1 /*
2  * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3  * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #ifndef _TLS_OFFLOAD_H
35 #define _TLS_OFFLOAD_H
36 
37 #include <linux/types.h>
38 #include <asm/byteorder.h>
39 #include <linux/crypto.h>
40 #include <linux/socket.h>
41 #include <linux/tcp.h>
42 #include <linux/mutex.h>
43 #include <linux/netdevice.h>
44 #include <linux/rcupdate.h>
45 
46 #include <net/net_namespace.h>
47 #include <net/tcp.h>
48 #include <net/strparser.h>
49 #include <crypto/aead.h>
50 #include <uapi/linux/tls.h>
51 
52 struct tls_rec;
53 
54 /* Maximum data size carried in a TLS record */
55 #define TLS_MAX_PAYLOAD_SIZE		((size_t)1 << 14)
56 
57 #define TLS_HEADER_SIZE			5
58 #define TLS_NONCE_OFFSET		TLS_HEADER_SIZE
59 
60 #define TLS_CRYPTO_INFO_READY(info)	((info)->cipher_type)
61 
62 #define TLS_AAD_SPACE_SIZE		13
63 
64 #define MAX_IV_SIZE			16
65 #define TLS_TAG_SIZE			16
66 #define TLS_MAX_REC_SEQ_SIZE		8
67 #define TLS_MAX_AAD_SIZE		TLS_AAD_SPACE_SIZE
68 
69 /* For CCM mode, the full 16-bytes of IV is made of '4' fields of given sizes.
70  *
71  * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
72  *
73  * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
74  * Hence b0 contains (3 - 1) = 2.
75  */
76 #define TLS_AES_CCM_IV_B0_BYTE		2
77 #define TLS_SM4_CCM_IV_B0_BYTE		2
78 
79 enum {
80 	TLS_BASE,
81 	TLS_SW,
82 	TLS_HW,
83 	TLS_HW_RECORD,
84 	TLS_NUM_CONFIG,
85 };
86 
87 struct tx_work {
88 	struct delayed_work work;
89 	struct sock *sk;
90 };
91 
92 struct tls_sw_context_tx {
93 	struct crypto_aead *aead_send;
94 	struct crypto_wait async_wait;
95 	struct tx_work tx_work;
96 	struct tls_rec *open_rec;
97 	struct list_head tx_list;
98 	atomic_t encrypt_pending;
99 	u8 async_capable:1;
100 
101 #define BIT_TX_SCHEDULED	0
102 #define BIT_TX_CLOSING		1
103 	unsigned long tx_bitmask;
104 };
105 
106 struct tls_strparser {
107 	struct sock *sk;
108 
109 	u32 mark : 8;
110 	u32 stopped : 1;
111 	u32 copy_mode : 1;
112 	u32 mixed_decrypted : 1;
113 	u32 msg_ready : 1;
114 
115 	struct strp_msg stm;
116 
117 	struct sk_buff *anchor;
118 	struct work_struct work;
119 };
120 
121 struct tls_sw_context_rx {
122 	struct crypto_aead *aead_recv;
123 	struct crypto_wait async_wait;
124 	struct sk_buff_head rx_list;	/* list of decrypted 'data' records */
125 	void (*saved_data_ready)(struct sock *sk);
126 
127 	u8 reader_present;
128 	u8 async_capable:1;
129 	u8 zc_capable:1;
130 	u8 reader_contended:1;
131 
132 	struct tls_strparser strp;
133 
134 	atomic_t decrypt_pending;
135 	struct sk_buff_head async_hold;
136 	struct wait_queue_head wq;
137 };
138 
139 struct tls_record_info {
140 	struct list_head list;
141 	u32 end_seq;
142 	int len;
143 	int num_frags;
144 	skb_frag_t frags[MAX_SKB_FRAGS];
145 };
146 
147 struct tls_offload_context_tx {
148 	struct crypto_aead *aead_send;
149 	spinlock_t lock;	/* protects records list */
150 	struct list_head records_list;
151 	struct tls_record_info *open_record;
152 	struct tls_record_info *retransmit_hint;
153 	u64 hint_record_sn;
154 	u64 unacked_record_sn;
155 
156 	struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
157 	void (*sk_destruct)(struct sock *sk);
158 	struct work_struct destruct_work;
159 	struct tls_context *ctx;
160 	u8 driver_state[] __aligned(8);
161 	/* The TLS layer reserves room for driver specific state
162 	 * Currently the belief is that there is not enough
163 	 * driver specific state to justify another layer of indirection
164 	 */
165 #define TLS_DRIVER_STATE_SIZE_TX	16
166 };
167 
168 #define TLS_OFFLOAD_CONTEXT_SIZE_TX                                            \
169 	(sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX)
170 
171 enum tls_context_flags {
172 	/* tls_device_down was called after the netdev went down, device state
173 	 * was released, and kTLS works in software, even though rx_conf is
174 	 * still TLS_HW (needed for transition).
175 	 */
176 	TLS_RX_DEV_DEGRADED = 0,
177 	/* Unlike RX where resync is driven entirely by the core in TX only
178 	 * the driver knows when things went out of sync, so we need the flag
179 	 * to be atomic.
180 	 */
181 	TLS_TX_SYNC_SCHED = 1,
182 	/* tls_dev_del was called for the RX side, device state was released,
183 	 * but tls_ctx->netdev might still be kept, because TX-side driver
184 	 * resources might not be released yet. Used to prevent the second
185 	 * tls_dev_del call in tls_device_down if it happens simultaneously.
186 	 */
187 	TLS_RX_DEV_CLOSED = 2,
188 };
189 
190 struct cipher_context {
191 	char *iv;
192 	char *rec_seq;
193 };
194 
195 union tls_crypto_context {
196 	struct tls_crypto_info info;
197 	union {
198 		struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
199 		struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
200 		struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305;
201 		struct tls12_crypto_info_sm4_gcm sm4_gcm;
202 		struct tls12_crypto_info_sm4_ccm sm4_ccm;
203 	};
204 };
205 
206 struct tls_prot_info {
207 	u16 version;
208 	u16 cipher_type;
209 	u16 prepend_size;
210 	u16 tag_size;
211 	u16 overhead_size;
212 	u16 iv_size;
213 	u16 salt_size;
214 	u16 rec_seq_size;
215 	u16 aad_size;
216 	u16 tail_size;
217 };
218 
219 struct tls_context {
220 	/* read-only cache line */
221 	struct tls_prot_info prot_info;
222 
223 	u8 tx_conf:3;
224 	u8 rx_conf:3;
225 	u8 zerocopy_sendfile:1;
226 	u8 rx_no_pad:1;
227 
228 	int (*push_pending_record)(struct sock *sk, int flags);
229 	void (*sk_write_space)(struct sock *sk);
230 
231 	void *priv_ctx_tx;
232 	void *priv_ctx_rx;
233 
234 	struct net_device __rcu *netdev;
235 
236 	/* rw cache line */
237 	struct cipher_context tx;
238 	struct cipher_context rx;
239 
240 	struct scatterlist *partially_sent_record;
241 	u16 partially_sent_offset;
242 
243 	bool splicing_pages;
244 	bool pending_open_record_frags;
245 
246 	struct mutex tx_lock; /* protects partially_sent_* fields and
247 			       * per-type TX fields
248 			       */
249 	unsigned long flags;
250 
251 	/* cache cold stuff */
252 	struct proto *sk_proto;
253 	struct sock *sk;
254 
255 	void (*sk_destruct)(struct sock *sk);
256 
257 	union tls_crypto_context crypto_send;
258 	union tls_crypto_context crypto_recv;
259 
260 	struct list_head list;
261 	refcount_t refcount;
262 	struct rcu_head rcu;
263 };
264 
265 enum tls_offload_ctx_dir {
266 	TLS_OFFLOAD_CTX_DIR_RX,
267 	TLS_OFFLOAD_CTX_DIR_TX,
268 };
269 
270 struct tlsdev_ops {
271 	int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
272 			   enum tls_offload_ctx_dir direction,
273 			   struct tls_crypto_info *crypto_info,
274 			   u32 start_offload_tcp_sn);
275 	void (*tls_dev_del)(struct net_device *netdev,
276 			    struct tls_context *ctx,
277 			    enum tls_offload_ctx_dir direction);
278 	int (*tls_dev_resync)(struct net_device *netdev,
279 			      struct sock *sk, u32 seq, u8 *rcd_sn,
280 			      enum tls_offload_ctx_dir direction);
281 };
282 
283 enum tls_offload_sync_type {
284 	TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
285 	TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
286 	TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2,
287 };
288 
289 #define TLS_DEVICE_RESYNC_NH_START_IVAL		2
290 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL		128
291 
292 #define TLS_DEVICE_RESYNC_ASYNC_LOGMAX		13
293 struct tls_offload_resync_async {
294 	atomic64_t req;
295 	u16 loglen;
296 	u16 rcd_delta;
297 	u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX];
298 };
299 
300 struct tls_offload_context_rx {
301 	/* sw must be the first member of tls_offload_context_rx */
302 	struct tls_sw_context_rx sw;
303 	enum tls_offload_sync_type resync_type;
304 	/* this member is set regardless of resync_type, to avoid branches */
305 	u8 resync_nh_reset:1;
306 	/* CORE_NEXT_HINT-only member, but use the hole here */
307 	u8 resync_nh_do_now:1;
308 	union {
309 		/* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
310 		struct {
311 			atomic64_t resync_req;
312 		};
313 		/* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
314 		struct {
315 			u32 decrypted_failed;
316 			u32 decrypted_tgt;
317 		} resync_nh;
318 		/* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC */
319 		struct {
320 			struct tls_offload_resync_async *resync_async;
321 		};
322 	};
323 	u8 driver_state[] __aligned(8);
324 	/* The TLS layer reserves room for driver specific state
325 	 * Currently the belief is that there is not enough
326 	 * driver specific state to justify another layer of indirection
327 	 */
328 #define TLS_DRIVER_STATE_SIZE_RX	8
329 };
330 
331 #define TLS_OFFLOAD_CONTEXT_SIZE_RX					\
332 	(sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX)
333 
334 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
335 				       u32 seq, u64 *p_record_sn);
336 
tls_record_is_start_marker(struct tls_record_info * rec)337 static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
338 {
339 	return rec->len == 0;
340 }
341 
tls_record_start_seq(struct tls_record_info * rec)342 static inline u32 tls_record_start_seq(struct tls_record_info *rec)
343 {
344 	return rec->end_seq - rec->len;
345 }
346 
347 struct sk_buff *
348 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
349 		      struct sk_buff *skb);
350 struct sk_buff *
351 tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev,
352 			 struct sk_buff *skb);
353 
tls_is_skb_tx_device_offloaded(const struct sk_buff * skb)354 static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb)
355 {
356 #ifdef CONFIG_TLS_DEVICE
357 	struct sock *sk = skb->sk;
358 
359 	return sk && sk_fullsock(sk) &&
360 	       (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
361 	       &tls_validate_xmit_skb);
362 #else
363 	return false;
364 #endif
365 }
366 
tls_get_ctx(const struct sock * sk)367 static inline struct tls_context *tls_get_ctx(const struct sock *sk)
368 {
369 	struct inet_connection_sock *icsk = inet_csk(sk);
370 
371 	/* Use RCU on icsk_ulp_data only for sock diag code,
372 	 * TLS data path doesn't need rcu_dereference().
373 	 */
374 	return (__force void *)icsk->icsk_ulp_data;
375 }
376 
tls_sw_ctx_rx(const struct tls_context * tls_ctx)377 static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
378 		const struct tls_context *tls_ctx)
379 {
380 	return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
381 }
382 
tls_sw_ctx_tx(const struct tls_context * tls_ctx)383 static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
384 		const struct tls_context *tls_ctx)
385 {
386 	return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
387 }
388 
389 static inline struct tls_offload_context_tx *
tls_offload_ctx_tx(const struct tls_context * tls_ctx)390 tls_offload_ctx_tx(const struct tls_context *tls_ctx)
391 {
392 	return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
393 }
394 
tls_sw_has_ctx_tx(const struct sock * sk)395 static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
396 {
397 	struct tls_context *ctx = tls_get_ctx(sk);
398 
399 	if (!ctx)
400 		return false;
401 	return !!tls_sw_ctx_tx(ctx);
402 }
403 
tls_sw_has_ctx_rx(const struct sock * sk)404 static inline bool tls_sw_has_ctx_rx(const struct sock *sk)
405 {
406 	struct tls_context *ctx = tls_get_ctx(sk);
407 
408 	if (!ctx)
409 		return false;
410 	return !!tls_sw_ctx_rx(ctx);
411 }
412 
413 static inline struct tls_offload_context_rx *
tls_offload_ctx_rx(const struct tls_context * tls_ctx)414 tls_offload_ctx_rx(const struct tls_context *tls_ctx)
415 {
416 	return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
417 }
418 
__tls_driver_ctx(struct tls_context * tls_ctx,enum tls_offload_ctx_dir direction)419 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
420 				     enum tls_offload_ctx_dir direction)
421 {
422 	if (direction == TLS_OFFLOAD_CTX_DIR_TX)
423 		return tls_offload_ctx_tx(tls_ctx)->driver_state;
424 	else
425 		return tls_offload_ctx_rx(tls_ctx)->driver_state;
426 }
427 
428 static inline void *
tls_driver_ctx(const struct sock * sk,enum tls_offload_ctx_dir direction)429 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
430 {
431 	return __tls_driver_ctx(tls_get_ctx(sk), direction);
432 }
433 
434 #define RESYNC_REQ BIT(0)
435 #define RESYNC_REQ_ASYNC BIT(1)
436 /* The TLS context is valid until sk_destruct is called */
tls_offload_rx_resync_request(struct sock * sk,__be32 seq)437 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
438 {
439 	struct tls_context *tls_ctx = tls_get_ctx(sk);
440 	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
441 
442 	atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ);
443 }
444 
445 /* Log all TLS record header TCP sequences in [seq, seq+len] */
446 static inline void
tls_offload_rx_resync_async_request_start(struct sock * sk,__be32 seq,u16 len)447 tls_offload_rx_resync_async_request_start(struct sock *sk, __be32 seq, u16 len)
448 {
449 	struct tls_context *tls_ctx = tls_get_ctx(sk);
450 	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
451 
452 	atomic64_set(&rx_ctx->resync_async->req, ((u64)ntohl(seq) << 32) |
453 		     ((u64)len << 16) | RESYNC_REQ | RESYNC_REQ_ASYNC);
454 	rx_ctx->resync_async->loglen = 0;
455 	rx_ctx->resync_async->rcd_delta = 0;
456 }
457 
458 static inline void
tls_offload_rx_resync_async_request_end(struct sock * sk,__be32 seq)459 tls_offload_rx_resync_async_request_end(struct sock *sk, __be32 seq)
460 {
461 	struct tls_context *tls_ctx = tls_get_ctx(sk);
462 	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
463 
464 	atomic64_set(&rx_ctx->resync_async->req,
465 		     ((u64)ntohl(seq) << 32) | RESYNC_REQ);
466 }
467 
468 static inline void
tls_offload_rx_resync_set_type(struct sock * sk,enum tls_offload_sync_type type)469 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
470 {
471 	struct tls_context *tls_ctx = tls_get_ctx(sk);
472 
473 	tls_offload_ctx_rx(tls_ctx)->resync_type = type;
474 }
475 
476 /* Driver's seq tracking has to be disabled until resync succeeded */
tls_offload_tx_resync_pending(struct sock * sk)477 static inline bool tls_offload_tx_resync_pending(struct sock *sk)
478 {
479 	struct tls_context *tls_ctx = tls_get_ctx(sk);
480 	bool ret;
481 
482 	ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
483 	smp_mb__after_atomic();
484 	return ret;
485 }
486 
487 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
488 
489 #ifdef CONFIG_TLS_DEVICE
490 void tls_device_sk_destruct(struct sock *sk);
491 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq);
492 
tls_is_sk_rx_device_offloaded(struct sock * sk)493 static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk)
494 {
495 	if (!sk_fullsock(sk) ||
496 	    smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct)
497 		return false;
498 	return tls_get_ctx(sk)->rx_conf == TLS_HW;
499 }
500 #endif
501 #endif /* _TLS_OFFLOAD_H */
502