xref: /netbsd-src/sys/dev/ld.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11)
1 /*	$NetBSD: ld.c,v 1.29 2004/08/30 00:34:41 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran and Charles M. Hannum.
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Disk driver for use by RAID controllers.
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.29 2004/08/30 00:34:41 thorpej Exp $");
45 
46 #include "rnd.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/device.h>
52 #include <sys/queue.h>
53 #include <sys/proc.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/disklabel.h>
57 #include <sys/disk.h>
58 #include <sys/dkio.h>
59 #include <sys/stat.h>
60 #include <sys/lock.h>
61 #include <sys/conf.h>
62 #include <sys/fcntl.h>
63 #include <sys/vnode.h>
64 #include <sys/syslog.h>
65 #if NRND > 0
66 #include <sys/rnd.h>
67 #endif
68 
69 #include <dev/ldvar.h>
70 
71 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
72 static void	ldgetdisklabel(struct ld_softc *);
73 static int	ldlock(struct ld_softc *);
74 static void	ldminphys(struct buf *bp);
75 static void	ldshutdown(void *);
76 static void	ldstart(struct ld_softc *);
77 static void	ldunlock(struct ld_softc *);
78 
79 extern struct	cfdriver ld_cd;
80 
81 static dev_type_open(ldopen);
82 static dev_type_close(ldclose);
83 static dev_type_read(ldread);
84 static dev_type_write(ldwrite);
85 static dev_type_ioctl(ldioctl);
86 static dev_type_strategy(ldstrategy);
87 static dev_type_dump(lddump);
88 static dev_type_size(ldsize);
89 
90 const struct bdevsw ld_bdevsw = {
91 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
92 };
93 
94 const struct cdevsw ld_cdevsw = {
95 	ldopen, ldclose, ldread, ldwrite, ldioctl,
96 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
97 };
98 
99 static struct	dkdriver lddkdriver = { ldstrategy };
100 static void	*ld_sdh;
101 
102 void
103 ldattach(struct ld_softc *sc)
104 {
105 	char buf[9];
106 
107 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
108 		printf("%s: disabled\n", sc->sc_dv.dv_xname);
109 		return;
110 	}
111 
112 	/* Initialise and attach the disk structure. */
113 	sc->sc_dk.dk_driver = &lddkdriver;
114 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
115 	disk_attach(&sc->sc_dk);
116 
117 	if (sc->sc_maxxfer > MAXPHYS)
118 		sc->sc_maxxfer = MAXPHYS;
119 
120 	/* Build synthetic geometry if necessary. */
121 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
122 	    sc->sc_ncylinders == 0) {
123 		uint64_t ncyl;
124 
125 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
126 			sc->sc_nheads = 16;
127 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
128 			sc->sc_nheads = 32;
129 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
130 			sc->sc_nheads = 64;
131 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
132 			sc->sc_nheads = 128;
133 		else
134 			sc->sc_nheads = 255;
135 
136 		sc->sc_nsectors = 63;
137 		sc->sc_ncylinders = INT_MAX;
138 		ncyl = sc->sc_secperunit /
139 		    (sc->sc_nheads * sc->sc_nsectors);
140 		if (ncyl < INT_MAX)
141 			sc->sc_ncylinders = (int)ncyl;
142 	}
143 
144 	format_bytes(buf, sizeof(buf), sc->sc_secperunit *
145 	    sc->sc_secsize);
146 	printf("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
147 	    sc->sc_dv.dv_xname, buf, sc->sc_ncylinders, sc->sc_nheads,
148 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
149 
150 #if NRND > 0
151 	/* Attach the device into the rnd source list. */
152 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
153 	    RND_TYPE_DISK, 0);
154 #endif
155 
156 	/* Set the `shutdownhook'. */
157 	if (ld_sdh == NULL)
158 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
159 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT()|BUFQ_SORT_RAWBLOCK);
160 }
161 
162 int
163 ldadjqparam(struct ld_softc *sc, int max)
164 {
165 	int s;
166 
167 	s = splbio();
168 	sc->sc_maxqueuecnt = max;
169 	splx(s);
170 
171 	return (0);
172 }
173 
174 int
175 ldbegindetach(struct ld_softc *sc, int flags)
176 {
177 	int s, rv = 0;
178 
179 	if ((sc->sc_flags & LDF_ENABLED) == 0)
180 		return (0);
181 
182 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
183 		return (EBUSY);
184 
185 	s = splbio();
186 	sc->sc_maxqueuecnt = 0;
187 	sc->sc_flags |= LDF_DETACH;
188 	while (sc->sc_queuecnt > 0) {
189 		sc->sc_flags |= LDF_DRAIN;
190 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
191 		if (rv)
192 			break;
193 	}
194 	splx(s);
195 
196 	return (rv);
197 }
198 
199 void
200 ldenddetach(struct ld_softc *sc)
201 {
202 	struct buf *bp;
203 	int s, bmaj, cmaj, i, mn;
204 
205 	if ((sc->sc_flags & LDF_ENABLED) == 0)
206 		return;
207 
208 	/* Wait for commands queued with the hardware to complete. */
209 	if (sc->sc_queuecnt != 0)
210 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
211 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
212 
213 	/* Locate the major numbers. */
214 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
215 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
216 
217 	/* Kill off any queued buffers. */
218 	s = splbio();
219 	while ((bp = BUFQ_GET(&sc->sc_bufq)) != NULL) {
220 		bp->b_error = EIO;
221 		bp->b_flags |= B_ERROR;
222 		bp->b_resid = bp->b_bcount;
223 		biodone(bp);
224 	}
225 	bufq_free(&sc->sc_bufq);
226 	splx(s);
227 
228 	/* Nuke the vnodes for any open instances. */
229 	for (i = 0; i < MAXPARTITIONS; i++) {
230 		mn = DISKMINOR(sc->sc_dv.dv_unit, i);
231 		vdevgone(bmaj, mn, mn, VBLK);
232 		vdevgone(cmaj, mn, mn, VCHR);
233 	}
234 
235 	/* Detach from the disk list. */
236 	disk_detach(&sc->sc_dk);
237 
238 #if NRND > 0
239 	/* Unhook the entropy source. */
240 	rnd_detach_source(&sc->sc_rnd_source);
241 #endif
242 
243 	/*
244 	 * XXX We can't really flush the cache here, beceause the
245 	 * XXX device may already be non-existent from the controller's
246 	 * XXX perspective.
247 	 */
248 #if 0
249 	/* Flush the device's cache. */
250 	if (sc->sc_flush != NULL)
251 		if ((*sc->sc_flush)(sc) != 0)
252 			printf("%s: unable to flush cache\n",
253 			    sc->sc_dv.dv_xname);
254 #endif
255 }
256 
257 /* ARGSUSED */
258 static void
259 ldshutdown(void *cookie)
260 {
261 	struct ld_softc *sc;
262 	int i;
263 
264 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
265 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
266 			continue;
267 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
268 			printf("%s: unable to flush cache\n",
269 			    sc->sc_dv.dv_xname);
270 	}
271 }
272 
273 /* ARGSUSED */
274 static int
275 ldopen(dev_t dev, int flags, int fmt, struct proc *p)
276 {
277 	struct ld_softc *sc;
278 	int unit, part;
279 
280 	unit = DISKUNIT(dev);
281 	if ((sc = device_lookup(&ld_cd, unit))== NULL)
282 		return (ENXIO);
283 	if ((sc->sc_flags & LDF_ENABLED) == 0)
284 		return (ENODEV);
285 	part = DISKPART(dev);
286 	ldlock(sc);
287 
288 	if (sc->sc_dk.dk_openmask == 0) {
289 		/* Load the partition info if not already loaded. */
290 		if ((sc->sc_flags & LDF_VLABEL) == 0)
291 			ldgetdisklabel(sc);
292 	}
293 
294 	/* Check that the partition exists. */
295 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
296 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
297 	     	ldunlock(sc);
298 		return (ENXIO);
299 	}
300 
301 	/* Ensure only one open at a time. */
302 	switch (fmt) {
303 	case S_IFCHR:
304 		sc->sc_dk.dk_copenmask |= (1 << part);
305 		break;
306 	case S_IFBLK:
307 		sc->sc_dk.dk_bopenmask |= (1 << part);
308 		break;
309 	}
310 	sc->sc_dk.dk_openmask =
311 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
312 
313 	ldunlock(sc);
314 	return (0);
315 }
316 
317 /* ARGSUSED */
318 static int
319 ldclose(dev_t dev, int flags, int fmt, struct proc *p)
320 {
321 	struct ld_softc *sc;
322 	int part, unit;
323 
324 	unit = DISKUNIT(dev);
325 	part = DISKPART(dev);
326 	sc = device_lookup(&ld_cd, unit);
327 	ldlock(sc);
328 
329 	switch (fmt) {
330 	case S_IFCHR:
331 		sc->sc_dk.dk_copenmask &= ~(1 << part);
332 		break;
333 	case S_IFBLK:
334 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
335 		break;
336 	}
337 	sc->sc_dk.dk_openmask =
338 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
339 
340 	if (sc->sc_dk.dk_openmask == 0) {
341 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
342 			printf("%s: unable to flush cache\n",
343 			    sc->sc_dv.dv_xname);
344 		if ((sc->sc_flags & LDF_KLABEL) == 0)
345 			sc->sc_flags &= ~LDF_VLABEL;
346 	}
347 
348 	ldunlock(sc);
349 	return (0);
350 }
351 
352 /* ARGSUSED */
353 static int
354 ldread(dev_t dev, struct uio *uio, int ioflag)
355 {
356 
357 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
358 }
359 
360 /* ARGSUSED */
361 static int
362 ldwrite(dev_t dev, struct uio *uio, int ioflag)
363 {
364 
365 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
366 }
367 
368 /* ARGSUSED */
369 static int
370 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct proc *p)
371 {
372 	struct ld_softc *sc;
373 	int part, unit, error;
374 #ifdef __HAVE_OLD_DISKLABEL
375 	struct disklabel newlabel;
376 #endif
377 	struct disklabel *lp;
378 
379 	unit = DISKUNIT(dev);
380 	part = DISKPART(dev);
381 	sc = device_lookup(&ld_cd, unit);
382 	error = 0;
383 
384 	switch (cmd) {
385 	case DIOCGDINFO:
386 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
387 		return (0);
388 
389 #ifdef __HAVE_OLD_DISKLABEL
390 	case ODIOCGDINFO:
391 		newlabel = *(sc->sc_dk.dk_label);
392 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
393 			return ENOTTY;
394 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
395 		return (0);
396 #endif
397 
398 	case DIOCGPART:
399 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
400 		((struct partinfo *)addr)->part =
401 		    &sc->sc_dk.dk_label->d_partitions[part];
402 		break;
403 
404 	case DIOCWDINFO:
405 	case DIOCSDINFO:
406 #ifdef __HAVE_OLD_DISKLABEL
407 	case ODIOCWDINFO:
408 	case ODIOCSDINFO:
409 
410 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
411 			memset(&newlabel, 0, sizeof newlabel);
412 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
413 			lp = &newlabel;
414 		} else
415 #endif
416 		lp = (struct disklabel *)addr;
417 
418 		if ((flag & FWRITE) == 0)
419 			return (EBADF);
420 
421 		if ((error = ldlock(sc)) != 0)
422 			return (error);
423 		sc->sc_flags |= LDF_LABELLING;
424 
425 		error = setdisklabel(sc->sc_dk.dk_label,
426 		    lp, /*sc->sc_dk.dk_openmask : */0,
427 		    sc->sc_dk.dk_cpulabel);
428 		if (error == 0 && (cmd == DIOCWDINFO
429 #ifdef __HAVE_OLD_DISKLABEL
430 		    || cmd == ODIOCWDINFO
431 #endif
432 		    ))
433 			error = writedisklabel(
434 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
435 			    ldstrategy, sc->sc_dk.dk_label,
436 			    sc->sc_dk.dk_cpulabel);
437 
438 		sc->sc_flags &= ~LDF_LABELLING;
439 		ldunlock(sc);
440 		break;
441 
442 	case DIOCKLABEL:
443 		if ((flag & FWRITE) == 0)
444 			return (EBADF);
445 		if (*(int *)addr)
446 			sc->sc_flags |= LDF_KLABEL;
447 		else
448 			sc->sc_flags &= ~LDF_KLABEL;
449 		break;
450 
451 	case DIOCWLABEL:
452 		if ((flag & FWRITE) == 0)
453 			return (EBADF);
454 		if (*(int *)addr)
455 			sc->sc_flags |= LDF_WLABEL;
456 		else
457 			sc->sc_flags &= ~LDF_WLABEL;
458 		break;
459 
460 	case DIOCGDEFLABEL:
461 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
462 		break;
463 
464 #ifdef __HAVE_OLD_DISKLABEL
465 	case ODIOCGDEFLABEL:
466 		ldgetdefaultlabel(sc, &newlabel);
467 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
468 			return ENOTTY;
469 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
470 		break;
471 #endif
472 
473 	default:
474 		error = ENOTTY;
475 		break;
476 	}
477 
478 	return (error);
479 }
480 
481 static void
482 ldstrategy(struct buf *bp)
483 {
484 	struct ld_softc *sc;
485 	struct disklabel *lp;
486 	daddr_t blkno;
487 	int s, part;
488 
489 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
490 	part = DISKPART(bp->b_dev);
491 
492 	if ((sc->sc_flags & LDF_DETACH) != 0) {
493 		bp->b_error = EIO;
494 		goto bad;
495 	}
496 
497 	lp = sc->sc_dk.dk_label;
498 
499 	/*
500 	 * The transfer must be a whole number of blocks and the offset must
501 	 * not be negative.
502 	 */
503 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
504 		bp->b_error = EINVAL;
505 		goto bad;
506 	}
507 
508 	/* If it's a null transfer, return immediately. */
509 	if (bp->b_bcount == 0)
510 		goto done;
511 
512 	/*
513 	 * Do bounds checking and adjust the transfer.  If error, process.
514 	 * If past the end of partition, just return.
515 	 */
516 	if (part != RAW_PART &&
517 	    bounds_check_with_label(&sc->sc_dk, bp,
518 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
519 		goto done;
520 	}
521 
522 	/*
523 	 * Convert the block number to absolute and put it in terms
524 	 * of the device's logical block size.
525 	 */
526 	if (lp->d_secsize == DEV_BSIZE)
527 		blkno = bp->b_blkno;
528 	else if (lp->d_secsize > DEV_BSIZE)
529 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
530 	else
531 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
532 
533 	if (part != RAW_PART)
534 		blkno += lp->d_partitions[part].p_offset;
535 
536 	bp->b_rawblkno = blkno;
537 
538 	s = splbio();
539 	BUFQ_PUT(&sc->sc_bufq, bp);
540 	ldstart(sc);
541 	splx(s);
542 	return;
543 
544  bad:
545 	bp->b_flags |= B_ERROR;
546  done:
547 	bp->b_resid = bp->b_bcount;
548 	biodone(bp);
549 }
550 
551 static void
552 ldstart(struct ld_softc *sc)
553 {
554 	struct buf *bp;
555 	int error;
556 
557 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
558 		/* See if there is work to do. */
559 		if ((bp = BUFQ_PEEK(&sc->sc_bufq)) == NULL)
560 			break;
561 
562 		disk_busy(&sc->sc_dk);
563 		sc->sc_queuecnt++;
564 
565 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
566 			/*
567 			 * The back-end is running the job; remove it from
568 			 * the queue.
569 			 */
570 			(void) BUFQ_GET(&sc->sc_bufq);
571 		} else  {
572 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
573 			sc->sc_queuecnt--;
574 			if (error == EAGAIN) {
575 				/*
576 				 * Temporary resource shortage in the
577 				 * back-end; just defer the job until
578 				 * later.
579 				 *
580 				 * XXX We might consider a watchdog timer
581 				 * XXX to make sure we are kicked into action.
582 				 */
583 				break;
584 			} else {
585 				(void) BUFQ_GET(&sc->sc_bufq);
586 				bp->b_error = error;
587 				bp->b_flags |= B_ERROR;
588 				bp->b_resid = bp->b_bcount;
589 				biodone(bp);
590 			}
591 		}
592 	}
593 }
594 
595 void
596 lddone(struct ld_softc *sc, struct buf *bp)
597 {
598 
599 	if ((bp->b_flags & B_ERROR) != 0) {
600 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
601 		printf("\n");
602 	}
603 
604 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
605 	    (bp->b_flags & B_READ));
606 #if NRND > 0
607 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
608 #endif
609 	biodone(bp);
610 
611 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
612 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
613 			sc->sc_flags &= ~LDF_DRAIN;
614 			wakeup(&sc->sc_queuecnt);
615 		}
616 		ldstart(sc);
617 	}
618 }
619 
620 static int
621 ldsize(dev_t dev)
622 {
623 	struct ld_softc *sc;
624 	int part, unit, omask, size;
625 
626 	unit = DISKUNIT(dev);
627 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
628 		return (ENODEV);
629 	if ((sc->sc_flags & LDF_ENABLED) == 0)
630 		return (ENODEV);
631 	part = DISKPART(dev);
632 
633 	omask = sc->sc_dk.dk_openmask & (1 << part);
634 
635 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
636 		return (-1);
637 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
638 		size = -1;
639 	else
640 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
641 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
642 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
643 		return (-1);
644 
645 	return (size);
646 }
647 
648 /*
649  * Load the label information from the specified device.
650  */
651 static void
652 ldgetdisklabel(struct ld_softc *sc)
653 {
654 	const char *errstring;
655 
656 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
657 
658 	/* Call the generic disklabel extraction routine. */
659 	errstring = readdisklabel(MAKEDISKDEV(0, sc->sc_dv.dv_unit, RAW_PART),
660 	    ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
661 	if (errstring != NULL)
662 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
663 
664 	/* In-core label now valid. */
665 	sc->sc_flags |= LDF_VLABEL;
666 }
667 
668 /*
669  * Construct a ficticious label.
670  */
671 static void
672 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
673 {
674 
675 	memset(lp, 0, sizeof(struct disklabel));
676 
677 	lp->d_secsize = sc->sc_secsize;
678 	lp->d_ntracks = sc->sc_nheads;
679 	lp->d_nsectors = sc->sc_nsectors;
680 	lp->d_ncylinders = sc->sc_ncylinders;
681 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
682 	lp->d_type = DTYPE_LD;
683 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
684 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
685 	lp->d_secperunit = sc->sc_secperunit;
686 	lp->d_rpm = 7200;
687 	lp->d_interleave = 1;
688 	lp->d_flags = 0;
689 
690 	lp->d_partitions[RAW_PART].p_offset = 0;
691 	lp->d_partitions[RAW_PART].p_size =
692 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
693 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
694 	lp->d_npartitions = RAW_PART + 1;
695 
696 	lp->d_magic = DISKMAGIC;
697 	lp->d_magic2 = DISKMAGIC;
698 	lp->d_checksum = dkcksum(lp);
699 }
700 
701 /*
702  * Wait interruptibly for an exclusive lock.
703  *
704  * XXX Several drivers do this; it should be abstracted and made MP-safe.
705  */
706 static int
707 ldlock(struct ld_softc *sc)
708 {
709 	int error;
710 
711 	while ((sc->sc_flags & LDF_LKHELD) != 0) {
712 		sc->sc_flags |= LDF_LKWANTED;
713 		if ((error = tsleep(sc, PRIBIO | PCATCH, "ldlck", 0)) != 0)
714 			return (error);
715 	}
716 	sc->sc_flags |= LDF_LKHELD;
717 	return (0);
718 }
719 
720 /*
721  * Unlock and wake up any waiters.
722  */
723 static void
724 ldunlock(struct ld_softc *sc)
725 {
726 
727 	sc->sc_flags &= ~LDF_LKHELD;
728 	if ((sc->sc_flags & LDF_LKWANTED) != 0) {
729 		sc->sc_flags &= ~LDF_LKWANTED;
730 		wakeup(sc);
731 	}
732 }
733 
734 /*
735  * Take a dump.
736  */
737 static int
738 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
739 {
740 	struct ld_softc *sc;
741 	struct disklabel *lp;
742 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
743 	static int dumping;
744 
745 	unit = DISKUNIT(dev);
746 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
747 		return (ENXIO);
748 	if ((sc->sc_flags & LDF_ENABLED) == 0)
749 		return (ENODEV);
750 	if (sc->sc_dump == NULL)
751 		return (ENXIO);
752 
753 	/* Check if recursive dump; if so, punt. */
754 	if (dumping)
755 		return (EFAULT);
756 	dumping = 1;
757 
758 	/* Convert to disk sectors.  Request must be a multiple of size. */
759 	part = DISKPART(dev);
760 	lp = sc->sc_dk.dk_label;
761 	if ((size % lp->d_secsize) != 0)
762 		return (EFAULT);
763 	towrt = size / lp->d_secsize;
764 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
765 
766 	nsects = lp->d_partitions[part].p_size;
767 	sectoff = lp->d_partitions[part].p_offset;
768 
769 	/* Check transfer bounds against partition size. */
770 	if ((blkno < 0) || ((blkno + towrt) > nsects))
771 		return (EINVAL);
772 
773 	/* Offset block number to start of partition. */
774 	blkno += sectoff;
775 
776 	/* Start dumping and return when done. */
777 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
778 	while (towrt > 0) {
779 		nblk = min(maxblkcnt, towrt);
780 
781 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
782 			return (rv);
783 
784 		towrt -= nblk;
785 		blkno += nblk;
786 		va += nblk * sc->sc_secsize;
787 	}
788 
789 	dumping = 0;
790 	return (0);
791 }
792 
793 /*
794  * Adjust the size of a transfer.
795  */
796 static void
797 ldminphys(struct buf *bp)
798 {
799 	struct ld_softc *sc;
800 
801 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
802 
803 	if (bp->b_bcount > sc->sc_maxxfer)
804 		bp->b_bcount = sc->sc_maxxfer;
805 	minphys(bp);
806 }
807