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>
44 /* device_offload_lock is used to synchronize tls_dev_add
45 * against NETDEV_DOWN notifications.
47 static DECLARE_RWSEM(device_offload_lock);
49 static struct workqueue_struct *destruct_wq __read_mostly;
51 static LIST_HEAD(tls_device_list);
52 static LIST_HEAD(tls_device_down_list);
53 static DEFINE_SPINLOCK(tls_device_lock);
55 static struct page *dummy_page;
57 static void tls_device_free_ctx(struct tls_context *ctx)
59 if (ctx->tx_conf == TLS_HW) {
60 kfree(tls_offload_ctx_tx(ctx));
64 if (ctx->rx_conf == TLS_HW)
65 kfree(tls_offload_ctx_rx(ctx));
67 tls_ctx_free(NULL, ctx);
70 static void tls_device_tx_del_task(struct work_struct *work)
72 struct tls_offload_context_tx *offload_ctx =
73 container_of(work, struct tls_offload_context_tx, destruct_work);
74 struct tls_context *ctx = offload_ctx->ctx;
75 struct net_device *netdev;
77 /* Safe, because this is the destroy flow, refcount is 0, so
78 * tls_device_down can't store this field in parallel.
80 netdev = rcu_dereference_protected(ctx->netdev,
81 !refcount_read(&ctx->refcount));
83 netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_TX);
86 tls_device_free_ctx(ctx);
89 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
91 struct net_device *netdev;
95 spin_lock_irqsave(&tls_device_lock, flags);
96 if (unlikely(!refcount_dec_and_test(&ctx->refcount))) {
97 spin_unlock_irqrestore(&tls_device_lock, flags);
101 list_del(&ctx->list); /* Remove from tls_device_list / tls_device_down_list */
103 /* Safe, because this is the destroy flow, refcount is 0, so
104 * tls_device_down can't store this field in parallel.
106 netdev = rcu_dereference_protected(ctx->netdev,
107 !refcount_read(&ctx->refcount));
109 async_cleanup = netdev && ctx->tx_conf == TLS_HW;
111 struct tls_offload_context_tx *offload_ctx = tls_offload_ctx_tx(ctx);
113 /* queue_work inside the spinlock
114 * to make sure tls_device_down waits for that work.
116 queue_work(destruct_wq, &offload_ctx->destruct_work);
118 spin_unlock_irqrestore(&tls_device_lock, flags);
121 tls_device_free_ctx(ctx);
124 /* We assume that the socket is already connected */
125 static struct net_device *get_netdev_for_sock(struct sock *sk)
127 struct dst_entry *dst = sk_dst_get(sk);
128 struct net_device *netdev = NULL;
131 netdev = netdev_sk_get_lowest_dev(dst->dev, sk);
140 static void destroy_record(struct tls_record_info *record)
144 for (i = 0; i < record->num_frags; i++)
145 __skb_frag_unref(&record->frags[i], false);
149 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
151 struct tls_record_info *info, *temp;
153 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
154 list_del(&info->list);
155 destroy_record(info);
158 offload_ctx->retransmit_hint = NULL;
161 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
163 struct tls_context *tls_ctx = tls_get_ctx(sk);
164 struct tls_record_info *info, *temp;
165 struct tls_offload_context_tx *ctx;
166 u64 deleted_records = 0;
172 ctx = tls_offload_ctx_tx(tls_ctx);
174 spin_lock_irqsave(&ctx->lock, flags);
175 info = ctx->retransmit_hint;
176 if (info && !before(acked_seq, info->end_seq))
177 ctx->retransmit_hint = NULL;
179 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
180 if (before(acked_seq, info->end_seq))
182 list_del(&info->list);
184 destroy_record(info);
188 ctx->unacked_record_sn += deleted_records;
189 spin_unlock_irqrestore(&ctx->lock, flags);
192 /* At this point, there should be no references on this
193 * socket and no in-flight SKBs associated with this
194 * socket, so it is safe to free all the resources.
196 void tls_device_sk_destruct(struct sock *sk)
198 struct tls_context *tls_ctx = tls_get_ctx(sk);
199 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
201 tls_ctx->sk_destruct(sk);
203 if (tls_ctx->tx_conf == TLS_HW) {
204 if (ctx->open_record)
205 destroy_record(ctx->open_record);
206 delete_all_records(ctx);
207 crypto_free_aead(ctx->aead_send);
208 clean_acked_data_disable(inet_csk(sk));
211 tls_device_queue_ctx_destruction(tls_ctx);
213 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
215 void tls_device_free_resources_tx(struct sock *sk)
217 struct tls_context *tls_ctx = tls_get_ctx(sk);
219 tls_free_partial_record(sk, tls_ctx);
222 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
224 struct tls_context *tls_ctx = tls_get_ctx(sk);
226 trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
227 WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
229 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
231 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
234 struct net_device *netdev;
239 skb = tcp_write_queue_tail(sk);
241 TCP_SKB_CB(skb)->eor = 1;
243 rcd_sn = tls_ctx->tx.rec_seq;
245 trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
246 down_read(&device_offload_lock);
247 netdev = rcu_dereference_protected(tls_ctx->netdev,
248 lockdep_is_held(&device_offload_lock));
250 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
252 TLS_OFFLOAD_CTX_DIR_TX);
253 up_read(&device_offload_lock);
257 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
260 static void tls_append_frag(struct tls_record_info *record,
261 struct page_frag *pfrag,
266 frag = &record->frags[record->num_frags - 1];
267 if (skb_frag_page(frag) == pfrag->page &&
268 skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
269 skb_frag_size_add(frag, size);
272 skb_frag_fill_page_desc(frag, pfrag->page, pfrag->offset,
275 get_page(pfrag->page);
278 pfrag->offset += size;
282 static int tls_push_record(struct sock *sk,
283 struct tls_context *ctx,
284 struct tls_offload_context_tx *offload_ctx,
285 struct tls_record_info *record,
288 struct tls_prot_info *prot = &ctx->prot_info;
289 struct tcp_sock *tp = tcp_sk(sk);
293 record->end_seq = tp->write_seq + record->len;
294 list_add_tail_rcu(&record->list, &offload_ctx->records_list);
295 offload_ctx->open_record = NULL;
297 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
298 tls_device_resync_tx(sk, ctx, tp->write_seq);
300 tls_advance_record_sn(sk, prot, &ctx->tx);
302 for (i = 0; i < record->num_frags; i++) {
303 frag = &record->frags[i];
304 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
305 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
306 skb_frag_size(frag), skb_frag_off(frag));
307 sk_mem_charge(sk, skb_frag_size(frag));
308 get_page(skb_frag_page(frag));
310 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
312 /* all ready, send */
313 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
316 static void tls_device_record_close(struct sock *sk,
317 struct tls_context *ctx,
318 struct tls_record_info *record,
319 struct page_frag *pfrag,
320 unsigned char record_type)
322 struct tls_prot_info *prot = &ctx->prot_info;
323 struct page_frag dummy_tag_frag;
326 * device will fill in the tag, we just need to append a placeholder
327 * use socket memory to improve coalescing (re-using a single buffer
328 * increases frag count)
329 * if we can't allocate memory now use the dummy page
331 if (unlikely(pfrag->size - pfrag->offset < prot->tag_size) &&
332 !skb_page_frag_refill(prot->tag_size, pfrag, sk->sk_allocation)) {
333 dummy_tag_frag.page = dummy_page;
334 dummy_tag_frag.offset = 0;
335 pfrag = &dummy_tag_frag;
337 tls_append_frag(record, pfrag, prot->tag_size);
340 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
341 record->len - prot->overhead_size,
345 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
346 struct page_frag *pfrag,
349 struct tls_record_info *record;
352 record = kmalloc(sizeof(*record), GFP_KERNEL);
356 frag = &record->frags[0];
357 skb_frag_fill_page_desc(frag, pfrag->page, pfrag->offset,
360 get_page(pfrag->page);
361 pfrag->offset += prepend_size;
363 record->num_frags = 1;
364 record->len = prepend_size;
365 offload_ctx->open_record = record;
369 static int tls_do_allocation(struct sock *sk,
370 struct tls_offload_context_tx *offload_ctx,
371 struct page_frag *pfrag,
376 if (!offload_ctx->open_record) {
377 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
378 sk->sk_allocation))) {
379 READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
380 sk_stream_moderate_sndbuf(sk);
384 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
388 if (pfrag->size > pfrag->offset)
392 if (!sk_page_frag_refill(sk, pfrag))
398 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
400 size_t pre_copy, nocache;
402 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
404 pre_copy = min(pre_copy, bytes);
405 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
411 nocache = round_down(bytes, SMP_CACHE_BYTES);
412 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
417 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
423 static int tls_push_data(struct sock *sk,
424 struct iov_iter *iter,
425 size_t size, int flags,
426 unsigned char record_type)
428 struct tls_context *tls_ctx = tls_get_ctx(sk);
429 struct tls_prot_info *prot = &tls_ctx->prot_info;
430 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
431 struct tls_record_info *record;
432 int tls_push_record_flags;
433 struct page_frag *pfrag;
434 size_t orig_size = size;
435 u32 max_open_record_len;
442 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
443 MSG_SPLICE_PAGES | MSG_EOR))
446 if ((flags & (MSG_MORE | MSG_EOR)) == (MSG_MORE | MSG_EOR))
449 if (unlikely(sk->sk_err))
452 flags |= MSG_SENDPAGE_DECRYPTED;
453 tls_push_record_flags = flags | MSG_MORE;
455 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
456 if (tls_is_partially_sent_record(tls_ctx)) {
457 rc = tls_push_partial_record(sk, tls_ctx, flags);
462 pfrag = sk_page_frag(sk);
464 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
465 * we need to leave room for an authentication tag.
467 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
470 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
472 rc = sk_stream_wait_memory(sk, &timeo);
476 record = ctx->open_record;
480 if (record_type != TLS_RECORD_TYPE_DATA) {
481 /* avoid sending partial
482 * record with type !=
486 destroy_record(record);
487 ctx->open_record = NULL;
488 } else if (record->len > prot->prepend_size) {
495 record = ctx->open_record;
497 copy = min_t(size_t, size, max_open_record_len - record->len);
498 if (copy && (flags & MSG_SPLICE_PAGES)) {
499 struct page_frag zc_pfrag;
500 struct page **pages = &zc_pfrag.page;
503 rc = iov_iter_extract_pages(iter, &pages,
512 if (WARN_ON_ONCE(!sendpage_ok(zc_pfrag.page))) {
513 iov_iter_revert(iter, copy);
518 zc_pfrag.offset = off;
519 zc_pfrag.size = copy;
520 tls_append_frag(record, &zc_pfrag, copy);
522 copy = min_t(size_t, copy, pfrag->size - pfrag->offset);
524 rc = tls_device_copy_data(page_address(pfrag->page) +
529 tls_append_frag(record, pfrag, copy);
535 tls_push_record_flags = flags;
536 if (flags & MSG_MORE) {
544 if (done || record->len >= max_open_record_len ||
545 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
546 tls_device_record_close(sk, tls_ctx, record,
549 rc = tls_push_record(sk,
553 tls_push_record_flags);
559 tls_ctx->pending_open_record_frags = more;
561 if (orig_size - size > 0)
562 rc = orig_size - size;
567 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
569 unsigned char record_type = TLS_RECORD_TYPE_DATA;
570 struct tls_context *tls_ctx = tls_get_ctx(sk);
573 if (!tls_ctx->zerocopy_sendfile)
574 msg->msg_flags &= ~MSG_SPLICE_PAGES;
576 mutex_lock(&tls_ctx->tx_lock);
579 if (unlikely(msg->msg_controllen)) {
580 rc = tls_process_cmsg(sk, msg, &record_type);
585 rc = tls_push_data(sk, &msg->msg_iter, size, msg->msg_flags,
590 mutex_unlock(&tls_ctx->tx_lock);
594 void tls_device_splice_eof(struct socket *sock)
596 struct sock *sk = sock->sk;
597 struct tls_context *tls_ctx = tls_get_ctx(sk);
598 struct iov_iter iter = {};
600 if (!tls_is_partially_sent_record(tls_ctx))
603 mutex_lock(&tls_ctx->tx_lock);
606 if (tls_is_partially_sent_record(tls_ctx)) {
607 iov_iter_bvec(&iter, ITER_SOURCE, NULL, 0, 0);
608 tls_push_data(sk, &iter, 0, 0, TLS_RECORD_TYPE_DATA);
612 mutex_unlock(&tls_ctx->tx_lock);
615 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
616 u32 seq, u64 *p_record_sn)
618 u64 record_sn = context->hint_record_sn;
619 struct tls_record_info *info, *last;
621 info = context->retransmit_hint;
623 before(seq, info->end_seq - info->len)) {
624 /* if retransmit_hint is irrelevant start
625 * from the beginning of the list
627 info = list_first_entry_or_null(&context->records_list,
628 struct tls_record_info, list);
631 /* send the start_marker record if seq number is before the
632 * tls offload start marker sequence number. This record is
633 * required to handle TCP packets which are before TLS offload
635 * And if it's not start marker, look if this seq number
636 * belongs to the list.
638 if (likely(!tls_record_is_start_marker(info))) {
639 /* we have the first record, get the last record to see
640 * if this seq number belongs to the list.
642 last = list_last_entry(&context->records_list,
643 struct tls_record_info, list);
645 if (!between(seq, tls_record_start_seq(info),
649 record_sn = context->unacked_record_sn;
652 /* We just need the _rcu for the READ_ONCE() */
654 list_for_each_entry_from_rcu(info, &context->records_list, list) {
655 if (before(seq, info->end_seq)) {
656 if (!context->retransmit_hint ||
658 context->retransmit_hint->end_seq)) {
659 context->hint_record_sn = record_sn;
660 context->retransmit_hint = info;
662 *p_record_sn = record_sn;
663 goto exit_rcu_unlock;
673 EXPORT_SYMBOL(tls_get_record);
675 static int tls_device_push_pending_record(struct sock *sk, int flags)
677 struct iov_iter iter;
679 iov_iter_kvec(&iter, ITER_SOURCE, NULL, 0, 0);
680 return tls_push_data(sk, &iter, 0, flags, TLS_RECORD_TYPE_DATA);
683 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
685 if (tls_is_partially_sent_record(ctx)) {
686 gfp_t sk_allocation = sk->sk_allocation;
688 WARN_ON_ONCE(sk->sk_write_pending);
690 sk->sk_allocation = GFP_ATOMIC;
691 tls_push_partial_record(sk, ctx,
692 MSG_DONTWAIT | MSG_NOSIGNAL |
693 MSG_SENDPAGE_DECRYPTED);
694 sk->sk_allocation = sk_allocation;
698 static void tls_device_resync_rx(struct tls_context *tls_ctx,
699 struct sock *sk, u32 seq, u8 *rcd_sn)
701 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
702 struct net_device *netdev;
704 trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
706 netdev = rcu_dereference(tls_ctx->netdev);
708 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
709 TLS_OFFLOAD_CTX_DIR_RX);
711 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
715 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
716 s64 resync_req, u32 *seq, u16 *rcd_delta)
718 u32 is_async = resync_req & RESYNC_REQ_ASYNC;
719 u32 req_seq = resync_req >> 32;
720 u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
726 /* shouldn't get to wraparound:
727 * too long in async stage, something bad happened
729 if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
732 /* asynchronous stage: log all headers seq such that
733 * req_seq <= seq <= end_seq, and wait for real resync request
735 if (before(*seq, req_seq))
737 if (!after(*seq, req_end) &&
738 resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
739 resync_async->log[resync_async->loglen++] = *seq;
741 resync_async->rcd_delta++;
746 /* synchronous stage: check against the logged entries and
747 * proceed to check the next entries if no match was found
749 for (i = 0; i < resync_async->loglen; i++)
750 if (req_seq == resync_async->log[i] &&
751 atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
752 *rcd_delta = resync_async->rcd_delta - i;
754 resync_async->loglen = 0;
755 resync_async->rcd_delta = 0;
759 resync_async->loglen = 0;
760 resync_async->rcd_delta = 0;
762 if (req_seq == *seq &&
763 atomic64_try_cmpxchg(&resync_async->req,
770 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
772 struct tls_context *tls_ctx = tls_get_ctx(sk);
773 struct tls_offload_context_rx *rx_ctx;
774 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
775 u32 sock_data, is_req_pending;
776 struct tls_prot_info *prot;
781 if (tls_ctx->rx_conf != TLS_HW)
783 if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
786 prot = &tls_ctx->prot_info;
787 rx_ctx = tls_offload_ctx_rx(tls_ctx);
788 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
790 switch (rx_ctx->resync_type) {
791 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
792 resync_req = atomic64_read(&rx_ctx->resync_req);
793 req_seq = resync_req >> 32;
794 seq += TLS_HEADER_SIZE - 1;
795 is_req_pending = resync_req;
797 if (likely(!is_req_pending) || req_seq != seq ||
798 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
801 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
802 if (likely(!rx_ctx->resync_nh_do_now))
805 /* head of next rec is already in, note that the sock_inq will
806 * include the currently parsed message when called from parser
808 sock_data = tcp_inq(sk);
809 if (sock_data > rcd_len) {
810 trace_tls_device_rx_resync_nh_delay(sk, sock_data,
815 rx_ctx->resync_nh_do_now = 0;
817 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
819 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
820 resync_req = atomic64_read(&rx_ctx->resync_async->req);
821 is_req_pending = resync_req;
822 if (likely(!is_req_pending))
825 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
826 resync_req, &seq, &rcd_delta))
828 tls_bigint_subtract(rcd_sn, rcd_delta);
832 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
835 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
836 struct tls_offload_context_rx *ctx,
837 struct sock *sk, struct sk_buff *skb)
839 struct strp_msg *rxm;
841 /* device will request resyncs by itself based on stream scan */
842 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
844 /* already scheduled */
845 if (ctx->resync_nh_do_now)
847 /* seen decrypted fragments since last fully-failed record */
848 if (ctx->resync_nh_reset) {
849 ctx->resync_nh_reset = 0;
850 ctx->resync_nh.decrypted_failed = 1;
851 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
855 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
858 /* doing resync, bump the next target in case it fails */
859 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
860 ctx->resync_nh.decrypted_tgt *= 2;
862 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
866 /* head of next rec is already in, parser will sync for us */
867 if (tcp_inq(sk) > rxm->full_len) {
868 trace_tls_device_rx_resync_nh_schedule(sk);
869 ctx->resync_nh_do_now = 1;
871 struct tls_prot_info *prot = &tls_ctx->prot_info;
872 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
874 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
875 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
877 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
883 tls_device_reencrypt(struct sock *sk, struct tls_context *tls_ctx)
885 struct tls_sw_context_rx *sw_ctx = tls_sw_ctx_rx(tls_ctx);
886 const struct tls_cipher_desc *cipher_desc;
887 int err, offset, copy, data_len, pos;
888 struct sk_buff *skb, *skb_iter;
889 struct scatterlist sg[1];
890 struct strp_msg *rxm;
891 char *orig_buf, *buf;
893 cipher_desc = get_cipher_desc(tls_ctx->crypto_recv.info.cipher_type);
894 DEBUG_NET_WARN_ON_ONCE(!cipher_desc || !cipher_desc->offloadable);
896 rxm = strp_msg(tls_strp_msg(sw_ctx));
897 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE + cipher_desc->iv,
903 err = tls_strp_msg_cow(sw_ctx);
907 skb = tls_strp_msg(sw_ctx);
909 offset = rxm->offset;
911 sg_init_table(sg, 1);
912 sg_set_buf(&sg[0], buf,
913 rxm->full_len + TLS_HEADER_SIZE + cipher_desc->iv);
914 err = skb_copy_bits(skb, offset, buf, TLS_HEADER_SIZE + cipher_desc->iv);
918 /* We are interested only in the decrypted data not the auth */
919 err = decrypt_skb(sk, sg);
925 data_len = rxm->full_len - cipher_desc->tag;
927 if (skb_pagelen(skb) > offset) {
928 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
930 if (skb->decrypted) {
931 err = skb_store_bits(skb, offset, buf, copy);
940 pos = skb_pagelen(skb);
941 skb_walk_frags(skb, skb_iter) {
944 /* Practically all frags must belong to msg if reencrypt
945 * is needed with current strparser and coalescing logic,
946 * but strparser may "get optimized", so let's be safe.
948 if (pos + skb_iter->len <= offset)
950 if (pos >= data_len + rxm->offset)
953 frag_pos = offset - pos;
954 copy = min_t(int, skb_iter->len - frag_pos,
955 data_len + rxm->offset - offset);
957 if (skb_iter->decrypted) {
958 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
966 pos += skb_iter->len;
974 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx)
976 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
977 struct tls_sw_context_rx *sw_ctx = tls_sw_ctx_rx(tls_ctx);
978 struct sk_buff *skb = tls_strp_msg(sw_ctx);
979 struct strp_msg *rxm = strp_msg(skb);
980 int is_decrypted, is_encrypted;
982 if (!tls_strp_msg_mixed_decrypted(sw_ctx)) {
983 is_decrypted = skb->decrypted;
984 is_encrypted = !is_decrypted;
990 trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
991 tls_ctx->rx.rec_seq, rxm->full_len,
992 is_encrypted, is_decrypted);
994 if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
995 if (likely(is_encrypted || is_decrypted))
998 /* After tls_device_down disables the offload, the next SKB will
999 * likely have initial fragments decrypted, and final ones not
1000 * decrypted. We need to reencrypt that single SKB.
1002 return tls_device_reencrypt(sk, tls_ctx);
1005 /* Return immediately if the record is either entirely plaintext or
1006 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
1010 ctx->resync_nh_reset = 1;
1011 return is_decrypted;
1014 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
1018 ctx->resync_nh_reset = 1;
1019 return tls_device_reencrypt(sk, tls_ctx);
1022 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
1023 struct net_device *netdev)
1025 if (sk->sk_destruct != tls_device_sk_destruct) {
1026 refcount_set(&ctx->refcount, 1);
1028 RCU_INIT_POINTER(ctx->netdev, netdev);
1029 spin_lock_irq(&tls_device_lock);
1030 list_add_tail(&ctx->list, &tls_device_list);
1031 spin_unlock_irq(&tls_device_lock);
1033 ctx->sk_destruct = sk->sk_destruct;
1034 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
1038 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
1040 struct tls_context *tls_ctx = tls_get_ctx(sk);
1041 struct tls_prot_info *prot = &tls_ctx->prot_info;
1042 const struct tls_cipher_desc *cipher_desc;
1043 struct tls_record_info *start_marker_record;
1044 struct tls_offload_context_tx *offload_ctx;
1045 struct tls_crypto_info *crypto_info;
1046 struct net_device *netdev;
1048 struct sk_buff *skb;
1055 if (ctx->priv_ctx_tx)
1058 netdev = get_netdev_for_sock(sk);
1060 pr_err_ratelimited("%s: netdev not found\n", __func__);
1064 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1066 goto release_netdev;
1069 crypto_info = &ctx->crypto_send.info;
1070 if (crypto_info->version != TLS_1_2_VERSION) {
1072 goto release_netdev;
1075 cipher_desc = get_cipher_desc(crypto_info->cipher_type);
1076 if (!cipher_desc || !cipher_desc->offloadable) {
1078 goto release_netdev;
1081 iv = crypto_info_iv(crypto_info, cipher_desc);
1082 rec_seq = crypto_info_rec_seq(crypto_info, cipher_desc);
1084 prot->version = crypto_info->version;
1085 prot->cipher_type = crypto_info->cipher_type;
1086 prot->prepend_size = TLS_HEADER_SIZE + cipher_desc->iv;
1087 prot->tag_size = cipher_desc->tag;
1088 prot->overhead_size = prot->prepend_size + prot->tag_size;
1089 prot->iv_size = cipher_desc->iv;
1090 prot->salt_size = cipher_desc->salt;
1091 ctx->tx.iv = kmalloc(cipher_desc->iv + cipher_desc->salt, GFP_KERNEL);
1094 goto release_netdev;
1097 memcpy(ctx->tx.iv + cipher_desc->salt, iv, cipher_desc->iv);
1099 prot->rec_seq_size = cipher_desc->rec_seq;
1100 memcpy(ctx->tx.rec_seq, rec_seq, cipher_desc->rec_seq);
1102 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1103 if (!start_marker_record) {
1108 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1111 goto free_marker_record;
1114 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1116 goto free_offload_ctx;
1118 /* start at rec_seq - 1 to account for the start marker record */
1119 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1120 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1122 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1123 start_marker_record->len = 0;
1124 start_marker_record->num_frags = 0;
1126 INIT_WORK(&offload_ctx->destruct_work, tls_device_tx_del_task);
1127 offload_ctx->ctx = ctx;
1129 INIT_LIST_HEAD(&offload_ctx->records_list);
1130 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1131 spin_lock_init(&offload_ctx->lock);
1132 sg_init_table(offload_ctx->sg_tx_data,
1133 ARRAY_SIZE(offload_ctx->sg_tx_data));
1135 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1136 ctx->push_pending_record = tls_device_push_pending_record;
1138 /* TLS offload is greatly simplified if we don't send
1139 * SKBs where only part of the payload needs to be encrypted.
1140 * So mark the last skb in the write queue as end of record.
1142 skb = tcp_write_queue_tail(sk);
1144 TCP_SKB_CB(skb)->eor = 1;
1146 /* Avoid offloading if the device is down
1147 * We don't want to offload new flows after
1148 * the NETDEV_DOWN event
1150 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1151 * handler thus protecting from the device going down before
1152 * ctx was added to tls_device_list.
1154 down_read(&device_offload_lock);
1155 if (!(netdev->flags & IFF_UP)) {
1160 ctx->priv_ctx_tx = offload_ctx;
1161 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1162 &ctx->crypto_send.info,
1163 tcp_sk(sk)->write_seq);
1164 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1165 tcp_sk(sk)->write_seq, rec_seq, rc);
1169 tls_device_attach(ctx, sk, netdev);
1170 up_read(&device_offload_lock);
1172 /* following this assignment tls_is_skb_tx_device_offloaded
1173 * will return true and the context might be accessed
1174 * by the netdev's xmit function.
1176 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1182 up_read(&device_offload_lock);
1183 clean_acked_data_disable(inet_csk(sk));
1184 crypto_free_aead(offload_ctx->aead_send);
1187 ctx->priv_ctx_tx = NULL;
1189 kfree(start_marker_record);
1197 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1199 struct tls12_crypto_info_aes_gcm_128 *info;
1200 struct tls_offload_context_rx *context;
1201 struct net_device *netdev;
1204 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1207 netdev = get_netdev_for_sock(sk);
1209 pr_err_ratelimited("%s: netdev not found\n", __func__);
1213 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1215 goto release_netdev;
1218 /* Avoid offloading if the device is down
1219 * We don't want to offload new flows after
1220 * the NETDEV_DOWN event
1222 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1223 * handler thus protecting from the device going down before
1224 * ctx was added to tls_device_list.
1226 down_read(&device_offload_lock);
1227 if (!(netdev->flags & IFF_UP)) {
1232 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1237 context->resync_nh_reset = 1;
1239 ctx->priv_ctx_rx = context;
1240 rc = tls_set_sw_offload(sk, ctx, 0);
1244 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1245 &ctx->crypto_recv.info,
1246 tcp_sk(sk)->copied_seq);
1247 info = (void *)&ctx->crypto_recv.info;
1248 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1249 tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1251 goto free_sw_resources;
1253 tls_device_attach(ctx, sk, netdev);
1254 up_read(&device_offload_lock);
1261 up_read(&device_offload_lock);
1262 tls_sw_free_resources_rx(sk);
1263 down_read(&device_offload_lock);
1265 ctx->priv_ctx_rx = NULL;
1267 up_read(&device_offload_lock);
1273 void tls_device_offload_cleanup_rx(struct sock *sk)
1275 struct tls_context *tls_ctx = tls_get_ctx(sk);
1276 struct net_device *netdev;
1278 down_read(&device_offload_lock);
1279 netdev = rcu_dereference_protected(tls_ctx->netdev,
1280 lockdep_is_held(&device_offload_lock));
1284 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1285 TLS_OFFLOAD_CTX_DIR_RX);
1287 if (tls_ctx->tx_conf != TLS_HW) {
1289 rcu_assign_pointer(tls_ctx->netdev, NULL);
1291 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1294 up_read(&device_offload_lock);
1295 tls_sw_release_resources_rx(sk);
1298 static int tls_device_down(struct net_device *netdev)
1300 struct tls_context *ctx, *tmp;
1301 unsigned long flags;
1304 /* Request a write lock to block new offload attempts */
1305 down_write(&device_offload_lock);
1307 spin_lock_irqsave(&tls_device_lock, flags);
1308 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1309 struct net_device *ctx_netdev =
1310 rcu_dereference_protected(ctx->netdev,
1311 lockdep_is_held(&device_offload_lock));
1313 if (ctx_netdev != netdev ||
1314 !refcount_inc_not_zero(&ctx->refcount))
1317 list_move(&ctx->list, &list);
1319 spin_unlock_irqrestore(&tls_device_lock, flags);
1321 list_for_each_entry_safe(ctx, tmp, &list, list) {
1322 /* Stop offloaded TX and switch to the fallback.
1323 * tls_is_skb_tx_device_offloaded will return false.
1325 WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1327 /* Stop the RX and TX resync.
1328 * tls_dev_resync must not be called after tls_dev_del.
1330 rcu_assign_pointer(ctx->netdev, NULL);
1332 /* Start skipping the RX resync logic completely. */
1333 set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1335 /* Sync with inflight packets. After this point:
1336 * TX: no non-encrypted packets will be passed to the driver.
1337 * RX: resync requests from the driver will be ignored.
1341 /* Release the offload context on the driver side. */
1342 if (ctx->tx_conf == TLS_HW)
1343 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1344 TLS_OFFLOAD_CTX_DIR_TX);
1345 if (ctx->rx_conf == TLS_HW &&
1346 !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1347 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1348 TLS_OFFLOAD_CTX_DIR_RX);
1352 /* Move the context to a separate list for two reasons:
1353 * 1. When the context is deallocated, list_del is called.
1354 * 2. It's no longer an offloaded context, so we don't want to
1355 * run offload-specific code on this context.
1357 spin_lock_irqsave(&tls_device_lock, flags);
1358 list_move_tail(&ctx->list, &tls_device_down_list);
1359 spin_unlock_irqrestore(&tls_device_lock, flags);
1361 /* Device contexts for RX and TX will be freed in on sk_destruct
1362 * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1363 * Now release the ref taken above.
1365 if (refcount_dec_and_test(&ctx->refcount)) {
1366 /* sk_destruct ran after tls_device_down took a ref, and
1367 * it returned early. Complete the destruction here.
1369 list_del(&ctx->list);
1370 tls_device_free_ctx(ctx);
1374 up_write(&device_offload_lock);
1376 flush_workqueue(destruct_wq);
1381 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1384 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1386 if (!dev->tlsdev_ops &&
1387 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1391 case NETDEV_REGISTER:
1392 case NETDEV_FEAT_CHANGE:
1393 if (netif_is_bond_master(dev))
1395 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1396 !dev->tlsdev_ops->tls_dev_resync)
1399 if (dev->tlsdev_ops &&
1400 dev->tlsdev_ops->tls_dev_add &&
1401 dev->tlsdev_ops->tls_dev_del)
1406 return tls_device_down(dev);
1411 static struct notifier_block tls_dev_notifier = {
1412 .notifier_call = tls_dev_event,
1415 int __init tls_device_init(void)
1419 dummy_page = alloc_page(GFP_KERNEL);
1423 destruct_wq = alloc_workqueue("ktls_device_destruct", 0, 0);
1426 goto err_free_dummy;
1429 err = register_netdevice_notifier(&tls_dev_notifier);
1431 goto err_destroy_wq;
1436 destroy_workqueue(destruct_wq);
1438 put_page(dummy_page);
1442 void __exit tls_device_cleanup(void)
1444 unregister_netdevice_notifier(&tls_dev_notifier);
1445 destroy_workqueue(destruct_wq);
1446 clean_acked_data_flush();
1447 put_page(dummy_page);