Commit | Line | Data |
---|---|---|
1b1c7a0e PK |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Multipath TCP | |
3 | * | |
4 | * Copyright (c) 2019, Intel Corporation. | |
5 | */ | |
c85adced GT |
6 | #define pr_fmt(fmt) "MPTCP: " fmt |
7 | ||
1b1c7a0e PK |
8 | #include <linux/kernel.h> |
9 | #include <net/tcp.h> | |
10 | #include <net/mptcp.h> | |
11 | #include "protocol.h" | |
12 | ||
1b1c7a0e PK |
13 | /* path manager command handlers */ |
14 | ||
15 | int mptcp_pm_announce_addr(struct mptcp_sock *msk, | |
6a6c05a8 | 16 | const struct mptcp_addr_info *addr, |
0f5c9e3f | 17 | bool echo, bool port) |
1b1c7a0e | 18 | { |
13ad9f01 | 19 | u8 add_addr = READ_ONCE(msk->pm.addr_signal); |
d91d322a | 20 | |
926bdeab PK |
21 | pr_debug("msk=%p, local_id=%d", msk, addr->id); |
22 | ||
3abc05d9 FW |
23 | lockdep_assert_held(&msk->pm.lock); |
24 | ||
42842a42 GT |
25 | if (add_addr) { |
26 | pr_warn("addr_signal error, add_addr=%d", add_addr); | |
27 | return -EINVAL; | |
28 | } | |
29 | ||
926bdeab | 30 | msk->pm.local = *addr; |
d91d322a GT |
31 | add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); |
32 | if (echo) | |
33 | add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); | |
84dfe367 GT |
34 | if (addr->family == AF_INET6) |
35 | add_addr |= BIT(MPTCP_ADD_ADDR_IPV6); | |
0f5c9e3f GT |
36 | if (port) |
37 | add_addr |= BIT(MPTCP_ADD_ADDR_PORT); | |
13ad9f01 | 38 | WRITE_ONCE(msk->pm.addr_signal, add_addr); |
926bdeab | 39 | return 0; |
1b1c7a0e PK |
40 | } |
41 | ||
42 | int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id) | |
43 | { | |
13ad9f01 | 44 | u8 rm_addr = READ_ONCE(msk->pm.addr_signal); |
42842a42 | 45 | |
b6c08380 GT |
46 | pr_debug("msk=%p, local_id=%d", msk, local_id); |
47 | ||
42842a42 GT |
48 | if (rm_addr) { |
49 | pr_warn("addr_signal error, rm_addr=%d", rm_addr); | |
50 | return -EINVAL; | |
51 | } | |
52 | ||
b6c08380 | 53 | msk->pm.rm_id = local_id; |
42842a42 | 54 | rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); |
13ad9f01 | 55 | WRITE_ONCE(msk->pm.addr_signal, rm_addr); |
b6c08380 | 56 | return 0; |
1b1c7a0e PK |
57 | } |
58 | ||
0ee4261a | 59 | int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 local_id) |
1b1c7a0e | 60 | { |
0ee4261a GT |
61 | pr_debug("msk=%p, local_id=%d", msk, local_id); |
62 | ||
63 | spin_lock_bh(&msk->pm.lock); | |
64 | mptcp_pm_nl_rm_subflow_received(msk, local_id); | |
65 | spin_unlock_bh(&msk->pm.lock); | |
66 | return 0; | |
1b1c7a0e PK |
67 | } |
68 | ||
69 | /* path manager event handlers */ | |
70 | ||
6c714f1b | 71 | void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) |
1b1c7a0e PK |
72 | { |
73 | struct mptcp_pm_data *pm = &msk->pm; | |
74 | ||
75 | pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side); | |
76 | ||
77 | WRITE_ONCE(pm->server_side, server_side); | |
b911c97c | 78 | mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); |
1b1c7a0e PK |
79 | } |
80 | ||
81 | bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) | |
82 | { | |
926bdeab | 83 | struct mptcp_pm_data *pm = &msk->pm; |
a914e586 | 84 | unsigned int subflows_max; |
f58f065a | 85 | int ret = 0; |
926bdeab | 86 | |
a914e586 GT |
87 | subflows_max = mptcp_pm_get_subflows_max(msk); |
88 | ||
926bdeab | 89 | pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows, |
a914e586 | 90 | subflows_max, READ_ONCE(pm->accept_subflow)); |
926bdeab PK |
91 | |
92 | /* try to avoid acquiring the lock below */ | |
93 | if (!READ_ONCE(pm->accept_subflow)) | |
94 | return false; | |
95 | ||
96 | spin_lock_bh(&pm->lock); | |
f58f065a | 97 | if (READ_ONCE(pm->accept_subflow)) { |
a914e586 GT |
98 | ret = pm->subflows < subflows_max; |
99 | if (ret && ++pm->subflows == subflows_max) | |
f58f065a GT |
100 | WRITE_ONCE(pm->accept_subflow, false); |
101 | } | |
926bdeab PK |
102 | spin_unlock_bh(&pm->lock); |
103 | ||
104 | return ret; | |
105 | } | |
106 | ||
107 | /* return true if the new status bit is currently cleared, that is, this event | |
108 | * can be server, eventually by an already scheduled work | |
109 | */ | |
110 | static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, | |
111 | enum mptcp_pm_status new_status) | |
112 | { | |
113 | pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status, | |
114 | BIT(new_status)); | |
115 | if (msk->pm.status & BIT(new_status)) | |
116 | return false; | |
117 | ||
118 | msk->pm.status |= BIT(new_status); | |
ba8f48f7 | 119 | mptcp_schedule_work((struct sock *)msk); |
926bdeab | 120 | return true; |
1b1c7a0e PK |
121 | } |
122 | ||
6c714f1b | 123 | void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk, gfp_t gfp) |
1b1c7a0e | 124 | { |
926bdeab | 125 | struct mptcp_pm_data *pm = &msk->pm; |
b911c97c | 126 | bool announce = false; |
926bdeab | 127 | |
1b1c7a0e | 128 | pr_debug("msk=%p", msk); |
926bdeab | 129 | |
926bdeab PK |
130 | spin_lock_bh(&pm->lock); |
131 | ||
5b950ff4 PA |
132 | /* mptcp_pm_fully_established() can be invoked by multiple |
133 | * racing paths - accept() and check_fully_established() | |
134 | * be sure to serve this event only once. | |
135 | */ | |
136 | if (READ_ONCE(pm->work_pending) && | |
137 | !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) | |
926bdeab PK |
138 | mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); |
139 | ||
b911c97c FW |
140 | if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) |
141 | announce = true; | |
142 | ||
143 | msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); | |
926bdeab | 144 | spin_unlock_bh(&pm->lock); |
b911c97c FW |
145 | |
146 | if (announce) | |
147 | mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, gfp); | |
1b1c7a0e PK |
148 | } |
149 | ||
150 | void mptcp_pm_connection_closed(struct mptcp_sock *msk) | |
151 | { | |
152 | pr_debug("msk=%p", msk); | |
153 | } | |
154 | ||
155 | void mptcp_pm_subflow_established(struct mptcp_sock *msk, | |
156 | struct mptcp_subflow_context *subflow) | |
157 | { | |
926bdeab PK |
158 | struct mptcp_pm_data *pm = &msk->pm; |
159 | ||
1b1c7a0e | 160 | pr_debug("msk=%p", msk); |
926bdeab PK |
161 | |
162 | if (!READ_ONCE(pm->work_pending)) | |
163 | return; | |
164 | ||
165 | spin_lock_bh(&pm->lock); | |
166 | ||
167 | if (READ_ONCE(pm->work_pending)) | |
168 | mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); | |
169 | ||
170 | spin_unlock_bh(&pm->lock); | |
1b1c7a0e PK |
171 | } |
172 | ||
173 | void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id) | |
174 | { | |
175 | pr_debug("msk=%p", msk); | |
176 | } | |
177 | ||
178 | void mptcp_pm_add_addr_received(struct mptcp_sock *msk, | |
179 | const struct mptcp_addr_info *addr) | |
180 | { | |
926bdeab PK |
181 | struct mptcp_pm_data *pm = &msk->pm; |
182 | ||
183 | pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id, | |
184 | READ_ONCE(pm->accept_addr)); | |
185 | ||
b911c97c FW |
186 | mptcp_event_addr_announced(msk, addr); |
187 | ||
926bdeab PK |
188 | spin_lock_bh(&pm->lock); |
189 | ||
84dfe367 | 190 | if (!READ_ONCE(pm->accept_addr)) { |
0f5c9e3f | 191 | mptcp_pm_announce_addr(msk, addr, true, addr->port); |
84dfe367 GT |
192 | mptcp_pm_add_addr_send_ack(msk); |
193 | } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { | |
926bdeab | 194 | pm->remote = *addr; |
84dfe367 | 195 | } |
926bdeab PK |
196 | |
197 | spin_unlock_bh(&pm->lock); | |
84dfe367 GT |
198 | } |
199 | ||
200 | void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) | |
201 | { | |
b5a7acd3 | 202 | if (!mptcp_pm_should_add_signal(msk)) |
84dfe367 GT |
203 | return; |
204 | ||
205 | mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); | |
1b1c7a0e PK |
206 | } |
207 | ||
d0876b22 GT |
208 | void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, u8 rm_id) |
209 | { | |
210 | struct mptcp_pm_data *pm = &msk->pm; | |
211 | ||
212 | pr_debug("msk=%p remote_id=%d", msk, rm_id); | |
213 | ||
b911c97c FW |
214 | mptcp_event_addr_removed(msk, rm_id); |
215 | ||
d0876b22 GT |
216 | spin_lock_bh(&pm->lock); |
217 | mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED); | |
218 | pm->rm_id = rm_id; | |
219 | spin_unlock_bh(&pm->lock); | |
220 | } | |
221 | ||
40453a5c GT |
222 | void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup) |
223 | { | |
224 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
225 | ||
226 | pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); | |
227 | subflow->backup = bkup; | |
b911c97c FW |
228 | |
229 | mptcp_event(MPTCP_EVENT_SUB_PRIORITY, mptcp_sk(subflow->conn), sk, GFP_ATOMIC); | |
40453a5c GT |
230 | } |
231 | ||
1b1c7a0e PK |
232 | /* path manager helpers */ |
233 | ||
f643b803 | 234 | bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining, |
4a2777a8 | 235 | struct mptcp_addr_info *saddr, bool *echo, bool *port) |
1b1c7a0e | 236 | { |
926bdeab PK |
237 | int ret = false; |
238 | ||
239 | spin_lock_bh(&msk->pm.lock); | |
240 | ||
241 | /* double check after the lock is acquired */ | |
f643b803 | 242 | if (!mptcp_pm_should_add_signal(msk)) |
926bdeab PK |
243 | goto out_unlock; |
244 | ||
d91d322a | 245 | *echo = mptcp_pm_should_add_signal_echo(msk); |
4a2777a8 | 246 | *port = mptcp_pm_should_add_signal_port(msk); |
456afe01 | 247 | |
4a2777a8 | 248 | if (remaining < mptcp_add_addr_len(msk->pm.local.family, *echo, *port)) |
926bdeab PK |
249 | goto out_unlock; |
250 | ||
251 | *saddr = msk->pm.local; | |
13ad9f01 | 252 | WRITE_ONCE(msk->pm.addr_signal, 0); |
926bdeab PK |
253 | ret = true; |
254 | ||
255 | out_unlock: | |
256 | spin_unlock_bh(&msk->pm.lock); | |
257 | return ret; | |
1b1c7a0e PK |
258 | } |
259 | ||
5cb104ae GT |
260 | bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, |
261 | u8 *rm_id) | |
262 | { | |
263 | int ret = false; | |
264 | ||
265 | spin_lock_bh(&msk->pm.lock); | |
266 | ||
267 | /* double check after the lock is acquired */ | |
268 | if (!mptcp_pm_should_rm_signal(msk)) | |
269 | goto out_unlock; | |
270 | ||
271 | if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE) | |
272 | goto out_unlock; | |
273 | ||
274 | *rm_id = msk->pm.rm_id; | |
13ad9f01 | 275 | WRITE_ONCE(msk->pm.addr_signal, 0); |
5cb104ae GT |
276 | ret = true; |
277 | ||
278 | out_unlock: | |
279 | spin_unlock_bh(&msk->pm.lock); | |
280 | return ret; | |
281 | } | |
282 | ||
1b1c7a0e PK |
283 | int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) |
284 | { | |
01cacb00 | 285 | return mptcp_pm_nl_get_local_id(msk, skc); |
1b1c7a0e PK |
286 | } |
287 | ||
1b1c7a0e PK |
288 | void mptcp_pm_data_init(struct mptcp_sock *msk) |
289 | { | |
290 | msk->pm.add_addr_signaled = 0; | |
291 | msk->pm.add_addr_accepted = 0; | |
292 | msk->pm.local_addr_used = 0; | |
293 | msk->pm.subflows = 0; | |
5cb104ae | 294 | msk->pm.rm_id = 0; |
1b1c7a0e | 295 | WRITE_ONCE(msk->pm.work_pending, false); |
13ad9f01 | 296 | WRITE_ONCE(msk->pm.addr_signal, 0); |
1b1c7a0e PK |
297 | WRITE_ONCE(msk->pm.accept_addr, false); |
298 | WRITE_ONCE(msk->pm.accept_subflow, false); | |
299 | msk->pm.status = 0; | |
300 | ||
301 | spin_lock_init(&msk->pm.lock); | |
b6c08380 | 302 | INIT_LIST_HEAD(&msk->pm.anno_list); |
01cacb00 PA |
303 | |
304 | mptcp_pm_nl_data_init(msk); | |
1b1c7a0e PK |
305 | } |
306 | ||
d39dceca | 307 | void __init mptcp_pm_init(void) |
1b1c7a0e | 308 | { |
01cacb00 | 309 | mptcp_pm_nl_init(); |
1b1c7a0e | 310 | } |