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;
78 rxrpc_seq_t hard_ack, top, seq;
83 tmp = atomic_xchg(&call->ackr_nr_unacked, 0);
84 tmp |= atomic_xchg(&call->ackr_nr_consumed, 0);
85 if (!tmp && (reason == RXRPC_ACK_DELAY ||
86 reason == RXRPC_ACK_IDLE))
89 /* Barrier against rxrpc_input_data(). */
90 serial = call->ackr_serial;
91 hard_ack = READ_ONCE(call->rx_hard_ack);
92 top = smp_load_acquire(&call->rx_top);
93 *_hard_ack = hard_ack;
96 pkt->ack.bufferSpace = htons(8);
97 pkt->ack.maxSkew = htons(0);
98 pkt->ack.firstPacket = htonl(hard_ack + 1);
99 pkt->ack.previousPacket = htonl(call->ackr_highest_seq);
100 pkt->ack.serial = htonl(serial);
101 pkt->ack.reason = reason;
102 pkt->ack.nAcks = top - hard_ack;
104 if (reason == RXRPC_ACK_PING)
105 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
107 if (after(top, hard_ack)) {
110 ix = seq & RXRPC_RXTX_BUFF_MASK;
111 if (call->rxtx_buffer[ix])
112 *ackp++ = RXRPC_ACK_TYPE_ACK;
114 *ackp++ = RXRPC_ACK_TYPE_NACK;
116 } while (before_eq(seq, top));
119 mtu = conn->params.peer->if_mtu;
120 mtu -= conn->params.peer->hdrsize;
121 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
122 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
123 pkt->ackinfo.maxMTU = htonl(mtu);
124 pkt->ackinfo.rwind = htonl(call->rx_winsize);
125 pkt->ackinfo.jumbo_max = htonl(jmax);
130 return top - hard_ack + 3;
134 * Record the beginning of an RTT probe.
136 static int rxrpc_begin_rtt_probe(struct rxrpc_call *call, rxrpc_serial_t serial,
137 enum rxrpc_rtt_tx_trace why)
139 unsigned long avail = call->rtt_avail;
142 if (!(avail & RXRPC_CALL_RTT_AVAIL_MASK))
145 rtt_slot = __ffs(avail & RXRPC_CALL_RTT_AVAIL_MASK);
146 if (!test_and_clear_bit(rtt_slot, &call->rtt_avail))
149 call->rtt_serial[rtt_slot] = serial;
150 call->rtt_sent_at[rtt_slot] = ktime_get_real();
151 smp_wmb(); /* Write data before avail bit */
152 set_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
154 trace_rxrpc_rtt_tx(call, why, rtt_slot, serial);
158 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_no_slot, rtt_slot, serial);
163 * Cancel an RTT probe.
165 static void rxrpc_cancel_rtt_probe(struct rxrpc_call *call,
166 rxrpc_serial_t serial, int rtt_slot)
168 if (rtt_slot != -1) {
169 clear_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
170 smp_wmb(); /* Clear pending bit before setting slot */
171 set_bit(rtt_slot, &call->rtt_avail);
172 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_cancel, rtt_slot, serial);
177 * Send an ACK call packet.
179 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
180 rxrpc_serial_t *_serial)
182 struct rxrpc_connection *conn;
183 struct rxrpc_ack_buffer *pkt;
186 rxrpc_serial_t serial;
187 rxrpc_seq_t hard_ack, top;
189 int ret, rtt_slot = -1;
192 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
195 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
201 msg.msg_name = &call->peer->srx.transport;
202 msg.msg_namelen = call->peer->srx.transport_len;
203 msg.msg_control = NULL;
204 msg.msg_controllen = 0;
207 pkt->whdr.epoch = htonl(conn->proto.epoch);
208 pkt->whdr.cid = htonl(call->cid);
209 pkt->whdr.callNumber = htonl(call->call_id);
211 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
212 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
213 pkt->whdr.userStatus = 0;
214 pkt->whdr.securityIndex = call->security_ix;
216 pkt->whdr.serviceId = htons(call->service_id);
218 spin_lock_bh(&call->lock);
220 reason = RXRPC_ACK_PING;
222 reason = call->ackr_reason;
223 if (!call->ackr_reason) {
224 spin_unlock_bh(&call->lock);
228 call->ackr_reason = 0;
230 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
232 spin_unlock_bh(&call->lock);
238 iov[0].iov_base = pkt;
239 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
240 iov[1].iov_base = &pkt->ackinfo;
241 iov[1].iov_len = sizeof(pkt->ackinfo);
242 len = iov[0].iov_len + iov[1].iov_len;
244 serial = atomic_inc_return(&conn->serial);
245 pkt->whdr.serial = htonl(serial);
246 trace_rxrpc_tx_ack(call->debug_id, serial,
247 ntohl(pkt->ack.firstPacket),
248 ntohl(pkt->ack.serial),
249 pkt->ack.reason, pkt->ack.nAcks);
254 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_ping);
256 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
257 conn->params.peer->last_tx_at = ktime_get_seconds();
259 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
260 rxrpc_tx_point_call_ack);
262 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
263 rxrpc_tx_point_call_ack);
264 rxrpc_tx_backoff(call, ret);
266 if (call->state < RXRPC_CALL_COMPLETE) {
268 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
269 rxrpc_propose_ACK(call, pkt->ack.reason,
270 ntohl(pkt->ack.serial),
272 rxrpc_propose_ack_retry_tx);
275 rxrpc_set_keepalive(call);
284 * Send an ABORT call packet.
286 int rxrpc_send_abort_packet(struct rxrpc_call *call)
288 struct rxrpc_connection *conn;
289 struct rxrpc_abort_buffer pkt;
292 rxrpc_serial_t serial;
295 /* Don't bother sending aborts for a client call once the server has
296 * hard-ACK'd all of its request data. After that point, we're not
297 * going to stop the operation proceeding, and whilst we might limit
298 * the reply, it's not worth it if we can send a new call on the same
299 * channel instead, thereby closing off this call.
301 if (rxrpc_is_client_call(call) &&
302 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
305 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
310 msg.msg_name = &call->peer->srx.transport;
311 msg.msg_namelen = call->peer->srx.transport_len;
312 msg.msg_control = NULL;
313 msg.msg_controllen = 0;
316 pkt.whdr.epoch = htonl(conn->proto.epoch);
317 pkt.whdr.cid = htonl(call->cid);
318 pkt.whdr.callNumber = htonl(call->call_id);
320 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
321 pkt.whdr.flags = conn->out_clientflag;
322 pkt.whdr.userStatus = 0;
323 pkt.whdr.securityIndex = call->security_ix;
325 pkt.whdr.serviceId = htons(call->service_id);
326 pkt.abort_code = htonl(call->abort_code);
328 iov[0].iov_base = &pkt;
329 iov[0].iov_len = sizeof(pkt);
331 serial = atomic_inc_return(&conn->serial);
332 pkt.whdr.serial = htonl(serial);
334 ret = kernel_sendmsg(conn->params.local->socket,
335 &msg, iov, 1, sizeof(pkt));
336 conn->params.peer->last_tx_at = ktime_get_seconds();
338 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
339 rxrpc_tx_point_call_abort);
341 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
342 rxrpc_tx_point_call_abort);
343 rxrpc_tx_backoff(call, ret);
348 * send a packet through the transport endpoint
350 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
353 struct rxrpc_connection *conn = call->conn;
354 struct rxrpc_wire_header whdr;
355 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
358 rxrpc_serial_t serial;
360 int ret, rtt_slot = -1;
362 _enter(",{%d}", skb->len);
364 if (hlist_unhashed(&call->error_link)) {
365 spin_lock_bh(&call->peer->lock);
366 hlist_add_head_rcu(&call->error_link, &call->peer->error_targets);
367 spin_unlock_bh(&call->peer->lock);
370 /* Each transmission of a Tx packet needs a new serial number */
371 serial = atomic_inc_return(&conn->serial);
373 whdr.epoch = htonl(conn->proto.epoch);
374 whdr.cid = htonl(call->cid);
375 whdr.callNumber = htonl(call->call_id);
376 whdr.seq = htonl(sp->hdr.seq);
377 whdr.serial = htonl(serial);
378 whdr.type = RXRPC_PACKET_TYPE_DATA;
379 whdr.flags = sp->hdr.flags;
381 whdr.securityIndex = call->security_ix;
382 whdr._rsvd = htons(sp->hdr._rsvd);
383 whdr.serviceId = htons(call->service_id);
385 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
387 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
389 iov[0].iov_base = &whdr;
390 iov[0].iov_len = sizeof(whdr);
391 iov[1].iov_base = skb->head;
392 iov[1].iov_len = skb->len;
393 len = iov[0].iov_len + iov[1].iov_len;
395 msg.msg_name = &call->peer->srx.transport;
396 msg.msg_namelen = call->peer->srx.transport_len;
397 msg.msg_control = NULL;
398 msg.msg_controllen = 0;
401 /* If our RTT cache needs working on, request an ACK. Also request
402 * ACKs if a DATA packet appears to have been lost.
404 * However, we mustn't request an ACK on the last reply packet of a
405 * service call, lest OpenAFS incorrectly send us an ACK with some
406 * soft-ACKs in it and then never follow up with a proper hard ACK.
408 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
411 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
413 call->cong_mode == RXRPC_CALL_SLOW_START ||
414 (call->peer->rtt_count < 3 && sp->hdr.seq & 1) ||
415 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
417 whdr.flags |= RXRPC_REQUEST_ACK;
419 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
421 if ((lose++ & 7) == 7) {
423 trace_rxrpc_tx_data(call, sp->hdr.seq, serial,
424 whdr.flags, retrans, true);
429 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags, retrans,
432 /* send the packet with the don't fragment bit set if we currently
433 * think it's small enough */
434 if (iov[1].iov_len >= call->peer->maxdata)
435 goto send_fragmentable;
437 down_read(&conn->params.local->defrag_sem);
439 sp->hdr.serial = serial;
440 smp_wmb(); /* Set serial before timestamp */
441 skb->tstamp = ktime_get_real();
442 if (whdr.flags & RXRPC_REQUEST_ACK)
443 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
445 /* send the packet by UDP
446 * - returns -EMSGSIZE if UDP would have to fragment the packet
447 * to go out of the interface
448 * - in which case, we'll have processed the ICMP error
449 * message and update the peer record
451 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
452 conn->params.peer->last_tx_at = ktime_get_seconds();
454 up_read(&conn->params.local->defrag_sem);
456 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
457 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
458 rxrpc_tx_point_call_data_nofrag);
460 trace_rxrpc_tx_packet(call->debug_id, &whdr,
461 rxrpc_tx_point_call_data_nofrag);
464 rxrpc_tx_backoff(call, ret);
465 if (ret == -EMSGSIZE)
466 goto send_fragmentable;
470 if (whdr.flags & RXRPC_REQUEST_ACK) {
471 call->peer->rtt_last_req = skb->tstamp;
472 if (call->peer->rtt_count > 1) {
473 unsigned long nowj = jiffies, ack_lost_at;
475 ack_lost_at = rxrpc_get_rto_backoff(call->peer, false);
477 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
478 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
479 rxrpc_timer_set_for_lost_ack);
483 if (sp->hdr.seq == 1 &&
484 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
486 unsigned long nowj = jiffies, expect_rx_by;
488 expect_rx_by = nowj + call->next_rx_timo;
489 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
490 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
491 rxrpc_timer_set_for_normal);
494 rxrpc_set_keepalive(call);
496 /* Cancel the call if the initial transmission fails,
497 * particularly if that's due to network routing issues that
498 * aren't going away anytime soon. The layer above can arrange
499 * the retransmission.
501 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
502 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
506 _leave(" = %d [%u]", ret, call->peer->maxdata);
510 /* attempt to send this message with fragmentation enabled */
511 _debug("send fragment");
513 down_write(&conn->params.local->defrag_sem);
515 sp->hdr.serial = serial;
516 smp_wmb(); /* Set serial before timestamp */
517 skb->tstamp = ktime_get_real();
518 if (whdr.flags & RXRPC_REQUEST_ACK)
519 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
521 switch (conn->params.local->srx.transport.family) {
524 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
526 ret = kernel_sendmsg(conn->params.local->socket, &msg,
528 conn->params.peer->last_tx_at = ktime_get_seconds();
530 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
539 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
540 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
541 rxrpc_tx_point_call_data_frag);
543 trace_rxrpc_tx_packet(call->debug_id, &whdr,
544 rxrpc_tx_point_call_data_frag);
546 rxrpc_tx_backoff(call, ret);
548 up_write(&conn->params.local->defrag_sem);
553 * reject packets through the local endpoint
555 void rxrpc_reject_packets(struct rxrpc_local *local)
557 struct sockaddr_rxrpc srx;
558 struct rxrpc_skb_priv *sp;
559 struct rxrpc_wire_header whdr;
567 _enter("%d", local->debug_id);
569 iov[0].iov_base = &whdr;
570 iov[0].iov_len = sizeof(whdr);
571 iov[1].iov_base = &code;
572 iov[1].iov_len = sizeof(code);
574 msg.msg_name = &srx.transport;
575 msg.msg_control = NULL;
576 msg.msg_controllen = 0;
579 memset(&whdr, 0, sizeof(whdr));
581 while ((skb = skb_dequeue(&local->reject_queue))) {
582 rxrpc_see_skb(skb, rxrpc_skb_seen);
586 case RXRPC_SKB_MARK_REJECT_BUSY:
587 whdr.type = RXRPC_PACKET_TYPE_BUSY;
591 case RXRPC_SKB_MARK_REJECT_ABORT:
592 whdr.type = RXRPC_PACKET_TYPE_ABORT;
593 code = htonl(skb->priority);
594 size = sizeof(whdr) + sizeof(code);
598 rxrpc_free_skb(skb, rxrpc_skb_freed);
602 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
603 msg.msg_namelen = srx.transport_len;
605 whdr.epoch = htonl(sp->hdr.epoch);
606 whdr.cid = htonl(sp->hdr.cid);
607 whdr.callNumber = htonl(sp->hdr.callNumber);
608 whdr.serviceId = htons(sp->hdr.serviceId);
609 whdr.flags = sp->hdr.flags;
610 whdr.flags ^= RXRPC_CLIENT_INITIATED;
611 whdr.flags &= RXRPC_CLIENT_INITIATED;
613 ret = kernel_sendmsg(local->socket, &msg,
616 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
617 rxrpc_tx_point_reject);
619 trace_rxrpc_tx_packet(local->debug_id, &whdr,
620 rxrpc_tx_point_reject);
623 rxrpc_free_skb(skb, rxrpc_skb_freed);
630 * Send a VERSION reply to a peer as a keepalive.
632 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
634 struct rxrpc_wire_header whdr;
642 msg.msg_name = &peer->srx.transport;
643 msg.msg_namelen = peer->srx.transport_len;
644 msg.msg_control = NULL;
645 msg.msg_controllen = 0;
648 whdr.epoch = htonl(peer->local->rxnet->epoch);
653 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
654 whdr.flags = RXRPC_LAST_PACKET;
656 whdr.securityIndex = 0;
660 iov[0].iov_base = &whdr;
661 iov[0].iov_len = sizeof(whdr);
662 iov[1].iov_base = (char *)rxrpc_keepalive_string;
663 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
665 len = iov[0].iov_len + iov[1].iov_len;
667 _proto("Tx VERSION (keepalive)");
669 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
671 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
672 rxrpc_tx_point_version_keepalive);
674 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
675 rxrpc_tx_point_version_keepalive);
677 peer->last_tx_at = ktime_get_seconds();