xref: /illumos-gate/usr/src/cmd/stat/iostat/iostat.c (revision 06e1a714)
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  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  *
25  * rewritten from UCB 4.13 83/09/25
26  * rewritten from SunOS 4.1 SID 1.18 89/10/06
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 #include <stdio.h>
32 #include <stdlib.h>
33 #include <stdarg.h>
34 #include <ctype.h>
35 #include <unistd.h>
36 #include <memory.h>
37 #include <errno.h>
38 #include <string.h>
39 #include <signal.h>
40 #include <sys/types.h>
41 #include <time.h>
42 #include <sys/time.h>
43 #include <sys/sysinfo.h>
44 #include <inttypes.h>
45 #include <strings.h>
46 #include <sys/systeminfo.h>
47 #include <kstat.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 /*
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 -Y) */
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	MAXNAMELEN
151 
152 static int	iodevs_nl;		/* name field width */
153 #define	IODEVS_NL_MIN		6	/* not too thin for "device" */
154 #define	IODEVS_NL_MAX		24	/* but keep full width under 80 */
155 
156 static	char	disk_header[132];
157 static	uint_t 	dh_len;			/* disk header length for centering */
158 static  int 	lineout;		/* data waiting to be printed? */
159 
160 static struct snapshot *newss;
161 static struct snapshot *oldss;
162 static	double	getime;			/* elapsed time */
163 static	double	percent;		/* 100 / etime */
164 
165 /*
166  * List of functions to be called which will construct the desired output
167  */
168 static format_t	*formatter_list;
169 static format_t *formatter_end;
170 
171 static u_longlong_t	ull_delta(u_longlong_t, u_longlong_t);
172 static uint_t 	u32_delta(uint_t, uint_t);
173 static void setup(void (*nfunc)(void));
174 static void print_timestamp(void);
175 static void print_tty_hdr1(void);
176 static void print_tty_hdr2(void);
177 static void print_cpu_hdr1(void);
178 static void print_cpu_hdr2(void);
179 static void print_tty_data(void);
180 static void print_cpu_data(void);
181 static void print_err_hdr(void);
182 static void print_disk_header(void);
183 static void hdrout(void);
184 static void disk_errors(void);
185 static void do_newline(void);
186 static void push_out(const char *, ...);
187 static void printhdr(int);
188 static void printxhdr(void);
189 static void usage(void);
190 static void do_args(int, char **);
191 static void do_format(void);
192 static void set_timer(int);
193 static void handle_sig(int);
194 static void show_all_disks(void);
195 static void show_first_disk(void);
196 static void show_other_disks(void);
197 static void show_disk_errors(void *, void *, void *);
198 static void write_core_header(void);
199 static int  fzero(double value);
200 static int  safe_strtoi(char const *val, char *errmsg);
201 
202 int
203 main(int argc, char **argv)
204 {
205 	enum snapshot_types types = SNAP_SYSTEM;
206 	kstat_ctl_t *kc;
207 	long hz;
208 	int iiter;
209 
210 	do_args(argc, argv);
211 
212 	/*
213 	 * iostat historically showed CPU changes, even though
214 	 * it doesn't provide much useful information
215 	 */
216 	types |= SNAP_CPUS;
217 
218 	if (do_disk)
219 		types |= SNAP_IODEVS;
220 
221 	if (do_disk && !do_partitions_only)
222 		df.if_allowed_types |= IODEV_DISK;
223 	if (do_disk & DISK_IOPATH_LI) {
224 		df.if_allowed_types |= IODEV_IOPATH_LTI;
225 		types |= SNAP_IOPATHS_LI;
226 	}
227 	if (do_disk & DISK_IOPATH_LTI) {
228 		df.if_allowed_types |= IODEV_IOPATH_LTI;
229 		types |= SNAP_IOPATHS_LTI;
230 	}
231 	if (do_disk & DISK_ERROR_MASK)
232 		types |= SNAP_IODEV_ERRORS;
233 	if (do_partitions || do_partitions_only)
234 		df.if_allowed_types |= IODEV_PARTITION;
235 	if (do_conversions)
236 		types |= SNAP_IODEV_PRETTY;
237 	if (do_devid)
238 		types |= SNAP_IODEV_DEVID;
239 	if (do_controller) {
240 		if (!(do_disk & PRINT_VERTICAL) ||
241 		    (do_disk & DISK_EXTENDED_ERRORS))
242 			fail(0, "-C can only be used with -e or -x.");
243 		types |= SNAP_CONTROLLERS;
244 		df.if_allowed_types |= IODEV_CONTROLLER;
245 	}
246 
247 	hz = sysconf(_SC_CLK_TCK);
248 
249 	/*
250 	 * Undocumented behavior - sending a SIGCONT will result
251 	 * in a new header being emitted. Used only if we're not
252 	 * doing extended headers. This is a historical
253 	 * artifact.
254 	 */
255 	if (!(do_disk & PRINT_VERTICAL))
256 		(void) signal(SIGCONT, printhdr);
257 
258 	if (interval)
259 		set_timer(interval);
260 
261 	kc = open_kstat();
262 	newss = acquire_snapshot(kc, types, &df);
263 
264 	/* compute width of "device" field */
265 	iodevs_nl = newss->s_iodevs_is_name_maxlen;
266 	iodevs_nl = (iodevs_nl < IODEVS_NL_MIN) ?
267 	    IODEVS_NL_MIN : iodevs_nl;
268 	iodevs_nl = (iodevs_nl > IODEVS_NL_MAX) ?
269 	    IODEVS_NL_MAX : iodevs_nl;
270 
271 	do_format();
272 
273 	iiter = iter;
274 	do {
275 		if (do_tty || do_cpu) {
276 			kstat_t *oldks;
277 			oldks = oldss ? &oldss->s_sys.ss_agg_sys : NULL;
278 			getime = cpu_ticks_delta(oldks,
279 			    &newss->s_sys.ss_agg_sys);
280 			percent = (getime > 0.0) ? 100.0 / getime : 0.0;
281 			getime = (getime / nr_active_cpus(newss)) / hz;
282 			if (getime == 0.0)
283 				getime = (double)interval;
284 			if (getime == 0.0 || do_interval)
285 				getime = 1.0;
286 		}
287 
288 		if (formatter_list) {
289 			format_t *tmp;
290 			tmp = formatter_list;
291 			while (tmp) {
292 				(tmp->nfunc)();
293 				tmp = tmp->next;
294 			}
295 			(void) fflush(stdout);
296 		}
297 
298 		/* only doing a single iteration, we are done */
299 		if (iiter == 1)
300 			continue;
301 
302 		if (interval > 0 && iter != 1)
303 			(void) pause();
304 
305 		free_snapshot(oldss);
306 		oldss = newss;
307 		newss = acquire_snapshot(kc, types, &df);
308 		iodevs_nl = (newss->s_iodevs_is_name_maxlen > iodevs_nl) ?
309 		    newss->s_iodevs_is_name_maxlen : iodevs_nl;
310 		iodevs_nl = (iodevs_nl < IODEVS_NL_MIN) ?
311 		    IODEVS_NL_MIN : iodevs_nl;
312 		iodevs_nl = (iodevs_nl > IODEVS_NL_MAX) ?
313 		    IODEVS_NL_MAX : iodevs_nl;
314 
315 		if (!suppress_state)
316 			snapshot_report_changes(oldss, newss);
317 
318 		/* if config changed, show stats from boot */
319 		if (snapshot_has_changed(oldss, newss)) {
320 			free_snapshot(oldss);
321 			oldss = NULL;
322 		}
323 
324 	} while (--iter);
325 
326 	free_snapshot(oldss);
327 	free_snapshot(newss);
328 	(void) kstat_close(kc);
329 	free(df.if_names);
330 	return (0);
331 }
332 
333 /*
334  * Some magic numbers used in header formatting.
335  *
336  * DISK_LEN = length of either "kps tps serv" or "wps rps util"
337  *	      using 0 as the first position
338  *
339  * DISK_ERROR_LEN = length of "s/w h/w trn tot" with one space on
340  *		either side. Does not use zero as first pos.
341  *
342  * DEVICE_LEN = length of "device" + 1 character.
343  */
344 
345 #define	DISK_LEN	11
346 #define	DISK_ERROR_LEN	16
347 #define	DEVICE_LEN	7
348 
349 /*ARGSUSED*/
350 static void
351 show_disk_name(void *v1, void *v2, void *data)
352 {
353 	struct iodev_snapshot *dev = (struct iodev_snapshot *)v2;
354 	size_t slen;
355 	char *name;
356 	char fbuf[SMALL_SCRATCH_BUFLEN];
357 
358 	if (dev == NULL)
359 		return;
360 
361 	name = do_conversions ? dev->is_pretty : dev->is_name;
362 	name = name ? name : dev->is_name;
363 
364 	if (!do_raw) {
365 		uint_t width;
366 
367 		slen = strlen(name);
368 		/*
369 		 * The length is less
370 		 * than the section
371 		 * which will be displayed
372 		 * on the next line.
373 		 * Center the entry.
374 		 */
375 
376 		width = (DISK_LEN + 1)/2 + (slen / 2);
377 		(void) snprintf(fbuf, sizeof (fbuf),
378 		    "%*s", width, name);
379 		name = fbuf;
380 		push_out("%-13.13s ", name);
381 	} else {
382 		push_out(name);
383 	}
384 }
385 
386 /*ARGSUSED*/
387 static void
388 show_disk_header(void *v1, void *v2, void *data)
389 {
390 	push_out(disk_header);
391 }
392 
393 /*
394  * Write out a two line header. What is written out depends on the flags
395  * selected but in the worst case consists of a tty header, a disk header
396  * providing information for 4 disks and a cpu header.
397  *
398  * The tty header consists of the word "tty" on the first line above the
399  * words "tin tout" on the next line. If present the tty portion consumes
400  * the first 10 characters of each line since "tin tout" is surrounded
401  * by single spaces.
402  *
403  * Each of the disk sections is a 14 character "block" in which the name of
404  * the disk is centered in the first 12 characters of the first line.
405  *
406  * The cpu section is an 11 character block with "cpu" centered over the
407  * section.
408  *
409  * The worst case should look as follows:
410  *
411  * 0---------1--------2---------3---------4---------5---------6---------7-------
412  *    tty        sd0           sd1           sd2           sd3           cpu
413  *  tin tout kps tps serv  kps tps serv  kps tps serv  kps tps serv  us sy wt id
414  *  NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NN NN NN NN
415  *
416  * When -D is specified, the disk header looks as follows (worst case):
417  *
418  * 0---------1--------2---------3---------4---------5---------6---------7-------
419  *     tty        sd0           sd1             sd2          sd3          cpu
420  *   tin tout rps wps util  rps wps util  rps wps util  rps wps util us sy wt id
421  *   NNN NNNN NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN  NNN NNN NNNN NN NN NN NN
422  */
423 static void
424 printhdr(int sig)
425 {
426 	/*
427 	 * If we're here because a signal fired, reenable the
428 	 * signal.
429 	 */
430 	if (sig)
431 		(void) signal(SIGCONT, printhdr);
432 	/*
433 	 * Horizontal mode headers
434 	 *
435 	 * First line
436 	 */
437 	if (do_tty)
438 		print_tty_hdr1();
439 
440 	if (do_disk & DISK_NORMAL) {
441 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
442 		    show_disk_name, NULL);
443 	}
444 
445 	if (do_cpu)
446 		print_cpu_hdr1();
447 	do_newline();
448 
449 	/*
450 	 * Second line
451 	 */
452 	if (do_tty)
453 		print_tty_hdr2();
454 
455 	if (do_disk & DISK_NORMAL) {
456 		(void) snapshot_walk(SNAP_IODEVS, NULL, newss,
457 		    show_disk_header, NULL);
458 	}
459 
460 	if (do_cpu)
461 		print_cpu_hdr2();
462 	do_newline();
463 
464 	tohdr = REPRINT;
465 }
466 
467 /*
468  * Write out the extended header centered over the core information.
469  */
470 static void
471 write_core_header(void)
472 {
473 	char *edev = "extended device statistics";
474 	uint_t lead_space_ct;
475 	uint_t follow_space_ct;
476 	size_t edevlen;
477 
478 	if (do_raw == 0) {
479 		/*
480 		 * The things we do to look nice...
481 		 *
482 		 * Center the core output header. Make sure we have the
483 		 * right number of trailing spaces for follow-on headers
484 		 * (i.e., cpu and/or tty and/or errors).
485 		 */
486 		edevlen = strlen(edev);
487 		lead_space_ct = dh_len - edevlen;
488 		lead_space_ct /= 2;
489 		if (lead_space_ct > 0) {
490 			follow_space_ct = dh_len - (lead_space_ct + edevlen);
491 			if (do_disk & DISK_ERRORS)
492 				follow_space_ct -= DISK_ERROR_LEN;
493 			if ((do_disk & DISK_EXTENDED) && do_conversions)
494 				follow_space_ct -= DEVICE_LEN;
495 
496 			push_out("%1$*2$.*2$s%3$s%4$*5$.*5$s", one_blank,
497 			    lead_space_ct, edev, one_blank, follow_space_ct);
498 		} else
499 			push_out("%56s", edev);
500 	} else
501 		push_out(edev);
502 }
503 
504 /*
505  * In extended mode headers, we don't want to reprint the header on
506  * signals as they are printed every time anyways.
507  */
508 static void
509 printxhdr(void)
510 {
511 
512 	/*
513 	 * Vertical mode headers
514 	 */
515 	if (do_disk & DISK_EXTENDED)
516 		setup(write_core_header);
517 	if (do_disk & DISK_ERRORS)
518 		setup(print_err_hdr);
519 
520 	if (do_conversions) {
521 		setup(do_newline);
522 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
523 			setup(print_disk_header);
524 		setup(do_newline);
525 	} else {
526 		if (do_tty)
527 			setup(print_tty_hdr1);
528 		if (do_cpu)
529 			setup(print_cpu_hdr1);
530 		setup(do_newline);
531 
532 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS))
533 			setup(print_disk_header);
534 		if (do_tty)
535 			setup(print_tty_hdr2);
536 		if (do_cpu)
537 			setup(print_cpu_hdr2);
538 		setup(do_newline);
539 	}
540 }
541 
542 /*
543  * Write out a line for this disk - note that show_disk writes out
544  * full lines or blocks for each selected disk.
545  */
546 static void
547 show_disk(void *v1, void *v2, void *data)
548 {
549 	struct iodev_snapshot *old = (struct iodev_snapshot *)v1;
550 	struct iodev_snapshot *new = (struct iodev_snapshot *)v2;
551 	int *count = (int *)data;
552 	double rps, wps, tps, mtps, krps, kwps, kps, avw, avr, w_pct, r_pct;
553 	double wserv, rserv, serv;
554 	double iosize;	/* kb/sec or MB/sec */
555 	double etime, hr_etime;
556 	char *disk_name;
557 	u_longlong_t ldeltas;
558 	uint_t udeltas;
559 	uint64_t t_delta;
560 	uint64_t w_delta;
561 	uint64_t r_delta;
562 	int doit = 1;
563 	int i;
564 	uint_t toterrs;
565 	char *fstr;
566 
567 	if (new == NULL)
568 		return;
569 
570 	switch (show_disk_mode) {
571 	case SHOW_FIRST_ONLY:
572 		if (count != NULL && *count)
573 			return;
574 		break;
575 
576 	case SHOW_SECOND_ONWARDS:
577 		if (count != NULL && !*count) {
578 			(*count)++;
579 			return;
580 		}
581 		break;
582 
583 	default:
584 		break;
585 	}
586 
587 	disk_name = do_conversions ? new->is_pretty : new->is_name;
588 	disk_name = disk_name ? disk_name : new->is_name;
589 
590 	/*
591 	 * Only do if we want IO stats - Avoids errors traveling this
592 	 * section if that's all we want to see.
593 	 */
594 	if (do_disk & DISK_IO_MASK) {
595 		if (old) {
596 			t_delta = hrtime_delta(old->is_snaptime,
597 			    new->is_snaptime);
598 		} else {
599 			t_delta = hrtime_delta(new->is_crtime,
600 			    new->is_snaptime);
601 		}
602 
603 		if (new->is_nr_children) {
604 			if (new->is_type == IODEV_CONTROLLER) {
605 				t_delta /= new->is_nr_children;
606 			} else if ((new->is_type == IODEV_IOPATH_LT) ||
607 			    (new->is_type == IODEV_IOPATH_LI)) {
608 				/* synthetic path */
609 				if (!old) {
610 					t_delta = new->is_crtime;
611 				}
612 				t_delta /= new->is_nr_children;
613 			}
614 		}
615 
616 		hr_etime = (double)t_delta;
617 		if (hr_etime == 0.0)
618 			hr_etime = (double)NANOSEC;
619 		etime = hr_etime / (double)NANOSEC;
620 
621 		/* reads per second */
622 		udeltas = u32_delta(old ? old->is_stats.reads : 0,
623 		    new->is_stats.reads);
624 		rps = (double)udeltas;
625 		rps /= etime;
626 
627 		/* writes per second */
628 		udeltas = u32_delta(old ? old->is_stats.writes : 0,
629 		    new->is_stats.writes);
630 		wps = (double)udeltas;
631 		wps /= etime;
632 
633 		tps = rps + wps;
634 			/* transactions per second */
635 
636 		/*
637 		 * report throughput as either kb/sec or MB/sec
638 		 */
639 
640 		if (!do_megabytes)
641 			iosize = 1024.0;
642 		else
643 			iosize = 1048576.0;
644 
645 		ldeltas = ull_delta(old ? old->is_stats.nread : 0,
646 		    new->is_stats.nread);
647 		if (ldeltas) {
648 			krps = (double)ldeltas;
649 			krps /= etime;
650 			krps /= iosize;
651 		} else
652 			krps = 0.0;
653 
654 		ldeltas = ull_delta(old ? old->is_stats.nwritten : 0,
655 		    new->is_stats.nwritten);
656 		if (ldeltas) {
657 			kwps = (double)ldeltas;
658 			kwps /= etime;
659 			kwps /= iosize;
660 		} else
661 			kwps = 0.0;
662 
663 		/*
664 		 * Blocks transferred per second
665 		 */
666 		kps = krps + kwps;
667 
668 		/*
669 		 * Average number of wait transactions waiting
670 		 */
671 		w_delta = hrtime_delta((u_longlong_t)
672 		    (old ? old->is_stats.wlentime : 0),
673 		    new->is_stats.wlentime);
674 		if (w_delta) {
675 			avw = (double)w_delta;
676 			avw /= hr_etime;
677 		} else
678 			avw = 0.0;
679 
680 		/*
681 		 * Average number of run transactions waiting
682 		 */
683 		r_delta = hrtime_delta(old ? old->is_stats.rlentime : 0,
684 		    new->is_stats.rlentime);
685 		if (r_delta) {
686 			avr = (double)r_delta;
687 			avr /= hr_etime;
688 		} else
689 			avr = 0.0;
690 
691 		/*
692 		 * Average wait service time in milliseconds
693 		 */
694 		if (tps > 0.0 && (avw != 0.0 || avr != 0.0)) {
695 			mtps = 1000.0 / tps;
696 			if (avw != 0.0)
697 				wserv = avw * mtps;
698 			else
699 				wserv = 0.0;
700 
701 			if (avr != 0.0)
702 				rserv = avr * mtps;
703 			else
704 				rserv = 0.0;
705 			serv = rserv + wserv;
706 		} else {
707 			rserv = 0.0;
708 			wserv = 0.0;
709 			serv = 0.0;
710 		}
711 
712 		/* % of time there is a transaction waiting for service */
713 		t_delta = hrtime_delta(old ? old->is_stats.wtime : 0,
714 		    new->is_stats.wtime);
715 		if (t_delta) {
716 			w_pct = (double)t_delta;
717 			w_pct /= hr_etime;
718 			w_pct *= 100.0;
719 
720 			/*
721 			 * Average the wait queue utilization over the
722 			 * the controller's devices, if this is a controller.
723 			 */
724 			if (new->is_type == IODEV_CONTROLLER)
725 				w_pct /= new->is_nr_children;
726 		} else
727 			w_pct = 0.0;
728 
729 		/* % of time there is a transaction running */
730 		t_delta = hrtime_delta(old ? old->is_stats.rtime : 0,
731 		    new->is_stats.rtime);
732 		if (t_delta) {
733 			r_pct = (double)t_delta;
734 			r_pct /= hr_etime;
735 			r_pct *= 100.0;
736 
737 			/*
738 			 * Average the percent busy over the controller's
739 			 * devices, if this is a controller.
740 			 */
741 			if (new->is_type == IODEV_CONTROLLER)
742 				w_pct /= new->is_nr_children;
743 		} else {
744 			r_pct = 0.0;
745 		}
746 
747 		/* % of time there is a transaction running */
748 		if (do_interval) {
749 			rps	*= etime;
750 			wps	*= etime;
751 			tps	*= etime;
752 			krps	*= etime;
753 			kwps	*= etime;
754 			kps	*= etime;
755 		}
756 	}
757 
758 	if (do_disk & (DISK_EXTENDED | DISK_ERRORS)) {
759 		if ((!do_conversions) && ((suppress_zero == 0) ||
760 		    ((do_disk & DISK_EXTENDED) == 0))) {
761 			if (do_raw == 0) {
762 				push_out("%-*.*s",
763 				    iodevs_nl, iodevs_nl, disk_name);
764 			} else {
765 				push_out(disk_name);
766 			}
767 		}
768 	}
769 
770 	switch (do_disk & DISK_IO_MASK) {
771 	    case DISK_OLD:
772 		if (do_raw == 0)
773 			fstr = "%3.0f %3.0f %4.0f  ";
774 		else
775 			fstr = "%.0f,%.0f,%.0f";
776 		push_out(fstr, kps, tps, serv);
777 		break;
778 	    case DISK_NEW:
779 		if (do_raw == 0)
780 			fstr = "%3.0f %3.0f %4.1f  ";
781 		else
782 			fstr = "%.0f,%.0f,%.1f";
783 		push_out(fstr, rps, wps, r_pct);
784 		break;
785 	    case DISK_EXTENDED:
786 		if (suppress_zero) {
787 			if (fzero(rps) && fzero(wps) && fzero(krps) &&
788 			    fzero(kwps) && fzero(avw) && fzero(avr) &&
789 			    fzero(serv) && fzero(w_pct) && fzero(r_pct)) {
790 				doit = 0;
791 			} else if (do_conversions == 0) {
792 				if (do_raw == 0) {
793 					push_out("%-*.*s",
794 					    iodevs_nl, iodevs_nl, disk_name);
795 				} else {
796 					push_out(disk_name);
797 				}
798 			}
799 		}
800 		if (doit) {
801 			if (!do_conversions) {
802 				if (do_raw == 0) {
803 					fstr = " %6.1f %6.1f %6.1f %6.1f "
804 						"%4.1f %4.1f %6.1f %3.0f "
805 						"%3.0f ";
806 				} else {
807 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
808 						"%.1f,%.0f,%.0f";
809 				}
810 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
811 				    serv, w_pct, r_pct);
812 			} else {
813 				if (do_raw == 0) {
814 					fstr = " %6.1f %6.1f %6.1f %6.1f "
815 						"%4.1f %4.1f %6.1f %6.1f "
816 						"%3.0f %3.0f ";
817 				} else {
818 					fstr = "%.1f,%.1f,%.1f,%.1f,%.1f,%.1f,"
819 						"%.1f,%.1f,%.0f,%.0f";
820 				}
821 				push_out(fstr, rps, wps, krps, kwps, avw, avr,
822 				    wserv, rserv, w_pct, r_pct);
823 			}
824 		}
825 		break;
826 	}
827 
828 	if (do_disk & DISK_ERRORS) {
829 		if ((do_disk == DISK_ERRORS)) {
830 			if (do_raw == 0)
831 				push_out(two_blanks);
832 		}
833 
834 		if (new->is_errors.ks_data) {
835 			kstat_named_t *knp;
836 			char *efstr;
837 
838 			if (do_raw == 0)
839 				efstr = "%3u ";
840 			else
841 				efstr = "%u";
842 			toterrs = 0;
843 			knp = KSTAT_NAMED_PTR(&new->is_errors);
844 			for (i = 0; i < 3; i++) {
845 				switch (knp[i].data_type) {
846 					case KSTAT_DATA_ULONG:
847 						push_out(efstr,
848 						    knp[i].value.ui32);
849 						toterrs += knp[i].value.ui32;
850 						break;
851 					case KSTAT_DATA_ULONGLONG:
852 						/*
853 						 * We're only set up to
854 						 * write out the low
855 						 * order 32-bits so
856 						 * just grab that.
857 						 */
858 						push_out(efstr,
859 						    knp[i].value.ui32);
860 						toterrs += knp[i].value.ui32;
861 						break;
862 					default:
863 						break;
864 				}
865 			}
866 			push_out(efstr, toterrs);
867 		} else {
868 			if (do_raw == 0)
869 				push_out("  0   0   0   0 ");
870 			else
871 				push_out("0,0,0,0");
872 		}
873 
874 	}
875 
876 	if (suppress_zero == 0 || doit == 1) {
877 		if ((do_disk & (DISK_EXTENDED | DISK_ERRORS)) &&
878 			do_conversions) {
879 			push_out("%s", disk_name);
880 			if (show_mountpts && new->is_dname) {
881 				mnt_t *mount_pt;
882 				char *lu;
883 				char lub[SMALL_SCRATCH_BUFLEN];
884 
885 				lu = strrchr(new->is_dname, '/');
886 				if (lu) {
887 					if (strcmp(disk_name, lu) == 0)
888 						lu = new->is_dname;
889 					else {
890 						*lu = 0;
891 						(void) strcpy(lub,
892 						    new->is_dname);
893 						*lu = '/';
894 						(void) strcat(lub, "/");
895 						(void) strcat(lub,
896 						    disk_name);
897 						lu = lub;
898 					}
899 				} else
900 					lu = disk_name;
901 				mount_pt = lookup_mntent_byname(lu);
902 				if (mount_pt) {
903 					if (do_raw == 0)
904 						push_out(" (%s)",
905 						    mount_pt->mount_point);
906 					else
907 						push_out("(%s)",
908 						    mount_pt->mount_point);
909 				}
910 			}
911 		}
912 	}
913 
914 	if ((do_disk & PRINT_VERTICAL) && show_disk_mode != SHOW_FIRST_ONLY)
915 		do_newline();
916 
917 	if (count != NULL)
918 		(*count)++;
919 }
920 
921 static void
922 usage(void)
923 {
924 	(void) fprintf(stderr,
925 	    "Usage: iostat [-cCdDeEiImMnpPrstxXYz] "
926 	    " [-l n] [-T d|u] [disk ...] [interval [count]]\n"
927 	    "\t\t-c: 	report percentage of time system has spent\n"
928 	    "\t\t\tin user/system/wait/idle mode\n"
929 	    "\t\t-C: 	report disk statistics by controller\n"
930 	    "\t\t-d: 	display disk Kb/sec, transfers/sec, avg. \n"
931 	    "\t\t\tservice time in milliseconds  \n"
932 	    "\t\t-D: 	display disk reads/sec, writes/sec, \n"
933 	    "\t\t\tpercentage disk utilization \n"
934 	    "\t\t-e: 	report device error summary statistics\n"
935 	    "\t\t-E: 	report extended device error statistics\n"
936 	    "\t\t-i:	show device IDs for -E output\n"
937 	    "\t\t-I: 	report the counts in each interval,\n"
938 	    "\t\t\tinstead of rates, where applicable\n"
939 	    "\t\t-l n:	Limit the number of disks to n\n"
940 	    "\t\t-m: 	Display mount points (most useful with -p)\n"
941 	    "\t\t-M: 	Display data throughput in MB/sec "
942 	    "instead of Kb/sec\n"
943 	    "\t\t-n: 	convert device names to cXdYtZ format\n"
944 	    "\t\t-p: 	report per-partition disk statistics\n"
945 	    "\t\t-P: 	report per-partition disk statistics only,\n"
946 	    "\t\t\tno per-device disk statistics\n"
947 	    "\t\t-r: 	Display data in comma separated format\n"
948 	    "\t\t-s: 	Suppress state change messages\n"
949 	    "\t\t-T d|u	Display a timestamp in date (d) or unix "
950 	    "time_t (u)\n"
951 	    "\t\t-t: 	display chars read/written to terminals\n"
952 	    "\t\t-x: 	display extended disk statistics\n"
953 	    "\t\t-X: 	display I/O path statistics\n"
954 	    "\t\t-Y: 	display I/O path (I/T/L) statistics\n"
955 	    "\t\t-z: 	Suppress entries with all zero values\n");
956 	exit(1);
957 }
958 
959 /*ARGSUSED*/
960 static void
961 show_disk_errors(void *v1, void *v2, void *d)
962 {
963 	struct iodev_snapshot *disk = (struct iodev_snapshot *)v2;
964 	kstat_named_t *knp;
965 	size_t  col;
966 	int	i, len;
967 	char	*dev_name;
968 
969 	if (disk->is_errors.ks_ndata == 0)
970 		return;
971 	if (disk->is_type == IODEV_CONTROLLER)
972 		return;
973 
974 	dev_name = do_conversions ? disk->is_pretty : disk->is_name;
975 	dev_name = dev_name ? dev_name : disk->is_name;
976 
977 	len = strlen(dev_name);
978 	if (len > 20)
979 		push_out("%s ", dev_name);
980 	else if (len > 16)
981 		push_out("%-20.20s ", dev_name);
982 	else {
983 		if (do_conversions)
984 			push_out("%-16.16s ", dev_name);
985 		else
986 			push_out("%-9.9s ", dev_name);
987 	}
988 	col = 0;
989 
990 	knp = KSTAT_NAMED_PTR(&disk->is_errors);
991 	for (i = 0; i < disk->is_errors.ks_ndata; i++) {
992 		/* skip kstats that the driver did not kstat_named_init */
993 		if (knp[i].name[0] == 0)
994 			continue;
995 
996 		col += strlen(knp[i].name);
997 
998 		switch (knp[i].data_type) {
999 			case KSTAT_DATA_CHAR:
1000 				if ((strcmp(knp[i].name, "Serial No") == 0) &&
1001 				    do_devid) {
1002 					if (disk->is_devid) {
1003 						push_out("Device Id: %s ",
1004 						    disk->is_devid);
1005 						col += strlen(disk->is_devid);
1006 					} else
1007 						push_out("Device Id: ");
1008 				} else {
1009 					push_out("%s: %-.16s ", knp[i].name,
1010 					    &knp[i].value.c[0]);
1011 					col += strlen(&knp[i].value.c[0]);
1012 				}
1013 				break;
1014 			case KSTAT_DATA_ULONG:
1015 				push_out("%s: %u ", knp[i].name,
1016 				    knp[i].value.ui32);
1017 				col += 4;
1018 				break;
1019 			case KSTAT_DATA_ULONGLONG:
1020 				if (strcmp(knp[i].name, "Size") == 0) {
1021 					push_out("%s: %2.2fGB <%llu bytes>\n",
1022 					    knp[i].name,
1023 					    (float)knp[i].value.ui64 /
1024 					    DISK_GIGABYTE,
1025 					    knp[i].value.ui64);
1026 					col = 0;
1027 					break;
1028 				}
1029 				push_out("%s: %u ", knp[i].name,
1030 				    knp[i].value.ui32);
1031 				col += 4;
1032 				break;
1033 			}
1034 		if ((col >= 62) || (i == 2)) {
1035 			do_newline();
1036 			col = 0;
1037 		}
1038 	}
1039 	if (col > 0) {
1040 		do_newline();
1041 	}
1042 	do_newline();
1043 }
1044 
1045 void
1046 do_args(int argc, char **argv)
1047 {
1048 	int 		c;
1049 	int 		errflg = 0;
1050 	extern char 	*optarg;
1051 	extern int 	optind;
1052 
1053 	while ((c = getopt(argc, argv, "tdDxXYCciIpPnmMeEszrT:l:")) != EOF)
1054 		switch (c) {
1055 		case 't':
1056 			do_tty++;
1057 			break;
1058 		case 'd':
1059 			do_disk |= DISK_OLD;
1060 			break;
1061 		case 'D':
1062 			do_disk |= DISK_NEW;
1063 			break;
1064 		case 'x':
1065 			do_disk |= DISK_EXTENDED;
1066 			break;
1067 		case 'X':
1068 			if (do_disk & DISK_IOPATH_LTI)
1069 				errflg++;	/* -Y already used */
1070 			else
1071 				do_disk |= DISK_IOPATH_LI;
1072 			break;
1073 		case 'Y':
1074 			if (do_disk & DISK_IOPATH_LI)
1075 				errflg++;	/* -X already used */
1076 			else
1077 				do_disk |= DISK_IOPATH_LTI;
1078 			break;
1079 		case 'C':
1080 			do_controller++;
1081 			break;
1082 		case 'c':
1083 			do_cpu++;
1084 			break;
1085 		case 'I':
1086 			do_interval++;
1087 			break;
1088 		case 'p':
1089 			do_partitions++;
1090 			break;
1091 		case 'P':
1092 			do_partitions_only++;
1093 			break;
1094 		case 'n':
1095 			do_conversions++;
1096 			break;
1097 		case 'M':
1098 			do_megabytes++;
1099 			break;
1100 		case 'e':
1101 			do_disk |= DISK_ERRORS;
1102 			break;
1103 		case 'E':
1104 			do_disk |= DISK_EXTENDED_ERRORS;
1105 			break;
1106 		case 'i':
1107 			do_devid = 1;
1108 			break;
1109 		case 's':
1110 			suppress_state = 1;
1111 			break;
1112 		case 'z':
1113 			suppress_zero = 1;
1114 			break;
1115 		case 'm':
1116 			show_mountpts = 1;
1117 			break;
1118 		case 'T':
1119 			if (optarg) {
1120 				if (*optarg == 'u')
1121 					do_timestamp = UDATE;
1122 				else if (*optarg == 'd')
1123 					do_timestamp = CDATE;
1124 				else
1125 					errflg++;
1126 			} else
1127 				errflg++;
1128 			break;
1129 		case 'r':
1130 			do_raw = 1;
1131 			break;
1132 		case 'l':
1133 			df.if_max_iodevs = safe_strtoi(optarg, "invalid limit");
1134 			if (df.if_max_iodevs < 1)
1135 				usage();
1136 			break;
1137 		case '?':
1138 			errflg++;
1139 	}
1140 
1141 	if ((do_disk & DISK_OLD) && (do_disk & DISK_NEW)) {
1142 		(void) fprintf(stderr, "-d and -D are incompatible.\n");
1143 		usage();
1144 	}
1145 
1146 	if (errflg) {
1147 		usage();
1148 	}
1149 
1150 	/* if no output classes explicity specified, use defaults */
1151 	if (do_tty == 0 && do_disk == 0 && do_cpu == 0)
1152 		do_tty = do_cpu = 1, do_disk = DISK_OLD;
1153 
1154 	/*
1155 	 * multi-path options (-X, -Y) without a specific vertical
1156 	 * output format (-x, -e, -E) imply extended -x format
1157 	 */
1158 	if ((do_disk & (DISK_IOPATH_LI | DISK_IOPATH_LTI)) &&
1159 	    !(do_disk & PRINT_VERTICAL))
1160 		do_disk |= DISK_EXTENDED;
1161 
1162 	/*
1163 	 * If conflicting options take the preferred
1164 	 * -D and -x result in -x
1165 	 * -d or -D and -e or -E gives only whatever -d or -D was specified
1166 	 */
1167 	if ((do_disk & DISK_EXTENDED) && (do_disk & DISK_NORMAL))
1168 		do_disk &= ~DISK_NORMAL;
1169 	if ((do_disk & DISK_NORMAL) && (do_disk & DISK_ERROR_MASK))
1170 		do_disk &= ~DISK_ERROR_MASK;
1171 
1172 	/* nfs, tape, always shown */
1173 	df.if_allowed_types = IODEV_NFS | IODEV_TAPE;
1174 
1175 	/*
1176 	 * If limit == 0 then no command line limit was set, else if any of
1177 	 * the flags that cause unlimited disks were not set,
1178 	 * use the default of 4
1179 	 */
1180 	if (df.if_max_iodevs == 0) {
1181 		df.if_max_iodevs = DEFAULT_LIMIT;
1182 		df.if_skip_floppy = 1;
1183 		if (do_disk & (DISK_EXTENDED | DISK_ERRORS |
1184 		    DISK_EXTENDED_ERRORS)) {
1185 			df.if_max_iodevs = UNLIMITED_IODEVS;
1186 			df.if_skip_floppy = 0;
1187 		}
1188 	}
1189 	if (do_disk) {
1190 		size_t count = 0;
1191 		size_t i = optind;
1192 
1193 		while (i < argc && !isdigit(argv[i][0])) {
1194 			count++;
1195 			i++;
1196 		}
1197 
1198 		/*
1199 		 * "Note:  disks  explicitly  requested
1200 		 * are not subject to this disk limit"
1201 		 */
1202 		if ((count > df.if_max_iodevs) ||
1203 		    (count && (df.if_max_iodevs == UNLIMITED_IODEVS)))
1204 			df.if_max_iodevs = count;
1205 
1206 		df.if_names = safe_alloc(count * sizeof (char *));
1207 		(void) memset(df.if_names, 0, count * sizeof (char *));
1208 
1209 		df.if_nr_names = 0;
1210 		while (optind < argc && !isdigit(argv[optind][0]))
1211 			df.if_names[df.if_nr_names++] = argv[optind++];
1212 	}
1213 	if (optind < argc) {
1214 		interval = safe_strtoi(argv[optind], "invalid interval");
1215 		if (interval < 1)
1216 			fail(0, "invalid interval");
1217 		optind++;
1218 
1219 		if (optind < argc) {
1220 			iter = safe_strtoi(argv[optind], "invalid count");
1221 			if (iter < 1)
1222 				fail(0, "invalid count");
1223 			optind++;
1224 		}
1225 	}
1226 	if (interval == 0)
1227 		iter = 1;
1228 	if (optind < argc)
1229 		usage();
1230 }
1231 
1232 /*
1233  * Driver for doing the extended header formatting. Will produce
1234  * the function stack needed to output an extended header based
1235  * on the options selected.
1236  */
1237 
1238 void
1239 do_format(void)
1240 {
1241 	char	header[SMALL_SCRATCH_BUFLEN];
1242 	char 	ch;
1243 	char 	iosz;
1244 	const char    *fstr;
1245 
1246 	disk_header[0] = 0;
1247 	ch = (do_interval ? 'i' : 's');
1248 	iosz = (do_megabytes ? 'M' : 'k');
1249 	if (do_disk & DISK_ERRORS) {
1250 		if (do_raw == 0) {
1251 			(void) sprintf(header, "s/w h/w trn tot ");
1252 		} else
1253 			(void) sprintf(header, "s/w,h/w,trn,tot");
1254 	} else
1255 		*header = NULL;
1256 	switch (do_disk & DISK_IO_MASK) {
1257 		case DISK_OLD:
1258 			if (do_raw == 0)
1259 				fstr = "%cp%c tp%c serv  ";
1260 			else
1261 				fstr = "%cp%c,tp%c,serv";
1262 			(void) snprintf(disk_header, sizeof (disk_header),
1263 			    fstr, iosz, ch, ch);
1264 			break;
1265 		case DISK_NEW:
1266 			if (do_raw == 0)
1267 				fstr = "rp%c wp%c util  ";
1268 			else
1269 				fstr = "%rp%c,wp%c,util";
1270 			(void) snprintf(disk_header, sizeof (disk_header),
1271 			    fstr, ch, ch);
1272 			break;
1273 		case DISK_EXTENDED:
1274 			/* This is -x option */
1275 			if (!do_conversions) {
1276 				/* without -n option */
1277 				if (do_raw == 0) {
1278 					/* without -r option */
1279 					(void) snprintf(disk_header,
1280 					    sizeof (disk_header),
1281 					    "%-*.*s    r/%c    w/%c   "
1282 					    "%cr/%c   %cw/%c wait actv  "
1283 					    "svc_t  %%%%w  %%%%b %s",
1284 					    iodevs_nl, iodevs_nl, "device",
1285 					    ch, ch, iosz, ch, iosz, ch, header);
1286 				} else {
1287 					/* with -r option */
1288 					(void) snprintf(disk_header,
1289 					    sizeof (disk_header),
1290 					    "device,r/%c,w/%c,%cr/%c,%cw/%c,"
1291 					    "wait,actv,svc_t,%%%%w,"
1292 					    "%%%%b,%s",
1293 					    ch, ch, iosz, ch, iosz, ch, header);
1294 				}
1295 			} else {
1296 				/* with -n option */
1297 				if (do_raw == 0) {
1298 					fstr = "    r/%c    w/%c   %cr/%c   "
1299 					    "%cw/%c wait actv wsvc_t asvc_t  "
1300 					    "%%%%w  %%%%b %sdevice";
1301 				} else {
1302 					fstr = "r/%c,w/%c,%cr/%c,%cw/%c,"
1303 					    "wait,actv,wsvc_t,asvc_t,"
1304 					    "%%%%w,%%%%b,%sdevice";
1305 				}
1306 				(void) snprintf(disk_header,
1307 				    sizeof (disk_header),
1308 				    fstr, ch, ch, iosz, ch, iosz,
1309 				    ch, header);
1310 			}
1311 			break;
1312 		default:
1313 			break;
1314 	}
1315 
1316 	/* do DISK_ERRORS header (already added above for DISK_EXTENDED) */
1317 	if ((do_disk & DISK_ERRORS) &&
1318 	    ((do_disk & DISK_IO_MASK) != DISK_EXTENDED)) {
1319 		if (!do_conversions) {
1320 			if (do_raw == 0)
1321 				(void) snprintf(disk_header,
1322 				    sizeof (disk_header), "%-*.*s  %s",
1323 				    iodevs_nl, iodevs_nl, "device", header);
1324 			else
1325 				(void) snprintf(disk_header,
1326 				    sizeof (disk_header), "device,%s", header);
1327 		} else {
1328 			if (do_raw == 0) {
1329 				(void) snprintf(disk_header,
1330 				    sizeof (disk_header),
1331 				    "  %sdevice", header);
1332 			} else {
1333 				(void) snprintf(disk_header,
1334 				    sizeof (disk_header),
1335 				    "%s,device", header);
1336 			}
1337 		}
1338 	} else {
1339 		/*
1340 		 * Need to subtract two characters for the % escape in
1341 		 * the string.
1342 		 */
1343 		dh_len = strlen(disk_header) - 2;
1344 	}
1345 
1346 	if (do_timestamp)
1347 		setup(print_timestamp);
1348 
1349 	/*
1350 	 * -n *and* (-E *or* -e *or* -x)
1351 	 */
1352 	if (do_conversions && (do_disk & PRINT_VERTICAL)) {
1353 		if (do_tty)
1354 			setup(print_tty_hdr1);
1355 		if (do_cpu)
1356 			setup(print_cpu_hdr1);
1357 		if (do_tty || do_cpu)
1358 			setup(do_newline);
1359 		if (do_tty)
1360 			setup(print_tty_hdr2);
1361 		if (do_cpu)
1362 			setup(print_cpu_hdr2);
1363 		if (do_tty || do_cpu)
1364 			setup(do_newline);
1365 		if (do_tty)
1366 			setup(print_tty_data);
1367 		if (do_cpu)
1368 			setup(print_cpu_data);
1369 		if (do_tty || do_cpu)
1370 			setup(do_newline);
1371 		printxhdr();
1372 
1373 		setup(show_all_disks);
1374 	} else {
1375 		/*
1376 		 * These unholy gymnastics are necessary to place CPU/tty
1377 		 * data to the right of the disks/errors for the first
1378 		 * line in vertical mode.
1379 		 */
1380 		if (do_disk & PRINT_VERTICAL) {
1381 			printxhdr();
1382 
1383 			setup(show_first_disk);
1384 			if (do_tty)
1385 				setup(print_tty_data);
1386 			if (do_cpu)
1387 				setup(print_cpu_data);
1388 			setup(do_newline);
1389 
1390 			setup(show_other_disks);
1391 		} else {
1392 			setup(hdrout);
1393 			if (do_tty)
1394 				setup(print_tty_data);
1395 			setup(show_all_disks);
1396 			if (do_cpu)
1397 				setup(print_cpu_data);
1398 		}
1399 
1400 		setup(do_newline);
1401 	}
1402 	if (do_disk & DISK_EXTENDED_ERRORS)
1403 		setup(disk_errors);
1404 }
1405 
1406 /*
1407  * Add a new function to the list of functions
1408  * for this invocation. Once on the stack the
1409  * function is never removed nor does its place
1410  * change.
1411  */
1412 void
1413 setup(void (*nfunc)(void))
1414 {
1415 	format_t *tmp;
1416 
1417 	tmp = safe_alloc(sizeof (format_t));
1418 	tmp->nfunc = nfunc;
1419 	tmp->next = 0;
1420 	if (formatter_end)
1421 		formatter_end->next = tmp;
1422 	else
1423 		formatter_list = tmp;
1424 	formatter_end = tmp;
1425 
1426 }
1427 
1428 /*
1429  * The functions after this comment are devoted to printing
1430  * various parts of the header. They are selected based on the
1431  * options provided when the program was invoked. The functions
1432  * are either directly invoked in printhdr() or are indirectly
1433  * invoked by being placed on the list of functions used when
1434  * extended headers are used.
1435  */
1436 void
1437 print_tty_hdr1(void)
1438 {
1439 	char *fstr;
1440 	char *dstr;
1441 
1442 	if (do_raw == 0) {
1443 		fstr = "%10.10s";
1444 		dstr = "tty    ";
1445 	} else {
1446 		fstr = "%s";
1447 		dstr = "tty";
1448 	}
1449 	push_out(fstr, dstr);
1450 }
1451 
1452 void
1453 print_tty_hdr2(void)
1454 {
1455 	if (do_raw == 0)
1456 		push_out("%-10.10s", " tin tout");
1457 	else
1458 		push_out("tin,tout");
1459 }
1460 
1461 void
1462 print_cpu_hdr1(void)
1463 {
1464 	char *dstr;
1465 
1466 	if (do_raw == 0)
1467 		dstr = "     cpu";
1468 	else
1469 		dstr = "cpu";
1470 	push_out(dstr);
1471 }
1472 
1473 void
1474 print_cpu_hdr2(void)
1475 {
1476 	char *dstr;
1477 
1478 	if (do_raw == 0)
1479 		dstr = " us sy wt id";
1480 	else
1481 		dstr = "us,sy,wt,id";
1482 	push_out(dstr);
1483 }
1484 
1485 /*
1486  * Assumption is that tty data is always first - no need for raw mode leading
1487  * comma.
1488  */
1489 void
1490 print_tty_data(void)
1491 {
1492 	char *fstr;
1493 	uint64_t deltas;
1494 	double raw;
1495 	double outch;
1496 	kstat_t *oldks = NULL;
1497 
1498 	if (oldss)
1499 		oldks = &oldss->s_sys.ss_agg_sys;
1500 
1501 	if (do_raw == 0)
1502 		fstr = " %3.0f %4.0f ";
1503 	else
1504 		fstr = "%.0f,%.0f";
1505 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "rawch");
1506 	raw = deltas;
1507 	raw /= getime;
1508 	deltas = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "outch");
1509 	outch = deltas;
1510 	outch /= getime;
1511 	push_out(fstr, raw, outch);
1512 }
1513 
1514 /*
1515  * Write out CPU data
1516  */
1517 void
1518 print_cpu_data(void)
1519 {
1520 	char *fstr;
1521 	uint64_t idle;
1522 	uint64_t user;
1523 	uint64_t kern;
1524 	uint64_t wait;
1525 	kstat_t *oldks = NULL;
1526 
1527 	if (oldss)
1528 		oldks = &oldss->s_sys.ss_agg_sys;
1529 
1530 	if (do_raw == 0)
1531 		fstr = " %2.0f %2.0f %2.0f %2.0f";
1532 	else
1533 		fstr = "%.0f,%.0f,%.0f,%.0f";
1534 
1535 	idle = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_idle");
1536 	user = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_user");
1537 	kern = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_kernel");
1538 	wait = kstat_delta(oldks, &newss->s_sys.ss_agg_sys, "cpu_ticks_wait");
1539 	push_out(fstr, user * percent, kern * percent,
1540 		wait * percent, idle * percent);
1541 }
1542 
1543 /*
1544  * Emit the appropriate header.
1545  */
1546 void
1547 hdrout(void)
1548 {
1549 	if (do_raw == 0) {
1550 		if (--tohdr == 0)
1551 			printhdr(0);
1552 	} else if (hdr_out == 0) {
1553 		printhdr(0);
1554 		hdr_out = 1;
1555 	}
1556 }
1557 
1558 /*
1559  * Write out disk errors when -E is specified.
1560  */
1561 void
1562 disk_errors(void)
1563 {
1564 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk_errors, NULL);
1565 }
1566 
1567 void
1568 show_first_disk(void)
1569 {
1570 	int count = 0;
1571 
1572 	show_disk_mode = SHOW_FIRST_ONLY;
1573 
1574 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1575 }
1576 
1577 void
1578 show_other_disks(void)
1579 {
1580 	int count = 0;
1581 
1582 	show_disk_mode = SHOW_SECOND_ONWARDS;
1583 
1584 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1585 }
1586 
1587 void
1588 show_all_disks(void)
1589 {
1590 	int count = 0;
1591 
1592 	show_disk_mode = SHOW_ALL;
1593 
1594 	(void) snapshot_walk(SNAP_IODEVS, oldss, newss, show_disk, &count);
1595 }
1596 
1597 /*
1598  * Write a newline out and clear the lineout flag.
1599  */
1600 static void
1601 do_newline(void)
1602 {
1603 	if (lineout) {
1604 		(void) putchar('\n');
1605 		lineout = 0;
1606 	}
1607 }
1608 
1609 /*
1610  * Generalized printf function that determines what extra
1611  * to print out if we're in raw mode. At this time we
1612  * don't care about errors.
1613  */
1614 static void
1615 push_out(const char *message, ...)
1616 {
1617 	va_list args;
1618 
1619 	va_start(args, message);
1620 	if (do_raw && lineout == 1)
1621 		(void) putchar(',');
1622 	(void) vprintf(message, args);
1623 	va_end(args);
1624 	lineout = 1;
1625 }
1626 
1627 /*
1628  * Emit the header string when -e is specified.
1629  */
1630 static void
1631 print_err_hdr(void)
1632 {
1633 	char obuf[SMALL_SCRATCH_BUFLEN];
1634 
1635 	if (do_raw) {
1636 		push_out("errors");
1637 		return;
1638 	}
1639 
1640 	if (do_conversions == 0) {
1641 		if (!(do_disk & DISK_EXTENDED)) {
1642 			(void) snprintf(obuf, sizeof (obuf),
1643 			    "%11s", one_blank);
1644 			push_out(obuf);
1645 		}
1646 	} else if (do_disk == DISK_ERRORS)
1647 		push_out(two_blanks);
1648 	else
1649 		push_out(one_blank);
1650 	push_out("---- errors --- ");
1651 }
1652 
1653 /*
1654  * Emit the header string when -e is specified.
1655  */
1656 static void
1657 print_disk_header(void)
1658 {
1659 	push_out(disk_header);
1660 }
1661 
1662 /*
1663  * Write out a timestamp. Format is all that goes out on
1664  * the line so no use of push_out.
1665  *
1666  * Write out as decimal reprentation of time_t value
1667  * (-T u was specified) or the string returned from
1668  * ctime() (-T d was specified).
1669  */
1670 static void
1671 print_timestamp(void)
1672 {
1673 	time_t t;
1674 
1675 	if (time(&t) != -1) {
1676 		if (do_timestamp == UDATE) {
1677 			(void) printf("%ld\n", t);
1678 		} else if (do_timestamp == CDATE) {
1679 			char *cpt;
1680 
1681 			cpt = ctime(&t);
1682 			if (cpt) {
1683 				(void) fputs(cpt, stdout);
1684 			}
1685 		}
1686 	}
1687 }
1688 
1689 /*
1690  * No, UINTMAX_MAX isn't the right thing here since
1691  * it is #defined to be either INT32_MAX or INT64_MAX
1692  * depending on the whether _LP64 is defined.
1693  *
1694  * We want to handle the odd future case of having
1695  * ulonglong_t be more than 64 bits but we have
1696  * no nice #define MAX value we can drop in place
1697  * without having to change this code in the future.
1698  */
1699 
1700 u_longlong_t
1701 ull_delta(u_longlong_t old, u_longlong_t new)
1702 {
1703 	if (new >= old)
1704 		return (new - old);
1705 	else
1706 		return ((UINT64_MAX - old) + new + 1);
1707 }
1708 
1709 /*
1710  * Take the difference of an unsigned 32
1711  * bit int attempting to cater for
1712  * overflow.
1713  */
1714 uint_t
1715 u32_delta(uint_t old, uint_t new)
1716 {
1717 	if (new >= old)
1718 		return (new - old);
1719 	else
1720 		return ((UINT32_MAX - old) + new + 1);
1721 }
1722 
1723 /*
1724  * Create and arm the timer. Used only when an interval has been specified.
1725  * Used in lieu of poll to ensure that we provide info for exactly the
1726  * desired period.
1727  */
1728 void
1729 set_timer(int interval)
1730 {
1731 	timer_t t_id;
1732 	itimerspec_t time_struct;
1733 	struct sigevent sig_struct;
1734 	struct sigaction act;
1735 
1736 	bzero(&sig_struct, sizeof (struct sigevent));
1737 	bzero(&act, sizeof (struct sigaction));
1738 
1739 	/* Create timer */
1740 	sig_struct.sigev_notify = SIGEV_SIGNAL;
1741 	sig_struct.sigev_signo = SIGUSR1;
1742 	sig_struct.sigev_value.sival_int = 0;
1743 
1744 	if (timer_create(CLOCK_REALTIME, &sig_struct, &t_id) != 0) {
1745 		fail(1, "Timer creation failed");
1746 	}
1747 
1748 	act.sa_handler = handle_sig;
1749 
1750 	if (sigaction(SIGUSR1, &act, NULL) != 0) {
1751 		fail(1, "Could not set up signal handler");
1752 	}
1753 
1754 	time_struct.it_value.tv_sec = interval;
1755 	time_struct.it_value.tv_nsec = 0;
1756 	time_struct.it_interval.tv_sec = interval;
1757 	time_struct.it_interval.tv_nsec = 0;
1758 
1759 	/* Arm timer */
1760 	if ((timer_settime(t_id, 0, &time_struct, NULL)) != 0) {
1761 		fail(1, "Setting timer failed");
1762 	}
1763 }
1764 /* ARGSUSED */
1765 void
1766 handle_sig(int x)
1767 {
1768 }
1769 
1770 /*
1771  * This is exactly what is needed for standard iostat output,
1772  * but make sure to use it only for that
1773  */
1774 #define	EPSILON	(0.1)
1775 static int
1776 fzero(double value)
1777 {
1778 	return (value >= 0.0 && value < EPSILON);
1779 }
1780 
1781 static int
1782 safe_strtoi(char const *val, char *errmsg)
1783 {
1784 	char *end;
1785 	long tmp;
1786 
1787 	errno = 0;
1788 	tmp = strtol(val, &end, 10);
1789 	if (*end != '\0' || errno)
1790 		fail(0, "%s %s", errmsg, val);
1791 	return ((int)tmp);
1792 }
1793