xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec.c (revision 307797159ac25fe5a2048bf5c6a5718298edca57)
1 /*
2  * Copyright (c) 2017 Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 
34 #include <crypto/internal/geniv.h>
35 #include <crypto/aead.h>
36 #include <linux/inetdevice.h>
37 #include <linux/netdevice.h>
38 #include <linux/module.h>
39 
40 #include "en.h"
41 #include "en_accel/ipsec.h"
42 #include "en_accel/ipsec_rxtx.h"
43 
44 
45 static struct mlx5e_ipsec_sa_entry *to_ipsec_sa_entry(struct xfrm_state *x)
46 {
47 	struct mlx5e_ipsec_sa_entry *sa;
48 
49 	if (!x)
50 		return NULL;
51 
52 	sa = (struct mlx5e_ipsec_sa_entry *)x->xso.offload_handle;
53 	if (!sa)
54 		return NULL;
55 
56 	WARN_ON(sa->x != x);
57 	return sa;
58 }
59 
60 struct xfrm_state *mlx5e_ipsec_sadb_rx_lookup(struct mlx5e_ipsec *ipsec,
61 					      unsigned int handle)
62 {
63 	struct mlx5e_ipsec_sa_entry *sa_entry;
64 	struct xfrm_state *ret = NULL;
65 
66 	rcu_read_lock();
67 	hash_for_each_possible_rcu(ipsec->sadb_rx, sa_entry, hlist, handle)
68 		if (sa_entry->handle == handle) {
69 			ret = sa_entry->x;
70 			xfrm_state_hold(ret);
71 			break;
72 		}
73 	rcu_read_unlock();
74 
75 	return ret;
76 }
77 
78 static int mlx5e_ipsec_sadb_rx_add(struct mlx5e_ipsec_sa_entry *sa_entry)
79 {
80 	struct mlx5e_ipsec *ipsec = sa_entry->ipsec;
81 	unsigned long flags;
82 	int ret;
83 
84 	ret = ida_simple_get(&ipsec->halloc, 1, 0, GFP_KERNEL);
85 	if (ret < 0)
86 		return ret;
87 
88 	spin_lock_irqsave(&ipsec->sadb_rx_lock, flags);
89 	sa_entry->handle = ret;
90 	hash_add_rcu(ipsec->sadb_rx, &sa_entry->hlist, sa_entry->handle);
91 	spin_unlock_irqrestore(&ipsec->sadb_rx_lock, flags);
92 
93 	return 0;
94 }
95 
96 static void mlx5e_ipsec_sadb_rx_del(struct mlx5e_ipsec_sa_entry *sa_entry)
97 {
98 	struct mlx5e_ipsec *ipsec = sa_entry->ipsec;
99 	unsigned long flags;
100 
101 	spin_lock_irqsave(&ipsec->sadb_rx_lock, flags);
102 	hash_del_rcu(&sa_entry->hlist);
103 	spin_unlock_irqrestore(&ipsec->sadb_rx_lock, flags);
104 }
105 
106 static void mlx5e_ipsec_sadb_rx_free(struct mlx5e_ipsec_sa_entry *sa_entry)
107 {
108 	struct mlx5e_ipsec *ipsec = sa_entry->ipsec;
109 
110 	/* xfrm already doing sync rcu between del and free callbacks */
111 
112 	ida_simple_remove(&ipsec->halloc, sa_entry->handle);
113 }
114 
115 static bool mlx5e_ipsec_update_esn_state(struct mlx5e_ipsec_sa_entry *sa_entry)
116 {
117 	struct xfrm_replay_state_esn *replay_esn;
118 	u32 seq_bottom;
119 	u8 overlap;
120 	u32 *esn;
121 
122 	if (!(sa_entry->x->props.flags & XFRM_STATE_ESN)) {
123 		sa_entry->esn_state.trigger = 0;
124 		return false;
125 	}
126 
127 	replay_esn = sa_entry->x->replay_esn;
128 	seq_bottom = replay_esn->seq - replay_esn->replay_window + 1;
129 	overlap = sa_entry->esn_state.overlap;
130 
131 	sa_entry->esn_state.esn = xfrm_replay_seqhi(sa_entry->x,
132 						    htonl(seq_bottom));
133 	esn = &sa_entry->esn_state.esn;
134 
135 	sa_entry->esn_state.trigger = 1;
136 	if (unlikely(overlap && seq_bottom < MLX5E_IPSEC_ESN_SCOPE_MID)) {
137 		++(*esn);
138 		sa_entry->esn_state.overlap = 0;
139 		return true;
140 	} else if (unlikely(!overlap &&
141 			    (seq_bottom >= MLX5E_IPSEC_ESN_SCOPE_MID))) {
142 		sa_entry->esn_state.overlap = 1;
143 		return true;
144 	}
145 
146 	return false;
147 }
148 
149 static void
150 mlx5e_ipsec_build_accel_xfrm_attrs(struct mlx5e_ipsec_sa_entry *sa_entry,
151 				   struct mlx5_accel_esp_xfrm_attrs *attrs)
152 {
153 	struct xfrm_state *x = sa_entry->x;
154 	struct aes_gcm_keymat *aes_gcm = &attrs->keymat.aes_gcm;
155 	struct aead_geniv_ctx *geniv_ctx;
156 	struct crypto_aead *aead;
157 	unsigned int crypto_data_len, key_len;
158 	int ivsize;
159 
160 	memset(attrs, 0, sizeof(*attrs));
161 
162 	/* key */
163 	crypto_data_len = (x->aead->alg_key_len + 7) / 8;
164 	key_len = crypto_data_len - 4; /* 4 bytes salt at end */
165 
166 	memcpy(aes_gcm->aes_key, x->aead->alg_key, key_len);
167 	aes_gcm->key_len = key_len * 8;
168 
169 	/* salt and seq_iv */
170 	aead = x->data;
171 	geniv_ctx = crypto_aead_ctx(aead);
172 	ivsize = crypto_aead_ivsize(aead);
173 	memcpy(&aes_gcm->seq_iv, &geniv_ctx->salt, ivsize);
174 	memcpy(&aes_gcm->salt, x->aead->alg_key + key_len,
175 	       sizeof(aes_gcm->salt));
176 
177 	/* iv len */
178 	aes_gcm->icv_len = x->aead->alg_icv_len;
179 
180 	/* esn */
181 	if (sa_entry->esn_state.trigger) {
182 		attrs->flags |= MLX5_ACCEL_ESP_FLAGS_ESN_TRIGGERED;
183 		attrs->esn = sa_entry->esn_state.esn;
184 		if (sa_entry->esn_state.overlap)
185 			attrs->flags |= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
186 	}
187 
188 	/* rx handle */
189 	attrs->sa_handle = sa_entry->handle;
190 
191 	/* algo type */
192 	attrs->keymat_type = MLX5_ACCEL_ESP_KEYMAT_AES_GCM;
193 
194 	/* action */
195 	attrs->action = (!(x->xso.flags & XFRM_OFFLOAD_INBOUND)) ?
196 			MLX5_ACCEL_ESP_ACTION_ENCRYPT :
197 			MLX5_ACCEL_ESP_ACTION_DECRYPT;
198 	/* flags */
199 	attrs->flags |= (x->props.mode == XFRM_MODE_TRANSPORT) ?
200 			MLX5_ACCEL_ESP_FLAGS_TRANSPORT :
201 			MLX5_ACCEL_ESP_FLAGS_TUNNEL;
202 }
203 
204 static inline int mlx5e_xfrm_validate_state(struct xfrm_state *x)
205 {
206 	struct net_device *netdev = x->xso.dev;
207 	struct mlx5e_priv *priv;
208 
209 	priv = netdev_priv(netdev);
210 
211 	if (x->props.aalgo != SADB_AALG_NONE) {
212 		netdev_info(netdev, "Cannot offload authenticated xfrm states\n");
213 		return -EINVAL;
214 	}
215 	if (x->props.ealgo != SADB_X_EALG_AES_GCM_ICV16) {
216 		netdev_info(netdev, "Only AES-GCM-ICV16 xfrm state may be offloaded\n");
217 		return -EINVAL;
218 	}
219 	if (x->props.calgo != SADB_X_CALG_NONE) {
220 		netdev_info(netdev, "Cannot offload compressed xfrm states\n");
221 		return -EINVAL;
222 	}
223 	if (x->props.flags & XFRM_STATE_ESN &&
224 	    !(mlx5_accel_ipsec_device_caps(priv->mdev) &
225 	    MLX5_ACCEL_IPSEC_CAP_ESN)) {
226 		netdev_info(netdev, "Cannot offload ESN xfrm states\n");
227 		return -EINVAL;
228 	}
229 	if (x->props.family != AF_INET &&
230 	    x->props.family != AF_INET6) {
231 		netdev_info(netdev, "Only IPv4/6 xfrm states may be offloaded\n");
232 		return -EINVAL;
233 	}
234 	if (x->props.mode != XFRM_MODE_TRANSPORT &&
235 	    x->props.mode != XFRM_MODE_TUNNEL) {
236 		dev_info(&netdev->dev, "Only transport and tunnel xfrm states may be offloaded\n");
237 		return -EINVAL;
238 	}
239 	if (x->id.proto != IPPROTO_ESP) {
240 		netdev_info(netdev, "Only ESP xfrm state may be offloaded\n");
241 		return -EINVAL;
242 	}
243 	if (x->encap) {
244 		netdev_info(netdev, "Encapsulated xfrm state may not be offloaded\n");
245 		return -EINVAL;
246 	}
247 	if (!x->aead) {
248 		netdev_info(netdev, "Cannot offload xfrm states without aead\n");
249 		return -EINVAL;
250 	}
251 	if (x->aead->alg_icv_len != 128) {
252 		netdev_info(netdev, "Cannot offload xfrm states with AEAD ICV length other than 128bit\n");
253 		return -EINVAL;
254 	}
255 	if ((x->aead->alg_key_len != 128 + 32) &&
256 	    (x->aead->alg_key_len != 256 + 32)) {
257 		netdev_info(netdev, "Cannot offload xfrm states with AEAD key length other than 128/256 bit\n");
258 		return -EINVAL;
259 	}
260 	if (x->tfcpad) {
261 		netdev_info(netdev, "Cannot offload xfrm states with tfc padding\n");
262 		return -EINVAL;
263 	}
264 	if (!x->geniv) {
265 		netdev_info(netdev, "Cannot offload xfrm states without geniv\n");
266 		return -EINVAL;
267 	}
268 	if (strcmp(x->geniv, "seqiv")) {
269 		netdev_info(netdev, "Cannot offload xfrm states with geniv other than seqiv\n");
270 		return -EINVAL;
271 	}
272 	if (x->props.family == AF_INET6 &&
273 	    !(mlx5_accel_ipsec_device_caps(priv->mdev) &
274 	     MLX5_ACCEL_IPSEC_CAP_IPV6)) {
275 		netdev_info(netdev, "IPv6 xfrm state offload is not supported by this device\n");
276 		return -EINVAL;
277 	}
278 	return 0;
279 }
280 
281 static int mlx5e_xfrm_add_state(struct xfrm_state *x)
282 {
283 	struct mlx5e_ipsec_sa_entry *sa_entry = NULL;
284 	struct net_device *netdev = x->xso.dev;
285 	struct mlx5_accel_esp_xfrm_attrs attrs;
286 	struct mlx5e_priv *priv;
287 	__be32 saddr[4] = {0}, daddr[4] = {0}, spi;
288 	bool is_ipv6 = false;
289 	int err;
290 
291 	priv = netdev_priv(netdev);
292 
293 	err = mlx5e_xfrm_validate_state(x);
294 	if (err)
295 		return err;
296 
297 	sa_entry = kzalloc(sizeof(*sa_entry), GFP_KERNEL);
298 	if (!sa_entry) {
299 		err = -ENOMEM;
300 		goto out;
301 	}
302 
303 	sa_entry->x = x;
304 	sa_entry->ipsec = priv->ipsec;
305 
306 	/* Add the SA to handle processed incoming packets before the add SA
307 	 * completion was received
308 	 */
309 	if (x->xso.flags & XFRM_OFFLOAD_INBOUND) {
310 		err = mlx5e_ipsec_sadb_rx_add(sa_entry);
311 		if (err) {
312 			netdev_info(netdev, "Failed adding to SADB_RX: %d\n", err);
313 			goto err_entry;
314 		}
315 	} else {
316 		sa_entry->set_iv_op = (x->props.flags & XFRM_STATE_ESN) ?
317 				mlx5e_ipsec_set_iv_esn : mlx5e_ipsec_set_iv;
318 	}
319 
320 	/* check esn */
321 	mlx5e_ipsec_update_esn_state(sa_entry);
322 
323 	/* create xfrm */
324 	mlx5e_ipsec_build_accel_xfrm_attrs(sa_entry, &attrs);
325 	sa_entry->xfrm =
326 		mlx5_accel_esp_create_xfrm(priv->mdev, &attrs,
327 					   MLX5_ACCEL_XFRM_FLAG_REQUIRE_METADATA);
328 	if (IS_ERR(sa_entry->xfrm)) {
329 		err = PTR_ERR(sa_entry->xfrm);
330 		goto err_sadb_rx;
331 	}
332 
333 	/* create hw context */
334 	if (x->props.family == AF_INET) {
335 		saddr[3] = x->props.saddr.a4;
336 		daddr[3] = x->id.daddr.a4;
337 	} else {
338 		memcpy(saddr, x->props.saddr.a6, sizeof(saddr));
339 		memcpy(daddr, x->id.daddr.a6, sizeof(daddr));
340 		is_ipv6 = true;
341 	}
342 	spi = x->id.spi;
343 	sa_entry->hw_context =
344 			mlx5_accel_esp_create_hw_context(priv->mdev,
345 							 sa_entry->xfrm,
346 							 saddr, daddr, spi,
347 							 is_ipv6);
348 	if (IS_ERR(sa_entry->hw_context)) {
349 		err = PTR_ERR(sa_entry->hw_context);
350 		goto err_xfrm;
351 	}
352 
353 	x->xso.offload_handle = (unsigned long)sa_entry;
354 	goto out;
355 
356 err_xfrm:
357 	mlx5_accel_esp_destroy_xfrm(sa_entry->xfrm);
358 err_sadb_rx:
359 	if (x->xso.flags & XFRM_OFFLOAD_INBOUND) {
360 		mlx5e_ipsec_sadb_rx_del(sa_entry);
361 		mlx5e_ipsec_sadb_rx_free(sa_entry);
362 	}
363 err_entry:
364 	kfree(sa_entry);
365 out:
366 	return err;
367 }
368 
369 static void mlx5e_xfrm_del_state(struct xfrm_state *x)
370 {
371 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
372 
373 	if (!sa_entry)
374 		return;
375 
376 	if (x->xso.flags & XFRM_OFFLOAD_INBOUND)
377 		mlx5e_ipsec_sadb_rx_del(sa_entry);
378 }
379 
380 static void mlx5e_xfrm_free_state(struct xfrm_state *x)
381 {
382 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
383 
384 	if (!sa_entry)
385 		return;
386 
387 	if (sa_entry->hw_context) {
388 		flush_workqueue(sa_entry->ipsec->wq);
389 		mlx5_accel_esp_free_hw_context(sa_entry->hw_context);
390 		mlx5_accel_esp_destroy_xfrm(sa_entry->xfrm);
391 	}
392 
393 	if (x->xso.flags & XFRM_OFFLOAD_INBOUND)
394 		mlx5e_ipsec_sadb_rx_free(sa_entry);
395 
396 	kfree(sa_entry);
397 }
398 
399 int mlx5e_ipsec_init(struct mlx5e_priv *priv)
400 {
401 	struct mlx5e_ipsec *ipsec = NULL;
402 
403 	if (!MLX5_IPSEC_DEV(priv->mdev)) {
404 		netdev_dbg(priv->netdev, "Not an IPSec offload device\n");
405 		return 0;
406 	}
407 
408 	ipsec = kzalloc(sizeof(*ipsec), GFP_KERNEL);
409 	if (!ipsec)
410 		return -ENOMEM;
411 
412 	hash_init(ipsec->sadb_rx);
413 	spin_lock_init(&ipsec->sadb_rx_lock);
414 	ida_init(&ipsec->halloc);
415 	ipsec->en_priv = priv;
416 	ipsec->en_priv->ipsec = ipsec;
417 	ipsec->no_trailer = !!(mlx5_accel_ipsec_device_caps(priv->mdev) &
418 			       MLX5_ACCEL_IPSEC_CAP_RX_NO_TRAILER);
419 	ipsec->wq = alloc_ordered_workqueue("mlx5e_ipsec: %s", 0,
420 					    priv->netdev->name);
421 	if (!ipsec->wq) {
422 		kfree(ipsec);
423 		return -ENOMEM;
424 	}
425 	netdev_dbg(priv->netdev, "IPSec attached to netdevice\n");
426 	return 0;
427 }
428 
429 void mlx5e_ipsec_cleanup(struct mlx5e_priv *priv)
430 {
431 	struct mlx5e_ipsec *ipsec = priv->ipsec;
432 
433 	if (!ipsec)
434 		return;
435 
436 	drain_workqueue(ipsec->wq);
437 	destroy_workqueue(ipsec->wq);
438 
439 	ida_destroy(&ipsec->halloc);
440 	kfree(ipsec);
441 	priv->ipsec = NULL;
442 }
443 
444 static bool mlx5e_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
445 {
446 	if (x->props.family == AF_INET) {
447 		/* Offload with IPv4 options is not supported yet */
448 		if (ip_hdr(skb)->ihl > 5)
449 			return false;
450 	} else {
451 		/* Offload with IPv6 extension headers is not support yet */
452 		if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr))
453 			return false;
454 	}
455 
456 	return true;
457 }
458 
459 struct mlx5e_ipsec_modify_state_work {
460 	struct work_struct		work;
461 	struct mlx5_accel_esp_xfrm_attrs attrs;
462 	struct mlx5e_ipsec_sa_entry	*sa_entry;
463 };
464 
465 static void _update_xfrm_state(struct work_struct *work)
466 {
467 	int ret;
468 	struct mlx5e_ipsec_modify_state_work *modify_work =
469 		container_of(work, struct mlx5e_ipsec_modify_state_work, work);
470 	struct mlx5e_ipsec_sa_entry *sa_entry = modify_work->sa_entry;
471 
472 	ret = mlx5_accel_esp_modify_xfrm(sa_entry->xfrm,
473 					 &modify_work->attrs);
474 	if (ret)
475 		netdev_warn(sa_entry->ipsec->en_priv->netdev,
476 			    "Not an IPSec offload device\n");
477 
478 	kfree(modify_work);
479 }
480 
481 static void mlx5e_xfrm_advance_esn_state(struct xfrm_state *x)
482 {
483 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
484 	struct mlx5e_ipsec_modify_state_work *modify_work;
485 	bool need_update;
486 
487 	if (!sa_entry)
488 		return;
489 
490 	need_update = mlx5e_ipsec_update_esn_state(sa_entry);
491 	if (!need_update)
492 		return;
493 
494 	modify_work = kzalloc(sizeof(*modify_work), GFP_ATOMIC);
495 	if (!modify_work)
496 		return;
497 
498 	mlx5e_ipsec_build_accel_xfrm_attrs(sa_entry, &modify_work->attrs);
499 	modify_work->sa_entry = sa_entry;
500 
501 	INIT_WORK(&modify_work->work, _update_xfrm_state);
502 	WARN_ON(!queue_work(sa_entry->ipsec->wq, &modify_work->work));
503 }
504 
505 static const struct xfrmdev_ops mlx5e_ipsec_xfrmdev_ops = {
506 	.xdo_dev_state_add	= mlx5e_xfrm_add_state,
507 	.xdo_dev_state_delete	= mlx5e_xfrm_del_state,
508 	.xdo_dev_state_free	= mlx5e_xfrm_free_state,
509 	.xdo_dev_offload_ok	= mlx5e_ipsec_offload_ok,
510 	.xdo_dev_state_advance_esn = mlx5e_xfrm_advance_esn_state,
511 };
512 
513 void mlx5e_ipsec_build_netdev(struct mlx5e_priv *priv)
514 {
515 	struct mlx5_core_dev *mdev = priv->mdev;
516 	struct net_device *netdev = priv->netdev;
517 
518 	if (!priv->ipsec)
519 		return;
520 
521 	if (!(mlx5_accel_ipsec_device_caps(mdev) & MLX5_ACCEL_IPSEC_CAP_ESP) ||
522 	    !MLX5_CAP_ETH(mdev, swp)) {
523 		mlx5_core_dbg(mdev, "mlx5e: ESP and SWP offload not supported\n");
524 		return;
525 	}
526 
527 	mlx5_core_info(mdev, "mlx5e: IPSec ESP acceleration enabled\n");
528 	netdev->xfrmdev_ops = &mlx5e_ipsec_xfrmdev_ops;
529 	netdev->features |= NETIF_F_HW_ESP;
530 	netdev->hw_enc_features |= NETIF_F_HW_ESP;
531 
532 	if (!MLX5_CAP_ETH(mdev, swp_csum)) {
533 		mlx5_core_dbg(mdev, "mlx5e: SWP checksum not supported\n");
534 		return;
535 	}
536 
537 	netdev->features |= NETIF_F_HW_ESP_TX_CSUM;
538 	netdev->hw_enc_features |= NETIF_F_HW_ESP_TX_CSUM;
539 
540 	if (!(mlx5_accel_ipsec_device_caps(mdev) & MLX5_ACCEL_IPSEC_CAP_LSO) ||
541 	    !MLX5_CAP_ETH(mdev, swp_lso)) {
542 		mlx5_core_dbg(mdev, "mlx5e: ESP LSO not supported\n");
543 		return;
544 	}
545 
546 	mlx5_core_dbg(mdev, "mlx5e: ESP GSO capability turned on\n");
547 	netdev->features |= NETIF_F_GSO_ESP;
548 	netdev->hw_features |= NETIF_F_GSO_ESP;
549 	netdev->hw_enc_features |= NETIF_F_GSO_ESP;
550 }
551