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));
64 tls_ctx_free(NULL, 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)
127 for (i = 0; i < record->num_frags; i++)
128 __skb_frag_unref(&record->frags[i]);
132 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
134 struct tls_record_info *info, *temp;
136 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
137 list_del(&info->list);
138 destroy_record(info);
141 offload_ctx->retransmit_hint = NULL;
144 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
146 struct tls_context *tls_ctx = tls_get_ctx(sk);
147 struct tls_record_info *info, *temp;
148 struct tls_offload_context_tx *ctx;
149 u64 deleted_records = 0;
155 ctx = tls_offload_ctx_tx(tls_ctx);
157 spin_lock_irqsave(&ctx->lock, flags);
158 info = ctx->retransmit_hint;
159 if (info && !before(acked_seq, info->end_seq))
160 ctx->retransmit_hint = NULL;
162 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
163 if (before(acked_seq, info->end_seq))
165 list_del(&info->list);
167 destroy_record(info);
171 ctx->unacked_record_sn += deleted_records;
172 spin_unlock_irqrestore(&ctx->lock, flags);
175 /* At this point, there should be no references on this
176 * socket and no in-flight SKBs associated with this
177 * socket, so it is safe to free all the resources.
179 static void tls_device_sk_destruct(struct sock *sk)
181 struct tls_context *tls_ctx = tls_get_ctx(sk);
182 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
184 tls_ctx->sk_destruct(sk);
186 if (tls_ctx->tx_conf == TLS_HW) {
187 if (ctx->open_record)
188 destroy_record(ctx->open_record);
189 delete_all_records(ctx);
190 crypto_free_aead(ctx->aead_send);
191 clean_acked_data_disable(inet_csk(sk));
194 if (refcount_dec_and_test(&tls_ctx->refcount))
195 tls_device_queue_ctx_destruction(tls_ctx);
198 void tls_device_free_resources_tx(struct sock *sk)
200 struct tls_context *tls_ctx = tls_get_ctx(sk);
202 tls_free_partial_record(sk, tls_ctx);
205 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
208 struct net_device *netdev;
213 skb = tcp_write_queue_tail(sk);
215 TCP_SKB_CB(skb)->eor = 1;
217 rcd_sn = tls_ctx->tx.rec_seq;
219 down_read(&device_offload_lock);
220 netdev = tls_ctx->netdev;
222 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
224 TLS_OFFLOAD_CTX_DIR_TX);
225 up_read(&device_offload_lock);
229 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
232 static void tls_append_frag(struct tls_record_info *record,
233 struct page_frag *pfrag,
238 frag = &record->frags[record->num_frags - 1];
239 if (skb_frag_page(frag) == pfrag->page &&
240 skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
241 skb_frag_size_add(frag, size);
244 __skb_frag_set_page(frag, pfrag->page);
245 skb_frag_off_set(frag, pfrag->offset);
246 skb_frag_size_set(frag, size);
248 get_page(pfrag->page);
251 pfrag->offset += size;
255 static int tls_push_record(struct sock *sk,
256 struct tls_context *ctx,
257 struct tls_offload_context_tx *offload_ctx,
258 struct tls_record_info *record,
261 struct tls_prot_info *prot = &ctx->prot_info;
262 struct tcp_sock *tp = tcp_sk(sk);
266 record->end_seq = tp->write_seq + record->len;
267 list_add_tail_rcu(&record->list, &offload_ctx->records_list);
268 offload_ctx->open_record = NULL;
270 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
271 tls_device_resync_tx(sk, ctx, tp->write_seq);
273 tls_advance_record_sn(sk, prot, &ctx->tx);
275 for (i = 0; i < record->num_frags; i++) {
276 frag = &record->frags[i];
277 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
278 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
279 skb_frag_size(frag), skb_frag_off(frag));
280 sk_mem_charge(sk, skb_frag_size(frag));
281 get_page(skb_frag_page(frag));
283 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
285 /* all ready, send */
286 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
289 static int tls_device_record_close(struct sock *sk,
290 struct tls_context *ctx,
291 struct tls_record_info *record,
292 struct page_frag *pfrag,
293 unsigned char record_type)
295 struct tls_prot_info *prot = &ctx->prot_info;
299 * device will fill in the tag, we just need to append a placeholder
300 * use socket memory to improve coalescing (re-using a single buffer
301 * increases frag count)
302 * if we can't allocate memory now, steal some back from data
304 if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
305 sk->sk_allocation))) {
307 tls_append_frag(record, pfrag, prot->tag_size);
309 ret = prot->tag_size;
310 if (record->len <= prot->overhead_size)
315 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
316 record->len - prot->overhead_size,
317 record_type, prot->version);
321 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
322 struct page_frag *pfrag,
325 struct tls_record_info *record;
328 record = kmalloc(sizeof(*record), GFP_KERNEL);
332 frag = &record->frags[0];
333 __skb_frag_set_page(frag, pfrag->page);
334 skb_frag_off_set(frag, pfrag->offset);
335 skb_frag_size_set(frag, prepend_size);
337 get_page(pfrag->page);
338 pfrag->offset += prepend_size;
340 record->num_frags = 1;
341 record->len = prepend_size;
342 offload_ctx->open_record = record;
346 static int tls_do_allocation(struct sock *sk,
347 struct tls_offload_context_tx *offload_ctx,
348 struct page_frag *pfrag,
353 if (!offload_ctx->open_record) {
354 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
355 sk->sk_allocation))) {
356 sk->sk_prot->enter_memory_pressure(sk);
357 sk_stream_moderate_sndbuf(sk);
361 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
365 if (pfrag->size > pfrag->offset)
369 if (!sk_page_frag_refill(sk, pfrag))
375 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
377 size_t pre_copy, nocache;
379 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
381 pre_copy = min(pre_copy, bytes);
382 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
388 nocache = round_down(bytes, SMP_CACHE_BYTES);
389 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
394 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
400 static int tls_push_data(struct sock *sk,
401 struct iov_iter *msg_iter,
402 size_t size, int flags,
403 unsigned char record_type)
405 struct tls_context *tls_ctx = tls_get_ctx(sk);
406 struct tls_prot_info *prot = &tls_ctx->prot_info;
407 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
408 struct tls_record_info *record = ctx->open_record;
409 int tls_push_record_flags;
410 struct page_frag *pfrag;
411 size_t orig_size = size;
412 u32 max_open_record_len;
419 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
425 flags |= MSG_SENDPAGE_DECRYPTED;
426 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
428 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
429 if (tls_is_partially_sent_record(tls_ctx)) {
430 rc = tls_push_partial_record(sk, tls_ctx, flags);
435 pfrag = sk_page_frag(sk);
437 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
438 * we need to leave room for an authentication tag.
440 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
443 rc = tls_do_allocation(sk, ctx, pfrag,
446 rc = sk_stream_wait_memory(sk, &timeo);
450 record = ctx->open_record;
454 if (record_type != TLS_RECORD_TYPE_DATA) {
455 /* avoid sending partial
456 * record with type !=
460 destroy_record(record);
461 ctx->open_record = NULL;
462 } else if (record->len > prot->prepend_size) {
469 record = ctx->open_record;
470 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
471 copy = min_t(size_t, copy, (max_open_record_len - record->len));
474 rc = tls_device_copy_data(page_address(pfrag->page) +
475 pfrag->offset, copy, msg_iter);
478 tls_append_frag(record, pfrag, copy);
484 tls_push_record_flags = flags;
485 if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
493 if (done || record->len >= max_open_record_len ||
494 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
495 rc = tls_device_record_close(sk, tls_ctx, record,
502 destroy_record(record);
503 ctx->open_record = NULL;
508 rc = tls_push_record(sk,
512 tls_push_record_flags);
518 tls_ctx->pending_open_record_frags = more;
520 if (orig_size - size > 0)
521 rc = orig_size - size;
526 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
528 unsigned char record_type = TLS_RECORD_TYPE_DATA;
529 struct tls_context *tls_ctx = tls_get_ctx(sk);
532 mutex_lock(&tls_ctx->tx_lock);
535 if (unlikely(msg->msg_controllen)) {
536 rc = tls_proccess_cmsg(sk, msg, &record_type);
541 rc = tls_push_data(sk, &msg->msg_iter, size,
542 msg->msg_flags, record_type);
546 mutex_unlock(&tls_ctx->tx_lock);
550 int tls_device_sendpage(struct sock *sk, struct page *page,
551 int offset, size_t size, int flags)
553 struct tls_context *tls_ctx = tls_get_ctx(sk);
554 struct iov_iter msg_iter;
559 if (flags & MSG_SENDPAGE_NOTLAST)
562 mutex_lock(&tls_ctx->tx_lock);
565 if (flags & MSG_OOB) {
571 iov.iov_base = kaddr + offset;
573 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
574 rc = tls_push_data(sk, &msg_iter, size,
575 flags, TLS_RECORD_TYPE_DATA);
580 mutex_unlock(&tls_ctx->tx_lock);
584 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
585 u32 seq, u64 *p_record_sn)
587 u64 record_sn = context->hint_record_sn;
588 struct tls_record_info *info, *last;
590 info = context->retransmit_hint;
592 before(seq, info->end_seq - info->len)) {
593 /* if retransmit_hint is irrelevant start
594 * from the beggining of the list
596 info = list_first_entry_or_null(&context->records_list,
597 struct tls_record_info, list);
600 /* send the start_marker record if seq number is before the
601 * tls offload start marker sequence number. This record is
602 * required to handle TCP packets which are before TLS offload
604 * And if it's not start marker, look if this seq number
605 * belongs to the list.
607 if (likely(!tls_record_is_start_marker(info))) {
608 /* we have the first record, get the last record to see
609 * if this seq number belongs to the list.
611 last = list_last_entry(&context->records_list,
612 struct tls_record_info, list);
614 if (!between(seq, tls_record_start_seq(info),
618 record_sn = context->unacked_record_sn;
621 /* We just need the _rcu for the READ_ONCE() */
623 list_for_each_entry_from_rcu(info, &context->records_list, list) {
624 if (before(seq, info->end_seq)) {
625 if (!context->retransmit_hint ||
627 context->retransmit_hint->end_seq)) {
628 context->hint_record_sn = record_sn;
629 context->retransmit_hint = info;
631 *p_record_sn = record_sn;
632 goto exit_rcu_unlock;
642 EXPORT_SYMBOL(tls_get_record);
644 static int tls_device_push_pending_record(struct sock *sk, int flags)
646 struct iov_iter msg_iter;
648 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
649 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
652 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
654 if (tls_is_partially_sent_record(ctx)) {
655 gfp_t sk_allocation = sk->sk_allocation;
657 WARN_ON_ONCE(sk->sk_write_pending);
659 sk->sk_allocation = GFP_ATOMIC;
660 tls_push_partial_record(sk, ctx,
661 MSG_DONTWAIT | MSG_NOSIGNAL |
662 MSG_SENDPAGE_DECRYPTED);
663 sk->sk_allocation = sk_allocation;
667 static void tls_device_resync_rx(struct tls_context *tls_ctx,
668 struct sock *sk, u32 seq, u8 *rcd_sn)
670 struct net_device *netdev;
672 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
674 netdev = READ_ONCE(tls_ctx->netdev);
676 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
677 TLS_OFFLOAD_CTX_DIR_RX);
678 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
681 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
683 struct tls_context *tls_ctx = tls_get_ctx(sk);
684 struct tls_offload_context_rx *rx_ctx;
685 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
686 struct tls_prot_info *prot;
691 if (tls_ctx->rx_conf != TLS_HW)
694 prot = &tls_ctx->prot_info;
695 rx_ctx = tls_offload_ctx_rx(tls_ctx);
696 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
698 switch (rx_ctx->resync_type) {
699 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
700 resync_req = atomic64_read(&rx_ctx->resync_req);
701 req_seq = resync_req >> 32;
702 seq += TLS_HEADER_SIZE - 1;
703 is_req_pending = resync_req;
705 if (likely(!is_req_pending) || req_seq != seq ||
706 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
709 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
710 if (likely(!rx_ctx->resync_nh_do_now))
713 /* head of next rec is already in, note that the sock_inq will
714 * include the currently parsed message when called from parser
716 if (tcp_inq(sk) > rcd_len)
719 rx_ctx->resync_nh_do_now = 0;
721 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
725 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
728 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
729 struct tls_offload_context_rx *ctx,
730 struct sock *sk, struct sk_buff *skb)
732 struct strp_msg *rxm;
734 /* device will request resyncs by itself based on stream scan */
735 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
737 /* already scheduled */
738 if (ctx->resync_nh_do_now)
740 /* seen decrypted fragments since last fully-failed record */
741 if (ctx->resync_nh_reset) {
742 ctx->resync_nh_reset = 0;
743 ctx->resync_nh.decrypted_failed = 1;
744 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
748 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
751 /* doing resync, bump the next target in case it fails */
752 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
753 ctx->resync_nh.decrypted_tgt *= 2;
755 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
759 /* head of next rec is already in, parser will sync for us */
760 if (tcp_inq(sk) > rxm->full_len) {
761 ctx->resync_nh_do_now = 1;
763 struct tls_prot_info *prot = &tls_ctx->prot_info;
764 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
766 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
767 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
769 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
774 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
776 struct strp_msg *rxm = strp_msg(skb);
777 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
778 struct sk_buff *skb_iter, *unused;
779 struct scatterlist sg[1];
780 char *orig_buf, *buf;
782 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
783 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
788 nsg = skb_cow_data(skb, 0, &unused);
789 if (unlikely(nsg < 0)) {
794 sg_init_table(sg, 1);
795 sg_set_buf(&sg[0], buf,
796 rxm->full_len + TLS_HEADER_SIZE +
797 TLS_CIPHER_AES_GCM_128_IV_SIZE);
798 err = skb_copy_bits(skb, offset, buf,
799 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
803 /* We are interested only in the decrypted data not the auth */
804 err = decrypt_skb(sk, skb, sg);
810 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
812 if (skb_pagelen(skb) > offset) {
813 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
815 if (skb->decrypted) {
816 err = skb_store_bits(skb, offset, buf, copy);
825 pos = skb_pagelen(skb);
826 skb_walk_frags(skb, skb_iter) {
829 /* Practically all frags must belong to msg if reencrypt
830 * is needed with current strparser and coalescing logic,
831 * but strparser may "get optimized", so let's be safe.
833 if (pos + skb_iter->len <= offset)
835 if (pos >= data_len + rxm->offset)
838 frag_pos = offset - pos;
839 copy = min_t(int, skb_iter->len - frag_pos,
840 data_len + rxm->offset - offset);
842 if (skb_iter->decrypted) {
843 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
851 pos += skb_iter->len;
859 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
861 struct tls_context *tls_ctx = tls_get_ctx(sk);
862 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
863 int is_decrypted = skb->decrypted;
864 int is_encrypted = !is_decrypted;
865 struct sk_buff *skb_iter;
867 /* Check if all the data is decrypted already */
868 skb_walk_frags(skb, skb_iter) {
869 is_decrypted &= skb_iter->decrypted;
870 is_encrypted &= !skb_iter->decrypted;
873 ctx->sw.decrypted |= is_decrypted;
875 /* Return immediately if the record is either entirely plaintext or
876 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
880 ctx->resync_nh_reset = 1;
884 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
888 ctx->resync_nh_reset = 1;
889 return tls_device_reencrypt(sk, skb);
892 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
893 struct net_device *netdev)
895 if (sk->sk_destruct != tls_device_sk_destruct) {
896 refcount_set(&ctx->refcount, 1);
898 ctx->netdev = netdev;
899 spin_lock_irq(&tls_device_lock);
900 list_add_tail(&ctx->list, &tls_device_list);
901 spin_unlock_irq(&tls_device_lock);
903 ctx->sk_destruct = sk->sk_destruct;
904 sk->sk_destruct = tls_device_sk_destruct;
908 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
910 u16 nonce_size, tag_size, iv_size, rec_seq_size;
911 struct tls_context *tls_ctx = tls_get_ctx(sk);
912 struct tls_prot_info *prot = &tls_ctx->prot_info;
913 struct tls_record_info *start_marker_record;
914 struct tls_offload_context_tx *offload_ctx;
915 struct tls_crypto_info *crypto_info;
916 struct net_device *netdev;
925 if (ctx->priv_ctx_tx)
928 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
929 if (!start_marker_record)
932 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
935 goto free_marker_record;
938 crypto_info = &ctx->crypto_send.info;
939 if (crypto_info->version != TLS_1_2_VERSION) {
941 goto free_offload_ctx;
944 switch (crypto_info->cipher_type) {
945 case TLS_CIPHER_AES_GCM_128:
946 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
947 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
948 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
949 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
950 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
952 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
956 goto free_offload_ctx;
959 /* Sanity-check the rec_seq_size for stack allocations */
960 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
962 goto free_offload_ctx;
965 prot->version = crypto_info->version;
966 prot->cipher_type = crypto_info->cipher_type;
967 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
968 prot->tag_size = tag_size;
969 prot->overhead_size = prot->prepend_size + prot->tag_size;
970 prot->iv_size = iv_size;
971 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
975 goto free_offload_ctx;
978 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
980 prot->rec_seq_size = rec_seq_size;
981 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
982 if (!ctx->tx.rec_seq) {
987 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
991 /* start at rec_seq - 1 to account for the start marker record */
992 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
993 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
995 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
996 start_marker_record->len = 0;
997 start_marker_record->num_frags = 0;
999 INIT_LIST_HEAD(&offload_ctx->records_list);
1000 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1001 spin_lock_init(&offload_ctx->lock);
1002 sg_init_table(offload_ctx->sg_tx_data,
1003 ARRAY_SIZE(offload_ctx->sg_tx_data));
1005 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1006 ctx->push_pending_record = tls_device_push_pending_record;
1008 /* TLS offload is greatly simplified if we don't send
1009 * SKBs where only part of the payload needs to be encrypted.
1010 * So mark the last skb in the write queue as end of record.
1012 skb = tcp_write_queue_tail(sk);
1014 TCP_SKB_CB(skb)->eor = 1;
1016 netdev = get_netdev_for_sock(sk);
1018 pr_err_ratelimited("%s: netdev not found\n", __func__);
1023 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1025 goto release_netdev;
1028 /* Avoid offloading if the device is down
1029 * We don't want to offload new flows after
1030 * the NETDEV_DOWN event
1032 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1033 * handler thus protecting from the device going down before
1034 * ctx was added to tls_device_list.
1036 down_read(&device_offload_lock);
1037 if (!(netdev->flags & IFF_UP)) {
1042 ctx->priv_ctx_tx = offload_ctx;
1043 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1044 &ctx->crypto_send.info,
1045 tcp_sk(sk)->write_seq);
1049 tls_device_attach(ctx, sk, netdev);
1050 up_read(&device_offload_lock);
1052 /* following this assignment tls_is_sk_tx_device_offloaded
1053 * will return true and the context might be accessed
1054 * by the netdev's xmit function.
1056 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1062 up_read(&device_offload_lock);
1066 clean_acked_data_disable(inet_csk(sk));
1067 crypto_free_aead(offload_ctx->aead_send);
1069 kfree(ctx->tx.rec_seq);
1074 ctx->priv_ctx_tx = NULL;
1076 kfree(start_marker_record);
1080 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1082 struct tls_offload_context_rx *context;
1083 struct net_device *netdev;
1086 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1089 netdev = get_netdev_for_sock(sk);
1091 pr_err_ratelimited("%s: netdev not found\n", __func__);
1095 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1097 goto release_netdev;
1100 /* Avoid offloading if the device is down
1101 * We don't want to offload new flows after
1102 * the NETDEV_DOWN event
1104 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1105 * handler thus protecting from the device going down before
1106 * ctx was added to tls_device_list.
1108 down_read(&device_offload_lock);
1109 if (!(netdev->flags & IFF_UP)) {
1114 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1119 context->resync_nh_reset = 1;
1121 ctx->priv_ctx_rx = context;
1122 rc = tls_set_sw_offload(sk, ctx, 0);
1126 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1127 &ctx->crypto_recv.info,
1128 tcp_sk(sk)->copied_seq);
1130 goto free_sw_resources;
1132 tls_device_attach(ctx, sk, netdev);
1133 up_read(&device_offload_lock);
1140 up_read(&device_offload_lock);
1141 tls_sw_free_resources_rx(sk);
1142 down_read(&device_offload_lock);
1144 ctx->priv_ctx_rx = NULL;
1146 up_read(&device_offload_lock);
1152 void tls_device_offload_cleanup_rx(struct sock *sk)
1154 struct tls_context *tls_ctx = tls_get_ctx(sk);
1155 struct net_device *netdev;
1157 down_read(&device_offload_lock);
1158 netdev = tls_ctx->netdev;
1162 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1163 TLS_OFFLOAD_CTX_DIR_RX);
1165 if (tls_ctx->tx_conf != TLS_HW) {
1167 tls_ctx->netdev = NULL;
1169 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1172 up_read(&device_offload_lock);
1173 tls_sw_release_resources_rx(sk);
1176 static int tls_device_down(struct net_device *netdev)
1178 struct tls_context *ctx, *tmp;
1179 unsigned long flags;
1182 /* Request a write lock to block new offload attempts */
1183 down_write(&device_offload_lock);
1185 spin_lock_irqsave(&tls_device_lock, flags);
1186 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1187 if (ctx->netdev != netdev ||
1188 !refcount_inc_not_zero(&ctx->refcount))
1191 list_move(&ctx->list, &list);
1193 spin_unlock_irqrestore(&tls_device_lock, flags);
1195 list_for_each_entry_safe(ctx, tmp, &list, list) {
1196 if (ctx->tx_conf == TLS_HW)
1197 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1198 TLS_OFFLOAD_CTX_DIR_TX);
1199 if (ctx->rx_conf == TLS_HW &&
1200 !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1201 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1202 TLS_OFFLOAD_CTX_DIR_RX);
1203 WRITE_ONCE(ctx->netdev, NULL);
1204 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1205 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1206 usleep_range(10, 200);
1208 list_del_init(&ctx->list);
1210 if (refcount_dec_and_test(&ctx->refcount))
1211 tls_device_free_ctx(ctx);
1214 up_write(&device_offload_lock);
1216 flush_work(&tls_device_gc_work);
1221 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1224 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1226 if (!dev->tlsdev_ops &&
1227 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1231 case NETDEV_REGISTER:
1232 case NETDEV_FEAT_CHANGE:
1233 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1234 !dev->tlsdev_ops->tls_dev_resync)
1237 if (dev->tlsdev_ops &&
1238 dev->tlsdev_ops->tls_dev_add &&
1239 dev->tlsdev_ops->tls_dev_del)
1244 return tls_device_down(dev);
1249 static struct notifier_block tls_dev_notifier = {
1250 .notifier_call = tls_dev_event,
1253 void __init tls_device_init(void)
1255 register_netdevice_notifier(&tls_dev_notifier);
1258 void __exit tls_device_cleanup(void)
1260 unregister_netdevice_notifier(&tls_dev_notifier);
1261 flush_work(&tls_device_gc_work);
1262 clean_acked_data_flush();