xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: sd.c,v 1.278 2009/04/07 18:35:17 dyoung 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Originally written by Julian Elischer (julian@dialix.oz.au)
34  * for TRW Financial Systems for use under the MACH(2.5) operating system.
35  *
36  * TRW Financial Systems, in accordance with their agreement with Carnegie
37  * Mellon University, makes this software available to CMU to distribute
38  * or use in any manner that they see fit as long as this message is kept with
39  * the software. For this reason TFS also grants any other persons or
40  * organisations permission to use or modify this software.
41  *
42  * TFS supplies this software to be publicly redistributed
43  * on the understanding that TFS is not responsible for the correct
44  * functioning of this software in any circumstances.
45  *
46  * Ported to run under 386BSD by Julian Elischer (julian@dialix.oz.au) Sept 1992
47  */
48 
49 #include <sys/cdefs.h>
50 __KERNEL_RCSID(0, "$NetBSD: sd.c,v 1.278 2009/04/07 18:35:17 dyoung Exp $");
51 
52 #include "opt_scsi.h"
53 #include "rnd.h"
54 
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/kernel.h>
58 #include <sys/file.h>
59 #include <sys/stat.h>
60 #include <sys/ioctl.h>
61 #include <sys/scsiio.h>
62 #include <sys/buf.h>
63 #include <sys/bufq.h>
64 #include <sys/uio.h>
65 #include <sys/malloc.h>
66 #include <sys/errno.h>
67 #include <sys/device.h>
68 #include <sys/disklabel.h>
69 #include <sys/disk.h>
70 #include <sys/proc.h>
71 #include <sys/conf.h>
72 #include <sys/vnode.h>
73 #if NRND > 0
74 #include <sys/rnd.h>
75 #endif
76 
77 #include <dev/scsipi/scsi_spc.h>
78 #include <dev/scsipi/scsipi_all.h>
79 #include <dev/scsipi/scsi_all.h>
80 #include <dev/scsipi/scsipi_disk.h>
81 #include <dev/scsipi/scsi_disk.h>
82 #include <dev/scsipi/scsiconf.h>
83 #include <dev/scsipi/scsipi_base.h>
84 #include <dev/scsipi/sdvar.h>
85 
86 #include <prop/proplib.h>
87 
88 #define	SDUNIT(dev)			DISKUNIT(dev)
89 #define	SDPART(dev)			DISKPART(dev)
90 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
91 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
92 
93 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
94 
95 #define	SD_DEFAULT_BLKSIZE	512
96 
97 static void	sdminphys(struct buf *);
98 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
99 static int	sdgetdisklabel(struct sd_softc *);
100 static void	sdstart(struct scsipi_periph *);
101 static void	sdrestart(void *);
102 static void	sddone(struct scsipi_xfer *, int);
103 static bool	sd_suspend(device_t PMF_FN_PROTO);
104 static void	sd_shutdown(void *);
105 static int	sd_interpret_sense(struct scsipi_xfer *);
106 
107 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
108 		    int, int *);
109 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
110 		    int);
111 static int	sd_validate_blksize(struct scsipi_periph *, int);
112 static u_int64_t sd_read_capacity(struct scsipi_periph *, int *, int flags);
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 static void	sd_set_properties(struct sd_softc *);
131 
132 CFATTACH_DECL3_NEW(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
133     sdactivate, NULL, NULL, DVF_DETACH_SHUTDOWN);
134 
135 extern struct cfdriver sd_cd;
136 
137 static const struct scsipi_inquiry_pattern sd_patterns[] = {
138 	{T_DIRECT, T_FIXED,
139 	 "",         "",                 ""},
140 	{T_DIRECT, T_REMOV,
141 	 "",         "",                 ""},
142 	{T_OPTICAL, T_FIXED,
143 	 "",         "",                 ""},
144 	{T_OPTICAL, T_REMOV,
145 	 "",         "",                 ""},
146 	{T_SIMPLE_DIRECT, T_FIXED,
147 	 "",         "",                 ""},
148 	{T_SIMPLE_DIRECT, T_REMOV,
149 	 "",         "",                 ""},
150 };
151 
152 static dev_type_open(sdopen);
153 static dev_type_close(sdclose);
154 static dev_type_read(sdread);
155 static dev_type_write(sdwrite);
156 static dev_type_ioctl(sdioctl);
157 static dev_type_strategy(sdstrategy);
158 static dev_type_dump(sddump);
159 static dev_type_size(sdsize);
160 
161 const struct bdevsw sd_bdevsw = {
162 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
163 };
164 
165 const struct cdevsw sd_cdevsw = {
166 	sdopen, sdclose, sdread, sdwrite, sdioctl,
167 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
168 };
169 
170 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
171 
172 static const struct scsipi_periphsw sd_switch = {
173 	sd_interpret_sense,	/* check our error handler first */
174 	sdstart,		/* have a queue, served by this */
175 	NULL,			/* have no async handler */
176 	sddone,			/* deal with stats at interrupt time */
177 };
178 
179 struct sd_mode_sense_data {
180 	/*
181 	 * XXX
182 	 * We are not going to parse this as-is -- it just has to be large
183 	 * enough.
184 	 */
185 	union {
186 		struct scsi_mode_parameter_header_6 small;
187 		struct scsi_mode_parameter_header_10 big;
188 	} header;
189 	struct scsi_general_block_descriptor blk_desc;
190 	union scsi_disk_pages pages;
191 };
192 
193 /*
194  * The routine called by the low level scsi routine when it discovers
195  * A device suitable for this driver
196  */
197 static int
198 sdmatch(struct device *parent, struct cfdata *match,
199     void *aux)
200 {
201 	struct scsipibus_attach_args *sa = aux;
202 	int priority;
203 
204 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
205 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
206 	    sizeof(sd_patterns[0]), &priority);
207 
208 	return (priority);
209 }
210 
211 /*
212  * Attach routine common to atapi & scsi.
213  */
214 static void
215 sdattach(struct device *parent, struct device *self, void *aux)
216 {
217 	struct sd_softc *sd = device_private(self);
218 	struct scsipibus_attach_args *sa = aux;
219 	struct scsipi_periph *periph = sa->sa_periph;
220 	int error, result;
221 	struct disk_parms *dp = &sd->params;
222 	char pbuf[9];
223 
224 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
225 
226 	sd->sc_dev = self;
227 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
228 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
229 	if (sd->type == T_SIMPLE_DIRECT)
230 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
231 
232 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
233 	    periph->periph_version == 0)
234 		sd->flags |= SDF_ANCIENT;
235 
236 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
237 
238 	callout_init(&sd->sc_callout, 0);
239 
240 	/*
241 	 * Store information needed to contact our base driver
242 	 */
243 	sd->sc_periph = periph;
244 
245 	periph->periph_dev = sd->sc_dev;
246 	periph->periph_switch = &sd_switch;
247 
248         /*
249          * Increase our openings to the maximum-per-periph
250          * supported by the adapter.  This will either be
251          * clamped down or grown by the adapter if necessary.
252          */
253 	periph->periph_openings =
254 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
255 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
256 
257 	/*
258 	 * Initialize and attach the disk structure.
259 	 */
260 	disk_init(&sd->sc_dk, device_xname(sd->sc_dev), &sddkdriver);
261 	disk_attach(&sd->sc_dk);
262 
263 	/*
264 	 * Use the subdriver to request information regarding the drive.
265 	 */
266 	aprint_naive("\n");
267 	aprint_normal("\n");
268 
269 	error = scsipi_test_unit_ready(periph,
270 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
271 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
272 
273 	if (error)
274 		result = SDGP_RESULT_OFFLINE;
275 	else
276 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
277 	aprint_normal_dev(sd->sc_dev, "");
278 	switch (result) {
279 	case SDGP_RESULT_OK:
280 		format_bytes(pbuf, sizeof(pbuf),
281 		    (u_int64_t)dp->disksize * dp->blksize);
282 	        aprint_normal(
283 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
284 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
285 		    (unsigned long long)dp->disksize);
286 		break;
287 
288 	case SDGP_RESULT_OFFLINE:
289 		aprint_normal("drive offline");
290 		break;
291 
292 	case SDGP_RESULT_UNFORMATTED:
293 		aprint_normal("unformatted media");
294 		break;
295 
296 #ifdef DIAGNOSTIC
297 	default:
298 		panic("sdattach: unknown result from get_parms");
299 		break;
300 #endif
301 	}
302 	aprint_normal("\n");
303 
304 	/*
305 	 * Establish a shutdown hook so that we can ensure that
306 	 * our data has actually made it onto the platter at
307 	 * shutdown time.  Note that this relies on the fact
308 	 * that the shutdown hook code puts us at the head of
309 	 * the list (thus guaranteeing that our hook runs before
310 	 * our ancestors').
311 	 */
312 	if ((sd->sc_sdhook =
313 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
314 		aprint_error_dev(sd->sc_dev,
315 			"WARNING: unable to establish shutdown hook\n");
316 
317 	if (!pmf_device_register(self, sd_suspend, NULL))
318 		aprint_error_dev(self, "couldn't establish power handler\n");
319 
320 #if NRND > 0
321 	/*
322 	 * attach the device into the random source list
323 	 */
324 	rnd_attach_source(&sd->rnd_source, device_xname(sd->sc_dev),
325 			  RND_TYPE_DISK, 0);
326 #endif
327 
328 	/* Discover wedges on this disk. */
329 	dkwedge_discover(&sd->sc_dk);
330 
331 	sd_set_properties(sd);
332 }
333 
334 static int
335 sdactivate(struct device *self, enum devact act)
336 {
337 	int rv = 0;
338 
339 	switch (act) {
340 	case DVACT_ACTIVATE:
341 		rv = EOPNOTSUPP;
342 		break;
343 
344 	case DVACT_DEACTIVATE:
345 		/*
346 		 * Nothing to do; we key off the device's DVF_ACTIVE.
347 		 */
348 		break;
349 	}
350 	return (rv);
351 }
352 
353 static int
354 sddetach(struct device *self, int flags)
355 {
356 	struct sd_softc *sd = device_private(self);
357 	int s, bmaj, cmaj, i, mn;
358 
359 	/* locate the major number */
360 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
361 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
362 
363 	/* Nuke the vnodes for any open instances */
364 	for (i = 0; i < MAXPARTITIONS; i++) {
365 		mn = SDMINOR(device_unit(self), i);
366 		vdevgone(bmaj, mn, mn, VBLK);
367 		vdevgone(cmaj, mn, mn, VCHR);
368 	}
369 
370 	/* kill any pending restart */
371 	callout_stop(&sd->sc_callout);
372 
373 	/* Delete all of our wedges. */
374 	dkwedge_delall(&sd->sc_dk);
375 
376 	s = splbio();
377 
378 	/* Kill off any queued buffers. */
379 	bufq_drain(sd->buf_queue);
380 
381 	bufq_free(sd->buf_queue);
382 
383 	/* Kill off any pending commands. */
384 	scsipi_kill_pending(sd->sc_periph);
385 
386 	splx(s);
387 
388 	/* Detach from the disk list. */
389 	disk_detach(&sd->sc_dk);
390 	disk_destroy(&sd->sc_dk);
391 
392 	callout_destroy(&sd->sc_callout);
393 
394 	pmf_device_deregister(self);
395 	shutdownhook_disestablish(sd->sc_sdhook);
396 
397 #if NRND > 0
398 	/* Unhook the entropy source. */
399 	rnd_detach_source(&sd->rnd_source);
400 #endif
401 
402 	return (0);
403 }
404 
405 /*
406  * open the device. Make sure the partition info is a up-to-date as can be.
407  */
408 static int
409 sdopen(dev_t dev, int flag, int fmt, struct lwp *l)
410 {
411 	struct sd_softc *sd;
412 	struct scsipi_periph *periph;
413 	struct scsipi_adapter *adapt;
414 	int unit, part;
415 	int error;
416 
417 	unit = SDUNIT(dev);
418 	sd = device_lookup_private(&sd_cd, unit);
419 	if (sd == NULL)
420 		return (ENXIO);
421 
422 	if (!device_is_active(sd->sc_dev))
423 		return (ENODEV);
424 
425 	part = SDPART(dev);
426 
427 	mutex_enter(&sd->sc_dk.dk_openlock);
428 
429 	/*
430 	 * If there are wedges, and this is not RAW_PART, then we
431 	 * need to fail.
432 	 */
433 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
434 		error = EBUSY;
435 		goto bad1;
436 	}
437 
438 	periph = sd->sc_periph;
439 	adapt = periph->periph_channel->chan_adapter;
440 
441 	SC_DEBUG(periph, SCSIPI_DB1,
442 	    ("sdopen: dev=0x%"PRIx64" (unit %d (of %d), partition %d)\n", dev, unit,
443 	    sd_cd.cd_ndevs, part));
444 
445 	/*
446 	 * If this is the first open of this device, add a reference
447 	 * to the adapter.
448 	 */
449 	if (sd->sc_dk.dk_openmask == 0 &&
450 	    (error = scsipi_adapter_addref(adapt)) != 0)
451 		goto bad1;
452 
453 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
454 		/*
455 		 * If any partition is open, but the disk has been invalidated,
456 		 * disallow further opens of non-raw partition
457 		 */
458 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
459 		    (part != RAW_PART || fmt != S_IFCHR)) {
460 			error = EIO;
461 			goto bad2;
462 		}
463 	} else {
464 		int silent;
465 
466 		if (part == RAW_PART && fmt == S_IFCHR)
467 			silent = XS_CTL_SILENT;
468 		else
469 			silent = 0;
470 
471 		/* Check that it is still responding and ok. */
472 		error = scsipi_test_unit_ready(periph,
473 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
474 		    silent);
475 
476 		/*
477 		 * Start the pack spinning if necessary. Always allow the
478 		 * raw parition to be opened, for raw IOCTLs. Data transfers
479 		 * will check for SDEV_MEDIA_LOADED.
480 		 */
481 		if (error == EIO) {
482 			int error2;
483 
484 			error2 = scsipi_start(periph, SSS_START, silent);
485 			switch (error2) {
486 			case 0:
487 				error = 0;
488 				break;
489 			case EIO:
490 			case EINVAL:
491 				break;
492 			default:
493 				error = error2;
494 				break;
495 			}
496 		}
497 		if (error) {
498 			if (silent)
499 				goto out;
500 			goto bad2;
501 		}
502 
503 		periph->periph_flags |= PERIPH_OPEN;
504 
505 		if (periph->periph_flags & PERIPH_REMOVABLE) {
506 			/* Lock the pack in. */
507 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
508 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
509 			    XS_CTL_IGNORE_MEDIA_CHANGE |
510 			    XS_CTL_SILENT);
511 			if (error)
512 				goto bad3;
513 		}
514 
515 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
516 			int param_error;
517 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
518 
519 			/*
520 			 * Load the physical device parameters.
521 			 *
522 			 * Note that if media is present but unformatted,
523 			 * we allow the open (so that it can be formatted!).
524 			 * The drive should refuse real I/O, if the media is
525 			 * unformatted.
526 			 */
527 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
528 			     == SDGP_RESULT_OFFLINE) {
529 				error = ENXIO;
530 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
531 				goto bad3;
532 			}
533 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
534 
535 			/* Load the partition info if not already loaded. */
536 			if (param_error == 0) {
537 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
538 					error = EIO;
539 					goto bad3;
540 				}
541 				SC_DEBUG(periph, SCSIPI_DB3,
542 				     ("Disklabel loaded "));
543 			}
544 		}
545 	}
546 
547 	/* Check that the partition exists. */
548 	if (part != RAW_PART &&
549 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
550 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
551 		error = ENXIO;
552 		goto bad3;
553 	}
554 
555  out:	/* Insure only one open at a time. */
556 	switch (fmt) {
557 	case S_IFCHR:
558 		sd->sc_dk.dk_copenmask |= (1 << part);
559 		break;
560 	case S_IFBLK:
561 		sd->sc_dk.dk_bopenmask |= (1 << part);
562 		break;
563 	}
564 	sd->sc_dk.dk_openmask =
565 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
566 
567 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
568 	mutex_exit(&sd->sc_dk.dk_openlock);
569 	return (0);
570 
571  bad3:
572 	if (sd->sc_dk.dk_openmask == 0) {
573 		if (periph->periph_flags & PERIPH_REMOVABLE)
574 			scsipi_prevent(periph, SPAMR_ALLOW,
575 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
576 			    XS_CTL_IGNORE_MEDIA_CHANGE |
577 			    XS_CTL_SILENT);
578 		periph->periph_flags &= ~PERIPH_OPEN;
579 	}
580 
581  bad2:
582 	if (sd->sc_dk.dk_openmask == 0)
583 		scsipi_adapter_delref(adapt);
584 
585  bad1:
586 	mutex_exit(&sd->sc_dk.dk_openlock);
587 	return (error);
588 }
589 
590 /*
591  * close the device.. only called if we are the LAST occurence of an open
592  * device.  Convenient now but usually a pain.
593  */
594 static int
595 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
596 {
597 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
598 	struct scsipi_periph *periph = sd->sc_periph;
599 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
600 	int part = SDPART(dev);
601 
602 	mutex_enter(&sd->sc_dk.dk_openlock);
603 	switch (fmt) {
604 	case S_IFCHR:
605 		sd->sc_dk.dk_copenmask &= ~(1 << part);
606 		break;
607 	case S_IFBLK:
608 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
609 		break;
610 	}
611 	sd->sc_dk.dk_openmask =
612 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
613 
614 	if (sd->sc_dk.dk_openmask == 0) {
615 		/*
616 		 * If the disk cache needs flushing, and the disk supports
617 		 * it, do it now.
618 		 */
619 		if ((sd->flags & SDF_DIRTY) != 0) {
620 			if (sd_flush(sd, 0)) {
621 				aprint_error_dev(sd->sc_dev,
622 					"cache synchronization failed\n");
623 				sd->flags &= ~SDF_FLUSHING;
624 			} else
625 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
626 		}
627 
628 		scsipi_wait_drain(periph);
629 
630 		if (periph->periph_flags & PERIPH_REMOVABLE)
631 			scsipi_prevent(periph, SPAMR_ALLOW,
632 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
633 			    XS_CTL_IGNORE_NOT_READY |
634 			    XS_CTL_SILENT);
635 		periph->periph_flags &= ~PERIPH_OPEN;
636 
637 		scsipi_wait_drain(periph);
638 
639 		scsipi_adapter_delref(adapt);
640 	}
641 
642 	mutex_exit(&sd->sc_dk.dk_openlock);
643 	return (0);
644 }
645 
646 /*
647  * Actually translate the requested transfer into one the physical driver
648  * can understand.  The transfer is described by a buf and will include
649  * only one physical transfer.
650  */
651 static void
652 sdstrategy(struct buf *bp)
653 {
654 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
655 	struct scsipi_periph *periph = sd->sc_periph;
656 	struct disklabel *lp;
657 	daddr_t blkno;
658 	int s;
659 	bool sector_aligned;
660 
661 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
662 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
663 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
664 	/*
665 	 * If the device has been made invalid, error out
666 	 */
667 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
668 	    !device_is_active(sd->sc_dev)) {
669 		if (periph->periph_flags & PERIPH_OPEN)
670 			bp->b_error = EIO;
671 		else
672 			bp->b_error = ENODEV;
673 		goto done;
674 	}
675 
676 	lp = sd->sc_dk.dk_label;
677 
678 	/*
679 	 * The transfer must be a whole number of blocks, offset must not be
680 	 * negative.
681 	 */
682 	if (lp->d_secsize == DEV_BSIZE) {
683 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
684 	} else {
685 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
686 	}
687 	if (!sector_aligned || bp->b_blkno < 0) {
688 		bp->b_error = EINVAL;
689 		goto done;
690 	}
691 	/*
692 	 * If it's a null transfer, return immediatly
693 	 */
694 	if (bp->b_bcount == 0)
695 		goto done;
696 
697 	/*
698 	 * Do bounds checking, adjust transfer. if error, process.
699 	 * If end of partition, just return.
700 	 */
701 	if (SDPART(bp->b_dev) == RAW_PART) {
702 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
703 		    sd->params.disksize512) <= 0)
704 			goto done;
705 	} else {
706 		if (bounds_check_with_label(&sd->sc_dk, bp,
707 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
708 			goto done;
709 	}
710 
711 	/*
712 	 * Now convert the block number to absolute and put it in
713 	 * terms of the device's logical block size.
714 	 */
715 	if (lp->d_secsize == DEV_BSIZE)
716 		blkno = bp->b_blkno;
717 	else if (lp->d_secsize > DEV_BSIZE)
718 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
719 	else
720 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
721 
722 	if (SDPART(bp->b_dev) != RAW_PART)
723 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
724 
725 	bp->b_rawblkno = blkno;
726 
727 	s = splbio();
728 
729 	/*
730 	 * Place it in the queue of disk activities for this disk.
731 	 *
732 	 * XXX Only do disksort() if the current operating mode does not
733 	 * XXX include tagged queueing.
734 	 */
735 	bufq_put(sd->buf_queue, bp);
736 
737 	/*
738 	 * Tell the device to get going on the transfer if it's
739 	 * not doing anything, otherwise just wait for completion
740 	 */
741 	sdstart(sd->sc_periph);
742 
743 	splx(s);
744 	return;
745 
746 done:
747 	/*
748 	 * Correctly set the buf to indicate a completed xfer
749 	 */
750 	bp->b_resid = bp->b_bcount;
751 	biodone(bp);
752 }
753 
754 /*
755  * sdstart looks to see if there is a buf waiting for the device
756  * and that the device is not already busy. If both are true,
757  * It dequeues the buf and creates a scsi command to perform the
758  * transfer in the buf. The transfer request will call scsipi_done
759  * on completion, which will in turn call this routine again
760  * so that the next queued transfer is performed.
761  * The bufs are queued by the strategy routine (sdstrategy)
762  *
763  * This routine is also called after other non-queued requests
764  * have been made of the scsi driver, to ensure that the queue
765  * continues to be drained.
766  *
767  * must be called at the correct (highish) spl level
768  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
769  */
770 static void
771 sdstart(struct scsipi_periph *periph)
772 {
773 	struct sd_softc *sd = device_private(periph->periph_dev);
774 	struct disklabel *lp = sd->sc_dk.dk_label;
775 	struct buf *bp = 0;
776 	struct scsipi_rw_16 cmd16;
777 	struct scsipi_rw_10 cmd_big;
778 	struct scsi_rw_6 cmd_small;
779 	struct scsipi_generic *cmdp;
780 	struct scsipi_xfer *xs;
781 	int nblks, cmdlen, error, flags;
782 
783 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
784 	/*
785 	 * Check if the device has room for another command
786 	 */
787 	while (periph->periph_active < periph->periph_openings) {
788 		/*
789 		 * there is excess capacity, but a special waits
790 		 * It'll need the adapter as soon as we clear out of the
791 		 * way and let it run (user level wait).
792 		 */
793 		if (periph->periph_flags & PERIPH_WAITING) {
794 			periph->periph_flags &= ~PERIPH_WAITING;
795 			wakeup((void *)periph);
796 			return;
797 		}
798 
799 		/*
800 		 * If the device has become invalid, abort all the
801 		 * reads and writes until all files have been closed and
802 		 * re-opened
803 		 */
804 		if (__predict_false(
805 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
806 			if ((bp = bufq_get(sd->buf_queue)) != NULL) {
807 				bp->b_error = EIO;
808 				bp->b_resid = bp->b_bcount;
809 				biodone(bp);
810 				continue;
811 			} else {
812 				return;
813 			}
814 		}
815 
816 		/*
817 		 * See if there is a buf with work for us to do..
818 		 */
819 		if ((bp = bufq_peek(sd->buf_queue)) == NULL)
820 			return;
821 
822 		/*
823 		 * We have a buf, now we should make a command.
824 		 */
825 
826 		if (lp->d_secsize == DEV_BSIZE)
827 			nblks = bp->b_bcount >> DEV_BSHIFT;
828 		else
829 			nblks = howmany(bp->b_bcount, lp->d_secsize);
830 
831 		/*
832 		 * Fill out the scsi command.  Use the smallest CDB possible
833 		 * (6-byte, 10-byte, or 16-byte).
834 		 */
835 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
836 		    ((nblks & 0xff) == nblks) &&
837 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
838 			/* 6-byte CDB */
839 			memset(&cmd_small, 0, sizeof(cmd_small));
840 			cmd_small.opcode = (bp->b_flags & B_READ) ?
841 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
842 			_lto3b(bp->b_rawblkno, cmd_small.addr);
843 			cmd_small.length = nblks & 0xff;
844 			cmdlen = sizeof(cmd_small);
845 			cmdp = (struct scsipi_generic *)&cmd_small;
846 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
847 			/* 10-byte CDB */
848 			memset(&cmd_big, 0, sizeof(cmd_big));
849 			cmd_big.opcode = (bp->b_flags & B_READ) ?
850 			    READ_10 : WRITE_10;
851 			_lto4b(bp->b_rawblkno, cmd_big.addr);
852 			_lto2b(nblks, cmd_big.length);
853 			cmdlen = sizeof(cmd_big);
854 			cmdp = (struct scsipi_generic *)&cmd_big;
855 		} else {
856 			/* 16-byte CDB */
857 			memset(&cmd16, 0, sizeof(cmd16));
858 			cmd16.opcode = (bp->b_flags & B_READ) ?
859 			    READ_16 : WRITE_16;
860 			_lto8b(bp->b_rawblkno, cmd16.addr);
861 			_lto4b(nblks, cmd16.length);
862 			cmdlen = sizeof(cmd16);
863 			cmdp = (struct scsipi_generic *)&cmd16;
864 		}
865 
866 		/* Instrumentation. */
867 		disk_busy(&sd->sc_dk);
868 
869 		/*
870 		 * Mark the disk dirty so that the cache will be
871 		 * flushed on close.
872 		 */
873 		if ((bp->b_flags & B_READ) == 0)
874 			sd->flags |= SDF_DIRTY;
875 
876 		/*
877 		 * Figure out what flags to use.
878 		 */
879 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
880 		if (bp->b_flags & B_READ)
881 			flags |= XS_CTL_DATA_IN;
882 		else
883 			flags |= XS_CTL_DATA_OUT;
884 
885 		/*
886 		 * Call the routine that chats with the adapter.
887 		 * Note: we cannot sleep as we may be an interrupt
888 		 */
889 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
890 		    (u_char *)bp->b_data, bp->b_bcount,
891 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
892 		if (__predict_false(xs == NULL)) {
893 			/*
894 			 * out of memory. Keep this buffer in the queue, and
895 			 * retry later.
896 			 */
897 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
898 			    periph);
899 			return;
900 		}
901 		/*
902 		 * need to dequeue the buffer before queuing the command,
903 		 * because cdstart may be called recursively from the
904 		 * HBA driver
905 		 */
906 #ifdef DIAGNOSTIC
907 		if (bufq_get(sd->buf_queue) != bp)
908 			panic("sdstart(): dequeued wrong buf");
909 #else
910 		bufq_get(sd->buf_queue);
911 #endif
912 		error = scsipi_execute_xs(xs);
913 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
914 		KASSERT(error == 0);
915 	}
916 }
917 
918 static void
919 sdrestart(void *v)
920 {
921 	int s = splbio();
922 	sdstart((struct scsipi_periph *)v);
923 	splx(s);
924 }
925 
926 static void
927 sddone(struct scsipi_xfer *xs, int error)
928 {
929 	struct sd_softc *sd = device_private(xs->xs_periph->periph_dev);
930 	struct buf *bp = xs->bp;
931 
932 	if (sd->flags & SDF_FLUSHING) {
933 		/* Flush completed, no longer dirty. */
934 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
935 	}
936 
937 	if (bp) {
938 		bp->b_error = error;
939 		bp->b_resid = xs->resid;
940 		if (error) {
941 			/* on a read/write error bp->b_resid is zero, so fix */
942 			bp->b_resid = bp->b_bcount;
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 = device_lookup_private(&sd_cd, 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, void *addr, int flag, struct lwp *l)
1004 {
1005 	struct sd_softc *sd = device_lookup_private(&sd_cd, 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 		mutex_enter(&sd->sc_dk.dk_openlock);
1093 		sd->flags |= SDF_LABELLING;
1094 
1095 		error = setdisklabel(sd->sc_dk.dk_label,
1096 		    lp, /*sd->sc_dk.dk_openmask : */0,
1097 		    sd->sc_dk.dk_cpulabel);
1098 		if (error == 0) {
1099 			if (cmd == DIOCWDINFO
1100 #ifdef __HAVE_OLD_DISKLABEL
1101 			    || cmd == ODIOCWDINFO
1102 #endif
1103 			   )
1104 				error = writedisklabel(SDLABELDEV(dev),
1105 				    sdstrategy, sd->sc_dk.dk_label,
1106 				    sd->sc_dk.dk_cpulabel);
1107 		}
1108 
1109 		sd->flags &= ~SDF_LABELLING;
1110 		mutex_exit(&sd->sc_dk.dk_openlock);
1111 #ifdef __HAVE_OLD_DISKLABEL
1112 		if (newlabel != NULL)
1113 			free(newlabel, M_TEMP);
1114 #endif
1115 		return (error);
1116 	}
1117 
1118 	case DIOCKLABEL:
1119 		if (*(int *)addr)
1120 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1121 		else
1122 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1123 		return (0);
1124 
1125 	case DIOCWLABEL:
1126 		if ((flag & FWRITE) == 0)
1127 			return (EBADF);
1128 		if (*(int *)addr)
1129 			sd->flags |= SDF_WLABEL;
1130 		else
1131 			sd->flags &= ~SDF_WLABEL;
1132 		return (0);
1133 
1134 	case DIOCLOCK:
1135 		if (periph->periph_flags & PERIPH_REMOVABLE)
1136 			return (scsipi_prevent(periph,
1137 			    (*(int *)addr) ?
1138 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1139 		else
1140 			return (ENOTTY);
1141 
1142 	case DIOCEJECT:
1143 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1144 			return (ENOTTY);
1145 		if (*(int *)addr == 0) {
1146 			/*
1147 			 * Don't force eject: check that we are the only
1148 			 * partition open. If so, unlock it.
1149 			 */
1150 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1151 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1152 			    sd->sc_dk.dk_openmask) {
1153 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1154 				    XS_CTL_IGNORE_NOT_READY);
1155 				if (error)
1156 					return (error);
1157 			} else {
1158 				return (EBUSY);
1159 			}
1160 		}
1161 		/* FALLTHROUGH */
1162 	case ODIOCEJECT:
1163 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1164 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1165 
1166 	case DIOCGDEFLABEL:
1167 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1168 		return (0);
1169 
1170 #ifdef __HAVE_OLD_DISKLABEL
1171 	case ODIOCGDEFLABEL:
1172 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1173 		if (newlabel == NULL)
1174 			return EIO;
1175 		sdgetdefaultlabel(sd, newlabel);
1176 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1177 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1178 		else
1179 			error = ENOTTY;
1180 		free(newlabel, M_TEMP);
1181 		return error;
1182 #endif
1183 
1184 	case DIOCGCACHE:
1185 		return (sd_getcache(sd, (int *) addr));
1186 
1187 	case DIOCSCACHE:
1188 		if ((flag & FWRITE) == 0)
1189 			return (EBADF);
1190 		return (sd_setcache(sd, *(int *) addr));
1191 
1192 	case DIOCCACHESYNC:
1193 		/*
1194 		 * XXX Do we really need to care about having a writable
1195 		 * file descriptor here?
1196 		 */
1197 		if ((flag & FWRITE) == 0)
1198 			return (EBADF);
1199 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1200 			error = sd_flush(sd, 0);
1201 			if (error)
1202 				sd->flags &= ~SDF_FLUSHING;
1203 			else
1204 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1205 		} else
1206 			error = 0;
1207 		return (error);
1208 
1209 	case DIOCAWEDGE:
1210 	    {
1211 	    	struct dkwedge_info *dkw = (void *) addr;
1212 
1213 		if ((flag & FWRITE) == 0)
1214 			return (EBADF);
1215 
1216 		/* If the ioctl happens here, the parent is us. */
1217 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1218 			sizeof(dkw->dkw_parent));
1219 		return (dkwedge_add(dkw));
1220 	    }
1221 
1222 	case DIOCDWEDGE:
1223 	    {
1224 	    	struct dkwedge_info *dkw = (void *) addr;
1225 
1226 		if ((flag & FWRITE) == 0)
1227 			return (EBADF);
1228 
1229 		/* If the ioctl happens here, the parent is us. */
1230 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1231 			sizeof(dkw->dkw_parent));
1232 		return (dkwedge_del(dkw));
1233 	    }
1234 
1235 	case DIOCLWEDGES:
1236 	    {
1237 	    	struct dkwedge_list *dkwl = (void *) addr;
1238 
1239 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1240 	    }
1241 
1242 	default:
1243 		if (part != RAW_PART)
1244 			return (ENOTTY);
1245 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1246 	}
1247 
1248 #ifdef DIAGNOSTIC
1249 	panic("sdioctl: impossible");
1250 #endif
1251 }
1252 
1253 static void
1254 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1255 {
1256 
1257 	memset(lp, 0, sizeof(struct disklabel));
1258 
1259 	lp->d_secsize = sd->params.blksize;
1260 	lp->d_ntracks = sd->params.heads;
1261 	lp->d_nsectors = sd->params.sectors;
1262 	lp->d_ncylinders = sd->params.cyls;
1263 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1264 
1265 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1266 	case SCSIPI_BUSTYPE_SCSI:
1267 		lp->d_type = DTYPE_SCSI;
1268 		break;
1269 	case SCSIPI_BUSTYPE_ATAPI:
1270 		lp->d_type = DTYPE_ATAPI;
1271 		break;
1272 	}
1273 	/*
1274 	 * XXX
1275 	 * We could probe the mode pages to figure out what kind of disc it is.
1276 	 * Is this worthwhile?
1277 	 */
1278 	strncpy(lp->d_typename, sd->name, 16);
1279 	strncpy(lp->d_packname, "fictitious", 16);
1280 	lp->d_secperunit = sd->params.disksize;
1281 	lp->d_rpm = sd->params.rot_rate;
1282 	lp->d_interleave = 1;
1283 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1284 	    D_REMOVABLE : 0;
1285 
1286 	lp->d_partitions[RAW_PART].p_offset = 0;
1287 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1288 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1289 	lp->d_npartitions = RAW_PART + 1;
1290 
1291 	lp->d_magic = DISKMAGIC;
1292 	lp->d_magic2 = DISKMAGIC;
1293 	lp->d_checksum = dkcksum(lp);
1294 }
1295 
1296 
1297 /*
1298  * Load the label information on the named device
1299  */
1300 static int
1301 sdgetdisklabel(struct sd_softc *sd)
1302 {
1303 	struct disklabel *lp = sd->sc_dk.dk_label;
1304 	const char *errstring;
1305 
1306 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1307 
1308 	sdgetdefaultlabel(sd, lp);
1309 
1310 	if (lp->d_secpercyl == 0) {
1311 		lp->d_secpercyl = 100;
1312 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1313 	}
1314 
1315 	/*
1316 	 * Call the generic disklabel extraction routine
1317 	 */
1318 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1319 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1320 	if (errstring) {
1321 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1322 		return EIO;
1323 	}
1324 	return 0;
1325 }
1326 
1327 static void
1328 sd_shutdown(void *arg)
1329 {
1330 	struct sd_softc *sd = arg;
1331 
1332 	/*
1333 	 * If the disk cache needs to be flushed, and the disk supports
1334 	 * it, flush it.  We're cold at this point, so we poll for
1335 	 * completion.
1336 	 */
1337 	if ((sd->flags & SDF_DIRTY) != 0) {
1338 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1339 			aprint_error_dev(sd->sc_dev,
1340 				"cache synchronization failed\n");
1341 			sd->flags &= ~SDF_FLUSHING;
1342 		} else
1343 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1344 	}
1345 }
1346 
1347 static bool
1348 sd_suspend(device_t dv PMF_FN_ARGS)
1349 {
1350 	struct sd_softc *sd = device_private(dv);
1351 
1352 	sd_shutdown(sd); /* XXX no need to poll */
1353 	return true;
1354 }
1355 
1356 /*
1357  * Check Errors
1358  */
1359 static int
1360 sd_interpret_sense(struct scsipi_xfer *xs)
1361 {
1362 	struct scsipi_periph *periph = xs->xs_periph;
1363 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1364 	struct sd_softc *sd = device_private(periph->periph_dev);
1365 	int s, error, retval = EJUSTRETURN;
1366 
1367 	/*
1368 	 * If the periph is already recovering, just do the normal
1369 	 * error processing.
1370 	 */
1371 	if (periph->periph_flags & PERIPH_RECOVERING)
1372 		return (retval);
1373 
1374 	/*
1375 	 * Ignore errors from accessing illegal fields (e.g. trying to
1376 	 * lock the door of a digicam, which doesn't have a door that
1377 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1378 	 */
1379 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1380 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1381 	    sense->asc == 0x24 &&
1382 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1383 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1384 			scsipi_printaddr(periph);
1385 			printf("no door lock\n");
1386 		}
1387 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1388 		return (retval);
1389 	}
1390 
1391 
1392 
1393 	/*
1394 	 * If the device is not open yet, let the generic code handle it.
1395 	 */
1396 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1397 		return (retval);
1398 
1399 	/*
1400 	 * If it isn't a extended or extended/deferred error, let
1401 	 * the generic code handle it.
1402 	 */
1403 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1404 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1405 		return (retval);
1406 
1407 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1408 	    sense->asc == 0x4) {
1409 		if (sense->ascq == 0x01)	{
1410 			/*
1411 			 * Unit In The Process Of Becoming Ready.
1412 			 */
1413 			printf("%s: waiting for pack to spin up...\n",
1414 			    device_xname(sd->sc_dev));
1415 			if (!callout_pending(&periph->periph_callout))
1416 				scsipi_periph_freeze(periph, 1);
1417 			callout_reset(&periph->periph_callout,
1418 			    5 * hz, scsipi_periph_timed_thaw, periph);
1419 			retval = ERESTART;
1420 		} else if (sense->ascq == 0x02) {
1421 			printf("%s: pack is stopped, restarting...\n",
1422 			    device_xname(sd->sc_dev));
1423 			s = splbio();
1424 			periph->periph_flags |= PERIPH_RECOVERING;
1425 			splx(s);
1426 			error = scsipi_start(periph, SSS_START,
1427 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1428 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1429 			if (error) {
1430 				aprint_error_dev(sd->sc_dev,
1431 					"unable to restart pack\n");
1432 				retval = error;
1433 			} else
1434 				retval = ERESTART;
1435 			s = splbio();
1436 			periph->periph_flags &= ~PERIPH_RECOVERING;
1437 			splx(s);
1438 		}
1439 	}
1440 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1441 	    sense->asc == 0x31 &&
1442 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1443 		/* Medium Format Corrupted */
1444 		retval = EFTYPE;
1445 	}
1446 	return (retval);
1447 }
1448 
1449 
1450 static int
1451 sdsize(dev_t dev)
1452 {
1453 	struct sd_softc *sd;
1454 	int part, unit, omask;
1455 	int size;
1456 
1457 	unit = SDUNIT(dev);
1458 	sd = device_lookup_private(&sd_cd, unit);
1459 	if (sd == NULL)
1460 		return (-1);
1461 
1462 	if (!device_is_active(sd->sc_dev))
1463 		return (-1);
1464 
1465 	part = SDPART(dev);
1466 	omask = sd->sc_dk.dk_openmask & (1 << part);
1467 
1468 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1469 		return (-1);
1470 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1471 		size = -1;
1472 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1473 		size = -1;
1474 	else
1475 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1476 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1477 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1478 		return (-1);
1479 	return (size);
1480 }
1481 
1482 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1483 static struct scsipi_xfer sx;
1484 static int sddoingadump;
1485 
1486 /*
1487  * dump all of physical memory into the partition specified, starting
1488  * at offset 'dumplo' into the partition.
1489  */
1490 static int
1491 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1492 {
1493 	struct sd_softc *sd;	/* disk unit to do the I/O */
1494 	struct disklabel *lp;	/* disk's disklabel */
1495 	int	unit, part;
1496 	int	sectorsize;	/* size of a disk sector */
1497 	int	nsects;		/* number of sectors in partition */
1498 	int	sectoff;	/* sector offset of partition */
1499 	int	totwrt;		/* total number of sectors left to write */
1500 	int	nwrt;		/* current number of sectors to write */
1501 	struct scsipi_rw_10 cmd;	/* write command */
1502 	struct scsipi_xfer *xs;	/* ... convenience */
1503 	struct scsipi_periph *periph;
1504 	struct scsipi_channel *chan;
1505 
1506 	/* Check if recursive dump; if so, punt. */
1507 	if (sddoingadump)
1508 		return (EFAULT);
1509 
1510 	/* Mark as active early. */
1511 	sddoingadump = 1;
1512 
1513 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1514 	part = SDPART(dev);
1515 
1516 	/* Check for acceptable drive number. */
1517 	sd = device_lookup_private(&sd_cd, unit);
1518 	if (sd == NULL)
1519 		return (ENXIO);
1520 
1521 	if (!device_is_active(sd->sc_dev))
1522 		return (ENODEV);
1523 
1524 	periph = sd->sc_periph;
1525 	chan = periph->periph_channel;
1526 
1527 	/* Make sure it was initialized. */
1528 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1529 		return (ENXIO);
1530 
1531 	/* Convert to disk sectors.  Request must be a multiple of size. */
1532 	lp = sd->sc_dk.dk_label;
1533 	sectorsize = lp->d_secsize;
1534 	if ((size % sectorsize) != 0)
1535 		return (EFAULT);
1536 	totwrt = size / sectorsize;
1537 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1538 
1539 	nsects = lp->d_partitions[part].p_size;
1540 	sectoff = lp->d_partitions[part].p_offset;
1541 
1542 	/* Check transfer bounds against partition size. */
1543 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1544 		return (EINVAL);
1545 
1546 	/* Offset block number to start of partition. */
1547 	blkno += sectoff;
1548 
1549 	xs = &sx;
1550 
1551 	while (totwrt > 0) {
1552 		nwrt = totwrt;		/* XXX */
1553 #ifndef	SD_DUMP_NOT_TRUSTED
1554 		/*
1555 		 *  Fill out the scsi command
1556 		 */
1557 		memset(&cmd, 0, sizeof(cmd));
1558 		cmd.opcode = WRITE_10;
1559 		_lto4b(blkno, cmd.addr);
1560 		_lto2b(nwrt, cmd.length);
1561 		/*
1562 		 * Fill out the scsipi_xfer structure
1563 		 *    Note: we cannot sleep as we may be an interrupt
1564 		 * don't use scsipi_command() as it may want to wait
1565 		 * for an xs.
1566 		 */
1567 		memset(xs, 0, sizeof(sx));
1568 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1569 		    XS_CTL_DATA_OUT;
1570 		xs->xs_status = 0;
1571 		xs->xs_periph = periph;
1572 		xs->xs_retries = SDRETRIES;
1573 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1574 		xs->cmd = (struct scsipi_generic *)&cmd;
1575 		xs->cmdlen = sizeof(cmd);
1576 		xs->resid = nwrt * sectorsize;
1577 		xs->error = XS_NOERROR;
1578 		xs->bp = 0;
1579 		xs->data = va;
1580 		xs->datalen = nwrt * sectorsize;
1581 		callout_init(&xs->xs_callout, 0);
1582 
1583 		/*
1584 		 * Pass all this info to the scsi driver.
1585 		 */
1586 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1587 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1588 		    xs->error != XS_NOERROR)
1589 			return (EIO);
1590 #else	/* SD_DUMP_NOT_TRUSTED */
1591 		/* Let's just talk about this first... */
1592 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1593 		delay(500 * 1000);	/* half a second */
1594 #endif	/* SD_DUMP_NOT_TRUSTED */
1595 
1596 		/* update block count */
1597 		totwrt -= nwrt;
1598 		blkno += nwrt;
1599 		va = (char *)va + sectorsize * nwrt;
1600 	}
1601 	sddoingadump = 0;
1602 	return (0);
1603 }
1604 
1605 static int
1606 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1607     int page, int flags, int *big)
1608 {
1609 
1610 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1611 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1612 		*big = 1;
1613 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1614 		    size + sizeof(struct scsi_mode_parameter_header_10),
1615 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1616 	} else {
1617 		*big = 0;
1618 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1619 		    size + sizeof(struct scsi_mode_parameter_header_6),
1620 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1621 	}
1622 }
1623 
1624 static int
1625 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1626     int flags, int big)
1627 {
1628 
1629 	if (big) {
1630 		struct scsi_mode_parameter_header_10 *header = sense;
1631 
1632 		_lto2b(0, header->data_length);
1633 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1634 		    size + sizeof(struct scsi_mode_parameter_header_10),
1635 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1636 	} else {
1637 		struct scsi_mode_parameter_header_6 *header = sense;
1638 
1639 		header->data_length = 0;
1640 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1641 		    size + sizeof(struct scsi_mode_parameter_header_6),
1642 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1643 	}
1644 }
1645 
1646 /*
1647  * sd_validate_blksize:
1648  *
1649  *	Validate the block size.  Print error if periph is specified,
1650  */
1651 static int
1652 sd_validate_blksize(struct scsipi_periph *periph, int len)
1653 {
1654 
1655 	switch (len) {
1656 	case 256:
1657 	case 512:
1658 	case 1024:
1659 	case 2048:
1660 	case 4096:
1661 		return 1;
1662 	}
1663 
1664 	if (periph) {
1665 		scsipi_printaddr(periph);
1666 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1667 		    (len ^ (1 << (ffs(len) - 1))) ?
1668 		    "preposterous" : "unsupported",
1669 		    len, SD_DEFAULT_BLKSIZE);
1670 	}
1671 
1672 	return 0;
1673 }
1674 
1675 /*
1676  * sd_read_capacity:
1677  *
1678  *	Find out from the device what its capacity is.
1679  */
1680 static u_int64_t
1681 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1682 {
1683 	union {
1684 		struct scsipi_read_capacity_10 cmd;
1685 		struct scsipi_read_capacity_16 cmd16;
1686 	} cmd;
1687 	union {
1688 		struct scsipi_read_capacity_10_data data;
1689 		struct scsipi_read_capacity_16_data data16;
1690 	} *datap;
1691 	uint64_t rv;
1692 
1693 	memset(&cmd, 0, sizeof(cmd));
1694 	cmd.cmd.opcode = READ_CAPACITY_10;
1695 
1696 	/*
1697 	 * Don't allocate data buffer on stack;
1698 	 * The lower driver layer might use the same stack and
1699 	 * if it uses region which is in the same cacheline,
1700 	 * cache flush ops against the data buffer won't work properly.
1701 	 */
1702 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1703 	if (datap == NULL)
1704 		return 0;
1705 
1706 	/*
1707 	 * If the command works, interpret the result as a 4 byte
1708 	 * number of blocks
1709 	 */
1710 	rv = 0;
1711 	memset(datap, 0, sizeof(datap->data));
1712 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1713 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1714 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1715 		goto out;
1716 
1717 	if (_4btol(datap->data.addr) != 0xffffffff) {
1718 		*blksize = _4btol(datap->data.length);
1719 		rv = _4btol(datap->data.addr) + 1;
1720 		goto out;
1721 	}
1722 
1723 	/*
1724 	 * Device is larger than can be reflected by READ CAPACITY (10).
1725 	 * Try READ CAPACITY (16).
1726 	 */
1727 
1728 	memset(&cmd, 0, sizeof(cmd));
1729 	cmd.cmd16.opcode = READ_CAPACITY_16;
1730 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1731 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1732 
1733 	memset(datap, 0, sizeof(datap->data16));
1734 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1735 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1736 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1737 		goto out;
1738 
1739 	*blksize = _4btol(datap->data16.length);
1740 	rv = _8btol(datap->data16.addr) + 1;
1741 
1742  out:
1743 	free(datap, M_TEMP);
1744 	return rv;
1745 }
1746 
1747 static int
1748 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1749 {
1750 	struct {
1751 		struct scsi_mode_parameter_header_6 header;
1752 		/* no block descriptor */
1753 		u_int8_t pg_code; /* page code (should be 6) */
1754 		u_int8_t pg_length; /* page length (should be 11) */
1755 		u_int8_t wcd; /* bit0: cache disable */
1756 		u_int8_t lbs[2]; /* logical block size */
1757 		u_int8_t size[5]; /* number of log. blocks */
1758 		u_int8_t pp; /* power/performance */
1759 		u_int8_t flags;
1760 		u_int8_t resvd;
1761 	} scsipi_sense;
1762 	u_int64_t blocks;
1763 	int error, blksize;
1764 
1765 	/*
1766 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1767 	 * give the same information. Do both for now, and check
1768 	 * for consistency.
1769 	 * XXX probably differs for removable media
1770 	 */
1771 	dp->blksize = SD_DEFAULT_BLKSIZE;
1772 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1773 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1774 
1775 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1776 	    &scsipi_sense.header, sizeof(scsipi_sense),
1777 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1778 
1779 	if (error != 0)
1780 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1781 
1782 	dp->blksize = blksize;
1783 	if (!sd_validate_blksize(NULL, dp->blksize))
1784 		dp->blksize = _2btol(scsipi_sense.lbs);
1785 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1786 		dp->blksize = SD_DEFAULT_BLKSIZE;
1787 
1788 	/*
1789 	 * Create a pseudo-geometry.
1790 	 */
1791 	dp->heads = 64;
1792 	dp->sectors = 32;
1793 	dp->cyls = blocks / (dp->heads * dp->sectors);
1794 	dp->disksize = _5btol(scsipi_sense.size);
1795 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1796 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1797 		       (unsigned long long)dp->disksize,
1798 		       (unsigned long long)blocks);
1799 		dp->disksize = blocks;
1800 	}
1801 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1802 
1803 	return (SDGP_RESULT_OK);
1804 }
1805 
1806 /*
1807  * Get the scsi driver to send a full inquiry to the * device and use the
1808  * results to fill out the disk parameter structure.
1809  */
1810 static int
1811 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1812 {
1813 	u_int64_t blocks;
1814 	int error, blksize;
1815 #if 0
1816 	int i;
1817 	u_int8_t *p;
1818 #endif
1819 
1820 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1821 	    flags);
1822 	if (blocks == 0) {
1823 		struct scsipi_read_format_capacities cmd;
1824 		struct {
1825 			struct scsipi_capacity_list_header header;
1826 			struct scsipi_capacity_descriptor desc;
1827 		} __packed data;
1828 
1829 		memset(&cmd, 0, sizeof(cmd));
1830 		memset(&data, 0, sizeof(data));
1831 		cmd.opcode = READ_FORMAT_CAPACITIES;
1832 		_lto2b(sizeof(data), cmd.length);
1833 
1834 		error = scsipi_command(sd->sc_periph,
1835 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1836 		    SDRETRIES, 20000, NULL,
1837 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1838 		if (error == EFTYPE) {
1839 			/* Medium Format Corrupted, handle as not formatted */
1840 			return (SDGP_RESULT_UNFORMATTED);
1841 		}
1842 		if (error || data.header.length == 0)
1843 			return (SDGP_RESULT_OFFLINE);
1844 
1845 #if 0
1846 printf("rfc: length=%d\n", data.header.length);
1847 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");
1848 #endif
1849 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1850 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1851 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1852 			break;
1853 
1854 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1855 			return (SDGP_RESULT_UNFORMATTED);
1856 
1857 		case SCSIPI_CAP_DESC_CODE_NONE:
1858 			return (SDGP_RESULT_OFFLINE);
1859 		}
1860 
1861 		dp->disksize = blocks = _4btol(data.desc.nblks);
1862 		if (blocks == 0)
1863 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1864 
1865 		blksize = _3btol(data.desc.blklen);
1866 
1867 	} else if (!sd_validate_blksize(NULL, blksize)) {
1868 		struct sd_mode_sense_data scsipi_sense;
1869 		int big, bsize;
1870 		struct scsi_general_block_descriptor *bdesc;
1871 
1872 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1873 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1874 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1875 		if (!error) {
1876 			if (big) {
1877 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1878 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1879 			} else {
1880 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1881 				bsize = scsipi_sense.header.small.blk_desc_len;
1882 			}
1883 
1884 #if 0
1885 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1886 printf("page 0 bsize=%d\n", bsize);
1887 printf("page 0 ok\n");
1888 #endif
1889 
1890 			if (bsize >= 8) {
1891 				blksize = _3btol(bdesc->blklen);
1892 			}
1893 		}
1894 	}
1895 
1896 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1897 		blksize = SD_DEFAULT_BLKSIZE;
1898 
1899 	dp->blksize = blksize;
1900 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1901 	return (0);
1902 }
1903 
1904 static int
1905 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1906 {
1907 	struct sd_mode_sense_data scsipi_sense;
1908 	int error;
1909 	int big, byte2;
1910 	size_t poffset;
1911 	union scsi_disk_pages *pages;
1912 
1913 	byte2 = SMS_DBD;
1914 again:
1915 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1916 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1917 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1918 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1919 	    flags | XS_CTL_SILENT, &big);
1920 	if (error) {
1921 		if (byte2 == SMS_DBD) {
1922 			/* No result; try once more with DBD off */
1923 			byte2 = 0;
1924 			goto again;
1925 		}
1926 		return (error);
1927 	}
1928 
1929 	if (big) {
1930 		poffset = sizeof scsipi_sense.header.big;
1931 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1932 	} else {
1933 		poffset = sizeof scsipi_sense.header.small;
1934 		poffset += scsipi_sense.header.small.blk_desc_len;
1935 	}
1936 
1937 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1938 		return ERESTART;
1939 
1940 	pages = (void *)((u_long)&scsipi_sense + poffset);
1941 #if 0
1942 	{
1943 		size_t i;
1944 		u_int8_t *p;
1945 
1946 		printf("page 4 sense:");
1947 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
1948 		    i--, p++)
1949 			printf(" %02x", *p);
1950 		printf("\n");
1951 		printf("page 4 pg_code=%d sense=%p/%p\n",
1952 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1953 	}
1954 #endif
1955 
1956 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1957 		return (ERESTART);
1958 
1959 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1960 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1961 	    _3btol(pages->rigid_geometry.ncyl),
1962 	    pages->rigid_geometry.nheads,
1963 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1964 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1965 	    _2btol(pages->rigid_geometry.land_zone)));
1966 
1967 	/*
1968 	 * KLUDGE!! (for zone recorded disks)
1969 	 * give a number of sectors so that sec * trks * cyls
1970 	 * is <= disk_size
1971 	 * can lead to wasted space! THINK ABOUT THIS !
1972 	 */
1973 	dp->heads = pages->rigid_geometry.nheads;
1974 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1975 	if (dp->heads == 0 || dp->cyls == 0)
1976 		return (ERESTART);
1977 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1978 
1979 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1980 	if (dp->rot_rate == 0)
1981 		dp->rot_rate = 3600;
1982 
1983 #if 0
1984 printf("page 4 ok\n");
1985 #endif
1986 	return (0);
1987 }
1988 
1989 static int
1990 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1991 {
1992 	struct sd_mode_sense_data scsipi_sense;
1993 	int error;
1994 	int big, byte2;
1995 	size_t poffset;
1996 	union scsi_disk_pages *pages;
1997 
1998 	byte2 = SMS_DBD;
1999 again:
2000 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2001 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2002 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2003 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2004 	    flags | XS_CTL_SILENT, &big);
2005 	if (error) {
2006 		if (byte2 == SMS_DBD) {
2007 			/* No result; try once more with DBD off */
2008 			byte2 = 0;
2009 			goto again;
2010 		}
2011 		return (error);
2012 	}
2013 
2014 	if (big) {
2015 		poffset = sizeof scsipi_sense.header.big;
2016 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2017 	} else {
2018 		poffset = sizeof scsipi_sense.header.small;
2019 		poffset += scsipi_sense.header.small.blk_desc_len;
2020 	}
2021 
2022 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2023 		return ERESTART;
2024 
2025 	pages = (void *)((u_long)&scsipi_sense + poffset);
2026 #if 0
2027 	{
2028 		size_t i;
2029 		u_int8_t *p;
2030 
2031 		printf("page 5 sense:");
2032 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2033 		    i--, p++)
2034 			printf(" %02x", *p);
2035 		printf("\n");
2036 		printf("page 5 pg_code=%d sense=%p/%p\n",
2037 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2038 	}
2039 #endif
2040 
2041 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2042 		return (ERESTART);
2043 
2044 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2045 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2046 	    _3btol(pages->flex_geometry.ncyl),
2047 	    pages->flex_geometry.nheads,
2048 	    pages->flex_geometry.ph_sec_tr,
2049 	    _2btol(pages->flex_geometry.bytes_s)));
2050 
2051 	dp->heads = pages->flex_geometry.nheads;
2052 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2053 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2054 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2055 		return (ERESTART);
2056 
2057 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2058 	if (dp->rot_rate == 0)
2059 		dp->rot_rate = 3600;
2060 
2061 #if 0
2062 printf("page 5 ok\n");
2063 #endif
2064 	return (0);
2065 }
2066 
2067 static int
2068 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2069 {
2070 	int error;
2071 
2072 	/*
2073 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2074 	 * cleared by the caller if necessary.
2075 	 */
2076 	if (sd->type == T_SIMPLE_DIRECT) {
2077 		error = sd_get_simplifiedparms(sd, dp, flags);
2078 		if (!error)
2079 			disk_blocksize(&sd->sc_dk, dp->blksize);
2080 		return (error);
2081 	}
2082 
2083 	error = sd_get_capacity(sd, dp, flags);
2084 	if (error)
2085 		return (error);
2086 
2087 	disk_blocksize(&sd->sc_dk, dp->blksize);
2088 
2089 	if (sd->type == T_OPTICAL)
2090 		goto page0;
2091 
2092 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2093 		if (!sd_get_parms_page5(sd, dp, flags) ||
2094 		    !sd_get_parms_page4(sd, dp, flags))
2095 			return (SDGP_RESULT_OK);
2096 	} else {
2097 		if (!sd_get_parms_page4(sd, dp, flags) ||
2098 		    !sd_get_parms_page5(sd, dp, flags))
2099 			return (SDGP_RESULT_OK);
2100 	}
2101 
2102 page0:
2103 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2104 	/* Try calling driver's method for figuring out geometry. */
2105 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2106 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2107 		(sd->sc_periph, dp, dp->disksize)) {
2108 		/*
2109 		 * Use adaptec standard fictitious geometry
2110 		 * this depends on which controller (e.g. 1542C is
2111 		 * different. but we have to put SOMETHING here..)
2112 		 */
2113 		dp->heads = 64;
2114 		dp->sectors = 32;
2115 		dp->cyls = dp->disksize / (64 * 32);
2116 	}
2117 	dp->rot_rate = 3600;
2118 	return (SDGP_RESULT_OK);
2119 }
2120 
2121 static int
2122 sd_flush(struct sd_softc *sd, int flags)
2123 {
2124 	struct scsipi_periph *periph = sd->sc_periph;
2125 	struct scsi_synchronize_cache_10 cmd;
2126 
2127 	/*
2128 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2129 	 * We issue with address 0 length 0, which should be
2130 	 * interpreted by the device as "all remaining blocks
2131 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2132 	 * in the event that the command is not supported by
2133 	 * the device, and poll for completion so that we know
2134 	 * that the cache has actually been flushed.
2135 	 *
2136 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2137 	 * command, as indicated by our quirks flags.
2138 	 *
2139 	 * XXX What about older devices?
2140 	 */
2141 	if (periph->periph_version < 2 ||
2142 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2143 		return (0);
2144 
2145 	sd->flags |= SDF_FLUSHING;
2146 	memset(&cmd, 0, sizeof(cmd));
2147 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2148 
2149 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2150 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2151 }
2152 
2153 static int
2154 sd_getcache(struct sd_softc *sd, int *bitsp)
2155 {
2156 	struct scsipi_periph *periph = sd->sc_periph;
2157 	struct sd_mode_sense_data scsipi_sense;
2158 	int error, bits = 0;
2159 	int big;
2160 	union scsi_disk_pages *pages;
2161 
2162 	if (periph->periph_version < 2)
2163 		return (EOPNOTSUPP);
2164 
2165 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2166 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2167 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2168 	if (error)
2169 		return (error);
2170 
2171 	if (big)
2172 		pages = (void *)(&scsipi_sense.header.big + 1);
2173 	else
2174 		pages = (void *)(&scsipi_sense.header.small + 1);
2175 
2176 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2177 		bits |= DKCACHE_READ;
2178 	if (pages->caching_params.flags & CACHING_WCE)
2179 		bits |= DKCACHE_WRITE;
2180 	if (pages->caching_params.pg_code & PGCODE_PS)
2181 		bits |= DKCACHE_SAVE;
2182 
2183 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2184 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2185 	    sizeof(scsipi_sense.pages.caching_params),
2186 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2187 	if (error == 0) {
2188 		if (big)
2189 			pages = (void *)(&scsipi_sense.header.big + 1);
2190 		else
2191 			pages = (void *)(&scsipi_sense.header.small + 1);
2192 
2193 		if (pages->caching_params.flags & CACHING_RCD)
2194 			bits |= DKCACHE_RCHANGE;
2195 		if (pages->caching_params.flags & CACHING_WCE)
2196 			bits |= DKCACHE_WCHANGE;
2197 	}
2198 
2199 	*bitsp = bits;
2200 
2201 	return (0);
2202 }
2203 
2204 static int
2205 sd_setcache(struct sd_softc *sd, int bits)
2206 {
2207 	struct scsipi_periph *periph = sd->sc_periph;
2208 	struct sd_mode_sense_data scsipi_sense;
2209 	int error;
2210 	uint8_t oflags, byte2 = 0;
2211 	int big;
2212 	union scsi_disk_pages *pages;
2213 
2214 	if (periph->periph_version < 2)
2215 		return (EOPNOTSUPP);
2216 
2217 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2218 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2219 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2220 	if (error)
2221 		return (error);
2222 
2223 	if (big)
2224 		pages = (void *)(&scsipi_sense.header.big + 1);
2225 	else
2226 		pages = (void *)(&scsipi_sense.header.small + 1);
2227 
2228 	oflags = pages->caching_params.flags;
2229 
2230 	if (bits & DKCACHE_READ)
2231 		pages->caching_params.flags &= ~CACHING_RCD;
2232 	else
2233 		pages->caching_params.flags |= CACHING_RCD;
2234 
2235 	if (bits & DKCACHE_WRITE)
2236 		pages->caching_params.flags |= CACHING_WCE;
2237 	else
2238 		pages->caching_params.flags &= ~CACHING_WCE;
2239 
2240 	if (oflags == pages->caching_params.flags)
2241 		return (0);
2242 
2243 	pages->caching_params.pg_code &= PGCODE_MASK;
2244 
2245 	if (bits & DKCACHE_SAVE)
2246 		byte2 |= SMS_SP;
2247 
2248 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2249 	    sizeof(struct scsi_mode_page_header) +
2250 	    pages->caching_params.pg_length, 0, big));
2251 }
2252 
2253 static void
2254 sd_set_properties(struct sd_softc *sd)
2255 {
2256 	prop_dictionary_t disk_info, odisk_info, geom;
2257 
2258 	disk_info = prop_dictionary_create();
2259 
2260 	geom = prop_dictionary_create();
2261 
2262 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2263 	    sd->params.disksize);
2264 
2265 	prop_dictionary_set_uint32(geom, "sector-size",
2266 	    sd->params.blksize);
2267 
2268 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2269 	    sd->params.sectors);
2270 
2271 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2272 	    sd->params.heads);
2273 
2274 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2275 	    sd->params.cyls);
2276 
2277 	prop_dictionary_set(disk_info, "geometry", geom);
2278 	prop_object_release(geom);
2279 
2280 	prop_dictionary_set(device_properties(sd->sc_dev),
2281 	    "disk-info", disk_info);
2282 
2283 	/*
2284 	 * Don't release disk_info here; we keep a reference to it.
2285 	 * disk_detach() will release it when we go away.
2286 	 */
2287 
2288 	odisk_info = sd->sc_dk.dk_info;
2289 	sd->sc_dk.dk_info = disk_info;
2290 	if (odisk_info)
2291 		prop_object_release(odisk_info);
2292 }
2293