1 /* Copyright (c) 2018, Mellanox Technologies All rights reserved.
3 * This software is available to you under a choice of one of two
4 * licenses. You may choose to be licensed under the terms of the GNU
5 * General Public License (GPL) Version 2, available from the file
6 * COPYING in the main directory of this source tree, or the
7 * OpenIB.org BSD license below:
9 * Redistribution and use in source and binary forms, with or
10 * without modification, are permitted provided that the following
13 * - Redistributions of source code must retain the above
14 * copyright notice, this list of conditions and the following
17 * - Redistributions in binary form must reproduce the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer in the documentation and/or other materials
20 * provided with the distribution.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
23 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
24 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
25 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
26 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
27 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
28 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <crypto/aead.h>
33 #include <linux/highmem.h>
34 #include <linux/module.h>
35 #include <linux/netdevice.h>
37 #include <net/inet_connection_sock.h>
41 /* device_offload_lock is used to synchronize tls_dev_add
42 * against NETDEV_DOWN notifications.
44 static DECLARE_RWSEM(device_offload_lock);
46 static void tls_device_gc_task(struct work_struct *work);
48 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
49 static LIST_HEAD(tls_device_gc_list);
50 static LIST_HEAD(tls_device_list);
51 static DEFINE_SPINLOCK(tls_device_lock);
53 static void tls_device_free_ctx(struct tls_context *ctx)
55 if (ctx->tx_conf == TLS_HW) {
56 kfree(tls_offload_ctx_tx(ctx));
57 kfree(ctx->tx.rec_seq);
61 if (ctx->rx_conf == TLS_HW)
62 kfree(tls_offload_ctx_rx(ctx));
67 static void tls_device_gc_task(struct work_struct *work)
69 struct tls_context *ctx, *tmp;
73 spin_lock_irqsave(&tls_device_lock, flags);
74 list_splice_init(&tls_device_gc_list, &gc_list);
75 spin_unlock_irqrestore(&tls_device_lock, flags);
77 list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
78 struct net_device *netdev = ctx->netdev;
80 if (netdev && ctx->tx_conf == TLS_HW) {
81 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
82 TLS_OFFLOAD_CTX_DIR_TX);
88 tls_device_free_ctx(ctx);
92 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
96 spin_lock_irqsave(&tls_device_lock, flags);
97 list_move_tail(&ctx->list, &tls_device_gc_list);
99 /* schedule_work inside the spinlock
100 * to make sure tls_device_down waits for that work.
102 schedule_work(&tls_device_gc_work);
104 spin_unlock_irqrestore(&tls_device_lock, flags);
107 /* We assume that the socket is already connected */
108 static struct net_device *get_netdev_for_sock(struct sock *sk)
110 struct dst_entry *dst = sk_dst_get(sk);
111 struct net_device *netdev = NULL;
123 static void destroy_record(struct tls_record_info *record)
125 int nr_frags = record->num_frags;
128 while (nr_frags-- > 0) {
129 frag = &record->frags[nr_frags];
130 __skb_frag_unref(frag);
135 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
137 struct tls_record_info *info, *temp;
139 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
140 list_del(&info->list);
141 destroy_record(info);
144 offload_ctx->retransmit_hint = NULL;
147 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
149 struct tls_context *tls_ctx = tls_get_ctx(sk);
150 struct tls_record_info *info, *temp;
151 struct tls_offload_context_tx *ctx;
152 u64 deleted_records = 0;
158 ctx = tls_offload_ctx_tx(tls_ctx);
160 spin_lock_irqsave(&ctx->lock, flags);
161 info = ctx->retransmit_hint;
162 if (info && !before(acked_seq, info->end_seq)) {
163 ctx->retransmit_hint = NULL;
164 list_del(&info->list);
165 destroy_record(info);
169 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
170 if (before(acked_seq, info->end_seq))
172 list_del(&info->list);
174 destroy_record(info);
178 ctx->unacked_record_sn += deleted_records;
179 spin_unlock_irqrestore(&ctx->lock, flags);
182 /* At this point, there should be no references on this
183 * socket and no in-flight SKBs associated with this
184 * socket, so it is safe to free all the resources.
186 static void tls_device_sk_destruct(struct sock *sk)
188 struct tls_context *tls_ctx = tls_get_ctx(sk);
189 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
191 tls_ctx->sk_destruct(sk);
193 if (tls_ctx->tx_conf == TLS_HW) {
194 if (ctx->open_record)
195 destroy_record(ctx->open_record);
196 delete_all_records(ctx);
197 crypto_free_aead(ctx->aead_send);
198 clean_acked_data_disable(inet_csk(sk));
201 if (refcount_dec_and_test(&tls_ctx->refcount))
202 tls_device_queue_ctx_destruction(tls_ctx);
205 void tls_device_free_resources_tx(struct sock *sk)
207 struct tls_context *tls_ctx = tls_get_ctx(sk);
209 tls_free_partial_record(sk, tls_ctx);
212 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
215 struct net_device *netdev;
220 skb = tcp_write_queue_tail(sk);
222 TCP_SKB_CB(skb)->eor = 1;
224 rcd_sn = tls_ctx->tx.rec_seq;
226 down_read(&device_offload_lock);
227 netdev = tls_ctx->netdev;
229 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
231 TLS_OFFLOAD_CTX_DIR_TX);
232 up_read(&device_offload_lock);
236 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
239 static void tls_append_frag(struct tls_record_info *record,
240 struct page_frag *pfrag,
245 frag = &record->frags[record->num_frags - 1];
246 if (frag->page.p == pfrag->page &&
247 frag->page_offset + frag->size == pfrag->offset) {
251 frag->page.p = pfrag->page;
252 frag->page_offset = pfrag->offset;
255 get_page(pfrag->page);
258 pfrag->offset += size;
262 static int tls_push_record(struct sock *sk,
263 struct tls_context *ctx,
264 struct tls_offload_context_tx *offload_ctx,
265 struct tls_record_info *record,
266 struct page_frag *pfrag,
268 unsigned char record_type)
270 struct tls_prot_info *prot = &ctx->prot_info;
271 struct tcp_sock *tp = tcp_sk(sk);
272 struct page_frag dummy_tag_frag;
277 frag = &record->frags[0];
278 tls_fill_prepend(ctx,
279 skb_frag_address(frag),
280 record->len - prot->prepend_size,
284 /* HW doesn't care about the data in the tag, because it fills it. */
285 dummy_tag_frag.page = skb_frag_page(frag);
286 dummy_tag_frag.offset = 0;
288 tls_append_frag(record, &dummy_tag_frag, prot->tag_size);
289 record->end_seq = tp->write_seq + record->len;
290 spin_lock_irq(&offload_ctx->lock);
291 list_add_tail(&record->list, &offload_ctx->records_list);
292 spin_unlock_irq(&offload_ctx->lock);
293 offload_ctx->open_record = NULL;
295 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
296 tls_device_resync_tx(sk, ctx, tp->write_seq);
298 tls_advance_record_sn(sk, prot, &ctx->tx);
300 for (i = 0; i < record->num_frags; i++) {
301 frag = &record->frags[i];
302 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
303 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
304 frag->size, frag->page_offset);
305 sk_mem_charge(sk, frag->size);
306 get_page(skb_frag_page(frag));
308 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
310 /* all ready, send */
311 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
314 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
315 struct page_frag *pfrag,
318 struct tls_record_info *record;
321 record = kmalloc(sizeof(*record), GFP_KERNEL);
325 frag = &record->frags[0];
326 __skb_frag_set_page(frag, pfrag->page);
327 frag->page_offset = pfrag->offset;
328 skb_frag_size_set(frag, prepend_size);
330 get_page(pfrag->page);
331 pfrag->offset += prepend_size;
333 record->num_frags = 1;
334 record->len = prepend_size;
335 offload_ctx->open_record = record;
339 static int tls_do_allocation(struct sock *sk,
340 struct tls_offload_context_tx *offload_ctx,
341 struct page_frag *pfrag,
346 if (!offload_ctx->open_record) {
347 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
348 sk->sk_allocation))) {
349 sk->sk_prot->enter_memory_pressure(sk);
350 sk_stream_moderate_sndbuf(sk);
354 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
358 if (pfrag->size > pfrag->offset)
362 if (!sk_page_frag_refill(sk, pfrag))
368 static int tls_push_data(struct sock *sk,
369 struct iov_iter *msg_iter,
370 size_t size, int flags,
371 unsigned char record_type)
373 struct tls_context *tls_ctx = tls_get_ctx(sk);
374 struct tls_prot_info *prot = &tls_ctx->prot_info;
375 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
376 int more = flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE);
377 struct tls_record_info *record = ctx->open_record;
378 int tls_push_record_flags;
379 struct page_frag *pfrag;
380 size_t orig_size = size;
381 u32 max_open_record_len;
387 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
393 flags |= MSG_SENDPAGE_DECRYPTED;
394 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
396 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
397 if (tls_is_partially_sent_record(tls_ctx)) {
398 rc = tls_push_partial_record(sk, tls_ctx, flags);
403 pfrag = sk_page_frag(sk);
405 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
406 * we need to leave room for an authentication tag.
408 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
411 rc = tls_do_allocation(sk, ctx, pfrag,
414 rc = sk_stream_wait_memory(sk, &timeo);
418 record = ctx->open_record;
422 if (record_type != TLS_RECORD_TYPE_DATA) {
423 /* avoid sending partial
424 * record with type !=
428 destroy_record(record);
429 ctx->open_record = NULL;
430 } else if (record->len > prot->prepend_size) {
437 record = ctx->open_record;
438 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
439 copy = min_t(size_t, copy, (max_open_record_len - record->len));
441 if (copy_from_iter_nocache(page_address(pfrag->page) +
443 copy, msg_iter) != copy) {
447 tls_append_frag(record, pfrag, copy);
452 tls_push_record_flags = flags;
454 tls_ctx->pending_open_record_frags =
462 if (done || record->len >= max_open_record_len ||
463 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
464 rc = tls_push_record(sk,
469 tls_push_record_flags,
476 if (orig_size - size > 0)
477 rc = orig_size - size;
482 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
484 unsigned char record_type = TLS_RECORD_TYPE_DATA;
489 if (unlikely(msg->msg_controllen)) {
490 rc = tls_proccess_cmsg(sk, msg, &record_type);
495 rc = tls_push_data(sk, &msg->msg_iter, size,
496 msg->msg_flags, record_type);
503 int tls_device_sendpage(struct sock *sk, struct page *page,
504 int offset, size_t size, int flags)
506 struct iov_iter msg_iter;
507 char *kaddr = kmap(page);
511 if (flags & MSG_SENDPAGE_NOTLAST)
516 if (flags & MSG_OOB) {
521 iov.iov_base = kaddr + offset;
523 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
524 rc = tls_push_data(sk, &msg_iter, size,
525 flags, TLS_RECORD_TYPE_DATA);
533 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
534 u32 seq, u64 *p_record_sn)
536 u64 record_sn = context->hint_record_sn;
537 struct tls_record_info *info;
539 info = context->retransmit_hint;
541 before(seq, info->end_seq - info->len)) {
542 /* if retransmit_hint is irrelevant start
543 * from the beggining of the list
545 info = list_first_entry(&context->records_list,
546 struct tls_record_info, list);
547 record_sn = context->unacked_record_sn;
550 list_for_each_entry_from(info, &context->records_list, list) {
551 if (before(seq, info->end_seq)) {
552 if (!context->retransmit_hint ||
554 context->retransmit_hint->end_seq)) {
555 context->hint_record_sn = record_sn;
556 context->retransmit_hint = info;
558 *p_record_sn = record_sn;
566 EXPORT_SYMBOL(tls_get_record);
568 static int tls_device_push_pending_record(struct sock *sk, int flags)
570 struct iov_iter msg_iter;
572 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
573 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
576 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
578 if (!sk->sk_write_pending && tls_is_partially_sent_record(ctx)) {
579 gfp_t sk_allocation = sk->sk_allocation;
581 sk->sk_allocation = GFP_ATOMIC;
582 tls_push_partial_record(sk, ctx,
583 MSG_DONTWAIT | MSG_NOSIGNAL |
584 MSG_SENDPAGE_DECRYPTED);
585 sk->sk_allocation = sk_allocation;
589 static void tls_device_resync_rx(struct tls_context *tls_ctx,
590 struct sock *sk, u32 seq, u8 *rcd_sn)
592 struct net_device *netdev;
594 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
596 netdev = READ_ONCE(tls_ctx->netdev);
598 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
599 TLS_OFFLOAD_CTX_DIR_RX);
600 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
603 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
605 struct tls_context *tls_ctx = tls_get_ctx(sk);
606 struct tls_offload_context_rx *rx_ctx;
607 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
608 struct tls_prot_info *prot;
613 if (tls_ctx->rx_conf != TLS_HW)
616 prot = &tls_ctx->prot_info;
617 rx_ctx = tls_offload_ctx_rx(tls_ctx);
618 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
620 switch (rx_ctx->resync_type) {
621 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
622 resync_req = atomic64_read(&rx_ctx->resync_req);
623 req_seq = resync_req >> 32;
624 seq += TLS_HEADER_SIZE - 1;
625 is_req_pending = resync_req;
627 if (likely(!is_req_pending) || req_seq != seq ||
628 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
631 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
632 if (likely(!rx_ctx->resync_nh_do_now))
635 /* head of next rec is already in, note that the sock_inq will
636 * include the currently parsed message when called from parser
638 if (tcp_inq(sk) > rcd_len)
641 rx_ctx->resync_nh_do_now = 0;
643 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
647 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
650 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
651 struct tls_offload_context_rx *ctx,
652 struct sock *sk, struct sk_buff *skb)
654 struct strp_msg *rxm;
656 /* device will request resyncs by itself based on stream scan */
657 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
659 /* already scheduled */
660 if (ctx->resync_nh_do_now)
662 /* seen decrypted fragments since last fully-failed record */
663 if (ctx->resync_nh_reset) {
664 ctx->resync_nh_reset = 0;
665 ctx->resync_nh.decrypted_failed = 1;
666 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
670 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
673 /* doing resync, bump the next target in case it fails */
674 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
675 ctx->resync_nh.decrypted_tgt *= 2;
677 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
681 /* head of next rec is already in, parser will sync for us */
682 if (tcp_inq(sk) > rxm->full_len) {
683 ctx->resync_nh_do_now = 1;
685 struct tls_prot_info *prot = &tls_ctx->prot_info;
686 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
688 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
689 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
691 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
696 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
698 struct strp_msg *rxm = strp_msg(skb);
699 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
700 struct sk_buff *skb_iter, *unused;
701 struct scatterlist sg[1];
702 char *orig_buf, *buf;
704 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
705 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
710 nsg = skb_cow_data(skb, 0, &unused);
711 if (unlikely(nsg < 0)) {
716 sg_init_table(sg, 1);
717 sg_set_buf(&sg[0], buf,
718 rxm->full_len + TLS_HEADER_SIZE +
719 TLS_CIPHER_AES_GCM_128_IV_SIZE);
720 err = skb_copy_bits(skb, offset, buf,
721 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
725 /* We are interested only in the decrypted data not the auth */
726 err = decrypt_skb(sk, skb, sg);
732 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
734 if (skb_pagelen(skb) > offset) {
735 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
737 if (skb->decrypted) {
738 err = skb_store_bits(skb, offset, buf, copy);
747 pos = skb_pagelen(skb);
748 skb_walk_frags(skb, skb_iter) {
751 /* Practically all frags must belong to msg if reencrypt
752 * is needed with current strparser and coalescing logic,
753 * but strparser may "get optimized", so let's be safe.
755 if (pos + skb_iter->len <= offset)
757 if (pos >= data_len + rxm->offset)
760 frag_pos = offset - pos;
761 copy = min_t(int, skb_iter->len - frag_pos,
762 data_len + rxm->offset - offset);
764 if (skb_iter->decrypted) {
765 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
773 pos += skb_iter->len;
781 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
783 struct tls_context *tls_ctx = tls_get_ctx(sk);
784 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
785 int is_decrypted = skb->decrypted;
786 int is_encrypted = !is_decrypted;
787 struct sk_buff *skb_iter;
789 /* Check if all the data is decrypted already */
790 skb_walk_frags(skb, skb_iter) {
791 is_decrypted &= skb_iter->decrypted;
792 is_encrypted &= !skb_iter->decrypted;
795 ctx->sw.decrypted |= is_decrypted;
797 /* Return immediately if the record is either entirely plaintext or
798 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
802 ctx->resync_nh_reset = 1;
806 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
810 ctx->resync_nh_reset = 1;
811 return tls_device_reencrypt(sk, skb);
814 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
815 struct net_device *netdev)
817 if (sk->sk_destruct != tls_device_sk_destruct) {
818 refcount_set(&ctx->refcount, 1);
820 ctx->netdev = netdev;
821 spin_lock_irq(&tls_device_lock);
822 list_add_tail(&ctx->list, &tls_device_list);
823 spin_unlock_irq(&tls_device_lock);
825 ctx->sk_destruct = sk->sk_destruct;
826 sk->sk_destruct = tls_device_sk_destruct;
830 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
832 u16 nonce_size, tag_size, iv_size, rec_seq_size;
833 struct tls_context *tls_ctx = tls_get_ctx(sk);
834 struct tls_prot_info *prot = &tls_ctx->prot_info;
835 struct tls_record_info *start_marker_record;
836 struct tls_offload_context_tx *offload_ctx;
837 struct tls_crypto_info *crypto_info;
838 struct net_device *netdev;
847 if (ctx->priv_ctx_tx) {
852 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
853 if (!start_marker_record) {
858 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
861 goto free_marker_record;
864 crypto_info = &ctx->crypto_send.info;
865 if (crypto_info->version != TLS_1_2_VERSION) {
867 goto free_offload_ctx;
870 switch (crypto_info->cipher_type) {
871 case TLS_CIPHER_AES_GCM_128:
872 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
873 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
874 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
875 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
876 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
878 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
882 goto free_offload_ctx;
885 /* Sanity-check the rec_seq_size for stack allocations */
886 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
888 goto free_offload_ctx;
891 prot->version = crypto_info->version;
892 prot->cipher_type = crypto_info->cipher_type;
893 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
894 prot->tag_size = tag_size;
895 prot->overhead_size = prot->prepend_size + prot->tag_size;
896 prot->iv_size = iv_size;
897 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
901 goto free_offload_ctx;
904 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
906 prot->rec_seq_size = rec_seq_size;
907 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
908 if (!ctx->tx.rec_seq) {
913 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
917 /* start at rec_seq - 1 to account for the start marker record */
918 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
919 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
921 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
922 start_marker_record->len = 0;
923 start_marker_record->num_frags = 0;
925 INIT_LIST_HEAD(&offload_ctx->records_list);
926 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
927 spin_lock_init(&offload_ctx->lock);
928 sg_init_table(offload_ctx->sg_tx_data,
929 ARRAY_SIZE(offload_ctx->sg_tx_data));
931 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
932 ctx->push_pending_record = tls_device_push_pending_record;
934 /* TLS offload is greatly simplified if we don't send
935 * SKBs where only part of the payload needs to be encrypted.
936 * So mark the last skb in the write queue as end of record.
938 skb = tcp_write_queue_tail(sk);
940 TCP_SKB_CB(skb)->eor = 1;
942 /* We support starting offload on multiple sockets
943 * concurrently, so we only need a read lock here.
944 * This lock must precede get_netdev_for_sock to prevent races between
945 * NETDEV_DOWN and setsockopt.
947 down_read(&device_offload_lock);
948 netdev = get_netdev_for_sock(sk);
950 pr_err_ratelimited("%s: netdev not found\n", __func__);
955 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
960 /* Avoid offloading if the device is down
961 * We don't want to offload new flows after
962 * the NETDEV_DOWN event
964 if (!(netdev->flags & IFF_UP)) {
969 ctx->priv_ctx_tx = offload_ctx;
970 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
971 &ctx->crypto_send.info,
972 tcp_sk(sk)->write_seq);
976 tls_device_attach(ctx, sk, netdev);
978 /* following this assignment tls_is_sk_tx_device_offloaded
979 * will return true and the context might be accessed
980 * by the netdev's xmit function.
982 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
984 up_read(&device_offload_lock);
990 up_read(&device_offload_lock);
991 clean_acked_data_disable(inet_csk(sk));
992 crypto_free_aead(offload_ctx->aead_send);
994 kfree(ctx->tx.rec_seq);
999 ctx->priv_ctx_tx = NULL;
1001 kfree(start_marker_record);
1006 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1008 struct tls_offload_context_rx *context;
1009 struct net_device *netdev;
1012 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1015 /* We support starting offload on multiple sockets
1016 * concurrently, so we only need a read lock here.
1017 * This lock must precede get_netdev_for_sock to prevent races between
1018 * NETDEV_DOWN and setsockopt.
1020 down_read(&device_offload_lock);
1021 netdev = get_netdev_for_sock(sk);
1023 pr_err_ratelimited("%s: netdev not found\n", __func__);
1028 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1030 goto release_netdev;
1033 /* Avoid offloading if the device is down
1034 * We don't want to offload new flows after
1035 * the NETDEV_DOWN event
1037 if (!(netdev->flags & IFF_UP)) {
1039 goto release_netdev;
1042 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1045 goto release_netdev;
1047 context->resync_nh_reset = 1;
1049 ctx->priv_ctx_rx = context;
1050 rc = tls_set_sw_offload(sk, ctx, 0);
1054 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1055 &ctx->crypto_recv.info,
1056 tcp_sk(sk)->copied_seq);
1058 goto free_sw_resources;
1060 tls_device_attach(ctx, sk, netdev);
1061 goto release_netdev;
1064 up_read(&device_offload_lock);
1065 tls_sw_free_resources_rx(sk);
1066 down_read(&device_offload_lock);
1068 ctx->priv_ctx_rx = NULL;
1072 up_read(&device_offload_lock);
1076 void tls_device_offload_cleanup_rx(struct sock *sk)
1078 struct tls_context *tls_ctx = tls_get_ctx(sk);
1079 struct net_device *netdev;
1081 down_read(&device_offload_lock);
1082 netdev = tls_ctx->netdev;
1086 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1087 TLS_OFFLOAD_CTX_DIR_RX);
1089 if (tls_ctx->tx_conf != TLS_HW) {
1091 tls_ctx->netdev = NULL;
1094 up_read(&device_offload_lock);
1095 tls_sw_release_resources_rx(sk);
1098 static int tls_device_down(struct net_device *netdev)
1100 struct tls_context *ctx, *tmp;
1101 unsigned long flags;
1104 /* Request a write lock to block new offload attempts */
1105 down_write(&device_offload_lock);
1107 spin_lock_irqsave(&tls_device_lock, flags);
1108 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1109 if (ctx->netdev != netdev ||
1110 !refcount_inc_not_zero(&ctx->refcount))
1113 list_move(&ctx->list, &list);
1115 spin_unlock_irqrestore(&tls_device_lock, flags);
1117 list_for_each_entry_safe(ctx, tmp, &list, list) {
1118 if (ctx->tx_conf == TLS_HW)
1119 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1120 TLS_OFFLOAD_CTX_DIR_TX);
1121 if (ctx->rx_conf == TLS_HW)
1122 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1123 TLS_OFFLOAD_CTX_DIR_RX);
1124 WRITE_ONCE(ctx->netdev, NULL);
1125 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1126 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1127 usleep_range(10, 200);
1129 list_del_init(&ctx->list);
1131 if (refcount_dec_and_test(&ctx->refcount))
1132 tls_device_free_ctx(ctx);
1135 up_write(&device_offload_lock);
1137 flush_work(&tls_device_gc_work);
1142 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1145 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1147 if (!dev->tlsdev_ops &&
1148 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1152 case NETDEV_REGISTER:
1153 case NETDEV_FEAT_CHANGE:
1154 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1155 !dev->tlsdev_ops->tls_dev_resync)
1158 if (dev->tlsdev_ops &&
1159 dev->tlsdev_ops->tls_dev_add &&
1160 dev->tlsdev_ops->tls_dev_del)
1165 return tls_device_down(dev);
1170 static struct notifier_block tls_dev_notifier = {
1171 .notifier_call = tls_dev_event,
1174 void __init tls_device_init(void)
1176 register_netdevice_notifier(&tls_dev_notifier);
1179 void __exit tls_device_cleanup(void)
1181 unregister_netdevice_notifier(&tls_dev_notifier);
1182 flush_work(&tls_device_gc_work);
1183 clean_acked_data_flush();