2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4 * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
5 * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37 #include <linux/module.h>
38 #include <crypto/aead.h>
42 static inline void tls_make_aad(int recv
,
45 char *record_sequence
,
46 int record_sequence_size
,
47 unsigned char record_type
)
49 memcpy(buf
, record_sequence
, record_sequence_size
);
52 buf
[9] = TLS_1_2_VERSION_MAJOR
;
53 buf
[10] = TLS_1_2_VERSION_MINOR
;
55 buf
[12] = size
& 0xFF;
58 static void trim_sg(struct sock
*sk
, struct scatterlist
*sg
,
59 int *sg_num_elem
, unsigned int *sg_size
, int target_size
)
61 int i
= *sg_num_elem
- 1;
62 int trim
= *sg_size
- target_size
;
69 *sg_size
= target_size
;
70 while (trim
>= sg
[i
].length
) {
72 sk_mem_uncharge(sk
, sg
[i
].length
);
73 put_page(sg_page(&sg
[i
]));
81 sk_mem_uncharge(sk
, trim
);
87 static void trim_both_sgl(struct sock
*sk
, int target_size
)
89 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
90 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
92 trim_sg(sk
, ctx
->sg_plaintext_data
,
93 &ctx
->sg_plaintext_num_elem
,
94 &ctx
->sg_plaintext_size
,
98 target_size
+= tls_ctx
->overhead_size
;
100 trim_sg(sk
, ctx
->sg_encrypted_data
,
101 &ctx
->sg_encrypted_num_elem
,
102 &ctx
->sg_encrypted_size
,
106 static int alloc_sg(struct sock
*sk
, int len
, struct scatterlist
*sg
,
107 int *sg_num_elem
, unsigned int *sg_size
,
110 struct page_frag
*pfrag
;
111 unsigned int size
= *sg_size
;
112 int num_elem
= *sg_num_elem
, use
= 0, rc
= 0;
113 struct scatterlist
*sge
;
114 unsigned int orig_offset
;
117 pfrag
= sk_page_frag(sk
);
120 if (!sk_page_frag_refill(sk
, pfrag
)) {
125 use
= min_t(int, len
, pfrag
->size
- pfrag
->offset
);
127 if (!sk_wmem_schedule(sk
, use
)) {
132 sk_mem_charge(sk
, use
);
134 orig_offset
= pfrag
->offset
;
135 pfrag
->offset
+= use
;
137 sge
= sg
+ num_elem
- 1;
139 if (num_elem
> first_coalesce
&& sg_page(sge
) == pfrag
->page
&&
140 sge
->offset
+ sge
->length
== orig_offset
) {
145 sg_set_page(sge
, pfrag
->page
, use
, orig_offset
);
146 get_page(pfrag
->page
);
148 if (num_elem
== MAX_SKB_FRAGS
) {
160 *sg_num_elem
= num_elem
;
164 static int alloc_encrypted_sg(struct sock
*sk
, int len
)
166 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
167 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
170 rc
= alloc_sg(sk
, len
, ctx
->sg_encrypted_data
,
171 &ctx
->sg_encrypted_num_elem
, &ctx
->sg_encrypted_size
, 0);
174 ctx
->sg_encrypted_num_elem
= ARRAY_SIZE(ctx
->sg_encrypted_data
);
179 static int alloc_plaintext_sg(struct sock
*sk
, int len
)
181 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
182 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
185 rc
= alloc_sg(sk
, len
, ctx
->sg_plaintext_data
,
186 &ctx
->sg_plaintext_num_elem
, &ctx
->sg_plaintext_size
,
187 tls_ctx
->pending_open_record_frags
);
190 ctx
->sg_plaintext_num_elem
= ARRAY_SIZE(ctx
->sg_plaintext_data
);
195 static void free_sg(struct sock
*sk
, struct scatterlist
*sg
,
196 int *sg_num_elem
, unsigned int *sg_size
)
198 int i
, n
= *sg_num_elem
;
200 for (i
= 0; i
< n
; ++i
) {
201 sk_mem_uncharge(sk
, sg
[i
].length
);
202 put_page(sg_page(&sg
[i
]));
208 static void tls_free_both_sg(struct sock
*sk
)
210 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
211 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
213 free_sg(sk
, ctx
->sg_encrypted_data
, &ctx
->sg_encrypted_num_elem
,
214 &ctx
->sg_encrypted_size
);
216 free_sg(sk
, ctx
->sg_plaintext_data
, &ctx
->sg_plaintext_num_elem
,
217 &ctx
->sg_plaintext_size
);
220 static int tls_do_encryption(struct tls_context
*tls_ctx
,
221 struct tls_sw_context
*ctx
,
222 struct aead_request
*aead_req
,
227 ctx
->sg_encrypted_data
[0].offset
+= tls_ctx
->prepend_size
;
228 ctx
->sg_encrypted_data
[0].length
-= tls_ctx
->prepend_size
;
230 aead_request_set_tfm(aead_req
, ctx
->aead_send
);
231 aead_request_set_ad(aead_req
, TLS_AAD_SPACE_SIZE
);
232 aead_request_set_crypt(aead_req
, ctx
->sg_aead_in
, ctx
->sg_aead_out
,
233 data_len
, tls_ctx
->iv
);
234 rc
= crypto_aead_encrypt(aead_req
);
236 ctx
->sg_encrypted_data
[0].offset
-= tls_ctx
->prepend_size
;
237 ctx
->sg_encrypted_data
[0].length
+= tls_ctx
->prepend_size
;
242 static int tls_push_record(struct sock
*sk
, int flags
,
243 unsigned char record_type
)
245 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
246 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
247 struct aead_request
*req
;
250 req
= kzalloc(sizeof(struct aead_request
) +
251 crypto_aead_reqsize(ctx
->aead_send
), sk
->sk_allocation
);
255 sg_mark_end(ctx
->sg_plaintext_data
+ ctx
->sg_plaintext_num_elem
- 1);
256 sg_mark_end(ctx
->sg_encrypted_data
+ ctx
->sg_encrypted_num_elem
- 1);
258 tls_make_aad(0, ctx
->aad_space
, ctx
->sg_plaintext_size
,
259 tls_ctx
->rec_seq
, tls_ctx
->rec_seq_size
,
262 tls_fill_prepend(tls_ctx
,
263 page_address(sg_page(&ctx
->sg_encrypted_data
[0])) +
264 ctx
->sg_encrypted_data
[0].offset
,
265 ctx
->sg_plaintext_size
, record_type
);
267 tls_ctx
->pending_open_record_frags
= 0;
268 set_bit(TLS_PENDING_CLOSED_RECORD
, &tls_ctx
->flags
);
270 rc
= tls_do_encryption(tls_ctx
, ctx
, req
, ctx
->sg_plaintext_size
);
272 /* If we are called from write_space and
273 * we fail, we need to set this SOCK_NOSPACE
274 * to trigger another write_space in the future.
276 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
280 free_sg(sk
, ctx
->sg_plaintext_data
, &ctx
->sg_plaintext_num_elem
,
281 &ctx
->sg_plaintext_size
);
283 ctx
->sg_encrypted_num_elem
= 0;
284 ctx
->sg_encrypted_size
= 0;
286 /* Only pass through MSG_DONTWAIT and MSG_NOSIGNAL flags */
287 rc
= tls_push_sg(sk
, tls_ctx
, ctx
->sg_encrypted_data
, 0, flags
);
288 if (rc
< 0 && rc
!= -EAGAIN
)
291 tls_advance_record_sn(sk
, tls_ctx
);
297 static int tls_sw_push_pending_record(struct sock
*sk
, int flags
)
299 return tls_push_record(sk
, flags
, TLS_RECORD_TYPE_DATA
);
302 static int zerocopy_from_iter(struct sock
*sk
, struct iov_iter
*from
,
305 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
306 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
307 struct page
*pages
[MAX_SKB_FRAGS
];
312 unsigned int size
= ctx
->sg_plaintext_size
;
313 int num_elem
= ctx
->sg_plaintext_num_elem
;
319 maxpages
= ARRAY_SIZE(ctx
->sg_plaintext_data
) - num_elem
;
324 copied
= iov_iter_get_pages(from
, pages
,
332 iov_iter_advance(from
, copied
);
337 use
= min_t(int, copied
, PAGE_SIZE
- offset
);
339 sg_set_page(&ctx
->sg_plaintext_data
[num_elem
],
340 pages
[i
], use
, offset
);
341 sg_unmark_end(&ctx
->sg_plaintext_data
[num_elem
]);
342 sk_mem_charge(sk
, use
);
353 ctx
->sg_plaintext_size
= size
;
354 ctx
->sg_plaintext_num_elem
= num_elem
;
358 static int memcopy_from_iter(struct sock
*sk
, struct iov_iter
*from
,
361 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
362 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
363 struct scatterlist
*sg
= ctx
->sg_plaintext_data
;
366 for (i
= tls_ctx
->pending_open_record_frags
;
367 i
< ctx
->sg_plaintext_num_elem
; ++i
) {
370 page_address(sg_page(&sg
[i
])) + sg
[i
].offset
,
371 copy
, from
) != copy
) {
377 ++tls_ctx
->pending_open_record_frags
;
387 int tls_sw_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
)
389 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
390 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
393 long timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
394 bool eor
= !(msg
->msg_flags
& MSG_MORE
);
395 size_t try_to_copy
, copied
= 0;
396 unsigned char record_type
= TLS_RECORD_TYPE_DATA
;
401 if (msg
->msg_flags
& ~(MSG_MORE
| MSG_DONTWAIT
| MSG_NOSIGNAL
))
406 ret
= tls_complete_pending_work(sk
, tls_ctx
, msg
->msg_flags
, &timeo
);
410 if (unlikely(msg
->msg_controllen
)) {
411 ret
= tls_proccess_cmsg(sk
, msg
, &record_type
);
416 while (msg_data_left(msg
)) {
422 orig_size
= ctx
->sg_plaintext_size
;
424 try_to_copy
= msg_data_left(msg
);
425 record_room
= TLS_MAX_PAYLOAD_SIZE
- ctx
->sg_plaintext_size
;
426 if (try_to_copy
>= record_room
) {
427 try_to_copy
= record_room
;
431 required_size
= ctx
->sg_plaintext_size
+ try_to_copy
+
432 tls_ctx
->overhead_size
;
434 if (!sk_stream_memory_free(sk
))
435 goto wait_for_sndbuf
;
437 ret
= alloc_encrypted_sg(sk
, required_size
);
440 goto wait_for_memory
;
442 /* Adjust try_to_copy according to the amount that was
443 * actually allocated. The difference is due
444 * to max sg elements limit
446 try_to_copy
-= required_size
- ctx
->sg_encrypted_size
;
450 if (full_record
|| eor
) {
451 ret
= zerocopy_from_iter(sk
, &msg
->msg_iter
,
454 goto fallback_to_reg_send
;
456 copied
+= try_to_copy
;
457 ret
= tls_push_record(sk
, msg
->msg_flags
, record_type
);
463 copied
-= try_to_copy
;
464 fallback_to_reg_send
:
465 iov_iter_revert(&msg
->msg_iter
,
466 ctx
->sg_plaintext_size
- orig_size
);
467 trim_sg(sk
, ctx
->sg_plaintext_data
,
468 &ctx
->sg_plaintext_num_elem
,
469 &ctx
->sg_plaintext_size
,
473 required_size
= ctx
->sg_plaintext_size
+ try_to_copy
;
475 ret
= alloc_plaintext_sg(sk
, required_size
);
478 goto wait_for_memory
;
480 /* Adjust try_to_copy according to the amount that was
481 * actually allocated. The difference is due
482 * to max sg elements limit
484 try_to_copy
-= required_size
- ctx
->sg_plaintext_size
;
487 trim_sg(sk
, ctx
->sg_encrypted_data
,
488 &ctx
->sg_encrypted_num_elem
,
489 &ctx
->sg_encrypted_size
,
490 ctx
->sg_plaintext_size
+
491 tls_ctx
->overhead_size
);
494 ret
= memcopy_from_iter(sk
, &msg
->msg_iter
, try_to_copy
);
498 copied
+= try_to_copy
;
499 if (full_record
|| eor
) {
501 ret
= tls_push_record(sk
, msg
->msg_flags
, record_type
);
504 goto wait_for_memory
;
513 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
515 ret
= sk_stream_wait_memory(sk
, &timeo
);
518 trim_both_sgl(sk
, orig_size
);
522 if (tls_is_pending_closed_record(tls_ctx
))
525 if (ctx
->sg_encrypted_size
< required_size
)
526 goto alloc_encrypted
;
528 goto alloc_plaintext
;
532 ret
= sk_stream_error(sk
, msg
->msg_flags
, ret
);
535 return copied
? copied
: ret
;
538 int tls_sw_sendpage(struct sock
*sk
, struct page
*page
,
539 int offset
, size_t size
, int flags
)
541 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
542 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
544 long timeo
= sock_sndtimeo(sk
, flags
& MSG_DONTWAIT
);
546 size_t orig_size
= size
;
547 unsigned char record_type
= TLS_RECORD_TYPE_DATA
;
548 struct scatterlist
*sg
;
552 if (flags
& ~(MSG_MORE
| MSG_DONTWAIT
| MSG_NOSIGNAL
|
553 MSG_SENDPAGE_NOTLAST
))
556 /* No MSG_EOR from splice, only look at MSG_MORE */
557 eor
= !(flags
& (MSG_MORE
| MSG_SENDPAGE_NOTLAST
));
561 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
563 ret
= tls_complete_pending_work(sk
, tls_ctx
, flags
, &timeo
);
567 /* Call the sk_stream functions to manage the sndbuf mem. */
569 size_t copy
, required_size
;
577 record_room
= TLS_MAX_PAYLOAD_SIZE
- ctx
->sg_plaintext_size
;
579 if (copy
>= record_room
) {
583 required_size
= ctx
->sg_plaintext_size
+ copy
+
584 tls_ctx
->overhead_size
;
586 if (!sk_stream_memory_free(sk
))
587 goto wait_for_sndbuf
;
589 ret
= alloc_encrypted_sg(sk
, required_size
);
592 goto wait_for_memory
;
594 /* Adjust copy according to the amount that was
595 * actually allocated. The difference is due
596 * to max sg elements limit
598 copy
-= required_size
- ctx
->sg_plaintext_size
;
603 sg
= ctx
->sg_plaintext_data
+ ctx
->sg_plaintext_num_elem
;
604 sg_set_page(sg
, page
, copy
, offset
);
605 ctx
->sg_plaintext_num_elem
++;
607 sk_mem_charge(sk
, copy
);
610 ctx
->sg_plaintext_size
+= copy
;
611 tls_ctx
->pending_open_record_frags
= ctx
->sg_plaintext_num_elem
;
613 if (full_record
|| eor
||
614 ctx
->sg_plaintext_num_elem
==
615 ARRAY_SIZE(ctx
->sg_plaintext_data
)) {
617 ret
= tls_push_record(sk
, flags
, record_type
);
620 goto wait_for_memory
;
627 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
629 ret
= sk_stream_wait_memory(sk
, &timeo
);
631 trim_both_sgl(sk
, ctx
->sg_plaintext_size
);
635 if (tls_is_pending_closed_record(tls_ctx
))
642 if (orig_size
> size
)
643 ret
= orig_size
- size
;
645 ret
= sk_stream_error(sk
, flags
, ret
);
651 void tls_sw_free_tx_resources(struct sock
*sk
)
653 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
654 struct tls_sw_context
*ctx
= tls_sw_ctx(tls_ctx
);
657 crypto_free_aead(ctx
->aead_send
);
659 tls_free_both_sg(sk
);
664 int tls_set_sw_offload(struct sock
*sk
, struct tls_context
*ctx
)
666 struct tls_crypto_info
*crypto_info
;
667 struct tls12_crypto_info_aes_gcm_128
*gcm_128_info
;
668 struct tls_sw_context
*sw_ctx
;
669 u16 nonce_size
, tag_size
, iv_size
, rec_seq_size
;
683 sw_ctx
= kzalloc(sizeof(*sw_ctx
), GFP_KERNEL
);
689 ctx
->priv_ctx
= (struct tls_offload_context
*)sw_ctx
;
691 crypto_info
= &ctx
->crypto_send
.info
;
692 switch (crypto_info
->cipher_type
) {
693 case TLS_CIPHER_AES_GCM_128
: {
694 nonce_size
= TLS_CIPHER_AES_GCM_128_IV_SIZE
;
695 tag_size
= TLS_CIPHER_AES_GCM_128_TAG_SIZE
;
696 iv_size
= TLS_CIPHER_AES_GCM_128_IV_SIZE
;
697 iv
= ((struct tls12_crypto_info_aes_gcm_128
*)crypto_info
)->iv
;
698 rec_seq_size
= TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE
;
700 ((struct tls12_crypto_info_aes_gcm_128
*)crypto_info
)->rec_seq
;
702 (struct tls12_crypto_info_aes_gcm_128
*)crypto_info
;
710 ctx
->prepend_size
= TLS_HEADER_SIZE
+ nonce_size
;
711 ctx
->tag_size
= tag_size
;
712 ctx
->overhead_size
= ctx
->prepend_size
+ ctx
->tag_size
;
713 ctx
->iv_size
= iv_size
;
714 ctx
->iv
= kmalloc(iv_size
+ TLS_CIPHER_AES_GCM_128_SALT_SIZE
, GFP_KERNEL
);
719 memcpy(ctx
->iv
, gcm_128_info
->salt
, TLS_CIPHER_AES_GCM_128_SALT_SIZE
);
720 memcpy(ctx
->iv
+ TLS_CIPHER_AES_GCM_128_SALT_SIZE
, iv
, iv_size
);
721 ctx
->rec_seq_size
= rec_seq_size
;
722 ctx
->rec_seq
= kmalloc(rec_seq_size
, GFP_KERNEL
);
727 memcpy(ctx
->rec_seq
, rec_seq
, rec_seq_size
);
729 sg_init_table(sw_ctx
->sg_encrypted_data
,
730 ARRAY_SIZE(sw_ctx
->sg_encrypted_data
));
731 sg_init_table(sw_ctx
->sg_plaintext_data
,
732 ARRAY_SIZE(sw_ctx
->sg_plaintext_data
));
734 sg_init_table(sw_ctx
->sg_aead_in
, 2);
735 sg_set_buf(&sw_ctx
->sg_aead_in
[0], sw_ctx
->aad_space
,
736 sizeof(sw_ctx
->aad_space
));
737 sg_unmark_end(&sw_ctx
->sg_aead_in
[1]);
738 sg_chain(sw_ctx
->sg_aead_in
, 2, sw_ctx
->sg_plaintext_data
);
739 sg_init_table(sw_ctx
->sg_aead_out
, 2);
740 sg_set_buf(&sw_ctx
->sg_aead_out
[0], sw_ctx
->aad_space
,
741 sizeof(sw_ctx
->aad_space
));
742 sg_unmark_end(&sw_ctx
->sg_aead_out
[1]);
743 sg_chain(sw_ctx
->sg_aead_out
, 2, sw_ctx
->sg_encrypted_data
);
745 if (!sw_ctx
->aead_send
) {
746 sw_ctx
->aead_send
= crypto_alloc_aead("gcm(aes)", 0, 0);
747 if (IS_ERR(sw_ctx
->aead_send
)) {
748 rc
= PTR_ERR(sw_ctx
->aead_send
);
749 sw_ctx
->aead_send
= NULL
;
754 ctx
->push_pending_record
= tls_sw_push_pending_record
;
756 rc
= crypto_aead_setkey(sw_ctx
->aead_send
, gcm_128_info
->key
,
757 TLS_CIPHER_AES_GCM_128_KEY_SIZE
);
761 rc
= crypto_aead_setauthsize(sw_ctx
->aead_send
, ctx
->tag_size
);
766 crypto_free_aead(sw_ctx
->aead_send
);
767 sw_ctx
->aead_send
= NULL
;
775 kfree(ctx
->priv_ctx
);
776 ctx
->priv_ctx
= NULL
;