xref: /netbsd-src/sys/kern/subr_disk.c (revision 7e30e94394d0994ab9534f68a8f91665045c91ce)
1 /*	$NetBSD: subr_disk.c,v 1.118 2017/03/05 23:07:12 mlelstv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1999, 2000, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: subr_disk.c,v 1.118 2017/03/05 23:07:12 mlelstv Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/kernel.h>
74 #include <sys/kmem.h>
75 #include <sys/buf.h>
76 #include <sys/fcntl.h>
77 #include <sys/syslog.h>
78 #include <sys/disklabel.h>
79 #include <sys/disk.h>
80 #include <sys/sysctl.h>
81 #include <lib/libkern/libkern.h>
82 
83 /*
84  * Compute checksum for disk label.
85  */
86 u_int
87 dkcksum(struct disklabel *lp)
88 {
89 
90 	return dkcksum_sized(lp, lp->d_npartitions);
91 }
92 
93 u_int
94 dkcksum_sized(struct disklabel *lp, size_t npartitions)
95 {
96 	uint16_t *start, *end;
97 	uint16_t sum = 0;
98 
99 	start = (uint16_t *)lp;
100 	end = (uint16_t *)&lp->d_partitions[npartitions];
101 	while (start < end)
102 		sum ^= *start++;
103 	return sum;
104 }
105 
106 /*
107  * Disk error is the preface to plaintive error messages
108  * about failing disk transfers.  It prints messages of the form
109 
110 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
111 
112  * if the offset of the error in the transfer and a disk label
113  * are both available.  blkdone should be -1 if the position of the error
114  * is unknown; the disklabel pointer may be null from drivers that have not
115  * been converted to use them.  The message is printed with printf
116  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
117  * The message should be completed (with at least a newline) with printf
118  * or addlog, respectively.  There is no trailing space.
119  */
120 #ifndef PRIdaddr
121 #define PRIdaddr PRId64
122 #endif
123 void
124 diskerr(const struct buf *bp, const char *dname, const char *what, int pri,
125     int blkdone, const struct disklabel *lp)
126 {
127 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
128 	void (*pr)(const char *, ...) __printflike(1, 2);
129 	char partname = 'a' + part;
130 	daddr_t sn;
131 
132 	if (/*CONSTCOND*/0)
133 		/* Compiler will error this is the format is wrong... */
134 		printf("%" PRIdaddr, bp->b_blkno);
135 
136 	if (pri != LOG_PRINTF) {
137 		static const char fmt[] = "";
138 		log(pri, fmt);
139 		pr = addlog;
140 	} else
141 		pr = printf;
142 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
143 	    bp->b_flags & B_READ ? "read" : "writ");
144 	sn = bp->b_blkno;
145 	if (bp->b_bcount <= DEV_BSIZE)
146 		(*pr)("%" PRIdaddr, sn);
147 	else {
148 		if (blkdone >= 0) {
149 			sn += blkdone;
150 			(*pr)("%" PRIdaddr " of ", sn);
151 		}
152 		(*pr)("%" PRIdaddr "-%" PRIdaddr "", bp->b_blkno,
153 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
154 	}
155 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
156 		sn += lp->d_partitions[part].p_offset;
157 		(*pr)(" (%s%d bn %" PRIdaddr "; cn %" PRIdaddr "",
158 		    dname, unit, sn, sn / lp->d_secpercyl);
159 		sn %= lp->d_secpercyl;
160 		(*pr)(" tn %" PRIdaddr " sn %" PRIdaddr ")",
161 		    sn / lp->d_nsectors, sn % lp->d_nsectors);
162 	}
163 }
164 
165 /*
166  * Searches the iostatlist for the disk corresponding to the
167  * name provided.
168  */
169 struct disk *
170 disk_find(const char *name)
171 {
172 	struct io_stats *stat;
173 
174 	stat = iostat_find(name);
175 
176 	if ((stat != NULL) && (stat->io_type == IOSTAT_DISK))
177 		return stat->io_parent;
178 
179 	return (NULL);
180 }
181 
182 void
183 disk_init(struct disk *diskp, const char *name, const struct dkdriver *driver)
184 {
185 	u_int blocksize = DEV_BSIZE;
186 
187 	/*
188 	 * Initialize the wedge-related locks and other fields.
189 	 */
190 	mutex_init(&diskp->dk_rawlock, MUTEX_DEFAULT, IPL_NONE);
191 	mutex_init(&diskp->dk_openlock, MUTEX_DEFAULT, IPL_NONE);
192 	LIST_INIT(&diskp->dk_wedges);
193 	diskp->dk_nwedges = 0;
194 	diskp->dk_labelsector = LABELSECTOR;
195 	diskp->dk_blkshift = DK_BSIZE2BLKSHIFT(blocksize);
196 	diskp->dk_byteshift = DK_BSIZE2BYTESHIFT(blocksize);
197 	diskp->dk_name = name;
198 	diskp->dk_driver = driver;
199 }
200 
201 /*
202  * Attach a disk.
203  */
204 void
205 disk_attach(struct disk *diskp)
206 {
207 
208 	/*
209 	 * Allocate and initialize the disklabel structures.
210 	 */
211 	diskp->dk_label = kmem_zalloc(sizeof(struct disklabel), KM_SLEEP);
212 	diskp->dk_cpulabel = kmem_zalloc(sizeof(struct cpu_disklabel),
213 	    KM_SLEEP);
214 	if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
215 		panic("disk_attach: can't allocate storage for disklabel");
216 
217 	/*
218 	 * Set up the stats collection.
219 	 */
220 	diskp->dk_stats = iostat_alloc(IOSTAT_DISK, diskp, diskp->dk_name);
221 }
222 
223 int
224 disk_begindetach(struct disk *dk, int (*lastclose)(device_t),
225     device_t self, int flags)
226 {
227 	int rc;
228 
229 	rc = 0;
230 	mutex_enter(&dk->dk_openlock);
231 	if (dk->dk_openmask == 0)
232 		;	/* nothing to do */
233 	else if ((flags & DETACH_FORCE) == 0)
234 		rc = EBUSY;
235 	else if (lastclose != NULL)
236 		rc = (*lastclose)(self);
237 	mutex_exit(&dk->dk_openlock);
238 
239 	return rc;
240 }
241 
242 /*
243  * Detach a disk.
244  */
245 void
246 disk_detach(struct disk *diskp)
247 {
248 
249 	/*
250 	 * Remove from the drivelist.
251 	 */
252 	iostat_free(diskp->dk_stats);
253 
254 	/*
255 	 * Release the disk-info dictionary.
256 	 */
257 	if (diskp->dk_info) {
258 		prop_object_release(diskp->dk_info);
259 		diskp->dk_info = NULL;
260 	}
261 
262 	/*
263 	 * Free the space used by the disklabel structures.
264 	 */
265 	kmem_free(diskp->dk_label, sizeof(*diskp->dk_label));
266 	kmem_free(diskp->dk_cpulabel, sizeof(*diskp->dk_cpulabel));
267 }
268 
269 void
270 disk_destroy(struct disk *diskp)
271 {
272 
273 	mutex_destroy(&diskp->dk_openlock);
274 	mutex_destroy(&diskp->dk_rawlock);
275 }
276 
277 /*
278  * Mark the disk as having work queued for metrics collection.
279  */
280 void
281 disk_wait(struct disk *diskp)
282 {
283 
284 	iostat_wait(diskp->dk_stats);
285 }
286 
287 /*
288  * Mark the disk as busy for metrics collection.
289  */
290 void
291 disk_busy(struct disk *diskp)
292 {
293 
294 	iostat_busy(diskp->dk_stats);
295 }
296 
297 /*
298  * Finished disk operations, gather metrics.
299  */
300 void
301 disk_unbusy(struct disk *diskp, long bcount, int read)
302 {
303 
304 	iostat_unbusy(diskp->dk_stats, bcount, read);
305 }
306 
307 /*
308  * Return true if disk has an I/O operation in flight.
309  */
310 bool
311 disk_isbusy(struct disk *diskp)
312 {
313 
314 	return iostat_isbusy(diskp->dk_stats);
315 }
316 
317 /*
318  * Bounds checking against the media size, used for the raw partition.
319  * secsize, mediasize and b_blkno must all be the same units.
320  * Possibly this has to be DEV_BSIZE (512).
321  */
322 int
323 bounds_check_with_mediasize(struct buf *bp, int secsize, uint64_t mediasize)
324 {
325 	int64_t sz;
326 
327 	if (bp->b_blkno < 0) {
328 		/* Reject negative offsets immediately. */
329 		bp->b_error = EINVAL;
330 		return 0;
331 	}
332 
333 	sz = howmany((int64_t)bp->b_bcount, secsize);
334 
335 	/*
336 	 * bp->b_bcount is a 32-bit value, and we rejected a negative
337 	 * bp->b_blkno already, so "bp->b_blkno + sz" cannot overflow.
338 	 */
339 
340 	if (bp->b_blkno + sz > mediasize) {
341 		sz = mediasize - bp->b_blkno;
342 		if (sz == 0) {
343 			/* If exactly at end of disk, return EOF. */
344 			bp->b_resid = bp->b_bcount;
345 			return 0;
346 		}
347 		if (sz < 0) {
348 			/* If past end of disk, return EINVAL. */
349 			bp->b_error = EINVAL;
350 			return 0;
351 		}
352 		/* Otherwise, truncate request. */
353 		bp->b_bcount = sz * secsize;
354 	}
355 
356 	return 1;
357 }
358 
359 /*
360  * Determine the size of the transfer, and make sure it is
361  * within the boundaries of the partition. Adjust transfer
362  * if needed, and signal errors or early completion.
363  */
364 int
365 bounds_check_with_label(struct disk *dk, struct buf *bp, int wlabel)
366 {
367 	struct disklabel *lp = dk->dk_label;
368 	struct partition *p = lp->d_partitions + DISKPART(bp->b_dev);
369 	uint64_t p_size, p_offset, labelsector;
370 	int64_t sz;
371 
372 	if (bp->b_blkno < 0) {
373 		/* Reject negative offsets immediately. */
374 		bp->b_error = EINVAL;
375 		return -1;
376 	}
377 
378 	/* Protect against division by zero. XXX: Should never happen?!?! */
379 	if (lp->d_secpercyl == 0) {
380 		bp->b_error = EINVAL;
381 		return -1;
382 	}
383 
384 	p_size = (uint64_t)p->p_size << dk->dk_blkshift;
385 	p_offset = (uint64_t)p->p_offset << dk->dk_blkshift;
386 #if RAW_PART == 3
387 	labelsector = lp->d_partitions[2].p_offset;
388 #else
389 	labelsector = lp->d_partitions[RAW_PART].p_offset;
390 #endif
391 	labelsector = (labelsector + dk->dk_labelsector) << dk->dk_blkshift;
392 
393 	sz = howmany((int64_t)bp->b_bcount, DEV_BSIZE);
394 
395 	/*
396 	 * bp->b_bcount is a 32-bit value, and we rejected a negative
397 	 * bp->b_blkno already, so "bp->b_blkno + sz" cannot overflow.
398 	 */
399 
400 	if (bp->b_blkno + sz > p_size) {
401 		sz = p_size - bp->b_blkno;
402 		if (sz == 0) {
403 			/* If exactly at end of disk, return EOF. */
404 			bp->b_resid = bp->b_bcount;
405 			return 0;
406 		}
407 		if (sz < 0) {
408 			/* If past end of disk, return EINVAL. */
409 			bp->b_error = EINVAL;
410 			return -1;
411 		}
412 		/* Otherwise, truncate request. */
413 		bp->b_bcount = sz << DEV_BSHIFT;
414 	}
415 
416 	/* Overwriting disk label? */
417 	if (bp->b_blkno + p_offset <= labelsector &&
418 	    bp->b_blkno + p_offset + sz > labelsector &&
419 	    (bp->b_flags & B_READ) == 0 && !wlabel) {
420 		bp->b_error = EROFS;
421 		return -1;
422 	}
423 
424 	/* calculate cylinder for disksort to order transfers with */
425 	bp->b_cylinder = (bp->b_blkno + p->p_offset) /
426 	    (lp->d_secsize / DEV_BSIZE) / lp->d_secpercyl;
427 	return 1;
428 }
429 
430 int
431 disk_read_sectors(void (*strat)(struct buf *), const struct disklabel *lp,
432     struct buf *bp, unsigned int sector, int count)
433 {
434 	bp->b_blkno = btodb((off_t)sector * lp->d_secsize);
435 	bp->b_bcount = count * lp->d_secsize;
436 	bp->b_flags = (bp->b_flags & ~B_WRITE) | B_READ;
437 	bp->b_oflags &= ~BO_DONE;
438 	bp->b_cylinder = sector / lp->d_secpercyl;
439 	(*strat)(bp);
440 	return biowait(bp);
441 }
442 
443 const char *
444 convertdisklabel(struct disklabel *lp, void (*strat)(struct buf *),
445     struct buf *bp, uint32_t secperunit)
446 {
447 	struct partition rp, *altp, *p;
448 	int geom_ok;
449 	const char *str;
450 
451 	memset(&rp, 0, sizeof(rp));
452 	rp.p_size = secperunit;
453 	rp.p_fstype = FS_UNUSED;
454 
455 	/* If we can seek to d_secperunit - 1, believe the disk geometry. */
456 	if (secperunit != 0 &&
457 	    disk_read_sectors(strat, lp, bp, secperunit - 1, 1) == 0)
458 		geom_ok = 1;
459 	else
460 		geom_ok = 0;
461 
462 #if 0
463 	printf("%s: secperunit (%" PRIu32 ") %s\n", __func__,
464 	    secperunit, geom_ok ? "ok" : "not ok");
465 #endif
466 
467 	p = &lp->d_partitions[RAW_PART];
468 	if (RAW_PART == 'c' - 'a')
469 		altp = &lp->d_partitions['d' - 'a'];
470 	else
471 		altp = &lp->d_partitions['c' - 'a'];
472 
473 	if (lp->d_npartitions > RAW_PART && p->p_offset == 0 && p->p_size != 0)
474 		return NULL;	/* already a raw partition */
475 	else if (lp->d_npartitions > MAX('c', 'd') - 'a' &&
476 		 altp->p_offset == 0 && altp->p_size != 0) {
477 		/* alternate partition ('c' or 'd') is suitable for raw slot,
478 		 * swap with 'd' or 'c'.
479 		 */
480 		rp = *p;
481 		*p = *altp;
482 		*altp = rp;
483 		return NULL;
484 	} else if (lp->d_npartitions <= RAW_PART &&
485 	           lp->d_npartitions > 'c' - 'a') {
486 		/* No raw partition is present, but the alternate is present.
487 		 * Copy alternate to raw partition.
488 		 */
489 		lp->d_npartitions = RAW_PART + 1;
490 		*p = *altp;
491 		return NULL;
492 	} else if (!geom_ok)
493 		str = "no raw partition and disk reports bad geometry";
494 	else if (lp->d_npartitions <= RAW_PART) {
495 		memset(&lp->d_partitions[lp->d_npartitions], 0,
496 		    sizeof(struct partition) * (RAW_PART - lp->d_npartitions));
497 		*p = rp;
498 		lp->d_npartitions = RAW_PART + 1;
499 		return NULL;
500 	} else if (lp->d_npartitions < MAXPARTITIONS) {
501 		memmove(p + 1, p,
502 		    sizeof(struct partition) * (lp->d_npartitions - RAW_PART));
503 		*p = rp;
504 		lp->d_npartitions++;
505 		return NULL;
506 	} else
507 		str = "no raw partition and partition table is full";
508 #ifdef DIAGNOSTIC
509 	printf("Bad partition: %s\n", str);
510 	printf("type = %u, subtype = %u, typename = %s\n",
511 	    lp->d_type, lp->d_subtype, lp->d_typename);
512 	printf("secsize = %u, nsectors = %u, ntracks = %u\n",
513 	    lp->d_secsize, lp->d_nsectors, lp->d_ntracks);
514 	printf("ncylinders = %u, secpercyl = %u, secperunit = %u\n",
515 	    lp->d_ncylinders, lp->d_secpercyl, lp->d_secperunit);
516 	printf("npartitions = %u\n", lp->d_npartitions);
517 
518 	for (size_t i = 0; i < MIN(lp->d_npartitions, MAXPARTITIONS); i++) {
519 		p = &lp->d_partitions[i];
520 		printf("\t%c: offset = %u size = %u fstype = %u\n",
521 		    (char)(i + 'a'), p->p_offset, p->p_size, p->p_fstype);
522 	}
523 #endif
524 	return str;
525 }
526 
527 /*
528  * disk_ioctl --
529  *	Generic disk ioctl handling.
530  */
531 int
532 disk_ioctl(struct disk *dk, dev_t dev, u_long cmd, void *data, int flag,
533     struct lwp *l)
534 {
535 	struct dkwedge_info *dkw;
536 	struct partinfo *pi;
537 	struct partition *dp;
538 #ifdef __HAVE_OLD_DISKLABEL
539 	struct disklabel newlabel;
540 #endif
541 
542 	switch (cmd) {
543 	case DIOCGDISKINFO:
544 		if (dk->dk_info == NULL)
545 			return ENOTSUP;
546 		return prop_dictionary_copyout_ioctl(data, cmd, dk->dk_info);
547 
548 	case DIOCGSECTORSIZE:
549 		*(u_int *)data = dk->dk_geom.dg_secsize;
550 		return 0;
551 
552 	case DIOCGMEDIASIZE:
553 		*(off_t *)data = (off_t)dk->dk_geom.dg_secsize *
554 		    dk->dk_geom.dg_secperunit;
555 		return 0;
556 	default:
557 		break;
558 	}
559 
560 	if (dev == NODEV)
561 		return EPASSTHROUGH;
562 
563 	/* The following should be moved to dk_ioctl */
564 	switch (cmd) {
565 	case DIOCGDINFO:
566 		if (dk->dk_label == NULL)
567 			return EBUSY;
568 		memcpy(data, dk->dk_label, sizeof (*dk->dk_label));
569 		return 0;
570 
571 #ifdef __HAVE_OLD_DISKLABEL
572 	case ODIOCGDINFO:
573 		if (dk->dk_label == NULL)
574 			return EBUSY;
575 		memcpy(&newlabel, dk->dk_label, sizeof(newlabel));
576 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
577 			return ENOTTY;
578 		memcpy(data, &newlabel, sizeof(struct olddisklabel));
579 		return 0;
580 #endif
581 
582 	case DIOCGPARTINFO:
583 		pi = data;
584 		memset(pi, 0, sizeof(*pi));
585 		pi->pi_secsize = dk->dk_geom.dg_secsize;
586 		pi->pi_bsize = MAX(BLKDEV_IOSIZE, pi->pi_secsize);
587 
588 		if (DISKPART(dev) == RAW_PART) {
589 			pi->pi_size = dk->dk_geom.dg_secperunit;
590 			return 0;
591 		}
592 
593 		if (dk->dk_label == NULL)
594 			return EBUSY;
595 
596 		dp = &dk->dk_label->d_partitions[DISKPART(dev)];
597 		pi->pi_offset = dp->p_offset;
598 		pi->pi_size = dp->p_size;
599 
600 		pi->pi_fstype = dp->p_fstype;
601 		pi->pi_frag = dp->p_frag;
602 		pi->pi_fsize = dp->p_fsize;
603 		pi->pi_cpg = dp->p_cpg;
604 
605 		/*
606 		 * dholland 20130616: XXX this logic should not be
607 		 * here. It is here because the old buffer cache
608 		 * demands that all accesses to the same blocks need
609 		 * to be the same size; but it only works for FFS and
610 		 * nowadays I think it'll fail silently if the size
611 		 * info in the disklabel is wrong. (Or missing.) The
612 		 * buffer cache needs to be smarter; or failing that
613 		 * we need a reliable way here to get the right block
614 		 * size; or a reliable way to guarantee that (a) the
615 		 * fs is not mounted when we get here and (b) any
616 		 * buffers generated here will get purged when the fs
617 		 * does get mounted.
618 		 */
619 		if (dp->p_fstype == FS_BSDFFS &&
620 		    dp->p_frag != 0 && dp->p_fsize != 0)
621 			pi->pi_bsize = dp->p_frag * dp->p_fsize;
622 		return 0;
623 
624 	case DIOCAWEDGE:
625 		if ((flag & FWRITE) == 0)
626 			return EBADF;
627 
628 		dkw = data;
629 		strlcpy(dkw->dkw_parent, dk->dk_name, sizeof(dkw->dkw_parent));
630 		return dkwedge_add(dkw);
631 
632 	case DIOCDWEDGE:
633 		if ((flag & FWRITE) == 0)
634 			return EBADF;
635 
636 		dkw = data;
637 		strlcpy(dkw->dkw_parent, dk->dk_name, sizeof(dkw->dkw_parent));
638 		return dkwedge_del(dkw);
639 
640 	case DIOCLWEDGES:
641 		return dkwedge_list(dk, data, l);
642 
643 	case DIOCMWEDGES:
644 		if ((flag & FWRITE) == 0)
645 			return EBADF;
646 
647 		dkwedge_discover(dk);
648 		return 0;
649 
650 	default:
651 		return EPASSTHROUGH;
652 	}
653 }
654 
655 void
656 disk_set_info(device_t dev, struct disk *dk, const char *type)
657 {
658 	struct disk_geom *dg = &dk->dk_geom;
659 
660 	if (dg->dg_secsize == 0) {
661 #ifdef DIAGNOSTIC
662 		printf("%s: fixing 0 sector size\n", dk->dk_name);
663 #endif
664 		dg->dg_secsize = DEV_BSIZE;
665 	}
666 
667 	dk->dk_blkshift = DK_BSIZE2BLKSHIFT(dg->dg_secsize);
668 	dk->dk_byteshift = DK_BSIZE2BYTESHIFT(dg->dg_secsize);
669 
670 	if (dg->dg_secperunit == 0 && dg->dg_ncylinders == 0) {
671 #ifdef DIAGNOSTIC
672 		printf("%s: secperunit and ncylinders are zero\n", dk->dk_name);
673 #endif
674 		return;
675 	}
676 
677 	if (dg->dg_secperunit == 0) {
678 		if (dg->dg_nsectors == 0 || dg->dg_ntracks == 0) {
679 #ifdef DIAGNOSTIC
680 			printf("%s: secperunit and (sectors or tracks) "
681 			    "are zero\n", dk->dk_name);
682 #endif
683 			return;
684 		}
685 		dg->dg_secperunit = (int64_t) dg->dg_nsectors *
686 		    dg->dg_ntracks * dg->dg_ncylinders;
687 	}
688 
689 	if (dg->dg_ncylinders == 0) {
690 		if (dg->dg_ntracks && dg->dg_nsectors)
691 			dg->dg_ncylinders = dg->dg_secperunit /
692 			    (dg->dg_ntracks * dg->dg_nsectors);
693 	}
694 
695 	prop_dictionary_t disk_info, odisk_info, geom;
696 
697 	disk_info = prop_dictionary_create();
698 	geom = prop_dictionary_create();
699 
700 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
701 	    dg->dg_secperunit);
702 
703 	prop_dictionary_set_uint32(geom, "sector-size", dg->dg_secsize);
704 
705 	if (dg->dg_nsectors)
706 		prop_dictionary_set_uint16(geom, "sectors-per-track",
707 		    dg->dg_nsectors);
708 
709 	if (dg->dg_ntracks)
710 		prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
711 		    dg->dg_ntracks);
712 
713 	if (dg->dg_ncylinders)
714 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
715 		    dg->dg_ncylinders);
716 
717 	prop_dictionary_set(disk_info, "geometry", geom);
718 
719 	if (type)
720 		prop_dictionary_set_cstring_nocopy(disk_info, "type", type);
721 
722 	prop_object_release(geom);
723 
724 	odisk_info = dk->dk_info;
725 	dk->dk_info = disk_info;
726 
727 	if (dev)
728 		prop_dictionary_set(device_properties(dev), "disk-info",
729 		    disk_info);
730 
731 	/*
732 	 * Don't release disk_info here; we keep a reference to it.
733 	 * disk_detach() will release it when we go away.
734 	 */
735 	if (odisk_info)
736 		prop_object_release(odisk_info);
737 }
738