xref: /netbsd-src/sys/dev/scsipi/sd.c (revision df0caa2637da0538ecdf6b878c4d08e684b43d8f)
1 /*	$NetBSD: sd.c,v 1.240 2005/05/29 22:00:50 christos 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 1.240 2005/05/29 22:00:50 christos 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 	    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 	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 	bufq_drain(&sd->buf_queue);
373 
374 	bufq_free(&sd->buf_queue);
375 
376 	/* Kill off any pending commands. */
377 	scsipi_kill_pending(sd->sc_periph);
378 
379 	splx(s);
380 
381 	/* Detach from the disk list. */
382 	disk_detach(&sd->sc_dk);
383 
384 	/* Get rid of the shutdown hook. */
385 	shutdownhook_disestablish(sd->sc_sdhook);
386 
387 #if NRND > 0
388 	/* Unhook the entropy source. */
389 	rnd_detach_source(&sd->rnd_source);
390 #endif
391 
392 	return (0);
393 }
394 
395 /*
396  * open the device. Make sure the partition info is a up-to-date as can be.
397  */
398 static int
399 sdopen(dev_t dev, int flag, int fmt, struct proc *p)
400 {
401 	struct sd_softc *sd;
402 	struct scsipi_periph *periph;
403 	struct scsipi_adapter *adapt;
404 	int unit, part;
405 	int error;
406 
407 	unit = SDUNIT(dev);
408 	if (unit >= sd_cd.cd_ndevs)
409 		return (ENXIO);
410 	sd = sd_cd.cd_devs[unit];
411 	if (sd == NULL)
412 		return (ENXIO);
413 
414 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
415 		return (ENODEV);
416 
417 	part = SDPART(dev);
418 
419 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
420 		return (error);
421 
422 	/*
423 	 * If there are wedges, and this is not RAW_PART, then we
424 	 * need to fail.
425 	 */
426 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
427 		error = EBUSY;
428 		goto bad1;
429 	}
430 
431 	periph = sd->sc_periph;
432 	adapt = periph->periph_channel->chan_adapter;
433 
434 	SC_DEBUG(periph, SCSIPI_DB1,
435 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
436 	    sd_cd.cd_ndevs, part));
437 
438 	/*
439 	 * If this is the first open of this device, add a reference
440 	 * to the adapter.
441 	 */
442 	if (sd->sc_dk.dk_openmask == 0 &&
443 	    (error = scsipi_adapter_addref(adapt)) != 0)
444 		goto bad1;
445 
446 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
447 		/*
448 		 * If any partition is open, but the disk has been invalidated,
449 		 * disallow further opens of non-raw partition
450 		 */
451 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
452 		    (part != RAW_PART || fmt != S_IFCHR)) {
453 			error = EIO;
454 			goto bad2;
455 		}
456 	} else {
457 		int silent;
458 
459 		if (part == RAW_PART && fmt == S_IFCHR)
460 			silent = XS_CTL_SILENT;
461 		else
462 			silent = 0;
463 
464 		/* Check that it is still responding and ok. */
465 		error = scsipi_test_unit_ready(periph,
466 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
467 		    silent);
468 
469 		/*
470 		 * Start the pack spinning if necessary. Always allow the
471 		 * raw parition to be opened, for raw IOCTLs. Data transfers
472 		 * will check for SDEV_MEDIA_LOADED.
473 		 */
474 		if (error == EIO) {
475 			int error2;
476 
477 			error2 = scsipi_start(periph, SSS_START, silent);
478 			switch (error2) {
479 			case 0:
480 				error = 0;
481 				break;
482 			case EIO:
483 			case EINVAL:
484 				break;
485 			default:
486 				error = error2;
487 				break;
488 			}
489 		}
490 		if (error) {
491 			if (silent)
492 				goto out;
493 			goto bad2;
494 		}
495 
496 		periph->periph_flags |= PERIPH_OPEN;
497 
498 		if (periph->periph_flags & PERIPH_REMOVABLE) {
499 			/* Lock the pack in. */
500 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
501 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
502 			    XS_CTL_IGNORE_MEDIA_CHANGE);
503 			if (error)
504 				goto bad3;
505 		}
506 
507 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
508 			int param_error;
509 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
510 
511 			/*
512 			 * Load the physical device parameters.
513 			 *
514 			 * Note that if media is present but unformatted,
515 			 * we allow the open (so that it can be formatted!).
516 			 * The drive should refuse real I/O, if the media is
517 			 * unformatted.
518 			 */
519 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
520 			     == SDGP_RESULT_OFFLINE) {
521 				error = ENXIO;
522 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
523 				goto bad3;
524 			}
525 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
526 
527 			/* Load the partition info if not already loaded. */
528 			if (param_error == 0) {
529 				sdgetdisklabel(sd);
530 				SC_DEBUG(periph, SCSIPI_DB3,
531 				     ("Disklabel loaded "));
532 			}
533 		}
534 	}
535 
536 	/* Check that the partition exists. */
537 	if (part != RAW_PART &&
538 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
539 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
540 		error = ENXIO;
541 		goto bad3;
542 	}
543 
544  out:	/* Insure only one open at a time. */
545 	switch (fmt) {
546 	case S_IFCHR:
547 		sd->sc_dk.dk_copenmask |= (1 << part);
548 		break;
549 	case S_IFBLK:
550 		sd->sc_dk.dk_bopenmask |= (1 << part);
551 		break;
552 	}
553 	sd->sc_dk.dk_openmask =
554 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
555 
556 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
557 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
558 	return (0);
559 
560  bad3:
561 	if (sd->sc_dk.dk_openmask == 0) {
562 		if (periph->periph_flags & PERIPH_REMOVABLE)
563 			scsipi_prevent(periph, SPAMR_ALLOW,
564 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
565 			    XS_CTL_IGNORE_MEDIA_CHANGE);
566 		periph->periph_flags &= ~PERIPH_OPEN;
567 	}
568 
569  bad2:
570 	if (sd->sc_dk.dk_openmask == 0)
571 		scsipi_adapter_delref(adapt);
572 
573  bad1:
574 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
575 	return (error);
576 }
577 
578 /*
579  * close the device.. only called if we are the LAST occurence of an open
580  * device.  Convenient now but usually a pain.
581  */
582 static int
583 sdclose(dev_t dev, int flag, int fmt, struct proc *p)
584 {
585 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
586 	struct scsipi_periph *periph = sd->sc_periph;
587 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
588 	int part = SDPART(dev);
589 	int error;
590 
591 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
592 		return (error);
593 
594 	switch (fmt) {
595 	case S_IFCHR:
596 		sd->sc_dk.dk_copenmask &= ~(1 << part);
597 		break;
598 	case S_IFBLK:
599 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
600 		break;
601 	}
602 	sd->sc_dk.dk_openmask =
603 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
604 
605 	if (sd->sc_dk.dk_openmask == 0) {
606 		/*
607 		 * If the disk cache needs flushing, and the disk supports
608 		 * it, do it now.
609 		 */
610 		if ((sd->flags & SDF_DIRTY) != 0) {
611 			if (sd_flush(sd, 0)) {
612 				printf("%s: cache synchronization failed\n",
613 				    sd->sc_dev.dv_xname);
614 				sd->flags &= ~SDF_FLUSHING;
615 			} else
616 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
617 		}
618 
619 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
620 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
621 
622 		scsipi_wait_drain(periph);
623 
624 		if (periph->periph_flags & PERIPH_REMOVABLE)
625 			scsipi_prevent(periph, SPAMR_ALLOW,
626 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
627 			    XS_CTL_IGNORE_NOT_READY);
628 		periph->periph_flags &= ~PERIPH_OPEN;
629 
630 		scsipi_wait_drain(periph);
631 
632 		scsipi_adapter_delref(adapt);
633 	}
634 
635 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
636 	return (0);
637 }
638 
639 /*
640  * Actually translate the requested transfer into one the physical driver
641  * can understand.  The transfer is described by a buf and will include
642  * only one physical transfer.
643  */
644 static void
645 sdstrategy(struct buf *bp)
646 {
647 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
648 	struct scsipi_periph *periph = sd->sc_periph;
649 	struct disklabel *lp;
650 	daddr_t blkno;
651 	int s;
652 	boolean_t sector_aligned;
653 
654 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
655 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
656 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
657 	/*
658 	 * If the device has been made invalid, error out
659 	 */
660 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
661 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
662 		if (periph->periph_flags & PERIPH_OPEN)
663 			bp->b_error = EIO;
664 		else
665 			bp->b_error = ENODEV;
666 		goto bad;
667 	}
668 
669 	lp = sd->sc_dk.dk_label;
670 
671 	/*
672 	 * The transfer must be a whole number of blocks, offset must not be
673 	 * negative.
674 	 */
675 	if (lp->d_secsize == DEV_BSIZE) {
676 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
677 	} else {
678 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
679 	}
680 	if (!sector_aligned || bp->b_blkno < 0) {
681 		bp->b_error = EINVAL;
682 		goto bad;
683 	}
684 	/*
685 	 * If it's a null transfer, return immediatly
686 	 */
687 	if (bp->b_bcount == 0)
688 		goto done;
689 
690 	/*
691 	 * Do bounds checking, adjust transfer. if error, process.
692 	 * If end of partition, just return.
693 	 */
694 	if (SDPART(bp->b_dev) == RAW_PART) {
695 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
696 		    sd->params.disksize512) <= 0)
697 			goto done;
698 	} else {
699 		if (bounds_check_with_label(&sd->sc_dk, bp,
700 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
701 			goto done;
702 	}
703 
704 	/*
705 	 * Now convert the block number to absolute and put it in
706 	 * terms of the device's logical block size.
707 	 */
708 	if (lp->d_secsize == DEV_BSIZE)
709 		blkno = bp->b_blkno;
710 	else if (lp->d_secsize > DEV_BSIZE)
711 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
712 	else
713 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
714 
715 	if (SDPART(bp->b_dev) != RAW_PART)
716 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
717 
718 	bp->b_rawblkno = blkno;
719 
720 	s = splbio();
721 
722 	/*
723 	 * Place it in the queue of disk activities for this disk.
724 	 *
725 	 * XXX Only do disksort() if the current operating mode does not
726 	 * XXX include tagged queueing.
727 	 */
728 	BUFQ_PUT(&sd->buf_queue, bp);
729 
730 	/*
731 	 * Tell the device to get going on the transfer if it's
732 	 * not doing anything, otherwise just wait for completion
733 	 */
734 	sdstart(sd->sc_periph);
735 
736 	splx(s);
737 	return;
738 
739 bad:
740 	bp->b_flags |= B_ERROR;
741 done:
742 	/*
743 	 * Correctly set the buf to indicate a completed xfer
744 	 */
745 	bp->b_resid = bp->b_bcount;
746 	biodone(bp);
747 }
748 
749 /*
750  * sdstart looks to see if there is a buf waiting for the device
751  * and that the device is not already busy. If both are true,
752  * It dequeues the buf and creates a scsi command to perform the
753  * transfer in the buf. The transfer request will call scsipi_done
754  * on completion, which will in turn call this routine again
755  * so that the next queued transfer is performed.
756  * The bufs are queued by the strategy routine (sdstrategy)
757  *
758  * This routine is also called after other non-queued requests
759  * have been made of the scsi driver, to ensure that the queue
760  * continues to be drained.
761  *
762  * must be called at the correct (highish) spl level
763  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
764  */
765 static void
766 sdstart(struct scsipi_periph *periph)
767 {
768 	struct sd_softc *sd = (void *)periph->periph_dev;
769 	struct disklabel *lp = sd->sc_dk.dk_label;
770 	struct buf *bp = 0;
771 	struct scsipi_rw_16 cmd16;
772 	struct scsipi_rw_10 cmd_big;
773 	struct scsi_rw_6 cmd_small;
774 	struct scsipi_generic *cmdp;
775 	struct scsipi_xfer *xs;
776 	int nblks, cmdlen, error, flags;
777 
778 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
779 	/*
780 	 * Check if the device has room for another command
781 	 */
782 	while (periph->periph_active < periph->periph_openings) {
783 		/*
784 		 * there is excess capacity, but a special waits
785 		 * It'll need the adapter as soon as we clear out of the
786 		 * way and let it run (user level wait).
787 		 */
788 		if (periph->periph_flags & PERIPH_WAITING) {
789 			periph->periph_flags &= ~PERIPH_WAITING;
790 			wakeup((caddr_t)periph);
791 			return;
792 		}
793 
794 		/*
795 		 * If the device has become invalid, abort all the
796 		 * reads and writes until all files have been closed and
797 		 * re-opened
798 		 */
799 		if (__predict_false(
800 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
801 			if ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
802 				bp->b_error = EIO;
803 				bp->b_flags |= B_ERROR;
804 				bp->b_resid = bp->b_bcount;
805 				biodone(bp);
806 				continue;
807 			} else {
808 				return;
809 			}
810 		}
811 
812 		/*
813 		 * See if there is a buf with work for us to do..
814 		 */
815 		if ((bp = BUFQ_PEEK(&sd->buf_queue)) == NULL)
816 			return;
817 
818 		/*
819 		 * We have a buf, now we should make a command.
820 		 */
821 
822 		if (lp->d_secsize == DEV_BSIZE)
823 			nblks = bp->b_bcount >> DEV_BSHIFT;
824 		else
825 			nblks = howmany(bp->b_bcount, lp->d_secsize);
826 
827 		/*
828 		 * Fill out the scsi command.  Use the smallest CDB possible
829 		 * (6-byte, 10-byte, or 16-byte).
830 		 */
831 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
832 		    ((nblks & 0xff) == nblks) &&
833 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
834 			/* 6-byte CDB */
835 			memset(&cmd_small, 0, sizeof(cmd_small));
836 			cmd_small.opcode = (bp->b_flags & B_READ) ?
837 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
838 			_lto3b(bp->b_rawblkno, cmd_small.addr);
839 			cmd_small.length = nblks & 0xff;
840 			cmdlen = sizeof(cmd_small);
841 			cmdp = (struct scsipi_generic *)&cmd_small;
842 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
843 			/* 10-byte CDB */
844 			memset(&cmd_big, 0, sizeof(cmd_big));
845 			cmd_big.opcode = (bp->b_flags & B_READ) ?
846 			    READ_10 : WRITE_10;
847 			_lto4b(bp->b_rawblkno, cmd_big.addr);
848 			_lto2b(nblks, cmd_big.length);
849 			cmdlen = sizeof(cmd_big);
850 			cmdp = (struct scsipi_generic *)&cmd_big;
851 		} else {
852 			/* 16-byte CDB */
853 			memset(&cmd16, 0, sizeof(cmd16));
854 			cmd16.opcode = (bp->b_flags & B_READ) ?
855 			    READ_16 : WRITE_16;
856 			_lto8b(bp->b_rawblkno, cmd16.addr);
857 			_lto4b(nblks, cmd16.length);
858 			cmdlen = sizeof(cmd16);
859 			cmdp = (struct scsipi_generic *)&cmd16;
860 		}
861 
862 		/* Instrumentation. */
863 		disk_busy(&sd->sc_dk);
864 
865 		/*
866 		 * Mark the disk dirty so that the cache will be
867 		 * flushed on close.
868 		 */
869 		if ((bp->b_flags & B_READ) == 0)
870 			sd->flags |= SDF_DIRTY;
871 
872 		/*
873 		 * Figure out what flags to use.
874 		 */
875 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
876 		if (bp->b_flags & B_READ)
877 			flags |= XS_CTL_DATA_IN;
878 		else
879 			flags |= XS_CTL_DATA_OUT;
880 
881 		/*
882 		 * Call the routine that chats with the adapter.
883 		 * Note: we cannot sleep as we may be an interrupt
884 		 */
885 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
886 		    (u_char *)bp->b_data, bp->b_bcount,
887 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
888 		if (__predict_false(xs == NULL)) {
889 			/*
890 			 * out of memory. Keep this buffer in the queue, and
891 			 * retry later.
892 			 */
893 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
894 			    periph);
895 			return;
896 		}
897 		/*
898 		 * need to dequeue the buffer before queuing the command,
899 		 * because cdstart may be called recursively from the
900 		 * HBA driver
901 		 */
902 #ifdef DIAGNOSTIC
903 		if (BUFQ_GET(&sd->buf_queue) != bp)
904 			panic("sdstart(): dequeued wrong buf");
905 #else
906 		BUFQ_GET(&sd->buf_queue);
907 #endif
908 		error = scsipi_execute_xs(xs);
909 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
910 		KASSERT(error == 0);
911 	}
912 }
913 
914 static void
915 sdrestart(void *v)
916 {
917 	int s = splbio();
918 	sdstart((struct scsipi_periph *)v);
919 	splx(s);
920 }
921 
922 static void
923 sddone(struct scsipi_xfer *xs, int error)
924 {
925 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
926 	struct buf *bp = xs->bp;
927 
928 	if (sd->flags & SDF_FLUSHING) {
929 		/* Flush completed, no longer dirty. */
930 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
931 	}
932 
933 	if (bp) {
934 		bp->b_error = error;
935 		bp->b_resid = xs->resid;
936 		if (error)
937 			bp->b_flags |= B_ERROR;
938 
939 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
940 		    (bp->b_flags & B_READ));
941 #if NRND > 0
942 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
943 #endif
944 
945 		biodone(bp);
946 	}
947 }
948 
949 static void
950 sdminphys(struct buf *bp)
951 {
952 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
953 	long xmax;
954 
955 	/*
956 	 * If the device is ancient, we want to make sure that
957 	 * the transfer fits into a 6-byte cdb.
958 	 *
959 	 * XXX Note that the SCSI-I spec says that 256-block transfers
960 	 * are allowed in a 6-byte read/write, and are specified
961 	 * by settng the "length" to 0.  However, we're conservative
962 	 * here, allowing only 255-block transfers in case an
963 	 * ancient device gets confused by length == 0.  A length of 0
964 	 * in a 10-byte read/write actually means 0 blocks.
965 	 */
966 	if ((sd->flags & SDF_ANCIENT) &&
967 	    ((sd->sc_periph->periph_flags &
968 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
969 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
970 
971 		if (bp->b_bcount > xmax)
972 			bp->b_bcount = xmax;
973 	}
974 
975 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
976 }
977 
978 static int
979 sdread(dev_t dev, struct uio *uio, int ioflag)
980 {
981 
982 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
983 }
984 
985 static int
986 sdwrite(dev_t dev, struct uio *uio, int ioflag)
987 {
988 
989 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
990 }
991 
992 /*
993  * Perform special action on behalf of the user
994  * Knows about the internals of this device
995  */
996 static int
997 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct proc *p)
998 {
999 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1000 	struct scsipi_periph *periph = sd->sc_periph;
1001 	int part = SDPART(dev);
1002 	int error = 0;
1003 #ifdef __HAVE_OLD_DISKLABEL
1004 	struct disklabel *newlabel = NULL;
1005 #endif
1006 
1007 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1008 
1009 	/*
1010 	 * If the device is not valid, some IOCTLs can still be
1011 	 * handled on the raw partition. Check this here.
1012 	 */
1013 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1014 		switch (cmd) {
1015 		case DIOCKLABEL:
1016 		case DIOCWLABEL:
1017 		case DIOCLOCK:
1018 		case DIOCEJECT:
1019 		case ODIOCEJECT:
1020 		case DIOCGCACHE:
1021 		case DIOCSCACHE:
1022 		case SCIOCIDENTIFY:
1023 		case OSCIOCIDENTIFY:
1024 		case SCIOCCOMMAND:
1025 		case SCIOCDEBUG:
1026 			if (part == RAW_PART)
1027 				break;
1028 		/* FALLTHROUGH */
1029 		default:
1030 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1031 				return (ENODEV);
1032 			else
1033 				return (EIO);
1034 		}
1035 	}
1036 
1037 	switch (cmd) {
1038 	case DIOCGDINFO:
1039 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1040 		return (0);
1041 
1042 #ifdef __HAVE_OLD_DISKLABEL
1043 	case ODIOCGDINFO:
1044 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1045 		if (newlabel == NULL)
1046 			return EIO;
1047 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1048 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1049 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1050 		else
1051 			error = ENOTTY;
1052 		free(newlabel, M_TEMP);
1053 		return error;
1054 #endif
1055 
1056 	case DIOCGPART:
1057 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1058 		((struct partinfo *)addr)->part =
1059 		    &sd->sc_dk.dk_label->d_partitions[part];
1060 		return (0);
1061 
1062 	case DIOCWDINFO:
1063 	case DIOCSDINFO:
1064 #ifdef __HAVE_OLD_DISKLABEL
1065 	case ODIOCWDINFO:
1066 	case ODIOCSDINFO:
1067 #endif
1068 	{
1069 		struct disklabel *lp;
1070 
1071 		if ((flag & FWRITE) == 0)
1072 			return (EBADF);
1073 
1074 #ifdef __HAVE_OLD_DISKLABEL
1075  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1076 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1077 			if (newlabel == NULL)
1078 				return EIO;
1079 			memset(newlabel, 0, sizeof newlabel);
1080 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1081 			lp = newlabel;
1082 		} else
1083 #endif
1084 		lp = (struct disklabel *)addr;
1085 
1086 		if ((error = lockmgr(&sd->sc_dk.dk_openlock,
1087 				     LK_EXCLUSIVE, NULL)) != 0)
1088 			goto bad;
1089 		sd->flags |= SDF_LABELLING;
1090 
1091 		error = setdisklabel(sd->sc_dk.dk_label,
1092 		    lp, /*sd->sc_dk.dk_openmask : */0,
1093 		    sd->sc_dk.dk_cpulabel);
1094 		if (error == 0) {
1095 			if (cmd == DIOCWDINFO
1096 #ifdef __HAVE_OLD_DISKLABEL
1097 			    || cmd == ODIOCWDINFO
1098 #endif
1099 			   )
1100 				error = writedisklabel(SDLABELDEV(dev),
1101 				    sdstrategy, sd->sc_dk.dk_label,
1102 				    sd->sc_dk.dk_cpulabel);
1103 		}
1104 
1105 		sd->flags &= ~SDF_LABELLING;
1106 		(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1107 bad:
1108 #ifdef __HAVE_OLD_DISKLABEL
1109 		if (newlabel != NULL)
1110 			free(newlabel, M_TEMP);
1111 #endif
1112 		return (error);
1113 	}
1114 
1115 	case DIOCKLABEL:
1116 		if (*(int *)addr)
1117 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1118 		else
1119 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1120 		return (0);
1121 
1122 	case DIOCWLABEL:
1123 		if ((flag & FWRITE) == 0)
1124 			return (EBADF);
1125 		if (*(int *)addr)
1126 			sd->flags |= SDF_WLABEL;
1127 		else
1128 			sd->flags &= ~SDF_WLABEL;
1129 		return (0);
1130 
1131 	case DIOCLOCK:
1132 		return (scsipi_prevent(periph,
1133 		    (*(int *)addr) ? SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1134 
1135 	case DIOCEJECT:
1136 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1137 			return (ENOTTY);
1138 		if (*(int *)addr == 0) {
1139 			/*
1140 			 * Don't force eject: check that we are the only
1141 			 * partition open. If so, unlock it.
1142 			 */
1143 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1144 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1145 			    sd->sc_dk.dk_openmask) {
1146 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1147 				    XS_CTL_IGNORE_NOT_READY);
1148 				if (error)
1149 					return (error);
1150 			} else {
1151 				return (EBUSY);
1152 			}
1153 		}
1154 		/* FALLTHROUGH */
1155 	case ODIOCEJECT:
1156 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1157 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1158 
1159 	case DIOCGDEFLABEL:
1160 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1161 		return (0);
1162 
1163 #ifdef __HAVE_OLD_DISKLABEL
1164 	case ODIOCGDEFLABEL:
1165 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1166 		if (newlabel == NULL)
1167 			return EIO;
1168 		sdgetdefaultlabel(sd, newlabel);
1169 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1170 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1171 		else
1172 			error = ENOTTY;
1173 		free(newlabel, M_TEMP);
1174 		return error;
1175 #endif
1176 
1177 	case DIOCGCACHE:
1178 		return (sd_getcache(sd, (int *) addr));
1179 
1180 	case DIOCSCACHE:
1181 		if ((flag & FWRITE) == 0)
1182 			return (EBADF);
1183 		return (sd_setcache(sd, *(int *) addr));
1184 
1185 	case DIOCCACHESYNC:
1186 		/*
1187 		 * XXX Do we really need to care about having a writable
1188 		 * file descriptor here?
1189 		 */
1190 		if ((flag & FWRITE) == 0)
1191 			return (EBADF);
1192 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1193 			error = sd_flush(sd, 0);
1194 			if (error)
1195 				sd->flags &= ~SDF_FLUSHING;
1196 			else
1197 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1198 		} else
1199 			error = 0;
1200 		return (error);
1201 
1202 	case DIOCAWEDGE:
1203 	    {
1204 	    	struct dkwedge_info *dkw = (void *) addr;
1205 
1206 		if ((flag & FWRITE) == 0)
1207 			return (EBADF);
1208 
1209 		/* If the ioctl happens here, the parent is us. */
1210 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1211 		return (dkwedge_add(dkw));
1212 	    }
1213 
1214 	case DIOCDWEDGE:
1215 	    {
1216 	    	struct dkwedge_info *dkw = (void *) addr;
1217 
1218 		if ((flag & FWRITE) == 0)
1219 			return (EBADF);
1220 
1221 		/* If the ioctl happens here, the parent is us. */
1222 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1223 		return (dkwedge_del(dkw));
1224 	    }
1225 
1226 	case DIOCLWEDGES:
1227 	    {
1228 	    	struct dkwedge_list *dkwl = (void *) addr;
1229 
1230 		return (dkwedge_list(&sd->sc_dk, dkwl, p));
1231 	    }
1232 
1233 	default:
1234 		if (part != RAW_PART)
1235 			return (ENOTTY);
1236 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1237 	}
1238 
1239 #ifdef DIAGNOSTIC
1240 	panic("sdioctl: impossible");
1241 #endif
1242 }
1243 
1244 static void
1245 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1246 {
1247 
1248 	memset(lp, 0, sizeof(struct disklabel));
1249 
1250 	lp->d_secsize = sd->params.blksize;
1251 	lp->d_ntracks = sd->params.heads;
1252 	lp->d_nsectors = sd->params.sectors;
1253 	lp->d_ncylinders = sd->params.cyls;
1254 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1255 
1256 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1257 	case SCSIPI_BUSTYPE_SCSI:
1258 		lp->d_type = DTYPE_SCSI;
1259 		break;
1260 	case SCSIPI_BUSTYPE_ATAPI:
1261 		lp->d_type = DTYPE_ATAPI;
1262 		break;
1263 	}
1264 	/*
1265 	 * XXX
1266 	 * We could probe the mode pages to figure out what kind of disc it is.
1267 	 * Is this worthwhile?
1268 	 */
1269 	strncpy(lp->d_typename, "mydisk", 16);
1270 	strncpy(lp->d_packname, "fictitious", 16);
1271 	lp->d_secperunit = sd->params.disksize;
1272 	lp->d_rpm = sd->params.rot_rate;
1273 	lp->d_interleave = 1;
1274 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1275 	    D_REMOVABLE : 0;
1276 
1277 	lp->d_partitions[RAW_PART].p_offset = 0;
1278 	lp->d_partitions[RAW_PART].p_size =
1279 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1280 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1281 	lp->d_npartitions = RAW_PART + 1;
1282 
1283 	lp->d_magic = DISKMAGIC;
1284 	lp->d_magic2 = DISKMAGIC;
1285 	lp->d_checksum = dkcksum(lp);
1286 }
1287 
1288 
1289 /*
1290  * Load the label information on the named device
1291  */
1292 static void
1293 sdgetdisklabel(struct sd_softc *sd)
1294 {
1295 	struct disklabel *lp = sd->sc_dk.dk_label;
1296 	const char *errstring;
1297 
1298 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1299 
1300 	sdgetdefaultlabel(sd, lp);
1301 
1302 	if (lp->d_secpercyl == 0) {
1303 		lp->d_secpercyl = 100;
1304 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1305 	}
1306 
1307 	/*
1308 	 * Call the generic disklabel extraction routine
1309 	 */
1310 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1311 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1312 	if (errstring) {
1313 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1314 		return;
1315 	}
1316 }
1317 
1318 static void
1319 sd_shutdown(void *arg)
1320 {
1321 	struct sd_softc *sd = arg;
1322 
1323 	/*
1324 	 * If the disk cache needs to be flushed, and the disk supports
1325 	 * it, flush it.  We're cold at this point, so we poll for
1326 	 * completion.
1327 	 */
1328 	if ((sd->flags & SDF_DIRTY) != 0) {
1329 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1330 			printf("%s: cache synchronization failed\n",
1331 			    sd->sc_dev.dv_xname);
1332 			sd->flags &= ~SDF_FLUSHING;
1333 		} else
1334 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1335 	}
1336 }
1337 
1338 /*
1339  * Check Errors
1340  */
1341 static int
1342 sd_interpret_sense(struct scsipi_xfer *xs)
1343 {
1344 	struct scsipi_periph *periph = xs->xs_periph;
1345 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1346 	struct sd_softc *sd = (void *)periph->periph_dev;
1347 	int s, error, retval = EJUSTRETURN;
1348 
1349 	/*
1350 	 * If the periph is already recovering, just do the normal
1351 	 * error processing.
1352 	 */
1353 	if (periph->periph_flags & PERIPH_RECOVERING)
1354 		return (retval);
1355 
1356 	/*
1357 	 * If the device is not open yet, let the generic code handle it.
1358 	 */
1359 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1360 		return (retval);
1361 
1362 	/*
1363 	 * If it isn't a extended or extended/deferred error, let
1364 	 * the generic code handle it.
1365 	 */
1366 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1367 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1368 		return (retval);
1369 
1370 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1371 	    sense->asc == 0x4) {
1372 		if (sense->ascq == 0x01)	{
1373 			/*
1374 			 * Unit In The Process Of Becoming Ready.
1375 			 */
1376 			printf("%s: waiting for pack to spin up...\n",
1377 			    sd->sc_dev.dv_xname);
1378 			if (!callout_pending(&periph->periph_callout))
1379 				scsipi_periph_freeze(periph, 1);
1380 			callout_reset(&periph->periph_callout,
1381 			    5 * hz, scsipi_periph_timed_thaw, periph);
1382 			retval = ERESTART;
1383 		} else if (sense->ascq == 0x02) {
1384 			printf("%s: pack is stopped, restarting...\n",
1385 			    sd->sc_dev.dv_xname);
1386 			s = splbio();
1387 			periph->periph_flags |= PERIPH_RECOVERING;
1388 			splx(s);
1389 			error = scsipi_start(periph, SSS_START,
1390 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1391 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1392 			if (error) {
1393 				printf("%s: unable to restart pack\n",
1394 				    sd->sc_dev.dv_xname);
1395 				retval = error;
1396 			} else
1397 				retval = ERESTART;
1398 			s = splbio();
1399 			periph->periph_flags &= ~PERIPH_RECOVERING;
1400 			splx(s);
1401 		}
1402 	}
1403 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1404 	    sense->asc == 0x31 &&
1405 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1406 		/* Medium Format Corrupted */
1407 		retval = EFTYPE;
1408 	}
1409 	return (retval);
1410 }
1411 
1412 
1413 static int
1414 sdsize(dev_t dev)
1415 {
1416 	struct sd_softc *sd;
1417 	int part, unit, omask;
1418 	int size;
1419 
1420 	unit = SDUNIT(dev);
1421 	if (unit >= sd_cd.cd_ndevs)
1422 		return (-1);
1423 	sd = sd_cd.cd_devs[unit];
1424 	if (sd == NULL)
1425 		return (-1);
1426 
1427 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1428 		return (-1);
1429 
1430 	part = SDPART(dev);
1431 	omask = sd->sc_dk.dk_openmask & (1 << part);
1432 
1433 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1434 		return (-1);
1435 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1436 		size = -1;
1437 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1438 		size = -1;
1439 	else
1440 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1441 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1442 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1443 		return (-1);
1444 	return (size);
1445 }
1446 
1447 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1448 static struct scsipi_xfer sx;
1449 static int sddoingadump;
1450 
1451 /*
1452  * dump all of physical memory into the partition specified, starting
1453  * at offset 'dumplo' into the partition.
1454  */
1455 static int
1456 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1457 {
1458 	struct sd_softc *sd;	/* disk unit to do the I/O */
1459 	struct disklabel *lp;	/* disk's disklabel */
1460 	int	unit, part;
1461 	int	sectorsize;	/* size of a disk sector */
1462 	int	nsects;		/* number of sectors in partition */
1463 	int	sectoff;	/* sector offset of partition */
1464 	int	totwrt;		/* total number of sectors left to write */
1465 	int	nwrt;		/* current number of sectors to write */
1466 	struct scsipi_rw_10 cmd;	/* write command */
1467 	struct scsipi_xfer *xs;	/* ... convenience */
1468 	struct scsipi_periph *periph;
1469 	struct scsipi_channel *chan;
1470 
1471 	/* Check if recursive dump; if so, punt. */
1472 	if (sddoingadump)
1473 		return (EFAULT);
1474 
1475 	/* Mark as active early. */
1476 	sddoingadump = 1;
1477 
1478 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1479 	part = SDPART(dev);
1480 
1481 	/* Check for acceptable drive number. */
1482 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1483 		return (ENXIO);
1484 
1485 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1486 		return (ENODEV);
1487 
1488 	periph = sd->sc_periph;
1489 	chan = periph->periph_channel;
1490 
1491 	/* Make sure it was initialized. */
1492 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1493 		return (ENXIO);
1494 
1495 	/* Convert to disk sectors.  Request must be a multiple of size. */
1496 	lp = sd->sc_dk.dk_label;
1497 	sectorsize = lp->d_secsize;
1498 	if ((size % sectorsize) != 0)
1499 		return (EFAULT);
1500 	totwrt = size / sectorsize;
1501 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1502 
1503 	nsects = lp->d_partitions[part].p_size;
1504 	sectoff = lp->d_partitions[part].p_offset;
1505 
1506 	/* Check transfer bounds against partition size. */
1507 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1508 		return (EINVAL);
1509 
1510 	/* Offset block number to start of partition. */
1511 	blkno += sectoff;
1512 
1513 	xs = &sx;
1514 
1515 	while (totwrt > 0) {
1516 		nwrt = totwrt;		/* XXX */
1517 #ifndef	SD_DUMP_NOT_TRUSTED
1518 		/*
1519 		 *  Fill out the scsi command
1520 		 */
1521 		memset(&cmd, 0, sizeof(cmd));
1522 		cmd.opcode = WRITE_10;
1523 		_lto4b(blkno, cmd.addr);
1524 		_lto2b(nwrt, cmd.length);
1525 		/*
1526 		 * Fill out the scsipi_xfer structure
1527 		 *    Note: we cannot sleep as we may be an interrupt
1528 		 * don't use scsipi_command() as it may want to wait
1529 		 * for an xs.
1530 		 */
1531 		memset(xs, 0, sizeof(sx));
1532 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1533 		    XS_CTL_DATA_OUT;
1534 		xs->xs_status = 0;
1535 		xs->xs_periph = periph;
1536 		xs->xs_retries = SDRETRIES;
1537 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1538 		xs->cmd = (struct scsipi_generic *)&cmd;
1539 		xs->cmdlen = sizeof(cmd);
1540 		xs->resid = nwrt * sectorsize;
1541 		xs->error = XS_NOERROR;
1542 		xs->bp = 0;
1543 		xs->data = va;
1544 		xs->datalen = nwrt * sectorsize;
1545 
1546 		/*
1547 		 * Pass all this info to the scsi driver.
1548 		 */
1549 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1550 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1551 		    xs->error != XS_NOERROR)
1552 			return (EIO);
1553 #else	/* SD_DUMP_NOT_TRUSTED */
1554 		/* Let's just talk about this first... */
1555 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1556 		delay(500 * 1000);	/* half a second */
1557 #endif	/* SD_DUMP_NOT_TRUSTED */
1558 
1559 		/* update block count */
1560 		totwrt -= nwrt;
1561 		blkno += nwrt;
1562 		va += sectorsize * nwrt;
1563 	}
1564 	sddoingadump = 0;
1565 	return (0);
1566 }
1567 
1568 static int
1569 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1570     int page, int flags, int *big)
1571 {
1572 
1573 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1574 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1575 		*big = 1;
1576 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1577 		    size + sizeof(struct scsi_mode_parameter_header_10),
1578 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1579 	} else {
1580 		*big = 0;
1581 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1582 		    size + sizeof(struct scsi_mode_parameter_header_6),
1583 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1584 	}
1585 }
1586 
1587 static int
1588 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1589     int flags, int big)
1590 {
1591 
1592 	if (big) {
1593 		struct scsi_mode_parameter_header_10 *header = sense;
1594 
1595 		_lto2b(0, header->data_length);
1596 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1597 		    size + sizeof(struct scsi_mode_parameter_header_10),
1598 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1599 	} else {
1600 		struct scsi_mode_parameter_header_6 *header = sense;
1601 
1602 		header->data_length = 0;
1603 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1604 		    size + sizeof(struct scsi_mode_parameter_header_6),
1605 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1606 	}
1607 }
1608 
1609 static int
1610 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1611 {
1612 	struct {
1613 		struct scsi_mode_parameter_header_6 header;
1614 		/* no block descriptor */
1615 		u_int8_t pg_code; /* page code (should be 6) */
1616 		u_int8_t pg_length; /* page length (should be 11) */
1617 		u_int8_t wcd; /* bit0: cache disable */
1618 		u_int8_t lbs[2]; /* logical block size */
1619 		u_int8_t size[5]; /* number of log. blocks */
1620 		u_int8_t pp; /* power/performance */
1621 		u_int8_t flags;
1622 		u_int8_t resvd;
1623 	} scsipi_sense;
1624 	u_int64_t sectors;
1625 	int error;
1626 
1627 	/*
1628 	 * scsipi_size (ie "read capacity") and mode sense page 6
1629 	 * give the same information. Do both for now, and check
1630 	 * for consistency.
1631 	 * XXX probably differs for removable media
1632 	 */
1633 	dp->blksize = 512;
1634 	if ((sectors = scsipi_size(sd->sc_periph, flags)) == 0)
1635 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1636 
1637 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1638 	    &scsipi_sense.header, sizeof(scsipi_sense),
1639 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1640 
1641 	if (error != 0)
1642 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1643 
1644 	dp->blksize = _2btol(scsipi_sense.lbs);
1645 	if (dp->blksize == 0)
1646 		dp->blksize = 512;
1647 
1648 	/*
1649 	 * Create a pseudo-geometry.
1650 	 */
1651 	dp->heads = 64;
1652 	dp->sectors = 32;
1653 	dp->cyls = sectors / (dp->heads * dp->sectors);
1654 	dp->disksize = _5btol(scsipi_sense.size);
1655 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1656 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1657 		       (unsigned long long)dp->disksize,
1658 		       (unsigned long long)sectors);
1659 		dp->disksize = sectors;
1660 	}
1661 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1662 
1663 	return (SDGP_RESULT_OK);
1664 }
1665 
1666 /*
1667  * Get the scsi driver to send a full inquiry to the * device and use the
1668  * results to fill out the disk parameter structure.
1669  */
1670 static int
1671 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1672 {
1673 	u_int64_t sectors;
1674 	int error;
1675 #if 0
1676 	int i;
1677 	u_int8_t *p;
1678 #endif
1679 
1680 	dp->disksize = sectors = scsipi_size(sd->sc_periph, flags);
1681 	if (sectors == 0) {
1682 		struct scsipi_read_format_capacities cmd;
1683 		struct {
1684 			struct scsipi_capacity_list_header header;
1685 			struct scsipi_capacity_descriptor desc;
1686 		} __attribute__((packed)) data;
1687 
1688 		memset(&cmd, 0, sizeof(cmd));
1689 		memset(&data, 0, sizeof(data));
1690 		cmd.opcode = READ_FORMAT_CAPACITIES;
1691 		_lto2b(sizeof(data), cmd.length);
1692 
1693 		error = scsipi_command(sd->sc_periph,
1694 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1695 		    SDRETRIES, 20000, NULL,
1696 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1697 		if (error == EFTYPE) {
1698 			/* Medium Format Corrupted, handle as not formatted */
1699 			return (SDGP_RESULT_UNFORMATTED);
1700 		}
1701 		if (error || data.header.length == 0)
1702 			return (SDGP_RESULT_OFFLINE);
1703 
1704 #if 0
1705 printf("rfc: length=%d\n", data.header.length);
1706 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");
1707 #endif
1708 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1709 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1710 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1711 			break;
1712 
1713 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1714 			return (SDGP_RESULT_UNFORMATTED);
1715 
1716 		case SCSIPI_CAP_DESC_CODE_NONE:
1717 			return (SDGP_RESULT_OFFLINE);
1718 		}
1719 
1720 		dp->disksize = sectors = _4btol(data.desc.nblks);
1721 		if (sectors == 0)
1722 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1723 
1724 		dp->blksize = _3btol(data.desc.blklen);
1725 		if (dp->blksize == 0)
1726 			dp->blksize = 512;
1727 	} else {
1728 		struct sd_mode_sense_data scsipi_sense;
1729 		int big, bsize;
1730 		struct scsi_general_block_descriptor *bdesc;
1731 
1732 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1733 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1734 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1735 		dp->blksize = 512;
1736 		if (!error) {
1737 			if (big) {
1738 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1739 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1740 			} else {
1741 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1742 				bsize = scsipi_sense.header.small.blk_desc_len;
1743 			}
1744 
1745 #if 0
1746 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1747 printf("page 0 bsize=%d\n", bsize);
1748 printf("page 0 ok\n");
1749 #endif
1750 
1751 			if (bsize >= 8) {
1752 				dp->blksize = _3btol(bdesc->blklen);
1753 				if (dp->blksize == 0)
1754 					dp->blksize = 512;
1755 			}
1756 		}
1757 	}
1758 
1759 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1760 	return (0);
1761 }
1762 
1763 static int
1764 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1765 {
1766 	struct sd_mode_sense_data scsipi_sense;
1767 	int error;
1768 	int big, poffset, byte2;
1769 	union scsi_disk_pages *pages;
1770 #if 0
1771 	int i;
1772 	u_int8_t *p;
1773 #endif
1774 
1775 	byte2 = SMS_DBD;
1776 again:
1777 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1778 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1779 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1780 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1781 	    flags | XS_CTL_SILENT, &big);
1782 	if (error) {
1783 		if (byte2 == SMS_DBD) {
1784 			/* No result; try once more with DBD off */
1785 			byte2 = 0;
1786 			goto again;
1787 		}
1788 		return (error);
1789 	}
1790 
1791 	if (big) {
1792 		poffset = sizeof scsipi_sense.header.big;
1793 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1794 	} else {
1795 		poffset = sizeof scsipi_sense.header.small;
1796 		poffset += scsipi_sense.header.small.blk_desc_len;
1797 	}
1798 
1799 	pages = (void *)((u_long)&scsipi_sense + poffset);
1800 #if 0
1801 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1802 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1803 #endif
1804 
1805 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1806 		return (ERESTART);
1807 
1808 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1809 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1810 	    _3btol(pages->rigid_geometry.ncyl),
1811 	    pages->rigid_geometry.nheads,
1812 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1813 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1814 	    _2btol(pages->rigid_geometry.land_zone)));
1815 
1816 	/*
1817 	 * KLUDGE!! (for zone recorded disks)
1818 	 * give a number of sectors so that sec * trks * cyls
1819 	 * is <= disk_size
1820 	 * can lead to wasted space! THINK ABOUT THIS !
1821 	 */
1822 	dp->heads = pages->rigid_geometry.nheads;
1823 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1824 	if (dp->heads == 0 || dp->cyls == 0)
1825 		return (ERESTART);
1826 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1827 
1828 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1829 	if (dp->rot_rate == 0)
1830 		dp->rot_rate = 3600;
1831 
1832 #if 0
1833 printf("page 4 ok\n");
1834 #endif
1835 	return (0);
1836 }
1837 
1838 static int
1839 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1840 {
1841 	struct sd_mode_sense_data scsipi_sense;
1842 	int error;
1843 	int big, poffset, byte2;
1844 	union scsi_disk_pages *pages;
1845 #if 0
1846 	int i;
1847 	u_int8_t *p;
1848 #endif
1849 
1850 	byte2 = SMS_DBD;
1851 again:
1852 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1853 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1854 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1855 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1856 	    flags | XS_CTL_SILENT, &big);
1857 	if (error) {
1858 		if (byte2 == SMS_DBD) {
1859 			/* No result; try once more with DBD off */
1860 			byte2 = 0;
1861 			goto again;
1862 		}
1863 		return (error);
1864 	}
1865 
1866 	if (big) {
1867 		poffset = sizeof scsipi_sense.header.big;
1868 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1869 	} else {
1870 		poffset = sizeof scsipi_sense.header.small;
1871 		poffset += scsipi_sense.header.small.blk_desc_len;
1872 	}
1873 
1874 	pages = (void *)((u_long)&scsipi_sense + poffset);
1875 #if 0
1876 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1877 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1878 #endif
1879 
1880 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1881 		return (ERESTART);
1882 
1883 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1884 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1885 	    _3btol(pages->flex_geometry.ncyl),
1886 	    pages->flex_geometry.nheads,
1887 	    pages->flex_geometry.ph_sec_tr,
1888 	    _2btol(pages->flex_geometry.bytes_s)));
1889 
1890 	dp->heads = pages->flex_geometry.nheads;
1891 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1892 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1893 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1894 		return (ERESTART);
1895 
1896 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1897 	if (dp->rot_rate == 0)
1898 		dp->rot_rate = 3600;
1899 
1900 #if 0
1901 printf("page 5 ok\n");
1902 #endif
1903 	return (0);
1904 }
1905 
1906 static int
1907 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1908 {
1909 	int error;
1910 
1911 	/*
1912 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1913 	 * cleared by the caller if necessary.
1914 	 */
1915 	if (sd->type == T_SIMPLE_DIRECT)
1916 		return (sd_get_simplifiedparms(sd, dp, flags));
1917 
1918 	error = sd_get_capacity(sd, dp, flags);
1919 	if (error)
1920 		return (error);
1921 
1922 	if (sd->type == T_OPTICAL)
1923 		goto page0;
1924 
1925 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1926 		if (!sd_get_parms_page5(sd, dp, flags) ||
1927 		    !sd_get_parms_page4(sd, dp, flags))
1928 			return (SDGP_RESULT_OK);
1929 	} else {
1930 		if (!sd_get_parms_page4(sd, dp, flags) ||
1931 		    !sd_get_parms_page5(sd, dp, flags))
1932 			return (SDGP_RESULT_OK);
1933 	}
1934 
1935 page0:
1936 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1937 	/* Try calling driver's method for figuring out geometry. */
1938 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1939 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1940 		(sd->sc_periph, dp, dp->disksize)) {
1941 		/*
1942 		 * Use adaptec standard fictitious geometry
1943 		 * this depends on which controller (e.g. 1542C is
1944 		 * different. but we have to put SOMETHING here..)
1945 		 */
1946 		dp->heads = 64;
1947 		dp->sectors = 32;
1948 		dp->cyls = dp->disksize / (64 * 32);
1949 	}
1950 	dp->rot_rate = 3600;
1951 	return (SDGP_RESULT_OK);
1952 }
1953 
1954 static int
1955 sd_flush(struct sd_softc *sd, int flags)
1956 {
1957 	struct scsipi_periph *periph = sd->sc_periph;
1958 	struct scsi_synchronize_cache_10 cmd;
1959 
1960 	/*
1961 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1962 	 * We issue with address 0 length 0, which should be
1963 	 * interpreted by the device as "all remaining blocks
1964 	 * starting at address 0".  We ignore ILLEGAL REQUEST
1965 	 * in the event that the command is not supported by
1966 	 * the device, and poll for completion so that we know
1967 	 * that the cache has actually been flushed.
1968 	 *
1969 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
1970 	 * command, as indicated by our quirks flags.
1971 	 *
1972 	 * XXX What about older devices?
1973 	 */
1974 	if (periph->periph_version < 2 ||
1975 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
1976 		return (0);
1977 
1978 	sd->flags |= SDF_FLUSHING;
1979 	memset(&cmd, 0, sizeof(cmd));
1980 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
1981 
1982 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
1983 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
1984 }
1985 
1986 static int
1987 sd_getcache(struct sd_softc *sd, int *bitsp)
1988 {
1989 	struct scsipi_periph *periph = sd->sc_periph;
1990 	struct sd_mode_sense_data scsipi_sense;
1991 	int error, bits = 0;
1992 	int big;
1993 	union scsi_disk_pages *pages;
1994 
1995 	if (periph->periph_version < 2)
1996 		return (EOPNOTSUPP);
1997 
1998 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1999 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2000 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2001 	if (error)
2002 		return (error);
2003 
2004 	if (big)
2005 		pages = (void *)(&scsipi_sense.header.big + 1);
2006 	else
2007 		pages = (void *)(&scsipi_sense.header.small + 1);
2008 
2009 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2010 		bits |= DKCACHE_READ;
2011 	if (pages->caching_params.flags & CACHING_WCE)
2012 		bits |= DKCACHE_WRITE;
2013 	if (pages->caching_params.pg_code & PGCODE_PS)
2014 		bits |= DKCACHE_SAVE;
2015 
2016 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2017 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2018 	    sizeof(scsipi_sense.pages.caching_params),
2019 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2020 	if (error == 0) {
2021 		if (big)
2022 			pages = (void *)(&scsipi_sense.header.big + 1);
2023 		else
2024 			pages = (void *)(&scsipi_sense.header.small + 1);
2025 
2026 		if (pages->caching_params.flags & CACHING_RCD)
2027 			bits |= DKCACHE_RCHANGE;
2028 		if (pages->caching_params.flags & CACHING_WCE)
2029 			bits |= DKCACHE_WCHANGE;
2030 	}
2031 
2032 	*bitsp = bits;
2033 
2034 	return (0);
2035 }
2036 
2037 static int
2038 sd_setcache(struct sd_softc *sd, int bits)
2039 {
2040 	struct scsipi_periph *periph = sd->sc_periph;
2041 	struct sd_mode_sense_data scsipi_sense;
2042 	int error;
2043 	uint8_t oflags, byte2 = 0;
2044 	int big;
2045 	union scsi_disk_pages *pages;
2046 
2047 	if (periph->periph_version < 2)
2048 		return (EOPNOTSUPP);
2049 
2050 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2051 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2052 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2053 	if (error)
2054 		return (error);
2055 
2056 	if (big)
2057 		pages = (void *)(&scsipi_sense.header.big + 1);
2058 	else
2059 		pages = (void *)(&scsipi_sense.header.small + 1);
2060 
2061 	oflags = pages->caching_params.flags;
2062 
2063 	if (bits & DKCACHE_READ)
2064 		pages->caching_params.flags &= ~CACHING_RCD;
2065 	else
2066 		pages->caching_params.flags |= CACHING_RCD;
2067 
2068 	if (bits & DKCACHE_WRITE)
2069 		pages->caching_params.flags |= CACHING_WCE;
2070 	else
2071 		pages->caching_params.flags &= ~CACHING_WCE;
2072 
2073 	if (oflags == pages->caching_params.flags)
2074 		return (0);
2075 
2076 	pages->caching_params.pg_code &= PGCODE_MASK;
2077 
2078 	if (bits & DKCACHE_SAVE)
2079 		byte2 |= SMS_SP;
2080 
2081 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2082 	    sizeof(struct scsi_mode_page_header) +
2083 	    pages->caching_params.pg_length, 0, big));
2084 }
2085