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_attach(struct tls_context *ctx, struct sock *sk,
93 struct net_device *netdev)
95 if (sk->sk_destruct != tls_device_sk_destruct) {
96 refcount_set(&ctx->refcount, 1);
99 spin_lock_irq(&tls_device_lock);
100 list_add_tail(&ctx->list, &tls_device_list);
101 spin_unlock_irq(&tls_device_lock);
103 ctx->sk_destruct = sk->sk_destruct;
104 sk->sk_destruct = tls_device_sk_destruct;
108 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
112 spin_lock_irqsave(&tls_device_lock, flags);
113 list_move_tail(&ctx->list, &tls_device_gc_list);
115 /* schedule_work inside the spinlock
116 * to make sure tls_device_down waits for that work.
118 schedule_work(&tls_device_gc_work);
120 spin_unlock_irqrestore(&tls_device_lock, flags);
123 /* We assume that the socket is already connected */
124 static struct net_device *get_netdev_for_sock(struct sock *sk)
126 struct dst_entry *dst = sk_dst_get(sk);
127 struct net_device *netdev = NULL;
139 static void destroy_record(struct tls_record_info *record)
141 int nr_frags = record->num_frags;
144 while (nr_frags-- > 0) {
145 frag = &record->frags[nr_frags];
146 __skb_frag_unref(frag);
151 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
153 struct tls_record_info *info, *temp;
155 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
156 list_del(&info->list);
157 destroy_record(info);
160 offload_ctx->retransmit_hint = NULL;
163 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
165 struct tls_context *tls_ctx = tls_get_ctx(sk);
166 struct tls_record_info *info, *temp;
167 struct tls_offload_context_tx *ctx;
168 u64 deleted_records = 0;
174 ctx = tls_offload_ctx_tx(tls_ctx);
176 spin_lock_irqsave(&ctx->lock, flags);
177 info = ctx->retransmit_hint;
178 if (info && !before(acked_seq, info->end_seq)) {
179 ctx->retransmit_hint = NULL;
180 list_del(&info->list);
181 destroy_record(info);
185 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
186 if (before(acked_seq, info->end_seq))
188 list_del(&info->list);
190 destroy_record(info);
194 ctx->unacked_record_sn += deleted_records;
195 spin_unlock_irqrestore(&ctx->lock, flags);
198 /* At this point, there should be no references on this
199 * socket and no in-flight SKBs associated with this
200 * socket, so it is safe to free all the resources.
202 void tls_device_sk_destruct(struct sock *sk)
204 struct tls_context *tls_ctx = tls_get_ctx(sk);
205 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
207 tls_ctx->sk_destruct(sk);
209 if (tls_ctx->tx_conf == TLS_HW) {
210 if (ctx->open_record)
211 destroy_record(ctx->open_record);
212 delete_all_records(ctx);
213 crypto_free_aead(ctx->aead_send);
214 clean_acked_data_disable(inet_csk(sk));
217 if (refcount_dec_and_test(&tls_ctx->refcount))
218 tls_device_queue_ctx_destruction(tls_ctx);
220 EXPORT_SYMBOL(tls_device_sk_destruct);
222 static void tls_append_frag(struct tls_record_info *record,
223 struct page_frag *pfrag,
228 frag = &record->frags[record->num_frags - 1];
229 if (frag->page.p == pfrag->page &&
230 frag->page_offset + frag->size == pfrag->offset) {
234 frag->page.p = pfrag->page;
235 frag->page_offset = pfrag->offset;
238 get_page(pfrag->page);
241 pfrag->offset += size;
245 static int tls_push_record(struct sock *sk,
246 struct tls_context *ctx,
247 struct tls_offload_context_tx *offload_ctx,
248 struct tls_record_info *record,
249 struct page_frag *pfrag,
251 unsigned char record_type)
253 struct tcp_sock *tp = tcp_sk(sk);
254 struct page_frag dummy_tag_frag;
259 frag = &record->frags[0];
260 tls_fill_prepend(ctx,
261 skb_frag_address(frag),
262 record->len - ctx->tx.prepend_size,
265 /* HW doesn't care about the data in the tag, because it fills it. */
266 dummy_tag_frag.page = skb_frag_page(frag);
267 dummy_tag_frag.offset = 0;
269 tls_append_frag(record, &dummy_tag_frag, ctx->tx.tag_size);
270 record->end_seq = tp->write_seq + record->len;
271 spin_lock_irq(&offload_ctx->lock);
272 list_add_tail(&record->list, &offload_ctx->records_list);
273 spin_unlock_irq(&offload_ctx->lock);
274 offload_ctx->open_record = NULL;
275 set_bit(TLS_PENDING_CLOSED_RECORD, &ctx->flags);
276 tls_advance_record_sn(sk, &ctx->tx);
278 for (i = 0; i < record->num_frags; i++) {
279 frag = &record->frags[i];
280 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
281 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
282 frag->size, frag->page_offset);
283 sk_mem_charge(sk, frag->size);
284 get_page(skb_frag_page(frag));
286 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
288 /* all ready, send */
289 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
292 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
293 struct page_frag *pfrag,
296 struct tls_record_info *record;
299 record = kmalloc(sizeof(*record), GFP_KERNEL);
303 frag = &record->frags[0];
304 __skb_frag_set_page(frag, pfrag->page);
305 frag->page_offset = pfrag->offset;
306 skb_frag_size_set(frag, prepend_size);
308 get_page(pfrag->page);
309 pfrag->offset += prepend_size;
311 record->num_frags = 1;
312 record->len = prepend_size;
313 offload_ctx->open_record = record;
317 static int tls_do_allocation(struct sock *sk,
318 struct tls_offload_context_tx *offload_ctx,
319 struct page_frag *pfrag,
324 if (!offload_ctx->open_record) {
325 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
326 sk->sk_allocation))) {
327 sk->sk_prot->enter_memory_pressure(sk);
328 sk_stream_moderate_sndbuf(sk);
332 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
336 if (pfrag->size > pfrag->offset)
340 if (!sk_page_frag_refill(sk, pfrag))
346 static int tls_push_data(struct sock *sk,
347 struct iov_iter *msg_iter,
348 size_t size, int flags,
349 unsigned char record_type)
351 struct tls_context *tls_ctx = tls_get_ctx(sk);
352 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
353 int tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
354 struct tls_record_info *record = ctx->open_record;
355 struct page_frag *pfrag;
356 size_t orig_size = size;
357 u32 max_open_record_len;
364 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
370 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
371 rc = tls_complete_pending_work(sk, tls_ctx, flags, &timeo);
375 pfrag = sk_page_frag(sk);
377 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
378 * we need to leave room for an authentication tag.
380 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
381 tls_ctx->tx.prepend_size;
383 rc = tls_do_allocation(sk, ctx, pfrag,
384 tls_ctx->tx.prepend_size);
386 rc = sk_stream_wait_memory(sk, &timeo);
390 record = ctx->open_record;
394 if (record_type != TLS_RECORD_TYPE_DATA) {
395 /* avoid sending partial
396 * record with type !=
400 destroy_record(record);
401 ctx->open_record = NULL;
402 } else if (record->len > tls_ctx->tx.prepend_size) {
409 record = ctx->open_record;
410 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
411 copy = min_t(size_t, copy, (max_open_record_len - record->len));
413 if (copy_from_iter_nocache(page_address(pfrag->page) +
415 copy, msg_iter) != copy) {
419 tls_append_frag(record, pfrag, copy);
424 tls_push_record_flags = flags;
425 if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
433 if (done || record->len >= max_open_record_len ||
434 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
435 rc = tls_push_record(sk,
440 tls_push_record_flags,
447 tls_ctx->pending_open_record_frags = more;
449 if (orig_size - size > 0)
450 rc = orig_size - size;
455 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
457 unsigned char record_type = TLS_RECORD_TYPE_DATA;
462 if (unlikely(msg->msg_controllen)) {
463 rc = tls_proccess_cmsg(sk, msg, &record_type);
468 rc = tls_push_data(sk, &msg->msg_iter, size,
469 msg->msg_flags, record_type);
476 int tls_device_sendpage(struct sock *sk, struct page *page,
477 int offset, size_t size, int flags)
479 struct iov_iter msg_iter;
484 if (flags & MSG_SENDPAGE_NOTLAST)
489 if (flags & MSG_OOB) {
495 iov.iov_base = kaddr + offset;
497 iov_iter_kvec(&msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
498 rc = tls_push_data(sk, &msg_iter, size,
499 flags, TLS_RECORD_TYPE_DATA);
507 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
508 u32 seq, u64 *p_record_sn)
510 u64 record_sn = context->hint_record_sn;
511 struct tls_record_info *info, *last;
513 info = context->retransmit_hint;
515 before(seq, info->end_seq - info->len)) {
516 /* if retransmit_hint is irrelevant start
517 * from the beggining of the list
519 info = list_first_entry(&context->records_list,
520 struct tls_record_info, list);
522 /* send the start_marker record if seq number is before the
523 * tls offload start marker sequence number. This record is
524 * required to handle TCP packets which are before TLS offload
526 * And if it's not start marker, look if this seq number
527 * belongs to the list.
529 if (likely(!tls_record_is_start_marker(info))) {
530 /* we have the first record, get the last record to see
531 * if this seq number belongs to the list.
533 last = list_last_entry(&context->records_list,
534 struct tls_record_info, list);
536 if (!between(seq, tls_record_start_seq(info),
540 record_sn = context->unacked_record_sn;
543 list_for_each_entry_from(info, &context->records_list, list) {
544 if (before(seq, info->end_seq)) {
545 if (!context->retransmit_hint ||
547 context->retransmit_hint->end_seq)) {
548 context->hint_record_sn = record_sn;
549 context->retransmit_hint = info;
551 *p_record_sn = record_sn;
559 EXPORT_SYMBOL(tls_get_record);
561 static int tls_device_push_pending_record(struct sock *sk, int flags)
563 struct iov_iter msg_iter;
565 iov_iter_kvec(&msg_iter, WRITE | ITER_KVEC, NULL, 0, 0);
566 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
569 static void tls_device_resync_rx(struct tls_context *tls_ctx,
570 struct sock *sk, u32 seq, u64 rcd_sn)
572 struct net_device *netdev;
574 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
576 netdev = READ_ONCE(tls_ctx->netdev);
578 netdev->tlsdev_ops->tls_dev_resync_rx(netdev, sk, seq, rcd_sn);
579 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
582 void handle_device_resync(struct sock *sk, u32 seq, u64 rcd_sn)
584 struct tls_context *tls_ctx = tls_get_ctx(sk);
585 struct tls_offload_context_rx *rx_ctx;
590 if (tls_ctx->rx_conf != TLS_HW)
593 rx_ctx = tls_offload_ctx_rx(tls_ctx);
594 resync_req = atomic64_read(&rx_ctx->resync_req);
595 req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
596 is_req_pending = resync_req;
598 if (unlikely(is_req_pending) && req_seq == seq &&
599 atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0)) {
600 seq += TLS_HEADER_SIZE - 1;
601 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
605 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
607 struct strp_msg *rxm = strp_msg(skb);
608 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
609 struct sk_buff *skb_iter, *unused;
610 struct scatterlist sg[1];
611 char *orig_buf, *buf;
613 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
614 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
619 nsg = skb_cow_data(skb, 0, &unused);
620 if (unlikely(nsg < 0)) {
625 sg_init_table(sg, 1);
626 sg_set_buf(&sg[0], buf,
627 rxm->full_len + TLS_HEADER_SIZE +
628 TLS_CIPHER_AES_GCM_128_IV_SIZE);
629 skb_copy_bits(skb, offset, buf,
630 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
632 /* We are interested only in the decrypted data not the auth */
633 err = decrypt_skb(sk, skb, sg);
639 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
641 if (skb_pagelen(skb) > offset) {
642 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
645 skb_store_bits(skb, offset, buf, copy);
651 pos = skb_pagelen(skb);
652 skb_walk_frags(skb, skb_iter) {
655 /* Practically all frags must belong to msg if reencrypt
656 * is needed with current strparser and coalescing logic,
657 * but strparser may "get optimized", so let's be safe.
659 if (pos + skb_iter->len <= offset)
661 if (pos >= data_len + rxm->offset)
664 frag_pos = offset - pos;
665 copy = min_t(int, skb_iter->len - frag_pos,
666 data_len + rxm->offset - offset);
668 if (skb_iter->decrypted)
669 skb_store_bits(skb_iter, frag_pos, buf, copy);
674 pos += skb_iter->len;
682 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
684 struct tls_context *tls_ctx = tls_get_ctx(sk);
685 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
686 int is_decrypted = skb->decrypted;
687 int is_encrypted = !is_decrypted;
688 struct sk_buff *skb_iter;
690 /* Skip if it is already decrypted */
691 if (ctx->sw.decrypted)
694 /* Check if all the data is decrypted already */
695 skb_walk_frags(skb, skb_iter) {
696 is_decrypted &= skb_iter->decrypted;
697 is_encrypted &= !skb_iter->decrypted;
700 ctx->sw.decrypted |= is_decrypted;
702 /* Return immedeatly if the record is either entirely plaintext or
703 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
706 return (is_encrypted || is_decrypted) ? 0 :
707 tls_device_reencrypt(sk, skb);
710 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
712 u16 nonce_size, tag_size, iv_size, rec_seq_size;
713 struct tls_record_info *start_marker_record;
714 struct tls_offload_context_tx *offload_ctx;
715 struct tls_crypto_info *crypto_info;
716 struct net_device *netdev;
725 if (ctx->priv_ctx_tx) {
730 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
731 if (!start_marker_record) {
736 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
739 goto free_marker_record;
742 crypto_info = &ctx->crypto_send.info;
743 switch (crypto_info->cipher_type) {
744 case TLS_CIPHER_AES_GCM_128:
745 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
746 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
747 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
748 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
749 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
751 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
755 goto free_offload_ctx;
758 ctx->tx.prepend_size = TLS_HEADER_SIZE + nonce_size;
759 ctx->tx.tag_size = tag_size;
760 ctx->tx.overhead_size = ctx->tx.prepend_size + ctx->tx.tag_size;
761 ctx->tx.iv_size = iv_size;
762 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
766 goto free_offload_ctx;
769 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
771 ctx->tx.rec_seq_size = rec_seq_size;
772 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
773 if (!ctx->tx.rec_seq) {
778 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
782 /* start at rec_seq - 1 to account for the start marker record */
783 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
784 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
786 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
787 start_marker_record->len = 0;
788 start_marker_record->num_frags = 0;
790 INIT_LIST_HEAD(&offload_ctx->records_list);
791 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
792 spin_lock_init(&offload_ctx->lock);
793 sg_init_table(offload_ctx->sg_tx_data,
794 ARRAY_SIZE(offload_ctx->sg_tx_data));
796 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
797 ctx->push_pending_record = tls_device_push_pending_record;
799 /* TLS offload is greatly simplified if we don't send
800 * SKBs where only part of the payload needs to be encrypted.
801 * So mark the last skb in the write queue as end of record.
803 skb = tcp_write_queue_tail(sk);
805 TCP_SKB_CB(skb)->eor = 1;
807 /* We support starting offload on multiple sockets
808 * concurrently, so we only need a read lock here.
809 * This lock must precede get_netdev_for_sock to prevent races between
810 * NETDEV_DOWN and setsockopt.
812 down_read(&device_offload_lock);
813 netdev = get_netdev_for_sock(sk);
815 pr_err_ratelimited("%s: netdev not found\n", __func__);
820 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
825 /* Avoid offloading if the device is down
826 * We don't want to offload new flows after
827 * the NETDEV_DOWN event
829 if (!(netdev->flags & IFF_UP)) {
834 ctx->priv_ctx_tx = offload_ctx;
835 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
836 &ctx->crypto_send.info,
837 tcp_sk(sk)->write_seq);
841 tls_device_attach(ctx, sk, netdev);
843 /* following this assignment tls_is_sk_tx_device_offloaded
844 * will return true and the context might be accessed
845 * by the netdev's xmit function.
847 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
849 up_read(&device_offload_lock);
855 up_read(&device_offload_lock);
856 clean_acked_data_disable(inet_csk(sk));
857 crypto_free_aead(offload_ctx->aead_send);
859 kfree(ctx->tx.rec_seq);
864 ctx->priv_ctx_tx = NULL;
866 kfree(start_marker_record);
871 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
873 struct tls_offload_context_rx *context;
874 struct net_device *netdev;
877 /* We support starting offload on multiple sockets
878 * concurrently, so we only need a read lock here.
879 * This lock must precede get_netdev_for_sock to prevent races between
880 * NETDEV_DOWN and setsockopt.
882 down_read(&device_offload_lock);
883 netdev = get_netdev_for_sock(sk);
885 pr_err_ratelimited("%s: netdev not found\n", __func__);
890 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
891 pr_err_ratelimited("%s: netdev %s with no TLS offload\n",
892 __func__, netdev->name);
897 /* Avoid offloading if the device is down
898 * We don't want to offload new flows after
899 * the NETDEV_DOWN event
901 if (!(netdev->flags & IFF_UP)) {
906 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
912 ctx->priv_ctx_rx = context;
913 rc = tls_set_sw_offload(sk, ctx, 0);
917 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
918 &ctx->crypto_recv.info,
919 tcp_sk(sk)->copied_seq);
921 pr_err_ratelimited("%s: The netdev has refused to offload this socket\n",
923 goto free_sw_resources;
926 tls_device_attach(ctx, sk, netdev);
930 up_read(&device_offload_lock);
931 tls_sw_free_resources_rx(sk);
932 down_read(&device_offload_lock);
934 ctx->priv_ctx_rx = NULL;
938 up_read(&device_offload_lock);
942 void tls_device_offload_cleanup_rx(struct sock *sk)
944 struct tls_context *tls_ctx = tls_get_ctx(sk);
945 struct net_device *netdev;
947 down_read(&device_offload_lock);
948 netdev = tls_ctx->netdev;
952 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
953 TLS_OFFLOAD_CTX_DIR_RX);
955 if (tls_ctx->tx_conf != TLS_HW) {
957 tls_ctx->netdev = NULL;
959 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
962 up_read(&device_offload_lock);
963 tls_sw_release_resources_rx(sk);
966 static int tls_device_down(struct net_device *netdev)
968 struct tls_context *ctx, *tmp;
972 /* Request a write lock to block new offload attempts */
973 down_write(&device_offload_lock);
975 spin_lock_irqsave(&tls_device_lock, flags);
976 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
977 if (ctx->netdev != netdev ||
978 !refcount_inc_not_zero(&ctx->refcount))
981 list_move(&ctx->list, &list);
983 spin_unlock_irqrestore(&tls_device_lock, flags);
985 list_for_each_entry_safe(ctx, tmp, &list, list) {
986 if (ctx->tx_conf == TLS_HW)
987 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
988 TLS_OFFLOAD_CTX_DIR_TX);
989 if (ctx->rx_conf == TLS_HW &&
990 !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
991 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
992 TLS_OFFLOAD_CTX_DIR_RX);
993 WRITE_ONCE(ctx->netdev, NULL);
994 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
995 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
996 usleep_range(10, 200);
998 list_del_init(&ctx->list);
1000 if (refcount_dec_and_test(&ctx->refcount))
1001 tls_device_free_ctx(ctx);
1004 up_write(&device_offload_lock);
1006 flush_work(&tls_device_gc_work);
1011 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1014 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1016 if (!dev->tlsdev_ops &&
1017 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1021 case NETDEV_REGISTER:
1022 case NETDEV_FEAT_CHANGE:
1023 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1024 !dev->tlsdev_ops->tls_dev_resync_rx)
1027 if (dev->tlsdev_ops &&
1028 dev->tlsdev_ops->tls_dev_add &&
1029 dev->tlsdev_ops->tls_dev_del)
1034 return tls_device_down(dev);
1039 static struct notifier_block tls_dev_notifier = {
1040 .notifier_call = tls_dev_event,
1043 void __init tls_device_init(void)
1045 register_netdevice_notifier(&tls_dev_notifier);
1048 void __exit tls_device_cleanup(void)
1050 unregister_netdevice_notifier(&tls_dev_notifier);
1051 flush_work(&tls_device_gc_work);