xref: /netbsd-src/sys/dev/scsipi/sd.c (revision aa73cae19608873cc4d1f712c4a0f8f8435f1ffa)
1 /*	$NetBSD: sd.c,v 1.237 2005/02/27 00:27:48 perry Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2003, 2004 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 1.237 2005/02/27 00:27:48 perry 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/bufq.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/scsi_spc.h>
85 #include <dev/scsipi/scsipi_all.h>
86 #include <dev/scsipi/scsi_all.h>
87 #include <dev/scsipi/scsipi_disk.h>
88 #include <dev/scsipi/scsi_disk.h>
89 #include <dev/scsipi/scsiconf.h>
90 #include <dev/scsipi/scsipi_base.h>
91 #include <dev/scsipi/sdvar.h>
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 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 *, int);
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, sdminphys };
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 scsi_mode_parameter_header_6 small;
186 		struct scsi_mode_parameter_header_10 big;
187 	} header;
188 	struct scsi_general_block_descriptor 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 	/* Discover wedges on this disk. */
324 	dkwedge_discover(&sd->sc_dk);
325 }
326 
327 static int
328 sdactivate(struct device *self, enum devact act)
329 {
330 	int rv = 0;
331 
332 	switch (act) {
333 	case DVACT_ACTIVATE:
334 		rv = EOPNOTSUPP;
335 		break;
336 
337 	case DVACT_DEACTIVATE:
338 		/*
339 		 * Nothing to do; we key off the device's DVF_ACTIVE.
340 		 */
341 		break;
342 	}
343 	return (rv);
344 }
345 
346 static int
347 sddetach(struct device *self, int flags)
348 {
349 	struct sd_softc *sd = (struct sd_softc *) self;
350 	struct buf *bp;
351 	int s, bmaj, cmaj, i, mn;
352 
353 	/* locate the major number */
354 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
355 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
356 
357 	/* Nuke the vnodes for any open instances */
358 	for (i = 0; i < MAXPARTITIONS; i++) {
359 		mn = SDMINOR(self->dv_unit, i);
360 		vdevgone(bmaj, mn, mn, VBLK);
361 		vdevgone(cmaj, mn, mn, VCHR);
362 	}
363 
364 	/* kill any pending restart */
365 	callout_stop(&sd->sc_callout);
366 
367 	/* Delete all of our wedges. */
368 	dkwedge_delall(&sd->sc_dk);
369 
370 	s = splbio();
371 
372 	/* Kill off any queued buffers. */
373 	while ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
374 		bp->b_error = EIO;
375 		bp->b_flags |= B_ERROR;
376 		bp->b_resid = bp->b_bcount;
377 		biodone(bp);
378 	}
379 
380 	bufq_free(&sd->buf_queue);
381 
382 	/* Kill off any pending commands. */
383 	scsipi_kill_pending(sd->sc_periph);
384 
385 	splx(s);
386 
387 	/* Detach from the disk list. */
388 	disk_detach(&sd->sc_dk);
389 
390 	/* Get rid of the shutdown hook. */
391 	shutdownhook_disestablish(sd->sc_sdhook);
392 
393 #if NRND > 0
394 	/* Unhook the entropy source. */
395 	rnd_detach_source(&sd->rnd_source);
396 #endif
397 
398 	return (0);
399 }
400 
401 /*
402  * open the device. Make sure the partition info is a up-to-date as can be.
403  */
404 static int
405 sdopen(dev_t dev, int flag, int fmt, struct proc *p)
406 {
407 	struct sd_softc *sd;
408 	struct scsipi_periph *periph;
409 	struct scsipi_adapter *adapt;
410 	int unit, part;
411 	int error;
412 
413 	unit = SDUNIT(dev);
414 	if (unit >= sd_cd.cd_ndevs)
415 		return (ENXIO);
416 	sd = sd_cd.cd_devs[unit];
417 	if (sd == NULL)
418 		return (ENXIO);
419 
420 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
421 		return (ENODEV);
422 
423 	part = SDPART(dev);
424 
425 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
426 		return (error);
427 
428 	/*
429 	 * If there are wedges, and this is not RAW_PART, then we
430 	 * need to fail.
431 	 */
432 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
433 		error = EBUSY;
434 		goto bad1;
435 	}
436 
437 	periph = sd->sc_periph;
438 	adapt = periph->periph_channel->chan_adapter;
439 
440 	SC_DEBUG(periph, SCSIPI_DB1,
441 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
442 	    sd_cd.cd_ndevs, part));
443 
444 	/*
445 	 * If this is the first open of this device, add a reference
446 	 * to the adapter.
447 	 */
448 	if (sd->sc_dk.dk_openmask == 0 &&
449 	    (error = scsipi_adapter_addref(adapt)) != 0)
450 		goto bad1;
451 
452 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
453 		/*
454 		 * If any partition is open, but the disk has been invalidated,
455 		 * disallow further opens of non-raw partition
456 		 */
457 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
458 		    (part != RAW_PART || fmt != S_IFCHR)) {
459 			error = EIO;
460 			goto bad2;
461 		}
462 	} else {
463 		int silent;
464 
465 		if (part == RAW_PART && fmt == S_IFCHR)
466 			silent = XS_CTL_SILENT;
467 		else
468 			silent = 0;
469 
470 		/* Check that it is still responding and ok. */
471 		error = scsipi_test_unit_ready(periph,
472 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
473 		    silent);
474 
475 		/*
476 		 * Start the pack spinning if necessary. Always allow the
477 		 * raw parition to be opened, for raw IOCTLs. Data transfers
478 		 * will check for SDEV_MEDIA_LOADED.
479 		 */
480 		if (error == EIO) {
481 			int error2;
482 
483 			error2 = scsipi_start(periph, SSS_START, silent);
484 			switch (error2) {
485 			case 0:
486 				error = 0;
487 				break;
488 			case EIO:
489 			case EINVAL:
490 				break;
491 			default:
492 				error = error2;
493 				break;
494 			}
495 		}
496 		if (error) {
497 			if (silent)
498 				goto out;
499 			goto bad2;
500 		}
501 
502 		periph->periph_flags |= PERIPH_OPEN;
503 
504 		if (periph->periph_flags & PERIPH_REMOVABLE) {
505 			/* Lock the pack in. */
506 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
507 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
508 			    XS_CTL_IGNORE_MEDIA_CHANGE);
509 			if (error)
510 				goto bad3;
511 		}
512 
513 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
514 			int param_error;
515 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
516 
517 			/*
518 			 * Load the physical device parameters.
519 			 *
520 			 * Note that if media is present but unformatted,
521 			 * we allow the open (so that it can be formatted!).
522 			 * The drive should refuse real I/O, if the media is
523 			 * unformatted.
524 			 */
525 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
526 			     == SDGP_RESULT_OFFLINE) {
527 				error = ENXIO;
528 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
529 				goto bad3;
530 			}
531 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
532 
533 			/* Load the partition info if not already loaded. */
534 			if (param_error == 0) {
535 				sdgetdisklabel(sd);
536 				SC_DEBUG(periph, SCSIPI_DB3,
537 				     ("Disklabel loaded "));
538 			}
539 		}
540 	}
541 
542 	/* Check that the partition exists. */
543 	if (part != RAW_PART &&
544 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
545 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
546 		error = ENXIO;
547 		goto bad3;
548 	}
549 
550  out:	/* Insure only one open at a time. */
551 	switch (fmt) {
552 	case S_IFCHR:
553 		sd->sc_dk.dk_copenmask |= (1 << part);
554 		break;
555 	case S_IFBLK:
556 		sd->sc_dk.dk_bopenmask |= (1 << part);
557 		break;
558 	}
559 	sd->sc_dk.dk_openmask =
560 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
561 
562 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
563 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
564 	return (0);
565 
566  bad3:
567 	if (sd->sc_dk.dk_openmask == 0) {
568 		if (periph->periph_flags & PERIPH_REMOVABLE)
569 			scsipi_prevent(periph, SPAMR_ALLOW,
570 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
571 			    XS_CTL_IGNORE_MEDIA_CHANGE);
572 		periph->periph_flags &= ~PERIPH_OPEN;
573 	}
574 
575  bad2:
576 	if (sd->sc_dk.dk_openmask == 0)
577 		scsipi_adapter_delref(adapt);
578 
579  bad1:
580 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
581 	return (error);
582 }
583 
584 /*
585  * close the device.. only called if we are the LAST occurence of an open
586  * device.  Convenient now but usually a pain.
587  */
588 static int
589 sdclose(dev_t dev, int flag, int fmt, struct proc *p)
590 {
591 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
592 	struct scsipi_periph *periph = sd->sc_periph;
593 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
594 	int part = SDPART(dev);
595 	int error;
596 
597 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
598 		return (error);
599 
600 	switch (fmt) {
601 	case S_IFCHR:
602 		sd->sc_dk.dk_copenmask &= ~(1 << part);
603 		break;
604 	case S_IFBLK:
605 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
606 		break;
607 	}
608 	sd->sc_dk.dk_openmask =
609 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
610 
611 	if (sd->sc_dk.dk_openmask == 0) {
612 		/*
613 		 * If the disk cache needs flushing, and the disk supports
614 		 * it, do it now.
615 		 */
616 		if ((sd->flags & SDF_DIRTY) != 0) {
617 			if (sd_flush(sd, 0)) {
618 				printf("%s: cache synchronization failed\n",
619 				    sd->sc_dev.dv_xname);
620 				sd->flags &= ~SDF_FLUSHING;
621 			} else
622 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
623 		}
624 
625 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
626 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
627 
628 		scsipi_wait_drain(periph);
629 
630 		if (periph->periph_flags & PERIPH_REMOVABLE)
631 			scsipi_prevent(periph, SPAMR_ALLOW,
632 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
633 			    XS_CTL_IGNORE_NOT_READY);
634 		periph->periph_flags &= ~PERIPH_OPEN;
635 
636 		scsipi_wait_drain(periph);
637 
638 		scsipi_adapter_delref(adapt);
639 	}
640 
641 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
642 	return (0);
643 }
644 
645 /*
646  * Actually translate the requested transfer into one the physical driver
647  * can understand.  The transfer is described by a buf and will include
648  * only one physical transfer.
649  */
650 static void
651 sdstrategy(struct buf *bp)
652 {
653 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
654 	struct scsipi_periph *periph = sd->sc_periph;
655 	struct disklabel *lp;
656 	daddr_t blkno;
657 	int s;
658 	boolean_t sector_aligned;
659 
660 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
661 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
662 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
663 	/*
664 	 * If the device has been made invalid, error out
665 	 */
666 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
667 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
668 		if (periph->periph_flags & PERIPH_OPEN)
669 			bp->b_error = EIO;
670 		else
671 			bp->b_error = ENODEV;
672 		goto bad;
673 	}
674 
675 	lp = sd->sc_dk.dk_label;
676 
677 	/*
678 	 * The transfer must be a whole number of blocks, offset must not be
679 	 * negative.
680 	 */
681 	if (lp->d_secsize == DEV_BSIZE) {
682 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
683 	} else {
684 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
685 	}
686 	if (!sector_aligned || bp->b_blkno < 0) {
687 		bp->b_error = EINVAL;
688 		goto bad;
689 	}
690 	/*
691 	 * If it's a null transfer, return immediatly
692 	 */
693 	if (bp->b_bcount == 0)
694 		goto done;
695 
696 	/*
697 	 * Do bounds checking, adjust transfer. if error, process.
698 	 * If end of partition, just return.
699 	 */
700 	if (SDPART(bp->b_dev) == RAW_PART) {
701 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
702 		    sd->params.disksize512) <= 0)
703 			goto done;
704 	} else {
705 		if (bounds_check_with_label(&sd->sc_dk, bp,
706 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
707 			goto done;
708 	}
709 
710 	/*
711 	 * Now convert the block number to absolute and put it in
712 	 * terms of the device's logical block size.
713 	 */
714 	if (lp->d_secsize == DEV_BSIZE)
715 		blkno = bp->b_blkno;
716 	else if (lp->d_secsize > DEV_BSIZE)
717 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
718 	else
719 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
720 
721 	if (SDPART(bp->b_dev) != RAW_PART)
722 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
723 
724 	bp->b_rawblkno = blkno;
725 
726 	s = splbio();
727 
728 	/*
729 	 * Place it in the queue of disk activities for this disk.
730 	 *
731 	 * XXX Only do disksort() if the current operating mode does not
732 	 * XXX include tagged queueing.
733 	 */
734 	BUFQ_PUT(&sd->buf_queue, bp);
735 
736 	/*
737 	 * Tell the device to get going on the transfer if it's
738 	 * not doing anything, otherwise just wait for completion
739 	 */
740 	sdstart(sd->sc_periph);
741 
742 	splx(s);
743 	return;
744 
745 bad:
746 	bp->b_flags |= B_ERROR;
747 done:
748 	/*
749 	 * Correctly set the buf to indicate a completed xfer
750 	 */
751 	bp->b_resid = bp->b_bcount;
752 	biodone(bp);
753 }
754 
755 /*
756  * sdstart looks to see if there is a buf waiting for the device
757  * and that the device is not already busy. If both are true,
758  * It dequeues the buf and creates a scsi command to perform the
759  * transfer in the buf. The transfer request will call scsipi_done
760  * on completion, which will in turn call this routine again
761  * so that the next queued transfer is performed.
762  * The bufs are queued by the strategy routine (sdstrategy)
763  *
764  * This routine is also called after other non-queued requests
765  * have been made of the scsi driver, to ensure that the queue
766  * continues to be drained.
767  *
768  * must be called at the correct (highish) spl level
769  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
770  */
771 static void
772 sdstart(struct scsipi_periph *periph)
773 {
774 	struct sd_softc *sd = (void *)periph->periph_dev;
775 	struct disklabel *lp = sd->sc_dk.dk_label;
776 	struct buf *bp = 0;
777 	struct scsipi_rw_16 cmd16;
778 	struct scsipi_rw_10 cmd_big;
779 	struct scsi_rw_6 cmd_small;
780 	struct scsipi_generic *cmdp;
781 	struct scsipi_xfer *xs;
782 	int nblks, cmdlen, error, flags;
783 
784 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
785 	/*
786 	 * Check if the device has room for another command
787 	 */
788 	while (periph->periph_active < periph->periph_openings) {
789 		/*
790 		 * there is excess capacity, but a special waits
791 		 * It'll need the adapter as soon as we clear out of the
792 		 * way and let it run (user level wait).
793 		 */
794 		if (periph->periph_flags & PERIPH_WAITING) {
795 			periph->periph_flags &= ~PERIPH_WAITING;
796 			wakeup((caddr_t)periph);
797 			return;
798 		}
799 
800 		/*
801 		 * If the device has become invalid, abort all the
802 		 * reads and writes until all files have been closed and
803 		 * re-opened
804 		 */
805 		if (__predict_false(
806 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
807 			if ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
808 				bp->b_error = EIO;
809 				bp->b_flags |= B_ERROR;
810 				bp->b_resid = bp->b_bcount;
811 				biodone(bp);
812 				continue;
813 			} else {
814 				return;
815 			}
816 		}
817 
818 		/*
819 		 * See if there is a buf with work for us to do..
820 		 */
821 		if ((bp = BUFQ_PEEK(&sd->buf_queue)) == NULL)
822 			return;
823 
824 		/*
825 		 * We have a buf, now we should make a command.
826 		 */
827 
828 		if (lp->d_secsize == DEV_BSIZE)
829 			nblks = bp->b_bcount >> DEV_BSHIFT;
830 		else
831 			nblks = howmany(bp->b_bcount, lp->d_secsize);
832 
833 		/*
834 		 * Fill out the scsi command.  Use the smallest CDB possible
835 		 * (6-byte, 10-byte, or 16-byte).
836 		 */
837 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
838 		    ((nblks & 0xff) == nblks) &&
839 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
840 			/* 6-byte CDB */
841 			memset(&cmd_small, 0, sizeof(cmd_small));
842 			cmd_small.opcode = (bp->b_flags & B_READ) ?
843 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
844 			_lto3b(bp->b_rawblkno, cmd_small.addr);
845 			cmd_small.length = nblks & 0xff;
846 			cmdlen = sizeof(cmd_small);
847 			cmdp = (struct scsipi_generic *)&cmd_small;
848 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
849 			/* 10-byte CDB */
850 			memset(&cmd_big, 0, sizeof(cmd_big));
851 			cmd_big.opcode = (bp->b_flags & B_READ) ?
852 			    READ_10 : WRITE_10;
853 			_lto4b(bp->b_rawblkno, cmd_big.addr);
854 			_lto2b(nblks, cmd_big.length);
855 			cmdlen = sizeof(cmd_big);
856 			cmdp = (struct scsipi_generic *)&cmd_big;
857 		} else {
858 			/* 16-byte CDB */
859 			memset(&cmd16, 0, sizeof(cmd16));
860 			cmd16.opcode = (bp->b_flags & B_READ) ?
861 			    READ_16 : WRITE_16;
862 			_lto8b(bp->b_rawblkno, cmd16.addr);
863 			_lto4b(nblks, cmd16.length);
864 			cmdlen = sizeof(cmd16);
865 			cmdp = (struct scsipi_generic *)&cmd16;
866 		}
867 
868 		/* Instrumentation. */
869 		disk_busy(&sd->sc_dk);
870 
871 		/*
872 		 * Mark the disk dirty so that the cache will be
873 		 * flushed on close.
874 		 */
875 		if ((bp->b_flags & B_READ) == 0)
876 			sd->flags |= SDF_DIRTY;
877 
878 		/*
879 		 * Figure out what flags to use.
880 		 */
881 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
882 		if (bp->b_flags & B_READ)
883 			flags |= XS_CTL_DATA_IN;
884 		else
885 			flags |= XS_CTL_DATA_OUT;
886 
887 		/*
888 		 * Call the routine that chats with the adapter.
889 		 * Note: we cannot sleep as we may be an interrupt
890 		 */
891 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
892 		    (u_char *)bp->b_data, bp->b_bcount,
893 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
894 		if (__predict_false(xs == NULL)) {
895 			/*
896 			 * out of memory. Keep this buffer in the queue, and
897 			 * retry later.
898 			 */
899 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
900 			    periph);
901 			return;
902 		}
903 		/*
904 		 * need to dequeue the buffer before queuing the command,
905 		 * because cdstart may be called recursively from the
906 		 * HBA driver
907 		 */
908 #ifdef DIAGNOSTIC
909 		if (BUFQ_GET(&sd->buf_queue) != bp)
910 			panic("sdstart(): dequeued wrong buf");
911 #else
912 		BUFQ_GET(&sd->buf_queue);
913 #endif
914 		error = scsipi_execute_xs(xs);
915 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
916 		KASSERT(error == 0);
917 	}
918 }
919 
920 static void
921 sdrestart(void *v)
922 {
923 	int s = splbio();
924 	sdstart((struct scsipi_periph *)v);
925 	splx(s);
926 }
927 
928 static void
929 sddone(struct scsipi_xfer *xs, int error)
930 {
931 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
932 	struct buf *bp = xs->bp;
933 
934 	if (sd->flags & SDF_FLUSHING) {
935 		/* Flush completed, no longer dirty. */
936 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
937 	}
938 
939 	if (bp) {
940 		bp->b_error = error;
941 		bp->b_resid = xs->resid;
942 		if (error)
943 			bp->b_flags |= B_ERROR;
944 
945 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
946 		    (bp->b_flags & B_READ));
947 #if NRND > 0
948 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
949 #endif
950 
951 		biodone(bp);
952 	}
953 }
954 
955 static void
956 sdminphys(struct buf *bp)
957 {
958 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
959 	long max;
960 
961 	/*
962 	 * If the device is ancient, we want to make sure that
963 	 * the transfer fits into a 6-byte cdb.
964 	 *
965 	 * XXX Note that the SCSI-I spec says that 256-block transfers
966 	 * are allowed in a 6-byte read/write, and are specified
967 	 * by settng the "length" to 0.  However, we're conservative
968 	 * here, allowing only 255-block transfers in case an
969 	 * ancient device gets confused by length == 0.  A length of 0
970 	 * in a 10-byte read/write actually means 0 blocks.
971 	 */
972 	if ((sd->flags & SDF_ANCIENT) &&
973 	    ((sd->sc_periph->periph_flags &
974 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
975 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
976 
977 		if (bp->b_bcount > max)
978 			bp->b_bcount = max;
979 	}
980 
981 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
982 }
983 
984 static int
985 sdread(dev_t dev, struct uio *uio, int ioflag)
986 {
987 
988 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
989 }
990 
991 static int
992 sdwrite(dev_t dev, struct uio *uio, int ioflag)
993 {
994 
995 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
996 }
997 
998 /*
999  * Perform special action on behalf of the user
1000  * Knows about the internals of this device
1001  */
1002 static int
1003 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct proc *p)
1004 {
1005 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1006 	struct scsipi_periph *periph = sd->sc_periph;
1007 	int part = SDPART(dev);
1008 	int error = 0;
1009 #ifdef __HAVE_OLD_DISKLABEL
1010 	struct disklabel *newlabel = NULL;
1011 #endif
1012 
1013 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1014 
1015 	/*
1016 	 * If the device is not valid, some IOCTLs can still be
1017 	 * handled on the raw partition. Check this here.
1018 	 */
1019 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1020 		switch (cmd) {
1021 		case DIOCKLABEL:
1022 		case DIOCWLABEL:
1023 		case DIOCLOCK:
1024 		case DIOCEJECT:
1025 		case ODIOCEJECT:
1026 		case DIOCGCACHE:
1027 		case DIOCSCACHE:
1028 		case SCIOCIDENTIFY:
1029 		case OSCIOCIDENTIFY:
1030 		case SCIOCCOMMAND:
1031 		case SCIOCDEBUG:
1032 			if (part == RAW_PART)
1033 				break;
1034 		/* FALLTHROUGH */
1035 		default:
1036 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1037 				return (ENODEV);
1038 			else
1039 				return (EIO);
1040 		}
1041 	}
1042 
1043 	switch (cmd) {
1044 	case DIOCGDINFO:
1045 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1046 		return (0);
1047 
1048 #ifdef __HAVE_OLD_DISKLABEL
1049 	case ODIOCGDINFO:
1050 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1051 		if (newlabel == NULL)
1052 			return EIO;
1053 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1054 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1055 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1056 		else
1057 			error = ENOTTY;
1058 		free(newlabel, M_TEMP);
1059 		return error;
1060 #endif
1061 
1062 	case DIOCGPART:
1063 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1064 		((struct partinfo *)addr)->part =
1065 		    &sd->sc_dk.dk_label->d_partitions[part];
1066 		return (0);
1067 
1068 	case DIOCWDINFO:
1069 	case DIOCSDINFO:
1070 #ifdef __HAVE_OLD_DISKLABEL
1071 	case ODIOCWDINFO:
1072 	case ODIOCSDINFO:
1073 #endif
1074 	{
1075 		struct disklabel *lp;
1076 
1077 		if ((flag & FWRITE) == 0)
1078 			return (EBADF);
1079 
1080 #ifdef __HAVE_OLD_DISKLABEL
1081  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1082 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1083 			if (newlabel == NULL)
1084 				return EIO;
1085 			memset(newlabel, 0, sizeof newlabel);
1086 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1087 			lp = newlabel;
1088 		} else
1089 #endif
1090 		lp = (struct disklabel *)addr;
1091 
1092 		if ((error = lockmgr(&sd->sc_dk.dk_openlock,
1093 				     LK_EXCLUSIVE, NULL)) != 0)
1094 			goto bad;
1095 		sd->flags |= SDF_LABELLING;
1096 
1097 		error = setdisklabel(sd->sc_dk.dk_label,
1098 		    lp, /*sd->sc_dk.dk_openmask : */0,
1099 		    sd->sc_dk.dk_cpulabel);
1100 		if (error == 0) {
1101 			if (cmd == DIOCWDINFO
1102 #ifdef __HAVE_OLD_DISKLABEL
1103 			    || cmd == ODIOCWDINFO
1104 #endif
1105 			   )
1106 				error = writedisklabel(SDLABELDEV(dev),
1107 				    sdstrategy, sd->sc_dk.dk_label,
1108 				    sd->sc_dk.dk_cpulabel);
1109 		}
1110 
1111 		sd->flags &= ~SDF_LABELLING;
1112 		(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1113 bad:
1114 #ifdef __HAVE_OLD_DISKLABEL
1115 		if (newlabel != NULL)
1116 			free(newlabel, M_TEMP);
1117 #endif
1118 		return (error);
1119 	}
1120 
1121 	case DIOCKLABEL:
1122 		if (*(int *)addr)
1123 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1124 		else
1125 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1126 		return (0);
1127 
1128 	case DIOCWLABEL:
1129 		if ((flag & FWRITE) == 0)
1130 			return (EBADF);
1131 		if (*(int *)addr)
1132 			sd->flags |= SDF_WLABEL;
1133 		else
1134 			sd->flags &= ~SDF_WLABEL;
1135 		return (0);
1136 
1137 	case DIOCLOCK:
1138 		return (scsipi_prevent(periph,
1139 		    (*(int *)addr) ? SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1140 
1141 	case DIOCEJECT:
1142 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1143 			return (ENOTTY);
1144 		if (*(int *)addr == 0) {
1145 			/*
1146 			 * Don't force eject: check that we are the only
1147 			 * partition open. If so, unlock it.
1148 			 */
1149 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1150 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1151 			    sd->sc_dk.dk_openmask) {
1152 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1153 				    XS_CTL_IGNORE_NOT_READY);
1154 				if (error)
1155 					return (error);
1156 			} else {
1157 				return (EBUSY);
1158 			}
1159 		}
1160 		/* FALLTHROUGH */
1161 	case ODIOCEJECT:
1162 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1163 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1164 
1165 	case DIOCGDEFLABEL:
1166 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1167 		return (0);
1168 
1169 #ifdef __HAVE_OLD_DISKLABEL
1170 	case ODIOCGDEFLABEL:
1171 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1172 		if (newlabel == NULL)
1173 			return EIO;
1174 		sdgetdefaultlabel(sd, newlabel);
1175 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1176 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1177 		else
1178 			error = ENOTTY;
1179 		free(newlabel, M_TEMP);
1180 		return error;
1181 #endif
1182 
1183 	case DIOCGCACHE:
1184 		return (sd_getcache(sd, (int *) addr));
1185 
1186 	case DIOCSCACHE:
1187 		if ((flag & FWRITE) == 0)
1188 			return (EBADF);
1189 		return (sd_setcache(sd, *(int *) addr));
1190 
1191 	case DIOCCACHESYNC:
1192 		/*
1193 		 * XXX Do we really need to care about having a writable
1194 		 * file descriptor here?
1195 		 */
1196 		if ((flag & FWRITE) == 0)
1197 			return (EBADF);
1198 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1199 			error = sd_flush(sd, 0);
1200 			if (error)
1201 				sd->flags &= ~SDF_FLUSHING;
1202 			else
1203 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1204 		} else
1205 			error = 0;
1206 		return (error);
1207 
1208 	case DIOCAWEDGE:
1209 	    {
1210 	    	struct dkwedge_info *dkw = (void *) addr;
1211 
1212 		if ((flag & FWRITE) == 0)
1213 			return (EBADF);
1214 
1215 		/* If the ioctl happens here, the parent is us. */
1216 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1217 		return (dkwedge_add(dkw));
1218 	    }
1219 
1220 	case DIOCDWEDGE:
1221 	    {
1222 	    	struct dkwedge_info *dkw = (void *) addr;
1223 
1224 		if ((flag & FWRITE) == 0)
1225 			return (EBADF);
1226 
1227 		/* If the ioctl happens here, the parent is us. */
1228 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1229 		return (dkwedge_del(dkw));
1230 	    }
1231 
1232 	case DIOCLWEDGES:
1233 	    {
1234 	    	struct dkwedge_list *dkwl = (void *) addr;
1235 
1236 		return (dkwedge_list(&sd->sc_dk, dkwl, p));
1237 	    }
1238 
1239 	default:
1240 		if (part != RAW_PART)
1241 			return (ENOTTY);
1242 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1243 	}
1244 
1245 #ifdef DIAGNOSTIC
1246 	panic("sdioctl: impossible");
1247 #endif
1248 }
1249 
1250 static void
1251 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1252 {
1253 
1254 	memset(lp, 0, sizeof(struct disklabel));
1255 
1256 	lp->d_secsize = sd->params.blksize;
1257 	lp->d_ntracks = sd->params.heads;
1258 	lp->d_nsectors = sd->params.sectors;
1259 	lp->d_ncylinders = sd->params.cyls;
1260 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1261 
1262 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1263 	case SCSIPI_BUSTYPE_SCSI:
1264 		lp->d_type = DTYPE_SCSI;
1265 		break;
1266 	case SCSIPI_BUSTYPE_ATAPI:
1267 		lp->d_type = DTYPE_ATAPI;
1268 		break;
1269 	}
1270 	/*
1271 	 * XXX
1272 	 * We could probe the mode pages to figure out what kind of disc it is.
1273 	 * Is this worthwhile?
1274 	 */
1275 	strncpy(lp->d_typename, "mydisk", 16);
1276 	strncpy(lp->d_packname, "fictitious", 16);
1277 	lp->d_secperunit = sd->params.disksize;
1278 	lp->d_rpm = sd->params.rot_rate;
1279 	lp->d_interleave = 1;
1280 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1281 	    D_REMOVABLE : 0;
1282 
1283 	lp->d_partitions[RAW_PART].p_offset = 0;
1284 	lp->d_partitions[RAW_PART].p_size =
1285 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1286 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1287 	lp->d_npartitions = RAW_PART + 1;
1288 
1289 	lp->d_magic = DISKMAGIC;
1290 	lp->d_magic2 = DISKMAGIC;
1291 	lp->d_checksum = dkcksum(lp);
1292 }
1293 
1294 
1295 /*
1296  * Load the label information on the named device
1297  */
1298 static void
1299 sdgetdisklabel(struct sd_softc *sd)
1300 {
1301 	struct disklabel *lp = sd->sc_dk.dk_label;
1302 	const char *errstring;
1303 
1304 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1305 
1306 	sdgetdefaultlabel(sd, lp);
1307 
1308 	if (lp->d_secpercyl == 0) {
1309 		lp->d_secpercyl = 100;
1310 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1311 	}
1312 
1313 	/*
1314 	 * Call the generic disklabel extraction routine
1315 	 */
1316 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1317 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1318 	if (errstring) {
1319 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1320 		return;
1321 	}
1322 }
1323 
1324 static void
1325 sd_shutdown(void *arg)
1326 {
1327 	struct sd_softc *sd = arg;
1328 
1329 	/*
1330 	 * If the disk cache needs to be flushed, and the disk supports
1331 	 * it, flush it.  We're cold at this point, so we poll for
1332 	 * completion.
1333 	 */
1334 	if ((sd->flags & SDF_DIRTY) != 0) {
1335 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1336 			printf("%s: cache synchronization failed\n",
1337 			    sd->sc_dev.dv_xname);
1338 			sd->flags &= ~SDF_FLUSHING;
1339 		} else
1340 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1341 	}
1342 }
1343 
1344 /*
1345  * Check Errors
1346  */
1347 static int
1348 sd_interpret_sense(struct scsipi_xfer *xs)
1349 {
1350 	struct scsipi_periph *periph = xs->xs_periph;
1351 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1352 	struct sd_softc *sd = (void *)periph->periph_dev;
1353 	int s, error, retval = EJUSTRETURN;
1354 
1355 	/*
1356 	 * If the periph is already recovering, just do the normal
1357 	 * error processing.
1358 	 */
1359 	if (periph->periph_flags & PERIPH_RECOVERING)
1360 		return (retval);
1361 
1362 	/*
1363 	 * If the device is not open yet, let the generic code handle it.
1364 	 */
1365 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1366 		return (retval);
1367 
1368 	/*
1369 	 * If it isn't a extended or extended/deferred error, let
1370 	 * the generic code handle it.
1371 	 */
1372 	if ((sense->response_code & SSD_RCODE_VALID) == 0 ||
1373 	    (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1374 	     SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED))
1375 		return (retval);
1376 
1377 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1378 	    sense->asc == 0x4) {
1379 		if (sense->ascq == 0x01)	{
1380 			/*
1381 			 * Unit In The Process Of Becoming Ready.
1382 			 */
1383 			printf("%s: waiting for pack to spin up...\n",
1384 			    sd->sc_dev.dv_xname);
1385 			if (!callout_pending(&periph->periph_callout))
1386 				scsipi_periph_freeze(periph, 1);
1387 			callout_reset(&periph->periph_callout,
1388 			    5 * hz, scsipi_periph_timed_thaw, periph);
1389 			retval = ERESTART;
1390 		} else if (sense->ascq == 0x02) {
1391 			printf("%s: pack is stopped, restarting...\n",
1392 			    sd->sc_dev.dv_xname);
1393 			s = splbio();
1394 			periph->periph_flags |= PERIPH_RECOVERING;
1395 			splx(s);
1396 			error = scsipi_start(periph, SSS_START,
1397 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1398 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1399 			if (error) {
1400 				printf("%s: unable to restart pack\n",
1401 				    sd->sc_dev.dv_xname);
1402 				retval = error;
1403 			} else
1404 				retval = ERESTART;
1405 			s = splbio();
1406 			periph->periph_flags &= ~PERIPH_RECOVERING;
1407 			splx(s);
1408 		}
1409 	}
1410 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1411 	    sense->asc == 0x31 &&
1412 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1413 		/* Medium Format Corrupted */
1414 		retval = EFTYPE;
1415 	}
1416 	return (retval);
1417 }
1418 
1419 
1420 static int
1421 sdsize(dev_t dev)
1422 {
1423 	struct sd_softc *sd;
1424 	int part, unit, omask;
1425 	int size;
1426 
1427 	unit = SDUNIT(dev);
1428 	if (unit >= sd_cd.cd_ndevs)
1429 		return (-1);
1430 	sd = sd_cd.cd_devs[unit];
1431 	if (sd == NULL)
1432 		return (-1);
1433 
1434 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1435 		return (-1);
1436 
1437 	part = SDPART(dev);
1438 	omask = sd->sc_dk.dk_openmask & (1 << part);
1439 
1440 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1441 		return (-1);
1442 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1443 		size = -1;
1444 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1445 		size = -1;
1446 	else
1447 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1448 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1449 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1450 		return (-1);
1451 	return (size);
1452 }
1453 
1454 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1455 static struct scsipi_xfer sx;
1456 static int sddoingadump;
1457 
1458 /*
1459  * dump all of physical memory into the partition specified, starting
1460  * at offset 'dumplo' into the partition.
1461  */
1462 static int
1463 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1464 {
1465 	struct sd_softc *sd;	/* disk unit to do the I/O */
1466 	struct disklabel *lp;	/* disk's disklabel */
1467 	int	unit, part;
1468 	int	sectorsize;	/* size of a disk sector */
1469 	int	nsects;		/* number of sectors in partition */
1470 	int	sectoff;	/* sector offset of partition */
1471 	int	totwrt;		/* total number of sectors left to write */
1472 	int	nwrt;		/* current number of sectors to write */
1473 	struct scsipi_rw_10 cmd;	/* write command */
1474 	struct scsipi_xfer *xs;	/* ... convenience */
1475 	struct scsipi_periph *periph;
1476 	struct scsipi_channel *chan;
1477 
1478 	/* Check if recursive dump; if so, punt. */
1479 	if (sddoingadump)
1480 		return (EFAULT);
1481 
1482 	/* Mark as active early. */
1483 	sddoingadump = 1;
1484 
1485 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1486 	part = SDPART(dev);
1487 
1488 	/* Check for acceptable drive number. */
1489 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1490 		return (ENXIO);
1491 
1492 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1493 		return (ENODEV);
1494 
1495 	periph = sd->sc_periph;
1496 	chan = periph->periph_channel;
1497 
1498 	/* Make sure it was initialized. */
1499 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1500 		return (ENXIO);
1501 
1502 	/* Convert to disk sectors.  Request must be a multiple of size. */
1503 	lp = sd->sc_dk.dk_label;
1504 	sectorsize = lp->d_secsize;
1505 	if ((size % sectorsize) != 0)
1506 		return (EFAULT);
1507 	totwrt = size / sectorsize;
1508 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1509 
1510 	nsects = lp->d_partitions[part].p_size;
1511 	sectoff = lp->d_partitions[part].p_offset;
1512 
1513 	/* Check transfer bounds against partition size. */
1514 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1515 		return (EINVAL);
1516 
1517 	/* Offset block number to start of partition. */
1518 	blkno += sectoff;
1519 
1520 	xs = &sx;
1521 
1522 	while (totwrt > 0) {
1523 		nwrt = totwrt;		/* XXX */
1524 #ifndef	SD_DUMP_NOT_TRUSTED
1525 		/*
1526 		 *  Fill out the scsi command
1527 		 */
1528 		memset(&cmd, 0, sizeof(cmd));
1529 		cmd.opcode = WRITE_10;
1530 		_lto4b(blkno, cmd.addr);
1531 		_lto2b(nwrt, cmd.length);
1532 		/*
1533 		 * Fill out the scsipi_xfer structure
1534 		 *    Note: we cannot sleep as we may be an interrupt
1535 		 * don't use scsipi_command() as it may want to wait
1536 		 * for an xs.
1537 		 */
1538 		memset(xs, 0, sizeof(sx));
1539 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1540 		    XS_CTL_DATA_OUT;
1541 		xs->xs_status = 0;
1542 		xs->xs_periph = periph;
1543 		xs->xs_retries = SDRETRIES;
1544 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1545 		xs->cmd = (struct scsipi_generic *)&cmd;
1546 		xs->cmdlen = sizeof(cmd);
1547 		xs->resid = nwrt * sectorsize;
1548 		xs->error = XS_NOERROR;
1549 		xs->bp = 0;
1550 		xs->data = va;
1551 		xs->datalen = nwrt * sectorsize;
1552 
1553 		/*
1554 		 * Pass all this info to the scsi driver.
1555 		 */
1556 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1557 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1558 		    xs->error != XS_NOERROR)
1559 			return (EIO);
1560 #else	/* SD_DUMP_NOT_TRUSTED */
1561 		/* Let's just talk about this first... */
1562 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1563 		delay(500 * 1000);	/* half a second */
1564 #endif	/* SD_DUMP_NOT_TRUSTED */
1565 
1566 		/* update block count */
1567 		totwrt -= nwrt;
1568 		blkno += nwrt;
1569 		va += sectorsize * nwrt;
1570 	}
1571 	sddoingadump = 0;
1572 	return (0);
1573 }
1574 
1575 static int
1576 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1577     int page, int flags, int *big)
1578 {
1579 
1580 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1581 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1582 		*big = 1;
1583 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1584 		    size + sizeof(struct scsi_mode_parameter_header_10),
1585 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1586 	} else {
1587 		*big = 0;
1588 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1589 		    size + sizeof(struct scsi_mode_parameter_header_6),
1590 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1591 	}
1592 }
1593 
1594 static int
1595 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1596     int flags, int big)
1597 {
1598 
1599 	if (big) {
1600 		struct scsi_mode_parameter_header_10 *header = sense;
1601 
1602 		_lto2b(0, header->data_length);
1603 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1604 		    size + sizeof(struct scsi_mode_parameter_header_10),
1605 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1606 	} else {
1607 		struct scsi_mode_parameter_header_6 *header = sense;
1608 
1609 		header->data_length = 0;
1610 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1611 		    size + sizeof(struct scsi_mode_parameter_header_6),
1612 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1613 	}
1614 }
1615 
1616 static int
1617 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1618 {
1619 	struct {
1620 		struct scsi_mode_parameter_header_6 header;
1621 		/* no block descriptor */
1622 		u_int8_t pg_code; /* page code (should be 6) */
1623 		u_int8_t pg_length; /* page length (should be 11) */
1624 		u_int8_t wcd; /* bit0: cache disable */
1625 		u_int8_t lbs[2]; /* logical block size */
1626 		u_int8_t size[5]; /* number of log. blocks */
1627 		u_int8_t pp; /* power/performance */
1628 		u_int8_t flags;
1629 		u_int8_t resvd;
1630 	} scsipi_sense;
1631 	u_int64_t sectors;
1632 	int error;
1633 
1634 	/*
1635 	 * scsipi_size (ie "read capacity") and mode sense page 6
1636 	 * give the same information. Do both for now, and check
1637 	 * for consistency.
1638 	 * XXX probably differs for removable media
1639 	 */
1640 	dp->blksize = 512;
1641 	if ((sectors = scsipi_size(sd->sc_periph, flags)) == 0)
1642 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1643 
1644 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1645 	    &scsipi_sense.header, sizeof(scsipi_sense),
1646 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1647 
1648 	if (error != 0)
1649 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1650 
1651 	dp->blksize = _2btol(scsipi_sense.lbs);
1652 	if (dp->blksize == 0)
1653 		dp->blksize = 512;
1654 
1655 	/*
1656 	 * Create a pseudo-geometry.
1657 	 */
1658 	dp->heads = 64;
1659 	dp->sectors = 32;
1660 	dp->cyls = sectors / (dp->heads * dp->sectors);
1661 	dp->disksize = _5btol(scsipi_sense.size);
1662 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1663 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1664 		       (unsigned long long)dp->disksize,
1665 		       (unsigned long long)sectors);
1666 		dp->disksize = sectors;
1667 	}
1668 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1669 
1670 	return (SDGP_RESULT_OK);
1671 }
1672 
1673 /*
1674  * Get the scsi driver to send a full inquiry to the * device and use the
1675  * results to fill out the disk parameter structure.
1676  */
1677 static int
1678 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1679 {
1680 	u_int64_t sectors;
1681 	int error;
1682 #if 0
1683 	int i;
1684 	u_int8_t *p;
1685 #endif
1686 
1687 	dp->disksize = sectors = scsipi_size(sd->sc_periph, flags);
1688 	if (sectors == 0) {
1689 		struct scsipi_read_format_capacities cmd;
1690 		struct {
1691 			struct scsipi_capacity_list_header header;
1692 			struct scsipi_capacity_descriptor desc;
1693 		} __attribute__((packed)) data;
1694 
1695 		memset(&cmd, 0, sizeof(cmd));
1696 		memset(&data, 0, sizeof(data));
1697 		cmd.opcode = READ_FORMAT_CAPACITIES;
1698 		_lto2b(sizeof(data), cmd.length);
1699 
1700 		error = scsipi_command(sd->sc_periph,
1701 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1702 		    SDRETRIES, 20000, NULL,
1703 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1704 		if (error == EFTYPE) {
1705 			/* Medium Format Corrupted, handle as not formatted */
1706 			return (SDGP_RESULT_UNFORMATTED);
1707 		}
1708 		if (error || data.header.length == 0)
1709 			return (SDGP_RESULT_OFFLINE);
1710 
1711 #if 0
1712 printf("rfc: length=%d\n", data.header.length);
1713 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");
1714 #endif
1715 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1716 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1717 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1718 			break;
1719 
1720 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1721 			return (SDGP_RESULT_UNFORMATTED);
1722 
1723 		case SCSIPI_CAP_DESC_CODE_NONE:
1724 			return (SDGP_RESULT_OFFLINE);
1725 		}
1726 
1727 		dp->disksize = sectors = _4btol(data.desc.nblks);
1728 		if (sectors == 0)
1729 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1730 
1731 		dp->blksize = _3btol(data.desc.blklen);
1732 		if (dp->blksize == 0)
1733 			dp->blksize = 512;
1734 	} else {
1735 		struct sd_mode_sense_data scsipi_sense;
1736 		int big, bsize;
1737 		struct scsi_general_block_descriptor *bdesc;
1738 
1739 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1740 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1741 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1742 		dp->blksize = 512;
1743 		if (!error) {
1744 			if (big) {
1745 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1746 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1747 			} else {
1748 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1749 				bsize = scsipi_sense.header.small.blk_desc_len;
1750 			}
1751 
1752 #if 0
1753 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1754 printf("page 0 bsize=%d\n", bsize);
1755 printf("page 0 ok\n");
1756 #endif
1757 
1758 			if (bsize >= 8) {
1759 				dp->blksize = _3btol(bdesc->blklen);
1760 				if (dp->blksize == 0)
1761 					dp->blksize = 512;
1762 			}
1763 		}
1764 	}
1765 
1766 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1767 	return (0);
1768 }
1769 
1770 static int
1771 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1772 {
1773 	struct sd_mode_sense_data scsipi_sense;
1774 	int error;
1775 	int big, poffset, byte2;
1776 	union scsi_disk_pages *pages;
1777 #if 0
1778 	int i;
1779 	u_int8_t *p;
1780 #endif
1781 
1782 	byte2 = SMS_DBD;
1783 again:
1784 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1785 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1786 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1787 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1788 	    flags | XS_CTL_SILENT, &big);
1789 	if (error) {
1790 		if (byte2 == SMS_DBD) {
1791 			/* No result; try once more with DBD off */
1792 			byte2 = 0;
1793 			goto again;
1794 		}
1795 		return (error);
1796 	}
1797 
1798 	if (big) {
1799 		poffset = sizeof scsipi_sense.header.big;
1800 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1801 	} else {
1802 		poffset = sizeof scsipi_sense.header.small;
1803 		poffset += scsipi_sense.header.small.blk_desc_len;
1804 	}
1805 
1806 	pages = (void *)((u_long)&scsipi_sense + poffset);
1807 #if 0
1808 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1809 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1810 #endif
1811 
1812 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1813 		return (ERESTART);
1814 
1815 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1816 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1817 	    _3btol(pages->rigid_geometry.ncyl),
1818 	    pages->rigid_geometry.nheads,
1819 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1820 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1821 	    _2btol(pages->rigid_geometry.land_zone)));
1822 
1823 	/*
1824 	 * KLUDGE!! (for zone recorded disks)
1825 	 * give a number of sectors so that sec * trks * cyls
1826 	 * is <= disk_size
1827 	 * can lead to wasted space! THINK ABOUT THIS !
1828 	 */
1829 	dp->heads = pages->rigid_geometry.nheads;
1830 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1831 	if (dp->heads == 0 || dp->cyls == 0)
1832 		return (ERESTART);
1833 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1834 
1835 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1836 	if (dp->rot_rate == 0)
1837 		dp->rot_rate = 3600;
1838 
1839 #if 0
1840 printf("page 4 ok\n");
1841 #endif
1842 	return (0);
1843 }
1844 
1845 static int
1846 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1847 {
1848 	struct sd_mode_sense_data scsipi_sense;
1849 	int error;
1850 	int big, poffset, byte2;
1851 	union scsi_disk_pages *pages;
1852 #if 0
1853 	int i;
1854 	u_int8_t *p;
1855 #endif
1856 
1857 	byte2 = SMS_DBD;
1858 again:
1859 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1860 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1861 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1862 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1863 	    flags | XS_CTL_SILENT, &big);
1864 	if (error) {
1865 		if (byte2 == SMS_DBD) {
1866 			/* No result; try once more with DBD off */
1867 			byte2 = 0;
1868 			goto again;
1869 		}
1870 		return (error);
1871 	}
1872 
1873 	if (big) {
1874 		poffset = sizeof scsipi_sense.header.big;
1875 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1876 	} else {
1877 		poffset = sizeof scsipi_sense.header.small;
1878 		poffset += scsipi_sense.header.small.blk_desc_len;
1879 	}
1880 
1881 	pages = (void *)((u_long)&scsipi_sense + poffset);
1882 #if 0
1883 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1884 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1885 #endif
1886 
1887 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1888 		return (ERESTART);
1889 
1890 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1891 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1892 	    _3btol(pages->flex_geometry.ncyl),
1893 	    pages->flex_geometry.nheads,
1894 	    pages->flex_geometry.ph_sec_tr,
1895 	    _2btol(pages->flex_geometry.bytes_s)));
1896 
1897 	dp->heads = pages->flex_geometry.nheads;
1898 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1899 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1900 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1901 		return (ERESTART);
1902 
1903 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1904 	if (dp->rot_rate == 0)
1905 		dp->rot_rate = 3600;
1906 
1907 #if 0
1908 printf("page 5 ok\n");
1909 #endif
1910 	return (0);
1911 }
1912 
1913 static int
1914 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1915 {
1916 	int error;
1917 
1918 	/*
1919 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1920 	 * cleared by the caller if necessary.
1921 	 */
1922 	if (sd->type == T_SIMPLE_DIRECT)
1923 		return (sd_get_simplifiedparms(sd, dp, flags));
1924 
1925 	error = sd_get_capacity(sd, dp, flags);
1926 	if (error)
1927 		return (error);
1928 
1929 	if (sd->type == T_OPTICAL)
1930 		goto page0;
1931 
1932 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1933 		if (!sd_get_parms_page5(sd, dp, flags) ||
1934 		    !sd_get_parms_page4(sd, dp, flags))
1935 			return (SDGP_RESULT_OK);
1936 	} else {
1937 		if (!sd_get_parms_page4(sd, dp, flags) ||
1938 		    !sd_get_parms_page5(sd, dp, flags))
1939 			return (SDGP_RESULT_OK);
1940 	}
1941 
1942 page0:
1943 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1944 	/* Try calling driver's method for figuring out geometry. */
1945 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1946 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1947 		(sd->sc_periph, dp, dp->disksize)) {
1948 		/*
1949 		 * Use adaptec standard fictitious geometry
1950 		 * this depends on which controller (e.g. 1542C is
1951 		 * different. but we have to put SOMETHING here..)
1952 		 */
1953 		dp->heads = 64;
1954 		dp->sectors = 32;
1955 		dp->cyls = dp->disksize / (64 * 32);
1956 	}
1957 	dp->rot_rate = 3600;
1958 	return (SDGP_RESULT_OK);
1959 }
1960 
1961 static int
1962 sd_flush(struct sd_softc *sd, int flags)
1963 {
1964 	struct scsipi_periph *periph = sd->sc_periph;
1965 	struct scsi_synchronize_cache_10 cmd;
1966 
1967 	/*
1968 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1969 	 * We issue with address 0 length 0, which should be
1970 	 * interpreted by the device as "all remaining blocks
1971 	 * starting at address 0".  We ignore ILLEGAL REQUEST
1972 	 * in the event that the command is not supported by
1973 	 * the device, and poll for completion so that we know
1974 	 * that the cache has actually been flushed.
1975 	 *
1976 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
1977 	 * command, as indicated by our quirks flags.
1978 	 *
1979 	 * XXX What about older devices?
1980 	 */
1981 	if (periph->periph_version < 2 ||
1982 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
1983 		return (0);
1984 
1985 	sd->flags |= SDF_FLUSHING;
1986 	memset(&cmd, 0, sizeof(cmd));
1987 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
1988 
1989 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
1990 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
1991 }
1992 
1993 static int
1994 sd_getcache(struct sd_softc *sd, int *bitsp)
1995 {
1996 	struct scsipi_periph *periph = sd->sc_periph;
1997 	struct sd_mode_sense_data scsipi_sense;
1998 	int error, bits = 0;
1999 	int big;
2000 	union scsi_disk_pages *pages;
2001 
2002 	if (periph->periph_version < 2)
2003 		return (EOPNOTSUPP);
2004 
2005 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2006 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2007 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2008 	if (error)
2009 		return (error);
2010 
2011 	if (big)
2012 		pages = (void *)(&scsipi_sense.header.big + 1);
2013 	else
2014 		pages = (void *)(&scsipi_sense.header.small + 1);
2015 
2016 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2017 		bits |= DKCACHE_READ;
2018 	if (pages->caching_params.flags & CACHING_WCE)
2019 		bits |= DKCACHE_WRITE;
2020 	if (pages->caching_params.pg_code & PGCODE_PS)
2021 		bits |= DKCACHE_SAVE;
2022 
2023 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2024 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2025 	    sizeof(scsipi_sense.pages.caching_params),
2026 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2027 	if (error == 0) {
2028 		if (big)
2029 			pages = (void *)(&scsipi_sense.header.big + 1);
2030 		else
2031 			pages = (void *)(&scsipi_sense.header.small + 1);
2032 
2033 		if (pages->caching_params.flags & CACHING_RCD)
2034 			bits |= DKCACHE_RCHANGE;
2035 		if (pages->caching_params.flags & CACHING_WCE)
2036 			bits |= DKCACHE_WCHANGE;
2037 	}
2038 
2039 	*bitsp = bits;
2040 
2041 	return (0);
2042 }
2043 
2044 static int
2045 sd_setcache(struct sd_softc *sd, int bits)
2046 {
2047 	struct scsipi_periph *periph = sd->sc_periph;
2048 	struct sd_mode_sense_data scsipi_sense;
2049 	int error;
2050 	uint8_t oflags, byte2 = 0;
2051 	int big;
2052 	union scsi_disk_pages *pages;
2053 
2054 	if (periph->periph_version < 2)
2055 		return (EOPNOTSUPP);
2056 
2057 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2058 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2059 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2060 	if (error)
2061 		return (error);
2062 
2063 	if (big)
2064 		pages = (void *)(&scsipi_sense.header.big + 1);
2065 	else
2066 		pages = (void *)(&scsipi_sense.header.small + 1);
2067 
2068 	oflags = pages->caching_params.flags;
2069 
2070 	if (bits & DKCACHE_READ)
2071 		pages->caching_params.flags &= ~CACHING_RCD;
2072 	else
2073 		pages->caching_params.flags |= CACHING_RCD;
2074 
2075 	if (bits & DKCACHE_WRITE)
2076 		pages->caching_params.flags |= CACHING_WCE;
2077 	else
2078 		pages->caching_params.flags &= ~CACHING_WCE;
2079 
2080 	if (oflags == pages->caching_params.flags)
2081 		return (0);
2082 
2083 	pages->caching_params.pg_code &= PGCODE_MASK;
2084 
2085 	if (bits & DKCACHE_SAVE)
2086 		byte2 |= SMS_SP;
2087 
2088 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2089 	    sizeof(struct scsi_mode_page_header) +
2090 	    pages->caching_params.pg_length, 0, big));
2091 }
2092