1 /* RxRPC packet transmission
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/net.h>
15 #include <linux/gfp.h>
16 #include <linux/skbuff.h>
17 #include <linux/export.h>
19 #include <net/af_rxrpc.h>
20 #include "ar-internal.h"
22 struct rxrpc_ack_buffer {
23 struct rxrpc_wire_header whdr;
24 struct rxrpc_ackpacket ack;
27 struct rxrpc_ackinfo ackinfo;
30 struct rxrpc_abort_buffer {
31 struct rxrpc_wire_header whdr;
35 static const char rxrpc_keepalive_string[] = "";
38 * Arrange for a keepalive ping a certain time after we last transmitted. This
39 * lets the far side know we're still interested in this call and helps keep
40 * the route through any intervening firewall open.
42 * Receiving a response to the ping will prevent the ->expect_rx_by timer from
45 static void rxrpc_set_keepalive(struct rxrpc_call *call)
47 unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
50 WRITE_ONCE(call->keepalive_at, keepalive_at);
51 rxrpc_reduce_call_timer(call, keepalive_at, now,
52 rxrpc_timer_set_for_keepalive);
56 * Fill out an ACK packet.
58 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
59 struct rxrpc_call *call,
60 struct rxrpc_ack_buffer *pkt,
61 rxrpc_seq_t *_hard_ack,
65 rxrpc_serial_t serial;
66 rxrpc_seq_t hard_ack, top, seq;
71 /* Barrier against rxrpc_input_data(). */
72 serial = call->ackr_serial;
73 hard_ack = READ_ONCE(call->rx_hard_ack);
74 top = smp_load_acquire(&call->rx_top);
75 *_hard_ack = hard_ack;
78 pkt->ack.bufferSpace = htons(8);
79 pkt->ack.maxSkew = htons(call->ackr_skew);
80 pkt->ack.firstPacket = htonl(hard_ack + 1);
81 pkt->ack.previousPacket = htonl(call->ackr_prev_seq);
82 pkt->ack.serial = htonl(serial);
83 pkt->ack.reason = reason;
84 pkt->ack.nAcks = top - hard_ack;
86 if (reason == RXRPC_ACK_PING)
87 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
89 if (after(top, hard_ack)) {
92 ix = seq & RXRPC_RXTX_BUFF_MASK;
93 if (call->rxtx_buffer[ix])
94 *ackp++ = RXRPC_ACK_TYPE_ACK;
96 *ackp++ = RXRPC_ACK_TYPE_NACK;
98 } while (before_eq(seq, top));
101 mtu = conn->params.peer->if_mtu;
102 mtu -= conn->params.peer->hdrsize;
103 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
104 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
105 pkt->ackinfo.maxMTU = htonl(mtu);
106 pkt->ackinfo.rwind = htonl(call->rx_winsize);
107 pkt->ackinfo.jumbo_max = htonl(jmax);
112 return top - hard_ack + 3;
116 * Send an ACK call packet.
118 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
119 rxrpc_serial_t *_serial)
121 struct rxrpc_connection *conn = NULL;
122 struct rxrpc_ack_buffer *pkt;
125 rxrpc_serial_t serial;
126 rxrpc_seq_t hard_ack, top;
132 spin_lock_bh(&call->lock);
134 conn = rxrpc_get_connection_maybe(call->conn);
135 spin_unlock_bh(&call->lock);
139 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
141 rxrpc_put_connection(conn);
145 msg.msg_name = &call->peer->srx.transport;
146 msg.msg_namelen = call->peer->srx.transport_len;
147 msg.msg_control = NULL;
148 msg.msg_controllen = 0;
151 pkt->whdr.epoch = htonl(conn->proto.epoch);
152 pkt->whdr.cid = htonl(call->cid);
153 pkt->whdr.callNumber = htonl(call->call_id);
155 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
156 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
157 pkt->whdr.userStatus = 0;
158 pkt->whdr.securityIndex = call->security_ix;
160 pkt->whdr.serviceId = htons(call->service_id);
162 spin_lock_bh(&call->lock);
164 reason = RXRPC_ACK_PING;
166 reason = call->ackr_reason;
167 if (!call->ackr_reason) {
168 spin_unlock_bh(&call->lock);
172 call->ackr_reason = 0;
174 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
176 spin_unlock_bh(&call->lock);
178 iov[0].iov_base = pkt;
179 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
180 iov[1].iov_base = &pkt->ackinfo;
181 iov[1].iov_len = sizeof(pkt->ackinfo);
182 len = iov[0].iov_len + iov[1].iov_len;
184 serial = atomic_inc_return(&conn->serial);
185 pkt->whdr.serial = htonl(serial);
186 trace_rxrpc_tx_ack(call, serial,
187 ntohl(pkt->ack.firstPacket),
188 ntohl(pkt->ack.serial),
189 pkt->ack.reason, pkt->ack.nAcks);
194 call->ping_serial = serial;
196 /* We need to stick a time in before we send the packet in case
197 * the reply gets back before kernel_sendmsg() completes - but
198 * asking UDP to send the packet can take a relatively long
199 * time, so we update the time after, on the assumption that
200 * the packet transmission is more likely to happen towards the
201 * end of the kernel_sendmsg() call.
203 call->ping_time = ktime_get_real();
204 set_bit(RXRPC_CALL_PINGING, &call->flags);
205 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
208 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
209 now = ktime_get_real();
211 call->ping_time = now;
212 conn->params.peer->last_tx_at = ktime_get_seconds();
214 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
215 rxrpc_tx_fail_call_ack);
217 if (call->state < RXRPC_CALL_COMPLETE) {
220 clear_bit(RXRPC_CALL_PINGING, &call->flags);
221 rxrpc_propose_ACK(call, pkt->ack.reason,
222 ntohs(pkt->ack.maxSkew),
223 ntohl(pkt->ack.serial),
225 rxrpc_propose_ack_retry_tx);
227 spin_lock_bh(&call->lock);
228 if (after(hard_ack, call->ackr_consumed))
229 call->ackr_consumed = hard_ack;
230 if (after(top, call->ackr_seen))
231 call->ackr_seen = top;
232 spin_unlock_bh(&call->lock);
235 rxrpc_set_keepalive(call);
239 rxrpc_put_connection(conn);
245 * Send an ABORT call packet.
247 int rxrpc_send_abort_packet(struct rxrpc_call *call)
249 struct rxrpc_connection *conn = NULL;
250 struct rxrpc_abort_buffer pkt;
253 rxrpc_serial_t serial;
256 /* Don't bother sending aborts for a client call once the server has
257 * hard-ACK'd all of its request data. After that point, we're not
258 * going to stop the operation proceeding, and whilst we might limit
259 * the reply, it's not worth it if we can send a new call on the same
260 * channel instead, thereby closing off this call.
262 if (rxrpc_is_client_call(call) &&
263 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
266 spin_lock_bh(&call->lock);
268 conn = rxrpc_get_connection_maybe(call->conn);
269 spin_unlock_bh(&call->lock);
273 msg.msg_name = &call->peer->srx.transport;
274 msg.msg_namelen = call->peer->srx.transport_len;
275 msg.msg_control = NULL;
276 msg.msg_controllen = 0;
279 pkt.whdr.epoch = htonl(conn->proto.epoch);
280 pkt.whdr.cid = htonl(call->cid);
281 pkt.whdr.callNumber = htonl(call->call_id);
283 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
284 pkt.whdr.flags = conn->out_clientflag;
285 pkt.whdr.userStatus = 0;
286 pkt.whdr.securityIndex = call->security_ix;
288 pkt.whdr.serviceId = htons(call->service_id);
289 pkt.abort_code = htonl(call->abort_code);
291 iov[0].iov_base = &pkt;
292 iov[0].iov_len = sizeof(pkt);
294 serial = atomic_inc_return(&conn->serial);
295 pkt.whdr.serial = htonl(serial);
297 ret = kernel_sendmsg(conn->params.local->socket,
298 &msg, iov, 1, sizeof(pkt));
299 conn->params.peer->last_tx_at = ktime_get_seconds();
301 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
302 rxrpc_tx_fail_call_abort);
305 rxrpc_put_connection(conn);
310 * send a packet through the transport endpoint
312 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
315 struct rxrpc_connection *conn = call->conn;
316 struct rxrpc_wire_header whdr;
317 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
320 rxrpc_serial_t serial;
325 _enter(",{%d}", skb->len);
327 /* Each transmission of a Tx packet needs a new serial number */
328 serial = atomic_inc_return(&conn->serial);
330 whdr.epoch = htonl(conn->proto.epoch);
331 whdr.cid = htonl(call->cid);
332 whdr.callNumber = htonl(call->call_id);
333 whdr.seq = htonl(sp->hdr.seq);
334 whdr.serial = htonl(serial);
335 whdr.type = RXRPC_PACKET_TYPE_DATA;
336 whdr.flags = sp->hdr.flags;
338 whdr.securityIndex = call->security_ix;
339 whdr._rsvd = htons(sp->hdr._rsvd);
340 whdr.serviceId = htons(call->service_id);
342 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
344 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
346 iov[0].iov_base = &whdr;
347 iov[0].iov_len = sizeof(whdr);
348 iov[1].iov_base = skb->head;
349 iov[1].iov_len = skb->len;
350 len = iov[0].iov_len + iov[1].iov_len;
352 msg.msg_name = &call->peer->srx.transport;
353 msg.msg_namelen = call->peer->srx.transport_len;
354 msg.msg_control = NULL;
355 msg.msg_controllen = 0;
358 /* If our RTT cache needs working on, request an ACK. Also request
359 * ACKs if a DATA packet appears to have been lost.
361 if (!(sp->hdr.flags & RXRPC_LAST_PACKET) &&
362 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
364 call->cong_mode == RXRPC_CALL_SLOW_START ||
365 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
366 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
368 whdr.flags |= RXRPC_REQUEST_ACK;
370 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
372 if ((lose++ & 7) == 7) {
379 _proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
381 /* send the packet with the don't fragment bit set if we currently
382 * think it's small enough */
383 if (iov[1].iov_len >= call->peer->maxdata)
384 goto send_fragmentable;
386 down_read(&conn->params.local->defrag_sem);
387 /* send the packet by UDP
388 * - returns -EMSGSIZE if UDP would have to fragment the packet
389 * to go out of the interface
390 * - in which case, we'll have processed the ICMP error
391 * message and update the peer record
393 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
394 conn->params.peer->last_tx_at = ktime_get_seconds();
396 up_read(&conn->params.local->defrag_sem);
398 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
399 rxrpc_tx_fail_call_data_nofrag);
400 if (ret == -EMSGSIZE)
401 goto send_fragmentable;
404 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
407 ktime_t now = ktime_get_real();
410 sp->hdr.serial = serial;
411 if (whdr.flags & RXRPC_REQUEST_ACK) {
412 call->peer->rtt_last_req = now;
413 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
414 if (call->peer->rtt_usage > 1) {
415 unsigned long nowj = jiffies, ack_lost_at;
417 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
422 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
423 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
424 rxrpc_timer_set_for_lost_ack);
428 if (sp->hdr.seq == 1 &&
429 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
431 unsigned long nowj = jiffies, expect_rx_by;
433 expect_rx_by = nowj + call->next_rx_timo;
434 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
435 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
436 rxrpc_timer_set_for_normal);
440 rxrpc_set_keepalive(call);
442 _leave(" = %d [%u]", ret, call->peer->maxdata);
446 /* attempt to send this message with fragmentation enabled */
447 _debug("send fragment");
449 down_write(&conn->params.local->defrag_sem);
451 switch (conn->params.local->srx.transport.family) {
453 opt = IP_PMTUDISC_DONT;
454 ret = kernel_setsockopt(conn->params.local->socket,
455 SOL_IP, IP_MTU_DISCOVER,
456 (char *)&opt, sizeof(opt));
458 ret = kernel_sendmsg(conn->params.local->socket, &msg,
460 conn->params.peer->last_tx_at = ktime_get_seconds();
462 opt = IP_PMTUDISC_DO;
463 kernel_setsockopt(conn->params.local->socket, SOL_IP,
465 (char *)&opt, sizeof(opt));
469 #ifdef CONFIG_AF_RXRPC_IPV6
471 opt = IPV6_PMTUDISC_DONT;
472 ret = kernel_setsockopt(conn->params.local->socket,
473 SOL_IPV6, IPV6_MTU_DISCOVER,
474 (char *)&opt, sizeof(opt));
476 ret = kernel_sendmsg(conn->params.local->socket, &msg,
478 conn->params.peer->last_tx_at = ktime_get_seconds();
480 opt = IPV6_PMTUDISC_DO;
481 kernel_setsockopt(conn->params.local->socket,
482 SOL_IPV6, IPV6_MTU_DISCOVER,
483 (char *)&opt, sizeof(opt));
490 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
491 rxrpc_tx_fail_call_data_frag);
493 up_write(&conn->params.local->defrag_sem);
498 * reject packets through the local endpoint
500 void rxrpc_reject_packets(struct rxrpc_local *local)
502 struct sockaddr_rxrpc srx;
503 struct rxrpc_skb_priv *sp;
504 struct rxrpc_wire_header whdr;
512 _enter("%d", local->debug_id);
514 iov[0].iov_base = &whdr;
515 iov[0].iov_len = sizeof(whdr);
516 iov[1].iov_base = &code;
517 iov[1].iov_len = sizeof(code);
518 size = sizeof(whdr) + sizeof(code);
520 msg.msg_name = &srx.transport;
521 msg.msg_control = NULL;
522 msg.msg_controllen = 0;
525 memset(&whdr, 0, sizeof(whdr));
526 whdr.type = RXRPC_PACKET_TYPE_ABORT;
528 while ((skb = skb_dequeue(&local->reject_queue))) {
529 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
532 if (rxrpc_extract_addr_from_skb(local, &srx, skb) == 0) {
533 msg.msg_namelen = srx.transport_len;
535 code = htonl(skb->priority);
537 whdr.epoch = htonl(sp->hdr.epoch);
538 whdr.cid = htonl(sp->hdr.cid);
539 whdr.callNumber = htonl(sp->hdr.callNumber);
540 whdr.serviceId = htons(sp->hdr.serviceId);
541 whdr.flags = sp->hdr.flags;
542 whdr.flags ^= RXRPC_CLIENT_INITIATED;
543 whdr.flags &= RXRPC_CLIENT_INITIATED;
545 ret = kernel_sendmsg(local->socket, &msg, iov, 2, size);
547 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
548 rxrpc_tx_fail_reject);
551 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
558 * Send a VERSION reply to a peer as a keepalive.
560 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
562 struct rxrpc_wire_header whdr;
570 msg.msg_name = &peer->srx.transport;
571 msg.msg_namelen = peer->srx.transport_len;
572 msg.msg_control = NULL;
573 msg.msg_controllen = 0;
576 whdr.epoch = htonl(peer->local->rxnet->epoch);
581 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
582 whdr.flags = RXRPC_LAST_PACKET;
584 whdr.securityIndex = 0;
588 iov[0].iov_base = &whdr;
589 iov[0].iov_len = sizeof(whdr);
590 iov[1].iov_base = (char *)rxrpc_keepalive_string;
591 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
593 len = iov[0].iov_len + iov[1].iov_len;
595 _proto("Tx VERSION (keepalive)");
597 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
599 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
600 rxrpc_tx_fail_version_keepalive);
602 peer->last_tx_at = ktime_get_seconds();