1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP token management 3 * Copyright (c) 2017 - 2019, Intel Corporation. 4 * 5 * Note: This code is based on mptcp_ctrl.c from multipath-tcp.org, 6 * authored by: 7 * 8 * Sébastien Barré <sebastien.barre@uclouvain.be> 9 * Christoph Paasch <christoph.paasch@uclouvain.be> 10 * Jaakko Korkeaniemi <jaakko.korkeaniemi@aalto.fi> 11 * Gregory Detal <gregory.detal@uclouvain.be> 12 * Fabien Duchêne <fabien.duchene@uclouvain.be> 13 * Andreas Seelinger <Andreas.Seelinger@rwth-aachen.de> 14 * Lavkesh Lahngir <lavkesh51@gmail.com> 15 * Andreas Ripke <ripke@neclab.eu> 16 * Vlad Dogaru <vlad.dogaru@intel.com> 17 * Octavian Purdila <octavian.purdila@intel.com> 18 * John Ronan <jronan@tssg.org> 19 * Catalin Nicutar <catalin.nicutar@gmail.com> 20 * Brandon Heller <brandonh@stanford.edu> 21 */ 22 23 #define pr_fmt(fmt) "MPTCP: " fmt 24 25 #include <linux/kernel.h> 26 #include <linux/module.h> 27 #include <linux/memblock.h> 28 #include <linux/ip.h> 29 #include <linux/tcp.h> 30 #include <net/sock.h> 31 #include <net/inet_common.h> 32 #include <net/protocol.h> 33 #include <net/mptcp.h> 34 #include "protocol.h" 35 36 #define TOKEN_MAX_CHAIN_LEN 4 37 38 struct token_bucket { 39 spinlock_t lock; 40 int chain_len; 41 struct hlist_nulls_head req_chain; 42 struct hlist_nulls_head msk_chain; 43 }; 44 45 static struct token_bucket *token_hash __read_mostly; 46 static unsigned int token_mask __read_mostly; 47 48 static struct token_bucket *token_bucket(u32 token) 49 { 50 return &token_hash[token & token_mask]; 51 } 52 53 /* called with bucket lock held */ 54 static struct mptcp_subflow_request_sock * 55 __token_lookup_req(struct token_bucket *t, u32 token) 56 { 57 struct mptcp_subflow_request_sock *req; 58 struct hlist_nulls_node *pos; 59 60 hlist_nulls_for_each_entry_rcu(req, pos, &t->req_chain, token_node) 61 if (req->token == token) 62 return req; 63 return NULL; 64 } 65 66 /* called with bucket lock held */ 67 static struct mptcp_sock * 68 __token_lookup_msk(struct token_bucket *t, u32 token) 69 { 70 struct hlist_nulls_node *pos; 71 struct sock *sk; 72 73 sk_nulls_for_each_rcu(sk, pos, &t->msk_chain) 74 if (mptcp_sk(sk)->token == token) 75 return mptcp_sk(sk); 76 return NULL; 77 } 78 79 static bool __token_bucket_busy(struct token_bucket *t, u32 token) 80 { 81 return !token || t->chain_len >= TOKEN_MAX_CHAIN_LEN || 82 __token_lookup_req(t, token) || __token_lookup_msk(t, token); 83 } 84 85 static void mptcp_crypto_key_gen_sha(u64 *key, u32 *token, u64 *idsn) 86 { 87 /* we might consider a faster version that computes the key as a 88 * hash of some information available in the MPTCP socket. Use 89 * random data at the moment, as it's probably the safest option 90 * in case multiple sockets are opened in different namespaces at 91 * the same time. 92 */ 93 get_random_bytes(key, sizeof(u64)); 94 mptcp_crypto_key_sha(*key, token, idsn); 95 } 96 97 /** 98 * mptcp_token_new_request - create new key/idsn/token for subflow_request 99 * @req: the request socket 100 * 101 * This function is called when a new mptcp connection is coming in. 102 * 103 * It creates a unique token to identify the new mptcp connection, 104 * a secret local key and the initial data sequence number (idsn). 105 * 106 * Return: 0 on success. 107 */ 108 int mptcp_token_new_request(struct request_sock *req) 109 { 110 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 111 struct token_bucket *bucket; 112 u32 token; 113 114 mptcp_crypto_key_sha(subflow_req->local_key, 115 &subflow_req->token, 116 &subflow_req->idsn); 117 pr_debug("req=%p local_key=%llu, token=%u, idsn=%llu\n", 118 req, subflow_req->local_key, subflow_req->token, 119 subflow_req->idsn); 120 121 token = subflow_req->token; 122 bucket = token_bucket(token); 123 spin_lock_bh(&bucket->lock); 124 if (__token_bucket_busy(bucket, token)) { 125 spin_unlock_bh(&bucket->lock); 126 return -EBUSY; 127 } 128 129 hlist_nulls_add_head_rcu(&subflow_req->token_node, &bucket->req_chain); 130 bucket->chain_len++; 131 spin_unlock_bh(&bucket->lock); 132 return 0; 133 } 134 135 /** 136 * mptcp_token_new_connect - create new key/idsn/token for subflow 137 * @ssk: the socket that will initiate a connection 138 * 139 * This function is called when a new outgoing mptcp connection is 140 * initiated. 141 * 142 * It creates a unique token to identify the new mptcp connection, 143 * a secret local key and the initial data sequence number (idsn). 144 * 145 * On success, the mptcp connection can be found again using 146 * the computed token at a later time, this is needed to process 147 * join requests. 148 * 149 * Return: 0 on success. 150 */ 151 int mptcp_token_new_connect(struct sock *ssk) 152 { 153 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 154 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 155 int retries = MPTCP_TOKEN_MAX_RETRIES; 156 struct sock *sk = subflow->conn; 157 struct token_bucket *bucket; 158 159 again: 160 mptcp_crypto_key_gen_sha(&subflow->local_key, &subflow->token, 161 &subflow->idsn); 162 163 bucket = token_bucket(subflow->token); 164 spin_lock_bh(&bucket->lock); 165 if (__token_bucket_busy(bucket, subflow->token)) { 166 spin_unlock_bh(&bucket->lock); 167 if (!--retries) 168 return -EBUSY; 169 goto again; 170 } 171 172 pr_debug("ssk=%p, local_key=%llu, token=%u, idsn=%llu\n", 173 ssk, subflow->local_key, subflow->token, subflow->idsn); 174 175 WRITE_ONCE(msk->token, subflow->token); 176 __sk_nulls_add_node_rcu((struct sock *)msk, &bucket->msk_chain); 177 bucket->chain_len++; 178 spin_unlock_bh(&bucket->lock); 179 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 180 return 0; 181 } 182 183 /** 184 * mptcp_token_move_request - move request token ownership to a clone 185 * @req: original request socket 186 * @new_req: cloned request socket 187 * 188 * Move the token hash entry from the original request to its clone. 189 */ 190 void mptcp_token_move_request(struct request_sock *req, 191 struct request_sock *new_req) 192 { 193 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 194 struct mptcp_subflow_request_sock *new_subflow_req; 195 struct mptcp_subflow_request_sock *pos; 196 struct token_bucket *bucket; 197 198 new_subflow_req = mptcp_subflow_rsk(new_req); 199 200 if (hlist_nulls_unhashed_lockless(&subflow_req->token_node)) { 201 mptcp_token_init_request(new_req); 202 return; 203 } 204 205 bucket = token_bucket(subflow_req->token); 206 spin_lock_bh(&bucket->lock); 207 if (hlist_nulls_unhashed(&subflow_req->token_node)) { 208 mptcp_token_init_request(new_req); 209 } else { 210 pos = __token_lookup_req(bucket, subflow_req->token); 211 if (pos == subflow_req) 212 hlist_nulls_replace_init_rcu(&subflow_req->token_node, 213 &new_subflow_req->token_node); 214 else 215 mptcp_token_init_request(new_req); 216 } 217 spin_unlock_bh(&bucket->lock); 218 } 219 220 /** 221 * mptcp_token_accept - replace a req sk with full sock in token hash 222 * @req: the request socket to be removed 223 * @msk: the just cloned socket linked to the new connection 224 * 225 * Called when a SYN packet creates a new logical connection, i.e. 226 * is not a join request. 227 * 228 * Return: true on success. 229 */ 230 bool mptcp_token_accept(struct mptcp_subflow_request_sock *req, 231 struct mptcp_sock *msk) 232 { 233 struct mptcp_subflow_request_sock *pos; 234 struct sock *sk = (struct sock *)msk; 235 struct token_bucket *bucket; 236 bool ret = false; 237 238 bucket = token_bucket(req->token); 239 spin_lock_bh(&bucket->lock); 240 if (hlist_nulls_unhashed(&req->token_node)) 241 goto unlock; 242 243 pos = __token_lookup_req(bucket, req->token); 244 if (pos != req) 245 goto unlock; 246 247 hlist_nulls_del_init_rcu(&req->token_node); 248 __sk_nulls_add_node_rcu((struct sock *)msk, &bucket->msk_chain); 249 ret = true; 250 251 unlock: 252 spin_unlock_bh(&bucket->lock); 253 if (ret) 254 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 255 256 return ret; 257 } 258 259 bool mptcp_token_exists(u32 token) 260 { 261 struct hlist_nulls_node *pos; 262 struct token_bucket *bucket; 263 struct mptcp_sock *msk; 264 struct sock *sk; 265 266 rcu_read_lock(); 267 bucket = token_bucket(token); 268 269 again: 270 sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { 271 msk = mptcp_sk(sk); 272 if (READ_ONCE(msk->token) == token) 273 goto found; 274 } 275 if (get_nulls_value(pos) != (token & token_mask)) 276 goto again; 277 278 rcu_read_unlock(); 279 return false; 280 found: 281 rcu_read_unlock(); 282 return true; 283 } 284 285 /** 286 * mptcp_token_get_sock - retrieve mptcp connection sock using its token 287 * @net: restrict to this namespace 288 * @token: token of the mptcp connection to retrieve 289 * 290 * This function returns the mptcp connection structure with the given token. 291 * A reference count on the mptcp socket returned is taken. 292 * 293 * Return: NULL if no connection with the given token value exists. 294 */ 295 struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token) 296 { 297 struct hlist_nulls_node *pos; 298 struct token_bucket *bucket; 299 struct mptcp_sock *msk; 300 struct sock *sk; 301 302 rcu_read_lock(); 303 bucket = token_bucket(token); 304 305 again: 306 sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { 307 msk = mptcp_sk(sk); 308 if (READ_ONCE(msk->token) != token || 309 !net_eq(sock_net(sk), net)) 310 continue; 311 312 if (!refcount_inc_not_zero(&sk->sk_refcnt)) 313 goto not_found; 314 315 if (READ_ONCE(msk->token) != token || 316 !net_eq(sock_net(sk), net)) { 317 sock_put(sk); 318 goto again; 319 } 320 goto found; 321 } 322 if (get_nulls_value(pos) != (token & token_mask)) 323 goto again; 324 325 not_found: 326 msk = NULL; 327 328 found: 329 rcu_read_unlock(); 330 return msk; 331 } 332 EXPORT_SYMBOL_GPL(mptcp_token_get_sock); 333 334 /** 335 * mptcp_token_iter_next - iterate over the token container from given pos 336 * @net: namespace to be iterated 337 * @s_slot: start slot number 338 * @s_num: start number inside the given lock 339 * 340 * Description: 341 * On successful iteration, the iterator is moved to the next position and a 342 * reference to the returned socket is acquired. 343 * 344 * Return: 345 * The first mptcp connection structure found inside the token container 346 * starting from the specified position, or NULL. 347 */ 348 struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot, 349 long *s_num) 350 { 351 struct mptcp_sock *ret = NULL; 352 struct hlist_nulls_node *pos; 353 int slot, num = 0; 354 355 for (slot = *s_slot; slot <= token_mask; *s_num = 0, slot++) { 356 struct token_bucket *bucket = &token_hash[slot]; 357 struct sock *sk; 358 359 num = 0; 360 361 if (hlist_nulls_empty(&bucket->msk_chain)) 362 continue; 363 364 rcu_read_lock(); 365 sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { 366 ++num; 367 if (!net_eq(sock_net(sk), net)) 368 continue; 369 370 if (num <= *s_num) 371 continue; 372 373 if (!refcount_inc_not_zero(&sk->sk_refcnt)) 374 continue; 375 376 if (!net_eq(sock_net(sk), net)) { 377 sock_put(sk); 378 continue; 379 } 380 381 ret = mptcp_sk(sk); 382 rcu_read_unlock(); 383 goto out; 384 } 385 rcu_read_unlock(); 386 } 387 388 out: 389 *s_slot = slot; 390 *s_num = num; 391 return ret; 392 } 393 EXPORT_SYMBOL_GPL(mptcp_token_iter_next); 394 395 /** 396 * mptcp_token_destroy_request - remove mptcp connection/token 397 * @req: mptcp request socket dropping the token 398 * 399 * Remove the token associated to @req. 400 */ 401 void mptcp_token_destroy_request(struct request_sock *req) 402 { 403 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 404 struct mptcp_subflow_request_sock *pos; 405 struct token_bucket *bucket; 406 407 if (hlist_nulls_unhashed_lockless(&subflow_req->token_node)) 408 return; 409 410 bucket = token_bucket(subflow_req->token); 411 spin_lock_bh(&bucket->lock); 412 if (hlist_nulls_unhashed(&subflow_req->token_node)) 413 goto unlock; 414 415 pos = __token_lookup_req(bucket, subflow_req->token); 416 if (pos == subflow_req) { 417 hlist_nulls_del_init_rcu(&pos->token_node); 418 bucket->chain_len--; 419 } 420 421 unlock: 422 spin_unlock_bh(&bucket->lock); 423 } 424 425 /** 426 * mptcp_token_destroy - remove mptcp connection/token 427 * @msk: mptcp connection dropping the token 428 * 429 * Remove the token associated to @msk 430 */ 431 void mptcp_token_destroy(struct mptcp_sock *msk) 432 { 433 struct sock *sk = (struct sock *)msk; 434 struct token_bucket *bucket; 435 struct mptcp_sock *pos; 436 437 if (sk_unhashed((struct sock *)msk)) 438 return; 439 440 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 441 bucket = token_bucket(msk->token); 442 spin_lock_bh(&bucket->lock); 443 pos = __token_lookup_msk(bucket, msk->token); 444 if (!WARN_ON_ONCE(pos != msk)) { 445 __sk_nulls_del_node_init_rcu((struct sock *)pos); 446 bucket->chain_len--; 447 } 448 spin_unlock_bh(&bucket->lock); 449 WRITE_ONCE(msk->token, 0); 450 } 451 452 void __init mptcp_token_init(void) 453 { 454 int i; 455 456 token_hash = alloc_large_system_hash("MPTCP token", 457 sizeof(struct token_bucket), 458 0, 459 20,/* one slot per 1MB of memory */ 460 HASH_ZERO, 461 NULL, 462 &token_mask, 463 0, 464 64 * 1024); 465 for (i = 0; i < token_mask + 1; ++i) { 466 INIT_HLIST_NULLS_HEAD(&token_hash[i].req_chain, i); 467 INIT_HLIST_NULLS_HEAD(&token_hash[i].msk_chain, i); 468 spin_lock_init(&token_hash[i].lock); 469 } 470 } 471 472 #if IS_MODULE(CONFIG_MPTCP_KUNIT_TEST) 473 EXPORT_SYMBOL_GPL(mptcp_token_new_request); 474 EXPORT_SYMBOL_GPL(mptcp_token_new_connect); 475 EXPORT_SYMBOL_GPL(mptcp_token_accept); 476 EXPORT_SYMBOL_GPL(mptcp_token_destroy_request); 477 EXPORT_SYMBOL_GPL(mptcp_token_destroy); 478 #endif 479