1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC packet transmission
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/net.h>
11 #include <linux/gfp.h>
12 #include <linux/skbuff.h>
13 #include <linux/export.h>
15 #include <net/af_rxrpc.h>
16 #include "ar-internal.h"
18 struct rxrpc_ack_buffer {
19 struct rxrpc_wire_header whdr;
20 struct rxrpc_ackpacket ack;
23 struct rxrpc_ackinfo ackinfo;
26 struct rxrpc_abort_buffer {
27 struct rxrpc_wire_header whdr;
31 static const char rxrpc_keepalive_string[] = "";
34 * Increase Tx backoff on transmission failure and clear it on success.
36 static void rxrpc_tx_backoff(struct rxrpc_call *call, int ret)
39 u16 tx_backoff = READ_ONCE(call->tx_backoff);
42 WRITE_ONCE(call->tx_backoff, tx_backoff + 1);
44 WRITE_ONCE(call->tx_backoff, 0);
49 * Arrange for a keepalive ping a certain time after we last transmitted. This
50 * lets the far side know we're still interested in this call and helps keep
51 * the route through any intervening firewall open.
53 * Receiving a response to the ping will prevent the ->expect_rx_by timer from
56 static void rxrpc_set_keepalive(struct rxrpc_call *call)
58 unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
61 WRITE_ONCE(call->keepalive_at, keepalive_at);
62 rxrpc_reduce_call_timer(call, keepalive_at, now,
63 rxrpc_timer_set_for_keepalive);
67 * Fill out an ACK packet.
69 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
70 struct rxrpc_call *call,
71 struct rxrpc_ack_buffer *pkt,
72 rxrpc_seq_t *_hard_ack,
76 rxrpc_serial_t serial;
77 rxrpc_seq_t hard_ack, top, seq;
82 /* Barrier against rxrpc_input_data(). */
83 serial = call->ackr_serial;
84 hard_ack = READ_ONCE(call->rx_hard_ack);
85 top = smp_load_acquire(&call->rx_top);
86 *_hard_ack = hard_ack;
89 pkt->ack.bufferSpace = htons(8);
90 pkt->ack.maxSkew = htons(call->ackr_skew);
91 pkt->ack.firstPacket = htonl(hard_ack + 1);
92 pkt->ack.previousPacket = htonl(call->ackr_prev_seq);
93 pkt->ack.serial = htonl(serial);
94 pkt->ack.reason = reason;
95 pkt->ack.nAcks = top - hard_ack;
97 if (reason == RXRPC_ACK_PING)
98 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
100 if (after(top, hard_ack)) {
103 ix = seq & RXRPC_RXTX_BUFF_MASK;
104 if (call->rxtx_buffer[ix])
105 *ackp++ = RXRPC_ACK_TYPE_ACK;
107 *ackp++ = RXRPC_ACK_TYPE_NACK;
109 } while (before_eq(seq, top));
112 mtu = conn->params.peer->if_mtu;
113 mtu -= conn->params.peer->hdrsize;
114 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
115 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
116 pkt->ackinfo.maxMTU = htonl(mtu);
117 pkt->ackinfo.rwind = htonl(call->rx_winsize);
118 pkt->ackinfo.jumbo_max = htonl(jmax);
123 return top - hard_ack + 3;
127 * Send an ACK call packet.
129 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
130 rxrpc_serial_t *_serial)
132 struct rxrpc_connection *conn = NULL;
133 struct rxrpc_ack_buffer *pkt;
136 rxrpc_serial_t serial;
137 rxrpc_seq_t hard_ack, top;
142 spin_lock_bh(&call->lock);
144 conn = rxrpc_get_connection_maybe(call->conn);
145 spin_unlock_bh(&call->lock);
149 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
151 rxrpc_put_connection(conn);
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);
248 rxrpc_put_connection(conn);
254 * Send an ABORT call packet.
256 int rxrpc_send_abort_packet(struct rxrpc_call *call)
258 struct rxrpc_connection *conn = NULL;
259 struct rxrpc_abort_buffer pkt;
262 rxrpc_serial_t serial;
265 /* Don't bother sending aborts for a client call once the server has
266 * hard-ACK'd all of its request data. After that point, we're not
267 * going to stop the operation proceeding, and whilst we might limit
268 * the reply, it's not worth it if we can send a new call on the same
269 * channel instead, thereby closing off this call.
271 if (rxrpc_is_client_call(call) &&
272 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
275 spin_lock_bh(&call->lock);
277 conn = rxrpc_get_connection_maybe(call->conn);
278 spin_unlock_bh(&call->lock);
282 msg.msg_name = &call->peer->srx.transport;
283 msg.msg_namelen = call->peer->srx.transport_len;
284 msg.msg_control = NULL;
285 msg.msg_controllen = 0;
288 pkt.whdr.epoch = htonl(conn->proto.epoch);
289 pkt.whdr.cid = htonl(call->cid);
290 pkt.whdr.callNumber = htonl(call->call_id);
292 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
293 pkt.whdr.flags = conn->out_clientflag;
294 pkt.whdr.userStatus = 0;
295 pkt.whdr.securityIndex = call->security_ix;
297 pkt.whdr.serviceId = htons(call->service_id);
298 pkt.abort_code = htonl(call->abort_code);
300 iov[0].iov_base = &pkt;
301 iov[0].iov_len = sizeof(pkt);
303 serial = atomic_inc_return(&conn->serial);
304 pkt.whdr.serial = htonl(serial);
306 ret = kernel_sendmsg(conn->params.local->socket,
307 &msg, iov, 1, sizeof(pkt));
308 conn->params.peer->last_tx_at = ktime_get_seconds();
310 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
311 rxrpc_tx_point_call_abort);
313 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
314 rxrpc_tx_point_call_abort);
315 rxrpc_tx_backoff(call, ret);
317 rxrpc_put_connection(conn);
322 * send a packet through the transport endpoint
324 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
327 struct rxrpc_connection *conn = call->conn;
328 struct rxrpc_wire_header whdr;
329 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
332 rxrpc_serial_t serial;
336 _enter(",{%d}", skb->len);
338 /* Each transmission of a Tx packet needs a new serial number */
339 serial = atomic_inc_return(&conn->serial);
341 whdr.epoch = htonl(conn->proto.epoch);
342 whdr.cid = htonl(call->cid);
343 whdr.callNumber = htonl(call->call_id);
344 whdr.seq = htonl(sp->hdr.seq);
345 whdr.serial = htonl(serial);
346 whdr.type = RXRPC_PACKET_TYPE_DATA;
347 whdr.flags = sp->hdr.flags;
349 whdr.securityIndex = call->security_ix;
350 whdr._rsvd = htons(sp->hdr._rsvd);
351 whdr.serviceId = htons(call->service_id);
353 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
355 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
357 iov[0].iov_base = &whdr;
358 iov[0].iov_len = sizeof(whdr);
359 iov[1].iov_base = skb->head;
360 iov[1].iov_len = skb->len;
361 len = iov[0].iov_len + iov[1].iov_len;
363 msg.msg_name = &call->peer->srx.transport;
364 msg.msg_namelen = call->peer->srx.transport_len;
365 msg.msg_control = NULL;
366 msg.msg_controllen = 0;
369 /* If our RTT cache needs working on, request an ACK. Also request
370 * ACKs if a DATA packet appears to have been lost.
372 * However, we mustn't request an ACK on the last reply packet of a
373 * service call, lest OpenAFS incorrectly send us an ACK with some
374 * soft-ACKs in it and then never follow up with a proper hard ACK.
376 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
379 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
381 call->cong_mode == RXRPC_CALL_SLOW_START ||
382 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
383 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
385 whdr.flags |= RXRPC_REQUEST_ACK;
387 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
389 if ((lose++ & 7) == 7) {
391 trace_rxrpc_tx_data(call, sp->hdr.seq, serial,
392 whdr.flags, retrans, true);
397 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags, retrans,
400 /* send the packet with the don't fragment bit set if we currently
401 * think it's small enough */
402 if (iov[1].iov_len >= call->peer->maxdata)
403 goto send_fragmentable;
405 down_read(&conn->params.local->defrag_sem);
407 sp->hdr.serial = serial;
408 smp_wmb(); /* Set serial before timestamp */
409 skb->tstamp = ktime_get_real();
411 /* send the packet by UDP
412 * - returns -EMSGSIZE if UDP would have to fragment the packet
413 * to go out of the interface
414 * - in which case, we'll have processed the ICMP error
415 * message and update the peer record
417 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
418 conn->params.peer->last_tx_at = ktime_get_seconds();
420 up_read(&conn->params.local->defrag_sem);
422 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
423 rxrpc_tx_point_call_data_nofrag);
425 trace_rxrpc_tx_packet(call->debug_id, &whdr,
426 rxrpc_tx_point_call_data_nofrag);
427 rxrpc_tx_backoff(call, ret);
428 if (ret == -EMSGSIZE)
429 goto send_fragmentable;
433 if (whdr.flags & RXRPC_REQUEST_ACK) {
434 call->peer->rtt_last_req = skb->tstamp;
435 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
436 if (call->peer->rtt_usage > 1) {
437 unsigned long nowj = jiffies, ack_lost_at;
439 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
444 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
445 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
446 rxrpc_timer_set_for_lost_ack);
450 if (sp->hdr.seq == 1 &&
451 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
453 unsigned long nowj = jiffies, expect_rx_by;
455 expect_rx_by = nowj + call->next_rx_timo;
456 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
457 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
458 rxrpc_timer_set_for_normal);
461 rxrpc_set_keepalive(call);
463 /* Cancel the call if the initial transmission fails,
464 * particularly if that's due to network routing issues that
465 * aren't going away anytime soon. The layer above can arrange
466 * the retransmission.
468 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
469 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
473 _leave(" = %d [%u]", ret, call->peer->maxdata);
477 /* attempt to send this message with fragmentation enabled */
478 _debug("send fragment");
480 down_write(&conn->params.local->defrag_sem);
482 sp->hdr.serial = serial;
483 smp_wmb(); /* Set serial before timestamp */
484 skb->tstamp = ktime_get_real();
486 switch (conn->params.local->srx.transport.family) {
488 opt = IP_PMTUDISC_DONT;
489 ret = kernel_setsockopt(conn->params.local->socket,
490 SOL_IP, IP_MTU_DISCOVER,
491 (char *)&opt, sizeof(opt));
493 ret = kernel_sendmsg(conn->params.local->socket, &msg,
495 conn->params.peer->last_tx_at = ktime_get_seconds();
497 opt = IP_PMTUDISC_DO;
498 kernel_setsockopt(conn->params.local->socket, SOL_IP,
500 (char *)&opt, sizeof(opt));
504 #ifdef CONFIG_AF_RXRPC_IPV6
506 opt = IPV6_PMTUDISC_DONT;
507 ret = kernel_setsockopt(conn->params.local->socket,
508 SOL_IPV6, IPV6_MTU_DISCOVER,
509 (char *)&opt, sizeof(opt));
511 ret = kernel_sendmsg(conn->params.local->socket, &msg,
513 conn->params.peer->last_tx_at = ktime_get_seconds();
515 opt = IPV6_PMTUDISC_DO;
516 kernel_setsockopt(conn->params.local->socket,
517 SOL_IPV6, IPV6_MTU_DISCOVER,
518 (char *)&opt, sizeof(opt));
528 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
529 rxrpc_tx_point_call_data_frag);
531 trace_rxrpc_tx_packet(call->debug_id, &whdr,
532 rxrpc_tx_point_call_data_frag);
533 rxrpc_tx_backoff(call, ret);
535 up_write(&conn->params.local->defrag_sem);
540 * reject packets through the local endpoint
542 void rxrpc_reject_packets(struct rxrpc_local *local)
544 struct sockaddr_rxrpc srx;
545 struct rxrpc_skb_priv *sp;
546 struct rxrpc_wire_header whdr;
554 _enter("%d", local->debug_id);
556 iov[0].iov_base = &whdr;
557 iov[0].iov_len = sizeof(whdr);
558 iov[1].iov_base = &code;
559 iov[1].iov_len = sizeof(code);
561 msg.msg_name = &srx.transport;
562 msg.msg_control = NULL;
563 msg.msg_controllen = 0;
566 memset(&whdr, 0, sizeof(whdr));
568 while ((skb = skb_dequeue(&local->reject_queue))) {
569 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
573 case RXRPC_SKB_MARK_REJECT_BUSY:
574 whdr.type = RXRPC_PACKET_TYPE_BUSY;
578 case RXRPC_SKB_MARK_REJECT_ABORT:
579 whdr.type = RXRPC_PACKET_TYPE_ABORT;
580 code = htonl(skb->priority);
581 size = sizeof(whdr) + sizeof(code);
585 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
589 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
590 msg.msg_namelen = srx.transport_len;
592 whdr.epoch = htonl(sp->hdr.epoch);
593 whdr.cid = htonl(sp->hdr.cid);
594 whdr.callNumber = htonl(sp->hdr.callNumber);
595 whdr.serviceId = htons(sp->hdr.serviceId);
596 whdr.flags = sp->hdr.flags;
597 whdr.flags ^= RXRPC_CLIENT_INITIATED;
598 whdr.flags &= RXRPC_CLIENT_INITIATED;
600 ret = kernel_sendmsg(local->socket, &msg,
603 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
604 rxrpc_tx_point_reject);
606 trace_rxrpc_tx_packet(local->debug_id, &whdr,
607 rxrpc_tx_point_reject);
610 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
617 * Send a VERSION reply to a peer as a keepalive.
619 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
621 struct rxrpc_wire_header whdr;
629 msg.msg_name = &peer->srx.transport;
630 msg.msg_namelen = peer->srx.transport_len;
631 msg.msg_control = NULL;
632 msg.msg_controllen = 0;
635 whdr.epoch = htonl(peer->local->rxnet->epoch);
640 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
641 whdr.flags = RXRPC_LAST_PACKET;
643 whdr.securityIndex = 0;
647 iov[0].iov_base = &whdr;
648 iov[0].iov_len = sizeof(whdr);
649 iov[1].iov_base = (char *)rxrpc_keepalive_string;
650 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
652 len = iov[0].iov_len + iov[1].iov_len;
654 _proto("Tx VERSION (keepalive)");
656 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
658 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
659 rxrpc_tx_point_version_keepalive);
661 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
662 rxrpc_tx_point_version_keepalive);
664 peer->last_tx_at = ktime_get_seconds();