xref: /netbsd-src/sys/dev/ld.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: ld.c,v 1.72 2014/03/16 05:20:26 dholland 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 1.72 2014/03/16 05:20:26 dholland 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_flag = D_DISK
92 };
93 
94 const struct cdevsw ld_cdevsw = {
95 	.d_open = ldopen,
96 	.d_close = ldclose,
97 	.d_read = ldread,
98 	.d_write = ldwrite,
99 	.d_ioctl = ldioctl,
100 	.d_stop = nostop,
101 	.d_tty = notty,
102 	.d_poll = nopoll,
103 	.d_mmap = nommap,
104 	.d_kqfilter = nokqfilter,
105 	.d_flag = D_DISK
106 };
107 
108 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
109 
110 void
111 ldattach(struct ld_softc *sc)
112 {
113 	char tbuf[9];
114 
115 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
116 
117 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
118 		aprint_normal_dev(sc->sc_dv, "disabled\n");
119 		return;
120 	}
121 
122 	/* Initialise and attach the disk structure. */
123 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
124 	disk_attach(&sc->sc_dk);
125 
126 	if (sc->sc_maxxfer > MAXPHYS)
127 		sc->sc_maxxfer = MAXPHYS;
128 
129 	/* Build synthetic geometry if necessary. */
130 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
131 	    sc->sc_ncylinders == 0) {
132 		uint64_t ncyl;
133 
134 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
135 			sc->sc_nheads = 16;
136 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
137 			sc->sc_nheads = 32;
138 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
139 			sc->sc_nheads = 64;
140 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
141 			sc->sc_nheads = 128;
142 		else
143 			sc->sc_nheads = 255;
144 
145 		sc->sc_nsectors = 63;
146 		sc->sc_ncylinders = INT_MAX;
147 		ncyl = sc->sc_secperunit /
148 		    (sc->sc_nheads * sc->sc_nsectors);
149 		if (ncyl < INT_MAX)
150 			sc->sc_ncylinders = (int)ncyl;
151 	}
152 
153 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
154 	    sc->sc_secsize);
155 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
156 	    "%d bytes/sect x %"PRIu64" sectors\n",
157 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
158 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
159 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
160 
161 	ld_set_geometry(sc);
162 
163 	/* Attach the device into the rnd source list. */
164 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
165 	    RND_TYPE_DISK, 0);
166 
167 	/* Register with PMF */
168 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
169 		aprint_error_dev(sc->sc_dv,
170 		    "couldn't establish power handler\n");
171 
172 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
173 
174 	/* Discover wedges on this disk. */
175 	config_interrupts(sc->sc_dv, ld_config_interrupts);
176 }
177 
178 int
179 ldadjqparam(struct ld_softc *sc, int xmax)
180 {
181 	int s;
182 
183 	s = splbio();
184 	sc->sc_maxqueuecnt = xmax;
185 	splx(s);
186 
187 	return (0);
188 }
189 
190 int
191 ldbegindetach(struct ld_softc *sc, int flags)
192 {
193 	int s, rv = 0;
194 
195 	if ((sc->sc_flags & LDF_ENABLED) == 0)
196 		return (0);
197 
198 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
199 
200 	if (rv != 0)
201 		return rv;
202 
203 	s = splbio();
204 	sc->sc_maxqueuecnt = 0;
205 	sc->sc_flags |= LDF_DETACH;
206 	while (sc->sc_queuecnt > 0) {
207 		sc->sc_flags |= LDF_DRAIN;
208 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
209 		if (rv)
210 			break;
211 	}
212 	splx(s);
213 
214 	return (rv);
215 }
216 
217 void
218 ldenddetach(struct ld_softc *sc)
219 {
220 	int s, bmaj, cmaj, i, mn;
221 
222 	if ((sc->sc_flags & LDF_ENABLED) == 0)
223 		return;
224 
225 	/* Wait for commands queued with the hardware to complete. */
226 	if (sc->sc_queuecnt != 0)
227 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
228 			printf("%s: not drained\n", device_xname(sc->sc_dv));
229 
230 	/* Locate the major numbers. */
231 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
232 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
233 
234 	/* Kill off any queued buffers. */
235 	s = splbio();
236 	bufq_drain(sc->sc_bufq);
237 	splx(s);
238 
239 	bufq_free(sc->sc_bufq);
240 
241 	/* Nuke the vnodes for any open instances. */
242 	for (i = 0; i < MAXPARTITIONS; i++) {
243 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
244 		vdevgone(bmaj, mn, mn, VBLK);
245 		vdevgone(cmaj, mn, mn, VCHR);
246 	}
247 
248 	/* Delete all of our wedges. */
249 	dkwedge_delall(&sc->sc_dk);
250 
251 	/* Detach from the disk list. */
252 	disk_detach(&sc->sc_dk);
253 	disk_destroy(&sc->sc_dk);
254 
255 	/* Unhook the entropy source. */
256 	rnd_detach_source(&sc->sc_rnd_source);
257 
258 	/* Deregister with PMF */
259 	pmf_device_deregister(sc->sc_dv);
260 
261 	/*
262 	 * XXX We can't really flush the cache here, beceause the
263 	 * XXX device may already be non-existent from the controller's
264 	 * XXX perspective.
265 	 */
266 #if 0
267 	/* Flush the device's cache. */
268 	if (sc->sc_flush != NULL)
269 		if ((*sc->sc_flush)(sc, 0) != 0)
270 			aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
271 #endif
272 	mutex_destroy(&sc->sc_mutex);
273 }
274 
275 /* ARGSUSED */
276 static bool
277 ld_suspend(device_t dev, const pmf_qual_t *qual)
278 {
279 	return ld_shutdown(dev, 0);
280 }
281 
282 /* ARGSUSED */
283 static bool
284 ld_shutdown(device_t dev, int flags)
285 {
286 	struct ld_softc *sc = device_private(dev);
287 
288 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
289 		printf("%s: unable to flush cache\n", device_xname(dev));
290 		return false;
291 	}
292 
293 	return true;
294 }
295 
296 /* ARGSUSED */
297 static int
298 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
299 {
300 	struct ld_softc *sc;
301 	int error, unit, part;
302 
303 	unit = DISKUNIT(dev);
304 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
305 		return (ENXIO);
306 	if ((sc->sc_flags & LDF_ENABLED) == 0)
307 		return (ENODEV);
308 	part = DISKPART(dev);
309 
310 	mutex_enter(&sc->sc_dk.dk_openlock);
311 
312 	if (sc->sc_dk.dk_openmask == 0) {
313 		/* Load the partition info if not already loaded. */
314 		if ((sc->sc_flags & LDF_VLABEL) == 0)
315 			ldgetdisklabel(sc);
316 	}
317 
318 	/* Check that the partition exists. */
319 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
320 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
321 		error = ENXIO;
322 		goto bad1;
323 	}
324 
325 	/* Ensure only one open at a time. */
326 	switch (fmt) {
327 	case S_IFCHR:
328 		sc->sc_dk.dk_copenmask |= (1 << part);
329 		break;
330 	case S_IFBLK:
331 		sc->sc_dk.dk_bopenmask |= (1 << part);
332 		break;
333 	}
334 	sc->sc_dk.dk_openmask =
335 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
336 
337 	error = 0;
338  bad1:
339 	mutex_exit(&sc->sc_dk.dk_openlock);
340 	return (error);
341 }
342 
343 static int
344 ldlastclose(device_t self)
345 {
346 	struct ld_softc *sc = device_private(self);
347 
348 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
349 		aprint_error_dev(self, "unable to flush cache\n");
350 	if ((sc->sc_flags & LDF_KLABEL) == 0)
351 		sc->sc_flags &= ~LDF_VLABEL;
352 
353 	return 0;
354 }
355 
356 /* ARGSUSED */
357 static int
358 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
359 {
360 	struct ld_softc *sc;
361 	int part, unit;
362 
363 	unit = DISKUNIT(dev);
364 	part = DISKPART(dev);
365 	sc = device_lookup_private(&ld_cd, unit);
366 
367 	mutex_enter(&sc->sc_dk.dk_openlock);
368 
369 	switch (fmt) {
370 	case S_IFCHR:
371 		sc->sc_dk.dk_copenmask &= ~(1 << part);
372 		break;
373 	case S_IFBLK:
374 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
375 		break;
376 	}
377 	sc->sc_dk.dk_openmask =
378 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
379 
380 	if (sc->sc_dk.dk_openmask == 0)
381 		ldlastclose(sc->sc_dv);
382 
383 	mutex_exit(&sc->sc_dk.dk_openlock);
384 	return (0);
385 }
386 
387 /* ARGSUSED */
388 static int
389 ldread(dev_t dev, struct uio *uio, int ioflag)
390 {
391 
392 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
393 }
394 
395 /* ARGSUSED */
396 static int
397 ldwrite(dev_t dev, struct uio *uio, int ioflag)
398 {
399 
400 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
401 }
402 
403 /* ARGSUSED */
404 static int
405 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
406 {
407 	struct ld_softc *sc;
408 	int part, unit, error;
409 #ifdef __HAVE_OLD_DISKLABEL
410 	struct disklabel newlabel;
411 #endif
412 	struct disklabel *lp;
413 
414 	unit = DISKUNIT(dev);
415 	part = DISKPART(dev);
416 	sc = device_lookup_private(&ld_cd, unit);
417 
418 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
419 	if (error != EPASSTHROUGH)
420 		return (error);
421 
422 	error = 0;
423 	switch (cmd) {
424 	case DIOCGDINFO:
425 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
426 		return (0);
427 
428 #ifdef __HAVE_OLD_DISKLABEL
429 	case ODIOCGDINFO:
430 		newlabel = *(sc->sc_dk.dk_label);
431 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
432 			return ENOTTY;
433 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
434 		return (0);
435 #endif
436 
437 	case DIOCGPART:
438 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
439 		((struct partinfo *)addr)->part =
440 		    &sc->sc_dk.dk_label->d_partitions[part];
441 		break;
442 
443 	case DIOCWDINFO:
444 	case DIOCSDINFO:
445 #ifdef __HAVE_OLD_DISKLABEL
446 	case ODIOCWDINFO:
447 	case ODIOCSDINFO:
448 
449 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
450 			memset(&newlabel, 0, sizeof newlabel);
451 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
452 			lp = &newlabel;
453 		} else
454 #endif
455 		lp = (struct disklabel *)addr;
456 
457 		if ((flag & FWRITE) == 0)
458 			return (EBADF);
459 
460 		mutex_enter(&sc->sc_dk.dk_openlock);
461 		sc->sc_flags |= LDF_LABELLING;
462 
463 		error = setdisklabel(sc->sc_dk.dk_label,
464 		    lp, /*sc->sc_dk.dk_openmask : */0,
465 		    sc->sc_dk.dk_cpulabel);
466 		if (error == 0 && (cmd == DIOCWDINFO
467 #ifdef __HAVE_OLD_DISKLABEL
468 		    || cmd == ODIOCWDINFO
469 #endif
470 		    ))
471 			error = writedisklabel(
472 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
473 			    ldstrategy, sc->sc_dk.dk_label,
474 			    sc->sc_dk.dk_cpulabel);
475 
476 		sc->sc_flags &= ~LDF_LABELLING;
477 		mutex_exit(&sc->sc_dk.dk_openlock);
478 		break;
479 
480 	case DIOCKLABEL:
481 		if ((flag & FWRITE) == 0)
482 			return (EBADF);
483 		if (*(int *)addr)
484 			sc->sc_flags |= LDF_KLABEL;
485 		else
486 			sc->sc_flags &= ~LDF_KLABEL;
487 		break;
488 
489 	case DIOCWLABEL:
490 		if ((flag & FWRITE) == 0)
491 			return (EBADF);
492 		if (*(int *)addr)
493 			sc->sc_flags |= LDF_WLABEL;
494 		else
495 			sc->sc_flags &= ~LDF_WLABEL;
496 		break;
497 
498 	case DIOCGDEFLABEL:
499 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
500 		break;
501 
502 #ifdef __HAVE_OLD_DISKLABEL
503 	case ODIOCGDEFLABEL:
504 		ldgetdefaultlabel(sc, &newlabel);
505 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
506 			return ENOTTY;
507 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
508 		break;
509 #endif
510 
511 	case DIOCCACHESYNC:
512 		/*
513 		 * XXX Do we really need to care about having a writable
514 		 * file descriptor here?
515 		 */
516 		if ((flag & FWRITE) == 0)
517 			error = EBADF;
518 		else if (sc->sc_flush)
519 			error = (*sc->sc_flush)(sc, 0);
520 		else
521 			error = 0;	/* XXX Error out instead? */
522 		break;
523 
524 	case DIOCAWEDGE:
525 	    {
526 	    	struct dkwedge_info *dkw = (void *) addr;
527 
528 		if ((flag & FWRITE) == 0)
529 			return (EBADF);
530 
531 		/* If the ioctl happens here, the parent is us. */
532 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
533 			sizeof(dkw->dkw_parent));
534 		return (dkwedge_add(dkw));
535 	    }
536 
537 	case DIOCDWEDGE:
538 	    {
539 	    	struct dkwedge_info *dkw = (void *) addr;
540 
541 		if ((flag & FWRITE) == 0)
542 			return (EBADF);
543 
544 		/* If the ioctl happens here, the parent is us. */
545 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
546 			sizeof(dkw->dkw_parent));
547 		return (dkwedge_del(dkw));
548 	    }
549 
550 	case DIOCLWEDGES:
551 	    {
552 	    	struct dkwedge_list *dkwl = (void *) addr;
553 
554 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
555 	    }
556 	case DIOCGSTRATEGY:
557 	    {
558 		struct disk_strategy *dks = (void *)addr;
559 
560 		mutex_enter(&sc->sc_mutex);
561 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
562 		    sizeof(dks->dks_name));
563 		mutex_exit(&sc->sc_mutex);
564 		dks->dks_paramlen = 0;
565 
566 		return 0;
567 	    }
568 	case DIOCSSTRATEGY:
569 	    {
570 		struct disk_strategy *dks = (void *)addr;
571 		struct bufq_state *new, *old;
572 
573 		if ((flag & FWRITE) == 0)
574 			return EPERM;
575 
576 		if (dks->dks_param != NULL)
577 			return EINVAL;
578 
579 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
580 		error = bufq_alloc(&new, dks->dks_name,
581 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
582 		if (error)
583 			return error;
584 
585 		mutex_enter(&sc->sc_mutex);
586 		old = sc->sc_bufq;
587 		bufq_move(new, old);
588 		sc->sc_bufq = new;
589 		mutex_exit(&sc->sc_mutex);
590 		bufq_free(old);
591 
592 		return 0;
593 	    }
594 	default:
595 		error = ENOTTY;
596 		break;
597 	}
598 
599 	return (error);
600 }
601 
602 static void
603 ldstrategy(struct buf *bp)
604 {
605 	struct ld_softc *sc;
606 	struct disklabel *lp;
607 	daddr_t blkno;
608 	int s, part;
609 
610 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
611 	part = DISKPART(bp->b_dev);
612 
613 	if ((sc->sc_flags & LDF_DETACH) != 0) {
614 		bp->b_error = EIO;
615 		goto done;
616 	}
617 
618 	lp = sc->sc_dk.dk_label;
619 
620 	/*
621 	 * The transfer must be a whole number of blocks and the offset must
622 	 * not be negative.
623 	 */
624 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
625 		bp->b_error = EINVAL;
626 		goto done;
627 	}
628 
629 	/* If it's a null transfer, return immediately. */
630 	if (bp->b_bcount == 0)
631 		goto done;
632 
633 	/*
634 	 * Do bounds checking and adjust the transfer.  If error, process.
635 	 * If past the end of partition, just return.
636 	 */
637 	if (part == RAW_PART) {
638 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
639 		    sc->sc_disksize512) <= 0)
640 			goto done;
641 	} else {
642 		if (bounds_check_with_label(&sc->sc_dk, bp,
643 		    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
644 			goto done;
645 	}
646 
647 	/*
648 	 * Convert the block number to absolute and put it in terms
649 	 * of the device's logical block size.
650 	 */
651 	if (lp->d_secsize == DEV_BSIZE)
652 		blkno = bp->b_blkno;
653 	else if (lp->d_secsize > DEV_BSIZE)
654 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
655 	else
656 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
657 
658 	if (part != RAW_PART)
659 		blkno += lp->d_partitions[part].p_offset;
660 
661 	bp->b_rawblkno = blkno;
662 
663 	s = splbio();
664 	ldstart(sc, bp);
665 	splx(s);
666 	return;
667 
668  done:
669 	bp->b_resid = bp->b_bcount;
670 	biodone(bp);
671 }
672 
673 static void
674 ldstart(struct ld_softc *sc, struct buf *bp)
675 {
676 	int error;
677 
678 	mutex_enter(&sc->sc_mutex);
679 
680 	if (bp != NULL)
681 		bufq_put(sc->sc_bufq, bp);
682 
683 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
684 		/* See if there is work to do. */
685 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
686 			break;
687 
688 		disk_busy(&sc->sc_dk);
689 		sc->sc_queuecnt++;
690 
691 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
692 			/*
693 			 * The back-end is running the job; remove it from
694 			 * the queue.
695 			 */
696 			(void) bufq_get(sc->sc_bufq);
697 		} else  {
698 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
699 			sc->sc_queuecnt--;
700 			if (error == EAGAIN) {
701 				/*
702 				 * Temporary resource shortage in the
703 				 * back-end; just defer the job until
704 				 * later.
705 				 *
706 				 * XXX We might consider a watchdog timer
707 				 * XXX to make sure we are kicked into action.
708 				 */
709 				break;
710 			} else {
711 				(void) bufq_get(sc->sc_bufq);
712 				bp->b_error = error;
713 				bp->b_resid = bp->b_bcount;
714 				mutex_exit(&sc->sc_mutex);
715 				biodone(bp);
716 				mutex_enter(&sc->sc_mutex);
717 			}
718 		}
719 	}
720 
721 	mutex_exit(&sc->sc_mutex);
722 }
723 
724 void
725 lddone(struct ld_softc *sc, struct buf *bp)
726 {
727 
728 	if (bp->b_error != 0) {
729 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
730 		printf("\n");
731 	}
732 
733 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
734 	    (bp->b_flags & B_READ));
735 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
736 	biodone(bp);
737 
738 	mutex_enter(&sc->sc_mutex);
739 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
740 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
741 			sc->sc_flags &= ~LDF_DRAIN;
742 			wakeup(&sc->sc_queuecnt);
743 		}
744 		mutex_exit(&sc->sc_mutex);
745 		ldstart(sc, NULL);
746 	} else
747 		mutex_exit(&sc->sc_mutex);
748 }
749 
750 static int
751 ldsize(dev_t dev)
752 {
753 	struct ld_softc *sc;
754 	int part, unit, omask, size;
755 
756 	unit = DISKUNIT(dev);
757 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
758 		return (ENODEV);
759 	if ((sc->sc_flags & LDF_ENABLED) == 0)
760 		return (ENODEV);
761 	part = DISKPART(dev);
762 
763 	omask = sc->sc_dk.dk_openmask & (1 << part);
764 
765 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
766 		return (-1);
767 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
768 		size = -1;
769 	else
770 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
771 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
772 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
773 		return (-1);
774 
775 	return (size);
776 }
777 
778 /*
779  * Load the label information from the specified device.
780  */
781 static void
782 ldgetdisklabel(struct ld_softc *sc)
783 {
784 	const char *errstring;
785 
786 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
787 
788 	/* Call the generic disklabel extraction routine. */
789 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
790 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
791 	if (errstring != NULL)
792 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
793 
794 	/* In-core label now valid. */
795 	sc->sc_flags |= LDF_VLABEL;
796 }
797 
798 /*
799  * Construct a ficticious label.
800  */
801 static void
802 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
803 {
804 
805 	memset(lp, 0, sizeof(struct disklabel));
806 
807 	lp->d_secsize = sc->sc_secsize;
808 	lp->d_ntracks = sc->sc_nheads;
809 	lp->d_nsectors = sc->sc_nsectors;
810 	lp->d_ncylinders = sc->sc_ncylinders;
811 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
812 	lp->d_type = DTYPE_LD;
813 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
814 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
815 	lp->d_secperunit = sc->sc_secperunit;
816 	lp->d_rpm = 7200;
817 	lp->d_interleave = 1;
818 	lp->d_flags = 0;
819 
820 	lp->d_partitions[RAW_PART].p_offset = 0;
821 	lp->d_partitions[RAW_PART].p_size =
822 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
823 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
824 	lp->d_npartitions = RAW_PART + 1;
825 
826 	lp->d_magic = DISKMAGIC;
827 	lp->d_magic2 = DISKMAGIC;
828 	lp->d_checksum = dkcksum(lp);
829 }
830 
831 /*
832  * Take a dump.
833  */
834 static int
835 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
836 {
837 	char *va = vav;
838 	struct ld_softc *sc;
839 	struct disklabel *lp;
840 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
841 	static int dumping;
842 
843 	unit = DISKUNIT(dev);
844 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
845 		return (ENXIO);
846 	if ((sc->sc_flags & LDF_ENABLED) == 0)
847 		return (ENODEV);
848 	if (sc->sc_dump == NULL)
849 		return (ENXIO);
850 
851 	/* Check if recursive dump; if so, punt. */
852 	if (dumping)
853 		return (EFAULT);
854 	dumping = 1;
855 
856 	/* Convert to disk sectors.  Request must be a multiple of size. */
857 	part = DISKPART(dev);
858 	lp = sc->sc_dk.dk_label;
859 	if ((size % lp->d_secsize) != 0)
860 		return (EFAULT);
861 	towrt = size / lp->d_secsize;
862 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
863 
864 	nsects = lp->d_partitions[part].p_size;
865 	sectoff = lp->d_partitions[part].p_offset;
866 
867 	/* Check transfer bounds against partition size. */
868 	if ((blkno < 0) || ((blkno + towrt) > nsects))
869 		return (EINVAL);
870 
871 	/* Offset block number to start of partition. */
872 	blkno += sectoff;
873 
874 	/* Start dumping and return when done. */
875 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
876 	while (towrt > 0) {
877 		nblk = min(maxblkcnt, towrt);
878 
879 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
880 			return (rv);
881 
882 		towrt -= nblk;
883 		blkno += nblk;
884 		va += nblk * sc->sc_secsize;
885 	}
886 
887 	dumping = 0;
888 	return (0);
889 }
890 
891 /*
892  * Adjust the size of a transfer.
893  */
894 static void
895 ldminphys(struct buf *bp)
896 {
897 	struct ld_softc *sc;
898 
899 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
900 
901 	if (bp->b_bcount > sc->sc_maxxfer)
902 		bp->b_bcount = sc->sc_maxxfer;
903 	minphys(bp);
904 }
905 
906 static void
907 ld_set_geometry(struct ld_softc *ld)
908 {
909 	struct disk_geom *dg = &ld->sc_dk.dk_geom;
910 
911 	memset(dg, 0, sizeof(*dg));
912 
913 	dg->dg_secperunit = ld->sc_secperunit;
914 	dg->dg_secsize = ld->sc_secsize;
915 	dg->dg_nsectors = ld->sc_nsectors;
916 	dg->dg_ntracks = ld->sc_nheads;
917 	dg->dg_ncylinders = ld->sc_ncylinders;
918 
919 	disk_set_info(ld->sc_dv, &ld->sc_dk, NULL);
920 }
921 
922 static void
923 ld_config_interrupts(device_t d)
924 {
925 	struct ld_softc *sc = device_private(d);
926 	dkwedge_discover(&sc->sc_dk);
927 }
928