xref: /netbsd-src/sys/dev/scsipi/sd.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: sd.c,v 1.195 2003/02/03 23:51:00 thorpej 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 1.195 2003/02/03 23:51:00 thorpej 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 (bounds_check_with_label(bp, lp,
635 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
636 		goto done;
637 
638 	/*
639 	 * Now convert the block number to absolute and put it in
640 	 * terms of the device's logical block size.
641 	 */
642 	if (lp->d_secsize == DEV_BSIZE)
643 		blkno = bp->b_blkno;
644 	else if (lp->d_secsize > DEV_BSIZE)
645 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
646 	else
647 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
648 
649 	if (SDPART(bp->b_dev) != RAW_PART)
650 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
651 
652 	bp->b_rawblkno = blkno;
653 
654 	s = splbio();
655 
656 	/*
657 	 * Place it in the queue of disk activities for this disk.
658 	 *
659 	 * XXX Only do disksort() if the current operating mode does not
660 	 * XXX include tagged queueing.
661 	 */
662 	BUFQ_PUT(&sd->buf_queue, bp);
663 
664 	/*
665 	 * Tell the device to get going on the transfer if it's
666 	 * not doing anything, otherwise just wait for completion
667 	 */
668 	sdstart(sd->sc_periph);
669 
670 	splx(s);
671 	return;
672 
673 bad:
674 	bp->b_flags |= B_ERROR;
675 done:
676 	/*
677 	 * Correctly set the buf to indicate a completed xfer
678 	 */
679 	bp->b_resid = bp->b_bcount;
680 	biodone(bp);
681 }
682 
683 /*
684  * sdstart looks to see if there is a buf waiting for the device
685  * and that the device is not already busy. If both are true,
686  * It dequeues the buf and creates a scsi command to perform the
687  * transfer in the buf. The transfer request will call scsipi_done
688  * on completion, which will in turn call this routine again
689  * so that the next queued transfer is performed.
690  * The bufs are queued by the strategy routine (sdstrategy)
691  *
692  * This routine is also called after other non-queued requests
693  * have been made of the scsi driver, to ensure that the queue
694  * continues to be drained.
695  *
696  * must be called at the correct (highish) spl level
697  * sdstart() is called at splbio from sdstrategy and scsipi_done
698  */
699 void
700 sdstart(periph)
701 	struct scsipi_periph *periph;
702 {
703 	struct sd_softc *sd = (void *)periph->periph_dev;
704 	struct disklabel *lp = sd->sc_dk.dk_label;
705 	struct buf *bp = 0;
706 	struct scsipi_rw_big cmd_big;
707 #if NSD_SCSIBUS > 0
708 	struct scsi_rw cmd_small;
709 #endif
710 	struct scsipi_generic *cmdp;
711 	int nblks, cmdlen, error, flags;
712 
713 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
714 	/*
715 	 * Check if the device has room for another command
716 	 */
717 	while (periph->periph_active < periph->periph_openings) {
718 		/*
719 		 * there is excess capacity, but a special waits
720 		 * It'll need the adapter as soon as we clear out of the
721 		 * way and let it run (user level wait).
722 		 */
723 		if (periph->periph_flags & PERIPH_WAITING) {
724 			periph->periph_flags &= ~PERIPH_WAITING;
725 			wakeup((caddr_t)periph);
726 			return;
727 		}
728 
729 		/*
730 		 * See if there is a buf with work for us to do..
731 		 */
732 		if ((bp = BUFQ_GET(&sd->buf_queue)) == NULL)
733 			return;
734 
735 		/*
736 		 * If the device has become invalid, abort all the
737 		 * reads and writes until all files have been closed and
738 		 * re-opened
739 		 */
740 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
741 			bp->b_error = EIO;
742 			bp->b_flags |= B_ERROR;
743 			bp->b_resid = bp->b_bcount;
744 			biodone(bp);
745 			continue;
746 		}
747 
748 		/*
749 		 * We have a buf, now we should make a command.
750 		 */
751 
752 		if (lp->d_secsize == DEV_BSIZE)
753 			nblks = bp->b_bcount >> DEV_BSHIFT;
754 		else
755 			nblks = howmany(bp->b_bcount, lp->d_secsize);
756 
757 #if NSD_SCSIBUS > 0
758 		/*
759 		 *  Fill out the scsi command.  If the transfer will
760 		 *  fit in a "small" cdb, use it.
761 		 */
762 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
763 		    ((nblks & 0xff) == nblks) &&
764 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
765 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
766 			/*
767 			 * We can fit in a small cdb.
768 			 */
769 			memset(&cmd_small, 0, sizeof(cmd_small));
770 			cmd_small.opcode = (bp->b_flags & B_READ) ?
771 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
772 			_lto3b(bp->b_rawblkno, cmd_small.addr);
773 			cmd_small.length = nblks & 0xff;
774 			cmdlen = sizeof(cmd_small);
775 			cmdp = (struct scsipi_generic *)&cmd_small;
776 		} else
777 #endif /* NSD_SCSIBUS > 0 */
778 		{
779 			/*
780 			 * Need a large cdb.
781 			 */
782 			memset(&cmd_big, 0, sizeof(cmd_big));
783 			cmd_big.opcode = (bp->b_flags & B_READ) ?
784 			    READ_BIG : WRITE_BIG;
785 			_lto4b(bp->b_rawblkno, cmd_big.addr);
786 			_lto2b(nblks, cmd_big.length);
787 			cmdlen = sizeof(cmd_big);
788 			cmdp = (struct scsipi_generic *)&cmd_big;
789 		}
790 
791 		/* Instrumentation. */
792 		disk_busy(&sd->sc_dk);
793 
794 		/*
795 		 * Mark the disk dirty so that the cache will be
796 		 * flushed on close.
797 		 */
798 		if ((bp->b_flags & B_READ) == 0)
799 			sd->flags |= SDF_DIRTY;
800 
801 		/*
802 		 * Figure out what flags to use.
803 		 */
804 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
805 		if (bp->b_flags & B_READ)
806 			flags |= XS_CTL_DATA_IN;
807 		else
808 			flags |= XS_CTL_DATA_OUT;
809 
810 		/*
811 		 * Call the routine that chats with the adapter.
812 		 * Note: we cannot sleep as we may be an interrupt
813 		 */
814 		error = scsipi_command(periph, cmdp, cmdlen,
815 		    (u_char *)bp->b_data, bp->b_bcount,
816 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
817 		if (error) {
818 			disk_unbusy(&sd->sc_dk, 0, 0);
819 			printf("%s: not queued, error %d\n",
820 			    sd->sc_dev.dv_xname, error);
821 		}
822 	}
823 }
824 
825 void
826 sddone(xs)
827 	struct scsipi_xfer *xs;
828 {
829 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
830 
831 	if (sd->flags & SDF_FLUSHING) {
832 		/* Flush completed, no longer dirty. */
833 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
834 	}
835 
836 	if (xs->bp != NULL) {
837 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid,
838 		    (xs->bp->b_flags & B_READ));
839 #if NRND > 0
840 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
841 #endif
842 	}
843 }
844 
845 void
846 sdminphys(bp)
847 	struct buf *bp;
848 {
849 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
850 	long max;
851 
852 	/*
853 	 * If the device is ancient, we want to make sure that
854 	 * the transfer fits into a 6-byte cdb.
855 	 *
856 	 * XXX Note that the SCSI-I spec says that 256-block transfers
857 	 * are allowed in a 6-byte read/write, and are specified
858 	 * by settng the "length" to 0.  However, we're conservative
859 	 * here, allowing only 255-block transfers in case an
860 	 * ancient device gets confused by length == 0.  A length of 0
861 	 * in a 10-byte read/write actually means 0 blocks.
862 	 */
863 	if ((sd->flags & SDF_ANCIENT) &&
864 	    ((sd->sc_periph->periph_flags &
865 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
866 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
867 
868 		if (bp->b_bcount > max)
869 			bp->b_bcount = max;
870 	}
871 
872 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
873 }
874 
875 int
876 sdread(dev, uio, ioflag)
877 	dev_t dev;
878 	struct uio *uio;
879 	int ioflag;
880 {
881 
882 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
883 }
884 
885 int
886 sdwrite(dev, uio, ioflag)
887 	dev_t dev;
888 	struct uio *uio;
889 	int ioflag;
890 {
891 
892 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
893 }
894 
895 /*
896  * Perform special action on behalf of the user
897  * Knows about the internals of this device
898  */
899 int
900 sdioctl(dev, cmd, addr, flag, p)
901 	dev_t dev;
902 	u_long cmd;
903 	caddr_t addr;
904 	int flag;
905 	struct proc *p;
906 {
907 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
908 	struct scsipi_periph *periph = sd->sc_periph;
909 	int part = SDPART(dev);
910 	int error = 0;
911 #ifdef __HAVE_OLD_DISKLABEL
912 	struct disklabel *newlabel = NULL;
913 #endif
914 
915 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
916 
917 	/*
918 	 * If the device is not valid, some IOCTLs can still be
919 	 * handled on the raw partition. Check this here.
920 	 */
921 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
922 		switch (cmd) {
923 		case DIOCKLABEL:
924 		case DIOCWLABEL:
925 		case DIOCLOCK:
926 		case DIOCEJECT:
927 		case ODIOCEJECT:
928 		case DIOCGCACHE:
929 		case DIOCSCACHE:
930 		case SCIOCIDENTIFY:
931 		case OSCIOCIDENTIFY:
932 		case SCIOCCOMMAND:
933 		case SCIOCDEBUG:
934 			if (part == RAW_PART)
935 				break;
936 		/* FALLTHROUGH */
937 		default:
938 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
939 				return (ENODEV);
940 			else
941 				return (EIO);
942 		}
943 	}
944 
945 	switch (cmd) {
946 	case DIOCGDINFO:
947 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
948 		return (0);
949 
950 #ifdef __HAVE_OLD_DISKLABEL
951 	case ODIOCGDINFO:
952 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
953 		if (newlabel == NULL)
954 			return EIO;
955 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
956 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
957 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
958 		else
959 			error = ENOTTY;
960 		free(newlabel, M_TEMP);
961 		return error;
962 #endif
963 
964 	case DIOCGPART:
965 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
966 		((struct partinfo *)addr)->part =
967 		    &sd->sc_dk.dk_label->d_partitions[part];
968 		return (0);
969 
970 	case DIOCWDINFO:
971 	case DIOCSDINFO:
972 #ifdef __HAVE_OLD_DISKLABEL
973 	case ODIOCWDINFO:
974 	case ODIOCSDINFO:
975 #endif
976 	{
977 		struct disklabel *lp;
978 
979 		if ((flag & FWRITE) == 0)
980 			return (EBADF);
981 
982 #ifdef __HAVE_OLD_DISKLABEL
983  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
984 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
985 			if (newlabel == NULL)
986 				return EIO;
987 			memset(newlabel, 0, sizeof newlabel);
988 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
989 			lp = newlabel;
990 		} else
991 #endif
992 		lp = (struct disklabel *)addr;
993 
994 		if ((error = sdlock(sd)) != 0)
995 			goto bad;
996 		sd->flags |= SDF_LABELLING;
997 
998 		error = setdisklabel(sd->sc_dk.dk_label,
999 		    lp, /*sd->sc_dk.dk_openmask : */0,
1000 		    sd->sc_dk.dk_cpulabel);
1001 		if (error == 0) {
1002 			if (cmd == DIOCWDINFO
1003 #ifdef __HAVE_OLD_DISKLABEL
1004 			    || cmd == ODIOCWDINFO
1005 #endif
1006 			   )
1007 				error = writedisklabel(SDLABELDEV(dev),
1008 				    sdstrategy, sd->sc_dk.dk_label,
1009 				    sd->sc_dk.dk_cpulabel);
1010 		}
1011 
1012 		sd->flags &= ~SDF_LABELLING;
1013 		sdunlock(sd);
1014 bad:
1015 #ifdef __HAVE_OLD_DISKLABEL
1016 		if (newlabel != NULL)
1017 			free(newlabel, M_TEMP);
1018 #endif
1019 		return (error);
1020 	}
1021 
1022 	case DIOCKLABEL:
1023 		if (*(int *)addr)
1024 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1025 		else
1026 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1027 		return (0);
1028 
1029 	case DIOCWLABEL:
1030 		if ((flag & FWRITE) == 0)
1031 			return (EBADF);
1032 		if (*(int *)addr)
1033 			sd->flags |= SDF_WLABEL;
1034 		else
1035 			sd->flags &= ~SDF_WLABEL;
1036 		return (0);
1037 
1038 	case DIOCLOCK:
1039 		return (scsipi_prevent(periph,
1040 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1041 
1042 	case DIOCEJECT:
1043 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1044 			return (ENOTTY);
1045 		if (*(int *)addr == 0) {
1046 			/*
1047 			 * Don't force eject: check that we are the only
1048 			 * partition open. If so, unlock it.
1049 			 */
1050 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1051 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1052 			    sd->sc_dk.dk_openmask) {
1053 				error = scsipi_prevent(periph, PR_ALLOW,
1054 				    XS_CTL_IGNORE_NOT_READY);
1055 				if (error)
1056 					return (error);
1057 			} else {
1058 				return (EBUSY);
1059 			}
1060 		}
1061 		/* FALLTHROUGH */
1062 	case ODIOCEJECT:
1063 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1064 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1065 
1066 	case DIOCGDEFLABEL:
1067 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1068 		return (0);
1069 
1070 #ifdef __HAVE_OLD_DISKLABEL
1071 	case ODIOCGDEFLABEL:
1072 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1073 		if (newlabel == NULL)
1074 			return EIO;
1075 		sdgetdefaultlabel(sd, newlabel);
1076 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1077 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1078 		else
1079 			error = ENOTTY;
1080 		free(newlabel, M_TEMP);
1081 		return error;
1082 #endif
1083 
1084 	case DIOCGCACHE:
1085 		if (sd->sc_ops->sdo_getcache != NULL)
1086 			return ((*sd->sc_ops->sdo_getcache)(sd, (int *) addr));
1087 
1088 		/* Not supported on this device. */
1089 		*(int *) addr = 0;
1090 		return (0);
1091 
1092 	case DIOCSCACHE:
1093 		if ((flag & FWRITE) == 0)
1094 			return (EBADF);
1095 		if (sd->sc_ops->sdo_setcache != NULL)
1096 			return ((*sd->sc_ops->sdo_setcache)(sd, *(int *) addr));
1097 
1098 		/* Not supported on this device. */
1099 		return (EOPNOTSUPP);
1100 
1101 	case DIOCCACHESYNC:
1102 		/*
1103 		 * XXX Do we really need to care about having a writable
1104 		 * file descriptor here?
1105 		 */
1106 		if ((flag & FWRITE) == 0)
1107 			return (EBADF);
1108 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0) &&
1109 		    sd->sc_ops->sdo_flush != NULL) {
1110 			error = (*sd->sc_ops->sdo_flush)(sd, 0);
1111 			if (error)
1112 				sd->flags &= ~SDF_FLUSHING;
1113 			else
1114 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1115 		} else
1116 			error = 0;
1117 		return (error);
1118 
1119 	default:
1120 		if (part != RAW_PART)
1121 			return (ENOTTY);
1122 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1123 	}
1124 
1125 #ifdef DIAGNOSTIC
1126 	panic("sdioctl: impossible");
1127 #endif
1128 }
1129 
1130 void
1131 sdgetdefaultlabel(sd, lp)
1132 	struct sd_softc *sd;
1133 	struct disklabel *lp;
1134 {
1135 
1136 	memset(lp, 0, sizeof(struct disklabel));
1137 
1138 	lp->d_secsize = sd->params.blksize;
1139 	lp->d_ntracks = sd->params.heads;
1140 	lp->d_nsectors = sd->params.sectors;
1141 	lp->d_ncylinders = sd->params.cyls;
1142 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1143 
1144 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1145 #if NSD_SCSIBUS > 0
1146 	case SCSIPI_BUSTYPE_SCSI:
1147 		lp->d_type = DTYPE_SCSI;
1148 		break;
1149 #endif
1150 #if NSD_ATAPIBUS > 0
1151 	case SCSIPI_BUSTYPE_ATAPI:
1152 		lp->d_type = DTYPE_ATAPI;
1153 		break;
1154 #endif
1155 	}
1156 	strncpy(lp->d_typename, sd->name, 16);
1157 	strncpy(lp->d_packname, "fictitious", 16);
1158 	lp->d_secperunit = sd->params.disksize;
1159 	lp->d_rpm = sd->params.rot_rate;
1160 	lp->d_interleave = 1;
1161 	lp->d_flags = 0;
1162 
1163 	lp->d_partitions[RAW_PART].p_offset = 0;
1164 	lp->d_partitions[RAW_PART].p_size =
1165 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1166 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1167 	lp->d_npartitions = RAW_PART + 1;
1168 
1169 	lp->d_magic = DISKMAGIC;
1170 	lp->d_magic2 = DISKMAGIC;
1171 	lp->d_checksum = dkcksum(lp);
1172 }
1173 
1174 
1175 /*
1176  * Load the label information on the named device
1177  */
1178 void
1179 sdgetdisklabel(sd)
1180 	struct sd_softc *sd;
1181 {
1182 	struct disklabel *lp = sd->sc_dk.dk_label;
1183 	char *errstring;
1184 
1185 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1186 
1187 	sdgetdefaultlabel(sd, lp);
1188 
1189 	if (lp->d_secpercyl == 0) {
1190 		lp->d_secpercyl = 100;
1191 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1192 	}
1193 
1194 	/*
1195 	 * Call the generic disklabel extraction routine
1196 	 */
1197 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1198 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1199 	if (errstring) {
1200 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1201 		return;
1202 	}
1203 }
1204 
1205 void
1206 sd_shutdown(arg)
1207 	void *arg;
1208 {
1209 	struct sd_softc *sd = arg;
1210 
1211 	/*
1212 	 * If the disk cache needs to be flushed, and the disk supports
1213 	 * it, flush it.  We're cold at this point, so we poll for
1214 	 * completion.
1215 	 */
1216 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
1217 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1218 			printf("%s: cache synchronization failed\n",
1219 			    sd->sc_dev.dv_xname);
1220 			sd->flags &= ~SDF_FLUSHING;
1221 		} else
1222 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1223 	}
1224 }
1225 
1226 /*
1227  * Tell the device to map out a defective block
1228  */
1229 int
1230 sd_reassign_blocks(sd, blkno)
1231 	struct sd_softc *sd;
1232 	u_long blkno;
1233 {
1234 	struct scsi_reassign_blocks scsipi_cmd;
1235 	struct scsi_reassign_blocks_data rbdata;
1236 
1237 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
1238 	memset(&rbdata, 0, sizeof(rbdata));
1239 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
1240 
1241 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
1242 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
1243 
1244 	return (scsipi_command(sd->sc_periph,
1245 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
1246 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
1247 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
1248 }
1249 
1250 /*
1251  * Check Errors
1252  */
1253 int
1254 sd_interpret_sense(xs)
1255 	struct scsipi_xfer *xs;
1256 {
1257 	struct scsipi_periph *periph = xs->xs_periph;
1258 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1259 	struct sd_softc *sd = (void *)periph->periph_dev;
1260 	int s, error, retval = EJUSTRETURN;
1261 
1262 	/*
1263 	 * If the periph is already recovering, just do the normal
1264 	 * error processing.
1265 	 */
1266 	if (periph->periph_flags & PERIPH_RECOVERING)
1267 		return (retval);
1268 
1269 	/*
1270 	 * If the device is not open yet, let the generic code handle it.
1271 	 */
1272 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1273 		return (retval);
1274 
1275 	/*
1276 	 * If it isn't a extended or extended/deferred error, let
1277 	 * the generic code handle it.
1278 	 */
1279 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1280 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1281 		return (retval);
1282 
1283 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1284 	    sense->add_sense_code == 0x4) {
1285 		if (sense->add_sense_code_qual == 0x01)	{
1286 			/*
1287 			 * Unit In The Process Of Becoming Ready.
1288 			 */
1289 			printf("%s: waiting for pack to spin up...\n",
1290 			    sd->sc_dev.dv_xname);
1291 			if (!callout_pending(&periph->periph_callout))
1292 				scsipi_periph_freeze(periph, 1);
1293 			callout_reset(&periph->periph_callout,
1294 			    5 * hz, scsipi_periph_timed_thaw, periph);
1295 			retval = ERESTART;
1296 		} else if ((sense->add_sense_code_qual == 0x2) &&
1297 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
1298 			printf("%s: pack is stopped, restarting...\n",
1299 			    sd->sc_dev.dv_xname);
1300 			s = splbio();
1301 			periph->periph_flags |= PERIPH_RECOVERING;
1302 			splx(s);
1303 			error = scsipi_start(periph, SSS_START,
1304 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1305 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1306 			if (error) {
1307 				printf("%s: unable to restart pack\n",
1308 				    sd->sc_dev.dv_xname);
1309 				retval = error;
1310 			} else
1311 				retval = ERESTART;
1312 			s = splbio();
1313 			periph->periph_flags &= ~PERIPH_RECOVERING;
1314 			splx(s);
1315 		}
1316 	}
1317 	return (retval);
1318 }
1319 
1320 
1321 int
1322 sdsize(dev)
1323 	dev_t dev;
1324 {
1325 	struct sd_softc *sd;
1326 	int part, unit, omask;
1327 	int size;
1328 
1329 	unit = SDUNIT(dev);
1330 	if (unit >= sd_cd.cd_ndevs)
1331 		return (-1);
1332 	sd = sd_cd.cd_devs[unit];
1333 	if (sd == NULL)
1334 		return (-1);
1335 
1336 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1337 		return (-1);
1338 
1339 	part = SDPART(dev);
1340 	omask = sd->sc_dk.dk_openmask & (1 << part);
1341 
1342 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1343 		return (-1);
1344 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1345 		size = -1;
1346 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1347 		size = -1;
1348 	else
1349 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1350 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1351 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1352 		return (-1);
1353 	return (size);
1354 }
1355 
1356 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1357 static struct scsipi_xfer sx;
1358 static int sddoingadump;
1359 
1360 /*
1361  * dump all of physical memory into the partition specified, starting
1362  * at offset 'dumplo' into the partition.
1363  */
1364 int
1365 sddump(dev, blkno, va, size)
1366 	dev_t dev;
1367 	daddr_t blkno;
1368 	caddr_t va;
1369 	size_t size;
1370 {
1371 	struct sd_softc *sd;	/* disk unit to do the I/O */
1372 	struct disklabel *lp;	/* disk's disklabel */
1373 	int	unit, part;
1374 	int	sectorsize;	/* size of a disk sector */
1375 	int	nsects;		/* number of sectors in partition */
1376 	int	sectoff;	/* sector offset of partition */
1377 	int	totwrt;		/* total number of sectors left to write */
1378 	int	nwrt;		/* current number of sectors to write */
1379 	struct scsipi_rw_big cmd;	/* write command */
1380 	struct scsipi_xfer *xs;	/* ... convenience */
1381 	struct scsipi_periph *periph;
1382 	struct scsipi_channel *chan;
1383 
1384 	/* Check if recursive dump; if so, punt. */
1385 	if (sddoingadump)
1386 		return (EFAULT);
1387 
1388 	/* Mark as active early. */
1389 	sddoingadump = 1;
1390 
1391 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1392 	part = SDPART(dev);
1393 
1394 	/* Check for acceptable drive number. */
1395 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1396 		return (ENXIO);
1397 
1398 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1399 		return (ENODEV);
1400 
1401 	periph = sd->sc_periph;
1402 	chan = periph->periph_channel;
1403 
1404 	/* Make sure it was initialized. */
1405 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1406 		return (ENXIO);
1407 
1408 	/* Convert to disk sectors.  Request must be a multiple of size. */
1409 	lp = sd->sc_dk.dk_label;
1410 	sectorsize = lp->d_secsize;
1411 	if ((size % sectorsize) != 0)
1412 		return (EFAULT);
1413 	totwrt = size / sectorsize;
1414 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1415 
1416 	nsects = lp->d_partitions[part].p_size;
1417 	sectoff = lp->d_partitions[part].p_offset;
1418 
1419 	/* Check transfer bounds against partition size. */
1420 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1421 		return (EINVAL);
1422 
1423 	/* Offset block number to start of partition. */
1424 	blkno += sectoff;
1425 
1426 	xs = &sx;
1427 
1428 	while (totwrt > 0) {
1429 		nwrt = totwrt;		/* XXX */
1430 #ifndef	SD_DUMP_NOT_TRUSTED
1431 		/*
1432 		 *  Fill out the scsi command
1433 		 */
1434 		memset(&cmd, 0, sizeof(cmd));
1435 		cmd.opcode = WRITE_BIG;
1436 		_lto4b(blkno, cmd.addr);
1437 		_lto2b(nwrt, cmd.length);
1438 		/*
1439 		 * Fill out the scsipi_xfer structure
1440 		 *    Note: we cannot sleep as we may be an interrupt
1441 		 * don't use scsipi_command() as it may want to wait
1442 		 * for an xs.
1443 		 */
1444 		memset(xs, 0, sizeof(sx));
1445 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1446 		    XS_CTL_DATA_OUT;
1447 		xs->xs_status = 0;
1448 		xs->xs_periph = periph;
1449 		xs->xs_retries = SDRETRIES;
1450 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1451 		xs->cmd = (struct scsipi_generic *)&cmd;
1452 		xs->cmdlen = sizeof(cmd);
1453 		xs->resid = nwrt * sectorsize;
1454 		xs->error = XS_NOERROR;
1455 		xs->bp = 0;
1456 		xs->data = va;
1457 		xs->datalen = nwrt * sectorsize;
1458 
1459 		/*
1460 		 * Pass all this info to the scsi driver.
1461 		 */
1462 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1463 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1464 		    xs->error != XS_NOERROR)
1465 			return (EIO);
1466 #else	/* SD_DUMP_NOT_TRUSTED */
1467 		/* Let's just talk about this first... */
1468 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1469 		delay(500 * 1000);	/* half a second */
1470 #endif	/* SD_DUMP_NOT_TRUSTED */
1471 
1472 		/* update block count */
1473 		totwrt -= nwrt;
1474 		blkno += nwrt;
1475 		va += sectorsize * nwrt;
1476 	}
1477 	sddoingadump = 0;
1478 	return (0);
1479 }
1480