xref: /netbsd-src/sys/dev/scsipi/sd.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /*	$NetBSD: sd.c,v 1.255 2006/11/25 12:06:31 scw 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 1.255 2006/11/25 12:06:31 scw 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,
198     void *aux)
199 {
200 	struct scsipibus_attach_args *sa = aux;
201 	int priority;
202 
203 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
204 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
205 	    sizeof(sd_patterns[0]), &priority);
206 
207 	return (priority);
208 }
209 
210 /*
211  * Attach routine common to atapi & scsi.
212  */
213 static void
214 sdattach(struct device *parent, struct device *self, void *aux)
215 {
216 	struct sd_softc *sd = device_private(self);
217 	struct scsipibus_attach_args *sa = aux;
218 	struct scsipi_periph *periph = sa->sa_periph;
219 	int error, result;
220 	struct disk_parms *dp = &sd->params;
221 	char pbuf[9];
222 
223 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
224 
225 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
226 	if (sd->type == T_SIMPLE_DIRECT)
227 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
228 
229 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
230 	    periph->periph_version == 0)
231 		sd->flags |= SDF_ANCIENT;
232 
233 	bufq_alloc(&sd->buf_queue, 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 = device_private(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(device_unit(self), 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 lwp *l)
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 (!device_is_active(&sd->sc_dev))
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 			    XS_CTL_SILENT);
504 			if (error)
505 				goto bad3;
506 		}
507 
508 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
509 			int param_error;
510 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
511 
512 			/*
513 			 * Load the physical device parameters.
514 			 *
515 			 * Note that if media is present but unformatted,
516 			 * we allow the open (so that it can be formatted!).
517 			 * The drive should refuse real I/O, if the media is
518 			 * unformatted.
519 			 */
520 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
521 			     == SDGP_RESULT_OFFLINE) {
522 				error = ENXIO;
523 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
524 				goto bad3;
525 			}
526 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
527 
528 			/* Load the partition info if not already loaded. */
529 			if (param_error == 0) {
530 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
531 					error = EIO;
532 					goto bad3;
533 				}
534 				SC_DEBUG(periph, SCSIPI_DB3,
535 				     ("Disklabel loaded "));
536 			}
537 		}
538 	}
539 
540 	/* Check that the partition exists. */
541 	if (part != RAW_PART &&
542 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
543 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
544 		error = ENXIO;
545 		goto bad3;
546 	}
547 
548  out:	/* Insure only one open at a time. */
549 	switch (fmt) {
550 	case S_IFCHR:
551 		sd->sc_dk.dk_copenmask |= (1 << part);
552 		break;
553 	case S_IFBLK:
554 		sd->sc_dk.dk_bopenmask |= (1 << part);
555 		break;
556 	}
557 	sd->sc_dk.dk_openmask =
558 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
559 
560 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
561 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
562 	return (0);
563 
564  bad3:
565 	if (sd->sc_dk.dk_openmask == 0) {
566 		if (periph->periph_flags & PERIPH_REMOVABLE)
567 			scsipi_prevent(periph, SPAMR_ALLOW,
568 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
569 			    XS_CTL_IGNORE_MEDIA_CHANGE |
570 			    XS_CTL_SILENT);
571 		periph->periph_flags &= ~PERIPH_OPEN;
572 	}
573 
574  bad2:
575 	if (sd->sc_dk.dk_openmask == 0)
576 		scsipi_adapter_delref(adapt);
577 
578  bad1:
579 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
580 	return (error);
581 }
582 
583 /*
584  * close the device.. only called if we are the LAST occurence of an open
585  * device.  Convenient now but usually a pain.
586  */
587 static int
588 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
589 {
590 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
591 	struct scsipi_periph *periph = sd->sc_periph;
592 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
593 	int part = SDPART(dev);
594 	int error;
595 
596 	if ((error = lockmgr(&sd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
597 		return (error);
598 
599 	switch (fmt) {
600 	case S_IFCHR:
601 		sd->sc_dk.dk_copenmask &= ~(1 << part);
602 		break;
603 	case S_IFBLK:
604 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
605 		break;
606 	}
607 	sd->sc_dk.dk_openmask =
608 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
609 
610 	if (sd->sc_dk.dk_openmask == 0) {
611 		/*
612 		 * If the disk cache needs flushing, and the disk supports
613 		 * it, do it now.
614 		 */
615 		if ((sd->flags & SDF_DIRTY) != 0) {
616 			if (sd_flush(sd, 0)) {
617 				printf("%s: cache synchronization failed\n",
618 				    sd->sc_dev.dv_xname);
619 				sd->flags &= ~SDF_FLUSHING;
620 			} else
621 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
622 		}
623 
624 		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 			    XS_CTL_SILENT);
631 		periph->periph_flags &= ~PERIPH_OPEN;
632 
633 		scsipi_wait_drain(periph);
634 
635 		scsipi_adapter_delref(adapt);
636 	}
637 
638 	(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
639 	return (0);
640 }
641 
642 /*
643  * Actually translate the requested transfer into one the physical driver
644  * can understand.  The transfer is described by a buf and will include
645  * only one physical transfer.
646  */
647 static void
648 sdstrategy(struct buf *bp)
649 {
650 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
651 	struct scsipi_periph *periph = sd->sc_periph;
652 	struct disklabel *lp;
653 	daddr_t blkno;
654 	int s;
655 	boolean_t sector_aligned;
656 
657 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
658 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
659 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
660 	/*
661 	 * If the device has been made invalid, error out
662 	 */
663 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
664 	    !device_is_active(&sd->sc_dev)) {
665 		if (periph->periph_flags & PERIPH_OPEN)
666 			bp->b_error = EIO;
667 		else
668 			bp->b_error = ENODEV;
669 		goto bad;
670 	}
671 
672 	lp = sd->sc_dk.dk_label;
673 
674 	/*
675 	 * The transfer must be a whole number of blocks, offset must not be
676 	 * negative.
677 	 */
678 	if (lp->d_secsize == DEV_BSIZE) {
679 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
680 	} else {
681 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
682 	}
683 	if (!sector_aligned || bp->b_blkno < 0) {
684 		bp->b_error = EINVAL;
685 		goto bad;
686 	}
687 	/*
688 	 * If it's a null transfer, return immediatly
689 	 */
690 	if (bp->b_bcount == 0)
691 		goto done;
692 
693 	/*
694 	 * Do bounds checking, adjust transfer. if error, process.
695 	 * If end of partition, just return.
696 	 */
697 	if (SDPART(bp->b_dev) == RAW_PART) {
698 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
699 		    sd->params.disksize512) <= 0)
700 			goto done;
701 	} else {
702 		if (bounds_check_with_label(&sd->sc_dk, bp,
703 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
704 			goto done;
705 	}
706 
707 	/*
708 	 * Now convert the block number to absolute and put it in
709 	 * terms of the device's logical block size.
710 	 */
711 	if (lp->d_secsize == DEV_BSIZE)
712 		blkno = bp->b_blkno;
713 	else if (lp->d_secsize > DEV_BSIZE)
714 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
715 	else
716 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
717 
718 	if (SDPART(bp->b_dev) != RAW_PART)
719 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
720 
721 	bp->b_rawblkno = blkno;
722 
723 	s = splbio();
724 
725 	/*
726 	 * Place it in the queue of disk activities for this disk.
727 	 *
728 	 * XXX Only do disksort() if the current operating mode does not
729 	 * XXX include tagged queueing.
730 	 */
731 	BUFQ_PUT(sd->buf_queue, bp);
732 
733 	/*
734 	 * Tell the device to get going on the transfer if it's
735 	 * not doing anything, otherwise just wait for completion
736 	 */
737 	sdstart(sd->sc_periph);
738 
739 	splx(s);
740 	return;
741 
742 bad:
743 	bp->b_flags |= B_ERROR;
744 done:
745 	/*
746 	 * Correctly set the buf to indicate a completed xfer
747 	 */
748 	bp->b_resid = bp->b_bcount;
749 	biodone(bp);
750 }
751 
752 /*
753  * sdstart looks to see if there is a buf waiting for the device
754  * and that the device is not already busy. If both are true,
755  * It dequeues the buf and creates a scsi command to perform the
756  * transfer in the buf. The transfer request will call scsipi_done
757  * on completion, which will in turn call this routine again
758  * so that the next queued transfer is performed.
759  * The bufs are queued by the strategy routine (sdstrategy)
760  *
761  * This routine is also called after other non-queued requests
762  * have been made of the scsi driver, to ensure that the queue
763  * continues to be drained.
764  *
765  * must be called at the correct (highish) spl level
766  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
767  */
768 static void
769 sdstart(struct scsipi_periph *periph)
770 {
771 	struct sd_softc *sd = (void *)periph->periph_dev;
772 	struct disklabel *lp = sd->sc_dk.dk_label;
773 	struct buf *bp = 0;
774 	struct scsipi_rw_16 cmd16;
775 	struct scsipi_rw_10 cmd_big;
776 	struct scsi_rw_6 cmd_small;
777 	struct scsipi_generic *cmdp;
778 	struct scsipi_xfer *xs;
779 	int nblks, cmdlen, error, flags;
780 
781 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
782 	/*
783 	 * Check if the device has room for another command
784 	 */
785 	while (periph->periph_active < periph->periph_openings) {
786 		/*
787 		 * there is excess capacity, but a special waits
788 		 * It'll need the adapter as soon as we clear out of the
789 		 * way and let it run (user level wait).
790 		 */
791 		if (periph->periph_flags & PERIPH_WAITING) {
792 			periph->periph_flags &= ~PERIPH_WAITING;
793 			wakeup((caddr_t)periph);
794 			return;
795 		}
796 
797 		/*
798 		 * If the device has become invalid, abort all the
799 		 * reads and writes until all files have been closed and
800 		 * re-opened
801 		 */
802 		if (__predict_false(
803 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
804 			if ((bp = BUFQ_GET(sd->buf_queue)) != NULL) {
805 				bp->b_error = EIO;
806 				bp->b_flags |= B_ERROR;
807 				bp->b_resid = bp->b_bcount;
808 				biodone(bp);
809 				continue;
810 			} else {
811 				return;
812 			}
813 		}
814 
815 		/*
816 		 * See if there is a buf with work for us to do..
817 		 */
818 		if ((bp = BUFQ_PEEK(sd->buf_queue)) == NULL)
819 			return;
820 
821 		/*
822 		 * We have a buf, now we should make a command.
823 		 */
824 
825 		if (lp->d_secsize == DEV_BSIZE)
826 			nblks = bp->b_bcount >> DEV_BSHIFT;
827 		else
828 			nblks = howmany(bp->b_bcount, lp->d_secsize);
829 
830 		/*
831 		 * Fill out the scsi command.  Use the smallest CDB possible
832 		 * (6-byte, 10-byte, or 16-byte).
833 		 */
834 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
835 		    ((nblks & 0xff) == nblks) &&
836 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
837 			/* 6-byte CDB */
838 			memset(&cmd_small, 0, sizeof(cmd_small));
839 			cmd_small.opcode = (bp->b_flags & B_READ) ?
840 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
841 			_lto3b(bp->b_rawblkno, cmd_small.addr);
842 			cmd_small.length = nblks & 0xff;
843 			cmdlen = sizeof(cmd_small);
844 			cmdp = (struct scsipi_generic *)&cmd_small;
845 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
846 			/* 10-byte CDB */
847 			memset(&cmd_big, 0, sizeof(cmd_big));
848 			cmd_big.opcode = (bp->b_flags & B_READ) ?
849 			    READ_10 : WRITE_10;
850 			_lto4b(bp->b_rawblkno, cmd_big.addr);
851 			_lto2b(nblks, cmd_big.length);
852 			cmdlen = sizeof(cmd_big);
853 			cmdp = (struct scsipi_generic *)&cmd_big;
854 		} else {
855 			/* 16-byte CDB */
856 			memset(&cmd16, 0, sizeof(cmd16));
857 			cmd16.opcode = (bp->b_flags & B_READ) ?
858 			    READ_16 : WRITE_16;
859 			_lto8b(bp->b_rawblkno, cmd16.addr);
860 			_lto4b(nblks, cmd16.length);
861 			cmdlen = sizeof(cmd16);
862 			cmdp = (struct scsipi_generic *)&cmd16;
863 		}
864 
865 		/* Instrumentation. */
866 		disk_busy(&sd->sc_dk);
867 
868 		/*
869 		 * Mark the disk dirty so that the cache will be
870 		 * flushed on close.
871 		 */
872 		if ((bp->b_flags & B_READ) == 0)
873 			sd->flags |= SDF_DIRTY;
874 
875 		/*
876 		 * Figure out what flags to use.
877 		 */
878 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
879 		if (bp->b_flags & B_READ)
880 			flags |= XS_CTL_DATA_IN;
881 		else
882 			flags |= XS_CTL_DATA_OUT;
883 
884 		/*
885 		 * Call the routine that chats with the adapter.
886 		 * Note: we cannot sleep as we may be an interrupt
887 		 */
888 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
889 		    (u_char *)bp->b_data, bp->b_bcount,
890 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
891 		if (__predict_false(xs == NULL)) {
892 			/*
893 			 * out of memory. Keep this buffer in the queue, and
894 			 * retry later.
895 			 */
896 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
897 			    periph);
898 			return;
899 		}
900 		/*
901 		 * need to dequeue the buffer before queuing the command,
902 		 * because cdstart may be called recursively from the
903 		 * HBA driver
904 		 */
905 #ifdef DIAGNOSTIC
906 		if (BUFQ_GET(sd->buf_queue) != bp)
907 			panic("sdstart(): dequeued wrong buf");
908 #else
909 		BUFQ_GET(sd->buf_queue);
910 #endif
911 		error = scsipi_execute_xs(xs);
912 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
913 		KASSERT(error == 0);
914 	}
915 }
916 
917 static void
918 sdrestart(void *v)
919 {
920 	int s = splbio();
921 	sdstart((struct scsipi_periph *)v);
922 	splx(s);
923 }
924 
925 static void
926 sddone(struct scsipi_xfer *xs, int error)
927 {
928 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
929 	struct buf *bp = xs->bp;
930 
931 	if (sd->flags & SDF_FLUSHING) {
932 		/* Flush completed, no longer dirty. */
933 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
934 	}
935 
936 	if (bp) {
937 		bp->b_error = error;
938 		bp->b_resid = xs->resid;
939 		if (error) {
940 			/* on a read/write error bp->b_resid is zero, so fix */
941 			bp->b_resid  =bp->b_bcount;
942 			bp->b_flags |= B_ERROR;
943 		}
944 
945 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
946 		    (bp->b_flags & B_READ));
947 #if NRND > 0
948 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
949 #endif
950 
951 		biodone(bp);
952 	}
953 }
954 
955 static void
956 sdminphys(struct buf *bp)
957 {
958 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
959 	long xmax;
960 
961 	/*
962 	 * If the device is ancient, we want to make sure that
963 	 * the transfer fits into a 6-byte cdb.
964 	 *
965 	 * XXX Note that the SCSI-I spec says that 256-block transfers
966 	 * are allowed in a 6-byte read/write, and are specified
967 	 * by settng the "length" to 0.  However, we're conservative
968 	 * here, allowing only 255-block transfers in case an
969 	 * ancient device gets confused by length == 0.  A length of 0
970 	 * in a 10-byte read/write actually means 0 blocks.
971 	 */
972 	if ((sd->flags & SDF_ANCIENT) &&
973 	    ((sd->sc_periph->periph_flags &
974 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
975 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
976 
977 		if (bp->b_bcount > xmax)
978 			bp->b_bcount = xmax;
979 	}
980 
981 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
982 }
983 
984 static int
985 sdread(dev_t dev, struct uio *uio, int ioflag)
986 {
987 
988 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
989 }
990 
991 static int
992 sdwrite(dev_t dev, struct uio *uio, int ioflag)
993 {
994 
995 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
996 }
997 
998 /*
999  * Perform special action on behalf of the user
1000  * Knows about the internals of this device
1001  */
1002 static int
1003 sdioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct lwp *l)
1004 {
1005 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1006 	struct scsipi_periph *periph = sd->sc_periph;
1007 	int part = SDPART(dev);
1008 	int error = 0;
1009 #ifdef __HAVE_OLD_DISKLABEL
1010 	struct disklabel *newlabel = NULL;
1011 #endif
1012 
1013 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1014 
1015 	/*
1016 	 * If the device is not valid, some IOCTLs can still be
1017 	 * handled on the raw partition. Check this here.
1018 	 */
1019 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1020 		switch (cmd) {
1021 		case DIOCKLABEL:
1022 		case DIOCWLABEL:
1023 		case DIOCLOCK:
1024 		case DIOCEJECT:
1025 		case ODIOCEJECT:
1026 		case DIOCGCACHE:
1027 		case DIOCSCACHE:
1028 		case SCIOCIDENTIFY:
1029 		case OSCIOCIDENTIFY:
1030 		case SCIOCCOMMAND:
1031 		case SCIOCDEBUG:
1032 			if (part == RAW_PART)
1033 				break;
1034 		/* FALLTHROUGH */
1035 		default:
1036 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1037 				return (ENODEV);
1038 			else
1039 				return (EIO);
1040 		}
1041 	}
1042 
1043 	switch (cmd) {
1044 	case DIOCGDINFO:
1045 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1046 		return (0);
1047 
1048 #ifdef __HAVE_OLD_DISKLABEL
1049 	case ODIOCGDINFO:
1050 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1051 		if (newlabel == NULL)
1052 			return EIO;
1053 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1054 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1055 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1056 		else
1057 			error = ENOTTY;
1058 		free(newlabel, M_TEMP);
1059 		return error;
1060 #endif
1061 
1062 	case DIOCGPART:
1063 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1064 		((struct partinfo *)addr)->part =
1065 		    &sd->sc_dk.dk_label->d_partitions[part];
1066 		return (0);
1067 
1068 	case DIOCWDINFO:
1069 	case DIOCSDINFO:
1070 #ifdef __HAVE_OLD_DISKLABEL
1071 	case ODIOCWDINFO:
1072 	case ODIOCSDINFO:
1073 #endif
1074 	{
1075 		struct disklabel *lp;
1076 
1077 		if ((flag & FWRITE) == 0)
1078 			return (EBADF);
1079 
1080 #ifdef __HAVE_OLD_DISKLABEL
1081  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1082 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1083 			if (newlabel == NULL)
1084 				return EIO;
1085 			memset(newlabel, 0, sizeof newlabel);
1086 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1087 			lp = newlabel;
1088 		} else
1089 #endif
1090 		lp = (struct disklabel *)addr;
1091 
1092 		if ((error = lockmgr(&sd->sc_dk.dk_openlock,
1093 				     LK_EXCLUSIVE, NULL)) != 0)
1094 			goto bad;
1095 		sd->flags |= SDF_LABELLING;
1096 
1097 		error = setdisklabel(sd->sc_dk.dk_label,
1098 		    lp, /*sd->sc_dk.dk_openmask : */0,
1099 		    sd->sc_dk.dk_cpulabel);
1100 		if (error == 0) {
1101 			if (cmd == DIOCWDINFO
1102 #ifdef __HAVE_OLD_DISKLABEL
1103 			    || cmd == ODIOCWDINFO
1104 #endif
1105 			   )
1106 				error = writedisklabel(SDLABELDEV(dev),
1107 				    sdstrategy, sd->sc_dk.dk_label,
1108 				    sd->sc_dk.dk_cpulabel);
1109 		}
1110 
1111 		sd->flags &= ~SDF_LABELLING;
1112 		(void) lockmgr(&sd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1113 bad:
1114 #ifdef __HAVE_OLD_DISKLABEL
1115 		if (newlabel != NULL)
1116 			free(newlabel, M_TEMP);
1117 #endif
1118 		return (error);
1119 	}
1120 
1121 	case DIOCKLABEL:
1122 		if (*(int *)addr)
1123 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1124 		else
1125 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1126 		return (0);
1127 
1128 	case DIOCWLABEL:
1129 		if ((flag & FWRITE) == 0)
1130 			return (EBADF);
1131 		if (*(int *)addr)
1132 			sd->flags |= SDF_WLABEL;
1133 		else
1134 			sd->flags &= ~SDF_WLABEL;
1135 		return (0);
1136 
1137 	case DIOCLOCK:
1138 		if (periph->periph_flags & PERIPH_REMOVABLE)
1139 			return (scsipi_prevent(periph,
1140 			    (*(int *)addr) ?
1141 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1142 		else
1143 			return (ENOTTY);
1144 
1145 	case DIOCEJECT:
1146 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1147 			return (ENOTTY);
1148 		if (*(int *)addr == 0) {
1149 			/*
1150 			 * Don't force eject: check that we are the only
1151 			 * partition open. If so, unlock it.
1152 			 */
1153 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1154 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1155 			    sd->sc_dk.dk_openmask) {
1156 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1157 				    XS_CTL_IGNORE_NOT_READY);
1158 				if (error)
1159 					return (error);
1160 			} else {
1161 				return (EBUSY);
1162 			}
1163 		}
1164 		/* FALLTHROUGH */
1165 	case ODIOCEJECT:
1166 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1167 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1168 
1169 	case DIOCGDEFLABEL:
1170 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1171 		return (0);
1172 
1173 #ifdef __HAVE_OLD_DISKLABEL
1174 	case ODIOCGDEFLABEL:
1175 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1176 		if (newlabel == NULL)
1177 			return EIO;
1178 		sdgetdefaultlabel(sd, newlabel);
1179 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1180 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1181 		else
1182 			error = ENOTTY;
1183 		free(newlabel, M_TEMP);
1184 		return error;
1185 #endif
1186 
1187 	case DIOCGCACHE:
1188 		return (sd_getcache(sd, (int *) addr));
1189 
1190 	case DIOCSCACHE:
1191 		if ((flag & FWRITE) == 0)
1192 			return (EBADF);
1193 		return (sd_setcache(sd, *(int *) addr));
1194 
1195 	case DIOCCACHESYNC:
1196 		/*
1197 		 * XXX Do we really need to care about having a writable
1198 		 * file descriptor here?
1199 		 */
1200 		if ((flag & FWRITE) == 0)
1201 			return (EBADF);
1202 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1203 			error = sd_flush(sd, 0);
1204 			if (error)
1205 				sd->flags &= ~SDF_FLUSHING;
1206 			else
1207 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1208 		} else
1209 			error = 0;
1210 		return (error);
1211 
1212 	case DIOCAWEDGE:
1213 	    {
1214 	    	struct dkwedge_info *dkw = (void *) addr;
1215 
1216 		if ((flag & FWRITE) == 0)
1217 			return (EBADF);
1218 
1219 		/* If the ioctl happens here, the parent is us. */
1220 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1221 		return (dkwedge_add(dkw));
1222 	    }
1223 
1224 	case DIOCDWEDGE:
1225 	    {
1226 	    	struct dkwedge_info *dkw = (void *) addr;
1227 
1228 		if ((flag & FWRITE) == 0)
1229 			return (EBADF);
1230 
1231 		/* If the ioctl happens here, the parent is us. */
1232 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1233 		return (dkwedge_del(dkw));
1234 	    }
1235 
1236 	case DIOCLWEDGES:
1237 	    {
1238 	    	struct dkwedge_list *dkwl = (void *) addr;
1239 
1240 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1241 	    }
1242 
1243 	default:
1244 		if (part != RAW_PART)
1245 			return (ENOTTY);
1246 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1247 	}
1248 
1249 #ifdef DIAGNOSTIC
1250 	panic("sdioctl: impossible");
1251 #endif
1252 }
1253 
1254 static void
1255 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1256 {
1257 
1258 	memset(lp, 0, sizeof(struct disklabel));
1259 
1260 	lp->d_secsize = sd->params.blksize;
1261 	lp->d_ntracks = sd->params.heads;
1262 	lp->d_nsectors = sd->params.sectors;
1263 	lp->d_ncylinders = sd->params.cyls;
1264 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1265 
1266 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1267 	case SCSIPI_BUSTYPE_SCSI:
1268 		lp->d_type = DTYPE_SCSI;
1269 		break;
1270 	case SCSIPI_BUSTYPE_ATAPI:
1271 		lp->d_type = DTYPE_ATAPI;
1272 		break;
1273 	}
1274 	/*
1275 	 * XXX
1276 	 * We could probe the mode pages to figure out what kind of disc it is.
1277 	 * Is this worthwhile?
1278 	 */
1279 	strncpy(lp->d_typename, "mydisk", 16);
1280 	strncpy(lp->d_packname, "fictitious", 16);
1281 	lp->d_secperunit = sd->params.disksize;
1282 	lp->d_rpm = sd->params.rot_rate;
1283 	lp->d_interleave = 1;
1284 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1285 	    D_REMOVABLE : 0;
1286 
1287 	lp->d_partitions[RAW_PART].p_offset = 0;
1288 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1289 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1290 	lp->d_npartitions = RAW_PART + 1;
1291 
1292 	lp->d_magic = DISKMAGIC;
1293 	lp->d_magic2 = DISKMAGIC;
1294 	lp->d_checksum = dkcksum(lp);
1295 }
1296 
1297 
1298 /*
1299  * Load the label information on the named device
1300  */
1301 static int
1302 sdgetdisklabel(struct sd_softc *sd)
1303 {
1304 	struct disklabel *lp = sd->sc_dk.dk_label;
1305 	const char *errstring;
1306 
1307 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1308 
1309 	sdgetdefaultlabel(sd, lp);
1310 
1311 	if (lp->d_secpercyl == 0) {
1312 		lp->d_secpercyl = 100;
1313 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1314 	}
1315 
1316 	/*
1317 	 * Call the generic disklabel extraction routine
1318 	 */
1319 	errstring = readdisklabel(MAKESDDEV(0, device_unit(&sd->sc_dev),
1320 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1321 	if (errstring) {
1322 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1323 		return EIO;
1324 	}
1325 	return 0;
1326 }
1327 
1328 static void
1329 sd_shutdown(void *arg)
1330 {
1331 	struct sd_softc *sd = arg;
1332 
1333 	/*
1334 	 * If the disk cache needs to be flushed, and the disk supports
1335 	 * it, flush it.  We're cold at this point, so we poll for
1336 	 * completion.
1337 	 */
1338 	if ((sd->flags & SDF_DIRTY) != 0) {
1339 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1340 			printf("%s: cache synchronization failed\n",
1341 			    sd->sc_dev.dv_xname);
1342 			sd->flags &= ~SDF_FLUSHING;
1343 		} else
1344 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1345 	}
1346 }
1347 
1348 /*
1349  * Check Errors
1350  */
1351 static int
1352 sd_interpret_sense(struct scsipi_xfer *xs)
1353 {
1354 	struct scsipi_periph *periph = xs->xs_periph;
1355 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1356 	struct sd_softc *sd = (void *)periph->periph_dev;
1357 	int s, error, retval = EJUSTRETURN;
1358 
1359 	/*
1360 	 * If the periph is already recovering, just do the normal
1361 	 * error processing.
1362 	 */
1363 	if (periph->periph_flags & PERIPH_RECOVERING)
1364 		return (retval);
1365 
1366 	/*
1367 	 * Ignore errors from accessing illegal fields (e.g. trying to
1368 	 * lock the door of a digicam, which doesn't have a door that
1369 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1370 	 */
1371 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1372 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1373 	    sense->asc == 0x24 &&
1374 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1375 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1376 			scsipi_printaddr(periph);
1377 			printf("no door lock\n");
1378 		}
1379 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1380 		return (retval);
1381 	}
1382 
1383 
1384 
1385 	/*
1386 	 * If the device is not open yet, let the generic code handle it.
1387 	 */
1388 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1389 		return (retval);
1390 
1391 	/*
1392 	 * If it isn't a extended or extended/deferred error, let
1393 	 * the generic code handle it.
1394 	 */
1395 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1396 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1397 		return (retval);
1398 
1399 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1400 	    sense->asc == 0x4) {
1401 		if (sense->ascq == 0x01)	{
1402 			/*
1403 			 * Unit In The Process Of Becoming Ready.
1404 			 */
1405 			printf("%s: waiting for pack to spin up...\n",
1406 			    sd->sc_dev.dv_xname);
1407 			if (!callout_pending(&periph->periph_callout))
1408 				scsipi_periph_freeze(periph, 1);
1409 			callout_reset(&periph->periph_callout,
1410 			    5 * hz, scsipi_periph_timed_thaw, periph);
1411 			retval = ERESTART;
1412 		} else if (sense->ascq == 0x02) {
1413 			printf("%s: pack is stopped, restarting...\n",
1414 			    sd->sc_dev.dv_xname);
1415 			s = splbio();
1416 			periph->periph_flags |= PERIPH_RECOVERING;
1417 			splx(s);
1418 			error = scsipi_start(periph, SSS_START,
1419 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1420 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1421 			if (error) {
1422 				printf("%s: unable to restart pack\n",
1423 				    sd->sc_dev.dv_xname);
1424 				retval = error;
1425 			} else
1426 				retval = ERESTART;
1427 			s = splbio();
1428 			periph->periph_flags &= ~PERIPH_RECOVERING;
1429 			splx(s);
1430 		}
1431 	}
1432 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1433 	    sense->asc == 0x31 &&
1434 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1435 		/* Medium Format Corrupted */
1436 		retval = EFTYPE;
1437 	}
1438 	return (retval);
1439 }
1440 
1441 
1442 static int
1443 sdsize(dev_t dev)
1444 {
1445 	struct sd_softc *sd;
1446 	int part, unit, omask;
1447 	int size;
1448 
1449 	unit = SDUNIT(dev);
1450 	if (unit >= sd_cd.cd_ndevs)
1451 		return (-1);
1452 	sd = sd_cd.cd_devs[unit];
1453 	if (sd == NULL)
1454 		return (-1);
1455 
1456 	if (!device_is_active(&sd->sc_dev))
1457 		return (-1);
1458 
1459 	part = SDPART(dev);
1460 	omask = sd->sc_dk.dk_openmask & (1 << part);
1461 
1462 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1463 		return (-1);
1464 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1465 		size = -1;
1466 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1467 		size = -1;
1468 	else
1469 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1470 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1471 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1472 		return (-1);
1473 	return (size);
1474 }
1475 
1476 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1477 static struct scsipi_xfer sx;
1478 static int sddoingadump;
1479 
1480 /*
1481  * dump all of physical memory into the partition specified, starting
1482  * at offset 'dumplo' into the partition.
1483  */
1484 static int
1485 sddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1486 {
1487 	struct sd_softc *sd;	/* disk unit to do the I/O */
1488 	struct disklabel *lp;	/* disk's disklabel */
1489 	int	unit, part;
1490 	int	sectorsize;	/* size of a disk sector */
1491 	int	nsects;		/* number of sectors in partition */
1492 	int	sectoff;	/* sector offset of partition */
1493 	int	totwrt;		/* total number of sectors left to write */
1494 	int	nwrt;		/* current number of sectors to write */
1495 	struct scsipi_rw_10 cmd;	/* write command */
1496 	struct scsipi_xfer *xs;	/* ... convenience */
1497 	struct scsipi_periph *periph;
1498 	struct scsipi_channel *chan;
1499 
1500 	/* Check if recursive dump; if so, punt. */
1501 	if (sddoingadump)
1502 		return (EFAULT);
1503 
1504 	/* Mark as active early. */
1505 	sddoingadump = 1;
1506 
1507 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1508 	part = SDPART(dev);
1509 
1510 	/* Check for acceptable drive number. */
1511 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1512 		return (ENXIO);
1513 
1514 	if (!device_is_active(&sd->sc_dev))
1515 		return (ENODEV);
1516 
1517 	periph = sd->sc_periph;
1518 	chan = periph->periph_channel;
1519 
1520 	/* Make sure it was initialized. */
1521 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1522 		return (ENXIO);
1523 
1524 	/* Convert to disk sectors.  Request must be a multiple of size. */
1525 	lp = sd->sc_dk.dk_label;
1526 	sectorsize = lp->d_secsize;
1527 	if ((size % sectorsize) != 0)
1528 		return (EFAULT);
1529 	totwrt = size / sectorsize;
1530 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1531 
1532 	nsects = lp->d_partitions[part].p_size;
1533 	sectoff = lp->d_partitions[part].p_offset;
1534 
1535 	/* Check transfer bounds against partition size. */
1536 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1537 		return (EINVAL);
1538 
1539 	/* Offset block number to start of partition. */
1540 	blkno += sectoff;
1541 
1542 	xs = &sx;
1543 
1544 	while (totwrt > 0) {
1545 		nwrt = totwrt;		/* XXX */
1546 #ifndef	SD_DUMP_NOT_TRUSTED
1547 		/*
1548 		 *  Fill out the scsi command
1549 		 */
1550 		memset(&cmd, 0, sizeof(cmd));
1551 		cmd.opcode = WRITE_10;
1552 		_lto4b(blkno, cmd.addr);
1553 		_lto2b(nwrt, cmd.length);
1554 		/*
1555 		 * Fill out the scsipi_xfer structure
1556 		 *    Note: we cannot sleep as we may be an interrupt
1557 		 * don't use scsipi_command() as it may want to wait
1558 		 * for an xs.
1559 		 */
1560 		memset(xs, 0, sizeof(sx));
1561 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1562 		    XS_CTL_DATA_OUT;
1563 		xs->xs_status = 0;
1564 		xs->xs_periph = periph;
1565 		xs->xs_retries = SDRETRIES;
1566 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1567 		xs->cmd = (struct scsipi_generic *)&cmd;
1568 		xs->cmdlen = sizeof(cmd);
1569 		xs->resid = nwrt * sectorsize;
1570 		xs->error = XS_NOERROR;
1571 		xs->bp = 0;
1572 		xs->data = va;
1573 		xs->datalen = nwrt * sectorsize;
1574 
1575 		/*
1576 		 * Pass all this info to the scsi driver.
1577 		 */
1578 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1579 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1580 		    xs->error != XS_NOERROR)
1581 			return (EIO);
1582 #else	/* SD_DUMP_NOT_TRUSTED */
1583 		/* Let's just talk about this first... */
1584 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1585 		delay(500 * 1000);	/* half a second */
1586 #endif	/* SD_DUMP_NOT_TRUSTED */
1587 
1588 		/* update block count */
1589 		totwrt -= nwrt;
1590 		blkno += nwrt;
1591 		va += sectorsize * nwrt;
1592 	}
1593 	sddoingadump = 0;
1594 	return (0);
1595 }
1596 
1597 static int
1598 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1599     int page, int flags, int *big)
1600 {
1601 
1602 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1603 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1604 		*big = 1;
1605 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1606 		    size + sizeof(struct scsi_mode_parameter_header_10),
1607 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1608 	} else {
1609 		*big = 0;
1610 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1611 		    size + sizeof(struct scsi_mode_parameter_header_6),
1612 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1613 	}
1614 }
1615 
1616 static int
1617 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1618     int flags, int big)
1619 {
1620 
1621 	if (big) {
1622 		struct scsi_mode_parameter_header_10 *header = sense;
1623 
1624 		_lto2b(0, header->data_length);
1625 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1626 		    size + sizeof(struct scsi_mode_parameter_header_10),
1627 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1628 	} else {
1629 		struct scsi_mode_parameter_header_6 *header = sense;
1630 
1631 		header->data_length = 0;
1632 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1633 		    size + sizeof(struct scsi_mode_parameter_header_6),
1634 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1635 	}
1636 }
1637 
1638 static int
1639 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1640 {
1641 	struct {
1642 		struct scsi_mode_parameter_header_6 header;
1643 		/* no block descriptor */
1644 		u_int8_t pg_code; /* page code (should be 6) */
1645 		u_int8_t pg_length; /* page length (should be 11) */
1646 		u_int8_t wcd; /* bit0: cache disable */
1647 		u_int8_t lbs[2]; /* logical block size */
1648 		u_int8_t size[5]; /* number of log. blocks */
1649 		u_int8_t pp; /* power/performance */
1650 		u_int8_t flags;
1651 		u_int8_t resvd;
1652 	} scsipi_sense;
1653 	u_int64_t sectors;
1654 	int error, blksize;
1655 
1656 	/*
1657 	 * scsipi_size (ie "read capacity") and mode sense page 6
1658 	 * give the same information. Do both for now, and check
1659 	 * for consistency.
1660 	 * XXX probably differs for removable media
1661 	 */
1662 	dp->blksize = 512;
1663 	if ((sectors = scsipi_size(sd->sc_periph, &blksize, 512, flags)) == 0)
1664 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1665 
1666 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1667 	    &scsipi_sense.header, sizeof(scsipi_sense),
1668 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1669 
1670 	if (error != 0)
1671 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1672 
1673 	dp->blksize = _2btol(scsipi_sense.lbs);
1674 	if (dp->blksize == 0)
1675 		dp->blksize = blksize;
1676 
1677 	/*
1678 	 * Create a pseudo-geometry.
1679 	 */
1680 	dp->heads = 64;
1681 	dp->sectors = 32;
1682 	dp->cyls = sectors / (dp->heads * dp->sectors);
1683 	dp->disksize = _5btol(scsipi_sense.size);
1684 	if (dp->disksize <= UINT32_MAX && dp->disksize != sectors) {
1685 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1686 		       (unsigned long long)dp->disksize,
1687 		       (unsigned long long)sectors);
1688 		dp->disksize = sectors;
1689 	}
1690 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1691 
1692 	return (SDGP_RESULT_OK);
1693 }
1694 
1695 /*
1696  * Get the scsi driver to send a full inquiry to the * device and use the
1697  * results to fill out the disk parameter structure.
1698  */
1699 static int
1700 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1701 {
1702 	u_int64_t sectors;
1703 	int error, blksize;
1704 #if 0
1705 	int i;
1706 	u_int8_t *p;
1707 #endif
1708 
1709 	dp->disksize = sectors = scsipi_size(sd->sc_periph, &blksize, 512,
1710 	    flags);
1711 	if (sectors == 0) {
1712 		struct scsipi_read_format_capacities cmd;
1713 		struct {
1714 			struct scsipi_capacity_list_header header;
1715 			struct scsipi_capacity_descriptor desc;
1716 		} __attribute__((packed)) data;
1717 
1718 		memset(&cmd, 0, sizeof(cmd));
1719 		memset(&data, 0, sizeof(data));
1720 		cmd.opcode = READ_FORMAT_CAPACITIES;
1721 		_lto2b(sizeof(data), cmd.length);
1722 
1723 		error = scsipi_command(sd->sc_periph,
1724 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1725 		    SDRETRIES, 20000, NULL,
1726 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1727 		if (error == EFTYPE) {
1728 			/* Medium Format Corrupted, handle as not formatted */
1729 			return (SDGP_RESULT_UNFORMATTED);
1730 		}
1731 		if (error || data.header.length == 0)
1732 			return (SDGP_RESULT_OFFLINE);
1733 
1734 #if 0
1735 printf("rfc: length=%d\n", data.header.length);
1736 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");
1737 #endif
1738 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1739 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1740 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1741 			break;
1742 
1743 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1744 			return (SDGP_RESULT_UNFORMATTED);
1745 
1746 		case SCSIPI_CAP_DESC_CODE_NONE:
1747 			return (SDGP_RESULT_OFFLINE);
1748 		}
1749 
1750 		dp->disksize = sectors = _4btol(data.desc.nblks);
1751 		if (sectors == 0)
1752 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1753 
1754 		dp->blksize = _3btol(data.desc.blklen);
1755 		if (dp->blksize == 0)
1756 			dp->blksize = 512;
1757 	} else {
1758 		struct sd_mode_sense_data scsipi_sense;
1759 		int big, bsize;
1760 		struct scsi_general_block_descriptor *bdesc;
1761 
1762 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1763 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1764 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1765 		dp->blksize = blksize;
1766 		if (!error) {
1767 			if (big) {
1768 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1769 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1770 			} else {
1771 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1772 				bsize = scsipi_sense.header.small.blk_desc_len;
1773 			}
1774 
1775 #if 0
1776 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1777 printf("page 0 bsize=%d\n", bsize);
1778 printf("page 0 ok\n");
1779 #endif
1780 
1781 			if (bsize >= 8) {
1782 				dp->blksize = _3btol(bdesc->blklen);
1783 				if (dp->blksize == 0)
1784 					dp->blksize = blksize;
1785 			}
1786 		}
1787 	}
1788 
1789 	dp->disksize512 = (sectors * dp->blksize) / DEV_BSIZE;
1790 	return (0);
1791 }
1792 
1793 static int
1794 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1795 {
1796 	struct sd_mode_sense_data scsipi_sense;
1797 	int error;
1798 	int big, poffset, byte2;
1799 	union scsi_disk_pages *pages;
1800 #if 0
1801 	int i;
1802 	u_int8_t *p;
1803 #endif
1804 
1805 	byte2 = SMS_DBD;
1806 again:
1807 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1808 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1809 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1810 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1811 	    flags | XS_CTL_SILENT, &big);
1812 	if (error) {
1813 		if (byte2 == SMS_DBD) {
1814 			/* No result; try once more with DBD off */
1815 			byte2 = 0;
1816 			goto again;
1817 		}
1818 		return (error);
1819 	}
1820 
1821 	if (big) {
1822 		poffset = sizeof scsipi_sense.header.big;
1823 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1824 	} else {
1825 		poffset = sizeof scsipi_sense.header.small;
1826 		poffset += scsipi_sense.header.small.blk_desc_len;
1827 	}
1828 
1829 	pages = (void *)((u_long)&scsipi_sense + poffset);
1830 #if 0
1831 printf("page 4 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1832 printf("page 4 pg_code=%d sense=%p/%p\n", pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1833 #endif
1834 
1835 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1836 		return (ERESTART);
1837 
1838 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1839 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1840 	    _3btol(pages->rigid_geometry.ncyl),
1841 	    pages->rigid_geometry.nheads,
1842 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1843 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1844 	    _2btol(pages->rigid_geometry.land_zone)));
1845 
1846 	/*
1847 	 * KLUDGE!! (for zone recorded disks)
1848 	 * give a number of sectors so that sec * trks * cyls
1849 	 * is <= disk_size
1850 	 * can lead to wasted space! THINK ABOUT THIS !
1851 	 */
1852 	dp->heads = pages->rigid_geometry.nheads;
1853 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1854 	if (dp->heads == 0 || dp->cyls == 0)
1855 		return (ERESTART);
1856 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1857 
1858 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1859 	if (dp->rot_rate == 0)
1860 		dp->rot_rate = 3600;
1861 
1862 #if 0
1863 printf("page 4 ok\n");
1864 #endif
1865 	return (0);
1866 }
1867 
1868 static int
1869 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1870 {
1871 	struct sd_mode_sense_data scsipi_sense;
1872 	int error;
1873 	int big, poffset, byte2;
1874 	union scsi_disk_pages *pages;
1875 #if 0
1876 	int i;
1877 	u_int8_t *p;
1878 #endif
1879 
1880 	byte2 = SMS_DBD;
1881 again:
1882 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1883 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1884 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1885 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1886 	    flags | XS_CTL_SILENT, &big);
1887 	if (error) {
1888 		if (byte2 == SMS_DBD) {
1889 			/* No result; try once more with DBD off */
1890 			byte2 = 0;
1891 			goto again;
1892 		}
1893 		return (error);
1894 	}
1895 
1896 	if (big) {
1897 		poffset = sizeof scsipi_sense.header.big;
1898 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1899 	} else {
1900 		poffset = sizeof scsipi_sense.header.small;
1901 		poffset += scsipi_sense.header.small.blk_desc_len;
1902 	}
1903 
1904 	pages = (void *)((u_long)&scsipi_sense + poffset);
1905 #if 0
1906 printf("page 5 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1907 printf("page 5 pg_code=%d sense=%p/%p\n", pages->flex_geometry.pg_code, &scsipi_sense, pages);
1908 #endif
1909 
1910 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
1911 		return (ERESTART);
1912 
1913 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1914 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
1915 	    _3btol(pages->flex_geometry.ncyl),
1916 	    pages->flex_geometry.nheads,
1917 	    pages->flex_geometry.ph_sec_tr,
1918 	    _2btol(pages->flex_geometry.bytes_s)));
1919 
1920 	dp->heads = pages->flex_geometry.nheads;
1921 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
1922 	dp->sectors = pages->flex_geometry.ph_sec_tr;
1923 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
1924 		return (ERESTART);
1925 
1926 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1927 	if (dp->rot_rate == 0)
1928 		dp->rot_rate = 3600;
1929 
1930 #if 0
1931 printf("page 5 ok\n");
1932 #endif
1933 	return (0);
1934 }
1935 
1936 static int
1937 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1938 {
1939 	int error;
1940 
1941 	/*
1942 	 * If offline, the SDEV_MEDIA_LOADED flag will be
1943 	 * cleared by the caller if necessary.
1944 	 */
1945 	if (sd->type == T_SIMPLE_DIRECT) {
1946 		error = sd_get_simplifiedparms(sd, dp, flags);
1947 		if (!error)
1948 			disk_blocksize(&sd->sc_dk, dp->blksize);
1949 		return (error);
1950 	}
1951 
1952 	error = sd_get_capacity(sd, dp, flags);
1953 	if (error)
1954 		return (error);
1955 
1956 	disk_blocksize(&sd->sc_dk, dp->blksize);
1957 
1958 	if (sd->type == T_OPTICAL)
1959 		goto page0;
1960 
1961 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
1962 		if (!sd_get_parms_page5(sd, dp, flags) ||
1963 		    !sd_get_parms_page4(sd, dp, flags))
1964 			return (SDGP_RESULT_OK);
1965 	} else {
1966 		if (!sd_get_parms_page4(sd, dp, flags) ||
1967 		    !sd_get_parms_page5(sd, dp, flags))
1968 			return (SDGP_RESULT_OK);
1969 	}
1970 
1971 page0:
1972 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
1973 	/* Try calling driver's method for figuring out geometry. */
1974 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
1975 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
1976 		(sd->sc_periph, dp, dp->disksize)) {
1977 		/*
1978 		 * Use adaptec standard fictitious geometry
1979 		 * this depends on which controller (e.g. 1542C is
1980 		 * different. but we have to put SOMETHING here..)
1981 		 */
1982 		dp->heads = 64;
1983 		dp->sectors = 32;
1984 		dp->cyls = dp->disksize / (64 * 32);
1985 	}
1986 	dp->rot_rate = 3600;
1987 	return (SDGP_RESULT_OK);
1988 }
1989 
1990 static int
1991 sd_flush(struct sd_softc *sd, int flags)
1992 {
1993 	struct scsipi_periph *periph = sd->sc_periph;
1994 	struct scsi_synchronize_cache_10 cmd;
1995 
1996 	/*
1997 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
1998 	 * We issue with address 0 length 0, which should be
1999 	 * interpreted by the device as "all remaining blocks
2000 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2001 	 * in the event that the command is not supported by
2002 	 * the device, and poll for completion so that we know
2003 	 * that the cache has actually been flushed.
2004 	 *
2005 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2006 	 * command, as indicated by our quirks flags.
2007 	 *
2008 	 * XXX What about older devices?
2009 	 */
2010 	if (periph->periph_version < 2 ||
2011 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2012 		return (0);
2013 
2014 	sd->flags |= SDF_FLUSHING;
2015 	memset(&cmd, 0, sizeof(cmd));
2016 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2017 
2018 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2019 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2020 }
2021 
2022 static int
2023 sd_getcache(struct sd_softc *sd, int *bitsp)
2024 {
2025 	struct scsipi_periph *periph = sd->sc_periph;
2026 	struct sd_mode_sense_data scsipi_sense;
2027 	int error, bits = 0;
2028 	int big;
2029 	union scsi_disk_pages *pages;
2030 
2031 	if (periph->periph_version < 2)
2032 		return (EOPNOTSUPP);
2033 
2034 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2035 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2036 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2037 	if (error)
2038 		return (error);
2039 
2040 	if (big)
2041 		pages = (void *)(&scsipi_sense.header.big + 1);
2042 	else
2043 		pages = (void *)(&scsipi_sense.header.small + 1);
2044 
2045 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2046 		bits |= DKCACHE_READ;
2047 	if (pages->caching_params.flags & CACHING_WCE)
2048 		bits |= DKCACHE_WRITE;
2049 	if (pages->caching_params.pg_code & PGCODE_PS)
2050 		bits |= DKCACHE_SAVE;
2051 
2052 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2053 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2054 	    sizeof(scsipi_sense.pages.caching_params),
2055 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2056 	if (error == 0) {
2057 		if (big)
2058 			pages = (void *)(&scsipi_sense.header.big + 1);
2059 		else
2060 			pages = (void *)(&scsipi_sense.header.small + 1);
2061 
2062 		if (pages->caching_params.flags & CACHING_RCD)
2063 			bits |= DKCACHE_RCHANGE;
2064 		if (pages->caching_params.flags & CACHING_WCE)
2065 			bits |= DKCACHE_WCHANGE;
2066 	}
2067 
2068 	*bitsp = bits;
2069 
2070 	return (0);
2071 }
2072 
2073 static int
2074 sd_setcache(struct sd_softc *sd, int bits)
2075 {
2076 	struct scsipi_periph *periph = sd->sc_periph;
2077 	struct sd_mode_sense_data scsipi_sense;
2078 	int error;
2079 	uint8_t oflags, byte2 = 0;
2080 	int big;
2081 	union scsi_disk_pages *pages;
2082 
2083 	if (periph->periph_version < 2)
2084 		return (EOPNOTSUPP);
2085 
2086 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2087 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2088 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2089 	if (error)
2090 		return (error);
2091 
2092 	if (big)
2093 		pages = (void *)(&scsipi_sense.header.big + 1);
2094 	else
2095 		pages = (void *)(&scsipi_sense.header.small + 1);
2096 
2097 	oflags = pages->caching_params.flags;
2098 
2099 	if (bits & DKCACHE_READ)
2100 		pages->caching_params.flags &= ~CACHING_RCD;
2101 	else
2102 		pages->caching_params.flags |= CACHING_RCD;
2103 
2104 	if (bits & DKCACHE_WRITE)
2105 		pages->caching_params.flags |= CACHING_WCE;
2106 	else
2107 		pages->caching_params.flags &= ~CACHING_WCE;
2108 
2109 	if (oflags == pages->caching_params.flags)
2110 		return (0);
2111 
2112 	pages->caching_params.pg_code &= PGCODE_MASK;
2113 
2114 	if (bits & DKCACHE_SAVE)
2115 		byte2 |= SMS_SP;
2116 
2117 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2118 	    sizeof(struct scsi_mode_page_header) +
2119 	    pages->caching_params.pg_length, 0, big));
2120 }
2121