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 | |
294de909 | 137 | * @ssk: 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 | */ | |
294de909 | 151 | int mptcp_token_new_connect(struct sock *ssk) |
79c0949e | 152 | { |
294de909 | 153 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); |
2c5ebd00 | 154 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); |
c68a0cd1 | 155 | int retries = MPTCP_TOKEN_MAX_RETRIES; |
c558246e | 156 | struct sock *sk = subflow->conn; |
2c5ebd00 | 157 | struct token_bucket *bucket; |
79c0949e | 158 | |
2c5ebd00 PA |
159 | again: |
160 | mptcp_crypto_key_gen_sha(&subflow->local_key, &subflow->token, | |
161 | &subflow->idsn); | |
79c0949e | 162 | |
2c5ebd00 PA |
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; | |
79c0949e | 170 | } |
79c0949e | 171 | |
2f1af441 | 172 | pr_debug("ssk=%p, local_key=%llu, token=%u, idsn=%llu\n", |
294de909 | 173 | ssk, subflow->local_key, subflow->token, subflow->idsn); |
2f1af441 | 174 | |
2c5ebd00 PA |
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); | |
c558246e | 179 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
2c5ebd00 | 180 | return 0; |
79c0949e PK |
181 | } |
182 | ||
183 | /** | |
2c5ebd00 PA |
184 | * mptcp_token_accept - replace a req sk with full sock in token hash |
185 | * @req: the request socket to be removed | |
186 | * @msk: the just cloned socket linked to the new connection | |
79c0949e PK |
187 | * |
188 | * Called when a SYN packet creates a new logical connection, i.e. | |
189 | * is not a join request. | |
79c0949e | 190 | */ |
2c5ebd00 PA |
191 | void mptcp_token_accept(struct mptcp_subflow_request_sock *req, |
192 | struct mptcp_sock *msk) | |
79c0949e | 193 | { |
2c5ebd00 | 194 | struct mptcp_subflow_request_sock *pos; |
c558246e | 195 | struct sock *sk = (struct sock *)msk; |
2c5ebd00 | 196 | struct token_bucket *bucket; |
79c0949e | 197 | |
c558246e | 198 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
2c5ebd00 PA |
199 | bucket = token_bucket(req->token); |
200 | spin_lock_bh(&bucket->lock); | |
79c0949e | 201 | |
2c5ebd00 PA |
202 | /* pedantic lookup check for the moved token */ |
203 | pos = __token_lookup_req(bucket, req->token); | |
204 | if (!WARN_ON_ONCE(pos != req)) | |
205 | hlist_nulls_del_init_rcu(&req->token_node); | |
206 | __sk_nulls_add_node_rcu((struct sock *)msk, &bucket->msk_chain); | |
207 | spin_unlock_bh(&bucket->lock); | |
79c0949e PK |
208 | } |
209 | ||
c83a47e5 FW |
210 | bool mptcp_token_exists(u32 token) |
211 | { | |
212 | struct hlist_nulls_node *pos; | |
213 | struct token_bucket *bucket; | |
214 | struct mptcp_sock *msk; | |
215 | struct sock *sk; | |
216 | ||
217 | rcu_read_lock(); | |
218 | bucket = token_bucket(token); | |
219 | ||
220 | again: | |
221 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
222 | msk = mptcp_sk(sk); | |
223 | if (READ_ONCE(msk->token) == token) | |
224 | goto found; | |
225 | } | |
226 | if (get_nulls_value(pos) != (token & token_mask)) | |
227 | goto again; | |
228 | ||
229 | rcu_read_unlock(); | |
230 | return false; | |
231 | found: | |
232 | rcu_read_unlock(); | |
233 | return true; | |
234 | } | |
235 | ||
f296234c PK |
236 | /** |
237 | * mptcp_token_get_sock - retrieve mptcp connection sock using its token | |
ea1300b9 | 238 | * @net: restrict to this namespace |
f296234c PK |
239 | * @token: token of the mptcp connection to retrieve |
240 | * | |
241 | * This function returns the mptcp connection structure with the given token. | |
242 | * A reference count on the mptcp socket returned is taken. | |
243 | * | |
244 | * returns NULL if no connection with the given token value exists. | |
245 | */ | |
ea1300b9 | 246 | struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token) |
f296234c | 247 | { |
2c5ebd00 PA |
248 | struct hlist_nulls_node *pos; |
249 | struct token_bucket *bucket; | |
250 | struct mptcp_sock *msk; | |
251 | struct sock *sk; | |
252 | ||
253 | rcu_read_lock(); | |
254 | bucket = token_bucket(token); | |
255 | ||
256 | again: | |
257 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
258 | msk = mptcp_sk(sk); | |
ea1300b9 FW |
259 | if (READ_ONCE(msk->token) != token || |
260 | !net_eq(sock_net(sk), net)) | |
2c5ebd00 | 261 | continue; |
ea1300b9 | 262 | |
2c5ebd00 PA |
263 | if (!refcount_inc_not_zero(&sk->sk_refcnt)) |
264 | goto not_found; | |
ea1300b9 FW |
265 | |
266 | if (READ_ONCE(msk->token) != token || | |
267 | !net_eq(sock_net(sk), net)) { | |
2c5ebd00 PA |
268 | sock_put(sk); |
269 | goto again; | |
270 | } | |
271 | goto found; | |
f296234c | 272 | } |
2c5ebd00 PA |
273 | if (get_nulls_value(pos) != (token & token_mask)) |
274 | goto again; | |
275 | ||
276 | not_found: | |
277 | msk = NULL; | |
f296234c | 278 | |
2c5ebd00 PA |
279 | found: |
280 | rcu_read_unlock(); | |
281 | return msk; | |
f296234c | 282 | } |
96d890da PA |
283 | EXPORT_SYMBOL_GPL(mptcp_token_get_sock); |
284 | ||
285 | /** | |
286 | * mptcp_token_iter_next - iterate over the token container from given pos | |
287 | * @net: namespace to be iterated | |
288 | * @s_slot: start slot number | |
289 | * @s_num: start number inside the given lock | |
290 | * | |
291 | * This function returns the first mptcp connection structure found inside the | |
292 | * token container starting from the specified position, or NULL. | |
293 | * | |
4373bf4b MM |
294 | * On successful iteration, the iterator is moved to the next position and |
295 | * a reference to the returned socket is acquired. | |
96d890da PA |
296 | */ |
297 | struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot, | |
298 | long *s_num) | |
299 | { | |
300 | struct mptcp_sock *ret = NULL; | |
301 | struct hlist_nulls_node *pos; | |
e16b874e | 302 | int slot, num = 0; |
96d890da PA |
303 | |
304 | for (slot = *s_slot; slot <= token_mask; *s_num = 0, slot++) { | |
305 | struct token_bucket *bucket = &token_hash[slot]; | |
306 | struct sock *sk; | |
307 | ||
308 | num = 0; | |
309 | ||
310 | if (hlist_nulls_empty(&bucket->msk_chain)) | |
311 | continue; | |
312 | ||
313 | rcu_read_lock(); | |
314 | sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) { | |
315 | ++num; | |
316 | if (!net_eq(sock_net(sk), net)) | |
317 | continue; | |
318 | ||
319 | if (num <= *s_num) | |
320 | continue; | |
321 | ||
322 | if (!refcount_inc_not_zero(&sk->sk_refcnt)) | |
323 | continue; | |
324 | ||
325 | if (!net_eq(sock_net(sk), net)) { | |
326 | sock_put(sk); | |
327 | continue; | |
328 | } | |
329 | ||
330 | ret = mptcp_sk(sk); | |
331 | rcu_read_unlock(); | |
332 | goto out; | |
333 | } | |
334 | rcu_read_unlock(); | |
335 | } | |
336 | ||
337 | out: | |
338 | *s_slot = slot; | |
339 | *s_num = num; | |
340 | return ret; | |
341 | } | |
342 | EXPORT_SYMBOL_GPL(mptcp_token_iter_next); | |
f296234c | 343 | |
79c0949e PK |
344 | /** |
345 | * mptcp_token_destroy_request - remove mptcp connection/token | |
2c5ebd00 | 346 | * @req: mptcp request socket dropping the token |
79c0949e | 347 | * |
2c5ebd00 | 348 | * Remove the token associated to @req. |
79c0949e | 349 | */ |
2c5ebd00 | 350 | void mptcp_token_destroy_request(struct request_sock *req) |
79c0949e | 351 | { |
2c5ebd00 PA |
352 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); |
353 | struct mptcp_subflow_request_sock *pos; | |
354 | struct token_bucket *bucket; | |
355 | ||
356 | if (hlist_nulls_unhashed(&subflow_req->token_node)) | |
357 | return; | |
358 | ||
359 | bucket = token_bucket(subflow_req->token); | |
360 | spin_lock_bh(&bucket->lock); | |
361 | pos = __token_lookup_req(bucket, subflow_req->token); | |
362 | if (!WARN_ON_ONCE(pos != subflow_req)) { | |
363 | hlist_nulls_del_init_rcu(&pos->token_node); | |
364 | bucket->chain_len--; | |
365 | } | |
366 | spin_unlock_bh(&bucket->lock); | |
79c0949e PK |
367 | } |
368 | ||
369 | /** | |
370 | * mptcp_token_destroy - remove mptcp connection/token | |
2c5ebd00 | 371 | * @msk: mptcp connection dropping the token |
79c0949e | 372 | * |
2c5ebd00 | 373 | * Remove the token associated to @msk |
79c0949e | 374 | */ |
2c5ebd00 | 375 | void mptcp_token_destroy(struct mptcp_sock *msk) |
79c0949e | 376 | { |
c558246e | 377 | struct sock *sk = (struct sock *)msk; |
2c5ebd00 PA |
378 | struct token_bucket *bucket; |
379 | struct mptcp_sock *pos; | |
380 | ||
381 | if (sk_unhashed((struct sock *)msk)) | |
382 | return; | |
383 | ||
c558246e | 384 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
2c5ebd00 PA |
385 | bucket = token_bucket(msk->token); |
386 | spin_lock_bh(&bucket->lock); | |
387 | pos = __token_lookup_msk(bucket, msk->token); | |
388 | if (!WARN_ON_ONCE(pos != msk)) { | |
389 | __sk_nulls_del_node_init_rcu((struct sock *)pos); | |
390 | bucket->chain_len--; | |
391 | } | |
392 | spin_unlock_bh(&bucket->lock); | |
b29fcfb5 | 393 | WRITE_ONCE(msk->token, 0); |
2c5ebd00 PA |
394 | } |
395 | ||
396 | void __init mptcp_token_init(void) | |
397 | { | |
398 | int i; | |
399 | ||
400 | token_hash = alloc_large_system_hash("MPTCP token", | |
401 | sizeof(struct token_bucket), | |
402 | 0, | |
403 | 20,/* one slot per 1MB of memory */ | |
6ab301c9 | 404 | HASH_ZERO, |
2c5ebd00 PA |
405 | NULL, |
406 | &token_mask, | |
407 | 0, | |
408 | 64 * 1024); | |
409 | for (i = 0; i < token_mask + 1; ++i) { | |
410 | INIT_HLIST_NULLS_HEAD(&token_hash[i].req_chain, i); | |
411 | INIT_HLIST_NULLS_HEAD(&token_hash[i].msk_chain, i); | |
412 | spin_lock_init(&token_hash[i].lock); | |
413 | } | |
79c0949e | 414 | } |
a8ee9c9b | 415 | |
3fcc8a25 | 416 | #if IS_MODULE(CONFIG_MPTCP_KUNIT_TEST) |
a8ee9c9b PA |
417 | EXPORT_SYMBOL_GPL(mptcp_token_new_request); |
418 | EXPORT_SYMBOL_GPL(mptcp_token_new_connect); | |
419 | EXPORT_SYMBOL_GPL(mptcp_token_accept); | |
a8ee9c9b PA |
420 | EXPORT_SYMBOL_GPL(mptcp_token_destroy_request); |
421 | EXPORT_SYMBOL_GPL(mptcp_token_destroy); | |
422 | #endif |