xref: /openbmc/linux/net/rxrpc/output.c (revision 4f6cce39)
1 /* RxRPC packet transmission
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
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.
10  */
11 
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 
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"
21 
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;
28 };
29 
30 struct rxrpc_abort_buffer {
31 	struct rxrpc_wire_header whdr;
32 	__be32 abort_code;
33 };
34 
35 /*
36  * Fill out an ACK packet.
37  */
38 static size_t rxrpc_fill_out_ack(struct rxrpc_call *call,
39 				 struct rxrpc_ack_buffer *pkt,
40 				 rxrpc_seq_t *_hard_ack,
41 				 rxrpc_seq_t *_top,
42 				 u8 reason)
43 {
44 	rxrpc_serial_t serial;
45 	rxrpc_seq_t hard_ack, top, seq;
46 	int ix;
47 	u32 mtu, jmax;
48 	u8 *ackp = pkt->acks;
49 
50 	/* Barrier against rxrpc_input_data(). */
51 	serial = call->ackr_serial;
52 	hard_ack = READ_ONCE(call->rx_hard_ack);
53 	top = smp_load_acquire(&call->rx_top);
54 	*_hard_ack = hard_ack;
55 	*_top = top;
56 
57 	pkt->ack.bufferSpace	= htons(8);
58 	pkt->ack.maxSkew	= htons(call->ackr_skew);
59 	pkt->ack.firstPacket	= htonl(hard_ack + 1);
60 	pkt->ack.previousPacket	= htonl(call->ackr_prev_seq);
61 	pkt->ack.serial		= htonl(serial);
62 	pkt->ack.reason		= reason;
63 	pkt->ack.nAcks		= top - hard_ack;
64 
65 	if (reason == RXRPC_ACK_PING)
66 		pkt->whdr.flags |= RXRPC_REQUEST_ACK;
67 
68 	if (after(top, hard_ack)) {
69 		seq = hard_ack + 1;
70 		do {
71 			ix = seq & RXRPC_RXTX_BUFF_MASK;
72 			if (call->rxtx_buffer[ix])
73 				*ackp++ = RXRPC_ACK_TYPE_ACK;
74 			else
75 				*ackp++ = RXRPC_ACK_TYPE_NACK;
76 			seq++;
77 		} while (before_eq(seq, top));
78 	}
79 
80 	mtu = call->conn->params.peer->if_mtu;
81 	mtu -= call->conn->params.peer->hdrsize;
82 	jmax = (call->nr_jumbo_bad > 3) ? 1 : rxrpc_rx_jumbo_max;
83 	pkt->ackinfo.rxMTU	= htonl(rxrpc_rx_mtu);
84 	pkt->ackinfo.maxMTU	= htonl(mtu);
85 	pkt->ackinfo.rwind	= htonl(call->rx_winsize);
86 	pkt->ackinfo.jumbo_max	= htonl(jmax);
87 
88 	*ackp++ = 0;
89 	*ackp++ = 0;
90 	*ackp++ = 0;
91 	return top - hard_ack + 3;
92 }
93 
94 /*
95  * Send an ACK call packet.
96  */
97 int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping)
98 {
99 	struct rxrpc_connection *conn = NULL;
100 	struct rxrpc_ack_buffer *pkt;
101 	struct msghdr msg;
102 	struct kvec iov[2];
103 	rxrpc_serial_t serial;
104 	rxrpc_seq_t hard_ack, top;
105 	size_t len, n;
106 	int ret;
107 	u8 reason;
108 
109 	spin_lock_bh(&call->lock);
110 	if (call->conn)
111 		conn = rxrpc_get_connection_maybe(call->conn);
112 	spin_unlock_bh(&call->lock);
113 	if (!conn)
114 		return -ECONNRESET;
115 
116 	pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
117 	if (!pkt) {
118 		rxrpc_put_connection(conn);
119 		return -ENOMEM;
120 	}
121 
122 	msg.msg_name	= &call->peer->srx.transport;
123 	msg.msg_namelen	= call->peer->srx.transport_len;
124 	msg.msg_control	= NULL;
125 	msg.msg_controllen = 0;
126 	msg.msg_flags	= 0;
127 
128 	pkt->whdr.epoch		= htonl(conn->proto.epoch);
129 	pkt->whdr.cid		= htonl(call->cid);
130 	pkt->whdr.callNumber	= htonl(call->call_id);
131 	pkt->whdr.seq		= 0;
132 	pkt->whdr.type		= RXRPC_PACKET_TYPE_ACK;
133 	pkt->whdr.flags		= RXRPC_SLOW_START_OK | conn->out_clientflag;
134 	pkt->whdr.userStatus	= 0;
135 	pkt->whdr.securityIndex	= call->security_ix;
136 	pkt->whdr._rsvd		= 0;
137 	pkt->whdr.serviceId	= htons(call->service_id);
138 
139 	spin_lock_bh(&call->lock);
140 	if (ping) {
141 		reason = RXRPC_ACK_PING;
142 	} else {
143 		reason = call->ackr_reason;
144 		if (!call->ackr_reason) {
145 			spin_unlock_bh(&call->lock);
146 			ret = 0;
147 			goto out;
148 		}
149 		call->ackr_reason = 0;
150 	}
151 	n = rxrpc_fill_out_ack(call, pkt, &hard_ack, &top, reason);
152 
153 	spin_unlock_bh(&call->lock);
154 
155 	iov[0].iov_base	= pkt;
156 	iov[0].iov_len	= sizeof(pkt->whdr) + sizeof(pkt->ack) + n;
157 	iov[1].iov_base = &pkt->ackinfo;
158 	iov[1].iov_len	= sizeof(pkt->ackinfo);
159 	len = iov[0].iov_len + iov[1].iov_len;
160 
161 	serial = atomic_inc_return(&conn->serial);
162 	pkt->whdr.serial = htonl(serial);
163 	trace_rxrpc_tx_ack(call, serial,
164 			   ntohl(pkt->ack.firstPacket),
165 			   ntohl(pkt->ack.serial),
166 			   pkt->ack.reason, pkt->ack.nAcks);
167 
168 	if (ping) {
169 		call->ping_serial = serial;
170 		smp_wmb();
171 		/* We need to stick a time in before we send the packet in case
172 		 * the reply gets back before kernel_sendmsg() completes - but
173 		 * asking UDP to send the packet can take a relatively long
174 		 * time, so we update the time after, on the assumption that
175 		 * the packet transmission is more likely to happen towards the
176 		 * end of the kernel_sendmsg() call.
177 		 */
178 		call->ping_time = ktime_get_real();
179 		set_bit(RXRPC_CALL_PINGING, &call->flags);
180 		trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_ping, serial);
181 	}
182 
183 	ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
184 	if (ping)
185 		call->ping_time = ktime_get_real();
186 
187 	if (call->state < RXRPC_CALL_COMPLETE) {
188 		if (ret < 0) {
189 			if (ping)
190 				clear_bit(RXRPC_CALL_PINGING, &call->flags);
191 			rxrpc_propose_ACK(call, pkt->ack.reason,
192 					  ntohs(pkt->ack.maxSkew),
193 					  ntohl(pkt->ack.serial),
194 					  true, true,
195 					  rxrpc_propose_ack_retry_tx);
196 		} else {
197 			spin_lock_bh(&call->lock);
198 			if (after(hard_ack, call->ackr_consumed))
199 				call->ackr_consumed = hard_ack;
200 			if (after(top, call->ackr_seen))
201 				call->ackr_seen = top;
202 			spin_unlock_bh(&call->lock);
203 		}
204 	}
205 
206 out:
207 	rxrpc_put_connection(conn);
208 	kfree(pkt);
209 	return ret;
210 }
211 
212 /*
213  * Send an ABORT call packet.
214  */
215 int rxrpc_send_abort_packet(struct rxrpc_call *call)
216 {
217 	struct rxrpc_connection *conn = NULL;
218 	struct rxrpc_abort_buffer pkt;
219 	struct msghdr msg;
220 	struct kvec iov[1];
221 	rxrpc_serial_t serial;
222 	int ret;
223 
224 	spin_lock_bh(&call->lock);
225 	if (call->conn)
226 		conn = rxrpc_get_connection_maybe(call->conn);
227 	spin_unlock_bh(&call->lock);
228 	if (!conn)
229 		return -ECONNRESET;
230 
231 	msg.msg_name	= &call->peer->srx.transport;
232 	msg.msg_namelen	= call->peer->srx.transport_len;
233 	msg.msg_control	= NULL;
234 	msg.msg_controllen = 0;
235 	msg.msg_flags	= 0;
236 
237 	pkt.whdr.epoch		= htonl(conn->proto.epoch);
238 	pkt.whdr.cid		= htonl(call->cid);
239 	pkt.whdr.callNumber	= htonl(call->call_id);
240 	pkt.whdr.seq		= 0;
241 	pkt.whdr.type		= RXRPC_PACKET_TYPE_ABORT;
242 	pkt.whdr.flags		= conn->out_clientflag;
243 	pkt.whdr.userStatus	= 0;
244 	pkt.whdr.securityIndex	= call->security_ix;
245 	pkt.whdr._rsvd		= 0;
246 	pkt.whdr.serviceId	= htons(call->service_id);
247 	pkt.abort_code		= htonl(call->abort_code);
248 
249 	iov[0].iov_base	= &pkt;
250 	iov[0].iov_len	= sizeof(pkt);
251 
252 	serial = atomic_inc_return(&conn->serial);
253 	pkt.whdr.serial = htonl(serial);
254 
255 	ret = kernel_sendmsg(conn->params.local->socket,
256 			     &msg, iov, 1, sizeof(pkt));
257 
258 	rxrpc_put_connection(conn);
259 	return ret;
260 }
261 
262 /*
263  * send a packet through the transport endpoint
264  */
265 int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
266 			   bool retrans)
267 {
268 	struct rxrpc_connection *conn = call->conn;
269 	struct rxrpc_wire_header whdr;
270 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
271 	struct msghdr msg;
272 	struct kvec iov[2];
273 	rxrpc_serial_t serial;
274 	size_t len;
275 	bool lost = false;
276 	int ret, opt;
277 
278 	_enter(",{%d}", skb->len);
279 
280 	/* Each transmission of a Tx packet needs a new serial number */
281 	serial = atomic_inc_return(&conn->serial);
282 
283 	whdr.epoch	= htonl(conn->proto.epoch);
284 	whdr.cid	= htonl(call->cid);
285 	whdr.callNumber	= htonl(call->call_id);
286 	whdr.seq	= htonl(sp->hdr.seq);
287 	whdr.serial	= htonl(serial);
288 	whdr.type	= RXRPC_PACKET_TYPE_DATA;
289 	whdr.flags	= sp->hdr.flags;
290 	whdr.userStatus	= 0;
291 	whdr.securityIndex = call->security_ix;
292 	whdr._rsvd	= htons(sp->hdr._rsvd);
293 	whdr.serviceId	= htons(call->service_id);
294 
295 	iov[0].iov_base = &whdr;
296 	iov[0].iov_len = sizeof(whdr);
297 	iov[1].iov_base = skb->head;
298 	iov[1].iov_len = skb->len;
299 	len = iov[0].iov_len + iov[1].iov_len;
300 
301 	msg.msg_name = &call->peer->srx.transport;
302 	msg.msg_namelen = call->peer->srx.transport_len;
303 	msg.msg_control = NULL;
304 	msg.msg_controllen = 0;
305 	msg.msg_flags = 0;
306 
307 	/* If our RTT cache needs working on, request an ACK.  Also request
308 	 * ACKs if a DATA packet appears to have been lost.
309 	 */
310 	if (!(sp->hdr.flags & RXRPC_LAST_PACKET) &&
311 	    (retrans ||
312 	     call->cong_mode == RXRPC_CALL_SLOW_START ||
313 	     (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
314 	     ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
315 			  ktime_get_real())))
316 		whdr.flags |= RXRPC_REQUEST_ACK;
317 
318 	if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
319 		static int lose;
320 		if ((lose++ & 7) == 7) {
321 			ret = 0;
322 			lost = true;
323 			goto done;
324 		}
325 	}
326 
327 	_proto("Tx DATA %%%u { #%u }", serial, sp->hdr.seq);
328 
329 	/* send the packet with the don't fragment bit set if we currently
330 	 * think it's small enough */
331 	if (iov[1].iov_len >= call->peer->maxdata)
332 		goto send_fragmentable;
333 
334 	down_read(&conn->params.local->defrag_sem);
335 	/* send the packet by UDP
336 	 * - returns -EMSGSIZE if UDP would have to fragment the packet
337 	 *   to go out of the interface
338 	 *   - in which case, we'll have processed the ICMP error
339 	 *     message and update the peer record
340 	 */
341 	ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
342 
343 	up_read(&conn->params.local->defrag_sem);
344 	if (ret == -EMSGSIZE)
345 		goto send_fragmentable;
346 
347 done:
348 	trace_rxrpc_tx_data(call, sp->hdr.seq, serial, whdr.flags,
349 			    retrans, lost);
350 	if (ret >= 0) {
351 		ktime_t now = ktime_get_real();
352 		skb->tstamp = now;
353 		smp_wmb();
354 		sp->hdr.serial = serial;
355 		if (whdr.flags & RXRPC_REQUEST_ACK) {
356 			call->peer->rtt_last_req = now;
357 			trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
358 		}
359 	}
360 	_leave(" = %d [%u]", ret, call->peer->maxdata);
361 	return ret;
362 
363 send_fragmentable:
364 	/* attempt to send this message with fragmentation enabled */
365 	_debug("send fragment");
366 
367 	down_write(&conn->params.local->defrag_sem);
368 
369 	switch (conn->params.local->srx.transport.family) {
370 	case AF_INET:
371 		opt = IP_PMTUDISC_DONT;
372 		ret = kernel_setsockopt(conn->params.local->socket,
373 					SOL_IP, IP_MTU_DISCOVER,
374 					(char *)&opt, sizeof(opt));
375 		if (ret == 0) {
376 			ret = kernel_sendmsg(conn->params.local->socket, &msg,
377 					     iov, 2, len);
378 
379 			opt = IP_PMTUDISC_DO;
380 			kernel_setsockopt(conn->params.local->socket, SOL_IP,
381 					  IP_MTU_DISCOVER,
382 					  (char *)&opt, sizeof(opt));
383 		}
384 		break;
385 
386 #ifdef CONFIG_AF_RXRPC_IPV6
387 	case AF_INET6:
388 		opt = IPV6_PMTUDISC_DONT;
389 		ret = kernel_setsockopt(conn->params.local->socket,
390 					SOL_IPV6, IPV6_MTU_DISCOVER,
391 					(char *)&opt, sizeof(opt));
392 		if (ret == 0) {
393 			ret = kernel_sendmsg(conn->params.local->socket, &msg,
394 					     iov, 1, iov[0].iov_len);
395 
396 			opt = IPV6_PMTUDISC_DO;
397 			kernel_setsockopt(conn->params.local->socket,
398 					  SOL_IPV6, IPV6_MTU_DISCOVER,
399 					  (char *)&opt, sizeof(opt));
400 		}
401 		break;
402 #endif
403 	}
404 
405 	up_write(&conn->params.local->defrag_sem);
406 	goto done;
407 }
408 
409 /*
410  * reject packets through the local endpoint
411  */
412 void rxrpc_reject_packets(struct rxrpc_local *local)
413 {
414 	struct sockaddr_rxrpc srx;
415 	struct rxrpc_skb_priv *sp;
416 	struct rxrpc_wire_header whdr;
417 	struct sk_buff *skb;
418 	struct msghdr msg;
419 	struct kvec iov[2];
420 	size_t size;
421 	__be32 code;
422 
423 	_enter("%d", local->debug_id);
424 
425 	iov[0].iov_base = &whdr;
426 	iov[0].iov_len = sizeof(whdr);
427 	iov[1].iov_base = &code;
428 	iov[1].iov_len = sizeof(code);
429 	size = sizeof(whdr) + sizeof(code);
430 
431 	msg.msg_name = &srx.transport;
432 	msg.msg_control = NULL;
433 	msg.msg_controllen = 0;
434 	msg.msg_flags = 0;
435 
436 	memset(&whdr, 0, sizeof(whdr));
437 	whdr.type = RXRPC_PACKET_TYPE_ABORT;
438 
439 	while ((skb = skb_dequeue(&local->reject_queue))) {
440 		rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
441 		sp = rxrpc_skb(skb);
442 
443 		if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
444 			msg.msg_namelen = srx.transport_len;
445 
446 			code = htonl(skb->priority);
447 
448 			whdr.epoch	= htonl(sp->hdr.epoch);
449 			whdr.cid	= htonl(sp->hdr.cid);
450 			whdr.callNumber	= htonl(sp->hdr.callNumber);
451 			whdr.serviceId	= htons(sp->hdr.serviceId);
452 			whdr.flags	= sp->hdr.flags;
453 			whdr.flags	^= RXRPC_CLIENT_INITIATED;
454 			whdr.flags	&= RXRPC_CLIENT_INITIATED;
455 
456 			kernel_sendmsg(local->socket, &msg, iov, 2, size);
457 		}
458 
459 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
460 	}
461 
462 	_leave("");
463 }
464