140 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
141 struct virtio_net_hdr *);
142 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
143 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
144 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
145 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
146 struct virtio_net_hdr *);
147 static int vtnet_rxq_eof(struct vtnet_rxq *);
148 static void vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
149 static void vtnet_rx_vq_intr(void *);
150 static void vtnet_rxq_tq_intr(void *, int);
151
152 static int vtnet_txq_intr_threshold(struct vtnet_txq *);
153 static int vtnet_txq_below_threshold(struct vtnet_txq *);
154 static int vtnet_txq_notify(struct vtnet_txq *);
155 static void vtnet_txq_free_mbufs(struct vtnet_txq *);
156 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
157 int *, int *, int *);
158 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
159 int, struct virtio_net_hdr *);
160 static struct mbuf *
161 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
162 struct virtio_net_hdr *);
163 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
164 struct vtnet_tx_header *);
165 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
166
167 /* Required for ALTQ */
168 static void vtnet_start_locked(struct vtnet_txq *, if_t);
169 static void vtnet_start(if_t);
170
171 /* Required for MQ */
172 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
173 static int vtnet_txq_mq_start(if_t, struct mbuf *);
174 static void vtnet_txq_tq_deferred(void *, int);
175 static void vtnet_qflush(if_t);
176
177
178 static void vtnet_txq_start(struct vtnet_txq *);
179 static void vtnet_txq_tq_intr(void *, int);
180 static int vtnet_txq_eof(struct vtnet_txq *);
181 static void vtnet_tx_vq_intr(void *);
182 static void vtnet_tx_start_all(struct vtnet_softc *);
183
184 static int vtnet_watchdog(struct vtnet_txq *);
185 static void vtnet_accum_stats(struct vtnet_softc *,
186 struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
187 static void vtnet_tick(void *);
188
189 static void vtnet_start_taskqueues(struct vtnet_softc *);
190 static void vtnet_free_taskqueues(struct vtnet_softc *);
191 static void vtnet_drain_taskqueues(struct vtnet_softc *);
192
193 static void vtnet_drain_rxtx_queues(struct vtnet_softc *);
194 static void vtnet_stop_rendezvous(struct vtnet_softc *);
195 static void vtnet_stop(struct vtnet_softc *);
196 static int vtnet_virtio_reinit(struct vtnet_softc *);
197 static void vtnet_init_rx_filters(struct vtnet_softc *);
198 static int vtnet_init_rx_queues(struct vtnet_softc *);
199 static int vtnet_init_tx_queues(struct vtnet_softc *);
200 static int vtnet_init_rxtx_queues(struct vtnet_softc *);
201 static void vtnet_set_active_vq_pairs(struct vtnet_softc *);
202 static void vtnet_update_rx_offloads(struct vtnet_softc *);
203 static int vtnet_reinit(struct vtnet_softc *);
204 static void vtnet_init_locked(struct vtnet_softc *, int);
205 static void vtnet_init(void *);
206
207 static void vtnet_free_ctrl_vq(struct vtnet_softc *);
208 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
209 struct sglist *, int, int);
210 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
211 static int vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
212 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
213 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
214 static int vtnet_set_promisc(struct vtnet_softc *, bool);
215 static int vtnet_set_allmulti(struct vtnet_softc *, bool);
216 static void vtnet_rx_filter(struct vtnet_softc *);
217 static void vtnet_rx_filter_mac(struct vtnet_softc *);
218 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
219 static void vtnet_rx_filter_vlan(struct vtnet_softc *);
220 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void vtnet_register_vlan(void *, if_t, uint16_t);
222 static void vtnet_unregister_vlan(void *, if_t, uint16_t);
223
224 static void vtnet_update_speed_duplex(struct vtnet_softc *);
225 static int vtnet_is_link_up(struct vtnet_softc *);
226 static void vtnet_update_link_status(struct vtnet_softc *);
227 static int vtnet_ifmedia_upd(if_t);
228 static void vtnet_ifmedia_sts(if_t, struct ifmediareq *);
229 static void vtnet_get_macaddr(struct vtnet_softc *);
230 static void vtnet_set_macaddr(struct vtnet_softc *);
231 static void vtnet_attached_set_macaddr(struct vtnet_softc *);
232 static void vtnet_vlan_tag_remove(struct mbuf *);
233 static void vtnet_set_rx_process_limit(struct vtnet_softc *);
234
235 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
236 struct sysctl_oid_list *, struct vtnet_rxq *);
237 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
238 struct sysctl_oid_list *, struct vtnet_txq *);
239 static void vtnet_setup_queue_sysctl(struct vtnet_softc *);
240 static void vtnet_load_tunables(struct vtnet_softc *);
241 static void vtnet_setup_sysctl(struct vtnet_softc *);
242
243 static int vtnet_rxq_enable_intr(struct vtnet_rxq *);
244 static void vtnet_rxq_disable_intr(struct vtnet_rxq *);
245 static int vtnet_txq_enable_intr(struct vtnet_txq *);
246 static void vtnet_txq_disable_intr(struct vtnet_txq *);
247 static void vtnet_enable_rx_interrupts(struct vtnet_softc *);
248 static void vtnet_enable_tx_interrupts(struct vtnet_softc *);
249 static void vtnet_enable_interrupts(struct vtnet_softc *);
250 static void vtnet_disable_rx_interrupts(struct vtnet_softc *);
251 static void vtnet_disable_tx_interrupts(struct vtnet_softc *);
252 static void vtnet_disable_interrupts(struct vtnet_softc *);
253
254 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int);
255
256 DEBUGNET_DEFINE(vtnet);
257
258 #define vtnet_htog16(_sc, _val) virtio_htog16(vtnet_modern(_sc), _val)
259 #define vtnet_htog32(_sc, _val) virtio_htog32(vtnet_modern(_sc), _val)
260 #define vtnet_htog64(_sc, _val) virtio_htog64(vtnet_modern(_sc), _val)
261 #define vtnet_gtoh16(_sc, _val) virtio_gtoh16(vtnet_modern(_sc), _val)
262 #define vtnet_gtoh32(_sc, _val) virtio_gtoh32(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh64(_sc, _val) virtio_gtoh64(vtnet_modern(_sc), _val)
264
265 /* Tunables. */
266 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
267 "VirtIO Net driver parameters");
268
269 static int vtnet_csum_disable = 0;
270 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
271 &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
272
273 static int vtnet_fixup_needs_csum = 0;
274 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
275 &vtnet_fixup_needs_csum, 0,
276 "Calculate valid checksum for NEEDS_CSUM packets");
277
278 static int vtnet_tso_disable = 0;
279 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
280 &vtnet_tso_disable, 0, "Disables TSO");
281
282 static int vtnet_lro_disable = 0;
283 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
284 &vtnet_lro_disable, 0, "Disables hardware LRO");
285
286 static int vtnet_mq_disable = 0;
287 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
288 &vtnet_mq_disable, 0, "Disables multiqueue support");
289
290 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
291 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
292 &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
293
294 static int vtnet_tso_maxlen = IP_MAXPACKET;
295 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
296 &vtnet_tso_maxlen, 0, "TSO burst limit");
297
298 static int vtnet_rx_process_limit = 1024;
299 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
300 &vtnet_rx_process_limit, 0,
301 "Number of RX segments processed in one pass");
302
303 static int vtnet_lro_entry_count = 128;
304 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
305 &vtnet_lro_entry_count, 0, "Software LRO entry count");
306
307 /* Enable sorted LRO, and the depth of the mbuf queue. */
308 static int vtnet_lro_mbufq_depth = 0;
309 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
310 &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
311
312 /* Deactivate ALTQ Support */
313 static int vtnet_altq_disable = 0;
314 SYSCTL_INT(_hw_vtnet, OID_AUTO, altq_disable, CTLFLAG_RDTUN,
315 &vtnet_altq_disable, 0, "Disables ALTQ Support");
316
317 /*
318 * For the driver to be considered as having altq enabled,
319 * it must be compiled with an ALTQ capable kernel,
320 * and the tunable hw.vtnet.altq_disable must be zero
321 */
322 #define VTNET_ALTQ_ENABLED (VTNET_ALTQ_CAPABLE && (!vtnet_altq_disable))
323
324
325 static uma_zone_t vtnet_tx_header_zone;
326
327 static struct virtio_feature_desc vtnet_feature_desc[] = {
328 { VIRTIO_NET_F_CSUM, "TxChecksum" },
329 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" },
330 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "CtrlRxOffloads" },
331 { VIRTIO_NET_F_MAC, "MAC" },
332 { VIRTIO_NET_F_GSO, "TxGSO" },
333 { VIRTIO_NET_F_GUEST_TSO4, "RxLROv4" },
334 { VIRTIO_NET_F_GUEST_TSO6, "RxLROv6" },
335 { VIRTIO_NET_F_GUEST_ECN, "RxLROECN" },
336 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" },
337 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" },
338 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" },
339 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" },
340 { VIRTIO_NET_F_HOST_UFO, "TxUFO" },
341 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" },
342 { VIRTIO_NET_F_STATUS, "Status" },
343 { VIRTIO_NET_F_CTRL_VQ, "CtrlVq" },
344 { VIRTIO_NET_F_CTRL_RX, "CtrlRxMode" },
345 { VIRTIO_NET_F_CTRL_VLAN, "CtrlVLANFilter" },
346 { VIRTIO_NET_F_CTRL_RX_EXTRA, "CtrlRxModeExtra" },
347 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" },
348 { VIRTIO_NET_F_MQ, "Multiqueue" },
349 { VIRTIO_NET_F_CTRL_MAC_ADDR, "CtrlMacAddr" },
350 { VIRTIO_NET_F_SPEED_DUPLEX, "SpeedDuplex" },
351
352 { 0, NULL }
353 };
354
355 static device_method_t vtnet_methods[] = {
356 /* Device methods. */
357 DEVMETHOD(device_probe, vtnet_probe),
358 DEVMETHOD(device_attach, vtnet_attach),
359 DEVMETHOD(device_detach, vtnet_detach),
360 DEVMETHOD(device_suspend, vtnet_suspend),
361 DEVMETHOD(device_resume, vtnet_resume),
362 DEVMETHOD(device_shutdown, vtnet_shutdown),
363
364 /* VirtIO methods. */
365 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed),
366 DEVMETHOD(virtio_config_change, vtnet_config_change),
367
368 DEVMETHOD_END
369 };
370
371 #ifdef DEV_NETMAP
372 #include <dev/netmap/if_vtnet_netmap.h>
373 #endif
374
375 static driver_t vtnet_driver = {
376 .name = "vtnet",
377 .methods = vtnet_methods,
378 .size = sizeof(struct vtnet_softc)
379 };
380 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
381 MODULE_VERSION(vtnet, 1);
382 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
383 #ifdef DEV_NETMAP
384 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
385 #endif
386
387 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
388
389 static int
vtnet_modevent(module_t mod __unused,int type,void * unused __unused)390 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
391 {
392 int error = 0;
393 static int loaded = 0;
394
395 switch (type) {
396 case MOD_LOAD:
397 if (loaded++ == 0) {
398 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
399 sizeof(struct vtnet_tx_header),
400 NULL, NULL, NULL, NULL, 0, 0);
401 #ifdef DEBUGNET
402 /*
403 * We need to allocate from this zone in the transmit path, so ensure
404 * that we have at least one item per header available.
405 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
406 * buckets
407 */
408 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
409 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
410 #endif
411 }
412 break;
413 case MOD_QUIESCE:
414 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
415 error = EBUSY;
416 break;
417 case MOD_UNLOAD:
418 if (--loaded == 0) {
419 uma_zdestroy(vtnet_tx_header_zone);
420 vtnet_tx_header_zone = NULL;
421 }
422 break;
423 case MOD_SHUTDOWN:
424 break;
425 default:
426 error = EOPNOTSUPP;
427 break;
428 }
429
430 return (error);
431 }
432
433 static int
vtnet_probe(device_t dev)434 vtnet_probe(device_t dev)
435 {
436 return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
437 }
438
439 static int
vtnet_attach(device_t dev)440 vtnet_attach(device_t dev)
441 {
442 struct vtnet_softc *sc;
443 int error;
444
445 sc = device_get_softc(dev);
446 sc->vtnet_dev = dev;
447 virtio_set_feature_desc(dev, vtnet_feature_desc);
448
449 VTNET_CORE_LOCK_INIT(sc);
450 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
451 vtnet_load_tunables(sc);
452
453 vtnet_alloc_interface(sc);
454 vtnet_setup_sysctl(sc);
455
456 error = vtnet_setup_features(sc);
457 if (error) {
458 device_printf(dev, "cannot setup features\n");
459 goto fail;
460 }
461
462 error = vtnet_alloc_rx_filters(sc);
463 if (error) {
464 device_printf(dev, "cannot allocate Rx filters\n");
465 goto fail;
466 }
467
468 error = vtnet_alloc_rxtx_queues(sc);
469 if (error) {
470 device_printf(dev, "cannot allocate queues\n");
471 goto fail;
472 }
473
474 error = vtnet_alloc_virtqueues(sc);
475 if (error) {
476 device_printf(dev, "cannot allocate virtqueues\n");
477 goto fail;
478 }
479
480 error = vtnet_setup_interface(sc);
481 if (error) {
482 device_printf(dev, "cannot setup interface\n");
483 goto fail;
484 }
485
486 error = virtio_setup_intr(dev, INTR_TYPE_NET);
487 if (error) {
488 device_printf(dev, "cannot setup interrupts\n");
489 ether_ifdetach(sc->vtnet_ifp);
490 goto fail;
491 }
492
493 #ifdef DEV_NETMAP
494 vtnet_netmap_attach(sc);
495 #endif
496 vtnet_start_taskqueues(sc);
497
498 fail:
499 if (error)
500 vtnet_detach(dev);
501
502 return (error);
503 }
504
505 static int
vtnet_detach(device_t dev)506 vtnet_detach(device_t dev)
507 {
508 struct vtnet_softc *sc;
509 if_t ifp;
510
511 sc = device_get_softc(dev);
512 ifp = sc->vtnet_ifp;
513
514 if (device_is_attached(dev)) {
515 VTNET_CORE_LOCK(sc);
516 vtnet_stop(sc);
517 VTNET_CORE_UNLOCK(sc);
518
519 callout_drain(&sc->vtnet_tick_ch);
520 vtnet_drain_taskqueues(sc);
521
522 ether_ifdetach(ifp);
523 }
524
525 #ifdef DEV_NETMAP
526 netmap_detach(ifp);
527 #endif
528
529 if (sc->vtnet_pfil != NULL) {
530 pfil_head_unregister(sc->vtnet_pfil);
531 sc->vtnet_pfil = NULL;
532 }
533
534 vtnet_free_taskqueues(sc);
535
536 if (sc->vtnet_vlan_attach != NULL) {
537 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
538 sc->vtnet_vlan_attach = NULL;
539 }
540 if (sc->vtnet_vlan_detach != NULL) {
541 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
542 sc->vtnet_vlan_detach = NULL;
543 }
544
545 ifmedia_removeall(&sc->vtnet_media);
546
547 if (ifp != NULL) {
548 if_free(ifp);
549 sc->vtnet_ifp = NULL;
550 }
551
552 vtnet_free_rxtx_queues(sc);
553 vtnet_free_rx_filters(sc);
554
555 if (sc->vtnet_ctrl_vq != NULL)
556 vtnet_free_ctrl_vq(sc);
557
558 VTNET_CORE_LOCK_DESTROY(sc);
559
560 return (0);
561 }
562
563 static int
vtnet_suspend(device_t dev)564 vtnet_suspend(device_t dev)
565 {
566 struct vtnet_softc *sc;
567
568 sc = device_get_softc(dev);
569
570 VTNET_CORE_LOCK(sc);
571 vtnet_stop(sc);
572 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
573 VTNET_CORE_UNLOCK(sc);
574
575 return (0);
576 }
577
578 static int
vtnet_resume(device_t dev)579 vtnet_resume(device_t dev)
580 {
581 struct vtnet_softc *sc;
582 if_t ifp;
583
584 sc = device_get_softc(dev);
585 ifp = sc->vtnet_ifp;
586
587 VTNET_CORE_LOCK(sc);
588 if (if_getflags(ifp) & IFF_UP)
589 vtnet_init_locked(sc, 0);
590 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
591 VTNET_CORE_UNLOCK(sc);
592
593 return (0);
594 }
595
596 static int
vtnet_shutdown(device_t dev)597 vtnet_shutdown(device_t dev)
598 {
599 /*
600 * Suspend already does all of what we need to
601 * do here; we just never expect to be resumed.
602 */
603 return (vtnet_suspend(dev));
604 }
605
606 static int
vtnet_attach_completed(device_t dev)607 vtnet_attach_completed(device_t dev)
608 {
609 struct vtnet_softc *sc;
610
611 sc = device_get_softc(dev);
612
613 VTNET_CORE_LOCK(sc);
614 vtnet_attached_set_macaddr(sc);
615 VTNET_CORE_UNLOCK(sc);
616
617 return (0);
618 }
619
620 static int
vtnet_config_change(device_t dev)621 vtnet_config_change(device_t dev)
622 {
623 struct vtnet_softc *sc;
624
625 sc = device_get_softc(dev);
626
627 VTNET_CORE_LOCK(sc);
628 vtnet_update_link_status(sc);
629 if (sc->vtnet_link_active != 0)
630 vtnet_tx_start_all(sc);
631 VTNET_CORE_UNLOCK(sc);
632
633 return (0);
634 }
635
636 static int
vtnet_negotiate_features(struct vtnet_softc * sc)637 vtnet_negotiate_features(struct vtnet_softc *sc)
638 {
639 device_t dev;
640 uint64_t features, negotiated_features;
641 int no_csum;
642
643 dev = sc->vtnet_dev;
644 features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
645 VTNET_LEGACY_FEATURES;
646
647 /*
648 * TSO and LRO are only available when their corresponding checksum
649 * offload feature is also negotiated.
650 */
651 no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
652 if (no_csum)
653 features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
654 if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
655 features &= ~VTNET_TSO_FEATURES;
656 if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
657 features &= ~VTNET_LRO_FEATURES;
658
659 /* Deactivate MQ Feature flag, if driver has ALTQ enabled, or MQ is explicitly disabled */
660 if (VTNET_ALTQ_ENABLED || vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
661 features &= ~VIRTIO_NET_F_MQ;
662
663 negotiated_features = virtio_negotiate_features(dev, features);
664
665 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
666 uint16_t mtu;
667
668 mtu = virtio_read_dev_config_2(dev,
669 offsetof(struct virtio_net_config, mtu));
670 if (mtu < VTNET_MIN_MTU) {
671 device_printf(dev, "Invalid MTU value: %d. "
672 "MTU feature disabled.\n", mtu);
673 features &= ~VIRTIO_NET_F_MTU;
674 negotiated_features =
675 virtio_negotiate_features(dev, features);
676 }
677 }
678
679 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
680 uint16_t npairs;
681
682 npairs = virtio_read_dev_config_2(dev,
683 offsetof(struct virtio_net_config, max_virtqueue_pairs));
684 if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
685 npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
686 device_printf(dev, "Invalid max_virtqueue_pairs value: "
687 "%d. Multiqueue feature disabled.\n", npairs);
688 features &= ~VIRTIO_NET_F_MQ;
689 negotiated_features =
690 virtio_negotiate_features(dev, features);
691 }
692 }
693
694 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
695 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
696 /*
697 * LRO without mergeable buffers requires special care. This
698 * is not ideal because every receive buffer must be large
699 * enough to hold the maximum TCP packet, the Ethernet header,
700 * and the header. This requires up to 34 descriptors with
701 * MCLBYTES clusters. If we do not have indirect descriptors,
702 * LRO is disabled since the virtqueue will not contain very
703 * many receive buffers.
704 */
705 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
706 device_printf(dev,
707 "Host LRO disabled since both mergeable buffers "
708 "and indirect descriptors were not negotiated\n");
709 features &= ~VTNET_LRO_FEATURES;
710 negotiated_features =
711 virtio_negotiate_features(dev, features);
712 } else
713 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
714 }
715
716 sc->vtnet_features = negotiated_features;
717 sc->vtnet_negotiated_features = negotiated_features;
718
719 return (virtio_finalize_features(dev));
720 }
721
722 static int
vtnet_setup_features(struct vtnet_softc * sc)723 vtnet_setup_features(struct vtnet_softc *sc)
724 {
725 device_t dev;
726 int error;
727
728 dev = sc->vtnet_dev;
729
730 error = vtnet_negotiate_features(sc);
731 if (error)
732 return (error);
733
734 if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
735 sc->vtnet_flags |= VTNET_FLAG_MODERN;
736 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
737 sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
738 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
739 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
740
741 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
742 /* This feature should always be negotiated. */
743 sc->vtnet_flags |= VTNET_FLAG_MAC;
744 }
745
746 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
747 sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
748 offsetof(struct virtio_net_config, mtu));
749 } else
750 sc->vtnet_max_mtu = VTNET_MAX_MTU;
751
752 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
753 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
754 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
755 } else if (vtnet_modern(sc)) {
756 /* This is identical to the mergeable header. */
757 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
758 } else
759 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
760
761 if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
762 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
763 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
764 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
765 else
766 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
767
768 /*
769 * Favor "hardware" LRO if negotiated, but support software LRO as
770 * a fallback; there is usually little benefit (or worse) with both.
771 */
772 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
773 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
774 sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
775
776 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
777 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
778 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
779 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
780 else
781 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
782
783 sc->vtnet_req_vq_pairs = 1;
784 sc->vtnet_max_vq_pairs = 1;
785
786 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
787 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
788
789 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
790 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
791 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
792 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
793 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
794 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
795
796 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
797 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
798 offsetof(struct virtio_net_config,
799 max_virtqueue_pairs));
800 }
801 }
802
803 if (sc->vtnet_max_vq_pairs > 1) {
804 int req;
805
806 /*
807 * Limit the maximum number of requested queue pairs to the
808 * number of CPUs and the configured maximum.
809 */
810 req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
811 if (req < 0)
812 req = 1;
813 if (req == 0)
814 req = mp_ncpus;
815 if (req > sc->vtnet_max_vq_pairs)
816 req = sc->vtnet_max_vq_pairs;
817 if (req > mp_ncpus)
818 req = mp_ncpus;
819 if (req > 1) {
820 sc->vtnet_req_vq_pairs = req;
821 sc->vtnet_flags |= VTNET_FLAG_MQ;
822 }
823 }
824
825 return (0);
826 }
827
828 static int
vtnet_init_rxq(struct vtnet_softc * sc,int id)829 vtnet_init_rxq(struct vtnet_softc *sc, int id)
830 {
831 struct vtnet_rxq *rxq;
832
833 rxq = &sc->vtnet_rxqs[id];
834
835 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
836 device_get_nameunit(sc->vtnet_dev), id);
837 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
838
839 rxq->vtnrx_sc = sc;
840 rxq->vtnrx_id = id;
841
842 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
843 if (rxq->vtnrx_sg == NULL)
844 return (ENOMEM);
845
846 #if defined(INET) || defined(INET6)
847 if (vtnet_software_lro(sc)) {
848 if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
849 sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
850 return (ENOMEM);
851 }
852 #endif
853
854 NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
855 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
856 taskqueue_thread_enqueue, &rxq->vtnrx_tq);
857
858 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
859 }
860
861 static int
vtnet_init_txq(struct vtnet_softc * sc,int id)862 vtnet_init_txq(struct vtnet_softc *sc, int id)
863 {
864 struct vtnet_txq *txq;
865
866 txq = &sc->vtnet_txqs[id];
867
868 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
869 device_get_nameunit(sc->vtnet_dev), id);
870 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
871
872 txq->vtntx_sc = sc;
873 txq->vtntx_id = id;
874
875 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
876 if (txq->vtntx_sg == NULL)
877 return (ENOMEM);
878
879 if (!VTNET_ALTQ_ENABLED) {
880 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
881 M_NOWAIT, &txq->vtntx_mtx);
882 if (txq->vtntx_br == NULL)
883 return (ENOMEM);
884
885 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
886 }
887 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
888 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
889 taskqueue_thread_enqueue, &txq->vtntx_tq);
890 if (txq->vtntx_tq == NULL)
891 return (ENOMEM);
892
893 return (0);
894 }
895
896 static int
vtnet_alloc_rxtx_queues(struct vtnet_softc * sc)897 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
898 {
899 int i, npairs, error;
900
901 npairs = sc->vtnet_max_vq_pairs;
902
903 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
904 M_NOWAIT | M_ZERO);
905 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
906 M_NOWAIT | M_ZERO);
907 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
908 return (ENOMEM);
909
910 for (i = 0; i < npairs; i++) {
911 error = vtnet_init_rxq(sc, i);
912 if (error)
913 return (error);
914 error = vtnet_init_txq(sc, i);
915 if (error)
916 return (error);
917 }
918
919 vtnet_set_rx_process_limit(sc);
920 vtnet_setup_queue_sysctl(sc);
921
922 return (0);
923 }
924
925 static void
vtnet_destroy_rxq(struct vtnet_rxq * rxq)926 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
927 {
928
929 rxq->vtnrx_sc = NULL;
930 rxq->vtnrx_id = -1;
931
932 #if defined(INET) || defined(INET6)
933 tcp_lro_free(&rxq->vtnrx_lro);
934 #endif
935
936 if (rxq->vtnrx_sg != NULL) {
937 sglist_free(rxq->vtnrx_sg);
938 rxq->vtnrx_sg = NULL;
939 }
940
941 if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
942 mtx_destroy(&rxq->vtnrx_mtx);
943 }
944
945 static void
vtnet_destroy_txq(struct vtnet_txq * txq)946 vtnet_destroy_txq(struct vtnet_txq *txq)
947 {
948
949 txq->vtntx_sc = NULL;
950 txq->vtntx_id = -1;
951
952 if (txq->vtntx_sg != NULL) {
953 sglist_free(txq->vtntx_sg);
954 txq->vtntx_sg = NULL;
955 }
956
957 if (!VTNET_ALTQ_ENABLED) {
958 if (txq->vtntx_br != NULL) {
959 buf_ring_free(txq->vtntx_br, M_DEVBUF);
960 txq->vtntx_br = NULL;
961 }
962 }
963
964 if (mtx_initialized(&txq->vtntx_mtx) != 0)
965 mtx_destroy(&txq->vtntx_mtx);
966 }
967
968 static void
vtnet_free_rxtx_queues(struct vtnet_softc * sc)969 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
970 {
971 int i;
972
973 if (sc->vtnet_rxqs != NULL) {
974 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
975 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
976 free(sc->vtnet_rxqs, M_DEVBUF);
977 sc->vtnet_rxqs = NULL;
978 }
979
980 if (sc->vtnet_txqs != NULL) {
981 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
982 vtnet_destroy_txq(&sc->vtnet_txqs[i]);
983 free(sc->vtnet_txqs, M_DEVBUF);
984 sc->vtnet_txqs = NULL;
985 }
986 }
987
988 static int
vtnet_alloc_rx_filters(struct vtnet_softc * sc)989 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
990 {
991
992 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
993 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
994 M_DEVBUF, M_NOWAIT | M_ZERO);
995 if (sc->vtnet_mac_filter == NULL)
996 return (ENOMEM);
997 }
998
999 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1000 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
1001 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
1002 if (sc->vtnet_vlan_filter == NULL)
1003 return (ENOMEM);
1004 }
1005
1006 return (0);
1007 }
1008
1009 static void
vtnet_free_rx_filters(struct vtnet_softc * sc)1010 vtnet_free_rx_filters(struct vtnet_softc *sc)
1011 {
1012
1013 if (sc->vtnet_mac_filter != NULL) {
1014 free(sc->vtnet_mac_filter, M_DEVBUF);
1015 sc->vtnet_mac_filter = NULL;
1016 }
1017
1018 if (sc->vtnet_vlan_filter != NULL) {
1019 free(sc->vtnet_vlan_filter, M_DEVBUF);
1020 sc->vtnet_vlan_filter = NULL;
1021 }
1022 }
1023
1024 static int
vtnet_alloc_virtqueues(struct vtnet_softc * sc)1025 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1026 {
1027 device_t dev;
1028 struct vq_alloc_info *info;
1029 struct vtnet_rxq *rxq;
1030 struct vtnet_txq *txq;
1031 int i, idx, nvqs, error;
1032
1033 dev = sc->vtnet_dev;
1034
1035 nvqs = sc->vtnet_max_vq_pairs * 2;
1036 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1037 nvqs++;
1038
1039 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1040 if (info == NULL)
1041 return (ENOMEM);
1042
1043 for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1044 rxq = &sc->vtnet_rxqs[i];
1045 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1046 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1047 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1048
1049 txq = &sc->vtnet_txqs[i];
1050 VQ_ALLOC_INFO_INIT(&info[idx + 1], sc->vtnet_tx_nsegs,
1051 vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1052 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1053 }
1054
1055 /* These queues will not be used so allocate the minimum resources. */
1056 for (; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1057 rxq = &sc->vtnet_rxqs[i];
1058 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1059 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1060
1061 txq = &sc->vtnet_txqs[i];
1062 VQ_ALLOC_INFO_INIT(&info[idx + 1], 0, NULL, txq, &txq->vtntx_vq,
1063 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1064 }
1065
1066 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1067 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1068 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1069 }
1070
1071 error = virtio_alloc_virtqueues(dev, nvqs, info);
1072 free(info, M_TEMP);
1073
1074 return (error);
1075 }
1076
1077 static void
vtnet_alloc_interface(struct vtnet_softc * sc)1078 vtnet_alloc_interface(struct vtnet_softc *sc)
1079 {
1080 device_t dev;
1081 if_t ifp;
1082
1083 dev = sc->vtnet_dev;
1084
1085 ifp = if_alloc(IFT_ETHER);
1086 sc->vtnet_ifp = ifp;
1087 if_setsoftc(ifp, sc);
1088 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1089 }
1090
1091 static int
vtnet_setup_interface(struct vtnet_softc * sc)1092 vtnet_setup_interface(struct vtnet_softc *sc)
1093 {
1094 device_t dev;
1095 struct pfil_head_args pa;
1096 if_t ifp;
1097
1098 dev = sc->vtnet_dev;
1099 ifp = sc->vtnet_ifp;
1100
1101 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1102 if_setbaudrate(ifp, IF_Gbps(10));
1103 if_setinitfn(ifp, vtnet_init);
1104 if_setioctlfn(ifp, vtnet_ioctl);
1105 if_setgetcounterfn(ifp, vtnet_get_counter);
1106
1107 if (!VTNET_ALTQ_ENABLED) {
1108 if_settransmitfn(ifp, vtnet_txq_mq_start);
1109 if_setqflushfn(ifp, vtnet_qflush);
1110 } else {
1111 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1112 if_setstartfn(ifp, vtnet_start);
1113 if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1114 if_setsendqready(ifp);
1115 }
1116
1117 vtnet_get_macaddr(sc);
1118
1119 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1120 if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1121
1122 ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1123 ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1124 ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1125
1126 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1127 int gso;
1128
1129 if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1130
1131 gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1132 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1133 if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1134 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1135 if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1136 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1137 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1138
1139 if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1140 int tso_maxlen;
1141
1142 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1143
1144 tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1145 vtnet_tso_maxlen);
1146 if_sethwtsomax(ifp, tso_maxlen -
1147 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1148 if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1149 if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1150 }
1151 }
1152
1153 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1154 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1155 #ifdef notyet
1156 /* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1157 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1158 #endif
1159
1160 if (vtnet_tunable_int(sc, "fixup_needs_csum",
1161 vtnet_fixup_needs_csum) != 0)
1162 sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1163
1164 /* Support either "hardware" or software LRO. */
1165 if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1166 }
1167
1168 if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1169 /*
1170 * VirtIO does not support VLAN tagging, but we can fake
1171 * it by inserting and removing the 802.1Q header during
1172 * transmit and receive. We are then able to do checksum
1173 * offloading of VLAN frames.
1174 */
1175 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1176 }
1177
1178 if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1179 if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1180 if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1181
1182 /*
1183 * Capabilities after here are not enabled by default.
1184 */
1185 if_setcapenable(ifp, if_getcapabilities(ifp));
1186
1187 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1188 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1189
1190 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1191 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1192 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1193 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1194 }
1195
1196 ether_ifattach(ifp, sc->vtnet_hwaddr);
1197
1198 /* Tell the upper layer(s) we support long frames. */
1199 if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1200
1201 DEBUGNET_SET(ifp, vtnet);
1202
1203 pa.pa_version = PFIL_VERSION;
1204 pa.pa_flags = PFIL_IN;
1205 pa.pa_type = PFIL_TYPE_ETHERNET;
1206 pa.pa_headname = if_name(ifp);
1207 sc->vtnet_pfil = pfil_head_register(&pa);
1208
1209 return (0);
1210 }
1211
1212 static int
vtnet_rx_cluster_size(struct vtnet_softc * sc,int mtu)1213 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1214 {
1215 int framesz;
1216
1217 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1218 return (MJUMPAGESIZE);
1219 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1220 return (MCLBYTES);
1221
1222 /*
1223 * Try to scale the receive mbuf cluster size from the MTU. We
1224 * could also use the VQ size to influence the selected size,
1225 * but that would only matter for very small queues.
1226 */
1227 if (vtnet_modern(sc)) {
1228 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1229 framesz = sizeof(struct virtio_net_hdr_v1);
1230 } else
1231 framesz = sizeof(struct vtnet_rx_header);
1232 framesz += sizeof(struct ether_vlan_header) + mtu;
1233 /*
1234 * Account for the offsetting we'll do elsewhere so we allocate the
1235 * right size for the mtu.
1236 */
1237 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1238 framesz += VTNET_ETHER_ALIGN;
1239 }
1240
1241 if (framesz <= MCLBYTES)
1242 return (MCLBYTES);
1243 else if (framesz <= MJUMPAGESIZE)
1244 return (MJUMPAGESIZE);
1245 else if (framesz <= MJUM9BYTES)
1246 return (MJUM9BYTES);
1247
1248 /* Sane default; avoid 16KB clusters. */
1249 return (MCLBYTES);
1250 }
1251
1252 static int
vtnet_ioctl_mtu(struct vtnet_softc * sc,u_int mtu)1253 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1254 {
1255 if_t ifp;
1256 int clustersz;
1257
1258 ifp = sc->vtnet_ifp;
1259 VTNET_CORE_LOCK_ASSERT(sc);
1260
1261 if (if_getmtu(ifp) == mtu)
1262 return (0);
1263 else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1264 return (EINVAL);
1265
1266 if_setmtu(ifp, mtu);
1267 clustersz = vtnet_rx_cluster_size(sc, mtu);
1268
1269 if (clustersz != sc->vtnet_rx_clustersz &&
1270 if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1271 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1272 vtnet_init_locked(sc, 0);
1273 }
1274
1275 return (0);
1276 }
1277
1278 static int
vtnet_ioctl_ifflags(struct vtnet_softc * sc)1279 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1280 {
1281 if_t ifp;
1282 int drv_running;
1283
1284 ifp = sc->vtnet_ifp;
1285 drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1286
1287 VTNET_CORE_LOCK_ASSERT(sc);
1288
1289 if ((if_getflags(ifp) & IFF_UP) == 0) {
1290 if (drv_running)
1291 vtnet_stop(sc);
1292 goto out;
1293 }
1294
1295 if (!drv_running) {
1296 vtnet_init_locked(sc, 0);
1297 goto out;
1298 }
1299
1300 if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1301 (IFF_PROMISC | IFF_ALLMULTI)) {
1302 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1303 vtnet_rx_filter(sc);
1304 else {
1305 /*
1306 * We don't support filtering out multicast, so
1307 * ALLMULTI is always set.
1308 */
1309 if_setflagbits(ifp, IFF_ALLMULTI, 0);
1310 if_setflagbits(ifp, IFF_PROMISC, 0);
1311 }
1312 }
1313
1314 out:
1315 sc->vtnet_if_flags = if_getflags(ifp);
1316 return (0);
1317 }
1318
1319 static int
vtnet_ioctl_multi(struct vtnet_softc * sc)1320 vtnet_ioctl_multi(struct vtnet_softc *sc)
1321 {
1322 if_t ifp;
1323
1324 ifp = sc->vtnet_ifp;
1325
1326 VTNET_CORE_LOCK_ASSERT(sc);
1327
1328 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1329 if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1330 vtnet_rx_filter_mac(sc);
1331
1332 return (0);
1333 }
1334
1335 static int
vtnet_ioctl_ifcap(struct vtnet_softc * sc,struct ifreq * ifr)1336 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1337 {
1338 if_t ifp;
1339 int mask, reinit, update;
1340
1341 ifp = sc->vtnet_ifp;
1342 mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1343 reinit = update = 0;
1344
1345 VTNET_CORE_LOCK_ASSERT(sc);
1346
1347 if (mask & IFCAP_TXCSUM)
1348 if_togglecapenable(ifp, IFCAP_TXCSUM);
1349 if (mask & IFCAP_TXCSUM_IPV6)
1350 if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1351 if (mask & IFCAP_TSO4)
1352 if_togglecapenable(ifp, IFCAP_TSO4);
1353 if (mask & IFCAP_TSO6)
1354 if_togglecapenable(ifp, IFCAP_TSO6);
1355
1356 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1357 /*
1358 * These Rx features require the negotiated features to
1359 * be updated. Avoid a full reinit if possible.
1360 */
1361 if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1362 update = 1;
1363 else
1364 reinit = 1;
1365
1366 /* BMV: Avoid needless renegotiation for just software LRO. */
1367 if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1368 IFCAP_LRO && vtnet_software_lro(sc))
1369 reinit = update = 0;
1370
1371 if (mask & IFCAP_RXCSUM)
1372 if_togglecapenable(ifp, IFCAP_RXCSUM);
1373 if (mask & IFCAP_RXCSUM_IPV6)
1374 if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1375 if (mask & IFCAP_LRO)
1376 if_togglecapenable(ifp, IFCAP_LRO);
1377
1378 /*
1379 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1380 * so treat them as a pair. Guest TSO (LRO) requires receive
1381 * checksums.
1382 */
1383 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1384 if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1385 #ifdef notyet
1386 if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1387 #endif
1388 } else
1389 if_setcapenablebit(ifp, 0,
1390 (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1391 }
1392
1393 if (mask & IFCAP_VLAN_HWFILTER) {
1394 /* These Rx features require renegotiation. */
1395 reinit = 1;
1396
1397 if (mask & IFCAP_VLAN_HWFILTER)
1398 if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1399 }
1400
1401 if (mask & IFCAP_VLAN_HWTSO)
1402 if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1403 if (mask & IFCAP_VLAN_HWTAGGING)
1404 if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1405
1406 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1407 if (reinit) {
1408 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1409 vtnet_init_locked(sc, 0);
1410 } else if (update)
1411 vtnet_update_rx_offloads(sc);
1412 }
1413
1414 return (0);
1415 }
1416
1417 static int
vtnet_ioctl(if_t ifp,u_long cmd,caddr_t data)1418 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1419 {
1420 struct vtnet_softc *sc;
1421 struct ifreq *ifr;
1422 int error;
1423
1424 sc = if_getsoftc(ifp);
1425 ifr = (struct ifreq *) data;
1426 error = 0;
1427
1428 switch (cmd) {
1429 case SIOCSIFMTU:
1430 VTNET_CORE_LOCK(sc);
1431 error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1432 VTNET_CORE_UNLOCK(sc);
1433 break;
1434
1435 case SIOCSIFFLAGS:
1436 VTNET_CORE_LOCK(sc);
1437 error = vtnet_ioctl_ifflags(sc);
1438 VTNET_CORE_UNLOCK(sc);
1439 break;
1440
1441 case SIOCADDMULTI:
1442 case SIOCDELMULTI:
1443 VTNET_CORE_LOCK(sc);
1444 error = vtnet_ioctl_multi(sc);
1445 VTNET_CORE_UNLOCK(sc);
1446 break;
1447
1448 case SIOCSIFMEDIA:
1449 case SIOCGIFMEDIA:
1450 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1451 break;
1452
1453 case SIOCSIFCAP:
1454 VTNET_CORE_LOCK(sc);
1455 error = vtnet_ioctl_ifcap(sc, ifr);
1456 VTNET_CORE_UNLOCK(sc);
1457 VLAN_CAPABILITIES(ifp);
1458 break;
1459
1460 default:
1461 error = ether_ioctl(ifp, cmd, data);
1462 break;
1463 }
1464
1465 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1466
1467 return (error);
1468 }
1469
1470 static int
vtnet_rxq_populate(struct vtnet_rxq * rxq)1471 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1472 {
1473 struct virtqueue *vq;
1474 int nbufs, error;
1475
1476 #ifdef DEV_NETMAP
1477 error = vtnet_netmap_rxq_populate(rxq);
1478 if (error >= 0)
1479 return (error);
1480 #endif /* DEV_NETMAP */
1481
1482 vq = rxq->vtnrx_vq;
1483 error = ENOSPC;
1484
1485 for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1486 error = vtnet_rxq_new_buf(rxq);
1487 if (error)
1488 break;
1489 }
1490
1491 if (nbufs > 0) {
1492 virtqueue_notify(vq);
1493 /*
1494 * EMSGSIZE signifies the virtqueue did not have enough
1495 * entries available to hold the last mbuf. This is not
1496 * an error.
1497 */
1498 if (error == EMSGSIZE)
1499 error = 0;
1500 }
1501
1502 return (error);
1503 }
1504
1505 static void
vtnet_rxq_free_mbufs(struct vtnet_rxq * rxq)1506 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1507 {
1508 struct virtqueue *vq;
1509 struct mbuf *m;
1510 int last;
1511 #ifdef DEV_NETMAP
1512 struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1513 rxq->vtnrx_id, NR_RX);
1514 #else /* !DEV_NETMAP */
1515 void *kring = NULL;
1516 #endif /* !DEV_NETMAP */
1517
1518 vq = rxq->vtnrx_vq;
1519 last = 0;
1520
1521 while ((m = virtqueue_drain(vq, &last)) != NULL) {
1522 if (kring == NULL)
1523 m_freem(m);
1524 }
1525
1526 KASSERT(virtqueue_empty(vq),
1527 ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1528 }
1529
1530 static struct mbuf *
vtnet_rx_alloc_buf(struct vtnet_softc * sc,int nbufs,struct mbuf ** m_tailp)1531 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1532 {
1533 struct mbuf *m_head, *m_tail, *m;
1534 int i, size;
1535
1536 m_head = NULL;
1537 size = sc->vtnet_rx_clustersz;
1538
1539 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1540 ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1541
1542 for (i = 0; i < nbufs; i++) {
1543 m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1544 if (m == NULL) {
1545 sc->vtnet_stats.mbuf_alloc_failed++;
1546 m_freem(m_head);
1547 return (NULL);
1548 }
1549
1550 m->m_len = size;
1551 /*
1552 * Need to offset the mbuf if the header we're going to add
1553 * will misalign.
1554 */
1555 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1556 m_adj(m, VTNET_ETHER_ALIGN);
1557 }
1558 if (m_head != NULL) {
1559 m_tail->m_next = m;
1560 m_tail = m;
1561 } else
1562 m_head = m_tail = m;
1563 }
1564
1565 if (m_tailp != NULL)
1566 *m_tailp = m_tail;
1567
1568 return (m_head);
1569 }
1570
1571 /*
1572 * Slow path for when LRO without mergeable buffers is negotiated.
1573 */
1574 static int
vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq * rxq,struct mbuf * m0,int len0)1575 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1576 int len0)
1577 {
1578 struct vtnet_softc *sc;
1579 struct mbuf *m, *m_prev, *m_new, *m_tail;
1580 int len, clustersz, nreplace, error;
1581
1582 sc = rxq->vtnrx_sc;
1583 clustersz = sc->vtnet_rx_clustersz;
1584 /*
1585 * Need to offset the mbuf if the header we're going to add will
1586 * misalign, account for that here.
1587 */
1588 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1589 clustersz -= VTNET_ETHER_ALIGN;
1590
1591 m_prev = NULL;
1592 m_tail = NULL;
1593 nreplace = 0;
1594
1595 m = m0;
1596 len = len0;
1597
1598 /*
1599 * Since these mbuf chains are so large, avoid allocating a complete
1600 * replacement when the received frame did not consume the entire
1601 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1602 */
1603 while (len > 0) {
1604 if (m == NULL) {
1605 sc->vtnet_stats.rx_frame_too_large++;
1606 return (EMSGSIZE);
1607 }
1608
1609 /*
1610 * Every mbuf should have the expected cluster size since that
1611 * is also used to allocate the replacements.
1612 */
1613 KASSERT(m->m_len == clustersz,
1614 ("%s: mbuf size %d not expected cluster size %d", __func__,
1615 m->m_len, clustersz));
1616
1617 m->m_len = MIN(m->m_len, len);
1618 len -= m->m_len;
1619
1620 m_prev = m;
1621 m = m->m_next;
1622 nreplace++;
1623 }
1624
1625 KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1626 ("%s: invalid replacement mbuf count %d max %d", __func__,
1627 nreplace, sc->vtnet_rx_nmbufs));
1628
1629 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1630 if (m_new == NULL) {
1631 m_prev->m_len = clustersz;
1632 return (ENOBUFS);
1633 }
1634
1635 /*
1636 * Move any unused mbufs from the received mbuf chain onto the
1637 * end of the replacement chain.
1638 */
1639 if (m_prev->m_next != NULL) {
1640 m_tail->m_next = m_prev->m_next;
1641 m_prev->m_next = NULL;
1642 }
1643
1644 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1645 if (error) {
1646 /*
1647 * The replacement is suppose to be an copy of the one
1648 * dequeued so this is a very unexpected error.
1649 *
1650 * Restore the m0 chain to the original state if it was
1651 * modified so we can then discard it.
1652 */
1653 if (m_tail->m_next != NULL) {
1654 m_prev->m_next = m_tail->m_next;
1655 m_tail->m_next = NULL;
1656 }
1657 m_prev->m_len = clustersz;
1658 sc->vtnet_stats.rx_enq_replacement_failed++;
1659 m_freem(m_new);
1660 }
1661
1662 return (error);
1663 }
1664
1665 static int
vtnet_rxq_replace_buf(struct vtnet_rxq * rxq,struct mbuf * m,int len)1666 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1667 {
1668 struct vtnet_softc *sc;
1669 struct mbuf *m_new;
1670 int error;
1671
1672 sc = rxq->vtnrx_sc;
1673
1674 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1675 return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1676
1677 MPASS(m->m_next == NULL);
1678 if (m->m_len < len)
1679 return (EMSGSIZE);
1680
1681 m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1682 if (m_new == NULL)
1683 return (ENOBUFS);
1684
1685 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1686 if (error) {
1687 sc->vtnet_stats.rx_enq_replacement_failed++;
1688 m_freem(m_new);
1689 } else
1690 m->m_len = len;
1691
1692 return (error);
1693 }
1694
1695 static int
vtnet_rxq_enqueue_buf(struct vtnet_rxq * rxq,struct mbuf * m)1696 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1697 {
1698 struct vtnet_softc *sc;
1699 struct sglist *sg;
1700 int header_inlined, error;
1701
1702 sc = rxq->vtnrx_sc;
1703 sg = rxq->vtnrx_sg;
1704
1705 KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1706 ("%s: mbuf chain without LRO_NOMRG", __func__));
1707 VTNET_RXQ_LOCK_ASSERT(rxq);
1708
1709 sglist_reset(sg);
1710 header_inlined = vtnet_modern(sc) ||
1711 (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1712
1713 /*
1714 * Note: The mbuf has been already adjusted when we allocate it if we
1715 * have to do strict alignment.
1716 */
1717 if (header_inlined)
1718 error = sglist_append_mbuf(sg, m);
1719 else {
1720 struct vtnet_rx_header *rxhdr =
1721 mtod(m, struct vtnet_rx_header *);
1722 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1723
1724 /* Append the header and remaining mbuf data. */
1725 error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1726 if (error)
1727 return (error);
1728 error = sglist_append(sg, &rxhdr[1],
1729 m->m_len - sizeof(struct vtnet_rx_header));
1730 if (error)
1731 return (error);
1732
1733 if (m->m_next != NULL)
1734 error = sglist_append_mbuf(sg, m->m_next);
1735 }
1736
1737 if (error)
1738 return (error);
1739
1740 return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1741 }
1742
1743 static int
vtnet_rxq_new_buf(struct vtnet_rxq * rxq)1744 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1745 {
1746 struct vtnet_softc *sc;
1747 struct mbuf *m;
1748 int error;
1749
1750 sc = rxq->vtnrx_sc;
1751
1752 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1753 if (m == NULL)
1754 return (ENOBUFS);
1755
1756 error = vtnet_rxq_enqueue_buf(rxq, m);
1757 if (error)
1758 m_freem(m);
1759
1760 return (error);
1761 }
1762
1763 static int
vtnet_rxq_csum_needs_csum(struct vtnet_rxq * rxq,struct mbuf * m,uint16_t etype,int hoff,struct virtio_net_hdr * hdr)1764 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1765 int hoff, struct virtio_net_hdr *hdr)
1766 {
1767 struct vtnet_softc *sc;
1768 int error;
1769
1770 sc = rxq->vtnrx_sc;
1771
1772 /*
1773 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1774 * not have an analogous CSUM flag. The checksum has been validated,
1775 * but is incomplete (TCP/UDP pseudo header).
1776 *
1777 * The packet is likely from another VM on the same host that itself
1778 * performed checksum offloading so Tx/Rx is basically a memcpy and
1779 * the checksum has little value.
1780 *
1781 * Default to receiving the packet as-is for performance reasons, but
1782 * this can cause issues if the packet is to be forwarded because it
1783 * does not contain a valid checksum. This patch may be helpful:
1784 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1785 * compute the checksum if requested.
1786 *
1787 * BMV: Need to add an CSUM_PARTIAL flag?
1788 */
1789 if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1790 error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1791 return (error);
1792 }
1793
1794 /*
1795 * Compute the checksum in the driver so the packet will contain a
1796 * valid checksum. The checksum is at csum_offset from csum_start.
1797 */
1798 switch (etype) {
1799 #if defined(INET) || defined(INET6)
1800 case ETHERTYPE_IP:
1801 case ETHERTYPE_IPV6: {
1802 int csum_off, csum_end;
1803 uint16_t csum;
1804
1805 csum_off = hdr->csum_start + hdr->csum_offset;
1806 csum_end = csum_off + sizeof(uint16_t);
1807
1808 /* Assume checksum will be in the first mbuf. */
1809 if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1810 return (1);
1811
1812 /*
1813 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1814 * checksum and write it at the specified offset. We could
1815 * try to verify the packet: csum_start should probably
1816 * correspond to the start of the TCP/UDP header.
1817 *
1818 * BMV: Need to properly handle UDP with zero checksum. Is
1819 * the IPv4 header checksum implicitly validated?
1820 */
1821 csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1822 *(uint16_t *)(mtodo(m, csum_off)) = csum;
1823 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1824 m->m_pkthdr.csum_data = 0xFFFF;
1825 break;
1826 }
1827 #endif
1828 default:
1829 sc->vtnet_stats.rx_csum_bad_ethtype++;
1830 return (1);
1831 }
1832
1833 return (0);
1834 }
1835
1836 static int
vtnet_rxq_csum_data_valid(struct vtnet_rxq * rxq,struct mbuf * m,uint16_t etype,int hoff,struct virtio_net_hdr * hdr __unused)1837 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1838 uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1839 {
1840 #if 0
1841 struct vtnet_softc *sc;
1842 #endif
1843 int protocol;
1844
1845 #if 0
1846 sc = rxq->vtnrx_sc;
1847 #endif
1848
1849 switch (etype) {
1850 #if defined(INET)
1851 case ETHERTYPE_IP:
1852 if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1853 protocol = IPPROTO_DONE;
1854 else {
1855 struct ip *ip = (struct ip *)(m->m_data + hoff);
1856 protocol = ip->ip_p;
1857 }
1858 break;
1859 #endif
1860 #if defined(INET6)
1861 case ETHERTYPE_IPV6:
1862 if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1863 || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1864 protocol = IPPROTO_DONE;
1865 break;
1866 #endif
1867 default:
1868 protocol = IPPROTO_DONE;
1869 break;
1870 }
1871
1872 switch (protocol) {
1873 case IPPROTO_TCP:
1874 case IPPROTO_UDP:
1875 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1876 m->m_pkthdr.csum_data = 0xFFFF;
1877 break;
1878 default:
1879 /*
1880 * FreeBSD does not support checksum offloading of this
1881 * protocol. Let the stack re-verify the checksum later
1882 * if the protocol is supported.
1883 */
1884 #if 0
1885 if_printf(sc->vtnet_ifp,
1886 "%s: checksum offload of unsupported protocol "
1887 "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1888 __func__, etype, protocol, hdr->csum_start,
1889 hdr->csum_offset);
1890 #endif
1891 break;
1892 }
1893
1894 return (0);
1895 }
1896
1897 static int
vtnet_rxq_csum(struct vtnet_rxq * rxq,struct mbuf * m,struct virtio_net_hdr * hdr)1898 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1899 struct virtio_net_hdr *hdr)
1900 {
1901 const struct ether_header *eh;
1902 int hoff;
1903 uint16_t etype;
1904
1905 eh = mtod(m, const struct ether_header *);
1906 etype = ntohs(eh->ether_type);
1907 if (etype == ETHERTYPE_VLAN) {
1908 /* TODO BMV: Handle QinQ. */
1909 const struct ether_vlan_header *evh =
1910 mtod(m, const struct ether_vlan_header *);
1911 etype = ntohs(evh->evl_proto);
1912 hoff = sizeof(struct ether_vlan_header);
1913 } else
1914 hoff = sizeof(struct ether_header);
1915
1916 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1917 return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1918 else /* VIRTIO_NET_HDR_F_DATA_VALID */
1919 return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1920 }
1921
1922 static void
vtnet_rxq_discard_merged_bufs(struct vtnet_rxq * rxq,int nbufs)1923 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1924 {
1925 struct mbuf *m;
1926
1927 while (--nbufs > 0) {
1928 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1929 if (m == NULL)
1930 break;
1931 vtnet_rxq_discard_buf(rxq, m);
1932 }
1933 }
1934
1935 static void
vtnet_rxq_discard_buf(struct vtnet_rxq * rxq,struct mbuf * m)1936 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1937 {
1938 int error __diagused;
1939
1940 /*
1941 * Requeue the discarded mbuf. This should always be successful
1942 * since it was just dequeued.
1943 */
1944 error = vtnet_rxq_enqueue_buf(rxq, m);
1945 KASSERT(error == 0,
1946 ("%s: cannot requeue discarded mbuf %d", __func__, error));
1947 }
1948
1949 static int
vtnet_rxq_merged_eof(struct vtnet_rxq * rxq,struct mbuf * m_head,int nbufs)1950 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1951 {
1952 struct vtnet_softc *sc;
1953 struct virtqueue *vq;
1954 struct mbuf *m_tail;
1955
1956 sc = rxq->vtnrx_sc;
1957 vq = rxq->vtnrx_vq;
1958 m_tail = m_head;
1959
1960 while (--nbufs > 0) {
1961 struct mbuf *m;
1962 uint32_t len;
1963
1964 m = virtqueue_dequeue(vq, &len);
1965 if (m == NULL) {
1966 rxq->vtnrx_stats.vrxs_ierrors++;
1967 goto fail;
1968 }
1969
1970 if (vtnet_rxq_new_buf(rxq) != 0) {
1971 rxq->vtnrx_stats.vrxs_iqdrops++;
1972 vtnet_rxq_discard_buf(rxq, m);
1973 if (nbufs > 1)
1974 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1975 goto fail;
1976 }
1977
1978 if (m->m_len < len)
1979 len = m->m_len;
1980
1981 m->m_len = len;
1982 m->m_flags &= ~M_PKTHDR;
1983
1984 m_head->m_pkthdr.len += len;
1985 m_tail->m_next = m;
1986 m_tail = m;
1987 }
1988
1989 return (0);
1990
1991 fail:
1992 sc->vtnet_stats.rx_mergeable_failed++;
1993 m_freem(m_head);
1994
1995 return (1);
1996 }
1997
1998 #if defined(INET) || defined(INET6)
1999 static int
vtnet_lro_rx(struct vtnet_rxq * rxq,struct mbuf * m)2000 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
2001 {
2002 struct lro_ctrl *lro;
2003
2004 lro = &rxq->vtnrx_lro;
2005
2006 if (lro->lro_mbuf_max != 0) {
2007 tcp_lro_queue_mbuf(lro, m);
2008 return (0);
2009 }
2010
2011 return (tcp_lro_rx(lro, m, 0));
2012 }
2013 #endif
2014
2015 static void
vtnet_rxq_input(struct vtnet_rxq * rxq,struct mbuf * m,struct virtio_net_hdr * hdr)2016 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2017 struct virtio_net_hdr *hdr)
2018 {
2019 struct vtnet_softc *sc;
2020 if_t ifp;
2021
2022 sc = rxq->vtnrx_sc;
2023 ifp = sc->vtnet_ifp;
2024
2025 if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2026 struct ether_header *eh = mtod(m, struct ether_header *);
2027 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2028 vtnet_vlan_tag_remove(m);
2029 /*
2030 * With the 802.1Q header removed, update the
2031 * checksum starting location accordingly.
2032 */
2033 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2034 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2035 }
2036 }
2037
2038 m->m_pkthdr.flowid = rxq->vtnrx_id;
2039 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2040
2041 if (hdr->flags &
2042 (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2043 if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2044 rxq->vtnrx_stats.vrxs_csum++;
2045 else
2046 rxq->vtnrx_stats.vrxs_csum_failed++;
2047 }
2048
2049 if (hdr->gso_size != 0) {
2050 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2051 case VIRTIO_NET_HDR_GSO_TCPV4:
2052 case VIRTIO_NET_HDR_GSO_TCPV6:
2053 m->m_pkthdr.lro_nsegs =
2054 howmany(m->m_pkthdr.len, hdr->gso_size);
2055 rxq->vtnrx_stats.vrxs_host_lro++;
2056 break;
2057 }
2058 }
2059
2060 rxq->vtnrx_stats.vrxs_ipackets++;
2061 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2062
2063 #if defined(INET) || defined(INET6)
2064 if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2065 if (vtnet_lro_rx(rxq, m) == 0)
2066 return;
2067 }
2068 #endif
2069
2070 if_input(ifp, m);
2071 }
2072
2073 static int
vtnet_rxq_eof(struct vtnet_rxq * rxq)2074 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2075 {
2076 struct virtio_net_hdr lhdr, *hdr;
2077 struct vtnet_softc *sc;
2078 if_t ifp;
2079 struct virtqueue *vq;
2080 int deq, count;
2081
2082 sc = rxq->vtnrx_sc;
2083 vq = rxq->vtnrx_vq;
2084 ifp = sc->vtnet_ifp;
2085 deq = 0;
2086 count = sc->vtnet_rx_process_limit;
2087
2088 VTNET_RXQ_LOCK_ASSERT(rxq);
2089
2090 CURVNET_SET(if_getvnet(ifp));
2091 while (count-- > 0) {
2092 struct mbuf *m;
2093 uint32_t len, nbufs, adjsz;
2094
2095 m = virtqueue_dequeue(vq, &len);
2096 if (m == NULL)
2097 break;
2098 deq++;
2099
2100 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2101 rxq->vtnrx_stats.vrxs_ierrors++;
2102 vtnet_rxq_discard_buf(rxq, m);
2103 continue;
2104 }
2105
2106 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2107 struct virtio_net_hdr_mrg_rxbuf *mhdr =
2108 mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2109 kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2110 nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2111 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2112 } else if (vtnet_modern(sc)) {
2113 nbufs = 1; /* num_buffers is always 1 */
2114 adjsz = sizeof(struct virtio_net_hdr_v1);
2115 } else {
2116 nbufs = 1;
2117 adjsz = sizeof(struct vtnet_rx_header);
2118 /*
2119 * Account for our gap between the header and start of
2120 * data to keep the segments separated.
2121 */
2122 len += VTNET_RX_HEADER_PAD;
2123 }
2124
2125 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2126 rxq->vtnrx_stats.vrxs_iqdrops++;
2127 vtnet_rxq_discard_buf(rxq, m);
2128 if (nbufs > 1)
2129 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2130 continue;
2131 }
2132
2133 m->m_pkthdr.len = len;
2134 m->m_pkthdr.rcvif = ifp;
2135 m->m_pkthdr.csum_flags = 0;
2136
2137 if (nbufs > 1) {
2138 /* Dequeue the rest of chain. */
2139 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2140 continue;
2141 }
2142
2143 kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2144
2145 /*
2146 * Save an endian swapped version of the header prior to it
2147 * being stripped. The header is always at the start of the
2148 * mbuf data. num_buffers was already saved (and not needed)
2149 * so use the standard header.
2150 */
2151 hdr = mtod(m, struct virtio_net_hdr *);
2152 lhdr.flags = hdr->flags;
2153 lhdr.gso_type = hdr->gso_type;
2154 lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2155 lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2156 lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2157 lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2158 m_adj(m, adjsz);
2159
2160 if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2161 pfil_return_t pfil;
2162
2163 pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2164 switch (pfil) {
2165 case PFIL_DROPPED:
2166 case PFIL_CONSUMED:
2167 continue;
2168 default:
2169 KASSERT(pfil == PFIL_PASS,
2170 ("Filter returned %d!", pfil));
2171 }
2172 }
2173
2174 vtnet_rxq_input(rxq, m, &lhdr);
2175 }
2176
2177 if (deq > 0) {
2178 #if defined(INET) || defined(INET6)
2179 if (vtnet_software_lro(sc))
2180 tcp_lro_flush_all(&rxq->vtnrx_lro);
2181 #endif
2182 virtqueue_notify(vq);
2183 }
2184 CURVNET_RESTORE();
2185
2186 return (count > 0 ? 0 : EAGAIN);
2187 }
2188
2189 static void
vtnet_rx_vq_process(struct vtnet_rxq * rxq,int tries)2190 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2191 {
2192 struct vtnet_softc *sc;
2193 if_t ifp;
2194 u_int more;
2195 #ifdef DEV_NETMAP
2196 int nmirq;
2197 #endif /* DEV_NETMAP */
2198
2199 sc = rxq->vtnrx_sc;
2200 ifp = sc->vtnet_ifp;
2201
2202 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2203 /*
2204 * Ignore this interrupt. Either this is a spurious interrupt
2205 * or multiqueue without per-VQ MSIX so every queue needs to
2206 * be polled (a brain dead configuration we could try harder
2207 * to avoid).
2208 */
2209 vtnet_rxq_disable_intr(rxq);
2210 return;
2211 }
2212
2213 VTNET_RXQ_LOCK(rxq);
2214
2215 #ifdef DEV_NETMAP
2216 /*
2217 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2218 * This is not necessary to serialize the access to the RX vq, but
2219 * rather to avoid races that may happen if this interface is
2220 * attached to a VALE switch, which would cause received packets
2221 * to stall in the RX queue (nm_kr_tryget() could find the kring
2222 * busy when called from netmap_bwrap_intr_notify()).
2223 */
2224 nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2225 if (nmirq != NM_IRQ_PASS) {
2226 VTNET_RXQ_UNLOCK(rxq);
2227 if (nmirq == NM_IRQ_RESCHED) {
2228 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2229 }
2230 return;
2231 }
2232 #endif /* DEV_NETMAP */
2233
2234 again:
2235 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2236 VTNET_RXQ_UNLOCK(rxq);
2237 return;
2238 }
2239
2240 more = vtnet_rxq_eof(rxq);
2241 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2242 if (!more)
2243 vtnet_rxq_disable_intr(rxq);
2244 /*
2245 * This is an occasional condition or race (when !more),
2246 * so retry a few times before scheduling the taskqueue.
2247 */
2248 if (tries-- > 0)
2249 goto again;
2250
2251 rxq->vtnrx_stats.vrxs_rescheduled++;
2252 VTNET_RXQ_UNLOCK(rxq);
2253 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2254 } else
2255 VTNET_RXQ_UNLOCK(rxq);
2256 }
2257
2258 static void
vtnet_rx_vq_intr(void * xrxq)2259 vtnet_rx_vq_intr(void *xrxq)
2260 {
2261 struct vtnet_rxq *rxq;
2262
2263 rxq = xrxq;
2264 vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2265 }
2266
2267 static void
vtnet_rxq_tq_intr(void * xrxq,int pending __unused)2268 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2269 {
2270 struct vtnet_rxq *rxq;
2271
2272 rxq = xrxq;
2273 vtnet_rx_vq_process(rxq, 0);
2274 }
2275
2276 static int
vtnet_txq_intr_threshold(struct vtnet_txq * txq)2277 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2278 {
2279 struct vtnet_softc *sc;
2280 int threshold;
2281
2282 sc = txq->vtntx_sc;
2283
2284 /*
2285 * The Tx interrupt is disabled until the queue free count falls
2286 * below our threshold. Completed frames are drained from the Tx
2287 * virtqueue before transmitting new frames and in the watchdog
2288 * callout, so the frequency of Tx interrupts is greatly reduced,
2289 * at the cost of not freeing mbufs as quickly as they otherwise
2290 * would be.
2291 */
2292 threshold = virtqueue_size(txq->vtntx_vq) / 4;
2293
2294 /*
2295 * Without indirect descriptors, leave enough room for the most
2296 * segments we handle.
2297 */
2298 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2299 threshold < sc->vtnet_tx_nsegs)
2300 threshold = sc->vtnet_tx_nsegs;
2301
2302 return (threshold);
2303 }
2304
2305 static int
vtnet_txq_below_threshold(struct vtnet_txq * txq)2306 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2307 {
2308 struct virtqueue *vq;
2309
2310 vq = txq->vtntx_vq;
2311
2312 return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2313 }
2314
2315 static int
vtnet_txq_notify(struct vtnet_txq * txq)2316 vtnet_txq_notify(struct vtnet_txq *txq)
2317 {
2318 struct virtqueue *vq;
2319
2320 vq = txq->vtntx_vq;
2321
2322 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2323 virtqueue_notify(vq);
2324
2325 if (vtnet_txq_enable_intr(txq) == 0)
2326 return (0);
2327
2328 /*
2329 * Drain frames that were completed since last checked. If this
2330 * causes the queue to go above the threshold, the caller should
2331 * continue transmitting.
2332 */
2333 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2334 virtqueue_disable_intr(vq);
2335 return (1);
2336 }
2337
2338 return (0);
2339 }
2340
2341 static void
vtnet_txq_free_mbufs(struct vtnet_txq * txq)2342 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2343 {
2344 struct virtqueue *vq;
2345 struct vtnet_tx_header *txhdr;
2346 int last;
2347 #ifdef DEV_NETMAP
2348 struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2349 txq->vtntx_id, NR_TX);
2350 #else /* !DEV_NETMAP */
2351 void *kring = NULL;
2352 #endif /* !DEV_NETMAP */
2353
2354 vq = txq->vtntx_vq;
2355 last = 0;
2356
2357 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2358 if (kring == NULL) {
2359 m_freem(txhdr->vth_mbuf);
2360 uma_zfree(vtnet_tx_header_zone, txhdr);
2361 }
2362 }
2363
2364 KASSERT(virtqueue_empty(vq),
2365 ("%s: mbufs remaining in tx queue %p", __func__, txq));
2366 }
2367
2368 /*
2369 * BMV: This can go away once we finally have offsets in the mbuf header.
2370 */
2371 static int
vtnet_txq_offload_ctx(struct vtnet_txq * txq,struct mbuf * m,int * etype,int * proto,int * start)2372 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2373 int *proto, int *start)
2374 {
2375 struct vtnet_softc *sc;
2376 struct ether_vlan_header *evh;
2377 #if defined(INET) || defined(INET6)
2378 int offset;
2379 #endif
2380
2381 sc = txq->vtntx_sc;
2382
2383 evh = mtod(m, struct ether_vlan_header *);
2384 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2385 /* BMV: We should handle nested VLAN tags too. */
2386 *etype = ntohs(evh->evl_proto);
2387 #if defined(INET) || defined(INET6)
2388 offset = sizeof(struct ether_vlan_header);
2389 #endif
2390 } else {
2391 *etype = ntohs(evh->evl_encap_proto);
2392 #if defined(INET) || defined(INET6)
2393 offset = sizeof(struct ether_header);
2394 #endif
2395 }
2396
2397 switch (*etype) {
2398 #if defined(INET)
2399 case ETHERTYPE_IP: {
2400 struct ip *ip, iphdr;
2401 if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2402 m_copydata(m, offset, sizeof(struct ip),
2403 (caddr_t) &iphdr);
2404 ip = &iphdr;
2405 } else
2406 ip = (struct ip *)(m->m_data + offset);
2407 *proto = ip->ip_p;
2408 *start = offset + (ip->ip_hl << 2);
2409 break;
2410 }
2411 #endif
2412 #if defined(INET6)
2413 case ETHERTYPE_IPV6:
2414 *proto = -1;
2415 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2416 /* Assert the network stack sent us a valid packet. */
2417 KASSERT(*start > offset,
2418 ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2419 *start, offset, *proto));
2420 break;
2421 #endif
2422 default:
2423 sc->vtnet_stats.tx_csum_unknown_ethtype++;
2424 return (EINVAL);
2425 }
2426
2427 return (0);
2428 }
2429
2430 static int
vtnet_txq_offload_tso(struct vtnet_txq * txq,struct mbuf * m,int eth_type,int offset,struct virtio_net_hdr * hdr)2431 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2432 int offset, struct virtio_net_hdr *hdr)
2433 {
2434 static struct timeval lastecn;
2435 static int curecn;
2436 struct vtnet_softc *sc;
2437 struct tcphdr *tcp, tcphdr;
2438
2439 sc = txq->vtntx_sc;
2440
2441 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2442 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2443 tcp = &tcphdr;
2444 } else
2445 tcp = (struct tcphdr *)(m->m_data + offset);
2446
2447 hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2448 hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2449 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2450 VIRTIO_NET_HDR_GSO_TCPV6;
2451
2452 if (__predict_false(tcp_get_flags(tcp) & TH_CWR)) {
2453 /*
2454 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2455 * FreeBSD, ECN support is not on a per-interface basis,
2456 * but globally via the net.inet.tcp.ecn.enable sysctl
2457 * knob. The default is off.
2458 */
2459 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2460 if (ppsratecheck(&lastecn, &curecn, 1))
2461 if_printf(sc->vtnet_ifp,
2462 "TSO with ECN not negotiated with host\n");
2463 return (ENOTSUP);
2464 }
2465 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2466 }
2467
2468 txq->vtntx_stats.vtxs_tso++;
2469
2470 return (0);
2471 }
2472
2473 static struct mbuf *
vtnet_txq_offload(struct vtnet_txq * txq,struct mbuf * m,struct virtio_net_hdr * hdr)2474 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2475 struct virtio_net_hdr *hdr)
2476 {
2477 struct vtnet_softc *sc;
2478 int flags, etype, csum_start, proto, error;
2479
2480 sc = txq->vtntx_sc;
2481 flags = m->m_pkthdr.csum_flags;
2482
2483 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2484 if (error)
2485 goto drop;
2486
2487 if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2488 /* Sanity check the parsed mbuf matches the offload flags. */
2489 if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2490 etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2491 && etype != ETHERTYPE_IPV6))) {
2492 sc->vtnet_stats.tx_csum_proto_mismatch++;
2493 goto drop;
2494 }
2495
2496 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2497 hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2498 hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2499 txq->vtntx_stats.vtxs_csum++;
2500 }
2501
2502 if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2503 /*
2504 * Sanity check the parsed mbuf IP protocol is TCP, and
2505 * VirtIO TSO reqires the checksum offloading above.
2506 */
2507 if (__predict_false(proto != IPPROTO_TCP)) {
2508 sc->vtnet_stats.tx_tso_not_tcp++;
2509 goto drop;
2510 } else if (__predict_false((hdr->flags &
2511 VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2512 sc->vtnet_stats.tx_tso_without_csum++;
2513 goto drop;
2514 }
2515
2516 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2517 if (error)
2518 goto drop;
2519 }
2520
2521 return (m);
2522
2523 drop:
2524 m_freem(m);
2525 return (NULL);
2526 }
2527
2528 static int
vtnet_txq_enqueue_buf(struct vtnet_txq * txq,struct mbuf ** m_head,struct vtnet_tx_header * txhdr)2529 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2530 struct vtnet_tx_header *txhdr)
2531 {
2532 struct vtnet_softc *sc;
2533 struct virtqueue *vq;
2534 struct sglist *sg;
2535 struct mbuf *m;
2536 int error;
2537
2538 sc = txq->vtntx_sc;
2539 vq = txq->vtntx_vq;
2540 sg = txq->vtntx_sg;
2541 m = *m_head;
2542
2543 sglist_reset(sg);
2544 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2545 if (error != 0 || sg->sg_nseg != 1) {
2546 KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2547 __func__, error, sg->sg_nseg));
2548 goto fail;
2549 }
2550
2551 error = sglist_append_mbuf(sg, m);
2552 if (error) {
2553 m = m_defrag(m, M_NOWAIT);
2554 if (m == NULL)
2555 goto fail;
2556
2557 *m_head = m;
2558 sc->vtnet_stats.tx_defragged++;
2559
2560 error = sglist_append_mbuf(sg, m);
2561 if (error)
2562 goto fail;
2563 }
2564
2565 txhdr->vth_mbuf = m;
2566 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2567
2568 return (error);
2569
2570 fail:
2571 sc->vtnet_stats.tx_defrag_failed++;
2572 m_freem(*m_head);
2573 *m_head = NULL;
2574
2575 return (ENOBUFS);
2576 }
2577
2578 static int
vtnet_txq_encap(struct vtnet_txq * txq,struct mbuf ** m_head,int flags)2579 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2580 {
2581 struct vtnet_tx_header *txhdr;
2582 struct virtio_net_hdr *hdr;
2583 struct mbuf *m;
2584 int error;
2585
2586 m = *m_head;
2587 M_ASSERTPKTHDR(m);
2588
2589 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2590 if (txhdr == NULL) {
2591 m_freem(m);
2592 *m_head = NULL;
2593 return (ENOMEM);
2594 }
2595
2596 /*
2597 * Always use the non-mergeable header, regardless if mergable headers
2598 * were negotiated, because for transmit num_buffers is always zero.
2599 * The vtnet_hdr_size is used to enqueue the right header size segment.
2600 */
2601 hdr = &txhdr->vth_uhdr.hdr;
2602
2603 if (m->m_flags & M_VLANTAG) {
2604 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2605 if ((*m_head = m) == NULL) {
2606 error = ENOBUFS;
2607 goto fail;
2608 }
2609 m->m_flags &= ~M_VLANTAG;
2610 }
2611
2612 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2613 m = vtnet_txq_offload(txq, m, hdr);
2614 if ((*m_head = m) == NULL) {
2615 error = ENOBUFS;
2616 goto fail;
2617 }
2618 }
2619
2620 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2621 fail:
2622 if (error)
2623 uma_zfree(vtnet_tx_header_zone, txhdr);
2624
2625 return (error);
2626 }
2627
2628
2629 static void
vtnet_start_locked(struct vtnet_txq * txq,if_t ifp)2630 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2631 {
2632 struct vtnet_softc *sc;
2633 struct virtqueue *vq;
2634 struct mbuf *m0;
2635 int tries, enq;
2636
2637 sc = txq->vtntx_sc;
2638 vq = txq->vtntx_vq;
2639 tries = 0;
2640
2641 VTNET_TXQ_LOCK_ASSERT(txq);
2642
2643 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2644 sc->vtnet_link_active == 0)
2645 return;
2646
2647 vtnet_txq_eof(txq);
2648
2649 again:
2650 enq = 0;
2651
2652 while (!if_sendq_empty(ifp)) {
2653 if (virtqueue_full(vq))
2654 break;
2655
2656 m0 = if_dequeue(ifp);
2657 if (m0 == NULL)
2658 break;
2659
2660 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2661 if (m0 != NULL)
2662 if_sendq_prepend(ifp, m0);
2663 break;
2664 }
2665
2666 enq++;
2667 ETHER_BPF_MTAP(ifp, m0);
2668 }
2669
2670 if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2671 if (tries++ < VTNET_NOTIFY_RETRIES)
2672 goto again;
2673
2674 txq->vtntx_stats.vtxs_rescheduled++;
2675 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2676 }
2677 }
2678
2679 static void
vtnet_start(if_t ifp)2680 vtnet_start(if_t ifp)
2681 {
2682 struct vtnet_softc *sc;
2683 struct vtnet_txq *txq;
2684
2685 sc = if_getsoftc(ifp);
2686 txq = &sc->vtnet_txqs[0];
2687
2688 VTNET_TXQ_LOCK(txq);
2689 vtnet_start_locked(txq, ifp);
2690 VTNET_TXQ_UNLOCK(txq);
2691 }
2692
2693
2694 static int
vtnet_txq_mq_start_locked(struct vtnet_txq * txq,struct mbuf * m)2695 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2696 {
2697 struct vtnet_softc *sc;
2698 struct virtqueue *vq;
2699 struct buf_ring *br;
2700 if_t ifp;
2701 int enq, tries, error;
2702
2703 sc = txq->vtntx_sc;
2704 vq = txq->vtntx_vq;
2705 br = txq->vtntx_br;
2706 ifp = sc->vtnet_ifp;
2707 tries = 0;
2708 error = 0;
2709
2710 VTNET_TXQ_LOCK_ASSERT(txq);
2711
2712 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2713 sc->vtnet_link_active == 0) {
2714 if (m != NULL)
2715 error = drbr_enqueue(ifp, br, m);
2716 return (error);
2717 }
2718
2719 if (m != NULL) {
2720 error = drbr_enqueue(ifp, br, m);
2721 if (error)
2722 return (error);
2723 }
2724
2725 vtnet_txq_eof(txq);
2726
2727 again:
2728 enq = 0;
2729
2730 while ((m = drbr_peek(ifp, br)) != NULL) {
2731 if (virtqueue_full(vq)) {
2732 drbr_putback(ifp, br, m);
2733 break;
2734 }
2735
2736 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2737 if (m != NULL)
2738 drbr_putback(ifp, br, m);
2739 else
2740 drbr_advance(ifp, br);
2741 break;
2742 }
2743 drbr_advance(ifp, br);
2744
2745 enq++;
2746 ETHER_BPF_MTAP(ifp, m);
2747 }
2748
2749 if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2750 if (tries++ < VTNET_NOTIFY_RETRIES)
2751 goto again;
2752
2753 txq->vtntx_stats.vtxs_rescheduled++;
2754 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2755 }
2756
2757 return (0);
2758 }
2759
2760 static int
vtnet_txq_mq_start(if_t ifp,struct mbuf * m)2761 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2762 {
2763 struct vtnet_softc *sc;
2764 struct vtnet_txq *txq;
2765 int i, npairs, error;
2766
2767 sc = if_getsoftc(ifp);
2768 npairs = sc->vtnet_act_vq_pairs;
2769
2770 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2771 i = m->m_pkthdr.flowid % npairs;
2772 else
2773 i = curcpu % npairs;
2774
2775 txq = &sc->vtnet_txqs[i];
2776
2777 if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2778 error = vtnet_txq_mq_start_locked(txq, m);
2779 VTNET_TXQ_UNLOCK(txq);
2780 } else {
2781 error = drbr_enqueue(ifp, txq->vtntx_br, m);
2782 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2783 }
2784
2785 return (error);
2786 }
2787
2788 static void
vtnet_txq_tq_deferred(void * xtxq,int pending __unused)2789 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2790 {
2791 struct vtnet_softc *sc;
2792 struct vtnet_txq *txq;
2793
2794 txq = xtxq;
2795 sc = txq->vtntx_sc;
2796
2797 VTNET_TXQ_LOCK(txq);
2798 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2799 vtnet_txq_mq_start_locked(txq, NULL);
2800 VTNET_TXQ_UNLOCK(txq);
2801 }
2802
2803
2804 static void
vtnet_txq_start(struct vtnet_txq * txq)2805 vtnet_txq_start(struct vtnet_txq *txq)
2806 {
2807 struct vtnet_softc *sc;
2808 if_t ifp;
2809
2810 sc = txq->vtntx_sc;
2811 ifp = sc->vtnet_ifp;
2812
2813 if (!VTNET_ALTQ_ENABLED) {
2814 if (!drbr_empty(ifp, txq->vtntx_br))
2815 vtnet_txq_mq_start_locked(txq, NULL);
2816 } else {
2817 if (!if_sendq_empty(ifp))
2818 vtnet_start_locked(txq, ifp);
2819
2820 }
2821 }
2822
2823 static void
vtnet_txq_tq_intr(void * xtxq,int pending __unused)2824 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2825 {
2826 struct vtnet_softc *sc;
2827 struct vtnet_txq *txq;
2828 if_t ifp;
2829
2830 txq = xtxq;
2831 sc = txq->vtntx_sc;
2832 ifp = sc->vtnet_ifp;
2833
2834 VTNET_TXQ_LOCK(txq);
2835
2836 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2837 VTNET_TXQ_UNLOCK(txq);
2838 return;
2839 }
2840
2841 vtnet_txq_eof(txq);
2842 vtnet_txq_start(txq);
2843
2844 VTNET_TXQ_UNLOCK(txq);
2845 }
2846
2847 static int
vtnet_txq_eof(struct vtnet_txq * txq)2848 vtnet_txq_eof(struct vtnet_txq *txq)
2849 {
2850 struct virtqueue *vq;
2851 struct vtnet_tx_header *txhdr;
2852 struct mbuf *m;
2853 int deq;
2854
2855 vq = txq->vtntx_vq;
2856 deq = 0;
2857 VTNET_TXQ_LOCK_ASSERT(txq);
2858
2859 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2860 m = txhdr->vth_mbuf;
2861 deq++;
2862
2863 txq->vtntx_stats.vtxs_opackets++;
2864 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2865 if (m->m_flags & M_MCAST)
2866 txq->vtntx_stats.vtxs_omcasts++;
2867
2868 m_freem(m);
2869 uma_zfree(vtnet_tx_header_zone, txhdr);
2870 }
2871
2872 if (virtqueue_empty(vq))
2873 txq->vtntx_watchdog = 0;
2874
2875 return (deq);
2876 }
2877
2878 static void
vtnet_tx_vq_intr(void * xtxq)2879 vtnet_tx_vq_intr(void *xtxq)
2880 {
2881 struct vtnet_softc *sc;
2882 struct vtnet_txq *txq;
2883 if_t ifp;
2884
2885 txq = xtxq;
2886 sc = txq->vtntx_sc;
2887 ifp = sc->vtnet_ifp;
2888
2889 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2890 /*
2891 * Ignore this interrupt. Either this is a spurious interrupt
2892 * or multiqueue without per-VQ MSIX so every queue needs to
2893 * be polled (a brain dead configuration we could try harder
2894 * to avoid).
2895 */
2896 vtnet_txq_disable_intr(txq);
2897 return;
2898 }
2899
2900 #ifdef DEV_NETMAP
2901 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2902 return;
2903 #endif /* DEV_NETMAP */
2904
2905 VTNET_TXQ_LOCK(txq);
2906
2907 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2908 VTNET_TXQ_UNLOCK(txq);
2909 return;
2910 }
2911
2912 vtnet_txq_eof(txq);
2913 vtnet_txq_start(txq);
2914
2915 VTNET_TXQ_UNLOCK(txq);
2916 }
2917
2918 static void
vtnet_tx_start_all(struct vtnet_softc * sc)2919 vtnet_tx_start_all(struct vtnet_softc *sc)
2920 {
2921 struct vtnet_txq *txq;
2922 int i;
2923
2924 VTNET_CORE_LOCK_ASSERT(sc);
2925
2926 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2927 txq = &sc->vtnet_txqs[i];
2928
2929 VTNET_TXQ_LOCK(txq);
2930 vtnet_txq_start(txq);
2931 VTNET_TXQ_UNLOCK(txq);
2932 }
2933 }
2934
2935 static void
vtnet_qflush(if_t ifp)2936 vtnet_qflush(if_t ifp)
2937 {
2938 struct vtnet_softc *sc;
2939 struct vtnet_txq *txq;
2940 struct mbuf *m;
2941 int i;
2942
2943 sc = if_getsoftc(ifp);
2944
2945 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2946 txq = &sc->vtnet_txqs[i];
2947
2948 VTNET_TXQ_LOCK(txq);
2949 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2950 m_freem(m);
2951 VTNET_TXQ_UNLOCK(txq);
2952 }
2953
2954 if_qflush(ifp);
2955 }
2956
2957 static int
vtnet_watchdog(struct vtnet_txq * txq)2958 vtnet_watchdog(struct vtnet_txq *txq)
2959 {
2960 if_t ifp;
2961
2962 ifp = txq->vtntx_sc->vtnet_ifp;
2963
2964 VTNET_TXQ_LOCK(txq);
2965 if (txq->vtntx_watchdog == 1) {
2966 /*
2967 * Only drain completed frames if the watchdog is about to
2968 * expire. If any frames were drained, there may be enough
2969 * free descriptors now available to transmit queued frames.
2970 * In that case, the timer will immediately be decremented
2971 * below, but the timeout is generous enough that should not
2972 * be a problem.
2973 */
2974 if (vtnet_txq_eof(txq) != 0)
2975 vtnet_txq_start(txq);
2976 }
2977
2978 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2979 VTNET_TXQ_UNLOCK(txq);
2980 return (0);
2981 }
2982 VTNET_TXQ_UNLOCK(txq);
2983
2984 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2985 return (1);
2986 }
2987
2988 static void
vtnet_accum_stats(struct vtnet_softc * sc,struct vtnet_rxq_stats * rxacc,struct vtnet_txq_stats * txacc)2989 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2990 struct vtnet_txq_stats *txacc)
2991 {
2992
2993 bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2994 bzero(txacc, sizeof(struct vtnet_txq_stats));
2995
2996 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2997 struct vtnet_rxq_stats *rxst;
2998 struct vtnet_txq_stats *txst;
2999
3000 rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
3001 rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
3002 rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
3003 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
3004 rxacc->vrxs_csum += rxst->vrxs_csum;
3005 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
3006 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3007
3008 txst = &sc->vtnet_txqs[i].vtntx_stats;
3009 txacc->vtxs_opackets += txst->vtxs_opackets;
3010 txacc->vtxs_obytes += txst->vtxs_obytes;
3011 txacc->vtxs_csum += txst->vtxs_csum;
3012 txacc->vtxs_tso += txst->vtxs_tso;
3013 txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3014 }
3015 }
3016
3017 static uint64_t
vtnet_get_counter(if_t ifp,ift_counter cnt)3018 vtnet_get_counter(if_t ifp, ift_counter cnt)
3019 {
3020 struct vtnet_softc *sc;
3021 struct vtnet_rxq_stats rxaccum;
3022 struct vtnet_txq_stats txaccum;
3023
3024 sc = if_getsoftc(ifp);
3025 vtnet_accum_stats(sc, &rxaccum, &txaccum);
3026
3027 switch (cnt) {
3028 case IFCOUNTER_IPACKETS:
3029 return (rxaccum.vrxs_ipackets);
3030 case IFCOUNTER_IQDROPS:
3031 return (rxaccum.vrxs_iqdrops);
3032 case IFCOUNTER_IERRORS:
3033 return (rxaccum.vrxs_ierrors);
3034 case IFCOUNTER_OPACKETS:
3035 return (txaccum.vtxs_opackets);
3036 case IFCOUNTER_OBYTES:
3037 if (!VTNET_ALTQ_ENABLED)
3038 return (txaccum.vtxs_obytes);
3039 /* FALLTHROUGH */
3040 case IFCOUNTER_OMCASTS:
3041 if (!VTNET_ALTQ_ENABLED)
3042 return (txaccum.vtxs_omcasts);
3043 /* FALLTHROUGH */
3044 default:
3045 return (if_get_counter_default(ifp, cnt));
3046 }
3047 }
3048
3049 static void
vtnet_tick(void * xsc)3050 vtnet_tick(void *xsc)
3051 {
3052 struct vtnet_softc *sc;
3053 if_t ifp;
3054 int i, timedout;
3055
3056 sc = xsc;
3057 ifp = sc->vtnet_ifp;
3058 timedout = 0;
3059
3060 VTNET_CORE_LOCK_ASSERT(sc);
3061
3062 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3063 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3064
3065 if (timedout != 0) {
3066 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3067 vtnet_init_locked(sc, 0);
3068 } else
3069 callout_schedule(&sc->vtnet_tick_ch, hz);
3070 }
3071
3072 static void
vtnet_start_taskqueues(struct vtnet_softc * sc)3073 vtnet_start_taskqueues(struct vtnet_softc *sc)
3074 {
3075 device_t dev;
3076 struct vtnet_rxq *rxq;
3077 struct vtnet_txq *txq;
3078 int i, error;
3079
3080 dev = sc->vtnet_dev;
3081
3082 /*
3083 * Errors here are very difficult to recover from - we cannot
3084 * easily fail because, if this is during boot, we will hang
3085 * when freeing any successfully started taskqueues because
3086 * the scheduler isn't up yet.
3087 *
3088 * Most drivers just ignore the return value - it only fails
3089 * with ENOMEM so an error is not likely.
3090 */
3091 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3092 rxq = &sc->vtnet_rxqs[i];
3093 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3094 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3095 if (error) {
3096 device_printf(dev, "failed to start rx taskq %d\n",
3097 rxq->vtnrx_id);
3098 }
3099
3100 txq = &sc->vtnet_txqs[i];
3101 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3102 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3103 if (error) {
3104 device_printf(dev, "failed to start tx taskq %d\n",
3105 txq->vtntx_id);
3106 }
3107 }
3108 }
3109
3110 static void
vtnet_free_taskqueues(struct vtnet_softc * sc)3111 vtnet_free_taskqueues(struct vtnet_softc *sc)
3112 {
3113 struct vtnet_rxq *rxq;
3114 struct vtnet_txq *txq;
3115 int i;
3116
3117 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3118 rxq = &sc->vtnet_rxqs[i];
3119 if (rxq->vtnrx_tq != NULL) {
3120 taskqueue_free(rxq->vtnrx_tq);
3121 rxq->vtnrx_tq = NULL;
3122 }
3123
3124 txq = &sc->vtnet_txqs[i];
3125 if (txq->vtntx_tq != NULL) {
3126 taskqueue_free(txq->vtntx_tq);
3127 txq->vtntx_tq = NULL;
3128 }
3129 }
3130 }
3131
3132 static void
vtnet_drain_taskqueues(struct vtnet_softc * sc)3133 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3134 {
3135 struct vtnet_rxq *rxq;
3136 struct vtnet_txq *txq;
3137 int i;
3138
3139 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3140 rxq = &sc->vtnet_rxqs[i];
3141 if (rxq->vtnrx_tq != NULL)
3142 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3143
3144 txq = &sc->vtnet_txqs[i];
3145 if (txq->vtntx_tq != NULL) {
3146 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3147 if (!VTNET_ALTQ_ENABLED)
3148 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3149 }
3150 }
3151 }
3152
3153 static void
vtnet_drain_rxtx_queues(struct vtnet_softc * sc)3154 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3155 {
3156 struct vtnet_rxq *rxq;
3157 struct vtnet_txq *txq;
3158 int i;
3159
3160 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3161 rxq = &sc->vtnet_rxqs[i];
3162 vtnet_rxq_free_mbufs(rxq);
3163
3164 txq = &sc->vtnet_txqs[i];
3165 vtnet_txq_free_mbufs(txq);
3166 }
3167 }
3168
3169 static void
vtnet_stop_rendezvous(struct vtnet_softc * sc)3170 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3171 {
3172 struct vtnet_rxq *rxq;
3173 struct vtnet_txq *txq;
3174 int i;
3175
3176 VTNET_CORE_LOCK_ASSERT(sc);
3177
3178 /*
3179 * Lock and unlock the per-queue mutex so we known the stop
3180 * state is visible. Doing only the active queues should be
3181 * sufficient, but it does not cost much extra to do all the
3182 * queues.
3183 */
3184 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3185 rxq = &sc->vtnet_rxqs[i];
3186 VTNET_RXQ_LOCK(rxq);
3187 VTNET_RXQ_UNLOCK(rxq);
3188
3189 txq = &sc->vtnet_txqs[i];
3190 VTNET_TXQ_LOCK(txq);
3191 VTNET_TXQ_UNLOCK(txq);
3192 }
3193 }
3194
3195 static void
vtnet_stop(struct vtnet_softc * sc)3196 vtnet_stop(struct vtnet_softc *sc)
3197 {
3198 device_t dev;
3199 if_t ifp;
3200
3201 dev = sc->vtnet_dev;
3202 ifp = sc->vtnet_ifp;
3203
3204 VTNET_CORE_LOCK_ASSERT(sc);
3205
3206 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3207 sc->vtnet_link_active = 0;
3208 callout_stop(&sc->vtnet_tick_ch);
3209
3210 /* Only advisory. */
3211 vtnet_disable_interrupts(sc);
3212
3213 #ifdef DEV_NETMAP
3214 /* Stop any pending txsync/rxsync and disable them. */
3215 netmap_disable_all_rings(ifp);
3216 #endif /* DEV_NETMAP */
3217
3218 /*
3219 * Stop the host adapter. This resets it to the pre-initialized
3220 * state. It will not generate any interrupts until after it is
3221 * reinitialized.
3222 */
3223 virtio_stop(dev);
3224 vtnet_stop_rendezvous(sc);
3225
3226 vtnet_drain_rxtx_queues(sc);
3227 sc->vtnet_act_vq_pairs = 1;
3228 }
3229
3230 static int
vtnet_virtio_reinit(struct vtnet_softc * sc)3231 vtnet_virtio_reinit(struct vtnet_softc *sc)
3232 {
3233 device_t dev;
3234 if_t ifp;
3235 uint64_t features;
3236 int error;
3237
3238 dev = sc->vtnet_dev;
3239 ifp = sc->vtnet_ifp;
3240 features = sc->vtnet_negotiated_features;
3241
3242 /*
3243 * Re-negotiate with the host, removing any disabled receive
3244 * features. Transmit features are disabled only on our side
3245 * via if_capenable and if_hwassist.
3246 */
3247
3248 if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3249 features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3250
3251 if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3252 features &= ~VTNET_LRO_FEATURES;
3253
3254 if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3255 features &= ~VIRTIO_NET_F_CTRL_VLAN;
3256
3257 error = virtio_reinit(dev, features);
3258 if (error) {
3259 device_printf(dev, "virtio reinit error %d\n", error);
3260 return (error);
3261 }
3262
3263 sc->vtnet_features = features;
3264 virtio_reinit_complete(dev);
3265
3266 return (0);
3267 }
3268
3269 static void
vtnet_init_rx_filters(struct vtnet_softc * sc)3270 vtnet_init_rx_filters(struct vtnet_softc *sc)
3271 {
3272 if_t ifp;
3273
3274 ifp = sc->vtnet_ifp;
3275
3276 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3277 vtnet_rx_filter(sc);
3278 vtnet_rx_filter_mac(sc);
3279 }
3280
3281 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3282 vtnet_rx_filter_vlan(sc);
3283 }
3284
3285 static int
vtnet_init_rx_queues(struct vtnet_softc * sc)3286 vtnet_init_rx_queues(struct vtnet_softc *sc)
3287 {
3288 device_t dev;
3289 if_t ifp;
3290 struct vtnet_rxq *rxq;
3291 int i, clustersz, error;
3292
3293 dev = sc->vtnet_dev;
3294 ifp = sc->vtnet_ifp;
3295
3296 clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3297 sc->vtnet_rx_clustersz = clustersz;
3298
3299 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3300 sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3301 VTNET_MAX_RX_SIZE, clustersz);
3302 KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3303 ("%s: too many rx mbufs %d for %d segments", __func__,
3304 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3305 } else
3306 sc->vtnet_rx_nmbufs = 1;
3307
3308 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3309 rxq = &sc->vtnet_rxqs[i];
3310
3311 /* Hold the lock to satisfy asserts. */
3312 VTNET_RXQ_LOCK(rxq);
3313 error = vtnet_rxq_populate(rxq);
3314 VTNET_RXQ_UNLOCK(rxq);
3315
3316 if (error) {
3317 device_printf(dev, "cannot populate Rx queue %d\n", i);
3318 return (error);
3319 }
3320 }
3321
3322 return (0);
3323 }
3324
3325 static int
vtnet_init_tx_queues(struct vtnet_softc * sc)3326 vtnet_init_tx_queues(struct vtnet_softc *sc)
3327 {
3328 struct vtnet_txq *txq;
3329 int i;
3330
3331 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3332 txq = &sc->vtnet_txqs[i];
3333 txq->vtntx_watchdog = 0;
3334 txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3335 #ifdef DEV_NETMAP
3336 netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3337 #endif /* DEV_NETMAP */
3338 }
3339
3340 return (0);
3341 }
3342
3343 static int
vtnet_init_rxtx_queues(struct vtnet_softc * sc)3344 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3345 {
3346 int error;
3347
3348 error = vtnet_init_rx_queues(sc);
3349 if (error)
3350 return (error);
3351
3352 error = vtnet_init_tx_queues(sc);
3353 if (error)
3354 return (error);
3355
3356 return (0);
3357 }
3358
3359 static void
vtnet_set_active_vq_pairs(struct vtnet_softc * sc)3360 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3361 {
3362 device_t dev;
3363 int npairs;
3364
3365 dev = sc->vtnet_dev;
3366
3367 if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3368 sc->vtnet_act_vq_pairs = 1;
3369 return;
3370 }
3371
3372 npairs = sc->vtnet_req_vq_pairs;
3373
3374 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3375 device_printf(dev, "cannot set active queue pairs to %d, "
3376 "falling back to 1 queue pair\n", npairs);
3377 npairs = 1;
3378 }
3379
3380 sc->vtnet_act_vq_pairs = npairs;
3381 }
3382
3383 static void
vtnet_update_rx_offloads(struct vtnet_softc * sc)3384 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3385 {
3386 if_t ifp;
3387 uint64_t features;
3388 int error;
3389
3390 ifp = sc->vtnet_ifp;
3391 features = sc->vtnet_features;
3392
3393 VTNET_CORE_LOCK_ASSERT(sc);
3394
3395 if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3396 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3397 features |= VIRTIO_NET_F_GUEST_CSUM;
3398 else
3399 features &= ~VIRTIO_NET_F_GUEST_CSUM;
3400 }
3401
3402 if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3403 if (if_getcapenable(ifp) & IFCAP_LRO)
3404 features |= VTNET_LRO_FEATURES;
3405 else
3406 features &= ~VTNET_LRO_FEATURES;
3407 }
3408
3409 error = vtnet_ctrl_guest_offloads(sc,
3410 features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3411 VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN |
3412 VIRTIO_NET_F_GUEST_UFO));
3413 if (error) {
3414 device_printf(sc->vtnet_dev,
3415 "%s: cannot update Rx features\n", __func__);
3416 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3417 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3418 vtnet_init_locked(sc, 0);
3419 }
3420 } else
3421 sc->vtnet_features = features;
3422 }
3423
3424 static int
vtnet_reinit(struct vtnet_softc * sc)3425 vtnet_reinit(struct vtnet_softc *sc)
3426 {
3427 if_t ifp;
3428 int error;
3429
3430 ifp = sc->vtnet_ifp;
3431
3432 bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3433
3434 error = vtnet_virtio_reinit(sc);
3435 if (error)
3436 return (error);
3437
3438 vtnet_set_macaddr(sc);
3439 vtnet_set_active_vq_pairs(sc);
3440
3441 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3442 vtnet_init_rx_filters(sc);
3443
3444 if_sethwassist(ifp, 0);
3445 if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3446 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3447 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3448 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3449 if (if_getcapenable(ifp) & IFCAP_TSO4)
3450 if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3451 if (if_getcapenable(ifp) & IFCAP_TSO6)
3452 if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3453
3454 error = vtnet_init_rxtx_queues(sc);
3455 if (error)
3456 return (error);
3457
3458 return (0);
3459 }
3460
3461 static void
vtnet_init_locked(struct vtnet_softc * sc,int init_mode)3462 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3463 {
3464 if_t ifp;
3465
3466 ifp = sc->vtnet_ifp;
3467
3468 VTNET_CORE_LOCK_ASSERT(sc);
3469
3470 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3471 return;
3472
3473 vtnet_stop(sc);
3474
3475 #ifdef DEV_NETMAP
3476 /* Once stopped we can update the netmap flags, if necessary. */
3477 switch (init_mode) {
3478 case VTNET_INIT_NETMAP_ENTER:
3479 nm_set_native_flags(NA(ifp));
3480 break;
3481 case VTNET_INIT_NETMAP_EXIT:
3482 nm_clear_native_flags(NA(ifp));
3483 break;
3484 }
3485 #endif /* DEV_NETMAP */
3486
3487 if (vtnet_reinit(sc) != 0) {
3488 vtnet_stop(sc);
3489 return;
3490 }
3491
3492 if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3493 vtnet_update_link_status(sc);
3494 vtnet_enable_interrupts(sc);
3495 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3496
3497 #ifdef DEV_NETMAP
3498 /* Re-enable txsync/rxsync. */
3499 netmap_enable_all_rings(ifp);
3500 #endif /* DEV_NETMAP */
3501 }
3502
3503 static void
vtnet_init(void * xsc)3504 vtnet_init(void *xsc)
3505 {
3506 struct vtnet_softc *sc;
3507
3508 sc = xsc;
3509
3510 VTNET_CORE_LOCK(sc);
3511 vtnet_init_locked(sc, 0);
3512 VTNET_CORE_UNLOCK(sc);
3513 }
3514
3515 static void
vtnet_free_ctrl_vq(struct vtnet_softc * sc)3516 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3517 {
3518
3519 /*
3520 * The control virtqueue is only polled and therefore it should
3521 * already be empty.
3522 */
3523 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3524 ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3525 }
3526
3527 static void
vtnet_exec_ctrl_cmd(struct vtnet_softc * sc,void * cookie,struct sglist * sg,int readable,int writable)3528 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3529 struct sglist *sg, int readable, int writable)
3530 {
3531 struct virtqueue *vq;
3532
3533 vq = sc->vtnet_ctrl_vq;
3534
3535 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3536 VTNET_CORE_LOCK_ASSERT(sc);
3537
3538 if (!virtqueue_empty(vq))
3539 return;
3540
3541 /*
3542 * Poll for the response, but the command is likely completed before
3543 * returning from the notify.
3544 */
3545 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0) {
3546 virtqueue_notify(vq);
3547 virtqueue_poll(vq, NULL);
3548 }
3549 }
3550
3551 static int
vtnet_ctrl_mac_cmd(struct vtnet_softc * sc,uint8_t * hwaddr)3552 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3553 {
3554 struct sglist_seg segs[3];
3555 struct sglist sg;
3556 struct {
3557 struct virtio_net_ctrl_hdr hdr __aligned(2);
3558 uint8_t pad1;
3559 uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3560 uint8_t pad2;
3561 uint8_t ack;
3562 } s;
3563 int error;
3564
3565 error = 0;
3566 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3567
3568 s.hdr.class = VIRTIO_NET_CTRL_MAC;
3569 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3570 bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3571 s.ack = VIRTIO_NET_ERR;
3572
3573 sglist_init(&sg, nitems(segs), segs);
3574 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3575 error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3576 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3577 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3578
3579 if (error == 0)
3580 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3581
3582 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3583 }
3584
3585 static int
vtnet_ctrl_guest_offloads(struct vtnet_softc * sc,uint64_t offloads)3586 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3587 {
3588 struct sglist_seg segs[3];
3589 struct sglist sg;
3590 struct {
3591 struct virtio_net_ctrl_hdr hdr __aligned(2);
3592 uint8_t pad1;
3593 uint64_t offloads __aligned(8);
3594 uint8_t pad2;
3595 uint8_t ack;
3596 } s;
3597 int error;
3598
3599 error = 0;
3600 MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3601
3602 s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3603 s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3604 s.offloads = vtnet_gtoh64(sc, offloads);
3605 s.ack = VIRTIO_NET_ERR;
3606
3607 sglist_init(&sg, nitems(segs), segs);
3608 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3609 error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3610 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3611 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3612
3613 if (error == 0)
3614 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3615
3616 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3617 }
3618
3619 static int
vtnet_ctrl_mq_cmd(struct vtnet_softc * sc,uint16_t npairs)3620 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3621 {
3622 struct sglist_seg segs[3];
3623 struct sglist sg;
3624 struct {
3625 struct virtio_net_ctrl_hdr hdr __aligned(2);
3626 uint8_t pad1;
3627 struct virtio_net_ctrl_mq mq __aligned(2);
3628 uint8_t pad2;
3629 uint8_t ack;
3630 } s;
3631 int error;
3632
3633 error = 0;
3634 MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3635
3636 s.hdr.class = VIRTIO_NET_CTRL_MQ;
3637 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3638 s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3639 s.ack = VIRTIO_NET_ERR;
3640
3641 sglist_init(&sg, nitems(segs), segs);
3642 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3643 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3644 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3645 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3646
3647 if (error == 0)
3648 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3649
3650 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3651 }
3652
3653 static int
vtnet_ctrl_rx_cmd(struct vtnet_softc * sc,uint8_t cmd,bool on)3654 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3655 {
3656 struct sglist_seg segs[3];
3657 struct sglist sg;
3658 struct {
3659 struct virtio_net_ctrl_hdr hdr __aligned(2);
3660 uint8_t pad1;
3661 uint8_t onoff;
3662 uint8_t pad2;
3663 uint8_t ack;
3664 } s;
3665 int error;
3666
3667 error = 0;
3668 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3669
3670 s.hdr.class = VIRTIO_NET_CTRL_RX;
3671 s.hdr.cmd = cmd;
3672 s.onoff = on;
3673 s.ack = VIRTIO_NET_ERR;
3674
3675 sglist_init(&sg, nitems(segs), segs);
3676 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3677 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3678 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3679 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3680
3681 if (error == 0)
3682 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3683
3684 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3685 }
3686
3687 static int
vtnet_set_promisc(struct vtnet_softc * sc,bool on)3688 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3689 {
3690 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3691 }
3692
3693 static int
vtnet_set_allmulti(struct vtnet_softc * sc,bool on)3694 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3695 {
3696 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3697 }
3698
3699 static void
vtnet_rx_filter(struct vtnet_softc * sc)3700 vtnet_rx_filter(struct vtnet_softc *sc)
3701 {
3702 device_t dev;
3703 if_t ifp;
3704
3705 dev = sc->vtnet_dev;
3706 ifp = sc->vtnet_ifp;
3707
3708 VTNET_CORE_LOCK_ASSERT(sc);
3709
3710 if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3711 device_printf(dev, "cannot %s promiscuous mode\n",
3712 if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3713 }
3714
3715 if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3716 device_printf(dev, "cannot %s all-multicast mode\n",
3717 if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3718 }
3719 }
3720
3721 static u_int
vtnet_copy_ifaddr(void * arg,struct sockaddr_dl * sdl,u_int ucnt)3722 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3723 {
3724 struct vtnet_softc *sc = arg;
3725
3726 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3727 return (0);
3728
3729 if (ucnt < VTNET_MAX_MAC_ENTRIES)
3730 bcopy(LLADDR(sdl),
3731 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3732 ETHER_ADDR_LEN);
3733
3734 return (1);
3735 }
3736
3737 static u_int
vtnet_copy_maddr(void * arg,struct sockaddr_dl * sdl,u_int mcnt)3738 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3739 {
3740 struct vtnet_mac_filter *filter = arg;
3741
3742 if (mcnt < VTNET_MAX_MAC_ENTRIES)
3743 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3744 ETHER_ADDR_LEN);
3745
3746 return (1);
3747 }
3748
3749 static void
vtnet_rx_filter_mac(struct vtnet_softc * sc)3750 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3751 {
3752 struct virtio_net_ctrl_hdr hdr __aligned(2);
3753 struct vtnet_mac_filter *filter;
3754 struct sglist_seg segs[4];
3755 struct sglist sg;
3756 if_t ifp;
3757 bool promisc, allmulti;
3758 u_int ucnt, mcnt;
3759 int error;
3760 uint8_t ack;
3761
3762 ifp = sc->vtnet_ifp;
3763 filter = sc->vtnet_mac_filter;
3764 error = 0;
3765
3766 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3767 VTNET_CORE_LOCK_ASSERT(sc);
3768
3769 /* Unicast MAC addresses: */
3770 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3771 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3772
3773 if (promisc) {
3774 ucnt = 0;
3775 if_printf(ifp, "more than %d MAC addresses assigned, "
3776 "falling back to promiscuous mode\n",
3777 VTNET_MAX_MAC_ENTRIES);
3778 }
3779
3780 /* Multicast MAC addresses: */
3781 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3782 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3783
3784 if (allmulti) {
3785 mcnt = 0;
3786 if_printf(ifp, "more than %d multicast MAC addresses "
3787 "assigned, falling back to all-multicast mode\n",
3788 VTNET_MAX_MAC_ENTRIES);
3789 }
3790
3791 if (promisc && allmulti)
3792 goto out;
3793
3794 filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3795 filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3796
3797 hdr.class = VIRTIO_NET_CTRL_MAC;
3798 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3799 ack = VIRTIO_NET_ERR;
3800
3801 sglist_init(&sg, nitems(segs), segs);
3802 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3803 error |= sglist_append(&sg, &filter->vmf_unicast,
3804 sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3805 error |= sglist_append(&sg, &filter->vmf_multicast,
3806 sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3807 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3808 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3809
3810 if (error == 0)
3811 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3812 if (ack != VIRTIO_NET_OK)
3813 if_printf(ifp, "error setting host MAC filter table\n");
3814
3815 out:
3816 if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3817 if_printf(ifp, "cannot enable promiscuous mode\n");
3818 if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3819 if_printf(ifp, "cannot enable all-multicast mode\n");
3820 }
3821
3822 static int
vtnet_exec_vlan_filter(struct vtnet_softc * sc,int add,uint16_t tag)3823 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3824 {
3825 struct sglist_seg segs[3];
3826 struct sglist sg;
3827 struct {
3828 struct virtio_net_ctrl_hdr hdr __aligned(2);
3829 uint8_t pad1;
3830 uint16_t tag __aligned(2);
3831 uint8_t pad2;
3832 uint8_t ack;
3833 } s;
3834 int error;
3835
3836 error = 0;
3837 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3838
3839 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3840 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3841 s.tag = vtnet_gtoh16(sc, tag);
3842 s.ack = VIRTIO_NET_ERR;
3843
3844 sglist_init(&sg, nitems(segs), segs);
3845 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3846 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3847 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3848 MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3849
3850 if (error == 0)
3851 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3852
3853 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3854 }
3855
3856 static void
vtnet_rx_filter_vlan(struct vtnet_softc * sc)3857 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3858 {
3859 int i, bit;
3860 uint32_t w;
3861 uint16_t tag;
3862
3863 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3864 VTNET_CORE_LOCK_ASSERT(sc);
3865
3866 /* Enable the filter for each configured VLAN. */
3867 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3868 w = sc->vtnet_vlan_filter[i];
3869
3870 while ((bit = ffs(w) - 1) != -1) {
3871 w &= ~(1 << bit);
3872 tag = sizeof(w) * CHAR_BIT * i + bit;
3873
3874 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3875 device_printf(sc->vtnet_dev,
3876 "cannot enable VLAN %d filter\n", tag);
3877 }
3878 }
3879 }
3880 }
3881
3882 static void
vtnet_update_vlan_filter(struct vtnet_softc * sc,int add,uint16_t tag)3883 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3884 {
3885 if_t ifp;
3886 int idx, bit;
3887
3888 ifp = sc->vtnet_ifp;
3889 idx = (tag >> 5) & 0x7F;
3890 bit = tag & 0x1F;
3891
3892 if (tag == 0 || tag > 4095)
3893 return;
3894
3895 VTNET_CORE_LOCK(sc);
3896
3897 if (add)
3898 sc->vtnet_vlan_filter[idx] |= (1 << bit);
3899 else
3900 sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3901
3902 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3903 if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3904 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3905 device_printf(sc->vtnet_dev,
3906 "cannot %s VLAN %d %s the host filter table\n",
3907 add ? "add" : "remove", tag, add ? "to" : "from");
3908 }
3909
3910 VTNET_CORE_UNLOCK(sc);
3911 }
3912
3913 static void
vtnet_register_vlan(void * arg,if_t ifp,uint16_t tag)3914 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3915 {
3916
3917 if (if_getsoftc(ifp) != arg)
3918 return;
3919
3920 vtnet_update_vlan_filter(arg, 1, tag);
3921 }
3922
3923 static void
vtnet_unregister_vlan(void * arg,if_t ifp,uint16_t tag)3924 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3925 {
3926
3927 if (if_getsoftc(ifp) != arg)
3928 return;
3929
3930 vtnet_update_vlan_filter(arg, 0, tag);
3931 }
3932
3933 static void
vtnet_update_speed_duplex(struct vtnet_softc * sc)3934 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3935 {
3936 if_t ifp;
3937 uint32_t speed;
3938
3939 ifp = sc->vtnet_ifp;
3940
3941 if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3942 return;
3943
3944 /* BMV: Ignore duplex. */
3945 speed = virtio_read_dev_config_4(sc->vtnet_dev,
3946 offsetof(struct virtio_net_config, speed));
3947 if (speed != UINT32_MAX)
3948 if_setbaudrate(ifp, IF_Mbps(speed));
3949 }
3950
3951 static int
vtnet_is_link_up(struct vtnet_softc * sc)3952 vtnet_is_link_up(struct vtnet_softc *sc)
3953 {
3954 uint16_t status;
3955
3956 if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3957 return (1);
3958
3959 status = virtio_read_dev_config_2(sc->vtnet_dev,
3960 offsetof(struct virtio_net_config, status));
3961
3962 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3963 }
3964
3965 static void
vtnet_update_link_status(struct vtnet_softc * sc)3966 vtnet_update_link_status(struct vtnet_softc *sc)
3967 {
3968 if_t ifp;
3969 int link;
3970
3971 ifp = sc->vtnet_ifp;
3972 VTNET_CORE_LOCK_ASSERT(sc);
3973 link = vtnet_is_link_up(sc);
3974
3975 /* Notify if the link status has changed. */
3976 if (link != 0 && sc->vtnet_link_active == 0) {
3977 vtnet_update_speed_duplex(sc);
3978 sc->vtnet_link_active = 1;
3979 if_link_state_change(ifp, LINK_STATE_UP);
3980 } else if (link == 0 && sc->vtnet_link_active != 0) {
3981 sc->vtnet_link_active = 0;
3982 if_link_state_change(ifp, LINK_STATE_DOWN);
3983 }
3984 }
3985
3986 static int
vtnet_ifmedia_upd(if_t ifp __unused)3987 vtnet_ifmedia_upd(if_t ifp __unused)
3988 {
3989 return (EOPNOTSUPP);
3990 }
3991
3992 static void
vtnet_ifmedia_sts(if_t ifp,struct ifmediareq * ifmr)3993 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3994 {
3995 struct vtnet_softc *sc;
3996
3997 sc = if_getsoftc(ifp);
3998
3999 ifmr->ifm_status = IFM_AVALID;
4000 ifmr->ifm_active = IFM_ETHER;
4001
4002 VTNET_CORE_LOCK(sc);
4003 if (vtnet_is_link_up(sc) != 0) {
4004 ifmr->ifm_status |= IFM_ACTIVE;
4005 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
4006 } else
4007 ifmr->ifm_active |= IFM_NONE;
4008 VTNET_CORE_UNLOCK(sc);
4009 }
4010
4011 static void
vtnet_get_macaddr(struct vtnet_softc * sc)4012 vtnet_get_macaddr(struct vtnet_softc *sc)
4013 {
4014
4015 if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4016 virtio_read_device_config_array(sc->vtnet_dev,
4017 offsetof(struct virtio_net_config, mac),
4018 &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4019 } else {
4020 /* Generate a random locally administered unicast address. */
4021 sc->vtnet_hwaddr[0] = 0xB2;
4022 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4023 }
4024 }
4025
4026 static void
vtnet_set_macaddr(struct vtnet_softc * sc)4027 vtnet_set_macaddr(struct vtnet_softc *sc)
4028 {
4029 device_t dev;
4030 int error;
4031
4032 dev = sc->vtnet_dev;
4033
4034 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4035 error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4036 if (error)
4037 device_printf(dev, "unable to set MAC address\n");
4038 return;
4039 }
4040
4041 /* MAC in config is read-only in modern VirtIO. */
4042 if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4043 for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4044 virtio_write_dev_config_1(dev,
4045 offsetof(struct virtio_net_config, mac) + i,
4046 sc->vtnet_hwaddr[i]);
4047 }
4048 }
4049 }
4050
4051 static void
vtnet_attached_set_macaddr(struct vtnet_softc * sc)4052 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4053 {
4054
4055 /* Assign MAC address if it was generated. */
4056 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4057 vtnet_set_macaddr(sc);
4058 }
4059
4060 static void
vtnet_vlan_tag_remove(struct mbuf * m)4061 vtnet_vlan_tag_remove(struct mbuf *m)
4062 {
4063 struct ether_vlan_header *evh;
4064
4065 evh = mtod(m, struct ether_vlan_header *);
4066 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4067 m->m_flags |= M_VLANTAG;
4068
4069 /* Strip the 802.1Q header. */
4070 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4071 ETHER_HDR_LEN - ETHER_TYPE_LEN);
4072 m_adj(m, ETHER_VLAN_ENCAP_LEN);
4073 }
4074
4075 static void
vtnet_set_rx_process_limit(struct vtnet_softc * sc)4076 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4077 {
4078 int limit;
4079
4080 limit = vtnet_tunable_int(sc, "rx_process_limit",
4081 vtnet_rx_process_limit);
4082 if (limit < 0)
4083 limit = INT_MAX;
4084 sc->vtnet_rx_process_limit = limit;
4085 }
4086
4087 static void
vtnet_setup_rxq_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_rxq * rxq)4088 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4089 struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4090 {
4091 struct sysctl_oid *node;
4092 struct sysctl_oid_list *list;
4093 struct vtnet_rxq_stats *stats;
4094 char namebuf[16];
4095
4096 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4097 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4098 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4099 list = SYSCTL_CHILDREN(node);
4100
4101 stats = &rxq->vtnrx_stats;
4102
4103 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4104 &stats->vrxs_ipackets, "Receive packets");
4105 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4106 &stats->vrxs_ibytes, "Receive bytes");
4107 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4108 &stats->vrxs_iqdrops, "Receive drops");
4109 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4110 &stats->vrxs_ierrors, "Receive errors");
4111 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4112 &stats->vrxs_csum, "Receive checksum offloaded");
4113 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4114 &stats->vrxs_csum_failed, "Receive checksum offload failed");
4115 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4116 &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4117 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4118 &stats->vrxs_rescheduled,
4119 "Receive interrupt handler rescheduled");
4120 }
4121
4122 static void
vtnet_setup_txq_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_txq * txq)4123 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4124 struct sysctl_oid_list *child, struct vtnet_txq *txq)
4125 {
4126 struct sysctl_oid *node;
4127 struct sysctl_oid_list *list;
4128 struct vtnet_txq_stats *stats;
4129 char namebuf[16];
4130
4131 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4132 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4133 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4134 list = SYSCTL_CHILDREN(node);
4135
4136 stats = &txq->vtntx_stats;
4137
4138 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4139 &stats->vtxs_opackets, "Transmit packets");
4140 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4141 &stats->vtxs_obytes, "Transmit bytes");
4142 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4143 &stats->vtxs_omcasts, "Transmit multicasts");
4144 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4145 &stats->vtxs_csum, "Transmit checksum offloaded");
4146 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4147 &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4148 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4149 &stats->vtxs_rescheduled,
4150 "Transmit interrupt handler rescheduled");
4151 }
4152
4153 static void
vtnet_setup_queue_sysctl(struct vtnet_softc * sc)4154 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4155 {
4156 device_t dev;
4157 struct sysctl_ctx_list *ctx;
4158 struct sysctl_oid *tree;
4159 struct sysctl_oid_list *child;
4160 int i;
4161
4162 dev = sc->vtnet_dev;
4163 ctx = device_get_sysctl_ctx(dev);
4164 tree = device_get_sysctl_tree(dev);
4165 child = SYSCTL_CHILDREN(tree);
4166
4167 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4168 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4169 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4170 }
4171 }
4172
4173 static void
vtnet_setup_stat_sysctl(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct vtnet_softc * sc)4174 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4175 struct sysctl_oid_list *child, struct vtnet_softc *sc)
4176 {
4177 struct vtnet_statistics *stats;
4178 struct vtnet_rxq_stats rxaccum;
4179 struct vtnet_txq_stats txaccum;
4180
4181 vtnet_accum_stats(sc, &rxaccum, &txaccum);
4182
4183 stats = &sc->vtnet_stats;
4184 stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4185 stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4186 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4187 stats->tx_csum_offloaded = txaccum.vtxs_csum;
4188 stats->tx_tso_offloaded = txaccum.vtxs_tso;
4189 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4190
4191 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4192 CTLFLAG_RD, &stats->mbuf_alloc_failed,
4193 "Mbuf cluster allocation failures");
4194
4195 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4196 CTLFLAG_RD, &stats->rx_frame_too_large,
4197 "Received frame larger than the mbuf chain");
4198 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4199 CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4200 "Enqueuing the replacement receive mbuf failed");
4201 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4202 CTLFLAG_RD, &stats->rx_mergeable_failed,
4203 "Mergeable buffers receive failures");
4204 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4205 CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4206 "Received checksum offloaded buffer with unsupported "
4207 "Ethernet type");
4208 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4209 CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4210 "Received checksum offloaded buffer with incorrect IP protocol");
4211 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4212 CTLFLAG_RD, &stats->rx_csum_bad_offset,
4213 "Received checksum offloaded buffer with incorrect offset");
4214 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4215 CTLFLAG_RD, &stats->rx_csum_bad_proto,
4216 "Received checksum offloaded buffer with incorrect protocol");
4217 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4218 CTLFLAG_RD, &stats->rx_csum_failed,
4219 "Received buffer checksum offload failed");
4220 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4221 CTLFLAG_RD, &stats->rx_csum_offloaded,
4222 "Received buffer checksum offload succeeded");
4223 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4224 CTLFLAG_RD, &stats->rx_task_rescheduled,
4225 "Times the receive interrupt task rescheduled itself");
4226
4227 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4228 CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4229 "Aborted transmit of checksum offloaded buffer with unknown "
4230 "Ethernet type");
4231 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4232 CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4233 "Aborted transmit of checksum offloaded buffer because mismatched "
4234 "protocols");
4235 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4236 CTLFLAG_RD, &stats->tx_tso_not_tcp,
4237 "Aborted transmit of TSO buffer with non TCP protocol");
4238 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4239 CTLFLAG_RD, &stats->tx_tso_without_csum,
4240 "Aborted transmit of TSO buffer without TCP checksum offload");
4241 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4242 CTLFLAG_RD, &stats->tx_defragged,
4243 "Transmit mbufs defragged");
4244 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4245 CTLFLAG_RD, &stats->tx_defrag_failed,
4246 "Aborted transmit of buffer because defrag failed");
4247 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4248 CTLFLAG_RD, &stats->tx_csum_offloaded,
4249 "Offloaded checksum of transmitted buffer");
4250 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4251 CTLFLAG_RD, &stats->tx_tso_offloaded,
4252 "Segmentation offload of transmitted buffer");
4253 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4254 CTLFLAG_RD, &stats->tx_task_rescheduled,
4255 "Times the transmit interrupt task rescheduled itself");
4256 }
4257
4258 static void
vtnet_setup_sysctl(struct vtnet_softc * sc)4259 vtnet_setup_sysctl(struct vtnet_softc *sc)
4260 {
4261 device_t dev;
4262 struct sysctl_ctx_list *ctx;
4263 struct sysctl_oid *tree;
4264 struct sysctl_oid_list *child;
4265
4266 dev = sc->vtnet_dev;
4267 ctx = device_get_sysctl_ctx(dev);
4268 tree = device_get_sysctl_tree(dev);
4269 child = SYSCTL_CHILDREN(tree);
4270
4271 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4272 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4273 "Number of maximum supported virtqueue pairs");
4274 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4275 CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4276 "Number of requested virtqueue pairs");
4277 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4278 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4279 "Number of active virtqueue pairs");
4280
4281 vtnet_setup_stat_sysctl(ctx, child, sc);
4282 }
4283
4284 static void
vtnet_load_tunables(struct vtnet_softc * sc)4285 vtnet_load_tunables(struct vtnet_softc *sc)
4286 {
4287
4288 sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4289 "lro_entry_count", vtnet_lro_entry_count);
4290 if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4291 sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4292
4293 sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4294 "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4295 }
4296
4297 static int
vtnet_rxq_enable_intr(struct vtnet_rxq * rxq)4298 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4299 {
4300
4301 return (virtqueue_enable_intr(rxq->vtnrx_vq));
4302 }
4303
4304 static void
vtnet_rxq_disable_intr(struct vtnet_rxq * rxq)4305 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4306 {
4307
4308 virtqueue_disable_intr(rxq->vtnrx_vq);
4309 }
4310
4311 static int
vtnet_txq_enable_intr(struct vtnet_txq * txq)4312 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4313 {
4314 struct virtqueue *vq;
4315
4316 vq = txq->vtntx_vq;
4317
4318 if (vtnet_txq_below_threshold(txq) != 0)
4319 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4320
4321 /*
4322 * The free count is above our threshold. Keep the Tx interrupt
4323 * disabled until the queue is fuller.
4324 */
4325 return (0);
4326 }
4327
4328 static void
vtnet_txq_disable_intr(struct vtnet_txq * txq)4329 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4330 {
4331
4332 virtqueue_disable_intr(txq->vtntx_vq);
4333 }
4334
4335 static void
vtnet_enable_rx_interrupts(struct vtnet_softc * sc)4336 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4337 {
4338 struct vtnet_rxq *rxq;
4339 int i;
4340
4341 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4342 rxq = &sc->vtnet_rxqs[i];
4343 if (vtnet_rxq_enable_intr(rxq) != 0)
4344 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4345 }
4346 }
4347
4348 static void
vtnet_enable_tx_interrupts(struct vtnet_softc * sc)4349 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4350 {
4351 int i;
4352
4353 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4354 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4355 }
4356
4357 static void
vtnet_enable_interrupts(struct vtnet_softc * sc)4358 vtnet_enable_interrupts(struct vtnet_softc *sc)
4359 {
4360
4361 vtnet_enable_rx_interrupts(sc);
4362 vtnet_enable_tx_interrupts(sc);
4363 }
4364
4365 static void
vtnet_disable_rx_interrupts(struct vtnet_softc * sc)4366 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4367 {
4368 int i;
4369
4370 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4371 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4372 }
4373
4374 static void
vtnet_disable_tx_interrupts(struct vtnet_softc * sc)4375 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4376 {
4377 int i;
4378
4379 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4380 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4381 }
4382
4383 static void
vtnet_disable_interrupts(struct vtnet_softc * sc)4384 vtnet_disable_interrupts(struct vtnet_softc *sc)
4385 {
4386
4387 vtnet_disable_rx_interrupts(sc);
4388 vtnet_disable_tx_interrupts(sc);
4389 }
4390
4391 static int
vtnet_tunable_int(struct vtnet_softc * sc,const char * knob,int def)4392 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4393 {
4394 char path[64];
4395
4396 snprintf(path, sizeof(path),
4397 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4398 TUNABLE_INT_FETCH(path, &def);
4399
4400 return (def);
4401 }
4402
4403 #ifdef DEBUGNET
4404 static void
vtnet_debugnet_init(if_t ifp,int * nrxr,int * ncl,int * clsize)4405 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4406 {
4407 struct vtnet_softc *sc;
4408
4409 sc = if_getsoftc(ifp);
4410
4411 VTNET_CORE_LOCK(sc);
4412 *nrxr = sc->vtnet_req_vq_pairs;
4413 *ncl = DEBUGNET_MAX_IN_FLIGHT;
4414 *clsize = sc->vtnet_rx_clustersz;
4415 VTNET_CORE_UNLOCK(sc);
4416 }
4417
4418 static void
vtnet_debugnet_event(if_t ifp __unused,enum debugnet_ev event)4419 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4420 {
4421 struct vtnet_softc *sc;
4422 static bool sw_lro_enabled = false;
4423
4424 /*
4425 * Disable software LRO, since it would require entering the network
4426 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4427 */
4428 sc = if_getsoftc(ifp);
4429 switch (event) {
4430 case DEBUGNET_START:
4431 sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4432 if (sw_lro_enabled)
4433 sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4434 break;
4435 case DEBUGNET_END:
4436 if (sw_lro_enabled)
4437 sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4438 break;
4439 }
4440 }
4441
4442 static int
vtnet_debugnet_transmit(if_t ifp,struct mbuf * m)4443 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4444 {
4445 struct vtnet_softc *sc;
4446 struct vtnet_txq *txq;
4447 int error;
4448
4449 sc = if_getsoftc(ifp);
4450 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4451 IFF_DRV_RUNNING)
4452 return (EBUSY);
4453
4454 txq = &sc->vtnet_txqs[0];
4455 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4456 if (error == 0)
4457 (void)vtnet_txq_notify(txq);
4458 return (error);
4459 }
4460
4461 static int
vtnet_debugnet_poll(if_t ifp,int count)4462 vtnet_debugnet_poll(if_t ifp, int count)
4463 {
4464 struct vtnet_softc *sc;
4465 int i;
4466
4467 sc = if_getsoftc(ifp);
4468 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4469 IFF_DRV_RUNNING)
4470 return (EBUSY);
4471
4472 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4473 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4474 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4475 return (0);
4476 }
4477 #endif /* DEBUGNET */
4478