xref: /freebsd/sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c (revision 02e9120893770924227138ba49df1edb3896112a)
1 /*-
2  * Copyright (c) 2019-2021 Mellanox Technologies. All rights reserved.
3  * Copyright (c) 2022 NVIDIA corporation & affiliates.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include "opt_kern_tls.h"
28 #include "opt_rss.h"
29 #include "opt_ratelimit.h"
30 
31 #include <dev/mlx5/mlx5_en/en.h>
32 
33 #include <dev/mlx5/tls.h>
34 
35 #include <linux/delay.h>
36 #include <sys/ktls.h>
37 #include <opencrypto/cryptodev.h>
38 
39 #ifdef KERN_TLS
40 
41 #ifdef RATELIMIT
42 static if_snd_tag_modify_t mlx5e_tls_rl_snd_tag_modify;
43 #endif
44 static if_snd_tag_query_t mlx5e_tls_snd_tag_query;
45 static if_snd_tag_free_t mlx5e_tls_snd_tag_free;
46 
47 static const struct if_snd_tag_sw mlx5e_tls_snd_tag_sw = {
48 	.snd_tag_query = mlx5e_tls_snd_tag_query,
49 	.snd_tag_free = mlx5e_tls_snd_tag_free,
50 	.type = IF_SND_TAG_TYPE_TLS
51 };
52 
53 #ifdef RATELIMIT
54 static const struct if_snd_tag_sw mlx5e_tls_rl_snd_tag_sw = {
55 	.snd_tag_modify = mlx5e_tls_rl_snd_tag_modify,
56 	.snd_tag_query = mlx5e_tls_snd_tag_query,
57 	.snd_tag_free = mlx5e_tls_snd_tag_free,
58 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
59 };
60 #endif
61 
62 MALLOC_DEFINE(M_MLX5E_TLS, "MLX5E_TLS", "MLX5 ethernet HW TLS");
63 
64 /* software TLS context */
65 struct mlx5_ifc_sw_tls_cntx_bits {
66 	struct mlx5_ifc_tls_static_params_bits param;
67 	struct mlx5_ifc_tls_progress_params_bits progress;
68 	struct {
69 		uint8_t key_data[8][0x20];
70 		uint8_t key_len[0x20];
71 	} key;
72 };
73 
74 CTASSERT(MLX5_ST_SZ_BYTES(sw_tls_cntx) <= sizeof(((struct mlx5e_tls_tag *)0)->crypto_params));
75 CTASSERT(MLX5_ST_SZ_BYTES(mkc) == sizeof(((struct mlx5e_tx_umr_wqe *)0)->mkc));
76 
77 static const char *mlx5e_tls_stats_desc[] = {
78 	MLX5E_TLS_STATS(MLX5E_STATS_DESC)
79 };
80 
81 static void mlx5e_tls_work(struct work_struct *);
82 
83 static int
84 mlx5e_tls_tag_import(void *arg, void **store, int cnt, int domain, int flags)
85 {
86 	struct mlx5e_tls_tag *ptag;
87 	int i;
88 
89 	for (i = 0; i != cnt; i++) {
90 		ptag = malloc_domainset(sizeof(*ptag), M_MLX5E_TLS,
91 		    mlx5_dev_domainset(arg), flags | M_ZERO);
92 		mtx_init(&ptag->mtx, "mlx5-tls-tag-mtx", NULL, MTX_DEF);
93 		INIT_WORK(&ptag->work, mlx5e_tls_work);
94 		store[i] = ptag;
95 	}
96 	return (i);
97 }
98 
99 static void
100 mlx5e_tls_tag_release(void *arg, void **store, int cnt)
101 {
102 	struct mlx5e_tls_tag *ptag;
103 	struct mlx5e_priv *priv;
104 	struct mlx5e_tls *ptls;
105 	int i;
106 
107 	for (i = 0; i != cnt; i++) {
108 		ptag = store[i];
109 		ptls = ptag->tls;
110 		priv = container_of(ptls, struct mlx5e_priv, tls);
111 
112 		flush_work(&ptag->work);
113 
114 		if (ptag->tisn != 0) {
115 			mlx5_tls_close_tis(priv->mdev, ptag->tisn);
116 			atomic_add_32(&ptls->num_resources, -1U);
117 		}
118 
119 		mtx_destroy(&ptag->mtx);
120 
121 		free(ptag, M_MLX5E_TLS);
122 	}
123 }
124 
125 static void
126 mlx5e_tls_tag_zfree(struct mlx5e_tls_tag *ptag)
127 {
128 	/* make sure any unhandled taskqueue events are ignored */
129 	ptag->state = MLX5E_TLS_ST_FREED;
130 
131 	/* reset some variables */
132 	ptag->dek_index = 0;
133 	ptag->dek_index_ok = 0;
134 
135 	/* avoid leaking keys */
136 	memset(ptag->crypto_params, 0, sizeof(ptag->crypto_params));
137 
138 	/* update number of TIS contexts */
139 	if (ptag->tisn == 0)
140 		atomic_add_32(&ptag->tls->num_resources, -1U);
141 
142 	/* return tag to UMA */
143 	uma_zfree(ptag->tls->zone, ptag);
144 }
145 
146 int
147 mlx5e_tls_init(struct mlx5e_priv *priv)
148 {
149 	struct mlx5e_tls *ptls = &priv->tls;
150 	struct sysctl_oid *node;
151 	uint32_t x;
152 
153 	if (MLX5_CAP_GEN(priv->mdev, tls_tx) == 0 ||
154 	    MLX5_CAP_GEN(priv->mdev, log_max_dek) == 0)
155 		return (0);
156 
157 	ptls->wq = create_singlethread_workqueue("mlx5-tls-wq");
158 	if (ptls->wq == NULL)
159 		return (ENOMEM);
160 
161 	sysctl_ctx_init(&ptls->ctx);
162 
163 	snprintf(ptls->zname, sizeof(ptls->zname),
164 	    "mlx5_%u_tls", device_get_unit(priv->mdev->pdev->dev.bsddev));
165 
166 	ptls->zone = uma_zcache_create(ptls->zname,
167 	     sizeof(struct mlx5e_tls_tag), NULL, NULL, NULL, NULL,
168 	     mlx5e_tls_tag_import, mlx5e_tls_tag_release, priv->mdev,
169 	     UMA_ZONE_UNMANAGED);
170 
171 	/* shared between RX and TX TLS */
172 	ptls->max_resources = 1U << (MLX5_CAP_GEN(priv->mdev, log_max_dek) - 1);
173 
174 	for (x = 0; x != MLX5E_TLS_STATS_NUM; x++)
175 		ptls->stats.arg[x] = counter_u64_alloc(M_WAITOK);
176 
177 	ptls->init = 1;
178 
179 	node = SYSCTL_ADD_NODE(&priv->sysctl_ctx,
180 	    SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO,
181 	    "tls", CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "Hardware TLS offload");
182 	if (node == NULL)
183 		return (0);
184 
185 	mlx5e_create_counter_stats(&ptls->ctx,
186 	    SYSCTL_CHILDREN(node), "stats",
187 	    mlx5e_tls_stats_desc, MLX5E_TLS_STATS_NUM,
188 	    ptls->stats.arg);
189 
190 	return (0);
191 }
192 
193 void
194 mlx5e_tls_cleanup(struct mlx5e_priv *priv)
195 {
196 	struct mlx5e_tls *ptls = &priv->tls;
197 	uint32_t x;
198 
199 	if (ptls->init == 0)
200 		return;
201 
202 	ptls->init = 0;
203 	flush_workqueue(ptls->wq);
204 	sysctl_ctx_free(&ptls->ctx);
205 	uma_zdestroy(ptls->zone);
206 	destroy_workqueue(ptls->wq);
207 
208 	/* check if all resources are freed */
209 	MPASS(priv->tls.num_resources == 0);
210 
211 	for (x = 0; x != MLX5E_TLS_STATS_NUM; x++)
212 		counter_u64_free(ptls->stats.arg[x]);
213 }
214 
215 static void
216 mlx5e_tls_work(struct work_struct *work)
217 {
218 	struct mlx5e_tls_tag *ptag;
219 	struct mlx5e_priv *priv;
220 	int err;
221 
222 	ptag = container_of(work, struct mlx5e_tls_tag, work);
223 	priv = container_of(ptag->tls, struct mlx5e_priv, tls);
224 
225 	switch (ptag->state) {
226 	case MLX5E_TLS_ST_INIT:
227 		/* try to open TIS, if not present */
228 		if (ptag->tisn == 0) {
229 			err = mlx5_tls_open_tis(priv->mdev, 0, priv->tdn,
230 			    priv->pdn, &ptag->tisn);
231 			if (err) {
232 				MLX5E_TLS_STAT_INC(ptag, tx_error, 1);
233 				break;
234 			}
235 		}
236 		MLX5_SET(sw_tls_cntx, ptag->crypto_params, progress.pd, ptag->tisn);
237 
238 		/* try to allocate a DEK context ID */
239 		err = mlx5_encryption_key_create(priv->mdev, priv->pdn,
240 		    MLX5_ADDR_OF(sw_tls_cntx, ptag->crypto_params, key.key_data),
241 		    MLX5_GET(sw_tls_cntx, ptag->crypto_params, key.key_len),
242 		    &ptag->dek_index);
243 		if (err) {
244 			MLX5E_TLS_STAT_INC(ptag, tx_error, 1);
245 			break;
246 		}
247 
248 		MLX5_SET(sw_tls_cntx, ptag->crypto_params, param.dek_index, ptag->dek_index);
249 
250 		ptag->dek_index_ok = 1;
251 
252 		MLX5E_TLS_TAG_LOCK(ptag);
253 		if (ptag->state == MLX5E_TLS_ST_INIT)
254 			ptag->state = MLX5E_TLS_ST_SETUP;
255 		MLX5E_TLS_TAG_UNLOCK(ptag);
256 		break;
257 
258 	case MLX5E_TLS_ST_RELEASE:
259 		/* try to destroy DEK context by ID */
260 		if (ptag->dek_index_ok)
261 			err = mlx5_encryption_key_destroy(priv->mdev, ptag->dek_index);
262 
263 		/* free tag */
264 		mlx5e_tls_tag_zfree(ptag);
265 		break;
266 
267 	default:
268 		break;
269 	}
270 }
271 
272 static int
273 mlx5e_tls_set_params(void *ctx, const struct tls_session_params *en)
274 {
275 
276 	MLX5_SET(sw_tls_cntx, ctx, param.const_2, 2);
277 	if (en->tls_vminor == TLS_MINOR_VER_TWO)
278 		MLX5_SET(sw_tls_cntx, ctx, param.tls_version, 2); /* v1.2 */
279 	else
280 		MLX5_SET(sw_tls_cntx, ctx, param.tls_version, 3); /* v1.3 */
281 	MLX5_SET(sw_tls_cntx, ctx, param.const_1, 1);
282 	MLX5_SET(sw_tls_cntx, ctx, param.encryption_standard, 1); /* TLS */
283 
284 	/* copy the initial vector in place */
285 	switch (en->iv_len) {
286 	case MLX5_FLD_SZ_BYTES(sw_tls_cntx, param.gcm_iv):
287 	case MLX5_FLD_SZ_BYTES(sw_tls_cntx, param.gcm_iv) +
288 	     MLX5_FLD_SZ_BYTES(sw_tls_cntx, param.implicit_iv):
289 		memcpy(MLX5_ADDR_OF(sw_tls_cntx, ctx, param.gcm_iv),
290 		    en->iv, en->iv_len);
291 		break;
292 	default:
293 		return (EINVAL);
294 	}
295 
296 	if (en->cipher_key_len <= MLX5_FLD_SZ_BYTES(sw_tls_cntx, key.key_data)) {
297 		memcpy(MLX5_ADDR_OF(sw_tls_cntx, ctx, key.key_data),
298 		    en->cipher_key, en->cipher_key_len);
299 		MLX5_SET(sw_tls_cntx, ctx, key.key_len, en->cipher_key_len);
300 	} else {
301 		return (EINVAL);
302 	}
303 	return (0);
304 }
305 
306 /* Verify zero default */
307 CTASSERT(MLX5E_TLS_ST_INIT == 0);
308 
309 int
310 mlx5e_tls_snd_tag_alloc(if_t ifp,
311     union if_snd_tag_alloc_params *params,
312     struct m_snd_tag **ppmt)
313 {
314 	union if_snd_tag_alloc_params rl_params;
315 	const struct if_snd_tag_sw *snd_tag_sw;
316 	struct mlx5e_priv *priv;
317 	struct mlx5e_tls_tag *ptag;
318 	const struct tls_session_params *en;
319 	int error;
320 
321 	priv = if_getsoftc(ifp);
322 
323 	if (priv->gone != 0 || priv->tls.init == 0)
324 		return (EOPNOTSUPP);
325 
326 	/* allocate new tag from zone, if any */
327 	ptag = uma_zalloc(priv->tls.zone, M_NOWAIT);
328 	if (ptag == NULL)
329 		return (ENOMEM);
330 
331 	/* sanity check default values */
332 	MPASS(ptag->dek_index == 0);
333 	MPASS(ptag->dek_index_ok == 0);
334 
335 	/* setup TLS tag */
336 	ptag->tls = &priv->tls;
337 
338 	/* check if there is no TIS context */
339 	if (ptag->tisn == 0) {
340 		uint32_t value;
341 
342 		value = atomic_fetchadd_32(&priv->tls.num_resources, 1U);
343 
344 		/* check resource limits */
345 		if (value >= priv->tls.max_resources) {
346 			error = ENOMEM;
347 			goto failure;
348 		}
349 	}
350 
351 	en = &params->tls.tls->params;
352 
353 	/* only TLS v1.2 and v1.3 is currently supported */
354 	if (en->tls_vmajor != TLS_MAJOR_VER_ONE ||
355 	    (en->tls_vminor != TLS_MINOR_VER_TWO
356 #ifdef TLS_MINOR_VER_THREE
357 	     && en->tls_vminor != TLS_MINOR_VER_THREE
358 #endif
359 	     )) {
360 		error = EPROTONOSUPPORT;
361 		goto failure;
362 	}
363 
364 	switch (en->cipher_algorithm) {
365 	case CRYPTO_AES_NIST_GCM_16:
366 		switch (en->cipher_key_len) {
367 		case 128 / 8:
368 			if (en->tls_vminor == TLS_MINOR_VER_TWO) {
369 				if (MLX5_CAP_TLS(priv->mdev, tls_1_2_aes_gcm_128) == 0) {
370 					error = EPROTONOSUPPORT;
371 					goto failure;
372 				}
373 			} else {
374 				if (MLX5_CAP_TLS(priv->mdev, tls_1_3_aes_gcm_128) == 0) {
375 					error = EPROTONOSUPPORT;
376 					goto failure;
377 				}
378 			}
379 			error = mlx5e_tls_set_params(ptag->crypto_params, en);
380 			if (error)
381 				goto failure;
382 			break;
383 
384 		case 256 / 8:
385 			if (en->tls_vminor == TLS_MINOR_VER_TWO) {
386 				if (MLX5_CAP_TLS(priv->mdev, tls_1_2_aes_gcm_256) == 0) {
387 					error = EPROTONOSUPPORT;
388 					goto failure;
389 				}
390 			} else {
391 				if (MLX5_CAP_TLS(priv->mdev, tls_1_3_aes_gcm_256) == 0) {
392 					error = EPROTONOSUPPORT;
393 					goto failure;
394 				}
395 			}
396 			error = mlx5e_tls_set_params(ptag->crypto_params, en);
397 			if (error)
398 				goto failure;
399 			break;
400 
401 		default:
402 			error = EINVAL;
403 			goto failure;
404 		}
405 		break;
406 	default:
407 		error = EPROTONOSUPPORT;
408 		goto failure;
409 	}
410 
411 	memset(&rl_params, 0, sizeof(rl_params));
412 	rl_params.hdr = params->hdr;
413 	switch (params->hdr.type) {
414 #ifdef RATELIMIT
415 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
416 		rl_params.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT;
417 		rl_params.rate_limit.max_rate = params->tls_rate_limit.max_rate;
418 		snd_tag_sw = &mlx5e_tls_rl_snd_tag_sw;
419 		break;
420 #endif
421 	case IF_SND_TAG_TYPE_TLS:
422 		rl_params.hdr.type = IF_SND_TAG_TYPE_UNLIMITED;
423 		snd_tag_sw = &mlx5e_tls_snd_tag_sw;
424 		break;
425 	default:
426 		error = EOPNOTSUPP;
427 		goto failure;
428 	}
429 
430 	error = m_snd_tag_alloc(ifp, &rl_params, &ptag->rl_tag);
431 	if (error)
432 		goto failure;
433 
434 	/* store pointer to mbuf tag */
435 	MPASS(ptag->tag.refcount == 0);
436 	m_snd_tag_init(&ptag->tag, ifp, snd_tag_sw);
437 	*ppmt = &ptag->tag;
438 
439 	/* reset state */
440 	ptag->state = MLX5E_TLS_ST_INIT;
441 
442 	queue_work(priv->tls.wq, &ptag->work);
443 	flush_work(&ptag->work);
444 
445 	return (0);
446 
447 failure:
448 	mlx5e_tls_tag_zfree(ptag);
449 	return (error);
450 }
451 
452 #ifdef RATELIMIT
453 static int
454 mlx5e_tls_rl_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
455 {
456 	union if_snd_tag_modify_params rl_params;
457 	struct mlx5e_tls_tag *ptag =
458 	    container_of(pmt, struct mlx5e_tls_tag, tag);
459 	int error;
460 
461 	memset(&rl_params, 0, sizeof(rl_params));
462 	rl_params.rate_limit.max_rate = params->tls_rate_limit.max_rate;
463 	error = ptag->rl_tag->sw->snd_tag_modify(ptag->rl_tag, &rl_params);
464 	return (error);
465 }
466 #endif
467 
468 static int
469 mlx5e_tls_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
470 {
471 	struct mlx5e_tls_tag *ptag =
472 	    container_of(pmt, struct mlx5e_tls_tag, tag);
473 
474 	return (ptag->rl_tag->sw->snd_tag_query(ptag->rl_tag, params));
475 }
476 
477 static void
478 mlx5e_tls_snd_tag_free(struct m_snd_tag *pmt)
479 {
480 	struct mlx5e_tls_tag *ptag =
481 	    container_of(pmt, struct mlx5e_tls_tag, tag);
482 	struct mlx5e_priv *priv;
483 
484 	m_snd_tag_rele(ptag->rl_tag);
485 
486 	MLX5E_TLS_TAG_LOCK(ptag);
487 	ptag->state = MLX5E_TLS_ST_RELEASE;
488 	MLX5E_TLS_TAG_UNLOCK(ptag);
489 
490 	priv = if_getsoftc(ptag->tag.ifp);
491 	queue_work(priv->tls.wq, &ptag->work);
492 }
493 
494 CTASSERT((MLX5_FLD_SZ_BYTES(sw_tls_cntx, param) % 16) == 0);
495 
496 static void
497 mlx5e_tls_send_static_parameters(struct mlx5e_sq *sq, struct mlx5e_tls_tag *ptag)
498 {
499 	const u32 ds_cnt = DIV_ROUND_UP(sizeof(struct mlx5e_tx_umr_wqe) +
500 	    MLX5_FLD_SZ_BYTES(sw_tls_cntx, param), MLX5_SEND_WQE_DS);
501 	struct mlx5e_tx_umr_wqe *wqe;
502 	u16 pi;
503 
504 	pi = sq->pc & sq->wq.sz_m1;
505 	wqe = mlx5_wq_cyc_get_wqe(&sq->wq, pi);
506 
507 	memset(wqe, 0, sizeof(*wqe));
508 
509 	wqe->ctrl.opmod_idx_opcode = cpu_to_be32((sq->pc << 8) |
510 	    MLX5_OPCODE_UMR | (MLX5_OPCODE_MOD_UMR_TLS_TIS_STATIC_PARAMS << 24));
511 	wqe->ctrl.qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
512 	wqe->ctrl.imm = cpu_to_be32(ptag->tisn << 8);
513 
514 	if (mlx5e_do_send_cqe(sq))
515 		wqe->ctrl.fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE | MLX5_FENCE_MODE_INITIATOR_SMALL;
516 	else
517 		wqe->ctrl.fm_ce_se = MLX5_FENCE_MODE_INITIATOR_SMALL;
518 
519 	/* fill out UMR control segment */
520 	wqe->umr.flags = 0x80;	/* inline data */
521 	wqe->umr.bsf_octowords = cpu_to_be16(MLX5_FLD_SZ_BYTES(sw_tls_cntx, param) / 16);
522 
523 	/* copy in the static crypto parameters */
524 	memcpy(wqe + 1, MLX5_ADDR_OF(sw_tls_cntx, ptag->crypto_params, param),
525 	    MLX5_FLD_SZ_BYTES(sw_tls_cntx, param));
526 
527 	/* copy data for doorbell */
528 	memcpy(sq->doorbell.d32, &wqe->ctrl, sizeof(sq->doorbell.d32));
529 
530 	sq->mbuf[pi].mbuf = NULL;
531 	sq->mbuf[pi].num_bytes = 0;
532 	sq->mbuf[pi].num_wqebbs = DIV_ROUND_UP(ds_cnt, MLX5_SEND_WQEBB_NUM_DS);
533 	sq->mbuf[pi].mst = m_snd_tag_ref(&ptag->tag);
534 
535 	sq->pc += sq->mbuf[pi].num_wqebbs;
536 }
537 
538 CTASSERT(MLX5_FLD_SZ_BYTES(sw_tls_cntx, progress) ==
539     sizeof(((struct mlx5e_tx_psv_wqe *)0)->psv));
540 
541 static void
542 mlx5e_tls_send_progress_parameters(struct mlx5e_sq *sq, struct mlx5e_tls_tag *ptag)
543 {
544 	const u32 ds_cnt = DIV_ROUND_UP(sizeof(struct mlx5e_tx_psv_wqe),
545 	    MLX5_SEND_WQE_DS);
546 	struct mlx5e_tx_psv_wqe *wqe;
547 	u16 pi;
548 
549 	pi = sq->pc & sq->wq.sz_m1;
550 	wqe = mlx5_wq_cyc_get_wqe(&sq->wq, pi);
551 
552 	memset(wqe, 0, sizeof(*wqe));
553 
554 	wqe->ctrl.opmod_idx_opcode = cpu_to_be32((sq->pc << 8) |
555 	    MLX5_OPCODE_SET_PSV | (MLX5_OPCODE_MOD_PSV_TLS_TIS_PROGRESS_PARAMS << 24));
556 	wqe->ctrl.qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
557 
558 	if (mlx5e_do_send_cqe(sq))
559 		wqe->ctrl.fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
560 
561 	/* copy in the PSV control segment */
562 	memcpy(&wqe->psv, MLX5_ADDR_OF(sw_tls_cntx, ptag->crypto_params, progress),
563 	    sizeof(wqe->psv));
564 
565 	/* copy data for doorbell */
566 	memcpy(sq->doorbell.d32, &wqe->ctrl, sizeof(sq->doorbell.d32));
567 
568 	sq->mbuf[pi].mbuf = NULL;
569 	sq->mbuf[pi].num_bytes = 0;
570 	sq->mbuf[pi].num_wqebbs = DIV_ROUND_UP(ds_cnt, MLX5_SEND_WQEBB_NUM_DS);
571 	sq->mbuf[pi].mst = m_snd_tag_ref(&ptag->tag);
572 
573 	sq->pc += sq->mbuf[pi].num_wqebbs;
574 }
575 
576 static void
577 mlx5e_tls_send_nop(struct mlx5e_sq *sq, struct mlx5e_tls_tag *ptag)
578 {
579 	const u32 ds_cnt = MLX5_SEND_WQEBB_NUM_DS;
580 	struct mlx5e_tx_wqe *wqe;
581 	u16 pi;
582 
583 	pi = sq->pc & sq->wq.sz_m1;
584 	wqe = mlx5_wq_cyc_get_wqe(&sq->wq, pi);
585 
586 	memset(&wqe->ctrl, 0, sizeof(wqe->ctrl));
587 
588 	wqe->ctrl.opmod_idx_opcode = cpu_to_be32((sq->pc << 8) | MLX5_OPCODE_NOP);
589 	wqe->ctrl.qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
590 	if (mlx5e_do_send_cqe(sq))
591 		wqe->ctrl.fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE | MLX5_FENCE_MODE_INITIATOR_SMALL;
592 	else
593 		wqe->ctrl.fm_ce_se = MLX5_FENCE_MODE_INITIATOR_SMALL;
594 
595 	/* Copy data for doorbell */
596 	memcpy(sq->doorbell.d32, &wqe->ctrl, sizeof(sq->doorbell.d32));
597 
598 	sq->mbuf[pi].mbuf = NULL;
599 	sq->mbuf[pi].num_bytes = 0;
600 	sq->mbuf[pi].num_wqebbs = DIV_ROUND_UP(ds_cnt, MLX5_SEND_WQEBB_NUM_DS);
601 	sq->mbuf[pi].mst = m_snd_tag_ref(&ptag->tag);
602 
603 	sq->pc += sq->mbuf[pi].num_wqebbs;
604 }
605 
606 #define	SBTLS_MBUF_NO_DATA ((struct mbuf *)1)
607 
608 static struct mbuf *
609 sbtls_recover_record(struct mbuf *mb, int wait, uint32_t tcp_old, uint32_t *ptcp_seq, bool *pis_start)
610 {
611 	struct mbuf *mr, *top;
612 	uint32_t offset;
613 	uint32_t delta;
614 
615 	/* check format of incoming mbuf */
616 	if (mb->m_next == NULL ||
617 	    (mb->m_next->m_flags & (M_EXTPG | M_EXT)) != (M_EXTPG | M_EXT)) {
618 		top = NULL;
619 		goto done;
620 	}
621 
622 	/* get unmapped data offset */
623 	offset = mtod(mb->m_next, uintptr_t);
624 
625 	/* check if we don't need to re-transmit anything */
626 	if (offset == 0) {
627 		top = SBTLS_MBUF_NO_DATA;
628 		*pis_start = true;
629 		goto done;
630 	}
631 
632 	/* try to get a new  packet header */
633 	top = m_gethdr(wait, MT_DATA);
634 	if (top == NULL)
635 		goto done;
636 
637 	mr = m_get(wait, MT_DATA);
638 	if (mr == NULL) {
639 		m_free(top);
640 		top = NULL;
641 		goto done;
642 	}
643 
644 	top->m_next = mr;
645 
646 	mb_dupcl(mr, mb->m_next);
647 
648 	/* the beginning of the TLS record */
649 	mr->m_data = NULL;
650 
651 	/* setup packet header length */
652 	top->m_pkthdr.len = mr->m_len = offset;
653 	top->m_len = 0;
654 
655 	/* check for partial re-transmit */
656 	delta = *ptcp_seq - tcp_old;
657 
658 	if (delta < offset) {
659 		m_adj(top, offset - delta);
660 		offset = delta;
661 
662 		/* continue where we left off */
663 		*pis_start = false;
664 	} else {
665 		*pis_start = true;
666 	}
667 
668 	/*
669 	 * Rewind the TCP sequence number by the amount of data
670 	 * retransmitted:
671 	 */
672 	*ptcp_seq -= offset;
673 done:
674 	return (top);
675 }
676 
677 static int
678 mlx5e_sq_tls_populate(struct mbuf *mb, uint64_t *pseq)
679 {
680 
681 	for (; mb != NULL; mb = mb->m_next) {
682 		if (!(mb->m_flags & M_EXTPG))
683 			continue;
684 		*pseq = mb->m_epg_seqno;
685 		return (1);
686 	}
687 	return (0);
688 }
689 
690 int
691 mlx5e_sq_tls_xmit(struct mlx5e_sq *sq, struct mlx5e_xmit_args *parg, struct mbuf **ppmb)
692 {
693 	struct mlx5e_tls_tag *ptls_tag;
694 	struct m_snd_tag *ptag;
695 	const struct tcphdr *th;
696 	struct mbuf *mb = *ppmb;
697 	u64 rcd_sn;
698 	u32 header_size;
699 	u32 mb_seq;
700 
701 	if ((mb->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0)
702 		return (MLX5E_TLS_CONTINUE);
703 
704 	ptag = mb->m_pkthdr.snd_tag;
705 
706 	if (
707 #ifdef RATELIMIT
708 	    ptag->sw->type != IF_SND_TAG_TYPE_TLS_RATE_LIMIT &&
709 #endif
710 	    ptag->sw->type != IF_SND_TAG_TYPE_TLS)
711 		return (MLX5E_TLS_CONTINUE);
712 
713 	ptls_tag = container_of(ptag, struct mlx5e_tls_tag, tag);
714 
715 	header_size = mlx5e_get_full_header_size(mb, &th);
716 	if (unlikely(header_size == 0 || th == NULL))
717 		return (MLX5E_TLS_FAILURE);
718 
719 	/*
720 	 * Send non-TLS TCP packets AS-IS:
721 	 */
722 	if (header_size == mb->m_pkthdr.len ||
723 	    mlx5e_sq_tls_populate(mb, &rcd_sn) == 0) {
724 		parg->tisn = 0;
725 		parg->ihs = header_size;
726 		return (MLX5E_TLS_CONTINUE);
727 	}
728 
729 	mb_seq = ntohl(th->th_seq);
730 
731 	MLX5E_TLS_TAG_LOCK(ptls_tag);
732 	switch (ptls_tag->state) {
733 	case MLX5E_TLS_ST_INIT:
734 		MLX5E_TLS_TAG_UNLOCK(ptls_tag);
735 		return (MLX5E_TLS_FAILURE);
736 	case MLX5E_TLS_ST_SETUP:
737 		ptls_tag->state = MLX5E_TLS_ST_TXRDY;
738 		ptls_tag->expected_seq = ~mb_seq;	/* force setup */
739 	default:
740 		MLX5E_TLS_TAG_UNLOCK(ptls_tag);
741 		break;
742 	}
743 
744 	if (unlikely(ptls_tag->expected_seq != mb_seq)) {
745 		bool is_start;
746 		struct mbuf *r_mb;
747 		uint32_t tcp_seq = mb_seq;
748 
749 		r_mb = sbtls_recover_record(mb, M_NOWAIT, ptls_tag->expected_seq, &tcp_seq, &is_start);
750 		if (r_mb == NULL) {
751 			MLX5E_TLS_STAT_INC(ptls_tag, tx_error, 1);
752 			return (MLX5E_TLS_FAILURE);
753 		}
754 
755 		MLX5E_TLS_STAT_INC(ptls_tag, tx_packets_ooo, 1);
756 
757 		/* check if this is the first fragment of a TLS record */
758 		if (is_start) {
759 			/* setup TLS static parameters */
760 			MLX5_SET64(sw_tls_cntx, ptls_tag->crypto_params,
761 			    param.initial_record_number, rcd_sn);
762 
763 			/*
764 			 * NOTE: The sendqueue should have enough room to
765 			 * carry both the static and the progress parameters
766 			 * when we get here!
767 			 */
768 			mlx5e_tls_send_static_parameters(sq, ptls_tag);
769 			mlx5e_tls_send_progress_parameters(sq, ptls_tag);
770 
771 			if (r_mb == SBTLS_MBUF_NO_DATA) {
772 				mlx5e_tls_send_nop(sq, ptls_tag);
773 				ptls_tag->expected_seq = mb_seq;
774 				return (MLX5E_TLS_LOOP);
775 			}
776 		}
777 
778 		MLX5E_TLS_STAT_INC(ptls_tag, tx_bytes_ooo, r_mb->m_pkthdr.len);
779 
780 		/* setup transmit arguments */
781 		parg->tisn = ptls_tag->tisn;
782 		parg->mst = &ptls_tag->tag;
783 
784 		/* try to send DUMP data */
785 		if (mlx5e_sq_dump_xmit(sq, parg, &r_mb) != 0) {
786 			m_freem(r_mb);
787 			ptls_tag->expected_seq = tcp_seq;
788 			return (MLX5E_TLS_FAILURE);
789 		} else {
790 			ptls_tag->expected_seq = mb_seq;
791 			return (MLX5E_TLS_LOOP);
792 		}
793 	} else {
794 		MLX5E_TLS_STAT_INC(ptls_tag, tx_packets, 1);
795 		MLX5E_TLS_STAT_INC(ptls_tag, tx_bytes, mb->m_pkthdr.len);
796 	}
797 	ptls_tag->expected_seq += mb->m_pkthdr.len - header_size;
798 
799 	parg->tisn = ptls_tag->tisn;
800 	parg->ihs = header_size;
801 	parg->mst = &ptls_tag->tag;
802 	return (MLX5E_TLS_CONTINUE);
803 }
804 
805 #else
806 
807 int
808 mlx5e_tls_init(struct mlx5e_priv *priv)
809 {
810 
811 	return (0);
812 }
813 
814 void
815 mlx5e_tls_cleanup(struct mlx5e_priv *priv)
816 {
817 	/* NOP */
818 }
819 
820 #endif		/* KERN_TLS */
821