Loading...
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/skmsg.h>
43#include <linux/mutex.h>
44#include <linux/netdevice.h>
45#include <linux/rcupdate.h>
46
47#include <net/tcp.h>
48#include <net/strparser.h>
49#include <crypto/aead.h>
50#include <uapi/linux/tls.h>
51
52
53/* Maximum data size carried in a TLS record */
54#define TLS_MAX_PAYLOAD_SIZE ((size_t)1 << 14)
55
56#define TLS_HEADER_SIZE 5
57#define TLS_NONCE_OFFSET TLS_HEADER_SIZE
58
59#define TLS_CRYPTO_INFO_READY(info) ((info)->cipher_type)
60
61#define TLS_RECORD_TYPE_DATA 0x17
62
63#define TLS_AAD_SPACE_SIZE 13
64#define TLS_DEVICE_NAME_MAX 32
65
66#define MAX_IV_SIZE 16
67#define TLS_MAX_REC_SEQ_SIZE 8
68
69/* For AES-CCM, the full 16-bytes of IV is made of '4' fields of given sizes.
70 *
71 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
72 *
73 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
74 * Hence b0 contains (3 - 1) = 2.
75 */
76#define TLS_AES_CCM_IV_B0_BYTE 2
77
78/*
79 * This structure defines the routines for Inline TLS driver.
80 * The following routines are optional and filled with a
81 * null pointer if not defined.
82 *
83 * @name: Its the name of registered Inline tls device
84 * @dev_list: Inline tls device list
85 * int (*feature)(struct tls_device *device);
86 * Called to return Inline TLS driver capability
87 *
88 * int (*hash)(struct tls_device *device, struct sock *sk);
89 * This function sets Inline driver for listen and program
90 * device specific functioanlity as required
91 *
92 * void (*unhash)(struct tls_device *device, struct sock *sk);
93 * This function cleans listen state set by Inline TLS driver
94 *
95 * void (*release)(struct kref *kref);
96 * Release the registered device and allocated resources
97 * @kref: Number of reference to tls_device
98 */
99struct tls_device {
100 char name[TLS_DEVICE_NAME_MAX];
101 struct list_head dev_list;
102 int (*feature)(struct tls_device *device);
103 int (*hash)(struct tls_device *device, struct sock *sk);
104 void (*unhash)(struct tls_device *device, struct sock *sk);
105 void (*release)(struct kref *kref);
106 struct kref kref;
107};
108
109enum {
110 TLS_BASE,
111 TLS_SW,
112 TLS_HW,
113 TLS_HW_RECORD,
114 TLS_NUM_CONFIG,
115};
116
117/* TLS records are maintained in 'struct tls_rec'. It stores the memory pages
118 * allocated or mapped for each TLS record. After encryption, the records are
119 * stores in a linked list.
120 */
121struct tls_rec {
122 struct list_head list;
123 int tx_ready;
124 int tx_flags;
125 int inplace_crypto;
126
127 struct sk_msg msg_plaintext;
128 struct sk_msg msg_encrypted;
129
130 /* AAD | msg_plaintext.sg.data | sg_tag */
131 struct scatterlist sg_aead_in[2];
132 /* AAD | msg_encrypted.sg.data (data contains overhead for hdr & iv & tag) */
133 struct scatterlist sg_aead_out[2];
134
135 char content_type;
136 struct scatterlist sg_content_type;
137
138 char aad_space[TLS_AAD_SPACE_SIZE];
139 u8 iv_data[MAX_IV_SIZE];
140 struct aead_request aead_req;
141 u8 aead_req_ctx[];
142};
143
144struct tls_msg {
145 struct strp_msg rxm;
146 u8 control;
147};
148
149struct tx_work {
150 struct delayed_work work;
151 struct sock *sk;
152};
153
154struct tls_sw_context_tx {
155 struct crypto_aead *aead_send;
156 struct crypto_wait async_wait;
157 struct tx_work tx_work;
158 struct tls_rec *open_rec;
159 struct list_head tx_list;
160 atomic_t encrypt_pending;
161 int async_notify;
162 int async_capable;
163
164#define BIT_TX_SCHEDULED 0
165#define BIT_TX_CLOSING 1
166 unsigned long tx_bitmask;
167};
168
169struct tls_sw_context_rx {
170 struct crypto_aead *aead_recv;
171 struct crypto_wait async_wait;
172 struct strparser strp;
173 struct sk_buff_head rx_list; /* list of decrypted 'data' records */
174 void (*saved_data_ready)(struct sock *sk);
175
176 struct sk_buff *recv_pkt;
177 u8 control;
178 int async_capable;
179 bool decrypted;
180 atomic_t decrypt_pending;
181 bool async_notify;
182};
183
184struct tls_record_info {
185 struct list_head list;
186 u32 end_seq;
187 int len;
188 int num_frags;
189 skb_frag_t frags[MAX_SKB_FRAGS];
190};
191
192struct tls_offload_context_tx {
193 struct crypto_aead *aead_send;
194 spinlock_t lock; /* protects records list */
195 struct list_head records_list;
196 struct tls_record_info *open_record;
197 struct tls_record_info *retransmit_hint;
198 u64 hint_record_sn;
199 u64 unacked_record_sn;
200
201 struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
202 void (*sk_destruct)(struct sock *sk);
203 u8 driver_state[] __aligned(8);
204 /* The TLS layer reserves room for driver specific state
205 * Currently the belief is that there is not enough
206 * driver specific state to justify another layer of indirection
207 */
208#define TLS_DRIVER_STATE_SIZE_TX 16
209};
210
211#define TLS_OFFLOAD_CONTEXT_SIZE_TX \
212 (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX)
213
214enum tls_context_flags {
215 TLS_RX_SYNC_RUNNING = 0,
216 /* Unlike RX where resync is driven entirely by the core in TX only
217 * the driver knows when things went out of sync, so we need the flag
218 * to be atomic.
219 */
220 TLS_TX_SYNC_SCHED = 1,
221};
222
223struct cipher_context {
224 char *iv;
225 char *rec_seq;
226};
227
228union tls_crypto_context {
229 struct tls_crypto_info info;
230 union {
231 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
232 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
233 };
234};
235
236struct tls_prot_info {
237 u16 version;
238 u16 cipher_type;
239 u16 prepend_size;
240 u16 tag_size;
241 u16 overhead_size;
242 u16 iv_size;
243 u16 salt_size;
244 u16 rec_seq_size;
245 u16 aad_size;
246 u16 tail_size;
247};
248
249struct tls_context {
250 /* read-only cache line */
251 struct tls_prot_info prot_info;
252
253 u8 tx_conf:3;
254 u8 rx_conf:3;
255
256 int (*push_pending_record)(struct sock *sk, int flags);
257 void (*sk_write_space)(struct sock *sk);
258
259 void *priv_ctx_tx;
260 void *priv_ctx_rx;
261
262 struct net_device *netdev;
263
264 /* rw cache line */
265 struct cipher_context tx;
266 struct cipher_context rx;
267
268 struct scatterlist *partially_sent_record;
269 u16 partially_sent_offset;
270
271 bool in_tcp_sendpages;
272 bool pending_open_record_frags;
273
274 struct mutex tx_lock; /* protects partially_sent_* fields and
275 * per-type TX fields
276 */
277 unsigned long flags;
278
279 /* cache cold stuff */
280 struct proto *sk_proto;
281
282 void (*sk_destruct)(struct sock *sk);
283
284 union tls_crypto_context crypto_send;
285 union tls_crypto_context crypto_recv;
286
287 struct list_head list;
288 refcount_t refcount;
289 struct rcu_head rcu;
290};
291
292enum tls_offload_ctx_dir {
293 TLS_OFFLOAD_CTX_DIR_RX,
294 TLS_OFFLOAD_CTX_DIR_TX,
295};
296
297struct tlsdev_ops {
298 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
299 enum tls_offload_ctx_dir direction,
300 struct tls_crypto_info *crypto_info,
301 u32 start_offload_tcp_sn);
302 void (*tls_dev_del)(struct net_device *netdev,
303 struct tls_context *ctx,
304 enum tls_offload_ctx_dir direction);
305 int (*tls_dev_resync)(struct net_device *netdev,
306 struct sock *sk, u32 seq, u8 *rcd_sn,
307 enum tls_offload_ctx_dir direction);
308};
309
310enum tls_offload_sync_type {
311 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
312 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
313};
314
315#define TLS_DEVICE_RESYNC_NH_START_IVAL 2
316#define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
317
318struct tls_offload_context_rx {
319 /* sw must be the first member of tls_offload_context_rx */
320 struct tls_sw_context_rx sw;
321 enum tls_offload_sync_type resync_type;
322 /* this member is set regardless of resync_type, to avoid branches */
323 u8 resync_nh_reset:1;
324 /* CORE_NEXT_HINT-only member, but use the hole here */
325 u8 resync_nh_do_now:1;
326 union {
327 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
328 struct {
329 atomic64_t resync_req;
330 };
331 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
332 struct {
333 u32 decrypted_failed;
334 u32 decrypted_tgt;
335 } resync_nh;
336 };
337 u8 driver_state[] __aligned(8);
338 /* The TLS layer reserves room for driver specific state
339 * Currently the belief is that there is not enough
340 * driver specific state to justify another layer of indirection
341 */
342#define TLS_DRIVER_STATE_SIZE_RX 8
343};
344
345#define TLS_OFFLOAD_CONTEXT_SIZE_RX \
346 (sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX)
347
348void tls_ctx_free(struct sock *sk, struct tls_context *ctx);
349int wait_on_pending_writer(struct sock *sk, long *timeo);
350int tls_sk_query(struct sock *sk, int optname, char __user *optval,
351 int __user *optlen);
352int tls_sk_attach(struct sock *sk, int optname, char __user *optval,
353 unsigned int optlen);
354
355int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx);
356void tls_sw_strparser_arm(struct sock *sk, struct tls_context *ctx);
357void tls_sw_strparser_done(struct tls_context *tls_ctx);
358int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
359int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
360 int offset, size_t size, int flags);
361int tls_sw_sendpage(struct sock *sk, struct page *page,
362 int offset, size_t size, int flags);
363void tls_sw_cancel_work_tx(struct tls_context *tls_ctx);
364void tls_sw_release_resources_tx(struct sock *sk);
365void tls_sw_free_ctx_tx(struct tls_context *tls_ctx);
366void tls_sw_free_resources_rx(struct sock *sk);
367void tls_sw_release_resources_rx(struct sock *sk);
368void tls_sw_free_ctx_rx(struct tls_context *tls_ctx);
369int tls_sw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
370 int nonblock, int flags, int *addr_len);
371bool tls_sw_stream_read(const struct sock *sk);
372ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
373 struct pipe_inode_info *pipe,
374 size_t len, unsigned int flags);
375
376int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
377int tls_device_sendpage(struct sock *sk, struct page *page,
378 int offset, size_t size, int flags);
379int tls_tx_records(struct sock *sk, int flags);
380
381struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
382 u32 seq, u64 *p_record_sn);
383
384static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
385{
386 return rec->len == 0;
387}
388
389static inline u32 tls_record_start_seq(struct tls_record_info *rec)
390{
391 return rec->end_seq - rec->len;
392}
393
394int tls_push_sg(struct sock *sk, struct tls_context *ctx,
395 struct scatterlist *sg, u16 first_offset,
396 int flags);
397int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
398 int flags);
399bool tls_free_partial_record(struct sock *sk, struct tls_context *ctx);
400
401static inline struct tls_msg *tls_msg(struct sk_buff *skb)
402{
403 return (struct tls_msg *)strp_msg(skb);
404}
405
406static inline bool tls_is_partially_sent_record(struct tls_context *ctx)
407{
408 return !!ctx->partially_sent_record;
409}
410
411static inline bool tls_is_pending_open_record(struct tls_context *tls_ctx)
412{
413 return tls_ctx->pending_open_record_frags;
414}
415
416static inline bool is_tx_ready(struct tls_sw_context_tx *ctx)
417{
418 struct tls_rec *rec;
419
420 rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
421 if (!rec)
422 return false;
423
424 return READ_ONCE(rec->tx_ready);
425}
426
427static inline u16 tls_user_config(struct tls_context *ctx, bool tx)
428{
429 u16 config = tx ? ctx->tx_conf : ctx->rx_conf;
430
431 switch (config) {
432 case TLS_BASE:
433 return TLS_CONF_BASE;
434 case TLS_SW:
435 return TLS_CONF_SW;
436 case TLS_HW:
437 return TLS_CONF_HW;
438 case TLS_HW_RECORD:
439 return TLS_CONF_HW_RECORD;
440 }
441 return 0;
442}
443
444struct sk_buff *
445tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
446 struct sk_buff *skb);
447
448static inline bool tls_is_sk_tx_device_offloaded(struct sock *sk)
449{
450#ifdef CONFIG_SOCK_VALIDATE_XMIT
451 return sk_fullsock(sk) &&
452 (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
453 &tls_validate_xmit_skb);
454#else
455 return false;
456#endif
457}
458
459static inline void tls_err_abort(struct sock *sk, int err)
460{
461 sk->sk_err = err;
462 sk->sk_error_report(sk);
463}
464
465static inline bool tls_bigint_increment(unsigned char *seq, int len)
466{
467 int i;
468
469 for (i = len - 1; i >= 0; i--) {
470 ++seq[i];
471 if (seq[i] != 0)
472 break;
473 }
474
475 return (i == -1);
476}
477
478static inline struct tls_context *tls_get_ctx(const struct sock *sk)
479{
480 struct inet_connection_sock *icsk = inet_csk(sk);
481
482 /* Use RCU on icsk_ulp_data only for sock diag code,
483 * TLS data path doesn't need rcu_dereference().
484 */
485 return (__force void *)icsk->icsk_ulp_data;
486}
487
488static inline void tls_advance_record_sn(struct sock *sk,
489 struct tls_prot_info *prot,
490 struct cipher_context *ctx)
491{
492 if (tls_bigint_increment(ctx->rec_seq, prot->rec_seq_size))
493 tls_err_abort(sk, EBADMSG);
494
495 if (prot->version != TLS_1_3_VERSION)
496 tls_bigint_increment(ctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
497 prot->iv_size);
498}
499
500static inline void tls_fill_prepend(struct tls_context *ctx,
501 char *buf,
502 size_t plaintext_len,
503 unsigned char record_type,
504 int version)
505{
506 struct tls_prot_info *prot = &ctx->prot_info;
507 size_t pkt_len, iv_size = prot->iv_size;
508
509 pkt_len = plaintext_len + prot->tag_size;
510 if (version != TLS_1_3_VERSION) {
511 pkt_len += iv_size;
512
513 memcpy(buf + TLS_NONCE_OFFSET,
514 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv_size);
515 }
516
517 /* we cover nonce explicit here as well, so buf should be of
518 * size KTLS_DTLS_HEADER_SIZE + KTLS_DTLS_NONCE_EXPLICIT_SIZE
519 */
520 buf[0] = version == TLS_1_3_VERSION ?
521 TLS_RECORD_TYPE_DATA : record_type;
522 /* Note that VERSION must be TLS_1_2 for both TLS1.2 and TLS1.3 */
523 buf[1] = TLS_1_2_VERSION_MINOR;
524 buf[2] = TLS_1_2_VERSION_MAJOR;
525 /* we can use IV for nonce explicit according to spec */
526 buf[3] = pkt_len >> 8;
527 buf[4] = pkt_len & 0xFF;
528}
529
530static inline void tls_make_aad(char *buf,
531 size_t size,
532 char *record_sequence,
533 int record_sequence_size,
534 unsigned char record_type,
535 int version)
536{
537 if (version != TLS_1_3_VERSION) {
538 memcpy(buf, record_sequence, record_sequence_size);
539 buf += 8;
540 } else {
541 size += TLS_CIPHER_AES_GCM_128_TAG_SIZE;
542 }
543
544 buf[0] = version == TLS_1_3_VERSION ?
545 TLS_RECORD_TYPE_DATA : record_type;
546 buf[1] = TLS_1_2_VERSION_MAJOR;
547 buf[2] = TLS_1_2_VERSION_MINOR;
548 buf[3] = size >> 8;
549 buf[4] = size & 0xFF;
550}
551
552static inline void xor_iv_with_seq(int version, char *iv, char *seq)
553{
554 int i;
555
556 if (version == TLS_1_3_VERSION) {
557 for (i = 0; i < 8; i++)
558 iv[i + 4] ^= seq[i];
559 }
560}
561
562
563static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
564 const struct tls_context *tls_ctx)
565{
566 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
567}
568
569static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
570 const struct tls_context *tls_ctx)
571{
572 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
573}
574
575static inline struct tls_offload_context_tx *
576tls_offload_ctx_tx(const struct tls_context *tls_ctx)
577{
578 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
579}
580
581static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
582{
583 struct tls_context *ctx = tls_get_ctx(sk);
584
585 if (!ctx)
586 return false;
587 return !!tls_sw_ctx_tx(ctx);
588}
589
590void tls_sw_write_space(struct sock *sk, struct tls_context *ctx);
591void tls_device_write_space(struct sock *sk, struct tls_context *ctx);
592
593static inline struct tls_offload_context_rx *
594tls_offload_ctx_rx(const struct tls_context *tls_ctx)
595{
596 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
597}
598
599#if IS_ENABLED(CONFIG_TLS_DEVICE)
600static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
601 enum tls_offload_ctx_dir direction)
602{
603 if (direction == TLS_OFFLOAD_CTX_DIR_TX)
604 return tls_offload_ctx_tx(tls_ctx)->driver_state;
605 else
606 return tls_offload_ctx_rx(tls_ctx)->driver_state;
607}
608
609static inline void *
610tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
611{
612 return __tls_driver_ctx(tls_get_ctx(sk), direction);
613}
614#endif
615
616/* The TLS context is valid until sk_destruct is called */
617static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
618{
619 struct tls_context *tls_ctx = tls_get_ctx(sk);
620 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
621
622 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | 1);
623}
624
625static inline void
626tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
627{
628 struct tls_context *tls_ctx = tls_get_ctx(sk);
629
630 tls_offload_ctx_rx(tls_ctx)->resync_type = type;
631}
632
633static inline void tls_offload_tx_resync_request(struct sock *sk)
634{
635 struct tls_context *tls_ctx = tls_get_ctx(sk);
636
637 WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
638}
639
640/* Driver's seq tracking has to be disabled until resync succeeded */
641static inline bool tls_offload_tx_resync_pending(struct sock *sk)
642{
643 struct tls_context *tls_ctx = tls_get_ctx(sk);
644 bool ret;
645
646 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
647 smp_mb__after_atomic();
648 return ret;
649}
650
651int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
652 unsigned char *record_type);
653void tls_register_device(struct tls_device *device);
654void tls_unregister_device(struct tls_device *device);
655int decrypt_skb(struct sock *sk, struct sk_buff *skb,
656 struct scatterlist *sgout);
657struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
658
659struct sk_buff *tls_validate_xmit_skb(struct sock *sk,
660 struct net_device *dev,
661 struct sk_buff *skb);
662
663int tls_sw_fallback_init(struct sock *sk,
664 struct tls_offload_context_tx *offload_ctx,
665 struct tls_crypto_info *crypto_info);
666
667#ifdef CONFIG_TLS_DEVICE
668void tls_device_init(void);
669void tls_device_cleanup(void);
670int tls_set_device_offload(struct sock *sk, struct tls_context *ctx);
671void tls_device_free_resources_tx(struct sock *sk);
672int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx);
673void tls_device_offload_cleanup_rx(struct sock *sk);
674void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq);
675int tls_device_decrypted(struct sock *sk, struct sk_buff *skb);
676#else
677static inline void tls_device_init(void) {}
678static inline void tls_device_cleanup(void) {}
679
680static inline int
681tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
682{
683 return -EOPNOTSUPP;
684}
685
686static inline void tls_device_free_resources_tx(struct sock *sk) {}
687
688static inline int
689tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
690{
691 return -EOPNOTSUPP;
692}
693
694static inline void tls_device_offload_cleanup_rx(struct sock *sk) {}
695static inline void
696tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq) {}
697
698static inline int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
699{
700 return 0;
701}
702#endif
703#endif /* _TLS_OFFLOAD_H */
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 */