xref: /openbmc/linux/net/tls/tls_sw.c (revision 8bd1369b)
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  * 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  *
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:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
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.
26  *
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
34  * SOFTWARE.
35  */
36 
37 #include <linux/sched/signal.h>
38 #include <linux/module.h>
39 #include <crypto/aead.h>
40 
41 #include <net/strparser.h>
42 #include <net/tls.h>
43 
44 #define MAX_IV_SIZE	TLS_CIPHER_AES_GCM_128_IV_SIZE
45 
46 static int tls_do_decryption(struct sock *sk,
47 			     struct scatterlist *sgin,
48 			     struct scatterlist *sgout,
49 			     char *iv_recv,
50 			     size_t data_len,
51 			     struct sk_buff *skb,
52 			     gfp_t flags)
53 {
54 	struct tls_context *tls_ctx = tls_get_ctx(sk);
55 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
56 	struct strp_msg *rxm = strp_msg(skb);
57 	struct aead_request *aead_req;
58 
59 	int ret;
60 	unsigned int req_size = sizeof(struct aead_request) +
61 		crypto_aead_reqsize(ctx->aead_recv);
62 
63 	aead_req = kzalloc(req_size, flags);
64 	if (!aead_req)
65 		return -ENOMEM;
66 
67 	aead_request_set_tfm(aead_req, ctx->aead_recv);
68 	aead_request_set_ad(aead_req, TLS_AAD_SPACE_SIZE);
69 	aead_request_set_crypt(aead_req, sgin, sgout,
70 			       data_len + tls_ctx->rx.tag_size,
71 			       (u8 *)iv_recv);
72 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
73 				  crypto_req_done, &ctx->async_wait);
74 
75 	ret = crypto_wait_req(crypto_aead_decrypt(aead_req), &ctx->async_wait);
76 
77 	if (ret < 0)
78 		goto out;
79 
80 	rxm->offset += tls_ctx->rx.prepend_size;
81 	rxm->full_len -= tls_ctx->rx.overhead_size;
82 	tls_advance_record_sn(sk, &tls_ctx->rx);
83 
84 	ctx->decrypted = true;
85 
86 	ctx->saved_data_ready(sk);
87 
88 out:
89 	kfree(aead_req);
90 	return ret;
91 }
92 
93 static void trim_sg(struct sock *sk, struct scatterlist *sg,
94 		    int *sg_num_elem, unsigned int *sg_size, int target_size)
95 {
96 	int i = *sg_num_elem - 1;
97 	int trim = *sg_size - target_size;
98 
99 	if (trim <= 0) {
100 		WARN_ON(trim < 0);
101 		return;
102 	}
103 
104 	*sg_size = target_size;
105 	while (trim >= sg[i].length) {
106 		trim -= sg[i].length;
107 		sk_mem_uncharge(sk, sg[i].length);
108 		put_page(sg_page(&sg[i]));
109 		i--;
110 
111 		if (i < 0)
112 			goto out;
113 	}
114 
115 	sg[i].length -= trim;
116 	sk_mem_uncharge(sk, trim);
117 
118 out:
119 	*sg_num_elem = i + 1;
120 }
121 
122 static void trim_both_sgl(struct sock *sk, int target_size)
123 {
124 	struct tls_context *tls_ctx = tls_get_ctx(sk);
125 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
126 
127 	trim_sg(sk, ctx->sg_plaintext_data,
128 		&ctx->sg_plaintext_num_elem,
129 		&ctx->sg_plaintext_size,
130 		target_size);
131 
132 	if (target_size > 0)
133 		target_size += tls_ctx->tx.overhead_size;
134 
135 	trim_sg(sk, ctx->sg_encrypted_data,
136 		&ctx->sg_encrypted_num_elem,
137 		&ctx->sg_encrypted_size,
138 		target_size);
139 }
140 
141 static int alloc_encrypted_sg(struct sock *sk, int len)
142 {
143 	struct tls_context *tls_ctx = tls_get_ctx(sk);
144 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
145 	int rc = 0;
146 
147 	rc = sk_alloc_sg(sk, len,
148 			 ctx->sg_encrypted_data, 0,
149 			 &ctx->sg_encrypted_num_elem,
150 			 &ctx->sg_encrypted_size, 0);
151 
152 	return rc;
153 }
154 
155 static int alloc_plaintext_sg(struct sock *sk, int len)
156 {
157 	struct tls_context *tls_ctx = tls_get_ctx(sk);
158 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
159 	int rc = 0;
160 
161 	rc = sk_alloc_sg(sk, len, ctx->sg_plaintext_data, 0,
162 			 &ctx->sg_plaintext_num_elem, &ctx->sg_plaintext_size,
163 			 tls_ctx->pending_open_record_frags);
164 
165 	return rc;
166 }
167 
168 static void free_sg(struct sock *sk, struct scatterlist *sg,
169 		    int *sg_num_elem, unsigned int *sg_size)
170 {
171 	int i, n = *sg_num_elem;
172 
173 	for (i = 0; i < n; ++i) {
174 		sk_mem_uncharge(sk, sg[i].length);
175 		put_page(sg_page(&sg[i]));
176 	}
177 	*sg_num_elem = 0;
178 	*sg_size = 0;
179 }
180 
181 static void tls_free_both_sg(struct sock *sk)
182 {
183 	struct tls_context *tls_ctx = tls_get_ctx(sk);
184 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
185 
186 	free_sg(sk, ctx->sg_encrypted_data, &ctx->sg_encrypted_num_elem,
187 		&ctx->sg_encrypted_size);
188 
189 	free_sg(sk, ctx->sg_plaintext_data, &ctx->sg_plaintext_num_elem,
190 		&ctx->sg_plaintext_size);
191 }
192 
193 static int tls_do_encryption(struct tls_context *tls_ctx,
194 			     struct tls_sw_context_tx *ctx,
195 			     struct aead_request *aead_req,
196 			     size_t data_len)
197 {
198 	int rc;
199 
200 	ctx->sg_encrypted_data[0].offset += tls_ctx->tx.prepend_size;
201 	ctx->sg_encrypted_data[0].length -= tls_ctx->tx.prepend_size;
202 
203 	aead_request_set_tfm(aead_req, ctx->aead_send);
204 	aead_request_set_ad(aead_req, TLS_AAD_SPACE_SIZE);
205 	aead_request_set_crypt(aead_req, ctx->sg_aead_in, ctx->sg_aead_out,
206 			       data_len, tls_ctx->tx.iv);
207 
208 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
209 				  crypto_req_done, &ctx->async_wait);
210 
211 	rc = crypto_wait_req(crypto_aead_encrypt(aead_req), &ctx->async_wait);
212 
213 	ctx->sg_encrypted_data[0].offset -= tls_ctx->tx.prepend_size;
214 	ctx->sg_encrypted_data[0].length += tls_ctx->tx.prepend_size;
215 
216 	return rc;
217 }
218 
219 static int tls_push_record(struct sock *sk, int flags,
220 			   unsigned char record_type)
221 {
222 	struct tls_context *tls_ctx = tls_get_ctx(sk);
223 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
224 	struct aead_request *req;
225 	int rc;
226 
227 	req = kzalloc(sizeof(struct aead_request) +
228 		      crypto_aead_reqsize(ctx->aead_send), sk->sk_allocation);
229 	if (!req)
230 		return -ENOMEM;
231 
232 	sg_mark_end(ctx->sg_plaintext_data + ctx->sg_plaintext_num_elem - 1);
233 	sg_mark_end(ctx->sg_encrypted_data + ctx->sg_encrypted_num_elem - 1);
234 
235 	tls_make_aad(ctx->aad_space, ctx->sg_plaintext_size,
236 		     tls_ctx->tx.rec_seq, tls_ctx->tx.rec_seq_size,
237 		     record_type);
238 
239 	tls_fill_prepend(tls_ctx,
240 			 page_address(sg_page(&ctx->sg_encrypted_data[0])) +
241 			 ctx->sg_encrypted_data[0].offset,
242 			 ctx->sg_plaintext_size, record_type);
243 
244 	tls_ctx->pending_open_record_frags = 0;
245 	set_bit(TLS_PENDING_CLOSED_RECORD, &tls_ctx->flags);
246 
247 	rc = tls_do_encryption(tls_ctx, ctx, req, ctx->sg_plaintext_size);
248 	if (rc < 0) {
249 		/* If we are called from write_space and
250 		 * we fail, we need to set this SOCK_NOSPACE
251 		 * to trigger another write_space in the future.
252 		 */
253 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
254 		goto out_req;
255 	}
256 
257 	free_sg(sk, ctx->sg_plaintext_data, &ctx->sg_plaintext_num_elem,
258 		&ctx->sg_plaintext_size);
259 
260 	ctx->sg_encrypted_num_elem = 0;
261 	ctx->sg_encrypted_size = 0;
262 
263 	/* Only pass through MSG_DONTWAIT and MSG_NOSIGNAL flags */
264 	rc = tls_push_sg(sk, tls_ctx, ctx->sg_encrypted_data, 0, flags);
265 	if (rc < 0 && rc != -EAGAIN)
266 		tls_err_abort(sk, EBADMSG);
267 
268 	tls_advance_record_sn(sk, &tls_ctx->tx);
269 out_req:
270 	kfree(req);
271 	return rc;
272 }
273 
274 static int tls_sw_push_pending_record(struct sock *sk, int flags)
275 {
276 	return tls_push_record(sk, flags, TLS_RECORD_TYPE_DATA);
277 }
278 
279 static int zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
280 			      int length, int *pages_used,
281 			      unsigned int *size_used,
282 			      struct scatterlist *to, int to_max_pages,
283 			      bool charge)
284 {
285 	struct page *pages[MAX_SKB_FRAGS];
286 
287 	size_t offset;
288 	ssize_t copied, use;
289 	int i = 0;
290 	unsigned int size = *size_used;
291 	int num_elem = *pages_used;
292 	int rc = 0;
293 	int maxpages;
294 
295 	while (length > 0) {
296 		i = 0;
297 		maxpages = to_max_pages - num_elem;
298 		if (maxpages == 0) {
299 			rc = -EFAULT;
300 			goto out;
301 		}
302 		copied = iov_iter_get_pages(from, pages,
303 					    length,
304 					    maxpages, &offset);
305 		if (copied <= 0) {
306 			rc = -EFAULT;
307 			goto out;
308 		}
309 
310 		iov_iter_advance(from, copied);
311 
312 		length -= copied;
313 		size += copied;
314 		while (copied) {
315 			use = min_t(int, copied, PAGE_SIZE - offset);
316 
317 			sg_set_page(&to[num_elem],
318 				    pages[i], use, offset);
319 			sg_unmark_end(&to[num_elem]);
320 			if (charge)
321 				sk_mem_charge(sk, use);
322 
323 			offset = 0;
324 			copied -= use;
325 
326 			++i;
327 			++num_elem;
328 		}
329 	}
330 
331 out:
332 	*size_used = size;
333 	*pages_used = num_elem;
334 
335 	return rc;
336 }
337 
338 static int memcopy_from_iter(struct sock *sk, struct iov_iter *from,
339 			     int bytes)
340 {
341 	struct tls_context *tls_ctx = tls_get_ctx(sk);
342 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
343 	struct scatterlist *sg = ctx->sg_plaintext_data;
344 	int copy, i, rc = 0;
345 
346 	for (i = tls_ctx->pending_open_record_frags;
347 	     i < ctx->sg_plaintext_num_elem; ++i) {
348 		copy = sg[i].length;
349 		if (copy_from_iter(
350 				page_address(sg_page(&sg[i])) + sg[i].offset,
351 				copy, from) != copy) {
352 			rc = -EFAULT;
353 			goto out;
354 		}
355 		bytes -= copy;
356 
357 		++tls_ctx->pending_open_record_frags;
358 
359 		if (!bytes)
360 			break;
361 	}
362 
363 out:
364 	return rc;
365 }
366 
367 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
368 {
369 	struct tls_context *tls_ctx = tls_get_ctx(sk);
370 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
371 	int ret = 0;
372 	int required_size;
373 	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
374 	bool eor = !(msg->msg_flags & MSG_MORE);
375 	size_t try_to_copy, copied = 0;
376 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
377 	int record_room;
378 	bool full_record;
379 	int orig_size;
380 
381 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
382 		return -ENOTSUPP;
383 
384 	lock_sock(sk);
385 
386 	if (tls_complete_pending_work(sk, tls_ctx, msg->msg_flags, &timeo))
387 		goto send_end;
388 
389 	if (unlikely(msg->msg_controllen)) {
390 		ret = tls_proccess_cmsg(sk, msg, &record_type);
391 		if (ret)
392 			goto send_end;
393 	}
394 
395 	while (msg_data_left(msg)) {
396 		if (sk->sk_err) {
397 			ret = -sk->sk_err;
398 			goto send_end;
399 		}
400 
401 		orig_size = ctx->sg_plaintext_size;
402 		full_record = false;
403 		try_to_copy = msg_data_left(msg);
404 		record_room = TLS_MAX_PAYLOAD_SIZE - ctx->sg_plaintext_size;
405 		if (try_to_copy >= record_room) {
406 			try_to_copy = record_room;
407 			full_record = true;
408 		}
409 
410 		required_size = ctx->sg_plaintext_size + try_to_copy +
411 				tls_ctx->tx.overhead_size;
412 
413 		if (!sk_stream_memory_free(sk))
414 			goto wait_for_sndbuf;
415 alloc_encrypted:
416 		ret = alloc_encrypted_sg(sk, required_size);
417 		if (ret) {
418 			if (ret != -ENOSPC)
419 				goto wait_for_memory;
420 
421 			/* Adjust try_to_copy according to the amount that was
422 			 * actually allocated. The difference is due
423 			 * to max sg elements limit
424 			 */
425 			try_to_copy -= required_size - ctx->sg_encrypted_size;
426 			full_record = true;
427 		}
428 
429 		if (full_record || eor) {
430 			ret = zerocopy_from_iter(sk, &msg->msg_iter,
431 				try_to_copy, &ctx->sg_plaintext_num_elem,
432 				&ctx->sg_plaintext_size,
433 				ctx->sg_plaintext_data,
434 				ARRAY_SIZE(ctx->sg_plaintext_data),
435 				true);
436 			if (ret)
437 				goto fallback_to_reg_send;
438 
439 			copied += try_to_copy;
440 			ret = tls_push_record(sk, msg->msg_flags, record_type);
441 			if (!ret)
442 				continue;
443 			if (ret < 0)
444 				goto send_end;
445 
446 			copied -= try_to_copy;
447 fallback_to_reg_send:
448 			iov_iter_revert(&msg->msg_iter,
449 					ctx->sg_plaintext_size - orig_size);
450 			trim_sg(sk, ctx->sg_plaintext_data,
451 				&ctx->sg_plaintext_num_elem,
452 				&ctx->sg_plaintext_size,
453 				orig_size);
454 		}
455 
456 		required_size = ctx->sg_plaintext_size + try_to_copy;
457 alloc_plaintext:
458 		ret = alloc_plaintext_sg(sk, required_size);
459 		if (ret) {
460 			if (ret != -ENOSPC)
461 				goto wait_for_memory;
462 
463 			/* Adjust try_to_copy according to the amount that was
464 			 * actually allocated. The difference is due
465 			 * to max sg elements limit
466 			 */
467 			try_to_copy -= required_size - ctx->sg_plaintext_size;
468 			full_record = true;
469 
470 			trim_sg(sk, ctx->sg_encrypted_data,
471 				&ctx->sg_encrypted_num_elem,
472 				&ctx->sg_encrypted_size,
473 				ctx->sg_plaintext_size +
474 				tls_ctx->tx.overhead_size);
475 		}
476 
477 		ret = memcopy_from_iter(sk, &msg->msg_iter, try_to_copy);
478 		if (ret)
479 			goto trim_sgl;
480 
481 		copied += try_to_copy;
482 		if (full_record || eor) {
483 push_record:
484 			ret = tls_push_record(sk, msg->msg_flags, record_type);
485 			if (ret) {
486 				if (ret == -ENOMEM)
487 					goto wait_for_memory;
488 
489 				goto send_end;
490 			}
491 		}
492 
493 		continue;
494 
495 wait_for_sndbuf:
496 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
497 wait_for_memory:
498 		ret = sk_stream_wait_memory(sk, &timeo);
499 		if (ret) {
500 trim_sgl:
501 			trim_both_sgl(sk, orig_size);
502 			goto send_end;
503 		}
504 
505 		if (tls_is_pending_closed_record(tls_ctx))
506 			goto push_record;
507 
508 		if (ctx->sg_encrypted_size < required_size)
509 			goto alloc_encrypted;
510 
511 		goto alloc_plaintext;
512 	}
513 
514 send_end:
515 	ret = sk_stream_error(sk, msg->msg_flags, ret);
516 
517 	release_sock(sk);
518 	return copied ? copied : ret;
519 }
520 
521 int tls_sw_sendpage(struct sock *sk, struct page *page,
522 		    int offset, size_t size, int flags)
523 {
524 	struct tls_context *tls_ctx = tls_get_ctx(sk);
525 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
526 	int ret = 0;
527 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
528 	bool eor;
529 	size_t orig_size = size;
530 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
531 	struct scatterlist *sg;
532 	bool full_record;
533 	int record_room;
534 
535 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
536 		      MSG_SENDPAGE_NOTLAST))
537 		return -ENOTSUPP;
538 
539 	/* No MSG_EOR from splice, only look at MSG_MORE */
540 	eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
541 
542 	lock_sock(sk);
543 
544 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
545 
546 	if (tls_complete_pending_work(sk, tls_ctx, flags, &timeo))
547 		goto sendpage_end;
548 
549 	/* Call the sk_stream functions to manage the sndbuf mem. */
550 	while (size > 0) {
551 		size_t copy, required_size;
552 
553 		if (sk->sk_err) {
554 			ret = -sk->sk_err;
555 			goto sendpage_end;
556 		}
557 
558 		full_record = false;
559 		record_room = TLS_MAX_PAYLOAD_SIZE - ctx->sg_plaintext_size;
560 		copy = size;
561 		if (copy >= record_room) {
562 			copy = record_room;
563 			full_record = true;
564 		}
565 		required_size = ctx->sg_plaintext_size + copy +
566 			      tls_ctx->tx.overhead_size;
567 
568 		if (!sk_stream_memory_free(sk))
569 			goto wait_for_sndbuf;
570 alloc_payload:
571 		ret = alloc_encrypted_sg(sk, required_size);
572 		if (ret) {
573 			if (ret != -ENOSPC)
574 				goto wait_for_memory;
575 
576 			/* Adjust copy according to the amount that was
577 			 * actually allocated. The difference is due
578 			 * to max sg elements limit
579 			 */
580 			copy -= required_size - ctx->sg_plaintext_size;
581 			full_record = true;
582 		}
583 
584 		get_page(page);
585 		sg = ctx->sg_plaintext_data + ctx->sg_plaintext_num_elem;
586 		sg_set_page(sg, page, copy, offset);
587 		sg_unmark_end(sg);
588 
589 		ctx->sg_plaintext_num_elem++;
590 
591 		sk_mem_charge(sk, copy);
592 		offset += copy;
593 		size -= copy;
594 		ctx->sg_plaintext_size += copy;
595 		tls_ctx->pending_open_record_frags = ctx->sg_plaintext_num_elem;
596 
597 		if (full_record || eor ||
598 		    ctx->sg_plaintext_num_elem ==
599 		    ARRAY_SIZE(ctx->sg_plaintext_data)) {
600 push_record:
601 			ret = tls_push_record(sk, flags, record_type);
602 			if (ret) {
603 				if (ret == -ENOMEM)
604 					goto wait_for_memory;
605 
606 				goto sendpage_end;
607 			}
608 		}
609 		continue;
610 wait_for_sndbuf:
611 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
612 wait_for_memory:
613 		ret = sk_stream_wait_memory(sk, &timeo);
614 		if (ret) {
615 			trim_both_sgl(sk, ctx->sg_plaintext_size);
616 			goto sendpage_end;
617 		}
618 
619 		if (tls_is_pending_closed_record(tls_ctx))
620 			goto push_record;
621 
622 		goto alloc_payload;
623 	}
624 
625 sendpage_end:
626 	if (orig_size > size)
627 		ret = orig_size - size;
628 	else
629 		ret = sk_stream_error(sk, flags, ret);
630 
631 	release_sock(sk);
632 	return ret;
633 }
634 
635 static struct sk_buff *tls_wait_data(struct sock *sk, int flags,
636 				     long timeo, int *err)
637 {
638 	struct tls_context *tls_ctx = tls_get_ctx(sk);
639 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
640 	struct sk_buff *skb;
641 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
642 
643 	while (!(skb = ctx->recv_pkt)) {
644 		if (sk->sk_err) {
645 			*err = sock_error(sk);
646 			return NULL;
647 		}
648 
649 		if (sock_flag(sk, SOCK_DONE))
650 			return NULL;
651 
652 		if ((flags & MSG_DONTWAIT) || !timeo) {
653 			*err = -EAGAIN;
654 			return NULL;
655 		}
656 
657 		add_wait_queue(sk_sleep(sk), &wait);
658 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
659 		sk_wait_event(sk, &timeo, ctx->recv_pkt != skb, &wait);
660 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
661 		remove_wait_queue(sk_sleep(sk), &wait);
662 
663 		/* Handle signals */
664 		if (signal_pending(current)) {
665 			*err = sock_intr_errno(timeo);
666 			return NULL;
667 		}
668 	}
669 
670 	return skb;
671 }
672 
673 static int decrypt_skb(struct sock *sk, struct sk_buff *skb,
674 		       struct scatterlist *sgout)
675 {
676 	struct tls_context *tls_ctx = tls_get_ctx(sk);
677 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
678 	char iv[TLS_CIPHER_AES_GCM_128_SALT_SIZE + MAX_IV_SIZE];
679 	struct scatterlist sgin_arr[MAX_SKB_FRAGS + 2];
680 	struct scatterlist *sgin = &sgin_arr[0];
681 	struct strp_msg *rxm = strp_msg(skb);
682 	int ret, nsg = ARRAY_SIZE(sgin_arr);
683 	struct sk_buff *unused;
684 
685 	ret = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
686 			    iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
687 			    tls_ctx->rx.iv_size);
688 	if (ret < 0)
689 		return ret;
690 
691 	memcpy(iv, tls_ctx->rx.iv, TLS_CIPHER_AES_GCM_128_SALT_SIZE);
692 	if (!sgout) {
693 		nsg = skb_cow_data(skb, 0, &unused) + 1;
694 		sgin = kmalloc_array(nsg, sizeof(*sgin), sk->sk_allocation);
695 		sgout = sgin;
696 	}
697 
698 	sg_init_table(sgin, nsg);
699 	sg_set_buf(&sgin[0], ctx->rx_aad_ciphertext, TLS_AAD_SPACE_SIZE);
700 
701 	nsg = skb_to_sgvec(skb, &sgin[1],
702 			   rxm->offset + tls_ctx->rx.prepend_size,
703 			   rxm->full_len - tls_ctx->rx.prepend_size);
704 	if (nsg < 0) {
705 		ret = nsg;
706 		goto out;
707 	}
708 
709 	tls_make_aad(ctx->rx_aad_ciphertext,
710 		     rxm->full_len - tls_ctx->rx.overhead_size,
711 		     tls_ctx->rx.rec_seq,
712 		     tls_ctx->rx.rec_seq_size,
713 		     ctx->control);
714 
715 	ret = tls_do_decryption(sk, sgin, sgout, iv,
716 				rxm->full_len - tls_ctx->rx.overhead_size,
717 				skb, sk->sk_allocation);
718 
719 out:
720 	if (sgin != &sgin_arr[0])
721 		kfree(sgin);
722 
723 	return ret;
724 }
725 
726 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
727 			       unsigned int len)
728 {
729 	struct tls_context *tls_ctx = tls_get_ctx(sk);
730 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
731 	struct strp_msg *rxm = strp_msg(skb);
732 
733 	if (len < rxm->full_len) {
734 		rxm->offset += len;
735 		rxm->full_len -= len;
736 
737 		return false;
738 	}
739 
740 	/* Finished with message */
741 	ctx->recv_pkt = NULL;
742 	kfree_skb(skb);
743 	__strp_unpause(&ctx->strp);
744 
745 	return true;
746 }
747 
748 int tls_sw_recvmsg(struct sock *sk,
749 		   struct msghdr *msg,
750 		   size_t len,
751 		   int nonblock,
752 		   int flags,
753 		   int *addr_len)
754 {
755 	struct tls_context *tls_ctx = tls_get_ctx(sk);
756 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
757 	unsigned char control;
758 	struct strp_msg *rxm;
759 	struct sk_buff *skb;
760 	ssize_t copied = 0;
761 	bool cmsg = false;
762 	int target, err = 0;
763 	long timeo;
764 
765 	flags |= nonblock;
766 
767 	if (unlikely(flags & MSG_ERRQUEUE))
768 		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
769 
770 	lock_sock(sk);
771 
772 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
773 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
774 	do {
775 		bool zc = false;
776 		int chunk = 0;
777 
778 		skb = tls_wait_data(sk, flags, timeo, &err);
779 		if (!skb)
780 			goto recv_end;
781 
782 		rxm = strp_msg(skb);
783 		if (!cmsg) {
784 			int cerr;
785 
786 			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
787 					sizeof(ctx->control), &ctx->control);
788 			cmsg = true;
789 			control = ctx->control;
790 			if (ctx->control != TLS_RECORD_TYPE_DATA) {
791 				if (cerr || msg->msg_flags & MSG_CTRUNC) {
792 					err = -EIO;
793 					goto recv_end;
794 				}
795 			}
796 		} else if (control != ctx->control) {
797 			goto recv_end;
798 		}
799 
800 		if (!ctx->decrypted) {
801 			int page_count;
802 			int to_copy;
803 
804 			page_count = iov_iter_npages(&msg->msg_iter,
805 						     MAX_SKB_FRAGS);
806 			to_copy = rxm->full_len - tls_ctx->rx.overhead_size;
807 			if (to_copy <= len && page_count < MAX_SKB_FRAGS &&
808 			    likely(!(flags & MSG_PEEK)))  {
809 				struct scatterlist sgin[MAX_SKB_FRAGS + 1];
810 				int pages = 0;
811 
812 				zc = true;
813 				sg_init_table(sgin, MAX_SKB_FRAGS + 1);
814 				sg_set_buf(&sgin[0], ctx->rx_aad_plaintext,
815 					   TLS_AAD_SPACE_SIZE);
816 
817 				err = zerocopy_from_iter(sk, &msg->msg_iter,
818 							 to_copy, &pages,
819 							 &chunk, &sgin[1],
820 							 MAX_SKB_FRAGS,	false);
821 				if (err < 0)
822 					goto fallback_to_reg_recv;
823 
824 				err = decrypt_skb(sk, skb, sgin);
825 				for (; pages > 0; pages--)
826 					put_page(sg_page(&sgin[pages]));
827 				if (err < 0) {
828 					tls_err_abort(sk, EBADMSG);
829 					goto recv_end;
830 				}
831 			} else {
832 fallback_to_reg_recv:
833 				err = decrypt_skb(sk, skb, NULL);
834 				if (err < 0) {
835 					tls_err_abort(sk, EBADMSG);
836 					goto recv_end;
837 				}
838 			}
839 			ctx->decrypted = true;
840 		}
841 
842 		if (!zc) {
843 			chunk = min_t(unsigned int, rxm->full_len, len);
844 			err = skb_copy_datagram_msg(skb, rxm->offset, msg,
845 						    chunk);
846 			if (err < 0)
847 				goto recv_end;
848 		}
849 
850 		copied += chunk;
851 		len -= chunk;
852 		if (likely(!(flags & MSG_PEEK))) {
853 			u8 control = ctx->control;
854 
855 			if (tls_sw_advance_skb(sk, skb, chunk)) {
856 				/* Return full control message to
857 				 * userspace before trying to parse
858 				 * another message type
859 				 */
860 				msg->msg_flags |= MSG_EOR;
861 				if (control != TLS_RECORD_TYPE_DATA)
862 					goto recv_end;
863 			}
864 		}
865 		/* If we have a new message from strparser, continue now. */
866 		if (copied >= target && !ctx->recv_pkt)
867 			break;
868 	} while (len);
869 
870 recv_end:
871 	release_sock(sk);
872 	return copied ? : err;
873 }
874 
875 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
876 			   struct pipe_inode_info *pipe,
877 			   size_t len, unsigned int flags)
878 {
879 	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
880 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
881 	struct strp_msg *rxm = NULL;
882 	struct sock *sk = sock->sk;
883 	struct sk_buff *skb;
884 	ssize_t copied = 0;
885 	int err = 0;
886 	long timeo;
887 	int chunk;
888 
889 	lock_sock(sk);
890 
891 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
892 
893 	skb = tls_wait_data(sk, flags, timeo, &err);
894 	if (!skb)
895 		goto splice_read_end;
896 
897 	/* splice does not support reading control messages */
898 	if (ctx->control != TLS_RECORD_TYPE_DATA) {
899 		err = -ENOTSUPP;
900 		goto splice_read_end;
901 	}
902 
903 	if (!ctx->decrypted) {
904 		err = decrypt_skb(sk, skb, NULL);
905 
906 		if (err < 0) {
907 			tls_err_abort(sk, EBADMSG);
908 			goto splice_read_end;
909 		}
910 		ctx->decrypted = true;
911 	}
912 	rxm = strp_msg(skb);
913 
914 	chunk = min_t(unsigned int, rxm->full_len, len);
915 	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
916 	if (copied < 0)
917 		goto splice_read_end;
918 
919 	if (likely(!(flags & MSG_PEEK)))
920 		tls_sw_advance_skb(sk, skb, copied);
921 
922 splice_read_end:
923 	release_sock(sk);
924 	return copied ? : err;
925 }
926 
927 unsigned int tls_sw_poll(struct file *file, struct socket *sock,
928 			 struct poll_table_struct *wait)
929 {
930 	unsigned int ret;
931 	struct sock *sk = sock->sk;
932 	struct tls_context *tls_ctx = tls_get_ctx(sk);
933 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
934 
935 	/* Grab POLLOUT and POLLHUP from the underlying socket */
936 	ret = ctx->sk_poll(file, sock, wait);
937 
938 	/* Clear POLLIN bits, and set based on recv_pkt */
939 	ret &= ~(POLLIN | POLLRDNORM);
940 	if (ctx->recv_pkt)
941 		ret |= POLLIN | POLLRDNORM;
942 
943 	return ret;
944 }
945 
946 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
947 {
948 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
949 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
950 	char header[tls_ctx->rx.prepend_size];
951 	struct strp_msg *rxm = strp_msg(skb);
952 	size_t cipher_overhead;
953 	size_t data_len = 0;
954 	int ret;
955 
956 	/* Verify that we have a full TLS header, or wait for more data */
957 	if (rxm->offset + tls_ctx->rx.prepend_size > skb->len)
958 		return 0;
959 
960 	/* Linearize header to local buffer */
961 	ret = skb_copy_bits(skb, rxm->offset, header, tls_ctx->rx.prepend_size);
962 
963 	if (ret < 0)
964 		goto read_failure;
965 
966 	ctx->control = header[0];
967 
968 	data_len = ((header[4] & 0xFF) | (header[3] << 8));
969 
970 	cipher_overhead = tls_ctx->rx.tag_size + tls_ctx->rx.iv_size;
971 
972 	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead) {
973 		ret = -EMSGSIZE;
974 		goto read_failure;
975 	}
976 	if (data_len < cipher_overhead) {
977 		ret = -EBADMSG;
978 		goto read_failure;
979 	}
980 
981 	if (header[1] != TLS_VERSION_MINOR(tls_ctx->crypto_recv.version) ||
982 	    header[2] != TLS_VERSION_MAJOR(tls_ctx->crypto_recv.version)) {
983 		ret = -EINVAL;
984 		goto read_failure;
985 	}
986 
987 	return data_len + TLS_HEADER_SIZE;
988 
989 read_failure:
990 	tls_err_abort(strp->sk, ret);
991 
992 	return ret;
993 }
994 
995 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
996 {
997 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
998 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
999 	struct strp_msg *rxm;
1000 
1001 	rxm = strp_msg(skb);
1002 
1003 	ctx->decrypted = false;
1004 
1005 	ctx->recv_pkt = skb;
1006 	strp_pause(strp);
1007 
1008 	strp->sk->sk_state_change(strp->sk);
1009 }
1010 
1011 static void tls_data_ready(struct sock *sk)
1012 {
1013 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1014 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1015 
1016 	strp_data_ready(&ctx->strp);
1017 }
1018 
1019 void tls_sw_free_resources_tx(struct sock *sk)
1020 {
1021 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1022 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1023 
1024 	if (ctx->aead_send)
1025 		crypto_free_aead(ctx->aead_send);
1026 	tls_free_both_sg(sk);
1027 
1028 	kfree(ctx);
1029 }
1030 
1031 void tls_sw_free_resources_rx(struct sock *sk)
1032 {
1033 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1034 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1035 
1036 	if (ctx->aead_recv) {
1037 		if (ctx->recv_pkt) {
1038 			kfree_skb(ctx->recv_pkt);
1039 			ctx->recv_pkt = NULL;
1040 		}
1041 		crypto_free_aead(ctx->aead_recv);
1042 		strp_stop(&ctx->strp);
1043 		write_lock_bh(&sk->sk_callback_lock);
1044 		sk->sk_data_ready = ctx->saved_data_ready;
1045 		write_unlock_bh(&sk->sk_callback_lock);
1046 		release_sock(sk);
1047 		strp_done(&ctx->strp);
1048 		lock_sock(sk);
1049 	}
1050 
1051 	kfree(ctx);
1052 }
1053 
1054 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
1055 {
1056 	char keyval[TLS_CIPHER_AES_GCM_128_KEY_SIZE];
1057 	struct tls_crypto_info *crypto_info;
1058 	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
1059 	struct tls_sw_context_tx *sw_ctx_tx = NULL;
1060 	struct tls_sw_context_rx *sw_ctx_rx = NULL;
1061 	struct cipher_context *cctx;
1062 	struct crypto_aead **aead;
1063 	struct strp_callbacks cb;
1064 	u16 nonce_size, tag_size, iv_size, rec_seq_size;
1065 	char *iv, *rec_seq;
1066 	int rc = 0;
1067 
1068 	if (!ctx) {
1069 		rc = -EINVAL;
1070 		goto out;
1071 	}
1072 
1073 	if (tx) {
1074 		sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
1075 		if (!sw_ctx_tx) {
1076 			rc = -ENOMEM;
1077 			goto out;
1078 		}
1079 		crypto_init_wait(&sw_ctx_tx->async_wait);
1080 		ctx->priv_ctx_tx = sw_ctx_tx;
1081 	} else {
1082 		sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
1083 		if (!sw_ctx_rx) {
1084 			rc = -ENOMEM;
1085 			goto out;
1086 		}
1087 		crypto_init_wait(&sw_ctx_rx->async_wait);
1088 		ctx->priv_ctx_rx = sw_ctx_rx;
1089 	}
1090 
1091 	if (tx) {
1092 		crypto_info = &ctx->crypto_send;
1093 		cctx = &ctx->tx;
1094 		aead = &sw_ctx_tx->aead_send;
1095 	} else {
1096 		crypto_info = &ctx->crypto_recv;
1097 		cctx = &ctx->rx;
1098 		aead = &sw_ctx_rx->aead_recv;
1099 	}
1100 
1101 	switch (crypto_info->cipher_type) {
1102 	case TLS_CIPHER_AES_GCM_128: {
1103 		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1104 		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1105 		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1106 		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1107 		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1108 		rec_seq =
1109 		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1110 		gcm_128_info =
1111 			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
1112 		break;
1113 	}
1114 	default:
1115 		rc = -EINVAL;
1116 		goto free_priv;
1117 	}
1118 
1119 	/* Sanity-check the IV size for stack allocations. */
1120 	if (iv_size > MAX_IV_SIZE) {
1121 		rc = -EINVAL;
1122 		goto free_priv;
1123 	}
1124 
1125 	cctx->prepend_size = TLS_HEADER_SIZE + nonce_size;
1126 	cctx->tag_size = tag_size;
1127 	cctx->overhead_size = cctx->prepend_size + cctx->tag_size;
1128 	cctx->iv_size = iv_size;
1129 	cctx->iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1130 			   GFP_KERNEL);
1131 	if (!cctx->iv) {
1132 		rc = -ENOMEM;
1133 		goto free_priv;
1134 	}
1135 	memcpy(cctx->iv, gcm_128_info->salt, TLS_CIPHER_AES_GCM_128_SALT_SIZE);
1136 	memcpy(cctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1137 	cctx->rec_seq_size = rec_seq_size;
1138 	cctx->rec_seq = kmalloc(rec_seq_size, GFP_KERNEL);
1139 	if (!cctx->rec_seq) {
1140 		rc = -ENOMEM;
1141 		goto free_iv;
1142 	}
1143 	memcpy(cctx->rec_seq, rec_seq, rec_seq_size);
1144 
1145 	if (sw_ctx_tx) {
1146 		sg_init_table(sw_ctx_tx->sg_encrypted_data,
1147 			      ARRAY_SIZE(sw_ctx_tx->sg_encrypted_data));
1148 		sg_init_table(sw_ctx_tx->sg_plaintext_data,
1149 			      ARRAY_SIZE(sw_ctx_tx->sg_plaintext_data));
1150 
1151 		sg_init_table(sw_ctx_tx->sg_aead_in, 2);
1152 		sg_set_buf(&sw_ctx_tx->sg_aead_in[0], sw_ctx_tx->aad_space,
1153 			   sizeof(sw_ctx_tx->aad_space));
1154 		sg_unmark_end(&sw_ctx_tx->sg_aead_in[1]);
1155 		sg_chain(sw_ctx_tx->sg_aead_in, 2,
1156 			 sw_ctx_tx->sg_plaintext_data);
1157 		sg_init_table(sw_ctx_tx->sg_aead_out, 2);
1158 		sg_set_buf(&sw_ctx_tx->sg_aead_out[0], sw_ctx_tx->aad_space,
1159 			   sizeof(sw_ctx_tx->aad_space));
1160 		sg_unmark_end(&sw_ctx_tx->sg_aead_out[1]);
1161 		sg_chain(sw_ctx_tx->sg_aead_out, 2,
1162 			 sw_ctx_tx->sg_encrypted_data);
1163 	}
1164 
1165 	if (!*aead) {
1166 		*aead = crypto_alloc_aead("gcm(aes)", 0, 0);
1167 		if (IS_ERR(*aead)) {
1168 			rc = PTR_ERR(*aead);
1169 			*aead = NULL;
1170 			goto free_rec_seq;
1171 		}
1172 	}
1173 
1174 	ctx->push_pending_record = tls_sw_push_pending_record;
1175 
1176 	memcpy(keyval, gcm_128_info->key, TLS_CIPHER_AES_GCM_128_KEY_SIZE);
1177 
1178 	rc = crypto_aead_setkey(*aead, keyval,
1179 				TLS_CIPHER_AES_GCM_128_KEY_SIZE);
1180 	if (rc)
1181 		goto free_aead;
1182 
1183 	rc = crypto_aead_setauthsize(*aead, cctx->tag_size);
1184 	if (rc)
1185 		goto free_aead;
1186 
1187 	if (sw_ctx_rx) {
1188 		/* Set up strparser */
1189 		memset(&cb, 0, sizeof(cb));
1190 		cb.rcv_msg = tls_queue;
1191 		cb.parse_msg = tls_read_size;
1192 
1193 		strp_init(&sw_ctx_rx->strp, sk, &cb);
1194 
1195 		write_lock_bh(&sk->sk_callback_lock);
1196 		sw_ctx_rx->saved_data_ready = sk->sk_data_ready;
1197 		sk->sk_data_ready = tls_data_ready;
1198 		write_unlock_bh(&sk->sk_callback_lock);
1199 
1200 		sw_ctx_rx->sk_poll = sk->sk_socket->ops->poll;
1201 
1202 		strp_check_rcv(&sw_ctx_rx->strp);
1203 	}
1204 
1205 	goto out;
1206 
1207 free_aead:
1208 	crypto_free_aead(*aead);
1209 	*aead = NULL;
1210 free_rec_seq:
1211 	kfree(cctx->rec_seq);
1212 	cctx->rec_seq = NULL;
1213 free_iv:
1214 	kfree(cctx->iv);
1215 	cctx->iv = NULL;
1216 free_priv:
1217 	if (tx) {
1218 		kfree(ctx->priv_ctx_tx);
1219 		ctx->priv_ctx_tx = NULL;
1220 	} else {
1221 		kfree(ctx->priv_ctx_rx);
1222 		ctx->priv_ctx_rx = NULL;
1223 	}
1224 out:
1225 	return rc;
1226 }
1227