1 /*
2 * Copyright (c) 2018 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 "en_accel/tls.h"
35 #include "en_accel/tls_rxtx.h"
36 #include "accel/accel.h"
37
38 #include <net/inet6_hashtables.h>
39 #include <linux/ipv6.h>
40
41 #define SYNDROM_DECRYPTED 0x30
42 #define SYNDROM_RESYNC_REQUEST 0x31
43 #define SYNDROM_AUTH_FAILED 0x32
44
45 #define SYNDROME_OFFLOAD_REQUIRED 32
46 #define SYNDROME_SYNC 33
47
48 struct sync_info {
49 u64 rcd_sn;
50 s32 sync_len;
51 int nr_frags;
52 skb_frag_t frags[MAX_SKB_FRAGS];
53 };
54
55 struct recv_metadata_content {
56 u8 syndrome;
57 u8 reserved;
58 __be32 sync_seq;
59 } __packed;
60
61 struct send_metadata_content {
62 /* One byte of syndrome followed by 3 bytes of swid */
63 __be32 syndrome_swid;
64 __be16 first_seq;
65 } __packed;
66
67 struct mlx5e_tls_metadata {
68 union {
69 /* from fpga to host */
70 struct recv_metadata_content recv;
71 /* from host to fpga */
72 struct send_metadata_content send;
73 unsigned char raw[6];
74 } __packed content;
75 /* packet type ID field */
76 __be16 ethertype;
77 } __packed;
78
mlx5e_tls_add_metadata(struct sk_buff * skb,__be32 swid)79 static int mlx5e_tls_add_metadata(struct sk_buff *skb, __be32 swid)
80 {
81 struct mlx5e_tls_metadata *pet;
82 struct ethhdr *eth;
83
84 if (skb_cow_head(skb, sizeof(struct mlx5e_tls_metadata)))
85 return -ENOMEM;
86
87 eth = (struct ethhdr *)skb_push(skb, sizeof(struct mlx5e_tls_metadata));
88 skb->mac_header -= sizeof(struct mlx5e_tls_metadata);
89 pet = (struct mlx5e_tls_metadata *)(eth + 1);
90
91 memmove(skb->data, skb->data + sizeof(struct mlx5e_tls_metadata),
92 2 * ETH_ALEN);
93
94 eth->h_proto = cpu_to_be16(MLX5E_METADATA_ETHER_TYPE);
95 pet->content.send.syndrome_swid =
96 htonl(SYNDROME_OFFLOAD_REQUIRED << 24) | swid;
97
98 return 0;
99 }
100
mlx5e_tls_get_sync_data(struct mlx5e_tls_offload_context_tx * context,u32 tcp_seq,struct sync_info * info)101 static int mlx5e_tls_get_sync_data(struct mlx5e_tls_offload_context_tx *context,
102 u32 tcp_seq, struct sync_info *info)
103 {
104 int remaining, i = 0, ret = -EINVAL;
105 struct tls_record_info *record;
106 unsigned long flags;
107 s32 sync_size;
108
109 spin_lock_irqsave(&context->base.lock, flags);
110 record = tls_get_record(&context->base, tcp_seq, &info->rcd_sn);
111
112 if (unlikely(!record))
113 goto out;
114
115 sync_size = tcp_seq - tls_record_start_seq(record);
116 info->sync_len = sync_size;
117 if (unlikely(sync_size < 0)) {
118 if (tls_record_is_start_marker(record))
119 goto done;
120
121 goto out;
122 }
123
124 remaining = sync_size;
125 while (remaining > 0) {
126 info->frags[i] = record->frags[i];
127 __skb_frag_ref(&info->frags[i]);
128 remaining -= skb_frag_size(&info->frags[i]);
129
130 if (remaining < 0)
131 skb_frag_size_add(&info->frags[i], remaining);
132
133 i++;
134 }
135 info->nr_frags = i;
136 done:
137 ret = 0;
138 out:
139 spin_unlock_irqrestore(&context->base.lock, flags);
140 return ret;
141 }
142
mlx5e_tls_complete_sync_skb(struct sk_buff * skb,struct sk_buff * nskb,u32 tcp_seq,int headln,__be64 rcd_sn)143 static void mlx5e_tls_complete_sync_skb(struct sk_buff *skb,
144 struct sk_buff *nskb, u32 tcp_seq,
145 int headln, __be64 rcd_sn)
146 {
147 struct mlx5e_tls_metadata *pet;
148 u8 syndrome = SYNDROME_SYNC;
149 struct iphdr *iph;
150 struct tcphdr *th;
151 int data_len, mss;
152
153 nskb->dev = skb->dev;
154 skb_reset_mac_header(nskb);
155 skb_set_network_header(nskb, skb_network_offset(skb));
156 skb_set_transport_header(nskb, skb_transport_offset(skb));
157 memcpy(nskb->data, skb->data, headln);
158 memcpy(nskb->data + headln, &rcd_sn, sizeof(rcd_sn));
159
160 iph = ip_hdr(nskb);
161 iph->tot_len = htons(nskb->len - skb_network_offset(nskb));
162 th = tcp_hdr(nskb);
163 data_len = nskb->len - headln;
164 tcp_seq -= data_len;
165 th->seq = htonl(tcp_seq);
166
167 mss = nskb->dev->mtu - (headln - skb_network_offset(nskb));
168 skb_shinfo(nskb)->gso_size = 0;
169 if (data_len > mss) {
170 skb_shinfo(nskb)->gso_size = mss;
171 skb_shinfo(nskb)->gso_segs = DIV_ROUND_UP(data_len, mss);
172 }
173 skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
174
175 pet = (struct mlx5e_tls_metadata *)(nskb->data + sizeof(struct ethhdr));
176 memcpy(pet, &syndrome, sizeof(syndrome));
177 pet->content.send.first_seq = htons(tcp_seq);
178
179 /* MLX5 devices don't care about the checksum partial start, offset
180 * and pseudo header
181 */
182 nskb->ip_summed = CHECKSUM_PARTIAL;
183
184 nskb->queue_mapping = skb->queue_mapping;
185 }
186
mlx5e_tls_handle_ooo(struct mlx5e_tls_offload_context_tx * context,struct mlx5e_txqsq * sq,struct sk_buff * skb,struct mlx5e_tls * tls)187 static bool mlx5e_tls_handle_ooo(struct mlx5e_tls_offload_context_tx *context,
188 struct mlx5e_txqsq *sq, struct sk_buff *skb,
189 struct mlx5e_tls *tls)
190 {
191 u32 tcp_seq = ntohl(tcp_hdr(skb)->seq);
192 struct sync_info info;
193 struct sk_buff *nskb;
194 int linear_len = 0;
195 int headln;
196 int i;
197
198 sq->stats->tls_ooo++;
199
200 if (mlx5e_tls_get_sync_data(context, tcp_seq, &info)) {
201 /* We might get here if a retransmission reaches the driver
202 * after the relevant record is acked.
203 * It should be safe to drop the packet in this case
204 */
205 atomic64_inc(&tls->sw_stats.tx_tls_drop_no_sync_data);
206 goto err_out;
207 }
208
209 if (unlikely(info.sync_len < 0)) {
210 u32 payload;
211
212 headln = skb_transport_offset(skb) + tcp_hdrlen(skb);
213 payload = skb->len - headln;
214 if (likely(payload <= -info.sync_len))
215 /* SKB payload doesn't require offload
216 */
217 return true;
218
219 atomic64_inc(&tls->sw_stats.tx_tls_drop_bypass_required);
220 goto err_out;
221 }
222
223 if (unlikely(mlx5e_tls_add_metadata(skb, context->swid))) {
224 atomic64_inc(&tls->sw_stats.tx_tls_drop_metadata);
225 goto err_out;
226 }
227
228 headln = skb_transport_offset(skb) + tcp_hdrlen(skb);
229 linear_len += headln + sizeof(info.rcd_sn);
230 nskb = alloc_skb(linear_len, GFP_ATOMIC);
231 if (unlikely(!nskb)) {
232 atomic64_inc(&tls->sw_stats.tx_tls_drop_resync_alloc);
233 goto err_out;
234 }
235
236 context->expected_seq = tcp_seq + skb->len - headln;
237 skb_put(nskb, linear_len);
238 for (i = 0; i < info.nr_frags; i++)
239 skb_shinfo(nskb)->frags[i] = info.frags[i];
240
241 skb_shinfo(nskb)->nr_frags = info.nr_frags;
242 nskb->data_len = info.sync_len;
243 nskb->len += info.sync_len;
244 sq->stats->tls_resync_bytes += nskb->len;
245 mlx5e_tls_complete_sync_skb(skb, nskb, tcp_seq, headln,
246 cpu_to_be64(info.rcd_sn));
247 mlx5e_sq_xmit_simple(sq, nskb, true);
248
249 return true;
250
251 err_out:
252 dev_kfree_skb_any(skb);
253 return false;
254 }
255
mlx5e_tls_handle_tx_skb(struct net_device * netdev,struct mlx5e_txqsq * sq,struct sk_buff * skb,struct mlx5e_accel_tx_tls_state * state)256 bool mlx5e_tls_handle_tx_skb(struct net_device *netdev, struct mlx5e_txqsq *sq,
257 struct sk_buff *skb, struct mlx5e_accel_tx_tls_state *state)
258 {
259 struct mlx5e_priv *priv = netdev_priv(netdev);
260 struct mlx5e_tls_offload_context_tx *context;
261 struct tls_context *tls_ctx;
262 u32 expected_seq;
263 int datalen;
264 u32 skb_seq;
265
266 datalen = skb->len - (skb_transport_offset(skb) + tcp_hdrlen(skb));
267 if (!datalen)
268 return true;
269
270 mlx5e_tx_mpwqe_ensure_complete(sq);
271
272 tls_ctx = tls_get_ctx(skb->sk);
273 if (WARN_ON_ONCE(tls_ctx->netdev != netdev))
274 goto err_out;
275
276 if (mlx5e_accel_is_ktls_tx(sq->mdev))
277 return mlx5e_ktls_handle_tx_skb(tls_ctx, sq, skb, datalen, state);
278
279 /* FPGA */
280 skb_seq = ntohl(tcp_hdr(skb)->seq);
281 context = mlx5e_get_tls_tx_context(tls_ctx);
282 expected_seq = context->expected_seq;
283
284 if (unlikely(expected_seq != skb_seq))
285 return mlx5e_tls_handle_ooo(context, sq, skb, priv->tls);
286
287 if (unlikely(mlx5e_tls_add_metadata(skb, context->swid))) {
288 atomic64_inc(&priv->tls->sw_stats.tx_tls_drop_metadata);
289 dev_kfree_skb_any(skb);
290 return false;
291 }
292
293 context->expected_seq = skb_seq + datalen;
294 return true;
295
296 err_out:
297 dev_kfree_skb_any(skb);
298 return false;
299 }
300
tls_update_resync_sn(struct net_device * netdev,struct sk_buff * skb,struct mlx5e_tls_metadata * mdata)301 static int tls_update_resync_sn(struct net_device *netdev,
302 struct sk_buff *skb,
303 struct mlx5e_tls_metadata *mdata)
304 {
305 struct sock *sk = NULL;
306 struct iphdr *iph;
307 struct tcphdr *th;
308 __be32 seq;
309
310 if (mdata->ethertype != htons(ETH_P_IP))
311 return -EINVAL;
312
313 iph = (struct iphdr *)(mdata + 1);
314
315 th = ((void *)iph) + iph->ihl * 4;
316
317 if (iph->version == 4) {
318 sk = inet_lookup_established(dev_net(netdev), &tcp_hashinfo,
319 iph->saddr, th->source, iph->daddr,
320 th->dest, netdev->ifindex);
321 #if IS_ENABLED(CONFIG_IPV6)
322 } else {
323 struct ipv6hdr *ipv6h = (struct ipv6hdr *)iph;
324
325 sk = __inet6_lookup_established(dev_net(netdev), &tcp_hashinfo,
326 &ipv6h->saddr, th->source,
327 &ipv6h->daddr, ntohs(th->dest),
328 netdev->ifindex, 0);
329 #endif
330 }
331 if (!sk || sk->sk_state == TCP_TIME_WAIT) {
332 struct mlx5e_priv *priv = netdev_priv(netdev);
333
334 atomic64_inc(&priv->tls->sw_stats.rx_tls_drop_resync_request);
335 goto out;
336 }
337
338 skb->sk = sk;
339 skb->destructor = sock_edemux;
340
341 memcpy(&seq, &mdata->content.recv.sync_seq, sizeof(seq));
342 tls_offload_rx_resync_request(sk, seq);
343 out:
344 return 0;
345 }
346
347 /* FPGA tls rx handler */
mlx5e_tls_handle_rx_skb_metadata(struct mlx5e_rq * rq,struct sk_buff * skb,u32 * cqe_bcnt)348 void mlx5e_tls_handle_rx_skb_metadata(struct mlx5e_rq *rq, struct sk_buff *skb,
349 u32 *cqe_bcnt)
350 {
351 struct mlx5e_tls_metadata *mdata;
352 struct mlx5e_priv *priv;
353
354 /* Use the metadata */
355 mdata = (struct mlx5e_tls_metadata *)(skb->data + ETH_HLEN);
356 switch (mdata->content.recv.syndrome) {
357 case SYNDROM_DECRYPTED:
358 skb->decrypted = 1;
359 break;
360 case SYNDROM_RESYNC_REQUEST:
361 tls_update_resync_sn(rq->netdev, skb, mdata);
362 priv = netdev_priv(rq->netdev);
363 atomic64_inc(&priv->tls->sw_stats.rx_tls_resync_request);
364 break;
365 case SYNDROM_AUTH_FAILED:
366 /* Authentication failure will be observed and verified by kTLS */
367 priv = netdev_priv(rq->netdev);
368 atomic64_inc(&priv->tls->sw_stats.rx_tls_auth_fail);
369 break;
370 default:
371 /* Bypass the metadata header to others */
372 return;
373 }
374
375 remove_metadata_hdr(skb);
376 *cqe_bcnt -= MLX5E_METADATA_ETHER_LEN;
377 }
378
mlx5e_tls_get_stop_room(struct mlx5_core_dev * mdev,struct mlx5e_params * params)379 u16 mlx5e_tls_get_stop_room(struct mlx5_core_dev *mdev, struct mlx5e_params *params)
380 {
381 if (!mlx5e_accel_is_tls_device(mdev))
382 return 0;
383
384 if (mlx5e_accel_is_ktls_device(mdev))
385 return mlx5e_ktls_get_stop_room(mdev, params);
386
387 /* FPGA */
388 /* Resync SKB. */
389 return mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
390 }
391