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 = NULL;
137 struct rxrpc_ack_buffer *pkt;
140 rxrpc_serial_t serial;
141 rxrpc_seq_t hard_ack, top;
146 spin_lock_bh(&call->lock);
148 conn = rxrpc_get_connection_maybe(call->conn);
149 spin_unlock_bh(&call->lock);
153 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
155 rxrpc_put_connection(conn);
159 msg.msg_name = &call->peer->srx.transport;
160 msg.msg_namelen = call->peer->srx.transport_len;
161 msg.msg_control = NULL;
162 msg.msg_controllen = 0;
165 pkt->whdr.epoch = htonl(conn->proto.epoch);
166 pkt->whdr.cid = htonl(call->cid);
167 pkt->whdr.callNumber = htonl(call->call_id);
169 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
170 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
171 pkt->whdr.userStatus = 0;
172 pkt->whdr.securityIndex = call->security_ix;
174 pkt->whdr.serviceId = htons(call->service_id);
176 spin_lock_bh(&call->lock);
178 reason = RXRPC_ACK_PING;
180 reason = call->ackr_reason;
181 if (!call->ackr_reason) {
182 spin_unlock_bh(&call->lock);
186 call->ackr_reason = 0;
188 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
190 spin_unlock_bh(&call->lock);
192 iov[0].iov_base = pkt;
193 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
194 iov[1].iov_base = &pkt->ackinfo;
195 iov[1].iov_len = sizeof(pkt->ackinfo);
196 len = iov[0].iov_len + iov[1].iov_len;
198 serial = atomic_inc_return(&conn->serial);
199 pkt->whdr.serial = htonl(serial);
200 trace_rxrpc_tx_ack(call->debug_id, serial,
201 ntohl(pkt->ack.firstPacket),
202 ntohl(pkt->ack.serial),
203 pkt->ack.reason, pkt->ack.nAcks);
208 call->ping_serial = serial;
210 /* We need to stick a time in before we send the packet in case
211 * the reply gets back before kernel_sendmsg() completes - but
212 * asking UDP to send the packet can take a relatively long
215 call->ping_time = ktime_get_real();
216 set_bit(RXRPC_CALL_PINGING, &call->flags);
217 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
220 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
221 conn->params.peer->last_tx_at = ktime_get_seconds();
223 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
224 rxrpc_tx_point_call_ack);
226 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
227 rxrpc_tx_point_call_ack);
228 rxrpc_tx_backoff(call, ret);
230 if (call->state < RXRPC_CALL_COMPLETE) {
233 clear_bit(RXRPC_CALL_PINGING, &call->flags);
234 rxrpc_propose_ACK(call, pkt->ack.reason,
235 ntohs(pkt->ack.maxSkew),
236 ntohl(pkt->ack.serial),
238 rxrpc_propose_ack_retry_tx);
240 spin_lock_bh(&call->lock);
241 if (after(hard_ack, call->ackr_consumed))
242 call->ackr_consumed = hard_ack;
243 if (after(top, call->ackr_seen))
244 call->ackr_seen = top;
245 spin_unlock_bh(&call->lock);
248 rxrpc_set_keepalive(call);
252 rxrpc_put_connection(conn);
258 * Send an ABORT call packet.
260 int rxrpc_send_abort_packet(struct rxrpc_call *call)
262 struct rxrpc_connection *conn = NULL;
263 struct rxrpc_abort_buffer pkt;
266 rxrpc_serial_t serial;
269 /* Don't bother sending aborts for a client call once the server has
270 * hard-ACK'd all of its request data. After that point, we're not
271 * going to stop the operation proceeding, and whilst we might limit
272 * the reply, it's not worth it if we can send a new call on the same
273 * channel instead, thereby closing off this call.
275 if (rxrpc_is_client_call(call) &&
276 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
279 spin_lock_bh(&call->lock);
281 conn = rxrpc_get_connection_maybe(call->conn);
282 spin_unlock_bh(&call->lock);
286 msg.msg_name = &call->peer->srx.transport;
287 msg.msg_namelen = call->peer->srx.transport_len;
288 msg.msg_control = NULL;
289 msg.msg_controllen = 0;
292 pkt.whdr.epoch = htonl(conn->proto.epoch);
293 pkt.whdr.cid = htonl(call->cid);
294 pkt.whdr.callNumber = htonl(call->call_id);
296 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
297 pkt.whdr.flags = conn->out_clientflag;
298 pkt.whdr.userStatus = 0;
299 pkt.whdr.securityIndex = call->security_ix;
301 pkt.whdr.serviceId = htons(call->service_id);
302 pkt.abort_code = htonl(call->abort_code);
304 iov[0].iov_base = &pkt;
305 iov[0].iov_len = sizeof(pkt);
307 serial = atomic_inc_return(&conn->serial);
308 pkt.whdr.serial = htonl(serial);
310 ret = kernel_sendmsg(conn->params.local->socket,
311 &msg, iov, 1, sizeof(pkt));
312 conn->params.peer->last_tx_at = ktime_get_seconds();
314 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
315 rxrpc_tx_point_call_abort);
317 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
318 rxrpc_tx_point_call_abort);
319 rxrpc_tx_backoff(call, ret);
321 rxrpc_put_connection(conn);
326 * send a packet through the transport endpoint
328 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
331 struct rxrpc_connection *conn = call->conn;
332 struct rxrpc_wire_header whdr;
333 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
336 rxrpc_serial_t serial;
340 _enter(",{%d}", skb->len);
342 /* Each transmission of a Tx packet needs a new serial number */
343 serial = atomic_inc_return(&conn->serial);
345 whdr.epoch = htonl(conn->proto.epoch);
346 whdr.cid = htonl(call->cid);
347 whdr.callNumber = htonl(call->call_id);
348 whdr.seq = htonl(sp->hdr.seq);
349 whdr.serial = htonl(serial);
350 whdr.type = RXRPC_PACKET_TYPE_DATA;
351 whdr.flags = sp->hdr.flags;
353 whdr.securityIndex = call->security_ix;
354 whdr._rsvd = htons(sp->hdr._rsvd);
355 whdr.serviceId = htons(call->service_id);
357 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
359 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
361 iov[0].iov_base = &whdr;
362 iov[0].iov_len = sizeof(whdr);
363 iov[1].iov_base = skb->head;
364 iov[1].iov_len = skb->len;
365 len = iov[0].iov_len + iov[1].iov_len;
367 msg.msg_name = &call->peer->srx.transport;
368 msg.msg_namelen = call->peer->srx.transport_len;
369 msg.msg_control = NULL;
370 msg.msg_controllen = 0;
373 /* If our RTT cache needs working on, request an ACK. Also request
374 * ACKs if a DATA packet appears to have been lost.
376 * However, we mustn't request an ACK on the last reply packet of a
377 * service call, lest OpenAFS incorrectly send us an ACK with some
378 * soft-ACKs in it and then never follow up with a proper hard ACK.
380 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
383 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
385 call->cong_mode == RXRPC_CALL_SLOW_START ||
386 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
387 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
389 whdr.flags |= RXRPC_REQUEST_ACK;
391 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
393 if ((lose++ & 7) == 7) {
395 trace_rxrpc_tx_data(call, sp->hdr.seq, serial,
396 whdr.flags, retrans, true);
401 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags, retrans,
404 /* send the packet with the don't fragment bit set if we currently
405 * think it's small enough */
406 if (iov[1].iov_len >= call->peer->maxdata)
407 goto send_fragmentable;
409 down_read(&conn->params.local->defrag_sem);
411 sp->hdr.serial = serial;
412 smp_wmb(); /* Set serial before timestamp */
413 skb->tstamp = ktime_get_real();
415 /* send the packet by UDP
416 * - returns -EMSGSIZE if UDP would have to fragment the packet
417 * to go out of the interface
418 * - in which case, we'll have processed the ICMP error
419 * message and update the peer record
421 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
422 conn->params.peer->last_tx_at = ktime_get_seconds();
424 up_read(&conn->params.local->defrag_sem);
426 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
427 rxrpc_tx_point_call_data_nofrag);
429 trace_rxrpc_tx_packet(call->debug_id, &whdr,
430 rxrpc_tx_point_call_data_nofrag);
431 rxrpc_tx_backoff(call, ret);
432 if (ret == -EMSGSIZE)
433 goto send_fragmentable;
437 if (whdr.flags & RXRPC_REQUEST_ACK) {
438 call->peer->rtt_last_req = skb->tstamp;
439 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
440 if (call->peer->rtt_usage > 1) {
441 unsigned long nowj = jiffies, ack_lost_at;
443 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
448 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
449 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
450 rxrpc_timer_set_for_lost_ack);
454 if (sp->hdr.seq == 1 &&
455 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
457 unsigned long nowj = jiffies, expect_rx_by;
459 expect_rx_by = nowj + call->next_rx_timo;
460 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
461 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
462 rxrpc_timer_set_for_normal);
465 rxrpc_set_keepalive(call);
467 /* Cancel the call if the initial transmission fails,
468 * particularly if that's due to network routing issues that
469 * aren't going away anytime soon. The layer above can arrange
470 * the retransmission.
472 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
473 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
477 _leave(" = %d [%u]", ret, call->peer->maxdata);
481 /* attempt to send this message with fragmentation enabled */
482 _debug("send fragment");
484 down_write(&conn->params.local->defrag_sem);
486 sp->hdr.serial = serial;
487 smp_wmb(); /* Set serial before timestamp */
488 skb->tstamp = ktime_get_real();
490 switch (conn->params.local->srx.transport.family) {
492 opt = IP_PMTUDISC_DONT;
493 ret = kernel_setsockopt(conn->params.local->socket,
494 SOL_IP, IP_MTU_DISCOVER,
495 (char *)&opt, sizeof(opt));
497 ret = kernel_sendmsg(conn->params.local->socket, &msg,
499 conn->params.peer->last_tx_at = ktime_get_seconds();
501 opt = IP_PMTUDISC_DO;
502 kernel_setsockopt(conn->params.local->socket, SOL_IP,
504 (char *)&opt, sizeof(opt));
508 #ifdef CONFIG_AF_RXRPC_IPV6
510 opt = IPV6_PMTUDISC_DONT;
511 ret = kernel_setsockopt(conn->params.local->socket,
512 SOL_IPV6, IPV6_MTU_DISCOVER,
513 (char *)&opt, sizeof(opt));
515 ret = kernel_sendmsg(conn->params.local->socket, &msg,
517 conn->params.peer->last_tx_at = ktime_get_seconds();
519 opt = IPV6_PMTUDISC_DO;
520 kernel_setsockopt(conn->params.local->socket,
521 SOL_IPV6, IPV6_MTU_DISCOVER,
522 (char *)&opt, sizeof(opt));
529 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
530 rxrpc_tx_point_call_data_frag);
532 trace_rxrpc_tx_packet(call->debug_id, &whdr,
533 rxrpc_tx_point_call_data_frag);
534 rxrpc_tx_backoff(call, ret);
536 up_write(&conn->params.local->defrag_sem);
541 * reject packets through the local endpoint
543 void rxrpc_reject_packets(struct rxrpc_local *local)
545 struct sockaddr_rxrpc srx;
546 struct rxrpc_skb_priv *sp;
547 struct rxrpc_wire_header whdr;
555 _enter("%d", local->debug_id);
557 iov[0].iov_base = &whdr;
558 iov[0].iov_len = sizeof(whdr);
559 iov[1].iov_base = &code;
560 iov[1].iov_len = sizeof(code);
562 msg.msg_name = &srx.transport;
563 msg.msg_control = NULL;
564 msg.msg_controllen = 0;
567 memset(&whdr, 0, sizeof(whdr));
569 while ((skb = skb_dequeue(&local->reject_queue))) {
570 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
574 case RXRPC_SKB_MARK_REJECT_BUSY:
575 whdr.type = RXRPC_PACKET_TYPE_BUSY;
579 case RXRPC_SKB_MARK_REJECT_ABORT:
580 whdr.type = RXRPC_PACKET_TYPE_ABORT;
581 code = htonl(skb->priority);
582 size = sizeof(whdr) + sizeof(code);
586 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
590 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
591 msg.msg_namelen = srx.transport_len;
593 whdr.epoch = htonl(sp->hdr.epoch);
594 whdr.cid = htonl(sp->hdr.cid);
595 whdr.callNumber = htonl(sp->hdr.callNumber);
596 whdr.serviceId = htons(sp->hdr.serviceId);
597 whdr.flags = sp->hdr.flags;
598 whdr.flags ^= RXRPC_CLIENT_INITIATED;
599 whdr.flags &= RXRPC_CLIENT_INITIATED;
601 ret = kernel_sendmsg(local->socket, &msg,
604 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
605 rxrpc_tx_point_reject);
607 trace_rxrpc_tx_packet(local->debug_id, &whdr,
608 rxrpc_tx_point_reject);
611 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
618 * Send a VERSION reply to a peer as a keepalive.
620 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
622 struct rxrpc_wire_header whdr;
630 msg.msg_name = &peer->srx.transport;
631 msg.msg_namelen = peer->srx.transport_len;
632 msg.msg_control = NULL;
633 msg.msg_controllen = 0;
636 whdr.epoch = htonl(peer->local->rxnet->epoch);
641 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
642 whdr.flags = RXRPC_LAST_PACKET;
644 whdr.securityIndex = 0;
648 iov[0].iov_base = &whdr;
649 iov[0].iov_len = sizeof(whdr);
650 iov[1].iov_base = (char *)rxrpc_keepalive_string;
651 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
653 len = iov[0].iov_len + iov[1].iov_len;
655 _proto("Tx VERSION (keepalive)");
657 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
659 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
660 rxrpc_tx_point_version_keepalive);
662 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
663 rxrpc_tx_point_version_keepalive);
665 peer->last_tx_at = ktime_get_seconds();