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
52struct 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 TLS_MAX_IV_SIZE			16
65#define TLS_MAX_SALT_SIZE		4
66#define TLS_TAG_SIZE			16
67#define TLS_MAX_REC_SEQ_SIZE		8
68#define TLS_MAX_AAD_SIZE		TLS_AAD_SPACE_SIZE
69
70/* For CCM mode, the full 16-bytes of IV is made of '4' fields of given sizes.
71 *
72 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
73 *
74 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
75 * Hence b0 contains (3 - 1) = 2.
76 */
77#define TLS_AES_CCM_IV_B0_BYTE		2
78#define TLS_SM4_CCM_IV_B0_BYTE		2
79
80enum {
81	TLS_BASE,
82	TLS_SW,
83	TLS_HW,
84	TLS_HW_RECORD,
85	TLS_NUM_CONFIG,
86};
87
88struct tx_work {
89	struct delayed_work work;
90	struct sock *sk;
91};
92
93struct tls_sw_context_tx {
94	struct crypto_aead *aead_send;
95	struct crypto_wait async_wait;
96	struct tx_work tx_work;
97	struct tls_rec *open_rec;
98	struct list_head tx_list;
99	atomic_t encrypt_pending;
100	u8 async_capable:1;
101
102#define BIT_TX_SCHEDULED	0
103#define BIT_TX_CLOSING		1
104	unsigned long tx_bitmask;
105};
106
107struct tls_strparser {
108	struct sock *sk;
109
110	u32 mark : 8;
111	u32 stopped : 1;
112	u32 copy_mode : 1;
113	u32 mixed_decrypted : 1;
114
115	bool msg_ready;
116
117	struct strp_msg stm;
118
119	struct sk_buff *anchor;
120	struct work_struct work;
121};
122
123struct tls_sw_context_rx {
124	struct crypto_aead *aead_recv;
125	struct crypto_wait async_wait;
126	struct sk_buff_head rx_list;	/* list of decrypted 'data' records */
127	void (*saved_data_ready)(struct sock *sk);
128
129	u8 reader_present;
130	u8 async_capable:1;
131	u8 zc_capable:1;
132	u8 reader_contended:1;
133
134	struct tls_strparser strp;
135
136	atomic_t decrypt_pending;
137	struct sk_buff_head async_hold;
138	struct wait_queue_head wq;
139};
140
141struct tls_record_info {
142	struct list_head list;
143	u32 end_seq;
144	int len;
145	int num_frags;
146	skb_frag_t frags[MAX_SKB_FRAGS];
147};
148
149#define TLS_DRIVER_STATE_SIZE_TX	16
150struct tls_offload_context_tx {
151	struct crypto_aead *aead_send;
152	spinlock_t lock;	/* protects records list */
153	struct list_head records_list;
154	struct tls_record_info *open_record;
155	struct tls_record_info *retransmit_hint;
156	u64 hint_record_sn;
157	u64 unacked_record_sn;
158
159	struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
160	void (*sk_destruct)(struct sock *sk);
161	struct work_struct destruct_work;
162	struct tls_context *ctx;
163	/* The TLS layer reserves room for driver specific state
164	 * Currently the belief is that there is not enough
165	 * driver specific state to justify another layer of indirection
166	 */
167	u8 driver_state[TLS_DRIVER_STATE_SIZE_TX] __aligned(8);
168};
169
170enum tls_context_flags {
171	/* tls_device_down was called after the netdev went down, device state
172	 * was released, and kTLS works in software, even though rx_conf is
173	 * still TLS_HW (needed for transition).
174	 */
175	TLS_RX_DEV_DEGRADED = 0,
176	/* Unlike RX where resync is driven entirely by the core in TX only
177	 * the driver knows when things went out of sync, so we need the flag
178	 * to be atomic.
179	 */
180	TLS_TX_SYNC_SCHED = 1,
181	/* tls_dev_del was called for the RX side, device state was released,
182	 * but tls_ctx->netdev might still be kept, because TX-side driver
183	 * resources might not be released yet. Used to prevent the second
184	 * tls_dev_del call in tls_device_down if it happens simultaneously.
185	 */
186	TLS_RX_DEV_CLOSED = 2,
187};
188
189struct cipher_context {
190	char iv[TLS_MAX_IV_SIZE + TLS_MAX_SALT_SIZE];
191	char rec_seq[TLS_MAX_REC_SEQ_SIZE];
192};
193
194union tls_crypto_context {
195	struct tls_crypto_info info;
196	union {
197		struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
198		struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
199		struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305;
200		struct tls12_crypto_info_sm4_gcm sm4_gcm;
201		struct tls12_crypto_info_sm4_ccm sm4_ccm;
202	};
203};
204
205struct tls_prot_info {
206	u16 version;
207	u16 cipher_type;
208	u16 prepend_size;
209	u16 tag_size;
210	u16 overhead_size;
211	u16 iv_size;
212	u16 salt_size;
213	u16 rec_seq_size;
214	u16 aad_size;
215	u16 tail_size;
216};
217
218struct tls_context {
219	/* read-only cache line */
220	struct tls_prot_info prot_info;
221
222	u8 tx_conf:3;
223	u8 rx_conf:3;
224	u8 zerocopy_sendfile:1;
225	u8 rx_no_pad:1;
226
227	int (*push_pending_record)(struct sock *sk, int flags);
228	void (*sk_write_space)(struct sock *sk);
229
230	void *priv_ctx_tx;
231	void *priv_ctx_rx;
232
233	struct net_device __rcu *netdev;
234
235	/* rw cache line */
236	struct cipher_context tx;
237	struct cipher_context rx;
238
239	struct scatterlist *partially_sent_record;
240	u16 partially_sent_offset;
241
242	bool splicing_pages;
243	bool pending_open_record_frags;
244
245	struct mutex tx_lock; /* protects partially_sent_* fields and
246			       * per-type TX fields
247			       */
248	unsigned long flags;
249
250	/* cache cold stuff */
251	struct proto *sk_proto;
252	struct sock *sk;
253
254	void (*sk_destruct)(struct sock *sk);
255
256	union tls_crypto_context crypto_send;
257	union tls_crypto_context crypto_recv;
258
259	struct list_head list;
260	refcount_t refcount;
261	struct rcu_head rcu;
262};
263
264enum tls_offload_ctx_dir {
265	TLS_OFFLOAD_CTX_DIR_RX,
266	TLS_OFFLOAD_CTX_DIR_TX,
267};
268
269struct tlsdev_ops {
270	int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
271			   enum tls_offload_ctx_dir direction,
272			   struct tls_crypto_info *crypto_info,
273			   u32 start_offload_tcp_sn);
274	void (*tls_dev_del)(struct net_device *netdev,
275			    struct tls_context *ctx,
276			    enum tls_offload_ctx_dir direction);
277	int (*tls_dev_resync)(struct net_device *netdev,
278			      struct sock *sk, u32 seq, u8 *rcd_sn,
279			      enum tls_offload_ctx_dir direction);
280};
281
282enum tls_offload_sync_type {
283	TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
284	TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
285	TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2,
286};
287
288#define TLS_DEVICE_RESYNC_NH_START_IVAL		2
289#define TLS_DEVICE_RESYNC_NH_MAX_IVAL		128
290
291#define TLS_DEVICE_RESYNC_ASYNC_LOGMAX		13
292struct tls_offload_resync_async {
293	atomic64_t req;
294	u16 loglen;
295	u16 rcd_delta;
296	u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX];
297};
298
299#define TLS_DRIVER_STATE_SIZE_RX	8
300struct 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	/* The TLS layer reserves room for driver specific state
324	 * Currently the belief is that there is not enough
325	 * driver specific state to justify another layer of indirection
326	 */
327	u8 driver_state[TLS_DRIVER_STATE_SIZE_RX] __aligned(8);
328};
329
330struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
331				       u32 seq, u64 *p_record_sn);
332
333static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
334{
335	return rec->len == 0;
336}
337
338static inline u32 tls_record_start_seq(struct tls_record_info *rec)
339{
340	return rec->end_seq - rec->len;
341}
342
343struct sk_buff *
344tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
345		      struct sk_buff *skb);
346struct sk_buff *
347tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev,
348			 struct sk_buff *skb);
349
350static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb)
351{
352#ifdef CONFIG_TLS_DEVICE
353	struct sock *sk = skb->sk;
354
355	return sk && sk_fullsock(sk) &&
356	       (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
357	       &tls_validate_xmit_skb);
358#else
359	return false;
360#endif
361}
362
363static inline struct tls_context *tls_get_ctx(const struct sock *sk)
364{
365	struct inet_connection_sock *icsk = inet_csk(sk);
366
367	/* Use RCU on icsk_ulp_data only for sock diag code,
368	 * TLS data path doesn't need rcu_dereference().
369	 */
370	return (__force void *)icsk->icsk_ulp_data;
371}
372
373static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
374		const struct tls_context *tls_ctx)
375{
376	return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
377}
378
379static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
380		const struct tls_context *tls_ctx)
381{
382	return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
383}
384
385static inline struct tls_offload_context_tx *
386tls_offload_ctx_tx(const struct tls_context *tls_ctx)
387{
388	return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
389}
390
391static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
392{
393	struct tls_context *ctx = tls_get_ctx(sk);
394
395	if (!ctx)
396		return false;
397	return !!tls_sw_ctx_tx(ctx);
398}
399
400static inline bool tls_sw_has_ctx_rx(const struct sock *sk)
401{
402	struct tls_context *ctx = tls_get_ctx(sk);
403
404	if (!ctx)
405		return false;
406	return !!tls_sw_ctx_rx(ctx);
407}
408
409static inline struct tls_offload_context_rx *
410tls_offload_ctx_rx(const struct tls_context *tls_ctx)
411{
412	return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
413}
414
415static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
416				     enum tls_offload_ctx_dir direction)
417{
418	if (direction == TLS_OFFLOAD_CTX_DIR_TX)
419		return tls_offload_ctx_tx(tls_ctx)->driver_state;
420	else
421		return tls_offload_ctx_rx(tls_ctx)->driver_state;
422}
423
424static inline void *
425tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
426{
427	return __tls_driver_ctx(tls_get_ctx(sk), direction);
428}
429
430#define RESYNC_REQ BIT(0)
431#define RESYNC_REQ_ASYNC BIT(1)
432/* The TLS context is valid until sk_destruct is called */
433static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
434{
435	struct tls_context *tls_ctx = tls_get_ctx(sk);
436	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
437
438	atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ);
439}
440
441/* Log all TLS record header TCP sequences in [seq, seq+len] */
442static inline void
443tls_offload_rx_resync_async_request_start(struct sock *sk, __be32 seq, u16 len)
444{
445	struct tls_context *tls_ctx = tls_get_ctx(sk);
446	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
447
448	atomic64_set(&rx_ctx->resync_async->req, ((u64)ntohl(seq) << 32) |
449		     ((u64)len << 16) | RESYNC_REQ | RESYNC_REQ_ASYNC);
450	rx_ctx->resync_async->loglen = 0;
451	rx_ctx->resync_async->rcd_delta = 0;
452}
453
454static inline void
455tls_offload_rx_resync_async_request_end(struct sock *sk, __be32 seq)
456{
457	struct tls_context *tls_ctx = tls_get_ctx(sk);
458	struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
459
460	atomic64_set(&rx_ctx->resync_async->req,
461		     ((u64)ntohl(seq) << 32) | RESYNC_REQ);
462}
463
464static inline void
465tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
466{
467	struct tls_context *tls_ctx = tls_get_ctx(sk);
468
469	tls_offload_ctx_rx(tls_ctx)->resync_type = type;
470}
471
472/* Driver's seq tracking has to be disabled until resync succeeded */
473static inline bool tls_offload_tx_resync_pending(struct sock *sk)
474{
475	struct tls_context *tls_ctx = tls_get_ctx(sk);
476	bool ret;
477
478	ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
479	smp_mb__after_atomic();
480	return ret;
481}
482
483struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
484
485#ifdef CONFIG_TLS_DEVICE
486void tls_device_sk_destruct(struct sock *sk);
487void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq);
488
489static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk)
490{
491	if (!sk_fullsock(sk) ||
492	    smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct)
493		return false;
494	return tls_get_ctx(sk)->rx_conf == TLS_HW;
495}
496#endif
497#endif /* _TLS_OFFLOAD_H */
498