Home | History | Annotate | Line # | Download | only in dev
ld.c revision 1.43
      1 /*	$NetBSD: ld.c,v 1.43 2007/02/08 03:19:42 riz 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 1.43 2007/02/08 03:19:42 riz 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 #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 void	ldshutdown(void *);
     78 static void	ldstart(struct ld_softc *);
     79 static void	ld_set_properties(struct ld_softc *);
     80 
     81 extern struct	cfdriver ld_cd;
     82 
     83 static dev_type_open(ldopen);
     84 static dev_type_close(ldclose);
     85 static dev_type_read(ldread);
     86 static dev_type_write(ldwrite);
     87 static dev_type_ioctl(ldioctl);
     88 static dev_type_strategy(ldstrategy);
     89 static dev_type_dump(lddump);
     90 static dev_type_size(ldsize);
     91 
     92 const struct bdevsw ld_bdevsw = {
     93 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
     94 };
     95 
     96 const struct cdevsw ld_cdevsw = {
     97 	ldopen, ldclose, ldread, ldwrite, ldioctl,
     98 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
     99 };
    100 
    101 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
    102 static void	*ld_sdh;
    103 
    104 void
    105 ldattach(struct ld_softc *sc)
    106 {
    107 	char tbuf[9];
    108 
    109 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
    110 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
    111 		return;
    112 	}
    113 
    114 	/* Initialise and attach the disk structure. */
    115 	sc->sc_dk.dk_driver = &lddkdriver;
    116 	sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
    117 	disk_attach(&sc->sc_dk);
    118 
    119 	if (sc->sc_maxxfer > MAXPHYS)
    120 		sc->sc_maxxfer = MAXPHYS;
    121 
    122 	/* Build synthetic geometry if necessary. */
    123 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
    124 	    sc->sc_ncylinders == 0) {
    125 		uint64_t ncyl;
    126 
    127 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
    128 			sc->sc_nheads = 16;
    129 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
    130 			sc->sc_nheads = 32;
    131 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
    132 			sc->sc_nheads = 64;
    133 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
    134 			sc->sc_nheads = 128;
    135 		else
    136 			sc->sc_nheads = 255;
    137 
    138 		sc->sc_nsectors = 63;
    139 		sc->sc_ncylinders = INT_MAX;
    140 		ncyl = sc->sc_secperunit /
    141 		    (sc->sc_nheads * sc->sc_nsectors);
    142 		if (ncyl < INT_MAX)
    143 			sc->sc_ncylinders = (int)ncyl;
    144 	}
    145 
    146 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
    147 	    sc->sc_secsize);
    148 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
    149 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
    150 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
    151 
    152 	ld_set_properties(sc);
    153 
    154 #if NRND > 0
    155 	/* Attach the device into the rnd source list. */
    156 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
    157 	    RND_TYPE_DISK, 0);
    158 #endif
    159 
    160 	/* Set the `shutdownhook'. */
    161 	if (ld_sdh == NULL)
    162 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
    163 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    164 
    165 	/* Discover wedges on this disk. */
    166 	dkwedge_discover(&sc->sc_dk);
    167 }
    168 
    169 int
    170 ldadjqparam(struct ld_softc *sc, int xmax)
    171 {
    172 	int s;
    173 
    174 	s = splbio();
    175 	sc->sc_maxqueuecnt = xmax;
    176 	splx(s);
    177 
    178 	return (0);
    179 }
    180 
    181 int
    182 ldbegindetach(struct ld_softc *sc, int flags)
    183 {
    184 	int s, rv = 0;
    185 
    186 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    187 		return (0);
    188 
    189 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
    190 		return (EBUSY);
    191 
    192 	s = splbio();
    193 	sc->sc_maxqueuecnt = 0;
    194 	sc->sc_flags |= LDF_DETACH;
    195 	while (sc->sc_queuecnt > 0) {
    196 		sc->sc_flags |= LDF_DRAIN;
    197 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
    198 		if (rv)
    199 			break;
    200 	}
    201 	splx(s);
    202 
    203 	return (rv);
    204 }
    205 
    206 void
    207 ldenddetach(struct ld_softc *sc)
    208 {
    209 	int s, bmaj, cmaj, i, mn;
    210 
    211 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    212 		return;
    213 
    214 	/* Wait for commands queued with the hardware to complete. */
    215 	if (sc->sc_queuecnt != 0)
    216 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
    217 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
    218 
    219 	/* Locate the major numbers. */
    220 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
    221 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
    222 
    223 	/* Kill off any queued buffers. */
    224 	s = splbio();
    225 	bufq_drain(sc->sc_bufq);
    226 	splx(s);
    227 
    228 	bufq_free(sc->sc_bufq);
    229 
    230 	/* Nuke the vnodes for any open instances. */
    231 	for (i = 0; i < MAXPARTITIONS; i++) {
    232 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
    233 		vdevgone(bmaj, mn, mn, VBLK);
    234 		vdevgone(cmaj, mn, mn, VCHR);
    235 	}
    236 
    237 	/* Delete all of our wedges. */
    238 	dkwedge_delall(&sc->sc_dk);
    239 
    240 	/* Detach from the disk list. */
    241 	disk_detach(&sc->sc_dk);
    242 
    243 #if NRND > 0
    244 	/* Unhook the entropy source. */
    245 	rnd_detach_source(&sc->sc_rnd_source);
    246 #endif
    247 
    248 	/*
    249 	 * XXX We can't really flush the cache here, beceause the
    250 	 * XXX device may already be non-existent from the controller's
    251 	 * XXX perspective.
    252 	 */
    253 #if 0
    254 	/* Flush the device's cache. */
    255 	if (sc->sc_flush != NULL)
    256 		if ((*sc->sc_flush)(sc) != 0)
    257 			printf("%s: unable to flush cache\n",
    258 			    sc->sc_dv.dv_xname);
    259 #endif
    260 }
    261 
    262 /* ARGSUSED */
    263 static void
    264 ldshutdown(void *cookie)
    265 {
    266 	struct ld_softc *sc;
    267 	int i;
    268 
    269 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
    270 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
    271 			continue;
    272 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    273 			printf("%s: unable to flush cache\n",
    274 			    sc->sc_dv.dv_xname);
    275 	}
    276 }
    277 
    278 /* ARGSUSED */
    279 static int
    280 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
    281 {
    282 	struct ld_softc *sc;
    283 	int error, unit, part;
    284 
    285 	unit = DISKUNIT(dev);
    286 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    287 		return (ENXIO);
    288 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    289 		return (ENODEV);
    290 	part = DISKPART(dev);
    291 
    292 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    293 		return (error);
    294 
    295 	if (sc->sc_dk.dk_openmask == 0) {
    296 		/* Load the partition info if not already loaded. */
    297 		if ((sc->sc_flags & LDF_VLABEL) == 0)
    298 			ldgetdisklabel(sc);
    299 	}
    300 
    301 	/* Check that the partition exists. */
    302 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
    303 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    304 		error = ENXIO;
    305 		goto bad1;
    306 	}
    307 
    308 	/* Ensure only one open at a time. */
    309 	switch (fmt) {
    310 	case S_IFCHR:
    311 		sc->sc_dk.dk_copenmask |= (1 << part);
    312 		break;
    313 	case S_IFBLK:
    314 		sc->sc_dk.dk_bopenmask |= (1 << part);
    315 		break;
    316 	}
    317 	sc->sc_dk.dk_openmask =
    318 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    319 
    320 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    321 	return (0);
    322 
    323  bad1:
    324 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    325 	return (error);
    326 }
    327 
    328 /* ARGSUSED */
    329 static int
    330 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
    331 {
    332 	struct ld_softc *sc;
    333 	int error, part, unit;
    334 
    335 	unit = DISKUNIT(dev);
    336 	part = DISKPART(dev);
    337 	sc = device_lookup(&ld_cd, unit);
    338 
    339 	if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    340 		return (error);
    341 
    342 	switch (fmt) {
    343 	case S_IFCHR:
    344 		sc->sc_dk.dk_copenmask &= ~(1 << part);
    345 		break;
    346 	case S_IFBLK:
    347 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
    348 		break;
    349 	}
    350 	sc->sc_dk.dk_openmask =
    351 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
    352 
    353 	if (sc->sc_dk.dk_openmask == 0) {
    354 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
    355 			printf("%s: unable to flush cache\n",
    356 			    sc->sc_dv.dv_xname);
    357 		if ((sc->sc_flags & LDF_KLABEL) == 0)
    358 			sc->sc_flags &= ~LDF_VLABEL;
    359 	}
    360 
    361 	(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    362 	return (0);
    363 }
    364 
    365 /* ARGSUSED */
    366 static int
    367 ldread(dev_t dev, struct uio *uio, int ioflag)
    368 {
    369 
    370 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
    371 }
    372 
    373 /* ARGSUSED */
    374 static int
    375 ldwrite(dev_t dev, struct uio *uio, int ioflag)
    376 {
    377 
    378 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
    379 }
    380 
    381 /* ARGSUSED */
    382 static int
    383 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct lwp *l)
    384 {
    385 	struct ld_softc *sc;
    386 	int part, unit, error;
    387 #ifdef __HAVE_OLD_DISKLABEL
    388 	struct disklabel newlabel;
    389 #endif
    390 	struct disklabel *lp;
    391 
    392 	unit = DISKUNIT(dev);
    393 	part = DISKPART(dev);
    394 	sc = device_lookup(&ld_cd, unit);
    395 	error = 0;
    396 
    397 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
    398 	if (error != EPASSTHROUGH)
    399 		return (error);
    400 
    401 	switch (cmd) {
    402 	case DIOCGDINFO:
    403 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
    404 		return (0);
    405 
    406 #ifdef __HAVE_OLD_DISKLABEL
    407 	case ODIOCGDINFO:
    408 		newlabel = *(sc->sc_dk.dk_label);
    409 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    410 			return ENOTTY;
    411 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
    412 		return (0);
    413 #endif
    414 
    415 	case DIOCGPART:
    416 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
    417 		((struct partinfo *)addr)->part =
    418 		    &sc->sc_dk.dk_label->d_partitions[part];
    419 		break;
    420 
    421 	case DIOCWDINFO:
    422 	case DIOCSDINFO:
    423 #ifdef __HAVE_OLD_DISKLABEL
    424 	case ODIOCWDINFO:
    425 	case ODIOCSDINFO:
    426 
    427 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    428 			memset(&newlabel, 0, sizeof newlabel);
    429 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
    430 			lp = &newlabel;
    431 		} else
    432 #endif
    433 		lp = (struct disklabel *)addr;
    434 
    435 		if ((flag & FWRITE) == 0)
    436 			return (EBADF);
    437 
    438 		if ((error = lockmgr(&sc->sc_dk.dk_openlock, LK_EXCLUSIVE,
    439 				     NULL)) != 0)
    440 			return (error);
    441 		sc->sc_flags |= LDF_LABELLING;
    442 
    443 		error = setdisklabel(sc->sc_dk.dk_label,
    444 		    lp, /*sc->sc_dk.dk_openmask : */0,
    445 		    sc->sc_dk.dk_cpulabel);
    446 		if (error == 0 && (cmd == DIOCWDINFO
    447 #ifdef __HAVE_OLD_DISKLABEL
    448 		    || cmd == ODIOCWDINFO
    449 #endif
    450 		    ))
    451 			error = writedisklabel(
    452 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
    453 			    ldstrategy, sc->sc_dk.dk_label,
    454 			    sc->sc_dk.dk_cpulabel);
    455 
    456 		sc->sc_flags &= ~LDF_LABELLING;
    457 		(void) lockmgr(&sc->sc_dk.dk_openlock, LK_RELEASE, NULL);
    458 		break;
    459 
    460 	case DIOCKLABEL:
    461 		if ((flag & FWRITE) == 0)
    462 			return (EBADF);
    463 		if (*(int *)addr)
    464 			sc->sc_flags |= LDF_KLABEL;
    465 		else
    466 			sc->sc_flags &= ~LDF_KLABEL;
    467 		break;
    468 
    469 	case DIOCWLABEL:
    470 		if ((flag & FWRITE) == 0)
    471 			return (EBADF);
    472 		if (*(int *)addr)
    473 			sc->sc_flags |= LDF_WLABEL;
    474 		else
    475 			sc->sc_flags &= ~LDF_WLABEL;
    476 		break;
    477 
    478 	case DIOCGDEFLABEL:
    479 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
    480 		break;
    481 
    482 #ifdef __HAVE_OLD_DISKLABEL
    483 	case ODIOCGDEFLABEL:
    484 		ldgetdefaultlabel(sc, &newlabel);
    485 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
    486 			return ENOTTY;
    487 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
    488 		break;
    489 #endif
    490 
    491 	case DIOCCACHESYNC:
    492 		/*
    493 		 * XXX Do we really need to care about having a writable
    494 		 * file descriptor here?
    495 		 */
    496 		if ((flag & FWRITE) == 0)
    497 			error = EBADF;
    498 		else if (sc->sc_flush)
    499 			error = (*sc->sc_flush)(sc);
    500 		else
    501 			error = 0;	/* XXX Error out instead? */
    502 		break;
    503 
    504 	case DIOCAWEDGE:
    505 	    {
    506 	    	struct dkwedge_info *dkw = (void *) addr;
    507 
    508 		if ((flag & FWRITE) == 0)
    509 			return (EBADF);
    510 
    511 		/* If the ioctl happens here, the parent is us. */
    512 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    513 		return (dkwedge_add(dkw));
    514 	    }
    515 
    516 	case DIOCDWEDGE:
    517 	    {
    518 	    	struct dkwedge_info *dkw = (void *) addr;
    519 
    520 		if ((flag & FWRITE) == 0)
    521 			return (EBADF);
    522 
    523 		/* If the ioctl happens here, the parent is us. */
    524 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
    525 		return (dkwedge_del(dkw));
    526 	    }
    527 
    528 	case DIOCLWEDGES:
    529 	    {
    530 	    	struct dkwedge_list *dkwl = (void *) addr;
    531 
    532 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
    533 	    }
    534 
    535 	default:
    536 		error = ENOTTY;
    537 		break;
    538 	}
    539 
    540 	return (error);
    541 }
    542 
    543 static void
    544 ldstrategy(struct buf *bp)
    545 {
    546 	struct ld_softc *sc;
    547 	struct disklabel *lp;
    548 	daddr_t blkno;
    549 	int s, part;
    550 
    551 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    552 	part = DISKPART(bp->b_dev);
    553 
    554 	if ((sc->sc_flags & LDF_DETACH) != 0) {
    555 		bp->b_error = EIO;
    556 		goto bad;
    557 	}
    558 
    559 	lp = sc->sc_dk.dk_label;
    560 
    561 	/*
    562 	 * The transfer must be a whole number of blocks and the offset must
    563 	 * not be negative.
    564 	 */
    565 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
    566 		bp->b_error = EINVAL;
    567 		goto bad;
    568 	}
    569 
    570 	/* If it's a null transfer, return immediately. */
    571 	if (bp->b_bcount == 0)
    572 		goto done;
    573 
    574 	/*
    575 	 * Do bounds checking and adjust the transfer.  If error, process.
    576 	 * If past the end of partition, just return.
    577 	 */
    578 	if (part != RAW_PART &&
    579 	    bounds_check_with_label(&sc->sc_dk, bp,
    580 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
    581 		goto done;
    582 	}
    583 
    584 	/*
    585 	 * Convert the block number to absolute and put it in terms
    586 	 * of the device's logical block size.
    587 	 */
    588 	if (lp->d_secsize == DEV_BSIZE)
    589 		blkno = bp->b_blkno;
    590 	else if (lp->d_secsize > DEV_BSIZE)
    591 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    592 	else
    593 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    594 
    595 	if (part != RAW_PART)
    596 		blkno += lp->d_partitions[part].p_offset;
    597 
    598 	bp->b_rawblkno = blkno;
    599 
    600 	s = splbio();
    601 	BUFQ_PUT(sc->sc_bufq, bp);
    602 	ldstart(sc);
    603 	splx(s);
    604 	return;
    605 
    606  bad:
    607 	bp->b_flags |= B_ERROR;
    608  done:
    609 	bp->b_resid = bp->b_bcount;
    610 	biodone(bp);
    611 }
    612 
    613 static void
    614 ldstart(struct ld_softc *sc)
    615 {
    616 	struct buf *bp;
    617 	int error;
    618 
    619 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
    620 		/* See if there is work to do. */
    621 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
    622 			break;
    623 
    624 		disk_busy(&sc->sc_dk);
    625 		sc->sc_queuecnt++;
    626 
    627 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
    628 			/*
    629 			 * The back-end is running the job; remove it from
    630 			 * the queue.
    631 			 */
    632 			(void) BUFQ_GET(sc->sc_bufq);
    633 		} else  {
    634 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
    635 			sc->sc_queuecnt--;
    636 			if (error == EAGAIN) {
    637 				/*
    638 				 * Temporary resource shortage in the
    639 				 * back-end; just defer the job until
    640 				 * later.
    641 				 *
    642 				 * XXX We might consider a watchdog timer
    643 				 * XXX to make sure we are kicked into action.
    644 				 */
    645 				break;
    646 			} else {
    647 				(void) BUFQ_GET(sc->sc_bufq);
    648 				bp->b_error = error;
    649 				bp->b_flags |= B_ERROR;
    650 				bp->b_resid = bp->b_bcount;
    651 				biodone(bp);
    652 			}
    653 		}
    654 	}
    655 }
    656 
    657 void
    658 lddone(struct ld_softc *sc, struct buf *bp)
    659 {
    660 
    661 	if ((bp->b_flags & B_ERROR) != 0) {
    662 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
    663 		printf("\n");
    664 	}
    665 
    666 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
    667 	    (bp->b_flags & B_READ));
    668 #if NRND > 0
    669 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
    670 #endif
    671 	biodone(bp);
    672 
    673 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
    674 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
    675 			sc->sc_flags &= ~LDF_DRAIN;
    676 			wakeup(&sc->sc_queuecnt);
    677 		}
    678 		ldstart(sc);
    679 	}
    680 }
    681 
    682 static int
    683 ldsize(dev_t dev)
    684 {
    685 	struct ld_softc *sc;
    686 	int part, unit, omask, size;
    687 
    688 	unit = DISKUNIT(dev);
    689 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    690 		return (ENODEV);
    691 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    692 		return (ENODEV);
    693 	part = DISKPART(dev);
    694 
    695 	omask = sc->sc_dk.dk_openmask & (1 << part);
    696 
    697 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
    698 		return (-1);
    699 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
    700 		size = -1;
    701 	else
    702 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
    703 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
    704 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
    705 		return (-1);
    706 
    707 	return (size);
    708 }
    709 
    710 /*
    711  * Load the label information from the specified device.
    712  */
    713 static void
    714 ldgetdisklabel(struct ld_softc *sc)
    715 {
    716 	const char *errstring;
    717 
    718 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
    719 
    720 	/* Call the generic disklabel extraction routine. */
    721 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
    722 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
    723 	if (errstring != NULL)
    724 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
    725 
    726 	/* In-core label now valid. */
    727 	sc->sc_flags |= LDF_VLABEL;
    728 }
    729 
    730 /*
    731  * Construct a ficticious label.
    732  */
    733 static void
    734 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
    735 {
    736 
    737 	memset(lp, 0, sizeof(struct disklabel));
    738 
    739 	lp->d_secsize = sc->sc_secsize;
    740 	lp->d_ntracks = sc->sc_nheads;
    741 	lp->d_nsectors = sc->sc_nsectors;
    742 	lp->d_ncylinders = sc->sc_ncylinders;
    743 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    744 	lp->d_type = DTYPE_LD;
    745 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
    746 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
    747 	lp->d_secperunit = sc->sc_secperunit;
    748 	lp->d_rpm = 7200;
    749 	lp->d_interleave = 1;
    750 	lp->d_flags = 0;
    751 
    752 	lp->d_partitions[RAW_PART].p_offset = 0;
    753 	lp->d_partitions[RAW_PART].p_size =
    754 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    755 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
    756 	lp->d_npartitions = RAW_PART + 1;
    757 
    758 	lp->d_magic = DISKMAGIC;
    759 	lp->d_magic2 = DISKMAGIC;
    760 	lp->d_checksum = dkcksum(lp);
    761 }
    762 
    763 /*
    764  * Take a dump.
    765  */
    766 static int
    767 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
    768 {
    769 	struct ld_softc *sc;
    770 	struct disklabel *lp;
    771 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
    772 	static int dumping;
    773 
    774 	unit = DISKUNIT(dev);
    775 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
    776 		return (ENXIO);
    777 	if ((sc->sc_flags & LDF_ENABLED) == 0)
    778 		return (ENODEV);
    779 	if (sc->sc_dump == NULL)
    780 		return (ENXIO);
    781 
    782 	/* Check if recursive dump; if so, punt. */
    783 	if (dumping)
    784 		return (EFAULT);
    785 	dumping = 1;
    786 
    787 	/* Convert to disk sectors.  Request must be a multiple of size. */
    788 	part = DISKPART(dev);
    789 	lp = sc->sc_dk.dk_label;
    790 	if ((size % lp->d_secsize) != 0)
    791 		return (EFAULT);
    792 	towrt = size / lp->d_secsize;
    793 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
    794 
    795 	nsects = lp->d_partitions[part].p_size;
    796 	sectoff = lp->d_partitions[part].p_offset;
    797 
    798 	/* Check transfer bounds against partition size. */
    799 	if ((blkno < 0) || ((blkno + towrt) > nsects))
    800 		return (EINVAL);
    801 
    802 	/* Offset block number to start of partition. */
    803 	blkno += sectoff;
    804 
    805 	/* Start dumping and return when done. */
    806 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
    807 	while (towrt > 0) {
    808 		nblk = min(maxblkcnt, towrt);
    809 
    810 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
    811 			return (rv);
    812 
    813 		towrt -= nblk;
    814 		blkno += nblk;
    815 		va += nblk * sc->sc_secsize;
    816 	}
    817 
    818 	dumping = 0;
    819 	return (0);
    820 }
    821 
    822 /*
    823  * Adjust the size of a transfer.
    824  */
    825 static void
    826 ldminphys(struct buf *bp)
    827 {
    828 	struct ld_softc *sc;
    829 
    830 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
    831 
    832 	if (bp->b_bcount > sc->sc_maxxfer)
    833 		bp->b_bcount = sc->sc_maxxfer;
    834 	minphys(bp);
    835 }
    836 
    837 static void
    838 ld_set_properties(struct ld_softc *ld)
    839 {
    840 	prop_dictionary_t disk_info, odisk_info, geom;
    841 
    842 	disk_info = prop_dictionary_create();
    843 
    844 	geom = prop_dictionary_create();
    845 
    846 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
    847 	    ld->sc_secperunit);
    848 
    849 	prop_dictionary_set_uint32(geom, "sector-size",
    850 	    ld->sc_secsize);
    851 
    852 	prop_dictionary_set_uint16(geom, "sectors-per-track",
    853 	    ld->sc_nsectors);
    854 
    855 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
    856 	    ld->sc_nheads);
    857 
    858 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
    859 	    ld->sc_ncylinders);
    860 
    861 	prop_dictionary_set(disk_info, "geometry", geom);
    862 	prop_object_release(geom);
    863 
    864 	prop_dictionary_set(device_properties(&ld->sc_dv),
    865 	    "disk-info", disk_info);
    866 
    867 	/*
    868 	 * Don't release disk_info here; we keep a reference to it.
    869 	 * disk_detach() will release it when we go away.
    870 	 */
    871 
    872 	odisk_info = ld->sc_dk.dk_info;
    873 	ld->sc_dk.dk_info = disk_info;
    874 	if (odisk_info)
    875 		prop_object_release(odisk_info);
    876 }
    877