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>
15 /* path manager command handlers */
17 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
18 const struct mptcp_addr_info *addr,
21 u8 add_addr = READ_ONCE(msk->pm.addr_signal);
23 pr_debug("msk=%p, local_id=%d, echo=%d", msk, addr->id, echo);
25 lockdep_assert_held(&msk->pm.lock);
28 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) {
29 MPTCP_INC_STATS(sock_net((struct sock *)msk),
30 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP);
35 msk->pm.remote = *addr;
36 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO);
38 msk->pm.local = *addr;
39 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL);
41 WRITE_ONCE(msk->pm.addr_signal, add_addr);
45 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
47 u8 rm_addr = READ_ONCE(msk->pm.addr_signal);
49 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
52 MPTCP_ADD_STATS(sock_net((struct sock *)msk),
53 MPTCP_MIB_RMADDRTXDROP, rm_list->nr);
57 msk->pm.rm_list_tx = *rm_list;
58 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL);
59 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
60 mptcp_pm_nl_addr_send_ack(msk);
64 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
66 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr);
68 spin_lock_bh(&msk->pm.lock);
69 mptcp_pm_nl_rm_subflow_received(msk, rm_list);
70 spin_unlock_bh(&msk->pm.lock);
74 /* path manager event handlers */
76 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side)
78 struct mptcp_pm_data *pm = &msk->pm;
80 pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side);
82 WRITE_ONCE(pm->server_side, server_side);
83 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC);
86 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk)
88 struct mptcp_pm_data *pm = &msk->pm;
89 unsigned int subflows_max;
92 if (mptcp_pm_is_userspace(msk)) {
93 if (mptcp_userspace_pm_active(msk)) {
94 spin_lock_bh(&pm->lock);
96 spin_unlock_bh(&pm->lock);
102 subflows_max = mptcp_pm_get_subflows_max(msk);
104 pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows,
105 subflows_max, READ_ONCE(pm->accept_subflow));
107 /* try to avoid acquiring the lock below */
108 if (!READ_ONCE(pm->accept_subflow))
111 spin_lock_bh(&pm->lock);
112 if (READ_ONCE(pm->accept_subflow)) {
113 ret = pm->subflows < subflows_max;
114 if (ret && ++pm->subflows == subflows_max)
115 WRITE_ONCE(pm->accept_subflow, false);
117 spin_unlock_bh(&pm->lock);
122 /* return true if the new status bit is currently cleared, that is, this event
123 * can be server, eventually by an already scheduled work
125 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk,
126 enum mptcp_pm_status new_status)
128 pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status,
130 if (msk->pm.status & BIT(new_status))
133 msk->pm.status |= BIT(new_status);
134 mptcp_schedule_work((struct sock *)msk);
138 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk)
140 struct mptcp_pm_data *pm = &msk->pm;
141 bool announce = false;
143 pr_debug("msk=%p", msk);
145 spin_lock_bh(&pm->lock);
147 /* mptcp_pm_fully_established() can be invoked by multiple
148 * racing paths - accept() and check_fully_established()
149 * be sure to serve this event only once.
151 if (READ_ONCE(pm->work_pending) &&
152 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)))
153 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED);
155 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
158 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED);
159 spin_unlock_bh(&pm->lock);
162 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC);
165 void mptcp_pm_connection_closed(struct mptcp_sock *msk)
167 pr_debug("msk=%p", msk);
170 void mptcp_pm_subflow_established(struct mptcp_sock *msk)
172 struct mptcp_pm_data *pm = &msk->pm;
174 pr_debug("msk=%p", msk);
176 if (!READ_ONCE(pm->work_pending))
179 spin_lock_bh(&pm->lock);
181 if (READ_ONCE(pm->work_pending))
182 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
184 spin_unlock_bh(&pm->lock);
187 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, const struct sock *ssk,
188 const struct mptcp_subflow_context *subflow)
190 struct mptcp_pm_data *pm = &msk->pm;
191 bool update_subflows;
193 update_subflows = subflow->request_join || subflow->mp_join;
194 if (mptcp_pm_is_userspace(msk)) {
195 if (update_subflows) {
196 spin_lock_bh(&pm->lock);
198 spin_unlock_bh(&pm->lock);
203 if (!READ_ONCE(pm->work_pending) && !update_subflows)
206 spin_lock_bh(&pm->lock);
208 __mptcp_pm_close_subflow(msk);
210 /* Even if this subflow is not really established, tell the PM to try
211 * to pick the next ones, if possible.
213 if (mptcp_pm_nl_check_work_pending(msk))
214 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
216 spin_unlock_bh(&pm->lock);
219 void mptcp_pm_add_addr_received(const struct sock *ssk,
220 const struct mptcp_addr_info *addr)
222 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
223 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
224 struct mptcp_pm_data *pm = &msk->pm;
226 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id,
227 READ_ONCE(pm->accept_addr));
229 mptcp_event_addr_announced(ssk, addr);
231 spin_lock_bh(&pm->lock);
233 if (mptcp_pm_is_userspace(msk)) {
234 if (mptcp_userspace_pm_active(msk)) {
235 mptcp_pm_announce_addr(msk, addr, true);
236 mptcp_pm_add_addr_send_ack(msk);
238 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP);
240 } else if (!READ_ONCE(pm->accept_addr)) {
241 mptcp_pm_announce_addr(msk, addr, true);
242 mptcp_pm_add_addr_send_ack(msk);
243 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) {
246 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP);
249 spin_unlock_bh(&pm->lock);
252 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
253 const struct mptcp_addr_info *addr)
255 struct mptcp_pm_data *pm = &msk->pm;
257 pr_debug("msk=%p", msk);
259 spin_lock_bh(&pm->lock);
261 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending))
262 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
264 spin_unlock_bh(&pm->lock);
267 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk)
269 if (!mptcp_pm_should_add_signal(msk))
272 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK);
275 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
276 const struct mptcp_rm_list *rm_list)
278 struct mptcp_pm_data *pm = &msk->pm;
281 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr);
283 for (i = 0; i < rm_list->nr; i++)
284 mptcp_event_addr_removed(msk, rm_list->ids[i]);
286 spin_lock_bh(&pm->lock);
287 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED))
288 pm->rm_list_rx = *rm_list;
290 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP);
291 spin_unlock_bh(&pm->lock);
294 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup)
296 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
297 struct sock *sk = subflow->conn;
298 struct mptcp_sock *msk;
300 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup);
302 if (subflow->backup != bkup) {
303 subflow->backup = bkup;
305 if (!sock_owned_by_user(sk))
306 msk->last_snd = NULL;
308 __set_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags);
309 mptcp_data_unlock(sk);
312 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC);
315 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
317 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
318 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
320 pr_debug("fail_seq=%llu", fail_seq);
322 if (!READ_ONCE(msk->allow_infinite_fallback))
325 if (!subflow->fail_tout) {
326 pr_debug("send MP_FAIL response and infinite map");
328 subflow->send_mp_fail = 1;
329 subflow->send_infinite_map = 1;
332 pr_debug("MP_FAIL response received");
333 WRITE_ONCE(subflow->fail_tout, 0);
337 /* path manager helpers */
339 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb,
340 unsigned int opt_size, unsigned int remaining,
341 struct mptcp_addr_info *addr, bool *echo,
342 bool *drop_other_suboptions)
349 spin_lock_bh(&msk->pm.lock);
351 /* double check after the lock is acquired */
352 if (!mptcp_pm_should_add_signal(msk))
355 /* always drop every other options for pure ack ADD_ADDR; this is a
356 * plain dup-ack from TCP perspective. The other MPTCP-relevant info,
357 * if any, will be carried by the 'original' TCP ack
359 if (skb && skb_is_tcp_pure_ack(skb)) {
360 remaining += opt_size;
361 *drop_other_suboptions = true;
364 *echo = mptcp_pm_should_add_signal_echo(msk);
365 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port);
367 family = *echo ? msk->pm.remote.family : msk->pm.local.family;
368 if (remaining < mptcp_add_addr_len(family, *echo, port))
372 *addr = msk->pm.remote;
373 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO);
375 *addr = msk->pm.local;
376 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL);
378 WRITE_ONCE(msk->pm.addr_signal, add_addr);
382 spin_unlock_bh(&msk->pm.lock);
386 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
387 struct mptcp_rm_list *rm_list)
389 int ret = false, len;
392 spin_lock_bh(&msk->pm.lock);
394 /* double check after the lock is acquired */
395 if (!mptcp_pm_should_rm_signal(msk))
398 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL);
399 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx);
401 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
407 *rm_list = msk->pm.rm_list_tx;
408 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
412 spin_unlock_bh(&msk->pm.lock);
416 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
418 struct mptcp_addr_info skc_local;
419 struct mptcp_addr_info msk_local;
421 if (WARN_ON_ONCE(!msk))
424 /* The 0 ID mapping is defined by the first subflow, copied into the msk
427 mptcp_local_address((struct sock_common *)msk, &msk_local);
428 mptcp_local_address((struct sock_common *)skc, &skc_local);
429 if (mptcp_addresses_equal(&msk_local, &skc_local, false))
432 if (mptcp_pm_is_userspace(msk))
433 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
434 return mptcp_pm_nl_get_local_id(msk, &skc_local);
437 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
438 u8 *flags, int *ifindex)
446 if (mptcp_pm_is_userspace(msk))
447 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, id, flags, ifindex);
448 return mptcp_pm_nl_get_flags_and_ifindex_by_id(msk, id, flags, ifindex);
451 int mptcp_pm_set_flags(struct net *net, struct nlattr *token,
452 struct mptcp_pm_addr_entry *loc,
453 struct mptcp_pm_addr_entry *rem, u8 bkup)
456 return mptcp_userspace_pm_set_flags(net, token, loc, rem, bkup);
457 return mptcp_pm_nl_set_flags(net, loc, bkup);
460 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
462 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
463 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp);
465 /* keep track of rtx periods with no progress */
466 if (!subflow->stale_count) {
467 subflow->stale_rcv_tstamp = rcv_tstamp;
468 subflow->stale_count++;
469 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) {
470 if (subflow->stale_count < U8_MAX)
471 subflow->stale_count++;
472 mptcp_pm_nl_subflow_chk_stale(msk, ssk);
474 subflow->stale_count = 0;
475 mptcp_subflow_set_active(subflow);
479 /* if sk is ipv4 or ipv6_only allows only same-family local and remote addresses,
480 * otherwise allow any matching local/remote pair
482 bool mptcp_pm_addr_families_match(const struct sock *sk,
483 const struct mptcp_addr_info *loc,
484 const struct mptcp_addr_info *rem)
486 bool mptcp_is_v4 = sk->sk_family == AF_INET;
488 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
489 bool loc_is_v4 = loc->family == AF_INET || ipv6_addr_v4mapped(&loc->addr6);
490 bool rem_is_v4 = rem->family == AF_INET || ipv6_addr_v4mapped(&rem->addr6);
493 return loc_is_v4 && rem_is_v4;
495 if (ipv6_only_sock(sk))
496 return !loc_is_v4 && !rem_is_v4;
498 return loc_is_v4 == rem_is_v4;
500 return mptcp_is_v4 && loc->family == AF_INET && rem->family == AF_INET;
504 void mptcp_pm_data_reset(struct mptcp_sock *msk)
506 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk));
507 struct mptcp_pm_data *pm = &msk->pm;
509 pm->add_addr_signaled = 0;
510 pm->add_addr_accepted = 0;
511 pm->local_addr_used = 0;
513 pm->rm_list_tx.nr = 0;
514 pm->rm_list_rx.nr = 0;
515 WRITE_ONCE(pm->pm_type, pm_type);
517 if (pm_type == MPTCP_PM_TYPE_KERNEL) {
518 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk);
520 /* pm->work_pending must be only be set to 'true' when
521 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL
523 WRITE_ONCE(pm->work_pending,
524 (!!mptcp_pm_get_local_addr_max(msk) &&
526 !!mptcp_pm_get_add_addr_signal_max(msk));
527 WRITE_ONCE(pm->accept_addr,
528 !!mptcp_pm_get_add_addr_accept_max(msk) &&
530 WRITE_ONCE(pm->accept_subflow, subflows_allowed);
532 WRITE_ONCE(pm->work_pending, 0);
533 WRITE_ONCE(pm->accept_addr, 0);
534 WRITE_ONCE(pm->accept_subflow, 0);
537 WRITE_ONCE(pm->addr_signal, 0);
538 WRITE_ONCE(pm->remote_deny_join_id0, false);
540 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
543 void mptcp_pm_data_init(struct mptcp_sock *msk)
545 spin_lock_init(&msk->pm.lock);
546 INIT_LIST_HEAD(&msk->pm.anno_list);
547 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list);
548 mptcp_pm_data_reset(msk);
551 void __init mptcp_pm_init(void)