xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 06be8101a16cc95f40783b3cb7afd12112103a9a)
1 /*	$NetBSD: sd.c,v 1.179 2001/11/13 06:56:40 lukem Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Originally written by Julian Elischer (julian@dialix.oz.au)
41  * for TRW Financial Systems for use under the MACH(2.5) operating system.
42  *
43  * TRW Financial Systems, in accordance with their agreement with Carnegie
44  * Mellon University, makes this software available to CMU to distribute
45  * or use in any manner that they see fit as long as this message is kept with
46  * the software. For this reason TFS also grants any other persons or
47  * organisations permission to use or modify this software.
48  *
49  * TFS supplies this software to be publicly redistributed
50  * on the understanding that TFS is not responsible for the correct
51  * functioning of this software in any circumstances.
52  *
53  * Ported to run under 386BSD by Julian Elischer (julian@dialix.oz.au) Sept 1992
54  */
55 
56 #include <sys/cdefs.h>
57 __KERNEL_RCSID(0, "$NetBSD: sd.c,v 1.179 2001/11/13 06:56:40 lukem Exp $");
58 
59 #include "opt_scsi.h"
60 #include "rnd.h"
61 
62 #include <sys/types.h>
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/file.h>
67 #include <sys/stat.h>
68 #include <sys/ioctl.h>
69 #include <sys/scsiio.h>
70 #include <sys/buf.h>
71 #include <sys/uio.h>
72 #include <sys/malloc.h>
73 #include <sys/errno.h>
74 #include <sys/device.h>
75 #include <sys/disklabel.h>
76 #include <sys/disk.h>
77 #include <sys/proc.h>
78 #include <sys/conf.h>
79 #include <sys/vnode.h>
80 #if NRND > 0
81 #include <sys/rnd.h>
82 #endif
83 
84 #include <dev/scsipi/scsipi_all.h>
85 #include <dev/scsipi/scsi_all.h>
86 #include <dev/scsipi/scsipi_disk.h>
87 #include <dev/scsipi/scsi_disk.h>
88 #include <dev/scsipi/scsiconf.h>
89 #include <dev/scsipi/sdvar.h>
90 
91 #include "sd.h"		/* NSD_SCSIBUS and NSD_ATAPIBUS come from here */
92 
93 #define	SDUNIT(dev)			DISKUNIT(dev)
94 #define	SDPART(dev)			DISKPART(dev)
95 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
96 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
97 
98 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
99 
100 int	sdlock __P((struct sd_softc *));
101 void	sdunlock __P((struct sd_softc *));
102 void	sdminphys __P((struct buf *));
103 void	sdgetdefaultlabel __P((struct sd_softc *, struct disklabel *));
104 void	sdgetdisklabel __P((struct sd_softc *));
105 void	sdstart __P((struct scsipi_periph *));
106 void	sddone __P((struct scsipi_xfer *));
107 void	sd_shutdown __P((void *));
108 int	sd_reassign_blocks __P((struct sd_softc *, u_long));
109 int	sd_interpret_sense __P((struct scsipi_xfer *));
110 
111 extern struct cfdriver sd_cd;
112 
113 struct dkdriver sddkdriver = { sdstrategy };
114 
115 const struct scsipi_periphsw sd_switch = {
116 	sd_interpret_sense,	/* check our error handler first */
117 	sdstart,		/* have a queue, served by this */
118 	NULL,			/* have no async handler */
119 	sddone,			/* deal with stats at interrupt time */
120 };
121 
122 /*
123  * Attach routine common to atapi & scsi.
124  */
125 void
126 sdattach(parent, sd, periph, ops)
127 	struct device *parent;
128 	struct sd_softc *sd;
129 	struct scsipi_periph *periph;
130 	const struct sd_ops *ops;
131 {
132 	int error, result;
133 	struct disk_parms *dp = &sd->params;
134 	char pbuf[9];
135 
136 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
137 
138 	BUFQ_INIT(&sd->buf_queue);
139 
140 	/*
141 	 * Store information needed to contact our base driver
142 	 */
143 	sd->sc_periph = periph;
144 	sd->sc_ops = ops;
145 
146 	periph->periph_dev = &sd->sc_dev;
147 	periph->periph_switch = &sd_switch;
148 
149         /*
150          * Increase our openings to the maximum-per-periph
151          * supported by the adapter.  This will either be
152          * clamped down or grown by the adapter if necessary.
153          */
154 	periph->periph_openings =
155 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
156 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
157 
158 	/*
159 	 * Initialize and attach the disk structure.
160 	 */
161 	sd->sc_dk.dk_driver = &sddkdriver;
162 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
163 	disk_attach(&sd->sc_dk);
164 
165 #ifdef __BROKEN_DK_ESTABLISH
166 	dk_establish(&sd->sc_dk, &sd->sc_dev);		/* XXX */
167 #endif
168 
169 	/*
170 	 * Use the subdriver to request information regarding the drive.
171 	 */
172 	printf("\n");
173 
174 	error = scsipi_start(periph, SSS_START,
175 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
176 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
177 
178 	if (error)
179 		result = SDGP_RESULT_OFFLINE;
180 	else
181 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
182 		    XS_CTL_DISCOVERY);
183 	printf("%s: ", sd->sc_dev.dv_xname);
184 	switch (result) {
185 	case SDGP_RESULT_OK:
186 		format_bytes(pbuf, sizeof(pbuf),
187 		    (u_int64_t)dp->disksize * dp->blksize);
188 	        printf(
189 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %ld sectors",
190 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
191 		    dp->disksize);
192 		break;
193 
194 	case SDGP_RESULT_OFFLINE:
195 		printf("drive offline");
196 		break;
197 
198 	case SDGP_RESULT_UNFORMATTED:
199 		printf("unformatted media");
200 		break;
201 
202 #ifdef DIAGNOSTIC
203 	default:
204 		panic("sdattach: unknown result from get_parms");
205 		break;
206 #endif
207 	}
208 	printf("\n");
209 
210 	/*
211 	 * Establish a shutdown hook so that we can ensure that
212 	 * our data has actually made it onto the platter at
213 	 * shutdown time.  Note that this relies on the fact
214 	 * that the shutdown hook code puts us at the head of
215 	 * the list (thus guaranteeing that our hook runs before
216 	 * our ancestors').
217 	 */
218 	if ((sd->sc_sdhook =
219 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
220 		printf("%s: WARNING: unable to establish shutdown hook\n",
221 		    sd->sc_dev.dv_xname);
222 
223 #if NRND > 0
224 	/*
225 	 * attach the device into the random source list
226 	 */
227 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
228 			  RND_TYPE_DISK, 0);
229 #endif
230 }
231 
232 int
233 sdactivate(self, act)
234 	struct device *self;
235 	enum devact act;
236 {
237 	int rv = 0;
238 
239 	switch (act) {
240 	case DVACT_ACTIVATE:
241 		rv = EOPNOTSUPP;
242 		break;
243 
244 	case DVACT_DEACTIVATE:
245 		/*
246 		 * Nothing to do; we key off the device's DVF_ACTIVE.
247 		 */
248 		break;
249 	}
250 	return (rv);
251 }
252 
253 int
254 sddetach(self, flags)
255 	struct device *self;
256 	int flags;
257 {
258 	struct sd_softc *sd = (struct sd_softc *) self;
259 	struct buf *bp;
260 	int s, bmaj, cmaj, i, mn;
261 
262 	/* locate the major number */
263 	for (bmaj = 0; bmaj <= nblkdev; bmaj++)
264 		if (bdevsw[bmaj].d_open == sdopen)
265 			break;
266 	for (cmaj = 0; cmaj <= nchrdev; cmaj++)
267 		if (cdevsw[cmaj].d_open == sdopen)
268 			break;
269 
270 	s = splbio();
271 
272 	/* Kill off any queued buffers. */
273 	while ((bp = BUFQ_FIRST(&sd->buf_queue)) != NULL) {
274 		BUFQ_REMOVE(&sd->buf_queue, bp);
275 		bp->b_error = EIO;
276 		bp->b_flags |= B_ERROR;
277 		bp->b_resid = bp->b_bcount;
278 		biodone(bp);
279 	}
280 
281 	/* Kill off any pending commands. */
282 	scsipi_kill_pending(sd->sc_periph);
283 
284 	splx(s);
285 
286 	/* Nuke the vnodes for any open instances */
287 	for (i = 0; i < MAXPARTITIONS; i++) {
288 		mn = SDMINOR(self->dv_unit, i);
289 		vdevgone(bmaj, mn, mn, VBLK);
290 		vdevgone(cmaj, mn, mn, VCHR);
291 	}
292 
293 	/* Detach from the disk list. */
294 	disk_detach(&sd->sc_dk);
295 
296 	/* Get rid of the shutdown hook. */
297 	shutdownhook_disestablish(sd->sc_sdhook);
298 
299 #if NRND > 0
300 	/* Unhook the entropy source. */
301 	rnd_detach_source(&sd->rnd_source);
302 #endif
303 
304 	return (0);
305 }
306 
307 /*
308  * Wait interruptibly for an exclusive lock.
309  *
310  * XXX
311  * Several drivers do this; it should be abstracted and made MP-safe.
312  */
313 int
314 sdlock(sd)
315 	struct sd_softc *sd;
316 {
317 	int error;
318 
319 	while ((sd->flags & SDF_LOCKED) != 0) {
320 		sd->flags |= SDF_WANTED;
321 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
322 			return (error);
323 	}
324 	sd->flags |= SDF_LOCKED;
325 	return (0);
326 }
327 
328 /*
329  * Unlock and wake up any waiters.
330  */
331 void
332 sdunlock(sd)
333 	struct sd_softc *sd;
334 {
335 
336 	sd->flags &= ~SDF_LOCKED;
337 	if ((sd->flags & SDF_WANTED) != 0) {
338 		sd->flags &= ~SDF_WANTED;
339 		wakeup(sd);
340 	}
341 }
342 
343 /*
344  * open the device. Make sure the partition info is a up-to-date as can be.
345  */
346 int
347 sdopen(dev, flag, fmt, p)
348 	dev_t dev;
349 	int flag, fmt;
350 	struct proc *p;
351 {
352 	struct sd_softc *sd;
353 	struct scsipi_periph *periph;
354 	struct scsipi_adapter *adapt;
355 	int unit, part;
356 	int error;
357 
358 	unit = SDUNIT(dev);
359 	if (unit >= sd_cd.cd_ndevs)
360 		return (ENXIO);
361 	sd = sd_cd.cd_devs[unit];
362 	if (sd == NULL)
363 		return (ENXIO);
364 
365 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
366 		return (ENODEV);
367 
368 	periph = sd->sc_periph;
369 	adapt = periph->periph_channel->chan_adapter;
370 	part = SDPART(dev);
371 
372 	SC_DEBUG(periph, SCSIPI_DB1,
373 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
374 	    sd_cd.cd_ndevs, part));
375 
376 	/*
377 	 * If this is the first open of this device, add a reference
378 	 * to the adapter.
379 	 */
380 	if (sd->sc_dk.dk_openmask == 0 &&
381 	    (error = scsipi_adapter_addref(adapt)) != 0)
382 		return (error);
383 
384 	if ((error = sdlock(sd)) != 0)
385 		goto bad4;
386 
387 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
388 		/*
389 		 * If any partition is open, but the disk has been invalidated,
390 		 * disallow further opens of non-raw partition
391 		 */
392 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
393 		    (part != RAW_PART || fmt != S_IFCHR)) {
394 			error = EIO;
395 			goto bad3;
396 		}
397 	} else {
398 		/* Check that it is still responding and ok. */
399 		error = scsipi_test_unit_ready(periph,
400 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
401 		    XS_CTL_IGNORE_NOT_READY);
402 		if (error)
403 			goto bad3;
404 
405 		/*
406 		 * Start the pack spinning if necessary. Always allow the
407 		 * raw parition to be opened, for raw IOCTLs. Data transfers
408 		 * will check for SDEV_MEDIA_LOADED.
409 		 */
410 		error = scsipi_start(periph, SSS_START,
411 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
412 		    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
413 		if (error) {
414 			if (part != RAW_PART || fmt != S_IFCHR)
415 				goto bad3;
416 			else
417 				goto out;
418 		}
419 
420 		periph->periph_flags |= PERIPH_OPEN;
421 
422 		if (periph->periph_flags & PERIPH_REMOVABLE) {
423 			/* Lock the pack in. */
424 			error = scsipi_prevent(periph, PR_PREVENT,
425 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
426 			    XS_CTL_IGNORE_MEDIA_CHANGE);
427 			if (error)
428 				goto bad;
429 		}
430 
431 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
432 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
433 
434 			/*
435 			 * Load the physical device parameters.
436 			 *
437 			 * Note that if media is present but unformatted,
438 			 * we allow the open (so that it can be formatted!).
439 			 * The drive should refuse real I/O, if the media is
440 			 * unformatted.
441 			 */
442 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
443 			    0) == SDGP_RESULT_OFFLINE) {
444 				error = ENXIO;
445 				goto bad2;
446 			}
447 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
448 
449 			/* Load the partition info if not already loaded. */
450 			sdgetdisklabel(sd);
451 			SC_DEBUG(periph, SCSIPI_DB3, ("Disklabel loaded "));
452 		}
453 	}
454 
455 	/* Check that the partition exists. */
456 	if (part != RAW_PART &&
457 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
458 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
459 		error = ENXIO;
460 		goto bad;
461 	}
462 
463 out:	/* Insure only one open at a time. */
464 	switch (fmt) {
465 	case S_IFCHR:
466 		sd->sc_dk.dk_copenmask |= (1 << part);
467 		break;
468 	case S_IFBLK:
469 		sd->sc_dk.dk_bopenmask |= (1 << part);
470 		break;
471 	}
472 	sd->sc_dk.dk_openmask =
473 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
474 
475 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
476 	sdunlock(sd);
477 	return (0);
478 
479 bad2:
480 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
481 
482 bad:
483 	if (sd->sc_dk.dk_openmask == 0) {
484 		scsipi_prevent(periph, PR_ALLOW,
485 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE);
486 		periph->periph_flags &= ~PERIPH_OPEN;
487 	}
488 
489 bad3:
490 	sdunlock(sd);
491 bad4:
492 	if (sd->sc_dk.dk_openmask == 0)
493 		scsipi_adapter_delref(adapt);
494 	return (error);
495 }
496 
497 /*
498  * close the device.. only called if we are the LAST occurence of an open
499  * device.  Convenient now but usually a pain.
500  */
501 int
502 sdclose(dev, flag, fmt, p)
503 	dev_t dev;
504 	int flag, fmt;
505 	struct proc *p;
506 {
507 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
508 	struct scsipi_periph *periph = sd->sc_periph;
509 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
510 	int part = SDPART(dev);
511 	int error;
512 
513 	if ((error = sdlock(sd)) != 0)
514 		return (error);
515 
516 	switch (fmt) {
517 	case S_IFCHR:
518 		sd->sc_dk.dk_copenmask &= ~(1 << part);
519 		break;
520 	case S_IFBLK:
521 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
522 		break;
523 	}
524 	sd->sc_dk.dk_openmask =
525 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
526 
527 	if (sd->sc_dk.dk_openmask == 0) {
528 		/*
529 		 * If the disk cache needs flushing, and the disk supports
530 		 * it, do it now.
531 		 */
532 		if ((sd->flags & SDF_DIRTY) != 0 &&
533 		    sd->sc_ops->sdo_flush != NULL) {
534 			if ((*sd->sc_ops->sdo_flush)(sd, 0)) {
535 				printf("%s: cache synchronization failed\n",
536 				    sd->sc_dev.dv_xname);
537 				sd->flags &= ~SDF_FLUSHING;
538 			} else
539 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
540 		}
541 
542 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
543 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
544 
545 		scsipi_wait_drain(periph);
546 
547 		if (periph->periph_flags & PERIPH_REMOVABLE) {
548 			scsipi_prevent(periph, PR_ALLOW,
549 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
550 			    XS_CTL_IGNORE_NOT_READY);
551 		}
552 		periph->periph_flags &= ~PERIPH_OPEN;
553 
554 		scsipi_wait_drain(periph);
555 
556 		scsipi_adapter_delref(adapt);
557 	}
558 
559 	sdunlock(sd);
560 	return (0);
561 }
562 
563 /*
564  * Actually translate the requested transfer into one the physical driver
565  * can understand.  The transfer is described by a buf and will include
566  * only one physical transfer.
567  */
568 void
569 sdstrategy(bp)
570 	struct buf *bp;
571 {
572 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
573 	struct scsipi_periph *periph = sd->sc_periph;
574 	struct disklabel *lp;
575 	daddr_t blkno;
576 	int s;
577 	boolean_t sector_aligned;
578 
579 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
580 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
581 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
582 	/*
583 	 * If the device has been made invalid, error out
584 	 */
585 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
586 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
587 		if (periph->periph_flags & PERIPH_OPEN)
588 			bp->b_error = EIO;
589 		else
590 			bp->b_error = ENODEV;
591 		goto bad;
592 	}
593 
594 	lp = sd->sc_dk.dk_label;
595 
596 	/*
597 	 * The transfer must be a whole number of blocks, offset must not be
598 	 * negative.
599 	 */
600 	if (lp->d_secsize == DEV_BSIZE) {
601 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
602 	} else {
603 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
604 	}
605 	if (!sector_aligned || bp->b_blkno < 0) {
606 		bp->b_error = EINVAL;
607 		goto bad;
608 	}
609 	/*
610 	 * If it's a null transfer, return immediatly
611 	 */
612 	if (bp->b_bcount == 0)
613 		goto done;
614 
615 	/*
616 	 * Do bounds checking, adjust transfer. if error, process.
617 	 * If end of partition, just return.
618 	 */
619 	if (SDPART(bp->b_dev) != RAW_PART &&
620 	    bounds_check_with_label(bp, lp,
621 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
622 		goto done;
623 
624 	/*
625 	 * Now convert the block number to absolute and put it in
626 	 * terms of the device's logical block size.
627 	 */
628 	if (lp->d_secsize == DEV_BSIZE)
629 		blkno = bp->b_blkno;
630 	else if (lp->d_secsize > DEV_BSIZE)
631 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
632 	else
633 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
634 
635 	if (SDPART(bp->b_dev) != RAW_PART)
636 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
637 
638 	bp->b_rawblkno = blkno;
639 
640 	s = splbio();
641 
642 	/*
643 	 * Place it in the queue of disk activities for this disk.
644 	 *
645 	 * XXX Only do disksort() if the current operating mode does not
646 	 * XXX include tagged queueing.
647 	 */
648 	disksort_blkno(&sd->buf_queue, bp);
649 
650 	/*
651 	 * Tell the device to get going on the transfer if it's
652 	 * not doing anything, otherwise just wait for completion
653 	 */
654 	sdstart(sd->sc_periph);
655 
656 	splx(s);
657 	return;
658 
659 bad:
660 	bp->b_flags |= B_ERROR;
661 done:
662 	/*
663 	 * Correctly set the buf to indicate a completed xfer
664 	 */
665 	bp->b_resid = bp->b_bcount;
666 	biodone(bp);
667 }
668 
669 /*
670  * sdstart looks to see if there is a buf waiting for the device
671  * and that the device is not already busy. If both are true,
672  * It dequeues the buf and creates a scsi command to perform the
673  * transfer in the buf. The transfer request will call scsipi_done
674  * on completion, which will in turn call this routine again
675  * so that the next queued transfer is performed.
676  * The bufs are queued by the strategy routine (sdstrategy)
677  *
678  * This routine is also called after other non-queued requests
679  * have been made of the scsi driver, to ensure that the queue
680  * continues to be drained.
681  *
682  * must be called at the correct (highish) spl level
683  * sdstart() is called at splbio from sdstrategy and scsipi_done
684  */
685 void
686 sdstart(periph)
687 	struct scsipi_periph *periph;
688 {
689 	struct sd_softc *sd = (void *)periph->periph_dev;
690 	struct disklabel *lp = sd->sc_dk.dk_label;
691 	struct buf *bp = 0;
692 	struct scsipi_rw_big cmd_big;
693 #if NSD_SCSIBUS > 0
694 	struct scsi_rw cmd_small;
695 #endif
696 	struct scsipi_generic *cmdp;
697 	int nblks, cmdlen, error, flags;
698 
699 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
700 	/*
701 	 * Check if the device has room for another command
702 	 */
703 	while (periph->periph_active < periph->periph_openings) {
704 		/*
705 		 * there is excess capacity, but a special waits
706 		 * It'll need the adapter as soon as we clear out of the
707 		 * way and let it run (user level wait).
708 		 */
709 		if (periph->periph_flags & PERIPH_WAITING) {
710 			periph->periph_flags &= ~PERIPH_WAITING;
711 			wakeup((caddr_t)periph);
712 			return;
713 		}
714 
715 		/*
716 		 * See if there is a buf with work for us to do..
717 		 */
718 		if ((bp = BUFQ_FIRST(&sd->buf_queue)) == NULL)
719 			return;
720 		BUFQ_REMOVE(&sd->buf_queue, bp);
721 
722 		/*
723 		 * If the device has become invalid, abort all the
724 		 * reads and writes until all files have been closed and
725 		 * re-opened
726 		 */
727 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
728 			bp->b_error = EIO;
729 			bp->b_flags |= B_ERROR;
730 			bp->b_resid = bp->b_bcount;
731 			biodone(bp);
732 			continue;
733 		}
734 
735 		/*
736 		 * We have a buf, now we should make a command.
737 		 */
738 
739 		if (lp->d_secsize == DEV_BSIZE)
740 			nblks = bp->b_bcount >> DEV_BSHIFT;
741 		else
742 			nblks = howmany(bp->b_bcount, lp->d_secsize);
743 
744 #if NSD_SCSIBUS > 0
745 		/*
746 		 *  Fill out the scsi command.  If the transfer will
747 		 *  fit in a "small" cdb, use it.
748 		 */
749 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
750 		    ((nblks & 0xff) == nblks) &&
751 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
752 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
753 			/*
754 			 * We can fit in a small cdb.
755 			 */
756 			memset(&cmd_small, 0, sizeof(cmd_small));
757 			cmd_small.opcode = (bp->b_flags & B_READ) ?
758 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
759 			_lto3b(bp->b_rawblkno, cmd_small.addr);
760 			cmd_small.length = nblks & 0xff;
761 			cmdlen = sizeof(cmd_small);
762 			cmdp = (struct scsipi_generic *)&cmd_small;
763 		} else
764 #endif
765 		{
766 			/*
767 			 * Need a large cdb.
768 			 */
769 			memset(&cmd_big, 0, sizeof(cmd_big));
770 			cmd_big.opcode = (bp->b_flags & B_READ) ?
771 			    READ_BIG : WRITE_BIG;
772 			_lto4b(bp->b_rawblkno, cmd_big.addr);
773 			_lto2b(nblks, cmd_big.length);
774 			cmdlen = sizeof(cmd_big);
775 			cmdp = (struct scsipi_generic *)&cmd_big;
776 		}
777 
778 		/* Instrumentation. */
779 		disk_busy(&sd->sc_dk);
780 
781 		/*
782 		 * Mark the disk dirty so that the cache will be
783 		 * flushed on close.
784 		 */
785 		if ((bp->b_flags & B_READ) == 0)
786 			sd->flags |= SDF_DIRTY;
787 
788 		/*
789 		 * Figure out what flags to use.
790 		 */
791 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC;
792 		if (bp->b_flags & B_READ)
793 			flags |= XS_CTL_DATA_IN;
794 		else
795 			flags |= XS_CTL_DATA_OUT;
796 		if (bp->b_flags & B_ORDERED)
797 			flags |= XS_CTL_ORDERED_TAG;
798 		else
799 			flags |= XS_CTL_SIMPLE_TAG;
800 
801 		/*
802 		 * Call the routine that chats with the adapter.
803 		 * Note: we cannot sleep as we may be an interrupt
804 		 */
805 		error = scsipi_command(periph, cmdp, cmdlen,
806 		    (u_char *)bp->b_data, bp->b_bcount,
807 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
808 		if (error) {
809 			disk_unbusy(&sd->sc_dk, 0);
810 			printf("%s: not queued, error %d\n",
811 			    sd->sc_dev.dv_xname, error);
812 		}
813 	}
814 }
815 
816 void
817 sddone(xs)
818 	struct scsipi_xfer *xs;
819 {
820 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
821 
822 	if (sd->flags & SDF_FLUSHING) {
823 		/* Flush completed, no longer dirty. */
824 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
825 	}
826 
827 	if (xs->bp != NULL) {
828 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid);
829 #if NRND > 0
830 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
831 #endif
832 	}
833 }
834 
835 void
836 sdminphys(bp)
837 	struct buf *bp;
838 {
839 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
840 	long max;
841 
842 	/*
843 	 * If the device is ancient, we want to make sure that
844 	 * the transfer fits into a 6-byte cdb.
845 	 *
846 	 * XXX Note that the SCSI-I spec says that 256-block transfers
847 	 * are allowed in a 6-byte read/write, and are specified
848 	 * by settng the "length" to 0.  However, we're conservative
849 	 * here, allowing only 255-block transfers in case an
850 	 * ancient device gets confused by length == 0.  A length of 0
851 	 * in a 10-byte read/write actually means 0 blocks.
852 	 */
853 	if ((sd->flags & SDF_ANCIENT) &&
854 	    ((sd->sc_periph->periph_flags &
855 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
856 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
857 
858 		if (bp->b_bcount > max)
859 			bp->b_bcount = max;
860 	}
861 
862 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
863 }
864 
865 int
866 sdread(dev, uio, ioflag)
867 	dev_t dev;
868 	struct uio *uio;
869 	int ioflag;
870 {
871 
872 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
873 }
874 
875 int
876 sdwrite(dev, uio, ioflag)
877 	dev_t dev;
878 	struct uio *uio;
879 	int ioflag;
880 {
881 
882 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
883 }
884 
885 /*
886  * Perform special action on behalf of the user
887  * Knows about the internals of this device
888  */
889 int
890 sdioctl(dev, cmd, addr, flag, p)
891 	dev_t dev;
892 	u_long cmd;
893 	caddr_t addr;
894 	int flag;
895 	struct proc *p;
896 {
897 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
898 	struct scsipi_periph *periph = sd->sc_periph;
899 	int part = SDPART(dev);
900 	int error;
901 #ifdef __HAVE_OLD_DISKLABEL
902 	struct disklabel newlabel;
903 #endif
904 
905 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
906 
907 	/*
908 	 * If the device is not valid, some IOCTLs can still be
909 	 * handled on the raw partition. Check this here.
910 	 */
911 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
912 		switch (cmd) {
913 		case DIOCKLABEL:
914 		case DIOCWLABEL:
915 		case DIOCLOCK:
916 		case DIOCEJECT:
917 		case ODIOCEJECT:
918 		case SCIOCIDENTIFY:
919 		case OSCIOCIDENTIFY:
920 		case SCIOCCOMMAND:
921 		case SCIOCDEBUG:
922 			if (part == RAW_PART)
923 				break;
924 		/* FALLTHROUGH */
925 		default:
926 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
927 				return (ENODEV);
928 			else
929 				return (EIO);
930 		}
931 	}
932 
933 	switch (cmd) {
934 	case DIOCGDINFO:
935 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
936 		return (0);
937 
938 #ifdef __HAVE_OLD_DISKLABEL
939 	case ODIOCGDINFO:
940 		newlabel = *(sd->sc_dk.dk_label);
941 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
942 			return ENOTTY;
943 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
944 		return (0);
945 #endif
946 
947 	case DIOCGPART:
948 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
949 		((struct partinfo *)addr)->part =
950 		    &sd->sc_dk.dk_label->d_partitions[part];
951 		return (0);
952 
953 	case DIOCWDINFO:
954 	case DIOCSDINFO:
955 #ifdef __HAVE_OLD_DISKLABEL
956 	case ODIOCWDINFO:
957 	case ODIOCSDINFO:
958 #endif
959 	{
960 		struct disklabel *lp;
961 
962 #ifdef __HAVE_OLD_DISKLABEL
963  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
964 			memset(&newlabel, 0, sizeof newlabel);
965 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
966 			lp = &newlabel;
967 		} else
968 #endif
969 		lp = (struct disklabel *)addr;
970 
971 		if ((flag & FWRITE) == 0)
972 			return (EBADF);
973 
974 		if ((error = sdlock(sd)) != 0)
975 			return (error);
976 		sd->flags |= SDF_LABELLING;
977 
978 		error = setdisklabel(sd->sc_dk.dk_label,
979 		    lp, /*sd->sc_dk.dk_openmask : */0,
980 		    sd->sc_dk.dk_cpulabel);
981 		if (error == 0) {
982 			if (cmd == DIOCWDINFO
983 #ifdef __HAVE_OLD_DISKLABEL
984 			    || cmd == ODIOCWDINFO
985 #endif
986 			   )
987 				error = writedisklabel(SDLABELDEV(dev),
988 				    sdstrategy, sd->sc_dk.dk_label,
989 				    sd->sc_dk.dk_cpulabel);
990 		}
991 
992 		sd->flags &= ~SDF_LABELLING;
993 		sdunlock(sd);
994 		return (error);
995 	}
996 
997 	case DIOCKLABEL:
998 		if (*(int *)addr)
999 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1000 		else
1001 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1002 		return (0);
1003 
1004 	case DIOCWLABEL:
1005 		if ((flag & FWRITE) == 0)
1006 			return (EBADF);
1007 		if (*(int *)addr)
1008 			sd->flags |= SDF_WLABEL;
1009 		else
1010 			sd->flags &= ~SDF_WLABEL;
1011 		return (0);
1012 
1013 	case DIOCLOCK:
1014 		return (scsipi_prevent(periph,
1015 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1016 
1017 	case DIOCEJECT:
1018 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1019 			return (ENOTTY);
1020 		if (*(int *)addr == 0) {
1021 			/*
1022 			 * Don't force eject: check that we are the only
1023 			 * partition open. If so, unlock it.
1024 			 */
1025 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1026 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1027 			    sd->sc_dk.dk_openmask) {
1028 				error = scsipi_prevent(periph, PR_ALLOW,
1029 				    XS_CTL_IGNORE_NOT_READY);
1030 				if (error)
1031 					return (error);
1032 			} else {
1033 				return (EBUSY);
1034 			}
1035 		}
1036 		/* FALLTHROUGH */
1037 	case ODIOCEJECT:
1038 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1039 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1040 
1041 	case DIOCGDEFLABEL:
1042 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1043 		return (0);
1044 
1045 #ifdef __HAVE_OLD_DISKLABEL
1046 	case ODIOCGDEFLABEL:
1047 		sdgetdefaultlabel(sd, &newlabel);
1048 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1049 			return ENOTTY;
1050 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1051 		return (0);
1052 #endif
1053 
1054 	default:
1055 		if (part != RAW_PART)
1056 			return (ENOTTY);
1057 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1058 	}
1059 
1060 #ifdef DIAGNOSTIC
1061 	panic("sdioctl: impossible");
1062 #endif
1063 }
1064 
1065 void
1066 sdgetdefaultlabel(sd, lp)
1067 	struct sd_softc *sd;
1068 	struct disklabel *lp;
1069 {
1070 
1071 	memset(lp, 0, sizeof(struct disklabel));
1072 
1073 	lp->d_secsize = sd->params.blksize;
1074 	lp->d_ntracks = sd->params.heads;
1075 	lp->d_nsectors = sd->params.sectors;
1076 	lp->d_ncylinders = sd->params.cyls;
1077 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1078 
1079 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1080 #if NSD_SCSIBUS > 0
1081 	case SCSIPI_BUSTYPE_SCSI:
1082 		lp->d_type = DTYPE_SCSI;
1083 		break;
1084 #endif
1085 #if NSD_ATAPIBUS > 0
1086 	case SCSIPI_BUSTYPE_ATAPI:
1087 		lp->d_type = DTYPE_ATAPI;
1088 		break;
1089 #endif
1090 	}
1091 	strncpy(lp->d_typename, sd->name, 16);
1092 	strncpy(lp->d_packname, "fictitious", 16);
1093 	lp->d_secperunit = sd->params.disksize;
1094 	lp->d_rpm = sd->params.rot_rate;
1095 	lp->d_interleave = 1;
1096 	lp->d_flags = 0;
1097 
1098 	lp->d_partitions[RAW_PART].p_offset = 0;
1099 	lp->d_partitions[RAW_PART].p_size =
1100 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1101 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1102 	lp->d_npartitions = RAW_PART + 1;
1103 
1104 	lp->d_magic = DISKMAGIC;
1105 	lp->d_magic2 = DISKMAGIC;
1106 	lp->d_checksum = dkcksum(lp);
1107 }
1108 
1109 
1110 /*
1111  * Load the label information on the named device
1112  */
1113 void
1114 sdgetdisklabel(sd)
1115 	struct sd_softc *sd;
1116 {
1117 	struct disklabel *lp = sd->sc_dk.dk_label;
1118 	char *errstring;
1119 
1120 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1121 
1122 	sdgetdefaultlabel(sd, lp);
1123 
1124 	if (lp->d_secpercyl == 0) {
1125 		lp->d_secpercyl = 100;
1126 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1127 	}
1128 
1129 	/*
1130 	 * Call the generic disklabel extraction routine
1131 	 */
1132 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1133 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1134 	if (errstring) {
1135 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1136 		return;
1137 	}
1138 }
1139 
1140 void
1141 sd_shutdown(arg)
1142 	void *arg;
1143 {
1144 	struct sd_softc *sd = arg;
1145 
1146 	/*
1147 	 * If the disk cache needs to be flushed, and the disk supports
1148 	 * it, flush it.  We're cold at this point, so we poll for
1149 	 * completion.
1150 	 */
1151 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
1152 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1153 			printf("%s: cache synchronization failed\n",
1154 			    sd->sc_dev.dv_xname);
1155 			sd->flags &= ~SDF_FLUSHING;
1156 		} else
1157 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1158 	}
1159 }
1160 
1161 /*
1162  * Tell the device to map out a defective block
1163  */
1164 int
1165 sd_reassign_blocks(sd, blkno)
1166 	struct sd_softc *sd;
1167 	u_long blkno;
1168 {
1169 	struct scsi_reassign_blocks scsipi_cmd;
1170 	struct scsi_reassign_blocks_data rbdata;
1171 
1172 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
1173 	memset(&rbdata, 0, sizeof(rbdata));
1174 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
1175 
1176 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
1177 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
1178 
1179 	return (scsipi_command(sd->sc_periph,
1180 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
1181 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
1182 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
1183 }
1184 
1185 /*
1186  * Check Errors
1187  */
1188 int
1189 sd_interpret_sense(xs)
1190 	struct scsipi_xfer *xs;
1191 {
1192 	struct scsipi_periph *periph = xs->xs_periph;
1193 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1194 	struct sd_softc *sd = (void *)periph->periph_dev;
1195 	int s, error, retval = EJUSTRETURN;
1196 
1197 	/*
1198 	 * If the periph is already recovering, just do the normal
1199 	 * error processing.
1200 	 */
1201 	if (periph->periph_flags & PERIPH_RECOVERING)
1202 		return (retval);
1203 
1204 	/*
1205 	 * If the device is not open yet, let the generic code handle it.
1206 	 */
1207 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1208 		return (retval);
1209 
1210 	/*
1211 	 * If it isn't a extended or extended/deferred error, let
1212 	 * the generic code handle it.
1213 	 */
1214 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1215 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1216 		return (retval);
1217 
1218 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1219 	    sense->add_sense_code == 0x4) {
1220 		if (sense->add_sense_code_qual == 0x01)	{
1221 			/*
1222 			 * Unit In The Process Of Becoming Ready.
1223 			 */
1224 			printf("%s: waiting for pack to spin up...\n",
1225 			    sd->sc_dev.dv_xname);
1226 			scsipi_periph_freeze(periph, 1);
1227 			callout_reset(&periph->periph_callout,
1228 			    5 * hz, scsipi_periph_timed_thaw, periph);
1229 			retval = ERESTART;
1230 		} else if ((sense->add_sense_code_qual == 0x2) &&
1231 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
1232 			printf("%s: pack is stopped, restarting...\n",
1233 			    sd->sc_dev.dv_xname);
1234 			s = splbio();
1235 			periph->periph_flags |= PERIPH_RECOVERING;
1236 			splx(s);
1237 			error = scsipi_start(periph, SSS_START,
1238 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1239 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1240 			if (error) {
1241 				printf("%s: unable to restart pack\n",
1242 				    sd->sc_dev.dv_xname);
1243 				retval = error;
1244 			} else
1245 				retval = ERESTART;
1246 			s = splbio();
1247 			periph->periph_flags &= ~PERIPH_RECOVERING;
1248 			splx(s);
1249 		}
1250 	}
1251 	return (retval);
1252 }
1253 
1254 
1255 int
1256 sdsize(dev)
1257 	dev_t dev;
1258 {
1259 	struct sd_softc *sd;
1260 	int part, unit, omask;
1261 	int size;
1262 
1263 	unit = SDUNIT(dev);
1264 	if (unit >= sd_cd.cd_ndevs)
1265 		return (-1);
1266 	sd = sd_cd.cd_devs[unit];
1267 	if (sd == NULL)
1268 		return (-1);
1269 
1270 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1271 		return (-1);
1272 
1273 	part = SDPART(dev);
1274 	omask = sd->sc_dk.dk_openmask & (1 << part);
1275 
1276 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1277 		return (-1);
1278 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1279 		size = -1;
1280 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1281 		size = -1;
1282 	else
1283 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1284 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1285 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1286 		return (-1);
1287 	return (size);
1288 }
1289 
1290 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1291 static struct scsipi_xfer sx;
1292 static int sddoingadump;
1293 
1294 /*
1295  * dump all of physical memory into the partition specified, starting
1296  * at offset 'dumplo' into the partition.
1297  */
1298 int
1299 sddump(dev, blkno, va, size)
1300 	dev_t dev;
1301 	daddr_t blkno;
1302 	caddr_t va;
1303 	size_t size;
1304 {
1305 	struct sd_softc *sd;	/* disk unit to do the I/O */
1306 	struct disklabel *lp;	/* disk's disklabel */
1307 	int	unit, part;
1308 	int	sectorsize;	/* size of a disk sector */
1309 	int	nsects;		/* number of sectors in partition */
1310 	int	sectoff;	/* sector offset of partition */
1311 	int	totwrt;		/* total number of sectors left to write */
1312 	int	nwrt;		/* current number of sectors to write */
1313 	struct scsipi_rw_big cmd;	/* write command */
1314 	struct scsipi_xfer *xs;	/* ... convenience */
1315 	struct scsipi_periph *periph;
1316 	struct scsipi_channel *chan;
1317 
1318 	/* Check if recursive dump; if so, punt. */
1319 	if (sddoingadump)
1320 		return (EFAULT);
1321 
1322 	/* Mark as active early. */
1323 	sddoingadump = 1;
1324 
1325 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1326 	part = SDPART(dev);
1327 
1328 	/* Check for acceptable drive number. */
1329 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1330 		return (ENXIO);
1331 
1332 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1333 		return (ENODEV);
1334 
1335 	periph = sd->sc_periph;
1336 	chan = periph->periph_channel;
1337 
1338 	/* Make sure it was initialized. */
1339 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1340 		return (ENXIO);
1341 
1342 	/* Convert to disk sectors.  Request must be a multiple of size. */
1343 	lp = sd->sc_dk.dk_label;
1344 	sectorsize = lp->d_secsize;
1345 	if ((size % sectorsize) != 0)
1346 		return (EFAULT);
1347 	totwrt = size / sectorsize;
1348 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1349 
1350 	nsects = lp->d_partitions[part].p_size;
1351 	sectoff = lp->d_partitions[part].p_offset;
1352 
1353 	/* Check transfer bounds against partition size. */
1354 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1355 		return (EINVAL);
1356 
1357 	/* Offset block number to start of partition. */
1358 	blkno += sectoff;
1359 
1360 	xs = &sx;
1361 
1362 	while (totwrt > 0) {
1363 		nwrt = totwrt;		/* XXX */
1364 #ifndef	SD_DUMP_NOT_TRUSTED
1365 		/*
1366 		 *  Fill out the scsi command
1367 		 */
1368 		memset(&cmd, 0, sizeof(cmd));
1369 		cmd.opcode = WRITE_BIG;
1370 		_lto4b(blkno, cmd.addr);
1371 		_lto2b(nwrt, cmd.length);
1372 		/*
1373 		 * Fill out the scsipi_xfer structure
1374 		 *    Note: we cannot sleep as we may be an interrupt
1375 		 * don't use scsipi_command() as it may want to wait
1376 		 * for an xs.
1377 		 */
1378 		memset(xs, 0, sizeof(sx));
1379 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1380 		    XS_CTL_DATA_OUT;
1381 		xs->xs_status = 0;
1382 		xs->xs_periph = periph;
1383 		xs->xs_retries = SDRETRIES;
1384 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1385 		xs->cmd = (struct scsipi_generic *)&cmd;
1386 		xs->cmdlen = sizeof(cmd);
1387 		xs->resid = nwrt * sectorsize;
1388 		xs->error = XS_NOERROR;
1389 		xs->bp = 0;
1390 		xs->data = va;
1391 		xs->datalen = nwrt * sectorsize;
1392 
1393 		/*
1394 		 * Pass all this info to the scsi driver.
1395 		 */
1396 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1397 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1398 		    xs->error != XS_NOERROR)
1399 			return (EIO);
1400 #else	/* SD_DUMP_NOT_TRUSTED */
1401 		/* Let's just talk about this first... */
1402 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1403 		delay(500 * 1000);	/* half a second */
1404 #endif	/* SD_DUMP_NOT_TRUSTED */
1405 
1406 		/* update block count */
1407 		totwrt -= nwrt;
1408 		blkno += nwrt;
1409 		va += sectorsize * nwrt;
1410 	}
1411 	sddoingadump = 0;
1412 	return (0);
1413 }
1414