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;
105 struct sk_msg msg_plaintext;
106 struct sk_msg msg_encrypted;
108 /* AAD | msg_plaintext.sg.data | sg_tag */
109 struct scatterlist sg_aead_in[2];
110 /* AAD | msg_encrypted.sg.data (data contains overhead for hdr & iv & tag) */
111 struct scatterlist sg_aead_out[2];
114 struct scatterlist sg_content_type;
116 char aad_space[TLS_AAD_SPACE_SIZE];
117 u8 iv_data[MAX_IV_SIZE];
118 struct aead_request aead_req;
128 struct delayed_work work;
132 struct tls_sw_context_tx {
133 struct crypto_aead *aead_send;
134 struct crypto_wait async_wait;
135 struct tx_work tx_work;
136 struct tls_rec *open_rec;
137 struct list_head tx_list;
138 atomic_t encrypt_pending;
142 #define BIT_TX_SCHEDULED 0
143 #define BIT_TX_CLOSING 1
144 unsigned long tx_bitmask;
147 struct tls_sw_context_rx {
148 struct crypto_aead *aead_recv;
149 struct crypto_wait async_wait;
150 struct strparser strp;
151 struct sk_buff_head rx_list; /* list of decrypted 'data' records */
152 void (*saved_data_ready)(struct sock *sk);
154 struct sk_buff *recv_pkt;
158 atomic_t decrypt_pending;
162 struct tls_record_info {
163 struct list_head list;
167 skb_frag_t frags[MAX_SKB_FRAGS];
170 struct tls_offload_context_tx {
171 struct crypto_aead *aead_send;
172 spinlock_t lock; /* protects records list */
173 struct list_head records_list;
174 struct tls_record_info *open_record;
175 struct tls_record_info *retransmit_hint;
177 u64 unacked_record_sn;
179 struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
180 void (*sk_destruct)(struct sock *sk);
181 u8 driver_state[] __aligned(8);
182 /* The TLS layer reserves room for driver specific state
183 * Currently the belief is that there is not enough
184 * driver specific state to justify another layer of indirection
186 #define TLS_DRIVER_STATE_SIZE_TX 16
189 #define TLS_OFFLOAD_CONTEXT_SIZE_TX \
190 (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX)
192 enum tls_context_flags {
193 TLS_RX_SYNC_RUNNING = 0,
194 /* Unlike RX where resync is driven entirely by the core in TX only
195 * the driver knows when things went out of sync, so we need the flag
198 TLS_TX_SYNC_SCHED = 1,
201 struct cipher_context {
206 union tls_crypto_context {
207 struct tls_crypto_info info;
209 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
210 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
214 struct tls_prot_info {
228 /* read-only cache line */
229 struct tls_prot_info prot_info;
234 int (*push_pending_record)(struct sock *sk, int flags);
235 void (*sk_write_space)(struct sock *sk);
240 struct net_device *netdev;
243 struct cipher_context tx;
244 struct cipher_context rx;
246 struct scatterlist *partially_sent_record;
247 u16 partially_sent_offset;
249 bool in_tcp_sendpages;
250 bool pending_open_record_frags;
252 struct mutex tx_lock; /* protects partially_sent_* fields and
257 /* cache cold stuff */
258 struct proto *sk_proto;
260 void (*sk_destruct)(struct sock *sk);
262 union tls_crypto_context crypto_send;
263 union tls_crypto_context crypto_recv;
265 struct list_head list;
270 enum tls_offload_ctx_dir {
271 TLS_OFFLOAD_CTX_DIR_RX,
272 TLS_OFFLOAD_CTX_DIR_TX,
276 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
277 enum tls_offload_ctx_dir direction,
278 struct tls_crypto_info *crypto_info,
279 u32 start_offload_tcp_sn);
280 void (*tls_dev_del)(struct net_device *netdev,
281 struct tls_context *ctx,
282 enum tls_offload_ctx_dir direction);
283 int (*tls_dev_resync)(struct net_device *netdev,
284 struct sock *sk, u32 seq, u8 *rcd_sn,
285 enum tls_offload_ctx_dir direction);
288 enum tls_offload_sync_type {
289 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
290 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
293 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2
294 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
296 struct tls_offload_context_rx {
297 /* sw must be the first member of tls_offload_context_rx */
298 struct tls_sw_context_rx sw;
299 enum tls_offload_sync_type resync_type;
300 /* this member is set regardless of resync_type, to avoid branches */
301 u8 resync_nh_reset:1;
302 /* CORE_NEXT_HINT-only member, but use the hole here */
303 u8 resync_nh_do_now:1;
305 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
307 atomic64_t resync_req;
309 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
311 u32 decrypted_failed;
315 u8 driver_state[] __aligned(8);
316 /* The TLS layer reserves room for driver specific state
317 * Currently the belief is that there is not enough
318 * driver specific state to justify another layer of indirection
320 #define TLS_DRIVER_STATE_SIZE_RX 8
323 #define TLS_OFFLOAD_CONTEXT_SIZE_RX \
324 (sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX)
326 struct tls_context *tls_ctx_create(struct sock *sk);
327 void tls_ctx_free(struct sock *sk, struct tls_context *ctx);
328 void update_sk_prot(struct sock *sk, struct tls_context *ctx);
330 int wait_on_pending_writer(struct sock *sk, long *timeo);
331 int tls_sk_query(struct sock *sk, int optname, char __user *optval,
333 int tls_sk_attach(struct sock *sk, int optname, char __user *optval,
334 unsigned int optlen);
336 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx);
337 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *ctx);
338 void tls_sw_strparser_done(struct tls_context *tls_ctx);
339 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
340 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
341 int offset, size_t size, int flags);
342 int tls_sw_sendpage(struct sock *sk, struct page *page,
343 int offset, size_t size, int flags);
344 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx);
345 void tls_sw_release_resources_tx(struct sock *sk);
346 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx);
347 void tls_sw_free_resources_rx(struct sock *sk);
348 void tls_sw_release_resources_rx(struct sock *sk);
349 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx);
350 int tls_sw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
351 int nonblock, int flags, int *addr_len);
352 bool tls_sw_stream_read(const struct sock *sk);
353 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
354 struct pipe_inode_info *pipe,
355 size_t len, unsigned int flags);
357 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
358 int tls_device_sendpage(struct sock *sk, struct page *page,
359 int offset, size_t size, int flags);
360 int tls_tx_records(struct sock *sk, int flags);
362 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
363 u32 seq, u64 *p_record_sn);
365 static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
367 return rec->len == 0;
370 static inline u32 tls_record_start_seq(struct tls_record_info *rec)
372 return rec->end_seq - rec->len;
375 int tls_push_sg(struct sock *sk, struct tls_context *ctx,
376 struct scatterlist *sg, u16 first_offset,
378 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
380 bool tls_free_partial_record(struct sock *sk, struct tls_context *ctx);
382 static inline struct tls_msg *tls_msg(struct sk_buff *skb)
384 return (struct tls_msg *)strp_msg(skb);
387 static inline bool tls_is_partially_sent_record(struct tls_context *ctx)
389 return !!ctx->partially_sent_record;
392 static inline bool tls_is_pending_open_record(struct tls_context *tls_ctx)
394 return tls_ctx->pending_open_record_frags;
397 static inline bool is_tx_ready(struct tls_sw_context_tx *ctx)
401 rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
405 return READ_ONCE(rec->tx_ready);
408 static inline u16 tls_user_config(struct tls_context *ctx, bool tx)
410 u16 config = tx ? ctx->tx_conf : ctx->rx_conf;
414 return TLS_CONF_BASE;
420 return TLS_CONF_HW_RECORD;
426 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
427 struct sk_buff *skb);
429 static inline bool tls_is_sk_tx_device_offloaded(struct sock *sk)
431 #ifdef CONFIG_SOCK_VALIDATE_XMIT
432 return sk_fullsock(sk) &&
433 (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
434 &tls_validate_xmit_skb);
440 static inline void tls_err_abort(struct sock *sk, int err)
443 sk->sk_error_report(sk);
446 static inline bool tls_bigint_increment(unsigned char *seq, int len)
450 for (i = len - 1; i >= 0; i--) {
459 static inline struct tls_context *tls_get_ctx(const struct sock *sk)
461 struct inet_connection_sock *icsk = inet_csk(sk);
463 /* Use RCU on icsk_ulp_data only for sock diag code,
464 * TLS data path doesn't need rcu_dereference().
466 return (__force void *)icsk->icsk_ulp_data;
469 static inline void tls_advance_record_sn(struct sock *sk,
470 struct tls_prot_info *prot,
471 struct cipher_context *ctx)
473 if (tls_bigint_increment(ctx->rec_seq, prot->rec_seq_size))
474 tls_err_abort(sk, EBADMSG);
476 if (prot->version != TLS_1_3_VERSION)
477 tls_bigint_increment(ctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
481 static inline void tls_fill_prepend(struct tls_context *ctx,
483 size_t plaintext_len,
484 unsigned char record_type,
487 struct tls_prot_info *prot = &ctx->prot_info;
488 size_t pkt_len, iv_size = prot->iv_size;
490 pkt_len = plaintext_len + prot->tag_size;
491 if (version != TLS_1_3_VERSION) {
494 memcpy(buf + TLS_NONCE_OFFSET,
495 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv_size);
498 /* we cover nonce explicit here as well, so buf should be of
499 * size KTLS_DTLS_HEADER_SIZE + KTLS_DTLS_NONCE_EXPLICIT_SIZE
501 buf[0] = version == TLS_1_3_VERSION ?
502 TLS_RECORD_TYPE_DATA : record_type;
503 /* Note that VERSION must be TLS_1_2 for both TLS1.2 and TLS1.3 */
504 buf[1] = TLS_1_2_VERSION_MINOR;
505 buf[2] = TLS_1_2_VERSION_MAJOR;
506 /* we can use IV for nonce explicit according to spec */
507 buf[3] = pkt_len >> 8;
508 buf[4] = pkt_len & 0xFF;
511 static inline void tls_make_aad(char *buf,
513 char *record_sequence,
514 int record_sequence_size,
515 unsigned char record_type,
518 if (version != TLS_1_3_VERSION) {
519 memcpy(buf, record_sequence, record_sequence_size);
522 size += TLS_CIPHER_AES_GCM_128_TAG_SIZE;
525 buf[0] = version == TLS_1_3_VERSION ?
526 TLS_RECORD_TYPE_DATA : record_type;
527 buf[1] = TLS_1_2_VERSION_MAJOR;
528 buf[2] = TLS_1_2_VERSION_MINOR;
530 buf[4] = size & 0xFF;
533 static inline void xor_iv_with_seq(int version, char *iv, char *seq)
537 if (version == TLS_1_3_VERSION) {
538 for (i = 0; i < 8; i++)
544 static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
545 const struct tls_context *tls_ctx)
547 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
550 static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
551 const struct tls_context *tls_ctx)
553 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
556 static inline struct tls_offload_context_tx *
557 tls_offload_ctx_tx(const struct tls_context *tls_ctx)
559 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
562 static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
564 struct tls_context *ctx = tls_get_ctx(sk);
568 return !!tls_sw_ctx_tx(ctx);
571 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx);
572 void tls_device_write_space(struct sock *sk, struct tls_context *ctx);
574 static inline struct tls_offload_context_rx *
575 tls_offload_ctx_rx(const struct tls_context *tls_ctx)
577 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
580 #if IS_ENABLED(CONFIG_TLS_DEVICE)
581 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
582 enum tls_offload_ctx_dir direction)
584 if (direction == TLS_OFFLOAD_CTX_DIR_TX)
585 return tls_offload_ctx_tx(tls_ctx)->driver_state;
587 return tls_offload_ctx_rx(tls_ctx)->driver_state;
591 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
593 return __tls_driver_ctx(tls_get_ctx(sk), direction);
597 /* The TLS context is valid until sk_destruct is called */
598 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
600 struct tls_context *tls_ctx = tls_get_ctx(sk);
601 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
603 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | 1);
607 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
609 struct tls_context *tls_ctx = tls_get_ctx(sk);
611 tls_offload_ctx_rx(tls_ctx)->resync_type = type;
614 /* Driver's seq tracking has to be disabled until resync succeeded */
615 static inline bool tls_offload_tx_resync_pending(struct sock *sk)
617 struct tls_context *tls_ctx = tls_get_ctx(sk);
620 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
621 smp_mb__after_atomic();
625 int __net_init tls_proc_init(struct net *net);
626 void __net_exit tls_proc_fini(struct net *net);
628 int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
629 unsigned char *record_type);
630 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
631 struct scatterlist *sgout);
632 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
634 struct sk_buff *tls_validate_xmit_skb(struct sock *sk,
635 struct net_device *dev,
636 struct sk_buff *skb);
638 int tls_sw_fallback_init(struct sock *sk,
639 struct tls_offload_context_tx *offload_ctx,
640 struct tls_crypto_info *crypto_info);
642 #ifdef CONFIG_TLS_DEVICE
643 void tls_device_init(void);
644 void tls_device_cleanup(void);
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 void tls_device_init(void) {}
655 static inline void tls_device_cleanup(void) {}
658 tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
663 static inline void tls_device_free_resources_tx(struct sock *sk) {}
666 tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
671 static inline void tls_device_offload_cleanup_rx(struct sock *sk) {}
673 tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq) {}
676 tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
677 struct sk_buff *skb, struct strp_msg *rxm)
682 #endif /* _TLS_OFFLOAD_H */