xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 466a16a118933bd295a8a104f095714fadf9cf68)
1 /*	$NetBSD: sd.c,v 1.275 2008/07/16 18:54:09 drochner 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 1.275 2008/07/16 18:54:09 drochner Exp $");
51 
52 #include "opt_scsi.h"
53 #include "rnd.h"
54 
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/kernel.h>
58 #include <sys/file.h>
59 #include <sys/stat.h>
60 #include <sys/ioctl.h>
61 #include <sys/scsiio.h>
62 #include <sys/buf.h>
63 #include <sys/bufq.h>
64 #include <sys/uio.h>
65 #include <sys/malloc.h>
66 #include <sys/errno.h>
67 #include <sys/device.h>
68 #include <sys/disklabel.h>
69 #include <sys/disk.h>
70 #include <sys/proc.h>
71 #include <sys/conf.h>
72 #include <sys/vnode.h>
73 #if NRND > 0
74 #include <sys/rnd.h>
75 #endif
76 
77 #include <dev/scsipi/scsi_spc.h>
78 #include <dev/scsipi/scsipi_all.h>
79 #include <dev/scsipi/scsi_all.h>
80 #include <dev/scsipi/scsipi_disk.h>
81 #include <dev/scsipi/scsi_disk.h>
82 #include <dev/scsipi/scsiconf.h>
83 #include <dev/scsipi/scsipi_base.h>
84 #include <dev/scsipi/sdvar.h>
85 
86 #include <prop/proplib.h>
87 
88 #define	SDUNIT(dev)			DISKUNIT(dev)
89 #define	SDPART(dev)			DISKPART(dev)
90 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
91 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
92 
93 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
94 
95 #define	SD_DEFAULT_BLKSIZE	512
96 
97 static void	sdminphys(struct buf *);
98 static void	sdgetdefaultlabel(struct sd_softc *, struct disklabel *);
99 static int	sdgetdisklabel(struct sd_softc *);
100 static void	sdstart(struct scsipi_periph *);
101 static void	sdrestart(void *);
102 static void	sddone(struct scsipi_xfer *, int);
103 static bool	sd_suspend(device_t PMF_FN_PROTO);
104 static void	sd_shutdown(void *);
105 static int	sd_interpret_sense(struct scsipi_xfer *);
106 
107 static int	sd_mode_sense(struct sd_softc *, u_int8_t, void *, size_t, int,
108 		    int, int *);
109 static int	sd_mode_select(struct sd_softc *, u_int8_t, void *, size_t, int,
110 		    int);
111 static int	sd_validate_blksize(struct scsipi_periph *, int);
112 static u_int64_t sd_read_capacity(struct scsipi_periph *, int *, int flags);
113 static int	sd_get_simplifiedparms(struct sd_softc *, struct disk_parms *,
114 		    int);
115 static int	sd_get_capacity(struct sd_softc *, struct disk_parms *, int);
116 static int	sd_get_parms(struct sd_softc *, struct disk_parms *, int);
117 static int	sd_get_parms_page4(struct sd_softc *, struct disk_parms *,
118 		    int);
119 static int	sd_get_parms_page5(struct sd_softc *, struct disk_parms *,
120 		    int);
121 
122 static int	sd_flush(struct sd_softc *, int);
123 static int	sd_getcache(struct sd_softc *, int *);
124 static int	sd_setcache(struct sd_softc *, int);
125 
126 static int	sdmatch(struct device *, struct cfdata *, void *);
127 static void	sdattach(struct device *, struct device *, void *);
128 static int	sdactivate(struct device *, enum devact);
129 static int	sddetach(struct device *, int);
130 static void	sd_set_properties(struct sd_softc *);
131 
132 CFATTACH_DECL_NEW(sd, sizeof(struct sd_softc), sdmatch, sdattach, sddetach,
133     sdactivate);
134 
135 extern struct cfdriver sd_cd;
136 
137 static const struct scsipi_inquiry_pattern sd_patterns[] = {
138 	{T_DIRECT, T_FIXED,
139 	 "",         "",                 ""},
140 	{T_DIRECT, T_REMOV,
141 	 "",         "",                 ""},
142 	{T_OPTICAL, T_FIXED,
143 	 "",         "",                 ""},
144 	{T_OPTICAL, T_REMOV,
145 	 "",         "",                 ""},
146 	{T_SIMPLE_DIRECT, T_FIXED,
147 	 "",         "",                 ""},
148 	{T_SIMPLE_DIRECT, T_REMOV,
149 	 "",         "",                 ""},
150 };
151 
152 static dev_type_open(sdopen);
153 static dev_type_close(sdclose);
154 static dev_type_read(sdread);
155 static dev_type_write(sdwrite);
156 static dev_type_ioctl(sdioctl);
157 static dev_type_strategy(sdstrategy);
158 static dev_type_dump(sddump);
159 static dev_type_size(sdsize);
160 
161 const struct bdevsw sd_bdevsw = {
162 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
163 };
164 
165 const struct cdevsw sd_cdevsw = {
166 	sdopen, sdclose, sdread, sdwrite, sdioctl,
167 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
168 };
169 
170 static struct dkdriver sddkdriver = { sdstrategy, sdminphys };
171 
172 static const struct scsipi_periphsw sd_switch = {
173 	sd_interpret_sense,	/* check our error handler first */
174 	sdstart,		/* have a queue, served by this */
175 	NULL,			/* have no async handler */
176 	sddone,			/* deal with stats at interrupt time */
177 };
178 
179 struct sd_mode_sense_data {
180 	/*
181 	 * XXX
182 	 * We are not going to parse this as-is -- it just has to be large
183 	 * enough.
184 	 */
185 	union {
186 		struct scsi_mode_parameter_header_6 small;
187 		struct scsi_mode_parameter_header_10 big;
188 	} header;
189 	struct scsi_general_block_descriptor blk_desc;
190 	union scsi_disk_pages pages;
191 };
192 
193 /*
194  * The routine called by the low level scsi routine when it discovers
195  * A device suitable for this driver
196  */
197 static int
198 sdmatch(struct device *parent, struct cfdata *match,
199     void *aux)
200 {
201 	struct scsipibus_attach_args *sa = aux;
202 	int priority;
203 
204 	(void)scsipi_inqmatch(&sa->sa_inqbuf,
205 	    sd_patterns, sizeof(sd_patterns) / sizeof(sd_patterns[0]),
206 	    sizeof(sd_patterns[0]), &priority);
207 
208 	return (priority);
209 }
210 
211 /*
212  * Attach routine common to atapi & scsi.
213  */
214 static void
215 sdattach(struct device *parent, struct device *self, void *aux)
216 {
217 	struct sd_softc *sd = device_private(self);
218 	struct scsipibus_attach_args *sa = aux;
219 	struct scsipi_periph *periph = sa->sa_periph;
220 	int error, result;
221 	struct disk_parms *dp = &sd->params;
222 	char pbuf[9];
223 
224 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
225 
226 	sd->sc_dev = self;
227 	sd->type = (sa->sa_inqbuf.type & SID_TYPE);
228 	strncpy(sd->name, sa->sa_inqbuf.product, sizeof(sd->name));
229 	if (sd->type == T_SIMPLE_DIRECT)
230 		periph->periph_quirks |= PQUIRK_ONLYBIG | PQUIRK_NOBIGMODESENSE;
231 
232 	if (scsipi_periph_bustype(sa->sa_periph) == SCSIPI_BUSTYPE_SCSI &&
233 	    periph->periph_version == 0)
234 		sd->flags |= SDF_ANCIENT;
235 
236 	bufq_alloc(&sd->buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
237 
238 	callout_init(&sd->sc_callout, 0);
239 
240 	/*
241 	 * Store information needed to contact our base driver
242 	 */
243 	sd->sc_periph = periph;
244 
245 	periph->periph_dev = sd->sc_dev;
246 	periph->periph_switch = &sd_switch;
247 
248         /*
249          * Increase our openings to the maximum-per-periph
250          * supported by the adapter.  This will either be
251          * clamped down or grown by the adapter if necessary.
252          */
253 	periph->periph_openings =
254 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
255 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
256 
257 	/*
258 	 * Initialize and attach the disk structure.
259 	 */
260 	disk_init(&sd->sc_dk, device_xname(sd->sc_dev), &sddkdriver);
261 	disk_attach(&sd->sc_dk);
262 
263 	/*
264 	 * Use the subdriver to request information regarding the drive.
265 	 */
266 	aprint_naive("\n");
267 	aprint_normal("\n");
268 
269 	error = scsipi_test_unit_ready(periph,
270 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
271 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT_NODEV);
272 
273 	if (error)
274 		result = SDGP_RESULT_OFFLINE;
275 	else
276 		result = sd_get_parms(sd, &sd->params, XS_CTL_DISCOVERY);
277 	aprint_normal_dev(sd->sc_dev, "");
278 	switch (result) {
279 	case SDGP_RESULT_OK:
280 		format_bytes(pbuf, sizeof(pbuf),
281 		    (u_int64_t)dp->disksize * dp->blksize);
282 	        aprint_normal(
283 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
284 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
285 		    (unsigned long long)dp->disksize);
286 		break;
287 
288 	case SDGP_RESULT_OFFLINE:
289 		aprint_normal("drive offline");
290 		break;
291 
292 	case SDGP_RESULT_UNFORMATTED:
293 		aprint_normal("unformatted media");
294 		break;
295 
296 #ifdef DIAGNOSTIC
297 	default:
298 		panic("sdattach: unknown result from get_parms");
299 		break;
300 #endif
301 	}
302 	aprint_normal("\n");
303 
304 	/*
305 	 * Establish a shutdown hook so that we can ensure that
306 	 * our data has actually made it onto the platter at
307 	 * shutdown time.  Note that this relies on the fact
308 	 * that the shutdown hook code puts us at the head of
309 	 * the list (thus guaranteeing that our hook runs before
310 	 * our ancestors').
311 	 */
312 	if ((sd->sc_sdhook =
313 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
314 		aprint_error_dev(sd->sc_dev,
315 			"WARNING: unable to establish shutdown hook\n");
316 
317 	if (!pmf_device_register(self, sd_suspend, NULL))
318 		aprint_error_dev(self, "couldn't establish power handler\n");
319 
320 #if NRND > 0
321 	/*
322 	 * attach the device into the random source list
323 	 */
324 	rnd_attach_source(&sd->rnd_source, device_xname(sd->sc_dev),
325 			  RND_TYPE_DISK, 0);
326 #endif
327 
328 	/* Discover wedges on this disk. */
329 	dkwedge_discover(&sd->sc_dk);
330 
331 	sd_set_properties(sd);
332 }
333 
334 static int
335 sdactivate(struct device *self, enum devact act)
336 {
337 	int rv = 0;
338 
339 	switch (act) {
340 	case DVACT_ACTIVATE:
341 		rv = EOPNOTSUPP;
342 		break;
343 
344 	case DVACT_DEACTIVATE:
345 		/*
346 		 * Nothing to do; we key off the device's DVF_ACTIVE.
347 		 */
348 		break;
349 	}
350 	return (rv);
351 }
352 
353 static int
354 sddetach(struct device *self, int flags)
355 {
356 	struct sd_softc *sd = device_private(self);
357 	int s, bmaj, cmaj, i, mn;
358 
359 	/* locate the major number */
360 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
361 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
362 
363 	/* Nuke the vnodes for any open instances */
364 	for (i = 0; i < MAXPARTITIONS; i++) {
365 		mn = SDMINOR(device_unit(self), i);
366 		vdevgone(bmaj, mn, mn, VBLK);
367 		vdevgone(cmaj, mn, mn, VCHR);
368 	}
369 
370 	/* kill any pending restart */
371 	callout_stop(&sd->sc_callout);
372 
373 	/* Delete all of our wedges. */
374 	dkwedge_delall(&sd->sc_dk);
375 
376 	s = splbio();
377 
378 	/* Kill off any queued buffers. */
379 	bufq_drain(sd->buf_queue);
380 
381 	bufq_free(sd->buf_queue);
382 
383 	/* Kill off any pending commands. */
384 	scsipi_kill_pending(sd->sc_periph);
385 
386 	splx(s);
387 
388 	/* Detach from the disk list. */
389 	disk_detach(&sd->sc_dk);
390 	disk_destroy(&sd->sc_dk);
391 
392 	pmf_device_deregister(self);
393 	shutdownhook_disestablish(sd->sc_sdhook);
394 
395 #if NRND > 0
396 	/* Unhook the entropy source. */
397 	rnd_detach_source(&sd->rnd_source);
398 #endif
399 
400 	return (0);
401 }
402 
403 /*
404  * open the device. Make sure the partition info is a up-to-date as can be.
405  */
406 static int
407 sdopen(dev_t dev, int flag, int fmt, struct lwp *l)
408 {
409 	struct sd_softc *sd;
410 	struct scsipi_periph *periph;
411 	struct scsipi_adapter *adapt;
412 	int unit, part;
413 	int error;
414 
415 	unit = SDUNIT(dev);
416 	sd = device_lookup_private(&sd_cd, unit);
417 	if (sd == NULL)
418 		return (ENXIO);
419 
420 	if (!device_is_active(sd->sc_dev))
421 		return (ENODEV);
422 
423 	part = SDPART(dev);
424 
425 	mutex_enter(&sd->sc_dk.dk_openlock);
426 
427 	/*
428 	 * If there are wedges, and this is not RAW_PART, then we
429 	 * need to fail.
430 	 */
431 	if (sd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
432 		error = EBUSY;
433 		goto bad1;
434 	}
435 
436 	periph = sd->sc_periph;
437 	adapt = periph->periph_channel->chan_adapter;
438 
439 	SC_DEBUG(periph, SCSIPI_DB1,
440 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
441 	    sd_cd.cd_ndevs, part));
442 
443 	/*
444 	 * If this is the first open of this device, add a reference
445 	 * to the adapter.
446 	 */
447 	if (sd->sc_dk.dk_openmask == 0 &&
448 	    (error = scsipi_adapter_addref(adapt)) != 0)
449 		goto bad1;
450 
451 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
452 		/*
453 		 * If any partition is open, but the disk has been invalidated,
454 		 * disallow further opens of non-raw partition
455 		 */
456 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
457 		    (part != RAW_PART || fmt != S_IFCHR)) {
458 			error = EIO;
459 			goto bad2;
460 		}
461 	} else {
462 		int silent;
463 
464 		if (part == RAW_PART && fmt == S_IFCHR)
465 			silent = XS_CTL_SILENT;
466 		else
467 			silent = 0;
468 
469 		/* Check that it is still responding and ok. */
470 		error = scsipi_test_unit_ready(periph,
471 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
472 		    silent);
473 
474 		/*
475 		 * Start the pack spinning if necessary. Always allow the
476 		 * raw parition to be opened, for raw IOCTLs. Data transfers
477 		 * will check for SDEV_MEDIA_LOADED.
478 		 */
479 		if (error == EIO) {
480 			int error2;
481 
482 			error2 = scsipi_start(periph, SSS_START, silent);
483 			switch (error2) {
484 			case 0:
485 				error = 0;
486 				break;
487 			case EIO:
488 			case EINVAL:
489 				break;
490 			default:
491 				error = error2;
492 				break;
493 			}
494 		}
495 		if (error) {
496 			if (silent)
497 				goto out;
498 			goto bad2;
499 		}
500 
501 		periph->periph_flags |= PERIPH_OPEN;
502 
503 		if (periph->periph_flags & PERIPH_REMOVABLE) {
504 			/* Lock the pack in. */
505 			error = scsipi_prevent(periph, SPAMR_PREVENT_DT,
506 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
507 			    XS_CTL_IGNORE_MEDIA_CHANGE |
508 			    XS_CTL_SILENT);
509 			if (error)
510 				goto bad3;
511 		}
512 
513 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
514 			int param_error;
515 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
516 
517 			/*
518 			 * Load the physical device parameters.
519 			 *
520 			 * Note that if media is present but unformatted,
521 			 * we allow the open (so that it can be formatted!).
522 			 * The drive should refuse real I/O, if the media is
523 			 * unformatted.
524 			 */
525 			if ((param_error = sd_get_parms(sd, &sd->params, 0))
526 			     == SDGP_RESULT_OFFLINE) {
527 				error = ENXIO;
528 				periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
529 				goto bad3;
530 			}
531 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
532 
533 			/* Load the partition info if not already loaded. */
534 			if (param_error == 0) {
535 				if ((sdgetdisklabel(sd) != 0) && (part != RAW_PART)) {
536 					error = EIO;
537 					goto bad3;
538 				}
539 				SC_DEBUG(periph, SCSIPI_DB3,
540 				     ("Disklabel loaded "));
541 			}
542 		}
543 	}
544 
545 	/* Check that the partition exists. */
546 	if (part != RAW_PART &&
547 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
548 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
549 		error = ENXIO;
550 		goto bad3;
551 	}
552 
553  out:	/* Insure only one open at a time. */
554 	switch (fmt) {
555 	case S_IFCHR:
556 		sd->sc_dk.dk_copenmask |= (1 << part);
557 		break;
558 	case S_IFBLK:
559 		sd->sc_dk.dk_bopenmask |= (1 << part);
560 		break;
561 	}
562 	sd->sc_dk.dk_openmask =
563 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
564 
565 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
566 	mutex_exit(&sd->sc_dk.dk_openlock);
567 	return (0);
568 
569  bad3:
570 	if (sd->sc_dk.dk_openmask == 0) {
571 		if (periph->periph_flags & PERIPH_REMOVABLE)
572 			scsipi_prevent(periph, SPAMR_ALLOW,
573 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
574 			    XS_CTL_IGNORE_MEDIA_CHANGE |
575 			    XS_CTL_SILENT);
576 		periph->periph_flags &= ~PERIPH_OPEN;
577 	}
578 
579  bad2:
580 	if (sd->sc_dk.dk_openmask == 0)
581 		scsipi_adapter_delref(adapt);
582 
583  bad1:
584 	mutex_exit(&sd->sc_dk.dk_openlock);
585 	return (error);
586 }
587 
588 /*
589  * close the device.. only called if we are the LAST occurence of an open
590  * device.  Convenient now but usually a pain.
591  */
592 static int
593 sdclose(dev_t dev, int flag, int fmt, struct lwp *l)
594 {
595 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
596 	struct scsipi_periph *periph = sd->sc_periph;
597 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
598 	int part = SDPART(dev);
599 
600 	mutex_enter(&sd->sc_dk.dk_openlock);
601 	switch (fmt) {
602 	case S_IFCHR:
603 		sd->sc_dk.dk_copenmask &= ~(1 << part);
604 		break;
605 	case S_IFBLK:
606 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
607 		break;
608 	}
609 	sd->sc_dk.dk_openmask =
610 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
611 
612 	if (sd->sc_dk.dk_openmask == 0) {
613 		/*
614 		 * If the disk cache needs flushing, and the disk supports
615 		 * it, do it now.
616 		 */
617 		if ((sd->flags & SDF_DIRTY) != 0) {
618 			if (sd_flush(sd, 0)) {
619 				aprint_error_dev(sd->sc_dev,
620 					"cache synchronization failed\n");
621 				sd->flags &= ~SDF_FLUSHING;
622 			} else
623 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
624 		}
625 
626 		scsipi_wait_drain(periph);
627 
628 		if (periph->periph_flags & PERIPH_REMOVABLE)
629 			scsipi_prevent(periph, SPAMR_ALLOW,
630 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
631 			    XS_CTL_IGNORE_NOT_READY |
632 			    XS_CTL_SILENT);
633 		periph->periph_flags &= ~PERIPH_OPEN;
634 
635 		scsipi_wait_drain(periph);
636 
637 		scsipi_adapter_delref(adapt);
638 	}
639 
640 	mutex_exit(&sd->sc_dk.dk_openlock);
641 	return (0);
642 }
643 
644 /*
645  * Actually translate the requested transfer into one the physical driver
646  * can understand.  The transfer is described by a buf and will include
647  * only one physical transfer.
648  */
649 static void
650 sdstrategy(struct buf *bp)
651 {
652 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
653 	struct scsipi_periph *periph = sd->sc_periph;
654 	struct disklabel *lp;
655 	daddr_t blkno;
656 	int s;
657 	bool sector_aligned;
658 
659 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
660 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
661 	    ("%d bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
662 	/*
663 	 * If the device has been made invalid, error out
664 	 */
665 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
666 	    !device_is_active(sd->sc_dev)) {
667 		if (periph->periph_flags & PERIPH_OPEN)
668 			bp->b_error = EIO;
669 		else
670 			bp->b_error = ENODEV;
671 		goto done;
672 	}
673 
674 	lp = sd->sc_dk.dk_label;
675 
676 	/*
677 	 * The transfer must be a whole number of blocks, offset must not be
678 	 * negative.
679 	 */
680 	if (lp->d_secsize == DEV_BSIZE) {
681 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
682 	} else {
683 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
684 	}
685 	if (!sector_aligned || bp->b_blkno < 0) {
686 		bp->b_error = EINVAL;
687 		goto done;
688 	}
689 	/*
690 	 * If it's a null transfer, return immediatly
691 	 */
692 	if (bp->b_bcount == 0)
693 		goto done;
694 
695 	/*
696 	 * Do bounds checking, adjust transfer. if error, process.
697 	 * If end of partition, just return.
698 	 */
699 	if (SDPART(bp->b_dev) == RAW_PART) {
700 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
701 		    sd->params.disksize512) <= 0)
702 			goto done;
703 	} else {
704 		if (bounds_check_with_label(&sd->sc_dk, bp,
705 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
706 			goto done;
707 	}
708 
709 	/*
710 	 * Now convert the block number to absolute and put it in
711 	 * terms of the device's logical block size.
712 	 */
713 	if (lp->d_secsize == DEV_BSIZE)
714 		blkno = bp->b_blkno;
715 	else if (lp->d_secsize > DEV_BSIZE)
716 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
717 	else
718 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
719 
720 	if (SDPART(bp->b_dev) != RAW_PART)
721 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
722 
723 	bp->b_rawblkno = blkno;
724 
725 	s = splbio();
726 
727 	/*
728 	 * Place it in the queue of disk activities for this disk.
729 	 *
730 	 * XXX Only do disksort() if the current operating mode does not
731 	 * XXX include tagged queueing.
732 	 */
733 	BUFQ_PUT(sd->buf_queue, bp);
734 
735 	/*
736 	 * Tell the device to get going on the transfer if it's
737 	 * not doing anything, otherwise just wait for completion
738 	 */
739 	sdstart(sd->sc_periph);
740 
741 	splx(s);
742 	return;
743 
744 done:
745 	/*
746 	 * Correctly set the buf to indicate a completed xfer
747 	 */
748 	bp->b_resid = bp->b_bcount;
749 	biodone(bp);
750 }
751 
752 /*
753  * sdstart looks to see if there is a buf waiting for the device
754  * and that the device is not already busy. If both are true,
755  * It dequeues the buf and creates a scsi command to perform the
756  * transfer in the buf. The transfer request will call scsipi_done
757  * on completion, which will in turn call this routine again
758  * so that the next queued transfer is performed.
759  * The bufs are queued by the strategy routine (sdstrategy)
760  *
761  * This routine is also called after other non-queued requests
762  * have been made of the scsi driver, to ensure that the queue
763  * continues to be drained.
764  *
765  * must be called at the correct (highish) spl level
766  * sdstart() is called at splbio from sdstrategy, sdrestart and scsipi_done
767  */
768 static void
769 sdstart(struct scsipi_periph *periph)
770 {
771 	struct sd_softc *sd = device_private(periph->periph_dev);
772 	struct disklabel *lp = sd->sc_dk.dk_label;
773 	struct buf *bp = 0;
774 	struct scsipi_rw_16 cmd16;
775 	struct scsipi_rw_10 cmd_big;
776 	struct scsi_rw_6 cmd_small;
777 	struct scsipi_generic *cmdp;
778 	struct scsipi_xfer *xs;
779 	int nblks, cmdlen, error, flags;
780 
781 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
782 	/*
783 	 * Check if the device has room for another command
784 	 */
785 	while (periph->periph_active < periph->periph_openings) {
786 		/*
787 		 * there is excess capacity, but a special waits
788 		 * It'll need the adapter as soon as we clear out of the
789 		 * way and let it run (user level wait).
790 		 */
791 		if (periph->periph_flags & PERIPH_WAITING) {
792 			periph->periph_flags &= ~PERIPH_WAITING;
793 			wakeup((void *)periph);
794 			return;
795 		}
796 
797 		/*
798 		 * If the device has become invalid, abort all the
799 		 * reads and writes until all files have been closed and
800 		 * re-opened
801 		 */
802 		if (__predict_false(
803 		    (periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)) {
804 			if ((bp = BUFQ_GET(sd->buf_queue)) != NULL) {
805 				bp->b_error = EIO;
806 				bp->b_resid = bp->b_bcount;
807 				biodone(bp);
808 				continue;
809 			} else {
810 				return;
811 			}
812 		}
813 
814 		/*
815 		 * See if there is a buf with work for us to do..
816 		 */
817 		if ((bp = BUFQ_PEEK(sd->buf_queue)) == NULL)
818 			return;
819 
820 		/*
821 		 * We have a buf, now we should make a command.
822 		 */
823 
824 		if (lp->d_secsize == DEV_BSIZE)
825 			nblks = bp->b_bcount >> DEV_BSHIFT;
826 		else
827 			nblks = howmany(bp->b_bcount, lp->d_secsize);
828 
829 		/*
830 		 * Fill out the scsi command.  Use the smallest CDB possible
831 		 * (6-byte, 10-byte, or 16-byte).
832 		 */
833 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
834 		    ((nblks & 0xff) == nblks) &&
835 		    !(periph->periph_quirks & PQUIRK_ONLYBIG)) {
836 			/* 6-byte CDB */
837 			memset(&cmd_small, 0, sizeof(cmd_small));
838 			cmd_small.opcode = (bp->b_flags & B_READ) ?
839 			    SCSI_READ_6_COMMAND : SCSI_WRITE_6_COMMAND;
840 			_lto3b(bp->b_rawblkno, cmd_small.addr);
841 			cmd_small.length = nblks & 0xff;
842 			cmdlen = sizeof(cmd_small);
843 			cmdp = (struct scsipi_generic *)&cmd_small;
844 		} else if ((bp->b_rawblkno & 0xffffffff) == bp->b_rawblkno) {
845 			/* 10-byte CDB */
846 			memset(&cmd_big, 0, sizeof(cmd_big));
847 			cmd_big.opcode = (bp->b_flags & B_READ) ?
848 			    READ_10 : WRITE_10;
849 			_lto4b(bp->b_rawblkno, cmd_big.addr);
850 			_lto2b(nblks, cmd_big.length);
851 			cmdlen = sizeof(cmd_big);
852 			cmdp = (struct scsipi_generic *)&cmd_big;
853 		} else {
854 			/* 16-byte CDB */
855 			memset(&cmd16, 0, sizeof(cmd16));
856 			cmd16.opcode = (bp->b_flags & B_READ) ?
857 			    READ_16 : WRITE_16;
858 			_lto8b(bp->b_rawblkno, cmd16.addr);
859 			_lto4b(nblks, cmd16.length);
860 			cmdlen = sizeof(cmd16);
861 			cmdp = (struct scsipi_generic *)&cmd16;
862 		}
863 
864 		/* Instrumentation. */
865 		disk_busy(&sd->sc_dk);
866 
867 		/*
868 		 * Mark the disk dirty so that the cache will be
869 		 * flushed on close.
870 		 */
871 		if ((bp->b_flags & B_READ) == 0)
872 			sd->flags |= SDF_DIRTY;
873 
874 		/*
875 		 * Figure out what flags to use.
876 		 */
877 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
878 		if (bp->b_flags & B_READ)
879 			flags |= XS_CTL_DATA_IN;
880 		else
881 			flags |= XS_CTL_DATA_OUT;
882 
883 		/*
884 		 * Call the routine that chats with the adapter.
885 		 * Note: we cannot sleep as we may be an interrupt
886 		 */
887 		xs = scsipi_make_xs(periph, cmdp, cmdlen,
888 		    (u_char *)bp->b_data, bp->b_bcount,
889 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
890 		if (__predict_false(xs == NULL)) {
891 			/*
892 			 * out of memory. Keep this buffer in the queue, and
893 			 * retry later.
894 			 */
895 			callout_reset(&sd->sc_callout, hz / 2, sdrestart,
896 			    periph);
897 			return;
898 		}
899 		/*
900 		 * need to dequeue the buffer before queuing the command,
901 		 * because cdstart may be called recursively from the
902 		 * HBA driver
903 		 */
904 #ifdef DIAGNOSTIC
905 		if (BUFQ_GET(sd->buf_queue) != bp)
906 			panic("sdstart(): dequeued wrong buf");
907 #else
908 		BUFQ_GET(sd->buf_queue);
909 #endif
910 		error = scsipi_execute_xs(xs);
911 		/* with a scsipi_xfer preallocated, scsipi_command can't fail */
912 		KASSERT(error == 0);
913 	}
914 }
915 
916 static void
917 sdrestart(void *v)
918 {
919 	int s = splbio();
920 	sdstart((struct scsipi_periph *)v);
921 	splx(s);
922 }
923 
924 static void
925 sddone(struct scsipi_xfer *xs, int error)
926 {
927 	struct sd_softc *sd = device_private(xs->xs_periph->periph_dev);
928 	struct buf *bp = xs->bp;
929 
930 	if (sd->flags & SDF_FLUSHING) {
931 		/* Flush completed, no longer dirty. */
932 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
933 	}
934 
935 	if (bp) {
936 		bp->b_error = error;
937 		bp->b_resid = xs->resid;
938 		if (error) {
939 			/* on a read/write error bp->b_resid is zero, so fix */
940 			bp->b_resid = bp->b_bcount;
941 		}
942 
943 		disk_unbusy(&sd->sc_dk, bp->b_bcount - bp->b_resid,
944 		    (bp->b_flags & B_READ));
945 #if NRND > 0
946 		rnd_add_uint32(&sd->rnd_source, bp->b_rawblkno);
947 #endif
948 
949 		biodone(bp);
950 	}
951 }
952 
953 static void
954 sdminphys(struct buf *bp)
955 {
956 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(bp->b_dev));
957 	long xmax;
958 
959 	/*
960 	 * If the device is ancient, we want to make sure that
961 	 * the transfer fits into a 6-byte cdb.
962 	 *
963 	 * XXX Note that the SCSI-I spec says that 256-block transfers
964 	 * are allowed in a 6-byte read/write, and are specified
965 	 * by settng the "length" to 0.  However, we're conservative
966 	 * here, allowing only 255-block transfers in case an
967 	 * ancient device gets confused by length == 0.  A length of 0
968 	 * in a 10-byte read/write actually means 0 blocks.
969 	 */
970 	if ((sd->flags & SDF_ANCIENT) &&
971 	    ((sd->sc_periph->periph_flags &
972 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
973 		xmax = sd->sc_dk.dk_label->d_secsize * 0xff;
974 
975 		if (bp->b_bcount > xmax)
976 			bp->b_bcount = xmax;
977 	}
978 
979 	scsipi_adapter_minphys(sd->sc_periph->periph_channel, bp);
980 }
981 
982 static int
983 sdread(dev_t dev, struct uio *uio, int ioflag)
984 {
985 
986 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
987 }
988 
989 static int
990 sdwrite(dev_t dev, struct uio *uio, int ioflag)
991 {
992 
993 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
994 }
995 
996 /*
997  * Perform special action on behalf of the user
998  * Knows about the internals of this device
999  */
1000 static int
1001 sdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1002 {
1003 	struct sd_softc *sd = device_lookup_private(&sd_cd, SDUNIT(dev));
1004 	struct scsipi_periph *periph = sd->sc_periph;
1005 	int part = SDPART(dev);
1006 	int error = 0;
1007 #ifdef __HAVE_OLD_DISKLABEL
1008 	struct disklabel *newlabel = NULL;
1009 #endif
1010 
1011 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
1012 
1013 	/*
1014 	 * If the device is not valid, some IOCTLs can still be
1015 	 * handled on the raw partition. Check this here.
1016 	 */
1017 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
1018 		switch (cmd) {
1019 		case DIOCKLABEL:
1020 		case DIOCWLABEL:
1021 		case DIOCLOCK:
1022 		case DIOCEJECT:
1023 		case ODIOCEJECT:
1024 		case DIOCGCACHE:
1025 		case DIOCSCACHE:
1026 		case SCIOCIDENTIFY:
1027 		case OSCIOCIDENTIFY:
1028 		case SCIOCCOMMAND:
1029 		case SCIOCDEBUG:
1030 			if (part == RAW_PART)
1031 				break;
1032 		/* FALLTHROUGH */
1033 		default:
1034 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
1035 				return (ENODEV);
1036 			else
1037 				return (EIO);
1038 		}
1039 	}
1040 
1041 	switch (cmd) {
1042 	case DIOCGDINFO:
1043 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
1044 		return (0);
1045 
1046 #ifdef __HAVE_OLD_DISKLABEL
1047 	case ODIOCGDINFO:
1048 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1049 		if (newlabel == NULL)
1050 			return EIO;
1051 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
1052 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1053 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1054 		else
1055 			error = ENOTTY;
1056 		free(newlabel, M_TEMP);
1057 		return error;
1058 #endif
1059 
1060 	case DIOCGPART:
1061 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
1062 		((struct partinfo *)addr)->part =
1063 		    &sd->sc_dk.dk_label->d_partitions[part];
1064 		return (0);
1065 
1066 	case DIOCWDINFO:
1067 	case DIOCSDINFO:
1068 #ifdef __HAVE_OLD_DISKLABEL
1069 	case ODIOCWDINFO:
1070 	case ODIOCSDINFO:
1071 #endif
1072 	{
1073 		struct disklabel *lp;
1074 
1075 		if ((flag & FWRITE) == 0)
1076 			return (EBADF);
1077 
1078 #ifdef __HAVE_OLD_DISKLABEL
1079  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1080 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1081 			if (newlabel == NULL)
1082 				return EIO;
1083 			memset(newlabel, 0, sizeof newlabel);
1084 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1085 			lp = newlabel;
1086 		} else
1087 #endif
1088 		lp = (struct disklabel *)addr;
1089 
1090 		mutex_enter(&sd->sc_dk.dk_openlock);
1091 		sd->flags |= SDF_LABELLING;
1092 
1093 		error = setdisklabel(sd->sc_dk.dk_label,
1094 		    lp, /*sd->sc_dk.dk_openmask : */0,
1095 		    sd->sc_dk.dk_cpulabel);
1096 		if (error == 0) {
1097 			if (cmd == DIOCWDINFO
1098 #ifdef __HAVE_OLD_DISKLABEL
1099 			    || cmd == ODIOCWDINFO
1100 #endif
1101 			   )
1102 				error = writedisklabel(SDLABELDEV(dev),
1103 				    sdstrategy, sd->sc_dk.dk_label,
1104 				    sd->sc_dk.dk_cpulabel);
1105 		}
1106 
1107 		sd->flags &= ~SDF_LABELLING;
1108 		mutex_exit(&sd->sc_dk.dk_openlock);
1109 #ifdef __HAVE_OLD_DISKLABEL
1110 		if (newlabel != NULL)
1111 			free(newlabel, M_TEMP);
1112 #endif
1113 		return (error);
1114 	}
1115 
1116 	case DIOCKLABEL:
1117 		if (*(int *)addr)
1118 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1119 		else
1120 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1121 		return (0);
1122 
1123 	case DIOCWLABEL:
1124 		if ((flag & FWRITE) == 0)
1125 			return (EBADF);
1126 		if (*(int *)addr)
1127 			sd->flags |= SDF_WLABEL;
1128 		else
1129 			sd->flags &= ~SDF_WLABEL;
1130 		return (0);
1131 
1132 	case DIOCLOCK:
1133 		if (periph->periph_flags & PERIPH_REMOVABLE)
1134 			return (scsipi_prevent(periph,
1135 			    (*(int *)addr) ?
1136 			    SPAMR_PREVENT_DT : SPAMR_ALLOW, 0));
1137 		else
1138 			return (ENOTTY);
1139 
1140 	case DIOCEJECT:
1141 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1142 			return (ENOTTY);
1143 		if (*(int *)addr == 0) {
1144 			/*
1145 			 * Don't force eject: check that we are the only
1146 			 * partition open. If so, unlock it.
1147 			 */
1148 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1149 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1150 			    sd->sc_dk.dk_openmask) {
1151 				error = scsipi_prevent(periph, SPAMR_ALLOW,
1152 				    XS_CTL_IGNORE_NOT_READY);
1153 				if (error)
1154 					return (error);
1155 			} else {
1156 				return (EBUSY);
1157 			}
1158 		}
1159 		/* FALLTHROUGH */
1160 	case ODIOCEJECT:
1161 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1162 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1163 
1164 	case DIOCGDEFLABEL:
1165 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1166 		return (0);
1167 
1168 #ifdef __HAVE_OLD_DISKLABEL
1169 	case ODIOCGDEFLABEL:
1170 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1171 		if (newlabel == NULL)
1172 			return EIO;
1173 		sdgetdefaultlabel(sd, newlabel);
1174 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1175 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1176 		else
1177 			error = ENOTTY;
1178 		free(newlabel, M_TEMP);
1179 		return error;
1180 #endif
1181 
1182 	case DIOCGCACHE:
1183 		return (sd_getcache(sd, (int *) addr));
1184 
1185 	case DIOCSCACHE:
1186 		if ((flag & FWRITE) == 0)
1187 			return (EBADF);
1188 		return (sd_setcache(sd, *(int *) addr));
1189 
1190 	case DIOCCACHESYNC:
1191 		/*
1192 		 * XXX Do we really need to care about having a writable
1193 		 * file descriptor here?
1194 		 */
1195 		if ((flag & FWRITE) == 0)
1196 			return (EBADF);
1197 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0)) {
1198 			error = sd_flush(sd, 0);
1199 			if (error)
1200 				sd->flags &= ~SDF_FLUSHING;
1201 			else
1202 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1203 		} else
1204 			error = 0;
1205 		return (error);
1206 
1207 	case DIOCAWEDGE:
1208 	    {
1209 	    	struct dkwedge_info *dkw = (void *) addr;
1210 
1211 		if ((flag & FWRITE) == 0)
1212 			return (EBADF);
1213 
1214 		/* If the ioctl happens here, the parent is us. */
1215 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1216 			sizeof(dkw->dkw_parent));
1217 		return (dkwedge_add(dkw));
1218 	    }
1219 
1220 	case DIOCDWEDGE:
1221 	    {
1222 	    	struct dkwedge_info *dkw = (void *) addr;
1223 
1224 		if ((flag & FWRITE) == 0)
1225 			return (EBADF);
1226 
1227 		/* If the ioctl happens here, the parent is us. */
1228 		strlcpy(dkw->dkw_parent, device_xname(sd->sc_dev),
1229 			sizeof(dkw->dkw_parent));
1230 		return (dkwedge_del(dkw));
1231 	    }
1232 
1233 	case DIOCLWEDGES:
1234 	    {
1235 	    	struct dkwedge_list *dkwl = (void *) addr;
1236 
1237 		return (dkwedge_list(&sd->sc_dk, dkwl, l));
1238 	    }
1239 
1240 	default:
1241 		if (part != RAW_PART)
1242 			return (ENOTTY);
1243 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, l));
1244 	}
1245 
1246 #ifdef DIAGNOSTIC
1247 	panic("sdioctl: impossible");
1248 #endif
1249 }
1250 
1251 static void
1252 sdgetdefaultlabel(struct sd_softc *sd, struct disklabel *lp)
1253 {
1254 
1255 	memset(lp, 0, sizeof(struct disklabel));
1256 
1257 	lp->d_secsize = sd->params.blksize;
1258 	lp->d_ntracks = sd->params.heads;
1259 	lp->d_nsectors = sd->params.sectors;
1260 	lp->d_ncylinders = sd->params.cyls;
1261 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1262 
1263 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1264 	case SCSIPI_BUSTYPE_SCSI:
1265 		lp->d_type = DTYPE_SCSI;
1266 		break;
1267 	case SCSIPI_BUSTYPE_ATAPI:
1268 		lp->d_type = DTYPE_ATAPI;
1269 		break;
1270 	}
1271 	/*
1272 	 * XXX
1273 	 * We could probe the mode pages to figure out what kind of disc it is.
1274 	 * Is this worthwhile?
1275 	 */
1276 	strncpy(lp->d_typename, sd->name, 16);
1277 	strncpy(lp->d_packname, "fictitious", 16);
1278 	lp->d_secperunit = sd->params.disksize;
1279 	lp->d_rpm = sd->params.rot_rate;
1280 	lp->d_interleave = 1;
1281 	lp->d_flags = sd->sc_periph->periph_flags & PERIPH_REMOVABLE ?
1282 	    D_REMOVABLE : 0;
1283 
1284 	lp->d_partitions[RAW_PART].p_offset = 0;
1285 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1286 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1287 	lp->d_npartitions = RAW_PART + 1;
1288 
1289 	lp->d_magic = DISKMAGIC;
1290 	lp->d_magic2 = DISKMAGIC;
1291 	lp->d_checksum = dkcksum(lp);
1292 }
1293 
1294 
1295 /*
1296  * Load the label information on the named device
1297  */
1298 static int
1299 sdgetdisklabel(struct sd_softc *sd)
1300 {
1301 	struct disklabel *lp = sd->sc_dk.dk_label;
1302 	const char *errstring;
1303 
1304 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1305 
1306 	sdgetdefaultlabel(sd, lp);
1307 
1308 	if (lp->d_secpercyl == 0) {
1309 		lp->d_secpercyl = 100;
1310 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1311 	}
1312 
1313 	/*
1314 	 * Call the generic disklabel extraction routine
1315 	 */
1316 	errstring = readdisklabel(MAKESDDEV(0, device_unit(sd->sc_dev),
1317 	    RAW_PART), sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1318 	if (errstring) {
1319 		aprint_error_dev(sd->sc_dev, "%s\n", errstring);
1320 		return EIO;
1321 	}
1322 	return 0;
1323 }
1324 
1325 static void
1326 sd_shutdown(void *arg)
1327 {
1328 	struct sd_softc *sd = arg;
1329 
1330 	/*
1331 	 * If the disk cache needs to be flushed, and the disk supports
1332 	 * it, flush it.  We're cold at this point, so we poll for
1333 	 * completion.
1334 	 */
1335 	if ((sd->flags & SDF_DIRTY) != 0) {
1336 		if (sd_flush(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1337 			aprint_error_dev(sd->sc_dev,
1338 				"cache synchronization failed\n");
1339 			sd->flags &= ~SDF_FLUSHING;
1340 		} else
1341 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1342 	}
1343 }
1344 
1345 static bool
1346 sd_suspend(device_t dv PMF_FN_ARGS)
1347 {
1348 	struct sd_softc *sd = device_private(dv);
1349 
1350 	sd_shutdown(sd); /* XXX no need to poll */
1351 	return true;
1352 }
1353 
1354 /*
1355  * Check Errors
1356  */
1357 static int
1358 sd_interpret_sense(struct scsipi_xfer *xs)
1359 {
1360 	struct scsipi_periph *periph = xs->xs_periph;
1361 	struct scsi_sense_data *sense = &xs->sense.scsi_sense;
1362 	struct sd_softc *sd = device_private(periph->periph_dev);
1363 	int s, error, retval = EJUSTRETURN;
1364 
1365 	/*
1366 	 * If the periph is already recovering, just do the normal
1367 	 * error processing.
1368 	 */
1369 	if (periph->periph_flags & PERIPH_RECOVERING)
1370 		return (retval);
1371 
1372 	/*
1373 	 * Ignore errors from accessing illegal fields (e.g. trying to
1374 	 * lock the door of a digicam, which doesn't have a door that
1375 	 * can be locked) for the SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL command.
1376 	 */
1377 	if (xs->cmd->opcode == SCSI_PREVENT_ALLOW_MEDIUM_REMOVAL &&
1378 	    SSD_SENSE_KEY(sense->flags) == SKEY_ILLEGAL_REQUEST &&
1379 	    sense->asc == 0x24 &&
1380 	    sense->ascq == 0x00) { /* Illegal field in CDB */
1381 		if (!(xs->xs_control & XS_CTL_SILENT)) {
1382 			scsipi_printaddr(periph);
1383 			printf("no door lock\n");
1384 		}
1385 		xs->xs_control |= XS_CTL_IGNORE_ILLEGAL_REQUEST;
1386 		return (retval);
1387 	}
1388 
1389 
1390 
1391 	/*
1392 	 * If the device is not open yet, let the generic code handle it.
1393 	 */
1394 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1395 		return (retval);
1396 
1397 	/*
1398 	 * If it isn't a extended or extended/deferred error, let
1399 	 * the generic code handle it.
1400 	 */
1401 	if (SSD_RCODE(sense->response_code) != SSD_RCODE_CURRENT &&
1402 	    SSD_RCODE(sense->response_code) != SSD_RCODE_DEFERRED)
1403 		return (retval);
1404 
1405 	if (SSD_SENSE_KEY(sense->flags) == SKEY_NOT_READY &&
1406 	    sense->asc == 0x4) {
1407 		if (sense->ascq == 0x01)	{
1408 			/*
1409 			 * Unit In The Process Of Becoming Ready.
1410 			 */
1411 			printf("%s: waiting for pack to spin up...\n",
1412 			    device_xname(sd->sc_dev));
1413 			if (!callout_pending(&periph->periph_callout))
1414 				scsipi_periph_freeze(periph, 1);
1415 			callout_reset(&periph->periph_callout,
1416 			    5 * hz, scsipi_periph_timed_thaw, periph);
1417 			retval = ERESTART;
1418 		} else if (sense->ascq == 0x02) {
1419 			printf("%s: pack is stopped, restarting...\n",
1420 			    device_xname(sd->sc_dev));
1421 			s = splbio();
1422 			periph->periph_flags |= PERIPH_RECOVERING;
1423 			splx(s);
1424 			error = scsipi_start(periph, SSS_START,
1425 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1426 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1427 			if (error) {
1428 				aprint_error_dev(sd->sc_dev,
1429 					"unable to restart pack\n");
1430 				retval = error;
1431 			} else
1432 				retval = ERESTART;
1433 			s = splbio();
1434 			periph->periph_flags &= ~PERIPH_RECOVERING;
1435 			splx(s);
1436 		}
1437 	}
1438 	if (SSD_SENSE_KEY(sense->flags) == SKEY_MEDIUM_ERROR &&
1439 	    sense->asc == 0x31 &&
1440 	    sense->ascq == 0x00)	{ /* maybe for any asq ? */
1441 		/* Medium Format Corrupted */
1442 		retval = EFTYPE;
1443 	}
1444 	return (retval);
1445 }
1446 
1447 
1448 static int
1449 sdsize(dev_t dev)
1450 {
1451 	struct sd_softc *sd;
1452 	int part, unit, omask;
1453 	int size;
1454 
1455 	unit = SDUNIT(dev);
1456 	sd = device_lookup_private(&sd_cd, unit);
1457 	if (sd == NULL)
1458 		return (-1);
1459 
1460 	if (!device_is_active(sd->sc_dev))
1461 		return (-1);
1462 
1463 	part = SDPART(dev);
1464 	omask = sd->sc_dk.dk_openmask & (1 << part);
1465 
1466 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1467 		return (-1);
1468 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1469 		size = -1;
1470 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1471 		size = -1;
1472 	else
1473 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1474 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1475 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1476 		return (-1);
1477 	return (size);
1478 }
1479 
1480 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1481 static struct scsipi_xfer sx;
1482 static int sddoingadump;
1483 
1484 /*
1485  * dump all of physical memory into the partition specified, starting
1486  * at offset 'dumplo' into the partition.
1487  */
1488 static int
1489 sddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1490 {
1491 	struct sd_softc *sd;	/* disk unit to do the I/O */
1492 	struct disklabel *lp;	/* disk's disklabel */
1493 	int	unit, part;
1494 	int	sectorsize;	/* size of a disk sector */
1495 	int	nsects;		/* number of sectors in partition */
1496 	int	sectoff;	/* sector offset of partition */
1497 	int	totwrt;		/* total number of sectors left to write */
1498 	int	nwrt;		/* current number of sectors to write */
1499 	struct scsipi_rw_10 cmd;	/* write command */
1500 	struct scsipi_xfer *xs;	/* ... convenience */
1501 	struct scsipi_periph *periph;
1502 	struct scsipi_channel *chan;
1503 
1504 	/* Check if recursive dump; if so, punt. */
1505 	if (sddoingadump)
1506 		return (EFAULT);
1507 
1508 	/* Mark as active early. */
1509 	sddoingadump = 1;
1510 
1511 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1512 	part = SDPART(dev);
1513 
1514 	/* Check for acceptable drive number. */
1515 	sd = device_lookup_private(&sd_cd, unit);
1516 	if (sd == NULL)
1517 		return (ENXIO);
1518 
1519 	if (!device_is_active(sd->sc_dev))
1520 		return (ENODEV);
1521 
1522 	periph = sd->sc_periph;
1523 	chan = periph->periph_channel;
1524 
1525 	/* Make sure it was initialized. */
1526 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1527 		return (ENXIO);
1528 
1529 	/* Convert to disk sectors.  Request must be a multiple of size. */
1530 	lp = sd->sc_dk.dk_label;
1531 	sectorsize = lp->d_secsize;
1532 	if ((size % sectorsize) != 0)
1533 		return (EFAULT);
1534 	totwrt = size / sectorsize;
1535 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1536 
1537 	nsects = lp->d_partitions[part].p_size;
1538 	sectoff = lp->d_partitions[part].p_offset;
1539 
1540 	/* Check transfer bounds against partition size. */
1541 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1542 		return (EINVAL);
1543 
1544 	/* Offset block number to start of partition. */
1545 	blkno += sectoff;
1546 
1547 	xs = &sx;
1548 
1549 	while (totwrt > 0) {
1550 		nwrt = totwrt;		/* XXX */
1551 #ifndef	SD_DUMP_NOT_TRUSTED
1552 		/*
1553 		 *  Fill out the scsi command
1554 		 */
1555 		memset(&cmd, 0, sizeof(cmd));
1556 		cmd.opcode = WRITE_10;
1557 		_lto4b(blkno, cmd.addr);
1558 		_lto2b(nwrt, cmd.length);
1559 		/*
1560 		 * Fill out the scsipi_xfer structure
1561 		 *    Note: we cannot sleep as we may be an interrupt
1562 		 * don't use scsipi_command() as it may want to wait
1563 		 * for an xs.
1564 		 */
1565 		memset(xs, 0, sizeof(sx));
1566 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1567 		    XS_CTL_DATA_OUT;
1568 		xs->xs_status = 0;
1569 		xs->xs_periph = periph;
1570 		xs->xs_retries = SDRETRIES;
1571 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1572 		xs->cmd = (struct scsipi_generic *)&cmd;
1573 		xs->cmdlen = sizeof(cmd);
1574 		xs->resid = nwrt * sectorsize;
1575 		xs->error = XS_NOERROR;
1576 		xs->bp = 0;
1577 		xs->data = va;
1578 		xs->datalen = nwrt * sectorsize;
1579 		callout_init(&xs->xs_callout, 0);
1580 
1581 		/*
1582 		 * Pass all this info to the scsi driver.
1583 		 */
1584 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1585 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1586 		    xs->error != XS_NOERROR)
1587 			return (EIO);
1588 #else	/* SD_DUMP_NOT_TRUSTED */
1589 		/* Let's just talk about this first... */
1590 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1591 		delay(500 * 1000);	/* half a second */
1592 #endif	/* SD_DUMP_NOT_TRUSTED */
1593 
1594 		/* update block count */
1595 		totwrt -= nwrt;
1596 		blkno += nwrt;
1597 		va = (char *)va + sectorsize * nwrt;
1598 	}
1599 	sddoingadump = 0;
1600 	return (0);
1601 }
1602 
1603 static int
1604 sd_mode_sense(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1605     int page, int flags, int *big)
1606 {
1607 
1608 	if ((sd->sc_periph->periph_quirks & PQUIRK_ONLYBIG) &&
1609 	    !(sd->sc_periph->periph_quirks & PQUIRK_NOBIGMODESENSE)) {
1610 		*big = 1;
1611 		return scsipi_mode_sense_big(sd->sc_periph, byte2, page, sense,
1612 		    size + sizeof(struct scsi_mode_parameter_header_10),
1613 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1614 	} else {
1615 		*big = 0;
1616 		return scsipi_mode_sense(sd->sc_periph, byte2, page, sense,
1617 		    size + sizeof(struct scsi_mode_parameter_header_6),
1618 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1619 	}
1620 }
1621 
1622 static int
1623 sd_mode_select(struct sd_softc *sd, u_int8_t byte2, void *sense, size_t size,
1624     int flags, int big)
1625 {
1626 
1627 	if (big) {
1628 		struct scsi_mode_parameter_header_10 *header = sense;
1629 
1630 		_lto2b(0, header->data_length);
1631 		return scsipi_mode_select_big(sd->sc_periph, byte2, sense,
1632 		    size + sizeof(struct scsi_mode_parameter_header_10),
1633 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1634 	} else {
1635 		struct scsi_mode_parameter_header_6 *header = sense;
1636 
1637 		header->data_length = 0;
1638 		return scsipi_mode_select(sd->sc_periph, byte2, sense,
1639 		    size + sizeof(struct scsi_mode_parameter_header_6),
1640 		    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1641 	}
1642 }
1643 
1644 /*
1645  * sd_validate_blksize:
1646  *
1647  *	Validate the block size.  Print error if periph is specified,
1648  */
1649 static int
1650 sd_validate_blksize(struct scsipi_periph *periph, int len)
1651 {
1652 
1653 	switch (len) {
1654 	case 256:
1655 	case 512:
1656 	case 1024:
1657 	case 2048:
1658 	case 4096:
1659 		return 1;
1660 	}
1661 
1662 	if (periph) {
1663 		scsipi_printaddr(periph);
1664 		printf("%s sector size: 0x%x.  Defaulting to %d bytes.\n",
1665 		    (len ^ (1 << (ffs(len) - 1))) ?
1666 		    "preposterous" : "unsupported",
1667 		    len, SD_DEFAULT_BLKSIZE);
1668 	}
1669 
1670 	return 0;
1671 }
1672 
1673 /*
1674  * sd_read_capacity:
1675  *
1676  *	Find out from the device what its capacity is.
1677  */
1678 static u_int64_t
1679 sd_read_capacity(struct scsipi_periph *periph, int *blksize, int flags)
1680 {
1681 	union {
1682 		struct scsipi_read_capacity_10 cmd;
1683 		struct scsipi_read_capacity_16 cmd16;
1684 	} cmd;
1685 	union {
1686 		struct scsipi_read_capacity_10_data data;
1687 		struct scsipi_read_capacity_16_data data16;
1688 	} *datap;
1689 	uint64_t rv;
1690 
1691 	memset(&cmd, 0, sizeof(cmd));
1692 	cmd.cmd.opcode = READ_CAPACITY_10;
1693 
1694 	/*
1695 	 * Don't allocate data buffer on stack;
1696 	 * The lower driver layer might use the same stack and
1697 	 * if it uses region which is in the same cacheline,
1698 	 * cache flush ops against the data buffer won't work properly.
1699 	 */
1700 	datap = malloc(sizeof(*datap), M_TEMP, M_WAITOK);
1701 	if (datap == NULL)
1702 		return 0;
1703 
1704 	/*
1705 	 * If the command works, interpret the result as a 4 byte
1706 	 * number of blocks
1707 	 */
1708 	rv = 0;
1709 	memset(datap, 0, sizeof(datap->data));
1710 	if (scsipi_command(periph, (void *)&cmd.cmd, sizeof(cmd.cmd),
1711 	    (void *)datap, sizeof(datap->data), SCSIPIRETRIES, 20000, NULL,
1712 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1713 		goto out;
1714 
1715 	if (_4btol(datap->data.addr) != 0xffffffff) {
1716 		*blksize = _4btol(datap->data.length);
1717 		rv = _4btol(datap->data.addr) + 1;
1718 		goto out;
1719 	}
1720 
1721 	/*
1722 	 * Device is larger than can be reflected by READ CAPACITY (10).
1723 	 * Try READ CAPACITY (16).
1724 	 */
1725 
1726 	memset(&cmd, 0, sizeof(cmd));
1727 	cmd.cmd16.opcode = READ_CAPACITY_16;
1728 	cmd.cmd16.byte2 = SRC16_SERVICE_ACTION;
1729 	_lto4b(sizeof(datap->data16), cmd.cmd16.len);
1730 
1731 	memset(datap, 0, sizeof(datap->data16));
1732 	if (scsipi_command(periph, (void *)&cmd.cmd16, sizeof(cmd.cmd16),
1733 	    (void *)datap, sizeof(datap->data16), SCSIPIRETRIES, 20000, NULL,
1734 	    flags | XS_CTL_DATA_IN | XS_CTL_SILENT) != 0)
1735 		goto out;
1736 
1737 	*blksize = _4btol(datap->data16.length);
1738 	rv = _8btol(datap->data16.addr) + 1;
1739 
1740  out:
1741 	free(datap, M_TEMP);
1742 	return rv;
1743 }
1744 
1745 static int
1746 sd_get_simplifiedparms(struct sd_softc *sd, struct disk_parms *dp, int flags)
1747 {
1748 	struct {
1749 		struct scsi_mode_parameter_header_6 header;
1750 		/* no block descriptor */
1751 		u_int8_t pg_code; /* page code (should be 6) */
1752 		u_int8_t pg_length; /* page length (should be 11) */
1753 		u_int8_t wcd; /* bit0: cache disable */
1754 		u_int8_t lbs[2]; /* logical block size */
1755 		u_int8_t size[5]; /* number of log. blocks */
1756 		u_int8_t pp; /* power/performance */
1757 		u_int8_t flags;
1758 		u_int8_t resvd;
1759 	} scsipi_sense;
1760 	u_int64_t blocks;
1761 	int error, blksize;
1762 
1763 	/*
1764 	 * sd_read_capacity (ie "read capacity") and mode sense page 6
1765 	 * give the same information. Do both for now, and check
1766 	 * for consistency.
1767 	 * XXX probably differs for removable media
1768 	 */
1769 	dp->blksize = SD_DEFAULT_BLKSIZE;
1770 	if ((blocks = sd_read_capacity(sd->sc_periph, &blksize, flags)) == 0)
1771 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1772 
1773 	error = scsipi_mode_sense(sd->sc_periph, SMS_DBD, 6,
1774 	    &scsipi_sense.header, sizeof(scsipi_sense),
1775 	    flags | XS_CTL_DATA_ONSTACK, SDRETRIES, 6000);
1776 
1777 	if (error != 0)
1778 		return (SDGP_RESULT_OFFLINE);		/* XXX? */
1779 
1780 	dp->blksize = blksize;
1781 	if (!sd_validate_blksize(NULL, dp->blksize))
1782 		dp->blksize = _2btol(scsipi_sense.lbs);
1783 	if (!sd_validate_blksize(sd->sc_periph, dp->blksize))
1784 		dp->blksize = SD_DEFAULT_BLKSIZE;
1785 
1786 	/*
1787 	 * Create a pseudo-geometry.
1788 	 */
1789 	dp->heads = 64;
1790 	dp->sectors = 32;
1791 	dp->cyls = blocks / (dp->heads * dp->sectors);
1792 	dp->disksize = _5btol(scsipi_sense.size);
1793 	if (dp->disksize <= UINT32_MAX && dp->disksize != blocks) {
1794 		printf("RBC size: mode sense=%llu, get cap=%llu\n",
1795 		       (unsigned long long)dp->disksize,
1796 		       (unsigned long long)blocks);
1797 		dp->disksize = blocks;
1798 	}
1799 	dp->disksize512 = (dp->disksize * dp->blksize) / DEV_BSIZE;
1800 
1801 	return (SDGP_RESULT_OK);
1802 }
1803 
1804 /*
1805  * Get the scsi driver to send a full inquiry to the * device and use the
1806  * results to fill out the disk parameter structure.
1807  */
1808 static int
1809 sd_get_capacity(struct sd_softc *sd, struct disk_parms *dp, int flags)
1810 {
1811 	u_int64_t blocks;
1812 	int error, blksize;
1813 #if 0
1814 	int i;
1815 	u_int8_t *p;
1816 #endif
1817 
1818 	dp->disksize = blocks = sd_read_capacity(sd->sc_periph, &blksize,
1819 	    flags);
1820 	if (blocks == 0) {
1821 		struct scsipi_read_format_capacities cmd;
1822 		struct {
1823 			struct scsipi_capacity_list_header header;
1824 			struct scsipi_capacity_descriptor desc;
1825 		} __packed data;
1826 
1827 		memset(&cmd, 0, sizeof(cmd));
1828 		memset(&data, 0, sizeof(data));
1829 		cmd.opcode = READ_FORMAT_CAPACITIES;
1830 		_lto2b(sizeof(data), cmd.length);
1831 
1832 		error = scsipi_command(sd->sc_periph,
1833 		    (void *)&cmd, sizeof(cmd), (void *)&data, sizeof(data),
1834 		    SDRETRIES, 20000, NULL,
1835 		    flags | XS_CTL_DATA_IN | XS_CTL_DATA_ONSTACK);
1836 		if (error == EFTYPE) {
1837 			/* Medium Format Corrupted, handle as not formatted */
1838 			return (SDGP_RESULT_UNFORMATTED);
1839 		}
1840 		if (error || data.header.length == 0)
1841 			return (SDGP_RESULT_OFFLINE);
1842 
1843 #if 0
1844 printf("rfc: length=%d\n", data.header.length);
1845 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");
1846 #endif
1847 		switch (data.desc.byte5 & SCSIPI_CAP_DESC_CODE_MASK) {
1848 		case SCSIPI_CAP_DESC_CODE_RESERVED:
1849 		case SCSIPI_CAP_DESC_CODE_FORMATTED:
1850 			break;
1851 
1852 		case SCSIPI_CAP_DESC_CODE_UNFORMATTED:
1853 			return (SDGP_RESULT_UNFORMATTED);
1854 
1855 		case SCSIPI_CAP_DESC_CODE_NONE:
1856 			return (SDGP_RESULT_OFFLINE);
1857 		}
1858 
1859 		dp->disksize = blocks = _4btol(data.desc.nblks);
1860 		if (blocks == 0)
1861 			return (SDGP_RESULT_OFFLINE);		/* XXX? */
1862 
1863 		blksize = _3btol(data.desc.blklen);
1864 
1865 	} else if (!sd_validate_blksize(NULL, blksize)) {
1866 		struct sd_mode_sense_data scsipi_sense;
1867 		int big, bsize;
1868 		struct scsi_general_block_descriptor *bdesc;
1869 
1870 		memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1871 		error = sd_mode_sense(sd, 0, &scsipi_sense,
1872 		    sizeof(scsipi_sense.blk_desc), 0, flags | XS_CTL_SILENT, &big);
1873 		if (!error) {
1874 			if (big) {
1875 				bdesc = (void *)(&scsipi_sense.header.big + 1);
1876 				bsize = _2btol(scsipi_sense.header.big.blk_desc_len);
1877 			} else {
1878 				bdesc = (void *)(&scsipi_sense.header.small + 1);
1879 				bsize = scsipi_sense.header.small.blk_desc_len;
1880 			}
1881 
1882 #if 0
1883 printf("page 0 sense:"); for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i; i--, p++) printf(" %02x", *p); printf("\n");
1884 printf("page 0 bsize=%d\n", bsize);
1885 printf("page 0 ok\n");
1886 #endif
1887 
1888 			if (bsize >= 8) {
1889 				blksize = _3btol(bdesc->blklen);
1890 			}
1891 		}
1892 	}
1893 
1894 	if (!sd_validate_blksize(sd->sc_periph, blksize))
1895 		blksize = SD_DEFAULT_BLKSIZE;
1896 
1897 	dp->blksize = blksize;
1898 	dp->disksize512 = (blocks * dp->blksize) / DEV_BSIZE;
1899 	return (0);
1900 }
1901 
1902 static int
1903 sd_get_parms_page4(struct sd_softc *sd, struct disk_parms *dp, int flags)
1904 {
1905 	struct sd_mode_sense_data scsipi_sense;
1906 	int error;
1907 	int big, byte2;
1908 	size_t poffset;
1909 	union scsi_disk_pages *pages;
1910 
1911 	byte2 = SMS_DBD;
1912 again:
1913 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1914 	error = sd_mode_sense(sd, byte2, &scsipi_sense,
1915 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
1916 	    sizeof(scsipi_sense.pages.rigid_geometry), 4,
1917 	    flags | XS_CTL_SILENT, &big);
1918 	if (error) {
1919 		if (byte2 == SMS_DBD) {
1920 			/* No result; try once more with DBD off */
1921 			byte2 = 0;
1922 			goto again;
1923 		}
1924 		return (error);
1925 	}
1926 
1927 	if (big) {
1928 		poffset = sizeof scsipi_sense.header.big;
1929 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
1930 	} else {
1931 		poffset = sizeof scsipi_sense.header.small;
1932 		poffset += scsipi_sense.header.small.blk_desc_len;
1933 	}
1934 
1935 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->rigid_geometry))
1936 		return ERESTART;
1937 
1938 	pages = (void *)((u_long)&scsipi_sense + poffset);
1939 #if 0
1940 	{
1941 		size_t i;
1942 		u_int8_t *p;
1943 
1944 		printf("page 4 sense:");
1945 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
1946 		    i--, p++)
1947 			printf(" %02x", *p);
1948 		printf("\n");
1949 		printf("page 4 pg_code=%d sense=%p/%p\n",
1950 		    pages->rigid_geometry.pg_code, &scsipi_sense, pages);
1951 	}
1952 #endif
1953 
1954 	if ((pages->rigid_geometry.pg_code & PGCODE_MASK) != 4)
1955 		return (ERESTART);
1956 
1957 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
1958 	    ("%d cyls, %d heads, %d precomp, %d red_write, %d land_zone\n",
1959 	    _3btol(pages->rigid_geometry.ncyl),
1960 	    pages->rigid_geometry.nheads,
1961 	    _2btol(pages->rigid_geometry.st_cyl_wp),
1962 	    _2btol(pages->rigid_geometry.st_cyl_rwc),
1963 	    _2btol(pages->rigid_geometry.land_zone)));
1964 
1965 	/*
1966 	 * KLUDGE!! (for zone recorded disks)
1967 	 * give a number of sectors so that sec * trks * cyls
1968 	 * is <= disk_size
1969 	 * can lead to wasted space! THINK ABOUT THIS !
1970 	 */
1971 	dp->heads = pages->rigid_geometry.nheads;
1972 	dp->cyls = _3btol(pages->rigid_geometry.ncyl);
1973 	if (dp->heads == 0 || dp->cyls == 0)
1974 		return (ERESTART);
1975 	dp->sectors = dp->disksize / (dp->heads * dp->cyls);	/* XXX */
1976 
1977 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
1978 	if (dp->rot_rate == 0)
1979 		dp->rot_rate = 3600;
1980 
1981 #if 0
1982 printf("page 4 ok\n");
1983 #endif
1984 	return (0);
1985 }
1986 
1987 static int
1988 sd_get_parms_page5(struct sd_softc *sd, struct disk_parms *dp, int flags)
1989 {
1990 	struct sd_mode_sense_data scsipi_sense;
1991 	int error;
1992 	int big, byte2;
1993 	size_t poffset;
1994 	union scsi_disk_pages *pages;
1995 
1996 	byte2 = SMS_DBD;
1997 again:
1998 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
1999 	error = sd_mode_sense(sd, 0, &scsipi_sense,
2000 	    (byte2 ? 0 : sizeof(scsipi_sense.blk_desc)) +
2001 	    sizeof(scsipi_sense.pages.flex_geometry), 5,
2002 	    flags | XS_CTL_SILENT, &big);
2003 	if (error) {
2004 		if (byte2 == SMS_DBD) {
2005 			/* No result; try once more with DBD off */
2006 			byte2 = 0;
2007 			goto again;
2008 		}
2009 		return (error);
2010 	}
2011 
2012 	if (big) {
2013 		poffset = sizeof scsipi_sense.header.big;
2014 		poffset += _2btol(scsipi_sense.header.big.blk_desc_len);
2015 	} else {
2016 		poffset = sizeof scsipi_sense.header.small;
2017 		poffset += scsipi_sense.header.small.blk_desc_len;
2018 	}
2019 
2020 	if (poffset > sizeof(scsipi_sense) - sizeof(pages->flex_geometry))
2021 		return ERESTART;
2022 
2023 	pages = (void *)((u_long)&scsipi_sense + poffset);
2024 #if 0
2025 	{
2026 		size_t i;
2027 		u_int8_t *p;
2028 
2029 		printf("page 5 sense:");
2030 		for (i = sizeof(scsipi_sense), p = (void *)&scsipi_sense; i;
2031 		    i--, p++)
2032 			printf(" %02x", *p);
2033 		printf("\n");
2034 		printf("page 5 pg_code=%d sense=%p/%p\n",
2035 		    pages->flex_geometry.pg_code, &scsipi_sense, pages);
2036 	}
2037 #endif
2038 
2039 	if ((pages->flex_geometry.pg_code & PGCODE_MASK) != 5)
2040 		return (ERESTART);
2041 
2042 	SC_DEBUG(sd->sc_periph, SCSIPI_DB3,
2043 	    ("%d cyls, %d heads, %d sec, %d bytes/sec\n",
2044 	    _3btol(pages->flex_geometry.ncyl),
2045 	    pages->flex_geometry.nheads,
2046 	    pages->flex_geometry.ph_sec_tr,
2047 	    _2btol(pages->flex_geometry.bytes_s)));
2048 
2049 	dp->heads = pages->flex_geometry.nheads;
2050 	dp->cyls = _2btol(pages->flex_geometry.ncyl);
2051 	dp->sectors = pages->flex_geometry.ph_sec_tr;
2052 	if (dp->heads == 0 || dp->cyls == 0 || dp->sectors == 0)
2053 		return (ERESTART);
2054 
2055 	dp->rot_rate = _2btol(pages->rigid_geometry.rpm);
2056 	if (dp->rot_rate == 0)
2057 		dp->rot_rate = 3600;
2058 
2059 #if 0
2060 printf("page 5 ok\n");
2061 #endif
2062 	return (0);
2063 }
2064 
2065 static int
2066 sd_get_parms(struct sd_softc *sd, struct disk_parms *dp, int flags)
2067 {
2068 	int error;
2069 
2070 	/*
2071 	 * If offline, the SDEV_MEDIA_LOADED flag will be
2072 	 * cleared by the caller if necessary.
2073 	 */
2074 	if (sd->type == T_SIMPLE_DIRECT) {
2075 		error = sd_get_simplifiedparms(sd, dp, flags);
2076 		if (!error)
2077 			disk_blocksize(&sd->sc_dk, dp->blksize);
2078 		return (error);
2079 	}
2080 
2081 	error = sd_get_capacity(sd, dp, flags);
2082 	if (error)
2083 		return (error);
2084 
2085 	disk_blocksize(&sd->sc_dk, dp->blksize);
2086 
2087 	if (sd->type == T_OPTICAL)
2088 		goto page0;
2089 
2090 	if (sd->sc_periph->periph_flags & PERIPH_REMOVABLE) {
2091 		if (!sd_get_parms_page5(sd, dp, flags) ||
2092 		    !sd_get_parms_page4(sd, dp, flags))
2093 			return (SDGP_RESULT_OK);
2094 	} else {
2095 		if (!sd_get_parms_page4(sd, dp, flags) ||
2096 		    !sd_get_parms_page5(sd, dp, flags))
2097 			return (SDGP_RESULT_OK);
2098 	}
2099 
2100 page0:
2101 	printf("%s: fabricating a geometry\n", device_xname(sd->sc_dev));
2102 	/* Try calling driver's method for figuring out geometry. */
2103 	if (!sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom ||
2104 	    !(*sd->sc_periph->periph_channel->chan_adapter->adapt_getgeom)
2105 		(sd->sc_periph, dp, dp->disksize)) {
2106 		/*
2107 		 * Use adaptec standard fictitious geometry
2108 		 * this depends on which controller (e.g. 1542C is
2109 		 * different. but we have to put SOMETHING here..)
2110 		 */
2111 		dp->heads = 64;
2112 		dp->sectors = 32;
2113 		dp->cyls = dp->disksize / (64 * 32);
2114 	}
2115 	dp->rot_rate = 3600;
2116 	return (SDGP_RESULT_OK);
2117 }
2118 
2119 static int
2120 sd_flush(struct sd_softc *sd, int flags)
2121 {
2122 	struct scsipi_periph *periph = sd->sc_periph;
2123 	struct scsi_synchronize_cache_10 cmd;
2124 
2125 	/*
2126 	 * If the device is SCSI-2, issue a SYNCHRONIZE CACHE.
2127 	 * We issue with address 0 length 0, which should be
2128 	 * interpreted by the device as "all remaining blocks
2129 	 * starting at address 0".  We ignore ILLEGAL REQUEST
2130 	 * in the event that the command is not supported by
2131 	 * the device, and poll for completion so that we know
2132 	 * that the cache has actually been flushed.
2133 	 *
2134 	 * Unless, that is, the device can't handle the SYNCHRONIZE CACHE
2135 	 * command, as indicated by our quirks flags.
2136 	 *
2137 	 * XXX What about older devices?
2138 	 */
2139 	if (periph->periph_version < 2 ||
2140 	    (periph->periph_quirks & PQUIRK_NOSYNCCACHE))
2141 		return (0);
2142 
2143 	sd->flags |= SDF_FLUSHING;
2144 	memset(&cmd, 0, sizeof(cmd));
2145 	cmd.opcode = SCSI_SYNCHRONIZE_CACHE_10;
2146 
2147 	return (scsipi_command(periph, (void *)&cmd, sizeof(cmd), 0, 0,
2148 	    SDRETRIES, 100000, NULL, flags | XS_CTL_IGNORE_ILLEGAL_REQUEST));
2149 }
2150 
2151 static int
2152 sd_getcache(struct sd_softc *sd, int *bitsp)
2153 {
2154 	struct scsipi_periph *periph = sd->sc_periph;
2155 	struct sd_mode_sense_data scsipi_sense;
2156 	int error, bits = 0;
2157 	int big;
2158 	union scsi_disk_pages *pages;
2159 
2160 	if (periph->periph_version < 2)
2161 		return (EOPNOTSUPP);
2162 
2163 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2164 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2165 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2166 	if (error)
2167 		return (error);
2168 
2169 	if (big)
2170 		pages = (void *)(&scsipi_sense.header.big + 1);
2171 	else
2172 		pages = (void *)(&scsipi_sense.header.small + 1);
2173 
2174 	if ((pages->caching_params.flags & CACHING_RCD) == 0)
2175 		bits |= DKCACHE_READ;
2176 	if (pages->caching_params.flags & CACHING_WCE)
2177 		bits |= DKCACHE_WRITE;
2178 	if (pages->caching_params.pg_code & PGCODE_PS)
2179 		bits |= DKCACHE_SAVE;
2180 
2181 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2182 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2183 	    sizeof(scsipi_sense.pages.caching_params),
2184 	    SMS_PCTRL_CHANGEABLE|8, 0, &big);
2185 	if (error == 0) {
2186 		if (big)
2187 			pages = (void *)(&scsipi_sense.header.big + 1);
2188 		else
2189 			pages = (void *)(&scsipi_sense.header.small + 1);
2190 
2191 		if (pages->caching_params.flags & CACHING_RCD)
2192 			bits |= DKCACHE_RCHANGE;
2193 		if (pages->caching_params.flags & CACHING_WCE)
2194 			bits |= DKCACHE_WCHANGE;
2195 	}
2196 
2197 	*bitsp = bits;
2198 
2199 	return (0);
2200 }
2201 
2202 static int
2203 sd_setcache(struct sd_softc *sd, int bits)
2204 {
2205 	struct scsipi_periph *periph = sd->sc_periph;
2206 	struct sd_mode_sense_data scsipi_sense;
2207 	int error;
2208 	uint8_t oflags, byte2 = 0;
2209 	int big;
2210 	union scsi_disk_pages *pages;
2211 
2212 	if (periph->periph_version < 2)
2213 		return (EOPNOTSUPP);
2214 
2215 	memset(&scsipi_sense, 0, sizeof(scsipi_sense));
2216 	error = sd_mode_sense(sd, SMS_DBD, &scsipi_sense,
2217 	    sizeof(scsipi_sense.pages.caching_params), 8, 0, &big);
2218 	if (error)
2219 		return (error);
2220 
2221 	if (big)
2222 		pages = (void *)(&scsipi_sense.header.big + 1);
2223 	else
2224 		pages = (void *)(&scsipi_sense.header.small + 1);
2225 
2226 	oflags = pages->caching_params.flags;
2227 
2228 	if (bits & DKCACHE_READ)
2229 		pages->caching_params.flags &= ~CACHING_RCD;
2230 	else
2231 		pages->caching_params.flags |= CACHING_RCD;
2232 
2233 	if (bits & DKCACHE_WRITE)
2234 		pages->caching_params.flags |= CACHING_WCE;
2235 	else
2236 		pages->caching_params.flags &= ~CACHING_WCE;
2237 
2238 	if (oflags == pages->caching_params.flags)
2239 		return (0);
2240 
2241 	pages->caching_params.pg_code &= PGCODE_MASK;
2242 
2243 	if (bits & DKCACHE_SAVE)
2244 		byte2 |= SMS_SP;
2245 
2246 	return (sd_mode_select(sd, byte2|SMS_PF, &scsipi_sense,
2247 	    sizeof(struct scsi_mode_page_header) +
2248 	    pages->caching_params.pg_length, 0, big));
2249 }
2250 
2251 static void
2252 sd_set_properties(struct sd_softc *sd)
2253 {
2254 	prop_dictionary_t disk_info, odisk_info, geom;
2255 
2256 	disk_info = prop_dictionary_create();
2257 
2258 	geom = prop_dictionary_create();
2259 
2260 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2261 	    sd->params.disksize);
2262 
2263 	prop_dictionary_set_uint32(geom, "sector-size",
2264 	    sd->params.blksize);
2265 
2266 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2267 	    sd->params.sectors);
2268 
2269 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2270 	    sd->params.heads);
2271 
2272 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2273 	    sd->params.cyls);
2274 
2275 	prop_dictionary_set(disk_info, "geometry", geom);
2276 	prop_object_release(geom);
2277 
2278 	prop_dictionary_set(device_properties(sd->sc_dev),
2279 	    "disk-info", disk_info);
2280 
2281 	/*
2282 	 * Don't release disk_info here; we keep a reference to it.
2283 	 * disk_detach() will release it when we go away.
2284 	 */
2285 
2286 	odisk_info = sd->sc_dk.dk_info;
2287 	sd->sc_dk.dk_info = disk_info;
2288 	if (odisk_info)
2289 		prop_object_release(odisk_info);
2290 }
2291