net/rxrpc/output.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* RxRPC packet transmission
   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/gfp.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/export.h>
  14 #include <net/sock.h>
  15 #include <net/af_rxrpc.h>
  16 #include <net/udp.h>
  17 #include "ar-internal.h"
  18
  19 extern int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
  20
  21 static ssize_t do_udp_sendmsg(struct socket *socket, struct msghdr *msg, size_t len)
  22 {
  23         struct sockaddr *sa = msg->msg_name;
  24         struct sock *sk = socket->sk;
  25
  26         if (IS_ENABLED(CONFIG_AF_RXRPC_IPV6)) {
  27                 if (sa->sa_family == AF_INET6) {
  28                         if (sk->sk_family != AF_INET6) {
  29                                 pr_warn("AF_INET6 address on AF_INET socket\n");
  30                                 return -ENOPROTOOPT;
  31                         }
  32                         return udpv6_sendmsg(sk, msg, len);
  33                 }
  34         }
  35         return udp_sendmsg(sk, msg, len);
  36 }
  37
  38 struct rxrpc_abort_buffer {
  39         struct rxrpc_wire_header whdr;
  40         __be32 abort_code;
  41 };
  42
  43 static const char rxrpc_keepalive_string[] = "";
  44
  45 /*
  46  * Increase Tx backoff on transmission failure and clear it on success.
  47  */
  48 static void rxrpc_tx_backoff(struct rxrpc_call *call, int ret)
  49 {
  50         if (ret < 0) {
  51                 u16 tx_backoff = READ_ONCE(call->tx_backoff);
  52
  53                 if (tx_backoff < HZ)
  54                         WRITE_ONCE(call->tx_backoff, tx_backoff + 1);
  55         } else {
  56                 WRITE_ONCE(call->tx_backoff, 0);
  57         }
  58 }
  59
  60 /*
  61  * Arrange for a keepalive ping a certain time after we last transmitted.  This
  62  * lets the far side know we're still interested in this call and helps keep
  63  * the route through any intervening firewall open.
  64  *
  65  * Receiving a response to the ping will prevent the ->expect_rx_by timer from
  66  * expiring.
  67  */
  68 static void rxrpc_set_keepalive(struct rxrpc_call *call)
  69 {
  70         unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
  71
  72         keepalive_at += now;
  73         WRITE_ONCE(call->keepalive_at, keepalive_at);
  74         rxrpc_reduce_call_timer(call, keepalive_at, now,
  75                                 rxrpc_timer_set_for_keepalive);
  76 }
  77
  78 /*
  79  * Fill out an ACK packet.
  80  */
  81 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
  82                                  struct rxrpc_call *call,
  83                                  struct rxrpc_txbuf *txb)
  84 {
  85         struct rxrpc_ackinfo ackinfo;
  86         unsigned int qsize;
  87         rxrpc_seq_t window, wtop, wrap_point, ix, first;
  88         int rsize;
  89         u64 wtmp;
  90         u32 mtu, jmax;
  91         u8 *ackp = txb->acks;
  92         u8 sack_buffer[sizeof(call->ackr_sack_table)] __aligned(8);
  93
  94         atomic_set(&call->ackr_nr_unacked, 0);
  95         atomic_set(&call->ackr_nr_consumed, 0);
  96         rxrpc_inc_stat(call->rxnet, stat_tx_ack_fill);
  97
  98         /* Barrier against rxrpc_input_data(). */
  99 retry:
 100         wtmp   = atomic64_read_acquire(&call->ackr_window);
 101         window = lower_32_bits(wtmp);
 102         wtop   = upper_32_bits(wtmp);
 103         txb->ack.firstPacket = htonl(window);
 104         txb->ack.nAcks = 0;
 105
 106         if (after(wtop, window)) {
 107                 /* Try to copy the SACK ring locklessly.  We can use the copy,
 108                  * only if the now-current top of the window didn't go past the
 109                  * previously read base - otherwise we can't know whether we
 110                  * have old data or new data.
 111                  */
 112                 memcpy(sack_buffer, call->ackr_sack_table, sizeof(sack_buffer));
 113                 wrap_point = window + RXRPC_SACK_SIZE - 1;
 114                 wtmp   = atomic64_read_acquire(&call->ackr_window);
 115                 window = lower_32_bits(wtmp);
 116                 wtop   = upper_32_bits(wtmp);
 117                 if (after(wtop, wrap_point)) {
 118                         cond_resched();
 119                         goto retry;
 120                 }
 121
 122                 /* The buffer is maintained as a ring with an invariant mapping
 123                  * between bit position and sequence number, so we'll probably
 124                  * need to rotate it.
 125                  */
 126                 txb->ack.nAcks = wtop - window;
 127                 ix = window % RXRPC_SACK_SIZE;
 128                 first = sizeof(sack_buffer) - ix;
 129
 130                 if (ix + txb->ack.nAcks <= RXRPC_SACK_SIZE) {
 131                         memcpy(txb->acks, sack_buffer + ix, txb->ack.nAcks);
 132                 } else {
 133                         memcpy(txb->acks, sack_buffer + ix, first);
 134                         memcpy(txb->acks + first, sack_buffer,
 135                                txb->ack.nAcks - first);
 136                 }
 137
 138                 ackp += txb->ack.nAcks;
 139         } else if (before(wtop, window)) {
 140                 pr_warn("ack window backward %x %x", window, wtop);
 141         } else if (txb->ack.reason == RXRPC_ACK_DELAY) {
 142                 txb->ack.reason = RXRPC_ACK_IDLE;
 143         }
 144
 145         mtu = conn->params.peer->if_mtu;
 146         mtu -= conn->params.peer->hdrsize;
 147         jmax = rxrpc_rx_jumbo_max;
 148         qsize = (window - 1) - call->rx_consumed;
 149         rsize = max_t(int, call->rx_winsize - qsize, 0);
 150         ackinfo.rxMTU           = htonl(rxrpc_rx_mtu);
 151         ackinfo.maxMTU          = htonl(mtu);
 152         ackinfo.rwind           = htonl(rsize);
 153         ackinfo.jumbo_max       = htonl(jmax);
 154
 155         *ackp++ = 0;
 156         *ackp++ = 0;
 157         *ackp++ = 0;
 158         memcpy(ackp, &ackinfo, sizeof(ackinfo));
 159         return txb->ack.nAcks + 3 + sizeof(ackinfo);
 160 }
 161
 162 /*
 163  * Record the beginning of an RTT probe.
 164  */
 165 static int rxrpc_begin_rtt_probe(struct rxrpc_call *call, rxrpc_serial_t serial,
 166                                  enum rxrpc_rtt_tx_trace why)
 167 {
 168         unsigned long avail = call->rtt_avail;
 169         int rtt_slot = 9;
 170
 171         if (!(avail & RXRPC_CALL_RTT_AVAIL_MASK))
 172                 goto no_slot;
 173
 174         rtt_slot = __ffs(avail & RXRPC_CALL_RTT_AVAIL_MASK);
 175         if (!test_and_clear_bit(rtt_slot, &call->rtt_avail))
 176                 goto no_slot;
 177
 178         call->rtt_serial[rtt_slot] = serial;
 179         call->rtt_sent_at[rtt_slot] = ktime_get_real();
 180         smp_wmb(); /* Write data before avail bit */
 181         set_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
 182
 183         trace_rxrpc_rtt_tx(call, why, rtt_slot, serial);
 184         return rtt_slot;
 185
 186 no_slot:
 187         trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_no_slot, rtt_slot, serial);
 188         return -1;
 189 }
 190
 191 /*
 192  * Cancel an RTT probe.
 193  */
 194 static void rxrpc_cancel_rtt_probe(struct rxrpc_call *call,
 195                                    rxrpc_serial_t serial, int rtt_slot)
 196 {
 197         if (rtt_slot != -1) {
 198                 clear_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
 199                 smp_wmb(); /* Clear pending bit before setting slot */
 200                 set_bit(rtt_slot, &call->rtt_avail);
 201                 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_cancel, rtt_slot, serial);
 202         }
 203 }
 204
 205 /*
 206  * Send an ACK call packet.
 207  */
 208 static int rxrpc_send_ack_packet(struct rxrpc_local *local, struct rxrpc_txbuf *txb)
 209 {
 210         struct rxrpc_connection *conn;
 211         struct rxrpc_ack_buffer *pkt;
 212         struct rxrpc_call *call = txb->call;
 213         struct msghdr msg;
 214         struct kvec iov[1];
 215         rxrpc_serial_t serial;
 216         size_t len, n;
 217         int ret, rtt_slot = -1;
 218
 219         if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
 220                 return -ECONNRESET;
 221
 222         conn = call->conn;
 223
 224         msg.msg_name    = &call->peer->srx.transport;
 225         msg.msg_namelen = call->peer->srx.transport_len;
 226         msg.msg_control = NULL;
 227         msg.msg_controllen = 0;
 228         msg.msg_flags   = 0;
 229
 230         if (txb->ack.reason == RXRPC_ACK_PING)
 231                 txb->wire.flags |= RXRPC_REQUEST_ACK;
 232
 233         if (txb->ack.reason == RXRPC_ACK_DELAY)
 234                 clear_bit(RXRPC_CALL_DELAY_ACK_PENDING, &call->flags);
 235         if (txb->ack.reason == RXRPC_ACK_IDLE)
 236                 clear_bit(RXRPC_CALL_IDLE_ACK_PENDING, &call->flags);
 237
 238         n = rxrpc_fill_out_ack(conn, call, txb);
 239         if (n == 0)
 240                 return 0;
 241
 242         iov[0].iov_base = &txb->wire;
 243         iov[0].iov_len  = sizeof(txb->wire) + sizeof(txb->ack) + n;
 244         len = iov[0].iov_len;
 245
 246         serial = atomic_inc_return(&conn->serial);
 247         txb->wire.serial = htonl(serial);
 248         trace_rxrpc_tx_ack(call->debug_id, serial,
 249                            ntohl(txb->ack.firstPacket),
 250                            ntohl(txb->ack.serial), txb->ack.reason, txb->ack.nAcks);
 251         if (txb->ack_why == rxrpc_propose_ack_ping_for_lost_ack)
 252                 call->acks_lost_ping = serial;
 253
 254         if (txb->ack.reason == RXRPC_ACK_PING)
 255                 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_ping);
 256
 257         rxrpc_inc_stat(call->rxnet, stat_tx_ack_send);
 258
 259         /* Grab the highest received seq as late as possible */
 260         txb->ack.previousPacket = htonl(call->rx_highest_seq);
 261
 262         iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, len);
 263         ret = do_udp_sendmsg(conn->params.local->socket, &msg, len);
 264         call->peer->last_tx_at = ktime_get_seconds();
 265         if (ret < 0)
 266                 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 267                                     rxrpc_tx_point_call_ack);
 268         else
 269                 trace_rxrpc_tx_packet(call->debug_id, &txb->wire,
 270                                       rxrpc_tx_point_call_ack);
 271         rxrpc_tx_backoff(call, ret);
 272
 273         if (call->state < RXRPC_CALL_COMPLETE) {
 274                 if (ret < 0)
 275                         rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
 276                 rxrpc_set_keepalive(call);
 277         }
 278
 279         kfree(pkt);
 280         return ret;
 281 }
 282
 283 /*
 284  * ACK transmitter for a local endpoint.  The UDP socket locks around each
 285  * transmission, so we can only transmit one packet at a time, ACK, DATA or
 286  * otherwise.
 287  */
 288 void rxrpc_transmit_ack_packets(struct rxrpc_local *local)
 289 {
 290         LIST_HEAD(queue);
 291         int ret;
 292
 293         trace_rxrpc_local(local->debug_id, rxrpc_local_tx_ack,
 294                           refcount_read(&local->ref), NULL);
 295
 296         if (list_empty(&local->ack_tx_queue))
 297                 return;
 298
 299         spin_lock_bh(&local->ack_tx_lock);
 300         list_splice_tail_init(&local->ack_tx_queue, &queue);
 301         spin_unlock_bh(&local->ack_tx_lock);
 302
 303         while (!list_empty(&queue)) {
 304                 struct rxrpc_txbuf *txb =
 305                         list_entry(queue.next, struct rxrpc_txbuf, tx_link);
 306
 307                 ret = rxrpc_send_ack_packet(local, txb);
 308                 if (ret < 0 && ret != -ECONNRESET) {
 309                         spin_lock_bh(&local->ack_tx_lock);
 310                         list_splice_init(&queue, &local->ack_tx_queue);
 311                         spin_unlock_bh(&local->ack_tx_lock);
 312                         break;
 313                 }
 314
 315                 list_del_init(&txb->tx_link);
 316                 rxrpc_put_call(txb->call, rxrpc_call_put);
 317                 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_ack_tx);
 318         }
 319 }
 320
 321 /*
 322  * Send an ABORT call packet.
 323  */
 324 int rxrpc_send_abort_packet(struct rxrpc_call *call)
 325 {
 326         struct rxrpc_connection *conn;
 327         struct rxrpc_abort_buffer pkt;
 328         struct msghdr msg;
 329         struct kvec iov[1];
 330         rxrpc_serial_t serial;
 331         int ret;
 332
 333         /* Don't bother sending aborts for a client call once the server has
 334          * hard-ACK'd all of its request data.  After that point, we're not
 335          * going to stop the operation proceeding, and whilst we might limit
 336          * the reply, it's not worth it if we can send a new call on the same
 337          * channel instead, thereby closing off this call.
 338          */
 339         if (rxrpc_is_client_call(call) &&
 340             test_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags))
 341                 return 0;
 342
 343         if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
 344                 return -ECONNRESET;
 345
 346         conn = call->conn;
 347
 348         msg.msg_name    = &call->peer->srx.transport;
 349         msg.msg_namelen = call->peer->srx.transport_len;
 350         msg.msg_control = NULL;
 351         msg.msg_controllen = 0;
 352         msg.msg_flags   = 0;
 353
 354         pkt.whdr.epoch          = htonl(conn->proto.epoch);
 355         pkt.whdr.cid            = htonl(call->cid);
 356         pkt.whdr.callNumber     = htonl(call->call_id);
 357         pkt.whdr.seq            = 0;
 358         pkt.whdr.type           = RXRPC_PACKET_TYPE_ABORT;
 359         pkt.whdr.flags          = conn->out_clientflag;
 360         pkt.whdr.userStatus     = 0;
 361         pkt.whdr.securityIndex  = call->security_ix;
 362         pkt.whdr._rsvd          = 0;
 363         pkt.whdr.serviceId      = htons(call->service_id);
 364         pkt.abort_code          = htonl(call->abort_code);
 365
 366         iov[0].iov_base = &pkt;
 367         iov[0].iov_len  = sizeof(pkt);
 368
 369         serial = atomic_inc_return(&conn->serial);
 370         pkt.whdr.serial = htonl(serial);
 371
 372         iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, sizeof(pkt));
 373         ret = do_udp_sendmsg(conn->params.local->socket, &msg, sizeof(pkt));
 374         conn->params.peer->last_tx_at = ktime_get_seconds();
 375         if (ret < 0)
 376                 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 377                                     rxrpc_tx_point_call_abort);
 378         else
 379                 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
 380                                       rxrpc_tx_point_call_abort);
 381         rxrpc_tx_backoff(call, ret);
 382         return ret;
 383 }
 384
 385 /*
 386  * send a packet through the transport endpoint
 387  */
 388 int rxrpc_send_data_packet(struct rxrpc_call *call, struct rxrpc_txbuf *txb)
 389 {
 390         enum rxrpc_req_ack_trace why;
 391         struct rxrpc_connection *conn = call->conn;
 392         struct msghdr msg;
 393         struct kvec iov[1];
 394         rxrpc_serial_t serial;
 395         size_t len;
 396         int ret, rtt_slot = -1;
 397
 398         _enter("%x,{%d}", txb->seq, txb->len);
 399
 400         if (hlist_unhashed(&call->error_link)) {
 401                 spin_lock_bh(&call->peer->lock);
 402                 hlist_add_head_rcu(&call->error_link, &call->peer->error_targets);
 403                 spin_unlock_bh(&call->peer->lock);
 404         }
 405
 406         /* Each transmission of a Tx packet needs a new serial number */
 407         serial = atomic_inc_return(&conn->serial);
 408         txb->wire.serial = htonl(serial);
 409
 410         if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
 411             txb->seq == 1)
 412                 txb->wire.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
 413
 414         iov[0].iov_base = &txb->wire;
 415         iov[0].iov_len = sizeof(txb->wire) + txb->len;
 416         len = iov[0].iov_len;
 417         iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, len);
 418
 419         msg.msg_name = &call->peer->srx.transport;
 420         msg.msg_namelen = call->peer->srx.transport_len;
 421         msg.msg_control = NULL;
 422         msg.msg_controllen = 0;
 423         msg.msg_flags = 0;
 424
 425         /* If our RTT cache needs working on, request an ACK.  Also request
 426          * ACKs if a DATA packet appears to have been lost.
 427          *
 428          * However, we mustn't request an ACK on the last reply packet of a
 429          * service call, lest OpenAFS incorrectly send us an ACK with some
 430          * soft-ACKs in it and then never follow up with a proper hard ACK.
 431          */
 432         if (txb->wire.flags & RXRPC_REQUEST_ACK)
 433                 why = rxrpc_reqack_already_on;
 434         else if (test_bit(RXRPC_TXBUF_LAST, &txb->flags) && rxrpc_sending_to_client(txb))
 435                 why = rxrpc_reqack_no_srv_last;
 436         else if (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events))
 437                 why = rxrpc_reqack_ack_lost;
 438         else if (test_bit(RXRPC_TXBUF_RESENT, &txb->flags))
 439                 why = rxrpc_reqack_retrans;
 440         else if (call->cong_mode == RXRPC_CALL_SLOW_START && call->cong_cwnd <= 2)
 441                 why = rxrpc_reqack_slow_start;
 442         else if (call->tx_winsize <= 2)
 443                 why = rxrpc_reqack_small_txwin;
 444         else if (call->peer->rtt_count < 3 && txb->seq & 1)
 445                 why = rxrpc_reqack_more_rtt;
 446         else if (ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), ktime_get_real()))
 447                 why = rxrpc_reqack_old_rtt;
 448         else
 449                 goto dont_set_request_ack;
 450
 451         rxrpc_inc_stat(call->rxnet, stat_why_req_ack[why]);
 452         trace_rxrpc_req_ack(call->debug_id, txb->seq, why);
 453         if (why != rxrpc_reqack_no_srv_last)
 454                 txb->wire.flags |= RXRPC_REQUEST_ACK;
 455 dont_set_request_ack:
 456
 457         if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
 458                 static int lose;
 459                 if ((lose++ & 7) == 7) {
 460                         ret = 0;
 461                         trace_rxrpc_tx_data(call, txb->seq, serial,
 462                                             txb->wire.flags,
 463                                             test_bit(RXRPC_TXBUF_RESENT, &txb->flags),
 464                                             true);
 465                         goto done;
 466                 }
 467         }
 468
 469         trace_rxrpc_tx_data(call, txb->seq, serial, txb->wire.flags,
 470                             test_bit(RXRPC_TXBUF_RESENT, &txb->flags), false);
 471         cmpxchg(&call->tx_transmitted, txb->seq - 1, txb->seq);
 472
 473         /* send the packet with the don't fragment bit set if we currently
 474          * think it's small enough */
 475         if (txb->len >= call->peer->maxdata)
 476                 goto send_fragmentable;
 477
 478         down_read(&conn->params.local->defrag_sem);
 479
 480         txb->last_sent = ktime_get_real();
 481         if (txb->wire.flags & RXRPC_REQUEST_ACK)
 482                 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
 483
 484         /* send the packet by UDP
 485          * - returns -EMSGSIZE if UDP would have to fragment the packet
 486          *   to go out of the interface
 487          *   - in which case, we'll have processed the ICMP error
 488          *     message and update the peer record
 489          */
 490         rxrpc_inc_stat(call->rxnet, stat_tx_data_send);
 491         ret = do_udp_sendmsg(conn->params.local->socket, &msg, len);
 492         conn->params.peer->last_tx_at = ktime_get_seconds();
 493
 494         up_read(&conn->params.local->defrag_sem);
 495         if (ret < 0) {
 496                 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
 497                 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 498                                     rxrpc_tx_point_call_data_nofrag);
 499         } else {
 500                 trace_rxrpc_tx_packet(call->debug_id, &txb->wire,
 501                                       rxrpc_tx_point_call_data_nofrag);
 502         }
 503
 504         rxrpc_tx_backoff(call, ret);
 505         if (ret == -EMSGSIZE)
 506                 goto send_fragmentable;
 507
 508 done:
 509         if (ret >= 0) {
 510                 call->tx_last_sent = txb->last_sent;
 511                 if (txb->wire.flags & RXRPC_REQUEST_ACK) {
 512                         call->peer->rtt_last_req = txb->last_sent;
 513                         if (call->peer->rtt_count > 1) {
 514                                 unsigned long nowj = jiffies, ack_lost_at;
 515
 516                                 ack_lost_at = rxrpc_get_rto_backoff(call->peer, false);
 517                                 ack_lost_at += nowj;
 518                                 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
 519                                 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
 520                                                         rxrpc_timer_set_for_lost_ack);
 521                         }
 522                 }
 523
 524                 if (txb->seq == 1 &&
 525                     !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
 526                                       &call->flags)) {
 527                         unsigned long nowj = jiffies, expect_rx_by;
 528
 529                         expect_rx_by = nowj + call->next_rx_timo;
 530                         WRITE_ONCE(call->expect_rx_by, expect_rx_by);
 531                         rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
 532                                                 rxrpc_timer_set_for_normal);
 533                 }
 534
 535                 rxrpc_set_keepalive(call);
 536         } else {
 537                 /* Cancel the call if the initial transmission fails,
 538                  * particularly if that's due to network routing issues that
 539                  * aren't going away anytime soon.  The layer above can arrange
 540                  * the retransmission.
 541                  */
 542                 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
 543                         rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
 544                                                   RX_USER_ABORT, ret);
 545         }
 546
 547         _leave(" = %d [%u]", ret, call->peer->maxdata);
 548         return ret;
 549
 550 send_fragmentable:
 551         /* attempt to send this message with fragmentation enabled */
 552         _debug("send fragment");
 553
 554         down_write(&conn->params.local->defrag_sem);
 555
 556         txb->last_sent = ktime_get_real();
 557         if (txb->wire.flags & RXRPC_REQUEST_ACK)
 558                 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
 559
 560         switch (conn->params.local->srx.transport.family) {
 561         case AF_INET6:
 562         case AF_INET:
 563                 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
 564                                          IP_PMTUDISC_DONT);
 565                 rxrpc_inc_stat(call->rxnet, stat_tx_data_send_frag);
 566                 ret = do_udp_sendmsg(conn->params.local->socket, &msg, len);
 567                 conn->params.peer->last_tx_at = ktime_get_seconds();
 568
 569                 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
 570                                          IP_PMTUDISC_DO);
 571                 break;
 572
 573         default:
 574                 BUG();
 575         }
 576
 577         if (ret < 0) {
 578                 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
 579                 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 580                                     rxrpc_tx_point_call_data_frag);
 581         } else {
 582                 trace_rxrpc_tx_packet(call->debug_id, &txb->wire,
 583                                       rxrpc_tx_point_call_data_frag);
 584         }
 585         rxrpc_tx_backoff(call, ret);
 586
 587         up_write(&conn->params.local->defrag_sem);
 588         goto done;
 589 }
 590
 591 /*
 592  * reject packets through the local endpoint
 593  */
 594 void rxrpc_reject_packets(struct rxrpc_local *local)
 595 {
 596         struct sockaddr_rxrpc srx;
 597         struct rxrpc_skb_priv *sp;
 598         struct rxrpc_wire_header whdr;
 599         struct sk_buff *skb;
 600         struct msghdr msg;
 601         struct kvec iov[2];
 602         size_t size;
 603         __be32 code;
 604         int ret, ioc;
 605
 606         _enter("%d", local->debug_id);
 607
 608         iov[0].iov_base = &whdr;
 609         iov[0].iov_len = sizeof(whdr);
 610         iov[1].iov_base = &code;
 611         iov[1].iov_len = sizeof(code);
 612
 613         msg.msg_name = &srx.transport;
 614         msg.msg_control = NULL;
 615         msg.msg_controllen = 0;
 616         msg.msg_flags = 0;
 617
 618         memset(&whdr, 0, sizeof(whdr));
 619
 620         while ((skb = skb_dequeue(&local->reject_queue))) {
 621                 rxrpc_see_skb(skb, rxrpc_skb_seen);
 622                 sp = rxrpc_skb(skb);
 623
 624                 switch (skb->mark) {
 625                 case RXRPC_SKB_MARK_REJECT_BUSY:
 626                         whdr.type = RXRPC_PACKET_TYPE_BUSY;
 627                         size = sizeof(whdr);
 628                         ioc = 1;
 629                         break;
 630                 case RXRPC_SKB_MARK_REJECT_ABORT:
 631                         whdr.type = RXRPC_PACKET_TYPE_ABORT;
 632                         code = htonl(skb->priority);
 633                         size = sizeof(whdr) + sizeof(code);
 634                         ioc = 2;
 635                         break;
 636                 default:
 637                         rxrpc_free_skb(skb, rxrpc_skb_freed);
 638                         continue;
 639                 }
 640
 641                 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
 642                         msg.msg_namelen = srx.transport_len;
 643
 644                         whdr.epoch      = htonl(sp->hdr.epoch);
 645                         whdr.cid        = htonl(sp->hdr.cid);
 646                         whdr.callNumber = htonl(sp->hdr.callNumber);
 647                         whdr.serviceId  = htons(sp->hdr.serviceId);
 648                         whdr.flags      = sp->hdr.flags;
 649                         whdr.flags      ^= RXRPC_CLIENT_INITIATED;
 650                         whdr.flags      &= RXRPC_CLIENT_INITIATED;
 651
 652                         iov_iter_kvec(&msg.msg_iter, WRITE, iov, ioc, size);
 653                         ret = do_udp_sendmsg(local->socket, &msg, size);
 654                         if (ret < 0)
 655                                 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
 656                                                     rxrpc_tx_point_reject);
 657                         else
 658                                 trace_rxrpc_tx_packet(local->debug_id, &whdr,
 659                                                       rxrpc_tx_point_reject);
 660                 }
 661
 662                 rxrpc_free_skb(skb, rxrpc_skb_freed);
 663         }
 664
 665         _leave("");
 666 }
 667
 668 /*
 669  * Send a VERSION reply to a peer as a keepalive.
 670  */
 671 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
 672 {
 673         struct rxrpc_wire_header whdr;
 674         struct msghdr msg;
 675         struct kvec iov[2];
 676         size_t len;
 677         int ret;
 678
 679         _enter("");
 680
 681         msg.msg_name    = &peer->srx.transport;
 682         msg.msg_namelen = peer->srx.transport_len;
 683         msg.msg_control = NULL;
 684         msg.msg_controllen = 0;
 685         msg.msg_flags   = 0;
 686
 687         whdr.epoch      = htonl(peer->local->rxnet->epoch);
 688         whdr.cid        = 0;
 689         whdr.callNumber = 0;
 690         whdr.seq        = 0;
 691         whdr.serial     = 0;
 692         whdr.type       = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
 693         whdr.flags      = RXRPC_LAST_PACKET;
 694         whdr.userStatus = 0;
 695         whdr.securityIndex = 0;
 696         whdr._rsvd      = 0;
 697         whdr.serviceId  = 0;
 698
 699         iov[0].iov_base = &whdr;
 700         iov[0].iov_len  = sizeof(whdr);
 701         iov[1].iov_base = (char *)rxrpc_keepalive_string;
 702         iov[1].iov_len  = sizeof(rxrpc_keepalive_string);
 703
 704         len = iov[0].iov_len + iov[1].iov_len;
 705
 706         _proto("Tx VERSION (keepalive)");
 707
 708         iov_iter_kvec(&msg.msg_iter, WRITE, iov, 2, len);
 709         ret = do_udp_sendmsg(peer->local->socket, &msg, len);
 710         if (ret < 0)
 711                 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
 712                                     rxrpc_tx_point_version_keepalive);
 713         else
 714                 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
 715                                       rxrpc_tx_point_version_keepalive);
 716
 717         peer->last_tx_at = ktime_get_seconds();
 718         _leave("");
 719 }