Linux 5.7.7
[linux/fpc-iii.git] / net / rxrpc / output.c
blobf8b632a5c61979fbd02828e2e9b6945f8d274e8a
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)
6 */
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>
14 #include <net/sock.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;
21 u8 acks[255];
22 u8 pad[3];
23 struct rxrpc_ackinfo ackinfo;
26 struct rxrpc_abort_buffer {
27 struct rxrpc_wire_header whdr;
28 __be32 abort_code;
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)
38 if (ret < 0) {
39 u16 tx_backoff = READ_ONCE(call->tx_backoff);
41 if (tx_backoff < HZ)
42 WRITE_ONCE(call->tx_backoff, tx_backoff + 1);
43 } else {
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
54 * expiring.
56 static void rxrpc_set_keepalive(struct rxrpc_call *call)
58 unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
60 keepalive_at += now;
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,
73 rxrpc_seq_t *_top,
74 u8 reason)
76 rxrpc_serial_t serial;
77 rxrpc_seq_t hard_ack, top, seq;
78 int ix;
79 u32 mtu, jmax;
80 u8 *ackp = pkt->acks;
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;
87 *_top = top;
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)) {
101 seq = hard_ack + 1;
102 do {
103 ix = seq & RXRPC_RXTX_BUFF_MASK;
104 if (call->rxtx_buffer[ix])
105 *ackp++ = RXRPC_ACK_TYPE_ACK;
106 else
107 *ackp++ = RXRPC_ACK_TYPE_NACK;
108 seq++;
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);
120 *ackp++ = 0;
121 *ackp++ = 0;
122 *ackp++ = 0;
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;
133 struct rxrpc_ack_buffer *pkt;
134 struct msghdr msg;
135 struct kvec iov[2];
136 rxrpc_serial_t serial;
137 rxrpc_seq_t hard_ack, top;
138 size_t len, n;
139 int ret;
140 u8 reason;
142 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
143 return -ECONNRESET;
145 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
146 if (!pkt)
147 return -ENOMEM;
149 conn = call->conn;
151 msg.msg_name = &call->peer->srx.transport;
152 msg.msg_namelen = call->peer->srx.transport_len;
153 msg.msg_control = NULL;
154 msg.msg_controllen = 0;
155 msg.msg_flags = 0;
157 pkt->whdr.epoch = htonl(conn->proto.epoch);
158 pkt->whdr.cid = htonl(call->cid);
159 pkt->whdr.callNumber = htonl(call->call_id);
160 pkt->whdr.seq = 0;
161 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
162 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
163 pkt->whdr.userStatus = 0;
164 pkt->whdr.securityIndex = call->security_ix;
165 pkt->whdr._rsvd = 0;
166 pkt->whdr.serviceId = htons(call->service_id);
168 spin_lock_bh(&call->lock);
169 if (ping) {
170 reason = RXRPC_ACK_PING;
171 } else {
172 reason = call->ackr_reason;
173 if (!call->ackr_reason) {
174 spin_unlock_bh(&call->lock);
175 ret = 0;
176 goto out;
178 call->ackr_reason = 0;
180 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
182 spin_unlock_bh(&call->lock);
184 iov[0].iov_base = pkt;
185 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
186 iov[1].iov_base = &pkt->ackinfo;
187 iov[1].iov_len = sizeof(pkt->ackinfo);
188 len = iov[0].iov_len + iov[1].iov_len;
190 serial = atomic_inc_return(&conn->serial);
191 pkt->whdr.serial = htonl(serial);
192 trace_rxrpc_tx_ack(call->debug_id, serial,
193 ntohl(pkt->ack.firstPacket),
194 ntohl(pkt->ack.serial),
195 pkt->ack.reason, pkt->ack.nAcks);
196 if (_serial)
197 *_serial = serial;
199 if (ping) {
200 call->ping_serial = serial;
201 smp_wmb();
202 /* We need to stick a time in before we send the packet in case
203 * the reply gets back before kernel_sendmsg() completes - but
204 * asking UDP to send the packet can take a relatively long
205 * time.
207 call->ping_time = ktime_get_real();
208 set_bit(RXRPC_CALL_PINGING, &call->flags);
209 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
212 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
213 conn->params.peer->last_tx_at = ktime_get_seconds();
214 if (ret < 0)
215 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
216 rxrpc_tx_point_call_ack);
217 else
218 trace_rxrpc_tx_packet(call->debug_id, &pkt->whdr,
219 rxrpc_tx_point_call_ack);
220 rxrpc_tx_backoff(call, ret);
222 if (call->state < RXRPC_CALL_COMPLETE) {
223 if (ret < 0) {
224 if (ping)
225 clear_bit(RXRPC_CALL_PINGING, &call->flags);
226 rxrpc_propose_ACK(call, pkt->ack.reason,
227 ntohl(pkt->ack.serial),
228 false, true,
229 rxrpc_propose_ack_retry_tx);
230 } else {
231 spin_lock_bh(&call->lock);
232 if (after(hard_ack, call->ackr_consumed))
233 call->ackr_consumed = hard_ack;
234 if (after(top, call->ackr_seen))
235 call->ackr_seen = top;
236 spin_unlock_bh(&call->lock);
239 rxrpc_set_keepalive(call);
242 out:
243 kfree(pkt);
244 return ret;
248 * Send an ABORT call packet.
250 int rxrpc_send_abort_packet(struct rxrpc_call *call)
252 struct rxrpc_connection *conn;
253 struct rxrpc_abort_buffer pkt;
254 struct msghdr msg;
255 struct kvec iov[1];
256 rxrpc_serial_t serial;
257 int ret;
259 /* Don't bother sending aborts for a client call once the server has
260 * hard-ACK'd all of its request data. After that point, we're not
261 * going to stop the operation proceeding, and whilst we might limit
262 * the reply, it's not worth it if we can send a new call on the same
263 * channel instead, thereby closing off this call.
265 if (rxrpc_is_client_call(call) &&
266 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
267 return 0;
269 if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
270 return -ECONNRESET;
272 conn = call->conn;
274 msg.msg_name = &call->peer->srx.transport;
275 msg.msg_namelen = call->peer->srx.transport_len;
276 msg.msg_control = NULL;
277 msg.msg_controllen = 0;
278 msg.msg_flags = 0;
280 pkt.whdr.epoch = htonl(conn->proto.epoch);
281 pkt.whdr.cid = htonl(call->cid);
282 pkt.whdr.callNumber = htonl(call->call_id);
283 pkt.whdr.seq = 0;
284 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
285 pkt.whdr.flags = conn->out_clientflag;
286 pkt.whdr.userStatus = 0;
287 pkt.whdr.securityIndex = call->security_ix;
288 pkt.whdr._rsvd = 0;
289 pkt.whdr.serviceId = htons(call->service_id);
290 pkt.abort_code = htonl(call->abort_code);
292 iov[0].iov_base = &pkt;
293 iov[0].iov_len = sizeof(pkt);
295 serial = atomic_inc_return(&conn->serial);
296 pkt.whdr.serial = htonl(serial);
298 ret = kernel_sendmsg(conn->params.local->socket,
299 &msg, iov, 1, sizeof(pkt));
300 conn->params.peer->last_tx_at = ktime_get_seconds();
301 if (ret < 0)
302 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
303 rxrpc_tx_point_call_abort);
304 else
305 trace_rxrpc_tx_packet(call->debug_id, &pkt.whdr,
306 rxrpc_tx_point_call_abort);
307 rxrpc_tx_backoff(call, ret);
308 return ret;
312 * send a packet through the transport endpoint
314 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
315 bool retrans)
317 struct rxrpc_connection *conn = call->conn;
318 struct rxrpc_wire_header whdr;
319 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
320 struct msghdr msg;
321 struct kvec iov[2];
322 rxrpc_serial_t serial;
323 size_t len;
324 int ret, opt;
326 _enter(",{%d}", skb->len);
328 /* Each transmission of a Tx packet needs a new serial number */
329 serial = atomic_inc_return(&conn->serial);
331 whdr.epoch = htonl(conn->proto.epoch);
332 whdr.cid = htonl(call->cid);
333 whdr.callNumber = htonl(call->call_id);
334 whdr.seq = htonl(sp->hdr.seq);
335 whdr.serial = htonl(serial);
336 whdr.type = RXRPC_PACKET_TYPE_DATA;
337 whdr.flags = sp->hdr.flags;
338 whdr.userStatus = 0;
339 whdr.securityIndex = call->security_ix;
340 whdr._rsvd = htons(sp->hdr._rsvd);
341 whdr.serviceId = htons(call->service_id);
343 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
344 sp->hdr.seq == 1)
345 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
347 iov[0].iov_base = &whdr;
348 iov[0].iov_len = sizeof(whdr);
349 iov[1].iov_base = skb->head;
350 iov[1].iov_len = skb->len;
351 len = iov[0].iov_len + iov[1].iov_len;
353 msg.msg_name = &call->peer->srx.transport;
354 msg.msg_namelen = call->peer->srx.transport_len;
355 msg.msg_control = NULL;
356 msg.msg_controllen = 0;
357 msg.msg_flags = 0;
359 /* If our RTT cache needs working on, request an ACK. Also request
360 * ACKs if a DATA packet appears to have been lost.
362 * However, we mustn't request an ACK on the last reply packet of a
363 * service call, lest OpenAFS incorrectly send us an ACK with some
364 * soft-ACKs in it and then never follow up with a proper hard ACK.
366 if ((!(sp->hdr.flags & RXRPC_LAST_PACKET) ||
367 rxrpc_to_server(sp)
368 ) &&
369 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
370 retrans ||
371 call->cong_mode == RXRPC_CALL_SLOW_START ||
372 (call->peer->rtt_count < 3 && sp->hdr.seq & 1) ||
373 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
374 ktime_get_real())))
375 whdr.flags |= RXRPC_REQUEST_ACK;
377 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
378 static int lose;
379 if ((lose++ & 7) == 7) {
380 ret = 0;
381 trace_rxrpc_tx_data(call, sp->hdr.seq, serial,
382 whdr.flags, retrans, true);
383 goto done;
387 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags, retrans,
388 false);
390 /* send the packet with the don't fragment bit set if we currently
391 * think it's small enough */
392 if (iov[1].iov_len >= call->peer->maxdata)
393 goto send_fragmentable;
395 down_read(&conn->params.local->defrag_sem);
397 sp->hdr.serial = serial;
398 smp_wmb(); /* Set serial before timestamp */
399 skb->tstamp = ktime_get_real();
401 /* send the packet by UDP
402 * - returns -EMSGSIZE if UDP would have to fragment the packet
403 * to go out of the interface
404 * - in which case, we'll have processed the ICMP error
405 * message and update the peer record
407 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
408 conn->params.peer->last_tx_at = ktime_get_seconds();
410 up_read(&conn->params.local->defrag_sem);
411 if (ret < 0)
412 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
413 rxrpc_tx_point_call_data_nofrag);
414 else
415 trace_rxrpc_tx_packet(call->debug_id, &whdr,
416 rxrpc_tx_point_call_data_nofrag);
417 rxrpc_tx_backoff(call, ret);
418 if (ret == -EMSGSIZE)
419 goto send_fragmentable;
421 done:
422 if (ret >= 0) {
423 if (whdr.flags & RXRPC_REQUEST_ACK) {
424 call->peer->rtt_last_req = skb->tstamp;
425 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
426 if (call->peer->rtt_count > 1) {
427 unsigned long nowj = jiffies, ack_lost_at;
429 ack_lost_at = rxrpc_get_rto_backoff(call->peer, retrans);
430 ack_lost_at += nowj;
431 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
432 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
433 rxrpc_timer_set_for_lost_ack);
437 if (sp->hdr.seq == 1 &&
438 !test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER,
439 &call->flags)) {
440 unsigned long nowj = jiffies, expect_rx_by;
442 expect_rx_by = nowj + call->next_rx_timo;
443 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
444 rxrpc_reduce_call_timer(call, expect_rx_by, nowj,
445 rxrpc_timer_set_for_normal);
448 rxrpc_set_keepalive(call);
449 } else {
450 /* Cancel the call if the initial transmission fails,
451 * particularly if that's due to network routing issues that
452 * aren't going away anytime soon. The layer above can arrange
453 * the retransmission.
455 if (!test_and_set_bit(RXRPC_CALL_BEGAN_RX_TIMER, &call->flags))
456 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
457 RX_USER_ABORT, ret);
460 _leave(" = %d [%u]", ret, call->peer->maxdata);
461 return ret;
463 send_fragmentable:
464 /* attempt to send this message with fragmentation enabled */
465 _debug("send fragment");
467 down_write(&conn->params.local->defrag_sem);
469 sp->hdr.serial = serial;
470 smp_wmb(); /* Set serial before timestamp */
471 skb->tstamp = ktime_get_real();
473 switch (conn->params.local->srx.transport.family) {
474 case AF_INET6:
475 case AF_INET:
476 opt = IP_PMTUDISC_DONT;
477 kernel_setsockopt(conn->params.local->socket,
478 SOL_IP, IP_MTU_DISCOVER,
479 (char *)&opt, sizeof(opt));
480 ret = kernel_sendmsg(conn->params.local->socket, &msg,
481 iov, 2, len);
482 conn->params.peer->last_tx_at = ktime_get_seconds();
484 opt = IP_PMTUDISC_DO;
485 kernel_setsockopt(conn->params.local->socket,
486 SOL_IP, IP_MTU_DISCOVER,
487 (char *)&opt, sizeof(opt));
488 break;
490 default:
491 BUG();
494 if (ret < 0)
495 trace_rxrpc_tx_fail(call->debug_id, serial, ret,
496 rxrpc_tx_point_call_data_frag);
497 else
498 trace_rxrpc_tx_packet(call->debug_id, &whdr,
499 rxrpc_tx_point_call_data_frag);
500 rxrpc_tx_backoff(call, ret);
502 up_write(&conn->params.local->defrag_sem);
503 goto done;
507 * reject packets through the local endpoint
509 void rxrpc_reject_packets(struct rxrpc_local *local)
511 struct sockaddr_rxrpc srx;
512 struct rxrpc_skb_priv *sp;
513 struct rxrpc_wire_header whdr;
514 struct sk_buff *skb;
515 struct msghdr msg;
516 struct kvec iov[2];
517 size_t size;
518 __be32 code;
519 int ret, ioc;
521 _enter("%d", local->debug_id);
523 iov[0].iov_base = &whdr;
524 iov[0].iov_len = sizeof(whdr);
525 iov[1].iov_base = &code;
526 iov[1].iov_len = sizeof(code);
528 msg.msg_name = &srx.transport;
529 msg.msg_control = NULL;
530 msg.msg_controllen = 0;
531 msg.msg_flags = 0;
533 memset(&whdr, 0, sizeof(whdr));
535 while ((skb = skb_dequeue(&local->reject_queue))) {
536 rxrpc_see_skb(skb, rxrpc_skb_seen);
537 sp = rxrpc_skb(skb);
539 switch (skb->mark) {
540 case RXRPC_SKB_MARK_REJECT_BUSY:
541 whdr.type = RXRPC_PACKET_TYPE_BUSY;
542 size = sizeof(whdr);
543 ioc = 1;
544 break;
545 case RXRPC_SKB_MARK_REJECT_ABORT:
546 whdr.type = RXRPC_PACKET_TYPE_ABORT;
547 code = htonl(skb->priority);
548 size = sizeof(whdr) + sizeof(code);
549 ioc = 2;
550 break;
551 default:
552 rxrpc_free_skb(skb, rxrpc_skb_freed);
553 continue;
556 if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
557 msg.msg_namelen = srx.transport_len;
559 whdr.epoch = htonl(sp->hdr.epoch);
560 whdr.cid = htonl(sp->hdr.cid);
561 whdr.callNumber = htonl(sp->hdr.callNumber);
562 whdr.serviceId = htons(sp->hdr.serviceId);
563 whdr.flags = sp->hdr.flags;
564 whdr.flags ^= RXRPC_CLIENT_INITIATED;
565 whdr.flags &= RXRPC_CLIENT_INITIATED;
567 ret = kernel_sendmsg(local->socket, &msg,
568 iov, ioc, size);
569 if (ret < 0)
570 trace_rxrpc_tx_fail(local->debug_id, 0, ret,
571 rxrpc_tx_point_reject);
572 else
573 trace_rxrpc_tx_packet(local->debug_id, &whdr,
574 rxrpc_tx_point_reject);
577 rxrpc_free_skb(skb, rxrpc_skb_freed);
580 _leave("");
584 * Send a VERSION reply to a peer as a keepalive.
586 void rxrpc_send_keepalive(struct rxrpc_peer *peer)
588 struct rxrpc_wire_header whdr;
589 struct msghdr msg;
590 struct kvec iov[2];
591 size_t len;
592 int ret;
594 _enter("");
596 msg.msg_name = &peer->srx.transport;
597 msg.msg_namelen = peer->srx.transport_len;
598 msg.msg_control = NULL;
599 msg.msg_controllen = 0;
600 msg.msg_flags = 0;
602 whdr.epoch = htonl(peer->local->rxnet->epoch);
603 whdr.cid = 0;
604 whdr.callNumber = 0;
605 whdr.seq = 0;
606 whdr.serial = 0;
607 whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */
608 whdr.flags = RXRPC_LAST_PACKET;
609 whdr.userStatus = 0;
610 whdr.securityIndex = 0;
611 whdr._rsvd = 0;
612 whdr.serviceId = 0;
614 iov[0].iov_base = &whdr;
615 iov[0].iov_len = sizeof(whdr);
616 iov[1].iov_base = (char *)rxrpc_keepalive_string;
617 iov[1].iov_len = sizeof(rxrpc_keepalive_string);
619 len = iov[0].iov_len + iov[1].iov_len;
621 _proto("Tx VERSION (keepalive)");
623 ret = kernel_sendmsg(peer->local->socket, &msg, iov, 2, len);
624 if (ret < 0)
625 trace_rxrpc_tx_fail(peer->debug_id, 0, ret,
626 rxrpc_tx_point_version_keepalive);
627 else
628 trace_rxrpc_tx_packet(peer->debug_id, &whdr,
629 rxrpc_tx_point_version_keepalive);
631 peer->last_tx_at = ktime_get_seconds();
632 _leave("");