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 LIST_HEAD(tls_device_down_list);
54 static DEFINE_SPINLOCK(tls_device_lock);
56 static void tls_device_free_ctx(struct tls_context *ctx)
58 if (ctx->tx_conf == TLS_HW) {
59 kfree(tls_offload_ctx_tx(ctx));
60 kfree(ctx->tx.rec_seq);
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_gc_task(struct work_struct *work)
72 struct tls_context *ctx, *tmp;
76 spin_lock_irqsave(&tls_device_lock, flags);
77 list_splice_init(&tls_device_gc_list, &gc_list);
78 spin_unlock_irqrestore(&tls_device_lock, flags);
80 list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
81 struct net_device *netdev = ctx->netdev;
83 if (netdev && ctx->tx_conf == TLS_HW) {
84 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
85 TLS_OFFLOAD_CTX_DIR_TX);
91 tls_device_free_ctx(ctx);
95 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
99 spin_lock_irqsave(&tls_device_lock, flags);
100 list_move_tail(&ctx->list, &tls_device_gc_list);
102 /* schedule_work inside the spinlock
103 * to make sure tls_device_down waits for that work.
105 schedule_work(&tls_device_gc_work);
107 spin_unlock_irqrestore(&tls_device_lock, flags);
110 /* We assume that the socket is already connected */
111 static struct net_device *get_netdev_for_sock(struct sock *sk)
113 struct dst_entry *dst = sk_dst_get(sk);
114 struct net_device *netdev = NULL;
117 netdev = netdev_sk_get_lowest_dev(dst->dev, sk);
126 static void destroy_record(struct tls_record_info *record)
130 for (i = 0; i < record->num_frags; i++)
131 __skb_frag_unref(&record->frags[i], false);
135 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
137 struct tls_record_info *info, *temp;
139 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
140 list_del(&info->list);
141 destroy_record(info);
144 offload_ctx->retransmit_hint = NULL;
147 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
149 struct tls_context *tls_ctx = tls_get_ctx(sk);
150 struct tls_record_info *info, *temp;
151 struct tls_offload_context_tx *ctx;
152 u64 deleted_records = 0;
158 ctx = tls_offload_ctx_tx(tls_ctx);
160 spin_lock_irqsave(&ctx->lock, flags);
161 info = ctx->retransmit_hint;
162 if (info && !before(acked_seq, info->end_seq))
163 ctx->retransmit_hint = NULL;
165 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
166 if (before(acked_seq, info->end_seq))
168 list_del(&info->list);
170 destroy_record(info);
174 ctx->unacked_record_sn += deleted_records;
175 spin_unlock_irqrestore(&ctx->lock, flags);
178 /* At this point, there should be no references on this
179 * socket and no in-flight SKBs associated with this
180 * socket, so it is safe to free all the resources.
182 void tls_device_sk_destruct(struct sock *sk)
184 struct tls_context *tls_ctx = tls_get_ctx(sk);
185 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
187 tls_ctx->sk_destruct(sk);
189 if (tls_ctx->tx_conf == TLS_HW) {
190 if (ctx->open_record)
191 destroy_record(ctx->open_record);
192 delete_all_records(ctx);
193 crypto_free_aead(ctx->aead_send);
194 clean_acked_data_disable(inet_csk(sk));
197 if (refcount_dec_and_test(&tls_ctx->refcount))
198 tls_device_queue_ctx_destruction(tls_ctx);
200 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
202 void tls_device_free_resources_tx(struct sock *sk)
204 struct tls_context *tls_ctx = tls_get_ctx(sk);
206 tls_free_partial_record(sk, tls_ctx);
209 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
211 struct tls_context *tls_ctx = tls_get_ctx(sk);
213 trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
214 WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
216 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
218 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
221 struct net_device *netdev;
226 skb = tcp_write_queue_tail(sk);
228 TCP_SKB_CB(skb)->eor = 1;
230 rcd_sn = tls_ctx->tx.rec_seq;
232 trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
233 down_read(&device_offload_lock);
234 netdev = tls_ctx->netdev;
236 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
238 TLS_OFFLOAD_CTX_DIR_TX);
239 up_read(&device_offload_lock);
243 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
246 static void tls_append_frag(struct tls_record_info *record,
247 struct page_frag *pfrag,
252 frag = &record->frags[record->num_frags - 1];
253 if (skb_frag_page(frag) == pfrag->page &&
254 skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
255 skb_frag_size_add(frag, size);
258 __skb_frag_set_page(frag, pfrag->page);
259 skb_frag_off_set(frag, pfrag->offset);
260 skb_frag_size_set(frag, size);
262 get_page(pfrag->page);
265 pfrag->offset += size;
269 static int tls_push_record(struct sock *sk,
270 struct tls_context *ctx,
271 struct tls_offload_context_tx *offload_ctx,
272 struct tls_record_info *record,
275 struct tls_prot_info *prot = &ctx->prot_info;
276 struct tcp_sock *tp = tcp_sk(sk);
280 record->end_seq = tp->write_seq + record->len;
281 list_add_tail_rcu(&record->list, &offload_ctx->records_list);
282 offload_ctx->open_record = NULL;
284 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
285 tls_device_resync_tx(sk, ctx, tp->write_seq);
287 tls_advance_record_sn(sk, prot, &ctx->tx);
289 for (i = 0; i < record->num_frags; i++) {
290 frag = &record->frags[i];
291 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
292 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
293 skb_frag_size(frag), skb_frag_off(frag));
294 sk_mem_charge(sk, skb_frag_size(frag));
295 get_page(skb_frag_page(frag));
297 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
299 /* all ready, send */
300 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
303 static int tls_device_record_close(struct sock *sk,
304 struct tls_context *ctx,
305 struct tls_record_info *record,
306 struct page_frag *pfrag,
307 unsigned char record_type)
309 struct tls_prot_info *prot = &ctx->prot_info;
313 * device will fill in the tag, we just need to append a placeholder
314 * use socket memory to improve coalescing (re-using a single buffer
315 * increases frag count)
316 * if we can't allocate memory now, steal some back from data
318 if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
319 sk->sk_allocation))) {
321 tls_append_frag(record, pfrag, prot->tag_size);
323 ret = prot->tag_size;
324 if (record->len <= prot->overhead_size)
329 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
330 record->len - prot->overhead_size,
335 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
336 struct page_frag *pfrag,
339 struct tls_record_info *record;
342 record = kmalloc(sizeof(*record), GFP_KERNEL);
346 frag = &record->frags[0];
347 __skb_frag_set_page(frag, pfrag->page);
348 skb_frag_off_set(frag, pfrag->offset);
349 skb_frag_size_set(frag, prepend_size);
351 get_page(pfrag->page);
352 pfrag->offset += prepend_size;
354 record->num_frags = 1;
355 record->len = prepend_size;
356 offload_ctx->open_record = record;
360 static int tls_do_allocation(struct sock *sk,
361 struct tls_offload_context_tx *offload_ctx,
362 struct page_frag *pfrag,
367 if (!offload_ctx->open_record) {
368 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
369 sk->sk_allocation))) {
370 READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
371 sk_stream_moderate_sndbuf(sk);
375 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
379 if (pfrag->size > pfrag->offset)
383 if (!sk_page_frag_refill(sk, pfrag))
389 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
391 size_t pre_copy, nocache;
393 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
395 pre_copy = min(pre_copy, bytes);
396 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
402 nocache = round_down(bytes, SMP_CACHE_BYTES);
403 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
408 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
414 static int tls_push_data(struct sock *sk,
415 struct iov_iter *msg_iter,
416 size_t size, int flags,
417 unsigned char record_type)
419 struct tls_context *tls_ctx = tls_get_ctx(sk);
420 struct tls_prot_info *prot = &tls_ctx->prot_info;
421 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
422 struct tls_record_info *record;
423 int tls_push_record_flags;
424 struct page_frag *pfrag;
425 size_t orig_size = size;
426 u32 max_open_record_len;
433 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
436 if (unlikely(sk->sk_err))
439 flags |= MSG_SENDPAGE_DECRYPTED;
440 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
442 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
443 if (tls_is_partially_sent_record(tls_ctx)) {
444 rc = tls_push_partial_record(sk, tls_ctx, flags);
449 pfrag = sk_page_frag(sk);
451 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
452 * we need to leave room for an authentication tag.
454 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
457 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
459 rc = sk_stream_wait_memory(sk, &timeo);
463 record = ctx->open_record;
467 if (record_type != TLS_RECORD_TYPE_DATA) {
468 /* avoid sending partial
469 * record with type !=
473 destroy_record(record);
474 ctx->open_record = NULL;
475 } else if (record->len > prot->prepend_size) {
482 record = ctx->open_record;
483 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
484 copy = min_t(size_t, copy, (max_open_record_len - record->len));
487 rc = tls_device_copy_data(page_address(pfrag->page) +
488 pfrag->offset, copy, msg_iter);
491 tls_append_frag(record, pfrag, copy);
497 tls_push_record_flags = flags;
498 if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
506 if (done || record->len >= max_open_record_len ||
507 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
508 rc = tls_device_record_close(sk, tls_ctx, record,
515 destroy_record(record);
516 ctx->open_record = NULL;
521 rc = tls_push_record(sk,
525 tls_push_record_flags);
531 tls_ctx->pending_open_record_frags = more;
533 if (orig_size - size > 0)
534 rc = orig_size - size;
539 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
541 unsigned char record_type = TLS_RECORD_TYPE_DATA;
542 struct tls_context *tls_ctx = tls_get_ctx(sk);
545 mutex_lock(&tls_ctx->tx_lock);
548 if (unlikely(msg->msg_controllen)) {
549 rc = tls_proccess_cmsg(sk, msg, &record_type);
554 rc = tls_push_data(sk, &msg->msg_iter, size,
555 msg->msg_flags, record_type);
559 mutex_unlock(&tls_ctx->tx_lock);
563 int tls_device_sendpage(struct sock *sk, struct page *page,
564 int offset, size_t size, int flags)
566 struct tls_context *tls_ctx = tls_get_ctx(sk);
567 struct iov_iter msg_iter;
572 if (flags & MSG_SENDPAGE_NOTLAST)
575 mutex_lock(&tls_ctx->tx_lock);
578 if (flags & MSG_OOB) {
584 iov.iov_base = kaddr + offset;
586 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
587 rc = tls_push_data(sk, &msg_iter, size,
588 flags, TLS_RECORD_TYPE_DATA);
593 mutex_unlock(&tls_ctx->tx_lock);
597 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
598 u32 seq, u64 *p_record_sn)
600 u64 record_sn = context->hint_record_sn;
601 struct tls_record_info *info, *last;
603 info = context->retransmit_hint;
605 before(seq, info->end_seq - info->len)) {
606 /* if retransmit_hint is irrelevant start
607 * from the beginning of the list
609 info = list_first_entry_or_null(&context->records_list,
610 struct tls_record_info, list);
613 /* send the start_marker record if seq number is before the
614 * tls offload start marker sequence number. This record is
615 * required to handle TCP packets which are before TLS offload
617 * And if it's not start marker, look if this seq number
618 * belongs to the list.
620 if (likely(!tls_record_is_start_marker(info))) {
621 /* we have the first record, get the last record to see
622 * if this seq number belongs to the list.
624 last = list_last_entry(&context->records_list,
625 struct tls_record_info, list);
627 if (!between(seq, tls_record_start_seq(info),
631 record_sn = context->unacked_record_sn;
634 /* We just need the _rcu for the READ_ONCE() */
636 list_for_each_entry_from_rcu(info, &context->records_list, list) {
637 if (before(seq, info->end_seq)) {
638 if (!context->retransmit_hint ||
640 context->retransmit_hint->end_seq)) {
641 context->hint_record_sn = record_sn;
642 context->retransmit_hint = info;
644 *p_record_sn = record_sn;
645 goto exit_rcu_unlock;
655 EXPORT_SYMBOL(tls_get_record);
657 static int tls_device_push_pending_record(struct sock *sk, int flags)
659 struct iov_iter msg_iter;
661 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
662 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
665 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
667 if (tls_is_partially_sent_record(ctx)) {
668 gfp_t sk_allocation = sk->sk_allocation;
670 WARN_ON_ONCE(sk->sk_write_pending);
672 sk->sk_allocation = GFP_ATOMIC;
673 tls_push_partial_record(sk, ctx,
674 MSG_DONTWAIT | MSG_NOSIGNAL |
675 MSG_SENDPAGE_DECRYPTED);
676 sk->sk_allocation = sk_allocation;
680 static void tls_device_resync_rx(struct tls_context *tls_ctx,
681 struct sock *sk, u32 seq, u8 *rcd_sn)
683 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
684 struct net_device *netdev;
686 trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
688 netdev = READ_ONCE(tls_ctx->netdev);
690 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
691 TLS_OFFLOAD_CTX_DIR_RX);
693 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
697 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
698 s64 resync_req, u32 *seq, u16 *rcd_delta)
700 u32 is_async = resync_req & RESYNC_REQ_ASYNC;
701 u32 req_seq = resync_req >> 32;
702 u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
708 /* shouldn't get to wraparound:
709 * too long in async stage, something bad happened
711 if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
714 /* asynchronous stage: log all headers seq such that
715 * req_seq <= seq <= end_seq, and wait for real resync request
717 if (before(*seq, req_seq))
719 if (!after(*seq, req_end) &&
720 resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
721 resync_async->log[resync_async->loglen++] = *seq;
723 resync_async->rcd_delta++;
728 /* synchronous stage: check against the logged entries and
729 * proceed to check the next entries if no match was found
731 for (i = 0; i < resync_async->loglen; i++)
732 if (req_seq == resync_async->log[i] &&
733 atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
734 *rcd_delta = resync_async->rcd_delta - i;
736 resync_async->loglen = 0;
737 resync_async->rcd_delta = 0;
741 resync_async->loglen = 0;
742 resync_async->rcd_delta = 0;
744 if (req_seq == *seq &&
745 atomic64_try_cmpxchg(&resync_async->req,
752 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
754 struct tls_context *tls_ctx = tls_get_ctx(sk);
755 struct tls_offload_context_rx *rx_ctx;
756 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
757 u32 sock_data, is_req_pending;
758 struct tls_prot_info *prot;
763 if (tls_ctx->rx_conf != TLS_HW)
765 if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
768 prot = &tls_ctx->prot_info;
769 rx_ctx = tls_offload_ctx_rx(tls_ctx);
770 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
772 switch (rx_ctx->resync_type) {
773 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
774 resync_req = atomic64_read(&rx_ctx->resync_req);
775 req_seq = resync_req >> 32;
776 seq += TLS_HEADER_SIZE - 1;
777 is_req_pending = resync_req;
779 if (likely(!is_req_pending) || req_seq != seq ||
780 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
783 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
784 if (likely(!rx_ctx->resync_nh_do_now))
787 /* head of next rec is already in, note that the sock_inq will
788 * include the currently parsed message when called from parser
790 sock_data = tcp_inq(sk);
791 if (sock_data > rcd_len) {
792 trace_tls_device_rx_resync_nh_delay(sk, sock_data,
797 rx_ctx->resync_nh_do_now = 0;
799 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
801 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
802 resync_req = atomic64_read(&rx_ctx->resync_async->req);
803 is_req_pending = resync_req;
804 if (likely(!is_req_pending))
807 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
808 resync_req, &seq, &rcd_delta))
810 tls_bigint_subtract(rcd_sn, rcd_delta);
814 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
817 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
818 struct tls_offload_context_rx *ctx,
819 struct sock *sk, struct sk_buff *skb)
821 struct strp_msg *rxm;
823 /* device will request resyncs by itself based on stream scan */
824 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
826 /* already scheduled */
827 if (ctx->resync_nh_do_now)
829 /* seen decrypted fragments since last fully-failed record */
830 if (ctx->resync_nh_reset) {
831 ctx->resync_nh_reset = 0;
832 ctx->resync_nh.decrypted_failed = 1;
833 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
837 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
840 /* doing resync, bump the next target in case it fails */
841 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
842 ctx->resync_nh.decrypted_tgt *= 2;
844 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
848 /* head of next rec is already in, parser will sync for us */
849 if (tcp_inq(sk) > rxm->full_len) {
850 trace_tls_device_rx_resync_nh_schedule(sk);
851 ctx->resync_nh_do_now = 1;
853 struct tls_prot_info *prot = &tls_ctx->prot_info;
854 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
856 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
857 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
859 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
864 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
866 struct strp_msg *rxm = strp_msg(skb);
867 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
868 struct sk_buff *skb_iter, *unused;
869 struct scatterlist sg[1];
870 char *orig_buf, *buf;
872 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
873 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
878 nsg = skb_cow_data(skb, 0, &unused);
879 if (unlikely(nsg < 0)) {
884 sg_init_table(sg, 1);
885 sg_set_buf(&sg[0], buf,
886 rxm->full_len + TLS_HEADER_SIZE +
887 TLS_CIPHER_AES_GCM_128_IV_SIZE);
888 err = skb_copy_bits(skb, offset, buf,
889 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
893 /* We are interested only in the decrypted data not the auth */
894 err = decrypt_skb(sk, skb, sg);
900 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
902 if (skb_pagelen(skb) > offset) {
903 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
905 if (skb->decrypted) {
906 err = skb_store_bits(skb, offset, buf, copy);
915 pos = skb_pagelen(skb);
916 skb_walk_frags(skb, skb_iter) {
919 /* Practically all frags must belong to msg if reencrypt
920 * is needed with current strparser and coalescing logic,
921 * but strparser may "get optimized", so let's be safe.
923 if (pos + skb_iter->len <= offset)
925 if (pos >= data_len + rxm->offset)
928 frag_pos = offset - pos;
929 copy = min_t(int, skb_iter->len - frag_pos,
930 data_len + rxm->offset - offset);
932 if (skb_iter->decrypted) {
933 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
941 pos += skb_iter->len;
949 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
950 struct sk_buff *skb, struct strp_msg *rxm)
952 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
953 int is_decrypted = skb->decrypted;
954 int is_encrypted = !is_decrypted;
955 struct sk_buff *skb_iter;
957 /* Check if all the data is decrypted already */
958 skb_walk_frags(skb, skb_iter) {
959 is_decrypted &= skb_iter->decrypted;
960 is_encrypted &= !skb_iter->decrypted;
963 trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
964 tls_ctx->rx.rec_seq, rxm->full_len,
965 is_encrypted, is_decrypted);
967 ctx->sw.decrypted |= is_decrypted;
969 if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
970 if (likely(is_encrypted || is_decrypted))
973 /* After tls_device_down disables the offload, the next SKB will
974 * likely have initial fragments decrypted, and final ones not
975 * decrypted. We need to reencrypt that single SKB.
977 return tls_device_reencrypt(sk, skb);
980 /* Return immediately if the record is either entirely plaintext or
981 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
985 ctx->resync_nh_reset = 1;
989 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
993 ctx->resync_nh_reset = 1;
994 return tls_device_reencrypt(sk, skb);
997 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
998 struct net_device *netdev)
1000 if (sk->sk_destruct != tls_device_sk_destruct) {
1001 refcount_set(&ctx->refcount, 1);
1003 ctx->netdev = netdev;
1004 spin_lock_irq(&tls_device_lock);
1005 list_add_tail(&ctx->list, &tls_device_list);
1006 spin_unlock_irq(&tls_device_lock);
1008 ctx->sk_destruct = sk->sk_destruct;
1009 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
1013 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
1015 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
1016 struct tls_context *tls_ctx = tls_get_ctx(sk);
1017 struct tls_prot_info *prot = &tls_ctx->prot_info;
1018 struct tls_record_info *start_marker_record;
1019 struct tls_offload_context_tx *offload_ctx;
1020 struct tls_crypto_info *crypto_info;
1021 struct net_device *netdev;
1023 struct sk_buff *skb;
1030 if (ctx->priv_ctx_tx)
1033 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1034 if (!start_marker_record)
1037 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1040 goto free_marker_record;
1043 crypto_info = &ctx->crypto_send.info;
1044 if (crypto_info->version != TLS_1_2_VERSION) {
1046 goto free_offload_ctx;
1049 switch (crypto_info->cipher_type) {
1050 case TLS_CIPHER_AES_GCM_128:
1051 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1052 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1053 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1054 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1055 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1056 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
1058 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1062 goto free_offload_ctx;
1065 /* Sanity-check the rec_seq_size for stack allocations */
1066 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1068 goto free_offload_ctx;
1071 prot->version = crypto_info->version;
1072 prot->cipher_type = crypto_info->cipher_type;
1073 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1074 prot->tag_size = tag_size;
1075 prot->overhead_size = prot->prepend_size + prot->tag_size;
1076 prot->iv_size = iv_size;
1077 prot->salt_size = salt_size;
1078 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1082 goto free_offload_ctx;
1085 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1087 prot->rec_seq_size = rec_seq_size;
1088 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1089 if (!ctx->tx.rec_seq) {
1094 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1098 /* start at rec_seq - 1 to account for the start marker record */
1099 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1100 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1102 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1103 start_marker_record->len = 0;
1104 start_marker_record->num_frags = 0;
1106 INIT_LIST_HEAD(&offload_ctx->records_list);
1107 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1108 spin_lock_init(&offload_ctx->lock);
1109 sg_init_table(offload_ctx->sg_tx_data,
1110 ARRAY_SIZE(offload_ctx->sg_tx_data));
1112 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1113 ctx->push_pending_record = tls_device_push_pending_record;
1115 /* TLS offload is greatly simplified if we don't send
1116 * SKBs where only part of the payload needs to be encrypted.
1117 * So mark the last skb in the write queue as end of record.
1119 skb = tcp_write_queue_tail(sk);
1121 TCP_SKB_CB(skb)->eor = 1;
1123 netdev = get_netdev_for_sock(sk);
1125 pr_err_ratelimited("%s: netdev not found\n", __func__);
1130 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1132 goto release_netdev;
1135 /* Avoid offloading if the device is down
1136 * We don't want to offload new flows after
1137 * the NETDEV_DOWN event
1139 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1140 * handler thus protecting from the device going down before
1141 * ctx was added to tls_device_list.
1143 down_read(&device_offload_lock);
1144 if (!(netdev->flags & IFF_UP)) {
1149 ctx->priv_ctx_tx = offload_ctx;
1150 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1151 &ctx->crypto_send.info,
1152 tcp_sk(sk)->write_seq);
1153 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1154 tcp_sk(sk)->write_seq, rec_seq, rc);
1158 tls_device_attach(ctx, sk, netdev);
1159 up_read(&device_offload_lock);
1161 /* following this assignment tls_is_sk_tx_device_offloaded
1162 * will return true and the context might be accessed
1163 * by the netdev's xmit function.
1165 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1171 up_read(&device_offload_lock);
1175 clean_acked_data_disable(inet_csk(sk));
1176 crypto_free_aead(offload_ctx->aead_send);
1178 kfree(ctx->tx.rec_seq);
1183 ctx->priv_ctx_tx = NULL;
1185 kfree(start_marker_record);
1189 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1191 struct tls12_crypto_info_aes_gcm_128 *info;
1192 struct tls_offload_context_rx *context;
1193 struct net_device *netdev;
1196 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1199 netdev = get_netdev_for_sock(sk);
1201 pr_err_ratelimited("%s: netdev not found\n", __func__);
1205 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1207 goto release_netdev;
1210 /* Avoid offloading if the device is down
1211 * We don't want to offload new flows after
1212 * the NETDEV_DOWN event
1214 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1215 * handler thus protecting from the device going down before
1216 * ctx was added to tls_device_list.
1218 down_read(&device_offload_lock);
1219 if (!(netdev->flags & IFF_UP)) {
1224 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1229 context->resync_nh_reset = 1;
1231 ctx->priv_ctx_rx = context;
1232 rc = tls_set_sw_offload(sk, ctx, 0);
1236 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1237 &ctx->crypto_recv.info,
1238 tcp_sk(sk)->copied_seq);
1239 info = (void *)&ctx->crypto_recv.info;
1240 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1241 tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1243 goto free_sw_resources;
1245 tls_device_attach(ctx, sk, netdev);
1246 up_read(&device_offload_lock);
1253 up_read(&device_offload_lock);
1254 tls_sw_free_resources_rx(sk);
1255 down_read(&device_offload_lock);
1257 ctx->priv_ctx_rx = NULL;
1259 up_read(&device_offload_lock);
1265 void tls_device_offload_cleanup_rx(struct sock *sk)
1267 struct tls_context *tls_ctx = tls_get_ctx(sk);
1268 struct net_device *netdev;
1270 down_read(&device_offload_lock);
1271 netdev = tls_ctx->netdev;
1275 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1276 TLS_OFFLOAD_CTX_DIR_RX);
1278 if (tls_ctx->tx_conf != TLS_HW) {
1280 tls_ctx->netdev = NULL;
1282 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1285 up_read(&device_offload_lock);
1286 tls_sw_release_resources_rx(sk);
1289 static int tls_device_down(struct net_device *netdev)
1291 struct tls_context *ctx, *tmp;
1292 unsigned long flags;
1295 /* Request a write lock to block new offload attempts */
1296 down_write(&device_offload_lock);
1298 spin_lock_irqsave(&tls_device_lock, flags);
1299 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1300 if (ctx->netdev != netdev ||
1301 !refcount_inc_not_zero(&ctx->refcount))
1304 list_move(&ctx->list, &list);
1306 spin_unlock_irqrestore(&tls_device_lock, flags);
1308 list_for_each_entry_safe(ctx, tmp, &list, list) {
1309 /* Stop offloaded TX and switch to the fallback.
1310 * tls_is_sk_tx_device_offloaded will return false.
1312 WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1314 /* Stop the RX and TX resync.
1315 * tls_dev_resync must not be called after tls_dev_del.
1317 WRITE_ONCE(ctx->netdev, NULL);
1319 /* Start skipping the RX resync logic completely. */
1320 set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1322 /* Sync with inflight packets. After this point:
1323 * TX: no non-encrypted packets will be passed to the driver.
1324 * RX: resync requests from the driver will be ignored.
1328 /* Release the offload context on the driver side. */
1329 if (ctx->tx_conf == TLS_HW)
1330 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1331 TLS_OFFLOAD_CTX_DIR_TX);
1332 if (ctx->rx_conf == TLS_HW &&
1333 !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1334 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1335 TLS_OFFLOAD_CTX_DIR_RX);
1339 /* Move the context to a separate list for two reasons:
1340 * 1. When the context is deallocated, list_del is called.
1341 * 2. It's no longer an offloaded context, so we don't want to
1342 * run offload-specific code on this context.
1344 spin_lock_irqsave(&tls_device_lock, flags);
1345 list_move_tail(&ctx->list, &tls_device_down_list);
1346 spin_unlock_irqrestore(&tls_device_lock, flags);
1348 /* Device contexts for RX and TX will be freed in on sk_destruct
1349 * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1353 up_write(&device_offload_lock);
1355 flush_work(&tls_device_gc_work);
1360 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1363 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1365 if (!dev->tlsdev_ops &&
1366 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1370 case NETDEV_REGISTER:
1371 case NETDEV_FEAT_CHANGE:
1372 if (netif_is_bond_master(dev))
1374 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1375 !dev->tlsdev_ops->tls_dev_resync)
1378 if (dev->tlsdev_ops &&
1379 dev->tlsdev_ops->tls_dev_add &&
1380 dev->tlsdev_ops->tls_dev_del)
1385 return tls_device_down(dev);
1390 static struct notifier_block tls_dev_notifier = {
1391 .notifier_call = tls_dev_event,
1394 void __init tls_device_init(void)
1396 register_netdevice_notifier(&tls_dev_notifier);
1399 void __exit tls_device_cleanup(void)
1401 unregister_netdevice_notifier(&tls_dev_notifier);
1402 flush_work(&tls_device_gc_work);
1403 clean_acked_data_flush();