xref: /freebsd/sys/dev/cxgbe/tom/t4_tls.c (revision 40dbb06fa73cac37d57563c07e55efd0cabbd488)
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, 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 	/* TLS 1.1 through TLS 1.3 are currently supported. */
232 	if (tls->params.tls_vmajor != TLS_MAJOR_VER_ONE ||
233 	    tls->params.tls_vminor < TLS_MINOR_VER_ONE ||
234 	    tls->params.tls_vminor > TLS_MINOR_VER_THREE) {
235 		return (EPROTONOSUPPORT);
236 	}
237 
238 	/* TLS 1.3 is only supported on T7+. */
239 	if (tls->params.tls_vminor == TLS_MINOR_VER_THREE) {
240 		if (is_t6(sc)) {
241 			return (EPROTONOSUPPORT);
242 		}
243 	}
244 
245 	/* Sanity check values in *tls. */
246 	switch (tls->params.cipher_algorithm) {
247 	case CRYPTO_AES_CBC:
248 		/* XXX: Explicitly ignore any provided IV. */
249 		switch (tls->params.cipher_key_len) {
250 		case 128 / 8:
251 		case 192 / 8:
252 		case 256 / 8:
253 			break;
254 		default:
255 			return (EINVAL);
256 		}
257 		switch (tls->params.auth_algorithm) {
258 		case CRYPTO_SHA1_HMAC:
259 		case CRYPTO_SHA2_256_HMAC:
260 		case CRYPTO_SHA2_384_HMAC:
261 			break;
262 		default:
263 			return (EPROTONOSUPPORT);
264 		}
265 		iv_size = AES_BLOCK_LEN;
266 		mac_first = 1;
267 		break;
268 	case CRYPTO_AES_NIST_GCM_16:
269 		switch (tls->params.cipher_key_len) {
270 		case 128 / 8:
271 		case 192 / 8:
272 		case 256 / 8:
273 			break;
274 		default:
275 			return (EINVAL);
276 		}
277 
278 		/*
279 		 * The IV size for TLS 1.2 is the explicit IV in the
280 		 * record header.  For TLS 1.3 it is the size of the
281 		 * sequence number.
282 		 */
283 		iv_size = 8;
284 		mac_first = 0;
285 		break;
286 	default:
287 		return (EPROTONOSUPPORT);
288 	}
289 
290 	/* Bail if we already have a key. */
291 	if (direction == KTLS_TX) {
292 		if (toep->tls.tx_key_addr != -1)
293 			return (EOPNOTSUPP);
294 	} else {
295 		if (toep->tls.rx_key_addr != -1)
296 			return (EOPNOTSUPP);
297 	}
298 
299 	error = tls_program_key_id(toep, tls, direction);
300 	if (error)
301 		return (error);
302 
303 	toep->tls.tls13 = tls->params.tls_vminor == TLS_MINOR_VER_THREE;
304 	if (direction == KTLS_TX) {
305 		toep->tls.scmd0.seqno_numivs =
306 			(V_SCMD_SEQ_NO_CTRL(3) |
307 			 V_SCMD_PROTO_VERSION(t4_tls_proto_ver(tls)) |
308 			 V_SCMD_ENC_DEC_CTRL(SCMD_ENCDECCTRL_ENCRYPT) |
309 			 V_SCMD_CIPH_AUTH_SEQ_CTRL((mac_first == 0)) |
310 			 V_SCMD_CIPH_MODE(t4_tls_cipher_mode(tls)) |
311 			 V_SCMD_AUTH_MODE(t4_tls_auth_mode(tls)) |
312 			 V_SCMD_HMAC_CTRL(t4_tls_hmac_ctrl(tls)) |
313 			 V_SCMD_IV_SIZE(iv_size / 2));
314 
315 		toep->tls.scmd0.ivgen_hdrlen =
316 			(V_SCMD_IV_GEN_CTRL(1) |
317 			 V_SCMD_KEY_CTX_INLINE(0) |
318 			 V_SCMD_TLS_FRAG_ENABLE(1));
319 
320 		toep->tls.iv_len = iv_size;
321 		toep->tls.frag_size = tls->params.max_frame_len;
322 		toep->tls.fcplenmax = get_tp_plen_max(tls);
323 		toep->tls.expn_per_ulp = tls->params.tls_hlen +
324 		    tls->params.tls_tlen;
325 		toep->tls.pdus_per_ulp = 1;
326 		toep->tls.adjusted_plen = toep->tls.expn_per_ulp +
327 		    tls->params.max_frame_len;
328 		toep->tls.tx_key_info_size = t4_tls_key_info_size(tls);
329 	} else {
330 		toep->flags |= TPF_TLS_STARTING | TPF_TLS_RX_QUIESCING;
331 		toep->tls.rx_version = tls->params.tls_vmajor << 8 |
332 		    tls->params.tls_vminor;
333 
334 		CTR2(KTR_CXGBE, "%s: tid %d setting RX_QUIESCE", __func__,
335 		    toep->tid);
336 		t4_set_rx_quiesce(toep);
337 	}
338 
339 	return (0);
340 }
341 
342 void
tls_init_toep(struct toepcb * toep)343 tls_init_toep(struct toepcb *toep)
344 {
345 	struct tls_ofld_info *tls_ofld = &toep->tls;
346 
347 	tls_ofld->rx_key_addr = -1;
348 	tls_ofld->tx_key_addr = -1;
349 }
350 
351 void
tls_uninit_toep(struct toepcb * toep)352 tls_uninit_toep(struct toepcb *toep)
353 {
354 
355 	clear_tls_keyid(toep);
356 }
357 
358 #define MAX_OFLD_TX_CREDITS (SGE_MAX_WR_LEN / 16)
359 #define	MIN_OFLD_TLSTX_CREDITS(toep)					\
360 	(howmany(sizeof(struct fw_tlstx_data_wr) +			\
361 	    sizeof(struct cpl_tx_tls_sfo) + sizeof(struct ulptx_idata) + \
362 	    sizeof(struct ulptx_sc_memrd) +				\
363 	    AES_BLOCK_LEN + 1, 16))
364 
365 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,int num_ivs)366 write_tlstx_wr(struct fw_tlstx_data_wr *txwr, struct toepcb *toep,
367     unsigned int plen, unsigned int expn, uint8_t credits, int shove,
368     int num_ivs)
369 {
370 	struct tls_ofld_info *tls_ofld = &toep->tls;
371 	unsigned int len = plen + expn;
372 
373 	txwr->op_to_immdlen = htobe32(V_WR_OP(FW_TLSTX_DATA_WR) |
374 	    V_FW_TLSTX_DATA_WR_COMPL(1) |
375 	    V_FW_TLSTX_DATA_WR_IMMDLEN(0));
376 	txwr->flowid_len16 = htobe32(V_FW_TLSTX_DATA_WR_FLOWID(toep->tid) |
377 	    V_FW_TLSTX_DATA_WR_LEN16(credits));
378 	txwr->plen = htobe32(len);
379 	txwr->lsodisable_to_flags = htobe32(V_TX_ULP_MODE(ULP_MODE_TLS) |
380 	    V_TX_URG(0) | /* F_T6_TX_FORCE | */ V_TX_SHOVE(shove));
381 	txwr->ctxloc_to_exp = htobe32(V_FW_TLSTX_DATA_WR_NUMIVS(num_ivs) |
382 	    V_FW_TLSTX_DATA_WR_EXP(expn) |
383 	    V_FW_TLSTX_DATA_WR_CTXLOC(TLS_SFO_WR_CONTEXTLOC_DDR) |
384 	    V_FW_TLSTX_DATA_WR_IVDSGL(0) |
385 	    V_FW_TLSTX_DATA_WR_KEYSIZE(tls_ofld->tx_key_info_size >> 4));
386 	txwr->mfs = htobe16(tls_ofld->frag_size);
387 	txwr->adjustedplen_pkd = htobe16(
388 	    V_FW_TLSTX_DATA_WR_ADJUSTEDPLEN(tls_ofld->adjusted_plen));
389 	txwr->expinplenmax_pkd = htobe16(
390 	    V_FW_TLSTX_DATA_WR_EXPINPLENMAX(tls_ofld->expn_per_ulp));
391 	txwr->pdusinplenmax_pkd =
392 	    V_FW_TLSTX_DATA_WR_PDUSINPLENMAX(tls_ofld->pdus_per_ulp);
393 }
394 
395 static void
write_tlstx_cpl(struct cpl_tx_tls_sfo * cpl,struct toepcb * toep,struct tls_hdr * tls_hdr,unsigned int plen,uint8_t rec_type,uint64_t seqno)396 write_tlstx_cpl(struct cpl_tx_tls_sfo *cpl, struct toepcb *toep,
397     struct tls_hdr *tls_hdr, unsigned int plen, uint8_t rec_type,
398     uint64_t seqno)
399 {
400 	struct tls_ofld_info *tls_ofld = &toep->tls;
401 	int data_type, seglen;
402 
403 	seglen = plen;
404 	data_type = tls_content_type(rec_type);
405 	cpl->op_to_seg_len = htobe32(V_CPL_TX_TLS_SFO_OPCODE(CPL_TX_TLS_SFO) |
406 	    V_CPL_TX_TLS_SFO_DATA_TYPE(data_type) |
407 	    V_CPL_TX_TLS_SFO_CPL_LEN(2) | V_CPL_TX_TLS_SFO_SEG_LEN(seglen));
408 	cpl->pld_len = htobe32(plen);
409 	if (data_type == CPL_TX_TLS_SFO_TYPE_CUSTOM)
410 		cpl->type_protover = htobe32(V_CPL_TX_TLS_SFO_TYPE(rec_type));
411 	cpl->seqno_numivs = htobe32(tls_ofld->scmd0.seqno_numivs |
412 	    V_SCMD_NUM_IVS(1));
413 	cpl->ivgen_hdrlen = htobe32(tls_ofld->scmd0.ivgen_hdrlen);
414 	cpl->scmd1 = htobe64(seqno);
415 }
416 
417 static int
count_ext_pgs_segs(struct mbuf * m)418 count_ext_pgs_segs(struct mbuf *m)
419 {
420 	vm_paddr_t nextpa;
421 	u_int i, nsegs;
422 
423 	MPASS(m->m_epg_npgs > 0);
424 	nsegs = 1;
425 	nextpa = m->m_epg_pa[0] + PAGE_SIZE;
426 	for (i = 1; i < m->m_epg_npgs; i++) {
427 		if (nextpa != m->m_epg_pa[i])
428 			nsegs++;
429 		nextpa = m->m_epg_pa[i] + PAGE_SIZE;
430 	}
431 	return (nsegs);
432 }
433 
434 static void
write_ktlstx_sgl(void * dst,struct mbuf * m,int nsegs)435 write_ktlstx_sgl(void *dst, struct mbuf *m, int nsegs)
436 {
437 	struct ulptx_sgl *usgl = dst;
438 	vm_paddr_t pa;
439 	uint32_t len;
440 	int i, j;
441 
442 	KASSERT(nsegs > 0, ("%s: nsegs 0", __func__));
443 
444 	usgl->cmd_nsge = htobe32(V_ULPTX_CMD(ULP_TX_SC_DSGL) |
445 	    V_ULPTX_NSGE(nsegs));
446 
447 	/* Figure out the first S/G length. */
448 	pa = m->m_epg_pa[0] + m->m_epg_1st_off;
449 	usgl->addr0 = htobe64(pa);
450 	len = m_epg_pagelen(m, 0, m->m_epg_1st_off);
451 	pa += len;
452 	for (i = 1; i < m->m_epg_npgs; i++) {
453 		if (m->m_epg_pa[i] != pa)
454 			break;
455 		len += m_epg_pagelen(m, i, 0);
456 		pa += m_epg_pagelen(m, i, 0);
457 	}
458 	usgl->len0 = htobe32(len);
459 #ifdef INVARIANTS
460 	nsegs--;
461 #endif
462 
463 	j = -1;
464 	for (; i < m->m_epg_npgs; i++) {
465 		if (j == -1 || m->m_epg_pa[i] != pa) {
466 			if (j >= 0)
467 				usgl->sge[j / 2].len[j & 1] = htobe32(len);
468 			j++;
469 #ifdef INVARIANTS
470 			nsegs--;
471 #endif
472 			pa = m->m_epg_pa[i];
473 			usgl->sge[j / 2].addr[j & 1] = htobe64(pa);
474 			len = m_epg_pagelen(m, i, 0);
475 			pa += len;
476 		} else {
477 			len += m_epg_pagelen(m, i, 0);
478 			pa += m_epg_pagelen(m, i, 0);
479 		}
480 	}
481 	if (j >= 0) {
482 		usgl->sge[j / 2].len[j & 1] = htobe32(len);
483 
484 		if ((j & 1) == 0)
485 			usgl->sge[j / 2].len[1] = htobe32(0);
486 	}
487 	KASSERT(nsegs == 0, ("%s: nsegs %d, m %p", __func__, nsegs, m));
488 }
489 
490 /*
491  * Similar to t4_push_frames() but handles sockets that contain TLS
492  * record mbufs.
493  */
494 void
t4_push_ktls(struct adapter * sc,struct toepcb * toep,int drop)495 t4_push_ktls(struct adapter *sc, struct toepcb *toep, int drop)
496 {
497 	struct tls_hdr *thdr;
498 	struct fw_tlstx_data_wr *txwr;
499 	struct cpl_tx_tls_sfo *cpl;
500 	struct ulptx_idata *idata;
501 	struct ulptx_sc_memrd *memrd;
502 	struct wrqe *wr;
503 	struct mbuf *m;
504 	u_int nsegs, credits, wr_len;
505 	u_int expn_size;
506 	struct inpcb *inp = toep->inp;
507 	struct tcpcb *tp = intotcpcb(inp);
508 	struct socket *so = inp->inp_socket;
509 	struct sockbuf *sb = &so->so_snd;
510 	struct mbufq *pduq = &toep->ulp_pduq;
511 	int tls_size, tx_credits, shove, sowwakeup;
512 	struct ofld_tx_sdesc *txsd;
513 	char *buf;
514 	bool tls13;
515 
516 	INP_WLOCK_ASSERT(inp);
517 	KASSERT(toep->flags & TPF_FLOWC_WR_SENT,
518 	    ("%s: flowc_wr not sent for tid %u.", __func__, toep->tid));
519 
520 	KASSERT(ulp_mode(toep) == ULP_MODE_NONE ||
521 	    ulp_mode(toep) == ULP_MODE_TCPDDP || ulp_mode(toep) == ULP_MODE_TLS,
522 	    ("%s: ulp_mode %u for toep %p", __func__, ulp_mode(toep), toep));
523 	KASSERT(tls_tx_key(toep),
524 	    ("%s: TX key not set for toep %p", __func__, toep));
525 
526 #ifdef VERBOSE_TRACES
527 	CTR4(KTR_CXGBE, "%s: tid %d toep flags %#x tp flags %#x drop %d",
528 	    __func__, toep->tid, toep->flags, tp->t_flags);
529 #endif
530 	if (__predict_false(toep->flags & TPF_ABORT_SHUTDOWN))
531 		return;
532 
533 #ifdef RATELIMIT
534 	if (__predict_false(inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) &&
535 	    (update_tx_rate_limit(sc, toep, so->so_max_pacing_rate) == 0)) {
536 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
537 	}
538 #endif
539 
540 	/*
541 	 * This function doesn't resume by itself.  Someone else must clear the
542 	 * flag and call this function.
543 	 */
544 	if (__predict_false(toep->flags & TPF_TX_SUSPENDED)) {
545 		KASSERT(drop == 0,
546 		    ("%s: drop (%d) != 0 but tx is suspended", __func__, drop));
547 		return;
548 	}
549 
550 	tls13 = toep->tls.tls13;
551 	txsd = &toep->txsd[toep->txsd_pidx];
552 	for (;;) {
553 		tx_credits = min(toep->tx_credits, MAX_OFLD_TX_CREDITS);
554 
555 		if (__predict_false((m = mbufq_first(pduq)) != NULL)) {
556 			if (!t4_push_raw_wr(sc, toep, m)) {
557 				toep->flags |= TPF_TX_SUSPENDED;
558 				return;
559 			}
560 
561 			(void)mbufq_dequeue(pduq);
562 
563 			txsd = &toep->txsd[toep->txsd_pidx];
564 			continue;
565 		}
566 
567 		SOCKBUF_LOCK(sb);
568 		sowwakeup = drop;
569 		if (drop) {
570 			sbdrop_locked(sb, drop);
571 			drop = 0;
572 		}
573 
574 		m = sb->sb_sndptr != NULL ? sb->sb_sndptr->m_next : sb->sb_mb;
575 
576 		/*
577 		 * Send a FIN if requested, but only if there's no
578 		 * more data to send.
579 		 */
580 		if (m == NULL && toep->flags & TPF_SEND_FIN) {
581 			if (sowwakeup)
582 				sowwakeup_locked(so);
583 			else
584 				SOCKBUF_UNLOCK(sb);
585 			SOCKBUF_UNLOCK_ASSERT(sb);
586 			t4_close_conn(sc, toep);
587 			return;
588 		}
589 
590 		/*
591 		 * If there is no ready data to send, wait until more
592 		 * data arrives.
593 		 */
594 		if (m == NULL || (m->m_flags & M_NOTREADY) != 0) {
595 			if (sowwakeup)
596 				sowwakeup_locked(so);
597 			else
598 				SOCKBUF_UNLOCK(sb);
599 			SOCKBUF_UNLOCK_ASSERT(sb);
600 #ifdef VERBOSE_TRACES
601 			CTR2(KTR_CXGBE, "%s: tid %d no ready data to send",
602 			    __func__, toep->tid);
603 #endif
604 			return;
605 		}
606 
607 		KASSERT(m->m_flags & M_EXTPG, ("%s: mbuf %p is not NOMAP",
608 		    __func__, m));
609 		KASSERT(m->m_epg_tls != NULL,
610 		    ("%s: mbuf %p doesn't have TLS session", __func__, m));
611 
612 		/* Calculate WR length. */
613 		wr_len = sizeof(struct fw_tlstx_data_wr) +
614 		    sizeof(struct cpl_tx_tls_sfo) +
615 		    sizeof(struct ulptx_idata) + sizeof(struct ulptx_sc_memrd);
616 
617 		if (!tls13) {
618 			/* Explicit IVs for AES-CBC and AES-GCM are <= 16. */
619 			MPASS(toep->tls.iv_len <= AES_BLOCK_LEN);
620 			wr_len += AES_BLOCK_LEN;
621 		}
622 
623 		/* Account for SGL in work request length. */
624 		nsegs = count_ext_pgs_segs(m);
625 		wr_len += sizeof(struct ulptx_sgl) +
626 		    ((3 * (nsegs - 1)) / 2 + ((nsegs - 1) & 1)) * 8;
627 
628 		/* Not enough credits for this work request. */
629 		if (howmany(wr_len, 16) > tx_credits) {
630 			if (sowwakeup)
631 				sowwakeup_locked(so);
632 			else
633 				SOCKBUF_UNLOCK(sb);
634 			SOCKBUF_UNLOCK_ASSERT(sb);
635 #ifdef VERBOSE_TRACES
636 			CTR5(KTR_CXGBE,
637 	    "%s: tid %d mbuf %p requires %d credits, but only %d available",
638 			    __func__, toep->tid, m, howmany(wr_len, 16),
639 			    tx_credits);
640 #endif
641 			toep->flags |= TPF_TX_SUSPENDED;
642 			return;
643 		}
644 
645 		/* Shove if there is no additional data pending. */
646 		shove = ((m->m_next == NULL ||
647 		    (m->m_next->m_flags & M_NOTREADY) != 0)) &&
648 		    (tp->t_flags & TF_MORETOCOME) == 0;
649 
650 		if (sb->sb_flags & SB_AUTOSIZE &&
651 		    V_tcp_do_autosndbuf &&
652 		    sb->sb_hiwat < V_tcp_autosndbuf_max &&
653 		    sbused(sb) >= sb->sb_hiwat * 7 / 8) {
654 			int newsize = min(sb->sb_hiwat + V_tcp_autosndbuf_inc,
655 			    V_tcp_autosndbuf_max);
656 
657 			if (!sbreserve_locked(so, SO_SND, newsize, NULL))
658 				sb->sb_flags &= ~SB_AUTOSIZE;
659 			else
660 				sowwakeup = 1;	/* room available */
661 		}
662 		if (sowwakeup)
663 			sowwakeup_locked(so);
664 		else
665 			SOCKBUF_UNLOCK(sb);
666 		SOCKBUF_UNLOCK_ASSERT(sb);
667 
668 		if (__predict_false(toep->flags & TPF_FIN_SENT))
669 			panic("%s: excess tx.", __func__);
670 
671 		wr = alloc_wrqe(roundup2(wr_len, 16), &toep->ofld_txq->wrq);
672 		if (wr == NULL) {
673 			/* XXX: how will we recover from this? */
674 			toep->flags |= TPF_TX_SUSPENDED;
675 			return;
676 		}
677 
678 		thdr = (struct tls_hdr *)&m->m_epg_hdr;
679 #ifdef VERBOSE_TRACES
680 		CTR5(KTR_CXGBE, "%s: tid %d TLS record %ju type %d len %#x",
681 		    __func__, toep->tid, m->m_epg_seqno, thdr->type,
682 		    m->m_len);
683 #endif
684 		txwr = wrtod(wr);
685 		cpl = (struct cpl_tx_tls_sfo *)(txwr + 1);
686 		memset(txwr, 0, roundup2(wr_len, 16));
687 		credits = howmany(wr_len, 16);
688 		expn_size = m->m_epg_hdrlen +
689 		    m->m_epg_trllen;
690 		tls_size = m->m_len - expn_size;
691 		write_tlstx_wr(txwr, toep, tls_size, expn_size, credits, shove,
692 		    tls13 ? 0 : 1);
693 		write_tlstx_cpl(cpl, toep, thdr, tls_size,
694 		    tls13 ? m->m_epg_record_type : thdr->type, m->m_epg_seqno);
695 
696 		idata = (struct ulptx_idata *)(cpl + 1);
697 		idata->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
698 		idata->len = htobe32(0);
699 		memrd = (struct ulptx_sc_memrd *)(idata + 1);
700 		memrd->cmd_to_len = htobe32(V_ULPTX_CMD(ULP_TX_SC_MEMRD) |
701 		    V_ULP_TX_SC_MORE(1) |
702 		    V_ULPTX_LEN16(toep->tls.tx_key_info_size >> 4));
703 		memrd->addr = htobe32(toep->tls.tx_key_addr >> 5);
704 
705 		buf = (char *)(memrd + 1);
706 		if (!tls13) {
707 			/* Copy IV. */
708 			memcpy(buf, thdr + 1, toep->tls.iv_len);
709 			buf += AES_BLOCK_LEN;
710 		}
711 
712 		write_ktlstx_sgl(buf, m, nsegs);
713 
714 		KASSERT(toep->tx_credits >= credits,
715 			("%s: not enough credits", __func__));
716 
717 		toep->tx_credits -= credits;
718 
719 		tp->snd_nxt += m->m_len;
720 		tp->snd_max += m->m_len;
721 
722 		SOCKBUF_LOCK(sb);
723 		sb->sb_sndptr = m;
724 		SOCKBUF_UNLOCK(sb);
725 
726 		toep->flags |= TPF_TX_DATA_SENT;
727 		if (toep->tx_credits < MIN_OFLD_TLSTX_CREDITS(toep))
728 			toep->flags |= TPF_TX_SUSPENDED;
729 
730 		KASSERT(toep->txsd_avail > 0, ("%s: no txsd", __func__));
731 		KASSERT(m->m_len <= MAX_OFLD_TX_SDESC_PLEN,
732 		    ("%s: plen %u too large", __func__, m->m_len));
733 		txsd->plen = m->m_len;
734 		txsd->tx_credits = credits;
735 		txsd++;
736 		if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) {
737 			toep->txsd_pidx = 0;
738 			txsd = &toep->txsd[0];
739 		}
740 		toep->txsd_avail--;
741 
742 		counter_u64_add(toep->ofld_txq->tx_toe_tls_records, 1);
743 		counter_u64_add(toep->ofld_txq->tx_toe_tls_octets, m->m_len);
744 
745 		t4_l2t_send(sc, wr, toep->l2te);
746 	}
747 }
748 
749 /*
750  * For TLS data we place received mbufs received via CPL_TLS_DATA into
751  * an mbufq in the TLS offload state.  When CPL_RX_TLS_CMP is
752  * received, the completed PDUs are placed into the socket receive
753  * buffer.
754  *
755  * The TLS code reuses the ulp_pdu_reclaimq to hold the pending mbufs.
756  */
757 static int
do_tls_data(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)758 do_tls_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
759 {
760 	struct adapter *sc = iq->adapter;
761 	const struct cpl_tls_data *cpl = mtod(m, const void *);
762 	unsigned int tid = GET_TID(cpl);
763 	struct toepcb *toep = lookup_tid(sc, tid);
764 	struct inpcb *inp = toep->inp;
765 	struct tcpcb *tp = intotcpcb(inp);
766 	int len;
767 
768 	/* XXX: Should this match do_rx_data instead? */
769 	KASSERT(!(toep->flags & TPF_SYNQE),
770 	    ("%s: toep %p claims to be a synq entry", __func__, toep));
771 
772 	KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
773 
774 	/* strip off CPL header */
775 	m_adj(m, sizeof(*cpl));
776 	len = m->m_pkthdr.len;
777 
778 	toep->ofld_rxq->rx_toe_tls_octets += len;
779 
780 	KASSERT(len == G_CPL_TLS_DATA_LENGTH(be32toh(cpl->length_pkd)),
781 	    ("%s: payload length mismatch", __func__));
782 
783 	INP_WLOCK(inp);
784 	if (tp->t_flags & TF_DISCONNECTED) {
785 		CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), t_flags 0x%x",
786 		    __func__, tid, len, tp->t_flags);
787 		INP_WUNLOCK(inp);
788 		m_freem(m);
789 		return (0);
790 	}
791 
792 	/* Save TCP sequence number. */
793 	m->m_pkthdr.tls_tcp_seq = be32toh(cpl->seq);
794 
795 	if (mbufq_enqueue(&toep->ulp_pdu_reclaimq, m)) {
796 #ifdef INVARIANTS
797 		panic("Failed to queue TLS data packet");
798 #else
799 		printf("%s: Failed to queue TLS data packet\n", __func__);
800 		INP_WUNLOCK(inp);
801 		m_freem(m);
802 		return (0);
803 #endif
804 	}
805 
806 	tp->t_rcvtime = ticks;
807 
808 #ifdef VERBOSE_TRACES
809 	CTR4(KTR_CXGBE, "%s: tid %u len %d seq %u", __func__, tid, len,
810 	    be32toh(cpl->seq));
811 #endif
812 
813 	INP_WUNLOCK(inp);
814 	return (0);
815 }
816 
817 static int
do_rx_tls_cmp(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)818 do_rx_tls_cmp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
819 {
820 	struct adapter *sc = iq->adapter;
821 	const struct cpl_rx_tls_cmp *cpl = mtod(m, const void *);
822 	struct tlsrx_hdr_pkt *tls_hdr_pkt;
823 	unsigned int tid = GET_TID(cpl);
824 	struct toepcb *toep = lookup_tid(sc, tid);
825 	struct inpcb *inp = toep->inp;
826 	struct tcpcb *tp = intotcpcb(inp);
827 	struct socket *so;
828 	struct sockbuf *sb;
829 	struct mbuf *tls_data;
830 	struct tls_get_record *tgr;
831 	struct mbuf *control, *n;
832 	int pdu_length, resid, trailer_len;
833 #if defined(KTR) || defined(INVARIANTS)
834 	int len;
835 #endif
836 
837 	KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
838 	KASSERT(!(toep->flags & TPF_SYNQE),
839 	    ("%s: toep %p claims to be a synq entry", __func__, toep));
840 
841 	/* strip off CPL header */
842 	m_adj(m, sizeof(*cpl));
843 #if defined(KTR) || defined(INVARIANTS)
844 	len = m->m_pkthdr.len;
845 #endif
846 
847 	toep->ofld_rxq->rx_toe_tls_records++;
848 
849 	KASSERT(len == G_CPL_RX_TLS_CMP_LENGTH(be32toh(cpl->pdulength_length)),
850 	    ("%s: payload length mismatch", __func__));
851 
852 	INP_WLOCK(inp);
853 	if (tp->t_flags & TF_DISCONNECTED) {
854 		CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), t_flags 0x%x",
855 		    __func__, tid, len, tp->t_flags);
856 		INP_WUNLOCK(inp);
857 		m_freem(m);
858 		return (0);
859 	}
860 
861 	pdu_length = G_CPL_RX_TLS_CMP_PDULENGTH(be32toh(cpl->pdulength_length));
862 
863 	so = inp_inpcbtosocket(inp);
864 
865 #ifdef VERBOSE_TRACES
866 	CTR6(KTR_CXGBE, "%s: tid %u PDU len %d len %d seq %u, rcv_nxt %u",
867 	    __func__, tid, pdu_length, len, be32toh(cpl->seq), tp->rcv_nxt);
868 #endif
869 
870 	tp->rcv_nxt += pdu_length;
871 	KASSERT(tp->rcv_wnd >= pdu_length,
872 	    ("%s: negative window size", __func__));
873 	tp->rcv_wnd -= pdu_length;
874 
875 	/* XXX: Not sure what to do about urgent data. */
876 
877 	/*
878 	 * The payload of this CPL is the TLS header followed by
879 	 * additional fields.  For TLS 1.3 the type field holds the
880 	 * inner record type and the length field has been updated to
881 	 * strip the inner record type, padding, and MAC.
882 	 */
883 	KASSERT(m->m_len >= sizeof(*tls_hdr_pkt),
884 	    ("%s: payload too small", __func__));
885 	tls_hdr_pkt = mtod(m, void *);
886 
887 	tls_data = mbufq_dequeue(&toep->ulp_pdu_reclaimq);
888 	if (tls_data != NULL) {
889 		KASSERT(be32toh(cpl->seq) == tls_data->m_pkthdr.tls_tcp_seq,
890 		    ("%s: sequence mismatch", __func__));
891 	}
892 
893 	/*
894 	 * Report decryption errors as EBADMSG.
895 	 *
896 	 * XXX: To support rekeying for TLS 1.3 this will eventually
897 	 * have to be updated to recrypt the data with the old key and
898 	 * then decrypt with the new key.  Punt for now as KTLS
899 	 * doesn't yet support rekeying.
900 	 */
901 	if ((tls_hdr_pkt->res_to_mac_error & M_TLSRX_HDR_PKT_ERROR) != 0) {
902 		CTR4(KTR_CXGBE, "%s: tid %u TLS error %#x ddp_vld %#x",
903 		    __func__, toep->tid, tls_hdr_pkt->res_to_mac_error,
904 		    be32toh(cpl->ddp_valid));
905 		m_freem(m);
906 		m_freem(tls_data);
907 
908 		CURVNET_SET(toep->vnet);
909 		so->so_error = EBADMSG;
910 		sorwakeup(so);
911 
912 		INP_WUNLOCK(inp);
913 		CURVNET_RESTORE();
914 
915 		return (0);
916 	}
917 
918 	/* For TLS 1.3 trim the header and trailer. */
919 	if (toep->tls.tls13) {
920 		KASSERT(tls_data != NULL, ("%s: TLS 1.3 record without data",
921 		    __func__));
922 		MPASS(tls_data->m_pkthdr.len == pdu_length);
923 		m_adj(tls_data, sizeof(struct tls_record_layer));
924 		if (tls_data->m_pkthdr.len > be16toh(tls_hdr_pkt->length))
925 			tls_data->m_pkthdr.len = be16toh(tls_hdr_pkt->length);
926 		resid = tls_data->m_pkthdr.len;
927 		if (resid == 0) {
928 			m_freem(tls_data);
929 			tls_data = NULL;
930 		} else {
931 			for (n = tls_data;; n = n->m_next) {
932 				if (n->m_len < resid) {
933 					resid -= n->m_len;
934 					continue;
935 				}
936 
937 				n->m_len = resid;
938 				m_freem(n->m_next);
939 				n->m_next = NULL;
940 				break;
941 			}
942 		}
943 	}
944 
945 	/* Handle data received after the socket is closed. */
946 	sb = &so->so_rcv;
947 	SOCKBUF_LOCK(sb);
948 	if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) {
949 		struct epoch_tracker et;
950 
951 		CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)",
952 		    __func__, tid, pdu_length);
953 		m_freem(m);
954 		m_freem(tls_data);
955 		SOCKBUF_UNLOCK(sb);
956 		INP_WUNLOCK(inp);
957 
958 		CURVNET_SET(toep->vnet);
959 		NET_EPOCH_ENTER(et);
960 		INP_WLOCK(inp);
961 		tp = tcp_drop(tp, ECONNRESET);
962 		if (tp != NULL)
963 			INP_WUNLOCK(inp);
964 		NET_EPOCH_EXIT(et);
965 		CURVNET_RESTORE();
966 
967 		return (0);
968 	}
969 
970 	/*
971 	 * If there is any data in the 'sb_mtls' chain of the socket
972 	 * or we aren't able to allocate the control mbuf, append the
973 	 * record as a CSUM_TLS_DECRYPTED packet to 'sb_mtls' rather
974 	 * than as a decrypted record to 'sb_m'.
975 	 */
976 	if (sb->sb_mtls != NULL)
977 		control = NULL;
978 	else
979 		control = sbcreatecontrol(NULL, sizeof(*tgr), TLS_GET_RECORD,
980 		    IPPROTO_TCP, M_NOWAIT);
981 
982 	if (control != NULL) {
983 		tgr = (struct tls_get_record *)
984 		    CMSG_DATA(mtod(control, struct cmsghdr *));
985 		memset(tgr, 0, sizeof(*tgr));
986 		tgr->tls_type = tls_hdr_pkt->type;
987 		tgr->tls_vmajor = be16toh(tls_hdr_pkt->version) >> 8;
988 		tgr->tls_vminor = be16toh(tls_hdr_pkt->version) & 0xff;
989 		if (tls_data != NULL) {
990 			m_last(tls_data)->m_flags |= M_EOR;
991 			tgr->tls_length = htobe16(tls_data->m_pkthdr.len);
992 		} else
993 			tgr->tls_length = 0;
994 
995 		m_freem(m);
996 		m = tls_data;
997 	} else {
998 		M_ASSERTPKTHDR(m);
999 
1000 		/* It's ok that any explicit IV is missing. */
1001 		m->m_len = sb->sb_tls_info->params.tls_hlen;
1002 		m->m_pkthdr.csum_flags |= CSUM_TLS_DECRYPTED;
1003 		m->m_pkthdr.len = m->m_len;
1004 		if (tls_data != NULL) {
1005 			m->m_pkthdr.len += tls_data->m_pkthdr.len;
1006 			m_demote_pkthdr(tls_data);
1007 			m->m_next = tls_data;
1008 		}
1009 
1010 		/*
1011 		 * Grow the chain by the trailer, but without
1012 		 * contents.  The trailer will be thrown away by
1013 		 * ktls_decrypt.  Note that ktls_decrypt assumes the
1014 		 * trailer is tls_tlen bytes long, so append that many
1015 		 * bytes not the actual trailer size computed from
1016 		 * pdu_length.
1017 		 */
1018 		trailer_len = sb->sb_tls_info->params.tls_tlen;
1019 		if (tls_data != NULL) {
1020 			m_last(tls_data)->m_len += trailer_len;
1021 			tls_data = NULL;
1022 		} else
1023 			m->m_len += trailer_len;
1024 		m->m_pkthdr.len += trailer_len;
1025 		tls_hdr_pkt->length = htobe16(m->m_pkthdr.len -
1026 		    sizeof(struct tls_record_layer));
1027 	}
1028 
1029 	/* receive buffer autosize */
1030 	MPASS(toep->vnet == so->so_vnet);
1031 	CURVNET_SET(toep->vnet);
1032 	if (sb->sb_flags & SB_AUTOSIZE &&
1033 	    V_tcp_do_autorcvbuf &&
1034 	    sb->sb_hiwat < V_tcp_autorcvbuf_max &&
1035 	    m->m_pkthdr.len > (sbspace(sb) / 8 * 7)) {
1036 		unsigned int hiwat = sb->sb_hiwat;
1037 		unsigned int newsize = min(hiwat + sc->tt.autorcvbuf_inc,
1038 		    V_tcp_autorcvbuf_max);
1039 
1040 		if (!sbreserve_locked(so, SO_RCV, newsize, NULL))
1041 			sb->sb_flags &= ~SB_AUTOSIZE;
1042 	}
1043 
1044 	if (control != NULL)
1045 		sbappendcontrol_locked(sb, m, control, 0);
1046 	else
1047 		sbappendstream_locked(sb, m, 0);
1048 	t4_rcvd_locked(&toep->td->tod, tp);
1049 
1050 	sorwakeup_locked(so);
1051 	SOCKBUF_UNLOCK_ASSERT(sb);
1052 
1053 	INP_WUNLOCK(inp);
1054 	CURVNET_RESTORE();
1055 	return (0);
1056 }
1057 
1058 void
do_rx_data_tls(const struct cpl_rx_data * cpl,struct toepcb * toep,struct mbuf * m)1059 do_rx_data_tls(const struct cpl_rx_data *cpl, struct toepcb *toep,
1060     struct mbuf *m)
1061 {
1062 	struct inpcb *inp = toep->inp;
1063 	struct tls_ofld_info *tls_ofld = &toep->tls;
1064 	struct tls_hdr *hdr;
1065 	struct tcpcb *tp;
1066 	struct socket *so;
1067 	struct sockbuf *sb;
1068 	int len;
1069 
1070 	len = m->m_pkthdr.len;
1071 
1072 	INP_WLOCK_ASSERT(inp);
1073 
1074 	so = inp_inpcbtosocket(inp);
1075 	tp = intotcpcb(inp);
1076 	sb = &so->so_rcv;
1077 	SOCKBUF_LOCK(sb);
1078 	CURVNET_SET(toep->vnet);
1079 
1080 	tp->rcv_nxt += len;
1081 	KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__));
1082 	tp->rcv_wnd -= len;
1083 
1084 	/* Do we have a full TLS header? */
1085 	if (len < sizeof(*hdr)) {
1086 		CTR3(KTR_CXGBE, "%s: tid %u len %d: too short for a TLS header",
1087 		    __func__, toep->tid, len);
1088 		so->so_error = EMSGSIZE;
1089 		goto out;
1090 	}
1091 	hdr = mtod(m, struct tls_hdr *);
1092 
1093 	/* Is the header valid? */
1094 	if (be16toh(hdr->version) != tls_ofld->rx_version) {
1095 		CTR3(KTR_CXGBE, "%s: tid %u invalid version %04x",
1096 		    __func__, toep->tid, be16toh(hdr->version));
1097 		so->so_error = EINVAL;
1098 		goto out;
1099 	}
1100 	if (be16toh(hdr->length) < sizeof(*hdr)) {
1101 		CTR3(KTR_CXGBE, "%s: tid %u invalid length %u",
1102 		    __func__, toep->tid, be16toh(hdr->length));
1103 		so->so_error = EBADMSG;
1104 		goto out;
1105 	}
1106 
1107 	/* Did we get a truncated record? */
1108 	if (len < be16toh(hdr->length)) {
1109 		CTR4(KTR_CXGBE, "%s: tid %u truncated TLS record (%d vs %u)",
1110 		    __func__, toep->tid, len, be16toh(hdr->length));
1111 
1112 		so->so_error = EMSGSIZE;
1113 		goto out;
1114 	}
1115 
1116 	/* Is the header type unknown? */
1117 	switch (hdr->type) {
1118 	case CONTENT_TYPE_CCS:
1119 	case CONTENT_TYPE_ALERT:
1120 	case CONTENT_TYPE_APP_DATA:
1121 	case CONTENT_TYPE_HANDSHAKE:
1122 		break;
1123 	default:
1124 		CTR3(KTR_CXGBE, "%s: tid %u invalid TLS record type %u",
1125 		    __func__, toep->tid, hdr->type);
1126 		so->so_error = EBADMSG;
1127 		goto out;
1128 	}
1129 
1130 	/*
1131 	 * Just punt.  Although this could fall back to software
1132 	 * decryption, this case should never really happen.
1133 	 */
1134 	CTR4(KTR_CXGBE, "%s: tid %u dropping TLS record type %u, length %u",
1135 	    __func__, toep->tid, hdr->type, be16toh(hdr->length));
1136 	so->so_error = EBADMSG;
1137 
1138 out:
1139 	sorwakeup_locked(so);
1140 	SOCKBUF_UNLOCK_ASSERT(sb);
1141 
1142 	INP_WUNLOCK(inp);
1143 	CURVNET_RESTORE();
1144 
1145 	m_freem(m);
1146 }
1147 
1148 /*
1149  * Send a work request setting one or more TCB fields to partially or
1150  * fully enable ULP_MODE_TLS.
1151  *
1152  * - If resid == 0, the socket buffer ends at a record boundary
1153  *   (either empty or contains one or more complete records).  Switch
1154  *   to ULP_MODE_TLS (if not already) and enable TLS decryption.
1155  *
1156  * - If resid != 0, the socket buffer contains a partial record.  In
1157  *   this case, switch to ULP_MODE_TLS partially and configure the TCB
1158  *   to pass along the remaining resid bytes undecrypted.  Once they
1159  *   arrive, this is called again with resid == 0 and enables TLS
1160  *   decryption.
1161  */
1162 static void
tls_update_tcb(struct adapter * sc,struct toepcb * toep,uint64_t seqno,size_t resid)1163 tls_update_tcb(struct adapter *sc, struct toepcb *toep, uint64_t seqno,
1164     size_t resid)
1165 {
1166 	struct mbuf *m;
1167 	struct work_request_hdr *wrh;
1168 	struct ulp_txpkt *ulpmc;
1169 	int fields, key_offset, len;
1170 
1171 	/*
1172 	 * If we are already in ULP_MODE_TLS, then we should now be at
1173 	 * a record boundary and ready to finish enabling TLS RX.
1174 	 */
1175 	KASSERT(resid == 0 || ulp_mode(toep) == ULP_MODE_NONE,
1176 	    ("%s: tid %d needs %zu more data but already ULP_MODE_TLS",
1177 	    __func__, toep->tid, resid));
1178 
1179 	fields = 0;
1180 	if (ulp_mode(toep) == ULP_MODE_NONE) {
1181 		/* 2 writes for the overlay region */
1182 		fields += 2;
1183 	}
1184 
1185 	if (resid == 0) {
1186 		/* W_TCB_TLS_SEQ */
1187 		fields++;
1188 
1189 		/* W_TCB_ULP_RAW */
1190 		fields++;
1191 	} else {
1192 		/* W_TCB_PDU_LEN */
1193 		fields++;
1194 
1195 		/* W_TCB_ULP_RAW */
1196 		fields++;
1197 	}
1198 
1199 	if (ulp_mode(toep) == ULP_MODE_NONE) {
1200 		/* W_TCB_ULP_TYPE */
1201 		fields ++;
1202 	}
1203 
1204 	/* W_TCB_T_FLAGS */
1205 	fields++;
1206 
1207 	len = sizeof(*wrh) + fields * roundup2(LEN__SET_TCB_FIELD_ULP, 16);
1208 	KASSERT(len <= SGE_MAX_WR_LEN,
1209 	    ("%s: WR with %d TCB field updates too large", __func__, fields));
1210 
1211 	m = alloc_raw_wr_mbuf(len);
1212 	if (m == NULL) {
1213 		/* XXX */
1214 		panic("%s: out of memory", __func__);
1215 	}
1216 
1217 	wrh = mtod(m, struct work_request_hdr *);
1218 	INIT_ULPTX_WRH(wrh, len, 1, toep->tid);	/* atomic */
1219 	ulpmc = (struct ulp_txpkt *)(wrh + 1);
1220 
1221 	if (ulp_mode(toep) == ULP_MODE_NONE) {
1222 		/*
1223 		 * Clear the TLS overlay region: 1023:832.
1224 		 *
1225 		 * Words 26/27 are always set to zero.  Words 28/29
1226 		 * contain seqno and are set when enabling TLS
1227 		 * decryption.  Word 30 is zero and Word 31 contains
1228 		 * the keyid.
1229 		 */
1230 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, 26,
1231 		    0xffffffffffffffff, 0);
1232 
1233 		/*
1234 		 * RX key tags are an index into the key portion of MA
1235 		 * memory stored as an offset from the base address in
1236 		 * units of 64 bytes.
1237 		 */
1238 		key_offset = toep->tls.rx_key_addr - sc->vres.key.start;
1239 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, 30,
1240 		    0xffffffffffffffff,
1241 		    (uint64_t)V_TCB_RX_TLS_KEY_TAG(key_offset / 64) << 32);
1242 	}
1243 
1244 	if (resid == 0) {
1245 		/*
1246 		 * The socket buffer is empty or only contains
1247 		 * complete TLS records: Set the sequence number and
1248 		 * enable TLS decryption.
1249 		 */
1250 		CTR3(KTR_CXGBE, "%s: tid %d enable TLS seqno %lu", __func__,
1251 		    toep->tid, seqno);
1252 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid,
1253 		    W_TCB_RX_TLS_SEQ, V_TCB_RX_TLS_SEQ(M_TCB_RX_TLS_SEQ),
1254 		    V_TCB_RX_TLS_SEQ(seqno));
1255 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid,
1256 		    W_TCB_ULP_RAW, V_TCB_ULP_RAW(M_TCB_ULP_RAW),
1257 		    V_TCB_ULP_RAW((V_TF_TLS_KEY_SIZE(3) | V_TF_TLS_CONTROL(1) |
1258 		    V_TF_TLS_ACTIVE(1) | V_TF_TLS_ENABLE(1))));
1259 
1260 		toep->flags &= ~TPF_TLS_STARTING;
1261 		toep->flags |= TPF_TLS_RECEIVE;
1262 	} else {
1263 		/*
1264 		 * The socket buffer ends with a partial record with a
1265 		 * full header and needs at least 6 bytes.
1266 		 *
1267 		 * Set PDU length.  This is treating the 'resid' bytes
1268 		 * as a TLS PDU, so the first 5 bytes are a fake
1269 		 * header and the rest are the PDU length.
1270 		 */
1271 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid,
1272 		    W_TCB_PDU_LEN, V_TCB_PDU_LEN(M_TCB_PDU_LEN),
1273 		    V_TCB_PDU_LEN(resid - sizeof(struct tls_hdr)));
1274 		CTR3(KTR_CXGBE, "%s: tid %d setting PDU_LEN to %zu",
1275 		    __func__, toep->tid, resid - sizeof(struct tls_hdr));
1276 
1277 		/* Clear all bits in ULP_RAW except for ENABLE. */
1278 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid,
1279 		    W_TCB_ULP_RAW, V_TCB_ULP_RAW(M_TCB_ULP_RAW),
1280 		    V_TCB_ULP_RAW(V_TF_TLS_ENABLE(1)));
1281 
1282 		/* Wait for 'resid' bytes to be delivered as CPL_RX_DATA. */
1283 		toep->tls.rx_resid = resid;
1284 	}
1285 
1286 	if (ulp_mode(toep) == ULP_MODE_NONE) {
1287 		/* Set the ULP mode to ULP_MODE_TLS. */
1288 		toep->params.ulp_mode = ULP_MODE_TLS;
1289 		ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid,
1290 		    W_TCB_ULP_TYPE, V_TCB_ULP_TYPE(M_TCB_ULP_TYPE),
1291 		    V_TCB_ULP_TYPE(ULP_MODE_TLS));
1292 	}
1293 
1294 	/* Clear TF_RX_QUIESCE. */
1295 	ulpmc = mk_set_tcb_field_ulp(sc, ulpmc, toep->tid, W_TCB_T_FLAGS,
1296 	    V_TF_RX_QUIESCE(1), 0);
1297 
1298 	t4_raw_wr_tx(sc, toep, m);
1299 }
1300 
1301 /*
1302  * Examine the pending data in the socket buffer and either enable TLS
1303  * RX or request more encrypted data.
1304  */
1305 static void
tls_check_rx_sockbuf(struct adapter * sc,struct toepcb * toep,struct sockbuf * sb)1306 tls_check_rx_sockbuf(struct adapter *sc, struct toepcb *toep,
1307     struct sockbuf *sb)
1308 {
1309 	uint64_t seqno;
1310 	size_t resid;
1311 	bool have_header;
1312 
1313 	SOCKBUF_LOCK_ASSERT(sb);
1314 	MPASS(toep->tls.rx_resid == 0);
1315 
1316 	have_header = ktls_pending_rx_info(sb, &seqno, &resid);
1317 	CTR5(KTR_CXGBE, "%s: tid %d have_header %d seqno %lu resid %zu",
1318 	    __func__, toep->tid, have_header, seqno, resid);
1319 
1320 	/*
1321 	 * If we have a partial header or we need fewer bytes than the
1322 	 * size of a TLS record, re-enable receive and pause again once
1323 	 * we get more data to try again.
1324 	 */
1325 	if (!have_header || (resid != 0 && (resid < sizeof(struct tls_hdr) ||
1326 	    is_t6(sc)))) {
1327 		CTR(KTR_CXGBE, "%s: tid %d waiting for more data", __func__,
1328 		    toep->tid);
1329 		toep->flags &= ~TPF_TLS_RX_QUIESCED;
1330 		t4_clear_rx_quiesce(toep);
1331 		return;
1332 	}
1333 
1334 	tls_update_tcb(sc, toep, seqno, resid);
1335 }
1336 
1337 void
tls_received_starting_data(struct adapter * sc,struct toepcb * toep,struct sockbuf * sb,int len)1338 tls_received_starting_data(struct adapter *sc, struct toepcb *toep,
1339     struct sockbuf *sb, int len)
1340 {
1341 	MPASS(toep->flags & TPF_TLS_STARTING);
1342 
1343 	/* Data was received before quiescing took effect. */
1344 	if ((toep->flags & TPF_TLS_RX_QUIESCING) != 0)
1345 		return;
1346 
1347 	/*
1348 	 * A previous call to tls_check_rx_sockbuf needed more data.
1349 	 * Now that more data has arrived, quiesce receive again and
1350 	 * check the state once the quiesce has completed.
1351 	 */
1352 	if ((toep->flags & TPF_TLS_RX_QUIESCED) == 0) {
1353 		CTR(KTR_CXGBE, "%s: tid %d quiescing", __func__, toep->tid);
1354 		toep->flags |= TPF_TLS_RX_QUIESCING;
1355 		t4_set_rx_quiesce(toep);
1356 		return;
1357 	}
1358 
1359 	KASSERT(len <= toep->tls.rx_resid,
1360 	    ("%s: received excess bytes %d (waiting for %zu)", __func__, len,
1361 	    toep->tls.rx_resid));
1362 	toep->tls.rx_resid -= len;
1363 	if (toep->tls.rx_resid != 0)
1364 		return;
1365 
1366 	tls_check_rx_sockbuf(sc, toep, sb);
1367 }
1368 
1369 static int
do_tls_tcb_rpl(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)1370 do_tls_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
1371 {
1372 	struct adapter *sc = iq->adapter;
1373 	const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1);
1374 	unsigned int tid = GET_TID(cpl);
1375 	struct toepcb *toep;
1376 	struct inpcb *inp;
1377 	struct socket *so;
1378 	struct sockbuf *sb;
1379 
1380 	if (cpl->status != CPL_ERR_NONE)
1381 		panic("XXX: tcp_rpl failed: %d", cpl->status);
1382 
1383 	toep = lookup_tid(sc, tid);
1384 	inp = toep->inp;
1385 	switch (cpl->cookie) {
1386 	case V_WORD(W_TCB_T_FLAGS) | V_COOKIE(CPL_COOKIE_TOM):
1387 		INP_WLOCK(inp);
1388 		if ((toep->flags & TPF_TLS_STARTING) == 0)
1389 			panic("%s: connection is not starting TLS RX\n",
1390 			    __func__);
1391 		MPASS((toep->flags & TPF_TLS_RX_QUIESCING) != 0);
1392 
1393 		toep->flags &= ~TPF_TLS_RX_QUIESCING;
1394 		toep->flags |= TPF_TLS_RX_QUIESCED;
1395 
1396 		so = inp->inp_socket;
1397 		sb = &so->so_rcv;
1398 		SOCKBUF_LOCK(sb);
1399 		tls_check_rx_sockbuf(sc, toep, sb);
1400 		SOCKBUF_UNLOCK(sb);
1401 		INP_WUNLOCK(inp);
1402 		break;
1403 	default:
1404 		panic("XXX: unknown tcb_rpl offset %#x, cookie %#x",
1405 		    G_WORD(cpl->cookie), G_COOKIE(cpl->cookie));
1406 	}
1407 
1408 	return (0);
1409 }
1410 
1411 void
t4_tls_mod_load(void)1412 t4_tls_mod_load(void)
1413 {
1414 
1415 	t4_register_cpl_handler(CPL_TLS_DATA, do_tls_data);
1416 	t4_register_cpl_handler(CPL_RX_TLS_CMP, do_rx_tls_cmp);
1417 	t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, do_tls_tcb_rpl,
1418 	    CPL_COOKIE_TOM);
1419 }
1420 
1421 void
t4_tls_mod_unload(void)1422 t4_tls_mod_unload(void)
1423 {
1424 
1425 	t4_register_cpl_handler(CPL_TLS_DATA, NULL);
1426 	t4_register_cpl_handler(CPL_RX_TLS_CMP, NULL);
1427 	t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, NULL, CPL_COOKIE_TOM);
1428 }
1429 #endif	/* TCP_OFFLOAD */
1430 #endif	/* KERN_TLS */
1431