xref: /netbsd-src/sys/kern/subr_disk.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*	$NetBSD: subr_disk.c,v 1.122 2018/03/07 21:13:24 kre 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 1.122 2018/03/07 21:13:24 kre 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 if 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 
215 	/*
216 	 * Set up the stats collection.
217 	 */
218 	diskp->dk_stats = iostat_alloc(IOSTAT_DISK, diskp, diskp->dk_name);
219 }
220 
221 int
222 disk_begindetach(struct disk *dk, int (*lastclose)(device_t),
223     device_t self, int flags)
224 {
225 	int rc;
226 
227 	rc = 0;
228 	mutex_enter(&dk->dk_openlock);
229 	if (dk->dk_openmask == 0)
230 		;	/* nothing to do */
231 	else if ((flags & DETACH_FORCE) == 0)
232 		rc = EBUSY;
233 	else if (lastclose != NULL)
234 		rc = (*lastclose)(self);
235 	mutex_exit(&dk->dk_openlock);
236 
237 	return rc;
238 }
239 
240 /*
241  * Detach a disk.
242  */
243 void
244 disk_detach(struct disk *diskp)
245 {
246 
247 	/*
248 	 * Remove from the drivelist.
249 	 */
250 	iostat_free(diskp->dk_stats);
251 
252 	/*
253 	 * Release the disk-info dictionary.
254 	 */
255 	if (diskp->dk_info) {
256 		prop_object_release(diskp->dk_info);
257 		diskp->dk_info = NULL;
258 	}
259 
260 	/*
261 	 * Free the space used by the disklabel structures.
262 	 */
263 	kmem_free(diskp->dk_label, sizeof(*diskp->dk_label));
264 	kmem_free(diskp->dk_cpulabel, sizeof(*diskp->dk_cpulabel));
265 }
266 
267 void
268 disk_destroy(struct disk *diskp)
269 {
270 
271 	mutex_destroy(&diskp->dk_openlock);
272 	mutex_destroy(&diskp->dk_rawlock);
273 }
274 
275 /*
276  * Mark the disk as having work queued for metrics collection.
277  */
278 void
279 disk_wait(struct disk *diskp)
280 {
281 
282 	iostat_wait(diskp->dk_stats);
283 }
284 
285 /*
286  * Mark the disk as busy for metrics collection.
287  */
288 void
289 disk_busy(struct disk *diskp)
290 {
291 
292 	iostat_busy(diskp->dk_stats);
293 }
294 
295 /*
296  * Finished disk operations, gather metrics.
297  */
298 void
299 disk_unbusy(struct disk *diskp, long bcount, int read)
300 {
301 
302 	iostat_unbusy(diskp->dk_stats, bcount, read);
303 }
304 
305 /*
306  * Return true if disk has an I/O operation in flight.
307  */
308 bool
309 disk_isbusy(struct disk *diskp)
310 {
311 
312 	return iostat_isbusy(diskp->dk_stats);
313 }
314 
315 /*
316  * Bounds checking against the media size, used for the raw partition.
317  * secsize, mediasize and b_blkno must all be the same units.
318  * Possibly this has to be DEV_BSIZE (512).
319  */
320 int
321 bounds_check_with_mediasize(struct buf *bp, int secsize, uint64_t mediasize)
322 {
323 	int64_t sz;
324 
325 	if (bp->b_blkno < 0) {
326 		/* Reject negative offsets immediately. */
327 		bp->b_error = EINVAL;
328 		return 0;
329 	}
330 
331 	sz = howmany((int64_t)bp->b_bcount, secsize);
332 
333 	/*
334 	 * bp->b_bcount is a 32-bit value, and we rejected a negative
335 	 * bp->b_blkno already, so "bp->b_blkno + sz" cannot overflow.
336 	 */
337 
338 	if (bp->b_blkno + sz > mediasize) {
339 		sz = mediasize - bp->b_blkno;
340 		if (sz == 0) {
341 			/* If exactly at end of disk, return EOF. */
342 			bp->b_resid = bp->b_bcount;
343 			return 0;
344 		}
345 		if (sz < 0) {
346 			/* If past end of disk, return EINVAL. */
347 			bp->b_error = EINVAL;
348 			return 0;
349 		}
350 		/* Otherwise, truncate request. */
351 		bp->b_bcount = sz * secsize;
352 	}
353 
354 	return 1;
355 }
356 
357 /*
358  * Determine the size of the transfer, and make sure it is
359  * within the boundaries of the partition. Adjust transfer
360  * if needed, and signal errors or early completion.
361  */
362 int
363 bounds_check_with_label(struct disk *dk, struct buf *bp, int wlabel)
364 {
365 	struct disklabel *lp = dk->dk_label;
366 	struct partition *p = lp->d_partitions + DISKPART(bp->b_dev);
367 	uint64_t p_size, p_offset, labelsector;
368 	int64_t sz;
369 
370 	if (bp->b_blkno < 0) {
371 		/* Reject negative offsets immediately. */
372 		bp->b_error = EINVAL;
373 		return -1;
374 	}
375 
376 	/* Protect against division by zero. XXX: Should never happen?!?! */
377 	if (lp->d_secpercyl == 0) {
378 		bp->b_error = EINVAL;
379 		return -1;
380 	}
381 
382 	p_size = (uint64_t)p->p_size << dk->dk_blkshift;
383 	p_offset = (uint64_t)p->p_offset << dk->dk_blkshift;
384 #if RAW_PART == 3
385 	labelsector = lp->d_partitions[2].p_offset;
386 #else
387 	labelsector = lp->d_partitions[RAW_PART].p_offset;
388 #endif
389 	labelsector = (labelsector + dk->dk_labelsector) << dk->dk_blkshift;
390 
391 	sz = howmany((int64_t)bp->b_bcount, DEV_BSIZE);
392 
393 	/*
394 	 * bp->b_bcount is a 32-bit value, and we rejected a negative
395 	 * bp->b_blkno already, so "bp->b_blkno + sz" cannot overflow.
396 	 */
397 
398 	if (bp->b_blkno + sz > p_size) {
399 		sz = p_size - bp->b_blkno;
400 		if (sz == 0) {
401 			/* If exactly at end of disk, return EOF. */
402 			bp->b_resid = bp->b_bcount;
403 			return 0;
404 		}
405 		if (sz < 0) {
406 			/* If past end of disk, return EINVAL. */
407 			bp->b_error = EINVAL;
408 			return -1;
409 		}
410 		/* Otherwise, truncate request. */
411 		bp->b_bcount = sz << DEV_BSHIFT;
412 	}
413 
414 	/* Overwriting disk label? */
415 	if (bp->b_blkno + p_offset <= labelsector &&
416 	    bp->b_blkno + p_offset + sz > labelsector &&
417 	    (bp->b_flags & B_READ) == 0 && !wlabel) {
418 		bp->b_error = EROFS;
419 		return -1;
420 	}
421 
422 	/* calculate cylinder for disksort to order transfers with */
423 	bp->b_cylinder = (bp->b_blkno + p->p_offset) /
424 	    (lp->d_secsize / DEV_BSIZE) / lp->d_secpercyl;
425 	return 1;
426 }
427 
428 int
429 disk_read_sectors(void (*strat)(struct buf *), const struct disklabel *lp,
430     struct buf *bp, unsigned int sector, int count)
431 {
432 	bp->b_blkno = btodb((off_t)sector * lp->d_secsize);
433 	bp->b_bcount = count * lp->d_secsize;
434 	bp->b_flags = (bp->b_flags & ~B_WRITE) | B_READ;
435 	bp->b_oflags &= ~BO_DONE;
436 	bp->b_cylinder = sector / lp->d_secpercyl;
437 	(*strat)(bp);
438 	return biowait(bp);
439 }
440 
441 const char *
442 convertdisklabel(struct disklabel *lp, void (*strat)(struct buf *),
443     struct buf *bp, uint32_t secperunit)
444 {
445 	struct partition rp, *altp, *p;
446 	int geom_ok;
447 	const char *str;
448 
449 	memset(&rp, 0, sizeof(rp));
450 	rp.p_size = secperunit;
451 	rp.p_fstype = FS_UNUSED;
452 
453 	/* If we can seek to d_secperunit - 1, believe the disk geometry. */
454 	if (secperunit != 0 &&
455 	    disk_read_sectors(strat, lp, bp, secperunit - 1, 1) == 0)
456 		geom_ok = 1;
457 	else
458 		geom_ok = 0;
459 
460 #if 0
461 	printf("%s: secperunit (%" PRIu32 ") %s\n", __func__,
462 	    secperunit, geom_ok ? "ok" : "not ok");
463 #endif
464 
465 	p = &lp->d_partitions[RAW_PART];
466 	if (RAW_PART == 'c' - 'a')
467 		altp = &lp->d_partitions['d' - 'a'];
468 	else
469 		altp = &lp->d_partitions['c' - 'a'];
470 
471 	if (lp->d_npartitions > RAW_PART && p->p_offset == 0 && p->p_size != 0)
472 		return NULL;	/* already a raw partition */
473 	else if (lp->d_npartitions > MAX('c', 'd') - 'a' &&
474 		 altp->p_offset == 0 && altp->p_size != 0) {
475 		/* alternate partition ('c' or 'd') is suitable for raw slot,
476 		 * swap with 'd' or 'c'.
477 		 */
478 		rp = *p;
479 		*p = *altp;
480 		*altp = rp;
481 		return NULL;
482 	} else if (lp->d_npartitions <= RAW_PART &&
483 	           lp->d_npartitions > 'c' - 'a') {
484 		/* No raw partition is present, but the alternate is present.
485 		 * Copy alternate to raw partition.
486 		 */
487 		lp->d_npartitions = RAW_PART + 1;
488 		*p = *altp;
489 		return NULL;
490 	} else if (!geom_ok)
491 		str = "no raw partition and disk reports bad geometry";
492 	else if (lp->d_npartitions <= RAW_PART) {
493 		memset(&lp->d_partitions[lp->d_npartitions], 0,
494 		    sizeof(struct partition) * (RAW_PART - lp->d_npartitions));
495 		*p = rp;
496 		lp->d_npartitions = RAW_PART + 1;
497 		return NULL;
498 	} else if (lp->d_npartitions < MAXPARTITIONS) {
499 		memmove(p + 1, p,
500 		    sizeof(struct partition) * (lp->d_npartitions - RAW_PART));
501 		*p = rp;
502 		lp->d_npartitions++;
503 		return NULL;
504 	} else
505 		str = "no raw partition and partition table is full";
506 #ifdef DIAGNOSTIC
507 	printf("Bad partition: %s\n", str);
508 	printf("type = %u, subtype = %u, typename = %s\n",
509 	    lp->d_type, lp->d_subtype, lp->d_typename);
510 	printf("secsize = %u, nsectors = %u, ntracks = %u\n",
511 	    lp->d_secsize, lp->d_nsectors, lp->d_ntracks);
512 	printf("ncylinders = %u, secpercyl = %u, secperunit = %u\n",
513 	    lp->d_ncylinders, lp->d_secpercyl, lp->d_secperunit);
514 	printf("npartitions = %u\n", lp->d_npartitions);
515 
516 	for (size_t i = 0; i < MIN(lp->d_npartitions, MAXPARTITIONS); i++) {
517 		p = &lp->d_partitions[i];
518 		printf("\t%c: offset = %u size = %u fstype = %u\n",
519 		    (char)(i + 'a'), p->p_offset, p->p_size, p->p_fstype);
520 	}
521 #endif
522 	return str;
523 }
524 
525 /*
526  * disk_ioctl --
527  *	Generic disk ioctl handling.
528  */
529 int
530 disk_ioctl(struct disk *dk, dev_t dev, u_long cmd, void *data, int flag,
531     struct lwp *l)
532 {
533 	struct dkwedge_info *dkw;
534 	struct partinfo *pi;
535 	struct partition *dp;
536 #ifdef __HAVE_OLD_DISKLABEL
537 	struct disklabel newlabel;
538 #endif
539 
540 	switch (cmd) {
541 	case DIOCGDISKINFO:
542 		if (dk->dk_info == NULL)
543 			return ENOTSUP;
544 		return prop_dictionary_copyout_ioctl(data, cmd, dk->dk_info);
545 
546 	case DIOCGSECTORSIZE:
547 		*(u_int *)data = dk->dk_geom.dg_secsize;
548 		return 0;
549 
550 	case DIOCGMEDIASIZE:
551 		*(off_t *)data = (off_t)dk->dk_geom.dg_secsize *
552 		    dk->dk_geom.dg_secperunit;
553 		return 0;
554 	default:
555 		break;
556 	}
557 
558 	if (dev == NODEV)
559 		return EPASSTHROUGH;
560 
561 	/* The following should be moved to dk_ioctl */
562 	switch (cmd) {
563 	case DIOCGDINFO:
564 		if (dk->dk_label == NULL)
565 			return EBUSY;
566 		memcpy(data, dk->dk_label, sizeof (*dk->dk_label));
567 		return 0;
568 
569 #ifdef __HAVE_OLD_DISKLABEL
570 	case ODIOCGDINFO:
571 		if (dk->dk_label == NULL)
572 			return EBUSY;
573 		memcpy(&newlabel, dk->dk_label, sizeof(newlabel));
574 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
575 			return ENOTTY;
576 		memcpy(data, &newlabel, sizeof(struct olddisklabel));
577 		return 0;
578 #endif
579 
580 	case DIOCGPARTINFO:
581 		pi = data;
582 		memset(pi, 0, sizeof(*pi));
583 		pi->pi_secsize = dk->dk_geom.dg_secsize;
584 		pi->pi_bsize = MAX(BLKDEV_IOSIZE, pi->pi_secsize);
585 
586 		if (DISKPART(dev) == RAW_PART) {
587 			pi->pi_size = dk->dk_geom.dg_secperunit;
588 			return 0;
589 		}
590 
591 		if (dk->dk_label == NULL)
592 			return EBUSY;
593 
594 		dp = &dk->dk_label->d_partitions[DISKPART(dev)];
595 		pi->pi_offset = dp->p_offset;
596 		pi->pi_size = dp->p_size;
597 
598 		pi->pi_fstype = dp->p_fstype;
599 		pi->pi_frag = dp->p_frag;
600 		pi->pi_fsize = dp->p_fsize;
601 		pi->pi_cpg = dp->p_cpg;
602 
603 		/*
604 		 * dholland 20130616: XXX this logic should not be
605 		 * here. It is here because the old buffer cache
606 		 * demands that all accesses to the same blocks need
607 		 * to be the same size; but it only works for FFS and
608 		 * nowadays I think it'll fail silently if the size
609 		 * info in the disklabel is wrong. (Or missing.) The
610 		 * buffer cache needs to be smarter; or failing that
611 		 * we need a reliable way here to get the right block
612 		 * size; or a reliable way to guarantee that (a) the
613 		 * fs is not mounted when we get here and (b) any
614 		 * buffers generated here will get purged when the fs
615 		 * does get mounted.
616 		 */
617 		if (dp->p_fstype == FS_BSDFFS &&
618 		    dp->p_frag != 0 && dp->p_fsize != 0)
619 			pi->pi_bsize = dp->p_frag * dp->p_fsize;
620 		return 0;
621 
622 	case DIOCAWEDGE:
623 		if ((flag & FWRITE) == 0)
624 			return EBADF;
625 
626 		dkw = data;
627 		strlcpy(dkw->dkw_parent, dk->dk_name, sizeof(dkw->dkw_parent));
628 		return dkwedge_add(dkw);
629 
630 	case DIOCDWEDGE:
631 		if ((flag & FWRITE) == 0)
632 			return EBADF;
633 
634 		dkw = data;
635 		strlcpy(dkw->dkw_parent, dk->dk_name, sizeof(dkw->dkw_parent));
636 		return dkwedge_del(dkw);
637 
638 	case DIOCLWEDGES:
639 		return dkwedge_list(dk, data, l);
640 
641 	case DIOCMWEDGES:
642 		if ((flag & FWRITE) == 0)
643 			return EBADF;
644 
645 		dkwedge_discover(dk);
646 		return 0;
647 
648 	default:
649 		return EPASSTHROUGH;
650 	}
651 }
652 
653 void
654 disk_set_info(device_t dev, struct disk *dk, const char *type)
655 {
656 	struct disk_geom *dg = &dk->dk_geom;
657 
658 	if (dg->dg_secsize == 0) {
659 #ifdef DIAGNOSTIC
660 		printf("%s: fixing 0 sector size\n", dk->dk_name);
661 #endif
662 		dg->dg_secsize = DEV_BSIZE;
663 	}
664 
665 	dk->dk_blkshift = DK_BSIZE2BLKSHIFT(dg->dg_secsize);
666 	dk->dk_byteshift = DK_BSIZE2BYTESHIFT(dg->dg_secsize);
667 
668 	if (dg->dg_secperunit == 0 && dg->dg_ncylinders == 0) {
669 #ifdef DIAGNOSTIC
670 		printf("%s: secperunit and ncylinders are zero\n", dk->dk_name);
671 #endif
672 		return;
673 	}
674 
675 	if (dg->dg_secperunit == 0) {
676 		if (dg->dg_nsectors == 0 || dg->dg_ntracks == 0) {
677 #ifdef DIAGNOSTIC
678 			printf("%s: secperunit and (sectors or tracks) "
679 			    "are zero\n", dk->dk_name);
680 #endif
681 			return;
682 		}
683 		dg->dg_secperunit = (int64_t) dg->dg_nsectors *
684 		    dg->dg_ntracks * dg->dg_ncylinders;
685 	}
686 
687 	if (dg->dg_ncylinders == 0) {
688 		if (dg->dg_ntracks && dg->dg_nsectors)
689 			dg->dg_ncylinders = dg->dg_secperunit /
690 			    (dg->dg_ntracks * dg->dg_nsectors);
691 	}
692 
693 	prop_dictionary_t disk_info, odisk_info, geom;
694 
695 	disk_info = prop_dictionary_create();
696 	geom = prop_dictionary_create();
697 
698 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
699 	    dg->dg_secperunit);
700 
701 	prop_dictionary_set_uint32(geom, "sector-size", dg->dg_secsize);
702 
703 	if (dg->dg_nsectors)
704 		prop_dictionary_set_uint16(geom, "sectors-per-track",
705 		    dg->dg_nsectors);
706 
707 	if (dg->dg_ntracks)
708 		prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
709 		    dg->dg_ntracks);
710 
711 	if (dg->dg_ncylinders)
712 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
713 		    dg->dg_ncylinders);
714 
715 	prop_dictionary_set(disk_info, "geometry", geom);
716 
717 	if (type)
718 		prop_dictionary_set_cstring_nocopy(disk_info, "type", type);
719 
720 	prop_object_release(geom);
721 
722 	odisk_info = dk->dk_info;
723 	dk->dk_info = disk_info;
724 
725 	if (dev)
726 		prop_dictionary_set(device_properties(dev), "disk-info",
727 		    disk_info);
728 
729 	/*
730 	 * Don't release disk_info here; we keep a reference to it.
731 	 * disk_detach() will release it when we go away.
732 	 */
733 	if (odisk_info)
734 		prop_object_release(odisk_info);
735 }
736