xref: /netbsd-src/sys/dev/ld.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: ld.c,v 1.68 2010/09/20 06:54:06 kiyohara 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  *
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 /*
33  * Disk driver for use by RAID controllers.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.68 2010/09/20 06:54:06 kiyohara Exp $");
38 
39 #include "rnd.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/device.h>
45 #include <sys/queue.h>
46 #include <sys/proc.h>
47 #include <sys/buf.h>
48 #include <sys/bufq.h>
49 #include <sys/endian.h>
50 #include <sys/disklabel.h>
51 #include <sys/disk.h>
52 #include <sys/dkio.h>
53 #include <sys/stat.h>
54 #include <sys/conf.h>
55 #include <sys/fcntl.h>
56 #include <sys/vnode.h>
57 #include <sys/syslog.h>
58 #include <sys/mutex.h>
59 #if NRND > 0
60 #include <sys/rnd.h>
61 #endif
62 
63 #include <dev/ldvar.h>
64 
65 #include <prop/proplib.h>
66 
67 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
68 static void	ldgetdisklabel(struct ld_softc *);
69 static void	ldminphys(struct buf *bp);
70 static bool	ld_suspend(device_t, const pmf_qual_t *);
71 static bool	ld_shutdown(device_t, int);
72 static void	ldstart(struct ld_softc *, struct buf *);
73 static void	ld_set_properties(struct ld_softc *);
74 static void	ld_config_interrupts (device_t);
75 static int	ldlastclose(device_t);
76 
77 extern struct	cfdriver ld_cd;
78 
79 static dev_type_open(ldopen);
80 static dev_type_close(ldclose);
81 static dev_type_read(ldread);
82 static dev_type_write(ldwrite);
83 static dev_type_ioctl(ldioctl);
84 static dev_type_strategy(ldstrategy);
85 static dev_type_dump(lddump);
86 static dev_type_size(ldsize);
87 
88 const struct bdevsw ld_bdevsw = {
89 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
90 };
91 
92 const struct cdevsw ld_cdevsw = {
93 	ldopen, ldclose, ldread, ldwrite, ldioctl,
94 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
95 };
96 
97 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
98 
99 void
100 ldattach(struct ld_softc *sc)
101 {
102 	char tbuf[9];
103 
104 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
105 
106 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
107 		aprint_normal_dev(sc->sc_dv, "disabled\n");
108 		return;
109 	}
110 
111 	/* Initialise and attach the disk structure. */
112 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
113 	disk_attach(&sc->sc_dk);
114 
115 	if (sc->sc_maxxfer > MAXPHYS)
116 		sc->sc_maxxfer = MAXPHYS;
117 
118 	/* Build synthetic geometry if necessary. */
119 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
120 	    sc->sc_ncylinders == 0) {
121 		uint64_t ncyl;
122 
123 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
124 			sc->sc_nheads = 16;
125 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
126 			sc->sc_nheads = 32;
127 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
128 			sc->sc_nheads = 64;
129 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
130 			sc->sc_nheads = 128;
131 		else
132 			sc->sc_nheads = 255;
133 
134 		sc->sc_nsectors = 63;
135 		sc->sc_ncylinders = INT_MAX;
136 		ncyl = sc->sc_secperunit /
137 		    (sc->sc_nheads * sc->sc_nsectors);
138 		if (ncyl < INT_MAX)
139 			sc->sc_ncylinders = (int)ncyl;
140 	}
141 
142 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
143 	    sc->sc_secsize);
144 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
145 	    "%d bytes/sect x %"PRIu64" sectors\n",
146 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
147 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
148 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
149 
150 	ld_set_properties(sc);
151 
152 #if NRND > 0
153 	/* Attach the device into the rnd source list. */
154 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
155 	    RND_TYPE_DISK, 0);
156 #endif
157 
158 	/* Register with PMF */
159 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
160 		aprint_error_dev(sc->sc_dv,
161 		    "couldn't establish power handler\n");
162 
163 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
164 
165 	/* Discover wedges on this disk. */
166 	config_interrupts(sc->sc_dv, ld_config_interrupts);
167 }
168 
169 int
170 ldadjqparam(struct ld_softc *sc, int xmax)
171 {
172 	int s;
173 
174 	s = splbio();
175 	sc->sc_maxqueuecnt = xmax;
176 	splx(s);
177 
178 	return (0);
179 }
180 
181 int
182 ldbegindetach(struct ld_softc *sc, int flags)
183 {
184 	int s, rv = 0;
185 
186 	if ((sc->sc_flags & LDF_ENABLED) == 0)
187 		return (0);
188 
189 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
190 
191 	if (rv != 0)
192 		return rv;
193 
194 	s = splbio();
195 	sc->sc_maxqueuecnt = 0;
196 	sc->sc_flags |= LDF_DETACH;
197 	while (sc->sc_queuecnt > 0) {
198 		sc->sc_flags |= LDF_DRAIN;
199 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
200 		if (rv)
201 			break;
202 	}
203 	splx(s);
204 
205 	return (rv);
206 }
207 
208 void
209 ldenddetach(struct ld_softc *sc)
210 {
211 	int s, bmaj, cmaj, i, mn;
212 
213 	if ((sc->sc_flags & LDF_ENABLED) == 0)
214 		return;
215 
216 	/* Wait for commands queued with the hardware to complete. */
217 	if (sc->sc_queuecnt != 0)
218 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
219 			printf("%s: not drained\n", device_xname(sc->sc_dv));
220 
221 	/* Locate the major numbers. */
222 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
223 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
224 
225 	/* Kill off any queued buffers. */
226 	s = splbio();
227 	bufq_drain(sc->sc_bufq);
228 	splx(s);
229 
230 	bufq_free(sc->sc_bufq);
231 
232 	/* Nuke the vnodes for any open instances. */
233 	for (i = 0; i < MAXPARTITIONS; i++) {
234 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
235 		vdevgone(bmaj, mn, mn, VBLK);
236 		vdevgone(cmaj, mn, mn, VCHR);
237 	}
238 
239 	/* Delete all of our wedges. */
240 	dkwedge_delall(&sc->sc_dk);
241 
242 	/* Detach from the disk list. */
243 	disk_detach(&sc->sc_dk);
244 	disk_destroy(&sc->sc_dk);
245 
246 #if NRND > 0
247 	/* Unhook the entropy source. */
248 	rnd_detach_source(&sc->sc_rnd_source);
249 #endif
250 
251 	/* Deregister with PMF */
252 	pmf_device_deregister(sc->sc_dv);
253 
254 	/*
255 	 * XXX We can't really flush the cache here, beceause the
256 	 * XXX device may already be non-existent from the controller's
257 	 * XXX perspective.
258 	 */
259 #if 0
260 	/* Flush the device's cache. */
261 	if (sc->sc_flush != NULL)
262 		if ((*sc->sc_flush)(sc, 0) != 0)
263 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
264 #endif
265 	mutex_destroy(&sc->sc_mutex);
266 }
267 
268 /* ARGSUSED */
269 static bool
270 ld_suspend(device_t dev, const pmf_qual_t *qual)
271 {
272 	return ld_shutdown(dev, 0);
273 }
274 
275 /* ARGSUSED */
276 static bool
277 ld_shutdown(device_t dev, int flags)
278 {
279 	struct ld_softc *sc = device_private(dev);
280 
281 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
282 		printf("%s: unable to flush cache\n", device_xname(dev));
283 		return false;
284 	}
285 
286 	return true;
287 }
288 
289 /* ARGSUSED */
290 static int
291 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
292 {
293 	struct ld_softc *sc;
294 	int error, unit, part;
295 
296 	unit = DISKUNIT(dev);
297 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
298 		return (ENXIO);
299 	if ((sc->sc_flags & LDF_ENABLED) == 0)
300 		return (ENODEV);
301 	part = DISKPART(dev);
302 
303 	mutex_enter(&sc->sc_dk.dk_openlock);
304 
305 	if (sc->sc_dk.dk_openmask == 0) {
306 		/* Load the partition info if not already loaded. */
307 		if ((sc->sc_flags & LDF_VLABEL) == 0)
308 			ldgetdisklabel(sc);
309 	}
310 
311 	/* Check that the partition exists. */
312 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
313 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
314 		error = ENXIO;
315 		goto bad1;
316 	}
317 
318 	/* Ensure only one open at a time. */
319 	switch (fmt) {
320 	case S_IFCHR:
321 		sc->sc_dk.dk_copenmask |= (1 << part);
322 		break;
323 	case S_IFBLK:
324 		sc->sc_dk.dk_bopenmask |= (1 << part);
325 		break;
326 	}
327 	sc->sc_dk.dk_openmask =
328 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
329 
330 	error = 0;
331  bad1:
332 	mutex_exit(&sc->sc_dk.dk_openlock);
333 	return (error);
334 }
335 
336 static int
337 ldlastclose(device_t self)
338 {
339 	struct ld_softc *sc = device_private(self);
340 
341 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
342 		aprint_error_dev(self, "unable to flush cache\n");
343 	if ((sc->sc_flags & LDF_KLABEL) == 0)
344 		sc->sc_flags &= ~LDF_VLABEL;
345 
346 	return 0;
347 }
348 
349 /* ARGSUSED */
350 static int
351 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
352 {
353 	struct ld_softc *sc;
354 	int part, unit;
355 
356 	unit = DISKUNIT(dev);
357 	part = DISKPART(dev);
358 	sc = device_lookup_private(&ld_cd, unit);
359 
360 	mutex_enter(&sc->sc_dk.dk_openlock);
361 
362 	switch (fmt) {
363 	case S_IFCHR:
364 		sc->sc_dk.dk_copenmask &= ~(1 << part);
365 		break;
366 	case S_IFBLK:
367 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
368 		break;
369 	}
370 	sc->sc_dk.dk_openmask =
371 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
372 
373 	if (sc->sc_dk.dk_openmask == 0)
374 		ldlastclose(sc->sc_dv);
375 
376 	mutex_exit(&sc->sc_dk.dk_openlock);
377 	return (0);
378 }
379 
380 /* ARGSUSED */
381 static int
382 ldread(dev_t dev, struct uio *uio, int ioflag)
383 {
384 
385 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
386 }
387 
388 /* ARGSUSED */
389 static int
390 ldwrite(dev_t dev, struct uio *uio, int ioflag)
391 {
392 
393 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
394 }
395 
396 /* ARGSUSED */
397 static int
398 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
399 {
400 	struct ld_softc *sc;
401 	int part, unit, error;
402 #ifdef __HAVE_OLD_DISKLABEL
403 	struct disklabel newlabel;
404 #endif
405 	struct disklabel *lp;
406 
407 	unit = DISKUNIT(dev);
408 	part = DISKPART(dev);
409 	sc = device_lookup_private(&ld_cd, unit);
410 
411 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
412 	if (error != EPASSTHROUGH)
413 		return (error);
414 
415 	error = 0;
416 	switch (cmd) {
417 	case DIOCGDINFO:
418 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
419 		return (0);
420 
421 #ifdef __HAVE_OLD_DISKLABEL
422 	case ODIOCGDINFO:
423 		newlabel = *(sc->sc_dk.dk_label);
424 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
425 			return ENOTTY;
426 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
427 		return (0);
428 #endif
429 
430 	case DIOCGPART:
431 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
432 		((struct partinfo *)addr)->part =
433 		    &sc->sc_dk.dk_label->d_partitions[part];
434 		break;
435 
436 	case DIOCWDINFO:
437 	case DIOCSDINFO:
438 #ifdef __HAVE_OLD_DISKLABEL
439 	case ODIOCWDINFO:
440 	case ODIOCSDINFO:
441 
442 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
443 			memset(&newlabel, 0, sizeof newlabel);
444 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
445 			lp = &newlabel;
446 		} else
447 #endif
448 		lp = (struct disklabel *)addr;
449 
450 		if ((flag & FWRITE) == 0)
451 			return (EBADF);
452 
453 		mutex_enter(&sc->sc_dk.dk_openlock);
454 		sc->sc_flags |= LDF_LABELLING;
455 
456 		error = setdisklabel(sc->sc_dk.dk_label,
457 		    lp, /*sc->sc_dk.dk_openmask : */0,
458 		    sc->sc_dk.dk_cpulabel);
459 		if (error == 0 && (cmd == DIOCWDINFO
460 #ifdef __HAVE_OLD_DISKLABEL
461 		    || cmd == ODIOCWDINFO
462 #endif
463 		    ))
464 			error = writedisklabel(
465 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
466 			    ldstrategy, sc->sc_dk.dk_label,
467 			    sc->sc_dk.dk_cpulabel);
468 
469 		sc->sc_flags &= ~LDF_LABELLING;
470 		mutex_exit(&sc->sc_dk.dk_openlock);
471 		break;
472 
473 	case DIOCKLABEL:
474 		if ((flag & FWRITE) == 0)
475 			return (EBADF);
476 		if (*(int *)addr)
477 			sc->sc_flags |= LDF_KLABEL;
478 		else
479 			sc->sc_flags &= ~LDF_KLABEL;
480 		break;
481 
482 	case DIOCWLABEL:
483 		if ((flag & FWRITE) == 0)
484 			return (EBADF);
485 		if (*(int *)addr)
486 			sc->sc_flags |= LDF_WLABEL;
487 		else
488 			sc->sc_flags &= ~LDF_WLABEL;
489 		break;
490 
491 	case DIOCGDEFLABEL:
492 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
493 		break;
494 
495 #ifdef __HAVE_OLD_DISKLABEL
496 	case ODIOCGDEFLABEL:
497 		ldgetdefaultlabel(sc, &newlabel);
498 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
499 			return ENOTTY;
500 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
501 		break;
502 #endif
503 
504 	case DIOCCACHESYNC:
505 		/*
506 		 * XXX Do we really need to care about having a writable
507 		 * file descriptor here?
508 		 */
509 		if ((flag & FWRITE) == 0)
510 			error = EBADF;
511 		else if (sc->sc_flush)
512 			error = (*sc->sc_flush)(sc, 0);
513 		else
514 			error = 0;	/* XXX Error out instead? */
515 		break;
516 
517 	case DIOCAWEDGE:
518 	    {
519 	    	struct dkwedge_info *dkw = (void *) addr;
520 
521 		if ((flag & FWRITE) == 0)
522 			return (EBADF);
523 
524 		/* If the ioctl happens here, the parent is us. */
525 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
526 			sizeof(dkw->dkw_parent));
527 		return (dkwedge_add(dkw));
528 	    }
529 
530 	case DIOCDWEDGE:
531 	    {
532 	    	struct dkwedge_info *dkw = (void *) addr;
533 
534 		if ((flag & FWRITE) == 0)
535 			return (EBADF);
536 
537 		/* If the ioctl happens here, the parent is us. */
538 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
539 			sizeof(dkw->dkw_parent));
540 		return (dkwedge_del(dkw));
541 	    }
542 
543 	case DIOCLWEDGES:
544 	    {
545 	    	struct dkwedge_list *dkwl = (void *) addr;
546 
547 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
548 	    }
549 	case DIOCGSTRATEGY:
550 	    {
551 		struct disk_strategy *dks = (void *)addr;
552 
553 		mutex_enter(&sc->sc_mutex);
554 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
555 		    sizeof(dks->dks_name));
556 		mutex_exit(&sc->sc_mutex);
557 		dks->dks_paramlen = 0;
558 
559 		return 0;
560 	    }
561 	case DIOCSSTRATEGY:
562 	    {
563 		struct disk_strategy *dks = (void *)addr;
564 		struct bufq_state *new, *old;
565 
566 		if ((flag & FWRITE) == 0)
567 			return EPERM;
568 
569 		if (dks->dks_param != NULL)
570 			return EINVAL;
571 
572 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
573 		error = bufq_alloc(&new, dks->dks_name,
574 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
575 		if (error)
576 			return error;
577 
578 		mutex_enter(&sc->sc_mutex);
579 		old = sc->sc_bufq;
580 		bufq_move(new, old);
581 		sc->sc_bufq = new;
582 		mutex_exit(&sc->sc_mutex);
583 		bufq_free(old);
584 
585 		return 0;
586 	    }
587 	default:
588 		error = ENOTTY;
589 		break;
590 	}
591 
592 	return (error);
593 }
594 
595 static void
596 ldstrategy(struct buf *bp)
597 {
598 	struct ld_softc *sc;
599 	struct disklabel *lp;
600 	daddr_t blkno;
601 	int s, part;
602 
603 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
604 	part = DISKPART(bp->b_dev);
605 
606 	if ((sc->sc_flags & LDF_DETACH) != 0) {
607 		bp->b_error = EIO;
608 		goto done;
609 	}
610 
611 	lp = sc->sc_dk.dk_label;
612 
613 	/*
614 	 * The transfer must be a whole number of blocks and the offset must
615 	 * not be negative.
616 	 */
617 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
618 		bp->b_error = EINVAL;
619 		goto done;
620 	}
621 
622 	/* If it's a null transfer, return immediately. */
623 	if (bp->b_bcount == 0)
624 		goto done;
625 
626 	/*
627 	 * Do bounds checking and adjust the transfer.  If error, process.
628 	 * If past the end of partition, just return.
629 	 */
630 	if (part == RAW_PART) {
631 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
632 		    sc->sc_disksize512) <= 0)
633 			goto done;
634 	} else {
635 		if (bounds_check_with_label(&sc->sc_dk, bp,
636 		    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
637 			goto done;
638 	}
639 
640 	/*
641 	 * Convert the block number to absolute and put it in terms
642 	 * of the device's logical block size.
643 	 */
644 	if (lp->d_secsize == DEV_BSIZE)
645 		blkno = bp->b_blkno;
646 	else if (lp->d_secsize > DEV_BSIZE)
647 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
648 	else
649 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
650 
651 	if (part != RAW_PART)
652 		blkno += lp->d_partitions[part].p_offset;
653 
654 	bp->b_rawblkno = blkno;
655 
656 	s = splbio();
657 	ldstart(sc, bp);
658 	splx(s);
659 	return;
660 
661  done:
662 	bp->b_resid = bp->b_bcount;
663 	biodone(bp);
664 }
665 
666 static void
667 ldstart(struct ld_softc *sc, struct buf *bp)
668 {
669 	int error;
670 
671 	mutex_enter(&sc->sc_mutex);
672 
673 	if (bp != NULL)
674 		bufq_put(sc->sc_bufq, bp);
675 
676 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
677 		/* See if there is work to do. */
678 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
679 			break;
680 
681 		disk_busy(&sc->sc_dk);
682 		sc->sc_queuecnt++;
683 
684 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
685 			/*
686 			 * The back-end is running the job; remove it from
687 			 * the queue.
688 			 */
689 			(void) bufq_get(sc->sc_bufq);
690 		} else  {
691 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
692 			sc->sc_queuecnt--;
693 			if (error == EAGAIN) {
694 				/*
695 				 * Temporary resource shortage in the
696 				 * back-end; just defer the job until
697 				 * later.
698 				 *
699 				 * XXX We might consider a watchdog timer
700 				 * XXX to make sure we are kicked into action.
701 				 */
702 				break;
703 			} else {
704 				(void) bufq_get(sc->sc_bufq);
705 				bp->b_error = error;
706 				bp->b_resid = bp->b_bcount;
707 				mutex_exit(&sc->sc_mutex);
708 				biodone(bp);
709 				mutex_enter(&sc->sc_mutex);
710 			}
711 		}
712 	}
713 
714 	mutex_exit(&sc->sc_mutex);
715 }
716 
717 void
718 lddone(struct ld_softc *sc, struct buf *bp)
719 {
720 
721 	if (bp->b_error != 0) {
722 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
723 		printf("\n");
724 	}
725 
726 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
727 	    (bp->b_flags & B_READ));
728 #if NRND > 0
729 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
730 #endif
731 	biodone(bp);
732 
733 	mutex_enter(&sc->sc_mutex);
734 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
735 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
736 			sc->sc_flags &= ~LDF_DRAIN;
737 			wakeup(&sc->sc_queuecnt);
738 		}
739 		mutex_exit(&sc->sc_mutex);
740 		ldstart(sc, NULL);
741 	} else
742 		mutex_exit(&sc->sc_mutex);
743 }
744 
745 static int
746 ldsize(dev_t dev)
747 {
748 	struct ld_softc *sc;
749 	int part, unit, omask, size;
750 
751 	unit = DISKUNIT(dev);
752 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
753 		return (ENODEV);
754 	if ((sc->sc_flags & LDF_ENABLED) == 0)
755 		return (ENODEV);
756 	part = DISKPART(dev);
757 
758 	omask = sc->sc_dk.dk_openmask & (1 << part);
759 
760 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
761 		return (-1);
762 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
763 		size = -1;
764 	else
765 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
766 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
767 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
768 		return (-1);
769 
770 	return (size);
771 }
772 
773 /*
774  * Load the label information from the specified device.
775  */
776 static void
777 ldgetdisklabel(struct ld_softc *sc)
778 {
779 	const char *errstring;
780 
781 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
782 
783 	/* Call the generic disklabel extraction routine. */
784 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
785 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
786 	if (errstring != NULL)
787 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
788 
789 	/* In-core label now valid. */
790 	sc->sc_flags |= LDF_VLABEL;
791 }
792 
793 /*
794  * Construct a ficticious label.
795  */
796 static void
797 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
798 {
799 
800 	memset(lp, 0, sizeof(struct disklabel));
801 
802 	lp->d_secsize = sc->sc_secsize;
803 	lp->d_ntracks = sc->sc_nheads;
804 	lp->d_nsectors = sc->sc_nsectors;
805 	lp->d_ncylinders = sc->sc_ncylinders;
806 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
807 	lp->d_type = DTYPE_LD;
808 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
809 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
810 	lp->d_secperunit = sc->sc_secperunit;
811 	lp->d_rpm = 7200;
812 	lp->d_interleave = 1;
813 	lp->d_flags = 0;
814 
815 	lp->d_partitions[RAW_PART].p_offset = 0;
816 	lp->d_partitions[RAW_PART].p_size =
817 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
818 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
819 	lp->d_npartitions = RAW_PART + 1;
820 
821 	lp->d_magic = DISKMAGIC;
822 	lp->d_magic2 = DISKMAGIC;
823 	lp->d_checksum = dkcksum(lp);
824 }
825 
826 /*
827  * Take a dump.
828  */
829 static int
830 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
831 {
832 	char *va = vav;
833 	struct ld_softc *sc;
834 	struct disklabel *lp;
835 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
836 	static int dumping;
837 
838 	unit = DISKUNIT(dev);
839 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
840 		return (ENXIO);
841 	if ((sc->sc_flags & LDF_ENABLED) == 0)
842 		return (ENODEV);
843 	if (sc->sc_dump == NULL)
844 		return (ENXIO);
845 
846 	/* Check if recursive dump; if so, punt. */
847 	if (dumping)
848 		return (EFAULT);
849 	dumping = 1;
850 
851 	/* Convert to disk sectors.  Request must be a multiple of size. */
852 	part = DISKPART(dev);
853 	lp = sc->sc_dk.dk_label;
854 	if ((size % lp->d_secsize) != 0)
855 		return (EFAULT);
856 	towrt = size / lp->d_secsize;
857 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
858 
859 	nsects = lp->d_partitions[part].p_size;
860 	sectoff = lp->d_partitions[part].p_offset;
861 
862 	/* Check transfer bounds against partition size. */
863 	if ((blkno < 0) || ((blkno + towrt) > nsects))
864 		return (EINVAL);
865 
866 	/* Offset block number to start of partition. */
867 	blkno += sectoff;
868 
869 	/* Start dumping and return when done. */
870 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
871 	while (towrt > 0) {
872 		nblk = min(maxblkcnt, towrt);
873 
874 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
875 			return (rv);
876 
877 		towrt -= nblk;
878 		blkno += nblk;
879 		va += nblk * sc->sc_secsize;
880 	}
881 
882 	dumping = 0;
883 	return (0);
884 }
885 
886 /*
887  * Adjust the size of a transfer.
888  */
889 static void
890 ldminphys(struct buf *bp)
891 {
892 	struct ld_softc *sc;
893 
894 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
895 
896 	if (bp->b_bcount > sc->sc_maxxfer)
897 		bp->b_bcount = sc->sc_maxxfer;
898 	minphys(bp);
899 }
900 
901 static void
902 ld_set_properties(struct ld_softc *ld)
903 {
904 	prop_dictionary_t disk_info, odisk_info, geom;
905 
906 	disk_info = prop_dictionary_create();
907 
908 	geom = prop_dictionary_create();
909 
910 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
911 	    ld->sc_secperunit);
912 
913 	prop_dictionary_set_uint32(geom, "sector-size",
914 	    ld->sc_secsize);
915 
916 	prop_dictionary_set_uint16(geom, "sectors-per-track",
917 	    ld->sc_nsectors);
918 
919 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
920 	    ld->sc_nheads);
921 
922 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
923 	    ld->sc_ncylinders);
924 
925 	prop_dictionary_set(disk_info, "geometry", geom);
926 	prop_object_release(geom);
927 
928 	prop_dictionary_set(device_properties(ld->sc_dv),
929 	    "disk-info", disk_info);
930 
931 	/*
932 	 * Don't release disk_info here; we keep a reference to it.
933 	 * disk_detach() will release it when we go away.
934 	 */
935 
936 	odisk_info = ld->sc_dk.dk_info;
937 	ld->sc_dk.dk_info = disk_info;
938 	if (odisk_info)
939 		prop_object_release(odisk_info);
940 }
941 
942 static void
943 ld_config_interrupts(device_t d)
944 {
945 	struct ld_softc *sc = device_private(d);
946 	dkwedge_discover(&sc->sc_dk);
947 }
948