net: ipv4: don't allow setting net.ipv4.route.min_pmtu below 68
[linux/fpc-iii.git] / net / rxrpc / output.c
blobcf73dc006c3bf0d3866461c63a5a7c51c161c90b
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>
18 #include <net/sock.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;
25 u8 acks[255];
26 u8 pad[3];
27 struct rxrpc_ackinfo ackinfo;
30 struct rxrpc_abort_buffer {
31 struct rxrpc_wire_header whdr;
32 __be32 abort_code;
36 * Arrange for a keepalive ping a certain time after we last transmitted. This
37 * lets the far side know we're still interested in this call and helps keep
38 * the route through any intervening firewall open.
40 * Receiving a response to the ping will prevent the ->expect_rx_by timer from
41 * expiring.
43 static void rxrpc_set_keepalive(struct rxrpc_call *call)
45 unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6;
47 keepalive_at += now;
48 WRITE_ONCE(call->keepalive_at, keepalive_at);
49 rxrpc_reduce_call_timer(call, keepalive_at, now,
50 rxrpc_timer_set_for_keepalive);
54 * Fill out an ACK packet.
56 static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn,
57 struct rxrpc_call *call,
58 struct rxrpc_ack_buffer *pkt,
59 rxrpc_seq_t *_hard_ack,
60 rxrpc_seq_t *_top,
61 u8 reason)
63 rxrpc_serial_t serial;
64 rxrpc_seq_t hard_ack, top, seq;
65 int ix;
66 u32 mtu, jmax;
67 u8 *ackp = pkt->acks;
69 /* Barrier against rxrpc_input_data(). */
70 serial = call->ackr_serial;
71 hard_ack = READ_ONCE(call->rx_hard_ack);
72 top = smp_load_acquire(&call->rx_top);
73 *_hard_ack = hard_ack;
74 *_top = top;
76 pkt->ack.bufferSpace = htons(8);
77 pkt->ack.maxSkew = htons(call->ackr_skew);
78 pkt->ack.firstPacket = htonl(hard_ack + 1);
79 pkt->ack.previousPacket = htonl(call->ackr_prev_seq);
80 pkt->ack.serial = htonl(serial);
81 pkt->ack.reason = reason;
82 pkt->ack.nAcks = top - hard_ack;
84 if (reason == RXRPC_ACK_PING)
85 pkt->whdr.flags |= RXRPC_REQUEST_ACK;
87 if (after(top, hard_ack)) {
88 seq = hard_ack + 1;
89 do {
90 ix = seq & RXRPC_RXTX_BUFF_MASK;
91 if (call->rxtx_buffer[ix])
92 *ackp++ = RXRPC_ACK_TYPE_ACK;
93 else
94 *ackp++ = RXRPC_ACK_TYPE_NACK;
95 seq++;
96 } while (before_eq(seq, top));
99 mtu = conn->params.peer->if_mtu;
100 mtu -= conn->params.peer->hdrsize;
101 jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
102 pkt->ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
103 pkt->ackinfo.maxMTU = htonl(mtu);
104 pkt->ackinfo.rwind = htonl(call->rx_winsize);
105 pkt->ackinfo.jumbo_max = htonl(jmax);
107 *ackp++ = 0;
108 *ackp++ = 0;
109 *ackp++ = 0;
110 return top - hard_ack + 3;
114 * Send an ACK call packet.
116 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
117 rxrpc_serial_t *_serial)
119 struct rxrpc_connection *conn = NULL;
120 struct rxrpc_ack_buffer *pkt;
121 struct msghdr msg;
122 struct kvec iov[2];
123 rxrpc_serial_t serial;
124 rxrpc_seq_t hard_ack, top;
125 size_t len, n;
126 int ret;
127 u8 reason;
129 spin_lock_bh(&call->lock);
130 if (call->conn)
131 conn = rxrpc_get_connection_maybe(call->conn);
132 spin_unlock_bh(&call->lock);
133 if (!conn)
134 return -ECONNRESET;
136 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
137 if (!pkt) {
138 rxrpc_put_connection(conn);
139 return -ENOMEM;
142 msg.msg_name = &call->peer->srx.transport;
143 msg.msg_namelen = call->peer->srx.transport_len;
144 msg.msg_control = NULL;
145 msg.msg_controllen = 0;
146 msg.msg_flags = 0;
148 pkt->whdr.epoch = htonl(conn->proto.epoch);
149 pkt->whdr.cid = htonl(call->cid);
150 pkt->whdr.callNumber = htonl(call->call_id);
151 pkt->whdr.seq = 0;
152 pkt->whdr.type = RXRPC_PACKET_TYPE_ACK;
153 pkt->whdr.flags = RXRPC_SLOW_START_OK | conn->out_clientflag;
154 pkt->whdr.userStatus = 0;
155 pkt->whdr.securityIndex = call->security_ix;
156 pkt->whdr._rsvd = 0;
157 pkt->whdr.serviceId = htons(call->service_id);
159 spin_lock_bh(&call->lock);
160 if (ping) {
161 reason = RXRPC_ACK_PING;
162 } else {
163 reason = call->ackr_reason;
164 if (!call->ackr_reason) {
165 spin_unlock_bh(&call->lock);
166 ret = 0;
167 goto out;
169 call->ackr_reason = 0;
171 n = rxrpc_fill_out_ack(conn, call, pkt, &hard_ack, &top, reason);
173 spin_unlock_bh(&call->lock);
175 iov[0].iov_base = pkt;
176 iov[0].iov_len = sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
177 iov[1].iov_base = &pkt->ackinfo;
178 iov[1].iov_len = sizeof(pkt->ackinfo);
179 len = iov[0].iov_len + iov[1].iov_len;
181 serial = atomic_inc_return(&conn->serial);
182 pkt->whdr.serial = htonl(serial);
183 trace_rxrpc_tx_ack(call, serial,
184 ntohl(pkt->ack.firstPacket),
185 ntohl(pkt->ack.serial),
186 pkt->ack.reason, pkt->ack.nAcks);
187 if (_serial)
188 *_serial = serial;
190 if (ping) {
191 call->ping_serial = serial;
192 smp_wmb();
193 /* We need to stick a time in before we send the packet in case
194 * the reply gets back before kernel_sendmsg() completes - but
195 * asking UDP to send the packet can take a relatively long
196 * time, so we update the time after, on the assumption that
197 * the packet transmission is more likely to happen towards the
198 * end of the kernel_sendmsg() call.
200 call->ping_time = ktime_get_real();
201 set_bit(RXRPC_CALL_PINGING, &call->flags);
202 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
205 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
206 if (ping)
207 call->ping_time = ktime_get_real();
209 if (call->state < RXRPC_CALL_COMPLETE) {
210 if (ret < 0) {
211 if (ping)
212 clear_bit(RXRPC_CALL_PINGING, &call->flags);
213 rxrpc_propose_ACK(call, pkt->ack.reason,
214 ntohs(pkt->ack.maxSkew),
215 ntohl(pkt->ack.serial),
216 true, true,
217 rxrpc_propose_ack_retry_tx);
218 } else {
219 spin_lock_bh(&call->lock);
220 if (after(hard_ack, call->ackr_consumed))
221 call->ackr_consumed = hard_ack;
222 if (after(top, call->ackr_seen))
223 call->ackr_seen = top;
224 spin_unlock_bh(&call->lock);
227 rxrpc_set_keepalive(call);
230 out:
231 rxrpc_put_connection(conn);
232 kfree(pkt);
233 return ret;
237 * Send an ABORT call packet.
239 int rxrpc_send_abort_packet(struct rxrpc_call *call)
241 struct rxrpc_connection *conn = NULL;
242 struct rxrpc_abort_buffer pkt;
243 struct msghdr msg;
244 struct kvec iov[1];
245 rxrpc_serial_t serial;
246 int ret;
248 /* Don't bother sending aborts for a client call once the server has
249 * hard-ACK'd all of its request data. After that point, we're not
250 * going to stop the operation proceeding, and whilst we might limit
251 * the reply, it's not worth it if we can send a new call on the same
252 * channel instead, thereby closing off this call.
254 if (rxrpc_is_client_call(call) &&
255 test_bit(RXRPC_CALL_TX_LAST, &call->flags))
256 return 0;
258 spin_lock_bh(&call->lock);
259 if (call->conn)
260 conn = rxrpc_get_connection_maybe(call->conn);
261 spin_unlock_bh(&call->lock);
262 if (!conn)
263 return -ECONNRESET;
265 msg.msg_name = &call->peer->srx.transport;
266 msg.msg_namelen = call->peer->srx.transport_len;
267 msg.msg_control = NULL;
268 msg.msg_controllen = 0;
269 msg.msg_flags = 0;
271 pkt.whdr.epoch = htonl(conn->proto.epoch);
272 pkt.whdr.cid = htonl(call->cid);
273 pkt.whdr.callNumber = htonl(call->call_id);
274 pkt.whdr.seq = 0;
275 pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT;
276 pkt.whdr.flags = conn->out_clientflag;
277 pkt.whdr.userStatus = 0;
278 pkt.whdr.securityIndex = call->security_ix;
279 pkt.whdr._rsvd = 0;
280 pkt.whdr.serviceId = htons(call->service_id);
281 pkt.abort_code = htonl(call->abort_code);
283 iov[0].iov_base = &pkt;
284 iov[0].iov_len = sizeof(pkt);
286 serial = atomic_inc_return(&conn->serial);
287 pkt.whdr.serial = htonl(serial);
289 ret = kernel_sendmsg(conn->params.local->socket,
290 &msg, iov, 1, sizeof(pkt));
292 rxrpc_put_connection(conn);
293 return ret;
297 * send a packet through the transport endpoint
299 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
300 bool retrans)
302 struct rxrpc_connection *conn = call->conn;
303 struct rxrpc_wire_header whdr;
304 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
305 struct msghdr msg;
306 struct kvec iov[2];
307 rxrpc_serial_t serial;
308 size_t len;
309 bool lost = false;
310 int ret, opt;
312 _enter(",{%d}", skb->len);
314 /* Each transmission of a Tx packet needs a new serial number */
315 serial = atomic_inc_return(&conn->serial);
317 whdr.epoch = htonl(conn->proto.epoch);
318 whdr.cid = htonl(call->cid);
319 whdr.callNumber = htonl(call->call_id);
320 whdr.seq = htonl(sp->hdr.seq);
321 whdr.serial = htonl(serial);
322 whdr.type = RXRPC_PACKET_TYPE_DATA;
323 whdr.flags = sp->hdr.flags;
324 whdr.userStatus = 0;
325 whdr.securityIndex = call->security_ix;
326 whdr._rsvd = htons(sp->hdr._rsvd);
327 whdr.serviceId = htons(call->service_id);
329 if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) &&
330 sp->hdr.seq == 1)
331 whdr.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE;
333 iov[0].iov_base = &whdr;
334 iov[0].iov_len = sizeof(whdr);
335 iov[1].iov_base = skb->head;
336 iov[1].iov_len = skb->len;
337 len = iov[0].iov_len + iov[1].iov_len;
339 msg.msg_name = &call->peer->srx.transport;
340 msg.msg_namelen = call->peer->srx.transport_len;
341 msg.msg_control = NULL;
342 msg.msg_controllen = 0;
343 msg.msg_flags = 0;
345 /* If our RTT cache needs working on, request an ACK. Also request
346 * ACKs if a DATA packet appears to have been lost.
348 if (!(sp->hdr.flags & RXRPC_LAST_PACKET) &&
349 (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
350 retrans ||
351 call->cong_mode == RXRPC_CALL_SLOW_START ||
352 (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
353 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
354 ktime_get_real())))
355 whdr.flags |= RXRPC_REQUEST_ACK;
357 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
358 static int lose;
359 if ((lose++ & 7) == 7) {
360 ret = 0;
361 lost = true;
362 goto done;
366 _proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
368 /* send the packet with the don't fragment bit set if we currently
369 * think it's small enough */
370 if (iov[1].iov_len >= call->peer->maxdata)
371 goto send_fragmentable;
373 down_read(&conn->params.local->defrag_sem);
374 /* send the packet by UDP
375 * - returns -EMSGSIZE if UDP would have to fragment the packet
376 * to go out of the interface
377 * - in which case, we'll have processed the ICMP error
378 * message and update the peer record
380 ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
382 up_read(&conn->params.local->defrag_sem);
383 if (ret == -EMSGSIZE)
384 goto send_fragmentable;
386 done:
387 trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
388 retrans, lost);
389 if (ret >= 0) {
390 ktime_t now = ktime_get_real();
391 skb->tstamp = now;
392 smp_wmb();
393 sp->hdr.serial = serial;
394 if (whdr.flags & RXRPC_REQUEST_ACK) {
395 call->peer->rtt_last_req = now;
396 trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
397 if (call->peer->rtt_usage > 1) {
398 unsigned long nowj = jiffies, ack_lost_at;
400 ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
401 if (ack_lost_at < 1)
402 ack_lost_at = 1;
404 ack_lost_at += nowj;
405 WRITE_ONCE(call->ack_lost_at, ack_lost_at);
406 rxrpc_reduce_call_timer(call, ack_lost_at, nowj,
407 rxrpc_timer_set_for_lost_ack);
412 rxrpc_set_keepalive(call);
414 _leave(" = %d [%u]", ret, call->peer->maxdata);
415 return ret;
417 send_fragmentable:
418 /* attempt to send this message with fragmentation enabled */
419 _debug("send fragment");
421 down_write(&conn->params.local->defrag_sem);
423 switch (conn->params.local->srx.transport.family) {
424 case AF_INET:
425 opt = IP_PMTUDISC_DONT;
426 ret = kernel_setsockopt(conn->params.local->socket,
427 SOL_IP, IP_MTU_DISCOVER,
428 (char *)&opt, sizeof(opt));
429 if (ret == 0) {
430 ret = kernel_sendmsg(conn->params.local->socket, &msg,
431 iov, 2, len);
433 opt = IP_PMTUDISC_DO;
434 kernel_setsockopt(conn->params.local->socket, SOL_IP,
435 IP_MTU_DISCOVER,
436 (char *)&opt, sizeof(opt));
438 break;
440 #ifdef CONFIG_AF_RXRPC_IPV6
441 case AF_INET6:
442 opt = IPV6_PMTUDISC_DONT;
443 ret = kernel_setsockopt(conn->params.local->socket,
444 SOL_IPV6, IPV6_MTU_DISCOVER,
445 (char *)&opt, sizeof(opt));
446 if (ret == 0) {
447 ret = kernel_sendmsg(conn->params.local->socket, &msg,
448 iov, 2, len);
450 opt = IPV6_PMTUDISC_DO;
451 kernel_setsockopt(conn->params.local->socket,
452 SOL_IPV6, IPV6_MTU_DISCOVER,
453 (char *)&opt, sizeof(opt));
455 break;
456 #endif
459 up_write(&conn->params.local->defrag_sem);
460 goto done;
464 * reject packets through the local endpoint
466 void rxrpc_reject_packets(struct rxrpc_local *local)
468 struct sockaddr_rxrpc srx;
469 struct rxrpc_skb_priv *sp;
470 struct rxrpc_wire_header whdr;
471 struct sk_buff *skb;
472 struct msghdr msg;
473 struct kvec iov[2];
474 size_t size;
475 __be32 code;
477 _enter("%d", local->debug_id);
479 iov[0].iov_base = &whdr;
480 iov[0].iov_len = sizeof(whdr);
481 iov[1].iov_base = &code;
482 iov[1].iov_len = sizeof(code);
483 size = sizeof(whdr) + sizeof(code);
485 msg.msg_name = &srx.transport;
486 msg.msg_control = NULL;
487 msg.msg_controllen = 0;
488 msg.msg_flags = 0;
490 memset(&whdr, 0, sizeof(whdr));
491 whdr.type = RXRPC_PACKET_TYPE_ABORT;
493 while ((skb = skb_dequeue(&local->reject_queue))) {
494 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
495 sp = rxrpc_skb(skb);
497 if (rxrpc_extract_addr_from_skb(local, &srx, skb) == 0) {
498 msg.msg_namelen = srx.transport_len;
500 code = htonl(skb->priority);
502 whdr.epoch = htonl(sp->hdr.epoch);
503 whdr.cid = htonl(sp->hdr.cid);
504 whdr.callNumber = htonl(sp->hdr.callNumber);
505 whdr.serviceId = htons(sp->hdr.serviceId);
506 whdr.flags = sp->hdr.flags;
507 whdr.flags ^= RXRPC_CLIENT_INITIATED;
508 whdr.flags &= RXRPC_CLIENT_INITIATED;
510 kernel_sendmsg(local->socket, &msg, iov, 2, size);
513 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
516 _leave("");