Home | History | Annotate | Line # | Download | only in scsipi
sd.c revision 1.192
      1 /*	$NetBSD: sd.c,v 1.192 2002/12/15 01:55:44 fvdl Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1998 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by 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  * Originally written by Julian Elischer (julian (at) dialix.oz.au)
     41  * for TRW Financial Systems for use under the MACH(2.5) operating system.
     42  *
     43  * TRW Financial Systems, in accordance with their agreement with Carnegie
     44  * Mellon University, makes this software available to CMU to distribute
     45  * or use in any manner that they see fit as long as this message is kept with
     46  * the software. For this reason TFS also grants any other persons or
     47  * organisations permission to use or modify this software.
     48  *
     49  * TFS supplies this software to be publicly redistributed
     50  * on the understanding that TFS is not responsible for the correct
     51  * functioning of this software in any circumstances.
     52  *
     53  * Ported to run under 386BSD by Julian Elischer (julian (at) dialix.oz.au) Sept 1992
     54  */
     55 
     56 #include <sys/cdefs.h>
     57 __KERNEL_RCSID(0, "$NetBSD: sd.c,v 1.192 2002/12/15 01:55:44 fvdl Exp $");
     58 
     59 #include "opt_scsi.h"
     60 #include "opt_bufq.h"
     61 #include "rnd.h"
     62 
     63 #include <sys/param.h>
     64 #include <sys/systm.h>
     65 #include <sys/kernel.h>
     66 #include <sys/file.h>
     67 #include <sys/stat.h>
     68 #include <sys/ioctl.h>
     69 #include <sys/scsiio.h>
     70 #include <sys/buf.h>
     71 #include <sys/uio.h>
     72 #include <sys/malloc.h>
     73 #include <sys/errno.h>
     74 #include <sys/device.h>
     75 #include <sys/disklabel.h>
     76 #include <sys/disk.h>
     77 #include <sys/proc.h>
     78 #include <sys/conf.h>
     79 #include <sys/vnode.h>
     80 #if NRND > 0
     81 #include <sys/rnd.h>
     82 #endif
     83 
     84 #include <dev/scsipi/scsipi_all.h>
     85 #include <dev/scsipi/scsi_all.h>
     86 #include <dev/scsipi/scsipi_disk.h>
     87 #include <dev/scsipi/scsi_disk.h>
     88 #include <dev/scsipi/scsiconf.h>
     89 #include <dev/scsipi/sdvar.h>
     90 
     91 #include "sd.h"		/* NSD_SCSIBUS and NSD_ATAPIBUS come from here */
     92 
     93 #define	SDUNIT(dev)			DISKUNIT(dev)
     94 #define	SDPART(dev)			DISKPART(dev)
     95 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
     96 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
     97 
     98 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
     99 
    100 int	sdlock __P((struct sd_softc *));
    101 void	sdunlock __P((struct sd_softc *));
    102 void	sdminphys __P((struct buf *));
    103 void	sdgetdefaultlabel __P((struct sd_softc *, struct disklabel *));
    104 void	sdgetdisklabel __P((struct sd_softc *));
    105 void	sdstart __P((struct scsipi_periph *));
    106 void	sddone __P((struct scsipi_xfer *));
    107 void	sd_shutdown __P((void *));
    108 int	sd_reassign_blocks __P((struct sd_softc *, u_long));
    109 int	sd_interpret_sense __P((struct scsipi_xfer *));
    110 
    111 extern struct cfdriver sd_cd;
    112 
    113 dev_type_open(sdopen);
    114 dev_type_close(sdclose);
    115 dev_type_read(sdread);
    116 dev_type_write(sdwrite);
    117 dev_type_ioctl(sdioctl);
    118 dev_type_strategy(sdstrategy);
    119 dev_type_dump(sddump);
    120 dev_type_size(sdsize);
    121 
    122 const struct bdevsw sd_bdevsw = {
    123 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
    124 };
    125 
    126 const struct cdevsw sd_cdevsw = {
    127 	sdopen, sdclose, sdread, sdwrite, sdioctl,
    128 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    129 };
    130 
    131 struct dkdriver sddkdriver = { sdstrategy };
    132 
    133 const struct scsipi_periphsw sd_switch = {
    134 	sd_interpret_sense,	/* check our error handler first */
    135 	sdstart,		/* have a queue, served by this */
    136 	NULL,			/* have no async handler */
    137 	sddone,			/* deal with stats at interrupt time */
    138 };
    139 
    140 /*
    141  * Attach routine common to atapi & scsi.
    142  */
    143 void
    144 sdattach(parent, sd, periph, ops)
    145 	struct device *parent;
    146 	struct sd_softc *sd;
    147 	struct scsipi_periph *periph;
    148 	const struct sd_ops *ops;
    149 {
    150 	int error, result;
    151 	struct disk_parms *dp = &sd->params;
    152 	char pbuf[9];
    153 
    154 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
    155 
    156 #ifdef NEW_BUFQ_STRATEGY
    157 	bufq_alloc(&sd->buf_queue, BUFQ_READ_PRIO|BUFQ_SORT_RAWBLOCK);
    158 #else
    159 	bufq_alloc(&sd->buf_queue, BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
    160 #endif
    161 
    162 	/*
    163 	 * Store information needed to contact our base driver
    164 	 */
    165 	sd->sc_periph = periph;
    166 	sd->sc_ops = ops;
    167 
    168 	periph->periph_dev = &sd->sc_dev;
    169 	periph->periph_switch = &sd_switch;
    170 
    171         /*
    172          * Increase our openings to the maximum-per-periph
    173          * supported by the adapter.  This will either be
    174          * clamped down or grown by the adapter if necessary.
    175          */
    176 	periph->periph_openings =
    177 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
    178 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
    179 
    180 	/*
    181 	 * Initialize and attach the disk structure.
    182 	 */
    183 	sd->sc_dk.dk_driver = &sddkdriver;
    184 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
    185 	disk_attach(&sd->sc_dk);
    186 
    187 	/*
    188 	 * Use the subdriver to request information regarding the drive.
    189 	 */
    190 	printf("\n");
    191 
    192 	error = scsipi_start(periph, SSS_START,
    193 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
    194 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
    195 
    196 	if (error)
    197 		result = SDGP_RESULT_OFFLINE;
    198 	else
    199 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
    200 		    XS_CTL_DISCOVERY);
    201 	printf("%s: ", sd->sc_dev.dv_xname);
    202 	switch (result) {
    203 	case SDGP_RESULT_OK:
    204 		format_bytes(pbuf, sizeof(pbuf),
    205 		    (u_int64_t)dp->disksize * dp->blksize);
    206 	        printf(
    207 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %lu sectors",
    208 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
    209 		    dp->disksize);
    210 		break;
    211 
    212 	case SDGP_RESULT_OFFLINE:
    213 		printf("drive offline");
    214 		break;
    215 
    216 	case SDGP_RESULT_UNFORMATTED:
    217 		printf("unformatted media");
    218 		break;
    219 
    220 #ifdef DIAGNOSTIC
    221 	default:
    222 		panic("sdattach: unknown result from get_parms");
    223 		break;
    224 #endif
    225 	}
    226 	printf("\n");
    227 
    228 	/*
    229 	 * Establish a shutdown hook so that we can ensure that
    230 	 * our data has actually made it onto the platter at
    231 	 * shutdown time.  Note that this relies on the fact
    232 	 * that the shutdown hook code puts us at the head of
    233 	 * the list (thus guaranteeing that our hook runs before
    234 	 * our ancestors').
    235 	 */
    236 	if ((sd->sc_sdhook =
    237 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
    238 		printf("%s: WARNING: unable to establish shutdown hook\n",
    239 		    sd->sc_dev.dv_xname);
    240 
    241 #if NRND > 0
    242 	/*
    243 	 * attach the device into the random source list
    244 	 */
    245 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
    246 			  RND_TYPE_DISK, 0);
    247 #endif
    248 }
    249 
    250 int
    251 sdactivate(self, act)
    252 	struct device *self;
    253 	enum devact act;
    254 {
    255 	int rv = 0;
    256 
    257 	switch (act) {
    258 	case DVACT_ACTIVATE:
    259 		rv = EOPNOTSUPP;
    260 		break;
    261 
    262 	case DVACT_DEACTIVATE:
    263 		/*
    264 		 * Nothing to do; we key off the device's DVF_ACTIVE.
    265 		 */
    266 		break;
    267 	}
    268 	return (rv);
    269 }
    270 
    271 int
    272 sddetach(self, flags)
    273 	struct device *self;
    274 	int flags;
    275 {
    276 	struct sd_softc *sd = (struct sd_softc *) self;
    277 	struct buf *bp;
    278 	int s, bmaj, cmaj, i, mn;
    279 
    280 	/* locate the major number */
    281 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
    282 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
    283 
    284 	s = splbio();
    285 
    286 	/* Kill off any queued buffers. */
    287 	while ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
    288 		bp->b_error = EIO;
    289 		bp->b_flags |= B_ERROR;
    290 		bp->b_resid = bp->b_bcount;
    291 		biodone(bp);
    292 	}
    293 
    294 	bufq_free(&sd->buf_queue);
    295 
    296 	/* Kill off any pending commands. */
    297 	scsipi_kill_pending(sd->sc_periph);
    298 
    299 	splx(s);
    300 
    301 	/* Nuke the vnodes for any open instances */
    302 	for (i = 0; i < MAXPARTITIONS; i++) {
    303 		mn = SDMINOR(self->dv_unit, i);
    304 		vdevgone(bmaj, mn, mn, VBLK);
    305 		vdevgone(cmaj, mn, mn, VCHR);
    306 	}
    307 
    308 	/* Detach from the disk list. */
    309 	disk_detach(&sd->sc_dk);
    310 
    311 	/* Get rid of the shutdown hook. */
    312 	shutdownhook_disestablish(sd->sc_sdhook);
    313 
    314 #if NRND > 0
    315 	/* Unhook the entropy source. */
    316 	rnd_detach_source(&sd->rnd_source);
    317 #endif
    318 
    319 	return (0);
    320 }
    321 
    322 /*
    323  * Wait interruptibly for an exclusive lock.
    324  *
    325  * XXX
    326  * Several drivers do this; it should be abstracted and made MP-safe.
    327  */
    328 int
    329 sdlock(sd)
    330 	struct sd_softc *sd;
    331 {
    332 	int error;
    333 
    334 	while ((sd->flags & SDF_LOCKED) != 0) {
    335 		sd->flags |= SDF_WANTED;
    336 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
    337 			return (error);
    338 	}
    339 	sd->flags |= SDF_LOCKED;
    340 	return (0);
    341 }
    342 
    343 /*
    344  * Unlock and wake up any waiters.
    345  */
    346 void
    347 sdunlock(sd)
    348 	struct sd_softc *sd;
    349 {
    350 
    351 	sd->flags &= ~SDF_LOCKED;
    352 	if ((sd->flags & SDF_WANTED) != 0) {
    353 		sd->flags &= ~SDF_WANTED;
    354 		wakeup(sd);
    355 	}
    356 }
    357 
    358 /*
    359  * open the device. Make sure the partition info is a up-to-date as can be.
    360  */
    361 int
    362 sdopen(dev, flag, fmt, p)
    363 	dev_t dev;
    364 	int flag, fmt;
    365 	struct proc *p;
    366 {
    367 	struct sd_softc *sd;
    368 	struct scsipi_periph *periph;
    369 	struct scsipi_adapter *adapt;
    370 	int unit, part;
    371 	int error;
    372 
    373 	unit = SDUNIT(dev);
    374 	if (unit >= sd_cd.cd_ndevs)
    375 		return (ENXIO);
    376 	sd = sd_cd.cd_devs[unit];
    377 	if (sd == NULL)
    378 		return (ENXIO);
    379 
    380 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
    381 		return (ENODEV);
    382 
    383 	periph = sd->sc_periph;
    384 	adapt = periph->periph_channel->chan_adapter;
    385 	part = SDPART(dev);
    386 
    387 	SC_DEBUG(periph, SCSIPI_DB1,
    388 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
    389 	    sd_cd.cd_ndevs, part));
    390 
    391 	/*
    392 	 * If this is the first open of this device, add a reference
    393 	 * to the adapter.
    394 	 */
    395 	if (sd->sc_dk.dk_openmask == 0 &&
    396 	    (error = scsipi_adapter_addref(adapt)) != 0)
    397 		return (error);
    398 
    399 	if ((error = sdlock(sd)) != 0)
    400 		goto bad4;
    401 
    402 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
    403 		/*
    404 		 * If any partition is open, but the disk has been invalidated,
    405 		 * disallow further opens of non-raw partition
    406 		 */
    407 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
    408 		    (part != RAW_PART || fmt != S_IFCHR)) {
    409 			error = EIO;
    410 			goto bad3;
    411 		}
    412 	} else {
    413 		/* Check that it is still responding and ok. */
    414 		error = scsipi_test_unit_ready(periph,
    415 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
    416 		    XS_CTL_IGNORE_NOT_READY);
    417 		if (error)
    418 			goto bad3;
    419 
    420 		/*
    421 		 * Start the pack spinning if necessary. Always allow the
    422 		 * raw parition to be opened, for raw IOCTLs. Data transfers
    423 		 * will check for SDEV_MEDIA_LOADED.
    424 		 */
    425 		error = scsipi_start(periph, SSS_START,
    426 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
    427 		    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
    428 		if (error) {
    429 			if (part != RAW_PART || fmt != S_IFCHR)
    430 				goto bad3;
    431 			else
    432 				goto out;
    433 		}
    434 
    435 		periph->periph_flags |= PERIPH_OPEN;
    436 
    437 		if (periph->periph_flags & PERIPH_REMOVABLE) {
    438 			/* Lock the pack in. */
    439 			error = scsipi_prevent(periph, PR_PREVENT,
    440 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
    441 			    XS_CTL_IGNORE_MEDIA_CHANGE);
    442 			if (error)
    443 				goto bad;
    444 		}
    445 
    446 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
    447 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
    448 
    449 			/*
    450 			 * Load the physical device parameters.
    451 			 *
    452 			 * Note that if media is present but unformatted,
    453 			 * we allow the open (so that it can be formatted!).
    454 			 * The drive should refuse real I/O, if the media is
    455 			 * unformatted.
    456 			 */
    457 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
    458 			    0) == SDGP_RESULT_OFFLINE) {
    459 				error = ENXIO;
    460 				goto bad2;
    461 			}
    462 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
    463 
    464 			/* Load the partition info if not already loaded. */
    465 			sdgetdisklabel(sd);
    466 			SC_DEBUG(periph, SCSIPI_DB3, ("Disklabel loaded "));
    467 		}
    468 	}
    469 
    470 	/* Check that the partition exists. */
    471 	if (part != RAW_PART &&
    472 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
    473 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    474 		error = ENXIO;
    475 		goto bad;
    476 	}
    477 
    478 out:	/* Insure only one open at a time. */
    479 	switch (fmt) {
    480 	case S_IFCHR:
    481 		sd->sc_dk.dk_copenmask |= (1 << part);
    482 		break;
    483 	case S_IFBLK:
    484 		sd->sc_dk.dk_bopenmask |= (1 << part);
    485 		break;
    486 	}
    487 	sd->sc_dk.dk_openmask =
    488 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
    489 
    490 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
    491 	sdunlock(sd);
    492 	return (0);
    493 
    494 bad2:
    495 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
    496 
    497 bad:
    498 	if (sd->sc_dk.dk_openmask == 0) {
    499 		scsipi_prevent(periph, PR_ALLOW,
    500 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE);
    501 		periph->periph_flags &= ~PERIPH_OPEN;
    502 	}
    503 
    504 bad3:
    505 	sdunlock(sd);
    506 bad4:
    507 	if (sd->sc_dk.dk_openmask == 0)
    508 		scsipi_adapter_delref(adapt);
    509 	return (error);
    510 }
    511 
    512 /*
    513  * close the device.. only called if we are the LAST occurence of an open
    514  * device.  Convenient now but usually a pain.
    515  */
    516 int
    517 sdclose(dev, flag, fmt, p)
    518 	dev_t dev;
    519 	int flag, fmt;
    520 	struct proc *p;
    521 {
    522 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
    523 	struct scsipi_periph *periph = sd->sc_periph;
    524 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
    525 	int part = SDPART(dev);
    526 	int error;
    527 
    528 	if ((error = sdlock(sd)) != 0)
    529 		return (error);
    530 
    531 	switch (fmt) {
    532 	case S_IFCHR:
    533 		sd->sc_dk.dk_copenmask &= ~(1 << part);
    534 		break;
    535 	case S_IFBLK:
    536 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
    537 		break;
    538 	}
    539 	sd->sc_dk.dk_openmask =
    540 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
    541 
    542 	if (sd->sc_dk.dk_openmask == 0) {
    543 		/*
    544 		 * If the disk cache needs flushing, and the disk supports
    545 		 * it, do it now.
    546 		 */
    547 		if ((sd->flags & SDF_DIRTY) != 0 &&
    548 		    sd->sc_ops->sdo_flush != NULL) {
    549 			if ((*sd->sc_ops->sdo_flush)(sd, 0)) {
    550 				printf("%s: cache synchronization failed\n",
    551 				    sd->sc_dev.dv_xname);
    552 				sd->flags &= ~SDF_FLUSHING;
    553 			} else
    554 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
    555 		}
    556 
    557 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
    558 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
    559 
    560 		scsipi_wait_drain(periph);
    561 
    562 		if (periph->periph_flags & PERIPH_REMOVABLE) {
    563 			scsipi_prevent(periph, PR_ALLOW,
    564 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
    565 			    XS_CTL_IGNORE_NOT_READY);
    566 		}
    567 		periph->periph_flags &= ~PERIPH_OPEN;
    568 
    569 		scsipi_wait_drain(periph);
    570 
    571 		scsipi_adapter_delref(adapt);
    572 	}
    573 
    574 	sdunlock(sd);
    575 	return (0);
    576 }
    577 
    578 /*
    579  * Actually translate the requested transfer into one the physical driver
    580  * can understand.  The transfer is described by a buf and will include
    581  * only one physical transfer.
    582  */
    583 void
    584 sdstrategy(bp)
    585 	struct buf *bp;
    586 {
    587 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
    588 	struct scsipi_periph *periph = sd->sc_periph;
    589 	struct disklabel *lp;
    590 	daddr_t blkno;
    591 	int s;
    592 	boolean_t sector_aligned;
    593 
    594 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
    595 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
    596 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
    597 	/*
    598 	 * If the device has been made invalid, error out
    599 	 */
    600 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
    601 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
    602 		if (periph->periph_flags & PERIPH_OPEN)
    603 			bp->b_error = EIO;
    604 		else
    605 			bp->b_error = ENODEV;
    606 		goto bad;
    607 	}
    608 
    609 	lp = sd->sc_dk.dk_label;
    610 
    611 	/*
    612 	 * The transfer must be a whole number of blocks, offset must not be
    613 	 * negative.
    614 	 */
    615 	if (lp->d_secsize == DEV_BSIZE) {
    616 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
    617 	} else {
    618 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
    619 	}
    620 	if (!sector_aligned || bp->b_blkno < 0) {
    621 		bp->b_error = EINVAL;
    622 		goto bad;
    623 	}
    624 	/*
    625 	 * If it's a null transfer, return immediatly
    626 	 */
    627 	if (bp->b_bcount == 0)
    628 		goto done;
    629 
    630 	/*
    631 	 * Do bounds checking, adjust transfer. if error, process.
    632 	 * If end of partition, just return.
    633 	 */
    634 	if (SDPART(bp->b_dev) != RAW_PART &&
    635 	    bounds_check_with_label(bp, lp,
    636 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
    637 		goto done;
    638 
    639 	/*
    640 	 * Now convert the block number to absolute and put it in
    641 	 * terms of the device's logical block size.
    642 	 */
    643 	if (lp->d_secsize == DEV_BSIZE)
    644 		blkno = bp->b_blkno;
    645 	else if (lp->d_secsize > DEV_BSIZE)
    646 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    647 	else
    648 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    649 
    650 	if (SDPART(bp->b_dev) != RAW_PART)
    651 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
    652 
    653 	bp->b_rawblkno = blkno;
    654 
    655 	s = splbio();
    656 
    657 	/*
    658 	 * Place it in the queue of disk activities for this disk.
    659 	 *
    660 	 * XXX Only do disksort() if the current operating mode does not
    661 	 * XXX include tagged queueing.
    662 	 */
    663 	BUFQ_PUT(&sd->buf_queue, bp);
    664 
    665 	/*
    666 	 * Tell the device to get going on the transfer if it's
    667 	 * not doing anything, otherwise just wait for completion
    668 	 */
    669 	sdstart(sd->sc_periph);
    670 
    671 	splx(s);
    672 	return;
    673 
    674 bad:
    675 	bp->b_flags |= B_ERROR;
    676 done:
    677 	/*
    678 	 * Correctly set the buf to indicate a completed xfer
    679 	 */
    680 	bp->b_resid = bp->b_bcount;
    681 	biodone(bp);
    682 }
    683 
    684 /*
    685  * sdstart looks to see if there is a buf waiting for the device
    686  * and that the device is not already busy. If both are true,
    687  * It dequeues the buf and creates a scsi command to perform the
    688  * transfer in the buf. The transfer request will call scsipi_done
    689  * on completion, which will in turn call this routine again
    690  * so that the next queued transfer is performed.
    691  * The bufs are queued by the strategy routine (sdstrategy)
    692  *
    693  * This routine is also called after other non-queued requests
    694  * have been made of the scsi driver, to ensure that the queue
    695  * continues to be drained.
    696  *
    697  * must be called at the correct (highish) spl level
    698  * sdstart() is called at splbio from sdstrategy and scsipi_done
    699  */
    700 void
    701 sdstart(periph)
    702 	struct scsipi_periph *periph;
    703 {
    704 	struct sd_softc *sd = (void *)periph->periph_dev;
    705 	struct disklabel *lp = sd->sc_dk.dk_label;
    706 	struct buf *bp = 0;
    707 	struct scsipi_rw_big cmd_big;
    708 #if NSD_SCSIBUS > 0
    709 	struct scsi_rw cmd_small;
    710 #endif
    711 	struct scsipi_generic *cmdp;
    712 	int nblks, cmdlen, error, flags;
    713 
    714 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
    715 	/*
    716 	 * Check if the device has room for another command
    717 	 */
    718 	while (periph->periph_active < periph->periph_openings) {
    719 		/*
    720 		 * there is excess capacity, but a special waits
    721 		 * It'll need the adapter as soon as we clear out of the
    722 		 * way and let it run (user level wait).
    723 		 */
    724 		if (periph->periph_flags & PERIPH_WAITING) {
    725 			periph->periph_flags &= ~PERIPH_WAITING;
    726 			wakeup((caddr_t)periph);
    727 			return;
    728 		}
    729 
    730 		/*
    731 		 * See if there is a buf with work for us to do..
    732 		 */
    733 		if ((bp = BUFQ_GET(&sd->buf_queue)) == NULL)
    734 			return;
    735 
    736 		/*
    737 		 * If the device has become invalid, abort all the
    738 		 * reads and writes until all files have been closed and
    739 		 * re-opened
    740 		 */
    741 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
    742 			bp->b_error = EIO;
    743 			bp->b_flags |= B_ERROR;
    744 			bp->b_resid = bp->b_bcount;
    745 			biodone(bp);
    746 			continue;
    747 		}
    748 
    749 		/*
    750 		 * We have a buf, now we should make a command.
    751 		 */
    752 
    753 		if (lp->d_secsize == DEV_BSIZE)
    754 			nblks = bp->b_bcount >> DEV_BSHIFT;
    755 		else
    756 			nblks = howmany(bp->b_bcount, lp->d_secsize);
    757 
    758 #if NSD_SCSIBUS > 0
    759 		/*
    760 		 *  Fill out the scsi command.  If the transfer will
    761 		 *  fit in a "small" cdb, use it.
    762 		 */
    763 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
    764 		    ((nblks & 0xff) == nblks) &&
    765 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
    766 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
    767 			/*
    768 			 * We can fit in a small cdb.
    769 			 */
    770 			memset(&cmd_small, 0, sizeof(cmd_small));
    771 			cmd_small.opcode = (bp->b_flags & B_READ) ?
    772 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
    773 			_lto3b(bp->b_rawblkno, cmd_small.addr);
    774 			cmd_small.length = nblks & 0xff;
    775 			cmdlen = sizeof(cmd_small);
    776 			cmdp = (struct scsipi_generic *)&cmd_small;
    777 		} else
    778 #endif /* NSD_SCSIBUS > 0 */
    779 		{
    780 			/*
    781 			 * Need a large cdb.
    782 			 */
    783 			memset(&cmd_big, 0, sizeof(cmd_big));
    784 			cmd_big.opcode = (bp->b_flags & B_READ) ?
    785 			    READ_BIG : WRITE_BIG;
    786 			_lto4b(bp->b_rawblkno, cmd_big.addr);
    787 			_lto2b(nblks, cmd_big.length);
    788 			cmdlen = sizeof(cmd_big);
    789 			cmdp = (struct scsipi_generic *)&cmd_big;
    790 		}
    791 
    792 		/* Instrumentation. */
    793 		disk_busy(&sd->sc_dk);
    794 
    795 		/*
    796 		 * Mark the disk dirty so that the cache will be
    797 		 * flushed on close.
    798 		 */
    799 		if ((bp->b_flags & B_READ) == 0)
    800 			sd->flags |= SDF_DIRTY;
    801 
    802 		/*
    803 		 * Figure out what flags to use.
    804 		 */
    805 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
    806 		if (bp->b_flags & B_READ)
    807 			flags |= XS_CTL_DATA_IN;
    808 		else
    809 			flags |= XS_CTL_DATA_OUT;
    810 
    811 		/*
    812 		 * Call the routine that chats with the adapter.
    813 		 * Note: we cannot sleep as we may be an interrupt
    814 		 */
    815 		error = scsipi_command(periph, cmdp, cmdlen,
    816 		    (u_char *)bp->b_data, bp->b_bcount,
    817 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
    818 		if (error) {
    819 			disk_unbusy(&sd->sc_dk, 0, 0);
    820 			printf("%s: not queued, error %d\n",
    821 			    sd->sc_dev.dv_xname, error);
    822 		}
    823 	}
    824 }
    825 
    826 void
    827 sddone(xs)
    828 	struct scsipi_xfer *xs;
    829 {
    830 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
    831 
    832 	if (sd->flags & SDF_FLUSHING) {
    833 		/* Flush completed, no longer dirty. */
    834 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
    835 	}
    836 
    837 	if (xs->bp != NULL) {
    838 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid,
    839 		    (xs->bp->b_flags & B_READ));
    840 #if NRND > 0
    841 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
    842 #endif
    843 	}
    844 }
    845 
    846 void
    847 sdminphys(bp)
    848 	struct buf *bp;
    849 {
    850 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
    851 	long max;
    852 
    853 	/*
    854 	 * If the device is ancient, we want to make sure that
    855 	 * the transfer fits into a 6-byte cdb.
    856 	 *
    857 	 * XXX Note that the SCSI-I spec says that 256-block transfers
    858 	 * are allowed in a 6-byte read/write, and are specified
    859 	 * by settng the "length" to 0.  However, we're conservative
    860 	 * here, allowing only 255-block transfers in case an
    861 	 * ancient device gets confused by length == 0.  A length of 0
    862 	 * in a 10-byte read/write actually means 0 blocks.
    863 	 */
    864 	if ((sd->flags & SDF_ANCIENT) &&
    865 	    ((sd->sc_periph->periph_flags &
    866 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
    867 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
    868 
    869 		if (bp->b_bcount > max)
    870 			bp->b_bcount = max;
    871 	}
    872 
    873 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
    874 }
    875 
    876 int
    877 sdread(dev, uio, ioflag)
    878 	dev_t dev;
    879 	struct uio *uio;
    880 	int ioflag;
    881 {
    882 
    883 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
    884 }
    885 
    886 int
    887 sdwrite(dev, uio, ioflag)
    888 	dev_t dev;
    889 	struct uio *uio;
    890 	int ioflag;
    891 {
    892 
    893 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
    894 }
    895 
    896 /*
    897  * Perform special action on behalf of the user
    898  * Knows about the internals of this device
    899  */
    900 int
    901 sdioctl(dev, cmd, addr, flag, p)
    902 	dev_t dev;
    903 	u_long cmd;
    904 	caddr_t addr;
    905 	int flag;
    906 	struct proc *p;
    907 {
    908 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
    909 	struct scsipi_periph *periph = sd->sc_periph;
    910 	int part = SDPART(dev);
    911 	int error = 0;
    912 #ifdef __HAVE_OLD_DISKLABEL
    913 	struct disklabel *newlabel = NULL;
    914 #endif
    915 
    916 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
    917 
    918 	/*
    919 	 * If the device is not valid, some IOCTLs can still be
    920 	 * handled on the raw partition. Check this here.
    921 	 */
    922 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
    923 		switch (cmd) {
    924 		case DIOCKLABEL:
    925 		case DIOCWLABEL:
    926 		case DIOCLOCK:
    927 		case DIOCEJECT:
    928 		case ODIOCEJECT:
    929 		case DIOCGCACHE:
    930 		case DIOCSCACHE:
    931 		case SCIOCIDENTIFY:
    932 		case OSCIOCIDENTIFY:
    933 		case SCIOCCOMMAND:
    934 		case SCIOCDEBUG:
    935 			if (part == RAW_PART)
    936 				break;
    937 		/* FALLTHROUGH */
    938 		default:
    939 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
    940 				return (ENODEV);
    941 			else
    942 				return (EIO);
    943 		}
    944 	}
    945 
    946 	switch (cmd) {
    947 	case DIOCGDINFO:
    948 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
    949 		return (0);
    950 
    951 #ifdef __HAVE_OLD_DISKLABEL
    952 	case ODIOCGDINFO:
    953 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
    954 		if (newlabel == NULL)
    955 			return EIO;
    956 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
    957 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
    958 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
    959 		else
    960 			error = ENOTTY;
    961 		free(newlabel, M_TEMP);
    962 		return error;
    963 #endif
    964 
    965 	case DIOCGPART:
    966 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
    967 		((struct partinfo *)addr)->part =
    968 		    &sd->sc_dk.dk_label->d_partitions[part];
    969 		return (0);
    970 
    971 	case DIOCWDINFO:
    972 	case DIOCSDINFO:
    973 #ifdef __HAVE_OLD_DISKLABEL
    974 	case ODIOCWDINFO:
    975 	case ODIOCSDINFO:
    976 #endif
    977 	{
    978 		struct disklabel *lp;
    979 
    980 		if ((flag & FWRITE) == 0)
    981 			return (EBADF);
    982 
    983 #ifdef __HAVE_OLD_DISKLABEL
    984  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
    985 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
    986 			if (newlabel == NULL)
    987 				return EIO;
    988 			memset(newlabel, 0, sizeof newlabel);
    989 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
    990 			lp = newlabel;
    991 		} else
    992 #endif
    993 		lp = (struct disklabel *)addr;
    994 
    995 		if ((error = sdlock(sd)) != 0)
    996 			goto bad;
    997 		sd->flags |= SDF_LABELLING;
    998 
    999 		error = setdisklabel(sd->sc_dk.dk_label,
   1000 		    lp, /*sd->sc_dk.dk_openmask : */0,
   1001 		    sd->sc_dk.dk_cpulabel);
   1002 		if (error == 0) {
   1003 			if (cmd == DIOCWDINFO
   1004 #ifdef __HAVE_OLD_DISKLABEL
   1005 			    || cmd == ODIOCWDINFO
   1006 #endif
   1007 			   )
   1008 				error = writedisklabel(SDLABELDEV(dev),
   1009 				    sdstrategy, sd->sc_dk.dk_label,
   1010 				    sd->sc_dk.dk_cpulabel);
   1011 		}
   1012 
   1013 		sd->flags &= ~SDF_LABELLING;
   1014 		sdunlock(sd);
   1015 bad:
   1016 #ifdef __HAVE_OLD_DISKLABEL
   1017 		if (newlabel != NULL)
   1018 			free(newlabel, M_TEMP);
   1019 #endif
   1020 		return (error);
   1021 	}
   1022 
   1023 	case DIOCKLABEL:
   1024 		if (*(int *)addr)
   1025 			periph->periph_flags |= PERIPH_KEEP_LABEL;
   1026 		else
   1027 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
   1028 		return (0);
   1029 
   1030 	case DIOCWLABEL:
   1031 		if ((flag & FWRITE) == 0)
   1032 			return (EBADF);
   1033 		if (*(int *)addr)
   1034 			sd->flags |= SDF_WLABEL;
   1035 		else
   1036 			sd->flags &= ~SDF_WLABEL;
   1037 		return (0);
   1038 
   1039 	case DIOCLOCK:
   1040 		return (scsipi_prevent(periph,
   1041 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
   1042 
   1043 	case DIOCEJECT:
   1044 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
   1045 			return (ENOTTY);
   1046 		if (*(int *)addr == 0) {
   1047 			/*
   1048 			 * Don't force eject: check that we are the only
   1049 			 * partition open. If so, unlock it.
   1050 			 */
   1051 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
   1052 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
   1053 			    sd->sc_dk.dk_openmask) {
   1054 				error = scsipi_prevent(periph, PR_ALLOW,
   1055 				    XS_CTL_IGNORE_NOT_READY);
   1056 				if (error)
   1057 					return (error);
   1058 			} else {
   1059 				return (EBUSY);
   1060 			}
   1061 		}
   1062 		/* FALLTHROUGH */
   1063 	case ODIOCEJECT:
   1064 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
   1065 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
   1066 
   1067 	case DIOCGDEFLABEL:
   1068 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
   1069 		return (0);
   1070 
   1071 #ifdef __HAVE_OLD_DISKLABEL
   1072 	case ODIOCGDEFLABEL:
   1073 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1074 		if (newlabel == NULL)
   1075 			return EIO;
   1076 		sdgetdefaultlabel(sd, newlabel);
   1077 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1078 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
   1079 		else
   1080 			error = ENOTTY;
   1081 		free(newlabel, M_TEMP);
   1082 		return error;
   1083 #endif
   1084 
   1085 	case DIOCGCACHE:
   1086 		if (sd->sc_ops->sdo_getcache != NULL)
   1087 			return ((*sd->sc_ops->sdo_getcache)(sd, (int *) addr));
   1088 
   1089 		/* Not supported on this device. */
   1090 		*(int *) addr = 0;
   1091 		return (0);
   1092 
   1093 	case DIOCSCACHE:
   1094 		if ((flag & FWRITE) == 0)
   1095 			return (EBADF);
   1096 		if (sd->sc_ops->sdo_setcache != NULL)
   1097 			return ((*sd->sc_ops->sdo_setcache)(sd, *(int *) addr));
   1098 
   1099 		/* Not supported on this device. */
   1100 		return (EOPNOTSUPP);
   1101 
   1102 	case DIOCCACHESYNC:
   1103 		/*
   1104 		 * XXX Do we really need to care about having a writeable
   1105 		 * file descriptor here?
   1106 		 */
   1107 		if ((flag & FWRITE) == 0)
   1108 			return (EBADF);
   1109 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0) &&
   1110 		    sd->sc_ops->sdo_flush != NULL) {
   1111 			error = (*sd->sc_ops->sdo_flush)(sd, 0);
   1112 			if (error)
   1113 				sd->flags &= ~SDF_FLUSHING;
   1114 			else
   1115 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
   1116 		} else
   1117 			error = 0;
   1118 		return (error);
   1119 
   1120 	default:
   1121 		if (part != RAW_PART)
   1122 			return (ENOTTY);
   1123 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
   1124 	}
   1125 
   1126 #ifdef DIAGNOSTIC
   1127 	panic("sdioctl: impossible");
   1128 #endif
   1129 }
   1130 
   1131 void
   1132 sdgetdefaultlabel(sd, lp)
   1133 	struct sd_softc *sd;
   1134 	struct disklabel *lp;
   1135 {
   1136 
   1137 	memset(lp, 0, sizeof(struct disklabel));
   1138 
   1139 	lp->d_secsize = sd->params.blksize;
   1140 	lp->d_ntracks = sd->params.heads;
   1141 	lp->d_nsectors = sd->params.sectors;
   1142 	lp->d_ncylinders = sd->params.cyls;
   1143 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1144 
   1145 	switch (scsipi_periph_bustype(sd->sc_periph)) {
   1146 #if NSD_SCSIBUS > 0
   1147 	case SCSIPI_BUSTYPE_SCSI:
   1148 		lp->d_type = DTYPE_SCSI;
   1149 		break;
   1150 #endif
   1151 #if NSD_ATAPIBUS > 0
   1152 	case SCSIPI_BUSTYPE_ATAPI:
   1153 		lp->d_type = DTYPE_ATAPI;
   1154 		break;
   1155 #endif
   1156 	}
   1157 	strncpy(lp->d_typename, sd->name, 16);
   1158 	strncpy(lp->d_packname, "fictitious", 16);
   1159 	lp->d_secperunit = sd->params.disksize;
   1160 	lp->d_rpm = sd->params.rot_rate;
   1161 	lp->d_interleave = 1;
   1162 	lp->d_flags = 0;
   1163 
   1164 	lp->d_partitions[RAW_PART].p_offset = 0;
   1165 	lp->d_partitions[RAW_PART].p_size =
   1166 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
   1167 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1168 	lp->d_npartitions = RAW_PART + 1;
   1169 
   1170 	lp->d_magic = DISKMAGIC;
   1171 	lp->d_magic2 = DISKMAGIC;
   1172 	lp->d_checksum = dkcksum(lp);
   1173 }
   1174 
   1175 
   1176 /*
   1177  * Load the label information on the named device
   1178  */
   1179 void
   1180 sdgetdisklabel(sd)
   1181 	struct sd_softc *sd;
   1182 {
   1183 	struct disklabel *lp = sd->sc_dk.dk_label;
   1184 	char *errstring;
   1185 
   1186 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1187 
   1188 	sdgetdefaultlabel(sd, lp);
   1189 
   1190 	if (lp->d_secpercyl == 0) {
   1191 		lp->d_secpercyl = 100;
   1192 		/* as long as it's not 0 - readdisklabel divides by it (?) */
   1193 	}
   1194 
   1195 	/*
   1196 	 * Call the generic disklabel extraction routine
   1197 	 */
   1198 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
   1199 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
   1200 	if (errstring) {
   1201 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
   1202 		return;
   1203 	}
   1204 }
   1205 
   1206 void
   1207 sd_shutdown(arg)
   1208 	void *arg;
   1209 {
   1210 	struct sd_softc *sd = arg;
   1211 
   1212 	/*
   1213 	 * If the disk cache needs to be flushed, and the disk supports
   1214 	 * it, flush it.  We're cold at this point, so we poll for
   1215 	 * completion.
   1216 	 */
   1217 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
   1218 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
   1219 			printf("%s: cache synchronization failed\n",
   1220 			    sd->sc_dev.dv_xname);
   1221 			sd->flags &= ~SDF_FLUSHING;
   1222 		} else
   1223 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
   1224 	}
   1225 }
   1226 
   1227 /*
   1228  * Tell the device to map out a defective block
   1229  */
   1230 int
   1231 sd_reassign_blocks(sd, blkno)
   1232 	struct sd_softc *sd;
   1233 	u_long blkno;
   1234 {
   1235 	struct scsi_reassign_blocks scsipi_cmd;
   1236 	struct scsi_reassign_blocks_data rbdata;
   1237 
   1238 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
   1239 	memset(&rbdata, 0, sizeof(rbdata));
   1240 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
   1241 
   1242 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
   1243 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
   1244 
   1245 	return (scsipi_command(sd->sc_periph,
   1246 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
   1247 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
   1248 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
   1249 }
   1250 
   1251 /*
   1252  * Check Errors
   1253  */
   1254 int
   1255 sd_interpret_sense(xs)
   1256 	struct scsipi_xfer *xs;
   1257 {
   1258 	struct scsipi_periph *periph = xs->xs_periph;
   1259 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
   1260 	struct sd_softc *sd = (void *)periph->periph_dev;
   1261 	int s, error, retval = EJUSTRETURN;
   1262 
   1263 	/*
   1264 	 * If the periph is already recovering, just do the normal
   1265 	 * error processing.
   1266 	 */
   1267 	if (periph->periph_flags & PERIPH_RECOVERING)
   1268 		return (retval);
   1269 
   1270 	/*
   1271 	 * If the device is not open yet, let the generic code handle it.
   1272 	 */
   1273 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
   1274 		return (retval);
   1275 
   1276 	/*
   1277 	 * If it isn't a extended or extended/deferred error, let
   1278 	 * the generic code handle it.
   1279 	 */
   1280 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
   1281 	    (sense->error_code & SSD_ERRCODE) != 0x71)
   1282 		return (retval);
   1283 
   1284 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
   1285 	    sense->add_sense_code == 0x4) {
   1286 		if (sense->add_sense_code_qual == 0x01)	{
   1287 			/*
   1288 			 * Unit In The Process Of Becoming Ready.
   1289 			 */
   1290 			printf("%s: waiting for pack to spin up...\n",
   1291 			    sd->sc_dev.dv_xname);
   1292 			if (!callout_active(&periph->periph_callout))
   1293 				scsipi_periph_freeze(periph, 1);
   1294 			callout_reset(&periph->periph_callout,
   1295 			    5 * hz, scsipi_periph_timed_thaw, periph);
   1296 			retval = ERESTART;
   1297 		} else if ((sense->add_sense_code_qual == 0x2) &&
   1298 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
   1299 			printf("%s: pack is stopped, restarting...\n",
   1300 			    sd->sc_dev.dv_xname);
   1301 			s = splbio();
   1302 			periph->periph_flags |= PERIPH_RECOVERING;
   1303 			splx(s);
   1304 			error = scsipi_start(periph, SSS_START,
   1305 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
   1306 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
   1307 			if (error) {
   1308 				printf("%s: unable to restart pack\n",
   1309 				    sd->sc_dev.dv_xname);
   1310 				retval = error;
   1311 			} else
   1312 				retval = ERESTART;
   1313 			s = splbio();
   1314 			periph->periph_flags &= ~PERIPH_RECOVERING;
   1315 			splx(s);
   1316 		}
   1317 	}
   1318 	return (retval);
   1319 }
   1320 
   1321 
   1322 int
   1323 sdsize(dev)
   1324 	dev_t dev;
   1325 {
   1326 	struct sd_softc *sd;
   1327 	int part, unit, omask;
   1328 	int size;
   1329 
   1330 	unit = SDUNIT(dev);
   1331 	if (unit >= sd_cd.cd_ndevs)
   1332 		return (-1);
   1333 	sd = sd_cd.cd_devs[unit];
   1334 	if (sd == NULL)
   1335 		return (-1);
   1336 
   1337 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
   1338 		return (-1);
   1339 
   1340 	part = SDPART(dev);
   1341 	omask = sd->sc_dk.dk_openmask & (1 << part);
   1342 
   1343 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
   1344 		return (-1);
   1345 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
   1346 		size = -1;
   1347 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1348 		size = -1;
   1349 	else
   1350 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
   1351 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1352 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
   1353 		return (-1);
   1354 	return (size);
   1355 }
   1356 
   1357 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
   1358 static struct scsipi_xfer sx;
   1359 static int sddoingadump;
   1360 
   1361 /*
   1362  * dump all of physical memory into the partition specified, starting
   1363  * at offset 'dumplo' into the partition.
   1364  */
   1365 int
   1366 sddump(dev, blkno, va, size)
   1367 	dev_t dev;
   1368 	daddr_t blkno;
   1369 	caddr_t va;
   1370 	size_t size;
   1371 {
   1372 	struct sd_softc *sd;	/* disk unit to do the I/O */
   1373 	struct disklabel *lp;	/* disk's disklabel */
   1374 	int	unit, part;
   1375 	int	sectorsize;	/* size of a disk sector */
   1376 	int	nsects;		/* number of sectors in partition */
   1377 	int	sectoff;	/* sector offset of partition */
   1378 	int	totwrt;		/* total number of sectors left to write */
   1379 	int	nwrt;		/* current number of sectors to write */
   1380 	struct scsipi_rw_big cmd;	/* write command */
   1381 	struct scsipi_xfer *xs;	/* ... convenience */
   1382 	struct scsipi_periph *periph;
   1383 	struct scsipi_channel *chan;
   1384 
   1385 	/* Check if recursive dump; if so, punt. */
   1386 	if (sddoingadump)
   1387 		return (EFAULT);
   1388 
   1389 	/* Mark as active early. */
   1390 	sddoingadump = 1;
   1391 
   1392 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
   1393 	part = SDPART(dev);
   1394 
   1395 	/* Check for acceptable drive number. */
   1396 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
   1397 		return (ENXIO);
   1398 
   1399 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
   1400 		return (ENODEV);
   1401 
   1402 	periph = sd->sc_periph;
   1403 	chan = periph->periph_channel;
   1404 
   1405 	/* Make sure it was initialized. */
   1406 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
   1407 		return (ENXIO);
   1408 
   1409 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1410 	lp = sd->sc_dk.dk_label;
   1411 	sectorsize = lp->d_secsize;
   1412 	if ((size % sectorsize) != 0)
   1413 		return (EFAULT);
   1414 	totwrt = size / sectorsize;
   1415 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
   1416 
   1417 	nsects = lp->d_partitions[part].p_size;
   1418 	sectoff = lp->d_partitions[part].p_offset;
   1419 
   1420 	/* Check transfer bounds against partition size. */
   1421 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
   1422 		return (EINVAL);
   1423 
   1424 	/* Offset block number to start of partition. */
   1425 	blkno += sectoff;
   1426 
   1427 	xs = &sx;
   1428 
   1429 	while (totwrt > 0) {
   1430 		nwrt = totwrt;		/* XXX */
   1431 #ifndef	SD_DUMP_NOT_TRUSTED
   1432 		/*
   1433 		 *  Fill out the scsi command
   1434 		 */
   1435 		memset(&cmd, 0, sizeof(cmd));
   1436 		cmd.opcode = WRITE_BIG;
   1437 		_lto4b(blkno, cmd.addr);
   1438 		_lto2b(nwrt, cmd.length);
   1439 		/*
   1440 		 * Fill out the scsipi_xfer structure
   1441 		 *    Note: we cannot sleep as we may be an interrupt
   1442 		 * don't use scsipi_command() as it may want to wait
   1443 		 * for an xs.
   1444 		 */
   1445 		memset(xs, 0, sizeof(sx));
   1446 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
   1447 		    XS_CTL_DATA_OUT;
   1448 		xs->xs_status = 0;
   1449 		xs->xs_periph = periph;
   1450 		xs->xs_retries = SDRETRIES;
   1451 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
   1452 		xs->cmd = (struct scsipi_generic *)&cmd;
   1453 		xs->cmdlen = sizeof(cmd);
   1454 		xs->resid = nwrt * sectorsize;
   1455 		xs->error = XS_NOERROR;
   1456 		xs->bp = 0;
   1457 		xs->data = va;
   1458 		xs->datalen = nwrt * sectorsize;
   1459 
   1460 		/*
   1461 		 * Pass all this info to the scsi driver.
   1462 		 */
   1463 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
   1464 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
   1465 		    xs->error != XS_NOERROR)
   1466 			return (EIO);
   1467 #else	/* SD_DUMP_NOT_TRUSTED */
   1468 		/* Let's just talk about this first... */
   1469 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
   1470 		delay(500 * 1000);	/* half a second */
   1471 #endif	/* SD_DUMP_NOT_TRUSTED */
   1472 
   1473 		/* update block count */
   1474 		totwrt -= nwrt;
   1475 		blkno += nwrt;
   1476 		va += sectorsize * nwrt;
   1477 	}
   1478 	sddoingadump = 0;
   1479 	return (0);
   1480 }
   1481