xref: /freebsd/sys/dev/cxgbe/tom/t4_tls.c (revision d869395ac4bd248da7c5bdc67afb19ca89fbeeee)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2017-2018 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: John Baldwin <jhb@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 "opt_inet.h"
31 #include "opt_kern_tls.h"
32 
33 #include <sys/cdefs.h>
34 #ifdef KERN_TLS
35 #include <sys/param.h>
36 #include <sys/ktr.h>
37 #include <sys/ktls.h>
38 #include <sys/sglist.h>
39 #include <sys/socket.h>
40 #include <sys/socketvar.h>
41 #include <sys/systm.h>
42 #include <netinet/in.h>
43 #include <netinet/in_pcb.h>
44 #include <netinet/tcp_var.h>
45 #include <netinet/toecore.h>
46 #include <opencrypto/cryptodev.h>
47 #include <opencrypto/xform.h>
48 
49 #ifdef TCP_OFFLOAD
50 #include "common/common.h"
51 #include "common/t4_tcb.h"
52 #include "crypto/t4_crypto.h"
53 #include "tom/t4_tom_l2t.h"
54 #include "tom/t4_tom.h"
55 
56 /*
57  * The TCP sequence number of a CPL_TLS_DATA mbuf is saved here while
58  * the mbuf is in the ulp_pdu_reclaimq.
59  */
60 #define	tls_tcp_seq	PH_loc.thirtytwo[0]
61 
62 static void
t4_set_tls_tcb_field(struct toepcb * toep,uint16_t word,uint64_t mask,uint64_t val,int reply,int cookie)63 t4_set_tls_tcb_field(struct toepcb *toep, uint16_t word, uint64_t mask,
64     uint64_t val, int reply, int cookie)
65 {
66 	struct adapter *sc = td_adapter(toep->td);
67 	struct mbuf *m;
68 
69 	m = alloc_raw_wr_mbuf(sizeof(struct cpl_set_tcb_field));
70 	if (m == NULL) {
71 		/* XXX */
72 		panic("%s: out of memory", __func__);
73 	}
74 
75 	write_set_tcb_field(sc, mtod(m, void *), toep, word, mask, val, reply,
76 	    cookie);
77 
78 	t4_raw_wr_tx(sc, toep, m);
79 }
80 
81 /* TLS and DTLS common routines */
82 bool
can_tls_offload(struct adapter * sc)83 can_tls_offload(struct adapter *sc)
84 {
85 
86 	return (sc->tt.tls && sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS);
87 }
88 
89 int
tls_tx_key(struct toepcb * toep)90 tls_tx_key(struct toepcb *toep)
91 {
92 	struct tls_ofld_info *tls_ofld = &toep->tls;
93 
94 	return (tls_ofld->tx_key_addr >= 0);
95 }
96 
97 /* Set TF_RX_QUIESCE to pause receive. */
98 static void
t4_set_rx_quiesce(struct toepcb * toep)99 t4_set_rx_quiesce(struct toepcb *toep)
100 {
101 
102 	t4_set_tls_tcb_field(toep, W_TCB_T_FLAGS, V_TF_RX_QUIESCE(1),
103 	    V_TF_RX_QUIESCE(1), 1, CPL_COOKIE_TOM);
104 }
105 
106 /* Clear TF_RX_QUIESCE to re-enable receive. */
107 static void
t4_clear_rx_quiesce(struct toepcb * toep)108 t4_clear_rx_quiesce(struct toepcb *toep)
109 {
110 
111 	t4_set_tls_tcb_field(toep, W_TCB_T_FLAGS, V_TF_RX_QUIESCE(1), 0, 0, 0);
112 }
113 
114 /* TLS/DTLS content type  for CPL SFO */
115 static inline unsigned char
tls_content_type(unsigned char content_type)116 tls_content_type(unsigned char content_type)
117 {
118 	switch (content_type) {
119 	case CONTENT_TYPE_CCS:
120 		return CPL_TX_TLS_SFO_TYPE_CCS;
121 	case CONTENT_TYPE_ALERT:
122 		return CPL_TX_TLS_SFO_TYPE_ALERT;
123 	case CONTENT_TYPE_HANDSHAKE:
124 		return CPL_TX_TLS_SFO_TYPE_HANDSHAKE;
125 	case CONTENT_TYPE_APP_DATA:
126 		return CPL_TX_TLS_SFO_TYPE_DATA;
127 	default:
128 		return CPL_TX_TLS_SFO_TYPE_CUSTOM;
129 	}
130 }
131 
132 /* TLS Key memory management */
133 static void
clear_tls_keyid(struct toepcb * toep)134 clear_tls_keyid(struct toepcb *toep)
135 {
136 	struct tls_ofld_info *tls_ofld = &toep->tls;
137 	struct adapter *sc = td_adapter(toep->td);
138 
139 	if (tls_ofld->rx_key_addr >= 0) {
140 		t4_free_tls_keyid(sc, tls_ofld->rx_key_addr);
141 		tls_ofld->rx_key_addr = -1;
142 	}
143 	if (tls_ofld->tx_key_addr >= 0) {
144 		t4_free_tls_keyid(sc, tls_ofld->tx_key_addr);
145 		tls_ofld->tx_key_addr = -1;
146 	}
147 }
148 
149 static int
get_tp_plen_max(struct ktls_session * tls)150 get_tp_plen_max(struct ktls_session *tls)
151 {
152 	int plen = ((min(3*4096, TP_TX_PG_SZ))/1448) * 1448;
153 
154 	return (tls->params.max_frame_len <= 8192 ? plen : FC_TP_PLEN_MAX);
155 }
156 
157 /* Send request to save the key in on-card memory. */
158 static int
tls_program_key_id(struct toepcb * toep,struct ktls_session * tls,int direction)159 tls_program_key_id(struct toepcb *toep, struct ktls_session *tls,
160     int direction)
161 {
162 	struct tls_ofld_info *tls_ofld = &toep->tls;
163 	struct adapter *sc = td_adapter(toep->td);
164 	int keyid;
165 	struct mbuf *m;
166 	struct tls_key_req *kwr;
167 	struct tls_keyctx *kctx;
168 
169 #ifdef INVARIANTS
170 	int kwrlen, kctxlen, len;
171 
172 	kwrlen = sizeof(*kwr);
173 	kctxlen = roundup2(sizeof(*kctx), 32);
174 	len = roundup2(kwrlen + kctxlen, 16);
175 	MPASS(TLS_KEY_WR_SZ == len);
176 #endif
177 	if (toep->txsd_avail == 0)
178 		return (EAGAIN);
179 
180 	if ((keyid = t4_alloc_tls_keyid(sc)) < 0) {
181 		return (ENOSPC);
182 	}
183 
184 	m = alloc_raw_wr_mbuf(TLS_KEY_WR_SZ);
185 	if (m == NULL) {
186 		t4_free_tls_keyid(sc, keyid);
187 		return (ENOMEM);
188 	}
189 	kwr = mtod(m, struct tls_key_req *);
190 	memset(kwr, 0, TLS_KEY_WR_SZ);
191 
192 	t4_write_tlskey_wr(tls, direction, toep->tid, F_FW_WR_COMPL, keyid,
193 	    kwr);
194 	kctx = (struct tls_keyctx *)(kwr + 1);
195 	if (direction == KTLS_TX)
196 		tls_ofld->tx_key_addr = keyid;
197 	else
198 		tls_ofld->rx_key_addr = keyid;
199 	t4_tls_key_ctx(tls, direction, kctx);
200 
201 	t4_raw_wr_tx(sc, toep, m);
202 
203 	return (0);
204 }
205 
206 int
tls_alloc_ktls(struct toepcb * toep,struct ktls_session * tls,int direction)207 tls_alloc_ktls(struct toepcb *toep, struct ktls_session *tls, int direction)
208 {
209 	struct adapter *sc = td_adapter(toep->td);
210 	int error, explicit_iv_size, mac_first;
211 
212 	if (!can_tls_offload(sc))
213 		return (EINVAL);
214 
215 	if (direction == KTLS_RX) {
216 		if (ulp_mode(toep) != ULP_MODE_NONE)
217 			return (EINVAL);
218 		if ((toep->flags & TPF_TLS_STARTING) != 0)
219 			return (EINVAL);
220 	} else {
221 		switch (ulp_mode(toep)) {
222 		case ULP_MODE_NONE:
223 		case ULP_MODE_TLS:
224 		case ULP_MODE_TCPDDP:
225 			break;
226 		default:
227 			return (EINVAL);
228 		}
229 	}
230 
231 	switch (tls->params.cipher_algorithm) {
232 	case CRYPTO_AES_CBC:
233 		/* XXX: Explicitly ignore any provided IV. */
234 		switch (tls->params.cipher_key_len) {
235 		case 128 / 8:
236 		case 192 / 8:
237 		case 256 / 8:
238 			break;
239 		default:
240 			return (EINVAL);
241 		}
242 		switch (tls->params.auth_algorithm) {
243 		case CRYPTO_SHA1_HMAC:
244 		case CRYPTO_SHA2_256_HMAC:
245 		case CRYPTO_SHA2_384_HMAC:
246 			break;
247 		default:
248 			return (EPROTONOSUPPORT);
249 		}
250 		explicit_iv_size = AES_BLOCK_LEN;
251 		mac_first = 1;
252 		break;
253 	case CRYPTO_AES_NIST_GCM_16:
254 		if (tls->params.iv_len != SALT_SIZE) {
255 			return (EINVAL);
256 		}
257 		switch (tls->params.cipher_key_len) {
258 		case 128 / 8:
259 		case 192 / 8:
260 		case 256 / 8:
261 			break;
262 		default:
263 			return (EINVAL);
264 		}
265 		explicit_iv_size = 8;
266 		mac_first = 0;
267 		break;
268 	default:
269 		return (EPROTONOSUPPORT);
270 	}
271 
272 	/* Only TLS 1.1 and TLS 1.2 are currently supported. */
273 	if (tls->params.tls_vmajor != TLS_MAJOR_VER_ONE ||
274 	    tls->params.tls_vminor < TLS_MINOR_VER_ONE ||
275 	    tls->params.tls_vminor > TLS_MINOR_VER_TWO) {
276 		return (EPROTONOSUPPORT);
277 	}
278 
279 	/* Bail if we already have a key. */
280 	if (direction == KTLS_TX) {
281 		if (toep->tls.tx_key_addr != -1)
282 			return (EOPNOTSUPP);
283 	} else {
284 		if (toep->tls.rx_key_addr != -1)
285 			return (EOPNOTSUPP);
286 	}
287 
288 	error = tls_program_key_id(toep, tls, direction);
289 	if (error)
290 		return (error);
291 
292 	if (direction == KTLS_TX) {
293 		toep->tls.scmd0.seqno_numivs =
294 			(V_SCMD_SEQ_NO_CTRL(3) |
295 			 V_SCMD_PROTO_VERSION(t4_tls_proto_ver(tls)) |
296 			 V_SCMD_ENC_DEC_CTRL(SCMD_ENCDECCTRL_ENCRYPT) |
297 			 V_SCMD_CIPH_AUTH_SEQ_CTRL((mac_first == 0)) |
298 			 V_SCMD_CIPH_MODE(t4_tls_cipher_mode(tls)) |
299 			 V_SCMD_AUTH_MODE(t4_tls_auth_mode(tls)) |
300 			 V_SCMD_HMAC_CTRL(t4_tls_hmac_ctrl(tls)) |
301 			 V_SCMD_IV_SIZE(explicit_iv_size / 2));
302 
303 		toep->tls.scmd0.ivgen_hdrlen =
304 			(V_SCMD_IV_GEN_CTRL(1) |
305 			 V_SCMD_KEY_CTX_INLINE(0) |
306 			 V_SCMD_TLS_FRAG_ENABLE(1));
307 
308 		toep->tls.iv_len = explicit_iv_size;
309 		toep->tls.frag_size = tls->params.max_frame_len;
310 		toep->tls.fcplenmax = get_tp_plen_max(tls);
311 		toep->tls.expn_per_ulp = tls->params.tls_hlen +
312 		    tls->params.tls_tlen;
313 		toep->tls.pdus_per_ulp = 1;
314 		toep->tls.adjusted_plen = toep->tls.expn_per_ulp +
315 		    tls->params.max_frame_len;
316 		toep->tls.tx_key_info_size = t4_tls_key_info_size(tls);
317 	} else {
318 		toep->flags |= TPF_TLS_STARTING | TPF_TLS_RX_QUIESCING;
319 		toep->tls.rx_version = tls->params.tls_vmajor << 8 |
320 		    tls->params.tls_vminor;
321 
322 		CTR2(KTR_CXGBE, "%s: tid %d setting RX_QUIESCE", __func__,
323 		    toep->tid);
324 		t4_set_rx_quiesce(toep);
325 	}
326 
327 	return (0);
328 }
329 
330 void
tls_init_toep(struct toepcb * toep)331 tls_init_toep(struct toepcb *toep)
332 {
333 	struct tls_ofld_info *tls_ofld = &toep->tls;
334 
335 	tls_ofld->rx_key_addr = -1;
336 	tls_ofld->tx_key_addr = -1;
337 }
338 
339 void
tls_uninit_toep(struct toepcb * toep)340 tls_uninit_toep(struct toepcb *toep)
341 {
342 
343 	clear_tls_keyid(toep);
344 }
345 
346 #define MAX_OFLD_TX_CREDITS (SGE_MAX_WR_LEN / 16)
347 #define	MIN_OFLD_TLSTX_CREDITS(toep)					\
348 	(howmany(sizeof(struct fw_tlstx_data_wr) +			\
349 	    sizeof(struct cpl_tx_tls_sfo) + sizeof(struct ulptx_idata) + \
350 	    sizeof(struct ulptx_sc_memrd) +				\
351 	    AES_BLOCK_LEN + 1, 16))
352 
353 static void
write_tlstx_wr(struct fw_tlstx_data_wr * txwr,struct toepcb * toep,unsigned int plen,unsigned int expn,uint8_t credits,int shove)354 write_tlstx_wr(struct fw_tlstx_data_wr *txwr, struct toepcb *toep,
355     unsigned int plen, unsigned int expn, uint8_t credits, int shove)
356 {
357 	struct tls_ofld_info *tls_ofld = &toep->tls;
358 	unsigned int len = plen + expn;
359 
360 	txwr->op_to_immdlen = htobe32(V_WR_OP(FW_TLSTX_DATA_WR) |
361 	    V_FW_TLSTX_DATA_WR_COMPL(1) |
362 	    V_FW_TLSTX_DATA_WR_IMMDLEN(0));
363 	txwr->flowid_len16 = htobe32(V_FW_TLSTX_DATA_WR_FLOWID(toep->tid) |
364 	    V_FW_TLSTX_DATA_WR_LEN16(credits));
365 	txwr->plen = htobe32(len);
366 	txwr->lsodisable_to_flags = htobe32(V_TX_ULP_MODE(ULP_MODE_TLS) |
367 	    V_TX_URG(0) | /* F_T6_TX_FORCE | */ V_TX_SHOVE(shove));
368 	txwr->ctxloc_to_exp = htobe32(V_FW_TLSTX_DATA_WR_NUMIVS(1) |
369 	    V_FW_TLSTX_DATA_WR_EXP(expn) |
370 	    V_FW_TLSTX_DATA_WR_CTXLOC(TLS_SFO_WR_CONTEXTLOC_DDR) |
371 	    V_FW_TLSTX_DATA_WR_IVDSGL(0) |
372 	    V_FW_TLSTX_DATA_WR_KEYSIZE(tls_ofld->tx_key_info_size >> 4));
373 	txwr->mfs = htobe16(tls_ofld->frag_size);
374 	txwr->adjustedplen_pkd = htobe16(
375 	    V_FW_TLSTX_DATA_WR_ADJUSTEDPLEN(tls_ofld->adjusted_plen));
376 	txwr->expinplenmax_pkd = htobe16(
377 	    V_FW_TLSTX_DATA_WR_EXPINPLENMAX(tls_ofld->expn_per_ulp));
378 	txwr->pdusinplenmax_pkd =
379 	    V_FW_TLSTX_DATA_WR_PDUSINPLENMAX(tls_ofld->pdus_per_ulp);
380 }
381 
382 static void
write_tlstx_cpl(struct cpl_tx_tls_sfo * cpl,struct toepcb * toep,struct tls_hdr * tls_hdr,unsigned int plen,uint64_t seqno)383 write_tlstx_cpl(struct cpl_tx_tls_sfo *cpl, struct toepcb *toep,
384     struct tls_hdr *tls_hdr, unsigned int plen, uint64_t seqno)
385 {
386 	struct tls_ofld_info *tls_ofld = &toep->tls;
387 	int data_type, seglen;
388 
389 	seglen = plen;
390 	data_type = tls_content_type(tls_hdr->type);
391 	cpl->op_to_seg_len = htobe32(V_CPL_TX_TLS_SFO_OPCODE(CPL_TX_TLS_SFO) |
392 	    V_CPL_TX_TLS_SFO_DATA_TYPE(data_type) |
393 	    V_CPL_TX_TLS_SFO_CPL_LEN(2) | V_CPL_TX_TLS_SFO_SEG_LEN(seglen));
394 	cpl->pld_len = htobe32(plen);
395 	if (data_type == CPL_TX_TLS_SFO_TYPE_CUSTOM)
396 		cpl->type_protover = htobe32(
397 		    V_CPL_TX_TLS_SFO_TYPE(tls_hdr->type));
398 	cpl->seqno_numivs = htobe32(tls_ofld->scmd0.seqno_numivs |
399 	    V_SCMD_NUM_IVS(1));
400 	cpl->ivgen_hdrlen = htobe32(tls_ofld->scmd0.ivgen_hdrlen);
401 	cpl->scmd1 = htobe64(seqno);
402 }
403 
404 static int
count_ext_pgs_segs(struct mbuf * m)405 count_ext_pgs_segs(struct mbuf *m)
406 {
407 	vm_paddr_t nextpa;
408 	u_int i, nsegs;
409 
410 	MPASS(m->m_epg_npgs > 0);
411 	nsegs = 1;
412 	nextpa = m->m_epg_pa[0] + PAGE_SIZE;
413 	for (i = 1; i < m->m_epg_npgs; i++) {
414 		if (nextpa != m->m_epg_pa[i])
415 			nsegs++;
416 		nextpa = m->m_epg_pa[i] + PAGE_SIZE;
417 	}
418 	return (nsegs);
419 }
420 
421 static void
write_ktlstx_sgl(void * dst,struct mbuf * m,int nsegs)422 write_ktlstx_sgl(void *dst, struct mbuf *m, int nsegs)
423 {
424 	struct ulptx_sgl *usgl = dst;
425 	vm_paddr_t pa;
426 	uint32_t len;
427 	int i, j;
428 
429 	KASSERT(nsegs > 0, ("%s: nsegs 0", __func__));
430 
431 	usgl->cmd_nsge = htobe32(V_ULPTX_CMD(ULP_TX_SC_DSGL) |
432 	    V_ULPTX_NSGE(nsegs));
433 
434 	/* Figure out the first S/G length. */
435 	pa = m->m_epg_pa[0] + m->m_epg_1st_off;
436 	usgl->addr0 = htobe64(pa);
437 	len = m_epg_pagelen(m, 0, m->m_epg_1st_off);
438 	pa += len;
439 	for (i = 1; i < m->m_epg_npgs; i++) {
440 		if (m->m_epg_pa[i] != pa)
441 			break;
442 		len += m_epg_pagelen(m, i, 0);
443 		pa += m_epg_pagelen(m, i, 0);
444 	}
445 	usgl->len0 = htobe32(len);
446 #ifdef INVARIANTS
447 	nsegs--;
448 #endif
449 
450 	j = -1;
451 	for (; i < m->m_epg_npgs; i++) {
452 		if (j == -1 || m->m_epg_pa[i] != pa) {
453 			if (j >= 0)
454 				usgl->sge[j / 2].len[j & 1] = htobe32(len);
455 			j++;
456 #ifdef INVARIANTS
457 			nsegs--;
458 #endif
459 			pa = m->m_epg_pa[i];
460 			usgl->sge[j / 2].addr[j & 1] = htobe64(pa);
461 			len = m_epg_pagelen(m, i, 0);
462 			pa += len;
463 		} else {
464 			len += m_epg_pagelen(m, i, 0);
465 			pa += m_epg_pagelen(m, i, 0);
466 		}
467 	}
468 	if (j >= 0) {
469 		usgl->sge[j / 2].len[j & 1] = htobe32(len);
470 
471 		if ((j & 1) == 0)
472 			usgl->sge[j / 2].len[1] = htobe32(0);
473 	}
474 	KASSERT(nsegs == 0, ("%s: nsegs %d, m %p", __func__, nsegs, m));
475 }
476 
477 /*
478  * Similar to t4_push_frames() but handles sockets that contain TLS
479  * record mbufs.
480  */
481 void
t4_push_ktls(struct adapter * sc,struct toepcb * toep,int drop)482 t4_push_ktls(struct adapter *sc, struct toepcb *toep, int drop)
483 {
484 	struct tls_hdr *thdr;
485 	struct fw_tlstx_data_wr *txwr;
486 	struct cpl_tx_tls_sfo *cpl;
487 	struct ulptx_idata *idata;
488 	struct ulptx_sc_memrd *memrd;
489 	struct wrqe *wr;
490 	struct mbuf *m;
491 	u_int nsegs, credits, wr_len;
492 	u_int expn_size;
493 	struct inpcb *inp = toep->inp;
494 	struct tcpcb *tp = intotcpcb(inp);
495 	struct socket *so = inp->inp_socket;
496 	struct sockbuf *sb = &so->so_snd;
497 	struct mbufq *pduq = &toep->ulp_pduq;
498 	int tls_size, tx_credits, shove, sowwakeup;
499 	struct ofld_tx_sdesc *txsd;
500 	char *buf;
501 
502 	INP_WLOCK_ASSERT(inp);
503 	KASSERT(toep->flags & TPF_FLOWC_WR_SENT,
504 	    ("%s: flowc_wr not sent for tid %u.", __func__, toep->tid));
505 
506 	KASSERT(ulp_mode(toep) == ULP_MODE_NONE ||
507 	    ulp_mode(toep) == ULP_MODE_TCPDDP || ulp_mode(toep) == ULP_MODE_TLS,
508 	    ("%s: ulp_mode %u for toep %p", __func__, ulp_mode(toep), toep));
509 	KASSERT(tls_tx_key(toep),
510 	    ("%s: TX key not set for toep %p", __func__, toep));
511 
512 #ifdef VERBOSE_TRACES
513 	CTR4(KTR_CXGBE, "%s: tid %d toep flags %#x tp flags %#x drop %d",
514 	    __func__, toep->tid, toep->flags, tp->t_flags);
515 #endif
516 	if (__predict_false(toep->flags & TPF_ABORT_SHUTDOWN))
517 		return;
518 
519 #ifdef RATELIMIT
520 	if (__predict_false(inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) &&
521 	    (update_tx_rate_limit(sc, toep, so->so_max_pacing_rate) == 0)) {
522 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
523 	}
524 #endif
525 
526 	/*
527 	 * This function doesn't resume by itself.  Someone else must clear the
528 	 * flag and call this function.
529 	 */
530 	if (__predict_false(toep->flags & TPF_TX_SUSPENDED)) {
531 		KASSERT(drop == 0,
532 		    ("%s: drop (%d) != 0 but tx is suspended", __func__, drop));
533 		return;
534 	}
535 
536 	txsd = &toep->txsd[toep->txsd_pidx];
537 	for (;;) {
538 		tx_credits = min(toep->tx_credits, MAX_OFLD_TX_CREDITS);
539 
540 		if (__predict_false((m = mbufq_first(pduq)) != NULL)) {
541 			if (!t4_push_raw_wr(sc, toep, m)) {
542 				toep->flags |= TPF_TX_SUSPENDED;
543 				return;
544 			}
545 
546 			(void)mbufq_dequeue(pduq);
547 
548 			txsd = &toep->txsd[toep->txsd_pidx];
549 			continue;
550 		}
551 
552 		SOCKBUF_LOCK(sb);
553 		sowwakeup = drop;
554 		if (drop) {
555 			sbdrop_locked(sb, drop);
556 			drop = 0;
557 		}
558 
559 		m = sb->sb_sndptr != NULL ? sb->sb_sndptr->m_next : sb->sb_mb;
560 
561 		/*
562 		 * Send a FIN if requested, but only if there's no
563 		 * more data to send.
564 		 */
565 		if (m == NULL && toep->flags & TPF_SEND_FIN) {
566 			if (sowwakeup)
567 				sowwakeup_locked(so);
568 			else
569 				SOCKBUF_UNLOCK(sb);
570 			SOCKBUF_UNLOCK_ASSERT(sb);
571 			t4_close_conn(sc, toep);
572 			return;
573 		}
574 
575 		/*
576 		 * If there is no ready data to send, wait until more
577 		 * data arrives.
578 		 */
579 		if (m == NULL || (m->m_flags & M_NOTREADY) != 0) {
580 			if (sowwakeup)
581 				sowwakeup_locked(so);
582 			else
583 				SOCKBUF_UNLOCK(sb);
584 			SOCKBUF_UNLOCK_ASSERT(sb);
585 #ifdef VERBOSE_TRACES
586 			CTR2(KTR_CXGBE, "%s: tid %d no ready data to send",
587 			    __func__, toep->tid);
588 #endif
589 			return;
590 		}
591 
592 		KASSERT(m->m_flags & M_EXTPG, ("%s: mbuf %p is not NOMAP",
593 		    __func__, m));
594 		KASSERT(m->m_epg_tls != NULL,
595 		    ("%s: mbuf %p doesn't have TLS session", __func__, m));
596 
597 		/* Calculate WR length. */
598 		wr_len = sizeof(struct fw_tlstx_data_wr) +
599 		    sizeof(struct cpl_tx_tls_sfo) +
600 		    sizeof(struct ulptx_idata) + sizeof(struct ulptx_sc_memrd);
601 
602 		/* Explicit IVs for AES-CBC and AES-GCM are <= 16. */
603 		MPASS(toep->tls.iv_len <= AES_BLOCK_LEN);
604 		wr_len += AES_BLOCK_LEN;
605 
606 		/* Account for SGL in work request length. */
607 		nsegs = count_ext_pgs_segs(m);
608 		wr_len += sizeof(struct ulptx_sgl) +
609 		    ((3 * (nsegs - 1)) / 2 + ((nsegs - 1) & 1)) * 8;
610 
611 		/* Not enough credits for this work request. */
612 		if (howmany(wr_len, 16) > tx_credits) {
613 			if (sowwakeup)
614 				sowwakeup_locked(so);
615 			else
616 				SOCKBUF_UNLOCK(sb);
617 			SOCKBUF_UNLOCK_ASSERT(sb);
618 #ifdef VERBOSE_TRACES
619 			CTR5(KTR_CXGBE,
620 	    "%s: tid %d mbuf %p requires %d credits, but only %d available",
621 			    __func__, toep->tid, m, howmany(wr_len, 16),
622 			    tx_credits);
623 #endif
624 			toep->flags |= TPF_TX_SUSPENDED;
625 			return;
626 		}
627 
628 		/* Shove if there is no additional data pending. */
629 		shove = ((m->m_next == NULL ||
630 		    (m->m_next->m_flags & M_NOTREADY) != 0)) &&
631 		    (tp->t_flags & TF_MORETOCOME) == 0;
632 
633 		if (sb->sb_flags & SB_AUTOSIZE &&
634 		    V_tcp_do_autosndbuf &&
635 		    sb->sb_hiwat < V_tcp_autosndbuf_max &&
636 		    sbused(sb) >= sb->sb_hiwat * 7 / 8) {
637 			int newsize = min(sb->sb_hiwat + V_tcp_autosndbuf_inc,
638 			    V_tcp_autosndbuf_max);
639 
640 			if (!sbreserve_locked(so, SO_SND, newsize, NULL))
641 				sb->sb_flags &= ~SB_AUTOSIZE;
642 			else
643 				sowwakeup = 1;	/* room available */
644 		}
645 		if (sowwakeup)
646 			sowwakeup_locked(so);
647 		else
648 			SOCKBUF_UNLOCK(sb);
649 		SOCKBUF_UNLOCK_ASSERT(sb);
650 
651 		if (__predict_false(toep->flags & TPF_FIN_SENT))
652 			panic("%s: excess tx.", __func__);
653 
654 		wr = alloc_wrqe(roundup2(wr_len, 16), &toep->ofld_txq->wrq);
655 		if (wr == NULL) {
656 			/* XXX: how will we recover from this? */
657 			toep->flags |= TPF_TX_SUSPENDED;
658 			return;
659 		}
660 
661 		thdr = (struct tls_hdr *)&m->m_epg_hdr;
662 #ifdef VERBOSE_TRACES
663 		CTR5(KTR_CXGBE, "%s: tid %d TLS record %ju type %d len %#x",
664 		    __func__, toep->tid, m->m_epg_seqno, thdr->type,
665 		    m->m_len);
666 #endif
667 		txwr = wrtod(wr);
668 		cpl = (struct cpl_tx_tls_sfo *)(txwr + 1);
669 		memset(txwr, 0, roundup2(wr_len, 16));
670 		credits = howmany(wr_len, 16);
671 		expn_size = m->m_epg_hdrlen +
672 		    m->m_epg_trllen;
673 		tls_size = m->m_len - expn_size;
674 		write_tlstx_wr(txwr, toep, tls_size, expn_size, credits, shove);
675 		write_tlstx_cpl(cpl, toep, thdr, tls_size, m->m_epg_seqno);
676 
677 		idata = (struct ulptx_idata *)(cpl + 1);
678 		idata->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
679 		idata->len = htobe32(0);
680 		memrd = (struct ulptx_sc_memrd *)(idata + 1);
681 		memrd->cmd_to_len = htobe32(V_ULPTX_CMD(ULP_TX_SC_MEMRD) |
682 		    V_ULP_TX_SC_MORE(1) |
683 		    V_ULPTX_LEN16(toep->tls.tx_key_info_size >> 4));
684 		memrd->addr = htobe32(toep->tls.tx_key_addr >> 5);
685 
686 		/* Copy IV. */
687 		buf = (char *)(memrd + 1);
688 		memcpy(buf, thdr + 1, toep->tls.iv_len);
689 		buf += AES_BLOCK_LEN;
690 
691 		write_ktlstx_sgl(buf, m, nsegs);
692 
693 		KASSERT(toep->tx_credits >= credits,
694 			("%s: not enough credits", __func__));
695 
696 		toep->tx_credits -= credits;
697 
698 		tp->snd_nxt += m->m_len;
699 		tp->snd_max += m->m_len;
700 
701 		SOCKBUF_LOCK(sb);
702 		sb->sb_sndptr = m;
703 		SOCKBUF_UNLOCK(sb);
704 
705 		toep->flags |= TPF_TX_DATA_SENT;
706 		if (toep->tx_credits < MIN_OFLD_TLSTX_CREDITS(toep))
707 			toep->flags |= TPF_TX_SUSPENDED;
708 
709 		KASSERT(toep->txsd_avail > 0, ("%s: no txsd", __func__));
710 		KASSERT(m->m_len <= MAX_OFLD_TX_SDESC_PLEN,
711 		    ("%s: plen %u too large", __func__, m->m_len));
712 		txsd->plen = m->m_len;
713 		txsd->tx_credits = credits;
714 		txsd++;
715 		if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) {
716 			toep->txsd_pidx = 0;
717 			txsd = &toep->txsd[0];
718 		}
719 		toep->txsd_avail--;
720 
721 		counter_u64_add(toep->ofld_txq->tx_toe_tls_records, 1);
722 		counter_u64_add(toep->ofld_txq->tx_toe_tls_octets, m->m_len);
723 
724 		t4_l2t_send(sc, wr, toep->l2te);
725 	}
726 }
727 
728 /*
729  * For TLS data we place received mbufs received via CPL_TLS_DATA into
730  * an mbufq in the TLS offload state.  When CPL_RX_TLS_CMP is
731  * received, the completed PDUs are placed into the socket receive
732  * buffer.
733  *
734  * The TLS code reuses the ulp_pdu_reclaimq to hold the pending mbufs.
735  */
736 static int
do_tls_data(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)737 do_tls_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
738 {
739 	struct adapter *sc = iq->adapter;
740 	const struct cpl_tls_data *cpl = mtod(m, const void *);
741 	unsigned int tid = GET_TID(cpl);
742 	struct toepcb *toep = lookup_tid(sc, tid);
743 	struct inpcb *inp = toep->inp;
744 	struct tcpcb *tp;
745 	int len;
746 
747 	/* XXX: Should this match do_rx_data instead? */
748 	KASSERT(!(toep->flags & TPF_SYNQE),
749 	    ("%s: toep %p claims to be a synq entry", __func__, toep));
750 
751 	KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
752 
753 	/* strip off CPL header */
754 	m_adj(m, sizeof(*cpl));
755 	len = m->m_pkthdr.len;
756 
757 	toep->ofld_rxq->rx_toe_tls_octets += len;
758 
759 	KASSERT(len == G_CPL_TLS_DATA_LENGTH(be32toh(cpl->length_pkd)),
760 	    ("%s: payload length mismatch", __func__));
761 
762 	INP_WLOCK(inp);
763 	if (inp->inp_flags & INP_DROPPED) {
764 		CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x",
765 		    __func__, tid, len, inp->inp_flags);
766 		INP_WUNLOCK(inp);
767 		m_freem(m);
768 		return (0);
769 	}
770 
771 	/* Save TCP sequence number. */
772 	m->m_pkthdr.tls_tcp_seq = be32toh(cpl->seq);
773 
774 	if (mbufq_enqueue(&toep->ulp_pdu_reclaimq, m)) {
775 #ifdef INVARIANTS
776 		panic("Failed to queue TLS data packet");
777 #else
778 		printf("%s: Failed to queue TLS data packet\n", __func__);
779 		INP_WUNLOCK(inp);
780 		m_freem(m);
781 		return (0);
782 #endif
783 	}
784 
785 	tp = intotcpcb(inp);
786 	tp->t_rcvtime = ticks;
787 
788 #ifdef VERBOSE_TRACES
789 	CTR4(KTR_CXGBE, "%s: tid %u len %d seq %u", __func__, tid, len,
790 	    be32toh(cpl->seq));
791 #endif
792 
793 	INP_WUNLOCK(inp);
794 	return (0);
795 }
796 
797 static int
do_rx_tls_cmp(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)798 do_rx_tls_cmp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
799 {
800 	struct adapter *sc = iq->adapter;
801 	const struct cpl_rx_tls_cmp *cpl = mtod(m, const void *);
802 	struct tlsrx_hdr_pkt *tls_hdr_pkt;
803 	unsigned int tid = GET_TID(cpl);
804 	struct toepcb *toep = lookup_tid(sc, tid);
805 	struct inpcb *inp = toep->inp;
806 	struct tcpcb *tp;
807 	struct socket *so;
808 	struct sockbuf *sb;
809 	struct mbuf *tls_data;
810 	struct tls_get_record *tgr;
811 	struct mbuf *control;
812 	int pdu_length, trailer_len;
813 #if defined(KTR) || defined(INVARIANTS)
814 	int len;
815 #endif
816 
817 	KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
818 	KASSERT(!(toep->flags & TPF_SYNQE),
819 	    ("%s: toep %p claims to be a synq entry", __func__, toep));
820 
821 	/* strip off CPL header */
822 	m_adj(m, sizeof(*cpl));
823 #if defined(KTR) || defined(INVARIANTS)
824 	len = m->m_pkthdr.len;
825 #endif
826 
827 	toep->ofld_rxq->rx_toe_tls_records++;
828 
829 	KASSERT(len == G_CPL_RX_TLS_CMP_LENGTH(be32toh(cpl->pdulength_length)),
830 	    ("%s: payload length mismatch", __func__));
831 
832 	INP_WLOCK(inp);
833 	if (inp->inp_flags & INP_DROPPED) {
834 		CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x",
835 		    __func__, tid, len, inp->inp_flags);
836 		INP_WUNLOCK(inp);
837 		m_freem(m);
838 		return (0);
839 	}
840 
841 	pdu_length = G_CPL_RX_TLS_CMP_PDULENGTH(be32toh(cpl->pdulength_length));
842 
843 	so = inp_inpcbtosocket(inp);
844 	tp = intotcpcb(inp);
845 
846 #ifdef VERBOSE_TRACES
847 	CTR6(KTR_CXGBE, "%s: tid %u PDU len %d len %d seq %u, rcv_nxt %u",
848 	    __func__, tid, pdu_length, len, be32toh(cpl->seq), tp->rcv_nxt);
849 #endif
850 
851 	tp->rcv_nxt += pdu_length;
852 	KASSERT(tp->rcv_wnd >= pdu_length,
853 	    ("%s: negative window size", __func__));
854 	tp->rcv_wnd -= pdu_length;
855 
856 	/* XXX: Not sure what to do about urgent data. */
857 
858 	/*
859 	 * The payload of this CPL is the TLS header followed by
860 	 * additional fields.
861 	 */
862 	KASSERT(m->m_len >= sizeof(*tls_hdr_pkt),
863 	    ("%s: payload too small", __func__));
864 	tls_hdr_pkt = mtod(m, void *);
865 
866 	tls_data = mbufq_dequeue(&toep->ulp_pdu_reclaimq);
867 	if (tls_data != NULL) {
868 		KASSERT(be32toh(cpl->seq) == tls_data->m_pkthdr.tls_tcp_seq,
869 		    ("%s: sequence mismatch", __func__));
870 	}
871 
872 	/* Report decryption errors as EBADMSG. */
873 	if ((tls_hdr_pkt->res_to_mac_error & M_TLSRX_HDR_PKT_ERROR) != 0) {
874 		CTR4(KTR_CXGBE, "%s: tid %u TLS error %#x ddp_vld %#x",
875 		    __func__, toep->tid, tls_hdr_pkt->res_to_mac_error,
876 		    be32toh(cpl->ddp_valid));
877 		m_freem(m);
878 		m_freem(tls_data);
879 
880 		CURVNET_SET(toep->vnet);
881 		so->so_error = EBADMSG;
882 		sorwakeup(so);
883 
884 		INP_WUNLOCK(inp);
885 		CURVNET_RESTORE();
886 
887 		return (0);
888 	}
889 
890 	/* Handle data received after the socket is closed. */
891 	sb = &so->so_rcv;
892 	SOCKBUF_LOCK(sb);
893 	if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) {
894 		struct epoch_tracker et;
895 
896 		CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)",
897 		    __func__, tid, pdu_length);
898 		m_freem(m);
899 		m_freem(tls_data);
900 		SOCKBUF_UNLOCK(sb);
901 		INP_WUNLOCK(inp);
902 
903 		CURVNET_SET(toep->vnet);
904 		NET_EPOCH_ENTER(et);
905 		INP_WLOCK(inp);
906 		tp = tcp_drop(tp, ECONNRESET);
907 		if (tp != NULL)
908 			INP_WUNLOCK(inp);
909 		NET_EPOCH_EXIT(et);
910 		CURVNET_RESTORE();
911 
912 		return (0);
913 	}
914 
915 	/*
916 	 * If there is any data in the 'sb_mtls' chain of the socket
917 	 * or we aren't able to allocate the control mbuf, append the
918 	 * record as a CSUM_TLS_DECRYPTED packet to 'sb_mtls' rather
919 	 * than as a decrypted record to 'sb_m'.
920 	 */
921 	if (sb->sb_mtls != NULL)
922 		control = NULL;
923 	else
924 		control = sbcreatecontrol(NULL, sizeof(*tgr), TLS_GET_RECORD,
925 		    IPPROTO_TCP, M_NOWAIT);
926 
927 	if (control != NULL) {
928 		tgr = (struct tls_get_record *)
929 		    CMSG_DATA(mtod(control, struct cmsghdr *));
930 		memset(tgr, 0, sizeof(*tgr));
931 		tgr->tls_type = tls_hdr_pkt->type;
932 		tgr->tls_vmajor = be16toh(tls_hdr_pkt->version) >> 8;
933 		tgr->tls_vminor = be16toh(tls_hdr_pkt->version) & 0xff;
934 		if (tls_data != NULL) {
935 			m_last(tls_data)->m_flags |= M_EOR;
936 			tgr->tls_length = htobe16(tls_data->m_pkthdr.len);
937 		} else
938 			tgr->tls_length = 0;
939 
940 		m_freem(m);
941 		m = tls_data;
942 	} else {
943 		M_ASSERTPKTHDR(m);
944 
945 		/* It's ok that any explicit IV is missing. */
946 		m->m_len = sb->sb_tls_info->params.tls_hlen;
947 		m->m_pkthdr.csum_flags |= CSUM_TLS_DECRYPTED;
948 		m->m_pkthdr.len = m->m_len;
949 		if (tls_data != NULL) {
950 			m->m_pkthdr.len += tls_data->m_pkthdr.len;
951 			m_demote_pkthdr(tls_data);
952 			m->m_next = tls_data;
953 		}
954 
955 		/*
956 		 * Grow the chain by the trailer, but without
957 		 * contents.  The trailer will be thrown away by
958 		 * ktls_decrypt.  Note that ktls_decrypt assumes the
959 		 * trailer is tls_tlen bytes long, so append that many
960 		 * bytes not the actual trailer size computed from
961 		 * pdu_length.
962 		 */
963 		trailer_len = sb->sb_tls_info->params.tls_tlen;
964 		if (tls_data != NULL) {
965 			m_last(tls_data)->m_len += trailer_len;
966 			tls_data = NULL;
967 		} else
968 			m->m_len += trailer_len;
969 		m->m_pkthdr.len += trailer_len;
970 		tls_hdr_pkt->length = htobe16(m->m_pkthdr.len -
971 		    sizeof(struct tls_record_layer));
972 	}
973 
974 	/* receive buffer autosize */
975 	MPASS(toep->vnet == so->so_vnet);
976 	CURVNET_SET(toep->vnet);
977 	if (sb->sb_flags & SB_AUTOSIZE &&
978 	    V_tcp_do_autorcvbuf &&
979 	    sb->sb_hiwat < V_tcp_autorcvbuf_max &&
980 	    m->m_pkthdr.len > (sbspace(sb) / 8 * 7)) {
981 		unsigned int hiwat = sb->sb_hiwat;
982 		unsigned int newsize = min(hiwat + sc->tt.autorcvbuf_inc,
983 		    V_tcp_autorcvbuf_max);
984 
985 		if (!sbreserve_locked(so, SO_RCV, newsize, NULL))
986 			sb->sb_flags &= ~SB_AUTOSIZE;
987 	}
988 
989 	if (control != NULL)
990 		sbappendcontrol_locked(sb, m, control, 0);
991 	else
992 		sbappendstream_locked(sb, m, 0);
993 	t4_rcvd_locked(&toep->td->tod, tp);
994 
995 	sorwakeup_locked(so);
996 	SOCKBUF_UNLOCK_ASSERT(sb);
997 
998 	INP_WUNLOCK(inp);
999 	CURVNET_RESTORE();
1000 	return (0);
1001 }
1002 
1003 void
do_rx_data_tls(const struct cpl_rx_data * cpl,struct toepcb * toep,struct mbuf * m)1004 do_rx_data_tls(const struct cpl_rx_data *cpl, struct toepcb *toep,
1005     struct mbuf *m)
1006 {
1007 	struct inpcb *inp = toep->inp;
1008 	struct tls_ofld_info *tls_ofld = &toep->tls;
1009 	struct tls_hdr *hdr;
1010 	struct tcpcb *tp;
1011 	struct socket *so;
1012 	struct sockbuf *sb;
1013 	int len;
1014 
1015 	len = m->m_pkthdr.len;
1016 
1017 	INP_WLOCK_ASSERT(inp);
1018 
1019 	so = inp_inpcbtosocket(inp);
1020 	tp = intotcpcb(inp);
1021 	sb = &so->so_rcv;
1022 	SOCKBUF_LOCK(sb);
1023 	CURVNET_SET(toep->vnet);
1024 
1025 	tp->rcv_nxt += len;
1026 	KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__));
1027 	tp->rcv_wnd -= len;
1028 
1029 	/* Do we have a full TLS header? */
1030 	if (len < sizeof(*hdr)) {
1031 		CTR3(KTR_CXGBE, "%s: tid %u len %d: too short for a TLS header",
1032 		    __func__, toep->tid, len);
1033 		so->so_error = EMSGSIZE;
1034 		goto out;
1035 	}
1036 	hdr = mtod(m, struct tls_hdr *);
1037 
1038 	/* Is the header valid? */
1039 	if (be16toh(hdr->version) != tls_ofld->rx_version) {
1040 		CTR3(KTR_CXGBE, "%s: tid %u invalid version %04x",
1041 		    __func__, toep->tid, be16toh(hdr->version));
1042 		so->so_error = EINVAL;
1043 		goto out;
1044 	}
1045 	if (be16toh(hdr->length) < sizeof(*hdr)) {
1046 		CTR3(KTR_CXGBE, "%s: tid %u invalid length %u",
1047 		    __func__, toep->tid, be16toh(hdr->length));
1048 		so->so_error = EBADMSG;
1049 		goto out;
1050 	}
1051 
1052 	/* Did we get a truncated record? */
1053 	if (len < be16toh(hdr->length)) {
1054 		CTR4(KTR_CXGBE, "%s: tid %u truncated TLS record (%d vs %u)",
1055 		    __func__, toep->tid, len, be16toh(hdr->length));
1056 
1057 		so->so_error = EMSGSIZE;
1058 		goto out;
1059 	}
1060 
1061 	/* Is the header type unknown? */
1062 	switch (hdr->type) {
1063 	case CONTENT_TYPE_CCS:
1064 	case CONTENT_TYPE_ALERT:
1065 	case CONTENT_TYPE_APP_DATA:
1066 	case CONTENT_TYPE_HANDSHAKE:
1067 		break;
1068 	default:
1069 		CTR3(KTR_CXGBE, "%s: tid %u invalid TLS record type %u",
1070 		    __func__, toep->tid, hdr->type);
1071 		so->so_error = EBADMSG;
1072 		goto out;
1073 	}
1074 
1075 	/*
1076 	 * Just punt.  Although this could fall back to software
1077 	 * decryption, this case should never really happen.
1078 	 */
1079 	CTR4(KTR_CXGBE, "%s: tid %u dropping TLS record type %u, length %u",
1080 	    __func__, toep->tid, hdr->type, be16toh(hdr->length));
1081 	so->so_error = EBADMSG;
1082 
1083 out:
1084 	sorwakeup_locked(so);
1085 	SOCKBUF_UNLOCK_ASSERT(sb);
1086 
1087 	INP_WUNLOCK(inp);
1088 	CURVNET_RESTORE();
1089 
1090 	m_freem(m);
1091 }
1092 
1093 /*
1094  * Send a work request setting multiple TCB fields to enable
1095  * ULP_MODE_TLS.
1096  */
1097 static void
tls_update_tcb(struct adapter * sc,struct toepcb * toep,uint64_t seqno)1098 tls_update_tcb(struct adapter *sc, struct toepcb *toep, uint64_t seqno)
1099 {
1100 	struct mbuf *m;
1101 	struct work_request_hdr *wrh;
1102 	struct ulp_txpkt *ulpmc;
1103 	int fields, key_offset, len;
1104 
1105 	KASSERT(ulp_mode(toep) == ULP_MODE_NONE,
1106 	    ("%s: tid %d already ULP_MODE_TLS", __func__, toep->tid));
1107 
1108 	fields = 0;
1109 
1110 	/* 2 writes for the overlay region */
1111 	fields += 2;
1112 
1113 	/* W_TCB_TLS_SEQ */
1114 	fields++;
1115 
1116 	/* W_TCB_ULP_RAW */
1117 	fields++;
1118 
1119 	/* W_TCB_ULP_TYPE */
1120 	fields ++;
1121 
1122 	/* W_TCB_T_FLAGS */
1123 	fields++;
1124 
1125 	len = sizeof(*wrh) + fields * roundup2(LEN__SET_TCB_FIELD_ULP, 16);
1126 	KASSERT(len <= SGE_MAX_WR_LEN,
1127 	    ("%s: WR with %d TCB field updates too large", __func__, fields));
1128 
1129 	m = alloc_raw_wr_mbuf(len);
1130 	if (m == NULL) {
1131 		/* XXX */
1132 		panic("%s: out of memory", __func__);
1133 	}
1134 
1135 	wrh = mtod(m, struct work_request_hdr *);
1136 	INIT_ULPTX_WRH(wrh, len, 1, toep->tid);	/* atomic */
1137 	ulpmc = (struct ulp_txpkt *)(wrh + 1);
1138 
1139 	/*
1140 	 * Clear the TLS overlay region: 1023:832.
1141 	 *
1142 	 * Words 26/27 are always set to zero.  Words 28/29
1143 	 * contain seqno and are set when enabling TLS
1144 	 * decryption.  Word 30 is zero and Word 31 contains
1145 	 * the keyid.
1146 	 */
1147 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, 26,
1148 	    0xffffffffffffffff, 0);
1149 
1150 	/*
1151 	 * RX key tags are an index into the key portion of MA
1152 	 * memory stored as an offset from the base address in
1153 	 * units of 64 bytes.
1154 	 */
1155 	key_offset = toep->tls.rx_key_addr - sc->vres.key.start;
1156 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, 30,
1157 	    0xffffffffffffffff,
1158 	    (uint64_t)V_TCB_RX_TLS_KEY_TAG(key_offset / 64) << 32);
1159 
1160 	CTR3(KTR_CXGBE, "%s: tid %d enable TLS seqno %lu", __func__,
1161 	    toep->tid, seqno);
1162 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, W_TCB_TLS_SEQ,
1163 	    V_TCB_TLS_SEQ(M_TCB_TLS_SEQ), V_TCB_TLS_SEQ(seqno));
1164 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, W_TCB_ULP_RAW,
1165 	    V_TCB_ULP_RAW(M_TCB_ULP_RAW),
1166 	    V_TCB_ULP_RAW((V_TF_TLS_KEY_SIZE(3) | V_TF_TLS_CONTROL(1) |
1167 	    V_TF_TLS_ACTIVE(1) | V_TF_TLS_ENABLE(1))));
1168 
1169 	toep->flags &= ~TPF_TLS_STARTING;
1170 	toep->flags |= TPF_TLS_RECEIVE;
1171 
1172 	/* Set the ULP mode to ULP_MODE_TLS. */
1173 	toep->params.ulp_mode = ULP_MODE_TLS;
1174 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, W_TCB_ULP_TYPE,
1175 	    V_TCB_ULP_TYPE(M_TCB_ULP_TYPE), V_TCB_ULP_TYPE(ULP_MODE_TLS));
1176 
1177 	/* Clear TF_RX_QUIESCE. */
1178 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, W_TCB_T_FLAGS,
1179 	    V_TF_RX_QUIESCE(1), 0);
1180 
1181 	t4_raw_wr_tx(sc, toep, m);
1182 }
1183 
1184 /*
1185  * Examine the pending data in the socket buffer and either enable TLS
1186  * RX or request more encrypted data.
1187  */
1188 static void
tls_check_rx_sockbuf(struct adapter * sc,struct toepcb * toep,struct sockbuf * sb)1189 tls_check_rx_sockbuf(struct adapter *sc, struct toepcb *toep,
1190     struct sockbuf *sb)
1191 {
1192 	uint64_t seqno;
1193 	size_t resid;
1194 	bool have_header;
1195 
1196 	SOCKBUF_LOCK_ASSERT(sb);
1197 	MPASS(toep->tls.rx_resid == 0);
1198 
1199 	have_header = ktls_pending_rx_info(sb, &seqno, &resid);
1200 	CTR5(KTR_CXGBE, "%s: tid %d have_header %d seqno %lu resid %zu",
1201 	    __func__, toep->tid, have_header, seqno, resid);
1202 
1203 	/*
1204 	 * If we have a partial header or we need fewer bytes than the
1205 	 * size of a TLS record, re-enable receive and pause again once
1206 	 * we get more data to try again.
1207 	 */
1208 	if (!have_header || resid != 0) {
1209 		CTR(KTR_CXGBE, "%s: tid %d waiting for more data", __func__,
1210 		    toep->tid);
1211 		toep->flags &= ~TPF_TLS_RX_QUIESCED;
1212 		t4_clear_rx_quiesce(toep);
1213 		return;
1214 	}
1215 
1216 	tls_update_tcb(sc, toep, seqno);
1217 }
1218 
1219 void
tls_received_starting_data(struct adapter * sc,struct toepcb * toep,struct sockbuf * sb,int len)1220 tls_received_starting_data(struct adapter *sc, struct toepcb *toep,
1221     struct sockbuf *sb, int len)
1222 {
1223 	MPASS(toep->flags & TPF_TLS_STARTING);
1224 
1225 	/* Data was received before quiescing took effect. */
1226 	if ((toep->flags & TPF_TLS_RX_QUIESCING) != 0)
1227 		return;
1228 
1229 	/*
1230 	 * A previous call to tls_check_rx_sockbuf needed more data.
1231 	 * Now that more data has arrived, quiesce receive again and
1232 	 * check the state once the quiesce has completed.
1233 	 */
1234 	if ((toep->flags & TPF_TLS_RX_QUIESCED) == 0) {
1235 		CTR(KTR_CXGBE, "%s: tid %d quiescing", __func__, toep->tid);
1236 		toep->flags |= TPF_TLS_RX_QUIESCING;
1237 		t4_set_rx_quiesce(toep);
1238 		return;
1239 	}
1240 
1241 	KASSERT(len <= toep->tls.rx_resid,
1242 	    ("%s: received excess bytes %d (waiting for %zu)", __func__, len,
1243 	    toep->tls.rx_resid));
1244 	toep->tls.rx_resid -= len;
1245 	if (toep->tls.rx_resid != 0)
1246 		return;
1247 
1248 	tls_check_rx_sockbuf(sc, toep, sb);
1249 }
1250 
1251 static int
do_tls_tcb_rpl(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)1252 do_tls_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
1253 {
1254 	struct adapter *sc = iq->adapter;
1255 	const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1);
1256 	unsigned int tid = GET_TID(cpl);
1257 	struct toepcb *toep;
1258 	struct inpcb *inp;
1259 	struct socket *so;
1260 	struct sockbuf *sb;
1261 
1262 	if (cpl->status != CPL_ERR_NONE)
1263 		panic("XXX: tcp_rpl failed: %d", cpl->status);
1264 
1265 	toep = lookup_tid(sc, tid);
1266 	inp = toep->inp;
1267 	switch (cpl->cookie) {
1268 	case V_WORD(W_TCB_T_FLAGS) | V_COOKIE(CPL_COOKIE_TOM):
1269 		INP_WLOCK(inp);
1270 		if ((toep->flags & TPF_TLS_STARTING) == 0)
1271 			panic("%s: connection is not starting TLS RX\n",
1272 			    __func__);
1273 		MPASS((toep->flags & TPF_TLS_RX_QUIESCING) != 0);
1274 
1275 		toep->flags &= ~TPF_TLS_RX_QUIESCING;
1276 		toep->flags |= TPF_TLS_RX_QUIESCED;
1277 
1278 		so = inp->inp_socket;
1279 		sb = &so->so_rcv;
1280 		SOCKBUF_LOCK(sb);
1281 		tls_check_rx_sockbuf(sc, toep, sb);
1282 		SOCKBUF_UNLOCK(sb);
1283 		INP_WUNLOCK(inp);
1284 		break;
1285 	default:
1286 		panic("XXX: unknown tcb_rpl offset %#x, cookie %#x",
1287 		    G_WORD(cpl->cookie), G_COOKIE(cpl->cookie));
1288 	}
1289 
1290 	return (0);
1291 }
1292 
1293 void
t4_tls_mod_load(void)1294 t4_tls_mod_load(void)
1295 {
1296 
1297 	t4_register_cpl_handler(CPL_TLS_DATA, do_tls_data);
1298 	t4_register_cpl_handler(CPL_RX_TLS_CMP, do_rx_tls_cmp);
1299 	t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, do_tls_tcb_rpl,
1300 	    CPL_COOKIE_TOM);
1301 }
1302 
1303 void
t4_tls_mod_unload(void)1304 t4_tls_mod_unload(void)
1305 {
1306 
1307 	t4_register_cpl_handler(CPL_TLS_DATA, NULL);
1308 	t4_register_cpl_handler(CPL_RX_TLS_CMP, NULL);
1309 	t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, NULL, CPL_COOKIE_TOM);
1310 }
1311 #endif	/* TCP_OFFLOAD */
1312 #endif	/* KERN_TLS */
1313