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->debug_id, 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_point_call_ack);
217 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
218 rxrpc_tx_point_call_ack);
220 if (call->state < RXRPC_CALL_COMPLETE) {
223 clear_bit(RXRPC_CALL_PINGING, &call->flags);
224 rxrpc_propose_ACK(call, pkt->ack.reason,
225 ntohs(pkt->ack.maxSkew),
226 ntohl(pkt->ack.serial),
228 rxrpc_propose_ack_retry_tx);
230 spin_lock_bh(&call->lock);
231 if (after(hard_ack, call->ackr_consumed))
232 call->ackr_consumed = hard_ack;
233 if (after(top, call->ackr_seen))
234 call->ackr_seen = top;
235 spin_unlock_bh(&call->lock);
238 rxrpc_set_keepalive(call);
242 rxrpc_put_connection(conn);
248 * Send an ABORT call packet.
250 int rxrpc_send_abort_packet(struct rxrpc_call *call)
252 struct rxrpc_connection *conn = NULL;
253 struct rxrpc_abort_buffer pkt;
256 rxrpc_serial_t serial;
259 /* Don't bother sending aborts for a client call once the server has
260 * hard-ACK'd all of its request data. After that point, we're not
261 * going to stop the operation proceeding, and whilst we might limit
262 * the reply, it's not worth it if we can send a new call on the same
263 * channel instead, thereby closing off this call.
265 if (rxrpc_is_client_call(call) &&
266 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
269 spin_lock_bh(&call->lock);
271 conn = rxrpc_get_connection_maybe(call->conn);
272 spin_unlock_bh(&call->lock);
276 msg.msg_name = &call->peer->srx.transport;
277 msg.msg_namelen = call->peer->srx.transport_len;
278 msg.msg_control = NULL;
279 msg.msg_controllen = 0;
282 pkt.whdr.epoch = htonl(conn->proto.epoch);
283 pkt.whdr.cid = htonl(call->cid);
284 pkt.whdr.callNumber = htonl(call->call_id);
286 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
287 pkt.whdr.flags = conn->out_clientflag;
288 pkt.whdr.userStatus = 0;
289 pkt.whdr.securityIndex = call->security_ix;
291 pkt.whdr.serviceId = htons(call->service_id);
292 pkt.abort_code = htonl(call->abort_code);
294 iov[0].iov_base = &pkt;
295 iov[0].iov_len = sizeof(pkt);
297 serial = atomic_inc_return(&conn->serial);
298 pkt.whdr.serial = htonl(serial);
300 ret = kernel_sendmsg(conn->params.local->socket,
301 &msg, iov, 1, sizeof(pkt));
302 conn->params.peer->last_tx_at = ktime_get_seconds();
304 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
305 rxrpc_tx_point_call_abort);
307 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
308 rxrpc_tx_point_call_abort);
311 rxrpc_put_connection(conn);
316 * send a packet through the transport endpoint
318 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
321 struct rxrpc_connection *conn = call->conn;
322 struct rxrpc_wire_header whdr;
323 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
326 rxrpc_serial_t serial;
331 _enter(",{%d}", skb->len);
333 /* Each transmission of a Tx packet needs a new serial number */
334 serial = atomic_inc_return(&conn->serial);
336 whdr.epoch = htonl(conn->proto.epoch);
337 whdr.cid = htonl(call->cid);
338 whdr.callNumber = htonl(call->call_id);
339 whdr.seq = htonl(sp->hdr.seq);
340 whdr.serial = htonl(serial);
341 whdr.type = RXRPC_PACKET_TYPE_DATA;
342 whdr.flags = sp->hdr.flags;
344 whdr.securityIndex = call->security_ix;
345 whdr._rsvd = htons(sp->hdr._rsvd);
346 whdr.serviceId = htons(call->service_id);
348 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
350 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
352 iov[0].iov_base = &whdr;
353 iov[0].iov_len = sizeof(whdr);
354 iov[1].iov_base = skb->head;
355 iov[1].iov_len = skb->len;
356 len = iov[0].iov_len + iov[1].iov_len;
358 msg.msg_name = &call->peer->srx.transport;
359 msg.msg_namelen = call->peer->srx.transport_len;
360 msg.msg_control = NULL;
361 msg.msg_controllen = 0;
364 /* If our RTT cache needs working on, request an ACK. Also request
365 * ACKs if a DATA packet appears to have been lost.
367 if (!(sp->hdr.flags & RXRPC_LAST_PACKET) &&
368 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
370 call->cong_mode == RXRPC_CALL_SLOW_START ||
371 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
372 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
374 whdr.flags |= RXRPC_REQUEST_ACK;
376 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
378 if ((lose++ & 7) == 7) {
385 _proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
387 /* send the packet with the don't fragment bit set if we currently
388 * think it's small enough */
389 if (iov[1].iov_len >= call->peer->maxdata)
390 goto send_fragmentable;
392 down_read(&conn->params.local->defrag_sem);
393 /* send the packet by UDP
394 * - returns -EMSGSIZE if UDP would have to fragment the packet
395 * to go out of the interface
396 * - in which case, we'll have processed the ICMP error
397 * message and update the peer record
399 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
400 conn->params.peer->last_tx_at = ktime_get_seconds();
402 up_read(&conn->params.local->defrag_sem);
404 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
405 rxrpc_tx_point_call_data_nofrag);
407 trace_rxrpc_tx_packet(call->debug_id, &whdr,
408 rxrpc_tx_point_call_data_nofrag);
409 if (ret == -EMSGSIZE)
410 goto send_fragmentable;
413 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
416 ktime_t now = ktime_get_real();
419 sp->hdr.serial = serial;
420 if (whdr.flags & RXRPC_REQUEST_ACK) {
421 call->peer->rtt_last_req = now;
422 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
423 if (call->peer->rtt_usage > 1) {
424 unsigned long nowj = jiffies, ack_lost_at;
426 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
431 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
432 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
433 rxrpc_timer_set_for_lost_ack);
437 if (sp->hdr.seq == 1 &&
438 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
440 unsigned long nowj = jiffies, expect_rx_by;
442 expect_rx_by = nowj + call->next_rx_timo;
443 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
444 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
445 rxrpc_timer_set_for_normal);
449 rxrpc_set_keepalive(call);
451 _leave(" = %d [%u]", ret, call->peer->maxdata);
455 /* attempt to send this message with fragmentation enabled */
456 _debug("send fragment");
458 down_write(&conn->params.local->defrag_sem);
460 switch (conn->params.local->srx.transport.family) {
462 opt = IP_PMTUDISC_DONT;
463 ret = kernel_setsockopt(conn->params.local->socket,
464 SOL_IP, IP_MTU_DISCOVER,
465 (char *)&opt, sizeof(opt));
467 ret = kernel_sendmsg(conn->params.local->socket, &msg,
469 conn->params.peer->last_tx_at = ktime_get_seconds();
471 opt = IP_PMTUDISC_DO;
472 kernel_setsockopt(conn->params.local->socket, SOL_IP,
474 (char *)&opt, sizeof(opt));
478 #ifdef CONFIG_AF_RXRPC_IPV6
480 opt = IPV6_PMTUDISC_DONT;
481 ret = kernel_setsockopt(conn->params.local->socket,
482 SOL_IPV6, IPV6_MTU_DISCOVER,
483 (char *)&opt, sizeof(opt));
485 ret = kernel_sendmsg(conn->params.local->socket, &msg,
487 conn->params.peer->last_tx_at = ktime_get_seconds();
489 opt = IPV6_PMTUDISC_DO;
490 kernel_setsockopt(conn->params.local->socket,
491 SOL_IPV6, IPV6_MTU_DISCOVER,
492 (char *)&opt, sizeof(opt));
499 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
500 rxrpc_tx_point_call_data_frag);
502 trace_rxrpc_tx_packet(call->debug_id, &whdr,
503 rxrpc_tx_point_call_data_frag);
505 up_write(&conn->params.local->defrag_sem);
510 * reject packets through the local endpoint
512 void rxrpc_reject_packets(struct rxrpc_local *local)
514 struct sockaddr_rxrpc srx;
515 struct rxrpc_skb_priv *sp;
516 struct rxrpc_wire_header whdr;
524 _enter("%d", local->debug_id);
526 iov[0].iov_base = &whdr;
527 iov[0].iov_len = sizeof(whdr);
528 iov[1].iov_base = &code;
529 iov[1].iov_len = sizeof(code);
530 size = sizeof(whdr) + sizeof(code);
532 msg.msg_name = &srx.transport;
533 msg.msg_control = NULL;
534 msg.msg_controllen = 0;
537 memset(&whdr, 0, sizeof(whdr));
538 whdr.type = RXRPC_PACKET_TYPE_ABORT;
540 while ((skb = skb_dequeue(&local->reject_queue))) {
541 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
544 if (rxrpc_extract_addr_from_skb(local, &srx, skb) == 0) {
545 msg.msg_namelen = srx.transport_len;
547 code = htonl(skb->priority);
549 whdr.epoch = htonl(sp->hdr.epoch);
550 whdr.cid = htonl(sp->hdr.cid);
551 whdr.callNumber = htonl(sp->hdr.callNumber);
552 whdr.serviceId = htons(sp->hdr.serviceId);
553 whdr.flags = sp->hdr.flags;
554 whdr.flags ^= RXRPC_CLIENT_INITIATED;
555 whdr.flags &= RXRPC_CLIENT_INITIATED;
557 ret = kernel_sendmsg(local->socket, &msg, iov, 2, size);
559 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
560 rxrpc_tx_point_reject);
562 trace_rxrpc_tx_packet(local->debug_id, &whdr,
563 rxrpc_tx_point_reject);
566 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
573 * Send a VERSION reply to a peer as a keepalive.
575 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
577 struct rxrpc_wire_header whdr;
585 msg.msg_name = &peer->srx.transport;
586 msg.msg_namelen = peer->srx.transport_len;
587 msg.msg_control = NULL;
588 msg.msg_controllen = 0;
591 whdr.epoch = htonl(peer->local->rxnet->epoch);
596 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
597 whdr.flags = RXRPC_LAST_PACKET;
599 whdr.securityIndex = 0;
603 iov[0].iov_base = &whdr;
604 iov[0].iov_len = sizeof(whdr);
605 iov[1].iov_base = (char *)rxrpc_keepalive_string;
606 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
608 len = iov[0].iov_len + iov[1].iov_len;
610 _proto("Tx VERSION (keepalive)");
612 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
614 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
615 rxrpc_tx_point_version_keepalive);
617 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
618 rxrpc_tx_point_version_keepalive);
620 peer->last_tx_at = ktime_get_seconds();