1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2019, Intel Corporation.
5 */
6 #define pr_fmt(fmt) "MPTCP: " fmt
7
8 #include <linux/kernel.h>
9 #include <net/tcp.h>
10 #include <net/mptcp.h>
11 #include "protocol.h"
12
13 #include "mib.h"
14
15 /* path manager command handlers */
16
mptcp_pm_announce_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool echo)17 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
18 const struct mptcp_addr_info *addr,
19 bool echo)
20 {
21 u8 add_addr = READ_ONCE(msk->pm.addr_signal);
22
23 pr_debug("msk=%p, local_id=%d, echo=%d", msk, addr->id, echo);
24
25 lockdep_assert_held(&msk->pm.lock);
26
27 if (add_addr &
28 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) {
29 pr_warn("addr_signal error, add_addr=%d, echo=%d", add_addr, echo);
30 return -EINVAL;
31 }
32
33 if (echo) {
34 msk->pm.remote = *addr;
35 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO);
36 } else {
37 msk->pm.local = *addr;
38 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL);
39 }
40 WRITE_ONCE(msk->pm.addr_signal, add_addr);
41 return 0;
42 }
43
mptcp_pm_remove_addr(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)44 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
45 {
46 u8 rm_addr = READ_ONCE(msk->pm.addr_signal);
47
48 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
49
50 if (rm_addr) {
51 pr_warn("addr_signal error, rm_addr=%d", rm_addr);
52 return -EINVAL;
53 }
54
55 msk->pm.rm_list_tx = *rm_list;
56 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL);
57 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
58 mptcp_pm_nl_addr_send_ack(msk);
59 return 0;
60 }
61
mptcp_pm_remove_subflow(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)62 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
63 {
64 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
65
66 spin_lock_bh(&msk->pm.lock);
67 mptcp_pm_nl_rm_subflow_received(msk, rm_list);
68 spin_unlock_bh(&msk->pm.lock);
69 return 0;
70 }
71
72 /* path manager event handlers */
73
mptcp_pm_new_connection(struct mptcp_sock * msk,const struct sock * ssk,int server_side)74 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side)
75 {
76 struct mptcp_pm_data *pm = &msk->pm;
77
78 pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side);
79
80 WRITE_ONCE(pm->server_side, server_side);
81 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC);
82 }
83
mptcp_pm_allow_new_subflow(struct mptcp_sock * msk)84 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk)
85 {
86 struct mptcp_pm_data *pm = &msk->pm;
87 unsigned int subflows_max;
88 int ret = 0;
89
90 subflows_max = mptcp_pm_get_subflows_max(msk);
91
92 pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows,
93 subflows_max, READ_ONCE(pm->accept_subflow));
94
95 /* try to avoid acquiring the lock below */
96 if (!READ_ONCE(pm->accept_subflow))
97 return false;
98
99 spin_lock_bh(&pm->lock);
100 if (READ_ONCE(pm->accept_subflow)) {
101 ret = pm->subflows < subflows_max;
102 if (ret && ++pm->subflows == subflows_max)
103 WRITE_ONCE(pm->accept_subflow, false);
104 }
105 spin_unlock_bh(&pm->lock);
106
107 return ret;
108 }
109
110 /* return true if the new status bit is currently cleared, that is, this event
111 * can be server, eventually by an already scheduled work
112 */
mptcp_pm_schedule_work(struct mptcp_sock * msk,enum mptcp_pm_status new_status)113 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk,
114 enum mptcp_pm_status new_status)
115 {
116 pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status,
117 BIT(new_status));
118 if (msk->pm.status & BIT(new_status))
119 return false;
120
121 msk->pm.status |= BIT(new_status);
122 mptcp_schedule_work((struct sock *)msk);
123 return true;
124 }
125
mptcp_pm_fully_established(struct mptcp_sock * msk,const struct sock * ssk,gfp_t gfp)126 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk, gfp_t gfp)
127 {
128 struct mptcp_pm_data *pm = &msk->pm;
129 bool announce = false;
130
131 pr_debug("msk=%p", msk);
132
133 spin_lock_bh(&pm->lock);
134
135 /* mptcp_pm_fully_established() can be invoked by multiple
136 * racing paths - accept() and check_fully_established()
137 * be sure to serve this event only once.
138 */
139 if (READ_ONCE(pm->work_pending) &&
140 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)))
141 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED);
142
143 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
144 announce = true;
145
146 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED);
147 spin_unlock_bh(&pm->lock);
148
149 if (announce)
150 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, gfp);
151 }
152
mptcp_pm_connection_closed(struct mptcp_sock * msk)153 void mptcp_pm_connection_closed(struct mptcp_sock *msk)
154 {
155 pr_debug("msk=%p", msk);
156 }
157
mptcp_pm_subflow_established(struct mptcp_sock * msk)158 void mptcp_pm_subflow_established(struct mptcp_sock *msk)
159 {
160 struct mptcp_pm_data *pm = &msk->pm;
161
162 pr_debug("msk=%p", msk);
163
164 if (!READ_ONCE(pm->work_pending))
165 return;
166
167 spin_lock_bh(&pm->lock);
168
169 if (READ_ONCE(pm->work_pending))
170 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
171
172 spin_unlock_bh(&pm->lock);
173 }
174
mptcp_pm_subflow_closed(struct mptcp_sock * msk,u8 id)175 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id)
176 {
177 pr_debug("msk=%p", msk);
178 }
179
mptcp_pm_add_addr_received(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)180 void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
181 const struct mptcp_addr_info *addr)
182 {
183 struct mptcp_pm_data *pm = &msk->pm;
184
185 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id,
186 READ_ONCE(pm->accept_addr));
187
188 mptcp_event_addr_announced(msk, addr);
189
190 spin_lock_bh(&pm->lock);
191
192 if (!READ_ONCE(pm->accept_addr)) {
193 mptcp_pm_announce_addr(msk, addr, true);
194 mptcp_pm_add_addr_send_ack(msk);
195 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) {
196 pm->remote = *addr;
197 }
198
199 spin_unlock_bh(&pm->lock);
200 }
201
mptcp_pm_add_addr_echoed(struct mptcp_sock * msk,struct mptcp_addr_info * addr)202 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
203 struct mptcp_addr_info *addr)
204 {
205 struct mptcp_pm_data *pm = &msk->pm;
206
207 pr_debug("msk=%p", msk);
208
209 spin_lock_bh(&pm->lock);
210
211 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending))
212 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
213
214 spin_unlock_bh(&pm->lock);
215 }
216
mptcp_pm_add_addr_send_ack(struct mptcp_sock * msk)217 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk)
218 {
219 if (!mptcp_pm_should_add_signal(msk))
220 return;
221
222 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK);
223 }
224
mptcp_pm_rm_addr_received(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)225 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
226 const struct mptcp_rm_list *rm_list)
227 {
228 struct mptcp_pm_data *pm = &msk->pm;
229 u8 i;
230
231 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr);
232
233 for (i = 0; i < rm_list->nr; i++)
234 mptcp_event_addr_removed(msk, rm_list->ids[i]);
235
236 spin_lock_bh(&pm->lock);
237 mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED);
238 pm->rm_list_rx = *rm_list;
239 spin_unlock_bh(&pm->lock);
240 }
241
mptcp_pm_mp_prio_received(struct sock * sk,u8 bkup)242 void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup)
243 {
244 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
245
246 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup);
247 subflow->backup = bkup;
248
249 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, mptcp_sk(subflow->conn), sk, GFP_ATOMIC);
250 }
251
mptcp_pm_mp_fail_received(struct sock * sk,u64 fail_seq)252 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
253 {
254 pr_debug("fail_seq=%llu", fail_seq);
255 }
256
257 /* path manager helpers */
258
mptcp_pm_add_addr_signal(struct mptcp_sock * msk,struct sk_buff * skb,unsigned int opt_size,unsigned int remaining,struct mptcp_addr_info * addr,bool * echo,bool * port,bool * drop_other_suboptions)259 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, struct sk_buff *skb,
260 unsigned int opt_size, unsigned int remaining,
261 struct mptcp_addr_info *addr, bool *echo,
262 bool *port, bool *drop_other_suboptions)
263 {
264 int ret = false;
265 u8 add_addr;
266 u8 family;
267
268 spin_lock_bh(&msk->pm.lock);
269
270 /* double check after the lock is acquired */
271 if (!mptcp_pm_should_add_signal(msk))
272 goto out_unlock;
273
274 /* always drop every other options for pure ack ADD_ADDR; this is a
275 * plain dup-ack from TCP perspective. The other MPTCP-relevant info,
276 * if any, will be carried by the 'original' TCP ack
277 */
278 if (skb && skb_is_tcp_pure_ack(skb)) {
279 remaining += opt_size;
280 *drop_other_suboptions = true;
281 }
282
283 *echo = mptcp_pm_should_add_signal_echo(msk);
284 *port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port);
285
286 family = *echo ? msk->pm.remote.family : msk->pm.local.family;
287 if (remaining < mptcp_add_addr_len(family, *echo, *port))
288 goto out_unlock;
289
290 if (*echo) {
291 *addr = msk->pm.remote;
292 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO);
293 } else {
294 *addr = msk->pm.local;
295 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL);
296 }
297 WRITE_ONCE(msk->pm.addr_signal, add_addr);
298 ret = true;
299
300 out_unlock:
301 spin_unlock_bh(&msk->pm.lock);
302 return ret;
303 }
304
mptcp_pm_rm_addr_signal(struct mptcp_sock * msk,unsigned int remaining,struct mptcp_rm_list * rm_list)305 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
306 struct mptcp_rm_list *rm_list)
307 {
308 int ret = false, len;
309 u8 rm_addr;
310
311 spin_lock_bh(&msk->pm.lock);
312
313 /* double check after the lock is acquired */
314 if (!mptcp_pm_should_rm_signal(msk))
315 goto out_unlock;
316
317 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL);
318 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx);
319 if (len < 0) {
320 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
321 goto out_unlock;
322 }
323 if (remaining < len)
324 goto out_unlock;
325
326 *rm_list = msk->pm.rm_list_tx;
327 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
328 ret = true;
329
330 out_unlock:
331 spin_unlock_bh(&msk->pm.lock);
332 return ret;
333 }
334
mptcp_pm_get_local_id(struct mptcp_sock * msk,struct sock_common * skc)335 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
336 {
337 return mptcp_pm_nl_get_local_id(msk, skc);
338 }
339
mptcp_pm_subflow_chk_stale(const struct mptcp_sock * msk,struct sock * ssk)340 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
341 {
342 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
343 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp);
344
345 /* keep track of rtx periods with no progress */
346 if (!subflow->stale_count) {
347 subflow->stale_rcv_tstamp = rcv_tstamp;
348 subflow->stale_count++;
349 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) {
350 if (subflow->stale_count < U8_MAX)
351 subflow->stale_count++;
352 mptcp_pm_nl_subflow_chk_stale(msk, ssk);
353 } else {
354 subflow->stale_count = 0;
355 mptcp_subflow_set_active(subflow);
356 }
357 }
358
mptcp_pm_data_init(struct mptcp_sock * msk)359 void mptcp_pm_data_init(struct mptcp_sock *msk)
360 {
361 msk->pm.add_addr_signaled = 0;
362 msk->pm.add_addr_accepted = 0;
363 msk->pm.local_addr_used = 0;
364 msk->pm.subflows = 0;
365 msk->pm.rm_list_tx.nr = 0;
366 msk->pm.rm_list_rx.nr = 0;
367 WRITE_ONCE(msk->pm.work_pending, false);
368 WRITE_ONCE(msk->pm.addr_signal, 0);
369 WRITE_ONCE(msk->pm.accept_addr, false);
370 WRITE_ONCE(msk->pm.accept_subflow, false);
371 WRITE_ONCE(msk->pm.remote_deny_join_id0, false);
372 msk->pm.status = 0;
373
374 spin_lock_init(&msk->pm.lock);
375 INIT_LIST_HEAD(&msk->pm.anno_list);
376
377 mptcp_pm_nl_data_init(msk);
378 }
379
mptcp_pm_init(void)380 void __init mptcp_pm_init(void)
381 {
382 mptcp_pm_nl_init();
383 }
384
