1/*- 2 * Copyright (c) 2016 Alexander Motin <mav@FreeBSD.org> 3 * Copyright (C) 2013 Intel Corporation 4 * Copyright (C) 2015 EMC Corporation 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, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29/* 30 * The Non-Transparent Bridge (NTB) is a device that allows you to connect 31 * two or more systems using a PCI-e links, providing remote memory access. 32 * 33 * This module contains a driver for simulated Ethernet device, using 34 * underlying NTB Transport device. 35 * 36 * NOTE: Much of the code in this module is shared with Linux. Any patches may 37 * be picked up and redistributed in Linux with a dual GPL/BSD license. 38 */ 39 40#include <sys/param.h> 41#include <sys/kernel.h> 42#include <sys/systm.h> 43#include <sys/buf_ring.h> 44#include <sys/bus.h> 45#include <sys/ktr.h> 46#include <sys/limits.h> 47#include <sys/module.h> 48#include <sys/socket.h> 49#include <sys/sockio.h> 50#include <sys/sysctl.h> 51#include <sys/taskqueue.h> 52 53#include <net/if.h> 54#include <net/if_media.h> 55#include <net/if_types.h> 56#include <net/if_media.h> 57#include <net/if_var.h> 58#include <net/bpf.h> 59#include <net/ethernet.h> 60 61#include <machine/bus.h> 62 63#include "../ntb_transport.h" 64 65#define KTR_NTB KTR_SPARE3 66#define NTB_MEDIATYPE (IFM_ETHER | IFM_AUTO | IFM_FDX) 67 68#define NTB_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP) 69#define NTB_CSUM_FEATURES6 (CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6) 70#define NTB_CSUM_SET (CSUM_DATA_VALID | CSUM_DATA_VALID_IPV6 | \ 71 CSUM_PSEUDO_HDR | \ 72 CSUM_IP_CHECKED | CSUM_IP_VALID | \ 73 CSUM_SCTP_VALID) 74 75static SYSCTL_NODE(_hw, OID_AUTO, if_ntb, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 76 "if_ntb"); 77 78static unsigned g_if_ntb_num_queues = UINT_MAX; 79SYSCTL_UINT(_hw_if_ntb, OID_AUTO, num_queues, CTLFLAG_RWTUN, 80 &g_if_ntb_num_queues, 0, "Number of queues per interface"); 81 82struct ntb_net_queue { 83 struct ntb_net_ctx *sc; 84 if_t ifp; 85 struct ntb_transport_qp *qp; 86 struct buf_ring *br; 87 struct task tx_task; 88 struct taskqueue *tx_tq; 89 struct mtx tx_lock; 90 struct callout queue_full; 91}; 92 93struct ntb_net_ctx { 94 if_t ifp; 95 struct ifmedia media; 96 u_char eaddr[ETHER_ADDR_LEN]; 97 int num_queues; 98 struct ntb_net_queue *queues; 99 int mtu; 100}; 101 102static int ntb_net_probe(device_t dev); 103static int ntb_net_attach(device_t dev); 104static int ntb_net_detach(device_t dev); 105static void ntb_net_init(void *arg); 106static int ntb_ifmedia_upd(struct ifnet *); 107static void ntb_ifmedia_sts(struct ifnet *, struct ifmediareq *); 108static int ntb_ioctl(if_t ifp, u_long command, caddr_t data); 109static int ntb_transmit(if_t ifp, struct mbuf *m); 110static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, 111 void *data, int len); 112static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, 113 void *data, int len); 114static void ntb_net_event_handler(void *data, enum ntb_link_event status); 115static void ntb_handle_tx(void *arg, int pending); 116static void ntb_qp_full(void *arg); 117static void ntb_qflush(if_t ifp); 118static void create_random_local_eui48(u_char *eaddr); 119 120static int 121ntb_net_probe(device_t dev) 122{ 123 124 device_set_desc(dev, "NTB Network Interface"); 125 return (0); 126} 127 128static int 129ntb_net_attach(device_t dev) 130{ 131 struct ntb_net_ctx *sc = device_get_softc(dev); 132 struct ntb_net_queue *q; 133 if_t ifp; 134 struct ntb_queue_handlers handlers = { ntb_net_rx_handler, 135 ntb_net_tx_handler, ntb_net_event_handler }; 136 int i; 137 138 ifp = sc->ifp = if_gethandle(IFT_ETHER); 139 if (ifp == NULL) { 140 printf("ntb: Cannot allocate ifnet structure\n"); 141 return (ENOMEM); 142 } 143 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 144 if_setdev(ifp, dev); 145 146 sc->num_queues = min(g_if_ntb_num_queues, 147 ntb_transport_queue_count(dev)); 148 sc->queues = malloc(sc->num_queues * sizeof(struct ntb_net_queue), 149 M_DEVBUF, M_WAITOK | M_ZERO); 150 sc->mtu = INT_MAX; 151 for (i = 0; i < sc->num_queues; i++) { 152 q = &sc->queues[i]; 153 q->sc = sc; 154 q->ifp = ifp; 155 q->qp = ntb_transport_create_queue(dev, i, &handlers, q); 156 if (q->qp == NULL) 157 break; 158 sc->mtu = imin(sc->mtu, ntb_transport_max_size(q->qp)); 159 mtx_init(&q->tx_lock, "ntb tx", NULL, MTX_DEF); 160 q->br = buf_ring_alloc(4096, M_DEVBUF, M_WAITOK, &q->tx_lock); 161 TASK_INIT(&q->tx_task, 0, ntb_handle_tx, q); 162 q->tx_tq = taskqueue_create_fast("ntb_txq", M_NOWAIT, 163 taskqueue_thread_enqueue, &q->tx_tq); 164 taskqueue_start_threads(&q->tx_tq, 1, PI_NET, "%s txq%d", 165 device_get_nameunit(dev), i); 166 callout_init(&q->queue_full, 1); 167 } 168 sc->num_queues = i; 169 device_printf(dev, "%d queue(s)\n", sc->num_queues); 170 171 if_setinitfn(ifp, ntb_net_init); 172 if_setsoftc(ifp, sc); 173 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); 174 if_setioctlfn(ifp, ntb_ioctl); 175 if_settransmitfn(ifp, ntb_transmit); 176 if_setqflushfn(ifp, ntb_qflush); 177 create_random_local_eui48(sc->eaddr); 178 ether_ifattach(ifp, sc->eaddr); 179 if_setcapabilities(ifp, IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 | 180 IFCAP_JUMBO_MTU | IFCAP_LINKSTATE); 181 if_setcapenable(ifp, IFCAP_JUMBO_MTU | IFCAP_LINKSTATE); 182 if_setmtu(ifp, sc->mtu - ETHER_HDR_LEN); 183 184 ifmedia_init(&sc->media, IFM_IMASK, ntb_ifmedia_upd, 185 ntb_ifmedia_sts); 186 ifmedia_add(&sc->media, NTB_MEDIATYPE, 0, NULL); 187 ifmedia_set(&sc->media, NTB_MEDIATYPE); 188 189 for (i = 0; i < sc->num_queues; i++) 190 ntb_transport_link_up(sc->queues[i].qp); 191 return (0); 192} 193 194static int 195ntb_net_detach(device_t dev) 196{ 197 struct ntb_net_ctx *sc = device_get_softc(dev); 198 struct ntb_net_queue *q; 199 int i; 200 201 for (i = 0; i < sc->num_queues; i++) 202 ntb_transport_link_down(sc->queues[i].qp); 203 ether_ifdetach(sc->ifp); 204 if_free(sc->ifp); 205 ifmedia_removeall(&sc->media); 206 for (i = 0; i < sc->num_queues; i++) { 207 q = &sc->queues[i]; 208 ntb_transport_free_queue(q->qp); 209 buf_ring_free(q->br, M_DEVBUF); 210 callout_drain(&q->queue_full); 211 taskqueue_drain_all(q->tx_tq); 212 mtx_destroy(&q->tx_lock); 213 } 214 free(sc->queues, M_DEVBUF); 215 return (0); 216} 217 218/* Network device interface */ 219 220static void 221ntb_net_init(void *arg) 222{ 223 struct ntb_net_ctx *sc = arg; 224 if_t ifp = sc->ifp; 225 226 if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE); 227 if_setbaudrate(ifp, ntb_transport_link_speed(sc->queues[0].qp)); 228 if_link_state_change(ifp, ntb_transport_link_query(sc->queues[0].qp) ? 229 LINK_STATE_UP : LINK_STATE_DOWN); 230} 231 232static int 233ntb_ioctl(if_t ifp, u_long command, caddr_t data) 234{ 235 struct ntb_net_ctx *sc = if_getsoftc(ifp); 236 struct ifreq *ifr = (struct ifreq *)data; 237 int error = 0; 238 239 switch (command) { 240 case SIOCSIFFLAGS: 241 case SIOCADDMULTI: 242 case SIOCDELMULTI: 243 break; 244 245 case SIOCSIFMTU: 246 { 247 if (ifr->ifr_mtu > sc->mtu - ETHER_HDR_LEN) { 248 error = EINVAL; 249 break; 250 } 251 252 if_setmtu(ifp, ifr->ifr_mtu); 253 break; 254 } 255 256 case SIOCSIFMEDIA: 257 case SIOCGIFMEDIA: 258 error = ifmedia_ioctl(ifp, ifr, &sc->media, command); 259 break; 260 261 case SIOCSIFCAP: 262 if (ifr->ifr_reqcap & IFCAP_RXCSUM) 263 if_setcapenablebit(ifp, IFCAP_RXCSUM, 0); 264 else 265 if_setcapenablebit(ifp, 0, IFCAP_RXCSUM); 266 if (ifr->ifr_reqcap & IFCAP_TXCSUM) { 267 if_setcapenablebit(ifp, IFCAP_TXCSUM, 0); 268 if_sethwassistbits(ifp, NTB_CSUM_FEATURES, 0); 269 } else { 270 if_setcapenablebit(ifp, 0, IFCAP_TXCSUM); 271 if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES); 272 } 273 if (ifr->ifr_reqcap & IFCAP_RXCSUM_IPV6) 274 if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0); 275 else 276 if_setcapenablebit(ifp, 0, IFCAP_RXCSUM_IPV6); 277 if (ifr->ifr_reqcap & IFCAP_TXCSUM_IPV6) { 278 if_setcapenablebit(ifp, IFCAP_TXCSUM_IPV6, 0); 279 if_sethwassistbits(ifp, NTB_CSUM_FEATURES6, 0); 280 } else { 281 if_setcapenablebit(ifp, 0, IFCAP_TXCSUM_IPV6); 282 if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES6); 283 } 284 break; 285 286 default: 287 error = ether_ioctl(ifp, command, data); 288 break; 289 } 290 291 return (error); 292} 293 294static int 295ntb_ifmedia_upd(struct ifnet *ifp) 296{ 297 struct ntb_net_ctx *sc = if_getsoftc(ifp); 298 struct ifmedia *ifm = &sc->media; 299 300 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 301 return (EINVAL); 302 303 return (0); 304} 305 306static void 307ntb_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 308{ 309 struct ntb_net_ctx *sc = if_getsoftc(ifp); 310 311 ifmr->ifm_status = IFM_AVALID; 312 ifmr->ifm_active = NTB_MEDIATYPE; 313 if (ntb_transport_link_query(sc->queues[0].qp)) 314 ifmr->ifm_status |= IFM_ACTIVE; 315} 316 317static void 318ntb_transmit_locked(struct ntb_net_queue *q) 319{ 320 if_t ifp = q->ifp; 321 struct mbuf *m; 322 int rc, len; 323 short mflags; 324 325 CTR0(KTR_NTB, "TX: ntb_transmit_locked"); 326 while ((m = drbr_peek(ifp, q->br)) != NULL) { 327 CTR1(KTR_NTB, "TX: start mbuf %p", m); 328 ether_bpf_mtap_if(ifp, m); 329 len = m->m_pkthdr.len; 330 mflags = m->m_flags; 331 rc = ntb_transport_tx_enqueue(q->qp, m, m, len); 332 if (rc != 0) { 333 CTR2(KTR_NTB, "TX: could not tx mbuf %p: %d", m, rc); 334 if (rc == EAGAIN) { 335 drbr_putback(ifp, q->br, m); 336 callout_reset_sbt(&q->queue_full, 337 SBT_1MS / 4, SBT_1MS / 4, 338 ntb_qp_full, q, 0); 339 } else { 340 m_freem(m); 341 drbr_advance(ifp, q->br); 342 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 343 } 344 break; 345 } 346 drbr_advance(ifp, q->br); 347 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 348 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 349 if (mflags & M_MCAST) 350 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 351 } 352} 353 354static int 355ntb_transmit(if_t ifp, struct mbuf *m) 356{ 357 struct ntb_net_ctx *sc = if_getsoftc(ifp); 358 struct ntb_net_queue *q; 359 int error, i; 360 361 CTR0(KTR_NTB, "TX: ntb_transmit"); 362 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 363 i = m->m_pkthdr.flowid % sc->num_queues; 364 else 365 i = curcpu % sc->num_queues; 366 q = &sc->queues[i]; 367 368 error = drbr_enqueue(ifp, q->br, m); 369 if (error) 370 return (error); 371 372 if (mtx_trylock(&q->tx_lock)) { 373 ntb_transmit_locked(q); 374 mtx_unlock(&q->tx_lock); 375 } else 376 taskqueue_enqueue(q->tx_tq, &q->tx_task); 377 return (0); 378} 379 380static void 381ntb_handle_tx(void *arg, int pending) 382{ 383 struct ntb_net_queue *q = arg; 384 385 mtx_lock(&q->tx_lock); 386 ntb_transmit_locked(q); 387 mtx_unlock(&q->tx_lock); 388} 389 390static void 391ntb_qp_full(void *arg) 392{ 393 struct ntb_net_queue *q = arg; 394 395 CTR0(KTR_NTB, "TX: qp_full callout"); 396 if (ntb_transport_tx_free_entry(q->qp) > 0) 397 taskqueue_enqueue(q->tx_tq, &q->tx_task); 398 else 399 callout_schedule_sbt(&q->queue_full, 400 SBT_1MS / 4, SBT_1MS / 4, 0); 401} 402 403static void 404ntb_qflush(if_t ifp) 405{ 406 struct ntb_net_ctx *sc = if_getsoftc(ifp); 407 struct ntb_net_queue *q; 408 struct mbuf *m; 409 int i; 410 411 for (i = 0; i < sc->num_queues; i++) { 412 q = &sc->queues[i]; 413 mtx_lock(&q->tx_lock); 414 while ((m = buf_ring_dequeue_sc(q->br)) != NULL) 415 m_freem(m); 416 mtx_unlock(&q->tx_lock); 417 } 418 if_qflush(ifp); 419} 420 421/* Network Device Callbacks */ 422static void 423ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, 424 int len) 425{ 426 427 m_freem(data); 428 CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data); 429} 430 431static void 432ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, 433 int len) 434{ 435 struct ntb_net_queue *q = qp_data; 436 struct ntb_net_ctx *sc = q->sc; 437 struct mbuf *m = data; 438 if_t ifp = q->ifp; 439 uint16_t proto; 440 441 CTR1(KTR_NTB, "RX: rx handler (%d)", len); 442 if (len < 0) { 443 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); 444 return; 445 } 446 447 m->m_pkthdr.rcvif = ifp; 448 if (sc->num_queues > 1) { 449 m->m_pkthdr.flowid = q - sc->queues; 450 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 451 } 452 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 453 m_copydata(m, 12, 2, (void *)&proto); 454 switch (ntohs(proto)) { 455 case ETHERTYPE_IP: 456 if (if_getcapenable(ifp) & IFCAP_RXCSUM) { 457 m->m_pkthdr.csum_data = 0xffff; 458 m->m_pkthdr.csum_flags = NTB_CSUM_SET; 459 } 460 break; 461 case ETHERTYPE_IPV6: 462 if (if_getcapenable(ifp) & IFCAP_RXCSUM_IPV6) { 463 m->m_pkthdr.csum_data = 0xffff; 464 m->m_pkthdr.csum_flags = NTB_CSUM_SET; 465 } 466 break; 467 } 468 } 469 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 470 if_input(ifp, m); 471} 472 473static void 474ntb_net_event_handler(void *data, enum ntb_link_event status) 475{ 476 struct ntb_net_queue *q = data; 477 478 if_setbaudrate(q->ifp, ntb_transport_link_speed(q->qp)); 479 if_link_state_change(q->ifp, (status == NTB_LINK_UP) ? LINK_STATE_UP : 480 LINK_STATE_DOWN); 481} 482 483/* Helper functions */ 484/* TODO: This too should really be part of the kernel */ 485#define EUI48_MULTICAST 1 << 0 486#define EUI48_LOCALLY_ADMINISTERED 1 << 1 487static void 488create_random_local_eui48(u_char *eaddr) 489{ 490 static uint8_t counter = 0; 491 492 eaddr[0] = EUI48_LOCALLY_ADMINISTERED; 493 arc4rand(&eaddr[1], 4, 0); 494 eaddr[5] = counter++; 495} 496 497static device_method_t ntb_net_methods[] = { 498 /* Device interface */ 499 DEVMETHOD(device_probe, ntb_net_probe), 500 DEVMETHOD(device_attach, ntb_net_attach), 501 DEVMETHOD(device_detach, ntb_net_detach), 502 DEVMETHOD_END 503}; 504 505static DEFINE_CLASS_0(ntb, ntb_net_driver, ntb_net_methods, 506 sizeof(struct ntb_net_ctx)); 507DRIVER_MODULE(if_ntb, ntb_transport, ntb_net_driver, NULL, NULL); 508MODULE_DEPEND(if_ntb, ntb_transport, 1, 1, 1); 509MODULE_VERSION(if_ntb, 1); 510