projects/linux·blob master · refs

516 lines · 13.6 KiB

1/*
2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <[email protected]>. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#ifndef _TLS_OFFLOAD_H
35#define _TLS_OFFLOAD_H
36
37#include <linux/types.h>
38#include <asm/byteorder.h>
39#include <linux/crypto.h>
40#include <linux/socket.h>
41#include <linux/tcp.h>
42#include <linux/mutex.h>
43#include <linux/netdevice.h>
44#include <linux/rcupdate.h>
45
46#include <net/net_namespace.h>
47#include <net/tcp.h>
48#include <net/strparser.h>
49#include <crypto/aead.h>
50#include <uapi/linux/tls.h>
51
52struct tls_rec;
53
54/* Maximum data size carried in a TLS record */
55#define TLS_MAX_PAYLOAD_SIZE ((size_t)1 << 14)
56/* Minimum record size limit as per RFC8449 */
57#define TLS_MIN_RECORD_SIZE_LIM ((size_t)1 << 6)
58
59#define TLS_HEADER_SIZE 5
60#define TLS_NONCE_OFFSET TLS_HEADER_SIZE
61
62#define TLS_CRYPTO_INFO_READY(info) ((info)->cipher_type)
63
64#define TLS_HANDSHAKE_KEYUPDATE 24 /* rfc8446 B.3: Key update */
65
66#define TLS_AAD_SPACE_SIZE 13
67
68#define TLS_MAX_IV_SIZE 16
69#define TLS_MAX_SALT_SIZE 4
70#define TLS_TAG_SIZE 16
71#define TLS_MAX_REC_SEQ_SIZE 8
72#define TLS_MAX_AAD_SIZE TLS_AAD_SPACE_SIZE
73
74/* For CCM mode, the full 16-bytes of IV is made of '4' fields of given sizes.
75 *
76 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3]
77 *
78 * The field 'length' is encoded in field 'b0' as '(length width - 1)'.
79 * Hence b0 contains (3 - 1) = 2.
80 */
81#define TLS_AES_CCM_IV_B0_BYTE 2
82#define TLS_SM4_CCM_IV_B0_BYTE 2
83
84enum {
85 TLS_BASE,
86 TLS_SW,
87 TLS_HW,
88 TLS_NUM_CONFIG,
89};
90
91struct tx_work {
92 struct delayed_work work;
93 struct sock *sk;
94};
95
96struct tls_sw_context_tx {
97 struct crypto_aead *aead_send;
98 struct crypto_wait async_wait;
99 struct tx_work tx_work;
100 struct tls_rec *open_rec;
101 struct list_head tx_list;
102 atomic_t encrypt_pending;
103 u8 async_capable:1;
104
105#define BIT_TX_SCHEDULED 0
106#define BIT_TX_CLOSING 1
107 unsigned long tx_bitmask;
108};
109
110struct tls_strparser {
111 struct sock *sk;
112
113 /* Bitfield word and msg_ready are serialized by the lower
114 * socket lock; BH and worker contexts both acquire it.
115 */
116 u32 mark : 8;
117 u32 stopped : 1;
118 u32 copy_mode : 1;
119 u32 mixed_decrypted : 1;
120
121 u32 msg_announced : 1;
122
123 bool msg_ready;
124
125 struct strp_msg stm;
126
127 struct sk_buff *anchor;
128 struct work_struct work;
129};
130
131struct tls_sw_context_rx {
132 struct crypto_aead *aead_recv;
133 struct crypto_wait async_wait;
134 struct sk_buff_head rx_list; /* list of decrypted 'data' records */
135 void (*saved_data_ready)(struct sock *sk);
136
137 u8 reader_present;
138 u8 async_capable:1;
139 u8 zc_capable:1;
140 u8 reader_contended:1;
141 bool key_update_pending;
142
143 struct tls_strparser strp;
144
145 atomic_t decrypt_pending;
146 struct sk_buff_head async_hold;
147 struct wait_queue_head wq;
148};
149
150struct tls_record_info {
151 struct list_head list;
152 u32 end_seq;
153 int len;
154 int num_frags;
155 skb_frag_t frags[MAX_SKB_FRAGS];
156};
157
158#define TLS_DRIVER_STATE_SIZE_TX 16
159struct tls_offload_context_tx {
160 struct crypto_aead *aead_send;
161 spinlock_t lock; /* protects records list */
162 struct list_head records_list;
163 struct tls_record_info *open_record;
164 struct tls_record_info *retransmit_hint;
165 u64 hint_record_sn;
166 u64 unacked_record_sn;
167
168 struct scatterlist sg_tx_data[MAX_SKB_FRAGS];
169 void (*sk_destruct)(struct sock *sk);
170 struct work_struct destruct_work;
171 struct tls_context *ctx;
172 /* The TLS layer reserves room for driver specific state
173 * Currently the belief is that there is not enough
174 * driver specific state to justify another layer of indirection
175 */
176 u8 driver_state[TLS_DRIVER_STATE_SIZE_TX] __aligned(8);
177};
178
179enum tls_context_flags {
180 /* tls_device_down was called after the netdev went down, device state
181 * was released, and kTLS works in software, even though rx_conf is
182 * still TLS_HW (needed for transition).
183 */
184 TLS_RX_DEV_DEGRADED = 0,
185 /* Unlike RX where resync is driven entirely by the core in TX only
186 * the driver knows when things went out of sync, so we need the flag
187 * to be atomic.
188 */
189 TLS_TX_SYNC_SCHED = 1,
190 /* tls_dev_del was called for the RX side, device state was released,
191 * but tls_ctx->netdev might still be kept, because TX-side driver
192 * resources might not be released yet. Used to prevent the second
193 * tls_dev_del call in tls_device_down if it happens simultaneously.
194 */
195 TLS_RX_DEV_CLOSED = 2,
196};
197
198struct cipher_context {
199 char iv[TLS_MAX_IV_SIZE + TLS_MAX_SALT_SIZE];
200 char rec_seq[TLS_MAX_REC_SEQ_SIZE];
201};
202
203union tls_crypto_context {
204 struct tls_crypto_info info;
205 union {
206 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128;
207 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256;
208 struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305;
209 struct tls12_crypto_info_sm4_gcm sm4_gcm;
210 struct tls12_crypto_info_sm4_ccm sm4_ccm;
211 };
212};
213
214struct tls_prot_info {
215 u16 version;
216 u16 cipher_type;
217 u16 prepend_size;
218 u16 tag_size;
219 u16 overhead_size;
220 u16 iv_size;
221 u16 salt_size;
222 u16 rec_seq_size;
223 u16 aad_size;
224 u16 tail_size;
225};
226
227struct tls_context {
228 /* read-only cache line */
229 struct tls_prot_info prot_info;
230
231 u8 tx_conf:3;
232 u8 rx_conf:3;
233 u8 zerocopy_sendfile:1;
234 u8 rx_no_pad:1;
235 u16 tx_max_payload_len;
236
237 int (*push_pending_record)(struct sock *sk, int flags);
238 void (*sk_write_space)(struct sock *sk);
239
240 void *priv_ctx_tx;
241 void *priv_ctx_rx;
242
243 struct net_device __rcu *netdev;
244
245 /* rw cache line */
246 struct cipher_context tx;
247 struct cipher_context rx;
248
249 struct scatterlist *partially_sent_record;
250 u16 partially_sent_offset;
251
252 bool splicing_pages;
253 bool pending_open_record_frags;
254
255 struct mutex tx_lock; /* protects partially_sent_* fields and
256 * per-type TX fields
257 */
258 unsigned long flags;
259
260 /* cache cold stuff */
261 struct proto *sk_proto;
262 struct sock *sk;
263
264 void (*sk_destruct)(struct sock *sk);
265
266 union tls_crypto_context crypto_send;
267 union tls_crypto_context crypto_recv;
268
269 struct list_head list;
270 refcount_t refcount;
271 struct rcu_head rcu;
272};
273
274enum tls_offload_ctx_dir {
275 TLS_OFFLOAD_CTX_DIR_RX,
276 TLS_OFFLOAD_CTX_DIR_TX,
277};
278
279struct tlsdev_ops {
280 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk,
281 enum tls_offload_ctx_dir direction,
282 struct tls_crypto_info *crypto_info,
283 u32 start_offload_tcp_sn);
284 void (*tls_dev_del)(struct net_device *netdev,
285 struct tls_context *ctx,
286 enum tls_offload_ctx_dir direction);
287 int (*tls_dev_resync)(struct net_device *netdev,
288 struct sock *sk, u32 seq, u8 *rcd_sn,
289 enum tls_offload_ctx_dir direction);
290};
291
292enum tls_offload_sync_type {
293 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0,
294 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1,
295 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2,
296};
297
298#define TLS_DEVICE_RESYNC_NH_START_IVAL 2
299#define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128
300
301#define TLS_DEVICE_RESYNC_ASYNC_LOGMAX 13
302struct tls_offload_resync_async {
303 atomic64_t req;
304 u16 loglen;
305 u16 rcd_delta;
306 u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX];
307};
308
309#define TLS_DRIVER_STATE_SIZE_RX 8
310struct tls_offload_context_rx {
311 /* sw must be the first member of tls_offload_context_rx */
312 struct tls_sw_context_rx sw;
313 enum tls_offload_sync_type resync_type;
314 /* this member is set regardless of resync_type, to avoid branches */
315 u8 resync_nh_reset:1;
316 /* CORE_NEXT_HINT-only member, but use the hole here */
317 u8 resync_nh_do_now:1;
318 union {
319 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */
320 struct {
321 atomic64_t resync_req;
322 };
323 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */
324 struct {
325 u32 decrypted_failed;
326 u32 decrypted_tgt;
327 } resync_nh;
328 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC */
329 struct {
330 struct tls_offload_resync_async *resync_async;
331 };
332 };
333 /* The TLS layer reserves room for driver specific state
334 * Currently the belief is that there is not enough
335 * driver specific state to justify another layer of indirection
336 */
337 u8 driver_state[TLS_DRIVER_STATE_SIZE_RX] __aligned(8);
338};
339
340struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
341 u32 seq, u64 *p_record_sn);
342
343static inline bool tls_record_is_start_marker(struct tls_record_info *rec)
344{
345 return rec->len == 0;
346}
347
348static inline u32 tls_record_start_seq(struct tls_record_info *rec)
349{
350 return rec->end_seq - rec->len;
351}
352
353struct sk_buff *
354tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
355 struct sk_buff *skb);
356struct sk_buff *
357tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev,
358 struct sk_buff *skb);
359
360static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb)
361{
362#ifdef CONFIG_TLS_DEVICE
363 struct sock *sk = skb->sk;
364
365 return sk && sk_fullsock(sk) &&
366 (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
367 &tls_validate_xmit_skb);
368#else
369 return false;
370#endif
371}
372
373static inline struct tls_context *tls_get_ctx(const struct sock *sk)
374{
375 const struct inet_connection_sock *icsk = inet_csk(sk);
376
377 /* Use RCU on icsk_ulp_data only for sock diag code,
378 * TLS data path doesn't need rcu_dereference().
379 */
380 return (__force void *)icsk->icsk_ulp_data;
381}
382
383static inline struct tls_sw_context_rx *tls_sw_ctx_rx(
384 const struct tls_context *tls_ctx)
385{
386 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx;
387}
388
389static inline struct tls_sw_context_tx *tls_sw_ctx_tx(
390 const struct tls_context *tls_ctx)
391{
392 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx;
393}
394
395static inline struct tls_offload_context_tx *
396tls_offload_ctx_tx(const struct tls_context *tls_ctx)
397{
398 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx;
399}
400
401static inline bool tls_sw_has_ctx_tx(const struct sock *sk)
402{
403 struct tls_context *ctx;
404
405 if (!sk_is_inet(sk) || !inet_test_bit(IS_ICSK, sk))
406 return false;
407
408 ctx = tls_get_ctx(sk);
409 if (!ctx)
410 return false;
411 return !!tls_sw_ctx_tx(ctx);
412}
413
414static inline bool tls_sw_has_ctx_rx(const struct sock *sk)
415{
416 struct tls_context *ctx;
417
418 if (!sk_is_inet(sk) || !inet_test_bit(IS_ICSK, sk))
419 return false;
420
421 ctx = tls_get_ctx(sk);
422 if (!ctx)
423 return false;
424 return !!tls_sw_ctx_rx(ctx);
425}
426
427static inline struct tls_offload_context_rx *
428tls_offload_ctx_rx(const struct tls_context *tls_ctx)
429{
430 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx;
431}
432
433static inline void *__tls_driver_ctx(struct tls_context *tls_ctx,
434 enum tls_offload_ctx_dir direction)
435{
436 if (direction == TLS_OFFLOAD_CTX_DIR_TX)
437 return tls_offload_ctx_tx(tls_ctx)->driver_state;
438 else
439 return tls_offload_ctx_rx(tls_ctx)->driver_state;
440}
441
442static inline void *
443tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction)
444{
445 return __tls_driver_ctx(tls_get_ctx(sk), direction);
446}
447
448#define RESYNC_REQ BIT(0)
449#define RESYNC_REQ_ASYNC BIT(1)
450/* The TLS context is valid until sk_destruct is called */
451static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
452{
453 struct tls_context *tls_ctx = tls_get_ctx(sk);
454 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
455
456 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ);
457}
458
459/* Log all TLS record header TCP sequences in [seq, seq+len] */
460static inline void
461tls_offload_rx_resync_async_request_start(struct tls_offload_resync_async *resync_async,
462 __be32 seq, u16 len)
463{
464 atomic64_set(&resync_async->req, ((u64)ntohl(seq) << 32) |
465 ((u64)len << 16) | RESYNC_REQ | RESYNC_REQ_ASYNC);
466 resync_async->loglen = 0;
467 resync_async->rcd_delta = 0;
468}
469
470static inline void
471tls_offload_rx_resync_async_request_end(struct tls_offload_resync_async *resync_async,
472 __be32 seq)
473{
474 atomic64_set(&resync_async->req, ((u64)ntohl(seq) << 32) | RESYNC_REQ);
475}
476
477static inline void
478tls_offload_rx_resync_async_request_cancel(struct tls_offload_resync_async *resync_async)
479{
480 atomic64_set(&resync_async->req, 0);
481}
482
483static inline void
484tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type)
485{
486 struct tls_context *tls_ctx = tls_get_ctx(sk);
487
488 tls_offload_ctx_rx(tls_ctx)->resync_type = type;
489}
490
491/* Driver's seq tracking has to be disabled until resync succeeded */
492static inline bool tls_offload_tx_resync_pending(struct sock *sk)
493{
494 struct tls_context *tls_ctx = tls_get_ctx(sk);
495 bool ret;
496
497 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
498 smp_mb__after_atomic();
499 return ret;
500}
501
502struct sk_buff *tls_encrypt_skb(struct sk_buff *skb);
503
504#ifdef CONFIG_TLS_DEVICE
505void tls_device_sk_destruct(struct sock *sk);
506void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq);
507
508static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk)
509{
510 if (!sk_fullsock(sk) ||
511 smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct)
512 return false;
513 return tls_get_ctx(sk)->rx_conf == TLS_HW;
514}
515#endif
516#endif /* _TLS_OFFLOAD_H */