1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC remote transport endpoint record management
3  *
4  * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/module.h>
11 #include <linux/net.h>
12 #include <linux/skbuff.h>
13 #include <linux/udp.h>
14 #include <linux/in.h>
15 #include <linux/in6.h>
16 #include <linux/slab.h>
17 #include <linux/hashtable.h>
18 #include <net/sock.h>
19 #include <net/af_rxrpc.h>
20 #include <net/ip.h>
21 #include <net/route.h>
22 #include <net/ip6_route.h>
23 #include "ar-internal.h"
24 
25 /*
26  * Hash a peer key.
27  */
rxrpc_peer_hash_key(struct rxrpc_local * local,const struct sockaddr_rxrpc * srx)28 static unsigned long rxrpc_peer_hash_key(struct rxrpc_local *local,
29 					 const struct sockaddr_rxrpc *srx)
30 {
31 	const u16 *p;
32 	unsigned int i, size;
33 	unsigned long hash_key;
34 
35 	_enter("");
36 
37 	hash_key = (unsigned long)local / __alignof__(*local);
38 	hash_key += srx->transport_type;
39 	hash_key += srx->transport_len;
40 	hash_key += srx->transport.family;
41 
42 	switch (srx->transport.family) {
43 	case AF_INET:
44 		hash_key += (u16 __force)srx->transport.sin.sin_port;
45 		size = sizeof(srx->transport.sin.sin_addr);
46 		p = (u16 *)&srx->transport.sin.sin_addr;
47 		break;
48 #ifdef CONFIG_AF_RXRPC_IPV6
49 	case AF_INET6:
50 		hash_key += (u16 __force)srx->transport.sin.sin_port;
51 		size = sizeof(srx->transport.sin6.sin6_addr);
52 		p = (u16 *)&srx->transport.sin6.sin6_addr;
53 		break;
54 #endif
55 	default:
56 		WARN(1, "AF_RXRPC: Unsupported transport address family\n");
57 		return 0;
58 	}
59 
60 	/* Step through the peer address in 16-bit portions for speed */
61 	for (i = 0; i < size; i += sizeof(*p), p++)
62 		hash_key += *p;
63 
64 	_leave(" 0x%lx", hash_key);
65 	return hash_key;
66 }
67 
68 /*
69  * Compare a peer to a key.  Return -ve, 0 or +ve to indicate less than, same
70  * or greater than.
71  *
72  * Unfortunately, the primitives in linux/hashtable.h don't allow for sorted
73  * buckets and mid-bucket insertion, so we don't make full use of this
74  * information at this point.
75  */
rxrpc_peer_cmp_key(const struct rxrpc_peer * peer,struct rxrpc_local * local,const struct sockaddr_rxrpc * srx,unsigned long hash_key)76 static long rxrpc_peer_cmp_key(const struct rxrpc_peer *peer,
77 			       struct rxrpc_local *local,
78 			       const struct sockaddr_rxrpc *srx,
79 			       unsigned long hash_key)
80 {
81 	long diff;
82 
83 	diff = ((peer->hash_key - hash_key) ?:
84 		((unsigned long)peer->local - (unsigned long)local) ?:
85 		(peer->srx.transport_type - srx->transport_type) ?:
86 		(peer->srx.transport_len - srx->transport_len) ?:
87 		(peer->srx.transport.family - srx->transport.family));
88 	if (diff != 0)
89 		return diff;
90 
91 	switch (srx->transport.family) {
92 	case AF_INET:
93 		return ((u16 __force)peer->srx.transport.sin.sin_port -
94 			(u16 __force)srx->transport.sin.sin_port) ?:
95 			memcmp(&peer->srx.transport.sin.sin_addr,
96 			       &srx->transport.sin.sin_addr,
97 			       sizeof(struct in_addr));
98 #ifdef CONFIG_AF_RXRPC_IPV6
99 	case AF_INET6:
100 		return ((u16 __force)peer->srx.transport.sin6.sin6_port -
101 			(u16 __force)srx->transport.sin6.sin6_port) ?:
102 			memcmp(&peer->srx.transport.sin6.sin6_addr,
103 			       &srx->transport.sin6.sin6_addr,
104 			       sizeof(struct in6_addr));
105 #endif
106 	default:
107 		BUG();
108 	}
109 }
110 
111 /*
112  * Look up a remote transport endpoint for the specified address using RCU.
113  */
__rxrpc_lookup_peer_rcu(struct rxrpc_local * local,const struct sockaddr_rxrpc * srx,unsigned long hash_key)114 static struct rxrpc_peer *__rxrpc_lookup_peer_rcu(
115 	struct rxrpc_local *local,
116 	const struct sockaddr_rxrpc *srx,
117 	unsigned long hash_key)
118 {
119 	struct rxrpc_peer *peer;
120 	struct rxrpc_net *rxnet = local->rxnet;
121 
122 	hash_for_each_possible_rcu(rxnet->peer_hash, peer, hash_link, hash_key) {
123 		if (rxrpc_peer_cmp_key(peer, local, srx, hash_key) == 0 &&
124 		    refcount_read(&peer->ref) > 0)
125 			return peer;
126 	}
127 
128 	return NULL;
129 }
130 
131 /*
132  * Look up a remote transport endpoint for the specified address using RCU.
133  */
rxrpc_lookup_peer_rcu(struct rxrpc_local * local,const struct sockaddr_rxrpc * srx)134 struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local,
135 					 const struct sockaddr_rxrpc *srx)
136 {
137 	struct rxrpc_peer *peer;
138 	unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
139 
140 	peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
141 	if (peer) {
142 		_net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport);
143 		_leave(" = %p {u=%d}", peer, refcount_read(&peer->ref));
144 	}
145 	return peer;
146 }
147 
148 /*
149  * assess the MTU size for the network interface through which this peer is
150  * reached
151  */
rxrpc_assess_MTU_size(struct rxrpc_sock * rx,struct rxrpc_peer * peer)152 static void rxrpc_assess_MTU_size(struct rxrpc_sock *rx,
153 				  struct rxrpc_peer *peer)
154 {
155 	struct net *net = sock_net(&rx->sk);
156 	struct dst_entry *dst;
157 	struct rtable *rt;
158 	struct flowi fl;
159 	struct flowi4 *fl4 = &fl.u.ip4;
160 #ifdef CONFIG_AF_RXRPC_IPV6
161 	struct flowi6 *fl6 = &fl.u.ip6;
162 #endif
163 
164 	peer->if_mtu = 1500;
165 
166 	memset(&fl, 0, sizeof(fl));
167 	switch (peer->srx.transport.family) {
168 	case AF_INET:
169 		rt = ip_route_output_ports(
170 			net, fl4, NULL,
171 			peer->srx.transport.sin.sin_addr.s_addr, 0,
172 			htons(7000), htons(7001), IPPROTO_UDP, 0, 0);
173 		if (IS_ERR(rt)) {
174 			_leave(" [route err %ld]", PTR_ERR(rt));
175 			return;
176 		}
177 		dst = &rt->dst;
178 		break;
179 
180 #ifdef CONFIG_AF_RXRPC_IPV6
181 	case AF_INET6:
182 		fl6->flowi6_iif = LOOPBACK_IFINDEX;
183 		fl6->flowi6_scope = RT_SCOPE_UNIVERSE;
184 		fl6->flowi6_proto = IPPROTO_UDP;
185 		memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr,
186 		       sizeof(struct in6_addr));
187 		fl6->fl6_dport = htons(7001);
188 		fl6->fl6_sport = htons(7000);
189 		dst = ip6_route_output(net, NULL, fl6);
190 		if (dst->error) {
191 			_leave(" [route err %d]", dst->error);
192 			return;
193 		}
194 		break;
195 #endif
196 
197 	default:
198 		BUG();
199 	}
200 
201 	peer->if_mtu = dst_mtu(dst);
202 	dst_release(dst);
203 
204 	_leave(" [if_mtu %u]", peer->if_mtu);
205 }
206 
207 /*
208  * Allocate a peer.
209  */
rxrpc_alloc_peer(struct rxrpc_local * local,gfp_t gfp)210 struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp)
211 {
212 	const void *here = __builtin_return_address(0);
213 	struct rxrpc_peer *peer;
214 
215 	_enter("");
216 
217 	peer = kzalloc(sizeof(struct rxrpc_peer), gfp);
218 	if (peer) {
219 		refcount_set(&peer->ref, 1);
220 		peer->local = rxrpc_get_local(local);
221 		INIT_HLIST_HEAD(&peer->error_targets);
222 		peer->service_conns = RB_ROOT;
223 		seqlock_init(&peer->service_conn_lock);
224 		spin_lock_init(&peer->lock);
225 		spin_lock_init(&peer->rtt_input_lock);
226 		peer->debug_id = atomic_inc_return(&rxrpc_debug_id);
227 
228 		rxrpc_peer_init_rtt(peer);
229 
230 		if (RXRPC_TX_SMSS > 2190)
231 			peer->cong_cwnd = 2;
232 		else if (RXRPC_TX_SMSS > 1095)
233 			peer->cong_cwnd = 3;
234 		else
235 			peer->cong_cwnd = 4;
236 		trace_rxrpc_peer(peer->debug_id, rxrpc_peer_new, 1, here);
237 	}
238 
239 	_leave(" = %p", peer);
240 	return peer;
241 }
242 
243 /*
244  * Initialise peer record.
245  */
rxrpc_init_peer(struct rxrpc_sock * rx,struct rxrpc_peer * peer,unsigned long hash_key)246 static void rxrpc_init_peer(struct rxrpc_sock *rx, struct rxrpc_peer *peer,
247 			    unsigned long hash_key)
248 {
249 	peer->hash_key = hash_key;
250 	rxrpc_assess_MTU_size(rx, peer);
251 	peer->mtu = peer->if_mtu;
252 	peer->rtt_last_req = ktime_get_real();
253 
254 	switch (peer->srx.transport.family) {
255 	case AF_INET:
256 		peer->hdrsize = sizeof(struct iphdr);
257 		break;
258 #ifdef CONFIG_AF_RXRPC_IPV6
259 	case AF_INET6:
260 		peer->hdrsize = sizeof(struct ipv6hdr);
261 		break;
262 #endif
263 	default:
264 		BUG();
265 	}
266 
267 	switch (peer->srx.transport_type) {
268 	case SOCK_DGRAM:
269 		peer->hdrsize += sizeof(struct udphdr);
270 		break;
271 	default:
272 		BUG();
273 	}
274 
275 	peer->hdrsize += sizeof(struct rxrpc_wire_header);
276 	peer->maxdata = peer->mtu - peer->hdrsize;
277 }
278 
279 /*
280  * Set up a new peer.
281  */
rxrpc_create_peer(struct rxrpc_sock * rx,struct rxrpc_local * local,struct sockaddr_rxrpc * srx,unsigned long hash_key,gfp_t gfp)282 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_sock *rx,
283 					    struct rxrpc_local *local,
284 					    struct sockaddr_rxrpc *srx,
285 					    unsigned long hash_key,
286 					    gfp_t gfp)
287 {
288 	struct rxrpc_peer *peer;
289 
290 	_enter("");
291 
292 	peer = rxrpc_alloc_peer(local, gfp);
293 	if (peer) {
294 		memcpy(&peer->srx, srx, sizeof(*srx));
295 		rxrpc_init_peer(rx, peer, hash_key);
296 	}
297 
298 	_leave(" = %p", peer);
299 	return peer;
300 }
301 
rxrpc_free_peer(struct rxrpc_peer * peer)302 static void rxrpc_free_peer(struct rxrpc_peer *peer)
303 {
304 	rxrpc_put_local(peer->local);
305 	kfree_rcu(peer, rcu);
306 }
307 
308 /*
309  * Set up a new incoming peer.  There shouldn't be any other matching peers
310  * since we've already done a search in the list from the non-reentrant context
311  * (the data_ready handler) that is the only place we can add new peers.
312  */
rxrpc_new_incoming_peer(struct rxrpc_sock * rx,struct rxrpc_local * local,struct rxrpc_peer * peer)313 void rxrpc_new_incoming_peer(struct rxrpc_sock *rx, struct rxrpc_local *local,
314 			     struct rxrpc_peer *peer)
315 {
316 	struct rxrpc_net *rxnet = local->rxnet;
317 	unsigned long hash_key;
318 
319 	hash_key = rxrpc_peer_hash_key(local, &peer->srx);
320 	rxrpc_init_peer(rx, peer, hash_key);
321 
322 	spin_lock(&rxnet->peer_hash_lock);
323 	hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key);
324 	list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new);
325 	spin_unlock(&rxnet->peer_hash_lock);
326 }
327 
328 /*
329  * obtain a remote transport endpoint for the specified address
330  */
rxrpc_lookup_peer(struct rxrpc_sock * rx,struct rxrpc_local * local,struct sockaddr_rxrpc * srx,gfp_t gfp)331 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_sock *rx,
332 				     struct rxrpc_local *local,
333 				     struct sockaddr_rxrpc *srx, gfp_t gfp)
334 {
335 	struct rxrpc_peer *peer, *candidate;
336 	struct rxrpc_net *rxnet = local->rxnet;
337 	unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
338 
339 	_enter("{%pISp}", &srx->transport);
340 
341 	/* search the peer list first */
342 	rcu_read_lock();
343 	peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
344 	if (peer && !rxrpc_get_peer_maybe(peer))
345 		peer = NULL;
346 	rcu_read_unlock();
347 
348 	if (!peer) {
349 		/* The peer is not yet present in hash - create a candidate
350 		 * for a new record and then redo the search.
351 		 */
352 		candidate = rxrpc_create_peer(rx, local, srx, hash_key, gfp);
353 		if (!candidate) {
354 			_leave(" = NULL [nomem]");
355 			return NULL;
356 		}
357 
358 		spin_lock_bh(&rxnet->peer_hash_lock);
359 
360 		/* Need to check that we aren't racing with someone else */
361 		peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
362 		if (peer && !rxrpc_get_peer_maybe(peer))
363 			peer = NULL;
364 		if (!peer) {
365 			hash_add_rcu(rxnet->peer_hash,
366 				     &candidate->hash_link, hash_key);
367 			list_add_tail(&candidate->keepalive_link,
368 				      &rxnet->peer_keepalive_new);
369 		}
370 
371 		spin_unlock_bh(&rxnet->peer_hash_lock);
372 
373 		if (peer)
374 			rxrpc_free_peer(candidate);
375 		else
376 			peer = candidate;
377 	}
378 
379 	_net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport);
380 
381 	_leave(" = %p {u=%d}", peer, refcount_read(&peer->ref));
382 	return peer;
383 }
384 
385 /*
386  * Get a ref on a peer record.
387  */
rxrpc_get_peer(struct rxrpc_peer * peer)388 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer)
389 {
390 	const void *here = __builtin_return_address(0);
391 	int r;
392 
393 	__refcount_inc(&peer->ref, &r);
394 	trace_rxrpc_peer(peer->debug_id, rxrpc_peer_got, r + 1, here);
395 	return peer;
396 }
397 
398 /*
399  * Get a ref on a peer record unless its usage has already reached 0.
400  */
rxrpc_get_peer_maybe(struct rxrpc_peer * peer)401 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer)
402 {
403 	const void *here = __builtin_return_address(0);
404 	int r;
405 
406 	if (peer) {
407 		if (__refcount_inc_not_zero(&peer->ref, &r))
408 			trace_rxrpc_peer(peer->debug_id, rxrpc_peer_got, r + 1, here);
409 		else
410 			peer = NULL;
411 	}
412 	return peer;
413 }
414 
415 /*
416  * Discard a peer record.
417  */
__rxrpc_put_peer(struct rxrpc_peer * peer)418 static void __rxrpc_put_peer(struct rxrpc_peer *peer)
419 {
420 	struct rxrpc_net *rxnet = peer->local->rxnet;
421 
422 	ASSERT(hlist_empty(&peer->error_targets));
423 
424 	spin_lock_bh(&rxnet->peer_hash_lock);
425 	hash_del_rcu(&peer->hash_link);
426 	list_del_init(&peer->keepalive_link);
427 	spin_unlock_bh(&rxnet->peer_hash_lock);
428 
429 	rxrpc_free_peer(peer);
430 }
431 
432 /*
433  * Drop a ref on a peer record.
434  */
rxrpc_put_peer(struct rxrpc_peer * peer)435 void rxrpc_put_peer(struct rxrpc_peer *peer)
436 {
437 	const void *here = __builtin_return_address(0);
438 	unsigned int debug_id;
439 	bool dead;
440 	int r;
441 
442 	if (peer) {
443 		debug_id = peer->debug_id;
444 		dead = __refcount_dec_and_test(&peer->ref, &r);
445 		trace_rxrpc_peer(debug_id, rxrpc_peer_put, r - 1, here);
446 		if (dead)
447 			__rxrpc_put_peer(peer);
448 	}
449 }
450 
451 /*
452  * Drop a ref on a peer record where the caller already holds the
453  * peer_hash_lock.
454  */
rxrpc_put_peer_locked(struct rxrpc_peer * peer)455 void rxrpc_put_peer_locked(struct rxrpc_peer *peer)
456 {
457 	const void *here = __builtin_return_address(0);
458 	unsigned int debug_id = peer->debug_id;
459 	bool dead;
460 	int r;
461 
462 	dead = __refcount_dec_and_test(&peer->ref, &r);
463 	trace_rxrpc_peer(debug_id, rxrpc_peer_put, r - 1, here);
464 	if (dead) {
465 		hash_del_rcu(&peer->hash_link);
466 		list_del_init(&peer->keepalive_link);
467 		rxrpc_free_peer(peer);
468 	}
469 }
470 
471 /*
472  * Make sure all peer records have been discarded.
473  */
rxrpc_destroy_all_peers(struct rxrpc_net * rxnet)474 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet)
475 {
476 	struct rxrpc_peer *peer;
477 	int i;
478 
479 	for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) {
480 		if (hlist_empty(&rxnet->peer_hash[i]))
481 			continue;
482 
483 		hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) {
484 			pr_err("Leaked peer %u {%u} %pISp\n",
485 			       peer->debug_id,
486 			       refcount_read(&peer->ref),
487 			       &peer->srx.transport);
488 		}
489 	}
490 }
491 
492 /**
493  * rxrpc_kernel_get_peer - Get the peer address of a call
494  * @sock: The socket on which the call is in progress.
495  * @call: The call to query
496  * @_srx: Where to place the result
497  *
498  * Get the address of the remote peer in a call.
499  */
rxrpc_kernel_get_peer(struct socket * sock,struct rxrpc_call * call,struct sockaddr_rxrpc * _srx)500 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call,
501 			   struct sockaddr_rxrpc *_srx)
502 {
503 	*_srx = call->peer->srx;
504 }
505 EXPORT_SYMBOL(rxrpc_kernel_get_peer);
506 
507 /**
508  * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT
509  * @sock: The socket on which the call is in progress.
510  * @call: The call to query
511  * @_srtt: Where to store the SRTT value.
512  *
513  * Get the call's peer smoothed RTT in uS.
514  */
rxrpc_kernel_get_srtt(struct socket * sock,struct rxrpc_call * call,u32 * _srtt)515 bool rxrpc_kernel_get_srtt(struct socket *sock, struct rxrpc_call *call,
516 			   u32 *_srtt)
517 {
518 	struct rxrpc_peer *peer = call->peer;
519 
520 	if (peer->rtt_count == 0) {
521 		*_srtt = 1000000; /* 1S */
522 		return false;
523 	}
524 
525 	*_srtt = call->peer->srtt_us >> 3;
526 	return true;
527 }
528 EXPORT_SYMBOL(rxrpc_kernel_get_srtt);
529