xref: /netbsd-src/sys/dev/scsipi/sd.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: sd.c,v 1.296 2012/02/02 19:43:06 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 1.296 2012/02/02 19:43:06 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_properties(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 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
161 };
162 
163 const struct cdevsw sd_cdevsw = {
164 	sdopen, sdclose, sdread, sdwrite, sdioctl,
165 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
166 };
167 
168 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
169 
170 static const struct scsipi_periphsw sd_switch = {
171 	sd_interpret_sense,	/* check our error handler first */
172 	sdstart,		/* have a queue, served by this */
173 	NULL,			/* have no async handler */
174 	sddone,			/* deal with stats at interrupt time */
175 };
176 
177 struct sd_mode_sense_data {
178 	/*
179 	 * XXX
180 	 * We are not going to parse this as-is -- it just has to be large
181 	 * enough.
182 	 */
183 	union {
184 		struct scsi_mode_parameter_header_6 small;
185 		struct scsi_mode_parameter_header_10 big;
186 	} header;
187 	struct scsi_general_block_descriptor blk_desc;
188 	union scsi_disk_pages pages;
189 };
190 
191 /*
192  * The routine called by the low level scsi routine when it discovers
193  * A device suitable for this driver
194  */
195 static int
196 sdmatch(device_t parent, cfdata_t match,
197     void *aux)
198 {
199 	struct scsipibus_attach_args *sa = aux;
200 	int priority;
201 
202 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
203 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
204 	    sizeof(sd_patterns[0]), &priority);
205 
206 	return (priority);
207 }
208 
209 /*
210  * Attach routine common to atapi & scsi.
211  */
212 static void
213 sdattach(device_t parent, device_t self, void *aux)
214 {
215 	struct sd_softc *sd = device_private(self);
216 	struct scsipibus_attach_args *sa = aux;
217 	struct scsipi_periph *periph = sa->sa_periph;
218 	int error, result, rndval = cprng_strong32();
219 	struct disk_parms *dp = &sd->params;
220 	char pbuf[9];
221 
222 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
223 
224 	sd->sc_dev = self;
225 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
226 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
227 	if (sd->type == T_SIMPLE_DIRECT)
228 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
229 
230 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
231 	    periph->periph_version == 0)
232 		sd->flags |= SDF_ANCIENT;
233 
234 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
235 
236 	callout_init(&sd->sc_callout, 0);
237 
238 	/*
239 	 * Store information needed to contact our base driver
240 	 */
241 	sd->sc_periph = periph;
242 
243 	periph->periph_dev = sd->sc_dev;
244 	periph->periph_switch = &sd_switch;
245 
246         /*
247          * Increase our openings to the maximum-per-periph
248          * supported by the adapter.  This will either be
249          * clamped down or grown by the adapter if necessary.
250          */
251 	periph->periph_openings =
252 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
253 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
254 
255 	/*
256 	 * Initialize and attach the disk structure.
257 	 */
258 	disk_init(&sd->sc_dk, device_xname(sd->sc_dev), &sddkdriver);
259 	disk_attach(&sd->sc_dk);
260 
261 	/*
262 	 * Use the subdriver to request information regarding the drive.
263 	 */
264 	aprint_naive("\n");
265 	aprint_normal("\n");
266 
267 	error = scsipi_test_unit_ready(periph,
268 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
269 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
270 
271 	if (error)
272 		result = SDGP_RESULT_OFFLINE;
273 	else
274 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
275 	aprint_normal_dev(sd->sc_dev, "");
276 	switch (result) {
277 	case SDGP_RESULT_OK:
278 		format_bytes(pbuf, sizeof(pbuf),
279 		    (u_int64_t)dp->disksize * dp->blksize);
280 	        aprint_normal(
281 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
282 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
283 		    (unsigned long long)dp->disksize);
284 		break;
285 
286 	case SDGP_RESULT_OFFLINE:
287 		aprint_normal("drive offline");
288 		break;
289 
290 	case SDGP_RESULT_UNFORMATTED:
291 		aprint_normal("unformatted media");
292 		break;
293 
294 #ifdef DIAGNOSTIC
295 	default:
296 		panic("sdattach: unknown result from get_parms");
297 		break;
298 #endif
299 	}
300 	aprint_normal("\n");
301 
302 	/*
303 	 * Establish a shutdown hook so that we can ensure that
304 	 * our data has actually made it onto the platter at
305 	 * shutdown time.  Note that this relies on the fact
306 	 * that the shutdown hooks at the "leaves" of the device tree
307 	 * are run, first (thus guaranteeing that our hook runs before
308 	 * our ancestors').
309 	 */
310 	if (!pmf_device_register1(self, sd_suspend, NULL, sd_shutdown))
311 		aprint_error_dev(self, "couldn't establish power handler\n");
312 
313 	/*
314 	 * attach the device into the random source list
315 	 */
316 	rnd_attach_source(&sd->rnd_source, device_xname(sd->sc_dev),
317 			  RND_TYPE_DISK, 0);
318 
319 	/* Discover wedges on this disk. */
320 	dkwedge_discover(&sd->sc_dk);
321 
322 	/*
323 	 * Disk insertion and removal times can be a useful source
324 	 * of entropy, though the estimator should never _count_
325 	 * these bits, on insertion, because the deltas to the
326 	 * nonexistent) previous event should never allow it.
327 	 */
328 	rnd_add_uint32(&sd->rnd_source, rndval);
329 }
330 
331 static int
332 sddetach(device_t self, int flags)
333 {
334 	struct sd_softc *sd = device_private(self);
335 	int s, bmaj, cmaj, i, mn, rc, rndval = cprng_strong32();
336 
337 	rnd_add_uint32(&sd->rnd_source, rndval);
338 
339 	if ((rc = disk_begindetach(&sd->sc_dk, sdlastclose, self, flags)) != 0)
340 		return rc;
341 
342 	/* locate the major number */
343 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
344 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
345 
346 	/* Nuke the vnodes for any open instances */
347 	for (i = 0; i < MAXPARTITIONS; i++) {
348 		mn = SDMINOR(device_unit(self), i);
349 		vdevgone(bmaj, mn, mn, VBLK);
350 		vdevgone(cmaj, mn, mn, VCHR);
351 	}
352 
353 	/* kill any pending restart */
354 	callout_stop(&sd->sc_callout);
355 
356 	/* Delete all of our wedges. */
357 	dkwedge_delall(&sd->sc_dk);
358 
359 	s = splbio();
360 
361 	/* Kill off any queued buffers. */
362 	bufq_drain(sd->buf_queue);
363 
364 	bufq_free(sd->buf_queue);
365 
366 	/* Kill off any pending commands. */
367 	scsipi_kill_pending(sd->sc_periph);
368 
369 	splx(s);
370 
371 	/* Detach from the disk list. */
372 	disk_detach(&sd->sc_dk);
373 	disk_destroy(&sd->sc_dk);
374 
375 	callout_destroy(&sd->sc_callout);
376 
377 	pmf_device_deregister(self);
378 
379 	/* Unhook the entropy source. */
380 	rnd_detach_source(&sd->rnd_source);
381 
382 	return (0);
383 }
384 
385 /*
386  * open the device. Make sure the partition info is a up-to-date as can be.
387  */
388 static int
389 sdopen(dev_t dev, int flag, int fmt, struct lwp *l)
390 {
391 	struct sd_softc *sd;
392 	struct scsipi_periph *periph;
393 	struct scsipi_adapter *adapt;
394 	int unit, part;
395 	int error;
396 
397 	unit = SDUNIT(dev);
398 	sd = device_lookup_private(&sd_cd, unit);
399 	if (sd == NULL)
400 		return (ENXIO);
401 
402 	if (!device_is_active(sd->sc_dev))
403 		return (ENODEV);
404 
405 	part = SDPART(dev);
406 
407 	mutex_enter(&sd->sc_dk.dk_openlock);
408 
409 	/*
410 	 * If there are wedges, and this is not RAW_PART, then we
411 	 * need to fail.
412 	 */
413 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
414 		error = EBUSY;
415 		goto bad1;
416 	}
417 
418 	periph = sd->sc_periph;
419 	adapt = periph->periph_channel->chan_adapter;
420 
421 	SC_DEBUG(periph, SCSIPI_DB1,
422 	    ("sdopen: dev=0x%"PRIx64" (unit %d (of %d), partition %d)\n", dev, unit,
423 	    sd_cd.cd_ndevs, part));
424 
425 	/*
426 	 * If this is the first open of this device, add a reference
427 	 * to the adapter.
428 	 */
429 	if (sd->sc_dk.dk_openmask == 0 &&
430 	    (error = scsipi_adapter_addref(adapt)) != 0)
431 		goto bad1;
432 
433 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
434 		/*
435 		 * If any partition is open, but the disk has been invalidated,
436 		 * disallow further opens of non-raw partition
437 		 */
438 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
439 		    (part != RAW_PART || fmt != S_IFCHR)) {
440 			error = EIO;
441 			goto bad2;
442 		}
443 	} else {
444 		int silent;
445 
446 		if ((part == RAW_PART && fmt == S_IFCHR) || (flag & FSILENT))
447 			silent = XS_CTL_SILENT;
448 		else
449 			silent = 0;
450 
451 		/* Check that it is still responding and ok. */
452 		error = scsipi_test_unit_ready(periph,
453 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
454 		    silent);
455 
456 		/*
457 		 * Start the pack spinning if necessary. Always allow the
458 		 * raw parition to be opened, for raw IOCTLs. Data transfers
459 		 * will check for SDEV_MEDIA_LOADED.
460 		 */
461 		if (error == EIO) {
462 			int error2;
463 
464 			error2 = scsipi_start(periph, SSS_START, silent);
465 			switch (error2) {
466 			case 0:
467 				error = 0;
468 				break;
469 			case EIO:
470 			case EINVAL:
471 				break;
472 			default:
473 				error = error2;
474 				break;
475 			}
476 		}
477 		if (error) {
478 			if (silent && (flag & FSILENT) == 0)
479 				goto out;
480 			goto bad2;
481 		}
482 
483 		periph->periph_flags |= PERIPH_OPEN;
484 
485 		if (periph->periph_flags & PERIPH_REMOVABLE) {
486 			/* Lock the pack in. */
487 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
488 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
489 			    XS_CTL_IGNORE_MEDIA_CHANGE |
490 			    XS_CTL_SILENT);
491 			if (error)
492 				goto bad3;
493 		}
494 
495 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
496 			int param_error;
497 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
498 
499 			/*
500 			 * Load the physical device parameters.
501 			 *
502 			 * Note that if media is present but unformatted,
503 			 * we allow the open (so that it can be formatted!).
504 			 * The drive should refuse real I/O, if the media is
505 			 * unformatted.
506 			 */
507 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
508 			     == SDGP_RESULT_OFFLINE) {
509 				error = ENXIO;
510 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
511 				goto bad3;
512 			}
513 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
514 
515 			/* Load the partition info if not already loaded. */
516 			if (param_error == 0) {
517 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
518 					error = EIO;
519 					goto bad3;
520 				}
521 				SC_DEBUG(periph, SCSIPI_DB3,
522 				     ("Disklabel loaded "));
523 			}
524 		}
525 	}
526 
527 	/* Check that the partition exists. */
528 	if (part != RAW_PART &&
529 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
530 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
531 		error = ENXIO;
532 		goto bad3;
533 	}
534 
535  out:	/* Insure only one open at a time. */
536 	switch (fmt) {
537 	case S_IFCHR:
538 		sd->sc_dk.dk_copenmask |= (1 << part);
539 		break;
540 	case S_IFBLK:
541 		sd->sc_dk.dk_bopenmask |= (1 << part);
542 		break;
543 	}
544 	sd->sc_dk.dk_openmask =
545 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
546 
547 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
548 	mutex_exit(&sd->sc_dk.dk_openlock);
549 	return (0);
550 
551  bad3:
552 	if (sd->sc_dk.dk_openmask == 0) {
553 		if (periph->periph_flags & PERIPH_REMOVABLE)
554 			scsipi_prevent(periph, SPAMR_ALLOW,
555 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
556 			    XS_CTL_IGNORE_MEDIA_CHANGE |
557 			    XS_CTL_SILENT);
558 		periph->periph_flags &= ~PERIPH_OPEN;
559 	}
560 
561  bad2:
562 	if (sd->sc_dk.dk_openmask == 0)
563 		scsipi_adapter_delref(adapt);
564 
565  bad1:
566 	mutex_exit(&sd->sc_dk.dk_openlock);
567 	return (error);
568 }
569 
570 /*
571  * Caller must hold sd->sc_dk.dk_openlock.
572  */
573 static int
574 sdlastclose(device_t self)
575 {
576 	struct sd_softc *sd = device_private(self);
577 	struct scsipi_periph *periph = sd->sc_periph;
578 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
579 
580 	/*
581 	 * If the disk cache needs flushing, and the disk supports
582 	 * it, do it now.
583 	 */
584 	if ((sd->flags & SDF_DIRTY) != 0) {
585 		if (sd_flush(sd, 0)) {
586 			aprint_error_dev(sd->sc_dev,
587 				"cache synchronization failed\n");
588 			sd->flags &= ~SDF_FLUSHING;
589 		} else
590 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
591 	}
592 
593 	scsipi_wait_drain(periph);
594 
595 	if (periph->periph_flags & PERIPH_REMOVABLE)
596 		scsipi_prevent(periph, SPAMR_ALLOW,
597 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
598 		    XS_CTL_IGNORE_NOT_READY |
599 		    XS_CTL_SILENT);
600 	periph->periph_flags &= ~PERIPH_OPEN;
601 
602 	scsipi_wait_drain(periph);
603 
604 	scsipi_adapter_delref(adapt);
605 
606 	return 0;
607 }
608 
609 /*
610  * close the device.. only called if we are the LAST occurence of an open
611  * device.  Convenient now but usually a pain.
612  */
613 static int
614 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
615 {
616 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
617 	int part = SDPART(dev);
618 
619 	mutex_enter(&sd->sc_dk.dk_openlock);
620 	switch (fmt) {
621 	case S_IFCHR:
622 		sd->sc_dk.dk_copenmask &= ~(1 << part);
623 		break;
624 	case S_IFBLK:
625 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
626 		break;
627 	}
628 	sd->sc_dk.dk_openmask =
629 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
630 
631 	if (sd->sc_dk.dk_openmask == 0)
632 		sdlastclose(sd->sc_dev);
633 
634 	mutex_exit(&sd->sc_dk.dk_openlock);
635 	return (0);
636 }
637 
638 /*
639  * Actually translate the requested transfer into one the physical driver
640  * can understand.  The transfer is described by a buf and will include
641  * only one physical transfer.
642  */
643 static void
644 sdstrategy(struct buf *bp)
645 {
646 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
647 	struct scsipi_periph *periph = sd->sc_periph;
648 	struct disklabel *lp;
649 	daddr_t blkno;
650 	int s;
651 	bool sector_aligned;
652 
653 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
654 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
655 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
656 	/*
657 	 * If the device has been made invalid, error out
658 	 */
659 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
660 	    !device_is_active(sd->sc_dev)) {
661 		if (periph->periph_flags & PERIPH_OPEN)
662 			bp->b_error = EIO;
663 		else
664 			bp->b_error = ENODEV;
665 		goto done;
666 	}
667 
668 	lp = sd->sc_dk.dk_label;
669 
670 	/*
671 	 * The transfer must be a whole number of blocks, offset must not be
672 	 * negative.
673 	 */
674 	if (lp->d_secsize == DEV_BSIZE) {
675 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
676 	} else {
677 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
678 	}
679 	if (!sector_aligned || bp->b_blkno < 0) {
680 		bp->b_error = EINVAL;
681 		goto done;
682 	}
683 	/*
684 	 * If it's a null transfer, return immediatly
685 	 */
686 	if (bp->b_bcount == 0)
687 		goto done;
688 
689 	/*
690 	 * Do bounds checking, adjust transfer. if error, process.
691 	 * If end of partition, just return.
692 	 */
693 	if (SDPART(bp->b_dev) == RAW_PART) {
694 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
695 		    sd->params.disksize512) <= 0)
696 			goto done;
697 	} else {
698 		if (bounds_check_with_label(&sd->sc_dk, bp,
699 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
700 			goto done;
701 	}
702 
703 	/*
704 	 * Now convert the block number to absolute and put it in
705 	 * terms of the device's logical block size.
706 	 */
707 	if (lp->d_secsize == DEV_BSIZE)
708 		blkno = bp->b_blkno;
709 	else if (lp->d_secsize > DEV_BSIZE)
710 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
711 	else
712 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
713 
714 	if (SDPART(bp->b_dev) != RAW_PART)
715 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
716 
717 	bp->b_rawblkno = blkno;
718 
719 	s = splbio();
720 
721 	/*
722 	 * Place it in the queue of disk activities for this disk.
723 	 *
724 	 * XXX Only do disksort() if the current operating mode does not
725 	 * XXX include tagged queueing.
726 	 */
727 	bufq_put(sd->buf_queue, bp);
728 
729 	/*
730 	 * Tell the device to get going on the transfer if it's
731 	 * not doing anything, otherwise just wait for completion
732 	 */
733 	sdstart(sd->sc_periph);
734 
735 	splx(s);
736 	return;
737 
738 done:
739 	/*
740 	 * Correctly set the buf to indicate a completed xfer
741 	 */
742 	bp->b_resid = bp->b_bcount;
743 	biodone(bp);
744 }
745 
746 /*
747  * sdstart looks to see if there is a buf waiting for the device
748  * and that the device is not already busy. If both are true,
749  * It dequeues the buf and creates a scsi command to perform the
750  * transfer in the buf. The transfer request will call scsipi_done
751  * on completion, which will in turn call this routine again
752  * so that the next queued transfer is performed.
753  * The bufs are queued by the strategy routine (sdstrategy)
754  *
755  * This routine is also called after other non-queued requests
756  * have been made of the scsi driver, to ensure that the queue
757  * continues to be drained.
758  *
759  * must be called at the correct (highish) spl level
760  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
761  */
762 static void
763 sdstart(struct scsipi_periph *periph)
764 {
765 	struct sd_softc *sd = device_private(periph->periph_dev);
766 	struct disklabel *lp = sd->sc_dk.dk_label;
767 	struct buf *bp = 0;
768 	struct scsipi_rw_16 cmd16;
769 	struct scsipi_rw_10 cmd_big;
770 	struct scsi_rw_6 cmd_small;
771 	struct scsipi_generic *cmdp;
772 	struct scsipi_xfer *xs;
773 	int nblks, cmdlen, error, flags;
774 
775 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
776 	/*
777 	 * Check if the device has room for another command
778 	 */
779 	while (periph->periph_active < periph->periph_openings) {
780 		/*
781 		 * there is excess capacity, but a special waits
782 		 * It'll need the adapter as soon as we clear out of the
783 		 * way and let it run (user level wait).
784 		 */
785 		if (periph->periph_flags & PERIPH_WAITING) {
786 			periph->periph_flags &= ~PERIPH_WAITING;
787 			wakeup((void *)periph);
788 			return;
789 		}
790 
791 		/*
792 		 * If the device has become invalid, abort all the
793 		 * reads and writes until all files have been closed and
794 		 * re-opened
795 		 */
796 		if (__predict_false(
797 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
798 			if ((bp = bufq_get(sd->buf_queue)) != NULL) {
799 				bp->b_error = EIO;
800 				bp->b_resid = bp->b_bcount;
801 				biodone(bp);
802 				continue;
803 			} else {
804 				return;
805 			}
806 		}
807 
808 		/*
809 		 * See if there is a buf with work for us to do..
810 		 */
811 		if ((bp = bufq_peek(sd->buf_queue)) == NULL)
812 			return;
813 
814 		/*
815 		 * We have a buf, now we should make a command.
816 		 */
817 
818 		if (lp->d_secsize == DEV_BSIZE)
819 			nblks = bp->b_bcount >> DEV_BSHIFT;
820 		else
821 			nblks = howmany(bp->b_bcount, lp->d_secsize);
822 
823 		/*
824 		 * Fill out the scsi command.  Use the smallest CDB possible
825 		 * (6-byte, 10-byte, or 16-byte).
826 		 */
827 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
828 		    ((nblks & 0xff) == nblks) &&
829 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
830 			/* 6-byte CDB */
831 			memset(&cmd_small, 0, sizeof(cmd_small));
832 			cmd_small.opcode = (bp->b_flags & B_READ) ?
833 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
834 			_lto3b(bp->b_rawblkno, cmd_small.addr);
835 			cmd_small.length = nblks & 0xff;
836 			cmdlen = sizeof(cmd_small);
837 			cmdp = (struct scsipi_generic *)&cmd_small;
838 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
839 			/* 10-byte CDB */
840 			memset(&cmd_big, 0, sizeof(cmd_big));
841 			cmd_big.opcode = (bp->b_flags & B_READ) ?
842 			    READ_10 : WRITE_10;
843 			_lto4b(bp->b_rawblkno, cmd_big.addr);
844 			_lto2b(nblks, cmd_big.length);
845 			cmdlen = sizeof(cmd_big);
846 			cmdp = (struct scsipi_generic *)&cmd_big;
847 		} else {
848 			/* 16-byte CDB */
849 			memset(&cmd16, 0, sizeof(cmd16));
850 			cmd16.opcode = (bp->b_flags & B_READ) ?
851 			    READ_16 : WRITE_16;
852 			_lto8b(bp->b_rawblkno, cmd16.addr);
853 			_lto4b(nblks, cmd16.length);
854 			cmdlen = sizeof(cmd16);
855 			cmdp = (struct scsipi_generic *)&cmd16;
856 		}
857 
858 		/* Instrumentation. */
859 		disk_busy(&sd->sc_dk);
860 
861 		/*
862 		 * Mark the disk dirty so that the cache will be
863 		 * flushed on close.
864 		 */
865 		if ((bp->b_flags & B_READ) == 0)
866 			sd->flags |= SDF_DIRTY;
867 
868 		/*
869 		 * Figure out what flags to use.
870 		 */
871 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
872 		if (bp->b_flags & B_READ)
873 			flags |= XS_CTL_DATA_IN;
874 		else
875 			flags |= XS_CTL_DATA_OUT;
876 
877 		/*
878 		 * Call the routine that chats with the adapter.
879 		 * Note: we cannot sleep as we may be an interrupt
880 		 */
881 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
882 		    (u_char *)bp->b_data, bp->b_bcount,
883 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
884 		if (__predict_false(xs == NULL)) {
885 			/*
886 			 * out of memory. Keep this buffer in the queue, and
887 			 * retry later.
888 			 */
889 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
890 			    periph);
891 			return;
892 		}
893 		/*
894 		 * need to dequeue the buffer before queuing the command,
895 		 * because cdstart may be called recursively from the
896 		 * HBA driver
897 		 */
898 #ifdef DIAGNOSTIC
899 		if (bufq_get(sd->buf_queue) != bp)
900 			panic("sdstart(): dequeued wrong buf");
901 #else
902 		bufq_get(sd->buf_queue);
903 #endif
904 		error = scsipi_execute_xs(xs);
905 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
906 		KASSERT(error == 0);
907 	}
908 }
909 
910 static void
911 sdrestart(void *v)
912 {
913 	int s = splbio();
914 	sdstart((struct scsipi_periph *)v);
915 	splx(s);
916 }
917 
918 static void
919 sddone(struct scsipi_xfer *xs, int error)
920 {
921 	struct sd_softc *sd = device_private(xs->xs_periph->periph_dev);
922 	struct buf *bp = xs->bp;
923 
924 	if (sd->flags & SDF_FLUSHING) {
925 		/* Flush completed, no longer dirty. */
926 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
927 	}
928 
929 	if (bp) {
930 		bp->b_error = error;
931 		bp->b_resid = xs->resid;
932 		if (error) {
933 			/* on a read/write error bp->b_resid is zero, so fix */
934 			bp->b_resid = bp->b_bcount;
935 		}
936 
937 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
938 		    (bp->b_flags & B_READ));
939 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
940 
941 		biodone(bp);
942 	}
943 }
944 
945 static void
946 sdminphys(struct buf *bp)
947 {
948 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
949 	long xmax;
950 
951 	/*
952 	 * If the device is ancient, we want to make sure that
953 	 * the transfer fits into a 6-byte cdb.
954 	 *
955 	 * XXX Note that the SCSI-I spec says that 256-block transfers
956 	 * are allowed in a 6-byte read/write, and are specified
957 	 * by settng the "length" to 0.  However, we're conservative
958 	 * here, allowing only 255-block transfers in case an
959 	 * ancient device gets confused by length == 0.  A length of 0
960 	 * in a 10-byte read/write actually means 0 blocks.
961 	 */
962 	if ((sd->flags & SDF_ANCIENT) &&
963 	    ((sd->sc_periph->periph_flags &
964 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
965 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
966 
967 		if (bp->b_bcount > xmax)
968 			bp->b_bcount = xmax;
969 	}
970 
971 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
972 }
973 
974 static int
975 sdread(dev_t dev, struct uio *uio, int ioflag)
976 {
977 
978 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
979 }
980 
981 static int
982 sdwrite(dev_t dev, struct uio *uio, int ioflag)
983 {
984 
985 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
986 }
987 
988 /*
989  * Perform special action on behalf of the user
990  * Knows about the internals of this device
991  */
992 static int
993 sdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
994 {
995 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
996 	struct scsipi_periph *periph = sd->sc_periph;
997 	int part = SDPART(dev);
998 	int error = 0;
999 	int s;
1000 #ifdef __HAVE_OLD_DISKLABEL
1001 	struct disklabel *newlabel = NULL;
1002 #endif
1003 
1004 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1005 
1006 	/*
1007 	 * If the device is not valid, some IOCTLs can still be
1008 	 * handled on the raw partition. Check this here.
1009 	 */
1010 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1011 		switch (cmd) {
1012 		case DIOCKLABEL:
1013 		case DIOCWLABEL:
1014 		case DIOCLOCK:
1015 		case DIOCEJECT:
1016 		case ODIOCEJECT:
1017 		case DIOCGCACHE:
1018 		case DIOCSCACHE:
1019 		case DIOCGSTRATEGY:
1020 		case DIOCSSTRATEGY:
1021 		case SCIOCIDENTIFY:
1022 		case OSCIOCIDENTIFY:
1023 		case SCIOCCOMMAND:
1024 		case SCIOCDEBUG:
1025 			if (part == RAW_PART)
1026 				break;
1027 		/* FALLTHROUGH */
1028 		default:
1029 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1030 				return (ENODEV);
1031 			else
1032 				return (EIO);
1033 		}
1034 	}
1035 
1036 	error = disk_ioctl(&sd->sc_dk, cmd, addr, flag, l);
1037 	if (error != EPASSTHROUGH)
1038 		return (error);
1039 
1040 	switch (cmd) {
1041 	case DIOCGDINFO:
1042 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1043 		return (0);
1044 
1045 #ifdef __HAVE_OLD_DISKLABEL
1046 	case ODIOCGDINFO:
1047 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1048 		if (newlabel == NULL)
1049 			return EIO;
1050 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1051 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1052 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1053 		else
1054 			error = ENOTTY;
1055 		free(newlabel, M_TEMP);
1056 		return error;
1057 #endif
1058 
1059 	case DIOCGPART:
1060 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1061 		((struct partinfo *)addr)->part =
1062 		    &sd->sc_dk.dk_label->d_partitions[part];
1063 		return (0);
1064 
1065 	case DIOCWDINFO:
1066 	case DIOCSDINFO:
1067 #ifdef __HAVE_OLD_DISKLABEL
1068 	case ODIOCWDINFO:
1069 	case ODIOCSDINFO:
1070 #endif
1071 	{
1072 		struct disklabel *lp;
1073 
1074 		if ((flag & FWRITE) == 0)
1075 			return (EBADF);
1076 
1077 #ifdef __HAVE_OLD_DISKLABEL
1078  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1079 			newlabel = malloc(sizeof *newlabel, M_TEMP,
1080 			    M_WAITOK | M_ZERO);
1081 			if (newlabel == NULL)
1082 				return EIO;
1083 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1084 			lp = newlabel;
1085 		} else
1086 #endif
1087 		lp = (struct disklabel *)addr;
1088 
1089 		mutex_enter(&sd->sc_dk.dk_openlock);
1090 		sd->flags |= SDF_LABELLING;
1091 
1092 		error = setdisklabel(sd->sc_dk.dk_label,
1093 		    lp, /*sd->sc_dk.dk_openmask : */0,
1094 		    sd->sc_dk.dk_cpulabel);
1095 		if (error == 0) {
1096 			if (cmd == DIOCWDINFO
1097 #ifdef __HAVE_OLD_DISKLABEL
1098 			    || cmd == ODIOCWDINFO
1099 #endif
1100 			   )
1101 				error = writedisklabel(SDLABELDEV(dev),
1102 				    sdstrategy, sd->sc_dk.dk_label,
1103 				    sd->sc_dk.dk_cpulabel);
1104 		}
1105 
1106 		sd->flags &= ~SDF_LABELLING;
1107 		mutex_exit(&sd->sc_dk.dk_openlock);
1108 #ifdef __HAVE_OLD_DISKLABEL
1109 		if (newlabel != NULL)
1110 			free(newlabel, M_TEMP);
1111 #endif
1112 		return (error);
1113 	}
1114 
1115 	case DIOCKLABEL:
1116 		if (*(int *)addr)
1117 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1118 		else
1119 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1120 		return (0);
1121 
1122 	case DIOCWLABEL:
1123 		if ((flag & FWRITE) == 0)
1124 			return (EBADF);
1125 		if (*(int *)addr)
1126 			sd->flags |= SDF_WLABEL;
1127 		else
1128 			sd->flags &= ~SDF_WLABEL;
1129 		return (0);
1130 
1131 	case DIOCLOCK:
1132 		if (periph->periph_flags & PERIPH_REMOVABLE)
1133 			return (scsipi_prevent(periph,
1134 			    (*(int *)addr) ?
1135 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1136 		else
1137 			return (ENOTTY);
1138 
1139 	case DIOCEJECT:
1140 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1141 			return (ENOTTY);
1142 		if (*(int *)addr == 0) {
1143 			/*
1144 			 * Don't force eject: check that we are the only
1145 			 * partition open. If so, unlock it.
1146 			 */
1147 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1148 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1149 			    sd->sc_dk.dk_openmask) {
1150 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1151 				    XS_CTL_IGNORE_NOT_READY);
1152 				if (error)
1153 					return (error);
1154 			} else {
1155 				return (EBUSY);
1156 			}
1157 		}
1158 		/* FALLTHROUGH */
1159 	case ODIOCEJECT:
1160 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1161 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1162 
1163 	case DIOCGDEFLABEL:
1164 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1165 		return (0);
1166 
1167 #ifdef __HAVE_OLD_DISKLABEL
1168 	case ODIOCGDEFLABEL:
1169 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1170 		if (newlabel == NULL)
1171 			return EIO;
1172 		sdgetdefaultlabel(sd, newlabel);
1173 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1174 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1175 		else
1176 			error = ENOTTY;
1177 		free(newlabel, M_TEMP);
1178 		return error;
1179 #endif
1180 
1181 	case DIOCGCACHE:
1182 		return (sd_getcache(sd, (int *) addr));
1183 
1184 	case DIOCSCACHE:
1185 		if ((flag & FWRITE) == 0)
1186 			return (EBADF);
1187 		return (sd_setcache(sd, *(int *) addr));
1188 
1189 	case DIOCCACHESYNC:
1190 		/*
1191 		 * XXX Do we really need to care about having a writable
1192 		 * file descriptor here?
1193 		 */
1194 		if ((flag & FWRITE) == 0)
1195 			return (EBADF);
1196 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1197 			error = sd_flush(sd, 0);
1198 			if (error)
1199 				sd->flags &= ~SDF_FLUSHING;
1200 			else
1201 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1202 		} else
1203 			error = 0;
1204 		return (error);
1205 
1206 	case DIOCAWEDGE:
1207 	    {
1208 	    	struct dkwedge_info *dkw = (void *) addr;
1209 
1210 		if ((flag & FWRITE) == 0)
1211 			return (EBADF);
1212 
1213 		/* If the ioctl happens here, the parent is us. */
1214 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1215 			sizeof(dkw->dkw_parent));
1216 		return (dkwedge_add(dkw));
1217 	    }
1218 
1219 	case DIOCDWEDGE:
1220 	    {
1221 	    	struct dkwedge_info *dkw = (void *) addr;
1222 
1223 		if ((flag & FWRITE) == 0)
1224 			return (EBADF);
1225 
1226 		/* If the ioctl happens here, the parent is us. */
1227 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1228 			sizeof(dkw->dkw_parent));
1229 		return (dkwedge_del(dkw));
1230 	    }
1231 
1232 	case DIOCLWEDGES:
1233 	    {
1234 	    	struct dkwedge_list *dkwl = (void *) addr;
1235 
1236 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1237 	    }
1238 
1239 	case DIOCGSTRATEGY:
1240 	    {
1241 		struct disk_strategy *dks = addr;
1242 
1243 		s = splbio();
1244 		strlcpy(dks->dks_name, bufq_getstrategyname(sd->buf_queue),
1245 		    sizeof(dks->dks_name));
1246 		splx(s);
1247 		dks->dks_paramlen = 0;
1248 
1249 		return 0;
1250 	    }
1251 
1252 	case DIOCSSTRATEGY:
1253 	    {
1254 		struct disk_strategy *dks = addr;
1255 		struct bufq_state *new;
1256 		struct bufq_state *old;
1257 
1258 		if ((flag & FWRITE) == 0) {
1259 			return EBADF;
1260 		}
1261 
1262 		if (dks->dks_param != NULL) {
1263 			return EINVAL;
1264 		}
1265 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1266 		error = bufq_alloc(&new, dks->dks_name,
1267 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1268 		if (error) {
1269 			return error;
1270 		}
1271 		s = splbio();
1272 		old = sd->buf_queue;
1273 		bufq_move(new, old);
1274 		sd->buf_queue = new;
1275 		splx(s);
1276 		bufq_free(old);
1277 
1278 		return 0;
1279 	    }
1280 
1281 	default:
1282 		if (part != RAW_PART)
1283 			return (ENOTTY);
1284 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1285 	}
1286 
1287 #ifdef DIAGNOSTIC
1288 	panic("sdioctl: impossible");
1289 #endif
1290 }
1291 
1292 static void
1293 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1294 {
1295 
1296 	memset(lp, 0, sizeof(struct disklabel));
1297 
1298 	lp->d_secsize = sd->params.blksize;
1299 	lp->d_ntracks = sd->params.heads;
1300 	lp->d_nsectors = sd->params.sectors;
1301 	lp->d_ncylinders = sd->params.cyls;
1302 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1303 
1304 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1305 	case SCSIPI_BUSTYPE_SCSI:
1306 		lp->d_type = DTYPE_SCSI;
1307 		break;
1308 	case SCSIPI_BUSTYPE_ATAPI:
1309 		lp->d_type = DTYPE_ATAPI;
1310 		break;
1311 	}
1312 	/*
1313 	 * XXX
1314 	 * We could probe the mode pages to figure out what kind of disc it is.
1315 	 * Is this worthwhile?
1316 	 */
1317 	strncpy(lp->d_typename, sd->name, 16);
1318 	strncpy(lp->d_packname, "fictitious", 16);
1319 	lp->d_secperunit = sd->params.disksize;
1320 	lp->d_rpm = sd->params.rot_rate;
1321 	lp->d_interleave = 1;
1322 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1323 	    D_REMOVABLE : 0;
1324 
1325 	lp->d_partitions[RAW_PART].p_offset = 0;
1326 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1327 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1328 	lp->d_npartitions = RAW_PART + 1;
1329 
1330 	lp->d_magic = DISKMAGIC;
1331 	lp->d_magic2 = DISKMAGIC;
1332 	lp->d_checksum = dkcksum(lp);
1333 }
1334 
1335 
1336 /*
1337  * Load the label information on the named device
1338  */
1339 static int
1340 sdgetdisklabel(struct sd_softc *sd)
1341 {
1342 	struct disklabel *lp = sd->sc_dk.dk_label;
1343 	const char *errstring;
1344 
1345 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1346 
1347 	sdgetdefaultlabel(sd, lp);
1348 
1349 	if (lp->d_secpercyl == 0) {
1350 		lp->d_secpercyl = 100;
1351 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1352 	}
1353 
1354 	/*
1355 	 * Call the generic disklabel extraction routine
1356 	 */
1357 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1358 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1359 	if (errstring) {
1360 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1361 		return EIO;
1362 	}
1363 	return 0;
1364 }
1365 
1366 static bool
1367 sd_shutdown(device_t self, int how)
1368 {
1369 	struct sd_softc *sd = device_private(self);
1370 
1371 	/*
1372 	 * If the disk cache needs to be flushed, and the disk supports
1373 	 * it, flush it.  We're cold at this point, so we poll for
1374 	 * completion.
1375 	 */
1376 	if ((sd->flags & SDF_DIRTY) != 0) {
1377 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1378 			aprint_error_dev(sd->sc_dev,
1379 				"cache synchronization failed\n");
1380 			sd->flags &= ~SDF_FLUSHING;
1381 		} else
1382 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1383 	}
1384 	return true;
1385 }
1386 
1387 static bool
1388 sd_suspend(device_t dv, const pmf_qual_t *qual)
1389 {
1390 	return sd_shutdown(dv, boothowto); /* XXX no need to poll */
1391 }
1392 
1393 /*
1394  * Check Errors
1395  */
1396 static int
1397 sd_interpret_sense(struct scsipi_xfer *xs)
1398 {
1399 	struct scsipi_periph *periph = xs->xs_periph;
1400 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1401 	struct sd_softc *sd = device_private(periph->periph_dev);
1402 	int s, error, retval = EJUSTRETURN;
1403 
1404 	/*
1405 	 * If the periph is already recovering, just do the normal
1406 	 * error processing.
1407 	 */
1408 	if (periph->periph_flags & PERIPH_RECOVERING)
1409 		return (retval);
1410 
1411 	/*
1412 	 * Ignore errors from accessing illegal fields (e.g. trying to
1413 	 * lock the door of a digicam, which doesn't have a door that
1414 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1415 	 */
1416 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1417 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1418 	    sense->asc == 0x24 &&
1419 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1420 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1421 			scsipi_printaddr(periph);
1422 			printf("no door lock\n");
1423 		}
1424 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1425 		return (retval);
1426 	}
1427 
1428 
1429 
1430 	/*
1431 	 * If the device is not open yet, let the generic code handle it.
1432 	 */
1433 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1434 		return (retval);
1435 
1436 	/*
1437 	 * If it isn't a extended or extended/deferred error, let
1438 	 * the generic code handle it.
1439 	 */
1440 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1441 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1442 		return (retval);
1443 
1444 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1445 	    sense->asc == 0x4) {
1446 		if (sense->ascq == 0x01)	{
1447 			/*
1448 			 * Unit In The Process Of Becoming Ready.
1449 			 */
1450 			printf("%s: waiting for pack to spin up...\n",
1451 			    device_xname(sd->sc_dev));
1452 			if (!callout_pending(&periph->periph_callout))
1453 				scsipi_periph_freeze(periph, 1);
1454 			callout_reset(&periph->periph_callout,
1455 			    5 * hz, scsipi_periph_timed_thaw, periph);
1456 			retval = ERESTART;
1457 		} else if (sense->ascq == 0x02) {
1458 			printf("%s: pack is stopped, restarting...\n",
1459 			    device_xname(sd->sc_dev));
1460 			s = splbio();
1461 			periph->periph_flags |= PERIPH_RECOVERING;
1462 			splx(s);
1463 			error = scsipi_start(periph, SSS_START,
1464 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1465 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1466 			if (error) {
1467 				aprint_error_dev(sd->sc_dev,
1468 					"unable to restart pack\n");
1469 				retval = error;
1470 			} else
1471 				retval = ERESTART;
1472 			s = splbio();
1473 			periph->periph_flags &= ~PERIPH_RECOVERING;
1474 			splx(s);
1475 		}
1476 	}
1477 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1478 	    sense->asc == 0x31 &&
1479 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1480 		/* Medium Format Corrupted */
1481 		retval = EFTYPE;
1482 	}
1483 	return (retval);
1484 }
1485 
1486 
1487 static int
1488 sdsize(dev_t dev)
1489 {
1490 	struct sd_softc *sd;
1491 	int part, unit, omask;
1492 	int size;
1493 
1494 	unit = SDUNIT(dev);
1495 	sd = device_lookup_private(&sd_cd, unit);
1496 	if (sd == NULL)
1497 		return (-1);
1498 
1499 	if (!device_is_active(sd->sc_dev))
1500 		return (-1);
1501 
1502 	part = SDPART(dev);
1503 	omask = sd->sc_dk.dk_openmask & (1 << part);
1504 
1505 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1506 		return (-1);
1507 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1508 		size = -1;
1509 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1510 		size = -1;
1511 	else
1512 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1513 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1514 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1515 		return (-1);
1516 	return (size);
1517 }
1518 
1519 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1520 static struct scsipi_xfer sx;
1521 static int sddoingadump;
1522 
1523 /*
1524  * dump all of physical memory into the partition specified, starting
1525  * at offset 'dumplo' into the partition.
1526  */
1527 static int
1528 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1529 {
1530 	struct sd_softc *sd;	/* disk unit to do the I/O */
1531 	struct disklabel *lp;	/* disk's disklabel */
1532 	int	unit, part;
1533 	int	sectorsize;	/* size of a disk sector */
1534 	int	nsects;		/* number of sectors in partition */
1535 	int	sectoff;	/* sector offset of partition */
1536 	int	totwrt;		/* total number of sectors left to write */
1537 	int	nwrt;		/* current number of sectors to write */
1538 	struct scsipi_rw_10 cmd;	/* write command */
1539 	struct scsipi_xfer *xs;	/* ... convenience */
1540 	struct scsipi_periph *periph;
1541 	struct scsipi_channel *chan;
1542 
1543 	/* Check if recursive dump; if so, punt. */
1544 	if (sddoingadump)
1545 		return (EFAULT);
1546 
1547 	/* Mark as active early. */
1548 	sddoingadump = 1;
1549 
1550 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1551 	part = SDPART(dev);
1552 
1553 	/* Check for acceptable drive number. */
1554 	sd = device_lookup_private(&sd_cd, unit);
1555 	if (sd == NULL)
1556 		return (ENXIO);
1557 
1558 	if (!device_is_active(sd->sc_dev))
1559 		return (ENODEV);
1560 
1561 	periph = sd->sc_periph;
1562 	chan = periph->periph_channel;
1563 
1564 	/* Make sure it was initialized. */
1565 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1566 		return (ENXIO);
1567 
1568 	/* Convert to disk sectors.  Request must be a multiple of size. */
1569 	lp = sd->sc_dk.dk_label;
1570 	sectorsize = lp->d_secsize;
1571 	if ((size % sectorsize) != 0)
1572 		return (EFAULT);
1573 	totwrt = size / sectorsize;
1574 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1575 
1576 	nsects = lp->d_partitions[part].p_size;
1577 	sectoff = lp->d_partitions[part].p_offset;
1578 
1579 	/* Check transfer bounds against partition size. */
1580 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1581 		return (EINVAL);
1582 
1583 	/* Offset block number to start of partition. */
1584 	blkno += sectoff;
1585 
1586 	xs = &sx;
1587 
1588 	while (totwrt > 0) {
1589 		nwrt = totwrt;		/* XXX */
1590 #ifndef	SD_DUMP_NOT_TRUSTED
1591 		/*
1592 		 *  Fill out the scsi command
1593 		 */
1594 		memset(&cmd, 0, sizeof(cmd));
1595 		cmd.opcode = WRITE_10;
1596 		_lto4b(blkno, cmd.addr);
1597 		_lto2b(nwrt, cmd.length);
1598 		/*
1599 		 * Fill out the scsipi_xfer structure
1600 		 *    Note: we cannot sleep as we may be an interrupt
1601 		 * don't use scsipi_command() as it may want to wait
1602 		 * for an xs.
1603 		 */
1604 		memset(xs, 0, sizeof(sx));
1605 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1606 		    XS_CTL_DATA_OUT;
1607 		xs->xs_status = 0;
1608 		xs->xs_periph = periph;
1609 		xs->xs_retries = SDRETRIES;
1610 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1611 		xs->cmd = (struct scsipi_generic *)&cmd;
1612 		xs->cmdlen = sizeof(cmd);
1613 		xs->resid = nwrt * sectorsize;
1614 		xs->error = XS_NOERROR;
1615 		xs->bp = 0;
1616 		xs->data = va;
1617 		xs->datalen = nwrt * sectorsize;
1618 		callout_init(&xs->xs_callout, 0);
1619 
1620 		/*
1621 		 * Pass all this info to the scsi driver.
1622 		 */
1623 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1624 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1625 		    xs->error != XS_NOERROR)
1626 			return (EIO);
1627 #else	/* SD_DUMP_NOT_TRUSTED */
1628 		/* Let's just talk about this first... */
1629 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1630 		delay(500 * 1000);	/* half a second */
1631 #endif	/* SD_DUMP_NOT_TRUSTED */
1632 
1633 		/* update block count */
1634 		totwrt -= nwrt;
1635 		blkno += nwrt;
1636 		va = (char *)va + sectorsize * nwrt;
1637 	}
1638 	sddoingadump = 0;
1639 	return (0);
1640 }
1641 
1642 static int
1643 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1644     int page, int flags, int *big)
1645 {
1646 
1647 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1648 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1649 		*big = 1;
1650 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1651 		    size + sizeof(struct scsi_mode_parameter_header_10),
1652 		    flags, SDRETRIES, 6000);
1653 	} else {
1654 		*big = 0;
1655 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1656 		    size + sizeof(struct scsi_mode_parameter_header_6),
1657 		    flags, SDRETRIES, 6000);
1658 	}
1659 }
1660 
1661 static int
1662 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1663     int flags, int big)
1664 {
1665 
1666 	if (big) {
1667 		struct scsi_mode_parameter_header_10 *header = sense;
1668 
1669 		_lto2b(0, header->data_length);
1670 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1671 		    size + sizeof(struct scsi_mode_parameter_header_10),
1672 		    flags, SDRETRIES, 6000);
1673 	} else {
1674 		struct scsi_mode_parameter_header_6 *header = sense;
1675 
1676 		header->data_length = 0;
1677 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1678 		    size + sizeof(struct scsi_mode_parameter_header_6),
1679 		    flags, SDRETRIES, 6000);
1680 	}
1681 }
1682 
1683 /*
1684  * sd_validate_blksize:
1685  *
1686  *	Validate the block size.  Print error if periph is specified,
1687  */
1688 static int
1689 sd_validate_blksize(struct scsipi_periph *periph, int len)
1690 {
1691 
1692 	switch (len) {
1693 	case 256:
1694 	case 512:
1695 	case 1024:
1696 	case 2048:
1697 	case 4096:
1698 		return 1;
1699 	}
1700 
1701 	if (periph) {
1702 		scsipi_printaddr(periph);
1703 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1704 		    (len ^ (1 << (ffs(len) - 1))) ?
1705 		    "preposterous" : "unsupported",
1706 		    len, SD_DEFAULT_BLKSIZE);
1707 	}
1708 
1709 	return 0;
1710 }
1711 
1712 /*
1713  * sd_read_capacity:
1714  *
1715  *	Find out from the device what its capacity is.
1716  */
1717 static u_int64_t
1718 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1719 {
1720 	union {
1721 		struct scsipi_read_capacity_10 cmd;
1722 		struct scsipi_read_capacity_16 cmd16;
1723 	} cmd;
1724 	union {
1725 		struct scsipi_read_capacity_10_data data;
1726 		struct scsipi_read_capacity_16_data data16;
1727 	} *datap;
1728 	uint64_t rv;
1729 
1730 	memset(&cmd, 0, sizeof(cmd));
1731 	cmd.cmd.opcode = READ_CAPACITY_10;
1732 
1733 	/*
1734 	 * Don't allocate data buffer on stack;
1735 	 * The lower driver layer might use the same stack and
1736 	 * if it uses region which is in the same cacheline,
1737 	 * cache flush ops against the data buffer won't work properly.
1738 	 */
1739 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1740 	if (datap == NULL)
1741 		return 0;
1742 
1743 	/*
1744 	 * If the command works, interpret the result as a 4 byte
1745 	 * number of blocks
1746 	 */
1747 	rv = 0;
1748 	memset(datap, 0, sizeof(datap->data));
1749 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1750 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1751 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1752 		goto out;
1753 
1754 	if (_4btol(datap->data.addr) != 0xffffffff) {
1755 		*blksize = _4btol(datap->data.length);
1756 		rv = _4btol(datap->data.addr) + 1;
1757 		goto out;
1758 	}
1759 
1760 	/*
1761 	 * Device is larger than can be reflected by READ CAPACITY (10).
1762 	 * Try READ CAPACITY (16).
1763 	 */
1764 
1765 	memset(&cmd, 0, sizeof(cmd));
1766 	cmd.cmd16.opcode = READ_CAPACITY_16;
1767 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1768 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1769 
1770 	memset(datap, 0, sizeof(datap->data16));
1771 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1772 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1773 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1774 		goto out;
1775 
1776 	*blksize = _4btol(datap->data16.length);
1777 	rv = _8btol(datap->data16.addr) + 1;
1778 
1779  out:
1780 	free(datap, M_TEMP);
1781 	return rv;
1782 }
1783 
1784 static int
1785 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1786 {
1787 	struct {
1788 		struct scsi_mode_parameter_header_6 header;
1789 		/* no block descriptor */
1790 		u_int8_t pg_code; /* page code (should be 6) */
1791 		u_int8_t pg_length; /* page length (should be 11) */
1792 		u_int8_t wcd; /* bit0: cache disable */
1793 		u_int8_t lbs[2]; /* logical block size */
1794 		u_int8_t size[5]; /* number of log. blocks */
1795 		u_int8_t pp; /* power/performance */
1796 		u_int8_t flags;
1797 		u_int8_t resvd;
1798 	} scsipi_sense;
1799 	u_int64_t blocks;
1800 	int error, blksize;
1801 
1802 	/*
1803 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1804 	 * give the same information. Do both for now, and check
1805 	 * for consistency.
1806 	 * XXX probably differs for removable media
1807 	 */
1808 	dp->blksize = SD_DEFAULT_BLKSIZE;
1809 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1810 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1811 
1812 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1813 	    &scsipi_sense.header, sizeof(scsipi_sense),
1814 	    flags, SDRETRIES, 6000);
1815 
1816 	if (error != 0)
1817 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1818 
1819 	dp->blksize = blksize;
1820 	if (!sd_validate_blksize(NULL, dp->blksize))
1821 		dp->blksize = _2btol(scsipi_sense.lbs);
1822 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1823 		dp->blksize = SD_DEFAULT_BLKSIZE;
1824 
1825 	/*
1826 	 * Create a pseudo-geometry.
1827 	 */
1828 	dp->heads = 64;
1829 	dp->sectors = 32;
1830 	dp->cyls = blocks / (dp->heads * dp->sectors);
1831 	dp->disksize = _5btol(scsipi_sense.size);
1832 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1833 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1834 		       (unsigned long long)dp->disksize,
1835 		       (unsigned long long)blocks);
1836 		dp->disksize = blocks;
1837 	}
1838 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1839 
1840 	return (SDGP_RESULT_OK);
1841 }
1842 
1843 /*
1844  * Get the scsi driver to send a full inquiry to the * device and use the
1845  * results to fill out the disk parameter structure.
1846  */
1847 static int
1848 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1849 {
1850 	u_int64_t blocks;
1851 	int error, blksize;
1852 #if 0
1853 	int i;
1854 	u_int8_t *p;
1855 #endif
1856 
1857 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1858 	    flags);
1859 	if (blocks == 0) {
1860 		struct scsipi_read_format_capacities cmd;
1861 		struct {
1862 			struct scsipi_capacity_list_header header;
1863 			struct scsipi_capacity_descriptor desc;
1864 		} __packed data;
1865 
1866 		memset(&cmd, 0, sizeof(cmd));
1867 		memset(&data, 0, sizeof(data));
1868 		cmd.opcode = READ_FORMAT_CAPACITIES;
1869 		_lto2b(sizeof(data), cmd.length);
1870 
1871 		error = scsipi_command(sd->sc_periph,
1872 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1873 		    SDRETRIES, 20000, NULL,
1874 		    flags | XS_CTL_DATA_IN);
1875 		if (error == EFTYPE) {
1876 			/* Medium Format Corrupted, handle as not formatted */
1877 			return (SDGP_RESULT_UNFORMATTED);
1878 		}
1879 		if (error || data.header.length == 0)
1880 			return (SDGP_RESULT_OFFLINE);
1881 
1882 #if 0
1883 printf("rfc: length=%d\n", data.header.length);
1884 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");
1885 #endif
1886 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1887 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1888 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1889 			break;
1890 
1891 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1892 			return (SDGP_RESULT_UNFORMATTED);
1893 
1894 		case SCSIPI_CAP_DESC_CODE_NONE:
1895 			return (SDGP_RESULT_OFFLINE);
1896 		}
1897 
1898 		dp->disksize = blocks = _4btol(data.desc.nblks);
1899 		if (blocks == 0)
1900 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1901 
1902 		blksize = _3btol(data.desc.blklen);
1903 
1904 	} else if (!sd_validate_blksize(NULL, blksize)) {
1905 		struct sd_mode_sense_data scsipi_sense;
1906 		int big, bsize;
1907 		struct scsi_general_block_descriptor *bdesc;
1908 
1909 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1910 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1911 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1912 		if (!error) {
1913 			if (big) {
1914 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1915 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1916 			} else {
1917 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1918 				bsize = scsipi_sense.header.small.blk_desc_len;
1919 			}
1920 
1921 #if 0
1922 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1923 printf("page 0 bsize=%d\n", bsize);
1924 printf("page 0 ok\n");
1925 #endif
1926 
1927 			if (bsize >= 8) {
1928 				blksize = _3btol(bdesc->blklen);
1929 			}
1930 		}
1931 	}
1932 
1933 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1934 		blksize = SD_DEFAULT_BLKSIZE;
1935 
1936 	dp->blksize = blksize;
1937 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1938 	return (0);
1939 }
1940 
1941 static int
1942 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1943 {
1944 	struct sd_mode_sense_data scsipi_sense;
1945 	int error;
1946 	int big, byte2;
1947 	size_t poffset;
1948 	union scsi_disk_pages *pages;
1949 
1950 	byte2 = SMS_DBD;
1951 again:
1952 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1953 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1954 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1955 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1956 	    flags | XS_CTL_SILENT, &big);
1957 	if (error) {
1958 		if (byte2 == SMS_DBD) {
1959 			/* No result; try once more with DBD off */
1960 			byte2 = 0;
1961 			goto again;
1962 		}
1963 		return (error);
1964 	}
1965 
1966 	if (big) {
1967 		poffset = sizeof scsipi_sense.header.big;
1968 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1969 	} else {
1970 		poffset = sizeof scsipi_sense.header.small;
1971 		poffset += scsipi_sense.header.small.blk_desc_len;
1972 	}
1973 
1974 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1975 		return ERESTART;
1976 
1977 	pages = (void *)((u_long)&scsipi_sense + poffset);
1978 #if 0
1979 	{
1980 		size_t i;
1981 		u_int8_t *p;
1982 
1983 		printf("page 4 sense:");
1984 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
1985 		    i--, p++)
1986 			printf(" %02x", *p);
1987 		printf("\n");
1988 		printf("page 4 pg_code=%d sense=%p/%p\n",
1989 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1990 	}
1991 #endif
1992 
1993 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1994 		return (ERESTART);
1995 
1996 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1997 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1998 	    _3btol(pages->rigid_geometry.ncyl),
1999 	    pages->rigid_geometry.nheads,
2000 	    _2btol(pages->rigid_geometry.st_cyl_wp),
2001 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
2002 	    _2btol(pages->rigid_geometry.land_zone)));
2003 
2004 	/*
2005 	 * KLUDGE!! (for zone recorded disks)
2006 	 * give a number of sectors so that sec * trks * cyls
2007 	 * is <= disk_size
2008 	 * can lead to wasted space! THINK ABOUT THIS !
2009 	 */
2010 	dp->heads = pages->rigid_geometry.nheads;
2011 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
2012 	if (dp->heads == 0 || dp->cyls == 0)
2013 		return (ERESTART);
2014 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
2015 
2016 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2017 	if (dp->rot_rate == 0)
2018 		dp->rot_rate = 3600;
2019 
2020 #if 0
2021 printf("page 4 ok\n");
2022 #endif
2023 	return (0);
2024 }
2025 
2026 static int
2027 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
2028 {
2029 	struct sd_mode_sense_data scsipi_sense;
2030 	int error;
2031 	int big, byte2;
2032 	size_t poffset;
2033 	union scsi_disk_pages *pages;
2034 
2035 	byte2 = SMS_DBD;
2036 again:
2037 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2038 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2039 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2040 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2041 	    flags | XS_CTL_SILENT, &big);
2042 	if (error) {
2043 		if (byte2 == SMS_DBD) {
2044 			/* No result; try once more with DBD off */
2045 			byte2 = 0;
2046 			goto again;
2047 		}
2048 		return (error);
2049 	}
2050 
2051 	if (big) {
2052 		poffset = sizeof scsipi_sense.header.big;
2053 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2054 	} else {
2055 		poffset = sizeof scsipi_sense.header.small;
2056 		poffset += scsipi_sense.header.small.blk_desc_len;
2057 	}
2058 
2059 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2060 		return ERESTART;
2061 
2062 	pages = (void *)((u_long)&scsipi_sense + poffset);
2063 #if 0
2064 	{
2065 		size_t i;
2066 		u_int8_t *p;
2067 
2068 		printf("page 5 sense:");
2069 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2070 		    i--, p++)
2071 			printf(" %02x", *p);
2072 		printf("\n");
2073 		printf("page 5 pg_code=%d sense=%p/%p\n",
2074 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2075 	}
2076 #endif
2077 
2078 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2079 		return (ERESTART);
2080 
2081 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2082 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2083 	    _3btol(pages->flex_geometry.ncyl),
2084 	    pages->flex_geometry.nheads,
2085 	    pages->flex_geometry.ph_sec_tr,
2086 	    _2btol(pages->flex_geometry.bytes_s)));
2087 
2088 	dp->heads = pages->flex_geometry.nheads;
2089 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2090 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2091 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2092 		return (ERESTART);
2093 
2094 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2095 	if (dp->rot_rate == 0)
2096 		dp->rot_rate = 3600;
2097 
2098 #if 0
2099 printf("page 5 ok\n");
2100 #endif
2101 	return (0);
2102 }
2103 
2104 static int
2105 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2106 {
2107 	int error;
2108 
2109 	/*
2110 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2111 	 * cleared by the caller if necessary.
2112 	 */
2113 	if (sd->type == T_SIMPLE_DIRECT) {
2114 		error = sd_get_simplifiedparms(sd, dp, flags);
2115 		if (!error)
2116 			disk_blocksize(&sd->sc_dk, dp->blksize);
2117 		return (error);
2118 	}
2119 
2120 	error = sd_get_capacity(sd, dp, flags);
2121 	if (error)
2122 		return (error);
2123 
2124 	disk_blocksize(&sd->sc_dk, dp->blksize);
2125 
2126 	if (sd->type == T_OPTICAL)
2127 		goto page0;
2128 
2129 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2130 		if (!sd_get_parms_page5(sd, dp, flags) ||
2131 		    !sd_get_parms_page4(sd, dp, flags))
2132 			goto setprops;
2133 	} else {
2134 		if (!sd_get_parms_page4(sd, dp, flags) ||
2135 		    !sd_get_parms_page5(sd, dp, flags))
2136 			goto setprops;
2137 	}
2138 
2139 page0:
2140 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2141 	/* Try calling driver's method for figuring out geometry. */
2142 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2143 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2144 		(sd->sc_periph, dp, dp->disksize)) {
2145 		/*
2146 		 * Use adaptec standard fictitious geometry
2147 		 * this depends on which controller (e.g. 1542C is
2148 		 * different. but we have to put SOMETHING here..)
2149 		 */
2150 		dp->heads = 64;
2151 		dp->sectors = 32;
2152 		dp->cyls = dp->disksize / (64 * 32);
2153 	}
2154 	dp->rot_rate = 3600;
2155 
2156 setprops:
2157 	sd_set_properties(sd);
2158 
2159 	return (SDGP_RESULT_OK);
2160 }
2161 
2162 static int
2163 sd_flush(struct sd_softc *sd, int flags)
2164 {
2165 	struct scsipi_periph *periph = sd->sc_periph;
2166 	struct scsi_synchronize_cache_10 cmd;
2167 
2168 	/*
2169 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2170 	 * We issue with address 0 length 0, which should be
2171 	 * interpreted by the device as "all remaining blocks
2172 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2173 	 * in the event that the command is not supported by
2174 	 * the device, and poll for completion so that we know
2175 	 * that the cache has actually been flushed.
2176 	 *
2177 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2178 	 * command, as indicated by our quirks flags.
2179 	 *
2180 	 * XXX What about older devices?
2181 	 */
2182 	if (periph->periph_version < 2 ||
2183 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2184 		return (0);
2185 
2186 	sd->flags |= SDF_FLUSHING;
2187 	memset(&cmd, 0, sizeof(cmd));
2188 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2189 
2190 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2191 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2192 }
2193 
2194 static int
2195 sd_getcache(struct sd_softc *sd, int *bitsp)
2196 {
2197 	struct scsipi_periph *periph = sd->sc_periph;
2198 	struct sd_mode_sense_data scsipi_sense;
2199 	int error, bits = 0;
2200 	int big;
2201 	union scsi_disk_pages *pages;
2202 
2203 	if (periph->periph_version < 2)
2204 		return (EOPNOTSUPP);
2205 
2206 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2207 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2208 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2209 	if (error)
2210 		return (error);
2211 
2212 	if (big)
2213 		pages = (void *)(&scsipi_sense.header.big + 1);
2214 	else
2215 		pages = (void *)(&scsipi_sense.header.small + 1);
2216 
2217 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2218 		bits |= DKCACHE_READ;
2219 	if (pages->caching_params.flags & CACHING_WCE)
2220 		bits |= DKCACHE_WRITE;
2221 	if (pages->caching_params.pg_code & PGCODE_PS)
2222 		bits |= DKCACHE_SAVE;
2223 
2224 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2225 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2226 	    sizeof(scsipi_sense.pages.caching_params),
2227 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2228 	if (error == 0) {
2229 		if (big)
2230 			pages = (void *)(&scsipi_sense.header.big + 1);
2231 		else
2232 			pages = (void *)(&scsipi_sense.header.small + 1);
2233 
2234 		if (pages->caching_params.flags & CACHING_RCD)
2235 			bits |= DKCACHE_RCHANGE;
2236 		if (pages->caching_params.flags & CACHING_WCE)
2237 			bits |= DKCACHE_WCHANGE;
2238 	}
2239 
2240 	*bitsp = bits;
2241 
2242 	return (0);
2243 }
2244 
2245 static int
2246 sd_setcache(struct sd_softc *sd, int bits)
2247 {
2248 	struct scsipi_periph *periph = sd->sc_periph;
2249 	struct sd_mode_sense_data scsipi_sense;
2250 	int error;
2251 	uint8_t oflags, byte2 = 0;
2252 	int big;
2253 	union scsi_disk_pages *pages;
2254 
2255 	if (periph->periph_version < 2)
2256 		return (EOPNOTSUPP);
2257 
2258 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2259 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2260 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2261 	if (error)
2262 		return (error);
2263 
2264 	if (big)
2265 		pages = (void *)(&scsipi_sense.header.big + 1);
2266 	else
2267 		pages = (void *)(&scsipi_sense.header.small + 1);
2268 
2269 	oflags = pages->caching_params.flags;
2270 
2271 	if (bits & DKCACHE_READ)
2272 		pages->caching_params.flags &= ~CACHING_RCD;
2273 	else
2274 		pages->caching_params.flags |= CACHING_RCD;
2275 
2276 	if (bits & DKCACHE_WRITE)
2277 		pages->caching_params.flags |= CACHING_WCE;
2278 	else
2279 		pages->caching_params.flags &= ~CACHING_WCE;
2280 
2281 	if (oflags == pages->caching_params.flags)
2282 		return (0);
2283 
2284 	pages->caching_params.pg_code &= PGCODE_MASK;
2285 
2286 	if (bits & DKCACHE_SAVE)
2287 		byte2 |= SMS_SP;
2288 
2289 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2290 	    sizeof(struct scsi_mode_page_header) +
2291 	    pages->caching_params.pg_length, 0, big));
2292 }
2293 
2294 static void
2295 sd_set_properties(struct sd_softc *sd)
2296 {
2297 	prop_dictionary_t disk_info, odisk_info, geom;
2298 
2299 	disk_info = prop_dictionary_create();
2300 
2301 	geom = prop_dictionary_create();
2302 
2303 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2304 	    sd->params.disksize);
2305 
2306 	prop_dictionary_set_uint32(geom, "sector-size",
2307 	    sd->params.blksize);
2308 
2309 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2310 	    sd->params.sectors);
2311 
2312 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2313 	    sd->params.heads);
2314 
2315 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2316 	    sd->params.cyls);
2317 
2318 	prop_dictionary_set(disk_info, "geometry", geom);
2319 	prop_object_release(geom);
2320 
2321 	prop_dictionary_set(device_properties(sd->sc_dev),
2322 	    "disk-info", disk_info);
2323 
2324 	/*
2325 	 * Don't release disk_info here; we keep a reference to it.
2326 	 * disk_detach() will release it when we go away.
2327 	 */
2328 
2329 	odisk_info = sd->sc_dk.dk_info;
2330 	sd->sc_dk.dk_info = disk_info;
2331 	if (odisk_info)
2332 		prop_object_release(odisk_info);
2333 }
2334