Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.55
      1 /*	$NetBSD: ld.c,v 1.55 2008/03/09 19:15:01 jmcneill 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 1.55 2008/03/09 19:15:01 jmcneill 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/conf.h>
     62 #include <sys/fcntl.h>
     63 #include <sys/vnode.h>
     64 #include <sys/syslog.h>
     65 #include <sys/mutex.h>
     66 #if NRND > 0
     67 #include <sys/rnd.h>
     68 #endif
     69 
     70 #include <dev/ldvar.h>
     71 
     72 #include <prop/proplib.h>
     73 
     74 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     75 static void	ldgetdisklabel(struct ld_softc *);
     76 static void	ldminphys(struct buf *bp);
     77 static bool	ld_shutdown(device_t, int);
     78 static void	ldstart(struct ld_softc *, struct buf *);
     79 static void	ld_set_properties(struct ld_softc *);
     80 static void	ld_config_interrupts (struct device *);
     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 
    104 void
    105 ldattach(struct ld_softc *sc)
    106 {
    107 	char tbuf[9];
    108 
    109 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
    110 
    111 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    112 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
    113 		return;
    114 	}
    115 
    116 	/* Initialise and attach the disk structure. */
    117 	disk_init(&sc->sc_dk, sc->sc_dv.dv_xname, &lddkdriver);
    118 	disk_attach(&sc->sc_dk);
    119 
    120 	if (sc->sc_maxxfer > MAXPHYS)
    121 		sc->sc_maxxfer = MAXPHYS;
    122 
    123 	/* Build synthetic geometry if necessary. */
    124 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    125 	    sc->sc_ncylinders == 0) {
    126 		uint64_t ncyl;
    127 
    128 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    129 			sc->sc_nheads = 16;
    130 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    131 			sc->sc_nheads = 32;
    132 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    133 			sc->sc_nheads = 64;
    134 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    135 			sc->sc_nheads = 128;
    136 		else
    137 			sc->sc_nheads = 255;
    138 
    139 		sc->sc_nsectors = 63;
    140 		sc->sc_ncylinders = INT_MAX;
    141 		ncyl = sc->sc_secperunit /
    142 		    (sc->sc_nheads * sc->sc_nsectors);
    143 		if (ncyl < INT_MAX)
    144 			sc->sc_ncylinders = (int)ncyl;
    145 	}
    146 
    147 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    148 	    sc->sc_secsize);
    149 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    150 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
    151 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    152 
    153 	ld_set_properties(sc);
    154 
    155 #if NRND > 0
    156 	/* Attach the device into the rnd source list. */
    157 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    158 	    RND_TYPE_DISK, 0);
    159 #endif
    160 
    161 	/* Register with PMF */
    162 	if (!pmf_device_register1(&sc->sc_dv, NULL, NULL, ld_shutdown))
    163 		aprint_error_dev(&sc->sc_dv,
    164 		    "couldn't establish power handler\n");
    165 
    166 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    167 
    168 	/* Discover wedges on this disk. */
    169 	config_interrupts(&sc->sc_dv, ld_config_interrupts);
    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 	disk_destroy(&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 bool
    268 ld_shutdown(device_t dev, int flags)
    269 {
    270 	struct ld_softc *sc = device_private(dev);
    271 
    272 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0) {
    273 		printf("%s: unable to flush cache\n", device_xname(dev));
    274 		return false;
    275 	}
    276 
    277 	return true;
    278 }
    279 
    280 /* ARGSUSED */
    281 static int
    282 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    283 {
    284 	struct ld_softc *sc;
    285 	int error, unit, part;
    286 
    287 	unit = DISKUNIT(dev);
    288 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    289 		return (ENXIO);
    290 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    291 		return (ENODEV);
    292 	part = DISKPART(dev);
    293 
    294 	mutex_enter(&sc->sc_dk.dk_openlock);
    295 
    296 	if (sc->sc_dk.dk_openmask == 0) {
    297 		/* Load the partition info if not already loaded. */
    298 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    299 			ldgetdisklabel(sc);
    300 	}
    301 
    302 	/* Check that the partition exists. */
    303 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    304 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    305 		error = ENXIO;
    306 		goto bad1;
    307 	}
    308 
    309 	/* Ensure only one open at a time. */
    310 	switch (fmt) {
    311 	case S_IFCHR:
    312 		sc->sc_dk.dk_copenmask |= (1 << part);
    313 		break;
    314 	case S_IFBLK:
    315 		sc->sc_dk.dk_bopenmask |= (1 << part);
    316 		break;
    317 	}
    318 	sc->sc_dk.dk_openmask =
    319 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    320 
    321 	error = 0;
    322  bad1:
    323 	mutex_exit(&sc->sc_dk.dk_openlock);
    324 	return (error);
    325 }
    326 
    327 /* ARGSUSED */
    328 static int
    329 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    330 {
    331 	struct ld_softc *sc;
    332 	int part, unit;
    333 
    334 	unit = DISKUNIT(dev);
    335 	part = DISKPART(dev);
    336 	sc = device_lookup(&ld_cd, unit);
    337 
    338 	mutex_enter(&sc->sc_dk.dk_openlock);
    339 
    340 	switch (fmt) {
    341 	case S_IFCHR:
    342 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    343 		break;
    344 	case S_IFBLK:
    345 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    346 		break;
    347 	}
    348 	sc->sc_dk.dk_openmask =
    349 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    350 
    351 	if (sc->sc_dk.dk_openmask == 0) {
    352 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    353 			printf("%s: unable to flush cache\n",
    354 			    sc->sc_dv.dv_xname);
    355 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    356 			sc->sc_flags &= ~LDF_VLABEL;
    357 	}
    358 
    359 	mutex_exit(&sc->sc_dk.dk_openlock);
    360 	return (0);
    361 }
    362 
    363 /* ARGSUSED */
    364 static int
    365 ldread(dev_t dev, struct uio *uio, int ioflag)
    366 {
    367 
    368 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    369 }
    370 
    371 /* ARGSUSED */
    372 static int
    373 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    374 {
    375 
    376 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    377 }
    378 
    379 /* ARGSUSED */
    380 static int
    381 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
    382 {
    383 	struct ld_softc *sc;
    384 	int part, unit, error;
    385 #ifdef __HAVE_OLD_DISKLABEL
    386 	struct disklabel newlabel;
    387 #endif
    388 	struct disklabel *lp;
    389 
    390 	unit = DISKUNIT(dev);
    391 	part = DISKPART(dev);
    392 	sc = device_lookup(&ld_cd, unit);
    393 
    394 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
    395 	if (error != EPASSTHROUGH)
    396 		return (error);
    397 
    398 	error = 0;
    399 	switch (cmd) {
    400 	case DIOCGDINFO:
    401 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    402 		return (0);
    403 
    404 #ifdef __HAVE_OLD_DISKLABEL
    405 	case ODIOCGDINFO:
    406 		newlabel = *(sc->sc_dk.dk_label);
    407 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    408 			return ENOTTY;
    409 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    410 		return (0);
    411 #endif
    412 
    413 	case DIOCGPART:
    414 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    415 		((struct partinfo *)addr)->part =
    416 		    &sc->sc_dk.dk_label->d_partitions[part];
    417 		break;
    418 
    419 	case DIOCWDINFO:
    420 	case DIOCSDINFO:
    421 #ifdef __HAVE_OLD_DISKLABEL
    422 	case ODIOCWDINFO:
    423 	case ODIOCSDINFO:
    424 
    425 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    426 			memset(&newlabel, 0, sizeof newlabel);
    427 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    428 			lp = &newlabel;
    429 		} else
    430 #endif
    431 		lp = (struct disklabel *)addr;
    432 
    433 		if ((flag & FWRITE) == 0)
    434 			return (EBADF);
    435 
    436 		mutex_enter(&sc->sc_dk.dk_openlock);
    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 		mutex_exit(&sc->sc_dk.dk_openlock);
    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, l));
    529 	    }
    530 	case DIOCGSTRATEGY:
    531 	    {
    532 		struct disk_strategy *dks = (void *)addr;
    533 
    534 		mutex_enter(&sc->sc_mutex);
    535 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
    536 		    sizeof(dks->dks_name));
    537 		mutex_exit(&sc->sc_mutex);
    538 		dks->dks_paramlen = 0;
    539 
    540 		return 0;
    541 	    }
    542 	case DIOCSSTRATEGY:
    543 	    {
    544 		struct disk_strategy *dks = (void *)addr;
    545 		struct bufq_state *new, *old;
    546 
    547 		if ((flag & FWRITE) == 0)
    548 			return EPERM;
    549 
    550 		if (dks->dks_param != NULL)
    551 			return EINVAL;
    552 
    553 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
    554 		error = bufq_alloc(&new, dks->dks_name,
    555 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
    556 		if (error)
    557 			return error;
    558 
    559 		mutex_enter(&sc->sc_mutex);
    560 		old = sc->sc_bufq;
    561 		bufq_move(new, old);
    562 		sc->sc_bufq = new;
    563 		mutex_exit(&sc->sc_mutex);
    564 		bufq_free(old);
    565 
    566 		return 0;
    567 	    }
    568 	default:
    569 		error = ENOTTY;
    570 		break;
    571 	}
    572 
    573 	return (error);
    574 }
    575 
    576 static void
    577 ldstrategy(struct buf *bp)
    578 {
    579 	struct ld_softc *sc;
    580 	struct disklabel *lp;
    581 	daddr_t blkno;
    582 	int s, part;
    583 
    584 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    585 	part = DISKPART(bp->b_dev);
    586 
    587 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    588 		bp->b_error = EIO;
    589 		goto done;
    590 	}
    591 
    592 	lp = sc->sc_dk.dk_label;
    593 
    594 	/*
    595 	 * The transfer must be a whole number of blocks and the offset must
    596 	 * not be negative.
    597 	 */
    598 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    599 		bp->b_error = EINVAL;
    600 		goto done;
    601 	}
    602 
    603 	/* If it's a null transfer, return immediately. */
    604 	if (bp->b_bcount == 0)
    605 		goto done;
    606 
    607 	/*
    608 	 * Do bounds checking and adjust the transfer.  If error, process.
    609 	 * If past the end of partition, just return.
    610 	 */
    611 	if (part != RAW_PART &&
    612 	    bounds_check_with_label(&sc->sc_dk, bp,
    613 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    614 		goto done;
    615 	}
    616 
    617 	/*
    618 	 * Convert the block number to absolute and put it in terms
    619 	 * of the device's logical block size.
    620 	 */
    621 	if (lp->d_secsize == DEV_BSIZE)
    622 		blkno = bp->b_blkno;
    623 	else if (lp->d_secsize > DEV_BSIZE)
    624 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    625 	else
    626 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    627 
    628 	if (part != RAW_PART)
    629 		blkno += lp->d_partitions[part].p_offset;
    630 
    631 	bp->b_rawblkno = blkno;
    632 
    633 	s = splbio();
    634 	ldstart(sc, bp);
    635 	splx(s);
    636 	return;
    637 
    638  done:
    639 	bp->b_resid = bp->b_bcount;
    640 	biodone(bp);
    641 }
    642 
    643 static void
    644 ldstart(struct ld_softc *sc, struct buf *bp)
    645 {
    646 	int error;
    647 
    648 	mutex_enter(&sc->sc_mutex);
    649 
    650 	if (bp != NULL)
    651 		BUFQ_PUT(sc->sc_bufq, bp);
    652 
    653 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    654 		/* See if there is work to do. */
    655 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    656 			break;
    657 
    658 		disk_busy(&sc->sc_dk);
    659 		sc->sc_queuecnt++;
    660 
    661 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    662 			/*
    663 			 * The back-end is running the job; remove it from
    664 			 * the queue.
    665 			 */
    666 			(void) BUFQ_GET(sc->sc_bufq);
    667 		} else  {
    668 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    669 			sc->sc_queuecnt--;
    670 			if (error == EAGAIN) {
    671 				/*
    672 				 * Temporary resource shortage in the
    673 				 * back-end; just defer the job until
    674 				 * later.
    675 				 *
    676 				 * XXX We might consider a watchdog timer
    677 				 * XXX to make sure we are kicked into action.
    678 				 */
    679 				break;
    680 			} else {
    681 				(void) BUFQ_GET(sc->sc_bufq);
    682 				bp->b_error = error;
    683 				bp->b_resid = bp->b_bcount;
    684 				mutex_exit(&sc->sc_mutex);
    685 				biodone(bp);
    686 				mutex_enter(&sc->sc_mutex);
    687 			}
    688 		}
    689 	}
    690 
    691 	mutex_exit(&sc->sc_mutex);
    692 }
    693 
    694 void
    695 lddone(struct ld_softc *sc, struct buf *bp)
    696 {
    697 
    698 	if (bp->b_error != 0) {
    699 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    700 		printf("\n");
    701 	}
    702 
    703 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    704 	    (bp->b_flags & B_READ));
    705 #if NRND > 0
    706 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    707 #endif
    708 	biodone(bp);
    709 
    710 	mutex_enter(&sc->sc_mutex);
    711 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    712 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    713 			sc->sc_flags &= ~LDF_DRAIN;
    714 			wakeup(&sc->sc_queuecnt);
    715 		}
    716 		mutex_exit(&sc->sc_mutex);
    717 		ldstart(sc, NULL);
    718 	} else
    719 		mutex_exit(&sc->sc_mutex);
    720 }
    721 
    722 static int
    723 ldsize(dev_t dev)
    724 {
    725 	struct ld_softc *sc;
    726 	int part, unit, omask, size;
    727 
    728 	unit = DISKUNIT(dev);
    729 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    730 		return (ENODEV);
    731 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    732 		return (ENODEV);
    733 	part = DISKPART(dev);
    734 
    735 	omask = sc->sc_dk.dk_openmask & (1 << part);
    736 
    737 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    738 		return (-1);
    739 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    740 		size = -1;
    741 	else
    742 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    743 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    744 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    745 		return (-1);
    746 
    747 	return (size);
    748 }
    749 
    750 /*
    751  * Load the label information from the specified device.
    752  */
    753 static void
    754 ldgetdisklabel(struct ld_softc *sc)
    755 {
    756 	const char *errstring;
    757 
    758 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    759 
    760 	/* Call the generic disklabel extraction routine. */
    761 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
    762 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    763 	if (errstring != NULL)
    764 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    765 
    766 	/* In-core label now valid. */
    767 	sc->sc_flags |= LDF_VLABEL;
    768 }
    769 
    770 /*
    771  * Construct a ficticious label.
    772  */
    773 static void
    774 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    775 {
    776 
    777 	memset(lp, 0, sizeof(struct disklabel));
    778 
    779 	lp->d_secsize = sc->sc_secsize;
    780 	lp->d_ntracks = sc->sc_nheads;
    781 	lp->d_nsectors = sc->sc_nsectors;
    782 	lp->d_ncylinders = sc->sc_ncylinders;
    783 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    784 	lp->d_type = DTYPE_LD;
    785 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    786 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    787 	lp->d_secperunit = sc->sc_secperunit;
    788 	lp->d_rpm = 7200;
    789 	lp->d_interleave = 1;
    790 	lp->d_flags = 0;
    791 
    792 	lp->d_partitions[RAW_PART].p_offset = 0;
    793 	lp->d_partitions[RAW_PART].p_size =
    794 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    795 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    796 	lp->d_npartitions = RAW_PART + 1;
    797 
    798 	lp->d_magic = DISKMAGIC;
    799 	lp->d_magic2 = DISKMAGIC;
    800 	lp->d_checksum = dkcksum(lp);
    801 }
    802 
    803 /*
    804  * Take a dump.
    805  */
    806 static int
    807 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
    808 {
    809 	char *va = vav;
    810 	struct ld_softc *sc;
    811 	struct disklabel *lp;
    812 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    813 	static int dumping;
    814 
    815 	unit = DISKUNIT(dev);
    816 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    817 		return (ENXIO);
    818 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    819 		return (ENODEV);
    820 	if (sc->sc_dump == NULL)
    821 		return (ENXIO);
    822 
    823 	/* Check if recursive dump; if so, punt. */
    824 	if (dumping)
    825 		return (EFAULT);
    826 	dumping = 1;
    827 
    828 	/* Convert to disk sectors.  Request must be a multiple of size. */
    829 	part = DISKPART(dev);
    830 	lp = sc->sc_dk.dk_label;
    831 	if ((size % lp->d_secsize) != 0)
    832 		return (EFAULT);
    833 	towrt = size / lp->d_secsize;
    834 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    835 
    836 	nsects = lp->d_partitions[part].p_size;
    837 	sectoff = lp->d_partitions[part].p_offset;
    838 
    839 	/* Check transfer bounds against partition size. */
    840 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    841 		return (EINVAL);
    842 
    843 	/* Offset block number to start of partition. */
    844 	blkno += sectoff;
    845 
    846 	/* Start dumping and return when done. */
    847 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    848 	while (towrt > 0) {
    849 		nblk = min(maxblkcnt, towrt);
    850 
    851 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    852 			return (rv);
    853 
    854 		towrt -= nblk;
    855 		blkno += nblk;
    856 		va += nblk * sc->sc_secsize;
    857 	}
    858 
    859 	dumping = 0;
    860 	return (0);
    861 }
    862 
    863 /*
    864  * Adjust the size of a transfer.
    865  */
    866 static void
    867 ldminphys(struct buf *bp)
    868 {
    869 	struct ld_softc *sc;
    870 
    871 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    872 
    873 	if (bp->b_bcount > sc->sc_maxxfer)
    874 		bp->b_bcount = sc->sc_maxxfer;
    875 	minphys(bp);
    876 }
    877 
    878 static void
    879 ld_set_properties(struct ld_softc *ld)
    880 {
    881 	prop_dictionary_t disk_info, odisk_info, geom;
    882 
    883 	disk_info = prop_dictionary_create();
    884 
    885 	geom = prop_dictionary_create();
    886 
    887 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
    888 	    ld->sc_secperunit);
    889 
    890 	prop_dictionary_set_uint32(geom, "sector-size",
    891 	    ld->sc_secsize);
    892 
    893 	prop_dictionary_set_uint16(geom, "sectors-per-track",
    894 	    ld->sc_nsectors);
    895 
    896 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
    897 	    ld->sc_nheads);
    898 
    899 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
    900 	    ld->sc_ncylinders);
    901 
    902 	prop_dictionary_set(disk_info, "geometry", geom);
    903 	prop_object_release(geom);
    904 
    905 	prop_dictionary_set(device_properties(&ld->sc_dv),
    906 	    "disk-info", disk_info);
    907 
    908 	/*
    909 	 * Don't release disk_info here; we keep a reference to it.
    910 	 * disk_detach() will release it when we go away.
    911 	 */
    912 
    913 	odisk_info = ld->sc_dk.dk_info;
    914 	ld->sc_dk.dk_info = disk_info;
    915 	if (odisk_info)
    916 		prop_object_release(odisk_info);
    917 }
    918 
    919 static void
    920 ld_config_interrupts (struct device *d)
    921 {
    922 	struct ld_softc *sc = (struct ld_softc *)d;
    923 	dkwedge_discover(&sc->sc_dk);
    924 }
    925