xref: /titanic_44/usr/src/cmd/stat/iostat/iostat.c (revision f169c0eae91b2ee787cf8d6dcf8edd9159d4c9e2)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  *
26  * rewritten from UCB 4.13 83/09/25
27  * rewritten from SunOS 4.1 SID 1.18 89/10/06
28  */
29 
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <stdarg.h>
33 #include <ctype.h>
34 #include <unistd.h>
35 #include <memory.h>
36 #include <errno.h>
37 #include <string.h>
38 #include <signal.h>
39 #include <sys/types.h>
40 #include <time.h>
41 #include <sys/time.h>
42 #include <sys/sysinfo.h>
43 #include <inttypes.h>
44 #include <strings.h>
45 #include <sys/systeminfo.h>
46 #include <kstat.h>
47 #include <locale.h>
48 
49 #include "dsr.h"
50 #include "statcommon.h"
51 
52 #define	DISK_OLD		0x0001
53 #define	DISK_NEW		0x0002
54 #define	DISK_EXTENDED		0x0004
55 #define	DISK_ERRORS		0x0008
56 #define	DISK_EXTENDED_ERRORS	0x0010
57 #define	DISK_IOPATH_LI		0x0020	/* LunInitiator */
58 #define	DISK_IOPATH_LTI		0x0040	/* LunTargetInitiator */
59 
60 #define	DISK_NORMAL		(DISK_OLD | DISK_NEW)
61 #define	DISK_IO_MASK		(DISK_OLD | DISK_NEW | DISK_EXTENDED)
62 #define	DISK_ERROR_MASK		(DISK_ERRORS | DISK_EXTENDED_ERRORS)
63 #define	PRINT_VERTICAL		(DISK_ERROR_MASK | DISK_EXTENDED)
64 
65 #define	REPRINT 19
66 
67 #define	NUMBER_OF_ERR_COUNTERS	3
68 
69 /*
70  * It's really a pseudo-gigabyte. We use 1000000000 bytes so that the disk
71  * labels don't look bad. 1GB is really 1073741824 bytes.
72  */
73 #define	DISK_GIGABYTE   1000000000.0
74 
75 /*
76  * Function desciptor to be called when extended
77  * headers are used.
78  */
79 typedef struct formatter {
80 	void (*nfunc)(void);
81 	struct formatter *next;
82 } format_t;
83 
84 /*
85  * Used to get formatting right when printing tty/cpu
86  * data to the right of disk data
87  */
88 enum show_disk_mode {
89 	SHOW_FIRST_ONLY,
90 	SHOW_SECOND_ONWARDS,
91 	SHOW_ALL
92 };
93 
94 enum show_disk_mode show_disk_mode = SHOW_ALL;
95 
96 char *cmdname = "iostat";
97 int caught_cont = 0;
98 
99 static char one_blank[] = " ";
100 static char two_blanks[] = "  ";
101 
102 /*
103  * count for number of lines to be emitted before a header is
104  * shown again. Only used for the basic format.
105  */
106 static	uint_t	tohdr = 1;
107 
108 /*
109  * If we're in raw format, have we printed a header? We only do it
110  * once for raw but we emit it every REPRINT lines in non-raw format.
111  * This applies only for the basic header. The extended header is
112  * done only once in both formats.
113  */
114 static	uint_t	hdr_out;
115 
116 /*
117  * Flags representing arguments from command line
118  */
119 static	uint_t	do_tty;			/* show tty info (-t) */
120 static	uint_t	do_disk;		/* show disk info per selected */
121 					/* format (-d, -D, -e, -E, -x -X -Y) */
122 static	uint_t	do_cpu;			/* show cpu info (-c) */
123 static	uint_t	do_interval;		/* do intervals (-I) */
124 static	int	do_partitions;		/* per-partition stats (-p) */
125 static	int	do_partitions_only;	/* per-partition stats only (-P) */
126 					/* no per-device stats for disks */
127 static	uint_t	do_conversions;		/* display disks as cXtYdZ (-n) */
128 static	uint_t	do_megabytes;		/* display data in MB/sec (-M) */
129 static  uint_t	do_controller;		/* display controller info (-C) */
130 static  uint_t	do_raw;			/* emit raw format (-r) */
131 static	uint_t	timestamp_fmt = NODATE;	/* timestamp  each display (-T) */
132 static	uint_t	do_devid;		/* -E should show devid */
133 
134 /*
135  * Default number of disk drives to be displayed in basic format
136  */
137 #define	DEFAULT_LIMIT	4
138 
139 struct iodev_filter df;
140 
141 static  uint_t	suppress_state;		/* skip state change messages */
142 static	uint_t	suppress_zero;		/* skip zero valued lines */
143 static  uint_t	show_mountpts;		/* show mount points */
144 static	int 	interval;		/* interval (seconds) to output */
145 static	int 	iter;			/* iterations from command line */
146 
147 #define	SMALL_SCRATCH_BUFLEN	MAXNAMELEN
148 
149 static int	iodevs_nl;		/* name field width */
150 #define	IODEVS_NL_MIN		6	/* not too thin for "device" */
151 #define	IODEVS_NL_MAX		24	/* but keep full width under 80 */
152 
153 static	char	disk_header[132];
154 static	uint_t 	dh_len;			/* disk header length for centering */
155 static  int 	lineout;		/* data waiting to be printed? */
156 
157 static struct snapshot *newss;
158 static struct snapshot *oldss;
159 static	double	getime;			/* elapsed time */
160 static	double	percent;		/* 100 / etime */
161 
162 /*
163  * List of functions to be called which will construct the desired output
164  */
165 static format_t	*formatter_list;
166 static format_t *formatter_end;
167 
168 static u_longlong_t	ull_delta(u_longlong_t, u_longlong_t);
169 static uint_t 	u32_delta(uint_t, uint_t);
170 static void setup(void (*nfunc)(void));
171 static void print_tty_hdr1(void);
172 static void print_tty_hdr2(void);
173 static void print_cpu_hdr1(void);
174 static void print_cpu_hdr2(void);
175 static void print_tty_data(void);
176 static void print_cpu_data(void);
177 static void print_err_hdr(void);
178 static void print_disk_header(void);
179 static void hdrout(void);
180 static void disk_errors(void);
181 static void do_newline(void);
182 static void push_out(const char *, ...);
183 static void printhdr(int);
184 static void printxhdr(void);
185 static void usage(void);
186 static void do_args(int, char **);
187 static void do_format(void);
188 static void show_all_disks(void);
189 static void show_first_disk(void);
190 static void show_other_disks(void);
191 static void show_disk_errors(void *, void *, void *);
192 static void write_core_header(void);
193 static int  fzero(double value);
194 static int  safe_strtoi(char const *val, char *errmsg);
195 
196 int
197 main(int argc, char **argv)
198 {
199 	enum snapshot_types types = SNAP_SYSTEM;
200 	kstat_ctl_t *kc;
201 	long hz;
202 	int forever;
203 	hrtime_t start_n;
204 	hrtime_t period_n;
205 
206 	(void) setlocale(LC_ALL, "");
207 #if !defined(TEXT_DOMAIN)		/* Should be defined by cc -D */
208 #define	TEXT_DOMAIN "SYS_TEST"		/* Use this only if it weren't */
209 #endif
210 	(void) textdomain(TEXT_DOMAIN);
211 
212 	do_args(argc, argv);
213 
214 	/*
215 	 * iostat historically showed CPU changes, even though
216 	 * it doesn't provide much useful information
217 	 */
218 	types |= SNAP_CPUS;
219 
220 	if (do_disk)
221 		types |= SNAP_IODEVS;
222 
223 	if (do_disk && !do_partitions_only)
224 		df.if_allowed_types |= IODEV_DISK;
225 	if (do_disk & DISK_IOPATH_LI) {
226 		df.if_allowed_types |= IODEV_IOPATH_LTI;
227 		types |= SNAP_IOPATHS_LI;
228 	}
229 	if (do_disk & DISK_IOPATH_LTI) {
230 		df.if_allowed_types |= IODEV_IOPATH_LTI;
231 		types |= SNAP_IOPATHS_LTI;
232 	}
233 	if (do_disk & DISK_ERROR_MASK)
234 		types |= SNAP_IODEV_ERRORS;
235 	if (do_partitions || do_partitions_only)
236 		df.if_allowed_types |= IODEV_PARTITION;
237 	if (do_conversions)
238 		types |= SNAP_IODEV_PRETTY;
239 	if (do_devid)
240 		types |= SNAP_IODEV_DEVID;
241 	if (do_controller) {
242 		if (!(do_disk & PRINT_VERTICAL) ||
243 		    (do_disk & DISK_EXTENDED_ERRORS))
244 			fail(0, "-C can only be used with -e or -x.");
245 		types |= SNAP_CONTROLLERS;
246 		df.if_allowed_types |= IODEV_CONTROLLER;
247 	}
248 
249 	hz = sysconf(_SC_CLK_TCK);
250 
251 	/*
252 	 * Undocumented behavior - sending a SIGCONT will result
253 	 * in a new header being emitted. Used only if we're not
254 	 * doing extended headers. This is a historical
255 	 * artifact.
256 	 */
257 	if (!(do_disk & PRINT_VERTICAL))
258 		(void) signal(SIGCONT, printhdr);
259 
260 	if (interval)
261 		period_n = (hrtime_t)interval * NANOSEC;
262 
263 	kc = open_kstat();
264 	if (interval)
265 		start_n = gethrtime();
266 	newss = acquire_snapshot(kc, types, &df);
267 
268 	/* compute width of "device" field */
269 	iodevs_nl = newss->s_iodevs_is_name_maxlen;
270 	iodevs_nl = (iodevs_nl < IODEVS_NL_MIN) ?
271 	    IODEVS_NL_MIN : iodevs_nl;
272 	iodevs_nl = (iodevs_nl > IODEVS_NL_MAX) ?
273 	    IODEVS_NL_MAX : iodevs_nl;
274 
275 	do_format();
276 
277 	forever = (iter == 0);
278 	do {
279 		if (do_conversions && show_mountpts)
280 			do_mnttab();
281 
282 		if (do_tty || do_cpu) {
283 			kstat_t *oldks;
284 			oldks = oldss ? &oldss->s_sys.ss_agg_sys : NULL;
285 			getime = cpu_ticks_delta(oldks,
286 			    &newss->s_sys.ss_agg_sys);
287 			percent = (getime > 0.0) ? 100.0 / getime : 0.0;
288 			getime = (getime / nr_active_cpus(newss)) / hz;
289 			if (getime == 0.0)
290 				getime = (double)interval;
291 			if (getime == 0.0 || do_interval)
292 				getime = 1.0;
293 		}
294 
295 		if (formatter_list) {
296 			format_t *tmp;
297 			tmp = formatter_list;
298 
299 			if (timestamp_fmt != NODATE)
300 				print_timestamp(timestamp_fmt);
301 
302 			while (tmp) {
303 				(tmp->nfunc)();
304 				tmp = tmp->next;
305 			}
306 			(void) fflush(stdout);
307 		}
308 
309 		/* only remaining/doing a single iteration, we are done */
310 		if (iter == 1)
311 			continue;
312 
313 		if (interval > 0)
314 			/* Have a kip */
315 			sleep_until(&start_n, period_n, forever, &caught_cont);
316 
317 		free_snapshot(oldss);
318 		oldss = newss;
319 		newss = acquire_snapshot(kc, types, &df);
320 		iodevs_nl = (newss->s_iodevs_is_name_maxlen > iodevs_nl) ?
321 		    newss->s_iodevs_is_name_maxlen : iodevs_nl;
322 		iodevs_nl = (iodevs_nl < IODEVS_NL_MIN) ?
323 		    IODEVS_NL_MIN : iodevs_nl;
324 		iodevs_nl = (iodevs_nl > IODEVS_NL_MAX) ?
325 		    IODEVS_NL_MAX : iodevs_nl;
326 
327 		if (!suppress_state)
328 			snapshot_report_changes(oldss, newss);
329 
330 		/* if config changed, show stats from boot */
331 		if (snapshot_has_changed(oldss, newss)) {
332 			free_snapshot(oldss);
333 			oldss = NULL;
334 		}
335 
336 	} while (--iter);
337 
338 	free_snapshot(oldss);
339 	free_snapshot(newss);
340 	(void) kstat_close(kc);
341 	free(df.if_names);
342 	return (0);
343 }
344 
345 /*
346  * Some magic numbers used in header formatting.
347  *
348  * DISK_LEN = length of either "kps tps serv" or "wps rps util"
349  *	      using 0 as the first position
350  *
351  * DISK_ERROR_LEN = length of "s/w h/w trn tot" with one space on
352  *		either side. Does not use zero as first pos.
353  *
354  * DEVICE_LEN = length of "device" + 1 character.
355  */
356 
357 #define	DISK_LEN	11
358 #define	DISK_ERROR_LEN	16
359 #define	DEVICE_LEN	7
360 
361 /*ARGSUSED*/
362 static void
363 show_disk_name(void *v1, void *v2, void *data)
364 {
365 	struct iodev_snapshot *dev = (struct iodev_snapshot *)v2;
366 	size_t slen;
367 	char *name;
368 	char fbuf[SMALL_SCRATCH_BUFLEN];
369 
370 	if (dev == NULL)
371 		return;
372 
373 	name = do_conversions ? dev->is_pretty : dev->is_name;
374 	name = name ? name : dev->is_name;
375 
376 	if (!do_raw) {
377 		uint_t width;
378 
379 		slen = strlen(name);
380 		/*
381 		 * The length is less
382 		 * than the section
383 		 * which will be displayed
384 		 * on the next line.
385 		 * Center the entry.
386 		 */
387 
388 		width = (DISK_LEN + 1)/2 + (slen / 2);
389 		(void) snprintf(fbuf, sizeof (fbuf),
390 		    "%*s", width, name);
391 		name = fbuf;
392 		push_out("%-13.13s ", name);
393 	} else {
394 		push_out(name);
395 	}
396 }
397 
398 /*ARGSUSED*/
399 static void
400 show_disk_header(void *v1, void *v2, void *data)
401 {
402 	push_out(disk_header);
403 }
404 
405 /*
406  * Write out a two line header. What is written out depends on the flags
407  * selected but in the worst case consists of a tty header, a disk header
408  * providing information for 4 disks and a cpu header.
409  *
410  * The tty header consists of the word "tty" on the first line above the
411  * words "tin tout" on the next line. If present the tty portion consumes
412  * the first 10 characters of each line since "tin tout" is surrounded
413  * by single spaces.
414  *
415  * Each of the disk sections is a 14 character "block" in which the name of
416  * the disk is centered in the first 12 characters of the first line.
417  *
418  * The cpu section is an 11 character block with "cpu" centered over the
419  * section.
420  *
421  * The worst case should look as follows:
422  *
423  * 0---------1--------2---------3---------4---------5---------6---------7-------
424  *    tty        sd0           sd1           sd2           sd3           cpu
425  *  tin tout kps tps serv  kps tps serv  kps tps serv  kps tps serv  us sy wt id
426  *  NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NN NN NN NN
427  *
428  * When -D is specified, the disk header looks as follows (worst case):
429  *
430  * 0---------1--------2---------3---------4---------5---------6---------7-------
431  *     tty        sd0           sd1             sd2          sd3          cpu
432  *   tin tout rps wps util  rps wps util  rps wps util  rps wps util us sy wt id
433  *   NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN NN NN NN NN
434  */
435 static void
436 printhdr(int sig)
437 {
438 	/*
439 	 * If we're here because a signal fired, reenable the
440 	 * signal.
441 	 */
442 	if (sig)
443 		(void) signal(SIGCONT, printhdr);
444 	if (sig == SIGCONT)
445 		caught_cont = 1;
446 	/*
447 	 * Horizontal mode headers
448 	 *
449 	 * First line
450 	 */
451 	if (do_tty)
452 		print_tty_hdr1();
453 
454 	if (do_disk & DISK_NORMAL) {
455 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
456 		    show_disk_name, NULL);
457 	}
458 
459 	if (do_cpu)
460 		print_cpu_hdr1();
461 	do_newline();
462 
463 	/*
464 	 * Second line
465 	 */
466 	if (do_tty)
467 		print_tty_hdr2();
468 
469 	if (do_disk & DISK_NORMAL) {
470 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
471 		    show_disk_header, NULL);
472 	}
473 
474 	if (do_cpu)
475 		print_cpu_hdr2();
476 	do_newline();
477 
478 	tohdr = REPRINT;
479 }
480 
481 /*
482  * Write out the extended header centered over the core information.
483  */
484 static void
485 write_core_header(void)
486 {
487 	char *edev = "extended device statistics";
488 	uint_t lead_space_ct;
489 	uint_t follow_space_ct;
490 	size_t edevlen;
491 
492 	if (do_raw == 0) {
493 		/*
494 		 * The things we do to look nice...
495 		 *
496 		 * Center the core output header. Make sure we have the
497 		 * right number of trailing spaces for follow-on headers
498 		 * (i.e., cpu and/or tty and/or errors).
499 		 */
500 		edevlen = strlen(edev);
501 		lead_space_ct = dh_len - edevlen;
502 		lead_space_ct /= 2;
503 		if (lead_space_ct > 0) {
504 			follow_space_ct = dh_len - (lead_space_ct + edevlen);
505 			if (do_disk & DISK_ERRORS)
506 				follow_space_ct -= DISK_ERROR_LEN;
507 			if ((do_disk & DISK_EXTENDED) && do_conversions)
508 				follow_space_ct -= DEVICE_LEN;
509 
510 			push_out("%1$*2$.*2$s%3$s%4$*5$.*5$s", one_blank,
511 			    lead_space_ct, edev, one_blank, follow_space_ct);
512 		} else
513 			push_out("%56s", edev);
514 	} else
515 		push_out(edev);
516 }
517 
518 /*
519  * In extended mode headers, we don't want to reprint the header on
520  * signals as they are printed every time anyways.
521  */
522 static void
523 printxhdr(void)
524 {
525 
526 	/*
527 	 * Vertical mode headers
528 	 */
529 	if (do_disk & DISK_EXTENDED)
530 		setup(write_core_header);
531 	if (do_disk & DISK_ERRORS)
532 		setup(print_err_hdr);
533 
534 	if (do_conversions) {
535 		setup(do_newline);
536 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
537 			setup(print_disk_header);
538 		setup(do_newline);
539 	} else {
540 		if (do_tty)
541 			setup(print_tty_hdr1);
542 		if (do_cpu)
543 			setup(print_cpu_hdr1);
544 		setup(do_newline);
545 
546 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
547 			setup(print_disk_header);
548 		if (do_tty)
549 			setup(print_tty_hdr2);
550 		if (do_cpu)
551 			setup(print_cpu_hdr2);
552 		setup(do_newline);
553 	}
554 }
555 
556 /*
557  * Write out a line for this disk - note that show_disk writes out
558  * full lines or blocks for each selected disk.
559  */
560 static void
561 show_disk(void *v1, void *v2, void *data)
562 {
563 	uint32_t err_counters[NUMBER_OF_ERR_COUNTERS];
564 	boolean_t display_err_counters = do_disk & DISK_ERRORS;
565 	struct iodev_snapshot *old = (struct iodev_snapshot *)v1;
566 	struct iodev_snapshot *new = (struct iodev_snapshot *)v2;
567 	int *count = (int *)data;
568 	double rps, wps, tps, mtps, krps, kwps, kps, avw, avr, w_pct, r_pct;
569 	double wserv, rserv, serv;
570 	double iosize;	/* kb/sec or MB/sec */
571 	double etime, hr_etime;
572 	char *disk_name;
573 	u_longlong_t ldeltas;
574 	uint_t udeltas;
575 	uint64_t t_delta;
576 	uint64_t w_delta;
577 	uint64_t r_delta;
578 	int doit = 1;
579 	uint_t toterrs;
580 	char *fstr;
581 
582 	if (new == NULL)
583 		return;
584 
585 	switch (show_disk_mode) {
586 	case SHOW_FIRST_ONLY:
587 		if (count != NULL && *count)
588 			return;
589 		break;
590 
591 	case SHOW_SECOND_ONWARDS:
592 		if (count != NULL && !*count) {
593 			(*count)++;
594 			return;
595 		}
596 		break;
597 
598 	default:
599 		break;
600 	}
601 
602 	disk_name = do_conversions ? new->is_pretty : new->is_name;
603 	disk_name = disk_name ? disk_name : new->is_name;
604 
605 	/*
606 	 * Only do if we want IO stats - Avoids errors traveling this
607 	 * section if that's all we want to see.
608 	 */
609 	if (do_disk & DISK_IO_MASK) {
610 		if (old) {
611 			t_delta = hrtime_delta(old->is_snaptime,
612 			    new->is_snaptime);
613 		} else {
614 			t_delta = hrtime_delta(new->is_crtime,
615 			    new->is_snaptime);
616 		}
617 
618 		if (new->is_nr_children) {
619 			if (new->is_type == IODEV_CONTROLLER) {
620 				t_delta /= new->is_nr_children;
621 			} else if ((new->is_type == IODEV_IOPATH_LT) ||
622 			    (new->is_type == IODEV_IOPATH_LI)) {
623 				/* synthetic path */
624 				if (!old) {
625 					t_delta = new->is_crtime;
626 				}
627 				t_delta /= new->is_nr_children;
628 			}
629 		}
630 
631 		hr_etime = (double)t_delta;
632 		if (hr_etime == 0.0)
633 			hr_etime = (double)NANOSEC;
634 		etime = hr_etime / (double)NANOSEC;
635 
636 		/* reads per second */
637 		udeltas = u32_delta(old ? old->is_stats.reads : 0,
638 		    new->is_stats.reads);
639 		rps = (double)udeltas;
640 		rps /= etime;
641 
642 		/* writes per second */
643 		udeltas = u32_delta(old ? old->is_stats.writes : 0,
644 		    new->is_stats.writes);
645 		wps = (double)udeltas;
646 		wps /= etime;
647 
648 		tps = rps + wps;
649 			/* transactions per second */
650 
651 		/*
652 		 * report throughput as either kb/sec or MB/sec
653 		 */
654 
655 		if (!do_megabytes)
656 			iosize = 1024.0;
657 		else
658 			iosize = 1048576.0;
659 
660 		ldeltas = ull_delta(old ? old->is_stats.nread : 0,
661 		    new->is_stats.nread);
662 		if (ldeltas) {
663 			krps = (double)ldeltas;
664 			krps /= etime;
665 			krps /= iosize;
666 		} else
667 			krps = 0.0;
668 
669 		ldeltas = ull_delta(old ? old->is_stats.nwritten : 0,
670 		    new->is_stats.nwritten);
671 		if (ldeltas) {
672 			kwps = (double)ldeltas;
673 			kwps /= etime;
674 			kwps /= iosize;
675 		} else
676 			kwps = 0.0;
677 
678 		/*
679 		 * Blocks transferred per second
680 		 */
681 		kps = krps + kwps;
682 
683 		/*
684 		 * Average number of wait transactions waiting
685 		 */
686 		w_delta = hrtime_delta((u_longlong_t)
687 		    (old ? old->is_stats.wlentime : 0),
688 		    new->is_stats.wlentime);
689 		if (w_delta) {
690 			avw = (double)w_delta;
691 			avw /= hr_etime;
692 		} else
693 			avw = 0.0;
694 
695 		/*
696 		 * Average number of run transactions waiting
697 		 */
698 		r_delta = hrtime_delta(old ? old->is_stats.rlentime : 0,
699 		    new->is_stats.rlentime);
700 		if (r_delta) {
701 			avr = (double)r_delta;
702 			avr /= hr_etime;
703 		} else
704 			avr = 0.0;
705 
706 		/*
707 		 * Average wait service time in milliseconds
708 		 */
709 		if (tps > 0.0 && (avw != 0.0 || avr != 0.0)) {
710 			mtps = 1000.0 / tps;
711 			if (avw != 0.0)
712 				wserv = avw * mtps;
713 			else
714 				wserv = 0.0;
715 
716 			if (avr != 0.0)
717 				rserv = avr * mtps;
718 			else
719 				rserv = 0.0;
720 			serv = rserv + wserv;
721 		} else {
722 			rserv = 0.0;
723 			wserv = 0.0;
724 			serv = 0.0;
725 		}
726 
727 		/* % of time there is a transaction waiting for service */
728 		t_delta = hrtime_delta(old ? old->is_stats.wtime : 0,
729 		    new->is_stats.wtime);
730 		if (t_delta) {
731 			w_pct = (double)t_delta;
732 			w_pct /= hr_etime;
733 			w_pct *= 100.0;
734 
735 			/*
736 			 * Average the wait queue utilization over the
737 			 * the controller's devices, if this is a controller.
738 			 */
739 			if (new->is_type == IODEV_CONTROLLER)
740 				w_pct /= new->is_nr_children;
741 		} else
742 			w_pct = 0.0;
743 
744 		/* % of time there is a transaction running */
745 		t_delta = hrtime_delta(old ? old->is_stats.rtime : 0,
746 		    new->is_stats.rtime);
747 		if (t_delta) {
748 			r_pct = (double)t_delta;
749 			r_pct /= hr_etime;
750 			r_pct *= 100.0;
751 
752 			/*
753 			 * Average the percent busy over the controller's
754 			 * devices, if this is a controller.
755 			 */
756 			if (new->is_type == IODEV_CONTROLLER)
757 				w_pct /= new->is_nr_children;
758 		} else {
759 			r_pct = 0.0;
760 		}
761 
762 		/* % of time there is a transaction running */
763 		if (do_interval) {
764 			rps	*= etime;
765 			wps	*= etime;
766 			tps	*= etime;
767 			krps	*= etime;
768 			kwps	*= etime;
769 			kps	*= etime;
770 		}
771 	}
772 
773 	if (do_disk & (DISK_EXTENDED | DISK_ERRORS)) {
774 		if ((!do_conversions) && ((suppress_zero == 0) ||
775 		    ((do_disk & DISK_EXTENDED) == 0))) {
776 			if (do_raw == 0) {
777 				push_out("%-*.*s",
778 				    iodevs_nl, iodevs_nl, disk_name);
779 			} else {
780 				push_out(disk_name);
781 			}
782 		}
783 	}
784 
785 	/*
786 	 * The error counters are read first (if asked for and if they are
787 	 * available).
788 	 */
789 	bzero(err_counters, sizeof (err_counters));
790 	toterrs = 0;
791 	if (display_err_counters && (new->is_errors.ks_data != NULL)) {
792 		kstat_named_t	*knp;
793 		int		i;
794 
795 		knp = KSTAT_NAMED_PTR(&new->is_errors);
796 		for (i = 0; i < NUMBER_OF_ERR_COUNTERS; i++) {
797 			switch (knp[i].data_type) {
798 				case KSTAT_DATA_ULONG:
799 				case KSTAT_DATA_ULONGLONG:
800 					err_counters[i] = knp[i].value.ui32;
801 					toterrs += knp[i].value.ui32;
802 					break;
803 				default:
804 					break;
805 			}
806 		}
807 	}
808 
809 	switch (do_disk & DISK_IO_MASK) {
810 	case DISK_OLD:
811 		if (do_raw == 0)
812 			fstr = "%3.0f %3.0f %4.0f  ";
813 		else
814 			fstr = "%.0f,%.0f,%.0f";
815 		push_out(fstr, kps, tps, serv);
816 		break;
817 	case DISK_NEW:
818 		if (do_raw == 0)
819 			fstr = "%3.0f %3.0f %4.1f  ";
820 		else
821 			fstr = "%.0f,%.0f,%.1f";
822 		push_out(fstr, rps, wps, r_pct);
823 		break;
824 	case DISK_EXTENDED:
825 		if (suppress_zero) {
826 			if (fzero(rps) && fzero(wps) && fzero(krps) &&
827 			    fzero(kwps) && fzero(avw) && fzero(avr) &&
828 			    fzero(serv) && fzero(w_pct) && fzero(r_pct) &&
829 			    (toterrs == 0)) {
830 				doit = 0;
831 				display_err_counters = B_FALSE;
832 			} else if (do_conversions == 0) {
833 				if (do_raw == 0) {
834 					push_out("%-*.*s",
835 					    iodevs_nl, iodevs_nl, disk_name);
836 				} else {
837 					push_out(disk_name);
838 				}
839 			}
840 		}
841 		if (doit) {
842 			if (!do_conversions) {
843 				if (do_raw == 0) {
844 					fstr = " %6.1f %6.1f %6.1f %6.1f "
845 					    "%4.1f %4.1f %6.1f %3.0f "
846 					    "%3.0f ";
847 				} else {
848 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
849 					    "%.1f,%.0f,%.0f";
850 				}
851 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
852 				    serv, w_pct, r_pct);
853 			} else {
854 				if (do_raw == 0) {
855 					fstr = " %6.1f %6.1f %6.1f %6.1f "
856 					    "%4.1f %4.1f %6.1f %6.1f "
857 					    "%3.0f %3.0f ";
858 				} else {
859 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
860 					    "%.1f,%.1f,%.0f,%.0f";
861 				}
862 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
863 				    wserv, rserv, w_pct, r_pct);
864 			}
865 		}
866 		break;
867 	}
868 
869 	if (display_err_counters) {
870 		char	*efstr;
871 		int	i;
872 
873 		if (do_raw == 0) {
874 			if (do_disk == DISK_ERRORS)
875 				push_out(two_blanks);
876 			efstr = "%3u ";
877 		} else {
878 			efstr = "%u";
879 		}
880 
881 		for (i = 0; i < NUMBER_OF_ERR_COUNTERS; i++)
882 			push_out(efstr, err_counters[i]);
883 
884 		push_out(efstr, toterrs);
885 	}
886 
887 	if (suppress_zero == 0 || doit == 1) {
888 		if ((do_disk & (DISK_EXTENDED | DISK_ERRORS)) &&
889 		    do_conversions) {
890 			push_out("%s", disk_name);
891 			if (show_mountpts && new->is_dname) {
892 				mnt_t *mount_pt;
893 				char *lu;
894 				char *dnlu;
895 				char lub[SMALL_SCRATCH_BUFLEN];
896 
897 				lu = strrchr(new->is_dname, '/');
898 				if (lu) {
899 					/* only the part after a possible '/' */
900 					dnlu = strrchr(disk_name, '/');
901 					if (dnlu != NULL &&
902 					    strcmp(dnlu, lu) == 0)
903 						lu = new->is_dname;
904 					else {
905 						*lu = 0;
906 						(void) strcpy(lub,
907 						    new->is_dname);
908 						*lu = '/';
909 						(void) strcat(lub, "/");
910 						(void) strcat(lub,
911 						    disk_name);
912 						lu = lub;
913 					}
914 				} else
915 					lu = disk_name;
916 				mount_pt = lookup_mntent_byname(lu);
917 				if (mount_pt) {
918 					if (do_raw == 0)
919 						push_out(" (%s)",
920 						    mount_pt->mount_point);
921 					else
922 						push_out("(%s)",
923 						    mount_pt->mount_point);
924 				}
925 			}
926 		}
927 	}
928 
929 	if ((do_disk & PRINT_VERTICAL) && show_disk_mode != SHOW_FIRST_ONLY)
930 		do_newline();
931 
932 	if (count != NULL)
933 		(*count)++;
934 }
935 
936 static void
937 usage(void)
938 {
939 	(void) fprintf(stderr,
940 	    "Usage: iostat [-cCdDeEiImMnpPrstxXYz] "
941 	    " [-l n] [-T d|u] [disk ...] [interval [count]]\n"
942 	    "\t\t-c: 	report percentage of time system has spent\n"
943 	    "\t\t\tin user/system/wait/idle mode\n"
944 	    "\t\t-C: 	report disk statistics by controller\n"
945 	    "\t\t-d: 	display disk Kb/sec, transfers/sec, avg. \n"
946 	    "\t\t\tservice time in milliseconds  \n"
947 	    "\t\t-D: 	display disk reads/sec, writes/sec, \n"
948 	    "\t\t\tpercentage disk utilization \n"
949 	    "\t\t-e: 	report device error summary statistics\n"
950 	    "\t\t-E: 	report extended device error statistics\n"
951 	    "\t\t-i:	show device IDs for -E output\n"
952 	    "\t\t-I: 	report the counts in each interval,\n"
953 	    "\t\t\tinstead of rates, where applicable\n"
954 	    "\t\t-l n:	Limit the number of disks to n\n"
955 	    "\t\t-m: 	Display mount points (most useful with -p)\n"
956 	    "\t\t-M: 	Display data throughput in MB/sec "
957 	    "instead of Kb/sec\n"
958 	    "\t\t-n: 	convert device names to cXdYtZ format\n"
959 	    "\t\t-p: 	report per-partition disk statistics\n"
960 	    "\t\t-P: 	report per-partition disk statistics only,\n"
961 	    "\t\t\tno per-device disk statistics\n"
962 	    "\t\t-r: 	Display data in comma separated format\n"
963 	    "\t\t-s: 	Suppress state change messages\n"
964 	    "\t\t-T d|u	Display a timestamp in date (d) or unix "
965 	    "time_t (u)\n"
966 	    "\t\t-t: 	display chars read/written to terminals\n"
967 	    "\t\t-x: 	display extended disk statistics\n"
968 	    "\t\t-X: 	display I/O path statistics\n"
969 	    "\t\t-Y: 	display I/O path (I/T/L) statistics\n"
970 	    "\t\t-z: 	Suppress entries with all zero values\n");
971 	exit(1);
972 }
973 
974 /*ARGSUSED*/
975 static void
976 show_disk_errors(void *v1, void *v2, void *d)
977 {
978 	struct iodev_snapshot *disk = (struct iodev_snapshot *)v2;
979 	kstat_named_t *knp;
980 	size_t  col;
981 	int	i, len;
982 	char	*dev_name;
983 
984 	if (disk->is_errors.ks_ndata == 0)
985 		return;
986 	if (disk->is_type == IODEV_CONTROLLER)
987 		return;
988 
989 	dev_name = do_conversions ? disk->is_pretty : disk->is_name;
990 	dev_name = dev_name ? dev_name : disk->is_name;
991 
992 	len = strlen(dev_name);
993 	if (len > 20)
994 		push_out("%s ", dev_name);
995 	else if (len > 16)
996 		push_out("%-20.20s ", dev_name);
997 	else {
998 		if (do_conversions)
999 			push_out("%-16.16s ", dev_name);
1000 		else
1001 			push_out("%-9.9s ", dev_name);
1002 	}
1003 	col = 0;
1004 
1005 	knp = KSTAT_NAMED_PTR(&disk->is_errors);
1006 	for (i = 0; i < disk->is_errors.ks_ndata; i++) {
1007 		/* skip kstats that the driver did not kstat_named_init */
1008 		if (knp[i].name[0] == 0)
1009 			continue;
1010 
1011 		col += strlen(knp[i].name);
1012 
1013 		switch (knp[i].data_type) {
1014 			case KSTAT_DATA_CHAR:
1015 				if ((strcmp(knp[i].name, "Serial No") == 0) &&
1016 				    do_devid) {
1017 					if (disk->is_devid) {
1018 						push_out("Device Id: %s ",
1019 						    disk->is_devid);
1020 						col += strlen(disk->is_devid);
1021 					} else
1022 						push_out("Device Id: ");
1023 				} else {
1024 					push_out("%s: %-.16s ", knp[i].name,
1025 					    &knp[i].value.c[0]);
1026 					col += strlen(&knp[i].value.c[0]);
1027 				}
1028 				break;
1029 			case KSTAT_DATA_ULONG:
1030 				push_out("%s: %u ", knp[i].name,
1031 				    knp[i].value.ui32);
1032 				col += 4;
1033 				break;
1034 			case KSTAT_DATA_ULONGLONG:
1035 				if (strcmp(knp[i].name, "Size") == 0) {
1036 					push_out("%s: %2.2fGB <%llu bytes>\n",
1037 					    knp[i].name,
1038 					    (float)knp[i].value.ui64 /
1039 					    DISK_GIGABYTE,
1040 					    knp[i].value.ui64);
1041 					col = 0;
1042 					break;
1043 				}
1044 				push_out("%s: %u ", knp[i].name,
1045 				    knp[i].value.ui32);
1046 				col += 4;
1047 				break;
1048 			}
1049 		if ((col >= 62) || (i == 2)) {
1050 			do_newline();
1051 			col = 0;
1052 		}
1053 	}
1054 	if (col > 0) {
1055 		do_newline();
1056 	}
1057 	do_newline();
1058 }
1059 
1060 void
1061 do_args(int argc, char **argv)
1062 {
1063 	int 		c;
1064 	int 		errflg = 0;
1065 	extern char 	*optarg;
1066 	extern int 	optind;
1067 
1068 	while ((c = getopt(argc, argv, "tdDxXYCciIpPnmMeEszrT:l:")) != EOF)
1069 		switch (c) {
1070 		case 't':
1071 			do_tty++;
1072 			break;
1073 		case 'd':
1074 			do_disk |= DISK_OLD;
1075 			break;
1076 		case 'D':
1077 			do_disk |= DISK_NEW;
1078 			break;
1079 		case 'x':
1080 			do_disk |= DISK_EXTENDED;
1081 			break;
1082 		case 'X':
1083 			if (do_disk & DISK_IOPATH_LTI)
1084 				errflg++;	/* -Y already used */
1085 			else
1086 				do_disk |= DISK_IOPATH_LI;
1087 			break;
1088 		case 'Y':
1089 			if (do_disk & DISK_IOPATH_LI)
1090 				errflg++;	/* -X already used */
1091 			else
1092 				do_disk |= DISK_IOPATH_LTI;
1093 			break;
1094 		case 'C':
1095 			do_controller++;
1096 			break;
1097 		case 'c':
1098 			do_cpu++;
1099 			break;
1100 		case 'I':
1101 			do_interval++;
1102 			break;
1103 		case 'p':
1104 			do_partitions++;
1105 			break;
1106 		case 'P':
1107 			do_partitions_only++;
1108 			break;
1109 		case 'n':
1110 			do_conversions++;
1111 			break;
1112 		case 'M':
1113 			do_megabytes++;
1114 			break;
1115 		case 'e':
1116 			do_disk |= DISK_ERRORS;
1117 			break;
1118 		case 'E':
1119 			do_disk |= DISK_EXTENDED_ERRORS;
1120 			break;
1121 		case 'i':
1122 			do_devid = 1;
1123 			break;
1124 		case 's':
1125 			suppress_state = 1;
1126 			break;
1127 		case 'z':
1128 			suppress_zero = 1;
1129 			break;
1130 		case 'm':
1131 			show_mountpts = 1;
1132 			break;
1133 		case 'T':
1134 			if (optarg) {
1135 				if (*optarg == 'u')
1136 					timestamp_fmt = UDATE;
1137 				else if (*optarg == 'd')
1138 					timestamp_fmt = DDATE;
1139 				else
1140 					errflg++;
1141 			} else {
1142 				errflg++;
1143 			}
1144 			break;
1145 		case 'r':
1146 			do_raw = 1;
1147 			break;
1148 		case 'l':
1149 			df.if_max_iodevs = safe_strtoi(optarg, "invalid limit");
1150 			if (df.if_max_iodevs < 1)
1151 				usage();
1152 			break;
1153 		case '?':
1154 			errflg++;
1155 	}
1156 
1157 	if ((do_disk & DISK_OLD) && (do_disk & DISK_NEW)) {
1158 		(void) fprintf(stderr, "-d and -D are incompatible.\n");
1159 		usage();
1160 	}
1161 
1162 	if (errflg) {
1163 		usage();
1164 	}
1165 
1166 	/* if no output classes explicity specified, use defaults */
1167 	if (do_tty == 0 && do_disk == 0 && do_cpu == 0)
1168 		do_tty = do_cpu = 1, do_disk = DISK_OLD;
1169 
1170 	/*
1171 	 * multi-path options (-X, -Y) without a specific vertical
1172 	 * output format (-x, -e, -E) imply extended -x format
1173 	 */
1174 	if ((do_disk & (DISK_IOPATH_LI | DISK_IOPATH_LTI)) &&
1175 	    !(do_disk & PRINT_VERTICAL))
1176 		do_disk |= DISK_EXTENDED;
1177 
1178 	/*
1179 	 * If conflicting options take the preferred
1180 	 * -D and -x result in -x
1181 	 * -d or -D and -e or -E gives only whatever -d or -D was specified
1182 	 */
1183 	if ((do_disk & DISK_EXTENDED) && (do_disk & DISK_NORMAL))
1184 		do_disk &= ~DISK_NORMAL;
1185 	if ((do_disk & DISK_NORMAL) && (do_disk & DISK_ERROR_MASK))
1186 		do_disk &= ~DISK_ERROR_MASK;
1187 
1188 	/* nfs, tape, always shown */
1189 	df.if_allowed_types = IODEV_NFS | IODEV_TAPE;
1190 
1191 	/*
1192 	 * If limit == 0 then no command line limit was set, else if any of
1193 	 * the flags that cause unlimited disks were not set,
1194 	 * use the default of 4
1195 	 */
1196 	if (df.if_max_iodevs == 0) {
1197 		df.if_max_iodevs = DEFAULT_LIMIT;
1198 		df.if_skip_floppy = 1;
1199 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS |
1200 		    DISK_EXTENDED_ERRORS)) {
1201 			df.if_max_iodevs = UNLIMITED_IODEVS;
1202 			df.if_skip_floppy = 0;
1203 		}
1204 	}
1205 	if (do_disk) {
1206 		size_t count = 0;
1207 		size_t i = optind;
1208 
1209 		while (i < argc && !isdigit(argv[i][0])) {
1210 			count++;
1211 			i++;
1212 		}
1213 
1214 		/*
1215 		 * "Note:  disks  explicitly  requested
1216 		 * are not subject to this disk limit"
1217 		 */
1218 		if ((count > df.if_max_iodevs) ||
1219 		    (count && (df.if_max_iodevs == UNLIMITED_IODEVS)))
1220 			df.if_max_iodevs = count;
1221 
1222 		df.if_names = safe_alloc(count * sizeof (char *));
1223 		(void) memset(df.if_names, 0, count * sizeof (char *));
1224 
1225 		df.if_nr_names = 0;
1226 		while (optind < argc && !isdigit(argv[optind][0]))
1227 			df.if_names[df.if_nr_names++] = argv[optind++];
1228 	}
1229 	if (optind < argc) {
1230 		interval = safe_strtoi(argv[optind], "invalid interval");
1231 		if (interval < 1)
1232 			fail(0, "invalid interval");
1233 		optind++;
1234 
1235 		if (optind < argc) {
1236 			iter = safe_strtoi(argv[optind], "invalid count");
1237 			if (iter < 1)
1238 				fail(0, "invalid count");
1239 			optind++;
1240 		}
1241 	}
1242 	if (interval == 0)
1243 		iter = 1;
1244 	if (optind < argc)
1245 		usage();
1246 }
1247 
1248 /*
1249  * Driver for doing the extended header formatting. Will produce
1250  * the function stack needed to output an extended header based
1251  * on the options selected.
1252  */
1253 
1254 void
1255 do_format(void)
1256 {
1257 	char	header[SMALL_SCRATCH_BUFLEN];
1258 	char 	ch;
1259 	char 	iosz;
1260 	const char    *fstr;
1261 
1262 	disk_header[0] = 0;
1263 	ch = (do_interval ? 'i' : 's');
1264 	iosz = (do_megabytes ? 'M' : 'k');
1265 	if (do_disk & DISK_ERRORS) {
1266 		if (do_raw == 0) {
1267 			(void) sprintf(header, "s/w h/w trn tot ");
1268 		} else
1269 			(void) sprintf(header, "s/w,h/w,trn,tot");
1270 	} else
1271 		*header = NULL;
1272 	switch (do_disk & DISK_IO_MASK) {
1273 		case DISK_OLD:
1274 			if (do_raw == 0)
1275 				fstr = "%cp%c tp%c serv  ";
1276 			else
1277 				fstr = "%cp%c,tp%c,serv";
1278 			(void) snprintf(disk_header, sizeof (disk_header),
1279 			    fstr, iosz, ch, ch);
1280 			break;
1281 		case DISK_NEW:
1282 			if (do_raw == 0)
1283 				fstr = "rp%c wp%c util  ";
1284 			else
1285 				fstr = "%rp%c,wp%c,util";
1286 			(void) snprintf(disk_header, sizeof (disk_header),
1287 			    fstr, ch, ch);
1288 			break;
1289 		case DISK_EXTENDED:
1290 			/* This is -x option */
1291 			if (!do_conversions) {
1292 				/* without -n option */
1293 				if (do_raw == 0) {
1294 					/* without -r option */
1295 					(void) snprintf(disk_header,
1296 					    sizeof (disk_header),
1297 					    "%-*.*s    r/%c    w/%c   "
1298 					    "%cr/%c   %cw/%c wait actv  "
1299 					    "svc_t  %%%%w  %%%%b %s",
1300 					    iodevs_nl, iodevs_nl, "device",
1301 					    ch, ch, iosz, ch, iosz, ch, header);
1302 				} else {
1303 					/* with -r option */
1304 					(void) snprintf(disk_header,
1305 					    sizeof (disk_header),
1306 					    "device,r/%c,w/%c,%cr/%c,%cw/%c,"
1307 					    "wait,actv,svc_t,%%%%w,"
1308 					    "%%%%b,%s",
1309 					    ch, ch, iosz, ch, iosz, ch, header);
1310 				}
1311 			} else {
1312 				/* with -n option */
1313 				if (do_raw == 0) {
1314 					fstr = "    r/%c    w/%c   %cr/%c   "
1315 					    "%cw/%c wait actv wsvc_t asvc_t  "
1316 					    "%%%%w  %%%%b %sdevice";
1317 				} else {
1318 					fstr = "r/%c,w/%c,%cr/%c,%cw/%c,"
1319 					    "wait,actv,wsvc_t,asvc_t,"
1320 					    "%%%%w,%%%%b,%sdevice";
1321 				}
1322 				(void) snprintf(disk_header,
1323 				    sizeof (disk_header),
1324 				    fstr, ch, ch, iosz, ch, iosz,
1325 				    ch, header);
1326 			}
1327 			break;
1328 		default:
1329 			break;
1330 	}
1331 
1332 	/* do DISK_ERRORS header (already added above for DISK_EXTENDED) */
1333 	if ((do_disk & DISK_ERRORS) &&
1334 	    ((do_disk & DISK_IO_MASK) != DISK_EXTENDED)) {
1335 		if (!do_conversions) {
1336 			if (do_raw == 0)
1337 				(void) snprintf(disk_header,
1338 				    sizeof (disk_header), "%-*.*s  %s",
1339 				    iodevs_nl, iodevs_nl, "device", header);
1340 			else
1341 				(void) snprintf(disk_header,
1342 				    sizeof (disk_header), "device,%s", header);
1343 		} else {
1344 			if (do_raw == 0) {
1345 				(void) snprintf(disk_header,
1346 				    sizeof (disk_header),
1347 				    "  %sdevice", header);
1348 			} else {
1349 				(void) snprintf(disk_header,
1350 				    sizeof (disk_header),
1351 				    "%s,device", header);
1352 			}
1353 		}
1354 	} else {
1355 		/*
1356 		 * Need to subtract two characters for the % escape in
1357 		 * the string.
1358 		 */
1359 		dh_len = strlen(disk_header) - 2;
1360 	}
1361 
1362 	/*
1363 	 * -n *and* (-E *or* -e *or* -x)
1364 	 */
1365 	if (do_conversions && (do_disk & PRINT_VERTICAL)) {
1366 		if (do_tty)
1367 			setup(print_tty_hdr1);
1368 		if (do_cpu)
1369 			setup(print_cpu_hdr1);
1370 		if (do_tty || do_cpu)
1371 			setup(do_newline);
1372 		if (do_tty)
1373 			setup(print_tty_hdr2);
1374 		if (do_cpu)
1375 			setup(print_cpu_hdr2);
1376 		if (do_tty || do_cpu)
1377 			setup(do_newline);
1378 		if (do_tty)
1379 			setup(print_tty_data);
1380 		if (do_cpu)
1381 			setup(print_cpu_data);
1382 		if (do_tty || do_cpu)
1383 			setup(do_newline);
1384 		printxhdr();
1385 
1386 		setup(show_all_disks);
1387 	} else {
1388 		/*
1389 		 * These unholy gymnastics are necessary to place CPU/tty
1390 		 * data to the right of the disks/errors for the first
1391 		 * line in vertical mode.
1392 		 */
1393 		if (do_disk & PRINT_VERTICAL) {
1394 			printxhdr();
1395 
1396 			setup(show_first_disk);
1397 			if (do_tty)
1398 				setup(print_tty_data);
1399 			if (do_cpu)
1400 				setup(print_cpu_data);
1401 			setup(do_newline);
1402 
1403 			setup(show_other_disks);
1404 		} else {
1405 			setup(hdrout);
1406 			if (do_tty)
1407 				setup(print_tty_data);
1408 			setup(show_all_disks);
1409 			if (do_cpu)
1410 				setup(print_cpu_data);
1411 		}
1412 
1413 		setup(do_newline);
1414 	}
1415 	if (do_disk & DISK_EXTENDED_ERRORS)
1416 		setup(disk_errors);
1417 }
1418 
1419 /*
1420  * Add a new function to the list of functions
1421  * for this invocation. Once on the stack the
1422  * function is never removed nor does its place
1423  * change.
1424  */
1425 void
1426 setup(void (*nfunc)(void))
1427 {
1428 	format_t *tmp;
1429 
1430 	tmp = safe_alloc(sizeof (format_t));
1431 	tmp->nfunc = nfunc;
1432 	tmp->next = 0;
1433 	if (formatter_end)
1434 		formatter_end->next = tmp;
1435 	else
1436 		formatter_list = tmp;
1437 	formatter_end = tmp;
1438 
1439 }
1440 
1441 /*
1442  * The functions after this comment are devoted to printing
1443  * various parts of the header. They are selected based on the
1444  * options provided when the program was invoked. The functions
1445  * are either directly invoked in printhdr() or are indirectly
1446  * invoked by being placed on the list of functions used when
1447  * extended headers are used.
1448  */
1449 void
1450 print_tty_hdr1(void)
1451 {
1452 	char *fstr;
1453 	char *dstr;
1454 
1455 	if (do_raw == 0) {
1456 		fstr = "%10.10s";
1457 		dstr = "tty    ";
1458 	} else {
1459 		fstr = "%s";
1460 		dstr = "tty";
1461 	}
1462 	push_out(fstr, dstr);
1463 }
1464 
1465 void
1466 print_tty_hdr2(void)
1467 {
1468 	if (do_raw == 0)
1469 		push_out("%-10.10s", " tin tout");
1470 	else
1471 		push_out("tin,tout");
1472 }
1473 
1474 void
1475 print_cpu_hdr1(void)
1476 {
1477 	char *dstr;
1478 
1479 	if (do_raw == 0)
1480 		dstr = "     cpu";
1481 	else
1482 		dstr = "cpu";
1483 	push_out(dstr);
1484 }
1485 
1486 void
1487 print_cpu_hdr2(void)
1488 {
1489 	char *dstr;
1490 
1491 	if (do_raw == 0)
1492 		dstr = " us sy wt id";
1493 	else
1494 		dstr = "us,sy,wt,id";
1495 	push_out(dstr);
1496 }
1497 
1498 /*
1499  * Assumption is that tty data is always first - no need for raw mode leading
1500  * comma.
1501  */
1502 void
1503 print_tty_data(void)
1504 {
1505 	char *fstr;
1506 	uint64_t deltas;
1507 	double raw;
1508 	double outch;
1509 	kstat_t *oldks = NULL;
1510 
1511 	if (oldss)
1512 		oldks = &oldss->s_sys.ss_agg_sys;
1513 
1514 	if (do_raw == 0)
1515 		fstr = " %3.0f %4.0f ";
1516 	else
1517 		fstr = "%.0f,%.0f";
1518 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "rawch");
1519 	raw = deltas;
1520 	raw /= getime;
1521 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "outch");
1522 	outch = deltas;
1523 	outch /= getime;
1524 	push_out(fstr, raw, outch);
1525 }
1526 
1527 /*
1528  * Write out CPU data
1529  */
1530 void
1531 print_cpu_data(void)
1532 {
1533 	char *fstr;
1534 	uint64_t idle;
1535 	uint64_t user;
1536 	uint64_t kern;
1537 	uint64_t wait;
1538 	kstat_t *oldks = NULL;
1539 
1540 	if (oldss)
1541 		oldks = &oldss->s_sys.ss_agg_sys;
1542 
1543 	if (do_raw == 0)
1544 		fstr = " %2.0f %2.0f %2.0f %2.0f";
1545 	else
1546 		fstr = "%.0f,%.0f,%.0f,%.0f";
1547 
1548 	idle = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_idle");
1549 	user = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_user");
1550 	kern = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_kernel");
1551 	wait = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_wait");
1552 	push_out(fstr, user * percent, kern * percent,
1553 	    wait * percent, idle * percent);
1554 }
1555 
1556 /*
1557  * Emit the appropriate header.
1558  */
1559 void
1560 hdrout(void)
1561 {
1562 	if (do_raw == 0) {
1563 		if (--tohdr == 0)
1564 			printhdr(0);
1565 	} else if (hdr_out == 0) {
1566 		printhdr(0);
1567 		hdr_out = 1;
1568 	}
1569 }
1570 
1571 /*
1572  * Write out disk errors when -E is specified.
1573  */
1574 void
1575 disk_errors(void)
1576 {
1577 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk_errors, NULL);
1578 }
1579 
1580 void
1581 show_first_disk(void)
1582 {
1583 	int count = 0;
1584 
1585 	show_disk_mode = SHOW_FIRST_ONLY;
1586 
1587 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1588 }
1589 
1590 void
1591 show_other_disks(void)
1592 {
1593 	int count = 0;
1594 
1595 	show_disk_mode = SHOW_SECOND_ONWARDS;
1596 
1597 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1598 }
1599 
1600 void
1601 show_all_disks(void)
1602 {
1603 	int count = 0;
1604 
1605 	show_disk_mode = SHOW_ALL;
1606 
1607 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1608 }
1609 
1610 /*
1611  * Write a newline out and clear the lineout flag.
1612  */
1613 static void
1614 do_newline(void)
1615 {
1616 	if (lineout) {
1617 		(void) putchar('\n');
1618 		lineout = 0;
1619 	}
1620 }
1621 
1622 /*
1623  * Generalized printf function that determines what extra
1624  * to print out if we're in raw mode. At this time we
1625  * don't care about errors.
1626  */
1627 static void
1628 push_out(const char *message, ...)
1629 {
1630 	va_list args;
1631 
1632 	va_start(args, message);
1633 	if (do_raw && lineout == 1)
1634 		(void) putchar(',');
1635 	(void) vprintf(message, args);
1636 	va_end(args);
1637 	lineout = 1;
1638 }
1639 
1640 /*
1641  * Emit the header string when -e is specified.
1642  */
1643 static void
1644 print_err_hdr(void)
1645 {
1646 	char obuf[SMALL_SCRATCH_BUFLEN];
1647 
1648 	if (do_raw) {
1649 		push_out("errors");
1650 		return;
1651 	}
1652 
1653 	if (do_conversions == 0) {
1654 		if (!(do_disk & DISK_EXTENDED)) {
1655 			(void) snprintf(obuf, sizeof (obuf),
1656 			    "%11s", one_blank);
1657 			push_out(obuf);
1658 		}
1659 	} else if (do_disk == DISK_ERRORS)
1660 		push_out(two_blanks);
1661 	else
1662 		push_out(one_blank);
1663 	push_out("---- errors --- ");
1664 }
1665 
1666 /*
1667  * Emit the header string when -e is specified.
1668  */
1669 static void
1670 print_disk_header(void)
1671 {
1672 	push_out(disk_header);
1673 }
1674 
1675 /*
1676  * No, UINTMAX_MAX isn't the right thing here since
1677  * it is #defined to be either INT32_MAX or INT64_MAX
1678  * depending on the whether _LP64 is defined.
1679  *
1680  * We want to handle the odd future case of having
1681  * ulonglong_t be more than 64 bits but we have
1682  * no nice #define MAX value we can drop in place
1683  * without having to change this code in the future.
1684  */
1685 
1686 u_longlong_t
1687 ull_delta(u_longlong_t old, u_longlong_t new)
1688 {
1689 	if (new >= old)
1690 		return (new - old);
1691 	else
1692 		return ((UINT64_MAX - old) + new + 1);
1693 }
1694 
1695 /*
1696  * Take the difference of an unsigned 32
1697  * bit int attempting to cater for
1698  * overflow.
1699  */
1700 uint_t
1701 u32_delta(uint_t old, uint_t new)
1702 {
1703 	if (new >= old)
1704 		return (new - old);
1705 	else
1706 		return ((UINT32_MAX - old) + new + 1);
1707 }
1708 
1709 /*
1710  * This is exactly what is needed for standard iostat output,
1711  * but make sure to use it only for that
1712  */
1713 #define	EPSILON	(0.1)
1714 static int
1715 fzero(double value)
1716 {
1717 	return (value >= 0.0 && value < EPSILON);
1718 }
1719 
1720 static int
1721 safe_strtoi(char const *val, char *errmsg)
1722 {
1723 	char *end;
1724 	long tmp;
1725 
1726 	errno = 0;
1727 	tmp = strtol(val, &end, 10);
1728 	if (*end != '\0' || errno)
1729 		fail(0, "%s %s", errmsg, val);
1730 	return ((int)tmp);
1731 }
1732