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/kernel.h>
42	#include <linux/splice.h>
43	#include <crypto/aead.h>
44
45	#include <net/strparser.h>
46	#include <net/tls.h>
47	#include <trace/events/sock.h>
48
49	#include "tls.h"
50
51	struct tls_decrypt_arg {
52	struct_group(inargs,
53	bool zc;
54	bool async;
55	bool async_done;
56	u8 tail;
57	);
58
59	struct sk_buff *skb;
60	};
61
62	struct tls_decrypt_ctx {
63	struct sock *sk;
64	u8 iv[TLS_MAX_IV_SIZE];
65	u8 aad[TLS_MAX_AAD_SIZE];
66	u8 tail;
67	bool free_sgout;
68	struct scatterlist sg[];
69	};
70
71	noinline void tls_err_abort(struct sock *sk, int err)
72	{
73	WARN_ON_ONCE(err >= 0);
74	/* sk->sk_err should contain a positive error code. */
75	WRITE_ONCE(sk->sk_err, -err);
76	/* Paired with smp_rmb() in tcp_poll() */
77	smp_wmb();
78	sk_error_report(sk);
79	}
80
81	static int __skb_nsg(struct sk_buff *skb, int offset, int len,
82	unsigned int recursion_level)
83	{
84	int start = skb_headlen(skb);
85	int i, chunk = start - offset;
86	struct sk_buff *frag_iter;
87	int elt = 0;
88
89	if (unlikely(recursion_level >= 24))
90	return -EMSGSIZE;
91
92	if (chunk > 0) {
93	if (chunk > len)
94	chunk = len;
95	elt++;
96	len -= chunk;
97	if (len == 0)
98	return elt;
99	offset += chunk;
100	}
101
102	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
103	int end;
104
105	WARN_ON(start > offset + len);
106
107	end = start + skb_frag_size(frag: &skb_shinfo(skb)->frags[i]);
108	chunk = end - offset;
109	if (chunk > 0) {
110	if (chunk > len)
111	chunk = len;
112	elt++;
113	len -= chunk;
114	if (len == 0)
115	return elt;
116	offset += chunk;
117	}
118	start = end;
119	}
120
121	if (unlikely(skb_has_frag_list(skb))) {
122	skb_walk_frags(skb, frag_iter) {
123	int end, ret;
124
125	WARN_ON(start > offset + len);
126
127	end = start + frag_iter->len;
128	chunk = end - offset;
129	if (chunk > 0) {
130	if (chunk > len)
131	chunk = len;
132	ret = __skb_nsg(skb: frag_iter, offset: offset - start, len: chunk,
133	recursion_level: recursion_level + 1);
134	if (unlikely(ret < 0))
135	return ret;
136	elt += ret;
137	len -= chunk;
138	if (len == 0)
139	return elt;
140	offset += chunk;
141	}
142	start = end;
143	}
144	}
145	BUG_ON(len);
146	return elt;
147	}
148
149	/* Return the number of scatterlist elements required to completely map the
150	* skb, or -EMSGSIZE if the recursion depth is exceeded.
151	*/
152	static int skb_nsg(struct sk_buff *skb, int offset, int len)
153	{
154	return __skb_nsg(skb, offset, len, recursion_level: 0);
155	}
156
157	static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
158	struct tls_decrypt_arg *darg)
159	{
160	struct strp_msg *rxm = strp_msg(skb);
161	struct tls_msg *tlm = tls_msg(skb);
162	int sub = 0;
163
164	/* Determine zero-padding length */
165	if (prot->version == TLS_1_3_VERSION) {
166	int offset = rxm->full_len - TLS_TAG_SIZE - 1;
167	char content_type = darg->zc ? darg->tail : 0;
168	int err;
169
170	while (content_type == 0) {
171	if (offset < prot->prepend_size)
172	return -EBADMSG;
173	err = skb_copy_bits(skb, offset: rxm->offset + offset,
174	to: &content_type, len: 1);
175	if (err)
176	return err;
177	if (content_type)
178	break;
179	sub++;
180	offset--;
181	}
182	tlm->control = content_type;
183	}
184	return sub;
185	}
186
187	static void tls_decrypt_done(void *data, int err)
188	{
189	struct aead_request *aead_req = data;
190	struct crypto_aead *aead = crypto_aead_reqtfm(req: aead_req);
191	struct scatterlist *sgout = aead_req->dst;
192	struct tls_sw_context_rx *ctx;
193	struct tls_decrypt_ctx *dctx;
194	struct tls_context *tls_ctx;
195	struct scatterlist *sg;
196	unsigned int pages;
197	struct sock *sk;
198	int aead_size;
199
200	/* If requests get too backlogged crypto API returns -EBUSY and calls
201	* ->complete(-EINPROGRESS) immediately followed by ->complete(0)
202	* to make waiting for backlog to flush with crypto_wait_req() easier.
203	* First wait converts -EBUSY -> -EINPROGRESS, and the second one
204	* -EINPROGRESS -> 0.
205	* We have a single struct crypto_async_request per direction, this
206	* scheme doesn't help us, so just ignore the first ->complete().
207	*/
208	if (err == -EINPROGRESS)
209	return;
210
211	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(tfm: aead);
212	aead_size = ALIGN(aead_size, __alignof__(*dctx));
213	dctx = (void *)((u8 *)aead_req + aead_size);
214
215	sk = dctx->sk;
216	tls_ctx = tls_get_ctx(sk);
217	ctx = tls_sw_ctx_rx(tls_ctx);
218
219	/* Propagate if there was an err */
220	if (err) {
221	if (err == -EBADMSG)
222	TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
223	ctx->async_wait.err = err;
224	tls_err_abort(sk, err);
225	}
226
227	/* Free the destination pages if skb was not decrypted inplace */
228	if (dctx->free_sgout) {
229	/* Skip the first S/G entry as it points to AAD */
230	for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
231	if (!sg)
232	break;
233	put_page(page: sg_page(sg));
234	}
235	}
236
237	kfree(objp: aead_req);
238
239	if (atomic_dec_and_test(v: &ctx->decrypt_pending))
240	complete(&ctx->async_wait.completion);
241	}
242
243	static int tls_decrypt_async_wait(struct tls_sw_context_rx *ctx)
244	{
245	if (!atomic_dec_and_test(v: &ctx->decrypt_pending))
246	crypto_wait_req(err: -EINPROGRESS, wait: &ctx->async_wait);
247	atomic_inc(v: &ctx->decrypt_pending);
248
249	return ctx->async_wait.err;
250	}
251
252	static int tls_do_decryption(struct sock *sk,
253	struct scatterlist *sgin,
254	struct scatterlist *sgout,
255	char *iv_recv,
256	size_t data_len,
257	struct aead_request *aead_req,
258	struct tls_decrypt_arg *darg)
259	{
260	struct tls_context *tls_ctx = tls_get_ctx(sk);
261	struct tls_prot_info *prot = &tls_ctx->prot_info;
262	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
263	int ret;
264
265	aead_request_set_tfm(req: aead_req, tfm: ctx->aead_recv);
266	aead_request_set_ad(req: aead_req, assoclen: prot->aad_size);
267	aead_request_set_crypt(req: aead_req, src: sgin, dst: sgout,
268	cryptlen: data_len + prot->tag_size,
269	iv: (u8 *)iv_recv);
270
271	if (darg->async) {
272	aead_request_set_callback(req: aead_req,
273	CRYPTO_TFM_REQ_MAY_BACKLOG,
274	compl: tls_decrypt_done, data: aead_req);
275	DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->decrypt_pending) < 1);
276	atomic_inc(v: &ctx->decrypt_pending);
277	} else {
278	DECLARE_CRYPTO_WAIT(wait);
279
280	aead_request_set_callback(req: aead_req,
281	CRYPTO_TFM_REQ_MAY_BACKLOG,
282	compl: crypto_req_done, data: &wait);
283	ret = crypto_aead_decrypt(req: aead_req);
284	if (ret == -EINPROGRESS || ret == -EBUSY)
285	ret = crypto_wait_req(err: ret, wait: &wait);
286	return ret;
287	}
288
289	ret = crypto_aead_decrypt(req: aead_req);
290	if (ret == -EINPROGRESS)
291	return 0;
292
293	if (ret == -EBUSY) {
294	ret = tls_decrypt_async_wait(ctx);
295	darg->async_done = true;
296	/* all completions have run, we're not doing async anymore */
297	darg->async = false;
298	return ret;
299	}
300
301	atomic_dec(v: &ctx->decrypt_pending);
302	darg->async = false;
303
304	return ret;
305	}
306
307	static void tls_trim_both_msgs(struct sock *sk, int target_size)
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
314	sk_msg_trim(sk, msg: &rec->msg_plaintext, len: target_size);
315	if (target_size > 0)
316	target_size += prot->overhead_size;
317	sk_msg_trim(sk, msg: &rec->msg_encrypted, len: target_size);
318	}
319
320	static int tls_alloc_encrypted_msg(struct sock *sk, int len)
321	{
322	struct tls_context *tls_ctx = tls_get_ctx(sk);
323	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
324	struct tls_rec *rec = ctx->open_rec;
325	struct sk_msg *msg_en = &rec->msg_encrypted;
326
327	return sk_msg_alloc(sk, msg: msg_en, len, elem_first_coalesce: 0);
328	}
329
330	static int tls_clone_plaintext_msg(struct sock *sk, int required)
331	{
332	struct tls_context *tls_ctx = tls_get_ctx(sk);
333	struct tls_prot_info *prot = &tls_ctx->prot_info;
334	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
335	struct tls_rec *rec = ctx->open_rec;
336	struct sk_msg *msg_pl = &rec->msg_plaintext;
337	struct sk_msg *msg_en = &rec->msg_encrypted;
338	int skip, len;
339
340	/* We add page references worth len bytes from encrypted sg
341	* at the end of plaintext sg. It is guaranteed that msg_en
342	* has enough required room (ensured by caller).
343	*/
344	len = required - msg_pl->sg.size;
345
346	/* Skip initial bytes in msg_en's data to be able to use
347	* same offset of both plain and encrypted data.
348	*/
349	skip = prot->prepend_size + msg_pl->sg.size;
350
351	return sk_msg_clone(sk, dst: msg_pl, src: msg_en, off: skip, len);
352	}
353
354	static struct tls_rec *tls_get_rec(struct sock *sk)
355	{
356	struct tls_context *tls_ctx = tls_get_ctx(sk);
357	struct tls_prot_info *prot = &tls_ctx->prot_info;
358	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
359	struct sk_msg *msg_pl, *msg_en;
360	struct tls_rec *rec;
361	int mem_size;
362
363	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(tfm: ctx->aead_send);
364
365	rec = kzalloc(mem_size, sk->sk_allocation);
366	if (!rec)
367	return NULL;
368
369	msg_pl = &rec->msg_plaintext;
370	msg_en = &rec->msg_encrypted;
371
372	sk_msg_init(msg: msg_pl);
373	sk_msg_init(msg: msg_en);
374
375	sg_init_table(rec->sg_aead_in, 2);
376	sg_set_buf(sg: &rec->sg_aead_in[0], buf: rec->aad_space, buflen: prot->aad_size);
377	sg_unmark_end(sg: &rec->sg_aead_in[1]);
378
379	sg_init_table(rec->sg_aead_out, 2);
380	sg_set_buf(sg: &rec->sg_aead_out[0], buf: rec->aad_space, buflen: prot->aad_size);
381	sg_unmark_end(sg: &rec->sg_aead_out[1]);
382
383	rec->sk = sk;
384
385	return rec;
386	}
387
388	static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
389	{
390	sk_msg_free(sk, msg: &rec->msg_encrypted);
391	sk_msg_free(sk, msg: &rec->msg_plaintext);
392	kfree(objp: rec);
393	}
394
395	static void tls_free_open_rec(struct sock *sk)
396	{
397	struct tls_context *tls_ctx = tls_get_ctx(sk);
398	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
399	struct tls_rec *rec = ctx->open_rec;
400
401	if (rec) {
402	tls_free_rec(sk, rec);
403	ctx->open_rec = NULL;
404	}
405	}
406
407	int tls_tx_records(struct sock *sk, int flags)
408	{
409	struct tls_context *tls_ctx = tls_get_ctx(sk);
410	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
411	struct tls_rec *rec, *tmp;
412	struct sk_msg *msg_en;
413	int tx_flags, rc = 0;
414
415	if (tls_is_partially_sent_record(ctx: tls_ctx)) {
416	rec = list_first_entry(&ctx->tx_list,
417	struct tls_rec, list);
418
419	if (flags == -1)
420	tx_flags = rec->tx_flags;
421	else
422	tx_flags = flags;
423
424	rc = tls_push_partial_record(sk, ctx: tls_ctx, flags: tx_flags);
425	if (rc)
426	goto tx_err;
427
428	/* Full record has been transmitted.
429	* Remove the head of tx_list
430	*/
431	list_del(entry: &rec->list);
432	sk_msg_free(sk, msg: &rec->msg_plaintext);
433	kfree(objp: rec);
434	}
435
436	/* Tx all ready records */
437	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
438	if (READ_ONCE(rec->tx_ready)) {
439	if (flags == -1)
440	tx_flags = rec->tx_flags;
441	else
442	tx_flags = flags;
443
444	msg_en = &rec->msg_encrypted;
445	rc = tls_push_sg(sk, ctx: tls_ctx,
446	sg: &msg_en->sg.data[msg_en->sg.curr],
447	first_offset: 0, flags: tx_flags);
448	if (rc)
449	goto tx_err;
450
451	list_del(entry: &rec->list);
452	sk_msg_free(sk, msg: &rec->msg_plaintext);
453	kfree(objp: rec);
454	} else {
455	break;
456	}
457	}
458
459	tx_err:
460	if (rc < 0 && rc != -EAGAIN)
461	tls_err_abort(sk, err: rc);
462
463	return rc;
464	}
465
466	static void tls_encrypt_done(void *data, int err)
467	{
468	struct tls_sw_context_tx *ctx;
469	struct tls_context *tls_ctx;
470	struct tls_prot_info *prot;
471	struct tls_rec *rec = data;
472	struct scatterlist *sge;
473	struct sk_msg *msg_en;
474	struct sock *sk;
475
476	if (err == -EINPROGRESS) /* see the comment in tls_decrypt_done() */
477	return;
478
479	msg_en = &rec->msg_encrypted;
480
481	sk = rec->sk;
482	tls_ctx = tls_get_ctx(sk);
483	prot = &tls_ctx->prot_info;
484	ctx = tls_sw_ctx_tx(tls_ctx);
485
486	sge = sk_msg_elem(msg: msg_en, which: msg_en->sg.curr);
487	sge->offset -= prot->prepend_size;
488	sge->length += prot->prepend_size;
489
490	/* Check if error is previously set on socket */
491	if (err || sk->sk_err) {
492	rec = NULL;
493
494	/* If err is already set on socket, return the same code */
495	if (sk->sk_err) {
496	ctx->async_wait.err = -sk->sk_err;
497	} else {
498	ctx->async_wait.err = err;
499	tls_err_abort(sk, err);
500	}
501	}
502
503	if (rec) {
504	struct tls_rec *first_rec;
505
506	/* Mark the record as ready for transmission */
507	smp_store_mb(rec->tx_ready, true);
508
509	/* If received record is at head of tx_list, schedule tx */
510	first_rec = list_first_entry(&ctx->tx_list,
511	struct tls_rec, list);
512	if (rec == first_rec) {
513	/* Schedule the transmission */
514	if (!test_and_set_bit(BIT_TX_SCHEDULED,
515	addr: &ctx->tx_bitmask))
516	schedule_delayed_work(dwork: &ctx->tx_work.work, delay: 1);
517	}
518	}
519
520	if (atomic_dec_and_test(v: &ctx->encrypt_pending))
521	complete(&ctx->async_wait.completion);
522	}
523
524	static int tls_encrypt_async_wait(struct tls_sw_context_tx *ctx)
525	{
526	if (!atomic_dec_and_test(v: &ctx->encrypt_pending))
527	crypto_wait_req(err: -EINPROGRESS, wait: &ctx->async_wait);
528	atomic_inc(v: &ctx->encrypt_pending);
529
530	return ctx->async_wait.err;
531	}
532
533	static int tls_do_encryption(struct sock *sk,
534	struct tls_context *tls_ctx,
535	struct tls_sw_context_tx *ctx,
536	struct aead_request *aead_req,
537	size_t data_len, u32 start)
538	{
539	struct tls_prot_info *prot = &tls_ctx->prot_info;
540	struct tls_rec *rec = ctx->open_rec;
541	struct sk_msg *msg_en = &rec->msg_encrypted;
542	struct scatterlist *sge = sk_msg_elem(msg: msg_en, which: start);
543	int rc, iv_offset = 0;
544
545	/* For CCM based ciphers, first byte of IV is a constant */
546	switch (prot->cipher_type) {
547	case TLS_CIPHER_AES_CCM_128:
548	rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
549	iv_offset = 1;
550	break;
551	case TLS_CIPHER_SM4_CCM:
552	rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
553	iv_offset = 1;
554	break;
555	}
556
557	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
558	prot->iv_size + prot->salt_size);
559
560	tls_xor_iv_with_seq(prot, iv: rec->iv_data + iv_offset,
561	seq: tls_ctx->tx.rec_seq);
562
563	sge->offset += prot->prepend_size;
564	sge->length -= prot->prepend_size;
565
566	msg_en->sg.curr = start;
567
568	aead_request_set_tfm(req: aead_req, tfm: ctx->aead_send);
569	aead_request_set_ad(req: aead_req, assoclen: prot->aad_size);
570	aead_request_set_crypt(req: aead_req, src: rec->sg_aead_in,
571	dst: rec->sg_aead_out,
572	cryptlen: data_len, iv: rec->iv_data);
573
574	aead_request_set_callback(req: aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
575	compl: tls_encrypt_done, data: rec);
576
577	/* Add the record in tx_list */
578	list_add_tail(new: (struct list_head *)&rec->list, head: &ctx->tx_list);
579	DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->encrypt_pending) < 1);
580	atomic_inc(v: &ctx->encrypt_pending);
581
582	rc = crypto_aead_encrypt(req: aead_req);
583	if (rc == -EBUSY) {
584	rc = tls_encrypt_async_wait(ctx);
585	rc = rc ?: -EINPROGRESS;
586	}
587	if (!rc || rc != -EINPROGRESS) {
588	atomic_dec(v: &ctx->encrypt_pending);
589	sge->offset -= prot->prepend_size;
590	sge->length += prot->prepend_size;
591	}
592
593	if (!rc) {
594	WRITE_ONCE(rec->tx_ready, true);
595	} else if (rc != -EINPROGRESS) {
596	list_del(entry: &rec->list);
597	return rc;
598	}
599
600	/* Unhook the record from context if encryption is not failure */
601	ctx->open_rec = NULL;
602	tls_advance_record_sn(sk, prot, ctx: &tls_ctx->tx);
603	return rc;
604	}
605
606	static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
607	struct tls_rec **to, struct sk_msg *msg_opl,
608	struct sk_msg *msg_oen, u32 split_point,
609	u32 tx_overhead_size, u32 *orig_end)
610	{
611	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
612	struct scatterlist *sge, *osge, *nsge;
613	u32 orig_size = msg_opl->sg.size;
614	struct scatterlist tmp = { };
615	struct sk_msg *msg_npl;
616	struct tls_rec *new;
617	int ret;
618
619	new = tls_get_rec(sk);
620	if (!new)
621	return -ENOMEM;
622	ret = sk_msg_alloc(sk, msg: &new->msg_encrypted, len: msg_opl->sg.size +
623	tx_overhead_size, elem_first_coalesce: 0);
624	if (ret < 0) {
625	tls_free_rec(sk, rec: new);
626	return ret;
627	}
628
629	*orig_end = msg_opl->sg.end;
630	i = msg_opl->sg.start;
631	sge = sk_msg_elem(msg: msg_opl, which: i);
632	while (apply && sge->length) {
633	if (sge->length > apply) {
634	u32 len = sge->length - apply;
635
636	get_page(page: sg_page(sg: sge));
637	sg_set_page(sg: &tmp, page: sg_page(sg: sge), len,
638	offset: sge->offset + apply);
639	sge->length = apply;
640	bytes += apply;
641	apply = 0;
642	} else {
643	apply -= sge->length;
644	bytes += sge->length;
645	}
646
647	sk_msg_iter_var_next(i);
648	if (i == msg_opl->sg.end)
649	break;
650	sge = sk_msg_elem(msg: msg_opl, which: i);
651	}
652
653	msg_opl->sg.end = i;
654	msg_opl->sg.curr = i;
655	msg_opl->sg.copybreak = 0;
656	msg_opl->apply_bytes = 0;
657	msg_opl->sg.size = bytes;
658
659	msg_npl = &new->msg_plaintext;
660	msg_npl->apply_bytes = apply;
661	msg_npl->sg.size = orig_size - bytes;
662
663	j = msg_npl->sg.start;
664	nsge = sk_msg_elem(msg: msg_npl, which: j);
665	if (tmp.length) {
666	memcpy(nsge, &tmp, sizeof(*nsge));
667	sk_msg_iter_var_next(j);
668	nsge = sk_msg_elem(msg: msg_npl, which: j);
669	}
670
671	osge = sk_msg_elem(msg: msg_opl, which: i);
672	while (osge->length) {
673	memcpy(nsge, osge, sizeof(*nsge));
674	sg_unmark_end(sg: nsge);
675	sk_msg_iter_var_next(i);
676	sk_msg_iter_var_next(j);
677	if (i == *orig_end)
678	break;
679	osge = sk_msg_elem(msg: msg_opl, which: i);
680	nsge = sk_msg_elem(msg: msg_npl, which: j);
681	}
682
683	msg_npl->sg.end = j;
684	msg_npl->sg.curr = j;
685	msg_npl->sg.copybreak = 0;
686
687	*to = new;
688	return 0;
689	}
690
691	static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
692	struct tls_rec *from, u32 orig_end)
693	{
694	struct sk_msg *msg_npl = &from->msg_plaintext;
695	struct sk_msg *msg_opl = &to->msg_plaintext;
696	struct scatterlist *osge, *nsge;
697	u32 i, j;
698
699	i = msg_opl->sg.end;
700	sk_msg_iter_var_prev(i);
701	j = msg_npl->sg.start;
702
703	osge = sk_msg_elem(msg: msg_opl, which: i);
704	nsge = sk_msg_elem(msg: msg_npl, which: j);
705
706	if (sg_page(sg: osge) == sg_page(sg: nsge) &&
707	osge->offset + osge->length == nsge->offset) {
708	osge->length += nsge->length;
709	put_page(page: sg_page(sg: nsge));
710	}
711
712	msg_opl->sg.end = orig_end;
713	msg_opl->sg.curr = orig_end;
714	msg_opl->sg.copybreak = 0;
715	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
716	msg_opl->sg.size += msg_npl->sg.size;
717
718	sk_msg_free(sk, msg: &to->msg_encrypted);
719	sk_msg_xfer_full(dst: &to->msg_encrypted, src: &from->msg_encrypted);
720
721	kfree(objp: from);
722	}
723
724	static int tls_push_record(struct sock *sk, int flags,
725	unsigned char record_type)
726	{
727	struct tls_context *tls_ctx = tls_get_ctx(sk);
728	struct tls_prot_info *prot = &tls_ctx->prot_info;
729	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
730	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
731	u32 i, split_point, orig_end;
732	struct sk_msg *msg_pl, *msg_en;
733	struct aead_request *req;
734	bool split;
735	int rc;
736
737	if (!rec)
738	return 0;
739
740	msg_pl = &rec->msg_plaintext;
741	msg_en = &rec->msg_encrypted;
742
743	split_point = msg_pl->apply_bytes;
744	split = split_point && split_point < msg_pl->sg.size;
745	if (unlikely((!split &&
746	msg_pl->sg.size +
747	prot->overhead_size > msg_en->sg.size) ||
748	(split &&
749	split_point +
750	prot->overhead_size > msg_en->sg.size))) {
751	split = true;
752	split_point = msg_en->sg.size;
753	}
754	if (split) {
755	rc = tls_split_open_record(sk, from: rec, to: &tmp, msg_opl: msg_pl, msg_oen: msg_en,
756	split_point, tx_overhead_size: prot->overhead_size,
757	orig_end: &orig_end);
758	if (rc < 0)
759	return rc;
760	/* This can happen if above tls_split_open_record allocates
761	* a single large encryption buffer instead of two smaller
762	* ones. In this case adjust pointers and continue without
763	* split.
764	*/
765	if (!msg_pl->sg.size) {
766	tls_merge_open_record(sk, to: rec, from: tmp, orig_end);
767	msg_pl = &rec->msg_plaintext;
768	msg_en = &rec->msg_encrypted;
769	split = false;
770	}
771	sk_msg_trim(sk, msg: msg_en, len: msg_pl->sg.size +
772	prot->overhead_size);
773	}
774
775	rec->tx_flags = flags;
776	req = &rec->aead_req;
777
778	i = msg_pl->sg.end;
779	sk_msg_iter_var_prev(i);
780
781	rec->content_type = record_type;
782	if (prot->version == TLS_1_3_VERSION) {
783	/* Add content type to end of message. No padding added */
784	sg_set_buf(sg: &rec->sg_content_type, buf: &rec->content_type, buflen: 1);
785	sg_mark_end(sg: &rec->sg_content_type);
786	sg_chain(prv: msg_pl->sg.data, prv_nents: msg_pl->sg.end + 1,
787	sgl: &rec->sg_content_type);
788	} else {
789	sg_mark_end(sg: sk_msg_elem(msg: msg_pl, which: i));
790	}
791
792	if (msg_pl->sg.end < msg_pl->sg.start) {
793	sg_chain(prv: &msg_pl->sg.data[msg_pl->sg.start],
794	MAX_SKB_FRAGS - msg_pl->sg.start + 1,
795	sgl: msg_pl->sg.data);
796	}
797
798	i = msg_pl->sg.start;
799	sg_chain(prv: rec->sg_aead_in, prv_nents: 2, sgl: &msg_pl->sg.data[i]);
800
801	i = msg_en->sg.end;
802	sk_msg_iter_var_prev(i);
803	sg_mark_end(sg: sk_msg_elem(msg: msg_en, which: i));
804
805	i = msg_en->sg.start;
806	sg_chain(prv: rec->sg_aead_out, prv_nents: 2, sgl: &msg_en->sg.data[i]);
807
808	tls_make_aad(buf: rec->aad_space, size: msg_pl->sg.size + prot->tail_size,
809	record_sequence: tls_ctx->tx.rec_seq, record_type, prot);
810
811	tls_fill_prepend(ctx: tls_ctx,
812	page_address(sg_page(&msg_en->sg.data[i])) +
813	msg_en->sg.data[i].offset,
814	plaintext_len: msg_pl->sg.size + prot->tail_size,
815	record_type);
816
817	tls_ctx->pending_open_record_frags = false;
818
819	rc = tls_do_encryption(sk, tls_ctx, ctx, aead_req: req,
820	data_len: msg_pl->sg.size + prot->tail_size, start: i);
821	if (rc < 0) {
822	if (rc != -EINPROGRESS) {
823	tls_err_abort(sk, err: -EBADMSG);
824	if (split) {
825	tls_ctx->pending_open_record_frags = true;
826	tls_merge_open_record(sk, to: rec, from: tmp, orig_end);
827	}
828	}
829	ctx->async_capable = 1;
830	return rc;
831	} else if (split) {
832	msg_pl = &tmp->msg_plaintext;
833	msg_en = &tmp->msg_encrypted;
834	sk_msg_trim(sk, msg: msg_en, len: msg_pl->sg.size + prot->overhead_size);
835	tls_ctx->pending_open_record_frags = true;
836	ctx->open_rec = tmp;
837	}
838
839	return tls_tx_records(sk, flags);
840	}
841
842	static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
843	bool full_record, u8 record_type,
844	ssize_t *copied, int flags)
845	{
846	struct tls_context *tls_ctx = tls_get_ctx(sk);
847	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
848	struct sk_msg msg_redir = { };
849	struct sk_psock *psock;
850	struct sock *sk_redir;
851	struct tls_rec *rec;
852	bool enospc, policy, redir_ingress;
853	int err = 0, send;
854	u32 delta = 0;
855
856	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
857	psock = sk_psock_get(sk);
858	if (!psock || !policy) {
859	err = tls_push_record(sk, flags, record_type);
860	if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
861	*copied -= sk_msg_free(sk, msg);
862	tls_free_open_rec(sk);
863	err = -sk->sk_err;
864	}
865	if (psock)
866	sk_psock_put(sk, psock);
867	return err;
868	}
869	more_data:
870	enospc = sk_msg_full(msg);
871	if (psock->eval == __SK_NONE) {
872	delta = msg->sg.size;
873	psock->eval = sk_psock_msg_verdict(sk, psock, msg);
874	delta -= msg->sg.size;
875
876	if ((s32)delta > 0) {
877	/* It indicates that we executed bpf_msg_pop_data(),
878	* causing the plaintext data size to decrease.
879	* Therefore the encrypted data size also needs to
880	* correspondingly decrease. We only need to subtract
881	* delta to calculate the new ciphertext length since
882	* ktls does not support block encryption.
883	*/
884	struct sk_msg *enc = &ctx->open_rec->msg_encrypted;
885
886	sk_msg_trim(sk, msg: enc, len: enc->sg.size - delta);
887	}
888	}
889	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
890	!enospc && !full_record) {
891	err = -ENOSPC;
892	goto out_err;
893	}
894	msg->cork_bytes = 0;
895	send = msg->sg.size;
896	if (msg->apply_bytes && msg->apply_bytes < send)
897	send = msg->apply_bytes;
898
899	switch (psock->eval) {
900	case __SK_PASS:
901	err = tls_push_record(sk, flags, record_type);
902	if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
903	*copied -= sk_msg_free(sk, msg);
904	tls_free_open_rec(sk);
905	err = -sk->sk_err;
906	goto out_err;
907	}
908	break;
909	case __SK_REDIRECT:
910	redir_ingress = psock->redir_ingress;
911	sk_redir = psock->sk_redir;
912	memcpy(&msg_redir, msg, sizeof(*msg));
913	if (msg->apply_bytes < send)
914	msg->apply_bytes = 0;
915	else
916	msg->apply_bytes -= send;
917	sk_msg_return_zero(sk, msg, bytes: send);
918	msg->sg.size -= send;
919	release_sock(sk);
920	err = tcp_bpf_sendmsg_redir(sk: sk_redir, ingress: redir_ingress,
921	msg: &msg_redir, bytes: send, flags);
922	lock_sock(sk);
923	if (err < 0) {
924	/* Regardless of whether the data represented by
925	* msg_redir is sent successfully, we have already
926	* uncharged it via sk_msg_return_zero(). The
927	* msg->sg.size represents the remaining unprocessed
928	* data, which needs to be uncharged here.
929	*/
930	sk_mem_uncharge(sk, size: msg->sg.size);
931	*copied -= sk_msg_free_nocharge(sk, msg: &msg_redir);
932	msg->sg.size = 0;
933	}
934	if (msg->sg.size == 0)
935	tls_free_open_rec(sk);
936	break;
937	case __SK_DROP:
938	default:
939	sk_msg_free_partial(sk, msg, bytes: send);
940	if (msg->apply_bytes < send)
941	msg->apply_bytes = 0;
942	else
943	msg->apply_bytes -= send;
944	if (msg->sg.size == 0)
945	tls_free_open_rec(sk);
946	*copied -= (send + delta);
947	err = -EACCES;
948	}
949
950	if (likely(!err)) {
951	bool reset_eval = !ctx->open_rec;
952
953	rec = ctx->open_rec;
954	if (rec) {
955	msg = &rec->msg_plaintext;
956	if (!msg->apply_bytes)
957	reset_eval = true;
958	}
959	if (reset_eval) {
960	psock->eval = __SK_NONE;
961	if (psock->sk_redir) {
962	sock_put(sk: psock->sk_redir);
963	psock->sk_redir = NULL;
964	}
965	}
966	if (rec)
967	goto more_data;
968	}
969	out_err:
970	sk_psock_put(sk, psock);
971	return err;
972	}
973
974	static int tls_sw_push_pending_record(struct sock *sk, int flags)
975	{
976	struct tls_context *tls_ctx = tls_get_ctx(sk);
977	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
978	struct tls_rec *rec = ctx->open_rec;
979	struct sk_msg *msg_pl;
980	size_t copied;
981
982	if (!rec)
983	return 0;
984
985	msg_pl = &rec->msg_plaintext;
986	copied = msg_pl->sg.size;
987	if (!copied)
988	return 0;
989
990	return bpf_exec_tx_verdict(msg: msg_pl, sk, full_record: true, record_type: TLS_RECORD_TYPE_DATA,
991	copied: &copied, flags);
992	}
993
994	static int tls_sw_sendmsg_splice(struct sock *sk, struct msghdr *msg,
995	struct sk_msg *msg_pl, size_t try_to_copy,
996	ssize_t *copied)
997	{
998	struct page *page = NULL, **pages = &page;
999
1000	do {
1001	ssize_t part;
1002	size_t off;
1003
1004	part = iov_iter_extract_pages(i: &msg->msg_iter, pages: &pages,
1005	maxsize: try_to_copy, maxpages: 1, extraction_flags: 0, offset0: &off);
1006	if (part <= 0)
1007	return part ?: -EIO;
1008
1009	if (WARN_ON_ONCE(!sendpage_ok(page))) {
1010	iov_iter_revert(i: &msg->msg_iter, bytes: part);
1011	return -EIO;
1012	}
1013
1014	sk_msg_page_add(msg: msg_pl, page, len: part, offset: off);
1015	msg_pl->sg.copybreak = 0;
1016	msg_pl->sg.curr = msg_pl->sg.end;
1017	sk_mem_charge(sk, size: part);
1018	*copied += part;
1019	try_to_copy -= part;
1020	} while (try_to_copy && !sk_msg_full(msg: msg_pl));
1021
1022	return 0;
1023	}
1024
1025	static int tls_sw_sendmsg_locked(struct sock *sk, struct msghdr *msg,
1026	size_t size)
1027	{
1028	long timeo = sock_sndtimeo(sk, noblock: msg->msg_flags & MSG_DONTWAIT);
1029	struct tls_context *tls_ctx = tls_get_ctx(sk);
1030	struct tls_prot_info *prot = &tls_ctx->prot_info;
1031	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1032	bool async_capable = ctx->async_capable;
1033	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1034	bool is_kvec = iov_iter_is_kvec(i: &msg->msg_iter);
1035	bool eor = !(msg->msg_flags & MSG_MORE);
1036	size_t try_to_copy;
1037	ssize_t copied = 0;
1038	struct sk_msg *msg_pl, *msg_en;
1039	struct tls_rec *rec;
1040	int required_size;
1041	int num_async = 0;
1042	bool full_record;
1043	int record_room;
1044	int num_zc = 0;
1045	int orig_size;
1046	int ret = 0;
1047
1048	if (!eor && (msg->msg_flags & MSG_EOR))
1049	return -EINVAL;
1050
1051	if (unlikely(msg->msg_controllen)) {
1052	ret = tls_process_cmsg(sk, msg, record_type: &record_type);
1053	if (ret) {
1054	if (ret == -EINPROGRESS)
1055	num_async++;
1056	else if (ret != -EAGAIN)
1057	goto end;
1058	}
1059	}
1060
1061	while (msg_data_left(msg)) {
1062	if (sk->sk_err) {
1063	ret = -sk->sk_err;
1064	goto send_end;
1065	}
1066
1067	if (ctx->open_rec)
1068	rec = ctx->open_rec;
1069	else
1070	rec = ctx->open_rec = tls_get_rec(sk);
1071	if (!rec) {
1072	ret = -ENOMEM;
1073	goto send_end;
1074	}
1075
1076	msg_pl = &rec->msg_plaintext;
1077	msg_en = &rec->msg_encrypted;
1078
1079	orig_size = msg_pl->sg.size;
1080	full_record = false;
1081	try_to_copy = msg_data_left(msg);
1082	record_room = tls_ctx->tx_max_payload_len - msg_pl->sg.size;
1083	if (try_to_copy >= record_room) {
1084	try_to_copy = record_room;
1085	full_record = true;
1086	}
1087
1088	required_size = msg_pl->sg.size + try_to_copy +
1089	prot->overhead_size;
1090
1091	if (!sk_stream_memory_free(sk))
1092	goto wait_for_sndbuf;
1093
1094	alloc_encrypted:
1095	ret = tls_alloc_encrypted_msg(sk, len: required_size);
1096	if (ret) {
1097	if (ret != -ENOSPC)
1098	goto wait_for_memory;
1099
1100	/* Adjust try_to_copy according to the amount that was
1101	* actually allocated. The difference is due
1102	* to max sg elements limit
1103	*/
1104	try_to_copy -= required_size - msg_en->sg.size;
1105	full_record = true;
1106	}
1107
1108	if (try_to_copy && (msg->msg_flags & MSG_SPLICE_PAGES)) {
1109	ret = tls_sw_sendmsg_splice(sk, msg, msg_pl,
1110	try_to_copy, copied: &copied);
1111	if (ret < 0)
1112	goto send_end;
1113	tls_ctx->pending_open_record_frags = true;
1114
1115	if (sk_msg_full(msg: msg_pl)) {
1116	full_record = true;
1117	sk_msg_trim(sk, msg: msg_en,
1118	len: msg_pl->sg.size + prot->overhead_size);
1119	}
1120
1121	if (full_record || eor)
1122	goto copied;
1123	continue;
1124	}
1125
1126	if (!is_kvec && (full_record || eor) && !async_capable) {
1127	u32 first = msg_pl->sg.end;
1128
1129	ret = sk_msg_zerocopy_from_iter(sk, from: &msg->msg_iter,
1130	msg: msg_pl, bytes: try_to_copy);
1131	if (ret)
1132	goto fallback_to_reg_send;
1133
1134	num_zc++;
1135	copied += try_to_copy;
1136
1137	sk_msg_sg_copy_set(msg: msg_pl, start: first);
1138	ret = bpf_exec_tx_verdict(msg: msg_pl, sk, full_record,
1139	record_type, copied: &copied,
1140	flags: msg->msg_flags);
1141	if (ret) {
1142	if (ret == -EINPROGRESS)
1143	num_async++;
1144	else if (ret == -ENOMEM)
1145	goto wait_for_memory;
1146	else if (ctx->open_rec && ret == -ENOSPC) {
1147	if (msg_pl->cork_bytes) {
1148	ret = 0;
1149	goto send_end;
1150	}
1151	goto rollback_iter;
1152	} else if (ret != -EAGAIN)
1153	goto send_end;
1154	}
1155
1156	/* Transmit if any encryptions have completed */
1157	if (test_and_clear_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask)) {
1158	cancel_delayed_work(dwork: &ctx->tx_work.work);
1159	tls_tx_records(sk, flags: msg->msg_flags);
1160	}
1161
1162	continue;
1163	rollback_iter:
1164	copied -= try_to_copy;
1165	sk_msg_sg_copy_clear(msg: msg_pl, start: first);
1166	iov_iter_revert(i: &msg->msg_iter,
1167	bytes: msg_pl->sg.size - orig_size);
1168	fallback_to_reg_send:
1169	sk_msg_trim(sk, msg: msg_pl, len: orig_size);
1170	}
1171
1172	required_size = msg_pl->sg.size + try_to_copy;
1173
1174	ret = tls_clone_plaintext_msg(sk, required: required_size);
1175	if (ret) {
1176	if (ret != -ENOSPC)
1177	goto send_end;
1178
1179	/* Adjust try_to_copy according to the amount that was
1180	* actually allocated. The difference is due
1181	* to max sg elements limit
1182	*/
1183	try_to_copy -= required_size - msg_pl->sg.size;
1184	full_record = true;
1185	sk_msg_trim(sk, msg: msg_en,
1186	len: msg_pl->sg.size + prot->overhead_size);
1187	}
1188
1189	if (try_to_copy) {
1190	ret = sk_msg_memcopy_from_iter(sk, from: &msg->msg_iter,
1191	msg: msg_pl, bytes: try_to_copy);
1192	if (ret < 0)
1193	goto trim_sgl;
1194	}
1195
1196	/* Open records defined only if successfully copied, otherwise
1197	* we would trim the sg but not reset the open record frags.
1198	*/
1199	tls_ctx->pending_open_record_frags = true;
1200	copied += try_to_copy;
1201	copied:
1202	if (full_record || eor) {
1203	ret = bpf_exec_tx_verdict(msg: msg_pl, sk, full_record,
1204	record_type, copied: &copied,
1205	flags: msg->msg_flags);
1206	if (ret) {
1207	if (ret == -EINPROGRESS)
1208	num_async++;
1209	else if (ret == -ENOMEM)
1210	goto wait_for_memory;
1211	else if (ret != -EAGAIN) {
1212	if (ret == -ENOSPC)
1213	ret = 0;
1214	goto send_end;
1215	}
1216	}
1217
1218	/* Transmit if any encryptions have completed */
1219	if (test_and_clear_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask)) {
1220	cancel_delayed_work(dwork: &ctx->tx_work.work);
1221	tls_tx_records(sk, flags: msg->msg_flags);
1222	}
1223	}
1224
1225	continue;
1226
1227	wait_for_sndbuf:
1228	set_bit(nr: SOCK_NOSPACE, addr: &sk->sk_socket->flags);
1229	wait_for_memory:
1230	ret = sk_stream_wait_memory(sk, timeo_p: &timeo);
1231	if (ret) {
1232	trim_sgl:
1233	if (ctx->open_rec)
1234	tls_trim_both_msgs(sk, target_size: orig_size);
1235	goto send_end;
1236	}
1237
1238	if (ctx->open_rec && msg_en->sg.size < required_size)
1239	goto alloc_encrypted;
1240	}
1241
1242	send_end:
1243	if (!num_async) {
1244	goto end;
1245	} else if (num_zc || eor) {
1246	int err;
1247
1248	/* Wait for pending encryptions to get completed */
1249	err = tls_encrypt_async_wait(ctx);
1250	if (err) {
1251	ret = err;
1252	copied = 0;
1253	}
1254	}
1255
1256	/* Transmit if any encryptions have completed */
1257	if (test_and_clear_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask)) {
1258	cancel_delayed_work(dwork: &ctx->tx_work.work);
1259	tls_tx_records(sk, flags: msg->msg_flags);
1260	}
1261
1262	end:
1263	ret = sk_stream_error(sk, flags: msg->msg_flags, err: ret);
1264	return copied > 0 ? copied : ret;
1265	}
1266
1267	int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1268	{
1269	struct tls_context *tls_ctx = tls_get_ctx(sk);
1270	int ret;
1271
1272	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273	MSG_CMSG_COMPAT | MSG_SPLICE_PAGES | MSG_EOR |
1274	MSG_SENDPAGE_NOPOLICY))
1275	return -EOPNOTSUPP;
1276
1277	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1278	if (ret)
1279	return ret;
1280	lock_sock(sk);
1281	ret = tls_sw_sendmsg_locked(sk, msg, size);
1282	release_sock(sk);
1283	mutex_unlock(lock: &tls_ctx->tx_lock);
1284	return ret;
1285	}
1286
1287	/*
1288	* Handle unexpected EOF during splice without SPLICE_F_MORE set.
1289	*/
1290	void tls_sw_splice_eof(struct socket *sock)
1291	{
1292	struct sock *sk = sock->sk;
1293	struct tls_context *tls_ctx = tls_get_ctx(sk);
1294	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1295	struct tls_rec *rec;
1296	struct sk_msg *msg_pl;
1297	ssize_t copied = 0;
1298	bool retrying = false;
1299	int ret = 0;
1300
1301	if (!ctx->open_rec)
1302	return;
1303
1304	mutex_lock(&tls_ctx->tx_lock);
1305	lock_sock(sk);
1306
1307	retry:
1308	/* same checks as in tls_sw_push_pending_record() */
1309	rec = ctx->open_rec;
1310	if (!rec)
1311	goto unlock;
1312
1313	msg_pl = &rec->msg_plaintext;
1314	if (msg_pl->sg.size == 0)
1315	goto unlock;
1316
1317	/* Check the BPF advisor and perform transmission. */
1318	ret = bpf_exec_tx_verdict(msg: msg_pl, sk, full_record: false, record_type: TLS_RECORD_TYPE_DATA,
1319	copied: &copied, flags: 0);
1320	switch (ret) {
1321	case 0:
1322	case -EAGAIN:
1323	if (retrying)
1324	goto unlock;
1325	retrying = true;
1326	goto retry;
1327	case -EINPROGRESS:
1328	break;
1329	default:
1330	goto unlock;
1331	}
1332
1333	/* Wait for pending encryptions to get completed */
1334	if (tls_encrypt_async_wait(ctx))
1335	goto unlock;
1336
1337	/* Transmit if any encryptions have completed */
1338	if (test_and_clear_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask)) {
1339	cancel_delayed_work(dwork: &ctx->tx_work.work);
1340	tls_tx_records(sk, flags: 0);
1341	}
1342
1343	unlock:
1344	release_sock(sk);
1345	mutex_unlock(lock: &tls_ctx->tx_lock);
1346	}
1347
1348	static int
1349	tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
1350	bool released)
1351	{
1352	struct tls_context *tls_ctx = tls_get_ctx(sk);
1353	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1354	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1355	int ret = 0;
1356	long timeo;
1357
1358	/* a rekey is pending, let userspace deal with it */
1359	if (unlikely(ctx->key_update_pending))
1360	return -EKEYEXPIRED;
1361
1362	timeo = sock_rcvtimeo(sk, noblock: nonblock);
1363
1364	while (!tls_strp_msg_ready(ctx)) {
1365	if (!sk_psock_queue_empty(psock))
1366	return 0;
1367
1368	if (sk->sk_err)
1369	return sock_error(sk);
1370
1371	if (ret < 0)
1372	return ret;
1373
1374	if (!skb_queue_empty(list: &sk->sk_receive_queue)) {
1375	tls_strp_check_rcv(strp: &ctx->strp);
1376	if (tls_strp_msg_ready(ctx))
1377	break;
1378	}
1379
1380	if (sk->sk_shutdown & RCV_SHUTDOWN)
1381	return 0;
1382
1383	if (sock_flag(sk, flag: SOCK_DONE))
1384	return 0;
1385
1386	if (!timeo)
1387	return -EAGAIN;
1388
1389	released = true;
1390	add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1391	sk_set_bit(nr: SOCKWQ_ASYNC_WAITDATA, sk);
1392	ret = sk_wait_event(sk, &timeo,
1393	tls_strp_msg_ready(ctx) ||
1394	!sk_psock_queue_empty(psock),
1395	&wait);
1396	sk_clear_bit(nr: SOCKWQ_ASYNC_WAITDATA, sk);
1397	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1398
1399	/* Handle signals */
1400	if (signal_pending(current))
1401	return sock_intr_errno(timeo);
1402	}
1403
1404	if (unlikely(!tls_strp_msg_load(&ctx->strp, released)))
1405	return tls_rx_rec_wait(sk, psock, nonblock, released: false);
1406
1407	return 1;
1408	}
1409
1410	static int tls_setup_from_iter(struct iov_iter *from,
1411	int length, int *pages_used,
1412	struct scatterlist *to,
1413	int to_max_pages)
1414	{
1415	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1416	struct page *pages[MAX_SKB_FRAGS];
1417	unsigned int size = 0;
1418	ssize_t copied, use;
1419	size_t offset;
1420
1421	while (length > 0) {
1422	i = 0;
1423	maxpages = to_max_pages - num_elem;
1424	if (maxpages == 0) {
1425	rc = -EFAULT;
1426	goto out;
1427	}
1428	copied = iov_iter_get_pages2(i: from, pages,
1429	maxsize: length,
1430	maxpages, start: &offset);
1431	if (copied <= 0) {
1432	rc = -EFAULT;
1433	goto out;
1434	}
1435
1436	length -= copied;
1437	size += copied;
1438	while (copied) {
1439	use = min_t(int, copied, PAGE_SIZE - offset);
1440
1441	sg_set_page(sg: &to[num_elem],
1442	page: pages[i], len: use, offset);
1443	sg_unmark_end(sg: &to[num_elem]);
1444	/* We do not uncharge memory from this API */
1445
1446	offset = 0;
1447	copied -= use;
1448
1449	i++;
1450	num_elem++;
1451	}
1452	}
1453	/* Mark the end in the last sg entry if newly added */
1454	if (num_elem > *pages_used)
1455	sg_mark_end(sg: &to[num_elem - 1]);
1456	out:
1457	if (rc)
1458	iov_iter_revert(i: from, bytes: size);
1459	*pages_used = num_elem;
1460
1461	return rc;
1462	}
1463
1464	static struct sk_buff *
1465	tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1466	unsigned int full_len)
1467	{
1468	struct strp_msg *clr_rxm;
1469	struct sk_buff *clr_skb;
1470	int err;
1471
1472	clr_skb = alloc_skb_with_frags(header_len: 0, data_len: full_len, TLS_PAGE_ORDER,
1473	errcode: &err, gfp_mask: sk->sk_allocation);
1474	if (!clr_skb)
1475	return NULL;
1476
1477	skb_copy_header(new: clr_skb, old: skb);
1478	clr_skb->len = full_len;
1479	clr_skb->data_len = full_len;
1480
1481	clr_rxm = strp_msg(skb: clr_skb);
1482	clr_rxm->offset = 0;
1483
1484	return clr_skb;
1485	}
1486
1487	/* Decrypt handlers
1488	*
1489	* tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1490	* They must transform the darg in/out argument are as follows:
1491	* | Input | Output
1492	* -------------------------------------------------------------------
1493	* zc | Zero-copy decrypt allowed | Zero-copy performed
1494	* async | Async decrypt allowed | Async crypto used / in progress
1495	* skb | * | Output skb
1496	*
1497	* If ZC decryption was performed darg.skb will point to the input skb.
1498	*/
1499
1500	/* This function decrypts the input skb into either out_iov or in out_sg
1501	* or in skb buffers itself. The input parameter 'darg->zc' indicates if
1502	* zero-copy mode needs to be tried or not. With zero-copy mode, either
1503	* out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1504	* NULL, then the decryption happens inside skb buffers itself, i.e.
1505	* zero-copy gets disabled and 'darg->zc' is updated.
1506	*/
1507	static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1508	struct scatterlist *out_sg,
1509	struct tls_decrypt_arg *darg)
1510	{
1511	struct tls_context *tls_ctx = tls_get_ctx(sk);
1512	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1513	struct tls_prot_info *prot = &tls_ctx->prot_info;
1514	int n_sgin, n_sgout, aead_size, err, pages = 0;
1515	struct sk_buff *skb = tls_strp_msg(ctx);
1516	const struct strp_msg *rxm = strp_msg(skb);
1517	const struct tls_msg *tlm = tls_msg(skb);
1518	struct aead_request *aead_req;
1519	struct scatterlist *sgin = NULL;
1520	struct scatterlist *sgout = NULL;
1521	const int data_len = rxm->full_len - prot->overhead_size;
1522	int tail_pages = !!prot->tail_size;
1523	struct tls_decrypt_ctx *dctx;
1524	struct sk_buff *clear_skb;
1525	int iv_offset = 0;
1526	u8 *mem;
1527
1528	n_sgin = skb_nsg(skb, offset: rxm->offset + prot->prepend_size,
1529	len: rxm->full_len - prot->prepend_size);
1530	if (n_sgin < 1)
1531	return n_sgin ?: -EBADMSG;
1532
1533	if (darg->zc && (out_iov || out_sg)) {
1534	clear_skb = NULL;
1535
1536	if (out_iov)
1537	n_sgout = 1 + tail_pages +
1538	iov_iter_npages_cap(i: out_iov, INT_MAX, max_bytes: data_len);
1539	else
1540	n_sgout = sg_nents(sg: out_sg);
1541	} else {
1542	darg->zc = false;
1543
1544	clear_skb = tls_alloc_clrtxt_skb(sk, skb, full_len: rxm->full_len);
1545	if (!clear_skb)
1546	return -ENOMEM;
1547
1548	n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1549	}
1550
1551	/* Increment to accommodate AAD */
1552	n_sgin = n_sgin + 1;
1553
1554	/* Allocate a single block of memory which contains
1555	* aead_req || tls_decrypt_ctx.
1556	* Both structs are variable length.
1557	*/
1558	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(tfm: ctx->aead_recv);
1559	aead_size = ALIGN(aead_size, __alignof__(*dctx));
1560	mem = kmalloc(aead_size + struct_size(dctx, sg, size_add(n_sgin, n_sgout)),
1561	sk->sk_allocation);
1562	if (!mem) {
1563	err = -ENOMEM;
1564	goto exit_free_skb;
1565	}
1566
1567	/* Segment the allocated memory */
1568	aead_req = (struct aead_request *)mem;
1569	dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1570	dctx->sk = sk;
1571	sgin = &dctx->sg[0];
1572	sgout = &dctx->sg[n_sgin];
1573
1574	/* For CCM based ciphers, first byte of nonce+iv is a constant */
1575	switch (prot->cipher_type) {
1576	case TLS_CIPHER_AES_CCM_128:
1577	dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1578	iv_offset = 1;
1579	break;
1580	case TLS_CIPHER_SM4_CCM:
1581	dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1582	iv_offset = 1;
1583	break;
1584	}
1585
1586	/* Prepare IV */
1587	if (prot->version == TLS_1_3_VERSION ||
1588	prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1589	memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1590	prot->iv_size + prot->salt_size);
1591	} else {
1592	err = skb_copy_bits(skb, offset: rxm->offset + TLS_HEADER_SIZE,
1593	to: &dctx->iv[iv_offset] + prot->salt_size,
1594	len: prot->iv_size);
1595	if (err < 0)
1596	goto exit_free;
1597	memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1598	}
1599	tls_xor_iv_with_seq(prot, iv: &dctx->iv[iv_offset], seq: tls_ctx->rx.rec_seq);
1600
1601	/* Prepare AAD */
1602	tls_make_aad(buf: dctx->aad, size: rxm->full_len - prot->overhead_size +
1603	prot->tail_size,
1604	record_sequence: tls_ctx->rx.rec_seq, record_type: tlm->control, prot);
1605
1606	/* Prepare sgin */
1607	sg_init_table(sgin, n_sgin);
1608	sg_set_buf(sg: &sgin[0], buf: dctx->aad, buflen: prot->aad_size);
1609	err = skb_to_sgvec(skb, sg: &sgin[1],
1610	offset: rxm->offset + prot->prepend_size,
1611	len: rxm->full_len - prot->prepend_size);
1612	if (err < 0)
1613	goto exit_free;
1614
1615	if (clear_skb) {
1616	sg_init_table(sgout, n_sgout);
1617	sg_set_buf(sg: &sgout[0], buf: dctx->aad, buflen: prot->aad_size);
1618
1619	err = skb_to_sgvec(skb: clear_skb, sg: &sgout[1], offset: prot->prepend_size,
1620	len: data_len + prot->tail_size);
1621	if (err < 0)
1622	goto exit_free;
1623	} else if (out_iov) {
1624	sg_init_table(sgout, n_sgout);
1625	sg_set_buf(sg: &sgout[0], buf: dctx->aad, buflen: prot->aad_size);
1626
1627	err = tls_setup_from_iter(from: out_iov, length: data_len, pages_used: &pages, to: &sgout[1],
1628	to_max_pages: (n_sgout - 1 - tail_pages));
1629	if (err < 0)
1630	goto exit_free_pages;
1631
1632	if (prot->tail_size) {
1633	sg_unmark_end(sg: &sgout[pages]);
1634	sg_set_buf(sg: &sgout[pages + 1], buf: &dctx->tail,
1635	buflen: prot->tail_size);
1636	sg_mark_end(sg: &sgout[pages + 1]);
1637	}
1638	} else if (out_sg) {
1639	memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1640	}
1641	dctx->free_sgout = !!pages;
1642
1643	/* Prepare and submit AEAD request */
1644	err = tls_do_decryption(sk, sgin, sgout, iv_recv: dctx->iv,
1645	data_len: data_len + prot->tail_size, aead_req, darg);
1646	if (err) {
1647	if (darg->async_done)
1648	goto exit_free_skb;
1649	goto exit_free_pages;
1650	}
1651
1652	darg->skb = clear_skb ?: tls_strp_msg(ctx);
1653	clear_skb = NULL;
1654
1655	if (unlikely(darg->async)) {
1656	err = tls_strp_msg_hold(strp: &ctx->strp, dst: &ctx->async_hold);
1657	if (err) {
1658	err = tls_decrypt_async_wait(ctx);
1659	darg->async = false;
1660	}
1661	return err;
1662	}
1663
1664	if (unlikely(darg->async_done))
1665	return 0;
1666
1667	if (prot->tail_size)
1668	darg->tail = dctx->tail;
1669
1670	exit_free_pages:
1671	/* Release the pages in case iov was mapped to pages */
1672	for (; pages > 0; pages--)
1673	put_page(page: sg_page(sg: &sgout[pages]));
1674	exit_free:
1675	kfree(objp: mem);
1676	exit_free_skb:
1677	consume_skb(skb: clear_skb);
1678	return err;
1679	}
1680
1681	static int
1682	tls_decrypt_sw(struct sock *sk, struct tls_context *tls_ctx,
1683	struct msghdr *msg, struct tls_decrypt_arg *darg)
1684	{
1685	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1686	struct tls_prot_info *prot = &tls_ctx->prot_info;
1687	struct strp_msg *rxm;
1688	int pad, err;
1689
1690	err = tls_decrypt_sg(sk, out_iov: &msg->msg_iter, NULL, darg);
1691	if (err < 0) {
1692	if (err == -EBADMSG)
1693	TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1694	return err;
1695	}
1696	/* keep going even for ->async, the code below is TLS 1.3 */
1697
1698	/* If opportunistic TLS 1.3 ZC failed retry without ZC */
1699	if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1700	darg->tail != TLS_RECORD_TYPE_DATA)) {
1701	darg->zc = false;
1702	if (!darg->tail)
1703	TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1704	TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1705	return tls_decrypt_sw(sk, tls_ctx, msg, darg);
1706	}
1707
1708	pad = tls_padding_length(prot, skb: darg->skb, darg);
1709	if (pad < 0) {
1710	if (darg->skb != tls_strp_msg(ctx))
1711	consume_skb(skb: darg->skb);
1712	return pad;
1713	}
1714
1715	rxm = strp_msg(skb: darg->skb);
1716	rxm->full_len -= pad;
1717
1718	return 0;
1719	}
1720
1721	static int
1722	tls_decrypt_device(struct sock *sk, struct msghdr *msg,
1723	struct tls_context *tls_ctx, struct tls_decrypt_arg *darg)
1724	{
1725	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1726	struct tls_prot_info *prot = &tls_ctx->prot_info;
1727	struct strp_msg *rxm;
1728	int pad, err;
1729
1730	if (tls_ctx->rx_conf != TLS_HW)
1731	return 0;
1732
1733	err = tls_device_decrypted(sk, tls_ctx);
1734	if (err <= 0)
1735	return err;
1736
1737	pad = tls_padding_length(prot, skb: tls_strp_msg(ctx), darg);
1738	if (pad < 0)
1739	return pad;
1740
1741	darg->async = false;
1742	darg->skb = tls_strp_msg(ctx);
1743	/* ->zc downgrade check, in case TLS 1.3 gets here */
1744	darg->zc &= !(prot->version == TLS_1_3_VERSION &&
1745	tls_msg(skb: darg->skb)->control != TLS_RECORD_TYPE_DATA);
1746
1747	rxm = strp_msg(skb: darg->skb);
1748	rxm->full_len -= pad;
1749
1750	if (!darg->zc) {
1751	/* Non-ZC case needs a real skb */
1752	darg->skb = tls_strp_msg_detach(ctx);
1753	if (!darg->skb)
1754	return -ENOMEM;
1755	} else {
1756	unsigned int off, len;
1757
1758	/* In ZC case nobody cares about the output skb.
1759	* Just copy the data here. Note the skb is not fully trimmed.
1760	*/
1761	off = rxm->offset + prot->prepend_size;
1762	len = rxm->full_len - prot->overhead_size;
1763
1764	err = skb_copy_datagram_msg(from: darg->skb, offset: off, msg, size: len);
1765	if (err)
1766	return err;
1767	}
1768	return 1;
1769	}
1770
1771	static int tls_check_pending_rekey(struct sock *sk, struct tls_context *ctx,
1772	struct sk_buff *skb)
1773	{
1774	const struct strp_msg *rxm = strp_msg(skb);
1775	const struct tls_msg *tlm = tls_msg(skb);
1776	char hs_type;
1777	int err;
1778
1779	if (likely(tlm->control != TLS_RECORD_TYPE_HANDSHAKE))
1780	return 0;
1781
1782	if (rxm->full_len < 1)
1783	return 0;
1784
1785	err = skb_copy_bits(skb, offset: rxm->offset, to: &hs_type, len: 1);
1786	if (err < 0) {
1787	DEBUG_NET_WARN_ON_ONCE(1);
1788	return err;
1789	}
1790
1791	if (hs_type == TLS_HANDSHAKE_KEYUPDATE) {
1792	struct tls_sw_context_rx *rx_ctx = ctx->priv_ctx_rx;
1793
1794	WRITE_ONCE(rx_ctx->key_update_pending, true);
1795	TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXREKEYRECEIVED);
1796	}
1797
1798	return 0;
1799	}
1800
1801	static int tls_rx_one_record(struct sock *sk, struct msghdr *msg,
1802	struct tls_decrypt_arg *darg)
1803	{
1804	struct tls_context *tls_ctx = tls_get_ctx(sk);
1805	struct tls_prot_info *prot = &tls_ctx->prot_info;
1806	struct strp_msg *rxm;
1807	int err;
1808
1809	err = tls_decrypt_device(sk, msg, tls_ctx, darg);
1810	if (!err)
1811	err = tls_decrypt_sw(sk, tls_ctx, msg, darg);
1812	if (err < 0)
1813	return err;
1814
1815	rxm = strp_msg(skb: darg->skb);
1816	rxm->offset += prot->prepend_size;
1817	rxm->full_len -= prot->overhead_size;
1818	tls_advance_record_sn(sk, prot, ctx: &tls_ctx->rx);
1819
1820	return tls_check_pending_rekey(sk, ctx: tls_ctx, skb: darg->skb);
1821	}
1822
1823	int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
1824	{
1825	struct tls_decrypt_arg darg = { .zc = true, };
1826
1827	return tls_decrypt_sg(sk, NULL, out_sg: sgout, darg: &darg);
1828	}
1829
1830	/* All records returned from a recvmsg() call must have the same type.
1831	* 0 is not a valid content type. Use it as "no type reported, yet".
1832	*/
1833	static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1834	u8 *control)
1835	{
1836	int err;
1837
1838	if (!*control) {
1839	*control = tlm->control;
1840	if (!*control)
1841	return -EBADMSG;
1842
1843	err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1844	len: sizeof(*control), data: control);
1845	if (*control != TLS_RECORD_TYPE_DATA) {
1846	if (err || msg->msg_flags & MSG_CTRUNC)
1847	return -EIO;
1848	}
1849	} else if (*control != tlm->control) {
1850	return 0;
1851	}
1852
1853	return 1;
1854	}
1855
1856	static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1857	{
1858	tls_strp_msg_done(strp: &ctx->strp);
1859	}
1860
1861	/* This function traverses the rx_list in tls receive context to copies the
1862	* decrypted records into the buffer provided by caller zero copy is not
1863	* true. Further, the records are removed from the rx_list if it is not a peek
1864	* case and the record has been consumed completely.
1865	*/
1866	static int process_rx_list(struct tls_sw_context_rx *ctx,
1867	struct msghdr *msg,
1868	u8 *control,
1869	size_t skip,
1870	size_t len,
1871	bool is_peek,
1872	bool *more)
1873	{
1874	struct sk_buff *skb = skb_peek(list_: &ctx->rx_list);
1875	struct tls_msg *tlm;
1876	ssize_t copied = 0;
1877	int err;
1878
1879	while (skip && skb) {
1880	struct strp_msg *rxm = strp_msg(skb);
1881	tlm = tls_msg(skb);
1882
1883	err = tls_record_content_type(msg, tlm, control);
1884	if (err <= 0)
1885	goto more;
1886
1887	if (skip < rxm->full_len)
1888	break;
1889
1890	skip = skip - rxm->full_len;
1891	skb = skb_peek_next(skb, list_: &ctx->rx_list);
1892	}
1893
1894	while (len && skb) {
1895	struct sk_buff *next_skb;
1896	struct strp_msg *rxm = strp_msg(skb);
1897	int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1898
1899	tlm = tls_msg(skb);
1900
1901	err = tls_record_content_type(msg, tlm, control);
1902	if (err <= 0)
1903	goto more;
1904
1905	err = skb_copy_datagram_msg(from: skb, offset: rxm->offset + skip,
1906	msg, size: chunk);
1907	if (err < 0)
1908	goto more;
1909
1910	len = len - chunk;
1911	copied = copied + chunk;
1912
1913	/* Consume the data from record if it is non-peek case*/
1914	if (!is_peek) {
1915	rxm->offset = rxm->offset + chunk;
1916	rxm->full_len = rxm->full_len - chunk;
1917
1918	/* Return if there is unconsumed data in the record */
1919	if (rxm->full_len - skip)
1920	break;
1921	}
1922
1923	/* The remaining skip-bytes must lie in 1st record in rx_list.
1924	* So from the 2nd record, 'skip' should be 0.
1925	*/
1926	skip = 0;
1927
1928	if (msg)
1929	msg->msg_flags |= MSG_EOR;
1930
1931	next_skb = skb_peek_next(skb, list_: &ctx->rx_list);
1932
1933	if (!is_peek) {
1934	__skb_unlink(skb, list: &ctx->rx_list);
1935	consume_skb(skb);
1936	}
1937
1938	skb = next_skb;
1939	}
1940	err = 0;
1941
1942	out:
1943	return copied ? : err;
1944	more:
1945	if (more)
1946	*more = true;
1947	goto out;
1948	}
1949
1950	static bool
1951	tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1952	size_t len_left, size_t decrypted, ssize_t done,
1953	size_t *flushed_at)
1954	{
1955	size_t max_rec;
1956
1957	if (len_left <= decrypted)
1958	return false;
1959
1960	max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1961	if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1962	return false;
1963
1964	*flushed_at = done;
1965	return sk_flush_backlog(sk);
1966	}
1967
1968	static int tls_rx_reader_acquire(struct sock *sk, struct tls_sw_context_rx *ctx,
1969	bool nonblock)
1970	{
1971	long timeo;
1972	int ret;
1973
1974	timeo = sock_rcvtimeo(sk, noblock: nonblock);
1975
1976	while (unlikely(ctx->reader_present)) {
1977	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1978
1979	ctx->reader_contended = 1;
1980
1981	add_wait_queue(wq_head: &ctx->wq, wq_entry: &wait);
1982	ret = sk_wait_event(sk, &timeo,
1983	!READ_ONCE(ctx->reader_present), &wait);
1984	remove_wait_queue(wq_head: &ctx->wq, wq_entry: &wait);
1985
1986	if (timeo <= 0)
1987	return -EAGAIN;
1988	if (signal_pending(current))
1989	return sock_intr_errno(timeo);
1990	if (ret < 0)
1991	return ret;
1992	}
1993
1994	WRITE_ONCE(ctx->reader_present, 1);
1995
1996	return 0;
1997	}
1998
1999	static int tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
2000	bool nonblock)
2001	{
2002	int err;
2003
2004	lock_sock(sk);
2005	err = tls_rx_reader_acquire(sk, ctx, nonblock);
2006	if (err)
2007	release_sock(sk);
2008	return err;
2009	}
2010
2011	static void tls_rx_reader_release(struct sock *sk, struct tls_sw_context_rx *ctx)
2012	{
2013	if (unlikely(ctx->reader_contended)) {
2014	if (wq_has_sleeper(wq_head: &ctx->wq))
2015	wake_up(&ctx->wq);
2016	else
2017	ctx->reader_contended = 0;
2018
2019	WARN_ON_ONCE(!ctx->reader_present);
2020	}
2021
2022	WRITE_ONCE(ctx->reader_present, 0);
2023	}
2024
2025	static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
2026	{
2027	tls_rx_reader_release(sk, ctx);
2028	release_sock(sk);
2029	}
2030
2031	int tls_sw_recvmsg(struct sock *sk,
2032	struct msghdr *msg,
2033	size_t len,
2034	int flags,
2035	int *addr_len)
2036	{
2037	struct tls_context *tls_ctx = tls_get_ctx(sk);
2038	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2039	struct tls_prot_info *prot = &tls_ctx->prot_info;
2040	ssize_t decrypted = 0, async_copy_bytes = 0;
2041	struct sk_psock *psock;
2042	unsigned char control = 0;
2043	size_t flushed_at = 0;
2044	struct strp_msg *rxm;
2045	struct tls_msg *tlm;
2046	ssize_t copied = 0;
2047	ssize_t peeked = 0;
2048	bool async = false;
2049	int target, err;
2050	bool is_kvec = iov_iter_is_kvec(i: &msg->msg_iter);
2051	bool is_peek = flags & MSG_PEEK;
2052	bool rx_more = false;
2053	bool released = true;
2054	bool bpf_strp_enabled;
2055	bool zc_capable;
2056
2057	if (unlikely(flags & MSG_ERRQUEUE))
2058	return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
2059
2060	err = tls_rx_reader_lock(sk, ctx, nonblock: flags & MSG_DONTWAIT);
2061	if (err < 0)
2062	return err;
2063	psock = sk_psock_get(sk);
2064	bpf_strp_enabled = sk_psock_strp_enabled(psock);
2065
2066	/* If crypto failed the connection is broken */
2067	err = ctx->async_wait.err;
2068	if (err)
2069	goto end;
2070
2071	/* Process pending decrypted records. It must be non-zero-copy */
2072	err = process_rx_list(ctx, msg, control: &control, skip: 0, len, is_peek, more: &rx_more);
2073	if (err < 0)
2074	goto end;
2075
2076	/* process_rx_list() will set @control if it processed any records */
2077	copied = err;
2078	if (len <= copied || rx_more ||
2079	(control && control != TLS_RECORD_TYPE_DATA))
2080	goto end;
2081
2082	target = sock_rcvlowat(sk, waitall: flags & MSG_WAITALL, len);
2083	len = len - copied;
2084
2085	zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
2086	ctx->zc_capable;
2087	decrypted = 0;
2088	while (len && (decrypted + copied < target || tls_strp_msg_ready(ctx))) {
2089	struct tls_decrypt_arg darg;
2090	int to_decrypt, chunk;
2091
2092	err = tls_rx_rec_wait(sk, psock, nonblock: flags & MSG_DONTWAIT,
2093	released);
2094	if (err <= 0) {
2095	if (psock) {
2096	chunk = sk_msg_recvmsg(sk, psock, msg, len,
2097	flags);
2098	if (chunk > 0) {
2099	decrypted += chunk;
2100	len -= chunk;
2101	continue;
2102	}
2103	}
2104	goto recv_end;
2105	}
2106
2107	memset(&darg.inargs, 0, sizeof(darg.inargs));
2108
2109	rxm = strp_msg(skb: tls_strp_msg(ctx));
2110	tlm = tls_msg(skb: tls_strp_msg(ctx));
2111
2112	to_decrypt = rxm->full_len - prot->overhead_size;
2113
2114	if (zc_capable && to_decrypt <= len &&
2115	tlm->control == TLS_RECORD_TYPE_DATA)
2116	darg.zc = true;
2117
2118	/* Do not use async mode if record is non-data */
2119	if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
2120	darg.async = ctx->async_capable;
2121	else
2122	darg.async = false;
2123
2124	err = tls_rx_one_record(sk, msg, darg: &darg);
2125	if (err < 0) {
2126	tls_err_abort(sk, err: -EBADMSG);
2127	goto recv_end;
2128	}
2129
2130	async |= darg.async;
2131
2132	/* If the type of records being processed is not known yet,
2133	* set it to record type just dequeued. If it is already known,
2134	* but does not match the record type just dequeued, go to end.
2135	* We always get record type here since for tls1.2, record type
2136	* is known just after record is dequeued from stream parser.
2137	* For tls1.3, we disable async.
2138	*/
2139	err = tls_record_content_type(msg, tlm: tls_msg(skb: darg.skb), control: &control);
2140	if (err <= 0) {
2141	DEBUG_NET_WARN_ON_ONCE(darg.zc);
2142	tls_rx_rec_done(ctx);
2143	put_on_rx_list_err:
2144	__skb_queue_tail(list: &ctx->rx_list, newsk: darg.skb);
2145	goto recv_end;
2146	}
2147
2148	/* periodically flush backlog, and feed strparser */
2149	released = tls_read_flush_backlog(sk, prot, len_left: len, decrypted: to_decrypt,
2150	done: decrypted + copied,
2151	flushed_at: &flushed_at);
2152
2153	/* TLS 1.3 may have updated the length by more than overhead */
2154	rxm = strp_msg(skb: darg.skb);
2155	chunk = rxm->full_len;
2156	tls_rx_rec_done(ctx);
2157
2158	if (!darg.zc) {
2159	bool partially_consumed = chunk > len;
2160	struct sk_buff *skb = darg.skb;
2161
2162	DEBUG_NET_WARN_ON_ONCE(darg.skb == ctx->strp.anchor);
2163
2164	if (async) {
2165	/* TLS 1.2-only, to_decrypt must be text len */
2166	chunk = min_t(int, to_decrypt, len);
2167	async_copy_bytes += chunk;
2168	put_on_rx_list:
2169	decrypted += chunk;
2170	len -= chunk;
2171	__skb_queue_tail(list: &ctx->rx_list, newsk: skb);
2172	if (unlikely(control != TLS_RECORD_TYPE_DATA))
2173	break;
2174	continue;
2175	}
2176
2177	if (bpf_strp_enabled) {
2178	released = true;
2179	err = sk_psock_tls_strp_read(psock, skb);
2180	if (err != __SK_PASS) {
2181	rxm->offset = rxm->offset + rxm->full_len;
2182	rxm->full_len = 0;
2183	if (err == __SK_DROP)
2184	consume_skb(skb);
2185	continue;
2186	}
2187	}
2188
2189	if (partially_consumed)
2190	chunk = len;
2191
2192	err = skb_copy_datagram_msg(from: skb, offset: rxm->offset,
2193	msg, size: chunk);
2194	if (err < 0)
2195	goto put_on_rx_list_err;
2196
2197	if (is_peek) {
2198	peeked += chunk;
2199	goto put_on_rx_list;
2200	}
2201
2202	if (partially_consumed) {
2203	rxm->offset += chunk;
2204	rxm->full_len -= chunk;
2205	goto put_on_rx_list;
2206	}
2207
2208	consume_skb(skb);
2209	}
2210
2211	decrypted += chunk;
2212	len -= chunk;
2213
2214	/* Return full control message to userspace before trying
2215	* to parse another message type
2216	*/
2217	msg->msg_flags |= MSG_EOR;
2218	if (control != TLS_RECORD_TYPE_DATA)
2219	break;
2220	}
2221
2222	recv_end:
2223	if (async) {
2224	int ret;
2225
2226	/* Wait for all previously submitted records to be decrypted */
2227	ret = tls_decrypt_async_wait(ctx);
2228	__skb_queue_purge(list: &ctx->async_hold);
2229
2230	if (ret) {
2231	if (err >= 0 || err == -EINPROGRESS)
2232	err = ret;
2233	goto end;
2234	}
2235
2236	/* Drain records from the rx_list & copy if required */
2237	if (is_peek)
2238	err = process_rx_list(ctx, msg, control: &control, skip: copied + peeked,
2239	len: decrypted - peeked, is_peek, NULL);
2240	else
2241	err = process_rx_list(ctx, msg, control: &control, skip: 0,
2242	len: async_copy_bytes, is_peek, NULL);
2243
2244	/* we could have copied less than we wanted, and possibly nothing */
2245	decrypted += max(err, 0) - async_copy_bytes;
2246	}
2247
2248	copied += decrypted;
2249
2250	end:
2251	tls_rx_reader_unlock(sk, ctx);
2252	if (psock)
2253	sk_psock_put(sk, psock);
2254	return copied ? : err;
2255	}
2256
2257	ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
2258	struct pipe_inode_info *pipe,
2259	size_t len, unsigned int flags)
2260	{
2261	struct tls_context *tls_ctx = tls_get_ctx(sk: sock->sk);
2262	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2263	struct strp_msg *rxm = NULL;
2264	struct sock *sk = sock->sk;
2265	struct tls_msg *tlm;
2266	struct sk_buff *skb;
2267	ssize_t copied = 0;
2268	int chunk;
2269	int err;
2270
2271	err = tls_rx_reader_lock(sk, ctx, nonblock: flags & SPLICE_F_NONBLOCK);
2272	if (err < 0)
2273	return err;
2274
2275	if (!skb_queue_empty(list: &ctx->rx_list)) {
2276	skb = __skb_dequeue(list: &ctx->rx_list);
2277	} else {
2278	struct tls_decrypt_arg darg;
2279
2280	err = tls_rx_rec_wait(sk, NULL, nonblock: flags & SPLICE_F_NONBLOCK,
2281	released: true);
2282	if (err <= 0)
2283	goto splice_read_end;
2284
2285	memset(&darg.inargs, 0, sizeof(darg.inargs));
2286
2287	err = tls_rx_one_record(sk, NULL, darg: &darg);
2288	if (err < 0) {
2289	tls_err_abort(sk, err: -EBADMSG);
2290	goto splice_read_end;
2291	}
2292
2293	tls_rx_rec_done(ctx);
2294	skb = darg.skb;
2295	}
2296
2297	rxm = strp_msg(skb);
2298	tlm = tls_msg(skb);
2299
2300	/* splice does not support reading control messages */
2301	if (tlm->control != TLS_RECORD_TYPE_DATA) {
2302	err = -EINVAL;
2303	goto splice_requeue;
2304	}
2305
2306	chunk = min_t(unsigned int, rxm->full_len, len);
2307	copied = skb_splice_bits(skb, sk, offset: rxm->offset, pipe, len: chunk, flags);
2308	if (copied < 0)
2309	goto splice_requeue;
2310
2311	if (chunk < rxm->full_len) {
2312	rxm->offset += len;
2313	rxm->full_len -= len;
2314	goto splice_requeue;
2315	}
2316
2317	consume_skb(skb);
2318
2319	splice_read_end:
2320	tls_rx_reader_unlock(sk, ctx);
2321	return copied ? : err;
2322
2323	splice_requeue:
2324	__skb_queue_head(list: &ctx->rx_list, newsk: skb);
2325	goto splice_read_end;
2326	}
2327
2328	int tls_sw_read_sock(struct sock *sk, read_descriptor_t *desc,
2329	sk_read_actor_t read_actor)
2330	{
2331	struct tls_context *tls_ctx = tls_get_ctx(sk);
2332	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2333	struct tls_prot_info *prot = &tls_ctx->prot_info;
2334	struct strp_msg *rxm = NULL;
2335	struct sk_buff *skb = NULL;
2336	struct sk_psock *psock;
2337	size_t flushed_at = 0;
2338	bool released = true;
2339	struct tls_msg *tlm;
2340	ssize_t copied = 0;
2341	ssize_t decrypted;
2342	int err, used;
2343
2344	psock = sk_psock_get(sk);
2345	if (psock) {
2346	sk_psock_put(sk, psock);
2347	return -EINVAL;
2348	}
2349	err = tls_rx_reader_acquire(sk, ctx, nonblock: true);
2350	if (err < 0)
2351	return err;
2352
2353	/* If crypto failed the connection is broken */
2354	err = ctx->async_wait.err;
2355	if (err)
2356	goto read_sock_end;
2357
2358	decrypted = 0;
2359	do {
2360	if (!skb_queue_empty(list: &ctx->rx_list)) {
2361	skb = __skb_dequeue(list: &ctx->rx_list);
2362	rxm = strp_msg(skb);
2363	tlm = tls_msg(skb);
2364	} else {
2365	struct tls_decrypt_arg darg;
2366
2367	err = tls_rx_rec_wait(sk, NULL, nonblock: true, released);
2368	if (err <= 0)
2369	goto read_sock_end;
2370
2371	memset(&darg.inargs, 0, sizeof(darg.inargs));
2372
2373	err = tls_rx_one_record(sk, NULL, darg: &darg);
2374	if (err < 0) {
2375	tls_err_abort(sk, err: -EBADMSG);
2376	goto read_sock_end;
2377	}
2378
2379	released = tls_read_flush_backlog(sk, prot, INT_MAX,
2380	decrypted: 0, done: decrypted,
2381	flushed_at: &flushed_at);
2382	skb = darg.skb;
2383	rxm = strp_msg(skb);
2384	tlm = tls_msg(skb);
2385	decrypted += rxm->full_len;
2386
2387	tls_rx_rec_done(ctx);
2388	}
2389
2390	/* read_sock does not support reading control messages */
2391	if (tlm->control != TLS_RECORD_TYPE_DATA) {
2392	err = -EINVAL;
2393	goto read_sock_requeue;
2394	}
2395
2396	used = read_actor(desc, skb, rxm->offset, rxm->full_len);
2397	if (used <= 0) {
2398	if (!copied)
2399	err = used;
2400	goto read_sock_requeue;
2401	}
2402	copied += used;
2403	if (used < rxm->full_len) {
2404	rxm->offset += used;
2405	rxm->full_len -= used;
2406	if (!desc->count)
2407	goto read_sock_requeue;
2408	} else {
2409	consume_skb(skb);
2410	if (!desc->count)
2411	skb = NULL;
2412	}
2413	} while (skb);
2414
2415	read_sock_end:
2416	tls_rx_reader_release(sk, ctx);
2417	return copied ? : err;
2418
2419	read_sock_requeue:
2420	__skb_queue_head(list: &ctx->rx_list, newsk: skb);
2421	goto read_sock_end;
2422	}
2423
2424	bool tls_sw_sock_is_readable(struct sock *sk)
2425	{
2426	struct tls_context *tls_ctx = tls_get_ctx(sk);
2427	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2428	bool ingress_empty = true;
2429	struct sk_psock *psock;
2430
2431	rcu_read_lock();
2432	psock = sk_psock(sk);
2433	if (psock)
2434	ingress_empty = list_empty(head: &psock->ingress_msg);
2435	rcu_read_unlock();
2436
2437	return !ingress_empty || tls_strp_msg_ready(ctx) ||
2438	!skb_queue_empty(list: &ctx->rx_list);
2439	}
2440
2441	int tls_rx_msg_size(struct tls_strparser *strp, struct sk_buff *skb)
2442	{
2443	struct tls_context *tls_ctx = tls_get_ctx(sk: strp->sk);
2444	struct tls_prot_info *prot = &tls_ctx->prot_info;
2445	char header[TLS_HEADER_SIZE + TLS_MAX_IV_SIZE];
2446	size_t cipher_overhead;
2447	size_t data_len = 0;
2448	int ret;
2449
2450	/* Verify that we have a full TLS header, or wait for more data */
2451	if (strp->stm.offset + prot->prepend_size > skb->len)
2452	return 0;
2453
2454	/* Sanity-check size of on-stack buffer. */
2455	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2456	ret = -EINVAL;
2457	goto read_failure;
2458	}
2459
2460	/* Linearize header to local buffer */
2461	ret = skb_copy_bits(skb, offset: strp->stm.offset, to: header, len: prot->prepend_size);
2462	if (ret < 0)
2463	goto read_failure;
2464
2465	strp->mark = header[0];
2466
2467	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2468
2469	cipher_overhead = prot->tag_size;
2470	if (prot->version != TLS_1_3_VERSION &&
2471	prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2472	cipher_overhead += prot->iv_size;
2473
2474	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2475	prot->tail_size) {
2476	ret = -EMSGSIZE;
2477	goto read_failure;
2478	}
2479	if (data_len < cipher_overhead) {
2480	ret = -EBADMSG;
2481	goto read_failure;
2482	}
2483
2484	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2485	if (header[1] != TLS_1_2_VERSION_MINOR ||
2486	header[2] != TLS_1_2_VERSION_MAJOR) {
2487	ret = -EINVAL;
2488	goto read_failure;
2489	}
2490
2491	tls_device_rx_resync_new_rec(sk: strp->sk, rcd_len: data_len + TLS_HEADER_SIZE,
2492	TCP_SKB_CB(skb)->seq + strp->stm.offset);
2493	return data_len + TLS_HEADER_SIZE;
2494
2495	read_failure:
2496	tls_strp_abort_strp(strp, err: ret);
2497	return ret;
2498	}
2499
2500	void tls_rx_msg_ready(struct tls_strparser *strp)
2501	{
2502	struct tls_sw_context_rx *ctx;
2503
2504	ctx = container_of(strp, struct tls_sw_context_rx, strp);
2505	ctx->saved_data_ready(strp->sk);
2506	}
2507
2508	static void tls_data_ready(struct sock *sk)
2509	{
2510	struct tls_context *tls_ctx = tls_get_ctx(sk);
2511	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2512	struct sk_psock *psock;
2513	gfp_t alloc_save;
2514
2515	trace_sk_data_ready(sk);
2516
2517	alloc_save = sk->sk_allocation;
2518	sk->sk_allocation = GFP_ATOMIC;
2519	tls_strp_data_ready(strp: &ctx->strp);
2520	sk->sk_allocation = alloc_save;
2521
2522	psock = sk_psock_get(sk);
2523	if (psock) {
2524	if (!list_empty(head: &psock->ingress_msg))
2525	ctx->saved_data_ready(sk);
2526	sk_psock_put(sk, psock);
2527	}
2528	}
2529
2530	void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2531	{
2532	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2533
2534	set_bit(BIT_TX_CLOSING, addr: &ctx->tx_bitmask);
2535	set_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask);
2536	cancel_delayed_work_sync(dwork: &ctx->tx_work.work);
2537	}
2538
2539	void tls_sw_release_resources_tx(struct sock *sk)
2540	{
2541	struct tls_context *tls_ctx = tls_get_ctx(sk);
2542	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2543	struct tls_rec *rec, *tmp;
2544
2545	/* Wait for any pending async encryptions to complete */
2546	tls_encrypt_async_wait(ctx);
2547
2548	tls_tx_records(sk, flags: -1);
2549
2550	/* Free up un-sent records in tx_list. First, free
2551	* the partially sent record if any at head of tx_list.
2552	*/
2553	if (tls_ctx->partially_sent_record) {
2554	tls_free_partial_record(sk, ctx: tls_ctx);
2555	rec = list_first_entry(&ctx->tx_list,
2556	struct tls_rec, list);
2557	list_del(entry: &rec->list);
2558	sk_msg_free(sk, msg: &rec->msg_plaintext);
2559	kfree(objp: rec);
2560	}
2561
2562	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2563	list_del(entry: &rec->list);
2564	sk_msg_free(sk, msg: &rec->msg_encrypted);
2565	sk_msg_free(sk, msg: &rec->msg_plaintext);
2566	kfree(objp: rec);
2567	}
2568
2569	crypto_free_aead(tfm: ctx->aead_send);
2570	tls_free_open_rec(sk);
2571	}
2572
2573	void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2574	{
2575	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2576
2577	kfree(objp: ctx);
2578	}
2579
2580	void tls_sw_release_resources_rx(struct sock *sk)
2581	{
2582	struct tls_context *tls_ctx = tls_get_ctx(sk);
2583	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2584
2585	if (ctx->aead_recv) {
2586	__skb_queue_purge(list: &ctx->rx_list);
2587	crypto_free_aead(tfm: ctx->aead_recv);
2588	tls_strp_stop(strp: &ctx->strp);
2589	/* If tls_sw_strparser_arm() was not called (cleanup paths)
2590	* we still want to tls_strp_stop(), but sk->sk_data_ready was
2591	* never swapped.
2592	*/
2593	if (ctx->saved_data_ready) {
2594	write_lock_bh(&sk->sk_callback_lock);
2595	sk->sk_data_ready = ctx->saved_data_ready;
2596	write_unlock_bh(&sk->sk_callback_lock);
2597	}
2598	}
2599	}
2600
2601	void tls_sw_strparser_done(struct tls_context *tls_ctx)
2602	{
2603	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2604
2605	tls_strp_done(strp: &ctx->strp);
2606	}
2607
2608	void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2609	{
2610	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2611
2612	kfree(objp: ctx);
2613	}
2614
2615	void tls_sw_free_resources_rx(struct sock *sk)
2616	{
2617	struct tls_context *tls_ctx = tls_get_ctx(sk);
2618
2619	tls_sw_release_resources_rx(sk);
2620	tls_sw_free_ctx_rx(tls_ctx);
2621	}
2622
2623	/* The work handler to transmitt the encrypted records in tx_list */
2624	static void tx_work_handler(struct work_struct *work)
2625	{
2626	struct delayed_work *delayed_work = to_delayed_work(work);
2627	struct tx_work *tx_work = container_of(delayed_work,
2628	struct tx_work, work);
2629	struct sock *sk = tx_work->sk;
2630	struct tls_context *tls_ctx = tls_get_ctx(sk);
2631	struct tls_sw_context_tx *ctx;
2632
2633	if (unlikely(!tls_ctx))
2634	return;
2635
2636	ctx = tls_sw_ctx_tx(tls_ctx);
2637	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2638	return;
2639
2640	if (!test_and_clear_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask))
2641	return;
2642
2643	if (mutex_trylock(&tls_ctx->tx_lock)) {
2644	lock_sock(sk);
2645	tls_tx_records(sk, flags: -1);
2646	release_sock(sk);
2647	mutex_unlock(lock: &tls_ctx->tx_lock);
2648	} else if (!test_and_set_bit(BIT_TX_SCHEDULED, addr: &ctx->tx_bitmask)) {
2649	/* Someone is holding the tx_lock, they will likely run Tx
2650	* and cancel the work on their way out of the lock section.
2651	* Schedule a long delay just in case.
2652	*/
2653	schedule_delayed_work(dwork: &ctx->tx_work.work, delay: msecs_to_jiffies(m: 10));
2654	}
2655	}
2656
2657	static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2658	{
2659	struct tls_rec *rec;
2660
2661	rec = list_first_entry_or_null(&ctx->tx_list, struct tls_rec, list);
2662	if (!rec)
2663	return false;
2664
2665	return READ_ONCE(rec->tx_ready);
2666	}
2667
2668	void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2669	{
2670	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(tls_ctx: ctx);
2671
2672	/* Schedule the transmission if tx list is ready */
2673	if (tls_is_tx_ready(ctx: tx_ctx) &&
2674	!test_and_set_bit(BIT_TX_SCHEDULED, addr: &tx_ctx->tx_bitmask))
2675	schedule_delayed_work(dwork: &tx_ctx->tx_work.work, delay: 0);
2676	}
2677
2678	void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2679	{
2680	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2681
2682	write_lock_bh(&sk->sk_callback_lock);
2683	rx_ctx->saved_data_ready = sk->sk_data_ready;
2684	sk->sk_data_ready = tls_data_ready;
2685	write_unlock_bh(&sk->sk_callback_lock);
2686	}
2687
2688	void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2689	{
2690	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2691
2692	rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2693	tls_ctx->prot_info.version != TLS_1_3_VERSION;
2694	}
2695
2696	static struct tls_sw_context_tx *init_ctx_tx(struct tls_context *ctx, struct sock *sk)
2697	{
2698	struct tls_sw_context_tx *sw_ctx_tx;
2699
2700	if (!ctx->priv_ctx_tx) {
2701	sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2702	if (!sw_ctx_tx)
2703	return NULL;
2704	} else {
2705	sw_ctx_tx = ctx->priv_ctx_tx;
2706	}
2707
2708	crypto_init_wait(wait: &sw_ctx_tx->async_wait);
2709	atomic_set(v: &sw_ctx_tx->encrypt_pending, i: 1);
2710	INIT_LIST_HEAD(list: &sw_ctx_tx->tx_list);
2711	INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2712	sw_ctx_tx->tx_work.sk = sk;
2713
2714	return sw_ctx_tx;
2715	}
2716
2717	static struct tls_sw_context_rx *init_ctx_rx(struct tls_context *ctx)
2718	{
2719	struct tls_sw_context_rx *sw_ctx_rx;
2720
2721	if (!ctx->priv_ctx_rx) {
2722	sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2723	if (!sw_ctx_rx)
2724	return NULL;
2725	} else {
2726	sw_ctx_rx = ctx->priv_ctx_rx;
2727	}
2728
2729	crypto_init_wait(wait: &sw_ctx_rx->async_wait);
2730	atomic_set(v: &sw_ctx_rx->decrypt_pending, i: 1);
2731	init_waitqueue_head(&sw_ctx_rx->wq);
2732	skb_queue_head_init(list: &sw_ctx_rx->rx_list);
2733	skb_queue_head_init(list: &sw_ctx_rx->async_hold);
2734
2735	return sw_ctx_rx;
2736	}
2737
2738	int init_prot_info(struct tls_prot_info *prot,
2739	const struct tls_crypto_info *crypto_info,
2740	const struct tls_cipher_desc *cipher_desc)
2741	{
2742	u16 nonce_size = cipher_desc->nonce;
2743
2744	if (crypto_info->version == TLS_1_3_VERSION) {
2745	nonce_size = 0;
2746	prot->aad_size = TLS_HEADER_SIZE;
2747	prot->tail_size = 1;
2748	} else {
2749	prot->aad_size = TLS_AAD_SPACE_SIZE;
2750	prot->tail_size = 0;
2751	}
2752
2753	/* Sanity-check the sizes for stack allocations. */
2754	if (nonce_size > TLS_MAX_IV_SIZE || prot->aad_size > TLS_MAX_AAD_SIZE)
2755	return -EINVAL;
2756
2757	prot->version = crypto_info->version;
2758	prot->cipher_type = crypto_info->cipher_type;
2759	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2760	prot->tag_size = cipher_desc->tag;
2761	prot->overhead_size = prot->prepend_size + prot->tag_size + prot->tail_size;
2762	prot->iv_size = cipher_desc->iv;
2763	prot->salt_size = cipher_desc->salt;
2764	prot->rec_seq_size = cipher_desc->rec_seq;
2765
2766	return 0;
2767	}
2768
2769	static void tls_finish_key_update(struct sock *sk, struct tls_context *tls_ctx)
2770	{
2771	struct tls_sw_context_rx *ctx = tls_ctx->priv_ctx_rx;
2772
2773	WRITE_ONCE(ctx->key_update_pending, false);
2774	/* wake-up pre-existing poll() */
2775	ctx->saved_data_ready(sk);
2776	}
2777
2778	int tls_set_sw_offload(struct sock *sk, int tx,
2779	struct tls_crypto_info *new_crypto_info)
2780	{
2781	struct tls_crypto_info *crypto_info, *src_crypto_info;
2782	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2783	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2784	const struct tls_cipher_desc *cipher_desc;
2785	char *iv, *rec_seq, *key, *salt;
2786	struct cipher_context *cctx;
2787	struct tls_prot_info *prot;
2788	struct crypto_aead **aead;
2789	struct tls_context *ctx;
2790	struct crypto_tfm *tfm;
2791	int rc = 0;
2792
2793	ctx = tls_get_ctx(sk);
2794	prot = &ctx->prot_info;
2795
2796	/* new_crypto_info != NULL means rekey */
2797	if (!new_crypto_info) {
2798	if (tx) {
2799	ctx->priv_ctx_tx = init_ctx_tx(ctx, sk);
2800	if (!ctx->priv_ctx_tx)
2801	return -ENOMEM;
2802	} else {
2803	ctx->priv_ctx_rx = init_ctx_rx(ctx);
2804	if (!ctx->priv_ctx_rx)
2805	return -ENOMEM;
2806	}
2807	}
2808
2809	if (tx) {
2810	sw_ctx_tx = ctx->priv_ctx_tx;
2811	crypto_info = &ctx->crypto_send.info;
2812	cctx = &ctx->tx;
2813	aead = &sw_ctx_tx->aead_send;
2814	} else {
2815	sw_ctx_rx = ctx->priv_ctx_rx;
2816	crypto_info = &ctx->crypto_recv.info;
2817	cctx = &ctx->rx;
2818	aead = &sw_ctx_rx->aead_recv;
2819	}
2820
2821	src_crypto_info = new_crypto_info ?: crypto_info;
2822
2823	cipher_desc = get_cipher_desc(cipher_type: src_crypto_info->cipher_type);
2824	if (!cipher_desc) {
2825	rc = -EINVAL;
2826	goto free_priv;
2827	}
2828
2829	rc = init_prot_info(prot, crypto_info: src_crypto_info, cipher_desc);
2830	if (rc)
2831	goto free_priv;
2832
2833	iv = crypto_info_iv(crypto_info: src_crypto_info, cipher_desc);
2834	key = crypto_info_key(crypto_info: src_crypto_info, cipher_desc);
2835	salt = crypto_info_salt(crypto_info: src_crypto_info, cipher_desc);
2836	rec_seq = crypto_info_rec_seq(crypto_info: src_crypto_info, cipher_desc);
2837
2838	if (!*aead) {
2839	*aead = crypto_alloc_aead(alg_name: cipher_desc->cipher_name, type: 0, mask: 0);
2840	if (IS_ERR(ptr: *aead)) {
2841	rc = PTR_ERR(ptr: *aead);
2842	*aead = NULL;
2843	goto free_priv;
2844	}
2845	}
2846
2847	ctx->push_pending_record = tls_sw_push_pending_record;
2848
2849	/* setkey is the last operation that could fail during a
2850	* rekey. if it succeeds, we can start modifying the
2851	* context.
2852	*/
2853	rc = crypto_aead_setkey(tfm: *aead, key, keylen: cipher_desc->key);
2854	if (rc) {
2855	if (new_crypto_info)
2856	goto out;
2857	else
2858	goto free_aead;
2859	}
2860
2861	if (!new_crypto_info) {
2862	rc = crypto_aead_setauthsize(tfm: *aead, authsize: prot->tag_size);
2863	if (rc)
2864	goto free_aead;
2865	}
2866
2867	if (!tx && !new_crypto_info) {
2868	tfm = crypto_aead_tfm(tfm: sw_ctx_rx->aead_recv);
2869
2870	tls_update_rx_zc_capable(tls_ctx: ctx);
2871	sw_ctx_rx->async_capable =
2872	src_crypto_info->version != TLS_1_3_VERSION &&
2873	!!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
2874
2875	rc = tls_strp_init(strp: &sw_ctx_rx->strp, sk);
2876	if (rc)
2877	goto free_aead;
2878	}
2879
2880	memcpy(cctx->iv, salt, cipher_desc->salt);
2881	memcpy(cctx->iv + cipher_desc->salt, iv, cipher_desc->iv);
2882	memcpy(cctx->rec_seq, rec_seq, cipher_desc->rec_seq);
2883
2884	if (new_crypto_info) {
2885	unsafe_memcpy(crypto_info, new_crypto_info,
2886	cipher_desc->crypto_info,
2887	/* size was checked in do_tls_setsockopt_conf */);
2888	memzero_explicit(s: new_crypto_info, count: cipher_desc->crypto_info);
2889	if (!tx)
2890	tls_finish_key_update(sk, tls_ctx: ctx);
2891	}
2892
2893	goto out;
2894
2895	free_aead:
2896	crypto_free_aead(tfm: *aead);
2897	*aead = NULL;
2898	free_priv:
2899	if (!new_crypto_info) {
2900	if (tx) {
2901	kfree(objp: ctx->priv_ctx_tx);
2902	ctx->priv_ctx_tx = NULL;
2903	} else {
2904	kfree(objp: ctx->priv_ctx_rx);
2905	ctx->priv_ctx_rx = NULL;
2906	}
2907	}
2908	out:
2909	return rc;
2910	}
2911