1 /* incoming call handling 2 * 3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 13 14 #include <linux/module.h> 15 #include <linux/net.h> 16 #include <linux/skbuff.h> 17 #include <linux/errqueue.h> 18 #include <linux/udp.h> 19 #include <linux/in.h> 20 #include <linux/in6.h> 21 #include <linux/icmp.h> 22 #include <linux/gfp.h> 23 #include <linux/circ_buf.h> 24 #include <net/sock.h> 25 #include <net/af_rxrpc.h> 26 #include <net/ip.h> 27 #include "ar-internal.h" 28 29 /* 30 * Preallocate a single service call, connection and peer and, if possible, 31 * give them a user ID and attach the user's side of the ID to them. 32 */ 33 static int rxrpc_service_prealloc_one(struct rxrpc_sock *rx, 34 struct rxrpc_backlog *b, 35 rxrpc_notify_rx_t notify_rx, 36 rxrpc_user_attach_call_t user_attach_call, 37 unsigned long user_call_ID, gfp_t gfp) 38 { 39 const void *here = __builtin_return_address(0); 40 struct rxrpc_call *call; 41 int max, tmp; 42 unsigned int size = RXRPC_BACKLOG_MAX; 43 unsigned int head, tail, call_head, call_tail; 44 45 max = rx->sk.sk_max_ack_backlog; 46 tmp = rx->sk.sk_ack_backlog; 47 if (tmp >= max) { 48 _leave(" = -ENOBUFS [full %u]", max); 49 return -ENOBUFS; 50 } 51 max -= tmp; 52 53 /* We don't need more conns and peers than we have calls, but on the 54 * other hand, we shouldn't ever use more peers than conns or conns 55 * than calls. 56 */ 57 call_head = b->call_backlog_head; 58 call_tail = READ_ONCE(b->call_backlog_tail); 59 tmp = CIRC_CNT(call_head, call_tail, size); 60 if (tmp >= max) { 61 _leave(" = -ENOBUFS [enough %u]", tmp); 62 return -ENOBUFS; 63 } 64 max = tmp + 1; 65 66 head = b->peer_backlog_head; 67 tail = READ_ONCE(b->peer_backlog_tail); 68 if (CIRC_CNT(head, tail, size) < max) { 69 struct rxrpc_peer *peer = rxrpc_alloc_peer(rx->local, gfp); 70 if (!peer) 71 return -ENOMEM; 72 b->peer_backlog[head] = peer; 73 smp_store_release(&b->peer_backlog_head, 74 (head + 1) & (size - 1)); 75 } 76 77 head = b->conn_backlog_head; 78 tail = READ_ONCE(b->conn_backlog_tail); 79 if (CIRC_CNT(head, tail, size) < max) { 80 struct rxrpc_connection *conn; 81 82 conn = rxrpc_prealloc_service_connection(gfp); 83 if (!conn) 84 return -ENOMEM; 85 b->conn_backlog[head] = conn; 86 smp_store_release(&b->conn_backlog_head, 87 (head + 1) & (size - 1)); 88 89 trace_rxrpc_conn(conn, rxrpc_conn_new_service, 90 atomic_read(&conn->usage), here); 91 } 92 93 /* Now it gets complicated, because calls get registered with the 94 * socket here, particularly if a user ID is preassigned by the user. 95 */ 96 call = rxrpc_alloc_call(gfp); 97 if (!call) 98 return -ENOMEM; 99 call->flags |= (1 << RXRPC_CALL_IS_SERVICE); 100 call->state = RXRPC_CALL_SERVER_PREALLOC; 101 102 trace_rxrpc_call(call, rxrpc_call_new_service, 103 atomic_read(&call->usage), 104 here, (const void *)user_call_ID); 105 106 write_lock(&rx->call_lock); 107 if (user_attach_call) { 108 struct rxrpc_call *xcall; 109 struct rb_node *parent, **pp; 110 111 /* Check the user ID isn't already in use */ 112 pp = &rx->calls.rb_node; 113 parent = NULL; 114 while (*pp) { 115 parent = *pp; 116 xcall = rb_entry(parent, struct rxrpc_call, sock_node); 117 if (user_call_ID < call->user_call_ID) 118 pp = &(*pp)->rb_left; 119 else if (user_call_ID > call->user_call_ID) 120 pp = &(*pp)->rb_right; 121 else 122 goto id_in_use; 123 } 124 125 call->user_call_ID = user_call_ID; 126 call->notify_rx = notify_rx; 127 rxrpc_get_call(call, rxrpc_call_got_kernel); 128 user_attach_call(call, user_call_ID); 129 rxrpc_get_call(call, rxrpc_call_got_userid); 130 rb_link_node(&call->sock_node, parent, pp); 131 rb_insert_color(&call->sock_node, &rx->calls); 132 set_bit(RXRPC_CALL_HAS_USERID, &call->flags); 133 } 134 135 list_add(&call->sock_link, &rx->sock_calls); 136 137 write_unlock(&rx->call_lock); 138 139 write_lock(&rxrpc_call_lock); 140 list_add_tail(&call->link, &rxrpc_calls); 141 write_unlock(&rxrpc_call_lock); 142 143 b->call_backlog[call_head] = call; 144 smp_store_release(&b->call_backlog_head, (call_head + 1) & (size - 1)); 145 _leave(" = 0 [%d -> %lx]", call->debug_id, user_call_ID); 146 return 0; 147 148 id_in_use: 149 write_unlock(&rx->call_lock); 150 rxrpc_cleanup_call(call); 151 _leave(" = -EBADSLT"); 152 return -EBADSLT; 153 } 154 155 /* 156 * Preallocate sufficient service connections, calls and peers to cover the 157 * entire backlog of a socket. When a new call comes in, if we don't have 158 * sufficient of each available, the call gets rejected as busy or ignored. 159 * 160 * The backlog is replenished when a connection is accepted or rejected. 161 */ 162 int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp) 163 { 164 struct rxrpc_backlog *b = rx->backlog; 165 166 if (!b) { 167 b = kzalloc(sizeof(struct rxrpc_backlog), gfp); 168 if (!b) 169 return -ENOMEM; 170 rx->backlog = b; 171 } 172 173 if (rx->discard_new_call) 174 return 0; 175 176 while (rxrpc_service_prealloc_one(rx, b, NULL, NULL, 0, gfp) == 0) 177 ; 178 179 return 0; 180 } 181 182 /* 183 * Discard the preallocation on a service. 184 */ 185 void rxrpc_discard_prealloc(struct rxrpc_sock *rx) 186 { 187 struct rxrpc_backlog *b = rx->backlog; 188 unsigned int size = RXRPC_BACKLOG_MAX, head, tail; 189 190 if (!b) 191 return; 192 rx->backlog = NULL; 193 194 /* Make sure that there aren't any incoming calls in progress before we 195 * clear the preallocation buffers. 196 */ 197 spin_lock_bh(&rx->incoming_lock); 198 spin_unlock_bh(&rx->incoming_lock); 199 200 head = b->peer_backlog_head; 201 tail = b->peer_backlog_tail; 202 while (CIRC_CNT(head, tail, size) > 0) { 203 struct rxrpc_peer *peer = b->peer_backlog[tail]; 204 kfree(peer); 205 tail = (tail + 1) & (size - 1); 206 } 207 208 head = b->conn_backlog_head; 209 tail = b->conn_backlog_tail; 210 while (CIRC_CNT(head, tail, size) > 0) { 211 struct rxrpc_connection *conn = b->conn_backlog[tail]; 212 write_lock(&rxrpc_connection_lock); 213 list_del(&conn->link); 214 list_del(&conn->proc_link); 215 write_unlock(&rxrpc_connection_lock); 216 kfree(conn); 217 tail = (tail + 1) & (size - 1); 218 } 219 220 head = b->call_backlog_head; 221 tail = b->call_backlog_tail; 222 while (CIRC_CNT(head, tail, size) > 0) { 223 struct rxrpc_call *call = b->call_backlog[tail]; 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 rxrpc_put_call(call, rxrpc_call_put_kernel); 228 } 229 rxrpc_call_completed(call); 230 rxrpc_release_call(rx, call); 231 rxrpc_put_call(call, rxrpc_call_put); 232 tail = (tail + 1) & (size - 1); 233 } 234 235 kfree(b); 236 } 237 238 /* 239 * Allocate a new incoming call from the prealloc pool, along with a connection 240 * and a peer as necessary. 241 */ 242 static struct rxrpc_call *rxrpc_alloc_incoming_call(struct rxrpc_sock *rx, 243 struct rxrpc_local *local, 244 struct rxrpc_connection *conn, 245 struct sk_buff *skb) 246 { 247 struct rxrpc_backlog *b = rx->backlog; 248 struct rxrpc_peer *peer, *xpeer; 249 struct rxrpc_call *call; 250 unsigned short call_head, conn_head, peer_head; 251 unsigned short call_tail, conn_tail, peer_tail; 252 unsigned short call_count, conn_count; 253 254 /* #calls >= #conns >= #peers must hold true. */ 255 call_head = smp_load_acquire(&b->call_backlog_head); 256 call_tail = b->call_backlog_tail; 257 call_count = CIRC_CNT(call_head, call_tail, RXRPC_BACKLOG_MAX); 258 conn_head = smp_load_acquire(&b->conn_backlog_head); 259 conn_tail = b->conn_backlog_tail; 260 conn_count = CIRC_CNT(conn_head, conn_tail, RXRPC_BACKLOG_MAX); 261 ASSERTCMP(conn_count, >=, call_count); 262 peer_head = smp_load_acquire(&b->peer_backlog_head); 263 peer_tail = b->peer_backlog_tail; 264 ASSERTCMP(CIRC_CNT(peer_head, peer_tail, RXRPC_BACKLOG_MAX), >=, 265 conn_count); 266 267 if (call_count == 0) 268 return NULL; 269 270 if (!conn) { 271 /* No connection. We're going to need a peer to start off 272 * with. If one doesn't yet exist, use a spare from the 273 * preallocation set. We dump the address into the spare in 274 * anticipation - and to save on stack space. 275 */ 276 xpeer = b->peer_backlog[peer_tail]; 277 if (rxrpc_extract_addr_from_skb(&xpeer->srx, skb) < 0) 278 return NULL; 279 280 peer = rxrpc_lookup_incoming_peer(local, xpeer); 281 if (peer == xpeer) { 282 b->peer_backlog[peer_tail] = NULL; 283 smp_store_release(&b->peer_backlog_tail, 284 (peer_tail + 1) & 285 (RXRPC_BACKLOG_MAX - 1)); 286 } 287 288 /* Now allocate and set up the connection */ 289 conn = b->conn_backlog[conn_tail]; 290 b->conn_backlog[conn_tail] = NULL; 291 smp_store_release(&b->conn_backlog_tail, 292 (conn_tail + 1) & (RXRPC_BACKLOG_MAX - 1)); 293 rxrpc_get_local(local); 294 conn->params.local = local; 295 conn->params.peer = peer; 296 rxrpc_see_connection(conn); 297 rxrpc_new_incoming_connection(conn, skb); 298 } else { 299 rxrpc_get_connection(conn); 300 } 301 302 /* And now we can allocate and set up a new call */ 303 call = b->call_backlog[call_tail]; 304 b->call_backlog[call_tail] = NULL; 305 smp_store_release(&b->call_backlog_tail, 306 (call_tail + 1) & (RXRPC_BACKLOG_MAX - 1)); 307 308 rxrpc_see_call(call); 309 call->conn = conn; 310 call->peer = rxrpc_get_peer(conn->params.peer); 311 return call; 312 } 313 314 /* 315 * Set up a new incoming call. Called in BH context with the RCU read lock 316 * held. 317 * 318 * If this is for a kernel service, when we allocate the call, it will have 319 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the 320 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace 321 * services only have the ref from the backlog buffer. We want to pass this 322 * ref to non-BH context to dispose of. 323 * 324 * If we want to report an error, we mark the skb with the packet type and 325 * abort code and return NULL. 326 * 327 * The call is returned with the user access mutex held. 328 */ 329 struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local, 330 struct rxrpc_connection *conn, 331 struct sk_buff *skb) 332 { 333 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 334 struct rxrpc_sock *rx; 335 struct rxrpc_call *call; 336 u16 service_id = sp->hdr.serviceId; 337 338 _enter(""); 339 340 /* Get the socket providing the service */ 341 rx = rcu_dereference(local->service); 342 if (rx && service_id == rx->srx.srx_service) 343 goto found_service; 344 345 trace_rxrpc_abort("INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 346 RX_INVALID_OPERATION, EOPNOTSUPP); 347 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT; 348 skb->priority = RX_INVALID_OPERATION; 349 _leave(" = NULL [service]"); 350 return NULL; 351 352 found_service: 353 spin_lock(&rx->incoming_lock); 354 if (rx->sk.sk_state == RXRPC_SERVER_LISTEN_DISABLED || 355 rx->sk.sk_state == RXRPC_CLOSE) { 356 trace_rxrpc_abort("CLS", sp->hdr.cid, sp->hdr.callNumber, 357 sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN); 358 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT; 359 skb->priority = RX_INVALID_OPERATION; 360 _leave(" = NULL [close]"); 361 call = NULL; 362 goto out; 363 } 364 365 call = rxrpc_alloc_incoming_call(rx, local, conn, skb); 366 if (!call) { 367 skb->mark = RXRPC_SKB_MARK_BUSY; 368 _leave(" = NULL [busy]"); 369 call = NULL; 370 goto out; 371 } 372 373 trace_rxrpc_receive(call, rxrpc_receive_incoming, 374 sp->hdr.serial, sp->hdr.seq); 375 376 /* Lock the call to prevent rxrpc_kernel_send/recv_data() and 377 * sendmsg()/recvmsg() inconveniently stealing the mutex once the 378 * notification is generated. 379 * 380 * The BUG should never happen because the kernel should be well 381 * behaved enough not to access the call before the first notification 382 * event and userspace is prevented from doing so until the state is 383 * appropriate. 384 */ 385 if (!mutex_trylock(&call->user_mutex)) 386 BUG(); 387 388 /* Make the call live. */ 389 rxrpc_incoming_call(rx, call, skb); 390 conn = call->conn; 391 392 if (rx->notify_new_call) 393 rx->notify_new_call(&rx->sk, call, call->user_call_ID); 394 else 395 sk_acceptq_added(&rx->sk); 396 397 spin_lock(&conn->state_lock); 398 switch (conn->state) { 399 case RXRPC_CONN_SERVICE_UNSECURED: 400 conn->state = RXRPC_CONN_SERVICE_CHALLENGING; 401 set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events); 402 rxrpc_queue_conn(call->conn); 403 break; 404 405 case RXRPC_CONN_SERVICE: 406 write_lock(&call->state_lock); 407 if (rx->discard_new_call) 408 call->state = RXRPC_CALL_SERVER_RECV_REQUEST; 409 else 410 call->state = RXRPC_CALL_SERVER_ACCEPTING; 411 write_unlock(&call->state_lock); 412 break; 413 414 case RXRPC_CONN_REMOTELY_ABORTED: 415 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED, 416 conn->remote_abort, ECONNABORTED); 417 break; 418 case RXRPC_CONN_LOCALLY_ABORTED: 419 rxrpc_abort_call("CON", call, sp->hdr.seq, 420 conn->local_abort, ECONNABORTED); 421 break; 422 default: 423 BUG(); 424 } 425 spin_unlock(&conn->state_lock); 426 427 if (call->state == RXRPC_CALL_SERVER_ACCEPTING) 428 rxrpc_notify_socket(call); 429 430 /* We have to discard the prealloc queue's ref here and rely on a 431 * combination of the RCU read lock and refs held either by the socket 432 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel 433 * service to prevent the call from being deallocated too early. 434 */ 435 rxrpc_put_call(call, rxrpc_call_put); 436 437 _leave(" = %p{%d}", call, call->debug_id); 438 out: 439 spin_unlock(&rx->incoming_lock); 440 return call; 441 } 442 443 /* 444 * handle acceptance of a call by userspace 445 * - assign the user call ID to the call at the front of the queue 446 * - called with the socket locked. 447 */ 448 struct rxrpc_call *rxrpc_accept_call(struct rxrpc_sock *rx, 449 unsigned long user_call_ID, 450 rxrpc_notify_rx_t notify_rx) 451 __releases(&rx->sk.sk_lock.slock) 452 { 453 struct rxrpc_call *call; 454 struct rb_node *parent, **pp; 455 int ret; 456 457 _enter(",%lx", user_call_ID); 458 459 ASSERT(!irqs_disabled()); 460 461 write_lock(&rx->call_lock); 462 463 if (list_empty(&rx->to_be_accepted)) { 464 write_unlock(&rx->call_lock); 465 release_sock(&rx->sk); 466 kleave(" = -ENODATA [empty]"); 467 return ERR_PTR(-ENODATA); 468 } 469 470 /* check the user ID isn't already in use */ 471 pp = &rx->calls.rb_node; 472 parent = NULL; 473 while (*pp) { 474 parent = *pp; 475 call = rb_entry(parent, struct rxrpc_call, sock_node); 476 477 if (user_call_ID < call->user_call_ID) 478 pp = &(*pp)->rb_left; 479 else if (user_call_ID > call->user_call_ID) 480 pp = &(*pp)->rb_right; 481 else 482 goto id_in_use; 483 } 484 485 /* Dequeue the first call and check it's still valid. We gain 486 * responsibility for the queue's reference. 487 */ 488 call = list_entry(rx->to_be_accepted.next, 489 struct rxrpc_call, accept_link); 490 write_unlock(&rx->call_lock); 491 492 /* We need to gain the mutex from the interrupt handler without 493 * upsetting lockdep, so we have to release it there and take it here. 494 * We are, however, still holding the socket lock, so other accepts 495 * must wait for us and no one can add the user ID behind our backs. 496 */ 497 if (mutex_lock_interruptible(&call->user_mutex) < 0) { 498 release_sock(&rx->sk); 499 kleave(" = -ERESTARTSYS"); 500 return ERR_PTR(-ERESTARTSYS); 501 } 502 503 write_lock(&rx->call_lock); 504 list_del_init(&call->accept_link); 505 sk_acceptq_removed(&rx->sk); 506 rxrpc_see_call(call); 507 508 /* Find the user ID insertion point. */ 509 pp = &rx->calls.rb_node; 510 parent = NULL; 511 while (*pp) { 512 parent = *pp; 513 call = rb_entry(parent, struct rxrpc_call, sock_node); 514 515 if (user_call_ID < call->user_call_ID) 516 pp = &(*pp)->rb_left; 517 else if (user_call_ID > call->user_call_ID) 518 pp = &(*pp)->rb_right; 519 else 520 BUG(); 521 } 522 523 write_lock_bh(&call->state_lock); 524 switch (call->state) { 525 case RXRPC_CALL_SERVER_ACCEPTING: 526 call->state = RXRPC_CALL_SERVER_RECV_REQUEST; 527 break; 528 case RXRPC_CALL_COMPLETE: 529 ret = call->error; 530 goto out_release; 531 default: 532 BUG(); 533 } 534 535 /* formalise the acceptance */ 536 call->notify_rx = notify_rx; 537 call->user_call_ID = user_call_ID; 538 rxrpc_get_call(call, rxrpc_call_got_userid); 539 rb_link_node(&call->sock_node, parent, pp); 540 rb_insert_color(&call->sock_node, &rx->calls); 541 if (test_and_set_bit(RXRPC_CALL_HAS_USERID, &call->flags)) 542 BUG(); 543 544 write_unlock_bh(&call->state_lock); 545 write_unlock(&rx->call_lock); 546 rxrpc_notify_socket(call); 547 rxrpc_service_prealloc(rx, GFP_KERNEL); 548 release_sock(&rx->sk); 549 _leave(" = %p{%d}", call, call->debug_id); 550 return call; 551 552 out_release: 553 _debug("release %p", call); 554 write_unlock_bh(&call->state_lock); 555 write_unlock(&rx->call_lock); 556 rxrpc_release_call(rx, call); 557 rxrpc_put_call(call, rxrpc_call_put); 558 goto out; 559 560 id_in_use: 561 ret = -EBADSLT; 562 write_unlock(&rx->call_lock); 563 out: 564 rxrpc_service_prealloc(rx, GFP_KERNEL); 565 release_sock(&rx->sk); 566 _leave(" = %d", ret); 567 return ERR_PTR(ret); 568 } 569 570 /* 571 * Handle rejection of a call by userspace 572 * - reject the call at the front of the queue 573 */ 574 int rxrpc_reject_call(struct rxrpc_sock *rx) 575 { 576 struct rxrpc_call *call; 577 bool abort = false; 578 int ret; 579 580 _enter(""); 581 582 ASSERT(!irqs_disabled()); 583 584 write_lock(&rx->call_lock); 585 586 if (list_empty(&rx->to_be_accepted)) { 587 write_unlock(&rx->call_lock); 588 return -ENODATA; 589 } 590 591 /* Dequeue the first call and check it's still valid. We gain 592 * responsibility for the queue's reference. 593 */ 594 call = list_entry(rx->to_be_accepted.next, 595 struct rxrpc_call, accept_link); 596 list_del_init(&call->accept_link); 597 sk_acceptq_removed(&rx->sk); 598 rxrpc_see_call(call); 599 600 write_lock_bh(&call->state_lock); 601 switch (call->state) { 602 case RXRPC_CALL_SERVER_ACCEPTING: 603 __rxrpc_abort_call("REJ", call, 1, RX_USER_ABORT, ECONNABORTED); 604 abort = true; 605 /* fall through */ 606 case RXRPC_CALL_COMPLETE: 607 ret = call->error; 608 goto out_discard; 609 default: 610 BUG(); 611 } 612 613 out_discard: 614 write_unlock_bh(&call->state_lock); 615 write_unlock(&rx->call_lock); 616 if (abort) { 617 rxrpc_send_abort_packet(call); 618 rxrpc_release_call(rx, call); 619 rxrpc_put_call(call, rxrpc_call_put); 620 } 621 rxrpc_service_prealloc(rx, GFP_KERNEL); 622 _leave(" = %d", ret); 623 return ret; 624 } 625 626 /* 627 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls 628 * @sock: The socket on which to preallocate 629 * @notify_rx: Event notification function for the call 630 * @user_attach_call: Func to attach call to user_call_ID 631 * @user_call_ID: The tag to attach to the preallocated call 632 * @gfp: The allocation conditions. 633 * 634 * Charge up the socket with preallocated calls, each with a user ID. A 635 * function should be provided to effect the attachment from the user's side. 636 * The user is given a ref to hold on the call. 637 * 638 * Note that the call may be come connected before this function returns. 639 */ 640 int rxrpc_kernel_charge_accept(struct socket *sock, 641 rxrpc_notify_rx_t notify_rx, 642 rxrpc_user_attach_call_t user_attach_call, 643 unsigned long user_call_ID, gfp_t gfp) 644 { 645 struct rxrpc_sock *rx = rxrpc_sk(sock->sk); 646 struct rxrpc_backlog *b = rx->backlog; 647 648 if (sock->sk->sk_state == RXRPC_CLOSE) 649 return -ESHUTDOWN; 650 651 return rxrpc_service_prealloc_one(rx, b, notify_rx, 652 user_attach_call, user_call_ID, 653 gfp); 654 } 655 EXPORT_SYMBOL(rxrpc_kernel_charge_accept); 656