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
25static const struct sockaddr_rxrpc rxrpc_null_addr;
26
27/*
28 * Hash a peer key.
29 */
30static unsigned long rxrpc_peer_hash_key(struct rxrpc_local *local,
31					 const struct sockaddr_rxrpc *srx)
32{
33	const u16 *p;
34	unsigned int i, size;
35	unsigned long hash_key;
36
37	_enter("");
38
39	hash_key = (unsigned long)local / __alignof__(*local);
40	hash_key += srx->transport_type;
41	hash_key += srx->transport_len;
42	hash_key += srx->transport.family;
43
44	switch (srx->transport.family) {
45	case AF_INET:
46		hash_key += (u16 __force)srx->transport.sin.sin_port;
47		size = sizeof(srx->transport.sin.sin_addr);
48		p = (u16 *)&srx->transport.sin.sin_addr;
49		break;
50#ifdef CONFIG_AF_RXRPC_IPV6
51	case AF_INET6:
52		hash_key += (u16 __force)srx->transport.sin.sin_port;
53		size = sizeof(srx->transport.sin6.sin6_addr);
54		p = (u16 *)&srx->transport.sin6.sin6_addr;
55		break;
56#endif
57	default:
58		WARN(1, "AF_RXRPC: Unsupported transport address family\n");
59		return 0;
60	}
61
62	/* Step through the peer address in 16-bit portions for speed */
63	for (i = 0; i < size; i += sizeof(*p), p++)
64		hash_key += *p;
65
66	_leave(" 0x%lx", hash_key);
67	return hash_key;
68}
69
70/*
71 * Compare a peer to a key.  Return -ve, 0 or +ve to indicate less than, same
72 * or greater than.
73 *
74 * Unfortunately, the primitives in linux/hashtable.h don't allow for sorted
75 * buckets and mid-bucket insertion, so we don't make full use of this
76 * information at this point.
77 */
78static long rxrpc_peer_cmp_key(const struct rxrpc_peer *peer,
79			       struct rxrpc_local *local,
80			       const struct sockaddr_rxrpc *srx,
81			       unsigned long hash_key)
82{
83	long diff;
84
85	diff = ((peer->hash_key - hash_key) ?:
86		((unsigned long)peer->local - (unsigned long)local) ?:
87		(peer->srx.transport_type - srx->transport_type) ?:
88		(peer->srx.transport_len - srx->transport_len) ?:
89		(peer->srx.transport.family - srx->transport.family));
90	if (diff != 0)
91		return diff;
92
93	switch (srx->transport.family) {
94	case AF_INET:
95		return ((u16 __force)peer->srx.transport.sin.sin_port -
96			(u16 __force)srx->transport.sin.sin_port) ?:
97			memcmp(&peer->srx.transport.sin.sin_addr,
98			       &srx->transport.sin.sin_addr,
99			       sizeof(struct in_addr));
100#ifdef CONFIG_AF_RXRPC_IPV6
101	case AF_INET6:
102		return ((u16 __force)peer->srx.transport.sin6.sin6_port -
103			(u16 __force)srx->transport.sin6.sin6_port) ?:
104			memcmp(&peer->srx.transport.sin6.sin6_addr,
105			       &srx->transport.sin6.sin6_addr,
106			       sizeof(struct in6_addr));
107#endif
108	default:
109		BUG();
110	}
111}
112
113/*
114 * Look up a remote transport endpoint for the specified address using RCU.
115 */
116static struct rxrpc_peer *__rxrpc_lookup_peer_rcu(
117	struct rxrpc_local *local,
118	const struct sockaddr_rxrpc *srx,
119	unsigned long hash_key)
120{
121	struct rxrpc_peer *peer;
122	struct rxrpc_net *rxnet = local->rxnet;
123
124	hash_for_each_possible_rcu(rxnet->peer_hash, peer, hash_link, hash_key) {
125		if (rxrpc_peer_cmp_key(peer, local, srx, hash_key) == 0 &&
126		    refcount_read(&peer->ref) > 0)
127			return peer;
128	}
129
130	return NULL;
131}
132
133/*
134 * Look up a remote transport endpoint for the specified address using RCU.
135 */
136struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local,
137					 const struct sockaddr_rxrpc *srx)
138{
139	struct rxrpc_peer *peer;
140	unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
141
142	peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
143	if (peer)
144		_leave(" = %p {u=%d}", peer, refcount_read(&peer->ref));
145	return peer;
146}
147
148/*
149 * assess the MTU size for the network interface through which this peer is
150 * reached
151 */
152static void rxrpc_assess_MTU_size(struct rxrpc_local *local,
153				  struct rxrpc_peer *peer)
154{
155	struct net *net = local->net;
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 */
210struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp,
211				    enum rxrpc_peer_trace why)
212{
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, rxrpc_local_get_peer);
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		peer->cong_ssthresh = RXRPC_TX_MAX_WINDOW;
231		trace_rxrpc_peer(peer->debug_id, 1, why);
232	}
233
234	_leave(" = %p", peer);
235	return peer;
236}
237
238/*
239 * Initialise peer record.
240 */
241static void rxrpc_init_peer(struct rxrpc_local *local, struct rxrpc_peer *peer,
242			    unsigned long hash_key)
243{
244	peer->hash_key = hash_key;
245	rxrpc_assess_MTU_size(local, peer);
246	peer->mtu = peer->if_mtu;
247	peer->rtt_last_req = ktime_get_real();
248
249	switch (peer->srx.transport.family) {
250	case AF_INET:
251		peer->hdrsize = sizeof(struct iphdr);
252		break;
253#ifdef CONFIG_AF_RXRPC_IPV6
254	case AF_INET6:
255		peer->hdrsize = sizeof(struct ipv6hdr);
256		break;
257#endif
258	default:
259		BUG();
260	}
261
262	switch (peer->srx.transport_type) {
263	case SOCK_DGRAM:
264		peer->hdrsize += sizeof(struct udphdr);
265		break;
266	default:
267		BUG();
268	}
269
270	peer->hdrsize += sizeof(struct rxrpc_wire_header);
271	peer->maxdata = peer->mtu - peer->hdrsize;
272}
273
274/*
275 * Set up a new peer.
276 */
277static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local,
278					    struct sockaddr_rxrpc *srx,
279					    unsigned long hash_key,
280					    gfp_t gfp)
281{
282	struct rxrpc_peer *peer;
283
284	_enter("");
285
286	peer = rxrpc_alloc_peer(local, gfp, rxrpc_peer_new_client);
287	if (peer) {
288		memcpy(&peer->srx, srx, sizeof(*srx));
289		rxrpc_init_peer(local, peer, hash_key);
290	}
291
292	_leave(" = %p", peer);
293	return peer;
294}
295
296static void rxrpc_free_peer(struct rxrpc_peer *peer)
297{
298	trace_rxrpc_peer(peer->debug_id, 0, rxrpc_peer_free);
299	rxrpc_put_local(peer->local, rxrpc_local_put_peer);
300	kfree_rcu(peer, rcu);
301}
302
303/*
304 * Set up a new incoming peer.  There shouldn't be any other matching peers
305 * since we've already done a search in the list from the non-reentrant context
306 * (the data_ready handler) that is the only place we can add new peers.
307 */
308void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer)
309{
310	struct rxrpc_net *rxnet = local->rxnet;
311	unsigned long hash_key;
312
313	hash_key = rxrpc_peer_hash_key(local, &peer->srx);
314	rxrpc_init_peer(local, peer, hash_key);
315
316	spin_lock(&rxnet->peer_hash_lock);
317	hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key);
318	list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new);
319	spin_unlock(&rxnet->peer_hash_lock);
320}
321
322/*
323 * obtain a remote transport endpoint for the specified address
324 */
325struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local,
326				     struct sockaddr_rxrpc *srx, gfp_t gfp)
327{
328	struct rxrpc_peer *peer, *candidate;
329	struct rxrpc_net *rxnet = local->rxnet;
330	unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
331
332	_enter("{%pISp}", &srx->transport);
333
334	/* search the peer list first */
335	rcu_read_lock();
336	peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
337	if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client))
338		peer = NULL;
339	rcu_read_unlock();
340
341	if (!peer) {
342		/* The peer is not yet present in hash - create a candidate
343		 * for a new record and then redo the search.
344		 */
345		candidate = rxrpc_create_peer(local, srx, hash_key, gfp);
346		if (!candidate) {
347			_leave(" = NULL [nomem]");
348			return NULL;
349		}
350
351		spin_lock(&rxnet->peer_hash_lock);
352
353		/* Need to check that we aren't racing with someone else */
354		peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
355		if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client))
356			peer = NULL;
357		if (!peer) {
358			hash_add_rcu(rxnet->peer_hash,
359				     &candidate->hash_link, hash_key);
360			list_add_tail(&candidate->keepalive_link,
361				      &rxnet->peer_keepalive_new);
362		}
363
364		spin_unlock(&rxnet->peer_hash_lock);
365
366		if (peer)
367			rxrpc_free_peer(candidate);
368		else
369			peer = candidate;
370	}
371
372	_leave(" = %p {u=%d}", peer, refcount_read(&peer->ref));
373	return peer;
374}
375
376/*
377 * Get a ref on a peer record.
378 */
379struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why)
380{
381	int r;
382
383	__refcount_inc(&peer->ref, &r);
384	trace_rxrpc_peer(peer->debug_id, r + 1, why);
385	return peer;
386}
387
388/*
389 * Get a ref on a peer record unless its usage has already reached 0.
390 */
391struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer,
392					enum rxrpc_peer_trace why)
393{
394	int r;
395
396	if (peer) {
397		if (__refcount_inc_not_zero(&peer->ref, &r))
398			trace_rxrpc_peer(peer->debug_id, r + 1, why);
399		else
400			peer = NULL;
401	}
402	return peer;
403}
404
405/*
406 * Discard a peer record.
407 */
408static void __rxrpc_put_peer(struct rxrpc_peer *peer)
409{
410	struct rxrpc_net *rxnet = peer->local->rxnet;
411
412	ASSERT(hlist_empty(&peer->error_targets));
413
414	spin_lock(&rxnet->peer_hash_lock);
415	hash_del_rcu(&peer->hash_link);
416	list_del_init(&peer->keepalive_link);
417	spin_unlock(&rxnet->peer_hash_lock);
418
419	rxrpc_free_peer(peer);
420}
421
422/*
423 * Drop a ref on a peer record.
424 */
425void rxrpc_put_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why)
426{
427	unsigned int debug_id;
428	bool dead;
429	int r;
430
431	if (peer) {
432		debug_id = peer->debug_id;
433		dead = __refcount_dec_and_test(&peer->ref, &r);
434		trace_rxrpc_peer(debug_id, r - 1, why);
435		if (dead)
436			__rxrpc_put_peer(peer);
437	}
438}
439
440/*
441 * Make sure all peer records have been discarded.
442 */
443void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet)
444{
445	struct rxrpc_peer *peer;
446	int i;
447
448	for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) {
449		if (hlist_empty(&rxnet->peer_hash[i]))
450			continue;
451
452		hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) {
453			pr_err("Leaked peer %u {%u} %pISp\n",
454			       peer->debug_id,
455			       refcount_read(&peer->ref),
456			       &peer->srx.transport);
457		}
458	}
459}
460
461/**
462 * rxrpc_kernel_get_call_peer - Get the peer address of a call
463 * @sock: The socket on which the call is in progress.
464 * @call: The call to query
465 *
466 * Get a record for the remote peer in a call.
467 */
468struct rxrpc_peer *rxrpc_kernel_get_call_peer(struct socket *sock, struct rxrpc_call *call)
469{
470	return call->peer;
471}
472EXPORT_SYMBOL(rxrpc_kernel_get_call_peer);
473
474/**
475 * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT
476 * @peer: The peer to query
477 *
478 * Get the call's peer smoothed RTT in uS or UINT_MAX if we have no samples.
479 */
480unsigned int rxrpc_kernel_get_srtt(const struct rxrpc_peer *peer)
481{
482	return peer->rtt_count > 0 ? peer->srtt_us >> 3 : UINT_MAX;
483}
484EXPORT_SYMBOL(rxrpc_kernel_get_srtt);
485
486/**
487 * rxrpc_kernel_remote_srx - Get the address of a peer
488 * @peer: The peer to query
489 *
490 * Get a pointer to the address from a peer record.  The caller is responsible
491 * for making sure that the address is not deallocated.
492 */
493const struct sockaddr_rxrpc *rxrpc_kernel_remote_srx(const struct rxrpc_peer *peer)
494{
495	return peer ? &peer->srx : &rxrpc_null_addr;
496}
497EXPORT_SYMBOL(rxrpc_kernel_remote_srx);
498
499/**
500 * rxrpc_kernel_remote_addr - Get the peer transport address of a call
501 * @peer: The peer to query
502 *
503 * Get a pointer to the transport address from a peer record.  The caller is
504 * responsible for making sure that the address is not deallocated.
505 */
506const struct sockaddr *rxrpc_kernel_remote_addr(const struct rxrpc_peer *peer)
507{
508	return (const struct sockaddr *)
509		(peer ? &peer->srx.transport : &rxrpc_null_addr.transport);
510}
511EXPORT_SYMBOL(rxrpc_kernel_remote_addr);
512