xref: /dragonfly/usr.bin/systat/netbw.c (revision fa71f50a)
1 /*
2  * Copyright (c) 2013 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 #include <sys/param.h>
35 #include <sys/queue.h>
36 #include <sys/tree.h>
37 #include <sys/socket.h>
38 #include <sys/socketvar.h>
39 #include <sys/protosw.h>
40 #include <sys/sysctl.h>
41 
42 #include <netinet/in.h>
43 #include <arpa/inet.h>
44 #include <net/route.h>
45 #include <netinet/in_systm.h>
46 #include <netinet/ip.h>
47 #ifdef INET6
48 #include <netinet/ip6.h>
49 #endif
50 #include <netinet/in_pcb.h>
51 #include <netinet/ip_icmp.h>
52 #include <netinet/icmp_var.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/tcp.h>
55 #include <netinet/tcpip.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/tcp_fsm.h>
58 #include <netinet/tcp_timer.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/tcp_debug.h>
61 #include <netinet/udp.h>
62 #include <netinet/udp_var.h>
63 
64 #include <err.h>
65 #include <errno.h>
66 #include <netdb.h>
67 #include <stdlib.h>
68 #include <string.h>
69 #include <nlist.h>
70 #include <paths.h>
71 #include "systat.h"
72 #include "extern.h"
73 
74 struct mytcpcb {
75 	RB_ENTRY(mytcpcb)	rb_node;
76 	int			seq;
77 	struct xtcpcb		xtcp;
78 	struct xtcpcb		last_xtcp;
79 };
80 
81 static int
82 mytcpcb_cmp(struct mytcpcb *tcp1, struct mytcpcb *tcp2)
83 {
84 	int r;
85 
86 	/*
87 	 * Low local or foreign port comes first (local has priority).
88 	 */
89 	if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_lport) >= 1024 &&
90 	    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_lport) >= 1024) {
91 		if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_fport) <
92 		    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_fport))
93 			return(-1);
94 		if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_fport) >
95 		    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_fport))
96 			return(1);
97 	}
98 
99 	if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_lport) <
100 	    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_lport))
101 		return(-1);
102 	if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_lport) >
103 	    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_lport))
104 		return(1);
105 	if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_fport) <
106 	    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_fport))
107 		return(-1);
108 	if (ntohs(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_fport) >
109 	    ntohs(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_fport))
110 		return(1);
111 
112 	/*
113 	 * Sort IPV4 vs IPV6 addresses
114 	 */
115 	if ((tcp1->xtcp.xt_inp.inp_vflag & (INP_IPV4|INP_IPV6)) <
116 	    (tcp2->xtcp.xt_inp.inp_vflag & (INP_IPV4|INP_IPV6)))
117 		return(-1);
118 	if ((tcp1->xtcp.xt_inp.inp_vflag & (INP_IPV4|INP_IPV6)) >
119 	    (tcp2->xtcp.xt_inp.inp_vflag & (INP_IPV4|INP_IPV6)))
120 		return(1);
121 
122 	/*
123 	 * Local and foreign addresses
124 	 */
125 	if (tcp1->xtcp.xt_inp.inp_vflag & INP_IPV4) {
126 		if (ntohl(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_laddr.s_addr) <
127 		    ntohl(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_laddr.s_addr))
128 			return(-1);
129 		if (ntohl(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_laddr.s_addr) >
130 		    ntohl(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_laddr.s_addr))
131 			return(1);
132 		if (ntohl(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_faddr.s_addr) <
133 		    ntohl(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_faddr.s_addr))
134 			return(-1);
135 		if (ntohl(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie_faddr.s_addr) >
136 		    ntohl(tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie_faddr.s_addr))
137 			return(1);
138 	} else if (tcp1->xtcp.xt_inp.inp_vflag & INP_IPV6) {
139 		r = bcmp(&tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr,
140 			 &tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr,
141 			 sizeof(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr));
142 		if (r)
143 			return(r);
144 	} else {
145 		r = bcmp(&tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr,
146 			 &tcp2->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr,
147 			 sizeof(tcp1->xtcp.xt_inp.inp_inc.inc_ie.ie6_faddr));
148 		if (r)
149 			return(r);
150 	}
151 	return(0);
152 }
153 
154 struct mytcpcb_tree;
155 RB_HEAD(mytcpcb_tree, mytcpcb);
156 RB_PROTOTYPE(mytcpcb_tree, mytcpcb, rb_node, mytcpcb_cmp);
157 RB_GENERATE(mytcpcb_tree, mytcpcb, rb_node, mytcpcb_cmp);
158 
159 static struct mytcpcb_tree mytcp_tree;
160 static struct timeval tv_curr;
161 static struct timeval tv_last;
162 static struct tcp_stats tcp_curr;
163 static struct tcp_stats tcp_last;
164 static int tcp_pcb_seq;
165 
166 static const char *numtok(double value);
167 static void netbwline(int row, struct mytcpcb *elm, double delta_time);
168 const char * netaddrstr(u_char vflags, union in_dependaddr *depaddr,
169 			u_int16_t port);
170 static void updatepcb(struct xtcpcb *xtcp);
171 
172 #define DELTARATE(field)	\
173 	((double)(tcp_curr.field - tcp_last.field) / delta_time)
174 
175 #define DELTAELM(field)		\
176 	((double)(tcp_seq_diff_t)(elm->xtcp.field -		\
177 				  elm->last_xtcp.field) /	\
178 	 delta_time)
179 
180 #define DELTAELMSCALE(field, scale)		\
181 	((double)((tcp_seq_diff_t)(elm->xtcp.field -		\
182 				   elm->last_xtcp.field) << scale) / \
183          delta_time)
184 
185 WINDOW *
186 opennetbw(void)
187 {
188 	RB_INIT(&mytcp_tree);
189 	return (subwin(stdscr, LINES-0-1, 0, 0, 0));
190 }
191 
192 void
193 closenetbw(WINDOW *w)
194 {
195 	struct mytcpcb *mytcp;
196 
197 	while ((mytcp = RB_ROOT(&mytcp_tree)) != NULL) {
198 		RB_REMOVE(mytcpcb_tree, &mytcp_tree, mytcp);
199 		free(mytcp);
200 	}
201 
202         if (w != NULL) {
203 		wclear(w);
204 		wrefresh(w);
205 		delwin(w);
206 	}
207 }
208 
209 int
210 initnetbw(void)
211 {
212 	return(1);
213 }
214 
215 void
216 fetchnetbw(void)
217 {
218 	struct tcp_stats tcp_array[SMP_MAXCPU];
219 	struct xtcpcb *tcp_pcbs;
220 	size_t npcbs;
221 	size_t len;
222 	size_t i;
223 	size_t j;
224 	size_t ncpus;
225 
226 	/*
227 	 * Extract PCB list
228 	 */
229 	len = 0;
230 	if (sysctlbyname("net.inet.tcp.pcblist", NULL, &len, NULL, 0) < 0)
231 		return;
232 	len += 128 * sizeof(tcp_pcbs[0]);
233 	tcp_pcbs = malloc(len);
234 	if (sysctlbyname("net.inet.tcp.pcblist", tcp_pcbs, &len, NULL, 0) < 0) {
235 		free(tcp_pcbs);
236 		return;
237 	}
238 	npcbs = len / sizeof(tcp_pcbs[0]);
239 	++tcp_pcb_seq;
240 
241 	for (i = 0; i < npcbs; ++i) {
242 		if (tcp_pcbs[i].xt_len != sizeof(tcp_pcbs[0]))
243 			break;
244 		updatepcb(&tcp_pcbs[i]);
245 	}
246 	free(tcp_pcbs);
247 
248 	/*
249 	 * General stats
250 	 */
251 	len = sizeof(tcp_array);
252 	if (sysctlbyname("net.inet.tcp.stats", tcp_array, &len, NULL, 0) < 0)
253 		return;
254 	ncpus = len / sizeof(tcp_array[0]);
255 	tcp_last = tcp_curr;
256 	tv_last = tv_curr;
257 	bzero(&tcp_curr, sizeof(tcp_curr));
258 	gettimeofday(&tv_curr, NULL);
259 
260 	for (i = 0; i < ncpus; ++i) {
261 		for (j = 0; j < sizeof(tcp_curr) / sizeof(u_long); ++j) {
262 			((u_long *)&tcp_curr)[j] +=
263 				((u_long *)&tcp_array[i])[j];
264 		}
265 	}
266 }
267 
268 void
269 labelnetbw(void)
270 {
271 	wmove(wnd, 0, 0);
272 	wclrtobot(wnd);
273 #if 0
274 	mvwaddstr(wnd, 0, LADDR, "Local Address");
275 	mvwaddstr(wnd, 0, FADDR, "Foreign Address");
276 	mvwaddstr(wnd, 0, PROTO, "Proto");
277 	mvwaddstr(wnd, 0, RCVCC, "Recv-Q");
278 	mvwaddstr(wnd, 0, SNDCC, "Send-Q");
279 	mvwaddstr(wnd, 0, STATE, "(state)");
280 #endif
281 }
282 
283 void
284 shownetbw(void)
285 {
286 	double delta_time;
287 	struct mytcpcb *elm;
288 	struct mytcpcb *delm;
289 	int row;
290 
291 	delta_time = (double)(tv_curr.tv_sec - tv_last.tv_sec) - 1.0 +
292 		     (tv_curr.tv_usec + 1000000 - tv_last.tv_usec) / 1e6;
293 	if (delta_time < 0.1)
294 		return;
295 
296 	mvwprintw(wnd, 0, 0,
297 		  "tcp accepts %s connects %s "
298 		  "         rcv %s snd %s rexmit %s",
299 		  numtok(DELTARATE(tcps_accepts)),
300 		  numtok(DELTARATE(tcps_connects) - DELTARATE(tcps_accepts)),
301 		  numtok(DELTARATE(tcps_rcvbyte)),
302 		  numtok(DELTARATE(tcps_sndbyte)),
303 		  numtok(DELTARATE(tcps_sndrexmitbyte)));
304 
305 	row = 2;
306 	delm = NULL;
307 	RB_FOREACH(elm, mytcpcb_tree, &mytcp_tree) {
308 		if (delm) {
309 			RB_REMOVE(mytcpcb_tree, &mytcp_tree, delm);
310 			free(delm);
311 			delm = NULL;
312 		}
313 		if (elm->seq == tcp_pcb_seq &&
314 		    (elm->xtcp.xt_socket.so_rcv.sb_cc ||
315 		     elm->xtcp.xt_socket.so_snd.sb_cc ||
316 		     DELTAELM(xt_tp.snd_max) ||
317 		     DELTAELM(xt_tp.rcv_nxt)
318 		    )) {
319 			if (row < LINES - 3)
320 				netbwline(row, elm, delta_time);
321 			++row;
322 		} else if (elm->seq != tcp_pcb_seq) {
323 			delm = elm;
324 		}
325 	}
326 	if (delm) {
327 		RB_REMOVE(mytcpcb_tree, &mytcp_tree, delm);
328 		free(delm);
329 		delm = NULL;
330 	}
331 	wmove(wnd, row, 0);
332 	wclrtobot(wnd);
333 	mvwprintw(wnd, LINES-2, 0,
334 		  "Rate/sec, "
335 		  "R=rxpend T=txpend N=nodelay T=tstmp "
336 		  "S=sack X=winscale F=fastrec");
337 }
338 
339 static
340 void
341 netbwline(int row, struct mytcpcb *elm, double delta_time)
342 {
343 	mvwprintw(wnd, row, 0,
344 		  "%s %s "
345 		  /*"rxb %s txb %s "*/
346 		  "rcv %s snd %s "
347 		  "[%c%c%c%c%c%c%c]",
348 		  netaddrstr(
349 		    elm->xtcp.xt_inp.inp_vflag,
350 		    &elm->xtcp.xt_inp.inp_inc.inc_ie.
351 			ie_dependladdr,
352 		    ntohs(elm->xtcp.xt_inp.inp_inc.inc_ie.ie_lport)),
353 		  netaddrstr(
354 		    elm->xtcp.xt_inp.inp_vflag,
355 		    &elm->xtcp.xt_inp.inp_inc.inc_ie.
356 			ie_dependfaddr,
357 		    ntohs(elm->xtcp.xt_inp.inp_inc.inc_ie.ie_fport)),
358 		/*
359 		  numtok(elm->xtcp.xt_socket.so_rcv.sb_cc),
360 		  numtok(elm->xtcp.xt_socket.so_snd.sb_cc),
361 		*/
362 		  numtok(DELTAELM(xt_tp.rcv_nxt)),
363 		  numtok(DELTAELM(xt_tp.snd_max)),
364 		  (elm->xtcp.xt_socket.so_rcv.sb_cc > 15000 ?
365 		   'R' : ' '),
366 		  (elm->xtcp.xt_socket.so_snd.sb_cc > 15000 ?
367 		   'T' : ' '),
368 		  ((elm->xtcp.xt_tp.t_flags & TF_NODELAY) ?
369 		   'N' : ' '),
370 		  ((elm->xtcp.xt_tp.t_flags & TF_RCVD_TSTMP) ?
371 		   'T' : ' '),
372 		  ((elm->xtcp.xt_tp.t_flags &
373 		   TF_SACK_PERMITTED) ?
374 		   'S' : ' '),
375 		  ((elm->xtcp.xt_tp.t_flags & TF_RCVD_SCALE) ?
376 		   'X' : ' '),
377 		  ((elm->xtcp.xt_tp.t_flags & TF_FASTRECOVERY) ?
378 				   'F' : ' ')
379 	);
380 	wclrtoeol(wnd);
381 }
382 
383 #if 0
384 int
385 cmdnetbw(const char *cmd __unused, char *args __unused)
386 {
387 	fetchnetbw();
388 	shownetbw();
389 	refresh();
390 
391 	return (0);
392 }
393 #endif
394 
395 #define MAXINDEXES 8
396 
397 static
398 const char *
399 numtok(double value)
400 {
401 	static char buf[MAXINDEXES][32];
402 	static int nexti;
403 	static const char *suffixes[] = { " ", "K", "M", "G", "T", NULL };
404 	int suffix = 0;
405 	const char *fmt;
406 
407 	while (value >= 1000.0 && suffixes[suffix+1]) {
408 		value /= 1000.0;
409 		++suffix;
410 	}
411 	nexti = (nexti + 1) % MAXINDEXES;
412 	if (value < 0.001) {
413 		fmt = "      ";
414 	} else if (value < 1.0) {
415 		fmt = "%5.3f%s";
416 	} else if (value < 10.0) {
417 		fmt = "%5.3f%s";
418 	} else if (value < 100.0) {
419 		fmt = "%5.2f%s";
420 	} else {
421 		fmt = "%5.1f%s";
422 	}
423 	snprintf(buf[nexti], sizeof(buf[nexti]),
424 		 fmt, value, suffixes[suffix]);
425 	return (buf[nexti]);
426 }
427 
428 const char *
429 netaddrstr(u_char vflags, union in_dependaddr *depaddr, u_int16_t port)
430 {
431 	static char buf[MAXINDEXES][64];
432 	static int nexta;
433 	char bufip[64];
434 
435 	nexta = (nexta + 1) % MAXINDEXES;
436 
437 	if (vflags & INP_IPV4) {
438 		snprintf(bufip, sizeof(bufip),
439 			 "%d.%d.%d.%d",
440 			 (ntohl(depaddr->id46_addr.ia46_addr4.s_addr) >> 24) &
441 			  255,
442 			 (ntohl(depaddr->id46_addr.ia46_addr4.s_addr) >> 16) &
443 			  255,
444 			 (ntohl(depaddr->id46_addr.ia46_addr4.s_addr) >> 8) &
445 			  255,
446 			 (ntohl(depaddr->id46_addr.ia46_addr4.s_addr) >> 0) &
447 			  255);
448 		snprintf(buf[nexta], sizeof(buf[nexta]),
449 			 "%15s:%-5d", bufip, port);
450 	} else if (vflags & INP_IPV6) {
451 		snprintf(bufip, sizeof(bufip),
452 			 "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x",
453 			 ntohs(depaddr->id6_addr.s6_addr16[0]),
454 			 ntohs(depaddr->id6_addr.s6_addr16[1]),
455 			 ntohs(depaddr->id6_addr.s6_addr16[2]),
456 			 ntohs(depaddr->id6_addr.s6_addr16[3]),
457 			 ntohs(depaddr->id6_addr.s6_addr16[4]),
458 			 ntohs(depaddr->id6_addr.s6_addr16[5]),
459 			 ntohs(depaddr->id6_addr.s6_addr16[6]),
460 			 ntohs(depaddr->id6_addr.s6_addr16[7]));
461 		snprintf(buf[nexta], sizeof(buf[nexta]),
462 			 "%39s:%-5d", bufip, port);
463 	} else {
464 		snprintf(bufip, sizeof(bufip), "<unknown>");
465 		snprintf(buf[nexta], sizeof(buf[nexta]),
466 			 "%15s:%-5d", bufip, port);
467 	}
468 	return (buf[nexta]);
469 }
470 
471 static
472 void
473 updatepcb(struct xtcpcb *xtcp)
474 {
475 	struct mytcpcb dummy;
476 	struct mytcpcb *elm;
477 
478 	dummy.xtcp = *xtcp;
479 	if ((elm = RB_FIND(mytcpcb_tree, &mytcp_tree, &dummy)) == NULL) {
480 		elm = malloc(sizeof(*elm));
481 		bzero(elm, sizeof(*elm));
482 		elm->xtcp = *xtcp;
483 		elm->last_xtcp = *xtcp;
484 		RB_INSERT(mytcpcb_tree, &mytcp_tree, elm);
485 	} else {
486 		elm->last_xtcp = elm->xtcp;
487 		elm->xtcp = *xtcp;
488 	}
489 	elm->seq = tcp_pcb_seq;
490 }
491