xref: /netbsd-src/sys/kern/subr_iostat.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*	$NetBSD: subr_iostat.c,v 1.24 2018/09/03 16:29:35 riastradh Exp $	*/
2 /*	NetBSD: subr_disk.c,v 1.69 2005/05/29 22:24:15 christos Exp	*/
3 
4 /*-
5  * Copyright (c) 1996, 1997, 1999, 2000, 2009 The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
10  * NASA Ames Research Center.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  * (c) UNIX System Laboratories, Inc.
38  * All or some portions of this file are derived from material licensed
39  * to the University of California by American Telephone and Telegraph
40  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41  * the permission of UNIX System Laboratories, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. Neither the name of the University nor the names of its contributors
52  *    may be used to endorse or promote products derived from this software
53  *    without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: subr_iostat.c,v 1.24 2018/09/03 16:29:35 riastradh Exp $");
72 
73 #include <sys/param.h>
74 #include <sys/kernel.h>
75 #include <sys/kmem.h>
76 #include <sys/iostat.h>
77 #include <sys/sysctl.h>
78 #include <sys/rwlock.h>
79 
80 /*
81  * Function prototypes for sysctl nodes
82  */
83 static int	sysctl_hw_disknames(SYSCTLFN_PROTO);
84 static int	sysctl_hw_iostatnames(SYSCTLFN_PROTO);
85 static int	sysctl_hw_iostats(SYSCTLFN_PROTO);
86 
87 static int
88 iostati_getnames(int disk_only, char *oldp, size_t *oldlenp, const void *newp,
89 		u_int namelen);
90 
91 /*
92  * A global list of all drives attached to the system.  May grow or
93  * shrink over time.
94  */
95 struct iostatlist_head iostatlist = TAILQ_HEAD_INITIALIZER(iostatlist);
96 int iostat_count;		/* number of drives in global drivelist */
97 krwlock_t iostatlist_lock;
98 
99 static void sysctl_io_stats_setup(struct sysctllog **);
100 
101 /*
102  * Initialise the iostat subsystem.
103  */
104 void
105 iostat_init(void)
106 {
107 
108 	rw_init(&iostatlist_lock);
109 	sysctl_io_stats_setup(NULL);
110 }
111 
112 /*
113  * Searches the iostatlist for the iostat corresponding to the
114  * name provided.
115  */
116 struct io_stats *
117 iostat_find(const char *name)
118 {
119 	struct io_stats *iostatp;
120 
121 	KASSERT(name != NULL);
122 
123 	rw_enter(&iostatlist_lock, RW_READER);
124 	TAILQ_FOREACH(iostatp, &iostatlist, io_link) {
125 		if (strcmp(iostatp->io_name, name) == 0) {
126 			break;
127 		}
128 	}
129 	rw_exit(&iostatlist_lock);
130 
131 	return iostatp;
132 }
133 
134 /*
135  * Allocate and initialise memory for the i/o statistics.
136  */
137 struct io_stats *
138 iostat_alloc(int32_t type, void *parent, const char *name)
139 {
140 	struct io_stats *stats;
141 
142 	stats = kmem_zalloc(sizeof(*stats), KM_SLEEP);
143 	stats->io_type = type;
144 	stats->io_parent = parent;
145 	(void)strlcpy(stats->io_name, name, sizeof(stats->io_name));
146 
147 	/*
148 	 * Set the attached timestamp.
149 	 */
150 	getmicrouptime(&stats->io_attachtime);
151 
152 	/*
153 	 * Link into the drivelist.
154 	 */
155 	rw_enter(&iostatlist_lock, RW_WRITER);
156 	TAILQ_INSERT_TAIL(&iostatlist, stats, io_link);
157 	iostat_count++;
158 	rw_exit(&iostatlist_lock);
159 
160 	return stats;
161 }
162 
163 /*
164  * Remove i/o from stats collection.
165  */
166 void
167 iostat_free(struct io_stats *stats)
168 {
169 
170 	/*
171 	 * Remove from the iostat list.
172 	 */
173 	if (iostat_count == 0)
174 		panic("iostat_free: iostat_count == 0");
175 	rw_enter(&iostatlist_lock, RW_WRITER);
176 	TAILQ_REMOVE(&iostatlist, stats, io_link);
177 	iostat_count--;
178 	rw_exit(&iostatlist_lock);
179 	kmem_free(stats, sizeof(*stats));
180 }
181 
182 /*
183  * multiply timeval by unsigned integer and add to result
184  */
185 static void
186 timermac(struct timeval *a, uint64_t count, struct timeval *res)
187 {
188 	struct timeval part = *a;
189 
190 	while (count) {
191 		if (count & 1)
192 			timeradd(res, &part, res);
193 		timeradd(&part, &part, &part);
194 		count >>= 1;
195 	}
196 }
197 
198 /*
199  * Increment the iostat wait counter.
200  * Accumulate wait time and timesum.
201  *
202  * Wait time is spent in the device bufq.
203  */
204 void
205 iostat_wait(struct io_stats *stats)
206 {
207 	struct timeval dv_time, diff_time;
208 	int32_t count;
209 
210 	KASSERT(stats->io_wait >= 0);
211 
212 	getmicrouptime(&dv_time);
213 
214 	timersub(&dv_time, &stats->io_waitstamp, &diff_time);
215 	count = stats->io_wait++;
216 	if (count != 0) {
217 		timermac(&diff_time, count, &stats->io_waitsum);
218 		timeradd(&stats->io_waittime, &diff_time, &stats->io_waittime);
219 	}
220 	stats->io_waitstamp = dv_time;
221 }
222 
223 /*
224  * Decrement the iostat wait counter.
225  * Increment the iostat busy counter.
226  * Accumulate wait and busy times and timesums.
227  *
228  * Busy time is spent being processed by the device.
229  *
230  * Old devices do not yet measure wait time, so skip
231  * processing it if the counter is still zero.
232  */
233 void
234 iostat_busy(struct io_stats *stats)
235 {
236 	struct timeval dv_time, diff_time;
237 	int32_t count;
238 
239 	KASSERT(stats->io_wait >= 0); /* > 0 when iostat_wait is used */
240 	KASSERT(stats->io_busy >= 0);
241 
242 	getmicrouptime(&dv_time);
243 
244 	timersub(&dv_time, &stats->io_waitstamp, &diff_time);
245 	if (stats->io_wait != 0) {
246 		count = stats->io_wait--;
247 		timermac(&diff_time, count, &stats->io_waitsum);
248 		timeradd(&stats->io_waittime, &diff_time, &stats->io_waittime);
249 	}
250 	stats->io_waitstamp = dv_time;
251 
252 	timersub(&dv_time, &stats->io_busystamp, &diff_time);
253 	count = stats->io_busy++;
254 	if (count != 0) {
255 		timermac(&diff_time, count, &stats->io_busysum);
256 		timeradd(&stats->io_busytime, &diff_time, &stats->io_busytime);
257 	}
258 	stats->io_busystamp = dv_time;
259 }
260 
261 /*
262  * Decrement the iostat busy counter, increment the byte count.
263  * Accumulate busy time and timesum.
264  */
265 void
266 iostat_unbusy(struct io_stats *stats, long bcount, int read)
267 {
268 	struct timeval dv_time, diff_time;
269 	int32_t count;
270 
271 	KASSERT(stats->io_busy > 0);
272 
273 	getmicrouptime(&dv_time);
274 	stats->io_timestamp = dv_time;
275 
276 	/* any op */
277 	timersub(&dv_time, &stats->io_busystamp, &diff_time);
278 	count = stats->io_busy--;
279 	timermac(&diff_time, count, &stats->io_busysum);
280 	timeradd(&stats->io_busytime, &diff_time, &stats->io_busytime);
281 	stats->io_busystamp = dv_time;
282 
283 	if (bcount > 0) {
284 		if (read) {
285 			stats->io_rbytes += bcount;
286 			stats->io_rxfer++;
287 		} else {
288 			stats->io_wbytes += bcount;
289 			stats->io_wxfer++;
290 		}
291 	}
292 }
293 
294 /*
295  * Return non-zero if a device has an I/O request in flight.
296  */
297 bool
298 iostat_isbusy(struct io_stats *stats)
299 {
300 
301 	return stats->io_busy != 0;
302 }
303 
304 /*
305  * Increment the seek counter.  This does look almost redundant but it
306  * abstracts the stats gathering.
307  */
308 void
309 iostat_seek(struct io_stats *stats)
310 {
311 
312 	stats->io_seek++;
313 }
314 
315 static int
316 sysctl_hw_disknames(SYSCTLFN_ARGS)
317 {
318 
319 	return iostati_getnames(1, oldp, oldlenp, newp, namelen);
320 }
321 
322 static int
323 sysctl_hw_iostatnames(SYSCTLFN_ARGS)
324 {
325 
326 	return iostati_getnames(0, oldp, oldlenp, newp, namelen);
327 }
328 
329 static int
330 iostati_getnames(int disk_only, char *oldp, size_t *oldlenp, const void *newp,
331 		 u_int namelen)
332 {
333 	char bf[IOSTATNAMELEN + 1];
334 	char *where = oldp;
335 	struct io_stats *stats;
336 	size_t needed, left, slen;
337 	int error, first;
338 
339 	if (newp != NULL)
340 		return (EPERM);
341 	if (namelen != 0)
342 		return (EINVAL);
343 
344 	first = 1;
345 	error = 0;
346 	needed = 0;
347 	left = *oldlenp;
348 
349 	rw_enter(&iostatlist_lock, RW_READER);
350 	for (stats = TAILQ_FIRST(&iostatlist); stats != NULL;
351 	    stats = TAILQ_NEXT(stats, io_link)) {
352 		if ((disk_only == 1) && (stats->io_type != IOSTAT_DISK))
353 			continue;
354 
355 		if (where == NULL)
356 			needed += strlen(stats->io_name) + 1;
357 		else {
358 			memset(bf, 0, sizeof(bf));
359 			if (first) {
360 				strncpy(bf, stats->io_name, sizeof(bf));
361 				first = 0;
362 			} else {
363 				bf[0] = ' ';
364 				strncpy(bf + 1, stats->io_name,
365 				    sizeof(bf) - 1);
366 			}
367 			bf[IOSTATNAMELEN] = '\0';
368 			slen = strlen(bf);
369 			if (left < slen + 1)
370 				break;
371 			/* +1 to copy out the trailing NUL byte */
372 			error = copyout(bf, where, slen + 1);
373 			if (error)
374 				break;
375 			where += slen;
376 			needed += slen;
377 			left -= slen;
378 		}
379 	}
380 	rw_exit(&iostatlist_lock);
381 	*oldlenp = needed;
382 	return (error);
383 }
384 
385 static int
386 sysctl_hw_iostats(SYSCTLFN_ARGS)
387 {
388 	struct io_sysctl sdrive;
389 	struct io_stats *stats;
390 	char *where = oldp;
391 	size_t tocopy, left;
392 	int error;
393 
394 	if (newp != NULL)
395 		return (EPERM);
396 
397 	/*
398 	 * The original hw.diskstats call was broken and did not require
399 	 * the userland to pass in its size of struct disk_sysctl.  This
400 	 * was fixed after NetBSD 1.6 was released.
401 	 */
402 	if (namelen == 0)
403 		tocopy = offsetof(struct io_sysctl, busy);
404 	else
405 		tocopy = name[0];
406 
407 	if (where == NULL) {
408 		*oldlenp = iostat_count * tocopy;
409 		return (0);
410 	}
411 
412 	error = 0;
413 	left = *oldlenp;
414 	memset(&sdrive, 0, sizeof(sdrive));
415 	*oldlenp = 0;
416 
417 	rw_enter(&iostatlist_lock, RW_READER);
418 	TAILQ_FOREACH(stats, &iostatlist, io_link) {
419 		if (left < tocopy)
420 			break;
421 
422 		strncpy(sdrive.name, stats->io_name, sizeof(sdrive.name));
423 		sdrive.attachtime_sec = stats->io_attachtime.tv_sec;
424 		sdrive.attachtime_usec = stats->io_attachtime.tv_usec;
425 		sdrive.timestamp_sec = stats->io_busystamp.tv_sec;
426 		sdrive.timestamp_usec = stats->io_busystamp.tv_usec;
427 
428 		sdrive.time_sec = stats->io_busytime.tv_sec;
429 		sdrive.time_usec = stats->io_busytime.tv_usec;
430 
431 		sdrive.seek = stats->io_seek;
432 
433 		sdrive.rxfer = stats->io_rxfer;
434 		sdrive.wxfer = stats->io_wxfer;
435 		sdrive.xfer = stats->io_rxfer + stats->io_wxfer;
436 
437 		sdrive.rbytes = stats->io_rbytes;
438 		sdrive.wbytes = stats->io_wbytes;
439 		sdrive.bytes = stats->io_rbytes + stats->io_wbytes;
440 
441 		sdrive.wait_sec = stats->io_waittime.tv_sec;
442 		sdrive.wait_usec = stats->io_waittime.tv_usec;
443 
444 		sdrive.time_sec = stats->io_busytime.tv_sec;
445 		sdrive.time_usec = stats->io_busytime.tv_usec;
446 
447 		sdrive.waitsum_sec = stats->io_waitsum.tv_sec;
448 		sdrive.waitsum_usec = stats->io_waitsum.tv_usec;
449 
450 		sdrive.busysum_sec = stats->io_busysum.tv_sec;
451 		sdrive.busysum_usec = stats->io_busysum.tv_usec;
452 
453 		sdrive.busy = stats->io_busy;
454 
455 		error = copyout(&sdrive, where, uimin(tocopy, sizeof(sdrive)));
456 		if (error)
457 			break;
458 		where += tocopy;
459 		*oldlenp += tocopy;
460 		left -= tocopy;
461 	}
462 	rw_exit(&iostatlist_lock);
463 	return (error);
464 }
465 
466 static void
467 sysctl_io_stats_setup(struct sysctllog **clog)
468 {
469 
470 	sysctl_createv(clog, 0, NULL, NULL,
471 		       CTLFLAG_PERMANENT,
472 		       CTLTYPE_STRING, "disknames",
473 		       SYSCTL_DESCR("List of disk drives present"),
474 		       sysctl_hw_disknames, 0, NULL, 0,
475 		       CTL_HW, HW_DISKNAMES, CTL_EOL);
476 	sysctl_createv(clog, 0, NULL, NULL,
477 		       CTLFLAG_PERMANENT,
478 		       CTLTYPE_STRING, "iostatnames",
479 		       SYSCTL_DESCR("I/O stats are being collected for these"
480 				    " devices"),
481 		       sysctl_hw_iostatnames, 0, NULL, 0,
482 		       CTL_HW, HW_IOSTATNAMES, CTL_EOL);
483 	sysctl_createv(clog, 0, NULL, NULL,
484 		       CTLFLAG_PERMANENT,
485 		       CTLTYPE_STRUCT, "iostats",
486 		       SYSCTL_DESCR("Statistics on device I/O operations"),
487 		       sysctl_hw_iostats, 0, NULL, 0,
488 		       CTL_HW, HW_IOSTATS, CTL_EOL);
489 }
490