xref: /netbsd-src/sys/dev/scsipi/sd.c (revision aad9773e38ed2370a628a6416e098f9008fc10a7)
1 /*	$NetBSD: sd.c,v 1.310 2014/11/04 07:51:55 mlelstv 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 1.310 2014/11/04 07:51:55 mlelstv Exp $");
51 
52 #include "opt_scsi.h"
53 
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/kernel.h>
57 #include <sys/file.h>
58 #include <sys/stat.h>
59 #include <sys/ioctl.h>
60 #include <sys/scsiio.h>
61 #include <sys/buf.h>
62 #include <sys/bufq.h>
63 #include <sys/uio.h>
64 #include <sys/malloc.h>
65 #include <sys/errno.h>
66 #include <sys/device.h>
67 #include <sys/disklabel.h>
68 #include <sys/disk.h>
69 #include <sys/proc.h>
70 #include <sys/conf.h>
71 #include <sys/vnode.h>
72 #include <sys/rnd.h>
73 #include <sys/cprng.h>
74 
75 #include <dev/scsipi/scsi_spc.h>
76 #include <dev/scsipi/scsipi_all.h>
77 #include <dev/scsipi/scsi_all.h>
78 #include <dev/scsipi/scsipi_disk.h>
79 #include <dev/scsipi/scsi_disk.h>
80 #include <dev/scsipi/scsiconf.h>
81 #include <dev/scsipi/scsipi_base.h>
82 #include <dev/scsipi/sdvar.h>
83 
84 #include <prop/proplib.h>
85 
86 #define	SDUNIT(dev)			DISKUNIT(dev)
87 #define	SDPART(dev)			DISKPART(dev)
88 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
89 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
90 
91 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
92 
93 #define	SD_DEFAULT_BLKSIZE	512
94 
95 static void	sdminphys(struct buf *);
96 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
97 static int	sdgetdisklabel(struct sd_softc *);
98 static void	sdstart(struct scsipi_periph *);
99 static void	sdrestart(void *);
100 static void	sddone(struct scsipi_xfer *, int);
101 static bool	sd_suspend(device_t, const pmf_qual_t *);
102 static bool	sd_shutdown(device_t, int);
103 static int	sd_interpret_sense(struct scsipi_xfer *);
104 static int	sdlastclose(device_t);
105 
106 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
107 		    int, int *);
108 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
109 		    int);
110 static int	sd_validate_blksize(struct scsipi_periph *, int);
111 static u_int64_t sd_read_capacity(struct scsipi_periph *, int *, int flags);
112 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
113 		    int);
114 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
115 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
116 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
117 		    int);
118 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
119 		    int);
120 
121 static int	sd_flush(struct sd_softc *, int);
122 static int	sd_getcache(struct sd_softc *, int *);
123 static int	sd_setcache(struct sd_softc *, int);
124 
125 static int	sdmatch(device_t, cfdata_t, void *);
126 static void	sdattach(device_t, device_t, void *);
127 static int	sddetach(device_t, int);
128 static void	sd_set_geometry(struct sd_softc *);
129 
130 CFATTACH_DECL3_NEW(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
131     NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
132 
133 extern struct cfdriver sd_cd;
134 
135 static const struct scsipi_inquiry_pattern sd_patterns[] = {
136 	{T_DIRECT, T_FIXED,
137 	 "",         "",                 ""},
138 	{T_DIRECT, T_REMOV,
139 	 "",         "",                 ""},
140 	{T_OPTICAL, T_FIXED,
141 	 "",         "",                 ""},
142 	{T_OPTICAL, T_REMOV,
143 	 "",         "",                 ""},
144 	{T_SIMPLE_DIRECT, T_FIXED,
145 	 "",         "",                 ""},
146 	{T_SIMPLE_DIRECT, T_REMOV,
147 	 "",         "",                 ""},
148 };
149 
150 static dev_type_open(sdopen);
151 static dev_type_close(sdclose);
152 static dev_type_read(sdread);
153 static dev_type_write(sdwrite);
154 static dev_type_ioctl(sdioctl);
155 static dev_type_strategy(sdstrategy);
156 static dev_type_dump(sddump);
157 static dev_type_size(sdsize);
158 
159 const struct bdevsw sd_bdevsw = {
160 	.d_open = sdopen,
161 	.d_close = sdclose,
162 	.d_strategy = sdstrategy,
163 	.d_ioctl = sdioctl,
164 	.d_dump = sddump,
165 	.d_psize = sdsize,
166 	.d_discard = nodiscard,
167 	.d_flag = D_DISK
168 };
169 
170 const struct cdevsw sd_cdevsw = {
171 	.d_open = sdopen,
172 	.d_close = sdclose,
173 	.d_read = sdread,
174 	.d_write = sdwrite,
175 	.d_ioctl = sdioctl,
176 	.d_stop = nostop,
177 	.d_tty = notty,
178 	.d_poll = nopoll,
179 	.d_mmap = nommap,
180 	.d_kqfilter = nokqfilter,
181 	.d_discard = nodiscard,
182 	.d_flag = D_DISK
183 };
184 
185 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
186 
187 static const struct scsipi_periphsw sd_switch = {
188 	sd_interpret_sense,	/* check our error handler first */
189 	sdstart,		/* have a queue, served by this */
190 	NULL,			/* have no async handler */
191 	sddone,			/* deal with stats at interrupt time */
192 };
193 
194 struct sd_mode_sense_data {
195 	/*
196 	 * XXX
197 	 * We are not going to parse this as-is -- it just has to be large
198 	 * enough.
199 	 */
200 	union {
201 		struct scsi_mode_parameter_header_6 small;
202 		struct scsi_mode_parameter_header_10 big;
203 	} header;
204 	struct scsi_general_block_descriptor blk_desc;
205 	union scsi_disk_pages pages;
206 };
207 
208 /*
209  * The routine called by the low level scsi routine when it discovers
210  * A device suitable for this driver
211  */
212 static int
213 sdmatch(device_t parent, cfdata_t match,
214     void *aux)
215 {
216 	struct scsipibus_attach_args *sa = aux;
217 	int priority;
218 
219 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
220 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
221 	    sizeof(sd_patterns[0]), &priority);
222 
223 	return (priority);
224 }
225 
226 /*
227  * Attach routine common to atapi & scsi.
228  */
229 static void
230 sdattach(device_t parent, device_t self, void *aux)
231 {
232 	struct sd_softc *sd = device_private(self);
233 	struct scsipibus_attach_args *sa = aux;
234 	struct scsipi_periph *periph = sa->sa_periph;
235 	int error, result;
236 	struct disk_parms *dp = &sd->params;
237 	char pbuf[9];
238 
239 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
240 
241 	sd->sc_dev = self;
242 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
243 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
244 	if (sd->type == T_SIMPLE_DIRECT)
245 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
246 
247 	if (SCSIPI_BUSTYPE_TYPE(scsipi_periph_bustype(sa->sa_periph)) ==
248 	    SCSIPI_BUSTYPE_SCSI && periph->periph_version == 0)
249 		sd->flags |= SDF_ANCIENT;
250 
251 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
252 
253 	callout_init(&sd->sc_callout, 0);
254 
255 	/*
256 	 * Store information needed to contact our base driver
257 	 */
258 	sd->sc_periph = periph;
259 
260 	periph->periph_dev = sd->sc_dev;
261 	periph->periph_switch = &sd_switch;
262 
263         /*
264          * Increase our openings to the maximum-per-periph
265          * supported by the adapter.  This will either be
266          * clamped down or grown by the adapter if necessary.
267          */
268 	periph->periph_openings =
269 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
270 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
271 
272 	/*
273 	 * Initialize and attach the disk structure.
274 	 */
275 	disk_init(&sd->sc_dk, device_xname(sd->sc_dev), &sddkdriver);
276 	disk_attach(&sd->sc_dk);
277 
278 	/*
279 	 * Use the subdriver to request information regarding the drive.
280 	 */
281 	aprint_naive("\n");
282 	aprint_normal("\n");
283 
284 	if (periph->periph_quirks & PQUIRK_START)
285 		(void)scsipi_start(periph, SSS_START, XS_CTL_SILENT);
286 
287 	error = scsipi_test_unit_ready(periph,
288 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
289 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
290 
291 	if (error)
292 		result = SDGP_RESULT_OFFLINE;
293 	else
294 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
295 	aprint_normal_dev(sd->sc_dev, "");
296 	switch (result) {
297 	case SDGP_RESULT_OK:
298 		format_bytes(pbuf, sizeof(pbuf),
299 		    (u_int64_t)dp->disksize * dp->blksize);
300 	        aprint_normal(
301 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
302 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
303 		    (unsigned long long)dp->disksize);
304 		break;
305 
306 	case SDGP_RESULT_OFFLINE:
307 		aprint_normal("drive offline");
308 		break;
309 
310 	case SDGP_RESULT_UNFORMATTED:
311 		aprint_normal("unformatted media");
312 		break;
313 
314 #ifdef DIAGNOSTIC
315 	default:
316 		panic("sdattach: unknown result from get_parms");
317 		break;
318 #endif
319 	}
320 	aprint_normal("\n");
321 
322 	/*
323 	 * Establish a shutdown hook so that we can ensure that
324 	 * our data has actually made it onto the platter at
325 	 * shutdown time.  Note that this relies on the fact
326 	 * that the shutdown hooks at the "leaves" of the device tree
327 	 * are run, first (thus guaranteeing that our hook runs before
328 	 * our ancestors').
329 	 */
330 	if (!pmf_device_register1(self, sd_suspend, NULL, sd_shutdown))
331 		aprint_error_dev(self, "couldn't establish power handler\n");
332 
333 	/*
334 	 * attach the device into the random source list
335 	 */
336 	rnd_attach_source(&sd->rnd_source, device_xname(sd->sc_dev),
337 			  RND_TYPE_DISK, RND_FLAG_DEFAULT);
338 
339 	/* Discover wedges on this disk. */
340 	dkwedge_discover(&sd->sc_dk);
341 
342 	/*
343 	 * Disk insertion and removal times can be a useful source
344 	 * of entropy, though the estimator should never _count_
345 	 * these bits, on insertion, because the deltas to the
346 	 * nonexistent) previous event should never allow it.
347 	 */
348 	rnd_add_uint32(&sd->rnd_source, 0);
349 }
350 
351 static int
352 sddetach(device_t self, int flags)
353 {
354 	struct sd_softc *sd = device_private(self);
355 	int s, bmaj, cmaj, i, mn, rc;
356 
357 	rnd_add_uint32(&sd->rnd_source, 0);
358 
359 	if ((rc = disk_begindetach(&sd->sc_dk, sdlastclose, self, flags)) != 0)
360 		return rc;
361 
362 	/* locate the major number */
363 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
364 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
365 
366 	/* Nuke the vnodes for any open instances */
367 	for (i = 0; i < MAXPARTITIONS; i++) {
368 		mn = SDMINOR(device_unit(self), i);
369 		vdevgone(bmaj, mn, mn, VBLK);
370 		vdevgone(cmaj, mn, mn, VCHR);
371 	}
372 
373 	/* kill any pending restart */
374 	callout_stop(&sd->sc_callout);
375 
376 	/* Delete all of our wedges. */
377 	dkwedge_delall(&sd->sc_dk);
378 
379 	s = splbio();
380 
381 	/* Kill off any queued buffers. */
382 	bufq_drain(sd->buf_queue);
383 
384 	bufq_free(sd->buf_queue);
385 
386 	/* Kill off any pending commands. */
387 	scsipi_kill_pending(sd->sc_periph);
388 
389 	splx(s);
390 
391 	/* Detach from the disk list. */
392 	disk_detach(&sd->sc_dk);
393 	disk_destroy(&sd->sc_dk);
394 
395 	callout_destroy(&sd->sc_callout);
396 
397 	pmf_device_deregister(self);
398 
399 	/* Unhook the entropy source. */
400 	rnd_detach_source(&sd->rnd_source);
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) || (flag & FSILENT))
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 && (flag & FSILENT) == 0)
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  * Caller must hold sd->sc_dk.dk_openlock.
592  */
593 static int
594 sdlastclose(device_t self)
595 {
596 	struct sd_softc *sd = device_private(self);
597 	struct scsipi_periph *periph = sd->sc_periph;
598 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
599 
600 	/*
601 	 * If the disk cache needs flushing, and the disk supports
602 	 * it, do it now.
603 	 */
604 	if ((sd->flags & SDF_DIRTY) != 0) {
605 		if (sd_flush(sd, 0)) {
606 			aprint_error_dev(sd->sc_dev,
607 				"cache synchronization failed\n");
608 			sd->flags &= ~SDF_FLUSHING;
609 		} else
610 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
611 	}
612 
613 	scsipi_wait_drain(periph);
614 
615 	if (periph->periph_flags & PERIPH_REMOVABLE)
616 		scsipi_prevent(periph, SPAMR_ALLOW,
617 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
618 		    XS_CTL_IGNORE_NOT_READY |
619 		    XS_CTL_SILENT);
620 	periph->periph_flags &= ~PERIPH_OPEN;
621 
622 	scsipi_wait_drain(periph);
623 
624 	scsipi_adapter_delref(adapt);
625 
626 	return 0;
627 }
628 
629 /*
630  * close the device.. only called if we are the LAST occurence of an open
631  * device.  Convenient now but usually a pain.
632  */
633 static int
634 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
635 {
636 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
637 	int part = SDPART(dev);
638 
639 	mutex_enter(&sd->sc_dk.dk_openlock);
640 	switch (fmt) {
641 	case S_IFCHR:
642 		sd->sc_dk.dk_copenmask &= ~(1 << part);
643 		break;
644 	case S_IFBLK:
645 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
646 		break;
647 	}
648 	sd->sc_dk.dk_openmask =
649 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
650 
651 	if (sd->sc_dk.dk_openmask == 0)
652 		sdlastclose(sd->sc_dev);
653 
654 	mutex_exit(&sd->sc_dk.dk_openlock);
655 	return (0);
656 }
657 
658 /*
659  * Actually translate the requested transfer into one the physical driver
660  * can understand.  The transfer is described by a buf and will include
661  * only one physical transfer.
662  */
663 static void
664 sdstrategy(struct buf *bp)
665 {
666 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
667 	struct scsipi_periph *periph = sd->sc_periph;
668 	struct disklabel *lp;
669 	daddr_t blkno;
670 	int s;
671 	bool sector_aligned;
672 
673 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
674 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
675 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
676 	/*
677 	 * If the device has been made invalid, error out
678 	 */
679 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
680 	    !device_is_active(sd->sc_dev)) {
681 		if (periph->periph_flags & PERIPH_OPEN)
682 			bp->b_error = EIO;
683 		else
684 			bp->b_error = ENODEV;
685 		goto done;
686 	}
687 
688 	lp = sd->sc_dk.dk_label;
689 
690 	/*
691 	 * The transfer must be a whole number of blocks, offset must not be
692 	 * negative.
693 	 */
694 	if (lp->d_secsize == DEV_BSIZE) {
695 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
696 	} else {
697 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
698 	}
699 	if (!sector_aligned || bp->b_blkno < 0) {
700 		bp->b_error = EINVAL;
701 		goto done;
702 	}
703 	/*
704 	 * If it's a null transfer, return immediatly
705 	 */
706 	if (bp->b_bcount == 0)
707 		goto done;
708 
709 	/*
710 	 * Do bounds checking, adjust transfer. if error, process.
711 	 * If end of partition, just return.
712 	 */
713 	if (SDPART(bp->b_dev) == RAW_PART) {
714 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
715 		    sd->params.disksize512) <= 0)
716 			goto done;
717 	} else {
718 		if (bounds_check_with_label(&sd->sc_dk, bp,
719 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
720 			goto done;
721 	}
722 
723 	/*
724 	 * Now convert the block number to absolute and put it in
725 	 * terms of the device's logical block size.
726 	 */
727 	if (lp->d_secsize == DEV_BSIZE)
728 		blkno = bp->b_blkno;
729 	else if (lp->d_secsize > DEV_BSIZE)
730 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
731 	else
732 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
733 
734 	if (SDPART(bp->b_dev) != RAW_PART)
735 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
736 
737 	bp->b_rawblkno = blkno;
738 
739 	s = splbio();
740 
741 	/*
742 	 * Place it in the queue of disk activities for this disk.
743 	 *
744 	 * XXX Only do disksort() if the current operating mode does not
745 	 * XXX include tagged queueing.
746 	 */
747 	bufq_put(sd->buf_queue, bp);
748 
749 	/*
750 	 * Tell the device to get going on the transfer if it's
751 	 * not doing anything, otherwise just wait for completion
752 	 */
753 	sdstart(sd->sc_periph);
754 
755 	splx(s);
756 	return;
757 
758 done:
759 	/*
760 	 * Correctly set the buf to indicate a completed xfer
761 	 */
762 	bp->b_resid = bp->b_bcount;
763 	biodone(bp);
764 }
765 
766 /*
767  * sdstart looks to see if there is a buf waiting for the device
768  * and that the device is not already busy. If both are true,
769  * It dequeues the buf and creates a scsi command to perform the
770  * transfer in the buf. The transfer request will call scsipi_done
771  * on completion, which will in turn call this routine again
772  * so that the next queued transfer is performed.
773  * The bufs are queued by the strategy routine (sdstrategy)
774  *
775  * This routine is also called after other non-queued requests
776  * have been made of the scsi driver, to ensure that the queue
777  * continues to be drained.
778  *
779  * must be called at the correct (highish) spl level
780  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
781  */
782 static void
783 sdstart(struct scsipi_periph *periph)
784 {
785 	struct sd_softc *sd = device_private(periph->periph_dev);
786 	struct disklabel *lp = sd->sc_dk.dk_label;
787 	struct buf *bp = 0;
788 	struct scsipi_rw_16 cmd16;
789 	struct scsipi_rw_10 cmd_big;
790 	struct scsi_rw_6 cmd_small;
791 	struct scsipi_generic *cmdp;
792 	struct scsipi_xfer *xs;
793 	int nblks, cmdlen, error __diagused, flags;
794 
795 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
796 	/*
797 	 * Check if the device has room for another command
798 	 */
799 	while (periph->periph_active < periph->periph_openings) {
800 		/*
801 		 * there is excess capacity, but a special waits
802 		 * It'll need the adapter as soon as we clear out of the
803 		 * way and let it run (user level wait).
804 		 */
805 		if (periph->periph_flags & PERIPH_WAITING) {
806 			periph->periph_flags &= ~PERIPH_WAITING;
807 			wakeup((void *)periph);
808 			return;
809 		}
810 
811 		/*
812 		 * If the device has become invalid, abort all the
813 		 * reads and writes until all files have been closed and
814 		 * re-opened
815 		 */
816 		if (__predict_false(
817 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
818 			if ((bp = bufq_get(sd->buf_queue)) != NULL) {
819 				bp->b_error = EIO;
820 				bp->b_resid = bp->b_bcount;
821 				biodone(bp);
822 				continue;
823 			} else {
824 				return;
825 			}
826 		}
827 
828 		/*
829 		 * See if there is a buf with work for us to do..
830 		 */
831 		if ((bp = bufq_peek(sd->buf_queue)) == NULL)
832 			return;
833 
834 		/*
835 		 * We have a buf, now we should make a command.
836 		 */
837 
838 		if (lp->d_secsize == DEV_BSIZE)
839 			nblks = bp->b_bcount >> DEV_BSHIFT;
840 		else
841 			nblks = howmany(bp->b_bcount, lp->d_secsize);
842 
843 		/*
844 		 * Fill out the scsi command.  Use the smallest CDB possible
845 		 * (6-byte, 10-byte, or 16-byte).
846 		 */
847 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
848 		    ((nblks & 0xff) == nblks) &&
849 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
850 			/* 6-byte CDB */
851 			memset(&cmd_small, 0, sizeof(cmd_small));
852 			cmd_small.opcode = (bp->b_flags & B_READ) ?
853 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
854 			_lto3b(bp->b_rawblkno, cmd_small.addr);
855 			cmd_small.length = nblks & 0xff;
856 			cmdlen = sizeof(cmd_small);
857 			cmdp = (struct scsipi_generic *)&cmd_small;
858 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
859 			/* 10-byte CDB */
860 			memset(&cmd_big, 0, sizeof(cmd_big));
861 			cmd_big.opcode = (bp->b_flags & B_READ) ?
862 			    READ_10 : WRITE_10;
863 			_lto4b(bp->b_rawblkno, cmd_big.addr);
864 			_lto2b(nblks, cmd_big.length);
865 			cmdlen = sizeof(cmd_big);
866 			cmdp = (struct scsipi_generic *)&cmd_big;
867 		} else {
868 			/* 16-byte CDB */
869 			memset(&cmd16, 0, sizeof(cmd16));
870 			cmd16.opcode = (bp->b_flags & B_READ) ?
871 			    READ_16 : WRITE_16;
872 			_lto8b(bp->b_rawblkno, cmd16.addr);
873 			_lto4b(nblks, cmd16.length);
874 			cmdlen = sizeof(cmd16);
875 			cmdp = (struct scsipi_generic *)&cmd16;
876 		}
877 
878 		/* Instrumentation. */
879 		disk_busy(&sd->sc_dk);
880 
881 		/*
882 		 * Mark the disk dirty so that the cache will be
883 		 * flushed on close.
884 		 */
885 		if ((bp->b_flags & B_READ) == 0)
886 			sd->flags |= SDF_DIRTY;
887 
888 		/*
889 		 * Figure out what flags to use.
890 		 */
891 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
892 		if (bp->b_flags & B_READ)
893 			flags |= XS_CTL_DATA_IN;
894 		else
895 			flags |= XS_CTL_DATA_OUT;
896 
897 		/*
898 		 * Call the routine that chats with the adapter.
899 		 * Note: we cannot sleep as we may be an interrupt
900 		 */
901 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
902 		    (u_char *)bp->b_data, bp->b_bcount,
903 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
904 		if (__predict_false(xs == NULL)) {
905 			/*
906 			 * out of memory. Keep this buffer in the queue, and
907 			 * retry later.
908 			 */
909 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
910 			    periph);
911 			return;
912 		}
913 		/*
914 		 * need to dequeue the buffer before queuing the command,
915 		 * because cdstart may be called recursively from the
916 		 * HBA driver
917 		 */
918 #ifdef DIAGNOSTIC
919 		if (bufq_get(sd->buf_queue) != bp)
920 			panic("sdstart(): dequeued wrong buf");
921 #else
922 		bufq_get(sd->buf_queue);
923 #endif
924 		error = scsipi_execute_xs(xs);
925 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
926 		KASSERT(error == 0);
927 	}
928 }
929 
930 static void
931 sdrestart(void *v)
932 {
933 	int s = splbio();
934 	sdstart((struct scsipi_periph *)v);
935 	splx(s);
936 }
937 
938 static void
939 sddone(struct scsipi_xfer *xs, int error)
940 {
941 	struct sd_softc *sd = device_private(xs->xs_periph->periph_dev);
942 	struct buf *bp = xs->bp;
943 
944 	if (sd->flags & SDF_FLUSHING) {
945 		/* Flush completed, no longer dirty. */
946 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
947 	}
948 
949 	if (bp) {
950 		bp->b_error = error;
951 		bp->b_resid = xs->resid;
952 		if (error) {
953 			/* on a read/write error bp->b_resid is zero, so fix */
954 			bp->b_resid = bp->b_bcount;
955 		}
956 
957 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
958 		    (bp->b_flags & B_READ));
959 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
960 
961 		biodone(bp);
962 	}
963 }
964 
965 static void
966 sdminphys(struct buf *bp)
967 {
968 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
969 	long xmax;
970 
971 	/*
972 	 * If the device is ancient, we want to make sure that
973 	 * the transfer fits into a 6-byte cdb.
974 	 *
975 	 * XXX Note that the SCSI-I spec says that 256-block transfers
976 	 * are allowed in a 6-byte read/write, and are specified
977 	 * by settng the "length" to 0.  However, we're conservative
978 	 * here, allowing only 255-block transfers in case an
979 	 * ancient device gets confused by length == 0.  A length of 0
980 	 * in a 10-byte read/write actually means 0 blocks.
981 	 */
982 	if ((sd->flags & SDF_ANCIENT) &&
983 	    ((sd->sc_periph->periph_flags &
984 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
985 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
986 
987 		if (bp->b_bcount > xmax)
988 			bp->b_bcount = xmax;
989 	}
990 
991 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
992 }
993 
994 static int
995 sdread(dev_t dev, struct uio *uio, int ioflag)
996 {
997 
998 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
999 }
1000 
1001 static int
1002 sdwrite(dev_t dev, struct uio *uio, int ioflag)
1003 {
1004 
1005 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
1006 }
1007 
1008 /*
1009  * Perform special action on behalf of the user
1010  * Knows about the internals of this device
1011  */
1012 static int
1013 sdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1014 {
1015 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
1016 	struct scsipi_periph *periph = sd->sc_periph;
1017 	int part = SDPART(dev);
1018 	int error;
1019 	int s;
1020 #ifdef __HAVE_OLD_DISKLABEL
1021 	struct disklabel *newlabel = NULL;
1022 #endif
1023 
1024 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1025 
1026 	/*
1027 	 * If the device is not valid, some IOCTLs can still be
1028 	 * handled on the raw partition. Check this here.
1029 	 */
1030 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1031 		switch (cmd) {
1032 		case DIOCKLABEL:
1033 		case DIOCWLABEL:
1034 		case DIOCLOCK:
1035 		case DIOCEJECT:
1036 		case ODIOCEJECT:
1037 		case DIOCGCACHE:
1038 		case DIOCSCACHE:
1039 		case DIOCGSTRATEGY:
1040 		case DIOCSSTRATEGY:
1041 		case SCIOCIDENTIFY:
1042 		case OSCIOCIDENTIFY:
1043 		case SCIOCCOMMAND:
1044 		case SCIOCDEBUG:
1045 			if (part == RAW_PART)
1046 				break;
1047 		/* FALLTHROUGH */
1048 		default:
1049 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1050 				return (ENODEV);
1051 			else
1052 				return (EIO);
1053 		}
1054 	}
1055 
1056 	error = disk_ioctl(&sd->sc_dk, cmd, addr, flag, l);
1057 	if (error != EPASSTHROUGH)
1058 		return (error);
1059 
1060 	error = 0;
1061 	switch (cmd) {
1062 	case DIOCGDINFO:
1063 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1064 		return (0);
1065 
1066 #ifdef __HAVE_OLD_DISKLABEL
1067 	case ODIOCGDINFO:
1068 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1069 		if (newlabel == NULL)
1070 			return EIO;
1071 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1072 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1073 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1074 		else
1075 			error = ENOTTY;
1076 		free(newlabel, M_TEMP);
1077 		return error;
1078 #endif
1079 
1080 	case DIOCGPART:
1081 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1082 		((struct partinfo *)addr)->part =
1083 		    &sd->sc_dk.dk_label->d_partitions[part];
1084 		return (0);
1085 
1086 	case DIOCWDINFO:
1087 	case DIOCSDINFO:
1088 #ifdef __HAVE_OLD_DISKLABEL
1089 	case ODIOCWDINFO:
1090 	case ODIOCSDINFO:
1091 #endif
1092 	{
1093 		struct disklabel *lp;
1094 
1095 		if ((flag & FWRITE) == 0)
1096 			return (EBADF);
1097 
1098 #ifdef __HAVE_OLD_DISKLABEL
1099  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1100 			newlabel = malloc(sizeof *newlabel, M_TEMP,
1101 			    M_WAITOK | M_ZERO);
1102 			if (newlabel == NULL)
1103 				return EIO;
1104 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1105 			lp = newlabel;
1106 		} else
1107 #endif
1108 		lp = (struct disklabel *)addr;
1109 
1110 		mutex_enter(&sd->sc_dk.dk_openlock);
1111 		sd->flags |= SDF_LABELLING;
1112 
1113 		error = setdisklabel(sd->sc_dk.dk_label,
1114 		    lp, /*sd->sc_dk.dk_openmask : */0,
1115 		    sd->sc_dk.dk_cpulabel);
1116 		if (error == 0) {
1117 			if (cmd == DIOCWDINFO
1118 #ifdef __HAVE_OLD_DISKLABEL
1119 			    || cmd == ODIOCWDINFO
1120 #endif
1121 			   )
1122 				error = writedisklabel(SDLABELDEV(dev),
1123 				    sdstrategy, sd->sc_dk.dk_label,
1124 				    sd->sc_dk.dk_cpulabel);
1125 		}
1126 
1127 		sd->flags &= ~SDF_LABELLING;
1128 		mutex_exit(&sd->sc_dk.dk_openlock);
1129 #ifdef __HAVE_OLD_DISKLABEL
1130 		if (newlabel != NULL)
1131 			free(newlabel, M_TEMP);
1132 #endif
1133 		return (error);
1134 	}
1135 
1136 	case DIOCKLABEL:
1137 		if (*(int *)addr)
1138 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1139 		else
1140 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1141 		return (0);
1142 
1143 	case DIOCWLABEL:
1144 		if ((flag & FWRITE) == 0)
1145 			return (EBADF);
1146 		if (*(int *)addr)
1147 			sd->flags |= SDF_WLABEL;
1148 		else
1149 			sd->flags &= ~SDF_WLABEL;
1150 		return (0);
1151 
1152 	case DIOCLOCK:
1153 		if (periph->periph_flags & PERIPH_REMOVABLE)
1154 			return (scsipi_prevent(periph,
1155 			    (*(int *)addr) ?
1156 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1157 		else
1158 			return (ENOTTY);
1159 
1160 	case DIOCEJECT:
1161 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1162 			return (ENOTTY);
1163 		if (*(int *)addr == 0) {
1164 			/*
1165 			 * Don't force eject: check that we are the only
1166 			 * partition open. If so, unlock it.
1167 			 */
1168 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1169 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1170 			    sd->sc_dk.dk_openmask) {
1171 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1172 				    XS_CTL_IGNORE_NOT_READY);
1173 				if (error)
1174 					return (error);
1175 			} else {
1176 				return (EBUSY);
1177 			}
1178 		}
1179 		/* FALLTHROUGH */
1180 	case ODIOCEJECT:
1181 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1182 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1183 
1184 	case DIOCGDEFLABEL:
1185 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1186 		return (0);
1187 
1188 #ifdef __HAVE_OLD_DISKLABEL
1189 	case ODIOCGDEFLABEL:
1190 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1191 		if (newlabel == NULL)
1192 			return EIO;
1193 		sdgetdefaultlabel(sd, newlabel);
1194 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1195 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1196 		else
1197 			error = ENOTTY;
1198 		free(newlabel, M_TEMP);
1199 		return error;
1200 #endif
1201 
1202 	case DIOCGCACHE:
1203 		return (sd_getcache(sd, (int *) addr));
1204 
1205 	case DIOCSCACHE:
1206 		if ((flag & FWRITE) == 0)
1207 			return (EBADF);
1208 		return (sd_setcache(sd, *(int *) addr));
1209 
1210 	case DIOCCACHESYNC:
1211 		/*
1212 		 * XXX Do we really need to care about having a writable
1213 		 * file descriptor here?
1214 		 */
1215 		if ((flag & FWRITE) == 0)
1216 			return (EBADF);
1217 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1218 			error = sd_flush(sd, 0);
1219 			if (error)
1220 				sd->flags &= ~SDF_FLUSHING;
1221 			else
1222 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1223 		}
1224 		return (error);
1225 
1226 	case DIOCAWEDGE:
1227 	    {
1228 	    	struct dkwedge_info *dkw = (void *) addr;
1229 
1230 		if ((flag & FWRITE) == 0)
1231 			return (EBADF);
1232 
1233 		/* If the ioctl happens here, the parent is us. */
1234 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1235 			sizeof(dkw->dkw_parent));
1236 		return (dkwedge_add(dkw));
1237 	    }
1238 
1239 	case DIOCDWEDGE:
1240 	    {
1241 	    	struct dkwedge_info *dkw = (void *) addr;
1242 
1243 		if ((flag & FWRITE) == 0)
1244 			return (EBADF);
1245 
1246 		/* If the ioctl happens here, the parent is us. */
1247 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1248 			sizeof(dkw->dkw_parent));
1249 		return (dkwedge_del(dkw));
1250 	    }
1251 
1252 	case DIOCLWEDGES:
1253 	    {
1254 	    	struct dkwedge_list *dkwl = (void *) addr;
1255 
1256 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1257 	    }
1258 
1259 	case DIOCMWEDGES:
1260 	    {
1261 		if ((flag & FWRITE) == 0)
1262 			return (EBADF);
1263 
1264 		dkwedge_discover(&sd->sc_dk);
1265 		return 0;
1266 	    }
1267 
1268 	case DIOCGSTRATEGY:
1269 	    {
1270 		struct disk_strategy *dks = addr;
1271 
1272 		s = splbio();
1273 		strlcpy(dks->dks_name, bufq_getstrategyname(sd->buf_queue),
1274 		    sizeof(dks->dks_name));
1275 		splx(s);
1276 		dks->dks_paramlen = 0;
1277 
1278 		return 0;
1279 	    }
1280 
1281 	case DIOCSSTRATEGY:
1282 	    {
1283 		struct disk_strategy *dks = addr;
1284 		struct bufq_state *new_bufq;
1285 		struct bufq_state *old_bufq;
1286 
1287 		if ((flag & FWRITE) == 0) {
1288 			return EBADF;
1289 		}
1290 
1291 		if (dks->dks_param != NULL) {
1292 			return EINVAL;
1293 		}
1294 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1295 		error = bufq_alloc(&new_bufq, dks->dks_name,
1296 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1297 		if (error) {
1298 			return error;
1299 		}
1300 		s = splbio();
1301 		old_bufq = sd->buf_queue;
1302 		bufq_move(new_bufq, old_bufq);
1303 		sd->buf_queue = new_bufq;
1304 		splx(s);
1305 		bufq_free(old_bufq);
1306 
1307 		return 0;
1308 	    }
1309 
1310 	default:
1311 		if (part != RAW_PART)
1312 			return (ENOTTY);
1313 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1314 	}
1315 
1316 #ifdef DIAGNOSTIC
1317 	panic("sdioctl: impossible");
1318 #endif
1319 }
1320 
1321 static void
1322 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1323 {
1324 
1325 	memset(lp, 0, sizeof(struct disklabel));
1326 
1327 	lp->d_secsize = sd->params.blksize;
1328 	lp->d_ntracks = sd->params.heads;
1329 	lp->d_nsectors = sd->params.sectors;
1330 	lp->d_ncylinders = sd->params.cyls;
1331 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1332 
1333 	switch (SCSIPI_BUSTYPE_TYPE(scsipi_periph_bustype(sd->sc_periph))) {
1334 	case SCSIPI_BUSTYPE_SCSI:
1335 		lp->d_type = DTYPE_SCSI;
1336 		break;
1337 	case SCSIPI_BUSTYPE_ATAPI:
1338 		lp->d_type = DTYPE_ATAPI;
1339 		break;
1340 	}
1341 	/*
1342 	 * XXX
1343 	 * We could probe the mode pages to figure out what kind of disc it is.
1344 	 * Is this worthwhile?
1345 	 */
1346 	strncpy(lp->d_typename, sd->name, 16);
1347 	strncpy(lp->d_packname, "fictitious", 16);
1348 	if (sd->params.disksize > UINT32_MAX)
1349 		lp->d_secperunit = UINT32_MAX;
1350 	else
1351 		lp->d_secperunit = sd->params.disksize;
1352 	lp->d_rpm = sd->params.rot_rate;
1353 	lp->d_interleave = 1;
1354 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1355 	    D_REMOVABLE : 0;
1356 
1357 	lp->d_partitions[RAW_PART].p_offset = 0;
1358 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1359 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1360 	lp->d_npartitions = RAW_PART + 1;
1361 
1362 	lp->d_magic = DISKMAGIC;
1363 	lp->d_magic2 = DISKMAGIC;
1364 	lp->d_checksum = dkcksum(lp);
1365 }
1366 
1367 
1368 /*
1369  * Load the label information on the named device
1370  */
1371 static int
1372 sdgetdisklabel(struct sd_softc *sd)
1373 {
1374 	struct disklabel *lp = sd->sc_dk.dk_label;
1375 	const char *errstring;
1376 
1377 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1378 
1379 	sdgetdefaultlabel(sd, lp);
1380 
1381 	if (lp->d_secpercyl == 0) {
1382 		lp->d_secpercyl = 100;
1383 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1384 	}
1385 
1386 	/*
1387 	 * Call the generic disklabel extraction routine
1388 	 */
1389 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1390 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1391 	if (errstring) {
1392 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1393 		return EIO;
1394 	}
1395 	return 0;
1396 }
1397 
1398 static bool
1399 sd_shutdown(device_t self, int how)
1400 {
1401 	struct sd_softc *sd = device_private(self);
1402 
1403 	/*
1404 	 * If the disk cache needs to be flushed, and the disk supports
1405 	 * it, flush it.  We're cold at this point, so we poll for
1406 	 * completion.
1407 	 */
1408 	if ((sd->flags & SDF_DIRTY) != 0) {
1409 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1410 			aprint_error_dev(sd->sc_dev,
1411 				"cache synchronization failed\n");
1412 			sd->flags &= ~SDF_FLUSHING;
1413 		} else
1414 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1415 	}
1416 	return true;
1417 }
1418 
1419 static bool
1420 sd_suspend(device_t dv, const pmf_qual_t *qual)
1421 {
1422 	return sd_shutdown(dv, boothowto); /* XXX no need to poll */
1423 }
1424 
1425 /*
1426  * Check Errors
1427  */
1428 static int
1429 sd_interpret_sense(struct scsipi_xfer *xs)
1430 {
1431 	struct scsipi_periph *periph = xs->xs_periph;
1432 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1433 	struct sd_softc *sd = device_private(periph->periph_dev);
1434 	int s, error, retval = EJUSTRETURN;
1435 
1436 	/*
1437 	 * If the periph is already recovering, just do the normal
1438 	 * error processing.
1439 	 */
1440 	if (periph->periph_flags & PERIPH_RECOVERING)
1441 		return (retval);
1442 
1443 	/*
1444 	 * Ignore errors from accessing illegal fields (e.g. trying to
1445 	 * lock the door of a digicam, which doesn't have a door that
1446 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1447 	 */
1448 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1449 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1450 	    sense->asc == 0x24 &&
1451 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1452 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1453 			scsipi_printaddr(periph);
1454 			printf("no door lock\n");
1455 		}
1456 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1457 		return (retval);
1458 	}
1459 
1460 
1461 
1462 	/*
1463 	 * If the device is not open yet, let the generic code handle it.
1464 	 */
1465 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1466 		return (retval);
1467 
1468 	/*
1469 	 * If it isn't a extended or extended/deferred error, let
1470 	 * the generic code handle it.
1471 	 */
1472 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1473 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1474 		return (retval);
1475 
1476 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1477 	    sense->asc == 0x4) {
1478 		if (sense->ascq == 0x01)	{
1479 			/*
1480 			 * Unit In The Process Of Becoming Ready.
1481 			 */
1482 			printf("%s: waiting for pack to spin up...\n",
1483 			    device_xname(sd->sc_dev));
1484 			if (!callout_pending(&periph->periph_callout))
1485 				scsipi_periph_freeze(periph, 1);
1486 			callout_reset(&periph->periph_callout,
1487 			    5 * hz, scsipi_periph_timed_thaw, periph);
1488 			retval = ERESTART;
1489 		} else if (sense->ascq == 0x02) {
1490 			printf("%s: pack is stopped, restarting...\n",
1491 			    device_xname(sd->sc_dev));
1492 			s = splbio();
1493 			periph->periph_flags |= PERIPH_RECOVERING;
1494 			splx(s);
1495 			error = scsipi_start(periph, SSS_START,
1496 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1497 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1498 			if (error) {
1499 				aprint_error_dev(sd->sc_dev,
1500 					"unable to restart pack\n");
1501 				retval = error;
1502 			} else
1503 				retval = ERESTART;
1504 			s = splbio();
1505 			periph->periph_flags &= ~PERIPH_RECOVERING;
1506 			splx(s);
1507 		}
1508 	}
1509 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1510 	    sense->asc == 0x31 &&
1511 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1512 		/* Medium Format Corrupted */
1513 		retval = EFTYPE;
1514 	}
1515 	return (retval);
1516 }
1517 
1518 
1519 static int
1520 sdsize(dev_t dev)
1521 {
1522 	struct sd_softc *sd;
1523 	int part, unit, omask;
1524 	int size;
1525 
1526 	unit = SDUNIT(dev);
1527 	sd = device_lookup_private(&sd_cd, unit);
1528 	if (sd == NULL)
1529 		return (-1);
1530 
1531 	if (!device_is_active(sd->sc_dev))
1532 		return (-1);
1533 
1534 	part = SDPART(dev);
1535 	omask = sd->sc_dk.dk_openmask & (1 << part);
1536 
1537 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1538 		return (-1);
1539 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1540 		size = -1;
1541 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1542 		size = -1;
1543 	else
1544 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1545 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1546 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1547 		return (-1);
1548 	return (size);
1549 }
1550 
1551 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1552 static struct scsipi_xfer sx;
1553 static int sddoingadump;
1554 
1555 /*
1556  * dump all of physical memory into the partition specified, starting
1557  * at offset 'dumplo' into the partition.
1558  */
1559 static int
1560 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1561 {
1562 	struct sd_softc *sd;	/* disk unit to do the I/O */
1563 	struct disklabel *lp;	/* disk's disklabel */
1564 	int	unit, part;
1565 	int	sectorsize;	/* size of a disk sector */
1566 	int	nsects;		/* number of sectors in partition */
1567 	int	sectoff;	/* sector offset of partition */
1568 	int	totwrt;		/* total number of sectors left to write */
1569 	int	nwrt;		/* current number of sectors to write */
1570 	struct scsipi_rw_10 cmd;	/* write command */
1571 	struct scsipi_xfer *xs;	/* ... convenience */
1572 	struct scsipi_periph *periph;
1573 	struct scsipi_channel *chan;
1574 
1575 	/* Check if recursive dump; if so, punt. */
1576 	if (sddoingadump)
1577 		return (EFAULT);
1578 
1579 	/* Mark as active early. */
1580 	sddoingadump = 1;
1581 
1582 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1583 	part = SDPART(dev);
1584 
1585 	/* Check for acceptable drive number. */
1586 	sd = device_lookup_private(&sd_cd, unit);
1587 	if (sd == NULL)
1588 		return (ENXIO);
1589 
1590 	if (!device_is_active(sd->sc_dev))
1591 		return (ENODEV);
1592 
1593 	periph = sd->sc_periph;
1594 	chan = periph->periph_channel;
1595 
1596 	/* Make sure it was initialized. */
1597 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1598 		return (ENXIO);
1599 
1600 	/* Convert to disk sectors.  Request must be a multiple of size. */
1601 	lp = sd->sc_dk.dk_label;
1602 	sectorsize = lp->d_secsize;
1603 	if ((size % sectorsize) != 0)
1604 		return (EFAULT);
1605 	totwrt = size / sectorsize;
1606 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1607 
1608 	nsects = lp->d_partitions[part].p_size;
1609 	sectoff = lp->d_partitions[part].p_offset;
1610 
1611 	/* Check transfer bounds against partition size. */
1612 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1613 		return (EINVAL);
1614 
1615 	/* Offset block number to start of partition. */
1616 	blkno += sectoff;
1617 
1618 	xs = &sx;
1619 
1620 	while (totwrt > 0) {
1621 		nwrt = totwrt;		/* XXX */
1622 #ifndef	SD_DUMP_NOT_TRUSTED
1623 		/*
1624 		 *  Fill out the scsi command
1625 		 */
1626 		memset(&cmd, 0, sizeof(cmd));
1627 		cmd.opcode = WRITE_10;
1628 		_lto4b(blkno, cmd.addr);
1629 		_lto2b(nwrt, cmd.length);
1630 		/*
1631 		 * Fill out the scsipi_xfer structure
1632 		 *    Note: we cannot sleep as we may be an interrupt
1633 		 * don't use scsipi_command() as it may want to wait
1634 		 * for an xs.
1635 		 */
1636 		memset(xs, 0, sizeof(sx));
1637 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1638 		    XS_CTL_DATA_OUT;
1639 		xs->xs_status = 0;
1640 		xs->xs_periph = periph;
1641 		xs->xs_retries = SDRETRIES;
1642 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1643 		xs->cmd = (struct scsipi_generic *)&cmd;
1644 		xs->cmdlen = sizeof(cmd);
1645 		xs->resid = nwrt * sectorsize;
1646 		xs->error = XS_NOERROR;
1647 		xs->bp = 0;
1648 		xs->data = va;
1649 		xs->datalen = nwrt * sectorsize;
1650 		callout_init(&xs->xs_callout, 0);
1651 
1652 		/*
1653 		 * Pass all this info to the scsi driver.
1654 		 */
1655 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1656 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1657 		    xs->error != XS_NOERROR)
1658 			return (EIO);
1659 #else	/* SD_DUMP_NOT_TRUSTED */
1660 		/* Let's just talk about this first... */
1661 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1662 		delay(500 * 1000);	/* half a second */
1663 #endif	/* SD_DUMP_NOT_TRUSTED */
1664 
1665 		/* update block count */
1666 		totwrt -= nwrt;
1667 		blkno += nwrt;
1668 		va = (char *)va + sectorsize * nwrt;
1669 	}
1670 	sddoingadump = 0;
1671 	return (0);
1672 }
1673 
1674 static int
1675 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1676     int page, int flags, int *big)
1677 {
1678 
1679 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1680 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1681 		*big = 1;
1682 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1683 		    size + sizeof(struct scsi_mode_parameter_header_10),
1684 		    flags, SDRETRIES, 6000);
1685 	} else {
1686 		*big = 0;
1687 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1688 		    size + sizeof(struct scsi_mode_parameter_header_6),
1689 		    flags, SDRETRIES, 6000);
1690 	}
1691 }
1692 
1693 static int
1694 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1695     int flags, int big)
1696 {
1697 
1698 	if (big) {
1699 		struct scsi_mode_parameter_header_10 *header = sense;
1700 
1701 		_lto2b(0, header->data_length);
1702 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1703 		    size + sizeof(struct scsi_mode_parameter_header_10),
1704 		    flags, SDRETRIES, 6000);
1705 	} else {
1706 		struct scsi_mode_parameter_header_6 *header = sense;
1707 
1708 		header->data_length = 0;
1709 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1710 		    size + sizeof(struct scsi_mode_parameter_header_6),
1711 		    flags, SDRETRIES, 6000);
1712 	}
1713 }
1714 
1715 /*
1716  * sd_validate_blksize:
1717  *
1718  *	Validate the block size.  Print error if periph is specified,
1719  */
1720 static int
1721 sd_validate_blksize(struct scsipi_periph *periph, int len)
1722 {
1723 
1724 	switch (len) {
1725 	case 256:
1726 	case 512:
1727 	case 1024:
1728 	case 2048:
1729 	case 4096:
1730 		return 1;
1731 	}
1732 
1733 	if (periph) {
1734 		scsipi_printaddr(periph);
1735 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1736 		    (len ^ (1 << (ffs(len) - 1))) ?
1737 		    "preposterous" : "unsupported",
1738 		    len, SD_DEFAULT_BLKSIZE);
1739 	}
1740 
1741 	return 0;
1742 }
1743 
1744 /*
1745  * sd_read_capacity:
1746  *
1747  *	Find out from the device what its capacity is.
1748  */
1749 static u_int64_t
1750 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1751 {
1752 	union {
1753 		struct scsipi_read_capacity_10 cmd;
1754 		struct scsipi_read_capacity_16 cmd16;
1755 	} cmd;
1756 	union {
1757 		struct scsipi_read_capacity_10_data data;
1758 		struct scsipi_read_capacity_16_data data16;
1759 	} *datap;
1760 	uint64_t rv;
1761 
1762 	memset(&cmd, 0, sizeof(cmd));
1763 	cmd.cmd.opcode = READ_CAPACITY_10;
1764 
1765 	/*
1766 	 * Don't allocate data buffer on stack;
1767 	 * The lower driver layer might use the same stack and
1768 	 * if it uses region which is in the same cacheline,
1769 	 * cache flush ops against the data buffer won't work properly.
1770 	 */
1771 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1772 	if (datap == NULL)
1773 		return 0;
1774 
1775 	/*
1776 	 * If the command works, interpret the result as a 4 byte
1777 	 * number of blocks
1778 	 */
1779 	rv = 0;
1780 	memset(datap, 0, sizeof(datap->data));
1781 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1782 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1783 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1784 		goto out;
1785 
1786 	if (_4btol(datap->data.addr) != 0xffffffff) {
1787 		*blksize = _4btol(datap->data.length);
1788 		rv = _4btol(datap->data.addr) + 1;
1789 		goto out;
1790 	}
1791 
1792 	/*
1793 	 * Device is larger than can be reflected by READ CAPACITY (10).
1794 	 * Try READ CAPACITY (16).
1795 	 */
1796 
1797 	memset(&cmd, 0, sizeof(cmd));
1798 	cmd.cmd16.opcode = READ_CAPACITY_16;
1799 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1800 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1801 
1802 	memset(datap, 0, sizeof(datap->data16));
1803 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1804 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1805 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1806 		goto out;
1807 
1808 	*blksize = _4btol(datap->data16.length);
1809 	rv = _8btol(datap->data16.addr) + 1;
1810 
1811  out:
1812 	free(datap, M_TEMP);
1813 	return rv;
1814 }
1815 
1816 static int
1817 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1818 {
1819 	struct {
1820 		struct scsi_mode_parameter_header_6 header;
1821 		/* no block descriptor */
1822 		u_int8_t pg_code; /* page code (should be 6) */
1823 		u_int8_t pg_length; /* page length (should be 11) */
1824 		u_int8_t wcd; /* bit0: cache disable */
1825 		u_int8_t lbs[2]; /* logical block size */
1826 		u_int8_t size[5]; /* number of log. blocks */
1827 		u_int8_t pp; /* power/performance */
1828 		u_int8_t flags;
1829 		u_int8_t resvd;
1830 	} scsipi_sense;
1831 	u_int64_t blocks;
1832 	int error, blksize;
1833 
1834 	/*
1835 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1836 	 * give the same information. Do both for now, and check
1837 	 * for consistency.
1838 	 * XXX probably differs for removable media
1839 	 */
1840 	dp->blksize = SD_DEFAULT_BLKSIZE;
1841 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1842 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1843 
1844 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1845 	    &scsipi_sense.header, sizeof(scsipi_sense),
1846 	    flags, SDRETRIES, 6000);
1847 
1848 	if (error != 0)
1849 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1850 
1851 	dp->blksize = blksize;
1852 	if (!sd_validate_blksize(NULL, dp->blksize))
1853 		dp->blksize = _2btol(scsipi_sense.lbs);
1854 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1855 		dp->blksize = SD_DEFAULT_BLKSIZE;
1856 
1857 	/*
1858 	 * Create a pseudo-geometry.
1859 	 */
1860 	dp->heads = 64;
1861 	dp->sectors = 32;
1862 	dp->cyls = blocks / (dp->heads * dp->sectors);
1863 	dp->disksize = _5btol(scsipi_sense.size);
1864 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1865 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1866 		       (unsigned long long)dp->disksize,
1867 		       (unsigned long long)blocks);
1868 		dp->disksize = blocks;
1869 	}
1870 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1871 
1872 	return (SDGP_RESULT_OK);
1873 }
1874 
1875 /*
1876  * Get the scsi driver to send a full inquiry to the * device and use the
1877  * results to fill out the disk parameter structure.
1878  */
1879 static int
1880 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1881 {
1882 	u_int64_t blocks;
1883 	int error, blksize;
1884 #if 0
1885 	int i;
1886 	u_int8_t *p;
1887 #endif
1888 
1889 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1890 	    flags);
1891 	if (blocks == 0) {
1892 		struct scsipi_read_format_capacities cmd;
1893 		struct {
1894 			struct scsipi_capacity_list_header header;
1895 			struct scsipi_capacity_descriptor desc;
1896 		} __packed data;
1897 
1898 		memset(&cmd, 0, sizeof(cmd));
1899 		memset(&data, 0, sizeof(data));
1900 		cmd.opcode = READ_FORMAT_CAPACITIES;
1901 		_lto2b(sizeof(data), cmd.length);
1902 
1903 		error = scsipi_command(sd->sc_periph,
1904 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1905 		    SDRETRIES, 20000, NULL,
1906 		    flags | XS_CTL_DATA_IN);
1907 		if (error == EFTYPE) {
1908 			/* Medium Format Corrupted, handle as not formatted */
1909 			return (SDGP_RESULT_UNFORMATTED);
1910 		}
1911 		if (error || data.header.length == 0)
1912 			return (SDGP_RESULT_OFFLINE);
1913 
1914 #if 0
1915 printf("rfc: length=%d\n", data.header.length);
1916 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");
1917 #endif
1918 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1919 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1920 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1921 			break;
1922 
1923 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1924 			return (SDGP_RESULT_UNFORMATTED);
1925 
1926 		case SCSIPI_CAP_DESC_CODE_NONE:
1927 			return (SDGP_RESULT_OFFLINE);
1928 		}
1929 
1930 		dp->disksize = blocks = _4btol(data.desc.nblks);
1931 		if (blocks == 0)
1932 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1933 
1934 		blksize = _3btol(data.desc.blklen);
1935 
1936 	} else if (!sd_validate_blksize(NULL, blksize)) {
1937 		struct sd_mode_sense_data scsipi_sense;
1938 		int big, bsize;
1939 		struct scsi_general_block_descriptor *bdesc;
1940 
1941 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1942 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1943 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1944 		if (!error) {
1945 			if (big) {
1946 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1947 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1948 			} else {
1949 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1950 				bsize = scsipi_sense.header.small.blk_desc_len;
1951 			}
1952 
1953 #if 0
1954 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1955 printf("page 0 bsize=%d\n", bsize);
1956 printf("page 0 ok\n");
1957 #endif
1958 
1959 			if (bsize >= 8) {
1960 				blksize = _3btol(bdesc->blklen);
1961 			}
1962 		}
1963 	}
1964 
1965 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1966 		blksize = SD_DEFAULT_BLKSIZE;
1967 
1968 	dp->blksize = blksize;
1969 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1970 	return (0);
1971 }
1972 
1973 static int
1974 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1975 {
1976 	struct sd_mode_sense_data scsipi_sense;
1977 	int error;
1978 	int big, byte2;
1979 	size_t poffset;
1980 	union scsi_disk_pages *pages;
1981 
1982 	byte2 = SMS_DBD;
1983 again:
1984 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1985 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1986 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1987 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1988 	    flags | XS_CTL_SILENT, &big);
1989 	if (error) {
1990 		if (byte2 == SMS_DBD) {
1991 			/* No result; try once more with DBD off */
1992 			byte2 = 0;
1993 			goto again;
1994 		}
1995 		return (error);
1996 	}
1997 
1998 	if (big) {
1999 		poffset = sizeof scsipi_sense.header.big;
2000 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2001 	} else {
2002 		poffset = sizeof scsipi_sense.header.small;
2003 		poffset += scsipi_sense.header.small.blk_desc_len;
2004 	}
2005 
2006 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
2007 		return ERESTART;
2008 
2009 	pages = (void *)((u_long)&scsipi_sense + poffset);
2010 #if 0
2011 	{
2012 		size_t i;
2013 		u_int8_t *p;
2014 
2015 		printf("page 4 sense:");
2016 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2017 		    i--, p++)
2018 			printf(" %02x", *p);
2019 		printf("\n");
2020 		printf("page 4 pg_code=%d sense=%p/%p\n",
2021 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
2022 	}
2023 #endif
2024 
2025 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
2026 		return (ERESTART);
2027 
2028 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2029 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
2030 	    _3btol(pages->rigid_geometry.ncyl),
2031 	    pages->rigid_geometry.nheads,
2032 	    _2btol(pages->rigid_geometry.st_cyl_wp),
2033 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
2034 	    _2btol(pages->rigid_geometry.land_zone)));
2035 
2036 	/*
2037 	 * KLUDGE!! (for zone recorded disks)
2038 	 * give a number of sectors so that sec * trks * cyls
2039 	 * is <= disk_size
2040 	 * can lead to wasted space! THINK ABOUT THIS !
2041 	 */
2042 	dp->heads = pages->rigid_geometry.nheads;
2043 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
2044 	if (dp->heads == 0 || dp->cyls == 0)
2045 		return (ERESTART);
2046 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
2047 
2048 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2049 	if (dp->rot_rate == 0)
2050 		dp->rot_rate = 3600;
2051 
2052 #if 0
2053 printf("page 4 ok\n");
2054 #endif
2055 	return (0);
2056 }
2057 
2058 static int
2059 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
2060 {
2061 	struct sd_mode_sense_data scsipi_sense;
2062 	int error;
2063 	int big, byte2;
2064 	size_t poffset;
2065 	union scsi_disk_pages *pages;
2066 
2067 	byte2 = SMS_DBD;
2068 again:
2069 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2070 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2071 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2072 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2073 	    flags | XS_CTL_SILENT, &big);
2074 	if (error) {
2075 		if (byte2 == SMS_DBD) {
2076 			/* No result; try once more with DBD off */
2077 			byte2 = 0;
2078 			goto again;
2079 		}
2080 		return (error);
2081 	}
2082 
2083 	if (big) {
2084 		poffset = sizeof scsipi_sense.header.big;
2085 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2086 	} else {
2087 		poffset = sizeof scsipi_sense.header.small;
2088 		poffset += scsipi_sense.header.small.blk_desc_len;
2089 	}
2090 
2091 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2092 		return ERESTART;
2093 
2094 	pages = (void *)((u_long)&scsipi_sense + poffset);
2095 #if 0
2096 	{
2097 		size_t i;
2098 		u_int8_t *p;
2099 
2100 		printf("page 5 sense:");
2101 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2102 		    i--, p++)
2103 			printf(" %02x", *p);
2104 		printf("\n");
2105 		printf("page 5 pg_code=%d sense=%p/%p\n",
2106 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2107 	}
2108 #endif
2109 
2110 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2111 		return (ERESTART);
2112 
2113 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2114 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2115 	    _3btol(pages->flex_geometry.ncyl),
2116 	    pages->flex_geometry.nheads,
2117 	    pages->flex_geometry.ph_sec_tr,
2118 	    _2btol(pages->flex_geometry.bytes_s)));
2119 
2120 	dp->heads = pages->flex_geometry.nheads;
2121 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2122 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2123 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2124 		return (ERESTART);
2125 
2126 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2127 	if (dp->rot_rate == 0)
2128 		dp->rot_rate = 3600;
2129 
2130 #if 0
2131 printf("page 5 ok\n");
2132 #endif
2133 	return (0);
2134 }
2135 
2136 static int
2137 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2138 {
2139 	int error;
2140 
2141 	/*
2142 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2143 	 * cleared by the caller if necessary.
2144 	 */
2145 	if (sd->type == T_SIMPLE_DIRECT) {
2146 		error = sd_get_simplifiedparms(sd, dp, flags);
2147 		if (!error)
2148 			disk_blocksize(&sd->sc_dk, dp->blksize);
2149 		return (error);
2150 	}
2151 
2152 	error = sd_get_capacity(sd, dp, flags);
2153 	if (error)
2154 		return (error);
2155 
2156 	disk_blocksize(&sd->sc_dk, dp->blksize);
2157 
2158 	if (sd->type == T_OPTICAL)
2159 		goto page0;
2160 
2161 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2162 		if (!sd_get_parms_page5(sd, dp, flags) ||
2163 		    !sd_get_parms_page4(sd, dp, flags))
2164 			goto setprops;
2165 	} else {
2166 		if (!sd_get_parms_page4(sd, dp, flags) ||
2167 		    !sd_get_parms_page5(sd, dp, flags))
2168 			goto setprops;
2169 	}
2170 
2171 page0:
2172 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2173 	/* Try calling driver's method for figuring out geometry. */
2174 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2175 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2176 		(sd->sc_periph, dp, dp->disksize)) {
2177 		/*
2178 		 * Use adaptec standard fictitious geometry
2179 		 * this depends on which controller (e.g. 1542C is
2180 		 * different. but we have to put SOMETHING here..)
2181 		 */
2182 		dp->heads = 64;
2183 		dp->sectors = 32;
2184 		dp->cyls = dp->disksize / (64 * 32);
2185 	}
2186 	dp->rot_rate = 3600;
2187 
2188 setprops:
2189 	sd_set_geometry(sd);
2190 
2191 	return (SDGP_RESULT_OK);
2192 }
2193 
2194 static int
2195 sd_flush(struct sd_softc *sd, int flags)
2196 {
2197 	struct scsipi_periph *periph = sd->sc_periph;
2198 	struct scsi_synchronize_cache_10 cmd;
2199 
2200 	/*
2201 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2202 	 * We issue with address 0 length 0, which should be
2203 	 * interpreted by the device as "all remaining blocks
2204 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2205 	 * in the event that the command is not supported by
2206 	 * the device, and poll for completion so that we know
2207 	 * that the cache has actually been flushed.
2208 	 *
2209 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2210 	 * command, as indicated by our quirks flags.
2211 	 *
2212 	 * XXX What about older devices?
2213 	 */
2214 	if (periph->periph_version < 2 ||
2215 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2216 		return (0);
2217 
2218 	sd->flags |= SDF_FLUSHING;
2219 	memset(&cmd, 0, sizeof(cmd));
2220 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2221 
2222 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2223 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2224 }
2225 
2226 static int
2227 sd_getcache(struct sd_softc *sd, int *bitsp)
2228 {
2229 	struct scsipi_periph *periph = sd->sc_periph;
2230 	struct sd_mode_sense_data scsipi_sense;
2231 	int error, bits = 0;
2232 	int big;
2233 	union scsi_disk_pages *pages;
2234 
2235 	if (periph->periph_version < 2)
2236 		return (EOPNOTSUPP);
2237 
2238 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2239 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2240 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2241 	if (error)
2242 		return (error);
2243 
2244 	if (big)
2245 		pages = (void *)(&scsipi_sense.header.big + 1);
2246 	else
2247 		pages = (void *)(&scsipi_sense.header.small + 1);
2248 
2249 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2250 		bits |= DKCACHE_READ;
2251 	if (pages->caching_params.flags & CACHING_WCE)
2252 		bits |= DKCACHE_WRITE;
2253 	if (pages->caching_params.pg_code & PGCODE_PS)
2254 		bits |= DKCACHE_SAVE;
2255 
2256 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2257 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2258 	    sizeof(scsipi_sense.pages.caching_params),
2259 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2260 	if (error == 0) {
2261 		if (big)
2262 			pages = (void *)(&scsipi_sense.header.big + 1);
2263 		else
2264 			pages = (void *)(&scsipi_sense.header.small + 1);
2265 
2266 		if (pages->caching_params.flags & CACHING_RCD)
2267 			bits |= DKCACHE_RCHANGE;
2268 		if (pages->caching_params.flags & CACHING_WCE)
2269 			bits |= DKCACHE_WCHANGE;
2270 	}
2271 
2272 	*bitsp = bits;
2273 
2274 	return (0);
2275 }
2276 
2277 static int
2278 sd_setcache(struct sd_softc *sd, int bits)
2279 {
2280 	struct scsipi_periph *periph = sd->sc_periph;
2281 	struct sd_mode_sense_data scsipi_sense;
2282 	int error;
2283 	uint8_t oflags, byte2 = 0;
2284 	int big;
2285 	union scsi_disk_pages *pages;
2286 
2287 	if (periph->periph_version < 2)
2288 		return (EOPNOTSUPP);
2289 
2290 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2291 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2292 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2293 	if (error)
2294 		return (error);
2295 
2296 	if (big)
2297 		pages = (void *)(&scsipi_sense.header.big + 1);
2298 	else
2299 		pages = (void *)(&scsipi_sense.header.small + 1);
2300 
2301 	oflags = pages->caching_params.flags;
2302 
2303 	if (bits & DKCACHE_READ)
2304 		pages->caching_params.flags &= ~CACHING_RCD;
2305 	else
2306 		pages->caching_params.flags |= CACHING_RCD;
2307 
2308 	if (bits & DKCACHE_WRITE)
2309 		pages->caching_params.flags |= CACHING_WCE;
2310 	else
2311 		pages->caching_params.flags &= ~CACHING_WCE;
2312 
2313 	if (oflags == pages->caching_params.flags)
2314 		return (0);
2315 
2316 	pages->caching_params.pg_code &= PGCODE_MASK;
2317 
2318 	if (bits & DKCACHE_SAVE)
2319 		byte2 |= SMS_SP;
2320 
2321 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2322 	    sizeof(struct scsi_mode_page_header) +
2323 	    pages->caching_params.pg_length, 0, big));
2324 }
2325 
2326 static void
2327 sd_set_geometry(struct sd_softc *sd)
2328 {
2329 	struct disk_geom *dg = &sd->sc_dk.dk_geom;
2330 
2331 	memset(dg, 0, sizeof(*dg));
2332 
2333 	dg->dg_secperunit = sd->params.disksize;
2334 	dg->dg_secsize = sd->params.blksize;
2335 	dg->dg_nsectors = sd->params.sectors;
2336 	dg->dg_ntracks = sd->params.heads;
2337 	dg->dg_ncylinders = sd->params.cyls;
2338 
2339 	disk_set_info(sd->sc_dev, &sd->sc_dk, NULL);
2340 }
2341