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>
43 /* device_offload_lock is used to synchronize tls_dev_add
44 * against NETDEV_DOWN notifications.
46 static DECLARE_RWSEM(device_offload_lock);
48 static void tls_device_gc_task(struct work_struct *work);
50 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
51 static LIST_HEAD(tls_device_gc_list);
52 static LIST_HEAD(tls_device_list);
53 static DEFINE_SPINLOCK(tls_device_lock);
55 static void tls_device_free_ctx(struct tls_context *ctx)
57 if (ctx->tx_conf == TLS_HW) {
58 kfree(tls_offload_ctx_tx(ctx));
59 kfree(ctx->tx.rec_seq);
63 if (ctx->rx_conf == TLS_HW)
64 kfree(tls_offload_ctx_rx(ctx));
66 tls_ctx_free(NULL, ctx);
69 static void tls_device_gc_task(struct work_struct *work)
71 struct tls_context *ctx, *tmp;
75 spin_lock_irqsave(&tls_device_lock, flags);
76 list_splice_init(&tls_device_gc_list, &gc_list);
77 spin_unlock_irqrestore(&tls_device_lock, flags);
79 list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
80 struct net_device *netdev = ctx->netdev;
82 if (netdev && ctx->tx_conf == TLS_HW) {
83 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
84 TLS_OFFLOAD_CTX_DIR_TX);
90 tls_device_free_ctx(ctx);
94 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
98 spin_lock_irqsave(&tls_device_lock, flags);
99 list_move_tail(&ctx->list, &tls_device_gc_list);
101 /* schedule_work inside the spinlock
102 * to make sure tls_device_down waits for that work.
104 schedule_work(&tls_device_gc_work);
106 spin_unlock_irqrestore(&tls_device_lock, flags);
109 /* We assume that the socket is already connected */
110 static struct net_device *get_netdev_for_sock(struct sock *sk)
112 struct dst_entry *dst = sk_dst_get(sk);
113 struct net_device *netdev = NULL;
125 static void destroy_record(struct tls_record_info *record)
129 for (i = 0; i < record->num_frags; i++)
130 __skb_frag_unref(&record->frags[i]);
134 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
136 struct tls_record_info *info, *temp;
138 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
139 list_del(&info->list);
140 destroy_record(info);
143 offload_ctx->retransmit_hint = NULL;
146 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
148 struct tls_context *tls_ctx = tls_get_ctx(sk);
149 struct tls_record_info *info, *temp;
150 struct tls_offload_context_tx *ctx;
151 u64 deleted_records = 0;
157 ctx = tls_offload_ctx_tx(tls_ctx);
159 spin_lock_irqsave(&ctx->lock, flags);
160 info = ctx->retransmit_hint;
161 if (info && !before(acked_seq, info->end_seq))
162 ctx->retransmit_hint = NULL;
164 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
165 if (before(acked_seq, info->end_seq))
167 list_del(&info->list);
169 destroy_record(info);
173 ctx->unacked_record_sn += deleted_records;
174 spin_unlock_irqrestore(&ctx->lock, flags);
177 /* At this point, there should be no references on this
178 * socket and no in-flight SKBs associated with this
179 * socket, so it is safe to free all the resources.
181 void tls_device_sk_destruct(struct sock *sk)
183 struct tls_context *tls_ctx = tls_get_ctx(sk);
184 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
186 tls_ctx->sk_destruct(sk);
188 if (tls_ctx->tx_conf == TLS_HW) {
189 if (ctx->open_record)
190 destroy_record(ctx->open_record);
191 delete_all_records(ctx);
192 crypto_free_aead(ctx->aead_send);
193 clean_acked_data_disable(inet_csk(sk));
196 if (refcount_dec_and_test(&tls_ctx->refcount))
197 tls_device_queue_ctx_destruction(tls_ctx);
199 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
201 void tls_device_free_resources_tx(struct sock *sk)
203 struct tls_context *tls_ctx = tls_get_ctx(sk);
205 tls_free_partial_record(sk, tls_ctx);
208 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
210 struct tls_context *tls_ctx = tls_get_ctx(sk);
212 trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
213 WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
215 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
217 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
220 struct net_device *netdev;
225 skb = tcp_write_queue_tail(sk);
227 TCP_SKB_CB(skb)->eor = 1;
229 rcd_sn = tls_ctx->tx.rec_seq;
231 trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
232 down_read(&device_offload_lock);
233 netdev = tls_ctx->netdev;
235 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
237 TLS_OFFLOAD_CTX_DIR_TX);
238 up_read(&device_offload_lock);
242 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
245 static void tls_append_frag(struct tls_record_info *record,
246 struct page_frag *pfrag,
251 frag = &record->frags[record->num_frags - 1];
252 if (skb_frag_page(frag) == pfrag->page &&
253 skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
254 skb_frag_size_add(frag, size);
257 __skb_frag_set_page(frag, pfrag->page);
258 skb_frag_off_set(frag, pfrag->offset);
259 skb_frag_size_set(frag, size);
261 get_page(pfrag->page);
264 pfrag->offset += size;
268 static int tls_push_record(struct sock *sk,
269 struct tls_context *ctx,
270 struct tls_offload_context_tx *offload_ctx,
271 struct tls_record_info *record,
274 struct tls_prot_info *prot = &ctx->prot_info;
275 struct tcp_sock *tp = tcp_sk(sk);
279 record->end_seq = tp->write_seq + record->len;
280 list_add_tail_rcu(&record->list, &offload_ctx->records_list);
281 offload_ctx->open_record = NULL;
283 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
284 tls_device_resync_tx(sk, ctx, tp->write_seq);
286 tls_advance_record_sn(sk, prot, &ctx->tx);
288 for (i = 0; i < record->num_frags; i++) {
289 frag = &record->frags[i];
290 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
291 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
292 skb_frag_size(frag), skb_frag_off(frag));
293 sk_mem_charge(sk, skb_frag_size(frag));
294 get_page(skb_frag_page(frag));
296 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
298 /* all ready, send */
299 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
302 static int tls_device_record_close(struct sock *sk,
303 struct tls_context *ctx,
304 struct tls_record_info *record,
305 struct page_frag *pfrag,
306 unsigned char record_type)
308 struct tls_prot_info *prot = &ctx->prot_info;
312 * device will fill in the tag, we just need to append a placeholder
313 * use socket memory to improve coalescing (re-using a single buffer
314 * increases frag count)
315 * if we can't allocate memory now, steal some back from data
317 if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
318 sk->sk_allocation))) {
320 tls_append_frag(record, pfrag, prot->tag_size);
322 ret = prot->tag_size;
323 if (record->len <= prot->overhead_size)
328 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
329 record->len - prot->overhead_size,
330 record_type, prot->version);
334 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
335 struct page_frag *pfrag,
338 struct tls_record_info *record;
341 record = kmalloc(sizeof(*record), GFP_KERNEL);
345 frag = &record->frags[0];
346 __skb_frag_set_page(frag, pfrag->page);
347 skb_frag_off_set(frag, pfrag->offset);
348 skb_frag_size_set(frag, prepend_size);
350 get_page(pfrag->page);
351 pfrag->offset += prepend_size;
353 record->num_frags = 1;
354 record->len = prepend_size;
355 offload_ctx->open_record = record;
359 static int tls_do_allocation(struct sock *sk,
360 struct tls_offload_context_tx *offload_ctx,
361 struct page_frag *pfrag,
366 if (!offload_ctx->open_record) {
367 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
368 sk->sk_allocation))) {
369 READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
370 sk_stream_moderate_sndbuf(sk);
374 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
378 if (pfrag->size > pfrag->offset)
382 if (!sk_page_frag_refill(sk, pfrag))
388 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
390 size_t pre_copy, nocache;
392 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
394 pre_copy = min(pre_copy, bytes);
395 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
401 nocache = round_down(bytes, SMP_CACHE_BYTES);
402 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
407 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
413 static int tls_push_data(struct sock *sk,
414 struct iov_iter *msg_iter,
415 size_t size, int flags,
416 unsigned char record_type)
418 struct tls_context *tls_ctx = tls_get_ctx(sk);
419 struct tls_prot_info *prot = &tls_ctx->prot_info;
420 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
421 struct tls_record_info *record = ctx->open_record;
422 int tls_push_record_flags;
423 struct page_frag *pfrag;
424 size_t orig_size = size;
425 u32 max_open_record_len;
432 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
435 if (unlikely(sk->sk_err))
438 flags |= MSG_SENDPAGE_DECRYPTED;
439 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
441 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
442 if (tls_is_partially_sent_record(tls_ctx)) {
443 rc = tls_push_partial_record(sk, tls_ctx, flags);
448 pfrag = sk_page_frag(sk);
450 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
451 * we need to leave room for an authentication tag.
453 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
456 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
458 rc = sk_stream_wait_memory(sk, &timeo);
462 record = ctx->open_record;
466 if (record_type != TLS_RECORD_TYPE_DATA) {
467 /* avoid sending partial
468 * record with type !=
472 destroy_record(record);
473 ctx->open_record = NULL;
474 } else if (record->len > prot->prepend_size) {
481 record = ctx->open_record;
482 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
483 copy = min_t(size_t, copy, (max_open_record_len - record->len));
485 rc = tls_device_copy_data(page_address(pfrag->page) +
486 pfrag->offset, copy, msg_iter);
489 tls_append_frag(record, pfrag, copy);
494 tls_push_record_flags = flags;
495 if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
503 if (done || record->len >= max_open_record_len ||
504 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
505 rc = tls_device_record_close(sk, tls_ctx, record,
512 destroy_record(record);
513 ctx->open_record = NULL;
518 rc = tls_push_record(sk,
522 tls_push_record_flags);
528 tls_ctx->pending_open_record_frags = more;
530 if (orig_size - size > 0)
531 rc = orig_size - size;
536 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
538 unsigned char record_type = TLS_RECORD_TYPE_DATA;
539 struct tls_context *tls_ctx = tls_get_ctx(sk);
542 mutex_lock(&tls_ctx->tx_lock);
545 if (unlikely(msg->msg_controllen)) {
546 rc = tls_proccess_cmsg(sk, msg, &record_type);
551 rc = tls_push_data(sk, &msg->msg_iter, size,
552 msg->msg_flags, record_type);
556 mutex_unlock(&tls_ctx->tx_lock);
560 int tls_device_sendpage(struct sock *sk, struct page *page,
561 int offset, size_t size, int flags)
563 struct tls_context *tls_ctx = tls_get_ctx(sk);
564 struct iov_iter msg_iter;
569 if (flags & MSG_SENDPAGE_NOTLAST)
572 mutex_lock(&tls_ctx->tx_lock);
575 if (flags & MSG_OOB) {
581 iov.iov_base = kaddr + offset;
583 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
584 rc = tls_push_data(sk, &msg_iter, size,
585 flags, TLS_RECORD_TYPE_DATA);
590 mutex_unlock(&tls_ctx->tx_lock);
594 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
595 u32 seq, u64 *p_record_sn)
597 u64 record_sn = context->hint_record_sn;
598 struct tls_record_info *info, *last;
600 info = context->retransmit_hint;
602 before(seq, info->end_seq - info->len)) {
603 /* if retransmit_hint is irrelevant start
604 * from the beggining of the list
606 info = list_first_entry_or_null(&context->records_list,
607 struct tls_record_info, list);
610 /* send the start_marker record if seq number is before the
611 * tls offload start marker sequence number. This record is
612 * required to handle TCP packets which are before TLS offload
614 * And if it's not start marker, look if this seq number
615 * belongs to the list.
617 if (likely(!tls_record_is_start_marker(info))) {
618 /* we have the first record, get the last record to see
619 * if this seq number belongs to the list.
621 last = list_last_entry(&context->records_list,
622 struct tls_record_info, list);
624 if (!between(seq, tls_record_start_seq(info),
628 record_sn = context->unacked_record_sn;
631 /* We just need the _rcu for the READ_ONCE() */
633 list_for_each_entry_from_rcu(info, &context->records_list, list) {
634 if (before(seq, info->end_seq)) {
635 if (!context->retransmit_hint ||
637 context->retransmit_hint->end_seq)) {
638 context->hint_record_sn = record_sn;
639 context->retransmit_hint = info;
641 *p_record_sn = record_sn;
642 goto exit_rcu_unlock;
652 EXPORT_SYMBOL(tls_get_record);
654 static int tls_device_push_pending_record(struct sock *sk, int flags)
656 struct iov_iter msg_iter;
658 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
659 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
662 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
664 if (tls_is_partially_sent_record(ctx)) {
665 gfp_t sk_allocation = sk->sk_allocation;
667 WARN_ON_ONCE(sk->sk_write_pending);
669 sk->sk_allocation = GFP_ATOMIC;
670 tls_push_partial_record(sk, ctx,
671 MSG_DONTWAIT | MSG_NOSIGNAL |
672 MSG_SENDPAGE_DECRYPTED);
673 sk->sk_allocation = sk_allocation;
677 static void tls_device_resync_rx(struct tls_context *tls_ctx,
678 struct sock *sk, u32 seq, u8 *rcd_sn)
680 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
681 struct net_device *netdev;
683 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
686 trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
687 netdev = READ_ONCE(tls_ctx->netdev);
689 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
690 TLS_OFFLOAD_CTX_DIR_RX);
691 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
692 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
696 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
697 s64 resync_req, u32 *seq)
699 u32 is_async = resync_req & RESYNC_REQ_ASYNC;
700 u32 req_seq = resync_req >> 32;
701 u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
704 /* asynchronous stage: log all headers seq such that
705 * req_seq <= seq <= end_seq, and wait for real resync request
707 if (between(*seq, req_seq, req_end) &&
708 resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
709 resync_async->log[resync_async->loglen++] = *seq;
714 /* synchronous stage: check against the logged entries and
715 * proceed to check the next entries if no match was found
717 while (resync_async->loglen) {
718 if (req_seq == resync_async->log[resync_async->loglen - 1] &&
719 atomic64_try_cmpxchg(&resync_async->req,
721 resync_async->loglen = 0;
725 resync_async->loglen--;
728 if (req_seq == *seq &&
729 atomic64_try_cmpxchg(&resync_async->req,
736 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
738 struct tls_context *tls_ctx = tls_get_ctx(sk);
739 struct tls_offload_context_rx *rx_ctx;
740 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
741 u32 sock_data, is_req_pending;
742 struct tls_prot_info *prot;
746 if (tls_ctx->rx_conf != TLS_HW)
749 prot = &tls_ctx->prot_info;
750 rx_ctx = tls_offload_ctx_rx(tls_ctx);
751 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
753 switch (rx_ctx->resync_type) {
754 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
755 resync_req = atomic64_read(&rx_ctx->resync_req);
756 req_seq = resync_req >> 32;
757 seq += TLS_HEADER_SIZE - 1;
758 is_req_pending = resync_req;
760 if (likely(!is_req_pending) || req_seq != seq ||
761 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
764 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
765 if (likely(!rx_ctx->resync_nh_do_now))
768 /* head of next rec is already in, note that the sock_inq will
769 * include the currently parsed message when called from parser
771 sock_data = tcp_inq(sk);
772 if (sock_data > rcd_len) {
773 trace_tls_device_rx_resync_nh_delay(sk, sock_data,
778 rx_ctx->resync_nh_do_now = 0;
780 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
782 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
783 resync_req = atomic64_read(&rx_ctx->resync_async->req);
784 is_req_pending = resync_req;
785 if (likely(!is_req_pending))
788 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
794 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
797 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
798 struct tls_offload_context_rx *ctx,
799 struct sock *sk, struct sk_buff *skb)
801 struct strp_msg *rxm;
803 /* device will request resyncs by itself based on stream scan */
804 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
806 /* already scheduled */
807 if (ctx->resync_nh_do_now)
809 /* seen decrypted fragments since last fully-failed record */
810 if (ctx->resync_nh_reset) {
811 ctx->resync_nh_reset = 0;
812 ctx->resync_nh.decrypted_failed = 1;
813 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
817 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
820 /* doing resync, bump the next target in case it fails */
821 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
822 ctx->resync_nh.decrypted_tgt *= 2;
824 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
828 /* head of next rec is already in, parser will sync for us */
829 if (tcp_inq(sk) > rxm->full_len) {
830 trace_tls_device_rx_resync_nh_schedule(sk);
831 ctx->resync_nh_do_now = 1;
833 struct tls_prot_info *prot = &tls_ctx->prot_info;
834 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
836 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
837 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
839 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
844 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
846 struct strp_msg *rxm = strp_msg(skb);
847 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
848 struct sk_buff *skb_iter, *unused;
849 struct scatterlist sg[1];
850 char *orig_buf, *buf;
852 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
853 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
858 nsg = skb_cow_data(skb, 0, &unused);
859 if (unlikely(nsg < 0)) {
864 sg_init_table(sg, 1);
865 sg_set_buf(&sg[0], buf,
866 rxm->full_len + TLS_HEADER_SIZE +
867 TLS_CIPHER_AES_GCM_128_IV_SIZE);
868 err = skb_copy_bits(skb, offset, buf,
869 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
873 /* We are interested only in the decrypted data not the auth */
874 err = decrypt_skb(sk, skb, sg);
880 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
882 if (skb_pagelen(skb) > offset) {
883 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
885 if (skb->decrypted) {
886 err = skb_store_bits(skb, offset, buf, copy);
895 pos = skb_pagelen(skb);
896 skb_walk_frags(skb, skb_iter) {
899 /* Practically all frags must belong to msg if reencrypt
900 * is needed with current strparser and coalescing logic,
901 * but strparser may "get optimized", so let's be safe.
903 if (pos + skb_iter->len <= offset)
905 if (pos >= data_len + rxm->offset)
908 frag_pos = offset - pos;
909 copy = min_t(int, skb_iter->len - frag_pos,
910 data_len + rxm->offset - offset);
912 if (skb_iter->decrypted) {
913 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
921 pos += skb_iter->len;
929 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
930 struct sk_buff *skb, struct strp_msg *rxm)
932 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
933 int is_decrypted = skb->decrypted;
934 int is_encrypted = !is_decrypted;
935 struct sk_buff *skb_iter;
937 /* Check if all the data is decrypted already */
938 skb_walk_frags(skb, skb_iter) {
939 is_decrypted &= skb_iter->decrypted;
940 is_encrypted &= !skb_iter->decrypted;
943 trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
944 tls_ctx->rx.rec_seq, rxm->full_len,
945 is_encrypted, is_decrypted);
947 ctx->sw.decrypted |= is_decrypted;
949 /* Return immediately if the record is either entirely plaintext or
950 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
954 ctx->resync_nh_reset = 1;
958 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
962 ctx->resync_nh_reset = 1;
963 return tls_device_reencrypt(sk, skb);
966 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
967 struct net_device *netdev)
969 if (sk->sk_destruct != tls_device_sk_destruct) {
970 refcount_set(&ctx->refcount, 1);
972 ctx->netdev = netdev;
973 spin_lock_irq(&tls_device_lock);
974 list_add_tail(&ctx->list, &tls_device_list);
975 spin_unlock_irq(&tls_device_lock);
977 ctx->sk_destruct = sk->sk_destruct;
978 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
982 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
984 u16 nonce_size, tag_size, iv_size, rec_seq_size;
985 struct tls_context *tls_ctx = tls_get_ctx(sk);
986 struct tls_prot_info *prot = &tls_ctx->prot_info;
987 struct tls_record_info *start_marker_record;
988 struct tls_offload_context_tx *offload_ctx;
989 struct tls_crypto_info *crypto_info;
990 struct net_device *netdev;
999 if (ctx->priv_ctx_tx)
1002 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1003 if (!start_marker_record)
1006 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1009 goto free_marker_record;
1012 crypto_info = &ctx->crypto_send.info;
1013 if (crypto_info->version != TLS_1_2_VERSION) {
1015 goto free_offload_ctx;
1018 switch (crypto_info->cipher_type) {
1019 case TLS_CIPHER_AES_GCM_128:
1020 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1021 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1022 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1023 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1024 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1026 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1030 goto free_offload_ctx;
1033 /* Sanity-check the rec_seq_size for stack allocations */
1034 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1036 goto free_offload_ctx;
1039 prot->version = crypto_info->version;
1040 prot->cipher_type = crypto_info->cipher_type;
1041 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1042 prot->tag_size = tag_size;
1043 prot->overhead_size = prot->prepend_size + prot->tag_size;
1044 prot->iv_size = iv_size;
1045 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1049 goto free_offload_ctx;
1052 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1054 prot->rec_seq_size = rec_seq_size;
1055 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1056 if (!ctx->tx.rec_seq) {
1061 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1065 /* start at rec_seq - 1 to account for the start marker record */
1066 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1067 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1069 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1070 start_marker_record->len = 0;
1071 start_marker_record->num_frags = 0;
1073 INIT_LIST_HEAD(&offload_ctx->records_list);
1074 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1075 spin_lock_init(&offload_ctx->lock);
1076 sg_init_table(offload_ctx->sg_tx_data,
1077 ARRAY_SIZE(offload_ctx->sg_tx_data));
1079 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1080 ctx->push_pending_record = tls_device_push_pending_record;
1082 /* TLS offload is greatly simplified if we don't send
1083 * SKBs where only part of the payload needs to be encrypted.
1084 * So mark the last skb in the write queue as end of record.
1086 skb = tcp_write_queue_tail(sk);
1088 TCP_SKB_CB(skb)->eor = 1;
1090 netdev = get_netdev_for_sock(sk);
1092 pr_err_ratelimited("%s: netdev not found\n", __func__);
1097 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1099 goto release_netdev;
1102 /* Avoid offloading if the device is down
1103 * We don't want to offload new flows after
1104 * the NETDEV_DOWN event
1106 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1107 * handler thus protecting from the device going down before
1108 * ctx was added to tls_device_list.
1110 down_read(&device_offload_lock);
1111 if (!(netdev->flags & IFF_UP)) {
1116 ctx->priv_ctx_tx = offload_ctx;
1117 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1118 &ctx->crypto_send.info,
1119 tcp_sk(sk)->write_seq);
1120 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1121 tcp_sk(sk)->write_seq, rec_seq, rc);
1125 tls_device_attach(ctx, sk, netdev);
1126 up_read(&device_offload_lock);
1128 /* following this assignment tls_is_sk_tx_device_offloaded
1129 * will return true and the context might be accessed
1130 * by the netdev's xmit function.
1132 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1138 up_read(&device_offload_lock);
1142 clean_acked_data_disable(inet_csk(sk));
1143 crypto_free_aead(offload_ctx->aead_send);
1145 kfree(ctx->tx.rec_seq);
1150 ctx->priv_ctx_tx = NULL;
1152 kfree(start_marker_record);
1156 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1158 struct tls12_crypto_info_aes_gcm_128 *info;
1159 struct tls_offload_context_rx *context;
1160 struct net_device *netdev;
1163 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1166 netdev = get_netdev_for_sock(sk);
1168 pr_err_ratelimited("%s: netdev not found\n", __func__);
1172 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1174 goto release_netdev;
1177 /* Avoid offloading if the device is down
1178 * We don't want to offload new flows after
1179 * the NETDEV_DOWN event
1181 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1182 * handler thus protecting from the device going down before
1183 * ctx was added to tls_device_list.
1185 down_read(&device_offload_lock);
1186 if (!(netdev->flags & IFF_UP)) {
1191 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1196 context->resync_nh_reset = 1;
1198 ctx->priv_ctx_rx = context;
1199 rc = tls_set_sw_offload(sk, ctx, 0);
1203 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1204 &ctx->crypto_recv.info,
1205 tcp_sk(sk)->copied_seq);
1206 info = (void *)&ctx->crypto_recv.info;
1207 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1208 tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1210 goto free_sw_resources;
1212 tls_device_attach(ctx, sk, netdev);
1213 up_read(&device_offload_lock);
1220 up_read(&device_offload_lock);
1221 tls_sw_free_resources_rx(sk);
1222 down_read(&device_offload_lock);
1224 ctx->priv_ctx_rx = NULL;
1226 up_read(&device_offload_lock);
1232 void tls_device_offload_cleanup_rx(struct sock *sk)
1234 struct tls_context *tls_ctx = tls_get_ctx(sk);
1235 struct net_device *netdev;
1237 down_read(&device_offload_lock);
1238 netdev = tls_ctx->netdev;
1242 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1243 TLS_OFFLOAD_CTX_DIR_RX);
1245 if (tls_ctx->tx_conf != TLS_HW) {
1247 tls_ctx->netdev = NULL;
1250 up_read(&device_offload_lock);
1251 tls_sw_release_resources_rx(sk);
1254 static int tls_device_down(struct net_device *netdev)
1256 struct tls_context *ctx, *tmp;
1257 unsigned long flags;
1260 /* Request a write lock to block new offload attempts */
1261 down_write(&device_offload_lock);
1263 spin_lock_irqsave(&tls_device_lock, flags);
1264 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1265 if (ctx->netdev != netdev ||
1266 !refcount_inc_not_zero(&ctx->refcount))
1269 list_move(&ctx->list, &list);
1271 spin_unlock_irqrestore(&tls_device_lock, flags);
1273 list_for_each_entry_safe(ctx, tmp, &list, list) {
1274 if (ctx->tx_conf == TLS_HW)
1275 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1276 TLS_OFFLOAD_CTX_DIR_TX);
1277 if (ctx->rx_conf == TLS_HW)
1278 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1279 TLS_OFFLOAD_CTX_DIR_RX);
1280 WRITE_ONCE(ctx->netdev, NULL);
1281 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1282 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1283 usleep_range(10, 200);
1285 list_del_init(&ctx->list);
1287 if (refcount_dec_and_test(&ctx->refcount))
1288 tls_device_free_ctx(ctx);
1291 up_write(&device_offload_lock);
1293 flush_work(&tls_device_gc_work);
1298 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1301 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1303 if (!dev->tlsdev_ops &&
1304 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1308 case NETDEV_REGISTER:
1309 case NETDEV_FEAT_CHANGE:
1310 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1311 !dev->tlsdev_ops->tls_dev_resync)
1314 if (dev->tlsdev_ops &&
1315 dev->tlsdev_ops->tls_dev_add &&
1316 dev->tlsdev_ops->tls_dev_del)
1321 return tls_device_down(dev);
1326 static struct notifier_block tls_dev_notifier = {
1327 .notifier_call = tls_dev_event,
1330 void __init tls_device_init(void)
1332 register_netdevice_notifier(&tls_dev_notifier);
1335 void __exit tls_device_cleanup(void)
1337 unregister_netdevice_notifier(&tls_dev_notifier);
1338 flush_work(&tls_device_gc_work);
1339 clean_acked_data_flush();