xref: /netbsd-src/sys/dev/scsipi/sd.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: sd.c,v 1.248 2006/06/01 15:53:09 nathanw 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 1.248 2006/06/01 15:53:09 nathanw 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 int	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 = device_private(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, 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 = device_private(self);
349 	int s, bmaj, cmaj, i, mn;
350 
351 	/* locate the major number */
352 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
353 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
354 
355 	/* Nuke the vnodes for any open instances */
356 	for (i = 0; i < MAXPARTITIONS; i++) {
357 		mn = SDMINOR(device_unit(self), i);
358 		vdevgone(bmaj, mn, mn, VBLK);
359 		vdevgone(cmaj, mn, mn, VCHR);
360 	}
361 
362 	/* kill any pending restart */
363 	callout_stop(&sd->sc_callout);
364 
365 	/* Delete all of our wedges. */
366 	dkwedge_delall(&sd->sc_dk);
367 
368 	s = splbio();
369 
370 	/* Kill off any queued buffers. */
371 	bufq_drain(sd->buf_queue);
372 
373 	bufq_free(sd->buf_queue);
374 
375 	/* Kill off any pending commands. */
376 	scsipi_kill_pending(sd->sc_periph);
377 
378 	splx(s);
379 
380 	/* Detach from the disk list. */
381 	disk_detach(&sd->sc_dk);
382 
383 	/* Get rid of the shutdown hook. */
384 	shutdownhook_disestablish(sd->sc_sdhook);
385 
386 #if NRND > 0
387 	/* Unhook the entropy source. */
388 	rnd_detach_source(&sd->rnd_source);
389 #endif
390 
391 	return (0);
392 }
393 
394 /*
395  * open the device. Make sure the partition info is a up-to-date as can be.
396  */
397 static int
398 sdopen(dev_t dev, int flag, int fmt, struct lwp *l)
399 {
400 	struct sd_softc *sd;
401 	struct scsipi_periph *periph;
402 	struct scsipi_adapter *adapt;
403 	int unit, part;
404 	int error;
405 
406 	unit = SDUNIT(dev);
407 	if (unit >= sd_cd.cd_ndevs)
408 		return (ENXIO);
409 	sd = sd_cd.cd_devs[unit];
410 	if (sd == NULL)
411 		return (ENXIO);
412 
413 	if (!device_is_active(&sd->sc_dev))
414 		return (ENODEV);
415 
416 	part = SDPART(dev);
417 
418 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
419 		return (error);
420 
421 	/*
422 	 * If there are wedges, and this is not RAW_PART, then we
423 	 * need to fail.
424 	 */
425 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
426 		error = EBUSY;
427 		goto bad1;
428 	}
429 
430 	periph = sd->sc_periph;
431 	adapt = periph->periph_channel->chan_adapter;
432 
433 	SC_DEBUG(periph, SCSIPI_DB1,
434 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
435 	    sd_cd.cd_ndevs, part));
436 
437 	/*
438 	 * If this is the first open of this device, add a reference
439 	 * to the adapter.
440 	 */
441 	if (sd->sc_dk.dk_openmask == 0 &&
442 	    (error = scsipi_adapter_addref(adapt)) != 0)
443 		goto bad1;
444 
445 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
446 		/*
447 		 * If any partition is open, but the disk has been invalidated,
448 		 * disallow further opens of non-raw partition
449 		 */
450 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
451 		    (part != RAW_PART || fmt != S_IFCHR)) {
452 			error = EIO;
453 			goto bad2;
454 		}
455 	} else {
456 		int silent;
457 
458 		if (part == RAW_PART && fmt == S_IFCHR)
459 			silent = XS_CTL_SILENT;
460 		else
461 			silent = 0;
462 
463 		/* Check that it is still responding and ok. */
464 		error = scsipi_test_unit_ready(periph,
465 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
466 		    silent);
467 
468 		/*
469 		 * Start the pack spinning if necessary. Always allow the
470 		 * raw parition to be opened, for raw IOCTLs. Data transfers
471 		 * will check for SDEV_MEDIA_LOADED.
472 		 */
473 		if (error == EIO) {
474 			int error2;
475 
476 			error2 = scsipi_start(periph, SSS_START, silent);
477 			switch (error2) {
478 			case 0:
479 				error = 0;
480 				break;
481 			case EIO:
482 			case EINVAL:
483 				break;
484 			default:
485 				error = error2;
486 				break;
487 			}
488 		}
489 		if (error) {
490 			if (silent)
491 				goto out;
492 			goto bad2;
493 		}
494 
495 		periph->periph_flags |= PERIPH_OPEN;
496 
497 		if (periph->periph_flags & PERIPH_REMOVABLE) {
498 			/* Lock the pack in. */
499 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
500 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
501 			    XS_CTL_IGNORE_MEDIA_CHANGE);
502 			if (error)
503 				goto bad3;
504 		}
505 
506 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
507 			int param_error;
508 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
509 
510 			/*
511 			 * Load the physical device parameters.
512 			 *
513 			 * Note that if media is present but unformatted,
514 			 * we allow the open (so that it can be formatted!).
515 			 * The drive should refuse real I/O, if the media is
516 			 * unformatted.
517 			 */
518 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
519 			     == SDGP_RESULT_OFFLINE) {
520 				error = ENXIO;
521 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
522 				goto bad3;
523 			}
524 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
525 
526 			/* Load the partition info if not already loaded. */
527 			if (param_error == 0) {
528 				if (sdgetdisklabel(sd) != 0) {
529 					error = EIO;
530 					goto bad3;
531 				}
532 				SC_DEBUG(periph, SCSIPI_DB3,
533 				     ("Disklabel loaded "));
534 			}
535 		}
536 	}
537 
538 	/* Check that the partition exists. */
539 	if (part != RAW_PART &&
540 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
541 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
542 		error = ENXIO;
543 		goto bad3;
544 	}
545 
546  out:	/* Insure only one open at a time. */
547 	switch (fmt) {
548 	case S_IFCHR:
549 		sd->sc_dk.dk_copenmask |= (1 << part);
550 		break;
551 	case S_IFBLK:
552 		sd->sc_dk.dk_bopenmask |= (1 << part);
553 		break;
554 	}
555 	sd->sc_dk.dk_openmask =
556 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
557 
558 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
559 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
560 	return (0);
561 
562  bad3:
563 	if (sd->sc_dk.dk_openmask == 0) {
564 		if (periph->periph_flags & PERIPH_REMOVABLE)
565 			scsipi_prevent(periph, SPAMR_ALLOW,
566 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
567 			    XS_CTL_IGNORE_MEDIA_CHANGE);
568 		periph->periph_flags &= ~PERIPH_OPEN;
569 	}
570 
571  bad2:
572 	if (sd->sc_dk.dk_openmask == 0)
573 		scsipi_adapter_delref(adapt);
574 
575  bad1:
576 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
577 	return (error);
578 }
579 
580 /*
581  * close the device.. only called if we are the LAST occurence of an open
582  * device.  Convenient now but usually a pain.
583  */
584 static int
585 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
586 {
587 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
588 	struct scsipi_periph *periph = sd->sc_periph;
589 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
590 	int part = SDPART(dev);
591 	int error;
592 
593 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
594 		return (error);
595 
596 	switch (fmt) {
597 	case S_IFCHR:
598 		sd->sc_dk.dk_copenmask &= ~(1 << part);
599 		break;
600 	case S_IFBLK:
601 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
602 		break;
603 	}
604 	sd->sc_dk.dk_openmask =
605 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
606 
607 	if (sd->sc_dk.dk_openmask == 0) {
608 		/*
609 		 * If the disk cache needs flushing, and the disk supports
610 		 * it, do it now.
611 		 */
612 		if ((sd->flags & SDF_DIRTY) != 0) {
613 			if (sd_flush(sd, 0)) {
614 				printf("%s: cache synchronization failed\n",
615 				    sd->sc_dev.dv_xname);
616 				sd->flags &= ~SDF_FLUSHING;
617 			} else
618 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
619 		}
620 
621 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
622 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
623 
624 		scsipi_wait_drain(periph);
625 
626 		if (periph->periph_flags & PERIPH_REMOVABLE)
627 			scsipi_prevent(periph, SPAMR_ALLOW,
628 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
629 			    XS_CTL_IGNORE_NOT_READY);
630 		periph->periph_flags &= ~PERIPH_OPEN;
631 
632 		scsipi_wait_drain(periph);
633 
634 		scsipi_adapter_delref(adapt);
635 	}
636 
637 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
638 	return (0);
639 }
640 
641 /*
642  * Actually translate the requested transfer into one the physical driver
643  * can understand.  The transfer is described by a buf and will include
644  * only one physical transfer.
645  */
646 static void
647 sdstrategy(struct buf *bp)
648 {
649 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
650 	struct scsipi_periph *periph = sd->sc_periph;
651 	struct disklabel *lp;
652 	daddr_t blkno;
653 	int s;
654 	boolean_t sector_aligned;
655 
656 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
657 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
658 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
659 	/*
660 	 * If the device has been made invalid, error out
661 	 */
662 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
663 	    !device_is_active(&sd->sc_dev)) {
664 		if (periph->periph_flags & PERIPH_OPEN)
665 			bp->b_error = EIO;
666 		else
667 			bp->b_error = ENODEV;
668 		goto bad;
669 	}
670 
671 	lp = sd->sc_dk.dk_label;
672 
673 	/*
674 	 * The transfer must be a whole number of blocks, offset must not be
675 	 * negative.
676 	 */
677 	if (lp->d_secsize == DEV_BSIZE) {
678 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
679 	} else {
680 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
681 	}
682 	if (!sector_aligned || bp->b_blkno < 0) {
683 		bp->b_error = EINVAL;
684 		goto bad;
685 	}
686 	/*
687 	 * If it's a null transfer, return immediatly
688 	 */
689 	if (bp->b_bcount == 0)
690 		goto done;
691 
692 	/*
693 	 * Do bounds checking, adjust transfer. if error, process.
694 	 * If end of partition, just return.
695 	 */
696 	if (SDPART(bp->b_dev) == RAW_PART) {
697 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
698 		    sd->params.disksize512) <= 0)
699 			goto done;
700 	} else {
701 		if (bounds_check_with_label(&sd->sc_dk, bp,
702 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
703 			goto done;
704 	}
705 
706 	/*
707 	 * Now convert the block number to absolute and put it in
708 	 * terms of the device's logical block size.
709 	 */
710 	if (lp->d_secsize == DEV_BSIZE)
711 		blkno = bp->b_blkno;
712 	else if (lp->d_secsize > DEV_BSIZE)
713 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
714 	else
715 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
716 
717 	if (SDPART(bp->b_dev) != RAW_PART)
718 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
719 
720 	bp->b_rawblkno = blkno;
721 
722 	s = splbio();
723 
724 	/*
725 	 * Place it in the queue of disk activities for this disk.
726 	 *
727 	 * XXX Only do disksort() if the current operating mode does not
728 	 * XXX include tagged queueing.
729 	 */
730 	BUFQ_PUT(sd->buf_queue, bp);
731 
732 	/*
733 	 * Tell the device to get going on the transfer if it's
734 	 * not doing anything, otherwise just wait for completion
735 	 */
736 	sdstart(sd->sc_periph);
737 
738 	splx(s);
739 	return;
740 
741 bad:
742 	bp->b_flags |= B_ERROR;
743 done:
744 	/*
745 	 * Correctly set the buf to indicate a completed xfer
746 	 */
747 	bp->b_resid = bp->b_bcount;
748 	biodone(bp);
749 }
750 
751 /*
752  * sdstart looks to see if there is a buf waiting for the device
753  * and that the device is not already busy. If both are true,
754  * It dequeues the buf and creates a scsi command to perform the
755  * transfer in the buf. The transfer request will call scsipi_done
756  * on completion, which will in turn call this routine again
757  * so that the next queued transfer is performed.
758  * The bufs are queued by the strategy routine (sdstrategy)
759  *
760  * This routine is also called after other non-queued requests
761  * have been made of the scsi driver, to ensure that the queue
762  * continues to be drained.
763  *
764  * must be called at the correct (highish) spl level
765  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
766  */
767 static void
768 sdstart(struct scsipi_periph *periph)
769 {
770 	struct sd_softc *sd = (void *)periph->periph_dev;
771 	struct disklabel *lp = sd->sc_dk.dk_label;
772 	struct buf *bp = 0;
773 	struct scsipi_rw_16 cmd16;
774 	struct scsipi_rw_10 cmd_big;
775 	struct scsi_rw_6 cmd_small;
776 	struct scsipi_generic *cmdp;
777 	struct scsipi_xfer *xs;
778 	int nblks, cmdlen, error, flags;
779 
780 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
781 	/*
782 	 * Check if the device has room for another command
783 	 */
784 	while (periph->periph_active < periph->periph_openings) {
785 		/*
786 		 * there is excess capacity, but a special waits
787 		 * It'll need the adapter as soon as we clear out of the
788 		 * way and let it run (user level wait).
789 		 */
790 		if (periph->periph_flags & PERIPH_WAITING) {
791 			periph->periph_flags &= ~PERIPH_WAITING;
792 			wakeup((caddr_t)periph);
793 			return;
794 		}
795 
796 		/*
797 		 * If the device has become invalid, abort all the
798 		 * reads and writes until all files have been closed and
799 		 * re-opened
800 		 */
801 		if (__predict_false(
802 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
803 			if ((bp = BUFQ_GET(sd->buf_queue)) != NULL) {
804 				bp->b_error = EIO;
805 				bp->b_flags |= B_ERROR;
806 				bp->b_resid = bp->b_bcount;
807 				biodone(bp);
808 				continue;
809 			} else {
810 				return;
811 			}
812 		}
813 
814 		/*
815 		 * See if there is a buf with work for us to do..
816 		 */
817 		if ((bp = BUFQ_PEEK(sd->buf_queue)) == NULL)
818 			return;
819 
820 		/*
821 		 * We have a buf, now we should make a command.
822 		 */
823 
824 		if (lp->d_secsize == DEV_BSIZE)
825 			nblks = bp->b_bcount >> DEV_BSHIFT;
826 		else
827 			nblks = howmany(bp->b_bcount, lp->d_secsize);
828 
829 		/*
830 		 * Fill out the scsi command.  Use the smallest CDB possible
831 		 * (6-byte, 10-byte, or 16-byte).
832 		 */
833 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
834 		    ((nblks & 0xff) == nblks) &&
835 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
836 			/* 6-byte CDB */
837 			memset(&cmd_small, 0, sizeof(cmd_small));
838 			cmd_small.opcode = (bp->b_flags & B_READ) ?
839 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
840 			_lto3b(bp->b_rawblkno, cmd_small.addr);
841 			cmd_small.length = nblks & 0xff;
842 			cmdlen = sizeof(cmd_small);
843 			cmdp = (struct scsipi_generic *)&cmd_small;
844 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
845 			/* 10-byte CDB */
846 			memset(&cmd_big, 0, sizeof(cmd_big));
847 			cmd_big.opcode = (bp->b_flags & B_READ) ?
848 			    READ_10 : WRITE_10;
849 			_lto4b(bp->b_rawblkno, cmd_big.addr);
850 			_lto2b(nblks, cmd_big.length);
851 			cmdlen = sizeof(cmd_big);
852 			cmdp = (struct scsipi_generic *)&cmd_big;
853 		} else {
854 			/* 16-byte CDB */
855 			memset(&cmd16, 0, sizeof(cmd16));
856 			cmd16.opcode = (bp->b_flags & B_READ) ?
857 			    READ_16 : WRITE_16;
858 			_lto8b(bp->b_rawblkno, cmd16.addr);
859 			_lto4b(nblks, cmd16.length);
860 			cmdlen = sizeof(cmd16);
861 			cmdp = (struct scsipi_generic *)&cmd16;
862 		}
863 
864 		/* Instrumentation. */
865 		disk_busy(&sd->sc_dk);
866 
867 		/*
868 		 * Mark the disk dirty so that the cache will be
869 		 * flushed on close.
870 		 */
871 		if ((bp->b_flags & B_READ) == 0)
872 			sd->flags |= SDF_DIRTY;
873 
874 		/*
875 		 * Figure out what flags to use.
876 		 */
877 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
878 		if (bp->b_flags & B_READ)
879 			flags |= XS_CTL_DATA_IN;
880 		else
881 			flags |= XS_CTL_DATA_OUT;
882 
883 		/*
884 		 * Call the routine that chats with the adapter.
885 		 * Note: we cannot sleep as we may be an interrupt
886 		 */
887 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
888 		    (u_char *)bp->b_data, bp->b_bcount,
889 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
890 		if (__predict_false(xs == NULL)) {
891 			/*
892 			 * out of memory. Keep this buffer in the queue, and
893 			 * retry later.
894 			 */
895 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
896 			    periph);
897 			return;
898 		}
899 		/*
900 		 * need to dequeue the buffer before queuing the command,
901 		 * because cdstart may be called recursively from the
902 		 * HBA driver
903 		 */
904 #ifdef DIAGNOSTIC
905 		if (BUFQ_GET(sd->buf_queue) != bp)
906 			panic("sdstart(): dequeued wrong buf");
907 #else
908 		BUFQ_GET(sd->buf_queue);
909 #endif
910 		error = scsipi_execute_xs(xs);
911 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
912 		KASSERT(error == 0);
913 	}
914 }
915 
916 static void
917 sdrestart(void *v)
918 {
919 	int s = splbio();
920 	sdstart((struct scsipi_periph *)v);
921 	splx(s);
922 }
923 
924 static void
925 sddone(struct scsipi_xfer *xs, int error)
926 {
927 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
928 	struct buf *bp = xs->bp;
929 
930 	if (sd->flags & SDF_FLUSHING) {
931 		/* Flush completed, no longer dirty. */
932 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
933 	}
934 
935 	if (bp) {
936 		bp->b_error = error;
937 		bp->b_resid = xs->resid;
938 		if (error)
939 			bp->b_flags |= B_ERROR;
940 
941 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
942 		    (bp->b_flags & B_READ));
943 #if NRND > 0
944 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
945 #endif
946 
947 		biodone(bp);
948 	}
949 }
950 
951 static void
952 sdminphys(struct buf *bp)
953 {
954 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
955 	long xmax;
956 
957 	/*
958 	 * If the device is ancient, we want to make sure that
959 	 * the transfer fits into a 6-byte cdb.
960 	 *
961 	 * XXX Note that the SCSI-I spec says that 256-block transfers
962 	 * are allowed in a 6-byte read/write, and are specified
963 	 * by settng the "length" to 0.  However, we're conservative
964 	 * here, allowing only 255-block transfers in case an
965 	 * ancient device gets confused by length == 0.  A length of 0
966 	 * in a 10-byte read/write actually means 0 blocks.
967 	 */
968 	if ((sd->flags & SDF_ANCIENT) &&
969 	    ((sd->sc_periph->periph_flags &
970 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
971 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
972 
973 		if (bp->b_bcount > xmax)
974 			bp->b_bcount = xmax;
975 	}
976 
977 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
978 }
979 
980 static int
981 sdread(dev_t dev, struct uio *uio, int ioflag)
982 {
983 
984 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
985 }
986 
987 static int
988 sdwrite(dev_t dev, struct uio *uio, int ioflag)
989 {
990 
991 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
992 }
993 
994 /*
995  * Perform special action on behalf of the user
996  * Knows about the internals of this device
997  */
998 static int
999 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct lwp *l)
1000 {
1001 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1002 	struct scsipi_periph *periph = sd->sc_periph;
1003 	int part = SDPART(dev);
1004 	int error = 0;
1005 #ifdef __HAVE_OLD_DISKLABEL
1006 	struct disklabel *newlabel = NULL;
1007 #endif
1008 
1009 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1010 
1011 	/*
1012 	 * If the device is not valid, some IOCTLs can still be
1013 	 * handled on the raw partition. Check this here.
1014 	 */
1015 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1016 		switch (cmd) {
1017 		case DIOCKLABEL:
1018 		case DIOCWLABEL:
1019 		case DIOCLOCK:
1020 		case DIOCEJECT:
1021 		case ODIOCEJECT:
1022 		case DIOCGCACHE:
1023 		case DIOCSCACHE:
1024 		case SCIOCIDENTIFY:
1025 		case OSCIOCIDENTIFY:
1026 		case SCIOCCOMMAND:
1027 		case SCIOCDEBUG:
1028 			if (part == RAW_PART)
1029 				break;
1030 		/* FALLTHROUGH */
1031 		default:
1032 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1033 				return (ENODEV);
1034 			else
1035 				return (EIO);
1036 		}
1037 	}
1038 
1039 	switch (cmd) {
1040 	case DIOCGDINFO:
1041 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1042 		return (0);
1043 
1044 #ifdef __HAVE_OLD_DISKLABEL
1045 	case ODIOCGDINFO:
1046 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1047 		if (newlabel == NULL)
1048 			return EIO;
1049 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1050 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1051 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1052 		else
1053 			error = ENOTTY;
1054 		free(newlabel, M_TEMP);
1055 		return error;
1056 #endif
1057 
1058 	case DIOCGPART:
1059 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1060 		((struct partinfo *)addr)->part =
1061 		    &sd->sc_dk.dk_label->d_partitions[part];
1062 		return (0);
1063 
1064 	case DIOCWDINFO:
1065 	case DIOCSDINFO:
1066 #ifdef __HAVE_OLD_DISKLABEL
1067 	case ODIOCWDINFO:
1068 	case ODIOCSDINFO:
1069 #endif
1070 	{
1071 		struct disklabel *lp;
1072 
1073 		if ((flag & FWRITE) == 0)
1074 			return (EBADF);
1075 
1076 #ifdef __HAVE_OLD_DISKLABEL
1077  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1078 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1079 			if (newlabel == NULL)
1080 				return EIO;
1081 			memset(newlabel, 0, sizeof newlabel);
1082 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1083 			lp = newlabel;
1084 		} else
1085 #endif
1086 		lp = (struct disklabel *)addr;
1087 
1088 		if ((error = lockmgr(&sd->sc_dk.dk_openlock,
1089 				     LK_EXCLUSIVE, NULL)) != 0)
1090 			goto bad;
1091 		sd->flags |= SDF_LABELLING;
1092 
1093 		error = setdisklabel(sd->sc_dk.dk_label,
1094 		    lp, /*sd->sc_dk.dk_openmask : */0,
1095 		    sd->sc_dk.dk_cpulabel);
1096 		if (error == 0) {
1097 			if (cmd == DIOCWDINFO
1098 #ifdef __HAVE_OLD_DISKLABEL
1099 			    || cmd == ODIOCWDINFO
1100 #endif
1101 			   )
1102 				error = writedisklabel(SDLABELDEV(dev),
1103 				    sdstrategy, sd->sc_dk.dk_label,
1104 				    sd->sc_dk.dk_cpulabel);
1105 		}
1106 
1107 		sd->flags &= ~SDF_LABELLING;
1108 		(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1109 bad:
1110 #ifdef __HAVE_OLD_DISKLABEL
1111 		if (newlabel != NULL)
1112 			free(newlabel, M_TEMP);
1113 #endif
1114 		return (error);
1115 	}
1116 
1117 	case DIOCKLABEL:
1118 		if (*(int *)addr)
1119 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1120 		else
1121 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1122 		return (0);
1123 
1124 	case DIOCWLABEL:
1125 		if ((flag & FWRITE) == 0)
1126 			return (EBADF);
1127 		if (*(int *)addr)
1128 			sd->flags |= SDF_WLABEL;
1129 		else
1130 			sd->flags &= ~SDF_WLABEL;
1131 		return (0);
1132 
1133 	case DIOCLOCK:
1134 		if (periph->periph_flags & PERIPH_REMOVABLE)
1135 			return (scsipi_prevent(periph,
1136 			    (*(int *)addr) ?
1137 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1138 		else
1139 			return (ENOTTY);
1140 
1141 	case DIOCEJECT:
1142 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1143 			return (ENOTTY);
1144 		if (*(int *)addr == 0) {
1145 			/*
1146 			 * Don't force eject: check that we are the only
1147 			 * partition open. If so, unlock it.
1148 			 */
1149 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1150 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1151 			    sd->sc_dk.dk_openmask) {
1152 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1153 				    XS_CTL_IGNORE_NOT_READY);
1154 				if (error)
1155 					return (error);
1156 			} else {
1157 				return (EBUSY);
1158 			}
1159 		}
1160 		/* FALLTHROUGH */
1161 	case ODIOCEJECT:
1162 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1163 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1164 
1165 	case DIOCGDEFLABEL:
1166 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1167 		return (0);
1168 
1169 #ifdef __HAVE_OLD_DISKLABEL
1170 	case ODIOCGDEFLABEL:
1171 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1172 		if (newlabel == NULL)
1173 			return EIO;
1174 		sdgetdefaultlabel(sd, newlabel);
1175 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1176 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1177 		else
1178 			error = ENOTTY;
1179 		free(newlabel, M_TEMP);
1180 		return error;
1181 #endif
1182 
1183 	case DIOCGCACHE:
1184 		return (sd_getcache(sd, (int *) addr));
1185 
1186 	case DIOCSCACHE:
1187 		if ((flag & FWRITE) == 0)
1188 			return (EBADF);
1189 		return (sd_setcache(sd, *(int *) addr));
1190 
1191 	case DIOCCACHESYNC:
1192 		/*
1193 		 * XXX Do we really need to care about having a writable
1194 		 * file descriptor here?
1195 		 */
1196 		if ((flag & FWRITE) == 0)
1197 			return (EBADF);
1198 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1199 			error = sd_flush(sd, 0);
1200 			if (error)
1201 				sd->flags &= ~SDF_FLUSHING;
1202 			else
1203 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1204 		} else
1205 			error = 0;
1206 		return (error);
1207 
1208 	case DIOCAWEDGE:
1209 	    {
1210 	    	struct dkwedge_info *dkw = (void *) addr;
1211 
1212 		if ((flag & FWRITE) == 0)
1213 			return (EBADF);
1214 
1215 		/* If the ioctl happens here, the parent is us. */
1216 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1217 		return (dkwedge_add(dkw));
1218 	    }
1219 
1220 	case DIOCDWEDGE:
1221 	    {
1222 	    	struct dkwedge_info *dkw = (void *) addr;
1223 
1224 		if ((flag & FWRITE) == 0)
1225 			return (EBADF);
1226 
1227 		/* If the ioctl happens here, the parent is us. */
1228 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1229 		return (dkwedge_del(dkw));
1230 	    }
1231 
1232 	case DIOCLWEDGES:
1233 	    {
1234 	    	struct dkwedge_list *dkwl = (void *) addr;
1235 
1236 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1237 	    }
1238 
1239 	default:
1240 		if (part != RAW_PART)
1241 			return (ENOTTY);
1242 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1243 	}
1244 
1245 #ifdef DIAGNOSTIC
1246 	panic("sdioctl: impossible");
1247 #endif
1248 }
1249 
1250 static void
1251 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1252 {
1253 
1254 	memset(lp, 0, sizeof(struct disklabel));
1255 
1256 	lp->d_secsize = sd->params.blksize;
1257 	lp->d_ntracks = sd->params.heads;
1258 	lp->d_nsectors = sd->params.sectors;
1259 	lp->d_ncylinders = sd->params.cyls;
1260 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1261 
1262 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1263 	case SCSIPI_BUSTYPE_SCSI:
1264 		lp->d_type = DTYPE_SCSI;
1265 		break;
1266 	case SCSIPI_BUSTYPE_ATAPI:
1267 		lp->d_type = DTYPE_ATAPI;
1268 		break;
1269 	}
1270 	/*
1271 	 * XXX
1272 	 * We could probe the mode pages to figure out what kind of disc it is.
1273 	 * Is this worthwhile?
1274 	 */
1275 	strncpy(lp->d_typename, "mydisk", 16);
1276 	strncpy(lp->d_packname, "fictitious", 16);
1277 	lp->d_secperunit = sd->params.disksize;
1278 	lp->d_rpm = sd->params.rot_rate;
1279 	lp->d_interleave = 1;
1280 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1281 	    D_REMOVABLE : 0;
1282 
1283 	lp->d_partitions[RAW_PART].p_offset = 0;
1284 	lp->d_partitions[RAW_PART].p_size =
1285 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1286 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1287 	lp->d_npartitions = RAW_PART + 1;
1288 
1289 	lp->d_magic = DISKMAGIC;
1290 	lp->d_magic2 = DISKMAGIC;
1291 	lp->d_checksum = dkcksum(lp);
1292 }
1293 
1294 
1295 /*
1296  * Load the label information on the named device
1297  */
1298 static int
1299 sdgetdisklabel(struct sd_softc *sd)
1300 {
1301 	struct disklabel *lp = sd->sc_dk.dk_label;
1302 	const char *errstring;
1303 
1304 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1305 
1306 	sdgetdefaultlabel(sd, lp);
1307 
1308 	if (lp->d_secpercyl == 0) {
1309 		lp->d_secpercyl = 100;
1310 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1311 	}
1312 
1313 	/*
1314 	 * Call the generic disklabel extraction routine
1315 	 */
1316 	errstring = readdisklabel(MAKESDDEV(0, device_unit(&sd->sc_dev),
1317 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1318 	if (errstring) {
1319 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1320 		return EIO;
1321 	}
1322 	return 0;
1323 }
1324 
1325 static void
1326 sd_shutdown(void *arg)
1327 {
1328 	struct sd_softc *sd = arg;
1329 
1330 	/*
1331 	 * If the disk cache needs to be flushed, and the disk supports
1332 	 * it, flush it.  We're cold at this point, so we poll for
1333 	 * completion.
1334 	 */
1335 	if ((sd->flags & SDF_DIRTY) != 0) {
1336 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1337 			printf("%s: cache synchronization failed\n",
1338 			    sd->sc_dev.dv_xname);
1339 			sd->flags &= ~SDF_FLUSHING;
1340 		} else
1341 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1342 	}
1343 }
1344 
1345 /*
1346  * Check Errors
1347  */
1348 static int
1349 sd_interpret_sense(struct scsipi_xfer *xs)
1350 {
1351 	struct scsipi_periph *periph = xs->xs_periph;
1352 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1353 	struct sd_softc *sd = (void *)periph->periph_dev;
1354 	int s, error, retval = EJUSTRETURN;
1355 
1356 	/*
1357 	 * If the periph is already recovering, just do the normal
1358 	 * error processing.
1359 	 */
1360 	if (periph->periph_flags & PERIPH_RECOVERING)
1361 		return (retval);
1362 
1363 	/*
1364 	 * Ignore errors from accessing illegal fields (e.g. trying to
1365 	 * lock the door of a digicam, which doesn't have a door that
1366 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1367 	 */
1368 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1369 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1370 	    sense->asc == 0x24 &&
1371 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1372 		scsipi_printaddr(periph);
1373 		printf("no door lock\n");
1374 		periph->periph_flags &= ~PERIPH_REMOVABLE;
1375 		return 0;
1376 	}
1377 
1378 
1379 
1380 	/*
1381 	 * If the device is not open yet, let the generic code handle it.
1382 	 */
1383 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1384 		return (retval);
1385 
1386 	/*
1387 	 * If it isn't a extended or extended/deferred error, let
1388 	 * the generic code handle it.
1389 	 */
1390 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1391 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1392 		return (retval);
1393 
1394 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1395 	    sense->asc == 0x4) {
1396 		if (sense->ascq == 0x01)	{
1397 			/*
1398 			 * Unit In The Process Of Becoming Ready.
1399 			 */
1400 			printf("%s: waiting for pack to spin up...\n",
1401 			    sd->sc_dev.dv_xname);
1402 			if (!callout_pending(&periph->periph_callout))
1403 				scsipi_periph_freeze(periph, 1);
1404 			callout_reset(&periph->periph_callout,
1405 			    5 * hz, scsipi_periph_timed_thaw, periph);
1406 			retval = ERESTART;
1407 		} else if (sense->ascq == 0x02) {
1408 			printf("%s: pack is stopped, restarting...\n",
1409 			    sd->sc_dev.dv_xname);
1410 			s = splbio();
1411 			periph->periph_flags |= PERIPH_RECOVERING;
1412 			splx(s);
1413 			error = scsipi_start(periph, SSS_START,
1414 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1415 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1416 			if (error) {
1417 				printf("%s: unable to restart pack\n",
1418 				    sd->sc_dev.dv_xname);
1419 				retval = error;
1420 			} else
1421 				retval = ERESTART;
1422 			s = splbio();
1423 			periph->periph_flags &= ~PERIPH_RECOVERING;
1424 			splx(s);
1425 		}
1426 	}
1427 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1428 	    sense->asc == 0x31 &&
1429 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1430 		/* Medium Format Corrupted */
1431 		retval = EFTYPE;
1432 	}
1433 	return (retval);
1434 }
1435 
1436 
1437 static int
1438 sdsize(dev_t dev)
1439 {
1440 	struct sd_softc *sd;
1441 	int part, unit, omask;
1442 	int size;
1443 
1444 	unit = SDUNIT(dev);
1445 	if (unit >= sd_cd.cd_ndevs)
1446 		return (-1);
1447 	sd = sd_cd.cd_devs[unit];
1448 	if (sd == NULL)
1449 		return (-1);
1450 
1451 	if (!device_is_active(&sd->sc_dev))
1452 		return (-1);
1453 
1454 	part = SDPART(dev);
1455 	omask = sd->sc_dk.dk_openmask & (1 << part);
1456 
1457 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1458 		return (-1);
1459 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1460 		size = -1;
1461 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1462 		size = -1;
1463 	else
1464 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1465 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1466 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1467 		return (-1);
1468 	return (size);
1469 }
1470 
1471 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1472 static struct scsipi_xfer sx;
1473 static int sddoingadump;
1474 
1475 /*
1476  * dump all of physical memory into the partition specified, starting
1477  * at offset 'dumplo' into the partition.
1478  */
1479 static int
1480 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1481 {
1482 	struct sd_softc *sd;	/* disk unit to do the I/O */
1483 	struct disklabel *lp;	/* disk's disklabel */
1484 	int	unit, part;
1485 	int	sectorsize;	/* size of a disk sector */
1486 	int	nsects;		/* number of sectors in partition */
1487 	int	sectoff;	/* sector offset of partition */
1488 	int	totwrt;		/* total number of sectors left to write */
1489 	int	nwrt;		/* current number of sectors to write */
1490 	struct scsipi_rw_10 cmd;	/* write command */
1491 	struct scsipi_xfer *xs;	/* ... convenience */
1492 	struct scsipi_periph *periph;
1493 	struct scsipi_channel *chan;
1494 
1495 	/* Check if recursive dump; if so, punt. */
1496 	if (sddoingadump)
1497 		return (EFAULT);
1498 
1499 	/* Mark as active early. */
1500 	sddoingadump = 1;
1501 
1502 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1503 	part = SDPART(dev);
1504 
1505 	/* Check for acceptable drive number. */
1506 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1507 		return (ENXIO);
1508 
1509 	if (!device_is_active(&sd->sc_dev))
1510 		return (ENODEV);
1511 
1512 	periph = sd->sc_periph;
1513 	chan = periph->periph_channel;
1514 
1515 	/* Make sure it was initialized. */
1516 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1517 		return (ENXIO);
1518 
1519 	/* Convert to disk sectors.  Request must be a multiple of size. */
1520 	lp = sd->sc_dk.dk_label;
1521 	sectorsize = lp->d_secsize;
1522 	if ((size % sectorsize) != 0)
1523 		return (EFAULT);
1524 	totwrt = size / sectorsize;
1525 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1526 
1527 	nsects = lp->d_partitions[part].p_size;
1528 	sectoff = lp->d_partitions[part].p_offset;
1529 
1530 	/* Check transfer bounds against partition size. */
1531 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1532 		return (EINVAL);
1533 
1534 	/* Offset block number to start of partition. */
1535 	blkno += sectoff;
1536 
1537 	xs = &sx;
1538 
1539 	while (totwrt > 0) {
1540 		nwrt = totwrt;		/* XXX */
1541 #ifndef	SD_DUMP_NOT_TRUSTED
1542 		/*
1543 		 *  Fill out the scsi command
1544 		 */
1545 		memset(&cmd, 0, sizeof(cmd));
1546 		cmd.opcode = WRITE_10;
1547 		_lto4b(blkno, cmd.addr);
1548 		_lto2b(nwrt, cmd.length);
1549 		/*
1550 		 * Fill out the scsipi_xfer structure
1551 		 *    Note: we cannot sleep as we may be an interrupt
1552 		 * don't use scsipi_command() as it may want to wait
1553 		 * for an xs.
1554 		 */
1555 		memset(xs, 0, sizeof(sx));
1556 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1557 		    XS_CTL_DATA_OUT;
1558 		xs->xs_status = 0;
1559 		xs->xs_periph = periph;
1560 		xs->xs_retries = SDRETRIES;
1561 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1562 		xs->cmd = (struct scsipi_generic *)&cmd;
1563 		xs->cmdlen = sizeof(cmd);
1564 		xs->resid = nwrt * sectorsize;
1565 		xs->error = XS_NOERROR;
1566 		xs->bp = 0;
1567 		xs->data = va;
1568 		xs->datalen = nwrt * sectorsize;
1569 
1570 		/*
1571 		 * Pass all this info to the scsi driver.
1572 		 */
1573 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1574 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1575 		    xs->error != XS_NOERROR)
1576 			return (EIO);
1577 #else	/* SD_DUMP_NOT_TRUSTED */
1578 		/* Let's just talk about this first... */
1579 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1580 		delay(500 * 1000);	/* half a second */
1581 #endif	/* SD_DUMP_NOT_TRUSTED */
1582 
1583 		/* update block count */
1584 		totwrt -= nwrt;
1585 		blkno += nwrt;
1586 		va += sectorsize * nwrt;
1587 	}
1588 	sddoingadump = 0;
1589 	return (0);
1590 }
1591 
1592 static int
1593 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1594     int page, int flags, int *big)
1595 {
1596 
1597 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1598 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1599 		*big = 1;
1600 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1601 		    size + sizeof(struct scsi_mode_parameter_header_10),
1602 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1603 	} else {
1604 		*big = 0;
1605 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1606 		    size + sizeof(struct scsi_mode_parameter_header_6),
1607 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1608 	}
1609 }
1610 
1611 static int
1612 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1613     int flags, int big)
1614 {
1615 
1616 	if (big) {
1617 		struct scsi_mode_parameter_header_10 *header = sense;
1618 
1619 		_lto2b(0, header->data_length);
1620 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1621 		    size + sizeof(struct scsi_mode_parameter_header_10),
1622 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1623 	} else {
1624 		struct scsi_mode_parameter_header_6 *header = sense;
1625 
1626 		header->data_length = 0;
1627 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1628 		    size + sizeof(struct scsi_mode_parameter_header_6),
1629 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1630 	}
1631 }
1632 
1633 static int
1634 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1635 {
1636 	struct {
1637 		struct scsi_mode_parameter_header_6 header;
1638 		/* no block descriptor */
1639 		u_int8_t pg_code; /* page code (should be 6) */
1640 		u_int8_t pg_length; /* page length (should be 11) */
1641 		u_int8_t wcd; /* bit0: cache disable */
1642 		u_int8_t lbs[2]; /* logical block size */
1643 		u_int8_t size[5]; /* number of log. blocks */
1644 		u_int8_t pp; /* power/performance */
1645 		u_int8_t flags;
1646 		u_int8_t resvd;
1647 	} scsipi_sense;
1648 	u_int64_t sectors;
1649 	int error;
1650 
1651 	/*
1652 	 * scsipi_size (ie "read capacity") and mode sense page 6
1653 	 * give the same information. Do both for now, and check
1654 	 * for consistency.
1655 	 * XXX probably differs for removable media
1656 	 */
1657 	dp->blksize = 512;
1658 	if ((sectors = scsipi_size(sd->sc_periph, flags)) == 0)
1659 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1660 
1661 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1662 	    &scsipi_sense.header, sizeof(scsipi_sense),
1663 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1664 
1665 	if (error != 0)
1666 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1667 
1668 	dp->blksize = _2btol(scsipi_sense.lbs);
1669 	if (dp->blksize == 0)
1670 		dp->blksize = 512;
1671 
1672 	/*
1673 	 * Create a pseudo-geometry.
1674 	 */
1675 	dp->heads = 64;
1676 	dp->sectors = 32;
1677 	dp->cyls = sectors / (dp->heads * dp->sectors);
1678 	dp->disksize = _5btol(scsipi_sense.size);
1679 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1680 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1681 		       (unsigned long long)dp->disksize,
1682 		       (unsigned long long)sectors);
1683 		dp->disksize = sectors;
1684 	}
1685 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1686 
1687 	return (SDGP_RESULT_OK);
1688 }
1689 
1690 /*
1691  * Get the scsi driver to send a full inquiry to the * device and use the
1692  * results to fill out the disk parameter structure.
1693  */
1694 static int
1695 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1696 {
1697 	u_int64_t sectors;
1698 	int error;
1699 #if 0
1700 	int i;
1701 	u_int8_t *p;
1702 #endif
1703 
1704 	dp->disksize = sectors = scsipi_size(sd->sc_periph, flags);
1705 	if (sectors == 0) {
1706 		struct scsipi_read_format_capacities cmd;
1707 		struct {
1708 			struct scsipi_capacity_list_header header;
1709 			struct scsipi_capacity_descriptor desc;
1710 		} __attribute__((packed)) data;
1711 
1712 		memset(&cmd, 0, sizeof(cmd));
1713 		memset(&data, 0, sizeof(data));
1714 		cmd.opcode = READ_FORMAT_CAPACITIES;
1715 		_lto2b(sizeof(data), cmd.length);
1716 
1717 		error = scsipi_command(sd->sc_periph,
1718 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1719 		    SDRETRIES, 20000, NULL,
1720 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1721 		if (error == EFTYPE) {
1722 			/* Medium Format Corrupted, handle as not formatted */
1723 			return (SDGP_RESULT_UNFORMATTED);
1724 		}
1725 		if (error || data.header.length == 0)
1726 			return (SDGP_RESULT_OFFLINE);
1727 
1728 #if 0
1729 printf("rfc: length=%d\n", data.header.length);
1730 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");
1731 #endif
1732 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1733 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1734 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1735 			break;
1736 
1737 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1738 			return (SDGP_RESULT_UNFORMATTED);
1739 
1740 		case SCSIPI_CAP_DESC_CODE_NONE:
1741 			return (SDGP_RESULT_OFFLINE);
1742 		}
1743 
1744 		dp->disksize = sectors = _4btol(data.desc.nblks);
1745 		if (sectors == 0)
1746 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1747 
1748 		dp->blksize = _3btol(data.desc.blklen);
1749 		if (dp->blksize == 0)
1750 			dp->blksize = 512;
1751 	} else {
1752 		struct sd_mode_sense_data scsipi_sense;
1753 		int big, bsize;
1754 		struct scsi_general_block_descriptor *bdesc;
1755 
1756 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1757 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1758 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1759 		dp->blksize = 512;
1760 		if (!error) {
1761 			if (big) {
1762 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1763 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1764 			} else {
1765 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1766 				bsize = scsipi_sense.header.small.blk_desc_len;
1767 			}
1768 
1769 #if 0
1770 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1771 printf("page 0 bsize=%d\n", bsize);
1772 printf("page 0 ok\n");
1773 #endif
1774 
1775 			if (bsize >= 8) {
1776 				dp->blksize = _3btol(bdesc->blklen);
1777 				if (dp->blksize == 0)
1778 					dp->blksize = 512;
1779 			}
1780 		}
1781 	}
1782 
1783 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1784 	return (0);
1785 }
1786 
1787 static int
1788 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1789 {
1790 	struct sd_mode_sense_data scsipi_sense;
1791 	int error;
1792 	int big, poffset, byte2;
1793 	union scsi_disk_pages *pages;
1794 #if 0
1795 	int i;
1796 	u_int8_t *p;
1797 #endif
1798 
1799 	byte2 = SMS_DBD;
1800 again:
1801 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1802 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1803 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1804 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1805 	    flags | XS_CTL_SILENT, &big);
1806 	if (error) {
1807 		if (byte2 == SMS_DBD) {
1808 			/* No result; try once more with DBD off */
1809 			byte2 = 0;
1810 			goto again;
1811 		}
1812 		return (error);
1813 	}
1814 
1815 	if (big) {
1816 		poffset = sizeof scsipi_sense.header.big;
1817 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1818 	} else {
1819 		poffset = sizeof scsipi_sense.header.small;
1820 		poffset += scsipi_sense.header.small.blk_desc_len;
1821 	}
1822 
1823 	pages = (void *)((u_long)&scsipi_sense + poffset);
1824 #if 0
1825 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1826 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1827 #endif
1828 
1829 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1830 		return (ERESTART);
1831 
1832 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1833 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1834 	    _3btol(pages->rigid_geometry.ncyl),
1835 	    pages->rigid_geometry.nheads,
1836 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1837 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1838 	    _2btol(pages->rigid_geometry.land_zone)));
1839 
1840 	/*
1841 	 * KLUDGE!! (for zone recorded disks)
1842 	 * give a number of sectors so that sec * trks * cyls
1843 	 * is <= disk_size
1844 	 * can lead to wasted space! THINK ABOUT THIS !
1845 	 */
1846 	dp->heads = pages->rigid_geometry.nheads;
1847 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1848 	if (dp->heads == 0 || dp->cyls == 0)
1849 		return (ERESTART);
1850 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1851 
1852 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1853 	if (dp->rot_rate == 0)
1854 		dp->rot_rate = 3600;
1855 
1856 #if 0
1857 printf("page 4 ok\n");
1858 #endif
1859 	return (0);
1860 }
1861 
1862 static int
1863 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1864 {
1865 	struct sd_mode_sense_data scsipi_sense;
1866 	int error;
1867 	int big, poffset, byte2;
1868 	union scsi_disk_pages *pages;
1869 #if 0
1870 	int i;
1871 	u_int8_t *p;
1872 #endif
1873 
1874 	byte2 = SMS_DBD;
1875 again:
1876 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1877 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1878 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1879 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1880 	    flags | XS_CTL_SILENT, &big);
1881 	if (error) {
1882 		if (byte2 == SMS_DBD) {
1883 			/* No result; try once more with DBD off */
1884 			byte2 = 0;
1885 			goto again;
1886 		}
1887 		return (error);
1888 	}
1889 
1890 	if (big) {
1891 		poffset = sizeof scsipi_sense.header.big;
1892 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1893 	} else {
1894 		poffset = sizeof scsipi_sense.header.small;
1895 		poffset += scsipi_sense.header.small.blk_desc_len;
1896 	}
1897 
1898 	pages = (void *)((u_long)&scsipi_sense + poffset);
1899 #if 0
1900 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1901 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1902 #endif
1903 
1904 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1905 		return (ERESTART);
1906 
1907 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1908 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1909 	    _3btol(pages->flex_geometry.ncyl),
1910 	    pages->flex_geometry.nheads,
1911 	    pages->flex_geometry.ph_sec_tr,
1912 	    _2btol(pages->flex_geometry.bytes_s)));
1913 
1914 	dp->heads = pages->flex_geometry.nheads;
1915 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1916 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1917 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1918 		return (ERESTART);
1919 
1920 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1921 	if (dp->rot_rate == 0)
1922 		dp->rot_rate = 3600;
1923 
1924 #if 0
1925 printf("page 5 ok\n");
1926 #endif
1927 	return (0);
1928 }
1929 
1930 static int
1931 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1932 {
1933 	int error;
1934 
1935 	/*
1936 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1937 	 * cleared by the caller if necessary.
1938 	 */
1939 	if (sd->type == T_SIMPLE_DIRECT)
1940 		return (sd_get_simplifiedparms(sd, dp, flags));
1941 
1942 	error = sd_get_capacity(sd, dp, flags);
1943 	if (error)
1944 		return (error);
1945 
1946 	if (sd->type == T_OPTICAL)
1947 		goto page0;
1948 
1949 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1950 		if (!sd_get_parms_page5(sd, dp, flags) ||
1951 		    !sd_get_parms_page4(sd, dp, flags))
1952 			return (SDGP_RESULT_OK);
1953 	} else {
1954 		if (!sd_get_parms_page4(sd, dp, flags) ||
1955 		    !sd_get_parms_page5(sd, dp, flags))
1956 			return (SDGP_RESULT_OK);
1957 	}
1958 
1959 page0:
1960 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1961 	/* Try calling driver's method for figuring out geometry. */
1962 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1963 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1964 		(sd->sc_periph, dp, dp->disksize)) {
1965 		/*
1966 		 * Use adaptec standard fictitious geometry
1967 		 * this depends on which controller (e.g. 1542C is
1968 		 * different. but we have to put SOMETHING here..)
1969 		 */
1970 		dp->heads = 64;
1971 		dp->sectors = 32;
1972 		dp->cyls = dp->disksize / (64 * 32);
1973 	}
1974 	dp->rot_rate = 3600;
1975 	return (SDGP_RESULT_OK);
1976 }
1977 
1978 static int
1979 sd_flush(struct sd_softc *sd, int flags)
1980 {
1981 	struct scsipi_periph *periph = sd->sc_periph;
1982 	struct scsi_synchronize_cache_10 cmd;
1983 
1984 	/*
1985 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1986 	 * We issue with address 0 length 0, which should be
1987 	 * interpreted by the device as "all remaining blocks
1988 	 * starting at address 0".  We ignore ILLEGAL REQUEST
1989 	 * in the event that the command is not supported by
1990 	 * the device, and poll for completion so that we know
1991 	 * that the cache has actually been flushed.
1992 	 *
1993 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
1994 	 * command, as indicated by our quirks flags.
1995 	 *
1996 	 * XXX What about older devices?
1997 	 */
1998 	if (periph->periph_version < 2 ||
1999 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2000 		return (0);
2001 
2002 	sd->flags |= SDF_FLUSHING;
2003 	memset(&cmd, 0, sizeof(cmd));
2004 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2005 
2006 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2007 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2008 }
2009 
2010 static int
2011 sd_getcache(struct sd_softc *sd, int *bitsp)
2012 {
2013 	struct scsipi_periph *periph = sd->sc_periph;
2014 	struct sd_mode_sense_data scsipi_sense;
2015 	int error, bits = 0;
2016 	int big;
2017 	union scsi_disk_pages *pages;
2018 
2019 	if (periph->periph_version < 2)
2020 		return (EOPNOTSUPP);
2021 
2022 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2023 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2024 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2025 	if (error)
2026 		return (error);
2027 
2028 	if (big)
2029 		pages = (void *)(&scsipi_sense.header.big + 1);
2030 	else
2031 		pages = (void *)(&scsipi_sense.header.small + 1);
2032 
2033 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2034 		bits |= DKCACHE_READ;
2035 	if (pages->caching_params.flags & CACHING_WCE)
2036 		bits |= DKCACHE_WRITE;
2037 	if (pages->caching_params.pg_code & PGCODE_PS)
2038 		bits |= DKCACHE_SAVE;
2039 
2040 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2041 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2042 	    sizeof(scsipi_sense.pages.caching_params),
2043 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2044 	if (error == 0) {
2045 		if (big)
2046 			pages = (void *)(&scsipi_sense.header.big + 1);
2047 		else
2048 			pages = (void *)(&scsipi_sense.header.small + 1);
2049 
2050 		if (pages->caching_params.flags & CACHING_RCD)
2051 			bits |= DKCACHE_RCHANGE;
2052 		if (pages->caching_params.flags & CACHING_WCE)
2053 			bits |= DKCACHE_WCHANGE;
2054 	}
2055 
2056 	*bitsp = bits;
2057 
2058 	return (0);
2059 }
2060 
2061 static int
2062 sd_setcache(struct sd_softc *sd, int bits)
2063 {
2064 	struct scsipi_periph *periph = sd->sc_periph;
2065 	struct sd_mode_sense_data scsipi_sense;
2066 	int error;
2067 	uint8_t oflags, byte2 = 0;
2068 	int big;
2069 	union scsi_disk_pages *pages;
2070 
2071 	if (periph->periph_version < 2)
2072 		return (EOPNOTSUPP);
2073 
2074 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2075 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2076 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2077 	if (error)
2078 		return (error);
2079 
2080 	if (big)
2081 		pages = (void *)(&scsipi_sense.header.big + 1);
2082 	else
2083 		pages = (void *)(&scsipi_sense.header.small + 1);
2084 
2085 	oflags = pages->caching_params.flags;
2086 
2087 	if (bits & DKCACHE_READ)
2088 		pages->caching_params.flags &= ~CACHING_RCD;
2089 	else
2090 		pages->caching_params.flags |= CACHING_RCD;
2091 
2092 	if (bits & DKCACHE_WRITE)
2093 		pages->caching_params.flags |= CACHING_WCE;
2094 	else
2095 		pages->caching_params.flags &= ~CACHING_WCE;
2096 
2097 	if (oflags == pages->caching_params.flags)
2098 		return (0);
2099 
2100 	pages->caching_params.pg_code &= PGCODE_MASK;
2101 
2102 	if (bits & DKCACHE_SAVE)
2103 		byte2 |= SMS_SP;
2104 
2105 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2106 	    sizeof(struct scsi_mode_page_header) +
2107 	    pages->caching_params.pg_length, 0, big));
2108 }
2109