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/netdevice.h>
46 #include <net/strparser.h>
47 #include <crypto/aead.h>
48 #include <uapi/linux/tls.h>
51 /* Maximum data size carried in a TLS record */
52 #define TLS_MAX_PAYLOAD_SIZE ((size_t)1 << 14)
54 #define TLS_HEADER_SIZE 5
55 #define TLS_NONCE_OFFSET TLS_HEADER_SIZE
57 #define TLS_CRYPTO_INFO_READY(info) ((info)->cipher_type)
59 #define TLS_RECORD_TYPE_DATA 0x17
61 #define TLS_AAD_SPACE_SIZE 13
62 #define TLS_DEVICE_NAME_MAX 32
64 #define MAX_IV_SIZE 16
65 #define TLS_MAX_REC_SEQ_SIZE 8
67 /* For AES-CCM, the full 16-bytes of IV is made of '4' fields of given sizes.
69 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
71 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
72 * Hence b0 contains (3 - 1) = 2.
74 #define TLS_AES_CCM_IV_B0_BYTE 2
77 * This structure defines the routines for Inline TLS driver.
78 * The following routines are optional and filled with a
79 * null pointer if not defined.
81 * @name: Its the name of registered Inline tls device
82 * @dev_list: Inline tls device list
83 * int (*feature)(struct tls_device *device);
84 * Called to return Inline TLS driver capability
86 * int (*hash)(struct tls_device *device, struct sock *sk);
87 * This function sets Inline driver for listen and program
88 * device specific functioanlity as required
90 * void (*unhash)(struct tls_device *device, struct sock *sk);
91 * This function cleans listen state set by Inline TLS driver
93 * void (*release)(struct kref *kref);
94 * Release the registered device and allocated resources
95 * @kref: Number of reference to tls_device
98 char name[TLS_DEVICE_NAME_MAX];
99 struct list_head dev_list;
100 int (*feature)(struct tls_device *device);
101 int (*hash)(struct tls_device *device, struct sock *sk);
102 void (*unhash)(struct tls_device *device, struct sock *sk);
103 void (*release)(struct kref *kref);
115 /* TLS records are maintained in 'struct tls_rec'. It stores the memory pages
116 * allocated or mapped for each TLS record. After encryption, the records are
117 * stores in a linked list.
120 struct list_head list;
125 struct sk_msg msg_plaintext;
126 struct sk_msg msg_encrypted;
128 /* AAD | msg_plaintext.sg.data | sg_tag */
129 struct scatterlist sg_aead_in[2];
130 /* AAD | msg_encrypted.sg.data (data contains overhead for hdr & iv & tag) */
131 struct scatterlist sg_aead_out[2];
134 struct scatterlist sg_content_type;
136 char aad_space[TLS_AAD_SPACE_SIZE];
137 u8 iv_data[MAX_IV_SIZE];
138 struct aead_request aead_req;
148 struct delayed_work work;
152 struct tls_sw_context_tx {
153 struct crypto_aead *aead_send;
154 struct crypto_wait async_wait;
155 struct tx_work tx_work;
156 struct tls_rec *open_rec;
157 struct list_head tx_list;
158 atomic_t encrypt_pending;
162 #define BIT_TX_SCHEDULED 0
163 #define BIT_TX_CLOSING 1
164 unsigned long tx_bitmask;
167 struct tls_sw_context_rx {
168 struct crypto_aead *aead_recv;
169 struct crypto_wait async_wait;
170 struct strparser strp;
171 struct sk_buff_head rx_list; /* list of decrypted 'data' records */
172 void (*saved_data_ready)(struct sock *sk);
174 struct sk_buff *recv_pkt;
178 atomic_t decrypt_pending;
182 struct tls_record_info {
183 struct list_head list;
187 skb_frag_t frags[MAX_SKB_FRAGS];
190 struct tls_offload_context_tx {
191 struct crypto_aead *aead_send;
192 spinlock_t lock; /* protects records list */
193 struct list_head records_list;
194 struct tls_record_info *open_record;
195 struct tls_record_info *retransmit_hint;
197 u64 unacked_record_sn;
199 struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
200 void (*sk_destruct)(struct sock *sk);
201 u8 driver_state[] __aligned(8);
202 /* The TLS layer reserves room for driver specific state
203 * Currently the belief is that there is not enough
204 * driver specific state to justify another layer of indirection
206 #define TLS_DRIVER_STATE_SIZE_TX 16
209 #define TLS_OFFLOAD_CONTEXT_SIZE_TX \
210 (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX)
212 enum tls_context_flags {
213 TLS_RX_SYNC_RUNNING = 0,
214 /* Unlike RX where resync is driven entirely by the core in TX only
215 * the driver knows when things went out of sync, so we need the flag
218 TLS_TX_SYNC_SCHED = 1,
221 struct cipher_context {
226 union tls_crypto_context {
227 struct tls_crypto_info info;
229 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
230 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
234 struct tls_prot_info {
248 /* read-only cache line */
249 struct tls_prot_info prot_info;
254 int (*push_pending_record)(struct sock *sk, int flags);
255 void (*sk_write_space)(struct sock *sk);
260 struct net_device *netdev;
263 struct cipher_context tx;
264 struct cipher_context rx;
266 struct scatterlist *partially_sent_record;
267 u16 partially_sent_offset;
269 bool in_tcp_sendpages;
270 bool pending_open_record_frags;
273 /* cache cold stuff */
274 struct proto *sk_proto;
276 void (*sk_destruct)(struct sock *sk);
277 void (*sk_proto_close)(struct sock *sk, long timeout);
279 int (*setsockopt)(struct sock *sk, int level,
280 int optname, char __user *optval,
281 unsigned int optlen);
282 int (*getsockopt)(struct sock *sk, int level,
283 int optname, char __user *optval,
285 int (*hash)(struct sock *sk);
286 void (*unhash)(struct sock *sk);
288 union tls_crypto_context crypto_send;
289 union tls_crypto_context crypto_recv;
291 struct list_head list;
294 struct work_struct gc;
297 enum tls_offload_ctx_dir {
298 TLS_OFFLOAD_CTX_DIR_RX,
299 TLS_OFFLOAD_CTX_DIR_TX,
303 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
304 enum tls_offload_ctx_dir direction,
305 struct tls_crypto_info *crypto_info,
306 u32 start_offload_tcp_sn);
307 void (*tls_dev_del)(struct net_device *netdev,
308 struct tls_context *ctx,
309 enum tls_offload_ctx_dir direction);
310 int (*tls_dev_resync)(struct net_device *netdev,
311 struct sock *sk, u32 seq, u8 *rcd_sn,
312 enum tls_offload_ctx_dir direction);
315 enum tls_offload_sync_type {
316 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
317 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
320 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2
321 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
323 struct tls_offload_context_rx {
324 /* sw must be the first member of tls_offload_context_rx */
325 struct tls_sw_context_rx sw;
326 enum tls_offload_sync_type resync_type;
327 /* this member is set regardless of resync_type, to avoid branches */
328 u8 resync_nh_reset:1;
329 /* CORE_NEXT_HINT-only member, but use the hole here */
330 u8 resync_nh_do_now:1;
332 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
334 atomic64_t resync_req;
336 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
338 u32 decrypted_failed;
342 u8 driver_state[] __aligned(8);
343 /* The TLS layer reserves room for driver specific state
344 * Currently the belief is that there is not enough
345 * driver specific state to justify another layer of indirection
347 #define TLS_DRIVER_STATE_SIZE_RX 8
350 #define TLS_OFFLOAD_CONTEXT_SIZE_RX \
351 (sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX)
353 void tls_ctx_free(struct tls_context *ctx);
354 int wait_on_pending_writer(struct sock *sk, long *timeo);
355 int tls_sk_query(struct sock *sk, int optname, char __user *optval,
357 int tls_sk_attach(struct sock *sk, int optname, char __user *optval,
358 unsigned int optlen);
360 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx);
361 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *ctx);
362 void tls_sw_strparser_done(struct tls_context *tls_ctx);
363 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
364 int tls_sw_sendpage(struct sock *sk, struct page *page,
365 int offset, size_t size, int flags);
366 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx);
367 void tls_sw_release_resources_tx(struct sock *sk);
368 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx);
369 void tls_sw_free_resources_rx(struct sock *sk);
370 void tls_sw_release_resources_rx(struct sock *sk);
371 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx);
372 int tls_sw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
373 int nonblock, int flags, int *addr_len);
374 bool tls_sw_stream_read(const struct sock *sk);
375 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
376 struct pipe_inode_info *pipe,
377 size_t len, unsigned int flags);
379 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx);
380 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
381 int tls_device_sendpage(struct sock *sk, struct page *page,
382 int offset, size_t size, int flags);
383 void tls_device_free_resources_tx(struct sock *sk);
384 void tls_device_init(void);
385 void tls_device_cleanup(void);
386 int tls_tx_records(struct sock *sk, int flags);
388 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
389 u32 seq, u64 *p_record_sn);
391 static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
393 return rec->len == 0;
396 static inline u32 tls_record_start_seq(struct tls_record_info *rec)
398 return rec->end_seq - rec->len;
401 int tls_push_sg(struct sock *sk, struct tls_context *ctx,
402 struct scatterlist *sg, u16 first_offset,
404 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
406 bool tls_free_partial_record(struct sock *sk, struct tls_context *ctx);
408 static inline struct tls_msg *tls_msg(struct sk_buff *skb)
410 return (struct tls_msg *)strp_msg(skb);
413 static inline bool tls_is_partially_sent_record(struct tls_context *ctx)
415 return !!ctx->partially_sent_record;
418 static inline bool tls_is_pending_open_record(struct tls_context *tls_ctx)
420 return tls_ctx->pending_open_record_frags;
423 static inline bool is_tx_ready(struct tls_sw_context_tx *ctx)
427 rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
431 return READ_ONCE(rec->tx_ready);
435 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
436 struct sk_buff *skb);
438 static inline bool tls_is_sk_tx_device_offloaded(struct sock *sk)
440 #ifdef CONFIG_SOCK_VALIDATE_XMIT
441 return sk_fullsock(sk) &&
442 (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
443 &tls_validate_xmit_skb);
449 static inline void tls_err_abort(struct sock *sk, int err)
452 sk->sk_error_report(sk);
455 static inline bool tls_bigint_increment(unsigned char *seq, int len)
459 for (i = len - 1; i >= 0; i--) {
468 static inline struct tls_context *tls_get_ctx(const struct sock *sk)
470 struct inet_connection_sock *icsk = inet_csk(sk);
472 return icsk->icsk_ulp_data;
475 static inline void tls_advance_record_sn(struct sock *sk,
476 struct tls_prot_info *prot,
477 struct cipher_context *ctx)
479 if (tls_bigint_increment(ctx->rec_seq, prot->rec_seq_size))
480 tls_err_abort(sk, EBADMSG);
482 if (prot->version != TLS_1_3_VERSION)
483 tls_bigint_increment(ctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
487 static inline void tls_fill_prepend(struct tls_context *ctx,
489 size_t plaintext_len,
490 unsigned char record_type,
493 struct tls_prot_info *prot = &ctx->prot_info;
494 size_t pkt_len, iv_size = prot->iv_size;
496 pkt_len = plaintext_len + prot->tag_size;
497 if (version != TLS_1_3_VERSION) {
500 memcpy(buf + TLS_NONCE_OFFSET,
501 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv_size);
504 /* we cover nonce explicit here as well, so buf should be of
505 * size KTLS_DTLS_HEADER_SIZE + KTLS_DTLS_NONCE_EXPLICIT_SIZE
507 buf[0] = version == TLS_1_3_VERSION ?
508 TLS_RECORD_TYPE_DATA : record_type;
509 /* Note that VERSION must be TLS_1_2 for both TLS1.2 and TLS1.3 */
510 buf[1] = TLS_1_2_VERSION_MINOR;
511 buf[2] = TLS_1_2_VERSION_MAJOR;
512 /* we can use IV for nonce explicit according to spec */
513 buf[3] = pkt_len >> 8;
514 buf[4] = pkt_len & 0xFF;
517 static inline void tls_make_aad(char *buf,
519 char *record_sequence,
520 int record_sequence_size,
521 unsigned char record_type,
524 if (version != TLS_1_3_VERSION) {
525 memcpy(buf, record_sequence, record_sequence_size);
528 size += TLS_CIPHER_AES_GCM_128_TAG_SIZE;
531 buf[0] = version == TLS_1_3_VERSION ?
532 TLS_RECORD_TYPE_DATA : record_type;
533 buf[1] = TLS_1_2_VERSION_MAJOR;
534 buf[2] = TLS_1_2_VERSION_MINOR;
536 buf[4] = size & 0xFF;
539 static inline void xor_iv_with_seq(int version, char *iv, char *seq)
543 if (version == TLS_1_3_VERSION) {
544 for (i = 0; i < 8; i++)
550 static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
551 const struct tls_context *tls_ctx)
553 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
556 static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
557 const struct tls_context *tls_ctx)
559 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
562 static inline struct tls_offload_context_tx *
563 tls_offload_ctx_tx(const struct tls_context *tls_ctx)
565 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
568 static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
570 struct tls_context *ctx = tls_get_ctx(sk);
574 return !!tls_sw_ctx_tx(ctx);
577 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx);
578 void tls_device_write_space(struct sock *sk, struct tls_context *ctx);
580 static inline struct tls_offload_context_rx *
581 tls_offload_ctx_rx(const struct tls_context *tls_ctx)
583 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
586 #if IS_ENABLED(CONFIG_TLS_DEVICE)
587 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
588 enum tls_offload_ctx_dir direction)
590 if (direction == TLS_OFFLOAD_CTX_DIR_TX)
591 return tls_offload_ctx_tx(tls_ctx)->driver_state;
593 return tls_offload_ctx_rx(tls_ctx)->driver_state;
597 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
599 return __tls_driver_ctx(tls_get_ctx(sk), direction);
603 /* The TLS context is valid until sk_destruct is called */
604 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
606 struct tls_context *tls_ctx = tls_get_ctx(sk);
607 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
609 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | 1);
613 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
615 struct tls_context *tls_ctx = tls_get_ctx(sk);
617 tls_offload_ctx_rx(tls_ctx)->resync_type = type;
620 static inline void tls_offload_tx_resync_request(struct sock *sk)
622 struct tls_context *tls_ctx = tls_get_ctx(sk);
624 WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
627 /* Driver's seq tracking has to be disabled until resync succeeded */
628 static inline bool tls_offload_tx_resync_pending(struct sock *sk)
630 struct tls_context *tls_ctx = tls_get_ctx(sk);
633 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
634 smp_mb__after_atomic();
638 int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
639 unsigned char *record_type);
640 void tls_register_device(struct tls_device *device);
641 void tls_unregister_device(struct tls_device *device);
642 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb);
643 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
644 struct scatterlist *sgout);
645 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
647 struct sk_buff *tls_validate_xmit_skb(struct sock *sk,
648 struct net_device *dev,
649 struct sk_buff *skb);
651 int tls_sw_fallback_init(struct sock *sk,
652 struct tls_offload_context_tx *offload_ctx,
653 struct tls_crypto_info *crypto_info);
655 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx);
657 void tls_device_offload_cleanup_rx(struct sock *sk);
658 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq);
660 #endif /* _TLS_OFFLOAD_H */