1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2019, Intel Corporation.
6 #define pr_fmt(fmt) "MPTCP: " fmt
8 #include <linux/kernel.h>
10 #include <net/mptcp.h>
13 /* path manager command handlers */
15 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
16 const struct mptcp_addr_info *addr,
19 u8 add_addr = READ_ONCE(msk->pm.addr_signal);
21 pr_debug("msk=%p, local_id=%d", msk, addr->id);
23 lockdep_assert_held(&msk->pm.lock);
26 pr_warn("addr_signal error, add_addr=%d", add_addr);
30 msk->pm.local = *addr;
31 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL);
33 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO);
34 if (addr->family == AF_INET6)
35 add_addr |= BIT(MPTCP_ADD_ADDR_IPV6);
37 add_addr |= BIT(MPTCP_ADD_ADDR_PORT);
38 WRITE_ONCE(msk->pm.addr_signal, add_addr);
42 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
44 u8 rm_addr = READ_ONCE(msk->pm.addr_signal);
46 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
49 pr_warn("addr_signal error, rm_addr=%d", rm_addr);
53 msk->pm.rm_list_tx = *rm_list;
54 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL);
55 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
56 mptcp_pm_nl_addr_send_ack(msk);
60 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
62 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
64 spin_lock_bh(&msk->pm.lock);
65 mptcp_pm_nl_rm_subflow_received(msk, rm_list);
66 spin_unlock_bh(&msk->pm.lock);
70 /* path manager event handlers */
72 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side)
74 struct mptcp_pm_data *pm = &msk->pm;
76 pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side);
78 WRITE_ONCE(pm->server_side, server_side);
79 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC);
82 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk)
84 struct mptcp_pm_data *pm = &msk->pm;
85 unsigned int subflows_max;
88 subflows_max = mptcp_pm_get_subflows_max(msk);
90 pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows,
91 subflows_max, READ_ONCE(pm->accept_subflow));
93 /* try to avoid acquiring the lock below */
94 if (!READ_ONCE(pm->accept_subflow))
97 spin_lock_bh(&pm->lock);
98 if (READ_ONCE(pm->accept_subflow)) {
99 ret = pm->subflows < subflows_max;
100 if (ret && ++pm->subflows == subflows_max)
101 WRITE_ONCE(pm->accept_subflow, false);
103 spin_unlock_bh(&pm->lock);
108 /* return true if the new status bit is currently cleared, that is, this event
109 * can be server, eventually by an already scheduled work
111 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk,
112 enum mptcp_pm_status new_status)
114 pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status,
116 if (msk->pm.status & BIT(new_status))
119 msk->pm.status |= BIT(new_status);
120 mptcp_schedule_work((struct sock *)msk);
124 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk, gfp_t gfp)
126 struct mptcp_pm_data *pm = &msk->pm;
127 bool announce = false;
129 pr_debug("msk=%p", msk);
131 spin_lock_bh(&pm->lock);
133 /* mptcp_pm_fully_established() can be invoked by multiple
134 * racing paths - accept() and check_fully_established()
135 * be sure to serve this event only once.
137 if (READ_ONCE(pm->work_pending) &&
138 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)))
139 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED);
141 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
144 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED);
145 spin_unlock_bh(&pm->lock);
148 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, gfp);
151 void mptcp_pm_connection_closed(struct mptcp_sock *msk)
153 pr_debug("msk=%p", msk);
156 void mptcp_pm_subflow_established(struct mptcp_sock *msk)
158 struct mptcp_pm_data *pm = &msk->pm;
160 pr_debug("msk=%p", msk);
162 if (!READ_ONCE(pm->work_pending))
165 spin_lock_bh(&pm->lock);
167 if (READ_ONCE(pm->work_pending))
168 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
170 spin_unlock_bh(&pm->lock);
173 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id)
175 pr_debug("msk=%p", msk);
178 void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
179 const struct mptcp_addr_info *addr)
181 struct mptcp_pm_data *pm = &msk->pm;
183 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id,
184 READ_ONCE(pm->accept_addr));
186 mptcp_event_addr_announced(msk, addr);
188 spin_lock_bh(&pm->lock);
190 if (!READ_ONCE(pm->accept_addr)) {
191 mptcp_pm_announce_addr(msk, addr, true);
192 mptcp_pm_add_addr_send_ack(msk);
193 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) {
197 spin_unlock_bh(&pm->lock);
200 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
201 struct mptcp_addr_info *addr)
203 struct mptcp_pm_data *pm = &msk->pm;
205 pr_debug("msk=%p", msk);
207 spin_lock_bh(&pm->lock);
209 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending))
210 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
212 spin_unlock_bh(&pm->lock);
215 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk)
217 if (!mptcp_pm_should_add_signal(msk))
220 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK);
223 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
224 const struct mptcp_rm_list *rm_list)
226 struct mptcp_pm_data *pm = &msk->pm;
229 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr);
231 for (i = 0; i < rm_list->nr; i++)
232 mptcp_event_addr_removed(msk, rm_list->ids[i]);
234 spin_lock_bh(&pm->lock);
235 mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED);
236 pm->rm_list_rx = *rm_list;
237 spin_unlock_bh(&pm->lock);
240 void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup)
242 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
244 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup);
245 subflow->backup = bkup;
247 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, mptcp_sk(subflow->conn), sk, GFP_ATOMIC);
250 /* path manager helpers */
252 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
253 struct mptcp_addr_info *saddr, bool *echo, bool *port)
257 spin_lock_bh(&msk->pm.lock);
259 /* double check after the lock is acquired */
260 if (!mptcp_pm_should_add_signal(msk))
263 *echo = mptcp_pm_should_add_signal_echo(msk);
264 *port = mptcp_pm_should_add_signal_port(msk);
266 if (remaining < mptcp_add_addr_len(msk->pm.local.family, *echo, *port))
269 *saddr = msk->pm.local;
270 WRITE_ONCE(msk->pm.addr_signal, 0);
274 spin_unlock_bh(&msk->pm.lock);
278 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
279 struct mptcp_rm_list *rm_list)
281 int ret = false, len;
283 spin_lock_bh(&msk->pm.lock);
285 /* double check after the lock is acquired */
286 if (!mptcp_pm_should_rm_signal(msk))
289 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx);
291 WRITE_ONCE(msk->pm.addr_signal, 0);
297 *rm_list = msk->pm.rm_list_tx;
298 WRITE_ONCE(msk->pm.addr_signal, 0);
302 spin_unlock_bh(&msk->pm.lock);
306 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
308 return mptcp_pm_nl_get_local_id(msk, skc);
311 void mptcp_pm_data_init(struct mptcp_sock *msk)
313 msk->pm.add_addr_signaled = 0;
314 msk->pm.add_addr_accepted = 0;
315 msk->pm.local_addr_used = 0;
316 msk->pm.subflows = 0;
317 msk->pm.rm_list_tx.nr = 0;
318 msk->pm.rm_list_rx.nr = 0;
319 WRITE_ONCE(msk->pm.work_pending, false);
320 WRITE_ONCE(msk->pm.addr_signal, 0);
321 WRITE_ONCE(msk->pm.accept_addr, false);
322 WRITE_ONCE(msk->pm.accept_subflow, false);
325 spin_lock_init(&msk->pm.lock);
326 INIT_LIST_HEAD(&msk->pm.anno_list);
328 mptcp_pm_nl_data_init(msk);
331 void __init mptcp_pm_init(void)