xref: /netbsd-src/sys/dev/scsipi/sd.c (revision 93f9db1b75d415b78f73ed629beeb86235153473)
1 /*	$NetBSD: sd.c,v 1.135 1998/08/17 19:30:38 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/buf.h>
67 #include <sys/uio.h>
68 #include <sys/malloc.h>
69 #include <sys/errno.h>
70 #include <sys/device.h>
71 #include <sys/disklabel.h>
72 #include <sys/disk.h>
73 #include <sys/proc.h>
74 #include <sys/conf.h>
75 #if NRND > 0
76 #include <sys/rnd.h>
77 #endif
78 
79 #include <dev/scsipi/scsipi_all.h>
80 #include <dev/scsipi/scsi_all.h>
81 #include <dev/scsipi/scsipi_disk.h>
82 #include <dev/scsipi/scsi_disk.h>
83 #include <dev/scsipi/scsiconf.h>
84 #include <dev/scsipi/sdvar.h>
85 
86 #include "sd.h"		/* NSD_SCSIBUS and NSD_ATAPIBUS come from here */
87 
88 #ifndef	SDOUTSTANDING
89 #define	SDOUTSTANDING	4
90 #endif
91 
92 #define	SDUNIT(dev)			DISKUNIT(dev)
93 #define	SDPART(dev)			DISKPART(dev)
94 #define	MAKESDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
95 
96 #define	SDLABELDEV(dev)	(MAKESDDEV(major(dev), SDUNIT(dev), RAW_PART))
97 
98 int	sdlock __P((struct sd_softc *));
99 void	sdunlock __P((struct sd_softc *));
100 void	sdminphys __P((struct buf *));
101 void	sdgetdefaultlabel __P((struct sd_softc *, struct disklabel *));
102 void	sdgetdisklabel __P((struct sd_softc *));
103 void	sdstart __P((void *));
104 void	sddone __P((struct scsipi_xfer *));
105 void	sd_shutdown __P((void *));
106 int	sd_reassign_blocks __P((struct sd_softc *, u_long));
107 int	sd_interpret_sense __P((struct scsipi_xfer *));
108 
109 extern struct cfdriver sd_cd;
110 
111 struct dkdriver sddkdriver = { sdstrategy };
112 
113 struct scsipi_device sd_switch = {
114 	sd_interpret_sense,	/* check our error handler first */
115 	sdstart,		/* have a queue, served by this */
116 	NULL,			/* have no async handler */
117 	sddone,			/* deal with stats at interrupt time */
118 };
119 
120 /*
121  * The routine called by the low level scsi routine when it discovers
122  * a device suitable for this driver.
123  */
124 void
125 sdattach(parent, sd, sc_link, ops)
126 	struct device *parent;
127 	struct sd_softc *sd;
128 	struct scsipi_link *sc_link;
129 	const struct sd_ops *ops;
130 {
131 	int error, result;
132 	struct disk_parms *dp = &sd->params;
133 
134 	SC_DEBUG(sc_link, SDEV_DB2, ("sdattach: "));
135 
136 	/*
137 	 * Store information needed to contact our base driver
138 	 */
139 	sd->sc_link = sc_link;
140 	sd->sc_ops = ops;
141 	sc_link->device = &sd_switch;
142 	sc_link->device_softc = sd;
143 	if (sc_link->openings > SDOUTSTANDING)
144 		sc_link->openings = SDOUTSTANDING;
145 
146 	/*
147 	 * Initialize and attach the disk structure.
148 	 */
149 	sd->sc_dk.dk_driver = &sddkdriver;
150 	sd->sc_dk.dk_name = sd->sc_dev.dv_xname;
151 	disk_attach(&sd->sc_dk);
152 
153 #if !defined(i386)
154 	dk_establish(&sd->sc_dk, &sd->sc_dev);		/* XXX */
155 #endif
156 
157 	/*
158 	 * Use the subdriver to request information regarding
159 	 * the drive. We cannot use interrupts yet, so the
160 	 * request must specify this.
161 	 */
162 	printf("\n");
163 
164 #if NSD_ATAPIBUS > 0
165 	/* XXX BEGIN HACK ALERT!!! */
166 	if (sc_link->type == BUS_ATAPI) {
167 		/*
168 		 * The ATAPI sense data handling is _TOTALLY_ broken.
169 		 * In particular, it _never_ does a REQUEST_SENSE to get
170 		 * sense data!  This causes UNIT ATTENTION to be not
171 		 * cleared properly for some ATAPI devices (e.g. Zip),
172 		 * which messes up the probe.
173 		 *
174 		 * This is the easiest way to get around the problem.
175 		 *
176 		 * YUCK!
177 		 */
178 		(void)scsipi_test_unit_ready(sc_link, SCSI_AUTOCONF |
179 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE |
180 		    SCSI_IGNORE_NOT_READY);
181 	}
182 	/* XXX END HACK ALERT!!! */
183 #endif /* NSD_ATAPIBUS > 0 */
184 
185 	if ((sd->sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
186 		error = scsipi_start(sd->sc_link, SSS_START,
187 		    SCSI_AUTOCONF | SCSI_IGNORE_ILLEGAL_REQUEST |
188 		    SCSI_IGNORE_MEDIA_CHANGE | SCSI_SILENT);
189 	} else
190 		error = 0;
191 
192 	if (error)
193 		result = SDGP_RESULT_OFFLINE;
194 	else
195 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
196 		    SCSI_AUTOCONF);
197 	printf("%s: ", sd->sc_dev.dv_xname);
198 	switch (result) {
199 	case SDGP_RESULT_OK:
200 	        printf("%ldMB, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %ld sectors",
201 		    dp->disksize / (1048576 / dp->blksize), dp->cyls,
202 		    dp->heads, dp->sectors, dp->blksize, dp->disksize);
203 		break;
204 
205 	case SDGP_RESULT_OFFLINE:
206 		printf("drive offline");
207 		break;
208 
209 	case SDGP_RESULT_UNFORMATTED:
210 		printf("unformatted media");
211 		break;
212 
213 #ifdef DIAGNOSTIC
214 	default:
215 		panic("sdattach: unknown result from get_parms");
216 		break;
217 #endif
218 	}
219 	printf("\n");
220 
221 	/*
222 	 * Establish a shutdown hook so that we can ensure that
223 	 * our data has actually made it onto the platter at
224 	 * shutdown time.  Note that this relies on the fact
225 	 * that the shutdown hook code puts us at the head of
226 	 * the list (thus guaranteeing that our hook runs before
227 	 * our ancestors').
228 	 */
229 	if ((sd->sc_sdhook =
230 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
231 		printf("%s: WARNING: unable to establish shutdown hook\n",
232 		    sd->sc_dev.dv_xname);
233 
234 #if NRND > 0
235 	/*
236 	 * attach the device into the random source list
237 	 */
238 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname, RND_TYPE_DISK);
239 #endif
240 }
241 
242 /*
243  * Wait interruptibly for an exclusive lock.
244  *
245  * XXX
246  * Several drivers do this; it should be abstracted and made MP-safe.
247  */
248 int
249 sdlock(sd)
250 	struct sd_softc *sd;
251 {
252 	int error;
253 
254 	while ((sd->flags & SDF_LOCKED) != 0) {
255 		sd->flags |= SDF_WANTED;
256 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
257 			return (error);
258 	}
259 	sd->flags |= SDF_LOCKED;
260 	return (0);
261 }
262 
263 /*
264  * Unlock and wake up any waiters.
265  */
266 void
267 sdunlock(sd)
268 	struct sd_softc *sd;
269 {
270 
271 	sd->flags &= ~SDF_LOCKED;
272 	if ((sd->flags & SDF_WANTED) != 0) {
273 		sd->flags &= ~SDF_WANTED;
274 		wakeup(sd);
275 	}
276 }
277 
278 /*
279  * open the device. Make sure the partition info is a up-to-date as can be.
280  */
281 int
282 sdopen(dev, flag, fmt, p)
283 	dev_t dev;
284 	int flag, fmt;
285 	struct proc *p;
286 {
287 	struct sd_softc *sd;
288 	struct scsipi_link *sc_link;
289 	int unit, part;
290 	int error;
291 
292 	unit = SDUNIT(dev);
293 	if (unit >= sd_cd.cd_ndevs)
294 		return (ENXIO);
295 	sd = sd_cd.cd_devs[unit];
296 	if (sd == NULL)
297 		return (ENXIO);
298 
299 	sc_link = sd->sc_link;
300 
301 	SC_DEBUG(sc_link, SDEV_DB1,
302 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
303 	    sd_cd.cd_ndevs, SDPART(dev)));
304 
305 	if ((error = sdlock(sd)) != 0)
306 		return (error);
307 
308 	if (sd->sc_dk.dk_openmask != 0) {
309 		/*
310 		 * If any partition is open, but the disk has been invalidated,
311 		 * disallow further opens.
312 		 */
313 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
314 			error = EIO;
315 			goto bad3;
316 		}
317 	} else {
318 		/* Check that it is still responding and ok. */
319 		error = scsipi_test_unit_ready(sc_link,
320 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE |
321 		    SCSI_IGNORE_NOT_READY);
322 		if (error)
323 			goto bad3;
324 
325 		/* Start the pack spinning if necessary. */
326 		if ((sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
327 			error = scsipi_start(sc_link, SSS_START,
328 			    SCSI_IGNORE_ILLEGAL_REQUEST |
329 			    SCSI_IGNORE_MEDIA_CHANGE | SCSI_SILENT);
330 			if (error)
331 				goto bad3;
332 		}
333 
334 		sc_link->flags |= SDEV_OPEN;
335 
336 		/* Lock the pack in. */
337 		error = scsipi_prevent(sc_link, PR_PREVENT,
338 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE);
339 		if (error)
340 			goto bad;
341 
342 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
343 			sc_link->flags |= SDEV_MEDIA_LOADED;
344 
345 			/*
346 			 * Load the physical device parameters.
347 			 *
348 			 * Note that if media is present but unformatted,
349 			 * we allow the open (so that it can be formatted!).
350 			 * The drive should refuse real I/O, if the media is
351 			 * unformatted.
352 			 */
353 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
354 			    0) == SDGP_RESULT_OFFLINE) {
355 				error = ENXIO;
356 				goto bad2;
357 			}
358 			SC_DEBUG(sc_link, SDEV_DB3, ("Params loaded "));
359 
360 			/* Load the partition info if not already loaded. */
361 			sdgetdisklabel(sd);
362 			SC_DEBUG(sc_link, SDEV_DB3, ("Disklabel loaded "));
363 		}
364 	}
365 
366 	part = SDPART(dev);
367 
368 	/* Check that the partition exists. */
369 	if (part != RAW_PART &&
370 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
371 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
372 		error = ENXIO;
373 		goto bad;
374 	}
375 
376 	/* Insure only one open at a time. */
377 	switch (fmt) {
378 	case S_IFCHR:
379 		sd->sc_dk.dk_copenmask |= (1 << part);
380 		break;
381 	case S_IFBLK:
382 		sd->sc_dk.dk_bopenmask |= (1 << part);
383 		break;
384 	}
385 	sd->sc_dk.dk_openmask =
386 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
387 
388 	SC_DEBUG(sc_link, SDEV_DB3, ("open complete\n"));
389 	sdunlock(sd);
390 	return (0);
391 
392 bad2:
393 	sc_link->flags &= ~SDEV_MEDIA_LOADED;
394 
395 bad:
396 	if (sd->sc_dk.dk_openmask == 0) {
397 		scsipi_prevent(sc_link, PR_ALLOW,
398 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE);
399 		sc_link->flags &= ~SDEV_OPEN;
400 	}
401 
402 bad3:
403 	sdunlock(sd);
404 	return (error);
405 }
406 
407 /*
408  * close the device.. only called if we are the LAST occurence of an open
409  * device.  Convenient now but usually a pain.
410  */
411 int
412 sdclose(dev, flag, fmt, p)
413 	dev_t dev;
414 	int flag, fmt;
415 	struct proc *p;
416 {
417 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
418 	int part = SDPART(dev);
419 	int error;
420 
421 	if ((error = sdlock(sd)) != 0)
422 		return (error);
423 
424 	switch (fmt) {
425 	case S_IFCHR:
426 		sd->sc_dk.dk_copenmask &= ~(1 << part);
427 		break;
428 	case S_IFBLK:
429 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
430 		break;
431 	}
432 	sd->sc_dk.dk_openmask =
433 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
434 
435 	if (sd->sc_dk.dk_openmask == 0) {
436 		/*
437 		 * If the disk cache needs flushing, and the disk supports
438 		 * it, do it now.
439 		 */
440 		if ((sd->flags & SDF_DIRTY) != 0 &&
441 		    sd->sc_ops->sdo_flush != NULL)
442 			(*sd->sc_ops->sdo_flush)(sd, 0);
443 
444 		/* XXXX Must wait for I/O to complete! */
445 
446 		scsipi_prevent(sd->sc_link, PR_ALLOW,
447 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_NOT_READY);
448 		sd->sc_link->flags &= ~(SDEV_OPEN|SDEV_MEDIA_LOADED);
449 	}
450 
451 	sdunlock(sd);
452 	return (0);
453 }
454 
455 /*
456  * Actually translate the requested transfer into one the physical driver
457  * can understand.  The transfer is described by a buf and will include
458  * only one physical transfer.
459  */
460 void
461 sdstrategy(bp)
462 	struct buf *bp;
463 {
464 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
465 	int s;
466 
467 	SC_DEBUG(sd->sc_link, SDEV_DB2, ("sdstrategy "));
468 	SC_DEBUG(sd->sc_link, SDEV_DB1,
469 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
470 	/*
471 	 * The transfer must be a whole number of blocks.
472 	 */
473 	if ((bp->b_bcount % sd->sc_dk.dk_label->d_secsize) != 0) {
474 		bp->b_error = EINVAL;
475 		goto bad;
476 	}
477 	/*
478 	 * If the device has been made invalid, error out
479 	 */
480 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
481 		bp->b_error = EIO;
482 		goto bad;
483 	}
484 	/*
485 	 * If it's a null transfer, return immediatly
486 	 */
487 	if (bp->b_bcount == 0)
488 		goto done;
489 
490 	/*
491 	 * Do bounds checking, adjust transfer. if error, process.
492 	 * If end of partition, just return.
493 	 */
494 	if (SDPART(bp->b_dev) != RAW_PART &&
495 	    bounds_check_with_label(bp, sd->sc_dk.dk_label,
496 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
497 		goto done;
498 
499 	s = splbio();
500 
501 	/*
502 	 * Place it in the queue of disk activities for this disk
503 	 */
504 	disksort(&sd->buf_queue, bp);
505 
506 	/*
507 	 * Tell the device to get going on the transfer if it's
508 	 * not doing anything, otherwise just wait for completion
509 	 */
510 	sdstart(sd);
511 
512 	splx(s);
513 	return;
514 
515 bad:
516 	bp->b_flags |= B_ERROR;
517 done:
518 	/*
519 	 * Correctly set the buf to indicate a completed xfer
520 	 */
521 	bp->b_resid = bp->b_bcount;
522 	biodone(bp);
523 }
524 
525 /*
526  * sdstart looks to see if there is a buf waiting for the device
527  * and that the device is not already busy. If both are true,
528  * It dequeues the buf and creates a scsi command to perform the
529  * transfer in the buf. The transfer request will call scsipi_done
530  * on completion, which will in turn call this routine again
531  * so that the next queued transfer is performed.
532  * The bufs are queued by the strategy routine (sdstrategy)
533  *
534  * This routine is also called after other non-queued requests
535  * have been made of the scsi driver, to ensure that the queue
536  * continues to be drained.
537  *
538  * must be called at the correct (highish) spl level
539  * sdstart() is called at splbio from sdstrategy and scsipi_done
540  */
541 void
542 sdstart(v)
543 	register void *v;
544 {
545 	register struct sd_softc *sd = v;
546 	register struct	scsipi_link *sc_link = sd->sc_link;
547 	struct disklabel *lp = sd->sc_dk.dk_label;
548 	struct buf *bp = 0;
549 	struct buf *dp;
550 	struct scsipi_rw_big cmd_big;
551 #if NSD_SCSIBUS > 0
552 	struct scsi_rw cmd_small;
553 #endif
554 	struct scsipi_generic *cmdp;
555 	int blkno, nblks, cmdlen, error;
556 	struct partition *p;
557 
558 	SC_DEBUG(sc_link, SDEV_DB2, ("sdstart "));
559 	/*
560 	 * Check if the device has room for another command
561 	 */
562 	while (sc_link->openings > 0) {
563 		/*
564 		 * there is excess capacity, but a special waits
565 		 * It'll need the adapter as soon as we clear out of the
566 		 * way and let it run (user level wait).
567 		 */
568 		if (sc_link->flags & SDEV_WAITING) {
569 			sc_link->flags &= ~SDEV_WAITING;
570 			wakeup((caddr_t)sc_link);
571 			return;
572 		}
573 
574 		/*
575 		 * See if there is a buf with work for us to do..
576 		 */
577 		dp = &sd->buf_queue;
578 		if ((bp = dp->b_actf) == NULL)	/* yes, an assign */
579 			return;
580 		dp->b_actf = bp->b_actf;
581 
582 		/*
583 		 * If the device has become invalid, abort all the
584 		 * reads and writes until all files have been closed and
585 		 * re-opened
586 		 */
587 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
588 			bp->b_error = EIO;
589 			bp->b_flags |= B_ERROR;
590 			bp->b_resid = bp->b_bcount;
591 			biodone(bp);
592 			continue;
593 		}
594 
595 		/*
596 		 * We have a buf, now we should make a command
597 		 *
598 		 * First, translate the block to absolute and put it in terms
599 		 * of the logical blocksize of the device.
600 		 */
601 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
602 		if (SDPART(bp->b_dev) != RAW_PART) {
603 			p = &lp->d_partitions[SDPART(bp->b_dev)];
604 			blkno += p->p_offset;
605 		}
606 		nblks = howmany(bp->b_bcount, lp->d_secsize);
607 
608 #if NSD_SCSIBUS > 0
609 		/*
610 		 *  Fill out the scsi command.  If the transfer will
611 		 *  fit in a "small" cdb, use it.
612 		 */
613 		if (((blkno & 0x1fffff) == blkno) &&
614 		    ((nblks & 0xff) == nblks) && sc_link->type == BUS_SCSI) {
615 			/*
616 			 * We can fit in a small cdb.
617 			 */
618 			bzero(&cmd_small, sizeof(cmd_small));
619 			cmd_small.opcode = (bp->b_flags & B_READ) ?
620 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
621 			_lto3b(blkno, cmd_small.addr);
622 			cmd_small.length = nblks & 0xff;
623 			cmdlen = sizeof(cmd_small);
624 			cmdp = (struct scsipi_generic *)&cmd_small;
625 		} else
626 #endif
627 		{
628 			/*
629 			 * Need a large cdb.
630 			 */
631 			bzero(&cmd_big, sizeof(cmd_big));
632 			cmd_big.opcode = (bp->b_flags & B_READ) ?
633 			    READ_BIG : WRITE_BIG;
634 			_lto4b(blkno, cmd_big.addr);
635 			_lto2b(nblks, cmd_big.length);
636 			cmdlen = sizeof(cmd_big);
637 			cmdp = (struct scsipi_generic *)&cmd_big;
638 		}
639 
640 		/* Instrumentation. */
641 		disk_busy(&sd->sc_dk);
642 
643 		/*
644 		 * Mark the disk dirty so that the cache will be
645 		 * flushed on close.
646 		 */
647 		if ((bp->b_flags & B_READ) == 0)
648 			sd->flags |= SDF_DIRTY;
649 
650 		/*
651 		 * Call the routine that chats with the adapter.
652 		 * Note: we cannot sleep as we may be an interrupt
653 		 */
654 		error = scsipi_command(sc_link, cmdp, cmdlen,
655 		    (u_char *)bp->b_data, bp->b_bcount,
656 		    SDRETRIES, 60000, bp, SCSI_NOSLEEP |
657 		    ((bp->b_flags & B_READ) ? SCSI_DATA_IN : SCSI_DATA_OUT));
658 		if (error) {
659 			disk_unbusy(&sd->sc_dk, 0);
660 			printf("%s: not queued, error %d\n",
661 			    sd->sc_dev.dv_xname, error);
662 		}
663 	}
664 }
665 
666 void
667 sddone(xs)
668 	struct scsipi_xfer *xs;
669 {
670 	struct sd_softc *sd = xs->sc_link->device_softc;
671 
672 	if (sd->flags & SDF_FLUSHING) {
673 		/* Flush completed, no longer dirty. */
674 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
675 	}
676 
677 	if (xs->bp != NULL) {
678 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid);
679 #if NRND > 0
680 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_blkno);
681 #endif
682 	}
683 }
684 
685 void
686 sdminphys(bp)
687 	struct buf *bp;
688 {
689 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
690 	long max;
691 
692 	/*
693 	 * If the device is ancient, we want to make sure that
694 	 * the transfer fits into a 6-byte cdb.
695 	 *
696 	 * XXX Note that the SCSI-I spec says that 256-block transfers
697 	 * are allowed in a 6-byte read/write, and are specified
698 	 * by settng the "length" to 0.  However, we're conservative
699 	 * here, allowing only 255-block transfers in case an
700 	 * ancient device gets confused by length == 0.  A length of 0
701 	 * in a 10-byte read/write actually means 0 blocks.
702 	 */
703 	if (sd->flags & SDF_ANCIENT) {
704 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
705 
706 		if (bp->b_bcount > max)
707 			bp->b_bcount = max;
708 	}
709 
710 	(*sd->sc_link->adapter->scsipi_minphys)(bp);
711 }
712 
713 int
714 sdread(dev, uio, ioflag)
715 	dev_t dev;
716 	struct uio *uio;
717 	int ioflag;
718 {
719 
720 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
721 }
722 
723 int
724 sdwrite(dev, uio, ioflag)
725 	dev_t dev;
726 	struct uio *uio;
727 	int ioflag;
728 {
729 
730 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
731 }
732 
733 /*
734  * Perform special action on behalf of the user
735  * Knows about the internals of this device
736  */
737 int
738 sdioctl(dev, cmd, addr, flag, p)
739 	dev_t dev;
740 	u_long cmd;
741 	caddr_t addr;
742 	int flag;
743 	struct proc *p;
744 {
745 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
746 	int error;
747 
748 	SC_DEBUG(sd->sc_link, SDEV_DB2, ("sdioctl 0x%lx ", cmd));
749 
750 	/*
751 	 * If the device is not valid.. abandon ship
752 	 */
753 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) == 0)
754 		return (EIO);
755 
756 	switch (cmd) {
757 	case DIOCGDINFO:
758 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
759 		return (0);
760 
761 	case DIOCGPART:
762 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
763 		((struct partinfo *)addr)->part =
764 		    &sd->sc_dk.dk_label->d_partitions[SDPART(dev)];
765 		return (0);
766 
767 	case DIOCWDINFO:
768 	case DIOCSDINFO:
769 		if ((flag & FWRITE) == 0)
770 			return (EBADF);
771 
772 		if ((error = sdlock(sd)) != 0)
773 			return (error);
774 		sd->flags |= SDF_LABELLING;
775 
776 		error = setdisklabel(sd->sc_dk.dk_label,
777 		    (struct disklabel *)addr, /*sd->sc_dk.dk_openmask : */0,
778 		    sd->sc_dk.dk_cpulabel);
779 		if (error == 0) {
780 			if (cmd == DIOCWDINFO)
781 				error = writedisklabel(SDLABELDEV(dev),
782 				    sdstrategy, sd->sc_dk.dk_label,
783 				    sd->sc_dk.dk_cpulabel);
784 		}
785 
786 		sd->flags &= ~SDF_LABELLING;
787 		sdunlock(sd);
788 		return (error);
789 
790 	case DIOCWLABEL:
791 		if ((flag & FWRITE) == 0)
792 			return (EBADF);
793 		if (*(int *)addr)
794 			sd->flags |= SDF_WLABEL;
795 		else
796 			sd->flags &= ~SDF_WLABEL;
797 		return (0);
798 
799 	case DIOCLOCK:
800 		return (scsipi_prevent(sd->sc_link,
801 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
802 
803 	case DIOCEJECT:
804 		return ((sd->sc_link->flags & SDEV_REMOVABLE) == 0 ? ENOTTY :
805 		    scsipi_start(sd->sc_link, SSS_STOP|SSS_LOEJ, 0));
806 
807 	case DIOCGDEFLABEL:
808 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
809 		return (0);
810 
811 	default:
812 		if (SDPART(dev) != RAW_PART)
813 			return (ENOTTY);
814 		return (scsipi_do_ioctl(sd->sc_link, dev, cmd, addr, flag, p));
815 	}
816 
817 #ifdef DIAGNOSTIC
818 	panic("sdioctl: impossible");
819 #endif
820 }
821 
822 void
823 sdgetdefaultlabel(sd, lp)
824 	struct sd_softc *sd;
825 	struct disklabel *lp;
826 {
827 
828 	bzero(lp, sizeof(struct disklabel));
829 
830 	lp->d_secsize = sd->params.blksize;
831 	lp->d_ntracks = sd->params.heads;
832 	lp->d_nsectors = sd->params.sectors;
833 	lp->d_ncylinders = sd->params.cyls;
834 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
835 
836 	switch (sd->sc_link->type) {
837 #if NSD_SCSIBUS > 0
838 	    case BUS_SCSI:
839 		lp->d_type = DTYPE_SCSI;
840 		break;
841 #endif
842 #if NSD_ATAPIBUS > 0
843 	    case BUS_ATAPI:
844 		lp->d_type = DTYPE_ATAPI;
845 		break;
846 #endif
847 	}
848 	strncpy(lp->d_typename, sd->name, 16);
849 	strncpy(lp->d_packname, "fictitious", 16);
850 	lp->d_secperunit = sd->params.disksize;
851 	lp->d_rpm = sd->params.rot_rate;
852 	lp->d_interleave = 1;
853 	lp->d_flags = 0;
854 
855 	lp->d_partitions[RAW_PART].p_offset = 0;
856 	lp->d_partitions[RAW_PART].p_size =
857 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
858 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
859 	lp->d_npartitions = RAW_PART + 1;
860 
861 	lp->d_magic = DISKMAGIC;
862 	lp->d_magic2 = DISKMAGIC;
863 	lp->d_checksum = dkcksum(lp);
864 }
865 
866 
867 /*
868  * Load the label information on the named device
869  */
870 void
871 sdgetdisklabel(sd)
872 	struct sd_softc *sd;
873 {
874 	struct disklabel *lp = sd->sc_dk.dk_label;
875 	char *errstring;
876 
877 	bzero(sd->sc_dk.dk_cpulabel, sizeof(struct cpu_disklabel));
878 
879 	sdgetdefaultlabel(sd, lp);
880 
881 	if (lp->d_secpercyl == 0) {
882 		lp->d_secpercyl = 100;
883 		/* as long as it's not 0 - readdisklabel divides by it (?) */
884 	}
885 
886 	/*
887 	 * Call the generic disklabel extraction routine
888 	 */
889 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
890 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
891 	if (errstring) {
892 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
893 		return;
894 	}
895 }
896 
897 void
898 sd_shutdown(arg)
899 	void *arg;
900 {
901 	struct sd_softc *sd = arg;
902 
903 	/*
904 	 * If the disk cache needs to be flushed, and the disk supports
905 	 * it, flush it.  We're cold at this point, so we poll for
906 	 * completion.
907 	 */
908 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL)
909 		(*sd->sc_ops->sdo_flush)(sd, SCSI_AUTOCONF);
910 }
911 
912 /*
913  * Tell the device to map out a defective block
914  */
915 int
916 sd_reassign_blocks(sd, blkno)
917 	struct sd_softc *sd;
918 	u_long blkno;
919 {
920 	struct scsi_reassign_blocks scsipi_cmd;
921 	struct scsi_reassign_blocks_data rbdata;
922 
923 	bzero(&scsipi_cmd, sizeof(scsipi_cmd));
924 	bzero(&rbdata, sizeof(rbdata));
925 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
926 
927 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
928 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
929 
930 	return (scsipi_command(sd->sc_link,
931 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
932 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
933 	    SCSI_DATA_OUT));
934 }
935 
936 /*
937  * Check Errors
938  */
939 int
940 sd_interpret_sense(xs)
941 	struct scsipi_xfer *xs;
942 {
943 	struct scsipi_link *sc_link = xs->sc_link;
944 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
945 	struct sd_softc *sd = sc_link->device_softc;
946 	int retval = SCSIRET_CONTINUE;
947 
948 	/*
949 	 * If the device is not open yet, let the generic code handle it.
950 	 */
951 	if ((sc_link->flags & SDEV_OPEN) == 0) {
952 		return (retval);
953 	}
954 
955 	/*
956 	 * If it isn't a extended or extended/deferred error, let
957 	 * the generic code handle it.
958 	 */
959 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
960 	    (sense->error_code & SSD_ERRCODE) != 0x71) {	/* DEFFERRED */
961 		return (retval);
962 	}
963 
964 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
965 	    sense->add_sense_code == 0x4) {
966 		if (sense->add_sense_code_qual == 0x01)	{
967 			printf("%s: ..is spinning up...waiting\n",
968 			    sd->sc_dev.dv_xname);
969 			/*
970 			 * I really need a sdrestart function I can call here.
971 			 */
972 			delay(1000000 * 5);	/* 5 seconds */
973 			retval = SCSIRET_RETRY;
974 		} else if ((sense->add_sense_code_qual == 0x2) &&
975 		    (sd->sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
976 			if (sd->sc_link->flags & SDEV_REMOVABLE) {
977 				printf("%s: removable disk stopped- not "
978 				    "restarting\n", sd->sc_dev.dv_xname);
979 				retval = EIO;
980 			} else {
981 				printf("%s: respinning up disk\n",
982 				    sd->sc_dev.dv_xname);
983 				retval = scsipi_start(sd->sc_link, SSS_START,
984 				    SCSI_URGENT | SCSI_NOSLEEP);
985 				if (retval != 0) {
986 					printf("%s: respin of disk failed-%d\n",
987 					    sd->sc_dev.dv_xname, retval);
988 					retval = EIO;
989 				} else {
990 					retval = SCSIRET_RETRY;
991 				}
992 			}
993 		}
994 	}
995 	return (retval);
996 }
997 
998 
999 int
1000 sdsize(dev)
1001 	dev_t dev;
1002 {
1003 	struct sd_softc *sd;
1004 	int part, unit, omask;
1005 	int size;
1006 
1007 	unit = SDUNIT(dev);
1008 	if (unit >= sd_cd.cd_ndevs)
1009 		return (-1);
1010 	sd = sd_cd.cd_devs[unit];
1011 	if (sd == NULL)
1012 		return (-1);
1013 
1014 	part = SDPART(dev);
1015 	omask = sd->sc_dk.dk_openmask & (1 << part);
1016 
1017 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1018 		return (-1);
1019 	if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1020 		size = -1;
1021 	else
1022 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1023 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1024 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1025 		return (-1);
1026 	return (size);
1027 }
1028 
1029 #ifndef __BDEVSW_DUMP_OLD_TYPE
1030 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1031 static struct scsipi_xfer sx;
1032 static int sddoingadump;
1033 
1034 /*
1035  * dump all of physical memory into the partition specified, starting
1036  * at offset 'dumplo' into the partition.
1037  */
1038 int
1039 sddump(dev, blkno, va, size)
1040 	dev_t dev;
1041 	daddr_t blkno;
1042 	caddr_t va;
1043 	size_t size;
1044 {
1045 	struct sd_softc *sd;	/* disk unit to do the I/O */
1046 	struct disklabel *lp;	/* disk's disklabel */
1047 	int	unit, part;
1048 	int	sectorsize;	/* size of a disk sector */
1049 	int	nsects;		/* number of sectors in partition */
1050 	int	sectoff;	/* sector offset of partition */
1051 	int	totwrt;		/* total number of sectors left to write */
1052 	int	nwrt;		/* current number of sectors to write */
1053 	struct scsipi_rw_big cmd;	/* write command */
1054 	struct scsipi_xfer *xs;	/* ... convenience */
1055 	int	retval;
1056 
1057 	/* Check if recursive dump; if so, punt. */
1058 	if (sddoingadump)
1059 		return (EFAULT);
1060 
1061 	/* Mark as active early. */
1062 	sddoingadump = 1;
1063 
1064 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1065 	part = SDPART(dev);
1066 
1067 	/* Check for acceptable drive number. */
1068 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1069 		return (ENXIO);
1070 
1071 	/*
1072 	 * XXX Can't do this check, since the media might have been
1073 	 * XXX marked `invalid' by successful unmounting of all
1074 	 * XXX filesystems.
1075 	 */
1076 #if 0
1077 	/* Make sure it was initialized. */
1078 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) != SDEV_MEDIA_LOADED)
1079 		return (ENXIO);
1080 #endif
1081 
1082 	/* Convert to disk sectors.  Request must be a multiple of size. */
1083 	lp = sd->sc_dk.dk_label;
1084 	sectorsize = lp->d_secsize;
1085 	if ((size % sectorsize) != 0)
1086 		return (EFAULT);
1087 	totwrt = size / sectorsize;
1088 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1089 
1090 	nsects = lp->d_partitions[part].p_size;
1091 	sectoff = lp->d_partitions[part].p_offset;
1092 
1093 	/* Check transfer bounds against partition size. */
1094 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1095 		return (EINVAL);
1096 
1097 	/* Offset block number to start of partition. */
1098 	blkno += sectoff;
1099 
1100 	xs = &sx;
1101 
1102 	while (totwrt > 0) {
1103 		nwrt = totwrt;		/* XXX */
1104 #ifndef	SD_DUMP_NOT_TRUSTED
1105 		/*
1106 		 *  Fill out the scsi command
1107 		 */
1108 		bzero(&cmd, sizeof(cmd));
1109 		cmd.opcode = WRITE_BIG;
1110 		_lto4b(blkno, cmd.addr);
1111 		_lto2b(nwrt, cmd.length);
1112 		/*
1113 		 * Fill out the scsipi_xfer structure
1114 		 *    Note: we cannot sleep as we may be an interrupt
1115 		 * don't use scsipi_command() as it may want to wait
1116 		 * for an xs.
1117 		 */
1118 		bzero(xs, sizeof(sx));
1119 		xs->flags |= SCSI_AUTOCONF | INUSE | SCSI_DATA_OUT;
1120 		xs->sc_link = sd->sc_link;
1121 		xs->retries = SDRETRIES;
1122 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1123 		xs->cmd = (struct scsipi_generic *)&cmd;
1124 		xs->cmdlen = sizeof(cmd);
1125 		xs->resid = nwrt * sectorsize;
1126 		xs->error = XS_NOERROR;
1127 		xs->bp = 0;
1128 		xs->data = va;
1129 		xs->datalen = nwrt * sectorsize;
1130 
1131 		/*
1132 		 * Pass all this info to the scsi driver.
1133 		 */
1134 		retval = scsipi_command_direct(xs);
1135 		if (retval != COMPLETE)
1136 			return (ENXIO);
1137 #else	/* SD_DUMP_NOT_TRUSTED */
1138 		/* Let's just talk about this first... */
1139 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1140 		delay(500 * 1000);	/* half a second */
1141 #endif	/* SD_DUMP_NOT_TRUSTED */
1142 
1143 		/* update block count */
1144 		totwrt -= nwrt;
1145 		blkno += nwrt;
1146 		va += sectorsize * nwrt;
1147 	}
1148 	sddoingadump = 0;
1149 	return (0);
1150 }
1151 #else	/* __BDEVSW_DUMP_NEW_TYPE */
1152 int
1153 sddump(dev, blkno, va, size)
1154 	dev_t dev;
1155 	daddr_t blkno;
1156 	caddr_t va;
1157 	size_t size;
1158 {
1159 
1160 	/* Not implemented. */
1161 	return (ENXIO);
1162 }
1163 #endif	/* __BDEVSW_DUMP_NEW_TYPE */
1164