xref: /freebsd-13-stable/sys/net/debugnet.c (revision 41e4389a3d0df7c31fbe2c80e2f00219c1b5bd6b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2019 Isilon Systems, LLC.
5  * Copyright (c) 2005-2014 Sandvine Incorporated. All rights reserved.
6  * Copyright (c) 2000 Darrell Anderson
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/endian.h>
38 #include <sys/errno.h>
39 #include <sys/eventhandler.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 
46 #ifdef DDB
47 #include <ddb/ddb.h>
48 #include <ddb/db_lex.h>
49 #endif
50 
51 #include <net/ethernet.h>
52 #include <net/if.h>
53 #include <net/if_arp.h>
54 #include <net/if_dl.h>
55 #include <net/if_types.h>
56 #include <net/if_var.h>
57 #include <net/vnet.h>
58 #include <net/route.h>
59 #include <net/route/nhop.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_fib.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_options.h>
68 #include <netinet/udp.h>
69 #include <netinet/udp_var.h>
70 
71 #include <machine/in_cksum.h>
72 #include <machine/pcb.h>
73 
74 #include <net/debugnet.h>
75 #define	DEBUGNET_INTERNAL
76 #include <net/debugnet_int.h>
77 
78 FEATURE(debugnet, "Debugnet support");
79 
80 SYSCTL_NODE(_net, OID_AUTO, debugnet, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
81     "debugnet parameters");
82 
83 unsigned debugnet_debug;
84 SYSCTL_UINT(_net_debugnet, OID_AUTO, debug, CTLFLAG_RWTUN,
85     &debugnet_debug, 0,
86     "Debug message verbosity (0: off; 1: on; 2: verbose)");
87 
88 int debugnet_npolls = 2000;
89 SYSCTL_INT(_net_debugnet, OID_AUTO, npolls, CTLFLAG_RWTUN,
90     &debugnet_npolls, 0,
91     "Number of times to poll before assuming packet loss (0.5ms per poll)");
92 int debugnet_nretries = 10;
93 SYSCTL_INT(_net_debugnet, OID_AUTO, nretries, CTLFLAG_RWTUN,
94     &debugnet_nretries, 0,
95     "Number of retransmit attempts before giving up");
96 int debugnet_fib = RT_DEFAULT_FIB;
97 SYSCTL_INT(_net_debugnet, OID_AUTO, fib, CTLFLAG_RWTUN,
98     &debugnet_fib, 0,
99     "Fib to use when sending dump");
100 
101 static bool g_debugnet_pcb_inuse;
102 static struct debugnet_pcb g_dnet_pcb;
103 
104 /*
105  * Simple accessors for opaque PCB.
106  */
107 const unsigned char *
debugnet_get_gw_mac(const struct debugnet_pcb * pcb)108 debugnet_get_gw_mac(const struct debugnet_pcb *pcb)
109 {
110 	MPASS(g_debugnet_pcb_inuse && pcb == &g_dnet_pcb &&
111 	    pcb->dp_state >= DN_STATE_HAVE_GW_MAC);
112 	return (pcb->dp_gw_mac.octet);
113 }
114 
115 /*
116  * Start of network primitives, beginning with output primitives.
117  */
118 
119 /*
120  * Handles creation of the ethernet header, then places outgoing packets into
121  * the tx buffer for the NIC
122  *
123  * Parameters:
124  *	m	The mbuf containing the packet to be sent (will be freed by
125  *		this function or the NIC driver)
126  *	ifp	The interface to send on
127  *	dst	The destination ethernet address (source address will be looked
128  *		up using ifp)
129  *	etype	The ETHERTYPE_* value for the protocol that is being sent
130  *
131  * Returns:
132  *	int	see errno.h, 0 for success
133  */
134 int
debugnet_ether_output(struct mbuf * m,struct ifnet * ifp,struct ether_addr dst,u_short etype)135 debugnet_ether_output(struct mbuf *m, struct ifnet *ifp, struct ether_addr dst,
136     u_short etype)
137 {
138 	struct ether_header *eh;
139 
140 	if (((ifp->if_flags & (IFF_MONITOR | IFF_UP)) != IFF_UP) ||
141 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING) {
142 		if_printf(ifp, "%s: interface isn't up\n", __func__);
143 		m_freem(m);
144 		return (ENETDOWN);
145 	}
146 
147 	/* Fill in the ethernet header. */
148 	M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
149 	if (m == NULL) {
150 		printf("%s: out of mbufs\n", __func__);
151 		return (ENOBUFS);
152 	}
153 	eh = mtod(m, struct ether_header *);
154 	memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN);
155 	memcpy(eh->ether_dhost, dst.octet, ETHER_ADDR_LEN);
156 	eh->ether_type = htons(etype);
157 	return (ifp->if_debugnet_methods->dn_transmit(ifp, m));
158 }
159 
160 /*
161  * Unreliable transmission of an mbuf chain to the debugnet server
162  * Note: can't handle fragmentation; fails if the packet is larger than
163  *	 ifp->if_mtu after adding the UDP/IP headers
164  *
165  * Parameters:
166  *	pcb	The debugnet context block
167  *	m	mbuf chain
168  *
169  * Returns:
170  *	int	see errno.h, 0 for success
171  */
172 static int
debugnet_udp_output(struct debugnet_pcb * pcb,struct mbuf * m)173 debugnet_udp_output(struct debugnet_pcb *pcb, struct mbuf *m)
174 {
175 	struct udphdr *udp;
176 
177 	MPASS(pcb->dp_state >= DN_STATE_HAVE_GW_MAC);
178 
179 	M_PREPEND(m, sizeof(*udp), M_NOWAIT);
180 	if (m == NULL) {
181 		printf("%s: out of mbufs\n", __func__);
182 		return (ENOBUFS);
183 	}
184 
185 	udp = mtod(m, void *);
186 	udp->uh_ulen = htons(m->m_pkthdr.len);
187 	/* Use this src port so that the server can connect() the socket */
188 	udp->uh_sport = htons(pcb->dp_client_port);
189 	udp->uh_dport = htons(pcb->dp_server_port);
190 	/* Computed later (protocol-dependent). */
191 	udp->uh_sum = 0;
192 
193 	return (debugnet_ip_output(pcb, m));
194 }
195 
196 int
debugnet_ack_output(struct debugnet_pcb * pcb,uint32_t seqno)197 debugnet_ack_output(struct debugnet_pcb *pcb, uint32_t seqno /* net endian */)
198 {
199 	struct debugnet_ack *dn_ack;
200 	struct mbuf *m;
201 
202 	DNETDEBUG("Acking with seqno %u\n", ntohl(seqno));
203 
204 	m = m_gethdr(M_NOWAIT, MT_DATA);
205 	if (m == NULL) {
206 		printf("%s: Out of mbufs\n", __func__);
207 		return (ENOBUFS);
208 	}
209 	m->m_len = sizeof(*dn_ack);
210 	m->m_pkthdr.len = sizeof(*dn_ack);
211 	MH_ALIGN(m, sizeof(*dn_ack));
212 	dn_ack = mtod(m, void *);
213 	dn_ack->da_seqno = seqno;
214 
215 	return (debugnet_udp_output(pcb, m));
216 }
217 
218 /*
219  * Dummy free function for debugnet clusters.
220  */
221 static void
debugnet_mbuf_free(struct mbuf * m __unused)222 debugnet_mbuf_free(struct mbuf *m __unused)
223 {
224 }
225 
226 /*
227  * Construct and reliably send a debugnet packet.  May fail from a resource
228  * shortage or extreme number of unacknowledged retransmissions.  Wait for
229  * an acknowledgement before returning.  Splits packets into chunks small
230  * enough to be sent without fragmentation (looks up the interface MTU)
231  *
232  * Parameters:
233  *	type	debugnet packet type (HERALD, FINISHED, ...)
234  *	data	data
235  *	datalen	data size (bytes)
236  *	auxdata	optional auxiliary information
237  *
238  * Returns:
239  *	int see errno.h, 0 for success
240  */
241 int
debugnet_send(struct debugnet_pcb * pcb,uint32_t type,const void * data,uint32_t datalen,const struct debugnet_proto_aux * auxdata)242 debugnet_send(struct debugnet_pcb *pcb, uint32_t type, const void *data,
243     uint32_t datalen, const struct debugnet_proto_aux *auxdata)
244 {
245 	struct debugnet_msg_hdr *dn_msg_hdr;
246 	struct mbuf *m, *m2;
247 	uint64_t want_acks;
248 	uint32_t i, pktlen, sent_so_far;
249 	int retries, polls, error;
250 
251 	if (pcb->dp_state == DN_STATE_REMOTE_CLOSED)
252 		return (ECONNRESET);
253 
254 	want_acks = 0;
255 	pcb->dp_rcvd_acks = 0;
256 	retries = 0;
257 
258 retransmit:
259 	/* Chunks can be too big to fit in packets. */
260 	for (i = sent_so_far = 0; sent_so_far < datalen ||
261 	    (i == 0 && datalen == 0); i++) {
262 		pktlen = datalen - sent_so_far;
263 
264 		/* Bound: the interface MTU (assume no IP options). */
265 		pktlen = min(pktlen, pcb->dp_ifp->if_mtu -
266 		    sizeof(struct udpiphdr) - sizeof(struct debugnet_msg_hdr));
267 
268 		/*
269 		 * Check if it is retransmitting and this has been ACKed
270 		 * already.
271 		 */
272 		if ((pcb->dp_rcvd_acks & (1 << i)) != 0) {
273 			sent_so_far += pktlen;
274 			continue;
275 		}
276 
277 		/*
278 		 * Get and fill a header mbuf, then chain data as an extended
279 		 * mbuf.
280 		 */
281 		m = m_gethdr(M_NOWAIT, MT_DATA);
282 		if (m == NULL) {
283 			printf("%s: Out of mbufs\n", __func__);
284 			return (ENOBUFS);
285 		}
286 		m->m_len = sizeof(struct debugnet_msg_hdr);
287 		m->m_pkthdr.len = sizeof(struct debugnet_msg_hdr);
288 		MH_ALIGN(m, sizeof(struct debugnet_msg_hdr));
289 		dn_msg_hdr = mtod(m, struct debugnet_msg_hdr *);
290 		dn_msg_hdr->mh_seqno = htonl(pcb->dp_seqno + i);
291 		dn_msg_hdr->mh_type = htonl(type);
292 		dn_msg_hdr->mh_len = htonl(pktlen);
293 
294 		if (auxdata != NULL) {
295 			dn_msg_hdr->mh_offset =
296 			    htobe64(auxdata->dp_offset_start + sent_so_far);
297 			dn_msg_hdr->mh_aux2 = htobe32(auxdata->dp_aux2);
298 		} else {
299 			dn_msg_hdr->mh_offset = htobe64(sent_so_far);
300 			dn_msg_hdr->mh_aux2 = 0;
301 		}
302 
303 		if (pktlen != 0) {
304 			m2 = m_get(M_NOWAIT, MT_DATA);
305 			if (m2 == NULL) {
306 				m_freem(m);
307 				printf("%s: Out of mbufs\n", __func__);
308 				return (ENOBUFS);
309 			}
310 			MEXTADD(m2, __DECONST(char *, data) + sent_so_far,
311 			    pktlen, debugnet_mbuf_free, NULL, NULL, 0,
312 			    EXT_DISPOSABLE);
313 			m2->m_len = pktlen;
314 
315 			m_cat(m, m2);
316 			m->m_pkthdr.len += pktlen;
317 		}
318 		error = debugnet_udp_output(pcb, m);
319 		if (error != 0)
320 			return (error);
321 
322 		/* Note that we're waiting for this packet in the bitfield. */
323 		want_acks |= (1 << i);
324 		sent_so_far += pktlen;
325 	}
326 	if (i >= DEBUGNET_MAX_IN_FLIGHT)
327 		printf("Warning: Sent more than %d packets (%d). "
328 		    "Acknowledgements will fail unless the size of "
329 		    "rcvd_acks/want_acks is increased.\n",
330 		    DEBUGNET_MAX_IN_FLIGHT, i);
331 
332 	/*
333 	 * Wait for acks.  A *real* window would speed things up considerably.
334 	 */
335 	polls = 0;
336 	while (pcb->dp_rcvd_acks != want_acks) {
337 		if (polls++ > debugnet_npolls) {
338 			if (retries++ > debugnet_nretries)
339 				return (ETIMEDOUT);
340 			printf(". ");
341 			goto retransmit;
342 		}
343 		debugnet_network_poll(pcb);
344 		DELAY(500);
345 		if (pcb->dp_state == DN_STATE_REMOTE_CLOSED)
346 			return (ECONNRESET);
347 	}
348 	pcb->dp_seqno += i;
349 	return (0);
350 }
351 
352 /*
353  * Network input primitives.
354  */
355 
356 /*
357  * Just introspect the header enough to fire off a seqno ack and validate
358  * length fits.
359  */
360 static void
debugnet_handle_rx_msg(struct debugnet_pcb * pcb,struct mbuf ** mb)361 debugnet_handle_rx_msg(struct debugnet_pcb *pcb, struct mbuf **mb)
362 {
363 	const struct debugnet_msg_hdr *dnh;
364 	struct mbuf *m;
365 	int error;
366 
367 	m = *mb;
368 
369 	if (m->m_pkthdr.len < sizeof(*dnh)) {
370 		DNETDEBUG("ignoring small debugnet_msg packet\n");
371 		return;
372 	}
373 
374 	/* Get ND header. */
375 	if (m->m_len < sizeof(*dnh)) {
376 		m = m_pullup(m, sizeof(*dnh));
377 		*mb = m;
378 		if (m == NULL) {
379 			DNETDEBUG("m_pullup failed\n");
380 			return;
381 		}
382 	}
383 	dnh = mtod(m, const void *);
384 
385 	if (ntohl(dnh->mh_len) + sizeof(*dnh) > m->m_pkthdr.len) {
386 		DNETDEBUG("Dropping short packet.\n");
387 		return;
388 	}
389 
390 	/*
391 	 * If the issue is transient (ENOBUFS), sender should resend.  If
392 	 * non-transient (like driver objecting to rx -> tx from the same
393 	 * thread), not much else we can do.
394 	 */
395 	error = debugnet_ack_output(pcb, dnh->mh_seqno);
396 	if (error != 0)
397 		return;
398 
399 	if (ntohl(dnh->mh_type) == DEBUGNET_FINISHED) {
400 		printf("Remote shut down the connection on us!\n");
401 		pcb->dp_state = DN_STATE_REMOTE_CLOSED;
402 
403 		/*
404 		 * Continue through to the user handler so they are signalled
405 		 * not to wait for further rx.
406 		 */
407 	}
408 
409 	pcb->dp_rx_handler(pcb, mb);
410 }
411 
412 static void
debugnet_handle_ack(struct debugnet_pcb * pcb,struct mbuf ** mb,uint16_t sport)413 debugnet_handle_ack(struct debugnet_pcb *pcb, struct mbuf **mb, uint16_t sport)
414 {
415 	const struct debugnet_ack *dn_ack;
416 	struct mbuf *m;
417 	uint32_t rcv_ackno;
418 
419 	m = *mb;
420 
421 	/* Get Ack. */
422 	if (m->m_len < sizeof(*dn_ack)) {
423 		m = m_pullup(m, sizeof(*dn_ack));
424 		*mb = m;
425 		if (m == NULL) {
426 			DNETDEBUG("m_pullup failed\n");
427 			return;
428 		}
429 	}
430 	dn_ack = mtod(m, const void *);
431 
432 	/* Debugnet processing. */
433 	/*
434 	 * Packet is meant for us.  Extract the ack sequence number and the
435 	 * port number if necessary.
436 	 */
437 	rcv_ackno = ntohl(dn_ack->da_seqno);
438 	if (pcb->dp_state < DN_STATE_GOT_HERALD_PORT) {
439 		pcb->dp_server_port = sport;
440 		pcb->dp_state = DN_STATE_GOT_HERALD_PORT;
441 	}
442 	if (rcv_ackno >= pcb->dp_seqno + DEBUGNET_MAX_IN_FLIGHT)
443 		printf("%s: ACK %u too far in future!\n", __func__, rcv_ackno);
444 	else if (rcv_ackno >= pcb->dp_seqno) {
445 		/* We're interested in this ack. Record it. */
446 		pcb->dp_rcvd_acks |= 1 << (rcv_ackno - pcb->dp_seqno);
447 	}
448 }
449 
450 void
debugnet_handle_udp(struct debugnet_pcb * pcb,struct mbuf ** mb)451 debugnet_handle_udp(struct debugnet_pcb *pcb, struct mbuf **mb)
452 {
453 	const struct udphdr *udp;
454 	struct mbuf *m;
455 	uint16_t sport, ulen;
456 
457 	/* UDP processing. */
458 
459 	m = *mb;
460 	if (m->m_pkthdr.len < sizeof(*udp)) {
461 		DNETDEBUG("ignoring small UDP packet\n");
462 		return;
463 	}
464 
465 	/* Get UDP headers. */
466 	if (m->m_len < sizeof(*udp)) {
467 		m = m_pullup(m, sizeof(*udp));
468 		*mb = m;
469 		if (m == NULL) {
470 			DNETDEBUG("m_pullup failed\n");
471 			return;
472 		}
473 	}
474 	udp = mtod(m, const void *);
475 
476 	/* We expect to receive UDP packets on the configured client port. */
477 	if (ntohs(udp->uh_dport) != pcb->dp_client_port) {
478 		DNETDEBUG("not on the expected port.\n");
479 		return;
480 	}
481 
482 	/* Check that ulen does not exceed actual size of data. */
483 	ulen = ntohs(udp->uh_ulen);
484 	if (m->m_pkthdr.len < ulen) {
485 		DNETDEBUG("ignoring runt UDP packet\n");
486 		return;
487 	}
488 
489 	sport = ntohs(udp->uh_sport);
490 
491 	m_adj(m, sizeof(*udp));
492 	ulen -= sizeof(*udp);
493 
494 	if (ulen == sizeof(struct debugnet_ack)) {
495 		debugnet_handle_ack(pcb, mb, sport);
496 		return;
497 	}
498 
499 	if (pcb->dp_rx_handler == NULL) {
500 		if (ulen < sizeof(struct debugnet_ack))
501 			DNETDEBUG("ignoring small ACK packet\n");
502 		else
503 			DNETDEBUG("ignoring unexpected non-ACK packet on "
504 			    "half-duplex connection.\n");
505 		return;
506 	}
507 
508 	debugnet_handle_rx_msg(pcb, mb);
509 }
510 
511 /*
512  * Handler for incoming packets directly from the network adapter
513  * Identifies the packet type (IP or ARP) and passes it along to one of the
514  * helper functions debugnet_handle_ip or debugnet_handle_arp.
515  *
516  * It needs to partially replicate the behaviour of ether_input() and
517  * ether_demux().
518  *
519  * Parameters:
520  *	ifp	the interface the packet came from
521  *	m	an mbuf containing the packet received
522  */
523 static void
debugnet_input_one(struct ifnet * ifp,struct mbuf * m)524 debugnet_input_one(struct ifnet *ifp, struct mbuf *m)
525 {
526 	struct ifreq ifr;
527 	struct ether_header *eh;
528 	u_short etype;
529 
530 	/* Ethernet processing. */
531 	if ((m->m_flags & M_PKTHDR) == 0) {
532 		DNETDEBUG_IF(ifp, "discard frame without packet header\n");
533 		goto done;
534 	}
535 	if (m->m_len < ETHER_HDR_LEN) {
536 		DNETDEBUG_IF(ifp,
537 	    "discard frame without leading eth header (len %d pktlen %d)\n",
538 		    m->m_len, m->m_pkthdr.len);
539 		goto done;
540 	}
541 	if ((m->m_flags & M_HASFCS) != 0) {
542 		m_adj(m, -ETHER_CRC_LEN);
543 		m->m_flags &= ~M_HASFCS;
544 	}
545 	eh = mtod(m, struct ether_header *);
546 	etype = ntohs(eh->ether_type);
547 	if ((m->m_flags & M_VLANTAG) != 0 || etype == ETHERTYPE_VLAN) {
548 		DNETDEBUG_IF(ifp, "ignoring vlan packets\n");
549 		goto done;
550 	}
551 	if (if_gethwaddr(ifp, &ifr) != 0) {
552 		DNETDEBUG_IF(ifp, "failed to get hw addr for interface\n");
553 		goto done;
554 	}
555 	if (memcmp(ifr.ifr_addr.sa_data, eh->ether_dhost,
556 	    ETHER_ADDR_LEN) != 0 &&
557 	    (etype != ETHERTYPE_ARP || !ETHER_IS_BROADCAST(eh->ether_dhost))) {
558 		DNETDEBUG_IF(ifp,
559 		    "discard frame with incorrect destination addr\n");
560 		goto done;
561 	}
562 
563 	MPASS(g_debugnet_pcb_inuse);
564 
565 	/* Done ethernet processing. Strip off the ethernet header. */
566 	m_adj(m, ETHER_HDR_LEN);
567 	switch (etype) {
568 	case ETHERTYPE_ARP:
569 		debugnet_handle_arp(&g_dnet_pcb, &m);
570 		break;
571 	case ETHERTYPE_IP:
572 		debugnet_handle_ip(&g_dnet_pcb, &m);
573 		break;
574 	default:
575 		DNETDEBUG_IF(ifp, "dropping unknown ethertype %hu\n", etype);
576 		break;
577 	}
578 done:
579 	if (m != NULL)
580 		m_freem(m);
581 }
582 
583 static void
debugnet_input(struct ifnet * ifp,struct mbuf * m)584 debugnet_input(struct ifnet *ifp, struct mbuf *m)
585 {
586 	struct mbuf *n;
587 
588 	do {
589 		n = m->m_nextpkt;
590 		m->m_nextpkt = NULL;
591 		debugnet_input_one(ifp, m);
592 		m = n;
593 	} while (m != NULL);
594 }
595 
596 /*
597  * Network polling primitive.
598  *
599  * Instead of assuming that most of the network stack is sane, we just poll the
600  * driver directly for packets.
601  */
602 void
debugnet_network_poll(struct debugnet_pcb * pcb)603 debugnet_network_poll(struct debugnet_pcb *pcb)
604 {
605 	struct ifnet *ifp;
606 
607 	ifp = pcb->dp_ifp;
608 	ifp->if_debugnet_methods->dn_poll(ifp, 1000);
609 }
610 
611 /*
612  * Start of consumer API surface.
613  */
614 void
debugnet_free(struct debugnet_pcb * pcb)615 debugnet_free(struct debugnet_pcb *pcb)
616 {
617 	struct ifnet *ifp;
618 
619 	MPASS(pcb == &g_dnet_pcb);
620 	MPASS(pcb->dp_drv_input == NULL || g_debugnet_pcb_inuse);
621 
622 	ifp = pcb->dp_ifp;
623 	if (ifp != NULL) {
624 		if (pcb->dp_drv_input != NULL)
625 			ifp->if_input = pcb->dp_drv_input;
626 		if (pcb->dp_event_started)
627 			ifp->if_debugnet_methods->dn_event(ifp, DEBUGNET_END);
628 	}
629 	debugnet_mbuf_finish();
630 
631 	g_debugnet_pcb_inuse = false;
632 	memset(&g_dnet_pcb, 0xfd, sizeof(g_dnet_pcb));
633 }
634 
635 int
debugnet_connect(const struct debugnet_conn_params * dcp,struct debugnet_pcb ** pcb_out)636 debugnet_connect(const struct debugnet_conn_params *dcp,
637     struct debugnet_pcb **pcb_out)
638 {
639 	struct debugnet_proto_aux herald_auxdata;
640 	struct debugnet_pcb *pcb;
641 	struct ifnet *ifp;
642 	int error;
643 
644 	if (g_debugnet_pcb_inuse) {
645 		printf("%s: Only one connection at a time.\n", __func__);
646 		return (EBUSY);
647 	}
648 
649 	pcb = &g_dnet_pcb;
650 	*pcb = (struct debugnet_pcb) {
651 		.dp_state = DN_STATE_INIT,
652 		.dp_client = dcp->dc_client,
653 		.dp_server = dcp->dc_server,
654 		.dp_gateway = dcp->dc_gateway,
655 		.dp_server_port = dcp->dc_herald_port,	/* Initially */
656 		.dp_client_port = dcp->dc_client_port,
657 		.dp_seqno = 1,
658 		.dp_ifp = dcp->dc_ifp,
659 		.dp_rx_handler = dcp->dc_rx_handler,
660 		.dp_drv_input = NULL,
661 	};
662 
663 	/* Switch to the debugnet mbuf zones. */
664 	debugnet_mbuf_start();
665 
666 	/* At least one needed parameter is missing; infer it. */
667 	if (pcb->dp_client == INADDR_ANY || pcb->dp_gateway == INADDR_ANY ||
668 	    pcb->dp_ifp == NULL) {
669 		struct sockaddr_in dest_sin, *gw_sin, *local_sin;
670 		struct ifnet *rt_ifp;
671 		struct nhop_object *nh;
672 
673 		memset(&dest_sin, 0, sizeof(dest_sin));
674 		dest_sin = (struct sockaddr_in) {
675 			.sin_len = sizeof(dest_sin),
676 			.sin_family = AF_INET,
677 			.sin_addr.s_addr = pcb->dp_server,
678 		};
679 
680 		CURVNET_SET(vnet0);
681 		nh = fib4_lookup_debugnet(debugnet_fib, dest_sin.sin_addr, 0,
682 		    NHR_NONE);
683 		CURVNET_RESTORE();
684 
685 		if (nh == NULL) {
686 			printf("%s: Could not get route for that server.\n",
687 			    __func__);
688 			error = ENOENT;
689 			goto cleanup;
690 		}
691 
692 		/* TODO support AF_INET6 */
693 		if (nh->gw_sa.sa_family == AF_INET)
694 			gw_sin = &nh->gw4_sa;
695 		else {
696 			if (nh->gw_sa.sa_family == AF_LINK)
697 				DNETDEBUG("Destination address is on link.\n");
698 			gw_sin = NULL;
699 		}
700 
701 		MPASS(nh->nh_ifa->ifa_addr->sa_family == AF_INET);
702 		local_sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
703 
704 		rt_ifp = nh->nh_ifp;
705 
706 		if (pcb->dp_client == INADDR_ANY)
707 			pcb->dp_client = local_sin->sin_addr.s_addr;
708 		if (pcb->dp_gateway == INADDR_ANY && gw_sin != NULL)
709 			pcb->dp_gateway = gw_sin->sin_addr.s_addr;
710 		if (pcb->dp_ifp == NULL)
711 			pcb->dp_ifp = rt_ifp;
712 	}
713 
714 	ifp = pcb->dp_ifp;
715 
716 	if (debugnet_debug > 0) {
717 		char serbuf[INET_ADDRSTRLEN], clibuf[INET_ADDRSTRLEN],
718 		    gwbuf[INET_ADDRSTRLEN];
719 		inet_ntop(AF_INET, &pcb->dp_server, serbuf, sizeof(serbuf));
720 		inet_ntop(AF_INET, &pcb->dp_client, clibuf, sizeof(clibuf));
721 		if (pcb->dp_gateway != INADDR_ANY)
722 			inet_ntop(AF_INET, &pcb->dp_gateway, gwbuf, sizeof(gwbuf));
723 		DNETDEBUG("Connecting to %s:%d%s%s from %s:%d on %s\n",
724 		    serbuf, pcb->dp_server_port,
725 		    (pcb->dp_gateway == INADDR_ANY) ? "" : " via ",
726 		    (pcb->dp_gateway == INADDR_ANY) ? "" : gwbuf,
727 		    clibuf, pcb->dp_client_port, if_name(ifp));
728 	}
729 
730 	/* Validate iface is online and supported. */
731 	if (!DEBUGNET_SUPPORTED_NIC(ifp)) {
732 		printf("%s: interface '%s' does not support debugnet\n",
733 		    __func__, if_name(ifp));
734 		error = ENODEV;
735 		goto cleanup;
736 	}
737 	if ((if_getflags(ifp) & IFF_UP) == 0) {
738 		printf("%s: interface '%s' link is down\n", __func__,
739 		    if_name(ifp));
740 		error = ENXIO;
741 		goto cleanup;
742 	}
743 
744 	ifp->if_debugnet_methods->dn_event(ifp, DEBUGNET_START);
745 	pcb->dp_event_started = true;
746 
747 	/*
748 	 * We maintain the invariant that g_debugnet_pcb_inuse is always true
749 	 * while the debugnet ifp's if_input is overridden with
750 	 * debugnet_input().
751 	 */
752 	g_debugnet_pcb_inuse = true;
753 
754 	/* Make the card use *our* receive callback. */
755 	pcb->dp_drv_input = ifp->if_input;
756 	ifp->if_input = debugnet_input;
757 
758 	printf("%s: searching for %s MAC...\n", __func__,
759 	    (dcp->dc_gateway == INADDR_ANY) ? "server" : "gateway");
760 
761 	error = debugnet_arp_gw(pcb);
762 	if (error != 0) {
763 		printf("%s: failed to locate MAC address\n", __func__);
764 		goto cleanup;
765 	}
766 	MPASS(pcb->dp_state == DN_STATE_HAVE_GW_MAC);
767 
768 	herald_auxdata = (struct debugnet_proto_aux) {
769 		.dp_offset_start = dcp->dc_herald_offset,
770 		.dp_aux2 = dcp->dc_herald_aux2,
771 	};
772 	error = debugnet_send(pcb, DEBUGNET_HERALD, dcp->dc_herald_data,
773 	    dcp->dc_herald_datalen, &herald_auxdata);
774 	if (error != 0) {
775 		printf("%s: failed to herald debugnet server\n", __func__);
776 		goto cleanup;
777 	}
778 
779 	*pcb_out = pcb;
780 	return (0);
781 
782 cleanup:
783 	debugnet_free(pcb);
784 	return (error);
785 }
786 
787 /*
788  * Pre-allocated dump-time mbuf tracking.
789  *
790  * We just track the high water mark we've ever seen and allocate appropriately
791  * for that iface/mtu combo.
792  */
793 static struct {
794 	int nmbuf;
795 	int ncl;
796 	int clsize;
797 } dn_hwm;
798 static struct mtx dn_hwm_lk;
799 MTX_SYSINIT(debugnet_hwm_lock, &dn_hwm_lk, "Debugnet HWM lock", MTX_DEF);
800 
801 static void
dn_maybe_reinit_mbufs(int nmbuf,int ncl,int clsize)802 dn_maybe_reinit_mbufs(int nmbuf, int ncl, int clsize)
803 {
804 	bool any;
805 
806 	any = false;
807 	mtx_lock(&dn_hwm_lk);
808 
809 	if (nmbuf > dn_hwm.nmbuf) {
810 		any = true;
811 		dn_hwm.nmbuf = nmbuf;
812 	} else
813 		nmbuf = dn_hwm.nmbuf;
814 
815 	if (ncl > dn_hwm.ncl) {
816 		any = true;
817 		dn_hwm.ncl = ncl;
818 	} else
819 		ncl = dn_hwm.ncl;
820 
821 	if (clsize > dn_hwm.clsize) {
822 		any = true;
823 		dn_hwm.clsize = clsize;
824 	} else
825 		clsize = dn_hwm.clsize;
826 
827 	mtx_unlock(&dn_hwm_lk);
828 
829 	if (any)
830 		debugnet_mbuf_reinit(nmbuf, ncl, clsize);
831 }
832 
833 void
debugnet_any_ifnet_update(struct ifnet * ifp)834 debugnet_any_ifnet_update(struct ifnet *ifp)
835 {
836 	int clsize, nmbuf, ncl, nrxr;
837 
838 	if (!DEBUGNET_SUPPORTED_NIC(ifp))
839 		return;
840 
841 	ifp->if_debugnet_methods->dn_init(ifp, &nrxr, &ncl, &clsize);
842 	KASSERT(nrxr > 0, ("invalid receive ring count %d", nrxr));
843 
844 	/*
845 	 * We need two headers per message on the transmit side. Multiply by
846 	 * four to give us some breathing room.
847 	 */
848 	nmbuf = ncl * (4 + nrxr);
849 	ncl *= nrxr;
850 
851 	/*
852 	 * Bandaid for drivers that (incorrectly) advertise LinkUp before their
853 	 * dn_init method is available.
854 	 */
855 	if (nmbuf == 0 || ncl == 0 || clsize == 0) {
856 #ifndef INVARIANTS
857 		if (bootverbose)
858 #endif
859 		printf("%s: Bad dn_init result from %s (ifp %p), ignoring.\n",
860 		    __func__, if_name(ifp), ifp);
861 		return;
862 	}
863 	dn_maybe_reinit_mbufs(nmbuf, ncl, clsize);
864 }
865 
866 /*
867  * Unfortunately, the ifnet_arrival_event eventhandler hook is mostly useless
868  * for us because drivers tend to if_attach before invoking DEBUGNET_SET().
869  *
870  * On the other hand, hooking DEBUGNET_SET() itself may still be too early,
871  * because the driver is still in attach.  Since we cannot use down interfaces,
872  * maybe hooking ifnet_event:IFNET_EVENT_UP is sufficient?  ... Nope, at least
873  * with vtnet and dhcpclient that event just never occurs.
874  *
875  * So that's how I've landed on the lower level ifnet_link_event.
876  */
877 
878 static void
dn_ifnet_event(void * arg __unused,struct ifnet * ifp,int link_state)879 dn_ifnet_event(void *arg __unused, struct ifnet *ifp, int link_state)
880 {
881 	if (link_state == LINK_STATE_UP)
882 		debugnet_any_ifnet_update(ifp);
883 }
884 
885 static eventhandler_tag dn_attach_cookie;
886 static void
dn_evh_init(void * ctx __unused)887 dn_evh_init(void *ctx __unused)
888 {
889 	dn_attach_cookie = EVENTHANDLER_REGISTER(ifnet_link_event,
890 	    dn_ifnet_event, NULL, EVENTHANDLER_PRI_ANY);
891 }
892 SYSINIT(dn_evh_init, SI_SUB_EVENTHANDLER + 1, SI_ORDER_ANY, dn_evh_init, NULL);
893 
894 /*
895  * DDB parsing helpers for debugnet(4) consumers.
896  */
897 #ifdef DDB
898 struct my_inet_opt {
899 	bool has_opt;
900 	const char *printname;
901 	in_addr_t *result;
902 };
903 
904 static int
dn_parse_optarg_ipv4(struct my_inet_opt * opt)905 dn_parse_optarg_ipv4(struct my_inet_opt *opt)
906 {
907 	in_addr_t tmp;
908 	unsigned octet;
909 	int t;
910 
911 	tmp = 0;
912 	for (octet = 0; octet < 4; octet++) {
913 		t = db_read_token_flags(DRT_WSPACE | DRT_DECIMAL);
914 		if (t != tNUMBER) {
915 			db_printf("%s:%s: octet %u expected number; found %d\n",
916 			    __func__, opt->printname, octet, t);
917 			return (EINVAL);
918 		}
919 		/*
920 		 * db_lex lexes '-' distinctly from the number itself, but
921 		 * let's document that invariant.
922 		 */
923 		MPASS(db_tok_number >= 0);
924 
925 		if (db_tok_number > UINT8_MAX) {
926 			db_printf("%s:%s: octet %u out of range: %jd\n", __func__,
927 			    opt->printname, octet, (intmax_t)db_tok_number);
928 			return (EDOM);
929 		}
930 
931 		/* Constructed host-endian and converted to network later. */
932 		tmp = (tmp << 8) | db_tok_number;
933 
934 		if (octet < 3) {
935 			t = db_read_token_flags(DRT_WSPACE);
936 			if (t != tDOT) {
937 				db_printf("%s:%s: octet %u expected '.'; found"
938 				    " %d\n", __func__, opt->printname, octet,
939 				    t);
940 				return (EINVAL);
941 			}
942 		}
943 	}
944 
945 	*opt->result = htonl(tmp);
946 	opt->has_opt = true;
947 	return (0);
948 }
949 
950 int
debugnet_parse_ddb_cmd(const char * cmd,struct debugnet_ddb_config * result)951 debugnet_parse_ddb_cmd(const char *cmd, struct debugnet_ddb_config *result)
952 {
953 	struct ifnet *ifp;
954 	int t, error;
955 	bool want_ifp;
956 	char ch;
957 
958 	struct my_inet_opt opt_client = {
959 		.printname = "client",
960 		.result = &result->dd_client,
961 	},
962 	opt_server = {
963 		.printname = "server",
964 		.result = &result->dd_server,
965 	},
966 	opt_gateway = {
967 		.printname = "gateway",
968 		.result = &result->dd_gateway,
969 	},
970 	*cur_inet_opt;
971 
972 	ifp = NULL;
973 	memset(result, 0, sizeof(*result));
974 
975 	/*
976 	 * command [space] [-] [opt] [[space] [optarg]] ...
977 	 *
978 	 * db_command has already lexed 'command' for us.
979 	 */
980 	t = db_read_token_flags(DRT_WSPACE);
981 	if (t == tWSPACE)
982 		t = db_read_token_flags(DRT_WSPACE);
983 
984 	while (t != tEOL) {
985 		if (t != tMINUS) {
986 			db_printf("%s: Bad syntax; expected '-', got %d\n",
987 			    cmd, t);
988 			goto usage;
989 		}
990 
991 		t = db_read_token_flags(DRT_WSPACE);
992 		if (t != tIDENT) {
993 			db_printf("%s: Bad syntax; expected tIDENT, got %d\n",
994 			    cmd, t);
995 			goto usage;
996 		}
997 
998 		if (strlen(db_tok_string) > 1) {
999 			db_printf("%s: Bad syntax; expected single option "
1000 			    "flag, got '%s'\n", cmd, db_tok_string);
1001 			goto usage;
1002 		}
1003 
1004 		want_ifp = false;
1005 		cur_inet_opt = NULL;
1006 		switch ((ch = db_tok_string[0])) {
1007 		default:
1008 			DNETDEBUG("Unexpected: '%c'\n", ch);
1009 			/* FALLTHROUGH */
1010 		case 'h':
1011 			goto usage;
1012 		case 'c':
1013 			cur_inet_opt = &opt_client;
1014 			break;
1015 		case 'g':
1016 			cur_inet_opt = &opt_gateway;
1017 			break;
1018 		case 's':
1019 			cur_inet_opt = &opt_server;
1020 			break;
1021 		case 'i':
1022 			want_ifp = true;
1023 			break;
1024 		}
1025 
1026 		t = db_read_token_flags(DRT_WSPACE);
1027 		if (t != tWSPACE) {
1028 			db_printf("%s: Bad syntax; expected space after "
1029 			    "flag %c, got %d\n", cmd, ch, t);
1030 			goto usage;
1031 		}
1032 
1033 		if (want_ifp) {
1034 			t = db_read_token_flags(DRT_WSPACE);
1035 			if (t != tIDENT) {
1036 				db_printf("%s: Expected interface but got %d\n",
1037 				    cmd, t);
1038 				goto usage;
1039 			}
1040 
1041 			CURVNET_SET(vnet0);
1042 			/*
1043 			 * We *don't* take a ref here because the only current
1044 			 * consumer, db_netdump_cmd, does not need it.  It
1045 			 * (somewhat redundantly) extracts the if_name(),
1046 			 * re-lookups the ifp, and takes its own reference.
1047 			 */
1048 			ifp = ifunit(db_tok_string);
1049 			CURVNET_RESTORE();
1050 			if (ifp == NULL) {
1051 				db_printf("Could not locate interface %s\n",
1052 				    db_tok_string);
1053 				error = ENOENT;
1054 				goto cleanup;
1055 			}
1056 		} else {
1057 			MPASS(cur_inet_opt != NULL);
1058 			/* Assume IPv4 for now. */
1059 			error = dn_parse_optarg_ipv4(cur_inet_opt);
1060 			if (error != 0)
1061 				goto cleanup;
1062 		}
1063 
1064 		/* Skip (mandatory) whitespace after option, if not EOL. */
1065 		t = db_read_token_flags(DRT_WSPACE);
1066 		if (t == tEOL)
1067 			break;
1068 		if (t != tWSPACE) {
1069 			db_printf("%s: Bad syntax; expected space after "
1070 			    "flag %c option; got %d\n", cmd, ch, t);
1071 			goto usage;
1072 		}
1073 		t = db_read_token_flags(DRT_WSPACE);
1074 	}
1075 
1076 	if (!opt_server.has_opt) {
1077 		db_printf("%s: need a destination server address\n", cmd);
1078 		goto usage;
1079 	}
1080 
1081 	result->dd_has_client = opt_client.has_opt;
1082 	result->dd_has_gateway = opt_gateway.has_opt;
1083 	result->dd_ifp = ifp;
1084 
1085 	/* We parsed the full line to tEOL already, or bailed with an error. */
1086 	return (0);
1087 
1088 usage:
1089 	db_printf("Usage: %s -s <server> [-g <gateway> -c <localip> "
1090 	    "-i <interface>]\n", cmd);
1091 	error = EINVAL;
1092 	/* FALLTHROUGH */
1093 cleanup:
1094 	db_skip_to_eol();
1095 	return (error);
1096 }
1097 #endif /* DDB */
1098