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 /* device_offload_lock is used to synchronize tls_dev_add
42 * against NETDEV_DOWN notifications.
43 */
44 static DECLARE_RWSEM(device_offload_lock);
45
46 static void tls_device_gc_task(struct work_struct *work);
47
48 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
49 static LIST_HEAD(tls_device_gc_list);
50 static LIST_HEAD(tls_device_list);
51 static DEFINE_SPINLOCK(tls_device_lock);
52
tls_device_free_ctx(struct tls_context * ctx)53 static void tls_device_free_ctx(struct tls_context *ctx)
54 {
55 if (ctx->tx_conf == TLS_HW) {
56 kfree(tls_offload_ctx_tx(ctx));
57 kfree(ctx->tx.rec_seq);
58 kfree(ctx->tx.iv);
59 }
60
61 if (ctx->rx_conf == TLS_HW)
62 kfree(tls_offload_ctx_rx(ctx));
63
64 tls_ctx_free(NULL, ctx);
65 }
66
tls_device_gc_task(struct work_struct * work)67 static void tls_device_gc_task(struct work_struct *work)
68 {
69 struct tls_context *ctx, *tmp;
70 unsigned long flags;
71 LIST_HEAD(gc_list);
72
73 spin_lock_irqsave(&tls_device_lock, flags);
74 list_splice_init(&tls_device_gc_list, &gc_list);
75 spin_unlock_irqrestore(&tls_device_lock, flags);
76
77 list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
78 struct net_device *netdev = ctx->netdev;
79
80 if (netdev && ctx->tx_conf == TLS_HW) {
81 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
82 TLS_OFFLOAD_CTX_DIR_TX);
83 dev_put(netdev);
84 ctx->netdev = NULL;
85 }
86
87 list_del(&ctx->list);
88 tls_device_free_ctx(ctx);
89 }
90 }
91
tls_device_queue_ctx_destruction(struct tls_context * ctx)92 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
93 {
94 unsigned long flags;
95
96 spin_lock_irqsave(&tls_device_lock, flags);
97 if (unlikely(!refcount_dec_and_test(&ctx->refcount)))
98 goto unlock;
99
100 list_move_tail(&ctx->list, &tls_device_gc_list);
101
102 /* schedule_work inside the spinlock
103 * to make sure tls_device_down waits for that work.
104 */
105 schedule_work(&tls_device_gc_work);
106 unlock:
107 spin_unlock_irqrestore(&tls_device_lock, flags);
108 }
109
110 /* We assume that the socket is already connected */
get_netdev_for_sock(struct sock * sk)111 static struct net_device *get_netdev_for_sock(struct sock *sk)
112 {
113 struct dst_entry *dst = sk_dst_get(sk);
114 struct net_device *netdev = NULL;
115
116 if (likely(dst)) {
117 netdev = dst->dev;
118 dev_hold(netdev);
119 }
120
121 dst_release(dst);
122
123 return netdev;
124 }
125
destroy_record(struct tls_record_info * record)126 static void destroy_record(struct tls_record_info *record)
127 {
128 int i;
129
130 for (i = 0; i < record->num_frags; i++)
131 __skb_frag_unref(&record->frags[i]);
132 kfree(record);
133 }
134
delete_all_records(struct tls_offload_context_tx * offload_ctx)135 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
136 {
137 struct tls_record_info *info, *temp;
138
139 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
140 list_del(&info->list);
141 destroy_record(info);
142 }
143
144 offload_ctx->retransmit_hint = NULL;
145 }
146
tls_icsk_clean_acked(struct sock * sk,u32 acked_seq)147 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
148 {
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;
153 unsigned long flags;
154
155 if (!tls_ctx)
156 return;
157
158 ctx = tls_offload_ctx_tx(tls_ctx);
159
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;
164
165 list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
166 if (before(acked_seq, info->end_seq))
167 break;
168 list_del(&info->list);
169
170 destroy_record(info);
171 deleted_records++;
172 }
173
174 ctx->unacked_record_sn += deleted_records;
175 spin_unlock_irqrestore(&ctx->lock, flags);
176 }
177
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.
181 */
tls_device_sk_destruct(struct sock * sk)182 static void tls_device_sk_destruct(struct sock *sk)
183 {
184 struct tls_context *tls_ctx = tls_get_ctx(sk);
185 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
186
187 tls_ctx->sk_destruct(sk);
188
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));
195 }
196
197 tls_device_queue_ctx_destruction(tls_ctx);
198 }
199
tls_device_free_resources_tx(struct sock * sk)200 void tls_device_free_resources_tx(struct sock *sk)
201 {
202 struct tls_context *tls_ctx = tls_get_ctx(sk);
203
204 tls_free_partial_record(sk, tls_ctx);
205 }
206
tls_device_resync_tx(struct sock * sk,struct tls_context * tls_ctx,u32 seq)207 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
208 u32 seq)
209 {
210 struct net_device *netdev;
211 struct sk_buff *skb;
212 int err = 0;
213 u8 *rcd_sn;
214
215 skb = tcp_write_queue_tail(sk);
216 if (skb)
217 TCP_SKB_CB(skb)->eor = 1;
218
219 rcd_sn = tls_ctx->tx.rec_seq;
220
221 down_read(&device_offload_lock);
222 netdev = tls_ctx->netdev;
223 if (netdev)
224 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
225 rcd_sn,
226 TLS_OFFLOAD_CTX_DIR_TX);
227 up_read(&device_offload_lock);
228 if (err)
229 return;
230
231 clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
232 }
233
tls_append_frag(struct tls_record_info * record,struct page_frag * pfrag,int size)234 static void tls_append_frag(struct tls_record_info *record,
235 struct page_frag *pfrag,
236 int size)
237 {
238 skb_frag_t *frag;
239
240 frag = &record->frags[record->num_frags - 1];
241 if (skb_frag_page(frag) == pfrag->page &&
242 skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
243 skb_frag_size_add(frag, size);
244 } else {
245 ++frag;
246 __skb_frag_set_page(frag, pfrag->page);
247 skb_frag_off_set(frag, pfrag->offset);
248 skb_frag_size_set(frag, size);
249 ++record->num_frags;
250 get_page(pfrag->page);
251 }
252
253 pfrag->offset += size;
254 record->len += size;
255 }
256
tls_push_record(struct sock * sk,struct tls_context * ctx,struct tls_offload_context_tx * offload_ctx,struct tls_record_info * record,int flags)257 static int tls_push_record(struct sock *sk,
258 struct tls_context *ctx,
259 struct tls_offload_context_tx *offload_ctx,
260 struct tls_record_info *record,
261 int flags)
262 {
263 struct tls_prot_info *prot = &ctx->prot_info;
264 struct tcp_sock *tp = tcp_sk(sk);
265 skb_frag_t *frag;
266 int i;
267
268 record->end_seq = tp->write_seq + record->len;
269 list_add_tail_rcu(&record->list, &offload_ctx->records_list);
270 offload_ctx->open_record = NULL;
271
272 if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
273 tls_device_resync_tx(sk, ctx, tp->write_seq);
274
275 tls_advance_record_sn(sk, prot, &ctx->tx);
276
277 for (i = 0; i < record->num_frags; i++) {
278 frag = &record->frags[i];
279 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
280 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
281 skb_frag_size(frag), skb_frag_off(frag));
282 sk_mem_charge(sk, skb_frag_size(frag));
283 get_page(skb_frag_page(frag));
284 }
285 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
286
287 /* all ready, send */
288 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
289 }
290
tls_device_record_close(struct sock * sk,struct tls_context * ctx,struct tls_record_info * record,struct page_frag * pfrag,unsigned char record_type)291 static int tls_device_record_close(struct sock *sk,
292 struct tls_context *ctx,
293 struct tls_record_info *record,
294 struct page_frag *pfrag,
295 unsigned char record_type)
296 {
297 struct tls_prot_info *prot = &ctx->prot_info;
298 int ret;
299
300 /* append tag
301 * device will fill in the tag, we just need to append a placeholder
302 * use socket memory to improve coalescing (re-using a single buffer
303 * increases frag count)
304 * if we can't allocate memory now, steal some back from data
305 */
306 if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
307 sk->sk_allocation))) {
308 ret = 0;
309 tls_append_frag(record, pfrag, prot->tag_size);
310 } else {
311 ret = prot->tag_size;
312 if (record->len <= prot->overhead_size)
313 return -ENOMEM;
314 }
315
316 /* fill prepend */
317 tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
318 record->len - prot->overhead_size,
319 record_type, prot->version);
320 return ret;
321 }
322
tls_create_new_record(struct tls_offload_context_tx * offload_ctx,struct page_frag * pfrag,size_t prepend_size)323 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
324 struct page_frag *pfrag,
325 size_t prepend_size)
326 {
327 struct tls_record_info *record;
328 skb_frag_t *frag;
329
330 record = kmalloc(sizeof(*record), GFP_KERNEL);
331 if (!record)
332 return -ENOMEM;
333
334 frag = &record->frags[0];
335 __skb_frag_set_page(frag, pfrag->page);
336 skb_frag_off_set(frag, pfrag->offset);
337 skb_frag_size_set(frag, prepend_size);
338
339 get_page(pfrag->page);
340 pfrag->offset += prepend_size;
341
342 record->num_frags = 1;
343 record->len = prepend_size;
344 offload_ctx->open_record = record;
345 return 0;
346 }
347
tls_do_allocation(struct sock * sk,struct tls_offload_context_tx * offload_ctx,struct page_frag * pfrag,size_t prepend_size)348 static int tls_do_allocation(struct sock *sk,
349 struct tls_offload_context_tx *offload_ctx,
350 struct page_frag *pfrag,
351 size_t prepend_size)
352 {
353 int ret;
354
355 if (!offload_ctx->open_record) {
356 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
357 sk->sk_allocation))) {
358 sk->sk_prot->enter_memory_pressure(sk);
359 sk_stream_moderate_sndbuf(sk);
360 return -ENOMEM;
361 }
362
363 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
364 if (ret)
365 return ret;
366
367 if (pfrag->size > pfrag->offset)
368 return 0;
369 }
370
371 if (!sk_page_frag_refill(sk, pfrag))
372 return -ENOMEM;
373
374 return 0;
375 }
376
tls_device_copy_data(void * addr,size_t bytes,struct iov_iter * i)377 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
378 {
379 size_t pre_copy, nocache;
380
381 pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
382 if (pre_copy) {
383 pre_copy = min(pre_copy, bytes);
384 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
385 return -EFAULT;
386 bytes -= pre_copy;
387 addr += pre_copy;
388 }
389
390 nocache = round_down(bytes, SMP_CACHE_BYTES);
391 if (copy_from_iter_nocache(addr, nocache, i) != nocache)
392 return -EFAULT;
393 bytes -= nocache;
394 addr += nocache;
395
396 if (bytes && copy_from_iter(addr, bytes, i) != bytes)
397 return -EFAULT;
398
399 return 0;
400 }
401
tls_push_data(struct sock * sk,struct iov_iter * msg_iter,size_t size,int flags,unsigned char record_type)402 static int tls_push_data(struct sock *sk,
403 struct iov_iter *msg_iter,
404 size_t size, int flags,
405 unsigned char record_type)
406 {
407 struct tls_context *tls_ctx = tls_get_ctx(sk);
408 struct tls_prot_info *prot = &tls_ctx->prot_info;
409 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
410 struct tls_record_info *record = ctx->open_record;
411 int tls_push_record_flags;
412 struct page_frag *pfrag;
413 size_t orig_size = size;
414 u32 max_open_record_len;
415 bool more = false;
416 bool done = false;
417 int copy, rc = 0;
418 long timeo;
419
420 if (flags &
421 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
422 return -EOPNOTSUPP;
423
424 if (sk->sk_err)
425 return -sk->sk_err;
426
427 flags |= MSG_SENDPAGE_DECRYPTED;
428 tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
429
430 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
431 if (tls_is_partially_sent_record(tls_ctx)) {
432 rc = tls_push_partial_record(sk, tls_ctx, flags);
433 if (rc < 0)
434 return rc;
435 }
436
437 pfrag = sk_page_frag(sk);
438
439 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
440 * we need to leave room for an authentication tag.
441 */
442 max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
443 prot->prepend_size;
444 do {
445 rc = tls_do_allocation(sk, ctx, pfrag,
446 prot->prepend_size);
447 if (rc) {
448 rc = sk_stream_wait_memory(sk, &timeo);
449 if (!rc)
450 continue;
451
452 record = ctx->open_record;
453 if (!record)
454 break;
455 handle_error:
456 if (record_type != TLS_RECORD_TYPE_DATA) {
457 /* avoid sending partial
458 * record with type !=
459 * application_data
460 */
461 size = orig_size;
462 destroy_record(record);
463 ctx->open_record = NULL;
464 } else if (record->len > prot->prepend_size) {
465 goto last_record;
466 }
467
468 break;
469 }
470
471 record = ctx->open_record;
472 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
473 copy = min_t(size_t, copy, (max_open_record_len - record->len));
474
475 if (copy) {
476 rc = tls_device_copy_data(page_address(pfrag->page) +
477 pfrag->offset, copy, msg_iter);
478 if (rc)
479 goto handle_error;
480 tls_append_frag(record, pfrag, copy);
481 }
482
483 size -= copy;
484 if (!size) {
485 last_record:
486 tls_push_record_flags = flags;
487 if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
488 more = true;
489 break;
490 }
491
492 done = true;
493 }
494
495 if (done || record->len >= max_open_record_len ||
496 (record->num_frags >= MAX_SKB_FRAGS - 1)) {
497 rc = tls_device_record_close(sk, tls_ctx, record,
498 pfrag, record_type);
499 if (rc) {
500 if (rc > 0) {
501 size += rc;
502 } else {
503 size = orig_size;
504 destroy_record(record);
505 ctx->open_record = NULL;
506 break;
507 }
508 }
509
510 rc = tls_push_record(sk,
511 tls_ctx,
512 ctx,
513 record,
514 tls_push_record_flags);
515 if (rc < 0)
516 break;
517 }
518 } while (!done);
519
520 tls_ctx->pending_open_record_frags = more;
521
522 if (orig_size - size > 0)
523 rc = orig_size - size;
524
525 return rc;
526 }
527
tls_device_sendmsg(struct sock * sk,struct msghdr * msg,size_t size)528 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
529 {
530 unsigned char record_type = TLS_RECORD_TYPE_DATA;
531 struct tls_context *tls_ctx = tls_get_ctx(sk);
532 int rc;
533
534 mutex_lock(&tls_ctx->tx_lock);
535 lock_sock(sk);
536
537 if (unlikely(msg->msg_controllen)) {
538 rc = tls_proccess_cmsg(sk, msg, &record_type);
539 if (rc)
540 goto out;
541 }
542
543 rc = tls_push_data(sk, &msg->msg_iter, size,
544 msg->msg_flags, record_type);
545
546 out:
547 release_sock(sk);
548 mutex_unlock(&tls_ctx->tx_lock);
549 return rc;
550 }
551
tls_device_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)552 int tls_device_sendpage(struct sock *sk, struct page *page,
553 int offset, size_t size, int flags)
554 {
555 struct tls_context *tls_ctx = tls_get_ctx(sk);
556 struct iov_iter msg_iter;
557 char *kaddr;
558 struct kvec iov;
559 int rc;
560
561 if (flags & MSG_SENDPAGE_NOTLAST)
562 flags |= MSG_MORE;
563
564 mutex_lock(&tls_ctx->tx_lock);
565 lock_sock(sk);
566
567 if (flags & MSG_OOB) {
568 rc = -EOPNOTSUPP;
569 goto out;
570 }
571
572 kaddr = kmap(page);
573 iov.iov_base = kaddr + offset;
574 iov.iov_len = size;
575 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
576 rc = tls_push_data(sk, &msg_iter, size,
577 flags, TLS_RECORD_TYPE_DATA);
578 kunmap(page);
579
580 out:
581 release_sock(sk);
582 mutex_unlock(&tls_ctx->tx_lock);
583 return rc;
584 }
585
tls_get_record(struct tls_offload_context_tx * context,u32 seq,u64 * p_record_sn)586 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
587 u32 seq, u64 *p_record_sn)
588 {
589 u64 record_sn = context->hint_record_sn;
590 struct tls_record_info *info, *last;
591
592 info = context->retransmit_hint;
593 if (!info ||
594 before(seq, info->end_seq - info->len)) {
595 /* if retransmit_hint is irrelevant start
596 * from the beggining of the list
597 */
598 info = list_first_entry_or_null(&context->records_list,
599 struct tls_record_info, list);
600 if (!info)
601 return NULL;
602 /* send the start_marker record if seq number is before the
603 * tls offload start marker sequence number. This record is
604 * required to handle TCP packets which are before TLS offload
605 * started.
606 * And if it's not start marker, look if this seq number
607 * belongs to the list.
608 */
609 if (likely(!tls_record_is_start_marker(info))) {
610 /* we have the first record, get the last record to see
611 * if this seq number belongs to the list.
612 */
613 last = list_last_entry(&context->records_list,
614 struct tls_record_info, list);
615
616 if (!between(seq, tls_record_start_seq(info),
617 last->end_seq))
618 return NULL;
619 }
620 record_sn = context->unacked_record_sn;
621 }
622
623 /* We just need the _rcu for the READ_ONCE() */
624 rcu_read_lock();
625 list_for_each_entry_from_rcu(info, &context->records_list, list) {
626 if (before(seq, info->end_seq)) {
627 if (!context->retransmit_hint ||
628 after(info->end_seq,
629 context->retransmit_hint->end_seq)) {
630 context->hint_record_sn = record_sn;
631 context->retransmit_hint = info;
632 }
633 *p_record_sn = record_sn;
634 goto exit_rcu_unlock;
635 }
636 record_sn++;
637 }
638 info = NULL;
639
640 exit_rcu_unlock:
641 rcu_read_unlock();
642 return info;
643 }
644 EXPORT_SYMBOL(tls_get_record);
645
tls_device_push_pending_record(struct sock * sk,int flags)646 static int tls_device_push_pending_record(struct sock *sk, int flags)
647 {
648 struct iov_iter msg_iter;
649
650 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
651 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
652 }
653
tls_device_write_space(struct sock * sk,struct tls_context * ctx)654 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
655 {
656 if (tls_is_partially_sent_record(ctx)) {
657 gfp_t sk_allocation = sk->sk_allocation;
658
659 WARN_ON_ONCE(sk->sk_write_pending);
660
661 sk->sk_allocation = GFP_ATOMIC;
662 tls_push_partial_record(sk, ctx,
663 MSG_DONTWAIT | MSG_NOSIGNAL |
664 MSG_SENDPAGE_DECRYPTED);
665 sk->sk_allocation = sk_allocation;
666 }
667 }
668
tls_device_resync_rx(struct tls_context * tls_ctx,struct sock * sk,u32 seq,u8 * rcd_sn)669 static void tls_device_resync_rx(struct tls_context *tls_ctx,
670 struct sock *sk, u32 seq, u8 *rcd_sn)
671 {
672 struct net_device *netdev;
673
674 if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
675 return;
676 netdev = READ_ONCE(tls_ctx->netdev);
677 if (netdev)
678 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
679 TLS_OFFLOAD_CTX_DIR_RX);
680 clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
681 }
682
tls_device_rx_resync_new_rec(struct sock * sk,u32 rcd_len,u32 seq)683 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
684 {
685 struct tls_context *tls_ctx = tls_get_ctx(sk);
686 struct tls_offload_context_rx *rx_ctx;
687 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
688 struct tls_prot_info *prot;
689 u32 is_req_pending;
690 s64 resync_req;
691 u32 req_seq;
692
693 if (tls_ctx->rx_conf != TLS_HW)
694 return;
695
696 prot = &tls_ctx->prot_info;
697 rx_ctx = tls_offload_ctx_rx(tls_ctx);
698 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
699
700 switch (rx_ctx->resync_type) {
701 case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
702 resync_req = atomic64_read(&rx_ctx->resync_req);
703 req_seq = resync_req >> 32;
704 seq += TLS_HEADER_SIZE - 1;
705 is_req_pending = resync_req;
706
707 if (likely(!is_req_pending) || req_seq != seq ||
708 !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
709 return;
710 break;
711 case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
712 if (likely(!rx_ctx->resync_nh_do_now))
713 return;
714
715 /* head of next rec is already in, note that the sock_inq will
716 * include the currently parsed message when called from parser
717 */
718 if (tcp_inq(sk) > rcd_len)
719 return;
720
721 rx_ctx->resync_nh_do_now = 0;
722 seq += rcd_len;
723 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
724 break;
725 }
726
727 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
728 }
729
tls_device_core_ctrl_rx_resync(struct tls_context * tls_ctx,struct tls_offload_context_rx * ctx,struct sock * sk,struct sk_buff * skb)730 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
731 struct tls_offload_context_rx *ctx,
732 struct sock *sk, struct sk_buff *skb)
733 {
734 struct strp_msg *rxm;
735
736 /* device will request resyncs by itself based on stream scan */
737 if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
738 return;
739 /* already scheduled */
740 if (ctx->resync_nh_do_now)
741 return;
742 /* seen decrypted fragments since last fully-failed record */
743 if (ctx->resync_nh_reset) {
744 ctx->resync_nh_reset = 0;
745 ctx->resync_nh.decrypted_failed = 1;
746 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
747 return;
748 }
749
750 if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
751 return;
752
753 /* doing resync, bump the next target in case it fails */
754 if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
755 ctx->resync_nh.decrypted_tgt *= 2;
756 else
757 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
758
759 rxm = strp_msg(skb);
760
761 /* head of next rec is already in, parser will sync for us */
762 if (tcp_inq(sk) > rxm->full_len) {
763 ctx->resync_nh_do_now = 1;
764 } else {
765 struct tls_prot_info *prot = &tls_ctx->prot_info;
766 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
767
768 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
769 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
770
771 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
772 rcd_sn);
773 }
774 }
775
tls_device_reencrypt(struct sock * sk,struct sk_buff * skb)776 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
777 {
778 struct strp_msg *rxm = strp_msg(skb);
779 int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
780 struct sk_buff *skb_iter, *unused;
781 struct scatterlist sg[1];
782 char *orig_buf, *buf;
783
784 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
785 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
786 if (!orig_buf)
787 return -ENOMEM;
788 buf = orig_buf;
789
790 nsg = skb_cow_data(skb, 0, &unused);
791 if (unlikely(nsg < 0)) {
792 err = nsg;
793 goto free_buf;
794 }
795
796 sg_init_table(sg, 1);
797 sg_set_buf(&sg[0], buf,
798 rxm->full_len + TLS_HEADER_SIZE +
799 TLS_CIPHER_AES_GCM_128_IV_SIZE);
800 err = skb_copy_bits(skb, offset, buf,
801 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
802 if (err)
803 goto free_buf;
804
805 /* We are interested only in the decrypted data not the auth */
806 err = decrypt_skb(sk, skb, sg);
807 if (err != -EBADMSG)
808 goto free_buf;
809 else
810 err = 0;
811
812 data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
813
814 if (skb_pagelen(skb) > offset) {
815 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
816
817 if (skb->decrypted) {
818 err = skb_store_bits(skb, offset, buf, copy);
819 if (err)
820 goto free_buf;
821 }
822
823 offset += copy;
824 buf += copy;
825 }
826
827 pos = skb_pagelen(skb);
828 skb_walk_frags(skb, skb_iter) {
829 int frag_pos;
830
831 /* Practically all frags must belong to msg if reencrypt
832 * is needed with current strparser and coalescing logic,
833 * but strparser may "get optimized", so let's be safe.
834 */
835 if (pos + skb_iter->len <= offset)
836 goto done_with_frag;
837 if (pos >= data_len + rxm->offset)
838 break;
839
840 frag_pos = offset - pos;
841 copy = min_t(int, skb_iter->len - frag_pos,
842 data_len + rxm->offset - offset);
843
844 if (skb_iter->decrypted) {
845 err = skb_store_bits(skb_iter, frag_pos, buf, copy);
846 if (err)
847 goto free_buf;
848 }
849
850 offset += copy;
851 buf += copy;
852 done_with_frag:
853 pos += skb_iter->len;
854 }
855
856 free_buf:
857 kfree(orig_buf);
858 return err;
859 }
860
tls_device_decrypted(struct sock * sk,struct sk_buff * skb)861 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
862 {
863 struct tls_context *tls_ctx = tls_get_ctx(sk);
864 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
865 int is_decrypted = skb->decrypted;
866 int is_encrypted = !is_decrypted;
867 struct sk_buff *skb_iter;
868
869 /* Check if all the data is decrypted already */
870 skb_walk_frags(skb, skb_iter) {
871 is_decrypted &= skb_iter->decrypted;
872 is_encrypted &= !skb_iter->decrypted;
873 }
874
875 ctx->sw.decrypted |= is_decrypted;
876
877 /* Return immediately if the record is either entirely plaintext or
878 * entirely ciphertext. Otherwise handle reencrypt partially decrypted
879 * record.
880 */
881 if (is_decrypted) {
882 ctx->resync_nh_reset = 1;
883 return 0;
884 }
885 if (is_encrypted) {
886 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
887 return 0;
888 }
889
890 ctx->resync_nh_reset = 1;
891 return tls_device_reencrypt(sk, skb);
892 }
893
tls_device_attach(struct tls_context * ctx,struct sock * sk,struct net_device * netdev)894 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
895 struct net_device *netdev)
896 {
897 if (sk->sk_destruct != tls_device_sk_destruct) {
898 refcount_set(&ctx->refcount, 1);
899 dev_hold(netdev);
900 ctx->netdev = netdev;
901 spin_lock_irq(&tls_device_lock);
902 list_add_tail(&ctx->list, &tls_device_list);
903 spin_unlock_irq(&tls_device_lock);
904
905 ctx->sk_destruct = sk->sk_destruct;
906 sk->sk_destruct = tls_device_sk_destruct;
907 }
908 }
909
tls_set_device_offload(struct sock * sk,struct tls_context * ctx)910 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
911 {
912 u16 nonce_size, tag_size, iv_size, rec_seq_size;
913 struct tls_context *tls_ctx = tls_get_ctx(sk);
914 struct tls_prot_info *prot = &tls_ctx->prot_info;
915 struct tls_record_info *start_marker_record;
916 struct tls_offload_context_tx *offload_ctx;
917 struct tls_crypto_info *crypto_info;
918 struct net_device *netdev;
919 char *iv, *rec_seq;
920 struct sk_buff *skb;
921 __be64 rcd_sn;
922 int rc;
923
924 if (!ctx)
925 return -EINVAL;
926
927 if (ctx->priv_ctx_tx)
928 return -EEXIST;
929
930 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
931 if (!start_marker_record)
932 return -ENOMEM;
933
934 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
935 if (!offload_ctx) {
936 rc = -ENOMEM;
937 goto free_marker_record;
938 }
939
940 crypto_info = &ctx->crypto_send.info;
941 if (crypto_info->version != TLS_1_2_VERSION) {
942 rc = -EOPNOTSUPP;
943 goto free_offload_ctx;
944 }
945
946 switch (crypto_info->cipher_type) {
947 case TLS_CIPHER_AES_GCM_128:
948 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
949 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
950 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
951 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
952 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
953 rec_seq =
954 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
955 break;
956 default:
957 rc = -EINVAL;
958 goto free_offload_ctx;
959 }
960
961 /* Sanity-check the rec_seq_size for stack allocations */
962 if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
963 rc = -EINVAL;
964 goto free_offload_ctx;
965 }
966
967 prot->version = crypto_info->version;
968 prot->cipher_type = crypto_info->cipher_type;
969 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
970 prot->tag_size = tag_size;
971 prot->overhead_size = prot->prepend_size + prot->tag_size;
972 prot->iv_size = iv_size;
973 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
974 GFP_KERNEL);
975 if (!ctx->tx.iv) {
976 rc = -ENOMEM;
977 goto free_offload_ctx;
978 }
979
980 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
981
982 prot->rec_seq_size = rec_seq_size;
983 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
984 if (!ctx->tx.rec_seq) {
985 rc = -ENOMEM;
986 goto free_iv;
987 }
988
989 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
990 if (rc)
991 goto free_rec_seq;
992
993 /* start at rec_seq - 1 to account for the start marker record */
994 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
995 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
996
997 start_marker_record->end_seq = tcp_sk(sk)->write_seq;
998 start_marker_record->len = 0;
999 start_marker_record->num_frags = 0;
1000
1001 INIT_LIST_HEAD(&offload_ctx->records_list);
1002 list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1003 spin_lock_init(&offload_ctx->lock);
1004 sg_init_table(offload_ctx->sg_tx_data,
1005 ARRAY_SIZE(offload_ctx->sg_tx_data));
1006
1007 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1008 ctx->push_pending_record = tls_device_push_pending_record;
1009
1010 /* TLS offload is greatly simplified if we don't send
1011 * SKBs where only part of the payload needs to be encrypted.
1012 * So mark the last skb in the write queue as end of record.
1013 */
1014 skb = tcp_write_queue_tail(sk);
1015 if (skb)
1016 TCP_SKB_CB(skb)->eor = 1;
1017
1018 netdev = get_netdev_for_sock(sk);
1019 if (!netdev) {
1020 pr_err_ratelimited("%s: netdev not found\n", __func__);
1021 rc = -EINVAL;
1022 goto disable_cad;
1023 }
1024
1025 if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1026 rc = -EOPNOTSUPP;
1027 goto release_netdev;
1028 }
1029
1030 /* Avoid offloading if the device is down
1031 * We don't want to offload new flows after
1032 * the NETDEV_DOWN event
1033 *
1034 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1035 * handler thus protecting from the device going down before
1036 * ctx was added to tls_device_list.
1037 */
1038 down_read(&device_offload_lock);
1039 if (!(netdev->flags & IFF_UP)) {
1040 rc = -EINVAL;
1041 goto release_lock;
1042 }
1043
1044 ctx->priv_ctx_tx = offload_ctx;
1045 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1046 &ctx->crypto_send.info,
1047 tcp_sk(sk)->write_seq);
1048 if (rc)
1049 goto release_lock;
1050
1051 tls_device_attach(ctx, sk, netdev);
1052 up_read(&device_offload_lock);
1053
1054 /* following this assignment tls_is_sk_tx_device_offloaded
1055 * will return true and the context might be accessed
1056 * by the netdev's xmit function.
1057 */
1058 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1059 dev_put(netdev);
1060
1061 return 0;
1062
1063 release_lock:
1064 up_read(&device_offload_lock);
1065 release_netdev:
1066 dev_put(netdev);
1067 disable_cad:
1068 clean_acked_data_disable(inet_csk(sk));
1069 crypto_free_aead(offload_ctx->aead_send);
1070 free_rec_seq:
1071 kfree(ctx->tx.rec_seq);
1072 free_iv:
1073 kfree(ctx->tx.iv);
1074 free_offload_ctx:
1075 kfree(offload_ctx);
1076 ctx->priv_ctx_tx = NULL;
1077 free_marker_record:
1078 kfree(start_marker_record);
1079 return rc;
1080 }
1081
tls_set_device_offload_rx(struct sock * sk,struct tls_context * ctx)1082 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1083 {
1084 struct tls_offload_context_rx *context;
1085 struct net_device *netdev;
1086 int rc = 0;
1087
1088 if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1089 return -EOPNOTSUPP;
1090
1091 netdev = get_netdev_for_sock(sk);
1092 if (!netdev) {
1093 pr_err_ratelimited("%s: netdev not found\n", __func__);
1094 return -EINVAL;
1095 }
1096
1097 if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1098 rc = -EOPNOTSUPP;
1099 goto release_netdev;
1100 }
1101
1102 /* Avoid offloading if the device is down
1103 * We don't want to offload new flows after
1104 * the NETDEV_DOWN event
1105 *
1106 * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1107 * handler thus protecting from the device going down before
1108 * ctx was added to tls_device_list.
1109 */
1110 down_read(&device_offload_lock);
1111 if (!(netdev->flags & IFF_UP)) {
1112 rc = -EINVAL;
1113 goto release_lock;
1114 }
1115
1116 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1117 if (!context) {
1118 rc = -ENOMEM;
1119 goto release_lock;
1120 }
1121 context->resync_nh_reset = 1;
1122
1123 ctx->priv_ctx_rx = context;
1124 rc = tls_set_sw_offload(sk, ctx, 0);
1125 if (rc)
1126 goto release_ctx;
1127
1128 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1129 &ctx->crypto_recv.info,
1130 tcp_sk(sk)->copied_seq);
1131 if (rc)
1132 goto free_sw_resources;
1133
1134 tls_device_attach(ctx, sk, netdev);
1135 up_read(&device_offload_lock);
1136
1137 dev_put(netdev);
1138
1139 return 0;
1140
1141 free_sw_resources:
1142 up_read(&device_offload_lock);
1143 tls_sw_free_resources_rx(sk);
1144 down_read(&device_offload_lock);
1145 release_ctx:
1146 ctx->priv_ctx_rx = NULL;
1147 release_lock:
1148 up_read(&device_offload_lock);
1149 release_netdev:
1150 dev_put(netdev);
1151 return rc;
1152 }
1153
tls_device_offload_cleanup_rx(struct sock * sk)1154 void tls_device_offload_cleanup_rx(struct sock *sk)
1155 {
1156 struct tls_context *tls_ctx = tls_get_ctx(sk);
1157 struct net_device *netdev;
1158
1159 down_read(&device_offload_lock);
1160 netdev = tls_ctx->netdev;
1161 if (!netdev)
1162 goto out;
1163
1164 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1165 TLS_OFFLOAD_CTX_DIR_RX);
1166
1167 if (tls_ctx->tx_conf != TLS_HW) {
1168 dev_put(netdev);
1169 tls_ctx->netdev = NULL;
1170 } else {
1171 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1172 }
1173 out:
1174 up_read(&device_offload_lock);
1175 tls_sw_release_resources_rx(sk);
1176 }
1177
tls_device_down(struct net_device * netdev)1178 static int tls_device_down(struct net_device *netdev)
1179 {
1180 struct tls_context *ctx, *tmp;
1181 unsigned long flags;
1182 LIST_HEAD(list);
1183
1184 /* Request a write lock to block new offload attempts */
1185 down_write(&device_offload_lock);
1186
1187 spin_lock_irqsave(&tls_device_lock, flags);
1188 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1189 if (ctx->netdev != netdev ||
1190 !refcount_inc_not_zero(&ctx->refcount))
1191 continue;
1192
1193 list_move(&ctx->list, &list);
1194 }
1195 spin_unlock_irqrestore(&tls_device_lock, flags);
1196
1197 list_for_each_entry_safe(ctx, tmp, &list, list) {
1198 if (ctx->tx_conf == TLS_HW)
1199 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1200 TLS_OFFLOAD_CTX_DIR_TX);
1201 if (ctx->rx_conf == TLS_HW &&
1202 !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1203 netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1204 TLS_OFFLOAD_CTX_DIR_RX);
1205 WRITE_ONCE(ctx->netdev, NULL);
1206 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1207 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1208 usleep_range(10, 200);
1209 dev_put(netdev);
1210 list_del_init(&ctx->list);
1211
1212 if (refcount_dec_and_test(&ctx->refcount))
1213 tls_device_free_ctx(ctx);
1214 }
1215
1216 up_write(&device_offload_lock);
1217
1218 flush_work(&tls_device_gc_work);
1219
1220 return NOTIFY_DONE;
1221 }
1222
tls_dev_event(struct notifier_block * this,unsigned long event,void * ptr)1223 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1224 void *ptr)
1225 {
1226 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1227
1228 if (!dev->tlsdev_ops &&
1229 !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1230 return NOTIFY_DONE;
1231
1232 switch (event) {
1233 case NETDEV_REGISTER:
1234 case NETDEV_FEAT_CHANGE:
1235 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1236 !dev->tlsdev_ops->tls_dev_resync)
1237 return NOTIFY_BAD;
1238
1239 if (dev->tlsdev_ops &&
1240 dev->tlsdev_ops->tls_dev_add &&
1241 dev->tlsdev_ops->tls_dev_del)
1242 return NOTIFY_DONE;
1243 else
1244 return NOTIFY_BAD;
1245 case NETDEV_DOWN:
1246 return tls_device_down(dev);
1247 }
1248 return NOTIFY_DONE;
1249 }
1250
1251 static struct notifier_block tls_dev_notifier = {
1252 .notifier_call = tls_dev_event,
1253 };
1254
tls_device_init(void)1255 void __init tls_device_init(void)
1256 {
1257 register_netdevice_notifier(&tls_dev_notifier);
1258 }
1259
tls_device_cleanup(void)1260 void __exit tls_device_cleanup(void)
1261 {
1262 unregister_netdevice_notifier(&tls_dev_notifier);
1263 flush_work(&tls_device_gc_work);
1264 clean_acked_data_flush();
1265 }
1266