1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2020, Red Hat, Inc.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/inet.h>
10 #include <linux/kernel.h>
11 #include <net/tcp.h>
12 #include <net/netns/generic.h>
13 #include <net/mptcp.h>
14 #include <net/genetlink.h>
15 #include <uapi/linux/mptcp.h>
16
17 #include "protocol.h"
18 #include "mib.h"
19
20 /* forward declaration */
21 static struct genl_family mptcp_genl_family;
22
23 static int pm_nl_pernet_id;
24
25 struct mptcp_pm_addr_entry {
26 struct list_head list;
27 struct mptcp_addr_info addr;
28 struct rcu_head rcu;
29 };
30
31 struct mptcp_pm_add_entry {
32 struct list_head list;
33 struct mptcp_addr_info addr;
34 struct timer_list add_timer;
35 struct mptcp_sock *sock;
36 u8 retrans_times;
37 };
38
39 struct pm_nl_pernet {
40 /* protects pernet updates */
41 spinlock_t lock;
42 struct list_head local_addr_list;
43 unsigned int addrs;
44 unsigned int add_addr_signal_max;
45 unsigned int add_addr_accept_max;
46 unsigned int local_addr_max;
47 unsigned int subflows_max;
48 unsigned int next_id;
49 };
50
51 #define MPTCP_PM_ADDR_MAX 8
52 #define ADD_ADDR_RETRANS_MAX 3
53
addresses_equal(const struct mptcp_addr_info * a,struct mptcp_addr_info * b,bool use_port)54 static bool addresses_equal(const struct mptcp_addr_info *a,
55 struct mptcp_addr_info *b, bool use_port)
56 {
57 bool addr_equals = false;
58
59 if (a->family != b->family)
60 return false;
61
62 if (a->family == AF_INET)
63 addr_equals = a->addr.s_addr == b->addr.s_addr;
64 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
65 else
66 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
67 #endif
68
69 if (!addr_equals)
70 return false;
71 if (!use_port)
72 return true;
73
74 return a->port == b->port;
75 }
76
address_zero(const struct mptcp_addr_info * addr)77 static bool address_zero(const struct mptcp_addr_info *addr)
78 {
79 struct mptcp_addr_info zero;
80
81 memset(&zero, 0, sizeof(zero));
82 zero.family = addr->family;
83
84 return addresses_equal(addr, &zero, false);
85 }
86
local_address(const struct sock_common * skc,struct mptcp_addr_info * addr)87 static void local_address(const struct sock_common *skc,
88 struct mptcp_addr_info *addr)
89 {
90 addr->port = 0;
91 addr->family = skc->skc_family;
92 if (addr->family == AF_INET)
93 addr->addr.s_addr = skc->skc_rcv_saddr;
94 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
95 else if (addr->family == AF_INET6)
96 addr->addr6 = skc->skc_v6_rcv_saddr;
97 #endif
98 }
99
remote_address(const struct sock_common * skc,struct mptcp_addr_info * addr)100 static void remote_address(const struct sock_common *skc,
101 struct mptcp_addr_info *addr)
102 {
103 addr->family = skc->skc_family;
104 addr->port = skc->skc_dport;
105 if (addr->family == AF_INET)
106 addr->addr.s_addr = skc->skc_daddr;
107 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
108 else if (addr->family == AF_INET6)
109 addr->addr6 = skc->skc_v6_daddr;
110 #endif
111 }
112
lookup_subflow_by_saddr(const struct list_head * list,struct mptcp_addr_info * saddr)113 static bool lookup_subflow_by_saddr(const struct list_head *list,
114 struct mptcp_addr_info *saddr)
115 {
116 struct mptcp_subflow_context *subflow;
117 struct mptcp_addr_info cur;
118 struct sock_common *skc;
119
120 list_for_each_entry(subflow, list, node) {
121 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
122
123 local_address(skc, &cur);
124 if (addresses_equal(&cur, saddr, false))
125 return true;
126 }
127
128 return false;
129 }
130
131 static struct mptcp_pm_addr_entry *
select_local_address(const struct pm_nl_pernet * pernet,struct mptcp_sock * msk)132 select_local_address(const struct pm_nl_pernet *pernet,
133 struct mptcp_sock *msk)
134 {
135 struct mptcp_pm_addr_entry *entry, *ret = NULL;
136
137 rcu_read_lock();
138 spin_lock_bh(&msk->join_list_lock);
139 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
140 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
141 continue;
142
143 /* avoid any address already in use by subflows and
144 * pending join
145 */
146 if (entry->addr.family == ((struct sock *)msk)->sk_family &&
147 !lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
148 !lookup_subflow_by_saddr(&msk->join_list, &entry->addr)) {
149 ret = entry;
150 break;
151 }
152 }
153 spin_unlock_bh(&msk->join_list_lock);
154 rcu_read_unlock();
155 return ret;
156 }
157
158 static struct mptcp_pm_addr_entry *
select_signal_address(struct pm_nl_pernet * pernet,unsigned int pos)159 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos)
160 {
161 struct mptcp_pm_addr_entry *entry, *ret = NULL;
162 int i = 0;
163
164 rcu_read_lock();
165 /* do not keep any additional per socket state, just signal
166 * the address list in order.
167 * Note: removal from the local address list during the msk life-cycle
168 * can lead to additional addresses not being announced.
169 */
170 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
171 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
172 continue;
173 if (i++ == pos) {
174 ret = entry;
175 break;
176 }
177 }
178 rcu_read_unlock();
179 return ret;
180 }
181
check_work_pending(struct mptcp_sock * msk)182 static void check_work_pending(struct mptcp_sock *msk)
183 {
184 if (msk->pm.add_addr_signaled == msk->pm.add_addr_signal_max &&
185 (msk->pm.local_addr_used == msk->pm.local_addr_max ||
186 msk->pm.subflows == msk->pm.subflows_max))
187 WRITE_ONCE(msk->pm.work_pending, false);
188 }
189
190 static struct mptcp_pm_add_entry *
lookup_anno_list_by_saddr(struct mptcp_sock * msk,struct mptcp_addr_info * addr)191 lookup_anno_list_by_saddr(struct mptcp_sock *msk,
192 struct mptcp_addr_info *addr)
193 {
194 struct mptcp_pm_add_entry *entry;
195
196 list_for_each_entry(entry, &msk->pm.anno_list, list) {
197 if (addresses_equal(&entry->addr, addr, false))
198 return entry;
199 }
200
201 return NULL;
202 }
203
mptcp_pm_add_timer(struct timer_list * timer)204 static void mptcp_pm_add_timer(struct timer_list *timer)
205 {
206 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
207 struct mptcp_sock *msk = entry->sock;
208 struct sock *sk = (struct sock *)msk;
209
210 pr_debug("msk=%p", msk);
211
212 if (!msk)
213 return;
214
215 if (inet_sk_state_load(sk) == TCP_CLOSE)
216 return;
217
218 if (!entry->addr.id)
219 return;
220
221 if (mptcp_pm_should_add_signal(msk)) {
222 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
223 goto out;
224 }
225
226 spin_lock_bh(&msk->pm.lock);
227
228 if (!mptcp_pm_should_add_signal(msk)) {
229 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
230 mptcp_pm_announce_addr(msk, &entry->addr, false);
231 entry->retrans_times++;
232 }
233
234 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
235 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX);
236
237 spin_unlock_bh(&msk->pm.lock);
238
239 out:
240 __sock_put(sk);
241 }
242
243 struct mptcp_pm_add_entry *
mptcp_pm_del_add_timer(struct mptcp_sock * msk,struct mptcp_addr_info * addr)244 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
245 struct mptcp_addr_info *addr)
246 {
247 struct mptcp_pm_add_entry *entry;
248 struct sock *sk = (struct sock *)msk;
249
250 spin_lock_bh(&msk->pm.lock);
251 entry = lookup_anno_list_by_saddr(msk, addr);
252 if (entry)
253 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
254 spin_unlock_bh(&msk->pm.lock);
255
256 if (entry)
257 sk_stop_timer_sync(sk, &entry->add_timer);
258
259 return entry;
260 }
261
mptcp_pm_alloc_anno_list(struct mptcp_sock * msk,struct mptcp_pm_addr_entry * entry)262 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
263 struct mptcp_pm_addr_entry *entry)
264 {
265 struct mptcp_pm_add_entry *add_entry = NULL;
266 struct sock *sk = (struct sock *)msk;
267
268 if (lookup_anno_list_by_saddr(msk, &entry->addr))
269 return false;
270
271 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
272 if (!add_entry)
273 return false;
274
275 list_add(&add_entry->list, &msk->pm.anno_list);
276
277 add_entry->addr = entry->addr;
278 add_entry->sock = msk;
279 add_entry->retrans_times = 0;
280
281 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
282 sk_reset_timer(sk, &add_entry->add_timer, jiffies + TCP_RTO_MAX);
283
284 return true;
285 }
286
mptcp_pm_free_anno_list(struct mptcp_sock * msk)287 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
288 {
289 struct mptcp_pm_add_entry *entry, *tmp;
290 struct sock *sk = (struct sock *)msk;
291 LIST_HEAD(free_list);
292
293 pr_debug("msk=%p", msk);
294
295 spin_lock_bh(&msk->pm.lock);
296 list_splice_init(&msk->pm.anno_list, &free_list);
297 spin_unlock_bh(&msk->pm.lock);
298
299 list_for_each_entry_safe(entry, tmp, &free_list, list) {
300 sk_stop_timer_sync(sk, &entry->add_timer);
301 kfree(entry);
302 }
303 }
304
mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock * msk)305 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
306 {
307 struct mptcp_addr_info remote = { 0 };
308 struct sock *sk = (struct sock *)msk;
309 struct mptcp_pm_addr_entry *local;
310 struct pm_nl_pernet *pernet;
311
312 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
313
314 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
315 msk->pm.local_addr_used, msk->pm.local_addr_max,
316 msk->pm.add_addr_signaled, msk->pm.add_addr_signal_max,
317 msk->pm.subflows, msk->pm.subflows_max);
318
319 /* check first for announce */
320 if (msk->pm.add_addr_signaled < msk->pm.add_addr_signal_max) {
321 local = select_signal_address(pernet,
322 msk->pm.add_addr_signaled);
323
324 if (local) {
325 if (mptcp_pm_alloc_anno_list(msk, local)) {
326 msk->pm.add_addr_signaled++;
327 mptcp_pm_announce_addr(msk, &local->addr, false);
328 }
329 } else {
330 /* pick failed, avoid fourther attempts later */
331 msk->pm.local_addr_used = msk->pm.add_addr_signal_max;
332 }
333
334 check_work_pending(msk);
335 }
336
337 /* check if should create a new subflow */
338 if (msk->pm.local_addr_used < msk->pm.local_addr_max &&
339 msk->pm.subflows < msk->pm.subflows_max) {
340 remote_address((struct sock_common *)sk, &remote);
341
342 local = select_local_address(pernet, msk);
343 if (local) {
344 msk->pm.local_addr_used++;
345 msk->pm.subflows++;
346 check_work_pending(msk);
347 spin_unlock_bh(&msk->pm.lock);
348 __mptcp_subflow_connect(sk, &local->addr, &remote);
349 spin_lock_bh(&msk->pm.lock);
350 return;
351 }
352
353 /* lookup failed, avoid fourther attempts later */
354 msk->pm.local_addr_used = msk->pm.local_addr_max;
355 check_work_pending(msk);
356 }
357 }
358
mptcp_pm_nl_fully_established(struct mptcp_sock * msk)359 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
360 {
361 mptcp_pm_create_subflow_or_signal_addr(msk);
362 }
363
mptcp_pm_nl_subflow_established(struct mptcp_sock * msk)364 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
365 {
366 mptcp_pm_create_subflow_or_signal_addr(msk);
367 }
368
mptcp_pm_nl_add_addr_received(struct mptcp_sock * msk)369 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
370 {
371 struct sock *sk = (struct sock *)msk;
372 struct mptcp_addr_info remote;
373 struct mptcp_addr_info local;
374
375 pr_debug("accepted %d:%d remote family %d",
376 msk->pm.add_addr_accepted, msk->pm.add_addr_accept_max,
377 msk->pm.remote.family);
378 msk->pm.add_addr_accepted++;
379 msk->pm.subflows++;
380 if (msk->pm.add_addr_accepted >= msk->pm.add_addr_accept_max ||
381 msk->pm.subflows >= msk->pm.subflows_max)
382 WRITE_ONCE(msk->pm.accept_addr, false);
383
384 /* connect to the specified remote address, using whatever
385 * local address the routing configuration will pick.
386 */
387 remote = msk->pm.remote;
388 if (!remote.port)
389 remote.port = sk->sk_dport;
390 memset(&local, 0, sizeof(local));
391 local.family = remote.family;
392
393 spin_unlock_bh(&msk->pm.lock);
394 __mptcp_subflow_connect((struct sock *)msk, &local, &remote);
395 spin_lock_bh(&msk->pm.lock);
396
397 mptcp_pm_announce_addr(msk, &remote, true);
398 }
399
mptcp_pm_nl_rm_addr_received(struct mptcp_sock * msk)400 void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
401 {
402 struct mptcp_subflow_context *subflow, *tmp;
403 struct sock *sk = (struct sock *)msk;
404
405 pr_debug("address rm_id %d", msk->pm.rm_id);
406
407 if (!msk->pm.rm_id)
408 return;
409
410 if (list_empty(&msk->conn_list))
411 return;
412
413 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
414 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
415 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
416 long timeout = 0;
417
418 if (msk->pm.rm_id != subflow->remote_id)
419 continue;
420
421 spin_unlock_bh(&msk->pm.lock);
422 mptcp_subflow_shutdown(sk, ssk, how);
423 __mptcp_close_ssk(sk, ssk, subflow, timeout);
424 spin_lock_bh(&msk->pm.lock);
425
426 msk->pm.add_addr_accepted--;
427 msk->pm.subflows--;
428 WRITE_ONCE(msk->pm.accept_addr, true);
429
430 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMADDR);
431
432 break;
433 }
434 }
435
mptcp_pm_nl_rm_subflow_received(struct mptcp_sock * msk,u8 rm_id)436 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, u8 rm_id)
437 {
438 struct mptcp_subflow_context *subflow, *tmp;
439 struct sock *sk = (struct sock *)msk;
440
441 pr_debug("subflow rm_id %d", rm_id);
442
443 if (!rm_id)
444 return;
445
446 if (list_empty(&msk->conn_list))
447 return;
448
449 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
450 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
451 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
452 long timeout = 0;
453
454 if (rm_id != subflow->local_id)
455 continue;
456
457 spin_unlock_bh(&msk->pm.lock);
458 mptcp_subflow_shutdown(sk, ssk, how);
459 __mptcp_close_ssk(sk, ssk, subflow, timeout);
460 spin_lock_bh(&msk->pm.lock);
461
462 msk->pm.local_addr_used--;
463 msk->pm.subflows--;
464
465 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMSUBFLOW);
466
467 break;
468 }
469 }
470
address_use_port(struct mptcp_pm_addr_entry * entry)471 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
472 {
473 return (entry->addr.flags &
474 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
475 MPTCP_PM_ADDR_FLAG_SIGNAL;
476 }
477
mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet * pernet,struct mptcp_pm_addr_entry * entry)478 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
479 struct mptcp_pm_addr_entry *entry)
480 {
481 struct mptcp_pm_addr_entry *cur;
482 int ret = -EINVAL;
483
484 spin_lock_bh(&pernet->lock);
485 /* to keep the code simple, don't do IDR-like allocation for address ID,
486 * just bail when we exceed limits
487 */
488 if (pernet->next_id > 255)
489 goto out;
490 if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
491 goto out;
492
493 /* do not insert duplicate address, differentiate on port only
494 * singled addresses
495 */
496 list_for_each_entry(cur, &pernet->local_addr_list, list) {
497 if (addresses_equal(&cur->addr, &entry->addr,
498 address_use_port(entry) &&
499 address_use_port(cur)))
500 goto out;
501 }
502
503 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)
504 pernet->add_addr_signal_max++;
505 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
506 pernet->local_addr_max++;
507
508 entry->addr.id = pernet->next_id++;
509 pernet->addrs++;
510 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
511 ret = entry->addr.id;
512
513 out:
514 spin_unlock_bh(&pernet->lock);
515 return ret;
516 }
517
mptcp_pm_nl_get_local_id(struct mptcp_sock * msk,struct sock_common * skc)518 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
519 {
520 struct mptcp_pm_addr_entry *entry;
521 struct mptcp_addr_info skc_local;
522 struct mptcp_addr_info msk_local;
523 struct pm_nl_pernet *pernet;
524 int ret = -1;
525
526 if (WARN_ON_ONCE(!msk))
527 return -1;
528
529 /* The 0 ID mapping is defined by the first subflow, copied into the msk
530 * addr
531 */
532 local_address((struct sock_common *)msk, &msk_local);
533 local_address((struct sock_common *)skc, &skc_local);
534 if (addresses_equal(&msk_local, &skc_local, false))
535 return 0;
536
537 if (address_zero(&skc_local))
538 return 0;
539
540 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
541
542 rcu_read_lock();
543 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
544 if (addresses_equal(&entry->addr, &skc_local, false)) {
545 ret = entry->addr.id;
546 break;
547 }
548 }
549 rcu_read_unlock();
550 if (ret >= 0)
551 return ret;
552
553 /* address not found, add to local list */
554 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
555 if (!entry)
556 return -ENOMEM;
557
558 entry->addr = skc_local;
559 entry->addr.ifindex = 0;
560 entry->addr.flags = 0;
561 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
562 if (ret < 0)
563 kfree(entry);
564
565 return ret;
566 }
567
mptcp_pm_nl_data_init(struct mptcp_sock * msk)568 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
569 {
570 struct mptcp_pm_data *pm = &msk->pm;
571 struct pm_nl_pernet *pernet;
572 bool subflows;
573
574 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
575
576 pm->add_addr_signal_max = READ_ONCE(pernet->add_addr_signal_max);
577 pm->add_addr_accept_max = READ_ONCE(pernet->add_addr_accept_max);
578 pm->local_addr_max = READ_ONCE(pernet->local_addr_max);
579 pm->subflows_max = READ_ONCE(pernet->subflows_max);
580 subflows = !!pm->subflows_max;
581 WRITE_ONCE(pm->work_pending, (!!pm->local_addr_max && subflows) ||
582 !!pm->add_addr_signal_max);
583 WRITE_ONCE(pm->accept_addr, !!pm->add_addr_accept_max && subflows);
584 WRITE_ONCE(pm->accept_subflow, subflows);
585 }
586
587 #define MPTCP_PM_CMD_GRP_OFFSET 0
588
589 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
590 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
591 };
592
593 static const struct nla_policy
594 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
595 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
596 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
597 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
598 [MPTCP_PM_ADDR_ATTR_ADDR6] =
599 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
600 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
601 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
602 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
603 };
604
605 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
606 [MPTCP_PM_ATTR_ADDR] =
607 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
608 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
609 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
610 };
611
mptcp_pm_family_to_addr(int family)612 static int mptcp_pm_family_to_addr(int family)
613 {
614 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
615 if (family == AF_INET6)
616 return MPTCP_PM_ADDR_ATTR_ADDR6;
617 #endif
618 return MPTCP_PM_ADDR_ATTR_ADDR4;
619 }
620
mptcp_pm_parse_addr(struct nlattr * attr,struct genl_info * info,bool require_family,struct mptcp_pm_addr_entry * entry)621 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
622 bool require_family,
623 struct mptcp_pm_addr_entry *entry)
624 {
625 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
626 int err, addr_addr;
627
628 if (!attr) {
629 GENL_SET_ERR_MSG(info, "missing address info");
630 return -EINVAL;
631 }
632
633 /* no validation needed - was already done via nested policy */
634 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
635 mptcp_pm_addr_policy, info->extack);
636 if (err)
637 return err;
638
639 memset(entry, 0, sizeof(*entry));
640 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
641 if (!require_family)
642 goto skip_family;
643
644 NL_SET_ERR_MSG_ATTR(info->extack, attr,
645 "missing family");
646 return -EINVAL;
647 }
648
649 entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
650 if (entry->addr.family != AF_INET
651 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
652 && entry->addr.family != AF_INET6
653 #endif
654 ) {
655 NL_SET_ERR_MSG_ATTR(info->extack, attr,
656 "unknown address family");
657 return -EINVAL;
658 }
659 addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
660 if (!tb[addr_addr]) {
661 NL_SET_ERR_MSG_ATTR(info->extack, attr,
662 "missing address data");
663 return -EINVAL;
664 }
665
666 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
667 if (entry->addr.family == AF_INET6)
668 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
669 else
670 #endif
671 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
672
673 skip_family:
674 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
675 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
676
677 entry->addr.ifindex = val;
678 }
679
680 if (tb[MPTCP_PM_ADDR_ATTR_ID])
681 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
682
683 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
684 entry->addr.flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
685
686 return 0;
687 }
688
genl_info_pm_nl(struct genl_info * info)689 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
690 {
691 return net_generic(genl_info_net(info), pm_nl_pernet_id);
692 }
693
mptcp_nl_cmd_add_addr(struct sk_buff * skb,struct genl_info * info)694 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
695 {
696 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
697 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
698 struct mptcp_pm_addr_entry addr, *entry;
699 int ret;
700
701 ret = mptcp_pm_parse_addr(attr, info, true, &addr);
702 if (ret < 0)
703 return ret;
704
705 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
706 if (!entry) {
707 GENL_SET_ERR_MSG(info, "can't allocate addr");
708 return -ENOMEM;
709 }
710
711 *entry = addr;
712 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
713 if (ret < 0) {
714 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
715 kfree(entry);
716 return ret;
717 }
718
719 return 0;
720 }
721
722 static struct mptcp_pm_addr_entry *
__lookup_addr_by_id(struct pm_nl_pernet * pernet,unsigned int id)723 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
724 {
725 struct mptcp_pm_addr_entry *entry;
726
727 list_for_each_entry(entry, &pernet->local_addr_list, list) {
728 if (entry->addr.id == id)
729 return entry;
730 }
731 return NULL;
732 }
733
remove_anno_list_by_saddr(struct mptcp_sock * msk,struct mptcp_addr_info * addr)734 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
735 struct mptcp_addr_info *addr)
736 {
737 struct mptcp_pm_add_entry *entry;
738
739 entry = mptcp_pm_del_add_timer(msk, addr);
740 if (entry) {
741 list_del(&entry->list);
742 kfree(entry);
743 return true;
744 }
745
746 return false;
747 }
748
mptcp_pm_remove_anno_addr(struct mptcp_sock * msk,struct mptcp_addr_info * addr,bool force)749 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
750 struct mptcp_addr_info *addr,
751 bool force)
752 {
753 bool ret;
754
755 ret = remove_anno_list_by_saddr(msk, addr);
756 if (ret || force) {
757 spin_lock_bh(&msk->pm.lock);
758 mptcp_pm_remove_addr(msk, addr->id);
759 spin_unlock_bh(&msk->pm.lock);
760 }
761 return ret;
762 }
763
mptcp_nl_remove_subflow_and_signal_addr(struct net * net,struct mptcp_addr_info * addr)764 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
765 struct mptcp_addr_info *addr)
766 {
767 struct mptcp_sock *msk;
768 long s_slot = 0, s_num = 0;
769
770 pr_debug("remove_id=%d", addr->id);
771
772 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
773 struct sock *sk = (struct sock *)msk;
774 bool remove_subflow;
775
776 if (list_empty(&msk->conn_list)) {
777 mptcp_pm_remove_anno_addr(msk, addr, false);
778 goto next;
779 }
780
781 lock_sock(sk);
782 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
783 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
784 if (remove_subflow)
785 mptcp_pm_remove_subflow(msk, addr->id);
786 release_sock(sk);
787
788 next:
789 sock_put(sk);
790 cond_resched();
791 }
792
793 return 0;
794 }
795
mptcp_nl_cmd_del_addr(struct sk_buff * skb,struct genl_info * info)796 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
797 {
798 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
799 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
800 struct mptcp_pm_addr_entry addr, *entry;
801 int ret;
802
803 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
804 if (ret < 0)
805 return ret;
806
807 spin_lock_bh(&pernet->lock);
808 entry = __lookup_addr_by_id(pernet, addr.addr.id);
809 if (!entry) {
810 GENL_SET_ERR_MSG(info, "address not found");
811 spin_unlock_bh(&pernet->lock);
812 return -EINVAL;
813 }
814 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)
815 pernet->add_addr_signal_max--;
816 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
817 pernet->local_addr_max--;
818
819 pernet->addrs--;
820 list_del_rcu(&entry->list);
821 spin_unlock_bh(&pernet->lock);
822
823 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
824 kfree_rcu(entry, rcu);
825
826 return ret;
827 }
828
__flush_addrs(struct pm_nl_pernet * pernet)829 static void __flush_addrs(struct pm_nl_pernet *pernet)
830 {
831 while (!list_empty(&pernet->local_addr_list)) {
832 struct mptcp_pm_addr_entry *cur;
833
834 cur = list_entry(pernet->local_addr_list.next,
835 struct mptcp_pm_addr_entry, list);
836 list_del_rcu(&cur->list);
837 kfree_rcu(cur, rcu);
838 }
839 }
840
__reset_counters(struct pm_nl_pernet * pernet)841 static void __reset_counters(struct pm_nl_pernet *pernet)
842 {
843 pernet->add_addr_signal_max = 0;
844 pernet->add_addr_accept_max = 0;
845 pernet->local_addr_max = 0;
846 pernet->addrs = 0;
847 }
848
mptcp_nl_cmd_flush_addrs(struct sk_buff * skb,struct genl_info * info)849 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
850 {
851 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
852
853 spin_lock_bh(&pernet->lock);
854 __flush_addrs(pernet);
855 __reset_counters(pernet);
856 spin_unlock_bh(&pernet->lock);
857 return 0;
858 }
859
mptcp_nl_fill_addr(struct sk_buff * skb,struct mptcp_pm_addr_entry * entry)860 static int mptcp_nl_fill_addr(struct sk_buff *skb,
861 struct mptcp_pm_addr_entry *entry)
862 {
863 struct mptcp_addr_info *addr = &entry->addr;
864 struct nlattr *attr;
865
866 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
867 if (!attr)
868 return -EMSGSIZE;
869
870 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
871 goto nla_put_failure;
872 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
873 goto nla_put_failure;
874 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->addr.flags))
875 goto nla_put_failure;
876 if (entry->addr.ifindex &&
877 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->addr.ifindex))
878 goto nla_put_failure;
879
880 if (addr->family == AF_INET &&
881 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
882 addr->addr.s_addr))
883 goto nla_put_failure;
884 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
885 else if (addr->family == AF_INET6 &&
886 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
887 goto nla_put_failure;
888 #endif
889 nla_nest_end(skb, attr);
890 return 0;
891
892 nla_put_failure:
893 nla_nest_cancel(skb, attr);
894 return -EMSGSIZE;
895 }
896
mptcp_nl_cmd_get_addr(struct sk_buff * skb,struct genl_info * info)897 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
898 {
899 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
900 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
901 struct mptcp_pm_addr_entry addr, *entry;
902 struct sk_buff *msg;
903 void *reply;
904 int ret;
905
906 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
907 if (ret < 0)
908 return ret;
909
910 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
911 if (!msg)
912 return -ENOMEM;
913
914 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
915 info->genlhdr->cmd);
916 if (!reply) {
917 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
918 ret = -EMSGSIZE;
919 goto fail;
920 }
921
922 spin_lock_bh(&pernet->lock);
923 entry = __lookup_addr_by_id(pernet, addr.addr.id);
924 if (!entry) {
925 GENL_SET_ERR_MSG(info, "address not found");
926 ret = -EINVAL;
927 goto unlock_fail;
928 }
929
930 ret = mptcp_nl_fill_addr(msg, entry);
931 if (ret)
932 goto unlock_fail;
933
934 genlmsg_end(msg, reply);
935 ret = genlmsg_reply(msg, info);
936 spin_unlock_bh(&pernet->lock);
937 return ret;
938
939 unlock_fail:
940 spin_unlock_bh(&pernet->lock);
941
942 fail:
943 nlmsg_free(msg);
944 return ret;
945 }
946
mptcp_nl_cmd_dump_addrs(struct sk_buff * msg,struct netlink_callback * cb)947 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
948 struct netlink_callback *cb)
949 {
950 struct net *net = sock_net(msg->sk);
951 struct mptcp_pm_addr_entry *entry;
952 struct pm_nl_pernet *pernet;
953 int id = cb->args[0];
954 void *hdr;
955
956 pernet = net_generic(net, pm_nl_pernet_id);
957
958 spin_lock_bh(&pernet->lock);
959 list_for_each_entry(entry, &pernet->local_addr_list, list) {
960 if (entry->addr.id <= id)
961 continue;
962
963 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
964 cb->nlh->nlmsg_seq, &mptcp_genl_family,
965 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
966 if (!hdr)
967 break;
968
969 if (mptcp_nl_fill_addr(msg, entry) < 0) {
970 genlmsg_cancel(msg, hdr);
971 break;
972 }
973
974 id = entry->addr.id;
975 genlmsg_end(msg, hdr);
976 }
977 spin_unlock_bh(&pernet->lock);
978
979 cb->args[0] = id;
980 return msg->len;
981 }
982
parse_limit(struct genl_info * info,int id,unsigned int * limit)983 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
984 {
985 struct nlattr *attr = info->attrs[id];
986
987 if (!attr)
988 return 0;
989
990 *limit = nla_get_u32(attr);
991 if (*limit > MPTCP_PM_ADDR_MAX) {
992 GENL_SET_ERR_MSG(info, "limit greater than maximum");
993 return -EINVAL;
994 }
995 return 0;
996 }
997
998 static int
mptcp_nl_cmd_set_limits(struct sk_buff * skb,struct genl_info * info)999 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1000 {
1001 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1002 unsigned int rcv_addrs, subflows;
1003 int ret;
1004
1005 spin_lock_bh(&pernet->lock);
1006 rcv_addrs = pernet->add_addr_accept_max;
1007 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1008 if (ret)
1009 goto unlock;
1010
1011 subflows = pernet->subflows_max;
1012 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1013 if (ret)
1014 goto unlock;
1015
1016 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1017 WRITE_ONCE(pernet->subflows_max, subflows);
1018
1019 unlock:
1020 spin_unlock_bh(&pernet->lock);
1021 return ret;
1022 }
1023
1024 static int
mptcp_nl_cmd_get_limits(struct sk_buff * skb,struct genl_info * info)1025 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1026 {
1027 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1028 struct sk_buff *msg;
1029 void *reply;
1030
1031 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1032 if (!msg)
1033 return -ENOMEM;
1034
1035 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1036 MPTCP_PM_CMD_GET_LIMITS);
1037 if (!reply)
1038 goto fail;
1039
1040 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1041 READ_ONCE(pernet->add_addr_accept_max)))
1042 goto fail;
1043
1044 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1045 READ_ONCE(pernet->subflows_max)))
1046 goto fail;
1047
1048 genlmsg_end(msg, reply);
1049 return genlmsg_reply(msg, info);
1050
1051 fail:
1052 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1053 nlmsg_free(msg);
1054 return -EMSGSIZE;
1055 }
1056
1057 static const struct genl_small_ops mptcp_pm_ops[] = {
1058 {
1059 .cmd = MPTCP_PM_CMD_ADD_ADDR,
1060 .doit = mptcp_nl_cmd_add_addr,
1061 .flags = GENL_ADMIN_PERM,
1062 },
1063 {
1064 .cmd = MPTCP_PM_CMD_DEL_ADDR,
1065 .doit = mptcp_nl_cmd_del_addr,
1066 .flags = GENL_ADMIN_PERM,
1067 },
1068 {
1069 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
1070 .doit = mptcp_nl_cmd_flush_addrs,
1071 .flags = GENL_ADMIN_PERM,
1072 },
1073 {
1074 .cmd = MPTCP_PM_CMD_GET_ADDR,
1075 .doit = mptcp_nl_cmd_get_addr,
1076 .dumpit = mptcp_nl_cmd_dump_addrs,
1077 },
1078 {
1079 .cmd = MPTCP_PM_CMD_SET_LIMITS,
1080 .doit = mptcp_nl_cmd_set_limits,
1081 .flags = GENL_ADMIN_PERM,
1082 },
1083 {
1084 .cmd = MPTCP_PM_CMD_GET_LIMITS,
1085 .doit = mptcp_nl_cmd_get_limits,
1086 },
1087 };
1088
1089 static struct genl_family mptcp_genl_family __ro_after_init = {
1090 .name = MPTCP_PM_NAME,
1091 .version = MPTCP_PM_VER,
1092 .maxattr = MPTCP_PM_ATTR_MAX,
1093 .policy = mptcp_pm_policy,
1094 .netnsok = true,
1095 .module = THIS_MODULE,
1096 .small_ops = mptcp_pm_ops,
1097 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
1098 .mcgrps = mptcp_pm_mcgrps,
1099 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
1100 };
1101
pm_nl_init_net(struct net * net)1102 static int __net_init pm_nl_init_net(struct net *net)
1103 {
1104 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
1105
1106 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
1107 __reset_counters(pernet);
1108 pernet->next_id = 1;
1109 spin_lock_init(&pernet->lock);
1110 return 0;
1111 }
1112
pm_nl_exit_net(struct list_head * net_list)1113 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
1114 {
1115 struct net *net;
1116
1117 list_for_each_entry(net, net_list, exit_list) {
1118 /* net is removed from namespace list, can't race with
1119 * other modifiers
1120 */
1121 __flush_addrs(net_generic(net, pm_nl_pernet_id));
1122 }
1123 }
1124
1125 static struct pernet_operations mptcp_pm_pernet_ops = {
1126 .init = pm_nl_init_net,
1127 .exit_batch = pm_nl_exit_net,
1128 .id = &pm_nl_pernet_id,
1129 .size = sizeof(struct pm_nl_pernet),
1130 };
1131
mptcp_pm_nl_init(void)1132 void __init mptcp_pm_nl_init(void)
1133 {
1134 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
1135 panic("Failed to register MPTCP PM pernet subsystem.\n");
1136
1137 if (genl_register_family(&mptcp_genl_family))
1138 panic("Failed to register MPTCP PM netlink family\n");
1139 }
1140