1 /* 2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved. 3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. 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 52 struct 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 84 enum { 85 TLS_BASE, 86 TLS_SW, 87 TLS_HW, 88 TLS_HW_RECORD, 89 TLS_NUM_CONFIG, 90 }; 91 92 struct tx_work { 93 struct delayed_work work; 94 struct sock *sk; 95 }; 96 97 struct tls_sw_context_tx { 98 struct crypto_aead *aead_send; 99 struct crypto_wait async_wait; 100 struct tx_work tx_work; 101 struct tls_rec *open_rec; 102 struct list_head tx_list; 103 atomic_t encrypt_pending; 104 u8 async_capable:1; 105 106 #define BIT_TX_SCHEDULED 0 107 #define BIT_TX_CLOSING 1 108 unsigned long tx_bitmask; 109 }; 110 111 struct tls_strparser { 112 struct sock *sk; 113 114 /* Bitfield word and msg_ready are serialized by the lower 115 * socket lock; BH and worker contexts both acquire it. 116 */ 117 u32 mark : 8; 118 u32 stopped : 1; 119 u32 copy_mode : 1; 120 u32 mixed_decrypted : 1; 121 122 u32 msg_announced : 1; 123 124 bool msg_ready; 125 126 struct strp_msg stm; 127 128 struct sk_buff *anchor; 129 struct work_struct work; 130 }; 131 132 struct tls_sw_context_rx { 133 struct crypto_aead *aead_recv; 134 struct crypto_wait async_wait; 135 struct sk_buff_head rx_list; /* list of decrypted 'data' records */ 136 void (*saved_data_ready)(struct sock *sk); 137 138 u8 reader_present; 139 u8 async_capable:1; 140 u8 zc_capable:1; 141 u8 reader_contended:1; 142 bool key_update_pending; 143 144 struct tls_strparser strp; 145 146 atomic_t decrypt_pending; 147 struct sk_buff_head async_hold; 148 struct wait_queue_head wq; 149 }; 150 151 struct tls_record_info { 152 struct list_head list; 153 u32 end_seq; 154 int len; 155 int num_frags; 156 skb_frag_t frags[MAX_SKB_FRAGS]; 157 }; 158 159 #define TLS_DRIVER_STATE_SIZE_TX 16 160 struct tls_offload_context_tx { 161 struct crypto_aead *aead_send; 162 spinlock_t lock; /* protects records list */ 163 struct list_head records_list; 164 struct tls_record_info *open_record; 165 struct tls_record_info *retransmit_hint; 166 u64 hint_record_sn; 167 u64 unacked_record_sn; 168 169 struct scatterlist sg_tx_data[MAX_SKB_FRAGS]; 170 void (*sk_destruct)(struct sock *sk); 171 struct work_struct destruct_work; 172 struct tls_context *ctx; 173 /* The TLS layer reserves room for driver specific state 174 * Currently the belief is that there is not enough 175 * driver specific state to justify another layer of indirection 176 */ 177 u8 driver_state[TLS_DRIVER_STATE_SIZE_TX] __aligned(8); 178 }; 179 180 enum tls_context_flags { 181 /* tls_device_down was called after the netdev went down, device state 182 * was released, and kTLS works in software, even though rx_conf is 183 * still TLS_HW (needed for transition). 184 */ 185 TLS_RX_DEV_DEGRADED = 0, 186 /* Unlike RX where resync is driven entirely by the core in TX only 187 * the driver knows when things went out of sync, so we need the flag 188 * to be atomic. 189 */ 190 TLS_TX_SYNC_SCHED = 1, 191 /* tls_dev_del was called for the RX side, device state was released, 192 * but tls_ctx->netdev might still be kept, because TX-side driver 193 * resources might not be released yet. Used to prevent the second 194 * tls_dev_del call in tls_device_down if it happens simultaneously. 195 */ 196 TLS_RX_DEV_CLOSED = 2, 197 }; 198 199 struct cipher_context { 200 char iv[TLS_MAX_IV_SIZE + TLS_MAX_SALT_SIZE]; 201 char rec_seq[TLS_MAX_REC_SEQ_SIZE]; 202 }; 203 204 union tls_crypto_context { 205 struct tls_crypto_info info; 206 union { 207 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128; 208 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256; 209 struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305; 210 struct tls12_crypto_info_sm4_gcm sm4_gcm; 211 struct tls12_crypto_info_sm4_ccm sm4_ccm; 212 }; 213 }; 214 215 struct tls_prot_info { 216 u16 version; 217 u16 cipher_type; 218 u16 prepend_size; 219 u16 tag_size; 220 u16 overhead_size; 221 u16 iv_size; 222 u16 salt_size; 223 u16 rec_seq_size; 224 u16 aad_size; 225 u16 tail_size; 226 }; 227 228 struct tls_context { 229 /* read-only cache line */ 230 struct tls_prot_info prot_info; 231 232 u8 tx_conf:3; 233 u8 rx_conf:3; 234 u8 zerocopy_sendfile:1; 235 u8 rx_no_pad:1; 236 u16 tx_max_payload_len; 237 238 int (*push_pending_record)(struct sock *sk, int flags); 239 void (*sk_write_space)(struct sock *sk); 240 241 void *priv_ctx_tx; 242 void *priv_ctx_rx; 243 244 struct net_device __rcu *netdev; 245 246 /* rw cache line */ 247 struct cipher_context tx; 248 struct cipher_context rx; 249 250 struct scatterlist *partially_sent_record; 251 u16 partially_sent_offset; 252 253 bool splicing_pages; 254 bool pending_open_record_frags; 255 256 struct mutex tx_lock; /* protects partially_sent_* fields and 257 * per-type TX fields 258 */ 259 unsigned long flags; 260 261 /* cache cold stuff */ 262 struct proto *sk_proto; 263 struct sock *sk; 264 265 void (*sk_destruct)(struct sock *sk); 266 267 union tls_crypto_context crypto_send; 268 union tls_crypto_context crypto_recv; 269 270 struct list_head list; 271 refcount_t refcount; 272 struct rcu_head rcu; 273 }; 274 275 enum tls_offload_ctx_dir { 276 TLS_OFFLOAD_CTX_DIR_RX, 277 TLS_OFFLOAD_CTX_DIR_TX, 278 }; 279 280 struct tlsdev_ops { 281 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk, 282 enum tls_offload_ctx_dir direction, 283 struct tls_crypto_info *crypto_info, 284 u32 start_offload_tcp_sn); 285 void (*tls_dev_del)(struct net_device *netdev, 286 struct tls_context *ctx, 287 enum tls_offload_ctx_dir direction); 288 int (*tls_dev_resync)(struct net_device *netdev, 289 struct sock *sk, u32 seq, u8 *rcd_sn, 290 enum tls_offload_ctx_dir direction); 291 }; 292 293 enum tls_offload_sync_type { 294 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0, 295 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1, 296 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2, 297 }; 298 299 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2 300 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128 301 302 #define TLS_DEVICE_RESYNC_ASYNC_LOGMAX 13 303 struct tls_offload_resync_async { 304 atomic64_t req; 305 u16 loglen; 306 u16 rcd_delta; 307 u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX]; 308 }; 309 310 #define TLS_DRIVER_STATE_SIZE_RX 8 311 struct tls_offload_context_rx { 312 /* sw must be the first member of tls_offload_context_rx */ 313 struct tls_sw_context_rx sw; 314 enum tls_offload_sync_type resync_type; 315 /* this member is set regardless of resync_type, to avoid branches */ 316 u8 resync_nh_reset:1; 317 /* CORE_NEXT_HINT-only member, but use the hole here */ 318 u8 resync_nh_do_now:1; 319 union { 320 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */ 321 struct { 322 atomic64_t resync_req; 323 }; 324 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */ 325 struct { 326 u32 decrypted_failed; 327 u32 decrypted_tgt; 328 } resync_nh; 329 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC */ 330 struct { 331 struct tls_offload_resync_async *resync_async; 332 }; 333 }; 334 /* The TLS layer reserves room for driver specific state 335 * Currently the belief is that there is not enough 336 * driver specific state to justify another layer of indirection 337 */ 338 u8 driver_state[TLS_DRIVER_STATE_SIZE_RX] __aligned(8); 339 }; 340 341 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context, 342 u32 seq, u64 *p_record_sn); 343 344 static inline bool tls_record_is_start_marker(struct tls_record_info *rec) 345 { 346 return rec->len == 0; 347 } 348 349 static inline u32 tls_record_start_seq(struct tls_record_info *rec) 350 { 351 return rec->end_seq - rec->len; 352 } 353 354 struct sk_buff * 355 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev, 356 struct sk_buff *skb); 357 struct sk_buff * 358 tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev, 359 struct sk_buff *skb); 360 361 static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb) 362 { 363 #ifdef CONFIG_TLS_DEVICE 364 struct sock *sk = skb->sk; 365 366 return sk && sk_fullsock(sk) && 367 (smp_load_acquire(&sk->sk_validate_xmit_skb) == 368 &tls_validate_xmit_skb); 369 #else 370 return false; 371 #endif 372 } 373 374 static inline struct tls_context *tls_get_ctx(const struct sock *sk) 375 { 376 const struct inet_connection_sock *icsk = inet_csk(sk); 377 378 /* Use RCU on icsk_ulp_data only for sock diag code, 379 * TLS data path doesn't need rcu_dereference(). 380 */ 381 return (__force void *)icsk->icsk_ulp_data; 382 } 383 384 static inline struct tls_sw_context_rx *tls_sw_ctx_rx( 385 const struct tls_context *tls_ctx) 386 { 387 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx; 388 } 389 390 static inline struct tls_sw_context_tx *tls_sw_ctx_tx( 391 const struct tls_context *tls_ctx) 392 { 393 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx; 394 } 395 396 static inline struct tls_offload_context_tx * 397 tls_offload_ctx_tx(const struct tls_context *tls_ctx) 398 { 399 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx; 400 } 401 402 static inline bool tls_sw_has_ctx_tx(const struct sock *sk) 403 { 404 struct tls_context *ctx; 405 406 if (!sk_is_inet(sk) || !inet_test_bit(IS_ICSK, sk)) 407 return false; 408 409 ctx = tls_get_ctx(sk); 410 if (!ctx) 411 return false; 412 return !!tls_sw_ctx_tx(ctx); 413 } 414 415 static inline bool tls_sw_has_ctx_rx(const struct sock *sk) 416 { 417 struct tls_context *ctx; 418 419 if (!sk_is_inet(sk) || !inet_test_bit(IS_ICSK, sk)) 420 return false; 421 422 ctx = tls_get_ctx(sk); 423 if (!ctx) 424 return false; 425 return !!tls_sw_ctx_rx(ctx); 426 } 427 428 static inline struct tls_offload_context_rx * 429 tls_offload_ctx_rx(const struct tls_context *tls_ctx) 430 { 431 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx; 432 } 433 434 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx, 435 enum tls_offload_ctx_dir direction) 436 { 437 if (direction == TLS_OFFLOAD_CTX_DIR_TX) 438 return tls_offload_ctx_tx(tls_ctx)->driver_state; 439 else 440 return tls_offload_ctx_rx(tls_ctx)->driver_state; 441 } 442 443 static inline void * 444 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction) 445 { 446 return __tls_driver_ctx(tls_get_ctx(sk), direction); 447 } 448 449 #define RESYNC_REQ BIT(0) 450 #define RESYNC_REQ_ASYNC BIT(1) 451 /* The TLS context is valid until sk_destruct is called */ 452 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq) 453 { 454 struct tls_context *tls_ctx = tls_get_ctx(sk); 455 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx); 456 457 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ); 458 } 459 460 /* Log all TLS record header TCP sequences in [seq, seq+len] */ 461 static inline void 462 tls_offload_rx_resync_async_request_start(struct tls_offload_resync_async *resync_async, 463 __be32 seq, u16 len) 464 { 465 atomic64_set(&resync_async->req, ((u64)ntohl(seq) << 32) | 466 ((u64)len << 16) | RESYNC_REQ | RESYNC_REQ_ASYNC); 467 resync_async->loglen = 0; 468 resync_async->rcd_delta = 0; 469 } 470 471 static inline void 472 tls_offload_rx_resync_async_request_end(struct tls_offload_resync_async *resync_async, 473 __be32 seq) 474 { 475 atomic64_set(&resync_async->req, ((u64)ntohl(seq) << 32) | RESYNC_REQ); 476 } 477 478 static inline void 479 tls_offload_rx_resync_async_request_cancel(struct tls_offload_resync_async *resync_async) 480 { 481 atomic64_set(&resync_async->req, 0); 482 } 483 484 static inline void 485 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type) 486 { 487 struct tls_context *tls_ctx = tls_get_ctx(sk); 488 489 tls_offload_ctx_rx(tls_ctx)->resync_type = type; 490 } 491 492 /* Driver's seq tracking has to be disabled until resync succeeded */ 493 static inline bool tls_offload_tx_resync_pending(struct sock *sk) 494 { 495 struct tls_context *tls_ctx = tls_get_ctx(sk); 496 bool ret; 497 498 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags); 499 smp_mb__after_atomic(); 500 return ret; 501 } 502 503 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb); 504 505 #ifdef CONFIG_TLS_DEVICE 506 void tls_device_sk_destruct(struct sock *sk); 507 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq); 508 509 static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk) 510 { 511 if (!sk_fullsock(sk) || 512 smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct) 513 return false; 514 return tls_get_ctx(sk)->rx_conf == TLS_HW; 515 } 516 #endif 517 #endif /* _TLS_OFFLOAD_H */ 518