• 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 
tls_err_abort(struct sock * sk,int err)47 noinline void tls_err_abort(struct sock *sk, int err)
48 {
49 	WARN_ON_ONCE(err >= 0);
50 	/* sk->sk_err should contain a positive error code. */
51 	sk->sk_err = -err;
52 	sk->sk_error_report(sk);
53 }
54 
__skb_nsg(struct sk_buff * skb,int offset,int len,unsigned int recursion_level)55 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
56                      unsigned int recursion_level)
57 {
58         int start = skb_headlen(skb);
59         int i, chunk = start - offset;
60         struct sk_buff *frag_iter;
61         int elt = 0;
62 
63         if (unlikely(recursion_level >= 24))
64                 return -EMSGSIZE;
65 
66         if (chunk > 0) {
67                 if (chunk > len)
68                         chunk = len;
69                 elt++;
70                 len -= chunk;
71                 if (len == 0)
72                         return elt;
73                 offset += chunk;
74         }
75 
76         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
77                 int end;
78 
79                 WARN_ON(start > offset + len);
80 
81                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
82                 chunk = end - offset;
83                 if (chunk > 0) {
84                         if (chunk > len)
85                                 chunk = len;
86                         elt++;
87                         len -= chunk;
88                         if (len == 0)
89                                 return elt;
90                         offset += chunk;
91                 }
92                 start = end;
93         }
94 
95         if (unlikely(skb_has_frag_list(skb))) {
96                 skb_walk_frags(skb, frag_iter) {
97                         int end, ret;
98 
99                         WARN_ON(start > offset + len);
100 
101                         end = start + frag_iter->len;
102                         chunk = end - offset;
103                         if (chunk > 0) {
104                                 if (chunk > len)
105                                         chunk = len;
106                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
107                                                 recursion_level + 1);
108                                 if (unlikely(ret < 0))
109                                         return ret;
110                                 elt += ret;
111                                 len -= chunk;
112                                 if (len == 0)
113                                         return elt;
114                                 offset += chunk;
115                         }
116                         start = end;
117                 }
118         }
119         BUG_ON(len);
120         return elt;
121 }
122 
123 /* Return the number of scatterlist elements required to completely map the
124  * skb, or -EMSGSIZE if the recursion depth is exceeded.
125  */
skb_nsg(struct sk_buff * skb,int offset,int len)126 static int skb_nsg(struct sk_buff *skb, int offset, int len)
127 {
128         return __skb_nsg(skb, offset, len, 0);
129 }
130 
padding_length(struct tls_sw_context_rx * ctx,struct tls_prot_info * prot,struct sk_buff * skb)131 static int padding_length(struct tls_sw_context_rx *ctx,
132 			  struct tls_prot_info *prot, struct sk_buff *skb)
133 {
134 	struct strp_msg *rxm = strp_msg(skb);
135 	int sub = 0;
136 
137 	/* Determine zero-padding length */
138 	if (prot->version == TLS_1_3_VERSION) {
139 		char content_type = 0;
140 		int err;
141 		int back = 17;
142 
143 		while (content_type == 0) {
144 			if (back > rxm->full_len - prot->prepend_size)
145 				return -EBADMSG;
146 			err = skb_copy_bits(skb,
147 					    rxm->offset + rxm->full_len - back,
148 					    &content_type, 1);
149 			if (err)
150 				return err;
151 			if (content_type)
152 				break;
153 			sub++;
154 			back++;
155 		}
156 		ctx->control = content_type;
157 	}
158 	return sub;
159 }
160 
tls_decrypt_done(struct crypto_async_request * req,int err)161 static void tls_decrypt_done(struct crypto_async_request *req, int err)
162 {
163 	struct aead_request *aead_req = (struct aead_request *)req;
164 	struct scatterlist *sgout = aead_req->dst;
165 	struct scatterlist *sgin = aead_req->src;
166 	struct tls_sw_context_rx *ctx;
167 	struct tls_context *tls_ctx;
168 	struct tls_prot_info *prot;
169 	struct scatterlist *sg;
170 	struct sk_buff *skb;
171 	unsigned int pages;
172 	int pending;
173 
174 	skb = (struct sk_buff *)req->data;
175 	tls_ctx = tls_get_ctx(skb->sk);
176 	ctx = tls_sw_ctx_rx(tls_ctx);
177 	prot = &tls_ctx->prot_info;
178 
179 	/* Propagate if there was an err */
180 	if (err) {
181 		if (err == -EBADMSG)
182 			TLS_INC_STATS(sock_net(skb->sk),
183 				      LINUX_MIB_TLSDECRYPTERROR);
184 		ctx->async_wait.err = err;
185 		tls_err_abort(skb->sk, err);
186 	} else {
187 		struct strp_msg *rxm = strp_msg(skb);
188 		int pad;
189 
190 		pad = padding_length(ctx, prot, skb);
191 		if (pad < 0) {
192 			ctx->async_wait.err = pad;
193 			tls_err_abort(skb->sk, pad);
194 		} else {
195 			rxm->full_len -= pad;
196 			rxm->offset += prot->prepend_size;
197 			rxm->full_len -= prot->overhead_size;
198 		}
199 	}
200 
201 	/* After using skb->sk to propagate sk through crypto async callback
202 	 * we need to NULL it again.
203 	 */
204 	skb->sk = NULL;
205 
206 
207 	/* Free the destination pages if skb was not decrypted inplace */
208 	if (sgout != sgin) {
209 		/* Skip the first S/G entry as it points to AAD */
210 		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
211 			if (!sg)
212 				break;
213 			put_page(sg_page(sg));
214 		}
215 	}
216 
217 	kfree(aead_req);
218 
219 	spin_lock_bh(&ctx->decrypt_compl_lock);
220 	pending = atomic_dec_return(&ctx->decrypt_pending);
221 
222 	if (!pending && ctx->async_notify)
223 		complete(&ctx->async_wait.completion);
224 	spin_unlock_bh(&ctx->decrypt_compl_lock);
225 }
226 
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,bool async)227 static int tls_do_decryption(struct sock *sk,
228 			     struct sk_buff *skb,
229 			     struct scatterlist *sgin,
230 			     struct scatterlist *sgout,
231 			     char *iv_recv,
232 			     size_t data_len,
233 			     struct aead_request *aead_req,
234 			     bool async)
235 {
236 	struct tls_context *tls_ctx = tls_get_ctx(sk);
237 	struct tls_prot_info *prot = &tls_ctx->prot_info;
238 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
239 	int ret;
240 
241 	aead_request_set_tfm(aead_req, ctx->aead_recv);
242 	aead_request_set_ad(aead_req, prot->aad_size);
243 	aead_request_set_crypt(aead_req, sgin, sgout,
244 			       data_len + prot->tag_size,
245 			       (u8 *)iv_recv);
246 
247 	if (async) {
248 		/* Using skb->sk to push sk through to crypto async callback
249 		 * handler. This allows propagating errors up to the socket
250 		 * if needed. It _must_ be cleared in the async handler
251 		 * before consume_skb is called. We _know_ skb->sk is NULL
252 		 * because it is a clone from strparser.
253 		 */
254 		skb->sk = sk;
255 		aead_request_set_callback(aead_req,
256 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
257 					  tls_decrypt_done, skb);
258 		atomic_inc(&ctx->decrypt_pending);
259 	} else {
260 		aead_request_set_callback(aead_req,
261 					  CRYPTO_TFM_REQ_MAY_BACKLOG,
262 					  crypto_req_done, &ctx->async_wait);
263 	}
264 
265 	ret = crypto_aead_decrypt(aead_req);
266 	if (ret == -EINPROGRESS) {
267 		if (async)
268 			return ret;
269 
270 		ret = crypto_wait_req(ret, &ctx->async_wait);
271 	}
272 
273 	if (async)
274 		atomic_dec(&ctx->decrypt_pending);
275 
276 	return ret;
277 }
278 
tls_trim_both_msgs(struct sock * sk,int target_size)279 static void tls_trim_both_msgs(struct sock *sk, int target_size)
280 {
281 	struct tls_context *tls_ctx = tls_get_ctx(sk);
282 	struct tls_prot_info *prot = &tls_ctx->prot_info;
283 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
284 	struct tls_rec *rec = ctx->open_rec;
285 
286 	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
287 	if (target_size > 0)
288 		target_size += prot->overhead_size;
289 	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
290 }
291 
tls_alloc_encrypted_msg(struct sock * sk,int len)292 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
293 {
294 	struct tls_context *tls_ctx = tls_get_ctx(sk);
295 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
296 	struct tls_rec *rec = ctx->open_rec;
297 	struct sk_msg *msg_en = &rec->msg_encrypted;
298 
299 	return sk_msg_alloc(sk, msg_en, len, 0);
300 }
301 
tls_clone_plaintext_msg(struct sock * sk,int required)302 static int tls_clone_plaintext_msg(struct sock *sk, int required)
303 {
304 	struct tls_context *tls_ctx = tls_get_ctx(sk);
305 	struct tls_prot_info *prot = &tls_ctx->prot_info;
306 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
307 	struct tls_rec *rec = ctx->open_rec;
308 	struct sk_msg *msg_pl = &rec->msg_plaintext;
309 	struct sk_msg *msg_en = &rec->msg_encrypted;
310 	int skip, len;
311 
312 	/* We add page references worth len bytes from encrypted sg
313 	 * at the end of plaintext sg. It is guaranteed that msg_en
314 	 * has enough required room (ensured by caller).
315 	 */
316 	len = required - msg_pl->sg.size;
317 
318 	/* Skip initial bytes in msg_en's data to be able to use
319 	 * same offset of both plain and encrypted data.
320 	 */
321 	skip = prot->prepend_size + msg_pl->sg.size;
322 
323 	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
324 }
325 
tls_get_rec(struct sock * sk)326 static struct tls_rec *tls_get_rec(struct sock *sk)
327 {
328 	struct tls_context *tls_ctx = tls_get_ctx(sk);
329 	struct tls_prot_info *prot = &tls_ctx->prot_info;
330 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
331 	struct sk_msg *msg_pl, *msg_en;
332 	struct tls_rec *rec;
333 	int mem_size;
334 
335 	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
336 
337 	rec = kzalloc(mem_size, sk->sk_allocation);
338 	if (!rec)
339 		return NULL;
340 
341 	msg_pl = &rec->msg_plaintext;
342 	msg_en = &rec->msg_encrypted;
343 
344 	sk_msg_init(msg_pl);
345 	sk_msg_init(msg_en);
346 
347 	sg_init_table(rec->sg_aead_in, 2);
348 	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
349 	sg_unmark_end(&rec->sg_aead_in[1]);
350 
351 	sg_init_table(rec->sg_aead_out, 2);
352 	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
353 	sg_unmark_end(&rec->sg_aead_out[1]);
354 
355 	return rec;
356 }
357 
tls_free_rec(struct sock * sk,struct tls_rec * rec)358 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
359 {
360 	sk_msg_free(sk, &rec->msg_encrypted);
361 	sk_msg_free(sk, &rec->msg_plaintext);
362 	kfree(rec);
363 }
364 
tls_free_open_rec(struct sock * sk)365 static void tls_free_open_rec(struct sock *sk)
366 {
367 	struct tls_context *tls_ctx = tls_get_ctx(sk);
368 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
369 	struct tls_rec *rec = ctx->open_rec;
370 
371 	if (rec) {
372 		tls_free_rec(sk, rec);
373 		ctx->open_rec = NULL;
374 	}
375 }
376 
tls_tx_records(struct sock * sk,int flags)377 int tls_tx_records(struct sock *sk, int flags)
378 {
379 	struct tls_context *tls_ctx = tls_get_ctx(sk);
380 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
381 	struct tls_rec *rec, *tmp;
382 	struct sk_msg *msg_en;
383 	int tx_flags, rc = 0;
384 
385 	if (tls_is_partially_sent_record(tls_ctx)) {
386 		rec = list_first_entry(&ctx->tx_list,
387 				       struct tls_rec, list);
388 
389 		if (flags == -1)
390 			tx_flags = rec->tx_flags;
391 		else
392 			tx_flags = flags;
393 
394 		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
395 		if (rc)
396 			goto tx_err;
397 
398 		/* Full record has been transmitted.
399 		 * Remove the head of tx_list
400 		 */
401 		list_del(&rec->list);
402 		sk_msg_free(sk, &rec->msg_plaintext);
403 		kfree(rec);
404 	}
405 
406 	/* Tx all ready records */
407 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
408 		if (READ_ONCE(rec->tx_ready)) {
409 			if (flags == -1)
410 				tx_flags = rec->tx_flags;
411 			else
412 				tx_flags = flags;
413 
414 			msg_en = &rec->msg_encrypted;
415 			rc = tls_push_sg(sk, tls_ctx,
416 					 &msg_en->sg.data[msg_en->sg.curr],
417 					 0, tx_flags);
418 			if (rc)
419 				goto tx_err;
420 
421 			list_del(&rec->list);
422 			sk_msg_free(sk, &rec->msg_plaintext);
423 			kfree(rec);
424 		} else {
425 			break;
426 		}
427 	}
428 
429 tx_err:
430 	if (rc < 0 && rc != -EAGAIN)
431 		tls_err_abort(sk, -EBADMSG);
432 
433 	return rc;
434 }
435 
tls_encrypt_done(struct crypto_async_request * req,int err)436 static void tls_encrypt_done(struct crypto_async_request *req, int err)
437 {
438 	struct aead_request *aead_req = (struct aead_request *)req;
439 	struct sock *sk = req->data;
440 	struct tls_context *tls_ctx = tls_get_ctx(sk);
441 	struct tls_prot_info *prot = &tls_ctx->prot_info;
442 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
443 	struct scatterlist *sge;
444 	struct sk_msg *msg_en;
445 	struct tls_rec *rec;
446 	bool ready = false;
447 	int pending;
448 
449 	rec = container_of(aead_req, struct tls_rec, aead_req);
450 	msg_en = &rec->msg_encrypted;
451 
452 	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
453 	sge->offset -= prot->prepend_size;
454 	sge->length += prot->prepend_size;
455 
456 	/* Check if error is previously set on socket */
457 	if (err || sk->sk_err) {
458 		rec = NULL;
459 
460 		/* If err is already set on socket, return the same code */
461 		if (sk->sk_err) {
462 			ctx->async_wait.err = -sk->sk_err;
463 		} else {
464 			ctx->async_wait.err = err;
465 			tls_err_abort(sk, err);
466 		}
467 	}
468 
469 	if (rec) {
470 		struct tls_rec *first_rec;
471 
472 		/* Mark the record as ready for transmission */
473 		smp_store_mb(rec->tx_ready, true);
474 
475 		/* If received record is at head of tx_list, schedule tx */
476 		first_rec = list_first_entry(&ctx->tx_list,
477 					     struct tls_rec, list);
478 		if (rec == first_rec)
479 			ready = true;
480 	}
481 
482 	spin_lock_bh(&ctx->encrypt_compl_lock);
483 	pending = atomic_dec_return(&ctx->encrypt_pending);
484 
485 	if (!pending && ctx->async_notify)
486 		complete(&ctx->async_wait.completion);
487 	spin_unlock_bh(&ctx->encrypt_compl_lock);
488 
489 	if (!ready)
490 		return;
491 
492 	/* Schedule the transmission */
493 	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
494 		schedule_delayed_work(&ctx->tx_work.work, 1);
495 }
496 
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)497 static int tls_do_encryption(struct sock *sk,
498 			     struct tls_context *tls_ctx,
499 			     struct tls_sw_context_tx *ctx,
500 			     struct aead_request *aead_req,
501 			     size_t data_len, u32 start)
502 {
503 	struct tls_prot_info *prot = &tls_ctx->prot_info;
504 	struct tls_rec *rec = ctx->open_rec;
505 	struct sk_msg *msg_en = &rec->msg_encrypted;
506 	struct scatterlist *sge = sk_msg_elem(msg_en, start);
507 	int rc, iv_offset = 0;
508 
509 	/* For CCM based ciphers, first byte of IV is a constant */
510 	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
511 		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
512 		iv_offset = 1;
513 	}
514 
515 	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
516 	       prot->iv_size + prot->salt_size);
517 
518 	xor_iv_with_seq(prot->version, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq);
519 
520 	sge->offset += prot->prepend_size;
521 	sge->length -= prot->prepend_size;
522 
523 	msg_en->sg.curr = start;
524 
525 	aead_request_set_tfm(aead_req, ctx->aead_send);
526 	aead_request_set_ad(aead_req, prot->aad_size);
527 	aead_request_set_crypt(aead_req, rec->sg_aead_in,
528 			       rec->sg_aead_out,
529 			       data_len, rec->iv_data);
530 
531 	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
532 				  tls_encrypt_done, sk);
533 
534 	/* Add the record in tx_list */
535 	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
536 	atomic_inc(&ctx->encrypt_pending);
537 
538 	rc = crypto_aead_encrypt(aead_req);
539 	if (!rc || rc != -EINPROGRESS) {
540 		atomic_dec(&ctx->encrypt_pending);
541 		sge->offset -= prot->prepend_size;
542 		sge->length += prot->prepend_size;
543 	}
544 
545 	if (!rc) {
546 		WRITE_ONCE(rec->tx_ready, true);
547 	} else if (rc != -EINPROGRESS) {
548 		list_del(&rec->list);
549 		return rc;
550 	}
551 
552 	/* Unhook the record from context if encryption is not failure */
553 	ctx->open_rec = NULL;
554 	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
555 	return rc;
556 }
557 
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)558 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
559 				 struct tls_rec **to, struct sk_msg *msg_opl,
560 				 struct sk_msg *msg_oen, u32 split_point,
561 				 u32 tx_overhead_size, u32 *orig_end)
562 {
563 	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
564 	struct scatterlist *sge, *osge, *nsge;
565 	u32 orig_size = msg_opl->sg.size;
566 	struct scatterlist tmp = { };
567 	struct sk_msg *msg_npl;
568 	struct tls_rec *new;
569 	int ret;
570 
571 	new = tls_get_rec(sk);
572 	if (!new)
573 		return -ENOMEM;
574 	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
575 			   tx_overhead_size, 0);
576 	if (ret < 0) {
577 		tls_free_rec(sk, new);
578 		return ret;
579 	}
580 
581 	*orig_end = msg_opl->sg.end;
582 	i = msg_opl->sg.start;
583 	sge = sk_msg_elem(msg_opl, i);
584 	while (apply && sge->length) {
585 		if (sge->length > apply) {
586 			u32 len = sge->length - apply;
587 
588 			get_page(sg_page(sge));
589 			sg_set_page(&tmp, sg_page(sge), len,
590 				    sge->offset + apply);
591 			sge->length = apply;
592 			bytes += apply;
593 			apply = 0;
594 		} else {
595 			apply -= sge->length;
596 			bytes += sge->length;
597 		}
598 
599 		sk_msg_iter_var_next(i);
600 		if (i == msg_opl->sg.end)
601 			break;
602 		sge = sk_msg_elem(msg_opl, i);
603 	}
604 
605 	msg_opl->sg.end = i;
606 	msg_opl->sg.curr = i;
607 	msg_opl->sg.copybreak = 0;
608 	msg_opl->apply_bytes = 0;
609 	msg_opl->sg.size = bytes;
610 
611 	msg_npl = &new->msg_plaintext;
612 	msg_npl->apply_bytes = apply;
613 	msg_npl->sg.size = orig_size - bytes;
614 
615 	j = msg_npl->sg.start;
616 	nsge = sk_msg_elem(msg_npl, j);
617 	if (tmp.length) {
618 		memcpy(nsge, &tmp, sizeof(*nsge));
619 		sk_msg_iter_var_next(j);
620 		nsge = sk_msg_elem(msg_npl, j);
621 	}
622 
623 	osge = sk_msg_elem(msg_opl, i);
624 	while (osge->length) {
625 		memcpy(nsge, osge, sizeof(*nsge));
626 		sg_unmark_end(nsge);
627 		sk_msg_iter_var_next(i);
628 		sk_msg_iter_var_next(j);
629 		if (i == *orig_end)
630 			break;
631 		osge = sk_msg_elem(msg_opl, i);
632 		nsge = sk_msg_elem(msg_npl, j);
633 	}
634 
635 	msg_npl->sg.end = j;
636 	msg_npl->sg.curr = j;
637 	msg_npl->sg.copybreak = 0;
638 
639 	*to = new;
640 	return 0;
641 }
642 
tls_merge_open_record(struct sock * sk,struct tls_rec * to,struct tls_rec * from,u32 orig_end)643 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
644 				  struct tls_rec *from, u32 orig_end)
645 {
646 	struct sk_msg *msg_npl = &from->msg_plaintext;
647 	struct sk_msg *msg_opl = &to->msg_plaintext;
648 	struct scatterlist *osge, *nsge;
649 	u32 i, j;
650 
651 	i = msg_opl->sg.end;
652 	sk_msg_iter_var_prev(i);
653 	j = msg_npl->sg.start;
654 
655 	osge = sk_msg_elem(msg_opl, i);
656 	nsge = sk_msg_elem(msg_npl, j);
657 
658 	if (sg_page(osge) == sg_page(nsge) &&
659 	    osge->offset + osge->length == nsge->offset) {
660 		osge->length += nsge->length;
661 		put_page(sg_page(nsge));
662 	}
663 
664 	msg_opl->sg.end = orig_end;
665 	msg_opl->sg.curr = orig_end;
666 	msg_opl->sg.copybreak = 0;
667 	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
668 	msg_opl->sg.size += msg_npl->sg.size;
669 
670 	sk_msg_free(sk, &to->msg_encrypted);
671 	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
672 
673 	kfree(from);
674 }
675 
tls_push_record(struct sock * sk,int flags,unsigned char record_type)676 static int tls_push_record(struct sock *sk, int flags,
677 			   unsigned char record_type)
678 {
679 	struct tls_context *tls_ctx = tls_get_ctx(sk);
680 	struct tls_prot_info *prot = &tls_ctx->prot_info;
681 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
682 	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
683 	u32 i, split_point, orig_end;
684 	struct sk_msg *msg_pl, *msg_en;
685 	struct aead_request *req;
686 	bool split;
687 	int rc;
688 
689 	if (!rec)
690 		return 0;
691 
692 	msg_pl = &rec->msg_plaintext;
693 	msg_en = &rec->msg_encrypted;
694 
695 	split_point = msg_pl->apply_bytes;
696 	split = split_point && split_point < msg_pl->sg.size;
697 	if (unlikely((!split &&
698 		      msg_pl->sg.size +
699 		      prot->overhead_size > msg_en->sg.size) ||
700 		     (split &&
701 		      split_point +
702 		      prot->overhead_size > msg_en->sg.size))) {
703 		split = true;
704 		split_point = msg_en->sg.size;
705 	}
706 	if (split) {
707 		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
708 					   split_point, prot->overhead_size,
709 					   &orig_end);
710 		if (rc < 0)
711 			return rc;
712 		/* This can happen if above tls_split_open_record allocates
713 		 * a single large encryption buffer instead of two smaller
714 		 * ones. In this case adjust pointers and continue without
715 		 * split.
716 		 */
717 		if (!msg_pl->sg.size) {
718 			tls_merge_open_record(sk, rec, tmp, orig_end);
719 			msg_pl = &rec->msg_plaintext;
720 			msg_en = &rec->msg_encrypted;
721 			split = false;
722 		}
723 		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
724 			    prot->overhead_size);
725 	}
726 
727 	rec->tx_flags = flags;
728 	req = &rec->aead_req;
729 
730 	i = msg_pl->sg.end;
731 	sk_msg_iter_var_prev(i);
732 
733 	rec->content_type = record_type;
734 	if (prot->version == TLS_1_3_VERSION) {
735 		/* Add content type to end of message.  No padding added */
736 		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
737 		sg_mark_end(&rec->sg_content_type);
738 		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
739 			 &rec->sg_content_type);
740 	} else {
741 		sg_mark_end(sk_msg_elem(msg_pl, i));
742 	}
743 
744 	if (msg_pl->sg.end < msg_pl->sg.start) {
745 		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
746 			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
747 			 msg_pl->sg.data);
748 	}
749 
750 	i = msg_pl->sg.start;
751 	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
752 
753 	i = msg_en->sg.end;
754 	sk_msg_iter_var_prev(i);
755 	sg_mark_end(sk_msg_elem(msg_en, i));
756 
757 	i = msg_en->sg.start;
758 	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
759 
760 	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
761 		     tls_ctx->tx.rec_seq, prot->rec_seq_size,
762 		     record_type, prot->version);
763 
764 	tls_fill_prepend(tls_ctx,
765 			 page_address(sg_page(&msg_en->sg.data[i])) +
766 			 msg_en->sg.data[i].offset,
767 			 msg_pl->sg.size + prot->tail_size,
768 			 record_type, prot->version);
769 
770 	tls_ctx->pending_open_record_frags = false;
771 
772 	rc = tls_do_encryption(sk, tls_ctx, ctx, req,
773 			       msg_pl->sg.size + prot->tail_size, i);
774 	if (rc < 0) {
775 		if (rc != -EINPROGRESS) {
776 			tls_err_abort(sk, -EBADMSG);
777 			if (split) {
778 				tls_ctx->pending_open_record_frags = true;
779 				tls_merge_open_record(sk, rec, tmp, orig_end);
780 			}
781 		}
782 		ctx->async_capable = 1;
783 		return rc;
784 	} else if (split) {
785 		msg_pl = &tmp->msg_plaintext;
786 		msg_en = &tmp->msg_encrypted;
787 		sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
788 		tls_ctx->pending_open_record_frags = true;
789 		ctx->open_rec = tmp;
790 	}
791 
792 	return tls_tx_records(sk, flags);
793 }
794 
bpf_exec_tx_verdict(struct sk_msg * msg,struct sock * sk,bool full_record,u8 record_type,ssize_t * copied,int flags)795 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
796 			       bool full_record, u8 record_type,
797 			       ssize_t *copied, int flags)
798 {
799 	struct tls_context *tls_ctx = tls_get_ctx(sk);
800 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
801 	struct sk_msg msg_redir = { };
802 	struct sk_psock *psock;
803 	struct sock *sk_redir;
804 	struct tls_rec *rec;
805 	bool enospc, policy;
806 	int err = 0, send;
807 	u32 delta = 0;
808 
809 	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
810 	psock = sk_psock_get(sk);
811 	if (!psock || !policy) {
812 		err = tls_push_record(sk, flags, record_type);
813 		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
814 			*copied -= sk_msg_free(sk, msg);
815 			tls_free_open_rec(sk);
816 			err = -sk->sk_err;
817 		}
818 		if (psock)
819 			sk_psock_put(sk, psock);
820 		return err;
821 	}
822 more_data:
823 	enospc = sk_msg_full(msg);
824 	if (psock->eval == __SK_NONE) {
825 		delta = msg->sg.size;
826 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
827 		delta -= msg->sg.size;
828 	}
829 	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
830 	    !enospc && !full_record) {
831 		err = -ENOSPC;
832 		goto out_err;
833 	}
834 	msg->cork_bytes = 0;
835 	send = msg->sg.size;
836 	if (msg->apply_bytes && msg->apply_bytes < send)
837 		send = msg->apply_bytes;
838 
839 	switch (psock->eval) {
840 	case __SK_PASS:
841 		err = tls_push_record(sk, flags, record_type);
842 		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
843 			*copied -= sk_msg_free(sk, msg);
844 			tls_free_open_rec(sk);
845 			err = -sk->sk_err;
846 			goto out_err;
847 		}
848 		break;
849 	case __SK_REDIRECT:
850 		sk_redir = psock->sk_redir;
851 		memcpy(&msg_redir, msg, sizeof(*msg));
852 		if (msg->apply_bytes < send)
853 			msg->apply_bytes = 0;
854 		else
855 			msg->apply_bytes -= send;
856 		sk_msg_return_zero(sk, msg, send);
857 		msg->sg.size -= send;
858 		release_sock(sk);
859 		err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
860 		lock_sock(sk);
861 		if (err < 0) {
862 			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
863 			msg->sg.size = 0;
864 		}
865 		if (msg->sg.size == 0)
866 			tls_free_open_rec(sk);
867 		break;
868 	case __SK_DROP:
869 	default:
870 		sk_msg_free_partial(sk, msg, send);
871 		if (msg->apply_bytes < send)
872 			msg->apply_bytes = 0;
873 		else
874 			msg->apply_bytes -= send;
875 		if (msg->sg.size == 0)
876 			tls_free_open_rec(sk);
877 		*copied -= (send + delta);
878 		err = -EACCES;
879 	}
880 
881 	if (likely(!err)) {
882 		bool reset_eval = !ctx->open_rec;
883 
884 		rec = ctx->open_rec;
885 		if (rec) {
886 			msg = &rec->msg_plaintext;
887 			if (!msg->apply_bytes)
888 				reset_eval = true;
889 		}
890 		if (reset_eval) {
891 			psock->eval = __SK_NONE;
892 			if (psock->sk_redir) {
893 				sock_put(psock->sk_redir);
894 				psock->sk_redir = NULL;
895 			}
896 		}
897 		if (rec)
898 			goto more_data;
899 	}
900  out_err:
901 	sk_psock_put(sk, psock);
902 	return err;
903 }
904 
tls_sw_push_pending_record(struct sock * sk,int flags)905 static int tls_sw_push_pending_record(struct sock *sk, int flags)
906 {
907 	struct tls_context *tls_ctx = tls_get_ctx(sk);
908 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
909 	struct tls_rec *rec = ctx->open_rec;
910 	struct sk_msg *msg_pl;
911 	size_t copied;
912 
913 	if (!rec)
914 		return 0;
915 
916 	msg_pl = &rec->msg_plaintext;
917 	copied = msg_pl->sg.size;
918 	if (!copied)
919 		return 0;
920 
921 	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
922 				   &copied, flags);
923 }
924 
tls_sw_sendmsg(struct sock * sk,struct msghdr * msg,size_t size)925 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
926 {
927 	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
928 	struct tls_context *tls_ctx = tls_get_ctx(sk);
929 	struct tls_prot_info *prot = &tls_ctx->prot_info;
930 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
931 	bool async_capable = ctx->async_capable;
932 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
933 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
934 	bool eor = !(msg->msg_flags & MSG_MORE);
935 	size_t try_to_copy;
936 	ssize_t copied = 0;
937 	struct sk_msg *msg_pl, *msg_en;
938 	struct tls_rec *rec;
939 	int required_size;
940 	int num_async = 0;
941 	bool full_record;
942 	int record_room;
943 	int num_zc = 0;
944 	int orig_size;
945 	int ret = 0;
946 	int pending;
947 
948 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
949 			       MSG_CMSG_COMPAT))
950 		return -EOPNOTSUPP;
951 
952 	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
953 	if (ret)
954 		return ret;
955 	lock_sock(sk);
956 
957 	if (unlikely(msg->msg_controllen)) {
958 		ret = tls_proccess_cmsg(sk, msg, &record_type);
959 		if (ret) {
960 			if (ret == -EINPROGRESS)
961 				num_async++;
962 			else if (ret != -EAGAIN)
963 				goto send_end;
964 		}
965 	}
966 
967 	while (msg_data_left(msg)) {
968 		if (sk->sk_err) {
969 			ret = -sk->sk_err;
970 			goto send_end;
971 		}
972 
973 		if (ctx->open_rec)
974 			rec = ctx->open_rec;
975 		else
976 			rec = ctx->open_rec = tls_get_rec(sk);
977 		if (!rec) {
978 			ret = -ENOMEM;
979 			goto send_end;
980 		}
981 
982 		msg_pl = &rec->msg_plaintext;
983 		msg_en = &rec->msg_encrypted;
984 
985 		orig_size = msg_pl->sg.size;
986 		full_record = false;
987 		try_to_copy = msg_data_left(msg);
988 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
989 		if (try_to_copy >= record_room) {
990 			try_to_copy = record_room;
991 			full_record = true;
992 		}
993 
994 		required_size = msg_pl->sg.size + try_to_copy +
995 				prot->overhead_size;
996 
997 		if (!sk_stream_memory_free(sk))
998 			goto wait_for_sndbuf;
999 
1000 alloc_encrypted:
1001 		ret = tls_alloc_encrypted_msg(sk, required_size);
1002 		if (ret) {
1003 			if (ret != -ENOSPC)
1004 				goto wait_for_memory;
1005 
1006 			/* Adjust try_to_copy according to the amount that was
1007 			 * actually allocated. The difference is due
1008 			 * to max sg elements limit
1009 			 */
1010 			try_to_copy -= required_size - msg_en->sg.size;
1011 			full_record = true;
1012 		}
1013 
1014 		if (!is_kvec && (full_record || eor) && !async_capable) {
1015 			u32 first = msg_pl->sg.end;
1016 
1017 			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1018 							msg_pl, try_to_copy);
1019 			if (ret)
1020 				goto fallback_to_reg_send;
1021 
1022 			num_zc++;
1023 			copied += try_to_copy;
1024 
1025 			sk_msg_sg_copy_set(msg_pl, first);
1026 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1027 						  record_type, &copied,
1028 						  msg->msg_flags);
1029 			if (ret) {
1030 				if (ret == -EINPROGRESS)
1031 					num_async++;
1032 				else if (ret == -ENOMEM)
1033 					goto wait_for_memory;
1034 				else if (ctx->open_rec && ret == -ENOSPC)
1035 					goto rollback_iter;
1036 				else if (ret != -EAGAIN)
1037 					goto send_end;
1038 			}
1039 			continue;
1040 rollback_iter:
1041 			copied -= try_to_copy;
1042 			sk_msg_sg_copy_clear(msg_pl, first);
1043 			iov_iter_revert(&msg->msg_iter,
1044 					msg_pl->sg.size - orig_size);
1045 fallback_to_reg_send:
1046 			sk_msg_trim(sk, msg_pl, orig_size);
1047 		}
1048 
1049 		required_size = msg_pl->sg.size + try_to_copy;
1050 
1051 		ret = tls_clone_plaintext_msg(sk, required_size);
1052 		if (ret) {
1053 			if (ret != -ENOSPC)
1054 				goto send_end;
1055 
1056 			/* Adjust try_to_copy according to the amount that was
1057 			 * actually allocated. The difference is due
1058 			 * to max sg elements limit
1059 			 */
1060 			try_to_copy -= required_size - msg_pl->sg.size;
1061 			full_record = true;
1062 			sk_msg_trim(sk, msg_en,
1063 				    msg_pl->sg.size + prot->overhead_size);
1064 		}
1065 
1066 		if (try_to_copy) {
1067 			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1068 						       msg_pl, try_to_copy);
1069 			if (ret < 0)
1070 				goto trim_sgl;
1071 		}
1072 
1073 		/* Open records defined only if successfully copied, otherwise
1074 		 * we would trim the sg but not reset the open record frags.
1075 		 */
1076 		tls_ctx->pending_open_record_frags = true;
1077 		copied += try_to_copy;
1078 		if (full_record || eor) {
1079 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1080 						  record_type, &copied,
1081 						  msg->msg_flags);
1082 			if (ret) {
1083 				if (ret == -EINPROGRESS)
1084 					num_async++;
1085 				else if (ret == -ENOMEM)
1086 					goto wait_for_memory;
1087 				else if (ret != -EAGAIN) {
1088 					if (ret == -ENOSPC)
1089 						ret = 0;
1090 					goto send_end;
1091 				}
1092 			}
1093 		}
1094 
1095 		continue;
1096 
1097 wait_for_sndbuf:
1098 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1099 wait_for_memory:
1100 		ret = sk_stream_wait_memory(sk, &timeo);
1101 		if (ret) {
1102 trim_sgl:
1103 			if (ctx->open_rec)
1104 				tls_trim_both_msgs(sk, orig_size);
1105 			goto send_end;
1106 		}
1107 
1108 		if (ctx->open_rec && msg_en->sg.size < required_size)
1109 			goto alloc_encrypted;
1110 	}
1111 
1112 	if (!num_async) {
1113 		goto send_end;
1114 	} else if (num_zc) {
1115 		/* Wait for pending encryptions to get completed */
1116 		spin_lock_bh(&ctx->encrypt_compl_lock);
1117 		ctx->async_notify = true;
1118 
1119 		pending = atomic_read(&ctx->encrypt_pending);
1120 		spin_unlock_bh(&ctx->encrypt_compl_lock);
1121 		if (pending)
1122 			crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1123 		else
1124 			reinit_completion(&ctx->async_wait.completion);
1125 
1126 		/* There can be no concurrent accesses, since we have no
1127 		 * pending encrypt operations
1128 		 */
1129 		WRITE_ONCE(ctx->async_notify, false);
1130 
1131 		if (ctx->async_wait.err) {
1132 			ret = ctx->async_wait.err;
1133 			copied = 0;
1134 		}
1135 	}
1136 
1137 	/* Transmit if any encryptions have completed */
1138 	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1139 		cancel_delayed_work(&ctx->tx_work.work);
1140 		tls_tx_records(sk, msg->msg_flags);
1141 	}
1142 
1143 send_end:
1144 	ret = sk_stream_error(sk, msg->msg_flags, ret);
1145 
1146 	release_sock(sk);
1147 	mutex_unlock(&tls_ctx->tx_lock);
1148 	return copied > 0 ? copied : ret;
1149 }
1150 
tls_sw_do_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)1151 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1152 			      int offset, size_t size, int flags)
1153 {
1154 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1155 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1156 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1157 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1158 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1159 	struct sk_msg *msg_pl;
1160 	struct tls_rec *rec;
1161 	int num_async = 0;
1162 	ssize_t copied = 0;
1163 	bool full_record;
1164 	int record_room;
1165 	int ret = 0;
1166 	bool eor;
1167 
1168 	eor = !(flags & MSG_SENDPAGE_NOTLAST);
1169 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1170 
1171 	/* Call the sk_stream functions to manage the sndbuf mem. */
1172 	while (size > 0) {
1173 		size_t copy, required_size;
1174 
1175 		if (sk->sk_err) {
1176 			ret = -sk->sk_err;
1177 			goto sendpage_end;
1178 		}
1179 
1180 		if (ctx->open_rec)
1181 			rec = ctx->open_rec;
1182 		else
1183 			rec = ctx->open_rec = tls_get_rec(sk);
1184 		if (!rec) {
1185 			ret = -ENOMEM;
1186 			goto sendpage_end;
1187 		}
1188 
1189 		msg_pl = &rec->msg_plaintext;
1190 
1191 		full_record = false;
1192 		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1193 		copy = size;
1194 		if (copy >= record_room) {
1195 			copy = record_room;
1196 			full_record = true;
1197 		}
1198 
1199 		required_size = msg_pl->sg.size + copy + prot->overhead_size;
1200 
1201 		if (!sk_stream_memory_free(sk))
1202 			goto wait_for_sndbuf;
1203 alloc_payload:
1204 		ret = tls_alloc_encrypted_msg(sk, required_size);
1205 		if (ret) {
1206 			if (ret != -ENOSPC)
1207 				goto wait_for_memory;
1208 
1209 			/* Adjust copy according to the amount that was
1210 			 * actually allocated. The difference is due
1211 			 * to max sg elements limit
1212 			 */
1213 			copy -= required_size - msg_pl->sg.size;
1214 			full_record = true;
1215 		}
1216 
1217 		sk_msg_page_add(msg_pl, page, copy, offset);
1218 		msg_pl->sg.copybreak = 0;
1219 		msg_pl->sg.curr = msg_pl->sg.end;
1220 		sk_mem_charge(sk, copy);
1221 
1222 		offset += copy;
1223 		size -= copy;
1224 		copied += copy;
1225 
1226 		tls_ctx->pending_open_record_frags = true;
1227 		if (full_record || eor || sk_msg_full(msg_pl)) {
1228 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1229 						  record_type, &copied, flags);
1230 			if (ret) {
1231 				if (ret == -EINPROGRESS)
1232 					num_async++;
1233 				else if (ret == -ENOMEM)
1234 					goto wait_for_memory;
1235 				else if (ret != -EAGAIN) {
1236 					if (ret == -ENOSPC)
1237 						ret = 0;
1238 					goto sendpage_end;
1239 				}
1240 			}
1241 		}
1242 		continue;
1243 wait_for_sndbuf:
1244 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1245 wait_for_memory:
1246 		ret = sk_stream_wait_memory(sk, &timeo);
1247 		if (ret) {
1248 			if (ctx->open_rec)
1249 				tls_trim_both_msgs(sk, msg_pl->sg.size);
1250 			goto sendpage_end;
1251 		}
1252 
1253 		if (ctx->open_rec)
1254 			goto alloc_payload;
1255 	}
1256 
1257 	if (num_async) {
1258 		/* Transmit if any encryptions have completed */
1259 		if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1260 			cancel_delayed_work(&ctx->tx_work.work);
1261 			tls_tx_records(sk, flags);
1262 		}
1263 	}
1264 sendpage_end:
1265 	ret = sk_stream_error(sk, flags, ret);
1266 	return copied > 0 ? copied : ret;
1267 }
1268 
tls_sw_sendpage_locked(struct sock * sk,struct page * page,int offset,size_t size,int flags)1269 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1270 			   int offset, size_t size, int flags)
1271 {
1272 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1274 		      MSG_NO_SHARED_FRAGS))
1275 		return -EOPNOTSUPP;
1276 
1277 	return tls_sw_do_sendpage(sk, page, offset, size, flags);
1278 }
1279 
tls_sw_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)1280 int tls_sw_sendpage(struct sock *sk, struct page *page,
1281 		    int offset, size_t size, int flags)
1282 {
1283 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1284 	int ret;
1285 
1286 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1287 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1288 		return -EOPNOTSUPP;
1289 
1290 	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1291 	if (ret)
1292 		return ret;
1293 	lock_sock(sk);
1294 	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1295 	release_sock(sk);
1296 	mutex_unlock(&tls_ctx->tx_lock);
1297 	return ret;
1298 }
1299 
tls_wait_data(struct sock * sk,struct sk_psock * psock,bool nonblock,long timeo,int * err)1300 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1301 				     bool nonblock, long timeo, int *err)
1302 {
1303 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1304 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1305 	struct sk_buff *skb;
1306 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1307 
1308 	while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1309 		if (sk->sk_err) {
1310 			*err = sock_error(sk);
1311 			return NULL;
1312 		}
1313 
1314 		if (!skb_queue_empty(&sk->sk_receive_queue)) {
1315 			__strp_unpause(&ctx->strp);
1316 			if (ctx->recv_pkt)
1317 				return ctx->recv_pkt;
1318 		}
1319 
1320 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1321 			return NULL;
1322 
1323 		if (sock_flag(sk, SOCK_DONE))
1324 			return NULL;
1325 
1326 		if (nonblock || !timeo) {
1327 			*err = -EAGAIN;
1328 			return NULL;
1329 		}
1330 
1331 		add_wait_queue(sk_sleep(sk), &wait);
1332 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1333 		sk_wait_event(sk, &timeo,
1334 			      ctx->recv_pkt != skb ||
1335 			      !sk_psock_queue_empty(psock),
1336 			      &wait);
1337 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1338 		remove_wait_queue(sk_sleep(sk), &wait);
1339 
1340 		/* Handle signals */
1341 		if (signal_pending(current)) {
1342 			*err = sock_intr_errno(timeo);
1343 			return NULL;
1344 		}
1345 	}
1346 
1347 	return skb;
1348 }
1349 
tls_setup_from_iter(struct sock * sk,struct iov_iter * from,int length,int * pages_used,unsigned int * size_used,struct scatterlist * to,int to_max_pages)1350 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1351 			       int length, int *pages_used,
1352 			       unsigned int *size_used,
1353 			       struct scatterlist *to,
1354 			       int to_max_pages)
1355 {
1356 	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1357 	struct page *pages[MAX_SKB_FRAGS];
1358 	unsigned int size = *size_used;
1359 	ssize_t copied, use;
1360 	size_t offset;
1361 
1362 	while (length > 0) {
1363 		i = 0;
1364 		maxpages = to_max_pages - num_elem;
1365 		if (maxpages == 0) {
1366 			rc = -EFAULT;
1367 			goto out;
1368 		}
1369 		copied = iov_iter_get_pages(from, pages,
1370 					    length,
1371 					    maxpages, &offset);
1372 		if (copied <= 0) {
1373 			rc = -EFAULT;
1374 			goto out;
1375 		}
1376 
1377 		iov_iter_advance(from, copied);
1378 
1379 		length -= copied;
1380 		size += copied;
1381 		while (copied) {
1382 			use = min_t(int, copied, PAGE_SIZE - offset);
1383 
1384 			sg_set_page(&to[num_elem],
1385 				    pages[i], use, offset);
1386 			sg_unmark_end(&to[num_elem]);
1387 			/* We do not uncharge memory from this API */
1388 
1389 			offset = 0;
1390 			copied -= use;
1391 
1392 			i++;
1393 			num_elem++;
1394 		}
1395 	}
1396 	/* Mark the end in the last sg entry if newly added */
1397 	if (num_elem > *pages_used)
1398 		sg_mark_end(&to[num_elem - 1]);
1399 out:
1400 	if (rc)
1401 		iov_iter_revert(from, size - *size_used);
1402 	*size_used = size;
1403 	*pages_used = num_elem;
1404 
1405 	return rc;
1406 }
1407 
1408 /* This function decrypts the input skb into either out_iov or in out_sg
1409  * or in skb buffers itself. The input parameter 'zc' indicates if
1410  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1411  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1412  * NULL, then the decryption happens inside skb buffers itself, i.e.
1413  * zero-copy gets disabled and 'zc' is updated.
1414  */
1415 
decrypt_internal(struct sock * sk,struct sk_buff * skb,struct iov_iter * out_iov,struct scatterlist * out_sg,int * chunk,bool * zc,bool async)1416 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1417 			    struct iov_iter *out_iov,
1418 			    struct scatterlist *out_sg,
1419 			    int *chunk, bool *zc, bool async)
1420 {
1421 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1422 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1423 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1424 	struct strp_msg *rxm = strp_msg(skb);
1425 	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1426 	struct aead_request *aead_req;
1427 	struct sk_buff *unused;
1428 	u8 *aad, *iv, *mem = NULL;
1429 	struct scatterlist *sgin = NULL;
1430 	struct scatterlist *sgout = NULL;
1431 	const int data_len = rxm->full_len - prot->overhead_size +
1432 			     prot->tail_size;
1433 	int iv_offset = 0;
1434 
1435 	if (*zc && (out_iov || out_sg)) {
1436 		if (out_iov)
1437 			n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1438 		else
1439 			n_sgout = sg_nents(out_sg);
1440 		n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1441 				 rxm->full_len - prot->prepend_size);
1442 	} else {
1443 		n_sgout = 0;
1444 		*zc = false;
1445 		n_sgin = skb_cow_data(skb, 0, &unused);
1446 	}
1447 
1448 	if (n_sgin < 1)
1449 		return -EBADMSG;
1450 
1451 	/* Increment to accommodate AAD */
1452 	n_sgin = n_sgin + 1;
1453 
1454 	nsg = n_sgin + n_sgout;
1455 
1456 	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1457 	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1458 	mem_size = mem_size + prot->aad_size;
1459 	mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1460 
1461 	/* Allocate a single block of memory which contains
1462 	 * aead_req || sgin[] || sgout[] || aad || iv.
1463 	 * This order achieves correct alignment for aead_req, sgin, sgout.
1464 	 */
1465 	mem = kmalloc(mem_size, sk->sk_allocation);
1466 	if (!mem)
1467 		return -ENOMEM;
1468 
1469 	/* Segment the allocated memory */
1470 	aead_req = (struct aead_request *)mem;
1471 	sgin = (struct scatterlist *)(mem + aead_size);
1472 	sgout = sgin + n_sgin;
1473 	aad = (u8 *)(sgout + n_sgout);
1474 	iv = aad + prot->aad_size;
1475 
1476 	/* For CCM based ciphers, first byte of nonce+iv is always '2' */
1477 	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1478 		iv[0] = 2;
1479 		iv_offset = 1;
1480 	}
1481 
1482 	/* Prepare IV */
1483 	err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1484 			    iv + iv_offset + prot->salt_size,
1485 			    prot->iv_size);
1486 	if (err < 0) {
1487 		kfree(mem);
1488 		return err;
1489 	}
1490 	if (prot->version == TLS_1_3_VERSION)
1491 		memcpy(iv + iv_offset, tls_ctx->rx.iv,
1492 		       prot->iv_size + prot->salt_size);
1493 	else
1494 		memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1495 
1496 	xor_iv_with_seq(prot->version, iv + iv_offset, tls_ctx->rx.rec_seq);
1497 
1498 	/* Prepare AAD */
1499 	tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1500 		     prot->tail_size,
1501 		     tls_ctx->rx.rec_seq, prot->rec_seq_size,
1502 		     ctx->control, prot->version);
1503 
1504 	/* Prepare sgin */
1505 	sg_init_table(sgin, n_sgin);
1506 	sg_set_buf(&sgin[0], aad, prot->aad_size);
1507 	err = skb_to_sgvec(skb, &sgin[1],
1508 			   rxm->offset + prot->prepend_size,
1509 			   rxm->full_len - prot->prepend_size);
1510 	if (err < 0) {
1511 		kfree(mem);
1512 		return err;
1513 	}
1514 
1515 	if (n_sgout) {
1516 		if (out_iov) {
1517 			sg_init_table(sgout, n_sgout);
1518 			sg_set_buf(&sgout[0], aad, prot->aad_size);
1519 
1520 			*chunk = 0;
1521 			err = tls_setup_from_iter(sk, out_iov, data_len,
1522 						  &pages, chunk, &sgout[1],
1523 						  (n_sgout - 1));
1524 			if (err < 0)
1525 				goto fallback_to_reg_recv;
1526 		} else if (out_sg) {
1527 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1528 		} else {
1529 			goto fallback_to_reg_recv;
1530 		}
1531 	} else {
1532 fallback_to_reg_recv:
1533 		sgout = sgin;
1534 		pages = 0;
1535 		*chunk = data_len;
1536 		*zc = false;
1537 	}
1538 
1539 	/* Prepare and submit AEAD request */
1540 	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1541 				data_len, aead_req, async);
1542 	if (err == -EINPROGRESS)
1543 		return err;
1544 
1545 	/* Release the pages in case iov was mapped to pages */
1546 	for (; pages > 0; pages--)
1547 		put_page(sg_page(&sgout[pages]));
1548 
1549 	kfree(mem);
1550 	return err;
1551 }
1552 
decrypt_skb_update(struct sock * sk,struct sk_buff * skb,struct iov_iter * dest,int * chunk,bool * zc,bool async)1553 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1554 			      struct iov_iter *dest, int *chunk, bool *zc,
1555 			      bool async)
1556 {
1557 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1558 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1559 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1560 	struct strp_msg *rxm = strp_msg(skb);
1561 	int pad, err = 0;
1562 
1563 	if (!ctx->decrypted) {
1564 		if (tls_ctx->rx_conf == TLS_HW) {
1565 			err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1566 			if (err < 0)
1567 				return err;
1568 		}
1569 
1570 		/* Still not decrypted after tls_device */
1571 		if (!ctx->decrypted) {
1572 			err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1573 					       async);
1574 			if (err < 0) {
1575 				if (err == -EINPROGRESS)
1576 					tls_advance_record_sn(sk, prot,
1577 							      &tls_ctx->rx);
1578 				else if (err == -EBADMSG)
1579 					TLS_INC_STATS(sock_net(sk),
1580 						      LINUX_MIB_TLSDECRYPTERROR);
1581 				return err;
1582 			}
1583 		} else {
1584 			*zc = false;
1585 		}
1586 
1587 		pad = padding_length(ctx, prot, skb);
1588 		if (pad < 0)
1589 			return pad;
1590 
1591 		rxm->full_len -= pad;
1592 		rxm->offset += prot->prepend_size;
1593 		rxm->full_len -= prot->overhead_size;
1594 		tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1595 		ctx->decrypted = 1;
1596 		ctx->saved_data_ready(sk);
1597 	} else {
1598 		*zc = false;
1599 	}
1600 
1601 	return err;
1602 }
1603 
decrypt_skb(struct sock * sk,struct sk_buff * skb,struct scatterlist * sgout)1604 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1605 		struct scatterlist *sgout)
1606 {
1607 	bool zc = true;
1608 	int chunk;
1609 
1610 	return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1611 }
1612 
tls_sw_advance_skb(struct sock * sk,struct sk_buff * skb,unsigned int len)1613 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1614 			       unsigned int len)
1615 {
1616 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1617 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1618 
1619 	if (skb) {
1620 		struct strp_msg *rxm = strp_msg(skb);
1621 
1622 		if (len < rxm->full_len) {
1623 			rxm->offset += len;
1624 			rxm->full_len -= len;
1625 			return false;
1626 		}
1627 		consume_skb(skb);
1628 	}
1629 
1630 	/* Finished with message */
1631 	ctx->recv_pkt = NULL;
1632 	__strp_unpause(&ctx->strp);
1633 
1634 	return true;
1635 }
1636 
1637 /* This function traverses the rx_list in tls receive context to copies the
1638  * decrypted records into the buffer provided by caller zero copy is not
1639  * true. Further, the records are removed from the rx_list if it is not a peek
1640  * case and the record has been consumed completely.
1641  */
process_rx_list(struct tls_sw_context_rx * ctx,struct msghdr * msg,u8 * control,bool * cmsg,size_t skip,size_t len,bool zc,bool is_peek)1642 static int process_rx_list(struct tls_sw_context_rx *ctx,
1643 			   struct msghdr *msg,
1644 			   u8 *control,
1645 			   bool *cmsg,
1646 			   size_t skip,
1647 			   size_t len,
1648 			   bool zc,
1649 			   bool is_peek)
1650 {
1651 	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1652 	u8 ctrl = *control;
1653 	u8 msgc = *cmsg;
1654 	struct tls_msg *tlm;
1655 	ssize_t copied = 0;
1656 
1657 	/* Set the record type in 'control' if caller didn't pass it */
1658 	if (!ctrl && skb) {
1659 		tlm = tls_msg(skb);
1660 		ctrl = tlm->control;
1661 	}
1662 
1663 	while (skip && skb) {
1664 		struct strp_msg *rxm = strp_msg(skb);
1665 		tlm = tls_msg(skb);
1666 
1667 		/* Cannot process a record of different type */
1668 		if (ctrl != tlm->control)
1669 			return 0;
1670 
1671 		if (skip < rxm->full_len)
1672 			break;
1673 
1674 		skip = skip - rxm->full_len;
1675 		skb = skb_peek_next(skb, &ctx->rx_list);
1676 	}
1677 
1678 	while (len && skb) {
1679 		struct sk_buff *next_skb;
1680 		struct strp_msg *rxm = strp_msg(skb);
1681 		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1682 
1683 		tlm = tls_msg(skb);
1684 
1685 		/* Cannot process a record of different type */
1686 		if (ctrl != tlm->control)
1687 			return 0;
1688 
1689 		/* Set record type if not already done. For a non-data record,
1690 		 * do not proceed if record type could not be copied.
1691 		 */
1692 		if (!msgc) {
1693 			int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1694 					    sizeof(ctrl), &ctrl);
1695 			msgc = true;
1696 			if (ctrl != TLS_RECORD_TYPE_DATA) {
1697 				if (cerr || msg->msg_flags & MSG_CTRUNC)
1698 					return -EIO;
1699 
1700 				*cmsg = msgc;
1701 			}
1702 		}
1703 
1704 		if (!zc || (rxm->full_len - skip) > len) {
1705 			int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1706 						    msg, chunk);
1707 			if (err < 0)
1708 				return err;
1709 		}
1710 
1711 		len = len - chunk;
1712 		copied = copied + chunk;
1713 
1714 		/* Consume the data from record if it is non-peek case*/
1715 		if (!is_peek) {
1716 			rxm->offset = rxm->offset + chunk;
1717 			rxm->full_len = rxm->full_len - chunk;
1718 
1719 			/* Return if there is unconsumed data in the record */
1720 			if (rxm->full_len - skip)
1721 				break;
1722 		}
1723 
1724 		/* The remaining skip-bytes must lie in 1st record in rx_list.
1725 		 * So from the 2nd record, 'skip' should be 0.
1726 		 */
1727 		skip = 0;
1728 
1729 		if (msg)
1730 			msg->msg_flags |= MSG_EOR;
1731 
1732 		next_skb = skb_peek_next(skb, &ctx->rx_list);
1733 
1734 		if (!is_peek) {
1735 			skb_unlink(skb, &ctx->rx_list);
1736 			consume_skb(skb);
1737 		}
1738 
1739 		skb = next_skb;
1740 	}
1741 
1742 	*control = ctrl;
1743 	return copied;
1744 }
1745 
tls_sw_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int nonblock,int flags,int * addr_len)1746 int tls_sw_recvmsg(struct sock *sk,
1747 		   struct msghdr *msg,
1748 		   size_t len,
1749 		   int nonblock,
1750 		   int flags,
1751 		   int *addr_len)
1752 {
1753 	struct tls_context *tls_ctx = tls_get_ctx(sk);
1754 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1755 	struct tls_prot_info *prot = &tls_ctx->prot_info;
1756 	struct sk_psock *psock;
1757 	int num_async, pending;
1758 	unsigned char control = 0;
1759 	ssize_t decrypted = 0;
1760 	struct strp_msg *rxm;
1761 	struct tls_msg *tlm;
1762 	struct sk_buff *skb;
1763 	ssize_t copied = 0;
1764 	bool cmsg = false;
1765 	int target, err = 0;
1766 	long timeo;
1767 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1768 	bool is_peek = flags & MSG_PEEK;
1769 	bool bpf_strp_enabled;
1770 
1771 	flags |= nonblock;
1772 
1773 	if (unlikely(flags & MSG_ERRQUEUE))
1774 		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1775 
1776 	psock = sk_psock_get(sk);
1777 	lock_sock(sk);
1778 	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1779 
1780 	/* Process pending decrypted records. It must be non-zero-copy */
1781 	err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1782 			      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 	num_async = 0;
1798 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1799 		bool retain_skb = false;
1800 		bool zc = false;
1801 		int to_decrypt;
1802 		int chunk = 0;
1803 		bool async_capable;
1804 		bool async = false;
1805 
1806 		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1807 		if (!skb) {
1808 			if (psock) {
1809 				int ret = __tcp_bpf_recvmsg(sk, psock,
1810 							    msg, len, flags);
1811 
1812 				if (ret > 0) {
1813 					decrypted += ret;
1814 					len -= ret;
1815 					continue;
1816 				}
1817 			}
1818 			goto recv_end;
1819 		} else {
1820 			tlm = tls_msg(skb);
1821 			if (prot->version == TLS_1_3_VERSION)
1822 				tlm->control = 0;
1823 			else
1824 				tlm->control = ctx->control;
1825 		}
1826 
1827 		rxm = strp_msg(skb);
1828 
1829 		to_decrypt = rxm->full_len - prot->overhead_size;
1830 
1831 		if (to_decrypt <= len && !is_kvec && !is_peek &&
1832 		    ctx->control == TLS_RECORD_TYPE_DATA &&
1833 		    prot->version != TLS_1_3_VERSION &&
1834 		    !bpf_strp_enabled)
1835 			zc = true;
1836 
1837 		/* Do not use async mode if record is non-data */
1838 		if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1839 			async_capable = ctx->async_capable;
1840 		else
1841 			async_capable = false;
1842 
1843 		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1844 					 &chunk, &zc, async_capable);
1845 		if (err < 0 && err != -EINPROGRESS) {
1846 			tls_err_abort(sk, -EBADMSG);
1847 			goto recv_end;
1848 		}
1849 
1850 		if (err == -EINPROGRESS) {
1851 			async = true;
1852 			num_async++;
1853 		} else if (prot->version == TLS_1_3_VERSION) {
1854 			tlm->control = ctx->control;
1855 		}
1856 
1857 		/* If the type of records being processed is not known yet,
1858 		 * set it to record type just dequeued. If it is already known,
1859 		 * but does not match the record type just dequeued, go to end.
1860 		 * We always get record type here since for tls1.2, record type
1861 		 * is known just after record is dequeued from stream parser.
1862 		 * For tls1.3, we disable async.
1863 		 */
1864 
1865 		if (!control)
1866 			control = tlm->control;
1867 		else if (control != tlm->control)
1868 			goto recv_end;
1869 
1870 		if (!cmsg) {
1871 			int cerr;
1872 
1873 			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1874 					sizeof(control), &control);
1875 			cmsg = true;
1876 			if (control != TLS_RECORD_TYPE_DATA) {
1877 				if (cerr || msg->msg_flags & MSG_CTRUNC) {
1878 					err = -EIO;
1879 					goto recv_end;
1880 				}
1881 			}
1882 		}
1883 
1884 		if (async)
1885 			goto pick_next_record;
1886 
1887 		if (!zc) {
1888 			if (bpf_strp_enabled) {
1889 				err = sk_psock_tls_strp_read(psock, skb);
1890 				if (err != __SK_PASS) {
1891 					rxm->offset = rxm->offset + rxm->full_len;
1892 					rxm->full_len = 0;
1893 					if (err == __SK_DROP)
1894 						consume_skb(skb);
1895 					ctx->recv_pkt = NULL;
1896 					__strp_unpause(&ctx->strp);
1897 					continue;
1898 				}
1899 			}
1900 
1901 			if (rxm->full_len > len) {
1902 				retain_skb = true;
1903 				chunk = len;
1904 			} else {
1905 				chunk = rxm->full_len;
1906 			}
1907 
1908 			err = skb_copy_datagram_msg(skb, rxm->offset,
1909 						    msg, chunk);
1910 			if (err < 0)
1911 				goto recv_end;
1912 
1913 			if (!is_peek) {
1914 				rxm->offset = rxm->offset + chunk;
1915 				rxm->full_len = rxm->full_len - chunk;
1916 			}
1917 		}
1918 
1919 pick_next_record:
1920 		if (chunk > len)
1921 			chunk = len;
1922 
1923 		decrypted += chunk;
1924 		len -= chunk;
1925 
1926 		/* For async or peek case, queue the current skb */
1927 		if (async || is_peek || retain_skb) {
1928 			skb_queue_tail(&ctx->rx_list, skb);
1929 			skb = NULL;
1930 		}
1931 
1932 		if (tls_sw_advance_skb(sk, skb, chunk)) {
1933 			/* Return full control message to
1934 			 * userspace before trying to parse
1935 			 * another message type
1936 			 */
1937 			msg->msg_flags |= MSG_EOR;
1938 			if (control != TLS_RECORD_TYPE_DATA)
1939 				goto recv_end;
1940 		} else {
1941 			break;
1942 		}
1943 	}
1944 
1945 recv_end:
1946 	if (num_async) {
1947 		/* Wait for all previously submitted records to be decrypted */
1948 		spin_lock_bh(&ctx->decrypt_compl_lock);
1949 		ctx->async_notify = true;
1950 		pending = atomic_read(&ctx->decrypt_pending);
1951 		spin_unlock_bh(&ctx->decrypt_compl_lock);
1952 		if (pending) {
1953 			err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1954 			if (err) {
1955 				/* one of async decrypt failed */
1956 				tls_err_abort(sk, err);
1957 				copied = 0;
1958 				decrypted = 0;
1959 				goto end;
1960 			}
1961 		} else {
1962 			reinit_completion(&ctx->async_wait.completion);
1963 		}
1964 
1965 		/* There can be no concurrent accesses, since we have no
1966 		 * pending decrypt operations
1967 		 */
1968 		WRITE_ONCE(ctx->async_notify, false);
1969 
1970 		/* Drain records from the rx_list & copy if required */
1971 		if (is_peek || is_kvec)
1972 			err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1973 					      decrypted, false, is_peek);
1974 		else
1975 			err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1976 					      decrypted, true, is_peek);
1977 		if (err < 0) {
1978 			tls_err_abort(sk, err);
1979 			copied = 0;
1980 			goto end;
1981 		}
1982 	}
1983 
1984 	copied += decrypted;
1985 
1986 end:
1987 	release_sock(sk);
1988 	if (psock)
1989 		sk_psock_put(sk, psock);
1990 	return copied ? : err;
1991 }
1992 
tls_sw_splice_read(struct socket * sock,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)1993 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
1994 			   struct pipe_inode_info *pipe,
1995 			   size_t len, unsigned int flags)
1996 {
1997 	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1998 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1999 	struct strp_msg *rxm = NULL;
2000 	struct sock *sk = sock->sk;
2001 	struct sk_buff *skb;
2002 	ssize_t copied = 0;
2003 	int err = 0;
2004 	long timeo;
2005 	int chunk;
2006 	bool zc = false;
2007 
2008 	lock_sock(sk);
2009 
2010 	timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
2011 
2012 	skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo, &err);
2013 	if (!skb)
2014 		goto splice_read_end;
2015 
2016 	err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
2017 	if (err < 0) {
2018 		tls_err_abort(sk, -EBADMSG);
2019 		goto splice_read_end;
2020 	}
2021 
2022 	/* splice does not support reading control messages */
2023 	if (ctx->control != TLS_RECORD_TYPE_DATA) {
2024 		err = -EINVAL;
2025 		goto splice_read_end;
2026 	}
2027 
2028 	rxm = strp_msg(skb);
2029 
2030 	chunk = min_t(unsigned int, rxm->full_len, len);
2031 	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2032 	if (copied < 0)
2033 		goto splice_read_end;
2034 
2035 	if (likely(!(flags & MSG_PEEK)))
2036 		tls_sw_advance_skb(sk, skb, copied);
2037 
2038 splice_read_end:
2039 	release_sock(sk);
2040 	return copied ? : err;
2041 }
2042 
tls_sw_stream_read(const struct sock * sk)2043 bool tls_sw_stream_read(const struct sock *sk)
2044 {
2045 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2046 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2047 	bool ingress_empty = true;
2048 	struct sk_psock *psock;
2049 
2050 	rcu_read_lock();
2051 	psock = sk_psock(sk);
2052 	if (psock)
2053 		ingress_empty = list_empty(&psock->ingress_msg);
2054 	rcu_read_unlock();
2055 
2056 	return !ingress_empty || ctx->recv_pkt ||
2057 		!skb_queue_empty(&ctx->rx_list);
2058 }
2059 
tls_read_size(struct strparser * strp,struct sk_buff * skb)2060 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2061 {
2062 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2063 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2064 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2065 	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2066 	struct strp_msg *rxm = strp_msg(skb);
2067 	size_t cipher_overhead;
2068 	size_t data_len = 0;
2069 	int ret;
2070 
2071 	/* Verify that we have a full TLS header, or wait for more data */
2072 	if (rxm->offset + prot->prepend_size > skb->len)
2073 		return 0;
2074 
2075 	/* Sanity-check size of on-stack buffer. */
2076 	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2077 		ret = -EINVAL;
2078 		goto read_failure;
2079 	}
2080 
2081 	/* Linearize header to local buffer */
2082 	ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2083 
2084 	if (ret < 0)
2085 		goto read_failure;
2086 
2087 	ctx->control = header[0];
2088 
2089 	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2090 
2091 	cipher_overhead = prot->tag_size;
2092 	if (prot->version != TLS_1_3_VERSION)
2093 		cipher_overhead += prot->iv_size;
2094 
2095 	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2096 	    prot->tail_size) {
2097 		ret = -EMSGSIZE;
2098 		goto read_failure;
2099 	}
2100 	if (data_len < cipher_overhead) {
2101 		ret = -EBADMSG;
2102 		goto read_failure;
2103 	}
2104 
2105 	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2106 	if (header[1] != TLS_1_2_VERSION_MINOR ||
2107 	    header[2] != TLS_1_2_VERSION_MAJOR) {
2108 		ret = -EINVAL;
2109 		goto read_failure;
2110 	}
2111 
2112 	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2113 				     TCP_SKB_CB(skb)->seq + rxm->offset);
2114 	return data_len + TLS_HEADER_SIZE;
2115 
2116 read_failure:
2117 	tls_err_abort(strp->sk, ret);
2118 
2119 	return ret;
2120 }
2121 
tls_queue(struct strparser * strp,struct sk_buff * skb)2122 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2123 {
2124 	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2125 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2126 
2127 	ctx->decrypted = 0;
2128 
2129 	ctx->recv_pkt = skb;
2130 	strp_pause(strp);
2131 
2132 	ctx->saved_data_ready(strp->sk);
2133 }
2134 
tls_data_ready(struct sock * sk)2135 static void tls_data_ready(struct sock *sk)
2136 {
2137 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2138 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2139 	struct sk_psock *psock;
2140 
2141 	strp_data_ready(&ctx->strp);
2142 
2143 	psock = sk_psock_get(sk);
2144 	if (psock) {
2145 		if (!list_empty(&psock->ingress_msg))
2146 			ctx->saved_data_ready(sk);
2147 		sk_psock_put(sk, psock);
2148 	}
2149 }
2150 
tls_sw_cancel_work_tx(struct tls_context * tls_ctx)2151 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2152 {
2153 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2154 
2155 	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2156 	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2157 	cancel_delayed_work_sync(&ctx->tx_work.work);
2158 }
2159 
tls_sw_release_resources_tx(struct sock * sk)2160 void tls_sw_release_resources_tx(struct sock *sk)
2161 {
2162 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2163 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2164 	struct tls_rec *rec, *tmp;
2165 	int pending;
2166 
2167 	/* Wait for any pending async encryptions to complete */
2168 	spin_lock_bh(&ctx->encrypt_compl_lock);
2169 	ctx->async_notify = true;
2170 	pending = atomic_read(&ctx->encrypt_pending);
2171 	spin_unlock_bh(&ctx->encrypt_compl_lock);
2172 
2173 	if (pending)
2174 		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2175 
2176 	tls_tx_records(sk, -1);
2177 
2178 	/* Free up un-sent records in tx_list. First, free
2179 	 * the partially sent record if any at head of tx_list.
2180 	 */
2181 	if (tls_ctx->partially_sent_record) {
2182 		tls_free_partial_record(sk, tls_ctx);
2183 		rec = list_first_entry(&ctx->tx_list,
2184 				       struct tls_rec, list);
2185 		list_del(&rec->list);
2186 		sk_msg_free(sk, &rec->msg_plaintext);
2187 		kfree(rec);
2188 	}
2189 
2190 	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2191 		list_del(&rec->list);
2192 		sk_msg_free(sk, &rec->msg_encrypted);
2193 		sk_msg_free(sk, &rec->msg_plaintext);
2194 		kfree(rec);
2195 	}
2196 
2197 	crypto_free_aead(ctx->aead_send);
2198 	tls_free_open_rec(sk);
2199 }
2200 
tls_sw_free_ctx_tx(struct tls_context * tls_ctx)2201 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2202 {
2203 	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2204 
2205 	kfree(ctx);
2206 }
2207 
tls_sw_release_resources_rx(struct sock * sk)2208 void tls_sw_release_resources_rx(struct sock *sk)
2209 {
2210 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2211 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2212 
2213 	kfree(tls_ctx->rx.rec_seq);
2214 	kfree(tls_ctx->rx.iv);
2215 
2216 	if (ctx->aead_recv) {
2217 		kfree_skb(ctx->recv_pkt);
2218 		ctx->recv_pkt = NULL;
2219 		skb_queue_purge(&ctx->rx_list);
2220 		crypto_free_aead(ctx->aead_recv);
2221 		strp_stop(&ctx->strp);
2222 		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2223 		 * we still want to strp_stop(), but sk->sk_data_ready was
2224 		 * never swapped.
2225 		 */
2226 		if (ctx->saved_data_ready) {
2227 			write_lock_bh(&sk->sk_callback_lock);
2228 			sk->sk_data_ready = ctx->saved_data_ready;
2229 			write_unlock_bh(&sk->sk_callback_lock);
2230 		}
2231 	}
2232 }
2233 
tls_sw_strparser_done(struct tls_context * tls_ctx)2234 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2235 {
2236 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2237 
2238 	strp_done(&ctx->strp);
2239 }
2240 
tls_sw_free_ctx_rx(struct tls_context * tls_ctx)2241 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2242 {
2243 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2244 
2245 	kfree(ctx);
2246 }
2247 
tls_sw_free_resources_rx(struct sock * sk)2248 void tls_sw_free_resources_rx(struct sock *sk)
2249 {
2250 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2251 
2252 	tls_sw_release_resources_rx(sk);
2253 	tls_sw_free_ctx_rx(tls_ctx);
2254 }
2255 
2256 /* The work handler to transmitt the encrypted records in tx_list */
tx_work_handler(struct work_struct * work)2257 static void tx_work_handler(struct work_struct *work)
2258 {
2259 	struct delayed_work *delayed_work = to_delayed_work(work);
2260 	struct tx_work *tx_work = container_of(delayed_work,
2261 					       struct tx_work, work);
2262 	struct sock *sk = tx_work->sk;
2263 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2264 	struct tls_sw_context_tx *ctx;
2265 
2266 	if (unlikely(!tls_ctx))
2267 		return;
2268 
2269 	ctx = tls_sw_ctx_tx(tls_ctx);
2270 	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2271 		return;
2272 
2273 	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2274 		return;
2275 
2276 	if (mutex_trylock(&tls_ctx->tx_lock)) {
2277 		lock_sock(sk);
2278 		tls_tx_records(sk, -1);
2279 		release_sock(sk);
2280 		mutex_unlock(&tls_ctx->tx_lock);
2281 	} else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2282 		/* Someone is holding the tx_lock, they will likely run Tx
2283 		 * and cancel the work on their way out of the lock section.
2284 		 * Schedule a long delay just in case.
2285 		 */
2286 		schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2287 	}
2288 }
2289 
tls_sw_write_space(struct sock * sk,struct tls_context * ctx)2290 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2291 {
2292 	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2293 
2294 	/* Schedule the transmission if tx list is ready */
2295 	if (is_tx_ready(tx_ctx) &&
2296 	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2297 		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2298 }
2299 
tls_sw_strparser_arm(struct sock * sk,struct tls_context * tls_ctx)2300 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2301 {
2302 	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2303 
2304 	write_lock_bh(&sk->sk_callback_lock);
2305 	rx_ctx->saved_data_ready = sk->sk_data_ready;
2306 	sk->sk_data_ready = tls_data_ready;
2307 	write_unlock_bh(&sk->sk_callback_lock);
2308 
2309 	strp_check_rcv(&rx_ctx->strp);
2310 }
2311 
tls_set_sw_offload(struct sock * sk,struct tls_context * ctx,int tx)2312 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2313 {
2314 	struct tls_context *tls_ctx = tls_get_ctx(sk);
2315 	struct tls_prot_info *prot = &tls_ctx->prot_info;
2316 	struct tls_crypto_info *crypto_info;
2317 	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2318 	struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2319 	struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2320 	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2321 	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2322 	struct cipher_context *cctx;
2323 	struct crypto_aead **aead;
2324 	struct strp_callbacks cb;
2325 	u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2326 	struct crypto_tfm *tfm;
2327 	char *iv, *rec_seq, *key, *salt, *cipher_name;
2328 	size_t keysize;
2329 	int rc = 0;
2330 
2331 	if (!ctx) {
2332 		rc = -EINVAL;
2333 		goto out;
2334 	}
2335 
2336 	if (tx) {
2337 		if (!ctx->priv_ctx_tx) {
2338 			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2339 			if (!sw_ctx_tx) {
2340 				rc = -ENOMEM;
2341 				goto out;
2342 			}
2343 			ctx->priv_ctx_tx = sw_ctx_tx;
2344 		} else {
2345 			sw_ctx_tx =
2346 				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2347 		}
2348 	} else {
2349 		if (!ctx->priv_ctx_rx) {
2350 			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2351 			if (!sw_ctx_rx) {
2352 				rc = -ENOMEM;
2353 				goto out;
2354 			}
2355 			ctx->priv_ctx_rx = sw_ctx_rx;
2356 		} else {
2357 			sw_ctx_rx =
2358 				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2359 		}
2360 	}
2361 
2362 	if (tx) {
2363 		crypto_init_wait(&sw_ctx_tx->async_wait);
2364 		spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2365 		crypto_info = &ctx->crypto_send.info;
2366 		cctx = &ctx->tx;
2367 		aead = &sw_ctx_tx->aead_send;
2368 		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2369 		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2370 		sw_ctx_tx->tx_work.sk = sk;
2371 	} else {
2372 		crypto_init_wait(&sw_ctx_rx->async_wait);
2373 		spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2374 		crypto_info = &ctx->crypto_recv.info;
2375 		cctx = &ctx->rx;
2376 		skb_queue_head_init(&sw_ctx_rx->rx_list);
2377 		aead = &sw_ctx_rx->aead_recv;
2378 	}
2379 
2380 	switch (crypto_info->cipher_type) {
2381 	case TLS_CIPHER_AES_GCM_128: {
2382 		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2383 		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2384 		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2385 		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2386 		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2387 		rec_seq =
2388 		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2389 		gcm_128_info =
2390 			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2391 		keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2392 		key = gcm_128_info->key;
2393 		salt = gcm_128_info->salt;
2394 		salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2395 		cipher_name = "gcm(aes)";
2396 		break;
2397 	}
2398 	case TLS_CIPHER_AES_GCM_256: {
2399 		nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2400 		tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2401 		iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2402 		iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2403 		rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2404 		rec_seq =
2405 		 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2406 		gcm_256_info =
2407 			(struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2408 		keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2409 		key = gcm_256_info->key;
2410 		salt = gcm_256_info->salt;
2411 		salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2412 		cipher_name = "gcm(aes)";
2413 		break;
2414 	}
2415 	case TLS_CIPHER_AES_CCM_128: {
2416 		nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2417 		tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2418 		iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2419 		iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2420 		rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2421 		rec_seq =
2422 		((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2423 		ccm_128_info =
2424 		(struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2425 		keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2426 		key = ccm_128_info->key;
2427 		salt = ccm_128_info->salt;
2428 		salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2429 		cipher_name = "ccm(aes)";
2430 		break;
2431 	}
2432 	default:
2433 		rc = -EINVAL;
2434 		goto free_priv;
2435 	}
2436 
2437 	/* Sanity-check the sizes for stack allocations. */
2438 	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2439 	    rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2440 		rc = -EINVAL;
2441 		goto free_priv;
2442 	}
2443 
2444 	if (crypto_info->version == TLS_1_3_VERSION) {
2445 		nonce_size = 0;
2446 		prot->aad_size = TLS_HEADER_SIZE;
2447 		prot->tail_size = 1;
2448 	} else {
2449 		prot->aad_size = TLS_AAD_SPACE_SIZE;
2450 		prot->tail_size = 0;
2451 	}
2452 
2453 	prot->version = crypto_info->version;
2454 	prot->cipher_type = crypto_info->cipher_type;
2455 	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2456 	prot->tag_size = tag_size;
2457 	prot->overhead_size = prot->prepend_size +
2458 			      prot->tag_size + prot->tail_size;
2459 	prot->iv_size = iv_size;
2460 	prot->salt_size = salt_size;
2461 	cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2462 	if (!cctx->iv) {
2463 		rc = -ENOMEM;
2464 		goto free_priv;
2465 	}
2466 	/* Note: 128 & 256 bit salt are the same size */
2467 	prot->rec_seq_size = rec_seq_size;
2468 	memcpy(cctx->iv, salt, salt_size);
2469 	memcpy(cctx->iv + salt_size, iv, iv_size);
2470 	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2471 	if (!cctx->rec_seq) {
2472 		rc = -ENOMEM;
2473 		goto free_iv;
2474 	}
2475 
2476 	if (!*aead) {
2477 		*aead = crypto_alloc_aead(cipher_name, 0, 0);
2478 		if (IS_ERR(*aead)) {
2479 			rc = PTR_ERR(*aead);
2480 			*aead = NULL;
2481 			goto free_rec_seq;
2482 		}
2483 	}
2484 
2485 	ctx->push_pending_record = tls_sw_push_pending_record;
2486 
2487 	rc = crypto_aead_setkey(*aead, key, keysize);
2488 
2489 	if (rc)
2490 		goto free_aead;
2491 
2492 	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2493 	if (rc)
2494 		goto free_aead;
2495 
2496 	if (sw_ctx_rx) {
2497 		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2498 
2499 		if (crypto_info->version == TLS_1_3_VERSION)
2500 			sw_ctx_rx->async_capable = 0;
2501 		else
2502 			sw_ctx_rx->async_capable =
2503 				!!(tfm->__crt_alg->cra_flags &
2504 				   CRYPTO_ALG_ASYNC);
2505 
2506 		/* Set up strparser */
2507 		memset(&cb, 0, sizeof(cb));
2508 		cb.rcv_msg = tls_queue;
2509 		cb.parse_msg = tls_read_size;
2510 
2511 		strp_init(&sw_ctx_rx->strp, sk, &cb);
2512 	}
2513 
2514 	goto out;
2515 
2516 free_aead:
2517 	crypto_free_aead(*aead);
2518 	*aead = NULL;
2519 free_rec_seq:
2520 	kfree(cctx->rec_seq);
2521 	cctx->rec_seq = NULL;
2522 free_iv:
2523 	kfree(cctx->iv);
2524 	cctx->iv = NULL;
2525 free_priv:
2526 	if (tx) {
2527 		kfree(ctx->priv_ctx_tx);
2528 		ctx->priv_ctx_tx = NULL;
2529 	} else {
2530 		kfree(ctx->priv_ctx_rx);
2531 		ctx->priv_ctx_rx = NULL;
2532 	}
2533 out:
2534 	return rc;
2535 }
2536