Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.49
      1 /*	$NetBSD: ld.c,v 1.49 2007/07/29 12:50:18 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 1.49 2007/07/29 12:50:18 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 #include <prop/proplib.h>
     74 
     75 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     76 static void	ldgetdisklabel(struct ld_softc *);
     77 static void	ldminphys(struct buf *bp);
     78 static void	ldshutdown(void *);
     79 static void	ldstart(struct ld_softc *, struct buf *);
     80 static void	ld_set_properties(struct ld_softc *);
     81 static void	ld_config_interrupts (struct device *);
     82 
     83 extern struct	cfdriver ld_cd;
     84 
     85 static dev_type_open(ldopen);
     86 static dev_type_close(ldclose);
     87 static dev_type_read(ldread);
     88 static dev_type_write(ldwrite);
     89 static dev_type_ioctl(ldioctl);
     90 static dev_type_strategy(ldstrategy);
     91 static dev_type_dump(lddump);
     92 static dev_type_size(ldsize);
     93 
     94 const struct bdevsw ld_bdevsw = {
     95 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
     96 };
     97 
     98 const struct cdevsw ld_cdevsw = {
     99 	ldopen, ldclose, ldread, ldwrite, ldioctl,
    100 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    101 };
    102 
    103 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
    104 static void	*ld_sdh;
    105 
    106 void
    107 ldattach(struct ld_softc *sc)
    108 {
    109 	char tbuf[9];
    110 
    111 	mutex_init(&sc->sc_mutex, MUTEX_DRIVER, IPL_BIO);
    112 
    113 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    114 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
    115 		return;
    116 	}
    117 
    118 	/* Initialise and attach the disk structure. */
    119 	sc->sc_dk.dk_driver = &lddkdriver;
    120 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
    121 	disk_attach(&sc->sc_dk);
    122 
    123 	if (sc->sc_maxxfer > MAXPHYS)
    124 		sc->sc_maxxfer = MAXPHYS;
    125 
    126 	/* Build synthetic geometry if necessary. */
    127 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    128 	    sc->sc_ncylinders == 0) {
    129 		uint64_t ncyl;
    130 
    131 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    132 			sc->sc_nheads = 16;
    133 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    134 			sc->sc_nheads = 32;
    135 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    136 			sc->sc_nheads = 64;
    137 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    138 			sc->sc_nheads = 128;
    139 		else
    140 			sc->sc_nheads = 255;
    141 
    142 		sc->sc_nsectors = 63;
    143 		sc->sc_ncylinders = INT_MAX;
    144 		ncyl = sc->sc_secperunit /
    145 		    (sc->sc_nheads * sc->sc_nsectors);
    146 		if (ncyl < INT_MAX)
    147 			sc->sc_ncylinders = (int)ncyl;
    148 	}
    149 
    150 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    151 	    sc->sc_secsize);
    152 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    153 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
    154 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    155 
    156 	ld_set_properties(sc);
    157 
    158 #if NRND > 0
    159 	/* Attach the device into the rnd source list. */
    160 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    161 	    RND_TYPE_DISK, 0);
    162 #endif
    163 
    164 	/* Set the `shutdownhook'. */
    165 	if (ld_sdh == NULL)
    166 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
    167 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    168 
    169 	/* Discover wedges on this disk. */
    170 	config_interrupts(&sc->sc_dv, ld_config_interrupts);
    171 }
    172 
    173 int
    174 ldadjqparam(struct ld_softc *sc, int xmax)
    175 {
    176 	int s;
    177 
    178 	s = splbio();
    179 	sc->sc_maxqueuecnt = xmax;
    180 	splx(s);
    181 
    182 	return (0);
    183 }
    184 
    185 int
    186 ldbegindetach(struct ld_softc *sc, int flags)
    187 {
    188 	int s, rv = 0;
    189 
    190 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    191 		return (0);
    192 
    193 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    194 		return (EBUSY);
    195 
    196 	s = splbio();
    197 	sc->sc_maxqueuecnt = 0;
    198 	sc->sc_flags |= LDF_DETACH;
    199 	while (sc->sc_queuecnt > 0) {
    200 		sc->sc_flags |= LDF_DRAIN;
    201 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    202 		if (rv)
    203 			break;
    204 	}
    205 	splx(s);
    206 
    207 	return (rv);
    208 }
    209 
    210 void
    211 ldenddetach(struct ld_softc *sc)
    212 {
    213 	int s, bmaj, cmaj, i, mn;
    214 
    215 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    216 		return;
    217 
    218 	/* Wait for commands queued with the hardware to complete. */
    219 	if (sc->sc_queuecnt != 0)
    220 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    221 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
    222 
    223 	/* Locate the major numbers. */
    224 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    225 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    226 
    227 	/* Kill off any queued buffers. */
    228 	s = splbio();
    229 	bufq_drain(sc->sc_bufq);
    230 	splx(s);
    231 
    232 	bufq_free(sc->sc_bufq);
    233 
    234 	/* Nuke the vnodes for any open instances. */
    235 	for (i = 0; i < MAXPARTITIONS; i++) {
    236 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
    237 		vdevgone(bmaj, mn, mn, VBLK);
    238 		vdevgone(cmaj, mn, mn, VCHR);
    239 	}
    240 
    241 	/* Delete all of our wedges. */
    242 	dkwedge_delall(&sc->sc_dk);
    243 
    244 	/* Detach from the disk list. */
    245 	disk_detach(&sc->sc_dk);
    246 
    247 #if NRND > 0
    248 	/* Unhook the entropy source. */
    249 	rnd_detach_source(&sc->sc_rnd_source);
    250 #endif
    251 
    252 	/*
    253 	 * XXX We can't really flush the cache here, beceause the
    254 	 * XXX device may already be non-existent from the controller's
    255 	 * XXX perspective.
    256 	 */
    257 #if 0
    258 	/* Flush the device's cache. */
    259 	if (sc->sc_flush != NULL)
    260 		if ((*sc->sc_flush)(sc) != 0)
    261 			printf("%s: unable to flush cache\n",
    262 			    sc->sc_dv.dv_xname);
    263 #endif
    264 }
    265 
    266 /* ARGSUSED */
    267 static void
    268 ldshutdown(void *cookie)
    269 {
    270 	struct ld_softc *sc;
    271 	int i;
    272 
    273 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    274 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    275 			continue;
    276 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    277 			printf("%s: unable to flush cache\n",
    278 			    sc->sc_dv.dv_xname);
    279 	}
    280 }
    281 
    282 /* ARGSUSED */
    283 static int
    284 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    285 {
    286 	struct ld_softc *sc;
    287 	int error, unit, part;
    288 
    289 	unit = DISKUNIT(dev);
    290 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    291 		return (ENXIO);
    292 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    293 		return (ENODEV);
    294 	part = DISKPART(dev);
    295 
    296 	mutex_enter(&sc->sc_dk.dk_openlock);
    297 
    298 	if (sc->sc_dk.dk_openmask == 0) {
    299 		/* Load the partition info if not already loaded. */
    300 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    301 			ldgetdisklabel(sc);
    302 	}
    303 
    304 	/* Check that the partition exists. */
    305 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    306 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    307 		error = ENXIO;
    308 		goto bad1;
    309 	}
    310 
    311 	/* Ensure only one open at a time. */
    312 	switch (fmt) {
    313 	case S_IFCHR:
    314 		sc->sc_dk.dk_copenmask |= (1 << part);
    315 		break;
    316 	case S_IFBLK:
    317 		sc->sc_dk.dk_bopenmask |= (1 << part);
    318 		break;
    319 	}
    320 	sc->sc_dk.dk_openmask =
    321 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    322 
    323 	error = 0;
    324  bad1:
    325 	mutex_exit(&sc->sc_dk.dk_openlock);
    326 	return (error);
    327 }
    328 
    329 /* ARGSUSED */
    330 static int
    331 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    332 {
    333 	struct ld_softc *sc;
    334 	int part, unit;
    335 
    336 	unit = DISKUNIT(dev);
    337 	part = DISKPART(dev);
    338 	sc = device_lookup(&ld_cd, unit);
    339 
    340 	mutex_enter(&sc->sc_dk.dk_openlock);
    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 	mutex_exit(&sc->sc_dk.dk_openlock);
    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, void *addr, int32_t flag, struct lwp *l)
    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 
    396 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
    397 	if (error != EPASSTHROUGH)
    398 		return (error);
    399 
    400 	error = 0;
    401 	switch (cmd) {
    402 	case DIOCGDINFO:
    403 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    404 		return (0);
    405 
    406 #ifdef __HAVE_OLD_DISKLABEL
    407 	case ODIOCGDINFO:
    408 		newlabel = *(sc->sc_dk.dk_label);
    409 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    410 			return ENOTTY;
    411 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    412 		return (0);
    413 #endif
    414 
    415 	case DIOCGPART:
    416 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    417 		((struct partinfo *)addr)->part =
    418 		    &sc->sc_dk.dk_label->d_partitions[part];
    419 		break;
    420 
    421 	case DIOCWDINFO:
    422 	case DIOCSDINFO:
    423 #ifdef __HAVE_OLD_DISKLABEL
    424 	case ODIOCWDINFO:
    425 	case ODIOCSDINFO:
    426 
    427 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    428 			memset(&newlabel, 0, sizeof newlabel);
    429 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    430 			lp = &newlabel;
    431 		} else
    432 #endif
    433 		lp = (struct disklabel *)addr;
    434 
    435 		if ((flag & FWRITE) == 0)
    436 			return (EBADF);
    437 
    438 		mutex_enter(&sc->sc_dk.dk_openlock);
    439 		sc->sc_flags |= LDF_LABELLING;
    440 
    441 		error = setdisklabel(sc->sc_dk.dk_label,
    442 		    lp, /*sc->sc_dk.dk_openmask : */0,
    443 		    sc->sc_dk.dk_cpulabel);
    444 		if (error == 0 && (cmd == DIOCWDINFO
    445 #ifdef __HAVE_OLD_DISKLABEL
    446 		    || cmd == ODIOCWDINFO
    447 #endif
    448 		    ))
    449 			error = writedisklabel(
    450 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    451 			    ldstrategy, sc->sc_dk.dk_label,
    452 			    sc->sc_dk.dk_cpulabel);
    453 
    454 		sc->sc_flags &= ~LDF_LABELLING;
    455 		mutex_exit(&sc->sc_dk.dk_openlock);
    456 		break;
    457 
    458 	case DIOCKLABEL:
    459 		if ((flag & FWRITE) == 0)
    460 			return (EBADF);
    461 		if (*(int *)addr)
    462 			sc->sc_flags |= LDF_KLABEL;
    463 		else
    464 			sc->sc_flags &= ~LDF_KLABEL;
    465 		break;
    466 
    467 	case DIOCWLABEL:
    468 		if ((flag & FWRITE) == 0)
    469 			return (EBADF);
    470 		if (*(int *)addr)
    471 			sc->sc_flags |= LDF_WLABEL;
    472 		else
    473 			sc->sc_flags &= ~LDF_WLABEL;
    474 		break;
    475 
    476 	case DIOCGDEFLABEL:
    477 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    478 		break;
    479 
    480 #ifdef __HAVE_OLD_DISKLABEL
    481 	case ODIOCGDEFLABEL:
    482 		ldgetdefaultlabel(sc, &newlabel);
    483 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    484 			return ENOTTY;
    485 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    486 		break;
    487 #endif
    488 
    489 	case DIOCCACHESYNC:
    490 		/*
    491 		 * XXX Do we really need to care about having a writable
    492 		 * file descriptor here?
    493 		 */
    494 		if ((flag & FWRITE) == 0)
    495 			error = EBADF;
    496 		else if (sc->sc_flush)
    497 			error = (*sc->sc_flush)(sc);
    498 		else
    499 			error = 0;	/* XXX Error out instead? */
    500 		break;
    501 
    502 	case DIOCAWEDGE:
    503 	    {
    504 	    	struct dkwedge_info *dkw = (void *) addr;
    505 
    506 		if ((flag & FWRITE) == 0)
    507 			return (EBADF);
    508 
    509 		/* If the ioctl happens here, the parent is us. */
    510 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    511 		return (dkwedge_add(dkw));
    512 	    }
    513 
    514 	case DIOCDWEDGE:
    515 	    {
    516 	    	struct dkwedge_info *dkw = (void *) addr;
    517 
    518 		if ((flag & FWRITE) == 0)
    519 			return (EBADF);
    520 
    521 		/* If the ioctl happens here, the parent is us. */
    522 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    523 		return (dkwedge_del(dkw));
    524 	    }
    525 
    526 	case DIOCLWEDGES:
    527 	    {
    528 	    	struct dkwedge_list *dkwl = (void *) addr;
    529 
    530 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
    531 	    }
    532 
    533 	default:
    534 		error = ENOTTY;
    535 		break;
    536 	}
    537 
    538 	return (error);
    539 }
    540 
    541 static void
    542 ldstrategy(struct buf *bp)
    543 {
    544 	struct ld_softc *sc;
    545 	struct disklabel *lp;
    546 	daddr_t blkno;
    547 	int s, part;
    548 
    549 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    550 	part = DISKPART(bp->b_dev);
    551 
    552 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    553 		bp->b_error = EIO;
    554 		goto done;
    555 	}
    556 
    557 	lp = sc->sc_dk.dk_label;
    558 
    559 	/*
    560 	 * The transfer must be a whole number of blocks and the offset must
    561 	 * not be negative.
    562 	 */
    563 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    564 		bp->b_error = EINVAL;
    565 		goto done;
    566 	}
    567 
    568 	/* If it's a null transfer, return immediately. */
    569 	if (bp->b_bcount == 0)
    570 		goto done;
    571 
    572 	/*
    573 	 * Do bounds checking and adjust the transfer.  If error, process.
    574 	 * If past the end of partition, just return.
    575 	 */
    576 	if (part != RAW_PART &&
    577 	    bounds_check_with_label(&sc->sc_dk, bp,
    578 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    579 		goto done;
    580 	}
    581 
    582 	/*
    583 	 * Convert the block number to absolute and put it in terms
    584 	 * of the device's logical block size.
    585 	 */
    586 	if (lp->d_secsize == DEV_BSIZE)
    587 		blkno = bp->b_blkno;
    588 	else if (lp->d_secsize > DEV_BSIZE)
    589 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    590 	else
    591 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    592 
    593 	if (part != RAW_PART)
    594 		blkno += lp->d_partitions[part].p_offset;
    595 
    596 	bp->b_rawblkno = blkno;
    597 
    598 	s = splbio();
    599 	ldstart(sc, bp);
    600 	splx(s);
    601 	return;
    602 
    603  done:
    604 	bp->b_resid = bp->b_bcount;
    605 	biodone(bp);
    606 }
    607 
    608 static void
    609 ldstart(struct ld_softc *sc, struct buf *bp)
    610 {
    611 	int error;
    612 
    613 	mutex_enter(&sc->sc_mutex);
    614 
    615 	if (bp != NULL)
    616 		BUFQ_PUT(sc->sc_bufq, bp);
    617 
    618 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    619 		/* See if there is work to do. */
    620 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    621 			break;
    622 
    623 		disk_busy(&sc->sc_dk);
    624 		sc->sc_queuecnt++;
    625 
    626 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    627 			/*
    628 			 * The back-end is running the job; remove it from
    629 			 * the queue.
    630 			 */
    631 			(void) BUFQ_GET(sc->sc_bufq);
    632 		} else  {
    633 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    634 			sc->sc_queuecnt--;
    635 			if (error == EAGAIN) {
    636 				/*
    637 				 * Temporary resource shortage in the
    638 				 * back-end; just defer the job until
    639 				 * later.
    640 				 *
    641 				 * XXX We might consider a watchdog timer
    642 				 * XXX to make sure we are kicked into action.
    643 				 */
    644 				break;
    645 			} else {
    646 				(void) BUFQ_GET(sc->sc_bufq);
    647 				bp->b_error = error;
    648 				bp->b_resid = bp->b_bcount;
    649 				mutex_exit(&sc->sc_mutex);
    650 				biodone(bp);
    651 				mutex_enter(&sc->sc_mutex);
    652 			}
    653 		}
    654 	}
    655 
    656 	mutex_exit(&sc->sc_mutex);
    657 }
    658 
    659 void
    660 lddone(struct ld_softc *sc, struct buf *bp)
    661 {
    662 
    663 	if (bp->b_error != 0) {
    664 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    665 		printf("\n");
    666 	}
    667 
    668 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    669 	    (bp->b_flags & B_READ));
    670 #if NRND > 0
    671 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    672 #endif
    673 	biodone(bp);
    674 
    675 	mutex_enter(&sc->sc_mutex);
    676 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    677 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    678 			sc->sc_flags &= ~LDF_DRAIN;
    679 			wakeup(&sc->sc_queuecnt);
    680 		}
    681 		mutex_exit(&sc->sc_mutex);
    682 		ldstart(sc, NULL);
    683 	} else
    684 		mutex_exit(&sc->sc_mutex);
    685 }
    686 
    687 static int
    688 ldsize(dev_t dev)
    689 {
    690 	struct ld_softc *sc;
    691 	int part, unit, omask, size;
    692 
    693 	unit = DISKUNIT(dev);
    694 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    695 		return (ENODEV);
    696 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    697 		return (ENODEV);
    698 	part = DISKPART(dev);
    699 
    700 	omask = sc->sc_dk.dk_openmask & (1 << part);
    701 
    702 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    703 		return (-1);
    704 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    705 		size = -1;
    706 	else
    707 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    708 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    709 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    710 		return (-1);
    711 
    712 	return (size);
    713 }
    714 
    715 /*
    716  * Load the label information from the specified device.
    717  */
    718 static void
    719 ldgetdisklabel(struct ld_softc *sc)
    720 {
    721 	const char *errstring;
    722 
    723 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    724 
    725 	/* Call the generic disklabel extraction routine. */
    726 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
    727 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    728 	if (errstring != NULL)
    729 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    730 
    731 	/* In-core label now valid. */
    732 	sc->sc_flags |= LDF_VLABEL;
    733 }
    734 
    735 /*
    736  * Construct a ficticious label.
    737  */
    738 static void
    739 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    740 {
    741 
    742 	memset(lp, 0, sizeof(struct disklabel));
    743 
    744 	lp->d_secsize = sc->sc_secsize;
    745 	lp->d_ntracks = sc->sc_nheads;
    746 	lp->d_nsectors = sc->sc_nsectors;
    747 	lp->d_ncylinders = sc->sc_ncylinders;
    748 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    749 	lp->d_type = DTYPE_LD;
    750 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    751 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    752 	lp->d_secperunit = sc->sc_secperunit;
    753 	lp->d_rpm = 7200;
    754 	lp->d_interleave = 1;
    755 	lp->d_flags = 0;
    756 
    757 	lp->d_partitions[RAW_PART].p_offset = 0;
    758 	lp->d_partitions[RAW_PART].p_size =
    759 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    760 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    761 	lp->d_npartitions = RAW_PART + 1;
    762 
    763 	lp->d_magic = DISKMAGIC;
    764 	lp->d_magic2 = DISKMAGIC;
    765 	lp->d_checksum = dkcksum(lp);
    766 }
    767 
    768 /*
    769  * Take a dump.
    770  */
    771 static int
    772 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
    773 {
    774 	char *va = vav;
    775 	struct ld_softc *sc;
    776 	struct disklabel *lp;
    777 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    778 	static int dumping;
    779 
    780 	unit = DISKUNIT(dev);
    781 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    782 		return (ENXIO);
    783 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    784 		return (ENODEV);
    785 	if (sc->sc_dump == NULL)
    786 		return (ENXIO);
    787 
    788 	/* Check if recursive dump; if so, punt. */
    789 	if (dumping)
    790 		return (EFAULT);
    791 	dumping = 1;
    792 
    793 	/* Convert to disk sectors.  Request must be a multiple of size. */
    794 	part = DISKPART(dev);
    795 	lp = sc->sc_dk.dk_label;
    796 	if ((size % lp->d_secsize) != 0)
    797 		return (EFAULT);
    798 	towrt = size / lp->d_secsize;
    799 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    800 
    801 	nsects = lp->d_partitions[part].p_size;
    802 	sectoff = lp->d_partitions[part].p_offset;
    803 
    804 	/* Check transfer bounds against partition size. */
    805 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    806 		return (EINVAL);
    807 
    808 	/* Offset block number to start of partition. */
    809 	blkno += sectoff;
    810 
    811 	/* Start dumping and return when done. */
    812 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    813 	while (towrt > 0) {
    814 		nblk = min(maxblkcnt, towrt);
    815 
    816 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    817 			return (rv);
    818 
    819 		towrt -= nblk;
    820 		blkno += nblk;
    821 		va += nblk * sc->sc_secsize;
    822 	}
    823 
    824 	dumping = 0;
    825 	return (0);
    826 }
    827 
    828 /*
    829  * Adjust the size of a transfer.
    830  */
    831 static void
    832 ldminphys(struct buf *bp)
    833 {
    834 	struct ld_softc *sc;
    835 
    836 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    837 
    838 	if (bp->b_bcount > sc->sc_maxxfer)
    839 		bp->b_bcount = sc->sc_maxxfer;
    840 	minphys(bp);
    841 }
    842 
    843 static void
    844 ld_set_properties(struct ld_softc *ld)
    845 {
    846 	prop_dictionary_t disk_info, odisk_info, geom;
    847 
    848 	disk_info = prop_dictionary_create();
    849 
    850 	geom = prop_dictionary_create();
    851 
    852 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
    853 	    ld->sc_secperunit);
    854 
    855 	prop_dictionary_set_uint32(geom, "sector-size",
    856 	    ld->sc_secsize);
    857 
    858 	prop_dictionary_set_uint16(geom, "sectors-per-track",
    859 	    ld->sc_nsectors);
    860 
    861 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
    862 	    ld->sc_nheads);
    863 
    864 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
    865 	    ld->sc_ncylinders);
    866 
    867 	prop_dictionary_set(disk_info, "geometry", geom);
    868 	prop_object_release(geom);
    869 
    870 	prop_dictionary_set(device_properties(&ld->sc_dv),
    871 	    "disk-info", disk_info);
    872 
    873 	/*
    874 	 * Don't release disk_info here; we keep a reference to it.
    875 	 * disk_detach() will release it when we go away.
    876 	 */
    877 
    878 	odisk_info = ld->sc_dk.dk_info;
    879 	ld->sc_dk.dk_info = disk_info;
    880 	if (odisk_info)
    881 		prop_object_release(odisk_info);
    882 }
    883 
    884 static void
    885 ld_config_interrupts (struct device *d)
    886 {
    887 	struct ld_softc *sc = (struct ld_softc *)d;
    888 	dkwedge_discover(&sc->sc_dk);
    889 }
    890