xref: /netbsd-src/sys/dev/dksubr.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /* $NetBSD: dksubr.c,v 1.88 2016/06/27 18:27:51 christos Exp $ */
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 1999, 2002, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and Roland C. Dowdeswell.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: dksubr.c,v 1.88 2016/06/27 18:27:51 christos Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/stat.h>
38 #include <sys/proc.h>
39 #include <sys/ioctl.h>
40 #include <sys/device.h>
41 #include <sys/disk.h>
42 #include <sys/disklabel.h>
43 #include <sys/buf.h>
44 #include <sys/bufq.h>
45 #include <sys/vnode.h>
46 #include <sys/fcntl.h>
47 #include <sys/namei.h>
48 #include <sys/module.h>
49 #include <sys/syslog.h>
50 
51 #include <dev/dkvar.h>
52 #include <miscfs/specfs/specdev.h> /* for v_rdev */
53 
54 int	dkdebug = 0;
55 
56 #ifdef DEBUG
57 #define DKDB_FOLLOW	0x1
58 #define DKDB_INIT	0x2
59 #define DKDB_VNODE	0x4
60 #define DKDB_DUMP	0x8
61 
62 #define IFDEBUG(x,y)		if (dkdebug & (x)) y
63 #define DPRINTF(x,y)		IFDEBUG(x, printf y)
64 #define DPRINTF_FOLLOW(y)	DPRINTF(DKDB_FOLLOW, y)
65 #else
66 #define IFDEBUG(x,y)
67 #define DPRINTF(x,y)
68 #define DPRINTF_FOLLOW(y)
69 #endif
70 
71 #define DKF_READYFORDUMP	(DKF_INITED|DKF_TAKEDUMP)
72 
73 static int dk_subr_modcmd(modcmd_t, void *);
74 
75 #define DKLABELDEV(dev)	\
76 	(MAKEDISKDEV(major((dev)), DISKUNIT((dev)), RAW_PART))
77 
78 static void	dk_makedisklabel(struct dk_softc *);
79 static int	dk_translate(struct dk_softc *, struct buf *);
80 static void	dk_done1(struct dk_softc *, struct buf *, bool);
81 
82 void
83 dk_init(struct dk_softc *dksc, device_t dev, int dtype)
84 {
85 
86 	memset(dksc, 0x0, sizeof(*dksc));
87 	dksc->sc_dtype = dtype;
88 	dksc->sc_dev = dev;
89 
90 	strlcpy(dksc->sc_xname, device_xname(dev), DK_XNAME_SIZE);
91 	dksc->sc_dkdev.dk_name = dksc->sc_xname;
92 }
93 
94 void
95 dk_attach(struct dk_softc *dksc)
96 {
97 	KASSERT(dksc->sc_dev != NULL);
98 
99 	mutex_init(&dksc->sc_iolock, MUTEX_DEFAULT, IPL_VM);
100 	dksc->sc_flags |= DKF_READYFORDUMP;
101 #ifdef DIAGNOSTIC
102 	dksc->sc_flags |= DKF_WARNLABEL | DKF_LABELSANITY;
103 #endif
104 
105 	/* Attach the device into the rnd source list. */
106 	rnd_attach_source(&dksc->sc_rnd_source, dksc->sc_xname,
107 	    RND_TYPE_DISK, RND_FLAG_DEFAULT);
108 }
109 
110 void
111 dk_detach(struct dk_softc *dksc)
112 {
113 	/* Unhook the entropy source. */
114 	rnd_detach_source(&dksc->sc_rnd_source);
115 
116 	dksc->sc_flags &= ~DKF_READYFORDUMP;
117 	mutex_destroy(&dksc->sc_iolock);
118 }
119 
120 /* ARGSUSED */
121 int
122 dk_open(struct dk_softc *dksc, dev_t dev,
123     int flags, int fmt, struct lwp *l)
124 {
125 	struct	disklabel *lp = dksc->sc_dkdev.dk_label;
126 	int	part = DISKPART(dev);
127 	int	pmask = 1 << part;
128 	int	ret = 0;
129 	struct disk *dk = &dksc->sc_dkdev;
130 
131 	DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__,
132 	    dksc->sc_xname, dksc, dev, flags));
133 
134 	mutex_enter(&dk->dk_openlock);
135 
136 	/*
137 	 * If there are wedges, and this is not RAW_PART, then we
138 	 * need to fail.
139 	 */
140 	if (dk->dk_nwedges != 0 && part != RAW_PART) {
141 		ret = EBUSY;
142 		goto done;
143 	}
144 
145 	/*
146 	 * If we're init'ed and there are no other open partitions then
147 	 * update the in-core disklabel.
148 	 */
149 	if ((dksc->sc_flags & DKF_INITED)) {
150 		if ((dksc->sc_flags & DKF_VLABEL) == 0) {
151 			dksc->sc_flags |= DKF_VLABEL;
152 			dk_getdisklabel(dksc, dev);
153 		}
154 	}
155 
156 	/* Fail if we can't find the partition. */
157 	if (part != RAW_PART &&
158 	    ((dksc->sc_flags & DKF_VLABEL) == 0 ||
159 	     part >= lp->d_npartitions ||
160 	     lp->d_partitions[part].p_fstype == FS_UNUSED)) {
161 		ret = ENXIO;
162 		goto done;
163 	}
164 
165 	/* Mark our unit as open. */
166 	switch (fmt) {
167 	case S_IFCHR:
168 		dk->dk_copenmask |= pmask;
169 		break;
170 	case S_IFBLK:
171 		dk->dk_bopenmask |= pmask;
172 		break;
173 	}
174 
175 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
176 
177 done:
178 	mutex_exit(&dk->dk_openlock);
179 	return ret;
180 }
181 
182 /* ARGSUSED */
183 int
184 dk_close(struct dk_softc *dksc, dev_t dev,
185     int flags, int fmt, struct lwp *l)
186 {
187 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
188 	int	part = DISKPART(dev);
189 	int	pmask = 1 << part;
190 	struct disk *dk = &dksc->sc_dkdev;
191 
192 	DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__,
193 	    dksc->sc_xname, dksc, dev, flags));
194 
195 	mutex_enter(&dk->dk_openlock);
196 
197 	switch (fmt) {
198 	case S_IFCHR:
199 		dk->dk_copenmask &= ~pmask;
200 		break;
201 	case S_IFBLK:
202 		dk->dk_bopenmask &= ~pmask;
203 		break;
204 	}
205 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
206 
207 	if (dk->dk_openmask == 0) {
208 		if (dkd->d_lastclose != NULL)
209 			(*dkd->d_lastclose)(dksc->sc_dev);
210 		if ((dksc->sc_flags & DKF_KLABEL) == 0)
211 			dksc->sc_flags &= ~DKF_VLABEL;
212 	}
213 
214 	mutex_exit(&dk->dk_openlock);
215 	return 0;
216 }
217 
218 static int
219 dk_translate(struct dk_softc *dksc, struct buf *bp)
220 {
221 	int	part;
222 	int	wlabel;
223 	daddr_t	blkno;
224 	struct disklabel *lp;
225 	struct disk *dk;
226 	uint64_t numsecs;
227 	unsigned secsize;
228 
229 	lp = dksc->sc_dkdev.dk_label;
230 	dk = &dksc->sc_dkdev;
231 
232 	part = DISKPART(bp->b_dev);
233 	numsecs = dk->dk_geom.dg_secperunit;
234 	secsize = dk->dk_geom.dg_secsize;
235 
236 	/*
237 	 * The transfer must be a whole number of blocks and the offset must
238 	 * not be negative.
239 	 */
240 	if ((bp->b_bcount % secsize) != 0 || bp->b_blkno < 0) {
241 		bp->b_error = EINVAL;
242 		goto done;
243 	}
244 
245 	/* If there is nothing to do, then we are done */
246 	if (bp->b_bcount == 0)
247 		goto done;
248 
249 	wlabel = dksc->sc_flags & (DKF_WLABEL|DKF_LABELLING);
250 	if (part == RAW_PART) {
251 		uint64_t numblocks = btodb(numsecs * secsize);
252 		if (bounds_check_with_mediasize(bp, DEV_BSIZE, numblocks) <= 0)
253 			goto done;
254 	} else {
255 		if (bounds_check_with_label(&dksc->sc_dkdev, bp, wlabel) <= 0)
256 			goto done;
257 	}
258 
259 	/*
260 	 * Convert the block number to absolute and put it in terms
261 	 * of the device's logical block size.
262 	 */
263 	if (secsize >= DEV_BSIZE)
264 		blkno = bp->b_blkno / (secsize / DEV_BSIZE);
265 	else
266 		blkno = bp->b_blkno * (DEV_BSIZE / secsize);
267 
268 	if (part != RAW_PART)
269 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
270 	bp->b_rawblkno = blkno;
271 
272 	return -1;
273 
274 done:
275 	bp->b_resid = bp->b_bcount;
276 	return bp->b_error;
277 }
278 
279 static int
280 dk_strategy1(struct dk_softc *dksc, struct buf *bp)
281 {
282 	int error;
283 
284 	DPRINTF_FOLLOW(("%s(%s, %p, %p)\n", __func__,
285 	    dksc->sc_xname, dksc, bp));
286 
287 	if (!(dksc->sc_flags & DKF_INITED)) {
288 		DPRINTF_FOLLOW(("%s: not inited\n", __func__));
289 		bp->b_error  = ENXIO;
290 		biodone(bp);
291 		return 1;
292 	}
293 
294 	error = dk_translate(dksc, bp);
295 	if (error >= 0) {
296 		biodone(bp);
297 		return 1;
298 	}
299 
300 	return 0;
301 }
302 
303 void
304 dk_strategy(struct dk_softc *dksc, struct buf *bp)
305 {
306 	int error;
307 
308 	error = dk_strategy1(dksc, bp);
309 	if (error)
310 		return;
311 
312 	/*
313 	 * Queue buffer and start unit
314 	 */
315 	dk_start(dksc, bp);
316 }
317 
318 int
319 dk_strategy_defer(struct dk_softc *dksc, struct buf *bp)
320 {
321 	int error;
322 
323 	error = dk_strategy1(dksc, bp);
324 	if (error)
325 		return error;
326 
327 	/*
328 	 * Queue buffer only
329 	 */
330 	mutex_enter(&dksc->sc_iolock);
331 	bufq_put(dksc->sc_bufq, bp);
332 	mutex_exit(&dksc->sc_iolock);
333 
334 	return 0;
335 }
336 
337 int
338 dk_strategy_pending(struct dk_softc *dksc)
339 {
340 	struct buf *bp;
341 
342 	if (!(dksc->sc_flags & DKF_INITED)) {
343 		DPRINTF_FOLLOW(("%s: not inited\n", __func__));
344 		return 0;
345 	}
346 
347 	mutex_enter(&dksc->sc_iolock);
348 	bp = bufq_peek(dksc->sc_bufq);
349 	mutex_exit(&dksc->sc_iolock);
350 
351 	return bp != NULL;
352 }
353 
354 void
355 dk_start(struct dk_softc *dksc, struct buf *bp)
356 {
357 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
358 	int error;
359 
360 	if (!(dksc->sc_flags & DKF_INITED)) {
361 		DPRINTF_FOLLOW(("%s: not inited\n", __func__));
362 		return;
363 	}
364 
365 	mutex_enter(&dksc->sc_iolock);
366 
367 	if (bp != NULL)
368 		bufq_put(dksc->sc_bufq, bp);
369 
370 	if (dksc->sc_busy)
371 		goto done;
372 	dksc->sc_busy = true;
373 
374 	/*
375 	 * Peeking at the buffer queue and committing the operation
376 	 * only after success isn't atomic.
377 	 *
378 	 * So when a diskstart fails, the buffer is saved
379 	 * and tried again before the next buffer is fetched.
380 	 * dk_drain() handles flushing of a saved buffer.
381 	 *
382 	 * This keeps order of I/O operations, unlike bufq_put.
383 	 */
384 
385 	bp = dksc->sc_deferred;
386 	dksc->sc_deferred = NULL;
387 
388 	if (bp == NULL)
389 		bp = bufq_get(dksc->sc_bufq);
390 
391 	while (bp != NULL) {
392 
393 		disk_busy(&dksc->sc_dkdev);
394 		mutex_exit(&dksc->sc_iolock);
395 		error = dkd->d_diskstart(dksc->sc_dev, bp);
396 		mutex_enter(&dksc->sc_iolock);
397 		if (error == EAGAIN) {
398 			dksc->sc_deferred = bp;
399 			disk_unbusy(&dksc->sc_dkdev, 0, (bp->b_flags & B_READ));
400 			break;
401 		}
402 
403 		if (error != 0) {
404 			bp->b_error = error;
405 			bp->b_resid = bp->b_bcount;
406 			dk_done1(dksc, bp, false);
407 		}
408 
409 		bp = bufq_get(dksc->sc_bufq);
410 	}
411 
412 	dksc->sc_busy = false;
413 done:
414 	mutex_exit(&dksc->sc_iolock);
415 }
416 
417 static void
418 dk_done1(struct dk_softc *dksc, struct buf *bp, bool lock)
419 {
420 	struct disk *dk = &dksc->sc_dkdev;
421 
422 	if (bp->b_error != 0) {
423 		struct cfdriver *cd = device_cfdriver(dksc->sc_dev);
424 
425 		diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0,
426 			dk->dk_label);
427 		printf("\n");
428 	}
429 
430 	if (lock)
431 		mutex_enter(&dksc->sc_iolock);
432 	disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ));
433 	if (lock)
434 		mutex_exit(&dksc->sc_iolock);
435 
436 	rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno);
437 
438 	biodone(bp);
439 }
440 
441 void
442 dk_done(struct dk_softc *dksc, struct buf *bp)
443 {
444 	dk_done1(dksc, bp, true);
445 }
446 
447 void
448 dk_drain(struct dk_softc *dksc)
449 {
450 	struct buf *bp;
451 
452 	mutex_enter(&dksc->sc_iolock);
453 	bp = dksc->sc_deferred;
454 	dksc->sc_deferred = NULL;
455 	if (bp != NULL) {
456 		bp->b_error = EIO;
457 		bp->b_resid = bp->b_bcount;
458 		biodone(bp);
459 	}
460 	bufq_drain(dksc->sc_bufq);
461 	mutex_exit(&dksc->sc_iolock);
462 }
463 
464 int
465 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len)
466 {
467 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
468 	unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize;
469 	struct buf tmp, *bp = &tmp;
470 	int error;
471 
472 	DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__,
473 	    dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len));
474 
475 	if (!(dksc->sc_flags & DKF_INITED)) {
476 		DPRINTF_FOLLOW(("%s: not inited\n", __func__));
477 		return ENXIO;
478 	}
479 
480 	if (secsize == 0 || (pos % secsize) != 0)
481 		return EINVAL;
482 
483 	/* enough data to please the bounds checking code */
484 	bp->b_dev = dev;
485 	bp->b_blkno = (daddr_t)(pos / secsize);
486 	bp->b_bcount = len;
487 	bp->b_flags = B_WRITE;
488 
489 	error = dk_translate(dksc, bp);
490 	if (error >= 0)
491 		return error;
492 
493 	error = dkd->d_discard(dksc->sc_dev,
494 		(off_t)bp->b_rawblkno * secsize,
495 		(off_t)bp->b_bcount);
496 
497 	return error;
498 }
499 
500 int
501 dk_size(struct dk_softc *dksc, dev_t dev)
502 {
503 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
504 	struct	disklabel *lp;
505 	int	is_open;
506 	int	part;
507 	int	size;
508 
509 	if ((dksc->sc_flags & DKF_INITED) == 0)
510 		return -1;
511 
512 	part = DISKPART(dev);
513 	is_open = dksc->sc_dkdev.dk_openmask & (1 << part);
514 
515 	if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp))
516 		return -1;
517 
518 	lp = dksc->sc_dkdev.dk_label;
519 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
520 		size = -1;
521 	else
522 		size = lp->d_partitions[part].p_size *
523 		    (lp->d_secsize / DEV_BSIZE);
524 
525 	if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp))
526 		return -1;
527 
528 	return size;
529 }
530 
531 int
532 dk_ioctl(struct dk_softc *dksc, dev_t dev,
533 	    u_long cmd, void *data, int flag, struct lwp *l)
534 {
535 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
536 	struct	disklabel *lp;
537 	struct	disk *dk = &dksc->sc_dkdev;
538 #ifdef __HAVE_OLD_DISKLABEL
539 	struct	disklabel newlabel;
540 #endif
541 	int	error;
542 
543 	DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__,
544 	    dksc->sc_xname, dksc, dev, cmd));
545 
546 	/* ensure that the pseudo disk is open for writes for these commands */
547 	switch (cmd) {
548 	case DIOCSDINFO:
549 	case DIOCWDINFO:
550 #ifdef __HAVE_OLD_DISKLABEL
551 	case ODIOCSDINFO:
552 	case ODIOCWDINFO:
553 #endif
554 	case DIOCKLABEL:
555 	case DIOCWLABEL:
556 	case DIOCAWEDGE:
557 	case DIOCDWEDGE:
558 	case DIOCSSTRATEGY:
559 		if ((flag & FWRITE) == 0)
560 			return EBADF;
561 	}
562 
563 	/* ensure that the pseudo-disk is initialized for these */
564 	switch (cmd) {
565 	case DIOCGDINFO:
566 	case DIOCSDINFO:
567 	case DIOCWDINFO:
568 	case DIOCGPARTINFO:
569 	case DIOCKLABEL:
570 	case DIOCWLABEL:
571 	case DIOCGDEFLABEL:
572 	case DIOCAWEDGE:
573 	case DIOCDWEDGE:
574 	case DIOCLWEDGES:
575 	case DIOCMWEDGES:
576 	case DIOCCACHESYNC:
577 #ifdef __HAVE_OLD_DISKLABEL
578 	case ODIOCGDINFO:
579 	case ODIOCSDINFO:
580 	case ODIOCWDINFO:
581 	case ODIOCGDEFLABEL:
582 #endif
583 		if ((dksc->sc_flags & DKF_INITED) == 0)
584 			return ENXIO;
585 	}
586 
587 	error = disk_ioctl(dk, dev, cmd, data, flag, l);
588 	if (error != EPASSTHROUGH)
589 		return error;
590 	else
591 		error = 0;
592 
593 	switch (cmd) {
594 	case DIOCWDINFO:
595 	case DIOCSDINFO:
596 #ifdef __HAVE_OLD_DISKLABEL
597 	case ODIOCWDINFO:
598 	case ODIOCSDINFO:
599 #endif
600 #ifdef __HAVE_OLD_DISKLABEL
601 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
602 			memset(&newlabel, 0, sizeof newlabel);
603 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
604 			lp = &newlabel;
605 		} else
606 #endif
607 		lp = (struct disklabel *)data;
608 
609 		mutex_enter(&dk->dk_openlock);
610 		dksc->sc_flags |= DKF_LABELLING;
611 
612 		error = setdisklabel(dksc->sc_dkdev.dk_label,
613 		    lp, 0, dksc->sc_dkdev.dk_cpulabel);
614 		if (error == 0) {
615 			if (cmd == DIOCWDINFO
616 #ifdef __HAVE_OLD_DISKLABEL
617 			    || cmd == ODIOCWDINFO
618 #endif
619 			   )
620 				error = writedisklabel(DKLABELDEV(dev),
621 				    dkd->d_strategy, dksc->sc_dkdev.dk_label,
622 				    dksc->sc_dkdev.dk_cpulabel);
623 		}
624 
625 		dksc->sc_flags &= ~DKF_LABELLING;
626 		mutex_exit(&dk->dk_openlock);
627 		break;
628 
629 	case DIOCKLABEL:
630 		if (*(int *)data != 0)
631 			dksc->sc_flags |= DKF_KLABEL;
632 		else
633 			dksc->sc_flags &= ~DKF_KLABEL;
634 		break;
635 
636 	case DIOCWLABEL:
637 		if (*(int *)data != 0)
638 			dksc->sc_flags |= DKF_WLABEL;
639 		else
640 			dksc->sc_flags &= ~DKF_WLABEL;
641 		break;
642 
643 	case DIOCGDEFLABEL:
644 		dk_getdefaultlabel(dksc, (struct disklabel *)data);
645 		break;
646 
647 #ifdef __HAVE_OLD_DISKLABEL
648 	case ODIOCGDEFLABEL:
649 		dk_getdefaultlabel(dksc, &newlabel);
650 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
651 			return ENOTTY;
652 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
653 		break;
654 #endif
655 
656 	case DIOCGSTRATEGY:
657 	    {
658 		struct disk_strategy *dks = (void *)data;
659 
660 		mutex_enter(&dksc->sc_iolock);
661 		if (dksc->sc_bufq != NULL)
662 			strlcpy(dks->dks_name,
663 			    bufq_getstrategyname(dksc->sc_bufq),
664 			    sizeof(dks->dks_name));
665 		else
666 			error = EINVAL;
667 		mutex_exit(&dksc->sc_iolock);
668 		dks->dks_paramlen = 0;
669 		break;
670 	    }
671 
672 	case DIOCSSTRATEGY:
673 	    {
674 		struct disk_strategy *dks = (void *)data;
675 		struct bufq_state *new;
676 		struct bufq_state *old;
677 
678 		if (dks->dks_param != NULL) {
679 			return EINVAL;
680 		}
681 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
682 		error = bufq_alloc(&new, dks->dks_name,
683 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
684 		if (error) {
685 			return error;
686 		}
687 		mutex_enter(&dksc->sc_iolock);
688 		old = dksc->sc_bufq;
689 		if (old)
690 			bufq_move(new, old);
691 		dksc->sc_bufq = new;
692 		mutex_exit(&dksc->sc_iolock);
693 		if (old)
694 			bufq_free(old);
695 		break;
696 	    }
697 
698 	default:
699 		error = ENOTTY;
700 	}
701 
702 	return error;
703 }
704 
705 /*
706  * dk_dump dumps all of physical memory into the partition specified.
707  * This requires substantially more framework than {s,w}ddump, and hence
708  * is probably much more fragile.
709  *
710  */
711 
712 #define DKFF_READYFORDUMP(x)	(((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP)
713 static volatile int	dk_dumping = 0;
714 
715 /* ARGSUSED */
716 int
717 dk_dump(struct dk_softc *dksc, dev_t dev,
718     daddr_t blkno, void *vav, size_t size)
719 {
720 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
721 	char *va = vav;
722 	struct disklabel *lp;
723 	struct partition *p;
724 	int part, towrt, nsects, sectoff, maxblkcnt, nblk;
725 	int maxxfer, rv = 0;
726 
727 	/*
728 	 * ensure that we consider this device to be safe for dumping,
729 	 * and that the device is configured.
730 	 */
731 	if (!DKFF_READYFORDUMP(dksc->sc_flags)) {
732 		DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__,
733 		    dksc->sc_flags));
734 		return ENXIO;
735 	}
736 
737 	/* ensure that we are not already dumping */
738 	if (dk_dumping)
739 		return EFAULT;
740 	dk_dumping = 1;
741 
742 	if (dkd->d_dumpblocks == NULL) {
743 		DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__));
744 		return ENXIO;
745 	}
746 
747 	/* device specific max transfer size */
748 	maxxfer = MAXPHYS;
749 	if (dkd->d_iosize != NULL)
750 		(*dkd->d_iosize)(dksc->sc_dev, &maxxfer);
751 
752 	/* Convert to disk sectors.  Request must be a multiple of size. */
753 	part = DISKPART(dev);
754 	lp = dksc->sc_dkdev.dk_label;
755 	if ((size % lp->d_secsize) != 0) {
756 		DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size));
757 		return EFAULT;
758 	}
759 	towrt = size / lp->d_secsize;
760 	blkno = dbtob(blkno) / lp->d_secsize;   /* blkno in secsize units */
761 
762 	p = &lp->d_partitions[part];
763 	if (p->p_fstype != FS_SWAP) {
764 		DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__,
765 		    p->p_fstype));
766 		return ENXIO;
767 	}
768 	nsects = p->p_size;
769 	sectoff = p->p_offset;
770 
771 	/* Check transfer bounds against partition size. */
772 	if ((blkno < 0) || ((blkno + towrt) > nsects)) {
773 		DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, "
774 		    "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects));
775 		return EINVAL;
776 	}
777 
778 	/* Offset block number to start of partition. */
779 	blkno += sectoff;
780 
781 	/* Start dumping and return when done. */
782 	maxblkcnt = howmany(maxxfer, lp->d_secsize);
783 	while (towrt > 0) {
784 		nblk = min(maxblkcnt, towrt);
785 
786 		if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk))
787 		    != 0) {
788 			DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__,
789 			    rv));
790 			return rv;
791 		}
792 
793 		towrt -= nblk;
794 		blkno += nblk;
795 		va += nblk * lp->d_secsize;
796 	}
797 
798 	dk_dumping = 0;
799 
800 	return 0;
801 }
802 
803 /* ARGSUSED */
804 void
805 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp)
806 {
807 	struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
808 
809 	memset(lp, 0, sizeof(*lp));
810 
811 	if (dg->dg_secperunit > UINT32_MAX)
812 		lp->d_secperunit = UINT32_MAX;
813 	else
814 		lp->d_secperunit = dg->dg_secperunit;
815 	lp->d_secsize = dg->dg_secsize;
816 	lp->d_nsectors = dg->dg_nsectors;
817 	lp->d_ntracks = dg->dg_ntracks;
818 	lp->d_ncylinders = dg->dg_ncylinders;
819 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
820 
821 	strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename));
822 	lp->d_type = dksc->sc_dtype;
823 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
824 	lp->d_rpm = 3600;
825 	lp->d_interleave = 1;
826 	lp->d_flags = 0;
827 
828 	lp->d_partitions[RAW_PART].p_offset = 0;
829 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
830 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
831 	lp->d_npartitions = RAW_PART + 1;
832 
833 	lp->d_magic = DISKMAGIC;
834 	lp->d_magic2 = DISKMAGIC;
835 	lp->d_checksum = dkcksum(dksc->sc_dkdev.dk_label);
836 }
837 
838 /* ARGSUSED */
839 void
840 dk_getdisklabel(struct dk_softc *dksc, dev_t dev)
841 {
842 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
843 	struct	 disklabel *lp = dksc->sc_dkdev.dk_label;
844 	struct	 cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel;
845 	struct   disk_geom *dg = &dksc->sc_dkdev.dk_geom;
846 	struct	 partition *pp;
847 	int	 i;
848 	const char	*errstring;
849 
850 	memset(clp, 0x0, sizeof(*clp));
851 	dk_getdefaultlabel(dksc, lp);
852 	errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy,
853 	    dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel);
854 	if (errstring) {
855 		dk_makedisklabel(dksc);
856 		if (dksc->sc_flags & DKF_WARNLABEL)
857 			printf("%s: %s\n", dksc->sc_xname, errstring);
858 		return;
859 	}
860 
861 	if ((dksc->sc_flags & DKF_LABELSANITY) == 0)
862 		return;
863 
864 	/* Sanity check */
865 	if (lp->d_secperunit < UINT32_MAX ?
866 		lp->d_secperunit != dg->dg_secperunit :
867 		lp->d_secperunit > dg->dg_secperunit)
868 		printf("WARNING: %s: total sector size in disklabel (%ju) "
869 		    "!= the size of %s (%ju)\n", dksc->sc_xname,
870 		    (uintmax_t)lp->d_secperunit, dksc->sc_xname,
871 		    (uintmax_t)dg->dg_secperunit);
872 
873 	for (i=0; i < lp->d_npartitions; i++) {
874 		pp = &lp->d_partitions[i];
875 		if (pp->p_offset + pp->p_size > dg->dg_secperunit)
876 			printf("WARNING: %s: end of partition `%c' exceeds "
877 			    "the size of %s (%ju)\n", dksc->sc_xname,
878 			    'a' + i, dksc->sc_xname,
879 			    (uintmax_t)dg->dg_secperunit);
880 	}
881 }
882 
883 /* ARGSUSED */
884 static void
885 dk_makedisklabel(struct dk_softc *dksc)
886 {
887 	struct	disklabel *lp = dksc->sc_dkdev.dk_label;
888 
889 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
890 	strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname));
891 	lp->d_checksum = dkcksum(lp);
892 }
893 
894 /* This function is taken from ccd.c:1.76  --rcd */
895 
896 /*
897  * XXX this function looks too generic for dksubr.c, shouldn't we
898  *     put it somewhere better?
899  */
900 
901 /*
902  * Lookup the provided name in the filesystem.  If the file exists,
903  * is a valid block device, and isn't being used by anyone else,
904  * set *vpp to the file's vnode.
905  */
906 int
907 dk_lookup(struct pathbuf *pb, struct lwp *l, struct vnode **vpp)
908 {
909 	struct nameidata nd;
910 	struct vnode *vp;
911 	int     error;
912 
913 	if (l == NULL)
914 		return ESRCH;	/* Is ESRCH the best choice? */
915 
916 	NDINIT(&nd, LOOKUP, FOLLOW, pb);
917 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
918 		DPRINTF((DKDB_FOLLOW|DKDB_INIT),
919 		    ("%s: vn_open error = %d\n", __func__, error));
920 		return error;
921 	}
922 
923 	vp = nd.ni_vp;
924 	if (vp->v_type != VBLK) {
925 		error = ENOTBLK;
926 		goto out;
927 	}
928 
929 	/* Reopen as anonymous vnode to protect against forced unmount. */
930 	if ((error = bdevvp(vp->v_rdev, vpp)) != 0)
931 		goto out;
932 	VOP_UNLOCK(vp);
933 	if ((error = vn_close(vp, FREAD | FWRITE, l->l_cred)) != 0) {
934 		vrele(*vpp);
935 		return error;
936 	}
937 	if ((error = VOP_OPEN(*vpp, FREAD | FWRITE, l->l_cred)) != 0) {
938 		vrele(*vpp);
939 		return error;
940 	}
941 	mutex_enter((*vpp)->v_interlock);
942 	(*vpp)->v_writecount++;
943 	mutex_exit((*vpp)->v_interlock);
944 
945 	IFDEBUG(DKDB_VNODE, vprint("dk_lookup: vnode info", *vpp));
946 
947 	return 0;
948 out:
949 	VOP_UNLOCK(vp);
950 	(void) vn_close(vp, FREAD | FWRITE, l->l_cred);
951 	return error;
952 }
953 
954 MODULE(MODULE_CLASS_MISC, dk_subr, NULL);
955 
956 static int
957 dk_subr_modcmd(modcmd_t cmd, void *arg)
958 {
959 	switch (cmd) {
960 	case MODULE_CMD_INIT:
961 	case MODULE_CMD_FINI:
962 		return 0;
963 	case MODULE_CMD_STAT:
964 	case MODULE_CMD_AUTOUNLOAD:
965 	default:
966 		return ENOTTY;
967 	}
968 }
969