xref: /freebsd/sys/dev/cxgbe/tom/t4_connect.c (revision a25896ca1270e25b657ceaa8d47d5699515f5c25)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2012 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 
36 #ifdef TCP_OFFLOAD
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/ktr.h>
41 #include <sys/module.h>
42 #include <sys/protosw.h>
43 #include <sys/domain.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/sysctl.h>
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_types.h>
50 #include <net/if_vlan_var.h>
51 #include <net/route.h>
52 #include <netinet/in.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip.h>
55 #define TCPSTATES
56 #include <netinet/tcp_fsm.h>
57 #include <netinet/tcp_var.h>
58 #include <netinet/toecore.h>
59 #include <netinet/cc/cc.h>
60 
61 #include "common/common.h"
62 #include "common/t4_msg.h"
63 #include "common/t4_regs.h"
64 #include "common/t4_regs_values.h"
65 #include "tom/t4_tom_l2t.h"
66 #include "tom/t4_tom.h"
67 
68 /*
69  * Active open succeeded.
70  */
71 static int
72 do_act_establish(struct sge_iq *iq, const struct rss_header *rss,
73     struct mbuf *m)
74 {
75 	struct adapter *sc = iq->adapter;
76 	const struct cpl_act_establish *cpl = (const void *)(rss + 1);
77 	u_int tid = GET_TID(cpl);
78 	u_int atid = G_TID_TID(ntohl(cpl->tos_atid));
79 	struct toepcb *toep = lookup_atid(sc, atid);
80 	struct inpcb *inp = toep->inp;
81 
82 	KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
83 	KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__));
84 
85 	CTR3(KTR_CXGBE, "%s: atid %u, tid %u", __func__, atid, tid);
86 	free_atid(sc, atid);
87 
88 	CURVNET_SET(toep->vnet);
89 	INP_WLOCK(inp);
90 	toep->tid = tid;
91 	insert_tid(sc, tid, toep, inp->inp_vflag & INP_IPV6 ? 2 : 1);
92 	if (inp->inp_flags & INP_DROPPED) {
93 
94 		/* socket closed by the kernel before hw told us it connected */
95 
96 		send_flowc_wr(toep, NULL);
97 		send_reset(sc, toep, be32toh(cpl->snd_isn));
98 		goto done;
99 	}
100 
101 	make_established(toep, cpl->snd_isn, cpl->rcv_isn, cpl->tcp_opt);
102 
103 	if (toep->ulp_mode == ULP_MODE_TLS)
104 		tls_establish(toep);
105 
106 done:
107 	INP_WUNLOCK(inp);
108 	CURVNET_RESTORE();
109 	return (0);
110 }
111 
112 void
113 act_open_failure_cleanup(struct adapter *sc, u_int atid, u_int status)
114 {
115 	struct toepcb *toep = lookup_atid(sc, atid);
116 	struct inpcb *inp = toep->inp;
117 	struct toedev *tod = &toep->td->tod;
118 	struct epoch_tracker et;
119 
120 	free_atid(sc, atid);
121 	toep->tid = -1;
122 
123 	CURVNET_SET(toep->vnet);
124 	if (status != EAGAIN)
125 		INP_INFO_RLOCK_ET(&V_tcbinfo, et);
126 	INP_WLOCK(inp);
127 	toe_connect_failed(tod, inp, status);
128 	final_cpl_received(toep);	/* unlocks inp */
129 	if (status != EAGAIN)
130 		INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
131 	CURVNET_RESTORE();
132 }
133 
134 /*
135  * Active open failed.
136  */
137 static int
138 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss,
139     struct mbuf *m)
140 {
141 	struct adapter *sc = iq->adapter;
142 	const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1);
143 	u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status)));
144 	u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status));
145 	struct toepcb *toep = lookup_atid(sc, atid);
146 	int rc;
147 
148 	KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
149 	KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__));
150 
151 	CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status);
152 
153 	/* Ignore negative advice */
154 	if (negative_advice(status))
155 		return (0);
156 
157 	if (status && act_open_has_tid(status))
158 		release_tid(sc, GET_TID(cpl), toep->ctrlq);
159 
160 	rc = act_open_rpl_status_to_errno(status);
161 	act_open_failure_cleanup(sc, atid, rc);
162 
163 	return (0);
164 }
165 
166 /*
167  * Options2 for active open.
168  */
169 static uint32_t
170 calc_opt2a(struct socket *so, struct toepcb *toep,
171     const struct offload_settings *s)
172 {
173 	struct tcpcb *tp = so_sototcpcb(so);
174 	struct port_info *pi = toep->vi->pi;
175 	struct adapter *sc = pi->adapter;
176 	uint32_t opt2 = 0;
177 
178 	/*
179 	 * rx flow control, rx coalesce, congestion control, and tx pace are all
180 	 * explicitly set by the driver.  On T5+ the ISS is also set by the
181 	 * driver to the value picked by the kernel.
182 	 */
183 	if (is_t4(sc)) {
184 		opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID;
185 		opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID;
186 	} else {
187 		opt2 |= F_T5_OPT_2_VALID;	/* all 4 valid */
188 		opt2 |= F_T5_ISS;		/* ISS provided in CPL */
189 	}
190 
191 	if (s->sack > 0 || (s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT)))
192 		opt2 |= F_SACK_EN;
193 
194 	if (s->tstamp > 0 || (s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP)))
195 		opt2 |= F_TSTAMPS_EN;
196 
197 	if (tp->t_flags & TF_REQ_SCALE)
198 		opt2 |= F_WND_SCALE_EN;
199 
200 	if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1))
201 		opt2 |= F_CCTRL_ECN;
202 
203 	/* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */
204 
205 	opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]);
206 
207 	/* These defaults are subject to ULP specific fixups later. */
208 	opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0);
209 
210 	opt2 |= V_PACE(0);
211 
212 	if (s->cong_algo >= 0)
213 		opt2 |= V_CONG_CNTRL(s->cong_algo);
214 	else if (sc->tt.cong_algorithm >= 0)
215 		opt2 |= V_CONG_CNTRL(sc->tt.cong_algorithm & M_CONG_CNTRL);
216 	else {
217 		struct cc_algo *cc = CC_ALGO(tp);
218 
219 		if (strcasecmp(cc->name, "reno") == 0)
220 			opt2 |= V_CONG_CNTRL(CONG_ALG_RENO);
221 		else if (strcasecmp(cc->name, "tahoe") == 0)
222 			opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
223 		if (strcasecmp(cc->name, "newreno") == 0)
224 			opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO);
225 		if (strcasecmp(cc->name, "highspeed") == 0)
226 			opt2 |= V_CONG_CNTRL(CONG_ALG_HIGHSPEED);
227 		else {
228 			/*
229 			 * Use newreno in case the algorithm selected by the
230 			 * host stack is not supported by the hardware.
231 			 */
232 			opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO);
233 		}
234 	}
235 
236 	if (s->rx_coalesce > 0 || (s->rx_coalesce < 0 && sc->tt.rx_coalesce))
237 		opt2 |= V_RX_COALESCE(M_RX_COALESCE);
238 
239 	/* Note that ofld_rxq is already set according to s->rxq. */
240 	opt2 |= F_RSS_QUEUE_VALID;
241 	opt2 |= V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id);
242 
243 #ifdef USE_DDP_RX_FLOW_CONTROL
244 	if (toep->ulp_mode == ULP_MODE_TCPDDP)
245 		opt2 |= F_RX_FC_DDP;
246 #endif
247 
248 	if (toep->ulp_mode == ULP_MODE_TLS) {
249 		opt2 &= ~V_RX_COALESCE(M_RX_COALESCE);
250 		opt2 |= F_RX_FC_DISABLE;
251 	}
252 
253 	return (htobe32(opt2));
254 }
255 
256 void
257 t4_init_connect_cpl_handlers(void)
258 {
259 
260 	t4_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
261 	t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl,
262 	    CPL_COOKIE_TOM);
263 }
264 
265 void
266 t4_uninit_connect_cpl_handlers(void)
267 {
268 
269 	t4_register_cpl_handler(CPL_ACT_ESTABLISH, NULL);
270 	t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL, NULL, CPL_COOKIE_TOM);
271 }
272 
273 #define DONT_OFFLOAD_ACTIVE_OPEN(x)	do { \
274 	reason = __LINE__; \
275 	rc = (x); \
276 	goto failed; \
277 } while (0)
278 
279 static inline int
280 act_open_cpl_size(struct adapter *sc, int isipv6)
281 {
282 	int idx;
283 	static const int sz_table[3][2] = {
284 		{
285 			sizeof (struct cpl_act_open_req),
286 			sizeof (struct cpl_act_open_req6)
287 		},
288 		{
289 			sizeof (struct cpl_t5_act_open_req),
290 			sizeof (struct cpl_t5_act_open_req6)
291 		},
292 		{
293 			sizeof (struct cpl_t6_act_open_req),
294 			sizeof (struct cpl_t6_act_open_req6)
295 		},
296 	};
297 
298 	MPASS(chip_id(sc) >= CHELSIO_T4);
299 	idx = min(chip_id(sc) - CHELSIO_T4, 2);
300 
301 	return (sz_table[idx][!!isipv6]);
302 }
303 
304 /*
305  * active open (soconnect).
306  *
307  * State of affairs on entry:
308  * soisconnecting (so_state |= SS_ISCONNECTING)
309  * tcbinfo not locked (This has changed - used to be WLOCKed)
310  * inp WLOCKed
311  * tp->t_state = TCPS_SYN_SENT
312  * rtalloc1, RT_UNLOCK on rt.
313  */
314 int
315 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt,
316     struct sockaddr *nam)
317 {
318 	struct adapter *sc = tod->tod_softc;
319 	struct tom_data *td = tod_td(tod);
320 	struct toepcb *toep = NULL;
321 	struct wrqe *wr = NULL;
322 	struct ifnet *rt_ifp = rt->rt_ifp;
323 	struct vi_info *vi;
324 	int mtu_idx, rscale, qid_atid, rc, isipv6, txqid, rxqid;
325 	struct inpcb *inp = sotoinpcb(so);
326 	struct tcpcb *tp = intotcpcb(inp);
327 	int reason;
328 	struct offload_settings settings;
329 	uint16_t vid = 0xfff, pcp = 0;
330 
331 	INP_WLOCK_ASSERT(inp);
332 	KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6,
333 	    ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family));
334 
335 	if (rt_ifp->if_type == IFT_ETHER)
336 		vi = rt_ifp->if_softc;
337 	else if (rt_ifp->if_type == IFT_L2VLAN) {
338 		struct ifnet *ifp = VLAN_TRUNKDEV(rt_ifp);
339 
340 		vi = ifp->if_softc;
341 		VLAN_TAG(rt_ifp, &vid);
342 		VLAN_PCP(rt_ifp, &pcp);
343 	} else if (rt_ifp->if_type == IFT_IEEE8023ADLAG)
344 		DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */
345 	else
346 		DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
347 
348 	rw_rlock(&sc->policy_lock);
349 	settings = *lookup_offload_policy(sc, OPEN_TYPE_ACTIVE, NULL,
350 	    EVL_MAKETAG(vid, pcp, 0), inp);
351 	rw_runlock(&sc->policy_lock);
352 	if (!settings.offload)
353 		DONT_OFFLOAD_ACTIVE_OPEN(EPERM);
354 
355 	if (settings.txq >= 0 && settings.txq < vi->nofldtxq)
356 		txqid = settings.txq;
357 	else
358 		txqid = arc4random() % vi->nofldtxq;
359 	txqid += vi->first_ofld_txq;
360 	if (settings.rxq >= 0 && settings.rxq < vi->nofldrxq)
361 		rxqid = settings.rxq;
362 	else
363 		rxqid = arc4random() % vi->nofldrxq;
364 	rxqid += vi->first_ofld_rxq;
365 
366 	toep = alloc_toepcb(vi, txqid, rxqid, M_NOWAIT | M_ZERO);
367 	if (toep == NULL)
368 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
369 
370 	toep->tid = alloc_atid(sc, toep);
371 	if (toep->tid < 0)
372 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
373 
374 	toep->l2te = t4_l2t_get(vi->pi, rt_ifp,
375 	    rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam);
376 	if (toep->l2te == NULL)
377 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
378 
379 	isipv6 = nam->sa_family == AF_INET6;
380 	wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq);
381 	if (wr == NULL)
382 		DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
383 
384 	toep->vnet = so->so_vnet;
385 	set_ulp_mode(toep, select_ulp_mode(so, sc, &settings));
386 	SOCKBUF_LOCK(&so->so_rcv);
387 	/* opt0 rcv_bufsiz initially, assumes its normal meaning later */
388 	toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ);
389 	SOCKBUF_UNLOCK(&so->so_rcv);
390 
391 	/*
392 	 * The kernel sets request_r_scale based on sb_max whereas we need to
393 	 * take hardware's MAX_RCV_WND into account too.  This is normally a
394 	 * no-op as MAX_RCV_WND is much larger than the default sb_max.
395 	 */
396 	if (tp->t_flags & TF_REQ_SCALE)
397 		rscale = tp->request_r_scale = select_rcv_wscale();
398 	else
399 		rscale = 0;
400 	mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, &settings);
401 	qid_atid = V_TID_QID(toep->ofld_rxq->iq.abs_id) | V_TID_TID(toep->tid) |
402 	    V_TID_COOKIE(CPL_COOKIE_TOM);
403 
404 	if (isipv6) {
405 		struct cpl_act_open_req6 *cpl = wrtod(wr);
406 		struct cpl_t5_act_open_req6 *cpl5 = (void *)cpl;
407 		struct cpl_t6_act_open_req6 *cpl6 = (void *)cpl;
408 
409 		if ((inp->inp_vflag & INP_IPV6) == 0)
410 			DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
411 
412 		toep->ce = hold_lip(td, &inp->in6p_laddr, NULL);
413 		if (toep->ce == NULL)
414 			DONT_OFFLOAD_ACTIVE_OPEN(ENOENT);
415 
416 		switch (chip_id(sc)) {
417 		case CHELSIO_T4:
418 			INIT_TP_WR(cpl, 0);
419 			cpl->params = select_ntuple(vi, toep->l2te);
420 			break;
421 		case CHELSIO_T5:
422 			INIT_TP_WR(cpl5, 0);
423 			cpl5->iss = htobe32(tp->iss);
424 			cpl5->params = select_ntuple(vi, toep->l2te);
425 			break;
426 		case CHELSIO_T6:
427 		default:
428 			INIT_TP_WR(cpl6, 0);
429 			cpl6->iss = htobe32(tp->iss);
430 			cpl6->params = select_ntuple(vi, toep->l2te);
431 			break;
432 		}
433 		OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
434 		    qid_atid));
435 		cpl->local_port = inp->inp_lport;
436 		cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0];
437 		cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8];
438 		cpl->peer_port = inp->inp_fport;
439 		cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0];
440 		cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8];
441 		cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale,
442 		    toep->rx_credits, toep->ulp_mode, &settings);
443 		cpl->opt2 = calc_opt2a(so, toep, &settings);
444 	} else {
445 		struct cpl_act_open_req *cpl = wrtod(wr);
446 		struct cpl_t5_act_open_req *cpl5 = (void *)cpl;
447 		struct cpl_t6_act_open_req *cpl6 = (void *)cpl;
448 
449 		switch (chip_id(sc)) {
450 		case CHELSIO_T4:
451 			INIT_TP_WR(cpl, 0);
452 			cpl->params = select_ntuple(vi, toep->l2te);
453 			break;
454 		case CHELSIO_T5:
455 			INIT_TP_WR(cpl5, 0);
456 			cpl5->iss = htobe32(tp->iss);
457 			cpl5->params = select_ntuple(vi, toep->l2te);
458 			break;
459 		case CHELSIO_T6:
460 		default:
461 			INIT_TP_WR(cpl6, 0);
462 			cpl6->iss = htobe32(tp->iss);
463 			cpl6->params = select_ntuple(vi, toep->l2te);
464 			break;
465 		}
466 		OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
467 		    qid_atid));
468 		inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port,
469 		    &cpl->peer_ip, &cpl->peer_port);
470 		cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale,
471 		    toep->rx_credits, toep->ulp_mode, &settings);
472 		cpl->opt2 = calc_opt2a(so, toep, &settings);
473 	}
474 
475 	CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__,
476 	    toep->tid, tcpstates[tp->t_state], toep, inp);
477 
478 	offload_socket(so, toep);
479 	rc = t4_l2t_send(sc, wr, toep->l2te);
480 	if (rc == 0) {
481 		toep->flags |= TPF_CPL_PENDING;
482 		return (0);
483 	}
484 
485 	undo_offload_socket(so);
486 	reason = __LINE__;
487 failed:
488 	CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc);
489 
490 	if (wr)
491 		free_wrqe(wr);
492 
493 	if (toep) {
494 		if (toep->tid >= 0)
495 			free_atid(sc, toep->tid);
496 		if (toep->l2te)
497 			t4_l2t_release(toep->l2te);
498 		if (toep->ce)
499 			release_lip(td, toep->ce);
500 		free_toepcb(toep);
501 	}
502 
503 	return (rc);
504 }
505 #endif
506