Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.33
      1 /*	$NetBSD: ld.c,v 1.33 2004/10/28 07:07:39 yamt 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 1.33 2004/10/28 07:07:39 yamt 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 buf[9];
    105 
    106 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    107 		printf("%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(buf, sizeof(buf), sc->sc_secperunit *
    144 	    sc->sc_secsize);
    145 	printf("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    146 	    sc->sc_dv.dv_xname, buf, 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 max)
    166 {
    167 	int s;
    168 
    169 	s = splbio();
    170 	sc->sc_maxqueuecnt = max;
    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 	struct buf *bp;
    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", sc->sc_dv.dv_xname);
    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 	while ((bp = BUFQ_GET(&sc->sc_bufq)) != NULL) {
    222 		bp->b_error = EIO;
    223 		bp->b_flags |= B_ERROR;
    224 		bp->b_resid = bp->b_bcount;
    225 		biodone(bp);
    226 	}
    227 	bufq_free(&sc->sc_bufq);
    228 	splx(s);
    229 
    230 	/* Nuke the vnodes for any open instances. */
    231 	for (i = 0; i < MAXPARTITIONS; i++) {
    232 		mn = DISKMINOR(sc->sc_dv.dv_unit, i);
    233 		vdevgone(bmaj, mn, mn, VBLK);
    234 		vdevgone(cmaj, mn, mn, VCHR);
    235 	}
    236 
    237 	/* Delete all of our wedges. */
    238 	dkwedge_delall(&sc->sc_dk);
    239 
    240 	/* Detach from the disk list. */
    241 	disk_detach(&sc->sc_dk);
    242 
    243 #if NRND > 0
    244 	/* Unhook the entropy source. */
    245 	rnd_detach_source(&sc->sc_rnd_source);
    246 #endif
    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 			printf("%s: unable to flush cache\n",
    258 			    sc->sc_dv.dv_xname);
    259 #endif
    260 }
    261 
    262 /* ARGSUSED */
    263 static void
    264 ldshutdown(void *cookie)
    265 {
    266 	struct ld_softc *sc;
    267 	int i;
    268 
    269 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    270 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    271 			continue;
    272 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    273 			printf("%s: unable to flush cache\n",
    274 			    sc->sc_dv.dv_xname);
    275 	}
    276 }
    277 
    278 /* ARGSUSED */
    279 static int
    280 ldopen(dev_t dev, int flags, int fmt, struct proc *p)
    281 {
    282 	struct ld_softc *sc;
    283 	int error, unit, part;
    284 
    285 	unit = DISKUNIT(dev);
    286 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    287 		return (ENXIO);
    288 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    289 		return (ENODEV);
    290 	part = DISKPART(dev);
    291 
    292 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    293 		return (error);
    294 
    295 	if (sc->sc_dk.dk_openmask == 0) {
    296 		/* Load the partition info if not already loaded. */
    297 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    298 			ldgetdisklabel(sc);
    299 	}
    300 
    301 	/* Check that the partition exists. */
    302 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    303 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    304 		error = ENXIO;
    305 		goto bad1;
    306 	}
    307 
    308 	/* Ensure only one open at a time. */
    309 	switch (fmt) {
    310 	case S_IFCHR:
    311 		sc->sc_dk.dk_copenmask |= (1 << part);
    312 		break;
    313 	case S_IFBLK:
    314 		sc->sc_dk.dk_bopenmask |= (1 << part);
    315 		break;
    316 	}
    317 	sc->sc_dk.dk_openmask =
    318 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    319 
    320 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    321 	return (0);
    322 
    323  bad1:
    324 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    325 	return (error);
    326 }
    327 
    328 /* ARGSUSED */
    329 static int
    330 ldclose(dev_t dev, int flags, int fmt, struct proc *p)
    331 {
    332 	struct ld_softc *sc;
    333 	int error, part, unit;
    334 
    335 	unit = DISKUNIT(dev);
    336 	part = DISKPART(dev);
    337 	sc = device_lookup(&ld_cd, unit);
    338 
    339 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    340 		return (error);
    341 
    342 	switch (fmt) {
    343 	case S_IFCHR:
    344 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    345 		break;
    346 	case S_IFBLK:
    347 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    348 		break;
    349 	}
    350 	sc->sc_dk.dk_openmask =
    351 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    352 
    353 	if (sc->sc_dk.dk_openmask == 0) {
    354 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    355 			printf("%s: unable to flush cache\n",
    356 			    sc->sc_dv.dv_xname);
    357 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    358 			sc->sc_flags &= ~LDF_VLABEL;
    359 	}
    360 
    361 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    362 	return (0);
    363 }
    364 
    365 /* ARGSUSED */
    366 static int
    367 ldread(dev_t dev, struct uio *uio, int ioflag)
    368 {
    369 
    370 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    371 }
    372 
    373 /* ARGSUSED */
    374 static int
    375 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    376 {
    377 
    378 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    379 }
    380 
    381 /* ARGSUSED */
    382 static int
    383 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct proc *p)
    384 {
    385 	struct ld_softc *sc;
    386 	int part, unit, error;
    387 #ifdef __HAVE_OLD_DISKLABEL
    388 	struct disklabel newlabel;
    389 #endif
    390 	struct disklabel *lp;
    391 
    392 	unit = DISKUNIT(dev);
    393 	part = DISKPART(dev);
    394 	sc = device_lookup(&ld_cd, unit);
    395 	error = 0;
    396 
    397 	switch (cmd) {
    398 	case DIOCGDINFO:
    399 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    400 		return (0);
    401 
    402 #ifdef __HAVE_OLD_DISKLABEL
    403 	case ODIOCGDINFO:
    404 		newlabel = *(sc->sc_dk.dk_label);
    405 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    406 			return ENOTTY;
    407 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    408 		return (0);
    409 #endif
    410 
    411 	case DIOCGPART:
    412 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    413 		((struct partinfo *)addr)->part =
    414 		    &sc->sc_dk.dk_label->d_partitions[part];
    415 		break;
    416 
    417 	case DIOCWDINFO:
    418 	case DIOCSDINFO:
    419 #ifdef __HAVE_OLD_DISKLABEL
    420 	case ODIOCWDINFO:
    421 	case ODIOCSDINFO:
    422 
    423 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    424 			memset(&newlabel, 0, sizeof newlabel);
    425 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    426 			lp = &newlabel;
    427 		} else
    428 #endif
    429 		lp = (struct disklabel *)addr;
    430 
    431 		if ((flag & FWRITE) == 0)
    432 			return (EBADF);
    433 
    434 		if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE,
    435 				     NULL)) != 0)
    436 			return (error);
    437 		sc->sc_flags |= LDF_LABELLING;
    438 
    439 		error = setdisklabel(sc->sc_dk.dk_label,
    440 		    lp, /*sc->sc_dk.dk_openmask : */0,
    441 		    sc->sc_dk.dk_cpulabel);
    442 		if (error == 0 && (cmd == DIOCWDINFO
    443 #ifdef __HAVE_OLD_DISKLABEL
    444 		    || cmd == ODIOCWDINFO
    445 #endif
    446 		    ))
    447 			error = writedisklabel(
    448 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    449 			    ldstrategy, sc->sc_dk.dk_label,
    450 			    sc->sc_dk.dk_cpulabel);
    451 
    452 		sc->sc_flags &= ~LDF_LABELLING;
    453 		(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    454 		break;
    455 
    456 	case DIOCKLABEL:
    457 		if ((flag & FWRITE) == 0)
    458 			return (EBADF);
    459 		if (*(int *)addr)
    460 			sc->sc_flags |= LDF_KLABEL;
    461 		else
    462 			sc->sc_flags &= ~LDF_KLABEL;
    463 		break;
    464 
    465 	case DIOCWLABEL:
    466 		if ((flag & FWRITE) == 0)
    467 			return (EBADF);
    468 		if (*(int *)addr)
    469 			sc->sc_flags |= LDF_WLABEL;
    470 		else
    471 			sc->sc_flags &= ~LDF_WLABEL;
    472 		break;
    473 
    474 	case DIOCGDEFLABEL:
    475 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    476 		break;
    477 
    478 #ifdef __HAVE_OLD_DISKLABEL
    479 	case ODIOCGDEFLABEL:
    480 		ldgetdefaultlabel(sc, &newlabel);
    481 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    482 			return ENOTTY;
    483 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    484 		break;
    485 #endif
    486 
    487 	case DIOCCACHESYNC:
    488 		/*
    489 		 * XXX Do we really need to care about having a writable
    490 		 * file descriptor here?
    491 		 */
    492 		if ((flag & FWRITE) == 0)
    493 			error = EBADF;
    494 		else if (sc->sc_flush)
    495 			error = (*sc->sc_flush)(sc);
    496 		else
    497 			error = 0;	/* XXX Error out instead? */
    498 		break;
    499 
    500 	case DIOCAWEDGE:
    501 	    {
    502 	    	struct dkwedge_info *dkw = (void *) addr;
    503 
    504 		if ((flag & FWRITE) == 0)
    505 			return (EBADF);
    506 
    507 		/* If the ioctl happens here, the parent is us. */
    508 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    509 		return (dkwedge_add(dkw));
    510 	    }
    511 
    512 	case DIOCDWEDGE:
    513 	    {
    514 	    	struct dkwedge_info *dkw = (void *) addr;
    515 
    516 		if ((flag & FWRITE) == 0)
    517 			return (EBADF);
    518 
    519 		/* If the ioctl happens here, the parent is us. */
    520 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    521 		return (dkwedge_del(dkw));
    522 	    }
    523 
    524 	case DIOCLWEDGES:
    525 	    {
    526 	    	struct dkwedge_list *dkwl = (void *) addr;
    527 
    528 		return (dkwedge_list(&sc->sc_dk, dkwl, p));
    529 	    }
    530 
    531 	default:
    532 		error = ENOTTY;
    533 		break;
    534 	}
    535 
    536 	return (error);
    537 }
    538 
    539 static void
    540 ldstrategy(struct buf *bp)
    541 {
    542 	struct ld_softc *sc;
    543 	struct disklabel *lp;
    544 	daddr_t blkno;
    545 	int s, part;
    546 
    547 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    548 	part = DISKPART(bp->b_dev);
    549 
    550 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    551 		bp->b_error = EIO;
    552 		goto bad;
    553 	}
    554 
    555 	lp = sc->sc_dk.dk_label;
    556 
    557 	/*
    558 	 * The transfer must be a whole number of blocks and the offset must
    559 	 * not be negative.
    560 	 */
    561 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    562 		bp->b_error = EINVAL;
    563 		goto bad;
    564 	}
    565 
    566 	/* If it's a null transfer, return immediately. */
    567 	if (bp->b_bcount == 0)
    568 		goto done;
    569 
    570 	/*
    571 	 * Do bounds checking and adjust the transfer.  If error, process.
    572 	 * If past the end of partition, just return.
    573 	 */
    574 	if (part != RAW_PART &&
    575 	    bounds_check_with_label(&sc->sc_dk, bp,
    576 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    577 		goto done;
    578 	}
    579 
    580 	/*
    581 	 * Convert the block number to absolute and put it in terms
    582 	 * of the device's logical block size.
    583 	 */
    584 	if (lp->d_secsize == DEV_BSIZE)
    585 		blkno = bp->b_blkno;
    586 	else if (lp->d_secsize > DEV_BSIZE)
    587 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    588 	else
    589 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    590 
    591 	if (part != RAW_PART)
    592 		blkno += lp->d_partitions[part].p_offset;
    593 
    594 	bp->b_rawblkno = blkno;
    595 
    596 	s = splbio();
    597 	BUFQ_PUT(&sc->sc_bufq, bp);
    598 	ldstart(sc);
    599 	splx(s);
    600 	return;
    601 
    602  bad:
    603 	bp->b_flags |= B_ERROR;
    604  done:
    605 	bp->b_resid = bp->b_bcount;
    606 	biodone(bp);
    607 }
    608 
    609 static void
    610 ldstart(struct ld_softc *sc)
    611 {
    612 	struct buf *bp;
    613 	int error;
    614 
    615 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    616 		/* See if there is work to do. */
    617 		if ((bp = BUFQ_PEEK(&sc->sc_bufq)) == NULL)
    618 			break;
    619 
    620 		disk_busy(&sc->sc_dk);
    621 		sc->sc_queuecnt++;
    622 
    623 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    624 			/*
    625 			 * The back-end is running the job; remove it from
    626 			 * the queue.
    627 			 */
    628 			(void) BUFQ_GET(&sc->sc_bufq);
    629 		} else  {
    630 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    631 			sc->sc_queuecnt--;
    632 			if (error == EAGAIN) {
    633 				/*
    634 				 * Temporary resource shortage in the
    635 				 * back-end; just defer the job until
    636 				 * later.
    637 				 *
    638 				 * XXX We might consider a watchdog timer
    639 				 * XXX to make sure we are kicked into action.
    640 				 */
    641 				break;
    642 			} else {
    643 				(void) BUFQ_GET(&sc->sc_bufq);
    644 				bp->b_error = error;
    645 				bp->b_flags |= B_ERROR;
    646 				bp->b_resid = bp->b_bcount;
    647 				biodone(bp);
    648 			}
    649 		}
    650 	}
    651 }
    652 
    653 void
    654 lddone(struct ld_softc *sc, struct buf *bp)
    655 {
    656 
    657 	if ((bp->b_flags & B_ERROR) != 0) {
    658 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    659 		printf("\n");
    660 	}
    661 
    662 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    663 	    (bp->b_flags & B_READ));
    664 #if NRND > 0
    665 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    666 #endif
    667 	biodone(bp);
    668 
    669 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    670 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    671 			sc->sc_flags &= ~LDF_DRAIN;
    672 			wakeup(&sc->sc_queuecnt);
    673 		}
    674 		ldstart(sc);
    675 	}
    676 }
    677 
    678 static int
    679 ldsize(dev_t dev)
    680 {
    681 	struct ld_softc *sc;
    682 	int part, unit, omask, size;
    683 
    684 	unit = DISKUNIT(dev);
    685 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    686 		return (ENODEV);
    687 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    688 		return (ENODEV);
    689 	part = DISKPART(dev);
    690 
    691 	omask = sc->sc_dk.dk_openmask & (1 << part);
    692 
    693 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    694 		return (-1);
    695 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    696 		size = -1;
    697 	else
    698 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    699 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    700 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    701 		return (-1);
    702 
    703 	return (size);
    704 }
    705 
    706 /*
    707  * Load the label information from the specified device.
    708  */
    709 static void
    710 ldgetdisklabel(struct ld_softc *sc)
    711 {
    712 	const char *errstring;
    713 
    714 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    715 
    716 	/* Call the generic disklabel extraction routine. */
    717 	errstring = readdisklabel(MAKEDISKDEV(0, sc->sc_dv.dv_unit, RAW_PART),
    718 	    ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    719 	if (errstring != NULL)
    720 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    721 
    722 	/* In-core label now valid. */
    723 	sc->sc_flags |= LDF_VLABEL;
    724 }
    725 
    726 /*
    727  * Construct a ficticious label.
    728  */
    729 static void
    730 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    731 {
    732 
    733 	memset(lp, 0, sizeof(struct disklabel));
    734 
    735 	lp->d_secsize = sc->sc_secsize;
    736 	lp->d_ntracks = sc->sc_nheads;
    737 	lp->d_nsectors = sc->sc_nsectors;
    738 	lp->d_ncylinders = sc->sc_ncylinders;
    739 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    740 	lp->d_type = DTYPE_LD;
    741 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    742 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    743 	lp->d_secperunit = sc->sc_secperunit;
    744 	lp->d_rpm = 7200;
    745 	lp->d_interleave = 1;
    746 	lp->d_flags = 0;
    747 
    748 	lp->d_partitions[RAW_PART].p_offset = 0;
    749 	lp->d_partitions[RAW_PART].p_size =
    750 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    751 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    752 	lp->d_npartitions = RAW_PART + 1;
    753 
    754 	lp->d_magic = DISKMAGIC;
    755 	lp->d_magic2 = DISKMAGIC;
    756 	lp->d_checksum = dkcksum(lp);
    757 }
    758 
    759 /*
    760  * Take a dump.
    761  */
    762 static int
    763 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
    764 {
    765 	struct ld_softc *sc;
    766 	struct disklabel *lp;
    767 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    768 	static int dumping;
    769 
    770 	unit = DISKUNIT(dev);
    771 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    772 		return (ENXIO);
    773 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    774 		return (ENODEV);
    775 	if (sc->sc_dump == NULL)
    776 		return (ENXIO);
    777 
    778 	/* Check if recursive dump; if so, punt. */
    779 	if (dumping)
    780 		return (EFAULT);
    781 	dumping = 1;
    782 
    783 	/* Convert to disk sectors.  Request must be a multiple of size. */
    784 	part = DISKPART(dev);
    785 	lp = sc->sc_dk.dk_label;
    786 	if ((size % lp->d_secsize) != 0)
    787 		return (EFAULT);
    788 	towrt = size / lp->d_secsize;
    789 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    790 
    791 	nsects = lp->d_partitions[part].p_size;
    792 	sectoff = lp->d_partitions[part].p_offset;
    793 
    794 	/* Check transfer bounds against partition size. */
    795 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    796 		return (EINVAL);
    797 
    798 	/* Offset block number to start of partition. */
    799 	blkno += sectoff;
    800 
    801 	/* Start dumping and return when done. */
    802 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    803 	while (towrt > 0) {
    804 		nblk = min(maxblkcnt, towrt);
    805 
    806 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    807 			return (rv);
    808 
    809 		towrt -= nblk;
    810 		blkno += nblk;
    811 		va += nblk * sc->sc_secsize;
    812 	}
    813 
    814 	dumping = 0;
    815 	return (0);
    816 }
    817 
    818 /*
    819  * Adjust the size of a transfer.
    820  */
    821 static void
    822 ldminphys(struct buf *bp)
    823 {
    824 	struct ld_softc *sc;
    825 
    826 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    827 
    828 	if (bp->b_bcount > sc->sc_maxxfer)
    829 		bp->b_bcount = sc->sc_maxxfer;
    830 	minphys(bp);
    831 }
    832