xref: /netbsd-src/sys/dev/ld.c (revision 80d9064ac03cbb6a4174695f0d5b237c8766d3d0)
1 /*	$NetBSD: ld.c,v 1.76 2014/09/05 05:27:23 matt 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 1.76 2014/09/05 05:27:23 matt Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/device.h>
43 #include <sys/queue.h>
44 #include <sys/proc.h>
45 #include <sys/buf.h>
46 #include <sys/bufq.h>
47 #include <sys/endian.h>
48 #include <sys/disklabel.h>
49 #include <sys/disk.h>
50 #include <sys/dkio.h>
51 #include <sys/stat.h>
52 #include <sys/conf.h>
53 #include <sys/fcntl.h>
54 #include <sys/vnode.h>
55 #include <sys/syslog.h>
56 #include <sys/mutex.h>
57 #include <sys/rnd.h>
58 
59 #include <dev/ldvar.h>
60 
61 #include <prop/proplib.h>
62 
63 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
64 static void	ldgetdisklabel(struct ld_softc *);
65 static void	ldminphys(struct buf *bp);
66 static bool	ld_suspend(device_t, const pmf_qual_t *);
67 static bool	ld_shutdown(device_t, int);
68 static void	ldstart(struct ld_softc *, struct buf *);
69 static void	ld_set_geometry(struct ld_softc *);
70 static void	ld_config_interrupts (device_t);
71 static int	ldlastclose(device_t);
72 
73 extern struct	cfdriver ld_cd;
74 
75 static dev_type_open(ldopen);
76 static dev_type_close(ldclose);
77 static dev_type_read(ldread);
78 static dev_type_write(ldwrite);
79 static dev_type_ioctl(ldioctl);
80 static dev_type_strategy(ldstrategy);
81 static dev_type_dump(lddump);
82 static dev_type_size(ldsize);
83 
84 const struct bdevsw ld_bdevsw = {
85 	.d_open = ldopen,
86 	.d_close = ldclose,
87 	.d_strategy = ldstrategy,
88 	.d_ioctl = ldioctl,
89 	.d_dump = lddump,
90 	.d_psize = ldsize,
91 	.d_discard = nodiscard,
92 	.d_flag = D_DISK
93 };
94 
95 const struct cdevsw ld_cdevsw = {
96 	.d_open = ldopen,
97 	.d_close = ldclose,
98 	.d_read = ldread,
99 	.d_write = ldwrite,
100 	.d_ioctl = ldioctl,
101 	.d_stop = nostop,
102 	.d_tty = notty,
103 	.d_poll = nopoll,
104 	.d_mmap = nommap,
105 	.d_kqfilter = nokqfilter,
106 	.d_discard = nodiscard,
107 	.d_flag = D_DISK
108 };
109 
110 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
111 
112 void
113 ldattach(struct ld_softc *sc)
114 {
115 	char tbuf[9];
116 
117 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
118 
119 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
120 		aprint_normal_dev(sc->sc_dv, "disabled\n");
121 		return;
122 	}
123 
124 	/* Initialise and attach the disk structure. */
125 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
126 	disk_attach(&sc->sc_dk);
127 
128 	if (sc->sc_maxxfer > MAXPHYS)
129 		sc->sc_maxxfer = MAXPHYS;
130 
131 	/* Build synthetic geometry if necessary. */
132 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
133 	    sc->sc_ncylinders == 0) {
134 		uint64_t ncyl;
135 
136 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
137 			sc->sc_nheads = 16;
138 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
139 			sc->sc_nheads = 32;
140 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
141 			sc->sc_nheads = 64;
142 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
143 			sc->sc_nheads = 128;
144 		else
145 			sc->sc_nheads = 255;
146 
147 		sc->sc_nsectors = 63;
148 		sc->sc_ncylinders = INT_MAX;
149 		ncyl = sc->sc_secperunit /
150 		    (sc->sc_nheads * sc->sc_nsectors);
151 		if (ncyl < INT_MAX)
152 			sc->sc_ncylinders = (int)ncyl;
153 	}
154 
155 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
156 	    sc->sc_secsize);
157 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
158 	    "%d bytes/sect x %"PRIu64" sectors\n",
159 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
160 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
161 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
162 
163 	ld_set_geometry(sc);
164 
165 	/* Attach the device into the rnd source list. */
166 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
167 	    RND_TYPE_DISK, RND_FLAG_DEFAULT);
168 
169 	/* Register with PMF */
170 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
171 		aprint_error_dev(sc->sc_dv,
172 		    "couldn't establish power handler\n");
173 
174 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
175 
176 	/* Discover wedges on this disk. */
177 	config_interrupts(sc->sc_dv, ld_config_interrupts);
178 }
179 
180 int
181 ldadjqparam(struct ld_softc *sc, int xmax)
182 {
183 	int s;
184 
185 	s = splbio();
186 	sc->sc_maxqueuecnt = xmax;
187 	splx(s);
188 
189 	return (0);
190 }
191 
192 int
193 ldbegindetach(struct ld_softc *sc, int flags)
194 {
195 	int s, rv = 0;
196 
197 	if ((sc->sc_flags & LDF_ENABLED) == 0)
198 		return (0);
199 
200 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
201 
202 	if (rv != 0)
203 		return rv;
204 
205 	s = splbio();
206 	sc->sc_maxqueuecnt = 0;
207 	sc->sc_flags |= LDF_DETACH;
208 	while (sc->sc_queuecnt > 0) {
209 		sc->sc_flags |= LDF_DRAIN;
210 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
211 		if (rv)
212 			break;
213 	}
214 	splx(s);
215 
216 	return (rv);
217 }
218 
219 void
220 ldenddetach(struct ld_softc *sc)
221 {
222 	int s, bmaj, cmaj, i, mn;
223 
224 	if ((sc->sc_flags & LDF_ENABLED) == 0)
225 		return;
226 
227 	/* Wait for commands queued with the hardware to complete. */
228 	if (sc->sc_queuecnt != 0)
229 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
230 			printf("%s: not drained\n", device_xname(sc->sc_dv));
231 
232 	/* Locate the major numbers. */
233 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
234 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
235 
236 	/* Kill off any queued buffers. */
237 	s = splbio();
238 	bufq_drain(sc->sc_bufq);
239 	splx(s);
240 
241 	bufq_free(sc->sc_bufq);
242 
243 	/* Nuke the vnodes for any open instances. */
244 	for (i = 0; i < MAXPARTITIONS; i++) {
245 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
246 		vdevgone(bmaj, mn, mn, VBLK);
247 		vdevgone(cmaj, mn, mn, VCHR);
248 	}
249 
250 	/* Delete all of our wedges. */
251 	dkwedge_delall(&sc->sc_dk);
252 
253 	/* Detach from the disk list. */
254 	disk_detach(&sc->sc_dk);
255 	disk_destroy(&sc->sc_dk);
256 
257 	/* Unhook the entropy source. */
258 	rnd_detach_source(&sc->sc_rnd_source);
259 
260 	/* Deregister with PMF */
261 	pmf_device_deregister(sc->sc_dv);
262 
263 	/*
264 	 * XXX We can't really flush the cache here, beceause the
265 	 * XXX device may already be non-existent from the controller's
266 	 * XXX perspective.
267 	 */
268 #if 0
269 	/* Flush the device's cache. */
270 	if (sc->sc_flush != NULL)
271 		if ((*sc->sc_flush)(sc, 0) != 0)
272 			aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
273 #endif
274 	mutex_destroy(&sc->sc_mutex);
275 }
276 
277 /* ARGSUSED */
278 static bool
279 ld_suspend(device_t dev, const pmf_qual_t *qual)
280 {
281 	return ld_shutdown(dev, 0);
282 }
283 
284 /* ARGSUSED */
285 static bool
286 ld_shutdown(device_t dev, int flags)
287 {
288 	struct ld_softc *sc = device_private(dev);
289 
290 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
291 		printf("%s: unable to flush cache\n", device_xname(dev));
292 		return false;
293 	}
294 
295 	return true;
296 }
297 
298 /* ARGSUSED */
299 static int
300 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
301 {
302 	struct ld_softc *sc;
303 	int error, unit, part;
304 
305 	unit = DISKUNIT(dev);
306 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
307 		return (ENXIO);
308 	if ((sc->sc_flags & LDF_ENABLED) == 0)
309 		return (ENODEV);
310 	part = DISKPART(dev);
311 
312 	mutex_enter(&sc->sc_dk.dk_openlock);
313 
314 	if (sc->sc_dk.dk_openmask == 0) {
315 		/* Load the partition info if not already loaded. */
316 		if ((sc->sc_flags & LDF_VLABEL) == 0)
317 			ldgetdisklabel(sc);
318 	}
319 
320 	/* Check that the partition exists. */
321 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
322 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
323 		error = ENXIO;
324 		goto bad1;
325 	}
326 
327 	/* Ensure only one open at a time. */
328 	switch (fmt) {
329 	case S_IFCHR:
330 		sc->sc_dk.dk_copenmask |= (1 << part);
331 		break;
332 	case S_IFBLK:
333 		sc->sc_dk.dk_bopenmask |= (1 << part);
334 		break;
335 	}
336 	sc->sc_dk.dk_openmask =
337 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
338 
339 	error = 0;
340  bad1:
341 	mutex_exit(&sc->sc_dk.dk_openlock);
342 	return (error);
343 }
344 
345 static int
346 ldlastclose(device_t self)
347 {
348 	struct ld_softc *sc = device_private(self);
349 
350 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
351 		aprint_error_dev(self, "unable to flush cache\n");
352 	if ((sc->sc_flags & LDF_KLABEL) == 0)
353 		sc->sc_flags &= ~LDF_VLABEL;
354 
355 	return 0;
356 }
357 
358 /* ARGSUSED */
359 static int
360 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
361 {
362 	struct ld_softc *sc;
363 	int part, unit;
364 
365 	unit = DISKUNIT(dev);
366 	part = DISKPART(dev);
367 	sc = device_lookup_private(&ld_cd, unit);
368 
369 	mutex_enter(&sc->sc_dk.dk_openlock);
370 
371 	switch (fmt) {
372 	case S_IFCHR:
373 		sc->sc_dk.dk_copenmask &= ~(1 << part);
374 		break;
375 	case S_IFBLK:
376 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
377 		break;
378 	}
379 	sc->sc_dk.dk_openmask =
380 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
381 
382 	if (sc->sc_dk.dk_openmask == 0)
383 		ldlastclose(sc->sc_dv);
384 
385 	mutex_exit(&sc->sc_dk.dk_openlock);
386 	return (0);
387 }
388 
389 /* ARGSUSED */
390 static int
391 ldread(dev_t dev, struct uio *uio, int ioflag)
392 {
393 
394 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
395 }
396 
397 /* ARGSUSED */
398 static int
399 ldwrite(dev_t dev, struct uio *uio, int ioflag)
400 {
401 
402 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
403 }
404 
405 /* ARGSUSED */
406 static int
407 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
408 {
409 	struct ld_softc *sc;
410 	int part, unit, error;
411 #ifdef __HAVE_OLD_DISKLABEL
412 	struct disklabel newlabel;
413 #endif
414 	struct disklabel *lp;
415 
416 	unit = DISKUNIT(dev);
417 	part = DISKPART(dev);
418 	sc = device_lookup_private(&ld_cd, unit);
419 
420 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
421 	if (error != EPASSTHROUGH)
422 		return (error);
423 
424 	error = 0;
425 	switch (cmd) {
426 	case DIOCGDINFO:
427 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
428 		return (0);
429 
430 #ifdef __HAVE_OLD_DISKLABEL
431 	case ODIOCGDINFO:
432 		newlabel = *(sc->sc_dk.dk_label);
433 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
434 			return ENOTTY;
435 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
436 		return (0);
437 #endif
438 
439 	case DIOCGPART:
440 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
441 		((struct partinfo *)addr)->part =
442 		    &sc->sc_dk.dk_label->d_partitions[part];
443 		break;
444 
445 	case DIOCWDINFO:
446 	case DIOCSDINFO:
447 #ifdef __HAVE_OLD_DISKLABEL
448 	case ODIOCWDINFO:
449 	case ODIOCSDINFO:
450 
451 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
452 			memset(&newlabel, 0, sizeof newlabel);
453 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
454 			lp = &newlabel;
455 		} else
456 #endif
457 		lp = (struct disklabel *)addr;
458 
459 		if ((flag & FWRITE) == 0)
460 			return (EBADF);
461 
462 		mutex_enter(&sc->sc_dk.dk_openlock);
463 		sc->sc_flags |= LDF_LABELLING;
464 
465 		error = setdisklabel(sc->sc_dk.dk_label,
466 		    lp, /*sc->sc_dk.dk_openmask : */0,
467 		    sc->sc_dk.dk_cpulabel);
468 		if (error == 0 && (cmd == DIOCWDINFO
469 #ifdef __HAVE_OLD_DISKLABEL
470 		    || cmd == ODIOCWDINFO
471 #endif
472 		    ))
473 			error = writedisklabel(
474 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
475 			    ldstrategy, sc->sc_dk.dk_label,
476 			    sc->sc_dk.dk_cpulabel);
477 
478 		sc->sc_flags &= ~LDF_LABELLING;
479 		mutex_exit(&sc->sc_dk.dk_openlock);
480 		break;
481 
482 	case DIOCKLABEL:
483 		if ((flag & FWRITE) == 0)
484 			return (EBADF);
485 		if (*(int *)addr)
486 			sc->sc_flags |= LDF_KLABEL;
487 		else
488 			sc->sc_flags &= ~LDF_KLABEL;
489 		break;
490 
491 	case DIOCWLABEL:
492 		if ((flag & FWRITE) == 0)
493 			return (EBADF);
494 		if (*(int *)addr)
495 			sc->sc_flags |= LDF_WLABEL;
496 		else
497 			sc->sc_flags &= ~LDF_WLABEL;
498 		break;
499 
500 	case DIOCGDEFLABEL:
501 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
502 		break;
503 
504 #ifdef __HAVE_OLD_DISKLABEL
505 	case ODIOCGDEFLABEL:
506 		ldgetdefaultlabel(sc, &newlabel);
507 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
508 			return ENOTTY;
509 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
510 		break;
511 #endif
512 
513 	case DIOCCACHESYNC:
514 		/*
515 		 * XXX Do we really need to care about having a writable
516 		 * file descriptor here?
517 		 */
518 		if ((flag & FWRITE) == 0)
519 			error = EBADF;
520 		else if (sc->sc_flush)
521 			error = (*sc->sc_flush)(sc, 0);
522 		else
523 			error = 0;	/* XXX Error out instead? */
524 		break;
525 
526 	case DIOCAWEDGE:
527 	    {
528 	    	struct dkwedge_info *dkw = (void *) addr;
529 
530 		if ((flag & FWRITE) == 0)
531 			return (EBADF);
532 
533 		/* If the ioctl happens here, the parent is us. */
534 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
535 			sizeof(dkw->dkw_parent));
536 		return (dkwedge_add(dkw));
537 	    }
538 
539 	case DIOCDWEDGE:
540 	    {
541 	    	struct dkwedge_info *dkw = (void *) addr;
542 
543 		if ((flag & FWRITE) == 0)
544 			return (EBADF);
545 
546 		/* If the ioctl happens here, the parent is us. */
547 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
548 			sizeof(dkw->dkw_parent));
549 		return (dkwedge_del(dkw));
550 	    }
551 
552 	case DIOCLWEDGES:
553 	    {
554 	    	struct dkwedge_list *dkwl = (void *) addr;
555 
556 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
557 	    }
558 	case DIOCGSTRATEGY:
559 	    {
560 		struct disk_strategy *dks = (void *)addr;
561 
562 		mutex_enter(&sc->sc_mutex);
563 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
564 		    sizeof(dks->dks_name));
565 		mutex_exit(&sc->sc_mutex);
566 		dks->dks_paramlen = 0;
567 
568 		return 0;
569 	    }
570 	case DIOCSSTRATEGY:
571 	    {
572 		struct disk_strategy *dks = (void *)addr;
573 		struct bufq_state *new_bufq, *old_bufq;
574 
575 		if ((flag & FWRITE) == 0)
576 			return EPERM;
577 
578 		if (dks->dks_param != NULL)
579 			return EINVAL;
580 
581 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
582 		error = bufq_alloc(&new_bufq, dks->dks_name,
583 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
584 		if (error)
585 			return error;
586 
587 		mutex_enter(&sc->sc_mutex);
588 		old_bufq = sc->sc_bufq;
589 		bufq_move(new_bufq, old_bufq);
590 		sc->sc_bufq = new_bufq;
591 		mutex_exit(&sc->sc_mutex);
592 		bufq_free(old_bufq);
593 
594 		return 0;
595 	    }
596 	default:
597 		error = ENOTTY;
598 		break;
599 	}
600 
601 	return (error);
602 }
603 
604 static void
605 ldstrategy(struct buf *bp)
606 {
607 	struct ld_softc *sc;
608 	struct disklabel *lp;
609 	daddr_t blkno;
610 	int s, part;
611 
612 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
613 	part = DISKPART(bp->b_dev);
614 
615 	if ((sc->sc_flags & LDF_DETACH) != 0) {
616 		bp->b_error = EIO;
617 		goto done;
618 	}
619 
620 	lp = sc->sc_dk.dk_label;
621 
622 	/*
623 	 * The transfer must be a whole number of blocks and the offset must
624 	 * not be negative.
625 	 */
626 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
627 		bp->b_error = EINVAL;
628 		goto done;
629 	}
630 
631 	/* If it's a null transfer, return immediately. */
632 	if (bp->b_bcount == 0)
633 		goto done;
634 
635 	/*
636 	 * Do bounds checking and adjust the transfer.  If error, process.
637 	 * If past the end of partition, just return.
638 	 */
639 	if (part == RAW_PART) {
640 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
641 		    sc->sc_disksize512) <= 0)
642 			goto done;
643 	} else {
644 		if (bounds_check_with_label(&sc->sc_dk, bp,
645 		    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
646 			goto done;
647 	}
648 
649 	/*
650 	 * Convert the block number to absolute and put it in terms
651 	 * of the device's logical block size.
652 	 */
653 	if (lp->d_secsize == DEV_BSIZE)
654 		blkno = bp->b_blkno;
655 	else if (lp->d_secsize > DEV_BSIZE)
656 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
657 	else
658 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
659 
660 	if (part != RAW_PART)
661 		blkno += lp->d_partitions[part].p_offset;
662 
663 	bp->b_rawblkno = blkno;
664 
665 	s = splbio();
666 	ldstart(sc, bp);
667 	splx(s);
668 	return;
669 
670  done:
671 	bp->b_resid = bp->b_bcount;
672 	biodone(bp);
673 }
674 
675 static void
676 ldstart(struct ld_softc *sc, struct buf *bp)
677 {
678 	int error;
679 
680 	mutex_enter(&sc->sc_mutex);
681 
682 	if (bp != NULL)
683 		bufq_put(sc->sc_bufq, bp);
684 
685 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
686 		/* See if there is work to do. */
687 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
688 			break;
689 
690 		disk_busy(&sc->sc_dk);
691 		sc->sc_queuecnt++;
692 
693 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
694 			/*
695 			 * The back-end is running the job; remove it from
696 			 * the queue.
697 			 */
698 			(void) bufq_get(sc->sc_bufq);
699 		} else  {
700 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
701 			sc->sc_queuecnt--;
702 			if (error == EAGAIN) {
703 				/*
704 				 * Temporary resource shortage in the
705 				 * back-end; just defer the job until
706 				 * later.
707 				 *
708 				 * XXX We might consider a watchdog timer
709 				 * XXX to make sure we are kicked into action.
710 				 */
711 				break;
712 			} else {
713 				(void) bufq_get(sc->sc_bufq);
714 				bp->b_error = error;
715 				bp->b_resid = bp->b_bcount;
716 				mutex_exit(&sc->sc_mutex);
717 				biodone(bp);
718 				mutex_enter(&sc->sc_mutex);
719 			}
720 		}
721 	}
722 
723 	mutex_exit(&sc->sc_mutex);
724 }
725 
726 void
727 lddone(struct ld_softc *sc, struct buf *bp)
728 {
729 
730 	if (bp->b_error != 0) {
731 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
732 		printf("\n");
733 	}
734 
735 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
736 	    (bp->b_flags & B_READ));
737 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
738 	biodone(bp);
739 
740 	mutex_enter(&sc->sc_mutex);
741 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
742 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
743 			sc->sc_flags &= ~LDF_DRAIN;
744 			wakeup(&sc->sc_queuecnt);
745 		}
746 		mutex_exit(&sc->sc_mutex);
747 		ldstart(sc, NULL);
748 	} else
749 		mutex_exit(&sc->sc_mutex);
750 }
751 
752 static int
753 ldsize(dev_t dev)
754 {
755 	struct ld_softc *sc;
756 	int part, unit, omask, size;
757 
758 	unit = DISKUNIT(dev);
759 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
760 		return (ENODEV);
761 	if ((sc->sc_flags & LDF_ENABLED) == 0)
762 		return (ENODEV);
763 	part = DISKPART(dev);
764 
765 	omask = sc->sc_dk.dk_openmask & (1 << part);
766 
767 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
768 		return (-1);
769 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
770 		size = -1;
771 	else
772 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
773 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
774 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
775 		return (-1);
776 
777 	return (size);
778 }
779 
780 /*
781  * Load the label information from the specified device.
782  */
783 static void
784 ldgetdisklabel(struct ld_softc *sc)
785 {
786 	const char *errstring;
787 
788 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
789 
790 	/* Call the generic disklabel extraction routine. */
791 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
792 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
793 	if (errstring != NULL)
794 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
795 
796 	/* In-core label now valid. */
797 	sc->sc_flags |= LDF_VLABEL;
798 }
799 
800 /*
801  * Construct a ficticious label.
802  */
803 static void
804 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
805 {
806 
807 	memset(lp, 0, sizeof(struct disklabel));
808 
809 	lp->d_secsize = sc->sc_secsize;
810 	lp->d_ntracks = sc->sc_nheads;
811 	lp->d_nsectors = sc->sc_nsectors;
812 	lp->d_ncylinders = sc->sc_ncylinders;
813 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
814 	lp->d_type = DTYPE_LD;
815 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
816 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
817 	lp->d_secperunit = sc->sc_secperunit;
818 	lp->d_rpm = 7200;
819 	lp->d_interleave = 1;
820 	lp->d_flags = 0;
821 
822 	lp->d_partitions[RAW_PART].p_offset = 0;
823 	lp->d_partitions[RAW_PART].p_size =
824 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
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(lp);
831 }
832 
833 /*
834  * Take a dump.
835  */
836 static int
837 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
838 {
839 	char *va = vav;
840 	struct ld_softc *sc;
841 	struct disklabel *lp;
842 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
843 	static int dumping;
844 
845 	unit = DISKUNIT(dev);
846 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
847 		return (ENXIO);
848 	if ((sc->sc_flags & LDF_ENABLED) == 0)
849 		return (ENODEV);
850 	if (sc->sc_dump == NULL)
851 		return (ENXIO);
852 
853 	/* Check if recursive dump; if so, punt. */
854 	if (dumping)
855 		return (EFAULT);
856 	dumping = 1;
857 
858 	/* Convert to disk sectors.  Request must be a multiple of size. */
859 	part = DISKPART(dev);
860 	lp = sc->sc_dk.dk_label;
861 	if ((size % lp->d_secsize) != 0)
862 		return (EFAULT);
863 	towrt = size / lp->d_secsize;
864 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
865 
866 	nsects = lp->d_partitions[part].p_size;
867 	sectoff = lp->d_partitions[part].p_offset;
868 
869 	/* Check transfer bounds against partition size. */
870 	if ((blkno < 0) || ((blkno + towrt) > nsects))
871 		return (EINVAL);
872 
873 	/* Offset block number to start of partition. */
874 	blkno += sectoff;
875 
876 	/* Start dumping and return when done. */
877 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
878 	while (towrt > 0) {
879 		nblk = min(maxblkcnt, towrt);
880 
881 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
882 			return (rv);
883 
884 		towrt -= nblk;
885 		blkno += nblk;
886 		va += nblk * sc->sc_secsize;
887 	}
888 
889 	dumping = 0;
890 	return (0);
891 }
892 
893 /*
894  * Adjust the size of a transfer.
895  */
896 static void
897 ldminphys(struct buf *bp)
898 {
899 	struct ld_softc *sc;
900 
901 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
902 
903 	if (bp->b_bcount > sc->sc_maxxfer)
904 		bp->b_bcount = sc->sc_maxxfer;
905 	minphys(bp);
906 }
907 
908 static void
909 ld_set_geometry(struct ld_softc *ld)
910 {
911 	struct disk_geom *dg = &ld->sc_dk.dk_geom;
912 
913 	memset(dg, 0, sizeof(*dg));
914 
915 	dg->dg_secperunit = ld->sc_secperunit;
916 	dg->dg_secsize = ld->sc_secsize;
917 	dg->dg_nsectors = ld->sc_nsectors;
918 	dg->dg_ntracks = ld->sc_nheads;
919 	dg->dg_ncylinders = ld->sc_ncylinders;
920 
921 	disk_set_info(ld->sc_dv, &ld->sc_dk, NULL);
922 }
923 
924 static void
925 ld_config_interrupts(device_t d)
926 {
927 	struct ld_softc *sc = device_private(d);
928 	dkwedge_discover(&sc->sc_dk);
929 }
930