xref: /netbsd-src/sys/dev/ld.c (revision 4b71a66d0f279143147d63ebfcfd8a59499a3684)
1 /*	$NetBSD: ld.c,v 1.58 2008/04/28 20:23:46 martin 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 1.58 2008/04/28 20:23:46 martin 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 }
260 
261 /* ARGSUSED */
262 static bool
263 ld_shutdown(device_t dev, int flags)
264 {
265 	struct ld_softc *sc = device_private(dev);
266 
267 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0) {
268 		printf("%s: unable to flush cache\n", device_xname(dev));
269 		return false;
270 	}
271 
272 	return true;
273 }
274 
275 /* ARGSUSED */
276 static int
277 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
278 {
279 	struct ld_softc *sc;
280 	int error, unit, part;
281 
282 	unit = DISKUNIT(dev);
283 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
284 		return (ENXIO);
285 	if ((sc->sc_flags & LDF_ENABLED) == 0)
286 		return (ENODEV);
287 	part = DISKPART(dev);
288 
289 	mutex_enter(&sc->sc_dk.dk_openlock);
290 
291 	if (sc->sc_dk.dk_openmask == 0) {
292 		/* Load the partition info if not already loaded. */
293 		if ((sc->sc_flags & LDF_VLABEL) == 0)
294 			ldgetdisklabel(sc);
295 	}
296 
297 	/* Check that the partition exists. */
298 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
299 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
300 		error = ENXIO;
301 		goto bad1;
302 	}
303 
304 	/* Ensure only one open at a time. */
305 	switch (fmt) {
306 	case S_IFCHR:
307 		sc->sc_dk.dk_copenmask |= (1 << part);
308 		break;
309 	case S_IFBLK:
310 		sc->sc_dk.dk_bopenmask |= (1 << part);
311 		break;
312 	}
313 	sc->sc_dk.dk_openmask =
314 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
315 
316 	error = 0;
317  bad1:
318 	mutex_exit(&sc->sc_dk.dk_openlock);
319 	return (error);
320 }
321 
322 /* ARGSUSED */
323 static int
324 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
325 {
326 	struct ld_softc *sc;
327 	int part, unit;
328 
329 	unit = DISKUNIT(dev);
330 	part = DISKPART(dev);
331 	sc = device_lookup(&ld_cd, unit);
332 
333 	mutex_enter(&sc->sc_dk.dk_openlock);
334 
335 	switch (fmt) {
336 	case S_IFCHR:
337 		sc->sc_dk.dk_copenmask &= ~(1 << part);
338 		break;
339 	case S_IFBLK:
340 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
341 		break;
342 	}
343 	sc->sc_dk.dk_openmask =
344 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
345 
346 	if (sc->sc_dk.dk_openmask == 0) {
347 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
348 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
349 		if ((sc->sc_flags & LDF_KLABEL) == 0)
350 			sc->sc_flags &= ~LDF_VLABEL;
351 	}
352 
353 	mutex_exit(&sc->sc_dk.dk_openlock);
354 	return (0);
355 }
356 
357 /* ARGSUSED */
358 static int
359 ldread(dev_t dev, struct uio *uio, int ioflag)
360 {
361 
362 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
363 }
364 
365 /* ARGSUSED */
366 static int
367 ldwrite(dev_t dev, struct uio *uio, int ioflag)
368 {
369 
370 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
371 }
372 
373 /* ARGSUSED */
374 static int
375 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
376 {
377 	struct ld_softc *sc;
378 	int part, unit, error;
379 #ifdef __HAVE_OLD_DISKLABEL
380 	struct disklabel newlabel;
381 #endif
382 	struct disklabel *lp;
383 
384 	unit = DISKUNIT(dev);
385 	part = DISKPART(dev);
386 	sc = device_lookup(&ld_cd, unit);
387 
388 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
389 	if (error != EPASSTHROUGH)
390 		return (error);
391 
392 	error = 0;
393 	switch (cmd) {
394 	case DIOCGDINFO:
395 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
396 		return (0);
397 
398 #ifdef __HAVE_OLD_DISKLABEL
399 	case ODIOCGDINFO:
400 		newlabel = *(sc->sc_dk.dk_label);
401 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
402 			return ENOTTY;
403 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
404 		return (0);
405 #endif
406 
407 	case DIOCGPART:
408 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
409 		((struct partinfo *)addr)->part =
410 		    &sc->sc_dk.dk_label->d_partitions[part];
411 		break;
412 
413 	case DIOCWDINFO:
414 	case DIOCSDINFO:
415 #ifdef __HAVE_OLD_DISKLABEL
416 	case ODIOCWDINFO:
417 	case ODIOCSDINFO:
418 
419 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
420 			memset(&newlabel, 0, sizeof newlabel);
421 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
422 			lp = &newlabel;
423 		} else
424 #endif
425 		lp = (struct disklabel *)addr;
426 
427 		if ((flag & FWRITE) == 0)
428 			return (EBADF);
429 
430 		mutex_enter(&sc->sc_dk.dk_openlock);
431 		sc->sc_flags |= LDF_LABELLING;
432 
433 		error = setdisklabel(sc->sc_dk.dk_label,
434 		    lp, /*sc->sc_dk.dk_openmask : */0,
435 		    sc->sc_dk.dk_cpulabel);
436 		if (error == 0 && (cmd == DIOCWDINFO
437 #ifdef __HAVE_OLD_DISKLABEL
438 		    || cmd == ODIOCWDINFO
439 #endif
440 		    ))
441 			error = writedisklabel(
442 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
443 			    ldstrategy, sc->sc_dk.dk_label,
444 			    sc->sc_dk.dk_cpulabel);
445 
446 		sc->sc_flags &= ~LDF_LABELLING;
447 		mutex_exit(&sc->sc_dk.dk_openlock);
448 		break;
449 
450 	case DIOCKLABEL:
451 		if ((flag & FWRITE) == 0)
452 			return (EBADF);
453 		if (*(int *)addr)
454 			sc->sc_flags |= LDF_KLABEL;
455 		else
456 			sc->sc_flags &= ~LDF_KLABEL;
457 		break;
458 
459 	case DIOCWLABEL:
460 		if ((flag & FWRITE) == 0)
461 			return (EBADF);
462 		if (*(int *)addr)
463 			sc->sc_flags |= LDF_WLABEL;
464 		else
465 			sc->sc_flags &= ~LDF_WLABEL;
466 		break;
467 
468 	case DIOCGDEFLABEL:
469 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
470 		break;
471 
472 #ifdef __HAVE_OLD_DISKLABEL
473 	case ODIOCGDEFLABEL:
474 		ldgetdefaultlabel(sc, &newlabel);
475 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
476 			return ENOTTY;
477 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
478 		break;
479 #endif
480 
481 	case DIOCCACHESYNC:
482 		/*
483 		 * XXX Do we really need to care about having a writable
484 		 * file descriptor here?
485 		 */
486 		if ((flag & FWRITE) == 0)
487 			error = EBADF;
488 		else if (sc->sc_flush)
489 			error = (*sc->sc_flush)(sc);
490 		else
491 			error = 0;	/* XXX Error out instead? */
492 		break;
493 
494 	case DIOCAWEDGE:
495 	    {
496 	    	struct dkwedge_info *dkw = (void *) addr;
497 
498 		if ((flag & FWRITE) == 0)
499 			return (EBADF);
500 
501 		/* If the ioctl happens here, the parent is us. */
502 		strlcpy(dkw->dkw_parent, device_xname(&sc->sc_dv),
503 			sizeof(dkw->dkw_parent));
504 		return (dkwedge_add(dkw));
505 	    }
506 
507 	case DIOCDWEDGE:
508 	    {
509 	    	struct dkwedge_info *dkw = (void *) addr;
510 
511 		if ((flag & FWRITE) == 0)
512 			return (EBADF);
513 
514 		/* If the ioctl happens here, the parent is us. */
515 		strlcpy(dkw->dkw_parent, device_xname(&sc->sc_dv),
516 			sizeof(dkw->dkw_parent));
517 		return (dkwedge_del(dkw));
518 	    }
519 
520 	case DIOCLWEDGES:
521 	    {
522 	    	struct dkwedge_list *dkwl = (void *) addr;
523 
524 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
525 	    }
526 	case DIOCGSTRATEGY:
527 	    {
528 		struct disk_strategy *dks = (void *)addr;
529 
530 		mutex_enter(&sc->sc_mutex);
531 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
532 		    sizeof(dks->dks_name));
533 		mutex_exit(&sc->sc_mutex);
534 		dks->dks_paramlen = 0;
535 
536 		return 0;
537 	    }
538 	case DIOCSSTRATEGY:
539 	    {
540 		struct disk_strategy *dks = (void *)addr;
541 		struct bufq_state *new, *old;
542 
543 		if ((flag & FWRITE) == 0)
544 			return EPERM;
545 
546 		if (dks->dks_param != NULL)
547 			return EINVAL;
548 
549 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
550 		error = bufq_alloc(&new, dks->dks_name,
551 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
552 		if (error)
553 			return error;
554 
555 		mutex_enter(&sc->sc_mutex);
556 		old = sc->sc_bufq;
557 		bufq_move(new, old);
558 		sc->sc_bufq = new;
559 		mutex_exit(&sc->sc_mutex);
560 		bufq_free(old);
561 
562 		return 0;
563 	    }
564 	default:
565 		error = ENOTTY;
566 		break;
567 	}
568 
569 	return (error);
570 }
571 
572 static void
573 ldstrategy(struct buf *bp)
574 {
575 	struct ld_softc *sc;
576 	struct disklabel *lp;
577 	daddr_t blkno;
578 	int s, part;
579 
580 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
581 	part = DISKPART(bp->b_dev);
582 
583 	if ((sc->sc_flags & LDF_DETACH) != 0) {
584 		bp->b_error = EIO;
585 		goto done;
586 	}
587 
588 	lp = sc->sc_dk.dk_label;
589 
590 	/*
591 	 * The transfer must be a whole number of blocks and the offset must
592 	 * not be negative.
593 	 */
594 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
595 		bp->b_error = EINVAL;
596 		goto done;
597 	}
598 
599 	/* If it's a null transfer, return immediately. */
600 	if (bp->b_bcount == 0)
601 		goto done;
602 
603 	/*
604 	 * Do bounds checking and adjust the transfer.  If error, process.
605 	 * If past the end of partition, just return.
606 	 */
607 	if (part != RAW_PART &&
608 	    bounds_check_with_label(&sc->sc_dk, bp,
609 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
610 		goto done;
611 	}
612 
613 	/*
614 	 * Convert the block number to absolute and put it in terms
615 	 * of the device's logical block size.
616 	 */
617 	if (lp->d_secsize == DEV_BSIZE)
618 		blkno = bp->b_blkno;
619 	else if (lp->d_secsize > DEV_BSIZE)
620 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
621 	else
622 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
623 
624 	if (part != RAW_PART)
625 		blkno += lp->d_partitions[part].p_offset;
626 
627 	bp->b_rawblkno = blkno;
628 
629 	s = splbio();
630 	ldstart(sc, bp);
631 	splx(s);
632 	return;
633 
634  done:
635 	bp->b_resid = bp->b_bcount;
636 	biodone(bp);
637 }
638 
639 static void
640 ldstart(struct ld_softc *sc, struct buf *bp)
641 {
642 	int error;
643 
644 	mutex_enter(&sc->sc_mutex);
645 
646 	if (bp != NULL)
647 		BUFQ_PUT(sc->sc_bufq, bp);
648 
649 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
650 		/* See if there is work to do. */
651 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
652 			break;
653 
654 		disk_busy(&sc->sc_dk);
655 		sc->sc_queuecnt++;
656 
657 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
658 			/*
659 			 * The back-end is running the job; remove it from
660 			 * the queue.
661 			 */
662 			(void) BUFQ_GET(sc->sc_bufq);
663 		} else  {
664 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
665 			sc->sc_queuecnt--;
666 			if (error == EAGAIN) {
667 				/*
668 				 * Temporary resource shortage in the
669 				 * back-end; just defer the job until
670 				 * later.
671 				 *
672 				 * XXX We might consider a watchdog timer
673 				 * XXX to make sure we are kicked into action.
674 				 */
675 				break;
676 			} else {
677 				(void) BUFQ_GET(sc->sc_bufq);
678 				bp->b_error = error;
679 				bp->b_resid = bp->b_bcount;
680 				mutex_exit(&sc->sc_mutex);
681 				biodone(bp);
682 				mutex_enter(&sc->sc_mutex);
683 			}
684 		}
685 	}
686 
687 	mutex_exit(&sc->sc_mutex);
688 }
689 
690 void
691 lddone(struct ld_softc *sc, struct buf *bp)
692 {
693 
694 	if (bp->b_error != 0) {
695 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
696 		printf("\n");
697 	}
698 
699 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
700 	    (bp->b_flags & B_READ));
701 #if NRND > 0
702 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
703 #endif
704 	biodone(bp);
705 
706 	mutex_enter(&sc->sc_mutex);
707 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
708 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
709 			sc->sc_flags &= ~LDF_DRAIN;
710 			wakeup(&sc->sc_queuecnt);
711 		}
712 		mutex_exit(&sc->sc_mutex);
713 		ldstart(sc, NULL);
714 	} else
715 		mutex_exit(&sc->sc_mutex);
716 }
717 
718 static int
719 ldsize(dev_t dev)
720 {
721 	struct ld_softc *sc;
722 	int part, unit, omask, size;
723 
724 	unit = DISKUNIT(dev);
725 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
726 		return (ENODEV);
727 	if ((sc->sc_flags & LDF_ENABLED) == 0)
728 		return (ENODEV);
729 	part = DISKPART(dev);
730 
731 	omask = sc->sc_dk.dk_openmask & (1 << part);
732 
733 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
734 		return (-1);
735 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
736 		size = -1;
737 	else
738 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
739 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
740 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
741 		return (-1);
742 
743 	return (size);
744 }
745 
746 /*
747  * Load the label information from the specified device.
748  */
749 static void
750 ldgetdisklabel(struct ld_softc *sc)
751 {
752 	const char *errstring;
753 
754 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
755 
756 	/* Call the generic disklabel extraction routine. */
757 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
758 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
759 	if (errstring != NULL)
760 		printf("%s: %s\n", device_xname(&sc->sc_dv), errstring);
761 
762 	/* In-core label now valid. */
763 	sc->sc_flags |= LDF_VLABEL;
764 }
765 
766 /*
767  * Construct a ficticious label.
768  */
769 static void
770 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
771 {
772 
773 	memset(lp, 0, sizeof(struct disklabel));
774 
775 	lp->d_secsize = sc->sc_secsize;
776 	lp->d_ntracks = sc->sc_nheads;
777 	lp->d_nsectors = sc->sc_nsectors;
778 	lp->d_ncylinders = sc->sc_ncylinders;
779 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
780 	lp->d_type = DTYPE_LD;
781 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
782 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
783 	lp->d_secperunit = sc->sc_secperunit;
784 	lp->d_rpm = 7200;
785 	lp->d_interleave = 1;
786 	lp->d_flags = 0;
787 
788 	lp->d_partitions[RAW_PART].p_offset = 0;
789 	lp->d_partitions[RAW_PART].p_size =
790 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
791 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
792 	lp->d_npartitions = RAW_PART + 1;
793 
794 	lp->d_magic = DISKMAGIC;
795 	lp->d_magic2 = DISKMAGIC;
796 	lp->d_checksum = dkcksum(lp);
797 }
798 
799 /*
800  * Take a dump.
801  */
802 static int
803 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
804 {
805 	char *va = vav;
806 	struct ld_softc *sc;
807 	struct disklabel *lp;
808 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
809 	static int dumping;
810 
811 	unit = DISKUNIT(dev);
812 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
813 		return (ENXIO);
814 	if ((sc->sc_flags & LDF_ENABLED) == 0)
815 		return (ENODEV);
816 	if (sc->sc_dump == NULL)
817 		return (ENXIO);
818 
819 	/* Check if recursive dump; if so, punt. */
820 	if (dumping)
821 		return (EFAULT);
822 	dumping = 1;
823 
824 	/* Convert to disk sectors.  Request must be a multiple of size. */
825 	part = DISKPART(dev);
826 	lp = sc->sc_dk.dk_label;
827 	if ((size % lp->d_secsize) != 0)
828 		return (EFAULT);
829 	towrt = size / lp->d_secsize;
830 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
831 
832 	nsects = lp->d_partitions[part].p_size;
833 	sectoff = lp->d_partitions[part].p_offset;
834 
835 	/* Check transfer bounds against partition size. */
836 	if ((blkno < 0) || ((blkno + towrt) > nsects))
837 		return (EINVAL);
838 
839 	/* Offset block number to start of partition. */
840 	blkno += sectoff;
841 
842 	/* Start dumping and return when done. */
843 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
844 	while (towrt > 0) {
845 		nblk = min(maxblkcnt, towrt);
846 
847 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
848 			return (rv);
849 
850 		towrt -= nblk;
851 		blkno += nblk;
852 		va += nblk * sc->sc_secsize;
853 	}
854 
855 	dumping = 0;
856 	return (0);
857 }
858 
859 /*
860  * Adjust the size of a transfer.
861  */
862 static void
863 ldminphys(struct buf *bp)
864 {
865 	struct ld_softc *sc;
866 
867 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
868 
869 	if (bp->b_bcount > sc->sc_maxxfer)
870 		bp->b_bcount = sc->sc_maxxfer;
871 	minphys(bp);
872 }
873 
874 static void
875 ld_set_properties(struct ld_softc *ld)
876 {
877 	prop_dictionary_t disk_info, odisk_info, geom;
878 
879 	disk_info = prop_dictionary_create();
880 
881 	geom = prop_dictionary_create();
882 
883 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
884 	    ld->sc_secperunit);
885 
886 	prop_dictionary_set_uint32(geom, "sector-size",
887 	    ld->sc_secsize);
888 
889 	prop_dictionary_set_uint16(geom, "sectors-per-track",
890 	    ld->sc_nsectors);
891 
892 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
893 	    ld->sc_nheads);
894 
895 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
896 	    ld->sc_ncylinders);
897 
898 	prop_dictionary_set(disk_info, "geometry", geom);
899 	prop_object_release(geom);
900 
901 	prop_dictionary_set(device_properties(&ld->sc_dv),
902 	    "disk-info", disk_info);
903 
904 	/*
905 	 * Don't release disk_info here; we keep a reference to it.
906 	 * disk_detach() will release it when we go away.
907 	 */
908 
909 	odisk_info = ld->sc_dk.dk_info;
910 	ld->sc_dk.dk_info = disk_info;
911 	if (odisk_info)
912 		prop_object_release(odisk_info);
913 }
914 
915 static void
916 ld_config_interrupts (struct device *d)
917 {
918 	struct ld_softc *sc = (struct ld_softc *)d;
919 	dkwedge_discover(&sc->sc_dk);
920 }
921