2 * Copyright (c) 2012 Chelsio Communications, Inc.
4 * Written by: Navdeep Parhar <np@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 #include "opt_inet6.h"
34 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/systm.h>
38 #include <sys/module.h>
39 #include <sys/types.h>
40 #include <sys/sockopt.h>
41 #include <sys/sysctl.h>
42 #include <sys/socket.h>
44 #include <net/ethernet.h>
46 #include <net/if_types.h>
47 #include <net/if_vlan_var.h>
48 #include <net/if_llatbl.h>
49 #include <net/route.h>
51 #include <netinet/if_ether.h>
52 #include <netinet/in.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/in_var.h>
55 #include <netinet6/in6_var.h>
56 #include <netinet6/in6_pcb.h>
57 #include <netinet6/nd6.h>
59 #include <netinet/tcp.h>
60 #include <netinet/tcp_fsm.h>
61 #include <netinet/tcp_timer.h>
62 #include <netinet/tcp_var.h>
63 #include <netinet/tcp_syncache.h>
64 #include <netinet/tcp_offload.h>
65 #include <netinet/toecore.h>
67 static struct mtx toedev_lock;
68 static TAILQ_HEAD(, toedev) toedev_list;
69 static eventhandler_tag listen_start_eh;
70 static eventhandler_tag listen_stop_eh;
71 static eventhandler_tag lle_event_eh;
72 static eventhandler_tag route_redirect_eh;
75 toedev_connect(struct toedev *tod __unused, struct socket *so __unused,
76 struct rtentry *rt __unused, struct sockaddr *nam __unused)
83 toedev_listen_start(struct toedev *tod __unused, struct tcpcb *tp __unused)
90 toedev_listen_stop(struct toedev *tod __unused, struct tcpcb *tp __unused)
97 toedev_input(struct toedev *tod __unused, struct tcpcb *tp __unused,
106 toedev_rcvd(struct toedev *tod __unused, struct tcpcb *tp __unused)
113 toedev_output(struct toedev *tod __unused, struct tcpcb *tp __unused)
120 toedev_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp __unused)
127 toedev_l2_update(struct toedev *tod __unused, struct ifnet *ifp __unused,
128 struct sockaddr *sa __unused, uint8_t *lladdr __unused,
129 uint16_t vtag __unused)
136 toedev_route_redirect(struct toedev *tod __unused, struct ifnet *ifp __unused,
137 struct rtentry *rt0 __unused, struct rtentry *rt1 __unused)
144 toedev_syncache_added(struct toedev *tod __unused, void *ctx __unused)
151 toedev_syncache_removed(struct toedev *tod __unused, void *ctx __unused)
158 toedev_syncache_respond(struct toedev *tod __unused, void *ctx __unused,
167 toedev_offload_socket(struct toedev *tod __unused, void *ctx __unused,
168 struct socket *so __unused)
175 toedev_ctloutput(struct toedev *tod __unused, struct tcpcb *tp __unused,
176 int sopt_dir __unused, int sopt_name __unused)
183 * Inform one or more TOE devices about a listening socket.
186 toe_listen_start(struct inpcb *inp, void *arg)
188 struct toedev *t, *tod;
191 INP_WLOCK_ASSERT(inp);
192 KASSERT(inp->inp_pcbinfo == &V_tcbinfo,
193 ("%s: inp is not a TCP inp", __func__));
195 if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))
199 if (tp->t_state != TCPS_LISTEN)
203 mtx_lock(&toedev_lock);
204 TAILQ_FOREACH(tod, &toedev_list, link) {
205 if (t == NULL || t == tod)
206 tod->tod_listen_start(tod, tp);
208 mtx_unlock(&toedev_lock);
212 toe_listen_start_event(void *arg __unused, struct tcpcb *tp)
214 struct inpcb *inp = tp->t_inpcb;
216 INP_WLOCK_ASSERT(inp);
217 KASSERT(tp->t_state == TCPS_LISTEN,
218 ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
220 toe_listen_start(inp, NULL);
224 toe_listen_stop_event(void *arg __unused, struct tcpcb *tp)
228 struct inpcb *inp = tp->t_inpcb;
231 INP_WLOCK_ASSERT(inp);
232 KASSERT(tp->t_state == TCPS_LISTEN,
233 ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
235 mtx_lock(&toedev_lock);
236 TAILQ_FOREACH(tod, &toedev_list, link)
237 tod->tod_listen_stop(tod, tp);
238 mtx_unlock(&toedev_lock);
242 * Fill up a freshly allocated toedev struct with reasonable defaults.
245 init_toedev(struct toedev *tod)
248 tod->tod_softc = NULL;
251 * Provide no-op defaults so that the kernel can call any toedev
252 * function without having to check whether the TOE driver supplied one
255 tod->tod_connect = toedev_connect;
256 tod->tod_listen_start = toedev_listen_start;
257 tod->tod_listen_stop = toedev_listen_stop;
258 tod->tod_input = toedev_input;
259 tod->tod_rcvd = toedev_rcvd;
260 tod->tod_output = toedev_output;
261 tod->tod_send_rst = toedev_output;
262 tod->tod_send_fin = toedev_output;
263 tod->tod_pcb_detach = toedev_pcb_detach;
264 tod->tod_l2_update = toedev_l2_update;
265 tod->tod_route_redirect = toedev_route_redirect;
266 tod->tod_syncache_added = toedev_syncache_added;
267 tod->tod_syncache_removed = toedev_syncache_removed;
268 tod->tod_syncache_respond = toedev_syncache_respond;
269 tod->tod_offload_socket = toedev_offload_socket;
270 tod->tod_ctloutput = toedev_ctloutput;
274 * Register an active TOE device with the system. This allows it to receive
275 * notifications from the kernel.
278 register_toedev(struct toedev *tod)
282 mtx_lock(&toedev_lock);
283 TAILQ_FOREACH(t, &toedev_list, link) {
285 mtx_unlock(&toedev_lock);
290 TAILQ_INSERT_TAIL(&toedev_list, tod, link);
291 registered_toedevs++;
292 mtx_unlock(&toedev_lock);
294 inp_apply_all(toe_listen_start, tod);
300 * Remove the TOE device from the global list of active TOE devices. It is the
301 * caller's responsibility to ensure that the TOE device is quiesced prior to
305 unregister_toedev(struct toedev *tod)
307 struct toedev *t, *t2;
310 mtx_lock(&toedev_lock);
311 TAILQ_FOREACH_SAFE(t, &toedev_list, link, t2) {
313 TAILQ_REMOVE(&toedev_list, tod, link);
314 registered_toedevs--;
319 KASSERT(registered_toedevs >= 0,
320 ("%s: registered_toedevs (%d) < 0", __func__, registered_toedevs));
321 mtx_unlock(&toedev_lock);
326 toe_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
327 struct inpcb *inp, void *tod, void *todctx)
329 struct socket *lso = inp->inp_socket;
331 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
332 INP_WLOCK_ASSERT(inp);
334 tcp_offload_syncache_add(inc, to, th, inp, &lso, tod, todctx);
338 toe_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
339 struct tcphdr *th, struct socket **lsop)
342 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
344 return (syncache_expand(inc, to, th, lsop, NULL));
348 * General purpose check to see if a 4-tuple is in use by the kernel. If a TCP
349 * header (presumably for an incoming SYN) is also provided, an existing 4-tuple
350 * in TIME_WAIT may be assassinated freeing it up for re-use.
352 * Note that the TCP header must have been run through tcp_fields_to_host() or
356 toe_4tuple_check(struct in_conninfo *inc, struct tcphdr *th, struct ifnet *ifp)
360 if (inc->inc_flags & INC_ISIPV6) {
361 inp = in6_pcblookup(&V_tcbinfo, &inc->inc6_faddr,
362 inc->inc_fport, &inc->inc6_laddr, inc->inc_lport,
363 INPLOOKUP_WLOCKPCB, ifp);
365 inp = in_pcblookup(&V_tcbinfo, inc->inc_faddr, inc->inc_fport,
366 inc->inc_laddr, inc->inc_lport, INPLOOKUP_WLOCKPCB, ifp);
369 INP_WLOCK_ASSERT(inp);
371 if ((inp->inp_flags & INP_TIMEWAIT) && th != NULL) {
373 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* for twcheck */
374 if (!tcp_twcheck(inp, NULL, th, NULL, 0))
386 toe_lle_event(void *arg __unused, struct llentry *lle, int evt)
394 LLE_WLOCK_ASSERT(lle);
396 ifp = lle->lle_tbl->llt_ifp;
399 KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
400 ("%s: lle_event %d for lle %p but sa %p !INET && !INET6",
401 __func__, evt, lle, sa));
404 * Not interested if the interface's TOE capability is not enabled.
406 if ((sa->sa_family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4)) ||
407 (sa->sa_family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6)))
415 if (evt != LLENTRY_RESOLVED) {
418 * LLENTRY_TIMEDOUT, LLENTRY_DELETED, LLENTRY_EXPIRED all mean
419 * this entry is going to be deleted.
425 KASSERT(lle->la_flags & LLE_VALID,
426 ("%s: %p resolved but not valid?", __func__, lle));
428 lladdr = (uint8_t *)&lle->ll_addr;
430 VLAN_TAG(ifp, &vtag);
434 tod->tod_l2_update(tod, ifp, sa, lladdr, vtag);
441 toe_route_redirect_event(void *arg __unused, struct rtentry *rt0,
442 struct rtentry *rt1, struct sockaddr *sa)
450 * XXX: no checks to verify that sa is really a neighbor because we assume it is
451 * the result of a route lookup and is on-link on the given ifp.
454 toe_nd6_resolve(struct ifnet *ifp, struct sockaddr *sa, uint8_t *lladdr)
457 struct sockaddr_in6 *sin6 = (void *)sa;
461 IF_AFDATA_RLOCK(ifp);
462 lle = lla_lookup(LLTABLE6(ifp), flags, sa);
463 IF_AFDATA_RUNLOCK(ifp);
466 lle = nd6_lookup(&sin6->sin6_addr, ND6_CREATE | ND6_EXCLUSIVE,
468 IF_AFDATA_UNLOCK(ifp);
470 return (ENOMEM); /* Couldn't create entry in cache. */
471 lle->ln_state = ND6_LLINFO_INCOMPLETE;
472 nd6_llinfo_settimer_locked(lle,
473 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
476 nd6_ns_output(ifp, NULL, &sin6->sin6_addr, NULL, 0);
478 return (EWOULDBLOCK);
481 if (lle->ln_state == ND6_LLINFO_STALE) {
482 if ((flags & LLE_EXCLUSIVE) == 0) {
484 flags |= LLE_EXCLUSIVE;
488 LLE_WLOCK_ASSERT(lle);
491 lle->ln_state = ND6_LLINFO_DELAY;
492 nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
495 if (lle->la_flags & LLE_VALID) {
496 memcpy(lladdr, &lle->ll_addr, ifp->if_addrlen);
501 if (flags & LLE_EXCLUSIVE)
511 * Returns 0 or EWOULDBLOCK on success (any other value is an error). 0 means
512 * lladdr and vtag are valid on return, EWOULDBLOCK means the TOE driver's
513 * tod_l2_update will be called later, when the entry is resolved or times out.
516 toe_l2_resolve(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa,
517 uint8_t *lladdr, uint16_t *vtag)
524 switch (sa->sa_family) {
527 rc = arpresolve(ifp, NULL, NULL, sa, lladdr, &lle);
532 rc = toe_nd6_resolve(ifp, sa, lladdr);
536 return (EPROTONOSUPPORT);
541 if (VLAN_TAG(ifp, vtag) != 0)
550 toe_connect_failed(struct toedev *tod, struct inpcb *inp, int err)
553 INP_WLOCK_ASSERT(inp);
555 if (!(inp->inp_flags & INP_DROPPED)) {
556 struct tcpcb *tp = intotcpcb(inp);
558 KASSERT(tp->t_flags & TF_TOE,
559 ("%s: tp %p not offloaded.", __func__, tp));
564 * Temporary failure during offload, take this PCB back.
565 * Detach from the TOE driver and do the rest of what
566 * TCP's pru_connect would have done if the connection
570 tod->tod_pcb_detach(tod, tp);
571 KASSERT(!(tp->t_flags & TF_TOE),
572 ("%s: tp %p still offloaded.", __func__, tp));
573 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
574 (void) tcp_output(tp);
577 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
578 tp = tcp_drop(tp, err);
580 INP_WLOCK(inp); /* re-acquire */
583 INP_WLOCK_ASSERT(inp);
590 mtx_init(&toedev_lock, "toedev lock", NULL, MTX_DEF);
591 TAILQ_INIT(&toedev_list);
593 listen_start_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_start,
594 toe_listen_start_event, NULL, EVENTHANDLER_PRI_ANY);
595 listen_stop_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_stop,
596 toe_listen_stop_event, NULL, EVENTHANDLER_PRI_ANY);
597 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, toe_lle_event, NULL,
598 EVENTHANDLER_PRI_ANY);
599 route_redirect_eh = EVENTHANDLER_REGISTER(route_redirect_event,
600 toe_route_redirect_event, NULL, EVENTHANDLER_PRI_ANY);
609 mtx_lock(&toedev_lock);
610 if (!TAILQ_EMPTY(&toedev_list)) {
611 mtx_unlock(&toedev_lock);
615 EVENTHANDLER_DEREGISTER(tcp_offload_listen_start, listen_start_eh);
616 EVENTHANDLER_DEREGISTER(tcp_offload_listen_stop, listen_stop_eh);
617 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
618 EVENTHANDLER_DEREGISTER(route_redirect_event, route_redirect_eh);
620 mtx_unlock(&toedev_lock);
621 mtx_destroy(&toedev_lock);
627 toecore_mod_handler(module_t mod, int cmd, void *arg)
631 return (toecore_load());
633 if (cmd == MOD_UNLOAD)
634 return (toecore_unload());
639 static moduledata_t mod_data= {
645 MODULE_VERSION(toecore, 1);
646 DECLARE_MODULE(toecore, mod_data, SI_SUB_EXEC, SI_ORDER_ANY);