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 * Increase Tx backoff on transmission failure and clear it on success.
40 static void rxrpc_tx_backoff(struct rxrpc_call *call, int ret)
43 u16 tx_backoff = READ_ONCE(call->tx_backoff);
46 WRITE_ONCE(call->tx_backoff, tx_backoff + 1);
48 WRITE_ONCE(call->tx_backoff, 0);
53 * Arrange for a keepalive ping a certain time after we last transmitted. This
54 * lets the far side know we're still interested in this call and helps keep
55 * the route through any intervening firewall open.
57 * Receiving a response to the ping will prevent the ->expect_rx_by timer from
60 static void rxrpc_set_keepalive(struct rxrpc_call *call)
62 unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
65 WRITE_ONCE(call->keepalive_at, keepalive_at);
66 rxrpc_reduce_call_timer(call, keepalive_at, now,
67 rxrpc_timer_set_for_keepalive);
71 * Fill out an ACK packet.
73 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
74 struct rxrpc_call *call,
75 struct rxrpc_ack_buffer *pkt,
76 rxrpc_seq_t *_hard_ack,
80 rxrpc_serial_t serial;
81 rxrpc_seq_t hard_ack, top, seq;
86 /* Barrier against rxrpc_input_data(). */
87 serial = call->ackr_serial;
88 hard_ack = READ_ONCE(call->rx_hard_ack);
89 top = smp_load_acquire(&call->rx_top);
90 *_hard_ack = hard_ack;
93 pkt->ack.bufferSpace = htons(8);
94 pkt->ack.maxSkew = htons(call->ackr_skew);
95 pkt->ack.firstPacket = htonl(hard_ack + 1);
96 pkt->ack.previousPacket = htonl(call->ackr_prev_seq);
97 pkt->ack.serial = htonl(serial);
98 pkt->ack.reason = reason;
99 pkt->ack.nAcks = top - hard_ack;
101 if (reason == RXRPC_ACK_PING)
102 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
104 if (after(top, hard_ack)) {
107 ix = seq & RXRPC_RXTX_BUFF_MASK;
108 if (call->rxtx_buffer[ix])
109 *ackp++ = RXRPC_ACK_TYPE_ACK;
111 *ackp++ = RXRPC_ACK_TYPE_NACK;
113 } while (before_eq(seq, top));
116 mtu = conn->params.peer->if_mtu;
117 mtu -= conn->params.peer->hdrsize;
118 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
119 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
120 pkt->ackinfo.maxMTU = htonl(mtu);
121 pkt->ackinfo.rwind = htonl(call->rx_winsize);
122 pkt->ackinfo.jumbo_max = htonl(jmax);
127 return top - hard_ack + 3;
131 * Send an ACK call packet.
133 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
134 rxrpc_serial_t *_serial)
136 struct rxrpc_connection *conn;
137 struct rxrpc_ack_buffer *pkt;
140 rxrpc_serial_t serial;
141 rxrpc_seq_t hard_ack, top;
146 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
149 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
155 msg.msg_name = &call->peer->srx.transport;
156 msg.msg_namelen = call->peer->srx.transport_len;
157 msg.msg_control = NULL;
158 msg.msg_controllen = 0;
161 pkt->whdr.epoch = htonl(conn->proto.epoch);
162 pkt->whdr.cid = htonl(call->cid);
163 pkt->whdr.callNumber = htonl(call->call_id);
165 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
166 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
167 pkt->whdr.userStatus = 0;
168 pkt->whdr.securityIndex = call->security_ix;
170 pkt->whdr.serviceId = htons(call->service_id);
172 spin_lock_bh(&call->lock);
174 reason = RXRPC_ACK_PING;
176 reason = call->ackr_reason;
177 if (!call->ackr_reason) {
178 spin_unlock_bh(&call->lock);
182 call->ackr_reason = 0;
184 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
186 spin_unlock_bh(&call->lock);
188 iov[0].iov_base = pkt;
189 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
190 iov[1].iov_base = &pkt->ackinfo;
191 iov[1].iov_len = sizeof(pkt->ackinfo);
192 len = iov[0].iov_len + iov[1].iov_len;
194 serial = atomic_inc_return(&conn->serial);
195 pkt->whdr.serial = htonl(serial);
196 trace_rxrpc_tx_ack(call->debug_id, serial,
197 ntohl(pkt->ack.firstPacket),
198 ntohl(pkt->ack.serial),
199 pkt->ack.reason, pkt->ack.nAcks);
204 call->ping_serial = serial;
206 /* We need to stick a time in before we send the packet in case
207 * the reply gets back before kernel_sendmsg() completes - but
208 * asking UDP to send the packet can take a relatively long
211 call->ping_time = ktime_get_real();
212 set_bit(RXRPC_CALL_PINGING, &call->flags);
213 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
216 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
217 conn->params.peer->last_tx_at = ktime_get_seconds();
219 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
220 rxrpc_tx_point_call_ack);
222 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
223 rxrpc_tx_point_call_ack);
224 rxrpc_tx_backoff(call, ret);
226 if (call->state < RXRPC_CALL_COMPLETE) {
229 clear_bit(RXRPC_CALL_PINGING, &call->flags);
230 rxrpc_propose_ACK(call, pkt->ack.reason,
231 ntohs(pkt->ack.maxSkew),
232 ntohl(pkt->ack.serial),
234 rxrpc_propose_ack_retry_tx);
236 spin_lock_bh(&call->lock);
237 if (after(hard_ack, call->ackr_consumed))
238 call->ackr_consumed = hard_ack;
239 if (after(top, call->ackr_seen))
240 call->ackr_seen = top;
241 spin_unlock_bh(&call->lock);
244 rxrpc_set_keepalive(call);
253 * Send an ABORT call packet.
255 int rxrpc_send_abort_packet(struct rxrpc_call *call)
257 struct rxrpc_connection *conn;
258 struct rxrpc_abort_buffer pkt;
261 rxrpc_serial_t serial;
264 /* Don't bother sending aborts for a client call once the server has
265 * hard-ACK'd all of its request data. After that point, we're not
266 * going to stop the operation proceeding, and whilst we might limit
267 * the reply, it's not worth it if we can send a new call on the same
268 * channel instead, thereby closing off this call.
270 if (rxrpc_is_client_call(call) &&
271 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
274 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
279 msg.msg_name = &call->peer->srx.transport;
280 msg.msg_namelen = call->peer->srx.transport_len;
281 msg.msg_control = NULL;
282 msg.msg_controllen = 0;
285 pkt.whdr.epoch = htonl(conn->proto.epoch);
286 pkt.whdr.cid = htonl(call->cid);
287 pkt.whdr.callNumber = htonl(call->call_id);
289 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
290 pkt.whdr.flags = conn->out_clientflag;
291 pkt.whdr.userStatus = 0;
292 pkt.whdr.securityIndex = call->security_ix;
294 pkt.whdr.serviceId = htons(call->service_id);
295 pkt.abort_code = htonl(call->abort_code);
297 iov[0].iov_base = &pkt;
298 iov[0].iov_len = sizeof(pkt);
300 serial = atomic_inc_return(&conn->serial);
301 pkt.whdr.serial = htonl(serial);
303 ret = kernel_sendmsg(conn->params.local->socket,
304 &msg, iov, 1, sizeof(pkt));
305 conn->params.peer->last_tx_at = ktime_get_seconds();
307 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
308 rxrpc_tx_point_call_abort);
310 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
311 rxrpc_tx_point_call_abort);
312 rxrpc_tx_backoff(call, ret);
317 * send a packet through the transport endpoint
319 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
322 struct rxrpc_connection *conn = call->conn;
323 struct rxrpc_wire_header whdr;
324 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
327 rxrpc_serial_t serial;
332 _enter(",{%d}", skb->len);
334 /* Each transmission of a Tx packet needs a new serial number */
335 serial = atomic_inc_return(&conn->serial);
337 whdr.epoch = htonl(conn->proto.epoch);
338 whdr.cid = htonl(call->cid);
339 whdr.callNumber = htonl(call->call_id);
340 whdr.seq = htonl(sp->hdr.seq);
341 whdr.serial = htonl(serial);
342 whdr.type = RXRPC_PACKET_TYPE_DATA;
343 whdr.flags = sp->hdr.flags;
345 whdr.securityIndex = call->security_ix;
346 whdr._rsvd = htons(sp->hdr._rsvd);
347 whdr.serviceId = htons(call->service_id);
349 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
351 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
353 iov[0].iov_base = &whdr;
354 iov[0].iov_len = sizeof(whdr);
355 iov[1].iov_base = skb->head;
356 iov[1].iov_len = skb->len;
357 len = iov[0].iov_len + iov[1].iov_len;
359 msg.msg_name = &call->peer->srx.transport;
360 msg.msg_namelen = call->peer->srx.transport_len;
361 msg.msg_control = NULL;
362 msg.msg_controllen = 0;
365 /* If our RTT cache needs working on, request an ACK. Also request
366 * ACKs if a DATA packet appears to have been lost.
368 * However, we mustn't request an ACK on the last reply packet of a
369 * service call, lest OpenAFS incorrectly send us an ACK with some
370 * soft-ACKs in it and then never follow up with a proper hard ACK.
372 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
375 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
377 call->cong_mode == RXRPC_CALL_SLOW_START ||
378 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
379 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
381 whdr.flags |= RXRPC_REQUEST_ACK;
383 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
385 if ((lose++ & 7) == 7) {
392 _proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
394 /* send the packet with the don't fragment bit set if we currently
395 * think it's small enough */
396 if (iov[1].iov_len >= call->peer->maxdata)
397 goto send_fragmentable;
399 down_read(&conn->params.local->defrag_sem);
401 sp->hdr.serial = serial;
402 smp_wmb(); /* Set serial before timestamp */
403 skb->tstamp = ktime_get_real();
405 /* send the packet by UDP
406 * - returns -EMSGSIZE if UDP would have to fragment the packet
407 * to go out of the interface
408 * - in which case, we'll have processed the ICMP error
409 * message and update the peer record
411 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
412 conn->params.peer->last_tx_at = ktime_get_seconds();
414 up_read(&conn->params.local->defrag_sem);
416 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
417 rxrpc_tx_point_call_data_nofrag);
419 trace_rxrpc_tx_packet(call->debug_id, &whdr,
420 rxrpc_tx_point_call_data_nofrag);
421 rxrpc_tx_backoff(call, ret);
422 if (ret == -EMSGSIZE)
423 goto send_fragmentable;
426 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
429 if (whdr.flags & RXRPC_REQUEST_ACK) {
430 call->peer->rtt_last_req = skb->tstamp;
431 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
432 if (call->peer->rtt_usage > 1) {
433 unsigned long nowj = jiffies, ack_lost_at;
435 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
440 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
441 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
442 rxrpc_timer_set_for_lost_ack);
446 if (sp->hdr.seq == 1 &&
447 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
449 unsigned long nowj = jiffies, expect_rx_by;
451 expect_rx_by = nowj + call->next_rx_timo;
452 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
453 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
454 rxrpc_timer_set_for_normal);
457 rxrpc_set_keepalive(call);
459 /* Cancel the call if the initial transmission fails,
460 * particularly if that's due to network routing issues that
461 * aren't going away anytime soon. The layer above can arrange
462 * the retransmission.
464 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
465 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
469 _leave(" = %d [%u]", ret, call->peer->maxdata);
473 /* attempt to send this message with fragmentation enabled */
474 _debug("send fragment");
476 down_write(&conn->params.local->defrag_sem);
478 sp->hdr.serial = serial;
479 smp_wmb(); /* Set serial before timestamp */
480 skb->tstamp = ktime_get_real();
482 switch (conn->params.local->srx.transport.family) {
485 opt = IP_PMTUDISC_DONT;
486 kernel_setsockopt(conn->params.local->socket,
487 SOL_IP, IP_MTU_DISCOVER,
488 (char *)&opt, sizeof(opt));
489 ret = kernel_sendmsg(conn->params.local->socket, &msg,
491 conn->params.peer->last_tx_at = ktime_get_seconds();
493 opt = IP_PMTUDISC_DO;
494 kernel_setsockopt(conn->params.local->socket,
495 SOL_IP, IP_MTU_DISCOVER,
496 (char *)&opt, sizeof(opt));
504 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
505 rxrpc_tx_point_call_data_frag);
507 trace_rxrpc_tx_packet(call->debug_id, &whdr,
508 rxrpc_tx_point_call_data_frag);
509 rxrpc_tx_backoff(call, ret);
511 up_write(&conn->params.local->defrag_sem);
516 * reject packets through the local endpoint
518 void rxrpc_reject_packets(struct rxrpc_local *local)
520 struct sockaddr_rxrpc srx;
521 struct rxrpc_skb_priv *sp;
522 struct rxrpc_wire_header whdr;
530 _enter("%d", local->debug_id);
532 iov[0].iov_base = &whdr;
533 iov[0].iov_len = sizeof(whdr);
534 iov[1].iov_base = &code;
535 iov[1].iov_len = sizeof(code);
537 msg.msg_name = &srx.transport;
538 msg.msg_control = NULL;
539 msg.msg_controllen = 0;
542 memset(&whdr, 0, sizeof(whdr));
544 while ((skb = skb_dequeue(&local->reject_queue))) {
545 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
549 case RXRPC_SKB_MARK_REJECT_BUSY:
550 whdr.type = RXRPC_PACKET_TYPE_BUSY;
554 case RXRPC_SKB_MARK_REJECT_ABORT:
555 whdr.type = RXRPC_PACKET_TYPE_ABORT;
556 code = htonl(skb->priority);
557 size = sizeof(whdr) + sizeof(code);
561 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
565 if (rxrpc_extract_addr_from_skb(local, &srx, skb) == 0) {
566 msg.msg_namelen = srx.transport_len;
568 whdr.epoch = htonl(sp->hdr.epoch);
569 whdr.cid = htonl(sp->hdr.cid);
570 whdr.callNumber = htonl(sp->hdr.callNumber);
571 whdr.serviceId = htons(sp->hdr.serviceId);
572 whdr.flags = sp->hdr.flags;
573 whdr.flags ^= RXRPC_CLIENT_INITIATED;
574 whdr.flags &= RXRPC_CLIENT_INITIATED;
576 ret = kernel_sendmsg(local->socket, &msg,
579 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
580 rxrpc_tx_point_reject);
582 trace_rxrpc_tx_packet(local->debug_id, &whdr,
583 rxrpc_tx_point_reject);
586 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
593 * Send a VERSION reply to a peer as a keepalive.
595 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
597 struct rxrpc_wire_header whdr;
605 msg.msg_name = &peer->srx.transport;
606 msg.msg_namelen = peer->srx.transport_len;
607 msg.msg_control = NULL;
608 msg.msg_controllen = 0;
611 whdr.epoch = htonl(peer->local->rxnet->epoch);
616 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
617 whdr.flags = RXRPC_LAST_PACKET;
619 whdr.securityIndex = 0;
623 iov[0].iov_base = &whdr;
624 iov[0].iov_len = sizeof(whdr);
625 iov[1].iov_base = (char *)rxrpc_keepalive_string;
626 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
628 len = iov[0].iov_len + iov[1].iov_len;
630 _proto("Tx VERSION (keepalive)");
632 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
634 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
635 rxrpc_tx_point_version_keepalive);
637 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
638 rxrpc_tx_point_version_keepalive);
640 peer->last_tx_at = ktime_get_seconds();