Merge tag 'pci-v5.12-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
[linux-2.6-microblaze.git] / net / rxrpc / call_accept.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* incoming call handling
3  *
4  * Copyright (C) 2007 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/errqueue.h>
14 #include <linux/udp.h>
15 #include <linux/in.h>
16 #include <linux/in6.h>
17 #include <linux/icmp.h>
18 #include <linux/gfp.h>
19 #include <linux/circ_buf.h>
20 #include <net/sock.h>
21 #include <net/af_rxrpc.h>
22 #include <net/ip.h>
23 #include "ar-internal.h"
24
25 static void rxrpc_dummy_notify(struct sock *sk, struct rxrpc_call *call,
26                                unsigned long user_call_ID)
27 {
28 }
29
30 /*
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.
33  */
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)
40 {
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;
45         int max, tmp;
46         unsigned int size = RXRPC_BACKLOG_MAX;
47         unsigned int head, tail, call_head, call_tail;
48
49         max = rx->sk.sk_max_ack_backlog;
50         tmp = rx->sk.sk_ack_backlog;
51         if (tmp >= max) {
52                 _leave(" = -ENOBUFS [full %u]", max);
53                 return -ENOBUFS;
54         }
55         max -= tmp;
56
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
59          * than calls.
60          */
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);
64         if (tmp >= max) {
65                 _leave(" = -ENOBUFS [enough %u]", tmp);
66                 return -ENOBUFS;
67         }
68         max = tmp + 1;
69
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);
74                 if (!peer)
75                         return -ENOMEM;
76                 b->peer_backlog[head] = peer;
77                 smp_store_release(&b->peer_backlog_head,
78                                   (head + 1) & (size - 1));
79         }
80
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;
85
86                 conn = rxrpc_prealloc_service_connection(rxnet, gfp);
87                 if (!conn)
88                         return -ENOMEM;
89                 b->conn_backlog[head] = conn;
90                 smp_store_release(&b->conn_backlog_head,
91                                   (head + 1) & (size - 1));
92
93                 trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_service,
94                                  atomic_read(&conn->usage), here);
95         }
96
97         /* Now it gets complicated, because calls get registered with the
98          * socket here, with a user ID preassigned by the user.
99          */
100         call = rxrpc_alloc_call(rx, gfp, debug_id);
101         if (!call)
102                 return -ENOMEM;
103         call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
104         call->state = RXRPC_CALL_SERVER_PREALLOC;
105
106         trace_rxrpc_call(call->debug_id, rxrpc_call_new_service,
107                          atomic_read(&call->usage),
108                          here, (const void *)user_call_ID);
109
110         write_lock(&rx->call_lock);
111
112         /* Check the user ID isn't already in use */
113         pp = &rx->calls.rb_node;
114         parent = NULL;
115         while (*pp) {
116                 parent = *pp;
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;
122                 else
123                         goto id_in_use;
124         }
125
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);
131         }
132
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);
137
138         list_add(&call->sock_link, &rx->sock_calls);
139
140         write_unlock(&rx->call_lock);
141
142         rxnet = call->rxnet;
143         write_lock(&rxnet->call_lock);
144         list_add_tail(&call->link, &rxnet->calls);
145         write_unlock(&rxnet->call_lock);
146
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);
150         return 0;
151
152 id_in_use:
153         write_unlock(&rx->call_lock);
154         rxrpc_cleanup_call(call);
155         _leave(" = -EBADSLT");
156         return -EBADSLT;
157 }
158
159 /*
160  * Allocate the preallocation buffers for incoming service calls.  These must
161  * be charged manually.
162  */
163 int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
164 {
165         struct rxrpc_backlog *b = rx->backlog;
166
167         if (!b) {
168                 b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
169                 if (!b)
170                         return -ENOMEM;
171                 rx->backlog = b;
172         }
173
174         return 0;
175 }
176
177 /*
178  * Discard the preallocation on a service.
179  */
180 void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
181 {
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;
185
186         if (!b)
187                 return;
188         rx->backlog = NULL;
189
190         /* Make sure that there aren't any incoming calls in progress before we
191          * clear the preallocation buffers.
192          */
193         spin_lock_bh(&rx->incoming_lock);
194         spin_unlock_bh(&rx->incoming_lock);
195
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);
201                 kfree(peer);
202                 tail = (tail + 1) & (size - 1);
203         }
204
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);
213                 kfree(conn);
214                 if (atomic_dec_and_test(&rxnet->nr_conns))
215                         wake_up_var(&rxnet->nr_conns);
216                 tail = (tail + 1) & (size - 1);
217         }
218
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);
227                         if (call->notify_rx)
228                                 call->notify_rx = rxrpc_dummy_notify;
229                         rxrpc_put_call(call, rxrpc_call_put_kernel);
230                 }
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);
235         }
236
237         kfree(b);
238 }
239
240 /*
241  * Ping the other end to fill our RTT cache and to retrieve the rwind
242  * and MTU parameters.
243  */
244 static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb)
245 {
246         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
247         ktime_t now = skb->tstamp;
248
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,
252                                   true, true,
253                                   rxrpc_propose_ack_ping_for_params);
254 }
255
256 /*
257  * Allocate a new incoming call from the prealloc pool, along with a connection
258  * and a peer as necessary.
259  */
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,
265                                                     struct sk_buff *skb)
266 {
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;
272
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), >=,
284                   conn_count);
285
286         if (call_count == 0)
287                 return NULL;
288
289         if (!conn) {
290                 if (peer && !rxrpc_get_peer_maybe(peer))
291                         peer = NULL;
292                 if (!peer) {
293                         peer = b->peer_backlog[peer_tail];
294                         if (rxrpc_extract_addr_from_skb(&peer->srx, skb) < 0)
295                                 return NULL;
296                         b->peer_backlog[peer_tail] = NULL;
297                         smp_store_release(&b->peer_backlog_tail,
298                                           (peer_tail + 1) &
299                                           (RXRPC_BACKLOG_MAX - 1));
300
301                         rxrpc_new_incoming_peer(rx, local, peer);
302                 }
303
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);
313         } else {
314                 rxrpc_get_connection(conn);
315         }
316
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));
322
323         rxrpc_see_call(call);
324         call->conn = conn;
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;
329         return call;
330 }
331
332 /*
333  * Set up a new incoming call.  Called in BH context with the RCU read lock
334  * held.
335  *
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.
341  *
342  * If we want to report an error, we mark the skb with the packet type and
343  * abort code and return NULL.
344  *
345  * The call is returned with the user access mutex held.
346  */
347 struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
348                                            struct rxrpc_sock *rx,
349                                            struct sk_buff *skb)
350 {
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;
356
357         _enter("");
358
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;
366                 goto no_call;
367         }
368
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.
373          */
374         conn = rxrpc_find_connection_rcu(local, skb, &peer);
375
376         if (!conn) {
377                 sec = rxrpc_get_incoming_security(rx, skb);
378                 if (!sec)
379                         goto no_call;
380         }
381
382         call = rxrpc_alloc_incoming_call(rx, local, peer, conn, sec, skb);
383         if (!call) {
384                 skb->mark = RXRPC_SKB_MARK_REJECT_BUSY;
385                 goto no_call;
386         }
387
388         trace_rxrpc_receive(call, rxrpc_receive_incoming,
389                             sp->hdr.serial, sp->hdr.seq);
390
391         /* Make the call live. */
392         rxrpc_incoming_call(rx, call, skb);
393         conn = call->conn;
394
395         if (rx->notify_new_call)
396                 rx->notify_new_call(&rx->sk, call, call->user_call_ID);
397
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);
404                 break;
405
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);
411                 break;
412
413         case RXRPC_CONN_REMOTELY_ABORTED:
414                 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
415                                           conn->abort_code, conn->error);
416                 break;
417         case RXRPC_CONN_LOCALLY_ABORTED:
418                 rxrpc_abort_call("CON", call, sp->hdr.seq,
419                                  conn->abort_code, conn->error);
420                 break;
421         default:
422                 BUG();
423         }
424         spin_unlock(&conn->state_lock);
425         spin_unlock(&rx->incoming_lock);
426
427         rxrpc_send_ping(call, skb);
428
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.
433          */
434         rxrpc_put_call(call, rxrpc_call_put);
435
436         _leave(" = %p{%d}", call, call->debug_id);
437         return call;
438
439 no_call:
440         spin_unlock(&rx->incoming_lock);
441         _leave(" = NULL [%u]", skb->mark);
442         return NULL;
443 }
444
445 /*
446  * Charge up socket with preallocated calls, attaching user call IDs.
447  */
448 int rxrpc_user_charge_accept(struct rxrpc_sock *rx, unsigned long user_call_ID)
449 {
450         struct rxrpc_backlog *b = rx->backlog;
451
452         if (rx->sk.sk_state == RXRPC_CLOSE)
453                 return -ESHUTDOWN;
454
455         return rxrpc_service_prealloc_one(rx, b, NULL, NULL, user_call_ID,
456                                           GFP_KERNEL,
457                                           atomic_inc_return(&rxrpc_debug_id));
458 }
459
460 /*
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.
468  *
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.
472  *
473  * Note that the call may be come connected before this function returns.
474  */
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)
480 {
481         struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
482         struct rxrpc_backlog *b = rx->backlog;
483
484         if (sock->sk->sk_state == RXRPC_CLOSE)
485                 return -ESHUTDOWN;
486
487         return rxrpc_service_prealloc_one(rx, b, notify_rx,
488                                           user_attach_call, user_call_ID,
489                                           gfp, debug_id);
490 }
491 EXPORT_SYMBOL(rxrpc_kernel_charge_accept);