1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1997, 1998, 1999 Kenneth D. Merry.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 *    derived from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD$");
33
34#include <sys/param.h>
35#include <sys/disk.h>
36#include <sys/kernel.h>
37#include <sys/systm.h>
38#include <sys/bio.h>
39#include <sys/devicestat.h>
40#include <sys/sdt.h>
41#include <sys/sysctl.h>
42#include <sys/malloc.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45#include <sys/conf.h>
46#include <vm/vm.h>
47#include <vm/pmap.h>
48
49#include <machine/atomic.h>
50
51SDT_PROVIDER_DEFINE(io);
52
53SDT_PROBE_DEFINE2(io, , , start, "struct bio *", "struct devstat *");
54SDT_PROBE_DEFINE2(io, , , done, "struct bio *", "struct devstat *");
55SDT_PROBE_DEFINE2(io, , , wait__start, "struct bio *",
56    "struct devstat *");
57SDT_PROBE_DEFINE2(io, , , wait__done, "struct bio *",
58    "struct devstat *");
59
60#define	DTRACE_DEVSTAT_START()		SDT_PROBE2(io, , , start, NULL, ds)
61#define	DTRACE_DEVSTAT_BIO_START()	SDT_PROBE2(io, , , start, bp, ds)
62#define	DTRACE_DEVSTAT_DONE()		SDT_PROBE2(io, , , done, NULL, ds)
63#define	DTRACE_DEVSTAT_BIO_DONE()	SDT_PROBE2(io, , , done, bp, ds)
64#define	DTRACE_DEVSTAT_WAIT_START()	SDT_PROBE2(io, , , wait__start, NULL, ds)
65#define	DTRACE_DEVSTAT_WAIT_DONE()	SDT_PROBE2(io, , , wait__done, NULL, ds)
66
67static int devstat_num_devs;
68static long devstat_generation = 1;
69static int devstat_version = DEVSTAT_VERSION;
70static int devstat_current_devnumber;
71static struct mtx devstat_mutex;
72MTX_SYSINIT(devstat_mutex, &devstat_mutex, "devstat", MTX_DEF);
73
74static struct devstatlist device_statq = STAILQ_HEAD_INITIALIZER(device_statq);
75static struct devstat *devstat_alloc(void);
76static void devstat_free(struct devstat *);
77static void devstat_add_entry(struct devstat *ds, const void *dev_name,
78		       int unit_number, uint32_t block_size,
79		       devstat_support_flags flags,
80		       devstat_type_flags device_type,
81		       devstat_priority priority);
82
83/*
84 * Allocate a devstat and initialize it
85 */
86struct devstat *
87devstat_new_entry(const void *dev_name,
88		  int unit_number, uint32_t block_size,
89		  devstat_support_flags flags,
90		  devstat_type_flags device_type,
91		  devstat_priority priority)
92{
93	struct devstat *ds;
94
95	mtx_assert(&devstat_mutex, MA_NOTOWNED);
96
97	ds = devstat_alloc();
98	mtx_lock(&devstat_mutex);
99	if (unit_number == -1) {
100		ds->unit_number = unit_number;
101		ds->id = dev_name;
102		binuptime(&ds->creation_time);
103		devstat_generation++;
104	} else {
105		devstat_add_entry(ds, dev_name, unit_number, block_size,
106				  flags, device_type, priority);
107	}
108	mtx_unlock(&devstat_mutex);
109	return (ds);
110}
111
112/*
113 * Take a malloced and zeroed devstat structure given to us, fill it in
114 * and add it to the queue of devices.
115 */
116static void
117devstat_add_entry(struct devstat *ds, const void *dev_name,
118		  int unit_number, uint32_t block_size,
119		  devstat_support_flags flags,
120		  devstat_type_flags device_type,
121		  devstat_priority priority)
122{
123	struct devstatlist *devstat_head;
124	struct devstat *ds_tmp;
125
126	mtx_assert(&devstat_mutex, MA_OWNED);
127	devstat_num_devs++;
128
129	devstat_head = &device_statq;
130
131	/*
132	 * Priority sort.  Each driver passes in its priority when it adds
133	 * its devstat entry.  Drivers are sorted first by priority, and
134	 * then by probe order.
135	 *
136	 * For the first device, we just insert it, since the priority
137	 * doesn't really matter yet.  Subsequent devices are inserted into
138	 * the list using the order outlined above.
139	 */
140	if (devstat_num_devs == 1)
141		STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
142	else {
143		STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
144			struct devstat *ds_next;
145
146			ds_next = STAILQ_NEXT(ds_tmp, dev_links);
147
148			/*
149			 * If we find a break between higher and lower
150			 * priority items, and if this item fits in the
151			 * break, insert it.  This also applies if the
152			 * "lower priority item" is the end of the list.
153			 */
154			if ((priority <= ds_tmp->priority)
155			 && ((ds_next == NULL)
156			   || (priority > ds_next->priority))) {
157				STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
158						    dev_links);
159				break;
160			} else if (priority > ds_tmp->priority) {
161				/*
162				 * If this is the case, we should be able
163				 * to insert ourselves at the head of the
164				 * list.  If we can't, something is wrong.
165				 */
166				if (ds_tmp == STAILQ_FIRST(devstat_head)) {
167					STAILQ_INSERT_HEAD(devstat_head,
168							   ds, dev_links);
169					break;
170				} else {
171					STAILQ_INSERT_TAIL(devstat_head,
172							   ds, dev_links);
173					printf("devstat_add_entry: HELP! "
174					       "sorting problem detected "
175					       "for name %p unit %d\n",
176					       dev_name, unit_number);
177					break;
178				}
179			}
180		}
181	}
182
183	ds->device_number = devstat_current_devnumber++;
184	ds->unit_number = unit_number;
185	strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
186	ds->block_size = block_size;
187	ds->flags = flags;
188	ds->device_type = device_type;
189	ds->priority = priority;
190	binuptime(&ds->creation_time);
191	devstat_generation++;
192}
193
194/*
195 * Remove a devstat structure from the list of devices.
196 */
197void
198devstat_remove_entry(struct devstat *ds)
199{
200	struct devstatlist *devstat_head;
201
202	mtx_assert(&devstat_mutex, MA_NOTOWNED);
203	if (ds == NULL)
204		return;
205
206	mtx_lock(&devstat_mutex);
207
208	devstat_head = &device_statq;
209
210	/* Remove this entry from the devstat queue */
211	atomic_add_acq_int(&ds->sequence1, 1);
212	if (ds->unit_number != -1) {
213		devstat_num_devs--;
214		STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
215	}
216	devstat_free(ds);
217	devstat_generation++;
218	mtx_unlock(&devstat_mutex);
219}
220
221/*
222 * Record a transaction start.
223 *
224 * See comments for devstat_end_transaction().  Ordering is very important
225 * here.
226 */
227void
228devstat_start_transaction(struct devstat *ds, const struct bintime *now)
229{
230
231	mtx_assert(&devstat_mutex, MA_NOTOWNED);
232
233	/* sanity check */
234	if (ds == NULL)
235		return;
236
237	atomic_add_acq_int(&ds->sequence1, 1);
238	/*
239	 * We only want to set the start time when we are going from idle
240	 * to busy.  The start time is really the start of the latest busy
241	 * period.
242	 */
243	if (ds->start_count == ds->end_count) {
244		if (now != NULL)
245			ds->busy_from = *now;
246		else
247			binuptime(&ds->busy_from);
248	}
249	ds->start_count++;
250	atomic_add_rel_int(&ds->sequence0, 1);
251	DTRACE_DEVSTAT_START();
252}
253
254void
255devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
256{
257
258	mtx_assert(&devstat_mutex, MA_NOTOWNED);
259
260	/* sanity check */
261	if (ds == NULL)
262		return;
263
264	binuptime(&bp->bio_t0);
265	devstat_start_transaction_bio_t0(ds, bp);
266}
267
268void
269devstat_start_transaction_bio_t0(struct devstat *ds, struct bio *bp)
270{
271
272	/* sanity check */
273	if (ds == NULL)
274		return;
275
276	devstat_start_transaction(ds, &bp->bio_t0);
277	DTRACE_DEVSTAT_BIO_START();
278}
279
280/*
281 * Record the ending of a transaction, and incrment the various counters.
282 *
283 * Ordering in this function, and in devstat_start_transaction() is VERY
284 * important.  The idea here is to run without locks, so we are very
285 * careful to only modify some fields on the way "down" (i.e. at
286 * transaction start) and some fields on the way "up" (i.e. at transaction
287 * completion).  One exception is busy_from, which we only modify in
288 * devstat_start_transaction() when there are no outstanding transactions,
289 * and thus it can't be modified in devstat_end_transaction()
290 * simultaneously.
291 *
292 * The sequence0 and sequence1 fields are provided to enable an application
293 * spying on the structures with mmap(2) to tell when a structure is in a
294 * consistent state or not.
295 *
296 * For this to work 100% reliably, it is important that the two fields
297 * are at opposite ends of the structure and that they are incremented
298 * in the opposite order of how a memcpy(3) in userland would copy them.
299 * We assume that the copying happens front to back, but there is actually
300 * no way short of writing your own memcpy(3) replacement to guarantee
301 * this will be the case.
302 *
303 * In addition to this, being a kind of locks, they must be updated with
304 * atomic instructions using appropriate memory barriers.
305 */
306void
307devstat_end_transaction(struct devstat *ds, uint32_t bytes,
308			devstat_tag_type tag_type, devstat_trans_flags flags,
309			const struct bintime *now, const struct bintime *then)
310{
311	struct bintime dt, lnow;
312
313	/* sanity check */
314	if (ds == NULL)
315		return;
316
317	if (now == NULL) {
318		binuptime(&lnow);
319		now = &lnow;
320	}
321
322	atomic_add_acq_int(&ds->sequence1, 1);
323	/* Update byte and operations counts */
324	ds->bytes[flags] += bytes;
325	ds->operations[flags]++;
326
327	/*
328	 * Keep a count of the various tag types sent.
329	 */
330	if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
331	    tag_type != DEVSTAT_TAG_NONE)
332		ds->tag_types[tag_type]++;
333
334	if (then != NULL) {
335		/* Update duration of operations */
336		dt = *now;
337		bintime_sub(&dt, then);
338		bintime_add(&ds->duration[flags], &dt);
339	}
340
341	/* Accumulate busy time */
342	dt = *now;
343	bintime_sub(&dt, &ds->busy_from);
344	bintime_add(&ds->busy_time, &dt);
345	ds->busy_from = *now;
346
347	ds->end_count++;
348	atomic_add_rel_int(&ds->sequence0, 1);
349	DTRACE_DEVSTAT_DONE();
350}
351
352void
353devstat_end_transaction_bio(struct devstat *ds, const struct bio *bp)
354{
355
356	devstat_end_transaction_bio_bt(ds, bp, NULL);
357}
358
359void
360devstat_end_transaction_bio_bt(struct devstat *ds, const struct bio *bp,
361    const struct bintime *now)
362{
363	devstat_trans_flags flg;
364	devstat_tag_type tag;
365
366	/* sanity check */
367	if (ds == NULL)
368		return;
369
370	if (bp->bio_flags & BIO_ORDERED)
371		tag = DEVSTAT_TAG_ORDERED;
372	else
373		tag = DEVSTAT_TAG_SIMPLE;
374	if (bp->bio_cmd == BIO_DELETE)
375		flg = DEVSTAT_FREE;
376	else if ((bp->bio_cmd == BIO_READ)
377	      || ((bp->bio_cmd == BIO_ZONE)
378	       && (bp->bio_zone.zone_cmd == DISK_ZONE_REPORT_ZONES)))
379		flg = DEVSTAT_READ;
380	else if (bp->bio_cmd == BIO_WRITE)
381		flg = DEVSTAT_WRITE;
382	else
383		flg = DEVSTAT_NO_DATA;
384
385	devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
386				tag, flg, now, &bp->bio_t0);
387	DTRACE_DEVSTAT_BIO_DONE();
388}
389
390/*
391 * This is the sysctl handler for the devstat package.  The data pushed out
392 * on the kern.devstat.all sysctl variable consists of the current devstat
393 * generation number, and then an array of devstat structures, one for each
394 * device in the system.
395 *
396 * This is more cryptic that obvious, but basically we neither can nor
397 * want to hold the devstat_mutex for any amount of time, so we grab it
398 * only when we need to and keep an eye on devstat_generation all the time.
399 */
400static int
401sysctl_devstat(SYSCTL_HANDLER_ARGS)
402{
403	int error;
404	long mygen;
405	struct devstat *nds;
406
407	mtx_assert(&devstat_mutex, MA_NOTOWNED);
408
409	/*
410	 * XXX devstat_generation should really be "volatile" but that
411	 * XXX freaks out the sysctl macro below.  The places where we
412	 * XXX change it and inspect it are bracketed in the mutex which
413	 * XXX guarantees us proper write barriers.  I don't believe the
414	 * XXX compiler is allowed to optimize mygen away across calls
415	 * XXX to other functions, so the following is belived to be safe.
416	 */
417	mygen = devstat_generation;
418
419	error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
420
421	if (devstat_num_devs == 0)
422		return(0);
423
424	if (error != 0)
425		return (error);
426
427	mtx_lock(&devstat_mutex);
428	nds = STAILQ_FIRST(&device_statq);
429	if (mygen != devstat_generation)
430		error = EBUSY;
431	mtx_unlock(&devstat_mutex);
432
433	if (error != 0)
434		return (error);
435
436	for (;nds != NULL;) {
437		error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
438		if (error != 0)
439			return (error);
440		mtx_lock(&devstat_mutex);
441		if (mygen != devstat_generation)
442			error = EBUSY;
443		else
444			nds = STAILQ_NEXT(nds, dev_links);
445		mtx_unlock(&devstat_mutex);
446		if (error != 0)
447			return (error);
448	}
449	return(error);
450}
451
452/*
453 * Sysctl entries for devstat.  The first one is a node that all the rest
454 * hang off of.
455 */
456static SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD, NULL,
457    "Device Statistics");
458
459SYSCTL_PROC(_kern_devstat, OID_AUTO, all, CTLFLAG_RD|CTLTYPE_OPAQUE,
460    NULL, 0, sysctl_devstat, "S,devstat", "All devices in the devstat list");
461/*
462 * Export the number of devices in the system so that userland utilities
463 * can determine how much memory to allocate to hold all the devices.
464 */
465SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
466    &devstat_num_devs, 0, "Number of devices in the devstat list");
467SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
468    &devstat_generation, 0, "Devstat list generation");
469SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
470    &devstat_version, 0, "Devstat list version number");
471
472/*
473 * Allocator for struct devstat structures.  We sub-allocate these from pages
474 * which we get from malloc.  These pages are exported for mmap(2)'ing through
475 * a miniature device driver
476 */
477
478#define statsperpage (PAGE_SIZE / sizeof(struct devstat))
479
480static d_ioctl_t devstat_ioctl;
481static d_mmap_t devstat_mmap;
482
483static struct cdevsw devstat_cdevsw = {
484	.d_version =	D_VERSION,
485	.d_ioctl =	devstat_ioctl,
486	.d_mmap =	devstat_mmap,
487	.d_name =	"devstat",
488};
489
490struct statspage {
491	TAILQ_ENTRY(statspage)	list;
492	struct devstat		*stat;
493	u_int			nfree;
494};
495
496static size_t pagelist_pages = 0;
497static TAILQ_HEAD(, statspage)	pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
498static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
499
500static int
501devstat_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
502    struct thread *td)
503{
504	int error = ENOTTY;
505
506	switch (cmd) {
507	case DIOCGMEDIASIZE:
508		error = 0;
509		*(off_t *)data = pagelist_pages * PAGE_SIZE;
510		break;
511	}
512
513	return (error);
514}
515
516static int
517devstat_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
518    int nprot, vm_memattr_t *memattr)
519{
520	struct statspage *spp;
521
522	if (nprot != VM_PROT_READ)
523		return (-1);
524	mtx_lock(&devstat_mutex);
525	TAILQ_FOREACH(spp, &pagelist, list) {
526		if (offset == 0) {
527			*paddr = vtophys(spp->stat);
528			mtx_unlock(&devstat_mutex);
529			return (0);
530		}
531		offset -= PAGE_SIZE;
532	}
533	mtx_unlock(&devstat_mutex);
534	return (-1);
535}
536
537static struct devstat *
538devstat_alloc(void)
539{
540	struct devstat *dsp;
541	struct statspage *spp, *spp2;
542	u_int u;
543	static int once;
544
545	mtx_assert(&devstat_mutex, MA_NOTOWNED);
546	if (!once) {
547		make_dev_credf(MAKEDEV_ETERNAL | MAKEDEV_CHECKNAME,
548		    &devstat_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0444,
549		    DEVSTAT_DEVICE_NAME);
550		once = 1;
551	}
552	spp2 = NULL;
553	mtx_lock(&devstat_mutex);
554	for (;;) {
555		TAILQ_FOREACH(spp, &pagelist, list) {
556			if (spp->nfree > 0)
557				break;
558		}
559		if (spp != NULL)
560			break;
561		mtx_unlock(&devstat_mutex);
562		spp2 = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
563		spp2->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
564		spp2->nfree = statsperpage;
565
566		/*
567		 * If free statspages were added while the lock was released
568		 * just reuse them.
569		 */
570		mtx_lock(&devstat_mutex);
571		TAILQ_FOREACH(spp, &pagelist, list)
572			if (spp->nfree > 0)
573				break;
574		if (spp == NULL) {
575			spp = spp2;
576
577			/*
578			 * It would make more sense to add the new page at the
579			 * head but the order on the list determine the
580			 * sequence of the mapping so we can't do that.
581			 */
582			pagelist_pages++;
583			TAILQ_INSERT_TAIL(&pagelist, spp, list);
584		} else
585			break;
586	}
587	dsp = spp->stat;
588	for (u = 0; u < statsperpage; u++) {
589		if (dsp->allocated == 0)
590			break;
591		dsp++;
592	}
593	spp->nfree--;
594	dsp->allocated = 1;
595	mtx_unlock(&devstat_mutex);
596	if (spp2 != NULL && spp2 != spp) {
597		free(spp2->stat, M_DEVSTAT);
598		free(spp2, M_DEVSTAT);
599	}
600	return (dsp);
601}
602
603static void
604devstat_free(struct devstat *dsp)
605{
606	struct statspage *spp;
607
608	mtx_assert(&devstat_mutex, MA_OWNED);
609	bzero(dsp, sizeof *dsp);
610	TAILQ_FOREACH(spp, &pagelist, list) {
611		if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
612			spp->nfree++;
613			return;
614		}
615	}
616}
617
618SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
619    SYSCTL_NULL_INT_PTR, sizeof(struct devstat), "sizeof(struct devstat)");
620