Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.40.8.2
      1 /*	$NetBSD: ld.c,v 1.40.8.2 2007/01/15 17:58:57 ad 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 1.40.8.2 2007/01/15 17:58:57 ad 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 #include <sys/mutex.h>
     67 #if NRND > 0
     68 #include <sys/rnd.h>
     69 #endif
     70 
     71 #include <dev/ldvar.h>
     72 
     73 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     74 static void	ldgetdisklabel(struct ld_softc *);
     75 static void	ldminphys(struct buf *bp);
     76 static void	ldshutdown(void *);
     77 static void	ldstart(struct ld_softc *, struct buf *);
     78 
     79 extern struct	cfdriver ld_cd;
     80 
     81 static dev_type_open(ldopen);
     82 static dev_type_close(ldclose);
     83 static dev_type_read(ldread);
     84 static dev_type_write(ldwrite);
     85 static dev_type_ioctl(ldioctl);
     86 static dev_type_strategy(ldstrategy);
     87 static dev_type_dump(lddump);
     88 static 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, ldminphys };
    100 static void	*ld_sdh;
    101 
    102 void
    103 ldattach(struct ld_softc *sc)
    104 {
    105 	char tbuf[9];
    106 
    107 	mutex_init(&sc->sc_mutex, MUTEX_DRIVER, IPL_BIO);
    108 
    109 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    110 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
    111 		return;
    112 	}
    113 
    114 	/* Initialise and attach the disk structure. */
    115 	sc->sc_dk.dk_driver = &lddkdriver;
    116 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
    117 	disk_attach(&sc->sc_dk);
    118 
    119 	if (sc->sc_maxxfer > MAXPHYS)
    120 		sc->sc_maxxfer = MAXPHYS;
    121 
    122 	/* Build synthetic geometry if necessary. */
    123 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    124 	    sc->sc_ncylinders == 0) {
    125 		uint64_t ncyl;
    126 
    127 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    128 			sc->sc_nheads = 16;
    129 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    130 			sc->sc_nheads = 32;
    131 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    132 			sc->sc_nheads = 64;
    133 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    134 			sc->sc_nheads = 128;
    135 		else
    136 			sc->sc_nheads = 255;
    137 
    138 		sc->sc_nsectors = 63;
    139 		sc->sc_ncylinders = INT_MAX;
    140 		ncyl = sc->sc_secperunit /
    141 		    (sc->sc_nheads * sc->sc_nsectors);
    142 		if (ncyl < INT_MAX)
    143 			sc->sc_ncylinders = (int)ncyl;
    144 	}
    145 
    146 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    147 	    sc->sc_secsize);
    148 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    149 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
    150 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    151 
    152 #if NRND > 0
    153 	/* Attach the device into the rnd source list. */
    154 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    155 	    RND_TYPE_DISK, 0);
    156 #endif
    157 
    158 	/* Set the `shutdownhook'. */
    159 	if (ld_sdh == NULL)
    160 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
    161 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    162 
    163 	/* Discover wedges on this disk. */
    164 	dkwedge_discover(&sc->sc_dk);
    165 }
    166 
    167 int
    168 ldadjqparam(struct ld_softc *sc, int xmax)
    169 {
    170 	int s;
    171 
    172 	s = splbio();
    173 	sc->sc_maxqueuecnt = xmax;
    174 	splx(s);
    175 
    176 	return (0);
    177 }
    178 
    179 int
    180 ldbegindetach(struct ld_softc *sc, int flags)
    181 {
    182 	int s, rv = 0;
    183 
    184 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    185 		return (0);
    186 
    187 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    188 		return (EBUSY);
    189 
    190 	s = splbio();
    191 	sc->sc_maxqueuecnt = 0;
    192 	sc->sc_flags |= LDF_DETACH;
    193 	while (sc->sc_queuecnt > 0) {
    194 		sc->sc_flags |= LDF_DRAIN;
    195 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    196 		if (rv)
    197 			break;
    198 	}
    199 	splx(s);
    200 
    201 	return (rv);
    202 }
    203 
    204 void
    205 ldenddetach(struct ld_softc *sc)
    206 {
    207 	int s, bmaj, cmaj, i, mn;
    208 
    209 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    210 		return;
    211 
    212 	/* Wait for commands queued with the hardware to complete. */
    213 	if (sc->sc_queuecnt != 0)
    214 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    215 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
    216 
    217 	/* Locate the major numbers. */
    218 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    219 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    220 
    221 	/* Kill off any queued buffers. */
    222 	s = splbio();
    223 	bufq_drain(sc->sc_bufq);
    224 	splx(s);
    225 
    226 	bufq_free(sc->sc_bufq);
    227 
    228 	/* Nuke the vnodes for any open instances. */
    229 	for (i = 0; i < MAXPARTITIONS; i++) {
    230 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
    231 		vdevgone(bmaj, mn, mn, VBLK);
    232 		vdevgone(cmaj, mn, mn, VCHR);
    233 	}
    234 
    235 	/* Delete all of our wedges. */
    236 	dkwedge_delall(&sc->sc_dk);
    237 
    238 	/* Detach from the disk list. */
    239 	disk_detach(&sc->sc_dk);
    240 
    241 #if NRND > 0
    242 	/* Unhook the entropy source. */
    243 	rnd_detach_source(&sc->sc_rnd_source);
    244 #endif
    245 
    246 	/*
    247 	 * XXX We can't really flush the cache here, beceause the
    248 	 * XXX device may already be non-existent from the controller's
    249 	 * XXX perspective.
    250 	 */
    251 #if 0
    252 	/* Flush the device's cache. */
    253 	if (sc->sc_flush != NULL)
    254 		if ((*sc->sc_flush)(sc) != 0)
    255 			printf("%s: unable to flush cache\n",
    256 			    sc->sc_dv.dv_xname);
    257 #endif
    258 }
    259 
    260 /* ARGSUSED */
    261 static void
    262 ldshutdown(void *cookie)
    263 {
    264 	struct ld_softc *sc;
    265 	int i;
    266 
    267 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    268 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    269 			continue;
    270 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    271 			printf("%s: unable to flush cache\n",
    272 			    sc->sc_dv.dv_xname);
    273 	}
    274 }
    275 
    276 /* ARGSUSED */
    277 static int
    278 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    279 {
    280 	struct ld_softc *sc;
    281 	int error, unit, part;
    282 
    283 	unit = DISKUNIT(dev);
    284 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    285 		return (ENXIO);
    286 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    287 		return (ENODEV);
    288 	part = DISKPART(dev);
    289 
    290 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    291 		return (error);
    292 
    293 	if (sc->sc_dk.dk_openmask == 0) {
    294 		/* Load the partition info if not already loaded. */
    295 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    296 			ldgetdisklabel(sc);
    297 	}
    298 
    299 	/* Check that the partition exists. */
    300 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    301 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    302 		error = ENXIO;
    303 		goto bad1;
    304 	}
    305 
    306 	/* Ensure only one open at a time. */
    307 	switch (fmt) {
    308 	case S_IFCHR:
    309 		sc->sc_dk.dk_copenmask |= (1 << part);
    310 		break;
    311 	case S_IFBLK:
    312 		sc->sc_dk.dk_bopenmask |= (1 << part);
    313 		break;
    314 	}
    315 	sc->sc_dk.dk_openmask =
    316 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    317 
    318 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    319 	return (0);
    320 
    321  bad1:
    322 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    323 	return (error);
    324 }
    325 
    326 /* ARGSUSED */
    327 static int
    328 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    329 {
    330 	struct ld_softc *sc;
    331 	int error, part, unit;
    332 
    333 	unit = DISKUNIT(dev);
    334 	part = DISKPART(dev);
    335 	sc = device_lookup(&ld_cd, unit);
    336 
    337 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    338 		return (error);
    339 
    340 	switch (fmt) {
    341 	case S_IFCHR:
    342 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    343 		break;
    344 	case S_IFBLK:
    345 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    346 		break;
    347 	}
    348 	sc->sc_dk.dk_openmask =
    349 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    350 
    351 	if (sc->sc_dk.dk_openmask == 0) {
    352 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    353 			printf("%s: unable to flush cache\n",
    354 			    sc->sc_dv.dv_xname);
    355 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    356 			sc->sc_flags &= ~LDF_VLABEL;
    357 	}
    358 
    359 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    360 	return (0);
    361 }
    362 
    363 /* ARGSUSED */
    364 static int
    365 ldread(dev_t dev, struct uio *uio, int ioflag)
    366 {
    367 
    368 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    369 }
    370 
    371 /* ARGSUSED */
    372 static int
    373 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    374 {
    375 
    376 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    377 }
    378 
    379 /* ARGSUSED */
    380 static int
    381 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct lwp *l)
    382 {
    383 	struct ld_softc *sc;
    384 	int part, unit, error;
    385 #ifdef __HAVE_OLD_DISKLABEL
    386 	struct disklabel newlabel;
    387 #endif
    388 	struct disklabel *lp;
    389 
    390 	unit = DISKUNIT(dev);
    391 	part = DISKPART(dev);
    392 	sc = device_lookup(&ld_cd, unit);
    393 	error = 0;
    394 
    395 	switch (cmd) {
    396 	case DIOCGDINFO:
    397 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    398 		return (0);
    399 
    400 #ifdef __HAVE_OLD_DISKLABEL
    401 	case ODIOCGDINFO:
    402 		newlabel = *(sc->sc_dk.dk_label);
    403 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    404 			return ENOTTY;
    405 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    406 		return (0);
    407 #endif
    408 
    409 	case DIOCGPART:
    410 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    411 		((struct partinfo *)addr)->part =
    412 		    &sc->sc_dk.dk_label->d_partitions[part];
    413 		break;
    414 
    415 	case DIOCWDINFO:
    416 	case DIOCSDINFO:
    417 #ifdef __HAVE_OLD_DISKLABEL
    418 	case ODIOCWDINFO:
    419 	case ODIOCSDINFO:
    420 
    421 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    422 			memset(&newlabel, 0, sizeof newlabel);
    423 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    424 			lp = &newlabel;
    425 		} else
    426 #endif
    427 		lp = (struct disklabel *)addr;
    428 
    429 		if ((flag & FWRITE) == 0)
    430 			return (EBADF);
    431 
    432 		if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE,
    433 				     NULL)) != 0)
    434 			return (error);
    435 		sc->sc_flags |= LDF_LABELLING;
    436 
    437 		error = setdisklabel(sc->sc_dk.dk_label,
    438 		    lp, /*sc->sc_dk.dk_openmask : */0,
    439 		    sc->sc_dk.dk_cpulabel);
    440 		if (error == 0 && (cmd == DIOCWDINFO
    441 #ifdef __HAVE_OLD_DISKLABEL
    442 		    || cmd == ODIOCWDINFO
    443 #endif
    444 		    ))
    445 			error = writedisklabel(
    446 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    447 			    ldstrategy, sc->sc_dk.dk_label,
    448 			    sc->sc_dk.dk_cpulabel);
    449 
    450 		sc->sc_flags &= ~LDF_LABELLING;
    451 		(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    452 		break;
    453 
    454 	case DIOCKLABEL:
    455 		if ((flag & FWRITE) == 0)
    456 			return (EBADF);
    457 		if (*(int *)addr)
    458 			sc->sc_flags |= LDF_KLABEL;
    459 		else
    460 			sc->sc_flags &= ~LDF_KLABEL;
    461 		break;
    462 
    463 	case DIOCWLABEL:
    464 		if ((flag & FWRITE) == 0)
    465 			return (EBADF);
    466 		if (*(int *)addr)
    467 			sc->sc_flags |= LDF_WLABEL;
    468 		else
    469 			sc->sc_flags &= ~LDF_WLABEL;
    470 		break;
    471 
    472 	case DIOCGDEFLABEL:
    473 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    474 		break;
    475 
    476 #ifdef __HAVE_OLD_DISKLABEL
    477 	case ODIOCGDEFLABEL:
    478 		ldgetdefaultlabel(sc, &newlabel);
    479 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    480 			return ENOTTY;
    481 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    482 		break;
    483 #endif
    484 
    485 	case DIOCCACHESYNC:
    486 		/*
    487 		 * XXX Do we really need to care about having a writable
    488 		 * file descriptor here?
    489 		 */
    490 		if ((flag & FWRITE) == 0)
    491 			error = EBADF;
    492 		else if (sc->sc_flush)
    493 			error = (*sc->sc_flush)(sc);
    494 		else
    495 			error = 0;	/* XXX Error out instead? */
    496 		break;
    497 
    498 	case DIOCAWEDGE:
    499 	    {
    500 	    	struct dkwedge_info *dkw = (void *) addr;
    501 
    502 		if ((flag & FWRITE) == 0)
    503 			return (EBADF);
    504 
    505 		/* If the ioctl happens here, the parent is us. */
    506 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    507 		return (dkwedge_add(dkw));
    508 	    }
    509 
    510 	case DIOCDWEDGE:
    511 	    {
    512 	    	struct dkwedge_info *dkw = (void *) addr;
    513 
    514 		if ((flag & FWRITE) == 0)
    515 			return (EBADF);
    516 
    517 		/* If the ioctl happens here, the parent is us. */
    518 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    519 		return (dkwedge_del(dkw));
    520 	    }
    521 
    522 	case DIOCLWEDGES:
    523 	    {
    524 	    	struct dkwedge_list *dkwl = (void *) addr;
    525 
    526 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
    527 	    }
    528 
    529 	default:
    530 		error = ENOTTY;
    531 		break;
    532 	}
    533 
    534 	return (error);
    535 }
    536 
    537 static void
    538 ldstrategy(struct buf *bp)
    539 {
    540 	struct ld_softc *sc;
    541 	struct disklabel *lp;
    542 	daddr_t blkno;
    543 	int s, part;
    544 
    545 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    546 	part = DISKPART(bp->b_dev);
    547 
    548 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    549 		bp->b_error = EIO;
    550 		goto bad;
    551 	}
    552 
    553 	lp = sc->sc_dk.dk_label;
    554 
    555 	/*
    556 	 * The transfer must be a whole number of blocks and the offset must
    557 	 * not be negative.
    558 	 */
    559 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    560 		bp->b_error = EINVAL;
    561 		goto bad;
    562 	}
    563 
    564 	/* If it's a null transfer, return immediately. */
    565 	if (bp->b_bcount == 0)
    566 		goto done;
    567 
    568 	/*
    569 	 * Do bounds checking and adjust the transfer.  If error, process.
    570 	 * If past the end of partition, just return.
    571 	 */
    572 	if (part != RAW_PART &&
    573 	    bounds_check_with_label(&sc->sc_dk, bp,
    574 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    575 		goto done;
    576 	}
    577 
    578 	/*
    579 	 * Convert the block number to absolute and put it in terms
    580 	 * of the device's logical block size.
    581 	 */
    582 	if (lp->d_secsize == DEV_BSIZE)
    583 		blkno = bp->b_blkno;
    584 	else if (lp->d_secsize > DEV_BSIZE)
    585 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    586 	else
    587 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    588 
    589 	if (part != RAW_PART)
    590 		blkno += lp->d_partitions[part].p_offset;
    591 
    592 	bp->b_rawblkno = blkno;
    593 
    594 	s = splbio();
    595 	ldstart(sc, bp);
    596 	splx(s);
    597 	return;
    598 
    599  bad:
    600 	bp->b_flags |= B_ERROR;
    601  done:
    602 	bp->b_resid = bp->b_bcount;
    603 	biodone(bp);
    604 }
    605 
    606 static void
    607 ldstart(struct ld_softc *sc, struct buf *bp)
    608 {
    609 	int error;
    610 
    611 	mutex_enter(&sc->sc_mutex);
    612 
    613 	if (bp != NULL)
    614 		BUFQ_PUT(sc->sc_bufq, bp);
    615 
    616 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    617 		/* See if there is work to do. */
    618 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    619 			break;
    620 
    621 		disk_busy(&sc->sc_dk);
    622 		sc->sc_queuecnt++;
    623 
    624 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    625 			/*
    626 			 * The back-end is running the job; remove it from
    627 			 * the queue.
    628 			 */
    629 			(void) BUFQ_GET(sc->sc_bufq);
    630 		} else  {
    631 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    632 			sc->sc_queuecnt--;
    633 			if (error == EAGAIN) {
    634 				/*
    635 				 * Temporary resource shortage in the
    636 				 * back-end; just defer the job until
    637 				 * later.
    638 				 *
    639 				 * XXX We might consider a watchdog timer
    640 				 * XXX to make sure we are kicked into action.
    641 				 */
    642 				break;
    643 			} else {
    644 				(void) BUFQ_GET(sc->sc_bufq);
    645 				bp->b_error = error;
    646 				bp->b_flags |= B_ERROR;
    647 				bp->b_resid = bp->b_bcount;
    648 				mutex_exit(&sc->sc_mutex);
    649 				biodone(bp);
    650 				mutex_enter(&sc->sc_mutex);
    651 			}
    652 		}
    653 	}
    654 
    655 	mutex_exit(&sc->sc_mutex);
    656 }
    657 
    658 void
    659 lddone(struct ld_softc *sc, struct buf *bp)
    660 {
    661 
    662 	if ((bp->b_flags & B_ERROR) != 0) {
    663 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    664 		printf("\n");
    665 	}
    666 
    667 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    668 	    (bp->b_flags & B_READ));
    669 #if NRND > 0
    670 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    671 #endif
    672 	biodone(bp);
    673 
    674 	mutex_enter(&sc->sc_mutex);
    675 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    676 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    677 			sc->sc_flags &= ~LDF_DRAIN;
    678 			wakeup(&sc->sc_queuecnt);
    679 		}
    680 		mutex_exit(&sc->sc_mutex);
    681 		ldstart(sc, NULL);
    682 	} else
    683 		mutex_exit(&sc->sc_mutex);
    684 }
    685 
    686 static int
    687 ldsize(dev_t dev)
    688 {
    689 	struct ld_softc *sc;
    690 	int part, unit, omask, size;
    691 
    692 	unit = DISKUNIT(dev);
    693 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    694 		return (ENODEV);
    695 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    696 		return (ENODEV);
    697 	part = DISKPART(dev);
    698 
    699 	omask = sc->sc_dk.dk_openmask & (1 << part);
    700 
    701 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    702 		return (-1);
    703 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    704 		size = -1;
    705 	else
    706 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    707 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    708 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    709 		return (-1);
    710 
    711 	return (size);
    712 }
    713 
    714 /*
    715  * Load the label information from the specified device.
    716  */
    717 static void
    718 ldgetdisklabel(struct ld_softc *sc)
    719 {
    720 	const char *errstring;
    721 
    722 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    723 
    724 	/* Call the generic disklabel extraction routine. */
    725 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
    726 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    727 	if (errstring != NULL)
    728 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    729 
    730 	/* In-core label now valid. */
    731 	sc->sc_flags |= LDF_VLABEL;
    732 }
    733 
    734 /*
    735  * Construct a ficticious label.
    736  */
    737 static void
    738 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    739 {
    740 
    741 	memset(lp, 0, sizeof(struct disklabel));
    742 
    743 	lp->d_secsize = sc->sc_secsize;
    744 	lp->d_ntracks = sc->sc_nheads;
    745 	lp->d_nsectors = sc->sc_nsectors;
    746 	lp->d_ncylinders = sc->sc_ncylinders;
    747 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    748 	lp->d_type = DTYPE_LD;
    749 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    750 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    751 	lp->d_secperunit = sc->sc_secperunit;
    752 	lp->d_rpm = 7200;
    753 	lp->d_interleave = 1;
    754 	lp->d_flags = 0;
    755 
    756 	lp->d_partitions[RAW_PART].p_offset = 0;
    757 	lp->d_partitions[RAW_PART].p_size =
    758 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    759 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    760 	lp->d_npartitions = RAW_PART + 1;
    761 
    762 	lp->d_magic = DISKMAGIC;
    763 	lp->d_magic2 = DISKMAGIC;
    764 	lp->d_checksum = dkcksum(lp);
    765 }
    766 
    767 /*
    768  * Take a dump.
    769  */
    770 static int
    771 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
    772 {
    773 	struct ld_softc *sc;
    774 	struct disklabel *lp;
    775 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    776 	static int dumping;
    777 
    778 	unit = DISKUNIT(dev);
    779 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    780 		return (ENXIO);
    781 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    782 		return (ENODEV);
    783 	if (sc->sc_dump == NULL)
    784 		return (ENXIO);
    785 
    786 	/* Check if recursive dump; if so, punt. */
    787 	if (dumping)
    788 		return (EFAULT);
    789 	dumping = 1;
    790 
    791 	/* Convert to disk sectors.  Request must be a multiple of size. */
    792 	part = DISKPART(dev);
    793 	lp = sc->sc_dk.dk_label;
    794 	if ((size % lp->d_secsize) != 0)
    795 		return (EFAULT);
    796 	towrt = size / lp->d_secsize;
    797 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    798 
    799 	nsects = lp->d_partitions[part].p_size;
    800 	sectoff = lp->d_partitions[part].p_offset;
    801 
    802 	/* Check transfer bounds against partition size. */
    803 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    804 		return (EINVAL);
    805 
    806 	/* Offset block number to start of partition. */
    807 	blkno += sectoff;
    808 
    809 	/* Start dumping and return when done. */
    810 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    811 	while (towrt > 0) {
    812 		nblk = min(maxblkcnt, towrt);
    813 
    814 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    815 			return (rv);
    816 
    817 		towrt -= nblk;
    818 		blkno += nblk;
    819 		va += nblk * sc->sc_secsize;
    820 	}
    821 
    822 	dumping = 0;
    823 	return (0);
    824 }
    825 
    826 /*
    827  * Adjust the size of a transfer.
    828  */
    829 static void
    830 ldminphys(struct buf *bp)
    831 {
    832 	struct ld_softc *sc;
    833 
    834 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    835 
    836 	if (bp->b_bcount > sc->sc_maxxfer)
    837 		bp->b_bcount = sc->sc_maxxfer;
    838 	minphys(bp);
    839 }
    840