xref: /netbsd/sys/dev/scsipi/sd.c (revision bf9ec67e)
1 /*	$NetBSD: sd.c,v 1.183 2002/05/15 13:01:27 bouyer 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@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@dialix.oz.au) Sept 1992
54  */
55 
56 #include <sys/cdefs.h>
57 __KERNEL_RCSID(0, "$NetBSD: sd.c,v 1.183 2002/05/15 13:01:27 bouyer Exp $");
58 
59 #include "opt_scsi.h"
60 #include "rnd.h"
61 
62 #include <sys/param.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/file.h>
66 #include <sys/stat.h>
67 #include <sys/ioctl.h>
68 #include <sys/scsiio.h>
69 #include <sys/buf.h>
70 #include <sys/uio.h>
71 #include <sys/malloc.h>
72 #include <sys/errno.h>
73 #include <sys/device.h>
74 #include <sys/disklabel.h>
75 #include <sys/disk.h>
76 #include <sys/proc.h>
77 #include <sys/conf.h>
78 #include <sys/vnode.h>
79 #if NRND > 0
80 #include <sys/rnd.h>
81 #endif
82 
83 #include <dev/scsipi/scsipi_all.h>
84 #include <dev/scsipi/scsi_all.h>
85 #include <dev/scsipi/scsipi_disk.h>
86 #include <dev/scsipi/scsi_disk.h>
87 #include <dev/scsipi/scsiconf.h>
88 #include <dev/scsipi/sdvar.h>
89 
90 #include "sd.h"		/* NSD_SCSIBUS and NSD_ATAPIBUS come from here */
91 
92 #define	SDUNIT(dev)			DISKUNIT(dev)
93 #define	SDPART(dev)			DISKPART(dev)
94 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
95 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
96 
97 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
98 
99 int	sdlock __P((struct sd_softc *));
100 void	sdunlock __P((struct sd_softc *));
101 void	sdminphys __P((struct buf *));
102 void	sdgetdefaultlabel __P((struct sd_softc *, struct disklabel *));
103 void	sdgetdisklabel __P((struct sd_softc *));
104 void	sdstart __P((struct scsipi_periph *));
105 void	sddone __P((struct scsipi_xfer *));
106 void	sd_shutdown __P((void *));
107 int	sd_reassign_blocks __P((struct sd_softc *, u_long));
108 int	sd_interpret_sense __P((struct scsipi_xfer *));
109 
110 extern struct cfdriver sd_cd;
111 
112 struct dkdriver sddkdriver = { sdstrategy };
113 
114 const struct scsipi_periphsw sd_switch = {
115 	sd_interpret_sense,	/* check our error handler first */
116 	sdstart,		/* have a queue, served by this */
117 	NULL,			/* have no async handler */
118 	sddone,			/* deal with stats at interrupt time */
119 };
120 
121 /*
122  * Attach routine common to atapi & scsi.
123  */
124 void
125 sdattach(parent, sd, periph, ops)
126 	struct device *parent;
127 	struct sd_softc *sd;
128 	struct scsipi_periph *periph;
129 	const struct sd_ops *ops;
130 {
131 	int error, result;
132 	struct disk_parms *dp = &sd->params;
133 	char pbuf[9];
134 
135 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
136 
137 	BUFQ_INIT(&sd->buf_queue);
138 
139 	/*
140 	 * Store information needed to contact our base driver
141 	 */
142 	sd->sc_periph = periph;
143 	sd->sc_ops = ops;
144 
145 	periph->periph_dev = &sd->sc_dev;
146 	periph->periph_switch = &sd_switch;
147 
148         /*
149          * Increase our openings to the maximum-per-periph
150          * supported by the adapter.  This will either be
151          * clamped down or grown by the adapter if necessary.
152          */
153 	periph->periph_openings =
154 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
155 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
156 
157 	/*
158 	 * Initialize and attach the disk structure.
159 	 */
160 	sd->sc_dk.dk_driver = &sddkdriver;
161 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
162 	disk_attach(&sd->sc_dk);
163 
164 #ifdef __BROKEN_DK_ESTABLISH
165 	dk_establish(&sd->sc_dk, &sd->sc_dev);		/* XXX */
166 #endif
167 
168 	/*
169 	 * Use the subdriver to request information regarding the drive.
170 	 */
171 	printf("\n");
172 
173 	error = scsipi_start(periph, SSS_START,
174 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
175 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
176 
177 	if (error)
178 		result = SDGP_RESULT_OFFLINE;
179 	else
180 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
181 		    XS_CTL_DISCOVERY);
182 	printf("%s: ", sd->sc_dev.dv_xname);
183 	switch (result) {
184 	case SDGP_RESULT_OK:
185 		format_bytes(pbuf, sizeof(pbuf),
186 		    (u_int64_t)dp->disksize * dp->blksize);
187 	        printf(
188 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %lu sectors",
189 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
190 		    dp->disksize);
191 		break;
192 
193 	case SDGP_RESULT_OFFLINE:
194 		printf("drive offline");
195 		break;
196 
197 	case SDGP_RESULT_UNFORMATTED:
198 		printf("unformatted media");
199 		break;
200 
201 #ifdef DIAGNOSTIC
202 	default:
203 		panic("sdattach: unknown result from get_parms");
204 		break;
205 #endif
206 	}
207 	printf("\n");
208 
209 	/*
210 	 * Establish a shutdown hook so that we can ensure that
211 	 * our data has actually made it onto the platter at
212 	 * shutdown time.  Note that this relies on the fact
213 	 * that the shutdown hook code puts us at the head of
214 	 * the list (thus guaranteeing that our hook runs before
215 	 * our ancestors').
216 	 */
217 	if ((sd->sc_sdhook =
218 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
219 		printf("%s: WARNING: unable to establish shutdown hook\n",
220 		    sd->sc_dev.dv_xname);
221 
222 #if NRND > 0
223 	/*
224 	 * attach the device into the random source list
225 	 */
226 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
227 			  RND_TYPE_DISK, 0);
228 #endif
229 }
230 
231 int
232 sdactivate(self, act)
233 	struct device *self;
234 	enum devact act;
235 {
236 	int rv = 0;
237 
238 	switch (act) {
239 	case DVACT_ACTIVATE:
240 		rv = EOPNOTSUPP;
241 		break;
242 
243 	case DVACT_DEACTIVATE:
244 		/*
245 		 * Nothing to do; we key off the device's DVF_ACTIVE.
246 		 */
247 		break;
248 	}
249 	return (rv);
250 }
251 
252 int
253 sddetach(self, flags)
254 	struct device *self;
255 	int flags;
256 {
257 	struct sd_softc *sd = (struct sd_softc *) self;
258 	struct buf *bp;
259 	int s, bmaj, cmaj, i, mn;
260 
261 	/* locate the major number */
262 	for (bmaj = 0; bmaj <= nblkdev; bmaj++)
263 		if (bdevsw[bmaj].d_open == sdopen)
264 			break;
265 	for (cmaj = 0; cmaj <= nchrdev; cmaj++)
266 		if (cdevsw[cmaj].d_open == sdopen)
267 			break;
268 
269 	s = splbio();
270 
271 	/* Kill off any queued buffers. */
272 	while ((bp = BUFQ_FIRST(&sd->buf_queue)) != NULL) {
273 		BUFQ_REMOVE(&sd->buf_queue, bp);
274 		bp->b_error = EIO;
275 		bp->b_flags |= B_ERROR;
276 		bp->b_resid = bp->b_bcount;
277 		biodone(bp);
278 	}
279 
280 	/* Kill off any pending commands. */
281 	scsipi_kill_pending(sd->sc_periph);
282 
283 	splx(s);
284 
285 	/* Nuke the vnodes for any open instances */
286 	for (i = 0; i < MAXPARTITIONS; i++) {
287 		mn = SDMINOR(self->dv_unit, i);
288 		vdevgone(bmaj, mn, mn, VBLK);
289 		vdevgone(cmaj, mn, mn, VCHR);
290 	}
291 
292 	/* Detach from the disk list. */
293 	disk_detach(&sd->sc_dk);
294 
295 	/* Get rid of the shutdown hook. */
296 	shutdownhook_disestablish(sd->sc_sdhook);
297 
298 #if NRND > 0
299 	/* Unhook the entropy source. */
300 	rnd_detach_source(&sd->rnd_source);
301 #endif
302 
303 	return (0);
304 }
305 
306 /*
307  * Wait interruptibly for an exclusive lock.
308  *
309  * XXX
310  * Several drivers do this; it should be abstracted and made MP-safe.
311  */
312 int
313 sdlock(sd)
314 	struct sd_softc *sd;
315 {
316 	int error;
317 
318 	while ((sd->flags & SDF_LOCKED) != 0) {
319 		sd->flags |= SDF_WANTED;
320 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
321 			return (error);
322 	}
323 	sd->flags |= SDF_LOCKED;
324 	return (0);
325 }
326 
327 /*
328  * Unlock and wake up any waiters.
329  */
330 void
331 sdunlock(sd)
332 	struct sd_softc *sd;
333 {
334 
335 	sd->flags &= ~SDF_LOCKED;
336 	if ((sd->flags & SDF_WANTED) != 0) {
337 		sd->flags &= ~SDF_WANTED;
338 		wakeup(sd);
339 	}
340 }
341 
342 /*
343  * open the device. Make sure the partition info is a up-to-date as can be.
344  */
345 int
346 sdopen(dev, flag, fmt, p)
347 	dev_t dev;
348 	int flag, fmt;
349 	struct proc *p;
350 {
351 	struct sd_softc *sd;
352 	struct scsipi_periph *periph;
353 	struct scsipi_adapter *adapt;
354 	int unit, part;
355 	int error;
356 
357 	unit = SDUNIT(dev);
358 	if (unit >= sd_cd.cd_ndevs)
359 		return (ENXIO);
360 	sd = sd_cd.cd_devs[unit];
361 	if (sd == NULL)
362 		return (ENXIO);
363 
364 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
365 		return (ENODEV);
366 
367 	periph = sd->sc_periph;
368 	adapt = periph->periph_channel->chan_adapter;
369 	part = SDPART(dev);
370 
371 	SC_DEBUG(periph, SCSIPI_DB1,
372 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
373 	    sd_cd.cd_ndevs, part));
374 
375 	/*
376 	 * If this is the first open of this device, add a reference
377 	 * to the adapter.
378 	 */
379 	if (sd->sc_dk.dk_openmask == 0 &&
380 	    (error = scsipi_adapter_addref(adapt)) != 0)
381 		return (error);
382 
383 	if ((error = sdlock(sd)) != 0)
384 		goto bad4;
385 
386 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
387 		/*
388 		 * If any partition is open, but the disk has been invalidated,
389 		 * disallow further opens of non-raw partition
390 		 */
391 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
392 		    (part != RAW_PART || fmt != S_IFCHR)) {
393 			error = EIO;
394 			goto bad3;
395 		}
396 	} else {
397 		/* Check that it is still responding and ok. */
398 		error = scsipi_test_unit_ready(periph,
399 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
400 		    XS_CTL_IGNORE_NOT_READY);
401 		if (error)
402 			goto bad3;
403 
404 		/*
405 		 * Start the pack spinning if necessary. Always allow the
406 		 * raw parition to be opened, for raw IOCTLs. Data transfers
407 		 * will check for SDEV_MEDIA_LOADED.
408 		 */
409 		error = scsipi_start(periph, SSS_START,
410 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
411 		    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
412 		if (error) {
413 			if (part != RAW_PART || fmt != S_IFCHR)
414 				goto bad3;
415 			else
416 				goto out;
417 		}
418 
419 		periph->periph_flags |= PERIPH_OPEN;
420 
421 		if (periph->periph_flags & PERIPH_REMOVABLE) {
422 			/* Lock the pack in. */
423 			error = scsipi_prevent(periph, PR_PREVENT,
424 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
425 			    XS_CTL_IGNORE_MEDIA_CHANGE);
426 			if (error)
427 				goto bad;
428 		}
429 
430 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
431 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
432 
433 			/*
434 			 * Load the physical device parameters.
435 			 *
436 			 * Note that if media is present but unformatted,
437 			 * we allow the open (so that it can be formatted!).
438 			 * The drive should refuse real I/O, if the media is
439 			 * unformatted.
440 			 */
441 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
442 			    0) == SDGP_RESULT_OFFLINE) {
443 				error = ENXIO;
444 				goto bad2;
445 			}
446 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
447 
448 			/* Load the partition info if not already loaded. */
449 			sdgetdisklabel(sd);
450 			SC_DEBUG(periph, SCSIPI_DB3, ("Disklabel loaded "));
451 		}
452 	}
453 
454 	/* Check that the partition exists. */
455 	if (part != RAW_PART &&
456 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
457 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
458 		error = ENXIO;
459 		goto bad;
460 	}
461 
462 out:	/* Insure only one open at a time. */
463 	switch (fmt) {
464 	case S_IFCHR:
465 		sd->sc_dk.dk_copenmask |= (1 << part);
466 		break;
467 	case S_IFBLK:
468 		sd->sc_dk.dk_bopenmask |= (1 << part);
469 		break;
470 	}
471 	sd->sc_dk.dk_openmask =
472 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
473 
474 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
475 	sdunlock(sd);
476 	return (0);
477 
478 bad2:
479 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
480 
481 bad:
482 	if (sd->sc_dk.dk_openmask == 0) {
483 		scsipi_prevent(periph, PR_ALLOW,
484 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE);
485 		periph->periph_flags &= ~PERIPH_OPEN;
486 	}
487 
488 bad3:
489 	sdunlock(sd);
490 bad4:
491 	if (sd->sc_dk.dk_openmask == 0)
492 		scsipi_adapter_delref(adapt);
493 	return (error);
494 }
495 
496 /*
497  * close the device.. only called if we are the LAST occurence of an open
498  * device.  Convenient now but usually a pain.
499  */
500 int
501 sdclose(dev, flag, fmt, p)
502 	dev_t dev;
503 	int flag, fmt;
504 	struct proc *p;
505 {
506 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
507 	struct scsipi_periph *periph = sd->sc_periph;
508 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
509 	int part = SDPART(dev);
510 	int error;
511 
512 	if ((error = sdlock(sd)) != 0)
513 		return (error);
514 
515 	switch (fmt) {
516 	case S_IFCHR:
517 		sd->sc_dk.dk_copenmask &= ~(1 << part);
518 		break;
519 	case S_IFBLK:
520 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
521 		break;
522 	}
523 	sd->sc_dk.dk_openmask =
524 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
525 
526 	if (sd->sc_dk.dk_openmask == 0) {
527 		/*
528 		 * If the disk cache needs flushing, and the disk supports
529 		 * it, do it now.
530 		 */
531 		if ((sd->flags & SDF_DIRTY) != 0 &&
532 		    sd->sc_ops->sdo_flush != NULL) {
533 			if ((*sd->sc_ops->sdo_flush)(sd, 0)) {
534 				printf("%s: cache synchronization failed\n",
535 				    sd->sc_dev.dv_xname);
536 				sd->flags &= ~SDF_FLUSHING;
537 			} else
538 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
539 		}
540 
541 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
542 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
543 
544 		scsipi_wait_drain(periph);
545 
546 		if (periph->periph_flags & PERIPH_REMOVABLE) {
547 			scsipi_prevent(periph, PR_ALLOW,
548 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
549 			    XS_CTL_IGNORE_NOT_READY);
550 		}
551 		periph->periph_flags &= ~PERIPH_OPEN;
552 
553 		scsipi_wait_drain(periph);
554 
555 		scsipi_adapter_delref(adapt);
556 	}
557 
558 	sdunlock(sd);
559 	return (0);
560 }
561 
562 /*
563  * Actually translate the requested transfer into one the physical driver
564  * can understand.  The transfer is described by a buf and will include
565  * only one physical transfer.
566  */
567 void
568 sdstrategy(bp)
569 	struct buf *bp;
570 {
571 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
572 	struct scsipi_periph *periph = sd->sc_periph;
573 	struct disklabel *lp;
574 	daddr_t blkno;
575 	int s;
576 	boolean_t sector_aligned;
577 
578 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
579 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
580 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
581 	/*
582 	 * If the device has been made invalid, error out
583 	 */
584 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
585 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
586 		if (periph->periph_flags & PERIPH_OPEN)
587 			bp->b_error = EIO;
588 		else
589 			bp->b_error = ENODEV;
590 		goto bad;
591 	}
592 
593 	lp = sd->sc_dk.dk_label;
594 
595 	/*
596 	 * The transfer must be a whole number of blocks, offset must not be
597 	 * negative.
598 	 */
599 	if (lp->d_secsize == DEV_BSIZE) {
600 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
601 	} else {
602 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
603 	}
604 	if (!sector_aligned || bp->b_blkno < 0) {
605 		bp->b_error = EINVAL;
606 		goto bad;
607 	}
608 	/*
609 	 * If it's a null transfer, return immediatly
610 	 */
611 	if (bp->b_bcount == 0)
612 		goto done;
613 
614 	/*
615 	 * Do bounds checking, adjust transfer. if error, process.
616 	 * If end of partition, just return.
617 	 */
618 	if (SDPART(bp->b_dev) != RAW_PART &&
619 	    bounds_check_with_label(bp, lp,
620 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
621 		goto done;
622 
623 	/*
624 	 * Now convert the block number to absolute and put it in
625 	 * terms of the device's logical block size.
626 	 */
627 	if (lp->d_secsize == DEV_BSIZE)
628 		blkno = bp->b_blkno;
629 	else if (lp->d_secsize > DEV_BSIZE)
630 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
631 	else
632 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
633 
634 	if (SDPART(bp->b_dev) != RAW_PART)
635 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
636 
637 	bp->b_rawblkno = blkno;
638 
639 	s = splbio();
640 
641 	/*
642 	 * Place it in the queue of disk activities for this disk.
643 	 *
644 	 * XXX Only do disksort() if the current operating mode does not
645 	 * XXX include tagged queueing.
646 	 */
647 	disksort_blkno(&sd->buf_queue, bp);
648 
649 	/*
650 	 * Tell the device to get going on the transfer if it's
651 	 * not doing anything, otherwise just wait for completion
652 	 */
653 	sdstart(sd->sc_periph);
654 
655 	splx(s);
656 	return;
657 
658 bad:
659 	bp->b_flags |= B_ERROR;
660 done:
661 	/*
662 	 * Correctly set the buf to indicate a completed xfer
663 	 */
664 	bp->b_resid = bp->b_bcount;
665 	biodone(bp);
666 }
667 
668 /*
669  * sdstart looks to see if there is a buf waiting for the device
670  * and that the device is not already busy. If both are true,
671  * It dequeues the buf and creates a scsi command to perform the
672  * transfer in the buf. The transfer request will call scsipi_done
673  * on completion, which will in turn call this routine again
674  * so that the next queued transfer is performed.
675  * The bufs are queued by the strategy routine (sdstrategy)
676  *
677  * This routine is also called after other non-queued requests
678  * have been made of the scsi driver, to ensure that the queue
679  * continues to be drained.
680  *
681  * must be called at the correct (highish) spl level
682  * sdstart() is called at splbio from sdstrategy and scsipi_done
683  */
684 void
685 sdstart(periph)
686 	struct scsipi_periph *periph;
687 {
688 	struct sd_softc *sd = (void *)periph->periph_dev;
689 	struct disklabel *lp = sd->sc_dk.dk_label;
690 	struct buf *bp = 0;
691 	struct scsipi_rw_big cmd_big;
692 #if NSD_SCSIBUS > 0
693 	struct scsi_rw cmd_small;
694 #endif
695 	struct scsipi_generic *cmdp;
696 	int nblks, cmdlen, error, flags;
697 
698 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
699 	/*
700 	 * Check if the device has room for another command
701 	 */
702 	while (periph->periph_active < periph->periph_openings) {
703 		/*
704 		 * there is excess capacity, but a special waits
705 		 * It'll need the adapter as soon as we clear out of the
706 		 * way and let it run (user level wait).
707 		 */
708 		if (periph->periph_flags & PERIPH_WAITING) {
709 			periph->periph_flags &= ~PERIPH_WAITING;
710 			wakeup((caddr_t)periph);
711 			return;
712 		}
713 
714 		/*
715 		 * See if there is a buf with work for us to do..
716 		 */
717 		if ((bp = BUFQ_FIRST(&sd->buf_queue)) == NULL)
718 			return;
719 		BUFQ_REMOVE(&sd->buf_queue, bp);
720 
721 		/*
722 		 * If the device has become invalid, abort all the
723 		 * reads and writes until all files have been closed and
724 		 * re-opened
725 		 */
726 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
727 			bp->b_error = EIO;
728 			bp->b_flags |= B_ERROR;
729 			bp->b_resid = bp->b_bcount;
730 			biodone(bp);
731 			continue;
732 		}
733 
734 		/*
735 		 * We have a buf, now we should make a command.
736 		 */
737 
738 		if (lp->d_secsize == DEV_BSIZE)
739 			nblks = bp->b_bcount >> DEV_BSHIFT;
740 		else
741 			nblks = howmany(bp->b_bcount, lp->d_secsize);
742 
743 #if NSD_SCSIBUS > 0
744 		/*
745 		 *  Fill out the scsi command.  If the transfer will
746 		 *  fit in a "small" cdb, use it.
747 		 */
748 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
749 		    ((nblks & 0xff) == nblks) &&
750 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
751 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
752 			/*
753 			 * We can fit in a small cdb.
754 			 */
755 			memset(&cmd_small, 0, sizeof(cmd_small));
756 			cmd_small.opcode = (bp->b_flags & B_READ) ?
757 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
758 			_lto3b(bp->b_rawblkno, cmd_small.addr);
759 			cmd_small.length = nblks & 0xff;
760 			cmdlen = sizeof(cmd_small);
761 			cmdp = (struct scsipi_generic *)&cmd_small;
762 		} else
763 #endif /* NSD_SCSIBUS > 0 */
764 		{
765 			/*
766 			 * Need a large cdb.
767 			 */
768 			memset(&cmd_big, 0, sizeof(cmd_big));
769 			cmd_big.opcode = (bp->b_flags & B_READ) ?
770 			    READ_BIG : WRITE_BIG;
771 			_lto4b(bp->b_rawblkno, cmd_big.addr);
772 			_lto2b(nblks, cmd_big.length);
773 			cmdlen = sizeof(cmd_big);
774 			cmdp = (struct scsipi_generic *)&cmd_big;
775 		}
776 
777 		/* Instrumentation. */
778 		disk_busy(&sd->sc_dk);
779 
780 		/*
781 		 * Mark the disk dirty so that the cache will be
782 		 * flushed on close.
783 		 */
784 		if ((bp->b_flags & B_READ) == 0)
785 			sd->flags |= SDF_DIRTY;
786 
787 		/*
788 		 * Figure out what flags to use.
789 		 */
790 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC;
791 		if (bp->b_flags & B_READ)
792 			flags |= XS_CTL_DATA_IN;
793 		else
794 			flags |= XS_CTL_DATA_OUT;
795 		if (bp->b_flags & B_ORDERED)
796 			flags |= XS_CTL_ORDERED_TAG;
797 		else
798 			flags |= XS_CTL_SIMPLE_TAG;
799 
800 		/*
801 		 * Call the routine that chats with the adapter.
802 		 * Note: we cannot sleep as we may be an interrupt
803 		 */
804 		error = scsipi_command(periph, cmdp, cmdlen,
805 		    (u_char *)bp->b_data, bp->b_bcount,
806 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
807 		if (error) {
808 			disk_unbusy(&sd->sc_dk, 0);
809 			printf("%s: not queued, error %d\n",
810 			    sd->sc_dev.dv_xname, error);
811 		}
812 	}
813 }
814 
815 void
816 sddone(xs)
817 	struct scsipi_xfer *xs;
818 {
819 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
820 
821 	if (sd->flags & SDF_FLUSHING) {
822 		/* Flush completed, no longer dirty. */
823 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
824 	}
825 
826 	if (xs->bp != NULL) {
827 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid);
828 #if NRND > 0
829 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
830 #endif
831 	}
832 }
833 
834 void
835 sdminphys(bp)
836 	struct buf *bp;
837 {
838 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
839 	long max;
840 
841 	/*
842 	 * If the device is ancient, we want to make sure that
843 	 * the transfer fits into a 6-byte cdb.
844 	 *
845 	 * XXX Note that the SCSI-I spec says that 256-block transfers
846 	 * are allowed in a 6-byte read/write, and are specified
847 	 * by settng the "length" to 0.  However, we're conservative
848 	 * here, allowing only 255-block transfers in case an
849 	 * ancient device gets confused by length == 0.  A length of 0
850 	 * in a 10-byte read/write actually means 0 blocks.
851 	 */
852 	if ((sd->flags & SDF_ANCIENT) &&
853 	    ((sd->sc_periph->periph_flags &
854 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
855 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
856 
857 		if (bp->b_bcount > max)
858 			bp->b_bcount = max;
859 	}
860 
861 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
862 }
863 
864 int
865 sdread(dev, uio, ioflag)
866 	dev_t dev;
867 	struct uio *uio;
868 	int ioflag;
869 {
870 
871 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
872 }
873 
874 int
875 sdwrite(dev, uio, ioflag)
876 	dev_t dev;
877 	struct uio *uio;
878 	int ioflag;
879 {
880 
881 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
882 }
883 
884 /*
885  * Perform special action on behalf of the user
886  * Knows about the internals of this device
887  */
888 int
889 sdioctl(dev, cmd, addr, flag, p)
890 	dev_t dev;
891 	u_long cmd;
892 	caddr_t addr;
893 	int flag;
894 	struct proc *p;
895 {
896 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
897 	struct scsipi_periph *periph = sd->sc_periph;
898 	int part = SDPART(dev);
899 	int error;
900 #ifdef __HAVE_OLD_DISKLABEL
901 	struct disklabel newlabel;
902 #endif
903 
904 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
905 
906 	/*
907 	 * If the device is not valid, some IOCTLs can still be
908 	 * handled on the raw partition. Check this here.
909 	 */
910 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
911 		switch (cmd) {
912 		case DIOCKLABEL:
913 		case DIOCWLABEL:
914 		case DIOCLOCK:
915 		case DIOCEJECT:
916 		case ODIOCEJECT:
917 		case DIOCGCACHE:
918 		case DIOCSCACHE:
919 		case SCIOCIDENTIFY:
920 		case OSCIOCIDENTIFY:
921 		case SCIOCCOMMAND:
922 		case SCIOCDEBUG:
923 			if (part == RAW_PART)
924 				break;
925 		/* FALLTHROUGH */
926 		default:
927 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
928 				return (ENODEV);
929 			else
930 				return (EIO);
931 		}
932 	}
933 
934 	switch (cmd) {
935 	case DIOCGDINFO:
936 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
937 		return (0);
938 
939 #ifdef __HAVE_OLD_DISKLABEL
940 	case ODIOCGDINFO:
941 		newlabel = *(sd->sc_dk.dk_label);
942 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
943 			return ENOTTY;
944 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
945 		return (0);
946 #endif
947 
948 	case DIOCGPART:
949 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
950 		((struct partinfo *)addr)->part =
951 		    &sd->sc_dk.dk_label->d_partitions[part];
952 		return (0);
953 
954 	case DIOCWDINFO:
955 	case DIOCSDINFO:
956 #ifdef __HAVE_OLD_DISKLABEL
957 	case ODIOCWDINFO:
958 	case ODIOCSDINFO:
959 #endif
960 	{
961 		struct disklabel *lp;
962 
963 #ifdef __HAVE_OLD_DISKLABEL
964  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
965 			memset(&newlabel, 0, sizeof newlabel);
966 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
967 			lp = &newlabel;
968 		} else
969 #endif
970 		lp = (struct disklabel *)addr;
971 
972 		if ((flag & FWRITE) == 0)
973 			return (EBADF);
974 
975 		if ((error = sdlock(sd)) != 0)
976 			return (error);
977 		sd->flags |= SDF_LABELLING;
978 
979 		error = setdisklabel(sd->sc_dk.dk_label,
980 		    lp, /*sd->sc_dk.dk_openmask : */0,
981 		    sd->sc_dk.dk_cpulabel);
982 		if (error == 0) {
983 			if (cmd == DIOCWDINFO
984 #ifdef __HAVE_OLD_DISKLABEL
985 			    || cmd == ODIOCWDINFO
986 #endif
987 			   )
988 				error = writedisklabel(SDLABELDEV(dev),
989 				    sdstrategy, sd->sc_dk.dk_label,
990 				    sd->sc_dk.dk_cpulabel);
991 		}
992 
993 		sd->flags &= ~SDF_LABELLING;
994 		sdunlock(sd);
995 		return (error);
996 	}
997 
998 	case DIOCKLABEL:
999 		if (*(int *)addr)
1000 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1001 		else
1002 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1003 		return (0);
1004 
1005 	case DIOCWLABEL:
1006 		if ((flag & FWRITE) == 0)
1007 			return (EBADF);
1008 		if (*(int *)addr)
1009 			sd->flags |= SDF_WLABEL;
1010 		else
1011 			sd->flags &= ~SDF_WLABEL;
1012 		return (0);
1013 
1014 	case DIOCLOCK:
1015 		return (scsipi_prevent(periph,
1016 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1017 
1018 	case DIOCEJECT:
1019 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1020 			return (ENOTTY);
1021 		if (*(int *)addr == 0) {
1022 			/*
1023 			 * Don't force eject: check that we are the only
1024 			 * partition open. If so, unlock it.
1025 			 */
1026 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1027 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1028 			    sd->sc_dk.dk_openmask) {
1029 				error = scsipi_prevent(periph, PR_ALLOW,
1030 				    XS_CTL_IGNORE_NOT_READY);
1031 				if (error)
1032 					return (error);
1033 			} else {
1034 				return (EBUSY);
1035 			}
1036 		}
1037 		/* FALLTHROUGH */
1038 	case ODIOCEJECT:
1039 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1040 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1041 
1042 	case DIOCGDEFLABEL:
1043 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1044 		return (0);
1045 
1046 #ifdef __HAVE_OLD_DISKLABEL
1047 	case ODIOCGDEFLABEL:
1048 		sdgetdefaultlabel(sd, &newlabel);
1049 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1050 			return ENOTTY;
1051 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1052 		return (0);
1053 #endif
1054 
1055 	case DIOCGCACHE:
1056 		if (sd->sc_ops->sdo_getcache != NULL)
1057 			return ((*sd->sc_ops->sdo_getcache)(sd, (int *) addr));
1058 
1059 		/* Not supported on this device. */
1060 		*(int *) addr = 0;
1061 		return (0);
1062 
1063 	case DIOCSCACHE:
1064 		if ((flag & FWRITE) == 0)
1065 			return (EBADF);
1066 		if (sd->sc_ops->sdo_setcache != NULL)
1067 			return ((*sd->sc_ops->sdo_setcache)(sd, *(int *) addr));
1068 
1069 		/* Not supported on this device. */
1070 		return (EOPNOTSUPP);
1071 
1072 	case DIOCCACHESYNC:
1073 		/*
1074 		 * XXX Do we really need to care about having a writeable
1075 		 * file descriptor here?
1076 		 */
1077 		if ((flag & FWRITE) == 0)
1078 			return (EBADF);
1079 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0) &&
1080 		    sd->sc_ops->sdo_flush != NULL) {
1081 			error = (*sd->sc_ops->sdo_flush)(sd, 0);
1082 			if (error)
1083 				sd->flags &= ~SDF_FLUSHING;
1084 			else
1085 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1086 		} else
1087 			error = 0;
1088 		return (error);
1089 
1090 	default:
1091 		if (part != RAW_PART)
1092 			return (ENOTTY);
1093 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1094 	}
1095 
1096 #ifdef DIAGNOSTIC
1097 	panic("sdioctl: impossible");
1098 #endif
1099 }
1100 
1101 void
1102 sdgetdefaultlabel(sd, lp)
1103 	struct sd_softc *sd;
1104 	struct disklabel *lp;
1105 {
1106 
1107 	memset(lp, 0, sizeof(struct disklabel));
1108 
1109 	lp->d_secsize = sd->params.blksize;
1110 	lp->d_ntracks = sd->params.heads;
1111 	lp->d_nsectors = sd->params.sectors;
1112 	lp->d_ncylinders = sd->params.cyls;
1113 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1114 
1115 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1116 #if NSD_SCSIBUS > 0
1117 	case SCSIPI_BUSTYPE_SCSI:
1118 		lp->d_type = DTYPE_SCSI;
1119 		break;
1120 #endif
1121 #if NSD_ATAPIBUS > 0
1122 	case SCSIPI_BUSTYPE_ATAPI:
1123 		lp->d_type = DTYPE_ATAPI;
1124 		break;
1125 #endif
1126 	}
1127 	strncpy(lp->d_typename, sd->name, 16);
1128 	strncpy(lp->d_packname, "fictitious", 16);
1129 	lp->d_secperunit = sd->params.disksize;
1130 	lp->d_rpm = sd->params.rot_rate;
1131 	lp->d_interleave = 1;
1132 	lp->d_flags = 0;
1133 
1134 	lp->d_partitions[RAW_PART].p_offset = 0;
1135 	lp->d_partitions[RAW_PART].p_size =
1136 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1137 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1138 	lp->d_npartitions = RAW_PART + 1;
1139 
1140 	lp->d_magic = DISKMAGIC;
1141 	lp->d_magic2 = DISKMAGIC;
1142 	lp->d_checksum = dkcksum(lp);
1143 }
1144 
1145 
1146 /*
1147  * Load the label information on the named device
1148  */
1149 void
1150 sdgetdisklabel(sd)
1151 	struct sd_softc *sd;
1152 {
1153 	struct disklabel *lp = sd->sc_dk.dk_label;
1154 	char *errstring;
1155 
1156 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1157 
1158 	sdgetdefaultlabel(sd, lp);
1159 
1160 	if (lp->d_secpercyl == 0) {
1161 		lp->d_secpercyl = 100;
1162 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1163 	}
1164 
1165 	/*
1166 	 * Call the generic disklabel extraction routine
1167 	 */
1168 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1169 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1170 	if (errstring) {
1171 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1172 		return;
1173 	}
1174 }
1175 
1176 void
1177 sd_shutdown(arg)
1178 	void *arg;
1179 {
1180 	struct sd_softc *sd = arg;
1181 
1182 	/*
1183 	 * If the disk cache needs to be flushed, and the disk supports
1184 	 * it, flush it.  We're cold at this point, so we poll for
1185 	 * completion.
1186 	 */
1187 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
1188 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1189 			printf("%s: cache synchronization failed\n",
1190 			    sd->sc_dev.dv_xname);
1191 			sd->flags &= ~SDF_FLUSHING;
1192 		} else
1193 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1194 	}
1195 }
1196 
1197 /*
1198  * Tell the device to map out a defective block
1199  */
1200 int
1201 sd_reassign_blocks(sd, blkno)
1202 	struct sd_softc *sd;
1203 	u_long blkno;
1204 {
1205 	struct scsi_reassign_blocks scsipi_cmd;
1206 	struct scsi_reassign_blocks_data rbdata;
1207 
1208 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
1209 	memset(&rbdata, 0, sizeof(rbdata));
1210 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
1211 
1212 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
1213 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
1214 
1215 	return (scsipi_command(sd->sc_periph,
1216 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
1217 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
1218 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
1219 }
1220 
1221 /*
1222  * Check Errors
1223  */
1224 int
1225 sd_interpret_sense(xs)
1226 	struct scsipi_xfer *xs;
1227 {
1228 	struct scsipi_periph *periph = xs->xs_periph;
1229 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1230 	struct sd_softc *sd = (void *)periph->periph_dev;
1231 	int s, error, retval = EJUSTRETURN;
1232 
1233 	/*
1234 	 * If the periph is already recovering, just do the normal
1235 	 * error processing.
1236 	 */
1237 	if (periph->periph_flags & PERIPH_RECOVERING)
1238 		return (retval);
1239 
1240 	/*
1241 	 * If the device is not open yet, let the generic code handle it.
1242 	 */
1243 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1244 		return (retval);
1245 
1246 	/*
1247 	 * If it isn't a extended or extended/deferred error, let
1248 	 * the generic code handle it.
1249 	 */
1250 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1251 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1252 		return (retval);
1253 
1254 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1255 	    sense->add_sense_code == 0x4) {
1256 		if (sense->add_sense_code_qual == 0x01)	{
1257 			/*
1258 			 * Unit In The Process Of Becoming Ready.
1259 			 */
1260 			printf("%s: waiting for pack to spin up...\n",
1261 			    sd->sc_dev.dv_xname);
1262 			if (!callout_active(&periph->periph_callout))
1263 				scsipi_periph_freeze(periph, 1);
1264 			callout_reset(&periph->periph_callout,
1265 			    5 * hz, scsipi_periph_timed_thaw, periph);
1266 			retval = ERESTART;
1267 		} else if ((sense->add_sense_code_qual == 0x2) &&
1268 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
1269 			printf("%s: pack is stopped, restarting...\n",
1270 			    sd->sc_dev.dv_xname);
1271 			s = splbio();
1272 			periph->periph_flags |= PERIPH_RECOVERING;
1273 			splx(s);
1274 			error = scsipi_start(periph, SSS_START,
1275 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1276 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1277 			if (error) {
1278 				printf("%s: unable to restart pack\n",
1279 				    sd->sc_dev.dv_xname);
1280 				retval = error;
1281 			} else
1282 				retval = ERESTART;
1283 			s = splbio();
1284 			periph->periph_flags &= ~PERIPH_RECOVERING;
1285 			splx(s);
1286 		}
1287 	}
1288 	return (retval);
1289 }
1290 
1291 
1292 int
1293 sdsize(dev)
1294 	dev_t dev;
1295 {
1296 	struct sd_softc *sd;
1297 	int part, unit, omask;
1298 	int size;
1299 
1300 	unit = SDUNIT(dev);
1301 	if (unit >= sd_cd.cd_ndevs)
1302 		return (-1);
1303 	sd = sd_cd.cd_devs[unit];
1304 	if (sd == NULL)
1305 		return (-1);
1306 
1307 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1308 		return (-1);
1309 
1310 	part = SDPART(dev);
1311 	omask = sd->sc_dk.dk_openmask & (1 << part);
1312 
1313 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1314 		return (-1);
1315 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1316 		size = -1;
1317 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1318 		size = -1;
1319 	else
1320 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1321 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1322 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1323 		return (-1);
1324 	return (size);
1325 }
1326 
1327 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1328 static struct scsipi_xfer sx;
1329 static int sddoingadump;
1330 
1331 /*
1332  * dump all of physical memory into the partition specified, starting
1333  * at offset 'dumplo' into the partition.
1334  */
1335 int
1336 sddump(dev, blkno, va, size)
1337 	dev_t dev;
1338 	daddr_t blkno;
1339 	caddr_t va;
1340 	size_t size;
1341 {
1342 	struct sd_softc *sd;	/* disk unit to do the I/O */
1343 	struct disklabel *lp;	/* disk's disklabel */
1344 	int	unit, part;
1345 	int	sectorsize;	/* size of a disk sector */
1346 	int	nsects;		/* number of sectors in partition */
1347 	int	sectoff;	/* sector offset of partition */
1348 	int	totwrt;		/* total number of sectors left to write */
1349 	int	nwrt;		/* current number of sectors to write */
1350 	struct scsipi_rw_big cmd;	/* write command */
1351 	struct scsipi_xfer *xs;	/* ... convenience */
1352 	struct scsipi_periph *periph;
1353 	struct scsipi_channel *chan;
1354 
1355 	/* Check if recursive dump; if so, punt. */
1356 	if (sddoingadump)
1357 		return (EFAULT);
1358 
1359 	/* Mark as active early. */
1360 	sddoingadump = 1;
1361 
1362 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1363 	part = SDPART(dev);
1364 
1365 	/* Check for acceptable drive number. */
1366 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1367 		return (ENXIO);
1368 
1369 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1370 		return (ENODEV);
1371 
1372 	periph = sd->sc_periph;
1373 	chan = periph->periph_channel;
1374 
1375 	/* Make sure it was initialized. */
1376 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1377 		return (ENXIO);
1378 
1379 	/* Convert to disk sectors.  Request must be a multiple of size. */
1380 	lp = sd->sc_dk.dk_label;
1381 	sectorsize = lp->d_secsize;
1382 	if ((size % sectorsize) != 0)
1383 		return (EFAULT);
1384 	totwrt = size / sectorsize;
1385 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1386 
1387 	nsects = lp->d_partitions[part].p_size;
1388 	sectoff = lp->d_partitions[part].p_offset;
1389 
1390 	/* Check transfer bounds against partition size. */
1391 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1392 		return (EINVAL);
1393 
1394 	/* Offset block number to start of partition. */
1395 	blkno += sectoff;
1396 
1397 	xs = &sx;
1398 
1399 	while (totwrt > 0) {
1400 		nwrt = totwrt;		/* XXX */
1401 #ifndef	SD_DUMP_NOT_TRUSTED
1402 		/*
1403 		 *  Fill out the scsi command
1404 		 */
1405 		memset(&cmd, 0, sizeof(cmd));
1406 		cmd.opcode = WRITE_BIG;
1407 		_lto4b(blkno, cmd.addr);
1408 		_lto2b(nwrt, cmd.length);
1409 		/*
1410 		 * Fill out the scsipi_xfer structure
1411 		 *    Note: we cannot sleep as we may be an interrupt
1412 		 * don't use scsipi_command() as it may want to wait
1413 		 * for an xs.
1414 		 */
1415 		memset(xs, 0, sizeof(sx));
1416 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1417 		    XS_CTL_DATA_OUT;
1418 		xs->xs_status = 0;
1419 		xs->xs_periph = periph;
1420 		xs->xs_retries = SDRETRIES;
1421 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1422 		xs->cmd = (struct scsipi_generic *)&cmd;
1423 		xs->cmdlen = sizeof(cmd);
1424 		xs->resid = nwrt * sectorsize;
1425 		xs->error = XS_NOERROR;
1426 		xs->bp = 0;
1427 		xs->data = va;
1428 		xs->datalen = nwrt * sectorsize;
1429 
1430 		/*
1431 		 * Pass all this info to the scsi driver.
1432 		 */
1433 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1434 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1435 		    xs->error != XS_NOERROR)
1436 			return (EIO);
1437 #else	/* SD_DUMP_NOT_TRUSTED */
1438 		/* Let's just talk about this first... */
1439 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1440 		delay(500 * 1000);	/* half a second */
1441 #endif	/* SD_DUMP_NOT_TRUSTED */
1442 
1443 		/* update block count */
1444 		totwrt -= nwrt;
1445 		blkno += nwrt;
1446 		va += sectorsize * nwrt;
1447 	}
1448 	sddoingadump = 0;
1449 	return (0);
1450 }
1451