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 <net/netevent.h>
39 
40 #include "en.h"
41 #include "eswitch.h"
42 #include "ipsec.h"
43 #include "ipsec_rxtx.h"
44 #include "en_rep.h"
45 
46 #define MLX5_IPSEC_RESCHED msecs_to_jiffies(1000)
47 #define MLX5E_IPSEC_TUNNEL_SA XA_MARK_1
48 
to_ipsec_sa_entry(struct xfrm_state * x)49 static struct mlx5e_ipsec_sa_entry *to_ipsec_sa_entry(struct xfrm_state *x)
50 {
51 	return (struct mlx5e_ipsec_sa_entry *)x->xso.offload_handle;
52 }
53 
to_ipsec_pol_entry(struct xfrm_policy * x)54 static struct mlx5e_ipsec_pol_entry *to_ipsec_pol_entry(struct xfrm_policy *x)
55 {
56 	return (struct mlx5e_ipsec_pol_entry *)x->xdo.offload_handle;
57 }
58 
mlx5e_ipsec_handle_tx_limit(struct work_struct * _work)59 static void mlx5e_ipsec_handle_tx_limit(struct work_struct *_work)
60 {
61 	struct mlx5e_ipsec_dwork *dwork =
62 		container_of(_work, struct mlx5e_ipsec_dwork, dwork.work);
63 	struct mlx5e_ipsec_sa_entry *sa_entry = dwork->sa_entry;
64 	struct xfrm_state *x = sa_entry->x;
65 
66 	if (sa_entry->attrs.drop)
67 		return;
68 
69 	spin_lock_bh(&x->lock);
70 	xfrm_state_check_expire(x);
71 	if (x->km.state == XFRM_STATE_EXPIRED) {
72 		sa_entry->attrs.drop = true;
73 		spin_unlock_bh(&x->lock);
74 
75 		mlx5e_accel_ipsec_fs_modify(sa_entry);
76 		return;
77 	}
78 	spin_unlock_bh(&x->lock);
79 
80 	queue_delayed_work(sa_entry->ipsec->wq, &dwork->dwork,
81 			   MLX5_IPSEC_RESCHED);
82 }
83 
mlx5e_ipsec_update_esn_state(struct mlx5e_ipsec_sa_entry * sa_entry)84 static bool mlx5e_ipsec_update_esn_state(struct mlx5e_ipsec_sa_entry *sa_entry)
85 {
86 	struct xfrm_state *x = sa_entry->x;
87 	u32 seq_bottom = 0;
88 	u32 esn, esn_msb;
89 	u8 overlap;
90 
91 	switch (x->xso.type) {
92 	case XFRM_DEV_OFFLOAD_PACKET:
93 		switch (x->xso.dir) {
94 		case XFRM_DEV_OFFLOAD_IN:
95 			esn = x->replay_esn->seq;
96 			esn_msb = x->replay_esn->seq_hi;
97 			break;
98 		case XFRM_DEV_OFFLOAD_OUT:
99 			esn = x->replay_esn->oseq;
100 			esn_msb = x->replay_esn->oseq_hi;
101 			break;
102 		default:
103 			WARN_ON(true);
104 			return false;
105 		}
106 		break;
107 	case XFRM_DEV_OFFLOAD_CRYPTO:
108 		/* Already parsed by XFRM core */
109 		esn = x->replay_esn->seq;
110 		break;
111 	default:
112 		WARN_ON(true);
113 		return false;
114 	}
115 
116 	overlap = sa_entry->esn_state.overlap;
117 
118 	if (esn >= x->replay_esn->replay_window)
119 		seq_bottom = esn - x->replay_esn->replay_window + 1;
120 
121 	if (x->xso.type == XFRM_DEV_OFFLOAD_CRYPTO)
122 		esn_msb = xfrm_replay_seqhi(x, htonl(seq_bottom));
123 
124 	if (sa_entry->esn_state.esn_msb)
125 		sa_entry->esn_state.esn = esn;
126 	else
127 		/* According to RFC4303, section "3.3.3. Sequence Number Generation",
128 		 * the first packet sent using a given SA will contain a sequence
129 		 * number of 1.
130 		 */
131 		sa_entry->esn_state.esn = max_t(u32, esn, 1);
132 	sa_entry->esn_state.esn_msb = esn_msb;
133 
134 	if (unlikely(overlap && seq_bottom < MLX5E_IPSEC_ESN_SCOPE_MID)) {
135 		sa_entry->esn_state.overlap = 0;
136 		return true;
137 	} else if (unlikely(!overlap &&
138 			    (seq_bottom >= MLX5E_IPSEC_ESN_SCOPE_MID))) {
139 		sa_entry->esn_state.overlap = 1;
140 		return true;
141 	}
142 
143 	return false;
144 }
145 
mlx5e_ipsec_init_limits(struct mlx5e_ipsec_sa_entry * sa_entry,struct mlx5_accel_esp_xfrm_attrs * attrs)146 static void mlx5e_ipsec_init_limits(struct mlx5e_ipsec_sa_entry *sa_entry,
147 				    struct mlx5_accel_esp_xfrm_attrs *attrs)
148 {
149 	struct xfrm_state *x = sa_entry->x;
150 	s64 start_value, n;
151 
152 	attrs->lft.hard_packet_limit = x->lft.hard_packet_limit;
153 	attrs->lft.soft_packet_limit = x->lft.soft_packet_limit;
154 	if (x->lft.soft_packet_limit == XFRM_INF)
155 		return;
156 
157 	/* Compute hard limit initial value and number of rounds.
158 	 *
159 	 * The counting pattern of hardware counter goes:
160 	 *                value  -> 2^31-1
161 	 *      2^31  | (2^31-1) -> 2^31-1
162 	 *      2^31  | (2^31-1) -> 2^31-1
163 	 *      [..]
164 	 *      2^31  | (2^31-1) -> 0
165 	 *
166 	 * The pattern is created by using an ASO operation to atomically set
167 	 * bit 31 after the down counter clears bit 31. This is effectively an
168 	 * atomic addition of 2**31 to the counter.
169 	 *
170 	 * We wish to configure the counter, within the above pattern, so that
171 	 * when it reaches 0, it has hit the hard limit. This is defined by this
172 	 * system of equations:
173 	 *
174 	 *      hard_limit == start_value + n * 2^31
175 	 *      n >= 0
176 	 *      start_value < 2^32, start_value >= 0
177 	 *
178 	 * These equations are not single-solution, there are often two choices:
179 	 *      hard_limit == start_value + n * 2^31
180 	 *      hard_limit == (start_value+2^31) + (n-1) * 2^31
181 	 *
182 	 * The algorithm selects the solution that keeps the counter value
183 	 * above 2^31 until the final iteration.
184 	 */
185 
186 	/* Start by estimating n and compute start_value */
187 	n = attrs->lft.hard_packet_limit / BIT_ULL(31);
188 	start_value = attrs->lft.hard_packet_limit - n * BIT_ULL(31);
189 
190 	/* Choose the best of the two solutions: */
191 	if (n >= 1)
192 		n -= 1;
193 
194 	/* Computed values solve the system of equations: */
195 	start_value = attrs->lft.hard_packet_limit - n * BIT_ULL(31);
196 
197 	/* The best solution means: when there are multiple iterations we must
198 	 * start above 2^31 and count down to 2**31 to get the interrupt.
199 	 */
200 	attrs->lft.hard_packet_limit = lower_32_bits(start_value);
201 	attrs->lft.numb_rounds_hard = (u64)n;
202 
203 	/* Compute soft limit initial value and number of rounds.
204 	 *
205 	 * The soft_limit is achieved by adjusting the counter's
206 	 * interrupt_value. This is embedded in the counting pattern created by
207 	 * hard packet calculations above.
208 	 *
209 	 * We wish to compute the interrupt_value for the soft_limit. This is
210 	 * defined by this system of equations:
211 	 *
212 	 *      soft_limit == start_value - soft_value + n * 2^31
213 	 *      n >= 0
214 	 *      soft_value < 2^32, soft_value >= 0
215 	 *      for n == 0 start_value > soft_value
216 	 *
217 	 * As with compute_hard_n_value() the equations are not single-solution.
218 	 * The algorithm selects the solution that has:
219 	 *      2^30 <= soft_limit < 2^31 + 2^30
220 	 * for the interior iterations, which guarantees a large guard band
221 	 * around the counter hard limit and next interrupt.
222 	 */
223 
224 	/* Start by estimating n and compute soft_value */
225 	n = (x->lft.soft_packet_limit - attrs->lft.hard_packet_limit) / BIT_ULL(31);
226 	start_value = attrs->lft.hard_packet_limit + n * BIT_ULL(31) -
227 		      x->lft.soft_packet_limit;
228 
229 	/* Compare against constraints and adjust n */
230 	if (n < 0)
231 		n = 0;
232 	else if (start_value >= BIT_ULL(32))
233 		n -= 1;
234 	else if (start_value < 0)
235 		n += 1;
236 
237 	/* Choose the best of the two solutions: */
238 	start_value = attrs->lft.hard_packet_limit + n * BIT_ULL(31) - start_value;
239 	if (n != attrs->lft.numb_rounds_hard && start_value < BIT_ULL(30))
240 		n += 1;
241 
242 	/* Note that the upper limit of soft_value happens naturally because we
243 	 * always select the lowest soft_value.
244 	 */
245 
246 	/* Computed values solve the system of equations: */
247 	start_value = attrs->lft.hard_packet_limit + n * BIT_ULL(31) - start_value;
248 
249 	/* The best solution means: when there are multiple iterations we must
250 	 * not fall below 2^30 as that would get too close to the false
251 	 * hard_limit and when we reach an interior iteration for soft_limit it
252 	 * has to be far away from 2**32-1 which is the counter reset point
253 	 * after the +2^31 to accommodate latency.
254 	 */
255 	attrs->lft.soft_packet_limit = lower_32_bits(start_value);
256 	attrs->lft.numb_rounds_soft = (u64)n;
257 }
258 
mlx5e_ipsec_init_macs(struct mlx5e_ipsec_sa_entry * sa_entry,struct mlx5_accel_esp_xfrm_attrs * attrs)259 static void mlx5e_ipsec_init_macs(struct mlx5e_ipsec_sa_entry *sa_entry,
260 				  struct mlx5_accel_esp_xfrm_attrs *attrs)
261 {
262 	struct mlx5_core_dev *mdev = mlx5e_ipsec_sa2dev(sa_entry);
263 	struct xfrm_state *x = sa_entry->x;
264 	struct net_device *netdev;
265 	struct neighbour *n;
266 	u8 addr[ETH_ALEN];
267 	const void *pkey;
268 	u8 *dst, *src;
269 
270 	if (attrs->mode != XFRM_MODE_TUNNEL ||
271 	    attrs->type != XFRM_DEV_OFFLOAD_PACKET)
272 		return;
273 
274 	netdev = x->xso.real_dev;
275 
276 	mlx5_query_mac_address(mdev, addr);
277 	switch (attrs->dir) {
278 	case XFRM_DEV_OFFLOAD_IN:
279 		src = attrs->dmac;
280 		dst = attrs->smac;
281 		pkey = &attrs->saddr.a4;
282 		break;
283 	case XFRM_DEV_OFFLOAD_OUT:
284 		src = attrs->smac;
285 		dst = attrs->dmac;
286 		pkey = &attrs->daddr.a4;
287 		break;
288 	default:
289 		return;
290 	}
291 
292 	ether_addr_copy(src, addr);
293 	n = neigh_lookup(&arp_tbl, pkey, netdev);
294 	if (!n) {
295 		n = neigh_create(&arp_tbl, pkey, netdev);
296 		if (IS_ERR(n))
297 			return;
298 		neigh_event_send(n, NULL);
299 		attrs->drop = true;
300 	} else {
301 		neigh_ha_snapshot(addr, n, netdev);
302 		ether_addr_copy(dst, addr);
303 	}
304 	neigh_release(n);
305 }
306 
mlx5e_ipsec_build_accel_xfrm_attrs(struct mlx5e_ipsec_sa_entry * sa_entry,struct mlx5_accel_esp_xfrm_attrs * attrs)307 void mlx5e_ipsec_build_accel_xfrm_attrs(struct mlx5e_ipsec_sa_entry *sa_entry,
308 					struct mlx5_accel_esp_xfrm_attrs *attrs)
309 {
310 	struct xfrm_state *x = sa_entry->x;
311 	struct aes_gcm_keymat *aes_gcm = &attrs->aes_gcm;
312 	struct aead_geniv_ctx *geniv_ctx;
313 	struct crypto_aead *aead;
314 	unsigned int crypto_data_len, key_len;
315 	int ivsize;
316 
317 	memset(attrs, 0, sizeof(*attrs));
318 
319 	/* key */
320 	crypto_data_len = (x->aead->alg_key_len + 7) / 8;
321 	key_len = crypto_data_len - 4; /* 4 bytes salt at end */
322 
323 	memcpy(aes_gcm->aes_key, x->aead->alg_key, key_len);
324 	aes_gcm->key_len = key_len * 8;
325 
326 	/* salt and seq_iv */
327 	aead = x->data;
328 	geniv_ctx = crypto_aead_ctx(aead);
329 	ivsize = crypto_aead_ivsize(aead);
330 	memcpy(&aes_gcm->seq_iv, &geniv_ctx->salt, ivsize);
331 	memcpy(&aes_gcm->salt, x->aead->alg_key + key_len,
332 	       sizeof(aes_gcm->salt));
333 
334 	attrs->authsize = crypto_aead_authsize(aead) / 4; /* in dwords */
335 
336 	/* iv len */
337 	aes_gcm->icv_len = x->aead->alg_icv_len;
338 
339 	attrs->dir = x->xso.dir;
340 
341 	/* esn */
342 	if (x->props.flags & XFRM_STATE_ESN) {
343 		attrs->replay_esn.trigger = true;
344 		attrs->replay_esn.esn = sa_entry->esn_state.esn;
345 		attrs->replay_esn.esn_msb = sa_entry->esn_state.esn_msb;
346 		attrs->replay_esn.overlap = sa_entry->esn_state.overlap;
347 		if (attrs->dir == XFRM_DEV_OFFLOAD_OUT)
348 			goto skip_replay_window;
349 
350 		switch (x->replay_esn->replay_window) {
351 		case 32:
352 			attrs->replay_esn.replay_window =
353 				MLX5_IPSEC_ASO_REPLAY_WIN_32BIT;
354 			break;
355 		case 64:
356 			attrs->replay_esn.replay_window =
357 				MLX5_IPSEC_ASO_REPLAY_WIN_64BIT;
358 			break;
359 		case 128:
360 			attrs->replay_esn.replay_window =
361 				MLX5_IPSEC_ASO_REPLAY_WIN_128BIT;
362 			break;
363 		case 256:
364 			attrs->replay_esn.replay_window =
365 				MLX5_IPSEC_ASO_REPLAY_WIN_256BIT;
366 			break;
367 		default:
368 			WARN_ON(true);
369 			return;
370 		}
371 	}
372 
373 skip_replay_window:
374 	/* spi */
375 	attrs->spi = be32_to_cpu(x->id.spi);
376 
377 	/* source , destination ips */
378 	memcpy(&attrs->saddr, x->props.saddr.a6, sizeof(attrs->saddr));
379 	memcpy(&attrs->daddr, x->id.daddr.a6, sizeof(attrs->daddr));
380 	attrs->family = x->props.family;
381 	attrs->type = x->xso.type;
382 	attrs->reqid = x->props.reqid;
383 	attrs->upspec.dport = ntohs(x->sel.dport);
384 	attrs->upspec.dport_mask = ntohs(x->sel.dport_mask);
385 	attrs->upspec.sport = ntohs(x->sel.sport);
386 	attrs->upspec.sport_mask = ntohs(x->sel.sport_mask);
387 	attrs->upspec.proto = x->sel.proto;
388 	attrs->mode = x->props.mode;
389 
390 	mlx5e_ipsec_init_limits(sa_entry, attrs);
391 	mlx5e_ipsec_init_macs(sa_entry, attrs);
392 
393 	if (x->encap) {
394 		attrs->encap = true;
395 		attrs->sport = x->encap->encap_sport;
396 		attrs->dport = x->encap->encap_dport;
397 	}
398 }
399 
mlx5e_xfrm_validate_state(struct mlx5_core_dev * mdev,struct xfrm_state * x,struct netlink_ext_ack * extack)400 static int mlx5e_xfrm_validate_state(struct mlx5_core_dev *mdev,
401 				     struct xfrm_state *x,
402 				     struct netlink_ext_ack *extack)
403 {
404 	if (x->props.aalgo != SADB_AALG_NONE) {
405 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload authenticated xfrm states");
406 		return -EINVAL;
407 	}
408 	if (x->props.ealgo != SADB_X_EALG_AES_GCM_ICV16) {
409 		NL_SET_ERR_MSG_MOD(extack, "Only AES-GCM-ICV16 xfrm state may be offloaded");
410 		return -EINVAL;
411 	}
412 	if (x->props.calgo != SADB_X_CALG_NONE) {
413 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload compressed xfrm states");
414 		return -EINVAL;
415 	}
416 	if (x->props.flags & XFRM_STATE_ESN &&
417 	    !(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_ESN)) {
418 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload ESN xfrm states");
419 		return -EINVAL;
420 	}
421 	if (x->props.family != AF_INET &&
422 	    x->props.family != AF_INET6) {
423 		NL_SET_ERR_MSG_MOD(extack, "Only IPv4/6 xfrm states may be offloaded");
424 		return -EINVAL;
425 	}
426 	if (x->id.proto != IPPROTO_ESP) {
427 		NL_SET_ERR_MSG_MOD(extack, "Only ESP xfrm state may be offloaded");
428 		return -EINVAL;
429 	}
430 	if (x->encap) {
431 		if (!(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_ESPINUDP)) {
432 			NL_SET_ERR_MSG_MOD(extack, "Encapsulation is not supported");
433 			return -EINVAL;
434 		}
435 
436 		if (x->encap->encap_type != UDP_ENCAP_ESPINUDP) {
437 			NL_SET_ERR_MSG_MOD(extack, "Encapsulation other than UDP is not supported");
438 			return -EINVAL;
439 		}
440 
441 		if (x->xso.type != XFRM_DEV_OFFLOAD_PACKET) {
442 			NL_SET_ERR_MSG_MOD(extack, "Encapsulation is supported in packet offload mode only");
443 			return -EINVAL;
444 		}
445 
446 		if (x->props.mode != XFRM_MODE_TRANSPORT) {
447 			NL_SET_ERR_MSG_MOD(extack, "Encapsulation is supported in transport mode only");
448 			return -EINVAL;
449 		}
450 	}
451 	if (!x->aead) {
452 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states without aead");
453 		return -EINVAL;
454 	}
455 	if (x->aead->alg_icv_len != 128) {
456 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states with AEAD ICV length other than 128bit");
457 		return -EINVAL;
458 	}
459 	if ((x->aead->alg_key_len != 128 + 32) &&
460 	    (x->aead->alg_key_len != 256 + 32)) {
461 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states with AEAD key length other than 128/256 bit");
462 		return -EINVAL;
463 	}
464 	if (x->tfcpad) {
465 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states with tfc padding");
466 		return -EINVAL;
467 	}
468 	if (!x->geniv) {
469 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states without geniv");
470 		return -EINVAL;
471 	}
472 	if (strcmp(x->geniv, "seqiv")) {
473 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload xfrm states with geniv other than seqiv");
474 		return -EINVAL;
475 	}
476 
477 	if (x->sel.proto != IPPROTO_IP && x->sel.proto != IPPROTO_UDP &&
478 	    x->sel.proto != IPPROTO_TCP) {
479 		NL_SET_ERR_MSG_MOD(extack, "Device does not support upper protocol other than TCP/UDP");
480 		return -EINVAL;
481 	}
482 
483 	if (x->props.mode != XFRM_MODE_TRANSPORT && x->props.mode != XFRM_MODE_TUNNEL) {
484 		NL_SET_ERR_MSG_MOD(extack, "Only transport and tunnel xfrm states may be offloaded");
485 		return -EINVAL;
486 	}
487 
488 	switch (x->xso.type) {
489 	case XFRM_DEV_OFFLOAD_CRYPTO:
490 		if (!(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_CRYPTO)) {
491 			NL_SET_ERR_MSG_MOD(extack, "Crypto offload is not supported");
492 			return -EINVAL;
493 		}
494 
495 		break;
496 	case XFRM_DEV_OFFLOAD_PACKET:
497 		if (!(mlx5_ipsec_device_caps(mdev) &
498 		      MLX5_IPSEC_CAP_PACKET_OFFLOAD)) {
499 			NL_SET_ERR_MSG_MOD(extack, "Packet offload is not supported");
500 			return -EINVAL;
501 		}
502 
503 		if (x->props.mode == XFRM_MODE_TUNNEL &&
504 		    !(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_TUNNEL)) {
505 			NL_SET_ERR_MSG_MOD(extack, "Packet offload is not supported for tunnel mode");
506 			return -EINVAL;
507 		}
508 
509 		if (x->replay_esn && x->xso.dir == XFRM_DEV_OFFLOAD_IN &&
510 		    x->replay_esn->replay_window != 32 &&
511 		    x->replay_esn->replay_window != 64 &&
512 		    x->replay_esn->replay_window != 128 &&
513 		    x->replay_esn->replay_window != 256) {
514 			NL_SET_ERR_MSG_MOD(extack, "Unsupported replay window size");
515 			return -EINVAL;
516 		}
517 
518 		if (!x->props.reqid) {
519 			NL_SET_ERR_MSG_MOD(extack, "Cannot offload without reqid");
520 			return -EINVAL;
521 		}
522 
523 		if (x->lft.hard_byte_limit != XFRM_INF ||
524 		    x->lft.soft_byte_limit != XFRM_INF) {
525 			NL_SET_ERR_MSG_MOD(extack, "Device doesn't support limits in bytes");
526 			return -EINVAL;
527 		}
528 
529 		if (x->lft.soft_packet_limit >= x->lft.hard_packet_limit &&
530 		    x->lft.hard_packet_limit != XFRM_INF) {
531 			/* XFRM stack doesn't prevent such configuration :(. */
532 			NL_SET_ERR_MSG_MOD(extack, "Hard packet limit must be greater than soft one");
533 			return -EINVAL;
534 		}
535 
536 		if (!x->lft.soft_packet_limit || !x->lft.hard_packet_limit) {
537 			NL_SET_ERR_MSG_MOD(extack, "Soft/hard packet limits can't be 0");
538 			return -EINVAL;
539 		}
540 		break;
541 	default:
542 		NL_SET_ERR_MSG_MOD(extack, "Unsupported xfrm offload type");
543 		return -EINVAL;
544 	}
545 	return 0;
546 }
547 
mlx5e_ipsec_modify_state(struct work_struct * _work)548 static void mlx5e_ipsec_modify_state(struct work_struct *_work)
549 {
550 	struct mlx5e_ipsec_work *work =
551 		container_of(_work, struct mlx5e_ipsec_work, work);
552 	struct mlx5e_ipsec_sa_entry *sa_entry = work->sa_entry;
553 	struct mlx5_accel_esp_xfrm_attrs *attrs;
554 
555 	attrs = &((struct mlx5e_ipsec_sa_entry *)work->data)->attrs;
556 
557 	mlx5_accel_esp_modify_xfrm(sa_entry, attrs);
558 }
559 
mlx5e_ipsec_set_esn_ops(struct mlx5e_ipsec_sa_entry * sa_entry)560 static void mlx5e_ipsec_set_esn_ops(struct mlx5e_ipsec_sa_entry *sa_entry)
561 {
562 	struct xfrm_state *x = sa_entry->x;
563 
564 	if (x->xso.type != XFRM_DEV_OFFLOAD_CRYPTO ||
565 	    x->xso.dir != XFRM_DEV_OFFLOAD_OUT)
566 		return;
567 
568 	if (x->props.flags & XFRM_STATE_ESN) {
569 		sa_entry->set_iv_op = mlx5e_ipsec_set_iv_esn;
570 		return;
571 	}
572 
573 	sa_entry->set_iv_op = mlx5e_ipsec_set_iv;
574 }
575 
mlx5e_ipsec_handle_netdev_event(struct work_struct * _work)576 static void mlx5e_ipsec_handle_netdev_event(struct work_struct *_work)
577 {
578 	struct mlx5e_ipsec_work *work =
579 		container_of(_work, struct mlx5e_ipsec_work, work);
580 	struct mlx5e_ipsec_sa_entry *sa_entry = work->sa_entry;
581 	struct mlx5e_ipsec_netevent_data *data = work->data;
582 	struct mlx5_accel_esp_xfrm_attrs *attrs;
583 
584 	attrs = &sa_entry->attrs;
585 
586 	switch (attrs->dir) {
587 	case XFRM_DEV_OFFLOAD_IN:
588 		ether_addr_copy(attrs->smac, data->addr);
589 		break;
590 	case XFRM_DEV_OFFLOAD_OUT:
591 		ether_addr_copy(attrs->dmac, data->addr);
592 		break;
593 	default:
594 		WARN_ON_ONCE(true);
595 	}
596 	attrs->drop = false;
597 	mlx5e_accel_ipsec_fs_modify(sa_entry);
598 }
599 
mlx5_ipsec_create_work(struct mlx5e_ipsec_sa_entry * sa_entry)600 static int mlx5_ipsec_create_work(struct mlx5e_ipsec_sa_entry *sa_entry)
601 {
602 	struct xfrm_state *x = sa_entry->x;
603 	struct mlx5e_ipsec_work *work;
604 	void *data = NULL;
605 
606 	switch (x->xso.type) {
607 	case XFRM_DEV_OFFLOAD_CRYPTO:
608 		if (!(x->props.flags & XFRM_STATE_ESN))
609 			return 0;
610 		break;
611 	case XFRM_DEV_OFFLOAD_PACKET:
612 		if (x->props.mode != XFRM_MODE_TUNNEL)
613 			return 0;
614 		break;
615 	default:
616 		break;
617 	}
618 
619 	work = kzalloc(sizeof(*work), GFP_KERNEL);
620 	if (!work)
621 		return -ENOMEM;
622 
623 	switch (x->xso.type) {
624 	case XFRM_DEV_OFFLOAD_CRYPTO:
625 		data = kzalloc(sizeof(*sa_entry), GFP_KERNEL);
626 		if (!data)
627 			goto free_work;
628 
629 		INIT_WORK(&work->work, mlx5e_ipsec_modify_state);
630 		break;
631 	case XFRM_DEV_OFFLOAD_PACKET:
632 		data = kzalloc(sizeof(struct mlx5e_ipsec_netevent_data),
633 			       GFP_KERNEL);
634 		if (!data)
635 			goto free_work;
636 
637 		INIT_WORK(&work->work, mlx5e_ipsec_handle_netdev_event);
638 		break;
639 	default:
640 		break;
641 	}
642 
643 	work->data = data;
644 	work->sa_entry = sa_entry;
645 	sa_entry->work = work;
646 	return 0;
647 
648 free_work:
649 	kfree(work);
650 	return -ENOMEM;
651 }
652 
mlx5e_ipsec_create_dwork(struct mlx5e_ipsec_sa_entry * sa_entry)653 static int mlx5e_ipsec_create_dwork(struct mlx5e_ipsec_sa_entry *sa_entry)
654 {
655 	struct xfrm_state *x = sa_entry->x;
656 	struct mlx5e_ipsec_dwork *dwork;
657 
658 	if (x->xso.type != XFRM_DEV_OFFLOAD_PACKET)
659 		return 0;
660 
661 	if (x->xso.dir != XFRM_DEV_OFFLOAD_OUT)
662 		return 0;
663 
664 	if (x->lft.soft_packet_limit == XFRM_INF &&
665 	    x->lft.hard_packet_limit == XFRM_INF)
666 		return 0;
667 
668 	dwork = kzalloc(sizeof(*dwork), GFP_KERNEL);
669 	if (!dwork)
670 		return -ENOMEM;
671 
672 	dwork->sa_entry = sa_entry;
673 	INIT_DELAYED_WORK(&dwork->dwork, mlx5e_ipsec_handle_tx_limit);
674 	sa_entry->dwork = dwork;
675 	return 0;
676 }
677 
mlx5e_xfrm_add_state(struct xfrm_state * x,struct netlink_ext_ack * extack)678 static int mlx5e_xfrm_add_state(struct xfrm_state *x,
679 				struct netlink_ext_ack *extack)
680 {
681 	struct mlx5e_ipsec_sa_entry *sa_entry = NULL;
682 	struct net_device *netdev = x->xso.real_dev;
683 	struct mlx5e_ipsec *ipsec;
684 	struct mlx5e_priv *priv;
685 	gfp_t gfp;
686 	int err;
687 
688 	priv = netdev_priv(netdev);
689 	if (!priv->ipsec)
690 		return -EOPNOTSUPP;
691 
692 	ipsec = priv->ipsec;
693 	gfp = (x->xso.flags & XFRM_DEV_OFFLOAD_FLAG_ACQ) ? GFP_ATOMIC : GFP_KERNEL;
694 	sa_entry = kzalloc(sizeof(*sa_entry), gfp);
695 	if (!sa_entry)
696 		return -ENOMEM;
697 
698 	sa_entry->x = x;
699 	sa_entry->ipsec = ipsec;
700 	/* Check if this SA is originated from acquire flow temporary SA */
701 	if (x->xso.flags & XFRM_DEV_OFFLOAD_FLAG_ACQ)
702 		goto out;
703 
704 	err = mlx5e_xfrm_validate_state(priv->mdev, x, extack);
705 	if (err)
706 		goto err_xfrm;
707 
708 	if (!mlx5_eswitch_block_ipsec(priv->mdev)) {
709 		err = -EBUSY;
710 		goto err_xfrm;
711 	}
712 
713 	/* check esn */
714 	if (x->props.flags & XFRM_STATE_ESN)
715 		mlx5e_ipsec_update_esn_state(sa_entry);
716 
717 	mlx5e_ipsec_build_accel_xfrm_attrs(sa_entry, &sa_entry->attrs);
718 
719 	err = mlx5_ipsec_create_work(sa_entry);
720 	if (err)
721 		goto unblock_ipsec;
722 
723 	err = mlx5e_ipsec_create_dwork(sa_entry);
724 	if (err)
725 		goto release_work;
726 
727 	/* create hw context */
728 	err = mlx5_ipsec_create_sa_ctx(sa_entry);
729 	if (err)
730 		goto release_dwork;
731 
732 	err = mlx5e_accel_ipsec_fs_add_rule(sa_entry);
733 	if (err)
734 		goto err_hw_ctx;
735 
736 	if (x->props.mode == XFRM_MODE_TUNNEL &&
737 	    x->xso.type == XFRM_DEV_OFFLOAD_PACKET &&
738 	    !mlx5e_ipsec_fs_tunnel_enabled(sa_entry)) {
739 		NL_SET_ERR_MSG_MOD(extack, "Packet offload tunnel mode is disabled due to encap settings");
740 		err = -EINVAL;
741 		goto err_add_rule;
742 	}
743 
744 	/* We use *_bh() variant because xfrm_timer_handler(), which runs
745 	 * in softirq context, can reach our state delete logic and we need
746 	 * xa_erase_bh() there.
747 	 */
748 	err = xa_insert_bh(&ipsec->sadb, sa_entry->ipsec_obj_id, sa_entry,
749 			   GFP_KERNEL);
750 	if (err)
751 		goto err_add_rule;
752 
753 	mlx5e_ipsec_set_esn_ops(sa_entry);
754 
755 	if (sa_entry->dwork)
756 		queue_delayed_work(ipsec->wq, &sa_entry->dwork->dwork,
757 				   MLX5_IPSEC_RESCHED);
758 
759 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET &&
760 	    x->props.mode == XFRM_MODE_TUNNEL)
761 		xa_set_mark(&ipsec->sadb, sa_entry->ipsec_obj_id,
762 			    MLX5E_IPSEC_TUNNEL_SA);
763 
764 out:
765 	x->xso.offload_handle = (unsigned long)sa_entry;
766 	return 0;
767 
768 err_add_rule:
769 	mlx5e_accel_ipsec_fs_del_rule(sa_entry);
770 err_hw_ctx:
771 	mlx5_ipsec_free_sa_ctx(sa_entry);
772 release_dwork:
773 	kfree(sa_entry->dwork);
774 release_work:
775 	if (sa_entry->work)
776 		kfree(sa_entry->work->data);
777 	kfree(sa_entry->work);
778 unblock_ipsec:
779 	mlx5_eswitch_unblock_ipsec(priv->mdev);
780 err_xfrm:
781 	kfree(sa_entry);
782 	NL_SET_ERR_MSG_WEAK_MOD(extack, "Device failed to offload this state");
783 	return err;
784 }
785 
mlx5e_xfrm_del_state(struct xfrm_state * x)786 static void mlx5e_xfrm_del_state(struct xfrm_state *x)
787 {
788 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
789 	struct mlx5_accel_esp_xfrm_attrs *attrs = &sa_entry->attrs;
790 	struct mlx5e_ipsec *ipsec = sa_entry->ipsec;
791 	struct mlx5e_ipsec_sa_entry *old;
792 
793 	if (x->xso.flags & XFRM_DEV_OFFLOAD_FLAG_ACQ)
794 		return;
795 
796 	old = xa_erase_bh(&ipsec->sadb, sa_entry->ipsec_obj_id);
797 	WARN_ON(old != sa_entry);
798 
799 	if (attrs->mode == XFRM_MODE_TUNNEL &&
800 	    attrs->type == XFRM_DEV_OFFLOAD_PACKET)
801 		/* Make sure that no ARP requests are running in parallel */
802 		flush_workqueue(ipsec->wq);
803 
804 }
805 
mlx5e_xfrm_free_state(struct xfrm_state * x)806 static void mlx5e_xfrm_free_state(struct xfrm_state *x)
807 {
808 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
809 	struct mlx5e_ipsec *ipsec = sa_entry->ipsec;
810 
811 	if (x->xso.flags & XFRM_DEV_OFFLOAD_FLAG_ACQ)
812 		goto sa_entry_free;
813 
814 	if (sa_entry->work)
815 		cancel_work_sync(&sa_entry->work->work);
816 
817 	if (sa_entry->dwork)
818 		cancel_delayed_work_sync(&sa_entry->dwork->dwork);
819 
820 	mlx5e_accel_ipsec_fs_del_rule(sa_entry);
821 	mlx5_ipsec_free_sa_ctx(sa_entry);
822 	kfree(sa_entry->dwork);
823 	if (sa_entry->work)
824 		kfree(sa_entry->work->data);
825 	kfree(sa_entry->work);
826 	mlx5_eswitch_unblock_ipsec(ipsec->mdev);
827 sa_entry_free:
828 	kfree(sa_entry);
829 }
830 
mlx5e_ipsec_netevent_event(struct notifier_block * nb,unsigned long event,void * ptr)831 static int mlx5e_ipsec_netevent_event(struct notifier_block *nb,
832 				      unsigned long event, void *ptr)
833 {
834 	struct mlx5_accel_esp_xfrm_attrs *attrs;
835 	struct mlx5e_ipsec_netevent_data *data;
836 	struct mlx5e_ipsec_sa_entry *sa_entry;
837 	struct mlx5e_ipsec *ipsec;
838 	struct neighbour *n = ptr;
839 	struct net_device *netdev;
840 	struct xfrm_state *x;
841 	unsigned long idx;
842 
843 	if (event != NETEVENT_NEIGH_UPDATE || !(n->nud_state & NUD_VALID))
844 		return NOTIFY_DONE;
845 
846 	ipsec = container_of(nb, struct mlx5e_ipsec, netevent_nb);
847 	xa_for_each_marked(&ipsec->sadb, idx, sa_entry, MLX5E_IPSEC_TUNNEL_SA) {
848 		attrs = &sa_entry->attrs;
849 
850 		if (attrs->family == AF_INET) {
851 			if (!neigh_key_eq32(n, &attrs->saddr.a4) &&
852 			    !neigh_key_eq32(n, &attrs->daddr.a4))
853 				continue;
854 		} else {
855 			if (!neigh_key_eq128(n, &attrs->saddr.a4) &&
856 			    !neigh_key_eq128(n, &attrs->daddr.a4))
857 				continue;
858 		}
859 
860 		x = sa_entry->x;
861 		netdev = x->xso.real_dev;
862 		data = sa_entry->work->data;
863 
864 		neigh_ha_snapshot(data->addr, n, netdev);
865 		queue_work(ipsec->wq, &sa_entry->work->work);
866 	}
867 
868 	return NOTIFY_DONE;
869 }
870 
mlx5e_ipsec_init(struct mlx5e_priv * priv)871 void mlx5e_ipsec_init(struct mlx5e_priv *priv)
872 {
873 	struct mlx5e_ipsec *ipsec;
874 	int ret = -ENOMEM;
875 
876 	if (!mlx5_ipsec_device_caps(priv->mdev)) {
877 		netdev_dbg(priv->netdev, "Not an IPSec offload device\n");
878 		return;
879 	}
880 
881 	ipsec = kzalloc(sizeof(*ipsec), GFP_KERNEL);
882 	if (!ipsec)
883 		return;
884 
885 	xa_init_flags(&ipsec->sadb, XA_FLAGS_ALLOC);
886 	ipsec->mdev = priv->mdev;
887 	ipsec->wq = alloc_workqueue("mlx5e_ipsec: %s", WQ_UNBOUND, 0,
888 				    priv->netdev->name);
889 	if (!ipsec->wq)
890 		goto err_wq;
891 
892 	if (mlx5_ipsec_device_caps(priv->mdev) &
893 	    MLX5_IPSEC_CAP_PACKET_OFFLOAD) {
894 		ret = mlx5e_ipsec_aso_init(ipsec);
895 		if (ret)
896 			goto err_aso;
897 	}
898 
899 	if (mlx5_ipsec_device_caps(priv->mdev) & MLX5_IPSEC_CAP_TUNNEL) {
900 		ipsec->netevent_nb.notifier_call = mlx5e_ipsec_netevent_event;
901 		ret = register_netevent_notifier(&ipsec->netevent_nb);
902 		if (ret)
903 			goto clear_aso;
904 	}
905 
906 	ipsec->is_uplink_rep = mlx5e_is_uplink_rep(priv);
907 	ret = mlx5e_accel_ipsec_fs_init(ipsec);
908 	if (ret)
909 		goto err_fs_init;
910 
911 	ipsec->fs = priv->fs;
912 	priv->ipsec = ipsec;
913 	netdev_dbg(priv->netdev, "IPSec attached to netdevice\n");
914 	return;
915 
916 err_fs_init:
917 	if (mlx5_ipsec_device_caps(priv->mdev) & MLX5_IPSEC_CAP_TUNNEL)
918 		unregister_netevent_notifier(&ipsec->netevent_nb);
919 clear_aso:
920 	if (mlx5_ipsec_device_caps(priv->mdev) & MLX5_IPSEC_CAP_PACKET_OFFLOAD)
921 		mlx5e_ipsec_aso_cleanup(ipsec);
922 err_aso:
923 	destroy_workqueue(ipsec->wq);
924 err_wq:
925 	kfree(ipsec);
926 	mlx5_core_err(priv->mdev, "IPSec initialization failed, %d\n", ret);
927 	return;
928 }
929 
mlx5e_ipsec_cleanup(struct mlx5e_priv * priv)930 void mlx5e_ipsec_cleanup(struct mlx5e_priv *priv)
931 {
932 	struct mlx5e_ipsec *ipsec = priv->ipsec;
933 
934 	if (!ipsec)
935 		return;
936 
937 	mlx5e_accel_ipsec_fs_cleanup(ipsec);
938 	if (ipsec->netevent_nb.notifier_call) {
939 		unregister_netevent_notifier(&ipsec->netevent_nb);
940 		ipsec->netevent_nb.notifier_call = NULL;
941 	}
942 	if (ipsec->aso)
943 		mlx5e_ipsec_aso_cleanup(ipsec);
944 	destroy_workqueue(ipsec->wq);
945 	kfree(ipsec);
946 	priv->ipsec = NULL;
947 }
948 
mlx5e_ipsec_offload_ok(struct sk_buff * skb,struct xfrm_state * x)949 static bool mlx5e_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
950 {
951 	if (x->props.family == AF_INET) {
952 		/* Offload with IPv4 options is not supported yet */
953 		if (ip_hdr(skb)->ihl > 5)
954 			return false;
955 	} else {
956 		/* Offload with IPv6 extension headers is not support yet */
957 		if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr))
958 			return false;
959 	}
960 
961 	return true;
962 }
963 
mlx5e_xfrm_advance_esn_state(struct xfrm_state * x)964 static void mlx5e_xfrm_advance_esn_state(struct xfrm_state *x)
965 {
966 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
967 	struct mlx5e_ipsec_work *work = sa_entry->work;
968 	struct mlx5e_ipsec_sa_entry *sa_entry_shadow;
969 	bool need_update;
970 
971 	need_update = mlx5e_ipsec_update_esn_state(sa_entry);
972 	if (!need_update)
973 		return;
974 
975 	sa_entry_shadow = work->data;
976 	memset(sa_entry_shadow, 0x00, sizeof(*sa_entry_shadow));
977 	mlx5e_ipsec_build_accel_xfrm_attrs(sa_entry, &sa_entry_shadow->attrs);
978 	queue_work(sa_entry->ipsec->wq, &work->work);
979 }
980 
mlx5e_xfrm_update_curlft(struct xfrm_state * x)981 static void mlx5e_xfrm_update_curlft(struct xfrm_state *x)
982 {
983 	struct mlx5e_ipsec_sa_entry *sa_entry = to_ipsec_sa_entry(x);
984 	struct mlx5e_ipsec_rule *ipsec_rule = &sa_entry->ipsec_rule;
985 	u64 packets, bytes, lastuse;
986 
987 	lockdep_assert(lockdep_is_held(&x->lock) ||
988 		       lockdep_is_held(&dev_net(x->xso.real_dev)->xfrm.xfrm_cfg_mutex));
989 
990 	if (x->xso.flags & XFRM_DEV_OFFLOAD_FLAG_ACQ)
991 		return;
992 
993 	mlx5_fc_query_cached(ipsec_rule->fc, &bytes, &packets, &lastuse);
994 	x->curlft.packets += packets;
995 	x->curlft.bytes += bytes;
996 }
997 
mlx5e_xfrm_validate_policy(struct mlx5_core_dev * mdev,struct xfrm_policy * x,struct netlink_ext_ack * extack)998 static int mlx5e_xfrm_validate_policy(struct mlx5_core_dev *mdev,
999 				      struct xfrm_policy *x,
1000 				      struct netlink_ext_ack *extack)
1001 {
1002 	struct xfrm_selector *sel = &x->selector;
1003 
1004 	if (x->type != XFRM_POLICY_TYPE_MAIN) {
1005 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload non-main policy types");
1006 		return -EINVAL;
1007 	}
1008 
1009 	/* Please pay attention that we support only one template */
1010 	if (x->xfrm_nr > 1) {
1011 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload more than one template");
1012 		return -EINVAL;
1013 	}
1014 
1015 	if (x->xdo.dir != XFRM_DEV_OFFLOAD_IN &&
1016 	    x->xdo.dir != XFRM_DEV_OFFLOAD_OUT) {
1017 		NL_SET_ERR_MSG_MOD(extack, "Cannot offload forward policy");
1018 		return -EINVAL;
1019 	}
1020 
1021 	if (!x->xfrm_vec[0].reqid && sel->proto == IPPROTO_IP &&
1022 	    addr6_all_zero(sel->saddr.a6) && addr6_all_zero(sel->daddr.a6)) {
1023 		NL_SET_ERR_MSG_MOD(extack, "Unsupported policy with reqid 0 without at least one of upper protocol or ip addr(s) different than 0");
1024 		return -EINVAL;
1025 	}
1026 
1027 	if (x->xdo.type != XFRM_DEV_OFFLOAD_PACKET) {
1028 		NL_SET_ERR_MSG_MOD(extack, "Unsupported xfrm offload type");
1029 		return -EINVAL;
1030 	}
1031 
1032 	if (x->selector.proto != IPPROTO_IP &&
1033 	    x->selector.proto != IPPROTO_UDP &&
1034 	    x->selector.proto != IPPROTO_TCP) {
1035 		NL_SET_ERR_MSG_MOD(extack, "Device does not support upper protocol other than TCP/UDP");
1036 		return -EINVAL;
1037 	}
1038 
1039 	if (x->priority) {
1040 		if (!(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_PRIO)) {
1041 			NL_SET_ERR_MSG_MOD(extack, "Device does not support policy priority");
1042 			return -EINVAL;
1043 		}
1044 
1045 		if (x->priority == U32_MAX) {
1046 			NL_SET_ERR_MSG_MOD(extack, "Device does not support requested policy priority");
1047 			return -EINVAL;
1048 		}
1049 	}
1050 
1051 	if (x->xdo.type == XFRM_DEV_OFFLOAD_PACKET &&
1052 	    !(mlx5_ipsec_device_caps(mdev) & MLX5_IPSEC_CAP_PACKET_OFFLOAD)) {
1053 		NL_SET_ERR_MSG_MOD(extack, "Packet offload is not supported");
1054 		return -EINVAL;
1055 	}
1056 
1057 	return 0;
1058 }
1059 
1060 static void
mlx5e_ipsec_build_accel_pol_attrs(struct mlx5e_ipsec_pol_entry * pol_entry,struct mlx5_accel_pol_xfrm_attrs * attrs)1061 mlx5e_ipsec_build_accel_pol_attrs(struct mlx5e_ipsec_pol_entry *pol_entry,
1062 				  struct mlx5_accel_pol_xfrm_attrs *attrs)
1063 {
1064 	struct xfrm_policy *x = pol_entry->x;
1065 	struct xfrm_selector *sel;
1066 
1067 	sel = &x->selector;
1068 	memset(attrs, 0, sizeof(*attrs));
1069 
1070 	memcpy(&attrs->saddr, sel->saddr.a6, sizeof(attrs->saddr));
1071 	memcpy(&attrs->daddr, sel->daddr.a6, sizeof(attrs->daddr));
1072 	attrs->family = sel->family;
1073 	attrs->dir = x->xdo.dir;
1074 	attrs->action = x->action;
1075 	attrs->type = XFRM_DEV_OFFLOAD_PACKET;
1076 	attrs->reqid = x->xfrm_vec[0].reqid;
1077 	attrs->upspec.dport = ntohs(sel->dport);
1078 	attrs->upspec.dport_mask = ntohs(sel->dport_mask);
1079 	attrs->upspec.sport = ntohs(sel->sport);
1080 	attrs->upspec.sport_mask = ntohs(sel->sport_mask);
1081 	attrs->upspec.proto = sel->proto;
1082 	attrs->prio = x->priority;
1083 }
1084 
mlx5e_xfrm_add_policy(struct xfrm_policy * x,struct netlink_ext_ack * extack)1085 static int mlx5e_xfrm_add_policy(struct xfrm_policy *x,
1086 				 struct netlink_ext_ack *extack)
1087 {
1088 	struct net_device *netdev = x->xdo.real_dev;
1089 	struct mlx5e_ipsec_pol_entry *pol_entry;
1090 	struct mlx5e_priv *priv;
1091 	int err;
1092 
1093 	priv = netdev_priv(netdev);
1094 	if (!priv->ipsec) {
1095 		NL_SET_ERR_MSG_MOD(extack, "Device doesn't support IPsec packet offload");
1096 		return -EOPNOTSUPP;
1097 	}
1098 
1099 	err = mlx5e_xfrm_validate_policy(priv->mdev, x, extack);
1100 	if (err)
1101 		return err;
1102 
1103 	pol_entry = kzalloc(sizeof(*pol_entry), GFP_KERNEL);
1104 	if (!pol_entry)
1105 		return -ENOMEM;
1106 
1107 	pol_entry->x = x;
1108 	pol_entry->ipsec = priv->ipsec;
1109 
1110 	if (!mlx5_eswitch_block_ipsec(priv->mdev)) {
1111 		err = -EBUSY;
1112 		goto ipsec_busy;
1113 	}
1114 
1115 	mlx5e_ipsec_build_accel_pol_attrs(pol_entry, &pol_entry->attrs);
1116 	err = mlx5e_accel_ipsec_fs_add_pol(pol_entry);
1117 	if (err)
1118 		goto err_fs;
1119 
1120 	x->xdo.offload_handle = (unsigned long)pol_entry;
1121 	return 0;
1122 
1123 err_fs:
1124 	mlx5_eswitch_unblock_ipsec(priv->mdev);
1125 ipsec_busy:
1126 	kfree(pol_entry);
1127 	NL_SET_ERR_MSG_MOD(extack, "Device failed to offload this policy");
1128 	return err;
1129 }
1130 
mlx5e_xfrm_del_policy(struct xfrm_policy * x)1131 static void mlx5e_xfrm_del_policy(struct xfrm_policy *x)
1132 {
1133 	struct mlx5e_ipsec_pol_entry *pol_entry = to_ipsec_pol_entry(x);
1134 
1135 	mlx5e_accel_ipsec_fs_del_pol(pol_entry);
1136 	mlx5_eswitch_unblock_ipsec(pol_entry->ipsec->mdev);
1137 }
1138 
mlx5e_xfrm_free_policy(struct xfrm_policy * x)1139 static void mlx5e_xfrm_free_policy(struct xfrm_policy *x)
1140 {
1141 	struct mlx5e_ipsec_pol_entry *pol_entry = to_ipsec_pol_entry(x);
1142 
1143 	kfree(pol_entry);
1144 }
1145 
1146 static const struct xfrmdev_ops mlx5e_ipsec_xfrmdev_ops = {
1147 	.xdo_dev_state_add	= mlx5e_xfrm_add_state,
1148 	.xdo_dev_state_delete	= mlx5e_xfrm_del_state,
1149 	.xdo_dev_state_free	= mlx5e_xfrm_free_state,
1150 	.xdo_dev_offload_ok	= mlx5e_ipsec_offload_ok,
1151 	.xdo_dev_state_advance_esn = mlx5e_xfrm_advance_esn_state,
1152 
1153 	.xdo_dev_state_update_curlft = mlx5e_xfrm_update_curlft,
1154 	.xdo_dev_policy_add = mlx5e_xfrm_add_policy,
1155 	.xdo_dev_policy_delete = mlx5e_xfrm_del_policy,
1156 	.xdo_dev_policy_free = mlx5e_xfrm_free_policy,
1157 };
1158 
mlx5e_ipsec_build_netdev(struct mlx5e_priv * priv)1159 void mlx5e_ipsec_build_netdev(struct mlx5e_priv *priv)
1160 {
1161 	struct mlx5_core_dev *mdev = priv->mdev;
1162 	struct net_device *netdev = priv->netdev;
1163 
1164 	if (!mlx5_ipsec_device_caps(mdev))
1165 		return;
1166 
1167 	mlx5_core_info(mdev, "mlx5e: IPSec ESP acceleration enabled\n");
1168 
1169 	netdev->xfrmdev_ops = &mlx5e_ipsec_xfrmdev_ops;
1170 	netdev->features |= NETIF_F_HW_ESP;
1171 	netdev->hw_enc_features |= NETIF_F_HW_ESP;
1172 
1173 	if (!MLX5_CAP_ETH(mdev, swp_csum)) {
1174 		mlx5_core_dbg(mdev, "mlx5e: SWP checksum not supported\n");
1175 		return;
1176 	}
1177 
1178 	netdev->features |= NETIF_F_HW_ESP_TX_CSUM;
1179 	netdev->hw_enc_features |= NETIF_F_HW_ESP_TX_CSUM;
1180 
1181 	if (!MLX5_CAP_ETH(mdev, swp_lso)) {
1182 		mlx5_core_dbg(mdev, "mlx5e: ESP LSO not supported\n");
1183 		return;
1184 	}
1185 
1186 	netdev->gso_partial_features |= NETIF_F_GSO_ESP;
1187 	mlx5_core_dbg(mdev, "mlx5e: ESP GSO capability turned on\n");
1188 	netdev->features |= NETIF_F_GSO_ESP;
1189 	netdev->hw_features |= NETIF_F_GSO_ESP;
1190 	netdev->hw_enc_features |= NETIF_F_GSO_ESP;
1191 }
1192