xref: /netbsd-src/sys/dev/scsipi/sd.c (revision f14316bcbc544b96a93e884bc5c2b15fd60e22ae)
1 /*	$NetBSD: sd.c,v 1.308 2014/08/10 16:44:36 tls 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 1.308 2014/08/10 16:44:36 tls 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 DIOCGSTRATEGY:
1260 	    {
1261 		struct disk_strategy *dks = addr;
1262 
1263 		s = splbio();
1264 		strlcpy(dks->dks_name, bufq_getstrategyname(sd->buf_queue),
1265 		    sizeof(dks->dks_name));
1266 		splx(s);
1267 		dks->dks_paramlen = 0;
1268 
1269 		return 0;
1270 	    }
1271 
1272 	case DIOCSSTRATEGY:
1273 	    {
1274 		struct disk_strategy *dks = addr;
1275 		struct bufq_state *new;
1276 		struct bufq_state *old;
1277 
1278 		if ((flag & FWRITE) == 0) {
1279 			return EBADF;
1280 		}
1281 
1282 		if (dks->dks_param != NULL) {
1283 			return EINVAL;
1284 		}
1285 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1286 		error = bufq_alloc(&new, dks->dks_name,
1287 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1288 		if (error) {
1289 			return error;
1290 		}
1291 		s = splbio();
1292 		old = sd->buf_queue;
1293 		bufq_move(new, old);
1294 		sd->buf_queue = new;
1295 		splx(s);
1296 		bufq_free(old);
1297 
1298 		return 0;
1299 	    }
1300 
1301 	default:
1302 		if (part != RAW_PART)
1303 			return (ENOTTY);
1304 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1305 	}
1306 
1307 #ifdef DIAGNOSTIC
1308 	panic("sdioctl: impossible");
1309 #endif
1310 }
1311 
1312 static void
1313 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1314 {
1315 
1316 	memset(lp, 0, sizeof(struct disklabel));
1317 
1318 	lp->d_secsize = sd->params.blksize;
1319 	lp->d_ntracks = sd->params.heads;
1320 	lp->d_nsectors = sd->params.sectors;
1321 	lp->d_ncylinders = sd->params.cyls;
1322 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1323 
1324 	switch (SCSIPI_BUSTYPE_TYPE(scsipi_periph_bustype(sd->sc_periph))) {
1325 	case SCSIPI_BUSTYPE_SCSI:
1326 		lp->d_type = DTYPE_SCSI;
1327 		break;
1328 	case SCSIPI_BUSTYPE_ATAPI:
1329 		lp->d_type = DTYPE_ATAPI;
1330 		break;
1331 	}
1332 	/*
1333 	 * XXX
1334 	 * We could probe the mode pages to figure out what kind of disc it is.
1335 	 * Is this worthwhile?
1336 	 */
1337 	strncpy(lp->d_typename, sd->name, 16);
1338 	strncpy(lp->d_packname, "fictitious", 16);
1339 	if (sd->params.disksize > UINT32_MAX)
1340 		lp->d_secperunit = UINT32_MAX;
1341 	else
1342 		lp->d_secperunit = sd->params.disksize;
1343 	lp->d_rpm = sd->params.rot_rate;
1344 	lp->d_interleave = 1;
1345 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1346 	    D_REMOVABLE : 0;
1347 
1348 	lp->d_partitions[RAW_PART].p_offset = 0;
1349 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1350 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1351 	lp->d_npartitions = RAW_PART + 1;
1352 
1353 	lp->d_magic = DISKMAGIC;
1354 	lp->d_magic2 = DISKMAGIC;
1355 	lp->d_checksum = dkcksum(lp);
1356 }
1357 
1358 
1359 /*
1360  * Load the label information on the named device
1361  */
1362 static int
1363 sdgetdisklabel(struct sd_softc *sd)
1364 {
1365 	struct disklabel *lp = sd->sc_dk.dk_label;
1366 	const char *errstring;
1367 
1368 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1369 
1370 	sdgetdefaultlabel(sd, lp);
1371 
1372 	if (lp->d_secpercyl == 0) {
1373 		lp->d_secpercyl = 100;
1374 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1375 	}
1376 
1377 	/*
1378 	 * Call the generic disklabel extraction routine
1379 	 */
1380 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1381 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1382 	if (errstring) {
1383 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1384 		return EIO;
1385 	}
1386 	return 0;
1387 }
1388 
1389 static bool
1390 sd_shutdown(device_t self, int how)
1391 {
1392 	struct sd_softc *sd = device_private(self);
1393 
1394 	/*
1395 	 * If the disk cache needs to be flushed, and the disk supports
1396 	 * it, flush it.  We're cold at this point, so we poll for
1397 	 * completion.
1398 	 */
1399 	if ((sd->flags & SDF_DIRTY) != 0) {
1400 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1401 			aprint_error_dev(sd->sc_dev,
1402 				"cache synchronization failed\n");
1403 			sd->flags &= ~SDF_FLUSHING;
1404 		} else
1405 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1406 	}
1407 	return true;
1408 }
1409 
1410 static bool
1411 sd_suspend(device_t dv, const pmf_qual_t *qual)
1412 {
1413 	return sd_shutdown(dv, boothowto); /* XXX no need to poll */
1414 }
1415 
1416 /*
1417  * Check Errors
1418  */
1419 static int
1420 sd_interpret_sense(struct scsipi_xfer *xs)
1421 {
1422 	struct scsipi_periph *periph = xs->xs_periph;
1423 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1424 	struct sd_softc *sd = device_private(periph->periph_dev);
1425 	int s, error, retval = EJUSTRETURN;
1426 
1427 	/*
1428 	 * If the periph is already recovering, just do the normal
1429 	 * error processing.
1430 	 */
1431 	if (periph->periph_flags & PERIPH_RECOVERING)
1432 		return (retval);
1433 
1434 	/*
1435 	 * Ignore errors from accessing illegal fields (e.g. trying to
1436 	 * lock the door of a digicam, which doesn't have a door that
1437 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1438 	 */
1439 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1440 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1441 	    sense->asc == 0x24 &&
1442 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1443 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1444 			scsipi_printaddr(periph);
1445 			printf("no door lock\n");
1446 		}
1447 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1448 		return (retval);
1449 	}
1450 
1451 
1452 
1453 	/*
1454 	 * If the device is not open yet, let the generic code handle it.
1455 	 */
1456 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1457 		return (retval);
1458 
1459 	/*
1460 	 * If it isn't a extended or extended/deferred error, let
1461 	 * the generic code handle it.
1462 	 */
1463 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1464 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1465 		return (retval);
1466 
1467 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1468 	    sense->asc == 0x4) {
1469 		if (sense->ascq == 0x01)	{
1470 			/*
1471 			 * Unit In The Process Of Becoming Ready.
1472 			 */
1473 			printf("%s: waiting for pack to spin up...\n",
1474 			    device_xname(sd->sc_dev));
1475 			if (!callout_pending(&periph->periph_callout))
1476 				scsipi_periph_freeze(periph, 1);
1477 			callout_reset(&periph->periph_callout,
1478 			    5 * hz, scsipi_periph_timed_thaw, periph);
1479 			retval = ERESTART;
1480 		} else if (sense->ascq == 0x02) {
1481 			printf("%s: pack is stopped, restarting...\n",
1482 			    device_xname(sd->sc_dev));
1483 			s = splbio();
1484 			periph->periph_flags |= PERIPH_RECOVERING;
1485 			splx(s);
1486 			error = scsipi_start(periph, SSS_START,
1487 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1488 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1489 			if (error) {
1490 				aprint_error_dev(sd->sc_dev,
1491 					"unable to restart pack\n");
1492 				retval = error;
1493 			} else
1494 				retval = ERESTART;
1495 			s = splbio();
1496 			periph->periph_flags &= ~PERIPH_RECOVERING;
1497 			splx(s);
1498 		}
1499 	}
1500 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1501 	    sense->asc == 0x31 &&
1502 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1503 		/* Medium Format Corrupted */
1504 		retval = EFTYPE;
1505 	}
1506 	return (retval);
1507 }
1508 
1509 
1510 static int
1511 sdsize(dev_t dev)
1512 {
1513 	struct sd_softc *sd;
1514 	int part, unit, omask;
1515 	int size;
1516 
1517 	unit = SDUNIT(dev);
1518 	sd = device_lookup_private(&sd_cd, unit);
1519 	if (sd == NULL)
1520 		return (-1);
1521 
1522 	if (!device_is_active(sd->sc_dev))
1523 		return (-1);
1524 
1525 	part = SDPART(dev);
1526 	omask = sd->sc_dk.dk_openmask & (1 << part);
1527 
1528 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1529 		return (-1);
1530 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1531 		size = -1;
1532 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1533 		size = -1;
1534 	else
1535 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1536 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1537 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1538 		return (-1);
1539 	return (size);
1540 }
1541 
1542 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1543 static struct scsipi_xfer sx;
1544 static int sddoingadump;
1545 
1546 /*
1547  * dump all of physical memory into the partition specified, starting
1548  * at offset 'dumplo' into the partition.
1549  */
1550 static int
1551 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1552 {
1553 	struct sd_softc *sd;	/* disk unit to do the I/O */
1554 	struct disklabel *lp;	/* disk's disklabel */
1555 	int	unit, part;
1556 	int	sectorsize;	/* size of a disk sector */
1557 	int	nsects;		/* number of sectors in partition */
1558 	int	sectoff;	/* sector offset of partition */
1559 	int	totwrt;		/* total number of sectors left to write */
1560 	int	nwrt;		/* current number of sectors to write */
1561 	struct scsipi_rw_10 cmd;	/* write command */
1562 	struct scsipi_xfer *xs;	/* ... convenience */
1563 	struct scsipi_periph *periph;
1564 	struct scsipi_channel *chan;
1565 
1566 	/* Check if recursive dump; if so, punt. */
1567 	if (sddoingadump)
1568 		return (EFAULT);
1569 
1570 	/* Mark as active early. */
1571 	sddoingadump = 1;
1572 
1573 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1574 	part = SDPART(dev);
1575 
1576 	/* Check for acceptable drive number. */
1577 	sd = device_lookup_private(&sd_cd, unit);
1578 	if (sd == NULL)
1579 		return (ENXIO);
1580 
1581 	if (!device_is_active(sd->sc_dev))
1582 		return (ENODEV);
1583 
1584 	periph = sd->sc_periph;
1585 	chan = periph->periph_channel;
1586 
1587 	/* Make sure it was initialized. */
1588 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1589 		return (ENXIO);
1590 
1591 	/* Convert to disk sectors.  Request must be a multiple of size. */
1592 	lp = sd->sc_dk.dk_label;
1593 	sectorsize = lp->d_secsize;
1594 	if ((size % sectorsize) != 0)
1595 		return (EFAULT);
1596 	totwrt = size / sectorsize;
1597 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1598 
1599 	nsects = lp->d_partitions[part].p_size;
1600 	sectoff = lp->d_partitions[part].p_offset;
1601 
1602 	/* Check transfer bounds against partition size. */
1603 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1604 		return (EINVAL);
1605 
1606 	/* Offset block number to start of partition. */
1607 	blkno += sectoff;
1608 
1609 	xs = &sx;
1610 
1611 	while (totwrt > 0) {
1612 		nwrt = totwrt;		/* XXX */
1613 #ifndef	SD_DUMP_NOT_TRUSTED
1614 		/*
1615 		 *  Fill out the scsi command
1616 		 */
1617 		memset(&cmd, 0, sizeof(cmd));
1618 		cmd.opcode = WRITE_10;
1619 		_lto4b(blkno, cmd.addr);
1620 		_lto2b(nwrt, cmd.length);
1621 		/*
1622 		 * Fill out the scsipi_xfer structure
1623 		 *    Note: we cannot sleep as we may be an interrupt
1624 		 * don't use scsipi_command() as it may want to wait
1625 		 * for an xs.
1626 		 */
1627 		memset(xs, 0, sizeof(sx));
1628 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1629 		    XS_CTL_DATA_OUT;
1630 		xs->xs_status = 0;
1631 		xs->xs_periph = periph;
1632 		xs->xs_retries = SDRETRIES;
1633 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1634 		xs->cmd = (struct scsipi_generic *)&cmd;
1635 		xs->cmdlen = sizeof(cmd);
1636 		xs->resid = nwrt * sectorsize;
1637 		xs->error = XS_NOERROR;
1638 		xs->bp = 0;
1639 		xs->data = va;
1640 		xs->datalen = nwrt * sectorsize;
1641 		callout_init(&xs->xs_callout, 0);
1642 
1643 		/*
1644 		 * Pass all this info to the scsi driver.
1645 		 */
1646 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1647 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1648 		    xs->error != XS_NOERROR)
1649 			return (EIO);
1650 #else	/* SD_DUMP_NOT_TRUSTED */
1651 		/* Let's just talk about this first... */
1652 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1653 		delay(500 * 1000);	/* half a second */
1654 #endif	/* SD_DUMP_NOT_TRUSTED */
1655 
1656 		/* update block count */
1657 		totwrt -= nwrt;
1658 		blkno += nwrt;
1659 		va = (char *)va + sectorsize * nwrt;
1660 	}
1661 	sddoingadump = 0;
1662 	return (0);
1663 }
1664 
1665 static int
1666 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1667     int page, int flags, int *big)
1668 {
1669 
1670 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1671 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1672 		*big = 1;
1673 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1674 		    size + sizeof(struct scsi_mode_parameter_header_10),
1675 		    flags, SDRETRIES, 6000);
1676 	} else {
1677 		*big = 0;
1678 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1679 		    size + sizeof(struct scsi_mode_parameter_header_6),
1680 		    flags, SDRETRIES, 6000);
1681 	}
1682 }
1683 
1684 static int
1685 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1686     int flags, int big)
1687 {
1688 
1689 	if (big) {
1690 		struct scsi_mode_parameter_header_10 *header = sense;
1691 
1692 		_lto2b(0, header->data_length);
1693 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1694 		    size + sizeof(struct scsi_mode_parameter_header_10),
1695 		    flags, SDRETRIES, 6000);
1696 	} else {
1697 		struct scsi_mode_parameter_header_6 *header = sense;
1698 
1699 		header->data_length = 0;
1700 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1701 		    size + sizeof(struct scsi_mode_parameter_header_6),
1702 		    flags, SDRETRIES, 6000);
1703 	}
1704 }
1705 
1706 /*
1707  * sd_validate_blksize:
1708  *
1709  *	Validate the block size.  Print error if periph is specified,
1710  */
1711 static int
1712 sd_validate_blksize(struct scsipi_periph *periph, int len)
1713 {
1714 
1715 	switch (len) {
1716 	case 256:
1717 	case 512:
1718 	case 1024:
1719 	case 2048:
1720 	case 4096:
1721 		return 1;
1722 	}
1723 
1724 	if (periph) {
1725 		scsipi_printaddr(periph);
1726 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1727 		    (len ^ (1 << (ffs(len) - 1))) ?
1728 		    "preposterous" : "unsupported",
1729 		    len, SD_DEFAULT_BLKSIZE);
1730 	}
1731 
1732 	return 0;
1733 }
1734 
1735 /*
1736  * sd_read_capacity:
1737  *
1738  *	Find out from the device what its capacity is.
1739  */
1740 static u_int64_t
1741 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1742 {
1743 	union {
1744 		struct scsipi_read_capacity_10 cmd;
1745 		struct scsipi_read_capacity_16 cmd16;
1746 	} cmd;
1747 	union {
1748 		struct scsipi_read_capacity_10_data data;
1749 		struct scsipi_read_capacity_16_data data16;
1750 	} *datap;
1751 	uint64_t rv;
1752 
1753 	memset(&cmd, 0, sizeof(cmd));
1754 	cmd.cmd.opcode = READ_CAPACITY_10;
1755 
1756 	/*
1757 	 * Don't allocate data buffer on stack;
1758 	 * The lower driver layer might use the same stack and
1759 	 * if it uses region which is in the same cacheline,
1760 	 * cache flush ops against the data buffer won't work properly.
1761 	 */
1762 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1763 	if (datap == NULL)
1764 		return 0;
1765 
1766 	/*
1767 	 * If the command works, interpret the result as a 4 byte
1768 	 * number of blocks
1769 	 */
1770 	rv = 0;
1771 	memset(datap, 0, sizeof(datap->data));
1772 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1773 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1774 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1775 		goto out;
1776 
1777 	if (_4btol(datap->data.addr) != 0xffffffff) {
1778 		*blksize = _4btol(datap->data.length);
1779 		rv = _4btol(datap->data.addr) + 1;
1780 		goto out;
1781 	}
1782 
1783 	/*
1784 	 * Device is larger than can be reflected by READ CAPACITY (10).
1785 	 * Try READ CAPACITY (16).
1786 	 */
1787 
1788 	memset(&cmd, 0, sizeof(cmd));
1789 	cmd.cmd16.opcode = READ_CAPACITY_16;
1790 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1791 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1792 
1793 	memset(datap, 0, sizeof(datap->data16));
1794 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1795 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1796 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1797 		goto out;
1798 
1799 	*blksize = _4btol(datap->data16.length);
1800 	rv = _8btol(datap->data16.addr) + 1;
1801 
1802  out:
1803 	free(datap, M_TEMP);
1804 	return rv;
1805 }
1806 
1807 static int
1808 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1809 {
1810 	struct {
1811 		struct scsi_mode_parameter_header_6 header;
1812 		/* no block descriptor */
1813 		u_int8_t pg_code; /* page code (should be 6) */
1814 		u_int8_t pg_length; /* page length (should be 11) */
1815 		u_int8_t wcd; /* bit0: cache disable */
1816 		u_int8_t lbs[2]; /* logical block size */
1817 		u_int8_t size[5]; /* number of log. blocks */
1818 		u_int8_t pp; /* power/performance */
1819 		u_int8_t flags;
1820 		u_int8_t resvd;
1821 	} scsipi_sense;
1822 	u_int64_t blocks;
1823 	int error, blksize;
1824 
1825 	/*
1826 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1827 	 * give the same information. Do both for now, and check
1828 	 * for consistency.
1829 	 * XXX probably differs for removable media
1830 	 */
1831 	dp->blksize = SD_DEFAULT_BLKSIZE;
1832 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1833 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1834 
1835 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1836 	    &scsipi_sense.header, sizeof(scsipi_sense),
1837 	    flags, SDRETRIES, 6000);
1838 
1839 	if (error != 0)
1840 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1841 
1842 	dp->blksize = blksize;
1843 	if (!sd_validate_blksize(NULL, dp->blksize))
1844 		dp->blksize = _2btol(scsipi_sense.lbs);
1845 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1846 		dp->blksize = SD_DEFAULT_BLKSIZE;
1847 
1848 	/*
1849 	 * Create a pseudo-geometry.
1850 	 */
1851 	dp->heads = 64;
1852 	dp->sectors = 32;
1853 	dp->cyls = blocks / (dp->heads * dp->sectors);
1854 	dp->disksize = _5btol(scsipi_sense.size);
1855 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1856 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1857 		       (unsigned long long)dp->disksize,
1858 		       (unsigned long long)blocks);
1859 		dp->disksize = blocks;
1860 	}
1861 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1862 
1863 	return (SDGP_RESULT_OK);
1864 }
1865 
1866 /*
1867  * Get the scsi driver to send a full inquiry to the * device and use the
1868  * results to fill out the disk parameter structure.
1869  */
1870 static int
1871 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1872 {
1873 	u_int64_t blocks;
1874 	int error, blksize;
1875 #if 0
1876 	int i;
1877 	u_int8_t *p;
1878 #endif
1879 
1880 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1881 	    flags);
1882 	if (blocks == 0) {
1883 		struct scsipi_read_format_capacities cmd;
1884 		struct {
1885 			struct scsipi_capacity_list_header header;
1886 			struct scsipi_capacity_descriptor desc;
1887 		} __packed data;
1888 
1889 		memset(&cmd, 0, sizeof(cmd));
1890 		memset(&data, 0, sizeof(data));
1891 		cmd.opcode = READ_FORMAT_CAPACITIES;
1892 		_lto2b(sizeof(data), cmd.length);
1893 
1894 		error = scsipi_command(sd->sc_periph,
1895 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1896 		    SDRETRIES, 20000, NULL,
1897 		    flags | XS_CTL_DATA_IN);
1898 		if (error == EFTYPE) {
1899 			/* Medium Format Corrupted, handle as not formatted */
1900 			return (SDGP_RESULT_UNFORMATTED);
1901 		}
1902 		if (error || data.header.length == 0)
1903 			return (SDGP_RESULT_OFFLINE);
1904 
1905 #if 0
1906 printf("rfc: length=%d\n", data.header.length);
1907 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");
1908 #endif
1909 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1910 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1911 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1912 			break;
1913 
1914 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1915 			return (SDGP_RESULT_UNFORMATTED);
1916 
1917 		case SCSIPI_CAP_DESC_CODE_NONE:
1918 			return (SDGP_RESULT_OFFLINE);
1919 		}
1920 
1921 		dp->disksize = blocks = _4btol(data.desc.nblks);
1922 		if (blocks == 0)
1923 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1924 
1925 		blksize = _3btol(data.desc.blklen);
1926 
1927 	} else if (!sd_validate_blksize(NULL, blksize)) {
1928 		struct sd_mode_sense_data scsipi_sense;
1929 		int big, bsize;
1930 		struct scsi_general_block_descriptor *bdesc;
1931 
1932 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1933 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1934 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1935 		if (!error) {
1936 			if (big) {
1937 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1938 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1939 			} else {
1940 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1941 				bsize = scsipi_sense.header.small.blk_desc_len;
1942 			}
1943 
1944 #if 0
1945 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1946 printf("page 0 bsize=%d\n", bsize);
1947 printf("page 0 ok\n");
1948 #endif
1949 
1950 			if (bsize >= 8) {
1951 				blksize = _3btol(bdesc->blklen);
1952 			}
1953 		}
1954 	}
1955 
1956 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1957 		blksize = SD_DEFAULT_BLKSIZE;
1958 
1959 	dp->blksize = blksize;
1960 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1961 	return (0);
1962 }
1963 
1964 static int
1965 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1966 {
1967 	struct sd_mode_sense_data scsipi_sense;
1968 	int error;
1969 	int big, byte2;
1970 	size_t poffset;
1971 	union scsi_disk_pages *pages;
1972 
1973 	byte2 = SMS_DBD;
1974 again:
1975 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1976 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1977 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1978 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1979 	    flags | XS_CTL_SILENT, &big);
1980 	if (error) {
1981 		if (byte2 == SMS_DBD) {
1982 			/* No result; try once more with DBD off */
1983 			byte2 = 0;
1984 			goto again;
1985 		}
1986 		return (error);
1987 	}
1988 
1989 	if (big) {
1990 		poffset = sizeof scsipi_sense.header.big;
1991 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1992 	} else {
1993 		poffset = sizeof scsipi_sense.header.small;
1994 		poffset += scsipi_sense.header.small.blk_desc_len;
1995 	}
1996 
1997 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1998 		return ERESTART;
1999 
2000 	pages = (void *)((u_long)&scsipi_sense + poffset);
2001 #if 0
2002 	{
2003 		size_t i;
2004 		u_int8_t *p;
2005 
2006 		printf("page 4 sense:");
2007 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2008 		    i--, p++)
2009 			printf(" %02x", *p);
2010 		printf("\n");
2011 		printf("page 4 pg_code=%d sense=%p/%p\n",
2012 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
2013 	}
2014 #endif
2015 
2016 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
2017 		return (ERESTART);
2018 
2019 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2020 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
2021 	    _3btol(pages->rigid_geometry.ncyl),
2022 	    pages->rigid_geometry.nheads,
2023 	    _2btol(pages->rigid_geometry.st_cyl_wp),
2024 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
2025 	    _2btol(pages->rigid_geometry.land_zone)));
2026 
2027 	/*
2028 	 * KLUDGE!! (for zone recorded disks)
2029 	 * give a number of sectors so that sec * trks * cyls
2030 	 * is <= disk_size
2031 	 * can lead to wasted space! THINK ABOUT THIS !
2032 	 */
2033 	dp->heads = pages->rigid_geometry.nheads;
2034 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
2035 	if (dp->heads == 0 || dp->cyls == 0)
2036 		return (ERESTART);
2037 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
2038 
2039 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2040 	if (dp->rot_rate == 0)
2041 		dp->rot_rate = 3600;
2042 
2043 #if 0
2044 printf("page 4 ok\n");
2045 #endif
2046 	return (0);
2047 }
2048 
2049 static int
2050 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
2051 {
2052 	struct sd_mode_sense_data scsipi_sense;
2053 	int error;
2054 	int big, byte2;
2055 	size_t poffset;
2056 	union scsi_disk_pages *pages;
2057 
2058 	byte2 = SMS_DBD;
2059 again:
2060 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2061 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2062 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2063 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2064 	    flags | XS_CTL_SILENT, &big);
2065 	if (error) {
2066 		if (byte2 == SMS_DBD) {
2067 			/* No result; try once more with DBD off */
2068 			byte2 = 0;
2069 			goto again;
2070 		}
2071 		return (error);
2072 	}
2073 
2074 	if (big) {
2075 		poffset = sizeof scsipi_sense.header.big;
2076 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2077 	} else {
2078 		poffset = sizeof scsipi_sense.header.small;
2079 		poffset += scsipi_sense.header.small.blk_desc_len;
2080 	}
2081 
2082 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2083 		return ERESTART;
2084 
2085 	pages = (void *)((u_long)&scsipi_sense + poffset);
2086 #if 0
2087 	{
2088 		size_t i;
2089 		u_int8_t *p;
2090 
2091 		printf("page 5 sense:");
2092 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2093 		    i--, p++)
2094 			printf(" %02x", *p);
2095 		printf("\n");
2096 		printf("page 5 pg_code=%d sense=%p/%p\n",
2097 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2098 	}
2099 #endif
2100 
2101 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2102 		return (ERESTART);
2103 
2104 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2105 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2106 	    _3btol(pages->flex_geometry.ncyl),
2107 	    pages->flex_geometry.nheads,
2108 	    pages->flex_geometry.ph_sec_tr,
2109 	    _2btol(pages->flex_geometry.bytes_s)));
2110 
2111 	dp->heads = pages->flex_geometry.nheads;
2112 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2113 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2114 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2115 		return (ERESTART);
2116 
2117 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2118 	if (dp->rot_rate == 0)
2119 		dp->rot_rate = 3600;
2120 
2121 #if 0
2122 printf("page 5 ok\n");
2123 #endif
2124 	return (0);
2125 }
2126 
2127 static int
2128 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2129 {
2130 	int error;
2131 
2132 	/*
2133 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2134 	 * cleared by the caller if necessary.
2135 	 */
2136 	if (sd->type == T_SIMPLE_DIRECT) {
2137 		error = sd_get_simplifiedparms(sd, dp, flags);
2138 		if (!error)
2139 			disk_blocksize(&sd->sc_dk, dp->blksize);
2140 		return (error);
2141 	}
2142 
2143 	error = sd_get_capacity(sd, dp, flags);
2144 	if (error)
2145 		return (error);
2146 
2147 	disk_blocksize(&sd->sc_dk, dp->blksize);
2148 
2149 	if (sd->type == T_OPTICAL)
2150 		goto page0;
2151 
2152 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2153 		if (!sd_get_parms_page5(sd, dp, flags) ||
2154 		    !sd_get_parms_page4(sd, dp, flags))
2155 			goto setprops;
2156 	} else {
2157 		if (!sd_get_parms_page4(sd, dp, flags) ||
2158 		    !sd_get_parms_page5(sd, dp, flags))
2159 			goto setprops;
2160 	}
2161 
2162 page0:
2163 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2164 	/* Try calling driver's method for figuring out geometry. */
2165 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2166 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2167 		(sd->sc_periph, dp, dp->disksize)) {
2168 		/*
2169 		 * Use adaptec standard fictitious geometry
2170 		 * this depends on which controller (e.g. 1542C is
2171 		 * different. but we have to put SOMETHING here..)
2172 		 */
2173 		dp->heads = 64;
2174 		dp->sectors = 32;
2175 		dp->cyls = dp->disksize / (64 * 32);
2176 	}
2177 	dp->rot_rate = 3600;
2178 
2179 setprops:
2180 	sd_set_geometry(sd);
2181 
2182 	return (SDGP_RESULT_OK);
2183 }
2184 
2185 static int
2186 sd_flush(struct sd_softc *sd, int flags)
2187 {
2188 	struct scsipi_periph *periph = sd->sc_periph;
2189 	struct scsi_synchronize_cache_10 cmd;
2190 
2191 	/*
2192 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2193 	 * We issue with address 0 length 0, which should be
2194 	 * interpreted by the device as "all remaining blocks
2195 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2196 	 * in the event that the command is not supported by
2197 	 * the device, and poll for completion so that we know
2198 	 * that the cache has actually been flushed.
2199 	 *
2200 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2201 	 * command, as indicated by our quirks flags.
2202 	 *
2203 	 * XXX What about older devices?
2204 	 */
2205 	if (periph->periph_version < 2 ||
2206 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2207 		return (0);
2208 
2209 	sd->flags |= SDF_FLUSHING;
2210 	memset(&cmd, 0, sizeof(cmd));
2211 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2212 
2213 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2214 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2215 }
2216 
2217 static int
2218 sd_getcache(struct sd_softc *sd, int *bitsp)
2219 {
2220 	struct scsipi_periph *periph = sd->sc_periph;
2221 	struct sd_mode_sense_data scsipi_sense;
2222 	int error, bits = 0;
2223 	int big;
2224 	union scsi_disk_pages *pages;
2225 
2226 	if (periph->periph_version < 2)
2227 		return (EOPNOTSUPP);
2228 
2229 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2230 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2231 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2232 	if (error)
2233 		return (error);
2234 
2235 	if (big)
2236 		pages = (void *)(&scsipi_sense.header.big + 1);
2237 	else
2238 		pages = (void *)(&scsipi_sense.header.small + 1);
2239 
2240 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2241 		bits |= DKCACHE_READ;
2242 	if (pages->caching_params.flags & CACHING_WCE)
2243 		bits |= DKCACHE_WRITE;
2244 	if (pages->caching_params.pg_code & PGCODE_PS)
2245 		bits |= DKCACHE_SAVE;
2246 
2247 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2248 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2249 	    sizeof(scsipi_sense.pages.caching_params),
2250 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2251 	if (error == 0) {
2252 		if (big)
2253 			pages = (void *)(&scsipi_sense.header.big + 1);
2254 		else
2255 			pages = (void *)(&scsipi_sense.header.small + 1);
2256 
2257 		if (pages->caching_params.flags & CACHING_RCD)
2258 			bits |= DKCACHE_RCHANGE;
2259 		if (pages->caching_params.flags & CACHING_WCE)
2260 			bits |= DKCACHE_WCHANGE;
2261 	}
2262 
2263 	*bitsp = bits;
2264 
2265 	return (0);
2266 }
2267 
2268 static int
2269 sd_setcache(struct sd_softc *sd, int bits)
2270 {
2271 	struct scsipi_periph *periph = sd->sc_periph;
2272 	struct sd_mode_sense_data scsipi_sense;
2273 	int error;
2274 	uint8_t oflags, byte2 = 0;
2275 	int big;
2276 	union scsi_disk_pages *pages;
2277 
2278 	if (periph->periph_version < 2)
2279 		return (EOPNOTSUPP);
2280 
2281 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2282 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2283 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2284 	if (error)
2285 		return (error);
2286 
2287 	if (big)
2288 		pages = (void *)(&scsipi_sense.header.big + 1);
2289 	else
2290 		pages = (void *)(&scsipi_sense.header.small + 1);
2291 
2292 	oflags = pages->caching_params.flags;
2293 
2294 	if (bits & DKCACHE_READ)
2295 		pages->caching_params.flags &= ~CACHING_RCD;
2296 	else
2297 		pages->caching_params.flags |= CACHING_RCD;
2298 
2299 	if (bits & DKCACHE_WRITE)
2300 		pages->caching_params.flags |= CACHING_WCE;
2301 	else
2302 		pages->caching_params.flags &= ~CACHING_WCE;
2303 
2304 	if (oflags == pages->caching_params.flags)
2305 		return (0);
2306 
2307 	pages->caching_params.pg_code &= PGCODE_MASK;
2308 
2309 	if (bits & DKCACHE_SAVE)
2310 		byte2 |= SMS_SP;
2311 
2312 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2313 	    sizeof(struct scsi_mode_page_header) +
2314 	    pages->caching_params.pg_length, 0, big));
2315 }
2316 
2317 static void
2318 sd_set_geometry(struct sd_softc *sd)
2319 {
2320 	struct disk_geom *dg = &sd->sc_dk.dk_geom;
2321 
2322 	memset(dg, 0, sizeof(*dg));
2323 
2324 	dg->dg_secperunit = sd->params.disksize;
2325 	dg->dg_secsize = sd->params.blksize;
2326 	dg->dg_nsectors = sd->params.sectors;
2327 	dg->dg_ntracks = sd->params.heads;
2328 	dg->dg_ncylinders = sd->params.cyls;
2329 
2330 	disk_set_info(sd->sc_dev, &sd->sc_dk, NULL);
2331 }
2332