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(0);
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 * Record the beginning of an RTT probe.
129 static int rxrpc_begin_rtt_probe(struct rxrpc_call *call, rxrpc_serial_t serial,
130 enum rxrpc_rtt_tx_trace why)
132 unsigned long avail = call->rtt_avail;
135 if (!(avail & RXRPC_CALL_RTT_AVAIL_MASK))
138 rtt_slot = __ffs(avail & RXRPC_CALL_RTT_AVAIL_MASK);
139 if (!test_and_clear_bit(rtt_slot, &call->rtt_avail))
142 call->rtt_serial[rtt_slot] = serial;
143 call->rtt_sent_at[rtt_slot] = ktime_get_real();
144 smp_wmb(); /* Write data before avail bit */
145 set_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
147 trace_rxrpc_rtt_tx(call, why, rtt_slot, serial);
151 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_no_slot, rtt_slot, serial);
156 * Cancel an RTT probe.
158 static void rxrpc_cancel_rtt_probe(struct rxrpc_call *call,
159 rxrpc_serial_t serial, int rtt_slot)
161 if (rtt_slot != -1) {
162 clear_bit(rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
163 smp_wmb(); /* Clear pending bit before setting slot */
164 set_bit(rtt_slot, &call->rtt_avail);
165 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_cancel, rtt_slot, serial);
170 * Send an ACK call packet.
172 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
173 rxrpc_serial_t *_serial)
175 struct rxrpc_connection *conn;
176 struct rxrpc_ack_buffer *pkt;
179 rxrpc_serial_t serial;
180 rxrpc_seq_t hard_ack, top;
182 int ret, rtt_slot = -1;
185 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
188 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
194 msg.msg_name = &call->peer->srx.transport;
195 msg.msg_namelen = call->peer->srx.transport_len;
196 msg.msg_control = NULL;
197 msg.msg_controllen = 0;
200 pkt->whdr.epoch = htonl(conn->proto.epoch);
201 pkt->whdr.cid = htonl(call->cid);
202 pkt->whdr.callNumber = htonl(call->call_id);
204 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
205 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
206 pkt->whdr.userStatus = 0;
207 pkt->whdr.securityIndex = call->security_ix;
209 pkt->whdr.serviceId = htons(call->service_id);
211 spin_lock_bh(&call->lock);
213 reason = RXRPC_ACK_PING;
215 reason = call->ackr_reason;
216 if (!call->ackr_reason) {
217 spin_unlock_bh(&call->lock);
221 call->ackr_reason = 0;
223 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
225 spin_unlock_bh(&call->lock);
227 iov[0].iov_base = pkt;
228 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
229 iov[1].iov_base = &pkt->ackinfo;
230 iov[1].iov_len = sizeof(pkt->ackinfo);
231 len = iov[0].iov_len + iov[1].iov_len;
233 serial = atomic_inc_return(&conn->serial);
234 pkt->whdr.serial = htonl(serial);
235 trace_rxrpc_tx_ack(call->debug_id, serial,
236 ntohl(pkt->ack.firstPacket),
237 ntohl(pkt->ack.serial),
238 pkt->ack.reason, pkt->ack.nAcks);
243 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_ping);
245 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
246 conn->params.peer->last_tx_at = ktime_get_seconds();
248 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
249 rxrpc_tx_point_call_ack);
251 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
252 rxrpc_tx_point_call_ack);
253 rxrpc_tx_backoff(call, ret);
255 if (call->state < RXRPC_CALL_COMPLETE) {
257 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
258 rxrpc_propose_ACK(call, pkt->ack.reason,
259 ntohl(pkt->ack.serial),
261 rxrpc_propose_ack_retry_tx);
263 spin_lock_bh(&call->lock);
264 if (after(hard_ack, call->ackr_consumed))
265 call->ackr_consumed = hard_ack;
266 if (after(top, call->ackr_seen))
267 call->ackr_seen = top;
268 spin_unlock_bh(&call->lock);
271 rxrpc_set_keepalive(call);
280 * Send an ABORT call packet.
282 int rxrpc_send_abort_packet(struct rxrpc_call *call)
284 struct rxrpc_connection *conn;
285 struct rxrpc_abort_buffer pkt;
288 rxrpc_serial_t serial;
291 /* Don't bother sending aborts for a client call once the server has
292 * hard-ACK'd all of its request data. After that point, we're not
293 * going to stop the operation proceeding, and whilst we might limit
294 * the reply, it's not worth it if we can send a new call on the same
295 * channel instead, thereby closing off this call.
297 if (rxrpc_is_client_call(call) &&
298 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
301 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
306 msg.msg_name = &call->peer->srx.transport;
307 msg.msg_namelen = call->peer->srx.transport_len;
308 msg.msg_control = NULL;
309 msg.msg_controllen = 0;
312 pkt.whdr.epoch = htonl(conn->proto.epoch);
313 pkt.whdr.cid = htonl(call->cid);
314 pkt.whdr.callNumber = htonl(call->call_id);
316 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
317 pkt.whdr.flags = conn->out_clientflag;
318 pkt.whdr.userStatus = 0;
319 pkt.whdr.securityIndex = call->security_ix;
321 pkt.whdr.serviceId = htons(call->service_id);
322 pkt.abort_code = htonl(call->abort_code);
324 iov[0].iov_base = &pkt;
325 iov[0].iov_len = sizeof(pkt);
327 serial = atomic_inc_return(&conn->serial);
328 pkt.whdr.serial = htonl(serial);
330 ret = kernel_sendmsg(conn->params.local->socket,
331 &msg, iov, 1, sizeof(pkt));
332 conn->params.peer->last_tx_at = ktime_get_seconds();
334 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
335 rxrpc_tx_point_call_abort);
337 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
338 rxrpc_tx_point_call_abort);
339 rxrpc_tx_backoff(call, ret);
344 * send a packet through the transport endpoint
346 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
349 struct rxrpc_connection *conn = call->conn;
350 struct rxrpc_wire_header whdr;
351 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
354 rxrpc_serial_t serial;
356 int ret, rtt_slot = -1;
358 _enter(",{%d}", skb->len);
360 /* Each transmission of a Tx packet needs a new serial number */
361 serial = atomic_inc_return(&conn->serial);
363 whdr.epoch = htonl(conn->proto.epoch);
364 whdr.cid = htonl(call->cid);
365 whdr.callNumber = htonl(call->call_id);
366 whdr.seq = htonl(sp->hdr.seq);
367 whdr.serial = htonl(serial);
368 whdr.type = RXRPC_PACKET_TYPE_DATA;
369 whdr.flags = sp->hdr.flags;
371 whdr.securityIndex = call->security_ix;
372 whdr._rsvd = htons(sp->hdr._rsvd);
373 whdr.serviceId = htons(call->service_id);
375 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
377 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
379 iov[0].iov_base = &whdr;
380 iov[0].iov_len = sizeof(whdr);
381 iov[1].iov_base = skb->head;
382 iov[1].iov_len = skb->len;
383 len = iov[0].iov_len + iov[1].iov_len;
385 msg.msg_name = &call->peer->srx.transport;
386 msg.msg_namelen = call->peer->srx.transport_len;
387 msg.msg_control = NULL;
388 msg.msg_controllen = 0;
391 /* If our RTT cache needs working on, request an ACK. Also request
392 * ACKs if a DATA packet appears to have been lost.
394 * However, we mustn't request an ACK on the last reply packet of a
395 * service call, lest OpenAFS incorrectly send us an ACK with some
396 * soft-ACKs in it and then never follow up with a proper hard ACK.
398 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
401 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
403 call->cong_mode == RXRPC_CALL_SLOW_START ||
404 (call->peer->rtt_count < 3 && sp->hdr.seq & 1) ||
405 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
407 whdr.flags |= RXRPC_REQUEST_ACK;
409 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
411 if ((lose++ & 7) == 7) {
413 trace_rxrpc_tx_data(call, sp->hdr.seq, serial,
414 whdr.flags, retrans, true);
419 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags, retrans,
422 /* send the packet with the don't fragment bit set if we currently
423 * think it's small enough */
424 if (iov[1].iov_len >= call->peer->maxdata)
425 goto send_fragmentable;
427 down_read(&conn->params.local->defrag_sem);
429 sp->hdr.serial = serial;
430 smp_wmb(); /* Set serial before timestamp */
431 skb->tstamp = ktime_get_real();
432 if (whdr.flags & RXRPC_REQUEST_ACK)
433 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
435 /* send the packet by UDP
436 * - returns -EMSGSIZE if UDP would have to fragment the packet
437 * to go out of the interface
438 * - in which case, we'll have processed the ICMP error
439 * message and update the peer record
441 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
442 conn->params.peer->last_tx_at = ktime_get_seconds();
444 up_read(&conn->params.local->defrag_sem);
446 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
447 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
448 rxrpc_tx_point_call_data_nofrag);
450 trace_rxrpc_tx_packet(call->debug_id, &whdr,
451 rxrpc_tx_point_call_data_nofrag);
454 rxrpc_tx_backoff(call, ret);
455 if (ret == -EMSGSIZE)
456 goto send_fragmentable;
460 if (whdr.flags & RXRPC_REQUEST_ACK) {
461 call->peer->rtt_last_req = skb->tstamp;
462 if (call->peer->rtt_count > 1) {
463 unsigned long nowj = jiffies, ack_lost_at;
465 ack_lost_at = rxrpc_get_rto_backoff(call->peer, retrans);
467 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
468 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
469 rxrpc_timer_set_for_lost_ack);
473 if (sp->hdr.seq == 1 &&
474 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
476 unsigned long nowj = jiffies, expect_rx_by;
478 expect_rx_by = nowj + call->next_rx_timo;
479 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
480 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
481 rxrpc_timer_set_for_normal);
484 rxrpc_set_keepalive(call);
486 /* Cancel the call if the initial transmission fails,
487 * particularly if that's due to network routing issues that
488 * aren't going away anytime soon. The layer above can arrange
489 * the retransmission.
491 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
492 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
496 _leave(" = %d [%u]", ret, call->peer->maxdata);
500 /* attempt to send this message with fragmentation enabled */
501 _debug("send fragment");
503 down_write(&conn->params.local->defrag_sem);
505 sp->hdr.serial = serial;
506 smp_wmb(); /* Set serial before timestamp */
507 skb->tstamp = ktime_get_real();
508 if (whdr.flags & RXRPC_REQUEST_ACK)
509 rtt_slot = rxrpc_begin_rtt_probe(call, serial, rxrpc_rtt_tx_data);
511 switch (conn->params.local->srx.transport.family) {
514 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
516 ret = kernel_sendmsg(conn->params.local->socket, &msg,
518 conn->params.peer->last_tx_at = ktime_get_seconds();
520 ip_sock_set_mtu_discover(conn->params.local->socket->sk,
529 rxrpc_cancel_rtt_probe(call, serial, rtt_slot);
530 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
531 rxrpc_tx_point_call_data_frag);
533 trace_rxrpc_tx_packet(call->debug_id, &whdr,
534 rxrpc_tx_point_call_data_frag);
536 rxrpc_tx_backoff(call, ret);
538 up_write(&conn->params.local->defrag_sem);
543 * reject packets through the local endpoint
545 void rxrpc_reject_packets(struct rxrpc_local *local)
547 struct sockaddr_rxrpc srx;
548 struct rxrpc_skb_priv *sp;
549 struct rxrpc_wire_header whdr;
557 _enter("%d", local->debug_id);
559 iov[0].iov_base = &whdr;
560 iov[0].iov_len = sizeof(whdr);
561 iov[1].iov_base = &code;
562 iov[1].iov_len = sizeof(code);
564 msg.msg_name = &srx.transport;
565 msg.msg_control = NULL;
566 msg.msg_controllen = 0;
569 memset(&whdr, 0, sizeof(whdr));
571 while ((skb = skb_dequeue(&local->reject_queue))) {
572 rxrpc_see_skb(skb, rxrpc_skb_seen);
576 case RXRPC_SKB_MARK_REJECT_BUSY:
577 whdr.type = RXRPC_PACKET_TYPE_BUSY;
581 case RXRPC_SKB_MARK_REJECT_ABORT:
582 whdr.type = RXRPC_PACKET_TYPE_ABORT;
583 code = htonl(skb->priority);
584 size = sizeof(whdr) + sizeof(code);
588 rxrpc_free_skb(skb, rxrpc_skb_freed);
592 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
593 msg.msg_namelen = srx.transport_len;
595 whdr.epoch = htonl(sp->hdr.epoch);
596 whdr.cid = htonl(sp->hdr.cid);
597 whdr.callNumber = htonl(sp->hdr.callNumber);
598 whdr.serviceId = htons(sp->hdr.serviceId);
599 whdr.flags = sp->hdr.flags;
600 whdr.flags ^= RXRPC_CLIENT_INITIATED;
601 whdr.flags &= RXRPC_CLIENT_INITIATED;
603 ret = kernel_sendmsg(local->socket, &msg,
606 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
607 rxrpc_tx_point_reject);
609 trace_rxrpc_tx_packet(local->debug_id, &whdr,
610 rxrpc_tx_point_reject);
613 rxrpc_free_skb(skb, rxrpc_skb_freed);
620 * Send a VERSION reply to a peer as a keepalive.
622 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
624 struct rxrpc_wire_header whdr;
632 msg.msg_name = &peer->srx.transport;
633 msg.msg_namelen = peer->srx.transport_len;
634 msg.msg_control = NULL;
635 msg.msg_controllen = 0;
638 whdr.epoch = htonl(peer->local->rxnet->epoch);
643 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
644 whdr.flags = RXRPC_LAST_PACKET;
646 whdr.securityIndex = 0;
650 iov[0].iov_base = &whdr;
651 iov[0].iov_len = sizeof(whdr);
652 iov[1].iov_base = (char *)rxrpc_keepalive_string;
653 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
655 len = iov[0].iov_len + iov[1].iov_len;
657 _proto("Tx VERSION (keepalive)");
659 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
661 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
662 rxrpc_tx_point_version_keepalive);
664 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
665 rxrpc_tx_point_version_keepalive);
667 peer->last_tx_at = ktime_get_seconds();