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
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:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
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.
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
38 #include <linux/sched/signal.h>
39 #include <linux/module.h>
40 #include <crypto/aead.h>
42 #include <net/strparser.h>
45 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
46 unsigned int recursion_level)
48 int start = skb_headlen(skb);
49 int i, chunk = start - offset;
50 struct sk_buff *frag_iter;
53 if (unlikely(recursion_level >= 24))
66 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
69 WARN_ON(start > offset + len);
71 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
85 if (unlikely(skb_has_frag_list(skb))) {
86 skb_walk_frags(skb, frag_iter) {
89 WARN_ON(start > offset + len);
91 end = start + frag_iter->len;
96 ret = __skb_nsg(frag_iter, offset - start, chunk,
98 if (unlikely(ret < 0))
113 /* Return the number of scatterlist elements required to completely map the
114 * skb, or -EMSGSIZE if the recursion depth is exceeded.
116 static int skb_nsg(struct sk_buff *skb, int offset, int len)
118 return __skb_nsg(skb, offset, len, 0);
121 static int padding_length(struct tls_sw_context_rx *ctx,
122 struct tls_prot_info *prot, struct sk_buff *skb)
124 struct strp_msg *rxm = strp_msg(skb);
127 /* Determine zero-padding length */
128 if (prot->version == TLS_1_3_VERSION) {
129 char content_type = 0;
133 while (content_type == 0) {
134 if (back > rxm->full_len - prot->prepend_size)
136 err = skb_copy_bits(skb,
137 rxm->offset + rxm->full_len - back,
146 ctx->control = content_type;
151 static void tls_decrypt_done(struct crypto_async_request *req, int err)
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;
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;
169 /* Propagate if there was an err */
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);
177 struct strp_msg *rxm = strp_msg(skb);
180 pad = padding_length(ctx, prot, skb);
182 ctx->async_wait.err = pad;
183 tls_err_abort(skb->sk, pad);
185 rxm->full_len -= pad;
186 rxm->offset += prot->prepend_size;
187 rxm->full_len -= prot->overhead_size;
191 /* After using skb->sk to propagate sk through crypto async callback
192 * we need to NULL it again.
197 /* Free the destination pages if skb was not decrypted inplace */
199 /* Skip the first S/G entry as it points to AAD */
200 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
203 put_page(sg_page(sg));
209 pending = atomic_dec_return(&ctx->decrypt_pending);
211 if (!pending && READ_ONCE(ctx->async_notify))
212 complete(&ctx->async_wait.completion);
215 static int tls_do_decryption(struct sock *sk,
217 struct scatterlist *sgin,
218 struct scatterlist *sgout,
221 struct aead_request *aead_req,
224 struct tls_context *tls_ctx = tls_get_ctx(sk);
225 struct tls_prot_info *prot = &tls_ctx->prot_info;
226 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
229 aead_request_set_tfm(aead_req, ctx->aead_recv);
230 aead_request_set_ad(aead_req, prot->aad_size);
231 aead_request_set_crypt(aead_req, sgin, sgout,
232 data_len + prot->tag_size,
236 /* Using skb->sk to push sk through to crypto async callback
237 * handler. This allows propagating errors up to the socket
238 * if needed. It _must_ be cleared in the async handler
239 * before consume_skb is called. We _know_ skb->sk is NULL
240 * because it is a clone from strparser.
243 aead_request_set_callback(aead_req,
244 CRYPTO_TFM_REQ_MAY_BACKLOG,
245 tls_decrypt_done, skb);
246 atomic_inc(&ctx->decrypt_pending);
248 aead_request_set_callback(aead_req,
249 CRYPTO_TFM_REQ_MAY_BACKLOG,
250 crypto_req_done, &ctx->async_wait);
253 ret = crypto_aead_decrypt(aead_req);
254 if (ret == -EINPROGRESS) {
258 ret = crypto_wait_req(ret, &ctx->async_wait);
262 atomic_dec(&ctx->decrypt_pending);
267 static void tls_trim_both_msgs(struct sock *sk, int target_size)
269 struct tls_context *tls_ctx = tls_get_ctx(sk);
270 struct tls_prot_info *prot = &tls_ctx->prot_info;
271 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
272 struct tls_rec *rec = ctx->open_rec;
274 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
276 target_size += prot->overhead_size;
277 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
280 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
282 struct tls_context *tls_ctx = tls_get_ctx(sk);
283 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
284 struct tls_rec *rec = ctx->open_rec;
285 struct sk_msg *msg_en = &rec->msg_encrypted;
287 return sk_msg_alloc(sk, msg_en, len, 0);
290 static int tls_clone_plaintext_msg(struct sock *sk, int required)
292 struct tls_context *tls_ctx = tls_get_ctx(sk);
293 struct tls_prot_info *prot = &tls_ctx->prot_info;
294 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
295 struct tls_rec *rec = ctx->open_rec;
296 struct sk_msg *msg_pl = &rec->msg_plaintext;
297 struct sk_msg *msg_en = &rec->msg_encrypted;
300 /* We add page references worth len bytes from encrypted sg
301 * at the end of plaintext sg. It is guaranteed that msg_en
302 * has enough required room (ensured by caller).
304 len = required - msg_pl->sg.size;
306 /* Skip initial bytes in msg_en's data to be able to use
307 * same offset of both plain and encrypted data.
309 skip = prot->prepend_size + msg_pl->sg.size;
311 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
314 static struct tls_rec *tls_get_rec(struct sock *sk)
316 struct tls_context *tls_ctx = tls_get_ctx(sk);
317 struct tls_prot_info *prot = &tls_ctx->prot_info;
318 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
319 struct sk_msg *msg_pl, *msg_en;
323 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
325 rec = kzalloc(mem_size, sk->sk_allocation);
329 msg_pl = &rec->msg_plaintext;
330 msg_en = &rec->msg_encrypted;
335 sg_init_table(rec->sg_aead_in, 2);
336 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
337 sg_unmark_end(&rec->sg_aead_in[1]);
339 sg_init_table(rec->sg_aead_out, 2);
340 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
341 sg_unmark_end(&rec->sg_aead_out[1]);
346 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
348 sk_msg_free(sk, &rec->msg_encrypted);
349 sk_msg_free(sk, &rec->msg_plaintext);
353 static void tls_free_open_rec(struct sock *sk)
355 struct tls_context *tls_ctx = tls_get_ctx(sk);
356 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
357 struct tls_rec *rec = ctx->open_rec;
360 tls_free_rec(sk, rec);
361 ctx->open_rec = NULL;
365 int tls_tx_records(struct sock *sk, int flags)
367 struct tls_context *tls_ctx = tls_get_ctx(sk);
368 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
369 struct tls_rec *rec, *tmp;
370 struct sk_msg *msg_en;
371 int tx_flags, rc = 0;
373 if (tls_is_partially_sent_record(tls_ctx)) {
374 rec = list_first_entry(&ctx->tx_list,
375 struct tls_rec, list);
378 tx_flags = rec->tx_flags;
382 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
386 /* Full record has been transmitted.
387 * Remove the head of tx_list
389 list_del(&rec->list);
390 sk_msg_free(sk, &rec->msg_plaintext);
394 /* Tx all ready records */
395 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
396 if (READ_ONCE(rec->tx_ready)) {
398 tx_flags = rec->tx_flags;
402 msg_en = &rec->msg_encrypted;
403 rc = tls_push_sg(sk, tls_ctx,
404 &msg_en->sg.data[msg_en->sg.curr],
409 list_del(&rec->list);
410 sk_msg_free(sk, &rec->msg_plaintext);
418 if (rc < 0 && rc != -EAGAIN)
419 tls_err_abort(sk, EBADMSG);
424 static void tls_encrypt_done(struct crypto_async_request *req, int err)
426 struct aead_request *aead_req = (struct aead_request *)req;
427 struct sock *sk = req->data;
428 struct tls_context *tls_ctx = tls_get_ctx(sk);
429 struct tls_prot_info *prot = &tls_ctx->prot_info;
430 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
431 struct scatterlist *sge;
432 struct sk_msg *msg_en;
437 rec = container_of(aead_req, struct tls_rec, aead_req);
438 msg_en = &rec->msg_encrypted;
440 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
441 sge->offset -= prot->prepend_size;
442 sge->length += prot->prepend_size;
444 /* Check if error is previously set on socket */
445 if (err || sk->sk_err) {
448 /* If err is already set on socket, return the same code */
450 ctx->async_wait.err = sk->sk_err;
452 ctx->async_wait.err = err;
453 tls_err_abort(sk, err);
458 struct tls_rec *first_rec;
460 /* Mark the record as ready for transmission */
461 smp_store_mb(rec->tx_ready, true);
463 /* If received record is at head of tx_list, schedule tx */
464 first_rec = list_first_entry(&ctx->tx_list,
465 struct tls_rec, list);
466 if (rec == first_rec)
470 pending = atomic_dec_return(&ctx->encrypt_pending);
472 if (!pending && READ_ONCE(ctx->async_notify))
473 complete(&ctx->async_wait.completion);
478 /* Schedule the transmission */
479 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
480 schedule_delayed_work(&ctx->tx_work.work, 1);
483 static int tls_do_encryption(struct sock *sk,
484 struct tls_context *tls_ctx,
485 struct tls_sw_context_tx *ctx,
486 struct aead_request *aead_req,
487 size_t data_len, u32 start)
489 struct tls_prot_info *prot = &tls_ctx->prot_info;
490 struct tls_rec *rec = ctx->open_rec;
491 struct sk_msg *msg_en = &rec->msg_encrypted;
492 struct scatterlist *sge = sk_msg_elem(msg_en, start);
493 int rc, iv_offset = 0;
495 /* For CCM based ciphers, first byte of IV is a constant */
496 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
497 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
501 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
502 prot->iv_size + prot->salt_size);
504 xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq);
506 sge->offset += prot->prepend_size;
507 sge->length -= prot->prepend_size;
509 msg_en->sg.curr = start;
511 aead_request_set_tfm(aead_req, ctx->aead_send);
512 aead_request_set_ad(aead_req, prot->aad_size);
513 aead_request_set_crypt(aead_req, rec->sg_aead_in,
515 data_len, rec->iv_data);
517 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
518 tls_encrypt_done, sk);
520 /* Add the record in tx_list */
521 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
522 atomic_inc(&ctx->encrypt_pending);
524 rc = crypto_aead_encrypt(aead_req);
525 if (!rc || rc != -EINPROGRESS) {
526 atomic_dec(&ctx->encrypt_pending);
527 sge->offset -= prot->prepend_size;
528 sge->length += prot->prepend_size;
532 WRITE_ONCE(rec->tx_ready, true);
533 } else if (rc != -EINPROGRESS) {
534 list_del(&rec->list);
538 /* Unhook the record from context if encryption is not failure */
539 ctx->open_rec = NULL;
540 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
544 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
545 struct tls_rec **to, struct sk_msg *msg_opl,
546 struct sk_msg *msg_oen, u32 split_point,
547 u32 tx_overhead_size, u32 *orig_end)
549 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
550 struct scatterlist *sge, *osge, *nsge;
551 u32 orig_size = msg_opl->sg.size;
552 struct scatterlist tmp = { };
553 struct sk_msg *msg_npl;
557 new = tls_get_rec(sk);
560 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
561 tx_overhead_size, 0);
563 tls_free_rec(sk, new);
567 *orig_end = msg_opl->sg.end;
568 i = msg_opl->sg.start;
569 sge = sk_msg_elem(msg_opl, i);
570 while (apply && sge->length) {
571 if (sge->length > apply) {
572 u32 len = sge->length - apply;
574 get_page(sg_page(sge));
575 sg_set_page(&tmp, sg_page(sge), len,
576 sge->offset + apply);
581 apply -= sge->length;
582 bytes += sge->length;
585 sk_msg_iter_var_next(i);
586 if (i == msg_opl->sg.end)
588 sge = sk_msg_elem(msg_opl, i);
592 msg_opl->sg.curr = i;
593 msg_opl->sg.copybreak = 0;
594 msg_opl->apply_bytes = 0;
595 msg_opl->sg.size = bytes;
597 msg_npl = &new->msg_plaintext;
598 msg_npl->apply_bytes = apply;
599 msg_npl->sg.size = orig_size - bytes;
601 j = msg_npl->sg.start;
602 nsge = sk_msg_elem(msg_npl, j);
604 memcpy(nsge, &tmp, sizeof(*nsge));
605 sk_msg_iter_var_next(j);
606 nsge = sk_msg_elem(msg_npl, j);
609 osge = sk_msg_elem(msg_opl, i);
610 while (osge->length) {
611 memcpy(nsge, osge, sizeof(*nsge));
613 sk_msg_iter_var_next(i);
614 sk_msg_iter_var_next(j);
617 osge = sk_msg_elem(msg_opl, i);
618 nsge = sk_msg_elem(msg_npl, j);
622 msg_npl->sg.curr = j;
623 msg_npl->sg.copybreak = 0;
629 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
630 struct tls_rec *from, u32 orig_end)
632 struct sk_msg *msg_npl = &from->msg_plaintext;
633 struct sk_msg *msg_opl = &to->msg_plaintext;
634 struct scatterlist *osge, *nsge;
638 sk_msg_iter_var_prev(i);
639 j = msg_npl->sg.start;
641 osge = sk_msg_elem(msg_opl, i);
642 nsge = sk_msg_elem(msg_npl, j);
644 if (sg_page(osge) == sg_page(nsge) &&
645 osge->offset + osge->length == nsge->offset) {
646 osge->length += nsge->length;
647 put_page(sg_page(nsge));
650 msg_opl->sg.end = orig_end;
651 msg_opl->sg.curr = orig_end;
652 msg_opl->sg.copybreak = 0;
653 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
654 msg_opl->sg.size += msg_npl->sg.size;
656 sk_msg_free(sk, &to->msg_encrypted);
657 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
662 static int tls_push_record(struct sock *sk, int flags,
663 unsigned char record_type)
665 struct tls_context *tls_ctx = tls_get_ctx(sk);
666 struct tls_prot_info *prot = &tls_ctx->prot_info;
667 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
668 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
669 u32 i, split_point, uninitialized_var(orig_end);
670 struct sk_msg *msg_pl, *msg_en;
671 struct aead_request *req;
678 msg_pl = &rec->msg_plaintext;
679 msg_en = &rec->msg_encrypted;
681 split_point = msg_pl->apply_bytes;
682 split = split_point && split_point < msg_pl->sg.size;
683 if (unlikely((!split &&
685 prot->overhead_size > msg_en->sg.size) ||
688 prot->overhead_size > msg_en->sg.size))) {
690 split_point = msg_en->sg.size;
693 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
694 split_point, prot->overhead_size,
698 /* This can happen if above tls_split_open_record allocates
699 * a single large encryption buffer instead of two smaller
700 * ones. In this case adjust pointers and continue without
703 if (!msg_pl->sg.size) {
704 tls_merge_open_record(sk, rec, tmp, orig_end);
705 msg_pl = &rec->msg_plaintext;
706 msg_en = &rec->msg_encrypted;
709 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
710 prot->overhead_size);
713 rec->tx_flags = flags;
714 req = &rec->aead_req;
717 sk_msg_iter_var_prev(i);
719 rec->content_type = record_type;
720 if (prot->version == TLS_1_3_VERSION) {
721 /* Add content type to end of message. No padding added */
722 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
723 sg_mark_end(&rec->sg_content_type);
724 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
725 &rec->sg_content_type);
727 sg_mark_end(sk_msg_elem(msg_pl, i));
730 if (msg_pl->sg.end < msg_pl->sg.start) {
731 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
732 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
736 i = msg_pl->sg.start;
737 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
740 sk_msg_iter_var_prev(i);
741 sg_mark_end(sk_msg_elem(msg_en, i));
743 i = msg_en->sg.start;
744 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
746 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
747 tls_ctx->tx.rec_seq, prot->rec_seq_size,
748 record_type, prot->version);
750 tls_fill_prepend(tls_ctx,
751 page_address(sg_page(&msg_en->sg.data[i])) +
752 msg_en->sg.data[i].offset,
753 msg_pl->sg.size + prot->tail_size,
754 record_type, prot->version);
756 tls_ctx->pending_open_record_frags = false;
758 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
759 msg_pl->sg.size + prot->tail_size, i);
761 if (rc != -EINPROGRESS) {
762 tls_err_abort(sk, EBADMSG);
764 tls_ctx->pending_open_record_frags = true;
765 tls_merge_open_record(sk, rec, tmp, orig_end);
768 ctx->async_capable = 1;
771 msg_pl = &tmp->msg_plaintext;
772 msg_en = &tmp->msg_encrypted;
773 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
774 tls_ctx->pending_open_record_frags = true;
778 return tls_tx_records(sk, flags);
781 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
782 bool full_record, u8 record_type,
783 ssize_t *copied, int flags)
785 struct tls_context *tls_ctx = tls_get_ctx(sk);
786 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
787 struct sk_msg msg_redir = { };
788 struct sk_psock *psock;
789 struct sock *sk_redir;
795 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
796 psock = sk_psock_get(sk);
797 if (!psock || !policy) {
798 err = tls_push_record(sk, flags, record_type);
799 if (err && sk->sk_err == EBADMSG) {
800 *copied -= sk_msg_free(sk, msg);
801 tls_free_open_rec(sk);
805 sk_psock_put(sk, psock);
809 enospc = sk_msg_full(msg);
810 if (psock->eval == __SK_NONE) {
811 delta = msg->sg.size;
812 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
813 delta -= msg->sg.size;
815 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
816 !enospc && !full_record) {
822 if (msg->apply_bytes && msg->apply_bytes < send)
823 send = msg->apply_bytes;
825 switch (psock->eval) {
827 err = tls_push_record(sk, flags, record_type);
828 if (err && sk->sk_err == EBADMSG) {
829 *copied -= sk_msg_free(sk, msg);
830 tls_free_open_rec(sk);
836 sk_redir = psock->sk_redir;
837 memcpy(&msg_redir, msg, sizeof(*msg));
838 if (msg->apply_bytes < send)
839 msg->apply_bytes = 0;
841 msg->apply_bytes -= send;
842 sk_msg_return_zero(sk, msg, send);
843 msg->sg.size -= send;
845 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
848 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
851 if (msg->sg.size == 0)
852 tls_free_open_rec(sk);
856 sk_msg_free_partial(sk, msg, send);
857 if (msg->apply_bytes < send)
858 msg->apply_bytes = 0;
860 msg->apply_bytes -= send;
861 if (msg->sg.size == 0)
862 tls_free_open_rec(sk);
863 *copied -= (send + delta);
868 bool reset_eval = !ctx->open_rec;
872 msg = &rec->msg_plaintext;
873 if (!msg->apply_bytes)
877 psock->eval = __SK_NONE;
878 if (psock->sk_redir) {
879 sock_put(psock->sk_redir);
880 psock->sk_redir = NULL;
887 sk_psock_put(sk, psock);
891 static int tls_sw_push_pending_record(struct sock *sk, int flags)
893 struct tls_context *tls_ctx = tls_get_ctx(sk);
894 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
895 struct tls_rec *rec = ctx->open_rec;
896 struct sk_msg *msg_pl;
902 msg_pl = &rec->msg_plaintext;
903 copied = msg_pl->sg.size;
907 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
911 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
913 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
914 struct tls_context *tls_ctx = tls_get_ctx(sk);
915 struct tls_prot_info *prot = &tls_ctx->prot_info;
916 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
917 bool async_capable = ctx->async_capable;
918 unsigned char record_type = TLS_RECORD_TYPE_DATA;
919 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
920 bool eor = !(msg->msg_flags & MSG_MORE);
923 struct sk_msg *msg_pl, *msg_en;
933 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
936 mutex_lock(&tls_ctx->tx_lock);
939 if (unlikely(msg->msg_controllen)) {
940 ret = tls_proccess_cmsg(sk, msg, &record_type);
942 if (ret == -EINPROGRESS)
944 else if (ret != -EAGAIN)
949 while (msg_data_left(msg)) {
958 rec = ctx->open_rec = tls_get_rec(sk);
964 msg_pl = &rec->msg_plaintext;
965 msg_en = &rec->msg_encrypted;
967 orig_size = msg_pl->sg.size;
969 try_to_copy = msg_data_left(msg);
970 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
971 if (try_to_copy >= record_room) {
972 try_to_copy = record_room;
976 required_size = msg_pl->sg.size + try_to_copy +
979 if (!sk_stream_memory_free(sk))
980 goto wait_for_sndbuf;
983 ret = tls_alloc_encrypted_msg(sk, required_size);
986 goto wait_for_memory;
988 /* Adjust try_to_copy according to the amount that was
989 * actually allocated. The difference is due
990 * to max sg elements limit
992 try_to_copy -= required_size - msg_en->sg.size;
996 if (!is_kvec && (full_record || eor) && !async_capable) {
997 u32 first = msg_pl->sg.end;
999 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1000 msg_pl, try_to_copy);
1002 goto fallback_to_reg_send;
1005 copied += try_to_copy;
1007 sk_msg_sg_copy_set(msg_pl, first);
1008 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1009 record_type, &copied,
1012 if (ret == -EINPROGRESS)
1014 else if (ret == -ENOMEM)
1015 goto wait_for_memory;
1016 else if (ctx->open_rec && ret == -ENOSPC)
1018 else if (ret != -EAGAIN)
1023 copied -= try_to_copy;
1024 sk_msg_sg_copy_clear(msg_pl, first);
1025 iov_iter_revert(&msg->msg_iter,
1026 msg_pl->sg.size - orig_size);
1027 fallback_to_reg_send:
1028 sk_msg_trim(sk, msg_pl, orig_size);
1031 required_size = msg_pl->sg.size + try_to_copy;
1033 ret = tls_clone_plaintext_msg(sk, required_size);
1038 /* Adjust try_to_copy according to the amount that was
1039 * actually allocated. The difference is due
1040 * to max sg elements limit
1042 try_to_copy -= required_size - msg_pl->sg.size;
1044 sk_msg_trim(sk, msg_en,
1045 msg_pl->sg.size + prot->overhead_size);
1049 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1050 msg_pl, try_to_copy);
1055 /* Open records defined only if successfully copied, otherwise
1056 * we would trim the sg but not reset the open record frags.
1058 tls_ctx->pending_open_record_frags = true;
1059 copied += try_to_copy;
1060 if (full_record || eor) {
1061 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1062 record_type, &copied,
1065 if (ret == -EINPROGRESS)
1067 else if (ret == -ENOMEM)
1068 goto wait_for_memory;
1069 else if (ret != -EAGAIN) {
1080 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1082 ret = sk_stream_wait_memory(sk, &timeo);
1086 tls_trim_both_msgs(sk, orig_size);
1090 if (ctx->open_rec && msg_en->sg.size < required_size)
1091 goto alloc_encrypted;
1096 } else if (num_zc) {
1097 /* Wait for pending encryptions to get completed */
1098 smp_store_mb(ctx->async_notify, true);
1100 if (atomic_read(&ctx->encrypt_pending))
1101 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1103 reinit_completion(&ctx->async_wait.completion);
1105 WRITE_ONCE(ctx->async_notify, false);
1107 if (ctx->async_wait.err) {
1108 ret = ctx->async_wait.err;
1113 /* Transmit if any encryptions have completed */
1114 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1115 cancel_delayed_work(&ctx->tx_work.work);
1116 tls_tx_records(sk, msg->msg_flags);
1120 ret = sk_stream_error(sk, msg->msg_flags, ret);
1123 mutex_unlock(&tls_ctx->tx_lock);
1124 return copied > 0 ? copied : ret;
1127 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1128 int offset, size_t size, int flags)
1130 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1131 struct tls_context *tls_ctx = tls_get_ctx(sk);
1132 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1133 struct tls_prot_info *prot = &tls_ctx->prot_info;
1134 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1135 struct sk_msg *msg_pl;
1136 struct tls_rec *rec;
1144 eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1145 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1147 /* Call the sk_stream functions to manage the sndbuf mem. */
1149 size_t copy, required_size;
1157 rec = ctx->open_rec;
1159 rec = ctx->open_rec = tls_get_rec(sk);
1165 msg_pl = &rec->msg_plaintext;
1167 full_record = false;
1168 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1170 if (copy >= record_room) {
1175 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1177 if (!sk_stream_memory_free(sk))
1178 goto wait_for_sndbuf;
1180 ret = tls_alloc_encrypted_msg(sk, required_size);
1183 goto wait_for_memory;
1185 /* Adjust copy according to the amount that was
1186 * actually allocated. The difference is due
1187 * to max sg elements limit
1189 copy -= required_size - msg_pl->sg.size;
1193 sk_msg_page_add(msg_pl, page, copy, offset);
1194 sk_mem_charge(sk, copy);
1200 tls_ctx->pending_open_record_frags = true;
1201 if (full_record || eor || sk_msg_full(msg_pl)) {
1202 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1203 record_type, &copied, flags);
1205 if (ret == -EINPROGRESS)
1207 else if (ret == -ENOMEM)
1208 goto wait_for_memory;
1209 else if (ret != -EAGAIN) {
1218 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1220 ret = sk_stream_wait_memory(sk, &timeo);
1223 tls_trim_both_msgs(sk, msg_pl->sg.size);
1232 /* Transmit if any encryptions have completed */
1233 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1234 cancel_delayed_work(&ctx->tx_work.work);
1235 tls_tx_records(sk, flags);
1239 ret = sk_stream_error(sk, flags, ret);
1240 return copied > 0 ? copied : ret;
1243 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1244 int offset, size_t size, int flags)
1246 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1247 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1248 MSG_NO_SHARED_FRAGS))
1251 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1254 int tls_sw_sendpage(struct sock *sk, struct page *page,
1255 int offset, size_t size, int flags)
1257 struct tls_context *tls_ctx = tls_get_ctx(sk);
1260 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1261 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1264 mutex_lock(&tls_ctx->tx_lock);
1266 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1268 mutex_unlock(&tls_ctx->tx_lock);
1272 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1273 int flags, long timeo, int *err)
1275 struct tls_context *tls_ctx = tls_get_ctx(sk);
1276 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1277 struct sk_buff *skb;
1278 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1280 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1282 *err = sock_error(sk);
1286 if (sk->sk_shutdown & RCV_SHUTDOWN)
1289 if (sock_flag(sk, SOCK_DONE))
1292 if ((flags & MSG_DONTWAIT) || !timeo) {
1297 add_wait_queue(sk_sleep(sk), &wait);
1298 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1299 sk_wait_event(sk, &timeo,
1300 ctx->recv_pkt != skb ||
1301 !sk_psock_queue_empty(psock),
1303 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1304 remove_wait_queue(sk_sleep(sk), &wait);
1306 /* Handle signals */
1307 if (signal_pending(current)) {
1308 *err = sock_intr_errno(timeo);
1316 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1317 int length, int *pages_used,
1318 unsigned int *size_used,
1319 struct scatterlist *to,
1322 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1323 struct page *pages[MAX_SKB_FRAGS];
1324 unsigned int size = *size_used;
1325 ssize_t copied, use;
1328 while (length > 0) {
1330 maxpages = to_max_pages - num_elem;
1331 if (maxpages == 0) {
1335 copied = iov_iter_get_pages(from, pages,
1343 iov_iter_advance(from, copied);
1348 use = min_t(int, copied, PAGE_SIZE - offset);
1350 sg_set_page(&to[num_elem],
1351 pages[i], use, offset);
1352 sg_unmark_end(&to[num_elem]);
1353 /* We do not uncharge memory from this API */
1362 /* Mark the end in the last sg entry if newly added */
1363 if (num_elem > *pages_used)
1364 sg_mark_end(&to[num_elem - 1]);
1367 iov_iter_revert(from, size - *size_used);
1369 *pages_used = num_elem;
1374 /* This function decrypts the input skb into either out_iov or in out_sg
1375 * or in skb buffers itself. The input parameter 'zc' indicates if
1376 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1377 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1378 * NULL, then the decryption happens inside skb buffers itself, i.e.
1379 * zero-copy gets disabled and 'zc' is updated.
1382 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1383 struct iov_iter *out_iov,
1384 struct scatterlist *out_sg,
1385 int *chunk, bool *zc, bool async)
1387 struct tls_context *tls_ctx = tls_get_ctx(sk);
1388 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1389 struct tls_prot_info *prot = &tls_ctx->prot_info;
1390 struct strp_msg *rxm = strp_msg(skb);
1391 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1392 struct aead_request *aead_req;
1393 struct sk_buff *unused;
1394 u8 *aad, *iv, *mem = NULL;
1395 struct scatterlist *sgin = NULL;
1396 struct scatterlist *sgout = NULL;
1397 const int data_len = rxm->full_len - prot->overhead_size +
1401 if (*zc && (out_iov || out_sg)) {
1403 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1405 n_sgout = sg_nents(out_sg);
1406 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1407 rxm->full_len - prot->prepend_size);
1411 n_sgin = skb_cow_data(skb, 0, &unused);
1417 /* Increment to accommodate AAD */
1418 n_sgin = n_sgin + 1;
1420 nsg = n_sgin + n_sgout;
1422 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1423 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1424 mem_size = mem_size + prot->aad_size;
1425 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1427 /* Allocate a single block of memory which contains
1428 * aead_req || sgin[] || sgout[] || aad || iv.
1429 * This order achieves correct alignment for aead_req, sgin, sgout.
1431 mem = kmalloc(mem_size, sk->sk_allocation);
1435 /* Segment the allocated memory */
1436 aead_req = (struct aead_request *)mem;
1437 sgin = (struct scatterlist *)(mem + aead_size);
1438 sgout = sgin + n_sgin;
1439 aad = (u8 *)(sgout + n_sgout);
1440 iv = aad + prot->aad_size;
1442 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1443 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1449 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1450 iv + iv_offset + prot->salt_size,
1456 if (prot->version == TLS_1_3_VERSION)
1457 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1458 crypto_aead_ivsize(ctx->aead_recv));
1460 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1462 xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq);
1465 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1467 tls_ctx->rx.rec_seq, prot->rec_seq_size,
1468 ctx->control, prot->version);
1471 sg_init_table(sgin, n_sgin);
1472 sg_set_buf(&sgin[0], aad, prot->aad_size);
1473 err = skb_to_sgvec(skb, &sgin[1],
1474 rxm->offset + prot->prepend_size,
1475 rxm->full_len - prot->prepend_size);
1483 sg_init_table(sgout, n_sgout);
1484 sg_set_buf(&sgout[0], aad, prot->aad_size);
1487 err = tls_setup_from_iter(sk, out_iov, data_len,
1488 &pages, chunk, &sgout[1],
1491 goto fallback_to_reg_recv;
1492 } else if (out_sg) {
1493 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1495 goto fallback_to_reg_recv;
1498 fallback_to_reg_recv:
1505 /* Prepare and submit AEAD request */
1506 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1507 data_len, aead_req, async);
1508 if (err == -EINPROGRESS)
1511 /* Release the pages in case iov was mapped to pages */
1512 for (; pages > 0; pages--)
1513 put_page(sg_page(&sgout[pages]));
1519 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1520 struct iov_iter *dest, int *chunk, bool *zc,
1523 struct tls_context *tls_ctx = tls_get_ctx(sk);
1524 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1525 struct tls_prot_info *prot = &tls_ctx->prot_info;
1526 struct strp_msg *rxm = strp_msg(skb);
1529 if (!ctx->decrypted) {
1530 if (tls_ctx->rx_conf == TLS_HW) {
1531 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1536 /* Still not decrypted after tls_device */
1537 if (!ctx->decrypted) {
1538 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1541 if (err == -EINPROGRESS)
1542 tls_advance_record_sn(sk, prot,
1544 else if (err == -EBADMSG)
1545 TLS_INC_STATS(sock_net(sk),
1546 LINUX_MIB_TLSDECRYPTERROR);
1553 pad = padding_length(ctx, prot, skb);
1557 rxm->full_len -= pad;
1558 rxm->offset += prot->prepend_size;
1559 rxm->full_len -= prot->overhead_size;
1560 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1562 ctx->saved_data_ready(sk);
1570 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1571 struct scatterlist *sgout)
1576 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1579 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1582 struct tls_context *tls_ctx = tls_get_ctx(sk);
1583 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1586 struct strp_msg *rxm = strp_msg(skb);
1588 if (len < rxm->full_len) {
1590 rxm->full_len -= len;
1596 /* Finished with message */
1597 ctx->recv_pkt = NULL;
1598 __strp_unpause(&ctx->strp);
1603 /* This function traverses the rx_list in tls receive context to copies the
1604 * decrypted records into the buffer provided by caller zero copy is not
1605 * true. Further, the records are removed from the rx_list if it is not a peek
1606 * case and the record has been consumed completely.
1608 static int process_rx_list(struct tls_sw_context_rx *ctx,
1617 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1620 struct tls_msg *tlm;
1623 /* Set the record type in 'control' if caller didn't pass it */
1626 ctrl = tlm->control;
1629 while (skip && skb) {
1630 struct strp_msg *rxm = strp_msg(skb);
1633 /* Cannot process a record of different type */
1634 if (ctrl != tlm->control)
1637 if (skip < rxm->full_len)
1640 skip = skip - rxm->full_len;
1641 skb = skb_peek_next(skb, &ctx->rx_list);
1644 while (len && skb) {
1645 struct sk_buff *next_skb;
1646 struct strp_msg *rxm = strp_msg(skb);
1647 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1651 /* Cannot process a record of different type */
1652 if (ctrl != tlm->control)
1655 /* Set record type if not already done. For a non-data record,
1656 * do not proceed if record type could not be copied.
1659 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1660 sizeof(ctrl), &ctrl);
1662 if (ctrl != TLS_RECORD_TYPE_DATA) {
1663 if (cerr || msg->msg_flags & MSG_CTRUNC)
1670 if (!zc || (rxm->full_len - skip) > len) {
1671 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1678 copied = copied + chunk;
1680 /* Consume the data from record if it is non-peek case*/
1682 rxm->offset = rxm->offset + chunk;
1683 rxm->full_len = rxm->full_len - chunk;
1685 /* Return if there is unconsumed data in the record */
1686 if (rxm->full_len - skip)
1690 /* The remaining skip-bytes must lie in 1st record in rx_list.
1691 * So from the 2nd record, 'skip' should be 0.
1696 msg->msg_flags |= MSG_EOR;
1698 next_skb = skb_peek_next(skb, &ctx->rx_list);
1701 skb_unlink(skb, &ctx->rx_list);
1712 int tls_sw_recvmsg(struct sock *sk,
1719 struct tls_context *tls_ctx = tls_get_ctx(sk);
1720 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1721 struct tls_prot_info *prot = &tls_ctx->prot_info;
1722 struct sk_psock *psock;
1723 unsigned char control = 0;
1724 ssize_t decrypted = 0;
1725 struct strp_msg *rxm;
1726 struct tls_msg *tlm;
1727 struct sk_buff *skb;
1730 int target, err = 0;
1732 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1733 bool is_peek = flags & MSG_PEEK;
1738 if (unlikely(flags & MSG_ERRQUEUE))
1739 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1741 psock = sk_psock_get(sk);
1744 /* Process pending decrypted records. It must be non-zero-copy */
1745 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1748 tls_err_abort(sk, err);
1757 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1759 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1761 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1762 bool retain_skb = false;
1769 skb = tls_wait_data(sk, psock, flags, timeo, &err);
1772 int ret = __tcp_bpf_recvmsg(sk, psock,
1784 if (prot->version == TLS_1_3_VERSION)
1787 tlm->control = ctx->control;
1790 rxm = strp_msg(skb);
1792 to_decrypt = rxm->full_len - prot->overhead_size;
1794 if (to_decrypt <= len && !is_kvec && !is_peek &&
1795 ctx->control == TLS_RECORD_TYPE_DATA &&
1796 prot->version != TLS_1_3_VERSION)
1799 /* Do not use async mode if record is non-data */
1800 if (ctx->control == TLS_RECORD_TYPE_DATA)
1801 async_capable = ctx->async_capable;
1803 async_capable = false;
1805 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1806 &chunk, &zc, async_capable);
1807 if (err < 0 && err != -EINPROGRESS) {
1808 tls_err_abort(sk, EBADMSG);
1812 if (err == -EINPROGRESS) {
1815 } else if (prot->version == TLS_1_3_VERSION) {
1816 tlm->control = ctx->control;
1819 /* If the type of records being processed is not known yet,
1820 * set it to record type just dequeued. If it is already known,
1821 * but does not match the record type just dequeued, go to end.
1822 * We always get record type here since for tls1.2, record type
1823 * is known just after record is dequeued from stream parser.
1824 * For tls1.3, we disable async.
1828 control = tlm->control;
1829 else if (control != tlm->control)
1835 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1836 sizeof(control), &control);
1838 if (control != TLS_RECORD_TYPE_DATA) {
1839 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1847 goto pick_next_record;
1850 if (rxm->full_len > len) {
1854 chunk = rxm->full_len;
1857 err = skb_copy_datagram_msg(skb, rxm->offset,
1863 rxm->offset = rxm->offset + chunk;
1864 rxm->full_len = rxm->full_len - chunk;
1875 /* For async or peek case, queue the current skb */
1876 if (async || is_peek || retain_skb) {
1877 skb_queue_tail(&ctx->rx_list, skb);
1881 if (tls_sw_advance_skb(sk, skb, chunk)) {
1882 /* Return full control message to
1883 * userspace before trying to parse
1884 * another message type
1886 msg->msg_flags |= MSG_EOR;
1887 if (ctx->control != TLS_RECORD_TYPE_DATA)
1896 /* Wait for all previously submitted records to be decrypted */
1897 smp_store_mb(ctx->async_notify, true);
1898 if (atomic_read(&ctx->decrypt_pending)) {
1899 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1901 /* one of async decrypt failed */
1902 tls_err_abort(sk, err);
1908 reinit_completion(&ctx->async_wait.completion);
1910 WRITE_ONCE(ctx->async_notify, false);
1912 /* Drain records from the rx_list & copy if required */
1913 if (is_peek || is_kvec)
1914 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1915 decrypted, false, is_peek);
1917 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1918 decrypted, true, is_peek);
1920 tls_err_abort(sk, err);
1926 copied += decrypted;
1931 sk_psock_put(sk, psock);
1932 return copied ? : err;
1935 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1936 struct pipe_inode_info *pipe,
1937 size_t len, unsigned int flags)
1939 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1940 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1941 struct strp_msg *rxm = NULL;
1942 struct sock *sk = sock->sk;
1943 struct sk_buff *skb;
1952 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1954 skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1956 goto splice_read_end;
1958 if (!ctx->decrypted) {
1959 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
1961 /* splice does not support reading control messages */
1962 if (ctx->control != TLS_RECORD_TYPE_DATA) {
1964 goto splice_read_end;
1968 tls_err_abort(sk, EBADMSG);
1969 goto splice_read_end;
1973 rxm = strp_msg(skb);
1975 chunk = min_t(unsigned int, rxm->full_len, len);
1976 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
1978 goto splice_read_end;
1980 if (likely(!(flags & MSG_PEEK)))
1981 tls_sw_advance_skb(sk, skb, copied);
1985 return copied ? : err;
1988 bool tls_sw_stream_read(const struct sock *sk)
1990 struct tls_context *tls_ctx = tls_get_ctx(sk);
1991 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1992 bool ingress_empty = true;
1993 struct sk_psock *psock;
1996 psock = sk_psock(sk);
1998 ingress_empty = list_empty(&psock->ingress_msg);
2001 return !ingress_empty || ctx->recv_pkt ||
2002 !skb_queue_empty(&ctx->rx_list);
2005 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2007 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2008 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2009 struct tls_prot_info *prot = &tls_ctx->prot_info;
2010 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2011 struct strp_msg *rxm = strp_msg(skb);
2012 size_t cipher_overhead;
2013 size_t data_len = 0;
2016 /* Verify that we have a full TLS header, or wait for more data */
2017 if (rxm->offset + prot->prepend_size > skb->len)
2020 /* Sanity-check size of on-stack buffer. */
2021 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2026 /* Linearize header to local buffer */
2027 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2032 ctx->control = header[0];
2034 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2036 cipher_overhead = prot->tag_size;
2037 if (prot->version != TLS_1_3_VERSION)
2038 cipher_overhead += prot->iv_size;
2040 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2045 if (data_len < cipher_overhead) {
2050 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2051 if (header[1] != TLS_1_2_VERSION_MINOR ||
2052 header[2] != TLS_1_2_VERSION_MAJOR) {
2057 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2058 TCP_SKB_CB(skb)->seq + rxm->offset);
2059 return data_len + TLS_HEADER_SIZE;
2062 tls_err_abort(strp->sk, ret);
2067 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2069 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2070 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2074 ctx->recv_pkt = skb;
2077 ctx->saved_data_ready(strp->sk);
2080 static void tls_data_ready(struct sock *sk)
2082 struct tls_context *tls_ctx = tls_get_ctx(sk);
2083 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2084 struct sk_psock *psock;
2086 strp_data_ready(&ctx->strp);
2088 psock = sk_psock_get(sk);
2090 if (!list_empty(&psock->ingress_msg))
2091 ctx->saved_data_ready(sk);
2092 sk_psock_put(sk, psock);
2096 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2098 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2100 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2101 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2102 cancel_delayed_work_sync(&ctx->tx_work.work);
2105 void tls_sw_release_resources_tx(struct sock *sk)
2107 struct tls_context *tls_ctx = tls_get_ctx(sk);
2108 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2109 struct tls_rec *rec, *tmp;
2111 /* Wait for any pending async encryptions to complete */
2112 smp_store_mb(ctx->async_notify, true);
2113 if (atomic_read(&ctx->encrypt_pending))
2114 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2116 tls_tx_records(sk, -1);
2118 /* Free up un-sent records in tx_list. First, free
2119 * the partially sent record if any at head of tx_list.
2121 if (tls_ctx->partially_sent_record) {
2122 tls_free_partial_record(sk, tls_ctx);
2123 rec = list_first_entry(&ctx->tx_list,
2124 struct tls_rec, list);
2125 list_del(&rec->list);
2126 sk_msg_free(sk, &rec->msg_plaintext);
2130 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2131 list_del(&rec->list);
2132 sk_msg_free(sk, &rec->msg_encrypted);
2133 sk_msg_free(sk, &rec->msg_plaintext);
2137 crypto_free_aead(ctx->aead_send);
2138 tls_free_open_rec(sk);
2141 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2143 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2148 void tls_sw_release_resources_rx(struct sock *sk)
2150 struct tls_context *tls_ctx = tls_get_ctx(sk);
2151 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2153 kfree(tls_ctx->rx.rec_seq);
2154 kfree(tls_ctx->rx.iv);
2156 if (ctx->aead_recv) {
2157 kfree_skb(ctx->recv_pkt);
2158 ctx->recv_pkt = NULL;
2159 skb_queue_purge(&ctx->rx_list);
2160 crypto_free_aead(ctx->aead_recv);
2161 strp_stop(&ctx->strp);
2162 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2163 * we still want to strp_stop(), but sk->sk_data_ready was
2166 if (ctx->saved_data_ready) {
2167 write_lock_bh(&sk->sk_callback_lock);
2168 sk->sk_data_ready = ctx->saved_data_ready;
2169 write_unlock_bh(&sk->sk_callback_lock);
2174 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2176 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2178 strp_done(&ctx->strp);
2181 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2183 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2188 void tls_sw_free_resources_rx(struct sock *sk)
2190 struct tls_context *tls_ctx = tls_get_ctx(sk);
2192 tls_sw_release_resources_rx(sk);
2193 tls_sw_free_ctx_rx(tls_ctx);
2196 /* The work handler to transmitt the encrypted records in tx_list */
2197 static void tx_work_handler(struct work_struct *work)
2199 struct delayed_work *delayed_work = to_delayed_work(work);
2200 struct tx_work *tx_work = container_of(delayed_work,
2201 struct tx_work, work);
2202 struct sock *sk = tx_work->sk;
2203 struct tls_context *tls_ctx = tls_get_ctx(sk);
2204 struct tls_sw_context_tx *ctx;
2206 if (unlikely(!tls_ctx))
2209 ctx = tls_sw_ctx_tx(tls_ctx);
2210 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2213 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2215 mutex_lock(&tls_ctx->tx_lock);
2217 tls_tx_records(sk, -1);
2219 mutex_unlock(&tls_ctx->tx_lock);
2222 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2224 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2226 /* Schedule the transmission if tx list is ready */
2227 if (is_tx_ready(tx_ctx) &&
2228 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2229 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2232 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2234 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2236 write_lock_bh(&sk->sk_callback_lock);
2237 rx_ctx->saved_data_ready = sk->sk_data_ready;
2238 sk->sk_data_ready = tls_data_ready;
2239 write_unlock_bh(&sk->sk_callback_lock);
2241 strp_check_rcv(&rx_ctx->strp);
2244 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2246 struct tls_context *tls_ctx = tls_get_ctx(sk);
2247 struct tls_prot_info *prot = &tls_ctx->prot_info;
2248 struct tls_crypto_info *crypto_info;
2249 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2250 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2251 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2252 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2253 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2254 struct cipher_context *cctx;
2255 struct crypto_aead **aead;
2256 struct strp_callbacks cb;
2257 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2258 struct crypto_tfm *tfm;
2259 char *iv, *rec_seq, *key, *salt, *cipher_name;
2269 if (!ctx->priv_ctx_tx) {
2270 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2275 ctx->priv_ctx_tx = sw_ctx_tx;
2278 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2281 if (!ctx->priv_ctx_rx) {
2282 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2287 ctx->priv_ctx_rx = sw_ctx_rx;
2290 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2295 crypto_init_wait(&sw_ctx_tx->async_wait);
2296 crypto_info = &ctx->crypto_send.info;
2298 aead = &sw_ctx_tx->aead_send;
2299 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2300 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2301 sw_ctx_tx->tx_work.sk = sk;
2303 crypto_init_wait(&sw_ctx_rx->async_wait);
2304 crypto_info = &ctx->crypto_recv.info;
2306 skb_queue_head_init(&sw_ctx_rx->rx_list);
2307 aead = &sw_ctx_rx->aead_recv;
2310 switch (crypto_info->cipher_type) {
2311 case TLS_CIPHER_AES_GCM_128: {
2312 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2313 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2314 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2315 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2316 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2318 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2320 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2321 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2322 key = gcm_128_info->key;
2323 salt = gcm_128_info->salt;
2324 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2325 cipher_name = "gcm(aes)";
2328 case TLS_CIPHER_AES_GCM_256: {
2329 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2330 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2331 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2332 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2333 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2335 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2337 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2338 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2339 key = gcm_256_info->key;
2340 salt = gcm_256_info->salt;
2341 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2342 cipher_name = "gcm(aes)";
2345 case TLS_CIPHER_AES_CCM_128: {
2346 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2347 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2348 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2349 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2350 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2352 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2354 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2355 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2356 key = ccm_128_info->key;
2357 salt = ccm_128_info->salt;
2358 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2359 cipher_name = "ccm(aes)";
2367 /* Sanity-check the sizes for stack allocations. */
2368 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2369 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2374 if (crypto_info->version == TLS_1_3_VERSION) {
2376 prot->aad_size = TLS_HEADER_SIZE;
2377 prot->tail_size = 1;
2379 prot->aad_size = TLS_AAD_SPACE_SIZE;
2380 prot->tail_size = 0;
2383 prot->version = crypto_info->version;
2384 prot->cipher_type = crypto_info->cipher_type;
2385 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2386 prot->tag_size = tag_size;
2387 prot->overhead_size = prot->prepend_size +
2388 prot->tag_size + prot->tail_size;
2389 prot->iv_size = iv_size;
2390 prot->salt_size = salt_size;
2391 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2396 /* Note: 128 & 256 bit salt are the same size */
2397 prot->rec_seq_size = rec_seq_size;
2398 memcpy(cctx->iv, salt, salt_size);
2399 memcpy(cctx->iv + salt_size, iv, iv_size);
2400 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2401 if (!cctx->rec_seq) {
2407 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2408 if (IS_ERR(*aead)) {
2409 rc = PTR_ERR(*aead);
2415 ctx->push_pending_record = tls_sw_push_pending_record;
2417 rc = crypto_aead_setkey(*aead, key, keysize);
2422 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2427 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2429 if (crypto_info->version == TLS_1_3_VERSION)
2430 sw_ctx_rx->async_capable = 0;
2432 sw_ctx_rx->async_capable =
2433 !!(tfm->__crt_alg->cra_flags &
2436 /* Set up strparser */
2437 memset(&cb, 0, sizeof(cb));
2438 cb.rcv_msg = tls_queue;
2439 cb.parse_msg = tls_read_size;
2441 strp_init(&sw_ctx_rx->strp, sk, &cb);
2447 crypto_free_aead(*aead);
2450 kfree(cctx->rec_seq);
2451 cctx->rec_seq = NULL;
2457 kfree(ctx->priv_ctx_tx);
2458 ctx->priv_ctx_tx = NULL;
2460 kfree(ctx->priv_ctx_rx);
2461 ctx->priv_ctx_rx = NULL;