docs: Fix empty parallelism argument
[linux-2.6-microblaze.git] / net / tls / tls_device.c
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  */
46 static DECLARE_RWSEM(device_offload_lock);
47
48 static void tls_device_gc_task(struct work_struct *work);
49
50 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
51 static LIST_HEAD(tls_device_gc_list);
52 static LIST_HEAD(tls_device_list);
53 static DEFINE_SPINLOCK(tls_device_lock);
54
55 static 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
69 static 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
94 static 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 */
110 static 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
125 static 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
134 static 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
146 static 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  */
181 void 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 }
199 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
200
201 void 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
208 void 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 }
215 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
216
217 static 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
245 static 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
268 static int tls_push_record(struct sock *sk,
269                            struct tls_context *ctx,
270                            struct tls_offload_context_tx *offload_ctx,
271                            struct tls_record_info *record,
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
302 static int tls_device_record_close(struct sock *sk,
303                                    struct tls_context *ctx,
304                                    struct tls_record_info *record,
305                                    struct page_frag *pfrag,
306                                    unsigned char record_type)
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
334 static 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
359 static 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                         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
388 static 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
413 static int tls_push_data(struct sock *sk,
414                          struct iov_iter *msg_iter,
415                          size_t size, int flags,
416                          unsigned char record_type)
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;
465 handle_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) {
493 last_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
535 int 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
553 out:
554         release_sock(sk);
555         mutex_unlock(&tls_ctx->tx_lock);
556         return rc;
557 }
558
559 int 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 = kmap(page);
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         iov.iov_base = kaddr + offset;
580         iov.iov_len = size;
581         iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
582         rc = tls_push_data(sk, &msg_iter, size,
583                            flags, TLS_RECORD_TYPE_DATA);
584         kunmap(page);
585
586 out:
587         release_sock(sk);
588         mutex_unlock(&tls_ctx->tx_lock);
589         return rc;
590 }
591
592 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
593                                        u32 seq, u64 *p_record_sn)
594 {
595         u64 record_sn = context->hint_record_sn;
596         struct tls_record_info *info;
597
598         info = context->retransmit_hint;
599         if (!info ||
600             before(seq, info->end_seq - info->len)) {
601                 /* if retransmit_hint is irrelevant start
602                  * from the beggining of the list
603                  */
604                 info = list_first_entry_or_null(&context->records_list,
605                                                 struct tls_record_info, list);
606                 if (!info)
607                         return NULL;
608                 record_sn = context->unacked_record_sn;
609         }
610
611         /* We just need the _rcu for the READ_ONCE() */
612         rcu_read_lock();
613         list_for_each_entry_from_rcu(info, &context->records_list, list) {
614                 if (before(seq, info->end_seq)) {
615                         if (!context->retransmit_hint ||
616                             after(info->end_seq,
617                                   context->retransmit_hint->end_seq)) {
618                                 context->hint_record_sn = record_sn;
619                                 context->retransmit_hint = info;
620                         }
621                         *p_record_sn = record_sn;
622                         goto exit_rcu_unlock;
623                 }
624                 record_sn++;
625         }
626         info = NULL;
627
628 exit_rcu_unlock:
629         rcu_read_unlock();
630         return info;
631 }
632 EXPORT_SYMBOL(tls_get_record);
633
634 static int tls_device_push_pending_record(struct sock *sk, int flags)
635 {
636         struct iov_iter msg_iter;
637
638         iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
639         return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
640 }
641
642 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
643 {
644         if (tls_is_partially_sent_record(ctx)) {
645                 gfp_t sk_allocation = sk->sk_allocation;
646
647                 WARN_ON_ONCE(sk->sk_write_pending);
648
649                 sk->sk_allocation = GFP_ATOMIC;
650                 tls_push_partial_record(sk, ctx,
651                                         MSG_DONTWAIT | MSG_NOSIGNAL |
652                                         MSG_SENDPAGE_DECRYPTED);
653                 sk->sk_allocation = sk_allocation;
654         }
655 }
656
657 static void tls_device_resync_rx(struct tls_context *tls_ctx,
658                                  struct sock *sk, u32 seq, u8 *rcd_sn)
659 {
660         struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
661         struct net_device *netdev;
662
663         if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
664                 return;
665
666         trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
667         netdev = READ_ONCE(tls_ctx->netdev);
668         if (netdev)
669                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
670                                                    TLS_OFFLOAD_CTX_DIR_RX);
671         clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
672         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
673 }
674
675 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
676 {
677         struct tls_context *tls_ctx = tls_get_ctx(sk);
678         struct tls_offload_context_rx *rx_ctx;
679         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
680         u32 sock_data, is_req_pending;
681         struct tls_prot_info *prot;
682         s64 resync_req;
683         u32 req_seq;
684
685         if (tls_ctx->rx_conf != TLS_HW)
686                 return;
687
688         prot = &tls_ctx->prot_info;
689         rx_ctx = tls_offload_ctx_rx(tls_ctx);
690         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
691
692         switch (rx_ctx->resync_type) {
693         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
694                 resync_req = atomic64_read(&rx_ctx->resync_req);
695                 req_seq = resync_req >> 32;
696                 seq += TLS_HEADER_SIZE - 1;
697                 is_req_pending = resync_req;
698
699                 if (likely(!is_req_pending) || req_seq != seq ||
700                     !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
701                         return;
702                 break;
703         case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
704                 if (likely(!rx_ctx->resync_nh_do_now))
705                         return;
706
707                 /* head of next rec is already in, note that the sock_inq will
708                  * include the currently parsed message when called from parser
709                  */
710                 sock_data = tcp_inq(sk);
711                 if (sock_data > rcd_len) {
712                         trace_tls_device_rx_resync_nh_delay(sk, sock_data,
713                                                             rcd_len);
714                         return;
715                 }
716
717                 rx_ctx->resync_nh_do_now = 0;
718                 seq += rcd_len;
719                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
720                 break;
721         }
722
723         tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
724 }
725
726 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
727                                            struct tls_offload_context_rx *ctx,
728                                            struct sock *sk, struct sk_buff *skb)
729 {
730         struct strp_msg *rxm;
731
732         /* device will request resyncs by itself based on stream scan */
733         if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
734                 return;
735         /* already scheduled */
736         if (ctx->resync_nh_do_now)
737                 return;
738         /* seen decrypted fragments since last fully-failed record */
739         if (ctx->resync_nh_reset) {
740                 ctx->resync_nh_reset = 0;
741                 ctx->resync_nh.decrypted_failed = 1;
742                 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
743                 return;
744         }
745
746         if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
747                 return;
748
749         /* doing resync, bump the next target in case it fails */
750         if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
751                 ctx->resync_nh.decrypted_tgt *= 2;
752         else
753                 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
754
755         rxm = strp_msg(skb);
756
757         /* head of next rec is already in, parser will sync for us */
758         if (tcp_inq(sk) > rxm->full_len) {
759                 trace_tls_device_rx_resync_nh_schedule(sk);
760                 ctx->resync_nh_do_now = 1;
761         } else {
762                 struct tls_prot_info *prot = &tls_ctx->prot_info;
763                 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
764
765                 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
766                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
767
768                 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
769                                      rcd_sn);
770         }
771 }
772
773 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
774 {
775         struct strp_msg *rxm = strp_msg(skb);
776         int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
777         struct sk_buff *skb_iter, *unused;
778         struct scatterlist sg[1];
779         char *orig_buf, *buf;
780
781         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
782                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
783         if (!orig_buf)
784                 return -ENOMEM;
785         buf = orig_buf;
786
787         nsg = skb_cow_data(skb, 0, &unused);
788         if (unlikely(nsg < 0)) {
789                 err = nsg;
790                 goto free_buf;
791         }
792
793         sg_init_table(sg, 1);
794         sg_set_buf(&sg[0], buf,
795                    rxm->full_len + TLS_HEADER_SIZE +
796                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
797         err = skb_copy_bits(skb, offset, buf,
798                             TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
799         if (err)
800                 goto free_buf;
801
802         /* We are interested only in the decrypted data not the auth */
803         err = decrypt_skb(sk, skb, sg);
804         if (err != -EBADMSG)
805                 goto free_buf;
806         else
807                 err = 0;
808
809         data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
810
811         if (skb_pagelen(skb) > offset) {
812                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
813
814                 if (skb->decrypted) {
815                         err = skb_store_bits(skb, offset, buf, copy);
816                         if (err)
817                                 goto free_buf;
818                 }
819
820                 offset += copy;
821                 buf += copy;
822         }
823
824         pos = skb_pagelen(skb);
825         skb_walk_frags(skb, skb_iter) {
826                 int frag_pos;
827
828                 /* Practically all frags must belong to msg if reencrypt
829                  * is needed with current strparser and coalescing logic,
830                  * but strparser may "get optimized", so let's be safe.
831                  */
832                 if (pos + skb_iter->len <= offset)
833                         goto done_with_frag;
834                 if (pos >= data_len + rxm->offset)
835                         break;
836
837                 frag_pos = offset - pos;
838                 copy = min_t(int, skb_iter->len - frag_pos,
839                              data_len + rxm->offset - offset);
840
841                 if (skb_iter->decrypted) {
842                         err = skb_store_bits(skb_iter, frag_pos, buf, copy);
843                         if (err)
844                                 goto free_buf;
845                 }
846
847                 offset += copy;
848                 buf += copy;
849 done_with_frag:
850                 pos += skb_iter->len;
851         }
852
853 free_buf:
854         kfree(orig_buf);
855         return err;
856 }
857
858 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
859                          struct sk_buff *skb, struct strp_msg *rxm)
860 {
861         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
862         int is_decrypted = skb->decrypted;
863         int is_encrypted = !is_decrypted;
864         struct sk_buff *skb_iter;
865
866         /* Check if all the data is decrypted already */
867         skb_walk_frags(skb, skb_iter) {
868                 is_decrypted &= skb_iter->decrypted;
869                 is_encrypted &= !skb_iter->decrypted;
870         }
871
872         trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
873                                    tls_ctx->rx.rec_seq, rxm->full_len,
874                                    is_encrypted, is_decrypted);
875
876         ctx->sw.decrypted |= is_decrypted;
877
878         /* Return immediately if the record is either entirely plaintext or
879          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
880          * record.
881          */
882         if (is_decrypted) {
883                 ctx->resync_nh_reset = 1;
884                 return 0;
885         }
886         if (is_encrypted) {
887                 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
888                 return 0;
889         }
890
891         ctx->resync_nh_reset = 1;
892         return tls_device_reencrypt(sk, skb);
893 }
894
895 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
896                               struct net_device *netdev)
897 {
898         if (sk->sk_destruct != tls_device_sk_destruct) {
899                 refcount_set(&ctx->refcount, 1);
900                 dev_hold(netdev);
901                 ctx->netdev = netdev;
902                 spin_lock_irq(&tls_device_lock);
903                 list_add_tail(&ctx->list, &tls_device_list);
904                 spin_unlock_irq(&tls_device_lock);
905
906                 ctx->sk_destruct = sk->sk_destruct;
907                 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
908         }
909 }
910
911 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
912 {
913         u16 nonce_size, tag_size, iv_size, rec_seq_size;
914         struct tls_context *tls_ctx = tls_get_ctx(sk);
915         struct tls_prot_info *prot = &tls_ctx->prot_info;
916         struct tls_record_info *start_marker_record;
917         struct tls_offload_context_tx *offload_ctx;
918         struct tls_crypto_info *crypto_info;
919         struct net_device *netdev;
920         char *iv, *rec_seq;
921         struct sk_buff *skb;
922         __be64 rcd_sn;
923         int rc;
924
925         if (!ctx)
926                 return -EINVAL;
927
928         if (ctx->priv_ctx_tx)
929                 return -EEXIST;
930
931         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
932         if (!start_marker_record)
933                 return -ENOMEM;
934
935         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
936         if (!offload_ctx) {
937                 rc = -ENOMEM;
938                 goto free_marker_record;
939         }
940
941         crypto_info = &ctx->crypto_send.info;
942         if (crypto_info->version != TLS_1_2_VERSION) {
943                 rc = -EOPNOTSUPP;
944                 goto free_offload_ctx;
945         }
946
947         switch (crypto_info->cipher_type) {
948         case TLS_CIPHER_AES_GCM_128:
949                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
950                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
951                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
952                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
953                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
954                 rec_seq =
955                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
956                 break;
957         default:
958                 rc = -EINVAL;
959                 goto free_offload_ctx;
960         }
961
962         /* Sanity-check the rec_seq_size for stack allocations */
963         if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
964                 rc = -EINVAL;
965                 goto free_offload_ctx;
966         }
967
968         prot->version = crypto_info->version;
969         prot->cipher_type = crypto_info->cipher_type;
970         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
971         prot->tag_size = tag_size;
972         prot->overhead_size = prot->prepend_size + prot->tag_size;
973         prot->iv_size = iv_size;
974         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
975                              GFP_KERNEL);
976         if (!ctx->tx.iv) {
977                 rc = -ENOMEM;
978                 goto free_offload_ctx;
979         }
980
981         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
982
983         prot->rec_seq_size = rec_seq_size;
984         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
985         if (!ctx->tx.rec_seq) {
986                 rc = -ENOMEM;
987                 goto free_iv;
988         }
989
990         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
991         if (rc)
992                 goto free_rec_seq;
993
994         /* start at rec_seq - 1 to account for the start marker record */
995         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
996         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
997
998         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
999         start_marker_record->len = 0;
1000         start_marker_record->num_frags = 0;
1001
1002         INIT_LIST_HEAD(&offload_ctx->records_list);
1003         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1004         spin_lock_init(&offload_ctx->lock);
1005         sg_init_table(offload_ctx->sg_tx_data,
1006                       ARRAY_SIZE(offload_ctx->sg_tx_data));
1007
1008         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1009         ctx->push_pending_record = tls_device_push_pending_record;
1010
1011         /* TLS offload is greatly simplified if we don't send
1012          * SKBs where only part of the payload needs to be encrypted.
1013          * So mark the last skb in the write queue as end of record.
1014          */
1015         skb = tcp_write_queue_tail(sk);
1016         if (skb)
1017                 TCP_SKB_CB(skb)->eor = 1;
1018
1019         netdev = get_netdev_for_sock(sk);
1020         if (!netdev) {
1021                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1022                 rc = -EINVAL;
1023                 goto disable_cad;
1024         }
1025
1026         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1027                 rc = -EOPNOTSUPP;
1028                 goto release_netdev;
1029         }
1030
1031         /* Avoid offloading if the device is down
1032          * We don't want to offload new flows after
1033          * the NETDEV_DOWN event
1034          *
1035          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1036          * handler thus protecting from the device going down before
1037          * ctx was added to tls_device_list.
1038          */
1039         down_read(&device_offload_lock);
1040         if (!(netdev->flags & IFF_UP)) {
1041                 rc = -EINVAL;
1042                 goto release_lock;
1043         }
1044
1045         ctx->priv_ctx_tx = offload_ctx;
1046         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1047                                              &ctx->crypto_send.info,
1048                                              tcp_sk(sk)->write_seq);
1049         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1050                                      tcp_sk(sk)->write_seq, rec_seq, rc);
1051         if (rc)
1052                 goto release_lock;
1053
1054         tls_device_attach(ctx, sk, netdev);
1055         up_read(&device_offload_lock);
1056
1057         /* following this assignment tls_is_sk_tx_device_offloaded
1058          * will return true and the context might be accessed
1059          * by the netdev's xmit function.
1060          */
1061         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1062         dev_put(netdev);
1063
1064         return 0;
1065
1066 release_lock:
1067         up_read(&device_offload_lock);
1068 release_netdev:
1069         dev_put(netdev);
1070 disable_cad:
1071         clean_acked_data_disable(inet_csk(sk));
1072         crypto_free_aead(offload_ctx->aead_send);
1073 free_rec_seq:
1074         kfree(ctx->tx.rec_seq);
1075 free_iv:
1076         kfree(ctx->tx.iv);
1077 free_offload_ctx:
1078         kfree(offload_ctx);
1079         ctx->priv_ctx_tx = NULL;
1080 free_marker_record:
1081         kfree(start_marker_record);
1082         return rc;
1083 }
1084
1085 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1086 {
1087         struct tls12_crypto_info_aes_gcm_128 *info;
1088         struct tls_offload_context_rx *context;
1089         struct net_device *netdev;
1090         int rc = 0;
1091
1092         if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1093                 return -EOPNOTSUPP;
1094
1095         netdev = get_netdev_for_sock(sk);
1096         if (!netdev) {
1097                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1098                 return -EINVAL;
1099         }
1100
1101         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1102                 rc = -EOPNOTSUPP;
1103                 goto release_netdev;
1104         }
1105
1106         /* Avoid offloading if the device is down
1107          * We don't want to offload new flows after
1108          * the NETDEV_DOWN event
1109          *
1110          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1111          * handler thus protecting from the device going down before
1112          * ctx was added to tls_device_list.
1113          */
1114         down_read(&device_offload_lock);
1115         if (!(netdev->flags & IFF_UP)) {
1116                 rc = -EINVAL;
1117                 goto release_lock;
1118         }
1119
1120         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1121         if (!context) {
1122                 rc = -ENOMEM;
1123                 goto release_lock;
1124         }
1125         context->resync_nh_reset = 1;
1126
1127         ctx->priv_ctx_rx = context;
1128         rc = tls_set_sw_offload(sk, ctx, 0);
1129         if (rc)
1130                 goto release_ctx;
1131
1132         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1133                                              &ctx->crypto_recv.info,
1134                                              tcp_sk(sk)->copied_seq);
1135         info = (void *)&ctx->crypto_recv.info;
1136         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1137                                      tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1138         if (rc)
1139                 goto free_sw_resources;
1140
1141         tls_device_attach(ctx, sk, netdev);
1142         up_read(&device_offload_lock);
1143
1144         dev_put(netdev);
1145
1146         return 0;
1147
1148 free_sw_resources:
1149         up_read(&device_offload_lock);
1150         tls_sw_free_resources_rx(sk);
1151         down_read(&device_offload_lock);
1152 release_ctx:
1153         ctx->priv_ctx_rx = NULL;
1154 release_lock:
1155         up_read(&device_offload_lock);
1156 release_netdev:
1157         dev_put(netdev);
1158         return rc;
1159 }
1160
1161 void tls_device_offload_cleanup_rx(struct sock *sk)
1162 {
1163         struct tls_context *tls_ctx = tls_get_ctx(sk);
1164         struct net_device *netdev;
1165
1166         down_read(&device_offload_lock);
1167         netdev = tls_ctx->netdev;
1168         if (!netdev)
1169                 goto out;
1170
1171         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1172                                         TLS_OFFLOAD_CTX_DIR_RX);
1173
1174         if (tls_ctx->tx_conf != TLS_HW) {
1175                 dev_put(netdev);
1176                 tls_ctx->netdev = NULL;
1177         }
1178 out:
1179         up_read(&device_offload_lock);
1180         tls_sw_release_resources_rx(sk);
1181 }
1182
1183 static int tls_device_down(struct net_device *netdev)
1184 {
1185         struct tls_context *ctx, *tmp;
1186         unsigned long flags;
1187         LIST_HEAD(list);
1188
1189         /* Request a write lock to block new offload attempts */
1190         down_write(&device_offload_lock);
1191
1192         spin_lock_irqsave(&tls_device_lock, flags);
1193         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1194                 if (ctx->netdev != netdev ||
1195                     !refcount_inc_not_zero(&ctx->refcount))
1196                         continue;
1197
1198                 list_move(&ctx->list, &list);
1199         }
1200         spin_unlock_irqrestore(&tls_device_lock, flags);
1201
1202         list_for_each_entry_safe(ctx, tmp, &list, list) {
1203                 if (ctx->tx_conf == TLS_HW)
1204                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1205                                                         TLS_OFFLOAD_CTX_DIR_TX);
1206                 if (ctx->rx_conf == TLS_HW)
1207                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1208                                                         TLS_OFFLOAD_CTX_DIR_RX);
1209                 WRITE_ONCE(ctx->netdev, NULL);
1210                 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1211                 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1212                         usleep_range(10, 200);
1213                 dev_put(netdev);
1214                 list_del_init(&ctx->list);
1215
1216                 if (refcount_dec_and_test(&ctx->refcount))
1217                         tls_device_free_ctx(ctx);
1218         }
1219
1220         up_write(&device_offload_lock);
1221
1222         flush_work(&tls_device_gc_work);
1223
1224         return NOTIFY_DONE;
1225 }
1226
1227 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1228                          void *ptr)
1229 {
1230         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1231
1232         if (!dev->tlsdev_ops &&
1233             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1234                 return NOTIFY_DONE;
1235
1236         switch (event) {
1237         case NETDEV_REGISTER:
1238         case NETDEV_FEAT_CHANGE:
1239                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1240                     !dev->tlsdev_ops->tls_dev_resync)
1241                         return NOTIFY_BAD;
1242
1243                 if  (dev->tlsdev_ops &&
1244                      dev->tlsdev_ops->tls_dev_add &&
1245                      dev->tlsdev_ops->tls_dev_del)
1246                         return NOTIFY_DONE;
1247                 else
1248                         return NOTIFY_BAD;
1249         case NETDEV_DOWN:
1250                 return tls_device_down(dev);
1251         }
1252         return NOTIFY_DONE;
1253 }
1254
1255 static struct notifier_block tls_dev_notifier = {
1256         .notifier_call  = tls_dev_event,
1257 };
1258
1259 void __init tls_device_init(void)
1260 {
1261         register_netdevice_notifier(&tls_dev_notifier);
1262 }
1263
1264 void __exit tls_device_cleanup(void)
1265 {
1266         unregister_netdevice_notifier(&tls_dev_notifier);
1267         flush_work(&tls_device_gc_work);
1268         clean_acked_data_flush();
1269 }