xref: /netbsd-src/sys/dev/ld.c (revision cac8e449158efc7261bebc8657cbb0125a2cfdde)
1 /*	$NetBSD: ld.c,v 1.61 2008/08/01 16:09:45 ws 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 1.61 2008/08/01 16:09:45 ws 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_shutdown(device_t, int);
71 static void	ldstart(struct ld_softc *, struct buf *);
72 static void	ld_set_properties(struct ld_softc *);
73 static void	ld_config_interrupts (struct device *);
74 
75 extern struct	cfdriver ld_cd;
76 
77 static dev_type_open(ldopen);
78 static dev_type_close(ldclose);
79 static dev_type_read(ldread);
80 static dev_type_write(ldwrite);
81 static dev_type_ioctl(ldioctl);
82 static dev_type_strategy(ldstrategy);
83 static dev_type_dump(lddump);
84 static dev_type_size(ldsize);
85 
86 const struct bdevsw ld_bdevsw = {
87 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
88 };
89 
90 const struct cdevsw ld_cdevsw = {
91 	ldopen, ldclose, ldread, ldwrite, ldioctl,
92 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
93 };
94 
95 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
96 
97 void
98 ldattach(struct ld_softc *sc)
99 {
100 	char tbuf[9];
101 
102 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
103 
104 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
105 		aprint_normal_dev(&sc->sc_dv, "disabled\n");
106 		return;
107 	}
108 
109 	/* Initialise and attach the disk structure. */
110 	disk_init(&sc->sc_dk, device_xname(&sc->sc_dv), &lddkdriver);
111 	disk_attach(&sc->sc_dk);
112 
113 	if (sc->sc_maxxfer > MAXPHYS)
114 		sc->sc_maxxfer = MAXPHYS;
115 
116 	/* Build synthetic geometry if necessary. */
117 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
118 	    sc->sc_ncylinders == 0) {
119 		uint64_t ncyl;
120 
121 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
122 			sc->sc_nheads = 16;
123 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
124 			sc->sc_nheads = 32;
125 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
126 			sc->sc_nheads = 64;
127 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
128 			sc->sc_nheads = 128;
129 		else
130 			sc->sc_nheads = 255;
131 
132 		sc->sc_nsectors = 63;
133 		sc->sc_ncylinders = INT_MAX;
134 		ncyl = sc->sc_secperunit /
135 		    (sc->sc_nheads * sc->sc_nsectors);
136 		if (ncyl < INT_MAX)
137 			sc->sc_ncylinders = (int)ncyl;
138 	}
139 
140 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
141 	    sc->sc_secsize);
142 	aprint_normal_dev(&sc->sc_dv, "%s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
143 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
144 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
145 
146 	ld_set_properties(sc);
147 
148 #if NRND > 0
149 	/* Attach the device into the rnd source list. */
150 	rnd_attach_source(&sc->sc_rnd_source, device_xname(&sc->sc_dv),
151 	    RND_TYPE_DISK, 0);
152 #endif
153 
154 	/* Register with PMF */
155 	if (!pmf_device_register1(&sc->sc_dv, NULL, NULL, ld_shutdown))
156 		aprint_error_dev(&sc->sc_dv,
157 		    "couldn't establish power handler\n");
158 
159 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
160 
161 	/* Discover wedges on this disk. */
162 	config_interrupts(&sc->sc_dv, ld_config_interrupts);
163 }
164 
165 int
166 ldadjqparam(struct ld_softc *sc, int xmax)
167 {
168 	int s;
169 
170 	s = splbio();
171 	sc->sc_maxqueuecnt = xmax;
172 	splx(s);
173 
174 	return (0);
175 }
176 
177 int
178 ldbegindetach(struct ld_softc *sc, int flags)
179 {
180 	int s, rv = 0;
181 
182 	if ((sc->sc_flags & LDF_ENABLED) == 0)
183 		return (0);
184 
185 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
186 		return (EBUSY);
187 
188 	s = splbio();
189 	sc->sc_maxqueuecnt = 0;
190 	sc->sc_flags |= LDF_DETACH;
191 	while (sc->sc_queuecnt > 0) {
192 		sc->sc_flags |= LDF_DRAIN;
193 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
194 		if (rv)
195 			break;
196 	}
197 	splx(s);
198 
199 	return (rv);
200 }
201 
202 void
203 ldenddetach(struct ld_softc *sc)
204 {
205 	int s, bmaj, cmaj, i, mn;
206 
207 	if ((sc->sc_flags & LDF_ENABLED) == 0)
208 		return;
209 
210 	/* Wait for commands queued with the hardware to complete. */
211 	if (sc->sc_queuecnt != 0)
212 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
213 			printf("%s: not drained\n", device_xname(&sc->sc_dv));
214 
215 	/* Locate the major numbers. */
216 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
217 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
218 
219 	/* Kill off any queued buffers. */
220 	s = splbio();
221 	bufq_drain(sc->sc_bufq);
222 	splx(s);
223 
224 	bufq_free(sc->sc_bufq);
225 
226 	/* Nuke the vnodes for any open instances. */
227 	for (i = 0; i < MAXPARTITIONS; i++) {
228 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
229 		vdevgone(bmaj, mn, mn, VBLK);
230 		vdevgone(cmaj, mn, mn, VCHR);
231 	}
232 
233 	/* Delete all of our wedges. */
234 	dkwedge_delall(&sc->sc_dk);
235 
236 	/* Detach from the disk list. */
237 	disk_detach(&sc->sc_dk);
238 	disk_destroy(&sc->sc_dk);
239 
240 #if NRND > 0
241 	/* Unhook the entropy source. */
242 	rnd_detach_source(&sc->sc_rnd_source);
243 #endif
244 
245 	/* Deregister with PMF */
246 	pmf_device_deregister(&sc->sc_dv);
247 
248 	/*
249 	 * XXX We can't really flush the cache here, beceause the
250 	 * XXX device may already be non-existent from the controller's
251 	 * XXX perspective.
252 	 */
253 #if 0
254 	/* Flush the device's cache. */
255 	if (sc->sc_flush != NULL)
256 		if ((*sc->sc_flush)(sc) != 0)
257 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
258 #endif
259 	mutex_destroy(&sc->sc_mutex);
260 }
261 
262 /* ARGSUSED */
263 static bool
264 ld_shutdown(device_t dev, int flags)
265 {
266 	struct ld_softc *sc = device_private(dev);
267 
268 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0) {
269 		printf("%s: unable to flush cache\n", device_xname(dev));
270 		return false;
271 	}
272 
273 	return true;
274 }
275 
276 /* ARGSUSED */
277 static int
278 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
279 {
280 	struct ld_softc *sc;
281 	int error, unit, part;
282 
283 	unit = DISKUNIT(dev);
284 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
285 		return (ENXIO);
286 	if ((sc->sc_flags & LDF_ENABLED) == 0)
287 		return (ENODEV);
288 	part = DISKPART(dev);
289 
290 	mutex_enter(&sc->sc_dk.dk_openlock);
291 
292 	if (sc->sc_dk.dk_openmask == 0) {
293 		/* Load the partition info if not already loaded. */
294 		if ((sc->sc_flags & LDF_VLABEL) == 0)
295 			ldgetdisklabel(sc);
296 	}
297 
298 	/* Check that the partition exists. */
299 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
300 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
301 		error = ENXIO;
302 		goto bad1;
303 	}
304 
305 	/* Ensure only one open at a time. */
306 	switch (fmt) {
307 	case S_IFCHR:
308 		sc->sc_dk.dk_copenmask |= (1 << part);
309 		break;
310 	case S_IFBLK:
311 		sc->sc_dk.dk_bopenmask |= (1 << part);
312 		break;
313 	}
314 	sc->sc_dk.dk_openmask =
315 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
316 
317 	error = 0;
318  bad1:
319 	mutex_exit(&sc->sc_dk.dk_openlock);
320 	return (error);
321 }
322 
323 /* ARGSUSED */
324 static int
325 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
326 {
327 	struct ld_softc *sc;
328 	int part, unit;
329 
330 	unit = DISKUNIT(dev);
331 	part = DISKPART(dev);
332 	sc = device_lookup_private(&ld_cd, unit);
333 
334 	mutex_enter(&sc->sc_dk.dk_openlock);
335 
336 	switch (fmt) {
337 	case S_IFCHR:
338 		sc->sc_dk.dk_copenmask &= ~(1 << part);
339 		break;
340 	case S_IFBLK:
341 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
342 		break;
343 	}
344 	sc->sc_dk.dk_openmask =
345 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
346 
347 	if (sc->sc_dk.dk_openmask == 0) {
348 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
349 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
350 		if ((sc->sc_flags & LDF_KLABEL) == 0)
351 			sc->sc_flags &= ~LDF_VLABEL;
352 	}
353 
354 	mutex_exit(&sc->sc_dk.dk_openlock);
355 	return (0);
356 }
357 
358 /* ARGSUSED */
359 static int
360 ldread(dev_t dev, struct uio *uio, int ioflag)
361 {
362 
363 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
364 }
365 
366 /* ARGSUSED */
367 static int
368 ldwrite(dev_t dev, struct uio *uio, int ioflag)
369 {
370 
371 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
372 }
373 
374 /* ARGSUSED */
375 static int
376 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
377 {
378 	struct ld_softc *sc;
379 	int part, unit, error;
380 #ifdef __HAVE_OLD_DISKLABEL
381 	struct disklabel newlabel;
382 #endif
383 	struct disklabel *lp;
384 
385 	unit = DISKUNIT(dev);
386 	part = DISKPART(dev);
387 	sc = device_lookup_private(&ld_cd, unit);
388 
389 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
390 	if (error != EPASSTHROUGH)
391 		return (error);
392 
393 	error = 0;
394 	switch (cmd) {
395 	case DIOCGDINFO:
396 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
397 		return (0);
398 
399 #ifdef __HAVE_OLD_DISKLABEL
400 	case ODIOCGDINFO:
401 		newlabel = *(sc->sc_dk.dk_label);
402 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
403 			return ENOTTY;
404 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
405 		return (0);
406 #endif
407 
408 	case DIOCGPART:
409 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
410 		((struct partinfo *)addr)->part =
411 		    &sc->sc_dk.dk_label->d_partitions[part];
412 		break;
413 
414 	case DIOCWDINFO:
415 	case DIOCSDINFO:
416 #ifdef __HAVE_OLD_DISKLABEL
417 	case ODIOCWDINFO:
418 	case ODIOCSDINFO:
419 
420 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
421 			memset(&newlabel, 0, sizeof newlabel);
422 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
423 			lp = &newlabel;
424 		} else
425 #endif
426 		lp = (struct disklabel *)addr;
427 
428 		if ((flag & FWRITE) == 0)
429 			return (EBADF);
430 
431 		mutex_enter(&sc->sc_dk.dk_openlock);
432 		sc->sc_flags |= LDF_LABELLING;
433 
434 		error = setdisklabel(sc->sc_dk.dk_label,
435 		    lp, /*sc->sc_dk.dk_openmask : */0,
436 		    sc->sc_dk.dk_cpulabel);
437 		if (error == 0 && (cmd == DIOCWDINFO
438 #ifdef __HAVE_OLD_DISKLABEL
439 		    || cmd == ODIOCWDINFO
440 #endif
441 		    ))
442 			error = writedisklabel(
443 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
444 			    ldstrategy, sc->sc_dk.dk_label,
445 			    sc->sc_dk.dk_cpulabel);
446 
447 		sc->sc_flags &= ~LDF_LABELLING;
448 		mutex_exit(&sc->sc_dk.dk_openlock);
449 		break;
450 
451 	case DIOCKLABEL:
452 		if ((flag & FWRITE) == 0)
453 			return (EBADF);
454 		if (*(int *)addr)
455 			sc->sc_flags |= LDF_KLABEL;
456 		else
457 			sc->sc_flags &= ~LDF_KLABEL;
458 		break;
459 
460 	case DIOCWLABEL:
461 		if ((flag & FWRITE) == 0)
462 			return (EBADF);
463 		if (*(int *)addr)
464 			sc->sc_flags |= LDF_WLABEL;
465 		else
466 			sc->sc_flags &= ~LDF_WLABEL;
467 		break;
468 
469 	case DIOCGDEFLABEL:
470 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
471 		break;
472 
473 #ifdef __HAVE_OLD_DISKLABEL
474 	case ODIOCGDEFLABEL:
475 		ldgetdefaultlabel(sc, &newlabel);
476 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
477 			return ENOTTY;
478 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
479 		break;
480 #endif
481 
482 	case DIOCCACHESYNC:
483 		/*
484 		 * XXX Do we really need to care about having a writable
485 		 * file descriptor here?
486 		 */
487 		if ((flag & FWRITE) == 0)
488 			error = EBADF;
489 		else if (sc->sc_flush)
490 			error = (*sc->sc_flush)(sc);
491 		else
492 			error = 0;	/* XXX Error out instead? */
493 		break;
494 
495 	case DIOCAWEDGE:
496 	    {
497 	    	struct dkwedge_info *dkw = (void *) addr;
498 
499 		if ((flag & FWRITE) == 0)
500 			return (EBADF);
501 
502 		/* If the ioctl happens here, the parent is us. */
503 		strlcpy(dkw->dkw_parent, device_xname(&sc->sc_dv),
504 			sizeof(dkw->dkw_parent));
505 		return (dkwedge_add(dkw));
506 	    }
507 
508 	case DIOCDWEDGE:
509 	    {
510 	    	struct dkwedge_info *dkw = (void *) addr;
511 
512 		if ((flag & FWRITE) == 0)
513 			return (EBADF);
514 
515 		/* If the ioctl happens here, the parent is us. */
516 		strlcpy(dkw->dkw_parent, device_xname(&sc->sc_dv),
517 			sizeof(dkw->dkw_parent));
518 		return (dkwedge_del(dkw));
519 	    }
520 
521 	case DIOCLWEDGES:
522 	    {
523 	    	struct dkwedge_list *dkwl = (void *) addr;
524 
525 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
526 	    }
527 	case DIOCGSTRATEGY:
528 	    {
529 		struct disk_strategy *dks = (void *)addr;
530 
531 		mutex_enter(&sc->sc_mutex);
532 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
533 		    sizeof(dks->dks_name));
534 		mutex_exit(&sc->sc_mutex);
535 		dks->dks_paramlen = 0;
536 
537 		return 0;
538 	    }
539 	case DIOCSSTRATEGY:
540 	    {
541 		struct disk_strategy *dks = (void *)addr;
542 		struct bufq_state *new, *old;
543 
544 		if ((flag & FWRITE) == 0)
545 			return EPERM;
546 
547 		if (dks->dks_param != NULL)
548 			return EINVAL;
549 
550 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
551 		error = bufq_alloc(&new, dks->dks_name,
552 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
553 		if (error)
554 			return error;
555 
556 		mutex_enter(&sc->sc_mutex);
557 		old = sc->sc_bufq;
558 		bufq_move(new, old);
559 		sc->sc_bufq = new;
560 		mutex_exit(&sc->sc_mutex);
561 		bufq_free(old);
562 
563 		return 0;
564 	    }
565 	default:
566 		error = ENOTTY;
567 		break;
568 	}
569 
570 	return (error);
571 }
572 
573 static void
574 ldstrategy(struct buf *bp)
575 {
576 	struct ld_softc *sc;
577 	struct disklabel *lp;
578 	daddr_t blkno;
579 	int s, part;
580 
581 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
582 	part = DISKPART(bp->b_dev);
583 
584 	if ((sc->sc_flags & LDF_DETACH) != 0) {
585 		bp->b_error = EIO;
586 		goto done;
587 	}
588 
589 	lp = sc->sc_dk.dk_label;
590 
591 	/*
592 	 * The transfer must be a whole number of blocks and the offset must
593 	 * not be negative.
594 	 */
595 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
596 		bp->b_error = EINVAL;
597 		goto done;
598 	}
599 
600 	/* If it's a null transfer, return immediately. */
601 	if (bp->b_bcount == 0)
602 		goto done;
603 
604 	/*
605 	 * Do bounds checking and adjust the transfer.  If error, process.
606 	 * If past the end of partition, just return.
607 	 */
608 	if (part != RAW_PART &&
609 	    bounds_check_with_label(&sc->sc_dk, bp,
610 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
611 		goto done;
612 	}
613 
614 	/*
615 	 * Convert the block number to absolute and put it in terms
616 	 * of the device's logical block size.
617 	 */
618 	if (lp->d_secsize == DEV_BSIZE)
619 		blkno = bp->b_blkno;
620 	else if (lp->d_secsize > DEV_BSIZE)
621 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
622 	else
623 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
624 
625 	if (part != RAW_PART)
626 		blkno += lp->d_partitions[part].p_offset;
627 
628 	bp->b_rawblkno = blkno;
629 
630 	s = splbio();
631 	ldstart(sc, bp);
632 	splx(s);
633 	return;
634 
635  done:
636 	bp->b_resid = bp->b_bcount;
637 	biodone(bp);
638 }
639 
640 static void
641 ldstart(struct ld_softc *sc, struct buf *bp)
642 {
643 	int error;
644 
645 	mutex_enter(&sc->sc_mutex);
646 
647 	if (bp != NULL)
648 		BUFQ_PUT(sc->sc_bufq, bp);
649 
650 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
651 		/* See if there is work to do. */
652 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
653 			break;
654 
655 		disk_busy(&sc->sc_dk);
656 		sc->sc_queuecnt++;
657 
658 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
659 			/*
660 			 * The back-end is running the job; remove it from
661 			 * the queue.
662 			 */
663 			(void) BUFQ_GET(sc->sc_bufq);
664 		} else  {
665 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
666 			sc->sc_queuecnt--;
667 			if (error == EAGAIN) {
668 				/*
669 				 * Temporary resource shortage in the
670 				 * back-end; just defer the job until
671 				 * later.
672 				 *
673 				 * XXX We might consider a watchdog timer
674 				 * XXX to make sure we are kicked into action.
675 				 */
676 				break;
677 			} else {
678 				(void) BUFQ_GET(sc->sc_bufq);
679 				bp->b_error = error;
680 				bp->b_resid = bp->b_bcount;
681 				mutex_exit(&sc->sc_mutex);
682 				biodone(bp);
683 				mutex_enter(&sc->sc_mutex);
684 			}
685 		}
686 	}
687 
688 	mutex_exit(&sc->sc_mutex);
689 }
690 
691 void
692 lddone(struct ld_softc *sc, struct buf *bp)
693 {
694 
695 	if (bp->b_error != 0) {
696 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
697 		printf("\n");
698 	}
699 
700 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
701 	    (bp->b_flags & B_READ));
702 #if NRND > 0
703 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
704 #endif
705 	biodone(bp);
706 
707 	mutex_enter(&sc->sc_mutex);
708 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
709 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
710 			sc->sc_flags &= ~LDF_DRAIN;
711 			wakeup(&sc->sc_queuecnt);
712 		}
713 		mutex_exit(&sc->sc_mutex);
714 		ldstart(sc, NULL);
715 	} else
716 		mutex_exit(&sc->sc_mutex);
717 }
718 
719 static int
720 ldsize(dev_t dev)
721 {
722 	struct ld_softc *sc;
723 	int part, unit, omask, size;
724 
725 	unit = DISKUNIT(dev);
726 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
727 		return (ENODEV);
728 	if ((sc->sc_flags & LDF_ENABLED) == 0)
729 		return (ENODEV);
730 	part = DISKPART(dev);
731 
732 	omask = sc->sc_dk.dk_openmask & (1 << part);
733 
734 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
735 		return (-1);
736 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
737 		size = -1;
738 	else
739 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
740 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
741 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
742 		return (-1);
743 
744 	return (size);
745 }
746 
747 /*
748  * Load the label information from the specified device.
749  */
750 static void
751 ldgetdisklabel(struct ld_softc *sc)
752 {
753 	const char *errstring;
754 
755 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
756 
757 	/* Call the generic disklabel extraction routine. */
758 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
759 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
760 	if (errstring != NULL)
761 		printf("%s: %s\n", device_xname(&sc->sc_dv), errstring);
762 
763 	/* In-core label now valid. */
764 	sc->sc_flags |= LDF_VLABEL;
765 }
766 
767 /*
768  * Construct a ficticious label.
769  */
770 static void
771 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
772 {
773 
774 	memset(lp, 0, sizeof(struct disklabel));
775 
776 	lp->d_secsize = sc->sc_secsize;
777 	lp->d_ntracks = sc->sc_nheads;
778 	lp->d_nsectors = sc->sc_nsectors;
779 	lp->d_ncylinders = sc->sc_ncylinders;
780 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
781 	lp->d_type = DTYPE_LD;
782 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
783 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
784 	lp->d_secperunit = sc->sc_secperunit;
785 	lp->d_rpm = 7200;
786 	lp->d_interleave = 1;
787 	lp->d_flags = 0;
788 
789 	lp->d_partitions[RAW_PART].p_offset = 0;
790 	lp->d_partitions[RAW_PART].p_size =
791 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
792 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
793 	lp->d_npartitions = RAW_PART + 1;
794 
795 	lp->d_magic = DISKMAGIC;
796 	lp->d_magic2 = DISKMAGIC;
797 	lp->d_checksum = dkcksum(lp);
798 }
799 
800 /*
801  * Take a dump.
802  */
803 static int
804 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
805 {
806 	char *va = vav;
807 	struct ld_softc *sc;
808 	struct disklabel *lp;
809 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
810 	static int dumping;
811 
812 	unit = DISKUNIT(dev);
813 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
814 		return (ENXIO);
815 	if ((sc->sc_flags & LDF_ENABLED) == 0)
816 		return (ENODEV);
817 	if (sc->sc_dump == NULL)
818 		return (ENXIO);
819 
820 	/* Check if recursive dump; if so, punt. */
821 	if (dumping)
822 		return (EFAULT);
823 	dumping = 1;
824 
825 	/* Convert to disk sectors.  Request must be a multiple of size. */
826 	part = DISKPART(dev);
827 	lp = sc->sc_dk.dk_label;
828 	if ((size % lp->d_secsize) != 0)
829 		return (EFAULT);
830 	towrt = size / lp->d_secsize;
831 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
832 
833 	nsects = lp->d_partitions[part].p_size;
834 	sectoff = lp->d_partitions[part].p_offset;
835 
836 	/* Check transfer bounds against partition size. */
837 	if ((blkno < 0) || ((blkno + towrt) > nsects))
838 		return (EINVAL);
839 
840 	/* Offset block number to start of partition. */
841 	blkno += sectoff;
842 
843 	/* Start dumping and return when done. */
844 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
845 	while (towrt > 0) {
846 		nblk = min(maxblkcnt, towrt);
847 
848 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
849 			return (rv);
850 
851 		towrt -= nblk;
852 		blkno += nblk;
853 		va += nblk * sc->sc_secsize;
854 	}
855 
856 	dumping = 0;
857 	return (0);
858 }
859 
860 /*
861  * Adjust the size of a transfer.
862  */
863 static void
864 ldminphys(struct buf *bp)
865 {
866 	struct ld_softc *sc;
867 
868 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
869 
870 	if (bp->b_bcount > sc->sc_maxxfer)
871 		bp->b_bcount = sc->sc_maxxfer;
872 	minphys(bp);
873 }
874 
875 static void
876 ld_set_properties(struct ld_softc *ld)
877 {
878 	prop_dictionary_t disk_info, odisk_info, geom;
879 
880 	disk_info = prop_dictionary_create();
881 
882 	geom = prop_dictionary_create();
883 
884 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
885 	    ld->sc_secperunit);
886 
887 	prop_dictionary_set_uint32(geom, "sector-size",
888 	    ld->sc_secsize);
889 
890 	prop_dictionary_set_uint16(geom, "sectors-per-track",
891 	    ld->sc_nsectors);
892 
893 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
894 	    ld->sc_nheads);
895 
896 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
897 	    ld->sc_ncylinders);
898 
899 	prop_dictionary_set(disk_info, "geometry", geom);
900 	prop_object_release(geom);
901 
902 	prop_dictionary_set(device_properties(&ld->sc_dv),
903 	    "disk-info", disk_info);
904 
905 	/*
906 	 * Don't release disk_info here; we keep a reference to it.
907 	 * disk_detach() will release it when we go away.
908 	 */
909 
910 	odisk_info = ld->sc_dk.dk_info;
911 	ld->sc_dk.dk_info = disk_info;
912 	if (odisk_info)
913 		prop_object_release(odisk_info);
914 }
915 
916 static void
917 ld_config_interrupts (struct device *d)
918 {
919 	struct ld_softc *sc = device_private(d);
920 	dkwedge_discover(&sc->sc_dk);
921 }
922