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/skmsg.h>
43 #include <linux/mutex.h>
44 #include <linux/netdevice.h>
45 #include <linux/rcupdate.h>
47 #include <net/net_namespace.h>
49 #include <net/strparser.h>
50 #include <crypto/aead.h>
51 #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_RECORD_TYPE_DATA 0x17
64 #define TLS_AAD_SPACE_SIZE 13
66 #define MAX_IV_SIZE 16
67 #define TLS_MAX_REC_SEQ_SIZE 8
69 /* For AES-CCM, the full 16-bytes of IV is made of '4' fields of given sizes.
71 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
73 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
74 * Hence b0 contains (3 - 1) = 2.
76 #define TLS_AES_CCM_IV_B0_BYTE 2
78 #define __TLS_INC_STATS(net, field) \
79 __SNMP_INC_STATS((net)->mib.tls_statistics, field)
80 #define TLS_INC_STATS(net, field) \
81 SNMP_INC_STATS((net)->mib.tls_statistics, field)
82 #define __TLS_DEC_STATS(net, field) \
83 __SNMP_DEC_STATS((net)->mib.tls_statistics, field)
84 #define TLS_DEC_STATS(net, field) \
85 SNMP_DEC_STATS((net)->mib.tls_statistics, field)
95 /* TLS records are maintained in 'struct tls_rec'. It stores the memory pages
96 * allocated or mapped for each TLS record. After encryption, the records are
97 * stores in a linked list.
100 struct list_head list
;
104 struct sk_msg msg_plaintext
;
105 struct sk_msg msg_encrypted
;
107 /* AAD | msg_plaintext.sg.data | sg_tag */
108 struct scatterlist sg_aead_in
[2];
109 /* AAD | msg_encrypted.sg.data (data contains overhead for hdr & iv & tag) */
110 struct scatterlist sg_aead_out
[2];
113 struct scatterlist sg_content_type
;
115 char aad_space
[TLS_AAD_SPACE_SIZE
];
116 u8 iv_data
[MAX_IV_SIZE
];
117 struct aead_request aead_req
;
127 struct delayed_work work
;
131 struct tls_sw_context_tx
{
132 struct crypto_aead
*aead_send
;
133 struct crypto_wait async_wait
;
134 struct tx_work tx_work
;
135 struct tls_rec
*open_rec
;
136 struct list_head tx_list
;
137 atomic_t encrypt_pending
;
141 #define BIT_TX_SCHEDULED 0
142 #define BIT_TX_CLOSING 1
143 unsigned long tx_bitmask
;
146 struct tls_sw_context_rx
{
147 struct crypto_aead
*aead_recv
;
148 struct crypto_wait async_wait
;
149 struct strparser strp
;
150 struct sk_buff_head rx_list
; /* list of decrypted 'data' records */
151 void (*saved_data_ready
)(struct sock
*sk
);
153 struct sk_buff
*recv_pkt
;
157 atomic_t decrypt_pending
;
161 struct tls_record_info
{
162 struct list_head list
;
166 skb_frag_t frags
[MAX_SKB_FRAGS
];
169 struct tls_offload_context_tx
{
170 struct crypto_aead
*aead_send
;
171 spinlock_t lock
; /* protects records list */
172 struct list_head records_list
;
173 struct tls_record_info
*open_record
;
174 struct tls_record_info
*retransmit_hint
;
176 u64 unacked_record_sn
;
178 struct scatterlist sg_tx_data
[MAX_SKB_FRAGS
];
179 void (*sk_destruct
)(struct sock
*sk
);
180 u8 driver_state
[] __aligned(8);
181 /* The TLS layer reserves room for driver specific state
182 * Currently the belief is that there is not enough
183 * driver specific state to justify another layer of indirection
185 #define TLS_DRIVER_STATE_SIZE_TX 16
188 #define TLS_OFFLOAD_CONTEXT_SIZE_TX \
189 (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX)
191 enum tls_context_flags
{
192 TLS_RX_SYNC_RUNNING
= 0,
193 /* Unlike RX where resync is driven entirely by the core in TX only
194 * the driver knows when things went out of sync, so we need the flag
197 TLS_TX_SYNC_SCHED
= 1,
200 struct cipher_context
{
205 union tls_crypto_context
{
206 struct tls_crypto_info info
;
208 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128
;
209 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256
;
213 struct tls_prot_info
{
227 /* read-only cache line */
228 struct tls_prot_info prot_info
;
233 int (*push_pending_record
)(struct sock
*sk
, int flags
);
234 void (*sk_write_space
)(struct sock
*sk
);
239 struct net_device
*netdev
;
242 struct cipher_context tx
;
243 struct cipher_context rx
;
245 struct scatterlist
*partially_sent_record
;
246 u16 partially_sent_offset
;
248 bool in_tcp_sendpages
;
249 bool pending_open_record_frags
;
251 struct mutex tx_lock
; /* protects partially_sent_* fields and
256 /* cache cold stuff */
257 struct proto
*sk_proto
;
259 void (*sk_destruct
)(struct sock
*sk
);
261 union tls_crypto_context crypto_send
;
262 union tls_crypto_context crypto_recv
;
264 struct list_head list
;
269 enum tls_offload_ctx_dir
{
270 TLS_OFFLOAD_CTX_DIR_RX
,
271 TLS_OFFLOAD_CTX_DIR_TX
,
275 int (*tls_dev_add
)(struct net_device
*netdev
, struct sock
*sk
,
276 enum tls_offload_ctx_dir direction
,
277 struct tls_crypto_info
*crypto_info
,
278 u32 start_offload_tcp_sn
);
279 void (*tls_dev_del
)(struct net_device
*netdev
,
280 struct tls_context
*ctx
,
281 enum tls_offload_ctx_dir direction
);
282 int (*tls_dev_resync
)(struct net_device
*netdev
,
283 struct sock
*sk
, u32 seq
, u8
*rcd_sn
,
284 enum tls_offload_ctx_dir direction
);
287 enum tls_offload_sync_type
{
288 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ
= 0,
289 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT
= 1,
292 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2
293 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
295 struct tls_offload_context_rx
{
296 /* sw must be the first member of tls_offload_context_rx */
297 struct tls_sw_context_rx sw
;
298 enum tls_offload_sync_type resync_type
;
299 /* this member is set regardless of resync_type, to avoid branches */
300 u8 resync_nh_reset
:1;
301 /* CORE_NEXT_HINT-only member, but use the hole here */
302 u8 resync_nh_do_now
:1;
304 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
306 atomic64_t resync_req
;
308 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
310 u32 decrypted_failed
;
314 u8 driver_state
[] __aligned(8);
315 /* The TLS layer reserves room for driver specific state
316 * Currently the belief is that there is not enough
317 * driver specific state to justify another layer of indirection
319 #define TLS_DRIVER_STATE_SIZE_RX 8
322 #define TLS_OFFLOAD_CONTEXT_SIZE_RX \
323 (sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX)
325 struct tls_context
*tls_ctx_create(struct sock
*sk
);
326 void tls_ctx_free(struct sock
*sk
, struct tls_context
*ctx
);
327 void update_sk_prot(struct sock
*sk
, struct tls_context
*ctx
);
329 int wait_on_pending_writer(struct sock
*sk
, long *timeo
);
330 int tls_sk_query(struct sock
*sk
, int optname
, char __user
*optval
,
332 int tls_sk_attach(struct sock
*sk
, int optname
, char __user
*optval
,
333 unsigned int optlen
);
335 int tls_set_sw_offload(struct sock
*sk
, struct tls_context
*ctx
, int tx
);
336 void tls_sw_strparser_arm(struct sock
*sk
, struct tls_context
*ctx
);
337 void tls_sw_strparser_done(struct tls_context
*tls_ctx
);
338 int tls_sw_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
339 int tls_sw_sendpage_locked(struct sock
*sk
, struct page
*page
,
340 int offset
, size_t size
, int flags
);
341 int tls_sw_sendpage(struct sock
*sk
, struct page
*page
,
342 int offset
, size_t size
, int flags
);
343 void tls_sw_cancel_work_tx(struct tls_context
*tls_ctx
);
344 void tls_sw_release_resources_tx(struct sock
*sk
);
345 void tls_sw_free_ctx_tx(struct tls_context
*tls_ctx
);
346 void tls_sw_free_resources_rx(struct sock
*sk
);
347 void tls_sw_release_resources_rx(struct sock
*sk
);
348 void tls_sw_free_ctx_rx(struct tls_context
*tls_ctx
);
349 int tls_sw_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
,
350 int nonblock
, int flags
, int *addr_len
);
351 bool tls_sw_stream_read(const struct sock
*sk
);
352 ssize_t
tls_sw_splice_read(struct socket
*sock
, loff_t
*ppos
,
353 struct pipe_inode_info
*pipe
,
354 size_t len
, unsigned int flags
);
356 int tls_device_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
357 int tls_device_sendpage(struct sock
*sk
, struct page
*page
,
358 int offset
, size_t size
, int flags
);
359 int tls_tx_records(struct sock
*sk
, int flags
);
361 struct tls_record_info
*tls_get_record(struct tls_offload_context_tx
*context
,
362 u32 seq
, u64
*p_record_sn
);
364 static inline bool tls_record_is_start_marker(struct tls_record_info
*rec
)
366 return rec
->len
== 0;
369 static inline u32
tls_record_start_seq(struct tls_record_info
*rec
)
371 return rec
->end_seq
- rec
->len
;
374 int tls_push_sg(struct sock
*sk
, struct tls_context
*ctx
,
375 struct scatterlist
*sg
, u16 first_offset
,
377 int tls_push_partial_record(struct sock
*sk
, struct tls_context
*ctx
,
379 void tls_free_partial_record(struct sock
*sk
, struct tls_context
*ctx
);
381 static inline struct tls_msg
*tls_msg(struct sk_buff
*skb
)
383 return (struct tls_msg
*)strp_msg(skb
);
386 static inline bool tls_is_partially_sent_record(struct tls_context
*ctx
)
388 return !!ctx
->partially_sent_record
;
391 static inline bool tls_is_pending_open_record(struct tls_context
*tls_ctx
)
393 return tls_ctx
->pending_open_record_frags
;
396 static inline bool is_tx_ready(struct tls_sw_context_tx
*ctx
)
400 rec
= list_first_entry(&ctx
->tx_list
, struct tls_rec
, list
);
404 return READ_ONCE(rec
->tx_ready
);
407 static inline u16
tls_user_config(struct tls_context
*ctx
, bool tx
)
409 u16 config
= tx
? ctx
->tx_conf
: ctx
->rx_conf
;
413 return TLS_CONF_BASE
;
419 return TLS_CONF_HW_RECORD
;
425 tls_validate_xmit_skb(struct sock
*sk
, struct net_device
*dev
,
426 struct sk_buff
*skb
);
428 static inline bool tls_is_sk_tx_device_offloaded(struct sock
*sk
)
430 #ifdef CONFIG_SOCK_VALIDATE_XMIT
431 return sk_fullsock(sk
) &&
432 (smp_load_acquire(&sk
->sk_validate_xmit_skb
) ==
433 &tls_validate_xmit_skb
);
439 static inline void tls_err_abort(struct sock
*sk
, int err
)
442 sk
->sk_error_report(sk
);
445 static inline bool tls_bigint_increment(unsigned char *seq
, int len
)
449 for (i
= len
- 1; i
>= 0; i
--) {
458 static inline struct tls_context
*tls_get_ctx(const struct sock
*sk
)
460 struct inet_connection_sock
*icsk
= inet_csk(sk
);
462 /* Use RCU on icsk_ulp_data only for sock diag code,
463 * TLS data path doesn't need rcu_dereference().
465 return (__force
void *)icsk
->icsk_ulp_data
;
468 static inline void tls_advance_record_sn(struct sock
*sk
,
469 struct tls_prot_info
*prot
,
470 struct cipher_context
*ctx
)
472 if (tls_bigint_increment(ctx
->rec_seq
, prot
->rec_seq_size
))
473 tls_err_abort(sk
, EBADMSG
);
475 if (prot
->version
!= TLS_1_3_VERSION
)
476 tls_bigint_increment(ctx
->iv
+ TLS_CIPHER_AES_GCM_128_SALT_SIZE
,
480 static inline void tls_fill_prepend(struct tls_context
*ctx
,
482 size_t plaintext_len
,
483 unsigned char record_type
,
486 struct tls_prot_info
*prot
= &ctx
->prot_info
;
487 size_t pkt_len
, iv_size
= prot
->iv_size
;
489 pkt_len
= plaintext_len
+ prot
->tag_size
;
490 if (version
!= TLS_1_3_VERSION
) {
493 memcpy(buf
+ TLS_NONCE_OFFSET
,
494 ctx
->tx
.iv
+ TLS_CIPHER_AES_GCM_128_SALT_SIZE
, iv_size
);
497 /* we cover nonce explicit here as well, so buf should be of
498 * size KTLS_DTLS_HEADER_SIZE + KTLS_DTLS_NONCE_EXPLICIT_SIZE
500 buf
[0] = version
== TLS_1_3_VERSION
?
501 TLS_RECORD_TYPE_DATA
: record_type
;
502 /* Note that VERSION must be TLS_1_2 for both TLS1.2 and TLS1.3 */
503 buf
[1] = TLS_1_2_VERSION_MINOR
;
504 buf
[2] = TLS_1_2_VERSION_MAJOR
;
505 /* we can use IV for nonce explicit according to spec */
506 buf
[3] = pkt_len
>> 8;
507 buf
[4] = pkt_len
& 0xFF;
510 static inline void tls_make_aad(char *buf
,
512 char *record_sequence
,
513 int record_sequence_size
,
514 unsigned char record_type
,
517 if (version
!= TLS_1_3_VERSION
) {
518 memcpy(buf
, record_sequence
, record_sequence_size
);
521 size
+= TLS_CIPHER_AES_GCM_128_TAG_SIZE
;
524 buf
[0] = version
== TLS_1_3_VERSION
?
525 TLS_RECORD_TYPE_DATA
: record_type
;
526 buf
[1] = TLS_1_2_VERSION_MAJOR
;
527 buf
[2] = TLS_1_2_VERSION_MINOR
;
529 buf
[4] = size
& 0xFF;
532 static inline void xor_iv_with_seq(int version
, char *iv
, char *seq
)
536 if (version
== TLS_1_3_VERSION
) {
537 for (i
= 0; i
< 8; i
++)
543 static inline struct tls_sw_context_rx
*tls_sw_ctx_rx(
544 const struct tls_context
*tls_ctx
)
546 return (struct tls_sw_context_rx
*)tls_ctx
->priv_ctx_rx
;
549 static inline struct tls_sw_context_tx
*tls_sw_ctx_tx(
550 const struct tls_context
*tls_ctx
)
552 return (struct tls_sw_context_tx
*)tls_ctx
->priv_ctx_tx
;
555 static inline struct tls_offload_context_tx
*
556 tls_offload_ctx_tx(const struct tls_context
*tls_ctx
)
558 return (struct tls_offload_context_tx
*)tls_ctx
->priv_ctx_tx
;
561 static inline bool tls_sw_has_ctx_tx(const struct sock
*sk
)
563 struct tls_context
*ctx
= tls_get_ctx(sk
);
567 return !!tls_sw_ctx_tx(ctx
);
570 void tls_sw_write_space(struct sock
*sk
, struct tls_context
*ctx
);
571 void tls_device_write_space(struct sock
*sk
, struct tls_context
*ctx
);
573 static inline struct tls_offload_context_rx
*
574 tls_offload_ctx_rx(const struct tls_context
*tls_ctx
)
576 return (struct tls_offload_context_rx
*)tls_ctx
->priv_ctx_rx
;
579 #if IS_ENABLED(CONFIG_TLS_DEVICE)
580 static inline void *__tls_driver_ctx(struct tls_context
*tls_ctx
,
581 enum tls_offload_ctx_dir direction
)
583 if (direction
== TLS_OFFLOAD_CTX_DIR_TX
)
584 return tls_offload_ctx_tx(tls_ctx
)->driver_state
;
586 return tls_offload_ctx_rx(tls_ctx
)->driver_state
;
590 tls_driver_ctx(const struct sock
*sk
, enum tls_offload_ctx_dir direction
)
592 return __tls_driver_ctx(tls_get_ctx(sk
), direction
);
596 /* The TLS context is valid until sk_destruct is called */
597 static inline void tls_offload_rx_resync_request(struct sock
*sk
, __be32 seq
)
599 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
600 struct tls_offload_context_rx
*rx_ctx
= tls_offload_ctx_rx(tls_ctx
);
602 atomic64_set(&rx_ctx
->resync_req
, ((u64
)ntohl(seq
) << 32) | 1);
606 tls_offload_rx_resync_set_type(struct sock
*sk
, enum tls_offload_sync_type type
)
608 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
610 tls_offload_ctx_rx(tls_ctx
)->resync_type
= type
;
613 /* Driver's seq tracking has to be disabled until resync succeeded */
614 static inline bool tls_offload_tx_resync_pending(struct sock
*sk
)
616 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
619 ret
= test_bit(TLS_TX_SYNC_SCHED
, &tls_ctx
->flags
);
620 smp_mb__after_atomic();
624 int __net_init
tls_proc_init(struct net
*net
);
625 void __net_exit
tls_proc_fini(struct net
*net
);
627 int tls_proccess_cmsg(struct sock
*sk
, struct msghdr
*msg
,
628 unsigned char *record_type
);
629 int decrypt_skb(struct sock
*sk
, struct sk_buff
*skb
,
630 struct scatterlist
*sgout
);
631 struct sk_buff
*tls_encrypt_skb(struct sk_buff
*skb
);
633 struct sk_buff
*tls_validate_xmit_skb(struct sock
*sk
,
634 struct net_device
*dev
,
635 struct sk_buff
*skb
);
637 int tls_sw_fallback_init(struct sock
*sk
,
638 struct tls_offload_context_tx
*offload_ctx
,
639 struct tls_crypto_info
*crypto_info
);
641 #ifdef CONFIG_TLS_DEVICE
642 void tls_device_init(void);
643 void tls_device_cleanup(void);
644 void tls_device_sk_destruct(struct sock
*sk
);
645 int tls_set_device_offload(struct sock
*sk
, struct tls_context
*ctx
);
646 void tls_device_free_resources_tx(struct sock
*sk
);
647 int tls_set_device_offload_rx(struct sock
*sk
, struct tls_context
*ctx
);
648 void tls_device_offload_cleanup_rx(struct sock
*sk
);
649 void tls_device_rx_resync_new_rec(struct sock
*sk
, u32 rcd_len
, u32 seq
);
650 void tls_offload_tx_resync_request(struct sock
*sk
, u32 got_seq
, u32 exp_seq
);
651 int tls_device_decrypted(struct sock
*sk
, struct tls_context
*tls_ctx
,
652 struct sk_buff
*skb
, struct strp_msg
*rxm
);
654 static inline bool tls_is_sk_rx_device_offloaded(struct sock
*sk
)
656 if (!sk_fullsock(sk
) ||
657 smp_load_acquire(&sk
->sk_destruct
) != tls_device_sk_destruct
)
659 return tls_get_ctx(sk
)->rx_conf
== TLS_HW
;
662 static inline void tls_device_init(void) {}
663 static inline void tls_device_cleanup(void) {}
666 tls_set_device_offload(struct sock
*sk
, struct tls_context
*ctx
)
671 static inline void tls_device_free_resources_tx(struct sock
*sk
) {}
674 tls_set_device_offload_rx(struct sock
*sk
, struct tls_context
*ctx
)
679 static inline void tls_device_offload_cleanup_rx(struct sock
*sk
) {}
681 tls_device_rx_resync_new_rec(struct sock
*sk
, u32 rcd_len
, u32 seq
) {}
684 tls_device_decrypted(struct sock
*sk
, struct tls_context
*tls_ctx
,
685 struct sk_buff
*skb
, struct strp_msg
*rxm
)
690 #endif /* _TLS_OFFLOAD_H */