xref: /netbsd-src/sys/dev/dksubr.c (revision 63aea4bd5b445e491ff0389fe27ec78b3099dba3)
1 /* $NetBSD: dksubr.c,v 1.85 2015/12/21 12:33:12 mlelstv 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 1.85 2015/12/21 12:33:12 mlelstv 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 	if (bp != NULL) {
455 		bp->b_error = EIO;
456 		bp->b_resid = bp->b_bcount;
457 		biodone(bp);
458 	}
459 	bufq_drain(dksc->sc_bufq);
460 	mutex_exit(&dksc->sc_iolock);
461 }
462 
463 int
464 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len)
465 {
466 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
467 	unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize;
468 	struct buf tmp, *bp = &tmp;
469 	int error;
470 
471 	DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__,
472 	    dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len));
473 
474 	if (!(dksc->sc_flags & DKF_INITED)) {
475 		DPRINTF_FOLLOW(("%s: not inited\n", __func__));
476 		return ENXIO;
477 	}
478 
479 	if (secsize == 0 || (pos % secsize) != 0)
480 		return EINVAL;
481 
482 	/* enough data to please the bounds checking code */
483 	bp->b_dev = dev;
484 	bp->b_blkno = (daddr_t)(pos / secsize);
485 	bp->b_bcount = len;
486 	bp->b_flags = B_WRITE;
487 
488 	error = dk_translate(dksc, bp);
489 	if (error >= 0)
490 		return error;
491 
492 	error = dkd->d_discard(dksc->sc_dev,
493 		(off_t)bp->b_rawblkno * secsize,
494 		(off_t)bp->b_bcount);
495 
496 	return error;
497 }
498 
499 int
500 dk_size(struct dk_softc *dksc, dev_t dev)
501 {
502 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
503 	struct	disklabel *lp;
504 	int	is_open;
505 	int	part;
506 	int	size;
507 
508 	if ((dksc->sc_flags & DKF_INITED) == 0)
509 		return -1;
510 
511 	part = DISKPART(dev);
512 	is_open = dksc->sc_dkdev.dk_openmask & (1 << part);
513 
514 	if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp))
515 		return -1;
516 
517 	lp = dksc->sc_dkdev.dk_label;
518 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
519 		size = -1;
520 	else
521 		size = lp->d_partitions[part].p_size *
522 		    (lp->d_secsize / DEV_BSIZE);
523 
524 	if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp))
525 		return -1;
526 
527 	return size;
528 }
529 
530 int
531 dk_ioctl(struct dk_softc *dksc, dev_t dev,
532 	    u_long cmd, void *data, int flag, struct lwp *l)
533 {
534 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
535 	struct	disklabel *lp;
536 	struct	disk *dk = &dksc->sc_dkdev;
537 #ifdef __HAVE_OLD_DISKLABEL
538 	struct	disklabel newlabel;
539 #endif
540 	int	error;
541 
542 	DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__,
543 	    dksc->sc_xname, dksc, dev, cmd));
544 
545 	/* ensure that the pseudo disk is open for writes for these commands */
546 	switch (cmd) {
547 	case DIOCSDINFO:
548 	case DIOCWDINFO:
549 #ifdef __HAVE_OLD_DISKLABEL
550 	case ODIOCSDINFO:
551 	case ODIOCWDINFO:
552 #endif
553 	case DIOCKLABEL:
554 	case DIOCWLABEL:
555 	case DIOCAWEDGE:
556 	case DIOCDWEDGE:
557 	case DIOCSSTRATEGY:
558 		if ((flag & FWRITE) == 0)
559 			return EBADF;
560 	}
561 
562 	/* ensure that the pseudo-disk is initialized for these */
563 	switch (cmd) {
564 	case DIOCGDINFO:
565 	case DIOCSDINFO:
566 	case DIOCWDINFO:
567 	case DIOCGPARTINFO:
568 	case DIOCKLABEL:
569 	case DIOCWLABEL:
570 	case DIOCGDEFLABEL:
571 	case DIOCAWEDGE:
572 	case DIOCDWEDGE:
573 	case DIOCLWEDGES:
574 	case DIOCMWEDGES:
575 	case DIOCCACHESYNC:
576 #ifdef __HAVE_OLD_DISKLABEL
577 	case ODIOCGDINFO:
578 	case ODIOCSDINFO:
579 	case ODIOCWDINFO:
580 	case ODIOCGDEFLABEL:
581 #endif
582 		if ((dksc->sc_flags & DKF_INITED) == 0)
583 			return ENXIO;
584 	}
585 
586 	error = disk_ioctl(dk, dev, cmd, data, flag, l);
587 	if (error != EPASSTHROUGH)
588 		return error;
589 	else
590 		error = 0;
591 
592 	switch (cmd) {
593 	case DIOCWDINFO:
594 	case DIOCSDINFO:
595 #ifdef __HAVE_OLD_DISKLABEL
596 	case ODIOCWDINFO:
597 	case ODIOCSDINFO:
598 #endif
599 #ifdef __HAVE_OLD_DISKLABEL
600 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
601 			memset(&newlabel, 0, sizeof newlabel);
602 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
603 			lp = &newlabel;
604 		} else
605 #endif
606 		lp = (struct disklabel *)data;
607 
608 		mutex_enter(&dk->dk_openlock);
609 		dksc->sc_flags |= DKF_LABELLING;
610 
611 		error = setdisklabel(dksc->sc_dkdev.dk_label,
612 		    lp, 0, dksc->sc_dkdev.dk_cpulabel);
613 		if (error == 0) {
614 			if (cmd == DIOCWDINFO
615 #ifdef __HAVE_OLD_DISKLABEL
616 			    || cmd == ODIOCWDINFO
617 #endif
618 			   )
619 				error = writedisklabel(DKLABELDEV(dev),
620 				    dkd->d_strategy, dksc->sc_dkdev.dk_label,
621 				    dksc->sc_dkdev.dk_cpulabel);
622 		}
623 
624 		dksc->sc_flags &= ~DKF_LABELLING;
625 		mutex_exit(&dk->dk_openlock);
626 		break;
627 
628 	case DIOCKLABEL:
629 		if (*(int *)data != 0)
630 			dksc->sc_flags |= DKF_KLABEL;
631 		else
632 			dksc->sc_flags &= ~DKF_KLABEL;
633 		break;
634 
635 	case DIOCWLABEL:
636 		if (*(int *)data != 0)
637 			dksc->sc_flags |= DKF_WLABEL;
638 		else
639 			dksc->sc_flags &= ~DKF_WLABEL;
640 		break;
641 
642 	case DIOCGDEFLABEL:
643 		dk_getdefaultlabel(dksc, (struct disklabel *)data);
644 		break;
645 
646 #ifdef __HAVE_OLD_DISKLABEL
647 	case ODIOCGDEFLABEL:
648 		dk_getdefaultlabel(dksc, &newlabel);
649 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
650 			return ENOTTY;
651 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
652 		break;
653 #endif
654 
655 	case DIOCGSTRATEGY:
656 	    {
657 		struct disk_strategy *dks = (void *)data;
658 
659 		mutex_enter(&dksc->sc_iolock);
660 		strlcpy(dks->dks_name, bufq_getstrategyname(dksc->sc_bufq),
661 		    sizeof(dks->dks_name));
662 		mutex_exit(&dksc->sc_iolock);
663 		dks->dks_paramlen = 0;
664 
665 		return 0;
666 	    }
667 
668 	case DIOCSSTRATEGY:
669 	    {
670 		struct disk_strategy *dks = (void *)data;
671 		struct bufq_state *new;
672 		struct bufq_state *old;
673 
674 		if (dks->dks_param != NULL) {
675 			return EINVAL;
676 		}
677 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
678 		error = bufq_alloc(&new, dks->dks_name,
679 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
680 		if (error) {
681 			return error;
682 		}
683 		mutex_enter(&dksc->sc_iolock);
684 		old = dksc->sc_bufq;
685 		bufq_move(new, old);
686 		dksc->sc_bufq = new;
687 		mutex_exit(&dksc->sc_iolock);
688 		bufq_free(old);
689 
690 		return 0;
691 	    }
692 
693 	default:
694 		error = ENOTTY;
695 	}
696 
697 	return error;
698 }
699 
700 /*
701  * dk_dump dumps all of physical memory into the partition specified.
702  * This requires substantially more framework than {s,w}ddump, and hence
703  * is probably much more fragile.
704  *
705  */
706 
707 #define DKFF_READYFORDUMP(x)	(((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP)
708 static volatile int	dk_dumping = 0;
709 
710 /* ARGSUSED */
711 int
712 dk_dump(struct dk_softc *dksc, dev_t dev,
713     daddr_t blkno, void *vav, size_t size)
714 {
715 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
716 	char *va = vav;
717 	struct disklabel *lp;
718 	struct partition *p;
719 	int part, towrt, nsects, sectoff, maxblkcnt, nblk;
720 	int maxxfer, rv = 0;
721 
722 	/*
723 	 * ensure that we consider this device to be safe for dumping,
724 	 * and that the device is configured.
725 	 */
726 	if (!DKFF_READYFORDUMP(dksc->sc_flags)) {
727 		DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__,
728 		    dksc->sc_flags));
729 		return ENXIO;
730 	}
731 
732 	/* ensure that we are not already dumping */
733 	if (dk_dumping)
734 		return EFAULT;
735 	dk_dumping = 1;
736 
737 	if (dkd->d_dumpblocks == NULL) {
738 		DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__));
739 		return ENXIO;
740 	}
741 
742 	/* device specific max transfer size */
743 	maxxfer = MAXPHYS;
744 	if (dkd->d_iosize != NULL)
745 		(*dkd->d_iosize)(dksc->sc_dev, &maxxfer);
746 
747 	/* Convert to disk sectors.  Request must be a multiple of size. */
748 	part = DISKPART(dev);
749 	lp = dksc->sc_dkdev.dk_label;
750 	if ((size % lp->d_secsize) != 0) {
751 		DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size));
752 		return EFAULT;
753 	}
754 	towrt = size / lp->d_secsize;
755 	blkno = dbtob(blkno) / lp->d_secsize;   /* blkno in secsize units */
756 
757 	p = &lp->d_partitions[part];
758 	if (p->p_fstype != FS_SWAP) {
759 		DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__,
760 		    p->p_fstype));
761 		return ENXIO;
762 	}
763 	nsects = p->p_size;
764 	sectoff = p->p_offset;
765 
766 	/* Check transfer bounds against partition size. */
767 	if ((blkno < 0) || ((blkno + towrt) > nsects)) {
768 		DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, "
769 		    "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects));
770 		return EINVAL;
771 	}
772 
773 	/* Offset block number to start of partition. */
774 	blkno += sectoff;
775 
776 	/* Start dumping and return when done. */
777 	maxblkcnt = howmany(maxxfer, lp->d_secsize);
778 	while (towrt > 0) {
779 		nblk = min(maxblkcnt, towrt);
780 
781 		if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk))
782 		    != 0) {
783 			DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__,
784 			    rv));
785 			return rv;
786 		}
787 
788 		towrt -= nblk;
789 		blkno += nblk;
790 		va += nblk * lp->d_secsize;
791 	}
792 
793 	dk_dumping = 0;
794 
795 	return 0;
796 }
797 
798 /* ARGSUSED */
799 void
800 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp)
801 {
802 	struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
803 
804 	memset(lp, 0, sizeof(*lp));
805 
806 	if (dg->dg_secperunit > UINT32_MAX)
807 		lp->d_secperunit = UINT32_MAX;
808 	else
809 		lp->d_secperunit = dg->dg_secperunit;
810 	lp->d_secsize = dg->dg_secsize;
811 	lp->d_nsectors = dg->dg_nsectors;
812 	lp->d_ntracks = dg->dg_ntracks;
813 	lp->d_ncylinders = dg->dg_ncylinders;
814 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
815 
816 	strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename));
817 	lp->d_type = dksc->sc_dtype;
818 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
819 	lp->d_rpm = 3600;
820 	lp->d_interleave = 1;
821 	lp->d_flags = 0;
822 
823 	lp->d_partitions[RAW_PART].p_offset = 0;
824 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
825 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
826 	lp->d_npartitions = RAW_PART + 1;
827 
828 	lp->d_magic = DISKMAGIC;
829 	lp->d_magic2 = DISKMAGIC;
830 	lp->d_checksum = dkcksum(dksc->sc_dkdev.dk_label);
831 }
832 
833 /* ARGSUSED */
834 void
835 dk_getdisklabel(struct dk_softc *dksc, dev_t dev)
836 {
837 	const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
838 	struct	 disklabel *lp = dksc->sc_dkdev.dk_label;
839 	struct	 cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel;
840 	struct   disk_geom *dg = &dksc->sc_dkdev.dk_geom;
841 	struct	 partition *pp;
842 	int	 i;
843 	const char	*errstring;
844 
845 	memset(clp, 0x0, sizeof(*clp));
846 	dk_getdefaultlabel(dksc, lp);
847 	errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy,
848 	    dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel);
849 	if (errstring) {
850 		dk_makedisklabel(dksc);
851 		if (dksc->sc_flags & DKF_WARNLABEL)
852 			printf("%s: %s\n", dksc->sc_xname, errstring);
853 		return;
854 	}
855 
856 	if ((dksc->sc_flags & DKF_LABELSANITY) == 0)
857 		return;
858 
859 	/* Sanity check */
860 	if (lp->d_secperunit < UINT32_MAX ?
861 		lp->d_secperunit != dg->dg_secperunit :
862 		lp->d_secperunit > dg->dg_secperunit)
863 		printf("WARNING: %s: total sector size in disklabel (%ju) "
864 		    "!= the size of %s (%ju)\n", dksc->sc_xname,
865 		    (uintmax_t)lp->d_secperunit, dksc->sc_xname,
866 		    (uintmax_t)dg->dg_secperunit);
867 
868 	for (i=0; i < lp->d_npartitions; i++) {
869 		pp = &lp->d_partitions[i];
870 		if (pp->p_offset + pp->p_size > dg->dg_secperunit)
871 			printf("WARNING: %s: end of partition `%c' exceeds "
872 			    "the size of %s (%ju)\n", dksc->sc_xname,
873 			    'a' + i, dksc->sc_xname,
874 			    (uintmax_t)dg->dg_secperunit);
875 	}
876 }
877 
878 /* ARGSUSED */
879 static void
880 dk_makedisklabel(struct dk_softc *dksc)
881 {
882 	struct	disklabel *lp = dksc->sc_dkdev.dk_label;
883 
884 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
885 	strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname));
886 	lp->d_checksum = dkcksum(lp);
887 }
888 
889 /* This function is taken from ccd.c:1.76  --rcd */
890 
891 /*
892  * XXX this function looks too generic for dksubr.c, shouldn't we
893  *     put it somewhere better?
894  */
895 
896 /*
897  * Lookup the provided name in the filesystem.  If the file exists,
898  * is a valid block device, and isn't being used by anyone else,
899  * set *vpp to the file's vnode.
900  */
901 int
902 dk_lookup(struct pathbuf *pb, struct lwp *l, struct vnode **vpp)
903 {
904 	struct nameidata nd;
905 	struct vnode *vp;
906 	int     error;
907 
908 	if (l == NULL)
909 		return ESRCH;	/* Is ESRCH the best choice? */
910 
911 	NDINIT(&nd, LOOKUP, FOLLOW, pb);
912 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
913 		DPRINTF((DKDB_FOLLOW|DKDB_INIT),
914 		    ("%s: vn_open error = %d\n", __func__, error));
915 		return error;
916 	}
917 
918 	vp = nd.ni_vp;
919 	if (vp->v_type != VBLK) {
920 		error = ENOTBLK;
921 		goto out;
922 	}
923 
924 	/* Reopen as anonymous vnode to protect against forced unmount. */
925 	if ((error = bdevvp(vp->v_rdev, vpp)) != 0)
926 		goto out;
927 	VOP_UNLOCK(vp);
928 	if ((error = vn_close(vp, FREAD | FWRITE, l->l_cred)) != 0) {
929 		vrele(*vpp);
930 		return error;
931 	}
932 	if ((error = VOP_OPEN(*vpp, FREAD | FWRITE, l->l_cred)) != 0) {
933 		vrele(*vpp);
934 		return error;
935 	}
936 	mutex_enter((*vpp)->v_interlock);
937 	(*vpp)->v_writecount++;
938 	mutex_exit((*vpp)->v_interlock);
939 
940 	IFDEBUG(DKDB_VNODE, vprint("dk_lookup: vnode info", *vpp));
941 
942 	return 0;
943 out:
944 	VOP_UNLOCK(vp);
945 	(void) vn_close(vp, FREAD | FWRITE, l->l_cred);
946 	return error;
947 }
948 
949 MODULE(MODULE_CLASS_MISC, dk_subr, NULL);
950 
951 static int
952 dk_subr_modcmd(modcmd_t cmd, void *arg)
953 {
954 	switch (cmd) {
955 	case MODULE_CMD_INIT:
956 	case MODULE_CMD_FINI:
957 		return 0;
958 	case MODULE_CMD_STAT:
959 	case MODULE_CMD_AUTOUNLOAD:
960 	default:
961 		return ENOTTY;
962 	}
963 }
964