Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.54.6.6
      1 /*	$NetBSD: ld.c,v 1.54.6.6 2008/09/28 10:40:18 mjf 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 1.54.6.6 2008/09/28 10:40:18 mjf Exp $");
     38 
     39 #include "rnd.h"
     40 
     41 #include <sys/param.h>
     42 #include <sys/systm.h>
     43 #include <sys/kernel.h>
     44 #include <sys/device.h>
     45 #include <sys/queue.h>
     46 #include <sys/proc.h>
     47 #include <sys/buf.h>
     48 #include <sys/bufq.h>
     49 #include <sys/endian.h>
     50 #include <sys/disklabel.h>
     51 #include <sys/disk.h>
     52 #include <sys/dkio.h>
     53 #include <sys/stat.h>
     54 #include <sys/conf.h>
     55 #include <sys/fcntl.h>
     56 #include <sys/vnode.h>
     57 #include <sys/syslog.h>
     58 #include <sys/mutex.h>
     59 #if NRND > 0
     60 #include <sys/rnd.h>
     61 #endif
     62 
     63 #include <dev/ldvar.h>
     64 
     65 #include <prop/proplib.h>
     66 
     67 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
     68 static void	ldgetdisklabel(struct ld_softc *);
     69 static void	ldminphys(struct buf *bp);
     70 static bool	ld_shutdown(device_t, int);
     71 static void	ldstart(struct ld_softc *, struct buf *);
     72 static void	ld_set_properties(struct ld_softc *);
     73 static void	ld_config_interrupts (struct device *);
     74 
     75 extern struct	cfdriver ld_cd;
     76 
     77 static dev_type_open(ldopen);
     78 static dev_type_close(ldclose);
     79 static dev_type_read(ldread);
     80 static dev_type_write(ldwrite);
     81 static dev_type_ioctl(ldioctl);
     82 static dev_type_strategy(ldstrategy);
     83 static dev_type_dump(lddump);
     84 static dev_type_size(ldsize);
     85 
     86 const struct bdevsw ld_bdevsw = {
     87 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
     88 };
     89 
     90 const struct cdevsw ld_cdevsw = {
     91 	ldopen, ldclose, ldread, ldwrite, ldioctl,
     92 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
     93 };
     94 
     95 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
     96 
     97 void
     98 ldattach(struct ld_softc *sc)
     99 {
    100 	char tbuf[9];
    101 	int i, unit, cmaj, bmaj;
    102 
    103 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
    104 
    105 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    106 		aprint_normal_dev(sc->sc_dv, "disabled\n");
    107 		return;
    108 	}
    109 
    110 	/* Initialise and attach the disk structure. */
    111 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
    112 	disk_attach(&sc->sc_dk);
    113 
    114 	if (sc->sc_maxxfer > MAXPHYS)
    115 		sc->sc_maxxfer = MAXPHYS;
    116 
    117 	/* Build synthetic geometry if necessary. */
    118 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    119 	    sc->sc_ncylinders == 0) {
    120 		uint64_t ncyl;
    121 
    122 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    123 			sc->sc_nheads = 16;
    124 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    125 			sc->sc_nheads = 32;
    126 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    127 			sc->sc_nheads = 64;
    128 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    129 			sc->sc_nheads = 128;
    130 		else
    131 			sc->sc_nheads = 255;
    132 
    133 		sc->sc_nsectors = 63;
    134 		sc->sc_ncylinders = INT_MAX;
    135 		ncyl = sc->sc_secperunit /
    136 		    (sc->sc_nheads * sc->sc_nsectors);
    137 		if (ncyl < INT_MAX)
    138 			sc->sc_ncylinders = (int)ncyl;
    139 	}
    140 
    141 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    142 	    sc->sc_secsize);
    143 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
    144 	    "%d bytes/sect x %"PRIu64" sectors\n",
    145 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
    146 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    147 
    148 	ld_set_properties(sc);
    149 
    150 #if NRND > 0
    151 	/* Attach the device into the rnd source list. */
    152 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
    153 	    RND_TYPE_DISK, 0);
    154 #endif
    155 
    156 	/* Register with PMF */
    157 	if (!pmf_device_register1(sc->sc_dv, NULL, NULL, ld_shutdown))
    158 		aprint_error_dev(sc->sc_dv,
    159 		    "couldn't establish power handler\n");
    160 
    161 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    162 
    163 	/* Discover wedges on this disk. */
    164 	config_interrupts(&sc->sc_dv, ld_config_interrupts);
    165 
    166 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    167 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    168 	unit = device_unit(&sc->sc_dv);
    169 
    170 	for (i = 0; i < MAXPARTITIONS; i++) {
    171 		device_register_name(MAKEDISKDEV(cmaj, unit, i), &sc->sc_dv,
    172 		    true, DEV_DISK, "rld%d%c", unit, 'a' + i);
    173 		device_register_name(MAKEDISKDEV(bmaj, unit, i), &sc->sc_dv,
    174 		    false, DEV_DISK, "ld%d%c", unit, 'a' + i);
    175 	}
    176 
    177 }
    178 
    179 int
    180 ldadjqparam(struct ld_softc *sc, int xmax)
    181 {
    182 	int s;
    183 
    184 	s = splbio();
    185 	sc->sc_maxqueuecnt = xmax;
    186 	splx(s);
    187 
    188 	return (0);
    189 }
    190 
    191 int
    192 ldbegindetach(struct ld_softc *sc, int flags)
    193 {
    194 	int s, rv = 0;
    195 
    196 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    197 		return (0);
    198 
    199 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    200 		return (EBUSY);
    201 
    202 	s = splbio();
    203 	sc->sc_maxqueuecnt = 0;
    204 	sc->sc_flags |= LDF_DETACH;
    205 	while (sc->sc_queuecnt > 0) {
    206 		sc->sc_flags |= LDF_DRAIN;
    207 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    208 		if (rv)
    209 			break;
    210 	}
    211 	splx(s);
    212 
    213 	return (rv);
    214 }
    215 
    216 void
    217 ldenddetach(struct ld_softc *sc)
    218 {
    219 	int s, bmaj, cmaj, i, mn;
    220 
    221 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    222 		return;
    223 
    224 	/* Wait for commands queued with the hardware to complete. */
    225 	if (sc->sc_queuecnt != 0)
    226 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    227 			printf("%s: not drained\n", device_xname(sc->sc_dv));
    228 
    229 	device_deregister_all(&sc->sc_dv);
    230 
    231 	/* Locate the major numbers. */
    232 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    233 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    234 
    235 	/* Kill off any queued buffers. */
    236 	s = splbio();
    237 	bufq_drain(sc->sc_bufq);
    238 	splx(s);
    239 
    240 	bufq_free(sc->sc_bufq);
    241 
    242 	/* Nuke the vnodes for any open instances. */
    243 	for (i = 0; i < MAXPARTITIONS; i++) {
    244 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
    245 		vdevgone(bmaj, mn, mn, VBLK);
    246 		vdevgone(cmaj, mn, mn, VCHR);
    247 	}
    248 
    249 	/* Delete all of our wedges. */
    250 	dkwedge_delall(&sc->sc_dk);
    251 
    252 	/* Detach from the disk list. */
    253 	disk_detach(&sc->sc_dk);
    254 	disk_destroy(&sc->sc_dk);
    255 
    256 #if NRND > 0
    257 	/* Unhook the entropy source. */
    258 	rnd_detach_source(&sc->sc_rnd_source);
    259 #endif
    260 
    261 	/* Deregister with PMF */
    262 	pmf_device_deregister(sc->sc_dv);
    263 
    264 	/*
    265 	 * XXX We can't really flush the cache here, beceause the
    266 	 * XXX device may already be non-existent from the controller's
    267 	 * XXX perspective.
    268 	 */
    269 #if 0
    270 	/* Flush the device's cache. */
    271 	if (sc->sc_flush != NULL)
    272 		if ((*sc->sc_flush)(sc, 0) != 0)
    273 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
    274 #endif
    275 	mutex_destroy(&sc->sc_mutex);
    276 }
    277 
    278 /* ARGSUSED */
    279 static bool
    280 ld_shutdown(device_t dev, int flags)
    281 {
    282 	struct ld_softc *sc = device_private(dev);
    283 
    284 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
    285 		printf("%s: unable to flush cache\n", device_xname(dev));
    286 		return false;
    287 	}
    288 
    289 	return true;
    290 }
    291 
    292 /* ARGSUSED */
    293 static int
    294 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    295 {
    296 	struct ld_softc *sc;
    297 	int error, unit, part;
    298 
    299 	unit = DISKUNIT(dev);
    300 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    301 		return (ENXIO);
    302 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    303 		return (ENODEV);
    304 	part = DISKPART(dev);
    305 
    306 	mutex_enter(&sc->sc_dk.dk_openlock);
    307 
    308 	if (sc->sc_dk.dk_openmask == 0) {
    309 		/* Load the partition info if not already loaded. */
    310 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    311 			ldgetdisklabel(sc);
    312 	}
    313 
    314 	/* Check that the partition exists. */
    315 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    316 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    317 		error = ENXIO;
    318 		goto bad1;
    319 	}
    320 
    321 	/* Ensure only one open at a time. */
    322 	switch (fmt) {
    323 	case S_IFCHR:
    324 		sc->sc_dk.dk_copenmask |= (1 << part);
    325 		break;
    326 	case S_IFBLK:
    327 		sc->sc_dk.dk_bopenmask |= (1 << part);
    328 		break;
    329 	}
    330 	sc->sc_dk.dk_openmask =
    331 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    332 
    333 	error = 0;
    334  bad1:
    335 	mutex_exit(&sc->sc_dk.dk_openlock);
    336 	return (error);
    337 }
    338 
    339 /* ARGSUSED */
    340 static int
    341 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    342 {
    343 	struct ld_softc *sc;
    344 	int part, unit;
    345 
    346 	unit = DISKUNIT(dev);
    347 	part = DISKPART(dev);
    348 	sc = device_lookup_private(&ld_cd, unit);
    349 
    350 	mutex_enter(&sc->sc_dk.dk_openlock);
    351 
    352 	switch (fmt) {
    353 	case S_IFCHR:
    354 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    355 		break;
    356 	case S_IFBLK:
    357 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    358 		break;
    359 	}
    360 	sc->sc_dk.dk_openmask =
    361 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    362 
    363 	if (sc->sc_dk.dk_openmask == 0) {
    364 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
    365 			aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
    366 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    367 			sc->sc_flags &= ~LDF_VLABEL;
    368 	}
    369 
    370 	mutex_exit(&sc->sc_dk.dk_openlock);
    371 	return (0);
    372 }
    373 
    374 /* ARGSUSED */
    375 static int
    376 ldread(dev_t dev, struct uio *uio, int ioflag)
    377 {
    378 
    379 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    380 }
    381 
    382 /* ARGSUSED */
    383 static int
    384 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    385 {
    386 
    387 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    388 }
    389 
    390 /* ARGSUSED */
    391 static int
    392 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
    393 {
    394 	struct ld_softc *sc;
    395 	int part, unit, error;
    396 #ifdef __HAVE_OLD_DISKLABEL
    397 	struct disklabel newlabel;
    398 #endif
    399 	struct disklabel *lp;
    400 
    401 	unit = DISKUNIT(dev);
    402 	part = DISKPART(dev);
    403 	sc = device_lookup_private(&ld_cd, unit);
    404 
    405 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
    406 	if (error != EPASSTHROUGH)
    407 		return (error);
    408 
    409 	error = 0;
    410 	switch (cmd) {
    411 	case DIOCGDINFO:
    412 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    413 		return (0);
    414 
    415 #ifdef __HAVE_OLD_DISKLABEL
    416 	case ODIOCGDINFO:
    417 		newlabel = *(sc->sc_dk.dk_label);
    418 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    419 			return ENOTTY;
    420 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    421 		return (0);
    422 #endif
    423 
    424 	case DIOCGPART:
    425 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    426 		((struct partinfo *)addr)->part =
    427 		    &sc->sc_dk.dk_label->d_partitions[part];
    428 		break;
    429 
    430 	case DIOCWDINFO:
    431 	case DIOCSDINFO:
    432 #ifdef __HAVE_OLD_DISKLABEL
    433 	case ODIOCWDINFO:
    434 	case ODIOCSDINFO:
    435 
    436 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    437 			memset(&newlabel, 0, sizeof newlabel);
    438 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    439 			lp = &newlabel;
    440 		} else
    441 #endif
    442 		lp = (struct disklabel *)addr;
    443 
    444 		if ((flag & FWRITE) == 0)
    445 			return (EBADF);
    446 
    447 		mutex_enter(&sc->sc_dk.dk_openlock);
    448 		sc->sc_flags |= LDF_LABELLING;
    449 
    450 		error = setdisklabel(sc->sc_dk.dk_label,
    451 		    lp, /*sc->sc_dk.dk_openmask : */0,
    452 		    sc->sc_dk.dk_cpulabel);
    453 		if (error == 0 && (cmd == DIOCWDINFO
    454 #ifdef __HAVE_OLD_DISKLABEL
    455 		    || cmd == ODIOCWDINFO
    456 #endif
    457 		    ))
    458 			error = writedisklabel(
    459 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    460 			    ldstrategy, sc->sc_dk.dk_label,
    461 			    sc->sc_dk.dk_cpulabel);
    462 
    463 		sc->sc_flags &= ~LDF_LABELLING;
    464 		mutex_exit(&sc->sc_dk.dk_openlock);
    465 		break;
    466 
    467 	case DIOCKLABEL:
    468 		if ((flag & FWRITE) == 0)
    469 			return (EBADF);
    470 		if (*(int *)addr)
    471 			sc->sc_flags |= LDF_KLABEL;
    472 		else
    473 			sc->sc_flags &= ~LDF_KLABEL;
    474 		break;
    475 
    476 	case DIOCWLABEL:
    477 		if ((flag & FWRITE) == 0)
    478 			return (EBADF);
    479 		if (*(int *)addr)
    480 			sc->sc_flags |= LDF_WLABEL;
    481 		else
    482 			sc->sc_flags &= ~LDF_WLABEL;
    483 		break;
    484 
    485 	case DIOCGDEFLABEL:
    486 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    487 		break;
    488 
    489 #ifdef __HAVE_OLD_DISKLABEL
    490 	case ODIOCGDEFLABEL:
    491 		ldgetdefaultlabel(sc, &newlabel);
    492 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    493 			return ENOTTY;
    494 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    495 		break;
    496 #endif
    497 
    498 	case DIOCCACHESYNC:
    499 		/*
    500 		 * XXX Do we really need to care about having a writable
    501 		 * file descriptor here?
    502 		 */
    503 		if ((flag & FWRITE) == 0)
    504 			error = EBADF;
    505 		else if (sc->sc_flush)
    506 			error = (*sc->sc_flush)(sc, 0);
    507 		else
    508 			error = 0;	/* XXX Error out instead? */
    509 		break;
    510 
    511 	case DIOCAWEDGE:
    512 	    {
    513 	    	struct dkwedge_info *dkw = (void *) addr;
    514 
    515 		if ((flag & FWRITE) == 0)
    516 			return (EBADF);
    517 
    518 		/* If the ioctl happens here, the parent is us. */
    519 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
    520 			sizeof(dkw->dkw_parent));
    521 		return (dkwedge_add(dkw));
    522 	    }
    523 
    524 	case DIOCDWEDGE:
    525 	    {
    526 	    	struct dkwedge_info *dkw = (void *) addr;
    527 
    528 		if ((flag & FWRITE) == 0)
    529 			return (EBADF);
    530 
    531 		/* If the ioctl happens here, the parent is us. */
    532 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
    533 			sizeof(dkw->dkw_parent));
    534 		return (dkwedge_del(dkw));
    535 	    }
    536 
    537 	case DIOCLWEDGES:
    538 	    {
    539 	    	struct dkwedge_list *dkwl = (void *) addr;
    540 
    541 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
    542 	    }
    543 	case DIOCGSTRATEGY:
    544 	    {
    545 		struct disk_strategy *dks = (void *)addr;
    546 
    547 		mutex_enter(&sc->sc_mutex);
    548 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
    549 		    sizeof(dks->dks_name));
    550 		mutex_exit(&sc->sc_mutex);
    551 		dks->dks_paramlen = 0;
    552 
    553 		return 0;
    554 	    }
    555 	case DIOCSSTRATEGY:
    556 	    {
    557 		struct disk_strategy *dks = (void *)addr;
    558 		struct bufq_state *new, *old;
    559 
    560 		if ((flag & FWRITE) == 0)
    561 			return EPERM;
    562 
    563 		if (dks->dks_param != NULL)
    564 			return EINVAL;
    565 
    566 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
    567 		error = bufq_alloc(&new, dks->dks_name,
    568 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
    569 		if (error)
    570 			return error;
    571 
    572 		mutex_enter(&sc->sc_mutex);
    573 		old = sc->sc_bufq;
    574 		bufq_move(new, old);
    575 		sc->sc_bufq = new;
    576 		mutex_exit(&sc->sc_mutex);
    577 		bufq_free(old);
    578 
    579 		return 0;
    580 	    }
    581 	default:
    582 		error = ENOTTY;
    583 		break;
    584 	}
    585 
    586 	return (error);
    587 }
    588 
    589 static void
    590 ldstrategy(struct buf *bp)
    591 {
    592 	struct ld_softc *sc;
    593 	struct disklabel *lp;
    594 	daddr_t blkno;
    595 	int s, part;
    596 
    597 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
    598 	part = DISKPART(bp->b_dev);
    599 
    600 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    601 		bp->b_error = EIO;
    602 		goto done;
    603 	}
    604 
    605 	lp = sc->sc_dk.dk_label;
    606 
    607 	/*
    608 	 * The transfer must be a whole number of blocks and the offset must
    609 	 * not be negative.
    610 	 */
    611 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    612 		bp->b_error = EINVAL;
    613 		goto done;
    614 	}
    615 
    616 	/* If it's a null transfer, return immediately. */
    617 	if (bp->b_bcount == 0)
    618 		goto done;
    619 
    620 	/*
    621 	 * Do bounds checking and adjust the transfer.  If error, process.
    622 	 * If past the end of partition, just return.
    623 	 */
    624 	if (part != RAW_PART &&
    625 	    bounds_check_with_label(&sc->sc_dk, bp,
    626 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    627 		goto done;
    628 	}
    629 
    630 	/*
    631 	 * Convert the block number to absolute and put it in terms
    632 	 * of the device's logical block size.
    633 	 */
    634 	if (lp->d_secsize == DEV_BSIZE)
    635 		blkno = bp->b_blkno;
    636 	else if (lp->d_secsize > DEV_BSIZE)
    637 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    638 	else
    639 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    640 
    641 	if (part != RAW_PART)
    642 		blkno += lp->d_partitions[part].p_offset;
    643 
    644 	bp->b_rawblkno = blkno;
    645 
    646 	s = splbio();
    647 	ldstart(sc, bp);
    648 	splx(s);
    649 	return;
    650 
    651  done:
    652 	bp->b_resid = bp->b_bcount;
    653 	biodone(bp);
    654 }
    655 
    656 static void
    657 ldstart(struct ld_softc *sc, struct buf *bp)
    658 {
    659 	int error;
    660 
    661 	mutex_enter(&sc->sc_mutex);
    662 
    663 	if (bp != NULL)
    664 		BUFQ_PUT(sc->sc_bufq, bp);
    665 
    666 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    667 		/* See if there is work to do. */
    668 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    669 			break;
    670 
    671 		disk_busy(&sc->sc_dk);
    672 		sc->sc_queuecnt++;
    673 
    674 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    675 			/*
    676 			 * The back-end is running the job; remove it from
    677 			 * the queue.
    678 			 */
    679 			(void) BUFQ_GET(sc->sc_bufq);
    680 		} else  {
    681 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    682 			sc->sc_queuecnt--;
    683 			if (error == EAGAIN) {
    684 				/*
    685 				 * Temporary resource shortage in the
    686 				 * back-end; just defer the job until
    687 				 * later.
    688 				 *
    689 				 * XXX We might consider a watchdog timer
    690 				 * XXX to make sure we are kicked into action.
    691 				 */
    692 				break;
    693 			} else {
    694 				(void) BUFQ_GET(sc->sc_bufq);
    695 				bp->b_error = error;
    696 				bp->b_resid = bp->b_bcount;
    697 				mutex_exit(&sc->sc_mutex);
    698 				biodone(bp);
    699 				mutex_enter(&sc->sc_mutex);
    700 			}
    701 		}
    702 	}
    703 
    704 	mutex_exit(&sc->sc_mutex);
    705 }
    706 
    707 void
    708 lddone(struct ld_softc *sc, struct buf *bp)
    709 {
    710 
    711 	if (bp->b_error != 0) {
    712 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    713 		printf("\n");
    714 	}
    715 
    716 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    717 	    (bp->b_flags & B_READ));
    718 #if NRND > 0
    719 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    720 #endif
    721 	biodone(bp);
    722 
    723 	mutex_enter(&sc->sc_mutex);
    724 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    725 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    726 			sc->sc_flags &= ~LDF_DRAIN;
    727 			wakeup(&sc->sc_queuecnt);
    728 		}
    729 		mutex_exit(&sc->sc_mutex);
    730 		ldstart(sc, NULL);
    731 	} else
    732 		mutex_exit(&sc->sc_mutex);
    733 }
    734 
    735 static int
    736 ldsize(dev_t dev)
    737 {
    738 	struct ld_softc *sc;
    739 	int part, unit, omask, size;
    740 
    741 	unit = DISKUNIT(dev);
    742 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    743 		return (ENODEV);
    744 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    745 		return (ENODEV);
    746 	part = DISKPART(dev);
    747 
    748 	omask = sc->sc_dk.dk_openmask & (1 << part);
    749 
    750 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    751 		return (-1);
    752 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    753 		size = -1;
    754 	else
    755 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    756 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    757 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    758 		return (-1);
    759 
    760 	return (size);
    761 }
    762 
    763 /*
    764  * Load the label information from the specified device.
    765  */
    766 static void
    767 ldgetdisklabel(struct ld_softc *sc)
    768 {
    769 	const char *errstring;
    770 
    771 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    772 
    773 	/* Call the generic disklabel extraction routine. */
    774 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
    775 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    776 	if (errstring != NULL)
    777 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
    778 
    779 	/* In-core label now valid. */
    780 	sc->sc_flags |= LDF_VLABEL;
    781 }
    782 
    783 /*
    784  * Construct a ficticious label.
    785  */
    786 static void
    787 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    788 {
    789 
    790 	memset(lp, 0, sizeof(struct disklabel));
    791 
    792 	lp->d_secsize = sc->sc_secsize;
    793 	lp->d_ntracks = sc->sc_nheads;
    794 	lp->d_nsectors = sc->sc_nsectors;
    795 	lp->d_ncylinders = sc->sc_ncylinders;
    796 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    797 	lp->d_type = DTYPE_LD;
    798 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    799 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    800 	lp->d_secperunit = sc->sc_secperunit;
    801 	lp->d_rpm = 7200;
    802 	lp->d_interleave = 1;
    803 	lp->d_flags = 0;
    804 
    805 	lp->d_partitions[RAW_PART].p_offset = 0;
    806 	lp->d_partitions[RAW_PART].p_size =
    807 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    808 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    809 	lp->d_npartitions = RAW_PART + 1;
    810 
    811 	lp->d_magic = DISKMAGIC;
    812 	lp->d_magic2 = DISKMAGIC;
    813 	lp->d_checksum = dkcksum(lp);
    814 }
    815 
    816 /*
    817  * Take a dump.
    818  */
    819 static int
    820 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
    821 {
    822 	char *va = vav;
    823 	struct ld_softc *sc;
    824 	struct disklabel *lp;
    825 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    826 	static int dumping;
    827 
    828 	unit = DISKUNIT(dev);
    829 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
    830 		return (ENXIO);
    831 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    832 		return (ENODEV);
    833 	if (sc->sc_dump == NULL)
    834 		return (ENXIO);
    835 
    836 	/* Check if recursive dump; if so, punt. */
    837 	if (dumping)
    838 		return (EFAULT);
    839 	dumping = 1;
    840 
    841 	/* Convert to disk sectors.  Request must be a multiple of size. */
    842 	part = DISKPART(dev);
    843 	lp = sc->sc_dk.dk_label;
    844 	if ((size % lp->d_secsize) != 0)
    845 		return (EFAULT);
    846 	towrt = size / lp->d_secsize;
    847 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    848 
    849 	nsects = lp->d_partitions[part].p_size;
    850 	sectoff = lp->d_partitions[part].p_offset;
    851 
    852 	/* Check transfer bounds against partition size. */
    853 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    854 		return (EINVAL);
    855 
    856 	/* Offset block number to start of partition. */
    857 	blkno += sectoff;
    858 
    859 	/* Start dumping and return when done. */
    860 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    861 	while (towrt > 0) {
    862 		nblk = min(maxblkcnt, towrt);
    863 
    864 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    865 			return (rv);
    866 
    867 		towrt -= nblk;
    868 		blkno += nblk;
    869 		va += nblk * sc->sc_secsize;
    870 	}
    871 
    872 	dumping = 0;
    873 	return (0);
    874 }
    875 
    876 /*
    877  * Adjust the size of a transfer.
    878  */
    879 static void
    880 ldminphys(struct buf *bp)
    881 {
    882 	struct ld_softc *sc;
    883 
    884 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
    885 
    886 	if (bp->b_bcount > sc->sc_maxxfer)
    887 		bp->b_bcount = sc->sc_maxxfer;
    888 	minphys(bp);
    889 }
    890 
    891 static void
    892 ld_set_properties(struct ld_softc *ld)
    893 {
    894 	prop_dictionary_t disk_info, odisk_info, geom;
    895 
    896 	disk_info = prop_dictionary_create();
    897 
    898 	geom = prop_dictionary_create();
    899 
    900 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
    901 	    ld->sc_secperunit);
    902 
    903 	prop_dictionary_set_uint32(geom, "sector-size",
    904 	    ld->sc_secsize);
    905 
    906 	prop_dictionary_set_uint16(geom, "sectors-per-track",
    907 	    ld->sc_nsectors);
    908 
    909 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
    910 	    ld->sc_nheads);
    911 
    912 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
    913 	    ld->sc_ncylinders);
    914 
    915 	prop_dictionary_set(disk_info, "geometry", geom);
    916 	prop_object_release(geom);
    917 
    918 	prop_dictionary_set(device_properties(ld->sc_dv),
    919 	    "disk-info", disk_info);
    920 
    921 	/*
    922 	 * Don't release disk_info here; we keep a reference to it.
    923 	 * disk_detach() will release it when we go away.
    924 	 */
    925 
    926 	odisk_info = ld->sc_dk.dk_info;
    927 	ld->sc_dk.dk_info = disk_info;
    928 	if (odisk_info)
    929 		prop_object_release(odisk_info);
    930 }
    931 
    932 static void
    933 ld_config_interrupts (struct device *d)
    934 {
    935 	struct ld_softc *sc = device_private(d);
    936 	dkwedge_discover(&sc->sc_dk);
    937 }
    938