1/*-
2 * Copyright (c) 2012 Chelsio Communications, Inc.
3 * All rights reserved.
4 * Written by: Navdeep Parhar <np@FreeBSD.org>
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
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.
14 *
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
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: stable/10/sys/netinet/toecore.c 329982 2018-02-25 11:29:55Z hselasky $");
30
31#include "opt_inet.h"
32#include "opt_inet6.h"
33
34#include <sys/param.h>
35#include <sys/kernel.h>
36#include <sys/systm.h>
37#include <sys/mbuf.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>
43
44#if defined(KLD_MODULE) || defined(INET) || defined(INET6)
45
46#include <net/ethernet.h>
47#include <net/if.h>
48#include <net/if_types.h>
49#include <net/if_vlan_var.h>
50#include <net/if_llatbl.h>
51#include <net/route.h>
52
53#include <netinet/if_ether.h>
54#include <netinet/in.h>
55#include <netinet/in_pcb.h>
56#include <netinet/in_var.h>
57#include <netinet6/in6_var.h>
58#include <netinet6/in6_pcb.h>
59#include <netinet6/nd6.h>
60#define TCPSTATES
61#include <netinet/tcp.h>
62#include <netinet/tcp_fsm.h>
63#include <netinet/tcp_timer.h>
64#include <netinet/tcp_var.h>
65#include <netinet/tcp_syncache.h>
66#include <netinet/tcp_offload.h>
67#include <netinet/toecore.h>
68
69static struct mtx toedev_lock;
70static TAILQ_HEAD(, toedev) toedev_list;
71static eventhandler_tag listen_start_eh;
72static eventhandler_tag listen_stop_eh;
73static eventhandler_tag lle_event_eh;
74static eventhandler_tag route_redirect_eh;
75
76static int
77toedev_connect(struct toedev *tod __unused, struct socket *so __unused,
78    struct rtentry *rt __unused, struct sockaddr *nam __unused)
79{
80
81	return (ENOTSUP);
82}
83
84static int
85toedev_listen_start(struct toedev *tod __unused, struct tcpcb *tp __unused)
86{
87
88	return (ENOTSUP);
89}
90
91static int
92toedev_listen_stop(struct toedev *tod __unused, struct tcpcb *tp __unused)
93{
94
95	return (ENOTSUP);
96}
97
98static void
99toedev_input(struct toedev *tod __unused, struct tcpcb *tp __unused,
100    struct mbuf *m)
101{
102
103	m_freem(m);
104	return;
105}
106
107static void
108toedev_rcvd(struct toedev *tod __unused, struct tcpcb *tp __unused)
109{
110
111	return;
112}
113
114static int
115toedev_output(struct toedev *tod __unused, struct tcpcb *tp __unused)
116{
117
118	return (ENOTSUP);
119}
120
121static void
122toedev_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp __unused)
123{
124
125	return;
126}
127
128static void
129toedev_l2_update(struct toedev *tod __unused, struct ifnet *ifp __unused,
130    struct sockaddr *sa __unused, uint8_t *lladdr __unused,
131    uint16_t vtag __unused)
132{
133
134	return;
135}
136
137static void
138toedev_route_redirect(struct toedev *tod __unused, struct ifnet *ifp __unused,
139    struct rtentry *rt0 __unused, struct rtentry *rt1 __unused)
140{
141
142	return;
143}
144
145static void
146toedev_syncache_added(struct toedev *tod __unused, void *ctx __unused)
147{
148
149	return;
150}
151
152static void
153toedev_syncache_removed(struct toedev *tod __unused, void *ctx __unused)
154{
155
156	return;
157}
158
159static int
160toedev_syncache_respond(struct toedev *tod __unused, void *ctx __unused,
161    struct mbuf *m)
162{
163
164	m_freem(m);
165	return (0);
166}
167
168static void
169toedev_offload_socket(struct toedev *tod __unused, void *ctx __unused,
170    struct socket *so __unused)
171{
172
173	return;
174}
175
176static void
177toedev_ctloutput(struct toedev *tod __unused, struct tcpcb *tp __unused,
178    int sopt_dir __unused, int sopt_name __unused)
179{
180
181	return;
182}
183
184/*
185 * Inform one or more TOE devices about a listening socket.
186 */
187static void
188toe_listen_start(struct inpcb *inp, void *arg)
189{
190	struct toedev *t, *tod;
191	struct tcpcb *tp;
192
193	INP_WLOCK_ASSERT(inp);
194	KASSERT(inp->inp_pcbinfo == &V_tcbinfo,
195	    ("%s: inp is not a TCP inp", __func__));
196
197	if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))
198		return;
199
200	tp = intotcpcb(inp);
201	if (tp->t_state != TCPS_LISTEN)
202		return;
203
204	t = arg;
205	mtx_lock(&toedev_lock);
206	TAILQ_FOREACH(tod, &toedev_list, link) {
207		if (t == NULL || t == tod)
208			tod->tod_listen_start(tod, tp);
209	}
210	mtx_unlock(&toedev_lock);
211}
212
213static void
214toe_listen_start_event(void *arg __unused, struct tcpcb *tp)
215{
216	struct inpcb *inp = tp->t_inpcb;
217
218	INP_WLOCK_ASSERT(inp);
219	KASSERT(tp->t_state == TCPS_LISTEN,
220	    ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
221
222	toe_listen_start(inp, NULL);
223}
224
225static void
226toe_listen_stop_event(void *arg __unused, struct tcpcb *tp)
227{
228	struct toedev *tod;
229#ifdef INVARIANTS
230	struct inpcb *inp = tp->t_inpcb;
231#endif
232
233	INP_WLOCK_ASSERT(inp);
234	KASSERT(tp->t_state == TCPS_LISTEN,
235	    ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
236
237	mtx_lock(&toedev_lock);
238	TAILQ_FOREACH(tod, &toedev_list, link)
239	    tod->tod_listen_stop(tod, tp);
240	mtx_unlock(&toedev_lock);
241}
242
243/*
244 * Fill up a freshly allocated toedev struct with reasonable defaults.
245 */
246void
247init_toedev(struct toedev *tod)
248{
249
250	tod->tod_softc = NULL;
251
252	/*
253	 * Provide no-op defaults so that the kernel can call any toedev
254	 * function without having to check whether the TOE driver supplied one
255	 * or not.
256	 */
257	tod->tod_connect = toedev_connect;
258	tod->tod_listen_start = toedev_listen_start;
259	tod->tod_listen_stop = toedev_listen_stop;
260	tod->tod_input = toedev_input;
261	tod->tod_rcvd = toedev_rcvd;
262	tod->tod_output = toedev_output;
263	tod->tod_send_rst = toedev_output;
264	tod->tod_send_fin = toedev_output;
265	tod->tod_pcb_detach = toedev_pcb_detach;
266	tod->tod_l2_update = toedev_l2_update;
267	tod->tod_route_redirect = toedev_route_redirect;
268	tod->tod_syncache_added = toedev_syncache_added;
269	tod->tod_syncache_removed = toedev_syncache_removed;
270	tod->tod_syncache_respond = toedev_syncache_respond;
271	tod->tod_offload_socket = toedev_offload_socket;
272	tod->tod_ctloutput = toedev_ctloutput;
273}
274
275/*
276 * Register an active TOE device with the system.  This allows it to receive
277 * notifications from the kernel.
278 */
279int
280register_toedev(struct toedev *tod)
281{
282	struct toedev *t;
283
284	mtx_lock(&toedev_lock);
285	TAILQ_FOREACH(t, &toedev_list, link) {
286		if (t == tod) {
287			mtx_unlock(&toedev_lock);
288			return (EEXIST);
289		}
290	}
291
292	TAILQ_INSERT_TAIL(&toedev_list, tod, link);
293	registered_toedevs++;
294	mtx_unlock(&toedev_lock);
295
296	inp_apply_all(toe_listen_start, tod);
297
298	return (0);
299}
300
301/*
302 * Remove the TOE device from the global list of active TOE devices.  It is the
303 * caller's responsibility to ensure that the TOE device is quiesced prior to
304 * this call.
305 */
306int
307unregister_toedev(struct toedev *tod)
308{
309	struct toedev *t, *t2;
310	int rc = ENODEV;
311
312	mtx_lock(&toedev_lock);
313	TAILQ_FOREACH_SAFE(t, &toedev_list, link, t2) {
314		if (t == tod) {
315			TAILQ_REMOVE(&toedev_list, tod, link);
316			registered_toedevs--;
317			rc = 0;
318			break;
319		}
320	}
321	KASSERT(registered_toedevs >= 0,
322	    ("%s: registered_toedevs (%d) < 0", __func__, registered_toedevs));
323	mtx_unlock(&toedev_lock);
324	return (rc);
325}
326
327void
328toe_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
329    struct inpcb *inp, void *tod, void *todctx)
330{
331	struct socket *lso = inp->inp_socket;
332
333	INP_WLOCK_ASSERT(inp);
334
335	syncache_add(inc, to, th, inp, &lso, NULL, tod, todctx);
336}
337
338int
339toe_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
340    struct tcphdr *th, struct socket **lsop)
341{
342
343	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
344
345	return (syncache_expand(inc, to, th, lsop, NULL));
346}
347
348/*
349 * General purpose check to see if a 4-tuple is in use by the kernel.  If a TCP
350 * header (presumably for an incoming SYN) is also provided, an existing 4-tuple
351 * in TIME_WAIT may be assassinated freeing it up for re-use.
352 *
353 * Note that the TCP header must have been run through tcp_fields_to_host() or
354 * equivalent.
355 */
356int
357toe_4tuple_check(struct in_conninfo *inc, struct tcphdr *th, struct ifnet *ifp)
358{
359	struct inpcb *inp;
360
361	if (inc->inc_flags & INC_ISIPV6) {
362#if defined(KLD_MODULE) || defined(INET6)
363		inp = in6_pcblookup(&V_tcbinfo, &inc->inc6_faddr,
364		    inc->inc_fport, &inc->inc6_laddr, inc->inc_lport,
365		    INPLOOKUP_WLOCKPCB, ifp);
366#else
367		inp = NULL;
368#endif
369	} else {
370#if defined(KLD_MODULE) || defined(INET)
371		inp = in_pcblookup(&V_tcbinfo, inc->inc_faddr, inc->inc_fport,
372		    inc->inc_laddr, inc->inc_lport, INPLOOKUP_WLOCKPCB, ifp);
373#else
374		inp = NULL;
375#endif
376	}
377	if (inp != NULL) {
378		INP_WLOCK_ASSERT(inp);
379
380		if ((inp->inp_flags & INP_TIMEWAIT) && th != NULL) {
381
382			INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* for twcheck */
383			if (!tcp_twcheck(inp, NULL, th, NULL, 0))
384				return (EADDRINUSE);
385		} else {
386			INP_WUNLOCK(inp);
387			return (EADDRINUSE);
388		}
389	}
390
391	return (0);
392}
393
394static void
395toe_lle_event(void *arg __unused, struct llentry *lle, int evt)
396{
397	struct toedev *tod;
398	struct ifnet *ifp;
399	struct sockaddr *sa;
400	uint8_t *lladdr;
401	uint16_t vtag;
402
403	LLE_WLOCK_ASSERT(lle);
404
405	ifp = lle->lle_tbl->llt_ifp;
406	sa = L3_ADDR(lle);
407
408	KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
409	    ("%s: lle_event %d for lle %p but sa %p !INET && !INET6",
410	    __func__, evt, lle, sa));
411
412	/*
413	 * Not interested if the interface's TOE capability is not enabled.
414	 */
415	if ((sa->sa_family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4)) ||
416	    (sa->sa_family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6)))
417		return;
418
419	tod = TOEDEV(ifp);
420	if (tod == NULL)
421		return;
422
423	vtag = 0xfff;
424	if (evt != LLENTRY_RESOLVED) {
425
426		/*
427		 * LLENTRY_TIMEDOUT, LLENTRY_DELETED, LLENTRY_EXPIRED all mean
428		 * this entry is going to be deleted.
429		 */
430
431		lladdr = NULL;
432	} else {
433
434		KASSERT(lle->la_flags & LLE_VALID,
435		    ("%s: %p resolved but not valid?", __func__, lle));
436
437		lladdr = (uint8_t *)&lle->ll_addr;
438#ifdef VLAN_TAG
439		VLAN_TAG(ifp, &vtag);
440#endif
441	}
442
443	tod->tod_l2_update(tod, ifp, sa, lladdr, vtag);
444}
445
446/*
447 * XXX: implement.
448 */
449static void
450toe_route_redirect_event(void *arg __unused, struct rtentry *rt0,
451    struct rtentry *rt1, struct sockaddr *sa)
452{
453
454	return;
455}
456
457#ifdef INET6
458/*
459 * XXX: no checks to verify that sa is really a neighbor because we assume it is
460 * the result of a route lookup and is on-link on the given ifp.
461 */
462static int
463toe_nd6_resolve(struct ifnet *ifp, struct sockaddr *sa, uint8_t *lladdr)
464{
465	struct llentry *lle;
466	struct sockaddr_in6 *sin6 = (void *)sa;
467	int rc, flags = 0;
468
469restart:
470	IF_AFDATA_RLOCK(ifp);
471	lle = lla_lookup(LLTABLE6(ifp), flags, sa);
472	IF_AFDATA_RUNLOCK(ifp);
473	if (lle == NULL) {
474		IF_AFDATA_LOCK(ifp);
475		lle = nd6_lookup(&sin6->sin6_addr, ND6_CREATE | ND6_EXCLUSIVE,
476		    ifp);
477		IF_AFDATA_UNLOCK(ifp);
478		if (lle == NULL)
479			return (ENOMEM); /* Couldn't create entry in cache. */
480		lle->ln_state = ND6_LLINFO_INCOMPLETE;
481		nd6_llinfo_settimer_locked(lle,
482		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
483		LLE_WUNLOCK(lle);
484
485		nd6_ns_output(ifp, NULL, &sin6->sin6_addr, NULL, 0);
486
487		return (EWOULDBLOCK);
488	}
489
490	if (lle->ln_state == ND6_LLINFO_STALE) {
491		if ((flags & LLE_EXCLUSIVE) == 0) {
492			LLE_RUNLOCK(lle);
493			flags |= LLE_EXCLUSIVE;
494			goto restart;
495		}
496
497		LLE_WLOCK_ASSERT(lle);
498
499		lle->la_asked = 0;
500		lle->ln_state = ND6_LLINFO_DELAY;
501		nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
502	}
503
504	if (lle->la_flags & LLE_VALID) {
505		memcpy(lladdr, &lle->ll_addr, ifp->if_addrlen);
506		rc = 0;
507	} else
508		rc = EWOULDBLOCK;
509
510	if (flags & LLE_EXCLUSIVE)
511		LLE_WUNLOCK(lle);
512	else
513		LLE_RUNLOCK(lle);
514
515	return (rc);
516}
517#endif
518
519/*
520 * Returns 0 or EWOULDBLOCK on success (any other value is an error).  0 means
521 * lladdr and vtag are valid on return, EWOULDBLOCK means the TOE driver's
522 * tod_l2_update will be called later, when the entry is resolved or times out.
523 */
524int
525toe_l2_resolve(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa,
526    uint8_t *lladdr, uint16_t *vtag)
527{
528#ifdef INET
529	struct llentry *lle;
530#endif
531	int rc;
532
533	switch (sa->sa_family) {
534#ifdef INET
535	case AF_INET:
536		rc = arpresolve(ifp, NULL, NULL, sa, lladdr, &lle);
537		break;
538#endif
539#ifdef INET6
540	case AF_INET6:
541		rc = toe_nd6_resolve(ifp, sa, lladdr);
542		break;
543#endif
544	default:
545		return (EPROTONOSUPPORT);
546	}
547
548	if (rc == 0) {
549#ifdef VLAN_TAG
550		if (VLAN_TAG(ifp, vtag) != 0)
551#endif
552			*vtag = 0xfff;
553	}
554
555	return (rc);
556}
557
558void
559toe_connect_failed(struct toedev *tod, struct inpcb *inp, int err)
560{
561
562	INP_WLOCK_ASSERT(inp);
563
564	if (!(inp->inp_flags & INP_DROPPED)) {
565		struct tcpcb *tp = intotcpcb(inp);
566
567		KASSERT(tp->t_flags & TF_TOE,
568		    ("%s: tp %p not offloaded.", __func__, tp));
569
570		if (err == EAGAIN) {
571
572			/*
573			 * Temporary failure during offload, take this PCB back.
574			 * Detach from the TOE driver and do the rest of what
575			 * TCP's pru_connect would have done if the connection
576			 * wasn't offloaded.
577			 */
578
579			tod->tod_pcb_detach(tod, tp);
580			KASSERT(!(tp->t_flags & TF_TOE),
581			    ("%s: tp %p still offloaded.", __func__, tp));
582			tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
583			(void) tcp_output(tp);
584		} else {
585
586			INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
587			tp = tcp_drop(tp, err);
588			if (tp == NULL)
589				INP_WLOCK(inp);	/* re-acquire */
590		}
591	}
592	INP_WLOCK_ASSERT(inp);
593}
594
595static int
596toecore_load(void)
597{
598
599	mtx_init(&toedev_lock, "toedev lock", NULL, MTX_DEF);
600	TAILQ_INIT(&toedev_list);
601
602	listen_start_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_start,
603	    toe_listen_start_event, NULL, EVENTHANDLER_PRI_ANY);
604	listen_stop_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_stop,
605	    toe_listen_stop_event, NULL, EVENTHANDLER_PRI_ANY);
606	lle_event_eh = EVENTHANDLER_REGISTER(lle_event, toe_lle_event, NULL,
607	    EVENTHANDLER_PRI_ANY);
608	route_redirect_eh = EVENTHANDLER_REGISTER(route_redirect_event,
609	    toe_route_redirect_event, NULL, EVENTHANDLER_PRI_ANY);
610
611	return (0);
612}
613
614static int
615toecore_unload(void)
616{
617
618	mtx_lock(&toedev_lock);
619	if (!TAILQ_EMPTY(&toedev_list)) {
620		mtx_unlock(&toedev_lock);
621		return (EBUSY);
622	}
623
624	EVENTHANDLER_DEREGISTER(tcp_offload_listen_start, listen_start_eh);
625	EVENTHANDLER_DEREGISTER(tcp_offload_listen_stop, listen_stop_eh);
626	EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
627	EVENTHANDLER_DEREGISTER(route_redirect_event, route_redirect_eh);
628
629	mtx_unlock(&toedev_lock);
630	mtx_destroy(&toedev_lock);
631
632	return (0);
633}
634
635static int
636toecore_mod_handler(module_t mod, int cmd, void *arg)
637{
638
639	if (cmd == MOD_LOAD)
640		return (toecore_load());
641
642	if (cmd == MOD_UNLOAD)
643		return (toecore_unload());
644
645	return (EOPNOTSUPP);
646}
647
648static moduledata_t mod_data= {
649	"toecore",
650	toecore_mod_handler,
651	0
652};
653
654DECLARE_MODULE(toecore, mod_data, SI_SUB_EXEC, SI_ORDER_ANY);
655#endif /* defined(KLD_MODULE) || defined(INET) || defined(INET6) */
656
657MODULE_VERSION(toecore, 1);
658
659