xref: /netbsd-src/sys/dev/scsipi/sd.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: sd.c,v 1.315 2015/04/13 16:33:25 riastradh 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 1.315 2015/04/13 16:33:25 riastradh 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/rndsource.h>
73 
74 #include <dev/scsipi/scsi_spc.h>
75 #include <dev/scsipi/scsipi_all.h>
76 #include <dev/scsipi/scsi_all.h>
77 #include <dev/scsipi/scsipi_disk.h>
78 #include <dev/scsipi/scsi_disk.h>
79 #include <dev/scsipi/scsiconf.h>
80 #include <dev/scsipi/scsipi_base.h>
81 #include <dev/scsipi/sdvar.h>
82 
83 #include <prop/proplib.h>
84 
85 #define	SDUNIT(dev)			DISKUNIT(dev)
86 #define	SDPART(dev)			DISKPART(dev)
87 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
88 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
89 
90 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
91 
92 #define	SD_DEFAULT_BLKSIZE	512
93 
94 static void	sdminphys(struct buf *);
95 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
96 static int	sdgetdisklabel(struct sd_softc *);
97 static void	sdstart(struct scsipi_periph *);
98 static void	sdrestart(void *);
99 static void	sddone(struct scsipi_xfer *, int);
100 static bool	sd_suspend(device_t, const pmf_qual_t *);
101 static bool	sd_shutdown(device_t, int);
102 static int	sd_interpret_sense(struct scsipi_xfer *);
103 static int	sdlastclose(device_t);
104 
105 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
106 		    int, int *);
107 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
108 		    int);
109 static int	sd_validate_blksize(struct scsipi_periph *, int);
110 static u_int64_t sd_read_capacity(struct scsipi_periph *, int *, int flags);
111 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
112 		    int);
113 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
114 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
115 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
116 		    int);
117 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
118 		    int);
119 
120 static int	sd_flush(struct sd_softc *, int);
121 static int	sd_getcache(struct sd_softc *, int *);
122 static int	sd_setcache(struct sd_softc *, int);
123 
124 static int	sdmatch(device_t, cfdata_t, void *);
125 static void	sdattach(device_t, device_t, void *);
126 static int	sddetach(device_t, int);
127 static void	sd_set_geometry(struct sd_softc *);
128 
129 CFATTACH_DECL3_NEW(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
130     NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
131 
132 extern struct cfdriver sd_cd;
133 
134 static const struct scsipi_inquiry_pattern sd_patterns[] = {
135 	{T_DIRECT, T_FIXED,
136 	 "",         "",                 ""},
137 	{T_DIRECT, T_REMOV,
138 	 "",         "",                 ""},
139 	{T_OPTICAL, T_FIXED,
140 	 "",         "",                 ""},
141 	{T_OPTICAL, T_REMOV,
142 	 "",         "",                 ""},
143 	{T_SIMPLE_DIRECT, T_FIXED,
144 	 "",         "",                 ""},
145 	{T_SIMPLE_DIRECT, T_REMOV,
146 	 "",         "",                 ""},
147 };
148 
149 static dev_type_open(sdopen);
150 static dev_type_close(sdclose);
151 static dev_type_read(sdread);
152 static dev_type_write(sdwrite);
153 static dev_type_ioctl(sdioctl);
154 static dev_type_strategy(sdstrategy);
155 static dev_type_dump(sddump);
156 static dev_type_size(sdsize);
157 
158 const struct bdevsw sd_bdevsw = {
159 	.d_open = sdopen,
160 	.d_close = sdclose,
161 	.d_strategy = sdstrategy,
162 	.d_ioctl = sdioctl,
163 	.d_dump = sddump,
164 	.d_psize = sdsize,
165 	.d_discard = nodiscard,
166 	.d_flag = D_DISK
167 };
168 
169 const struct cdevsw sd_cdevsw = {
170 	.d_open = sdopen,
171 	.d_close = sdclose,
172 	.d_read = sdread,
173 	.d_write = sdwrite,
174 	.d_ioctl = sdioctl,
175 	.d_stop = nostop,
176 	.d_tty = notty,
177 	.d_poll = nopoll,
178 	.d_mmap = nommap,
179 	.d_kqfilter = nokqfilter,
180 	.d_discard = nodiscard,
181 	.d_flag = D_DISK
182 };
183 
184 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
185 
186 static const struct scsipi_periphsw sd_switch = {
187 	sd_interpret_sense,	/* check our error handler first */
188 	sdstart,		/* have a queue, served by this */
189 	NULL,			/* have no async handler */
190 	sddone,			/* deal with stats at interrupt time */
191 };
192 
193 struct sd_mode_sense_data {
194 	/*
195 	 * XXX
196 	 * We are not going to parse this as-is -- it just has to be large
197 	 * enough.
198 	 */
199 	union {
200 		struct scsi_mode_parameter_header_6 small;
201 		struct scsi_mode_parameter_header_10 big;
202 	} header;
203 	struct scsi_general_block_descriptor blk_desc;
204 	union scsi_disk_pages pages;
205 };
206 
207 /*
208  * The routine called by the low level scsi routine when it discovers
209  * A device suitable for this driver
210  */
211 static int
212 sdmatch(device_t parent, cfdata_t match,
213     void *aux)
214 {
215 	struct scsipibus_attach_args *sa = aux;
216 	int priority;
217 
218 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
219 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
220 	    sizeof(sd_patterns[0]), &priority);
221 
222 	return (priority);
223 }
224 
225 /*
226  * Attach routine common to atapi & scsi.
227  */
228 static void
229 sdattach(device_t parent, device_t self, void *aux)
230 {
231 	struct sd_softc *sd = device_private(self);
232 	struct scsipibus_attach_args *sa = aux;
233 	struct scsipi_periph *periph = sa->sa_periph;
234 	int error, result;
235 	struct disk_parms *dp = &sd->params;
236 	char pbuf[9];
237 
238 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
239 
240 	sd->sc_dev = self;
241 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
242 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
243 	if (sd->type == T_SIMPLE_DIRECT)
244 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
245 
246 	if (SCSIPI_BUSTYPE_TYPE(scsipi_periph_bustype(sa->sa_periph)) ==
247 	    SCSIPI_BUSTYPE_SCSI && periph->periph_version == 0)
248 		sd->flags |= SDF_ANCIENT;
249 
250 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
251 
252 	callout_init(&sd->sc_callout, 0);
253 
254 	/*
255 	 * Store information needed to contact our base driver
256 	 */
257 	sd->sc_periph = periph;
258 
259 	periph->periph_dev = sd->sc_dev;
260 	periph->periph_switch = &sd_switch;
261 
262         /*
263          * Increase our openings to the maximum-per-periph
264          * supported by the adapter.  This will either be
265          * clamped down or grown by the adapter if necessary.
266          */
267 	periph->periph_openings =
268 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
269 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
270 
271 	/*
272 	 * Initialize and attach the disk structure.
273 	 */
274 	disk_init(&sd->sc_dk, device_xname(sd->sc_dev), &sddkdriver);
275 	disk_attach(&sd->sc_dk);
276 
277 	/*
278 	 * Use the subdriver to request information regarding the drive.
279 	 */
280 	aprint_naive("\n");
281 	aprint_normal("\n");
282 
283 	if (periph->periph_quirks & PQUIRK_START)
284 		(void)scsipi_start(periph, SSS_START, XS_CTL_SILENT);
285 
286 	error = scsipi_test_unit_ready(periph,
287 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
288 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
289 
290 	if (error)
291 		result = SDGP_RESULT_OFFLINE;
292 	else
293 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
294 	aprint_normal_dev(sd->sc_dev, "");
295 	switch (result) {
296 	case SDGP_RESULT_OK:
297 		format_bytes(pbuf, sizeof(pbuf),
298 		    (u_int64_t)dp->disksize * dp->blksize);
299 	        aprint_normal(
300 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
301 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
302 		    (unsigned long long)dp->disksize);
303 		break;
304 
305 	case SDGP_RESULT_OFFLINE:
306 		aprint_normal("drive offline");
307 		break;
308 
309 	case SDGP_RESULT_UNFORMATTED:
310 		aprint_normal("unformatted media");
311 		break;
312 
313 #ifdef DIAGNOSTIC
314 	default:
315 		panic("sdattach: unknown result from get_parms");
316 		break;
317 #endif
318 	}
319 	aprint_normal("\n");
320 
321 	/*
322 	 * Establish a shutdown hook so that we can ensure that
323 	 * our data has actually made it onto the platter at
324 	 * shutdown time.  Note that this relies on the fact
325 	 * that the shutdown hooks at the "leaves" of the device tree
326 	 * are run, first (thus guaranteeing that our hook runs before
327 	 * our ancestors').
328 	 */
329 	if (!pmf_device_register1(self, sd_suspend, NULL, sd_shutdown))
330 		aprint_error_dev(self, "couldn't establish power handler\n");
331 
332 	/*
333 	 * attach the device into the random source list
334 	 */
335 	rnd_attach_source(&sd->rnd_source, device_xname(sd->sc_dev),
336 			  RND_TYPE_DISK, RND_FLAG_DEFAULT);
337 
338 	/* Discover wedges on this disk. */
339 	dkwedge_discover(&sd->sc_dk);
340 
341 	/*
342 	 * Disk insertion and removal times can be a useful source
343 	 * of entropy, though the estimator should never _count_
344 	 * these bits, on insertion, because the deltas to the
345 	 * nonexistent) previous event should never allow it.
346 	 */
347 	rnd_add_uint32(&sd->rnd_source, 0);
348 }
349 
350 static int
351 sddetach(device_t self, int flags)
352 {
353 	struct sd_softc *sd = device_private(self);
354 	int s, bmaj, cmaj, i, mn, rc;
355 
356 	rnd_add_uint32(&sd->rnd_source, 0);
357 
358 	if ((rc = disk_begindetach(&sd->sc_dk, sdlastclose, self, flags)) != 0)
359 		return rc;
360 
361 	/* locate the major number */
362 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
363 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
364 
365 	/* Nuke the vnodes for any open instances */
366 	for (i = 0; i < MAXPARTITIONS; i++) {
367 		mn = SDMINOR(device_unit(self), i);
368 		vdevgone(bmaj, mn, mn, VBLK);
369 		vdevgone(cmaj, mn, mn, VCHR);
370 	}
371 
372 	/* kill any pending restart */
373 	callout_stop(&sd->sc_callout);
374 
375 	/* Delete all of our wedges. */
376 	dkwedge_delall(&sd->sc_dk);
377 
378 	s = splbio();
379 
380 	/* Kill off any queued buffers. */
381 	bufq_drain(sd->buf_queue);
382 
383 	bufq_free(sd->buf_queue);
384 
385 	/* Kill off any pending commands. */
386 	scsipi_kill_pending(sd->sc_periph);
387 
388 	splx(s);
389 
390 	/* Detach from the disk list. */
391 	disk_detach(&sd->sc_dk);
392 	disk_destroy(&sd->sc_dk);
393 
394 	callout_destroy(&sd->sc_callout);
395 
396 	pmf_device_deregister(self);
397 
398 	/* Unhook the entropy source. */
399 	rnd_detach_source(&sd->rnd_source);
400 
401 	return (0);
402 }
403 
404 /*
405  * open the device. Make sure the partition info is a up-to-date as can be.
406  */
407 static int
408 sdopen(dev_t dev, int flag, int fmt, struct lwp *l)
409 {
410 	struct sd_softc *sd;
411 	struct scsipi_periph *periph;
412 	struct scsipi_adapter *adapt;
413 	int unit, part;
414 	int error;
415 
416 	unit = SDUNIT(dev);
417 	sd = device_lookup_private(&sd_cd, unit);
418 	if (sd == NULL)
419 		return (ENXIO);
420 
421 	if (!device_is_active(sd->sc_dev))
422 		return (ENODEV);
423 
424 	part = SDPART(dev);
425 
426 	mutex_enter(&sd->sc_dk.dk_openlock);
427 
428 	/*
429 	 * If there are wedges, and this is not RAW_PART, then we
430 	 * need to fail.
431 	 */
432 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
433 		error = EBUSY;
434 		goto bad1;
435 	}
436 
437 	periph = sd->sc_periph;
438 	adapt = periph->periph_channel->chan_adapter;
439 
440 	SC_DEBUG(periph, SCSIPI_DB1,
441 	    ("sdopen: dev=0x%"PRIx64" (unit %d (of %d), partition %d)\n", dev, unit,
442 	    sd_cd.cd_ndevs, part));
443 
444 	/*
445 	 * If this is the first open of this device, add a reference
446 	 * to the adapter.
447 	 */
448 	if (sd->sc_dk.dk_openmask == 0 &&
449 	    (error = scsipi_adapter_addref(adapt)) != 0)
450 		goto bad1;
451 
452 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
453 		/*
454 		 * If any partition is open, but the disk has been invalidated,
455 		 * disallow further opens of non-raw partition
456 		 */
457 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
458 		    (part != RAW_PART || fmt != S_IFCHR)) {
459 			error = EIO;
460 			goto bad2;
461 		}
462 	} else {
463 		int silent;
464 
465 		if ((part == RAW_PART && fmt == S_IFCHR) || (flag & FSILENT))
466 			silent = XS_CTL_SILENT;
467 		else
468 			silent = 0;
469 
470 		/* Check that it is still responding and ok. */
471 		error = scsipi_test_unit_ready(periph,
472 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
473 		    silent);
474 
475 		/*
476 		 * Start the pack spinning if necessary. Always allow the
477 		 * raw parition to be opened, for raw IOCTLs. Data transfers
478 		 * will check for SDEV_MEDIA_LOADED.
479 		 */
480 		if (error == EIO) {
481 			int error2;
482 
483 			error2 = scsipi_start(periph, SSS_START, silent);
484 			switch (error2) {
485 			case 0:
486 				error = 0;
487 				break;
488 			case EIO:
489 			case EINVAL:
490 				break;
491 			default:
492 				error = error2;
493 				break;
494 			}
495 		}
496 		if (error) {
497 			if (silent && (flag & FSILENT) == 0)
498 				goto out;
499 			goto bad2;
500 		}
501 
502 		periph->periph_flags |= PERIPH_OPEN;
503 
504 		if (periph->periph_flags & PERIPH_REMOVABLE) {
505 			/* Lock the pack in. */
506 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
507 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
508 			    XS_CTL_IGNORE_MEDIA_CHANGE |
509 			    XS_CTL_SILENT);
510 			if (error)
511 				goto bad3;
512 		}
513 
514 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
515 			int param_error;
516 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
517 
518 			/*
519 			 * Load the physical device parameters.
520 			 *
521 			 * Note that if media is present but unformatted,
522 			 * we allow the open (so that it can be formatted!).
523 			 * The drive should refuse real I/O, if the media is
524 			 * unformatted.
525 			 */
526 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
527 			     == SDGP_RESULT_OFFLINE) {
528 				error = ENXIO;
529 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
530 				goto bad3;
531 			}
532 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
533 
534 			/* Load the partition info if not already loaded. */
535 			if (param_error == 0) {
536 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
537 					error = EIO;
538 					goto bad3;
539 				}
540 				SC_DEBUG(periph, SCSIPI_DB3,
541 				     ("Disklabel loaded "));
542 			}
543 		}
544 	}
545 
546 	/* Check that the partition exists. */
547 	if (part != RAW_PART &&
548 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
549 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
550 		error = ENXIO;
551 		goto bad3;
552 	}
553 
554  out:	/* Insure only one open at a time. */
555 	switch (fmt) {
556 	case S_IFCHR:
557 		sd->sc_dk.dk_copenmask |= (1 << part);
558 		break;
559 	case S_IFBLK:
560 		sd->sc_dk.dk_bopenmask |= (1 << part);
561 		break;
562 	}
563 	sd->sc_dk.dk_openmask =
564 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
565 
566 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
567 	mutex_exit(&sd->sc_dk.dk_openlock);
568 	return (0);
569 
570  bad3:
571 	if (sd->sc_dk.dk_openmask == 0) {
572 		if (periph->periph_flags & PERIPH_REMOVABLE)
573 			scsipi_prevent(periph, SPAMR_ALLOW,
574 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
575 			    XS_CTL_IGNORE_MEDIA_CHANGE |
576 			    XS_CTL_SILENT);
577 		periph->periph_flags &= ~PERIPH_OPEN;
578 	}
579 
580  bad2:
581 	if (sd->sc_dk.dk_openmask == 0)
582 		scsipi_adapter_delref(adapt);
583 
584  bad1:
585 	mutex_exit(&sd->sc_dk.dk_openlock);
586 	return (error);
587 }
588 
589 /*
590  * Caller must hold sd->sc_dk.dk_openlock.
591  */
592 static int
593 sdlastclose(device_t self)
594 {
595 	struct sd_softc *sd = device_private(self);
596 	struct scsipi_periph *periph = sd->sc_periph;
597 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
598 
599 	/*
600 	 * If the disk cache needs flushing, and the disk supports
601 	 * it, do it now.
602 	 */
603 	if ((sd->flags & SDF_DIRTY) != 0) {
604 		if (sd_flush(sd, 0)) {
605 			aprint_error_dev(sd->sc_dev,
606 				"cache synchronization failed\n");
607 			sd->flags &= ~SDF_FLUSHING;
608 		} else
609 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
610 	}
611 
612 	scsipi_wait_drain(periph);
613 
614 	if (periph->periph_flags & PERIPH_REMOVABLE)
615 		scsipi_prevent(periph, SPAMR_ALLOW,
616 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
617 		    XS_CTL_IGNORE_NOT_READY |
618 		    XS_CTL_SILENT);
619 	periph->periph_flags &= ~PERIPH_OPEN;
620 
621 	scsipi_wait_drain(periph);
622 
623 	scsipi_adapter_delref(adapt);
624 
625 	return 0;
626 }
627 
628 /*
629  * close the device.. only called if we are the LAST occurence of an open
630  * device.  Convenient now but usually a pain.
631  */
632 static int
633 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
634 {
635 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
636 	int part = SDPART(dev);
637 
638 	mutex_enter(&sd->sc_dk.dk_openlock);
639 	switch (fmt) {
640 	case S_IFCHR:
641 		sd->sc_dk.dk_copenmask &= ~(1 << part);
642 		break;
643 	case S_IFBLK:
644 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
645 		break;
646 	}
647 	sd->sc_dk.dk_openmask =
648 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
649 
650 	if (sd->sc_dk.dk_openmask == 0)
651 		sdlastclose(sd->sc_dev);
652 
653 	mutex_exit(&sd->sc_dk.dk_openlock);
654 	return (0);
655 }
656 
657 /*
658  * Actually translate the requested transfer into one the physical driver
659  * can understand.  The transfer is described by a buf and will include
660  * only one physical transfer.
661  */
662 static void
663 sdstrategy(struct buf *bp)
664 {
665 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
666 	struct scsipi_periph *periph = sd->sc_periph;
667 	struct disklabel *lp;
668 	daddr_t blkno;
669 	int s;
670 	bool sector_aligned;
671 
672 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
673 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
674 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
675 	/*
676 	 * If the device has been made invalid, error out
677 	 */
678 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
679 	    !device_is_active(sd->sc_dev)) {
680 		if (periph->periph_flags & PERIPH_OPEN)
681 			bp->b_error = EIO;
682 		else
683 			bp->b_error = ENODEV;
684 		goto done;
685 	}
686 
687 	lp = sd->sc_dk.dk_label;
688 
689 	/*
690 	 * The transfer must be a whole number of blocks, offset must not be
691 	 * negative.
692 	 */
693 	if (lp->d_secsize == DEV_BSIZE) {
694 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
695 	} else {
696 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
697 	}
698 	if (!sector_aligned || bp->b_blkno < 0) {
699 		bp->b_error = EINVAL;
700 		goto done;
701 	}
702 	/*
703 	 * If it's a null transfer, return immediatly
704 	 */
705 	if (bp->b_bcount == 0)
706 		goto done;
707 
708 	/*
709 	 * Do bounds checking, adjust transfer. if error, process.
710 	 * If end of partition, just return.
711 	 */
712 	if (SDPART(bp->b_dev) == RAW_PART) {
713 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
714 		    sd->params.disksize512) <= 0)
715 			goto done;
716 	} else {
717 		if (bounds_check_with_label(&sd->sc_dk, bp,
718 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
719 			goto done;
720 	}
721 
722 	/*
723 	 * Now convert the block number to absolute and put it in
724 	 * terms of the device's logical block size.
725 	 */
726 	if (lp->d_secsize == DEV_BSIZE)
727 		blkno = bp->b_blkno;
728 	else if (lp->d_secsize > DEV_BSIZE)
729 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
730 	else
731 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
732 
733 	if (SDPART(bp->b_dev) != RAW_PART)
734 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
735 
736 	bp->b_rawblkno = blkno;
737 
738 	s = splbio();
739 
740 	/*
741 	 * Place it in the queue of disk activities for this disk.
742 	 *
743 	 * XXX Only do disksort() if the current operating mode does not
744 	 * XXX include tagged queueing.
745 	 */
746 	bufq_put(sd->buf_queue, bp);
747 
748 	/*
749 	 * Tell the device to get going on the transfer if it's
750 	 * not doing anything, otherwise just wait for completion
751 	 */
752 	sdstart(sd->sc_periph);
753 
754 	splx(s);
755 	return;
756 
757 done:
758 	/*
759 	 * Correctly set the buf to indicate a completed xfer
760 	 */
761 	bp->b_resid = bp->b_bcount;
762 	biodone(bp);
763 }
764 
765 /*
766  * sdstart looks to see if there is a buf waiting for the device
767  * and that the device is not already busy. If both are true,
768  * It dequeues the buf and creates a scsi command to perform the
769  * transfer in the buf. The transfer request will call scsipi_done
770  * on completion, which will in turn call this routine again
771  * so that the next queued transfer is performed.
772  * The bufs are queued by the strategy routine (sdstrategy)
773  *
774  * This routine is also called after other non-queued requests
775  * have been made of the scsi driver, to ensure that the queue
776  * continues to be drained.
777  *
778  * must be called at the correct (highish) spl level
779  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
780  */
781 static void
782 sdstart(struct scsipi_periph *periph)
783 {
784 	struct sd_softc *sd = device_private(periph->periph_dev);
785 	struct disklabel *lp = sd->sc_dk.dk_label;
786 	struct buf *bp = 0;
787 	struct scsipi_rw_16 cmd16;
788 	struct scsipi_rw_10 cmd_big;
789 	struct scsi_rw_6 cmd_small;
790 	struct scsipi_generic *cmdp;
791 	struct scsipi_xfer *xs;
792 	int nblks, cmdlen, error __diagused, flags;
793 
794 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
795 	/*
796 	 * Check if the device has room for another command
797 	 */
798 	while (periph->periph_active < periph->periph_openings) {
799 		/*
800 		 * there is excess capacity, but a special waits
801 		 * It'll need the adapter as soon as we clear out of the
802 		 * way and let it run (user level wait).
803 		 */
804 		if (periph->periph_flags & PERIPH_WAITING) {
805 			periph->periph_flags &= ~PERIPH_WAITING;
806 			wakeup((void *)periph);
807 			return;
808 		}
809 
810 		/*
811 		 * If the device has become invalid, abort all the
812 		 * reads and writes until all files have been closed and
813 		 * re-opened
814 		 */
815 		if (__predict_false(
816 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
817 			if ((bp = bufq_get(sd->buf_queue)) != NULL) {
818 				bp->b_error = EIO;
819 				bp->b_resid = bp->b_bcount;
820 				biodone(bp);
821 				continue;
822 			} else {
823 				return;
824 			}
825 		}
826 
827 		/*
828 		 * See if there is a buf with work for us to do..
829 		 */
830 		if ((bp = bufq_peek(sd->buf_queue)) == NULL)
831 			return;
832 
833 		/*
834 		 * We have a buf, now we should make a command.
835 		 */
836 
837 		if (lp->d_secsize == DEV_BSIZE)
838 			nblks = bp->b_bcount >> DEV_BSHIFT;
839 		else
840 			nblks = howmany(bp->b_bcount, lp->d_secsize);
841 
842 		/*
843 		 * Fill out the scsi command.  Use the smallest CDB possible
844 		 * (6-byte, 10-byte, or 16-byte).
845 		 */
846 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
847 		    ((nblks & 0xff) == nblks) &&
848 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
849 			/* 6-byte CDB */
850 			memset(&cmd_small, 0, sizeof(cmd_small));
851 			cmd_small.opcode = (bp->b_flags & B_READ) ?
852 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
853 			_lto3b(bp->b_rawblkno, cmd_small.addr);
854 			cmd_small.length = nblks & 0xff;
855 			cmdlen = sizeof(cmd_small);
856 			cmdp = (struct scsipi_generic *)&cmd_small;
857 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
858 			/* 10-byte CDB */
859 			memset(&cmd_big, 0, sizeof(cmd_big));
860 			cmd_big.opcode = (bp->b_flags & B_READ) ?
861 			    READ_10 : WRITE_10;
862 			_lto4b(bp->b_rawblkno, cmd_big.addr);
863 			_lto2b(nblks, cmd_big.length);
864 			cmdlen = sizeof(cmd_big);
865 			cmdp = (struct scsipi_generic *)&cmd_big;
866 		} else {
867 			/* 16-byte CDB */
868 			memset(&cmd16, 0, sizeof(cmd16));
869 			cmd16.opcode = (bp->b_flags & B_READ) ?
870 			    READ_16 : WRITE_16;
871 			_lto8b(bp->b_rawblkno, cmd16.addr);
872 			_lto4b(nblks, cmd16.length);
873 			cmdlen = sizeof(cmd16);
874 			cmdp = (struct scsipi_generic *)&cmd16;
875 		}
876 
877 		/* Instrumentation. */
878 		disk_busy(&sd->sc_dk);
879 
880 		/*
881 		 * Mark the disk dirty so that the cache will be
882 		 * flushed on close.
883 		 */
884 		if ((bp->b_flags & B_READ) == 0)
885 			sd->flags |= SDF_DIRTY;
886 
887 		/*
888 		 * Figure out what flags to use.
889 		 */
890 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
891 		if (bp->b_flags & B_READ)
892 			flags |= XS_CTL_DATA_IN;
893 		else
894 			flags |= XS_CTL_DATA_OUT;
895 
896 		/*
897 		 * Call the routine that chats with the adapter.
898 		 * Note: we cannot sleep as we may be an interrupt
899 		 */
900 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
901 		    (u_char *)bp->b_data, bp->b_bcount,
902 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
903 		if (__predict_false(xs == NULL)) {
904 			/*
905 			 * out of memory. Keep this buffer in the queue, and
906 			 * retry later.
907 			 */
908 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
909 			    periph);
910 			return;
911 		}
912 		/*
913 		 * need to dequeue the buffer before queuing the command,
914 		 * because cdstart may be called recursively from the
915 		 * HBA driver
916 		 */
917 #ifdef DIAGNOSTIC
918 		if (bufq_get(sd->buf_queue) != bp)
919 			panic("sdstart(): dequeued wrong buf");
920 #else
921 		bufq_get(sd->buf_queue);
922 #endif
923 		error = scsipi_execute_xs(xs);
924 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
925 		KASSERT(error == 0);
926 	}
927 }
928 
929 static void
930 sdrestart(void *v)
931 {
932 	int s = splbio();
933 	sdstart((struct scsipi_periph *)v);
934 	splx(s);
935 }
936 
937 static void
938 sddone(struct scsipi_xfer *xs, int error)
939 {
940 	struct sd_softc *sd = device_private(xs->xs_periph->periph_dev);
941 	struct buf *bp = xs->bp;
942 
943 	if (sd->flags & SDF_FLUSHING) {
944 		/* Flush completed, no longer dirty. */
945 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
946 	}
947 
948 	if (bp) {
949 		bp->b_error = error;
950 		bp->b_resid = xs->resid;
951 		if (error) {
952 			/* on a read/write error bp->b_resid is zero, so fix */
953 			bp->b_resid = bp->b_bcount;
954 		}
955 
956 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
957 		    (bp->b_flags & B_READ));
958 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
959 
960 		biodone(bp);
961 	}
962 }
963 
964 static void
965 sdminphys(struct buf *bp)
966 {
967 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
968 	long xmax;
969 
970 	/*
971 	 * If the device is ancient, we want to make sure that
972 	 * the transfer fits into a 6-byte cdb.
973 	 *
974 	 * XXX Note that the SCSI-I spec says that 256-block transfers
975 	 * are allowed in a 6-byte read/write, and are specified
976 	 * by settng the "length" to 0.  However, we're conservative
977 	 * here, allowing only 255-block transfers in case an
978 	 * ancient device gets confused by length == 0.  A length of 0
979 	 * in a 10-byte read/write actually means 0 blocks.
980 	 */
981 	if ((sd->flags & SDF_ANCIENT) &&
982 	    ((sd->sc_periph->periph_flags &
983 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
984 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
985 
986 		if (bp->b_bcount > xmax)
987 			bp->b_bcount = xmax;
988 	}
989 
990 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
991 }
992 
993 static int
994 sdread(dev_t dev, struct uio *uio, int ioflag)
995 {
996 
997 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
998 }
999 
1000 static int
1001 sdwrite(dev_t dev, struct uio *uio, int ioflag)
1002 {
1003 
1004 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
1005 }
1006 
1007 /*
1008  * Perform special action on behalf of the user
1009  * Knows about the internals of this device
1010  */
1011 static int
1012 sdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1013 {
1014 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
1015 	struct scsipi_periph *periph = sd->sc_periph;
1016 	int part = SDPART(dev);
1017 	int error;
1018 	int s;
1019 #ifdef __HAVE_OLD_DISKLABEL
1020 	struct disklabel *newlabel = NULL;
1021 #endif
1022 
1023 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1024 
1025 	/*
1026 	 * If the device is not valid, some IOCTLs can still be
1027 	 * handled on the raw partition. Check this here.
1028 	 */
1029 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1030 		switch (cmd) {
1031 		case DIOCKLABEL:
1032 		case DIOCWLABEL:
1033 		case DIOCLOCK:
1034 		case DIOCEJECT:
1035 		case ODIOCEJECT:
1036 		case DIOCGCACHE:
1037 		case DIOCSCACHE:
1038 		case DIOCGSTRATEGY:
1039 		case DIOCSSTRATEGY:
1040 		case SCIOCIDENTIFY:
1041 		case OSCIOCIDENTIFY:
1042 		case SCIOCCOMMAND:
1043 		case SCIOCDEBUG:
1044 			if (part == RAW_PART)
1045 				break;
1046 		/* FALLTHROUGH */
1047 		default:
1048 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1049 				return (ENODEV);
1050 			else
1051 				return (EIO);
1052 		}
1053 	}
1054 
1055 	error = disk_ioctl(&sd->sc_dk, dev, cmd, addr, flag, l);
1056 	if (error != EPASSTHROUGH)
1057 		return (error);
1058 
1059 	error = 0;
1060 	switch (cmd) {
1061 	case DIOCWDINFO:
1062 	case DIOCSDINFO:
1063 #ifdef __HAVE_OLD_DISKLABEL
1064 	case ODIOCWDINFO:
1065 	case ODIOCSDINFO:
1066 #endif
1067 	{
1068 		struct disklabel *lp;
1069 
1070 		if ((flag & FWRITE) == 0)
1071 			return (EBADF);
1072 
1073 #ifdef __HAVE_OLD_DISKLABEL
1074  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1075 			newlabel = malloc(sizeof *newlabel, M_TEMP,
1076 			    M_WAITOK | M_ZERO);
1077 			if (newlabel == NULL)
1078 				return EIO;
1079 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1080 			lp = newlabel;
1081 		} else
1082 #endif
1083 		lp = (struct disklabel *)addr;
1084 
1085 		mutex_enter(&sd->sc_dk.dk_openlock);
1086 		sd->flags |= SDF_LABELLING;
1087 
1088 		error = setdisklabel(sd->sc_dk.dk_label,
1089 		    lp, /*sd->sc_dk.dk_openmask : */0,
1090 		    sd->sc_dk.dk_cpulabel);
1091 		if (error == 0) {
1092 			if (cmd == DIOCWDINFO
1093 #ifdef __HAVE_OLD_DISKLABEL
1094 			    || cmd == ODIOCWDINFO
1095 #endif
1096 			   )
1097 				error = writedisklabel(SDLABELDEV(dev),
1098 				    sdstrategy, sd->sc_dk.dk_label,
1099 				    sd->sc_dk.dk_cpulabel);
1100 		}
1101 
1102 		sd->flags &= ~SDF_LABELLING;
1103 		mutex_exit(&sd->sc_dk.dk_openlock);
1104 #ifdef __HAVE_OLD_DISKLABEL
1105 		if (newlabel != NULL)
1106 			free(newlabel, M_TEMP);
1107 #endif
1108 		return (error);
1109 	}
1110 
1111 	case DIOCKLABEL:
1112 		if (*(int *)addr)
1113 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1114 		else
1115 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1116 		return (0);
1117 
1118 	case DIOCWLABEL:
1119 		if ((flag & FWRITE) == 0)
1120 			return (EBADF);
1121 		if (*(int *)addr)
1122 			sd->flags |= SDF_WLABEL;
1123 		else
1124 			sd->flags &= ~SDF_WLABEL;
1125 		return (0);
1126 
1127 	case DIOCLOCK:
1128 		if (periph->periph_flags & PERIPH_REMOVABLE)
1129 			return (scsipi_prevent(periph,
1130 			    (*(int *)addr) ?
1131 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1132 		else
1133 			return (ENOTTY);
1134 
1135 	case DIOCEJECT:
1136 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1137 			return (ENOTTY);
1138 		if (*(int *)addr == 0) {
1139 			/*
1140 			 * Don't force eject: check that we are the only
1141 			 * partition open. If so, unlock it.
1142 			 */
1143 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1144 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1145 			    sd->sc_dk.dk_openmask) {
1146 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1147 				    XS_CTL_IGNORE_NOT_READY);
1148 				if (error)
1149 					return (error);
1150 			} else {
1151 				return (EBUSY);
1152 			}
1153 		}
1154 		/* FALLTHROUGH */
1155 	case ODIOCEJECT:
1156 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1157 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1158 
1159 	case DIOCGDEFLABEL:
1160 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1161 		return (0);
1162 
1163 #ifdef __HAVE_OLD_DISKLABEL
1164 	case ODIOCGDEFLABEL:
1165 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1166 		if (newlabel == NULL)
1167 			return EIO;
1168 		sdgetdefaultlabel(sd, newlabel);
1169 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1170 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1171 		else
1172 			error = ENOTTY;
1173 		free(newlabel, M_TEMP);
1174 		return error;
1175 #endif
1176 
1177 	case DIOCGCACHE:
1178 		return (sd_getcache(sd, (int *) addr));
1179 
1180 	case DIOCSCACHE:
1181 		if ((flag & FWRITE) == 0)
1182 			return (EBADF);
1183 		return (sd_setcache(sd, *(int *) addr));
1184 
1185 	case DIOCCACHESYNC:
1186 		/*
1187 		 * XXX Do we really need to care about having a writable
1188 		 * file descriptor here?
1189 		 */
1190 		if ((flag & FWRITE) == 0)
1191 			return (EBADF);
1192 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1193 			error = sd_flush(sd, 0);
1194 			if (error)
1195 				sd->flags &= ~SDF_FLUSHING;
1196 			else
1197 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1198 		}
1199 		return (error);
1200 
1201 	case DIOCGSTRATEGY:
1202 	    {
1203 		struct disk_strategy *dks = addr;
1204 
1205 		s = splbio();
1206 		strlcpy(dks->dks_name, bufq_getstrategyname(sd->buf_queue),
1207 		    sizeof(dks->dks_name));
1208 		splx(s);
1209 		dks->dks_paramlen = 0;
1210 
1211 		return 0;
1212 	    }
1213 
1214 	case DIOCSSTRATEGY:
1215 	    {
1216 		struct disk_strategy *dks = addr;
1217 		struct bufq_state *new_bufq;
1218 		struct bufq_state *old_bufq;
1219 
1220 		if ((flag & FWRITE) == 0) {
1221 			return EBADF;
1222 		}
1223 
1224 		if (dks->dks_param != NULL) {
1225 			return EINVAL;
1226 		}
1227 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1228 		error = bufq_alloc(&new_bufq, dks->dks_name,
1229 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1230 		if (error) {
1231 			return error;
1232 		}
1233 		s = splbio();
1234 		old_bufq = sd->buf_queue;
1235 		bufq_move(new_bufq, old_bufq);
1236 		sd->buf_queue = new_bufq;
1237 		splx(s);
1238 		bufq_free(old_bufq);
1239 
1240 		return 0;
1241 	    }
1242 
1243 	default:
1244 		if (part != RAW_PART)
1245 			return (ENOTTY);
1246 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1247 	}
1248 
1249 #ifdef DIAGNOSTIC
1250 	panic("sdioctl: impossible");
1251 #endif
1252 }
1253 
1254 static void
1255 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1256 {
1257 
1258 	memset(lp, 0, sizeof(struct disklabel));
1259 
1260 	lp->d_secsize = sd->params.blksize;
1261 	lp->d_ntracks = sd->params.heads;
1262 	lp->d_nsectors = sd->params.sectors;
1263 	lp->d_ncylinders = sd->params.cyls;
1264 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1265 
1266 	switch (SCSIPI_BUSTYPE_TYPE(scsipi_periph_bustype(sd->sc_periph))) {
1267 	case SCSIPI_BUSTYPE_SCSI:
1268 		lp->d_type = DKTYPE_SCSI;
1269 		break;
1270 	case SCSIPI_BUSTYPE_ATAPI:
1271 		lp->d_type = DKTYPE_ATAPI;
1272 		break;
1273 	}
1274 	/*
1275 	 * XXX
1276 	 * We could probe the mode pages to figure out what kind of disc it is.
1277 	 * Is this worthwhile?
1278 	 */
1279 	strncpy(lp->d_typename, sd->name, 16);
1280 	strncpy(lp->d_packname, "fictitious", 16);
1281 	if (sd->params.disksize > UINT32_MAX)
1282 		lp->d_secperunit = UINT32_MAX;
1283 	else
1284 		lp->d_secperunit = sd->params.disksize;
1285 	lp->d_rpm = sd->params.rot_rate;
1286 	lp->d_interleave = 1;
1287 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1288 	    D_REMOVABLE : 0;
1289 
1290 	lp->d_partitions[RAW_PART].p_offset = 0;
1291 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1292 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1293 	lp->d_npartitions = RAW_PART + 1;
1294 
1295 	lp->d_magic = DISKMAGIC;
1296 	lp->d_magic2 = DISKMAGIC;
1297 	lp->d_checksum = dkcksum(lp);
1298 }
1299 
1300 
1301 /*
1302  * Load the label information on the named device
1303  */
1304 static int
1305 sdgetdisklabel(struct sd_softc *sd)
1306 {
1307 	struct disklabel *lp = sd->sc_dk.dk_label;
1308 	const char *errstring;
1309 
1310 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1311 
1312 	sdgetdefaultlabel(sd, lp);
1313 
1314 	if (lp->d_secpercyl == 0) {
1315 		lp->d_secpercyl = 100;
1316 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1317 	}
1318 
1319 	/*
1320 	 * Call the generic disklabel extraction routine
1321 	 */
1322 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1323 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1324 	if (errstring) {
1325 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1326 		return EIO;
1327 	}
1328 	return 0;
1329 }
1330 
1331 static bool
1332 sd_shutdown(device_t self, int how)
1333 {
1334 	struct sd_softc *sd = device_private(self);
1335 
1336 	/*
1337 	 * If the disk cache needs to be flushed, and the disk supports
1338 	 * it, flush it.  We're cold at this point, so we poll for
1339 	 * completion.
1340 	 */
1341 	if ((sd->flags & SDF_DIRTY) != 0) {
1342 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1343 			aprint_error_dev(sd->sc_dev,
1344 				"cache synchronization failed\n");
1345 			sd->flags &= ~SDF_FLUSHING;
1346 		} else
1347 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1348 	}
1349 	return true;
1350 }
1351 
1352 static bool
1353 sd_suspend(device_t dv, const pmf_qual_t *qual)
1354 {
1355 	return sd_shutdown(dv, boothowto); /* XXX no need to poll */
1356 }
1357 
1358 /*
1359  * Check Errors
1360  */
1361 static int
1362 sd_interpret_sense(struct scsipi_xfer *xs)
1363 {
1364 	struct scsipi_periph *periph = xs->xs_periph;
1365 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1366 	struct sd_softc *sd = device_private(periph->periph_dev);
1367 	int s, error, retval = EJUSTRETURN;
1368 
1369 	/*
1370 	 * If the periph is already recovering, just do the normal
1371 	 * error processing.
1372 	 */
1373 	if (periph->periph_flags & PERIPH_RECOVERING)
1374 		return (retval);
1375 
1376 	/*
1377 	 * Ignore errors from accessing illegal fields (e.g. trying to
1378 	 * lock the door of a digicam, which doesn't have a door that
1379 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1380 	 */
1381 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1382 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1383 	    sense->asc == 0x24 &&
1384 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1385 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1386 			scsipi_printaddr(periph);
1387 			printf("no door lock\n");
1388 		}
1389 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1390 		return (retval);
1391 	}
1392 
1393 
1394 
1395 	/*
1396 	 * If the device is not open yet, let the generic code handle it.
1397 	 */
1398 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1399 		return (retval);
1400 
1401 	/*
1402 	 * If it isn't a extended or extended/deferred error, let
1403 	 * the generic code handle it.
1404 	 */
1405 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1406 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1407 		return (retval);
1408 
1409 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1410 	    sense->asc == 0x4) {
1411 		if (sense->ascq == 0x01)	{
1412 			/*
1413 			 * Unit In The Process Of Becoming Ready.
1414 			 */
1415 			printf("%s: waiting for pack to spin up...\n",
1416 			    device_xname(sd->sc_dev));
1417 			if (!callout_pending(&periph->periph_callout))
1418 				scsipi_periph_freeze(periph, 1);
1419 			callout_reset(&periph->periph_callout,
1420 			    5 * hz, scsipi_periph_timed_thaw, periph);
1421 			retval = ERESTART;
1422 		} else if (sense->ascq == 0x02) {
1423 			printf("%s: pack is stopped, restarting...\n",
1424 			    device_xname(sd->sc_dev));
1425 			s = splbio();
1426 			periph->periph_flags |= PERIPH_RECOVERING;
1427 			splx(s);
1428 			error = scsipi_start(periph, SSS_START,
1429 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1430 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1431 			if (error) {
1432 				aprint_error_dev(sd->sc_dev,
1433 					"unable to restart pack\n");
1434 				retval = error;
1435 			} else
1436 				retval = ERESTART;
1437 			s = splbio();
1438 			periph->periph_flags &= ~PERIPH_RECOVERING;
1439 			splx(s);
1440 		}
1441 	}
1442 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1443 	    sense->asc == 0x31 &&
1444 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1445 		/* Medium Format Corrupted */
1446 		retval = EFTYPE;
1447 	}
1448 	return (retval);
1449 }
1450 
1451 
1452 static int
1453 sdsize(dev_t dev)
1454 {
1455 	struct sd_softc *sd;
1456 	int part, unit, omask;
1457 	int size;
1458 
1459 	unit = SDUNIT(dev);
1460 	sd = device_lookup_private(&sd_cd, unit);
1461 	if (sd == NULL)
1462 		return (-1);
1463 
1464 	if (!device_is_active(sd->sc_dev))
1465 		return (-1);
1466 
1467 	part = SDPART(dev);
1468 	omask = sd->sc_dk.dk_openmask & (1 << part);
1469 
1470 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1471 		return (-1);
1472 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1473 		size = -1;
1474 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1475 		size = -1;
1476 	else
1477 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1478 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1479 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1480 		return (-1);
1481 	return (size);
1482 }
1483 
1484 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1485 static struct scsipi_xfer sx;
1486 static int sddoingadump;
1487 
1488 /*
1489  * dump all of physical memory into the partition specified, starting
1490  * at offset 'dumplo' into the partition.
1491  */
1492 static int
1493 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1494 {
1495 	struct sd_softc *sd;	/* disk unit to do the I/O */
1496 	struct disklabel *lp;	/* disk's disklabel */
1497 	int	unit, part;
1498 	int	sectorsize;	/* size of a disk sector */
1499 	int	nsects;		/* number of sectors in partition */
1500 	int	sectoff;	/* sector offset of partition */
1501 	int	totwrt;		/* total number of sectors left to write */
1502 	int	nwrt;		/* current number of sectors to write */
1503 	struct scsipi_rw_10 cmd;	/* write command */
1504 	struct scsipi_xfer *xs;	/* ... convenience */
1505 	struct scsipi_periph *periph;
1506 	struct scsipi_channel *chan;
1507 
1508 	/* Check if recursive dump; if so, punt. */
1509 	if (sddoingadump)
1510 		return (EFAULT);
1511 
1512 	/* Mark as active early. */
1513 	sddoingadump = 1;
1514 
1515 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1516 	part = SDPART(dev);
1517 
1518 	/* Check for acceptable drive number. */
1519 	sd = device_lookup_private(&sd_cd, unit);
1520 	if (sd == NULL)
1521 		return (ENXIO);
1522 
1523 	if (!device_is_active(sd->sc_dev))
1524 		return (ENODEV);
1525 
1526 	periph = sd->sc_periph;
1527 	chan = periph->periph_channel;
1528 
1529 	/* Make sure it was initialized. */
1530 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1531 		return (ENXIO);
1532 
1533 	/* Convert to disk sectors.  Request must be a multiple of size. */
1534 	lp = sd->sc_dk.dk_label;
1535 	sectorsize = lp->d_secsize;
1536 	if ((size % sectorsize) != 0)
1537 		return (EFAULT);
1538 	totwrt = size / sectorsize;
1539 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1540 
1541 	nsects = lp->d_partitions[part].p_size;
1542 	sectoff = lp->d_partitions[part].p_offset;
1543 
1544 	/* Check transfer bounds against partition size. */
1545 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1546 		return (EINVAL);
1547 
1548 	/* Offset block number to start of partition. */
1549 	blkno += sectoff;
1550 
1551 	xs = &sx;
1552 
1553 	while (totwrt > 0) {
1554 		nwrt = totwrt;		/* XXX */
1555 #ifndef	SD_DUMP_NOT_TRUSTED
1556 		/*
1557 		 *  Fill out the scsi command
1558 		 */
1559 		memset(&cmd, 0, sizeof(cmd));
1560 		cmd.opcode = WRITE_10;
1561 		_lto4b(blkno, cmd.addr);
1562 		_lto2b(nwrt, cmd.length);
1563 		/*
1564 		 * Fill out the scsipi_xfer structure
1565 		 *    Note: we cannot sleep as we may be an interrupt
1566 		 * don't use scsipi_command() as it may want to wait
1567 		 * for an xs.
1568 		 */
1569 		memset(xs, 0, sizeof(sx));
1570 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1571 		    XS_CTL_DATA_OUT;
1572 		xs->xs_status = 0;
1573 		xs->xs_periph = periph;
1574 		xs->xs_retries = SDRETRIES;
1575 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1576 		xs->cmd = (struct scsipi_generic *)&cmd;
1577 		xs->cmdlen = sizeof(cmd);
1578 		xs->resid = nwrt * sectorsize;
1579 		xs->error = XS_NOERROR;
1580 		xs->bp = 0;
1581 		xs->data = va;
1582 		xs->datalen = nwrt * sectorsize;
1583 		callout_init(&xs->xs_callout, 0);
1584 
1585 		/*
1586 		 * Pass all this info to the scsi driver.
1587 		 */
1588 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1589 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1590 		    xs->error != XS_NOERROR)
1591 			return (EIO);
1592 #else	/* SD_DUMP_NOT_TRUSTED */
1593 		/* Let's just talk about this first... */
1594 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1595 		delay(500 * 1000);	/* half a second */
1596 #endif	/* SD_DUMP_NOT_TRUSTED */
1597 
1598 		/* update block count */
1599 		totwrt -= nwrt;
1600 		blkno += nwrt;
1601 		va = (char *)va + sectorsize * nwrt;
1602 	}
1603 	sddoingadump = 0;
1604 	return (0);
1605 }
1606 
1607 static int
1608 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1609     int page, int flags, int *big)
1610 {
1611 
1612 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1613 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1614 		*big = 1;
1615 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1616 		    size + sizeof(struct scsi_mode_parameter_header_10),
1617 		    flags, SDRETRIES, 6000);
1618 	} else {
1619 		*big = 0;
1620 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1621 		    size + sizeof(struct scsi_mode_parameter_header_6),
1622 		    flags, SDRETRIES, 6000);
1623 	}
1624 }
1625 
1626 static int
1627 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1628     int flags, int big)
1629 {
1630 
1631 	if (big) {
1632 		struct scsi_mode_parameter_header_10 *header = sense;
1633 
1634 		_lto2b(0, header->data_length);
1635 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1636 		    size + sizeof(struct scsi_mode_parameter_header_10),
1637 		    flags, SDRETRIES, 6000);
1638 	} else {
1639 		struct scsi_mode_parameter_header_6 *header = sense;
1640 
1641 		header->data_length = 0;
1642 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1643 		    size + sizeof(struct scsi_mode_parameter_header_6),
1644 		    flags, SDRETRIES, 6000);
1645 	}
1646 }
1647 
1648 /*
1649  * sd_validate_blksize:
1650  *
1651  *	Validate the block size.  Print error if periph is specified,
1652  */
1653 static int
1654 sd_validate_blksize(struct scsipi_periph *periph, int len)
1655 {
1656 
1657 	switch (len) {
1658 	case 256:
1659 	case 512:
1660 	case 1024:
1661 	case 2048:
1662 	case 4096:
1663 		return 1;
1664 	}
1665 
1666 	if (periph) {
1667 		scsipi_printaddr(periph);
1668 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1669 		    (len ^ (1 << (ffs(len) - 1))) ?
1670 		    "preposterous" : "unsupported",
1671 		    len, SD_DEFAULT_BLKSIZE);
1672 	}
1673 
1674 	return 0;
1675 }
1676 
1677 /*
1678  * sd_read_capacity:
1679  *
1680  *	Find out from the device what its capacity is.
1681  */
1682 static u_int64_t
1683 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1684 {
1685 	union {
1686 		struct scsipi_read_capacity_10 cmd;
1687 		struct scsipi_read_capacity_16 cmd16;
1688 	} cmd;
1689 	union {
1690 		struct scsipi_read_capacity_10_data data;
1691 		struct scsipi_read_capacity_16_data data16;
1692 	} *datap;
1693 	uint64_t rv;
1694 
1695 	memset(&cmd, 0, sizeof(cmd));
1696 	cmd.cmd.opcode = READ_CAPACITY_10;
1697 
1698 	/*
1699 	 * Don't allocate data buffer on stack;
1700 	 * The lower driver layer might use the same stack and
1701 	 * if it uses region which is in the same cacheline,
1702 	 * cache flush ops against the data buffer won't work properly.
1703 	 */
1704 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1705 	if (datap == NULL)
1706 		return 0;
1707 
1708 	/*
1709 	 * If the command works, interpret the result as a 4 byte
1710 	 * number of blocks
1711 	 */
1712 	rv = 0;
1713 	memset(datap, 0, sizeof(datap->data));
1714 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1715 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1716 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1717 		goto out;
1718 
1719 	if (_4btol(datap->data.addr) != 0xffffffff) {
1720 		*blksize = _4btol(datap->data.length);
1721 		rv = _4btol(datap->data.addr) + 1;
1722 		goto out;
1723 	}
1724 
1725 	/*
1726 	 * Device is larger than can be reflected by READ CAPACITY (10).
1727 	 * Try READ CAPACITY (16).
1728 	 */
1729 
1730 	memset(&cmd, 0, sizeof(cmd));
1731 	cmd.cmd16.opcode = READ_CAPACITY_16;
1732 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1733 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1734 
1735 	memset(datap, 0, sizeof(datap->data16));
1736 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1737 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1738 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1739 		goto out;
1740 
1741 	*blksize = _4btol(datap->data16.length);
1742 	rv = _8btol(datap->data16.addr) + 1;
1743 
1744  out:
1745 	free(datap, M_TEMP);
1746 	return rv;
1747 }
1748 
1749 static int
1750 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1751 {
1752 	struct {
1753 		struct scsi_mode_parameter_header_6 header;
1754 		/* no block descriptor */
1755 		u_int8_t pg_code; /* page code (should be 6) */
1756 		u_int8_t pg_length; /* page length (should be 11) */
1757 		u_int8_t wcd; /* bit0: cache disable */
1758 		u_int8_t lbs[2]; /* logical block size */
1759 		u_int8_t size[5]; /* number of log. blocks */
1760 		u_int8_t pp; /* power/performance */
1761 		u_int8_t flags;
1762 		u_int8_t resvd;
1763 	} scsipi_sense;
1764 	u_int64_t blocks;
1765 	int error, blksize;
1766 
1767 	/*
1768 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1769 	 * give the same information. Do both for now, and check
1770 	 * for consistency.
1771 	 * XXX probably differs for removable media
1772 	 */
1773 	dp->blksize = SD_DEFAULT_BLKSIZE;
1774 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1775 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1776 
1777 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1778 	    &scsipi_sense.header, sizeof(scsipi_sense),
1779 	    flags, SDRETRIES, 6000);
1780 
1781 	if (error != 0)
1782 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1783 
1784 	dp->blksize = blksize;
1785 	if (!sd_validate_blksize(NULL, dp->blksize))
1786 		dp->blksize = _2btol(scsipi_sense.lbs);
1787 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1788 		dp->blksize = SD_DEFAULT_BLKSIZE;
1789 
1790 	/*
1791 	 * Create a pseudo-geometry.
1792 	 */
1793 	dp->heads = 64;
1794 	dp->sectors = 32;
1795 	dp->cyls = blocks / (dp->heads * dp->sectors);
1796 	dp->disksize = _5btol(scsipi_sense.size);
1797 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1798 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1799 		       (unsigned long long)dp->disksize,
1800 		       (unsigned long long)blocks);
1801 		dp->disksize = blocks;
1802 	}
1803 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1804 
1805 	return (SDGP_RESULT_OK);
1806 }
1807 
1808 /*
1809  * Get the scsi driver to send a full inquiry to the * device and use the
1810  * results to fill out the disk parameter structure.
1811  */
1812 static int
1813 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1814 {
1815 	u_int64_t blocks;
1816 	int error, blksize;
1817 #if 0
1818 	int i;
1819 	u_int8_t *p;
1820 #endif
1821 
1822 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1823 	    flags);
1824 	if (blocks == 0) {
1825 		struct scsipi_read_format_capacities cmd;
1826 		struct {
1827 			struct scsipi_capacity_list_header header;
1828 			struct scsipi_capacity_descriptor desc;
1829 		} __packed data;
1830 
1831 		memset(&cmd, 0, sizeof(cmd));
1832 		memset(&data, 0, sizeof(data));
1833 		cmd.opcode = READ_FORMAT_CAPACITIES;
1834 		_lto2b(sizeof(data), cmd.length);
1835 
1836 		error = scsipi_command(sd->sc_periph,
1837 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1838 		    SDRETRIES, 20000, NULL,
1839 		    flags | XS_CTL_DATA_IN);
1840 		if (error == EFTYPE) {
1841 			/* Medium Format Corrupted, handle as not formatted */
1842 			return (SDGP_RESULT_UNFORMATTED);
1843 		}
1844 		if (error || data.header.length == 0)
1845 			return (SDGP_RESULT_OFFLINE);
1846 
1847 #if 0
1848 printf("rfc: length=%d\n", data.header.length);
1849 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");
1850 #endif
1851 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1852 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1853 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1854 			break;
1855 
1856 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1857 			return (SDGP_RESULT_UNFORMATTED);
1858 
1859 		case SCSIPI_CAP_DESC_CODE_NONE:
1860 			return (SDGP_RESULT_OFFLINE);
1861 		}
1862 
1863 		dp->disksize = blocks = _4btol(data.desc.nblks);
1864 		if (blocks == 0)
1865 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1866 
1867 		blksize = _3btol(data.desc.blklen);
1868 
1869 	} else if (!sd_validate_blksize(NULL, blksize)) {
1870 		struct sd_mode_sense_data scsipi_sense;
1871 		int big, bsize;
1872 		struct scsi_general_block_descriptor *bdesc;
1873 
1874 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1875 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1876 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1877 		if (!error) {
1878 			if (big) {
1879 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1880 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1881 			} else {
1882 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1883 				bsize = scsipi_sense.header.small.blk_desc_len;
1884 			}
1885 
1886 #if 0
1887 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1888 printf("page 0 bsize=%d\n", bsize);
1889 printf("page 0 ok\n");
1890 #endif
1891 
1892 			if (bsize >= 8) {
1893 				blksize = _3btol(bdesc->blklen);
1894 			}
1895 		}
1896 	}
1897 
1898 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1899 		blksize = SD_DEFAULT_BLKSIZE;
1900 
1901 	dp->blksize = blksize;
1902 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1903 	return (0);
1904 }
1905 
1906 static int
1907 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1908 {
1909 	struct sd_mode_sense_data scsipi_sense;
1910 	int error;
1911 	int big, byte2;
1912 	size_t poffset;
1913 	union scsi_disk_pages *pages;
1914 
1915 	byte2 = SMS_DBD;
1916 again:
1917 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1918 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1919 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1920 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1921 	    flags | XS_CTL_SILENT, &big);
1922 	if (error) {
1923 		if (byte2 == SMS_DBD) {
1924 			/* No result; try once more with DBD off */
1925 			byte2 = 0;
1926 			goto again;
1927 		}
1928 		return (error);
1929 	}
1930 
1931 	if (big) {
1932 		poffset = sizeof scsipi_sense.header.big;
1933 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1934 	} else {
1935 		poffset = sizeof scsipi_sense.header.small;
1936 		poffset += scsipi_sense.header.small.blk_desc_len;
1937 	}
1938 
1939 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1940 		return ERESTART;
1941 
1942 	pages = (void *)((u_long)&scsipi_sense + poffset);
1943 #if 0
1944 	{
1945 		size_t i;
1946 		u_int8_t *p;
1947 
1948 		printf("page 4 sense:");
1949 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
1950 		    i--, p++)
1951 			printf(" %02x", *p);
1952 		printf("\n");
1953 		printf("page 4 pg_code=%d sense=%p/%p\n",
1954 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1955 	}
1956 #endif
1957 
1958 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1959 		return (ERESTART);
1960 
1961 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1962 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1963 	    _3btol(pages->rigid_geometry.ncyl),
1964 	    pages->rigid_geometry.nheads,
1965 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1966 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1967 	    _2btol(pages->rigid_geometry.land_zone)));
1968 
1969 	/*
1970 	 * KLUDGE!! (for zone recorded disks)
1971 	 * give a number of sectors so that sec * trks * cyls
1972 	 * is <= disk_size
1973 	 * can lead to wasted space! THINK ABOUT THIS !
1974 	 */
1975 	dp->heads = pages->rigid_geometry.nheads;
1976 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1977 	if (dp->heads == 0 || dp->cyls == 0)
1978 		return (ERESTART);
1979 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1980 
1981 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1982 	if (dp->rot_rate == 0)
1983 		dp->rot_rate = 3600;
1984 
1985 #if 0
1986 printf("page 4 ok\n");
1987 #endif
1988 	return (0);
1989 }
1990 
1991 static int
1992 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1993 {
1994 	struct sd_mode_sense_data scsipi_sense;
1995 	int error;
1996 	int big, byte2;
1997 	size_t poffset;
1998 	union scsi_disk_pages *pages;
1999 
2000 	byte2 = SMS_DBD;
2001 again:
2002 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2003 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2004 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2005 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2006 	    flags | XS_CTL_SILENT, &big);
2007 	if (error) {
2008 		if (byte2 == SMS_DBD) {
2009 			/* No result; try once more with DBD off */
2010 			byte2 = 0;
2011 			goto again;
2012 		}
2013 		return (error);
2014 	}
2015 
2016 	if (big) {
2017 		poffset = sizeof scsipi_sense.header.big;
2018 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2019 	} else {
2020 		poffset = sizeof scsipi_sense.header.small;
2021 		poffset += scsipi_sense.header.small.blk_desc_len;
2022 	}
2023 
2024 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2025 		return ERESTART;
2026 
2027 	pages = (void *)((u_long)&scsipi_sense + poffset);
2028 #if 0
2029 	{
2030 		size_t i;
2031 		u_int8_t *p;
2032 
2033 		printf("page 5 sense:");
2034 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2035 		    i--, p++)
2036 			printf(" %02x", *p);
2037 		printf("\n");
2038 		printf("page 5 pg_code=%d sense=%p/%p\n",
2039 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2040 	}
2041 #endif
2042 
2043 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2044 		return (ERESTART);
2045 
2046 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2047 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2048 	    _3btol(pages->flex_geometry.ncyl),
2049 	    pages->flex_geometry.nheads,
2050 	    pages->flex_geometry.ph_sec_tr,
2051 	    _2btol(pages->flex_geometry.bytes_s)));
2052 
2053 	dp->heads = pages->flex_geometry.nheads;
2054 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2055 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2056 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2057 		return (ERESTART);
2058 
2059 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2060 	if (dp->rot_rate == 0)
2061 		dp->rot_rate = 3600;
2062 
2063 #if 0
2064 printf("page 5 ok\n");
2065 #endif
2066 	return (0);
2067 }
2068 
2069 static int
2070 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2071 {
2072 	int error;
2073 
2074 	/*
2075 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2076 	 * cleared by the caller if necessary.
2077 	 */
2078 	if (sd->type == T_SIMPLE_DIRECT) {
2079 		error = sd_get_simplifiedparms(sd, dp, flags);
2080 		if (!error)
2081 			goto setprops;
2082 		return (error);
2083 	}
2084 
2085 	error = sd_get_capacity(sd, dp, flags);
2086 	if (error)
2087 		return (error);
2088 
2089 	if (sd->type == T_OPTICAL)
2090 		goto page0;
2091 
2092 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2093 		if (!sd_get_parms_page5(sd, dp, flags) ||
2094 		    !sd_get_parms_page4(sd, dp, flags))
2095 			goto setprops;
2096 	} else {
2097 		if (!sd_get_parms_page4(sd, dp, flags) ||
2098 		    !sd_get_parms_page5(sd, dp, flags))
2099 			goto setprops;
2100 	}
2101 
2102 page0:
2103 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2104 	/* Try calling driver's method for figuring out geometry. */
2105 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2106 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2107 		(sd->sc_periph, dp, dp->disksize)) {
2108 		/*
2109 		 * Use adaptec standard fictitious geometry
2110 		 * this depends on which controller (e.g. 1542C is
2111 		 * different. but we have to put SOMETHING here..)
2112 		 */
2113 		dp->heads = 64;
2114 		dp->sectors = 32;
2115 		dp->cyls = dp->disksize / (64 * 32);
2116 	}
2117 	dp->rot_rate = 3600;
2118 
2119 setprops:
2120 	sd_set_geometry(sd);
2121 
2122 	return (SDGP_RESULT_OK);
2123 }
2124 
2125 static int
2126 sd_flush(struct sd_softc *sd, int flags)
2127 {
2128 	struct scsipi_periph *periph = sd->sc_periph;
2129 	struct scsi_synchronize_cache_10 cmd;
2130 
2131 	/*
2132 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2133 	 * We issue with address 0 length 0, which should be
2134 	 * interpreted by the device as "all remaining blocks
2135 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2136 	 * in the event that the command is not supported by
2137 	 * the device, and poll for completion so that we know
2138 	 * that the cache has actually been flushed.
2139 	 *
2140 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2141 	 * command, as indicated by our quirks flags.
2142 	 *
2143 	 * XXX What about older devices?
2144 	 */
2145 	if (periph->periph_version < 2 ||
2146 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2147 		return (0);
2148 
2149 	sd->flags |= SDF_FLUSHING;
2150 	memset(&cmd, 0, sizeof(cmd));
2151 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2152 
2153 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2154 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2155 }
2156 
2157 static int
2158 sd_getcache(struct sd_softc *sd, int *bitsp)
2159 {
2160 	struct scsipi_periph *periph = sd->sc_periph;
2161 	struct sd_mode_sense_data scsipi_sense;
2162 	int error, bits = 0;
2163 	int big;
2164 	union scsi_disk_pages *pages;
2165 
2166 	if (periph->periph_version < 2)
2167 		return (EOPNOTSUPP);
2168 
2169 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2170 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2171 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2172 	if (error)
2173 		return (error);
2174 
2175 	if (big)
2176 		pages = (void *)(&scsipi_sense.header.big + 1);
2177 	else
2178 		pages = (void *)(&scsipi_sense.header.small + 1);
2179 
2180 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2181 		bits |= DKCACHE_READ;
2182 	if (pages->caching_params.flags & CACHING_WCE)
2183 		bits |= DKCACHE_WRITE;
2184 	if (pages->caching_params.pg_code & PGCODE_PS)
2185 		bits |= DKCACHE_SAVE;
2186 
2187 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2188 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2189 	    sizeof(scsipi_sense.pages.caching_params),
2190 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2191 	if (error == 0) {
2192 		if (big)
2193 			pages = (void *)(&scsipi_sense.header.big + 1);
2194 		else
2195 			pages = (void *)(&scsipi_sense.header.small + 1);
2196 
2197 		if (pages->caching_params.flags & CACHING_RCD)
2198 			bits |= DKCACHE_RCHANGE;
2199 		if (pages->caching_params.flags & CACHING_WCE)
2200 			bits |= DKCACHE_WCHANGE;
2201 	}
2202 
2203 	*bitsp = bits;
2204 
2205 	return (0);
2206 }
2207 
2208 static int
2209 sd_setcache(struct sd_softc *sd, int bits)
2210 {
2211 	struct scsipi_periph *periph = sd->sc_periph;
2212 	struct sd_mode_sense_data scsipi_sense;
2213 	int error;
2214 	uint8_t oflags, byte2 = 0;
2215 	int big;
2216 	union scsi_disk_pages *pages;
2217 
2218 	if (periph->periph_version < 2)
2219 		return (EOPNOTSUPP);
2220 
2221 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2222 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2223 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2224 	if (error)
2225 		return (error);
2226 
2227 	if (big)
2228 		pages = (void *)(&scsipi_sense.header.big + 1);
2229 	else
2230 		pages = (void *)(&scsipi_sense.header.small + 1);
2231 
2232 	oflags = pages->caching_params.flags;
2233 
2234 	if (bits & DKCACHE_READ)
2235 		pages->caching_params.flags &= ~CACHING_RCD;
2236 	else
2237 		pages->caching_params.flags |= CACHING_RCD;
2238 
2239 	if (bits & DKCACHE_WRITE)
2240 		pages->caching_params.flags |= CACHING_WCE;
2241 	else
2242 		pages->caching_params.flags &= ~CACHING_WCE;
2243 
2244 	if (oflags == pages->caching_params.flags)
2245 		return (0);
2246 
2247 	pages->caching_params.pg_code &= PGCODE_MASK;
2248 
2249 	if (bits & DKCACHE_SAVE)
2250 		byte2 |= SMS_SP;
2251 
2252 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2253 	    sizeof(struct scsi_mode_page_header) +
2254 	    pages->caching_params.pg_length, 0, big));
2255 }
2256 
2257 static void
2258 sd_set_geometry(struct sd_softc *sd)
2259 {
2260 	struct disk_geom *dg = &sd->sc_dk.dk_geom;
2261 
2262 	memset(dg, 0, sizeof(*dg));
2263 
2264 	dg->dg_secperunit = sd->params.disksize;
2265 	dg->dg_secsize = sd->params.blksize;
2266 	dg->dg_nsectors = sd->params.sectors;
2267 	dg->dg_ntracks = sd->params.heads;
2268 	dg->dg_ncylinders = sd->params.cyls;
2269 
2270 	disk_set_info(sd->sc_dev, &sd->sc_dk, NULL);
2271 }
2272