xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: sd.c,v 1.178 2001/07/18 18:21:05 thorpej 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 "opt_scsi.h"
57 #include "rnd.h"
58 
59 #include <sys/types.h>
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/file.h>
64 #include <sys/stat.h>
65 #include <sys/ioctl.h>
66 #include <sys/scsiio.h>
67 #include <sys/buf.h>
68 #include <sys/uio.h>
69 #include <sys/malloc.h>
70 #include <sys/errno.h>
71 #include <sys/device.h>
72 #include <sys/disklabel.h>
73 #include <sys/disk.h>
74 #include <sys/proc.h>
75 #include <sys/conf.h>
76 #include <sys/vnode.h>
77 #if NRND > 0
78 #include <sys/rnd.h>
79 #endif
80 
81 #include <dev/scsipi/scsipi_all.h>
82 #include <dev/scsipi/scsi_all.h>
83 #include <dev/scsipi/scsipi_disk.h>
84 #include <dev/scsipi/scsi_disk.h>
85 #include <dev/scsipi/scsiconf.h>
86 #include <dev/scsipi/sdvar.h>
87 
88 #include "sd.h"		/* NSD_SCSIBUS and NSD_ATAPIBUS come from here */
89 
90 #define	SDUNIT(dev)			DISKUNIT(dev)
91 #define	SDPART(dev)			DISKPART(dev)
92 #define	SDMINOR(unit, part)		DISKMINOR(unit, part)
93 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
94 
95 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
96 
97 int	sdlock __P((struct sd_softc *));
98 void	sdunlock __P((struct sd_softc *));
99 void	sdminphys __P((struct buf *));
100 void	sdgetdefaultlabel __P((struct sd_softc *, struct disklabel *));
101 void	sdgetdisklabel __P((struct sd_softc *));
102 void	sdstart __P((struct scsipi_periph *));
103 void	sddone __P((struct scsipi_xfer *));
104 void	sd_shutdown __P((void *));
105 int	sd_reassign_blocks __P((struct sd_softc *, u_long));
106 int	sd_interpret_sense __P((struct scsipi_xfer *));
107 
108 extern struct cfdriver sd_cd;
109 
110 struct dkdriver sddkdriver = { sdstrategy };
111 
112 const struct scsipi_periphsw sd_switch = {
113 	sd_interpret_sense,	/* check our error handler first */
114 	sdstart,		/* have a queue, served by this */
115 	NULL,			/* have no async handler */
116 	sddone,			/* deal with stats at interrupt time */
117 };
118 
119 /*
120  * Attach routine common to atapi & scsi.
121  */
122 void
123 sdattach(parent, sd, periph, ops)
124 	struct device *parent;
125 	struct sd_softc *sd;
126 	struct scsipi_periph *periph;
127 	const struct sd_ops *ops;
128 {
129 	int error, result;
130 	struct disk_parms *dp = &sd->params;
131 	char pbuf[9];
132 
133 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
134 
135 	BUFQ_INIT(&sd->buf_queue);
136 
137 	/*
138 	 * Store information needed to contact our base driver
139 	 */
140 	sd->sc_periph = periph;
141 	sd->sc_ops = ops;
142 
143 	periph->periph_dev = &sd->sc_dev;
144 	periph->periph_switch = &sd_switch;
145 
146         /*
147          * Increase our openings to the maximum-per-periph
148          * supported by the adapter.  This will either be
149          * clamped down or grown by the adapter if necessary.
150          */
151 	periph->periph_openings =
152 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
153 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
154 
155 	/*
156 	 * Initialize and attach the disk structure.
157 	 */
158 	sd->sc_dk.dk_driver = &sddkdriver;
159 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
160 	disk_attach(&sd->sc_dk);
161 
162 #ifdef __BROKEN_DK_ESTABLISH
163 	dk_establish(&sd->sc_dk, &sd->sc_dev);		/* XXX */
164 #endif
165 
166 	/*
167 	 * Use the subdriver to request information regarding the drive.
168 	 */
169 	printf("\n");
170 
171 	error = scsipi_start(periph, SSS_START,
172 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
173 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
174 
175 	if (error)
176 		result = SDGP_RESULT_OFFLINE;
177 	else
178 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
179 		    XS_CTL_DISCOVERY);
180 	printf("%s: ", sd->sc_dev.dv_xname);
181 	switch (result) {
182 	case SDGP_RESULT_OK:
183 		format_bytes(pbuf, sizeof(pbuf),
184 		    (u_int64_t)dp->disksize * dp->blksize);
185 	        printf(
186 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %ld sectors",
187 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
188 		    dp->disksize);
189 		break;
190 
191 	case SDGP_RESULT_OFFLINE:
192 		printf("drive offline");
193 		break;
194 
195 	case SDGP_RESULT_UNFORMATTED:
196 		printf("unformatted media");
197 		break;
198 
199 #ifdef DIAGNOSTIC
200 	default:
201 		panic("sdattach: unknown result from get_parms");
202 		break;
203 #endif
204 	}
205 	printf("\n");
206 
207 	/*
208 	 * Establish a shutdown hook so that we can ensure that
209 	 * our data has actually made it onto the platter at
210 	 * shutdown time.  Note that this relies on the fact
211 	 * that the shutdown hook code puts us at the head of
212 	 * the list (thus guaranteeing that our hook runs before
213 	 * our ancestors').
214 	 */
215 	if ((sd->sc_sdhook =
216 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
217 		printf("%s: WARNING: unable to establish shutdown hook\n",
218 		    sd->sc_dev.dv_xname);
219 
220 #if NRND > 0
221 	/*
222 	 * attach the device into the random source list
223 	 */
224 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
225 			  RND_TYPE_DISK, 0);
226 #endif
227 }
228 
229 int
230 sdactivate(self, act)
231 	struct device *self;
232 	enum devact act;
233 {
234 	int rv = 0;
235 
236 	switch (act) {
237 	case DVACT_ACTIVATE:
238 		rv = EOPNOTSUPP;
239 		break;
240 
241 	case DVACT_DEACTIVATE:
242 		/*
243 		 * Nothing to do; we key off the device's DVF_ACTIVE.
244 		 */
245 		break;
246 	}
247 	return (rv);
248 }
249 
250 int
251 sddetach(self, flags)
252 	struct device *self;
253 	int flags;
254 {
255 	struct sd_softc *sd = (struct sd_softc *) self;
256 	struct buf *bp;
257 	int s, bmaj, cmaj, i, mn;
258 
259 	/* locate the major number */
260 	for (bmaj = 0; bmaj <= nblkdev; bmaj++)
261 		if (bdevsw[bmaj].d_open == sdopen)
262 			break;
263 	for (cmaj = 0; cmaj <= nchrdev; cmaj++)
264 		if (cdevsw[cmaj].d_open == sdopen)
265 			break;
266 
267 	s = splbio();
268 
269 	/* Kill off any queued buffers. */
270 	while ((bp = BUFQ_FIRST(&sd->buf_queue)) != NULL) {
271 		BUFQ_REMOVE(&sd->buf_queue, bp);
272 		bp->b_error = EIO;
273 		bp->b_flags |= B_ERROR;
274 		bp->b_resid = bp->b_bcount;
275 		biodone(bp);
276 	}
277 
278 	/* Kill off any pending commands. */
279 	scsipi_kill_pending(sd->sc_periph);
280 
281 	splx(s);
282 
283 	/* Nuke the vnodes for any open instances */
284 	for (i = 0; i < MAXPARTITIONS; i++) {
285 		mn = SDMINOR(self->dv_unit, i);
286 		vdevgone(bmaj, mn, mn, VBLK);
287 		vdevgone(cmaj, mn, mn, VCHR);
288 	}
289 
290 	/* Detach from the disk list. */
291 	disk_detach(&sd->sc_dk);
292 
293 	/* Get rid of the shutdown hook. */
294 	shutdownhook_disestablish(sd->sc_sdhook);
295 
296 #if NRND > 0
297 	/* Unhook the entropy source. */
298 	rnd_detach_source(&sd->rnd_source);
299 #endif
300 
301 	return (0);
302 }
303 
304 /*
305  * Wait interruptibly for an exclusive lock.
306  *
307  * XXX
308  * Several drivers do this; it should be abstracted and made MP-safe.
309  */
310 int
311 sdlock(sd)
312 	struct sd_softc *sd;
313 {
314 	int error;
315 
316 	while ((sd->flags & SDF_LOCKED) != 0) {
317 		sd->flags |= SDF_WANTED;
318 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
319 			return (error);
320 	}
321 	sd->flags |= SDF_LOCKED;
322 	return (0);
323 }
324 
325 /*
326  * Unlock and wake up any waiters.
327  */
328 void
329 sdunlock(sd)
330 	struct sd_softc *sd;
331 {
332 
333 	sd->flags &= ~SDF_LOCKED;
334 	if ((sd->flags & SDF_WANTED) != 0) {
335 		sd->flags &= ~SDF_WANTED;
336 		wakeup(sd);
337 	}
338 }
339 
340 /*
341  * open the device. Make sure the partition info is a up-to-date as can be.
342  */
343 int
344 sdopen(dev, flag, fmt, p)
345 	dev_t dev;
346 	int flag, fmt;
347 	struct proc *p;
348 {
349 	struct sd_softc *sd;
350 	struct scsipi_periph *periph;
351 	struct scsipi_adapter *adapt;
352 	int unit, part;
353 	int error;
354 
355 	unit = SDUNIT(dev);
356 	if (unit >= sd_cd.cd_ndevs)
357 		return (ENXIO);
358 	sd = sd_cd.cd_devs[unit];
359 	if (sd == NULL)
360 		return (ENXIO);
361 
362 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
363 		return (ENODEV);
364 
365 	periph = sd->sc_periph;
366 	adapt = periph->periph_channel->chan_adapter;
367 	part = SDPART(dev);
368 
369 	SC_DEBUG(periph, SCSIPI_DB1,
370 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
371 	    sd_cd.cd_ndevs, part));
372 
373 	/*
374 	 * If this is the first open of this device, add a reference
375 	 * to the adapter.
376 	 */
377 	if (sd->sc_dk.dk_openmask == 0 &&
378 	    (error = scsipi_adapter_addref(adapt)) != 0)
379 		return (error);
380 
381 	if ((error = sdlock(sd)) != 0)
382 		goto bad4;
383 
384 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
385 		/*
386 		 * If any partition is open, but the disk has been invalidated,
387 		 * disallow further opens of non-raw partition
388 		 */
389 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
390 		    (part != RAW_PART || fmt != S_IFCHR)) {
391 			error = EIO;
392 			goto bad3;
393 		}
394 	} else {
395 		/* Check that it is still responding and ok. */
396 		error = scsipi_test_unit_ready(periph,
397 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
398 		    XS_CTL_IGNORE_NOT_READY);
399 		if (error)
400 			goto bad3;
401 
402 		/*
403 		 * Start the pack spinning if necessary. Always allow the
404 		 * raw parition to be opened, for raw IOCTLs. Data transfers
405 		 * will check for SDEV_MEDIA_LOADED.
406 		 */
407 		error = scsipi_start(periph, SSS_START,
408 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
409 		    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
410 		if (error) {
411 			if (part != RAW_PART || fmt != S_IFCHR)
412 				goto bad3;
413 			else
414 				goto out;
415 		}
416 
417 		periph->periph_flags |= PERIPH_OPEN;
418 
419 		if (periph->periph_flags & PERIPH_REMOVABLE) {
420 			/* Lock the pack in. */
421 			error = scsipi_prevent(periph, PR_PREVENT,
422 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
423 			    XS_CTL_IGNORE_MEDIA_CHANGE);
424 			if (error)
425 				goto bad;
426 		}
427 
428 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
429 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
430 
431 			/*
432 			 * Load the physical device parameters.
433 			 *
434 			 * Note that if media is present but unformatted,
435 			 * we allow the open (so that it can be formatted!).
436 			 * The drive should refuse real I/O, if the media is
437 			 * unformatted.
438 			 */
439 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
440 			    0) == SDGP_RESULT_OFFLINE) {
441 				error = ENXIO;
442 				goto bad2;
443 			}
444 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
445 
446 			/* Load the partition info if not already loaded. */
447 			sdgetdisklabel(sd);
448 			SC_DEBUG(periph, SCSIPI_DB3, ("Disklabel loaded "));
449 		}
450 	}
451 
452 	/* Check that the partition exists. */
453 	if (part != RAW_PART &&
454 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
455 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
456 		error = ENXIO;
457 		goto bad;
458 	}
459 
460 out:	/* Insure only one open at a time. */
461 	switch (fmt) {
462 	case S_IFCHR:
463 		sd->sc_dk.dk_copenmask |= (1 << part);
464 		break;
465 	case S_IFBLK:
466 		sd->sc_dk.dk_bopenmask |= (1 << part);
467 		break;
468 	}
469 	sd->sc_dk.dk_openmask =
470 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
471 
472 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
473 	sdunlock(sd);
474 	return (0);
475 
476 bad2:
477 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
478 
479 bad:
480 	if (sd->sc_dk.dk_openmask == 0) {
481 		scsipi_prevent(periph, PR_ALLOW,
482 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE);
483 		periph->periph_flags &= ~PERIPH_OPEN;
484 	}
485 
486 bad3:
487 	sdunlock(sd);
488 bad4:
489 	if (sd->sc_dk.dk_openmask == 0)
490 		scsipi_adapter_delref(adapt);
491 	return (error);
492 }
493 
494 /*
495  * close the device.. only called if we are the LAST occurence of an open
496  * device.  Convenient now but usually a pain.
497  */
498 int
499 sdclose(dev, flag, fmt, p)
500 	dev_t dev;
501 	int flag, fmt;
502 	struct proc *p;
503 {
504 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
505 	struct scsipi_periph *periph = sd->sc_periph;
506 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
507 	int part = SDPART(dev);
508 	int error;
509 
510 	if ((error = sdlock(sd)) != 0)
511 		return (error);
512 
513 	switch (fmt) {
514 	case S_IFCHR:
515 		sd->sc_dk.dk_copenmask &= ~(1 << part);
516 		break;
517 	case S_IFBLK:
518 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
519 		break;
520 	}
521 	sd->sc_dk.dk_openmask =
522 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
523 
524 	if (sd->sc_dk.dk_openmask == 0) {
525 		/*
526 		 * If the disk cache needs flushing, and the disk supports
527 		 * it, do it now.
528 		 */
529 		if ((sd->flags & SDF_DIRTY) != 0 &&
530 		    sd->sc_ops->sdo_flush != NULL) {
531 			if ((*sd->sc_ops->sdo_flush)(sd, 0)) {
532 				printf("%s: cache synchronization failed\n",
533 				    sd->sc_dev.dv_xname);
534 				sd->flags &= ~SDF_FLUSHING;
535 			} else
536 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
537 		}
538 
539 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
540 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
541 
542 		scsipi_wait_drain(periph);
543 
544 		if (periph->periph_flags & PERIPH_REMOVABLE) {
545 			scsipi_prevent(periph, PR_ALLOW,
546 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
547 			    XS_CTL_IGNORE_NOT_READY);
548 		}
549 		periph->periph_flags &= ~PERIPH_OPEN;
550 
551 		scsipi_wait_drain(periph);
552 
553 		scsipi_adapter_delref(adapt);
554 	}
555 
556 	sdunlock(sd);
557 	return (0);
558 }
559 
560 /*
561  * Actually translate the requested transfer into one the physical driver
562  * can understand.  The transfer is described by a buf and will include
563  * only one physical transfer.
564  */
565 void
566 sdstrategy(bp)
567 	struct buf *bp;
568 {
569 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
570 	struct scsipi_periph *periph = sd->sc_periph;
571 	struct disklabel *lp;
572 	daddr_t blkno;
573 	int s;
574 	boolean_t sector_aligned;
575 
576 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
577 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
578 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
579 	/*
580 	 * If the device has been made invalid, error out
581 	 */
582 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
583 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
584 		if (periph->periph_flags & PERIPH_OPEN)
585 			bp->b_error = EIO;
586 		else
587 			bp->b_error = ENODEV;
588 		goto bad;
589 	}
590 
591 	lp = sd->sc_dk.dk_label;
592 
593 	/*
594 	 * The transfer must be a whole number of blocks, offset must not be
595 	 * negative.
596 	 */
597 	if (lp->d_secsize == DEV_BSIZE) {
598 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
599 	} else {
600 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
601 	}
602 	if (!sector_aligned || bp->b_blkno < 0) {
603 		bp->b_error = EINVAL;
604 		goto bad;
605 	}
606 	/*
607 	 * If it's a null transfer, return immediatly
608 	 */
609 	if (bp->b_bcount == 0)
610 		goto done;
611 
612 	/*
613 	 * Do bounds checking, adjust transfer. if error, process.
614 	 * If end of partition, just return.
615 	 */
616 	if (SDPART(bp->b_dev) != RAW_PART &&
617 	    bounds_check_with_label(bp, lp,
618 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
619 		goto done;
620 
621 	/*
622 	 * Now convert the block number to absolute and put it in
623 	 * terms of the device's logical block size.
624 	 */
625 	if (lp->d_secsize == DEV_BSIZE)
626 		blkno = bp->b_blkno;
627 	else if (lp->d_secsize > DEV_BSIZE)
628 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
629 	else
630 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
631 
632 	if (SDPART(bp->b_dev) != RAW_PART)
633 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
634 
635 	bp->b_rawblkno = blkno;
636 
637 	s = splbio();
638 
639 	/*
640 	 * Place it in the queue of disk activities for this disk.
641 	 *
642 	 * XXX Only do disksort() if the current operating mode does not
643 	 * XXX include tagged queueing.
644 	 */
645 	disksort_blkno(&sd->buf_queue, bp);
646 
647 	/*
648 	 * Tell the device to get going on the transfer if it's
649 	 * not doing anything, otherwise just wait for completion
650 	 */
651 	sdstart(sd->sc_periph);
652 
653 	splx(s);
654 	return;
655 
656 bad:
657 	bp->b_flags |= B_ERROR;
658 done:
659 	/*
660 	 * Correctly set the buf to indicate a completed xfer
661 	 */
662 	bp->b_resid = bp->b_bcount;
663 	biodone(bp);
664 }
665 
666 /*
667  * sdstart looks to see if there is a buf waiting for the device
668  * and that the device is not already busy. If both are true,
669  * It dequeues the buf and creates a scsi command to perform the
670  * transfer in the buf. The transfer request will call scsipi_done
671  * on completion, which will in turn call this routine again
672  * so that the next queued transfer is performed.
673  * The bufs are queued by the strategy routine (sdstrategy)
674  *
675  * This routine is also called after other non-queued requests
676  * have been made of the scsi driver, to ensure that the queue
677  * continues to be drained.
678  *
679  * must be called at the correct (highish) spl level
680  * sdstart() is called at splbio from sdstrategy and scsipi_done
681  */
682 void
683 sdstart(periph)
684 	struct scsipi_periph *periph;
685 {
686 	struct sd_softc *sd = (void *)periph->periph_dev;
687 	struct disklabel *lp = sd->sc_dk.dk_label;
688 	struct buf *bp = 0;
689 	struct scsipi_rw_big cmd_big;
690 #if NSD_SCSIBUS > 0
691 	struct scsi_rw cmd_small;
692 #endif
693 	struct scsipi_generic *cmdp;
694 	int nblks, cmdlen, error, flags;
695 
696 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
697 	/*
698 	 * Check if the device has room for another command
699 	 */
700 	while (periph->periph_active < periph->periph_openings) {
701 		/*
702 		 * there is excess capacity, but a special waits
703 		 * It'll need the adapter as soon as we clear out of the
704 		 * way and let it run (user level wait).
705 		 */
706 		if (periph->periph_flags & PERIPH_WAITING) {
707 			periph->periph_flags &= ~PERIPH_WAITING;
708 			wakeup((caddr_t)periph);
709 			return;
710 		}
711 
712 		/*
713 		 * See if there is a buf with work for us to do..
714 		 */
715 		if ((bp = BUFQ_FIRST(&sd->buf_queue)) == NULL)
716 			return;
717 		BUFQ_REMOVE(&sd->buf_queue, bp);
718 
719 		/*
720 		 * If the device has become invalid, abort all the
721 		 * reads and writes until all files have been closed and
722 		 * re-opened
723 		 */
724 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
725 			bp->b_error = EIO;
726 			bp->b_flags |= B_ERROR;
727 			bp->b_resid = bp->b_bcount;
728 			biodone(bp);
729 			continue;
730 		}
731 
732 		/*
733 		 * We have a buf, now we should make a command.
734 		 */
735 
736 		if (lp->d_secsize == DEV_BSIZE)
737 			nblks = bp->b_bcount >> DEV_BSHIFT;
738 		else
739 			nblks = howmany(bp->b_bcount, lp->d_secsize);
740 
741 #if NSD_SCSIBUS > 0
742 		/*
743 		 *  Fill out the scsi command.  If the transfer will
744 		 *  fit in a "small" cdb, use it.
745 		 */
746 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
747 		    ((nblks & 0xff) == nblks) &&
748 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
749 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
750 			/*
751 			 * We can fit in a small cdb.
752 			 */
753 			memset(&cmd_small, 0, sizeof(cmd_small));
754 			cmd_small.opcode = (bp->b_flags & B_READ) ?
755 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
756 			_lto3b(bp->b_rawblkno, cmd_small.addr);
757 			cmd_small.length = nblks & 0xff;
758 			cmdlen = sizeof(cmd_small);
759 			cmdp = (struct scsipi_generic *)&cmd_small;
760 		} else
761 #endif
762 		{
763 			/*
764 			 * Need a large cdb.
765 			 */
766 			memset(&cmd_big, 0, sizeof(cmd_big));
767 			cmd_big.opcode = (bp->b_flags & B_READ) ?
768 			    READ_BIG : WRITE_BIG;
769 			_lto4b(bp->b_rawblkno, cmd_big.addr);
770 			_lto2b(nblks, cmd_big.length);
771 			cmdlen = sizeof(cmd_big);
772 			cmdp = (struct scsipi_generic *)&cmd_big;
773 		}
774 
775 		/* Instrumentation. */
776 		disk_busy(&sd->sc_dk);
777 
778 		/*
779 		 * Mark the disk dirty so that the cache will be
780 		 * flushed on close.
781 		 */
782 		if ((bp->b_flags & B_READ) == 0)
783 			sd->flags |= SDF_DIRTY;
784 
785 		/*
786 		 * Figure out what flags to use.
787 		 */
788 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC;
789 		if (bp->b_flags & B_READ)
790 			flags |= XS_CTL_DATA_IN;
791 		else
792 			flags |= XS_CTL_DATA_OUT;
793 		if (bp->b_flags & B_ORDERED)
794 			flags |= XS_CTL_ORDERED_TAG;
795 		else
796 			flags |= XS_CTL_SIMPLE_TAG;
797 
798 		/*
799 		 * Call the routine that chats with the adapter.
800 		 * Note: we cannot sleep as we may be an interrupt
801 		 */
802 		error = scsipi_command(periph, cmdp, cmdlen,
803 		    (u_char *)bp->b_data, bp->b_bcount,
804 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
805 		if (error) {
806 			disk_unbusy(&sd->sc_dk, 0);
807 			printf("%s: not queued, error %d\n",
808 			    sd->sc_dev.dv_xname, error);
809 		}
810 	}
811 }
812 
813 void
814 sddone(xs)
815 	struct scsipi_xfer *xs;
816 {
817 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
818 
819 	if (sd->flags & SDF_FLUSHING) {
820 		/* Flush completed, no longer dirty. */
821 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
822 	}
823 
824 	if (xs->bp != NULL) {
825 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid);
826 #if NRND > 0
827 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
828 #endif
829 	}
830 }
831 
832 void
833 sdminphys(bp)
834 	struct buf *bp;
835 {
836 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
837 	long max;
838 
839 	/*
840 	 * If the device is ancient, we want to make sure that
841 	 * the transfer fits into a 6-byte cdb.
842 	 *
843 	 * XXX Note that the SCSI-I spec says that 256-block transfers
844 	 * are allowed in a 6-byte read/write, and are specified
845 	 * by settng the "length" to 0.  However, we're conservative
846 	 * here, allowing only 255-block transfers in case an
847 	 * ancient device gets confused by length == 0.  A length of 0
848 	 * in a 10-byte read/write actually means 0 blocks.
849 	 */
850 	if ((sd->flags & SDF_ANCIENT) &&
851 	    ((sd->sc_periph->periph_flags &
852 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
853 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
854 
855 		if (bp->b_bcount > max)
856 			bp->b_bcount = max;
857 	}
858 
859 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
860 }
861 
862 int
863 sdread(dev, uio, ioflag)
864 	dev_t dev;
865 	struct uio *uio;
866 	int ioflag;
867 {
868 
869 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
870 }
871 
872 int
873 sdwrite(dev, uio, ioflag)
874 	dev_t dev;
875 	struct uio *uio;
876 	int ioflag;
877 {
878 
879 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
880 }
881 
882 /*
883  * Perform special action on behalf of the user
884  * Knows about the internals of this device
885  */
886 int
887 sdioctl(dev, cmd, addr, flag, p)
888 	dev_t dev;
889 	u_long cmd;
890 	caddr_t addr;
891 	int flag;
892 	struct proc *p;
893 {
894 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
895 	struct scsipi_periph *periph = sd->sc_periph;
896 	int part = SDPART(dev);
897 	int error;
898 #ifdef __HAVE_OLD_DISKLABEL
899 	struct disklabel newlabel;
900 #endif
901 
902 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
903 
904 	/*
905 	 * If the device is not valid, some IOCTLs can still be
906 	 * handled on the raw partition. Check this here.
907 	 */
908 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
909 		switch (cmd) {
910 		case DIOCKLABEL:
911 		case DIOCWLABEL:
912 		case DIOCLOCK:
913 		case DIOCEJECT:
914 		case ODIOCEJECT:
915 		case SCIOCIDENTIFY:
916 		case OSCIOCIDENTIFY:
917 		case SCIOCCOMMAND:
918 		case SCIOCDEBUG:
919 			if (part == RAW_PART)
920 				break;
921 		/* FALLTHROUGH */
922 		default:
923 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
924 				return (ENODEV);
925 			else
926 				return (EIO);
927 		}
928 	}
929 
930 	switch (cmd) {
931 	case DIOCGDINFO:
932 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
933 		return (0);
934 
935 #ifdef __HAVE_OLD_DISKLABEL
936 	case ODIOCGDINFO:
937 		newlabel = *(sd->sc_dk.dk_label);
938 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
939 			return ENOTTY;
940 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
941 		return (0);
942 #endif
943 
944 	case DIOCGPART:
945 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
946 		((struct partinfo *)addr)->part =
947 		    &sd->sc_dk.dk_label->d_partitions[part];
948 		return (0);
949 
950 	case DIOCWDINFO:
951 	case DIOCSDINFO:
952 #ifdef __HAVE_OLD_DISKLABEL
953 	case ODIOCWDINFO:
954 	case ODIOCSDINFO:
955 #endif
956 	{
957 		struct disklabel *lp;
958 
959 #ifdef __HAVE_OLD_DISKLABEL
960  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
961 			memset(&newlabel, 0, sizeof newlabel);
962 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
963 			lp = &newlabel;
964 		} else
965 #endif
966 		lp = (struct disklabel *)addr;
967 
968 		if ((flag & FWRITE) == 0)
969 			return (EBADF);
970 
971 		if ((error = sdlock(sd)) != 0)
972 			return (error);
973 		sd->flags |= SDF_LABELLING;
974 
975 		error = setdisklabel(sd->sc_dk.dk_label,
976 		    lp, /*sd->sc_dk.dk_openmask : */0,
977 		    sd->sc_dk.dk_cpulabel);
978 		if (error == 0) {
979 			if (cmd == DIOCWDINFO
980 #ifdef __HAVE_OLD_DISKLABEL
981 			    || cmd == ODIOCWDINFO
982 #endif
983 			   )
984 				error = writedisklabel(SDLABELDEV(dev),
985 				    sdstrategy, sd->sc_dk.dk_label,
986 				    sd->sc_dk.dk_cpulabel);
987 		}
988 
989 		sd->flags &= ~SDF_LABELLING;
990 		sdunlock(sd);
991 		return (error);
992 	}
993 
994 	case DIOCKLABEL:
995 		if (*(int *)addr)
996 			periph->periph_flags |= PERIPH_KEEP_LABEL;
997 		else
998 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
999 		return (0);
1000 
1001 	case DIOCWLABEL:
1002 		if ((flag & FWRITE) == 0)
1003 			return (EBADF);
1004 		if (*(int *)addr)
1005 			sd->flags |= SDF_WLABEL;
1006 		else
1007 			sd->flags &= ~SDF_WLABEL;
1008 		return (0);
1009 
1010 	case DIOCLOCK:
1011 		return (scsipi_prevent(periph,
1012 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1013 
1014 	case DIOCEJECT:
1015 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1016 			return (ENOTTY);
1017 		if (*(int *)addr == 0) {
1018 			/*
1019 			 * Don't force eject: check that we are the only
1020 			 * partition open. If so, unlock it.
1021 			 */
1022 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1023 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1024 			    sd->sc_dk.dk_openmask) {
1025 				error = scsipi_prevent(periph, PR_ALLOW,
1026 				    XS_CTL_IGNORE_NOT_READY);
1027 				if (error)
1028 					return (error);
1029 			} else {
1030 				return (EBUSY);
1031 			}
1032 		}
1033 		/* FALLTHROUGH */
1034 	case ODIOCEJECT:
1035 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1036 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1037 
1038 	case DIOCGDEFLABEL:
1039 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1040 		return (0);
1041 
1042 #ifdef __HAVE_OLD_DISKLABEL
1043 	case ODIOCGDEFLABEL:
1044 		sdgetdefaultlabel(sd, &newlabel);
1045 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1046 			return ENOTTY;
1047 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1048 		return (0);
1049 #endif
1050 
1051 	default:
1052 		if (part != RAW_PART)
1053 			return (ENOTTY);
1054 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1055 	}
1056 
1057 #ifdef DIAGNOSTIC
1058 	panic("sdioctl: impossible");
1059 #endif
1060 }
1061 
1062 void
1063 sdgetdefaultlabel(sd, lp)
1064 	struct sd_softc *sd;
1065 	struct disklabel *lp;
1066 {
1067 
1068 	memset(lp, 0, sizeof(struct disklabel));
1069 
1070 	lp->d_secsize = sd->params.blksize;
1071 	lp->d_ntracks = sd->params.heads;
1072 	lp->d_nsectors = sd->params.sectors;
1073 	lp->d_ncylinders = sd->params.cyls;
1074 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1075 
1076 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1077 #if NSD_SCSIBUS > 0
1078 	case SCSIPI_BUSTYPE_SCSI:
1079 		lp->d_type = DTYPE_SCSI;
1080 		break;
1081 #endif
1082 #if NSD_ATAPIBUS > 0
1083 	case SCSIPI_BUSTYPE_ATAPI:
1084 		lp->d_type = DTYPE_ATAPI;
1085 		break;
1086 #endif
1087 	}
1088 	strncpy(lp->d_typename, sd->name, 16);
1089 	strncpy(lp->d_packname, "fictitious", 16);
1090 	lp->d_secperunit = sd->params.disksize;
1091 	lp->d_rpm = sd->params.rot_rate;
1092 	lp->d_interleave = 1;
1093 	lp->d_flags = 0;
1094 
1095 	lp->d_partitions[RAW_PART].p_offset = 0;
1096 	lp->d_partitions[RAW_PART].p_size =
1097 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1098 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1099 	lp->d_npartitions = RAW_PART + 1;
1100 
1101 	lp->d_magic = DISKMAGIC;
1102 	lp->d_magic2 = DISKMAGIC;
1103 	lp->d_checksum = dkcksum(lp);
1104 }
1105 
1106 
1107 /*
1108  * Load the label information on the named device
1109  */
1110 void
1111 sdgetdisklabel(sd)
1112 	struct sd_softc *sd;
1113 {
1114 	struct disklabel *lp = sd->sc_dk.dk_label;
1115 	char *errstring;
1116 
1117 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1118 
1119 	sdgetdefaultlabel(sd, lp);
1120 
1121 	if (lp->d_secpercyl == 0) {
1122 		lp->d_secpercyl = 100;
1123 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1124 	}
1125 
1126 	/*
1127 	 * Call the generic disklabel extraction routine
1128 	 */
1129 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1130 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1131 	if (errstring) {
1132 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1133 		return;
1134 	}
1135 }
1136 
1137 void
1138 sd_shutdown(arg)
1139 	void *arg;
1140 {
1141 	struct sd_softc *sd = arg;
1142 
1143 	/*
1144 	 * If the disk cache needs to be flushed, and the disk supports
1145 	 * it, flush it.  We're cold at this point, so we poll for
1146 	 * completion.
1147 	 */
1148 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
1149 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1150 			printf("%s: cache synchronization failed\n",
1151 			    sd->sc_dev.dv_xname);
1152 			sd->flags &= ~SDF_FLUSHING;
1153 		} else
1154 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1155 	}
1156 }
1157 
1158 /*
1159  * Tell the device to map out a defective block
1160  */
1161 int
1162 sd_reassign_blocks(sd, blkno)
1163 	struct sd_softc *sd;
1164 	u_long blkno;
1165 {
1166 	struct scsi_reassign_blocks scsipi_cmd;
1167 	struct scsi_reassign_blocks_data rbdata;
1168 
1169 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
1170 	memset(&rbdata, 0, sizeof(rbdata));
1171 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
1172 
1173 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
1174 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
1175 
1176 	return (scsipi_command(sd->sc_periph,
1177 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
1178 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
1179 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
1180 }
1181 
1182 /*
1183  * Check Errors
1184  */
1185 int
1186 sd_interpret_sense(xs)
1187 	struct scsipi_xfer *xs;
1188 {
1189 	struct scsipi_periph *periph = xs->xs_periph;
1190 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1191 	struct sd_softc *sd = (void *)periph->periph_dev;
1192 	int s, error, retval = EJUSTRETURN;
1193 
1194 	/*
1195 	 * If the periph is already recovering, just do the normal
1196 	 * error processing.
1197 	 */
1198 	if (periph->periph_flags & PERIPH_RECOVERING)
1199 		return (retval);
1200 
1201 	/*
1202 	 * If the device is not open yet, let the generic code handle it.
1203 	 */
1204 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1205 		return (retval);
1206 
1207 	/*
1208 	 * If it isn't a extended or extended/deferred error, let
1209 	 * the generic code handle it.
1210 	 */
1211 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1212 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1213 		return (retval);
1214 
1215 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1216 	    sense->add_sense_code == 0x4) {
1217 		if (sense->add_sense_code_qual == 0x01)	{
1218 			/*
1219 			 * Unit In The Process Of Becoming Ready.
1220 			 */
1221 			printf("%s: waiting for pack to spin up...\n",
1222 			    sd->sc_dev.dv_xname);
1223 			scsipi_periph_freeze(periph, 1);
1224 			callout_reset(&periph->periph_callout,
1225 			    5 * hz, scsipi_periph_timed_thaw, periph);
1226 			retval = ERESTART;
1227 		} else if ((sense->add_sense_code_qual == 0x2) &&
1228 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
1229 			printf("%s: pack is stopped, restarting...\n",
1230 			    sd->sc_dev.dv_xname);
1231 			s = splbio();
1232 			periph->periph_flags |= PERIPH_RECOVERING;
1233 			splx(s);
1234 			error = scsipi_start(periph, SSS_START,
1235 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1236 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1237 			if (error) {
1238 				printf("%s: unable to restart pack\n",
1239 				    sd->sc_dev.dv_xname);
1240 				retval = error;
1241 			} else
1242 				retval = ERESTART;
1243 			s = splbio();
1244 			periph->periph_flags &= ~PERIPH_RECOVERING;
1245 			splx(s);
1246 		}
1247 	}
1248 	return (retval);
1249 }
1250 
1251 
1252 int
1253 sdsize(dev)
1254 	dev_t dev;
1255 {
1256 	struct sd_softc *sd;
1257 	int part, unit, omask;
1258 	int size;
1259 
1260 	unit = SDUNIT(dev);
1261 	if (unit >= sd_cd.cd_ndevs)
1262 		return (-1);
1263 	sd = sd_cd.cd_devs[unit];
1264 	if (sd == NULL)
1265 		return (-1);
1266 
1267 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1268 		return (-1);
1269 
1270 	part = SDPART(dev);
1271 	omask = sd->sc_dk.dk_openmask & (1 << part);
1272 
1273 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1274 		return (-1);
1275 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1276 		size = -1;
1277 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1278 		size = -1;
1279 	else
1280 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1281 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1282 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1283 		return (-1);
1284 	return (size);
1285 }
1286 
1287 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1288 static struct scsipi_xfer sx;
1289 static int sddoingadump;
1290 
1291 /*
1292  * dump all of physical memory into the partition specified, starting
1293  * at offset 'dumplo' into the partition.
1294  */
1295 int
1296 sddump(dev, blkno, va, size)
1297 	dev_t dev;
1298 	daddr_t blkno;
1299 	caddr_t va;
1300 	size_t size;
1301 {
1302 	struct sd_softc *sd;	/* disk unit to do the I/O */
1303 	struct disklabel *lp;	/* disk's disklabel */
1304 	int	unit, part;
1305 	int	sectorsize;	/* size of a disk sector */
1306 	int	nsects;		/* number of sectors in partition */
1307 	int	sectoff;	/* sector offset of partition */
1308 	int	totwrt;		/* total number of sectors left to write */
1309 	int	nwrt;		/* current number of sectors to write */
1310 	struct scsipi_rw_big cmd;	/* write command */
1311 	struct scsipi_xfer *xs;	/* ... convenience */
1312 	struct scsipi_periph *periph;
1313 	struct scsipi_channel *chan;
1314 
1315 	/* Check if recursive dump; if so, punt. */
1316 	if (sddoingadump)
1317 		return (EFAULT);
1318 
1319 	/* Mark as active early. */
1320 	sddoingadump = 1;
1321 
1322 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1323 	part = SDPART(dev);
1324 
1325 	/* Check for acceptable drive number. */
1326 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1327 		return (ENXIO);
1328 
1329 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1330 		return (ENODEV);
1331 
1332 	periph = sd->sc_periph;
1333 	chan = periph->periph_channel;
1334 
1335 	/* Make sure it was initialized. */
1336 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1337 		return (ENXIO);
1338 
1339 	/* Convert to disk sectors.  Request must be a multiple of size. */
1340 	lp = sd->sc_dk.dk_label;
1341 	sectorsize = lp->d_secsize;
1342 	if ((size % sectorsize) != 0)
1343 		return (EFAULT);
1344 	totwrt = size / sectorsize;
1345 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1346 
1347 	nsects = lp->d_partitions[part].p_size;
1348 	sectoff = lp->d_partitions[part].p_offset;
1349 
1350 	/* Check transfer bounds against partition size. */
1351 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1352 		return (EINVAL);
1353 
1354 	/* Offset block number to start of partition. */
1355 	blkno += sectoff;
1356 
1357 	xs = &sx;
1358 
1359 	while (totwrt > 0) {
1360 		nwrt = totwrt;		/* XXX */
1361 #ifndef	SD_DUMP_NOT_TRUSTED
1362 		/*
1363 		 *  Fill out the scsi command
1364 		 */
1365 		memset(&cmd, 0, sizeof(cmd));
1366 		cmd.opcode = WRITE_BIG;
1367 		_lto4b(blkno, cmd.addr);
1368 		_lto2b(nwrt, cmd.length);
1369 		/*
1370 		 * Fill out the scsipi_xfer structure
1371 		 *    Note: we cannot sleep as we may be an interrupt
1372 		 * don't use scsipi_command() as it may want to wait
1373 		 * for an xs.
1374 		 */
1375 		memset(xs, 0, sizeof(sx));
1376 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1377 		    XS_CTL_DATA_OUT;
1378 		xs->xs_status = 0;
1379 		xs->xs_periph = periph;
1380 		xs->xs_retries = SDRETRIES;
1381 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1382 		xs->cmd = (struct scsipi_generic *)&cmd;
1383 		xs->cmdlen = sizeof(cmd);
1384 		xs->resid = nwrt * sectorsize;
1385 		xs->error = XS_NOERROR;
1386 		xs->bp = 0;
1387 		xs->data = va;
1388 		xs->datalen = nwrt * sectorsize;
1389 
1390 		/*
1391 		 * Pass all this info to the scsi driver.
1392 		 */
1393 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1394 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1395 		    xs->error != XS_NOERROR)
1396 			return (EIO);
1397 #else	/* SD_DUMP_NOT_TRUSTED */
1398 		/* Let's just talk about this first... */
1399 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1400 		delay(500 * 1000);	/* half a second */
1401 #endif	/* SD_DUMP_NOT_TRUSTED */
1402 
1403 		/* update block count */
1404 		totwrt -= nwrt;
1405 		blkno += nwrt;
1406 		va += sectorsize * nwrt;
1407 	}
1408 	sddoingadump = 0;
1409 	return (0);
1410 }
1411