Commit | Line | Data |
---|---|---|
79c0949e PK |
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> | |
2c5ebd00 | 27 | #include <linux/memblock.h> |
79c0949e PK |
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 | ||
2c5ebd00 PA |
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 | } | |
79c0949e | 84 | |
c1d069e3 FW |
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 | ||
79c0949e PK |
97 | /** |
98 | * mptcp_token_new_request - create new key/idsn/token for subflow_request | |
564cf2f3 | 99 | * @req: the request socket |
79c0949e PK |
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 | * Returns 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); | |
2c5ebd00 PA |
111 | struct token_bucket *bucket; |
112 | u32 token; | |
113 | ||
535fb815 FW |
114 | mptcp_crypto_key_sha(subflow_req->local_key, |
115 | &subflow_req->token, | |
116 | &subflow_req->idsn); | |
2c5ebd00 PA |
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); | |
535fb815 | 126 | return -EBUSY; |
79c0949e PK |
127 | } |
128 | ||
2c5ebd00 PA |
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; | |
79c0949e PK |
133 | } |
134 | ||
135 | /** | |
136 | * mptcp_token_new_connect - create new key/idsn/token for subflow | |
564cf2f3 | 137 | * @sk: the socket that will initiate a connection |
79c0949e PK |
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 | * returns 0 on success. | |
150 | */ | |
151 | int mptcp_token_new_connect(struct sock *sk) | |
152 | { | |
153 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
2c5ebd00 | 154 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); |
c68a0cd1 | 155 | int retries = MPTCP_TOKEN_MAX_RETRIES; |
2c5ebd00 | 156 | struct token_bucket *bucket; |
79c0949e | 157 | |
2c5ebd00 PA |
158 | again: |
159 | mptcp_crypto_key_gen_sha(&subflow->local_key, &subflow->token, | |
160 | &subflow->idsn); | |
79c0949e | 161 | |
2c5ebd00 PA |
162 | bucket = token_bucket(subflow->token); |
163 | spin_lock_bh(&bucket->lock); | |
164 | if (__token_bucket_busy(bucket, subflow->token)) { | |
165 | spin_unlock_bh(&bucket->lock); | |
166 | if (!--retries) | |
167 | return -EBUSY; | |
168 | goto again; | |
79c0949e | 169 | } |
79c0949e | 170 | |
2f1af441 JW |
171 | pr_debug("ssk=%p, local_key=%llu, token=%u, idsn=%llu\n", |
172 | sk, subflow->local_key, subflow->token, subflow->idsn); | |
173 | ||
2c5ebd00 PA |
174 | WRITE_ONCE(msk->token, subflow->token); |
175 | __sk_nulls_add_node_rcu((struct sock *)msk, &bucket->msk_chain); | |
176 | bucket->chain_len++; | |
177 | spin_unlock_bh(&bucket->lock); | |
178 | return 0; | |
79c0949e PK |
179 | } |
180 | ||
181 | /** | |
2c5ebd00 PA |
182 | * mptcp_token_accept - replace a req sk with full sock in token hash |
183 | * @req: the request socket to be removed | |
184 | * @msk: the just cloned socket linked to the new connection | |
79c0949e PK |
185 | * |
186 | * Called when a SYN packet creates a new logical connection, i.e. | |
187 | * is not a join request. | |
79c0949e | 188 | */ |
2c5ebd00 PA |
189 | void mptcp_token_accept(struct mptcp_subflow_request_sock *req, |
190 | struct mptcp_sock *msk) | |
79c0949e | 191 | { |
2c5ebd00 PA |
192 | struct mptcp_subflow_request_sock *pos; |
193 | struct token_bucket *bucket; | |
79c0949e | 194 | |
2c5ebd00 PA |
195 | bucket = token_bucket(req->token); |
196 | spin_lock_bh(&bucket->lock); | |
79c0949e | 197 | |
2c5ebd00 PA |
198 | /* pedantic lookup check for the moved token */ |
199 | pos = __token_lookup_req(bucket, req->token); | |
200 | if (!WARN_ON_ONCE(pos != req)) | |
201 | hlist_nulls_del_init_rcu(&req->token_node); | |
202 | __sk_nulls_add_node_rcu((struct sock *)msk, &bucket->msk_chain); | |
203 | spin_unlock_bh(&bucket->lock); | |
79c0949e PK |
204 | } |
205 | ||
c83a47e5 FW |
206 | bool mptcp_token_exists(u32 token) |
207 | { | |
208 | struct hlist_nulls_node *pos; | |
209 | struct token_bucket *bucket; | |
210 | struct mptcp_sock *msk; | |
211 | struct sock *sk; | |
212 | ||
213 | rcu_read_lock(); | |
214 | bucket = token_bucket(token); | |
215 | ||
216 | again: | |
217 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
218 | msk = mptcp_sk(sk); | |
219 | if (READ_ONCE(msk->token) == token) | |
220 | goto found; | |
221 | } | |
222 | if (get_nulls_value(pos) != (token & token_mask)) | |
223 | goto again; | |
224 | ||
225 | rcu_read_unlock(); | |
226 | return false; | |
227 | found: | |
228 | rcu_read_unlock(); | |
229 | return true; | |
230 | } | |
231 | ||
f296234c PK |
232 | /** |
233 | * mptcp_token_get_sock - retrieve mptcp connection sock using its token | |
ea1300b9 | 234 | * @net: restrict to this namespace |
f296234c PK |
235 | * @token: token of the mptcp connection to retrieve |
236 | * | |
237 | * This function returns the mptcp connection structure with the given token. | |
238 | * A reference count on the mptcp socket returned is taken. | |
239 | * | |
240 | * returns NULL if no connection with the given token value exists. | |
241 | */ | |
ea1300b9 | 242 | struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token) |
f296234c | 243 | { |
2c5ebd00 PA |
244 | struct hlist_nulls_node *pos; |
245 | struct token_bucket *bucket; | |
246 | struct mptcp_sock *msk; | |
247 | struct sock *sk; | |
248 | ||
249 | rcu_read_lock(); | |
250 | bucket = token_bucket(token); | |
251 | ||
252 | again: | |
253 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
254 | msk = mptcp_sk(sk); | |
ea1300b9 FW |
255 | if (READ_ONCE(msk->token) != token || |
256 | !net_eq(sock_net(sk), net)) | |
2c5ebd00 | 257 | continue; |
ea1300b9 | 258 | |
2c5ebd00 PA |
259 | if (!refcount_inc_not_zero(&sk->sk_refcnt)) |
260 | goto not_found; | |
ea1300b9 FW |
261 | |
262 | if (READ_ONCE(msk->token) != token || | |
263 | !net_eq(sock_net(sk), net)) { | |
2c5ebd00 PA |
264 | sock_put(sk); |
265 | goto again; | |
266 | } | |
267 | goto found; | |
f296234c | 268 | } |
2c5ebd00 PA |
269 | if (get_nulls_value(pos) != (token & token_mask)) |
270 | goto again; | |
271 | ||
272 | not_found: | |
273 | msk = NULL; | |
f296234c | 274 | |
2c5ebd00 PA |
275 | found: |
276 | rcu_read_unlock(); | |
277 | return msk; | |
f296234c | 278 | } |
96d890da PA |
279 | EXPORT_SYMBOL_GPL(mptcp_token_get_sock); |
280 | ||
281 | /** | |
282 | * mptcp_token_iter_next - iterate over the token container from given pos | |
283 | * @net: namespace to be iterated | |
284 | * @s_slot: start slot number | |
285 | * @s_num: start number inside the given lock | |
286 | * | |
287 | * This function returns the first mptcp connection structure found inside the | |
288 | * token container starting from the specified position, or NULL. | |
289 | * | |
290 | * On successful iteration, the iterator is move to the next position and the | |
291 | * the acquires a reference to the returned socket. | |
292 | */ | |
293 | struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot, | |
294 | long *s_num) | |
295 | { | |
296 | struct mptcp_sock *ret = NULL; | |
297 | struct hlist_nulls_node *pos; | |
e16b874e | 298 | int slot, num = 0; |
96d890da PA |
299 | |
300 | for (slot = *s_slot; slot <= token_mask; *s_num = 0, slot++) { | |
301 | struct token_bucket *bucket = &token_hash[slot]; | |
302 | struct sock *sk; | |
303 | ||
304 | num = 0; | |
305 | ||
306 | if (hlist_nulls_empty(&bucket->msk_chain)) | |
307 | continue; | |
308 | ||
309 | rcu_read_lock(); | |
310 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
311 | ++num; | |
312 | if (!net_eq(sock_net(sk), net)) | |
313 | continue; | |
314 | ||
315 | if (num <= *s_num) | |
316 | continue; | |
317 | ||
318 | if (!refcount_inc_not_zero(&sk->sk_refcnt)) | |
319 | continue; | |
320 | ||
321 | if (!net_eq(sock_net(sk), net)) { | |
322 | sock_put(sk); | |
323 | continue; | |
324 | } | |
325 | ||
326 | ret = mptcp_sk(sk); | |
327 | rcu_read_unlock(); | |
328 | goto out; | |
329 | } | |
330 | rcu_read_unlock(); | |
331 | } | |
332 | ||
333 | out: | |
334 | *s_slot = slot; | |
335 | *s_num = num; | |
336 | return ret; | |
337 | } | |
338 | EXPORT_SYMBOL_GPL(mptcp_token_iter_next); | |
f296234c | 339 | |
79c0949e PK |
340 | /** |
341 | * mptcp_token_destroy_request - remove mptcp connection/token | |
2c5ebd00 | 342 | * @req: mptcp request socket dropping the token |
79c0949e | 343 | * |
2c5ebd00 | 344 | * Remove the token associated to @req. |
79c0949e | 345 | */ |
2c5ebd00 | 346 | void mptcp_token_destroy_request(struct request_sock *req) |
79c0949e | 347 | { |
2c5ebd00 PA |
348 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); |
349 | struct mptcp_subflow_request_sock *pos; | |
350 | struct token_bucket *bucket; | |
351 | ||
352 | if (hlist_nulls_unhashed(&subflow_req->token_node)) | |
353 | return; | |
354 | ||
355 | bucket = token_bucket(subflow_req->token); | |
356 | spin_lock_bh(&bucket->lock); | |
357 | pos = __token_lookup_req(bucket, subflow_req->token); | |
358 | if (!WARN_ON_ONCE(pos != subflow_req)) { | |
359 | hlist_nulls_del_init_rcu(&pos->token_node); | |
360 | bucket->chain_len--; | |
361 | } | |
362 | spin_unlock_bh(&bucket->lock); | |
79c0949e PK |
363 | } |
364 | ||
365 | /** | |
366 | * mptcp_token_destroy - remove mptcp connection/token | |
2c5ebd00 | 367 | * @msk: mptcp connection dropping the token |
79c0949e | 368 | * |
2c5ebd00 | 369 | * Remove the token associated to @msk |
79c0949e | 370 | */ |
2c5ebd00 | 371 | void mptcp_token_destroy(struct mptcp_sock *msk) |
79c0949e | 372 | { |
2c5ebd00 PA |
373 | struct token_bucket *bucket; |
374 | struct mptcp_sock *pos; | |
375 | ||
376 | if (sk_unhashed((struct sock *)msk)) | |
377 | return; | |
378 | ||
379 | bucket = token_bucket(msk->token); | |
380 | spin_lock_bh(&bucket->lock); | |
381 | pos = __token_lookup_msk(bucket, msk->token); | |
382 | if (!WARN_ON_ONCE(pos != msk)) { | |
383 | __sk_nulls_del_node_init_rcu((struct sock *)pos); | |
384 | bucket->chain_len--; | |
385 | } | |
386 | spin_unlock_bh(&bucket->lock); | |
387 | } | |
388 | ||
389 | void __init mptcp_token_init(void) | |
390 | { | |
391 | int i; | |
392 | ||
393 | token_hash = alloc_large_system_hash("MPTCP token", | |
394 | sizeof(struct token_bucket), | |
395 | 0, | |
396 | 20,/* one slot per 1MB of memory */ | |
6ab301c9 | 397 | HASH_ZERO, |
2c5ebd00 PA |
398 | NULL, |
399 | &token_mask, | |
400 | 0, | |
401 | 64 * 1024); | |
402 | for (i = 0; i < token_mask + 1; ++i) { | |
403 | INIT_HLIST_NULLS_HEAD(&token_hash[i].req_chain, i); | |
404 | INIT_HLIST_NULLS_HEAD(&token_hash[i].msk_chain, i); | |
405 | spin_lock_init(&token_hash[i].lock); | |
406 | } | |
79c0949e | 407 | } |
a8ee9c9b | 408 | |
3fcc8a25 | 409 | #if IS_MODULE(CONFIG_MPTCP_KUNIT_TEST) |
a8ee9c9b PA |
410 | EXPORT_SYMBOL_GPL(mptcp_token_new_request); |
411 | EXPORT_SYMBOL_GPL(mptcp_token_new_connect); | |
412 | EXPORT_SYMBOL_GPL(mptcp_token_accept); | |
a8ee9c9b PA |
413 | EXPORT_SYMBOL_GPL(mptcp_token_destroy_request); |
414 | EXPORT_SYMBOL_GPL(mptcp_token_destroy); | |
415 | #endif |