xref: /freebsd/sys/dev/cxgbe/tom/t4_connect.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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 #ifdef TCP_OFFLOAD
35 #include <sys/param.h>
36 #include <sys/types.h>
37 #include <sys/kernel.h>
38 #include <sys/ktr.h>
39 #include <sys/module.h>
40 #include <sys/protosw.h>
41 #include <sys/domain.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_types.h>
47 #include <net/if_vlan_var.h>
48 #include <net/route.h>
49 #include <netinet/in.h>
50 #include <netinet/in_pcb.h>
51 #include <netinet/ip.h>
52 #include <netinet/tcp_var.h>
53 #define TCPSTATES
54 #include <netinet/tcp_fsm.h>
55 #include <netinet/toecore.h>
56 
57 #include "common/common.h"
58 #include "common/t4_msg.h"
59 #include "common/t4_regs.h"
60 #include "tom/t4_tom_l2t.h"
61 #include "tom/t4_tom.h"
62 
63 /* atid services */
64 static int alloc_atid(struct adapter *, void *);
65 static void *lookup_atid(struct adapter *, int);
66 static void free_atid(struct adapter *, int);
67 
68 static int
69 alloc_atid(struct adapter *sc, void *ctx)
70 {
71 	struct tid_info *t = &sc->tids;
72 	int atid = -1;
73 
74 	mtx_lock(&t->atid_lock);
75 	if (t->afree) {
76 		union aopen_entry *p = t->afree;
77 
78 		atid = p - t->atid_tab;
79 		t->afree = p->next;
80 		p->data = ctx;
81 		t->atids_in_use++;
82 	}
83 	mtx_unlock(&t->atid_lock);
84 	return (atid);
85 }
86 
87 static void *
88 lookup_atid(struct adapter *sc, int atid)
89 {
90 	struct tid_info *t = &sc->tids;
91 
92 	return (t->atid_tab[atid].data);
93 }
94 
95 static void
96 free_atid(struct adapter *sc, int atid)
97 {
98 	struct tid_info *t = &sc->tids;
99 	union aopen_entry *p = &t->atid_tab[atid];
100 
101 	mtx_lock(&t->atid_lock);
102 	p->next = t->afree;
103 	t->afree = p;
104 	t->atids_in_use--;
105 	mtx_unlock(&t->atid_lock);
106 }
107 
108 /*
109  * Active open failed.
110  */
111 static int
112 do_act_establish(struct sge_iq *iq, const struct rss_header *rss,
113     struct mbuf *m)
114 {
115 	struct adapter *sc = iq->adapter;
116 	const struct cpl_act_establish *cpl = (const void *)(rss + 1);
117 	unsigned int tid = GET_TID(cpl);
118 	unsigned int atid = G_TID_TID(ntohl(cpl->tos_atid));
119 	struct toepcb *toep = lookup_atid(sc, atid);
120 	struct inpcb *inp = toep->inp;
121 
122 	KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
123 	KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__));
124 
125 	CTR3(KTR_CXGBE, "%s: atid %u, tid %u", __func__, atid, tid);
126 	free_atid(sc, atid);
127 
128 	INP_WLOCK(inp);
129 	toep->tid = tid;
130 	insert_tid(sc, tid, toep);
131 	if (inp->inp_flags & INP_DROPPED) {
132 
133 		/* socket closed by the kernel before hw told us it connected */
134 
135 		send_flowc_wr(toep, NULL);
136 		send_reset(sc, toep, be32toh(cpl->snd_isn));
137 		goto done;
138 	}
139 
140 	make_established(toep, cpl->snd_isn, cpl->rcv_isn, cpl->tcp_opt);
141 done:
142 	INP_WUNLOCK(inp);
143 	return (0);
144 }
145 
146 static inline int
147 act_open_has_tid(unsigned int status)
148 {
149 
150 	return (status != CPL_ERR_TCAM_FULL &&
151 	    status != CPL_ERR_TCAM_PARITY &&
152 	    status != CPL_ERR_CONN_EXIST &&
153 	    status != CPL_ERR_ARP_MISS);
154 }
155 
156 /*
157  * Convert an ACT_OPEN_RPL status to an errno.
158  */
159 static inline int
160 act_open_rpl_status_to_errno(int status)
161 {
162 
163 	switch (status) {
164 	case CPL_ERR_CONN_RESET:
165 		return (ECONNREFUSED);
166 	case CPL_ERR_ARP_MISS:
167 		return (EHOSTUNREACH);
168 	case CPL_ERR_CONN_TIMEDOUT:
169 		return (ETIMEDOUT);
170 	case CPL_ERR_TCAM_FULL:
171 		return (EAGAIN);
172 	case CPL_ERR_CONN_EXIST:
173 		log(LOG_ERR, "ACTIVE_OPEN_RPL: 4-tuple in use\n");
174 		return (EAGAIN);
175 	default:
176 		return (EIO);
177 	}
178 }
179 
180 static int
181 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss,
182     struct mbuf *m)
183 {
184 	struct adapter *sc = iq->adapter;
185 	const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1);
186 	unsigned int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status)));
187 	unsigned int status = G_AOPEN_STATUS(be32toh(cpl->atid_status));
188 	struct toepcb *toep = lookup_atid(sc, atid);
189 	struct inpcb *inp = toep->inp;
190 	struct toedev *tod = &toep->td->tod;
191 	int rc;
192 
193 	KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
194 	KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__));
195 
196 	CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status);
197 
198 	/* Ignore negative advice */
199 	if (negative_advice(status))
200 		return (0);
201 
202 	free_atid(sc, atid);
203 	toep->tid = -1;
204 
205 	if (status && act_open_has_tid(status))
206 		release_tid(sc, GET_TID(cpl), toep->ctrlq);
207 
208 	rc = act_open_rpl_status_to_errno(status);
209 	if (rc != EAGAIN)
210 		INP_INFO_WLOCK(&V_tcbinfo);
211 	INP_WLOCK(inp);
212 	toe_connect_failed(tod, inp, rc);
213 	final_cpl_received(toep);	/* unlocks inp */
214 	if (rc != EAGAIN)
215 		INP_INFO_WUNLOCK(&V_tcbinfo);
216 
217 	return (0);
218 }
219 
220 /*
221  * Options2 for active open.
222  */
223 static uint32_t
224 calc_opt2a(struct socket *so, struct toepcb *toep)
225 {
226 	struct tcpcb *tp = so_sototcpcb(so);
227 	struct port_info *pi = toep->port;
228 	struct adapter *sc = pi->adapter;
229 	uint32_t opt2;
230 
231 	opt2 = V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]) |
232 	    F_RSS_QUEUE_VALID | V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id);
233 
234 	if (tp->t_flags & TF_SACK_PERMIT)
235 		opt2 |= F_SACK_EN;
236 
237 	if (tp->t_flags & TF_REQ_TSTMP)
238 		opt2 |= F_TSTAMPS_EN;
239 
240 	if (tp->t_flags & TF_REQ_SCALE)
241 		opt2 |= F_WND_SCALE_EN;
242 
243 	if (V_tcp_do_ecn)
244 		opt2 |= F_CCTRL_ECN;
245 
246 	/* RX_COALESCE is always a valid value (M_RX_COALESCE). */
247 	if (is_t4(sc))
248 		opt2 |= F_RX_COALESCE_VALID;
249 	else
250 		opt2 |= F_T5_OPT_2_VALID;
251 	opt2 |= V_RX_COALESCE(M_RX_COALESCE);
252 
253 #ifdef USE_DDP_RX_FLOW_CONTROL
254 	if (toep->ulp_mode == ULP_MODE_TCPDDP)
255 		opt2 |= F_RX_FC_VALID | F_RX_FC_DDP;
256 #endif
257 
258 	return (htobe32(opt2));
259 }
260 
261 void
262 t4_init_connect_cpl_handlers(struct adapter *sc)
263 {
264 
265 	t4_register_cpl_handler(sc, CPL_ACT_ESTABLISH, do_act_establish);
266 	t4_register_cpl_handler(sc, CPL_ACT_OPEN_RPL, do_act_open_rpl);
267 }
268 
269 #define DONT_OFFLOAD_ACTIVE_OPEN(x)	do { \
270 	reason = __LINE__; \
271 	rc = (x); \
272 	goto failed; \
273 } while (0)
274 
275 static inline int
276 act_open_cpl_size(struct adapter *sc, int isipv6)
277 {
278 	static const int sz_t4[] = {
279 		sizeof (struct cpl_act_open_req),
280 		sizeof (struct cpl_act_open_req6)
281 	};
282 	static const int sz_t5[] = {
283 		sizeof (struct cpl_t5_act_open_req),
284 		sizeof (struct cpl_t5_act_open_req6)
285 	};
286 
287 	if (is_t4(sc))
288 		return (sz_t4[!!isipv6]);
289 	else
290 		return (sz_t5[!!isipv6]);
291 }
292 
293 /*
294  * active open (soconnect).
295  *
296  * State of affairs on entry:
297  * soisconnecting (so_state |= SS_ISCONNECTING)
298  * tcbinfo not locked (This has changed - used to be WLOCKed)
299  * inp WLOCKed
300  * tp->t_state = TCPS_SYN_SENT
301  * rtalloc1, RT_UNLOCK on rt.
302  */
303 int
304 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt,
305     struct sockaddr *nam)
306 {
307 	struct adapter *sc = tod->tod_softc;
308 	struct tom_data *td = tod_td(tod);
309 	struct toepcb *toep = NULL;
310 	struct wrqe *wr = NULL;
311 	struct ifnet *rt_ifp = rt->rt_ifp;
312 	struct port_info *pi;
313 	int mtu_idx, rscale, qid_atid, rc, isipv6;
314 	struct inpcb *inp = sotoinpcb(so);
315 	struct tcpcb *tp = intotcpcb(inp);
316 	int reason;
317 
318 	INP_WLOCK_ASSERT(inp);
319 	KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6,
320 	    ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family));
321 
322 	if (rt_ifp->if_type == IFT_ETHER)
323 		pi = rt_ifp->if_softc;
324 	else if (rt_ifp->if_type == IFT_L2VLAN) {
325 		struct ifnet *ifp = VLAN_COOKIE(rt_ifp);
326 
327 		pi = ifp->if_softc;
328 	} else if (rt_ifp->if_type == IFT_IEEE8023ADLAG)
329 		DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */
330 	else
331 		DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
332 
333 	toep = alloc_toepcb(pi, -1, -1, M_NOWAIT);
334 	if (toep == NULL)
335 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
336 
337 	toep->tid = alloc_atid(sc, toep);
338 	if (toep->tid < 0)
339 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
340 
341 	toep->l2te = t4_l2t_get(pi, rt_ifp,
342 	    rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam);
343 	if (toep->l2te == NULL)
344 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
345 
346 	isipv6 = nam->sa_family == AF_INET6;
347 	wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq);
348 	if (wr == NULL)
349 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
350 
351 	if (sc->tt.ddp && (so->so_options & SO_NO_DDP) == 0)
352 		set_tcpddp_ulp_mode(toep);
353 	else
354 		toep->ulp_mode = ULP_MODE_NONE;
355 	SOCKBUF_LOCK(&so->so_rcv);
356 	/* opt0 rcv_bufsiz initially, assumes its normal meaning later */
357 	toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ);
358 	SOCKBUF_UNLOCK(&so->so_rcv);
359 
360 	/*
361 	 * The kernel sets request_r_scale based on sb_max whereas we need to
362 	 * take hardware's MAX_RCV_WND into account too.  This is normally a
363 	 * no-op as MAX_RCV_WND is much larger than the default sb_max.
364 	 */
365 	if (tp->t_flags & TF_REQ_SCALE)
366 		rscale = tp->request_r_scale = select_rcv_wscale();
367 	else
368 		rscale = 0;
369 	mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, 0);
370 	qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid;
371 
372 	if (isipv6) {
373 		struct cpl_act_open_req6 *cpl = wrtod(wr);
374 
375 		if ((inp->inp_vflag & INP_IPV6) == 0) {
376 			/* XXX think about this a bit more */
377 			log(LOG_ERR,
378 			    "%s: time to think about AF_INET6 + vflag 0x%x.\n",
379 			    __func__, inp->inp_vflag);
380 			DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
381 		}
382 
383 		toep->ce = hold_lip(td, &inp->in6p_laddr);
384 		if (toep->ce == NULL)
385 			DONT_OFFLOAD_ACTIVE_OPEN(ENOENT);
386 
387 		INIT_TP_WR(cpl, 0);
388 		OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
389 		    qid_atid));
390 
391 		cpl->local_port = inp->inp_lport;
392 		cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0];
393 		cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8];
394 		cpl->peer_port = inp->inp_fport;
395 		cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0];
396 		cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8];
397 		cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale,
398 		    toep->rx_credits, toep->ulp_mode);
399 		cpl->opt2 = calc_opt2a(so, toep);
400 		if (is_t4(sc)) {
401 			cpl->params = select_ntuple(pi, toep->l2te,
402 			    sc->filter_mode);
403 		} else {
404 			struct cpl_t5_act_open_req6 *c5 = (void *)cpl;
405 
406 			c5->rsvd = 0;
407 			c5->params = select_ntuple(pi, toep->l2te,
408 			    sc->filter_mode);
409 		}
410 	} else {
411 		struct cpl_act_open_req *cpl = wrtod(wr);
412 
413 		INIT_TP_WR(cpl, 0);
414 		OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
415 		    qid_atid));
416 		inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port,
417 		    &cpl->peer_ip, &cpl->peer_port);
418 		cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale,
419 		    toep->rx_credits, toep->ulp_mode);
420 		cpl->opt2 = calc_opt2a(so, toep);
421 		if (is_t4(sc)) {
422 			cpl->params = select_ntuple(pi, toep->l2te,
423 			    sc->filter_mode);
424 		} else {
425 			struct cpl_t5_act_open_req6 *c5 = (void *)cpl;
426 
427 			c5->rsvd = 0;
428 			c5->params = select_ntuple(pi, toep->l2te,
429 			    sc->filter_mode);
430 		}
431 	}
432 
433 	CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__,
434 	    toep->tid, tcpstates[tp->t_state], toep, inp);
435 
436 	offload_socket(so, toep);
437 	rc = t4_l2t_send(sc, wr, toep->l2te);
438 	if (rc == 0) {
439 		toep->flags |= TPF_CPL_PENDING;
440 		return (0);
441 	}
442 
443 	undo_offload_socket(so);
444 	reason = __LINE__;
445 failed:
446 	CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc);
447 
448 	if (wr)
449 		free_wrqe(wr);
450 
451 	if (toep) {
452 		if (toep->tid >= 0)
453 			free_atid(sc, toep->tid);
454 		if (toep->l2te)
455 			t4_l2t_release(toep->l2te);
456 		if (toep->ce)
457 			release_lip(td, toep->ce);
458 		free_toepcb(toep);
459 	}
460 
461 	return (rc);
462 }
463 #endif
464