• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 struct tls_decrypt_arg {
48 	bool zc;
49 	bool async;
50 };
51 
tls_err_abort(struct sock * sk,int err)52 noinline void tls_err_abort(struct sock *sk, int err)
53 {
54 	WARN_ON_ONCE(err >= 0);
55 	/* sk->sk_err should contain a positive error code. */
56 	sk->sk_err = -err;
57 	sk_error_report(sk);
58 }
59 
__skb_nsg(struct sk_buff * skb,int offset,int len,unsigned int recursion_level)60 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
61                      unsigned int recursion_level)
62 {
63         int start = skb_headlen(skb);
64         int i, chunk = start - offset;
65         struct sk_buff *frag_iter;
66         int elt = 0;
67 
68         if (unlikely(recursion_level >= 24))
69                 return -EMSGSIZE;
70 
71         if (chunk > 0) {
72                 if (chunk > len)
73                         chunk = len;
74                 elt++;
75                 len -= chunk;
76                 if (len == 0)
77                         return elt;
78                 offset += chunk;
79         }
80 
81         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
82                 int end;
83 
84                 WARN_ON(start > offset + len);
85 
86                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
87                 chunk = end - offset;
88                 if (chunk > 0) {
89                         if (chunk > len)
90                                 chunk = len;
91                         elt++;
92                         len -= chunk;
93                         if (len == 0)
94                                 return elt;
95                         offset += chunk;
96                 }
97                 start = end;
98         }
99 
100         if (unlikely(skb_has_frag_list(skb))) {
101                 skb_walk_frags(skb, frag_iter) {
102                         int end, ret;
103 
104                         WARN_ON(start > offset + len);
105 
106                         end = start + frag_iter->len;
107                         chunk = end - offset;
108                         if (chunk > 0) {
109                                 if (chunk > len)
110                                         chunk = len;
111                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
112                                                 recursion_level + 1);
113                                 if (unlikely(ret < 0))
114                                         return ret;
115                                 elt += ret;
116                                 len -= chunk;
117                                 if (len == 0)
118                                         return elt;
119                                 offset += chunk;
120                         }
121                         start = end;
122                 }
123         }
124         BUG_ON(len);
125         return elt;
126 }
127 
128 /* Return the number of scatterlist elements required to completely map the
129  * skb, or -EMSGSIZE if the recursion depth is exceeded.
130  */
skb_nsg(struct sk_buff * skb,int offset,int len)131 static int skb_nsg(struct sk_buff *skb, int offset, int len)
132 {
133         return __skb_nsg(skb, offset, len, 0);
134 }
135 
padding_length(struct tls_prot_info * prot,struct sk_buff * skb)136 static int padding_length(struct tls_prot_info *prot, struct sk_buff *skb)
137 {
138 	struct strp_msg *rxm = strp_msg(skb);
139 	struct tls_msg *tlm = tls_msg(skb);
140 	int sub = 0;
141 
142 	/* Determine zero-padding length */
143 	if (prot->version == TLS_1_3_VERSION) {
144 		char content_type = 0;
145 		int err;
146 		int back = 17;
147 
148 		while (content_type == 0) {
149 			if (back > rxm->full_len - prot->prepend_size)
150 				return -EBADMSG;
151 			err = skb_copy_bits(skb,
152 					    rxm->offset + rxm->full_len - back,
153 					    &content_type, 1);
154 			if (err)
155 				return err;
156 			if (content_type)
157 				break;
158 			sub++;
159 			back++;
160 		}
161 		tlm->control = content_type;
162 	}
163 	return sub;
164 }
165 
tls_decrypt_done(struct crypto_async_request * req,int err)166 static void tls_decrypt_done(struct crypto_async_request *req, int err)
167 {
168 	struct aead_request *aead_req = (struct aead_request *)req;
169 	struct scatterlist *sgout = aead_req->dst;
170 	struct scatterlist *sgin = aead_req->src;
171 	struct tls_sw_context_rx *ctx;
172 	struct tls_context *tls_ctx;
173 	struct tls_prot_info *prot;
174 	struct scatterlist *sg;
175 	struct sk_buff *skb;
176 	unsigned int pages;
177 	int pending;
178 
179 	skb = (struct sk_buff *)req->data;
180 	tls_ctx = tls_get_ctx(skb->sk);
181 	ctx = tls_sw_ctx_rx(tls_ctx);
182 	prot = &tls_ctx->prot_info;
183 
184 	/* Propagate if there was an err */
185 	if (err) {
186 		if (err == -EBADMSG)
187 			TLS_INC_STATS(sock_net(skb->sk),
188 				      LINUX_MIB_TLSDECRYPTERROR);
189 		ctx->async_wait.err = err;
190 		tls_err_abort(skb->sk, err);
191 	} else {
192 		struct strp_msg *rxm = strp_msg(skb);
193 		int pad;
194 
195 		pad = padding_length(prot, skb);
196 		if (pad < 0) {
197 			ctx->async_wait.err = pad;
198 			tls_err_abort(skb->sk, pad);
199 		} else {
200 			rxm->full_len -= pad;
201 			rxm->offset += prot->prepend_size;
202 			rxm->full_len -= prot->overhead_size;
203 		}
204 	}
205 
206 	/* After using skb->sk to propagate sk through crypto async callback
207 	 * we need to NULL it again.
208 	 */
209 	skb->sk = NULL;
210 
211 
212 	/* Free the destination pages if skb was not decrypted inplace */
213 	if (sgout != sgin) {
214 		/* Skip the first S/G entry as it points to AAD */
215 		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
216 			if (!sg)
217 				break;
218 			put_page(sg_page(sg));
219 		}
220 	}
221 
222 	kfree(aead_req);
223 
224 	spin_lock_bh(&ctx->decrypt_compl_lock);
225 	pending = atomic_dec_return(&ctx->decrypt_pending);
226 
227 	if (!pending && ctx->async_notify)
228 		complete(&ctx->async_wait.completion);
229 	spin_unlock_bh(&ctx->decrypt_compl_lock);
230 }
231 
tls_do_decryption(struct sock * sk,struct sk_buff * skb,struct scatterlist * sgin,struct scatterlist * sgout,char * iv_recv,size_t data_len,struct aead_request * aead_req,struct tls_decrypt_arg * darg)232 static int tls_do_decryption(struct sock *sk,
233 			     struct sk_buff *skb,
234 			     struct scatterlist *sgin,
235 			     struct scatterlist *sgout,
236 			     char *iv_recv,
237 			     size_t data_len,
238 			     struct aead_request *aead_req,
239 			     struct tls_decrypt_arg *darg)
240 {
241 	struct tls_context *tls_ctx = tls_get_ctx(sk);
242 	struct tls_prot_info *prot = &tls_ctx->prot_info;
243 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
244 	int ret;
245 
246 	aead_request_set_tfm(aead_req, ctx->aead_recv);
247 	aead_request_set_ad(aead_req, prot->aad_size);
248 	aead_request_set_crypt(aead_req, sgin, sgout,
249 			       data_len + prot->tag_size,
250 			       (u8 *)iv_recv);
251 
252 	if (darg->async) {
253 		/* Using skb->sk to push sk through to crypto async callback
254 		 * handler. This allows propagating errors up to the socket
255 		 * if needed. It _must_ be cleared in the async handler
256 		 * before consume_skb is called. We _know_ skb->sk is NULL
257 		 * because it is a clone from strparser.
258 		 */
259 		skb->sk = sk;
260 		aead_request_set_callback(aead_req,
261 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
262 					  tls_decrypt_done, skb);
263 		atomic_inc(&ctx->decrypt_pending);
264 	} else {
265 		aead_request_set_callback(aead_req,
266 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
267 					  crypto_req_done, &ctx->async_wait);
268 	}
269 
270 	ret = crypto_aead_decrypt(aead_req);
271 	if (ret == -EINPROGRESS) {
272 		if (darg->async)
273 			return 0;
274 
275 		ret = crypto_wait_req(ret, &ctx->async_wait);
276 	} else if (darg->async) {
277 		atomic_dec(&ctx->decrypt_pending);
278 	}
279 	darg->async = false;
280 
281 	return ret;
282 }
283 
tls_trim_both_msgs(struct sock * sk,int target_size)284 static void tls_trim_both_msgs(struct sock *sk, int target_size)
285 {
286 	struct tls_context *tls_ctx = tls_get_ctx(sk);
287 	struct tls_prot_info *prot = &tls_ctx->prot_info;
288 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
289 	struct tls_rec *rec = ctx->open_rec;
290 
291 	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
292 	if (target_size > 0)
293 		target_size += prot->overhead_size;
294 	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
295 }
296 
tls_alloc_encrypted_msg(struct sock * sk,int len)297 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
298 {
299 	struct tls_context *tls_ctx = tls_get_ctx(sk);
300 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
301 	struct tls_rec *rec = ctx->open_rec;
302 	struct sk_msg *msg_en = &rec->msg_encrypted;
303 
304 	return sk_msg_alloc(sk, msg_en, len, 0);
305 }
306 
tls_clone_plaintext_msg(struct sock * sk,int required)307 static int tls_clone_plaintext_msg(struct sock *sk, int required)
308 {
309 	struct tls_context *tls_ctx = tls_get_ctx(sk);
310 	struct tls_prot_info *prot = &tls_ctx->prot_info;
311 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
312 	struct tls_rec *rec = ctx->open_rec;
313 	struct sk_msg *msg_pl = &rec->msg_plaintext;
314 	struct sk_msg *msg_en = &rec->msg_encrypted;
315 	int skip, len;
316 
317 	/* We add page references worth len bytes from encrypted sg
318 	 * at the end of plaintext sg. It is guaranteed that msg_en
319 	 * has enough required room (ensured by caller).
320 	 */
321 	len = required - msg_pl->sg.size;
322 
323 	/* Skip initial bytes in msg_en's data to be able to use
324 	 * same offset of both plain and encrypted data.
325 	 */
326 	skip = prot->prepend_size + msg_pl->sg.size;
327 
328 	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
329 }
330 
tls_get_rec(struct sock * sk)331 static struct tls_rec *tls_get_rec(struct sock *sk)
332 {
333 	struct tls_context *tls_ctx = tls_get_ctx(sk);
334 	struct tls_prot_info *prot = &tls_ctx->prot_info;
335 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
336 	struct sk_msg *msg_pl, *msg_en;
337 	struct tls_rec *rec;
338 	int mem_size;
339 
340 	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
341 
342 	rec = kzalloc(mem_size, sk->sk_allocation);
343 	if (!rec)
344 		return NULL;
345 
346 	msg_pl = &rec->msg_plaintext;
347 	msg_en = &rec->msg_encrypted;
348 
349 	sk_msg_init(msg_pl);
350 	sk_msg_init(msg_en);
351 
352 	sg_init_table(rec->sg_aead_in, 2);
353 	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
354 	sg_unmark_end(&rec->sg_aead_in[1]);
355 
356 	sg_init_table(rec->sg_aead_out, 2);
357 	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
358 	sg_unmark_end(&rec->sg_aead_out[1]);
359 
360 	return rec;
361 }
362 
tls_free_rec(struct sock * sk,struct tls_rec * rec)363 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
364 {
365 	sk_msg_free(sk, &rec->msg_encrypted);
366 	sk_msg_free(sk, &rec->msg_plaintext);
367 	kfree(rec);
368 }
369 
tls_free_open_rec(struct sock * sk)370 static void tls_free_open_rec(struct sock *sk)
371 {
372 	struct tls_context *tls_ctx = tls_get_ctx(sk);
373 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
374 	struct tls_rec *rec = ctx->open_rec;
375 
376 	if (rec) {
377 		tls_free_rec(sk, rec);
378 		ctx->open_rec = NULL;
379 	}
380 }
381 
tls_tx_records(struct sock * sk,int flags)382 int tls_tx_records(struct sock *sk, int flags)
383 {
384 	struct tls_context *tls_ctx = tls_get_ctx(sk);
385 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
386 	struct tls_rec *rec, *tmp;
387 	struct sk_msg *msg_en;
388 	int tx_flags, rc = 0;
389 
390 	if (tls_is_partially_sent_record(tls_ctx)) {
391 		rec = list_first_entry(&ctx->tx_list,
392 				       struct tls_rec, list);
393 
394 		if (flags == -1)
395 			tx_flags = rec->tx_flags;
396 		else
397 			tx_flags = flags;
398 
399 		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
400 		if (rc)
401 			goto tx_err;
402 
403 		/* Full record has been transmitted.
404 		 * Remove the head of tx_list
405 		 */
406 		list_del(&rec->list);
407 		sk_msg_free(sk, &rec->msg_plaintext);
408 		kfree(rec);
409 	}
410 
411 	/* Tx all ready records */
412 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
413 		if (READ_ONCE(rec->tx_ready)) {
414 			if (flags == -1)
415 				tx_flags = rec->tx_flags;
416 			else
417 				tx_flags = flags;
418 
419 			msg_en = &rec->msg_encrypted;
420 			rc = tls_push_sg(sk, tls_ctx,
421 					 &msg_en->sg.data[msg_en->sg.curr],
422 					 0, tx_flags);
423 			if (rc)
424 				goto tx_err;
425 
426 			list_del(&rec->list);
427 			sk_msg_free(sk, &rec->msg_plaintext);
428 			kfree(rec);
429 		} else {
430 			break;
431 		}
432 	}
433 
434 tx_err:
435 	if (rc < 0 && rc != -EAGAIN)
436 		tls_err_abort(sk, -EBADMSG);
437 
438 	return rc;
439 }
440 
tls_encrypt_done(struct crypto_async_request * req,int err)441 static void tls_encrypt_done(struct crypto_async_request *req, int err)
442 {
443 	struct aead_request *aead_req = (struct aead_request *)req;
444 	struct sock *sk = req->data;
445 	struct tls_context *tls_ctx = tls_get_ctx(sk);
446 	struct tls_prot_info *prot = &tls_ctx->prot_info;
447 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
448 	struct scatterlist *sge;
449 	struct sk_msg *msg_en;
450 	struct tls_rec *rec;
451 	bool ready = false;
452 	int pending;
453 
454 	rec = container_of(aead_req, struct tls_rec, aead_req);
455 	msg_en = &rec->msg_encrypted;
456 
457 	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
458 	sge->offset -= prot->prepend_size;
459 	sge->length += prot->prepend_size;
460 
461 	/* Check if error is previously set on socket */
462 	if (err || sk->sk_err) {
463 		rec = NULL;
464 
465 		/* If err is already set on socket, return the same code */
466 		if (sk->sk_err) {
467 			ctx->async_wait.err = -sk->sk_err;
468 		} else {
469 			ctx->async_wait.err = err;
470 			tls_err_abort(sk, err);
471 		}
472 	}
473 
474 	if (rec) {
475 		struct tls_rec *first_rec;
476 
477 		/* Mark the record as ready for transmission */
478 		smp_store_mb(rec->tx_ready, true);
479 
480 		/* If received record is at head of tx_list, schedule tx */
481 		first_rec = list_first_entry(&ctx->tx_list,
482 					     struct tls_rec, list);
483 		if (rec == first_rec)
484 			ready = true;
485 	}
486 
487 	spin_lock_bh(&ctx->encrypt_compl_lock);
488 	pending = atomic_dec_return(&ctx->encrypt_pending);
489 
490 	if (!pending && ctx->async_notify)
491 		complete(&ctx->async_wait.completion);
492 	spin_unlock_bh(&ctx->encrypt_compl_lock);
493 
494 	if (!ready)
495 		return;
496 
497 	/* Schedule the transmission */
498 	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
499 		schedule_delayed_work(&ctx->tx_work.work, 1);
500 }
501 
tls_do_encryption(struct sock * sk,struct tls_context * tls_ctx,struct tls_sw_context_tx * ctx,struct aead_request * aead_req,size_t data_len,u32 start)502 static int tls_do_encryption(struct sock *sk,
503 			     struct tls_context *tls_ctx,
504 			     struct tls_sw_context_tx *ctx,
505 			     struct aead_request *aead_req,
506 			     size_t data_len, u32 start)
507 {
508 	struct tls_prot_info *prot = &tls_ctx->prot_info;
509 	struct tls_rec *rec = ctx->open_rec;
510 	struct sk_msg *msg_en = &rec->msg_encrypted;
511 	struct scatterlist *sge = sk_msg_elem(msg_en, start);
512 	int rc, iv_offset = 0;
513 
514 	/* For CCM based ciphers, first byte of IV is a constant */
515 	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
516 		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
517 		iv_offset = 1;
518 	}
519 
520 	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
521 	       prot->iv_size + prot->salt_size);
522 
523 	xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq);
524 
525 	sge->offset += prot->prepend_size;
526 	sge->length -= prot->prepend_size;
527 
528 	msg_en->sg.curr = start;
529 
530 	aead_request_set_tfm(aead_req, ctx->aead_send);
531 	aead_request_set_ad(aead_req, prot->aad_size);
532 	aead_request_set_crypt(aead_req, rec->sg_aead_in,
533 			       rec->sg_aead_out,
534 			       data_len, rec->iv_data);
535 
536 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
537 				  tls_encrypt_done, sk);
538 
539 	/* Add the record in tx_list */
540 	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
541 	atomic_inc(&ctx->encrypt_pending);
542 
543 	rc = crypto_aead_encrypt(aead_req);
544 	if (!rc || rc != -EINPROGRESS) {
545 		atomic_dec(&ctx->encrypt_pending);
546 		sge->offset -= prot->prepend_size;
547 		sge->length += prot->prepend_size;
548 	}
549 
550 	if (!rc) {
551 		WRITE_ONCE(rec->tx_ready, true);
552 	} else if (rc != -EINPROGRESS) {
553 		list_del(&rec->list);
554 		return rc;
555 	}
556 
557 	/* Unhook the record from context if encryption is not failure */
558 	ctx->open_rec = NULL;
559 	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
560 	return rc;
561 }
562 
tls_split_open_record(struct sock * sk,struct tls_rec * from,struct tls_rec ** to,struct sk_msg * msg_opl,struct sk_msg * msg_oen,u32 split_point,u32 tx_overhead_size,u32 * orig_end)563 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
564 				 struct tls_rec **to, struct sk_msg *msg_opl,
565 				 struct sk_msg *msg_oen, u32 split_point,
566 				 u32 tx_overhead_size, u32 *orig_end)
567 {
568 	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
569 	struct scatterlist *sge, *osge, *nsge;
570 	u32 orig_size = msg_opl->sg.size;
571 	struct scatterlist tmp = { };
572 	struct sk_msg *msg_npl;
573 	struct tls_rec *new;
574 	int ret;
575 
576 	new = tls_get_rec(sk);
577 	if (!new)
578 		return -ENOMEM;
579 	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
580 			   tx_overhead_size, 0);
581 	if (ret < 0) {
582 		tls_free_rec(sk, new);
583 		return ret;
584 	}
585 
586 	*orig_end = msg_opl->sg.end;
587 	i = msg_opl->sg.start;
588 	sge = sk_msg_elem(msg_opl, i);
589 	while (apply && sge->length) {
590 		if (sge->length > apply) {
591 			u32 len = sge->length - apply;
592 
593 			get_page(sg_page(sge));
594 			sg_set_page(&tmp, sg_page(sge), len,
595 				    sge->offset + apply);
596 			sge->length = apply;
597 			bytes += apply;
598 			apply = 0;
599 		} else {
600 			apply -= sge->length;
601 			bytes += sge->length;
602 		}
603 
604 		sk_msg_iter_var_next(i);
605 		if (i == msg_opl->sg.end)
606 			break;
607 		sge = sk_msg_elem(msg_opl, i);
608 	}
609 
610 	msg_opl->sg.end = i;
611 	msg_opl->sg.curr = i;
612 	msg_opl->sg.copybreak = 0;
613 	msg_opl->apply_bytes = 0;
614 	msg_opl->sg.size = bytes;
615 
616 	msg_npl = &new->msg_plaintext;
617 	msg_npl->apply_bytes = apply;
618 	msg_npl->sg.size = orig_size - bytes;
619 
620 	j = msg_npl->sg.start;
621 	nsge = sk_msg_elem(msg_npl, j);
622 	if (tmp.length) {
623 		memcpy(nsge, &tmp, sizeof(*nsge));
624 		sk_msg_iter_var_next(j);
625 		nsge = sk_msg_elem(msg_npl, j);
626 	}
627 
628 	osge = sk_msg_elem(msg_opl, i);
629 	while (osge->length) {
630 		memcpy(nsge, osge, sizeof(*nsge));
631 		sg_unmark_end(nsge);
632 		sk_msg_iter_var_next(i);
633 		sk_msg_iter_var_next(j);
634 		if (i == *orig_end)
635 			break;
636 		osge = sk_msg_elem(msg_opl, i);
637 		nsge = sk_msg_elem(msg_npl, j);
638 	}
639 
640 	msg_npl->sg.end = j;
641 	msg_npl->sg.curr = j;
642 	msg_npl->sg.copybreak = 0;
643 
644 	*to = new;
645 	return 0;
646 }
647 
tls_merge_open_record(struct sock * sk,struct tls_rec * to,struct tls_rec * from,u32 orig_end)648 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
649 				  struct tls_rec *from, u32 orig_end)
650 {
651 	struct sk_msg *msg_npl = &from->msg_plaintext;
652 	struct sk_msg *msg_opl = &to->msg_plaintext;
653 	struct scatterlist *osge, *nsge;
654 	u32 i, j;
655 
656 	i = msg_opl->sg.end;
657 	sk_msg_iter_var_prev(i);
658 	j = msg_npl->sg.start;
659 
660 	osge = sk_msg_elem(msg_opl, i);
661 	nsge = sk_msg_elem(msg_npl, j);
662 
663 	if (sg_page(osge) == sg_page(nsge) &&
664 	    osge->offset + osge->length == nsge->offset) {
665 		osge->length += nsge->length;
666 		put_page(sg_page(nsge));
667 	}
668 
669 	msg_opl->sg.end = orig_end;
670 	msg_opl->sg.curr = orig_end;
671 	msg_opl->sg.copybreak = 0;
672 	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
673 	msg_opl->sg.size += msg_npl->sg.size;
674 
675 	sk_msg_free(sk, &to->msg_encrypted);
676 	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
677 
678 	kfree(from);
679 }
680 
tls_push_record(struct sock * sk,int flags,unsigned char record_type)681 static int tls_push_record(struct sock *sk, int flags,
682 			   unsigned char record_type)
683 {
684 	struct tls_context *tls_ctx = tls_get_ctx(sk);
685 	struct tls_prot_info *prot = &tls_ctx->prot_info;
686 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
687 	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
688 	u32 i, split_point, orig_end;
689 	struct sk_msg *msg_pl, *msg_en;
690 	struct aead_request *req;
691 	bool split;
692 	int rc;
693 
694 	if (!rec)
695 		return 0;
696 
697 	msg_pl = &rec->msg_plaintext;
698 	msg_en = &rec->msg_encrypted;
699 
700 	split_point = msg_pl->apply_bytes;
701 	split = split_point && split_point < msg_pl->sg.size;
702 	if (unlikely((!split &&
703 		      msg_pl->sg.size +
704 		      prot->overhead_size > msg_en->sg.size) ||
705 		     (split &&
706 		      split_point +
707 		      prot->overhead_size > msg_en->sg.size))) {
708 		split = true;
709 		split_point = msg_en->sg.size;
710 	}
711 	if (split) {
712 		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
713 					   split_point, prot->overhead_size,
714 					   &orig_end);
715 		if (rc < 0)
716 			return rc;
717 		/* This can happen if above tls_split_open_record allocates
718 		 * a single large encryption buffer instead of two smaller
719 		 * ones. In this case adjust pointers and continue without
720 		 * split.
721 		 */
722 		if (!msg_pl->sg.size) {
723 			tls_merge_open_record(sk, rec, tmp, orig_end);
724 			msg_pl = &rec->msg_plaintext;
725 			msg_en = &rec->msg_encrypted;
726 			split = false;
727 		}
728 		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
729 			    prot->overhead_size);
730 	}
731 
732 	rec->tx_flags = flags;
733 	req = &rec->aead_req;
734 
735 	i = msg_pl->sg.end;
736 	sk_msg_iter_var_prev(i);
737 
738 	rec->content_type = record_type;
739 	if (prot->version == TLS_1_3_VERSION) {
740 		/* Add content type to end of message.  No padding added */
741 		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
742 		sg_mark_end(&rec->sg_content_type);
743 		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
744 			 &rec->sg_content_type);
745 	} else {
746 		sg_mark_end(sk_msg_elem(msg_pl, i));
747 	}
748 
749 	if (msg_pl->sg.end < msg_pl->sg.start) {
750 		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
751 			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
752 			 msg_pl->sg.data);
753 	}
754 
755 	i = msg_pl->sg.start;
756 	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
757 
758 	i = msg_en->sg.end;
759 	sk_msg_iter_var_prev(i);
760 	sg_mark_end(sk_msg_elem(msg_en, i));
761 
762 	i = msg_en->sg.start;
763 	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
764 
765 	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
766 		     tls_ctx->tx.rec_seq, record_type, prot);
767 
768 	tls_fill_prepend(tls_ctx,
769 			 page_address(sg_page(&msg_en->sg.data[i])) +
770 			 msg_en->sg.data[i].offset,
771 			 msg_pl->sg.size + prot->tail_size,
772 			 record_type);
773 
774 	tls_ctx->pending_open_record_frags = false;
775 
776 	rc = tls_do_encryption(sk, tls_ctx, ctx, req,
777 			       msg_pl->sg.size + prot->tail_size, i);
778 	if (rc < 0) {
779 		if (rc != -EINPROGRESS) {
780 			tls_err_abort(sk, -EBADMSG);
781 			if (split) {
782 				tls_ctx->pending_open_record_frags = true;
783 				tls_merge_open_record(sk, rec, tmp, orig_end);
784 			}
785 		}
786 		ctx->async_capable = 1;
787 		return rc;
788 	} else if (split) {
789 		msg_pl = &tmp->msg_plaintext;
790 		msg_en = &tmp->msg_encrypted;
791 		sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
792 		tls_ctx->pending_open_record_frags = true;
793 		ctx->open_rec = tmp;
794 	}
795 
796 	return tls_tx_records(sk, flags);
797 }
798 
bpf_exec_tx_verdict(struct sk_msg * msg,struct sock * sk,bool full_record,u8 record_type,ssize_t * copied,int flags)799 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
800 			       bool full_record, u8 record_type,
801 			       ssize_t *copied, int flags)
802 {
803 	struct tls_context *tls_ctx = tls_get_ctx(sk);
804 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
805 	struct sk_msg msg_redir = { };
806 	struct sk_psock *psock;
807 	struct sock *sk_redir;
808 	struct tls_rec *rec;
809 	bool enospc, policy, redir_ingress;
810 	int err = 0, send;
811 	u32 delta = 0;
812 
813 	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
814 	psock = sk_psock_get(sk);
815 	if (!psock || !policy) {
816 		err = tls_push_record(sk, flags, record_type);
817 		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
818 			*copied -= sk_msg_free(sk, msg);
819 			tls_free_open_rec(sk);
820 			err = -sk->sk_err;
821 		}
822 		if (psock)
823 			sk_psock_put(sk, psock);
824 		return err;
825 	}
826 more_data:
827 	enospc = sk_msg_full(msg);
828 	if (psock->eval == __SK_NONE) {
829 		delta = msg->sg.size;
830 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
831 		delta -= msg->sg.size;
832 	}
833 	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
834 	    !enospc && !full_record) {
835 		err = -ENOSPC;
836 		goto out_err;
837 	}
838 	msg->cork_bytes = 0;
839 	send = msg->sg.size;
840 	if (msg->apply_bytes && msg->apply_bytes < send)
841 		send = msg->apply_bytes;
842 
843 	switch (psock->eval) {
844 	case __SK_PASS:
845 		err = tls_push_record(sk, flags, record_type);
846 		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
847 			*copied -= sk_msg_free(sk, msg);
848 			tls_free_open_rec(sk);
849 			err = -sk->sk_err;
850 			goto out_err;
851 		}
852 		break;
853 	case __SK_REDIRECT:
854 		redir_ingress = psock->redir_ingress;
855 		sk_redir = psock->sk_redir;
856 		memcpy(&msg_redir, msg, sizeof(*msg));
857 		if (msg->apply_bytes < send)
858 			msg->apply_bytes = 0;
859 		else
860 			msg->apply_bytes -= send;
861 		sk_msg_return_zero(sk, msg, send);
862 		msg->sg.size -= send;
863 		release_sock(sk);
864 		err = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
865 					    &msg_redir, send, flags);
866 		lock_sock(sk);
867 		if (err < 0) {
868 			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
869 			msg->sg.size = 0;
870 		}
871 		if (msg->sg.size == 0)
872 			tls_free_open_rec(sk);
873 		break;
874 	case __SK_DROP:
875 	default:
876 		sk_msg_free_partial(sk, msg, send);
877 		if (msg->apply_bytes < send)
878 			msg->apply_bytes = 0;
879 		else
880 			msg->apply_bytes -= send;
881 		if (msg->sg.size == 0)
882 			tls_free_open_rec(sk);
883 		*copied -= (send + delta);
884 		err = -EACCES;
885 	}
886 
887 	if (likely(!err)) {
888 		bool reset_eval = !ctx->open_rec;
889 
890 		rec = ctx->open_rec;
891 		if (rec) {
892 			msg = &rec->msg_plaintext;
893 			if (!msg->apply_bytes)
894 				reset_eval = true;
895 		}
896 		if (reset_eval) {
897 			psock->eval = __SK_NONE;
898 			if (psock->sk_redir) {
899 				sock_put(psock->sk_redir);
900 				psock->sk_redir = NULL;
901 			}
902 		}
903 		if (rec)
904 			goto more_data;
905 	}
906  out_err:
907 	sk_psock_put(sk, psock);
908 	return err;
909 }
910 
tls_sw_push_pending_record(struct sock * sk,int flags)911 static int tls_sw_push_pending_record(struct sock *sk, int flags)
912 {
913 	struct tls_context *tls_ctx = tls_get_ctx(sk);
914 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
915 	struct tls_rec *rec = ctx->open_rec;
916 	struct sk_msg *msg_pl;
917 	size_t copied;
918 
919 	if (!rec)
920 		return 0;
921 
922 	msg_pl = &rec->msg_plaintext;
923 	copied = msg_pl->sg.size;
924 	if (!copied)
925 		return 0;
926 
927 	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
928 				   &copied, flags);
929 }
930 
tls_sw_sendmsg(struct sock * sk,struct msghdr * msg,size_t size)931 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
932 {
933 	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
934 	struct tls_context *tls_ctx = tls_get_ctx(sk);
935 	struct tls_prot_info *prot = &tls_ctx->prot_info;
936 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
937 	bool async_capable = ctx->async_capable;
938 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
939 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
940 	bool eor = !(msg->msg_flags & MSG_MORE);
941 	size_t try_to_copy;
942 	ssize_t copied = 0;
943 	struct sk_msg *msg_pl, *msg_en;
944 	struct tls_rec *rec;
945 	int required_size;
946 	int num_async = 0;
947 	bool full_record;
948 	int record_room;
949 	int num_zc = 0;
950 	int orig_size;
951 	int ret = 0;
952 	int pending;
953 
954 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
955 			       MSG_CMSG_COMPAT))
956 		return -EOPNOTSUPP;
957 
958 	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
959 	if (ret)
960 		return ret;
961 	lock_sock(sk);
962 
963 	if (unlikely(msg->msg_controllen)) {
964 		ret = tls_proccess_cmsg(sk, msg, &record_type);
965 		if (ret) {
966 			if (ret == -EINPROGRESS)
967 				num_async++;
968 			else if (ret != -EAGAIN)
969 				goto send_end;
970 		}
971 	}
972 
973 	while (msg_data_left(msg)) {
974 		if (sk->sk_err) {
975 			ret = -sk->sk_err;
976 			goto send_end;
977 		}
978 
979 		if (ctx->open_rec)
980 			rec = ctx->open_rec;
981 		else
982 			rec = ctx->open_rec = tls_get_rec(sk);
983 		if (!rec) {
984 			ret = -ENOMEM;
985 			goto send_end;
986 		}
987 
988 		msg_pl = &rec->msg_plaintext;
989 		msg_en = &rec->msg_encrypted;
990 
991 		orig_size = msg_pl->sg.size;
992 		full_record = false;
993 		try_to_copy = msg_data_left(msg);
994 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
995 		if (try_to_copy >= record_room) {
996 			try_to_copy = record_room;
997 			full_record = true;
998 		}
999 
1000 		required_size = msg_pl->sg.size + try_to_copy +
1001 				prot->overhead_size;
1002 
1003 		if (!sk_stream_memory_free(sk))
1004 			goto wait_for_sndbuf;
1005 
1006 alloc_encrypted:
1007 		ret = tls_alloc_encrypted_msg(sk, required_size);
1008 		if (ret) {
1009 			if (ret != -ENOSPC)
1010 				goto wait_for_memory;
1011 
1012 			/* Adjust try_to_copy according to the amount that was
1013 			 * actually allocated. The difference is due
1014 			 * to max sg elements limit
1015 			 */
1016 			try_to_copy -= required_size - msg_en->sg.size;
1017 			full_record = true;
1018 		}
1019 
1020 		if (!is_kvec && (full_record || eor) && !async_capable) {
1021 			u32 first = msg_pl->sg.end;
1022 
1023 			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1024 							msg_pl, try_to_copy);
1025 			if (ret)
1026 				goto fallback_to_reg_send;
1027 
1028 			num_zc++;
1029 			copied += try_to_copy;
1030 
1031 			sk_msg_sg_copy_set(msg_pl, first);
1032 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1033 						  record_type, &copied,
1034 						  msg->msg_flags);
1035 			if (ret) {
1036 				if (ret == -EINPROGRESS)
1037 					num_async++;
1038 				else if (ret == -ENOMEM)
1039 					goto wait_for_memory;
1040 				else if (ctx->open_rec && ret == -ENOSPC)
1041 					goto rollback_iter;
1042 				else if (ret != -EAGAIN)
1043 					goto send_end;
1044 			}
1045 			continue;
1046 rollback_iter:
1047 			copied -= try_to_copy;
1048 			sk_msg_sg_copy_clear(msg_pl, first);
1049 			iov_iter_revert(&msg->msg_iter,
1050 					msg_pl->sg.size - orig_size);
1051 fallback_to_reg_send:
1052 			sk_msg_trim(sk, msg_pl, orig_size);
1053 		}
1054 
1055 		required_size = msg_pl->sg.size + try_to_copy;
1056 
1057 		ret = tls_clone_plaintext_msg(sk, required_size);
1058 		if (ret) {
1059 			if (ret != -ENOSPC)
1060 				goto send_end;
1061 
1062 			/* Adjust try_to_copy according to the amount that was
1063 			 * actually allocated. The difference is due
1064 			 * to max sg elements limit
1065 			 */
1066 			try_to_copy -= required_size - msg_pl->sg.size;
1067 			full_record = true;
1068 			sk_msg_trim(sk, msg_en,
1069 				    msg_pl->sg.size + prot->overhead_size);
1070 		}
1071 
1072 		if (try_to_copy) {
1073 			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1074 						       msg_pl, try_to_copy);
1075 			if (ret < 0)
1076 				goto trim_sgl;
1077 		}
1078 
1079 		/* Open records defined only if successfully copied, otherwise
1080 		 * we would trim the sg but not reset the open record frags.
1081 		 */
1082 		tls_ctx->pending_open_record_frags = true;
1083 		copied += try_to_copy;
1084 		if (full_record || eor) {
1085 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1086 						  record_type, &copied,
1087 						  msg->msg_flags);
1088 			if (ret) {
1089 				if (ret == -EINPROGRESS)
1090 					num_async++;
1091 				else if (ret == -ENOMEM)
1092 					goto wait_for_memory;
1093 				else if (ret != -EAGAIN) {
1094 					if (ret == -ENOSPC)
1095 						ret = 0;
1096 					goto send_end;
1097 				}
1098 			}
1099 		}
1100 
1101 		continue;
1102 
1103 wait_for_sndbuf:
1104 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1105 wait_for_memory:
1106 		ret = sk_stream_wait_memory(sk, &timeo);
1107 		if (ret) {
1108 trim_sgl:
1109 			if (ctx->open_rec)
1110 				tls_trim_both_msgs(sk, orig_size);
1111 			goto send_end;
1112 		}
1113 
1114 		if (ctx->open_rec && msg_en->sg.size < required_size)
1115 			goto alloc_encrypted;
1116 	}
1117 
1118 	if (!num_async) {
1119 		goto send_end;
1120 	} else if (num_zc) {
1121 		/* Wait for pending encryptions to get completed */
1122 		spin_lock_bh(&ctx->encrypt_compl_lock);
1123 		ctx->async_notify = true;
1124 
1125 		pending = atomic_read(&ctx->encrypt_pending);
1126 		spin_unlock_bh(&ctx->encrypt_compl_lock);
1127 		if (pending)
1128 			crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1129 		else
1130 			reinit_completion(&ctx->async_wait.completion);
1131 
1132 		/* There can be no concurrent accesses, since we have no
1133 		 * pending encrypt operations
1134 		 */
1135 		WRITE_ONCE(ctx->async_notify, false);
1136 
1137 		if (ctx->async_wait.err) {
1138 			ret = ctx->async_wait.err;
1139 			copied = 0;
1140 		}
1141 	}
1142 
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, msg->msg_flags);
1147 	}
1148 
1149 send_end:
1150 	ret = sk_stream_error(sk, msg->msg_flags, ret);
1151 
1152 	release_sock(sk);
1153 	mutex_unlock(&tls_ctx->tx_lock);
1154 	return copied > 0 ? copied : ret;
1155 }
1156 
tls_sw_do_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)1157 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1158 			      int offset, size_t size, int flags)
1159 {
1160 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1161 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1162 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1163 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1164 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1165 	struct sk_msg *msg_pl;
1166 	struct tls_rec *rec;
1167 	int num_async = 0;
1168 	ssize_t copied = 0;
1169 	bool full_record;
1170 	int record_room;
1171 	int ret = 0;
1172 	bool eor;
1173 
1174 	eor = !(flags & MSG_SENDPAGE_NOTLAST);
1175 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1176 
1177 	/* Call the sk_stream functions to manage the sndbuf mem. */
1178 	while (size > 0) {
1179 		size_t copy, required_size;
1180 
1181 		if (sk->sk_err) {
1182 			ret = -sk->sk_err;
1183 			goto sendpage_end;
1184 		}
1185 
1186 		if (ctx->open_rec)
1187 			rec = ctx->open_rec;
1188 		else
1189 			rec = ctx->open_rec = tls_get_rec(sk);
1190 		if (!rec) {
1191 			ret = -ENOMEM;
1192 			goto sendpage_end;
1193 		}
1194 
1195 		msg_pl = &rec->msg_plaintext;
1196 
1197 		full_record = false;
1198 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1199 		copy = size;
1200 		if (copy >= record_room) {
1201 			copy = record_room;
1202 			full_record = true;
1203 		}
1204 
1205 		required_size = msg_pl->sg.size + copy + prot->overhead_size;
1206 
1207 		if (!sk_stream_memory_free(sk))
1208 			goto wait_for_sndbuf;
1209 alloc_payload:
1210 		ret = tls_alloc_encrypted_msg(sk, required_size);
1211 		if (ret) {
1212 			if (ret != -ENOSPC)
1213 				goto wait_for_memory;
1214 
1215 			/* Adjust copy according to the amount that was
1216 			 * actually allocated. The difference is due
1217 			 * to max sg elements limit
1218 			 */
1219 			copy -= required_size - msg_pl->sg.size;
1220 			full_record = true;
1221 		}
1222 
1223 		sk_msg_page_add(msg_pl, page, copy, offset);
1224 		msg_pl->sg.copybreak = 0;
1225 		msg_pl->sg.curr = msg_pl->sg.end;
1226 		sk_mem_charge(sk, copy);
1227 
1228 		offset += copy;
1229 		size -= copy;
1230 		copied += copy;
1231 
1232 		tls_ctx->pending_open_record_frags = true;
1233 		if (full_record || eor || sk_msg_full(msg_pl)) {
1234 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1235 						  record_type, &copied, flags);
1236 			if (ret) {
1237 				if (ret == -EINPROGRESS)
1238 					num_async++;
1239 				else if (ret == -ENOMEM)
1240 					goto wait_for_memory;
1241 				else if (ret != -EAGAIN) {
1242 					if (ret == -ENOSPC)
1243 						ret = 0;
1244 					goto sendpage_end;
1245 				}
1246 			}
1247 		}
1248 		continue;
1249 wait_for_sndbuf:
1250 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1251 wait_for_memory:
1252 		ret = sk_stream_wait_memory(sk, &timeo);
1253 		if (ret) {
1254 			if (ctx->open_rec)
1255 				tls_trim_both_msgs(sk, msg_pl->sg.size);
1256 			goto sendpage_end;
1257 		}
1258 
1259 		if (ctx->open_rec)
1260 			goto alloc_payload;
1261 	}
1262 
1263 	if (num_async) {
1264 		/* Transmit if any encryptions have completed */
1265 		if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1266 			cancel_delayed_work(&ctx->tx_work.work);
1267 			tls_tx_records(sk, flags);
1268 		}
1269 	}
1270 sendpage_end:
1271 	ret = sk_stream_error(sk, flags, ret);
1272 	return copied > 0 ? copied : ret;
1273 }
1274 
tls_sw_sendpage_locked(struct sock * sk,struct page * page,int offset,size_t size,int flags)1275 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1276 			   int offset, size_t size, int flags)
1277 {
1278 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1279 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1280 		      MSG_NO_SHARED_FRAGS))
1281 		return -EOPNOTSUPP;
1282 
1283 	return tls_sw_do_sendpage(sk, page, offset, size, flags);
1284 }
1285 
tls_sw_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)1286 int tls_sw_sendpage(struct sock *sk, struct page *page,
1287 		    int offset, size_t size, int flags)
1288 {
1289 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1290 	int ret;
1291 
1292 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1293 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1294 		return -EOPNOTSUPP;
1295 
1296 	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1297 	if (ret)
1298 		return ret;
1299 	lock_sock(sk);
1300 	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1301 	release_sock(sk);
1302 	mutex_unlock(&tls_ctx->tx_lock);
1303 	return ret;
1304 }
1305 
tls_wait_data(struct sock * sk,struct sk_psock * psock,bool nonblock,long timeo,int * err)1306 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1307 				     bool nonblock, long timeo, int *err)
1308 {
1309 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1310 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1311 	struct sk_buff *skb;
1312 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1313 
1314 	while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1315 		if (sk->sk_err) {
1316 			*err = sock_error(sk);
1317 			return NULL;
1318 		}
1319 
1320 		if (!skb_queue_empty(&sk->sk_receive_queue)) {
1321 			__strp_unpause(&ctx->strp);
1322 			if (ctx->recv_pkt)
1323 				return ctx->recv_pkt;
1324 		}
1325 
1326 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1327 			return NULL;
1328 
1329 		if (sock_flag(sk, SOCK_DONE))
1330 			return NULL;
1331 
1332 		if (nonblock || !timeo) {
1333 			*err = -EAGAIN;
1334 			return NULL;
1335 		}
1336 
1337 		add_wait_queue(sk_sleep(sk), &wait);
1338 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1339 		sk_wait_event(sk, &timeo,
1340 			      ctx->recv_pkt != skb ||
1341 			      !sk_psock_queue_empty(psock),
1342 			      &wait);
1343 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1344 		remove_wait_queue(sk_sleep(sk), &wait);
1345 
1346 		/* Handle signals */
1347 		if (signal_pending(current)) {
1348 			*err = sock_intr_errno(timeo);
1349 			return NULL;
1350 		}
1351 	}
1352 
1353 	return skb;
1354 }
1355 
tls_setup_from_iter(struct iov_iter * from,int length,int * pages_used,struct scatterlist * to,int to_max_pages)1356 static int tls_setup_from_iter(struct iov_iter *from,
1357 			       int length, int *pages_used,
1358 			       struct scatterlist *to,
1359 			       int to_max_pages)
1360 {
1361 	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1362 	struct page *pages[MAX_SKB_FRAGS];
1363 	unsigned int size = 0;
1364 	ssize_t copied, use;
1365 	size_t offset;
1366 
1367 	while (length > 0) {
1368 		i = 0;
1369 		maxpages = to_max_pages - num_elem;
1370 		if (maxpages == 0) {
1371 			rc = -EFAULT;
1372 			goto out;
1373 		}
1374 		copied = iov_iter_get_pages(from, pages,
1375 					    length,
1376 					    maxpages, &offset);
1377 		if (copied <= 0) {
1378 			rc = -EFAULT;
1379 			goto out;
1380 		}
1381 
1382 		iov_iter_advance(from, copied);
1383 
1384 		length -= copied;
1385 		size += copied;
1386 		while (copied) {
1387 			use = min_t(int, copied, PAGE_SIZE - offset);
1388 
1389 			sg_set_page(&to[num_elem],
1390 				    pages[i], use, offset);
1391 			sg_unmark_end(&to[num_elem]);
1392 			/* We do not uncharge memory from this API */
1393 
1394 			offset = 0;
1395 			copied -= use;
1396 
1397 			i++;
1398 			num_elem++;
1399 		}
1400 	}
1401 	/* Mark the end in the last sg entry if newly added */
1402 	if (num_elem > *pages_used)
1403 		sg_mark_end(&to[num_elem - 1]);
1404 out:
1405 	if (rc)
1406 		iov_iter_revert(from, size);
1407 	*pages_used = num_elem;
1408 
1409 	return rc;
1410 }
1411 
1412 /* This function decrypts the input skb into either out_iov or in out_sg
1413  * or in skb buffers itself. The input parameter 'zc' indicates if
1414  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1415  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1416  * NULL, then the decryption happens inside skb buffers itself, i.e.
1417  * zero-copy gets disabled and 'zc' is updated.
1418  */
1419 
decrypt_internal(struct sock * sk,struct sk_buff * skb,struct iov_iter * out_iov,struct scatterlist * out_sg,struct tls_decrypt_arg * darg)1420 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1421 			    struct iov_iter *out_iov,
1422 			    struct scatterlist *out_sg,
1423 			    struct tls_decrypt_arg *darg)
1424 {
1425 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1426 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1427 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1428 	struct strp_msg *rxm = strp_msg(skb);
1429 	struct tls_msg *tlm = tls_msg(skb);
1430 	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1431 	struct aead_request *aead_req;
1432 	struct sk_buff *unused;
1433 	u8 *aad, *iv, *mem = NULL;
1434 	struct scatterlist *sgin = NULL;
1435 	struct scatterlist *sgout = NULL;
1436 	const int data_len = rxm->full_len - prot->overhead_size +
1437 			     prot->tail_size;
1438 	int iv_offset = 0;
1439 
1440 	if (darg->zc && (out_iov || out_sg)) {
1441 		if (out_iov)
1442 			n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1443 		else
1444 			n_sgout = sg_nents(out_sg);
1445 		n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1446 				 rxm->full_len - prot->prepend_size);
1447 	} else {
1448 		n_sgout = 0;
1449 		darg->zc = false;
1450 		n_sgin = skb_cow_data(skb, 0, &unused);
1451 	}
1452 
1453 	if (n_sgin < 1)
1454 		return -EBADMSG;
1455 
1456 	/* Increment to accommodate AAD */
1457 	n_sgin = n_sgin + 1;
1458 
1459 	nsg = n_sgin + n_sgout;
1460 
1461 	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1462 	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1463 	mem_size = mem_size + prot->aad_size;
1464 	mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1465 
1466 	/* Allocate a single block of memory which contains
1467 	 * aead_req || sgin[] || sgout[] || aad || iv.
1468 	 * This order achieves correct alignment for aead_req, sgin, sgout.
1469 	 */
1470 	mem = kmalloc(mem_size, sk->sk_allocation);
1471 	if (!mem)
1472 		return -ENOMEM;
1473 
1474 	/* Segment the allocated memory */
1475 	aead_req = (struct aead_request *)mem;
1476 	sgin = (struct scatterlist *)(mem + aead_size);
1477 	sgout = sgin + n_sgin;
1478 	aad = (u8 *)(sgout + n_sgout);
1479 	iv = aad + prot->aad_size;
1480 
1481 	/* For CCM based ciphers, first byte of nonce+iv is always '2' */
1482 	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1483 		iv[0] = 2;
1484 		iv_offset = 1;
1485 	}
1486 
1487 	/* Prepare IV */
1488 	err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1489 			    iv + iv_offset + prot->salt_size,
1490 			    prot->iv_size);
1491 	if (err < 0) {
1492 		kfree(mem);
1493 		return err;
1494 	}
1495 	if (prot->version == TLS_1_3_VERSION ||
1496 	    prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1497 		memcpy(iv + iv_offset, tls_ctx->rx.iv,
1498 		       prot->iv_size + prot->salt_size);
1499 	else
1500 		memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1501 
1502 	xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq);
1503 
1504 	/* Prepare AAD */
1505 	tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1506 		     prot->tail_size,
1507 		     tls_ctx->rx.rec_seq, tlm->control, prot);
1508 
1509 	/* Prepare sgin */
1510 	sg_init_table(sgin, n_sgin);
1511 	sg_set_buf(&sgin[0], aad, prot->aad_size);
1512 	err = skb_to_sgvec(skb, &sgin[1],
1513 			   rxm->offset + prot->prepend_size,
1514 			   rxm->full_len - prot->prepend_size);
1515 	if (err < 0) {
1516 		kfree(mem);
1517 		return err;
1518 	}
1519 
1520 	if (n_sgout) {
1521 		if (out_iov) {
1522 			sg_init_table(sgout, n_sgout);
1523 			sg_set_buf(&sgout[0], aad, prot->aad_size);
1524 
1525 			err = tls_setup_from_iter(out_iov, data_len,
1526 						  &pages, &sgout[1],
1527 						  (n_sgout - 1));
1528 			if (err < 0)
1529 				goto fallback_to_reg_recv;
1530 		} else if (out_sg) {
1531 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1532 		} else {
1533 			goto fallback_to_reg_recv;
1534 		}
1535 	} else {
1536 fallback_to_reg_recv:
1537 		sgout = sgin;
1538 		pages = 0;
1539 		darg->zc = false;
1540 	}
1541 
1542 	/* Prepare and submit AEAD request */
1543 	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1544 				data_len, aead_req, darg);
1545 	if (darg->async)
1546 		return 0;
1547 
1548 	/* Release the pages in case iov was mapped to pages */
1549 	for (; pages > 0; pages--)
1550 		put_page(sg_page(&sgout[pages]));
1551 
1552 	kfree(mem);
1553 	return err;
1554 }
1555 
decrypt_skb_update(struct sock * sk,struct sk_buff * skb,struct iov_iter * dest,struct tls_decrypt_arg * darg)1556 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1557 			      struct iov_iter *dest,
1558 			      struct tls_decrypt_arg *darg)
1559 {
1560 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1561 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1562 	struct strp_msg *rxm = strp_msg(skb);
1563 	struct tls_msg *tlm = tls_msg(skb);
1564 	int pad, err;
1565 
1566 	if (tlm->decrypted) {
1567 		darg->zc = false;
1568 		darg->async = false;
1569 		return 0;
1570 	}
1571 
1572 	if (tls_ctx->rx_conf == TLS_HW) {
1573 		err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1574 		if (err < 0)
1575 			return err;
1576 		if (err > 0) {
1577 			tlm->decrypted = 1;
1578 			darg->zc = false;
1579 			darg->async = false;
1580 			goto decrypt_done;
1581 		}
1582 	}
1583 
1584 	err = decrypt_internal(sk, skb, dest, NULL, darg);
1585 	if (err < 0) {
1586 		if (err == -EBADMSG)
1587 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1588 		return err;
1589 	}
1590 	if (darg->async)
1591 		goto decrypt_next;
1592 
1593 decrypt_done:
1594 	pad = padding_length(prot, skb);
1595 	if (pad < 0)
1596 		return pad;
1597 
1598 	rxm->full_len -= pad;
1599 	rxm->offset += prot->prepend_size;
1600 	rxm->full_len -= prot->overhead_size;
1601 	tlm->decrypted = 1;
1602 decrypt_next:
1603 	tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1604 
1605 	return 0;
1606 }
1607 
decrypt_skb(struct sock * sk,struct sk_buff * skb,struct scatterlist * sgout)1608 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1609 		struct scatterlist *sgout)
1610 {
1611 	struct tls_decrypt_arg darg = { .zc = true, };
1612 
1613 	return decrypt_internal(sk, skb, NULL, sgout, &darg);
1614 }
1615 
tls_sw_advance_skb(struct sock * sk,struct sk_buff * skb,unsigned int len)1616 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1617 			       unsigned int len)
1618 {
1619 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1620 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1621 
1622 	if (skb) {
1623 		struct strp_msg *rxm = strp_msg(skb);
1624 
1625 		if (len < rxm->full_len) {
1626 			rxm->offset += len;
1627 			rxm->full_len -= len;
1628 			return false;
1629 		}
1630 		consume_skb(skb);
1631 	}
1632 
1633 	/* Finished with message */
1634 	ctx->recv_pkt = NULL;
1635 	__strp_unpause(&ctx->strp);
1636 
1637 	return true;
1638 }
1639 
tls_record_content_type(struct msghdr * msg,struct tls_msg * tlm,u8 * control)1640 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1641 				   u8 *control)
1642 {
1643 	int err;
1644 
1645 	if (!*control) {
1646 		*control = tlm->control;
1647 		if (!*control)
1648 			return -EBADMSG;
1649 
1650 		err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1651 			       sizeof(*control), control);
1652 		if (*control != TLS_RECORD_TYPE_DATA) {
1653 			if (err || msg->msg_flags & MSG_CTRUNC)
1654 				return -EIO;
1655 		}
1656 	} else if (*control != tlm->control) {
1657 		return 0;
1658 	}
1659 
1660 	return 1;
1661 }
1662 
1663 /* This function traverses the rx_list in tls receive context to copies the
1664  * decrypted records into the buffer provided by caller zero copy is not
1665  * true. Further, the records are removed from the rx_list if it is not a peek
1666  * case and the record has been consumed completely.
1667  */
process_rx_list(struct tls_sw_context_rx * ctx,struct msghdr * msg,u8 * control,size_t skip,size_t len,bool zc,bool is_peek)1668 static int process_rx_list(struct tls_sw_context_rx *ctx,
1669 			   struct msghdr *msg,
1670 			   u8 *control,
1671 			   size_t skip,
1672 			   size_t len,
1673 			   bool zc,
1674 			   bool is_peek)
1675 {
1676 	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1677 	struct tls_msg *tlm;
1678 	ssize_t copied = 0;
1679 	int err;
1680 
1681 	while (skip && skb) {
1682 		struct strp_msg *rxm = strp_msg(skb);
1683 		tlm = tls_msg(skb);
1684 
1685 		err = tls_record_content_type(msg, tlm, control);
1686 		if (err <= 0)
1687 			return err;
1688 
1689 		if (skip < rxm->full_len)
1690 			break;
1691 
1692 		skip = skip - rxm->full_len;
1693 		skb = skb_peek_next(skb, &ctx->rx_list);
1694 	}
1695 
1696 	while (len && skb) {
1697 		struct sk_buff *next_skb;
1698 		struct strp_msg *rxm = strp_msg(skb);
1699 		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1700 
1701 		tlm = tls_msg(skb);
1702 
1703 		err = tls_record_content_type(msg, tlm, control);
1704 		if (err <= 0)
1705 			return err;
1706 
1707 		if (!zc || (rxm->full_len - skip) > len) {
1708 			err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1709 						    msg, chunk);
1710 			if (err < 0)
1711 				return err;
1712 		}
1713 
1714 		len = len - chunk;
1715 		copied = copied + chunk;
1716 
1717 		/* Consume the data from record if it is non-peek case*/
1718 		if (!is_peek) {
1719 			rxm->offset = rxm->offset + chunk;
1720 			rxm->full_len = rxm->full_len - chunk;
1721 
1722 			/* Return if there is unconsumed data in the record */
1723 			if (rxm->full_len - skip)
1724 				break;
1725 		}
1726 
1727 		/* The remaining skip-bytes must lie in 1st record in rx_list.
1728 		 * So from the 2nd record, 'skip' should be 0.
1729 		 */
1730 		skip = 0;
1731 
1732 		if (msg)
1733 			msg->msg_flags |= MSG_EOR;
1734 
1735 		next_skb = skb_peek_next(skb, &ctx->rx_list);
1736 
1737 		if (!is_peek) {
1738 			skb_unlink(skb, &ctx->rx_list);
1739 			consume_skb(skb);
1740 		}
1741 
1742 		skb = next_skb;
1743 	}
1744 
1745 	return copied;
1746 }
1747 
tls_sw_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int nonblock,int flags,int * addr_len)1748 int tls_sw_recvmsg(struct sock *sk,
1749 		   struct msghdr *msg,
1750 		   size_t len,
1751 		   int nonblock,
1752 		   int flags,
1753 		   int *addr_len)
1754 {
1755 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1756 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1757 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1758 	struct sk_psock *psock;
1759 	unsigned char control = 0;
1760 	ssize_t decrypted = 0;
1761 	struct strp_msg *rxm;
1762 	struct tls_msg *tlm;
1763 	struct sk_buff *skb;
1764 	ssize_t copied = 0;
1765 	bool async = false;
1766 	int target, err = 0;
1767 	long timeo;
1768 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1769 	bool is_peek = flags & MSG_PEEK;
1770 	bool bpf_strp_enabled;
1771 
1772 	flags |= nonblock;
1773 
1774 	if (unlikely(flags & MSG_ERRQUEUE))
1775 		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1776 
1777 	psock = sk_psock_get(sk);
1778 	lock_sock(sk);
1779 	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1780 
1781 	/* Process pending decrypted records. It must be non-zero-copy */
1782 	err = process_rx_list(ctx, msg, &control, 0, len, false, is_peek);
1783 	if (err < 0) {
1784 		tls_err_abort(sk, err);
1785 		goto end;
1786 	}
1787 
1788 	copied = err;
1789 	if (len <= copied || (copied && control != TLS_RECORD_TYPE_DATA))
1790 		goto end;
1791 
1792 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1793 	len = len - copied;
1794 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1795 
1796 	decrypted = 0;
1797 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1798 		struct tls_decrypt_arg darg = {};
1799 		bool retain_skb = false;
1800 		int to_decrypt, chunk;
1801 
1802 		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1803 		if (!skb) {
1804 			if (psock) {
1805 				int ret = sk_msg_recvmsg(sk, psock, msg, len,
1806 							 flags);
1807 
1808 				if (ret > 0) {
1809 					decrypted += ret;
1810 					len -= ret;
1811 					continue;
1812 				}
1813 			}
1814 			goto recv_end;
1815 		}
1816 
1817 		rxm = strp_msg(skb);
1818 		tlm = tls_msg(skb);
1819 
1820 		to_decrypt = rxm->full_len - prot->overhead_size;
1821 
1822 		if (to_decrypt <= len && !is_kvec && !is_peek &&
1823 		    tlm->control == TLS_RECORD_TYPE_DATA &&
1824 		    prot->version != TLS_1_3_VERSION &&
1825 		    !bpf_strp_enabled)
1826 			darg.zc = true;
1827 
1828 		/* Do not use async mode if record is non-data */
1829 		if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1830 			darg.async = ctx->async_capable;
1831 		else
1832 			darg.async = false;
1833 
1834 		err = decrypt_skb_update(sk, skb, &msg->msg_iter, &darg);
1835 		if (err < 0) {
1836 			tls_err_abort(sk, -EBADMSG);
1837 			goto recv_end;
1838 		}
1839 
1840 		async |= darg.async;
1841 
1842 		/* If the type of records being processed is not known yet,
1843 		 * set it to record type just dequeued. If it is already known,
1844 		 * but does not match the record type just dequeued, go to end.
1845 		 * We always get record type here since for tls1.2, record type
1846 		 * is known just after record is dequeued from stream parser.
1847 		 * For tls1.3, we disable async.
1848 		 */
1849 		err = tls_record_content_type(msg, tlm, &control);
1850 		if (err <= 0)
1851 			goto recv_end;
1852 
1853 		if (async) {
1854 			/* TLS 1.2-only, to_decrypt must be text length */
1855 			chunk = min_t(int, to_decrypt, len);
1856 			goto pick_next_record;
1857 		}
1858 		/* TLS 1.3 may have updated the length by more than overhead */
1859 		chunk = rxm->full_len;
1860 
1861 		if (!darg.zc) {
1862 			if (bpf_strp_enabled) {
1863 				err = sk_psock_tls_strp_read(psock, skb);
1864 				if (err != __SK_PASS) {
1865 					rxm->offset = rxm->offset + rxm->full_len;
1866 					rxm->full_len = 0;
1867 					if (err == __SK_DROP)
1868 						consume_skb(skb);
1869 					ctx->recv_pkt = NULL;
1870 					__strp_unpause(&ctx->strp);
1871 					continue;
1872 				}
1873 			}
1874 
1875 			if (chunk > len) {
1876 				retain_skb = true;
1877 				chunk = len;
1878 			}
1879 
1880 			err = skb_copy_datagram_msg(skb, rxm->offset,
1881 						    msg, chunk);
1882 			if (err < 0)
1883 				goto recv_end;
1884 
1885 			if (!is_peek) {
1886 				rxm->offset = rxm->offset + chunk;
1887 				rxm->full_len = rxm->full_len - chunk;
1888 			}
1889 		}
1890 
1891 pick_next_record:
1892 		decrypted += chunk;
1893 		len -= chunk;
1894 
1895 		/* For async or peek case, queue the current skb */
1896 		if (async || is_peek || retain_skb) {
1897 			skb_queue_tail(&ctx->rx_list, skb);
1898 			skb = NULL;
1899 		}
1900 
1901 		if (tls_sw_advance_skb(sk, skb, chunk)) {
1902 			/* Return full control message to
1903 			 * userspace before trying to parse
1904 			 * another message type
1905 			 */
1906 			msg->msg_flags |= MSG_EOR;
1907 			if (control != TLS_RECORD_TYPE_DATA)
1908 				goto recv_end;
1909 		} else {
1910 			break;
1911 		}
1912 	}
1913 
1914 recv_end:
1915 	if (async) {
1916 		int pending;
1917 
1918 		/* Wait for all previously submitted records to be decrypted */
1919 		spin_lock_bh(&ctx->decrypt_compl_lock);
1920 		ctx->async_notify = true;
1921 		pending = atomic_read(&ctx->decrypt_pending);
1922 		spin_unlock_bh(&ctx->decrypt_compl_lock);
1923 		if (pending) {
1924 			err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1925 			if (err) {
1926 				/* one of async decrypt failed */
1927 				tls_err_abort(sk, err);
1928 				copied = 0;
1929 				decrypted = 0;
1930 				goto end;
1931 			}
1932 		} else {
1933 			reinit_completion(&ctx->async_wait.completion);
1934 		}
1935 
1936 		/* There can be no concurrent accesses, since we have no
1937 		 * pending decrypt operations
1938 		 */
1939 		WRITE_ONCE(ctx->async_notify, false);
1940 
1941 		/* Drain records from the rx_list & copy if required */
1942 		if (is_peek || is_kvec)
1943 			err = process_rx_list(ctx, msg, &control, copied,
1944 					      decrypted, false, is_peek);
1945 		else
1946 			err = process_rx_list(ctx, msg, &control, 0,
1947 					      decrypted, true, is_peek);
1948 		if (err < 0) {
1949 			tls_err_abort(sk, err);
1950 			copied = 0;
1951 			goto end;
1952 		}
1953 	}
1954 
1955 	copied += decrypted;
1956 
1957 end:
1958 	release_sock(sk);
1959 	if (psock)
1960 		sk_psock_put(sk, psock);
1961 	return copied ? : err;
1962 }
1963 
tls_sw_splice_read(struct socket * sock,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)1964 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
1965 			   struct pipe_inode_info *pipe,
1966 			   size_t len, unsigned int flags)
1967 {
1968 	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1969 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1970 	struct strp_msg *rxm = NULL;
1971 	struct sock *sk = sock->sk;
1972 	struct tls_msg *tlm;
1973 	struct sk_buff *skb;
1974 	ssize_t copied = 0;
1975 	bool from_queue;
1976 	int err = 0;
1977 	long timeo;
1978 	int chunk;
1979 
1980 	lock_sock(sk);
1981 
1982 	timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
1983 
1984 	from_queue = !skb_queue_empty(&ctx->rx_list);
1985 	if (from_queue) {
1986 		skb = __skb_dequeue(&ctx->rx_list);
1987 	} else {
1988 		struct tls_decrypt_arg darg = {};
1989 
1990 		skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
1991 				    &err);
1992 		if (!skb)
1993 			goto splice_read_end;
1994 
1995 		err = decrypt_skb_update(sk, skb, NULL, &darg);
1996 		if (err < 0) {
1997 			tls_err_abort(sk, -EBADMSG);
1998 			goto splice_read_end;
1999 		}
2000 	}
2001 
2002 	rxm = strp_msg(skb);
2003 	tlm = tls_msg(skb);
2004 
2005 	/* splice does not support reading control messages */
2006 	if (tlm->control != TLS_RECORD_TYPE_DATA) {
2007 		err = -EINVAL;
2008 		goto splice_read_end;
2009 	}
2010 
2011 	chunk = min_t(unsigned int, rxm->full_len, len);
2012 	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2013 	if (copied < 0)
2014 		goto splice_read_end;
2015 
2016 	if (!from_queue) {
2017 		ctx->recv_pkt = NULL;
2018 		__strp_unpause(&ctx->strp);
2019 	}
2020 	if (chunk < rxm->full_len) {
2021 		__skb_queue_head(&ctx->rx_list, skb);
2022 		rxm->offset += len;
2023 		rxm->full_len -= len;
2024 	} else {
2025 		consume_skb(skb);
2026 	}
2027 
2028 splice_read_end:
2029 	release_sock(sk);
2030 	return copied ? : err;
2031 }
2032 
tls_sw_sock_is_readable(struct sock * sk)2033 bool tls_sw_sock_is_readable(struct sock *sk)
2034 {
2035 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2036 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2037 	bool ingress_empty = true;
2038 	struct sk_psock *psock;
2039 
2040 	rcu_read_lock();
2041 	psock = sk_psock(sk);
2042 	if (psock)
2043 		ingress_empty = list_empty(&psock->ingress_msg);
2044 	rcu_read_unlock();
2045 
2046 	return !ingress_empty || ctx->recv_pkt ||
2047 		!skb_queue_empty(&ctx->rx_list);
2048 }
2049 
tls_read_size(struct strparser * strp,struct sk_buff * skb)2050 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2051 {
2052 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2053 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2054 	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2055 	struct strp_msg *rxm = strp_msg(skb);
2056 	struct tls_msg *tlm = tls_msg(skb);
2057 	size_t cipher_overhead;
2058 	size_t data_len = 0;
2059 	int ret;
2060 
2061 	/* Verify that we have a full TLS header, or wait for more data */
2062 	if (rxm->offset + prot->prepend_size > skb->len)
2063 		return 0;
2064 
2065 	/* Sanity-check size of on-stack buffer. */
2066 	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2067 		ret = -EINVAL;
2068 		goto read_failure;
2069 	}
2070 
2071 	/* Linearize header to local buffer */
2072 	ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2073 
2074 	if (ret < 0)
2075 		goto read_failure;
2076 
2077 	tlm->control = header[0];
2078 
2079 	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2080 
2081 	cipher_overhead = prot->tag_size;
2082 	if (prot->version != TLS_1_3_VERSION &&
2083 	    prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2084 		cipher_overhead += prot->iv_size;
2085 
2086 	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2087 	    prot->tail_size) {
2088 		ret = -EMSGSIZE;
2089 		goto read_failure;
2090 	}
2091 	if (data_len < cipher_overhead) {
2092 		ret = -EBADMSG;
2093 		goto read_failure;
2094 	}
2095 
2096 	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2097 	if (header[1] != TLS_1_2_VERSION_MINOR ||
2098 	    header[2] != TLS_1_2_VERSION_MAJOR) {
2099 		ret = -EINVAL;
2100 		goto read_failure;
2101 	}
2102 
2103 	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2104 				     TCP_SKB_CB(skb)->seq + rxm->offset);
2105 	return data_len + TLS_HEADER_SIZE;
2106 
2107 read_failure:
2108 	tls_err_abort(strp->sk, ret);
2109 
2110 	return ret;
2111 }
2112 
tls_queue(struct strparser * strp,struct sk_buff * skb)2113 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2114 {
2115 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2116 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2117 	struct tls_msg *tlm = tls_msg(skb);
2118 
2119 	tlm->decrypted = 0;
2120 
2121 	ctx->recv_pkt = skb;
2122 	strp_pause(strp);
2123 
2124 	ctx->saved_data_ready(strp->sk);
2125 }
2126 
tls_data_ready(struct sock * sk)2127 static void tls_data_ready(struct sock *sk)
2128 {
2129 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2130 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2131 	struct sk_psock *psock;
2132 
2133 	strp_data_ready(&ctx->strp);
2134 
2135 	psock = sk_psock_get(sk);
2136 	if (psock) {
2137 		if (!list_empty(&psock->ingress_msg))
2138 			ctx->saved_data_ready(sk);
2139 		sk_psock_put(sk, psock);
2140 	}
2141 }
2142 
tls_sw_cancel_work_tx(struct tls_context * tls_ctx)2143 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2144 {
2145 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2146 
2147 	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2148 	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2149 	cancel_delayed_work_sync(&ctx->tx_work.work);
2150 }
2151 
tls_sw_release_resources_tx(struct sock * sk)2152 void tls_sw_release_resources_tx(struct sock *sk)
2153 {
2154 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2155 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2156 	struct tls_rec *rec, *tmp;
2157 	int pending;
2158 
2159 	/* Wait for any pending async encryptions to complete */
2160 	spin_lock_bh(&ctx->encrypt_compl_lock);
2161 	ctx->async_notify = true;
2162 	pending = atomic_read(&ctx->encrypt_pending);
2163 	spin_unlock_bh(&ctx->encrypt_compl_lock);
2164 
2165 	if (pending)
2166 		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2167 
2168 	tls_tx_records(sk, -1);
2169 
2170 	/* Free up un-sent records in tx_list. First, free
2171 	 * the partially sent record if any at head of tx_list.
2172 	 */
2173 	if (tls_ctx->partially_sent_record) {
2174 		tls_free_partial_record(sk, tls_ctx);
2175 		rec = list_first_entry(&ctx->tx_list,
2176 				       struct tls_rec, list);
2177 		list_del(&rec->list);
2178 		sk_msg_free(sk, &rec->msg_plaintext);
2179 		kfree(rec);
2180 	}
2181 
2182 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2183 		list_del(&rec->list);
2184 		sk_msg_free(sk, &rec->msg_encrypted);
2185 		sk_msg_free(sk, &rec->msg_plaintext);
2186 		kfree(rec);
2187 	}
2188 
2189 	crypto_free_aead(ctx->aead_send);
2190 	tls_free_open_rec(sk);
2191 }
2192 
tls_sw_free_ctx_tx(struct tls_context * tls_ctx)2193 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2194 {
2195 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2196 
2197 	kfree(ctx);
2198 }
2199 
tls_sw_release_resources_rx(struct sock * sk)2200 void tls_sw_release_resources_rx(struct sock *sk)
2201 {
2202 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2203 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2204 
2205 	kfree(tls_ctx->rx.rec_seq);
2206 	kfree(tls_ctx->rx.iv);
2207 
2208 	if (ctx->aead_recv) {
2209 		kfree_skb(ctx->recv_pkt);
2210 		ctx->recv_pkt = NULL;
2211 		skb_queue_purge(&ctx->rx_list);
2212 		crypto_free_aead(ctx->aead_recv);
2213 		strp_stop(&ctx->strp);
2214 		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2215 		 * we still want to strp_stop(), but sk->sk_data_ready was
2216 		 * never swapped.
2217 		 */
2218 		if (ctx->saved_data_ready) {
2219 			write_lock_bh(&sk->sk_callback_lock);
2220 			sk->sk_data_ready = ctx->saved_data_ready;
2221 			write_unlock_bh(&sk->sk_callback_lock);
2222 		}
2223 	}
2224 }
2225 
tls_sw_strparser_done(struct tls_context * tls_ctx)2226 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2227 {
2228 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2229 
2230 	strp_done(&ctx->strp);
2231 }
2232 
tls_sw_free_ctx_rx(struct tls_context * tls_ctx)2233 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2234 {
2235 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2236 
2237 	kfree(ctx);
2238 }
2239 
tls_sw_free_resources_rx(struct sock * sk)2240 void tls_sw_free_resources_rx(struct sock *sk)
2241 {
2242 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2243 
2244 	tls_sw_release_resources_rx(sk);
2245 	tls_sw_free_ctx_rx(tls_ctx);
2246 }
2247 
2248 /* The work handler to transmitt the encrypted records in tx_list */
tx_work_handler(struct work_struct * work)2249 static void tx_work_handler(struct work_struct *work)
2250 {
2251 	struct delayed_work *delayed_work = to_delayed_work(work);
2252 	struct tx_work *tx_work = container_of(delayed_work,
2253 					       struct tx_work, work);
2254 	struct sock *sk = tx_work->sk;
2255 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2256 	struct tls_sw_context_tx *ctx;
2257 
2258 	if (unlikely(!tls_ctx))
2259 		return;
2260 
2261 	ctx = tls_sw_ctx_tx(tls_ctx);
2262 	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2263 		return;
2264 
2265 	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2266 		return;
2267 
2268 	if (mutex_trylock(&tls_ctx->tx_lock)) {
2269 		lock_sock(sk);
2270 		tls_tx_records(sk, -1);
2271 		release_sock(sk);
2272 		mutex_unlock(&tls_ctx->tx_lock);
2273 	} else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2274 		/* Someone is holding the tx_lock, they will likely run Tx
2275 		 * and cancel the work on their way out of the lock section.
2276 		 * Schedule a long delay just in case.
2277 		 */
2278 		schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2279 	}
2280 }
2281 
tls_sw_write_space(struct sock * sk,struct tls_context * ctx)2282 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2283 {
2284 	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2285 
2286 	/* Schedule the transmission if tx list is ready */
2287 	if (is_tx_ready(tx_ctx) &&
2288 	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2289 		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2290 }
2291 
tls_sw_strparser_arm(struct sock * sk,struct tls_context * tls_ctx)2292 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2293 {
2294 	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2295 
2296 	write_lock_bh(&sk->sk_callback_lock);
2297 	rx_ctx->saved_data_ready = sk->sk_data_ready;
2298 	sk->sk_data_ready = tls_data_ready;
2299 	write_unlock_bh(&sk->sk_callback_lock);
2300 
2301 	strp_check_rcv(&rx_ctx->strp);
2302 }
2303 
tls_set_sw_offload(struct sock * sk,struct tls_context * ctx,int tx)2304 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2305 {
2306 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2307 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2308 	struct tls_crypto_info *crypto_info;
2309 	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2310 	struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2311 	struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2312 	struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2313 	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2314 	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2315 	struct cipher_context *cctx;
2316 	struct crypto_aead **aead;
2317 	struct strp_callbacks cb;
2318 	u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2319 	struct crypto_tfm *tfm;
2320 	char *iv, *rec_seq, *key, *salt, *cipher_name;
2321 	size_t keysize;
2322 	int rc = 0;
2323 
2324 	if (!ctx) {
2325 		rc = -EINVAL;
2326 		goto out;
2327 	}
2328 
2329 	if (tx) {
2330 		if (!ctx->priv_ctx_tx) {
2331 			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2332 			if (!sw_ctx_tx) {
2333 				rc = -ENOMEM;
2334 				goto out;
2335 			}
2336 			ctx->priv_ctx_tx = sw_ctx_tx;
2337 		} else {
2338 			sw_ctx_tx =
2339 				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2340 		}
2341 	} else {
2342 		if (!ctx->priv_ctx_rx) {
2343 			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2344 			if (!sw_ctx_rx) {
2345 				rc = -ENOMEM;
2346 				goto out;
2347 			}
2348 			ctx->priv_ctx_rx = sw_ctx_rx;
2349 		} else {
2350 			sw_ctx_rx =
2351 				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2352 		}
2353 	}
2354 
2355 	if (tx) {
2356 		crypto_init_wait(&sw_ctx_tx->async_wait);
2357 		spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2358 		crypto_info = &ctx->crypto_send.info;
2359 		cctx = &ctx->tx;
2360 		aead = &sw_ctx_tx->aead_send;
2361 		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2362 		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2363 		sw_ctx_tx->tx_work.sk = sk;
2364 	} else {
2365 		crypto_init_wait(&sw_ctx_rx->async_wait);
2366 		spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2367 		crypto_info = &ctx->crypto_recv.info;
2368 		cctx = &ctx->rx;
2369 		skb_queue_head_init(&sw_ctx_rx->rx_list);
2370 		aead = &sw_ctx_rx->aead_recv;
2371 	}
2372 
2373 	switch (crypto_info->cipher_type) {
2374 	case TLS_CIPHER_AES_GCM_128: {
2375 		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2376 		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2377 		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2378 		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2379 		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2380 		rec_seq =
2381 		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2382 		gcm_128_info =
2383 			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2384 		keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2385 		key = gcm_128_info->key;
2386 		salt = gcm_128_info->salt;
2387 		salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2388 		cipher_name = "gcm(aes)";
2389 		break;
2390 	}
2391 	case TLS_CIPHER_AES_GCM_256: {
2392 		nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2393 		tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2394 		iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2395 		iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2396 		rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2397 		rec_seq =
2398 		 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2399 		gcm_256_info =
2400 			(struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2401 		keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2402 		key = gcm_256_info->key;
2403 		salt = gcm_256_info->salt;
2404 		salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2405 		cipher_name = "gcm(aes)";
2406 		break;
2407 	}
2408 	case TLS_CIPHER_AES_CCM_128: {
2409 		nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2410 		tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2411 		iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2412 		iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2413 		rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2414 		rec_seq =
2415 		((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2416 		ccm_128_info =
2417 		(struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2418 		keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2419 		key = ccm_128_info->key;
2420 		salt = ccm_128_info->salt;
2421 		salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2422 		cipher_name = "ccm(aes)";
2423 		break;
2424 	}
2425 	case TLS_CIPHER_CHACHA20_POLY1305: {
2426 		chacha20_poly1305_info = (void *)crypto_info;
2427 		nonce_size = 0;
2428 		tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2429 		iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2430 		iv = chacha20_poly1305_info->iv;
2431 		rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2432 		rec_seq = chacha20_poly1305_info->rec_seq;
2433 		keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2434 		key = chacha20_poly1305_info->key;
2435 		salt = chacha20_poly1305_info->salt;
2436 		salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2437 		cipher_name = "rfc7539(chacha20,poly1305)";
2438 		break;
2439 	}
2440 	default:
2441 		rc = -EINVAL;
2442 		goto free_priv;
2443 	}
2444 
2445 	/* Sanity-check the sizes for stack allocations. */
2446 	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2447 	    rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2448 		rc = -EINVAL;
2449 		goto free_priv;
2450 	}
2451 
2452 	if (crypto_info->version == TLS_1_3_VERSION) {
2453 		nonce_size = 0;
2454 		prot->aad_size = TLS_HEADER_SIZE;
2455 		prot->tail_size = 1;
2456 	} else {
2457 		prot->aad_size = TLS_AAD_SPACE_SIZE;
2458 		prot->tail_size = 0;
2459 	}
2460 
2461 	prot->version = crypto_info->version;
2462 	prot->cipher_type = crypto_info->cipher_type;
2463 	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2464 	prot->tag_size = tag_size;
2465 	prot->overhead_size = prot->prepend_size +
2466 			      prot->tag_size + prot->tail_size;
2467 	prot->iv_size = iv_size;
2468 	prot->salt_size = salt_size;
2469 	cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2470 	if (!cctx->iv) {
2471 		rc = -ENOMEM;
2472 		goto free_priv;
2473 	}
2474 	/* Note: 128 & 256 bit salt are the same size */
2475 	prot->rec_seq_size = rec_seq_size;
2476 	memcpy(cctx->iv, salt, salt_size);
2477 	memcpy(cctx->iv + salt_size, iv, iv_size);
2478 	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2479 	if (!cctx->rec_seq) {
2480 		rc = -ENOMEM;
2481 		goto free_iv;
2482 	}
2483 
2484 	if (!*aead) {
2485 		*aead = crypto_alloc_aead(cipher_name, 0, 0);
2486 		if (IS_ERR(*aead)) {
2487 			rc = PTR_ERR(*aead);
2488 			*aead = NULL;
2489 			goto free_rec_seq;
2490 		}
2491 	}
2492 
2493 	ctx->push_pending_record = tls_sw_push_pending_record;
2494 
2495 	rc = crypto_aead_setkey(*aead, key, keysize);
2496 
2497 	if (rc)
2498 		goto free_aead;
2499 
2500 	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2501 	if (rc)
2502 		goto free_aead;
2503 
2504 	if (sw_ctx_rx) {
2505 		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2506 
2507 		if (crypto_info->version == TLS_1_3_VERSION)
2508 			sw_ctx_rx->async_capable = 0;
2509 		else
2510 			sw_ctx_rx->async_capable =
2511 				!!(tfm->__crt_alg->cra_flags &
2512 				   CRYPTO_ALG_ASYNC);
2513 
2514 		/* Set up strparser */
2515 		memset(&cb, 0, sizeof(cb));
2516 		cb.rcv_msg = tls_queue;
2517 		cb.parse_msg = tls_read_size;
2518 
2519 		strp_init(&sw_ctx_rx->strp, sk, &cb);
2520 	}
2521 
2522 	goto out;
2523 
2524 free_aead:
2525 	crypto_free_aead(*aead);
2526 	*aead = NULL;
2527 free_rec_seq:
2528 	kfree(cctx->rec_seq);
2529 	cctx->rec_seq = NULL;
2530 free_iv:
2531 	kfree(cctx->iv);
2532 	cctx->iv = NULL;
2533 free_priv:
2534 	if (tx) {
2535 		kfree(ctx->priv_ctx_tx);
2536 		ctx->priv_ctx_tx = NULL;
2537 	} else {
2538 		kfree(ctx->priv_ctx_rx);
2539 		ctx->priv_ctx_rx = NULL;
2540 	}
2541 out:
2542 	return rc;
2543 }
2544