Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.24
      1 /*	$NetBSD: ld.c,v 1.24 2003/06/13 02:32:27 thorpej 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 1.24 2003/06/13 02:32:27 thorpej 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/endian.h>
     56 #include <sys/disklabel.h>
     57 #include <sys/disk.h>
     58 #include <sys/dkio.h>
     59 #include <sys/stat.h>
     60 #include <sys/lock.h>
     61 #include <sys/conf.h>
     62 #include <sys/fcntl.h>
     63 #include <sys/vnode.h>
     64 #include <sys/syslog.h>
     65 #if NRND > 0
     66 #include <sys/rnd.h>
     67 #endif
     68 
     69 #include <dev/ldvar.h>
     70 
     71 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     72 static void	ldgetdisklabel(struct ld_softc *);
     73 static int	ldlock(struct ld_softc *);
     74 static void	ldminphys(struct buf *bp);
     75 static void	ldshutdown(void *);
     76 static void	ldstart(struct ld_softc *);
     77 static void	ldunlock(struct ld_softc *);
     78 
     79 extern struct	cfdriver ld_cd;
     80 
     81 dev_type_open(ldopen);
     82 dev_type_close(ldclose);
     83 dev_type_read(ldread);
     84 dev_type_write(ldwrite);
     85 dev_type_ioctl(ldioctl);
     86 dev_type_strategy(ldstrategy);
     87 dev_type_dump(lddump);
     88 dev_type_size(ldsize);
     89 
     90 const struct bdevsw ld_bdevsw = {
     91 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
     92 };
     93 
     94 const struct cdevsw ld_cdevsw = {
     95 	ldopen, ldclose, ldread, ldwrite, ldioctl,
     96 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
     97 };
     98 
     99 static struct	dkdriver lddkdriver = { ldstrategy };
    100 static void	*ld_sdh;
    101 
    102 void
    103 ldattach(struct ld_softc *sc)
    104 {
    105 	char buf[9];
    106 
    107 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    108 		printf("%s: disabled\n", sc->sc_dv.dv_xname);
    109 		return;
    110 	}
    111 
    112 	/* Initialise and attach the disk structure. */
    113 	sc->sc_dk.dk_driver = &lddkdriver;
    114 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
    115 	disk_attach(&sc->sc_dk);
    116 
    117 	if (sc->sc_maxxfer > MAXPHYS)
    118 		sc->sc_maxxfer = MAXPHYS;
    119 
    120 	/* Build synthetic geometry if necessary. */
    121 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    122 	    sc->sc_ncylinders == 0) {
    123 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    124 			sc->sc_nheads = 16;
    125 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    126 			sc->sc_nheads = 32;
    127 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    128 			sc->sc_nheads = 64;
    129 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    130 			sc->sc_nheads = 128;
    131 		else
    132 			sc->sc_nheads = 255;
    133 
    134 		sc->sc_nsectors = 63;
    135 		sc->sc_ncylinders = sc->sc_secperunit /
    136 		    (sc->sc_nheads * sc->sc_nsectors);
    137 	}
    138 
    139 	format_bytes(buf, sizeof(buf), (u_int64_t)sc->sc_secperunit *
    140 	    sc->sc_secsize);
    141 	printf("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %d sectors\n",
    142 	    sc->sc_dv.dv_xname, buf, sc->sc_ncylinders, sc->sc_nheads,
    143 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    144 
    145 #if NRND > 0
    146 	/* Attach the device into the rnd source list. */
    147 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    148 	    RND_TYPE_DISK, 0);
    149 #endif
    150 
    151 	/* Set the `shutdownhook'. */
    152 	if (ld_sdh == NULL)
    153 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
    154 	bufq_alloc(&sc->sc_bufq, BUFQ_FCFS);
    155 }
    156 
    157 int
    158 ldadjqparam(struct ld_softc *sc, int max)
    159 {
    160 	int s;
    161 
    162 	s = splbio();
    163 	sc->sc_maxqueuecnt = max;
    164 	splx(s);
    165 
    166 	return (0);
    167 }
    168 
    169 int
    170 ldbegindetach(struct ld_softc *sc, int flags)
    171 {
    172 	int s, rv = 0;
    173 
    174 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    175 		return (0);
    176 
    177 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    178 		return (EBUSY);
    179 
    180 	s = splbio();
    181 	sc->sc_maxqueuecnt = 0;
    182 	sc->sc_flags |= LDF_DETACH;
    183 	while (sc->sc_queuecnt > 0) {
    184 		sc->sc_flags |= LDF_DRAIN;
    185 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    186 		if (rv)
    187 			break;
    188 	}
    189 	splx(s);
    190 
    191 	return (rv);
    192 }
    193 
    194 void
    195 ldenddetach(struct ld_softc *sc)
    196 {
    197 	struct buf *bp;
    198 	int s, bmaj, cmaj, i, mn;
    199 
    200 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    201 		return;
    202 
    203 	/* Wait for commands queued with the hardware to complete. */
    204 	if (sc->sc_queuecnt != 0)
    205 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    206 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
    207 
    208 	/* Locate the major numbers. */
    209 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    210 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    211 
    212 	/* Kill off any queued buffers. */
    213 	s = splbio();
    214 	while ((bp = BUFQ_GET(&sc->sc_bufq)) != NULL) {
    215 		bp->b_error = EIO;
    216 		bp->b_flags |= B_ERROR;
    217 		bp->b_resid = bp->b_bcount;
    218 		biodone(bp);
    219 	}
    220 	bufq_free(&sc->sc_bufq);
    221 	splx(s);
    222 
    223 	/* Nuke the vnodes for any open instances. */
    224 	for (i = 0; i < MAXPARTITIONS; i++) {
    225 		mn = DISKMINOR(sc->sc_dv.dv_unit, i);
    226 		vdevgone(bmaj, mn, mn, VBLK);
    227 		vdevgone(cmaj, mn, mn, VCHR);
    228 	}
    229 
    230 	/* Detach from the disk list. */
    231 	disk_detach(&sc->sc_dk);
    232 
    233 #if NRND > 0
    234 	/* Unhook the entropy source. */
    235 	rnd_detach_source(&sc->sc_rnd_source);
    236 #endif
    237 
    238 	/*
    239 	 * XXX We can't really flush the cache here, beceause the
    240 	 * XXX device may already be non-existent from the controller's
    241 	 * XXX perspective.
    242 	 */
    243 #if 0
    244 	/* Flush the device's cache. */
    245 	if (sc->sc_flush != NULL)
    246 		if ((*sc->sc_flush)(sc) != 0)
    247 			printf("%s: unable to flush cache\n",
    248 			    sc->sc_dv.dv_xname);
    249 #endif
    250 }
    251 
    252 /* ARGSUSED */
    253 static void
    254 ldshutdown(void *cookie)
    255 {
    256 	struct ld_softc *sc;
    257 	int i;
    258 
    259 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    260 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    261 			continue;
    262 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    263 			printf("%s: unable to flush cache\n",
    264 			    sc->sc_dv.dv_xname);
    265 	}
    266 }
    267 
    268 /* ARGSUSED */
    269 int
    270 ldopen(dev_t dev, int flags, int fmt, struct proc *p)
    271 {
    272 	struct ld_softc *sc;
    273 	int unit, part;
    274 
    275 	unit = DISKUNIT(dev);
    276 	if ((sc = device_lookup(&ld_cd, unit))== NULL)
    277 		return (ENXIO);
    278 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    279 		return (ENODEV);
    280 	part = DISKPART(dev);
    281 	ldlock(sc);
    282 
    283 	if (sc->sc_dk.dk_openmask == 0) {
    284 		/* Load the partition info if not already loaded. */
    285 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    286 			ldgetdisklabel(sc);
    287 	}
    288 
    289 	/* Check that the partition exists. */
    290 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    291 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    292 	     	ldunlock(sc);
    293 		return (ENXIO);
    294 	}
    295 
    296 	/* Ensure only one open at a time. */
    297 	switch (fmt) {
    298 	case S_IFCHR:
    299 		sc->sc_dk.dk_copenmask |= (1 << part);
    300 		break;
    301 	case S_IFBLK:
    302 		sc->sc_dk.dk_bopenmask |= (1 << part);
    303 		break;
    304 	}
    305 	sc->sc_dk.dk_openmask =
    306 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    307 
    308 	ldunlock(sc);
    309 	return (0);
    310 }
    311 
    312 /* ARGSUSED */
    313 int
    314 ldclose(dev_t dev, int flags, int fmt, struct proc *p)
    315 {
    316 	struct ld_softc *sc;
    317 	int part, unit;
    318 
    319 	unit = DISKUNIT(dev);
    320 	part = DISKPART(dev);
    321 	sc = device_lookup(&ld_cd, unit);
    322 	ldlock(sc);
    323 
    324 	switch (fmt) {
    325 	case S_IFCHR:
    326 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    327 		break;
    328 	case S_IFBLK:
    329 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    330 		break;
    331 	}
    332 	sc->sc_dk.dk_openmask =
    333 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    334 
    335 	if (sc->sc_dk.dk_openmask == 0) {
    336 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    337 			printf("%s: unable to flush cache\n",
    338 			    sc->sc_dv.dv_xname);
    339 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    340 			sc->sc_flags &= ~LDF_VLABEL;
    341 	}
    342 
    343 	ldunlock(sc);
    344 	return (0);
    345 }
    346 
    347 /* ARGSUSED */
    348 int
    349 ldread(dev_t dev, struct uio *uio, int ioflag)
    350 {
    351 
    352 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    353 }
    354 
    355 /* ARGSUSED */
    356 int
    357 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    358 {
    359 
    360 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    361 }
    362 
    363 /* ARGSUSED */
    364 int
    365 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct proc *p)
    366 {
    367 	struct ld_softc *sc;
    368 	int part, unit, error;
    369 #ifdef __HAVE_OLD_DISKLABEL
    370 	struct disklabel newlabel;
    371 #endif
    372 	struct disklabel *lp;
    373 
    374 	unit = DISKUNIT(dev);
    375 	part = DISKPART(dev);
    376 	sc = device_lookup(&ld_cd, unit);
    377 	error = 0;
    378 
    379 	switch (cmd) {
    380 	case DIOCGDINFO:
    381 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    382 		return (0);
    383 
    384 #ifdef __HAVE_OLD_DISKLABEL
    385 	case ODIOCGDINFO:
    386 		newlabel = *(sc->sc_dk.dk_label);
    387 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    388 			return ENOTTY;
    389 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    390 		return (0);
    391 #endif
    392 
    393 	case DIOCGPART:
    394 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    395 		((struct partinfo *)addr)->part =
    396 		    &sc->sc_dk.dk_label->d_partitions[part];
    397 		break;
    398 
    399 	case DIOCWDINFO:
    400 	case DIOCSDINFO:
    401 #ifdef __HAVE_OLD_DISKLABEL
    402 	case ODIOCWDINFO:
    403 	case ODIOCSDINFO:
    404 
    405 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    406 			memset(&newlabel, 0, sizeof newlabel);
    407 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    408 			lp = &newlabel;
    409 		} else
    410 #endif
    411 		lp = (struct disklabel *)addr;
    412 
    413 		if ((flag & FWRITE) == 0)
    414 			return (EBADF);
    415 
    416 		if ((error = ldlock(sc)) != 0)
    417 			return (error);
    418 		sc->sc_flags |= LDF_LABELLING;
    419 
    420 		error = setdisklabel(sc->sc_dk.dk_label,
    421 		    lp, /*sc->sc_dk.dk_openmask : */0,
    422 		    sc->sc_dk.dk_cpulabel);
    423 		if (error == 0 && (cmd == DIOCWDINFO
    424 #ifdef __HAVE_OLD_DISKLABEL
    425 		    || cmd == ODIOCWDINFO
    426 #endif
    427 		    ))
    428 			error = writedisklabel(
    429 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    430 			    ldstrategy, sc->sc_dk.dk_label,
    431 			    sc->sc_dk.dk_cpulabel);
    432 
    433 		sc->sc_flags &= ~LDF_LABELLING;
    434 		ldunlock(sc);
    435 		break;
    436 
    437 	case DIOCKLABEL:
    438 		if ((flag & FWRITE) == 0)
    439 			return (EBADF);
    440 		if (*(int *)addr)
    441 			sc->sc_flags |= LDF_KLABEL;
    442 		else
    443 			sc->sc_flags &= ~LDF_KLABEL;
    444 		break;
    445 
    446 	case DIOCWLABEL:
    447 		if ((flag & FWRITE) == 0)
    448 			return (EBADF);
    449 		if (*(int *)addr)
    450 			sc->sc_flags |= LDF_WLABEL;
    451 		else
    452 			sc->sc_flags &= ~LDF_WLABEL;
    453 		break;
    454 
    455 	case DIOCGDEFLABEL:
    456 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    457 		break;
    458 
    459 #ifdef __HAVE_OLD_DISKLABEL
    460 	case ODIOCGDEFLABEL:
    461 		ldgetdefaultlabel(sc, &newlabel);
    462 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    463 			return ENOTTY;
    464 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    465 		break;
    466 #endif
    467 
    468 	default:
    469 		error = ENOTTY;
    470 		break;
    471 	}
    472 
    473 	return (error);
    474 }
    475 
    476 void
    477 ldstrategy(struct buf *bp)
    478 {
    479 	struct ld_softc *sc;
    480 	struct disklabel *lp;
    481 	daddr_t blkno;
    482 	int s, part;
    483 
    484 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    485 	part = DISKPART(bp->b_dev);
    486 
    487 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    488 		bp->b_error = EIO;
    489 		goto bad;
    490 	}
    491 
    492 	lp = sc->sc_dk.dk_label;
    493 
    494 	/*
    495 	 * The transfer must be a whole number of blocks and the offset must
    496 	 * not be negative.
    497 	 */
    498 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    499 		bp->b_error = EINVAL;
    500 		goto bad;
    501 	}
    502 
    503 	/* If it's a null transfer, return immediately. */
    504 	if (bp->b_bcount == 0)
    505 		goto done;
    506 
    507 	/*
    508 	 * Do bounds checking and adjust the transfer.  If error, process.
    509 	 * If past the end of partition, just return.
    510 	 */
    511 	if (part != RAW_PART &&
    512 	    bounds_check_with_label(&sc->sc_dk, bp,
    513 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    514 		goto done;
    515 	}
    516 
    517 	/*
    518 	 * Convert the block number to absolute and put it in terms
    519 	 * of the device's logical block size.
    520 	 */
    521 	if (lp->d_secsize == DEV_BSIZE)
    522 		blkno = bp->b_blkno;
    523 	else if (lp->d_secsize > DEV_BSIZE)
    524 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    525 	else
    526 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    527 
    528 	if (part != RAW_PART)
    529 		blkno += lp->d_partitions[part].p_offset;
    530 
    531 	bp->b_rawblkno = blkno;
    532 
    533 	s = splbio();
    534 	BUFQ_PUT(&sc->sc_bufq, bp);
    535 	ldstart(sc);
    536 	splx(s);
    537 	return;
    538 
    539  bad:
    540 	bp->b_flags |= B_ERROR;
    541  done:
    542 	bp->b_resid = bp->b_bcount;
    543 	biodone(bp);
    544 }
    545 
    546 static void
    547 ldstart(struct ld_softc *sc)
    548 {
    549 	struct buf *bp;
    550 	int error;
    551 
    552 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    553 		/* See if there is work to do. */
    554 		if ((bp = BUFQ_PEEK(&sc->sc_bufq)) == NULL)
    555 			break;
    556 
    557 		disk_busy(&sc->sc_dk);
    558 		sc->sc_queuecnt++;
    559 
    560 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    561 			/*
    562 			 * The back-end is running the job; remove it from
    563 			 * the queue.
    564 			 */
    565 			(void) BUFQ_GET(&sc->sc_bufq);
    566 		} else  {
    567 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    568 			sc->sc_queuecnt--;
    569 			if (error == EAGAIN) {
    570 				/*
    571 				 * Temporary resource shortage in the
    572 				 * back-end; just defer the job until
    573 				 * later.
    574 				 *
    575 				 * XXX We might consider a watchdog timer
    576 				 * XXX to make sure we are kicked into action.
    577 				 */
    578 				break;
    579 			} else {
    580 				(void) BUFQ_GET(&sc->sc_bufq);
    581 				bp->b_error = error;
    582 				bp->b_flags |= B_ERROR;
    583 				bp->b_resid = bp->b_bcount;
    584 				biodone(bp);
    585 			}
    586 		}
    587 	}
    588 }
    589 
    590 void
    591 lddone(struct ld_softc *sc, struct buf *bp)
    592 {
    593 
    594 	if ((bp->b_flags & B_ERROR) != 0) {
    595 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    596 		printf("\n");
    597 	}
    598 
    599 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    600 	    (bp->b_flags & B_READ));
    601 #if NRND > 0
    602 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    603 #endif
    604 	biodone(bp);
    605 
    606 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    607 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    608 			sc->sc_flags &= ~LDF_DRAIN;
    609 			wakeup(&sc->sc_queuecnt);
    610 		}
    611 		ldstart(sc);
    612 	}
    613 }
    614 
    615 int
    616 ldsize(dev_t dev)
    617 {
    618 	struct ld_softc *sc;
    619 	int part, unit, omask, size;
    620 
    621 	unit = DISKUNIT(dev);
    622 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    623 		return (ENODEV);
    624 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    625 		return (ENODEV);
    626 	part = DISKPART(dev);
    627 
    628 	omask = sc->sc_dk.dk_openmask & (1 << part);
    629 
    630 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    631 		return (-1);
    632 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    633 		size = -1;
    634 	else
    635 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    636 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    637 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    638 		return (-1);
    639 
    640 	return (size);
    641 }
    642 
    643 /*
    644  * Load the label information from the specified device.
    645  */
    646 static void
    647 ldgetdisklabel(struct ld_softc *sc)
    648 {
    649 	const char *errstring;
    650 
    651 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    652 
    653 	/* Call the generic disklabel extraction routine. */
    654 	errstring = readdisklabel(MAKEDISKDEV(0, sc->sc_dv.dv_unit, RAW_PART),
    655 	    ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    656 	if (errstring != NULL)
    657 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    658 
    659 	/* In-core label now valid. */
    660 	sc->sc_flags |= LDF_VLABEL;
    661 }
    662 
    663 /*
    664  * Construct a ficticious label.
    665  */
    666 static void
    667 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    668 {
    669 
    670 	memset(lp, 0, sizeof(struct disklabel));
    671 
    672 	lp->d_secsize = sc->sc_secsize;
    673 	lp->d_ntracks = sc->sc_nheads;
    674 	lp->d_nsectors = sc->sc_nsectors;
    675 	lp->d_ncylinders = sc->sc_ncylinders;
    676 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    677 	lp->d_type = DTYPE_LD;
    678 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    679 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    680 	lp->d_secperunit = sc->sc_secperunit;
    681 	lp->d_rpm = 7200;
    682 	lp->d_interleave = 1;
    683 	lp->d_flags = 0;
    684 
    685 	lp->d_partitions[RAW_PART].p_offset = 0;
    686 	lp->d_partitions[RAW_PART].p_size =
    687 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    688 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    689 	lp->d_npartitions = RAW_PART + 1;
    690 
    691 	lp->d_magic = DISKMAGIC;
    692 	lp->d_magic2 = DISKMAGIC;
    693 	lp->d_checksum = dkcksum(lp);
    694 }
    695 
    696 /*
    697  * Wait interruptibly for an exclusive lock.
    698  *
    699  * XXX Several drivers do this; it should be abstracted and made MP-safe.
    700  */
    701 static int
    702 ldlock(struct ld_softc *sc)
    703 {
    704 	int error;
    705 
    706 	while ((sc->sc_flags & LDF_LKHELD) != 0) {
    707 		sc->sc_flags |= LDF_LKWANTED;
    708 		if ((error = tsleep(sc, PRIBIO | PCATCH, "ldlck", 0)) != 0)
    709 			return (error);
    710 	}
    711 	sc->sc_flags |= LDF_LKHELD;
    712 	return (0);
    713 }
    714 
    715 /*
    716  * Unlock and wake up any waiters.
    717  */
    718 static void
    719 ldunlock(struct ld_softc *sc)
    720 {
    721 
    722 	sc->sc_flags &= ~LDF_LKHELD;
    723 	if ((sc->sc_flags & LDF_LKWANTED) != 0) {
    724 		sc->sc_flags &= ~LDF_LKWANTED;
    725 		wakeup(sc);
    726 	}
    727 }
    728 
    729 /*
    730  * Take a dump.
    731  */
    732 int
    733 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
    734 {
    735 	struct ld_softc *sc;
    736 	struct disklabel *lp;
    737 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    738 	static int dumping;
    739 
    740 	unit = DISKUNIT(dev);
    741 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    742 		return (ENXIO);
    743 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    744 		return (ENODEV);
    745 	if (sc->sc_dump == NULL)
    746 		return (ENXIO);
    747 
    748 	/* Check if recursive dump; if so, punt. */
    749 	if (dumping)
    750 		return (EFAULT);
    751 	dumping = 1;
    752 
    753 	/* Convert to disk sectors.  Request must be a multiple of size. */
    754 	part = DISKPART(dev);
    755 	lp = sc->sc_dk.dk_label;
    756 	if ((size % lp->d_secsize) != 0)
    757 		return (EFAULT);
    758 	towrt = size / lp->d_secsize;
    759 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    760 
    761 	nsects = lp->d_partitions[part].p_size;
    762 	sectoff = lp->d_partitions[part].p_offset;
    763 
    764 	/* Check transfer bounds against partition size. */
    765 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    766 		return (EINVAL);
    767 
    768 	/* Offset block number to start of partition. */
    769 	blkno += sectoff;
    770 
    771 	/* Start dumping and return when done. */
    772 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    773 	while (towrt > 0) {
    774 		nblk = min(maxblkcnt, towrt);
    775 
    776 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    777 			return (rv);
    778 
    779 		towrt -= nblk;
    780 		blkno += nblk;
    781 		va += nblk * sc->sc_secsize;
    782 	}
    783 
    784 	dumping = 0;
    785 	return (0);
    786 }
    787 
    788 /*
    789  * Adjust the size of a transfer.
    790  */
    791 static void
    792 ldminphys(struct buf *bp)
    793 {
    794 	struct ld_softc *sc;
    795 
    796 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    797 
    798 	if (bp->b_bcount > sc->sc_maxxfer)
    799 		bp->b_bcount = sc->sc_maxxfer;
    800 	minphys(bp);
    801 }
    802