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("$FreeBSD$");
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/malloc.h>
38#include <sys/mbuf.h>
39#include <sys/module.h>
40#include <sys/types.h>
41#include <sys/sockopt.h>
42#include <sys/sysctl.h>
43#include <sys/socket.h>
44
45#include <net/ethernet.h>
46#include <net/if.h>
47#include <net/if_var.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;
74
75static int
76toedev_connect(struct toedev *tod __unused, struct socket *so __unused,
77    struct rtentry *rt __unused, struct sockaddr *nam __unused)
78{
79
80	return (ENOTSUP);
81}
82
83static int
84toedev_listen_start(struct toedev *tod __unused, struct tcpcb *tp __unused)
85{
86
87	return (ENOTSUP);
88}
89
90static int
91toedev_listen_stop(struct toedev *tod __unused, struct tcpcb *tp __unused)
92{
93
94	return (ENOTSUP);
95}
96
97static void
98toedev_input(struct toedev *tod __unused, struct tcpcb *tp __unused,
99    struct mbuf *m)
100{
101
102	m_freem(m);
103	return;
104}
105
106static void
107toedev_rcvd(struct toedev *tod __unused, struct tcpcb *tp __unused)
108{
109
110	return;
111}
112
113static int
114toedev_output(struct toedev *tod __unused, struct tcpcb *tp __unused)
115{
116
117	return (ENOTSUP);
118}
119
120static void
121toedev_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp __unused)
122{
123
124	return;
125}
126
127static void
128toedev_l2_update(struct toedev *tod __unused, struct ifnet *ifp __unused,
129    struct sockaddr *sa __unused, uint8_t *lladdr __unused,
130    uint16_t vtag __unused)
131{
132
133	return;
134}
135
136static void
137toedev_route_redirect(struct toedev *tod __unused, struct ifnet *ifp __unused,
138    struct rtentry *rt0 __unused, struct rtentry *rt1 __unused)
139{
140
141	return;
142}
143
144static void
145toedev_syncache_added(struct toedev *tod __unused, void *ctx __unused)
146{
147
148	return;
149}
150
151static void
152toedev_syncache_removed(struct toedev *tod __unused, void *ctx __unused)
153{
154
155	return;
156}
157
158static int
159toedev_syncache_respond(struct toedev *tod __unused, void *ctx __unused,
160    struct mbuf *m)
161{
162
163	m_freem(m);
164	return (0);
165}
166
167static void
168toedev_offload_socket(struct toedev *tod __unused, void *ctx __unused,
169    struct socket *so __unused)
170{
171
172	return;
173}
174
175static void
176toedev_ctloutput(struct toedev *tod __unused, struct tcpcb *tp __unused,
177    int sopt_dir __unused, int sopt_name __unused)
178{
179
180	return;
181}
182
183/*
184 * Inform one or more TOE devices about a listening socket.
185 */
186static void
187toe_listen_start(struct inpcb *inp, void *arg)
188{
189	struct toedev *t, *tod;
190	struct tcpcb *tp;
191
192	INP_WLOCK_ASSERT(inp);
193	KASSERT(inp->inp_pcbinfo == &V_tcbinfo,
194	    ("%s: inp is not a TCP inp", __func__));
195
196	if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))
197		return;
198
199	tp = intotcpcb(inp);
200	if (tp->t_state != TCPS_LISTEN)
201		return;
202
203	t = arg;
204	mtx_lock(&toedev_lock);
205	TAILQ_FOREACH(tod, &toedev_list, link) {
206		if (t == NULL || t == tod)
207			tod->tod_listen_start(tod, tp);
208	}
209	mtx_unlock(&toedev_lock);
210}
211
212static void
213toe_listen_start_event(void *arg __unused, struct tcpcb *tp)
214{
215	struct inpcb *inp = tp->t_inpcb;
216
217	INP_WLOCK_ASSERT(inp);
218	KASSERT(tp->t_state == TCPS_LISTEN,
219	    ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
220
221	toe_listen_start(inp, NULL);
222}
223
224static void
225toe_listen_stop_event(void *arg __unused, struct tcpcb *tp)
226{
227	struct toedev *tod;
228#ifdef INVARIANTS
229	struct inpcb *inp = tp->t_inpcb;
230#endif
231
232	INP_WLOCK_ASSERT(inp);
233	KASSERT(tp->t_state == TCPS_LISTEN,
234	    ("%s: t_state %s", __func__, tcpstates[tp->t_state]));
235
236	mtx_lock(&toedev_lock);
237	TAILQ_FOREACH(tod, &toedev_list, link)
238	    tod->tod_listen_stop(tod, tp);
239	mtx_unlock(&toedev_lock);
240}
241
242/*
243 * Fill up a freshly allocated toedev struct with reasonable defaults.
244 */
245void
246init_toedev(struct toedev *tod)
247{
248
249	tod->tod_softc = NULL;
250
251	/*
252	 * Provide no-op defaults so that the kernel can call any toedev
253	 * function without having to check whether the TOE driver supplied one
254	 * or not.
255	 */
256	tod->tod_connect = toedev_connect;
257	tod->tod_listen_start = toedev_listen_start;
258	tod->tod_listen_stop = toedev_listen_stop;
259	tod->tod_input = toedev_input;
260	tod->tod_rcvd = toedev_rcvd;
261	tod->tod_output = toedev_output;
262	tod->tod_send_rst = toedev_output;
263	tod->tod_send_fin = toedev_output;
264	tod->tod_pcb_detach = toedev_pcb_detach;
265	tod->tod_l2_update = toedev_l2_update;
266	tod->tod_route_redirect = toedev_route_redirect;
267	tod->tod_syncache_added = toedev_syncache_added;
268	tod->tod_syncache_removed = toedev_syncache_removed;
269	tod->tod_syncache_respond = toedev_syncache_respond;
270	tod->tod_offload_socket = toedev_offload_socket;
271	tod->tod_ctloutput = toedev_ctloutput;
272}
273
274/*
275 * Register an active TOE device with the system.  This allows it to receive
276 * notifications from the kernel.
277 */
278int
279register_toedev(struct toedev *tod)
280{
281	struct toedev *t;
282
283	mtx_lock(&toedev_lock);
284	TAILQ_FOREACH(t, &toedev_list, link) {
285		if (t == tod) {
286			mtx_unlock(&toedev_lock);
287			return (EEXIST);
288		}
289	}
290
291	TAILQ_INSERT_TAIL(&toedev_list, tod, link);
292	registered_toedevs++;
293	mtx_unlock(&toedev_lock);
294
295	inp_apply_all(toe_listen_start, tod);
296
297	return (0);
298}
299
300/*
301 * Remove the TOE device from the global list of active TOE devices.  It is the
302 * caller's responsibility to ensure that the TOE device is quiesced prior to
303 * this call.
304 */
305int
306unregister_toedev(struct toedev *tod)
307{
308	struct toedev *t, *t2;
309	int rc = ENODEV;
310
311	mtx_lock(&toedev_lock);
312	TAILQ_FOREACH_SAFE(t, &toedev_list, link, t2) {
313		if (t == tod) {
314			TAILQ_REMOVE(&toedev_list, tod, link);
315			registered_toedevs--;
316			rc = 0;
317			break;
318		}
319	}
320	KASSERT(registered_toedevs >= 0,
321	    ("%s: registered_toedevs (%d) < 0", __func__, registered_toedevs));
322	mtx_unlock(&toedev_lock);
323	return (rc);
324}
325
326void
327toe_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
328    struct inpcb *inp, void *tod, void *todctx)
329{
330	struct socket *lso = inp->inp_socket;
331
332	INP_WLOCK_ASSERT(inp);
333
334	syncache_add(inc, to, th, inp, &lso, NULL, tod, todctx);
335}
336
337int
338toe_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
339    struct tcphdr *th, struct socket **lsop)
340{
341
342	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
343
344	return (syncache_expand(inc, to, th, lsop, NULL));
345}
346
347/*
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.
351 *
352 * Note that the TCP header must have been run through tcp_fields_to_host() or
353 * equivalent.
354 */
355int
356toe_4tuple_check(struct in_conninfo *inc, struct tcphdr *th, struct ifnet *ifp)
357{
358	struct inpcb *inp;
359
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);
364	} else {
365		inp = in_pcblookup(&V_tcbinfo, inc->inc_faddr, inc->inc_fport,
366		    inc->inc_laddr, inc->inc_lport, INPLOOKUP_WLOCKPCB, ifp);
367	}
368	if (inp != NULL) {
369		INP_WLOCK_ASSERT(inp);
370
371		if ((inp->inp_flags & INP_TIMEWAIT) && th != NULL) {
372
373			INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* for twcheck */
374			if (!tcp_twcheck(inp, NULL, th, NULL, 0))
375				return (EADDRINUSE);
376		} else {
377			INP_WUNLOCK(inp);
378			return (EADDRINUSE);
379		}
380	}
381
382	return (0);
383}
384
385static void
386toe_lle_event(void *arg __unused, struct llentry *lle, int evt)
387{
388	struct toedev *tod;
389	struct ifnet *ifp;
390	struct sockaddr *sa;
391	uint8_t *lladdr;
392	uint16_t vtag;
393	int family;
394	struct sockaddr_in6 sin6;
395
396	LLE_WLOCK_ASSERT(lle);
397
398	ifp = lltable_get_ifp(lle->lle_tbl);
399	family = lltable_get_af(lle->lle_tbl);
400
401	if (family != AF_INET && family != AF_INET6)
402		return;
403	/*
404	 * Not interested if the interface's TOE capability is not enabled.
405	 */
406	if ((family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4)) ||
407	    (family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6)))
408		return;
409
410	tod = TOEDEV(ifp);
411	if (tod == NULL)
412		return;
413
414	sa = (struct sockaddr *)&sin6;
415	lltable_fill_sa_entry(lle, sa);
416
417	vtag = 0xfff;
418	if (evt != LLENTRY_RESOLVED) {
419
420		/*
421		 * LLENTRY_TIMEDOUT, LLENTRY_DELETED, LLENTRY_EXPIRED all mean
422		 * this entry is going to be deleted.
423		 */
424
425		lladdr = NULL;
426	} else {
427
428		KASSERT(lle->la_flags & LLE_VALID,
429		    ("%s: %p resolved but not valid?", __func__, lle));
430
431		lladdr = (uint8_t *)lle->ll_addr;
432#ifdef VLAN_TAG
433		VLAN_TAG(ifp, &vtag);
434#endif
435	}
436
437	tod->tod_l2_update(tod, ifp, sa, lladdr, vtag);
438}
439
440/*
441 * Returns 0 or EWOULDBLOCK on success (any other value is an error).  0 means
442 * lladdr and vtag are valid on return, EWOULDBLOCK means the TOE driver's
443 * tod_l2_update will be called later, when the entry is resolved or times out.
444 */
445int
446toe_l2_resolve(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa,
447    uint8_t *lladdr, uint16_t *vtag)
448{
449	int rc;
450
451	switch (sa->sa_family) {
452#ifdef INET
453	case AF_INET:
454		rc = arpresolve(ifp, 0, NULL, sa, lladdr, NULL, NULL);
455		break;
456#endif
457#ifdef INET6
458	case AF_INET6:
459		rc = nd6_resolve(ifp, 0, NULL, sa, lladdr, NULL, NULL);
460		break;
461#endif
462	default:
463		return (EPROTONOSUPPORT);
464	}
465
466	if (rc == 0) {
467#ifdef VLAN_TAG
468		if (VLAN_TAG(ifp, vtag) != 0)
469#endif
470			*vtag = 0xfff;
471	}
472
473	return (rc);
474}
475
476void
477toe_connect_failed(struct toedev *tod, struct inpcb *inp, int err)
478{
479
480	INP_WLOCK_ASSERT(inp);
481
482	if (!(inp->inp_flags & INP_DROPPED)) {
483		struct tcpcb *tp = intotcpcb(inp);
484
485		KASSERT(tp->t_flags & TF_TOE,
486		    ("%s: tp %p not offloaded.", __func__, tp));
487
488		if (err == EAGAIN) {
489
490			/*
491			 * Temporary failure during offload, take this PCB back.
492			 * Detach from the TOE driver and do the rest of what
493			 * TCP's pru_connect would have done if the connection
494			 * wasn't offloaded.
495			 */
496
497			tod->tod_pcb_detach(tod, tp);
498			KASSERT(!(tp->t_flags & TF_TOE),
499			    ("%s: tp %p still offloaded.", __func__, tp));
500			tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
501			(void) tp->t_fb->tfb_tcp_output(tp);
502		} else {
503
504			INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
505			tp = tcp_drop(tp, err);
506			if (tp == NULL)
507				INP_WLOCK(inp);	/* re-acquire */
508		}
509	}
510	INP_WLOCK_ASSERT(inp);
511}
512
513static int
514toecore_load(void)
515{
516
517	mtx_init(&toedev_lock, "toedev lock", NULL, MTX_DEF);
518	TAILQ_INIT(&toedev_list);
519
520	listen_start_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_start,
521	    toe_listen_start_event, NULL, EVENTHANDLER_PRI_ANY);
522	listen_stop_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_stop,
523	    toe_listen_stop_event, NULL, EVENTHANDLER_PRI_ANY);
524	lle_event_eh = EVENTHANDLER_REGISTER(lle_event, toe_lle_event, NULL,
525	    EVENTHANDLER_PRI_ANY);
526
527	return (0);
528}
529
530static int
531toecore_unload(void)
532{
533
534	mtx_lock(&toedev_lock);
535	if (!TAILQ_EMPTY(&toedev_list)) {
536		mtx_unlock(&toedev_lock);
537		return (EBUSY);
538	}
539
540	EVENTHANDLER_DEREGISTER(tcp_offload_listen_start, listen_start_eh);
541	EVENTHANDLER_DEREGISTER(tcp_offload_listen_stop, listen_stop_eh);
542	EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
543
544	mtx_unlock(&toedev_lock);
545	mtx_destroy(&toedev_lock);
546
547	return (0);
548}
549
550static int
551toecore_mod_handler(module_t mod, int cmd, void *arg)
552{
553
554	if (cmd == MOD_LOAD)
555		return (toecore_load());
556
557	if (cmd == MOD_UNLOAD)
558		return (toecore_unload());
559
560	return (EOPNOTSUPP);
561}
562
563static moduledata_t mod_data= {
564	"toecore",
565	toecore_mod_handler,
566	0
567};
568
569MODULE_VERSION(toecore, 1);
570DECLARE_MODULE(toecore, mod_data, SI_SUB_EXEC, SI_ORDER_ANY);
571