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.
7 * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
9 * This software is available to you under a choice of one of two
10 * licenses. You may choose to be licensed under the terms of the GNU
11 * General Public License (GPL) Version 2, available from the file
12 * COPYING in the main directory of this source tree, or the
13 * OpenIB.org BSD license below:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
23 * - Redistributions in binary form must reproduce the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer in the documentation and/or other materials
26 * provided with the distribution.
28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 #include <linux/bug.h>
39 #include <linux/sched/signal.h>
40 #include <linux/module.h>
41 #include <linux/kernel.h>
42 #include <linux/splice.h>
43 #include <crypto/aead.h>
45 #include <net/strparser.h>
47 #include <trace/events/sock.h>
51 struct tls_decrypt_arg
{
61 struct tls_decrypt_ctx
{
63 u8 iv
[TLS_MAX_IV_SIZE
];
64 u8 aad
[TLS_MAX_AAD_SIZE
];
66 struct scatterlist sg
[];
69 noinline
void tls_err_abort(struct sock
*sk
, int err
)
71 WARN_ON_ONCE(err
>= 0);
72 /* sk->sk_err should contain a positive error code. */
73 WRITE_ONCE(sk
->sk_err
, -err
);
74 /* Paired with smp_rmb() in tcp_poll() */
79 static int __skb_nsg(struct sk_buff
*skb
, int offset
, int len
,
80 unsigned int recursion_level
)
82 int start
= skb_headlen(skb
);
83 int i
, chunk
= start
- offset
;
84 struct sk_buff
*frag_iter
;
87 if (unlikely(recursion_level
>= 24))
100 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
103 WARN_ON(start
> offset
+ len
);
105 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
106 chunk
= end
- offset
;
119 if (unlikely(skb_has_frag_list(skb
))) {
120 skb_walk_frags(skb
, frag_iter
) {
123 WARN_ON(start
> offset
+ len
);
125 end
= start
+ frag_iter
->len
;
126 chunk
= end
- offset
;
130 ret
= __skb_nsg(frag_iter
, offset
- start
, chunk
,
131 recursion_level
+ 1);
132 if (unlikely(ret
< 0))
147 /* Return the number of scatterlist elements required to completely map the
148 * skb, or -EMSGSIZE if the recursion depth is exceeded.
150 static int skb_nsg(struct sk_buff
*skb
, int offset
, int len
)
152 return __skb_nsg(skb
, offset
, len
, 0);
155 static int tls_padding_length(struct tls_prot_info
*prot
, struct sk_buff
*skb
,
156 struct tls_decrypt_arg
*darg
)
158 struct strp_msg
*rxm
= strp_msg(skb
);
159 struct tls_msg
*tlm
= tls_msg(skb
);
162 /* Determine zero-padding length */
163 if (prot
->version
== TLS_1_3_VERSION
) {
164 int offset
= rxm
->full_len
- TLS_TAG_SIZE
- 1;
165 char content_type
= darg
->zc
? darg
->tail
: 0;
168 while (content_type
== 0) {
169 if (offset
< prot
->prepend_size
)
171 err
= skb_copy_bits(skb
, rxm
->offset
+ offset
,
180 tlm
->control
= content_type
;
185 static void tls_decrypt_done(void *data
, int err
)
187 struct aead_request
*aead_req
= data
;
188 struct crypto_aead
*aead
= crypto_aead_reqtfm(aead_req
);
189 struct scatterlist
*sgout
= aead_req
->dst
;
190 struct scatterlist
*sgin
= aead_req
->src
;
191 struct tls_sw_context_rx
*ctx
;
192 struct tls_decrypt_ctx
*dctx
;
193 struct tls_context
*tls_ctx
;
194 struct scatterlist
*sg
;
199 aead_size
= sizeof(*aead_req
) + crypto_aead_reqsize(aead
);
200 aead_size
= ALIGN(aead_size
, __alignof__(*dctx
));
201 dctx
= (void *)((u8
*)aead_req
+ aead_size
);
204 tls_ctx
= tls_get_ctx(sk
);
205 ctx
= tls_sw_ctx_rx(tls_ctx
);
207 /* Propagate if there was an err */
210 TLS_INC_STATS(sock_net(sk
), LINUX_MIB_TLSDECRYPTERROR
);
211 ctx
->async_wait
.err
= err
;
212 tls_err_abort(sk
, err
);
215 /* Free the destination pages if skb was not decrypted inplace */
217 /* Skip the first S/G entry as it points to AAD */
218 for_each_sg(sg_next(sgout
), sg
, UINT_MAX
, pages
) {
221 put_page(sg_page(sg
));
227 spin_lock_bh(&ctx
->decrypt_compl_lock
);
228 if (!atomic_dec_return(&ctx
->decrypt_pending
))
229 complete(&ctx
->async_wait
.completion
);
230 spin_unlock_bh(&ctx
->decrypt_compl_lock
);
233 static int tls_do_decryption(struct sock
*sk
,
234 struct scatterlist
*sgin
,
235 struct scatterlist
*sgout
,
238 struct aead_request
*aead_req
,
239 struct tls_decrypt_arg
*darg
)
241 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
242 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
243 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
246 aead_request_set_tfm(aead_req
, ctx
->aead_recv
);
247 aead_request_set_ad(aead_req
, prot
->aad_size
);
248 aead_request_set_crypt(aead_req
, sgin
, sgout
,
249 data_len
+ prot
->tag_size
,
253 aead_request_set_callback(aead_req
,
254 CRYPTO_TFM_REQ_MAY_BACKLOG
,
255 tls_decrypt_done
, aead_req
);
256 atomic_inc(&ctx
->decrypt_pending
);
258 aead_request_set_callback(aead_req
,
259 CRYPTO_TFM_REQ_MAY_BACKLOG
,
260 crypto_req_done
, &ctx
->async_wait
);
263 ret
= crypto_aead_decrypt(aead_req
);
264 if (ret
== -EINPROGRESS
) {
268 ret
= crypto_wait_req(ret
, &ctx
->async_wait
);
275 static void tls_trim_both_msgs(struct sock
*sk
, int target_size
)
277 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
278 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
279 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
280 struct tls_rec
*rec
= ctx
->open_rec
;
282 sk_msg_trim(sk
, &rec
->msg_plaintext
, target_size
);
284 target_size
+= prot
->overhead_size
;
285 sk_msg_trim(sk
, &rec
->msg_encrypted
, target_size
);
288 static int tls_alloc_encrypted_msg(struct sock
*sk
, int len
)
290 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
291 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
292 struct tls_rec
*rec
= ctx
->open_rec
;
293 struct sk_msg
*msg_en
= &rec
->msg_encrypted
;
295 return sk_msg_alloc(sk
, msg_en
, len
, 0);
298 static int tls_clone_plaintext_msg(struct sock
*sk
, int required
)
300 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
301 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
302 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
303 struct tls_rec
*rec
= ctx
->open_rec
;
304 struct sk_msg
*msg_pl
= &rec
->msg_plaintext
;
305 struct sk_msg
*msg_en
= &rec
->msg_encrypted
;
308 /* We add page references worth len bytes from encrypted sg
309 * at the end of plaintext sg. It is guaranteed that msg_en
310 * has enough required room (ensured by caller).
312 len
= required
- msg_pl
->sg
.size
;
314 /* Skip initial bytes in msg_en's data to be able to use
315 * same offset of both plain and encrypted data.
317 skip
= prot
->prepend_size
+ msg_pl
->sg
.size
;
319 return sk_msg_clone(sk
, msg_pl
, msg_en
, skip
, len
);
322 static struct tls_rec
*tls_get_rec(struct sock
*sk
)
324 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
325 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
326 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
327 struct sk_msg
*msg_pl
, *msg_en
;
331 mem_size
= sizeof(struct tls_rec
) + crypto_aead_reqsize(ctx
->aead_send
);
333 rec
= kzalloc(mem_size
, sk
->sk_allocation
);
337 msg_pl
= &rec
->msg_plaintext
;
338 msg_en
= &rec
->msg_encrypted
;
343 sg_init_table(rec
->sg_aead_in
, 2);
344 sg_set_buf(&rec
->sg_aead_in
[0], rec
->aad_space
, prot
->aad_size
);
345 sg_unmark_end(&rec
->sg_aead_in
[1]);
347 sg_init_table(rec
->sg_aead_out
, 2);
348 sg_set_buf(&rec
->sg_aead_out
[0], rec
->aad_space
, prot
->aad_size
);
349 sg_unmark_end(&rec
->sg_aead_out
[1]);
356 static void tls_free_rec(struct sock
*sk
, struct tls_rec
*rec
)
358 sk_msg_free(sk
, &rec
->msg_encrypted
);
359 sk_msg_free(sk
, &rec
->msg_plaintext
);
363 static void tls_free_open_rec(struct sock
*sk
)
365 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
366 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
367 struct tls_rec
*rec
= ctx
->open_rec
;
370 tls_free_rec(sk
, rec
);
371 ctx
->open_rec
= NULL
;
375 int tls_tx_records(struct sock
*sk
, int flags
)
377 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
378 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
379 struct tls_rec
*rec
, *tmp
;
380 struct sk_msg
*msg_en
;
381 int tx_flags
, rc
= 0;
383 if (tls_is_partially_sent_record(tls_ctx
)) {
384 rec
= list_first_entry(&ctx
->tx_list
,
385 struct tls_rec
, list
);
388 tx_flags
= rec
->tx_flags
;
392 rc
= tls_push_partial_record(sk
, tls_ctx
, tx_flags
);
396 /* Full record has been transmitted.
397 * Remove the head of tx_list
399 list_del(&rec
->list
);
400 sk_msg_free(sk
, &rec
->msg_plaintext
);
404 /* Tx all ready records */
405 list_for_each_entry_safe(rec
, tmp
, &ctx
->tx_list
, list
) {
406 if (READ_ONCE(rec
->tx_ready
)) {
408 tx_flags
= rec
->tx_flags
;
412 msg_en
= &rec
->msg_encrypted
;
413 rc
= tls_push_sg(sk
, tls_ctx
,
414 &msg_en
->sg
.data
[msg_en
->sg
.curr
],
419 list_del(&rec
->list
);
420 sk_msg_free(sk
, &rec
->msg_plaintext
);
428 if (rc
< 0 && rc
!= -EAGAIN
)
429 tls_err_abort(sk
, -EBADMSG
);
434 static void tls_encrypt_done(void *data
, int err
)
436 struct tls_sw_context_tx
*ctx
;
437 struct tls_context
*tls_ctx
;
438 struct tls_prot_info
*prot
;
439 struct tls_rec
*rec
= data
;
440 struct scatterlist
*sge
;
441 struct sk_msg
*msg_en
;
446 msg_en
= &rec
->msg_encrypted
;
449 tls_ctx
= tls_get_ctx(sk
);
450 prot
= &tls_ctx
->prot_info
;
451 ctx
= tls_sw_ctx_tx(tls_ctx
);
453 sge
= sk_msg_elem(msg_en
, msg_en
->sg
.curr
);
454 sge
->offset
-= prot
->prepend_size
;
455 sge
->length
+= prot
->prepend_size
;
457 /* Check if error is previously set on socket */
458 if (err
|| sk
->sk_err
) {
461 /* If err is already set on socket, return the same code */
463 ctx
->async_wait
.err
= -sk
->sk_err
;
465 ctx
->async_wait
.err
= err
;
466 tls_err_abort(sk
, err
);
471 struct tls_rec
*first_rec
;
473 /* Mark the record as ready for transmission */
474 smp_store_mb(rec
->tx_ready
, true);
476 /* If received record is at head of tx_list, schedule tx */
477 first_rec
= list_first_entry(&ctx
->tx_list
,
478 struct tls_rec
, list
);
479 if (rec
== first_rec
)
483 spin_lock_bh(&ctx
->encrypt_compl_lock
);
484 pending
= atomic_dec_return(&ctx
->encrypt_pending
);
486 if (!pending
&& ctx
->async_notify
)
487 complete(&ctx
->async_wait
.completion
);
488 spin_unlock_bh(&ctx
->encrypt_compl_lock
);
493 /* Schedule the transmission */
494 if (!test_and_set_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
))
495 schedule_delayed_work(&ctx
->tx_work
.work
, 1);
498 static int tls_do_encryption(struct sock
*sk
,
499 struct tls_context
*tls_ctx
,
500 struct tls_sw_context_tx
*ctx
,
501 struct aead_request
*aead_req
,
502 size_t data_len
, u32 start
)
504 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
505 struct tls_rec
*rec
= ctx
->open_rec
;
506 struct sk_msg
*msg_en
= &rec
->msg_encrypted
;
507 struct scatterlist
*sge
= sk_msg_elem(msg_en
, start
);
508 int rc
, iv_offset
= 0;
510 /* For CCM based ciphers, first byte of IV is a constant */
511 switch (prot
->cipher_type
) {
512 case TLS_CIPHER_AES_CCM_128
:
513 rec
->iv_data
[0] = TLS_AES_CCM_IV_B0_BYTE
;
516 case TLS_CIPHER_SM4_CCM
:
517 rec
->iv_data
[0] = TLS_SM4_CCM_IV_B0_BYTE
;
522 memcpy(&rec
->iv_data
[iv_offset
], tls_ctx
->tx
.iv
,
523 prot
->iv_size
+ prot
->salt_size
);
525 tls_xor_iv_with_seq(prot
, rec
->iv_data
+ iv_offset
,
526 tls_ctx
->tx
.rec_seq
);
528 sge
->offset
+= prot
->prepend_size
;
529 sge
->length
-= prot
->prepend_size
;
531 msg_en
->sg
.curr
= start
;
533 aead_request_set_tfm(aead_req
, ctx
->aead_send
);
534 aead_request_set_ad(aead_req
, prot
->aad_size
);
535 aead_request_set_crypt(aead_req
, rec
->sg_aead_in
,
537 data_len
, rec
->iv_data
);
539 aead_request_set_callback(aead_req
, CRYPTO_TFM_REQ_MAY_BACKLOG
,
540 tls_encrypt_done
, rec
);
542 /* Add the record in tx_list */
543 list_add_tail((struct list_head
*)&rec
->list
, &ctx
->tx_list
);
544 atomic_inc(&ctx
->encrypt_pending
);
546 rc
= crypto_aead_encrypt(aead_req
);
547 if (!rc
|| rc
!= -EINPROGRESS
) {
548 atomic_dec(&ctx
->encrypt_pending
);
549 sge
->offset
-= prot
->prepend_size
;
550 sge
->length
+= prot
->prepend_size
;
554 WRITE_ONCE(rec
->tx_ready
, true);
555 } else if (rc
!= -EINPROGRESS
) {
556 list_del(&rec
->list
);
560 /* Unhook the record from context if encryption is not failure */
561 ctx
->open_rec
= NULL
;
562 tls_advance_record_sn(sk
, prot
, &tls_ctx
->tx
);
566 static int tls_split_open_record(struct sock
*sk
, struct tls_rec
*from
,
567 struct tls_rec
**to
, struct sk_msg
*msg_opl
,
568 struct sk_msg
*msg_oen
, u32 split_point
,
569 u32 tx_overhead_size
, u32
*orig_end
)
571 u32 i
, j
, bytes
= 0, apply
= msg_opl
->apply_bytes
;
572 struct scatterlist
*sge
, *osge
, *nsge
;
573 u32 orig_size
= msg_opl
->sg
.size
;
574 struct scatterlist tmp
= { };
575 struct sk_msg
*msg_npl
;
579 new = tls_get_rec(sk
);
582 ret
= sk_msg_alloc(sk
, &new->msg_encrypted
, msg_opl
->sg
.size
+
583 tx_overhead_size
, 0);
585 tls_free_rec(sk
, new);
589 *orig_end
= msg_opl
->sg
.end
;
590 i
= msg_opl
->sg
.start
;
591 sge
= sk_msg_elem(msg_opl
, i
);
592 while (apply
&& sge
->length
) {
593 if (sge
->length
> apply
) {
594 u32 len
= sge
->length
- apply
;
596 get_page(sg_page(sge
));
597 sg_set_page(&tmp
, sg_page(sge
), len
,
598 sge
->offset
+ apply
);
603 apply
-= sge
->length
;
604 bytes
+= sge
->length
;
607 sk_msg_iter_var_next(i
);
608 if (i
== msg_opl
->sg
.end
)
610 sge
= sk_msg_elem(msg_opl
, i
);
614 msg_opl
->sg
.curr
= i
;
615 msg_opl
->sg
.copybreak
= 0;
616 msg_opl
->apply_bytes
= 0;
617 msg_opl
->sg
.size
= bytes
;
619 msg_npl
= &new->msg_plaintext
;
620 msg_npl
->apply_bytes
= apply
;
621 msg_npl
->sg
.size
= orig_size
- bytes
;
623 j
= msg_npl
->sg
.start
;
624 nsge
= sk_msg_elem(msg_npl
, j
);
626 memcpy(nsge
, &tmp
, sizeof(*nsge
));
627 sk_msg_iter_var_next(j
);
628 nsge
= sk_msg_elem(msg_npl
, j
);
631 osge
= sk_msg_elem(msg_opl
, i
);
632 while (osge
->length
) {
633 memcpy(nsge
, osge
, sizeof(*nsge
));
635 sk_msg_iter_var_next(i
);
636 sk_msg_iter_var_next(j
);
639 osge
= sk_msg_elem(msg_opl
, i
);
640 nsge
= sk_msg_elem(msg_npl
, j
);
644 msg_npl
->sg
.curr
= j
;
645 msg_npl
->sg
.copybreak
= 0;
651 static void tls_merge_open_record(struct sock
*sk
, struct tls_rec
*to
,
652 struct tls_rec
*from
, u32 orig_end
)
654 struct sk_msg
*msg_npl
= &from
->msg_plaintext
;
655 struct sk_msg
*msg_opl
= &to
->msg_plaintext
;
656 struct scatterlist
*osge
, *nsge
;
660 sk_msg_iter_var_prev(i
);
661 j
= msg_npl
->sg
.start
;
663 osge
= sk_msg_elem(msg_opl
, i
);
664 nsge
= sk_msg_elem(msg_npl
, j
);
666 if (sg_page(osge
) == sg_page(nsge
) &&
667 osge
->offset
+ osge
->length
== nsge
->offset
) {
668 osge
->length
+= nsge
->length
;
669 put_page(sg_page(nsge
));
672 msg_opl
->sg
.end
= orig_end
;
673 msg_opl
->sg
.curr
= orig_end
;
674 msg_opl
->sg
.copybreak
= 0;
675 msg_opl
->apply_bytes
= msg_opl
->sg
.size
+ msg_npl
->sg
.size
;
676 msg_opl
->sg
.size
+= msg_npl
->sg
.size
;
678 sk_msg_free(sk
, &to
->msg_encrypted
);
679 sk_msg_xfer_full(&to
->msg_encrypted
, &from
->msg_encrypted
);
684 static int tls_push_record(struct sock
*sk
, int flags
,
685 unsigned char record_type
)
687 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
688 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
689 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
690 struct tls_rec
*rec
= ctx
->open_rec
, *tmp
= NULL
;
691 u32 i
, split_point
, orig_end
;
692 struct sk_msg
*msg_pl
, *msg_en
;
693 struct aead_request
*req
;
700 msg_pl
= &rec
->msg_plaintext
;
701 msg_en
= &rec
->msg_encrypted
;
703 split_point
= msg_pl
->apply_bytes
;
704 split
= split_point
&& split_point
< msg_pl
->sg
.size
;
705 if (unlikely((!split
&&
707 prot
->overhead_size
> msg_en
->sg
.size
) ||
710 prot
->overhead_size
> msg_en
->sg
.size
))) {
712 split_point
= msg_en
->sg
.size
;
715 rc
= tls_split_open_record(sk
, rec
, &tmp
, msg_pl
, msg_en
,
716 split_point
, prot
->overhead_size
,
720 /* This can happen if above tls_split_open_record allocates
721 * a single large encryption buffer instead of two smaller
722 * ones. In this case adjust pointers and continue without
725 if (!msg_pl
->sg
.size
) {
726 tls_merge_open_record(sk
, rec
, tmp
, orig_end
);
727 msg_pl
= &rec
->msg_plaintext
;
728 msg_en
= &rec
->msg_encrypted
;
731 sk_msg_trim(sk
, msg_en
, msg_pl
->sg
.size
+
732 prot
->overhead_size
);
735 rec
->tx_flags
= flags
;
736 req
= &rec
->aead_req
;
739 sk_msg_iter_var_prev(i
);
741 rec
->content_type
= record_type
;
742 if (prot
->version
== TLS_1_3_VERSION
) {
743 /* Add content type to end of message. No padding added */
744 sg_set_buf(&rec
->sg_content_type
, &rec
->content_type
, 1);
745 sg_mark_end(&rec
->sg_content_type
);
746 sg_chain(msg_pl
->sg
.data
, msg_pl
->sg
.end
+ 1,
747 &rec
->sg_content_type
);
749 sg_mark_end(sk_msg_elem(msg_pl
, i
));
752 if (msg_pl
->sg
.end
< msg_pl
->sg
.start
) {
753 sg_chain(&msg_pl
->sg
.data
[msg_pl
->sg
.start
],
754 MAX_SKB_FRAGS
- msg_pl
->sg
.start
+ 1,
758 i
= msg_pl
->sg
.start
;
759 sg_chain(rec
->sg_aead_in
, 2, &msg_pl
->sg
.data
[i
]);
762 sk_msg_iter_var_prev(i
);
763 sg_mark_end(sk_msg_elem(msg_en
, i
));
765 i
= msg_en
->sg
.start
;
766 sg_chain(rec
->sg_aead_out
, 2, &msg_en
->sg
.data
[i
]);
768 tls_make_aad(rec
->aad_space
, msg_pl
->sg
.size
+ prot
->tail_size
,
769 tls_ctx
->tx
.rec_seq
, record_type
, prot
);
771 tls_fill_prepend(tls_ctx
,
772 page_address(sg_page(&msg_en
->sg
.data
[i
])) +
773 msg_en
->sg
.data
[i
].offset
,
774 msg_pl
->sg
.size
+ prot
->tail_size
,
777 tls_ctx
->pending_open_record_frags
= false;
779 rc
= tls_do_encryption(sk
, tls_ctx
, ctx
, req
,
780 msg_pl
->sg
.size
+ prot
->tail_size
, i
);
782 if (rc
!= -EINPROGRESS
) {
783 tls_err_abort(sk
, -EBADMSG
);
785 tls_ctx
->pending_open_record_frags
= true;
786 tls_merge_open_record(sk
, rec
, tmp
, orig_end
);
789 ctx
->async_capable
= 1;
792 msg_pl
= &tmp
->msg_plaintext
;
793 msg_en
= &tmp
->msg_encrypted
;
794 sk_msg_trim(sk
, msg_en
, msg_pl
->sg
.size
+ prot
->overhead_size
);
795 tls_ctx
->pending_open_record_frags
= true;
799 return tls_tx_records(sk
, flags
);
802 static int bpf_exec_tx_verdict(struct sk_msg
*msg
, struct sock
*sk
,
803 bool full_record
, u8 record_type
,
804 ssize_t
*copied
, int flags
)
806 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
807 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
808 struct sk_msg msg_redir
= { };
809 struct sk_psock
*psock
;
810 struct sock
*sk_redir
;
812 bool enospc
, policy
, redir_ingress
;
816 policy
= !(flags
& MSG_SENDPAGE_NOPOLICY
);
817 psock
= sk_psock_get(sk
);
818 if (!psock
|| !policy
) {
819 err
= tls_push_record(sk
, flags
, record_type
);
820 if (err
&& err
!= -EINPROGRESS
&& sk
->sk_err
== EBADMSG
) {
821 *copied
-= sk_msg_free(sk
, msg
);
822 tls_free_open_rec(sk
);
826 sk_psock_put(sk
, psock
);
830 enospc
= sk_msg_full(msg
);
831 if (psock
->eval
== __SK_NONE
) {
832 delta
= msg
->sg
.size
;
833 psock
->eval
= sk_psock_msg_verdict(sk
, psock
, msg
);
834 delta
-= msg
->sg
.size
;
836 if (msg
->cork_bytes
&& msg
->cork_bytes
> msg
->sg
.size
&&
837 !enospc
&& !full_record
) {
843 if (msg
->apply_bytes
&& msg
->apply_bytes
< send
)
844 send
= msg
->apply_bytes
;
846 switch (psock
->eval
) {
848 err
= tls_push_record(sk
, flags
, record_type
);
849 if (err
&& err
!= -EINPROGRESS
&& sk
->sk_err
== EBADMSG
) {
850 *copied
-= sk_msg_free(sk
, msg
);
851 tls_free_open_rec(sk
);
857 redir_ingress
= psock
->redir_ingress
;
858 sk_redir
= psock
->sk_redir
;
859 memcpy(&msg_redir
, msg
, sizeof(*msg
));
860 if (msg
->apply_bytes
< send
)
861 msg
->apply_bytes
= 0;
863 msg
->apply_bytes
-= send
;
864 sk_msg_return_zero(sk
, msg
, send
);
865 msg
->sg
.size
-= send
;
867 err
= tcp_bpf_sendmsg_redir(sk_redir
, redir_ingress
,
868 &msg_redir
, send
, flags
);
871 *copied
-= sk_msg_free_nocharge(sk
, &msg_redir
);
874 if (msg
->sg
.size
== 0)
875 tls_free_open_rec(sk
);
879 sk_msg_free_partial(sk
, msg
, send
);
880 if (msg
->apply_bytes
< send
)
881 msg
->apply_bytes
= 0;
883 msg
->apply_bytes
-= send
;
884 if (msg
->sg
.size
== 0)
885 tls_free_open_rec(sk
);
886 *copied
-= (send
+ delta
);
891 bool reset_eval
= !ctx
->open_rec
;
895 msg
= &rec
->msg_plaintext
;
896 if (!msg
->apply_bytes
)
900 psock
->eval
= __SK_NONE
;
901 if (psock
->sk_redir
) {
902 sock_put(psock
->sk_redir
);
903 psock
->sk_redir
= NULL
;
910 sk_psock_put(sk
, psock
);
914 static int tls_sw_push_pending_record(struct sock
*sk
, int flags
)
916 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
917 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
918 struct tls_rec
*rec
= ctx
->open_rec
;
919 struct sk_msg
*msg_pl
;
925 msg_pl
= &rec
->msg_plaintext
;
926 copied
= msg_pl
->sg
.size
;
930 return bpf_exec_tx_verdict(msg_pl
, sk
, true, TLS_RECORD_TYPE_DATA
,
934 static int tls_sw_sendmsg_splice(struct sock
*sk
, struct msghdr
*msg
,
935 struct sk_msg
*msg_pl
, size_t try_to_copy
,
938 struct page
*page
= NULL
, **pages
= &page
;
944 part
= iov_iter_extract_pages(&msg
->msg_iter
, &pages
,
945 try_to_copy
, 1, 0, &off
);
949 if (WARN_ON_ONCE(!sendpage_ok(page
))) {
950 iov_iter_revert(&msg
->msg_iter
, part
);
954 sk_msg_page_add(msg_pl
, page
, part
, off
);
955 msg_pl
->sg
.copybreak
= 0;
956 msg_pl
->sg
.curr
= msg_pl
->sg
.end
;
957 sk_mem_charge(sk
, part
);
960 } while (try_to_copy
&& !sk_msg_full(msg_pl
));
965 static int tls_sw_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
,
968 long timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
969 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
970 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
971 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
972 bool async_capable
= ctx
->async_capable
;
973 unsigned char record_type
= TLS_RECORD_TYPE_DATA
;
974 bool is_kvec
= iov_iter_is_kvec(&msg
->msg_iter
);
975 bool eor
= !(msg
->msg_flags
& MSG_MORE
);
978 struct sk_msg
*msg_pl
, *msg_en
;
989 if (!eor
&& (msg
->msg_flags
& MSG_EOR
))
992 if (unlikely(msg
->msg_controllen
)) {
993 ret
= tls_process_cmsg(sk
, msg
, &record_type
);
995 if (ret
== -EINPROGRESS
)
997 else if (ret
!= -EAGAIN
)
1002 while (msg_data_left(msg
)) {
1009 rec
= ctx
->open_rec
;
1011 rec
= ctx
->open_rec
= tls_get_rec(sk
);
1017 msg_pl
= &rec
->msg_plaintext
;
1018 msg_en
= &rec
->msg_encrypted
;
1020 orig_size
= msg_pl
->sg
.size
;
1021 full_record
= false;
1022 try_to_copy
= msg_data_left(msg
);
1023 record_room
= TLS_MAX_PAYLOAD_SIZE
- msg_pl
->sg
.size
;
1024 if (try_to_copy
>= record_room
) {
1025 try_to_copy
= record_room
;
1029 required_size
= msg_pl
->sg
.size
+ try_to_copy
+
1030 prot
->overhead_size
;
1032 if (!sk_stream_memory_free(sk
))
1033 goto wait_for_sndbuf
;
1036 ret
= tls_alloc_encrypted_msg(sk
, required_size
);
1039 goto wait_for_memory
;
1041 /* Adjust try_to_copy according to the amount that was
1042 * actually allocated. The difference is due
1043 * to max sg elements limit
1045 try_to_copy
-= required_size
- msg_en
->sg
.size
;
1049 if (try_to_copy
&& (msg
->msg_flags
& MSG_SPLICE_PAGES
)) {
1050 ret
= tls_sw_sendmsg_splice(sk
, msg
, msg_pl
,
1051 try_to_copy
, &copied
);
1054 tls_ctx
->pending_open_record_frags
= true;
1055 if (full_record
|| eor
|| sk_msg_full(msg_pl
))
1060 if (!is_kvec
&& (full_record
|| eor
) && !async_capable
) {
1061 u32 first
= msg_pl
->sg
.end
;
1063 ret
= sk_msg_zerocopy_from_iter(sk
, &msg
->msg_iter
,
1064 msg_pl
, try_to_copy
);
1066 goto fallback_to_reg_send
;
1069 copied
+= try_to_copy
;
1071 sk_msg_sg_copy_set(msg_pl
, first
);
1072 ret
= bpf_exec_tx_verdict(msg_pl
, sk
, full_record
,
1073 record_type
, &copied
,
1076 if (ret
== -EINPROGRESS
)
1078 else if (ret
== -ENOMEM
)
1079 goto wait_for_memory
;
1080 else if (ctx
->open_rec
&& ret
== -ENOSPC
)
1082 else if (ret
!= -EAGAIN
)
1087 copied
-= try_to_copy
;
1088 sk_msg_sg_copy_clear(msg_pl
, first
);
1089 iov_iter_revert(&msg
->msg_iter
,
1090 msg_pl
->sg
.size
- orig_size
);
1091 fallback_to_reg_send
:
1092 sk_msg_trim(sk
, msg_pl
, orig_size
);
1095 required_size
= msg_pl
->sg
.size
+ try_to_copy
;
1097 ret
= tls_clone_plaintext_msg(sk
, required_size
);
1102 /* Adjust try_to_copy according to the amount that was
1103 * actually allocated. The difference is due
1104 * to max sg elements limit
1106 try_to_copy
-= required_size
- msg_pl
->sg
.size
;
1108 sk_msg_trim(sk
, msg_en
,
1109 msg_pl
->sg
.size
+ prot
->overhead_size
);
1113 ret
= sk_msg_memcopy_from_iter(sk
, &msg
->msg_iter
,
1114 msg_pl
, try_to_copy
);
1119 /* Open records defined only if successfully copied, otherwise
1120 * we would trim the sg but not reset the open record frags.
1122 tls_ctx
->pending_open_record_frags
= true;
1123 copied
+= try_to_copy
;
1125 if (full_record
|| eor
) {
1126 ret
= bpf_exec_tx_verdict(msg_pl
, sk
, full_record
,
1127 record_type
, &copied
,
1130 if (ret
== -EINPROGRESS
)
1132 else if (ret
== -ENOMEM
)
1133 goto wait_for_memory
;
1134 else if (ret
!= -EAGAIN
) {
1145 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1147 ret
= sk_stream_wait_memory(sk
, &timeo
);
1151 tls_trim_both_msgs(sk
, orig_size
);
1155 if (ctx
->open_rec
&& msg_en
->sg
.size
< required_size
)
1156 goto alloc_encrypted
;
1161 } else if (num_zc
) {
1162 /* Wait for pending encryptions to get completed */
1163 spin_lock_bh(&ctx
->encrypt_compl_lock
);
1164 ctx
->async_notify
= true;
1166 pending
= atomic_read(&ctx
->encrypt_pending
);
1167 spin_unlock_bh(&ctx
->encrypt_compl_lock
);
1169 crypto_wait_req(-EINPROGRESS
, &ctx
->async_wait
);
1171 reinit_completion(&ctx
->async_wait
.completion
);
1173 /* There can be no concurrent accesses, since we have no
1174 * pending encrypt operations
1176 WRITE_ONCE(ctx
->async_notify
, false);
1178 if (ctx
->async_wait
.err
) {
1179 ret
= ctx
->async_wait
.err
;
1184 /* Transmit if any encryptions have completed */
1185 if (test_and_clear_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
)) {
1186 cancel_delayed_work(&ctx
->tx_work
.work
);
1187 tls_tx_records(sk
, msg
->msg_flags
);
1191 ret
= sk_stream_error(sk
, msg
->msg_flags
, ret
);
1192 return copied
> 0 ? copied
: ret
;
1195 int tls_sw_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
)
1197 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1200 if (msg
->msg_flags
& ~(MSG_MORE
| MSG_DONTWAIT
| MSG_NOSIGNAL
|
1201 MSG_CMSG_COMPAT
| MSG_SPLICE_PAGES
| MSG_EOR
|
1202 MSG_SENDPAGE_NOPOLICY
))
1205 ret
= mutex_lock_interruptible(&tls_ctx
->tx_lock
);
1209 ret
= tls_sw_sendmsg_locked(sk
, msg
, size
);
1211 mutex_unlock(&tls_ctx
->tx_lock
);
1216 * Handle unexpected EOF during splice without SPLICE_F_MORE set.
1218 void tls_sw_splice_eof(struct socket
*sock
)
1220 struct sock
*sk
= sock
->sk
;
1221 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1222 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
1223 struct tls_rec
*rec
;
1224 struct sk_msg
*msg_pl
;
1226 bool retrying
= false;
1233 mutex_lock(&tls_ctx
->tx_lock
);
1237 /* same checks as in tls_sw_push_pending_record() */
1238 rec
= ctx
->open_rec
;
1242 msg_pl
= &rec
->msg_plaintext
;
1243 if (msg_pl
->sg
.size
== 0)
1246 /* Check the BPF advisor and perform transmission. */
1247 ret
= bpf_exec_tx_verdict(msg_pl
, sk
, false, TLS_RECORD_TYPE_DATA
,
1262 /* Wait for pending encryptions to get completed */
1263 spin_lock_bh(&ctx
->encrypt_compl_lock
);
1264 ctx
->async_notify
= true;
1266 pending
= atomic_read(&ctx
->encrypt_pending
);
1267 spin_unlock_bh(&ctx
->encrypt_compl_lock
);
1269 crypto_wait_req(-EINPROGRESS
, &ctx
->async_wait
);
1271 reinit_completion(&ctx
->async_wait
.completion
);
1273 /* There can be no concurrent accesses, since we have no pending
1274 * encrypt operations
1276 WRITE_ONCE(ctx
->async_notify
, false);
1278 if (ctx
->async_wait
.err
)
1281 /* Transmit if any encryptions have completed */
1282 if (test_and_clear_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
)) {
1283 cancel_delayed_work(&ctx
->tx_work
.work
);
1284 tls_tx_records(sk
, 0);
1289 mutex_unlock(&tls_ctx
->tx_lock
);
1293 tls_rx_rec_wait(struct sock
*sk
, struct sk_psock
*psock
, bool nonblock
,
1296 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1297 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
1298 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
1302 timeo
= sock_rcvtimeo(sk
, nonblock
);
1304 while (!tls_strp_msg_ready(ctx
)) {
1305 if (!sk_psock_queue_empty(psock
))
1309 return sock_error(sk
);
1314 if (!skb_queue_empty(&sk
->sk_receive_queue
)) {
1315 tls_strp_check_rcv(&ctx
->strp
);
1316 if (tls_strp_msg_ready(ctx
))
1320 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
1323 if (sock_flag(sk
, SOCK_DONE
))
1330 add_wait_queue(sk_sleep(sk
), &wait
);
1331 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
1332 ret
= sk_wait_event(sk
, &timeo
,
1333 tls_strp_msg_ready(ctx
) ||
1334 !sk_psock_queue_empty(psock
),
1336 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
1337 remove_wait_queue(sk_sleep(sk
), &wait
);
1339 /* Handle signals */
1340 if (signal_pending(current
))
1341 return sock_intr_errno(timeo
);
1344 tls_strp_msg_load(&ctx
->strp
, released
);
1349 static int tls_setup_from_iter(struct iov_iter
*from
,
1350 int length
, int *pages_used
,
1351 struct scatterlist
*to
,
1354 int rc
= 0, i
= 0, num_elem
= *pages_used
, maxpages
;
1355 struct page
*pages
[MAX_SKB_FRAGS
];
1356 unsigned int size
= 0;
1357 ssize_t copied
, use
;
1360 while (length
> 0) {
1362 maxpages
= to_max_pages
- num_elem
;
1363 if (maxpages
== 0) {
1367 copied
= iov_iter_get_pages2(from
, pages
,
1378 use
= min_t(int, copied
, PAGE_SIZE
- offset
);
1380 sg_set_page(&to
[num_elem
],
1381 pages
[i
], use
, offset
);
1382 sg_unmark_end(&to
[num_elem
]);
1383 /* We do not uncharge memory from this API */
1392 /* Mark the end in the last sg entry if newly added */
1393 if (num_elem
> *pages_used
)
1394 sg_mark_end(&to
[num_elem
- 1]);
1397 iov_iter_revert(from
, size
);
1398 *pages_used
= num_elem
;
1403 static struct sk_buff
*
1404 tls_alloc_clrtxt_skb(struct sock
*sk
, struct sk_buff
*skb
,
1405 unsigned int full_len
)
1407 struct strp_msg
*clr_rxm
;
1408 struct sk_buff
*clr_skb
;
1411 clr_skb
= alloc_skb_with_frags(0, full_len
, TLS_PAGE_ORDER
,
1412 &err
, sk
->sk_allocation
);
1416 skb_copy_header(clr_skb
, skb
);
1417 clr_skb
->len
= full_len
;
1418 clr_skb
->data_len
= full_len
;
1420 clr_rxm
= strp_msg(clr_skb
);
1421 clr_rxm
->offset
= 0;
1428 * tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1429 * They must transform the darg in/out argument are as follows:
1431 * -------------------------------------------------------------------
1432 * zc | Zero-copy decrypt allowed | Zero-copy performed
1433 * async | Async decrypt allowed | Async crypto used / in progress
1434 * skb | * | Output skb
1436 * If ZC decryption was performed darg.skb will point to the input skb.
1439 /* This function decrypts the input skb into either out_iov or in out_sg
1440 * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1441 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1442 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1443 * NULL, then the decryption happens inside skb buffers itself, i.e.
1444 * zero-copy gets disabled and 'darg->zc' is updated.
1446 static int tls_decrypt_sg(struct sock
*sk
, struct iov_iter
*out_iov
,
1447 struct scatterlist
*out_sg
,
1448 struct tls_decrypt_arg
*darg
)
1450 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1451 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
1452 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
1453 int n_sgin
, n_sgout
, aead_size
, err
, pages
= 0;
1454 struct sk_buff
*skb
= tls_strp_msg(ctx
);
1455 const struct strp_msg
*rxm
= strp_msg(skb
);
1456 const struct tls_msg
*tlm
= tls_msg(skb
);
1457 struct aead_request
*aead_req
;
1458 struct scatterlist
*sgin
= NULL
;
1459 struct scatterlist
*sgout
= NULL
;
1460 const int data_len
= rxm
->full_len
- prot
->overhead_size
;
1461 int tail_pages
= !!prot
->tail_size
;
1462 struct tls_decrypt_ctx
*dctx
;
1463 struct sk_buff
*clear_skb
;
1467 n_sgin
= skb_nsg(skb
, rxm
->offset
+ prot
->prepend_size
,
1468 rxm
->full_len
- prot
->prepend_size
);
1470 return n_sgin
?: -EBADMSG
;
1472 if (darg
->zc
&& (out_iov
|| out_sg
)) {
1476 n_sgout
= 1 + tail_pages
+
1477 iov_iter_npages_cap(out_iov
, INT_MAX
, data_len
);
1479 n_sgout
= sg_nents(out_sg
);
1483 clear_skb
= tls_alloc_clrtxt_skb(sk
, skb
, rxm
->full_len
);
1487 n_sgout
= 1 + skb_shinfo(clear_skb
)->nr_frags
;
1490 /* Increment to accommodate AAD */
1491 n_sgin
= n_sgin
+ 1;
1493 /* Allocate a single block of memory which contains
1494 * aead_req || tls_decrypt_ctx.
1495 * Both structs are variable length.
1497 aead_size
= sizeof(*aead_req
) + crypto_aead_reqsize(ctx
->aead_recv
);
1498 aead_size
= ALIGN(aead_size
, __alignof__(*dctx
));
1499 mem
= kmalloc(aead_size
+ struct_size(dctx
, sg
, size_add(n_sgin
, n_sgout
)),
1506 /* Segment the allocated memory */
1507 aead_req
= (struct aead_request
*)mem
;
1508 dctx
= (struct tls_decrypt_ctx
*)(mem
+ aead_size
);
1510 sgin
= &dctx
->sg
[0];
1511 sgout
= &dctx
->sg
[n_sgin
];
1513 /* For CCM based ciphers, first byte of nonce+iv is a constant */
1514 switch (prot
->cipher_type
) {
1515 case TLS_CIPHER_AES_CCM_128
:
1516 dctx
->iv
[0] = TLS_AES_CCM_IV_B0_BYTE
;
1519 case TLS_CIPHER_SM4_CCM
:
1520 dctx
->iv
[0] = TLS_SM4_CCM_IV_B0_BYTE
;
1526 if (prot
->version
== TLS_1_3_VERSION
||
1527 prot
->cipher_type
== TLS_CIPHER_CHACHA20_POLY1305
) {
1528 memcpy(&dctx
->iv
[iv_offset
], tls_ctx
->rx
.iv
,
1529 prot
->iv_size
+ prot
->salt_size
);
1531 err
= skb_copy_bits(skb
, rxm
->offset
+ TLS_HEADER_SIZE
,
1532 &dctx
->iv
[iv_offset
] + prot
->salt_size
,
1536 memcpy(&dctx
->iv
[iv_offset
], tls_ctx
->rx
.iv
, prot
->salt_size
);
1538 tls_xor_iv_with_seq(prot
, &dctx
->iv
[iv_offset
], tls_ctx
->rx
.rec_seq
);
1541 tls_make_aad(dctx
->aad
, rxm
->full_len
- prot
->overhead_size
+
1543 tls_ctx
->rx
.rec_seq
, tlm
->control
, prot
);
1546 sg_init_table(sgin
, n_sgin
);
1547 sg_set_buf(&sgin
[0], dctx
->aad
, prot
->aad_size
);
1548 err
= skb_to_sgvec(skb
, &sgin
[1],
1549 rxm
->offset
+ prot
->prepend_size
,
1550 rxm
->full_len
- prot
->prepend_size
);
1555 sg_init_table(sgout
, n_sgout
);
1556 sg_set_buf(&sgout
[0], dctx
->aad
, prot
->aad_size
);
1558 err
= skb_to_sgvec(clear_skb
, &sgout
[1], prot
->prepend_size
,
1559 data_len
+ prot
->tail_size
);
1562 } else if (out_iov
) {
1563 sg_init_table(sgout
, n_sgout
);
1564 sg_set_buf(&sgout
[0], dctx
->aad
, prot
->aad_size
);
1566 err
= tls_setup_from_iter(out_iov
, data_len
, &pages
, &sgout
[1],
1567 (n_sgout
- 1 - tail_pages
));
1569 goto exit_free_pages
;
1571 if (prot
->tail_size
) {
1572 sg_unmark_end(&sgout
[pages
]);
1573 sg_set_buf(&sgout
[pages
+ 1], &dctx
->tail
,
1575 sg_mark_end(&sgout
[pages
+ 1]);
1577 } else if (out_sg
) {
1578 memcpy(sgout
, out_sg
, n_sgout
* sizeof(*sgout
));
1581 /* Prepare and submit AEAD request */
1582 err
= tls_do_decryption(sk
, sgin
, sgout
, dctx
->iv
,
1583 data_len
+ prot
->tail_size
, aead_req
, darg
);
1585 goto exit_free_pages
;
1587 darg
->skb
= clear_skb
?: tls_strp_msg(ctx
);
1590 if (unlikely(darg
->async
)) {
1591 err
= tls_strp_msg_hold(&ctx
->strp
, &ctx
->async_hold
);
1593 __skb_queue_tail(&ctx
->async_hold
, darg
->skb
);
1597 if (prot
->tail_size
)
1598 darg
->tail
= dctx
->tail
;
1601 /* Release the pages in case iov was mapped to pages */
1602 for (; pages
> 0; pages
--)
1603 put_page(sg_page(&sgout
[pages
]));
1607 consume_skb(clear_skb
);
1612 tls_decrypt_sw(struct sock
*sk
, struct tls_context
*tls_ctx
,
1613 struct msghdr
*msg
, struct tls_decrypt_arg
*darg
)
1615 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
1616 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
1617 struct strp_msg
*rxm
;
1620 err
= tls_decrypt_sg(sk
, &msg
->msg_iter
, NULL
, darg
);
1622 if (err
== -EBADMSG
)
1623 TLS_INC_STATS(sock_net(sk
), LINUX_MIB_TLSDECRYPTERROR
);
1626 /* keep going even for ->async, the code below is TLS 1.3 */
1628 /* If opportunistic TLS 1.3 ZC failed retry without ZC */
1629 if (unlikely(darg
->zc
&& prot
->version
== TLS_1_3_VERSION
&&
1630 darg
->tail
!= TLS_RECORD_TYPE_DATA
)) {
1633 TLS_INC_STATS(sock_net(sk
), LINUX_MIB_TLSRXNOPADVIOL
);
1634 TLS_INC_STATS(sock_net(sk
), LINUX_MIB_TLSDECRYPTRETRY
);
1635 return tls_decrypt_sw(sk
, tls_ctx
, msg
, darg
);
1638 pad
= tls_padding_length(prot
, darg
->skb
, darg
);
1640 if (darg
->skb
!= tls_strp_msg(ctx
))
1641 consume_skb(darg
->skb
);
1645 rxm
= strp_msg(darg
->skb
);
1646 rxm
->full_len
-= pad
;
1652 tls_decrypt_device(struct sock
*sk
, struct msghdr
*msg
,
1653 struct tls_context
*tls_ctx
, struct tls_decrypt_arg
*darg
)
1655 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
1656 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
1657 struct strp_msg
*rxm
;
1660 if (tls_ctx
->rx_conf
!= TLS_HW
)
1663 err
= tls_device_decrypted(sk
, tls_ctx
);
1667 pad
= tls_padding_length(prot
, tls_strp_msg(ctx
), darg
);
1671 darg
->async
= false;
1672 darg
->skb
= tls_strp_msg(ctx
);
1673 /* ->zc downgrade check, in case TLS 1.3 gets here */
1674 darg
->zc
&= !(prot
->version
== TLS_1_3_VERSION
&&
1675 tls_msg(darg
->skb
)->control
!= TLS_RECORD_TYPE_DATA
);
1677 rxm
= strp_msg(darg
->skb
);
1678 rxm
->full_len
-= pad
;
1681 /* Non-ZC case needs a real skb */
1682 darg
->skb
= tls_strp_msg_detach(ctx
);
1686 unsigned int off
, len
;
1688 /* In ZC case nobody cares about the output skb.
1689 * Just copy the data here. Note the skb is not fully trimmed.
1691 off
= rxm
->offset
+ prot
->prepend_size
;
1692 len
= rxm
->full_len
- prot
->overhead_size
;
1694 err
= skb_copy_datagram_msg(darg
->skb
, off
, msg
, len
);
1701 static int tls_rx_one_record(struct sock
*sk
, struct msghdr
*msg
,
1702 struct tls_decrypt_arg
*darg
)
1704 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1705 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
1706 struct strp_msg
*rxm
;
1709 err
= tls_decrypt_device(sk
, msg
, tls_ctx
, darg
);
1711 err
= tls_decrypt_sw(sk
, tls_ctx
, msg
, darg
);
1715 rxm
= strp_msg(darg
->skb
);
1716 rxm
->offset
+= prot
->prepend_size
;
1717 rxm
->full_len
-= prot
->overhead_size
;
1718 tls_advance_record_sn(sk
, prot
, &tls_ctx
->rx
);
1723 int decrypt_skb(struct sock
*sk
, struct scatterlist
*sgout
)
1725 struct tls_decrypt_arg darg
= { .zc
= true, };
1727 return tls_decrypt_sg(sk
, NULL
, sgout
, &darg
);
1730 static int tls_record_content_type(struct msghdr
*msg
, struct tls_msg
*tlm
,
1736 *control
= tlm
->control
;
1740 err
= put_cmsg(msg
, SOL_TLS
, TLS_GET_RECORD_TYPE
,
1741 sizeof(*control
), control
);
1742 if (*control
!= TLS_RECORD_TYPE_DATA
) {
1743 if (err
|| msg
->msg_flags
& MSG_CTRUNC
)
1746 } else if (*control
!= tlm
->control
) {
1753 static void tls_rx_rec_done(struct tls_sw_context_rx
*ctx
)
1755 tls_strp_msg_done(&ctx
->strp
);
1758 /* This function traverses the rx_list in tls receive context to copies the
1759 * decrypted records into the buffer provided by caller zero copy is not
1760 * true. Further, the records are removed from the rx_list if it is not a peek
1761 * case and the record has been consumed completely.
1763 static int process_rx_list(struct tls_sw_context_rx
*ctx
,
1770 struct sk_buff
*skb
= skb_peek(&ctx
->rx_list
);
1771 struct tls_msg
*tlm
;
1775 while (skip
&& skb
) {
1776 struct strp_msg
*rxm
= strp_msg(skb
);
1779 err
= tls_record_content_type(msg
, tlm
, control
);
1783 if (skip
< rxm
->full_len
)
1786 skip
= skip
- rxm
->full_len
;
1787 skb
= skb_peek_next(skb
, &ctx
->rx_list
);
1790 while (len
&& skb
) {
1791 struct sk_buff
*next_skb
;
1792 struct strp_msg
*rxm
= strp_msg(skb
);
1793 int chunk
= min_t(unsigned int, rxm
->full_len
- skip
, len
);
1797 err
= tls_record_content_type(msg
, tlm
, control
);
1801 err
= skb_copy_datagram_msg(skb
, rxm
->offset
+ skip
,
1807 copied
= copied
+ chunk
;
1809 /* Consume the data from record if it is non-peek case*/
1811 rxm
->offset
= rxm
->offset
+ chunk
;
1812 rxm
->full_len
= rxm
->full_len
- chunk
;
1814 /* Return if there is unconsumed data in the record */
1815 if (rxm
->full_len
- skip
)
1819 /* The remaining skip-bytes must lie in 1st record in rx_list.
1820 * So from the 2nd record, 'skip' should be 0.
1825 msg
->msg_flags
|= MSG_EOR
;
1827 next_skb
= skb_peek_next(skb
, &ctx
->rx_list
);
1830 __skb_unlink(skb
, &ctx
->rx_list
);
1839 return copied
? : err
;
1843 tls_read_flush_backlog(struct sock
*sk
, struct tls_prot_info
*prot
,
1844 size_t len_left
, size_t decrypted
, ssize_t done
,
1849 if (len_left
<= decrypted
)
1852 max_rec
= prot
->overhead_size
- prot
->tail_size
+ TLS_MAX_PAYLOAD_SIZE
;
1853 if (done
- *flushed_at
< SZ_128K
&& tcp_inq(sk
) > max_rec
)
1857 return sk_flush_backlog(sk
);
1860 static int tls_rx_reader_acquire(struct sock
*sk
, struct tls_sw_context_rx
*ctx
,
1866 timeo
= sock_rcvtimeo(sk
, nonblock
);
1868 while (unlikely(ctx
->reader_present
)) {
1869 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
1871 ctx
->reader_contended
= 1;
1873 add_wait_queue(&ctx
->wq
, &wait
);
1874 ret
= sk_wait_event(sk
, &timeo
,
1875 !READ_ONCE(ctx
->reader_present
), &wait
);
1876 remove_wait_queue(&ctx
->wq
, &wait
);
1880 if (signal_pending(current
))
1881 return sock_intr_errno(timeo
);
1886 WRITE_ONCE(ctx
->reader_present
, 1);
1891 static int tls_rx_reader_lock(struct sock
*sk
, struct tls_sw_context_rx
*ctx
,
1897 err
= tls_rx_reader_acquire(sk
, ctx
, nonblock
);
1903 static void tls_rx_reader_release(struct sock
*sk
, struct tls_sw_context_rx
*ctx
)
1905 if (unlikely(ctx
->reader_contended
)) {
1906 if (wq_has_sleeper(&ctx
->wq
))
1909 ctx
->reader_contended
= 0;
1911 WARN_ON_ONCE(!ctx
->reader_present
);
1914 WRITE_ONCE(ctx
->reader_present
, 0);
1917 static void tls_rx_reader_unlock(struct sock
*sk
, struct tls_sw_context_rx
*ctx
)
1919 tls_rx_reader_release(sk
, ctx
);
1923 int tls_sw_recvmsg(struct sock
*sk
,
1929 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
1930 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
1931 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
1932 ssize_t decrypted
= 0, async_copy_bytes
= 0;
1933 struct sk_psock
*psock
;
1934 unsigned char control
= 0;
1935 size_t flushed_at
= 0;
1936 struct strp_msg
*rxm
;
1937 struct tls_msg
*tlm
;
1941 bool is_kvec
= iov_iter_is_kvec(&msg
->msg_iter
);
1942 bool is_peek
= flags
& MSG_PEEK
;
1943 bool released
= true;
1944 bool bpf_strp_enabled
;
1947 if (unlikely(flags
& MSG_ERRQUEUE
))
1948 return sock_recv_errqueue(sk
, msg
, len
, SOL_IP
, IP_RECVERR
);
1950 psock
= sk_psock_get(sk
);
1951 err
= tls_rx_reader_lock(sk
, ctx
, flags
& MSG_DONTWAIT
);
1954 bpf_strp_enabled
= sk_psock_strp_enabled(psock
);
1956 /* If crypto failed the connection is broken */
1957 err
= ctx
->async_wait
.err
;
1961 /* Process pending decrypted records. It must be non-zero-copy */
1962 err
= process_rx_list(ctx
, msg
, &control
, 0, len
, is_peek
);
1970 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, len
);
1973 zc_capable
= !bpf_strp_enabled
&& !is_kvec
&& !is_peek
&&
1976 while (len
&& (decrypted
+ copied
< target
|| tls_strp_msg_ready(ctx
))) {
1977 struct tls_decrypt_arg darg
;
1978 int to_decrypt
, chunk
;
1980 err
= tls_rx_rec_wait(sk
, psock
, flags
& MSG_DONTWAIT
,
1984 chunk
= sk_msg_recvmsg(sk
, psock
, msg
, len
,
1995 memset(&darg
.inargs
, 0, sizeof(darg
.inargs
));
1997 rxm
= strp_msg(tls_strp_msg(ctx
));
1998 tlm
= tls_msg(tls_strp_msg(ctx
));
2000 to_decrypt
= rxm
->full_len
- prot
->overhead_size
;
2002 if (zc_capable
&& to_decrypt
<= len
&&
2003 tlm
->control
== TLS_RECORD_TYPE_DATA
)
2006 /* Do not use async mode if record is non-data */
2007 if (tlm
->control
== TLS_RECORD_TYPE_DATA
&& !bpf_strp_enabled
)
2008 darg
.async
= ctx
->async_capable
;
2012 err
= tls_rx_one_record(sk
, msg
, &darg
);
2014 tls_err_abort(sk
, -EBADMSG
);
2018 async
|= darg
.async
;
2020 /* If the type of records being processed is not known yet,
2021 * set it to record type just dequeued. If it is already known,
2022 * but does not match the record type just dequeued, go to end.
2023 * We always get record type here since for tls1.2, record type
2024 * is known just after record is dequeued from stream parser.
2025 * For tls1.3, we disable async.
2027 err
= tls_record_content_type(msg
, tls_msg(darg
.skb
), &control
);
2029 DEBUG_NET_WARN_ON_ONCE(darg
.zc
);
2030 tls_rx_rec_done(ctx
);
2032 __skb_queue_tail(&ctx
->rx_list
, darg
.skb
);
2036 /* periodically flush backlog, and feed strparser */
2037 released
= tls_read_flush_backlog(sk
, prot
, len
, to_decrypt
,
2041 /* TLS 1.3 may have updated the length by more than overhead */
2042 rxm
= strp_msg(darg
.skb
);
2043 chunk
= rxm
->full_len
;
2044 tls_rx_rec_done(ctx
);
2047 bool partially_consumed
= chunk
> len
;
2048 struct sk_buff
*skb
= darg
.skb
;
2050 DEBUG_NET_WARN_ON_ONCE(darg
.skb
== ctx
->strp
.anchor
);
2053 /* TLS 1.2-only, to_decrypt must be text len */
2054 chunk
= min_t(int, to_decrypt
, len
);
2055 async_copy_bytes
+= chunk
;
2059 __skb_queue_tail(&ctx
->rx_list
, skb
);
2063 if (bpf_strp_enabled
) {
2065 err
= sk_psock_tls_strp_read(psock
, skb
);
2066 if (err
!= __SK_PASS
) {
2067 rxm
->offset
= rxm
->offset
+ rxm
->full_len
;
2069 if (err
== __SK_DROP
)
2075 if (partially_consumed
)
2078 err
= skb_copy_datagram_msg(skb
, rxm
->offset
,
2081 goto put_on_rx_list_err
;
2084 goto put_on_rx_list
;
2086 if (partially_consumed
) {
2087 rxm
->offset
+= chunk
;
2088 rxm
->full_len
-= chunk
;
2089 goto put_on_rx_list
;
2098 /* Return full control message to userspace before trying
2099 * to parse another message type
2101 msg
->msg_flags
|= MSG_EOR
;
2102 if (control
!= TLS_RECORD_TYPE_DATA
)
2110 /* Wait for all previously submitted records to be decrypted */
2111 spin_lock_bh(&ctx
->decrypt_compl_lock
);
2112 reinit_completion(&ctx
->async_wait
.completion
);
2113 pending
= atomic_read(&ctx
->decrypt_pending
);
2114 spin_unlock_bh(&ctx
->decrypt_compl_lock
);
2117 ret
= crypto_wait_req(-EINPROGRESS
, &ctx
->async_wait
);
2118 __skb_queue_purge(&ctx
->async_hold
);
2121 if (err
>= 0 || err
== -EINPROGRESS
)
2127 /* Drain records from the rx_list & copy if required */
2128 if (is_peek
|| is_kvec
)
2129 err
= process_rx_list(ctx
, msg
, &control
, copied
,
2130 decrypted
, is_peek
);
2132 err
= process_rx_list(ctx
, msg
, &control
, 0,
2133 async_copy_bytes
, is_peek
);
2134 decrypted
+= max(err
, 0);
2137 copied
+= decrypted
;
2140 tls_rx_reader_unlock(sk
, ctx
);
2142 sk_psock_put(sk
, psock
);
2143 return copied
? : err
;
2146 ssize_t
tls_sw_splice_read(struct socket
*sock
, loff_t
*ppos
,
2147 struct pipe_inode_info
*pipe
,
2148 size_t len
, unsigned int flags
)
2150 struct tls_context
*tls_ctx
= tls_get_ctx(sock
->sk
);
2151 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2152 struct strp_msg
*rxm
= NULL
;
2153 struct sock
*sk
= sock
->sk
;
2154 struct tls_msg
*tlm
;
2155 struct sk_buff
*skb
;
2160 err
= tls_rx_reader_lock(sk
, ctx
, flags
& SPLICE_F_NONBLOCK
);
2164 if (!skb_queue_empty(&ctx
->rx_list
)) {
2165 skb
= __skb_dequeue(&ctx
->rx_list
);
2167 struct tls_decrypt_arg darg
;
2169 err
= tls_rx_rec_wait(sk
, NULL
, flags
& SPLICE_F_NONBLOCK
,
2172 goto splice_read_end
;
2174 memset(&darg
.inargs
, 0, sizeof(darg
.inargs
));
2176 err
= tls_rx_one_record(sk
, NULL
, &darg
);
2178 tls_err_abort(sk
, -EBADMSG
);
2179 goto splice_read_end
;
2182 tls_rx_rec_done(ctx
);
2186 rxm
= strp_msg(skb
);
2189 /* splice does not support reading control messages */
2190 if (tlm
->control
!= TLS_RECORD_TYPE_DATA
) {
2192 goto splice_requeue
;
2195 chunk
= min_t(unsigned int, rxm
->full_len
, len
);
2196 copied
= skb_splice_bits(skb
, sk
, rxm
->offset
, pipe
, chunk
, flags
);
2198 goto splice_requeue
;
2200 if (chunk
< rxm
->full_len
) {
2202 rxm
->full_len
-= len
;
2203 goto splice_requeue
;
2209 tls_rx_reader_unlock(sk
, ctx
);
2210 return copied
? : err
;
2213 __skb_queue_head(&ctx
->rx_list
, skb
);
2214 goto splice_read_end
;
2217 int tls_sw_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
2218 sk_read_actor_t read_actor
)
2220 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2221 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2222 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
2223 struct strp_msg
*rxm
= NULL
;
2224 struct sk_buff
*skb
= NULL
;
2225 struct sk_psock
*psock
;
2226 size_t flushed_at
= 0;
2227 bool released
= true;
2228 struct tls_msg
*tlm
;
2233 psock
= sk_psock_get(sk
);
2235 sk_psock_put(sk
, psock
);
2238 err
= tls_rx_reader_acquire(sk
, ctx
, true);
2242 /* If crypto failed the connection is broken */
2243 err
= ctx
->async_wait
.err
;
2249 if (!skb_queue_empty(&ctx
->rx_list
)) {
2250 skb
= __skb_dequeue(&ctx
->rx_list
);
2251 rxm
= strp_msg(skb
);
2254 struct tls_decrypt_arg darg
;
2256 err
= tls_rx_rec_wait(sk
, NULL
, true, released
);
2260 memset(&darg
.inargs
, 0, sizeof(darg
.inargs
));
2262 err
= tls_rx_one_record(sk
, NULL
, &darg
);
2264 tls_err_abort(sk
, -EBADMSG
);
2268 released
= tls_read_flush_backlog(sk
, prot
, INT_MAX
,
2272 rxm
= strp_msg(skb
);
2274 decrypted
+= rxm
->full_len
;
2276 tls_rx_rec_done(ctx
);
2279 /* read_sock does not support reading control messages */
2280 if (tlm
->control
!= TLS_RECORD_TYPE_DATA
) {
2282 goto read_sock_requeue
;
2285 used
= read_actor(desc
, skb
, rxm
->offset
, rxm
->full_len
);
2289 goto read_sock_requeue
;
2292 if (used
< rxm
->full_len
) {
2293 rxm
->offset
+= used
;
2294 rxm
->full_len
-= used
;
2296 goto read_sock_requeue
;
2305 tls_rx_reader_release(sk
, ctx
);
2306 return copied
? : err
;
2309 __skb_queue_head(&ctx
->rx_list
, skb
);
2313 bool tls_sw_sock_is_readable(struct sock
*sk
)
2315 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2316 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2317 bool ingress_empty
= true;
2318 struct sk_psock
*psock
;
2321 psock
= sk_psock(sk
);
2323 ingress_empty
= list_empty(&psock
->ingress_msg
);
2326 return !ingress_empty
|| tls_strp_msg_ready(ctx
) ||
2327 !skb_queue_empty(&ctx
->rx_list
);
2330 int tls_rx_msg_size(struct tls_strparser
*strp
, struct sk_buff
*skb
)
2332 struct tls_context
*tls_ctx
= tls_get_ctx(strp
->sk
);
2333 struct tls_prot_info
*prot
= &tls_ctx
->prot_info
;
2334 char header
[TLS_HEADER_SIZE
+ TLS_MAX_IV_SIZE
];
2335 size_t cipher_overhead
;
2336 size_t data_len
= 0;
2339 /* Verify that we have a full TLS header, or wait for more data */
2340 if (strp
->stm
.offset
+ prot
->prepend_size
> skb
->len
)
2343 /* Sanity-check size of on-stack buffer. */
2344 if (WARN_ON(prot
->prepend_size
> sizeof(header
))) {
2349 /* Linearize header to local buffer */
2350 ret
= skb_copy_bits(skb
, strp
->stm
.offset
, header
, prot
->prepend_size
);
2354 strp
->mark
= header
[0];
2356 data_len
= ((header
[4] & 0xFF) | (header
[3] << 8));
2358 cipher_overhead
= prot
->tag_size
;
2359 if (prot
->version
!= TLS_1_3_VERSION
&&
2360 prot
->cipher_type
!= TLS_CIPHER_CHACHA20_POLY1305
)
2361 cipher_overhead
+= prot
->iv_size
;
2363 if (data_len
> TLS_MAX_PAYLOAD_SIZE
+ cipher_overhead
+
2368 if (data_len
< cipher_overhead
) {
2373 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2374 if (header
[1] != TLS_1_2_VERSION_MINOR
||
2375 header
[2] != TLS_1_2_VERSION_MAJOR
) {
2380 tls_device_rx_resync_new_rec(strp
->sk
, data_len
+ TLS_HEADER_SIZE
,
2381 TCP_SKB_CB(skb
)->seq
+ strp
->stm
.offset
);
2382 return data_len
+ TLS_HEADER_SIZE
;
2385 tls_err_abort(strp
->sk
, ret
);
2390 void tls_rx_msg_ready(struct tls_strparser
*strp
)
2392 struct tls_sw_context_rx
*ctx
;
2394 ctx
= container_of(strp
, struct tls_sw_context_rx
, strp
);
2395 ctx
->saved_data_ready(strp
->sk
);
2398 static void tls_data_ready(struct sock
*sk
)
2400 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2401 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2402 struct sk_psock
*psock
;
2405 trace_sk_data_ready(sk
);
2407 alloc_save
= sk
->sk_allocation
;
2408 sk
->sk_allocation
= GFP_ATOMIC
;
2409 tls_strp_data_ready(&ctx
->strp
);
2410 sk
->sk_allocation
= alloc_save
;
2412 psock
= sk_psock_get(sk
);
2414 if (!list_empty(&psock
->ingress_msg
))
2415 ctx
->saved_data_ready(sk
);
2416 sk_psock_put(sk
, psock
);
2420 void tls_sw_cancel_work_tx(struct tls_context
*tls_ctx
)
2422 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
2424 set_bit(BIT_TX_CLOSING
, &ctx
->tx_bitmask
);
2425 set_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
);
2426 cancel_delayed_work_sync(&ctx
->tx_work
.work
);
2429 void tls_sw_release_resources_tx(struct sock
*sk
)
2431 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2432 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
2433 struct tls_rec
*rec
, *tmp
;
2436 /* Wait for any pending async encryptions to complete */
2437 spin_lock_bh(&ctx
->encrypt_compl_lock
);
2438 ctx
->async_notify
= true;
2439 pending
= atomic_read(&ctx
->encrypt_pending
);
2440 spin_unlock_bh(&ctx
->encrypt_compl_lock
);
2443 crypto_wait_req(-EINPROGRESS
, &ctx
->async_wait
);
2445 tls_tx_records(sk
, -1);
2447 /* Free up un-sent records in tx_list. First, free
2448 * the partially sent record if any at head of tx_list.
2450 if (tls_ctx
->partially_sent_record
) {
2451 tls_free_partial_record(sk
, tls_ctx
);
2452 rec
= list_first_entry(&ctx
->tx_list
,
2453 struct tls_rec
, list
);
2454 list_del(&rec
->list
);
2455 sk_msg_free(sk
, &rec
->msg_plaintext
);
2459 list_for_each_entry_safe(rec
, tmp
, &ctx
->tx_list
, list
) {
2460 list_del(&rec
->list
);
2461 sk_msg_free(sk
, &rec
->msg_encrypted
);
2462 sk_msg_free(sk
, &rec
->msg_plaintext
);
2466 crypto_free_aead(ctx
->aead_send
);
2467 tls_free_open_rec(sk
);
2470 void tls_sw_free_ctx_tx(struct tls_context
*tls_ctx
)
2472 struct tls_sw_context_tx
*ctx
= tls_sw_ctx_tx(tls_ctx
);
2477 void tls_sw_release_resources_rx(struct sock
*sk
)
2479 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2480 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2482 if (ctx
->aead_recv
) {
2483 __skb_queue_purge(&ctx
->rx_list
);
2484 crypto_free_aead(ctx
->aead_recv
);
2485 tls_strp_stop(&ctx
->strp
);
2486 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2487 * we still want to tls_strp_stop(), but sk->sk_data_ready was
2490 if (ctx
->saved_data_ready
) {
2491 write_lock_bh(&sk
->sk_callback_lock
);
2492 sk
->sk_data_ready
= ctx
->saved_data_ready
;
2493 write_unlock_bh(&sk
->sk_callback_lock
);
2498 void tls_sw_strparser_done(struct tls_context
*tls_ctx
)
2500 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2502 tls_strp_done(&ctx
->strp
);
2505 void tls_sw_free_ctx_rx(struct tls_context
*tls_ctx
)
2507 struct tls_sw_context_rx
*ctx
= tls_sw_ctx_rx(tls_ctx
);
2512 void tls_sw_free_resources_rx(struct sock
*sk
)
2514 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2516 tls_sw_release_resources_rx(sk
);
2517 tls_sw_free_ctx_rx(tls_ctx
);
2520 /* The work handler to transmitt the encrypted records in tx_list */
2521 static void tx_work_handler(struct work_struct
*work
)
2523 struct delayed_work
*delayed_work
= to_delayed_work(work
);
2524 struct tx_work
*tx_work
= container_of(delayed_work
,
2525 struct tx_work
, work
);
2526 struct sock
*sk
= tx_work
->sk
;
2527 struct tls_context
*tls_ctx
= tls_get_ctx(sk
);
2528 struct tls_sw_context_tx
*ctx
;
2530 if (unlikely(!tls_ctx
))
2533 ctx
= tls_sw_ctx_tx(tls_ctx
);
2534 if (test_bit(BIT_TX_CLOSING
, &ctx
->tx_bitmask
))
2537 if (!test_and_clear_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
))
2540 if (mutex_trylock(&tls_ctx
->tx_lock
)) {
2542 tls_tx_records(sk
, -1);
2544 mutex_unlock(&tls_ctx
->tx_lock
);
2545 } else if (!test_and_set_bit(BIT_TX_SCHEDULED
, &ctx
->tx_bitmask
)) {
2546 /* Someone is holding the tx_lock, they will likely run Tx
2547 * and cancel the work on their way out of the lock section.
2548 * Schedule a long delay just in case.
2550 schedule_delayed_work(&ctx
->tx_work
.work
, msecs_to_jiffies(10));
2554 static bool tls_is_tx_ready(struct tls_sw_context_tx
*ctx
)
2556 struct tls_rec
*rec
;
2558 rec
= list_first_entry_or_null(&ctx
->tx_list
, struct tls_rec
, list
);
2562 return READ_ONCE(rec
->tx_ready
);
2565 void tls_sw_write_space(struct sock
*sk
, struct tls_context
*ctx
)
2567 struct tls_sw_context_tx
*tx_ctx
= tls_sw_ctx_tx(ctx
);
2569 /* Schedule the transmission if tx list is ready */
2570 if (tls_is_tx_ready(tx_ctx
) &&
2571 !test_and_set_bit(BIT_TX_SCHEDULED
, &tx_ctx
->tx_bitmask
))
2572 schedule_delayed_work(&tx_ctx
->tx_work
.work
, 0);
2575 void tls_sw_strparser_arm(struct sock
*sk
, struct tls_context
*tls_ctx
)
2577 struct tls_sw_context_rx
*rx_ctx
= tls_sw_ctx_rx(tls_ctx
);
2579 write_lock_bh(&sk
->sk_callback_lock
);
2580 rx_ctx
->saved_data_ready
= sk
->sk_data_ready
;
2581 sk
->sk_data_ready
= tls_data_ready
;
2582 write_unlock_bh(&sk
->sk_callback_lock
);
2585 void tls_update_rx_zc_capable(struct tls_context
*tls_ctx
)
2587 struct tls_sw_context_rx
*rx_ctx
= tls_sw_ctx_rx(tls_ctx
);
2589 rx_ctx
->zc_capable
= tls_ctx
->rx_no_pad
||
2590 tls_ctx
->prot_info
.version
!= TLS_1_3_VERSION
;
2593 static struct tls_sw_context_tx
*init_ctx_tx(struct tls_context
*ctx
, struct sock
*sk
)
2595 struct tls_sw_context_tx
*sw_ctx_tx
;
2597 if (!ctx
->priv_ctx_tx
) {
2598 sw_ctx_tx
= kzalloc(sizeof(*sw_ctx_tx
), GFP_KERNEL
);
2602 sw_ctx_tx
= ctx
->priv_ctx_tx
;
2605 crypto_init_wait(&sw_ctx_tx
->async_wait
);
2606 spin_lock_init(&sw_ctx_tx
->encrypt_compl_lock
);
2607 INIT_LIST_HEAD(&sw_ctx_tx
->tx_list
);
2608 INIT_DELAYED_WORK(&sw_ctx_tx
->tx_work
.work
, tx_work_handler
);
2609 sw_ctx_tx
->tx_work
.sk
= sk
;
2614 static struct tls_sw_context_rx
*init_ctx_rx(struct tls_context
*ctx
)
2616 struct tls_sw_context_rx
*sw_ctx_rx
;
2618 if (!ctx
->priv_ctx_rx
) {
2619 sw_ctx_rx
= kzalloc(sizeof(*sw_ctx_rx
), GFP_KERNEL
);
2623 sw_ctx_rx
= ctx
->priv_ctx_rx
;
2626 crypto_init_wait(&sw_ctx_rx
->async_wait
);
2627 spin_lock_init(&sw_ctx_rx
->decrypt_compl_lock
);
2628 init_waitqueue_head(&sw_ctx_rx
->wq
);
2629 skb_queue_head_init(&sw_ctx_rx
->rx_list
);
2630 skb_queue_head_init(&sw_ctx_rx
->async_hold
);
2635 int init_prot_info(struct tls_prot_info
*prot
,
2636 const struct tls_crypto_info
*crypto_info
,
2637 const struct tls_cipher_desc
*cipher_desc
)
2639 u16 nonce_size
= cipher_desc
->nonce
;
2641 if (crypto_info
->version
== TLS_1_3_VERSION
) {
2643 prot
->aad_size
= TLS_HEADER_SIZE
;
2644 prot
->tail_size
= 1;
2646 prot
->aad_size
= TLS_AAD_SPACE_SIZE
;
2647 prot
->tail_size
= 0;
2650 /* Sanity-check the sizes for stack allocations. */
2651 if (nonce_size
> TLS_MAX_IV_SIZE
|| prot
->aad_size
> TLS_MAX_AAD_SIZE
)
2654 prot
->version
= crypto_info
->version
;
2655 prot
->cipher_type
= crypto_info
->cipher_type
;
2656 prot
->prepend_size
= TLS_HEADER_SIZE
+ nonce_size
;
2657 prot
->tag_size
= cipher_desc
->tag
;
2658 prot
->overhead_size
= prot
->prepend_size
+ prot
->tag_size
+ prot
->tail_size
;
2659 prot
->iv_size
= cipher_desc
->iv
;
2660 prot
->salt_size
= cipher_desc
->salt
;
2661 prot
->rec_seq_size
= cipher_desc
->rec_seq
;
2666 int tls_set_sw_offload(struct sock
*sk
, int tx
)
2668 struct tls_sw_context_tx
*sw_ctx_tx
= NULL
;
2669 struct tls_sw_context_rx
*sw_ctx_rx
= NULL
;
2670 const struct tls_cipher_desc
*cipher_desc
;
2671 struct tls_crypto_info
*crypto_info
;
2672 char *iv
, *rec_seq
, *key
, *salt
;
2673 struct cipher_context
*cctx
;
2674 struct tls_prot_info
*prot
;
2675 struct crypto_aead
**aead
;
2676 struct tls_context
*ctx
;
2677 struct crypto_tfm
*tfm
;
2680 ctx
= tls_get_ctx(sk
);
2681 prot
= &ctx
->prot_info
;
2684 ctx
->priv_ctx_tx
= init_ctx_tx(ctx
, sk
);
2685 if (!ctx
->priv_ctx_tx
)
2688 sw_ctx_tx
= ctx
->priv_ctx_tx
;
2689 crypto_info
= &ctx
->crypto_send
.info
;
2691 aead
= &sw_ctx_tx
->aead_send
;
2693 ctx
->priv_ctx_rx
= init_ctx_rx(ctx
);
2694 if (!ctx
->priv_ctx_rx
)
2697 sw_ctx_rx
= ctx
->priv_ctx_rx
;
2698 crypto_info
= &ctx
->crypto_recv
.info
;
2700 aead
= &sw_ctx_rx
->aead_recv
;
2703 cipher_desc
= get_cipher_desc(crypto_info
->cipher_type
);
2709 rc
= init_prot_info(prot
, crypto_info
, cipher_desc
);
2713 iv
= crypto_info_iv(crypto_info
, cipher_desc
);
2714 key
= crypto_info_key(crypto_info
, cipher_desc
);
2715 salt
= crypto_info_salt(crypto_info
, cipher_desc
);
2716 rec_seq
= crypto_info_rec_seq(crypto_info
, cipher_desc
);
2718 memcpy(cctx
->iv
, salt
, cipher_desc
->salt
);
2719 memcpy(cctx
->iv
+ cipher_desc
->salt
, iv
, cipher_desc
->iv
);
2720 memcpy(cctx
->rec_seq
, rec_seq
, cipher_desc
->rec_seq
);
2723 *aead
= crypto_alloc_aead(cipher_desc
->cipher_name
, 0, 0);
2724 if (IS_ERR(*aead
)) {
2725 rc
= PTR_ERR(*aead
);
2731 ctx
->push_pending_record
= tls_sw_push_pending_record
;
2733 rc
= crypto_aead_setkey(*aead
, key
, cipher_desc
->key
);
2737 rc
= crypto_aead_setauthsize(*aead
, prot
->tag_size
);
2742 tfm
= crypto_aead_tfm(sw_ctx_rx
->aead_recv
);
2744 tls_update_rx_zc_capable(ctx
);
2745 sw_ctx_rx
->async_capable
=
2746 crypto_info
->version
!= TLS_1_3_VERSION
&&
2747 !!(tfm
->__crt_alg
->cra_flags
& CRYPTO_ALG_ASYNC
);
2749 rc
= tls_strp_init(&sw_ctx_rx
->strp
, sk
);
2757 crypto_free_aead(*aead
);
2761 kfree(ctx
->priv_ctx_tx
);
2762 ctx
->priv_ctx_tx
= NULL
;
2764 kfree(ctx
->priv_ctx_rx
);
2765 ctx
->priv_ctx_rx
= NULL
;