?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'trace-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[linux/fpc-iii.git] / net / tls / tls_sw.c
blob01d933ae5f164ef3343bc2a5813bb4d13b065c38
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/sched/signal.h>
39 #include <linux/module.h>
40 #include <crypto/aead.h>
41
42 #include <net/strparser.h>
43 #include <net/tls.h>
44
45 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
46 unsigned int recursion_level)
47 {
48 int start = skb_headlen(skb);
49 int i, chunk = start - offset;
50 struct sk_buff *frag_iter;
51 int elt = 0;
52
53 if (unlikely(recursion_level >= 24))
54 return -EMSGSIZE;
55
56 if (chunk > 0) {
57 if (chunk > len)
58 chunk = len;
59 elt++;
60 len -= chunk;
61 if (len == 0)
62 return elt;
63 offset += chunk;
64 }
65
66 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
67 int end;
68
69 WARN_ON(start > offset + len);
70
71 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
72 chunk = end - offset;
73 if (chunk > 0) {
74 if (chunk > len)
75 chunk = len;
76 elt++;
77 len -= chunk;
78 if (len == 0)
79 return elt;
80 offset += chunk;
81 }
82 start = end;
83 }
84
85 if (unlikely(skb_has_frag_list(skb))) {
86 skb_walk_frags(skb, frag_iter) {
87 int end, ret;
88
89 WARN_ON(start > offset + len);
90
91 end = start + frag_iter->len;
92 chunk = end - offset;
93 if (chunk > 0) {
94 if (chunk > len)
95 chunk = len;
96 ret = __skb_nsg(frag_iter, offset - start, chunk,
97 recursion_level + 1);
98 if (unlikely(ret < 0))
99 return ret;
100 elt += ret;
101 len -= chunk;
102 if (len == 0)
103 return elt;
104 offset += chunk;
105 }
106 start = end;
107 }
108 }
109 BUG_ON(len);
110 return elt;
111 }
112
113 /* Return the number of scatterlist elements required to completely map the
114 * skb, or -EMSGSIZE if the recursion depth is exceeded.
115 */
116 static int skb_nsg(struct sk_buff *skb, int offset, int len)
117 {
118 return __skb_nsg(skb, offset, len, 0);
119 }
120
121 static int padding_length(struct tls_sw_context_rx *ctx,
122 struct tls_prot_info *prot, struct sk_buff *skb)
123 {
124 struct strp_msg *rxm = strp_msg(skb);
125 int sub = 0;
126
127 /* Determine zero-padding length */
128 if (prot->version == TLS_1_3_VERSION) {
129 char content_type = 0;
130 int err;
131 int back = 17;
132
133 while (content_type == 0) {
134 if (back > rxm->full_len - prot->prepend_size)
135 return -EBADMSG;
136 err = skb_copy_bits(skb,
137 rxm->offset + rxm->full_len - back,
138 &content_type, 1);
139 if (err)
140 return err;
141 if (content_type)
142 break;
143 sub++;
144 back++;
145 }
146 ctx->control = content_type;
147 }
148 return sub;
149 }
150
151 static void tls_decrypt_done(struct crypto_async_request *req, int err)
152 {
153 struct aead_request *aead_req = (struct aead_request *)req;
154 struct scatterlist *sgout = aead_req->dst;
155 struct scatterlist *sgin = aead_req->src;
156 struct tls_sw_context_rx *ctx;
157 struct tls_context *tls_ctx;
158 struct tls_prot_info *prot;
159 struct scatterlist *sg;
160 struct sk_buff *skb;
161 unsigned int pages;
162 int pending;
163
164 skb = (struct sk_buff *)req->data;
165 tls_ctx = tls_get_ctx(skb->sk);
166 ctx = tls_sw_ctx_rx(tls_ctx);
167 prot = &tls_ctx->prot_info;
168
169 /* Propagate if there was an err */
170 if (err) {
171 if (err == -EBADMSG)
172 TLS_INC_STATS(sock_net(skb->sk),
173 LINUX_MIB_TLSDECRYPTERROR);
174 ctx->async_wait.err = err;
175 tls_err_abort(skb->sk, err);
176 } else {
177 struct strp_msg *rxm = strp_msg(skb);
178 int pad;
179
180 pad = padding_length(ctx, prot, skb);
181 if (pad < 0) {
182 ctx->async_wait.err = pad;
183 tls_err_abort(skb->sk, pad);
184 } else {
185 rxm->full_len -= pad;
186 rxm->offset += prot->prepend_size;
187 rxm->full_len -= prot->overhead_size;
188 }
189 }
190
191 /* After using skb->sk to propagate sk through crypto async callback
192 * we need to NULL it again.
193 */
194 skb->sk = NULL;
195
196
197 /* Free the destination pages if skb was not decrypted inplace */
198 if (sgout != sgin) {
199 /* Skip the first S/G entry as it points to AAD */
200 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
201 if (!sg)
202 break;
203 put_page(sg_page(sg));
204 }
205 }
206
207 kfree(aead_req);
208
209 spin_lock_bh(&ctx->decrypt_compl_lock);
210 pending = atomic_dec_return(&ctx->decrypt_pending);
211
212 if (!pending && ctx->async_notify)
213 complete(&ctx->async_wait.completion);
214 spin_unlock_bh(&ctx->decrypt_compl_lock);
215 }
216
217 static int tls_do_decryption(struct sock *sk,
218 struct sk_buff *skb,
219 struct scatterlist *sgin,
220 struct scatterlist *sgout,
221 char *iv_recv,
222 size_t data_len,
223 struct aead_request *aead_req,
224 bool async)
225 {
226 struct tls_context *tls_ctx = tls_get_ctx(sk);
227 struct tls_prot_info *prot = &tls_ctx->prot_info;
228 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
229 int ret;
230
231 aead_request_set_tfm(aead_req, ctx->aead_recv);
232 aead_request_set_ad(aead_req, prot->aad_size);
233 aead_request_set_crypt(aead_req, sgin, sgout,
234 data_len + prot->tag_size,
235 (u8 *)iv_recv);
236
237 if (async) {
238 /* Using skb->sk to push sk through to crypto async callback
239 * handler. This allows propagating errors up to the socket
240 * if needed. It _must_ be cleared in the async handler
241 * before consume_skb is called. We _know_ skb->sk is NULL
242 * because it is a clone from strparser.
243 */
244 skb->sk = sk;
245 aead_request_set_callback(aead_req,
246 CRYPTO_TFM_REQ_MAY_BACKLOG,
247 tls_decrypt_done, skb);
248 atomic_inc(&ctx->decrypt_pending);
249 } else {
250 aead_request_set_callback(aead_req,
251 CRYPTO_TFM_REQ_MAY_BACKLOG,
252 crypto_req_done, &ctx->async_wait);
253 }
254
255 ret = crypto_aead_decrypt(aead_req);
256 if (ret == -EINPROGRESS) {
257 if (async)
258 return ret;
259
260 ret = crypto_wait_req(ret, &ctx->async_wait);
261 }
262
263 if (async)
264 atomic_dec(&ctx->decrypt_pending);
265
266 return ret;
267 }
268
269 static void tls_trim_both_msgs(struct sock *sk, int target_size)
270 {
271 struct tls_context *tls_ctx = tls_get_ctx(sk);
272 struct tls_prot_info *prot = &tls_ctx->prot_info;
273 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
274 struct tls_rec *rec = ctx->open_rec;
275
276 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
277 if (target_size > 0)
278 target_size += prot->overhead_size;
279 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
280 }
281
282 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
283 {
284 struct tls_context *tls_ctx = tls_get_ctx(sk);
285 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
286 struct tls_rec *rec = ctx->open_rec;
287 struct sk_msg *msg_en = &rec->msg_encrypted;
288
289 return sk_msg_alloc(sk, msg_en, len, 0);
290 }
291
292 static int tls_clone_plaintext_msg(struct sock *sk, int required)
293 {
294 struct tls_context *tls_ctx = tls_get_ctx(sk);
295 struct tls_prot_info *prot = &tls_ctx->prot_info;
296 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
297 struct tls_rec *rec = ctx->open_rec;
298 struct sk_msg *msg_pl = &rec->msg_plaintext;
299 struct sk_msg *msg_en = &rec->msg_encrypted;
300 int skip, len;
301
302 /* We add page references worth len bytes from encrypted sg
303 * at the end of plaintext sg. It is guaranteed that msg_en
304 * has enough required room (ensured by caller).
305 */
306 len = required - msg_pl->sg.size;
307
308 /* Skip initial bytes in msg_en's data to be able to use
309 * same offset of both plain and encrypted data.
310 */
311 skip = prot->prepend_size + msg_pl->sg.size;
312
313 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
314 }
315
316 static struct tls_rec *tls_get_rec(struct sock *sk)
317 {
318 struct tls_context *tls_ctx = tls_get_ctx(sk);
319 struct tls_prot_info *prot = &tls_ctx->prot_info;
320 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
321 struct sk_msg *msg_pl, *msg_en;
322 struct tls_rec *rec;
323 int mem_size;
324
325 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
326
327 rec = kzalloc(mem_size, sk->sk_allocation);
328 if (!rec)
329 return NULL;
330
331 msg_pl = &rec->msg_plaintext;
332 msg_en = &rec->msg_encrypted;
333
334 sk_msg_init(msg_pl);
335 sk_msg_init(msg_en);
336
337 sg_init_table(rec->sg_aead_in, 2);
338 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
339 sg_unmark_end(&rec->sg_aead_in[1]);
340
341 sg_init_table(rec->sg_aead_out, 2);
342 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
343 sg_unmark_end(&rec->sg_aead_out[1]);
344
345 return rec;
346 }
347
348 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
349 {
350 sk_msg_free(sk, &rec->msg_encrypted);
351 sk_msg_free(sk, &rec->msg_plaintext);
352 kfree(rec);
353 }
354
355 static void tls_free_open_rec(struct sock *sk)
356 {
357 struct tls_context *tls_ctx = tls_get_ctx(sk);
358 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
359 struct tls_rec *rec = ctx->open_rec;
360
361 if (rec) {
362 tls_free_rec(sk, rec);
363 ctx->open_rec = NULL;
364 }
365 }
366
367 int tls_tx_records(struct sock *sk, int flags)
368 {
369 struct tls_context *tls_ctx = tls_get_ctx(sk);
370 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
371 struct tls_rec *rec, *tmp;
372 struct sk_msg *msg_en;
373 int tx_flags, rc = 0;
374
375 if (tls_is_partially_sent_record(tls_ctx)) {
376 rec = list_first_entry(&ctx->tx_list,
377 struct tls_rec, list);
378
379 if (flags == -1)
380 tx_flags = rec->tx_flags;
381 else
382 tx_flags = flags;
383
384 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
385 if (rc)
386 goto tx_err;
387
388 /* Full record has been transmitted.
389 * Remove the head of tx_list
390 */
391 list_del(&rec->list);
392 sk_msg_free(sk, &rec->msg_plaintext);
393 kfree(rec);
394 }
395
396 /* Tx all ready records */
397 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
398 if (READ_ONCE(rec->tx_ready)) {
399 if (flags == -1)
400 tx_flags = rec->tx_flags;
401 else
402 tx_flags = flags;
403
404 msg_en = &rec->msg_encrypted;
405 rc = tls_push_sg(sk, tls_ctx,
406 &msg_en->sg.data[msg_en->sg.curr],
407 0, tx_flags);
408 if (rc)
409 goto tx_err;
410
411 list_del(&rec->list);
412 sk_msg_free(sk, &rec->msg_plaintext);
413 kfree(rec);
414 } else {
415 break;
416 }
417 }
418
419 tx_err:
420 if (rc < 0 && rc != -EAGAIN)
421 tls_err_abort(sk, EBADMSG);
422
423 return rc;
424 }
425
426 static void tls_encrypt_done(struct crypto_async_request *req, int err)
427 {
428 struct aead_request *aead_req = (struct aead_request *)req;
429 struct sock *sk = req->data;
430 struct tls_context *tls_ctx = tls_get_ctx(sk);
431 struct tls_prot_info *prot = &tls_ctx->prot_info;
432 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
433 struct scatterlist *sge;
434 struct sk_msg *msg_en;
435 struct tls_rec *rec;
436 bool ready = false;
437 int pending;
438
439 rec = container_of(aead_req, struct tls_rec, aead_req);
440 msg_en = &rec->msg_encrypted;
441
442 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
443 sge->offset -= prot->prepend_size;
444 sge->length += prot->prepend_size;
445
446 /* Check if error is previously set on socket */
447 if (err || sk->sk_err) {
448 rec = NULL;
449
450 /* If err is already set on socket, return the same code */
451 if (sk->sk_err) {
452 ctx->async_wait.err = sk->sk_err;
453 } else {
454 ctx->async_wait.err = err;
455 tls_err_abort(sk, err);
456 }
457 }
458
459 if (rec) {
460 struct tls_rec *first_rec;
461
462 /* Mark the record as ready for transmission */
463 smp_store_mb(rec->tx_ready, true);
464
465 /* If received record is at head of tx_list, schedule tx */
466 first_rec = list_first_entry(&ctx->tx_list,
467 struct tls_rec, list);
468 if (rec == first_rec)
469 ready = true;
470 }
471
472 spin_lock_bh(&ctx->encrypt_compl_lock);
473 pending = atomic_dec_return(&ctx->encrypt_pending);
474
475 if (!pending && ctx->async_notify)
476 complete(&ctx->async_wait.completion);
477 spin_unlock_bh(&ctx->encrypt_compl_lock);
478
479 if (!ready)
480 return;
481
482 /* Schedule the transmission */
483 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
484 schedule_delayed_work(&ctx->tx_work.work, 1);
485 }
486
487 static int tls_do_encryption(struct sock *sk,
488 struct tls_context *tls_ctx,
489 struct tls_sw_context_tx *ctx,
490 struct aead_request *aead_req,
491 size_t data_len, u32 start)
492 {
493 struct tls_prot_info *prot = &tls_ctx->prot_info;
494 struct tls_rec *rec = ctx->open_rec;
495 struct sk_msg *msg_en = &rec->msg_encrypted;
496 struct scatterlist *sge = sk_msg_elem(msg_en, start);
497 int rc, iv_offset = 0;
498
499 /* For CCM based ciphers, first byte of IV is a constant */
500 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
501 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
502 iv_offset = 1;
503 }
504
505 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
506 prot->iv_size + prot->salt_size);
507
508 xor_iv_with_seq(prot, rec->iv_data, tls_ctx->tx.rec_seq);
509
510 sge->offset += prot->prepend_size;
511 sge->length -= prot->prepend_size;
512
513 msg_en->sg.curr = start;
514
515 aead_request_set_tfm(aead_req, ctx->aead_send);
516 aead_request_set_ad(aead_req, prot->aad_size);
517 aead_request_set_crypt(aead_req, rec->sg_aead_in,
518 rec->sg_aead_out,
519 data_len, rec->iv_data);
520
521 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
522 tls_encrypt_done, sk);
523
524 /* Add the record in tx_list */
525 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
526 atomic_inc(&ctx->encrypt_pending);
527
528 rc = crypto_aead_encrypt(aead_req);
529 if (!rc || rc != -EINPROGRESS) {
530 atomic_dec(&ctx->encrypt_pending);
531 sge->offset -= prot->prepend_size;
532 sge->length += prot->prepend_size;
533 }
534
535 if (!rc) {
536 WRITE_ONCE(rec->tx_ready, true);
537 } else if (rc != -EINPROGRESS) {
538 list_del(&rec->list);
539 return rc;
540 }
541
542 /* Unhook the record from context if encryption is not failure */
543 ctx->open_rec = NULL;
544 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
545 return rc;
546 }
547
548 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
549 struct tls_rec **to, struct sk_msg *msg_opl,
550 struct sk_msg *msg_oen, u32 split_point,
551 u32 tx_overhead_size, u32 *orig_end)
552 {
553 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
554 struct scatterlist *sge, *osge, *nsge;
555 u32 orig_size = msg_opl->sg.size;
556 struct scatterlist tmp = { };
557 struct sk_msg *msg_npl;
558 struct tls_rec *new;
559 int ret;
560
561 new = tls_get_rec(sk);
562 if (!new)
563 return -ENOMEM;
564 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
565 tx_overhead_size, 0);
566 if (ret < 0) {
567 tls_free_rec(sk, new);
568 return ret;
569 }
570
571 *orig_end = msg_opl->sg.end;
572 i = msg_opl->sg.start;
573 sge = sk_msg_elem(msg_opl, i);
574 while (apply && sge->length) {
575 if (sge->length > apply) {
576 u32 len = sge->length - apply;
577
578 get_page(sg_page(sge));
579 sg_set_page(&tmp, sg_page(sge), len,
580 sge->offset + apply);
581 sge->length = apply;
582 bytes += apply;
583 apply = 0;
584 } else {
585 apply -= sge->length;
586 bytes += sge->length;
587 }
588
589 sk_msg_iter_var_next(i);
590 if (i == msg_opl->sg.end)
591 break;
592 sge = sk_msg_elem(msg_opl, i);
593 }
594
595 msg_opl->sg.end = i;
596 msg_opl->sg.curr = i;
597 msg_opl->sg.copybreak = 0;
598 msg_opl->apply_bytes = 0;
599 msg_opl->sg.size = bytes;
600
601 msg_npl = &new->msg_plaintext;
602 msg_npl->apply_bytes = apply;
603 msg_npl->sg.size = orig_size - bytes;
604
605 j = msg_npl->sg.start;
606 nsge = sk_msg_elem(msg_npl, j);
607 if (tmp.length) {
608 memcpy(nsge, &tmp, sizeof(*nsge));
609 sk_msg_iter_var_next(j);
610 nsge = sk_msg_elem(msg_npl, j);
611 }
612
613 osge = sk_msg_elem(msg_opl, i);
614 while (osge->length) {
615 memcpy(nsge, osge, sizeof(*nsge));
616 sg_unmark_end(nsge);
617 sk_msg_iter_var_next(i);
618 sk_msg_iter_var_next(j);
619 if (i == *orig_end)
620 break;
621 osge = sk_msg_elem(msg_opl, i);
622 nsge = sk_msg_elem(msg_npl, j);
623 }
624
625 msg_npl->sg.end = j;
626 msg_npl->sg.curr = j;
627 msg_npl->sg.copybreak = 0;
628
629 *to = new;
630 return 0;
631 }
632
633 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
634 struct tls_rec *from, u32 orig_end)
635 {
636 struct sk_msg *msg_npl = &from->msg_plaintext;
637 struct sk_msg *msg_opl = &to->msg_plaintext;
638 struct scatterlist *osge, *nsge;
639 u32 i, j;
640
641 i = msg_opl->sg.end;
642 sk_msg_iter_var_prev(i);
643 j = msg_npl->sg.start;
644
645 osge = sk_msg_elem(msg_opl, i);
646 nsge = sk_msg_elem(msg_npl, j);
647
648 if (sg_page(osge) == sg_page(nsge) &&
649 osge->offset + osge->length == nsge->offset) {
650 osge->length += nsge->length;
651 put_page(sg_page(nsge));
652 }
653
654 msg_opl->sg.end = orig_end;
655 msg_opl->sg.curr = orig_end;
656 msg_opl->sg.copybreak = 0;
657 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
658 msg_opl->sg.size += msg_npl->sg.size;
659
660 sk_msg_free(sk, &to->msg_encrypted);
661 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
662
663 kfree(from);
664 }
665
666 static int tls_push_record(struct sock *sk, int flags,
667 unsigned char record_type)
668 {
669 struct tls_context *tls_ctx = tls_get_ctx(sk);
670 struct tls_prot_info *prot = &tls_ctx->prot_info;
671 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
672 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
673 u32 i, split_point, orig_end;
674 struct sk_msg *msg_pl, *msg_en;
675 struct aead_request *req;
676 bool split;
677 int rc;
678
679 if (!rec)
680 return 0;
681
682 msg_pl = &rec->msg_plaintext;
683 msg_en = &rec->msg_encrypted;
684
685 split_point = msg_pl->apply_bytes;
686 split = split_point && split_point < msg_pl->sg.size;
687 if (unlikely((!split &&
688 msg_pl->sg.size +
689 prot->overhead_size > msg_en->sg.size) ||
690 (split &&
691 split_point +
692 prot->overhead_size > msg_en->sg.size))) {
693 split = true;
694 split_point = msg_en->sg.size;
695 }
696 if (split) {
697 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
698 split_point, prot->overhead_size,
699 &orig_end);
700 if (rc < 0)
701 return rc;
702 /* This can happen if above tls_split_open_record allocates
703 * a single large encryption buffer instead of two smaller
704 * ones. In this case adjust pointers and continue without
705 * split.
706 */
707 if (!msg_pl->sg.size) {
708 tls_merge_open_record(sk, rec, tmp, orig_end);
709 msg_pl = &rec->msg_plaintext;
710 msg_en = &rec->msg_encrypted;
711 split = false;
712 }
713 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
714 prot->overhead_size);
715 }
716
717 rec->tx_flags = flags;
718 req = &rec->aead_req;
719
720 i = msg_pl->sg.end;
721 sk_msg_iter_var_prev(i);
722
723 rec->content_type = record_type;
724 if (prot->version == TLS_1_3_VERSION) {
725 /* Add content type to end of message. No padding added */
726 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
727 sg_mark_end(&rec->sg_content_type);
728 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
729 &rec->sg_content_type);
730 } else {
731 sg_mark_end(sk_msg_elem(msg_pl, i));
732 }
733
734 if (msg_pl->sg.end < msg_pl->sg.start) {
735 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
736 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
737 msg_pl->sg.data);
738 }
739
740 i = msg_pl->sg.start;
741 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
742
743 i = msg_en->sg.end;
744 sk_msg_iter_var_prev(i);
745 sg_mark_end(sk_msg_elem(msg_en, i));
746
747 i = msg_en->sg.start;
748 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
749
750 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
751 tls_ctx->tx.rec_seq, record_type, prot);
752
753 tls_fill_prepend(tls_ctx,
754 page_address(sg_page(&msg_en->sg.data[i])) +
755 msg_en->sg.data[i].offset,
756 msg_pl->sg.size + prot->tail_size,
757 record_type);
758
759 tls_ctx->pending_open_record_frags = false;
760
761 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
762 msg_pl->sg.size + prot->tail_size, i);
763 if (rc < 0) {
764 if (rc != -EINPROGRESS) {
765 tls_err_abort(sk, EBADMSG);
766 if (split) {
767 tls_ctx->pending_open_record_frags = true;
768 tls_merge_open_record(sk, rec, tmp, orig_end);
769 }
770 }
771 ctx->async_capable = 1;
772 return rc;
773 } else if (split) {
774 msg_pl = &tmp->msg_plaintext;
775 msg_en = &tmp->msg_encrypted;
776 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
777 tls_ctx->pending_open_record_frags = true;
778 ctx->open_rec = tmp;
779 }
780
781 return tls_tx_records(sk, flags);
782 }
783
784 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
785 bool full_record, u8 record_type,
786 ssize_t *copied, int flags)
787 {
788 struct tls_context *tls_ctx = tls_get_ctx(sk);
789 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
790 struct sk_msg msg_redir = { };
791 struct sk_psock *psock;
792 struct sock *sk_redir;
793 struct tls_rec *rec;
794 bool enospc, policy;
795 int err = 0, send;
796 u32 delta = 0;
797
798 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
799 psock = sk_psock_get(sk);
800 if (!psock || !policy) {
801 err = tls_push_record(sk, flags, record_type);
802 if (err && sk->sk_err == EBADMSG) {
803 *copied -= sk_msg_free(sk, msg);
804 tls_free_open_rec(sk);
805 err = -sk->sk_err;
806 }
807 if (psock)
808 sk_psock_put(sk, psock);
809 return err;
810 }
811 more_data:
812 enospc = sk_msg_full(msg);
813 if (psock->eval == __SK_NONE) {
814 delta = msg->sg.size;
815 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
816 delta -= msg->sg.size;
817 }
818 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
819 !enospc && !full_record) {
820 err = -ENOSPC;
821 goto out_err;
822 }
823 msg->cork_bytes = 0;
824 send = msg->sg.size;
825 if (msg->apply_bytes && msg->apply_bytes < send)
826 send = msg->apply_bytes;
827
828 switch (psock->eval) {
829 case __SK_PASS:
830 err = tls_push_record(sk, flags, record_type);
831 if (err && sk->sk_err == EBADMSG) {
832 *copied -= sk_msg_free(sk, msg);
833 tls_free_open_rec(sk);
834 err = -sk->sk_err;
835 goto out_err;
836 }
837 break;
838 case __SK_REDIRECT:
839 sk_redir = psock->sk_redir;
840 memcpy(&msg_redir, msg, sizeof(*msg));
841 if (msg->apply_bytes < send)
842 msg->apply_bytes = 0;
843 else
844 msg->apply_bytes -= send;
845 sk_msg_return_zero(sk, msg, send);
846 msg->sg.size -= send;
847 release_sock(sk);
848 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
849 lock_sock(sk);
850 if (err < 0) {
851 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
852 msg->sg.size = 0;
853 }
854 if (msg->sg.size == 0)
855 tls_free_open_rec(sk);
856 break;
857 case __SK_DROP:
858 default:
859 sk_msg_free_partial(sk, msg, send);
860 if (msg->apply_bytes < send)
861 msg->apply_bytes = 0;
862 else
863 msg->apply_bytes -= send;
864 if (msg->sg.size == 0)
865 tls_free_open_rec(sk);
866 *copied -= (send + delta);
867 err = -EACCES;
868 }
869
870 if (likely(!err)) {
871 bool reset_eval = !ctx->open_rec;
872
873 rec = ctx->open_rec;
874 if (rec) {
875 msg = &rec->msg_plaintext;
876 if (!msg->apply_bytes)
877 reset_eval = true;
878 }
879 if (reset_eval) {
880 psock->eval = __SK_NONE;
881 if (psock->sk_redir) {
882 sock_put(psock->sk_redir);
883 psock->sk_redir = NULL;
884 }
885 }
886 if (rec)
887 goto more_data;
888 }
889 out_err:
890 sk_psock_put(sk, psock);
891 return err;
892 }
893
894 static int tls_sw_push_pending_record(struct sock *sk, int flags)
895 {
896 struct tls_context *tls_ctx = tls_get_ctx(sk);
897 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
898 struct tls_rec *rec = ctx->open_rec;
899 struct sk_msg *msg_pl;
900 size_t copied;
901
902 if (!rec)
903 return 0;
904
905 msg_pl = &rec->msg_plaintext;
906 copied = msg_pl->sg.size;
907 if (!copied)
908 return 0;
909
910 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
911 &copied, flags);
912 }
913
914 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
915 {
916 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
917 struct tls_context *tls_ctx = tls_get_ctx(sk);
918 struct tls_prot_info *prot = &tls_ctx->prot_info;
919 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
920 bool async_capable = ctx->async_capable;
921 unsigned char record_type = TLS_RECORD_TYPE_DATA;
922 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
923 bool eor = !(msg->msg_flags & MSG_MORE);
924 size_t try_to_copy;
925 ssize_t copied = 0;
926 struct sk_msg *msg_pl, *msg_en;
927 struct tls_rec *rec;
928 int required_size;
929 int num_async = 0;
930 bool full_record;
931 int record_room;
932 int num_zc = 0;
933 int orig_size;
934 int ret = 0;
935 int pending;
936
937 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
938 MSG_CMSG_COMPAT))
939 return -EOPNOTSUPP;
940
941 mutex_lock(&tls_ctx->tx_lock);
942 lock_sock(sk);
943
944 if (unlikely(msg->msg_controllen)) {
945 ret = tls_proccess_cmsg(sk, msg, &record_type);
946 if (ret) {
947 if (ret == -EINPROGRESS)
948 num_async++;
949 else if (ret != -EAGAIN)
950 goto send_end;
951 }
952 }
953
954 while (msg_data_left(msg)) {
955 if (sk->sk_err) {
956 ret = -sk->sk_err;
957 goto send_end;
958 }
959
960 if (ctx->open_rec)
961 rec = ctx->open_rec;
962 else
963 rec = ctx->open_rec = tls_get_rec(sk);
964 if (!rec) {
965 ret = -ENOMEM;
966 goto send_end;
967 }
968
969 msg_pl = &rec->msg_plaintext;
970 msg_en = &rec->msg_encrypted;
971
972 orig_size = msg_pl->sg.size;
973 full_record = false;
974 try_to_copy = msg_data_left(msg);
975 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
976 if (try_to_copy >= record_room) {
977 try_to_copy = record_room;
978 full_record = true;
979 }
980
981 required_size = msg_pl->sg.size + try_to_copy +
982 prot->overhead_size;
983
984 if (!sk_stream_memory_free(sk))
985 goto wait_for_sndbuf;
986
987 alloc_encrypted:
988 ret = tls_alloc_encrypted_msg(sk, required_size);
989 if (ret) {
990 if (ret != -ENOSPC)
991 goto wait_for_memory;
992
993 /* Adjust try_to_copy according to the amount that was
994 * actually allocated. The difference is due
995 * to max sg elements limit
996 */
997 try_to_copy -= required_size - msg_en->sg.size;
998 full_record = true;
999 }
1000
1001 if (!is_kvec && (full_record || eor) && !async_capable) {
1002 u32 first = msg_pl->sg.end;
1003
1004 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1005 msg_pl, try_to_copy);
1006 if (ret)
1007 goto fallback_to_reg_send;
1008
1009 num_zc++;
1010 copied += try_to_copy;
1011
1012 sk_msg_sg_copy_set(msg_pl, first);
1013 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1014 record_type, &copied,
1015 msg->msg_flags);
1016 if (ret) {
1017 if (ret == -EINPROGRESS)
1018 num_async++;
1019 else if (ret == -ENOMEM)
1020 goto wait_for_memory;
1021 else if (ctx->open_rec && ret == -ENOSPC)
1022 goto rollback_iter;
1023 else if (ret != -EAGAIN)
1024 goto send_end;
1025 }
1026 continue;
1027 rollback_iter:
1028 copied -= try_to_copy;
1029 sk_msg_sg_copy_clear(msg_pl, first);
1030 iov_iter_revert(&msg->msg_iter,
1031 msg_pl->sg.size - orig_size);
1032 fallback_to_reg_send:
1033 sk_msg_trim(sk, msg_pl, orig_size);
1034 }
1035
1036 required_size = msg_pl->sg.size + try_to_copy;
1037
1038 ret = tls_clone_plaintext_msg(sk, required_size);
1039 if (ret) {
1040 if (ret != -ENOSPC)
1041 goto send_end;
1042
1043 /* Adjust try_to_copy according to the amount that was
1044 * actually allocated. The difference is due
1045 * to max sg elements limit
1046 */
1047 try_to_copy -= required_size - msg_pl->sg.size;
1048 full_record = true;
1049 sk_msg_trim(sk, msg_en,
1050 msg_pl->sg.size + prot->overhead_size);
1051 }
1052
1053 if (try_to_copy) {
1054 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1055 msg_pl, try_to_copy);
1056 if (ret < 0)
1057 goto trim_sgl;
1058 }
1059
1060 /* Open records defined only if successfully copied, otherwise
1061 * we would trim the sg but not reset the open record frags.
1062 */
1063 tls_ctx->pending_open_record_frags = true;
1064 copied += try_to_copy;
1065 if (full_record || eor) {
1066 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1067 record_type, &copied,
1068 msg->msg_flags);
1069 if (ret) {
1070 if (ret == -EINPROGRESS)
1071 num_async++;
1072 else if (ret == -ENOMEM)
1073 goto wait_for_memory;
1074 else if (ret != -EAGAIN) {
1075 if (ret == -ENOSPC)
1076 ret = 0;
1077 goto send_end;
1078 }
1079 }
1080 }
1081
1082 continue;
1083
1084 wait_for_sndbuf:
1085 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1086 wait_for_memory:
1087 ret = sk_stream_wait_memory(sk, &timeo);
1088 if (ret) {
1089 trim_sgl:
1090 if (ctx->open_rec)
1091 tls_trim_both_msgs(sk, orig_size);
1092 goto send_end;
1093 }
1094
1095 if (ctx->open_rec && msg_en->sg.size < required_size)
1096 goto alloc_encrypted;
1097 }
1098
1099 if (!num_async) {
1100 goto send_end;
1101 } else if (num_zc) {
1102 /* Wait for pending encryptions to get completed */
1103 spin_lock_bh(&ctx->encrypt_compl_lock);
1104 ctx->async_notify = true;
1105
1106 pending = atomic_read(&ctx->encrypt_pending);
1107 spin_unlock_bh(&ctx->encrypt_compl_lock);
1108 if (pending)
1109 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1110 else
1111 reinit_completion(&ctx->async_wait.completion);
1112
1113 /* There can be no concurrent accesses, since we have no
1114 * pending encrypt operations
1115 */
1116 WRITE_ONCE(ctx->async_notify, false);
1117
1118 if (ctx->async_wait.err) {
1119 ret = ctx->async_wait.err;
1120 copied = 0;
1121 }
1122 }
1123
1124 /* Transmit if any encryptions have completed */
1125 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1126 cancel_delayed_work(&ctx->tx_work.work);
1127 tls_tx_records(sk, msg->msg_flags);
1128 }
1129
1130 send_end:
1131 ret = sk_stream_error(sk, msg->msg_flags, ret);
1132
1133 release_sock(sk);
1134 mutex_unlock(&tls_ctx->tx_lock);
1135 return copied > 0 ? copied : ret;
1136 }
1137
1138 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1139 int offset, size_t size, int flags)
1140 {
1141 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1142 struct tls_context *tls_ctx = tls_get_ctx(sk);
1143 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1144 struct tls_prot_info *prot = &tls_ctx->prot_info;
1145 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1146 struct sk_msg *msg_pl;
1147 struct tls_rec *rec;
1148 int num_async = 0;
1149 ssize_t copied = 0;
1150 bool full_record;
1151 int record_room;
1152 int ret = 0;
1153 bool eor;
1154
1155 eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1156 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1157
1158 /* Call the sk_stream functions to manage the sndbuf mem. */
1159 while (size > 0) {
1160 size_t copy, required_size;
1161
1162 if (sk->sk_err) {
1163 ret = -sk->sk_err;
1164 goto sendpage_end;
1165 }
1166
1167 if (ctx->open_rec)
1168 rec = ctx->open_rec;
1169 else
1170 rec = ctx->open_rec = tls_get_rec(sk);
1171 if (!rec) {
1172 ret = -ENOMEM;
1173 goto sendpage_end;
1174 }
1175
1176 msg_pl = &rec->msg_plaintext;
1177
1178 full_record = false;
1179 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1180 copy = size;
1181 if (copy >= record_room) {
1182 copy = record_room;
1183 full_record = true;
1184 }
1185
1186 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1187
1188 if (!sk_stream_memory_free(sk))
1189 goto wait_for_sndbuf;
1190 alloc_payload:
1191 ret = tls_alloc_encrypted_msg(sk, required_size);
1192 if (ret) {
1193 if (ret != -ENOSPC)
1194 goto wait_for_memory;
1195
1196 /* Adjust copy according to the amount that was
1197 * actually allocated. The difference is due
1198 * to max sg elements limit
1199 */
1200 copy -= required_size - msg_pl->sg.size;
1201 full_record = true;
1202 }
1203
1204 sk_msg_page_add(msg_pl, page, copy, offset);
1205 sk_mem_charge(sk, copy);
1206
1207 offset += copy;
1208 size -= copy;
1209 copied += copy;
1210
1211 tls_ctx->pending_open_record_frags = true;
1212 if (full_record || eor || sk_msg_full(msg_pl)) {
1213 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1214 record_type, &copied, flags);
1215 if (ret) {
1216 if (ret == -EINPROGRESS)
1217 num_async++;
1218 else if (ret == -ENOMEM)
1219 goto wait_for_memory;
1220 else if (ret != -EAGAIN) {
1221 if (ret == -ENOSPC)
1222 ret = 0;
1223 goto sendpage_end;
1224 }
1225 }
1226 }
1227 continue;
1228 wait_for_sndbuf:
1229 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1230 wait_for_memory:
1231 ret = sk_stream_wait_memory(sk, &timeo);
1232 if (ret) {
1233 if (ctx->open_rec)
1234 tls_trim_both_msgs(sk, msg_pl->sg.size);
1235 goto sendpage_end;
1236 }
1237
1238 if (ctx->open_rec)
1239 goto alloc_payload;
1240 }
1241
1242 if (num_async) {
1243 /* Transmit if any encryptions have completed */
1244 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1245 cancel_delayed_work(&ctx->tx_work.work);
1246 tls_tx_records(sk, flags);
1247 }
1248 }
1249 sendpage_end:
1250 ret = sk_stream_error(sk, flags, ret);
1251 return copied > 0 ? copied : ret;
1252 }
1253
1254 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1255 int offset, size_t size, int flags)
1256 {
1257 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1258 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1259 MSG_NO_SHARED_FRAGS))
1260 return -EOPNOTSUPP;
1261
1262 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1263 }
1264
1265 int tls_sw_sendpage(struct sock *sk, struct page *page,
1266 int offset, size_t size, int flags)
1267 {
1268 struct tls_context *tls_ctx = tls_get_ctx(sk);
1269 int ret;
1270
1271 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1272 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1273 return -EOPNOTSUPP;
1274
1275 mutex_lock(&tls_ctx->tx_lock);
1276 lock_sock(sk);
1277 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1278 release_sock(sk);
1279 mutex_unlock(&tls_ctx->tx_lock);
1280 return ret;
1281 }
1282
1283 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1284 int flags, long timeo, int *err)
1285 {
1286 struct tls_context *tls_ctx = tls_get_ctx(sk);
1287 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1288 struct sk_buff *skb;
1289 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1290
1291 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1292 if (sk->sk_err) {
1293 *err = sock_error(sk);
1294 return NULL;
1295 }
1296
1297 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1298 __strp_unpause(&ctx->strp);
1299 if (ctx->recv_pkt)
1300 return ctx->recv_pkt;
1301 }
1302
1303 if (sk->sk_shutdown & RCV_SHUTDOWN)
1304 return NULL;
1305
1306 if (sock_flag(sk, SOCK_DONE))
1307 return NULL;
1308
1309 if ((flags & MSG_DONTWAIT) || !timeo) {
1310 *err = -EAGAIN;
1311 return NULL;
1312 }
1313
1314 add_wait_queue(sk_sleep(sk), &wait);
1315 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1316 sk_wait_event(sk, &timeo,
1317 ctx->recv_pkt != skb ||
1318 !sk_psock_queue_empty(psock),
1319 &wait);
1320 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1321 remove_wait_queue(sk_sleep(sk), &wait);
1322
1323 /* Handle signals */
1324 if (signal_pending(current)) {
1325 *err = sock_intr_errno(timeo);
1326 return NULL;
1327 }
1328 }
1329
1330 return skb;
1331 }
1332
1333 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1334 int length, int *pages_used,
1335 unsigned int *size_used,
1336 struct scatterlist *to,
1337 int to_max_pages)
1338 {
1339 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1340 struct page *pages[MAX_SKB_FRAGS];
1341 unsigned int size = *size_used;
1342 ssize_t copied, use;
1343 size_t offset;
1344
1345 while (length > 0) {
1346 i = 0;
1347 maxpages = to_max_pages - num_elem;
1348 if (maxpages == 0) {
1349 rc = -EFAULT;
1350 goto out;
1351 }
1352 copied = iov_iter_get_pages(from, pages,
1353 length,
1354 maxpages, &offset);
1355 if (copied <= 0) {
1356 rc = -EFAULT;
1357 goto out;
1358 }
1359
1360 iov_iter_advance(from, copied);
1361
1362 length -= copied;
1363 size += copied;
1364 while (copied) {
1365 use = min_t(int, copied, PAGE_SIZE - offset);
1366
1367 sg_set_page(&to[num_elem],
1368 pages[i], use, offset);
1369 sg_unmark_end(&to[num_elem]);
1370 /* We do not uncharge memory from this API */
1371
1372 offset = 0;
1373 copied -= use;
1374
1375 i++;
1376 num_elem++;
1377 }
1378 }
1379 /* Mark the end in the last sg entry if newly added */
1380 if (num_elem > *pages_used)
1381 sg_mark_end(&to[num_elem - 1]);
1382 out:
1383 if (rc)
1384 iov_iter_revert(from, size - *size_used);
1385 *size_used = size;
1386 *pages_used = num_elem;
1387
1388 return rc;
1389 }
1390
1391 /* This function decrypts the input skb into either out_iov or in out_sg
1392 * or in skb buffers itself. The input parameter 'zc' indicates if
1393 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1394 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1395 * NULL, then the decryption happens inside skb buffers itself, i.e.
1396 * zero-copy gets disabled and 'zc' is updated.
1397 */
1398
1399 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1400 struct iov_iter *out_iov,
1401 struct scatterlist *out_sg,
1402 int *chunk, bool *zc, bool async)
1403 {
1404 struct tls_context *tls_ctx = tls_get_ctx(sk);
1405 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1406 struct tls_prot_info *prot = &tls_ctx->prot_info;
1407 struct strp_msg *rxm = strp_msg(skb);
1408 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1409 struct aead_request *aead_req;
1410 struct sk_buff *unused;
1411 u8 *aad, *iv, *mem = NULL;
1412 struct scatterlist *sgin = NULL;
1413 struct scatterlist *sgout = NULL;
1414 const int data_len = rxm->full_len - prot->overhead_size +
1415 prot->tail_size;
1416 int iv_offset = 0;
1417
1418 if (*zc && (out_iov || out_sg)) {
1419 if (out_iov)
1420 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1421 else
1422 n_sgout = sg_nents(out_sg);
1423 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1424 rxm->full_len - prot->prepend_size);
1425 } else {
1426 n_sgout = 0;
1427 *zc = false;
1428 n_sgin = skb_cow_data(skb, 0, &unused);
1429 }
1430
1431 if (n_sgin < 1)
1432 return -EBADMSG;
1433
1434 /* Increment to accommodate AAD */
1435 n_sgin = n_sgin + 1;
1436
1437 nsg = n_sgin + n_sgout;
1438
1439 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1440 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1441 mem_size = mem_size + prot->aad_size;
1442 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1443
1444 /* Allocate a single block of memory which contains
1445 * aead_req || sgin[] || sgout[] || aad || iv.
1446 * This order achieves correct alignment for aead_req, sgin, sgout.
1447 */
1448 mem = kmalloc(mem_size, sk->sk_allocation);
1449 if (!mem)
1450 return -ENOMEM;
1451
1452 /* Segment the allocated memory */
1453 aead_req = (struct aead_request *)mem;
1454 sgin = (struct scatterlist *)(mem + aead_size);
1455 sgout = sgin + n_sgin;
1456 aad = (u8 *)(sgout + n_sgout);
1457 iv = aad + prot->aad_size;
1458
1459 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1460 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1461 iv[0] = 2;
1462 iv_offset = 1;
1463 }
1464
1465 /* Prepare IV */
1466 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1467 iv + iv_offset + prot->salt_size,
1468 prot->iv_size);
1469 if (err < 0) {
1470 kfree(mem);
1471 return err;
1472 }
1473 if (prot->version == TLS_1_3_VERSION ||
1474 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1475 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1476 crypto_aead_ivsize(ctx->aead_recv));
1477 else
1478 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1479
1480 xor_iv_with_seq(prot, iv, tls_ctx->rx.rec_seq);
1481
1482 /* Prepare AAD */
1483 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1484 prot->tail_size,
1485 tls_ctx->rx.rec_seq, ctx->control, prot);
1486
1487 /* Prepare sgin */
1488 sg_init_table(sgin, n_sgin);
1489 sg_set_buf(&sgin[0], aad, prot->aad_size);
1490 err = skb_to_sgvec(skb, &sgin[1],
1491 rxm->offset + prot->prepend_size,
1492 rxm->full_len - prot->prepend_size);
1493 if (err < 0) {
1494 kfree(mem);
1495 return err;
1496 }
1497
1498 if (n_sgout) {
1499 if (out_iov) {
1500 sg_init_table(sgout, n_sgout);
1501 sg_set_buf(&sgout[0], aad, prot->aad_size);
1502
1503 *chunk = 0;
1504 err = tls_setup_from_iter(sk, out_iov, data_len,
1505 &pages, chunk, &sgout[1],
1506 (n_sgout - 1));
1507 if (err < 0)
1508 goto fallback_to_reg_recv;
1509 } else if (out_sg) {
1510 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1511 } else {
1512 goto fallback_to_reg_recv;
1513 }
1514 } else {
1515 fallback_to_reg_recv:
1516 sgout = sgin;
1517 pages = 0;
1518 *chunk = data_len;
1519 *zc = false;
1520 }
1521
1522 /* Prepare and submit AEAD request */
1523 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1524 data_len, aead_req, async);
1525 if (err == -EINPROGRESS)
1526 return err;
1527
1528 /* Release the pages in case iov was mapped to pages */
1529 for (; pages > 0; pages--)
1530 put_page(sg_page(&sgout[pages]));
1531
1532 kfree(mem);
1533 return err;
1534 }
1535
1536 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1537 struct iov_iter *dest, int *chunk, bool *zc,
1538 bool async)
1539 {
1540 struct tls_context *tls_ctx = tls_get_ctx(sk);
1541 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1542 struct tls_prot_info *prot = &tls_ctx->prot_info;
1543 struct strp_msg *rxm = strp_msg(skb);
1544 int pad, err = 0;
1545
1546 if (!ctx->decrypted) {
1547 if (tls_ctx->rx_conf == TLS_HW) {
1548 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1549 if (err < 0)
1550 return err;
1551 }
1552
1553 /* Still not decrypted after tls_device */
1554 if (!ctx->decrypted) {
1555 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1556 async);
1557 if (err < 0) {
1558 if (err == -EINPROGRESS)
1559 tls_advance_record_sn(sk, prot,
1560 &tls_ctx->rx);
1561 else if (err == -EBADMSG)
1562 TLS_INC_STATS(sock_net(sk),
1563 LINUX_MIB_TLSDECRYPTERROR);
1564 return err;
1565 }
1566 } else {
1567 *zc = false;
1568 }
1569
1570 pad = padding_length(ctx, prot, skb);
1571 if (pad < 0)
1572 return pad;
1573
1574 rxm->full_len -= pad;
1575 rxm->offset += prot->prepend_size;
1576 rxm->full_len -= prot->overhead_size;
1577 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1578 ctx->decrypted = 1;
1579 ctx->saved_data_ready(sk);
1580 } else {
1581 *zc = false;
1582 }
1583
1584 return err;
1585 }
1586
1587 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1588 struct scatterlist *sgout)
1589 {
1590 bool zc = true;
1591 int chunk;
1592
1593 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1594 }
1595
1596 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1597 unsigned int len)
1598 {
1599 struct tls_context *tls_ctx = tls_get_ctx(sk);
1600 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1601
1602 if (skb) {
1603 struct strp_msg *rxm = strp_msg(skb);
1604
1605 if (len < rxm->full_len) {
1606 rxm->offset += len;
1607 rxm->full_len -= len;
1608 return false;
1609 }
1610 consume_skb(skb);
1611 }
1612
1613 /* Finished with message */
1614 ctx->recv_pkt = NULL;
1615 __strp_unpause(&ctx->strp);
1616
1617 return true;
1618 }
1619
1620 /* This function traverses the rx_list in tls receive context to copies the
1621 * decrypted records into the buffer provided by caller zero copy is not
1622 * true. Further, the records are removed from the rx_list if it is not a peek
1623 * case and the record has been consumed completely.
1624 */
1625 static int process_rx_list(struct tls_sw_context_rx *ctx,
1626 struct msghdr *msg,
1627 u8 *control,
1628 bool *cmsg,
1629 size_t skip,
1630 size_t len,
1631 bool zc,
1632 bool is_peek)
1633 {
1634 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1635 u8 ctrl = *control;
1636 u8 msgc = *cmsg;
1637 struct tls_msg *tlm;
1638 ssize_t copied = 0;
1639
1640 /* Set the record type in 'control' if caller didn't pass it */
1641 if (!ctrl && skb) {
1642 tlm = tls_msg(skb);
1643 ctrl = tlm->control;
1644 }
1645
1646 while (skip && skb) {
1647 struct strp_msg *rxm = strp_msg(skb);
1648 tlm = tls_msg(skb);
1649
1650 /* Cannot process a record of different type */
1651 if (ctrl != tlm->control)
1652 return 0;
1653
1654 if (skip < rxm->full_len)
1655 break;
1656
1657 skip = skip - rxm->full_len;
1658 skb = skb_peek_next(skb, &ctx->rx_list);
1659 }
1660
1661 while (len && skb) {
1662 struct sk_buff *next_skb;
1663 struct strp_msg *rxm = strp_msg(skb);
1664 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1665
1666 tlm = tls_msg(skb);
1667
1668 /* Cannot process a record of different type */
1669 if (ctrl != tlm->control)
1670 return 0;
1671
1672 /* Set record type if not already done. For a non-data record,
1673 * do not proceed if record type could not be copied.
1674 */
1675 if (!msgc) {
1676 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1677 sizeof(ctrl), &ctrl);
1678 msgc = true;
1679 if (ctrl != TLS_RECORD_TYPE_DATA) {
1680 if (cerr || msg->msg_flags & MSG_CTRUNC)
1681 return -EIO;
1682
1683 *cmsg = msgc;
1684 }
1685 }
1686
1687 if (!zc || (rxm->full_len - skip) > len) {
1688 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1689 msg, chunk);
1690 if (err < 0)
1691 return err;
1692 }
1693
1694 len = len - chunk;
1695 copied = copied + chunk;
1696
1697 /* Consume the data from record if it is non-peek case*/
1698 if (!is_peek) {
1699 rxm->offset = rxm->offset + chunk;
1700 rxm->full_len = rxm->full_len - chunk;
1701
1702 /* Return if there is unconsumed data in the record */
1703 if (rxm->full_len - skip)
1704 break;
1705 }
1706
1707 /* The remaining skip-bytes must lie in 1st record in rx_list.
1708 * So from the 2nd record, 'skip' should be 0.
1709 */
1710 skip = 0;
1711
1712 if (msg)
1713 msg->msg_flags |= MSG_EOR;
1714
1715 next_skb = skb_peek_next(skb, &ctx->rx_list);
1716
1717 if (!is_peek) {
1718 skb_unlink(skb, &ctx->rx_list);
1719 consume_skb(skb);
1720 }
1721
1722 skb = next_skb;
1723 }
1724
1725 *control = ctrl;
1726 return copied;
1727 }
1728
1729 int tls_sw_recvmsg(struct sock *sk,
1730 struct msghdr *msg,
1731 size_t len,
1732 int nonblock,
1733 int flags,
1734 int *addr_len)
1735 {
1736 struct tls_context *tls_ctx = tls_get_ctx(sk);
1737 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1738 struct tls_prot_info *prot = &tls_ctx->prot_info;
1739 struct sk_psock *psock;
1740 unsigned char control = 0;
1741 ssize_t decrypted = 0;
1742 struct strp_msg *rxm;
1743 struct tls_msg *tlm;
1744 struct sk_buff *skb;
1745 ssize_t copied = 0;
1746 bool cmsg = false;
1747 int target, err = 0;
1748 long timeo;
1749 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1750 bool is_peek = flags & MSG_PEEK;
1751 bool bpf_strp_enabled;
1752 int num_async = 0;
1753 int pending;
1754
1755 flags |= nonblock;
1756
1757 if (unlikely(flags & MSG_ERRQUEUE))
1758 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1759
1760 psock = sk_psock_get(sk);
1761 lock_sock(sk);
1762 bpf_strp_enabled = sk_psock_strp_enabled(psock);
1763
1764 /* Process pending decrypted records. It must be non-zero-copy */
1765 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1766 is_peek);
1767 if (err < 0) {
1768 tls_err_abort(sk, err);
1769 goto end;
1770 } else {
1771 copied = err;
1772 }
1773
1774 if (len <= copied)
1775 goto recv_end;
1776
1777 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1778 len = len - copied;
1779 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1780
1781 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1782 bool retain_skb = false;
1783 bool zc = false;
1784 int to_decrypt;
1785 int chunk = 0;
1786 bool async_capable;
1787 bool async = false;
1788
1789 skb = tls_wait_data(sk, psock, flags, timeo, &err);
1790 if (!skb) {
1791 if (psock) {
1792 int ret = __tcp_bpf_recvmsg(sk, psock,
1793 msg, len, flags);
1794
1795 if (ret > 0) {
1796 decrypted += ret;
1797 len -= ret;
1798 continue;
1799 }
1800 }
1801 goto recv_end;
1802 } else {
1803 tlm = tls_msg(skb);
1804 if (prot->version == TLS_1_3_VERSION)
1805 tlm->control = 0;
1806 else
1807 tlm->control = ctx->control;
1808 }
1809
1810 rxm = strp_msg(skb);
1811
1812 to_decrypt = rxm->full_len - prot->overhead_size;
1813
1814 if (to_decrypt <= len && !is_kvec && !is_peek &&
1815 ctx->control == TLS_RECORD_TYPE_DATA &&
1816 prot->version != TLS_1_3_VERSION &&
1817 !bpf_strp_enabled)
1818 zc = true;
1819
1820 /* Do not use async mode if record is non-data */
1821 if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1822 async_capable = ctx->async_capable;
1823 else
1824 async_capable = false;
1825
1826 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1827 &chunk, &zc, async_capable);
1828 if (err < 0 && err != -EINPROGRESS) {
1829 tls_err_abort(sk, EBADMSG);
1830 goto recv_end;
1831 }
1832
1833 if (err == -EINPROGRESS) {
1834 async = true;
1835 num_async++;
1836 } else if (prot->version == TLS_1_3_VERSION) {
1837 tlm->control = ctx->control;
1838 }
1839
1840 /* If the type of records being processed is not known yet,
1841 * set it to record type just dequeued. If it is already known,
1842 * but does not match the record type just dequeued, go to end.
1843 * We always get record type here since for tls1.2, record type
1844 * is known just after record is dequeued from stream parser.
1845 * For tls1.3, we disable async.
1846 */
1847
1848 if (!control)
1849 control = tlm->control;
1850 else if (control != tlm->control)
1851 goto recv_end;
1852
1853 if (!cmsg) {
1854 int cerr;
1855
1856 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1857 sizeof(control), &control);
1858 cmsg = true;
1859 if (control != TLS_RECORD_TYPE_DATA) {
1860 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1861 err = -EIO;
1862 goto recv_end;
1863 }
1864 }
1865 }
1866
1867 if (async)
1868 goto pick_next_record;
1869
1870 if (!zc) {
1871 if (bpf_strp_enabled) {
1872 err = sk_psock_tls_strp_read(psock, skb);
1873 if (err != __SK_PASS) {
1874 rxm->offset = rxm->offset + rxm->full_len;
1875 rxm->full_len = 0;
1876 if (err == __SK_DROP)
1877 consume_skb(skb);
1878 ctx->recv_pkt = NULL;
1879 __strp_unpause(&ctx->strp);
1880 continue;
1881 }
1882 }
1883
1884 if (rxm->full_len > len) {
1885 retain_skb = true;
1886 chunk = len;
1887 } else {
1888 chunk = rxm->full_len;
1889 }
1890
1891 err = skb_copy_datagram_msg(skb, rxm->offset,
1892 msg, chunk);
1893 if (err < 0)
1894 goto recv_end;
1895
1896 if (!is_peek) {
1897 rxm->offset = rxm->offset + chunk;
1898 rxm->full_len = rxm->full_len - chunk;
1899 }
1900 }
1901
1902 pick_next_record:
1903 if (chunk > len)
1904 chunk = len;
1905
1906 decrypted += chunk;
1907 len -= chunk;
1908
1909 /* For async or peek case, queue the current skb */
1910 if (async || is_peek || retain_skb) {
1911 skb_queue_tail(&ctx->rx_list, skb);
1912 skb = NULL;
1913 }
1914
1915 if (tls_sw_advance_skb(sk, skb, chunk)) {
1916 /* Return full control message to
1917 * userspace before trying to parse
1918 * another message type
1919 */
1920 msg->msg_flags |= MSG_EOR;
1921 if (control != TLS_RECORD_TYPE_DATA)
1922 goto recv_end;
1923 } else {
1924 break;
1925 }
1926 }
1927
1928 recv_end:
1929 if (num_async) {
1930 /* Wait for all previously submitted records to be decrypted */
1931 spin_lock_bh(&ctx->decrypt_compl_lock);
1932 ctx->async_notify = true;
1933 pending = atomic_read(&ctx->decrypt_pending);
1934 spin_unlock_bh(&ctx->decrypt_compl_lock);
1935 if (pending) {
1936 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1937 if (err) {
1938 /* one of async decrypt failed */
1939 tls_err_abort(sk, err);
1940 copied = 0;
1941 decrypted = 0;
1942 goto end;
1943 }
1944 } else {
1945 reinit_completion(&ctx->async_wait.completion);
1946 }
1947
1948 /* There can be no concurrent accesses, since we have no
1949 * pending decrypt operations
1950 */
1951 WRITE_ONCE(ctx->async_notify, false);
1952
1953 /* Drain records from the rx_list & copy if required */
1954 if (is_peek || is_kvec)
1955 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1956 decrypted, false, is_peek);
1957 else
1958 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1959 decrypted, true, is_peek);
1960 if (err < 0) {
1961 tls_err_abort(sk, err);
1962 copied = 0;
1963 goto end;
1964 }
1965 }
1966
1967 copied += decrypted;
1968
1969 end:
1970 release_sock(sk);
1971 if (psock)
1972 sk_psock_put(sk, psock);
1973 return copied ? : err;
1974 }
1975
1976 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1977 struct pipe_inode_info *pipe,
1978 size_t len, unsigned int flags)
1979 {
1980 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1981 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1982 struct strp_msg *rxm = NULL;
1983 struct sock *sk = sock->sk;
1984 struct sk_buff *skb;
1985 ssize_t copied = 0;
1986 int err = 0;
1987 long timeo;
1988 int chunk;
1989 bool zc = false;
1990
1991 lock_sock(sk);
1992
1993 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1994
1995 skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1996 if (!skb)
1997 goto splice_read_end;
1998
1999 if (!ctx->decrypted) {
2000 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
2001
2002 /* splice does not support reading control messages */
2003 if (ctx->control != TLS_RECORD_TYPE_DATA) {
2004 err = -EINVAL;
2005 goto splice_read_end;
2006 }
2007
2008 if (err < 0) {
2009 tls_err_abort(sk, EBADMSG);
2010 goto splice_read_end;
2011 }
2012 ctx->decrypted = 1;
2013 }
2014 rxm = strp_msg(skb);
2015
2016 chunk = min_t(unsigned int, rxm->full_len, len);
2017 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2018 if (copied < 0)
2019 goto splice_read_end;
2020
2021 if (likely(!(flags & MSG_PEEK)))
2022 tls_sw_advance_skb(sk, skb, copied);
2023
2024 splice_read_end:
2025 release_sock(sk);
2026 return copied ? : err;
2027 }
2028
2029 bool tls_sw_stream_read(const struct sock *sk)
2030 {
2031 struct tls_context *tls_ctx = tls_get_ctx(sk);
2032 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2033 bool ingress_empty = true;
2034 struct sk_psock *psock;
2035
2036 rcu_read_lock();
2037 psock = sk_psock(sk);
2038 if (psock)
2039 ingress_empty = list_empty(&psock->ingress_msg);
2040 rcu_read_unlock();
2041
2042 return !ingress_empty || ctx->recv_pkt ||
2043 !skb_queue_empty(&ctx->rx_list);
2044 }
2045
2046 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2047 {
2048 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2049 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2050 struct tls_prot_info *prot = &tls_ctx->prot_info;
2051 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2052 struct strp_msg *rxm = strp_msg(skb);
2053 size_t cipher_overhead;
2054 size_t data_len = 0;
2055 int ret;
2056
2057 /* Verify that we have a full TLS header, or wait for more data */
2058 if (rxm->offset + prot->prepend_size > skb->len)
2059 return 0;
2060
2061 /* Sanity-check size of on-stack buffer. */
2062 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2063 ret = -EINVAL;
2064 goto read_failure;
2065 }
2066
2067 /* Linearize header to local buffer */
2068 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2069
2070 if (ret < 0)
2071 goto read_failure;
2072
2073 ctx->control = header[0];
2074
2075 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2076
2077 cipher_overhead = prot->tag_size;
2078 if (prot->version != TLS_1_3_VERSION &&
2079 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2080 cipher_overhead += prot->iv_size;
2081
2082 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2083 prot->tail_size) {
2084 ret = -EMSGSIZE;
2085 goto read_failure;
2086 }
2087 if (data_len < cipher_overhead) {
2088 ret = -EBADMSG;
2089 goto read_failure;
2090 }
2091
2092 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2093 if (header[1] != TLS_1_2_VERSION_MINOR ||
2094 header[2] != TLS_1_2_VERSION_MAJOR) {
2095 ret = -EINVAL;
2096 goto read_failure;
2097 }
2098
2099 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2100 TCP_SKB_CB(skb)->seq + rxm->offset);
2101 return data_len + TLS_HEADER_SIZE;
2102
2103 read_failure:
2104 tls_err_abort(strp->sk, ret);
2105
2106 return ret;
2107 }
2108
2109 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2110 {
2111 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2112 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2113
2114 ctx->decrypted = 0;
2115
2116 ctx->recv_pkt = skb;
2117 strp_pause(strp);
2118
2119 ctx->saved_data_ready(strp->sk);
2120 }
2121
2122 static void tls_data_ready(struct sock *sk)
2123 {
2124 struct tls_context *tls_ctx = tls_get_ctx(sk);
2125 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2126 struct sk_psock *psock;
2127
2128 strp_data_ready(&ctx->strp);
2129
2130 psock = sk_psock_get(sk);
2131 if (psock) {
2132 if (!list_empty(&psock->ingress_msg))
2133 ctx->saved_data_ready(sk);
2134 sk_psock_put(sk, psock);
2135 }
2136 }
2137
2138 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2139 {
2140 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2141
2142 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2143 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2144 cancel_delayed_work_sync(&ctx->tx_work.work);
2145 }
2146
2147 void tls_sw_release_resources_tx(struct sock *sk)
2148 {
2149 struct tls_context *tls_ctx = tls_get_ctx(sk);
2150 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2151 struct tls_rec *rec, *tmp;
2152 int pending;
2153
2154 /* Wait for any pending async encryptions to complete */
2155 spin_lock_bh(&ctx->encrypt_compl_lock);
2156 ctx->async_notify = true;
2157 pending = atomic_read(&ctx->encrypt_pending);
2158 spin_unlock_bh(&ctx->encrypt_compl_lock);
2159
2160 if (pending)
2161 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2162
2163 tls_tx_records(sk, -1);
2164
2165 /* Free up un-sent records in tx_list. First, free
2166 * the partially sent record if any at head of tx_list.
2167 */
2168 if (tls_ctx->partially_sent_record) {
2169 tls_free_partial_record(sk, tls_ctx);
2170 rec = list_first_entry(&ctx->tx_list,
2171 struct tls_rec, list);
2172 list_del(&rec->list);
2173 sk_msg_free(sk, &rec->msg_plaintext);
2174 kfree(rec);
2175 }
2176
2177 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2178 list_del(&rec->list);
2179 sk_msg_free(sk, &rec->msg_encrypted);
2180 sk_msg_free(sk, &rec->msg_plaintext);
2181 kfree(rec);
2182 }
2183
2184 crypto_free_aead(ctx->aead_send);
2185 tls_free_open_rec(sk);
2186 }
2187
2188 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2189 {
2190 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2191
2192 kfree(ctx);
2193 }
2194
2195 void tls_sw_release_resources_rx(struct sock *sk)
2196 {
2197 struct tls_context *tls_ctx = tls_get_ctx(sk);
2198 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2199
2200 kfree(tls_ctx->rx.rec_seq);
2201 kfree(tls_ctx->rx.iv);
2202
2203 if (ctx->aead_recv) {
2204 kfree_skb(ctx->recv_pkt);
2205 ctx->recv_pkt = NULL;
2206 skb_queue_purge(&ctx->rx_list);
2207 crypto_free_aead(ctx->aead_recv);
2208 strp_stop(&ctx->strp);
2209 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2210 * we still want to strp_stop(), but sk->sk_data_ready was
2211 * never swapped.
2212 */
2213 if (ctx->saved_data_ready) {
2214 write_lock_bh(&sk->sk_callback_lock);
2215 sk->sk_data_ready = ctx->saved_data_ready;
2216 write_unlock_bh(&sk->sk_callback_lock);
2217 }
2218 }
2219 }
2220
2221 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2222 {
2223 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2224
2225 strp_done(&ctx->strp);
2226 }
2227
2228 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2229 {
2230 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2231
2232 kfree(ctx);
2233 }
2234
2235 void tls_sw_free_resources_rx(struct sock *sk)
2236 {
2237 struct tls_context *tls_ctx = tls_get_ctx(sk);
2238
2239 tls_sw_release_resources_rx(sk);
2240 tls_sw_free_ctx_rx(tls_ctx);
2241 }
2242
2243 /* The work handler to transmitt the encrypted records in tx_list */
2244 static void tx_work_handler(struct work_struct *work)
2245 {
2246 struct delayed_work *delayed_work = to_delayed_work(work);
2247 struct tx_work *tx_work = container_of(delayed_work,
2248 struct tx_work, work);
2249 struct sock *sk = tx_work->sk;
2250 struct tls_context *tls_ctx = tls_get_ctx(sk);
2251 struct tls_sw_context_tx *ctx;
2252
2253 if (unlikely(!tls_ctx))
2254 return;
2255
2256 ctx = tls_sw_ctx_tx(tls_ctx);
2257 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2258 return;
2259
2260 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2261 return;
2262 mutex_lock(&tls_ctx->tx_lock);
2263 lock_sock(sk);
2264 tls_tx_records(sk, -1);
2265 release_sock(sk);
2266 mutex_unlock(&tls_ctx->tx_lock);
2267 }
2268
2269 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2270 {
2271 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2272
2273 /* Schedule the transmission if tx list is ready */
2274 if (is_tx_ready(tx_ctx) &&
2275 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2276 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2277 }
2278
2279 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2280 {
2281 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2282
2283 write_lock_bh(&sk->sk_callback_lock);
2284 rx_ctx->saved_data_ready = sk->sk_data_ready;
2285 sk->sk_data_ready = tls_data_ready;
2286 write_unlock_bh(&sk->sk_callback_lock);
2287
2288 strp_check_rcv(&rx_ctx->strp);
2289 }
2290
2291 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2292 {
2293 struct tls_context *tls_ctx = tls_get_ctx(sk);
2294 struct tls_prot_info *prot = &tls_ctx->prot_info;
2295 struct tls_crypto_info *crypto_info;
2296 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2297 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2298 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2299 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2300 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2301 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2302 struct cipher_context *cctx;
2303 struct crypto_aead **aead;
2304 struct strp_callbacks cb;
2305 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2306 struct crypto_tfm *tfm;
2307 char *iv, *rec_seq, *key, *salt, *cipher_name;
2308 size_t keysize;
2309 int rc = 0;
2310
2311 if (!ctx) {
2312 rc = -EINVAL;
2313 goto out;
2314 }
2315
2316 if (tx) {
2317 if (!ctx->priv_ctx_tx) {
2318 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2319 if (!sw_ctx_tx) {
2320 rc = -ENOMEM;
2321 goto out;
2322 }
2323 ctx->priv_ctx_tx = sw_ctx_tx;
2324 } else {
2325 sw_ctx_tx =
2326 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2327 }
2328 } else {
2329 if (!ctx->priv_ctx_rx) {
2330 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2331 if (!sw_ctx_rx) {
2332 rc = -ENOMEM;
2333 goto out;
2334 }
2335 ctx->priv_ctx_rx = sw_ctx_rx;
2336 } else {
2337 sw_ctx_rx =
2338 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2339 }
2340 }
2341
2342 if (tx) {
2343 crypto_init_wait(&sw_ctx_tx->async_wait);
2344 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2345 crypto_info = &ctx->crypto_send.info;
2346 cctx = &ctx->tx;
2347 aead = &sw_ctx_tx->aead_send;
2348 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2349 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2350 sw_ctx_tx->tx_work.sk = sk;
2351 } else {
2352 crypto_init_wait(&sw_ctx_rx->async_wait);
2353 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2354 crypto_info = &ctx->crypto_recv.info;
2355 cctx = &ctx->rx;
2356 skb_queue_head_init(&sw_ctx_rx->rx_list);
2357 aead = &sw_ctx_rx->aead_recv;
2358 }
2359
2360 switch (crypto_info->cipher_type) {
2361 case TLS_CIPHER_AES_GCM_128: {
2362 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2363 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2364 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2365 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2366 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2367 rec_seq =
2368 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2369 gcm_128_info =
2370 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2371 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2372 key = gcm_128_info->key;
2373 salt = gcm_128_info->salt;
2374 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2375 cipher_name = "gcm(aes)";
2376 break;
2377 }
2378 case TLS_CIPHER_AES_GCM_256: {
2379 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2380 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2381 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2382 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2383 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2384 rec_seq =
2385 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2386 gcm_256_info =
2387 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2388 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2389 key = gcm_256_info->key;
2390 salt = gcm_256_info->salt;
2391 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2392 cipher_name = "gcm(aes)";
2393 break;
2394 }
2395 case TLS_CIPHER_AES_CCM_128: {
2396 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2397 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2398 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2399 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2400 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2401 rec_seq =
2402 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2403 ccm_128_info =
2404 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2405 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2406 key = ccm_128_info->key;
2407 salt = ccm_128_info->salt;
2408 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2409 cipher_name = "ccm(aes)";
2410 break;
2411 }
2412 case TLS_CIPHER_CHACHA20_POLY1305: {
2413 chacha20_poly1305_info = (void *)crypto_info;
2414 nonce_size = 0;
2415 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2416 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2417 iv = chacha20_poly1305_info->iv;
2418 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2419 rec_seq = chacha20_poly1305_info->rec_seq;
2420 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2421 key = chacha20_poly1305_info->key;
2422 salt = chacha20_poly1305_info->salt;
2423 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2424 cipher_name = "rfc7539(chacha20,poly1305)";
2425 break;
2426 }
2427 default:
2428 rc = -EINVAL;
2429 goto free_priv;
2430 }
2431
2432 /* Sanity-check the sizes for stack allocations. */
2433 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2434 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2435 rc = -EINVAL;
2436 goto free_priv;
2437 }
2438
2439 if (crypto_info->version == TLS_1_3_VERSION) {
2440 nonce_size = 0;
2441 prot->aad_size = TLS_HEADER_SIZE;
2442 prot->tail_size = 1;
2443 } else {
2444 prot->aad_size = TLS_AAD_SPACE_SIZE;
2445 prot->tail_size = 0;
2446 }
2447
2448 prot->version = crypto_info->version;
2449 prot->cipher_type = crypto_info->cipher_type;
2450 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2451 prot->tag_size = tag_size;
2452 prot->overhead_size = prot->prepend_size +
2453 prot->tag_size + prot->tail_size;
2454 prot->iv_size = iv_size;
2455 prot->salt_size = salt_size;
2456 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2457 if (!cctx->iv) {
2458 rc = -ENOMEM;
2459 goto free_priv;
2460 }
2461 /* Note: 128 & 256 bit salt are the same size */
2462 prot->rec_seq_size = rec_seq_size;
2463 memcpy(cctx->iv, salt, salt_size);
2464 memcpy(cctx->iv + salt_size, iv, iv_size);
2465 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2466 if (!cctx->rec_seq) {
2467 rc = -ENOMEM;
2468 goto free_iv;
2469 }
2470
2471 if (!*aead) {
2472 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2473 if (IS_ERR(*aead)) {
2474 rc = PTR_ERR(*aead);
2475 *aead = NULL;
2476 goto free_rec_seq;
2477 }
2478 }
2479
2480 ctx->push_pending_record = tls_sw_push_pending_record;
2481
2482 rc = crypto_aead_setkey(*aead, key, keysize);
2483
2484 if (rc)
2485 goto free_aead;
2486
2487 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2488 if (rc)
2489 goto free_aead;
2490
2491 if (sw_ctx_rx) {
2492 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2493
2494 if (crypto_info->version == TLS_1_3_VERSION)
2495 sw_ctx_rx->async_capable = 0;
2496 else
2497 sw_ctx_rx->async_capable =
2498 !!(tfm->__crt_alg->cra_flags &
2499 CRYPTO_ALG_ASYNC);
2500
2501 /* Set up strparser */
2502 memset(&cb, 0, sizeof(cb));
2503 cb.rcv_msg = tls_queue;
2504 cb.parse_msg = tls_read_size;
2505
2506 strp_init(&sw_ctx_rx->strp, sk, &cb);
2507 }
2508
2509 goto out;
2510
2511 free_aead:
2512 crypto_free_aead(*aead);
2513 *aead = NULL;
2514 free_rec_seq:
2515 kfree(cctx->rec_seq);
2516 cctx->rec_seq = NULL;
2517 free_iv:
2518 kfree(cctx->iv);
2519 cctx->iv = NULL;
2520 free_priv:
2521 if (tx) {
2522 kfree(ctx->priv_ctx_tx);
2523 ctx->priv_ctx_tx = NULL;
2524 } else {
2525 kfree(ctx->priv_ctx_rx);
2526 ctx->priv_ctx_rx = NULL;
2527 }
2528 out:
2529 return rc;
2530 }
Linux with added FPC-III support