xref: /openbmc/linux/net/tls/tls_sw.c (revision 22fc4c4c9fd60427bcda00878cee94e7622cfa7a)
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  * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
8  *
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:
14  *
15  *     Redistribution and use in source and binary forms, with or
16  *     without modification, are permitted provided that the following
17  *     conditions are met:
18  *
19  *      - Redistributions of source code must retain the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer.
22  *
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.
27  *
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
35  * SOFTWARE.
36  */
37 
38 #include <linux/sched/signal.h>
39 #include <linux/module.h>
40 #include <crypto/aead.h>
41 
42 #include <net/strparser.h>
43 #include <net/tls.h>
44 
45 #define MAX_IV_SIZE	TLS_CIPHER_AES_GCM_128_IV_SIZE
46 
47 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
48                      unsigned int recursion_level)
49 {
50         int start = skb_headlen(skb);
51         int i, chunk = start - offset;
52         struct sk_buff *frag_iter;
53         int elt = 0;
54 
55         if (unlikely(recursion_level >= 24))
56                 return -EMSGSIZE;
57 
58         if (chunk > 0) {
59                 if (chunk > len)
60                         chunk = len;
61                 elt++;
62                 len -= chunk;
63                 if (len == 0)
64                         return elt;
65                 offset += chunk;
66         }
67 
68         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
69                 int end;
70 
71                 WARN_ON(start > offset + len);
72 
73                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
74                 chunk = end - offset;
75                 if (chunk > 0) {
76                         if (chunk > len)
77                                 chunk = len;
78                         elt++;
79                         len -= chunk;
80                         if (len == 0)
81                                 return elt;
82                         offset += chunk;
83                 }
84                 start = end;
85         }
86 
87         if (unlikely(skb_has_frag_list(skb))) {
88                 skb_walk_frags(skb, frag_iter) {
89                         int end, ret;
90 
91                         WARN_ON(start > offset + len);
92 
93                         end = start + frag_iter->len;
94                         chunk = end - offset;
95                         if (chunk > 0) {
96                                 if (chunk > len)
97                                         chunk = len;
98                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
99                                                 recursion_level + 1);
100                                 if (unlikely(ret < 0))
101                                         return ret;
102                                 elt += ret;
103                                 len -= chunk;
104                                 if (len == 0)
105                                         return elt;
106                                 offset += chunk;
107                         }
108                         start = end;
109                 }
110         }
111         BUG_ON(len);
112         return elt;
113 }
114 
115 /* Return the number of scatterlist elements required to completely map the
116  * skb, or -EMSGSIZE if the recursion depth is exceeded.
117  */
118 static int skb_nsg(struct sk_buff *skb, int offset, int len)
119 {
120         return __skb_nsg(skb, offset, len, 0);
121 }
122 
123 static void tls_decrypt_done(struct crypto_async_request *req, int err)
124 {
125 	struct aead_request *aead_req = (struct aead_request *)req;
126 	struct scatterlist *sgout = aead_req->dst;
127 	struct scatterlist *sgin = aead_req->src;
128 	struct tls_sw_context_rx *ctx;
129 	struct tls_context *tls_ctx;
130 	struct scatterlist *sg;
131 	struct sk_buff *skb;
132 	unsigned int pages;
133 	int pending;
134 
135 	skb = (struct sk_buff *)req->data;
136 	tls_ctx = tls_get_ctx(skb->sk);
137 	ctx = tls_sw_ctx_rx(tls_ctx);
138 
139 	/* Propagate if there was an err */
140 	if (err) {
141 		ctx->async_wait.err = err;
142 		tls_err_abort(skb->sk, err);
143 	} else {
144 		struct strp_msg *rxm = strp_msg(skb);
145 
146 		rxm->offset += tls_ctx->rx.prepend_size;
147 		rxm->full_len -= tls_ctx->rx.overhead_size;
148 	}
149 
150 	/* After using skb->sk to propagate sk through crypto async callback
151 	 * we need to NULL it again.
152 	 */
153 	skb->sk = NULL;
154 
155 
156 	/* Free the destination pages if skb was not decrypted inplace */
157 	if (sgout != sgin) {
158 		/* Skip the first S/G entry as it points to AAD */
159 		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
160 			if (!sg)
161 				break;
162 			put_page(sg_page(sg));
163 		}
164 	}
165 
166 	kfree(aead_req);
167 
168 	pending = atomic_dec_return(&ctx->decrypt_pending);
169 
170 	if (!pending && READ_ONCE(ctx->async_notify))
171 		complete(&ctx->async_wait.completion);
172 }
173 
174 static int tls_do_decryption(struct sock *sk,
175 			     struct sk_buff *skb,
176 			     struct scatterlist *sgin,
177 			     struct scatterlist *sgout,
178 			     char *iv_recv,
179 			     size_t data_len,
180 			     struct aead_request *aead_req,
181 			     bool async)
182 {
183 	struct tls_context *tls_ctx = tls_get_ctx(sk);
184 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
185 	int ret;
186 
187 	aead_request_set_tfm(aead_req, ctx->aead_recv);
188 	aead_request_set_ad(aead_req, TLS_AAD_SPACE_SIZE);
189 	aead_request_set_crypt(aead_req, sgin, sgout,
190 			       data_len + tls_ctx->rx.tag_size,
191 			       (u8 *)iv_recv);
192 
193 	if (async) {
194 		/* Using skb->sk to push sk through to crypto async callback
195 		 * handler. This allows propagating errors up to the socket
196 		 * if needed. It _must_ be cleared in the async handler
197 		 * before kfree_skb is called. We _know_ skb->sk is NULL
198 		 * because it is a clone from strparser.
199 		 */
200 		skb->sk = sk;
201 		aead_request_set_callback(aead_req,
202 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
203 					  tls_decrypt_done, skb);
204 		atomic_inc(&ctx->decrypt_pending);
205 	} else {
206 		aead_request_set_callback(aead_req,
207 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
208 					  crypto_req_done, &ctx->async_wait);
209 	}
210 
211 	ret = crypto_aead_decrypt(aead_req);
212 	if (ret == -EINPROGRESS) {
213 		if (async)
214 			return ret;
215 
216 		ret = crypto_wait_req(ret, &ctx->async_wait);
217 	}
218 
219 	if (async)
220 		atomic_dec(&ctx->decrypt_pending);
221 
222 	return ret;
223 }
224 
225 static void tls_trim_both_msgs(struct sock *sk, int target_size)
226 {
227 	struct tls_context *tls_ctx = tls_get_ctx(sk);
228 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
229 	struct tls_rec *rec = ctx->open_rec;
230 
231 	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
232 	if (target_size > 0)
233 		target_size += tls_ctx->tx.overhead_size;
234 	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
235 }
236 
237 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
238 {
239 	struct tls_context *tls_ctx = tls_get_ctx(sk);
240 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
241 	struct tls_rec *rec = ctx->open_rec;
242 	struct sk_msg *msg_en = &rec->msg_encrypted;
243 
244 	return sk_msg_alloc(sk, msg_en, len, 0);
245 }
246 
247 static int tls_clone_plaintext_msg(struct sock *sk, int required)
248 {
249 	struct tls_context *tls_ctx = tls_get_ctx(sk);
250 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
251 	struct tls_rec *rec = ctx->open_rec;
252 	struct sk_msg *msg_pl = &rec->msg_plaintext;
253 	struct sk_msg *msg_en = &rec->msg_encrypted;
254 	int skip, len;
255 
256 	/* We add page references worth len bytes from encrypted sg
257 	 * at the end of plaintext sg. It is guaranteed that msg_en
258 	 * has enough required room (ensured by caller).
259 	 */
260 	len = required - msg_pl->sg.size;
261 
262 	/* Skip initial bytes in msg_en's data to be able to use
263 	 * same offset of both plain and encrypted data.
264 	 */
265 	skip = tls_ctx->tx.prepend_size + msg_pl->sg.size;
266 
267 	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
268 }
269 
270 static struct tls_rec *tls_get_rec(struct sock *sk)
271 {
272 	struct tls_context *tls_ctx = tls_get_ctx(sk);
273 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
274 	struct sk_msg *msg_pl, *msg_en;
275 	struct tls_rec *rec;
276 	int mem_size;
277 
278 	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
279 
280 	rec = kzalloc(mem_size, sk->sk_allocation);
281 	if (!rec)
282 		return NULL;
283 
284 	msg_pl = &rec->msg_plaintext;
285 	msg_en = &rec->msg_encrypted;
286 
287 	sk_msg_init(msg_pl);
288 	sk_msg_init(msg_en);
289 
290 	sg_init_table(rec->sg_aead_in, 2);
291 	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space,
292 		   sizeof(rec->aad_space));
293 	sg_unmark_end(&rec->sg_aead_in[1]);
294 
295 	sg_init_table(rec->sg_aead_out, 2);
296 	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space,
297 		   sizeof(rec->aad_space));
298 	sg_unmark_end(&rec->sg_aead_out[1]);
299 
300 	return rec;
301 }
302 
303 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
304 {
305 	sk_msg_free(sk, &rec->msg_encrypted);
306 	sk_msg_free(sk, &rec->msg_plaintext);
307 	kfree(rec);
308 }
309 
310 static void tls_free_open_rec(struct sock *sk)
311 {
312 	struct tls_context *tls_ctx = tls_get_ctx(sk);
313 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
314 	struct tls_rec *rec = ctx->open_rec;
315 
316 	if (rec) {
317 		tls_free_rec(sk, rec);
318 		ctx->open_rec = NULL;
319 	}
320 }
321 
322 int tls_tx_records(struct sock *sk, int flags)
323 {
324 	struct tls_context *tls_ctx = tls_get_ctx(sk);
325 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
326 	struct tls_rec *rec, *tmp;
327 	struct sk_msg *msg_en;
328 	int tx_flags, rc = 0;
329 
330 	if (tls_is_partially_sent_record(tls_ctx)) {
331 		rec = list_first_entry(&ctx->tx_list,
332 				       struct tls_rec, list);
333 
334 		if (flags == -1)
335 			tx_flags = rec->tx_flags;
336 		else
337 			tx_flags = flags;
338 
339 		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
340 		if (rc)
341 			goto tx_err;
342 
343 		/* Full record has been transmitted.
344 		 * Remove the head of tx_list
345 		 */
346 		list_del(&rec->list);
347 		sk_msg_free(sk, &rec->msg_plaintext);
348 		kfree(rec);
349 	}
350 
351 	/* Tx all ready records */
352 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
353 		if (READ_ONCE(rec->tx_ready)) {
354 			if (flags == -1)
355 				tx_flags = rec->tx_flags;
356 			else
357 				tx_flags = flags;
358 
359 			msg_en = &rec->msg_encrypted;
360 			rc = tls_push_sg(sk, tls_ctx,
361 					 &msg_en->sg.data[msg_en->sg.curr],
362 					 0, tx_flags);
363 			if (rc)
364 				goto tx_err;
365 
366 			list_del(&rec->list);
367 			sk_msg_free(sk, &rec->msg_plaintext);
368 			kfree(rec);
369 		} else {
370 			break;
371 		}
372 	}
373 
374 tx_err:
375 	if (rc < 0 && rc != -EAGAIN)
376 		tls_err_abort(sk, EBADMSG);
377 
378 	return rc;
379 }
380 
381 static void tls_encrypt_done(struct crypto_async_request *req, int err)
382 {
383 	struct aead_request *aead_req = (struct aead_request *)req;
384 	struct sock *sk = req->data;
385 	struct tls_context *tls_ctx = tls_get_ctx(sk);
386 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
387 	struct scatterlist *sge;
388 	struct sk_msg *msg_en;
389 	struct tls_rec *rec;
390 	bool ready = false;
391 	int pending;
392 
393 	rec = container_of(aead_req, struct tls_rec, aead_req);
394 	msg_en = &rec->msg_encrypted;
395 
396 	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
397 	sge->offset -= tls_ctx->tx.prepend_size;
398 	sge->length += tls_ctx->tx.prepend_size;
399 
400 	/* Check if error is previously set on socket */
401 	if (err || sk->sk_err) {
402 		rec = NULL;
403 
404 		/* If err is already set on socket, return the same code */
405 		if (sk->sk_err) {
406 			ctx->async_wait.err = sk->sk_err;
407 		} else {
408 			ctx->async_wait.err = err;
409 			tls_err_abort(sk, err);
410 		}
411 	}
412 
413 	if (rec) {
414 		struct tls_rec *first_rec;
415 
416 		/* Mark the record as ready for transmission */
417 		smp_store_mb(rec->tx_ready, true);
418 
419 		/* If received record is at head of tx_list, schedule tx */
420 		first_rec = list_first_entry(&ctx->tx_list,
421 					     struct tls_rec, list);
422 		if (rec == first_rec)
423 			ready = true;
424 	}
425 
426 	pending = atomic_dec_return(&ctx->encrypt_pending);
427 
428 	if (!pending && READ_ONCE(ctx->async_notify))
429 		complete(&ctx->async_wait.completion);
430 
431 	if (!ready)
432 		return;
433 
434 	/* Schedule the transmission */
435 	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
436 		schedule_delayed_work(&ctx->tx_work.work, 1);
437 }
438 
439 static int tls_do_encryption(struct sock *sk,
440 			     struct tls_context *tls_ctx,
441 			     struct tls_sw_context_tx *ctx,
442 			     struct aead_request *aead_req,
443 			     size_t data_len, u32 start)
444 {
445 	struct tls_rec *rec = ctx->open_rec;
446 	struct sk_msg *msg_en = &rec->msg_encrypted;
447 	struct scatterlist *sge = sk_msg_elem(msg_en, start);
448 	int rc;
449 
450 	sge->offset += tls_ctx->tx.prepend_size;
451 	sge->length -= tls_ctx->tx.prepend_size;
452 
453 	msg_en->sg.curr = start;
454 
455 	aead_request_set_tfm(aead_req, ctx->aead_send);
456 	aead_request_set_ad(aead_req, TLS_AAD_SPACE_SIZE);
457 	aead_request_set_crypt(aead_req, rec->sg_aead_in,
458 			       rec->sg_aead_out,
459 			       data_len, tls_ctx->tx.iv);
460 
461 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
462 				  tls_encrypt_done, sk);
463 
464 	/* Add the record in tx_list */
465 	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
466 	atomic_inc(&ctx->encrypt_pending);
467 
468 	rc = crypto_aead_encrypt(aead_req);
469 	if (!rc || rc != -EINPROGRESS) {
470 		atomic_dec(&ctx->encrypt_pending);
471 		sge->offset -= tls_ctx->tx.prepend_size;
472 		sge->length += tls_ctx->tx.prepend_size;
473 	}
474 
475 	if (!rc) {
476 		WRITE_ONCE(rec->tx_ready, true);
477 	} else if (rc != -EINPROGRESS) {
478 		list_del(&rec->list);
479 		return rc;
480 	}
481 
482 	/* Unhook the record from context if encryption is not failure */
483 	ctx->open_rec = NULL;
484 	tls_advance_record_sn(sk, &tls_ctx->tx);
485 	return rc;
486 }
487 
488 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
489 				 struct tls_rec **to, struct sk_msg *msg_opl,
490 				 struct sk_msg *msg_oen, u32 split_point,
491 				 u32 tx_overhead_size, u32 *orig_end)
492 {
493 	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
494 	struct scatterlist *sge, *osge, *nsge;
495 	u32 orig_size = msg_opl->sg.size;
496 	struct scatterlist tmp = { };
497 	struct sk_msg *msg_npl;
498 	struct tls_rec *new;
499 	int ret;
500 
501 	new = tls_get_rec(sk);
502 	if (!new)
503 		return -ENOMEM;
504 	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
505 			   tx_overhead_size, 0);
506 	if (ret < 0) {
507 		tls_free_rec(sk, new);
508 		return ret;
509 	}
510 
511 	*orig_end = msg_opl->sg.end;
512 	i = msg_opl->sg.start;
513 	sge = sk_msg_elem(msg_opl, i);
514 	while (apply && sge->length) {
515 		if (sge->length > apply) {
516 			u32 len = sge->length - apply;
517 
518 			get_page(sg_page(sge));
519 			sg_set_page(&tmp, sg_page(sge), len,
520 				    sge->offset + apply);
521 			sge->length = apply;
522 			bytes += apply;
523 			apply = 0;
524 		} else {
525 			apply -= sge->length;
526 			bytes += sge->length;
527 		}
528 
529 		sk_msg_iter_var_next(i);
530 		if (i == msg_opl->sg.end)
531 			break;
532 		sge = sk_msg_elem(msg_opl, i);
533 	}
534 
535 	msg_opl->sg.end = i;
536 	msg_opl->sg.curr = i;
537 	msg_opl->sg.copybreak = 0;
538 	msg_opl->apply_bytes = 0;
539 	msg_opl->sg.size = bytes;
540 
541 	msg_npl = &new->msg_plaintext;
542 	msg_npl->apply_bytes = apply;
543 	msg_npl->sg.size = orig_size - bytes;
544 
545 	j = msg_npl->sg.start;
546 	nsge = sk_msg_elem(msg_npl, j);
547 	if (tmp.length) {
548 		memcpy(nsge, &tmp, sizeof(*nsge));
549 		sk_msg_iter_var_next(j);
550 		nsge = sk_msg_elem(msg_npl, j);
551 	}
552 
553 	osge = sk_msg_elem(msg_opl, i);
554 	while (osge->length) {
555 		memcpy(nsge, osge, sizeof(*nsge));
556 		sg_unmark_end(nsge);
557 		sk_msg_iter_var_next(i);
558 		sk_msg_iter_var_next(j);
559 		if (i == *orig_end)
560 			break;
561 		osge = sk_msg_elem(msg_opl, i);
562 		nsge = sk_msg_elem(msg_npl, j);
563 	}
564 
565 	msg_npl->sg.end = j;
566 	msg_npl->sg.curr = j;
567 	msg_npl->sg.copybreak = 0;
568 
569 	*to = new;
570 	return 0;
571 }
572 
573 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
574 				  struct tls_rec *from, u32 orig_end)
575 {
576 	struct sk_msg *msg_npl = &from->msg_plaintext;
577 	struct sk_msg *msg_opl = &to->msg_plaintext;
578 	struct scatterlist *osge, *nsge;
579 	u32 i, j;
580 
581 	i = msg_opl->sg.end;
582 	sk_msg_iter_var_prev(i);
583 	j = msg_npl->sg.start;
584 
585 	osge = sk_msg_elem(msg_opl, i);
586 	nsge = sk_msg_elem(msg_npl, j);
587 
588 	if (sg_page(osge) == sg_page(nsge) &&
589 	    osge->offset + osge->length == nsge->offset) {
590 		osge->length += nsge->length;
591 		put_page(sg_page(nsge));
592 	}
593 
594 	msg_opl->sg.end = orig_end;
595 	msg_opl->sg.curr = orig_end;
596 	msg_opl->sg.copybreak = 0;
597 	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
598 	msg_opl->sg.size += msg_npl->sg.size;
599 
600 	sk_msg_free(sk, &to->msg_encrypted);
601 	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
602 
603 	kfree(from);
604 }
605 
606 static int tls_push_record(struct sock *sk, int flags,
607 			   unsigned char record_type)
608 {
609 	struct tls_context *tls_ctx = tls_get_ctx(sk);
610 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
611 	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
612 	u32 i, split_point, uninitialized_var(orig_end);
613 	struct sk_msg *msg_pl, *msg_en;
614 	struct aead_request *req;
615 	bool split;
616 	int rc;
617 
618 	if (!rec)
619 		return 0;
620 
621 	msg_pl = &rec->msg_plaintext;
622 	msg_en = &rec->msg_encrypted;
623 
624 	split_point = msg_pl->apply_bytes;
625 	split = split_point && split_point < msg_pl->sg.size;
626 	if (split) {
627 		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
628 					   split_point, tls_ctx->tx.overhead_size,
629 					   &orig_end);
630 		if (rc < 0)
631 			return rc;
632 		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
633 			    tls_ctx->tx.overhead_size);
634 	}
635 
636 	rec->tx_flags = flags;
637 	req = &rec->aead_req;
638 
639 	i = msg_pl->sg.end;
640 	sk_msg_iter_var_prev(i);
641 	sg_mark_end(sk_msg_elem(msg_pl, i));
642 
643 	i = msg_pl->sg.start;
644 	sg_chain(rec->sg_aead_in, 2, rec->inplace_crypto ?
645 		 &msg_en->sg.data[i] : &msg_pl->sg.data[i]);
646 
647 	i = msg_en->sg.end;
648 	sk_msg_iter_var_prev(i);
649 	sg_mark_end(sk_msg_elem(msg_en, i));
650 
651 	i = msg_en->sg.start;
652 	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
653 
654 	tls_make_aad(rec->aad_space, msg_pl->sg.size,
655 		     tls_ctx->tx.rec_seq, tls_ctx->tx.rec_seq_size,
656 		     record_type);
657 
658 	tls_fill_prepend(tls_ctx,
659 			 page_address(sg_page(&msg_en->sg.data[i])) +
660 			 msg_en->sg.data[i].offset, msg_pl->sg.size,
661 			 record_type);
662 
663 	tls_ctx->pending_open_record_frags = false;
664 
665 	rc = tls_do_encryption(sk, tls_ctx, ctx, req, msg_pl->sg.size, i);
666 	if (rc < 0) {
667 		if (rc != -EINPROGRESS) {
668 			tls_err_abort(sk, EBADMSG);
669 			if (split) {
670 				tls_ctx->pending_open_record_frags = true;
671 				tls_merge_open_record(sk, rec, tmp, orig_end);
672 			}
673 		}
674 		return rc;
675 	} else if (split) {
676 		msg_pl = &tmp->msg_plaintext;
677 		msg_en = &tmp->msg_encrypted;
678 		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
679 			    tls_ctx->tx.overhead_size);
680 		tls_ctx->pending_open_record_frags = true;
681 		ctx->open_rec = tmp;
682 	}
683 
684 	return tls_tx_records(sk, flags);
685 }
686 
687 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
688 			       bool full_record, u8 record_type,
689 			       size_t *copied, int flags)
690 {
691 	struct tls_context *tls_ctx = tls_get_ctx(sk);
692 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
693 	struct sk_msg msg_redir = { };
694 	struct sk_psock *psock;
695 	struct sock *sk_redir;
696 	struct tls_rec *rec;
697 	bool enospc, policy;
698 	int err = 0, send;
699 	u32 delta = 0;
700 
701 	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
702 	psock = sk_psock_get(sk);
703 	if (!psock || !policy)
704 		return tls_push_record(sk, flags, record_type);
705 more_data:
706 	enospc = sk_msg_full(msg);
707 	if (psock->eval == __SK_NONE) {
708 		delta = msg->sg.size;
709 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
710 		if (delta < msg->sg.size)
711 			delta -= msg->sg.size;
712 		else
713 			delta = 0;
714 	}
715 	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
716 	    !enospc && !full_record) {
717 		err = -ENOSPC;
718 		goto out_err;
719 	}
720 	msg->cork_bytes = 0;
721 	send = msg->sg.size;
722 	if (msg->apply_bytes && msg->apply_bytes < send)
723 		send = msg->apply_bytes;
724 
725 	switch (psock->eval) {
726 	case __SK_PASS:
727 		err = tls_push_record(sk, flags, record_type);
728 		if (err < 0) {
729 			*copied -= sk_msg_free(sk, msg);
730 			tls_free_open_rec(sk);
731 			goto out_err;
732 		}
733 		break;
734 	case __SK_REDIRECT:
735 		sk_redir = psock->sk_redir;
736 		memcpy(&msg_redir, msg, sizeof(*msg));
737 		if (msg->apply_bytes < send)
738 			msg->apply_bytes = 0;
739 		else
740 			msg->apply_bytes -= send;
741 		sk_msg_return_zero(sk, msg, send);
742 		msg->sg.size -= send;
743 		release_sock(sk);
744 		err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
745 		lock_sock(sk);
746 		if (err < 0) {
747 			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
748 			msg->sg.size = 0;
749 		}
750 		if (msg->sg.size == 0)
751 			tls_free_open_rec(sk);
752 		break;
753 	case __SK_DROP:
754 	default:
755 		sk_msg_free_partial(sk, msg, send);
756 		if (msg->apply_bytes < send)
757 			msg->apply_bytes = 0;
758 		else
759 			msg->apply_bytes -= send;
760 		if (msg->sg.size == 0)
761 			tls_free_open_rec(sk);
762 		*copied -= (send + delta);
763 		err = -EACCES;
764 	}
765 
766 	if (likely(!err)) {
767 		bool reset_eval = !ctx->open_rec;
768 
769 		rec = ctx->open_rec;
770 		if (rec) {
771 			msg = &rec->msg_plaintext;
772 			if (!msg->apply_bytes)
773 				reset_eval = true;
774 		}
775 		if (reset_eval) {
776 			psock->eval = __SK_NONE;
777 			if (psock->sk_redir) {
778 				sock_put(psock->sk_redir);
779 				psock->sk_redir = NULL;
780 			}
781 		}
782 		if (rec)
783 			goto more_data;
784 	}
785  out_err:
786 	sk_psock_put(sk, psock);
787 	return err;
788 }
789 
790 static int tls_sw_push_pending_record(struct sock *sk, int flags)
791 {
792 	struct tls_context *tls_ctx = tls_get_ctx(sk);
793 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
794 	struct tls_rec *rec = ctx->open_rec;
795 	struct sk_msg *msg_pl;
796 	size_t copied;
797 
798 	if (!rec)
799 		return 0;
800 
801 	msg_pl = &rec->msg_plaintext;
802 	copied = msg_pl->sg.size;
803 	if (!copied)
804 		return 0;
805 
806 	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
807 				   &copied, flags);
808 }
809 
810 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
811 {
812 	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
813 	struct tls_context *tls_ctx = tls_get_ctx(sk);
814 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
815 	struct crypto_tfm *tfm = crypto_aead_tfm(ctx->aead_send);
816 	bool async_capable = tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC;
817 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
818 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
819 	bool eor = !(msg->msg_flags & MSG_MORE);
820 	size_t try_to_copy, copied = 0;
821 	struct sk_msg *msg_pl, *msg_en;
822 	struct tls_rec *rec;
823 	int required_size;
824 	int num_async = 0;
825 	bool full_record;
826 	int record_room;
827 	int num_zc = 0;
828 	int orig_size;
829 	int ret = 0;
830 
831 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
832 		return -ENOTSUPP;
833 
834 	lock_sock(sk);
835 
836 	/* Wait till there is any pending write on socket */
837 	if (unlikely(sk->sk_write_pending)) {
838 		ret = wait_on_pending_writer(sk, &timeo);
839 		if (unlikely(ret))
840 			goto send_end;
841 	}
842 
843 	if (unlikely(msg->msg_controllen)) {
844 		ret = tls_proccess_cmsg(sk, msg, &record_type);
845 		if (ret) {
846 			if (ret == -EINPROGRESS)
847 				num_async++;
848 			else if (ret != -EAGAIN)
849 				goto send_end;
850 		}
851 	}
852 
853 	while (msg_data_left(msg)) {
854 		if (sk->sk_err) {
855 			ret = -sk->sk_err;
856 			goto send_end;
857 		}
858 
859 		if (ctx->open_rec)
860 			rec = ctx->open_rec;
861 		else
862 			rec = ctx->open_rec = tls_get_rec(sk);
863 		if (!rec) {
864 			ret = -ENOMEM;
865 			goto send_end;
866 		}
867 
868 		msg_pl = &rec->msg_plaintext;
869 		msg_en = &rec->msg_encrypted;
870 
871 		orig_size = msg_pl->sg.size;
872 		full_record = false;
873 		try_to_copy = msg_data_left(msg);
874 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
875 		if (try_to_copy >= record_room) {
876 			try_to_copy = record_room;
877 			full_record = true;
878 		}
879 
880 		required_size = msg_pl->sg.size + try_to_copy +
881 				tls_ctx->tx.overhead_size;
882 
883 		if (!sk_stream_memory_free(sk))
884 			goto wait_for_sndbuf;
885 
886 alloc_encrypted:
887 		ret = tls_alloc_encrypted_msg(sk, required_size);
888 		if (ret) {
889 			if (ret != -ENOSPC)
890 				goto wait_for_memory;
891 
892 			/* Adjust try_to_copy according to the amount that was
893 			 * actually allocated. The difference is due
894 			 * to max sg elements limit
895 			 */
896 			try_to_copy -= required_size - msg_en->sg.size;
897 			full_record = true;
898 		}
899 
900 		if (!is_kvec && (full_record || eor) && !async_capable) {
901 			u32 first = msg_pl->sg.end;
902 
903 			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
904 							msg_pl, try_to_copy);
905 			if (ret)
906 				goto fallback_to_reg_send;
907 
908 			rec->inplace_crypto = 0;
909 
910 			num_zc++;
911 			copied += try_to_copy;
912 
913 			sk_msg_sg_copy_set(msg_pl, first);
914 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
915 						  record_type, &copied,
916 						  msg->msg_flags);
917 			if (ret) {
918 				if (ret == -EINPROGRESS)
919 					num_async++;
920 				else if (ret == -ENOMEM)
921 					goto wait_for_memory;
922 				else if (ret == -ENOSPC)
923 					goto rollback_iter;
924 				else if (ret != -EAGAIN)
925 					goto send_end;
926 			}
927 			continue;
928 rollback_iter:
929 			copied -= try_to_copy;
930 			sk_msg_sg_copy_clear(msg_pl, first);
931 			iov_iter_revert(&msg->msg_iter,
932 					msg_pl->sg.size - orig_size);
933 fallback_to_reg_send:
934 			sk_msg_trim(sk, msg_pl, orig_size);
935 		}
936 
937 		required_size = msg_pl->sg.size + try_to_copy;
938 
939 		ret = tls_clone_plaintext_msg(sk, required_size);
940 		if (ret) {
941 			if (ret != -ENOSPC)
942 				goto send_end;
943 
944 			/* Adjust try_to_copy according to the amount that was
945 			 * actually allocated. The difference is due
946 			 * to max sg elements limit
947 			 */
948 			try_to_copy -= required_size - msg_pl->sg.size;
949 			full_record = true;
950 			sk_msg_trim(sk, msg_en, msg_pl->sg.size +
951 				    tls_ctx->tx.overhead_size);
952 		}
953 
954 		if (try_to_copy) {
955 			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
956 						       msg_pl, try_to_copy);
957 			if (ret < 0)
958 				goto trim_sgl;
959 		}
960 
961 		/* Open records defined only if successfully copied, otherwise
962 		 * we would trim the sg but not reset the open record frags.
963 		 */
964 		tls_ctx->pending_open_record_frags = true;
965 		copied += try_to_copy;
966 		if (full_record || eor) {
967 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
968 						  record_type, &copied,
969 						  msg->msg_flags);
970 			if (ret) {
971 				if (ret == -EINPROGRESS)
972 					num_async++;
973 				else if (ret == -ENOMEM)
974 					goto wait_for_memory;
975 				else if (ret != -EAGAIN) {
976 					if (ret == -ENOSPC)
977 						ret = 0;
978 					goto send_end;
979 				}
980 			}
981 		}
982 
983 		continue;
984 
985 wait_for_sndbuf:
986 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
987 wait_for_memory:
988 		ret = sk_stream_wait_memory(sk, &timeo);
989 		if (ret) {
990 trim_sgl:
991 			tls_trim_both_msgs(sk, orig_size);
992 			goto send_end;
993 		}
994 
995 		if (msg_en->sg.size < required_size)
996 			goto alloc_encrypted;
997 	}
998 
999 	if (!num_async) {
1000 		goto send_end;
1001 	} else if (num_zc) {
1002 		/* Wait for pending encryptions to get completed */
1003 		smp_store_mb(ctx->async_notify, true);
1004 
1005 		if (atomic_read(&ctx->encrypt_pending))
1006 			crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1007 		else
1008 			reinit_completion(&ctx->async_wait.completion);
1009 
1010 		WRITE_ONCE(ctx->async_notify, false);
1011 
1012 		if (ctx->async_wait.err) {
1013 			ret = ctx->async_wait.err;
1014 			copied = 0;
1015 		}
1016 	}
1017 
1018 	/* Transmit if any encryptions have completed */
1019 	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1020 		cancel_delayed_work(&ctx->tx_work.work);
1021 		tls_tx_records(sk, msg->msg_flags);
1022 	}
1023 
1024 send_end:
1025 	ret = sk_stream_error(sk, msg->msg_flags, ret);
1026 
1027 	release_sock(sk);
1028 	return copied ? copied : ret;
1029 }
1030 
1031 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1032 			      int offset, size_t size, int flags)
1033 {
1034 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1035 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1036 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1037 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1038 	struct sk_msg *msg_pl;
1039 	struct tls_rec *rec;
1040 	int num_async = 0;
1041 	size_t copied = 0;
1042 	bool full_record;
1043 	int record_room;
1044 	int ret = 0;
1045 	bool eor;
1046 
1047 	eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1048 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1049 
1050 	/* Wait till there is any pending write on socket */
1051 	if (unlikely(sk->sk_write_pending)) {
1052 		ret = wait_on_pending_writer(sk, &timeo);
1053 		if (unlikely(ret))
1054 			goto sendpage_end;
1055 	}
1056 
1057 	/* Call the sk_stream functions to manage the sndbuf mem. */
1058 	while (size > 0) {
1059 		size_t copy, required_size;
1060 
1061 		if (sk->sk_err) {
1062 			ret = -sk->sk_err;
1063 			goto sendpage_end;
1064 		}
1065 
1066 		if (ctx->open_rec)
1067 			rec = ctx->open_rec;
1068 		else
1069 			rec = ctx->open_rec = tls_get_rec(sk);
1070 		if (!rec) {
1071 			ret = -ENOMEM;
1072 			goto sendpage_end;
1073 		}
1074 
1075 		msg_pl = &rec->msg_plaintext;
1076 
1077 		full_record = false;
1078 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1079 		copied = 0;
1080 		copy = size;
1081 		if (copy >= record_room) {
1082 			copy = record_room;
1083 			full_record = true;
1084 		}
1085 
1086 		required_size = msg_pl->sg.size + copy +
1087 				tls_ctx->tx.overhead_size;
1088 
1089 		if (!sk_stream_memory_free(sk))
1090 			goto wait_for_sndbuf;
1091 alloc_payload:
1092 		ret = tls_alloc_encrypted_msg(sk, required_size);
1093 		if (ret) {
1094 			if (ret != -ENOSPC)
1095 				goto wait_for_memory;
1096 
1097 			/* Adjust copy according to the amount that was
1098 			 * actually allocated. The difference is due
1099 			 * to max sg elements limit
1100 			 */
1101 			copy -= required_size - msg_pl->sg.size;
1102 			full_record = true;
1103 		}
1104 
1105 		sk_msg_page_add(msg_pl, page, copy, offset);
1106 		sk_mem_charge(sk, copy);
1107 
1108 		offset += copy;
1109 		size -= copy;
1110 		copied += copy;
1111 
1112 		tls_ctx->pending_open_record_frags = true;
1113 		if (full_record || eor || sk_msg_full(msg_pl)) {
1114 			rec->inplace_crypto = 0;
1115 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1116 						  record_type, &copied, flags);
1117 			if (ret) {
1118 				if (ret == -EINPROGRESS)
1119 					num_async++;
1120 				else if (ret == -ENOMEM)
1121 					goto wait_for_memory;
1122 				else if (ret != -EAGAIN) {
1123 					if (ret == -ENOSPC)
1124 						ret = 0;
1125 					goto sendpage_end;
1126 				}
1127 			}
1128 		}
1129 		continue;
1130 wait_for_sndbuf:
1131 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1132 wait_for_memory:
1133 		ret = sk_stream_wait_memory(sk, &timeo);
1134 		if (ret) {
1135 			tls_trim_both_msgs(sk, msg_pl->sg.size);
1136 			goto sendpage_end;
1137 		}
1138 
1139 		goto alloc_payload;
1140 	}
1141 
1142 	if (num_async) {
1143 		/* Transmit if any encryptions have completed */
1144 		if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1145 			cancel_delayed_work(&ctx->tx_work.work);
1146 			tls_tx_records(sk, flags);
1147 		}
1148 	}
1149 sendpage_end:
1150 	ret = sk_stream_error(sk, flags, ret);
1151 	return copied ? copied : ret;
1152 }
1153 
1154 int tls_sw_sendpage(struct sock *sk, struct page *page,
1155 		    int offset, size_t size, int flags)
1156 {
1157 	int ret;
1158 
1159 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1160 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1161 		return -ENOTSUPP;
1162 
1163 	lock_sock(sk);
1164 	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1165 	release_sock(sk);
1166 	return ret;
1167 }
1168 
1169 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1170 				     int flags, long timeo, int *err)
1171 {
1172 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1173 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1174 	struct sk_buff *skb;
1175 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1176 
1177 	while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1178 		if (sk->sk_err) {
1179 			*err = sock_error(sk);
1180 			return NULL;
1181 		}
1182 
1183 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1184 			return NULL;
1185 
1186 		if (sock_flag(sk, SOCK_DONE))
1187 			return NULL;
1188 
1189 		if ((flags & MSG_DONTWAIT) || !timeo) {
1190 			*err = -EAGAIN;
1191 			return NULL;
1192 		}
1193 
1194 		add_wait_queue(sk_sleep(sk), &wait);
1195 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1196 		sk_wait_event(sk, &timeo,
1197 			      ctx->recv_pkt != skb ||
1198 			      !sk_psock_queue_empty(psock),
1199 			      &wait);
1200 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1201 		remove_wait_queue(sk_sleep(sk), &wait);
1202 
1203 		/* Handle signals */
1204 		if (signal_pending(current)) {
1205 			*err = sock_intr_errno(timeo);
1206 			return NULL;
1207 		}
1208 	}
1209 
1210 	return skb;
1211 }
1212 
1213 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1214 			       int length, int *pages_used,
1215 			       unsigned int *size_used,
1216 			       struct scatterlist *to,
1217 			       int to_max_pages)
1218 {
1219 	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1220 	struct page *pages[MAX_SKB_FRAGS];
1221 	unsigned int size = *size_used;
1222 	ssize_t copied, use;
1223 	size_t offset;
1224 
1225 	while (length > 0) {
1226 		i = 0;
1227 		maxpages = to_max_pages - num_elem;
1228 		if (maxpages == 0) {
1229 			rc = -EFAULT;
1230 			goto out;
1231 		}
1232 		copied = iov_iter_get_pages(from, pages,
1233 					    length,
1234 					    maxpages, &offset);
1235 		if (copied <= 0) {
1236 			rc = -EFAULT;
1237 			goto out;
1238 		}
1239 
1240 		iov_iter_advance(from, copied);
1241 
1242 		length -= copied;
1243 		size += copied;
1244 		while (copied) {
1245 			use = min_t(int, copied, PAGE_SIZE - offset);
1246 
1247 			sg_set_page(&to[num_elem],
1248 				    pages[i], use, offset);
1249 			sg_unmark_end(&to[num_elem]);
1250 			/* We do not uncharge memory from this API */
1251 
1252 			offset = 0;
1253 			copied -= use;
1254 
1255 			i++;
1256 			num_elem++;
1257 		}
1258 	}
1259 	/* Mark the end in the last sg entry if newly added */
1260 	if (num_elem > *pages_used)
1261 		sg_mark_end(&to[num_elem - 1]);
1262 out:
1263 	if (rc)
1264 		iov_iter_revert(from, size - *size_used);
1265 	*size_used = size;
1266 	*pages_used = num_elem;
1267 
1268 	return rc;
1269 }
1270 
1271 /* This function decrypts the input skb into either out_iov or in out_sg
1272  * or in skb buffers itself. The input parameter 'zc' indicates if
1273  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1274  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1275  * NULL, then the decryption happens inside skb buffers itself, i.e.
1276  * zero-copy gets disabled and 'zc' is updated.
1277  */
1278 
1279 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1280 			    struct iov_iter *out_iov,
1281 			    struct scatterlist *out_sg,
1282 			    int *chunk, bool *zc, bool async)
1283 {
1284 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1285 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1286 	struct strp_msg *rxm = strp_msg(skb);
1287 	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1288 	struct aead_request *aead_req;
1289 	struct sk_buff *unused;
1290 	u8 *aad, *iv, *mem = NULL;
1291 	struct scatterlist *sgin = NULL;
1292 	struct scatterlist *sgout = NULL;
1293 	const int data_len = rxm->full_len - tls_ctx->rx.overhead_size;
1294 
1295 	if (*zc && (out_iov || out_sg)) {
1296 		if (out_iov)
1297 			n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1298 		else
1299 			n_sgout = sg_nents(out_sg);
1300 		n_sgin = skb_nsg(skb, rxm->offset + tls_ctx->rx.prepend_size,
1301 				 rxm->full_len - tls_ctx->rx.prepend_size);
1302 	} else {
1303 		n_sgout = 0;
1304 		*zc = false;
1305 		n_sgin = skb_cow_data(skb, 0, &unused);
1306 	}
1307 
1308 	if (n_sgin < 1)
1309 		return -EBADMSG;
1310 
1311 	/* Increment to accommodate AAD */
1312 	n_sgin = n_sgin + 1;
1313 
1314 	nsg = n_sgin + n_sgout;
1315 
1316 	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1317 	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1318 	mem_size = mem_size + TLS_AAD_SPACE_SIZE;
1319 	mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1320 
1321 	/* Allocate a single block of memory which contains
1322 	 * aead_req || sgin[] || sgout[] || aad || iv.
1323 	 * This order achieves correct alignment for aead_req, sgin, sgout.
1324 	 */
1325 	mem = kmalloc(mem_size, sk->sk_allocation);
1326 	if (!mem)
1327 		return -ENOMEM;
1328 
1329 	/* Segment the allocated memory */
1330 	aead_req = (struct aead_request *)mem;
1331 	sgin = (struct scatterlist *)(mem + aead_size);
1332 	sgout = sgin + n_sgin;
1333 	aad = (u8 *)(sgout + n_sgout);
1334 	iv = aad + TLS_AAD_SPACE_SIZE;
1335 
1336 	/* Prepare IV */
1337 	err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1338 			    iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1339 			    tls_ctx->rx.iv_size);
1340 	if (err < 0) {
1341 		kfree(mem);
1342 		return err;
1343 	}
1344 	memcpy(iv, tls_ctx->rx.iv, TLS_CIPHER_AES_GCM_128_SALT_SIZE);
1345 
1346 	/* Prepare AAD */
1347 	tls_make_aad(aad, rxm->full_len - tls_ctx->rx.overhead_size,
1348 		     tls_ctx->rx.rec_seq, tls_ctx->rx.rec_seq_size,
1349 		     ctx->control);
1350 
1351 	/* Prepare sgin */
1352 	sg_init_table(sgin, n_sgin);
1353 	sg_set_buf(&sgin[0], aad, TLS_AAD_SPACE_SIZE);
1354 	err = skb_to_sgvec(skb, &sgin[1],
1355 			   rxm->offset + tls_ctx->rx.prepend_size,
1356 			   rxm->full_len - tls_ctx->rx.prepend_size);
1357 	if (err < 0) {
1358 		kfree(mem);
1359 		return err;
1360 	}
1361 
1362 	if (n_sgout) {
1363 		if (out_iov) {
1364 			sg_init_table(sgout, n_sgout);
1365 			sg_set_buf(&sgout[0], aad, TLS_AAD_SPACE_SIZE);
1366 
1367 			*chunk = 0;
1368 			err = tls_setup_from_iter(sk, out_iov, data_len,
1369 						  &pages, chunk, &sgout[1],
1370 						  (n_sgout - 1));
1371 			if (err < 0)
1372 				goto fallback_to_reg_recv;
1373 		} else if (out_sg) {
1374 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1375 		} else {
1376 			goto fallback_to_reg_recv;
1377 		}
1378 	} else {
1379 fallback_to_reg_recv:
1380 		sgout = sgin;
1381 		pages = 0;
1382 		*chunk = data_len;
1383 		*zc = false;
1384 	}
1385 
1386 	/* Prepare and submit AEAD request */
1387 	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1388 				data_len, aead_req, async);
1389 	if (err == -EINPROGRESS)
1390 		return err;
1391 
1392 	/* Release the pages in case iov was mapped to pages */
1393 	for (; pages > 0; pages--)
1394 		put_page(sg_page(&sgout[pages]));
1395 
1396 	kfree(mem);
1397 	return err;
1398 }
1399 
1400 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1401 			      struct iov_iter *dest, int *chunk, bool *zc,
1402 			      bool async)
1403 {
1404 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1405 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1406 	struct strp_msg *rxm = strp_msg(skb);
1407 	int err = 0;
1408 
1409 #ifdef CONFIG_TLS_DEVICE
1410 	err = tls_device_decrypted(sk, skb);
1411 	if (err < 0)
1412 		return err;
1413 #endif
1414 	if (!ctx->decrypted) {
1415 		err = decrypt_internal(sk, skb, dest, NULL, chunk, zc, async);
1416 		if (err < 0) {
1417 			if (err == -EINPROGRESS)
1418 				tls_advance_record_sn(sk, &tls_ctx->rx);
1419 
1420 			return err;
1421 		}
1422 	} else {
1423 		*zc = false;
1424 	}
1425 
1426 	rxm->offset += tls_ctx->rx.prepend_size;
1427 	rxm->full_len -= tls_ctx->rx.overhead_size;
1428 	tls_advance_record_sn(sk, &tls_ctx->rx);
1429 	ctx->decrypted = true;
1430 	ctx->saved_data_ready(sk);
1431 
1432 	return err;
1433 }
1434 
1435 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1436 		struct scatterlist *sgout)
1437 {
1438 	bool zc = true;
1439 	int chunk;
1440 
1441 	return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1442 }
1443 
1444 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1445 			       unsigned int len)
1446 {
1447 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1448 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1449 
1450 	if (skb) {
1451 		struct strp_msg *rxm = strp_msg(skb);
1452 
1453 		if (len < rxm->full_len) {
1454 			rxm->offset += len;
1455 			rxm->full_len -= len;
1456 			return false;
1457 		}
1458 		kfree_skb(skb);
1459 	}
1460 
1461 	/* Finished with message */
1462 	ctx->recv_pkt = NULL;
1463 	__strp_unpause(&ctx->strp);
1464 
1465 	return true;
1466 }
1467 
1468 /* This function traverses the rx_list in tls receive context to copies the
1469  * decrypted data records into the buffer provided by caller zero copy is not
1470  * true. Further, the records are removed from the rx_list if it is not a peek
1471  * case and the record has been consumed completely.
1472  */
1473 static int process_rx_list(struct tls_sw_context_rx *ctx,
1474 			   struct msghdr *msg,
1475 			   size_t skip,
1476 			   size_t len,
1477 			   bool zc,
1478 			   bool is_peek)
1479 {
1480 	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1481 	ssize_t copied = 0;
1482 
1483 	while (skip && skb) {
1484 		struct strp_msg *rxm = strp_msg(skb);
1485 
1486 		if (skip < rxm->full_len)
1487 			break;
1488 
1489 		skip = skip - rxm->full_len;
1490 		skb = skb_peek_next(skb, &ctx->rx_list);
1491 	}
1492 
1493 	while (len && skb) {
1494 		struct sk_buff *next_skb;
1495 		struct strp_msg *rxm = strp_msg(skb);
1496 		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1497 
1498 		if (!zc || (rxm->full_len - skip) > len) {
1499 			int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1500 						    msg, chunk);
1501 			if (err < 0)
1502 				return err;
1503 		}
1504 
1505 		len = len - chunk;
1506 		copied = copied + chunk;
1507 
1508 		/* Consume the data from record if it is non-peek case*/
1509 		if (!is_peek) {
1510 			rxm->offset = rxm->offset + chunk;
1511 			rxm->full_len = rxm->full_len - chunk;
1512 
1513 			/* Return if there is unconsumed data in the record */
1514 			if (rxm->full_len - skip)
1515 				break;
1516 		}
1517 
1518 		/* The remaining skip-bytes must lie in 1st record in rx_list.
1519 		 * So from the 2nd record, 'skip' should be 0.
1520 		 */
1521 		skip = 0;
1522 
1523 		if (msg)
1524 			msg->msg_flags |= MSG_EOR;
1525 
1526 		next_skb = skb_peek_next(skb, &ctx->rx_list);
1527 
1528 		if (!is_peek) {
1529 			skb_unlink(skb, &ctx->rx_list);
1530 			kfree_skb(skb);
1531 		}
1532 
1533 		skb = next_skb;
1534 	}
1535 
1536 	return copied;
1537 }
1538 
1539 int tls_sw_recvmsg(struct sock *sk,
1540 		   struct msghdr *msg,
1541 		   size_t len,
1542 		   int nonblock,
1543 		   int flags,
1544 		   int *addr_len)
1545 {
1546 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1547 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1548 	struct sk_psock *psock;
1549 	unsigned char control = 0;
1550 	ssize_t decrypted = 0;
1551 	struct strp_msg *rxm;
1552 	struct sk_buff *skb;
1553 	ssize_t copied = 0;
1554 	bool cmsg = false;
1555 	int target, err = 0;
1556 	long timeo;
1557 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1558 	bool is_peek = flags & MSG_PEEK;
1559 	int num_async = 0;
1560 
1561 	flags |= nonblock;
1562 
1563 	if (unlikely(flags & MSG_ERRQUEUE))
1564 		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1565 
1566 	psock = sk_psock_get(sk);
1567 	lock_sock(sk);
1568 
1569 	/* Process pending decrypted records. It must be non-zero-copy */
1570 	err = process_rx_list(ctx, msg, 0, len, false, is_peek);
1571 	if (err < 0) {
1572 		tls_err_abort(sk, err);
1573 		goto end;
1574 	} else {
1575 		copied = err;
1576 	}
1577 
1578 	len = len - copied;
1579 	if (len) {
1580 		target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1581 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1582 	} else {
1583 		goto recv_end;
1584 	}
1585 
1586 	do {
1587 		bool retain_skb = false;
1588 		bool async = false;
1589 		bool zc = false;
1590 		int to_decrypt;
1591 		int chunk = 0;
1592 
1593 		skb = tls_wait_data(sk, psock, flags, timeo, &err);
1594 		if (!skb) {
1595 			if (psock) {
1596 				int ret = __tcp_bpf_recvmsg(sk, psock,
1597 							    msg, len, flags);
1598 
1599 				if (ret > 0) {
1600 					decrypted += ret;
1601 					len -= ret;
1602 					continue;
1603 				}
1604 			}
1605 			goto recv_end;
1606 		}
1607 
1608 		rxm = strp_msg(skb);
1609 
1610 		if (!cmsg) {
1611 			int cerr;
1612 
1613 			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1614 					sizeof(ctx->control), &ctx->control);
1615 			cmsg = true;
1616 			control = ctx->control;
1617 			if (ctx->control != TLS_RECORD_TYPE_DATA) {
1618 				if (cerr || msg->msg_flags & MSG_CTRUNC) {
1619 					err = -EIO;
1620 					goto recv_end;
1621 				}
1622 			}
1623 		} else if (control != ctx->control) {
1624 			goto recv_end;
1625 		}
1626 
1627 		to_decrypt = rxm->full_len - tls_ctx->rx.overhead_size;
1628 
1629 		if (to_decrypt <= len && !is_kvec && !is_peek)
1630 			zc = true;
1631 
1632 		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1633 					 &chunk, &zc, ctx->async_capable);
1634 		if (err < 0 && err != -EINPROGRESS) {
1635 			tls_err_abort(sk, EBADMSG);
1636 			goto recv_end;
1637 		}
1638 
1639 		if (err == -EINPROGRESS) {
1640 			async = true;
1641 			num_async++;
1642 			goto pick_next_record;
1643 		} else {
1644 			if (!zc) {
1645 				if (rxm->full_len > len) {
1646 					retain_skb = true;
1647 					chunk = len;
1648 				} else {
1649 					chunk = rxm->full_len;
1650 				}
1651 
1652 				err = skb_copy_datagram_msg(skb, rxm->offset,
1653 							    msg, chunk);
1654 				if (err < 0)
1655 					goto recv_end;
1656 
1657 				if (!is_peek) {
1658 					rxm->offset = rxm->offset + chunk;
1659 					rxm->full_len = rxm->full_len - chunk;
1660 				}
1661 			}
1662 		}
1663 
1664 pick_next_record:
1665 		if (chunk > len)
1666 			chunk = len;
1667 
1668 		decrypted += chunk;
1669 		len -= chunk;
1670 
1671 		/* For async or peek case, queue the current skb */
1672 		if (async || is_peek || retain_skb) {
1673 			skb_queue_tail(&ctx->rx_list, skb);
1674 			skb = NULL;
1675 		}
1676 
1677 		if (tls_sw_advance_skb(sk, skb, chunk)) {
1678 			/* Return full control message to
1679 			 * userspace before trying to parse
1680 			 * another message type
1681 			 */
1682 			msg->msg_flags |= MSG_EOR;
1683 			if (ctx->control != TLS_RECORD_TYPE_DATA)
1684 				goto recv_end;
1685 		} else {
1686 			break;
1687 		}
1688 
1689 		/* If we have a new message from strparser, continue now. */
1690 		if (decrypted >= target && !ctx->recv_pkt)
1691 			break;
1692 	} while (len);
1693 
1694 recv_end:
1695 	if (num_async) {
1696 		/* Wait for all previously submitted records to be decrypted */
1697 		smp_store_mb(ctx->async_notify, true);
1698 		if (atomic_read(&ctx->decrypt_pending)) {
1699 			err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1700 			if (err) {
1701 				/* one of async decrypt failed */
1702 				tls_err_abort(sk, err);
1703 				copied = 0;
1704 				decrypted = 0;
1705 				goto end;
1706 			}
1707 		} else {
1708 			reinit_completion(&ctx->async_wait.completion);
1709 		}
1710 		WRITE_ONCE(ctx->async_notify, false);
1711 
1712 		/* Drain records from the rx_list & copy if required */
1713 		if (is_peek || is_kvec)
1714 			err = process_rx_list(ctx, msg, copied,
1715 					      decrypted, false, is_peek);
1716 		else
1717 			err = process_rx_list(ctx, msg, 0,
1718 					      decrypted, true, is_peek);
1719 		if (err < 0) {
1720 			tls_err_abort(sk, err);
1721 			copied = 0;
1722 			goto end;
1723 		}
1724 
1725 		WARN_ON(decrypted != err);
1726 	}
1727 
1728 	copied += decrypted;
1729 
1730 end:
1731 	release_sock(sk);
1732 	if (psock)
1733 		sk_psock_put(sk, psock);
1734 	return copied ? : err;
1735 }
1736 
1737 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
1738 			   struct pipe_inode_info *pipe,
1739 			   size_t len, unsigned int flags)
1740 {
1741 	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1742 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1743 	struct strp_msg *rxm = NULL;
1744 	struct sock *sk = sock->sk;
1745 	struct sk_buff *skb;
1746 	ssize_t copied = 0;
1747 	int err = 0;
1748 	long timeo;
1749 	int chunk;
1750 	bool zc = false;
1751 
1752 	lock_sock(sk);
1753 
1754 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1755 
1756 	skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1757 	if (!skb)
1758 		goto splice_read_end;
1759 
1760 	/* splice does not support reading control messages */
1761 	if (ctx->control != TLS_RECORD_TYPE_DATA) {
1762 		err = -ENOTSUPP;
1763 		goto splice_read_end;
1764 	}
1765 
1766 	if (!ctx->decrypted) {
1767 		err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
1768 
1769 		if (err < 0) {
1770 			tls_err_abort(sk, EBADMSG);
1771 			goto splice_read_end;
1772 		}
1773 		ctx->decrypted = true;
1774 	}
1775 	rxm = strp_msg(skb);
1776 
1777 	chunk = min_t(unsigned int, rxm->full_len, len);
1778 	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
1779 	if (copied < 0)
1780 		goto splice_read_end;
1781 
1782 	if (likely(!(flags & MSG_PEEK)))
1783 		tls_sw_advance_skb(sk, skb, copied);
1784 
1785 splice_read_end:
1786 	release_sock(sk);
1787 	return copied ? : err;
1788 }
1789 
1790 bool tls_sw_stream_read(const struct sock *sk)
1791 {
1792 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1793 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1794 	bool ingress_empty = true;
1795 	struct sk_psock *psock;
1796 
1797 	rcu_read_lock();
1798 	psock = sk_psock(sk);
1799 	if (psock)
1800 		ingress_empty = list_empty(&psock->ingress_msg);
1801 	rcu_read_unlock();
1802 
1803 	return !ingress_empty || ctx->recv_pkt;
1804 }
1805 
1806 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
1807 {
1808 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
1809 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1810 	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
1811 	struct strp_msg *rxm = strp_msg(skb);
1812 	size_t cipher_overhead;
1813 	size_t data_len = 0;
1814 	int ret;
1815 
1816 	/* Verify that we have a full TLS header, or wait for more data */
1817 	if (rxm->offset + tls_ctx->rx.prepend_size > skb->len)
1818 		return 0;
1819 
1820 	/* Sanity-check size of on-stack buffer. */
1821 	if (WARN_ON(tls_ctx->rx.prepend_size > sizeof(header))) {
1822 		ret = -EINVAL;
1823 		goto read_failure;
1824 	}
1825 
1826 	/* Linearize header to local buffer */
1827 	ret = skb_copy_bits(skb, rxm->offset, header, tls_ctx->rx.prepend_size);
1828 
1829 	if (ret < 0)
1830 		goto read_failure;
1831 
1832 	ctx->control = header[0];
1833 
1834 	data_len = ((header[4] & 0xFF) | (header[3] << 8));
1835 
1836 	cipher_overhead = tls_ctx->rx.tag_size + tls_ctx->rx.iv_size;
1837 
1838 	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead) {
1839 		ret = -EMSGSIZE;
1840 		goto read_failure;
1841 	}
1842 	if (data_len < cipher_overhead) {
1843 		ret = -EBADMSG;
1844 		goto read_failure;
1845 	}
1846 
1847 	if (header[1] != TLS_VERSION_MINOR(tls_ctx->crypto_recv.info.version) ||
1848 	    header[2] != TLS_VERSION_MAJOR(tls_ctx->crypto_recv.info.version)) {
1849 		ret = -EINVAL;
1850 		goto read_failure;
1851 	}
1852 
1853 #ifdef CONFIG_TLS_DEVICE
1854 	handle_device_resync(strp->sk, TCP_SKB_CB(skb)->seq + rxm->offset,
1855 			     *(u64*)tls_ctx->rx.rec_seq);
1856 #endif
1857 	return data_len + TLS_HEADER_SIZE;
1858 
1859 read_failure:
1860 	tls_err_abort(strp->sk, ret);
1861 
1862 	return ret;
1863 }
1864 
1865 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
1866 {
1867 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
1868 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1869 
1870 	ctx->decrypted = false;
1871 
1872 	ctx->recv_pkt = skb;
1873 	strp_pause(strp);
1874 
1875 	ctx->saved_data_ready(strp->sk);
1876 }
1877 
1878 static void tls_data_ready(struct sock *sk)
1879 {
1880 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1881 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1882 	struct sk_psock *psock;
1883 
1884 	strp_data_ready(&ctx->strp);
1885 
1886 	psock = sk_psock_get(sk);
1887 	if (psock && !list_empty(&psock->ingress_msg)) {
1888 		ctx->saved_data_ready(sk);
1889 		sk_psock_put(sk, psock);
1890 	}
1891 }
1892 
1893 void tls_sw_free_resources_tx(struct sock *sk)
1894 {
1895 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1896 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1897 	struct tls_rec *rec, *tmp;
1898 
1899 	/* Wait for any pending async encryptions to complete */
1900 	smp_store_mb(ctx->async_notify, true);
1901 	if (atomic_read(&ctx->encrypt_pending))
1902 		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1903 
1904 	cancel_delayed_work_sync(&ctx->tx_work.work);
1905 
1906 	/* Tx whatever records we can transmit and abandon the rest */
1907 	tls_tx_records(sk, -1);
1908 
1909 	/* Free up un-sent records in tx_list. First, free
1910 	 * the partially sent record if any at head of tx_list.
1911 	 */
1912 	if (tls_ctx->partially_sent_record) {
1913 		struct scatterlist *sg = tls_ctx->partially_sent_record;
1914 
1915 		while (1) {
1916 			put_page(sg_page(sg));
1917 			sk_mem_uncharge(sk, sg->length);
1918 
1919 			if (sg_is_last(sg))
1920 				break;
1921 			sg++;
1922 		}
1923 
1924 		tls_ctx->partially_sent_record = NULL;
1925 
1926 		rec = list_first_entry(&ctx->tx_list,
1927 				       struct tls_rec, list);
1928 		list_del(&rec->list);
1929 		sk_msg_free(sk, &rec->msg_plaintext);
1930 		kfree(rec);
1931 	}
1932 
1933 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
1934 		list_del(&rec->list);
1935 		sk_msg_free(sk, &rec->msg_encrypted);
1936 		sk_msg_free(sk, &rec->msg_plaintext);
1937 		kfree(rec);
1938 	}
1939 
1940 	crypto_free_aead(ctx->aead_send);
1941 	tls_free_open_rec(sk);
1942 
1943 	kfree(ctx);
1944 }
1945 
1946 void tls_sw_release_resources_rx(struct sock *sk)
1947 {
1948 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1949 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1950 
1951 	if (ctx->aead_recv) {
1952 		kfree_skb(ctx->recv_pkt);
1953 		ctx->recv_pkt = NULL;
1954 		skb_queue_purge(&ctx->rx_list);
1955 		crypto_free_aead(ctx->aead_recv);
1956 		strp_stop(&ctx->strp);
1957 		write_lock_bh(&sk->sk_callback_lock);
1958 		sk->sk_data_ready = ctx->saved_data_ready;
1959 		write_unlock_bh(&sk->sk_callback_lock);
1960 		release_sock(sk);
1961 		strp_done(&ctx->strp);
1962 		lock_sock(sk);
1963 	}
1964 }
1965 
1966 void tls_sw_free_resources_rx(struct sock *sk)
1967 {
1968 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1969 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1970 
1971 	tls_sw_release_resources_rx(sk);
1972 
1973 	kfree(ctx);
1974 }
1975 
1976 /* The work handler to transmitt the encrypted records in tx_list */
1977 static void tx_work_handler(struct work_struct *work)
1978 {
1979 	struct delayed_work *delayed_work = to_delayed_work(work);
1980 	struct tx_work *tx_work = container_of(delayed_work,
1981 					       struct tx_work, work);
1982 	struct sock *sk = tx_work->sk;
1983 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1984 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1985 
1986 	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
1987 		return;
1988 
1989 	lock_sock(sk);
1990 	tls_tx_records(sk, -1);
1991 	release_sock(sk);
1992 }
1993 
1994 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
1995 {
1996 	struct tls_crypto_info *crypto_info;
1997 	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
1998 	struct tls_sw_context_tx *sw_ctx_tx = NULL;
1999 	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2000 	struct cipher_context *cctx;
2001 	struct crypto_aead **aead;
2002 	struct strp_callbacks cb;
2003 	u16 nonce_size, tag_size, iv_size, rec_seq_size;
2004 	struct crypto_tfm *tfm;
2005 	char *iv, *rec_seq;
2006 	int rc = 0;
2007 
2008 	if (!ctx) {
2009 		rc = -EINVAL;
2010 		goto out;
2011 	}
2012 
2013 	if (tx) {
2014 		if (!ctx->priv_ctx_tx) {
2015 			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2016 			if (!sw_ctx_tx) {
2017 				rc = -ENOMEM;
2018 				goto out;
2019 			}
2020 			ctx->priv_ctx_tx = sw_ctx_tx;
2021 		} else {
2022 			sw_ctx_tx =
2023 				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2024 		}
2025 	} else {
2026 		if (!ctx->priv_ctx_rx) {
2027 			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2028 			if (!sw_ctx_rx) {
2029 				rc = -ENOMEM;
2030 				goto out;
2031 			}
2032 			ctx->priv_ctx_rx = sw_ctx_rx;
2033 		} else {
2034 			sw_ctx_rx =
2035 				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2036 		}
2037 	}
2038 
2039 	if (tx) {
2040 		crypto_init_wait(&sw_ctx_tx->async_wait);
2041 		crypto_info = &ctx->crypto_send.info;
2042 		cctx = &ctx->tx;
2043 		aead = &sw_ctx_tx->aead_send;
2044 		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2045 		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2046 		sw_ctx_tx->tx_work.sk = sk;
2047 	} else {
2048 		crypto_init_wait(&sw_ctx_rx->async_wait);
2049 		crypto_info = &ctx->crypto_recv.info;
2050 		cctx = &ctx->rx;
2051 		skb_queue_head_init(&sw_ctx_rx->rx_list);
2052 		aead = &sw_ctx_rx->aead_recv;
2053 	}
2054 
2055 	switch (crypto_info->cipher_type) {
2056 	case TLS_CIPHER_AES_GCM_128: {
2057 		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2058 		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2059 		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2060 		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2061 		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2062 		rec_seq =
2063 		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2064 		gcm_128_info =
2065 			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2066 		break;
2067 	}
2068 	default:
2069 		rc = -EINVAL;
2070 		goto free_priv;
2071 	}
2072 
2073 	/* Sanity-check the IV size for stack allocations. */
2074 	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE) {
2075 		rc = -EINVAL;
2076 		goto free_priv;
2077 	}
2078 
2079 	cctx->prepend_size = TLS_HEADER_SIZE + nonce_size;
2080 	cctx->tag_size = tag_size;
2081 	cctx->overhead_size = cctx->prepend_size + cctx->tag_size;
2082 	cctx->iv_size = iv_size;
2083 	cctx->iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
2084 			   GFP_KERNEL);
2085 	if (!cctx->iv) {
2086 		rc = -ENOMEM;
2087 		goto free_priv;
2088 	}
2089 	memcpy(cctx->iv, gcm_128_info->salt, TLS_CIPHER_AES_GCM_128_SALT_SIZE);
2090 	memcpy(cctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
2091 	cctx->rec_seq_size = rec_seq_size;
2092 	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2093 	if (!cctx->rec_seq) {
2094 		rc = -ENOMEM;
2095 		goto free_iv;
2096 	}
2097 
2098 	if (!*aead) {
2099 		*aead = crypto_alloc_aead("gcm(aes)", 0, 0);
2100 		if (IS_ERR(*aead)) {
2101 			rc = PTR_ERR(*aead);
2102 			*aead = NULL;
2103 			goto free_rec_seq;
2104 		}
2105 	}
2106 
2107 	ctx->push_pending_record = tls_sw_push_pending_record;
2108 
2109 	rc = crypto_aead_setkey(*aead, gcm_128_info->key,
2110 				TLS_CIPHER_AES_GCM_128_KEY_SIZE);
2111 	if (rc)
2112 		goto free_aead;
2113 
2114 	rc = crypto_aead_setauthsize(*aead, cctx->tag_size);
2115 	if (rc)
2116 		goto free_aead;
2117 
2118 	if (sw_ctx_rx) {
2119 		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2120 		sw_ctx_rx->async_capable =
2121 			tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC;
2122 
2123 		/* Set up strparser */
2124 		memset(&cb, 0, sizeof(cb));
2125 		cb.rcv_msg = tls_queue;
2126 		cb.parse_msg = tls_read_size;
2127 
2128 		strp_init(&sw_ctx_rx->strp, sk, &cb);
2129 
2130 		write_lock_bh(&sk->sk_callback_lock);
2131 		sw_ctx_rx->saved_data_ready = sk->sk_data_ready;
2132 		sk->sk_data_ready = tls_data_ready;
2133 		write_unlock_bh(&sk->sk_callback_lock);
2134 
2135 		strp_check_rcv(&sw_ctx_rx->strp);
2136 	}
2137 
2138 	goto out;
2139 
2140 free_aead:
2141 	crypto_free_aead(*aead);
2142 	*aead = NULL;
2143 free_rec_seq:
2144 	kfree(cctx->rec_seq);
2145 	cctx->rec_seq = NULL;
2146 free_iv:
2147 	kfree(cctx->iv);
2148 	cctx->iv = NULL;
2149 free_priv:
2150 	if (tx) {
2151 		kfree(ctx->priv_ctx_tx);
2152 		ctx->priv_ctx_tx = NULL;
2153 	} else {
2154 		kfree(ctx->priv_ctx_rx);
2155 		ctx->priv_ctx_rx = NULL;
2156 	}
2157 out:
2158 	return rc;
2159 }
2160