xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: sd.c,v 1.267 2007/10/08 16:41:14 ad 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 1.267 2007/10/08 16:41:14 ad 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 #include <prop/proplib.h>
94 
95 #define	SDUNIT(dev)			DISKUNIT(dev)
96 #define	SDPART(dev)			DISKPART(dev)
97 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
98 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
99 
100 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
101 
102 #define	SD_DEFAULT_BLKSIZE	512
103 
104 static void	sdminphys(struct buf *);
105 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
106 static int	sdgetdisklabel(struct sd_softc *);
107 static void	sdstart(struct scsipi_periph *);
108 static void	sdrestart(void *);
109 static void	sddone(struct scsipi_xfer *, int);
110 static void	sd_shutdown(void *);
111 static int	sd_interpret_sense(struct scsipi_xfer *);
112 
113 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
114 		    int, int *);
115 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
116 		    int);
117 static int	sd_validate_blksize(struct scsipi_periph *, int);
118 static u_int64_t sd_read_capacity(struct scsipi_periph *, int *, int flags);
119 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
120 		    int);
121 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
122 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
123 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
124 		    int);
125 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
126 		    int);
127 
128 static int	sd_flush(struct sd_softc *, int);
129 static int	sd_getcache(struct sd_softc *, int *);
130 static int	sd_setcache(struct sd_softc *, int);
131 
132 static int	sdmatch(struct device *, struct cfdata *, void *);
133 static void	sdattach(struct device *, struct device *, void *);
134 static int	sdactivate(struct device *, enum devact);
135 static int	sddetach(struct device *, int);
136 static void	sd_set_properties(struct sd_softc *);
137 
138 CFATTACH_DECL(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
139     sdactivate);
140 
141 extern struct cfdriver sd_cd;
142 
143 static const struct scsipi_inquiry_pattern sd_patterns[] = {
144 	{T_DIRECT, T_FIXED,
145 	 "",         "",                 ""},
146 	{T_DIRECT, T_REMOV,
147 	 "",         "",                 ""},
148 	{T_OPTICAL, T_FIXED,
149 	 "",         "",                 ""},
150 	{T_OPTICAL, T_REMOV,
151 	 "",         "",                 ""},
152 	{T_SIMPLE_DIRECT, T_FIXED,
153 	 "",         "",                 ""},
154 	{T_SIMPLE_DIRECT, T_REMOV,
155 	 "",         "",                 ""},
156 };
157 
158 static dev_type_open(sdopen);
159 static dev_type_close(sdclose);
160 static dev_type_read(sdread);
161 static dev_type_write(sdwrite);
162 static dev_type_ioctl(sdioctl);
163 static dev_type_strategy(sdstrategy);
164 static dev_type_dump(sddump);
165 static dev_type_size(sdsize);
166 
167 const struct bdevsw sd_bdevsw = {
168 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
169 };
170 
171 const struct cdevsw sd_cdevsw = {
172 	sdopen, sdclose, sdread, sdwrite, sdioctl,
173 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
174 };
175 
176 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
177 
178 static const struct scsipi_periphsw sd_switch = {
179 	sd_interpret_sense,	/* check our error handler first */
180 	sdstart,		/* have a queue, served by this */
181 	NULL,			/* have no async handler */
182 	sddone,			/* deal with stats at interrupt time */
183 };
184 
185 struct sd_mode_sense_data {
186 	/*
187 	 * XXX
188 	 * We are not going to parse this as-is -- it just has to be large
189 	 * enough.
190 	 */
191 	union {
192 		struct scsi_mode_parameter_header_6 small;
193 		struct scsi_mode_parameter_header_10 big;
194 	} header;
195 	struct scsi_general_block_descriptor blk_desc;
196 	union scsi_disk_pages pages;
197 };
198 
199 /*
200  * The routine called by the low level scsi routine when it discovers
201  * A device suitable for this driver
202  */
203 static int
204 sdmatch(struct device *parent, struct cfdata *match,
205     void *aux)
206 {
207 	struct scsipibus_attach_args *sa = aux;
208 	int priority;
209 
210 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
211 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
212 	    sizeof(sd_patterns[0]), &priority);
213 
214 	return (priority);
215 }
216 
217 /*
218  * Attach routine common to atapi & scsi.
219  */
220 static void
221 sdattach(struct device *parent, struct device *self, void *aux)
222 {
223 	struct sd_softc *sd = device_private(self);
224 	struct scsipibus_attach_args *sa = aux;
225 	struct scsipi_periph *periph = sa->sa_periph;
226 	int error, result;
227 	struct disk_parms *dp = &sd->params;
228 	char pbuf[9];
229 
230 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
231 
232 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
233 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
234 	if (sd->type == T_SIMPLE_DIRECT)
235 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
236 
237 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
238 	    periph->periph_version == 0)
239 		sd->flags |= SDF_ANCIENT;
240 
241 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
242 
243 	callout_init(&sd->sc_callout, 0);
244 
245 	/*
246 	 * Store information needed to contact our base driver
247 	 */
248 	sd->sc_periph = periph;
249 
250 	periph->periph_dev = &sd->sc_dev;
251 	periph->periph_switch = &sd_switch;
252 
253         /*
254          * Increase our openings to the maximum-per-periph
255          * supported by the adapter.  This will either be
256          * clamped down or grown by the adapter if necessary.
257          */
258 	periph->periph_openings =
259 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
260 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
261 
262 	/*
263 	 * Initialize and attach the disk structure.
264 	 */
265 	disk_init(&sd->sc_dk, sd->sc_dev.dv_xname, &sddkdriver);
266 	disk_attach(&sd->sc_dk);
267 
268 	/*
269 	 * Use the subdriver to request information regarding the drive.
270 	 */
271 	aprint_naive("\n");
272 	aprint_normal("\n");
273 
274 	error = scsipi_test_unit_ready(periph,
275 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
276 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
277 
278 	if (error)
279 		result = SDGP_RESULT_OFFLINE;
280 	else
281 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
282 	aprint_normal("%s: ", sd->sc_dev.dv_xname);
283 	switch (result) {
284 	case SDGP_RESULT_OK:
285 		format_bytes(pbuf, sizeof(pbuf),
286 		    (u_int64_t)dp->disksize * dp->blksize);
287 	        aprint_normal(
288 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
289 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
290 		    (unsigned long long)dp->disksize);
291 		break;
292 
293 	case SDGP_RESULT_OFFLINE:
294 		aprint_normal("drive offline");
295 		break;
296 
297 	case SDGP_RESULT_UNFORMATTED:
298 		aprint_normal("unformatted media");
299 		break;
300 
301 #ifdef DIAGNOSTIC
302 	default:
303 		panic("sdattach: unknown result from get_parms");
304 		break;
305 #endif
306 	}
307 	aprint_normal("\n");
308 
309 	/*
310 	 * Establish a shutdown hook so that we can ensure that
311 	 * our data has actually made it onto the platter at
312 	 * shutdown time.  Note that this relies on the fact
313 	 * that the shutdown hook code puts us at the head of
314 	 * the list (thus guaranteeing that our hook runs before
315 	 * our ancestors').
316 	 */
317 	if ((sd->sc_sdhook =
318 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
319 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
320 		    sd->sc_dev.dv_xname);
321 
322 #if NRND > 0
323 	/*
324 	 * attach the device into the random source list
325 	 */
326 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
327 			  RND_TYPE_DISK, 0);
328 #endif
329 
330 	/* Discover wedges on this disk. */
331 	dkwedge_discover(&sd->sc_dk);
332 
333 	sd_set_properties(sd);
334 }
335 
336 static int
337 sdactivate(struct device *self, enum devact act)
338 {
339 	int rv = 0;
340 
341 	switch (act) {
342 	case DVACT_ACTIVATE:
343 		rv = EOPNOTSUPP;
344 		break;
345 
346 	case DVACT_DEACTIVATE:
347 		/*
348 		 * Nothing to do; we key off the device's DVF_ACTIVE.
349 		 */
350 		break;
351 	}
352 	return (rv);
353 }
354 
355 static int
356 sddetach(struct device *self, int flags)
357 {
358 	struct sd_softc *sd = device_private(self);
359 	int s, bmaj, cmaj, i, mn;
360 
361 	/* locate the major number */
362 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
363 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
364 
365 	/* Nuke the vnodes for any open instances */
366 	for (i = 0; i < MAXPARTITIONS; i++) {
367 		mn = SDMINOR(device_unit(self), i);
368 		vdevgone(bmaj, mn, mn, VBLK);
369 		vdevgone(cmaj, mn, mn, VCHR);
370 	}
371 
372 	/* kill any pending restart */
373 	callout_stop(&sd->sc_callout);
374 
375 	/* Delete all of our wedges. */
376 	dkwedge_delall(&sd->sc_dk);
377 
378 	s = splbio();
379 
380 	/* Kill off any queued buffers. */
381 	bufq_drain(sd->buf_queue);
382 
383 	bufq_free(sd->buf_queue);
384 
385 	/* Kill off any pending commands. */
386 	scsipi_kill_pending(sd->sc_periph);
387 
388 	splx(s);
389 
390 	/* Detach from the disk list. */
391 	disk_detach(&sd->sc_dk);
392 	disk_destroy(&sd->sc_dk);
393 
394 	/* Get rid of the shutdown hook. */
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 	if (unit >= sd_cd.cd_ndevs)
419 		return (ENXIO);
420 	sd = sd_cd.cd_devs[unit];
421 	if (sd == NULL)
422 		return (ENXIO);
423 
424 	if (!device_is_active(&sd->sc_dev))
425 		return (ENODEV);
426 
427 	part = SDPART(dev);
428 
429 	mutex_enter(&sd->sc_dk.dk_openlock);
430 
431 	/*
432 	 * If there are wedges, and this is not RAW_PART, then we
433 	 * need to fail.
434 	 */
435 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
436 		error = EBUSY;
437 		goto bad1;
438 	}
439 
440 	periph = sd->sc_periph;
441 	adapt = periph->periph_channel->chan_adapter;
442 
443 	SC_DEBUG(periph, SCSIPI_DB1,
444 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
445 	    sd_cd.cd_ndevs, part));
446 
447 	/*
448 	 * If this is the first open of this device, add a reference
449 	 * to the adapter.
450 	 */
451 	if (sd->sc_dk.dk_openmask == 0 &&
452 	    (error = scsipi_adapter_addref(adapt)) != 0)
453 		goto bad1;
454 
455 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
456 		/*
457 		 * If any partition is open, but the disk has been invalidated,
458 		 * disallow further opens of non-raw partition
459 		 */
460 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
461 		    (part != RAW_PART || fmt != S_IFCHR)) {
462 			error = EIO;
463 			goto bad2;
464 		}
465 	} else {
466 		int silent;
467 
468 		if (part == RAW_PART && fmt == S_IFCHR)
469 			silent = XS_CTL_SILENT;
470 		else
471 			silent = 0;
472 
473 		/* Check that it is still responding and ok. */
474 		error = scsipi_test_unit_ready(periph,
475 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
476 		    silent);
477 
478 		/*
479 		 * Start the pack spinning if necessary. Always allow the
480 		 * raw parition to be opened, for raw IOCTLs. Data transfers
481 		 * will check for SDEV_MEDIA_LOADED.
482 		 */
483 		if (error == EIO) {
484 			int error2;
485 
486 			error2 = scsipi_start(periph, SSS_START, silent);
487 			switch (error2) {
488 			case 0:
489 				error = 0;
490 				break;
491 			case EIO:
492 			case EINVAL:
493 				break;
494 			default:
495 				error = error2;
496 				break;
497 			}
498 		}
499 		if (error) {
500 			if (silent)
501 				goto out;
502 			goto bad2;
503 		}
504 
505 		periph->periph_flags |= PERIPH_OPEN;
506 
507 		if (periph->periph_flags & PERIPH_REMOVABLE) {
508 			/* Lock the pack in. */
509 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
510 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
511 			    XS_CTL_IGNORE_MEDIA_CHANGE |
512 			    XS_CTL_SILENT);
513 			if (error)
514 				goto bad3;
515 		}
516 
517 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
518 			int param_error;
519 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
520 
521 			/*
522 			 * Load the physical device parameters.
523 			 *
524 			 * Note that if media is present but unformatted,
525 			 * we allow the open (so that it can be formatted!).
526 			 * The drive should refuse real I/O, if the media is
527 			 * unformatted.
528 			 */
529 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
530 			     == SDGP_RESULT_OFFLINE) {
531 				error = ENXIO;
532 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
533 				goto bad3;
534 			}
535 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
536 
537 			/* Load the partition info if not already loaded. */
538 			if (param_error == 0) {
539 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
540 					error = EIO;
541 					goto bad3;
542 				}
543 				SC_DEBUG(periph, SCSIPI_DB3,
544 				     ("Disklabel loaded "));
545 			}
546 		}
547 	}
548 
549 	/* Check that the partition exists. */
550 	if (part != RAW_PART &&
551 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
552 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
553 		error = ENXIO;
554 		goto bad3;
555 	}
556 
557  out:	/* Insure only one open at a time. */
558 	switch (fmt) {
559 	case S_IFCHR:
560 		sd->sc_dk.dk_copenmask |= (1 << part);
561 		break;
562 	case S_IFBLK:
563 		sd->sc_dk.dk_bopenmask |= (1 << part);
564 		break;
565 	}
566 	sd->sc_dk.dk_openmask =
567 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
568 
569 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
570 	mutex_exit(&sd->sc_dk.dk_openlock);
571 	return (0);
572 
573  bad3:
574 	if (sd->sc_dk.dk_openmask == 0) {
575 		if (periph->periph_flags & PERIPH_REMOVABLE)
576 			scsipi_prevent(periph, SPAMR_ALLOW,
577 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
578 			    XS_CTL_IGNORE_MEDIA_CHANGE |
579 			    XS_CTL_SILENT);
580 		periph->periph_flags &= ~PERIPH_OPEN;
581 	}
582 
583  bad2:
584 	if (sd->sc_dk.dk_openmask == 0)
585 		scsipi_adapter_delref(adapt);
586 
587  bad1:
588 	mutex_exit(&sd->sc_dk.dk_openlock);
589 	return (error);
590 }
591 
592 /*
593  * close the device.. only called if we are the LAST occurence of an open
594  * device.  Convenient now but usually a pain.
595  */
596 static int
597 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
598 {
599 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
600 	struct scsipi_periph *periph = sd->sc_periph;
601 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
602 	int part = SDPART(dev);
603 
604 	mutex_enter(&sd->sc_dk.dk_openlock);
605 	switch (fmt) {
606 	case S_IFCHR:
607 		sd->sc_dk.dk_copenmask &= ~(1 << part);
608 		break;
609 	case S_IFBLK:
610 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
611 		break;
612 	}
613 	sd->sc_dk.dk_openmask =
614 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
615 
616 	if (sd->sc_dk.dk_openmask == 0) {
617 		/*
618 		 * If the disk cache needs flushing, and the disk supports
619 		 * it, do it now.
620 		 */
621 		if ((sd->flags & SDF_DIRTY) != 0) {
622 			if (sd_flush(sd, 0)) {
623 				printf("%s: cache synchronization failed\n",
624 				    sd->sc_dev.dv_xname);
625 				sd->flags &= ~SDF_FLUSHING;
626 			} else
627 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
628 		}
629 
630 		scsipi_wait_drain(periph);
631 
632 		if (periph->periph_flags & PERIPH_REMOVABLE)
633 			scsipi_prevent(periph, SPAMR_ALLOW,
634 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
635 			    XS_CTL_IGNORE_NOT_READY |
636 			    XS_CTL_SILENT);
637 		periph->periph_flags &= ~PERIPH_OPEN;
638 
639 		scsipi_wait_drain(periph);
640 
641 		scsipi_adapter_delref(adapt);
642 	}
643 
644 	mutex_exit(&sd->sc_dk.dk_openlock);
645 	return (0);
646 }
647 
648 /*
649  * Actually translate the requested transfer into one the physical driver
650  * can understand.  The transfer is described by a buf and will include
651  * only one physical transfer.
652  */
653 static void
654 sdstrategy(struct buf *bp)
655 {
656 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
657 	struct scsipi_periph *periph = sd->sc_periph;
658 	struct disklabel *lp;
659 	daddr_t blkno;
660 	int s;
661 	bool sector_aligned;
662 
663 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
664 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
665 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
666 	/*
667 	 * If the device has been made invalid, error out
668 	 */
669 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
670 	    !device_is_active(&sd->sc_dev)) {
671 		if (periph->periph_flags & PERIPH_OPEN)
672 			bp->b_error = EIO;
673 		else
674 			bp->b_error = ENODEV;
675 		goto done;
676 	}
677 
678 	lp = sd->sc_dk.dk_label;
679 
680 	/*
681 	 * The transfer must be a whole number of blocks, offset must not be
682 	 * negative.
683 	 */
684 	if (lp->d_secsize == DEV_BSIZE) {
685 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
686 	} else {
687 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
688 	}
689 	if (!sector_aligned || bp->b_blkno < 0) {
690 		bp->b_error = EINVAL;
691 		goto done;
692 	}
693 	/*
694 	 * If it's a null transfer, return immediatly
695 	 */
696 	if (bp->b_bcount == 0)
697 		goto done;
698 
699 	/*
700 	 * Do bounds checking, adjust transfer. if error, process.
701 	 * If end of partition, just return.
702 	 */
703 	if (SDPART(bp->b_dev) == RAW_PART) {
704 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
705 		    sd->params.disksize512) <= 0)
706 			goto done;
707 	} else {
708 		if (bounds_check_with_label(&sd->sc_dk, bp,
709 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
710 			goto done;
711 	}
712 
713 	/*
714 	 * Now convert the block number to absolute and put it in
715 	 * terms of the device's logical block size.
716 	 */
717 	if (lp->d_secsize == DEV_BSIZE)
718 		blkno = bp->b_blkno;
719 	else if (lp->d_secsize > DEV_BSIZE)
720 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
721 	else
722 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
723 
724 	if (SDPART(bp->b_dev) != RAW_PART)
725 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
726 
727 	bp->b_rawblkno = blkno;
728 
729 	s = splbio();
730 
731 	/*
732 	 * Place it in the queue of disk activities for this disk.
733 	 *
734 	 * XXX Only do disksort() if the current operating mode does not
735 	 * XXX include tagged queueing.
736 	 */
737 	BUFQ_PUT(sd->buf_queue, bp);
738 
739 	/*
740 	 * Tell the device to get going on the transfer if it's
741 	 * not doing anything, otherwise just wait for completion
742 	 */
743 	sdstart(sd->sc_periph);
744 
745 	splx(s);
746 	return;
747 
748 done:
749 	/*
750 	 * Correctly set the buf to indicate a completed xfer
751 	 */
752 	bp->b_resid = bp->b_bcount;
753 	biodone(bp);
754 }
755 
756 /*
757  * sdstart looks to see if there is a buf waiting for the device
758  * and that the device is not already busy. If both are true,
759  * It dequeues the buf and creates a scsi command to perform the
760  * transfer in the buf. The transfer request will call scsipi_done
761  * on completion, which will in turn call this routine again
762  * so that the next queued transfer is performed.
763  * The bufs are queued by the strategy routine (sdstrategy)
764  *
765  * This routine is also called after other non-queued requests
766  * have been made of the scsi driver, to ensure that the queue
767  * continues to be drained.
768  *
769  * must be called at the correct (highish) spl level
770  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
771  */
772 static void
773 sdstart(struct scsipi_periph *periph)
774 {
775 	struct sd_softc *sd = (void *)periph->periph_dev;
776 	struct disklabel *lp = sd->sc_dk.dk_label;
777 	struct buf *bp = 0;
778 	struct scsipi_rw_16 cmd16;
779 	struct scsipi_rw_10 cmd_big;
780 	struct scsi_rw_6 cmd_small;
781 	struct scsipi_generic *cmdp;
782 	struct scsipi_xfer *xs;
783 	int nblks, cmdlen, error, flags;
784 
785 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
786 	/*
787 	 * Check if the device has room for another command
788 	 */
789 	while (periph->periph_active < periph->periph_openings) {
790 		/*
791 		 * there is excess capacity, but a special waits
792 		 * It'll need the adapter as soon as we clear out of the
793 		 * way and let it run (user level wait).
794 		 */
795 		if (periph->periph_flags & PERIPH_WAITING) {
796 			periph->periph_flags &= ~PERIPH_WAITING;
797 			wakeup((void *)periph);
798 			return;
799 		}
800 
801 		/*
802 		 * If the device has become invalid, abort all the
803 		 * reads and writes until all files have been closed and
804 		 * re-opened
805 		 */
806 		if (__predict_false(
807 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
808 			if ((bp = BUFQ_GET(sd->buf_queue)) != NULL) {
809 				bp->b_error = EIO;
810 				bp->b_resid = bp->b_bcount;
811 				biodone(bp);
812 				continue;
813 			} else {
814 				return;
815 			}
816 		}
817 
818 		/*
819 		 * See if there is a buf with work for us to do..
820 		 */
821 		if ((bp = BUFQ_PEEK(sd->buf_queue)) == NULL)
822 			return;
823 
824 		/*
825 		 * We have a buf, now we should make a command.
826 		 */
827 
828 		if (lp->d_secsize == DEV_BSIZE)
829 			nblks = bp->b_bcount >> DEV_BSHIFT;
830 		else
831 			nblks = howmany(bp->b_bcount, lp->d_secsize);
832 
833 		/*
834 		 * Fill out the scsi command.  Use the smallest CDB possible
835 		 * (6-byte, 10-byte, or 16-byte).
836 		 */
837 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
838 		    ((nblks & 0xff) == nblks) &&
839 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
840 			/* 6-byte CDB */
841 			memset(&cmd_small, 0, sizeof(cmd_small));
842 			cmd_small.opcode = (bp->b_flags & B_READ) ?
843 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
844 			_lto3b(bp->b_rawblkno, cmd_small.addr);
845 			cmd_small.length = nblks & 0xff;
846 			cmdlen = sizeof(cmd_small);
847 			cmdp = (struct scsipi_generic *)&cmd_small;
848 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
849 			/* 10-byte CDB */
850 			memset(&cmd_big, 0, sizeof(cmd_big));
851 			cmd_big.opcode = (bp->b_flags & B_READ) ?
852 			    READ_10 : WRITE_10;
853 			_lto4b(bp->b_rawblkno, cmd_big.addr);
854 			_lto2b(nblks, cmd_big.length);
855 			cmdlen = sizeof(cmd_big);
856 			cmdp = (struct scsipi_generic *)&cmd_big;
857 		} else {
858 			/* 16-byte CDB */
859 			memset(&cmd16, 0, sizeof(cmd16));
860 			cmd16.opcode = (bp->b_flags & B_READ) ?
861 			    READ_16 : WRITE_16;
862 			_lto8b(bp->b_rawblkno, cmd16.addr);
863 			_lto4b(nblks, cmd16.length);
864 			cmdlen = sizeof(cmd16);
865 			cmdp = (struct scsipi_generic *)&cmd16;
866 		}
867 
868 		/* Instrumentation. */
869 		disk_busy(&sd->sc_dk);
870 
871 		/*
872 		 * Mark the disk dirty so that the cache will be
873 		 * flushed on close.
874 		 */
875 		if ((bp->b_flags & B_READ) == 0)
876 			sd->flags |= SDF_DIRTY;
877 
878 		/*
879 		 * Figure out what flags to use.
880 		 */
881 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
882 		if (bp->b_flags & B_READ)
883 			flags |= XS_CTL_DATA_IN;
884 		else
885 			flags |= XS_CTL_DATA_OUT;
886 
887 		/*
888 		 * Call the routine that chats with the adapter.
889 		 * Note: we cannot sleep as we may be an interrupt
890 		 */
891 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
892 		    (u_char *)bp->b_data, bp->b_bcount,
893 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
894 		if (__predict_false(xs == NULL)) {
895 			/*
896 			 * out of memory. Keep this buffer in the queue, and
897 			 * retry later.
898 			 */
899 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
900 			    periph);
901 			return;
902 		}
903 		/*
904 		 * need to dequeue the buffer before queuing the command,
905 		 * because cdstart may be called recursively from the
906 		 * HBA driver
907 		 */
908 #ifdef DIAGNOSTIC
909 		if (BUFQ_GET(sd->buf_queue) != bp)
910 			panic("sdstart(): dequeued wrong buf");
911 #else
912 		BUFQ_GET(sd->buf_queue);
913 #endif
914 		error = scsipi_execute_xs(xs);
915 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
916 		KASSERT(error == 0);
917 	}
918 }
919 
920 static void
921 sdrestart(void *v)
922 {
923 	int s = splbio();
924 	sdstart((struct scsipi_periph *)v);
925 	splx(s);
926 }
927 
928 static void
929 sddone(struct scsipi_xfer *xs, int error)
930 {
931 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
932 	struct buf *bp = xs->bp;
933 
934 	if (sd->flags & SDF_FLUSHING) {
935 		/* Flush completed, no longer dirty. */
936 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
937 	}
938 
939 	if (bp) {
940 		bp->b_error = error;
941 		bp->b_resid = xs->resid;
942 		if (error) {
943 			/* on a read/write error bp->b_resid is zero, so fix */
944 			bp->b_resid = bp->b_bcount;
945 		}
946 
947 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
948 		    (bp->b_flags & B_READ));
949 #if NRND > 0
950 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
951 #endif
952 
953 		biodone(bp);
954 	}
955 }
956 
957 static void
958 sdminphys(struct buf *bp)
959 {
960 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
961 	long xmax;
962 
963 	/*
964 	 * If the device is ancient, we want to make sure that
965 	 * the transfer fits into a 6-byte cdb.
966 	 *
967 	 * XXX Note that the SCSI-I spec says that 256-block transfers
968 	 * are allowed in a 6-byte read/write, and are specified
969 	 * by settng the "length" to 0.  However, we're conservative
970 	 * here, allowing only 255-block transfers in case an
971 	 * ancient device gets confused by length == 0.  A length of 0
972 	 * in a 10-byte read/write actually means 0 blocks.
973 	 */
974 	if ((sd->flags & SDF_ANCIENT) &&
975 	    ((sd->sc_periph->periph_flags &
976 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
977 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
978 
979 		if (bp->b_bcount > xmax)
980 			bp->b_bcount = xmax;
981 	}
982 
983 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
984 }
985 
986 static int
987 sdread(dev_t dev, struct uio *uio, int ioflag)
988 {
989 
990 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
991 }
992 
993 static int
994 sdwrite(dev_t dev, struct uio *uio, int ioflag)
995 {
996 
997 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
998 }
999 
1000 /*
1001  * Perform special action on behalf of the user
1002  * Knows about the internals of this device
1003  */
1004 static int
1005 sdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1006 {
1007 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
1008 	struct scsipi_periph *periph = sd->sc_periph;
1009 	int part = SDPART(dev);
1010 	int error = 0;
1011 #ifdef __HAVE_OLD_DISKLABEL
1012 	struct disklabel *newlabel = NULL;
1013 #endif
1014 
1015 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1016 
1017 	/*
1018 	 * If the device is not valid, some IOCTLs can still be
1019 	 * handled on the raw partition. Check this here.
1020 	 */
1021 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1022 		switch (cmd) {
1023 		case DIOCKLABEL:
1024 		case DIOCWLABEL:
1025 		case DIOCLOCK:
1026 		case DIOCEJECT:
1027 		case ODIOCEJECT:
1028 		case DIOCGCACHE:
1029 		case DIOCSCACHE:
1030 		case SCIOCIDENTIFY:
1031 		case OSCIOCIDENTIFY:
1032 		case SCIOCCOMMAND:
1033 		case SCIOCDEBUG:
1034 			if (part == RAW_PART)
1035 				break;
1036 		/* FALLTHROUGH */
1037 		default:
1038 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1039 				return (ENODEV);
1040 			else
1041 				return (EIO);
1042 		}
1043 	}
1044 
1045 	switch (cmd) {
1046 	case DIOCGDINFO:
1047 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1048 		return (0);
1049 
1050 #ifdef __HAVE_OLD_DISKLABEL
1051 	case ODIOCGDINFO:
1052 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1053 		if (newlabel == NULL)
1054 			return EIO;
1055 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1056 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1057 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1058 		else
1059 			error = ENOTTY;
1060 		free(newlabel, M_TEMP);
1061 		return error;
1062 #endif
1063 
1064 	case DIOCGPART:
1065 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1066 		((struct partinfo *)addr)->part =
1067 		    &sd->sc_dk.dk_label->d_partitions[part];
1068 		return (0);
1069 
1070 	case DIOCWDINFO:
1071 	case DIOCSDINFO:
1072 #ifdef __HAVE_OLD_DISKLABEL
1073 	case ODIOCWDINFO:
1074 	case ODIOCSDINFO:
1075 #endif
1076 	{
1077 		struct disklabel *lp;
1078 
1079 		if ((flag & FWRITE) == 0)
1080 			return (EBADF);
1081 
1082 #ifdef __HAVE_OLD_DISKLABEL
1083  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1084 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1085 			if (newlabel == NULL)
1086 				return EIO;
1087 			memset(newlabel, 0, sizeof newlabel);
1088 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1089 			lp = newlabel;
1090 		} else
1091 #endif
1092 		lp = (struct disklabel *)addr;
1093 
1094 		mutex_enter(&sd->sc_dk.dk_openlock);
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 		mutex_exit(&sd->sc_dk.dk_openlock);
1113 #ifdef __HAVE_OLD_DISKLABEL
1114 		if (newlabel != NULL)
1115 			free(newlabel, M_TEMP);
1116 #endif
1117 		return (error);
1118 	}
1119 
1120 	case DIOCKLABEL:
1121 		if (*(int *)addr)
1122 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1123 		else
1124 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1125 		return (0);
1126 
1127 	case DIOCWLABEL:
1128 		if ((flag & FWRITE) == 0)
1129 			return (EBADF);
1130 		if (*(int *)addr)
1131 			sd->flags |= SDF_WLABEL;
1132 		else
1133 			sd->flags &= ~SDF_WLABEL;
1134 		return (0);
1135 
1136 	case DIOCLOCK:
1137 		if (periph->periph_flags & PERIPH_REMOVABLE)
1138 			return (scsipi_prevent(periph,
1139 			    (*(int *)addr) ?
1140 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1141 		else
1142 			return (ENOTTY);
1143 
1144 	case DIOCEJECT:
1145 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1146 			return (ENOTTY);
1147 		if (*(int *)addr == 0) {
1148 			/*
1149 			 * Don't force eject: check that we are the only
1150 			 * partition open. If so, unlock it.
1151 			 */
1152 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1153 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1154 			    sd->sc_dk.dk_openmask) {
1155 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1156 				    XS_CTL_IGNORE_NOT_READY);
1157 				if (error)
1158 					return (error);
1159 			} else {
1160 				return (EBUSY);
1161 			}
1162 		}
1163 		/* FALLTHROUGH */
1164 	case ODIOCEJECT:
1165 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1166 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1167 
1168 	case DIOCGDEFLABEL:
1169 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1170 		return (0);
1171 
1172 #ifdef __HAVE_OLD_DISKLABEL
1173 	case ODIOCGDEFLABEL:
1174 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1175 		if (newlabel == NULL)
1176 			return EIO;
1177 		sdgetdefaultlabel(sd, newlabel);
1178 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1179 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1180 		else
1181 			error = ENOTTY;
1182 		free(newlabel, M_TEMP);
1183 		return error;
1184 #endif
1185 
1186 	case DIOCGCACHE:
1187 		return (sd_getcache(sd, (int *) addr));
1188 
1189 	case DIOCSCACHE:
1190 		if ((flag & FWRITE) == 0)
1191 			return (EBADF);
1192 		return (sd_setcache(sd, *(int *) addr));
1193 
1194 	case DIOCCACHESYNC:
1195 		/*
1196 		 * XXX Do we really need to care about having a writable
1197 		 * file descriptor here?
1198 		 */
1199 		if ((flag & FWRITE) == 0)
1200 			return (EBADF);
1201 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1202 			error = sd_flush(sd, 0);
1203 			if (error)
1204 				sd->flags &= ~SDF_FLUSHING;
1205 			else
1206 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1207 		} else
1208 			error = 0;
1209 		return (error);
1210 
1211 	case DIOCAWEDGE:
1212 	    {
1213 	    	struct dkwedge_info *dkw = (void *) addr;
1214 
1215 		if ((flag & FWRITE) == 0)
1216 			return (EBADF);
1217 
1218 		/* If the ioctl happens here, the parent is us. */
1219 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
1220 		return (dkwedge_add(dkw));
1221 	    }
1222 
1223 	case DIOCDWEDGE:
1224 	    {
1225 	    	struct dkwedge_info *dkw = (void *) addr;
1226 
1227 		if ((flag & FWRITE) == 0)
1228 			return (EBADF);
1229 
1230 		/* If the ioctl happens here, the parent is us. */
1231 		strcpy(dkw->dkw_parent, sd->sc_dev.dv_xname);
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 		printf("%s: %s\n", sd->sc_dev.dv_xname, 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 			printf("%s: cache synchronization failed\n",
1340 			    sd->sc_dev.dv_xname);
1341 			sd->flags &= ~SDF_FLUSHING;
1342 		} else
1343 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1344 	}
1345 }
1346 
1347 /*
1348  * Check Errors
1349  */
1350 static int
1351 sd_interpret_sense(struct scsipi_xfer *xs)
1352 {
1353 	struct scsipi_periph *periph = xs->xs_periph;
1354 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1355 	struct sd_softc *sd = (void *)periph->periph_dev;
1356 	int s, error, retval = EJUSTRETURN;
1357 
1358 	/*
1359 	 * If the periph is already recovering, just do the normal
1360 	 * error processing.
1361 	 */
1362 	if (periph->periph_flags & PERIPH_RECOVERING)
1363 		return (retval);
1364 
1365 	/*
1366 	 * Ignore errors from accessing illegal fields (e.g. trying to
1367 	 * lock the door of a digicam, which doesn't have a door that
1368 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1369 	 */
1370 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1371 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1372 	    sense->asc == 0x24 &&
1373 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1374 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1375 			scsipi_printaddr(periph);
1376 			printf("no door lock\n");
1377 		}
1378 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1379 		return (retval);
1380 	}
1381 
1382 
1383 
1384 	/*
1385 	 * If the device is not open yet, let the generic code handle it.
1386 	 */
1387 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1388 		return (retval);
1389 
1390 	/*
1391 	 * If it isn't a extended or extended/deferred error, let
1392 	 * the generic code handle it.
1393 	 */
1394 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1395 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1396 		return (retval);
1397 
1398 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1399 	    sense->asc == 0x4) {
1400 		if (sense->ascq == 0x01)	{
1401 			/*
1402 			 * Unit In The Process Of Becoming Ready.
1403 			 */
1404 			printf("%s: waiting for pack to spin up...\n",
1405 			    sd->sc_dev.dv_xname);
1406 			if (!callout_pending(&periph->periph_callout))
1407 				scsipi_periph_freeze(periph, 1);
1408 			callout_reset(&periph->periph_callout,
1409 			    5 * hz, scsipi_periph_timed_thaw, periph);
1410 			retval = ERESTART;
1411 		} else if (sense->ascq == 0x02) {
1412 			printf("%s: pack is stopped, restarting...\n",
1413 			    sd->sc_dev.dv_xname);
1414 			s = splbio();
1415 			periph->periph_flags |= PERIPH_RECOVERING;
1416 			splx(s);
1417 			error = scsipi_start(periph, SSS_START,
1418 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1419 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1420 			if (error) {
1421 				printf("%s: unable to restart pack\n",
1422 				    sd->sc_dev.dv_xname);
1423 				retval = error;
1424 			} else
1425 				retval = ERESTART;
1426 			s = splbio();
1427 			periph->periph_flags &= ~PERIPH_RECOVERING;
1428 			splx(s);
1429 		}
1430 	}
1431 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1432 	    sense->asc == 0x31 &&
1433 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1434 		/* Medium Format Corrupted */
1435 		retval = EFTYPE;
1436 	}
1437 	return (retval);
1438 }
1439 
1440 
1441 static int
1442 sdsize(dev_t dev)
1443 {
1444 	struct sd_softc *sd;
1445 	int part, unit, omask;
1446 	int size;
1447 
1448 	unit = SDUNIT(dev);
1449 	if (unit >= sd_cd.cd_ndevs)
1450 		return (-1);
1451 	sd = sd_cd.cd_devs[unit];
1452 	if (sd == NULL)
1453 		return (-1);
1454 
1455 	if (!device_is_active(&sd->sc_dev))
1456 		return (-1);
1457 
1458 	part = SDPART(dev);
1459 	omask = sd->sc_dk.dk_openmask & (1 << part);
1460 
1461 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1462 		return (-1);
1463 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1464 		size = -1;
1465 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1466 		size = -1;
1467 	else
1468 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1469 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1470 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1471 		return (-1);
1472 	return (size);
1473 }
1474 
1475 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1476 static struct scsipi_xfer sx;
1477 static int sddoingadump;
1478 
1479 /*
1480  * dump all of physical memory into the partition specified, starting
1481  * at offset 'dumplo' into the partition.
1482  */
1483 static int
1484 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1485 {
1486 	struct sd_softc *sd;	/* disk unit to do the I/O */
1487 	struct disklabel *lp;	/* disk's disklabel */
1488 	int	unit, part;
1489 	int	sectorsize;	/* size of a disk sector */
1490 	int	nsects;		/* number of sectors in partition */
1491 	int	sectoff;	/* sector offset of partition */
1492 	int	totwrt;		/* total number of sectors left to write */
1493 	int	nwrt;		/* current number of sectors to write */
1494 	struct scsipi_rw_10 cmd;	/* write command */
1495 	struct scsipi_xfer *xs;	/* ... convenience */
1496 	struct scsipi_periph *periph;
1497 	struct scsipi_channel *chan;
1498 
1499 	/* Check if recursive dump; if so, punt. */
1500 	if (sddoingadump)
1501 		return (EFAULT);
1502 
1503 	/* Mark as active early. */
1504 	sddoingadump = 1;
1505 
1506 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1507 	part = SDPART(dev);
1508 
1509 	/* Check for acceptable drive number. */
1510 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1511 		return (ENXIO);
1512 
1513 	if (!device_is_active(&sd->sc_dev))
1514 		return (ENODEV);
1515 
1516 	periph = sd->sc_periph;
1517 	chan = periph->periph_channel;
1518 
1519 	/* Make sure it was initialized. */
1520 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1521 		return (ENXIO);
1522 
1523 	/* Convert to disk sectors.  Request must be a multiple of size. */
1524 	lp = sd->sc_dk.dk_label;
1525 	sectorsize = lp->d_secsize;
1526 	if ((size % sectorsize) != 0)
1527 		return (EFAULT);
1528 	totwrt = size / sectorsize;
1529 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1530 
1531 	nsects = lp->d_partitions[part].p_size;
1532 	sectoff = lp->d_partitions[part].p_offset;
1533 
1534 	/* Check transfer bounds against partition size. */
1535 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1536 		return (EINVAL);
1537 
1538 	/* Offset block number to start of partition. */
1539 	blkno += sectoff;
1540 
1541 	xs = &sx;
1542 
1543 	while (totwrt > 0) {
1544 		nwrt = totwrt;		/* XXX */
1545 #ifndef	SD_DUMP_NOT_TRUSTED
1546 		/*
1547 		 *  Fill out the scsi command
1548 		 */
1549 		memset(&cmd, 0, sizeof(cmd));
1550 		cmd.opcode = WRITE_10;
1551 		_lto4b(blkno, cmd.addr);
1552 		_lto2b(nwrt, cmd.length);
1553 		/*
1554 		 * Fill out the scsipi_xfer structure
1555 		 *    Note: we cannot sleep as we may be an interrupt
1556 		 * don't use scsipi_command() as it may want to wait
1557 		 * for an xs.
1558 		 */
1559 		memset(xs, 0, sizeof(sx));
1560 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1561 		    XS_CTL_DATA_OUT;
1562 		xs->xs_status = 0;
1563 		xs->xs_periph = periph;
1564 		xs->xs_retries = SDRETRIES;
1565 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1566 		xs->cmd = (struct scsipi_generic *)&cmd;
1567 		xs->cmdlen = sizeof(cmd);
1568 		xs->resid = nwrt * sectorsize;
1569 		xs->error = XS_NOERROR;
1570 		xs->bp = 0;
1571 		xs->data = va;
1572 		xs->datalen = nwrt * sectorsize;
1573 		callout_init(&xs->xs_callout, 0);
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 = (char *)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 /*
1639  * sd_validate_blksize:
1640  *
1641  *	Validate the block size.  Print error if periph is specified,
1642  */
1643 static int
1644 sd_validate_blksize(struct scsipi_periph *periph, int len)
1645 {
1646 
1647 	switch (len) {
1648 	case 256:
1649 	case 512:
1650 	case 1024:
1651 	case 2048:
1652 	case 4096:
1653 		return 1;
1654 	}
1655 
1656 	if (periph) {
1657 		scsipi_printaddr(periph);
1658 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1659 		    (len ^ (1 << (ffs(len) - 1))) ?
1660 		    "preposterous" : "unsupported",
1661 		    len, SD_DEFAULT_BLKSIZE);
1662 	}
1663 
1664 	return 0;
1665 }
1666 
1667 /*
1668  * sd_read_capacity:
1669  *
1670  *	Find out from the device what its capacity is.
1671  */
1672 static u_int64_t
1673 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1674 {
1675 	union {
1676 		struct scsipi_read_capacity_10 cmd;
1677 		struct scsipi_read_capacity_16 cmd16;
1678 	} cmd;
1679 	union {
1680 		struct scsipi_read_capacity_10_data data;
1681 		struct scsipi_read_capacity_16_data data16;
1682 	} *datap;
1683 	uint64_t rv;
1684 
1685 	memset(&cmd, 0, sizeof(cmd));
1686 	cmd.cmd.opcode = READ_CAPACITY_10;
1687 
1688 	/*
1689 	 * Don't allocate data buffer on stack;
1690 	 * The lower driver layer might use the same stack and
1691 	 * if it uses region which is in the same cacheline,
1692 	 * cache flush ops against the data buffer won't work properly.
1693 	 */
1694 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1695 	if (datap == NULL)
1696 		return 0;
1697 
1698 	/*
1699 	 * If the command works, interpret the result as a 4 byte
1700 	 * number of blocks
1701 	 */
1702 	rv = 0;
1703 	memset(datap, 0, sizeof(datap->data));
1704 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1705 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1706 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1707 		goto out;
1708 
1709 	if (_4btol(datap->data.addr) != 0xffffffff) {
1710 		*blksize = _4btol(datap->data.length);
1711 		rv = _4btol(datap->data.addr) + 1;
1712 		goto out;
1713 	}
1714 
1715 	/*
1716 	 * Device is larger than can be reflected by READ CAPACITY (10).
1717 	 * Try READ CAPACITY (16).
1718 	 */
1719 
1720 	memset(&cmd, 0, sizeof(cmd));
1721 	cmd.cmd16.opcode = READ_CAPACITY_16;
1722 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1723 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1724 
1725 	memset(datap, 0, sizeof(datap->data16));
1726 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1727 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1728 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1729 		goto out;
1730 
1731 	*blksize = _4btol(datap->data16.length);
1732 	rv = _8btol(datap->data16.addr) + 1;
1733 
1734  out:
1735 	free(datap, M_TEMP);
1736 	return rv;
1737 }
1738 
1739 static int
1740 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1741 {
1742 	struct {
1743 		struct scsi_mode_parameter_header_6 header;
1744 		/* no block descriptor */
1745 		u_int8_t pg_code; /* page code (should be 6) */
1746 		u_int8_t pg_length; /* page length (should be 11) */
1747 		u_int8_t wcd; /* bit0: cache disable */
1748 		u_int8_t lbs[2]; /* logical block size */
1749 		u_int8_t size[5]; /* number of log. blocks */
1750 		u_int8_t pp; /* power/performance */
1751 		u_int8_t flags;
1752 		u_int8_t resvd;
1753 	} scsipi_sense;
1754 	u_int64_t blocks;
1755 	int error, blksize;
1756 
1757 	/*
1758 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1759 	 * give the same information. Do both for now, and check
1760 	 * for consistency.
1761 	 * XXX probably differs for removable media
1762 	 */
1763 	dp->blksize = SD_DEFAULT_BLKSIZE;
1764 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1765 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1766 
1767 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1768 	    &scsipi_sense.header, sizeof(scsipi_sense),
1769 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1770 
1771 	if (error != 0)
1772 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1773 
1774 	dp->blksize = blksize;
1775 	if (!sd_validate_blksize(NULL, dp->blksize))
1776 		dp->blksize = _2btol(scsipi_sense.lbs);
1777 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1778 		dp->blksize = SD_DEFAULT_BLKSIZE;
1779 
1780 	/*
1781 	 * Create a pseudo-geometry.
1782 	 */
1783 	dp->heads = 64;
1784 	dp->sectors = 32;
1785 	dp->cyls = blocks / (dp->heads * dp->sectors);
1786 	dp->disksize = _5btol(scsipi_sense.size);
1787 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1788 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1789 		       (unsigned long long)dp->disksize,
1790 		       (unsigned long long)blocks);
1791 		dp->disksize = blocks;
1792 	}
1793 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1794 
1795 	return (SDGP_RESULT_OK);
1796 }
1797 
1798 /*
1799  * Get the scsi driver to send a full inquiry to the * device and use the
1800  * results to fill out the disk parameter structure.
1801  */
1802 static int
1803 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1804 {
1805 	u_int64_t blocks;
1806 	int error, blksize;
1807 #if 0
1808 	int i;
1809 	u_int8_t *p;
1810 #endif
1811 
1812 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1813 	    flags);
1814 	if (blocks == 0) {
1815 		struct scsipi_read_format_capacities cmd;
1816 		struct {
1817 			struct scsipi_capacity_list_header header;
1818 			struct scsipi_capacity_descriptor desc;
1819 		} __attribute__((packed)) data;
1820 
1821 		memset(&cmd, 0, sizeof(cmd));
1822 		memset(&data, 0, sizeof(data));
1823 		cmd.opcode = READ_FORMAT_CAPACITIES;
1824 		_lto2b(sizeof(data), cmd.length);
1825 
1826 		error = scsipi_command(sd->sc_periph,
1827 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1828 		    SDRETRIES, 20000, NULL,
1829 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1830 		if (error == EFTYPE) {
1831 			/* Medium Format Corrupted, handle as not formatted */
1832 			return (SDGP_RESULT_UNFORMATTED);
1833 		}
1834 		if (error || data.header.length == 0)
1835 			return (SDGP_RESULT_OFFLINE);
1836 
1837 #if 0
1838 printf("rfc: length=%d\n", data.header.length);
1839 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");
1840 #endif
1841 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1842 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1843 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1844 			break;
1845 
1846 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1847 			return (SDGP_RESULT_UNFORMATTED);
1848 
1849 		case SCSIPI_CAP_DESC_CODE_NONE:
1850 			return (SDGP_RESULT_OFFLINE);
1851 		}
1852 
1853 		dp->disksize = blocks = _4btol(data.desc.nblks);
1854 		if (blocks == 0)
1855 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1856 
1857 		blksize = _3btol(data.desc.blklen);
1858 
1859 	} else if (!sd_validate_blksize(NULL, blksize)) {
1860 		struct sd_mode_sense_data scsipi_sense;
1861 		int big, bsize;
1862 		struct scsi_general_block_descriptor *bdesc;
1863 
1864 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1865 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1866 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1867 		if (!error) {
1868 			if (big) {
1869 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1870 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1871 			} else {
1872 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1873 				bsize = scsipi_sense.header.small.blk_desc_len;
1874 			}
1875 
1876 #if 0
1877 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1878 printf("page 0 bsize=%d\n", bsize);
1879 printf("page 0 ok\n");
1880 #endif
1881 
1882 			if (bsize >= 8) {
1883 				blksize = _3btol(bdesc->blklen);
1884 			}
1885 		}
1886 	}
1887 
1888 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1889 		blksize = SD_DEFAULT_BLKSIZE;
1890 
1891 	dp->blksize = blksize;
1892 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1893 	return (0);
1894 }
1895 
1896 static int
1897 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1898 {
1899 	struct sd_mode_sense_data scsipi_sense;
1900 	int error;
1901 	int big, byte2;
1902 	size_t poffset;
1903 	union scsi_disk_pages *pages;
1904 
1905 	byte2 = SMS_DBD;
1906 again:
1907 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1908 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1909 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1910 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1911 	    flags | XS_CTL_SILENT, &big);
1912 	if (error) {
1913 		if (byte2 == SMS_DBD) {
1914 			/* No result; try once more with DBD off */
1915 			byte2 = 0;
1916 			goto again;
1917 		}
1918 		return (error);
1919 	}
1920 
1921 	if (big) {
1922 		poffset = sizeof scsipi_sense.header.big;
1923 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1924 	} else {
1925 		poffset = sizeof scsipi_sense.header.small;
1926 		poffset += scsipi_sense.header.small.blk_desc_len;
1927 	}
1928 
1929 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1930 		return ERESTART;
1931 
1932 	pages = (void *)((u_long)&scsipi_sense + poffset);
1933 #if 0
1934 	{
1935 		size_t i;
1936 		u_int8_t *p;
1937 
1938 		printf("page 4 sense:");
1939 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
1940 		    i--, p++)
1941 			printf(" %02x", *p);
1942 		printf("\n");
1943 		printf("page 4 pg_code=%d sense=%p/%p\n",
1944 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1945 	}
1946 #endif
1947 
1948 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1949 		return (ERESTART);
1950 
1951 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1952 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1953 	    _3btol(pages->rigid_geometry.ncyl),
1954 	    pages->rigid_geometry.nheads,
1955 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1956 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1957 	    _2btol(pages->rigid_geometry.land_zone)));
1958 
1959 	/*
1960 	 * KLUDGE!! (for zone recorded disks)
1961 	 * give a number of sectors so that sec * trks * cyls
1962 	 * is <= disk_size
1963 	 * can lead to wasted space! THINK ABOUT THIS !
1964 	 */
1965 	dp->heads = pages->rigid_geometry.nheads;
1966 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1967 	if (dp->heads == 0 || dp->cyls == 0)
1968 		return (ERESTART);
1969 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1970 
1971 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1972 	if (dp->rot_rate == 0)
1973 		dp->rot_rate = 3600;
1974 
1975 #if 0
1976 printf("page 4 ok\n");
1977 #endif
1978 	return (0);
1979 }
1980 
1981 static int
1982 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1983 {
1984 	struct sd_mode_sense_data scsipi_sense;
1985 	int error;
1986 	int big, byte2;
1987 	size_t poffset;
1988 	union scsi_disk_pages *pages;
1989 
1990 	byte2 = SMS_DBD;
1991 again:
1992 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1993 	error = sd_mode_sense(sd, 0, &scsipi_sense,
1994 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1995 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
1996 	    flags | XS_CTL_SILENT, &big);
1997 	if (error) {
1998 		if (byte2 == SMS_DBD) {
1999 			/* No result; try once more with DBD off */
2000 			byte2 = 0;
2001 			goto again;
2002 		}
2003 		return (error);
2004 	}
2005 
2006 	if (big) {
2007 		poffset = sizeof scsipi_sense.header.big;
2008 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2009 	} else {
2010 		poffset = sizeof scsipi_sense.header.small;
2011 		poffset += scsipi_sense.header.small.blk_desc_len;
2012 	}
2013 
2014 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2015 		return ERESTART;
2016 
2017 	pages = (void *)((u_long)&scsipi_sense + poffset);
2018 #if 0
2019 	{
2020 		size_t i;
2021 		u_int8_t *p;
2022 
2023 		printf("page 5 sense:");
2024 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2025 		    i--, p++)
2026 			printf(" %02x", *p);
2027 		printf("\n");
2028 		printf("page 5 pg_code=%d sense=%p/%p\n",
2029 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2030 	}
2031 #endif
2032 
2033 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2034 		return (ERESTART);
2035 
2036 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2037 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2038 	    _3btol(pages->flex_geometry.ncyl),
2039 	    pages->flex_geometry.nheads,
2040 	    pages->flex_geometry.ph_sec_tr,
2041 	    _2btol(pages->flex_geometry.bytes_s)));
2042 
2043 	dp->heads = pages->flex_geometry.nheads;
2044 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2045 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2046 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2047 		return (ERESTART);
2048 
2049 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2050 	if (dp->rot_rate == 0)
2051 		dp->rot_rate = 3600;
2052 
2053 #if 0
2054 printf("page 5 ok\n");
2055 #endif
2056 	return (0);
2057 }
2058 
2059 static int
2060 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2061 {
2062 	int error;
2063 
2064 	/*
2065 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2066 	 * cleared by the caller if necessary.
2067 	 */
2068 	if (sd->type == T_SIMPLE_DIRECT) {
2069 		error = sd_get_simplifiedparms(sd, dp, flags);
2070 		if (!error)
2071 			disk_blocksize(&sd->sc_dk, dp->blksize);
2072 		return (error);
2073 	}
2074 
2075 	error = sd_get_capacity(sd, dp, flags);
2076 	if (error)
2077 		return (error);
2078 
2079 	disk_blocksize(&sd->sc_dk, dp->blksize);
2080 
2081 	if (sd->type == T_OPTICAL)
2082 		goto page0;
2083 
2084 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2085 		if (!sd_get_parms_page5(sd, dp, flags) ||
2086 		    !sd_get_parms_page4(sd, dp, flags))
2087 			return (SDGP_RESULT_OK);
2088 	} else {
2089 		if (!sd_get_parms_page4(sd, dp, flags) ||
2090 		    !sd_get_parms_page5(sd, dp, flags))
2091 			return (SDGP_RESULT_OK);
2092 	}
2093 
2094 page0:
2095 	printf("%s: fabricating a geometry\n", sd->sc_dev.dv_xname);
2096 	/* Try calling driver's method for figuring out geometry. */
2097 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2098 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2099 		(sd->sc_periph, dp, dp->disksize)) {
2100 		/*
2101 		 * Use adaptec standard fictitious geometry
2102 		 * this depends on which controller (e.g. 1542C is
2103 		 * different. but we have to put SOMETHING here..)
2104 		 */
2105 		dp->heads = 64;
2106 		dp->sectors = 32;
2107 		dp->cyls = dp->disksize / (64 * 32);
2108 	}
2109 	dp->rot_rate = 3600;
2110 	return (SDGP_RESULT_OK);
2111 }
2112 
2113 static int
2114 sd_flush(struct sd_softc *sd, int flags)
2115 {
2116 	struct scsipi_periph *periph = sd->sc_periph;
2117 	struct scsi_synchronize_cache_10 cmd;
2118 
2119 	/*
2120 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2121 	 * We issue with address 0 length 0, which should be
2122 	 * interpreted by the device as "all remaining blocks
2123 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2124 	 * in the event that the command is not supported by
2125 	 * the device, and poll for completion so that we know
2126 	 * that the cache has actually been flushed.
2127 	 *
2128 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2129 	 * command, as indicated by our quirks flags.
2130 	 *
2131 	 * XXX What about older devices?
2132 	 */
2133 	if (periph->periph_version < 2 ||
2134 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2135 		return (0);
2136 
2137 	sd->flags |= SDF_FLUSHING;
2138 	memset(&cmd, 0, sizeof(cmd));
2139 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2140 
2141 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2142 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2143 }
2144 
2145 static int
2146 sd_getcache(struct sd_softc *sd, int *bitsp)
2147 {
2148 	struct scsipi_periph *periph = sd->sc_periph;
2149 	struct sd_mode_sense_data scsipi_sense;
2150 	int error, bits = 0;
2151 	int big;
2152 	union scsi_disk_pages *pages;
2153 
2154 	if (periph->periph_version < 2)
2155 		return (EOPNOTSUPP);
2156 
2157 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2158 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2159 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2160 	if (error)
2161 		return (error);
2162 
2163 	if (big)
2164 		pages = (void *)(&scsipi_sense.header.big + 1);
2165 	else
2166 		pages = (void *)(&scsipi_sense.header.small + 1);
2167 
2168 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2169 		bits |= DKCACHE_READ;
2170 	if (pages->caching_params.flags & CACHING_WCE)
2171 		bits |= DKCACHE_WRITE;
2172 	if (pages->caching_params.pg_code & PGCODE_PS)
2173 		bits |= DKCACHE_SAVE;
2174 
2175 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2176 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2177 	    sizeof(scsipi_sense.pages.caching_params),
2178 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2179 	if (error == 0) {
2180 		if (big)
2181 			pages = (void *)(&scsipi_sense.header.big + 1);
2182 		else
2183 			pages = (void *)(&scsipi_sense.header.small + 1);
2184 
2185 		if (pages->caching_params.flags & CACHING_RCD)
2186 			bits |= DKCACHE_RCHANGE;
2187 		if (pages->caching_params.flags & CACHING_WCE)
2188 			bits |= DKCACHE_WCHANGE;
2189 	}
2190 
2191 	*bitsp = bits;
2192 
2193 	return (0);
2194 }
2195 
2196 static int
2197 sd_setcache(struct sd_softc *sd, int bits)
2198 {
2199 	struct scsipi_periph *periph = sd->sc_periph;
2200 	struct sd_mode_sense_data scsipi_sense;
2201 	int error;
2202 	uint8_t oflags, byte2 = 0;
2203 	int big;
2204 	union scsi_disk_pages *pages;
2205 
2206 	if (periph->periph_version < 2)
2207 		return (EOPNOTSUPP);
2208 
2209 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2210 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2211 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2212 	if (error)
2213 		return (error);
2214 
2215 	if (big)
2216 		pages = (void *)(&scsipi_sense.header.big + 1);
2217 	else
2218 		pages = (void *)(&scsipi_sense.header.small + 1);
2219 
2220 	oflags = pages->caching_params.flags;
2221 
2222 	if (bits & DKCACHE_READ)
2223 		pages->caching_params.flags &= ~CACHING_RCD;
2224 	else
2225 		pages->caching_params.flags |= CACHING_RCD;
2226 
2227 	if (bits & DKCACHE_WRITE)
2228 		pages->caching_params.flags |= CACHING_WCE;
2229 	else
2230 		pages->caching_params.flags &= ~CACHING_WCE;
2231 
2232 	if (oflags == pages->caching_params.flags)
2233 		return (0);
2234 
2235 	pages->caching_params.pg_code &= PGCODE_MASK;
2236 
2237 	if (bits & DKCACHE_SAVE)
2238 		byte2 |= SMS_SP;
2239 
2240 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2241 	    sizeof(struct scsi_mode_page_header) +
2242 	    pages->caching_params.pg_length, 0, big));
2243 }
2244 
2245 static void
2246 sd_set_properties(struct sd_softc *sd)
2247 {
2248 	prop_dictionary_t disk_info, odisk_info, geom;
2249 
2250 	disk_info = prop_dictionary_create();
2251 
2252 	geom = prop_dictionary_create();
2253 
2254 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2255 	    sd->params.disksize);
2256 
2257 	prop_dictionary_set_uint32(geom, "sector-size",
2258 	    sd->params.blksize);
2259 
2260 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2261 	    sd->params.sectors);
2262 
2263 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2264 	    sd->params.heads);
2265 
2266 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2267 	    sd->params.cyls);
2268 
2269 	prop_dictionary_set(disk_info, "geometry", geom);
2270 	prop_object_release(geom);
2271 
2272 	prop_dictionary_set(device_properties(&sd->sc_dev),
2273 	    "disk-info", disk_info);
2274 
2275 	/*
2276 	 * Don't release disk_info here; we keep a reference to it.
2277 	 * disk_detach() will release it when we go away.
2278 	 */
2279 
2280 	odisk_info = sd->sc_dk.dk_info;
2281 	sd->sc_dk.dk_info = disk_info;
2282 	if (odisk_info)
2283 		prop_object_release(odisk_info);
2284 }
2285