projects/linux·blob master · refs

424 lines · 11.1 KiB

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 <[email protected]>
9 * Christoph Paasch <[email protected]>
10 * Jaakko Korkeaniemi <[email protected]>
11 * Gregory Detal <[email protected]>
12 * Fabien Duchêne <[email protected]>
13 * Andreas Seelinger <[email protected]>
14 * Lavkesh Lahngir <[email protected]>
15 * Andreas Ripke <[email protected]>
16 * Vlad Dogaru <[email protected]>
17 * Octavian Purdila <[email protected]>
18 * John Ronan <[email protected]>
19 * Catalin Nicutar <[email protected]>
20 * Brandon Heller <[email protected]>
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
38struct 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
45static struct token_bucket *token_hash __read_mostly;
46static unsigned int token_mask __read_mostly;
47
48static struct token_bucket *token_bucket(u32 token)
49{
50 return &token_hash[token & token_mask];
51}
52
53/* called with bucket lock held */
54static 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 */
67static 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
79static 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
85static 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 */
108int 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 */
151int 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
159again:
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_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
187 *
188 * Called when a SYN packet creates a new logical connection, i.e.
189 * is not a join request.
190 */
191void mptcp_token_accept(struct mptcp_subflow_request_sock *req,
192 struct mptcp_sock *msk)
193{
194 struct mptcp_subflow_request_sock *pos;
195 struct sock *sk = (struct sock *)msk;
196 struct token_bucket *bucket;
197
198 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
199 bucket = token_bucket(req->token);
200 spin_lock_bh(&bucket->lock);
201
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);
208}
209
210bool 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
220again:
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;
231found:
232 rcu_read_unlock();
233 return true;
234}
235
236/**
237 * mptcp_token_get_sock - retrieve mptcp connection sock using its token
238 * @net: restrict to this namespace
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 * Return: NULL if no connection with the given token value exists.
245 */
246struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token)
247{
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
256again:
257 sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) {
258 msk = mptcp_sk(sk);
259 if (READ_ONCE(msk->token) != token ||
260 !net_eq(sock_net(sk), net))
261 continue;
262
263 if (!refcount_inc_not_zero(&sk->sk_refcnt))
264 goto not_found;
265
266 if (READ_ONCE(msk->token) != token ||
267 !net_eq(sock_net(sk), net)) {
268 sock_put(sk);
269 goto again;
270 }
271 goto found;
272 }
273 if (get_nulls_value(pos) != (token & token_mask))
274 goto again;
275
276not_found:
277 msk = NULL;
278
279found:
280 rcu_read_unlock();
281 return msk;
282}
283EXPORT_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 * Description:
292 * On successful iteration, the iterator is moved to the next position and a
293 * reference to the returned socket is acquired.
294 *
295 * Return:
296 * The first mptcp connection structure found inside the token container
297 * starting from the specified position, or NULL.
298 */
299struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
300 long *s_num)
301{
302 struct mptcp_sock *ret = NULL;
303 struct hlist_nulls_node *pos;
304 int slot, num = 0;
305
306 for (slot = *s_slot; slot <= token_mask; *s_num = 0, slot++) {
307 struct token_bucket *bucket = &token_hash[slot];
308 struct sock *sk;
309
310 num = 0;
311
312 if (hlist_nulls_empty(&bucket->msk_chain))
313 continue;
314
315 rcu_read_lock();
316 sk_nulls_for_each_rcu(sk, pos, &bucket->msk_chain) {
317 ++num;
318 if (!net_eq(sock_net(sk), net))
319 continue;
320
321 if (num <= *s_num)
322 continue;
323
324 if (!refcount_inc_not_zero(&sk->sk_refcnt))
325 continue;
326
327 if (!net_eq(sock_net(sk), net)) {
328 sock_put(sk);
329 continue;
330 }
331
332 ret = mptcp_sk(sk);
333 rcu_read_unlock();
334 goto out;
335 }
336 rcu_read_unlock();
337 }
338
339out:
340 *s_slot = slot;
341 *s_num = num;
342 return ret;
343}
344EXPORT_SYMBOL_GPL(mptcp_token_iter_next);
345
346/**
347 * mptcp_token_destroy_request - remove mptcp connection/token
348 * @req: mptcp request socket dropping the token
349 *
350 * Remove the token associated to @req.
351 */
352void mptcp_token_destroy_request(struct request_sock *req)
353{
354 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
355 struct mptcp_subflow_request_sock *pos;
356 struct token_bucket *bucket;
357
358 if (hlist_nulls_unhashed(&subflow_req->token_node))
359 return;
360
361 bucket = token_bucket(subflow_req->token);
362 spin_lock_bh(&bucket->lock);
363 pos = __token_lookup_req(bucket, subflow_req->token);
364 if (!WARN_ON_ONCE(pos != subflow_req)) {
365 hlist_nulls_del_init_rcu(&pos->token_node);
366 bucket->chain_len--;
367 }
368 spin_unlock_bh(&bucket->lock);
369}
370
371/**
372 * mptcp_token_destroy - remove mptcp connection/token
373 * @msk: mptcp connection dropping the token
374 *
375 * Remove the token associated to @msk
376 */
377void mptcp_token_destroy(struct mptcp_sock *msk)
378{
379 struct sock *sk = (struct sock *)msk;
380 struct token_bucket *bucket;
381 struct mptcp_sock *pos;
382
383 if (sk_unhashed((struct sock *)msk))
384 return;
385
386 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
387 bucket = token_bucket(msk->token);
388 spin_lock_bh(&bucket->lock);
389 pos = __token_lookup_msk(bucket, msk->token);
390 if (!WARN_ON_ONCE(pos != msk)) {
391 __sk_nulls_del_node_init_rcu((struct sock *)pos);
392 bucket->chain_len--;
393 }
394 spin_unlock_bh(&bucket->lock);
395 WRITE_ONCE(msk->token, 0);
396}
397
398void __init mptcp_token_init(void)
399{
400 int i;
401
402 token_hash = alloc_large_system_hash("MPTCP token",
403 sizeof(struct token_bucket),
404 0,
405 20,/* one slot per 1MB of memory */
406 HASH_ZERO,
407 NULL,
408 &token_mask,
409 0,
410 64 * 1024);
411 for (i = 0; i < token_mask + 1; ++i) {
412 INIT_HLIST_NULLS_HEAD(&token_hash[i].req_chain, i);
413 INIT_HLIST_NULLS_HEAD(&token_hash[i].msk_chain, i);
414 spin_lock_init(&token_hash[i].lock);
415 }
416}
417
418#if IS_MODULE(CONFIG_MPTCP_KUNIT_TEST)
419EXPORT_SYMBOL_GPL(mptcp_token_new_request);
420EXPORT_SYMBOL_GPL(mptcp_token_new_connect);
421EXPORT_SYMBOL_GPL(mptcp_token_accept);
422EXPORT_SYMBOL_GPL(mptcp_token_destroy_request);
423EXPORT_SYMBOL_GPL(mptcp_token_destroy);
424#endif