xref: /netbsd-src/sys/dev/scsipi/sd.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11)
1 /*	$NetBSD: sd.c,v 1.223 2004/09/09 19:35:32 bouyer Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2003 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 1.223 2004/09/09 19:35:32 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/scsipi_base.h>
89 #include <dev/scsipi/sdvar.h>
90 
91 #define	SDUNIT(dev)			DISKUNIT(dev)
92 #define	SDPART(dev)			DISKPART(dev)
93 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
94 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
95 
96 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
97 
98 static int	sdlock(struct sd_softc *);
99 static void	sdunlock(struct sd_softc *);
100 static void	sdminphys(struct buf *);
101 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
102 static void	sdgetdisklabel(struct sd_softc *);
103 static void	sdstart(struct scsipi_periph *);
104 static void	sdrestart(void *);
105 static void	sddone(struct scsipi_xfer *);
106 static void	sd_shutdown(void *);
107 static int	sd_interpret_sense(struct scsipi_xfer *);
108 
109 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
110 		    int, int *);
111 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
112 		    int);
113 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
114 		    int);
115 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
116 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
117 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
118 		    int);
119 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
120 		    int);
121 
122 static int	sd_flush(struct sd_softc *, int);
123 static int	sd_getcache(struct sd_softc *, int *);
124 static int	sd_setcache(struct sd_softc *, int);
125 
126 static int	sdmatch(struct device *, struct cfdata *, void *);
127 static void	sdattach(struct device *, struct device *, void *);
128 static int	sdactivate(struct device *, enum devact);
129 static int	sddetach(struct device *, int);
130 
131 CFATTACH_DECL(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
132     sdactivate);
133 
134 extern struct cfdriver sd_cd;
135 
136 static const struct scsipi_inquiry_pattern sd_patterns[] = {
137 	{T_DIRECT, T_FIXED,
138 	 "",         "",                 ""},
139 	{T_DIRECT, T_REMOV,
140 	 "",         "",                 ""},
141 	{T_OPTICAL, T_FIXED,
142 	 "",         "",                 ""},
143 	{T_OPTICAL, T_REMOV,
144 	 "",         "",                 ""},
145 	{T_SIMPLE_DIRECT, T_FIXED,
146 	 "",         "",                 ""},
147 	{T_SIMPLE_DIRECT, T_REMOV,
148 	 "",         "",                 ""},
149 };
150 
151 static dev_type_open(sdopen);
152 static dev_type_close(sdclose);
153 static dev_type_read(sdread);
154 static dev_type_write(sdwrite);
155 static dev_type_ioctl(sdioctl);
156 static dev_type_strategy(sdstrategy);
157 static dev_type_dump(sddump);
158 static dev_type_size(sdsize);
159 
160 const struct bdevsw sd_bdevsw = {
161 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
162 };
163 
164 const struct cdevsw sd_cdevsw = {
165 	sdopen, sdclose, sdread, sdwrite, sdioctl,
166 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
167 };
168 
169 static struct dkdriver sddkdriver = { sdstrategy };
170 
171 static const struct scsipi_periphsw sd_switch = {
172 	sd_interpret_sense,	/* check our error handler first */
173 	sdstart,		/* have a queue, served by this */
174 	NULL,			/* have no async handler */
175 	sddone,			/* deal with stats at interrupt time */
176 };
177 
178 struct sd_mode_sense_data {
179 	/*
180 	 * XXX
181 	 * We are not going to parse this as-is -- it just has to be large
182 	 * enough.
183 	 */
184 	union {
185 		struct scsipi_mode_header small;
186 		struct scsipi_mode_header_big big;
187 	} header;
188 	struct scsi_blk_desc blk_desc;
189 	union scsi_disk_pages pages;
190 };
191 
192 /*
193  * The routine called by the low level scsi routine when it discovers
194  * A device suitable for this driver
195  */
196 static int
197 sdmatch(struct device *parent, struct cfdata *match, void *aux)
198 {
199 	struct scsipibus_attach_args *sa = aux;
200 	int priority;
201 
202 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
203 	    (caddr_t)sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
204 	    sizeof(sd_patterns[0]), &priority);
205 
206 	return (priority);
207 }
208 
209 /*
210  * Attach routine common to atapi & scsi.
211  */
212 static void
213 sdattach(struct device *parent, struct device *self, void *aux)
214 {
215 	struct sd_softc *sd = (void *)self;
216 	struct scsipibus_attach_args *sa = aux;
217 	struct scsipi_periph *periph = sa->sa_periph;
218 	int error, result;
219 	struct disk_parms *dp = &sd->params;
220 	char pbuf[9];
221 
222 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
223 
224 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
225 	if (sd->type == T_SIMPLE_DIRECT)
226 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
227 
228 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
229 	    periph->periph_version == 0)
230 		sd->flags |= SDF_ANCIENT;
231 
232 	bufq_alloc(&sd->buf_queue,
233 	    BUFQ_DISK_DEFAULT_STRAT()|BUFQ_SORT_RAWBLOCK);
234 
235 	callout_init(&sd->sc_callout);
236 
237 	/*
238 	 * Store information needed to contact our base driver
239 	 */
240 	sd->sc_periph = periph;
241 
242 	periph->periph_dev = &sd->sc_dev;
243 	periph->periph_switch = &sd_switch;
244 
245         /*
246          * Increase our openings to the maximum-per-periph
247          * supported by the adapter.  This will either be
248          * clamped down or grown by the adapter if necessary.
249          */
250 	periph->periph_openings =
251 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
252 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
253 
254 	/*
255 	 * Initialize and attach the disk structure.
256 	 */
257 	sd->sc_dk.dk_driver = &sddkdriver;
258 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
259 	disk_attach(&sd->sc_dk);
260 
261 	/*
262 	 * Use the subdriver to request information regarding the drive.
263 	 */
264 	aprint_naive("\n");
265 	aprint_normal("\n");
266 
267 	error = scsipi_test_unit_ready(periph,
268 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
269 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
270 
271 	if (error)
272 		result = SDGP_RESULT_OFFLINE;
273 	else
274 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
275 	aprint_normal("%s: ", sd->sc_dev.dv_xname);
276 	switch (result) {
277 	case SDGP_RESULT_OK:
278 		format_bytes(pbuf, sizeof(pbuf),
279 		    (u_int64_t)dp->disksize * dp->blksize);
280 	        aprint_normal(
281 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
282 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
283 		    (unsigned long long)dp->disksize);
284 		break;
285 
286 	case SDGP_RESULT_OFFLINE:
287 		aprint_normal("drive offline");
288 		break;
289 
290 	case SDGP_RESULT_UNFORMATTED:
291 		aprint_normal("unformatted media");
292 		break;
293 
294 #ifdef DIAGNOSTIC
295 	default:
296 		panic("sdattach: unknown result from get_parms");
297 		break;
298 #endif
299 	}
300 	aprint_normal("\n");
301 
302 	/*
303 	 * Establish a shutdown hook so that we can ensure that
304 	 * our data has actually made it onto the platter at
305 	 * shutdown time.  Note that this relies on the fact
306 	 * that the shutdown hook code puts us at the head of
307 	 * the list (thus guaranteeing that our hook runs before
308 	 * our ancestors').
309 	 */
310 	if ((sd->sc_sdhook =
311 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
312 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
313 		    sd->sc_dev.dv_xname);
314 
315 #if NRND > 0
316 	/*
317 	 * attach the device into the random source list
318 	 */
319 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
320 			  RND_TYPE_DISK, 0);
321 #endif
322 }
323 
324 static int
325 sdactivate(struct device *self, enum devact act)
326 {
327 	int rv = 0;
328 
329 	switch (act) {
330 	case DVACT_ACTIVATE:
331 		rv = EOPNOTSUPP;
332 		break;
333 
334 	case DVACT_DEACTIVATE:
335 		/*
336 		 * Nothing to do; we key off the device's DVF_ACTIVE.
337 		 */
338 		break;
339 	}
340 	return (rv);
341 }
342 
343 static int
344 sddetach(struct device *self, int flags)
345 {
346 	struct sd_softc *sd = (struct sd_softc *) self;
347 	struct buf *bp;
348 	int s, bmaj, cmaj, i, mn;
349 
350 	/* locate the major number */
351 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
352 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
353 
354 	/* kill any pending restart */
355 	callout_stop(&sd->sc_callout);
356 
357 	s = splbio();
358 
359 	/* Kill off any queued buffers. */
360 	while ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
361 		bp->b_error = EIO;
362 		bp->b_flags |= B_ERROR;
363 		bp->b_resid = bp->b_bcount;
364 		biodone(bp);
365 	}
366 
367 	bufq_free(&sd->buf_queue);
368 
369 	/* Kill off any pending commands. */
370 	scsipi_kill_pending(sd->sc_periph);
371 
372 	splx(s);
373 
374 	/* Nuke the vnodes for any open instances */
375 	for (i = 0; i < MAXPARTITIONS; i++) {
376 		mn = SDMINOR(self->dv_unit, i);
377 		vdevgone(bmaj, mn, mn, VBLK);
378 		vdevgone(cmaj, mn, mn, VCHR);
379 	}
380 
381 	/* Detach from the disk list. */
382 	disk_detach(&sd->sc_dk);
383 
384 	/* Get rid of the shutdown hook. */
385 	shutdownhook_disestablish(sd->sc_sdhook);
386 
387 #if NRND > 0
388 	/* Unhook the entropy source. */
389 	rnd_detach_source(&sd->rnd_source);
390 #endif
391 
392 	return (0);
393 }
394 
395 /*
396  * Wait interruptibly for an exclusive lock.
397  *
398  * XXX
399  * Several drivers do this; it should be abstracted and made MP-safe.
400  */
401 static int
402 sdlock(struct sd_softc *sd)
403 {
404 	int error;
405 
406 	while ((sd->flags & SDF_LOCKED) != 0) {
407 		sd->flags |= SDF_WANTED;
408 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
409 			return (error);
410 	}
411 	sd->flags |= SDF_LOCKED;
412 	return (0);
413 }
414 
415 /*
416  * Unlock and wake up any waiters.
417  */
418 static void
419 sdunlock(struct sd_softc *sd)
420 {
421 
422 	sd->flags &= ~SDF_LOCKED;
423 	if ((sd->flags & SDF_WANTED) != 0) {
424 		sd->flags &= ~SDF_WANTED;
425 		wakeup(sd);
426 	}
427 }
428 
429 /*
430  * open the device. Make sure the partition info is a up-to-date as can be.
431  */
432 static int
433 sdopen(dev_t dev, int flag, int fmt, struct proc *p)
434 {
435 	struct sd_softc *sd;
436 	struct scsipi_periph *periph;
437 	struct scsipi_adapter *adapt;
438 	int unit, part;
439 	int error;
440 
441 	unit = SDUNIT(dev);
442 	if (unit >= sd_cd.cd_ndevs)
443 		return (ENXIO);
444 	sd = sd_cd.cd_devs[unit];
445 	if (sd == NULL)
446 		return (ENXIO);
447 
448 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
449 		return (ENODEV);
450 
451 	periph = sd->sc_periph;
452 	adapt = periph->periph_channel->chan_adapter;
453 	part = SDPART(dev);
454 
455 	SC_DEBUG(periph, SCSIPI_DB1,
456 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
457 	    sd_cd.cd_ndevs, part));
458 
459 	/*
460 	 * If this is the first open of this device, add a reference
461 	 * to the adapter.
462 	 */
463 	if (sd->sc_dk.dk_openmask == 0 &&
464 	    (error = scsipi_adapter_addref(adapt)) != 0)
465 		return (error);
466 
467 	if ((error = sdlock(sd)) != 0)
468 		goto bad4;
469 
470 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
471 		/*
472 		 * If any partition is open, but the disk has been invalidated,
473 		 * disallow further opens of non-raw partition
474 		 */
475 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
476 		    (part != RAW_PART || fmt != S_IFCHR)) {
477 			error = EIO;
478 			goto bad3;
479 		}
480 	} else {
481 		int silent;
482 
483 		if (part == RAW_PART && fmt == S_IFCHR)
484 			silent = XS_CTL_SILENT;
485 		else
486 			silent = 0;
487 
488 		/* Check that it is still responding and ok. */
489 		error = scsipi_test_unit_ready(periph,
490 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
491 		    silent);
492 
493 		/*
494 		 * Start the pack spinning if necessary. Always allow the
495 		 * raw parition to be opened, for raw IOCTLs. Data transfers
496 		 * will check for SDEV_MEDIA_LOADED.
497 		 */
498 		if (error == EIO) {
499 			int error2;
500 
501 			error2 = scsipi_start(periph, SSS_START, silent);
502 			switch (error2) {
503 			case 0:
504 				error = 0;
505 				break;
506 			case EIO:
507 			case EINVAL:
508 				break;
509 			default:
510 				error = error2;
511 				break;
512 			}
513 		}
514 		if (error) {
515 			if (silent)
516 				goto out;
517 			goto bad3;
518 		}
519 
520 		periph->periph_flags |= PERIPH_OPEN;
521 
522 		if (periph->periph_flags & PERIPH_REMOVABLE) {
523 			/* Lock the pack in. */
524 			error = scsipi_prevent(periph, PR_PREVENT,
525 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
526 			    XS_CTL_IGNORE_MEDIA_CHANGE);
527 			if (error)
528 				goto bad;
529 		}
530 
531 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
532 			int param_error;
533 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
534 
535 			/*
536 			 * Load the physical device parameters.
537 			 *
538 			 * Note that if media is present but unformatted,
539 			 * we allow the open (so that it can be formatted!).
540 			 * The drive should refuse real I/O, if the media is
541 			 * unformatted.
542 			 */
543 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
544 			     == SDGP_RESULT_OFFLINE) {
545 				error = ENXIO;
546 				goto bad2;
547 			}
548 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
549 
550 			/* Load the partition info if not already loaded. */
551 			if (param_error == 0) {
552 				sdgetdisklabel(sd);
553 				SC_DEBUG(periph, SCSIPI_DB3,
554 				     ("Disklabel loaded "));
555 			}
556 		}
557 	}
558 
559 	/* Check that the partition exists. */
560 	if (part != RAW_PART &&
561 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
562 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
563 		error = ENXIO;
564 		goto bad;
565 	}
566 
567 out:	/* Insure only one open at a time. */
568 	switch (fmt) {
569 	case S_IFCHR:
570 		sd->sc_dk.dk_copenmask |= (1 << part);
571 		break;
572 	case S_IFBLK:
573 		sd->sc_dk.dk_bopenmask |= (1 << part);
574 		break;
575 	}
576 	sd->sc_dk.dk_openmask =
577 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
578 
579 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
580 	sdunlock(sd);
581 	return (0);
582 
583 bad2:
584 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
585 
586 bad:
587 	if (sd->sc_dk.dk_openmask == 0) {
588 		if (periph->periph_flags & PERIPH_REMOVABLE)
589 			scsipi_prevent(periph, PR_ALLOW,
590 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
591 			    XS_CTL_IGNORE_MEDIA_CHANGE);
592 		periph->periph_flags &= ~PERIPH_OPEN;
593 	}
594 
595 bad3:
596 	sdunlock(sd);
597 bad4:
598 	if (sd->sc_dk.dk_openmask == 0)
599 		scsipi_adapter_delref(adapt);
600 	return (error);
601 }
602 
603 /*
604  * close the device.. only called if we are the LAST occurence of an open
605  * device.  Convenient now but usually a pain.
606  */
607 static int
608 sdclose(dev_t dev, int flag, int fmt, struct proc *p)
609 {
610 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
611 	struct scsipi_periph *periph = sd->sc_periph;
612 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
613 	int part = SDPART(dev);
614 	int error;
615 
616 	if ((error = sdlock(sd)) != 0)
617 		return (error);
618 
619 	switch (fmt) {
620 	case S_IFCHR:
621 		sd->sc_dk.dk_copenmask &= ~(1 << part);
622 		break;
623 	case S_IFBLK:
624 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
625 		break;
626 	}
627 	sd->sc_dk.dk_openmask =
628 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
629 
630 	if (sd->sc_dk.dk_openmask == 0) {
631 		/*
632 		 * If the disk cache needs flushing, and the disk supports
633 		 * it, do it now.
634 		 */
635 		if ((sd->flags & SDF_DIRTY) != 0) {
636 			if (sd_flush(sd, 0)) {
637 				printf("%s: cache synchronization failed\n",
638 				    sd->sc_dev.dv_xname);
639 				sd->flags &= ~SDF_FLUSHING;
640 			} else
641 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
642 		}
643 
644 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
645 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
646 
647 		scsipi_wait_drain(periph);
648 
649 		if (periph->periph_flags & PERIPH_REMOVABLE)
650 			scsipi_prevent(periph, PR_ALLOW,
651 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
652 			    XS_CTL_IGNORE_NOT_READY);
653 		periph->periph_flags &= ~PERIPH_OPEN;
654 
655 		scsipi_wait_drain(periph);
656 
657 		scsipi_adapter_delref(adapt);
658 	}
659 
660 	sdunlock(sd);
661 	return (0);
662 }
663 
664 /*
665  * Actually translate the requested transfer into one the physical driver
666  * can understand.  The transfer is described by a buf and will include
667  * only one physical transfer.
668  */
669 static void
670 sdstrategy(struct buf *bp)
671 {
672 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
673 	struct scsipi_periph *periph = sd->sc_periph;
674 	struct disklabel *lp;
675 	daddr_t blkno;
676 	int s;
677 	boolean_t sector_aligned;
678 
679 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
680 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
681 	    ("%ld bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
682 	/*
683 	 * If the device has been made invalid, error out
684 	 */
685 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
686 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
687 		if (periph->periph_flags & PERIPH_OPEN)
688 			bp->b_error = EIO;
689 		else
690 			bp->b_error = ENODEV;
691 		goto bad;
692 	}
693 
694 	lp = sd->sc_dk.dk_label;
695 
696 	/*
697 	 * The transfer must be a whole number of blocks, offset must not be
698 	 * negative.
699 	 */
700 	if (lp->d_secsize == DEV_BSIZE) {
701 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
702 	} else {
703 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
704 	}
705 	if (!sector_aligned || bp->b_blkno < 0) {
706 		bp->b_error = EINVAL;
707 		goto bad;
708 	}
709 	/*
710 	 * If it's a null transfer, return immediatly
711 	 */
712 	if (bp->b_bcount == 0)
713 		goto done;
714 
715 	/*
716 	 * Do bounds checking, adjust transfer. if error, process.
717 	 * If end of partition, just return.
718 	 */
719 	if (SDPART(bp->b_dev) == RAW_PART) {
720 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
721 		    sd->params.disksize512) <= 0)
722 			goto done;
723 	} else {
724 		if (bounds_check_with_label(&sd->sc_dk, bp,
725 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
726 			goto done;
727 	}
728 
729 	/*
730 	 * Now convert the block number to absolute and put it in
731 	 * terms of the device's logical block size.
732 	 */
733 	if (lp->d_secsize == DEV_BSIZE)
734 		blkno = bp->b_blkno;
735 	else if (lp->d_secsize > DEV_BSIZE)
736 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
737 	else
738 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
739 
740 	if (SDPART(bp->b_dev) != RAW_PART)
741 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
742 
743 	bp->b_rawblkno = blkno;
744 
745 	s = splbio();
746 
747 	/*
748 	 * Place it in the queue of disk activities for this disk.
749 	 *
750 	 * XXX Only do disksort() if the current operating mode does not
751 	 * XXX include tagged queueing.
752 	 */
753 	BUFQ_PUT(&sd->buf_queue, bp);
754 
755 	/*
756 	 * Tell the device to get going on the transfer if it's
757 	 * not doing anything, otherwise just wait for completion
758 	 */
759 	sdstart(sd->sc_periph);
760 
761 	splx(s);
762 	return;
763 
764 bad:
765 	bp->b_flags |= B_ERROR;
766 done:
767 	/*
768 	 * Correctly set the buf to indicate a completed xfer
769 	 */
770 	bp->b_resid = bp->b_bcount;
771 	biodone(bp);
772 }
773 
774 /*
775  * sdstart looks to see if there is a buf waiting for the device
776  * and that the device is not already busy. If both are true,
777  * It dequeues the buf and creates a scsi command to perform the
778  * transfer in the buf. The transfer request will call scsipi_done
779  * on completion, which will in turn call this routine again
780  * so that the next queued transfer is performed.
781  * The bufs are queued by the strategy routine (sdstrategy)
782  *
783  * This routine is also called after other non-queued requests
784  * have been made of the scsi driver, to ensure that the queue
785  * continues to be drained.
786  *
787  * must be called at the correct (highish) spl level
788  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
789  */
790 static void
791 sdstart(struct scsipi_periph *periph)
792 {
793 	struct sd_softc *sd = (void *)periph->periph_dev;
794 	struct disklabel *lp = sd->sc_dk.dk_label;
795 	struct buf *bp = 0;
796 	struct scsipi_rw_big cmd_big;
797 	struct scsi_rw cmd_small;
798 	struct scsipi_generic *cmdp;
799 	struct scsipi_xfer *xs;
800 	int nblks, cmdlen, error, flags;
801 
802 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
803 	/*
804 	 * Check if the device has room for another command
805 	 */
806 	while (periph->periph_active < periph->periph_openings) {
807 		/*
808 		 * there is excess capacity, but a special waits
809 		 * It'll need the adapter as soon as we clear out of the
810 		 * way and let it run (user level wait).
811 		 */
812 		if (periph->periph_flags & PERIPH_WAITING) {
813 			periph->periph_flags &= ~PERIPH_WAITING;
814 			wakeup((caddr_t)periph);
815 			return;
816 		}
817 
818 		/*
819 		 * If the device has become invalid, abort all the
820 		 * reads and writes until all files have been closed and
821 		 * re-opened
822 		 */
823 		if (__predict_false(
824 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
825 			if ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
826 				bp->b_error = EIO;
827 				bp->b_flags |= B_ERROR;
828 				bp->b_resid = bp->b_bcount;
829 				biodone(bp);
830 				continue;
831 			} else {
832 				return;
833 			}
834 		}
835 
836 		/*
837 		 * See if there is a buf with work for us to do..
838 		 */
839 		if ((bp = BUFQ_PEEK(&sd->buf_queue)) == NULL)
840 			return;
841 
842 		/*
843 		 * We have a buf, now we should make a command.
844 		 */
845 
846 		if (lp->d_secsize == DEV_BSIZE)
847 			nblks = bp->b_bcount >> DEV_BSHIFT;
848 		else
849 			nblks = howmany(bp->b_bcount, lp->d_secsize);
850 
851 		/*
852 		 *  Fill out the scsi command.  If the transfer will
853 		 *  fit in a "small" cdb, use it.
854 		 */
855 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
856 		    ((nblks & 0xff) == nblks) &&
857 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
858 			/*
859 			 * We can fit in a small cdb.
860 			 */
861 			memset(&cmd_small, 0, sizeof(cmd_small));
862 			cmd_small.opcode = (bp->b_flags & B_READ) ?
863 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
864 			_lto3b(bp->b_rawblkno, cmd_small.addr);
865 			cmd_small.length = nblks & 0xff;
866 			cmdlen = sizeof(cmd_small);
867 			cmdp = (struct scsipi_generic *)&cmd_small;
868 		} else {
869 			/*
870 			 * Need a large cdb.
871 			 */
872 			memset(&cmd_big, 0, sizeof(cmd_big));
873 			cmd_big.opcode = (bp->b_flags & B_READ) ?
874 			    READ_BIG : WRITE_BIG;
875 			_lto4b(bp->b_rawblkno, cmd_big.addr);
876 			_lto2b(nblks, cmd_big.length);
877 			cmdlen = sizeof(cmd_big);
878 			cmdp = (struct scsipi_generic *)&cmd_big;
879 		}
880 
881 		/* Instrumentation. */
882 		disk_busy(&sd->sc_dk);
883 
884 		/*
885 		 * Mark the disk dirty so that the cache will be
886 		 * flushed on close.
887 		 */
888 		if ((bp->b_flags & B_READ) == 0)
889 			sd->flags |= SDF_DIRTY;
890 
891 		/*
892 		 * Figure out what flags to use.
893 		 */
894 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
895 		if (bp->b_flags & B_READ)
896 			flags |= XS_CTL_DATA_IN;
897 		else
898 			flags |= XS_CTL_DATA_OUT;
899 
900 		/*
901 		 * Call the routine that chats with the adapter.
902 		 * Note: we cannot sleep as we may be an interrupt
903 		 */
904 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
905 		    (u_char *)bp->b_data, bp->b_bcount,
906 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
907 		if (__predict_false(xs == NULL)) {
908 			/*
909 			 * out of memory. Keep this buffer in the queue, and
910 			 * retry later.
911 			 */
912 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
913 			    periph);
914 			return;
915 		}
916 		/*
917 		 * need to dequeue the buffer before queuing the command,
918 		 * because cdstart may be called recursively from the
919 		 * HBA driver
920 		 */
921 #ifdef DIAGNOSTIC
922 		if (BUFQ_GET(&sd->buf_queue) != bp)
923 			panic("sdstart(): dequeued wrong buf");
924 #else
925 		BUFQ_GET(&sd->buf_queue);
926 #endif
927 		error = scsipi_command(periph, xs, cmdp, cmdlen,
928 		    (u_char *)bp->b_data, bp->b_bcount,
929 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
930 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
931 		KASSERT(error == 0);
932 	}
933 }
934 
935 static void
936 sdrestart(void *v)
937 {
938 	int s = splbio();
939 	sdstart((struct scsipi_periph *)v);
940 	splx(s);
941 }
942 
943 static void
944 sddone(struct scsipi_xfer *xs)
945 {
946 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
947 
948 	if (sd->flags & SDF_FLUSHING) {
949 		/* Flush completed, no longer dirty. */
950 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
951 	}
952 
953 	if (xs->bp != NULL) {
954 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid,
955 		    (xs->bp->b_flags & B_READ));
956 #if NRND > 0
957 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
958 #endif
959 	}
960 }
961 
962 static void
963 sdminphys(struct buf *bp)
964 {
965 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
966 	long max;
967 
968 	/*
969 	 * If the device is ancient, we want to make sure that
970 	 * the transfer fits into a 6-byte cdb.
971 	 *
972 	 * XXX Note that the SCSI-I spec says that 256-block transfers
973 	 * are allowed in a 6-byte read/write, and are specified
974 	 * by settng the "length" to 0.  However, we're conservative
975 	 * here, allowing only 255-block transfers in case an
976 	 * ancient device gets confused by length == 0.  A length of 0
977 	 * in a 10-byte read/write actually means 0 blocks.
978 	 */
979 	if ((sd->flags & SDF_ANCIENT) &&
980 	    ((sd->sc_periph->periph_flags &
981 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
982 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
983 
984 		if (bp->b_bcount > max)
985 			bp->b_bcount = max;
986 	}
987 
988 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
989 }
990 
991 static int
992 sdread(dev_t dev, struct uio *uio, int ioflag)
993 {
994 
995 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
996 }
997 
998 static int
999 sdwrite(dev_t dev, struct uio *uio, int ioflag)
1000 {
1001 
1002 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
1003 }
1004 
1005 /*
1006  * Perform special action on behalf of the user
1007  * Knows about the internals of this device
1008  */
1009 static int
1010 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct proc *p)
1011 {
1012 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1013 	struct scsipi_periph *periph = sd->sc_periph;
1014 	int part = SDPART(dev);
1015 	int error = 0;
1016 #ifdef __HAVE_OLD_DISKLABEL
1017 	struct disklabel *newlabel = NULL;
1018 #endif
1019 
1020 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1021 
1022 	/*
1023 	 * If the device is not valid, some IOCTLs can still be
1024 	 * handled on the raw partition. Check this here.
1025 	 */
1026 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1027 		switch (cmd) {
1028 		case DIOCKLABEL:
1029 		case DIOCWLABEL:
1030 		case DIOCLOCK:
1031 		case DIOCEJECT:
1032 		case ODIOCEJECT:
1033 		case DIOCGCACHE:
1034 		case DIOCSCACHE:
1035 		case SCIOCIDENTIFY:
1036 		case OSCIOCIDENTIFY:
1037 		case SCIOCCOMMAND:
1038 		case SCIOCDEBUG:
1039 			if (part == RAW_PART)
1040 				break;
1041 		/* FALLTHROUGH */
1042 		default:
1043 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1044 				return (ENODEV);
1045 			else
1046 				return (EIO);
1047 		}
1048 	}
1049 
1050 	switch (cmd) {
1051 	case DIOCGDINFO:
1052 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1053 		return (0);
1054 
1055 #ifdef __HAVE_OLD_DISKLABEL
1056 	case ODIOCGDINFO:
1057 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1058 		if (newlabel == NULL)
1059 			return EIO;
1060 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1061 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1062 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1063 		else
1064 			error = ENOTTY;
1065 		free(newlabel, M_TEMP);
1066 		return error;
1067 #endif
1068 
1069 	case DIOCGPART:
1070 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1071 		((struct partinfo *)addr)->part =
1072 		    &sd->sc_dk.dk_label->d_partitions[part];
1073 		return (0);
1074 
1075 	case DIOCWDINFO:
1076 	case DIOCSDINFO:
1077 #ifdef __HAVE_OLD_DISKLABEL
1078 	case ODIOCWDINFO:
1079 	case ODIOCSDINFO:
1080 #endif
1081 	{
1082 		struct disklabel *lp;
1083 
1084 		if ((flag & FWRITE) == 0)
1085 			return (EBADF);
1086 
1087 #ifdef __HAVE_OLD_DISKLABEL
1088  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1089 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1090 			if (newlabel == NULL)
1091 				return EIO;
1092 			memset(newlabel, 0, sizeof newlabel);
1093 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1094 			lp = newlabel;
1095 		} else
1096 #endif
1097 		lp = (struct disklabel *)addr;
1098 
1099 		if ((error = sdlock(sd)) != 0)
1100 			goto bad;
1101 		sd->flags |= SDF_LABELLING;
1102 
1103 		error = setdisklabel(sd->sc_dk.dk_label,
1104 		    lp, /*sd->sc_dk.dk_openmask : */0,
1105 		    sd->sc_dk.dk_cpulabel);
1106 		if (error == 0) {
1107 			if (cmd == DIOCWDINFO
1108 #ifdef __HAVE_OLD_DISKLABEL
1109 			    || cmd == ODIOCWDINFO
1110 #endif
1111 			   )
1112 				error = writedisklabel(SDLABELDEV(dev),
1113 				    sdstrategy, sd->sc_dk.dk_label,
1114 				    sd->sc_dk.dk_cpulabel);
1115 		}
1116 
1117 		sd->flags &= ~SDF_LABELLING;
1118 		sdunlock(sd);
1119 bad:
1120 #ifdef __HAVE_OLD_DISKLABEL
1121 		if (newlabel != NULL)
1122 			free(newlabel, M_TEMP);
1123 #endif
1124 		return (error);
1125 	}
1126 
1127 	case DIOCKLABEL:
1128 		if (*(int *)addr)
1129 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1130 		else
1131 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1132 		return (0);
1133 
1134 	case DIOCWLABEL:
1135 		if ((flag & FWRITE) == 0)
1136 			return (EBADF);
1137 		if (*(int *)addr)
1138 			sd->flags |= SDF_WLABEL;
1139 		else
1140 			sd->flags &= ~SDF_WLABEL;
1141 		return (0);
1142 
1143 	case DIOCLOCK:
1144 		return (scsipi_prevent(periph,
1145 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1146 
1147 	case DIOCEJECT:
1148 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1149 			return (ENOTTY);
1150 		if (*(int *)addr == 0) {
1151 			/*
1152 			 * Don't force eject: check that we are the only
1153 			 * partition open. If so, unlock it.
1154 			 */
1155 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1156 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1157 			    sd->sc_dk.dk_openmask) {
1158 				error = scsipi_prevent(periph, PR_ALLOW,
1159 				    XS_CTL_IGNORE_NOT_READY);
1160 				if (error)
1161 					return (error);
1162 			} else {
1163 				return (EBUSY);
1164 			}
1165 		}
1166 		/* FALLTHROUGH */
1167 	case ODIOCEJECT:
1168 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1169 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1170 
1171 	case DIOCGDEFLABEL:
1172 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1173 		return (0);
1174 
1175 #ifdef __HAVE_OLD_DISKLABEL
1176 	case ODIOCGDEFLABEL:
1177 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1178 		if (newlabel == NULL)
1179 			return EIO;
1180 		sdgetdefaultlabel(sd, newlabel);
1181 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1182 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1183 		else
1184 			error = ENOTTY;
1185 		free(newlabel, M_TEMP);
1186 		return error;
1187 #endif
1188 
1189 	case DIOCGCACHE:
1190 		return (sd_getcache(sd, (int *) addr));
1191 
1192 	case DIOCSCACHE:
1193 		if ((flag & FWRITE) == 0)
1194 			return (EBADF);
1195 		return (sd_setcache(sd, *(int *) addr));
1196 
1197 	case DIOCCACHESYNC:
1198 		/*
1199 		 * XXX Do we really need to care about having a writable
1200 		 * file descriptor here?
1201 		 */
1202 		if ((flag & FWRITE) == 0)
1203 			return (EBADF);
1204 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1205 			error = sd_flush(sd, 0);
1206 			if (error)
1207 				sd->flags &= ~SDF_FLUSHING;
1208 			else
1209 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1210 		} else
1211 			error = 0;
1212 		return (error);
1213 
1214 	default:
1215 		if (part != RAW_PART)
1216 			return (ENOTTY);
1217 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1218 	}
1219 
1220 #ifdef DIAGNOSTIC
1221 	panic("sdioctl: impossible");
1222 #endif
1223 }
1224 
1225 static void
1226 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1227 {
1228 
1229 	memset(lp, 0, sizeof(struct disklabel));
1230 
1231 	lp->d_secsize = sd->params.blksize;
1232 	lp->d_ntracks = sd->params.heads;
1233 	lp->d_nsectors = sd->params.sectors;
1234 	lp->d_ncylinders = sd->params.cyls;
1235 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1236 
1237 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1238 	case SCSIPI_BUSTYPE_SCSI:
1239 		lp->d_type = DTYPE_SCSI;
1240 		break;
1241 	case SCSIPI_BUSTYPE_ATAPI:
1242 		lp->d_type = DTYPE_ATAPI;
1243 		break;
1244 	}
1245 	/*
1246 	 * XXX
1247 	 * We could probe the mode pages to figure out what kind of disc it is.
1248 	 * Is this worthwhile?
1249 	 */
1250 	strncpy(lp->d_typename, "mydisk", 16);
1251 	strncpy(lp->d_packname, "fictitious", 16);
1252 	lp->d_secperunit = sd->params.disksize;
1253 	lp->d_rpm = sd->params.rot_rate;
1254 	lp->d_interleave = 1;
1255 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1256 	    D_REMOVABLE : 0;
1257 
1258 	lp->d_partitions[RAW_PART].p_offset = 0;
1259 	lp->d_partitions[RAW_PART].p_size =
1260 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1261 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1262 	lp->d_npartitions = RAW_PART + 1;
1263 
1264 	lp->d_magic = DISKMAGIC;
1265 	lp->d_magic2 = DISKMAGIC;
1266 	lp->d_checksum = dkcksum(lp);
1267 }
1268 
1269 
1270 /*
1271  * Load the label information on the named device
1272  */
1273 static void
1274 sdgetdisklabel(struct sd_softc *sd)
1275 {
1276 	struct disklabel *lp = sd->sc_dk.dk_label;
1277 	const char *errstring;
1278 
1279 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1280 
1281 	sdgetdefaultlabel(sd, lp);
1282 
1283 	if (lp->d_secpercyl == 0) {
1284 		lp->d_secpercyl = 100;
1285 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1286 	}
1287 
1288 	/*
1289 	 * Call the generic disklabel extraction routine
1290 	 */
1291 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1292 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1293 	if (errstring) {
1294 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1295 		return;
1296 	}
1297 }
1298 
1299 static void
1300 sd_shutdown(void *arg)
1301 {
1302 	struct sd_softc *sd = arg;
1303 
1304 	/*
1305 	 * If the disk cache needs to be flushed, and the disk supports
1306 	 * it, flush it.  We're cold at this point, so we poll for
1307 	 * completion.
1308 	 */
1309 	if ((sd->flags & SDF_DIRTY) != 0) {
1310 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1311 			printf("%s: cache synchronization failed\n",
1312 			    sd->sc_dev.dv_xname);
1313 			sd->flags &= ~SDF_FLUSHING;
1314 		} else
1315 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1316 	}
1317 }
1318 
1319 /*
1320  * Check Errors
1321  */
1322 static int
1323 sd_interpret_sense(struct scsipi_xfer *xs)
1324 {
1325 	struct scsipi_periph *periph = xs->xs_periph;
1326 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1327 	struct sd_softc *sd = (void *)periph->periph_dev;
1328 	int s, error, retval = EJUSTRETURN;
1329 
1330 	/*
1331 	 * If the periph is already recovering, just do the normal
1332 	 * error processing.
1333 	 */
1334 	if (periph->periph_flags & PERIPH_RECOVERING)
1335 		return (retval);
1336 
1337 	/*
1338 	 * If the device is not open yet, let the generic code handle it.
1339 	 */
1340 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1341 		return (retval);
1342 
1343 	/*
1344 	 * If it isn't a extended or extended/deferred error, let
1345 	 * the generic code handle it.
1346 	 */
1347 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1348 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1349 		return (retval);
1350 
1351 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1352 	    sense->add_sense_code == 0x4) {
1353 		if (sense->add_sense_code_qual == 0x01)	{
1354 			/*
1355 			 * Unit In The Process Of Becoming Ready.
1356 			 */
1357 			printf("%s: waiting for pack to spin up...\n",
1358 			    sd->sc_dev.dv_xname);
1359 			if (!callout_pending(&periph->periph_callout))
1360 				scsipi_periph_freeze(periph, 1);
1361 			callout_reset(&periph->periph_callout,
1362 			    5 * hz, scsipi_periph_timed_thaw, periph);
1363 			retval = ERESTART;
1364 		} else if (sense->add_sense_code_qual == 0x02) {
1365 			printf("%s: pack is stopped, restarting...\n",
1366 			    sd->sc_dev.dv_xname);
1367 			s = splbio();
1368 			periph->periph_flags |= PERIPH_RECOVERING;
1369 			splx(s);
1370 			error = scsipi_start(periph, SSS_START,
1371 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1372 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1373 			if (error) {
1374 				printf("%s: unable to restart pack\n",
1375 				    sd->sc_dev.dv_xname);
1376 				retval = error;
1377 			} else
1378 				retval = ERESTART;
1379 			s = splbio();
1380 			periph->periph_flags &= ~PERIPH_RECOVERING;
1381 			splx(s);
1382 		}
1383 	}
1384 	if ((sense->flags & SSD_KEY) == SKEY_MEDIUM_ERROR &&
1385 	    sense->add_sense_code == 0x31 &&
1386 	    sense->add_sense_code_qual == 0x00)	{ /* maybe for any asq ? */
1387 		/* Medium Format Corrupted */
1388 		retval = EFTYPE;
1389 	}
1390 	return (retval);
1391 }
1392 
1393 
1394 static int
1395 sdsize(dev_t dev)
1396 {
1397 	struct sd_softc *sd;
1398 	int part, unit, omask;
1399 	int size;
1400 
1401 	unit = SDUNIT(dev);
1402 	if (unit >= sd_cd.cd_ndevs)
1403 		return (-1);
1404 	sd = sd_cd.cd_devs[unit];
1405 	if (sd == NULL)
1406 		return (-1);
1407 
1408 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1409 		return (-1);
1410 
1411 	part = SDPART(dev);
1412 	omask = sd->sc_dk.dk_openmask & (1 << part);
1413 
1414 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1415 		return (-1);
1416 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1417 		size = -1;
1418 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1419 		size = -1;
1420 	else
1421 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1422 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1423 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1424 		return (-1);
1425 	return (size);
1426 }
1427 
1428 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1429 static struct scsipi_xfer sx;
1430 static int sddoingadump;
1431 
1432 /*
1433  * dump all of physical memory into the partition specified, starting
1434  * at offset 'dumplo' into the partition.
1435  */
1436 static int
1437 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1438 {
1439 	struct sd_softc *sd;	/* disk unit to do the I/O */
1440 	struct disklabel *lp;	/* disk's disklabel */
1441 	int	unit, part;
1442 	int	sectorsize;	/* size of a disk sector */
1443 	int	nsects;		/* number of sectors in partition */
1444 	int	sectoff;	/* sector offset of partition */
1445 	int	totwrt;		/* total number of sectors left to write */
1446 	int	nwrt;		/* current number of sectors to write */
1447 	struct scsipi_rw_big cmd;	/* write command */
1448 	struct scsipi_xfer *xs;	/* ... convenience */
1449 	struct scsipi_periph *periph;
1450 	struct scsipi_channel *chan;
1451 
1452 	/* Check if recursive dump; if so, punt. */
1453 	if (sddoingadump)
1454 		return (EFAULT);
1455 
1456 	/* Mark as active early. */
1457 	sddoingadump = 1;
1458 
1459 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1460 	part = SDPART(dev);
1461 
1462 	/* Check for acceptable drive number. */
1463 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1464 		return (ENXIO);
1465 
1466 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1467 		return (ENODEV);
1468 
1469 	periph = sd->sc_periph;
1470 	chan = periph->periph_channel;
1471 
1472 	/* Make sure it was initialized. */
1473 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1474 		return (ENXIO);
1475 
1476 	/* Convert to disk sectors.  Request must be a multiple of size. */
1477 	lp = sd->sc_dk.dk_label;
1478 	sectorsize = lp->d_secsize;
1479 	if ((size % sectorsize) != 0)
1480 		return (EFAULT);
1481 	totwrt = size / sectorsize;
1482 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1483 
1484 	nsects = lp->d_partitions[part].p_size;
1485 	sectoff = lp->d_partitions[part].p_offset;
1486 
1487 	/* Check transfer bounds against partition size. */
1488 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1489 		return (EINVAL);
1490 
1491 	/* Offset block number to start of partition. */
1492 	blkno += sectoff;
1493 
1494 	xs = &sx;
1495 
1496 	while (totwrt > 0) {
1497 		nwrt = totwrt;		/* XXX */
1498 #ifndef	SD_DUMP_NOT_TRUSTED
1499 		/*
1500 		 *  Fill out the scsi command
1501 		 */
1502 		memset(&cmd, 0, sizeof(cmd));
1503 		cmd.opcode = WRITE_BIG;
1504 		_lto4b(blkno, cmd.addr);
1505 		_lto2b(nwrt, cmd.length);
1506 		/*
1507 		 * Fill out the scsipi_xfer structure
1508 		 *    Note: we cannot sleep as we may be an interrupt
1509 		 * don't use scsipi_command() as it may want to wait
1510 		 * for an xs.
1511 		 */
1512 		memset(xs, 0, sizeof(sx));
1513 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1514 		    XS_CTL_DATA_OUT;
1515 		xs->xs_status = 0;
1516 		xs->xs_periph = periph;
1517 		xs->xs_retries = SDRETRIES;
1518 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1519 		xs->cmd = (struct scsipi_generic *)&cmd;
1520 		xs->cmdlen = sizeof(cmd);
1521 		xs->resid = nwrt * sectorsize;
1522 		xs->error = XS_NOERROR;
1523 		xs->bp = 0;
1524 		xs->data = va;
1525 		xs->datalen = nwrt * sectorsize;
1526 
1527 		/*
1528 		 * Pass all this info to the scsi driver.
1529 		 */
1530 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1531 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1532 		    xs->error != XS_NOERROR)
1533 			return (EIO);
1534 #else	/* SD_DUMP_NOT_TRUSTED */
1535 		/* Let's just talk about this first... */
1536 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1537 		delay(500 * 1000);	/* half a second */
1538 #endif	/* SD_DUMP_NOT_TRUSTED */
1539 
1540 		/* update block count */
1541 		totwrt -= nwrt;
1542 		blkno += nwrt;
1543 		va += sectorsize * nwrt;
1544 	}
1545 	sddoingadump = 0;
1546 	return (0);
1547 }
1548 
1549 static int
1550 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1551     int page, int flags, int *big)
1552 {
1553 
1554 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1555 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1556 		*big = 1;
1557 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1558 		    size + sizeof(struct scsipi_mode_header_big),
1559 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1560 	} else {
1561 		*big = 0;
1562 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1563 		    size + sizeof(struct scsipi_mode_header),
1564 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1565 	}
1566 }
1567 
1568 static int
1569 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1570     int flags, int big)
1571 {
1572 
1573 	if (big) {
1574 		struct scsipi_mode_header_big *header = sense;
1575 
1576 		_lto2b(0, header->data_length);
1577 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1578 		    size + sizeof(struct scsipi_mode_header_big),
1579 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1580 	} else {
1581 		struct scsipi_mode_header *header = sense;
1582 
1583 		header->data_length = 0;
1584 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1585 		    size + sizeof(struct scsipi_mode_header),
1586 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1587 	}
1588 }
1589 
1590 static int
1591 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1592 {
1593 	struct {
1594 		struct scsipi_mode_header header;
1595 		/* no block descriptor */
1596 		u_int8_t pg_code; /* page code (should be 6) */
1597 		u_int8_t pg_length; /* page length (should be 11) */
1598 		u_int8_t wcd; /* bit0: cache disable */
1599 		u_int8_t lbs[2]; /* logical block size */
1600 		u_int8_t size[5]; /* number of log. blocks */
1601 		u_int8_t pp; /* power/performance */
1602 		u_int8_t flags;
1603 		u_int8_t resvd;
1604 	} scsipi_sense;
1605 	u_int64_t sectors;
1606 	int error;
1607 
1608 	/*
1609 	 * scsipi_size (ie "read capacity") and mode sense page 6
1610 	 * give the same information. Do both for now, and check
1611 	 * for consistency.
1612 	 * XXX probably differs for removable media
1613 	 */
1614 	dp->blksize = 512;
1615 	if ((sectors = scsipi_size(sd->sc_periph, flags)) == 0)
1616 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1617 
1618 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1619 	    &scsipi_sense.header, sizeof(scsipi_sense),
1620 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1621 
1622 	if (error != 0)
1623 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1624 
1625 	dp->blksize = _2btol(scsipi_sense.lbs);
1626 	if (dp->blksize == 0)
1627 		dp->blksize = 512;
1628 
1629 	/*
1630 	 * Create a pseudo-geometry.
1631 	 */
1632 	dp->heads = 64;
1633 	dp->sectors = 32;
1634 	dp->cyls = sectors / (dp->heads * dp->sectors);
1635 	dp->disksize = _5btol(scsipi_sense.size);
1636 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1637 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1638 		       (unsigned long long)dp->disksize,
1639 		       (unsigned long long)sectors);
1640 		dp->disksize = sectors;
1641 	}
1642 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1643 
1644 	return (SDGP_RESULT_OK);
1645 }
1646 
1647 /*
1648  * Get the scsi driver to send a full inquiry to the * device and use the
1649  * results to fill out the disk parameter structure.
1650  */
1651 static int
1652 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1653 {
1654 	u_int64_t sectors;
1655 	int error;
1656 #if 0
1657 	int i;
1658 	u_int8_t *p;
1659 #endif
1660 
1661 	dp->disksize = sectors = scsipi_size(sd->sc_periph, flags);
1662 	if (sectors == 0) {
1663 		struct scsipi_read_format_capacities cmd;
1664 		struct {
1665 			struct scsipi_capacity_list_header header;
1666 			struct scsipi_capacity_descriptor desc;
1667 		} __attribute__((packed)) data;
1668 
1669 		memset(&cmd, 0, sizeof(cmd));
1670 		memset(&data, 0, sizeof(data));
1671 		cmd.opcode = READ_FORMAT_CAPACITIES;
1672 		_lto2b(sizeof(data), cmd.length);
1673 
1674 		error = scsipi_command(sd->sc_periph, NULL,
1675 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1676 		    SDRETRIES, 20000, NULL,
1677 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1678 		if (error == EFTYPE) {
1679 			/* Medium Format Corrupted, handle as not formatted */
1680 			return (SDGP_RESULT_UNFORMATTED);
1681 		}
1682 		if (error || data.header.length == 0)
1683 			return (SDGP_RESULT_OFFLINE);
1684 
1685 #if 0
1686 printf("rfc: length=%d\n", data.header.length);
1687 printf("rfc result:"); for (i = sizeof(struct scsipi_capacity_list_header) + data.header.length, p = (void *)&data; i; i--, p++) printf(" %02x", *p); printf("\n");
1688 #endif
1689 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1690 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1691 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1692 			break;
1693 
1694 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1695 			return (SDGP_RESULT_UNFORMATTED);
1696 
1697 		case SCSIPI_CAP_DESC_CODE_NONE:
1698 			return (SDGP_RESULT_OFFLINE);
1699 		}
1700 
1701 		dp->disksize = sectors = _4btol(data.desc.nblks);
1702 		if (sectors == 0)
1703 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1704 
1705 		dp->blksize = _3btol(data.desc.blklen);
1706 		if (dp->blksize == 0)
1707 			dp->blksize = 512;
1708 	} else {
1709 		struct sd_mode_sense_data scsipi_sense;
1710 		int big, bsize;
1711 		struct scsi_blk_desc *bdesc;
1712 
1713 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1714 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1715 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1716 		dp->blksize = 512;
1717 		if (!error) {
1718 			if (big) {
1719 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1720 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1721 			} else {
1722 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1723 				bsize = scsipi_sense.header.small.blk_desc_len;
1724 			}
1725 
1726 #if 0
1727 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1728 printf("page 0 bsize=%d\n", bsize);
1729 printf("page 0 ok\n");
1730 #endif
1731 
1732 			if (bsize >= 8) {
1733 				dp->blksize = _3btol(bdesc->blklen);
1734 				if (dp->blksize == 0)
1735 					dp->blksize = 512;
1736 			}
1737 		}
1738 	}
1739 
1740 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1741 	return (0);
1742 }
1743 
1744 static int
1745 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1746 {
1747 	struct sd_mode_sense_data scsipi_sense;
1748 	int error;
1749 	int big, poffset, byte2;
1750 	union scsi_disk_pages *pages;
1751 #if 0
1752 	int i;
1753 	u_int8_t *p;
1754 #endif
1755 
1756 	byte2 = SMS_DBD;
1757 again:
1758 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1759 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1760 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1761 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1762 	    flags | XS_CTL_SILENT, &big);
1763 	if (error) {
1764 		if (byte2 == SMS_DBD) {
1765 			/* No result; try once more with DBD off */
1766 			byte2 = 0;
1767 			goto again;
1768 		}
1769 		return (error);
1770 	}
1771 
1772 	if (big) {
1773 		poffset = sizeof scsipi_sense.header.big;
1774 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1775 	} else {
1776 		poffset = sizeof scsipi_sense.header.small;
1777 		poffset += scsipi_sense.header.small.blk_desc_len;
1778 	}
1779 
1780 	pages = (void *)((u_long)&scsipi_sense + poffset);
1781 #if 0
1782 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1783 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1784 #endif
1785 
1786 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1787 		return (ERESTART);
1788 
1789 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1790 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1791 	    _3btol(pages->rigid_geometry.ncyl),
1792 	    pages->rigid_geometry.nheads,
1793 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1794 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1795 	    _2btol(pages->rigid_geometry.land_zone)));
1796 
1797 	/*
1798 	 * KLUDGE!! (for zone recorded disks)
1799 	 * give a number of sectors so that sec * trks * cyls
1800 	 * is <= disk_size
1801 	 * can lead to wasted space! THINK ABOUT THIS !
1802 	 */
1803 	dp->heads = pages->rigid_geometry.nheads;
1804 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1805 	if (dp->heads == 0 || dp->cyls == 0)
1806 		return (ERESTART);
1807 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1808 
1809 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1810 	if (dp->rot_rate == 0)
1811 		dp->rot_rate = 3600;
1812 
1813 #if 0
1814 printf("page 4 ok\n");
1815 #endif
1816 	return (0);
1817 }
1818 
1819 static int
1820 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1821 {
1822 	struct sd_mode_sense_data scsipi_sense;
1823 	int error;
1824 	int big, poffset, byte2;
1825 	union scsi_disk_pages *pages;
1826 #if 0
1827 	int i;
1828 	u_int8_t *p;
1829 #endif
1830 
1831 	byte2 = SMS_DBD;
1832 again:
1833 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1834 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1835 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1836 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1837 	    flags | XS_CTL_SILENT, &big);
1838 	if (error) {
1839 		if (byte2 == SMS_DBD) {
1840 			/* No result; try once more with DBD off */
1841 			byte2 = 0;
1842 			goto again;
1843 		}
1844 		return (error);
1845 	}
1846 
1847 	if (big) {
1848 		poffset = sizeof scsipi_sense.header.big;
1849 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1850 	} else {
1851 		poffset = sizeof scsipi_sense.header.small;
1852 		poffset += scsipi_sense.header.small.blk_desc_len;
1853 	}
1854 
1855 	pages = (void *)((u_long)&scsipi_sense + poffset);
1856 #if 0
1857 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1858 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1859 #endif
1860 
1861 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1862 		return (ERESTART);
1863 
1864 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1865 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1866 	    _3btol(pages->flex_geometry.ncyl),
1867 	    pages->flex_geometry.nheads,
1868 	    pages->flex_geometry.ph_sec_tr,
1869 	    _2btol(pages->flex_geometry.bytes_s)));
1870 
1871 	dp->heads = pages->flex_geometry.nheads;
1872 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1873 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1874 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1875 		return (ERESTART);
1876 
1877 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1878 	if (dp->rot_rate == 0)
1879 		dp->rot_rate = 3600;
1880 
1881 #if 0
1882 printf("page 5 ok\n");
1883 #endif
1884 	return (0);
1885 }
1886 
1887 static int
1888 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1889 {
1890 	int error;
1891 
1892 	/*
1893 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1894 	 * cleared by the caller if necessary.
1895 	 */
1896 	if (sd->type == T_SIMPLE_DIRECT)
1897 		return (sd_get_simplifiedparms(sd, dp, flags));
1898 
1899 	error = sd_get_capacity(sd, dp, flags);
1900 	if (error)
1901 		return (error);
1902 
1903 	if (sd->type == T_OPTICAL)
1904 		goto page0;
1905 
1906 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1907 		if (!sd_get_parms_page5(sd, dp, flags) ||
1908 		    !sd_get_parms_page4(sd, dp, flags))
1909 			return (SDGP_RESULT_OK);
1910 	} else {
1911 		if (!sd_get_parms_page4(sd, dp, flags) ||
1912 		    !sd_get_parms_page5(sd, dp, flags))
1913 			return (SDGP_RESULT_OK);
1914 	}
1915 
1916 page0:
1917 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1918 	/* Try calling driver's method for figuring out geometry. */
1919 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1920 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1921 		(sd->sc_periph, dp, dp->disksize)) {
1922 		/*
1923 		 * Use adaptec standard fictitious geometry
1924 		 * this depends on which controller (e.g. 1542C is
1925 		 * different. but we have to put SOMETHING here..)
1926 		 */
1927 		dp->heads = 64;
1928 		dp->sectors = 32;
1929 		dp->cyls = dp->disksize / (64 * 32);
1930 	}
1931 	dp->rot_rate = 3600;
1932 	return (SDGP_RESULT_OK);
1933 }
1934 
1935 static int
1936 sd_flush(struct sd_softc *sd, int flags)
1937 {
1938 	struct scsipi_periph *periph = sd->sc_periph;
1939 	struct scsi_synchronize_cache cmd;
1940 
1941 	/*
1942 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1943 	 * We issue with address 0 length 0, which should be
1944 	 * interpreted by the device as "all remaining blocks
1945 	 * starting at address 0".  We ignore ILLEGAL REQUEST
1946 	 * in the event that the command is not supported by
1947 	 * the device, and poll for completion so that we know
1948 	 * that the cache has actually been flushed.
1949 	 *
1950 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
1951 	 * command, as indicated by our quirks flags.
1952 	 *
1953 	 * XXX What about older devices?
1954 	 */
1955 	if (periph->periph_version < 2 ||
1956 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
1957 		return (0);
1958 
1959 	sd->flags |= SDF_FLUSHING;
1960 	memset(&cmd, 0, sizeof(cmd));
1961 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE;
1962 
1963 	return (scsipi_command(periph, NULL, (void *)&cmd, sizeof(cmd), 0, 0,
1964 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
1965 }
1966 
1967 static int
1968 sd_getcache(struct sd_softc *sd, int *bitsp)
1969 {
1970 	struct scsipi_periph *periph = sd->sc_periph;
1971 	struct sd_mode_sense_data scsipi_sense;
1972 	int error, bits = 0;
1973 	int big;
1974 	union scsi_disk_pages *pages;
1975 
1976 	if (periph->periph_version < 2)
1977 		return (EOPNOTSUPP);
1978 
1979 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1980 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
1981 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
1982 	if (error)
1983 		return (error);
1984 
1985 	if (big)
1986 		pages = (void *)(&scsipi_sense.header.big + 1);
1987 	else
1988 		pages = (void *)(&scsipi_sense.header.small + 1);
1989 
1990 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
1991 		bits |= DKCACHE_READ;
1992 	if (pages->caching_params.flags & CACHING_WCE)
1993 		bits |= DKCACHE_WRITE;
1994 	if (pages->caching_params.pg_code & PGCODE_PS)
1995 		bits |= DKCACHE_SAVE;
1996 
1997 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1998 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
1999 	    sizeof(scsipi_sense.pages.caching_params),
2000 	    SMS_PAGE_CTRL_CHANGEABLE|8, 0, &big);
2001 	if (error == 0) {
2002 		if (big)
2003 			pages = (void *)(&scsipi_sense.header.big + 1);
2004 		else
2005 			pages = (void *)(&scsipi_sense.header.small + 1);
2006 
2007 		if (pages->caching_params.flags & CACHING_RCD)
2008 			bits |= DKCACHE_RCHANGE;
2009 		if (pages->caching_params.flags & CACHING_WCE)
2010 			bits |= DKCACHE_WCHANGE;
2011 	}
2012 
2013 	*bitsp = bits;
2014 
2015 	return (0);
2016 }
2017 
2018 static int
2019 sd_setcache(struct sd_softc *sd, int bits)
2020 {
2021 	struct scsipi_periph *periph = sd->sc_periph;
2022 	struct sd_mode_sense_data scsipi_sense;
2023 	int error;
2024 	uint8_t oflags, byte2 = 0;
2025 	int big;
2026 	union scsi_disk_pages *pages;
2027 
2028 	if (periph->periph_version < 2)
2029 		return (EOPNOTSUPP);
2030 
2031 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2032 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2033 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2034 	if (error)
2035 		return (error);
2036 
2037 	if (big)
2038 		pages = (void *)(&scsipi_sense.header.big + 1);
2039 	else
2040 		pages = (void *)(&scsipi_sense.header.small + 1);
2041 
2042 	oflags = pages->caching_params.flags;
2043 
2044 	if (bits & DKCACHE_READ)
2045 		pages->caching_params.flags &= ~CACHING_RCD;
2046 	else
2047 		pages->caching_params.flags |= CACHING_RCD;
2048 
2049 	if (bits & DKCACHE_WRITE)
2050 		pages->caching_params.flags |= CACHING_WCE;
2051 	else
2052 		pages->caching_params.flags &= ~CACHING_WCE;
2053 
2054 	if (oflags == pages->caching_params.flags)
2055 		return (0);
2056 
2057 	pages->caching_params.pg_code &= PGCODE_MASK;
2058 
2059 	if (bits & DKCACHE_SAVE)
2060 		byte2 |= SMS_SP;
2061 
2062 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2063 	    sizeof(struct scsipi_mode_page_header) +
2064 	    pages->caching_params.pg_length, 0, big));
2065 }
2066