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 struct tls_cipher_size_desc { 55 unsigned int iv; 56 unsigned int key; 57 unsigned int salt; 58 unsigned int tag; 59 unsigned int rec_seq; 60 }; 61 62 extern const struct tls_cipher_size_desc tls_cipher_size_desc[]; 63 64 /* Maximum data size carried in a TLS record */ 65 #define TLS_MAX_PAYLOAD_SIZE ((size_t)1 << 14) 66 67 #define TLS_HEADER_SIZE 5 68 #define TLS_NONCE_OFFSET TLS_HEADER_SIZE 69 70 #define TLS_CRYPTO_INFO_READY(info) ((info)->cipher_type) 71 72 #define TLS_RECORD_TYPE_ALERT 0x15 73 #define TLS_RECORD_TYPE_HANDSHAKE 0x16 74 #define TLS_RECORD_TYPE_DATA 0x17 75 76 #define TLS_AAD_SPACE_SIZE 13 77 78 #define MAX_IV_SIZE 16 79 #define TLS_TAG_SIZE 16 80 #define TLS_MAX_REC_SEQ_SIZE 8 81 #define TLS_MAX_AAD_SIZE TLS_AAD_SPACE_SIZE 82 83 /* For CCM mode, the full 16-bytes of IV is made of '4' fields of given sizes. 84 * 85 * IV[16] = b0[1] || implicit nonce[4] || explicit nonce[8] || length[3] 86 * 87 * The field 'length' is encoded in field 'b0' as '(length width - 1)'. 88 * Hence b0 contains (3 - 1) = 2. 89 */ 90 #define TLS_AES_CCM_IV_B0_BYTE 2 91 #define TLS_SM4_CCM_IV_B0_BYTE 2 92 93 enum { 94 TLS_BASE, 95 TLS_SW, 96 TLS_HW, 97 TLS_HW_RECORD, 98 TLS_NUM_CONFIG, 99 }; 100 101 struct tx_work { 102 struct delayed_work work; 103 struct sock *sk; 104 }; 105 106 struct tls_sw_context_tx { 107 struct crypto_aead *aead_send; 108 struct crypto_wait async_wait; 109 struct tx_work tx_work; 110 struct tls_rec *open_rec; 111 struct list_head tx_list; 112 atomic_t encrypt_pending; 113 /* protect crypto_wait with encrypt_pending */ 114 spinlock_t encrypt_compl_lock; 115 int async_notify; 116 u8 async_capable:1; 117 118 #define BIT_TX_SCHEDULED 0 119 #define BIT_TX_CLOSING 1 120 unsigned long tx_bitmask; 121 }; 122 123 struct tls_strparser { 124 struct sock *sk; 125 126 u32 mark : 8; 127 u32 stopped : 1; 128 u32 copy_mode : 1; 129 u32 mixed_decrypted : 1; 130 u32 msg_ready : 1; 131 132 struct strp_msg stm; 133 134 struct sk_buff *anchor; 135 struct work_struct work; 136 }; 137 138 struct tls_sw_context_rx { 139 struct crypto_aead *aead_recv; 140 struct crypto_wait async_wait; 141 struct sk_buff_head rx_list; /* list of decrypted 'data' records */ 142 void (*saved_data_ready)(struct sock *sk); 143 144 u8 reader_present; 145 u8 async_capable:1; 146 u8 zc_capable:1; 147 u8 reader_contended:1; 148 149 struct tls_strparser strp; 150 151 atomic_t decrypt_pending; 152 /* protect crypto_wait with decrypt_pending*/ 153 spinlock_t decrypt_compl_lock; 154 struct sk_buff_head async_hold; 155 struct wait_queue_head wq; 156 }; 157 158 struct tls_record_info { 159 struct list_head list; 160 u32 end_seq; 161 int len; 162 int num_frags; 163 skb_frag_t frags[MAX_SKB_FRAGS]; 164 }; 165 166 struct tls_offload_context_tx { 167 struct crypto_aead *aead_send; 168 spinlock_t lock; /* protects records list */ 169 struct list_head records_list; 170 struct tls_record_info *open_record; 171 struct tls_record_info *retransmit_hint; 172 u64 hint_record_sn; 173 u64 unacked_record_sn; 174 175 struct scatterlist sg_tx_data[MAX_SKB_FRAGS]; 176 void (*sk_destruct)(struct sock *sk); 177 struct work_struct destruct_work; 178 struct tls_context *ctx; 179 u8 driver_state[] __aligned(8); 180 /* The TLS layer reserves room for driver specific state 181 * Currently the belief is that there is not enough 182 * driver specific state to justify another layer of indirection 183 */ 184 #define TLS_DRIVER_STATE_SIZE_TX 16 185 }; 186 187 #define TLS_OFFLOAD_CONTEXT_SIZE_TX \ 188 (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX) 189 190 enum tls_context_flags { 191 /* tls_device_down was called after the netdev went down, device state 192 * was released, and kTLS works in software, even though rx_conf is 193 * still TLS_HW (needed for transition). 194 */ 195 TLS_RX_DEV_DEGRADED = 0, 196 /* Unlike RX where resync is driven entirely by the core in TX only 197 * the driver knows when things went out of sync, so we need the flag 198 * to be atomic. 199 */ 200 TLS_TX_SYNC_SCHED = 1, 201 /* tls_dev_del was called for the RX side, device state was released, 202 * but tls_ctx->netdev might still be kept, because TX-side driver 203 * resources might not be released yet. Used to prevent the second 204 * tls_dev_del call in tls_device_down if it happens simultaneously. 205 */ 206 TLS_RX_DEV_CLOSED = 2, 207 }; 208 209 struct cipher_context { 210 char *iv; 211 char *rec_seq; 212 }; 213 214 union tls_crypto_context { 215 struct tls_crypto_info info; 216 union { 217 struct tls12_crypto_info_aes_gcm_128 aes_gcm_128; 218 struct tls12_crypto_info_aes_gcm_256 aes_gcm_256; 219 struct tls12_crypto_info_chacha20_poly1305 chacha20_poly1305; 220 struct tls12_crypto_info_sm4_gcm sm4_gcm; 221 struct tls12_crypto_info_sm4_ccm sm4_ccm; 222 }; 223 }; 224 225 struct tls_prot_info { 226 u16 version; 227 u16 cipher_type; 228 u16 prepend_size; 229 u16 tag_size; 230 u16 overhead_size; 231 u16 iv_size; 232 u16 salt_size; 233 u16 rec_seq_size; 234 u16 aad_size; 235 u16 tail_size; 236 }; 237 238 struct tls_context { 239 /* read-only cache line */ 240 struct tls_prot_info prot_info; 241 242 u8 tx_conf:3; 243 u8 rx_conf:3; 244 u8 zerocopy_sendfile:1; 245 u8 rx_no_pad:1; 246 247 int (*push_pending_record)(struct sock *sk, int flags); 248 void (*sk_write_space)(struct sock *sk); 249 250 void *priv_ctx_tx; 251 void *priv_ctx_rx; 252 253 struct net_device __rcu *netdev; 254 255 /* rw cache line */ 256 struct cipher_context tx; 257 struct cipher_context rx; 258 259 struct scatterlist *partially_sent_record; 260 u16 partially_sent_offset; 261 262 bool splicing_pages; 263 bool pending_open_record_frags; 264 265 struct mutex tx_lock; /* protects partially_sent_* fields and 266 * per-type TX fields 267 */ 268 unsigned long flags; 269 270 /* cache cold stuff */ 271 struct proto *sk_proto; 272 struct sock *sk; 273 274 void (*sk_destruct)(struct sock *sk); 275 276 union tls_crypto_context crypto_send; 277 union tls_crypto_context crypto_recv; 278 279 struct list_head list; 280 refcount_t refcount; 281 struct rcu_head rcu; 282 }; 283 284 enum tls_offload_ctx_dir { 285 TLS_OFFLOAD_CTX_DIR_RX, 286 TLS_OFFLOAD_CTX_DIR_TX, 287 }; 288 289 struct tlsdev_ops { 290 int (*tls_dev_add)(struct net_device *netdev, struct sock *sk, 291 enum tls_offload_ctx_dir direction, 292 struct tls_crypto_info *crypto_info, 293 u32 start_offload_tcp_sn); 294 void (*tls_dev_del)(struct net_device *netdev, 295 struct tls_context *ctx, 296 enum tls_offload_ctx_dir direction); 297 int (*tls_dev_resync)(struct net_device *netdev, 298 struct sock *sk, u32 seq, u8 *rcd_sn, 299 enum tls_offload_ctx_dir direction); 300 }; 301 302 enum tls_offload_sync_type { 303 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ = 0, 304 TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT = 1, 305 TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC = 2, 306 }; 307 308 #define TLS_DEVICE_RESYNC_NH_START_IVAL 2 309 #define TLS_DEVICE_RESYNC_NH_MAX_IVAL 128 310 311 #define TLS_DEVICE_RESYNC_ASYNC_LOGMAX 13 312 struct tls_offload_resync_async { 313 atomic64_t req; 314 u16 loglen; 315 u16 rcd_delta; 316 u32 log[TLS_DEVICE_RESYNC_ASYNC_LOGMAX]; 317 }; 318 319 struct tls_offload_context_rx { 320 /* sw must be the first member of tls_offload_context_rx */ 321 struct tls_sw_context_rx sw; 322 enum tls_offload_sync_type resync_type; 323 /* this member is set regardless of resync_type, to avoid branches */ 324 u8 resync_nh_reset:1; 325 /* CORE_NEXT_HINT-only member, but use the hole here */ 326 u8 resync_nh_do_now:1; 327 union { 328 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ */ 329 struct { 330 atomic64_t resync_req; 331 }; 332 /* TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT */ 333 struct { 334 u32 decrypted_failed; 335 u32 decrypted_tgt; 336 } resync_nh; 337 /* TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC */ 338 struct { 339 struct tls_offload_resync_async *resync_async; 340 }; 341 }; 342 u8 driver_state[] __aligned(8); 343 /* The TLS layer reserves room for driver specific state 344 * Currently the belief is that there is not enough 345 * driver specific state to justify another layer of indirection 346 */ 347 #define TLS_DRIVER_STATE_SIZE_RX 8 348 }; 349 350 #define TLS_OFFLOAD_CONTEXT_SIZE_RX \ 351 (sizeof(struct tls_offload_context_rx) + TLS_DRIVER_STATE_SIZE_RX) 352 353 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context, 354 u32 seq, u64 *p_record_sn); 355 356 static inline bool tls_record_is_start_marker(struct tls_record_info *rec) 357 { 358 return rec->len == 0; 359 } 360 361 static inline u32 tls_record_start_seq(struct tls_record_info *rec) 362 { 363 return rec->end_seq - rec->len; 364 } 365 366 struct sk_buff * 367 tls_validate_xmit_skb(struct sock *sk, struct net_device *dev, 368 struct sk_buff *skb); 369 struct sk_buff * 370 tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev, 371 struct sk_buff *skb); 372 373 static inline bool tls_is_skb_tx_device_offloaded(const struct sk_buff *skb) 374 { 375 #ifdef CONFIG_TLS_DEVICE 376 struct sock *sk = skb->sk; 377 378 return sk && sk_fullsock(sk) && 379 (smp_load_acquire(&sk->sk_validate_xmit_skb) == 380 &tls_validate_xmit_skb); 381 #else 382 return false; 383 #endif 384 } 385 386 static inline struct tls_context *tls_get_ctx(const struct sock *sk) 387 { 388 struct inet_connection_sock *icsk = inet_csk(sk); 389 390 /* Use RCU on icsk_ulp_data only for sock diag code, 391 * TLS data path doesn't need rcu_dereference(). 392 */ 393 return (__force void *)icsk->icsk_ulp_data; 394 } 395 396 static inline struct tls_sw_context_rx *tls_sw_ctx_rx( 397 const struct tls_context *tls_ctx) 398 { 399 return (struct tls_sw_context_rx *)tls_ctx->priv_ctx_rx; 400 } 401 402 static inline struct tls_sw_context_tx *tls_sw_ctx_tx( 403 const struct tls_context *tls_ctx) 404 { 405 return (struct tls_sw_context_tx *)tls_ctx->priv_ctx_tx; 406 } 407 408 static inline struct tls_offload_context_tx * 409 tls_offload_ctx_tx(const struct tls_context *tls_ctx) 410 { 411 return (struct tls_offload_context_tx *)tls_ctx->priv_ctx_tx; 412 } 413 414 static inline bool tls_sw_has_ctx_tx(const struct sock *sk) 415 { 416 struct tls_context *ctx = tls_get_ctx(sk); 417 418 if (!ctx) 419 return false; 420 return !!tls_sw_ctx_tx(ctx); 421 } 422 423 static inline bool tls_sw_has_ctx_rx(const struct sock *sk) 424 { 425 struct tls_context *ctx = tls_get_ctx(sk); 426 427 if (!ctx) 428 return false; 429 return !!tls_sw_ctx_rx(ctx); 430 } 431 432 static inline struct tls_offload_context_rx * 433 tls_offload_ctx_rx(const struct tls_context *tls_ctx) 434 { 435 return (struct tls_offload_context_rx *)tls_ctx->priv_ctx_rx; 436 } 437 438 static inline void *__tls_driver_ctx(struct tls_context *tls_ctx, 439 enum tls_offload_ctx_dir direction) 440 { 441 if (direction == TLS_OFFLOAD_CTX_DIR_TX) 442 return tls_offload_ctx_tx(tls_ctx)->driver_state; 443 else 444 return tls_offload_ctx_rx(tls_ctx)->driver_state; 445 } 446 447 static inline void * 448 tls_driver_ctx(const struct sock *sk, enum tls_offload_ctx_dir direction) 449 { 450 return __tls_driver_ctx(tls_get_ctx(sk), direction); 451 } 452 453 #define RESYNC_REQ BIT(0) 454 #define RESYNC_REQ_ASYNC BIT(1) 455 /* The TLS context is valid until sk_destruct is called */ 456 static inline void tls_offload_rx_resync_request(struct sock *sk, __be32 seq) 457 { 458 struct tls_context *tls_ctx = tls_get_ctx(sk); 459 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx); 460 461 atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | RESYNC_REQ); 462 } 463 464 /* Log all TLS record header TCP sequences in [seq, seq+len] */ 465 static inline void 466 tls_offload_rx_resync_async_request_start(struct sock *sk, __be32 seq, u16 len) 467 { 468 struct tls_context *tls_ctx = tls_get_ctx(sk); 469 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx); 470 471 atomic64_set(&rx_ctx->resync_async->req, ((u64)ntohl(seq) << 32) | 472 ((u64)len << 16) | RESYNC_REQ | RESYNC_REQ_ASYNC); 473 rx_ctx->resync_async->loglen = 0; 474 rx_ctx->resync_async->rcd_delta = 0; 475 } 476 477 static inline void 478 tls_offload_rx_resync_async_request_end(struct sock *sk, __be32 seq) 479 { 480 struct tls_context *tls_ctx = tls_get_ctx(sk); 481 struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx); 482 483 atomic64_set(&rx_ctx->resync_async->req, 484 ((u64)ntohl(seq) << 32) | RESYNC_REQ); 485 } 486 487 static inline void 488 tls_offload_rx_resync_set_type(struct sock *sk, enum tls_offload_sync_type type) 489 { 490 struct tls_context *tls_ctx = tls_get_ctx(sk); 491 492 tls_offload_ctx_rx(tls_ctx)->resync_type = type; 493 } 494 495 /* Driver's seq tracking has to be disabled until resync succeeded */ 496 static inline bool tls_offload_tx_resync_pending(struct sock *sk) 497 { 498 struct tls_context *tls_ctx = tls_get_ctx(sk); 499 bool ret; 500 501 ret = test_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags); 502 smp_mb__after_atomic(); 503 return ret; 504 } 505 506 struct sk_buff *tls_encrypt_skb(struct sk_buff *skb); 507 508 #ifdef CONFIG_TLS_DEVICE 509 void tls_device_sk_destruct(struct sock *sk); 510 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq); 511 512 static inline bool tls_is_sk_rx_device_offloaded(struct sock *sk) 513 { 514 if (!sk_fullsock(sk) || 515 smp_load_acquire(&sk->sk_destruct) != tls_device_sk_destruct) 516 return false; 517 return tls_get_ctx(sk)->rx_conf == TLS_HW; 518 } 519 #endif 520 #endif /* _TLS_OFFLOAD_H */ 521