xref: /netbsd-src/sys/dev/scsipi/sd.c (revision fd5cb0acea84d278e04e640d37ca2398f894991f)
1 /*	$NetBSD: sd.c,v 1.235 2005/02/01 00:19:34 reinoud 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 1.235 2005/02/01 00:19:34 reinoud 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/scsipi_all.h>
85 #include <dev/scsipi/scsi_all.h>
86 #include <dev/scsipi/scsipi_disk.h>
87 #include <dev/scsipi/scsi_disk.h>
88 #include <dev/scsipi/scsiconf.h>
89 #include <dev/scsipi/scsipi_base.h>
90 #include <dev/scsipi/sdvar.h>
91 
92 #define	SDUNIT(dev)			DISKUNIT(dev)
93 #define	SDPART(dev)			DISKPART(dev)
94 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
95 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
96 
97 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
98 
99 static void	sdminphys(struct buf *);
100 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
101 static void	sdgetdisklabel(struct sd_softc *);
102 static void	sdstart(struct scsipi_periph *);
103 static void	sdrestart(void *);
104 static void	sddone(struct scsipi_xfer *, int);
105 static void	sd_shutdown(void *);
106 static int	sd_interpret_sense(struct scsipi_xfer *);
107 
108 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
109 		    int, int *);
110 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
111 		    int);
112 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
113 		    int);
114 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
115 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
116 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
117 		    int);
118 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
119 		    int);
120 
121 static int	sd_flush(struct sd_softc *, int);
122 static int	sd_getcache(struct sd_softc *, int *);
123 static int	sd_setcache(struct sd_softc *, int);
124 
125 static int	sdmatch(struct device *, struct cfdata *, void *);
126 static void	sdattach(struct device *, struct device *, void *);
127 static int	sdactivate(struct device *, enum devact);
128 static int	sddetach(struct device *, int);
129 
130 CFATTACH_DECL(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
131     sdactivate);
132 
133 extern struct cfdriver sd_cd;
134 
135 static const struct scsipi_inquiry_pattern sd_patterns[] = {
136 	{T_DIRECT, T_FIXED,
137 	 "",         "",                 ""},
138 	{T_DIRECT, T_REMOV,
139 	 "",         "",                 ""},
140 	{T_OPTICAL, T_FIXED,
141 	 "",         "",                 ""},
142 	{T_OPTICAL, T_REMOV,
143 	 "",         "",                 ""},
144 	{T_SIMPLE_DIRECT, T_FIXED,
145 	 "",         "",                 ""},
146 	{T_SIMPLE_DIRECT, T_REMOV,
147 	 "",         "",                 ""},
148 };
149 
150 static dev_type_open(sdopen);
151 static dev_type_close(sdclose);
152 static dev_type_read(sdread);
153 static dev_type_write(sdwrite);
154 static dev_type_ioctl(sdioctl);
155 static dev_type_strategy(sdstrategy);
156 static dev_type_dump(sddump);
157 static dev_type_size(sdsize);
158 
159 const struct bdevsw sd_bdevsw = {
160 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
161 };
162 
163 const struct cdevsw sd_cdevsw = {
164 	sdopen, sdclose, sdread, sdwrite, sdioctl,
165 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
166 };
167 
168 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
169 
170 static const struct scsipi_periphsw sd_switch = {
171 	sd_interpret_sense,	/* check our error handler first */
172 	sdstart,		/* have a queue, served by this */
173 	NULL,			/* have no async handler */
174 	sddone,			/* deal with stats at interrupt time */
175 };
176 
177 struct sd_mode_sense_data {
178 	/*
179 	 * XXX
180 	 * We are not going to parse this as-is -- it just has to be large
181 	 * enough.
182 	 */
183 	union {
184 		struct scsipi_mode_header small;
185 		struct scsipi_mode_header_big big;
186 	} header;
187 	struct scsi_blk_desc blk_desc;
188 	union scsi_disk_pages pages;
189 };
190 
191 /*
192  * The routine called by the low level scsi routine when it discovers
193  * A device suitable for this driver
194  */
195 static int
196 sdmatch(struct device *parent, struct cfdata *match, void *aux)
197 {
198 	struct scsipibus_attach_args *sa = aux;
199 	int priority;
200 
201 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
202 	    (caddr_t)sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
203 	    sizeof(sd_patterns[0]), &priority);
204 
205 	return (priority);
206 }
207 
208 /*
209  * Attach routine common to atapi & scsi.
210  */
211 static void
212 sdattach(struct device *parent, struct device *self, void *aux)
213 {
214 	struct sd_softc *sd = (void *)self;
215 	struct scsipibus_attach_args *sa = aux;
216 	struct scsipi_periph *periph = sa->sa_periph;
217 	int error, result;
218 	struct disk_parms *dp = &sd->params;
219 	char pbuf[9];
220 
221 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
222 
223 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
224 	if (sd->type == T_SIMPLE_DIRECT)
225 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
226 
227 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
228 	    periph->periph_version == 0)
229 		sd->flags |= SDF_ANCIENT;
230 
231 	bufq_alloc(&sd->buf_queue,
232 	    BUFQ_DISK_DEFAULT_STRAT()|BUFQ_SORT_RAWBLOCK);
233 
234 	callout_init(&sd->sc_callout);
235 
236 	/*
237 	 * Store information needed to contact our base driver
238 	 */
239 	sd->sc_periph = periph;
240 
241 	periph->periph_dev = &sd->sc_dev;
242 	periph->periph_switch = &sd_switch;
243 
244         /*
245          * Increase our openings to the maximum-per-periph
246          * supported by the adapter.  This will either be
247          * clamped down or grown by the adapter if necessary.
248          */
249 	periph->periph_openings =
250 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
251 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
252 
253 	/*
254 	 * Initialize and attach the disk structure.
255 	 */
256 	sd->sc_dk.dk_driver = &sddkdriver;
257 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
258 	disk_attach(&sd->sc_dk);
259 
260 	/*
261 	 * Use the subdriver to request information regarding the drive.
262 	 */
263 	aprint_naive("\n");
264 	aprint_normal("\n");
265 
266 	error = scsipi_test_unit_ready(periph,
267 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
268 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
269 
270 	if (error)
271 		result = SDGP_RESULT_OFFLINE;
272 	else
273 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
274 	aprint_normal("%s: ", sd->sc_dev.dv_xname);
275 	switch (result) {
276 	case SDGP_RESULT_OK:
277 		format_bytes(pbuf, sizeof(pbuf),
278 		    (u_int64_t)dp->disksize * dp->blksize);
279 	        aprint_normal(
280 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
281 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
282 		    (unsigned long long)dp->disksize);
283 		break;
284 
285 	case SDGP_RESULT_OFFLINE:
286 		aprint_normal("drive offline");
287 		break;
288 
289 	case SDGP_RESULT_UNFORMATTED:
290 		aprint_normal("unformatted media");
291 		break;
292 
293 #ifdef DIAGNOSTIC
294 	default:
295 		panic("sdattach: unknown result from get_parms");
296 		break;
297 #endif
298 	}
299 	aprint_normal("\n");
300 
301 	/*
302 	 * Establish a shutdown hook so that we can ensure that
303 	 * our data has actually made it onto the platter at
304 	 * shutdown time.  Note that this relies on the fact
305 	 * that the shutdown hook code puts us at the head of
306 	 * the list (thus guaranteeing that our hook runs before
307 	 * our ancestors').
308 	 */
309 	if ((sd->sc_sdhook =
310 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
311 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
312 		    sd->sc_dev.dv_xname);
313 
314 #if NRND > 0
315 	/*
316 	 * attach the device into the random source list
317 	 */
318 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
319 			  RND_TYPE_DISK, 0);
320 #endif
321 
322 	/* Discover wedges on this disk. */
323 	dkwedge_discover(&sd->sc_dk);
324 }
325 
326 static int
327 sdactivate(struct device *self, enum devact act)
328 {
329 	int rv = 0;
330 
331 	switch (act) {
332 	case DVACT_ACTIVATE:
333 		rv = EOPNOTSUPP;
334 		break;
335 
336 	case DVACT_DEACTIVATE:
337 		/*
338 		 * Nothing to do; we key off the device's DVF_ACTIVE.
339 		 */
340 		break;
341 	}
342 	return (rv);
343 }
344 
345 static int
346 sddetach(struct device *self, int flags)
347 {
348 	struct sd_softc *sd = (struct sd_softc *) self;
349 	struct buf *bp;
350 	int s, bmaj, cmaj, i, mn;
351 
352 	/* locate the major number */
353 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
354 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
355 
356 	/* Nuke the vnodes for any open instances */
357 	for (i = 0; i < MAXPARTITIONS; i++) {
358 		mn = SDMINOR(self->dv_unit, i);
359 		vdevgone(bmaj, mn, mn, VBLK);
360 		vdevgone(cmaj, mn, mn, VCHR);
361 	}
362 
363 	/* kill any pending restart */
364 	callout_stop(&sd->sc_callout);
365 
366 	/* Delete all of our wedges. */
367 	dkwedge_delall(&sd->sc_dk);
368 
369 	s = splbio();
370 
371 	/* Kill off any queued buffers. */
372 	while ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
373 		bp->b_error = EIO;
374 		bp->b_flags |= B_ERROR;
375 		bp->b_resid = bp->b_bcount;
376 		biodone(bp);
377 	}
378 
379 	bufq_free(&sd->buf_queue);
380 
381 	/* Kill off any pending commands. */
382 	scsipi_kill_pending(sd->sc_periph);
383 
384 	splx(s);
385 
386 	/* Detach from the disk list. */
387 	disk_detach(&sd->sc_dk);
388 
389 	/* Get rid of the shutdown hook. */
390 	shutdownhook_disestablish(sd->sc_sdhook);
391 
392 #if NRND > 0
393 	/* Unhook the entropy source. */
394 	rnd_detach_source(&sd->rnd_source);
395 #endif
396 
397 	return (0);
398 }
399 
400 /*
401  * open the device. Make sure the partition info is a up-to-date as can be.
402  */
403 static int
404 sdopen(dev_t dev, int flag, int fmt, struct proc *p)
405 {
406 	struct sd_softc *sd;
407 	struct scsipi_periph *periph;
408 	struct scsipi_adapter *adapt;
409 	int unit, part;
410 	int error;
411 
412 	unit = SDUNIT(dev);
413 	if (unit >= sd_cd.cd_ndevs)
414 		return (ENXIO);
415 	sd = sd_cd.cd_devs[unit];
416 	if (sd == NULL)
417 		return (ENXIO);
418 
419 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
420 		return (ENODEV);
421 
422 	part = SDPART(dev);
423 
424 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
425 		return (error);
426 
427 	/*
428 	 * If there are wedges, and this is not RAW_PART, then we
429 	 * need to fail.
430 	 */
431 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
432 		error = EBUSY;
433 		goto bad1;
434 	}
435 
436 	periph = sd->sc_periph;
437 	adapt = periph->periph_channel->chan_adapter;
438 
439 	SC_DEBUG(periph, SCSIPI_DB1,
440 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
441 	    sd_cd.cd_ndevs, part));
442 
443 	/*
444 	 * If this is the first open of this device, add a reference
445 	 * to the adapter.
446 	 */
447 	if (sd->sc_dk.dk_openmask == 0 &&
448 	    (error = scsipi_adapter_addref(adapt)) != 0)
449 		goto bad1;
450 
451 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
452 		/*
453 		 * If any partition is open, but the disk has been invalidated,
454 		 * disallow further opens of non-raw partition
455 		 */
456 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
457 		    (part != RAW_PART || fmt != S_IFCHR)) {
458 			error = EIO;
459 			goto bad2;
460 		}
461 	} else {
462 		int silent;
463 
464 		if (part == RAW_PART && fmt == S_IFCHR)
465 			silent = XS_CTL_SILENT;
466 		else
467 			silent = 0;
468 
469 		/* Check that it is still responding and ok. */
470 		error = scsipi_test_unit_ready(periph,
471 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
472 		    silent);
473 
474 		/*
475 		 * Start the pack spinning if necessary. Always allow the
476 		 * raw parition to be opened, for raw IOCTLs. Data transfers
477 		 * will check for SDEV_MEDIA_LOADED.
478 		 */
479 		if (error == EIO) {
480 			int error2;
481 
482 			error2 = scsipi_start(periph, SSS_START, silent);
483 			switch (error2) {
484 			case 0:
485 				error = 0;
486 				break;
487 			case EIO:
488 			case EINVAL:
489 				break;
490 			default:
491 				error = error2;
492 				break;
493 			}
494 		}
495 		if (error) {
496 			if (silent)
497 				goto out;
498 			goto bad2;
499 		}
500 
501 		periph->periph_flags |= PERIPH_OPEN;
502 
503 		if (periph->periph_flags & PERIPH_REMOVABLE) {
504 			/* Lock the pack in. */
505 			error = scsipi_prevent(periph, PR_PREVENT,
506 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
507 			    XS_CTL_IGNORE_MEDIA_CHANGE);
508 			if (error)
509 				goto bad3;
510 		}
511 
512 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
513 			int param_error;
514 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
515 
516 			/*
517 			 * Load the physical device parameters.
518 			 *
519 			 * Note that if media is present but unformatted,
520 			 * we allow the open (so that it can be formatted!).
521 			 * The drive should refuse real I/O, if the media is
522 			 * unformatted.
523 			 */
524 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
525 			     == SDGP_RESULT_OFFLINE) {
526 				error = ENXIO;
527 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
528 				goto bad3;
529 			}
530 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
531 
532 			/* Load the partition info if not already loaded. */
533 			if (param_error == 0) {
534 				sdgetdisklabel(sd);
535 				SC_DEBUG(periph, SCSIPI_DB3,
536 				     ("Disklabel loaded "));
537 			}
538 		}
539 	}
540 
541 	/* Check that the partition exists. */
542 	if (part != RAW_PART &&
543 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
544 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
545 		error = ENXIO;
546 		goto bad3;
547 	}
548 
549  out:	/* Insure only one open at a time. */
550 	switch (fmt) {
551 	case S_IFCHR:
552 		sd->sc_dk.dk_copenmask |= (1 << part);
553 		break;
554 	case S_IFBLK:
555 		sd->sc_dk.dk_bopenmask |= (1 << part);
556 		break;
557 	}
558 	sd->sc_dk.dk_openmask =
559 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
560 
561 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
562 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
563 	return (0);
564 
565  bad3:
566 	if (sd->sc_dk.dk_openmask == 0) {
567 		if (periph->periph_flags & PERIPH_REMOVABLE)
568 			scsipi_prevent(periph, PR_ALLOW,
569 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
570 			    XS_CTL_IGNORE_MEDIA_CHANGE);
571 		periph->periph_flags &= ~PERIPH_OPEN;
572 	}
573 
574  bad2:
575 	if (sd->sc_dk.dk_openmask == 0)
576 		scsipi_adapter_delref(adapt);
577 
578  bad1:
579 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
580 	return (error);
581 }
582 
583 /*
584  * close the device.. only called if we are the LAST occurence of an open
585  * device.  Convenient now but usually a pain.
586  */
587 static int
588 sdclose(dev_t dev, int flag, int fmt, struct proc *p)
589 {
590 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
591 	struct scsipi_periph *periph = sd->sc_periph;
592 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
593 	int part = SDPART(dev);
594 	int error;
595 
596 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
597 		return (error);
598 
599 	switch (fmt) {
600 	case S_IFCHR:
601 		sd->sc_dk.dk_copenmask &= ~(1 << part);
602 		break;
603 	case S_IFBLK:
604 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
605 		break;
606 	}
607 	sd->sc_dk.dk_openmask =
608 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
609 
610 	if (sd->sc_dk.dk_openmask == 0) {
611 		/*
612 		 * If the disk cache needs flushing, and the disk supports
613 		 * it, do it now.
614 		 */
615 		if ((sd->flags & SDF_DIRTY) != 0) {
616 			if (sd_flush(sd, 0)) {
617 				printf("%s: cache synchronization failed\n",
618 				    sd->sc_dev.dv_xname);
619 				sd->flags &= ~SDF_FLUSHING;
620 			} else
621 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
622 		}
623 
624 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
625 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
626 
627 		scsipi_wait_drain(periph);
628 
629 		if (periph->periph_flags & PERIPH_REMOVABLE)
630 			scsipi_prevent(periph, PR_ALLOW,
631 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
632 			    XS_CTL_IGNORE_NOT_READY);
633 		periph->periph_flags &= ~PERIPH_OPEN;
634 
635 		scsipi_wait_drain(periph);
636 
637 		scsipi_adapter_delref(adapt);
638 	}
639 
640 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
641 	return (0);
642 }
643 
644 /*
645  * Actually translate the requested transfer into one the physical driver
646  * can understand.  The transfer is described by a buf and will include
647  * only one physical transfer.
648  */
649 static void
650 sdstrategy(struct buf *bp)
651 {
652 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
653 	struct scsipi_periph *periph = sd->sc_periph;
654 	struct disklabel *lp;
655 	daddr_t blkno;
656 	int s;
657 	boolean_t sector_aligned;
658 
659 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
660 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
661 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
662 	/*
663 	 * If the device has been made invalid, error out
664 	 */
665 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
666 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
667 		if (periph->periph_flags & PERIPH_OPEN)
668 			bp->b_error = EIO;
669 		else
670 			bp->b_error = ENODEV;
671 		goto bad;
672 	}
673 
674 	lp = sd->sc_dk.dk_label;
675 
676 	/*
677 	 * The transfer must be a whole number of blocks, offset must not be
678 	 * negative.
679 	 */
680 	if (lp->d_secsize == DEV_BSIZE) {
681 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
682 	} else {
683 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
684 	}
685 	if (!sector_aligned || bp->b_blkno < 0) {
686 		bp->b_error = EINVAL;
687 		goto bad;
688 	}
689 	/*
690 	 * If it's a null transfer, return immediatly
691 	 */
692 	if (bp->b_bcount == 0)
693 		goto done;
694 
695 	/*
696 	 * Do bounds checking, adjust transfer. if error, process.
697 	 * If end of partition, just return.
698 	 */
699 	if (SDPART(bp->b_dev) == RAW_PART) {
700 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
701 		    sd->params.disksize512) <= 0)
702 			goto done;
703 	} else {
704 		if (bounds_check_with_label(&sd->sc_dk, bp,
705 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
706 			goto done;
707 	}
708 
709 	/*
710 	 * Now convert the block number to absolute and put it in
711 	 * terms of the device's logical block size.
712 	 */
713 	if (lp->d_secsize == DEV_BSIZE)
714 		blkno = bp->b_blkno;
715 	else if (lp->d_secsize > DEV_BSIZE)
716 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
717 	else
718 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
719 
720 	if (SDPART(bp->b_dev) != RAW_PART)
721 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
722 
723 	bp->b_rawblkno = blkno;
724 
725 	s = splbio();
726 
727 	/*
728 	 * Place it in the queue of disk activities for this disk.
729 	 *
730 	 * XXX Only do disksort() if the current operating mode does not
731 	 * XXX include tagged queueing.
732 	 */
733 	BUFQ_PUT(&sd->buf_queue, bp);
734 
735 	/*
736 	 * Tell the device to get going on the transfer if it's
737 	 * not doing anything, otherwise just wait for completion
738 	 */
739 	sdstart(sd->sc_periph);
740 
741 	splx(s);
742 	return;
743 
744 bad:
745 	bp->b_flags |= B_ERROR;
746 done:
747 	/*
748 	 * Correctly set the buf to indicate a completed xfer
749 	 */
750 	bp->b_resid = bp->b_bcount;
751 	biodone(bp);
752 }
753 
754 /*
755  * sdstart looks to see if there is a buf waiting for the device
756  * and that the device is not already busy. If both are true,
757  * It dequeues the buf and creates a scsi command to perform the
758  * transfer in the buf. The transfer request will call scsipi_done
759  * on completion, which will in turn call this routine again
760  * so that the next queued transfer is performed.
761  * The bufs are queued by the strategy routine (sdstrategy)
762  *
763  * This routine is also called after other non-queued requests
764  * have been made of the scsi driver, to ensure that the queue
765  * continues to be drained.
766  *
767  * must be called at the correct (highish) spl level
768  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
769  */
770 static void
771 sdstart(struct scsipi_periph *periph)
772 {
773 	struct sd_softc *sd = (void *)periph->periph_dev;
774 	struct disklabel *lp = sd->sc_dk.dk_label;
775 	struct buf *bp = 0;
776 	struct scsipi_rw_16 cmd16;
777 	struct scsipi_rw_10 cmd_big;
778 	struct scsi_rw_6 cmd_small;
779 	struct scsipi_generic *cmdp;
780 	struct scsipi_xfer *xs;
781 	int nblks, cmdlen, error, flags;
782 
783 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
784 	/*
785 	 * Check if the device has room for another command
786 	 */
787 	while (periph->periph_active < periph->periph_openings) {
788 		/*
789 		 * there is excess capacity, but a special waits
790 		 * It'll need the adapter as soon as we clear out of the
791 		 * way and let it run (user level wait).
792 		 */
793 		if (periph->periph_flags & PERIPH_WAITING) {
794 			periph->periph_flags &= ~PERIPH_WAITING;
795 			wakeup((caddr_t)periph);
796 			return;
797 		}
798 
799 		/*
800 		 * If the device has become invalid, abort all the
801 		 * reads and writes until all files have been closed and
802 		 * re-opened
803 		 */
804 		if (__predict_false(
805 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
806 			if ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
807 				bp->b_error = EIO;
808 				bp->b_flags |= B_ERROR;
809 				bp->b_resid = bp->b_bcount;
810 				biodone(bp);
811 				continue;
812 			} else {
813 				return;
814 			}
815 		}
816 
817 		/*
818 		 * See if there is a buf with work for us to do..
819 		 */
820 		if ((bp = BUFQ_PEEK(&sd->buf_queue)) == NULL)
821 			return;
822 
823 		/*
824 		 * We have a buf, now we should make a command.
825 		 */
826 
827 		if (lp->d_secsize == DEV_BSIZE)
828 			nblks = bp->b_bcount >> DEV_BSHIFT;
829 		else
830 			nblks = howmany(bp->b_bcount, lp->d_secsize);
831 
832 		/*
833 		 * Fill out the scsi command.  Use the smallest CDB possible
834 		 * (6-byte, 10-byte, or 16-byte).
835 		 */
836 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
837 		    ((nblks & 0xff) == nblks) &&
838 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
839 			/* 6-byte CDB */
840 			memset(&cmd_small, 0, sizeof(cmd_small));
841 			cmd_small.opcode = (bp->b_flags & B_READ) ?
842 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
843 			_lto3b(bp->b_rawblkno, cmd_small.addr);
844 			cmd_small.length = nblks & 0xff;
845 			cmdlen = sizeof(cmd_small);
846 			cmdp = (struct scsipi_generic *)&cmd_small;
847 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
848 			/* 10-byte CDB */
849 			memset(&cmd_big, 0, sizeof(cmd_big));
850 			cmd_big.opcode = (bp->b_flags & B_READ) ?
851 			    READ_10 : WRITE_10;
852 			_lto4b(bp->b_rawblkno, cmd_big.addr);
853 			_lto2b(nblks, cmd_big.length);
854 			cmdlen = sizeof(cmd_big);
855 			cmdp = (struct scsipi_generic *)&cmd_big;
856 		} else {
857 			/* 16-byte CDB */
858 			memset(&cmd16, 0, sizeof(cmd16));
859 			cmd16.opcode = (bp->b_flags & B_READ) ?
860 			    READ_16 : WRITE_16;
861 			_lto8b(bp->b_rawblkno, cmd16.addr);
862 			_lto4b(nblks, cmd16.length);
863 			cmdlen = sizeof(cmd16);
864 			cmdp = (struct scsipi_generic *)&cmd16;
865 		}
866 
867 		/* Instrumentation. */
868 		disk_busy(&sd->sc_dk);
869 
870 		/*
871 		 * Mark the disk dirty so that the cache will be
872 		 * flushed on close.
873 		 */
874 		if ((bp->b_flags & B_READ) == 0)
875 			sd->flags |= SDF_DIRTY;
876 
877 		/*
878 		 * Figure out what flags to use.
879 		 */
880 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
881 		if (bp->b_flags & B_READ)
882 			flags |= XS_CTL_DATA_IN;
883 		else
884 			flags |= XS_CTL_DATA_OUT;
885 
886 		/*
887 		 * Call the routine that chats with the adapter.
888 		 * Note: we cannot sleep as we may be an interrupt
889 		 */
890 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
891 		    (u_char *)bp->b_data, bp->b_bcount,
892 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
893 		if (__predict_false(xs == NULL)) {
894 			/*
895 			 * out of memory. Keep this buffer in the queue, and
896 			 * retry later.
897 			 */
898 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
899 			    periph);
900 			return;
901 		}
902 		/*
903 		 * need to dequeue the buffer before queuing the command,
904 		 * because cdstart may be called recursively from the
905 		 * HBA driver
906 		 */
907 #ifdef DIAGNOSTIC
908 		if (BUFQ_GET(&sd->buf_queue) != bp)
909 			panic("sdstart(): dequeued wrong buf");
910 #else
911 		BUFQ_GET(&sd->buf_queue);
912 #endif
913 		error = scsipi_execute_xs(xs);
914 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
915 		KASSERT(error == 0);
916 	}
917 }
918 
919 static void
920 sdrestart(void *v)
921 {
922 	int s = splbio();
923 	sdstart((struct scsipi_periph *)v);
924 	splx(s);
925 }
926 
927 static void
928 sddone(struct scsipi_xfer *xs, int error)
929 {
930 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
931 	struct buf *bp = xs->bp;
932 
933 	if (sd->flags & SDF_FLUSHING) {
934 		/* Flush completed, no longer dirty. */
935 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
936 	}
937 
938 	if (bp) {
939 		bp->b_error = error;
940 		bp->b_resid = xs->resid;
941 		if (error)
942 			bp->b_flags |= B_ERROR;
943 
944 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
945 		    (bp->b_flags & B_READ));
946 #if NRND > 0
947 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
948 #endif
949 
950 		biodone(bp);
951 	}
952 }
953 
954 static void
955 sdminphys(struct buf *bp)
956 {
957 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
958 	long max;
959 
960 	/*
961 	 * If the device is ancient, we want to make sure that
962 	 * the transfer fits into a 6-byte cdb.
963 	 *
964 	 * XXX Note that the SCSI-I spec says that 256-block transfers
965 	 * are allowed in a 6-byte read/write, and are specified
966 	 * by settng the "length" to 0.  However, we're conservative
967 	 * here, allowing only 255-block transfers in case an
968 	 * ancient device gets confused by length == 0.  A length of 0
969 	 * in a 10-byte read/write actually means 0 blocks.
970 	 */
971 	if ((sd->flags & SDF_ANCIENT) &&
972 	    ((sd->sc_periph->periph_flags &
973 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
974 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
975 
976 		if (bp->b_bcount > max)
977 			bp->b_bcount = max;
978 	}
979 
980 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
981 }
982 
983 static int
984 sdread(dev_t dev, struct uio *uio, int ioflag)
985 {
986 
987 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
988 }
989 
990 static int
991 sdwrite(dev_t dev, struct uio *uio, int ioflag)
992 {
993 
994 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
995 }
996 
997 /*
998  * Perform special action on behalf of the user
999  * Knows about the internals of this device
1000  */
1001 static int
1002 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct proc *p)
1003 {
1004 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1005 	struct scsipi_periph *periph = sd->sc_periph;
1006 	int part = SDPART(dev);
1007 	int error = 0;
1008 #ifdef __HAVE_OLD_DISKLABEL
1009 	struct disklabel *newlabel = NULL;
1010 #endif
1011 
1012 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1013 
1014 	/*
1015 	 * If the device is not valid, some IOCTLs can still be
1016 	 * handled on the raw partition. Check this here.
1017 	 */
1018 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1019 		switch (cmd) {
1020 		case DIOCKLABEL:
1021 		case DIOCWLABEL:
1022 		case DIOCLOCK:
1023 		case DIOCEJECT:
1024 		case ODIOCEJECT:
1025 		case DIOCGCACHE:
1026 		case DIOCSCACHE:
1027 		case SCIOCIDENTIFY:
1028 		case OSCIOCIDENTIFY:
1029 		case SCIOCCOMMAND:
1030 		case SCIOCDEBUG:
1031 			if (part == RAW_PART)
1032 				break;
1033 		/* FALLTHROUGH */
1034 		default:
1035 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1036 				return (ENODEV);
1037 			else
1038 				return (EIO);
1039 		}
1040 	}
1041 
1042 	switch (cmd) {
1043 	case DIOCGDINFO:
1044 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1045 		return (0);
1046 
1047 #ifdef __HAVE_OLD_DISKLABEL
1048 	case ODIOCGDINFO:
1049 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1050 		if (newlabel == NULL)
1051 			return EIO;
1052 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1053 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1054 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1055 		else
1056 			error = ENOTTY;
1057 		free(newlabel, M_TEMP);
1058 		return error;
1059 #endif
1060 
1061 	case DIOCGPART:
1062 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1063 		((struct partinfo *)addr)->part =
1064 		    &sd->sc_dk.dk_label->d_partitions[part];
1065 		return (0);
1066 
1067 	case DIOCWDINFO:
1068 	case DIOCSDINFO:
1069 #ifdef __HAVE_OLD_DISKLABEL
1070 	case ODIOCWDINFO:
1071 	case ODIOCSDINFO:
1072 #endif
1073 	{
1074 		struct disklabel *lp;
1075 
1076 		if ((flag & FWRITE) == 0)
1077 			return (EBADF);
1078 
1079 #ifdef __HAVE_OLD_DISKLABEL
1080  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1081 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1082 			if (newlabel == NULL)
1083 				return EIO;
1084 			memset(newlabel, 0, sizeof newlabel);
1085 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1086 			lp = newlabel;
1087 		} else
1088 #endif
1089 		lp = (struct disklabel *)addr;
1090 
1091 		if ((error = lockmgr(&sd->sc_dk.dk_openlock,
1092 				     LK_EXCLUSIVE, NULL)) != 0)
1093 			goto bad;
1094 		sd->flags |= SDF_LABELLING;
1095 
1096 		error = setdisklabel(sd->sc_dk.dk_label,
1097 		    lp, /*sd->sc_dk.dk_openmask : */0,
1098 		    sd->sc_dk.dk_cpulabel);
1099 		if (error == 0) {
1100 			if (cmd == DIOCWDINFO
1101 #ifdef __HAVE_OLD_DISKLABEL
1102 			    || cmd == ODIOCWDINFO
1103 #endif
1104 			   )
1105 				error = writedisklabel(SDLABELDEV(dev),
1106 				    sdstrategy, sd->sc_dk.dk_label,
1107 				    sd->sc_dk.dk_cpulabel);
1108 		}
1109 
1110 		sd->flags &= ~SDF_LABELLING;
1111 		(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1112 bad:
1113 #ifdef __HAVE_OLD_DISKLABEL
1114 		if (newlabel != NULL)
1115 			free(newlabel, M_TEMP);
1116 #endif
1117 		return (error);
1118 	}
1119 
1120 	case DIOCKLABEL:
1121 		if (*(int *)addr)
1122 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1123 		else
1124 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1125 		return (0);
1126 
1127 	case DIOCWLABEL:
1128 		if ((flag & FWRITE) == 0)
1129 			return (EBADF);
1130 		if (*(int *)addr)
1131 			sd->flags |= SDF_WLABEL;
1132 		else
1133 			sd->flags &= ~SDF_WLABEL;
1134 		return (0);
1135 
1136 	case DIOCLOCK:
1137 		return (scsipi_prevent(periph,
1138 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1139 
1140 	case DIOCEJECT:
1141 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1142 			return (ENOTTY);
1143 		if (*(int *)addr == 0) {
1144 			/*
1145 			 * Don't force eject: check that we are the only
1146 			 * partition open. If so, unlock it.
1147 			 */
1148 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1149 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1150 			    sd->sc_dk.dk_openmask) {
1151 				error = scsipi_prevent(periph, PR_ALLOW,
1152 				    XS_CTL_IGNORE_NOT_READY);
1153 				if (error)
1154 					return (error);
1155 			} else {
1156 				return (EBUSY);
1157 			}
1158 		}
1159 		/* FALLTHROUGH */
1160 	case ODIOCEJECT:
1161 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1162 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1163 
1164 	case DIOCGDEFLABEL:
1165 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1166 		return (0);
1167 
1168 #ifdef __HAVE_OLD_DISKLABEL
1169 	case ODIOCGDEFLABEL:
1170 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1171 		if (newlabel == NULL)
1172 			return EIO;
1173 		sdgetdefaultlabel(sd, newlabel);
1174 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1175 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1176 		else
1177 			error = ENOTTY;
1178 		free(newlabel, M_TEMP);
1179 		return error;
1180 #endif
1181 
1182 	case DIOCGCACHE:
1183 		return (sd_getcache(sd, (int *) addr));
1184 
1185 	case DIOCSCACHE:
1186 		if ((flag & FWRITE) == 0)
1187 			return (EBADF);
1188 		return (sd_setcache(sd, *(int *) addr));
1189 
1190 	case DIOCCACHESYNC:
1191 		/*
1192 		 * XXX Do we really need to care about having a writable
1193 		 * file descriptor here?
1194 		 */
1195 		if ((flag & FWRITE) == 0)
1196 			return (EBADF);
1197 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1198 			error = sd_flush(sd, 0);
1199 			if (error)
1200 				sd->flags &= ~SDF_FLUSHING;
1201 			else
1202 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1203 		} else
1204 			error = 0;
1205 		return (error);
1206 
1207 	case DIOCAWEDGE:
1208 	    {
1209 	    	struct dkwedge_info *dkw = (void *) addr;
1210 
1211 		if ((flag & FWRITE) == 0)
1212 			return (EBADF);
1213 
1214 		/* If the ioctl happens here, the parent is us. */
1215 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1216 		return (dkwedge_add(dkw));
1217 	    }
1218 
1219 	case DIOCDWEDGE:
1220 	    {
1221 	    	struct dkwedge_info *dkw = (void *) addr;
1222 
1223 		if ((flag & FWRITE) == 0)
1224 			return (EBADF);
1225 
1226 		/* If the ioctl happens here, the parent is us. */
1227 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1228 		return (dkwedge_del(dkw));
1229 	    }
1230 
1231 	case DIOCLWEDGES:
1232 	    {
1233 	    	struct dkwedge_list *dkwl = (void *) addr;
1234 
1235 		return (dkwedge_list(&sd->sc_dk, dkwl, p));
1236 	    }
1237 
1238 	default:
1239 		if (part != RAW_PART)
1240 			return (ENOTTY);
1241 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1242 	}
1243 
1244 #ifdef DIAGNOSTIC
1245 	panic("sdioctl: impossible");
1246 #endif
1247 }
1248 
1249 static void
1250 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1251 {
1252 
1253 	memset(lp, 0, sizeof(struct disklabel));
1254 
1255 	lp->d_secsize = sd->params.blksize;
1256 	lp->d_ntracks = sd->params.heads;
1257 	lp->d_nsectors = sd->params.sectors;
1258 	lp->d_ncylinders = sd->params.cyls;
1259 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1260 
1261 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1262 	case SCSIPI_BUSTYPE_SCSI:
1263 		lp->d_type = DTYPE_SCSI;
1264 		break;
1265 	case SCSIPI_BUSTYPE_ATAPI:
1266 		lp->d_type = DTYPE_ATAPI;
1267 		break;
1268 	}
1269 	/*
1270 	 * XXX
1271 	 * We could probe the mode pages to figure out what kind of disc it is.
1272 	 * Is this worthwhile?
1273 	 */
1274 	strncpy(lp->d_typename, "mydisk", 16);
1275 	strncpy(lp->d_packname, "fictitious", 16);
1276 	lp->d_secperunit = sd->params.disksize;
1277 	lp->d_rpm = sd->params.rot_rate;
1278 	lp->d_interleave = 1;
1279 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1280 	    D_REMOVABLE : 0;
1281 
1282 	lp->d_partitions[RAW_PART].p_offset = 0;
1283 	lp->d_partitions[RAW_PART].p_size =
1284 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1285 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1286 	lp->d_npartitions = RAW_PART + 1;
1287 
1288 	lp->d_magic = DISKMAGIC;
1289 	lp->d_magic2 = DISKMAGIC;
1290 	lp->d_checksum = dkcksum(lp);
1291 }
1292 
1293 
1294 /*
1295  * Load the label information on the named device
1296  */
1297 static void
1298 sdgetdisklabel(struct sd_softc *sd)
1299 {
1300 	struct disklabel *lp = sd->sc_dk.dk_label;
1301 	const char *errstring;
1302 
1303 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1304 
1305 	sdgetdefaultlabel(sd, lp);
1306 
1307 	if (lp->d_secpercyl == 0) {
1308 		lp->d_secpercyl = 100;
1309 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1310 	}
1311 
1312 	/*
1313 	 * Call the generic disklabel extraction routine
1314 	 */
1315 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1316 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1317 	if (errstring) {
1318 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1319 		return;
1320 	}
1321 }
1322 
1323 static void
1324 sd_shutdown(void *arg)
1325 {
1326 	struct sd_softc *sd = arg;
1327 
1328 	/*
1329 	 * If the disk cache needs to be flushed, and the disk supports
1330 	 * it, flush it.  We're cold at this point, so we poll for
1331 	 * completion.
1332 	 */
1333 	if ((sd->flags & SDF_DIRTY) != 0) {
1334 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1335 			printf("%s: cache synchronization failed\n",
1336 			    sd->sc_dev.dv_xname);
1337 			sd->flags &= ~SDF_FLUSHING;
1338 		} else
1339 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1340 	}
1341 }
1342 
1343 /*
1344  * Check Errors
1345  */
1346 static int
1347 sd_interpret_sense(struct scsipi_xfer *xs)
1348 {
1349 	struct scsipi_periph *periph = xs->xs_periph;
1350 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1351 	struct sd_softc *sd = (void *)periph->periph_dev;
1352 	int s, error, retval = EJUSTRETURN;
1353 
1354 	/*
1355 	 * If the periph is already recovering, just do the normal
1356 	 * error processing.
1357 	 */
1358 	if (periph->periph_flags & PERIPH_RECOVERING)
1359 		return (retval);
1360 
1361 	/*
1362 	 * If the device is not open yet, let the generic code handle it.
1363 	 */
1364 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1365 		return (retval);
1366 
1367 	/*
1368 	 * If it isn't a extended or extended/deferred error, let
1369 	 * the generic code handle it.
1370 	 */
1371 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1372 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1373 		return (retval);
1374 
1375 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1376 	    sense->add_sense_code == 0x4) {
1377 		if (sense->add_sense_code_qual == 0x01)	{
1378 			/*
1379 			 * Unit In The Process Of Becoming Ready.
1380 			 */
1381 			printf("%s: waiting for pack to spin up...\n",
1382 			    sd->sc_dev.dv_xname);
1383 			if (!callout_pending(&periph->periph_callout))
1384 				scsipi_periph_freeze(periph, 1);
1385 			callout_reset(&periph->periph_callout,
1386 			    5 * hz, scsipi_periph_timed_thaw, periph);
1387 			retval = ERESTART;
1388 		} else if (sense->add_sense_code_qual == 0x02) {
1389 			printf("%s: pack is stopped, restarting...\n",
1390 			    sd->sc_dev.dv_xname);
1391 			s = splbio();
1392 			periph->periph_flags |= PERIPH_RECOVERING;
1393 			splx(s);
1394 			error = scsipi_start(periph, SSS_START,
1395 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1396 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1397 			if (error) {
1398 				printf("%s: unable to restart pack\n",
1399 				    sd->sc_dev.dv_xname);
1400 				retval = error;
1401 			} else
1402 				retval = ERESTART;
1403 			s = splbio();
1404 			periph->periph_flags &= ~PERIPH_RECOVERING;
1405 			splx(s);
1406 		}
1407 	}
1408 	if ((sense->flags & SSD_KEY) == SKEY_MEDIUM_ERROR &&
1409 	    sense->add_sense_code == 0x31 &&
1410 	    sense->add_sense_code_qual == 0x00)	{ /* maybe for any asq ? */
1411 		/* Medium Format Corrupted */
1412 		retval = EFTYPE;
1413 	}
1414 	return (retval);
1415 }
1416 
1417 
1418 static int
1419 sdsize(dev_t dev)
1420 {
1421 	struct sd_softc *sd;
1422 	int part, unit, omask;
1423 	int size;
1424 
1425 	unit = SDUNIT(dev);
1426 	if (unit >= sd_cd.cd_ndevs)
1427 		return (-1);
1428 	sd = sd_cd.cd_devs[unit];
1429 	if (sd == NULL)
1430 		return (-1);
1431 
1432 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1433 		return (-1);
1434 
1435 	part = SDPART(dev);
1436 	omask = sd->sc_dk.dk_openmask & (1 << part);
1437 
1438 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1439 		return (-1);
1440 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1441 		size = -1;
1442 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1443 		size = -1;
1444 	else
1445 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1446 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1447 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1448 		return (-1);
1449 	return (size);
1450 }
1451 
1452 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1453 static struct scsipi_xfer sx;
1454 static int sddoingadump;
1455 
1456 /*
1457  * dump all of physical memory into the partition specified, starting
1458  * at offset 'dumplo' into the partition.
1459  */
1460 static int
1461 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1462 {
1463 	struct sd_softc *sd;	/* disk unit to do the I/O */
1464 	struct disklabel *lp;	/* disk's disklabel */
1465 	int	unit, part;
1466 	int	sectorsize;	/* size of a disk sector */
1467 	int	nsects;		/* number of sectors in partition */
1468 	int	sectoff;	/* sector offset of partition */
1469 	int	totwrt;		/* total number of sectors left to write */
1470 	int	nwrt;		/* current number of sectors to write */
1471 	struct scsipi_rw_10 cmd;	/* write command */
1472 	struct scsipi_xfer *xs;	/* ... convenience */
1473 	struct scsipi_periph *periph;
1474 	struct scsipi_channel *chan;
1475 
1476 	/* Check if recursive dump; if so, punt. */
1477 	if (sddoingadump)
1478 		return (EFAULT);
1479 
1480 	/* Mark as active early. */
1481 	sddoingadump = 1;
1482 
1483 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1484 	part = SDPART(dev);
1485 
1486 	/* Check for acceptable drive number. */
1487 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1488 		return (ENXIO);
1489 
1490 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1491 		return (ENODEV);
1492 
1493 	periph = sd->sc_periph;
1494 	chan = periph->periph_channel;
1495 
1496 	/* Make sure it was initialized. */
1497 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1498 		return (ENXIO);
1499 
1500 	/* Convert to disk sectors.  Request must be a multiple of size. */
1501 	lp = sd->sc_dk.dk_label;
1502 	sectorsize = lp->d_secsize;
1503 	if ((size % sectorsize) != 0)
1504 		return (EFAULT);
1505 	totwrt = size / sectorsize;
1506 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1507 
1508 	nsects = lp->d_partitions[part].p_size;
1509 	sectoff = lp->d_partitions[part].p_offset;
1510 
1511 	/* Check transfer bounds against partition size. */
1512 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1513 		return (EINVAL);
1514 
1515 	/* Offset block number to start of partition. */
1516 	blkno += sectoff;
1517 
1518 	xs = &sx;
1519 
1520 	while (totwrt > 0) {
1521 		nwrt = totwrt;		/* XXX */
1522 #ifndef	SD_DUMP_NOT_TRUSTED
1523 		/*
1524 		 *  Fill out the scsi command
1525 		 */
1526 		memset(&cmd, 0, sizeof(cmd));
1527 		cmd.opcode = WRITE_10;
1528 		_lto4b(blkno, cmd.addr);
1529 		_lto2b(nwrt, cmd.length);
1530 		/*
1531 		 * Fill out the scsipi_xfer structure
1532 		 *    Note: we cannot sleep as we may be an interrupt
1533 		 * don't use scsipi_command() as it may want to wait
1534 		 * for an xs.
1535 		 */
1536 		memset(xs, 0, sizeof(sx));
1537 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1538 		    XS_CTL_DATA_OUT;
1539 		xs->xs_status = 0;
1540 		xs->xs_periph = periph;
1541 		xs->xs_retries = SDRETRIES;
1542 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1543 		xs->cmd = (struct scsipi_generic *)&cmd;
1544 		xs->cmdlen = sizeof(cmd);
1545 		xs->resid = nwrt * sectorsize;
1546 		xs->error = XS_NOERROR;
1547 		xs->bp = 0;
1548 		xs->data = va;
1549 		xs->datalen = nwrt * sectorsize;
1550 
1551 		/*
1552 		 * Pass all this info to the scsi driver.
1553 		 */
1554 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1555 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1556 		    xs->error != XS_NOERROR)
1557 			return (EIO);
1558 #else	/* SD_DUMP_NOT_TRUSTED */
1559 		/* Let's just talk about this first... */
1560 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1561 		delay(500 * 1000);	/* half a second */
1562 #endif	/* SD_DUMP_NOT_TRUSTED */
1563 
1564 		/* update block count */
1565 		totwrt -= nwrt;
1566 		blkno += nwrt;
1567 		va += sectorsize * nwrt;
1568 	}
1569 	sddoingadump = 0;
1570 	return (0);
1571 }
1572 
1573 static int
1574 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1575     int page, int flags, int *big)
1576 {
1577 
1578 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1579 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1580 		*big = 1;
1581 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1582 		    size + sizeof(struct scsipi_mode_header_big),
1583 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1584 	} else {
1585 		*big = 0;
1586 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1587 		    size + sizeof(struct scsipi_mode_header),
1588 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1589 	}
1590 }
1591 
1592 static int
1593 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1594     int flags, int big)
1595 {
1596 
1597 	if (big) {
1598 		struct scsipi_mode_header_big *header = sense;
1599 
1600 		_lto2b(0, header->data_length);
1601 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1602 		    size + sizeof(struct scsipi_mode_header_big),
1603 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1604 	} else {
1605 		struct scsipi_mode_header *header = sense;
1606 
1607 		header->data_length = 0;
1608 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1609 		    size + sizeof(struct scsipi_mode_header),
1610 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1611 	}
1612 }
1613 
1614 static int
1615 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1616 {
1617 	struct {
1618 		struct scsipi_mode_header header;
1619 		/* no block descriptor */
1620 		u_int8_t pg_code; /* page code (should be 6) */
1621 		u_int8_t pg_length; /* page length (should be 11) */
1622 		u_int8_t wcd; /* bit0: cache disable */
1623 		u_int8_t lbs[2]; /* logical block size */
1624 		u_int8_t size[5]; /* number of log. blocks */
1625 		u_int8_t pp; /* power/performance */
1626 		u_int8_t flags;
1627 		u_int8_t resvd;
1628 	} scsipi_sense;
1629 	u_int64_t sectors;
1630 	int error;
1631 
1632 	/*
1633 	 * scsipi_size (ie "read capacity") and mode sense page 6
1634 	 * give the same information. Do both for now, and check
1635 	 * for consistency.
1636 	 * XXX probably differs for removable media
1637 	 */
1638 	dp->blksize = 512;
1639 	if ((sectors = scsipi_size(sd->sc_periph, flags)) == 0)
1640 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1641 
1642 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1643 	    &scsipi_sense.header, sizeof(scsipi_sense),
1644 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1645 
1646 	if (error != 0)
1647 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1648 
1649 	dp->blksize = _2btol(scsipi_sense.lbs);
1650 	if (dp->blksize == 0)
1651 		dp->blksize = 512;
1652 
1653 	/*
1654 	 * Create a pseudo-geometry.
1655 	 */
1656 	dp->heads = 64;
1657 	dp->sectors = 32;
1658 	dp->cyls = sectors / (dp->heads * dp->sectors);
1659 	dp->disksize = _5btol(scsipi_sense.size);
1660 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1661 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1662 		       (unsigned long long)dp->disksize,
1663 		       (unsigned long long)sectors);
1664 		dp->disksize = sectors;
1665 	}
1666 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1667 
1668 	return (SDGP_RESULT_OK);
1669 }
1670 
1671 /*
1672  * Get the scsi driver to send a full inquiry to the * device and use the
1673  * results to fill out the disk parameter structure.
1674  */
1675 static int
1676 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1677 {
1678 	u_int64_t sectors;
1679 	int error;
1680 #if 0
1681 	int i;
1682 	u_int8_t *p;
1683 #endif
1684 
1685 	dp->disksize = sectors = scsipi_size(sd->sc_periph, flags);
1686 	if (sectors == 0) {
1687 		struct scsipi_read_format_capacities cmd;
1688 		struct {
1689 			struct scsipi_capacity_list_header header;
1690 			struct scsipi_capacity_descriptor desc;
1691 		} __attribute__((packed)) data;
1692 
1693 		memset(&cmd, 0, sizeof(cmd));
1694 		memset(&data, 0, sizeof(data));
1695 		cmd.opcode = READ_FORMAT_CAPACITIES;
1696 		_lto2b(sizeof(data), cmd.length);
1697 
1698 		error = scsipi_command(sd->sc_periph,
1699 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1700 		    SDRETRIES, 20000, NULL,
1701 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1702 		if (error == EFTYPE) {
1703 			/* Medium Format Corrupted, handle as not formatted */
1704 			return (SDGP_RESULT_UNFORMATTED);
1705 		}
1706 		if (error || data.header.length == 0)
1707 			return (SDGP_RESULT_OFFLINE);
1708 
1709 #if 0
1710 printf("rfc: length=%d\n", data.header.length);
1711 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");
1712 #endif
1713 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1714 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1715 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1716 			break;
1717 
1718 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1719 			return (SDGP_RESULT_UNFORMATTED);
1720 
1721 		case SCSIPI_CAP_DESC_CODE_NONE:
1722 			return (SDGP_RESULT_OFFLINE);
1723 		}
1724 
1725 		dp->disksize = sectors = _4btol(data.desc.nblks);
1726 		if (sectors == 0)
1727 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1728 
1729 		dp->blksize = _3btol(data.desc.blklen);
1730 		if (dp->blksize == 0)
1731 			dp->blksize = 512;
1732 	} else {
1733 		struct sd_mode_sense_data scsipi_sense;
1734 		int big, bsize;
1735 		struct scsi_blk_desc *bdesc;
1736 
1737 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1738 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1739 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1740 		dp->blksize = 512;
1741 		if (!error) {
1742 			if (big) {
1743 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1744 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1745 			} else {
1746 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1747 				bsize = scsipi_sense.header.small.blk_desc_len;
1748 			}
1749 
1750 #if 0
1751 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1752 printf("page 0 bsize=%d\n", bsize);
1753 printf("page 0 ok\n");
1754 #endif
1755 
1756 			if (bsize >= 8) {
1757 				dp->blksize = _3btol(bdesc->blklen);
1758 				if (dp->blksize == 0)
1759 					dp->blksize = 512;
1760 			}
1761 		}
1762 	}
1763 
1764 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1765 	return (0);
1766 }
1767 
1768 static int
1769 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1770 {
1771 	struct sd_mode_sense_data scsipi_sense;
1772 	int error;
1773 	int big, poffset, byte2;
1774 	union scsi_disk_pages *pages;
1775 #if 0
1776 	int i;
1777 	u_int8_t *p;
1778 #endif
1779 
1780 	byte2 = SMS_DBD;
1781 again:
1782 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1783 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1784 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1785 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1786 	    flags | XS_CTL_SILENT, &big);
1787 	if (error) {
1788 		if (byte2 == SMS_DBD) {
1789 			/* No result; try once more with DBD off */
1790 			byte2 = 0;
1791 			goto again;
1792 		}
1793 		return (error);
1794 	}
1795 
1796 	if (big) {
1797 		poffset = sizeof scsipi_sense.header.big;
1798 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1799 	} else {
1800 		poffset = sizeof scsipi_sense.header.small;
1801 		poffset += scsipi_sense.header.small.blk_desc_len;
1802 	}
1803 
1804 	pages = (void *)((u_long)&scsipi_sense + poffset);
1805 #if 0
1806 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1807 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1808 #endif
1809 
1810 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1811 		return (ERESTART);
1812 
1813 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1814 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1815 	    _3btol(pages->rigid_geometry.ncyl),
1816 	    pages->rigid_geometry.nheads,
1817 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1818 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1819 	    _2btol(pages->rigid_geometry.land_zone)));
1820 
1821 	/*
1822 	 * KLUDGE!! (for zone recorded disks)
1823 	 * give a number of sectors so that sec * trks * cyls
1824 	 * is <= disk_size
1825 	 * can lead to wasted space! THINK ABOUT THIS !
1826 	 */
1827 	dp->heads = pages->rigid_geometry.nheads;
1828 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1829 	if (dp->heads == 0 || dp->cyls == 0)
1830 		return (ERESTART);
1831 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1832 
1833 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1834 	if (dp->rot_rate == 0)
1835 		dp->rot_rate = 3600;
1836 
1837 #if 0
1838 printf("page 4 ok\n");
1839 #endif
1840 	return (0);
1841 }
1842 
1843 static int
1844 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1845 {
1846 	struct sd_mode_sense_data scsipi_sense;
1847 	int error;
1848 	int big, poffset, byte2;
1849 	union scsi_disk_pages *pages;
1850 #if 0
1851 	int i;
1852 	u_int8_t *p;
1853 #endif
1854 
1855 	byte2 = SMS_DBD;
1856 again:
1857 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1858 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1859 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1860 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1861 	    flags | XS_CTL_SILENT, &big);
1862 	if (error) {
1863 		if (byte2 == SMS_DBD) {
1864 			/* No result; try once more with DBD off */
1865 			byte2 = 0;
1866 			goto again;
1867 		}
1868 		return (error);
1869 	}
1870 
1871 	if (big) {
1872 		poffset = sizeof scsipi_sense.header.big;
1873 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1874 	} else {
1875 		poffset = sizeof scsipi_sense.header.small;
1876 		poffset += scsipi_sense.header.small.blk_desc_len;
1877 	}
1878 
1879 	pages = (void *)((u_long)&scsipi_sense + poffset);
1880 #if 0
1881 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1882 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1883 #endif
1884 
1885 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1886 		return (ERESTART);
1887 
1888 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1889 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1890 	    _3btol(pages->flex_geometry.ncyl),
1891 	    pages->flex_geometry.nheads,
1892 	    pages->flex_geometry.ph_sec_tr,
1893 	    _2btol(pages->flex_geometry.bytes_s)));
1894 
1895 	dp->heads = pages->flex_geometry.nheads;
1896 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1897 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1898 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1899 		return (ERESTART);
1900 
1901 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1902 	if (dp->rot_rate == 0)
1903 		dp->rot_rate = 3600;
1904 
1905 #if 0
1906 printf("page 5 ok\n");
1907 #endif
1908 	return (0);
1909 }
1910 
1911 static int
1912 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1913 {
1914 	int error;
1915 
1916 	/*
1917 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1918 	 * cleared by the caller if necessary.
1919 	 */
1920 	if (sd->type == T_SIMPLE_DIRECT)
1921 		return (sd_get_simplifiedparms(sd, dp, flags));
1922 
1923 	error = sd_get_capacity(sd, dp, flags);
1924 	if (error)
1925 		return (error);
1926 
1927 	if (sd->type == T_OPTICAL)
1928 		goto page0;
1929 
1930 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1931 		if (!sd_get_parms_page5(sd, dp, flags) ||
1932 		    !sd_get_parms_page4(sd, dp, flags))
1933 			return (SDGP_RESULT_OK);
1934 	} else {
1935 		if (!sd_get_parms_page4(sd, dp, flags) ||
1936 		    !sd_get_parms_page5(sd, dp, flags))
1937 			return (SDGP_RESULT_OK);
1938 	}
1939 
1940 page0:
1941 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1942 	/* Try calling driver's method for figuring out geometry. */
1943 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1944 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1945 		(sd->sc_periph, dp, dp->disksize)) {
1946 		/*
1947 		 * Use adaptec standard fictitious geometry
1948 		 * this depends on which controller (e.g. 1542C is
1949 		 * different. but we have to put SOMETHING here..)
1950 		 */
1951 		dp->heads = 64;
1952 		dp->sectors = 32;
1953 		dp->cyls = dp->disksize / (64 * 32);
1954 	}
1955 	dp->rot_rate = 3600;
1956 	return (SDGP_RESULT_OK);
1957 }
1958 
1959 static int
1960 sd_flush(struct sd_softc *sd, int flags)
1961 {
1962 	struct scsipi_periph *periph = sd->sc_periph;
1963 	struct scsi_synchronize_cache_10 cmd;
1964 
1965 	/*
1966 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1967 	 * We issue with address 0 length 0, which should be
1968 	 * interpreted by the device as "all remaining blocks
1969 	 * starting at address 0".  We ignore ILLEGAL REQUEST
1970 	 * in the event that the command is not supported by
1971 	 * the device, and poll for completion so that we know
1972 	 * that the cache has actually been flushed.
1973 	 *
1974 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
1975 	 * command, as indicated by our quirks flags.
1976 	 *
1977 	 * XXX What about older devices?
1978 	 */
1979 	if (periph->periph_version < 2 ||
1980 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
1981 		return (0);
1982 
1983 	sd->flags |= SDF_FLUSHING;
1984 	memset(&cmd, 0, sizeof(cmd));
1985 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
1986 
1987 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
1988 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
1989 }
1990 
1991 static int
1992 sd_getcache(struct sd_softc *sd, int *bitsp)
1993 {
1994 	struct scsipi_periph *periph = sd->sc_periph;
1995 	struct sd_mode_sense_data scsipi_sense;
1996 	int error, bits = 0;
1997 	int big;
1998 	union scsi_disk_pages *pages;
1999 
2000 	if (periph->periph_version < 2)
2001 		return (EOPNOTSUPP);
2002 
2003 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2004 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2005 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2006 	if (error)
2007 		return (error);
2008 
2009 	if (big)
2010 		pages = (void *)(&scsipi_sense.header.big + 1);
2011 	else
2012 		pages = (void *)(&scsipi_sense.header.small + 1);
2013 
2014 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2015 		bits |= DKCACHE_READ;
2016 	if (pages->caching_params.flags & CACHING_WCE)
2017 		bits |= DKCACHE_WRITE;
2018 	if (pages->caching_params.pg_code & PGCODE_PS)
2019 		bits |= DKCACHE_SAVE;
2020 
2021 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2022 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2023 	    sizeof(scsipi_sense.pages.caching_params),
2024 	    SMS_PAGE_CTRL_CHANGEABLE|8, 0, &big);
2025 	if (error == 0) {
2026 		if (big)
2027 			pages = (void *)(&scsipi_sense.header.big + 1);
2028 		else
2029 			pages = (void *)(&scsipi_sense.header.small + 1);
2030 
2031 		if (pages->caching_params.flags & CACHING_RCD)
2032 			bits |= DKCACHE_RCHANGE;
2033 		if (pages->caching_params.flags & CACHING_WCE)
2034 			bits |= DKCACHE_WCHANGE;
2035 	}
2036 
2037 	*bitsp = bits;
2038 
2039 	return (0);
2040 }
2041 
2042 static int
2043 sd_setcache(struct sd_softc *sd, int bits)
2044 {
2045 	struct scsipi_periph *periph = sd->sc_periph;
2046 	struct sd_mode_sense_data scsipi_sense;
2047 	int error;
2048 	uint8_t oflags, byte2 = 0;
2049 	int big;
2050 	union scsi_disk_pages *pages;
2051 
2052 	if (periph->periph_version < 2)
2053 		return (EOPNOTSUPP);
2054 
2055 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2056 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2057 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2058 	if (error)
2059 		return (error);
2060 
2061 	if (big)
2062 		pages = (void *)(&scsipi_sense.header.big + 1);
2063 	else
2064 		pages = (void *)(&scsipi_sense.header.small + 1);
2065 
2066 	oflags = pages->caching_params.flags;
2067 
2068 	if (bits & DKCACHE_READ)
2069 		pages->caching_params.flags &= ~CACHING_RCD;
2070 	else
2071 		pages->caching_params.flags |= CACHING_RCD;
2072 
2073 	if (bits & DKCACHE_WRITE)
2074 		pages->caching_params.flags |= CACHING_WCE;
2075 	else
2076 		pages->caching_params.flags &= ~CACHING_WCE;
2077 
2078 	if (oflags == pages->caching_params.flags)
2079 		return (0);
2080 
2081 	pages->caching_params.pg_code &= PGCODE_MASK;
2082 
2083 	if (bits & DKCACHE_SAVE)
2084 		byte2 |= SMS_SP;
2085 
2086 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2087 	    sizeof(struct scsipi_mode_page_header) +
2088 	    pages->caching_params.pg_length, 0, big));
2089 }
2090