xref: /netbsd-src/sys/dev/scsipi/sd.c (revision dc306354b0b29af51801a7632f1e95265a68cd81)
1 /*	$NetBSD: sd.c,v 1.138 1998/12/08 00:18:46 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 ((sd->sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
165 		error = scsipi_start(sd->sc_link, SSS_START,
166 		    SCSI_AUTOCONF | SCSI_IGNORE_ILLEGAL_REQUEST |
167 		    SCSI_IGNORE_MEDIA_CHANGE | SCSI_SILENT);
168 	} else
169 		error = 0;
170 
171 	if (error)
172 		result = SDGP_RESULT_OFFLINE;
173 	else
174 		result = (*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
175 		    SCSI_AUTOCONF);
176 	printf("%s: ", sd->sc_dev.dv_xname);
177 	switch (result) {
178 	case SDGP_RESULT_OK:
179 	        printf("%ldMB, %ld cyl, %ld head, %ld sec, %ld bytes/sect x %ld sectors",
180 		    dp->disksize / (1048576 / dp->blksize), dp->cyls,
181 		    dp->heads, dp->sectors, dp->blksize, dp->disksize);
182 		break;
183 
184 	case SDGP_RESULT_OFFLINE:
185 		printf("drive offline");
186 		break;
187 
188 	case SDGP_RESULT_UNFORMATTED:
189 		printf("unformatted media");
190 		break;
191 
192 #ifdef DIAGNOSTIC
193 	default:
194 		panic("sdattach: unknown result from get_parms");
195 		break;
196 #endif
197 	}
198 	printf("\n");
199 
200 	/*
201 	 * Establish a shutdown hook so that we can ensure that
202 	 * our data has actually made it onto the platter at
203 	 * shutdown time.  Note that this relies on the fact
204 	 * that the shutdown hook code puts us at the head of
205 	 * the list (thus guaranteeing that our hook runs before
206 	 * our ancestors').
207 	 */
208 	if ((sd->sc_sdhook =
209 	    shutdownhook_establish(sd_shutdown, sd)) == NULL)
210 		printf("%s: WARNING: unable to establish shutdown hook\n",
211 		    sd->sc_dev.dv_xname);
212 
213 #if NRND > 0
214 	/*
215 	 * attach the device into the random source list
216 	 */
217 	rnd_attach_source(&sd->rnd_source, sd->sc_dev.dv_xname, RND_TYPE_DISK);
218 #endif
219 }
220 
221 /*
222  * Wait interruptibly for an exclusive lock.
223  *
224  * XXX
225  * Several drivers do this; it should be abstracted and made MP-safe.
226  */
227 int
228 sdlock(sd)
229 	struct sd_softc *sd;
230 {
231 	int error;
232 
233 	while ((sd->flags & SDF_LOCKED) != 0) {
234 		sd->flags |= SDF_WANTED;
235 		if ((error = tsleep(sd, PRIBIO | PCATCH, "sdlck", 0)) != 0)
236 			return (error);
237 	}
238 	sd->flags |= SDF_LOCKED;
239 	return (0);
240 }
241 
242 /*
243  * Unlock and wake up any waiters.
244  */
245 void
246 sdunlock(sd)
247 	struct sd_softc *sd;
248 {
249 
250 	sd->flags &= ~SDF_LOCKED;
251 	if ((sd->flags & SDF_WANTED) != 0) {
252 		sd->flags &= ~SDF_WANTED;
253 		wakeup(sd);
254 	}
255 }
256 
257 /*
258  * open the device. Make sure the partition info is a up-to-date as can be.
259  */
260 int
261 sdopen(dev, flag, fmt, p)
262 	dev_t dev;
263 	int flag, fmt;
264 	struct proc *p;
265 {
266 	struct sd_softc *sd;
267 	struct scsipi_link *sc_link;
268 	int unit, part;
269 	int error;
270 
271 	unit = SDUNIT(dev);
272 	if (unit >= sd_cd.cd_ndevs)
273 		return (ENXIO);
274 	sd = sd_cd.cd_devs[unit];
275 	if (sd == NULL)
276 		return (ENXIO);
277 
278 	sc_link = sd->sc_link;
279 
280 	SC_DEBUG(sc_link, SDEV_DB1,
281 	    ("sdopen: dev=0x%x (unit %d (of %d), partition %d)\n", dev, unit,
282 	    sd_cd.cd_ndevs, SDPART(dev)));
283 
284 	/*
285 	 * If this is the first open of this device, add a reference
286 	 * to the adapter.
287 	 */
288 	if (sd->sc_dk.dk_openmask == 0 &&
289 	    (error = scsipi_adapter_addref(sc_link)) != 0)
290 		return (error);
291 
292 	if ((error = sdlock(sd)) != 0)
293 		goto bad4;
294 
295 	if (sd->sc_dk.dk_openmask != 0) {
296 		/*
297 		 * If any partition is open, but the disk has been invalidated,
298 		 * disallow further opens.
299 		 */
300 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
301 			error = EIO;
302 			goto bad3;
303 		}
304 	} else {
305 		/* Check that it is still responding and ok. */
306 		error = scsipi_test_unit_ready(sc_link,
307 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE |
308 		    SCSI_IGNORE_NOT_READY);
309 		if (error)
310 			goto bad3;
311 
312 		/* Start the pack spinning if necessary. */
313 		if ((sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
314 			error = scsipi_start(sc_link, SSS_START,
315 			    SCSI_IGNORE_ILLEGAL_REQUEST |
316 			    SCSI_IGNORE_MEDIA_CHANGE | SCSI_SILENT);
317 			if (error)
318 				goto bad3;
319 		}
320 
321 		sc_link->flags |= SDEV_OPEN;
322 
323 		/* Lock the pack in. */
324 		error = scsipi_prevent(sc_link, PR_PREVENT,
325 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE);
326 		if (error)
327 			goto bad;
328 
329 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
330 			sc_link->flags |= SDEV_MEDIA_LOADED;
331 
332 			/*
333 			 * Load the physical device parameters.
334 			 *
335 			 * Note that if media is present but unformatted,
336 			 * we allow the open (so that it can be formatted!).
337 			 * The drive should refuse real I/O, if the media is
338 			 * unformatted.
339 			 */
340 			if ((*sd->sc_ops->sdo_get_parms)(sd, &sd->params,
341 			    0) == SDGP_RESULT_OFFLINE) {
342 				error = ENXIO;
343 				goto bad2;
344 			}
345 			SC_DEBUG(sc_link, SDEV_DB3, ("Params loaded "));
346 
347 			/* Load the partition info if not already loaded. */
348 			sdgetdisklabel(sd);
349 			SC_DEBUG(sc_link, SDEV_DB3, ("Disklabel loaded "));
350 		}
351 	}
352 
353 	part = SDPART(dev);
354 
355 	/* Check that the partition exists. */
356 	if (part != RAW_PART &&
357 	    (part >= sd->sc_dk.dk_label->d_npartitions ||
358 	     sd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
359 		error = ENXIO;
360 		goto bad;
361 	}
362 
363 	/* Insure only one open at a time. */
364 	switch (fmt) {
365 	case S_IFCHR:
366 		sd->sc_dk.dk_copenmask |= (1 << part);
367 		break;
368 	case S_IFBLK:
369 		sd->sc_dk.dk_bopenmask |= (1 << part);
370 		break;
371 	}
372 	sd->sc_dk.dk_openmask =
373 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
374 
375 	SC_DEBUG(sc_link, SDEV_DB3, ("open complete\n"));
376 	sdunlock(sd);
377 	return (0);
378 
379 bad2:
380 	sc_link->flags &= ~SDEV_MEDIA_LOADED;
381 
382 bad:
383 	if (sd->sc_dk.dk_openmask == 0) {
384 		scsipi_prevent(sc_link, PR_ALLOW,
385 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_MEDIA_CHANGE);
386 		sc_link->flags &= ~SDEV_OPEN;
387 	}
388 
389 bad3:
390 	sdunlock(sd);
391 bad4:
392 	if (sd->sc_dk.dk_openmask == 0)
393 		scsipi_adapter_delref(sc_link);
394 	return (error);
395 }
396 
397 /*
398  * close the device.. only called if we are the LAST occurence of an open
399  * device.  Convenient now but usually a pain.
400  */
401 int
402 sdclose(dev, flag, fmt, p)
403 	dev_t dev;
404 	int flag, fmt;
405 	struct proc *p;
406 {
407 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
408 	int part = SDPART(dev);
409 	int error;
410 
411 	if ((error = sdlock(sd)) != 0)
412 		return (error);
413 
414 	switch (fmt) {
415 	case S_IFCHR:
416 		sd->sc_dk.dk_copenmask &= ~(1 << part);
417 		break;
418 	case S_IFBLK:
419 		sd->sc_dk.dk_bopenmask &= ~(1 << part);
420 		break;
421 	}
422 	sd->sc_dk.dk_openmask =
423 	    sd->sc_dk.dk_copenmask | sd->sc_dk.dk_bopenmask;
424 
425 	if (sd->sc_dk.dk_openmask == 0) {
426 		/*
427 		 * If the disk cache needs flushing, and the disk supports
428 		 * it, do it now.
429 		 */
430 		if ((sd->flags & SDF_DIRTY) != 0 &&
431 		    sd->sc_ops->sdo_flush != NULL)
432 			(*sd->sc_ops->sdo_flush)(sd, 0);
433 
434 		scsipi_wait_drain(sd->sc_link);
435 
436 		scsipi_prevent(sd->sc_link, PR_ALLOW,
437 		    SCSI_IGNORE_ILLEGAL_REQUEST | SCSI_IGNORE_NOT_READY);
438 		sd->sc_link->flags &= ~(SDEV_OPEN|SDEV_MEDIA_LOADED);
439 
440 		scsipi_wait_drain(sd->sc_link);
441 
442 		scsipi_adapter_delref(sd->sc_link);
443 	}
444 
445 	sdunlock(sd);
446 	return (0);
447 }
448 
449 /*
450  * Actually translate the requested transfer into one the physical driver
451  * can understand.  The transfer is described by a buf and will include
452  * only one physical transfer.
453  */
454 void
455 sdstrategy(bp)
456 	struct buf *bp;
457 {
458 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
459 	int s;
460 
461 	SC_DEBUG(sd->sc_link, SDEV_DB2, ("sdstrategy "));
462 	SC_DEBUG(sd->sc_link, SDEV_DB1,
463 	    ("%ld bytes @ blk %d\n", bp->b_bcount, bp->b_blkno));
464 	/*
465 	 * The transfer must be a whole number of blocks.
466 	 */
467 	if ((bp->b_bcount % sd->sc_dk.dk_label->d_secsize) != 0) {
468 		bp->b_error = EINVAL;
469 		goto bad;
470 	}
471 	/*
472 	 * If the device has been made invalid, error out
473 	 */
474 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
475 		bp->b_error = EIO;
476 		goto bad;
477 	}
478 	/*
479 	 * If it's a null transfer, return immediatly
480 	 */
481 	if (bp->b_bcount == 0)
482 		goto done;
483 
484 	/*
485 	 * Do bounds checking, adjust transfer. if error, process.
486 	 * If end of partition, just return.
487 	 */
488 	if (SDPART(bp->b_dev) != RAW_PART &&
489 	    bounds_check_with_label(bp, sd->sc_dk.dk_label,
490 	    (sd->flags & (SDF_WLABEL|SDF_LABELLING)) != 0) <= 0)
491 		goto done;
492 
493 	s = splbio();
494 
495 	/*
496 	 * Place it in the queue of disk activities for this disk
497 	 */
498 	disksort(&sd->buf_queue, bp);
499 
500 	/*
501 	 * Tell the device to get going on the transfer if it's
502 	 * not doing anything, otherwise just wait for completion
503 	 */
504 	sdstart(sd);
505 
506 	splx(s);
507 	return;
508 
509 bad:
510 	bp->b_flags |= B_ERROR;
511 done:
512 	/*
513 	 * Correctly set the buf to indicate a completed xfer
514 	 */
515 	bp->b_resid = bp->b_bcount;
516 	biodone(bp);
517 }
518 
519 /*
520  * sdstart looks to see if there is a buf waiting for the device
521  * and that the device is not already busy. If both are true,
522  * It dequeues the buf and creates a scsi command to perform the
523  * transfer in the buf. The transfer request will call scsipi_done
524  * on completion, which will in turn call this routine again
525  * so that the next queued transfer is performed.
526  * The bufs are queued by the strategy routine (sdstrategy)
527  *
528  * This routine is also called after other non-queued requests
529  * have been made of the scsi driver, to ensure that the queue
530  * continues to be drained.
531  *
532  * must be called at the correct (highish) spl level
533  * sdstart() is called at splbio from sdstrategy and scsipi_done
534  */
535 void
536 sdstart(v)
537 	register void *v;
538 {
539 	register struct sd_softc *sd = v;
540 	register struct	scsipi_link *sc_link = sd->sc_link;
541 	struct disklabel *lp = sd->sc_dk.dk_label;
542 	struct buf *bp = 0;
543 	struct buf *dp;
544 	struct scsipi_rw_big cmd_big;
545 #if NSD_SCSIBUS > 0
546 	struct scsi_rw cmd_small;
547 #endif
548 	struct scsipi_generic *cmdp;
549 	int blkno, nblks, cmdlen, error;
550 	struct partition *p;
551 
552 	SC_DEBUG(sc_link, SDEV_DB2, ("sdstart "));
553 	/*
554 	 * Check if the device has room for another command
555 	 */
556 	while (sc_link->openings > 0) {
557 		/*
558 		 * there is excess capacity, but a special waits
559 		 * It'll need the adapter as soon as we clear out of the
560 		 * way and let it run (user level wait).
561 		 */
562 		if (sc_link->flags & SDEV_WAITING) {
563 			sc_link->flags &= ~SDEV_WAITING;
564 			wakeup((caddr_t)sc_link);
565 			return;
566 		}
567 
568 		/*
569 		 * See if there is a buf with work for us to do..
570 		 */
571 		dp = &sd->buf_queue;
572 		if ((bp = dp->b_actf) == NULL)	/* yes, an assign */
573 			return;
574 		dp->b_actf = bp->b_actf;
575 
576 		/*
577 		 * If the device has become invalid, abort all the
578 		 * reads and writes until all files have been closed and
579 		 * re-opened
580 		 */
581 		if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
582 			bp->b_error = EIO;
583 			bp->b_flags |= B_ERROR;
584 			bp->b_resid = bp->b_bcount;
585 			biodone(bp);
586 			continue;
587 		}
588 
589 		/*
590 		 * We have a buf, now we should make a command
591 		 *
592 		 * First, translate the block to absolute and put it in terms
593 		 * of the logical blocksize of the device.
594 		 */
595 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
596 		if (SDPART(bp->b_dev) != RAW_PART) {
597 			p = &lp->d_partitions[SDPART(bp->b_dev)];
598 			blkno += p->p_offset;
599 		}
600 		nblks = howmany(bp->b_bcount, lp->d_secsize);
601 
602 #if NSD_SCSIBUS > 0
603 		/*
604 		 *  Fill out the scsi command.  If the transfer will
605 		 *  fit in a "small" cdb, use it.
606 		 */
607 		if (((blkno & 0x1fffff) == blkno) &&
608 		    ((nblks & 0xff) == nblks) && sc_link->type == BUS_SCSI) {
609 			/*
610 			 * We can fit in a small cdb.
611 			 */
612 			bzero(&cmd_small, sizeof(cmd_small));
613 			cmd_small.opcode = (bp->b_flags & B_READ) ?
614 			    SCSI_READ_COMMAND : SCSI_WRITE_COMMAND;
615 			_lto3b(blkno, cmd_small.addr);
616 			cmd_small.length = nblks & 0xff;
617 			cmdlen = sizeof(cmd_small);
618 			cmdp = (struct scsipi_generic *)&cmd_small;
619 		} else
620 #endif
621 		{
622 			/*
623 			 * Need a large cdb.
624 			 */
625 			bzero(&cmd_big, sizeof(cmd_big));
626 			cmd_big.opcode = (bp->b_flags & B_READ) ?
627 			    READ_BIG : WRITE_BIG;
628 			_lto4b(blkno, cmd_big.addr);
629 			_lto2b(nblks, cmd_big.length);
630 			cmdlen = sizeof(cmd_big);
631 			cmdp = (struct scsipi_generic *)&cmd_big;
632 		}
633 
634 		/* Instrumentation. */
635 		disk_busy(&sd->sc_dk);
636 
637 		/*
638 		 * Mark the disk dirty so that the cache will be
639 		 * flushed on close.
640 		 */
641 		if ((bp->b_flags & B_READ) == 0)
642 			sd->flags |= SDF_DIRTY;
643 
644 		/*
645 		 * Call the routine that chats with the adapter.
646 		 * Note: we cannot sleep as we may be an interrupt
647 		 */
648 		error = scsipi_command(sc_link, cmdp, cmdlen,
649 		    (u_char *)bp->b_data, bp->b_bcount,
650 		    SDRETRIES, 60000, bp, SCSI_NOSLEEP |
651 		    ((bp->b_flags & B_READ) ? SCSI_DATA_IN : SCSI_DATA_OUT));
652 		if (error) {
653 			disk_unbusy(&sd->sc_dk, 0);
654 			printf("%s: not queued, error %d\n",
655 			    sd->sc_dev.dv_xname, error);
656 		}
657 	}
658 }
659 
660 void
661 sddone(xs)
662 	struct scsipi_xfer *xs;
663 {
664 	struct sd_softc *sd = xs->sc_link->device_softc;
665 
666 	if (sd->flags & SDF_FLUSHING) {
667 		/* Flush completed, no longer dirty. */
668 		sd->flags &= ~(SDF_FLUSHING|SDF_DIRTY);
669 	}
670 
671 	if (xs->bp != NULL) {
672 		disk_unbusy(&sd->sc_dk, xs->bp->b_bcount - xs->bp->b_resid);
673 #if NRND > 0
674 		rnd_add_uint32(&sd->rnd_source, xs->bp->b_blkno);
675 #endif
676 	}
677 }
678 
679 void
680 sdminphys(bp)
681 	struct buf *bp;
682 {
683 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(bp->b_dev)];
684 	long max;
685 
686 	/*
687 	 * If the device is ancient, we want to make sure that
688 	 * the transfer fits into a 6-byte cdb.
689 	 *
690 	 * XXX Note that the SCSI-I spec says that 256-block transfers
691 	 * are allowed in a 6-byte read/write, and are specified
692 	 * by settng the "length" to 0.  However, we're conservative
693 	 * here, allowing only 255-block transfers in case an
694 	 * ancient device gets confused by length == 0.  A length of 0
695 	 * in a 10-byte read/write actually means 0 blocks.
696 	 */
697 	if (sd->flags & SDF_ANCIENT) {
698 		max = sd->sc_dk.dk_label->d_secsize * 0xff;
699 
700 		if (bp->b_bcount > max)
701 			bp->b_bcount = max;
702 	}
703 
704 	(*sd->sc_link->adapter->scsipi_minphys)(bp);
705 }
706 
707 int
708 sdread(dev, uio, ioflag)
709 	dev_t dev;
710 	struct uio *uio;
711 	int ioflag;
712 {
713 
714 	return (physio(sdstrategy, NULL, dev, B_READ, sdminphys, uio));
715 }
716 
717 int
718 sdwrite(dev, uio, ioflag)
719 	dev_t dev;
720 	struct uio *uio;
721 	int ioflag;
722 {
723 
724 	return (physio(sdstrategy, NULL, dev, B_WRITE, sdminphys, uio));
725 }
726 
727 /*
728  * Perform special action on behalf of the user
729  * Knows about the internals of this device
730  */
731 int
732 sdioctl(dev, cmd, addr, flag, p)
733 	dev_t dev;
734 	u_long cmd;
735 	caddr_t addr;
736 	int flag;
737 	struct proc *p;
738 {
739 	struct sd_softc *sd = sd_cd.cd_devs[SDUNIT(dev)];
740 	int error;
741 
742 	SC_DEBUG(sd->sc_link, SDEV_DB2, ("sdioctl 0x%lx ", cmd));
743 
744 	/*
745 	 * If the device is not valid.. abandon ship
746 	 */
747 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) == 0)
748 		return (EIO);
749 
750 	switch (cmd) {
751 	case DIOCGDINFO:
752 		*(struct disklabel *)addr = *(sd->sc_dk.dk_label);
753 		return (0);
754 
755 	case DIOCGPART:
756 		((struct partinfo *)addr)->disklab = sd->sc_dk.dk_label;
757 		((struct partinfo *)addr)->part =
758 		    &sd->sc_dk.dk_label->d_partitions[SDPART(dev)];
759 		return (0);
760 
761 	case DIOCWDINFO:
762 	case DIOCSDINFO:
763 		if ((flag & FWRITE) == 0)
764 			return (EBADF);
765 
766 		if ((error = sdlock(sd)) != 0)
767 			return (error);
768 		sd->flags |= SDF_LABELLING;
769 
770 		error = setdisklabel(sd->sc_dk.dk_label,
771 		    (struct disklabel *)addr, /*sd->sc_dk.dk_openmask : */0,
772 		    sd->sc_dk.dk_cpulabel);
773 		if (error == 0) {
774 			if (cmd == DIOCWDINFO)
775 				error = writedisklabel(SDLABELDEV(dev),
776 				    sdstrategy, sd->sc_dk.dk_label,
777 				    sd->sc_dk.dk_cpulabel);
778 		}
779 
780 		sd->flags &= ~SDF_LABELLING;
781 		sdunlock(sd);
782 		return (error);
783 
784 	case DIOCWLABEL:
785 		if ((flag & FWRITE) == 0)
786 			return (EBADF);
787 		if (*(int *)addr)
788 			sd->flags |= SDF_WLABEL;
789 		else
790 			sd->flags &= ~SDF_WLABEL;
791 		return (0);
792 
793 	case DIOCLOCK:
794 		return (scsipi_prevent(sd->sc_link,
795 		    (*(int *)addr) ? PR_PREVENT : PR_ALLOW, 0));
796 
797 	case DIOCEJECT:
798 		return ((sd->sc_link->flags & SDEV_REMOVABLE) == 0 ? ENOTTY :
799 		    scsipi_start(sd->sc_link, SSS_STOP|SSS_LOEJ, 0));
800 
801 	case DIOCGDEFLABEL:
802 		sdgetdefaultlabel(sd, (struct disklabel *)addr);
803 		return (0);
804 
805 	default:
806 		if (SDPART(dev) != RAW_PART)
807 			return (ENOTTY);
808 		return (scsipi_do_ioctl(sd->sc_link, dev, cmd, addr, flag, p));
809 	}
810 
811 #ifdef DIAGNOSTIC
812 	panic("sdioctl: impossible");
813 #endif
814 }
815 
816 void
817 sdgetdefaultlabel(sd, lp)
818 	struct sd_softc *sd;
819 	struct disklabel *lp;
820 {
821 
822 	bzero(lp, sizeof(struct disklabel));
823 
824 	lp->d_secsize = sd->params.blksize;
825 	lp->d_ntracks = sd->params.heads;
826 	lp->d_nsectors = sd->params.sectors;
827 	lp->d_ncylinders = sd->params.cyls;
828 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
829 
830 	switch (sd->sc_link->type) {
831 #if NSD_SCSIBUS > 0
832 	    case BUS_SCSI:
833 		lp->d_type = DTYPE_SCSI;
834 		break;
835 #endif
836 #if NSD_ATAPIBUS > 0
837 	    case BUS_ATAPI:
838 		lp->d_type = DTYPE_ATAPI;
839 		break;
840 #endif
841 	}
842 	strncpy(lp->d_typename, sd->name, 16);
843 	strncpy(lp->d_packname, "fictitious", 16);
844 	lp->d_secperunit = sd->params.disksize;
845 	lp->d_rpm = sd->params.rot_rate;
846 	lp->d_interleave = 1;
847 	lp->d_flags = 0;
848 
849 	lp->d_partitions[RAW_PART].p_offset = 0;
850 	lp->d_partitions[RAW_PART].p_size =
851 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
852 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
853 	lp->d_npartitions = RAW_PART + 1;
854 
855 	lp->d_magic = DISKMAGIC;
856 	lp->d_magic2 = DISKMAGIC;
857 	lp->d_checksum = dkcksum(lp);
858 }
859 
860 
861 /*
862  * Load the label information on the named device
863  */
864 void
865 sdgetdisklabel(sd)
866 	struct sd_softc *sd;
867 {
868 	struct disklabel *lp = sd->sc_dk.dk_label;
869 	char *errstring;
870 
871 	bzero(sd->sc_dk.dk_cpulabel, sizeof(struct cpu_disklabel));
872 
873 	sdgetdefaultlabel(sd, lp);
874 
875 	if (lp->d_secpercyl == 0) {
876 		lp->d_secpercyl = 100;
877 		/* as long as it's not 0 - readdisklabel divides by it (?) */
878 	}
879 
880 	/*
881 	 * Call the generic disklabel extraction routine
882 	 */
883 	errstring = readdisklabel(MAKESDDEV(0, sd->sc_dev.dv_unit, RAW_PART),
884 	    sdstrategy, lp, sd->sc_dk.dk_cpulabel);
885 	if (errstring) {
886 		printf("%s: %s\n", sd->sc_dev.dv_xname, errstring);
887 		return;
888 	}
889 }
890 
891 void
892 sd_shutdown(arg)
893 	void *arg;
894 {
895 	struct sd_softc *sd = arg;
896 
897 	/*
898 	 * If the disk cache needs to be flushed, and the disk supports
899 	 * it, flush it.  We're cold at this point, so we poll for
900 	 * completion.
901 	 */
902 	if ((sd->flags & SDF_DIRTY) != 0 && sd->sc_ops->sdo_flush != NULL)
903 		(*sd->sc_ops->sdo_flush)(sd, SCSI_AUTOCONF);
904 }
905 
906 /*
907  * Tell the device to map out a defective block
908  */
909 int
910 sd_reassign_blocks(sd, blkno)
911 	struct sd_softc *sd;
912 	u_long blkno;
913 {
914 	struct scsi_reassign_blocks scsipi_cmd;
915 	struct scsi_reassign_blocks_data rbdata;
916 
917 	bzero(&scsipi_cmd, sizeof(scsipi_cmd));
918 	bzero(&rbdata, sizeof(rbdata));
919 	scsipi_cmd.opcode = SCSI_REASSIGN_BLOCKS;
920 
921 	_lto2b(sizeof(rbdata.defect_descriptor[0]), rbdata.length);
922 	_lto4b(blkno, rbdata.defect_descriptor[0].dlbaddr);
923 
924 	return (scsipi_command(sd->sc_link,
925 	    (struct scsipi_generic *)&scsipi_cmd, sizeof(scsipi_cmd),
926 	    (u_char *)&rbdata, sizeof(rbdata), SDRETRIES, 5000, NULL,
927 	    SCSI_DATA_OUT));
928 }
929 
930 /*
931  * Check Errors
932  */
933 int
934 sd_interpret_sense(xs)
935 	struct scsipi_xfer *xs;
936 {
937 	struct scsipi_link *sc_link = xs->sc_link;
938 	struct scsipi_sense_data *sense = &xs->sense.scsi_sense;
939 	struct sd_softc *sd = sc_link->device_softc;
940 	int retval = SCSIRET_CONTINUE;
941 
942 	/*
943 	 * If the device is not open yet, let the generic code handle it.
944 	 */
945 	if ((sc_link->flags & SDEV_OPEN) == 0) {
946 		return (retval);
947 	}
948 
949 	/*
950 	 * If it isn't a extended or extended/deferred error, let
951 	 * the generic code handle it.
952 	 */
953 	if ((sense->error_code & SSD_ERRCODE) != 0x70 &&
954 	    (sense->error_code & SSD_ERRCODE) != 0x71) {	/* DEFFERRED */
955 		return (retval);
956 	}
957 
958 	if ((sense->flags & SSD_KEY) == SKEY_NOT_READY &&
959 	    sense->add_sense_code == 0x4) {
960 		if (sense->add_sense_code_qual == 0x01)	{
961 			printf("%s: ..is spinning up...waiting\n",
962 			    sd->sc_dev.dv_xname);
963 			/*
964 			 * I really need a sdrestart function I can call here.
965 			 */
966 			delay(1000000 * 5);	/* 5 seconds */
967 			retval = SCSIRET_RETRY;
968 		} else if ((sense->add_sense_code_qual == 0x2) &&
969 		    (sd->sc_link->quirks & SDEV_NOSTARTUNIT) == 0) {
970 			if (sd->sc_link->flags & SDEV_REMOVABLE) {
971 				printf("%s: removable disk stopped- not "
972 				    "restarting\n", sd->sc_dev.dv_xname);
973 				retval = EIO;
974 			} else {
975 				printf("%s: respinning up disk\n",
976 				    sd->sc_dev.dv_xname);
977 				retval = scsipi_start(sd->sc_link, SSS_START,
978 				    SCSI_URGENT | SCSI_NOSLEEP);
979 				if (retval != 0) {
980 					printf("%s: respin of disk failed-%d\n",
981 					    sd->sc_dev.dv_xname, retval);
982 					retval = EIO;
983 				} else {
984 					retval = SCSIRET_RETRY;
985 				}
986 			}
987 		}
988 	}
989 	return (retval);
990 }
991 
992 
993 int
994 sdsize(dev)
995 	dev_t dev;
996 {
997 	struct sd_softc *sd;
998 	int part, unit, omask;
999 	int size;
1000 
1001 	unit = SDUNIT(dev);
1002 	if (unit >= sd_cd.cd_ndevs)
1003 		return (-1);
1004 	sd = sd_cd.cd_devs[unit];
1005 	if (sd == NULL)
1006 		return (-1);
1007 
1008 	part = SDPART(dev);
1009 	omask = sd->sc_dk.dk_openmask & (1 << part);
1010 
1011 	if (omask == 0 && sdopen(dev, 0, S_IFBLK, NULL) != 0)
1012 		return (-1);
1013 	if (sd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1014 		size = -1;
1015 	else
1016 		size = sd->sc_dk.dk_label->d_partitions[part].p_size *
1017 		    (sd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1018 	if (omask == 0 && sdclose(dev, 0, S_IFBLK, NULL) != 0)
1019 		return (-1);
1020 	return (size);
1021 }
1022 
1023 #ifndef __BDEVSW_DUMP_OLD_TYPE
1024 /* #define SD_DUMP_NOT_TRUSTED if you just want to watch */
1025 static struct scsipi_xfer sx;
1026 static int sddoingadump;
1027 
1028 /*
1029  * dump all of physical memory into the partition specified, starting
1030  * at offset 'dumplo' into the partition.
1031  */
1032 int
1033 sddump(dev, blkno, va, size)
1034 	dev_t dev;
1035 	daddr_t blkno;
1036 	caddr_t va;
1037 	size_t size;
1038 {
1039 	struct sd_softc *sd;	/* disk unit to do the I/O */
1040 	struct disklabel *lp;	/* disk's disklabel */
1041 	int	unit, part;
1042 	int	sectorsize;	/* size of a disk sector */
1043 	int	nsects;		/* number of sectors in partition */
1044 	int	sectoff;	/* sector offset of partition */
1045 	int	totwrt;		/* total number of sectors left to write */
1046 	int	nwrt;		/* current number of sectors to write */
1047 	struct scsipi_rw_big cmd;	/* write command */
1048 	struct scsipi_xfer *xs;	/* ... convenience */
1049 	int	retval;
1050 
1051 	/* Check if recursive dump; if so, punt. */
1052 	if (sddoingadump)
1053 		return (EFAULT);
1054 
1055 	/* Mark as active early. */
1056 	sddoingadump = 1;
1057 
1058 	unit = SDUNIT(dev);	/* Decompose unit & partition. */
1059 	part = SDPART(dev);
1060 
1061 	/* Check for acceptable drive number. */
1062 	if (unit >= sd_cd.cd_ndevs || (sd = sd_cd.cd_devs[unit]) == NULL)
1063 		return (ENXIO);
1064 
1065 	/*
1066 	 * XXX Can't do this check, since the media might have been
1067 	 * XXX marked `invalid' by successful unmounting of all
1068 	 * XXX filesystems.
1069 	 */
1070 #if 0
1071 	/* Make sure it was initialized. */
1072 	if ((sd->sc_link->flags & SDEV_MEDIA_LOADED) != SDEV_MEDIA_LOADED)
1073 		return (ENXIO);
1074 #endif
1075 
1076 	/* Convert to disk sectors.  Request must be a multiple of size. */
1077 	lp = sd->sc_dk.dk_label;
1078 	sectorsize = lp->d_secsize;
1079 	if ((size % sectorsize) != 0)
1080 		return (EFAULT);
1081 	totwrt = size / sectorsize;
1082 	blkno = dbtob(blkno) / sectorsize;	/* blkno in DEV_BSIZE units */
1083 
1084 	nsects = lp->d_partitions[part].p_size;
1085 	sectoff = lp->d_partitions[part].p_offset;
1086 
1087 	/* Check transfer bounds against partition size. */
1088 	if ((blkno < 0) || ((blkno + totwrt) > nsects))
1089 		return (EINVAL);
1090 
1091 	/* Offset block number to start of partition. */
1092 	blkno += sectoff;
1093 
1094 	xs = &sx;
1095 
1096 	while (totwrt > 0) {
1097 		nwrt = totwrt;		/* XXX */
1098 #ifndef	SD_DUMP_NOT_TRUSTED
1099 		/*
1100 		 *  Fill out the scsi command
1101 		 */
1102 		bzero(&cmd, sizeof(cmd));
1103 		cmd.opcode = WRITE_BIG;
1104 		_lto4b(blkno, cmd.addr);
1105 		_lto2b(nwrt, cmd.length);
1106 		/*
1107 		 * Fill out the scsipi_xfer structure
1108 		 *    Note: we cannot sleep as we may be an interrupt
1109 		 * don't use scsipi_command() as it may want to wait
1110 		 * for an xs.
1111 		 */
1112 		bzero(xs, sizeof(sx));
1113 		xs->flags |= SCSI_AUTOCONF | INUSE | SCSI_DATA_OUT;
1114 		xs->sc_link = sd->sc_link;
1115 		xs->retries = SDRETRIES;
1116 		xs->timeout = 10000;	/* 10000 millisecs for a disk ! */
1117 		xs->cmd = (struct scsipi_generic *)&cmd;
1118 		xs->cmdlen = sizeof(cmd);
1119 		xs->resid = nwrt * sectorsize;
1120 		xs->error = XS_NOERROR;
1121 		xs->bp = 0;
1122 		xs->data = va;
1123 		xs->datalen = nwrt * sectorsize;
1124 
1125 		/*
1126 		 * Pass all this info to the scsi driver.
1127 		 */
1128 		retval = scsipi_command_direct(xs);
1129 		if (retval != COMPLETE)
1130 			return (ENXIO);
1131 #else	/* SD_DUMP_NOT_TRUSTED */
1132 		/* Let's just talk about this first... */
1133 		printf("sd%d: dump addr 0x%x, blk %d\n", unit, va, blkno);
1134 		delay(500 * 1000);	/* half a second */
1135 #endif	/* SD_DUMP_NOT_TRUSTED */
1136 
1137 		/* update block count */
1138 		totwrt -= nwrt;
1139 		blkno += nwrt;
1140 		va += sectorsize * nwrt;
1141 	}
1142 	sddoingadump = 0;
1143 	return (0);
1144 }
1145 #else	/* __BDEVSW_DUMP_NEW_TYPE */
1146 int
1147 sddump(dev, blkno, va, size)
1148 	dev_t dev;
1149 	daddr_t blkno;
1150 	caddr_t va;
1151 	size_t size;
1152 {
1153 
1154 	/* Not implemented. */
1155 	return (ENXIO);
1156 }
1157 #endif	/* __BDEVSW_DUMP_NEW_TYPE */
1158