xref: /linux/net/tls/tls_sw.c (revision 9e7c9b8eb719835638ee74d93dccc2173581324c)
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/bug.h>
39 #include <linux/sched/signal.h>
40 #include <linux/module.h>
41 #include <linux/splice.h>
42 #include <crypto/aead.h>
43 
44 #include <net/strparser.h>
45 #include <net/tls.h>
46 
47 #include "tls.h"
48 
49 struct tls_decrypt_arg {
50 	struct_group(inargs,
51 	bool zc;
52 	bool async;
53 	u8 tail;
54 	);
55 
56 	struct sk_buff *skb;
57 };
58 
59 struct tls_decrypt_ctx {
60 	u8 iv[MAX_IV_SIZE];
61 	u8 aad[TLS_MAX_AAD_SIZE];
62 	u8 tail;
63 	struct scatterlist sg[];
64 };
65 
66 noinline void tls_err_abort(struct sock *sk, int err)
67 {
68 	WARN_ON_ONCE(err >= 0);
69 	/* sk->sk_err should contain a positive error code. */
70 	sk->sk_err = -err;
71 	sk_error_report(sk);
72 }
73 
74 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
75                      unsigned int recursion_level)
76 {
77         int start = skb_headlen(skb);
78         int i, chunk = start - offset;
79         struct sk_buff *frag_iter;
80         int elt = 0;
81 
82         if (unlikely(recursion_level >= 24))
83                 return -EMSGSIZE;
84 
85         if (chunk > 0) {
86                 if (chunk > len)
87                         chunk = len;
88                 elt++;
89                 len -= chunk;
90                 if (len == 0)
91                         return elt;
92                 offset += chunk;
93         }
94 
95         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
96                 int end;
97 
98                 WARN_ON(start > offset + len);
99 
100                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
101                 chunk = end - offset;
102                 if (chunk > 0) {
103                         if (chunk > len)
104                                 chunk = len;
105                         elt++;
106                         len -= chunk;
107                         if (len == 0)
108                                 return elt;
109                         offset += chunk;
110                 }
111                 start = end;
112         }
113 
114         if (unlikely(skb_has_frag_list(skb))) {
115                 skb_walk_frags(skb, frag_iter) {
116                         int end, ret;
117 
118                         WARN_ON(start > offset + len);
119 
120                         end = start + frag_iter->len;
121                         chunk = end - offset;
122                         if (chunk > 0) {
123                                 if (chunk > len)
124                                         chunk = len;
125                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
126                                                 recursion_level + 1);
127                                 if (unlikely(ret < 0))
128                                         return ret;
129                                 elt += ret;
130                                 len -= chunk;
131                                 if (len == 0)
132                                         return elt;
133                                 offset += chunk;
134                         }
135                         start = end;
136                 }
137         }
138         BUG_ON(len);
139         return elt;
140 }
141 
142 /* Return the number of scatterlist elements required to completely map the
143  * skb, or -EMSGSIZE if the recursion depth is exceeded.
144  */
145 static int skb_nsg(struct sk_buff *skb, int offset, int len)
146 {
147         return __skb_nsg(skb, offset, len, 0);
148 }
149 
150 static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
151 			      struct tls_decrypt_arg *darg)
152 {
153 	struct strp_msg *rxm = strp_msg(skb);
154 	struct tls_msg *tlm = tls_msg(skb);
155 	int sub = 0;
156 
157 	/* Determine zero-padding length */
158 	if (prot->version == TLS_1_3_VERSION) {
159 		int offset = rxm->full_len - TLS_TAG_SIZE - 1;
160 		char content_type = darg->zc ? darg->tail : 0;
161 		int err;
162 
163 		while (content_type == 0) {
164 			if (offset < prot->prepend_size)
165 				return -EBADMSG;
166 			err = skb_copy_bits(skb, rxm->offset + offset,
167 					    &content_type, 1);
168 			if (err)
169 				return err;
170 			if (content_type)
171 				break;
172 			sub++;
173 			offset--;
174 		}
175 		tlm->control = content_type;
176 	}
177 	return sub;
178 }
179 
180 static void tls_decrypt_done(struct crypto_async_request *req, int err)
181 {
182 	struct aead_request *aead_req = (struct aead_request *)req;
183 	struct scatterlist *sgout = aead_req->dst;
184 	struct scatterlist *sgin = aead_req->src;
185 	struct tls_sw_context_rx *ctx;
186 	struct tls_context *tls_ctx;
187 	struct scatterlist *sg;
188 	unsigned int pages;
189 	struct sock *sk;
190 
191 	sk = (struct sock *)req->data;
192 	tls_ctx = tls_get_ctx(sk);
193 	ctx = tls_sw_ctx_rx(tls_ctx);
194 
195 	/* Propagate if there was an err */
196 	if (err) {
197 		if (err == -EBADMSG)
198 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
199 		ctx->async_wait.err = err;
200 		tls_err_abort(sk, err);
201 	}
202 
203 	/* Free the destination pages if skb was not decrypted inplace */
204 	if (sgout != sgin) {
205 		/* Skip the first S/G entry as it points to AAD */
206 		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
207 			if (!sg)
208 				break;
209 			put_page(sg_page(sg));
210 		}
211 	}
212 
213 	kfree(aead_req);
214 
215 	spin_lock_bh(&ctx->decrypt_compl_lock);
216 	if (!atomic_dec_return(&ctx->decrypt_pending))
217 		complete(&ctx->async_wait.completion);
218 	spin_unlock_bh(&ctx->decrypt_compl_lock);
219 }
220 
221 static int tls_do_decryption(struct sock *sk,
222 			     struct scatterlist *sgin,
223 			     struct scatterlist *sgout,
224 			     char *iv_recv,
225 			     size_t data_len,
226 			     struct aead_request *aead_req,
227 			     struct tls_decrypt_arg *darg)
228 {
229 	struct tls_context *tls_ctx = tls_get_ctx(sk);
230 	struct tls_prot_info *prot = &tls_ctx->prot_info;
231 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
232 	int ret;
233 
234 	aead_request_set_tfm(aead_req, ctx->aead_recv);
235 	aead_request_set_ad(aead_req, prot->aad_size);
236 	aead_request_set_crypt(aead_req, sgin, sgout,
237 			       data_len + prot->tag_size,
238 			       (u8 *)iv_recv);
239 
240 	if (darg->async) {
241 		aead_request_set_callback(aead_req,
242 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
243 					  tls_decrypt_done, sk);
244 		atomic_inc(&ctx->decrypt_pending);
245 	} else {
246 		aead_request_set_callback(aead_req,
247 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
248 					  crypto_req_done, &ctx->async_wait);
249 	}
250 
251 	ret = crypto_aead_decrypt(aead_req);
252 	if (ret == -EINPROGRESS) {
253 		if (darg->async)
254 			return 0;
255 
256 		ret = crypto_wait_req(ret, &ctx->async_wait);
257 	}
258 	darg->async = false;
259 
260 	return ret;
261 }
262 
263 static void tls_trim_both_msgs(struct sock *sk, int target_size)
264 {
265 	struct tls_context *tls_ctx = tls_get_ctx(sk);
266 	struct tls_prot_info *prot = &tls_ctx->prot_info;
267 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
268 	struct tls_rec *rec = ctx->open_rec;
269 
270 	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
271 	if (target_size > 0)
272 		target_size += prot->overhead_size;
273 	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
274 }
275 
276 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
277 {
278 	struct tls_context *tls_ctx = tls_get_ctx(sk);
279 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
280 	struct tls_rec *rec = ctx->open_rec;
281 	struct sk_msg *msg_en = &rec->msg_encrypted;
282 
283 	return sk_msg_alloc(sk, msg_en, len, 0);
284 }
285 
286 static int tls_clone_plaintext_msg(struct sock *sk, int required)
287 {
288 	struct tls_context *tls_ctx = tls_get_ctx(sk);
289 	struct tls_prot_info *prot = &tls_ctx->prot_info;
290 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
291 	struct tls_rec *rec = ctx->open_rec;
292 	struct sk_msg *msg_pl = &rec->msg_plaintext;
293 	struct sk_msg *msg_en = &rec->msg_encrypted;
294 	int skip, len;
295 
296 	/* We add page references worth len bytes from encrypted sg
297 	 * at the end of plaintext sg. It is guaranteed that msg_en
298 	 * has enough required room (ensured by caller).
299 	 */
300 	len = required - msg_pl->sg.size;
301 
302 	/* Skip initial bytes in msg_en's data to be able to use
303 	 * same offset of both plain and encrypted data.
304 	 */
305 	skip = prot->prepend_size + msg_pl->sg.size;
306 
307 	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
308 }
309 
310 static struct tls_rec *tls_get_rec(struct sock *sk)
311 {
312 	struct tls_context *tls_ctx = tls_get_ctx(sk);
313 	struct tls_prot_info *prot = &tls_ctx->prot_info;
314 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
315 	struct sk_msg *msg_pl, *msg_en;
316 	struct tls_rec *rec;
317 	int mem_size;
318 
319 	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
320 
321 	rec = kzalloc(mem_size, sk->sk_allocation);
322 	if (!rec)
323 		return NULL;
324 
325 	msg_pl = &rec->msg_plaintext;
326 	msg_en = &rec->msg_encrypted;
327 
328 	sk_msg_init(msg_pl);
329 	sk_msg_init(msg_en);
330 
331 	sg_init_table(rec->sg_aead_in, 2);
332 	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
333 	sg_unmark_end(&rec->sg_aead_in[1]);
334 
335 	sg_init_table(rec->sg_aead_out, 2);
336 	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
337 	sg_unmark_end(&rec->sg_aead_out[1]);
338 
339 	return rec;
340 }
341 
342 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
343 {
344 	sk_msg_free(sk, &rec->msg_encrypted);
345 	sk_msg_free(sk, &rec->msg_plaintext);
346 	kfree(rec);
347 }
348 
349 static void tls_free_open_rec(struct sock *sk)
350 {
351 	struct tls_context *tls_ctx = tls_get_ctx(sk);
352 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
353 	struct tls_rec *rec = ctx->open_rec;
354 
355 	if (rec) {
356 		tls_free_rec(sk, rec);
357 		ctx->open_rec = NULL;
358 	}
359 }
360 
361 int tls_tx_records(struct sock *sk, int flags)
362 {
363 	struct tls_context *tls_ctx = tls_get_ctx(sk);
364 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
365 	struct tls_rec *rec, *tmp;
366 	struct sk_msg *msg_en;
367 	int tx_flags, rc = 0;
368 
369 	if (tls_is_partially_sent_record(tls_ctx)) {
370 		rec = list_first_entry(&ctx->tx_list,
371 				       struct tls_rec, list);
372 
373 		if (flags == -1)
374 			tx_flags = rec->tx_flags;
375 		else
376 			tx_flags = flags;
377 
378 		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
379 		if (rc)
380 			goto tx_err;
381 
382 		/* Full record has been transmitted.
383 		 * Remove the head of tx_list
384 		 */
385 		list_del(&rec->list);
386 		sk_msg_free(sk, &rec->msg_plaintext);
387 		kfree(rec);
388 	}
389 
390 	/* Tx all ready records */
391 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
392 		if (READ_ONCE(rec->tx_ready)) {
393 			if (flags == -1)
394 				tx_flags = rec->tx_flags;
395 			else
396 				tx_flags = flags;
397 
398 			msg_en = &rec->msg_encrypted;
399 			rc = tls_push_sg(sk, tls_ctx,
400 					 &msg_en->sg.data[msg_en->sg.curr],
401 					 0, tx_flags);
402 			if (rc)
403 				goto tx_err;
404 
405 			list_del(&rec->list);
406 			sk_msg_free(sk, &rec->msg_plaintext);
407 			kfree(rec);
408 		} else {
409 			break;
410 		}
411 	}
412 
413 tx_err:
414 	if (rc < 0 && rc != -EAGAIN)
415 		tls_err_abort(sk, -EBADMSG);
416 
417 	return rc;
418 }
419 
420 static void tls_encrypt_done(struct crypto_async_request *req, int err)
421 {
422 	struct aead_request *aead_req = (struct aead_request *)req;
423 	struct sock *sk = req->data;
424 	struct tls_context *tls_ctx = tls_get_ctx(sk);
425 	struct tls_prot_info *prot = &tls_ctx->prot_info;
426 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
427 	struct scatterlist *sge;
428 	struct sk_msg *msg_en;
429 	struct tls_rec *rec;
430 	bool ready = false;
431 	int pending;
432 
433 	rec = container_of(aead_req, struct tls_rec, aead_req);
434 	msg_en = &rec->msg_encrypted;
435 
436 	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
437 	sge->offset -= prot->prepend_size;
438 	sge->length += prot->prepend_size;
439 
440 	/* Check if error is previously set on socket */
441 	if (err || sk->sk_err) {
442 		rec = NULL;
443 
444 		/* If err is already set on socket, return the same code */
445 		if (sk->sk_err) {
446 			ctx->async_wait.err = -sk->sk_err;
447 		} else {
448 			ctx->async_wait.err = err;
449 			tls_err_abort(sk, err);
450 		}
451 	}
452 
453 	if (rec) {
454 		struct tls_rec *first_rec;
455 
456 		/* Mark the record as ready for transmission */
457 		smp_store_mb(rec->tx_ready, true);
458 
459 		/* If received record is at head of tx_list, schedule tx */
460 		first_rec = list_first_entry(&ctx->tx_list,
461 					     struct tls_rec, list);
462 		if (rec == first_rec)
463 			ready = true;
464 	}
465 
466 	spin_lock_bh(&ctx->encrypt_compl_lock);
467 	pending = atomic_dec_return(&ctx->encrypt_pending);
468 
469 	if (!pending && ctx->async_notify)
470 		complete(&ctx->async_wait.completion);
471 	spin_unlock_bh(&ctx->encrypt_compl_lock);
472 
473 	if (!ready)
474 		return;
475 
476 	/* Schedule the transmission */
477 	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
478 		schedule_delayed_work(&ctx->tx_work.work, 1);
479 }
480 
481 static int tls_do_encryption(struct sock *sk,
482 			     struct tls_context *tls_ctx,
483 			     struct tls_sw_context_tx *ctx,
484 			     struct aead_request *aead_req,
485 			     size_t data_len, u32 start)
486 {
487 	struct tls_prot_info *prot = &tls_ctx->prot_info;
488 	struct tls_rec *rec = ctx->open_rec;
489 	struct sk_msg *msg_en = &rec->msg_encrypted;
490 	struct scatterlist *sge = sk_msg_elem(msg_en, start);
491 	int rc, iv_offset = 0;
492 
493 	/* For CCM based ciphers, first byte of IV is a constant */
494 	switch (prot->cipher_type) {
495 	case TLS_CIPHER_AES_CCM_128:
496 		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
497 		iv_offset = 1;
498 		break;
499 	case TLS_CIPHER_SM4_CCM:
500 		rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
501 		iv_offset = 1;
502 		break;
503 	}
504 
505 	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
506 	       prot->iv_size + prot->salt_size);
507 
508 	tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
509 			    tls_ctx->tx.rec_seq);
510 
511 	sge->offset += prot->prepend_size;
512 	sge->length -= prot->prepend_size;
513 
514 	msg_en->sg.curr = start;
515 
516 	aead_request_set_tfm(aead_req, ctx->aead_send);
517 	aead_request_set_ad(aead_req, prot->aad_size);
518 	aead_request_set_crypt(aead_req, rec->sg_aead_in,
519 			       rec->sg_aead_out,
520 			       data_len, rec->iv_data);
521 
522 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
523 				  tls_encrypt_done, sk);
524 
525 	/* Add the record in tx_list */
526 	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
527 	atomic_inc(&ctx->encrypt_pending);
528 
529 	rc = crypto_aead_encrypt(aead_req);
530 	if (!rc || rc != -EINPROGRESS) {
531 		atomic_dec(&ctx->encrypt_pending);
532 		sge->offset -= prot->prepend_size;
533 		sge->length += prot->prepend_size;
534 	}
535 
536 	if (!rc) {
537 		WRITE_ONCE(rec->tx_ready, true);
538 	} else if (rc != -EINPROGRESS) {
539 		list_del(&rec->list);
540 		return rc;
541 	}
542 
543 	/* Unhook the record from context if encryption is not failure */
544 	ctx->open_rec = NULL;
545 	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
546 	return rc;
547 }
548 
549 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
550 				 struct tls_rec **to, struct sk_msg *msg_opl,
551 				 struct sk_msg *msg_oen, u32 split_point,
552 				 u32 tx_overhead_size, u32 *orig_end)
553 {
554 	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
555 	struct scatterlist *sge, *osge, *nsge;
556 	u32 orig_size = msg_opl->sg.size;
557 	struct scatterlist tmp = { };
558 	struct sk_msg *msg_npl;
559 	struct tls_rec *new;
560 	int ret;
561 
562 	new = tls_get_rec(sk);
563 	if (!new)
564 		return -ENOMEM;
565 	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
566 			   tx_overhead_size, 0);
567 	if (ret < 0) {
568 		tls_free_rec(sk, new);
569 		return ret;
570 	}
571 
572 	*orig_end = msg_opl->sg.end;
573 	i = msg_opl->sg.start;
574 	sge = sk_msg_elem(msg_opl, i);
575 	while (apply && sge->length) {
576 		if (sge->length > apply) {
577 			u32 len = sge->length - apply;
578 
579 			get_page(sg_page(sge));
580 			sg_set_page(&tmp, sg_page(sge), len,
581 				    sge->offset + apply);
582 			sge->length = apply;
583 			bytes += apply;
584 			apply = 0;
585 		} else {
586 			apply -= sge->length;
587 			bytes += sge->length;
588 		}
589 
590 		sk_msg_iter_var_next(i);
591 		if (i == msg_opl->sg.end)
592 			break;
593 		sge = sk_msg_elem(msg_opl, i);
594 	}
595 
596 	msg_opl->sg.end = i;
597 	msg_opl->sg.curr = i;
598 	msg_opl->sg.copybreak = 0;
599 	msg_opl->apply_bytes = 0;
600 	msg_opl->sg.size = bytes;
601 
602 	msg_npl = &new->msg_plaintext;
603 	msg_npl->apply_bytes = apply;
604 	msg_npl->sg.size = orig_size - bytes;
605 
606 	j = msg_npl->sg.start;
607 	nsge = sk_msg_elem(msg_npl, j);
608 	if (tmp.length) {
609 		memcpy(nsge, &tmp, sizeof(*nsge));
610 		sk_msg_iter_var_next(j);
611 		nsge = sk_msg_elem(msg_npl, j);
612 	}
613 
614 	osge = sk_msg_elem(msg_opl, i);
615 	while (osge->length) {
616 		memcpy(nsge, osge, sizeof(*nsge));
617 		sg_unmark_end(nsge);
618 		sk_msg_iter_var_next(i);
619 		sk_msg_iter_var_next(j);
620 		if (i == *orig_end)
621 			break;
622 		osge = sk_msg_elem(msg_opl, i);
623 		nsge = sk_msg_elem(msg_npl, j);
624 	}
625 
626 	msg_npl->sg.end = j;
627 	msg_npl->sg.curr = j;
628 	msg_npl->sg.copybreak = 0;
629 
630 	*to = new;
631 	return 0;
632 }
633 
634 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
635 				  struct tls_rec *from, u32 orig_end)
636 {
637 	struct sk_msg *msg_npl = &from->msg_plaintext;
638 	struct sk_msg *msg_opl = &to->msg_plaintext;
639 	struct scatterlist *osge, *nsge;
640 	u32 i, j;
641 
642 	i = msg_opl->sg.end;
643 	sk_msg_iter_var_prev(i);
644 	j = msg_npl->sg.start;
645 
646 	osge = sk_msg_elem(msg_opl, i);
647 	nsge = sk_msg_elem(msg_npl, j);
648 
649 	if (sg_page(osge) == sg_page(nsge) &&
650 	    osge->offset + osge->length == nsge->offset) {
651 		osge->length += nsge->length;
652 		put_page(sg_page(nsge));
653 	}
654 
655 	msg_opl->sg.end = orig_end;
656 	msg_opl->sg.curr = orig_end;
657 	msg_opl->sg.copybreak = 0;
658 	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
659 	msg_opl->sg.size += msg_npl->sg.size;
660 
661 	sk_msg_free(sk, &to->msg_encrypted);
662 	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
663 
664 	kfree(from);
665 }
666 
667 static int tls_push_record(struct sock *sk, int flags,
668 			   unsigned char record_type)
669 {
670 	struct tls_context *tls_ctx = tls_get_ctx(sk);
671 	struct tls_prot_info *prot = &tls_ctx->prot_info;
672 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
673 	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
674 	u32 i, split_point, orig_end;
675 	struct sk_msg *msg_pl, *msg_en;
676 	struct aead_request *req;
677 	bool split;
678 	int rc;
679 
680 	if (!rec)
681 		return 0;
682 
683 	msg_pl = &rec->msg_plaintext;
684 	msg_en = &rec->msg_encrypted;
685 
686 	split_point = msg_pl->apply_bytes;
687 	split = split_point && split_point < msg_pl->sg.size;
688 	if (unlikely((!split &&
689 		      msg_pl->sg.size +
690 		      prot->overhead_size > msg_en->sg.size) ||
691 		     (split &&
692 		      split_point +
693 		      prot->overhead_size > msg_en->sg.size))) {
694 		split = true;
695 		split_point = msg_en->sg.size;
696 	}
697 	if (split) {
698 		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
699 					   split_point, prot->overhead_size,
700 					   &orig_end);
701 		if (rc < 0)
702 			return rc;
703 		/* This can happen if above tls_split_open_record allocates
704 		 * a single large encryption buffer instead of two smaller
705 		 * ones. In this case adjust pointers and continue without
706 		 * split.
707 		 */
708 		if (!msg_pl->sg.size) {
709 			tls_merge_open_record(sk, rec, tmp, orig_end);
710 			msg_pl = &rec->msg_plaintext;
711 			msg_en = &rec->msg_encrypted;
712 			split = false;
713 		}
714 		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
715 			    prot->overhead_size);
716 	}
717 
718 	rec->tx_flags = flags;
719 	req = &rec->aead_req;
720 
721 	i = msg_pl->sg.end;
722 	sk_msg_iter_var_prev(i);
723 
724 	rec->content_type = record_type;
725 	if (prot->version == TLS_1_3_VERSION) {
726 		/* Add content type to end of message.  No padding added */
727 		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
728 		sg_mark_end(&rec->sg_content_type);
729 		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
730 			 &rec->sg_content_type);
731 	} else {
732 		sg_mark_end(sk_msg_elem(msg_pl, i));
733 	}
734 
735 	if (msg_pl->sg.end < msg_pl->sg.start) {
736 		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
737 			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
738 			 msg_pl->sg.data);
739 	}
740 
741 	i = msg_pl->sg.start;
742 	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
743 
744 	i = msg_en->sg.end;
745 	sk_msg_iter_var_prev(i);
746 	sg_mark_end(sk_msg_elem(msg_en, i));
747 
748 	i = msg_en->sg.start;
749 	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
750 
751 	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
752 		     tls_ctx->tx.rec_seq, record_type, prot);
753 
754 	tls_fill_prepend(tls_ctx,
755 			 page_address(sg_page(&msg_en->sg.data[i])) +
756 			 msg_en->sg.data[i].offset,
757 			 msg_pl->sg.size + prot->tail_size,
758 			 record_type);
759 
760 	tls_ctx->pending_open_record_frags = false;
761 
762 	rc = tls_do_encryption(sk, tls_ctx, ctx, req,
763 			       msg_pl->sg.size + prot->tail_size, i);
764 	if (rc < 0) {
765 		if (rc != -EINPROGRESS) {
766 			tls_err_abort(sk, -EBADMSG);
767 			if (split) {
768 				tls_ctx->pending_open_record_frags = true;
769 				tls_merge_open_record(sk, rec, tmp, orig_end);
770 			}
771 		}
772 		ctx->async_capable = 1;
773 		return rc;
774 	} else if (split) {
775 		msg_pl = &tmp->msg_plaintext;
776 		msg_en = &tmp->msg_encrypted;
777 		sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
778 		tls_ctx->pending_open_record_frags = true;
779 		ctx->open_rec = tmp;
780 	}
781 
782 	return tls_tx_records(sk, flags);
783 }
784 
785 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
786 			       bool full_record, u8 record_type,
787 			       ssize_t *copied, int flags)
788 {
789 	struct tls_context *tls_ctx = tls_get_ctx(sk);
790 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
791 	struct sk_msg msg_redir = { };
792 	struct sk_psock *psock;
793 	struct sock *sk_redir;
794 	struct tls_rec *rec;
795 	bool enospc, policy;
796 	int err = 0, send;
797 	u32 delta = 0;
798 
799 	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
800 	psock = sk_psock_get(sk);
801 	if (!psock || !policy) {
802 		err = tls_push_record(sk, flags, record_type);
803 		if (err && sk->sk_err == EBADMSG) {
804 			*copied -= sk_msg_free(sk, msg);
805 			tls_free_open_rec(sk);
806 			err = -sk->sk_err;
807 		}
808 		if (psock)
809 			sk_psock_put(sk, psock);
810 		return err;
811 	}
812 more_data:
813 	enospc = sk_msg_full(msg);
814 	if (psock->eval == __SK_NONE) {
815 		delta = msg->sg.size;
816 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
817 		delta -= msg->sg.size;
818 	}
819 	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
820 	    !enospc && !full_record) {
821 		err = -ENOSPC;
822 		goto out_err;
823 	}
824 	msg->cork_bytes = 0;
825 	send = msg->sg.size;
826 	if (msg->apply_bytes && msg->apply_bytes < send)
827 		send = msg->apply_bytes;
828 
829 	switch (psock->eval) {
830 	case __SK_PASS:
831 		err = tls_push_record(sk, flags, record_type);
832 		if (err && sk->sk_err == EBADMSG) {
833 			*copied -= sk_msg_free(sk, msg);
834 			tls_free_open_rec(sk);
835 			err = -sk->sk_err;
836 			goto out_err;
837 		}
838 		break;
839 	case __SK_REDIRECT:
840 		sk_redir = psock->sk_redir;
841 		memcpy(&msg_redir, msg, sizeof(*msg));
842 		if (msg->apply_bytes < send)
843 			msg->apply_bytes = 0;
844 		else
845 			msg->apply_bytes -= send;
846 		sk_msg_return_zero(sk, msg, send);
847 		msg->sg.size -= send;
848 		release_sock(sk);
849 		err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
850 		lock_sock(sk);
851 		if (err < 0) {
852 			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
853 			msg->sg.size = 0;
854 		}
855 		if (msg->sg.size == 0)
856 			tls_free_open_rec(sk);
857 		break;
858 	case __SK_DROP:
859 	default:
860 		sk_msg_free_partial(sk, msg, send);
861 		if (msg->apply_bytes < send)
862 			msg->apply_bytes = 0;
863 		else
864 			msg->apply_bytes -= send;
865 		if (msg->sg.size == 0)
866 			tls_free_open_rec(sk);
867 		*copied -= (send + delta);
868 		err = -EACCES;
869 	}
870 
871 	if (likely(!err)) {
872 		bool reset_eval = !ctx->open_rec;
873 
874 		rec = ctx->open_rec;
875 		if (rec) {
876 			msg = &rec->msg_plaintext;
877 			if (!msg->apply_bytes)
878 				reset_eval = true;
879 		}
880 		if (reset_eval) {
881 			psock->eval = __SK_NONE;
882 			if (psock->sk_redir) {
883 				sock_put(psock->sk_redir);
884 				psock->sk_redir = NULL;
885 			}
886 		}
887 		if (rec)
888 			goto more_data;
889 	}
890  out_err:
891 	sk_psock_put(sk, psock);
892 	return err;
893 }
894 
895 static int tls_sw_push_pending_record(struct sock *sk, int flags)
896 {
897 	struct tls_context *tls_ctx = tls_get_ctx(sk);
898 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
899 	struct tls_rec *rec = ctx->open_rec;
900 	struct sk_msg *msg_pl;
901 	size_t copied;
902 
903 	if (!rec)
904 		return 0;
905 
906 	msg_pl = &rec->msg_plaintext;
907 	copied = msg_pl->sg.size;
908 	if (!copied)
909 		return 0;
910 
911 	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
912 				   &copied, flags);
913 }
914 
915 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
916 {
917 	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
918 	struct tls_context *tls_ctx = tls_get_ctx(sk);
919 	struct tls_prot_info *prot = &tls_ctx->prot_info;
920 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
921 	bool async_capable = ctx->async_capable;
922 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
923 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
924 	bool eor = !(msg->msg_flags & MSG_MORE);
925 	size_t try_to_copy;
926 	ssize_t copied = 0;
927 	struct sk_msg *msg_pl, *msg_en;
928 	struct tls_rec *rec;
929 	int required_size;
930 	int num_async = 0;
931 	bool full_record;
932 	int record_room;
933 	int num_zc = 0;
934 	int orig_size;
935 	int ret = 0;
936 	int pending;
937 
938 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
939 			       MSG_CMSG_COMPAT))
940 		return -EOPNOTSUPP;
941 
942 	mutex_lock(&tls_ctx->tx_lock);
943 	lock_sock(sk);
944 
945 	if (unlikely(msg->msg_controllen)) {
946 		ret = tls_process_cmsg(sk, msg, &record_type);
947 		if (ret) {
948 			if (ret == -EINPROGRESS)
949 				num_async++;
950 			else if (ret != -EAGAIN)
951 				goto send_end;
952 		}
953 	}
954 
955 	while (msg_data_left(msg)) {
956 		if (sk->sk_err) {
957 			ret = -sk->sk_err;
958 			goto send_end;
959 		}
960 
961 		if (ctx->open_rec)
962 			rec = ctx->open_rec;
963 		else
964 			rec = ctx->open_rec = tls_get_rec(sk);
965 		if (!rec) {
966 			ret = -ENOMEM;
967 			goto send_end;
968 		}
969 
970 		msg_pl = &rec->msg_plaintext;
971 		msg_en = &rec->msg_encrypted;
972 
973 		orig_size = msg_pl->sg.size;
974 		full_record = false;
975 		try_to_copy = msg_data_left(msg);
976 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
977 		if (try_to_copy >= record_room) {
978 			try_to_copy = record_room;
979 			full_record = true;
980 		}
981 
982 		required_size = msg_pl->sg.size + try_to_copy +
983 				prot->overhead_size;
984 
985 		if (!sk_stream_memory_free(sk))
986 			goto wait_for_sndbuf;
987 
988 alloc_encrypted:
989 		ret = tls_alloc_encrypted_msg(sk, required_size);
990 		if (ret) {
991 			if (ret != -ENOSPC)
992 				goto wait_for_memory;
993 
994 			/* Adjust try_to_copy according to the amount that was
995 			 * actually allocated. The difference is due
996 			 * to max sg elements limit
997 			 */
998 			try_to_copy -= required_size - msg_en->sg.size;
999 			full_record = true;
1000 		}
1001 
1002 		if (!is_kvec && (full_record || eor) && !async_capable) {
1003 			u32 first = msg_pl->sg.end;
1004 
1005 			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1006 							msg_pl, try_to_copy);
1007 			if (ret)
1008 				goto fallback_to_reg_send;
1009 
1010 			num_zc++;
1011 			copied += try_to_copy;
1012 
1013 			sk_msg_sg_copy_set(msg_pl, first);
1014 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1015 						  record_type, &copied,
1016 						  msg->msg_flags);
1017 			if (ret) {
1018 				if (ret == -EINPROGRESS)
1019 					num_async++;
1020 				else if (ret == -ENOMEM)
1021 					goto wait_for_memory;
1022 				else if (ctx->open_rec && ret == -ENOSPC)
1023 					goto rollback_iter;
1024 				else if (ret != -EAGAIN)
1025 					goto send_end;
1026 			}
1027 			continue;
1028 rollback_iter:
1029 			copied -= try_to_copy;
1030 			sk_msg_sg_copy_clear(msg_pl, first);
1031 			iov_iter_revert(&msg->msg_iter,
1032 					msg_pl->sg.size - orig_size);
1033 fallback_to_reg_send:
1034 			sk_msg_trim(sk, msg_pl, orig_size);
1035 		}
1036 
1037 		required_size = msg_pl->sg.size + try_to_copy;
1038 
1039 		ret = tls_clone_plaintext_msg(sk, required_size);
1040 		if (ret) {
1041 			if (ret != -ENOSPC)
1042 				goto send_end;
1043 
1044 			/* Adjust try_to_copy according to the amount that was
1045 			 * actually allocated. The difference is due
1046 			 * to max sg elements limit
1047 			 */
1048 			try_to_copy -= required_size - msg_pl->sg.size;
1049 			full_record = true;
1050 			sk_msg_trim(sk, msg_en,
1051 				    msg_pl->sg.size + prot->overhead_size);
1052 		}
1053 
1054 		if (try_to_copy) {
1055 			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1056 						       msg_pl, try_to_copy);
1057 			if (ret < 0)
1058 				goto trim_sgl;
1059 		}
1060 
1061 		/* Open records defined only if successfully copied, otherwise
1062 		 * we would trim the sg but not reset the open record frags.
1063 		 */
1064 		tls_ctx->pending_open_record_frags = true;
1065 		copied += try_to_copy;
1066 		if (full_record || eor) {
1067 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1068 						  record_type, &copied,
1069 						  msg->msg_flags);
1070 			if (ret) {
1071 				if (ret == -EINPROGRESS)
1072 					num_async++;
1073 				else if (ret == -ENOMEM)
1074 					goto wait_for_memory;
1075 				else if (ret != -EAGAIN) {
1076 					if (ret == -ENOSPC)
1077 						ret = 0;
1078 					goto send_end;
1079 				}
1080 			}
1081 		}
1082 
1083 		continue;
1084 
1085 wait_for_sndbuf:
1086 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1087 wait_for_memory:
1088 		ret = sk_stream_wait_memory(sk, &timeo);
1089 		if (ret) {
1090 trim_sgl:
1091 			if (ctx->open_rec)
1092 				tls_trim_both_msgs(sk, orig_size);
1093 			goto send_end;
1094 		}
1095 
1096 		if (ctx->open_rec && msg_en->sg.size < required_size)
1097 			goto alloc_encrypted;
1098 	}
1099 
1100 	if (!num_async) {
1101 		goto send_end;
1102 	} else if (num_zc) {
1103 		/* Wait for pending encryptions to get completed */
1104 		spin_lock_bh(&ctx->encrypt_compl_lock);
1105 		ctx->async_notify = true;
1106 
1107 		pending = atomic_read(&ctx->encrypt_pending);
1108 		spin_unlock_bh(&ctx->encrypt_compl_lock);
1109 		if (pending)
1110 			crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1111 		else
1112 			reinit_completion(&ctx->async_wait.completion);
1113 
1114 		/* There can be no concurrent accesses, since we have no
1115 		 * pending encrypt operations
1116 		 */
1117 		WRITE_ONCE(ctx->async_notify, false);
1118 
1119 		if (ctx->async_wait.err) {
1120 			ret = ctx->async_wait.err;
1121 			copied = 0;
1122 		}
1123 	}
1124 
1125 	/* Transmit if any encryptions have completed */
1126 	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1127 		cancel_delayed_work(&ctx->tx_work.work);
1128 		tls_tx_records(sk, msg->msg_flags);
1129 	}
1130 
1131 send_end:
1132 	ret = sk_stream_error(sk, msg->msg_flags, ret);
1133 
1134 	release_sock(sk);
1135 	mutex_unlock(&tls_ctx->tx_lock);
1136 	return copied > 0 ? copied : ret;
1137 }
1138 
1139 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1140 			      int offset, size_t size, int flags)
1141 {
1142 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1143 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1144 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1145 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1146 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1147 	struct sk_msg *msg_pl;
1148 	struct tls_rec *rec;
1149 	int num_async = 0;
1150 	ssize_t copied = 0;
1151 	bool full_record;
1152 	int record_room;
1153 	int ret = 0;
1154 	bool eor;
1155 
1156 	eor = !(flags & MSG_SENDPAGE_NOTLAST);
1157 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1158 
1159 	/* Call the sk_stream functions to manage the sndbuf mem. */
1160 	while (size > 0) {
1161 		size_t copy, required_size;
1162 
1163 		if (sk->sk_err) {
1164 			ret = -sk->sk_err;
1165 			goto sendpage_end;
1166 		}
1167 
1168 		if (ctx->open_rec)
1169 			rec = ctx->open_rec;
1170 		else
1171 			rec = ctx->open_rec = tls_get_rec(sk);
1172 		if (!rec) {
1173 			ret = -ENOMEM;
1174 			goto sendpage_end;
1175 		}
1176 
1177 		msg_pl = &rec->msg_plaintext;
1178 
1179 		full_record = false;
1180 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1181 		copy = size;
1182 		if (copy >= record_room) {
1183 			copy = record_room;
1184 			full_record = true;
1185 		}
1186 
1187 		required_size = msg_pl->sg.size + copy + prot->overhead_size;
1188 
1189 		if (!sk_stream_memory_free(sk))
1190 			goto wait_for_sndbuf;
1191 alloc_payload:
1192 		ret = tls_alloc_encrypted_msg(sk, required_size);
1193 		if (ret) {
1194 			if (ret != -ENOSPC)
1195 				goto wait_for_memory;
1196 
1197 			/* Adjust copy according to the amount that was
1198 			 * actually allocated. The difference is due
1199 			 * to max sg elements limit
1200 			 */
1201 			copy -= required_size - msg_pl->sg.size;
1202 			full_record = true;
1203 		}
1204 
1205 		sk_msg_page_add(msg_pl, page, copy, offset);
1206 		sk_mem_charge(sk, copy);
1207 
1208 		offset += copy;
1209 		size -= copy;
1210 		copied += copy;
1211 
1212 		tls_ctx->pending_open_record_frags = true;
1213 		if (full_record || eor || sk_msg_full(msg_pl)) {
1214 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1215 						  record_type, &copied, flags);
1216 			if (ret) {
1217 				if (ret == -EINPROGRESS)
1218 					num_async++;
1219 				else if (ret == -ENOMEM)
1220 					goto wait_for_memory;
1221 				else if (ret != -EAGAIN) {
1222 					if (ret == -ENOSPC)
1223 						ret = 0;
1224 					goto sendpage_end;
1225 				}
1226 			}
1227 		}
1228 		continue;
1229 wait_for_sndbuf:
1230 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1231 wait_for_memory:
1232 		ret = sk_stream_wait_memory(sk, &timeo);
1233 		if (ret) {
1234 			if (ctx->open_rec)
1235 				tls_trim_both_msgs(sk, msg_pl->sg.size);
1236 			goto sendpage_end;
1237 		}
1238 
1239 		if (ctx->open_rec)
1240 			goto alloc_payload;
1241 	}
1242 
1243 	if (num_async) {
1244 		/* Transmit if any encryptions have completed */
1245 		if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1246 			cancel_delayed_work(&ctx->tx_work.work);
1247 			tls_tx_records(sk, flags);
1248 		}
1249 	}
1250 sendpage_end:
1251 	ret = sk_stream_error(sk, flags, ret);
1252 	return copied > 0 ? copied : ret;
1253 }
1254 
1255 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1256 			   int offset, size_t size, int flags)
1257 {
1258 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1259 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1260 		      MSG_NO_SHARED_FRAGS))
1261 		return -EOPNOTSUPP;
1262 
1263 	return tls_sw_do_sendpage(sk, page, offset, size, flags);
1264 }
1265 
1266 int tls_sw_sendpage(struct sock *sk, struct page *page,
1267 		    int offset, size_t size, int flags)
1268 {
1269 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1270 	int ret;
1271 
1272 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1274 		return -EOPNOTSUPP;
1275 
1276 	mutex_lock(&tls_ctx->tx_lock);
1277 	lock_sock(sk);
1278 	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1279 	release_sock(sk);
1280 	mutex_unlock(&tls_ctx->tx_lock);
1281 	return ret;
1282 }
1283 
1284 static int
1285 tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
1286 		long timeo)
1287 {
1288 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1289 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1290 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1291 
1292 	while (!ctx->recv_pkt) {
1293 		if (!sk_psock_queue_empty(psock))
1294 			return 0;
1295 
1296 		if (sk->sk_err)
1297 			return sock_error(sk);
1298 
1299 		if (!skb_queue_empty(&sk->sk_receive_queue)) {
1300 			__strp_unpause(&ctx->strp);
1301 			if (ctx->recv_pkt)
1302 				break;
1303 		}
1304 
1305 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1306 			return 0;
1307 
1308 		if (sock_flag(sk, SOCK_DONE))
1309 			return 0;
1310 
1311 		if (nonblock || !timeo)
1312 			return -EAGAIN;
1313 
1314 		add_wait_queue(sk_sleep(sk), &wait);
1315 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1316 		sk_wait_event(sk, &timeo,
1317 			      ctx->recv_pkt || !sk_psock_queue_empty(psock),
1318 			      &wait);
1319 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1320 		remove_wait_queue(sk_sleep(sk), &wait);
1321 
1322 		/* Handle signals */
1323 		if (signal_pending(current))
1324 			return sock_intr_errno(timeo);
1325 	}
1326 
1327 	return 1;
1328 }
1329 
1330 static int tls_setup_from_iter(struct iov_iter *from,
1331 			       int length, int *pages_used,
1332 			       struct scatterlist *to,
1333 			       int to_max_pages)
1334 {
1335 	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1336 	struct page *pages[MAX_SKB_FRAGS];
1337 	unsigned int size = 0;
1338 	ssize_t copied, use;
1339 	size_t offset;
1340 
1341 	while (length > 0) {
1342 		i = 0;
1343 		maxpages = to_max_pages - num_elem;
1344 		if (maxpages == 0) {
1345 			rc = -EFAULT;
1346 			goto out;
1347 		}
1348 		copied = iov_iter_get_pages(from, pages,
1349 					    length,
1350 					    maxpages, &offset);
1351 		if (copied <= 0) {
1352 			rc = -EFAULT;
1353 			goto out;
1354 		}
1355 
1356 		iov_iter_advance(from, copied);
1357 
1358 		length -= copied;
1359 		size += copied;
1360 		while (copied) {
1361 			use = min_t(int, copied, PAGE_SIZE - offset);
1362 
1363 			sg_set_page(&to[num_elem],
1364 				    pages[i], use, offset);
1365 			sg_unmark_end(&to[num_elem]);
1366 			/* We do not uncharge memory from this API */
1367 
1368 			offset = 0;
1369 			copied -= use;
1370 
1371 			i++;
1372 			num_elem++;
1373 		}
1374 	}
1375 	/* Mark the end in the last sg entry if newly added */
1376 	if (num_elem > *pages_used)
1377 		sg_mark_end(&to[num_elem - 1]);
1378 out:
1379 	if (rc)
1380 		iov_iter_revert(from, size);
1381 	*pages_used = num_elem;
1382 
1383 	return rc;
1384 }
1385 
1386 static struct sk_buff *
1387 tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1388 		     unsigned int full_len)
1389 {
1390 	struct strp_msg *clr_rxm;
1391 	struct sk_buff *clr_skb;
1392 	int err;
1393 
1394 	clr_skb = alloc_skb_with_frags(0, full_len, TLS_PAGE_ORDER,
1395 				       &err, sk->sk_allocation);
1396 	if (!clr_skb)
1397 		return NULL;
1398 
1399 	skb_copy_header(clr_skb, skb);
1400 	clr_skb->len = full_len;
1401 	clr_skb->data_len = full_len;
1402 
1403 	clr_rxm = strp_msg(clr_skb);
1404 	clr_rxm->offset = 0;
1405 
1406 	return clr_skb;
1407 }
1408 
1409 /* Decrypt handlers
1410  *
1411  * tls_decrypt_sg() and tls_decrypt_device() are decrypt handlers.
1412  * They must transform the darg in/out argument are as follows:
1413  *       |          Input            |         Output
1414  * -------------------------------------------------------------------
1415  *    zc | Zero-copy decrypt allowed | Zero-copy performed
1416  * async | Async decrypt allowed     | Async crypto used / in progress
1417  *   skb |            *              | Output skb
1418  */
1419 
1420 /* This function decrypts the input skb into either out_iov or in out_sg
1421  * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1422  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1423  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1424  * NULL, then the decryption happens inside skb buffers itself, i.e.
1425  * zero-copy gets disabled and 'darg->zc' is updated.
1426  */
1427 static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1428 			  struct scatterlist *out_sg,
1429 			  struct tls_decrypt_arg *darg)
1430 {
1431 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1432 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1433 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1434 	int n_sgin, n_sgout, aead_size, err, pages = 0;
1435 	struct sk_buff *skb = tls_strp_msg(ctx);
1436 	const struct strp_msg *rxm = strp_msg(skb);
1437 	const struct tls_msg *tlm = tls_msg(skb);
1438 	struct aead_request *aead_req;
1439 	struct scatterlist *sgin = NULL;
1440 	struct scatterlist *sgout = NULL;
1441 	const int data_len = rxm->full_len - prot->overhead_size;
1442 	int tail_pages = !!prot->tail_size;
1443 	struct tls_decrypt_ctx *dctx;
1444 	struct sk_buff *clear_skb;
1445 	int iv_offset = 0;
1446 	u8 *mem;
1447 
1448 	n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1449 			 rxm->full_len - prot->prepend_size);
1450 	if (n_sgin < 1)
1451 		return n_sgin ?: -EBADMSG;
1452 
1453 	if (darg->zc && (out_iov || out_sg)) {
1454 		clear_skb = NULL;
1455 
1456 		if (out_iov)
1457 			n_sgout = 1 + tail_pages +
1458 				iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1459 		else
1460 			n_sgout = sg_nents(out_sg);
1461 	} else {
1462 		darg->zc = false;
1463 
1464 		clear_skb = tls_alloc_clrtxt_skb(sk, skb, rxm->full_len);
1465 		if (!clear_skb)
1466 			return -ENOMEM;
1467 
1468 		n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1469 	}
1470 
1471 	/* Increment to accommodate AAD */
1472 	n_sgin = n_sgin + 1;
1473 
1474 	/* Allocate a single block of memory which contains
1475 	 *   aead_req || tls_decrypt_ctx.
1476 	 * Both structs are variable length.
1477 	 */
1478 	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1479 	mem = kmalloc(aead_size + struct_size(dctx, sg, n_sgin + n_sgout),
1480 		      sk->sk_allocation);
1481 	if (!mem) {
1482 		err = -ENOMEM;
1483 		goto exit_free_skb;
1484 	}
1485 
1486 	/* Segment the allocated memory */
1487 	aead_req = (struct aead_request *)mem;
1488 	dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1489 	sgin = &dctx->sg[0];
1490 	sgout = &dctx->sg[n_sgin];
1491 
1492 	/* For CCM based ciphers, first byte of nonce+iv is a constant */
1493 	switch (prot->cipher_type) {
1494 	case TLS_CIPHER_AES_CCM_128:
1495 		dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1496 		iv_offset = 1;
1497 		break;
1498 	case TLS_CIPHER_SM4_CCM:
1499 		dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1500 		iv_offset = 1;
1501 		break;
1502 	}
1503 
1504 	/* Prepare IV */
1505 	if (prot->version == TLS_1_3_VERSION ||
1506 	    prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1507 		memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1508 		       prot->iv_size + prot->salt_size);
1509 	} else {
1510 		err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1511 				    &dctx->iv[iv_offset] + prot->salt_size,
1512 				    prot->iv_size);
1513 		if (err < 0)
1514 			goto exit_free;
1515 		memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1516 	}
1517 	tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
1518 
1519 	/* Prepare AAD */
1520 	tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
1521 		     prot->tail_size,
1522 		     tls_ctx->rx.rec_seq, tlm->control, prot);
1523 
1524 	/* Prepare sgin */
1525 	sg_init_table(sgin, n_sgin);
1526 	sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
1527 	err = skb_to_sgvec(skb, &sgin[1],
1528 			   rxm->offset + prot->prepend_size,
1529 			   rxm->full_len - prot->prepend_size);
1530 	if (err < 0)
1531 		goto exit_free;
1532 
1533 	if (clear_skb) {
1534 		sg_init_table(sgout, n_sgout);
1535 		sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1536 
1537 		err = skb_to_sgvec(clear_skb, &sgout[1], prot->prepend_size,
1538 				   data_len + prot->tail_size);
1539 		if (err < 0)
1540 			goto exit_free;
1541 	} else if (out_iov) {
1542 		sg_init_table(sgout, n_sgout);
1543 		sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1544 
1545 		err = tls_setup_from_iter(out_iov, data_len, &pages, &sgout[1],
1546 					  (n_sgout - 1 - tail_pages));
1547 		if (err < 0)
1548 			goto exit_free_pages;
1549 
1550 		if (prot->tail_size) {
1551 			sg_unmark_end(&sgout[pages]);
1552 			sg_set_buf(&sgout[pages + 1], &dctx->tail,
1553 				   prot->tail_size);
1554 			sg_mark_end(&sgout[pages + 1]);
1555 		}
1556 	} else if (out_sg) {
1557 		memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1558 	}
1559 
1560 	/* Prepare and submit AEAD request */
1561 	err = tls_do_decryption(sk, sgin, sgout, dctx->iv,
1562 				data_len + prot->tail_size, aead_req, darg);
1563 	if (err)
1564 		goto exit_free_pages;
1565 
1566 	darg->skb = clear_skb ?: tls_strp_msg(ctx);
1567 	clear_skb = NULL;
1568 
1569 	if (unlikely(darg->async)) {
1570 		err = tls_strp_msg_hold(sk, skb, &ctx->async_hold);
1571 		if (err)
1572 			__skb_queue_tail(&ctx->async_hold, darg->skb);
1573 		return err;
1574 	}
1575 
1576 	if (prot->tail_size)
1577 		darg->tail = dctx->tail;
1578 
1579 exit_free_pages:
1580 	/* Release the pages in case iov was mapped to pages */
1581 	for (; pages > 0; pages--)
1582 		put_page(sg_page(&sgout[pages]));
1583 exit_free:
1584 	kfree(mem);
1585 exit_free_skb:
1586 	consume_skb(clear_skb);
1587 	return err;
1588 }
1589 
1590 static int
1591 tls_decrypt_device(struct sock *sk, struct tls_context *tls_ctx,
1592 		   struct tls_decrypt_arg *darg)
1593 {
1594 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1595 	int err;
1596 
1597 	if (tls_ctx->rx_conf != TLS_HW)
1598 		return 0;
1599 
1600 	err = tls_device_decrypted(sk, tls_ctx);
1601 	if (err <= 0)
1602 		return err;
1603 
1604 	darg->zc = false;
1605 	darg->async = false;
1606 	darg->skb = tls_strp_msg(ctx);
1607 	ctx->recv_pkt = NULL;
1608 	return 1;
1609 }
1610 
1611 static int tls_rx_one_record(struct sock *sk, struct iov_iter *dest,
1612 			     struct tls_decrypt_arg *darg)
1613 {
1614 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1615 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1616 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1617 	struct strp_msg *rxm;
1618 	int pad, err;
1619 
1620 	err = tls_decrypt_device(sk, tls_ctx, darg);
1621 	if (err < 0)
1622 		return err;
1623 	if (err)
1624 		goto decrypt_done;
1625 
1626 	err = tls_decrypt_sg(sk, dest, NULL, darg);
1627 	if (err < 0) {
1628 		if (err == -EBADMSG)
1629 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1630 		return err;
1631 	}
1632 	if (darg->async)
1633 		goto decrypt_done;
1634 	/* If opportunistic TLS 1.3 ZC failed retry without ZC */
1635 	if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1636 		     darg->tail != TLS_RECORD_TYPE_DATA)) {
1637 		darg->zc = false;
1638 		if (!darg->tail)
1639 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1640 		TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1641 		return tls_rx_one_record(sk, dest, darg);
1642 	}
1643 
1644 decrypt_done:
1645 	if (darg->skb == ctx->recv_pkt)
1646 		ctx->recv_pkt = NULL;
1647 
1648 	pad = tls_padding_length(prot, darg->skb, darg);
1649 	if (pad < 0) {
1650 		consume_skb(darg->skb);
1651 		return pad;
1652 	}
1653 
1654 	rxm = strp_msg(darg->skb);
1655 	rxm->full_len -= pad;
1656 	rxm->offset += prot->prepend_size;
1657 	rxm->full_len -= prot->overhead_size;
1658 	tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1659 
1660 	return 0;
1661 }
1662 
1663 int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
1664 {
1665 	struct tls_decrypt_arg darg = { .zc = true, };
1666 
1667 	return tls_decrypt_sg(sk, NULL, sgout, &darg);
1668 }
1669 
1670 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1671 				   u8 *control)
1672 {
1673 	int err;
1674 
1675 	if (!*control) {
1676 		*control = tlm->control;
1677 		if (!*control)
1678 			return -EBADMSG;
1679 
1680 		err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1681 			       sizeof(*control), control);
1682 		if (*control != TLS_RECORD_TYPE_DATA) {
1683 			if (err || msg->msg_flags & MSG_CTRUNC)
1684 				return -EIO;
1685 		}
1686 	} else if (*control != tlm->control) {
1687 		return 0;
1688 	}
1689 
1690 	return 1;
1691 }
1692 
1693 static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1694 {
1695 	consume_skb(ctx->recv_pkt);
1696 	ctx->recv_pkt = NULL;
1697 	__strp_unpause(&ctx->strp);
1698 }
1699 
1700 /* This function traverses the rx_list in tls receive context to copies the
1701  * decrypted records into the buffer provided by caller zero copy is not
1702  * true. Further, the records are removed from the rx_list if it is not a peek
1703  * case and the record has been consumed completely.
1704  */
1705 static int process_rx_list(struct tls_sw_context_rx *ctx,
1706 			   struct msghdr *msg,
1707 			   u8 *control,
1708 			   size_t skip,
1709 			   size_t len,
1710 			   bool is_peek)
1711 {
1712 	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1713 	struct tls_msg *tlm;
1714 	ssize_t copied = 0;
1715 	int err;
1716 
1717 	while (skip && skb) {
1718 		struct strp_msg *rxm = strp_msg(skb);
1719 		tlm = tls_msg(skb);
1720 
1721 		err = tls_record_content_type(msg, tlm, control);
1722 		if (err <= 0)
1723 			goto out;
1724 
1725 		if (skip < rxm->full_len)
1726 			break;
1727 
1728 		skip = skip - rxm->full_len;
1729 		skb = skb_peek_next(skb, &ctx->rx_list);
1730 	}
1731 
1732 	while (len && skb) {
1733 		struct sk_buff *next_skb;
1734 		struct strp_msg *rxm = strp_msg(skb);
1735 		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1736 
1737 		tlm = tls_msg(skb);
1738 
1739 		err = tls_record_content_type(msg, tlm, control);
1740 		if (err <= 0)
1741 			goto out;
1742 
1743 		err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1744 					    msg, chunk);
1745 		if (err < 0)
1746 			goto out;
1747 
1748 		len = len - chunk;
1749 		copied = copied + chunk;
1750 
1751 		/* Consume the data from record if it is non-peek case*/
1752 		if (!is_peek) {
1753 			rxm->offset = rxm->offset + chunk;
1754 			rxm->full_len = rxm->full_len - chunk;
1755 
1756 			/* Return if there is unconsumed data in the record */
1757 			if (rxm->full_len - skip)
1758 				break;
1759 		}
1760 
1761 		/* The remaining skip-bytes must lie in 1st record in rx_list.
1762 		 * So from the 2nd record, 'skip' should be 0.
1763 		 */
1764 		skip = 0;
1765 
1766 		if (msg)
1767 			msg->msg_flags |= MSG_EOR;
1768 
1769 		next_skb = skb_peek_next(skb, &ctx->rx_list);
1770 
1771 		if (!is_peek) {
1772 			__skb_unlink(skb, &ctx->rx_list);
1773 			consume_skb(skb);
1774 		}
1775 
1776 		skb = next_skb;
1777 	}
1778 	err = 0;
1779 
1780 out:
1781 	return copied ? : err;
1782 }
1783 
1784 static void
1785 tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1786 		       size_t len_left, size_t decrypted, ssize_t done,
1787 		       size_t *flushed_at)
1788 {
1789 	size_t max_rec;
1790 
1791 	if (len_left <= decrypted)
1792 		return;
1793 
1794 	max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1795 	if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1796 		return;
1797 
1798 	*flushed_at = done;
1799 	sk_flush_backlog(sk);
1800 }
1801 
1802 static long tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
1803 			       bool nonblock)
1804 {
1805 	long timeo;
1806 
1807 	lock_sock(sk);
1808 
1809 	timeo = sock_rcvtimeo(sk, nonblock);
1810 
1811 	while (unlikely(ctx->reader_present)) {
1812 		DEFINE_WAIT_FUNC(wait, woken_wake_function);
1813 
1814 		ctx->reader_contended = 1;
1815 
1816 		add_wait_queue(&ctx->wq, &wait);
1817 		sk_wait_event(sk, &timeo,
1818 			      !READ_ONCE(ctx->reader_present), &wait);
1819 		remove_wait_queue(&ctx->wq, &wait);
1820 
1821 		if (!timeo)
1822 			return -EAGAIN;
1823 		if (signal_pending(current))
1824 			return sock_intr_errno(timeo);
1825 	}
1826 
1827 	WRITE_ONCE(ctx->reader_present, 1);
1828 
1829 	return timeo;
1830 }
1831 
1832 static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
1833 {
1834 	if (unlikely(ctx->reader_contended)) {
1835 		if (wq_has_sleeper(&ctx->wq))
1836 			wake_up(&ctx->wq);
1837 		else
1838 			ctx->reader_contended = 0;
1839 
1840 		WARN_ON_ONCE(!ctx->reader_present);
1841 	}
1842 
1843 	WRITE_ONCE(ctx->reader_present, 0);
1844 	release_sock(sk);
1845 }
1846 
1847 int tls_sw_recvmsg(struct sock *sk,
1848 		   struct msghdr *msg,
1849 		   size_t len,
1850 		   int flags,
1851 		   int *addr_len)
1852 {
1853 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1854 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1855 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1856 	ssize_t decrypted = 0, async_copy_bytes = 0;
1857 	struct sk_psock *psock;
1858 	unsigned char control = 0;
1859 	size_t flushed_at = 0;
1860 	struct strp_msg *rxm;
1861 	struct tls_msg *tlm;
1862 	struct sk_buff *skb;
1863 	ssize_t copied = 0;
1864 	bool async = false;
1865 	int target, err = 0;
1866 	long timeo;
1867 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1868 	bool is_peek = flags & MSG_PEEK;
1869 	bool bpf_strp_enabled;
1870 	bool zc_capable;
1871 
1872 	if (unlikely(flags & MSG_ERRQUEUE))
1873 		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1874 
1875 	psock = sk_psock_get(sk);
1876 	timeo = tls_rx_reader_lock(sk, ctx, flags & MSG_DONTWAIT);
1877 	if (timeo < 0)
1878 		return timeo;
1879 	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1880 
1881 	/* If crypto failed the connection is broken */
1882 	err = ctx->async_wait.err;
1883 	if (err)
1884 		goto end;
1885 
1886 	/* Process pending decrypted records. It must be non-zero-copy */
1887 	err = process_rx_list(ctx, msg, &control, 0, len, is_peek);
1888 	if (err < 0)
1889 		goto end;
1890 
1891 	copied = err;
1892 	if (len <= copied)
1893 		goto end;
1894 
1895 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1896 	len = len - copied;
1897 
1898 	zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
1899 		ctx->zc_capable;
1900 	decrypted = 0;
1901 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1902 		struct tls_decrypt_arg darg;
1903 		int to_decrypt, chunk;
1904 
1905 		err = tls_rx_rec_wait(sk, psock, flags & MSG_DONTWAIT, timeo);
1906 		if (err <= 0) {
1907 			if (psock) {
1908 				chunk = sk_msg_recvmsg(sk, psock, msg, len,
1909 						       flags);
1910 				if (chunk > 0) {
1911 					decrypted += chunk;
1912 					len -= chunk;
1913 					continue;
1914 				}
1915 			}
1916 			goto recv_end;
1917 		}
1918 
1919 		memset(&darg.inargs, 0, sizeof(darg.inargs));
1920 
1921 		rxm = strp_msg(ctx->recv_pkt);
1922 		tlm = tls_msg(ctx->recv_pkt);
1923 
1924 		to_decrypt = rxm->full_len - prot->overhead_size;
1925 
1926 		if (zc_capable && to_decrypt <= len &&
1927 		    tlm->control == TLS_RECORD_TYPE_DATA)
1928 			darg.zc = true;
1929 
1930 		/* Do not use async mode if record is non-data */
1931 		if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1932 			darg.async = ctx->async_capable;
1933 		else
1934 			darg.async = false;
1935 
1936 		err = tls_rx_one_record(sk, &msg->msg_iter, &darg);
1937 		if (err < 0) {
1938 			tls_err_abort(sk, -EBADMSG);
1939 			goto recv_end;
1940 		}
1941 
1942 		skb = darg.skb;
1943 		rxm = strp_msg(skb);
1944 		tlm = tls_msg(skb);
1945 
1946 		async |= darg.async;
1947 
1948 		/* If the type of records being processed is not known yet,
1949 		 * set it to record type just dequeued. If it is already known,
1950 		 * but does not match the record type just dequeued, go to end.
1951 		 * We always get record type here since for tls1.2, record type
1952 		 * is known just after record is dequeued from stream parser.
1953 		 * For tls1.3, we disable async.
1954 		 */
1955 		err = tls_record_content_type(msg, tlm, &control);
1956 		if (err <= 0) {
1957 			tls_rx_rec_done(ctx);
1958 put_on_rx_list_err:
1959 			__skb_queue_tail(&ctx->rx_list, skb);
1960 			goto recv_end;
1961 		}
1962 
1963 		/* periodically flush backlog, and feed strparser */
1964 		tls_read_flush_backlog(sk, prot, len, to_decrypt,
1965 				       decrypted + copied, &flushed_at);
1966 
1967 		/* TLS 1.3 may have updated the length by more than overhead */
1968 		chunk = rxm->full_len;
1969 		tls_rx_rec_done(ctx);
1970 
1971 		if (!darg.zc) {
1972 			bool partially_consumed = chunk > len;
1973 
1974 			if (async) {
1975 				/* TLS 1.2-only, to_decrypt must be text len */
1976 				chunk = min_t(int, to_decrypt, len);
1977 				async_copy_bytes += chunk;
1978 put_on_rx_list:
1979 				decrypted += chunk;
1980 				len -= chunk;
1981 				__skb_queue_tail(&ctx->rx_list, skb);
1982 				continue;
1983 			}
1984 
1985 			if (bpf_strp_enabled) {
1986 				err = sk_psock_tls_strp_read(psock, skb);
1987 				if (err != __SK_PASS) {
1988 					rxm->offset = rxm->offset + rxm->full_len;
1989 					rxm->full_len = 0;
1990 					if (err == __SK_DROP)
1991 						consume_skb(skb);
1992 					continue;
1993 				}
1994 			}
1995 
1996 			if (partially_consumed)
1997 				chunk = len;
1998 
1999 			err = skb_copy_datagram_msg(skb, rxm->offset,
2000 						    msg, chunk);
2001 			if (err < 0)
2002 				goto put_on_rx_list_err;
2003 
2004 			if (is_peek)
2005 				goto put_on_rx_list;
2006 
2007 			if (partially_consumed) {
2008 				rxm->offset += chunk;
2009 				rxm->full_len -= chunk;
2010 				goto put_on_rx_list;
2011 			}
2012 		}
2013 
2014 		decrypted += chunk;
2015 		len -= chunk;
2016 
2017 		consume_skb(skb);
2018 
2019 		/* Return full control message to userspace before trying
2020 		 * to parse another message type
2021 		 */
2022 		msg->msg_flags |= MSG_EOR;
2023 		if (control != TLS_RECORD_TYPE_DATA)
2024 			break;
2025 	}
2026 
2027 recv_end:
2028 	if (async) {
2029 		int ret, pending;
2030 
2031 		/* Wait for all previously submitted records to be decrypted */
2032 		spin_lock_bh(&ctx->decrypt_compl_lock);
2033 		reinit_completion(&ctx->async_wait.completion);
2034 		pending = atomic_read(&ctx->decrypt_pending);
2035 		spin_unlock_bh(&ctx->decrypt_compl_lock);
2036 		ret = 0;
2037 		if (pending)
2038 			ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2039 		__skb_queue_purge(&ctx->async_hold);
2040 
2041 		if (ret) {
2042 			if (err >= 0 || err == -EINPROGRESS)
2043 				err = ret;
2044 			decrypted = 0;
2045 			goto end;
2046 		}
2047 
2048 		/* Drain records from the rx_list & copy if required */
2049 		if (is_peek || is_kvec)
2050 			err = process_rx_list(ctx, msg, &control, copied,
2051 					      decrypted, is_peek);
2052 		else
2053 			err = process_rx_list(ctx, msg, &control, 0,
2054 					      async_copy_bytes, is_peek);
2055 		decrypted = max(err, 0);
2056 	}
2057 
2058 	copied += decrypted;
2059 
2060 end:
2061 	tls_rx_reader_unlock(sk, ctx);
2062 	if (psock)
2063 		sk_psock_put(sk, psock);
2064 	return copied ? : err;
2065 }
2066 
2067 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
2068 			   struct pipe_inode_info *pipe,
2069 			   size_t len, unsigned int flags)
2070 {
2071 	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2072 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2073 	struct strp_msg *rxm = NULL;
2074 	struct sock *sk = sock->sk;
2075 	struct tls_msg *tlm;
2076 	struct sk_buff *skb;
2077 	ssize_t copied = 0;
2078 	int err = 0;
2079 	long timeo;
2080 	int chunk;
2081 
2082 	timeo = tls_rx_reader_lock(sk, ctx, flags & SPLICE_F_NONBLOCK);
2083 	if (timeo < 0)
2084 		return timeo;
2085 
2086 	if (!skb_queue_empty(&ctx->rx_list)) {
2087 		skb = __skb_dequeue(&ctx->rx_list);
2088 	} else {
2089 		struct tls_decrypt_arg darg;
2090 
2091 		err = tls_rx_rec_wait(sk, NULL, flags & SPLICE_F_NONBLOCK,
2092 				      timeo);
2093 		if (err <= 0)
2094 			goto splice_read_end;
2095 
2096 		memset(&darg.inargs, 0, sizeof(darg.inargs));
2097 
2098 		err = tls_rx_one_record(sk, NULL, &darg);
2099 		if (err < 0) {
2100 			tls_err_abort(sk, -EBADMSG);
2101 			goto splice_read_end;
2102 		}
2103 
2104 		tls_rx_rec_done(ctx);
2105 		skb = darg.skb;
2106 	}
2107 
2108 	rxm = strp_msg(skb);
2109 	tlm = tls_msg(skb);
2110 
2111 	/* splice does not support reading control messages */
2112 	if (tlm->control != TLS_RECORD_TYPE_DATA) {
2113 		err = -EINVAL;
2114 		goto splice_requeue;
2115 	}
2116 
2117 	chunk = min_t(unsigned int, rxm->full_len, len);
2118 	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2119 	if (copied < 0)
2120 		goto splice_requeue;
2121 
2122 	if (chunk < rxm->full_len) {
2123 		rxm->offset += len;
2124 		rxm->full_len -= len;
2125 		goto splice_requeue;
2126 	}
2127 
2128 	consume_skb(skb);
2129 
2130 splice_read_end:
2131 	tls_rx_reader_unlock(sk, ctx);
2132 	return copied ? : err;
2133 
2134 splice_requeue:
2135 	__skb_queue_head(&ctx->rx_list, skb);
2136 	goto splice_read_end;
2137 }
2138 
2139 bool tls_sw_sock_is_readable(struct sock *sk)
2140 {
2141 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2142 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2143 	bool ingress_empty = true;
2144 	struct sk_psock *psock;
2145 
2146 	rcu_read_lock();
2147 	psock = sk_psock(sk);
2148 	if (psock)
2149 		ingress_empty = list_empty(&psock->ingress_msg);
2150 	rcu_read_unlock();
2151 
2152 	return !ingress_empty || ctx->recv_pkt ||
2153 		!skb_queue_empty(&ctx->rx_list);
2154 }
2155 
2156 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2157 {
2158 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2159 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2160 	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2161 	struct strp_msg *rxm = strp_msg(skb);
2162 	struct tls_msg *tlm = tls_msg(skb);
2163 	size_t cipher_overhead;
2164 	size_t data_len = 0;
2165 	int ret;
2166 
2167 	/* Verify that we have a full TLS header, or wait for more data */
2168 	if (rxm->offset + prot->prepend_size > skb->len)
2169 		return 0;
2170 
2171 	/* Sanity-check size of on-stack buffer. */
2172 	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2173 		ret = -EINVAL;
2174 		goto read_failure;
2175 	}
2176 
2177 	/* Linearize header to local buffer */
2178 	ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2179 	if (ret < 0)
2180 		goto read_failure;
2181 
2182 	tlm->control = header[0];
2183 
2184 	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2185 
2186 	cipher_overhead = prot->tag_size;
2187 	if (prot->version != TLS_1_3_VERSION &&
2188 	    prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2189 		cipher_overhead += prot->iv_size;
2190 
2191 	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2192 	    prot->tail_size) {
2193 		ret = -EMSGSIZE;
2194 		goto read_failure;
2195 	}
2196 	if (data_len < cipher_overhead) {
2197 		ret = -EBADMSG;
2198 		goto read_failure;
2199 	}
2200 
2201 	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2202 	if (header[1] != TLS_1_2_VERSION_MINOR ||
2203 	    header[2] != TLS_1_2_VERSION_MAJOR) {
2204 		ret = -EINVAL;
2205 		goto read_failure;
2206 	}
2207 
2208 	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2209 				     TCP_SKB_CB(skb)->seq + rxm->offset);
2210 	return data_len + TLS_HEADER_SIZE;
2211 
2212 read_failure:
2213 	tls_err_abort(strp->sk, ret);
2214 
2215 	return ret;
2216 }
2217 
2218 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2219 {
2220 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2221 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2222 
2223 	ctx->recv_pkt = skb;
2224 	strp_pause(strp);
2225 
2226 	ctx->saved_data_ready(strp->sk);
2227 }
2228 
2229 static void tls_data_ready(struct sock *sk)
2230 {
2231 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2232 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2233 	struct sk_psock *psock;
2234 
2235 	strp_data_ready(&ctx->strp);
2236 
2237 	psock = sk_psock_get(sk);
2238 	if (psock) {
2239 		if (!list_empty(&psock->ingress_msg))
2240 			ctx->saved_data_ready(sk);
2241 		sk_psock_put(sk, psock);
2242 	}
2243 }
2244 
2245 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2246 {
2247 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2248 
2249 	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2250 	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2251 	cancel_delayed_work_sync(&ctx->tx_work.work);
2252 }
2253 
2254 void tls_sw_release_resources_tx(struct sock *sk)
2255 {
2256 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2257 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2258 	struct tls_rec *rec, *tmp;
2259 	int pending;
2260 
2261 	/* Wait for any pending async encryptions to complete */
2262 	spin_lock_bh(&ctx->encrypt_compl_lock);
2263 	ctx->async_notify = true;
2264 	pending = atomic_read(&ctx->encrypt_pending);
2265 	spin_unlock_bh(&ctx->encrypt_compl_lock);
2266 
2267 	if (pending)
2268 		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2269 
2270 	tls_tx_records(sk, -1);
2271 
2272 	/* Free up un-sent records in tx_list. First, free
2273 	 * the partially sent record if any at head of tx_list.
2274 	 */
2275 	if (tls_ctx->partially_sent_record) {
2276 		tls_free_partial_record(sk, tls_ctx);
2277 		rec = list_first_entry(&ctx->tx_list,
2278 				       struct tls_rec, list);
2279 		list_del(&rec->list);
2280 		sk_msg_free(sk, &rec->msg_plaintext);
2281 		kfree(rec);
2282 	}
2283 
2284 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2285 		list_del(&rec->list);
2286 		sk_msg_free(sk, &rec->msg_encrypted);
2287 		sk_msg_free(sk, &rec->msg_plaintext);
2288 		kfree(rec);
2289 	}
2290 
2291 	crypto_free_aead(ctx->aead_send);
2292 	tls_free_open_rec(sk);
2293 }
2294 
2295 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2296 {
2297 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2298 
2299 	kfree(ctx);
2300 }
2301 
2302 void tls_sw_release_resources_rx(struct sock *sk)
2303 {
2304 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2305 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2306 
2307 	kfree(tls_ctx->rx.rec_seq);
2308 	kfree(tls_ctx->rx.iv);
2309 
2310 	if (ctx->aead_recv) {
2311 		kfree_skb(ctx->recv_pkt);
2312 		ctx->recv_pkt = NULL;
2313 		__skb_queue_purge(&ctx->rx_list);
2314 		crypto_free_aead(ctx->aead_recv);
2315 		strp_stop(&ctx->strp);
2316 		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2317 		 * we still want to strp_stop(), but sk->sk_data_ready was
2318 		 * never swapped.
2319 		 */
2320 		if (ctx->saved_data_ready) {
2321 			write_lock_bh(&sk->sk_callback_lock);
2322 			sk->sk_data_ready = ctx->saved_data_ready;
2323 			write_unlock_bh(&sk->sk_callback_lock);
2324 		}
2325 	}
2326 }
2327 
2328 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2329 {
2330 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2331 
2332 	strp_done(&ctx->strp);
2333 }
2334 
2335 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2336 {
2337 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2338 
2339 	kfree(ctx);
2340 }
2341 
2342 void tls_sw_free_resources_rx(struct sock *sk)
2343 {
2344 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2345 
2346 	tls_sw_release_resources_rx(sk);
2347 	tls_sw_free_ctx_rx(tls_ctx);
2348 }
2349 
2350 /* The work handler to transmitt the encrypted records in tx_list */
2351 static void tx_work_handler(struct work_struct *work)
2352 {
2353 	struct delayed_work *delayed_work = to_delayed_work(work);
2354 	struct tx_work *tx_work = container_of(delayed_work,
2355 					       struct tx_work, work);
2356 	struct sock *sk = tx_work->sk;
2357 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2358 	struct tls_sw_context_tx *ctx;
2359 
2360 	if (unlikely(!tls_ctx))
2361 		return;
2362 
2363 	ctx = tls_sw_ctx_tx(tls_ctx);
2364 	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2365 		return;
2366 
2367 	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2368 		return;
2369 	mutex_lock(&tls_ctx->tx_lock);
2370 	lock_sock(sk);
2371 	tls_tx_records(sk, -1);
2372 	release_sock(sk);
2373 	mutex_unlock(&tls_ctx->tx_lock);
2374 }
2375 
2376 static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2377 {
2378 	struct tls_rec *rec;
2379 
2380 	rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
2381 	if (!rec)
2382 		return false;
2383 
2384 	return READ_ONCE(rec->tx_ready);
2385 }
2386 
2387 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2388 {
2389 	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2390 
2391 	/* Schedule the transmission if tx list is ready */
2392 	if (tls_is_tx_ready(tx_ctx) &&
2393 	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2394 		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2395 }
2396 
2397 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2398 {
2399 	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2400 
2401 	write_lock_bh(&sk->sk_callback_lock);
2402 	rx_ctx->saved_data_ready = sk->sk_data_ready;
2403 	sk->sk_data_ready = tls_data_ready;
2404 	write_unlock_bh(&sk->sk_callback_lock);
2405 
2406 	strp_check_rcv(&rx_ctx->strp);
2407 }
2408 
2409 void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2410 {
2411 	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2412 
2413 	rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2414 		tls_ctx->prot_info.version != TLS_1_3_VERSION;
2415 }
2416 
2417 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2418 {
2419 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2420 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2421 	struct tls_crypto_info *crypto_info;
2422 	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2423 	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2424 	struct cipher_context *cctx;
2425 	struct crypto_aead **aead;
2426 	struct strp_callbacks cb;
2427 	u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2428 	struct crypto_tfm *tfm;
2429 	char *iv, *rec_seq, *key, *salt, *cipher_name;
2430 	size_t keysize;
2431 	int rc = 0;
2432 
2433 	if (!ctx) {
2434 		rc = -EINVAL;
2435 		goto out;
2436 	}
2437 
2438 	if (tx) {
2439 		if (!ctx->priv_ctx_tx) {
2440 			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2441 			if (!sw_ctx_tx) {
2442 				rc = -ENOMEM;
2443 				goto out;
2444 			}
2445 			ctx->priv_ctx_tx = sw_ctx_tx;
2446 		} else {
2447 			sw_ctx_tx =
2448 				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2449 		}
2450 	} else {
2451 		if (!ctx->priv_ctx_rx) {
2452 			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2453 			if (!sw_ctx_rx) {
2454 				rc = -ENOMEM;
2455 				goto out;
2456 			}
2457 			ctx->priv_ctx_rx = sw_ctx_rx;
2458 		} else {
2459 			sw_ctx_rx =
2460 				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2461 		}
2462 	}
2463 
2464 	if (tx) {
2465 		crypto_init_wait(&sw_ctx_tx->async_wait);
2466 		spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2467 		crypto_info = &ctx->crypto_send.info;
2468 		cctx = &ctx->tx;
2469 		aead = &sw_ctx_tx->aead_send;
2470 		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2471 		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2472 		sw_ctx_tx->tx_work.sk = sk;
2473 	} else {
2474 		crypto_init_wait(&sw_ctx_rx->async_wait);
2475 		spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2476 		init_waitqueue_head(&sw_ctx_rx->wq);
2477 		crypto_info = &ctx->crypto_recv.info;
2478 		cctx = &ctx->rx;
2479 		skb_queue_head_init(&sw_ctx_rx->rx_list);
2480 		skb_queue_head_init(&sw_ctx_rx->async_hold);
2481 		aead = &sw_ctx_rx->aead_recv;
2482 	}
2483 
2484 	switch (crypto_info->cipher_type) {
2485 	case TLS_CIPHER_AES_GCM_128: {
2486 		struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2487 
2488 		gcm_128_info = (void *)crypto_info;
2489 		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2490 		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2491 		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2492 		iv = gcm_128_info->iv;
2493 		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2494 		rec_seq = gcm_128_info->rec_seq;
2495 		keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2496 		key = gcm_128_info->key;
2497 		salt = gcm_128_info->salt;
2498 		salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2499 		cipher_name = "gcm(aes)";
2500 		break;
2501 	}
2502 	case TLS_CIPHER_AES_GCM_256: {
2503 		struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2504 
2505 		gcm_256_info = (void *)crypto_info;
2506 		nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2507 		tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2508 		iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2509 		iv = gcm_256_info->iv;
2510 		rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2511 		rec_seq = gcm_256_info->rec_seq;
2512 		keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2513 		key = gcm_256_info->key;
2514 		salt = gcm_256_info->salt;
2515 		salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2516 		cipher_name = "gcm(aes)";
2517 		break;
2518 	}
2519 	case TLS_CIPHER_AES_CCM_128: {
2520 		struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2521 
2522 		ccm_128_info = (void *)crypto_info;
2523 		nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2524 		tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2525 		iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2526 		iv = ccm_128_info->iv;
2527 		rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2528 		rec_seq = ccm_128_info->rec_seq;
2529 		keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2530 		key = ccm_128_info->key;
2531 		salt = ccm_128_info->salt;
2532 		salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2533 		cipher_name = "ccm(aes)";
2534 		break;
2535 	}
2536 	case TLS_CIPHER_CHACHA20_POLY1305: {
2537 		struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2538 
2539 		chacha20_poly1305_info = (void *)crypto_info;
2540 		nonce_size = 0;
2541 		tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2542 		iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2543 		iv = chacha20_poly1305_info->iv;
2544 		rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2545 		rec_seq = chacha20_poly1305_info->rec_seq;
2546 		keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2547 		key = chacha20_poly1305_info->key;
2548 		salt = chacha20_poly1305_info->salt;
2549 		salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2550 		cipher_name = "rfc7539(chacha20,poly1305)";
2551 		break;
2552 	}
2553 	case TLS_CIPHER_SM4_GCM: {
2554 		struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
2555 
2556 		sm4_gcm_info = (void *)crypto_info;
2557 		nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2558 		tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
2559 		iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2560 		iv = sm4_gcm_info->iv;
2561 		rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
2562 		rec_seq = sm4_gcm_info->rec_seq;
2563 		keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
2564 		key = sm4_gcm_info->key;
2565 		salt = sm4_gcm_info->salt;
2566 		salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
2567 		cipher_name = "gcm(sm4)";
2568 		break;
2569 	}
2570 	case TLS_CIPHER_SM4_CCM: {
2571 		struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
2572 
2573 		sm4_ccm_info = (void *)crypto_info;
2574 		nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2575 		tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
2576 		iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2577 		iv = sm4_ccm_info->iv;
2578 		rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
2579 		rec_seq = sm4_ccm_info->rec_seq;
2580 		keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
2581 		key = sm4_ccm_info->key;
2582 		salt = sm4_ccm_info->salt;
2583 		salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
2584 		cipher_name = "ccm(sm4)";
2585 		break;
2586 	}
2587 	default:
2588 		rc = -EINVAL;
2589 		goto free_priv;
2590 	}
2591 
2592 	if (crypto_info->version == TLS_1_3_VERSION) {
2593 		nonce_size = 0;
2594 		prot->aad_size = TLS_HEADER_SIZE;
2595 		prot->tail_size = 1;
2596 	} else {
2597 		prot->aad_size = TLS_AAD_SPACE_SIZE;
2598 		prot->tail_size = 0;
2599 	}
2600 
2601 	/* Sanity-check the sizes for stack allocations. */
2602 	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2603 	    rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE ||
2604 	    prot->aad_size > TLS_MAX_AAD_SIZE) {
2605 		rc = -EINVAL;
2606 		goto free_priv;
2607 	}
2608 
2609 	prot->version = crypto_info->version;
2610 	prot->cipher_type = crypto_info->cipher_type;
2611 	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2612 	prot->tag_size = tag_size;
2613 	prot->overhead_size = prot->prepend_size +
2614 			      prot->tag_size + prot->tail_size;
2615 	prot->iv_size = iv_size;
2616 	prot->salt_size = salt_size;
2617 	cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2618 	if (!cctx->iv) {
2619 		rc = -ENOMEM;
2620 		goto free_priv;
2621 	}
2622 	/* Note: 128 & 256 bit salt are the same size */
2623 	prot->rec_seq_size = rec_seq_size;
2624 	memcpy(cctx->iv, salt, salt_size);
2625 	memcpy(cctx->iv + salt_size, iv, iv_size);
2626 	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2627 	if (!cctx->rec_seq) {
2628 		rc = -ENOMEM;
2629 		goto free_iv;
2630 	}
2631 
2632 	if (!*aead) {
2633 		*aead = crypto_alloc_aead(cipher_name, 0, 0);
2634 		if (IS_ERR(*aead)) {
2635 			rc = PTR_ERR(*aead);
2636 			*aead = NULL;
2637 			goto free_rec_seq;
2638 		}
2639 	}
2640 
2641 	ctx->push_pending_record = tls_sw_push_pending_record;
2642 
2643 	rc = crypto_aead_setkey(*aead, key, keysize);
2644 
2645 	if (rc)
2646 		goto free_aead;
2647 
2648 	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2649 	if (rc)
2650 		goto free_aead;
2651 
2652 	if (sw_ctx_rx) {
2653 		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2654 
2655 		tls_update_rx_zc_capable(ctx);
2656 		sw_ctx_rx->async_capable =
2657 			crypto_info->version != TLS_1_3_VERSION &&
2658 			!!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
2659 
2660 		/* Set up strparser */
2661 		memset(&cb, 0, sizeof(cb));
2662 		cb.rcv_msg = tls_queue;
2663 		cb.parse_msg = tls_read_size;
2664 
2665 		strp_init(&sw_ctx_rx->strp, sk, &cb);
2666 	}
2667 
2668 	goto out;
2669 
2670 free_aead:
2671 	crypto_free_aead(*aead);
2672 	*aead = NULL;
2673 free_rec_seq:
2674 	kfree(cctx->rec_seq);
2675 	cctx->rec_seq = NULL;
2676 free_iv:
2677 	kfree(cctx->iv);
2678 	cctx->iv = NULL;
2679 free_priv:
2680 	if (tx) {
2681 		kfree(ctx->priv_ctx_tx);
2682 		ctx->priv_ctx_tx = NULL;
2683 	} else {
2684 		kfree(ctx->priv_ctx_rx);
2685 		ctx->priv_ctx_rx = NULL;
2686 	}
2687 out:
2688 	return rc;
2689 }
2690