net/rxrpc/recvmsg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* RxRPC recvmsg() implementation
   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/net.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/export.h>
  13 #include <linux/sched/signal.h>
  14
  15 #include <net/sock.h>
  16 #include <net/af_rxrpc.h>
  17 #include "ar-internal.h"
  18
  19 /*
  20  * Post a call for attention by the socket or kernel service.  Further
  21  * notifications are suppressed by putting recvmsg_link on a dummy queue.
  22  */
  23 void rxrpc_notify_socket(struct rxrpc_call *call)
  24 {
  25         struct rxrpc_sock *rx;
  26         struct sock *sk;
  27
  28         _enter("%d", call->debug_id);
  29
  30         if (!list_empty(&call->recvmsg_link))
  31                 return;
  32
  33         rcu_read_lock();
  34
  35         rx = rcu_dereference(call->socket);
  36         sk = &rx->sk;
  37         if (rx && sk->sk_state < RXRPC_CLOSE) {
  38                 if (call->notify_rx) {
  39                         spin_lock_bh(&call->notify_lock);
  40                         call->notify_rx(sk, call, call->user_call_ID);
  41                         spin_unlock_bh(&call->notify_lock);
  42                 } else {
  43                         write_lock_bh(&rx->recvmsg_lock);
  44                         if (list_empty(&call->recvmsg_link)) {
  45                                 rxrpc_get_call(call, rxrpc_call_got);
  46                                 list_add_tail(&call->recvmsg_link, &rx->recvmsg_q);
  47                         }
  48                         write_unlock_bh(&rx->recvmsg_lock);
  49
  50                         if (!sock_flag(sk, SOCK_DEAD)) {
  51                                 _debug("call %ps", sk->sk_data_ready);
  52                                 sk->sk_data_ready(sk);
  53                         }
  54                 }
  55         }
  56
  57         rcu_read_unlock();
  58         _leave("");
  59 }
  60
  61 /*
  62  * Transition a call to the complete state.
  63  */
  64 bool __rxrpc_set_call_completion(struct rxrpc_call *call,
  65                                  enum rxrpc_call_completion compl,
  66                                  u32 abort_code,
  67                                  int error)
  68 {
  69         if (call->state < RXRPC_CALL_COMPLETE) {
  70                 call->abort_code = abort_code;
  71                 call->error = error;
  72                 call->completion = compl;
  73                 call->state = RXRPC_CALL_COMPLETE;
  74                 trace_rxrpc_call_complete(call);
  75                 wake_up(&call->waitq);
  76                 rxrpc_notify_socket(call);
  77                 return true;
  78         }
  79         return false;
  80 }
  81
  82 bool rxrpc_set_call_completion(struct rxrpc_call *call,
  83                                enum rxrpc_call_completion compl,
  84                                u32 abort_code,
  85                                int error)
  86 {
  87         bool ret = false;
  88
  89         if (call->state < RXRPC_CALL_COMPLETE) {
  90                 write_lock_bh(&call->state_lock);
  91                 ret = __rxrpc_set_call_completion(call, compl, abort_code, error);
  92                 write_unlock_bh(&call->state_lock);
  93         }
  94         return ret;
  95 }
  96
  97 /*
  98  * Record that a call successfully completed.
  99  */
 100 bool __rxrpc_call_completed(struct rxrpc_call *call)
 101 {
 102         return __rxrpc_set_call_completion(call, RXRPC_CALL_SUCCEEDED, 0, 0);
 103 }
 104
 105 bool rxrpc_call_completed(struct rxrpc_call *call)
 106 {
 107         bool ret = false;
 108
 109         if (call->state < RXRPC_CALL_COMPLETE) {
 110                 write_lock_bh(&call->state_lock);
 111                 ret = __rxrpc_call_completed(call);
 112                 write_unlock_bh(&call->state_lock);
 113         }
 114         return ret;
 115 }
 116
 117 /*
 118  * Record that a call is locally aborted.
 119  */
 120 bool __rxrpc_abort_call(const char *why, struct rxrpc_call *call,
 121                         rxrpc_seq_t seq, u32 abort_code, int error)
 122 {
 123         trace_rxrpc_abort(call->debug_id, why, call->cid, call->call_id, seq,
 124                           abort_code, error);
 125         return __rxrpc_set_call_completion(call, RXRPC_CALL_LOCALLY_ABORTED,
 126                                            abort_code, error);
 127 }
 128
 129 bool rxrpc_abort_call(const char *why, struct rxrpc_call *call,
 130                       rxrpc_seq_t seq, u32 abort_code, int error)
 131 {
 132         bool ret;
 133
 134         write_lock_bh(&call->state_lock);
 135         ret = __rxrpc_abort_call(why, call, seq, abort_code, error);
 136         write_unlock_bh(&call->state_lock);
 137         return ret;
 138 }
 139
 140 /*
 141  * Pass a call terminating message to userspace.
 142  */
 143 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
 144 {
 145         u32 tmp = 0;
 146         int ret;
 147
 148         switch (call->completion) {
 149         case RXRPC_CALL_SUCCEEDED:
 150                 ret = 0;
 151                 if (rxrpc_is_service_call(call))
 152                         ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
 153                 break;
 154         case RXRPC_CALL_REMOTELY_ABORTED:
 155                 tmp = call->abort_code;
 156                 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
 157                 break;
 158         case RXRPC_CALL_LOCALLY_ABORTED:
 159                 tmp = call->abort_code;
 160                 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
 161                 break;
 162         case RXRPC_CALL_NETWORK_ERROR:
 163                 tmp = -call->error;
 164                 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
 165                 break;
 166         case RXRPC_CALL_LOCAL_ERROR:
 167                 tmp = -call->error;
 168                 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
 169                 break;
 170         default:
 171                 pr_err("Invalid terminal call state %u\n", call->state);
 172                 BUG();
 173                 break;
 174         }
 175
 176         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_terminal, call->rx_hard_ack,
 177                             call->rx_pkt_offset, call->rx_pkt_len, ret);
 178         return ret;
 179 }
 180
 181 /*
 182  * End the packet reception phase.
 183  */
 184 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
 185 {
 186         _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
 187
 188         trace_rxrpc_receive(call, rxrpc_receive_end, 0, call->rx_top);
 189         ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
 190
 191         if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
 192                 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, serial, false, true,
 193                                   rxrpc_propose_ack_terminal_ack);
 194                 //rxrpc_send_ack_packet(call, false, NULL);
 195         }
 196
 197         write_lock_bh(&call->state_lock);
 198
 199         switch (call->state) {
 200         case RXRPC_CALL_CLIENT_RECV_REPLY:
 201                 __rxrpc_call_completed(call);
 202                 write_unlock_bh(&call->state_lock);
 203                 break;
 204
 205         case RXRPC_CALL_SERVER_RECV_REQUEST:
 206                 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
 207                 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
 208                 write_unlock_bh(&call->state_lock);
 209                 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial, false, true,
 210                                   rxrpc_propose_ack_processing_op);
 211                 break;
 212         default:
 213                 write_unlock_bh(&call->state_lock);
 214                 break;
 215         }
 216 }
 217
 218 /*
 219  * Discard a packet we've used up and advance the Rx window by one.
 220  */
 221 static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
 222 {
 223         struct rxrpc_skb_priv *sp;
 224         struct sk_buff *skb;
 225         rxrpc_serial_t serial;
 226         rxrpc_seq_t hard_ack, top;
 227         bool last = false;
 228         u8 subpacket;
 229         int ix;
 230
 231         _enter("%d", call->debug_id);
 232
 233         hard_ack = call->rx_hard_ack;
 234         top = smp_load_acquire(&call->rx_top);
 235         ASSERT(before(hard_ack, top));
 236
 237         hard_ack++;
 238         ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
 239         skb = call->rxtx_buffer[ix];
 240         rxrpc_see_skb(skb, rxrpc_skb_rotated);
 241         sp = rxrpc_skb(skb);
 242
 243         subpacket = call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
 244         serial = sp->hdr.serial + subpacket;
 245
 246         if (subpacket == sp->nr_subpackets - 1 &&
 247             sp->rx_flags & RXRPC_SKB_INCL_LAST)
 248                 last = true;
 249
 250         call->rxtx_buffer[ix] = NULL;
 251         call->rxtx_annotations[ix] = 0;
 252         /* Barrier against rxrpc_input_data(). */
 253         smp_store_release(&call->rx_hard_ack, hard_ack);
 254
 255         rxrpc_free_skb(skb, rxrpc_skb_freed);
 256
 257         trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
 258         if (last) {
 259                 rxrpc_end_rx_phase(call, serial);
 260         } else {
 261                 /* Check to see if there's an ACK that needs sending. */
 262                 if (atomic_inc_return(&call->ackr_nr_consumed) > 2)
 263                         rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, serial,
 264                                           true, false,
 265                                           rxrpc_propose_ack_rotate_rx);
 266                 if (call->ackr_reason && call->ackr_reason != RXRPC_ACK_DELAY)
 267                         rxrpc_send_ack_packet(call, false, NULL);
 268         }
 269 }
 270
 271 /*
 272  * Decrypt and verify a (sub)packet.  The packet's length may be changed due to
 273  * padding, but if this is the case, the packet length will be resident in the
 274  * socket buffer.  Note that we can't modify the master skb info as the skb may
 275  * be the home to multiple subpackets.
 276  */
 277 static int rxrpc_verify_packet(struct rxrpc_call *call, struct sk_buff *skb,
 278                                u8 annotation,
 279                                unsigned int offset, unsigned int len)
 280 {
 281         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
 282         rxrpc_seq_t seq = sp->hdr.seq;
 283         u16 cksum = sp->hdr.cksum;
 284         u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
 285
 286         _enter("");
 287
 288         /* For all but the head jumbo subpacket, the security checksum is in a
 289          * jumbo header immediately prior to the data.
 290          */
 291         if (subpacket > 0) {
 292                 __be16 tmp;
 293                 if (skb_copy_bits(skb, offset - 2, &tmp, 2) < 0)
 294                         BUG();
 295                 cksum = ntohs(tmp);
 296                 seq += subpacket;
 297         }
 298
 299         return call->security->verify_packet(call, skb, offset, len,
 300                                              seq, cksum);
 301 }
 302
 303 /*
 304  * Locate the data within a packet.  This is complicated by:
 305  *
 306  * (1) An skb may contain a jumbo packet - so we have to find the appropriate
 307  *     subpacket.
 308  *
 309  * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
 310  *     contains an extra header which includes the true length of the data,
 311  *     excluding any encrypted padding.
 312  */
 313 static int rxrpc_locate_data(struct rxrpc_call *call, struct sk_buff *skb,
 314                              u8 *_annotation,
 315                              unsigned int *_offset, unsigned int *_len,
 316                              bool *_last)
 317 {
 318         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
 319         unsigned int offset = sizeof(struct rxrpc_wire_header);
 320         unsigned int len;
 321         bool last = false;
 322         int ret;
 323         u8 annotation = *_annotation;
 324         u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
 325
 326         /* Locate the subpacket */
 327         offset += subpacket * RXRPC_JUMBO_SUBPKTLEN;
 328         len = skb->len - offset;
 329         if (subpacket < sp->nr_subpackets - 1)
 330                 len = RXRPC_JUMBO_DATALEN;
 331         else if (sp->rx_flags & RXRPC_SKB_INCL_LAST)
 332                 last = true;
 333
 334         if (!(annotation & RXRPC_RX_ANNO_VERIFIED)) {
 335                 ret = rxrpc_verify_packet(call, skb, annotation, offset, len);
 336                 if (ret < 0)
 337                         return ret;
 338                 *_annotation |= RXRPC_RX_ANNO_VERIFIED;
 339         }
 340
 341         *_offset = offset;
 342         *_len = len;
 343         *_last = last;
 344         call->security->locate_data(call, skb, _offset, _len);
 345         return 0;
 346 }
 347
 348 /*
 349  * Deliver messages to a call.  This keeps processing packets until the buffer
 350  * is filled and we find either more DATA (returns 0) or the end of the DATA
 351  * (returns 1).  If more packets are required, it returns -EAGAIN.
 352  */
 353 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
 354                               struct msghdr *msg, struct iov_iter *iter,
 355                               size_t len, int flags, size_t *_offset)
 356 {
 357         struct rxrpc_skb_priv *sp;
 358         struct sk_buff *skb;
 359         rxrpc_serial_t serial;
 360         rxrpc_seq_t hard_ack, top, seq;
 361         size_t remain;
 362         bool rx_pkt_last;
 363         unsigned int rx_pkt_offset, rx_pkt_len;
 364         int ix, copy, ret = -EAGAIN, ret2;
 365
 366         if (test_and_clear_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags) &&
 367             call->ackr_reason)
 368                 rxrpc_send_ack_packet(call, false, NULL);
 369
 370         rx_pkt_offset = call->rx_pkt_offset;
 371         rx_pkt_len = call->rx_pkt_len;
 372         rx_pkt_last = call->rx_pkt_last;
 373
 374         if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
 375                 seq = call->rx_hard_ack;
 376                 ret = 1;
 377                 goto done;
 378         }
 379
 380         /* Barriers against rxrpc_input_data(). */
 381         hard_ack = call->rx_hard_ack;
 382         seq = hard_ack + 1;
 383
 384         while (top = smp_load_acquire(&call->rx_top),
 385                before_eq(seq, top)
 386                ) {
 387                 ix = seq & RXRPC_RXTX_BUFF_MASK;
 388                 skb = call->rxtx_buffer[ix];
 389                 if (!skb) {
 390                         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_hole, seq,
 391                                             rx_pkt_offset, rx_pkt_len, 0);
 392                         break;
 393                 }
 394                 smp_rmb();
 395                 rxrpc_see_skb(skb, rxrpc_skb_seen);
 396                 sp = rxrpc_skb(skb);
 397
 398                 if (!(flags & MSG_PEEK)) {
 399                         serial = sp->hdr.serial;
 400                         serial += call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
 401                         trace_rxrpc_receive(call, rxrpc_receive_front,
 402                                             serial, seq);
 403                 }
 404
 405                 if (msg)
 406                         sock_recv_timestamp(msg, sock->sk, skb);
 407
 408                 if (rx_pkt_offset == 0) {
 409                         ret2 = rxrpc_locate_data(call, skb,
 410                                                  &call->rxtx_annotations[ix],
 411                                                  &rx_pkt_offset, &rx_pkt_len,
 412                                                  &rx_pkt_last);
 413                         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_next, seq,
 414                                             rx_pkt_offset, rx_pkt_len, ret2);
 415                         if (ret2 < 0) {
 416                                 ret = ret2;
 417                                 goto out;
 418                         }
 419                 } else {
 420                         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_cont, seq,
 421                                             rx_pkt_offset, rx_pkt_len, 0);
 422                 }
 423
 424                 /* We have to handle short, empty and used-up DATA packets. */
 425                 remain = len - *_offset;
 426                 copy = rx_pkt_len;
 427                 if (copy > remain)
 428                         copy = remain;
 429                 if (copy > 0) {
 430                         ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
 431                                                       copy);
 432                         if (ret2 < 0) {
 433                                 ret = ret2;
 434                                 goto out;
 435                         }
 436
 437                         /* handle piecemeal consumption of data packets */
 438                         rx_pkt_offset += copy;
 439                         rx_pkt_len -= copy;
 440                         *_offset += copy;
 441                 }
 442
 443                 if (rx_pkt_len > 0) {
 444                         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_full, seq,
 445                                             rx_pkt_offset, rx_pkt_len, 0);
 446                         ASSERTCMP(*_offset, ==, len);
 447                         ret = 0;
 448                         break;
 449                 }
 450
 451                 /* The whole packet has been transferred. */
 452                 if (!(flags & MSG_PEEK))
 453                         rxrpc_rotate_rx_window(call);
 454                 rx_pkt_offset = 0;
 455                 rx_pkt_len = 0;
 456
 457                 if (rx_pkt_last) {
 458                         ASSERTCMP(seq, ==, READ_ONCE(call->rx_top));
 459                         ret = 1;
 460                         goto out;
 461                 }
 462
 463                 seq++;
 464         }
 465
 466 out:
 467         if (!(flags & MSG_PEEK)) {
 468                 call->rx_pkt_offset = rx_pkt_offset;
 469                 call->rx_pkt_len = rx_pkt_len;
 470                 call->rx_pkt_last = rx_pkt_last;
 471         }
 472 done:
 473         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_data_return, seq,
 474                             rx_pkt_offset, rx_pkt_len, ret);
 475         if (ret == -EAGAIN)
 476                 set_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags);
 477         return ret;
 478 }
 479
 480 /*
 481  * Receive a message from an RxRPC socket
 482  * - we need to be careful about two or more threads calling recvmsg
 483  *   simultaneously
 484  */
 485 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 486                   int flags)
 487 {
 488         struct rxrpc_call *call;
 489         struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
 490         struct list_head *l;
 491         size_t copied = 0;
 492         long timeo;
 493         int ret;
 494
 495         DEFINE_WAIT(wait);
 496
 497         trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_enter, 0, 0, 0, 0);
 498
 499         if (flags & (MSG_OOB | MSG_TRUNC))
 500                 return -EOPNOTSUPP;
 501
 502         timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
 503
 504 try_again:
 505         lock_sock(&rx->sk);
 506
 507         /* Return immediately if a client socket has no outstanding calls */
 508         if (RB_EMPTY_ROOT(&rx->calls) &&
 509             list_empty(&rx->recvmsg_q) &&
 510             rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
 511                 release_sock(&rx->sk);
 512                 return -EAGAIN;
 513         }
 514
 515         if (list_empty(&rx->recvmsg_q)) {
 516                 ret = -EWOULDBLOCK;
 517                 if (timeo == 0) {
 518                         call = NULL;
 519                         goto error_no_call;
 520                 }
 521
 522                 release_sock(&rx->sk);
 523
 524                 /* Wait for something to happen */
 525                 prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
 526                                           TASK_INTERRUPTIBLE);
 527                 ret = sock_error(&rx->sk);
 528                 if (ret)
 529                         goto wait_error;
 530
 531                 if (list_empty(&rx->recvmsg_q)) {
 532                         if (signal_pending(current))
 533                                 goto wait_interrupted;
 534                         trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_wait,
 535                                             0, 0, 0, 0);
 536                         timeo = schedule_timeout(timeo);
 537                 }
 538                 finish_wait(sk_sleep(&rx->sk), &wait);
 539                 goto try_again;
 540         }
 541
 542         /* Find the next call and dequeue it if we're not just peeking.  If we
 543          * do dequeue it, that comes with a ref that we will need to release.
 544          */
 545         write_lock_bh(&rx->recvmsg_lock);
 546         l = rx->recvmsg_q.next;
 547         call = list_entry(l, struct rxrpc_call, recvmsg_link);
 548         if (!(flags & MSG_PEEK))
 549                 list_del_init(&call->recvmsg_link);
 550         else
 551                 rxrpc_get_call(call, rxrpc_call_got);
 552         write_unlock_bh(&rx->recvmsg_lock);
 553
 554         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_dequeue, 0, 0, 0, 0);
 555
 556         /* We're going to drop the socket lock, so we need to lock the call
 557          * against interference by sendmsg.
 558          */
 559         if (!mutex_trylock(&call->user_mutex)) {
 560                 ret = -EWOULDBLOCK;
 561                 if (flags & MSG_DONTWAIT)
 562                         goto error_requeue_call;
 563                 ret = -ERESTARTSYS;
 564                 if (mutex_lock_interruptible(&call->user_mutex) < 0)
 565                         goto error_requeue_call;
 566         }
 567
 568         release_sock(&rx->sk);
 569
 570         if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
 571                 BUG();
 572
 573         if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
 574                 if (flags & MSG_CMSG_COMPAT) {
 575                         unsigned int id32 = call->user_call_ID;
 576
 577                         ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
 578                                        sizeof(unsigned int), &id32);
 579                 } else {
 580                         unsigned long idl = call->user_call_ID;
 581
 582                         ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
 583                                        sizeof(unsigned long), &idl);
 584                 }
 585                 if (ret < 0)
 586                         goto error_unlock_call;
 587         }
 588
 589         if (msg->msg_name && call->peer) {
 590                 struct sockaddr_rxrpc *srx = msg->msg_name;
 591                 size_t len = sizeof(call->peer->srx);
 592
 593                 memcpy(msg->msg_name, &call->peer->srx, len);
 594                 srx->srx_service = call->service_id;
 595                 msg->msg_namelen = len;
 596         }
 597
 598         switch (READ_ONCE(call->state)) {
 599         case RXRPC_CALL_CLIENT_RECV_REPLY:
 600         case RXRPC_CALL_SERVER_RECV_REQUEST:
 601         case RXRPC_CALL_SERVER_ACK_REQUEST:
 602                 ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
 603                                          flags, &copied);
 604                 if (ret == -EAGAIN)
 605                         ret = 0;
 606
 607                 if (after(call->rx_top, call->rx_hard_ack) &&
 608                     call->rxtx_buffer[(call->rx_hard_ack + 1) & RXRPC_RXTX_BUFF_MASK])
 609                         rxrpc_notify_socket(call);
 610                 break;
 611         default:
 612                 ret = 0;
 613                 break;
 614         }
 615
 616         if (ret < 0)
 617                 goto error_unlock_call;
 618
 619         if (call->state == RXRPC_CALL_COMPLETE) {
 620                 ret = rxrpc_recvmsg_term(call, msg);
 621                 if (ret < 0)
 622                         goto error_unlock_call;
 623                 if (!(flags & MSG_PEEK))
 624                         rxrpc_release_call(rx, call);
 625                 msg->msg_flags |= MSG_EOR;
 626                 ret = 1;
 627         }
 628
 629         if (ret == 0)
 630                 msg->msg_flags |= MSG_MORE;
 631         else
 632                 msg->msg_flags &= ~MSG_MORE;
 633         ret = copied;
 634
 635 error_unlock_call:
 636         mutex_unlock(&call->user_mutex);
 637         rxrpc_put_call(call, rxrpc_call_put);
 638         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
 639         return ret;
 640
 641 error_requeue_call:
 642         if (!(flags & MSG_PEEK)) {
 643                 write_lock_bh(&rx->recvmsg_lock);
 644                 list_add(&call->recvmsg_link, &rx->recvmsg_q);
 645                 write_unlock_bh(&rx->recvmsg_lock);
 646                 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_requeue, 0, 0, 0, 0);
 647         } else {
 648                 rxrpc_put_call(call, rxrpc_call_put);
 649         }
 650 error_no_call:
 651         release_sock(&rx->sk);
 652 error_trace:
 653         trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
 654         return ret;
 655
 656 wait_interrupted:
 657         ret = sock_intr_errno(timeo);
 658 wait_error:
 659         finish_wait(sk_sleep(&rx->sk), &wait);
 660         call = NULL;
 661         goto error_trace;
 662 }
 663
 664 /**
 665  * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
 666  * @sock: The socket that the call exists on
 667  * @call: The call to send data through
 668  * @iter: The buffer to receive into
 669  * @_len: The amount of data we want to receive (decreased on return)
 670  * @want_more: True if more data is expected to be read
 671  * @_abort: Where the abort code is stored if -ECONNABORTED is returned
 672  * @_service: Where to store the actual service ID (may be upgraded)
 673  *
 674  * Allow a kernel service to receive data and pick up information about the
 675  * state of a call.  Returns 0 if got what was asked for and there's more
 676  * available, 1 if we got what was asked for and we're at the end of the data
 677  * and -EAGAIN if we need more data.
 678  *
 679  * Note that we may return -EAGAIN to drain empty packets at the end of the
 680  * data, even if we've already copied over the requested data.
 681  *
 682  * *_abort should also be initialised to 0.
 683  */
 684 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
 685                            struct iov_iter *iter, size_t *_len,
 686                            bool want_more, u32 *_abort, u16 *_service)
 687 {
 688         size_t offset = 0;
 689         int ret;
 690
 691         _enter("{%d,%s},%zu,%d",
 692                call->debug_id, rxrpc_call_states[call->state],
 693                *_len, want_more);
 694
 695         ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_SECURING);
 696
 697         mutex_lock(&call->user_mutex);
 698
 699         switch (READ_ONCE(call->state)) {
 700         case RXRPC_CALL_CLIENT_RECV_REPLY:
 701         case RXRPC_CALL_SERVER_RECV_REQUEST:
 702         case RXRPC_CALL_SERVER_ACK_REQUEST:
 703                 ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
 704                                          *_len, 0, &offset);
 705                 *_len -= offset;
 706                 if (ret < 0)
 707                         goto out;
 708
 709                 /* We can only reach here with a partially full buffer if we
 710                  * have reached the end of the data.  We must otherwise have a
 711                  * full buffer or have been given -EAGAIN.
 712                  */
 713                 if (ret == 1) {
 714                         if (iov_iter_count(iter) > 0)
 715                                 goto short_data;
 716                         if (!want_more)
 717                                 goto read_phase_complete;
 718                         ret = 0;
 719                         goto out;
 720                 }
 721
 722                 if (!want_more)
 723                         goto excess_data;
 724                 goto out;
 725
 726         case RXRPC_CALL_COMPLETE:
 727                 goto call_complete;
 728
 729         default:
 730                 ret = -EINPROGRESS;
 731                 goto out;
 732         }
 733
 734 read_phase_complete:
 735         ret = 1;
 736 out:
 737         switch (call->ackr_reason) {
 738         case RXRPC_ACK_IDLE:
 739                 break;
 740         case RXRPC_ACK_DELAY:
 741                 if (ret != -EAGAIN)
 742                         break;
 743                 fallthrough;
 744         default:
 745                 rxrpc_send_ack_packet(call, false, NULL);
 746         }
 747
 748         if (_service)
 749                 *_service = call->service_id;
 750         mutex_unlock(&call->user_mutex);
 751         _leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
 752         return ret;
 753
 754 short_data:
 755         trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
 756         ret = -EBADMSG;
 757         goto out;
 758 excess_data:
 759         trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
 760         ret = -EMSGSIZE;
 761         goto out;
 762 call_complete:
 763         *_abort = call->abort_code;
 764         ret = call->error;
 765         if (call->completion == RXRPC_CALL_SUCCEEDED) {
 766                 ret = 1;
 767                 if (iov_iter_count(iter) > 0)
 768                         ret = -ECONNRESET;
 769         }
 770         goto out;
 771 }
 772 EXPORT_SYMBOL(rxrpc_kernel_recv_data);