2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #ifndef _TLS_OFFLOAD_H
35 #define _TLS_OFFLOAD_H
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>
46 #include <net/net_namespace.h>
48 #include <net/strparser.h>
49 #include <crypto/aead.h>
50 #include <uapi/linux/tls.h>
54 /* Maximum data size carried in a TLS record */
55 #define TLS_MAX_PAYLOAD_SIZE ((size_t)1 << 14)
57 #define TLS_HEADER_SIZE 5
58 #define TLS_NONCE_OFFSET TLS_HEADER_SIZE
60 #define TLS_CRYPTO_INFO_READY(info) ((info)->cipher_type)
62 #define TLS_AAD_SPACE_SIZE 13
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
70 /* For CCM mode, the full 16-bytes of IV is made of '4' fields of given sizes.
72 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
74 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
75 * Hence b0 contains (3 - 1) = 2.
77 #define TLS_AES_CCM_IV_B0_BYTE 2
78 #define TLS_SM4_CCM_IV_B0_BYTE 2
89 struct delayed_work work;
93 struct 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 /* protect crypto_wait with encrypt_pending */
101 spinlock_t encrypt_compl_lock;
105 #define BIT_TX_SCHEDULED 0
106 #define BIT_TX_CLOSING 1
107 unsigned long tx_bitmask;
110 struct tls_strparser {
116 u32 mixed_decrypted : 1;
121 struct sk_buff *anchor;
122 struct work_struct work;
125 struct tls_sw_context_rx {
126 struct crypto_aead *aead_recv;
127 struct crypto_wait async_wait;
128 struct sk_buff_head rx_list; /* list of decrypted 'data' records */
129 void (*saved_data_ready)(struct sock *sk);
134 u8 reader_contended:1;
136 struct tls_strparser strp;
138 atomic_t decrypt_pending;
139 /* protect crypto_wait with decrypt_pending*/
140 spinlock_t decrypt_compl_lock;
141 struct sk_buff_head async_hold;
142 struct wait_queue_head wq;
145 struct tls_record_info {
146 struct list_head list;
150 skb_frag_t frags[MAX_SKB_FRAGS];
153 #define TLS_DRIVER_STATE_SIZE_TX 16
154 struct tls_offload_context_tx {
155 struct crypto_aead *aead_send;
156 spinlock_t lock; /* protects records list */
157 struct list_head records_list;
158 struct tls_record_info *open_record;
159 struct tls_record_info *retransmit_hint;
161 u64 unacked_record_sn;
163 struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
164 void (*sk_destruct)(struct sock *sk);
165 struct work_struct destruct_work;
166 struct tls_context *ctx;
167 /* The TLS layer reserves room for driver specific state
168 * Currently the belief is that there is not enough
169 * driver specific state to justify another layer of indirection
171 u8 driver_state[TLS_DRIVER_STATE_SIZE_TX] __aligned(8);
174 enum tls_context_flags {
175 /* tls_device_down was called after the netdev went down, device state
176 * was released, and kTLS works in software, even though rx_conf is
177 * still TLS_HW (needed for transition).
179 TLS_RX_DEV_DEGRADED = 0,
180 /* Unlike RX where resync is driven entirely by the core in TX only
181 * the driver knows when things went out of sync, so we need the flag
184 TLS_TX_SYNC_SCHED = 1,
185 /* tls_dev_del was called for the RX side, device state was released,
186 * but tls_ctx->netdev might still be kept, because TX-side driver
187 * resources might not be released yet. Used to prevent the second
188 * tls_dev_del call in tls_device_down if it happens simultaneously.
190 TLS_RX_DEV_CLOSED = 2,
193 struct cipher_context {
194 char iv[TLS_MAX_IV_SIZE + TLS_MAX_SALT_SIZE];
195 char rec_seq[TLS_MAX_REC_SEQ_SIZE];
198 union tls_crypto_context {
199 struct tls_crypto_info info;
201 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
202 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
203 struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305;
204 struct tls12_crypto_info_sm4_gcm sm4_gcm;
205 struct tls12_crypto_info_sm4_ccm sm4_ccm;
209 struct tls_prot_info {
223 /* read-only cache line */
224 struct tls_prot_info prot_info;
228 u8 zerocopy_sendfile:1;
231 int (*push_pending_record)(struct sock *sk, int flags);
232 void (*sk_write_space)(struct sock *sk);
237 struct net_device __rcu *netdev;
240 struct cipher_context tx;
241 struct cipher_context rx;
243 struct scatterlist *partially_sent_record;
244 u16 partially_sent_offset;
247 bool pending_open_record_frags;
249 struct mutex tx_lock; /* protects partially_sent_* fields and
254 /* cache cold stuff */
255 struct proto *sk_proto;
258 void (*sk_destruct)(struct sock *sk);
260 union tls_crypto_context crypto_send;
261 union tls_crypto_context crypto_recv;
263 struct list_head list;
268 enum tls_offload_ctx_dir {
269 TLS_OFFLOAD_CTX_DIR_RX,
270 TLS_OFFLOAD_CTX_DIR_TX,
274 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
275 enum tls_offload_ctx_dir direction,
276 struct tls_crypto_info *crypto_info,
277 u32 start_offload_tcp_sn);
278 void (*tls_dev_del)(struct net_device *netdev,
279 struct tls_context *ctx,
280 enum tls_offload_ctx_dir direction);
281 int (*tls_dev_resync)(struct net_device *netdev,
282 struct sock *sk, u32 seq, u8 *rcd_sn,
283 enum tls_offload_ctx_dir direction);
286 enum tls_offload_sync_type {
287 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
288 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
289 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2,
292 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2
293 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
295 #define TLS_DEVICE_RESYNC_ASYNC_LOGMAX 13
296 struct tls_offload_resync_async {
300 u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX];
303 #define TLS_DRIVER_STATE_SIZE_RX 8
304 struct tls_offload_context_rx {
305 /* sw must be the first member of tls_offload_context_rx */
306 struct tls_sw_context_rx sw;
307 enum tls_offload_sync_type resync_type;
308 /* this member is set regardless of resync_type, to avoid branches */
309 u8 resync_nh_reset:1;
310 /* CORE_NEXT_HINT-only member, but use the hole here */
311 u8 resync_nh_do_now:1;
313 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
315 atomic64_t resync_req;
317 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
319 u32 decrypted_failed;
322 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC */
324 struct tls_offload_resync_async *resync_async;
327 /* The TLS layer reserves room for driver specific state
328 * Currently the belief is that there is not enough
329 * driver specific state to justify another layer of indirection
331 u8 driver_state[TLS_DRIVER_STATE_SIZE_RX] __aligned(8);
334 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
335 u32 seq, u64 *p_record_sn);
337 static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
339 return rec->len == 0;
342 static inline u32 tls_record_start_seq(struct tls_record_info *rec)
344 return rec->end_seq - rec->len;
348 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
349 struct sk_buff *skb);
351 tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev,
352 struct sk_buff *skb);
354 static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb)
356 #ifdef CONFIG_TLS_DEVICE
357 struct sock *sk = skb->sk;
359 return sk && sk_fullsock(sk) &&
360 (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
361 &tls_validate_xmit_skb);
367 static inline struct tls_context *tls_get_ctx(const struct sock *sk)
369 struct inet_connection_sock *icsk = inet_csk(sk);
371 /* Use RCU on icsk_ulp_data only for sock diag code,
372 * TLS data path doesn't need rcu_dereference().
374 return (__force void *)icsk->icsk_ulp_data;
377 static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
378 const struct tls_context *tls_ctx)
380 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
383 static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
384 const struct tls_context *tls_ctx)
386 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
389 static inline struct tls_offload_context_tx *
390 tls_offload_ctx_tx(const struct tls_context *tls_ctx)
392 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
395 static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
397 struct tls_context *ctx = tls_get_ctx(sk);
401 return !!tls_sw_ctx_tx(ctx);
404 static inline bool tls_sw_has_ctx_rx(const struct sock *sk)
406 struct tls_context *ctx = tls_get_ctx(sk);
410 return !!tls_sw_ctx_rx(ctx);
413 static inline struct tls_offload_context_rx *
414 tls_offload_ctx_rx(const struct tls_context *tls_ctx)
416 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
419 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
420 enum tls_offload_ctx_dir direction)
422 if (direction == TLS_OFFLOAD_CTX_DIR_TX)
423 return tls_offload_ctx_tx(tls_ctx)->driver_state;
425 return tls_offload_ctx_rx(tls_ctx)->driver_state;
429 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
431 return __tls_driver_ctx(tls_get_ctx(sk), direction);
434 #define RESYNC_REQ BIT(0)
435 #define RESYNC_REQ_ASYNC BIT(1)
436 /* The TLS context is valid until sk_destruct is called */
437 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
439 struct tls_context *tls_ctx = tls_get_ctx(sk);
440 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
442 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ);
445 /* Log all TLS record header TCP sequences in [seq, seq+len] */
447 tls_offload_rx_resync_async_request_start(struct sock *sk, __be32 seq, u16 len)
449 struct tls_context *tls_ctx = tls_get_ctx(sk);
450 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
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;
459 tls_offload_rx_resync_async_request_end(struct sock *sk, __be32 seq)
461 struct tls_context *tls_ctx = tls_get_ctx(sk);
462 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
464 atomic64_set(&rx_ctx->resync_async->req,
465 ((u64)ntohl(seq) << 32) | RESYNC_REQ);
469 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
471 struct tls_context *tls_ctx = tls_get_ctx(sk);
473 tls_offload_ctx_rx(tls_ctx)->resync_type = type;
476 /* Driver's seq tracking has to be disabled until resync succeeded */
477 static inline bool tls_offload_tx_resync_pending(struct sock *sk)
479 struct tls_context *tls_ctx = tls_get_ctx(sk);
482 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
483 smp_mb__after_atomic();
487 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
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);
493 static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk)
495 if (!sk_fullsock(sk) ||
496 smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct)
498 return tls_get_ctx(sk)->rx_conf == TLS_HW;
501 #endif /* _TLS_OFFLOAD_H */