Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.81
      1 /*	$NetBSD: ld.c,v 1.81 2015/01/02 19:42:06 christos 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  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Disk driver for use by RAID controllers.
     34  */
     35 
     36 #include <sys/cdefs.h>
     37 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.81 2015/01/02 19:42:06 christos Exp $");
     38 
     39 #include <sys/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/kernel.h>
     42 #include <sys/device.h>
     43 #include <sys/queue.h>
     44 #include <sys/proc.h>
     45 #include <sys/buf.h>
     46 #include <sys/bufq.h>
     47 #include <sys/endian.h>
     48 #include <sys/disklabel.h>
     49 #include <sys/disk.h>
     50 #include <sys/dkio.h>
     51 #include <sys/stat.h>
     52 #include <sys/conf.h>
     53 #include <sys/fcntl.h>
     54 #include <sys/vnode.h>
     55 #include <sys/syslog.h>
     56 #include <sys/mutex.h>
     57 #include <sys/rnd.h>
     58 
     59 #include <dev/ldvar.h>
     60 
     61 #include <prop/proplib.h>
     62 
     63 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     64 static void	ldgetdisklabel(struct ld_softc *);
     65 static void	ldminphys(struct buf *bp);
     66 static bool	ld_suspend(device_t, const pmf_qual_t *);
     67 static bool	ld_shutdown(device_t, int);
     68 static void	ldstart(struct ld_softc *, struct buf *);
     69 static void	ld_set_geometry(struct ld_softc *);
     70 static void	ld_config_interrupts (device_t);
     71 static int	ldlastclose(device_t);
     72 
     73 extern struct	cfdriver ld_cd;
     74 
     75 static dev_type_open(ldopen);
     76 static dev_type_close(ldclose);
     77 static dev_type_read(ldread);
     78 static dev_type_write(ldwrite);
     79 static dev_type_ioctl(ldioctl);
     80 static dev_type_strategy(ldstrategy);
     81 static dev_type_dump(lddump);
     82 static dev_type_size(ldsize);
     83 
     84 const struct bdevsw ld_bdevsw = {
     85 	.d_open = ldopen,
     86 	.d_close = ldclose,
     87 	.d_strategy = ldstrategy,
     88 	.d_ioctl = ldioctl,
     89 	.d_dump = lddump,
     90 	.d_psize = ldsize,
     91 	.d_discard = nodiscard,
     92 	.d_flag = D_DISK
     93 };
     94 
     95 const struct cdevsw ld_cdevsw = {
     96 	.d_open = ldopen,
     97 	.d_close = ldclose,
     98 	.d_read = ldread,
     99 	.d_write = ldwrite,
    100 	.d_ioctl = ldioctl,
    101 	.d_stop = nostop,
    102 	.d_tty = notty,
    103 	.d_poll = nopoll,
    104 	.d_mmap = nommap,
    105 	.d_kqfilter = nokqfilter,
    106 	.d_discard = nodiscard,
    107 	.d_flag = D_DISK
    108 };
    109 
    110 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
    111 
    112 void
    113 ldattach(struct ld_softc *sc)
    114 {
    115 	char tbuf[9];
    116 
    117 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
    118 
    119 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    120 		aprint_normal_dev(sc->sc_dv, "disabled\n");
    121 		return;
    122 	}
    123 
    124 	/* Initialise and attach the disk structure. */
    125 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
    126 	disk_attach(&sc->sc_dk);
    127 
    128 	if (sc->sc_maxxfer > MAXPHYS)
    129 		sc->sc_maxxfer = MAXPHYS;
    130 
    131 	/* Build synthetic geometry if necessary. */
    132 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    133 	    sc->sc_ncylinders == 0) {
    134 		uint64_t ncyl;
    135 
    136 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    137 			sc->sc_nheads = 16;
    138 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    139 			sc->sc_nheads = 32;
    140 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    141 			sc->sc_nheads = 64;
    142 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    143 			sc->sc_nheads = 128;
    144 		else
    145 			sc->sc_nheads = 255;
    146 
    147 		sc->sc_nsectors = 63;
    148 		sc->sc_ncylinders = INT_MAX;
    149 		ncyl = sc->sc_secperunit /
    150 		    (sc->sc_nheads * sc->sc_nsectors);
    151 		if (ncyl < INT_MAX)
    152 			sc->sc_ncylinders = (int)ncyl;
    153 	}
    154 
    155 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    156 	    sc->sc_secsize);
    157 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
    158 	    "%d bytes/sect x %"PRIu64" sectors\n",
    159 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
    160 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    161 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
    162 
    163 	ld_set_geometry(sc);
    164 
    165 	/* Attach the device into the rnd source list. */
    166 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
    167 	    RND_TYPE_DISK, RND_FLAG_DEFAULT);
    168 
    169 	/* Register with PMF */
    170 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
    171 		aprint_error_dev(sc->sc_dv,
    172 		    "couldn't establish power handler\n");
    173 
    174 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    175 
    176 	/* Discover wedges on this disk. */
    177 	config_interrupts(sc->sc_dv, ld_config_interrupts);
    178 }
    179 
    180 int
    181 ldadjqparam(struct ld_softc *sc, int xmax)
    182 {
    183 	int s;
    184 
    185 	s = splbio();
    186 	sc->sc_maxqueuecnt = xmax;
    187 	splx(s);
    188 
    189 	return (0);
    190 }
    191 
    192 int
    193 ldbegindetach(struct ld_softc *sc, int flags)
    194 {
    195 	int s, rv = 0;
    196 
    197 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    198 		return (0);
    199 
    200 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
    201 
    202 	if (rv != 0)
    203 		return rv;
    204 
    205 	s = splbio();
    206 	sc->sc_maxqueuecnt = 0;
    207 	sc->sc_flags |= LDF_DETACH;
    208 	while (sc->sc_queuecnt > 0) {
    209 		sc->sc_flags |= LDF_DRAIN;
    210 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    211 		if (rv)
    212 			break;
    213 	}
    214 	splx(s);
    215 
    216 	return (rv);
    217 }
    218 
    219 void
    220 ldenddetach(struct ld_softc *sc)
    221 {
    222 	int s, bmaj, cmaj, i, mn;
    223 
    224 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    225 		return;
    226 
    227 	/* Wait for commands queued with the hardware to complete. */
    228 	if (sc->sc_queuecnt != 0)
    229 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    230 			printf("%s: not drained\n", device_xname(sc->sc_dv));
    231 
    232 	/* Locate the major numbers. */
    233 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    234 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    235 
    236 	/* Kill off any queued buffers. */
    237 	s = splbio();
    238 	bufq_drain(sc->sc_bufq);
    239 	splx(s);
    240 
    241 	bufq_free(sc->sc_bufq);
    242 
    243 	/* Nuke the vnodes for any open instances. */
    244 	for (i = 0; i < MAXPARTITIONS; i++) {
    245 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
    246 		vdevgone(bmaj, mn, mn, VBLK);
    247 		vdevgone(cmaj, mn, mn, VCHR);
    248 	}
    249 
    250 	/* Delete all of our wedges. */
    251 	dkwedge_delall(&sc->sc_dk);
    252 
    253 	/* Detach from the disk list. */
    254 	disk_detach(&sc->sc_dk);
    255 	disk_destroy(&sc->sc_dk);
    256 
    257 	/* Unhook the entropy source. */
    258 	rnd_detach_source(&sc->sc_rnd_source);
    259 
    260 	/* Deregister with PMF */
    261 	pmf_device_deregister(sc->sc_dv);
    262 
    263 	/*
    264 	 * XXX We can't really flush the cache here, beceause the
    265 	 * XXX device may already be non-existent from the controller's
    266 	 * XXX perspective.
    267 	 */
    268 #if 0
    269 	/* Flush the device's cache. */
    270 	if (sc->sc_flush != NULL)
    271 		if ((*sc->sc_flush)(sc, 0) != 0)
    272 			aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
    273 #endif
    274 	mutex_destroy(&sc->sc_mutex);
    275 }
    276 
    277 /* ARGSUSED */
    278 static bool
    279 ld_suspend(device_t dev, const pmf_qual_t *qual)
    280 {
    281 	return ld_shutdown(dev, 0);
    282 }
    283 
    284 /* ARGSUSED */
    285 static bool
    286 ld_shutdown(device_t dev, int flags)
    287 {
    288 	struct ld_softc *sc = device_private(dev);
    289 
    290 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
    291 		printf("%s: unable to flush cache\n", device_xname(dev));
    292 		return false;
    293 	}
    294 
    295 	return true;
    296 }
    297 
    298 /* ARGSUSED */
    299 static int
    300 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    301 {
    302 	struct ld_softc *sc;
    303 	int error, unit, part;
    304 
    305 	unit = DISKUNIT(dev);
    306 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    307 		return (ENXIO);
    308 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    309 		return (ENODEV);
    310 	part = DISKPART(dev);
    311 
    312 	mutex_enter(&sc->sc_dk.dk_openlock);
    313 
    314 	if (sc->sc_dk.dk_openmask == 0) {
    315 		/* Load the partition info if not already loaded. */
    316 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    317 			ldgetdisklabel(sc);
    318 	}
    319 
    320 	/* Check that the partition exists. */
    321 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    322 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    323 		error = ENXIO;
    324 		goto bad1;
    325 	}
    326 
    327 	/* Ensure only one open at a time. */
    328 	switch (fmt) {
    329 	case S_IFCHR:
    330 		sc->sc_dk.dk_copenmask |= (1 << part);
    331 		break;
    332 	case S_IFBLK:
    333 		sc->sc_dk.dk_bopenmask |= (1 << part);
    334 		break;
    335 	}
    336 	sc->sc_dk.dk_openmask =
    337 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    338 
    339 	error = 0;
    340  bad1:
    341 	mutex_exit(&sc->sc_dk.dk_openlock);
    342 	return (error);
    343 }
    344 
    345 static int
    346 ldlastclose(device_t self)
    347 {
    348 	struct ld_softc *sc = device_private(self);
    349 
    350 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
    351 		aprint_error_dev(self, "unable to flush cache\n");
    352 	if ((sc->sc_flags & LDF_KLABEL) == 0)
    353 		sc->sc_flags &= ~LDF_VLABEL;
    354 
    355 	return 0;
    356 }
    357 
    358 /* ARGSUSED */
    359 static int
    360 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    361 {
    362 	struct ld_softc *sc;
    363 	int part, unit;
    364 
    365 	unit = DISKUNIT(dev);
    366 	part = DISKPART(dev);
    367 	sc = device_lookup_private(&ld_cd, unit);
    368 
    369 	mutex_enter(&sc->sc_dk.dk_openlock);
    370 
    371 	switch (fmt) {
    372 	case S_IFCHR:
    373 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    374 		break;
    375 	case S_IFBLK:
    376 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    377 		break;
    378 	}
    379 	sc->sc_dk.dk_openmask =
    380 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    381 
    382 	if (sc->sc_dk.dk_openmask == 0)
    383 		ldlastclose(sc->sc_dv);
    384 
    385 	mutex_exit(&sc->sc_dk.dk_openlock);
    386 	return (0);
    387 }
    388 
    389 /* ARGSUSED */
    390 static int
    391 ldread(dev_t dev, struct uio *uio, int ioflag)
    392 {
    393 
    394 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    395 }
    396 
    397 /* ARGSUSED */
    398 static int
    399 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    400 {
    401 
    402 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    403 }
    404 
    405 /* ARGSUSED */
    406 static int
    407 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
    408 {
    409 	struct ld_softc *sc;
    410 	int unit, error;
    411 #ifdef __HAVE_OLD_DISKLABEL
    412 	struct disklabel newlabel;
    413 #endif
    414 	struct disklabel *lp;
    415 
    416 	unit = DISKUNIT(dev);
    417 	sc = device_lookup_private(&ld_cd, unit);
    418 
    419 	error = disk_ioctl(&sc->sc_dk, dev, cmd, addr, flag, l);
    420 	if (error != EPASSTHROUGH)
    421 		return (error);
    422 
    423 	error = 0;
    424 	switch (cmd) {
    425 	case DIOCWDINFO:
    426 	case DIOCSDINFO:
    427 #ifdef __HAVE_OLD_DISKLABEL
    428 	case ODIOCWDINFO:
    429 	case ODIOCSDINFO:
    430 
    431 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    432 			memset(&newlabel, 0, sizeof newlabel);
    433 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    434 			lp = &newlabel;
    435 		} else
    436 #endif
    437 		lp = (struct disklabel *)addr;
    438 
    439 		if ((flag & FWRITE) == 0)
    440 			return (EBADF);
    441 
    442 		mutex_enter(&sc->sc_dk.dk_openlock);
    443 		sc->sc_flags |= LDF_LABELLING;
    444 
    445 		error = setdisklabel(sc->sc_dk.dk_label,
    446 		    lp, /*sc->sc_dk.dk_openmask : */0,
    447 		    sc->sc_dk.dk_cpulabel);
    448 		if (error == 0 && (cmd == DIOCWDINFO
    449 #ifdef __HAVE_OLD_DISKLABEL
    450 		    || cmd == ODIOCWDINFO
    451 #endif
    452 		    ))
    453 			error = writedisklabel(
    454 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    455 			    ldstrategy, sc->sc_dk.dk_label,
    456 			    sc->sc_dk.dk_cpulabel);
    457 
    458 		sc->sc_flags &= ~LDF_LABELLING;
    459 		mutex_exit(&sc->sc_dk.dk_openlock);
    460 		break;
    461 
    462 	case DIOCKLABEL:
    463 		if ((flag & FWRITE) == 0)
    464 			return (EBADF);
    465 		if (*(int *)addr)
    466 			sc->sc_flags |= LDF_KLABEL;
    467 		else
    468 			sc->sc_flags &= ~LDF_KLABEL;
    469 		break;
    470 
    471 	case DIOCWLABEL:
    472 		if ((flag & FWRITE) == 0)
    473 			return (EBADF);
    474 		if (*(int *)addr)
    475 			sc->sc_flags |= LDF_WLABEL;
    476 		else
    477 			sc->sc_flags &= ~LDF_WLABEL;
    478 		break;
    479 
    480 	case DIOCGDEFLABEL:
    481 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    482 		break;
    483 
    484 #ifdef __HAVE_OLD_DISKLABEL
    485 	case ODIOCGDEFLABEL:
    486 		ldgetdefaultlabel(sc, &newlabel);
    487 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    488 			return ENOTTY;
    489 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    490 		break;
    491 #endif
    492 
    493 	case DIOCCACHESYNC:
    494 		/*
    495 		 * XXX Do we really need to care about having a writable
    496 		 * file descriptor here?
    497 		 */
    498 		if ((flag & FWRITE) == 0)
    499 			error = EBADF;
    500 		else if (sc->sc_flush)
    501 			error = (*sc->sc_flush)(sc, 0);
    502 		else
    503 			error = 0;	/* XXX Error out instead? */
    504 		break;
    505 
    506 	case DIOCGSTRATEGY:
    507 	    {
    508 		struct disk_strategy *dks = (void *)addr;
    509 
    510 		mutex_enter(&sc->sc_mutex);
    511 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
    512 		    sizeof(dks->dks_name));
    513 		mutex_exit(&sc->sc_mutex);
    514 		dks->dks_paramlen = 0;
    515 
    516 		return 0;
    517 	    }
    518 	case DIOCSSTRATEGY:
    519 	    {
    520 		struct disk_strategy *dks = (void *)addr;
    521 		struct bufq_state *new_bufq, *old_bufq;
    522 
    523 		if ((flag & FWRITE) == 0)
    524 			return EPERM;
    525 
    526 		if (dks->dks_param != NULL)
    527 			return EINVAL;
    528 
    529 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
    530 		error = bufq_alloc(&new_bufq, dks->dks_name,
    531 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
    532 		if (error)
    533 			return error;
    534 
    535 		mutex_enter(&sc->sc_mutex);
    536 		old_bufq = sc->sc_bufq;
    537 		bufq_move(new_bufq, old_bufq);
    538 		sc->sc_bufq = new_bufq;
    539 		mutex_exit(&sc->sc_mutex);
    540 		bufq_free(old_bufq);
    541 
    542 		return 0;
    543 	    }
    544 	default:
    545 		error = ENOTTY;
    546 		break;
    547 	}
    548 
    549 	return (error);
    550 }
    551 
    552 static void
    553 ldstrategy(struct buf *bp)
    554 {
    555 	struct ld_softc *sc;
    556 	struct disklabel *lp;
    557 	daddr_t blkno;
    558 	int s, part;
    559 
    560 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
    561 	part = DISKPART(bp->b_dev);
    562 
    563 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    564 		bp->b_error = EIO;
    565 		goto done;
    566 	}
    567 
    568 	lp = sc->sc_dk.dk_label;
    569 
    570 	/*
    571 	 * The transfer must be a whole number of blocks and the offset must
    572 	 * not be negative.
    573 	 */
    574 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    575 		bp->b_error = EINVAL;
    576 		goto done;
    577 	}
    578 
    579 	/* If it's a null transfer, return immediately. */
    580 	if (bp->b_bcount == 0)
    581 		goto done;
    582 
    583 	/*
    584 	 * Do bounds checking and adjust the transfer.  If error, process.
    585 	 * If past the end of partition, just return.
    586 	 */
    587 	if (part == RAW_PART) {
    588 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    589 		    sc->sc_disksize512) <= 0)
    590 			goto done;
    591 	} else {
    592 		if (bounds_check_with_label(&sc->sc_dk, bp,
    593 		    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
    594 			goto done;
    595 	}
    596 
    597 	/*
    598 	 * Convert the block number to absolute and put it in terms
    599 	 * of the device's logical block size.
    600 	 */
    601 	if (lp->d_secsize == DEV_BSIZE)
    602 		blkno = bp->b_blkno;
    603 	else if (lp->d_secsize > DEV_BSIZE)
    604 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    605 	else
    606 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    607 
    608 	if (part != RAW_PART)
    609 		blkno += lp->d_partitions[part].p_offset;
    610 
    611 	bp->b_rawblkno = blkno;
    612 
    613 	s = splbio();
    614 	ldstart(sc, bp);
    615 	splx(s);
    616 	return;
    617 
    618  done:
    619 	bp->b_resid = bp->b_bcount;
    620 	biodone(bp);
    621 }
    622 
    623 static void
    624 ldstart(struct ld_softc *sc, struct buf *bp)
    625 {
    626 	int error;
    627 
    628 	mutex_enter(&sc->sc_mutex);
    629 
    630 	if (bp != NULL)
    631 		bufq_put(sc->sc_bufq, bp);
    632 
    633 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    634 		/* See if there is work to do. */
    635 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
    636 			break;
    637 
    638 		disk_busy(&sc->sc_dk);
    639 		sc->sc_queuecnt++;
    640 
    641 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    642 			/*
    643 			 * The back-end is running the job; remove it from
    644 			 * the queue.
    645 			 */
    646 			(void) bufq_get(sc->sc_bufq);
    647 		} else  {
    648 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    649 			sc->sc_queuecnt--;
    650 			if (error == EAGAIN) {
    651 				/*
    652 				 * Temporary resource shortage in the
    653 				 * back-end; just defer the job until
    654 				 * later.
    655 				 *
    656 				 * XXX We might consider a watchdog timer
    657 				 * XXX to make sure we are kicked into action.
    658 				 */
    659 				break;
    660 			} else {
    661 				(void) bufq_get(sc->sc_bufq);
    662 				bp->b_error = error;
    663 				bp->b_resid = bp->b_bcount;
    664 				mutex_exit(&sc->sc_mutex);
    665 				biodone(bp);
    666 				mutex_enter(&sc->sc_mutex);
    667 			}
    668 		}
    669 	}
    670 
    671 	mutex_exit(&sc->sc_mutex);
    672 }
    673 
    674 void
    675 lddone(struct ld_softc *sc, struct buf *bp)
    676 {
    677 
    678 	if (bp->b_error != 0) {
    679 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    680 		printf("\n");
    681 	}
    682 
    683 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    684 	    (bp->b_flags & B_READ));
    685 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    686 	biodone(bp);
    687 
    688 	mutex_enter(&sc->sc_mutex);
    689 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    690 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    691 			sc->sc_flags &= ~LDF_DRAIN;
    692 			wakeup(&sc->sc_queuecnt);
    693 		}
    694 		mutex_exit(&sc->sc_mutex);
    695 		ldstart(sc, NULL);
    696 	} else
    697 		mutex_exit(&sc->sc_mutex);
    698 }
    699 
    700 static int
    701 ldsize(dev_t dev)
    702 {
    703 	struct ld_softc *sc;
    704 	int part, unit, omask, size;
    705 
    706 	unit = DISKUNIT(dev);
    707 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    708 		return (ENODEV);
    709 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    710 		return (ENODEV);
    711 	part = DISKPART(dev);
    712 
    713 	omask = sc->sc_dk.dk_openmask & (1 << part);
    714 
    715 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    716 		return (-1);
    717 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    718 		size = -1;
    719 	else
    720 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    721 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    722 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    723 		return (-1);
    724 
    725 	return (size);
    726 }
    727 
    728 /*
    729  * Load the label information from the specified device.
    730  */
    731 static void
    732 ldgetdisklabel(struct ld_softc *sc)
    733 {
    734 	const char *errstring;
    735 
    736 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    737 
    738 	/* Call the generic disklabel extraction routine. */
    739 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
    740 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    741 	if (errstring != NULL)
    742 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
    743 
    744 	/* In-core label now valid. */
    745 	sc->sc_flags |= LDF_VLABEL;
    746 }
    747 
    748 /*
    749  * Construct a ficticious label.
    750  */
    751 static void
    752 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    753 {
    754 
    755 	memset(lp, 0, sizeof(struct disklabel));
    756 
    757 	lp->d_secsize = sc->sc_secsize;
    758 	lp->d_ntracks = sc->sc_nheads;
    759 	lp->d_nsectors = sc->sc_nsectors;
    760 	lp->d_ncylinders = sc->sc_ncylinders;
    761 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    762 	lp->d_type = DKTYPE_LD;
    763 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    764 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    765 	if (sc->sc_secperunit > UINT32_MAX)
    766 		lp->d_secperunit = UINT32_MAX;
    767 	else
    768 		lp->d_secperunit = sc->sc_secperunit;
    769 	lp->d_rpm = 7200;
    770 	lp->d_interleave = 1;
    771 	lp->d_flags = 0;
    772 
    773 	lp->d_partitions[RAW_PART].p_offset = 0;
    774 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
    775 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    776 	lp->d_npartitions = RAW_PART + 1;
    777 
    778 	lp->d_magic = DISKMAGIC;
    779 	lp->d_magic2 = DISKMAGIC;
    780 	lp->d_checksum = dkcksum(lp);
    781 }
    782 
    783 /*
    784  * Take a dump.
    785  */
    786 static int
    787 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
    788 {
    789 	char *va = vav;
    790 	struct ld_softc *sc;
    791 	struct disklabel *lp;
    792 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    793 	static int dumping;
    794 
    795 	unit = DISKUNIT(dev);
    796 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    797 		return (ENXIO);
    798 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    799 		return (ENODEV);
    800 	if (sc->sc_dump == NULL)
    801 		return (ENXIO);
    802 
    803 	/* Check if recursive dump; if so, punt. */
    804 	if (dumping)
    805 		return (EFAULT);
    806 	dumping = 1;
    807 
    808 	/* Convert to disk sectors.  Request must be a multiple of size. */
    809 	part = DISKPART(dev);
    810 	lp = sc->sc_dk.dk_label;
    811 	if ((size % lp->d_secsize) != 0)
    812 		return (EFAULT);
    813 	towrt = size / lp->d_secsize;
    814 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    815 
    816 	nsects = lp->d_partitions[part].p_size;
    817 	sectoff = lp->d_partitions[part].p_offset;
    818 
    819 	/* Check transfer bounds against partition size. */
    820 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    821 		return (EINVAL);
    822 
    823 	/* Offset block number to start of partition. */
    824 	blkno += sectoff;
    825 
    826 	/* Start dumping and return when done. */
    827 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    828 	while (towrt > 0) {
    829 		nblk = min(maxblkcnt, towrt);
    830 
    831 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    832 			return (rv);
    833 
    834 		towrt -= nblk;
    835 		blkno += nblk;
    836 		va += nblk * sc->sc_secsize;
    837 	}
    838 
    839 	dumping = 0;
    840 	return (0);
    841 }
    842 
    843 /*
    844  * Adjust the size of a transfer.
    845  */
    846 static void
    847 ldminphys(struct buf *bp)
    848 {
    849 	struct ld_softc *sc;
    850 
    851 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
    852 
    853 	if (bp->b_bcount > sc->sc_maxxfer)
    854 		bp->b_bcount = sc->sc_maxxfer;
    855 	minphys(bp);
    856 }
    857 
    858 static void
    859 ld_set_geometry(struct ld_softc *ld)
    860 {
    861 	struct disk_geom *dg = &ld->sc_dk.dk_geom;
    862 
    863 	memset(dg, 0, sizeof(*dg));
    864 
    865 	dg->dg_secperunit = ld->sc_secperunit;
    866 	dg->dg_secsize = ld->sc_secsize;
    867 	dg->dg_nsectors = ld->sc_nsectors;
    868 	dg->dg_ntracks = ld->sc_nheads;
    869 	dg->dg_ncylinders = ld->sc_ncylinders;
    870 
    871 	disk_set_info(ld->sc_dv, &ld->sc_dk, NULL);
    872 }
    873 
    874 static void
    875 ld_config_interrupts(device_t d)
    876 {
    877 	struct ld_softc *sc = device_private(d);
    878 	dkwedge_discover(&sc->sc_dk);
    879 }
    880