projects/linux·blob master · refs

2657 lines · 66.6 KiB

1/*
2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <[email protected]>. All rights reserved.
4 * Copyright (c) 2016-2017, Lance Chao <[email protected]>. All rights reserved.
5 * Copyright (c) 2016, Fridolin Pokorny <[email protected]>. All rights reserved.
6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <[email protected]>. 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
51struct 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
62struct 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
71noinline 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
81static 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(&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(frag_iter, offset - start, chunk,
133 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 */
152static int skb_nsg(struct sk_buff *skb, int offset, int len)
153{
154 return __skb_nsg(skb, offset, len, 0);
155}
156
157static 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, rxm->offset + offset,
174 &content_type, 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
187static void tls_decrypt_done(void *data, int err)
188{
189 struct aead_request *aead_req = data;
190 struct crypto_aead *aead = crypto_aead_reqtfm(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(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(sg_page(sg));
234 }
235 }
236
237 kfree(aead_req);
238
239 if (atomic_dec_and_test(&ctx->decrypt_pending))
240 complete(&ctx->async_wait.completion);
241}
242
243static int tls_decrypt_async_wait(struct tls_sw_context_rx *ctx)
244{
245 if (!atomic_dec_and_test(&ctx->decrypt_pending))
246 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
247 atomic_inc(&ctx->decrypt_pending);
248
249 __skb_queue_purge(&ctx->async_hold);
250 return ctx->async_wait.err;
251}
252
253static int tls_do_decryption(struct sock *sk,
254 struct scatterlist *sgin,
255 struct scatterlist *sgout,
256 char *iv_recv,
257 size_t data_len,
258 struct aead_request *aead_req,
259 struct tls_decrypt_arg *darg)
260{
261 struct tls_context *tls_ctx = tls_get_ctx(sk);
262 struct tls_prot_info *prot = &tls_ctx->prot_info;
263 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
264 int ret;
265
266 aead_request_set_tfm(aead_req, ctx->aead_recv);
267 aead_request_set_ad(aead_req, prot->aad_size);
268 aead_request_set_crypt(aead_req, sgin, sgout,
269 data_len + prot->tag_size,
270 (u8 *)iv_recv);
271
272 if (darg->async) {
273 aead_request_set_callback(aead_req,
274 CRYPTO_TFM_REQ_MAY_BACKLOG,
275 tls_decrypt_done, aead_req);
276 DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->decrypt_pending) < 1);
277 atomic_inc(&ctx->decrypt_pending);
278 } else {
279 DECLARE_CRYPTO_WAIT(wait);
280
281 aead_request_set_callback(aead_req,
282 CRYPTO_TFM_REQ_MAY_BACKLOG,
283 crypto_req_done, &wait);
284 ret = crypto_aead_decrypt(aead_req);
285 if (ret == -EINPROGRESS || ret == -EBUSY)
286 ret = crypto_wait_req(ret, &wait);
287 return ret;
288 }
289
290 ret = crypto_aead_decrypt(aead_req);
291 if (ret == -EINPROGRESS)
292 return 0;
293
294 if (ret == -EBUSY) {
295 ret = tls_decrypt_async_wait(ctx);
296 darg->async_done = true;
297 /* all completions have run, we're not doing async anymore */
298 darg->async = false;
299 return ret;
300 }
301
302 atomic_dec(&ctx->decrypt_pending);
303 darg->async = false;
304
305 return ret;
306}
307
308static void tls_trim_both_msgs(struct sock *sk, int target_size)
309{
310 struct tls_context *tls_ctx = tls_get_ctx(sk);
311 struct tls_prot_info *prot = &tls_ctx->prot_info;
312 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
313 struct tls_rec *rec = ctx->open_rec;
314
315 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
316 if (target_size > 0)
317 target_size += prot->overhead_size;
318 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
319}
320
321static int tls_alloc_encrypted_msg(struct sock *sk, int len)
322{
323 struct tls_context *tls_ctx = tls_get_ctx(sk);
324 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
325 struct tls_rec *rec = ctx->open_rec;
326 struct sk_msg *msg_en = &rec->msg_encrypted;
327
328 return sk_msg_alloc(sk, msg_en, len, 0);
329}
330
331static int tls_clone_plaintext_msg(struct sock *sk, int required)
332{
333 struct tls_context *tls_ctx = tls_get_ctx(sk);
334 struct tls_prot_info *prot = &tls_ctx->prot_info;
335 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
336 struct tls_rec *rec = ctx->open_rec;
337 struct sk_msg *msg_pl = &rec->msg_plaintext;
338 struct sk_msg *msg_en = &rec->msg_encrypted;
339 int skip, len;
340
341 /* We add page references worth len bytes from encrypted sg
342 * at the end of plaintext sg. It is guaranteed that msg_en
343 * has enough required room (ensured by caller).
344 */
345 len = required - msg_pl->sg.size;
346
347 /* Skip initial bytes in msg_en's data to be able to use
348 * same offset of both plain and encrypted data.
349 */
350 skip = prot->prepend_size + msg_pl->sg.size;
351
352 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
353}
354
355static struct tls_rec *tls_get_rec(struct sock *sk)
356{
357 struct tls_context *tls_ctx = tls_get_ctx(sk);
358 struct tls_prot_info *prot = &tls_ctx->prot_info;
359 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
360 struct sk_msg *msg_pl, *msg_en;
361 struct tls_rec *rec;
362 int mem_size;
363
364 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
365
366 rec = kzalloc(mem_size, sk->sk_allocation);
367 if (!rec)
368 return NULL;
369
370 msg_pl = &rec->msg_plaintext;
371 msg_en = &rec->msg_encrypted;
372
373 sk_msg_init(msg_pl);
374 sk_msg_init(msg_en);
375
376 sg_init_table(rec->sg_aead_in, 2);
377 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
378 sg_unmark_end(&rec->sg_aead_in[1]);
379
380 sg_init_table(rec->sg_aead_out, 2);
381 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
382 sg_unmark_end(&rec->sg_aead_out[1]);
383
384 rec->sk = sk;
385
386 return rec;
387}
388
389static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
390{
391 sk_msg_free(sk, &rec->msg_encrypted);
392 sk_msg_free(sk, &rec->msg_plaintext);
393 kfree(rec);
394}
395
396static void tls_free_open_rec(struct sock *sk)
397{
398 struct tls_context *tls_ctx = tls_get_ctx(sk);
399 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
400 struct tls_rec *rec = ctx->open_rec;
401
402 if (rec) {
403 tls_free_rec(sk, rec);
404 ctx->open_rec = NULL;
405 }
406}
407
408int tls_tx_records(struct sock *sk, int flags)
409{
410 struct tls_context *tls_ctx = tls_get_ctx(sk);
411 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
412 struct tls_rec *rec, *tmp;
413 struct sk_msg *msg_en;
414 int tx_flags, rc = 0;
415
416 if (tls_is_partially_sent_record(tls_ctx)) {
417 rec = list_first_entry(&ctx->tx_list,
418 struct tls_rec, list);
419
420 if (flags == -1)
421 tx_flags = rec->tx_flags;
422 else
423 tx_flags = flags;
424
425 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
426 if (rc)
427 goto tx_err;
428
429 /* Full record has been transmitted.
430 * Remove the head of tx_list
431 */
432 list_del(&rec->list);
433 sk_msg_free(sk, &rec->msg_plaintext);
434 kfree(rec);
435 }
436
437 /* Tx all ready records */
438 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
439 if (READ_ONCE(rec->tx_ready)) {
440 if (flags == -1)
441 tx_flags = rec->tx_flags;
442 else
443 tx_flags = flags;
444
445 msg_en = &rec->msg_encrypted;
446 rc = tls_push_sg(sk, tls_ctx,
447 &msg_en->sg.data[msg_en->sg.curr],
448 0, tx_flags);
449 if (rc)
450 goto tx_err;
451
452 list_del(&rec->list);
453 sk_msg_free(sk, &rec->msg_plaintext);
454 kfree(rec);
455 } else {
456 break;
457 }
458 }
459
460tx_err:
461 if (rc < 0 && rc != -EAGAIN && rc != -EINTR && rc != -ERESTARTSYS)
462 tls_err_abort(sk, rc);
463
464 return rc;
465}
466
467static void tls_encrypt_done(void *data, int err)
468{
469 struct tls_sw_context_tx *ctx;
470 struct tls_context *tls_ctx;
471 struct tls_prot_info *prot;
472 struct tls_rec *rec = data;
473 struct scatterlist *sge;
474 struct sk_msg *msg_en;
475 struct sock *sk;
476
477 if (err == -EINPROGRESS) /* see the comment in tls_decrypt_done() */
478 return;
479
480 msg_en = &rec->msg_encrypted;
481
482 sk = rec->sk;
483 tls_ctx = tls_get_ctx(sk);
484 prot = &tls_ctx->prot_info;
485 ctx = tls_sw_ctx_tx(tls_ctx);
486
487 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
488 sge->offset -= prot->prepend_size;
489 sge->length += prot->prepend_size;
490
491 /* Check if error is previously set on socket */
492 if (err || sk->sk_err) {
493 rec = NULL;
494
495 /* If err is already set on socket, return the same code */
496 if (sk->sk_err) {
497 ctx->async_wait.err = -sk->sk_err;
498 } else {
499 ctx->async_wait.err = err;
500 tls_err_abort(sk, err);
501 }
502 }
503
504 if (rec) {
505 struct tls_rec *first_rec;
506
507 /* Mark the record as ready for transmission */
508 smp_store_mb(rec->tx_ready, true);
509
510 /* If received record is at head of tx_list, schedule tx */
511 first_rec = list_first_entry(&ctx->tx_list,
512 struct tls_rec, list);
513 if (rec == first_rec) {
514 /* Schedule the transmission */
515 if (!test_and_set_bit(BIT_TX_SCHEDULED,
516 &ctx->tx_bitmask))
517 schedule_delayed_work(&ctx->tx_work.work, 1);
518 }
519 }
520
521 if (atomic_dec_and_test(&ctx->encrypt_pending))
522 complete(&ctx->async_wait.completion);
523}
524
525static int tls_encrypt_async_wait(struct tls_sw_context_tx *ctx)
526{
527 if (!atomic_dec_and_test(&ctx->encrypt_pending))
528 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
529 atomic_inc(&ctx->encrypt_pending);
530
531 return ctx->async_wait.err;
532}
533
534static int tls_do_encryption(struct sock *sk,
535 struct tls_context *tls_ctx,
536 struct tls_sw_context_tx *ctx,
537 struct aead_request *aead_req,
538 size_t data_len, u32 start)
539{
540 struct tls_prot_info *prot = &tls_ctx->prot_info;
541 struct tls_rec *rec = ctx->open_rec;
542 struct sk_msg *msg_en = &rec->msg_encrypted;
543 struct scatterlist *sge = sk_msg_elem(msg_en, start);
544 int rc, iv_offset = 0;
545
546 /* For CCM based ciphers, first byte of IV is a constant */
547 switch (prot->cipher_type) {
548 case TLS_CIPHER_AES_CCM_128:
549 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
550 iv_offset = 1;
551 break;
552 case TLS_CIPHER_SM4_CCM:
553 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
554 iv_offset = 1;
555 break;
556 }
557
558 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
559 prot->iv_size + prot->salt_size);
560
561 tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
562 tls_ctx->tx.rec_seq);
563
564 sge->offset += prot->prepend_size;
565 sge->length -= prot->prepend_size;
566
567 msg_en->sg.curr = start;
568
569 aead_request_set_tfm(aead_req, ctx->aead_send);
570 aead_request_set_ad(aead_req, prot->aad_size);
571 aead_request_set_crypt(aead_req, rec->sg_aead_in,
572 rec->sg_aead_out,
573 data_len, rec->iv_data);
574
575 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
576 tls_encrypt_done, rec);
577
578 /* Add the record in tx_list */
579 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
580 DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->encrypt_pending) < 1);
581 atomic_inc(&ctx->encrypt_pending);
582
583 rc = crypto_aead_encrypt(aead_req);
584 if (rc == -EBUSY) {
585 rc = tls_encrypt_async_wait(ctx);
586 rc = rc ?: -EINPROGRESS;
587 /*
588 * The async callback tls_encrypt_done() has already
589 * decremented encrypt_pending and restored the sge on
590 * both success and error. Skip the synchronous cleanup
591 * below on error, just remove the record and return.
592 */
593 if (rc != -EINPROGRESS) {
594 list_del(&rec->list);
595 return rc;
596 }
597 }
598 if (!rc || rc != -EINPROGRESS) {
599 atomic_dec(&ctx->encrypt_pending);
600 sge->offset -= prot->prepend_size;
601 sge->length += prot->prepend_size;
602 }
603
604 if (!rc) {
605 WRITE_ONCE(rec->tx_ready, true);
606 } else if (rc != -EINPROGRESS) {
607 list_del(&rec->list);
608 return rc;
609 }
610
611 /* Unhook the record from context if encryption is not failure */
612 ctx->open_rec = NULL;
613 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
614 return rc;
615}
616
617static int tls_push_record(struct sock *sk, int flags,
618 unsigned char record_type)
619{
620 struct tls_context *tls_ctx = tls_get_ctx(sk);
621 struct tls_prot_info *prot = &tls_ctx->prot_info;
622 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
623 struct tls_rec *rec = ctx->open_rec;
624 struct sk_msg *msg_pl, *msg_en;
625 struct aead_request *req;
626 int rc;
627 u32 i;
628
629 if (!rec)
630 return 0;
631
632 msg_pl = &rec->msg_plaintext;
633 msg_en = &rec->msg_encrypted;
634
635 rec->tx_flags = flags;
636 req = &rec->aead_req;
637
638 i = msg_pl->sg.end;
639 sk_msg_iter_var_prev(i);
640
641 /* msg_pl->sg.data is a ring; data[MAX+1] is reserved for the wrap
642 * link (frags won't use it). 'i' is now the last filled entry:
643 *
644 * i end start
645 * v v v [ rsv ]
646 * [ d ][ d ][ ][ ]...[ ][ d ][ d ][ d ][chain]
647 * ^ END v
648 * `-----------------------------------------'
649 *
650 * Note that SGL does not allow chain-after-chain, so for TLS 1.3,
651 * we must make sure we don't create the wrap entry and then chain
652 * link to content_type immediately at index 0.
653 */
654 if (i < msg_pl->sg.start)
655 sg_chain(msg_pl->sg.data, ARRAY_SIZE(msg_pl->sg.data),
656 msg_pl->sg.data);
657
658 rec->content_type = record_type;
659 if (prot->version == TLS_1_3_VERSION) {
660 /* Add content type to end of message. No padding added */
661 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
662 sg_mark_end(&rec->sg_content_type);
663 sg_chain(msg_pl->sg.data, i + 2, &rec->sg_content_type);
664 } else {
665 sg_mark_end(sk_msg_elem(msg_pl, i));
666 }
667
668 i = msg_pl->sg.start;
669 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
670
671 i = msg_en->sg.end;
672 sk_msg_iter_var_prev(i);
673 sg_mark_end(sk_msg_elem(msg_en, i));
674
675 i = msg_en->sg.start;
676 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
677
678 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
679 tls_ctx->tx.rec_seq, record_type, prot);
680
681 tls_fill_prepend(tls_ctx,
682 page_address(sg_page(&msg_en->sg.data[i])) +
683 msg_en->sg.data[i].offset,
684 msg_pl->sg.size + prot->tail_size,
685 record_type);
686
687 tls_ctx->pending_open_record_frags = false;
688
689 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
690 msg_pl->sg.size + prot->tail_size, i);
691 if (rc < 0) {
692 if (rc != -EINPROGRESS)
693 tls_err_abort(sk, -EBADMSG);
694 ctx->async_capable = 1;
695 return rc;
696 }
697
698 return tls_tx_records(sk, flags);
699}
700
701static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
702 u8 record_type, ssize_t *copied, int flags)
703{
704 int err;
705
706 err = tls_push_record(sk, flags, record_type);
707 if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
708 *copied -= sk_msg_free(sk, msg);
709 tls_free_open_rec(sk);
710 err = -sk->sk_err;
711 }
712 return err;
713}
714
715static int tls_sw_push_pending_record(struct sock *sk, int flags)
716{
717 struct tls_context *tls_ctx = tls_get_ctx(sk);
718 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
719 struct tls_rec *rec = ctx->open_rec;
720 struct sk_msg *msg_pl;
721 size_t copied;
722
723 if (!rec)
724 return 0;
725
726 msg_pl = &rec->msg_plaintext;
727 copied = msg_pl->sg.size;
728 if (!copied)
729 return 0;
730
731 return bpf_exec_tx_verdict(msg_pl, sk, TLS_RECORD_TYPE_DATA,
732 &copied, flags);
733}
734
735static int tls_sw_sendmsg_splice(struct sock *sk, struct msghdr *msg,
736 struct sk_msg *msg_pl, size_t try_to_copy,
737 ssize_t *copied)
738{
739 struct page *page = NULL, **pages = &page;
740
741 do {
742 ssize_t part;
743 size_t off;
744
745 part = iov_iter_extract_pages(&msg->msg_iter, &pages,
746 try_to_copy, 1, 0, &off);
747 if (part <= 0)
748 return part ?: -EIO;
749
750 if (WARN_ON_ONCE(!sendpage_ok(page))) {
751 iov_iter_revert(&msg->msg_iter, part);
752 return -EIO;
753 }
754
755 sk_msg_page_add(msg_pl, page, part, off);
756 msg_pl->sg.copybreak = 0;
757 msg_pl->sg.curr = msg_pl->sg.end;
758 sk_mem_charge(sk, part);
759 *copied += part;
760 try_to_copy -= part;
761 } while (try_to_copy && !sk_msg_full(msg_pl));
762
763 return 0;
764}
765
766static int tls_sw_sendmsg_locked(struct sock *sk, struct msghdr *msg,
767 size_t size)
768{
769 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
770 struct tls_context *tls_ctx = tls_get_ctx(sk);
771 struct tls_prot_info *prot = &tls_ctx->prot_info;
772 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
773 bool async_capable = ctx->async_capable;
774 unsigned char record_type = TLS_RECORD_TYPE_DATA;
775 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
776 bool eor = !(msg->msg_flags & MSG_MORE);
777 size_t try_to_copy;
778 ssize_t copied = 0;
779 struct sk_msg *msg_pl, *msg_en;
780 struct tls_rec *rec;
781 int required_size;
782 int num_async = 0;
783 bool full_record;
784 int record_room;
785 int num_zc = 0;
786 int orig_size;
787 int ret = 0;
788
789 if (!eor && (msg->msg_flags & MSG_EOR))
790 return -EINVAL;
791
792 if (unlikely(msg->msg_controllen)) {
793 ret = tls_process_cmsg(sk, msg, &record_type);
794 if (ret) {
795 if (ret == -EINPROGRESS)
796 num_async++;
797 else if (ret != -EAGAIN)
798 goto end;
799 }
800 }
801
802 while (msg_data_left(msg)) {
803 if (sk->sk_err) {
804 ret = -sk->sk_err;
805 goto send_end;
806 }
807
808 if (ctx->open_rec)
809 rec = ctx->open_rec;
810 else
811 rec = ctx->open_rec = tls_get_rec(sk);
812 if (!rec) {
813 ret = -ENOMEM;
814 goto send_end;
815 }
816
817 msg_pl = &rec->msg_plaintext;
818 msg_en = &rec->msg_encrypted;
819
820 orig_size = msg_pl->sg.size;
821 full_record = false;
822 try_to_copy = msg_data_left(msg);
823 record_room = tls_ctx->tx_max_payload_len - msg_pl->sg.size;
824 if (try_to_copy >= record_room) {
825 try_to_copy = record_room;
826 full_record = true;
827 }
828
829 required_size = msg_pl->sg.size + try_to_copy +
830 prot->overhead_size;
831
832 if (!sk_stream_memory_free(sk))
833 goto wait_for_sndbuf;
834
835 /* open record may be full if we couldn't push it in the last sendmsg call */
836 if (sk_msg_full(msg_pl)) {
837 full_record = true;
838 sk_msg_trim(sk, msg_en,
839 msg_pl->sg.size + prot->overhead_size);
840 goto copied;
841 }
842
843alloc_encrypted:
844 ret = tls_alloc_encrypted_msg(sk, required_size);
845 if (ret) {
846 if (ret != -ENOSPC)
847 goto wait_for_memory;
848
849 /* Adjust try_to_copy according to the amount that was
850 * actually allocated. The difference is due
851 * to max sg elements limit
852 */
853 try_to_copy -= required_size - msg_en->sg.size;
854 full_record = true;
855 }
856
857 if (try_to_copy && (msg->msg_flags & MSG_SPLICE_PAGES)) {
858 ret = tls_sw_sendmsg_splice(sk, msg, msg_pl,
859 try_to_copy, &copied);
860 if (ret < 0)
861 goto send_end;
862 tls_ctx->pending_open_record_frags = true;
863
864 if (sk_msg_full(msg_pl)) {
865 full_record = true;
866 sk_msg_trim(sk, msg_en,
867 msg_pl->sg.size + prot->overhead_size);
868 }
869
870 if (full_record || eor)
871 goto copied;
872 continue;
873 }
874
875 if (!is_kvec && (full_record || eor) && !async_capable) {
876 u32 first = msg_pl->sg.end;
877
878 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
879 msg_pl, try_to_copy);
880 if (ret)
881 goto fallback_to_reg_send;
882
883 num_zc++;
884 copied += try_to_copy;
885
886 sk_msg_sg_copy_set(msg_pl, first);
887 ret = bpf_exec_tx_verdict(msg_pl, sk,
888 record_type, &copied,
889 msg->msg_flags);
890 if (ret) {
891 if (ret == -EINPROGRESS)
892 num_async++;
893 else if (ret == -ENOMEM)
894 goto wait_for_memory;
895 else if (ret != -EAGAIN)
896 goto send_end;
897 }
898
899 /* Transmit if any encryptions have completed */
900 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
901 cancel_delayed_work(&ctx->tx_work.work);
902 tls_tx_records(sk, msg->msg_flags);
903 }
904
905 continue;
906fallback_to_reg_send:
907 sk_msg_trim(sk, msg_pl, orig_size);
908 }
909
910 required_size = msg_pl->sg.size + try_to_copy;
911
912 ret = tls_clone_plaintext_msg(sk, required_size);
913 if (ret) {
914 if (ret != -ENOSPC)
915 goto send_end;
916
917 /* Adjust try_to_copy according to the amount that was
918 * actually allocated. The difference is due
919 * to max sg elements limit
920 */
921 try_to_copy -= required_size - msg_pl->sg.size;
922 full_record = true;
923 sk_msg_trim(sk, msg_en,
924 msg_pl->sg.size + prot->overhead_size);
925 }
926
927 if (try_to_copy) {
928 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
929 msg_pl, try_to_copy);
930 if (ret < 0)
931 goto trim_sgl;
932
933 if (sk_msg_full(msg_pl)) {
934 full_record = true;
935 sk_msg_trim(sk, msg_en,
936 msg_pl->sg.size + prot->overhead_size);
937 }
938 }
939
940 /* Open records defined only if successfully copied, otherwise
941 * we would trim the sg but not reset the open record frags.
942 */
943 tls_ctx->pending_open_record_frags = true;
944 copied += try_to_copy;
945copied:
946 if (full_record || eor) {
947 ret = bpf_exec_tx_verdict(msg_pl, sk,
948 record_type, &copied,
949 msg->msg_flags);
950 if (ret) {
951 if (ret == -EINPROGRESS)
952 num_async++;
953 else if (ret == -ENOMEM)
954 goto wait_for_memory;
955 else if (ret != -EAGAIN)
956 goto send_end;
957 }
958
959 /* Transmit if any encryptions have completed */
960 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
961 cancel_delayed_work(&ctx->tx_work.work);
962 tls_tx_records(sk, msg->msg_flags);
963 }
964 }
965
966 continue;
967
968wait_for_sndbuf:
969 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
970wait_for_memory:
971 ret = sk_stream_wait_memory(sk, &timeo);
972 if (ret) {
973trim_sgl:
974 if (ctx->open_rec)
975 tls_trim_both_msgs(sk, orig_size);
976 goto send_end;
977 }
978
979 if (ctx->open_rec && msg_en->sg.size < required_size)
980 goto alloc_encrypted;
981 }
982
983send_end:
984 if (!num_async) {
985 goto end;
986 } else if (num_zc || eor) {
987 int err;
988
989 /* Wait for pending encryptions to get completed */
990 err = tls_encrypt_async_wait(ctx);
991 if (err) {
992 ret = err;
993 copied = 0;
994 }
995 }
996
997 /* Transmit if any encryptions have completed */
998 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
999 cancel_delayed_work(&ctx->tx_work.work);
1000 tls_tx_records(sk, msg->msg_flags);
1001 }
1002
1003end:
1004 ret = sk_stream_error(sk, msg->msg_flags, ret);
1005 return copied > 0 ? copied : ret;
1006}
1007
1008int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1009{
1010 struct tls_context *tls_ctx = tls_get_ctx(sk);
1011 int ret;
1012
1013 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1014 MSG_CMSG_COMPAT | MSG_SPLICE_PAGES | MSG_EOR |
1015 MSG_SENDPAGE_NOPOLICY))
1016 return -EOPNOTSUPP;
1017
1018 ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1019 if (ret)
1020 return ret;
1021 lock_sock(sk);
1022 ret = tls_sw_sendmsg_locked(sk, msg, size);
1023 release_sock(sk);
1024 mutex_unlock(&tls_ctx->tx_lock);
1025 return ret;
1026}
1027
1028/*
1029 * Handle unexpected EOF during splice without SPLICE_F_MORE set.
1030 */
1031void tls_sw_splice_eof(struct socket *sock)
1032{
1033 struct sock *sk = sock->sk;
1034 struct tls_context *tls_ctx = tls_get_ctx(sk);
1035 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1036 struct tls_rec *rec;
1037 struct sk_msg *msg_pl;
1038 ssize_t copied = 0;
1039 bool retrying = false;
1040 int ret = 0;
1041
1042 if (!ctx->open_rec)
1043 return;
1044
1045 mutex_lock(&tls_ctx->tx_lock);
1046 lock_sock(sk);
1047
1048retry:
1049 /* same checks as in tls_sw_push_pending_record() */
1050 rec = ctx->open_rec;
1051 if (!rec)
1052 goto unlock;
1053
1054 msg_pl = &rec->msg_plaintext;
1055 if (msg_pl->sg.size == 0)
1056 goto unlock;
1057
1058 /* Perform transmission. */
1059 ret = bpf_exec_tx_verdict(msg_pl, sk, TLS_RECORD_TYPE_DATA,
1060 &copied, 0);
1061 switch (ret) {
1062 case 0:
1063 case -EAGAIN:
1064 if (retrying)
1065 goto unlock;
1066 retrying = true;
1067 goto retry;
1068 case -EINPROGRESS:
1069 break;
1070 default:
1071 goto unlock;
1072 }
1073
1074 /* Wait for pending encryptions to get completed */
1075 if (tls_encrypt_async_wait(ctx))
1076 goto unlock;
1077
1078 /* Transmit if any encryptions have completed */
1079 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1080 cancel_delayed_work(&ctx->tx_work.work);
1081 tls_tx_records(sk, 0);
1082 }
1083
1084unlock:
1085 release_sock(sk);
1086 mutex_unlock(&tls_ctx->tx_lock);
1087}
1088
1089/* When has_copied is true the caller has already moved bytes to
1090 * userspace. Report sk_err but leave it set so the next read
1091 * surfaces it instead of a spurious EOF, otherwise sk_err is
1092 * consumed via sock_error().
1093 */
1094static int
1095tls_rx_rec_wait(struct sock *sk, bool nonblock, bool released, bool has_copied)
1096{
1097 struct tls_context *tls_ctx = tls_get_ctx(sk);
1098 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1099 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1100 int ret = 0;
1101 long timeo;
1102
1103 /* a rekey is pending, let userspace deal with it */
1104 if (unlikely(ctx->key_update_pending))
1105 return -EKEYEXPIRED;
1106
1107 timeo = sock_rcvtimeo(sk, nonblock);
1108
1109 while (!tls_strp_msg_ready(ctx)) {
1110 if (sk->sk_err) {
1111 if (has_copied)
1112 return -READ_ONCE(sk->sk_err);
1113 return sock_error(sk);
1114 }
1115
1116 if (ret < 0)
1117 return ret;
1118
1119 if (sk_flush_backlog(sk))
1120 released = true;
1121 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1122 /* Defer notification to the exit point; this thread
1123 * will consume the record directly.
1124 */
1125 tls_strp_check_rcv(&ctx->strp, false);
1126 if (tls_strp_msg_ready(ctx))
1127 break;
1128 }
1129
1130 /* sk_flush_backlog() can run tcp_reset(), which sets
1131 * sk_err and then sk_shutdown via tcp_done(). Recheck
1132 * sk_err here so a connection abort surfaces as the
1133 * actual error rather than a clean EOF.
1134 */
1135 if (sk->sk_err) {
1136 if (has_copied)
1137 return -READ_ONCE(sk->sk_err);
1138 return sock_error(sk);
1139 }
1140 if (sk->sk_shutdown & RCV_SHUTDOWN)
1141 return 0;
1142
1143 if (sock_flag(sk, SOCK_DONE))
1144 return 0;
1145
1146 if (!timeo)
1147 return -EAGAIN;
1148
1149 released = true;
1150 add_wait_queue(sk_sleep(sk), &wait);
1151 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1152 ret = sk_wait_event(sk, &timeo,
1153 tls_strp_msg_ready(ctx), &wait);
1154 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1155 remove_wait_queue(sk_sleep(sk), &wait);
1156
1157 /* Handle signals */
1158 if (signal_pending(current))
1159 return sock_intr_errno(timeo);
1160 }
1161
1162 if (unlikely(!tls_strp_msg_load(&ctx->strp, released)))
1163 return tls_rx_rec_wait(sk, nonblock, false, has_copied);
1164
1165 return 1;
1166}
1167
1168static int tls_setup_from_iter(struct iov_iter *from,
1169 int length, int *pages_used,
1170 struct scatterlist *to,
1171 int to_max_pages)
1172{
1173 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1174 struct page *pages[MAX_SKB_FRAGS];
1175 unsigned int size = 0;
1176 ssize_t copied, use;
1177 size_t offset;
1178
1179 while (length > 0) {
1180 i = 0;
1181 maxpages = to_max_pages - num_elem;
1182 if (maxpages == 0) {
1183 rc = -EFAULT;
1184 goto out;
1185 }
1186 copied = iov_iter_get_pages2(from, pages,
1187 length,
1188 maxpages, &offset);
1189 if (copied <= 0) {
1190 rc = -EFAULT;
1191 goto out;
1192 }
1193
1194 length -= copied;
1195 size += copied;
1196 while (copied) {
1197 use = min_t(int, copied, PAGE_SIZE - offset);
1198
1199 sg_set_page(&to[num_elem],
1200 pages[i], use, offset);
1201 sg_unmark_end(&to[num_elem]);
1202 /* We do not uncharge memory from this API */
1203
1204 offset = 0;
1205 copied -= use;
1206
1207 i++;
1208 num_elem++;
1209 }
1210 }
1211 /* Mark the end in the last sg entry if newly added */
1212 if (num_elem > *pages_used)
1213 sg_mark_end(&to[num_elem - 1]);
1214out:
1215 if (rc)
1216 iov_iter_revert(from, size);
1217 *pages_used = num_elem;
1218
1219 return rc;
1220}
1221
1222static struct sk_buff *
1223tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1224 unsigned int full_len)
1225{
1226 struct strp_msg *clr_rxm;
1227 struct sk_buff *clr_skb;
1228 int err;
1229
1230 clr_skb = alloc_skb_with_frags(0, full_len, TLS_PAGE_ORDER,
1231 &err, sk->sk_allocation);
1232 if (!clr_skb)
1233 return NULL;
1234
1235 skb_copy_header(clr_skb, skb);
1236 clr_skb->len = full_len;
1237 clr_skb->data_len = full_len;
1238
1239 clr_rxm = strp_msg(clr_skb);
1240 clr_rxm->offset = 0;
1241
1242 return clr_skb;
1243}
1244
1245/* Decrypt handlers
1246 *
1247 * tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1248 * They must transform the darg in/out argument are as follows:
1249 * | Input | Output
1250 * -------------------------------------------------------------------
1251 * zc | Zero-copy decrypt allowed | Zero-copy performed
1252 * async | Async decrypt allowed | Async crypto used / in progress
1253 * skb | * | Output skb
1254 *
1255 * If ZC decryption was performed darg.skb will point to the input skb.
1256 */
1257
1258/* This function decrypts the input skb into either out_iov or in out_sg
1259 * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1260 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1261 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1262 * NULL, then the decryption happens inside skb buffers itself, i.e.
1263 * zero-copy gets disabled and 'darg->zc' is updated.
1264 */
1265static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1266 struct scatterlist *out_sg,
1267 struct tls_decrypt_arg *darg)
1268{
1269 struct tls_context *tls_ctx = tls_get_ctx(sk);
1270 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1271 struct tls_prot_info *prot = &tls_ctx->prot_info;
1272 int n_sgin, n_sgout, aead_size, err, pages = 0;
1273 struct sk_buff *skb = tls_strp_msg(ctx);
1274 const struct strp_msg *rxm = strp_msg(skb);
1275 const struct tls_msg *tlm = tls_msg(skb);
1276 struct aead_request *aead_req;
1277 struct scatterlist *sgin = NULL;
1278 struct scatterlist *sgout = NULL;
1279 const int data_len = rxm->full_len - prot->overhead_size;
1280 int tail_pages = !!prot->tail_size;
1281 struct tls_decrypt_ctx *dctx;
1282 struct sk_buff *clear_skb;
1283 int iv_offset = 0;
1284 u8 *mem;
1285
1286 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1287 rxm->full_len - prot->prepend_size);
1288 if (n_sgin < 1)
1289 return n_sgin ?: -EBADMSG;
1290
1291 if (darg->zc && (out_iov || out_sg)) {
1292 clear_skb = NULL;
1293
1294 if (out_iov)
1295 n_sgout = 1 + tail_pages +
1296 iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1297 else
1298 n_sgout = sg_nents(out_sg);
1299 } else {
1300 darg->zc = false;
1301
1302 clear_skb = tls_alloc_clrtxt_skb(sk, skb, rxm->full_len);
1303 if (!clear_skb)
1304 return -ENOMEM;
1305
1306 n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1307 }
1308
1309 /* Increment to accommodate AAD */
1310 n_sgin = n_sgin + 1;
1311
1312 /* Allocate a single block of memory which contains
1313 * aead_req || tls_decrypt_ctx.
1314 * Both structs are variable length.
1315 */
1316 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1317 aead_size = ALIGN(aead_size, __alignof__(*dctx));
1318 mem = kmalloc(aead_size + struct_size(dctx, sg, size_add(n_sgin, n_sgout)),
1319 sk->sk_allocation);
1320 if (!mem) {
1321 err = -ENOMEM;
1322 goto exit_free_skb;
1323 }
1324
1325 /* Segment the allocated memory */
1326 aead_req = (struct aead_request *)mem;
1327 dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1328 dctx->sk = sk;
1329 sgin = &dctx->sg[0];
1330 sgout = &dctx->sg[n_sgin];
1331
1332 /* For CCM based ciphers, first byte of nonce+iv is a constant */
1333 switch (prot->cipher_type) {
1334 case TLS_CIPHER_AES_CCM_128:
1335 dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1336 iv_offset = 1;
1337 break;
1338 case TLS_CIPHER_SM4_CCM:
1339 dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1340 iv_offset = 1;
1341 break;
1342 }
1343
1344 /* Prepare IV */
1345 if (prot->version == TLS_1_3_VERSION ||
1346 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1347 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1348 prot->iv_size + prot->salt_size);
1349 } else {
1350 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1351 &dctx->iv[iv_offset] + prot->salt_size,
1352 prot->iv_size);
1353 if (err < 0)
1354 goto exit_free;
1355 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1356 }
1357 tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
1358
1359 /* Prepare AAD */
1360 tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
1361 prot->tail_size,
1362 tls_ctx->rx.rec_seq, tlm->control, prot);
1363
1364 /* Prepare sgin */
1365 sg_init_table(sgin, n_sgin);
1366 sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
1367 err = skb_to_sgvec(skb, &sgin[1],
1368 rxm->offset + prot->prepend_size,
1369 rxm->full_len - prot->prepend_size);
1370 if (err < 0)
1371 goto exit_free;
1372
1373 if (clear_skb) {
1374 sg_init_table(sgout, n_sgout);
1375 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1376
1377 err = skb_to_sgvec(clear_skb, &sgout[1], prot->prepend_size,
1378 data_len + prot->tail_size);
1379 if (err < 0)
1380 goto exit_free;
1381 } else if (out_iov) {
1382 sg_init_table(sgout, n_sgout);
1383 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1384
1385 err = tls_setup_from_iter(out_iov, data_len, &pages, &sgout[1],
1386 (n_sgout - 1 - tail_pages));
1387 if (err < 0)
1388 goto exit_free_pages;
1389
1390 if (prot->tail_size) {
1391 sg_unmark_end(&sgout[pages]);
1392 sg_set_buf(&sgout[pages + 1], &dctx->tail,
1393 prot->tail_size);
1394 sg_mark_end(&sgout[pages + 1]);
1395 }
1396 } else if (out_sg) {
1397 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1398 }
1399 dctx->free_sgout = !!pages;
1400
1401 /* Prepare and submit AEAD request */
1402 err = tls_do_decryption(sk, sgin, sgout, dctx->iv,
1403 data_len + prot->tail_size, aead_req, darg);
1404 if (err) {
1405 if (darg->async_done)
1406 goto exit_free_skb;
1407 goto exit_free_pages;
1408 }
1409
1410 darg->skb = clear_skb ?: tls_strp_msg(ctx);
1411 clear_skb = NULL;
1412
1413 if (unlikely(darg->async)) {
1414 err = tls_strp_msg_hold(&ctx->strp, &ctx->async_hold);
1415 if (err) {
1416 err = tls_decrypt_async_wait(ctx);
1417 darg->async = false;
1418 }
1419 return err;
1420 }
1421
1422 if (unlikely(darg->async_done))
1423 return 0;
1424
1425 if (prot->tail_size)
1426 darg->tail = dctx->tail;
1427
1428exit_free_pages:
1429 /* Release the pages in case iov was mapped to pages */
1430 for (; pages > 0; pages--)
1431 put_page(sg_page(&sgout[pages]));
1432exit_free:
1433 kfree(mem);
1434exit_free_skb:
1435 consume_skb(clear_skb);
1436 return err;
1437}
1438
1439static int
1440tls_decrypt_sw(struct sock *sk, struct tls_context *tls_ctx,
1441 struct msghdr *msg, struct tls_decrypt_arg *darg)
1442{
1443 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1444 struct tls_prot_info *prot = &tls_ctx->prot_info;
1445 struct strp_msg *rxm;
1446 int pad, err;
1447
1448 err = tls_decrypt_sg(sk, &msg->msg_iter, NULL, darg);
1449 if (err < 0) {
1450 if (err == -EBADMSG)
1451 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1452 return err;
1453 }
1454 /* keep going even for ->async, the code below is TLS 1.3 */
1455
1456 /* If opportunistic TLS 1.3 ZC failed retry without ZC */
1457 if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1458 darg->tail != TLS_RECORD_TYPE_DATA)) {
1459 iov_iter_revert(&msg->msg_iter, strp_msg(darg->skb)->full_len -
1460 prot->overhead_size);
1461 darg->zc = false;
1462 if (!darg->tail)
1463 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1464 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1465 return tls_decrypt_sw(sk, tls_ctx, msg, darg);
1466 }
1467
1468 pad = tls_padding_length(prot, darg->skb, darg);
1469 if (pad < 0) {
1470 if (darg->skb != tls_strp_msg(ctx))
1471 consume_skb(darg->skb);
1472 return pad;
1473 }
1474
1475 rxm = strp_msg(darg->skb);
1476 rxm->full_len -= pad;
1477
1478 return 0;
1479}
1480
1481static int
1482tls_decrypt_device(struct sock *sk, struct msghdr *msg,
1483 struct tls_context *tls_ctx, struct tls_decrypt_arg *darg)
1484{
1485 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1486 struct tls_prot_info *prot = &tls_ctx->prot_info;
1487 struct strp_msg *rxm;
1488 int pad, err;
1489
1490 if (tls_ctx->rx_conf != TLS_HW)
1491 return 0;
1492
1493 err = tls_device_decrypted(sk, tls_ctx);
1494 if (err <= 0)
1495 return err;
1496
1497 pad = tls_padding_length(prot, tls_strp_msg(ctx), darg);
1498 if (pad < 0)
1499 return pad;
1500
1501 darg->async = false;
1502 darg->skb = tls_strp_msg(ctx);
1503 /* ->zc downgrade check, in case TLS 1.3 gets here */
1504 darg->zc &= !(prot->version == TLS_1_3_VERSION &&
1505 tls_msg(darg->skb)->control != TLS_RECORD_TYPE_DATA);
1506
1507 rxm = strp_msg(darg->skb);
1508 rxm->full_len -= pad;
1509
1510 if (!darg->zc) {
1511 /* Non-ZC case needs a real skb */
1512 darg->skb = tls_strp_msg_detach(ctx);
1513 if (!darg->skb)
1514 return -ENOMEM;
1515 } else {
1516 unsigned int off, len;
1517
1518 /* In ZC case nobody cares about the output skb.
1519 * Just copy the data here. Note the skb is not fully trimmed.
1520 */
1521 off = rxm->offset + prot->prepend_size;
1522 len = rxm->full_len - prot->overhead_size;
1523
1524 err = skb_copy_datagram_msg(darg->skb, off, msg, len);
1525 if (err)
1526 return err;
1527 }
1528 return 1;
1529}
1530
1531static int tls_check_pending_rekey(struct sock *sk, struct tls_context *ctx,
1532 struct sk_buff *skb)
1533{
1534 const struct strp_msg *rxm = strp_msg(skb);
1535 const struct tls_msg *tlm = tls_msg(skb);
1536 char hs_type;
1537 int err;
1538
1539 if (likely(tlm->control != TLS_RECORD_TYPE_HANDSHAKE))
1540 return 0;
1541
1542 if (rxm->full_len < 1)
1543 return 0;
1544
1545 err = skb_copy_bits(skb, rxm->offset, &hs_type, 1);
1546 if (err < 0) {
1547 DEBUG_NET_WARN_ON_ONCE(1);
1548 return err;
1549 }
1550
1551 if (hs_type == TLS_HANDSHAKE_KEYUPDATE) {
1552 struct tls_sw_context_rx *rx_ctx = ctx->priv_ctx_rx;
1553
1554 WRITE_ONCE(rx_ctx->key_update_pending, true);
1555 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXREKEYRECEIVED);
1556 }
1557
1558 return 0;
1559}
1560
1561/* On decrypt failure the connection is aborted (sk_err set) before
1562 * returning a negative errno.
1563 */
1564static int tls_rx_one_record(struct sock *sk, struct msghdr *msg,
1565 struct tls_decrypt_arg *darg)
1566{
1567 struct tls_context *tls_ctx = tls_get_ctx(sk);
1568 struct tls_prot_info *prot = &tls_ctx->prot_info;
1569 struct strp_msg *rxm;
1570 int err;
1571
1572 err = tls_decrypt_device(sk, msg, tls_ctx, darg);
1573 if (!err)
1574 err = tls_decrypt_sw(sk, tls_ctx, msg, darg);
1575 if (err < 0) {
1576 tls_err_abort(sk, -EBADMSG);
1577 return err;
1578 }
1579
1580 rxm = strp_msg(darg->skb);
1581 rxm->offset += prot->prepend_size;
1582 rxm->full_len -= prot->overhead_size;
1583 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1584
1585 return tls_check_pending_rekey(sk, tls_ctx, darg->skb);
1586}
1587
1588int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
1589{
1590 struct tls_decrypt_arg darg = { .zc = true, };
1591
1592 return tls_decrypt_sg(sk, NULL, sgout, &darg);
1593}
1594
1595/* All records returned from a recvmsg() call must have the same type.
1596 * 0 is not a valid content type. Use it as "no type reported, yet".
1597 */
1598static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1599 u8 *control)
1600{
1601 int err;
1602
1603 if (!*control) {
1604 *control = tlm->control;
1605 if (!*control)
1606 return -EBADMSG;
1607
1608 err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1609 sizeof(*control), control);
1610 if (*control != TLS_RECORD_TYPE_DATA) {
1611 if (err || msg->msg_flags & MSG_CTRUNC)
1612 return -EIO;
1613 }
1614 } else if (*control != tlm->control) {
1615 return 0;
1616 }
1617
1618 return 1;
1619}
1620
1621/* The deferred announce is fired once on reader exit by
1622 * tls_rx_reader_release().
1623 */
1624static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1625{
1626 tls_strp_msg_consume(&ctx->strp);
1627 tls_strp_check_rcv(&ctx->strp, false);
1628}
1629
1630/* This function traverses the rx_list in tls receive context to copies the
1631 * decrypted records into the buffer provided by caller zero copy is not
1632 * true. Further, the records are removed from the rx_list if it is not a peek
1633 * case and the record has been consumed completely.
1634 */
1635static int process_rx_list(struct tls_sw_context_rx *ctx,
1636 struct msghdr *msg,
1637 u8 *control,
1638 size_t skip,
1639 size_t len,
1640 bool is_peek,
1641 bool *more)
1642{
1643 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1644 struct tls_msg *tlm;
1645 ssize_t copied = 0;
1646 int err;
1647
1648 while (skip && skb) {
1649 struct strp_msg *rxm = strp_msg(skb);
1650 tlm = tls_msg(skb);
1651
1652 err = tls_record_content_type(msg, tlm, control);
1653 if (err <= 0)
1654 goto more;
1655
1656 if (skip < rxm->full_len)
1657 break;
1658
1659 skip = skip - rxm->full_len;
1660 skb = skb_peek_next(skb, &ctx->rx_list);
1661 }
1662
1663 while (len && skb) {
1664 struct sk_buff *next_skb;
1665 struct strp_msg *rxm = strp_msg(skb);
1666 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1667
1668 tlm = tls_msg(skb);
1669
1670 err = tls_record_content_type(msg, tlm, control);
1671 if (err <= 0)
1672 goto more;
1673
1674 err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1675 msg, chunk);
1676 if (err < 0)
1677 goto more;
1678
1679 len = len - chunk;
1680 copied = copied + chunk;
1681
1682 /* Consume the data from record if it is non-peek case*/
1683 if (!is_peek) {
1684 rxm->offset = rxm->offset + chunk;
1685 rxm->full_len = rxm->full_len - chunk;
1686
1687 /* Return if there is unconsumed data in the record */
1688 if (rxm->full_len - skip)
1689 break;
1690 }
1691
1692 /* The remaining skip-bytes must lie in 1st record in rx_list.
1693 * So from the 2nd record, 'skip' should be 0.
1694 */
1695 skip = 0;
1696
1697 if (msg)
1698 msg->msg_flags |= MSG_EOR;
1699
1700 next_skb = skb_peek_next(skb, &ctx->rx_list);
1701
1702 if (!is_peek) {
1703 __skb_unlink(skb, &ctx->rx_list);
1704 consume_skb(skb);
1705 }
1706
1707 skb = next_skb;
1708 }
1709 err = 0;
1710
1711out:
1712 return copied ? : err;
1713more:
1714 if (more)
1715 *more = true;
1716 goto out;
1717}
1718
1719static bool
1720tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1721 size_t len_left, size_t decrypted, ssize_t done,
1722 size_t *flushed_at)
1723{
1724 size_t max_rec;
1725
1726 if (len_left <= decrypted)
1727 return false;
1728
1729 max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1730 if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1731 return false;
1732
1733 *flushed_at = done;
1734 return sk_flush_backlog(sk);
1735}
1736
1737static int tls_rx_reader_acquire(struct sock *sk, struct tls_sw_context_rx *ctx,
1738 bool nonblock)
1739{
1740 long timeo;
1741 int ret;
1742
1743 timeo = sock_rcvtimeo(sk, nonblock);
1744
1745 while (unlikely(ctx->reader_present)) {
1746 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1747
1748 ctx->reader_contended = 1;
1749
1750 add_wait_queue(&ctx->wq, &wait);
1751 ret = sk_wait_event(sk, &timeo,
1752 !READ_ONCE(ctx->reader_present), &wait);
1753 remove_wait_queue(&ctx->wq, &wait);
1754
1755 if (timeo <= 0)
1756 return -EAGAIN;
1757 if (signal_pending(current))
1758 return sock_intr_errno(timeo);
1759 if (ret < 0)
1760 return ret;
1761 }
1762
1763 WRITE_ONCE(ctx->reader_present, 1);
1764
1765 return 0;
1766}
1767
1768static int tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
1769 bool nonblock)
1770{
1771 int err;
1772
1773 lock_sock(sk);
1774 err = tls_rx_reader_acquire(sk, ctx, nonblock);
1775 if (err)
1776 release_sock(sk);
1777 return err;
1778}
1779
1780static void tls_rx_reader_release(struct sock *sk, struct tls_sw_context_rx *ctx)
1781{
1782 /* Fire any deferred announce once per reader so that a record
1783 * parsed but not yet announced becomes visible to the next
1784 * reader. The call is idempotent through msg_announced.
1785 */
1786 tls_rx_msg_maybe_announce(&ctx->strp);
1787
1788 if (unlikely(ctx->reader_contended)) {
1789 if (wq_has_sleeper(&ctx->wq))
1790 wake_up(&ctx->wq);
1791 else
1792 ctx->reader_contended = 0;
1793
1794 WARN_ON_ONCE(!ctx->reader_present);
1795 }
1796
1797 WRITE_ONCE(ctx->reader_present, 0);
1798}
1799
1800static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
1801{
1802 tls_rx_reader_release(sk, ctx);
1803 release_sock(sk);
1804}
1805
1806int tls_sw_recvmsg(struct sock *sk,
1807 struct msghdr *msg,
1808 size_t len,
1809 int flags)
1810{
1811 struct tls_context *tls_ctx = tls_get_ctx(sk);
1812 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1813 struct tls_prot_info *prot = &tls_ctx->prot_info;
1814 ssize_t decrypted = 0, async_copy_bytes = 0;
1815 unsigned char control = 0;
1816 size_t flushed_at = 0;
1817 struct strp_msg *rxm;
1818 struct tls_msg *tlm;
1819 ssize_t copied = 0;
1820 ssize_t peeked = 0;
1821 bool async = false;
1822 int target, err;
1823 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1824 bool is_peek = flags & MSG_PEEK;
1825 bool rx_more = false;
1826 bool released = true;
1827 bool zc_capable;
1828
1829 if (unlikely(flags & MSG_ERRQUEUE))
1830 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1831
1832 err = tls_rx_reader_lock(sk, ctx, flags & MSG_DONTWAIT);
1833 if (err < 0)
1834 return err;
1835
1836 /* If crypto failed the connection is broken */
1837 err = ctx->async_wait.err;
1838 if (err)
1839 goto end;
1840
1841 /* Process pending decrypted records. It must be non-zero-copy */
1842 err = process_rx_list(ctx, msg, &control, 0, len, is_peek, &rx_more);
1843 if (err < 0)
1844 goto end;
1845
1846 /* process_rx_list() will set @control if it processed any records */
1847 copied = err;
1848 if (len <= copied || rx_more ||
1849 (control && control != TLS_RECORD_TYPE_DATA))
1850 goto end;
1851
1852 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1853 len = len - copied;
1854
1855 zc_capable = !is_kvec && !is_peek && ctx->zc_capable;
1856 decrypted = 0;
1857 while (len && (decrypted + copied < target || tls_strp_msg_ready(ctx))) {
1858 struct tls_decrypt_arg darg;
1859 int to_decrypt, chunk;
1860
1861 err = tls_rx_rec_wait(sk, flags & MSG_DONTWAIT,
1862 released, !!(decrypted + copied));
1863 if (err <= 0)
1864 goto recv_end;
1865
1866 memset(&darg.inargs, 0, sizeof(darg.inargs));
1867
1868 rxm = strp_msg(tls_strp_msg(ctx));
1869 tlm = tls_msg(tls_strp_msg(ctx));
1870
1871 to_decrypt = rxm->full_len - prot->overhead_size;
1872
1873 if (zc_capable && to_decrypt <= len &&
1874 tlm->control == TLS_RECORD_TYPE_DATA)
1875 darg.zc = true;
1876
1877 /* Do not use async mode if record is non-data */
1878 if (tlm->control == TLS_RECORD_TYPE_DATA)
1879 darg.async = ctx->async_capable;
1880 else
1881 darg.async = false;
1882
1883 err = tls_rx_one_record(sk, msg, &darg);
1884 if (err < 0)
1885 goto recv_end;
1886
1887 async |= darg.async;
1888
1889 /* If the type of records being processed is not known yet,
1890 * set it to record type just dequeued. If it is already known,
1891 * but does not match the record type just dequeued, go to end.
1892 * We always get record type here since for tls1.2, record type
1893 * is known just after record is dequeued from stream parser.
1894 * For tls1.3, we disable async.
1895 */
1896 err = tls_record_content_type(msg, tls_msg(darg.skb), &control);
1897 if (err <= 0) {
1898 DEBUG_NET_WARN_ON_ONCE(darg.zc);
1899 tls_rx_rec_done(ctx);
1900put_on_rx_list_err:
1901 __skb_queue_tail(&ctx->rx_list, darg.skb);
1902 goto recv_end;
1903 }
1904
1905 /* periodically flush backlog, and feed strparser */
1906 released = tls_read_flush_backlog(sk, prot, len, to_decrypt,
1907 decrypted + copied,
1908 &flushed_at);
1909
1910 /* TLS 1.3 may have updated the length by more than overhead */
1911 rxm = strp_msg(darg.skb);
1912 chunk = rxm->full_len;
1913 tls_rx_rec_done(ctx);
1914
1915 if (!darg.zc) {
1916 bool partially_consumed = chunk > len;
1917 struct sk_buff *skb = darg.skb;
1918
1919 DEBUG_NET_WARN_ON_ONCE(darg.skb == ctx->strp.anchor);
1920
1921 if (async) {
1922 /* TLS 1.2-only, to_decrypt must be text len */
1923 chunk = min_t(int, to_decrypt, len);
1924 async_copy_bytes += chunk;
1925put_on_rx_list:
1926 decrypted += chunk;
1927 len -= chunk;
1928 __skb_queue_tail(&ctx->rx_list, skb);
1929 if (unlikely(control != TLS_RECORD_TYPE_DATA))
1930 break;
1931 continue;
1932 }
1933
1934 if (partially_consumed)
1935 chunk = len;
1936
1937 err = skb_copy_datagram_msg(skb, rxm->offset,
1938 msg, chunk);
1939 if (err < 0)
1940 goto put_on_rx_list_err;
1941
1942 if (is_peek) {
1943 peeked += chunk;
1944 goto put_on_rx_list;
1945 }
1946
1947 if (partially_consumed) {
1948 rxm->offset += chunk;
1949 rxm->full_len -= chunk;
1950 goto put_on_rx_list;
1951 }
1952
1953 consume_skb(skb);
1954 }
1955
1956 decrypted += chunk;
1957 len -= chunk;
1958
1959 /* Return full control message to userspace before trying
1960 * to parse another message type
1961 */
1962 msg->msg_flags |= MSG_EOR;
1963 if (control != TLS_RECORD_TYPE_DATA)
1964 break;
1965 }
1966
1967recv_end:
1968 if (async) {
1969 int ret;
1970
1971 /* Wait for all previously submitted records to be decrypted */
1972 ret = tls_decrypt_async_wait(ctx);
1973
1974 if (ret) {
1975 if (err >= 0 || err == -EINPROGRESS)
1976 err = ret;
1977 goto end;
1978 }
1979
1980 /* Drain records from the rx_list & copy if required */
1981 if (is_peek)
1982 err = process_rx_list(ctx, msg, &control, copied + peeked,
1983 decrypted - peeked, is_peek, NULL);
1984 else
1985 err = process_rx_list(ctx, msg, &control, 0,
1986 async_copy_bytes, is_peek, NULL);
1987
1988 /* we could have copied less than we wanted, and possibly nothing */
1989 decrypted += max(err, 0) - async_copy_bytes;
1990 }
1991
1992 copied += decrypted;
1993
1994end:
1995 tls_rx_reader_unlock(sk, ctx);
1996 return copied ? : err;
1997}
1998
1999ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
2000 struct pipe_inode_info *pipe,
2001 size_t len, unsigned int flags)
2002{
2003 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2004 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2005 struct strp_msg *rxm = NULL;
2006 struct sock *sk = sock->sk;
2007 struct tls_msg *tlm;
2008 struct sk_buff *skb;
2009 ssize_t copied = 0;
2010 int chunk;
2011 int err;
2012
2013 err = tls_rx_reader_lock(sk, ctx, flags & SPLICE_F_NONBLOCK);
2014 if (err < 0)
2015 return err;
2016
2017 /* If crypto failed the connection is broken */
2018 err = ctx->async_wait.err;
2019 if (err)
2020 goto splice_read_end;
2021
2022 if (!skb_queue_empty(&ctx->rx_list)) {
2023 skb = __skb_dequeue(&ctx->rx_list);
2024 } else {
2025 struct tls_decrypt_arg darg;
2026
2027 err = tls_rx_rec_wait(sk, flags & SPLICE_F_NONBLOCK,
2028 true, false);
2029 if (err <= 0)
2030 goto splice_read_end;
2031
2032 memset(&darg.inargs, 0, sizeof(darg.inargs));
2033
2034 err = tls_rx_one_record(sk, NULL, &darg);
2035 if (err < 0)
2036 goto splice_read_end;
2037
2038 tls_rx_rec_done(ctx);
2039 skb = darg.skb;
2040 }
2041
2042 rxm = strp_msg(skb);
2043 tlm = tls_msg(skb);
2044
2045 /* splice does not support reading control messages */
2046 if (tlm->control != TLS_RECORD_TYPE_DATA) {
2047 err = -EINVAL;
2048 goto splice_requeue;
2049 }
2050
2051 chunk = min_t(unsigned int, rxm->full_len, len);
2052 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2053 if (copied < 0)
2054 goto splice_requeue;
2055
2056 if (copied < rxm->full_len) {
2057 rxm->offset += copied;
2058 rxm->full_len -= copied;
2059 goto splice_requeue;
2060 }
2061
2062 consume_skb(skb);
2063
2064splice_read_end:
2065 tls_rx_reader_unlock(sk, ctx);
2066 return copied ? : err;
2067
2068splice_requeue:
2069 __skb_queue_head(&ctx->rx_list, skb);
2070 goto splice_read_end;
2071}
2072
2073int tls_sw_read_sock(struct sock *sk, read_descriptor_t *desc,
2074 sk_read_actor_t read_actor)
2075{
2076 struct tls_context *tls_ctx = tls_get_ctx(sk);
2077 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2078 struct tls_prot_info *prot = &tls_ctx->prot_info;
2079 struct strp_msg *rxm = NULL;
2080 struct sk_buff *skb = NULL;
2081 struct sk_psock *psock;
2082 size_t flushed_at = 0;
2083 bool released = true;
2084 struct tls_msg *tlm;
2085 ssize_t copied = 0;
2086 ssize_t decrypted;
2087 int err, used;
2088
2089 psock = sk_psock_get(sk);
2090 if (psock) {
2091 sk_psock_put(sk, psock);
2092 return -EINVAL;
2093 }
2094 err = tls_rx_reader_acquire(sk, ctx, true);
2095 if (err < 0)
2096 return err;
2097
2098 /* If crypto failed the connection is broken */
2099 err = ctx->async_wait.err;
2100 if (err)
2101 goto read_sock_end;
2102
2103 decrypted = 0;
2104 while (desc->count) {
2105 if (!skb_queue_empty(&ctx->rx_list)) {
2106 skb = __skb_dequeue(&ctx->rx_list);
2107 rxm = strp_msg(skb);
2108 tlm = tls_msg(skb);
2109 } else {
2110 struct tls_decrypt_arg darg;
2111
2112 err = tls_rx_rec_wait(sk, true, released, !!copied);
2113 if (err <= 0)
2114 goto read_sock_end;
2115
2116 memset(&darg.inargs, 0, sizeof(darg.inargs));
2117
2118 err = tls_rx_one_record(sk, NULL, &darg);
2119 if (err < 0)
2120 goto read_sock_end;
2121
2122 released = tls_read_flush_backlog(sk, prot, INT_MAX,
2123 0, decrypted,
2124 &flushed_at);
2125 skb = darg.skb;
2126 rxm = strp_msg(skb);
2127 tlm = tls_msg(skb);
2128 decrypted += rxm->full_len;
2129
2130 tls_rx_rec_done(ctx);
2131 }
2132
2133 /* read_sock does not support reading control messages */
2134 if (tlm->control != TLS_RECORD_TYPE_DATA) {
2135 err = -EINVAL;
2136 goto read_sock_requeue;
2137 }
2138
2139 /* An empty data record (legal in TLS 1.3) gives a zero
2140 * read_actor return, indistinguishable from the consumer
2141 * stalling; the used <= 0 path would requeue it at the
2142 * head of rx_list and block all later records. Consume it
2143 * here instead.
2144 */
2145 if (rxm->full_len == 0) {
2146 consume_skb(skb);
2147 continue;
2148 }
2149
2150 used = read_actor(desc, skb, rxm->offset, rxm->full_len);
2151 if (used <= 0) {
2152 if (!copied)
2153 err = used;
2154 goto read_sock_requeue;
2155 }
2156 copied += used;
2157 if (used < rxm->full_len) {
2158 rxm->offset += used;
2159 rxm->full_len -= used;
2160 __skb_queue_head(&ctx->rx_list, skb);
2161 } else {
2162 consume_skb(skb);
2163 }
2164 }
2165
2166read_sock_end:
2167 tls_rx_reader_release(sk, ctx);
2168 return copied ? : err;
2169
2170read_sock_requeue:
2171 __skb_queue_head(&ctx->rx_list, skb);
2172 goto read_sock_end;
2173}
2174
2175bool tls_sw_sock_is_readable(struct sock *sk)
2176{
2177 struct tls_context *tls_ctx = tls_get_ctx(sk);
2178 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2179
2180 return tls_strp_msg_ready(ctx) ||
2181 !skb_queue_empty(&ctx->rx_list);
2182}
2183
2184int tls_rx_msg_size(struct tls_strparser *strp, struct sk_buff *skb)
2185{
2186 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2187 struct tls_prot_info *prot = &tls_ctx->prot_info;
2188 char header[TLS_HEADER_SIZE + TLS_MAX_IV_SIZE];
2189 size_t cipher_overhead;
2190 size_t data_len = 0;
2191 int ret;
2192
2193 /* Verify that we have a full TLS header, or wait for more data */
2194 if (strp->stm.offset + prot->prepend_size > skb->len)
2195 return 0;
2196
2197 /* Sanity-check size of on-stack buffer. */
2198 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2199 ret = -EINVAL;
2200 goto read_failure;
2201 }
2202
2203 /* Linearize header to local buffer */
2204 ret = skb_copy_bits(skb, strp->stm.offset, header, prot->prepend_size);
2205 if (ret < 0)
2206 goto read_failure;
2207
2208 strp->mark = header[0];
2209
2210 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2211
2212 cipher_overhead = prot->tag_size;
2213 if (prot->version != TLS_1_3_VERSION &&
2214 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2215 cipher_overhead += prot->iv_size;
2216
2217 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2218 prot->tail_size) {
2219 ret = -EMSGSIZE;
2220 goto read_failure;
2221 }
2222 if (data_len < cipher_overhead) {
2223 ret = -EBADMSG;
2224 goto read_failure;
2225 }
2226
2227 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2228 if (header[1] != TLS_1_2_VERSION_MINOR ||
2229 header[2] != TLS_1_2_VERSION_MAJOR) {
2230 ret = -EINVAL;
2231 goto read_failure;
2232 }
2233
2234 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2235 TCP_SKB_CB(skb)->seq + strp->stm.offset);
2236 return data_len + TLS_HEADER_SIZE;
2237
2238read_failure:
2239 tls_strp_abort_strp(strp, ret);
2240 return ret;
2241}
2242
2243/* Fire saved_data_ready() at most once per parsed record. The
2244 * msg_announced bit is cleared by tls_strp_msg_consume() when the
2245 * record is consumed, arming the next announcement.
2246 */
2247void tls_rx_msg_maybe_announce(struct tls_strparser *strp)
2248{
2249 struct tls_sw_context_rx *ctx;
2250
2251 if (!READ_ONCE(strp->msg_ready) || strp->msg_announced)
2252 return;
2253 strp->msg_announced = 1;
2254
2255 ctx = container_of(strp, struct tls_sw_context_rx, strp);
2256 ctx->saved_data_ready(strp->sk);
2257}
2258
2259static void tls_data_ready(struct sock *sk)
2260{
2261 struct tls_context *tls_ctx = tls_get_ctx(sk);
2262 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2263 gfp_t alloc_save;
2264
2265 trace_sk_data_ready(sk);
2266
2267 alloc_save = sk->sk_allocation;
2268 sk->sk_allocation = GFP_ATOMIC;
2269 tls_strp_data_ready(&ctx->strp);
2270 sk->sk_allocation = alloc_save;
2271}
2272
2273void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2274{
2275 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2276
2277 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2278 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2279 disable_delayed_work_sync(&ctx->tx_work.work);
2280}
2281
2282void tls_sw_release_resources_tx(struct sock *sk)
2283{
2284 struct tls_context *tls_ctx = tls_get_ctx(sk);
2285 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2286 struct tls_rec *rec, *tmp;
2287
2288 /* Wait for any pending async encryptions to complete */
2289 tls_encrypt_async_wait(ctx);
2290
2291 tls_tx_records(sk, -1);
2292
2293 /* Free up un-sent records in tx_list. First, free
2294 * the partially sent record if any at head of tx_list.
2295 */
2296 if (tls_ctx->partially_sent_record) {
2297 tls_free_partial_record(sk, tls_ctx);
2298 rec = list_first_entry(&ctx->tx_list,
2299 struct tls_rec, list);
2300 list_del(&rec->list);
2301 sk_msg_free(sk, &rec->msg_plaintext);
2302 kfree(rec);
2303 }
2304
2305 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2306 list_del(&rec->list);
2307 sk_msg_free(sk, &rec->msg_encrypted);
2308 sk_msg_free(sk, &rec->msg_plaintext);
2309 kfree(rec);
2310 }
2311
2312 crypto_free_aead(ctx->aead_send);
2313 tls_free_open_rec(sk);
2314}
2315
2316void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2317{
2318 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2319
2320 kfree(ctx);
2321}
2322
2323void tls_sw_release_resources_rx(struct sock *sk)
2324{
2325 struct tls_context *tls_ctx = tls_get_ctx(sk);
2326 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2327
2328 if (ctx->aead_recv) {
2329 __skb_queue_purge(&ctx->rx_list);
2330 crypto_free_aead(ctx->aead_recv);
2331 tls_strp_stop(&ctx->strp);
2332 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2333 * we still want to tls_strp_stop(), but sk->sk_data_ready was
2334 * never swapped.
2335 */
2336 if (ctx->saved_data_ready) {
2337 write_lock_bh(&sk->sk_callback_lock);
2338 sk->sk_data_ready = ctx->saved_data_ready;
2339 write_unlock_bh(&sk->sk_callback_lock);
2340 }
2341 }
2342}
2343
2344void tls_sw_strparser_done(struct tls_context *tls_ctx)
2345{
2346 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2347
2348 tls_strp_done(&ctx->strp);
2349}
2350
2351void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2352{
2353 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2354
2355 kfree(ctx);
2356}
2357
2358void tls_sw_free_resources_rx(struct sock *sk)
2359{
2360 struct tls_context *tls_ctx = tls_get_ctx(sk);
2361 struct tls_sw_context_rx *ctx;
2362
2363 ctx = tls_sw_ctx_rx(tls_ctx);
2364
2365 tls_sw_release_resources_rx(sk);
2366 __tls_strp_done(&ctx->strp);
2367 tls_sw_free_ctx_rx(tls_ctx);
2368}
2369
2370/* The work handler to transmitt the encrypted records in tx_list */
2371static void tx_work_handler(struct work_struct *work)
2372{
2373 struct delayed_work *delayed_work = to_delayed_work(work);
2374 struct tx_work *tx_work = container_of(delayed_work,
2375 struct tx_work, work);
2376 struct sock *sk = tx_work->sk;
2377 struct tls_context *tls_ctx = tls_get_ctx(sk);
2378 struct tls_sw_context_tx *ctx;
2379
2380 if (unlikely(!tls_ctx))
2381 return;
2382
2383 ctx = tls_sw_ctx_tx(tls_ctx);
2384 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2385 return;
2386
2387 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2388 return;
2389
2390 if (mutex_trylock(&tls_ctx->tx_lock)) {
2391 lock_sock(sk);
2392 tls_tx_records(sk, -1);
2393 release_sock(sk);
2394 mutex_unlock(&tls_ctx->tx_lock);
2395 } else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2396 /* Someone is holding the tx_lock, they will likely run Tx
2397 * and cancel the work on their way out of the lock section.
2398 * Schedule a long delay just in case.
2399 */
2400 schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2401 }
2402}
2403
2404static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2405{
2406 struct tls_rec *rec;
2407
2408 rec = list_first_entry_or_null(&ctx->tx_list, struct tls_rec, list);
2409 if (!rec)
2410 return false;
2411
2412 return READ_ONCE(rec->tx_ready);
2413}
2414
2415void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2416{
2417 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2418
2419 /* Schedule the transmission if tx list is ready */
2420 if (tls_is_tx_ready(tx_ctx) &&
2421 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2422 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2423}
2424
2425void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2426{
2427 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2428
2429 write_lock_bh(&sk->sk_callback_lock);
2430 rx_ctx->saved_data_ready = sk->sk_data_ready;
2431 sk->sk_data_ready = tls_data_ready;
2432 write_unlock_bh(&sk->sk_callback_lock);
2433}
2434
2435void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2436{
2437 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2438
2439 rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2440 tls_ctx->prot_info.version != TLS_1_3_VERSION;
2441}
2442
2443static struct tls_sw_context_tx *init_ctx_tx(struct tls_context *ctx, struct sock *sk)
2444{
2445 struct tls_sw_context_tx *sw_ctx_tx;
2446
2447 if (!ctx->priv_ctx_tx) {
2448 sw_ctx_tx = kzalloc_obj(*sw_ctx_tx);
2449 if (!sw_ctx_tx)
2450 return NULL;
2451 } else {
2452 sw_ctx_tx = ctx->priv_ctx_tx;
2453 }
2454
2455 crypto_init_wait(&sw_ctx_tx->async_wait);
2456 atomic_set(&sw_ctx_tx->encrypt_pending, 1);
2457 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2458 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2459 sw_ctx_tx->tx_work.sk = sk;
2460
2461 return sw_ctx_tx;
2462}
2463
2464static struct tls_sw_context_rx *init_ctx_rx(struct tls_context *ctx)
2465{
2466 struct tls_sw_context_rx *sw_ctx_rx;
2467
2468 if (!ctx->priv_ctx_rx) {
2469 sw_ctx_rx = kzalloc_obj(*sw_ctx_rx);
2470 if (!sw_ctx_rx)
2471 return NULL;
2472 } else {
2473 sw_ctx_rx = ctx->priv_ctx_rx;
2474 }
2475
2476 crypto_init_wait(&sw_ctx_rx->async_wait);
2477 atomic_set(&sw_ctx_rx->decrypt_pending, 1);
2478 init_waitqueue_head(&sw_ctx_rx->wq);
2479 skb_queue_head_init(&sw_ctx_rx->rx_list);
2480 skb_queue_head_init(&sw_ctx_rx->async_hold);
2481
2482 return sw_ctx_rx;
2483}
2484
2485int init_prot_info(struct tls_prot_info *prot,
2486 const struct tls_crypto_info *crypto_info,
2487 const struct tls_cipher_desc *cipher_desc)
2488{
2489 u16 nonce_size = cipher_desc->nonce;
2490
2491 if (crypto_info->version == TLS_1_3_VERSION) {
2492 nonce_size = 0;
2493 prot->aad_size = TLS_HEADER_SIZE;
2494 prot->tail_size = 1;
2495 } else {
2496 prot->aad_size = TLS_AAD_SPACE_SIZE;
2497 prot->tail_size = 0;
2498 }
2499
2500 /* Sanity-check the sizes for stack allocations. */
2501 if (nonce_size > TLS_MAX_IV_SIZE || prot->aad_size > TLS_MAX_AAD_SIZE)
2502 return -EINVAL;
2503
2504 prot->version = crypto_info->version;
2505 prot->cipher_type = crypto_info->cipher_type;
2506 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2507 prot->tag_size = cipher_desc->tag;
2508 prot->overhead_size = prot->prepend_size + prot->tag_size + prot->tail_size;
2509 prot->iv_size = cipher_desc->iv;
2510 prot->salt_size = cipher_desc->salt;
2511 prot->rec_seq_size = cipher_desc->rec_seq;
2512
2513 return 0;
2514}
2515
2516static void tls_finish_key_update(struct sock *sk, struct tls_context *tls_ctx)
2517{
2518 struct tls_sw_context_rx *ctx = tls_ctx->priv_ctx_rx;
2519
2520 WRITE_ONCE(ctx->key_update_pending, false);
2521 /* wake-up pre-existing poll() */
2522 ctx->saved_data_ready(sk);
2523}
2524
2525int tls_set_sw_offload(struct sock *sk, int tx,
2526 struct tls_crypto_info *new_crypto_info)
2527{
2528 struct tls_crypto_info *crypto_info, *src_crypto_info;
2529 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2530 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2531 const struct tls_cipher_desc *cipher_desc;
2532 char *iv, *rec_seq, *key, *salt;
2533 struct cipher_context *cctx;
2534 struct tls_prot_info *prot;
2535 struct crypto_aead **aead;
2536 struct tls_context *ctx;
2537 struct crypto_tfm *tfm;
2538 int rc = 0;
2539
2540 ctx = tls_get_ctx(sk);
2541 prot = &ctx->prot_info;
2542
2543 /* new_crypto_info != NULL means rekey */
2544 if (!new_crypto_info) {
2545 if (tx) {
2546 ctx->priv_ctx_tx = init_ctx_tx(ctx, sk);
2547 if (!ctx->priv_ctx_tx)
2548 return -ENOMEM;
2549 } else {
2550 ctx->priv_ctx_rx = init_ctx_rx(ctx);
2551 if (!ctx->priv_ctx_rx)
2552 return -ENOMEM;
2553 }
2554 }
2555
2556 if (tx) {
2557 sw_ctx_tx = ctx->priv_ctx_tx;
2558 crypto_info = &ctx->crypto_send.info;
2559 cctx = &ctx->tx;
2560 aead = &sw_ctx_tx->aead_send;
2561 } else {
2562 sw_ctx_rx = ctx->priv_ctx_rx;
2563 crypto_info = &ctx->crypto_recv.info;
2564 cctx = &ctx->rx;
2565 aead = &sw_ctx_rx->aead_recv;
2566 }
2567
2568 src_crypto_info = new_crypto_info ?: crypto_info;
2569
2570 cipher_desc = get_cipher_desc(src_crypto_info->cipher_type);
2571 if (!cipher_desc) {
2572 rc = -EINVAL;
2573 goto free_priv;
2574 }
2575
2576 rc = init_prot_info(prot, src_crypto_info, cipher_desc);
2577 if (rc)
2578 goto free_priv;
2579
2580 iv = crypto_info_iv(src_crypto_info, cipher_desc);
2581 key = crypto_info_key(src_crypto_info, cipher_desc);
2582 salt = crypto_info_salt(src_crypto_info, cipher_desc);
2583 rec_seq = crypto_info_rec_seq(src_crypto_info, cipher_desc);
2584
2585 if (!*aead) {
2586 *aead = crypto_alloc_aead(cipher_desc->cipher_name, 0, 0);
2587 if (IS_ERR(*aead)) {
2588 rc = PTR_ERR(*aead);
2589 *aead = NULL;
2590 goto free_priv;
2591 }
2592 }
2593
2594 ctx->push_pending_record = tls_sw_push_pending_record;
2595
2596 /* setkey is the last operation that could fail during a
2597 * rekey. if it succeeds, we can start modifying the
2598 * context.
2599 */
2600 rc = crypto_aead_setkey(*aead, key, cipher_desc->key);
2601 if (rc) {
2602 if (new_crypto_info)
2603 goto out;
2604 else
2605 goto free_aead;
2606 }
2607
2608 if (!new_crypto_info) {
2609 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2610 if (rc)
2611 goto free_aead;
2612 }
2613
2614 if (!tx && !new_crypto_info) {
2615 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2616
2617 tls_update_rx_zc_capable(ctx);
2618 sw_ctx_rx->async_capable =
2619 src_crypto_info->version != TLS_1_3_VERSION &&
2620 !!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
2621
2622 rc = tls_strp_init(&sw_ctx_rx->strp, sk);
2623 if (rc)
2624 goto free_aead;
2625 }
2626
2627 memcpy(cctx->iv, salt, cipher_desc->salt);
2628 memcpy(cctx->iv + cipher_desc->salt, iv, cipher_desc->iv);
2629 memcpy(cctx->rec_seq, rec_seq, cipher_desc->rec_seq);
2630
2631 if (new_crypto_info) {
2632 unsafe_memcpy(crypto_info, new_crypto_info,
2633 cipher_desc->crypto_info,
2634 /* size was checked in do_tls_setsockopt_conf */);
2635 memzero_explicit(new_crypto_info, cipher_desc->crypto_info);
2636 if (!tx)
2637 tls_finish_key_update(sk, ctx);
2638 }
2639
2640 goto out;
2641
2642free_aead:
2643 crypto_free_aead(*aead);
2644 *aead = NULL;
2645free_priv:
2646 if (!new_crypto_info) {
2647 if (tx) {
2648 kfree(ctx->priv_ctx_tx);
2649 ctx->priv_ctx_tx = NULL;
2650 } else {
2651 kfree(ctx->priv_ctx_rx);
2652 ctx->priv_ctx_rx = NULL;
2653 }
2654 }
2655out:
2656 return rc;
2657}