xref: /netbsd-src/sys/dev/ld.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: ld.c,v 1.39 2005/12/11 12:20:53 christos 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Disk driver for use by RAID controllers.
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.39 2005/12/11 12:20:53 christos Exp $");
45 
46 #include "rnd.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/device.h>
52 #include <sys/queue.h>
53 #include <sys/proc.h>
54 #include <sys/buf.h>
55 #include <sys/bufq.h>
56 #include <sys/endian.h>
57 #include <sys/disklabel.h>
58 #include <sys/disk.h>
59 #include <sys/dkio.h>
60 #include <sys/stat.h>
61 #include <sys/lock.h>
62 #include <sys/conf.h>
63 #include <sys/fcntl.h>
64 #include <sys/vnode.h>
65 #include <sys/syslog.h>
66 #if NRND > 0
67 #include <sys/rnd.h>
68 #endif
69 
70 #include <dev/ldvar.h>
71 
72 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
73 static void	ldgetdisklabel(struct ld_softc *);
74 static void	ldminphys(struct buf *bp);
75 static void	ldshutdown(void *);
76 static void	ldstart(struct ld_softc *);
77 
78 extern struct	cfdriver ld_cd;
79 
80 static dev_type_open(ldopen);
81 static dev_type_close(ldclose);
82 static dev_type_read(ldread);
83 static dev_type_write(ldwrite);
84 static dev_type_ioctl(ldioctl);
85 static dev_type_strategy(ldstrategy);
86 static dev_type_dump(lddump);
87 static dev_type_size(ldsize);
88 
89 const struct bdevsw ld_bdevsw = {
90 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
91 };
92 
93 const struct cdevsw ld_cdevsw = {
94 	ldopen, ldclose, ldread, ldwrite, ldioctl,
95 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
96 };
97 
98 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
99 static void	*ld_sdh;
100 
101 void
102 ldattach(struct ld_softc *sc)
103 {
104 	char tbuf[9];
105 
106 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
107 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
108 		return;
109 	}
110 
111 	/* Initialise and attach the disk structure. */
112 	sc->sc_dk.dk_driver = &lddkdriver;
113 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
114 	disk_attach(&sc->sc_dk);
115 
116 	if (sc->sc_maxxfer > MAXPHYS)
117 		sc->sc_maxxfer = MAXPHYS;
118 
119 	/* Build synthetic geometry if necessary. */
120 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
121 	    sc->sc_ncylinders == 0) {
122 		uint64_t ncyl;
123 
124 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
125 			sc->sc_nheads = 16;
126 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
127 			sc->sc_nheads = 32;
128 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
129 			sc->sc_nheads = 64;
130 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
131 			sc->sc_nheads = 128;
132 		else
133 			sc->sc_nheads = 255;
134 
135 		sc->sc_nsectors = 63;
136 		sc->sc_ncylinders = INT_MAX;
137 		ncyl = sc->sc_secperunit /
138 		    (sc->sc_nheads * sc->sc_nsectors);
139 		if (ncyl < INT_MAX)
140 			sc->sc_ncylinders = (int)ncyl;
141 	}
142 
143 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
144 	    sc->sc_secsize);
145 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
146 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
147 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
148 
149 #if NRND > 0
150 	/* Attach the device into the rnd source list. */
151 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
152 	    RND_TYPE_DISK, 0);
153 #endif
154 
155 	/* Set the `shutdownhook'. */
156 	if (ld_sdh == NULL)
157 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
158 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
159 
160 	/* Discover wedges on this disk. */
161 	dkwedge_discover(&sc->sc_dk);
162 }
163 
164 int
165 ldadjqparam(struct ld_softc *sc, int xmax)
166 {
167 	int s;
168 
169 	s = splbio();
170 	sc->sc_maxqueuecnt = xmax;
171 	splx(s);
172 
173 	return (0);
174 }
175 
176 int
177 ldbegindetach(struct ld_softc *sc, int flags)
178 {
179 	int s, rv = 0;
180 
181 	if ((sc->sc_flags & LDF_ENABLED) == 0)
182 		return (0);
183 
184 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
185 		return (EBUSY);
186 
187 	s = splbio();
188 	sc->sc_maxqueuecnt = 0;
189 	sc->sc_flags |= LDF_DETACH;
190 	while (sc->sc_queuecnt > 0) {
191 		sc->sc_flags |= LDF_DRAIN;
192 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
193 		if (rv)
194 			break;
195 	}
196 	splx(s);
197 
198 	return (rv);
199 }
200 
201 void
202 ldenddetach(struct ld_softc *sc)
203 {
204 	int s, bmaj, cmaj, i, mn;
205 
206 	if ((sc->sc_flags & LDF_ENABLED) == 0)
207 		return;
208 
209 	/* Wait for commands queued with the hardware to complete. */
210 	if (sc->sc_queuecnt != 0)
211 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
212 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
213 
214 	/* Locate the major numbers. */
215 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
216 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
217 
218 	/* Kill off any queued buffers. */
219 	s = splbio();
220 	bufq_drain(sc->sc_bufq);
221 	splx(s);
222 
223 	bufq_free(sc->sc_bufq);
224 
225 	/* Nuke the vnodes for any open instances. */
226 	for (i = 0; i < MAXPARTITIONS; i++) {
227 		mn = DISKMINOR(sc->sc_dv.dv_unit, i);
228 		vdevgone(bmaj, mn, mn, VBLK);
229 		vdevgone(cmaj, mn, mn, VCHR);
230 	}
231 
232 	/* Delete all of our wedges. */
233 	dkwedge_delall(&sc->sc_dk);
234 
235 	/* Detach from the disk list. */
236 	disk_detach(&sc->sc_dk);
237 
238 #if NRND > 0
239 	/* Unhook the entropy source. */
240 	rnd_detach_source(&sc->sc_rnd_source);
241 #endif
242 
243 	/*
244 	 * XXX We can't really flush the cache here, beceause the
245 	 * XXX device may already be non-existent from the controller's
246 	 * XXX perspective.
247 	 */
248 #if 0
249 	/* Flush the device's cache. */
250 	if (sc->sc_flush != NULL)
251 		if ((*sc->sc_flush)(sc) != 0)
252 			printf("%s: unable to flush cache\n",
253 			    sc->sc_dv.dv_xname);
254 #endif
255 }
256 
257 /* ARGSUSED */
258 static void
259 ldshutdown(void *cookie)
260 {
261 	struct ld_softc *sc;
262 	int i;
263 
264 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
265 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
266 			continue;
267 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
268 			printf("%s: unable to flush cache\n",
269 			    sc->sc_dv.dv_xname);
270 	}
271 }
272 
273 /* ARGSUSED */
274 static int
275 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
276 {
277 	struct ld_softc *sc;
278 	int error, unit, part;
279 
280 	unit = DISKUNIT(dev);
281 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
282 		return (ENXIO);
283 	if ((sc->sc_flags & LDF_ENABLED) == 0)
284 		return (ENODEV);
285 	part = DISKPART(dev);
286 
287 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
288 		return (error);
289 
290 	if (sc->sc_dk.dk_openmask == 0) {
291 		/* Load the partition info if not already loaded. */
292 		if ((sc->sc_flags & LDF_VLABEL) == 0)
293 			ldgetdisklabel(sc);
294 	}
295 
296 	/* Check that the partition exists. */
297 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
298 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
299 		error = ENXIO;
300 		goto bad1;
301 	}
302 
303 	/* Ensure only one open at a time. */
304 	switch (fmt) {
305 	case S_IFCHR:
306 		sc->sc_dk.dk_copenmask |= (1 << part);
307 		break;
308 	case S_IFBLK:
309 		sc->sc_dk.dk_bopenmask |= (1 << part);
310 		break;
311 	}
312 	sc->sc_dk.dk_openmask =
313 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
314 
315 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
316 	return (0);
317 
318  bad1:
319 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
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 error, part, unit;
329 
330 	unit = DISKUNIT(dev);
331 	part = DISKPART(dev);
332 	sc = device_lookup(&ld_cd, unit);
333 
334 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
335 		return (error);
336 
337 	switch (fmt) {
338 	case S_IFCHR:
339 		sc->sc_dk.dk_copenmask &= ~(1 << part);
340 		break;
341 	case S_IFBLK:
342 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
343 		break;
344 	}
345 	sc->sc_dk.dk_openmask =
346 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
347 
348 	if (sc->sc_dk.dk_openmask == 0) {
349 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
350 			printf("%s: unable to flush cache\n",
351 			    sc->sc_dv.dv_xname);
352 		if ((sc->sc_flags & LDF_KLABEL) == 0)
353 			sc->sc_flags &= ~LDF_VLABEL;
354 	}
355 
356 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
357 	return (0);
358 }
359 
360 /* ARGSUSED */
361 static int
362 ldread(dev_t dev, struct uio *uio, int ioflag)
363 {
364 
365 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
366 }
367 
368 /* ARGSUSED */
369 static int
370 ldwrite(dev_t dev, struct uio *uio, int ioflag)
371 {
372 
373 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
374 }
375 
376 /* ARGSUSED */
377 static int
378 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct lwp *l)
379 {
380 	struct ld_softc *sc;
381 	int part, unit, error;
382 #ifdef __HAVE_OLD_DISKLABEL
383 	struct disklabel newlabel;
384 #endif
385 	struct disklabel *lp;
386 
387 	unit = DISKUNIT(dev);
388 	part = DISKPART(dev);
389 	sc = device_lookup(&ld_cd, unit);
390 	error = 0;
391 
392 	switch (cmd) {
393 	case DIOCGDINFO:
394 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
395 		return (0);
396 
397 #ifdef __HAVE_OLD_DISKLABEL
398 	case ODIOCGDINFO:
399 		newlabel = *(sc->sc_dk.dk_label);
400 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
401 			return ENOTTY;
402 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
403 		return (0);
404 #endif
405 
406 	case DIOCGPART:
407 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
408 		((struct partinfo *)addr)->part =
409 		    &sc->sc_dk.dk_label->d_partitions[part];
410 		break;
411 
412 	case DIOCWDINFO:
413 	case DIOCSDINFO:
414 #ifdef __HAVE_OLD_DISKLABEL
415 	case ODIOCWDINFO:
416 	case ODIOCSDINFO:
417 
418 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
419 			memset(&newlabel, 0, sizeof newlabel);
420 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
421 			lp = &newlabel;
422 		} else
423 #endif
424 		lp = (struct disklabel *)addr;
425 
426 		if ((flag & FWRITE) == 0)
427 			return (EBADF);
428 
429 		if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE,
430 				     NULL)) != 0)
431 			return (error);
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 		(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
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 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
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 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
516 		return (dkwedge_del(dkw));
517 	    }
518 
519 	case DIOCLWEDGES:
520 	    {
521 	    	struct dkwedge_list *dkwl = (void *) addr;
522 
523 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
524 	    }
525 
526 	default:
527 		error = ENOTTY;
528 		break;
529 	}
530 
531 	return (error);
532 }
533 
534 static void
535 ldstrategy(struct buf *bp)
536 {
537 	struct ld_softc *sc;
538 	struct disklabel *lp;
539 	daddr_t blkno;
540 	int s, part;
541 
542 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
543 	part = DISKPART(bp->b_dev);
544 
545 	if ((sc->sc_flags & LDF_DETACH) != 0) {
546 		bp->b_error = EIO;
547 		goto bad;
548 	}
549 
550 	lp = sc->sc_dk.dk_label;
551 
552 	/*
553 	 * The transfer must be a whole number of blocks and the offset must
554 	 * not be negative.
555 	 */
556 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
557 		bp->b_error = EINVAL;
558 		goto bad;
559 	}
560 
561 	/* If it's a null transfer, return immediately. */
562 	if (bp->b_bcount == 0)
563 		goto done;
564 
565 	/*
566 	 * Do bounds checking and adjust the transfer.  If error, process.
567 	 * If past the end of partition, just return.
568 	 */
569 	if (part != RAW_PART &&
570 	    bounds_check_with_label(&sc->sc_dk, bp,
571 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
572 		goto done;
573 	}
574 
575 	/*
576 	 * Convert the block number to absolute and put it in terms
577 	 * of the device's logical block size.
578 	 */
579 	if (lp->d_secsize == DEV_BSIZE)
580 		blkno = bp->b_blkno;
581 	else if (lp->d_secsize > DEV_BSIZE)
582 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
583 	else
584 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
585 
586 	if (part != RAW_PART)
587 		blkno += lp->d_partitions[part].p_offset;
588 
589 	bp->b_rawblkno = blkno;
590 
591 	s = splbio();
592 	BUFQ_PUT(sc->sc_bufq, bp);
593 	ldstart(sc);
594 	splx(s);
595 	return;
596 
597  bad:
598 	bp->b_flags |= B_ERROR;
599  done:
600 	bp->b_resid = bp->b_bcount;
601 	biodone(bp);
602 }
603 
604 static void
605 ldstart(struct ld_softc *sc)
606 {
607 	struct buf *bp;
608 	int error;
609 
610 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
611 		/* See if there is work to do. */
612 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
613 			break;
614 
615 		disk_busy(&sc->sc_dk);
616 		sc->sc_queuecnt++;
617 
618 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
619 			/*
620 			 * The back-end is running the job; remove it from
621 			 * the queue.
622 			 */
623 			(void) BUFQ_GET(sc->sc_bufq);
624 		} else  {
625 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
626 			sc->sc_queuecnt--;
627 			if (error == EAGAIN) {
628 				/*
629 				 * Temporary resource shortage in the
630 				 * back-end; just defer the job until
631 				 * later.
632 				 *
633 				 * XXX We might consider a watchdog timer
634 				 * XXX to make sure we are kicked into action.
635 				 */
636 				break;
637 			} else {
638 				(void) BUFQ_GET(sc->sc_bufq);
639 				bp->b_error = error;
640 				bp->b_flags |= B_ERROR;
641 				bp->b_resid = bp->b_bcount;
642 				biodone(bp);
643 			}
644 		}
645 	}
646 }
647 
648 void
649 lddone(struct ld_softc *sc, struct buf *bp)
650 {
651 
652 	if ((bp->b_flags & B_ERROR) != 0) {
653 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
654 		printf("\n");
655 	}
656 
657 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
658 	    (bp->b_flags & B_READ));
659 #if NRND > 0
660 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
661 #endif
662 	biodone(bp);
663 
664 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
665 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
666 			sc->sc_flags &= ~LDF_DRAIN;
667 			wakeup(&sc->sc_queuecnt);
668 		}
669 		ldstart(sc);
670 	}
671 }
672 
673 static int
674 ldsize(dev_t dev)
675 {
676 	struct ld_softc *sc;
677 	int part, unit, omask, size;
678 
679 	unit = DISKUNIT(dev);
680 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
681 		return (ENODEV);
682 	if ((sc->sc_flags & LDF_ENABLED) == 0)
683 		return (ENODEV);
684 	part = DISKPART(dev);
685 
686 	omask = sc->sc_dk.dk_openmask & (1 << part);
687 
688 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
689 		return (-1);
690 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
691 		size = -1;
692 	else
693 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
694 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
695 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
696 		return (-1);
697 
698 	return (size);
699 }
700 
701 /*
702  * Load the label information from the specified device.
703  */
704 static void
705 ldgetdisklabel(struct ld_softc *sc)
706 {
707 	const char *errstring;
708 
709 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
710 
711 	/* Call the generic disklabel extraction routine. */
712 	errstring = readdisklabel(MAKEDISKDEV(0, sc->sc_dv.dv_unit, RAW_PART),
713 	    ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
714 	if (errstring != NULL)
715 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
716 
717 	/* In-core label now valid. */
718 	sc->sc_flags |= LDF_VLABEL;
719 }
720 
721 /*
722  * Construct a ficticious label.
723  */
724 static void
725 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
726 {
727 
728 	memset(lp, 0, sizeof(struct disklabel));
729 
730 	lp->d_secsize = sc->sc_secsize;
731 	lp->d_ntracks = sc->sc_nheads;
732 	lp->d_nsectors = sc->sc_nsectors;
733 	lp->d_ncylinders = sc->sc_ncylinders;
734 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
735 	lp->d_type = DTYPE_LD;
736 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
737 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
738 	lp->d_secperunit = sc->sc_secperunit;
739 	lp->d_rpm = 7200;
740 	lp->d_interleave = 1;
741 	lp->d_flags = 0;
742 
743 	lp->d_partitions[RAW_PART].p_offset = 0;
744 	lp->d_partitions[RAW_PART].p_size =
745 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
746 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
747 	lp->d_npartitions = RAW_PART + 1;
748 
749 	lp->d_magic = DISKMAGIC;
750 	lp->d_magic2 = DISKMAGIC;
751 	lp->d_checksum = dkcksum(lp);
752 }
753 
754 /*
755  * Take a dump.
756  */
757 static int
758 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
759 {
760 	struct ld_softc *sc;
761 	struct disklabel *lp;
762 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
763 	static int dumping;
764 
765 	unit = DISKUNIT(dev);
766 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
767 		return (ENXIO);
768 	if ((sc->sc_flags & LDF_ENABLED) == 0)
769 		return (ENODEV);
770 	if (sc->sc_dump == NULL)
771 		return (ENXIO);
772 
773 	/* Check if recursive dump; if so, punt. */
774 	if (dumping)
775 		return (EFAULT);
776 	dumping = 1;
777 
778 	/* Convert to disk sectors.  Request must be a multiple of size. */
779 	part = DISKPART(dev);
780 	lp = sc->sc_dk.dk_label;
781 	if ((size % lp->d_secsize) != 0)
782 		return (EFAULT);
783 	towrt = size / lp->d_secsize;
784 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
785 
786 	nsects = lp->d_partitions[part].p_size;
787 	sectoff = lp->d_partitions[part].p_offset;
788 
789 	/* Check transfer bounds against partition size. */
790 	if ((blkno < 0) || ((blkno + towrt) > nsects))
791 		return (EINVAL);
792 
793 	/* Offset block number to start of partition. */
794 	blkno += sectoff;
795 
796 	/* Start dumping and return when done. */
797 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
798 	while (towrt > 0) {
799 		nblk = min(maxblkcnt, towrt);
800 
801 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
802 			return (rv);
803 
804 		towrt -= nblk;
805 		blkno += nblk;
806 		va += nblk * sc->sc_secsize;
807 	}
808 
809 	dumping = 0;
810 	return (0);
811 }
812 
813 /*
814  * Adjust the size of a transfer.
815  */
816 static void
817 ldminphys(struct buf *bp)
818 {
819 	struct ld_softc *sc;
820 
821 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
822 
823 	if (bp->b_bcount > sc->sc_maxxfer)
824 		bp->b_bcount = sc->sc_maxxfer;
825 	minphys(bp);
826 }
827