1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* incoming call handling
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/module.h>
11 #include <linux/net.h>
12 #include <linux/skbuff.h>
13 #include <linux/errqueue.h>
14 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/icmp.h>
18 #include <linux/gfp.h>
19 #include <linux/circ_buf.h>
21 #include <net/af_rxrpc.h>
23 #include "ar-internal.h"
25 static void rxrpc_dummy_notify(struct sock
*sk
, struct rxrpc_call
*call
,
26 unsigned long user_call_ID
)
31 * Preallocate a single service call, connection and peer and, if possible,
32 * give them a user ID and attach the user's side of the ID to them.
34 static int rxrpc_service_prealloc_one(struct rxrpc_sock
*rx
,
35 struct rxrpc_backlog
*b
,
36 rxrpc_notify_rx_t notify_rx
,
37 rxrpc_user_attach_call_t user_attach_call
,
38 unsigned long user_call_ID
, gfp_t gfp
,
39 unsigned int debug_id
)
41 const void *here
= __builtin_return_address(0);
42 struct rxrpc_call
*call
, *xcall
;
43 struct rxrpc_net
*rxnet
= rxrpc_net(sock_net(&rx
->sk
));
44 struct rb_node
*parent
, **pp
;
46 unsigned int size
= RXRPC_BACKLOG_MAX
;
47 unsigned int head
, tail
, call_head
, call_tail
;
49 max
= rx
->sk
.sk_max_ack_backlog
;
50 tmp
= rx
->sk
.sk_ack_backlog
;
52 _leave(" = -ENOBUFS [full %u]", max
);
57 /* We don't need more conns and peers than we have calls, but on the
58 * other hand, we shouldn't ever use more peers than conns or conns
61 call_head
= b
->call_backlog_head
;
62 call_tail
= READ_ONCE(b
->call_backlog_tail
);
63 tmp
= CIRC_CNT(call_head
, call_tail
, size
);
65 _leave(" = -ENOBUFS [enough %u]", tmp
);
70 head
= b
->peer_backlog_head
;
71 tail
= READ_ONCE(b
->peer_backlog_tail
);
72 if (CIRC_CNT(head
, tail
, size
) < max
) {
73 struct rxrpc_peer
*peer
= rxrpc_alloc_peer(rx
->local
, gfp
);
76 b
->peer_backlog
[head
] = peer
;
77 smp_store_release(&b
->peer_backlog_head
,
78 (head
+ 1) & (size
- 1));
81 head
= b
->conn_backlog_head
;
82 tail
= READ_ONCE(b
->conn_backlog_tail
);
83 if (CIRC_CNT(head
, tail
, size
) < max
) {
84 struct rxrpc_connection
*conn
;
86 conn
= rxrpc_prealloc_service_connection(rxnet
, gfp
);
89 b
->conn_backlog
[head
] = conn
;
90 smp_store_release(&b
->conn_backlog_head
,
91 (head
+ 1) & (size
- 1));
93 trace_rxrpc_conn(conn
->debug_id
, rxrpc_conn_new_service
,
94 atomic_read(&conn
->usage
), here
);
97 /* Now it gets complicated, because calls get registered with the
98 * socket here, with a user ID preassigned by the user.
100 call
= rxrpc_alloc_call(rx
, gfp
, debug_id
);
103 call
->flags
|= (1 << RXRPC_CALL_IS_SERVICE
);
104 call
->state
= RXRPC_CALL_SERVER_PREALLOC
;
106 trace_rxrpc_call(call
->debug_id
, rxrpc_call_new_service
,
107 atomic_read(&call
->usage
),
108 here
, (const void *)user_call_ID
);
110 write_lock(&rx
->call_lock
);
112 /* Check the user ID isn't already in use */
113 pp
= &rx
->calls
.rb_node
;
117 xcall
= rb_entry(parent
, struct rxrpc_call
, sock_node
);
118 if (user_call_ID
< xcall
->user_call_ID
)
119 pp
= &(*pp
)->rb_left
;
120 else if (user_call_ID
> xcall
->user_call_ID
)
121 pp
= &(*pp
)->rb_right
;
126 call
->user_call_ID
= user_call_ID
;
127 call
->notify_rx
= notify_rx
;
128 if (user_attach_call
) {
129 rxrpc_get_call(call
, rxrpc_call_got_kernel
);
130 user_attach_call(call
, user_call_ID
);
133 rxrpc_get_call(call
, rxrpc_call_got_userid
);
134 rb_link_node(&call
->sock_node
, parent
, pp
);
135 rb_insert_color(&call
->sock_node
, &rx
->calls
);
136 set_bit(RXRPC_CALL_HAS_USERID
, &call
->flags
);
138 list_add(&call
->sock_link
, &rx
->sock_calls
);
140 write_unlock(&rx
->call_lock
);
143 write_lock(&rxnet
->call_lock
);
144 list_add_tail(&call
->link
, &rxnet
->calls
);
145 write_unlock(&rxnet
->call_lock
);
147 b
->call_backlog
[call_head
] = call
;
148 smp_store_release(&b
->call_backlog_head
, (call_head
+ 1) & (size
- 1));
149 _leave(" = 0 [%d -> %lx]", call
->debug_id
, user_call_ID
);
153 write_unlock(&rx
->call_lock
);
154 rxrpc_cleanup_call(call
);
155 _leave(" = -EBADSLT");
160 * Allocate the preallocation buffers for incoming service calls. These must
161 * be charged manually.
163 int rxrpc_service_prealloc(struct rxrpc_sock
*rx
, gfp_t gfp
)
165 struct rxrpc_backlog
*b
= rx
->backlog
;
168 b
= kzalloc(sizeof(struct rxrpc_backlog
), gfp
);
178 * Discard the preallocation on a service.
180 void rxrpc_discard_prealloc(struct rxrpc_sock
*rx
)
182 struct rxrpc_backlog
*b
= rx
->backlog
;
183 struct rxrpc_net
*rxnet
= rxrpc_net(sock_net(&rx
->sk
));
184 unsigned int size
= RXRPC_BACKLOG_MAX
, head
, tail
;
190 /* Make sure that there aren't any incoming calls in progress before we
191 * clear the preallocation buffers.
193 spin_lock_bh(&rx
->incoming_lock
);
194 spin_unlock_bh(&rx
->incoming_lock
);
196 head
= b
->peer_backlog_head
;
197 tail
= b
->peer_backlog_tail
;
198 while (CIRC_CNT(head
, tail
, size
) > 0) {
199 struct rxrpc_peer
*peer
= b
->peer_backlog
[tail
];
200 rxrpc_put_local(peer
->local
);
202 tail
= (tail
+ 1) & (size
- 1);
205 head
= b
->conn_backlog_head
;
206 tail
= b
->conn_backlog_tail
;
207 while (CIRC_CNT(head
, tail
, size
) > 0) {
208 struct rxrpc_connection
*conn
= b
->conn_backlog
[tail
];
209 write_lock(&rxnet
->conn_lock
);
210 list_del(&conn
->link
);
211 list_del(&conn
->proc_link
);
212 write_unlock(&rxnet
->conn_lock
);
214 if (atomic_dec_and_test(&rxnet
->nr_conns
))
215 wake_up_var(&rxnet
->nr_conns
);
216 tail
= (tail
+ 1) & (size
- 1);
219 head
= b
->call_backlog_head
;
220 tail
= b
->call_backlog_tail
;
221 while (CIRC_CNT(head
, tail
, size
) > 0) {
222 struct rxrpc_call
*call
= b
->call_backlog
[tail
];
223 rcu_assign_pointer(call
->socket
, rx
);
224 if (rx
->discard_new_call
) {
225 _debug("discard %lx", call
->user_call_ID
);
226 rx
->discard_new_call(call
, call
->user_call_ID
);
228 call
->notify_rx
= rxrpc_dummy_notify
;
229 rxrpc_put_call(call
, rxrpc_call_put_kernel
);
231 rxrpc_call_completed(call
);
232 rxrpc_release_call(rx
, call
);
233 rxrpc_put_call(call
, rxrpc_call_put
);
234 tail
= (tail
+ 1) & (size
- 1);
241 * Ping the other end to fill our RTT cache and to retrieve the rwind
242 * and MTU parameters.
244 static void rxrpc_send_ping(struct rxrpc_call
*call
, struct sk_buff
*skb
)
246 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
247 ktime_t now
= skb
->tstamp
;
249 if (call
->peer
->rtt_count
< 3 ||
250 ktime_before(ktime_add_ms(call
->peer
->rtt_last_req
, 1000), now
))
251 rxrpc_propose_ACK(call
, RXRPC_ACK_PING
, sp
->hdr
.serial
,
253 rxrpc_propose_ack_ping_for_params
);
257 * Allocate a new incoming call from the prealloc pool, along with a connection
258 * and a peer as necessary.
260 static struct rxrpc_call
*rxrpc_alloc_incoming_call(struct rxrpc_sock
*rx
,
261 struct rxrpc_local
*local
,
262 struct rxrpc_peer
*peer
,
263 struct rxrpc_connection
*conn
,
264 const struct rxrpc_security
*sec
,
267 struct rxrpc_backlog
*b
= rx
->backlog
;
268 struct rxrpc_call
*call
;
269 unsigned short call_head
, conn_head
, peer_head
;
270 unsigned short call_tail
, conn_tail
, peer_tail
;
271 unsigned short call_count
, conn_count
;
273 /* #calls >= #conns >= #peers must hold true. */
274 call_head
= smp_load_acquire(&b
->call_backlog_head
);
275 call_tail
= b
->call_backlog_tail
;
276 call_count
= CIRC_CNT(call_head
, call_tail
, RXRPC_BACKLOG_MAX
);
277 conn_head
= smp_load_acquire(&b
->conn_backlog_head
);
278 conn_tail
= b
->conn_backlog_tail
;
279 conn_count
= CIRC_CNT(conn_head
, conn_tail
, RXRPC_BACKLOG_MAX
);
280 ASSERTCMP(conn_count
, >=, call_count
);
281 peer_head
= smp_load_acquire(&b
->peer_backlog_head
);
282 peer_tail
= b
->peer_backlog_tail
;
283 ASSERTCMP(CIRC_CNT(peer_head
, peer_tail
, RXRPC_BACKLOG_MAX
), >=,
290 if (peer
&& !rxrpc_get_peer_maybe(peer
))
293 peer
= b
->peer_backlog
[peer_tail
];
294 if (rxrpc_extract_addr_from_skb(&peer
->srx
, skb
) < 0)
296 b
->peer_backlog
[peer_tail
] = NULL
;
297 smp_store_release(&b
->peer_backlog_tail
,
299 (RXRPC_BACKLOG_MAX
- 1));
301 rxrpc_new_incoming_peer(rx
, local
, peer
);
304 /* Now allocate and set up the connection */
305 conn
= b
->conn_backlog
[conn_tail
];
306 b
->conn_backlog
[conn_tail
] = NULL
;
307 smp_store_release(&b
->conn_backlog_tail
,
308 (conn_tail
+ 1) & (RXRPC_BACKLOG_MAX
- 1));
309 conn
->params
.local
= rxrpc_get_local(local
);
310 conn
->params
.peer
= peer
;
311 rxrpc_see_connection(conn
);
312 rxrpc_new_incoming_connection(rx
, conn
, sec
, skb
);
314 rxrpc_get_connection(conn
);
317 /* And now we can allocate and set up a new call */
318 call
= b
->call_backlog
[call_tail
];
319 b
->call_backlog
[call_tail
] = NULL
;
320 smp_store_release(&b
->call_backlog_tail
,
321 (call_tail
+ 1) & (RXRPC_BACKLOG_MAX
- 1));
323 rxrpc_see_call(call
);
325 call
->security
= conn
->security
;
326 call
->security_ix
= conn
->security_ix
;
327 call
->peer
= rxrpc_get_peer(conn
->params
.peer
);
328 call
->cong_cwnd
= call
->peer
->cong_cwnd
;
333 * Set up a new incoming call. Called in BH context with the RCU read lock
336 * If this is for a kernel service, when we allocate the call, it will have
337 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
338 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace
339 * services only have the ref from the backlog buffer. We want to pass this
340 * ref to non-BH context to dispose of.
342 * If we want to report an error, we mark the skb with the packet type and
343 * abort code and return NULL.
345 * The call is returned with the user access mutex held.
347 struct rxrpc_call
*rxrpc_new_incoming_call(struct rxrpc_local
*local
,
348 struct rxrpc_sock
*rx
,
351 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
352 const struct rxrpc_security
*sec
= NULL
;
353 struct rxrpc_connection
*conn
;
354 struct rxrpc_peer
*peer
= NULL
;
355 struct rxrpc_call
*call
= NULL
;
359 spin_lock(&rx
->incoming_lock
);
360 if (rx
->sk
.sk_state
== RXRPC_SERVER_LISTEN_DISABLED
||
361 rx
->sk
.sk_state
== RXRPC_CLOSE
) {
362 trace_rxrpc_abort(0, "CLS", sp
->hdr
.cid
, sp
->hdr
.callNumber
,
363 sp
->hdr
.seq
, RX_INVALID_OPERATION
, ESHUTDOWN
);
364 skb
->mark
= RXRPC_SKB_MARK_REJECT_ABORT
;
365 skb
->priority
= RX_INVALID_OPERATION
;
369 /* The peer, connection and call may all have sprung into existence due
370 * to a duplicate packet being handled on another CPU in parallel, so
371 * we have to recheck the routing. However, we're now holding
372 * rx->incoming_lock, so the values should remain stable.
374 conn
= rxrpc_find_connection_rcu(local
, skb
, &peer
);
377 sec
= rxrpc_get_incoming_security(rx
, skb
);
382 call
= rxrpc_alloc_incoming_call(rx
, local
, peer
, conn
, sec
, skb
);
384 skb
->mark
= RXRPC_SKB_MARK_REJECT_BUSY
;
388 trace_rxrpc_receive(call
, rxrpc_receive_incoming
,
389 sp
->hdr
.serial
, sp
->hdr
.seq
);
391 /* Make the call live. */
392 rxrpc_incoming_call(rx
, call
, skb
);
395 if (rx
->notify_new_call
)
396 rx
->notify_new_call(&rx
->sk
, call
, call
->user_call_ID
);
398 spin_lock(&conn
->state_lock
);
399 switch (conn
->state
) {
400 case RXRPC_CONN_SERVICE_UNSECURED
:
401 conn
->state
= RXRPC_CONN_SERVICE_CHALLENGING
;
402 set_bit(RXRPC_CONN_EV_CHALLENGE
, &call
->conn
->events
);
403 rxrpc_queue_conn(call
->conn
);
406 case RXRPC_CONN_SERVICE
:
407 write_lock(&call
->state_lock
);
408 if (call
->state
< RXRPC_CALL_COMPLETE
)
409 call
->state
= RXRPC_CALL_SERVER_RECV_REQUEST
;
410 write_unlock(&call
->state_lock
);
413 case RXRPC_CONN_REMOTELY_ABORTED
:
414 rxrpc_set_call_completion(call
, RXRPC_CALL_REMOTELY_ABORTED
,
415 conn
->abort_code
, conn
->error
);
417 case RXRPC_CONN_LOCALLY_ABORTED
:
418 rxrpc_abort_call("CON", call
, sp
->hdr
.seq
,
419 conn
->abort_code
, conn
->error
);
424 spin_unlock(&conn
->state_lock
);
425 spin_unlock(&rx
->incoming_lock
);
427 rxrpc_send_ping(call
, skb
);
429 /* We have to discard the prealloc queue's ref here and rely on a
430 * combination of the RCU read lock and refs held either by the socket
431 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel
432 * service to prevent the call from being deallocated too early.
434 rxrpc_put_call(call
, rxrpc_call_put
);
436 _leave(" = %p{%d}", call
, call
->debug_id
);
440 spin_unlock(&rx
->incoming_lock
);
441 _leave(" = NULL [%u]", skb
->mark
);
446 * Charge up socket with preallocated calls, attaching user call IDs.
448 int rxrpc_user_charge_accept(struct rxrpc_sock
*rx
, unsigned long user_call_ID
)
450 struct rxrpc_backlog
*b
= rx
->backlog
;
452 if (rx
->sk
.sk_state
== RXRPC_CLOSE
)
455 return rxrpc_service_prealloc_one(rx
, b
, NULL
, NULL
, user_call_ID
,
457 atomic_inc_return(&rxrpc_debug_id
));
461 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
462 * @sock: The socket on which to preallocate
463 * @notify_rx: Event notification function for the call
464 * @user_attach_call: Func to attach call to user_call_ID
465 * @user_call_ID: The tag to attach to the preallocated call
466 * @gfp: The allocation conditions.
467 * @debug_id: The tracing debug ID.
469 * Charge up the socket with preallocated calls, each with a user ID. A
470 * function should be provided to effect the attachment from the user's side.
471 * The user is given a ref to hold on the call.
473 * Note that the call may be come connected before this function returns.
475 int rxrpc_kernel_charge_accept(struct socket
*sock
,
476 rxrpc_notify_rx_t notify_rx
,
477 rxrpc_user_attach_call_t user_attach_call
,
478 unsigned long user_call_ID
, gfp_t gfp
,
479 unsigned int debug_id
)
481 struct rxrpc_sock
*rx
= rxrpc_sk(sock
->sk
);
482 struct rxrpc_backlog
*b
= rx
->backlog
;
484 if (sock
->sk
->sk_state
== RXRPC_CLOSE
)
487 return rxrpc_service_prealloc_one(rx
, b
, notify_rx
,
488 user_attach_call
, user_call_ID
,
491 EXPORT_SYMBOL(rxrpc_kernel_charge_accept
);