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/* path manager command handlers */
 14
 15int mptcp_pm_announce_addr(struct mptcp_sock *msk,
 16			   const struct mptcp_addr_info *addr,
 17			   bool echo)
 18{
 19	u8 add_addr = READ_ONCE(msk->pm.addr_signal);
 20
 21	pr_debug("msk=%p, local_id=%d", msk, addr->id);
 22
 23	lockdep_assert_held(&msk->pm.lock);
 24
 25	if (add_addr) {
 26		pr_warn("addr_signal error, add_addr=%d", add_addr);
 27		return -EINVAL;
 28	}
 29
 30	msk->pm.local = *addr;
 31	add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL);
 32	if (echo)
 33		add_addr |= BIT(MPTCP_ADD_ADDR_ECHO);
 34	if (addr->family == AF_INET6)
 35		add_addr |= BIT(MPTCP_ADD_ADDR_IPV6);
 36	if (addr->port)
 37		add_addr |= BIT(MPTCP_ADD_ADDR_PORT);
 38	WRITE_ONCE(msk->pm.addr_signal, add_addr);
 39	return 0;
 40}
 41
 42int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
 43{
 44	u8 rm_addr = READ_ONCE(msk->pm.addr_signal);
 45
 46	pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
 47
 48	if (rm_addr) {
 49		pr_warn("addr_signal error, rm_addr=%d", rm_addr);
 50		return -EINVAL;
 51	}
 52
 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);
 57	return 0;
 58}
 59
 60int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
 61{
 62	pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
 63
 64	spin_lock_bh(&msk->pm.lock);
 65	mptcp_pm_nl_rm_subflow_received(msk, rm_list);
 66	spin_unlock_bh(&msk->pm.lock);
 67	return 0;
 68}
 69
 70/* path manager event handlers */
 71
 72void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side)
 73{
 74	struct mptcp_pm_data *pm = &msk->pm;
 75
 76	pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side);
 77
 78	WRITE_ONCE(pm->server_side, server_side);
 79	mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC);
 80}
 81
 82bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk)
 83{
 84	struct mptcp_pm_data *pm = &msk->pm;
 85	unsigned int subflows_max;
 86	int ret = 0;
 87
 88	subflows_max = mptcp_pm_get_subflows_max(msk);
 89
 90	pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows,
 91		 subflows_max, READ_ONCE(pm->accept_subflow));
 92
 93	/* try to avoid acquiring the lock below */
 94	if (!READ_ONCE(pm->accept_subflow))
 95		return false;
 96
 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);
102	}
103	spin_unlock_bh(&pm->lock);
104
105	return ret;
106}
107
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
110 */
111static bool mptcp_pm_schedule_work(struct mptcp_sock *msk,
112				   enum mptcp_pm_status new_status)
113{
114	pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status,
115		 BIT(new_status));
116	if (msk->pm.status & BIT(new_status))
117		return false;
118
119	msk->pm.status |= BIT(new_status);
120	mptcp_schedule_work((struct sock *)msk);
121	return true;
122}
123
124void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk, gfp_t gfp)
125{
126	struct mptcp_pm_data *pm = &msk->pm;
127	bool announce = false;
128
129	pr_debug("msk=%p", msk);
130
131	spin_lock_bh(&pm->lock);
132
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.
136	 */
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);
140
141	if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
142		announce = true;
143
144	msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED);
145	spin_unlock_bh(&pm->lock);
146
147	if (announce)
148		mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, gfp);
149}
150
151void mptcp_pm_connection_closed(struct mptcp_sock *msk)
152{
153	pr_debug("msk=%p", msk);
154}
155
156void mptcp_pm_subflow_established(struct mptcp_sock *msk)
157{
158	struct mptcp_pm_data *pm = &msk->pm;
159
160	pr_debug("msk=%p", msk);
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);
171}
172
173void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id)
174{
175	pr_debug("msk=%p", msk);
176}
177
178void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
179				const struct mptcp_addr_info *addr)
180{
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
186	mptcp_event_addr_announced(msk, addr);
187
188	spin_lock_bh(&pm->lock);
189
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)) {
194		pm->remote = *addr;
195	}
196
197	spin_unlock_bh(&pm->lock);
198}
199
200void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
201			      struct mptcp_addr_info *addr)
202{
203	struct mptcp_pm_data *pm = &msk->pm;
204
205	pr_debug("msk=%p", msk);
206
207	spin_lock_bh(&pm->lock);
208
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);
211
212	spin_unlock_bh(&pm->lock);
213}
214
215void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk)
216{
217	if (!mptcp_pm_should_add_signal(msk))
218		return;
219
220	mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK);
221}
222
223void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
224			       const struct mptcp_rm_list *rm_list)
225{
226	struct mptcp_pm_data *pm = &msk->pm;
227	u8 i;
228
229	pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr);
230
231	for (i = 0; i < rm_list->nr; i++)
232		mptcp_event_addr_removed(msk, rm_list->ids[i]);
233
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);
238}
239
240void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup)
241{
242	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
243
244	pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup);
245	subflow->backup = bkup;
246
247	mptcp_event(MPTCP_EVENT_SUB_PRIORITY, mptcp_sk(subflow->conn), sk, GFP_ATOMIC);
248}
249
250/* path manager helpers */
251
252bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
253			      struct mptcp_addr_info *saddr, bool *echo, bool *port)
254{
255	int ret = false;
256
257	spin_lock_bh(&msk->pm.lock);
258
259	/* double check after the lock is acquired */
260	if (!mptcp_pm_should_add_signal(msk))
261		goto out_unlock;
262
263	*echo = mptcp_pm_should_add_signal_echo(msk);
264	*port = mptcp_pm_should_add_signal_port(msk);
265
266	if (remaining < mptcp_add_addr_len(msk->pm.local.family, *echo, *port))
267		goto out_unlock;
268
269	*saddr = msk->pm.local;
270	WRITE_ONCE(msk->pm.addr_signal, 0);
271	ret = true;
272
273out_unlock:
274	spin_unlock_bh(&msk->pm.lock);
275	return ret;
276}
277
278bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
279			     struct mptcp_rm_list *rm_list)
280{
281	int ret = false, len;
282
283	spin_lock_bh(&msk->pm.lock);
284
285	/* double check after the lock is acquired */
286	if (!mptcp_pm_should_rm_signal(msk))
287		goto out_unlock;
288
289	len = mptcp_rm_addr_len(&msk->pm.rm_list_tx);
290	if (len < 0) {
291		WRITE_ONCE(msk->pm.addr_signal, 0);
292		goto out_unlock;
293	}
294	if (remaining < len)
295		goto out_unlock;
296
297	*rm_list = msk->pm.rm_list_tx;
298	WRITE_ONCE(msk->pm.addr_signal, 0);
299	ret = true;
300
301out_unlock:
302	spin_unlock_bh(&msk->pm.lock);
303	return ret;
304}
305
306int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
307{
308	return mptcp_pm_nl_get_local_id(msk, skc);
309}
310
311void mptcp_pm_data_init(struct mptcp_sock *msk)
312{
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);
323	WRITE_ONCE(msk->pm.remote_deny_join_id0, false);
324	msk->pm.status = 0;
325
326	spin_lock_init(&msk->pm.lock);
327	INIT_LIST_HEAD(&msk->pm.anno_list);
328
329	mptcp_pm_nl_data_init(msk);
330}
331
332void __init mptcp_pm_init(void)
333{
334	mptcp_pm_nl_init();
335}