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