xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: sd.c,v 1.202 2003/06/29 22:30:42 fvdl 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 1.202 2003/06/29 22:30:42 fvdl Exp $");
58 
59 #include "opt_scsi.h"
60 #include "opt_bufq.h"
61 #include "rnd.h"
62 
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 dev_type_open(sdopen);
114 dev_type_close(sdclose);
115 dev_type_read(sdread);
116 dev_type_write(sdwrite);
117 dev_type_ioctl(sdioctl);
118 dev_type_strategy(sdstrategy);
119 dev_type_dump(sddump);
120 dev_type_size(sdsize);
121 
122 const struct bdevsw sd_bdevsw = {
123 	sdopen, sdclose, sdstrategy, sdioctl, sddump, sdsize, D_DISK
124 };
125 
126 const struct cdevsw sd_cdevsw = {
127 	sdopen, sdclose, sdread, sdwrite, sdioctl,
128 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
129 };
130 
131 struct dkdriver sddkdriver = { sdstrategy };
132 
133 const struct scsipi_periphsw sd_switch = {
134 	sd_interpret_sense,	/* check our error handler first */
135 	sdstart,		/* have a queue, served by this */
136 	NULL,			/* have no async handler */
137 	sddone,			/* deal with stats at interrupt time */
138 };
139 
140 /*
141  * Attach routine common to atapi & scsi.
142  */
143 void
144 sdattach(parent, sd, periph, ops)
145 	struct device *parent;
146 	struct sd_softc *sd;
147 	struct scsipi_periph *periph;
148 	const struct sd_ops *ops;
149 {
150 	int error, result;
151 	struct disk_parms *dp = &sd->params;
152 	char pbuf[9];
153 
154 	SC_DEBUG(periph, SCSIPI_DB2, ("sdattach: "));
155 
156 #ifdef NEW_BUFQ_STRATEGY
157 	bufq_alloc(&sd->buf_queue, BUFQ_READ_PRIO|BUFQ_SORT_RAWBLOCK);
158 #else
159 	bufq_alloc(&sd->buf_queue, BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
160 #endif
161 
162 	/*
163 	 * Store information needed to contact our base driver
164 	 */
165 	sd->sc_periph = periph;
166 	sd->sc_ops = ops;
167 
168 	periph->periph_dev = &sd->sc_dev;
169 	periph->periph_switch = &sd_switch;
170 
171         /*
172          * Increase our openings to the maximum-per-periph
173          * supported by the adapter.  This will either be
174          * clamped down or grown by the adapter if necessary.
175          */
176 	periph->periph_openings =
177 	    SCSIPI_CHAN_MAX_PERIPH(periph->periph_channel);
178 	periph->periph_flags |= PERIPH_GROW_OPENINGS;
179 
180 	/*
181 	 * Initialize and attach the disk structure.
182 	 */
183 	sd->sc_dk.dk_driver = &sddkdriver;
184 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
185 	disk_attach(&sd->sc_dk);
186 
187 	/*
188 	 * Use the subdriver to request information regarding the drive.
189 	 */
190 	aprint_naive("\n");
191 	aprint_normal("\n");
192 
193 	error = scsipi_start(periph, SSS_START,
194 	    XS_CTL_DISCOVERY | XS_CTL_IGNORE_ILLEGAL_REQUEST |
195 	    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
196 
197 	if (error)
198 		result = SDGP_RESULT_OFFLINE;
199 	else
200 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
201 		    XS_CTL_DISCOVERY);
202 	aprint_normal("%s: ", sd->sc_dev.dv_xname);
203 	switch (result) {
204 	case SDGP_RESULT_OK:
205 		format_bytes(pbuf, sizeof(pbuf),
206 		    (u_int64_t)dp->disksize * dp->blksize);
207 	        aprint_normal(
208 		"%s, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %llu sectors",
209 		    pbuf, dp->cyls, dp->heads, dp->sectors, dp->blksize,
210 		    (unsigned long long)dp->disksize);
211 		break;
212 
213 	case SDGP_RESULT_OFFLINE:
214 		aprint_normal("drive offline");
215 		break;
216 
217 	case SDGP_RESULT_UNFORMATTED:
218 		aprint_normal("unformatted media");
219 		break;
220 
221 #ifdef DIAGNOSTIC
222 	default:
223 		panic("sdattach: unknown result from get_parms");
224 		break;
225 #endif
226 	}
227 	aprint_normal("\n");
228 
229 	/*
230 	 * Establish a shutdown hook so that we can ensure that
231 	 * our data has actually made it onto the platter at
232 	 * shutdown time.  Note that this relies on the fact
233 	 * that the shutdown hook code puts us at the head of
234 	 * the list (thus guaranteeing that our hook runs before
235 	 * our ancestors').
236 	 */
237 	if ((sd->sc_sdhook =
238 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
239 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
240 		    sd->sc_dev.dv_xname);
241 
242 #if NRND > 0
243 	/*
244 	 * attach the device into the random source list
245 	 */
246 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname,
247 			  RND_TYPE_DISK, 0);
248 #endif
249 }
250 
251 int
252 sdactivate(self, act)
253 	struct device *self;
254 	enum devact act;
255 {
256 	int rv = 0;
257 
258 	switch (act) {
259 	case DVACT_ACTIVATE:
260 		rv = EOPNOTSUPP;
261 		break;
262 
263 	case DVACT_DEACTIVATE:
264 		/*
265 		 * Nothing to do; we key off the device's DVF_ACTIVE.
266 		 */
267 		break;
268 	}
269 	return (rv);
270 }
271 
272 int
273 sddetach(self, flags)
274 	struct device *self;
275 	int flags;
276 {
277 	struct sd_softc *sd = (struct sd_softc *) self;
278 	struct buf *bp;
279 	int s, bmaj, cmaj, i, mn;
280 
281 	/* locate the major number */
282 	bmaj = bdevsw_lookup_major(&sd_bdevsw);
283 	cmaj = cdevsw_lookup_major(&sd_cdevsw);
284 
285 	s = splbio();
286 
287 	/* Kill off any queued buffers. */
288 	while ((bp = BUFQ_GET(&sd->buf_queue)) != NULL) {
289 		bp->b_error = EIO;
290 		bp->b_flags |= B_ERROR;
291 		bp->b_resid = bp->b_bcount;
292 		biodone(bp);
293 	}
294 
295 	bufq_free(&sd->buf_queue);
296 
297 	/* Kill off any pending commands. */
298 	scsipi_kill_pending(sd->sc_periph);
299 
300 	splx(s);
301 
302 	/* Nuke the vnodes for any open instances */
303 	for (i = 0; i < MAXPARTITIONS; i++) {
304 		mn = SDMINOR(self->dv_unit, i);
305 		vdevgone(bmaj, mn, mn, VBLK);
306 		vdevgone(cmaj, mn, mn, VCHR);
307 	}
308 
309 	/* Detach from the disk list. */
310 	disk_detach(&sd->sc_dk);
311 
312 	/* Get rid of the shutdown hook. */
313 	shutdownhook_disestablish(sd->sc_sdhook);
314 
315 #if NRND > 0
316 	/* Unhook the entropy source. */
317 	rnd_detach_source(&sd->rnd_source);
318 #endif
319 
320 	return (0);
321 }
322 
323 /*
324  * Wait interruptibly for an exclusive lock.
325  *
326  * XXX
327  * Several drivers do this; it should be abstracted and made MP-safe.
328  */
329 int
330 sdlock(sd)
331 	struct sd_softc *sd;
332 {
333 	int error;
334 
335 	while ((sd->flags & SDF_LOCKED) != 0) {
336 		sd->flags |= SDF_WANTED;
337 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
338 			return (error);
339 	}
340 	sd->flags |= SDF_LOCKED;
341 	return (0);
342 }
343 
344 /*
345  * Unlock and wake up any waiters.
346  */
347 void
348 sdunlock(sd)
349 	struct sd_softc *sd;
350 {
351 
352 	sd->flags &= ~SDF_LOCKED;
353 	if ((sd->flags & SDF_WANTED) != 0) {
354 		sd->flags &= ~SDF_WANTED;
355 		wakeup(sd);
356 	}
357 }
358 
359 /*
360  * open the device. Make sure the partition info is a up-to-date as can be.
361  */
362 int
363 sdopen(dev, flag, fmt, p)
364 	dev_t dev;
365 	int flag, fmt;
366 	struct proc *p;
367 {
368 	struct sd_softc *sd;
369 	struct scsipi_periph *periph;
370 	struct scsipi_adapter *adapt;
371 	int unit, part;
372 	int error;
373 
374 	unit = SDUNIT(dev);
375 	if (unit >= sd_cd.cd_ndevs)
376 		return (ENXIO);
377 	sd = sd_cd.cd_devs[unit];
378 	if (sd == NULL)
379 		return (ENXIO);
380 
381 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
382 		return (ENODEV);
383 
384 	periph = sd->sc_periph;
385 	adapt = periph->periph_channel->chan_adapter;
386 	part = SDPART(dev);
387 
388 	SC_DEBUG(periph, SCSIPI_DB1,
389 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
390 	    sd_cd.cd_ndevs, part));
391 
392 	/*
393 	 * If this is the first open of this device, add a reference
394 	 * to the adapter.
395 	 */
396 	if (sd->sc_dk.dk_openmask == 0 &&
397 	    (error = scsipi_adapter_addref(adapt)) != 0)
398 		return (error);
399 
400 	if ((error = sdlock(sd)) != 0)
401 		goto bad4;
402 
403 	if ((periph->periph_flags & PERIPH_OPEN) != 0) {
404 		/*
405 		 * If any partition is open, but the disk has been invalidated,
406 		 * disallow further opens of non-raw partition
407 		 */
408 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 &&
409 		    (part != RAW_PART || fmt != S_IFCHR)) {
410 			error = EIO;
411 			goto bad3;
412 		}
413 	} else {
414 		/* Check that it is still responding and ok. */
415 		error = scsipi_test_unit_ready(periph,
416 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE |
417 		    XS_CTL_IGNORE_NOT_READY);
418 		if (error)
419 			goto bad3;
420 
421 		/*
422 		 * Start the pack spinning if necessary. Always allow the
423 		 * raw parition to be opened, for raw IOCTLs. Data transfers
424 		 * will check for SDEV_MEDIA_LOADED.
425 		 */
426 		error = scsipi_start(periph, SSS_START,
427 		    XS_CTL_IGNORE_ILLEGAL_REQUEST |
428 		    XS_CTL_IGNORE_MEDIA_CHANGE | XS_CTL_SILENT);
429 		if (error) {
430 			if (part != RAW_PART || fmt != S_IFCHR)
431 				goto bad3;
432 			else
433 				goto out;
434 		}
435 
436 		periph->periph_flags |= PERIPH_OPEN;
437 
438 		if (periph->periph_flags & PERIPH_REMOVABLE) {
439 			/* Lock the pack in. */
440 			error = scsipi_prevent(periph, PR_PREVENT,
441 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
442 			    XS_CTL_IGNORE_MEDIA_CHANGE);
443 			if (error)
444 				goto bad;
445 		}
446 
447 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
448 			periph->periph_flags |= PERIPH_MEDIA_LOADED;
449 
450 			/*
451 			 * Load the physical device parameters.
452 			 *
453 			 * Note that if media is present but unformatted,
454 			 * we allow the open (so that it can be formatted!).
455 			 * The drive should refuse real I/O, if the media is
456 			 * unformatted.
457 			 */
458 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
459 			    0) == SDGP_RESULT_OFFLINE) {
460 				error = ENXIO;
461 				goto bad2;
462 			}
463 			SC_DEBUG(periph, SCSIPI_DB3, ("Params loaded "));
464 
465 			/* Load the partition info if not already loaded. */
466 			sdgetdisklabel(sd);
467 			SC_DEBUG(periph, SCSIPI_DB3, ("Disklabel loaded "));
468 		}
469 	}
470 
471 	/* Check that the partition exists. */
472 	if (part != RAW_PART &&
473 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
474 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
475 		error = ENXIO;
476 		goto bad;
477 	}
478 
479 out:	/* Insure only one open at a time. */
480 	switch (fmt) {
481 	case S_IFCHR:
482 		sd->sc_dk.dk_copenmask |= (1 << part);
483 		break;
484 	case S_IFBLK:
485 		sd->sc_dk.dk_bopenmask |= (1 << part);
486 		break;
487 	}
488 	sd->sc_dk.dk_openmask =
489 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
490 
491 	SC_DEBUG(periph, SCSIPI_DB3, ("open complete\n"));
492 	sdunlock(sd);
493 	return (0);
494 
495 bad2:
496 	periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
497 
498 bad:
499 	if (sd->sc_dk.dk_openmask == 0) {
500 		scsipi_prevent(periph, PR_ALLOW,
501 		    XS_CTL_IGNORE_ILLEGAL_REQUEST | XS_CTL_IGNORE_MEDIA_CHANGE);
502 		periph->periph_flags &= ~PERIPH_OPEN;
503 	}
504 
505 bad3:
506 	sdunlock(sd);
507 bad4:
508 	if (sd->sc_dk.dk_openmask == 0)
509 		scsipi_adapter_delref(adapt);
510 	return (error);
511 }
512 
513 /*
514  * close the device.. only called if we are the LAST occurence of an open
515  * device.  Convenient now but usually a pain.
516  */
517 int
518 sdclose(dev, flag, fmt, p)
519 	dev_t dev;
520 	int flag, fmt;
521 	struct proc *p;
522 {
523 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
524 	struct scsipi_periph *periph = sd->sc_periph;
525 	struct scsipi_adapter *adapt = periph->periph_channel->chan_adapter;
526 	int part = SDPART(dev);
527 	int error;
528 
529 	if ((error = sdlock(sd)) != 0)
530 		return (error);
531 
532 	switch (fmt) {
533 	case S_IFCHR:
534 		sd->sc_dk.dk_copenmask &= ~(1 << part);
535 		break;
536 	case S_IFBLK:
537 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
538 		break;
539 	}
540 	sd->sc_dk.dk_openmask =
541 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
542 
543 	if (sd->sc_dk.dk_openmask == 0) {
544 		/*
545 		 * If the disk cache needs flushing, and the disk supports
546 		 * it, do it now.
547 		 */
548 		if ((sd->flags & SDF_DIRTY) != 0 &&
549 		    sd->sc_ops->sdo_flush != NULL) {
550 			if ((*sd->sc_ops->sdo_flush)(sd, 0)) {
551 				printf("%s: cache synchronization failed\n",
552 				    sd->sc_dev.dv_xname);
553 				sd->flags &= ~SDF_FLUSHING;
554 			} else
555 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
556 		}
557 
558 		if (! (periph->periph_flags & PERIPH_KEEP_LABEL))
559 			periph->periph_flags &= ~PERIPH_MEDIA_LOADED;
560 
561 		scsipi_wait_drain(periph);
562 
563 		if (periph->periph_flags & PERIPH_REMOVABLE) {
564 			scsipi_prevent(periph, PR_ALLOW,
565 			    XS_CTL_IGNORE_ILLEGAL_REQUEST |
566 			    XS_CTL_IGNORE_NOT_READY);
567 		}
568 		periph->periph_flags &= ~PERIPH_OPEN;
569 
570 		scsipi_wait_drain(periph);
571 
572 		scsipi_adapter_delref(adapt);
573 	}
574 
575 	sdunlock(sd);
576 	return (0);
577 }
578 
579 /*
580  * Actually translate the requested transfer into one the physical driver
581  * can understand.  The transfer is described by a buf and will include
582  * only one physical transfer.
583  */
584 void
585 sdstrategy(bp)
586 	struct buf *bp;
587 {
588 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
589 	struct scsipi_periph *periph = sd->sc_periph;
590 	struct disklabel *lp;
591 	daddr_t blkno;
592 	int s;
593 	boolean_t sector_aligned;
594 
595 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdstrategy "));
596 	SC_DEBUG(sd->sc_periph, SCSIPI_DB1,
597 	    ("%ld bytes @ blk %" PRId64 "\n", bp->b_bcount, bp->b_blkno));
598 	/*
599 	 * If the device has been made invalid, error out
600 	 */
601 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0 ||
602 	    (sd->sc_dev.dv_flags & DVF_ACTIVE) == 0) {
603 		if (periph->periph_flags & PERIPH_OPEN)
604 			bp->b_error = EIO;
605 		else
606 			bp->b_error = ENODEV;
607 		goto bad;
608 	}
609 
610 	lp = sd->sc_dk.dk_label;
611 
612 	/*
613 	 * The transfer must be a whole number of blocks, offset must not be
614 	 * negative.
615 	 */
616 	if (lp->d_secsize == DEV_BSIZE) {
617 		sector_aligned = (bp->b_bcount & (DEV_BSIZE - 1)) == 0;
618 	} else {
619 		sector_aligned = (bp->b_bcount % lp->d_secsize) == 0;
620 	}
621 	if (!sector_aligned || bp->b_blkno < 0) {
622 		bp->b_error = EINVAL;
623 		goto bad;
624 	}
625 	/*
626 	 * If it's a null transfer, return immediatly
627 	 */
628 	if (bp->b_bcount == 0)
629 		goto done;
630 
631 	/*
632 	 * Do bounds checking, adjust transfer. if error, process.
633 	 * If end of partition, just return.
634 	 */
635 	if (SDPART(bp->b_dev) == RAW_PART) {
636 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
637 		    sd->params.disksize512) <= 0)
638 			goto done;
639 	} else {
640 		if (bounds_check_with_label(&sd->sc_dk, bp,
641 		    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
642 			goto done;
643 	}
644 
645 	/*
646 	 * Now convert the block number to absolute and put it in
647 	 * terms of the device's logical block size.
648 	 */
649 	if (lp->d_secsize == DEV_BSIZE)
650 		blkno = bp->b_blkno;
651 	else if (lp->d_secsize > DEV_BSIZE)
652 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
653 	else
654 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
655 
656 	if (SDPART(bp->b_dev) != RAW_PART)
657 		blkno += lp->d_partitions[SDPART(bp->b_dev)].p_offset;
658 
659 	bp->b_rawblkno = blkno;
660 
661 	s = splbio();
662 
663 	/*
664 	 * Place it in the queue of disk activities for this disk.
665 	 *
666 	 * XXX Only do disksort() if the current operating mode does not
667 	 * XXX include tagged queueing.
668 	 */
669 	BUFQ_PUT(&sd->buf_queue, bp);
670 
671 	/*
672 	 * Tell the device to get going on the transfer if it's
673 	 * not doing anything, otherwise just wait for completion
674 	 */
675 	sdstart(sd->sc_periph);
676 
677 	splx(s);
678 	return;
679 
680 bad:
681 	bp->b_flags |= B_ERROR;
682 done:
683 	/*
684 	 * Correctly set the buf to indicate a completed xfer
685 	 */
686 	bp->b_resid = bp->b_bcount;
687 	biodone(bp);
688 }
689 
690 /*
691  * sdstart looks to see if there is a buf waiting for the device
692  * and that the device is not already busy. If both are true,
693  * It dequeues the buf and creates a scsi command to perform the
694  * transfer in the buf. The transfer request will call scsipi_done
695  * on completion, which will in turn call this routine again
696  * so that the next queued transfer is performed.
697  * The bufs are queued by the strategy routine (sdstrategy)
698  *
699  * This routine is also called after other non-queued requests
700  * have been made of the scsi driver, to ensure that the queue
701  * continues to be drained.
702  *
703  * must be called at the correct (highish) spl level
704  * sdstart() is called at splbio from sdstrategy and scsipi_done
705  */
706 void
707 sdstart(periph)
708 	struct scsipi_periph *periph;
709 {
710 	struct sd_softc *sd = (void *)periph->periph_dev;
711 	struct disklabel *lp = sd->sc_dk.dk_label;
712 	struct buf *bp = 0;
713 	struct scsipi_rw_big cmd_big;
714 #if NSD_SCSIBUS > 0
715 	struct scsi_rw cmd_small;
716 #endif
717 	struct scsipi_generic *cmdp;
718 	int nblks, cmdlen, error, flags;
719 
720 	SC_DEBUG(periph, SCSIPI_DB2, ("sdstart "));
721 	/*
722 	 * Check if the device has room for another command
723 	 */
724 	while (periph->periph_active < periph->periph_openings) {
725 		/*
726 		 * there is excess capacity, but a special waits
727 		 * It'll need the adapter as soon as we clear out of the
728 		 * way and let it run (user level wait).
729 		 */
730 		if (periph->periph_flags & PERIPH_WAITING) {
731 			periph->periph_flags &= ~PERIPH_WAITING;
732 			wakeup((caddr_t)periph);
733 			return;
734 		}
735 
736 		/*
737 		 * See if there is a buf with work for us to do..
738 		 */
739 		if ((bp = BUFQ_GET(&sd->buf_queue)) == NULL)
740 			return;
741 
742 		/*
743 		 * If the device has become invalid, abort all the
744 		 * reads and writes until all files have been closed and
745 		 * re-opened
746 		 */
747 		if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
748 			bp->b_error = EIO;
749 			bp->b_flags |= B_ERROR;
750 			bp->b_resid = bp->b_bcount;
751 			biodone(bp);
752 			continue;
753 		}
754 
755 		/*
756 		 * We have a buf, now we should make a command.
757 		 */
758 
759 		if (lp->d_secsize == DEV_BSIZE)
760 			nblks = bp->b_bcount >> DEV_BSHIFT;
761 		else
762 			nblks = howmany(bp->b_bcount, lp->d_secsize);
763 
764 #if NSD_SCSIBUS > 0
765 		/*
766 		 *  Fill out the scsi command.  If the transfer will
767 		 *  fit in a "small" cdb, use it.
768 		 */
769 		if (((bp->b_rawblkno & 0x1fffff) == bp->b_rawblkno) &&
770 		    ((nblks & 0xff) == nblks) &&
771 		    !(periph->periph_quirks & PQUIRK_ONLYBIG) &&
772 		    scsipi_periph_bustype(periph) == SCSIPI_BUSTYPE_SCSI) {
773 			/*
774 			 * We can fit in a small cdb.
775 			 */
776 			memset(&cmd_small, 0, sizeof(cmd_small));
777 			cmd_small.opcode = (bp->b_flags & B_READ) ?
778 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
779 			_lto3b(bp->b_rawblkno, cmd_small.addr);
780 			cmd_small.length = nblks & 0xff;
781 			cmdlen = sizeof(cmd_small);
782 			cmdp = (struct scsipi_generic *)&cmd_small;
783 		} else
784 #endif /* NSD_SCSIBUS > 0 */
785 		{
786 			/*
787 			 * Need a large cdb.
788 			 */
789 			memset(&cmd_big, 0, sizeof(cmd_big));
790 			cmd_big.opcode = (bp->b_flags & B_READ) ?
791 			    READ_BIG : WRITE_BIG;
792 			_lto4b(bp->b_rawblkno, cmd_big.addr);
793 			_lto2b(nblks, cmd_big.length);
794 			cmdlen = sizeof(cmd_big);
795 			cmdp = (struct scsipi_generic *)&cmd_big;
796 		}
797 
798 		/* Instrumentation. */
799 		disk_busy(&sd->sc_dk);
800 
801 		/*
802 		 * Mark the disk dirty so that the cache will be
803 		 * flushed on close.
804 		 */
805 		if ((bp->b_flags & B_READ) == 0)
806 			sd->flags |= SDF_DIRTY;
807 
808 		/*
809 		 * Figure out what flags to use.
810 		 */
811 		flags = XS_CTL_NOSLEEP|XS_CTL_ASYNC|XS_CTL_SIMPLE_TAG;
812 		if (bp->b_flags & B_READ)
813 			flags |= XS_CTL_DATA_IN;
814 		else
815 			flags |= XS_CTL_DATA_OUT;
816 
817 		/*
818 		 * Call the routine that chats with the adapter.
819 		 * Note: we cannot sleep as we may be an interrupt
820 		 */
821 		error = scsipi_command(periph, cmdp, cmdlen,
822 		    (u_char *)bp->b_data, bp->b_bcount,
823 		    SDRETRIES, SD_IO_TIMEOUT, bp, flags);
824 		if (error) {
825 			disk_unbusy(&sd->sc_dk, 0, 0);
826 			printf("%s: not queued, error %d\n",
827 			    sd->sc_dev.dv_xname, error);
828 		}
829 	}
830 }
831 
832 void
833 sddone(xs)
834 	struct scsipi_xfer *xs;
835 {
836 	struct sd_softc *sd = (void *)xs->xs_periph->periph_dev;
837 
838 	if (sd->flags & SDF_FLUSHING) {
839 		/* Flush completed, no longer dirty. */
840 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
841 	}
842 
843 	if (xs->bp != NULL) {
844 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid,
845 		    (xs->bp->b_flags & B_READ));
846 #if NRND > 0
847 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_rawblkno);
848 #endif
849 	}
850 }
851 
852 void
853 sdminphys(bp)
854 	struct buf *bp;
855 {
856 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
857 	long max;
858 
859 	/*
860 	 * If the device is ancient, we want to make sure that
861 	 * the transfer fits into a 6-byte cdb.
862 	 *
863 	 * XXX Note that the SCSI-I spec says that 256-block transfers
864 	 * are allowed in a 6-byte read/write, and are specified
865 	 * by settng the "length" to 0.  However, we're conservative
866 	 * here, allowing only 255-block transfers in case an
867 	 * ancient device gets confused by length == 0.  A length of 0
868 	 * in a 10-byte read/write actually means 0 blocks.
869 	 */
870 	if ((sd->flags & SDF_ANCIENT) &&
871 	    ((sd->sc_periph->periph_flags &
872 	    (PERIPH_REMOVABLE | PERIPH_MEDIA_LOADED)) != PERIPH_REMOVABLE)) {
873 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
874 
875 		if (bp->b_bcount > max)
876 			bp->b_bcount = max;
877 	}
878 
879 	(*sd->sc_periph->periph_channel->chan_adapter->adapt_minphys)(bp);
880 }
881 
882 int
883 sdread(dev, uio, ioflag)
884 	dev_t dev;
885 	struct uio *uio;
886 	int ioflag;
887 {
888 
889 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
890 }
891 
892 int
893 sdwrite(dev, uio, ioflag)
894 	dev_t dev;
895 	struct uio *uio;
896 	int ioflag;
897 {
898 
899 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
900 }
901 
902 /*
903  * Perform special action on behalf of the user
904  * Knows about the internals of this device
905  */
906 int
907 sdioctl(dev, cmd, addr, flag, p)
908 	dev_t dev;
909 	u_long cmd;
910 	caddr_t addr;
911 	int flag;
912 	struct proc *p;
913 {
914 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
915 	struct scsipi_periph *periph = sd->sc_periph;
916 	int part = SDPART(dev);
917 	int error = 0;
918 #ifdef __HAVE_OLD_DISKLABEL
919 	struct disklabel *newlabel = NULL;
920 #endif
921 
922 	SC_DEBUG(sd->sc_periph, SCSIPI_DB2, ("sdioctl 0x%lx ", cmd));
923 
924 	/*
925 	 * If the device is not valid, some IOCTLs can still be
926 	 * handled on the raw partition. Check this here.
927 	 */
928 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0) {
929 		switch (cmd) {
930 		case DIOCKLABEL:
931 		case DIOCWLABEL:
932 		case DIOCLOCK:
933 		case DIOCEJECT:
934 		case ODIOCEJECT:
935 		case DIOCGCACHE:
936 		case DIOCSCACHE:
937 		case SCIOCIDENTIFY:
938 		case OSCIOCIDENTIFY:
939 		case SCIOCCOMMAND:
940 		case SCIOCDEBUG:
941 			if (part == RAW_PART)
942 				break;
943 		/* FALLTHROUGH */
944 		default:
945 			if ((periph->periph_flags & PERIPH_OPEN) == 0)
946 				return (ENODEV);
947 			else
948 				return (EIO);
949 		}
950 	}
951 
952 	switch (cmd) {
953 	case DIOCGDINFO:
954 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
955 		return (0);
956 
957 #ifdef __HAVE_OLD_DISKLABEL
958 	case ODIOCGDINFO:
959 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
960 		if (newlabel == NULL)
961 			return EIO;
962 		memcpy(newlabel, sd->sc_dk.dk_label, sizeof (*newlabel));
963 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
964 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
965 		else
966 			error = ENOTTY;
967 		free(newlabel, M_TEMP);
968 		return error;
969 #endif
970 
971 	case DIOCGPART:
972 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
973 		((struct partinfo *)addr)->part =
974 		    &sd->sc_dk.dk_label->d_partitions[part];
975 		return (0);
976 
977 	case DIOCWDINFO:
978 	case DIOCSDINFO:
979 #ifdef __HAVE_OLD_DISKLABEL
980 	case ODIOCWDINFO:
981 	case ODIOCSDINFO:
982 #endif
983 	{
984 		struct disklabel *lp;
985 
986 		if ((flag & FWRITE) == 0)
987 			return (EBADF);
988 
989 #ifdef __HAVE_OLD_DISKLABEL
990  		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
991 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
992 			if (newlabel == NULL)
993 				return EIO;
994 			memset(newlabel, 0, sizeof newlabel);
995 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
996 			lp = newlabel;
997 		} else
998 #endif
999 		lp = (struct disklabel *)addr;
1000 
1001 		if ((error = sdlock(sd)) != 0)
1002 			goto bad;
1003 		sd->flags |= SDF_LABELLING;
1004 
1005 		error = setdisklabel(sd->sc_dk.dk_label,
1006 		    lp, /*sd->sc_dk.dk_openmask : */0,
1007 		    sd->sc_dk.dk_cpulabel);
1008 		if (error == 0) {
1009 			if (cmd == DIOCWDINFO
1010 #ifdef __HAVE_OLD_DISKLABEL
1011 			    || cmd == ODIOCWDINFO
1012 #endif
1013 			   )
1014 				error = writedisklabel(SDLABELDEV(dev),
1015 				    sdstrategy, sd->sc_dk.dk_label,
1016 				    sd->sc_dk.dk_cpulabel);
1017 		}
1018 
1019 		sd->flags &= ~SDF_LABELLING;
1020 		sdunlock(sd);
1021 bad:
1022 #ifdef __HAVE_OLD_DISKLABEL
1023 		if (newlabel != NULL)
1024 			free(newlabel, M_TEMP);
1025 #endif
1026 		return (error);
1027 	}
1028 
1029 	case DIOCKLABEL:
1030 		if (*(int *)addr)
1031 			periph->periph_flags |= PERIPH_KEEP_LABEL;
1032 		else
1033 			periph->periph_flags &= ~PERIPH_KEEP_LABEL;
1034 		return (0);
1035 
1036 	case DIOCWLABEL:
1037 		if ((flag & FWRITE) == 0)
1038 			return (EBADF);
1039 		if (*(int *)addr)
1040 			sd->flags |= SDF_WLABEL;
1041 		else
1042 			sd->flags &= ~SDF_WLABEL;
1043 		return (0);
1044 
1045 	case DIOCLOCK:
1046 		return (scsipi_prevent(periph,
1047 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
1048 
1049 	case DIOCEJECT:
1050 		if ((periph->periph_flags & PERIPH_REMOVABLE) == 0)
1051 			return (ENOTTY);
1052 		if (*(int *)addr == 0) {
1053 			/*
1054 			 * Don't force eject: check that we are the only
1055 			 * partition open. If so, unlock it.
1056 			 */
1057 			if ((sd->sc_dk.dk_openmask & ~(1 << part)) == 0 &&
1058 			    sd->sc_dk.dk_bopenmask + sd->sc_dk.dk_copenmask ==
1059 			    sd->sc_dk.dk_openmask) {
1060 				error = scsipi_prevent(periph, PR_ALLOW,
1061 				    XS_CTL_IGNORE_NOT_READY);
1062 				if (error)
1063 					return (error);
1064 			} else {
1065 				return (EBUSY);
1066 			}
1067 		}
1068 		/* FALLTHROUGH */
1069 	case ODIOCEJECT:
1070 		return ((periph->periph_flags & PERIPH_REMOVABLE) == 0 ?
1071 		    ENOTTY : scsipi_start(periph, SSS_STOP|SSS_LOEJ, 0));
1072 
1073 	case DIOCGDEFLABEL:
1074 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
1075 		return (0);
1076 
1077 #ifdef __HAVE_OLD_DISKLABEL
1078 	case ODIOCGDEFLABEL:
1079 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1080 		if (newlabel == NULL)
1081 			return EIO;
1082 		sdgetdefaultlabel(sd, newlabel);
1083 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1084 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1085 		else
1086 			error = ENOTTY;
1087 		free(newlabel, M_TEMP);
1088 		return error;
1089 #endif
1090 
1091 	case DIOCGCACHE:
1092 		if (sd->sc_ops->sdo_getcache != NULL)
1093 			return ((*sd->sc_ops->sdo_getcache)(sd, (int *) addr));
1094 
1095 		/* Not supported on this device. */
1096 		*(int *) addr = 0;
1097 		return (0);
1098 
1099 	case DIOCSCACHE:
1100 		if ((flag & FWRITE) == 0)
1101 			return (EBADF);
1102 		if (sd->sc_ops->sdo_setcache != NULL)
1103 			return ((*sd->sc_ops->sdo_setcache)(sd, *(int *) addr));
1104 
1105 		/* Not supported on this device. */
1106 		return (EOPNOTSUPP);
1107 
1108 	case DIOCCACHESYNC:
1109 		/*
1110 		 * XXX Do we really need to care about having a writable
1111 		 * file descriptor here?
1112 		 */
1113 		if ((flag & FWRITE) == 0)
1114 			return (EBADF);
1115 		if (((sd->flags & SDF_DIRTY) != 0 || *(int *)addr != 0) &&
1116 		    sd->sc_ops->sdo_flush != NULL) {
1117 			error = (*sd->sc_ops->sdo_flush)(sd, 0);
1118 			if (error)
1119 				sd->flags &= ~SDF_FLUSHING;
1120 			else
1121 				sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1122 		} else
1123 			error = 0;
1124 		return (error);
1125 
1126 	default:
1127 		if (part != RAW_PART)
1128 			return (ENOTTY);
1129 		return (scsipi_do_ioctl(periph, dev, cmd, addr, flag, p));
1130 	}
1131 
1132 #ifdef DIAGNOSTIC
1133 	panic("sdioctl: impossible");
1134 #endif
1135 }
1136 
1137 void
1138 sdgetdefaultlabel(sd, lp)
1139 	struct sd_softc *sd;
1140 	struct disklabel *lp;
1141 {
1142 
1143 	memset(lp, 0, sizeof(struct disklabel));
1144 
1145 	lp->d_secsize = sd->params.blksize;
1146 	lp->d_ntracks = sd->params.heads;
1147 	lp->d_nsectors = sd->params.sectors;
1148 	lp->d_ncylinders = sd->params.cyls;
1149 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1150 
1151 	switch (scsipi_periph_bustype(sd->sc_periph)) {
1152 #if NSD_SCSIBUS > 0
1153 	case SCSIPI_BUSTYPE_SCSI:
1154 		lp->d_type = DTYPE_SCSI;
1155 		break;
1156 #endif
1157 #if NSD_ATAPIBUS > 0
1158 	case SCSIPI_BUSTYPE_ATAPI:
1159 		lp->d_type = DTYPE_ATAPI;
1160 		break;
1161 #endif
1162 	}
1163 	strncpy(lp->d_typename, sd->name, 16);
1164 	strncpy(lp->d_packname, "fictitious", 16);
1165 	lp->d_secperunit = sd->params.disksize;
1166 	lp->d_rpm = sd->params.rot_rate;
1167 	lp->d_interleave = 1;
1168 	lp->d_flags = 0;
1169 
1170 	lp->d_partitions[RAW_PART].p_offset = 0;
1171 	lp->d_partitions[RAW_PART].p_size =
1172 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1173 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1174 	lp->d_npartitions = RAW_PART + 1;
1175 
1176 	lp->d_magic = DISKMAGIC;
1177 	lp->d_magic2 = DISKMAGIC;
1178 	lp->d_checksum = dkcksum(lp);
1179 }
1180 
1181 
1182 /*
1183  * Load the label information on the named device
1184  */
1185 void
1186 sdgetdisklabel(sd)
1187 	struct sd_softc *sd;
1188 {
1189 	struct disklabel *lp = sd->sc_dk.dk_label;
1190 	const char *errstring;
1191 
1192 	memset(sd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1193 
1194 	sdgetdefaultlabel(sd, lp);
1195 
1196 	if (lp->d_secpercyl == 0) {
1197 		lp->d_secpercyl = 100;
1198 		/* as long as it's not 0 - readdisklabel divides by it (?) */
1199 	}
1200 
1201 	/*
1202 	 * Call the generic disklabel extraction routine
1203 	 */
1204 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
1205 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
1206 	if (errstring) {
1207 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
1208 		return;
1209 	}
1210 }
1211 
1212 void
1213 sd_shutdown(arg)
1214 	void *arg;
1215 {
1216 	struct sd_softc *sd = arg;
1217 
1218 	/*
1219 	 * If the disk cache needs to be flushed, and the disk supports
1220 	 * it, flush it.  We're cold at this point, so we poll for
1221 	 * completion.
1222 	 */
1223 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL) {
1224 		if ((*sd->sc_ops->sdo_flush)(sd, XS_CTL_NOSLEEP|XS_CTL_POLL)) {
1225 			printf("%s: cache synchronization failed\n",
1226 			    sd->sc_dev.dv_xname);
1227 			sd->flags &= ~SDF_FLUSHING;
1228 		} else
1229 			sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
1230 	}
1231 }
1232 
1233 /*
1234  * Tell the device to map out a defective block
1235  */
1236 int
1237 sd_reassign_blocks(sd, blkno)
1238 	struct sd_softc *sd;
1239 	u_long blkno;
1240 {
1241 	struct scsi_reassign_blocks scsipi_cmd;
1242 	struct scsi_reassign_blocks_data rbdata;
1243 
1244 	memset(&scsipi_cmd, 0, sizeof(scsipi_cmd));
1245 	memset(&rbdata, 0, sizeof(rbdata));
1246 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
1247 
1248 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
1249 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
1250 
1251 	return (scsipi_command(sd->sc_periph,
1252 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
1253 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
1254 	    XS_CTL_DATA_OUT | XS_CTL_DATA_ONSTACK));
1255 }
1256 
1257 /*
1258  * Check Errors
1259  */
1260 int
1261 sd_interpret_sense(xs)
1262 	struct scsipi_xfer *xs;
1263 {
1264 	struct scsipi_periph *periph = xs->xs_periph;
1265 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
1266 	struct sd_softc *sd = (void *)periph->periph_dev;
1267 	int s, error, retval = EJUSTRETURN;
1268 
1269 	/*
1270 	 * If the periph is already recovering, just do the normal
1271 	 * error processing.
1272 	 */
1273 	if (periph->periph_flags & PERIPH_RECOVERING)
1274 		return (retval);
1275 
1276 	/*
1277 	 * If the device is not open yet, let the generic code handle it.
1278 	 */
1279 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1280 		return (retval);
1281 
1282 	/*
1283 	 * If it isn't a extended or extended/deferred error, let
1284 	 * the generic code handle it.
1285 	 */
1286 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
1287 	    (sense->error_code & SSD_ERRCODE) != 0x71)
1288 		return (retval);
1289 
1290 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
1291 	    sense->add_sense_code == 0x4) {
1292 		if (sense->add_sense_code_qual == 0x01)	{
1293 			/*
1294 			 * Unit In The Process Of Becoming Ready.
1295 			 */
1296 			printf("%s: waiting for pack to spin up...\n",
1297 			    sd->sc_dev.dv_xname);
1298 			if (!callout_pending(&periph->periph_callout))
1299 				scsipi_periph_freeze(periph, 1);
1300 			callout_reset(&periph->periph_callout,
1301 			    5 * hz, scsipi_periph_timed_thaw, periph);
1302 			retval = ERESTART;
1303 		} else if ((sense->add_sense_code_qual == 0x2) &&
1304 		    (periph->periph_quirks & PQUIRK_NOSTARTUNIT) == 0) {
1305 			printf("%s: pack is stopped, restarting...\n",
1306 			    sd->sc_dev.dv_xname);
1307 			s = splbio();
1308 			periph->periph_flags |= PERIPH_RECOVERING;
1309 			splx(s);
1310 			error = scsipi_start(periph, SSS_START,
1311 			    XS_CTL_URGENT|XS_CTL_HEAD_TAG|
1312 			    XS_CTL_THAW_PERIPH|XS_CTL_FREEZE_PERIPH);
1313 			if (error) {
1314 				printf("%s: unable to restart pack\n",
1315 				    sd->sc_dev.dv_xname);
1316 				retval = error;
1317 			} else
1318 				retval = ERESTART;
1319 			s = splbio();
1320 			periph->periph_flags &= ~PERIPH_RECOVERING;
1321 			splx(s);
1322 		}
1323 	}
1324 	return (retval);
1325 }
1326 
1327 
1328 int
1329 sdsize(dev)
1330 	dev_t dev;
1331 {
1332 	struct sd_softc *sd;
1333 	int part, unit, omask;
1334 	int size;
1335 
1336 	unit = SDUNIT(dev);
1337 	if (unit >= sd_cd.cd_ndevs)
1338 		return (-1);
1339 	sd = sd_cd.cd_devs[unit];
1340 	if (sd == NULL)
1341 		return (-1);
1342 
1343 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1344 		return (-1);
1345 
1346 	part = SDPART(dev);
1347 	omask = sd->sc_dk.dk_openmask & (1 << part);
1348 
1349 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1350 		return (-1);
1351 	if ((sd->sc_periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1352 		size = -1;
1353 	else if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1354 		size = -1;
1355 	else
1356 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1357 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1358 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1359 		return (-1);
1360 	return (size);
1361 }
1362 
1363 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1364 static struct scsipi_xfer sx;
1365 static int sddoingadump;
1366 
1367 /*
1368  * dump all of physical memory into the partition specified, starting
1369  * at offset 'dumplo' into the partition.
1370  */
1371 int
1372 sddump(dev, blkno, va, size)
1373 	dev_t dev;
1374 	daddr_t blkno;
1375 	caddr_t va;
1376 	size_t size;
1377 {
1378 	struct sd_softc *sd;	/* disk unit to do the I/O */
1379 	struct disklabel *lp;	/* disk's disklabel */
1380 	int	unit, part;
1381 	int	sectorsize;	/* size of a disk sector */
1382 	int	nsects;		/* number of sectors in partition */
1383 	int	sectoff;	/* sector offset of partition */
1384 	int	totwrt;		/* total number of sectors left to write */
1385 	int	nwrt;		/* current number of sectors to write */
1386 	struct scsipi_rw_big cmd;	/* write command */
1387 	struct scsipi_xfer *xs;	/* ... convenience */
1388 	struct scsipi_periph *periph;
1389 	struct scsipi_channel *chan;
1390 
1391 	/* Check if recursive dump; if so, punt. */
1392 	if (sddoingadump)
1393 		return (EFAULT);
1394 
1395 	/* Mark as active early. */
1396 	sddoingadump = 1;
1397 
1398 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1399 	part = SDPART(dev);
1400 
1401 	/* Check for acceptable drive number. */
1402 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1403 		return (ENXIO);
1404 
1405 	if ((sd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
1406 		return (ENODEV);
1407 
1408 	periph = sd->sc_periph;
1409 	chan = periph->periph_channel;
1410 
1411 	/* Make sure it was initialized. */
1412 	if ((periph->periph_flags & PERIPH_MEDIA_LOADED) == 0)
1413 		return (ENXIO);
1414 
1415 	/* Convert to disk sectors.  Request must be a multiple of size. */
1416 	lp = sd->sc_dk.dk_label;
1417 	sectorsize = lp->d_secsize;
1418 	if ((size % sectorsize) != 0)
1419 		return (EFAULT);
1420 	totwrt = size / sectorsize;
1421 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1422 
1423 	nsects = lp->d_partitions[part].p_size;
1424 	sectoff = lp->d_partitions[part].p_offset;
1425 
1426 	/* Check transfer bounds against partition size. */
1427 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1428 		return (EINVAL);
1429 
1430 	/* Offset block number to start of partition. */
1431 	blkno += sectoff;
1432 
1433 	xs = &sx;
1434 
1435 	while (totwrt > 0) {
1436 		nwrt = totwrt;		/* XXX */
1437 #ifndef	SD_DUMP_NOT_TRUSTED
1438 		/*
1439 		 *  Fill out the scsi command
1440 		 */
1441 		memset(&cmd, 0, sizeof(cmd));
1442 		cmd.opcode = WRITE_BIG;
1443 		_lto4b(blkno, cmd.addr);
1444 		_lto2b(nwrt, cmd.length);
1445 		/*
1446 		 * Fill out the scsipi_xfer structure
1447 		 *    Note: we cannot sleep as we may be an interrupt
1448 		 * don't use scsipi_command() as it may want to wait
1449 		 * for an xs.
1450 		 */
1451 		memset(xs, 0, sizeof(sx));
1452 		xs->xs_control |= XS_CTL_NOSLEEP | XS_CTL_POLL |
1453 		    XS_CTL_DATA_OUT;
1454 		xs->xs_status = 0;
1455 		xs->xs_periph = periph;
1456 		xs->xs_retries = SDRETRIES;
1457 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1458 		xs->cmd = (struct scsipi_generic *)&cmd;
1459 		xs->cmdlen = sizeof(cmd);
1460 		xs->resid = nwrt * sectorsize;
1461 		xs->error = XS_NOERROR;
1462 		xs->bp = 0;
1463 		xs->data = va;
1464 		xs->datalen = nwrt * sectorsize;
1465 
1466 		/*
1467 		 * Pass all this info to the scsi driver.
1468 		 */
1469 		scsipi_adapter_request(chan, ADAPTER_REQ_RUN_XFER, xs);
1470 		if ((xs->xs_status & XS_STS_DONE) == 0 ||
1471 		    xs->error != XS_NOERROR)
1472 			return (EIO);
1473 #else	/* SD_DUMP_NOT_TRUSTED */
1474 		/* Let's just talk about this first... */
1475 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1476 		delay(500 * 1000);	/* half a second */
1477 #endif	/* SD_DUMP_NOT_TRUSTED */
1478 
1479 		/* update block count */
1480 		totwrt -= nwrt;
1481 		blkno += nwrt;
1482 		va += sectorsize * nwrt;
1483 	}
1484 	sddoingadump = 0;
1485 	return (0);
1486 }
1487