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 | */ |
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 | */ |
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(p: &peer->srx.transport.sin.sin_addr, |
96 | q: &srx->transport.sin.sin_addr, |
97 | size: 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(p: &peer->srx.transport.sin6.sin6_addr, |
103 | q: &srx->transport.sin6.sin6_addr, |
104 | size: 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 | */ |
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(r: &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 | */ |
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 | _leave(" = %p {u=%d}" , peer, refcount_read(&peer->ref)); |
143 | return peer; |
144 | } |
145 | |
146 | /* |
147 | * assess the MTU size for the network interface through which this peer is |
148 | * reached |
149 | */ |
150 | static void rxrpc_assess_MTU_size(struct rxrpc_local *local, |
151 | struct rxrpc_peer *peer) |
152 | { |
153 | struct net *net = local->net; |
154 | struct dst_entry *dst; |
155 | struct rtable *rt; |
156 | struct flowi fl; |
157 | struct flowi4 *fl4 = &fl.u.ip4; |
158 | #ifdef CONFIG_AF_RXRPC_IPV6 |
159 | struct flowi6 *fl6 = &fl.u.ip6; |
160 | #endif |
161 | |
162 | peer->if_mtu = 1500; |
163 | |
164 | memset(&fl, 0, sizeof(fl)); |
165 | switch (peer->srx.transport.family) { |
166 | case AF_INET: |
167 | rt = ip_route_output_ports( |
168 | net, fl4, NULL, |
169 | daddr: peer->srx.transport.sin.sin_addr.s_addr, saddr: 0, |
170 | htons(7000), htons(7001), IPPROTO_UDP, tos: 0, oif: 0); |
171 | if (IS_ERR(ptr: rt)) { |
172 | _leave(" [route err %ld]" , PTR_ERR(rt)); |
173 | return; |
174 | } |
175 | dst = &rt->dst; |
176 | break; |
177 | |
178 | #ifdef CONFIG_AF_RXRPC_IPV6 |
179 | case AF_INET6: |
180 | fl6->flowi6_iif = LOOPBACK_IFINDEX; |
181 | fl6->flowi6_scope = RT_SCOPE_UNIVERSE; |
182 | fl6->flowi6_proto = IPPROTO_UDP; |
183 | memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr, |
184 | sizeof(struct in6_addr)); |
185 | fl6->fl6_dport = htons(7001); |
186 | fl6->fl6_sport = htons(7000); |
187 | dst = ip6_route_output(net, NULL, fl6); |
188 | if (dst->error) { |
189 | _leave(" [route err %d]" , dst->error); |
190 | return; |
191 | } |
192 | break; |
193 | #endif |
194 | |
195 | default: |
196 | BUG(); |
197 | } |
198 | |
199 | peer->if_mtu = dst_mtu(dst); |
200 | dst_release(dst); |
201 | |
202 | _leave(" [if_mtu %u]" , peer->if_mtu); |
203 | } |
204 | |
205 | /* |
206 | * Allocate a peer. |
207 | */ |
208 | struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp, |
209 | enum rxrpc_peer_trace why) |
210 | { |
211 | struct rxrpc_peer *peer; |
212 | |
213 | _enter("" ); |
214 | |
215 | peer = kzalloc(size: sizeof(struct rxrpc_peer), flags: gfp); |
216 | if (peer) { |
217 | refcount_set(r: &peer->ref, n: 1); |
218 | peer->local = rxrpc_get_local(local, rxrpc_local_get_peer); |
219 | INIT_HLIST_HEAD(&peer->error_targets); |
220 | peer->service_conns = RB_ROOT; |
221 | seqlock_init(&peer->service_conn_lock); |
222 | spin_lock_init(&peer->lock); |
223 | spin_lock_init(&peer->rtt_input_lock); |
224 | peer->debug_id = atomic_inc_return(v: &rxrpc_debug_id); |
225 | |
226 | rxrpc_peer_init_rtt(peer); |
227 | |
228 | peer->cong_ssthresh = RXRPC_TX_MAX_WINDOW; |
229 | trace_rxrpc_peer(peer_debug_id: peer->debug_id, ref: 1, why); |
230 | } |
231 | |
232 | _leave(" = %p" , peer); |
233 | return peer; |
234 | } |
235 | |
236 | /* |
237 | * Initialise peer record. |
238 | */ |
239 | static void rxrpc_init_peer(struct rxrpc_local *local, struct rxrpc_peer *peer, |
240 | unsigned long hash_key) |
241 | { |
242 | peer->hash_key = hash_key; |
243 | rxrpc_assess_MTU_size(local, peer); |
244 | peer->mtu = peer->if_mtu; |
245 | peer->rtt_last_req = ktime_get_real(); |
246 | |
247 | switch (peer->srx.transport.family) { |
248 | case AF_INET: |
249 | peer->hdrsize = sizeof(struct iphdr); |
250 | break; |
251 | #ifdef CONFIG_AF_RXRPC_IPV6 |
252 | case AF_INET6: |
253 | peer->hdrsize = sizeof(struct ipv6hdr); |
254 | break; |
255 | #endif |
256 | default: |
257 | BUG(); |
258 | } |
259 | |
260 | switch (peer->srx.transport_type) { |
261 | case SOCK_DGRAM: |
262 | peer->hdrsize += sizeof(struct udphdr); |
263 | break; |
264 | default: |
265 | BUG(); |
266 | } |
267 | |
268 | peer->hdrsize += sizeof(struct rxrpc_wire_header); |
269 | peer->maxdata = peer->mtu - peer->hdrsize; |
270 | } |
271 | |
272 | /* |
273 | * Set up a new peer. |
274 | */ |
275 | static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local, |
276 | struct sockaddr_rxrpc *srx, |
277 | unsigned long hash_key, |
278 | gfp_t gfp) |
279 | { |
280 | struct rxrpc_peer *peer; |
281 | |
282 | _enter("" ); |
283 | |
284 | peer = rxrpc_alloc_peer(local, gfp, why: rxrpc_peer_new_client); |
285 | if (peer) { |
286 | memcpy(&peer->srx, srx, sizeof(*srx)); |
287 | rxrpc_init_peer(local, peer, hash_key); |
288 | } |
289 | |
290 | _leave(" = %p" , peer); |
291 | return peer; |
292 | } |
293 | |
294 | static void rxrpc_free_peer(struct rxrpc_peer *peer) |
295 | { |
296 | trace_rxrpc_peer(peer_debug_id: peer->debug_id, ref: 0, why: rxrpc_peer_free); |
297 | rxrpc_put_local(peer->local, rxrpc_local_put_peer); |
298 | kfree_rcu(peer, rcu); |
299 | } |
300 | |
301 | /* |
302 | * Set up a new incoming peer. There shouldn't be any other matching peers |
303 | * since we've already done a search in the list from the non-reentrant context |
304 | * (the data_ready handler) that is the only place we can add new peers. |
305 | */ |
306 | void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer) |
307 | { |
308 | struct rxrpc_net *rxnet = local->rxnet; |
309 | unsigned long hash_key; |
310 | |
311 | hash_key = rxrpc_peer_hash_key(local, srx: &peer->srx); |
312 | rxrpc_init_peer(local, peer, hash_key); |
313 | |
314 | spin_lock(lock: &rxnet->peer_hash_lock); |
315 | hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); |
316 | list_add_tail(new: &peer->keepalive_link, head: &rxnet->peer_keepalive_new); |
317 | spin_unlock(lock: &rxnet->peer_hash_lock); |
318 | } |
319 | |
320 | /* |
321 | * obtain a remote transport endpoint for the specified address |
322 | */ |
323 | struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local, |
324 | struct sockaddr_rxrpc *srx, gfp_t gfp) |
325 | { |
326 | struct rxrpc_peer *peer, *candidate; |
327 | struct rxrpc_net *rxnet = local->rxnet; |
328 | unsigned long hash_key = rxrpc_peer_hash_key(local, srx); |
329 | |
330 | _enter("{%pISp}" , &srx->transport); |
331 | |
332 | /* search the peer list first */ |
333 | rcu_read_lock(); |
334 | peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); |
335 | if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) |
336 | peer = NULL; |
337 | rcu_read_unlock(); |
338 | |
339 | if (!peer) { |
340 | /* The peer is not yet present in hash - create a candidate |
341 | * for a new record and then redo the search. |
342 | */ |
343 | candidate = rxrpc_create_peer(local, srx, hash_key, gfp); |
344 | if (!candidate) { |
345 | _leave(" = NULL [nomem]" ); |
346 | return NULL; |
347 | } |
348 | |
349 | spin_lock(lock: &rxnet->peer_hash_lock); |
350 | |
351 | /* Need to check that we aren't racing with someone else */ |
352 | peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); |
353 | if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) |
354 | peer = NULL; |
355 | if (!peer) { |
356 | hash_add_rcu(rxnet->peer_hash, |
357 | &candidate->hash_link, hash_key); |
358 | list_add_tail(new: &candidate->keepalive_link, |
359 | head: &rxnet->peer_keepalive_new); |
360 | } |
361 | |
362 | spin_unlock(lock: &rxnet->peer_hash_lock); |
363 | |
364 | if (peer) |
365 | rxrpc_free_peer(peer: candidate); |
366 | else |
367 | peer = candidate; |
368 | } |
369 | |
370 | _leave(" = %p {u=%d}" , peer, refcount_read(&peer->ref)); |
371 | return peer; |
372 | } |
373 | |
374 | /* |
375 | * Get a ref on a peer record. |
376 | */ |
377 | struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) |
378 | { |
379 | int r; |
380 | |
381 | __refcount_inc(r: &peer->ref, oldp: &r); |
382 | trace_rxrpc_peer(peer_debug_id: peer->debug_id, ref: r + 1, why); |
383 | return peer; |
384 | } |
385 | |
386 | /* |
387 | * Get a ref on a peer record unless its usage has already reached 0. |
388 | */ |
389 | struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer, |
390 | enum rxrpc_peer_trace why) |
391 | { |
392 | int r; |
393 | |
394 | if (peer) { |
395 | if (__refcount_inc_not_zero(r: &peer->ref, oldp: &r)) |
396 | trace_rxrpc_peer(peer_debug_id: peer->debug_id, ref: r + 1, why); |
397 | else |
398 | peer = NULL; |
399 | } |
400 | return peer; |
401 | } |
402 | |
403 | /* |
404 | * Discard a peer record. |
405 | */ |
406 | static void __rxrpc_put_peer(struct rxrpc_peer *peer) |
407 | { |
408 | struct rxrpc_net *rxnet = peer->local->rxnet; |
409 | |
410 | ASSERT(hlist_empty(&peer->error_targets)); |
411 | |
412 | spin_lock(lock: &rxnet->peer_hash_lock); |
413 | hash_del_rcu(node: &peer->hash_link); |
414 | list_del_init(entry: &peer->keepalive_link); |
415 | spin_unlock(lock: &rxnet->peer_hash_lock); |
416 | |
417 | rxrpc_free_peer(peer); |
418 | } |
419 | |
420 | /* |
421 | * Drop a ref on a peer record. |
422 | */ |
423 | void rxrpc_put_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) |
424 | { |
425 | unsigned int debug_id; |
426 | bool dead; |
427 | int r; |
428 | |
429 | if (peer) { |
430 | debug_id = peer->debug_id; |
431 | dead = __refcount_dec_and_test(r: &peer->ref, oldp: &r); |
432 | trace_rxrpc_peer(peer_debug_id: debug_id, ref: r - 1, why); |
433 | if (dead) |
434 | __rxrpc_put_peer(peer); |
435 | } |
436 | } |
437 | |
438 | /* |
439 | * Make sure all peer records have been discarded. |
440 | */ |
441 | void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) |
442 | { |
443 | struct rxrpc_peer *peer; |
444 | int i; |
445 | |
446 | for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { |
447 | if (hlist_empty(h: &rxnet->peer_hash[i])) |
448 | continue; |
449 | |
450 | hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { |
451 | pr_err("Leaked peer %u {%u} %pISp\n" , |
452 | peer->debug_id, |
453 | refcount_read(&peer->ref), |
454 | &peer->srx.transport); |
455 | } |
456 | } |
457 | } |
458 | |
459 | /** |
460 | * rxrpc_kernel_get_peer - Get the peer address of a call |
461 | * @sock: The socket on which the call is in progress. |
462 | * @call: The call to query |
463 | * @_srx: Where to place the result |
464 | * |
465 | * Get the address of the remote peer in a call. |
466 | */ |
467 | void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call, |
468 | struct sockaddr_rxrpc *_srx) |
469 | { |
470 | *_srx = call->peer->srx; |
471 | } |
472 | EXPORT_SYMBOL(rxrpc_kernel_get_peer); |
473 | |
474 | /** |
475 | * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT |
476 | * @sock: The socket on which the call is in progress. |
477 | * @call: The call to query |
478 | * @_srtt: Where to store the SRTT value. |
479 | * |
480 | * Get the call's peer smoothed RTT in uS. |
481 | */ |
482 | bool rxrpc_kernel_get_srtt(struct socket *sock, struct rxrpc_call *call, |
483 | u32 *_srtt) |
484 | { |
485 | struct rxrpc_peer *peer = call->peer; |
486 | |
487 | if (peer->rtt_count == 0) { |
488 | *_srtt = 1000000; /* 1S */ |
489 | return false; |
490 | } |
491 | |
492 | *_srtt = call->peer->srtt_us >> 3; |
493 | return true; |
494 | } |
495 | EXPORT_SYMBOL(rxrpc_kernel_get_srtt); |
496 | |