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;
36 * Fill out an ACK packet.
38 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
39 struct rxrpc_call *call,
40 struct rxrpc_ack_buffer *pkt,
41 rxrpc_seq_t *_hard_ack,
45 rxrpc_serial_t serial;
46 rxrpc_seq_t hard_ack, top, seq;
51 /* Barrier against rxrpc_input_data(). */
52 serial = call->ackr_serial;
53 hard_ack = READ_ONCE(call->rx_hard_ack);
54 top = smp_load_acquire(&call->rx_top);
55 *_hard_ack = hard_ack;
58 pkt->ack.bufferSpace = htons(8);
59 pkt->ack.maxSkew = htons(call->ackr_skew);
60 pkt->ack.firstPacket = htonl(hard_ack + 1);
61 pkt->ack.previousPacket = htonl(call->ackr_prev_seq);
62 pkt->ack.serial = htonl(serial);
63 pkt->ack.reason = reason;
64 pkt->ack.nAcks = top - hard_ack;
66 if (reason == RXRPC_ACK_PING)
67 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
69 if (after(top, hard_ack)) {
72 ix = seq & RXRPC_RXTX_BUFF_MASK;
73 if (call->rxtx_buffer[ix])
74 *ackp++ = RXRPC_ACK_TYPE_ACK;
76 *ackp++ = RXRPC_ACK_TYPE_NACK;
78 } while (before_eq(seq, top));
81 mtu = conn->params.peer->if_mtu;
82 mtu -= conn->params.peer->hdrsize;
83 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
84 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
85 pkt->ackinfo.maxMTU = htonl(mtu);
86 pkt->ackinfo.rwind = htonl(call->rx_winsize);
87 pkt->ackinfo.jumbo_max = htonl(jmax);
92 return top - hard_ack + 3;
96 * Send an ACK call packet.
98 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping)
100 struct rxrpc_connection *conn = NULL;
101 struct rxrpc_ack_buffer *pkt;
104 rxrpc_serial_t serial;
105 rxrpc_seq_t hard_ack, top;
110 spin_lock_bh(&call->lock);
112 conn = rxrpc_get_connection_maybe(call->conn);
113 spin_unlock_bh(&call->lock);
117 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
119 rxrpc_put_connection(conn);
123 msg.msg_name = &call->peer->srx.transport;
124 msg.msg_namelen = call->peer->srx.transport_len;
125 msg.msg_control = NULL;
126 msg.msg_controllen = 0;
129 pkt->whdr.epoch = htonl(conn->proto.epoch);
130 pkt->whdr.cid = htonl(call->cid);
131 pkt->whdr.callNumber = htonl(call->call_id);
133 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
134 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
135 pkt->whdr.userStatus = 0;
136 pkt->whdr.securityIndex = call->security_ix;
138 pkt->whdr.serviceId = htons(call->service_id);
140 spin_lock_bh(&call->lock);
142 reason = RXRPC_ACK_PING;
144 reason = call->ackr_reason;
145 if (!call->ackr_reason) {
146 spin_unlock_bh(&call->lock);
150 call->ackr_reason = 0;
152 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
154 spin_unlock_bh(&call->lock);
156 iov[0].iov_base = pkt;
157 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
158 iov[1].iov_base = &pkt->ackinfo;
159 iov[1].iov_len = sizeof(pkt->ackinfo);
160 len = iov[0].iov_len + iov[1].iov_len;
162 serial = atomic_inc_return(&conn->serial);
163 pkt->whdr.serial = htonl(serial);
164 trace_rxrpc_tx_ack(call, serial,
165 ntohl(pkt->ack.firstPacket),
166 ntohl(pkt->ack.serial),
167 pkt->ack.reason, pkt->ack.nAcks);
170 call->ping_serial = serial;
172 /* We need to stick a time in before we send the packet in case
173 * the reply gets back before kernel_sendmsg() completes - but
174 * asking UDP to send the packet can take a relatively long
175 * time, so we update the time after, on the assumption that
176 * the packet transmission is more likely to happen towards the
177 * end of the kernel_sendmsg() call.
179 call->ping_time = ktime_get_real();
180 set_bit(RXRPC_CALL_PINGING, &call->flags);
181 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
184 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
186 call->ping_time = ktime_get_real();
188 if (call->state < RXRPC_CALL_COMPLETE) {
191 clear_bit(RXRPC_CALL_PINGING, &call->flags);
192 rxrpc_propose_ACK(call, pkt->ack.reason,
193 ntohs(pkt->ack.maxSkew),
194 ntohl(pkt->ack.serial),
196 rxrpc_propose_ack_retry_tx);
198 spin_lock_bh(&call->lock);
199 if (after(hard_ack, call->ackr_consumed))
200 call->ackr_consumed = hard_ack;
201 if (after(top, call->ackr_seen))
202 call->ackr_seen = top;
203 spin_unlock_bh(&call->lock);
208 rxrpc_put_connection(conn);
214 * Send an ABORT call packet.
216 int rxrpc_send_abort_packet(struct rxrpc_call *call)
218 struct rxrpc_connection *conn = NULL;
219 struct rxrpc_abort_buffer pkt;
222 rxrpc_serial_t serial;
225 /* Don't bother sending aborts for a client call once the server has
226 * hard-ACK'd all of its request data. After that point, we're not
227 * going to stop the operation proceeding, and whilst we might limit
228 * the reply, it's not worth it if we can send a new call on the same
229 * channel instead, thereby closing off this call.
231 if (rxrpc_is_client_call(call) &&
232 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
235 spin_lock_bh(&call->lock);
237 conn = rxrpc_get_connection_maybe(call->conn);
238 spin_unlock_bh(&call->lock);
242 msg.msg_name = &call->peer->srx.transport;
243 msg.msg_namelen = call->peer->srx.transport_len;
244 msg.msg_control = NULL;
245 msg.msg_controllen = 0;
248 pkt.whdr.epoch = htonl(conn->proto.epoch);
249 pkt.whdr.cid = htonl(call->cid);
250 pkt.whdr.callNumber = htonl(call->call_id);
252 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
253 pkt.whdr.flags = conn->out_clientflag;
254 pkt.whdr.userStatus = 0;
255 pkt.whdr.securityIndex = call->security_ix;
257 pkt.whdr.serviceId = htons(call->service_id);
258 pkt.abort_code = htonl(call->abort_code);
260 iov[0].iov_base = &pkt;
261 iov[0].iov_len = sizeof(pkt);
263 serial = atomic_inc_return(&conn->serial);
264 pkt.whdr.serial = htonl(serial);
266 ret = kernel_sendmsg(conn->params.local->socket,
267 &msg, iov, 1, sizeof(pkt));
269 rxrpc_put_connection(conn);
274 * send a packet through the transport endpoint
276 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
279 struct rxrpc_connection *conn = call->conn;
280 struct rxrpc_wire_header whdr;
281 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
284 rxrpc_serial_t serial;
289 _enter(",{%d}", skb->len);
291 /* Each transmission of a Tx packet needs a new serial number */
292 serial = atomic_inc_return(&conn->serial);
294 whdr.epoch = htonl(conn->proto.epoch);
295 whdr.cid = htonl(call->cid);
296 whdr.callNumber = htonl(call->call_id);
297 whdr.seq = htonl(sp->hdr.seq);
298 whdr.serial = htonl(serial);
299 whdr.type = RXRPC_PACKET_TYPE_DATA;
300 whdr.flags = sp->hdr.flags;
302 whdr.securityIndex = call->security_ix;
303 whdr._rsvd = htons(sp->hdr._rsvd);
304 whdr.serviceId = htons(call->service_id);
306 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
308 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
310 iov[0].iov_base = &whdr;
311 iov[0].iov_len = sizeof(whdr);
312 iov[1].iov_base = skb->head;
313 iov[1].iov_len = skb->len;
314 len = iov[0].iov_len + iov[1].iov_len;
316 msg.msg_name = &call->peer->srx.transport;
317 msg.msg_namelen = call->peer->srx.transport_len;
318 msg.msg_control = NULL;
319 msg.msg_controllen = 0;
322 /* If our RTT cache needs working on, request an ACK. Also request
323 * ACKs if a DATA packet appears to have been lost.
325 if (!(sp->hdr.flags & RXRPC_LAST_PACKET) &&
327 call->cong_mode == RXRPC_CALL_SLOW_START ||
328 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
329 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
331 whdr.flags |= RXRPC_REQUEST_ACK;
333 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
335 if ((lose++ & 7) == 7) {
342 _proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
344 /* send the packet with the don't fragment bit set if we currently
345 * think it's small enough */
346 if (iov[1].iov_len >= call->peer->maxdata)
347 goto send_fragmentable;
349 down_read(&conn->params.local->defrag_sem);
350 /* send the packet by UDP
351 * - returns -EMSGSIZE if UDP would have to fragment the packet
352 * to go out of the interface
353 * - in which case, we'll have processed the ICMP error
354 * message and update the peer record
356 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
358 up_read(&conn->params.local->defrag_sem);
359 if (ret == -EMSGSIZE)
360 goto send_fragmentable;
363 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
366 ktime_t now = ktime_get_real();
369 sp->hdr.serial = serial;
370 if (whdr.flags & RXRPC_REQUEST_ACK) {
371 call->peer->rtt_last_req = now;
372 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
375 _leave(" = %d [%u]", ret, call->peer->maxdata);
379 /* attempt to send this message with fragmentation enabled */
380 _debug("send fragment");
382 down_write(&conn->params.local->defrag_sem);
384 switch (conn->params.local->srx.transport.family) {
386 opt = IP_PMTUDISC_DONT;
387 ret = kernel_setsockopt(conn->params.local->socket,
388 SOL_IP, IP_MTU_DISCOVER,
389 (char *)&opt, sizeof(opt));
391 ret = kernel_sendmsg(conn->params.local->socket, &msg,
394 opt = IP_PMTUDISC_DO;
395 kernel_setsockopt(conn->params.local->socket, SOL_IP,
397 (char *)&opt, sizeof(opt));
401 #ifdef CONFIG_AF_RXRPC_IPV6
403 opt = IPV6_PMTUDISC_DONT;
404 ret = kernel_setsockopt(conn->params.local->socket,
405 SOL_IPV6, IPV6_MTU_DISCOVER,
406 (char *)&opt, sizeof(opt));
408 ret = kernel_sendmsg(conn->params.local->socket, &msg,
409 iov, 1, iov[0].iov_len);
411 opt = IPV6_PMTUDISC_DO;
412 kernel_setsockopt(conn->params.local->socket,
413 SOL_IPV6, IPV6_MTU_DISCOVER,
414 (char *)&opt, sizeof(opt));
420 up_write(&conn->params.local->defrag_sem);
425 * reject packets through the local endpoint
427 void rxrpc_reject_packets(struct rxrpc_local *local)
429 struct sockaddr_rxrpc srx;
430 struct rxrpc_skb_priv *sp;
431 struct rxrpc_wire_header whdr;
438 _enter("%d", local->debug_id);
440 iov[0].iov_base = &whdr;
441 iov[0].iov_len = sizeof(whdr);
442 iov[1].iov_base = &code;
443 iov[1].iov_len = sizeof(code);
444 size = sizeof(whdr) + sizeof(code);
446 msg.msg_name = &srx.transport;
447 msg.msg_control = NULL;
448 msg.msg_controllen = 0;
451 memset(&whdr, 0, sizeof(whdr));
452 whdr.type = RXRPC_PACKET_TYPE_ABORT;
454 while ((skb = skb_dequeue(&local->reject_queue))) {
455 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
458 if (rxrpc_extract_addr_from_skb(local, &srx, skb) == 0) {
459 msg.msg_namelen = srx.transport_len;
461 code = htonl(skb->priority);
463 whdr.epoch = htonl(sp->hdr.epoch);
464 whdr.cid = htonl(sp->hdr.cid);
465 whdr.callNumber = htonl(sp->hdr.callNumber);
466 whdr.serviceId = htons(sp->hdr.serviceId);
467 whdr.flags = sp->hdr.flags;
468 whdr.flags ^= RXRPC_CLIENT_INITIATED;
469 whdr.flags &= RXRPC_CLIENT_INITIATED;
471 kernel_sendmsg(local->socket, &msg, iov, 2, size);
474 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);