xref: /titanic_51/usr/src/cmd/stat/iostat/iostat.c (revision 43a291055ab3951f6372241323fd4e2486098fff)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <stdarg.h>
35 #include <ctype.h>
36 #include <unistd.h>
37 #include <memory.h>
38 #include <errno.h>
39 #include <string.h>
40 #include <signal.h>
41 #include <sys/types.h>
42 #include <time.h>
43 #include <sys/time.h>
44 #include <sys/sysinfo.h>
45 #include <inttypes.h>
46 #include <strings.h>
47 #include <sys/systeminfo.h>
48 #include <kstat.h>
49 
50 #include "dsr.h"
51 #include "statcommon.h"
52 
53 #define	DISK_OLD		0x0001
54 #define	DISK_NEW		0x0002
55 #define	DISK_EXTENDED		0x0004
56 #define	DISK_ERRORS		0x0008
57 #define	DISK_EXTENDED_ERRORS	0x0010
58 #define	DISK_IOPATH		0x0020
59 #define	DISK_NORMAL		(DISK_OLD | DISK_NEW)
60 
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 | DISK_IOPATH)
64 
65 #define	REPRINT 19
66 
67 /*
68  * It's really a pseudo-gigabyte. We use 1000000000 bytes so that the disk
69  * labels don't look bad. 1GB is really 1073741824 bytes.
70  */
71 #define	DISK_GIGABYTE   1000000000.0
72 
73 /*
74  * Function desciptor to be called when extended
75  * headers are used.
76  */
77 typedef struct formatter {
78 	void (*nfunc)(void);
79 	struct formatter *next;
80 } format_t;
81 
82 /*
83  * Used to get formatting right when printing tty/cpu
84  * data to the right of disk data
85  */
86 enum show_disk_mode {
87 	SHOW_FIRST_ONLY,
88 	SHOW_SECOND_ONWARDS,
89 	SHOW_ALL
90 };
91 
92 enum show_disk_mode show_disk_mode = SHOW_ALL;
93 
94 char cmdname[] = "iostat";
95 
96 static char one_blank[] = " ";
97 static char two_blanks[] = "  ";
98 
99 /*
100  * count for number of lines to be emitted before a header is
101  * shown again. Only used for the basic format.
102  */
103 static	uint_t	tohdr = 1;
104 
105 /*
106  * If we're in raw format, have we printed a header? We only do it
107  * once for raw but we emit it every REPRINT lines in non-raw format.
108  * This applies only for the basic header. The extended header is
109  * done only once in both formats.
110  */
111 static	uint_t	hdr_out;
112 
113 /*
114  * Flags representing arguments from command line
115  */
116 static	uint_t	do_tty;			/* show tty info (-t) */
117 static	uint_t	do_disk;		/* show disk info per selected */
118 					/* format (-d, -D, -e, -E, -x -X) */
119 static	uint_t	do_cpu;			/* show cpu info (-c) */
120 static	uint_t	do_interval;		/* do intervals (-I) */
121 static	int	do_partitions;		/* per-partition stats (-p) */
122 static	int	do_partitions_only;	/* per-partition stats only (-P) */
123 					/* no per-device stats for disks */
124 static	uint_t	do_conversions;		/* display disks as cXtYdZ (-n) */
125 static	uint_t	do_megabytes;		/* display data in MB/sec (-M) */
126 static  uint_t	do_controller;		/* display controller info (-C) */
127 static  uint_t	do_raw;			/* emit raw format (-r) */
128 static  uint_t	do_timestamp;		/* timestamp  each display (-T) */
129 static	uint_t	do_devid;		/* -E should show devid */
130 
131 /*
132  * Definition of allowable types of timestamps
133  */
134 #define	CDATE 1
135 #define	UDATE 2
136 
137 /*
138  * Default number of disk drives to be displayed in basic format
139  */
140 #define	DEFAULT_LIMIT	4
141 
142 struct iodev_filter df;
143 
144 static  uint_t	suppress_state;		/* skip state change messages */
145 static	uint_t	suppress_zero;		/* skip zero valued lines */
146 static  uint_t	show_mountpts;		/* show mount points */
147 static	int 	interval;		/* interval (seconds) to output */
148 static	int 	iter;			/* iterations from command line */
149 
150 #define	SMALL_SCRATCH_BUFLEN	64
151 #define	DISPLAYED_NAME_FORMAT "%-9.9s"
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 uint64_t	hrtime_delta(hrtime_t, hrtime_t);
169 static u_longlong_t	ull_delta(u_longlong_t, u_longlong_t);
170 static uint_t 	u32_delta(uint_t, uint_t);
171 static void setup(void (*nfunc)(void));
172 static void print_timestamp(void);
173 static void print_tty_hdr1(void);
174 static void print_tty_hdr2(void);
175 static void print_cpu_hdr1(void);
176 static void print_cpu_hdr2(void);
177 static void print_tty_data(void);
178 static void print_cpu_data(void);
179 static void print_err_hdr(void);
180 static void print_disk_header(void);
181 static void hdrout(void);
182 static void disk_errors(void);
183 static void do_newline(void);
184 static void push_out(const char *, ...);
185 static void printhdr(int);
186 static void printxhdr(void);
187 static void usage(void);
188 static void do_args(int, char **);
189 static void do_format(void);
190 static void set_timer(int);
191 static void handle_sig(int);
192 static void show_all_disks(void);
193 static void show_first_disk(void);
194 static void show_other_disks(void);
195 static void show_disk_errors(void *, void *, void *);
196 static void write_core_header(void);
197 static int  fzero(double value);
198 static int  safe_strtoi(char const *val, char *errmsg);
199 
200 int
201 main(int argc, char **argv)
202 {
203 	enum snapshot_types types = SNAP_SYSTEM;
204 	kstat_ctl_t *kc;
205 	long hz;
206 
207 	do_args(argc, argv);
208 	do_format();
209 
210 	/*
211 	 * iostat historically showed CPU changes, even though
212 	 * it doesn't provide much useful information
213 	 */
214 	types |= SNAP_CPUS;
215 
216 	if (do_disk)
217 		types |= SNAP_IODEVS;
218 
219 	if (do_disk && !do_partitions_only)
220 		df.if_allowed_types |= IODEV_DISK;
221 	if (do_disk & DISK_IOPATH) {
222 		df.if_allowed_types |= IODEV_IOPATH;
223 		types |= SNAP_IOPATHS;
224 	}
225 	if (do_disk & DISK_ERROR_MASK)
226 		types |= SNAP_IODEV_ERRORS;
227 	if (do_partitions || do_partitions_only)
228 		df.if_allowed_types |= IODEV_PARTITION;
229 	if (do_conversions)
230 		types |= SNAP_IODEV_PRETTY;
231 	if (do_controller) {
232 		if (!(do_disk & PRINT_VERTICAL) ||
233 			(do_disk & DISK_EXTENDED_ERRORS))
234 			fail(0, "-C can only be used with -e or -x.");
235 		types |= SNAP_CONTROLLERS;
236 		df.if_allowed_types |= IODEV_CONTROLLER;
237 	}
238 
239 	hz = sysconf(_SC_CLK_TCK);
240 
241 	/*
242 	 * Undocumented behavior - sending a SIGCONT will result
243 	 * in a new header being emitted. Used only if we're not
244 	 * doing extended headers. This is a historical
245 	 * artifact.
246 	 */
247 	if (!(do_disk & PRINT_VERTICAL))
248 		(void) signal(SIGCONT, printhdr);
249 
250 	if (interval)
251 		set_timer(interval);
252 
253 	kc = open_kstat();
254 	newss = acquire_snapshot(kc, types, &df);
255 
256 	do {
257 		if (do_tty || do_cpu) {
258 			kstat_t *oldks;
259 			oldks = oldss ? &oldss->s_sys.ss_agg_sys : NULL;
260 			getime = cpu_ticks_delta(oldks,
261 			    &newss->s_sys.ss_agg_sys);
262 			percent = (getime > 0.0) ? 100.0 / getime : 0.0;
263 			getime = (getime / nr_active_cpus(newss)) / hz;
264 			if (getime == 0.0)
265 				getime = (double)interval;
266 			if (getime == 0.0 || do_interval)
267 				getime = 1.0;
268 		}
269 
270 		if (formatter_list) {
271 			format_t *tmp;
272 			tmp = formatter_list;
273 			while (tmp) {
274 				(tmp->nfunc)();
275 				tmp = tmp->next;
276 			}
277 			(void) fflush(stdout);
278 		}
279 
280 		if (interval > 0 && iter != 1)
281 			(void) pause();
282 
283 		free_snapshot(oldss);
284 		oldss = newss;
285 		newss = acquire_snapshot(kc, types, &df);
286 
287 		if (!suppress_state)
288 			snapshot_report_changes(oldss, newss);
289 
290 		/* if config changed, show stats from boot */
291 		if (snapshot_has_changed(oldss, newss)) {
292 			free_snapshot(oldss);
293 			oldss = NULL;
294 		}
295 
296 	} while (--iter);
297 
298 	free_snapshot(oldss);
299 	free_snapshot(newss);
300 	return (0);
301 }
302 
303 /*
304  * Some magic numbers used in header formatting.
305  *
306  * DISK_LEN = length of either "kps tps serv" or "wps rps util"
307  *	      using 0 as the first position
308  *
309  * DISK_ERROR_LEN = length of "s/w h/w trn tot" with one space on
310  *		either side. Does not use zero as first pos.
311  *
312  * DEVICE_LEN = length of "device" + 1 character.
313  */
314 
315 #define	DISK_LEN	11
316 #define	DISK_ERROR_LEN	16
317 #define	DEVICE_LEN	7
318 
319 /*ARGSUSED*/
320 static void
321 show_disk_name(void *v1, void *v2, void *data)
322 {
323 	struct iodev_snapshot *dev = (struct iodev_snapshot *)v2;
324 	size_t slen;
325 	char *name;
326 	char fbuf[SMALL_SCRATCH_BUFLEN];
327 
328 	if (dev == NULL)
329 		return;
330 
331 	name = dev->is_pretty ? dev->is_pretty : dev->is_name;
332 	if (!do_raw) {
333 		uint_t width;
334 
335 		slen = strlen(name);
336 		/*
337 		 * The length is less
338 		 * than the section
339 		 * which will be displayed
340 		 * on the next line.
341 		 * Center the entry.
342 		 */
343 
344 		width = (DISK_LEN + 1)/2 + (slen / 2);
345 		(void) snprintf(fbuf, sizeof (fbuf),
346 		    "%*s", width, name);
347 		name = fbuf;
348 		push_out("%-14.14s", name);
349 	} else {
350 		push_out(name);
351 	}
352 }
353 
354 /*ARGSUSED*/
355 static void
356 show_disk_header(void *v1, void *v2, void *data)
357 {
358 	push_out(disk_header);
359 }
360 
361 /*
362  * Write out a two line header. What is written out depends on the flags
363  * selected but in the worst case consists of a tty header, a disk header
364  * providing information for 4 disks and a cpu header.
365  *
366  * The tty header consists of the word "tty" on the first line above the
367  * words "tin tout" on the next line. If present the tty portion consumes
368  * the first 10 characters of each line since "tin tout" is surrounded
369  * by single spaces.
370  *
371  * Each of the disk sections is a 14 character "block" in which the name of
372  * the disk is centered in the first 12 characters of the first line.
373  *
374  * The cpu section is an 11 character block with "cpu" centered over the
375  * section.
376  *
377  * The worst case should look as follows:
378  *
379  * 0---------1--------2---------3---------4---------5---------6---------7-------
380  *    tty        sd0           sd1           sd2           sd3           cpu
381  *  tin tout kps tps serv  kps tps serv  kps tps serv  kps tps serv  us sy wt id
382  *  NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NN NN NN NN
383  *
384  * When -D is specified, the disk header looks as follows (worst case):
385  *
386  * 0---------1--------2---------3---------4---------5---------6---------7-------
387  *     tty        sd0           sd1             sd2          sd3          cpu
388  *   tin tout rps wps util  rps wps util  rps wps util  rps wps util us sy wt id
389  *   NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN NN NN NN NN
390  */
391 static void
392 printhdr(int sig)
393 {
394 	/*
395 	 * If we're here because a signal fired, reenable the
396 	 * signal.
397 	 */
398 	if (sig)
399 		(void) signal(SIGCONT, printhdr);
400 	/*
401 	 * Horizontal mode headers
402 	 *
403 	 * First line
404 	 */
405 	if (do_tty)
406 		print_tty_hdr1();
407 
408 	if (do_disk & DISK_NORMAL) {
409 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
410 		    show_disk_name, NULL);
411 	}
412 
413 	if (do_cpu)
414 		print_cpu_hdr1();
415 	do_newline();
416 
417 	/*
418 	 * Second line
419 	 */
420 	if (do_tty)
421 		print_tty_hdr2();
422 
423 	if (do_disk & DISK_NORMAL) {
424 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
425 		    show_disk_header, NULL);
426 	}
427 
428 	if (do_cpu)
429 		print_cpu_hdr2();
430 	do_newline();
431 
432 	tohdr = REPRINT;
433 }
434 
435 /*
436  * Write out the extended header centered over the core information.
437  */
438 static void
439 write_core_header(void)
440 {
441 	char *edev = "extended device statistics";
442 	uint_t lead_space_ct;
443 	uint_t follow_space_ct;
444 	size_t edevlen;
445 
446 	if (do_raw == 0) {
447 		/*
448 		 * The things we do to look nice...
449 		 *
450 		 * Center the core output header. Make sure we have the
451 		 * right number of trailing spaces for follow-on headers
452 		 * (i.e., cpu and/or tty and/or errors).
453 		 */
454 		edevlen = strlen(edev);
455 		lead_space_ct = dh_len - edevlen;
456 		lead_space_ct /= 2;
457 		if (lead_space_ct > 0) {
458 			follow_space_ct = dh_len - (lead_space_ct + edevlen);
459 			if (do_disk & DISK_ERRORS)
460 				follow_space_ct -= DISK_ERROR_LEN;
461 			if ((do_disk & DISK_EXTENDED) && do_conversions)
462 				follow_space_ct -= DEVICE_LEN;
463 
464 			push_out("%1$*2$.*2$s%3$s%4$*5$.*5$s", one_blank,
465 			    lead_space_ct, edev, one_blank, follow_space_ct);
466 		} else
467 			push_out("%56s", edev);
468 	} else
469 		push_out(edev);
470 }
471 
472 /*
473  * In extended mode headers, we don't want to reprint the header on
474  * signals as they are printed every time anyways.
475  */
476 static void
477 printxhdr(void)
478 {
479 
480 	/*
481 	 * Vertical mode headers
482 	 */
483 	if (do_disk & DISK_EXTENDED)
484 		setup(write_core_header);
485 	if (do_disk & DISK_ERRORS)
486 		setup(print_err_hdr);
487 
488 	if (do_conversions) {
489 		setup(do_newline);
490 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
491 			setup(print_disk_header);
492 		setup(do_newline);
493 	} else {
494 		if (do_tty)
495 			setup(print_tty_hdr1);
496 		if (do_cpu)
497 			setup(print_cpu_hdr1);
498 		setup(do_newline);
499 
500 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
501 			setup(print_disk_header);
502 		if (do_tty)
503 			setup(print_tty_hdr2);
504 		if (do_cpu)
505 			setup(print_cpu_hdr2);
506 		setup(do_newline);
507 	}
508 }
509 
510 /*
511  * Write out a line for this disk - note that show_disk writes out
512  * full lines or blocks for each selected disk.
513  */
514 static void
515 show_disk(void *v1, void *v2, void *data)
516 {
517 	struct iodev_snapshot *old = (struct iodev_snapshot *)v1;
518 	struct iodev_snapshot *new = (struct iodev_snapshot *)v2;
519 	int *count = (int *)data;
520 	double rps, wps, tps, mtps, krps, kwps, kps, avw, avr, w_pct, r_pct;
521 	double wserv, rserv, serv;
522 	double iosize;	/* kb/sec or MB/sec */
523 	double etime, hr_etime;
524 	char *disk_name;
525 	u_longlong_t ldeltas;
526 	uint_t udeltas;
527 	uint64_t t_delta;
528 	uint64_t w_delta;
529 	uint64_t r_delta;
530 	int doit = 1;
531 	int i;
532 	uint_t toterrs;
533 	char *fstr;
534 
535 	if (new == NULL)
536 		return;
537 
538 	switch (show_disk_mode) {
539 	case SHOW_FIRST_ONLY:
540 		if (count != NULL && *count)
541 			return;
542 		break;
543 
544 	case SHOW_SECOND_ONWARDS:
545 		if (count != NULL && !*count) {
546 			(*count)++;
547 			return;
548 		}
549 		break;
550 
551 	default:
552 		break;
553 	}
554 
555 	disk_name = new->is_pretty ? new->is_pretty : new->is_name;
556 
557 	/*
558 	 * Only do if we want IO stats - Avoids errors traveling this
559 	 * section if that's all we want to see.
560 	 */
561 	if (do_disk & DISK_IO_MASK) {
562 		if (old) {
563 			t_delta = hrtime_delta(old->is_snaptime,
564 			    new->is_snaptime);
565 		} else {
566 			t_delta = hrtime_delta(new->is_crtime,
567 			    new->is_snaptime);
568 		}
569 
570 		if (new->is_type == IODEV_CONTROLLER && new->is_nr_children)
571 			t_delta /= new->is_nr_children;
572 
573 		hr_etime = (double)t_delta;
574 		if (hr_etime == 0.0)
575 			hr_etime = (double)NANOSEC;
576 		etime = hr_etime / (double)NANOSEC;
577 
578 		/* reads per second */
579 		udeltas = u32_delta(old ? old->is_stats.reads : 0,
580 		    new->is_stats.reads);
581 		rps = (double)udeltas;
582 		rps /= etime;
583 
584 		/* writes per second */
585 		udeltas = u32_delta(old ? old->is_stats.writes : 0,
586 		    new->is_stats.writes);
587 		wps = (double)udeltas;
588 		wps /= etime;
589 
590 		tps = rps + wps;
591 			/* transactions per second */
592 
593 		/*
594 		 * report throughput as either kb/sec or MB/sec
595 		 */
596 
597 		if (!do_megabytes)
598 			iosize = 1024.0;
599 		else
600 			iosize = 1048576.0;
601 
602 		ldeltas = ull_delta(old ? old->is_stats.nread : 0,
603 		    new->is_stats.nread);
604 		if (ldeltas) {
605 			krps = (double)ldeltas;
606 			krps /= etime;
607 			krps /= iosize;
608 		} else
609 			krps = 0.0;
610 
611 		ldeltas = ull_delta(old ? old->is_stats.nwritten : 0,
612 		    new->is_stats.nwritten);
613 		if (ldeltas) {
614 			kwps = (double)ldeltas;
615 			kwps /= etime;
616 			kwps /= iosize;
617 		} else
618 			kwps = 0.0;
619 
620 		/*
621 		 * Blocks transferred per second
622 		 */
623 		kps = krps + kwps;
624 
625 		/*
626 		 * Average number of wait transactions waiting
627 		 */
628 		w_delta = hrtime_delta((u_longlong_t)
629 		    (old ? old->is_stats.wlentime : 0),
630 		    new->is_stats.wlentime);
631 		if (w_delta) {
632 			avw = (double)w_delta;
633 			avw /= hr_etime;
634 		} else
635 			avw = 0.0;
636 
637 		/*
638 		 * Average number of run transactions waiting
639 		 */
640 		r_delta = hrtime_delta(old ? old->is_stats.rlentime : 0,
641 		    new->is_stats.rlentime);
642 		if (r_delta) {
643 			avr = (double)r_delta;
644 			avr /= hr_etime;
645 		} else
646 			avr = 0.0;
647 
648 		/*
649 		 * Average wait service time in milliseconds
650 		 */
651 		if (tps > 0.0 && (avw != 0.0 || avr != 0.0)) {
652 			mtps = 1000.0 / tps;
653 			if (avw != 0.0)
654 				wserv = avw * mtps;
655 			else
656 				wserv = 0.0;
657 
658 			if (avr != 0.0)
659 				rserv = avr * mtps;
660 			else
661 				rserv = 0.0;
662 			serv = rserv + wserv;
663 		} else {
664 			rserv = 0.0;
665 			wserv = 0.0;
666 			serv = 0.0;
667 		}
668 
669 		/* % of time there is a transaction waiting for service */
670 		t_delta = hrtime_delta(old ? old->is_stats.wtime : 0,
671 		    new->is_stats.wtime);
672 		if (t_delta) {
673 			w_pct = (double)t_delta;
674 			w_pct /= hr_etime;
675 			w_pct *= 100.0;
676 
677 			/*
678 			 * Average the wait queue utilization over the
679 			 * the controller's devices, if this is a controller.
680 			 */
681 			if (new->is_type == IODEV_CONTROLLER)
682 				w_pct /= new->is_nr_children;
683 		} else
684 			w_pct = 0.0;
685 
686 		/* % of time there is a transaction running */
687 		t_delta = hrtime_delta(old ? old->is_stats.rtime : 0,
688 		    new->is_stats.rtime);
689 		if (t_delta) {
690 			r_pct = (double)t_delta;
691 			r_pct /= hr_etime;
692 			r_pct *= 100.0;
693 
694 			/*
695 			 * Average the percent busy over the controller's
696 			 * devices, if this is a controller.
697 			 */
698 			if (new->is_type == IODEV_CONTROLLER)
699 				w_pct /= new->is_nr_children;
700 		} else {
701 			r_pct = 0.0;
702 		}
703 
704 		/* % of time there is a transaction running */
705 		if (do_interval) {
706 			rps	*= etime;
707 			wps	*= etime;
708 			tps	*= etime;
709 			krps	*= etime;
710 			kwps	*= etime;
711 			kps	*= etime;
712 		}
713 	}
714 
715 	if (do_disk & (DISK_EXTENDED | DISK_ERRORS)) {
716 		if ((!do_conversions) && ((suppress_zero == 0) ||
717 		    ((do_disk & DISK_EXTENDED) == 0))) {
718 			if (do_raw == 0)
719 				push_out(DISPLAYED_NAME_FORMAT,
720 				    disk_name);
721 			else
722 				push_out(disk_name);
723 		}
724 	}
725 
726 	switch (do_disk & DISK_IO_MASK) {
727 	    case DISK_OLD:
728 		if (do_raw == 0)
729 			fstr = "%3.0f %3.0f %4.0f  ";
730 		else
731 			fstr = "%.0f,%.0f,%.0f";
732 		push_out(fstr, kps, tps, serv);
733 		break;
734 	    case DISK_NEW:
735 		if (do_raw == 0)
736 			fstr = "%3.0f %3.0f %4.1f  ";
737 		else
738 			fstr = "%.0f,%.0f,%.1f";
739 		push_out(fstr, rps, wps, r_pct);
740 		break;
741 	    case DISK_EXTENDED:
742 		if (suppress_zero) {
743 			if (fzero(rps) && fzero(wps) && fzero(krps) &&
744 			    fzero(kwps) && fzero(avw) && fzero(avr) &&
745 			    fzero(serv) && fzero(w_pct) && fzero(r_pct))
746 				doit = 0;
747 			else if (do_conversions == 0) {
748 				if (do_raw == 0)
749 					push_out(DISPLAYED_NAME_FORMAT,
750 					    disk_name);
751 				else
752 					push_out(disk_name);
753 			}
754 		}
755 		if (doit) {
756 			if (!do_conversions) {
757 				if (do_raw == 0) {
758 					fstr = " %6.1f %6.1f %6.1f %6.1f "
759 						"%4.1f %4.1f %6.1f %3.0f "
760 						"%3.0f ";
761 				} else {
762 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
763 						"%.1f,%.0f,%.0f";
764 				}
765 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
766 				    serv, w_pct, r_pct);
767 			} else {
768 				if (do_raw == 0) {
769 					fstr = " %6.1f %6.1f %6.1f %6.1f "
770 						"%4.1f %4.1f %6.1f %6.1f "
771 						"%3.0f %3.0f ";
772 				} else {
773 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
774 						"%.1f,%.1f,%.0f,%.0f";
775 				}
776 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
777 				    wserv, rserv, w_pct, r_pct);
778 			}
779 		}
780 		break;
781 	}
782 
783 	if (do_disk & DISK_ERRORS) {
784 		if ((do_disk == DISK_ERRORS)) {
785 			if (do_raw == 0)
786 				push_out(two_blanks);
787 		}
788 
789 		if (new->is_errors.ks_data) {
790 			kstat_named_t *knp;
791 			char *efstr;
792 
793 			if (do_raw == 0)
794 				efstr = "%3u ";
795 			else
796 				efstr = "%u";
797 			toterrs = 0;
798 			knp = KSTAT_NAMED_PTR(&new->is_errors);
799 			for (i = 0; i < 3; i++) {
800 				switch (knp[i].data_type) {
801 					case KSTAT_DATA_ULONG:
802 						push_out(efstr,
803 						    knp[i].value.ui32);
804 						toterrs += knp[i].value.ui32;
805 						break;
806 					case KSTAT_DATA_ULONGLONG:
807 						/*
808 						 * We're only set up to
809 						 * write out the low
810 						 * order 32-bits so
811 						 * just grab that.
812 						 */
813 						push_out(efstr,
814 						    knp[i].value.ui32);
815 						toterrs += knp[i].value.ui32;
816 						break;
817 					default:
818 						break;
819 				}
820 			}
821 			push_out(efstr, toterrs);
822 		} else {
823 			if (do_raw == 0)
824 				push_out("  0   0   0   0 ");
825 			else
826 				push_out("0,0,0,0");
827 		}
828 
829 	}
830 
831 	if (suppress_zero == 0 || doit == 1) {
832 		if ((do_disk & (DISK_EXTENDED | DISK_ERRORS)) &&
833 			do_conversions) {
834 			push_out("%s", disk_name);
835 			if (show_mountpts && new->is_dname) {
836 				mnt_t *mount_pt;
837 				char *lu;
838 				char lub[SMALL_SCRATCH_BUFLEN];
839 
840 				lu = strrchr(new->is_dname, '/');
841 				if (lu) {
842 					if (strcmp(disk_name, lu) == 0)
843 						lu = new->is_dname;
844 					else {
845 						*lu = 0;
846 						(void) strcpy(lub,
847 						    new->is_dname);
848 						*lu = '/';
849 						(void) strcat(lub, "/");
850 						(void) strcat(lub,
851 						    disk_name);
852 						lu = lub;
853 					}
854 				} else
855 					lu = disk_name;
856 				mount_pt = lookup_mntent_byname(lu);
857 				if (mount_pt) {
858 					if (do_raw == 0)
859 						push_out(" (%s)",
860 						    mount_pt->mount_point);
861 					else
862 						push_out("(%s)",
863 						    mount_pt->mount_point);
864 				}
865 			}
866 		}
867 	}
868 
869 	if ((do_disk & PRINT_VERTICAL) && show_disk_mode != SHOW_FIRST_ONLY)
870 		do_newline();
871 
872 	if (count != NULL)
873 		(*count)++;
874 }
875 
876 static void
877 usage(void)
878 {
879 	(void) fprintf(stderr,
880 	    "Usage: iostat [-cCdDeEiImMnpPrstxXz] "
881 	    " [-l n] [-T d|u] [disk ...] [interval [count]]\n"
882 	    "\t\t-c: 	report percentage of time system has spent\n"
883 	    "\t\t\tin user/system/wait/idle mode\n"
884 	    "\t\t-C: 	report disk statistics by controller\n"
885 	    "\t\t-d: 	display disk Kb/sec, transfers/sec, avg. \n"
886 	    "\t\t\tservice time in milliseconds  \n"
887 	    "\t\t-D: 	display disk reads/sec, writes/sec, \n"
888 	    "\t\t\tpercentage disk utilization \n"
889 	    "\t\t-e: 	report device error summary statistics\n"
890 	    "\t\t-E: 	report extended device error statistics\n"
891 	    "\t\t-i:	show device IDs for -E output\n"
892 	    "\t\t-I: 	report the counts in each interval,\n"
893 	    "\t\t\tinstead of rates, where applicable\n"
894 	    "\t\t-l n:	Limit the number of disks to n\n"
895 	    "\t\t-m: 	Display mount points (most useful with -p)\n"
896 	    "\t\t-M: 	Display data throughput in MB/sec "
897 	    "instead of Kb/sec\n"
898 	    "\t\t-n: 	convert device names to cXdYtZ format\n"
899 	    "\t\t-p: 	report per-partition disk statistics\n"
900 	    "\t\t-P: 	report per-partition disk statistics only,\n"
901 	    "\t\t\tno per-device disk statistics\n"
902 	    "\t\t-r: 	Display data in comma separated format\n"
903 	    "\t\t-s: 	Suppress state change messages\n"
904 	    "\t\t-T d|u	Display a timestamp in date (d) or unix "
905 	    "time_t (u)\n"
906 	    "\t\t-t: 	display chars read/written to terminals\n"
907 	    "\t\t-x: 	display extended disk statistics\n"
908 	    "\t\t-X: 	display I/O path statistics\n"
909 	    "\t\t-z: 	Suppress entries with all zero values\n");
910 	exit(1);
911 }
912 
913 /*ARGSUSED*/
914 static void
915 show_disk_errors(void *v1, void *v2, void *d)
916 {
917 	struct iodev_snapshot *disk = (struct iodev_snapshot *)v2;
918 	kstat_named_t *knp;
919 	size_t  col;
920 	int	i, len;
921 	char	*dev_name = disk->is_name;
922 
923 	if (disk->is_errors.ks_ndata == 0)
924 		return;
925 	if (disk->is_type == IODEV_CONTROLLER)
926 		return;
927 
928 	if (disk->is_pretty)
929 		dev_name = disk->is_pretty;
930 
931 	len = strlen(dev_name);
932 	if (len > 20)
933 		push_out("%s ", dev_name);
934 	else if (len > 16)
935 		push_out("%-20.20s ", dev_name);
936 	else {
937 		if (do_conversions)
938 			push_out("%-16.16s ", dev_name);
939 		else
940 			push_out("%-9.9s ", dev_name);
941 	}
942 	col = 0;
943 
944 	knp = KSTAT_NAMED_PTR(&disk->is_errors);
945 	for (i = 0; i < disk->is_errors.ks_ndata; i++) {
946 		/* skip kstats that the driver did not kstat_named_init */
947 		if (knp[i].name[0] == 0)
948 			continue;
949 
950 		col += strlen(knp[i].name);
951 
952 		switch (knp[i].data_type) {
953 			case KSTAT_DATA_CHAR:
954 				if ((strcmp(knp[i].name, "Serial No") == 0) &&
955 				    do_devid) {
956 					if (disk->is_devid) {
957 						push_out("Device Id: %s ",
958 						    disk->is_devid);
959 						col += strlen(disk->is_devid);
960 					} else
961 						push_out("Device Id: ");
962 				} else {
963 					push_out("%s: %-.16s ", knp[i].name,
964 					    &knp[i].value.c[0]);
965 					col += strlen(&knp[i].value.c[0]);
966 				}
967 				break;
968 			case KSTAT_DATA_ULONG:
969 				push_out("%s: %u ", knp[i].name,
970 				    knp[i].value.ui32);
971 				col += 4;
972 				break;
973 			case KSTAT_DATA_ULONGLONG:
974 				if (strcmp(knp[i].name, "Size") == 0) {
975 					push_out("%s: %2.2fGB <%llu bytes>\n",
976 					    knp[i].name,
977 					    (float)knp[i].value.ui64 /
978 					    DISK_GIGABYTE,
979 					    knp[i].value.ui64);
980 					col = 0;
981 					break;
982 				}
983 				push_out("%s: %u ", knp[i].name,
984 				    knp[i].value.ui32);
985 				col += 4;
986 				break;
987 			}
988 		if ((col >= 62) || (i == 2)) {
989 			do_newline();
990 			col = 0;
991 		}
992 	}
993 	if (col > 0) {
994 		do_newline();
995 	}
996 	do_newline();
997 }
998 
999 void
1000 do_args(int argc, char **argv)
1001 {
1002 	int 		c;
1003 	int 		errflg = 0;
1004 	extern char 	*optarg;
1005 	extern int 	optind;
1006 
1007 	while ((c = getopt(argc, argv, "tdDxXCciIpPnmMeEszrT:l:")) != EOF)
1008 		switch (c) {
1009 		case 't':
1010 			do_tty++;
1011 			break;
1012 		case 'd':
1013 			do_disk |= DISK_OLD;
1014 			break;
1015 		case 'D':
1016 			do_disk |= DISK_NEW;
1017 			break;
1018 		case 'x':
1019 			do_disk |= DISK_EXTENDED;
1020 			break;
1021 		case 'X':
1022 			do_disk |= DISK_IOPATH;
1023 			break;
1024 		case 'C':
1025 			do_controller++;
1026 			break;
1027 		case 'c':
1028 			do_cpu++;
1029 			break;
1030 		case 'I':
1031 			do_interval++;
1032 			break;
1033 		case 'p':
1034 			do_partitions++;
1035 			break;
1036 		case 'P':
1037 			do_partitions_only++;
1038 			break;
1039 		case 'n':
1040 			do_conversions++;
1041 			break;
1042 		case 'M':
1043 			do_megabytes++;
1044 			break;
1045 		case 'e':
1046 			do_disk |= DISK_ERRORS;
1047 			break;
1048 		case 'E':
1049 			do_disk |= DISK_EXTENDED_ERRORS;
1050 			break;
1051 		case 'i':
1052 			do_devid = 1;
1053 			break;
1054 		case 's':
1055 			suppress_state = 1;
1056 			break;
1057 		case 'z':
1058 			suppress_zero = 1;
1059 			break;
1060 		case 'm':
1061 			show_mountpts = 1;
1062 			break;
1063 		case 'T':
1064 			if (optarg) {
1065 				if (*optarg == 'u')
1066 					do_timestamp = UDATE;
1067 				else if (*optarg == 'd')
1068 					do_timestamp = CDATE;
1069 				else
1070 					errflg++;
1071 			} else
1072 				errflg++;
1073 			break;
1074 		case 'r':
1075 			do_raw = 1;
1076 			break;
1077 		case 'l':
1078 			df.if_max_iodevs = safe_strtoi(optarg, "invalid limit");
1079 			if (df.if_max_iodevs < 1)
1080 				usage();
1081 			break;
1082 		case '?':
1083 			errflg++;
1084 	}
1085 
1086 	if ((do_disk & DISK_OLD) && (do_disk & DISK_NEW)) {
1087 		(void) fprintf(stderr, "-d and -D are incompatible.\n");
1088 		usage();
1089 	}
1090 
1091 	if (errflg) {
1092 		usage();
1093 	}
1094 	/* if no output classes explicity specified, use defaults */
1095 	if (do_tty == 0 && do_disk == 0 && do_cpu == 0)
1096 		do_tty = do_cpu = 1, do_disk = DISK_OLD;
1097 
1098 	/*
1099 	 * If conflicting options take the preferred
1100 	 * -D and -x result in -x
1101 	 * -d or -D and -e or -E gives only whatever -d or -D was specified
1102 	 */
1103 	if ((do_disk & DISK_EXTENDED) && (do_disk & DISK_NORMAL))
1104 		do_disk &= ~DISK_NORMAL;
1105 	if ((do_disk & DISK_NORMAL) && (do_disk & DISK_ERROR_MASK))
1106 		do_disk &= ~DISK_ERROR_MASK;
1107 
1108 	/*
1109 	 * I/O path stats are only available with extended (-x) stats
1110 	 */
1111 	if ((do_disk & DISK_IOPATH) && !(do_disk & DISK_EXTENDED))
1112 		do_disk &= ~DISK_IOPATH;
1113 
1114 	/* nfs, tape, always shown */
1115 	df.if_allowed_types = IODEV_NFS | IODEV_TAPE;
1116 
1117 	/*
1118 	 * If limit == 0 then no command line limit was set, else if any of
1119 	 * the flags that cause unlimited disks were not set,
1120 	 * use the default of 4
1121 	 */
1122 	if (df.if_max_iodevs == 0) {
1123 		df.if_max_iodevs = DEFAULT_LIMIT;
1124 		df.if_skip_floppy = 1;
1125 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS |
1126 		    DISK_EXTENDED_ERRORS)) {
1127 			df.if_max_iodevs = UNLIMITED_IODEVS;
1128 			df.if_skip_floppy = 0;
1129 		}
1130 	}
1131 	if (do_disk) {
1132 		size_t count = 0;
1133 		size_t i = optind;
1134 
1135 		while (i < argc && !isdigit(argv[i][0])) {
1136 			count++;
1137 			i++;
1138 		}
1139 
1140 		/*
1141 		 * "Note:  disks  explicitly  requested
1142 		 * are not subject to this disk limit"
1143 		 */
1144 		if (count > df.if_max_iodevs)
1145 			df.if_max_iodevs = count;
1146 		df.if_names = safe_alloc(count * sizeof (char *));
1147 		(void) memset(df.if_names, 0, count * sizeof (char *));
1148 
1149 		while (optind < argc && !isdigit(argv[optind][0]))
1150 			df.if_names[df.if_nr_names++] = argv[optind++];
1151 	}
1152 	if (optind < argc) {
1153 		interval = safe_strtoi(argv[optind], "invalid interval");
1154 		if (interval < 1)
1155 			fail(0, "invalid interval");
1156 		optind++;
1157 
1158 		if (optind < argc) {
1159 			iter = safe_strtoi(argv[optind], "invalid count");
1160 			if (iter < 1)
1161 				fail(0, "invalid count");
1162 			optind++;
1163 		}
1164 	}
1165 	if (interval == 0)
1166 		iter = 1;
1167 	if (optind < argc)
1168 		usage();
1169 }
1170 
1171 /*
1172  * Driver for doing the extended header formatting. Will produce
1173  * the function stack needed to output an extended header based
1174  * on the options selected.
1175  */
1176 
1177 void
1178 do_format(void)
1179 {
1180 	char	header[SMALL_SCRATCH_BUFLEN];
1181 	char 	ch;
1182 	char 	iosz;
1183 	const char    *fstr;
1184 
1185 	disk_header[0] = 0;
1186 	ch = (do_interval ? 'i' : 's');
1187 	iosz = (do_megabytes ? 'M' : 'k');
1188 	if (do_disk & DISK_ERRORS) {
1189 		if (do_raw == 0) {
1190 			(void) sprintf(header, "s/w h/w trn tot ");
1191 		} else
1192 			(void) sprintf(header, "s/w,h/w,trn,tot");
1193 	} else
1194 		*header = NULL;
1195 	switch (do_disk & DISK_IO_MASK) {
1196 		case DISK_OLD:
1197 			if (do_raw == 0)
1198 				fstr = "%cp%c tp%c serv  ";
1199 			else
1200 				fstr = "%cp%c,tp%c,serv";
1201 			(void) snprintf(disk_header, sizeof (disk_header),
1202 			    fstr, iosz, ch, ch);
1203 			break;
1204 		case DISK_NEW:
1205 			if (do_raw == 0)
1206 				fstr = "rp%c wp%c util  ";
1207 			else
1208 				fstr = "%rp%c,wp%c,util";
1209 			(void) snprintf(disk_header, sizeof (disk_header),
1210 			    fstr, ch, ch);
1211 			break;
1212 		case DISK_EXTENDED:
1213 			if (!do_conversions) {
1214 				if (do_raw == 0)
1215 					fstr = "device       r/%c    w/%c   "
1216 					    "%cr/%c   %cw/%c wait actv  "
1217 					    "svc_t  %%%%w  %%%%b %s";
1218 				else
1219 					fstr = "device,r/%c,w/%c,%cr/%c,%cw/%c,"
1220 						"wait,actv,svc_t,%%%%w,"
1221 						"%%%%b,%s";
1222 				(void) snprintf(disk_header,
1223 				    sizeof (disk_header),
1224 				    fstr, ch, ch, iosz, ch, iosz,
1225 				    ch, header);
1226 			} else {
1227 				if (do_raw == 0) {
1228 					fstr = "    r/%c    w/%c   %cr/%c   "
1229 					    "%cw/%c wait actv wsvc_t asvc_t  "
1230 					    "%%%%w  %%%%b %sdevice";
1231 				} else {
1232 					fstr = "r/%c,w/%c,%cr/%c,%cw/%c,"
1233 					    "wait,actv,wsvc_t,asvc_t,"
1234 					    "%%%%w,%%%%b,%sdevice";
1235 				}
1236 				(void) snprintf(disk_header,
1237 				    sizeof (disk_header),
1238 				    fstr, ch, ch, iosz, ch, iosz,
1239 				    ch, header);
1240 			}
1241 			break;
1242 		default:
1243 			break;
1244 	}
1245 	if (do_disk == DISK_ERRORS) {
1246 		char *sep;
1247 
1248 		if (!do_conversions) {
1249 			if (do_raw == 0) {
1250 				sep = "     ";
1251 			} else
1252 				sep = ",";
1253 			(void) snprintf(disk_header, sizeof (disk_header),
1254 			    "%s%s%s", "device", sep, header);
1255 		} else {
1256 			if (do_raw == 0) {
1257 				(void) snprintf(disk_header,
1258 				    sizeof (disk_header),
1259 				    "  %s", header);
1260 			} else
1261 				(void) strcpy(disk_header, header);
1262 		}
1263 	} else {
1264 		/*
1265 		 * Need to subtract two characters for the % escape in
1266 		 * the string.
1267 		 */
1268 		dh_len = strlen(disk_header) - 2;
1269 	}
1270 
1271 	if (do_timestamp)
1272 		setup(print_timestamp);
1273 
1274 	/*
1275 	 * -n *and* (-E *or* -e *or* -x)
1276 	 */
1277 	if (do_conversions && (do_disk & PRINT_VERTICAL)) {
1278 		if (do_tty)
1279 			setup(print_tty_hdr1);
1280 		if (do_cpu)
1281 			setup(print_cpu_hdr1);
1282 		if (do_tty || do_cpu)
1283 			setup(do_newline);
1284 		if (do_tty)
1285 			setup(print_tty_hdr2);
1286 		if (do_cpu)
1287 			setup(print_cpu_hdr2);
1288 		if (do_tty || do_cpu)
1289 			setup(do_newline);
1290 		if (do_tty)
1291 			setup(print_tty_data);
1292 		if (do_cpu)
1293 			setup(print_cpu_data);
1294 		if (do_tty || do_cpu)
1295 			setup(do_newline);
1296 		printxhdr();
1297 
1298 		setup(show_all_disks);
1299 	} else {
1300 		/*
1301 		 * These unholy gymnastics are necessary to place CPU/tty
1302 		 * data to the right of the disks/errors for the first
1303 		 * line in vertical mode.
1304 		 */
1305 		if (do_disk & PRINT_VERTICAL) {
1306 			printxhdr();
1307 
1308 			setup(show_first_disk);
1309 			if (do_tty)
1310 				setup(print_tty_data);
1311 			if (do_cpu)
1312 				setup(print_cpu_data);
1313 			setup(do_newline);
1314 
1315 			setup(show_other_disks);
1316 		} else {
1317 			setup(hdrout);
1318 			if (do_tty)
1319 				setup(print_tty_data);
1320 			setup(show_all_disks);
1321 			if (do_cpu)
1322 				setup(print_cpu_data);
1323 		}
1324 
1325 		setup(do_newline);
1326 	}
1327 	if (do_disk & DISK_EXTENDED_ERRORS)
1328 		setup(disk_errors);
1329 }
1330 
1331 /*
1332  * Add a new function to the list of functions
1333  * for this invocation. Once on the stack the
1334  * function is never removed nor does its place
1335  * change.
1336  */
1337 void
1338 setup(void (*nfunc)(void))
1339 {
1340 	format_t *tmp;
1341 
1342 	tmp = safe_alloc(sizeof (format_t));
1343 	tmp->nfunc = nfunc;
1344 	tmp->next = 0;
1345 	if (formatter_end)
1346 		formatter_end->next = tmp;
1347 	else
1348 		formatter_list = tmp;
1349 	formatter_end = tmp;
1350 
1351 }
1352 
1353 /*
1354  * The functions after this comment are devoted to printing
1355  * various parts of the header. They are selected based on the
1356  * options provided when the program was invoked. The functions
1357  * are either directly invoked in printhdr() or are indirectly
1358  * invoked by being placed on the list of functions used when
1359  * extended headers are used.
1360  */
1361 void
1362 print_tty_hdr1(void)
1363 {
1364 	char *fstr;
1365 	char *dstr;
1366 
1367 	if (do_raw == 0) {
1368 		fstr = "%10.10s";
1369 		dstr = "tty    ";
1370 	} else {
1371 		fstr = "%s";
1372 		dstr = "tty";
1373 	}
1374 	push_out(fstr, dstr);
1375 }
1376 
1377 void
1378 print_tty_hdr2(void)
1379 {
1380 	if (do_raw == 0)
1381 		push_out("%-10.10s", " tin tout");
1382 	else
1383 		push_out("tin,tout");
1384 }
1385 
1386 void
1387 print_cpu_hdr1(void)
1388 {
1389 	char *dstr;
1390 
1391 	if (do_raw == 0)
1392 		dstr = "     cpu";
1393 	else
1394 		dstr = "cpu";
1395 	push_out(dstr);
1396 }
1397 
1398 void
1399 print_cpu_hdr2(void)
1400 {
1401 	char *dstr;
1402 
1403 	if (do_raw == 0)
1404 		dstr = " us sy wt id";
1405 	else
1406 		dstr = "us,sy,wt,id";
1407 	push_out(dstr);
1408 }
1409 
1410 /*
1411  * Assumption is that tty data is always first - no need for raw mode leading
1412  * comma.
1413  */
1414 void
1415 print_tty_data(void)
1416 {
1417 	char *fstr;
1418 	uint64_t deltas;
1419 	double raw;
1420 	double outch;
1421 	kstat_t *oldks = NULL;
1422 
1423 	if (oldss)
1424 		oldks = &oldss->s_sys.ss_agg_sys;
1425 
1426 	if (do_raw == 0)
1427 		fstr = " %3.0f %4.0f ";
1428 	else
1429 		fstr = "%.0f,%.0f";
1430 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "rawch");
1431 	raw = deltas;
1432 	raw /= getime;
1433 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "outch");
1434 	outch = deltas;
1435 	outch /= getime;
1436 	push_out(fstr, raw, outch);
1437 }
1438 
1439 /*
1440  * Write out CPU data
1441  */
1442 void
1443 print_cpu_data(void)
1444 {
1445 	char *fstr;
1446 	uint64_t idle;
1447 	uint64_t user;
1448 	uint64_t kern;
1449 	uint64_t wait;
1450 	kstat_t *oldks = NULL;
1451 
1452 	if (oldss)
1453 		oldks = &oldss->s_sys.ss_agg_sys;
1454 
1455 	if (do_raw == 0)
1456 		fstr = " %2.0f %2.0f %2.0f %2.0f";
1457 	else
1458 		fstr = "%.0f,%.0f,%.0f,%.0f";
1459 
1460 	idle = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_idle");
1461 	user = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_user");
1462 	kern = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_kernel");
1463 	wait = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_wait");
1464 	push_out(fstr, user * percent, kern * percent,
1465 		wait * percent, idle * percent);
1466 }
1467 
1468 /*
1469  * Emit the appropriate header.
1470  */
1471 void
1472 hdrout(void)
1473 {
1474 	if (do_raw == 0) {
1475 		if (--tohdr == 0)
1476 			printhdr(0);
1477 	} else if (hdr_out == 0) {
1478 		printhdr(0);
1479 		hdr_out = 1;
1480 	}
1481 }
1482 
1483 /*
1484  * Write out disk errors when -E is specified.
1485  */
1486 void
1487 disk_errors(void)
1488 {
1489 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk_errors, NULL);
1490 }
1491 
1492 void
1493 show_first_disk(void)
1494 {
1495 	int count = 0;
1496 
1497 	show_disk_mode = SHOW_FIRST_ONLY;
1498 
1499 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1500 }
1501 
1502 void
1503 show_other_disks(void)
1504 {
1505 	int count = 0;
1506 
1507 	show_disk_mode = SHOW_SECOND_ONWARDS;
1508 
1509 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1510 }
1511 
1512 void
1513 show_all_disks(void)
1514 {
1515 	int count = 0;
1516 
1517 	show_disk_mode = SHOW_ALL;
1518 
1519 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1520 }
1521 
1522 /*
1523  * Write a newline out and clear the lineout flag.
1524  */
1525 static void
1526 do_newline(void)
1527 {
1528 	if (lineout) {
1529 		(void) putchar('\n');
1530 		lineout = 0;
1531 	}
1532 }
1533 
1534 /*
1535  * Generalized printf function that determines what extra
1536  * to print out if we're in raw mode. At this time we
1537  * don't care about errors.
1538  */
1539 static void
1540 push_out(const char *message, ...)
1541 {
1542 	va_list args;
1543 
1544 	va_start(args, message);
1545 	if (do_raw && lineout == 1)
1546 		(void) putchar(',');
1547 	(void) vprintf(message, args);
1548 	va_end(args);
1549 	lineout = 1;
1550 }
1551 
1552 /*
1553  * Emit the header string when -e is specified.
1554  */
1555 static void
1556 print_err_hdr(void)
1557 {
1558 	char obuf[SMALL_SCRATCH_BUFLEN];
1559 
1560 	if (do_conversions == 0) {
1561 		if (!(do_disk & DISK_EXTENDED)) {
1562 			(void) snprintf(obuf, sizeof (obuf),
1563 			    "%11s", one_blank);
1564 			push_out(obuf);
1565 		}
1566 	} else if (do_disk == DISK_ERRORS)
1567 		push_out(two_blanks);
1568 	else
1569 		push_out(one_blank);
1570 	push_out("---- errors --- ");
1571 }
1572 
1573 /*
1574  * Emit the header string when -e is specified.
1575  */
1576 static void
1577 print_disk_header(void)
1578 {
1579 	push_out(disk_header);
1580 }
1581 
1582 /*
1583  * Write out a timestamp. Format is all that goes out on
1584  * the line so no use of push_out.
1585  *
1586  * Write out as decimal reprentation of time_t value
1587  * (-T u was specified) or the string returned from
1588  * ctime() (-T d was specified).
1589  */
1590 static void
1591 print_timestamp(void)
1592 {
1593 	time_t t;
1594 
1595 	if (time(&t) != -1) {
1596 		if (do_timestamp == UDATE) {
1597 			(void) printf("%ld\n", t);
1598 		} else if (do_timestamp == CDATE) {
1599 			char *cpt;
1600 
1601 			cpt = ctime(&t);
1602 			if (cpt) {
1603 				(void) fputs(cpt, stdout);
1604 			}
1605 		}
1606 	}
1607 }
1608 
1609 /*
1610  * No, UINTMAX_MAX isn't the right thing here since
1611  * it is #defined to be either INT32_MAX or INT64_MAX
1612  * depending on the whether _LP64 is defined.
1613  *
1614  * We want to handle the odd future case of having
1615  * ulonglong_t be more than 64 bits but we have
1616  * no nice #define MAX value we can drop in place
1617  * without having to change this code in the future.
1618  */
1619 
1620 u_longlong_t
1621 ull_delta(u_longlong_t old, u_longlong_t new)
1622 {
1623 	if (new >= old)
1624 		return (new - old);
1625 	else
1626 		return ((UINT64_MAX - old) + new + 1);
1627 }
1628 
1629 /*
1630  * Return the number of ticks delta between two hrtime_t
1631  * values. Attempt to cater for various kinds of overflow
1632  * in hrtime_t - no matter how improbable.
1633  */
1634 uint64_t
1635 hrtime_delta(hrtime_t old, hrtime_t new)
1636 {
1637 	uint64_t del;
1638 
1639 	if ((new >= old) && (old >= 0L))
1640 		return (new - old);
1641 	else {
1642 		/*
1643 		 * We've overflowed the positive portion of an
1644 		 * hrtime_t.
1645 		 */
1646 		if (new < 0L) {
1647 			/*
1648 			 * The new value is negative. Handle the
1649 			 * case where the old value is positive or
1650 			 * negative.
1651 			 */
1652 			uint64_t n1;
1653 			uint64_t o1;
1654 
1655 			n1 = -new;
1656 			if (old > 0L)
1657 				return (n1 - old);
1658 			else {
1659 				o1 = -old;
1660 				del = n1 - o1;
1661 				return (del);
1662 			}
1663 		} else {
1664 			/*
1665 			 * Either we've just gone from being negative
1666 			 * to positive *or* the last entry was positive
1667 			 * and the new entry is also positive but *less*
1668 			 * than the old entry. This implies we waited
1669 			 * quite a few days on a very fast system between
1670 			 * iostat displays.
1671 			 */
1672 			if (old < 0L) {
1673 				uint64_t o2;
1674 
1675 				o2 = -old;
1676 				del = UINT64_MAX - o2;
1677 			} else {
1678 				del = UINT64_MAX - old;
1679 			}
1680 			del += new;
1681 			return (del);
1682 		}
1683 	}
1684 }
1685 
1686 /*
1687  * Take the difference of an unsigned 32
1688  * bit int attempting to cater for
1689  * overflow.
1690  */
1691 uint_t
1692 u32_delta(uint_t old, uint_t new)
1693 {
1694 	if (new >= old)
1695 		return (new - old);
1696 	else
1697 		return ((UINT32_MAX - old) + new + 1);
1698 }
1699 
1700 /*
1701  * Create and arm the timer. Used only when an interval has been specified.
1702  * Used in lieu of poll to ensure that we provide info for exactly the
1703  * desired period.
1704  */
1705 void
1706 set_timer(int interval)
1707 {
1708 	timer_t t_id;
1709 	itimerspec_t time_struct;
1710 	struct sigevent sig_struct;
1711 	struct sigaction act;
1712 
1713 	bzero(&sig_struct, sizeof (struct sigevent));
1714 	bzero(&act, sizeof (struct sigaction));
1715 
1716 	/* Create timer */
1717 	sig_struct.sigev_notify = SIGEV_SIGNAL;
1718 	sig_struct.sigev_signo = SIGUSR1;
1719 	sig_struct.sigev_value.sival_int = 0;
1720 
1721 	if (timer_create(CLOCK_REALTIME, &sig_struct, &t_id) != 0) {
1722 		fail(1, "Timer creation failed");
1723 	}
1724 
1725 	act.sa_handler = handle_sig;
1726 
1727 	if (sigaction(SIGUSR1, &act, NULL) != 0) {
1728 		fail(1, "Could not set up signal handler");
1729 	}
1730 
1731 	time_struct.it_value.tv_sec = interval;
1732 	time_struct.it_value.tv_nsec = 0;
1733 	time_struct.it_interval.tv_sec = interval;
1734 	time_struct.it_interval.tv_nsec = 0;
1735 
1736 	/* Arm timer */
1737 	if ((timer_settime(t_id, 0, &time_struct, NULL)) != 0) {
1738 		fail(1, "Setting timer failed");
1739 	}
1740 }
1741 /* ARGSUSED */
1742 void
1743 handle_sig(int x)
1744 {
1745 }
1746 
1747 /*
1748  * This is exactly what is needed for standard iostat output,
1749  * but make sure to use it only for that
1750  */
1751 #define	EPSILON	(0.1)
1752 static int
1753 fzero(double value)
1754 {
1755 	return (value >= 0.0 && value < EPSILON);
1756 }
1757 
1758 static int
1759 safe_strtoi(char const *val, char *errmsg)
1760 {
1761 	char *end;
1762 	long tmp;
1763 
1764 	errno = 0;
1765 	tmp = strtol(val, &end, 10);
1766 	if (*end != '\0' || errno)
1767 		fail(0, "%s %s", errmsg, val);
1768 	return ((int)tmp);
1769 }
1770