Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.44
      1 /*	$NetBSD: ld.c,v 1.44 2007/02/09 21:55:26 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 1.44 2007/02/09 21:55:26 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 
     82 extern struct	cfdriver ld_cd;
     83 
     84 static dev_type_open(ldopen);
     85 static dev_type_close(ldclose);
     86 static dev_type_read(ldread);
     87 static dev_type_write(ldwrite);
     88 static dev_type_ioctl(ldioctl);
     89 static dev_type_strategy(ldstrategy);
     90 static dev_type_dump(lddump);
     91 static dev_type_size(ldsize);
     92 
     93 const struct bdevsw ld_bdevsw = {
     94 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
     95 };
     96 
     97 const struct cdevsw ld_cdevsw = {
     98 	ldopen, ldclose, ldread, ldwrite, ldioctl,
     99 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    100 };
    101 
    102 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
    103 static void	*ld_sdh;
    104 
    105 void
    106 ldattach(struct ld_softc *sc)
    107 {
    108 	char tbuf[9];
    109 
    110 	mutex_init(&sc->sc_mutex, MUTEX_DRIVER, IPL_BIO);
    111 
    112 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    113 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
    114 		return;
    115 	}
    116 
    117 	/* Initialise and attach the disk structure. */
    118 	sc->sc_dk.dk_driver = &lddkdriver;
    119 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
    120 	disk_attach(&sc->sc_dk);
    121 
    122 	if (sc->sc_maxxfer > MAXPHYS)
    123 		sc->sc_maxxfer = MAXPHYS;
    124 
    125 	/* Build synthetic geometry if necessary. */
    126 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    127 	    sc->sc_ncylinders == 0) {
    128 		uint64_t ncyl;
    129 
    130 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    131 			sc->sc_nheads = 16;
    132 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    133 			sc->sc_nheads = 32;
    134 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    135 			sc->sc_nheads = 64;
    136 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    137 			sc->sc_nheads = 128;
    138 		else
    139 			sc->sc_nheads = 255;
    140 
    141 		sc->sc_nsectors = 63;
    142 		sc->sc_ncylinders = INT_MAX;
    143 		ncyl = sc->sc_secperunit /
    144 		    (sc->sc_nheads * sc->sc_nsectors);
    145 		if (ncyl < INT_MAX)
    146 			sc->sc_ncylinders = (int)ncyl;
    147 	}
    148 
    149 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    150 	    sc->sc_secsize);
    151 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    152 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
    153 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    154 
    155 	ld_set_properties(sc);
    156 
    157 #if NRND > 0
    158 	/* Attach the device into the rnd source list. */
    159 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    160 	    RND_TYPE_DISK, 0);
    161 #endif
    162 
    163 	/* Set the `shutdownhook'. */
    164 	if (ld_sdh == NULL)
    165 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
    166 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    167 
    168 	/* Discover wedges on this disk. */
    169 	dkwedge_discover(&sc->sc_dk);
    170 }
    171 
    172 int
    173 ldadjqparam(struct ld_softc *sc, int xmax)
    174 {
    175 	int s;
    176 
    177 	s = splbio();
    178 	sc->sc_maxqueuecnt = xmax;
    179 	splx(s);
    180 
    181 	return (0);
    182 }
    183 
    184 int
    185 ldbegindetach(struct ld_softc *sc, int flags)
    186 {
    187 	int s, rv = 0;
    188 
    189 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    190 		return (0);
    191 
    192 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    193 		return (EBUSY);
    194 
    195 	s = splbio();
    196 	sc->sc_maxqueuecnt = 0;
    197 	sc->sc_flags |= LDF_DETACH;
    198 	while (sc->sc_queuecnt > 0) {
    199 		sc->sc_flags |= LDF_DRAIN;
    200 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    201 		if (rv)
    202 			break;
    203 	}
    204 	splx(s);
    205 
    206 	return (rv);
    207 }
    208 
    209 void
    210 ldenddetach(struct ld_softc *sc)
    211 {
    212 	int s, bmaj, cmaj, i, mn;
    213 
    214 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    215 		return;
    216 
    217 	/* Wait for commands queued with the hardware to complete. */
    218 	if (sc->sc_queuecnt != 0)
    219 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    220 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
    221 
    222 	/* Locate the major numbers. */
    223 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    224 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    225 
    226 	/* Kill off any queued buffers. */
    227 	s = splbio();
    228 	bufq_drain(sc->sc_bufq);
    229 	splx(s);
    230 
    231 	bufq_free(sc->sc_bufq);
    232 
    233 	/* Nuke the vnodes for any open instances. */
    234 	for (i = 0; i < MAXPARTITIONS; i++) {
    235 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
    236 		vdevgone(bmaj, mn, mn, VBLK);
    237 		vdevgone(cmaj, mn, mn, VCHR);
    238 	}
    239 
    240 	/* Delete all of our wedges. */
    241 	dkwedge_delall(&sc->sc_dk);
    242 
    243 	/* Detach from the disk list. */
    244 	disk_detach(&sc->sc_dk);
    245 
    246 #if NRND > 0
    247 	/* Unhook the entropy source. */
    248 	rnd_detach_source(&sc->sc_rnd_source);
    249 #endif
    250 
    251 	/*
    252 	 * XXX We can't really flush the cache here, beceause the
    253 	 * XXX device may already be non-existent from the controller's
    254 	 * XXX perspective.
    255 	 */
    256 #if 0
    257 	/* Flush the device's cache. */
    258 	if (sc->sc_flush != NULL)
    259 		if ((*sc->sc_flush)(sc) != 0)
    260 			printf("%s: unable to flush cache\n",
    261 			    sc->sc_dv.dv_xname);
    262 #endif
    263 }
    264 
    265 /* ARGSUSED */
    266 static void
    267 ldshutdown(void *cookie)
    268 {
    269 	struct ld_softc *sc;
    270 	int i;
    271 
    272 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    273 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    274 			continue;
    275 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    276 			printf("%s: unable to flush cache\n",
    277 			    sc->sc_dv.dv_xname);
    278 	}
    279 }
    280 
    281 /* ARGSUSED */
    282 static int
    283 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    284 {
    285 	struct ld_softc *sc;
    286 	int error, unit, part;
    287 
    288 	unit = DISKUNIT(dev);
    289 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    290 		return (ENXIO);
    291 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    292 		return (ENODEV);
    293 	part = DISKPART(dev);
    294 
    295 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    296 		return (error);
    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 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    324 	return (0);
    325 
    326  bad1:
    327 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    328 	return (error);
    329 }
    330 
    331 /* ARGSUSED */
    332 static int
    333 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    334 {
    335 	struct ld_softc *sc;
    336 	int error, part, unit;
    337 
    338 	unit = DISKUNIT(dev);
    339 	part = DISKPART(dev);
    340 	sc = device_lookup(&ld_cd, unit);
    341 
    342 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    343 		return (error);
    344 
    345 	switch (fmt) {
    346 	case S_IFCHR:
    347 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    348 		break;
    349 	case S_IFBLK:
    350 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    351 		break;
    352 	}
    353 	sc->sc_dk.dk_openmask =
    354 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    355 
    356 	if (sc->sc_dk.dk_openmask == 0) {
    357 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    358 			printf("%s: unable to flush cache\n",
    359 			    sc->sc_dv.dv_xname);
    360 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    361 			sc->sc_flags &= ~LDF_VLABEL;
    362 	}
    363 
    364 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    365 	return (0);
    366 }
    367 
    368 /* ARGSUSED */
    369 static int
    370 ldread(dev_t dev, struct uio *uio, int ioflag)
    371 {
    372 
    373 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    374 }
    375 
    376 /* ARGSUSED */
    377 static int
    378 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    379 {
    380 
    381 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    382 }
    383 
    384 /* ARGSUSED */
    385 static int
    386 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct lwp *l)
    387 {
    388 	struct ld_softc *sc;
    389 	int part, unit, error;
    390 #ifdef __HAVE_OLD_DISKLABEL
    391 	struct disklabel newlabel;
    392 #endif
    393 	struct disklabel *lp;
    394 
    395 	unit = DISKUNIT(dev);
    396 	part = DISKPART(dev);
    397 	sc = device_lookup(&ld_cd, unit);
    398 	error = 0;
    399 
    400 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
    401 	if (error != EPASSTHROUGH)
    402 		return (error);
    403 
    404 	switch (cmd) {
    405 	case DIOCGDINFO:
    406 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    407 		return (0);
    408 
    409 #ifdef __HAVE_OLD_DISKLABEL
    410 	case ODIOCGDINFO:
    411 		newlabel = *(sc->sc_dk.dk_label);
    412 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    413 			return ENOTTY;
    414 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    415 		return (0);
    416 #endif
    417 
    418 	case DIOCGPART:
    419 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    420 		((struct partinfo *)addr)->part =
    421 		    &sc->sc_dk.dk_label->d_partitions[part];
    422 		break;
    423 
    424 	case DIOCWDINFO:
    425 	case DIOCSDINFO:
    426 #ifdef __HAVE_OLD_DISKLABEL
    427 	case ODIOCWDINFO:
    428 	case ODIOCSDINFO:
    429 
    430 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    431 			memset(&newlabel, 0, sizeof newlabel);
    432 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    433 			lp = &newlabel;
    434 		} else
    435 #endif
    436 		lp = (struct disklabel *)addr;
    437 
    438 		if ((flag & FWRITE) == 0)
    439 			return (EBADF);
    440 
    441 		if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE,
    442 				     NULL)) != 0)
    443 			return (error);
    444 		sc->sc_flags |= LDF_LABELLING;
    445 
    446 		error = setdisklabel(sc->sc_dk.dk_label,
    447 		    lp, /*sc->sc_dk.dk_openmask : */0,
    448 		    sc->sc_dk.dk_cpulabel);
    449 		if (error == 0 && (cmd == DIOCWDINFO
    450 #ifdef __HAVE_OLD_DISKLABEL
    451 		    || cmd == ODIOCWDINFO
    452 #endif
    453 		    ))
    454 			error = writedisklabel(
    455 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    456 			    ldstrategy, sc->sc_dk.dk_label,
    457 			    sc->sc_dk.dk_cpulabel);
    458 
    459 		sc->sc_flags &= ~LDF_LABELLING;
    460 		(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    461 		break;
    462 
    463 	case DIOCKLABEL:
    464 		if ((flag & FWRITE) == 0)
    465 			return (EBADF);
    466 		if (*(int *)addr)
    467 			sc->sc_flags |= LDF_KLABEL;
    468 		else
    469 			sc->sc_flags &= ~LDF_KLABEL;
    470 		break;
    471 
    472 	case DIOCWLABEL:
    473 		if ((flag & FWRITE) == 0)
    474 			return (EBADF);
    475 		if (*(int *)addr)
    476 			sc->sc_flags |= LDF_WLABEL;
    477 		else
    478 			sc->sc_flags &= ~LDF_WLABEL;
    479 		break;
    480 
    481 	case DIOCGDEFLABEL:
    482 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    483 		break;
    484 
    485 #ifdef __HAVE_OLD_DISKLABEL
    486 	case ODIOCGDEFLABEL:
    487 		ldgetdefaultlabel(sc, &newlabel);
    488 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    489 			return ENOTTY;
    490 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    491 		break;
    492 #endif
    493 
    494 	case DIOCCACHESYNC:
    495 		/*
    496 		 * XXX Do we really need to care about having a writable
    497 		 * file descriptor here?
    498 		 */
    499 		if ((flag & FWRITE) == 0)
    500 			error = EBADF;
    501 		else if (sc->sc_flush)
    502 			error = (*sc->sc_flush)(sc);
    503 		else
    504 			error = 0;	/* XXX Error out instead? */
    505 		break;
    506 
    507 	case DIOCAWEDGE:
    508 	    {
    509 	    	struct dkwedge_info *dkw = (void *) addr;
    510 
    511 		if ((flag & FWRITE) == 0)
    512 			return (EBADF);
    513 
    514 		/* If the ioctl happens here, the parent is us. */
    515 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    516 		return (dkwedge_add(dkw));
    517 	    }
    518 
    519 	case DIOCDWEDGE:
    520 	    {
    521 	    	struct dkwedge_info *dkw = (void *) addr;
    522 
    523 		if ((flag & FWRITE) == 0)
    524 			return (EBADF);
    525 
    526 		/* If the ioctl happens here, the parent is us. */
    527 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    528 		return (dkwedge_del(dkw));
    529 	    }
    530 
    531 	case DIOCLWEDGES:
    532 	    {
    533 	    	struct dkwedge_list *dkwl = (void *) addr;
    534 
    535 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
    536 	    }
    537 
    538 	default:
    539 		error = ENOTTY;
    540 		break;
    541 	}
    542 
    543 	return (error);
    544 }
    545 
    546 static void
    547 ldstrategy(struct buf *bp)
    548 {
    549 	struct ld_softc *sc;
    550 	struct disklabel *lp;
    551 	daddr_t blkno;
    552 	int s, part;
    553 
    554 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    555 	part = DISKPART(bp->b_dev);
    556 
    557 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    558 		bp->b_error = EIO;
    559 		goto bad;
    560 	}
    561 
    562 	lp = sc->sc_dk.dk_label;
    563 
    564 	/*
    565 	 * The transfer must be a whole number of blocks and the offset must
    566 	 * not be negative.
    567 	 */
    568 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    569 		bp->b_error = EINVAL;
    570 		goto bad;
    571 	}
    572 
    573 	/* If it's a null transfer, return immediately. */
    574 	if (bp->b_bcount == 0)
    575 		goto done;
    576 
    577 	/*
    578 	 * Do bounds checking and adjust the transfer.  If error, process.
    579 	 * If past the end of partition, just return.
    580 	 */
    581 	if (part != RAW_PART &&
    582 	    bounds_check_with_label(&sc->sc_dk, bp,
    583 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    584 		goto done;
    585 	}
    586 
    587 	/*
    588 	 * Convert the block number to absolute and put it in terms
    589 	 * of the device's logical block size.
    590 	 */
    591 	if (lp->d_secsize == DEV_BSIZE)
    592 		blkno = bp->b_blkno;
    593 	else if (lp->d_secsize > DEV_BSIZE)
    594 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    595 	else
    596 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    597 
    598 	if (part != RAW_PART)
    599 		blkno += lp->d_partitions[part].p_offset;
    600 
    601 	bp->b_rawblkno = blkno;
    602 
    603 	s = splbio();
    604 	ldstart(sc, bp);
    605 	splx(s);
    606 	return;
    607 
    608  bad:
    609 	bp->b_flags |= B_ERROR;
    610  done:
    611 	bp->b_resid = bp->b_bcount;
    612 	biodone(bp);
    613 }
    614 
    615 static void
    616 ldstart(struct ld_softc *sc, struct buf *bp)
    617 {
    618 	int error;
    619 
    620 	mutex_enter(&sc->sc_mutex);
    621 
    622 	if (bp != NULL)
    623 		BUFQ_PUT(sc->sc_bufq, bp);
    624 
    625 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    626 		/* See if there is work to do. */
    627 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    628 			break;
    629 
    630 		disk_busy(&sc->sc_dk);
    631 		sc->sc_queuecnt++;
    632 
    633 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    634 			/*
    635 			 * The back-end is running the job; remove it from
    636 			 * the queue.
    637 			 */
    638 			(void) BUFQ_GET(sc->sc_bufq);
    639 		} else  {
    640 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    641 			sc->sc_queuecnt--;
    642 			if (error == EAGAIN) {
    643 				/*
    644 				 * Temporary resource shortage in the
    645 				 * back-end; just defer the job until
    646 				 * later.
    647 				 *
    648 				 * XXX We might consider a watchdog timer
    649 				 * XXX to make sure we are kicked into action.
    650 				 */
    651 				break;
    652 			} else {
    653 				(void) BUFQ_GET(sc->sc_bufq);
    654 				bp->b_error = error;
    655 				bp->b_flags |= B_ERROR;
    656 				bp->b_resid = bp->b_bcount;
    657 				mutex_exit(&sc->sc_mutex);
    658 				biodone(bp);
    659 				mutex_enter(&sc->sc_mutex);
    660 			}
    661 		}
    662 	}
    663 
    664 	mutex_exit(&sc->sc_mutex);
    665 }
    666 
    667 void
    668 lddone(struct ld_softc *sc, struct buf *bp)
    669 {
    670 
    671 	if ((bp->b_flags & B_ERROR) != 0) {
    672 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    673 		printf("\n");
    674 	}
    675 
    676 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    677 	    (bp->b_flags & B_READ));
    678 #if NRND > 0
    679 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    680 #endif
    681 	biodone(bp);
    682 
    683 	mutex_enter(&sc->sc_mutex);
    684 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    685 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    686 			sc->sc_flags &= ~LDF_DRAIN;
    687 			wakeup(&sc->sc_queuecnt);
    688 		}
    689 		mutex_exit(&sc->sc_mutex);
    690 		ldstart(sc, NULL);
    691 	} else
    692 		mutex_exit(&sc->sc_mutex);
    693 }
    694 
    695 static int
    696 ldsize(dev_t dev)
    697 {
    698 	struct ld_softc *sc;
    699 	int part, unit, omask, size;
    700 
    701 	unit = DISKUNIT(dev);
    702 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    703 		return (ENODEV);
    704 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    705 		return (ENODEV);
    706 	part = DISKPART(dev);
    707 
    708 	omask = sc->sc_dk.dk_openmask & (1 << part);
    709 
    710 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    711 		return (-1);
    712 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    713 		size = -1;
    714 	else
    715 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    716 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    717 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    718 		return (-1);
    719 
    720 	return (size);
    721 }
    722 
    723 /*
    724  * Load the label information from the specified device.
    725  */
    726 static void
    727 ldgetdisklabel(struct ld_softc *sc)
    728 {
    729 	const char *errstring;
    730 
    731 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    732 
    733 	/* Call the generic disklabel extraction routine. */
    734 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
    735 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    736 	if (errstring != NULL)
    737 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    738 
    739 	/* In-core label now valid. */
    740 	sc->sc_flags |= LDF_VLABEL;
    741 }
    742 
    743 /*
    744  * Construct a ficticious label.
    745  */
    746 static void
    747 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    748 {
    749 
    750 	memset(lp, 0, sizeof(struct disklabel));
    751 
    752 	lp->d_secsize = sc->sc_secsize;
    753 	lp->d_ntracks = sc->sc_nheads;
    754 	lp->d_nsectors = sc->sc_nsectors;
    755 	lp->d_ncylinders = sc->sc_ncylinders;
    756 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    757 	lp->d_type = DTYPE_LD;
    758 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    759 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    760 	lp->d_secperunit = sc->sc_secperunit;
    761 	lp->d_rpm = 7200;
    762 	lp->d_interleave = 1;
    763 	lp->d_flags = 0;
    764 
    765 	lp->d_partitions[RAW_PART].p_offset = 0;
    766 	lp->d_partitions[RAW_PART].p_size =
    767 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    768 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    769 	lp->d_npartitions = RAW_PART + 1;
    770 
    771 	lp->d_magic = DISKMAGIC;
    772 	lp->d_magic2 = DISKMAGIC;
    773 	lp->d_checksum = dkcksum(lp);
    774 }
    775 
    776 /*
    777  * Take a dump.
    778  */
    779 static int
    780 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
    781 {
    782 	struct ld_softc *sc;
    783 	struct disklabel *lp;
    784 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    785 	static int dumping;
    786 
    787 	unit = DISKUNIT(dev);
    788 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    789 		return (ENXIO);
    790 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    791 		return (ENODEV);
    792 	if (sc->sc_dump == NULL)
    793 		return (ENXIO);
    794 
    795 	/* Check if recursive dump; if so, punt. */
    796 	if (dumping)
    797 		return (EFAULT);
    798 	dumping = 1;
    799 
    800 	/* Convert to disk sectors.  Request must be a multiple of size. */
    801 	part = DISKPART(dev);
    802 	lp = sc->sc_dk.dk_label;
    803 	if ((size % lp->d_secsize) != 0)
    804 		return (EFAULT);
    805 	towrt = size / lp->d_secsize;
    806 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    807 
    808 	nsects = lp->d_partitions[part].p_size;
    809 	sectoff = lp->d_partitions[part].p_offset;
    810 
    811 	/* Check transfer bounds against partition size. */
    812 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    813 		return (EINVAL);
    814 
    815 	/* Offset block number to start of partition. */
    816 	blkno += sectoff;
    817 
    818 	/* Start dumping and return when done. */
    819 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    820 	while (towrt > 0) {
    821 		nblk = min(maxblkcnt, towrt);
    822 
    823 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    824 			return (rv);
    825 
    826 		towrt -= nblk;
    827 		blkno += nblk;
    828 		va += nblk * sc->sc_secsize;
    829 	}
    830 
    831 	dumping = 0;
    832 	return (0);
    833 }
    834 
    835 /*
    836  * Adjust the size of a transfer.
    837  */
    838 static void
    839 ldminphys(struct buf *bp)
    840 {
    841 	struct ld_softc *sc;
    842 
    843 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    844 
    845 	if (bp->b_bcount > sc->sc_maxxfer)
    846 		bp->b_bcount = sc->sc_maxxfer;
    847 	minphys(bp);
    848 }
    849 
    850 static void
    851 ld_set_properties(struct ld_softc *ld)
    852 {
    853 	prop_dictionary_t disk_info, odisk_info, geom;
    854 
    855 	disk_info = prop_dictionary_create();
    856 
    857 	geom = prop_dictionary_create();
    858 
    859 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
    860 	    ld->sc_secperunit);
    861 
    862 	prop_dictionary_set_uint32(geom, "sector-size",
    863 	    ld->sc_secsize);
    864 
    865 	prop_dictionary_set_uint16(geom, "sectors-per-track",
    866 	    ld->sc_nsectors);
    867 
    868 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
    869 	    ld->sc_nheads);
    870 
    871 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
    872 	    ld->sc_ncylinders);
    873 
    874 	prop_dictionary_set(disk_info, "geometry", geom);
    875 	prop_object_release(geom);
    876 
    877 	prop_dictionary_set(device_properties(&ld->sc_dv),
    878 	    "disk-info", disk_info);
    879 
    880 	/*
    881 	 * Don't release disk_info here; we keep a reference to it.
    882 	 * disk_detach() will release it when we go away.
    883 	 */
    884 
    885 	odisk_info = ld->sc_dk.dk_info;
    886 	ld->sc_dk.dk_info = disk_info;
    887 	if (odisk_info)
    888 		prop_object_release(odisk_info);
    889 }
    890