Loading...
Note: File does not exist in v3.15.
1/* Copyright (c) 2018, Mellanox Technologies All rights reserved.
2 *
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:
8 *
9 * Redistribution and use in source and binary forms, with or
10 * without modification, are permitted provided that the following
11 * conditions are met:
12 *
13 * - Redistributions of source code must retain the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer.
16 *
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.
21 *
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
29 * SOFTWARE.
30 */
31
32#include <crypto/aead.h>
33#include <linux/highmem.h>
34#include <linux/module.h>
35#include <linux/netdevice.h>
36#include <net/dst.h>
37#include <net/inet_connection_sock.h>
38#include <net/tcp.h>
39#include <net/tls.h>
40
41#include "trace.h"
42
43/* device_offload_lock is used to synchronize tls_dev_add
44 * against NETDEV_DOWN notifications.
45 */
46static DECLARE_RWSEM(device_offload_lock);
47
48static void tls_device_gc_task(struct work_struct *work);
49
50static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
51static LIST_HEAD(tls_device_gc_list);
52static LIST_HEAD(tls_device_list);
53static DEFINE_SPINLOCK(tls_device_lock);
54
55static void tls_device_free_ctx(struct tls_context *ctx)
56{
57 if (ctx->tx_conf == TLS_HW) {
58 kfree(tls_offload_ctx_tx(ctx));
59 kfree(ctx->tx.rec_seq);
60 kfree(ctx->tx.iv);
61 }
62
63 if (ctx->rx_conf == TLS_HW)
64 kfree(tls_offload_ctx_rx(ctx));
65
66 tls_ctx_free(NULL, ctx);
67}
68
69static void tls_device_gc_task(struct work_struct *work)
70{
71 struct tls_context *ctx, *tmp;
72 unsigned long flags;
73 LIST_HEAD(gc_list);
74
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);
78
79 list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
80 struct net_device *netdev = ctx->netdev;
81
82 if (netdev && ctx->tx_conf == TLS_HW) {
83 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
84 TLS_OFFLOAD_CTX_DIR_TX);
85 dev_put(netdev);
86 ctx->netdev = NULL;
87 }
88
89 list_del(&ctx->list);
90 tls_device_free_ctx(ctx);
91 }
92}
93
94static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
95{
96 unsigned long flags;
97
98 spin_lock_irqsave(&tls_device_lock, flags);
99 list_move_tail(&ctx->list, &tls_device_gc_list);
100
101 /* schedule_work inside the spinlock
102 * to make sure tls_device_down waits for that work.
103 */
104 schedule_work(&tls_device_gc_work);
105
106 spin_unlock_irqrestore(&tls_device_lock, flags);
107}
108
109/* We assume that the socket is already connected */
110static struct net_device *get_netdev_for_sock(struct sock *sk)
111{
112 struct dst_entry *dst = sk_dst_get(sk);
113 struct net_device *netdev = NULL;
114
115 if (likely(dst)) {
116 netdev = dst->dev;
117 dev_hold(netdev);
118 }
119
120 dst_release(dst);
121
122 return netdev;
123}
124
125static void destroy_record(struct tls_record_info *record)
126{
127 int i;
128
129 for (i = 0; i < record->num_frags; i++)
130 __skb_frag_unref(&record->frags[i]);
131 kfree(record);
132}
133
134static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
135{
136 struct tls_record_info *info, *temp;
137
138 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
139 list_del(&info->list);
140 destroy_record(info);
141 }
142
143 offload_ctx->retransmit_hint = NULL;
144}
145
146static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
147{
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;
152 unsigned long flags;
153
154 if (!tls_ctx)
155 return;
156
157 ctx = tls_offload_ctx_tx(tls_ctx);
158
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;
163
164 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
165 if (before(acked_seq, info->end_seq))
166 break;
167 list_del(&info->list);
168
169 destroy_record(info);
170 deleted_records++;
171 }
172
173 ctx->unacked_record_sn += deleted_records;
174 spin_unlock_irqrestore(&ctx->lock, flags);
175}
176
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.
180 */
181void tls_device_sk_destruct(struct sock *sk)
182{
183 struct tls_context *tls_ctx = tls_get_ctx(sk);
184 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
185
186 tls_ctx->sk_destruct(sk);
187
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));
194 }
195
196 if (refcount_dec_and_test(&tls_ctx->refcount))
197 tls_device_queue_ctx_destruction(tls_ctx);
198}
199EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
200
201void tls_device_free_resources_tx(struct sock *sk)
202{
203 struct tls_context *tls_ctx = tls_get_ctx(sk);
204
205 tls_free_partial_record(sk, tls_ctx);
206}
207
208void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
209{
210 struct tls_context *tls_ctx = tls_get_ctx(sk);
211
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));
214}
215EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
216
217static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
218 u32 seq)
219{
220 struct net_device *netdev;
221 struct sk_buff *skb;
222 int err = 0;
223 u8 *rcd_sn;
224
225 skb = tcp_write_queue_tail(sk);
226 if (skb)
227 TCP_SKB_CB(skb)->eor = 1;
228
229 rcd_sn = tls_ctx->tx.rec_seq;
230
231 trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
232 down_read(&device_offload_lock);
233 netdev = tls_ctx->netdev;
234 if (netdev)
235 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
236 rcd_sn,
237 TLS_OFFLOAD_CTX_DIR_TX);
238 up_read(&device_offload_lock);
239 if (err)
240 return;
241
242 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
243}
244
245static void tls_append_frag(struct tls_record_info *record,
246 struct page_frag *pfrag,
247 int size)
248{
249 skb_frag_t *frag;
250
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);
255 } else {
256 ++frag;
257 __skb_frag_set_page(frag, pfrag->page);
258 skb_frag_off_set(frag, pfrag->offset);
259 skb_frag_size_set(frag, size);
260 ++record->num_frags;
261 get_page(pfrag->page);
262 }
263
264 pfrag->offset += size;
265 record->len += size;
266}
267
268static 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,
272 int flags)
273{
274 struct tls_prot_info *prot = &ctx->prot_info;
275 struct tcp_sock *tp = tcp_sk(sk);
276 skb_frag_t *frag;
277 int i;
278
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;
282
283 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
284 tls_device_resync_tx(sk, ctx, tp->write_seq);
285
286 tls_advance_record_sn(sk, prot, &ctx->tx);
287
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));
295 }
296 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
297
298 /* all ready, send */
299 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
300}
301
302static 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)
307{
308 struct tls_prot_info *prot = &ctx->prot_info;
309 int ret;
310
311 /* append tag
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
316 */
317 if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
318 sk->sk_allocation))) {
319 ret = 0;
320 tls_append_frag(record, pfrag, prot->tag_size);
321 } else {
322 ret = prot->tag_size;
323 if (record->len <= prot->overhead_size)
324 return -ENOMEM;
325 }
326
327 /* fill prepend */
328 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
329 record->len - prot->overhead_size,
330 record_type, prot->version);
331 return ret;
332}
333
334static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
335 struct page_frag *pfrag,
336 size_t prepend_size)
337{
338 struct tls_record_info *record;
339 skb_frag_t *frag;
340
341 record = kmalloc(sizeof(*record), GFP_KERNEL);
342 if (!record)
343 return -ENOMEM;
344
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);
349
350 get_page(pfrag->page);
351 pfrag->offset += prepend_size;
352
353 record->num_frags = 1;
354 record->len = prepend_size;
355 offload_ctx->open_record = record;
356 return 0;
357}
358
359static int tls_do_allocation(struct sock *sk,
360 struct tls_offload_context_tx *offload_ctx,
361 struct page_frag *pfrag,
362 size_t prepend_size)
363{
364 int ret;
365
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);
371 return -ENOMEM;
372 }
373
374 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
375 if (ret)
376 return ret;
377
378 if (pfrag->size > pfrag->offset)
379 return 0;
380 }
381
382 if (!sk_page_frag_refill(sk, pfrag))
383 return -ENOMEM;
384
385 return 0;
386}
387
388static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
389{
390 size_t pre_copy, nocache;
391
392 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
393 if (pre_copy) {
394 pre_copy = min(pre_copy, bytes);
395 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
396 return -EFAULT;
397 bytes -= pre_copy;
398 addr += pre_copy;
399 }
400
401 nocache = round_down(bytes, SMP_CACHE_BYTES);
402 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
403 return -EFAULT;
404 bytes -= nocache;
405 addr += nocache;
406
407 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
408 return -EFAULT;
409
410 return 0;
411}
412
413static int tls_push_data(struct sock *sk,
414 struct iov_iter *msg_iter,
415 size_t size, int flags,
416 unsigned char record_type)
417{
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 int more = flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE);
422 struct tls_record_info *record = ctx->open_record;
423 int tls_push_record_flags;
424 struct page_frag *pfrag;
425 size_t orig_size = size;
426 u32 max_open_record_len;
427 int copy, rc = 0;
428 bool done = false;
429 long timeo;
430
431 if (flags &
432 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
433 return -EOPNOTSUPP;
434
435 if (unlikely(sk->sk_err))
436 return -sk->sk_err;
437
438 flags |= MSG_SENDPAGE_DECRYPTED;
439 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
440
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);
444 if (rc < 0)
445 return rc;
446 }
447
448 pfrag = sk_page_frag(sk);
449
450 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
451 * we need to leave room for an authentication tag.
452 */
453 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
454 prot->prepend_size;
455 do {
456 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
457 if (unlikely(rc)) {
458 rc = sk_stream_wait_memory(sk, &timeo);
459 if (!rc)
460 continue;
461
462 record = ctx->open_record;
463 if (!record)
464 break;
465handle_error:
466 if (record_type != TLS_RECORD_TYPE_DATA) {
467 /* avoid sending partial
468 * record with type !=
469 * application_data
470 */
471 size = orig_size;
472 destroy_record(record);
473 ctx->open_record = NULL;
474 } else if (record->len > prot->prepend_size) {
475 goto last_record;
476 }
477
478 break;
479 }
480
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));
484
485 rc = tls_device_copy_data(page_address(pfrag->page) +
486 pfrag->offset, copy, msg_iter);
487 if (rc)
488 goto handle_error;
489 tls_append_frag(record, pfrag, copy);
490
491 size -= copy;
492 if (!size) {
493last_record:
494 tls_push_record_flags = flags;
495 if (more) {
496 tls_ctx->pending_open_record_frags =
497 !!record->num_frags;
498 break;
499 }
500
501 done = true;
502 }
503
504 if (done || record->len >= max_open_record_len ||
505 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
506 rc = tls_device_record_close(sk, tls_ctx, record,
507 pfrag, record_type);
508 if (rc) {
509 if (rc > 0) {
510 size += rc;
511 } else {
512 size = orig_size;
513 destroy_record(record);
514 ctx->open_record = NULL;
515 break;
516 }
517 }
518
519 rc = tls_push_record(sk,
520 tls_ctx,
521 ctx,
522 record,
523 tls_push_record_flags);
524 if (rc < 0)
525 break;
526 }
527 } while (!done);
528
529 if (orig_size - size > 0)
530 rc = orig_size - size;
531
532 return rc;
533}
534
535int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
536{
537 unsigned char record_type = TLS_RECORD_TYPE_DATA;
538 struct tls_context *tls_ctx = tls_get_ctx(sk);
539 int rc;
540
541 mutex_lock(&tls_ctx->tx_lock);
542 lock_sock(sk);
543
544 if (unlikely(msg->msg_controllen)) {
545 rc = tls_proccess_cmsg(sk, msg, &record_type);
546 if (rc)
547 goto out;
548 }
549
550 rc = tls_push_data(sk, &msg->msg_iter, size,
551 msg->msg_flags, record_type);
552
553out:
554 release_sock(sk);
555 mutex_unlock(&tls_ctx->tx_lock);
556 return rc;
557}
558
559int tls_device_sendpage(struct sock *sk, struct page *page,
560 int offset, size_t size, int flags)
561{
562 struct tls_context *tls_ctx = tls_get_ctx(sk);
563 struct iov_iter msg_iter;
564 char *kaddr;
565 struct kvec iov;
566 int rc;
567
568 if (flags & MSG_SENDPAGE_NOTLAST)
569 flags |= MSG_MORE;
570
571 mutex_lock(&tls_ctx->tx_lock);
572 lock_sock(sk);
573
574 if (flags & MSG_OOB) {
575 rc = -EOPNOTSUPP;
576 goto out;
577 }
578
579 kaddr = kmap(page);
580 iov.iov_base = kaddr + offset;
581 iov.iov_len = size;
582 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
583 rc = tls_push_data(sk, &msg_iter, size,
584 flags, TLS_RECORD_TYPE_DATA);
585 kunmap(page);
586
587out:
588 release_sock(sk);
589 mutex_unlock(&tls_ctx->tx_lock);
590 return rc;
591}
592
593struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
594 u32 seq, u64 *p_record_sn)
595{
596 u64 record_sn = context->hint_record_sn;
597 struct tls_record_info *info, *last;
598
599 info = context->retransmit_hint;
600 if (!info ||
601 before(seq, info->end_seq - info->len)) {
602 /* if retransmit_hint is irrelevant start
603 * from the beggining of the list
604 */
605 info = list_first_entry_or_null(&context->records_list,
606 struct tls_record_info, list);
607 if (!info)
608 return NULL;
609 /* send the start_marker record if seq number is before the
610 * tls offload start marker sequence number. This record is
611 * required to handle TCP packets which are before TLS offload
612 * started.
613 * And if it's not start marker, look if this seq number
614 * belongs to the list.
615 */
616 if (likely(!tls_record_is_start_marker(info))) {
617 /* we have the first record, get the last record to see
618 * if this seq number belongs to the list.
619 */
620 last = list_last_entry(&context->records_list,
621 struct tls_record_info, list);
622
623 if (!between(seq, tls_record_start_seq(info),
624 last->end_seq))
625 return NULL;
626 }
627 record_sn = context->unacked_record_sn;
628 }
629
630 /* We just need the _rcu for the READ_ONCE() */
631 rcu_read_lock();
632 list_for_each_entry_from_rcu(info, &context->records_list, list) {
633 if (before(seq, info->end_seq)) {
634 if (!context->retransmit_hint ||
635 after(info->end_seq,
636 context->retransmit_hint->end_seq)) {
637 context->hint_record_sn = record_sn;
638 context->retransmit_hint = info;
639 }
640 *p_record_sn = record_sn;
641 goto exit_rcu_unlock;
642 }
643 record_sn++;
644 }
645 info = NULL;
646
647exit_rcu_unlock:
648 rcu_read_unlock();
649 return info;
650}
651EXPORT_SYMBOL(tls_get_record);
652
653static int tls_device_push_pending_record(struct sock *sk, int flags)
654{
655 struct iov_iter msg_iter;
656
657 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
658 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
659}
660
661void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
662{
663 if (tls_is_partially_sent_record(ctx)) {
664 gfp_t sk_allocation = sk->sk_allocation;
665
666 WARN_ON_ONCE(sk->sk_write_pending);
667
668 sk->sk_allocation = GFP_ATOMIC;
669 tls_push_partial_record(sk, ctx,
670 MSG_DONTWAIT | MSG_NOSIGNAL |
671 MSG_SENDPAGE_DECRYPTED);
672 sk->sk_allocation = sk_allocation;
673 }
674}
675
676static void tls_device_resync_rx(struct tls_context *tls_ctx,
677 struct sock *sk, u32 seq, u8 *rcd_sn)
678{
679 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
680 struct net_device *netdev;
681
682 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
683 return;
684
685 trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
686 netdev = READ_ONCE(tls_ctx->netdev);
687 if (netdev)
688 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
689 TLS_OFFLOAD_CTX_DIR_RX);
690 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
691 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
692}
693
694static bool
695tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
696 s64 resync_req, u32 *seq)
697{
698 u32 is_async = resync_req & RESYNC_REQ_ASYNC;
699 u32 req_seq = resync_req >> 32;
700 u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
701
702 if (is_async) {
703 /* asynchronous stage: log all headers seq such that
704 * req_seq <= seq <= end_seq, and wait for real resync request
705 */
706 if (between(*seq, req_seq, req_end) &&
707 resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
708 resync_async->log[resync_async->loglen++] = *seq;
709
710 return false;
711 }
712
713 /* synchronous stage: check against the logged entries and
714 * proceed to check the next entries if no match was found
715 */
716 while (resync_async->loglen) {
717 if (req_seq == resync_async->log[resync_async->loglen - 1] &&
718 atomic64_try_cmpxchg(&resync_async->req,
719 &resync_req, 0)) {
720 resync_async->loglen = 0;
721 *seq = req_seq;
722 return true;
723 }
724 resync_async->loglen--;
725 }
726
727 if (req_seq == *seq &&
728 atomic64_try_cmpxchg(&resync_async->req,
729 &resync_req, 0))
730 return true;
731
732 return false;
733}
734
735void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
736{
737 struct tls_context *tls_ctx = tls_get_ctx(sk);
738 struct tls_offload_context_rx *rx_ctx;
739 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
740 u32 sock_data, is_req_pending;
741 struct tls_prot_info *prot;
742 s64 resync_req;
743 u32 req_seq;
744
745 if (tls_ctx->rx_conf != TLS_HW)
746 return;
747
748 prot = &tls_ctx->prot_info;
749 rx_ctx = tls_offload_ctx_rx(tls_ctx);
750 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
751
752 switch (rx_ctx->resync_type) {
753 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
754 resync_req = atomic64_read(&rx_ctx->resync_req);
755 req_seq = resync_req >> 32;
756 seq += TLS_HEADER_SIZE - 1;
757 is_req_pending = resync_req;
758
759 if (likely(!is_req_pending) || req_seq != seq ||
760 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
761 return;
762 break;
763 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
764 if (likely(!rx_ctx->resync_nh_do_now))
765 return;
766
767 /* head of next rec is already in, note that the sock_inq will
768 * include the currently parsed message when called from parser
769 */
770 sock_data = tcp_inq(sk);
771 if (sock_data > rcd_len) {
772 trace_tls_device_rx_resync_nh_delay(sk, sock_data,
773 rcd_len);
774 return;
775 }
776
777 rx_ctx->resync_nh_do_now = 0;
778 seq += rcd_len;
779 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
780 break;
781 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
782 resync_req = atomic64_read(&rx_ctx->resync_async->req);
783 is_req_pending = resync_req;
784 if (likely(!is_req_pending))
785 return;
786
787 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
788 resync_req, &seq))
789 return;
790 break;
791 }
792
793 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
794}
795
796static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
797 struct tls_offload_context_rx *ctx,
798 struct sock *sk, struct sk_buff *skb)
799{
800 struct strp_msg *rxm;
801
802 /* device will request resyncs by itself based on stream scan */
803 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
804 return;
805 /* already scheduled */
806 if (ctx->resync_nh_do_now)
807 return;
808 /* seen decrypted fragments since last fully-failed record */
809 if (ctx->resync_nh_reset) {
810 ctx->resync_nh_reset = 0;
811 ctx->resync_nh.decrypted_failed = 1;
812 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
813 return;
814 }
815
816 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
817 return;
818
819 /* doing resync, bump the next target in case it fails */
820 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
821 ctx->resync_nh.decrypted_tgt *= 2;
822 else
823 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
824
825 rxm = strp_msg(skb);
826
827 /* head of next rec is already in, parser will sync for us */
828 if (tcp_inq(sk) > rxm->full_len) {
829 trace_tls_device_rx_resync_nh_schedule(sk);
830 ctx->resync_nh_do_now = 1;
831 } else {
832 struct tls_prot_info *prot = &tls_ctx->prot_info;
833 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
834
835 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
836 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
837
838 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
839 rcd_sn);
840 }
841}
842
843static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
844{
845 struct strp_msg *rxm = strp_msg(skb);
846 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
847 struct sk_buff *skb_iter, *unused;
848 struct scatterlist sg[1];
849 char *orig_buf, *buf;
850
851 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
852 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
853 if (!orig_buf)
854 return -ENOMEM;
855 buf = orig_buf;
856
857 nsg = skb_cow_data(skb, 0, &unused);
858 if (unlikely(nsg < 0)) {
859 err = nsg;
860 goto free_buf;
861 }
862
863 sg_init_table(sg, 1);
864 sg_set_buf(&sg[0], buf,
865 rxm->full_len + TLS_HEADER_SIZE +
866 TLS_CIPHER_AES_GCM_128_IV_SIZE);
867 err = skb_copy_bits(skb, offset, buf,
868 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
869 if (err)
870 goto free_buf;
871
872 /* We are interested only in the decrypted data not the auth */
873 err = decrypt_skb(sk, skb, sg);
874 if (err != -EBADMSG)
875 goto free_buf;
876 else
877 err = 0;
878
879 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
880
881 if (skb_pagelen(skb) > offset) {
882 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
883
884 if (skb->decrypted) {
885 err = skb_store_bits(skb, offset, buf, copy);
886 if (err)
887 goto free_buf;
888 }
889
890 offset += copy;
891 buf += copy;
892 }
893
894 pos = skb_pagelen(skb);
895 skb_walk_frags(skb, skb_iter) {
896 int frag_pos;
897
898 /* Practically all frags must belong to msg if reencrypt
899 * is needed with current strparser and coalescing logic,
900 * but strparser may "get optimized", so let's be safe.
901 */
902 if (pos + skb_iter->len <= offset)
903 goto done_with_frag;
904 if (pos >= data_len + rxm->offset)
905 break;
906
907 frag_pos = offset - pos;
908 copy = min_t(int, skb_iter->len - frag_pos,
909 data_len + rxm->offset - offset);
910
911 if (skb_iter->decrypted) {
912 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
913 if (err)
914 goto free_buf;
915 }
916
917 offset += copy;
918 buf += copy;
919done_with_frag:
920 pos += skb_iter->len;
921 }
922
923free_buf:
924 kfree(orig_buf);
925 return err;
926}
927
928int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
929 struct sk_buff *skb, struct strp_msg *rxm)
930{
931 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
932 int is_decrypted = skb->decrypted;
933 int is_encrypted = !is_decrypted;
934 struct sk_buff *skb_iter;
935
936 /* Check if all the data is decrypted already */
937 skb_walk_frags(skb, skb_iter) {
938 is_decrypted &= skb_iter->decrypted;
939 is_encrypted &= !skb_iter->decrypted;
940 }
941
942 trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
943 tls_ctx->rx.rec_seq, rxm->full_len,
944 is_encrypted, is_decrypted);
945
946 ctx->sw.decrypted |= is_decrypted;
947
948 /* Return immediately if the record is either entirely plaintext or
949 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
950 * record.
951 */
952 if (is_decrypted) {
953 ctx->resync_nh_reset = 1;
954 return 0;
955 }
956 if (is_encrypted) {
957 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
958 return 0;
959 }
960
961 ctx->resync_nh_reset = 1;
962 return tls_device_reencrypt(sk, skb);
963}
964
965static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
966 struct net_device *netdev)
967{
968 if (sk->sk_destruct != tls_device_sk_destruct) {
969 refcount_set(&ctx->refcount, 1);
970 dev_hold(netdev);
971 ctx->netdev = netdev;
972 spin_lock_irq(&tls_device_lock);
973 list_add_tail(&ctx->list, &tls_device_list);
974 spin_unlock_irq(&tls_device_lock);
975
976 ctx->sk_destruct = sk->sk_destruct;
977 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
978 }
979}
980
981int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
982{
983 u16 nonce_size, tag_size, iv_size, rec_seq_size;
984 struct tls_context *tls_ctx = tls_get_ctx(sk);
985 struct tls_prot_info *prot = &tls_ctx->prot_info;
986 struct tls_record_info *start_marker_record;
987 struct tls_offload_context_tx *offload_ctx;
988 struct tls_crypto_info *crypto_info;
989 struct net_device *netdev;
990 char *iv, *rec_seq;
991 struct sk_buff *skb;
992 __be64 rcd_sn;
993 int rc;
994
995 if (!ctx)
996 return -EINVAL;
997
998 if (ctx->priv_ctx_tx)
999 return -EEXIST;
1000
1001 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1002 if (!start_marker_record)
1003 return -ENOMEM;
1004
1005 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1006 if (!offload_ctx) {
1007 rc = -ENOMEM;
1008 goto free_marker_record;
1009 }
1010
1011 crypto_info = &ctx->crypto_send.info;
1012 if (crypto_info->version != TLS_1_2_VERSION) {
1013 rc = -EOPNOTSUPP;
1014 goto free_offload_ctx;
1015 }
1016
1017 switch (crypto_info->cipher_type) {
1018 case TLS_CIPHER_AES_GCM_128:
1019 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1020 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1021 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1022 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1023 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1024 rec_seq =
1025 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1026 break;
1027 default:
1028 rc = -EINVAL;
1029 goto free_offload_ctx;
1030 }
1031
1032 /* Sanity-check the rec_seq_size for stack allocations */
1033 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1034 rc = -EINVAL;
1035 goto free_offload_ctx;
1036 }
1037
1038 prot->version = crypto_info->version;
1039 prot->cipher_type = crypto_info->cipher_type;
1040 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1041 prot->tag_size = tag_size;
1042 prot->overhead_size = prot->prepend_size + prot->tag_size;
1043 prot->iv_size = iv_size;
1044 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1045 GFP_KERNEL);
1046 if (!ctx->tx.iv) {
1047 rc = -ENOMEM;
1048 goto free_offload_ctx;
1049 }
1050
1051 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1052
1053 prot->rec_seq_size = rec_seq_size;
1054 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1055 if (!ctx->tx.rec_seq) {
1056 rc = -ENOMEM;
1057 goto free_iv;
1058 }
1059
1060 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1061 if (rc)
1062 goto free_rec_seq;
1063
1064 /* start at rec_seq - 1 to account for the start marker record */
1065 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1066 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1067
1068 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1069 start_marker_record->len = 0;
1070 start_marker_record->num_frags = 0;
1071
1072 INIT_LIST_HEAD(&offload_ctx->records_list);
1073 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1074 spin_lock_init(&offload_ctx->lock);
1075 sg_init_table(offload_ctx->sg_tx_data,
1076 ARRAY_SIZE(offload_ctx->sg_tx_data));
1077
1078 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1079 ctx->push_pending_record = tls_device_push_pending_record;
1080
1081 /* TLS offload is greatly simplified if we don't send
1082 * SKBs where only part of the payload needs to be encrypted.
1083 * So mark the last skb in the write queue as end of record.
1084 */
1085 skb = tcp_write_queue_tail(sk);
1086 if (skb)
1087 TCP_SKB_CB(skb)->eor = 1;
1088
1089 netdev = get_netdev_for_sock(sk);
1090 if (!netdev) {
1091 pr_err_ratelimited("%s: netdev not found\n", __func__);
1092 rc = -EINVAL;
1093 goto disable_cad;
1094 }
1095
1096 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1097 rc = -EOPNOTSUPP;
1098 goto release_netdev;
1099 }
1100
1101 /* Avoid offloading if the device is down
1102 * We don't want to offload new flows after
1103 * the NETDEV_DOWN event
1104 *
1105 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1106 * handler thus protecting from the device going down before
1107 * ctx was added to tls_device_list.
1108 */
1109 down_read(&device_offload_lock);
1110 if (!(netdev->flags & IFF_UP)) {
1111 rc = -EINVAL;
1112 goto release_lock;
1113 }
1114
1115 ctx->priv_ctx_tx = offload_ctx;
1116 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1117 &ctx->crypto_send.info,
1118 tcp_sk(sk)->write_seq);
1119 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1120 tcp_sk(sk)->write_seq, rec_seq, rc);
1121 if (rc)
1122 goto release_lock;
1123
1124 tls_device_attach(ctx, sk, netdev);
1125 up_read(&device_offload_lock);
1126
1127 /* following this assignment tls_is_sk_tx_device_offloaded
1128 * will return true and the context might be accessed
1129 * by the netdev's xmit function.
1130 */
1131 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1132 dev_put(netdev);
1133
1134 return 0;
1135
1136release_lock:
1137 up_read(&device_offload_lock);
1138release_netdev:
1139 dev_put(netdev);
1140disable_cad:
1141 clean_acked_data_disable(inet_csk(sk));
1142 crypto_free_aead(offload_ctx->aead_send);
1143free_rec_seq:
1144 kfree(ctx->tx.rec_seq);
1145free_iv:
1146 kfree(ctx->tx.iv);
1147free_offload_ctx:
1148 kfree(offload_ctx);
1149 ctx->priv_ctx_tx = NULL;
1150free_marker_record:
1151 kfree(start_marker_record);
1152 return rc;
1153}
1154
1155int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1156{
1157 struct tls12_crypto_info_aes_gcm_128 *info;
1158 struct tls_offload_context_rx *context;
1159 struct net_device *netdev;
1160 int rc = 0;
1161
1162 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1163 return -EOPNOTSUPP;
1164
1165 netdev = get_netdev_for_sock(sk);
1166 if (!netdev) {
1167 pr_err_ratelimited("%s: netdev not found\n", __func__);
1168 return -EINVAL;
1169 }
1170
1171 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1172 rc = -EOPNOTSUPP;
1173 goto release_netdev;
1174 }
1175
1176 /* Avoid offloading if the device is down
1177 * We don't want to offload new flows after
1178 * the NETDEV_DOWN event
1179 *
1180 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1181 * handler thus protecting from the device going down before
1182 * ctx was added to tls_device_list.
1183 */
1184 down_read(&device_offload_lock);
1185 if (!(netdev->flags & IFF_UP)) {
1186 rc = -EINVAL;
1187 goto release_lock;
1188 }
1189
1190 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1191 if (!context) {
1192 rc = -ENOMEM;
1193 goto release_lock;
1194 }
1195 context->resync_nh_reset = 1;
1196
1197 ctx->priv_ctx_rx = context;
1198 rc = tls_set_sw_offload(sk, ctx, 0);
1199 if (rc)
1200 goto release_ctx;
1201
1202 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1203 &ctx->crypto_recv.info,
1204 tcp_sk(sk)->copied_seq);
1205 info = (void *)&ctx->crypto_recv.info;
1206 trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1207 tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1208 if (rc)
1209 goto free_sw_resources;
1210
1211 tls_device_attach(ctx, sk, netdev);
1212 up_read(&device_offload_lock);
1213
1214 dev_put(netdev);
1215
1216 return 0;
1217
1218free_sw_resources:
1219 up_read(&device_offload_lock);
1220 tls_sw_free_resources_rx(sk);
1221 down_read(&device_offload_lock);
1222release_ctx:
1223 ctx->priv_ctx_rx = NULL;
1224release_lock:
1225 up_read(&device_offload_lock);
1226release_netdev:
1227 dev_put(netdev);
1228 return rc;
1229}
1230
1231void tls_device_offload_cleanup_rx(struct sock *sk)
1232{
1233 struct tls_context *tls_ctx = tls_get_ctx(sk);
1234 struct net_device *netdev;
1235
1236 down_read(&device_offload_lock);
1237 netdev = tls_ctx->netdev;
1238 if (!netdev)
1239 goto out;
1240
1241 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1242 TLS_OFFLOAD_CTX_DIR_RX);
1243
1244 if (tls_ctx->tx_conf != TLS_HW) {
1245 dev_put(netdev);
1246 tls_ctx->netdev = NULL;
1247 }
1248out:
1249 up_read(&device_offload_lock);
1250 tls_sw_release_resources_rx(sk);
1251}
1252
1253static int tls_device_down(struct net_device *netdev)
1254{
1255 struct tls_context *ctx, *tmp;
1256 unsigned long flags;
1257 LIST_HEAD(list);
1258
1259 /* Request a write lock to block new offload attempts */
1260 down_write(&device_offload_lock);
1261
1262 spin_lock_irqsave(&tls_device_lock, flags);
1263 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1264 if (ctx->netdev != netdev ||
1265 !refcount_inc_not_zero(&ctx->refcount))
1266 continue;
1267
1268 list_move(&ctx->list, &list);
1269 }
1270 spin_unlock_irqrestore(&tls_device_lock, flags);
1271
1272 list_for_each_entry_safe(ctx, tmp, &list, list) {
1273 if (ctx->tx_conf == TLS_HW)
1274 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1275 TLS_OFFLOAD_CTX_DIR_TX);
1276 if (ctx->rx_conf == TLS_HW)
1277 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1278 TLS_OFFLOAD_CTX_DIR_RX);
1279 WRITE_ONCE(ctx->netdev, NULL);
1280 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1281 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1282 usleep_range(10, 200);
1283 dev_put(netdev);
1284 list_del_init(&ctx->list);
1285
1286 if (refcount_dec_and_test(&ctx->refcount))
1287 tls_device_free_ctx(ctx);
1288 }
1289
1290 up_write(&device_offload_lock);
1291
1292 flush_work(&tls_device_gc_work);
1293
1294 return NOTIFY_DONE;
1295}
1296
1297static int tls_dev_event(struct notifier_block *this, unsigned long event,
1298 void *ptr)
1299{
1300 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1301
1302 if (!dev->tlsdev_ops &&
1303 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1304 return NOTIFY_DONE;
1305
1306 switch (event) {
1307 case NETDEV_REGISTER:
1308 case NETDEV_FEAT_CHANGE:
1309 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1310 !dev->tlsdev_ops->tls_dev_resync)
1311 return NOTIFY_BAD;
1312
1313 if (dev->tlsdev_ops &&
1314 dev->tlsdev_ops->tls_dev_add &&
1315 dev->tlsdev_ops->tls_dev_del)
1316 return NOTIFY_DONE;
1317 else
1318 return NOTIFY_BAD;
1319 case NETDEV_DOWN:
1320 return tls_device_down(dev);
1321 }
1322 return NOTIFY_DONE;
1323}
1324
1325static struct notifier_block tls_dev_notifier = {
1326 .notifier_call = tls_dev_event,
1327};
1328
1329void __init tls_device_init(void)
1330{
1331 register_netdevice_notifier(&tls_dev_notifier);
1332}
1333
1334void __exit tls_device_cleanup(void)
1335{
1336 unregister_netdevice_notifier(&tls_dev_notifier);
1337 flush_work(&tls_device_gc_work);
1338 clean_acked_data_flush();
1339}