xref: /dflybsd-src/sys/kern/subr_diskslice.c (revision 286555ab3724a60987968ac954f4c750c1fbe560)
1 /*-
2  * Copyright (c) 1994 Bruce D. Evans.
3  * All rights reserved.
4  *
5  * Copyright (c) 1990 The Regents of the University of California.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * William Jolitz.
10  *
11  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
12  * All rights reserved.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *	This product includes software developed by the University of
25  *	California, Berkeley and its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	from: @(#)wd.c	7.2 (Berkeley) 5/9/91
43  *	from: wd.c,v 1.55 1994/10/22 01:57:12 phk Exp $
44  *	from: @(#)ufs_disksubr.c	7.16 (Berkeley) 5/4/91
45  *	from: ufs_disksubr.c,v 1.8 1994/06/07 01:21:39 phk Exp $
46  * $FreeBSD: src/sys/kern/subr_diskslice.c,v 1.82.2.6 2001/07/24 09:49:41 dd Exp $
47  * $DragonFly: src/sys/kern/subr_diskslice.c,v 1.14 2006/03/24 18:35:33 dillon Exp $
48  */
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/buf.h>
53 #include <sys/conf.h>
54 #include <sys/disklabel.h>
55 #include <sys/diskslice.h>
56 #include <sys/diskmbr.h>
57 #include <sys/fcntl.h>
58 #include <sys/malloc.h>
59 #include <sys/stat.h>
60 #include <sys/syslog.h>
61 #include <sys/vnode.h>
62 #include <sys/device.h>
63 #include <sys/thread2.h>
64 
65 #include <vfs/ufs/fs.h>
66 
67 #define TRACE(str)	do { if (ds_debug) printf str; } while (0)
68 
69 typedef	u_char	bool_t;
70 
71 static volatile bool_t ds_debug;
72 
73 static struct disklabel *clone_label (struct disklabel *lp);
74 static void dsiodone (struct bio *bio);
75 static char *fixlabel (char *sname, struct diskslice *sp,
76 			   struct disklabel *lp, int writeflag);
77 static void free_ds_label (struct diskslices *ssp, int slice);
78 static void partition_info (char *sname, int part, struct partition *pp);
79 static void slice_info (char *sname, struct diskslice *sp);
80 static void set_ds_label (struct diskslices *ssp, int slice,
81 			      struct disklabel *lp);
82 static void set_ds_wlabel (struct diskslices *ssp, int slice, int wlabel);
83 
84 /*
85  * Duplicate a label for the whole disk, and initialize defaults in the
86  * copy for fields that are not already initialized.  The caller only
87  * needs to initialize d_secsize and d_secperunit, and zero the fields
88  * that are to be defaulted.
89  */
90 static struct disklabel *
91 clone_label(struct disklabel *lp)
92 {
93 	struct disklabel *lp1;
94 
95 	lp1 = malloc(sizeof *lp1, M_DEVBUF, M_WAITOK);
96 	*lp1 = *lp;
97 	lp = NULL;
98 	if (lp1->d_typename[0] == '\0')
99 		strncpy(lp1->d_typename, "amnesiac", sizeof(lp1->d_typename));
100 	if (lp1->d_packname[0] == '\0')
101 		strncpy(lp1->d_packname, "fictitious", sizeof(lp1->d_packname));
102 	if (lp1->d_nsectors == 0)
103 		lp1->d_nsectors = 32;
104 	if (lp1->d_ntracks == 0)
105 		lp1->d_ntracks = 64;
106 	lp1->d_secpercyl = lp1->d_nsectors * lp1->d_ntracks;
107 	lp1->d_ncylinders = lp1->d_secperunit / lp1->d_secpercyl;
108 	if (lp1->d_rpm == 0)
109 		lp1->d_rpm = 3600;
110 	if (lp1->d_interleave == 0)
111 		lp1->d_interleave = 1;
112 	if (lp1->d_npartitions < RAW_PART + 1)
113 		lp1->d_npartitions = MAXPARTITIONS;
114 	if (lp1->d_bbsize == 0)
115 		lp1->d_bbsize = BBSIZE;
116 	if (lp1->d_sbsize == 0)
117 		lp1->d_sbsize = SBSIZE;
118 	lp1->d_partitions[RAW_PART].p_size = lp1->d_secperunit;
119 	lp1->d_magic = DISKMAGIC;
120 	lp1->d_magic2 = DISKMAGIC;
121 	lp1->d_checksum = dkcksum(lp1);
122 	return (lp1);
123 }
124 
125 /*
126  * Determine the size of the transfer, and make sure it is
127  * within the boundaries of the partition. Adjust transfer
128  * if needed, and signal errors or early completion.
129  *
130  * XXX TODO:
131  *	o Split buffers that are too big for the device.
132  *	o Check for overflow.
133  *	o Finish cleaning this up.
134  *
135  * This function returns 1 on success, 0 if transfer equates
136  * to EOF (end of disk) or -1 on failure.  The appropriate
137  * 'errno' value is also set in bp->b_error and bp->b_flags
138  * is marked with B_ERROR.
139  */
140 struct bio *
141 dscheck(dev_t dev, struct bio *bio, struct diskslices *ssp)
142 {
143 	struct buf *bp = bio->bio_buf;
144 	struct bio *nbio;
145 	u_long	endsecno;
146 	daddr_t	labelsect;
147 	struct disklabel *lp;
148 	char *msg;
149 	long nsec;
150 	struct partition *pp;
151 	daddr_t	secno;
152 	daddr_t	slicerel_secno;
153 	struct diskslice *sp;
154 	int shift;
155 	int mask;
156 
157 	if (bio->bio_offset < 0) {
158 		printf("dscheck(%s): negative bio_offset %lld\n",
159 		    devtoname(dev), bio->bio_offset);
160 		bp->b_error = EINVAL;
161 		goto bad;
162 	}
163 	sp = &ssp->dss_slices[dkslice(dev)];
164 	lp = sp->ds_label;
165 
166 	if (ssp->dss_secmult == 1) {
167 		shift = DEV_BSHIFT;
168 		goto doshift;
169 	} else if (ssp->dss_secshift != -1) {
170 		shift = DEV_BSHIFT + ssp->dss_secshift;
171 doshift:
172 		mask = (1 << shift) - 1;
173 		if ((int)bp->b_bcount & mask)
174 			goto bad_bcount;
175 		if ((int)bio->bio_offset & mask)
176 			goto bad_blkno;
177 		secno = (daddr_t)(bio->bio_offset >> shift);
178 		nsec = bp->b_bcount >> shift;
179 	} else {
180 		if (bp->b_bcount % ssp->dss_secsize)
181 			goto bad_bcount;
182 		if (bio->bio_offset % ssp->dss_secsize)
183 			goto bad_blkno;
184 		secno = (daddr_t)(bio->bio_offset / ssp->dss_secsize);
185 		nsec = bp->b_bcount / ssp->dss_secsize;
186 	}
187 	if (lp == NULL) {
188 		labelsect = -LABELSECTOR - 1;
189 		endsecno = sp->ds_size;
190 		slicerel_secno = secno;
191 	} else {
192 		labelsect = lp->d_partitions[LABEL_PART].p_offset;
193 		if (labelsect != 0)
194 			Debugger("labelsect != 0 in dscheck()");
195 		pp = &lp->d_partitions[dkpart(dev)];
196 		endsecno = pp->p_size;
197 		slicerel_secno = pp->p_offset + secno;
198 	}
199 
200 	/* overwriting disk label ? */
201 	/* XXX should also protect bootstrap in first 8K */
202 	if (slicerel_secno <= LABELSECTOR + labelsect &&
203 #if LABELSECTOR != 0
204 	    slicerel_secno + nsec > LABELSECTOR + labelsect &&
205 #endif
206 	    (bp->b_flags & B_READ) == 0 && sp->ds_wlabel == 0) {
207 		bp->b_error = EROFS;
208 		goto bad;
209 	}
210 
211 #if defined(DOSBBSECTOR) && defined(notyet)
212 	/* overwriting master boot record? */
213 	if (slicerel_secno <= DOSBBSECTOR && (bp->b_flags & B_READ) == 0 &&
214 	    sp->ds_wlabel == 0) {
215 		bp->b_error = EROFS;
216 		goto bad;
217 	}
218 #endif
219 
220 	/* beyond partition? */
221 	if (secno + nsec > endsecno) {
222 		/* if exactly at end of disk, return an EOF */
223 		if (secno == endsecno) {
224 			bp->b_resid = bp->b_bcount;
225 			return (0);
226 		}
227 		/* or truncate if part of it fits */
228 		nsec = endsecno - secno;
229 		if (nsec <= 0) {
230 			bp->b_error = EINVAL;
231 			goto bad;
232 		}
233 		bp->b_bcount = nsec * ssp->dss_secsize;
234 	}
235 
236 	nbio = push_bio(bio);
237 	nbio->bio_offset = (off_t)(sp->ds_offset + slicerel_secno) * ssp->dss_secsize;
238 
239 	/*
240 	 * Snoop on label accesses if the slice offset is nonzero.  Fudge
241 	 * offsets in the label to keep the in-core label coherent with
242 	 * the on-disk one.
243 	 */
244 	if (slicerel_secno <= LABELSECTOR + labelsect
245 #if LABELSECTOR != 0
246 	    && slicerel_secno + nsec > LABELSECTOR + labelsect
247 #endif
248 	    && sp->ds_offset != 0) {
249 		nbio->bio_done = dsiodone;
250 		nbio->bio_caller_info1.ptr = sp;
251 		nbio->bio_caller_info2.offset = (off_t)(LABELSECTOR + labelsect -
252 					 slicerel_secno) * ssp->dss_secsize;
253 		if ((bp->b_flags & B_READ) == 0) {
254 			/*
255 			 * XXX even disklabel(8) writes directly so we need
256 			 * to adjust writes.  Perhaps we should drop support
257 			 * for DIOCWLABEL (always write protect labels) and
258 			 * require the use of DIOCWDINFO.
259 			 *
260 			 * XXX probably need to copy the data to avoid even
261 			 * temporarily corrupting the in-core copy.
262 			 */
263 			/* XXX need name here. */
264 			msg = fixlabel(
265 				NULL, sp,
266 			       (struct disklabel *)
267 			       (bp->b_data + (int)nbio->bio_caller_info2.offset),
268 			       TRUE);
269 			if (msg != NULL) {
270 				printf("dscheck(%s): %s\n",
271 				    devtoname(dev), msg);
272 				bp->b_error = EROFS;
273 				pop_bio(nbio);
274 				goto bad;
275 			}
276 		}
277 	}
278 	return (nbio);
279 
280 bad_bcount:
281 	printf(
282 	"dscheck(%s): b_bcount %d is not on a sector boundary (ssize %d)\n",
283 	    devtoname(dev), bp->b_bcount, ssp->dss_secsize);
284 	bp->b_error = EINVAL;
285 	goto bad;
286 
287 bad_blkno:
288 	printf(
289 	"dscheck(%s): bio_offset %lld is not on a sector boundary (ssize %d)\n",
290 	    devtoname(dev), bio->bio_offset, ssp->dss_secsize);
291 	bp->b_error = EINVAL;
292 	goto bad;
293 
294 bad:
295 	bp->b_resid = bp->b_bcount;
296 	bp->b_flags |= B_ERROR;
297 	return (NULL);
298 }
299 
300 void
301 dsclose(dev_t dev, int mode, struct diskslices *ssp)
302 {
303 	u_char mask;
304 	struct diskslice *sp;
305 
306 	sp = &ssp->dss_slices[dkslice(dev)];
307 	mask = 1 << dkpart(dev);
308 	sp->ds_openmask &= ~mask;
309 }
310 
311 void
312 dsgone(struct diskslices **sspp)
313 {
314 	int slice;
315 	struct diskslice *sp;
316 	struct diskslices *ssp;
317 
318 	for (slice = 0, ssp = *sspp; slice < ssp->dss_nslices; slice++) {
319 		sp = &ssp->dss_slices[slice];
320 		free_ds_label(ssp, slice);
321 	}
322 	free(ssp, M_DEVBUF);
323 	*sspp = NULL;
324 }
325 
326 /*
327  * For the "write" commands (DIOCSDINFO and DIOCWDINFO), this
328  * is subject to the same restriction as dsopen().
329  */
330 int
331 dsioctl(dev_t dev, u_long cmd, caddr_t data,
332 	int flags, struct diskslices **sspp)
333 {
334 	int error;
335 	struct disklabel *lp;
336 	int old_wlabel;
337 	u_char openmask;
338 	int part;
339 	int slice;
340 	struct diskslice *sp;
341 	struct diskslices *ssp;
342 	struct partition *pp;
343 
344 	slice = dkslice(dev);
345 	ssp = *sspp;
346 	sp = &ssp->dss_slices[slice];
347 	lp = sp->ds_label;
348 	switch (cmd) {
349 
350 	case DIOCGDVIRGIN:
351 		lp = (struct disklabel *)data;
352 		if (ssp->dss_slices[WHOLE_DISK_SLICE].ds_label) {
353 			*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
354 		} else {
355 			bzero(lp, sizeof(struct disklabel));
356 		}
357 
358 		lp->d_magic = DISKMAGIC;
359 		lp->d_magic2 = DISKMAGIC;
360 		pp = &lp->d_partitions[RAW_PART];
361 		pp->p_offset = 0;
362 		pp->p_size = sp->ds_size;
363 
364 		lp->d_npartitions = MAXPARTITIONS;
365 		if (lp->d_interleave == 0)
366 			lp->d_interleave = 1;
367 		if (lp->d_rpm == 0)
368 			lp->d_rpm = 3600;
369 		if (lp->d_nsectors == 0)
370 			lp->d_nsectors = 32;
371 		if (lp->d_ntracks == 0)
372 			lp->d_ntracks = 64;
373 
374 		lp->d_bbsize = BBSIZE;
375 		lp->d_sbsize = SBSIZE;
376 		lp->d_secpercyl = lp->d_nsectors * lp->d_ntracks;
377 		lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
378 		lp->d_secperunit = sp->ds_size;
379 		lp->d_checksum = 0;
380 		lp->d_checksum = dkcksum(lp);
381 		return (0);
382 
383 	case DIOCGDINFO:
384 		if (lp == NULL)
385 			return (EINVAL);
386 		*(struct disklabel *)data = *lp;
387 		return (0);
388 
389 #ifdef notyet
390 	case DIOCGDINFOP:
391 		if (lp == NULL)
392 			return (EINVAL);
393 		*(struct disklabel **)data = lp;
394 		return (0);
395 #endif
396 
397 	case DIOCGPART:
398 		if (lp == NULL)
399 			return (EINVAL);
400 		((struct partinfo *)data)->disklab = lp;
401 		((struct partinfo *)data)->part
402 			= &lp->d_partitions[dkpart(dev)];
403 		return (0);
404 
405 	case DIOCGSLICEINFO:
406 		bcopy(ssp, data, (char *)&ssp->dss_slices[ssp->dss_nslices] -
407 				 (char *)ssp);
408 		return (0);
409 
410 	case DIOCSDINFO:
411 		if (slice == WHOLE_DISK_SLICE)
412 			return (ENODEV);
413 		if (!(flags & FWRITE))
414 			return (EBADF);
415 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
416 		if (sp->ds_label == NULL)
417 			bzero(lp, sizeof *lp);
418 		else
419 			bcopy(sp->ds_label, lp, sizeof *lp);
420 		if (sp->ds_label == NULL)
421 			openmask = 0;
422 		else {
423 			openmask = sp->ds_openmask;
424 			if (slice == COMPATIBILITY_SLICE)
425 				openmask |= ssp->dss_slices[
426 				    ssp->dss_first_bsd_slice].ds_openmask;
427 			else if (slice == ssp->dss_first_bsd_slice)
428 				openmask |= ssp->dss_slices[
429 				    COMPATIBILITY_SLICE].ds_openmask;
430 		}
431 		error = setdisklabel(lp, (struct disklabel *)data,
432 				     (u_long)openmask);
433 		/* XXX why doesn't setdisklabel() check this? */
434 		if (error == 0 && lp->d_partitions[RAW_PART].p_offset != 0)
435 			error = EXDEV;
436 		if (error == 0) {
437 			if (lp->d_secperunit > sp->ds_size)
438 				error = ENOSPC;
439 			for (part = 0; part < lp->d_npartitions; part++)
440 				if (lp->d_partitions[part].p_size > sp->ds_size)
441 					error = ENOSPC;
442 		}
443 		if (error != 0) {
444 			free(lp, M_DEVBUF);
445 			return (error);
446 		}
447 		free_ds_label(ssp, slice);
448 		set_ds_label(ssp, slice, lp);
449 		return (0);
450 
451 	case DIOCSYNCSLICEINFO:
452 		if (slice != WHOLE_DISK_SLICE || dkpart(dev) != RAW_PART)
453 			return (EINVAL);
454 		if (!*(int *)data)
455 			for (slice = 0; slice < ssp->dss_nslices; slice++) {
456 				openmask = ssp->dss_slices[slice].ds_openmask;
457 				if (openmask
458 				    && (slice != WHOLE_DISK_SLICE
459 					|| openmask & ~(1 << RAW_PART)))
460 					return (EBUSY);
461 			}
462 
463 		/*
464 		 * Temporarily forget the current slices struct and read
465 		 * the current one.
466 		 * XXX should wait for current accesses on this disk to
467 		 * complete, then lock out future accesses and opens.
468 		 */
469 		*sspp = NULL;
470 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
471 		*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
472 		error = dsopen(dev, S_IFCHR, ssp->dss_oflags, sspp, lp);
473 		if (error != 0) {
474 			free(lp, M_DEVBUF);
475 			*sspp = ssp;
476 			return (error);
477 		}
478 
479 		/*
480 		 * Reopen everything.  This is a no-op except in the "force"
481 		 * case and when the raw bdev and cdev are both open.  Abort
482 		 * if anything fails.
483 		 */
484 		for (slice = 0; slice < ssp->dss_nslices; slice++) {
485 			for (openmask = ssp->dss_slices[slice].ds_openmask,
486 			     part = 0; openmask; openmask >>= 1, part++) {
487 				if (!(openmask & 1))
488 					continue;
489 				error = dsopen(dkmodslice(dkmodpart(dev, part),
490 							  slice),
491 					       S_IFCHR, ssp->dss_oflags, sspp,
492 					       lp);
493 				if (error != 0) {
494 					free(lp, M_DEVBUF);
495 					*sspp = ssp;
496 					return (EBUSY);
497 				}
498 			}
499 		}
500 
501 		free(lp, M_DEVBUF);
502 		dsgone(&ssp);
503 		return (0);
504 
505 	case DIOCWDINFO:
506 		error = dsioctl(dev, DIOCSDINFO, data, flags, &ssp);
507 		if (error != 0)
508 			return (error);
509 		/*
510 		 * XXX this used to hack on dk_openpart to fake opening
511 		 * partition 0 in case that is used instead of dkpart(dev).
512 		 */
513 		old_wlabel = sp->ds_wlabel;
514 		set_ds_wlabel(ssp, slice, TRUE);
515 		error = writedisklabel(dev, sp->ds_label);
516 		/* XXX should invalidate in-core label if write failed. */
517 		set_ds_wlabel(ssp, slice, old_wlabel);
518 		return (error);
519 
520 	case DIOCWLABEL:
521 		if (slice == WHOLE_DISK_SLICE)
522 			return (ENODEV);
523 		if (!(flags & FWRITE))
524 			return (EBADF);
525 		set_ds_wlabel(ssp, slice, *(int *)data != 0);
526 		return (0);
527 
528 	default:
529 		return (ENOIOCTL);
530 	}
531 }
532 
533 static void
534 dsiodone(struct bio *bio)
535 {
536 	struct buf *bp = bio->bio_buf;
537 	char *msg;
538 
539 	bp->b_flags = bp->b_flags & ~B_DONE;
540 	if (!(bp->b_flags & B_READ)
541 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
542 		msg = fixlabel(NULL, bio->bio_caller_info1.ptr,
543 			       (struct disklabel *)
544 			       (bp->b_data + (int)bio->bio_caller_info2.offset),
545 			       FALSE);
546 		if (msg != NULL)
547 			printf("%s\n", msg);
548 	}
549 	biodone(bio->bio_prev);
550 }
551 
552 int
553 dsisopen(struct diskslices *ssp)
554 {
555 	int slice;
556 
557 	if (ssp == NULL)
558 		return (0);
559 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
560 		if (ssp->dss_slices[slice].ds_openmask)
561 			return (1);
562 	}
563 	return (0);
564 }
565 
566 /*
567  * Allocate a slices "struct" and initialize it to contain only an empty
568  * compatibility slice (pointing to itself), a whole disk slice (covering
569  * the disk as described by the label), and (nslices - BASE_SLICES) empty
570  * slices beginning at BASE_SLICE.
571  */
572 struct diskslices *
573 dsmakeslicestruct(int nslices, struct disklabel *lp)
574 {
575 	struct diskslice *sp;
576 	struct diskslices *ssp;
577 
578 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
579 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
580 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
581 	ssp->dss_nslices = nslices;
582 	ssp->dss_oflags = 0;
583 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
584 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
585 		ssp->dss_secshift = -1;
586 	else
587 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
588 	ssp->dss_secsize = lp->d_secsize;
589 	sp = &ssp->dss_slices[0];
590 	bzero(sp, nslices * sizeof *sp);
591 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
592 	return (ssp);
593 }
594 
595 char *
596 dsname(dev_t dev, int unit, int slice, int part, char *partname)
597 {
598 	static char name[32];
599 	const char *dname;
600 
601 	dname = dev_dname(dev);
602 	if (strlen(dname) > 16)
603 		dname = "nametoolong";
604 	snprintf(name, sizeof(name), "%s%d", dname, unit);
605 	partname[0] = '\0';
606 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
607 		partname[0] = 'a' + part;
608 		partname[1] = '\0';
609 		if (slice != COMPATIBILITY_SLICE) {
610 			snprintf(name + strlen(name),
611 			    sizeof(name) - strlen(name), "s%d", slice - 1);
612 		}
613 	}
614 	return (name);
615 }
616 
617 /*
618  * This should only be called when the unit is inactive and the strategy
619  * routine should not allow it to become active unless we call it.  Our
620  * strategy routine must be special to allow activity.
621  */
622 int
623 dsopen(dev_t dev, int mode, u_int flags,
624 	struct diskslices **sspp, struct disklabel *lp)
625 {
626 	dev_t dev1;
627 	int error;
628 	struct disklabel *lp1;
629 	char *msg;
630 	u_char mask;
631 	bool_t need_init;
632 	int part;
633 	char partname[2];
634 	int slice;
635 	char *sname;
636 	struct diskslice *sp;
637 	struct diskslices *ssp;
638 	int unit;
639 
640 	dev->si_bsize_phys = lp->d_secsize;
641 
642 	unit = dkunit(dev);
643 	if (lp->d_secsize % DEV_BSIZE) {
644 		printf("%s: invalid sector size %lu\n", devtoname(dev),
645 		    (u_long)lp->d_secsize);
646 		return (EINVAL);
647 	}
648 
649 	/*
650 	 * XXX reinitialize the slice table unless there is an open device
651 	 * on the unit.  This should only be done if the media has changed.
652 	 */
653 	ssp = *sspp;
654 	need_init = !dsisopen(ssp);
655 	if (ssp != NULL && need_init)
656 		dsgone(sspp);
657 	if (need_init) {
658 		/*
659 		 * Allocate a minimal slices "struct".  This will become
660 		 * the final slices "struct" if we don't want real slices
661 		 * or if we can't find any real slices.
662 		 */
663 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
664 
665 		if (!(flags & DSO_ONESLICE)) {
666 			TRACE(("dsinit\n"));
667 			error = dsinit(dev, lp, sspp);
668 			if (error != 0) {
669 				dsgone(sspp);
670 				return (error);
671 			}
672 		}
673 		ssp = *sspp;
674 		ssp->dss_oflags = flags;
675 
676 		/*
677 		 * If there are no real slices, then make the compatiblity
678 		 * slice cover the whole disk.
679 		 */
680 		if (ssp->dss_nslices == BASE_SLICE)
681 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
682 				= lp->d_secperunit;
683 
684 		/* Point the compatibility slice at the BSD slice, if any. */
685 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
686 			sp = &ssp->dss_slices[slice];
687 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
688 				ssp->dss_first_bsd_slice = slice;
689 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
690 					= sp->ds_offset;
691 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
692 					= sp->ds_size;
693 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
694 					= sp->ds_type;
695 				break;
696 			}
697 		}
698 
699 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
700 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
701 	}
702 
703 	/*
704 	 * Initialize secondary info for all slices.  It is needed for more
705 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
706 	 */
707 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
708 		sp = &ssp->dss_slices[slice];
709 		if (sp->ds_label != NULL)
710 			continue;
711 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
712 		sname = dsname(dev, unit, slice, RAW_PART, partname);
713 		/*
714 		 * XXX this should probably only be done for the need_init
715 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
716 		 */
717 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
718 		lp1 = clone_label(lp);
719 		TRACE(("readdisklabel\n"));
720 		if (flags & DSO_NOLABELS)
721 			msg = NULL;
722 		else {
723 			msg = readdisklabel(dev1, lp1);
724 
725 			/*
726 			 * readdisklabel() returns NULL for success, and an
727 			 * error string for failure.
728 			 *
729 			 * If there isn't a label on the disk, and if the
730 			 * DSO_COMPATLABEL is set, we want to use the
731 			 * faked-up label provided by the caller.
732 			 *
733 			 * So we set msg to NULL to indicate that there is
734 			 * no failure (since we have a faked-up label),
735 			 * free lp1, and then clone it again from lp.
736 			 * (In case readdisklabel() modified lp1.)
737 			 */
738 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
739 				msg = NULL;
740 				free(lp1, M_DEVBUF);
741 				lp1 = clone_label(lp);
742 			}
743 		}
744 		if (msg == NULL)
745 			msg = fixlabel(sname, sp, lp1, FALSE);
746 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
747 			msg = "inconsistent sector size";
748 		if (msg != NULL) {
749 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
750 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
751 				    sname, msg);
752 			free(lp1, M_DEVBUF);
753 			continue;
754 		}
755 		if (lp1->d_flags & D_BADSECT) {
756 			log(LOG_ERR, "%s: bad sector table not supported\n",
757 			    sname);
758 			free(lp1, M_DEVBUF);
759 			continue;
760 		}
761 		set_ds_label(ssp, slice, lp1);
762 		set_ds_wlabel(ssp, slice, FALSE);
763 	}
764 
765 	slice = dkslice(dev);
766 	if (slice >= ssp->dss_nslices)
767 		return (ENXIO);
768 	sp = &ssp->dss_slices[slice];
769 	part = dkpart(dev);
770 	if (part != RAW_PART
771 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
772 		return (EINVAL);	/* XXX needs translation */
773 	mask = 1 << part;
774 	sp->ds_openmask |= mask;
775 	return (0);
776 }
777 
778 int
779 dssize(dev_t dev, struct diskslices **sspp)
780 {
781 	struct disklabel *lp;
782 	int part;
783 	int slice;
784 	struct diskslices *ssp;
785 
786 	slice = dkslice(dev);
787 	part = dkpart(dev);
788 	ssp = *sspp;
789 	if (ssp == NULL || slice >= ssp->dss_nslices
790 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
791 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
792 			return (-1);
793 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
794 		ssp = *sspp;
795 	}
796 	lp = ssp->dss_slices[slice].ds_label;
797 	if (lp == NULL)
798 		return (-1);
799 	return ((int)lp->d_partitions[part].p_size);
800 }
801 
802 static void
803 free_ds_label(struct diskslices *ssp, int slice)
804 {
805 	struct disklabel *lp;
806 	struct diskslice *sp;
807 
808 	sp = &ssp->dss_slices[slice];
809 	lp = sp->ds_label;
810 	if (lp == NULL)
811 		return;
812 	free(lp, M_DEVBUF);
813 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
814 }
815 
816 static char *
817 fixlabel(char *sname, struct diskslice *sp, struct disklabel *lp, int writeflag)
818 {
819 	u_long end;
820 	u_long offset;
821 	int part;
822 	struct partition *pp;
823 	u_long start;
824 	bool_t warned;
825 
826 	/* These errors "can't happen" so don't bother reporting details. */
827 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
828 		return ("fixlabel: invalid magic");
829 	if (dkcksum(lp) != 0)
830 		return ("fixlabel: invalid checksum");
831 
832 	pp = &lp->d_partitions[RAW_PART];
833 	if (writeflag) {
834 		start = 0;
835 		offset = sp->ds_offset;
836 	} else {
837 		start = sp->ds_offset;
838 		offset = -sp->ds_offset;
839 	}
840 	if (pp->p_offset != start) {
841 		if (sname != NULL) {
842 			printf(
843 "%s: rejecting BSD label: raw partition offset != slice offset\n",
844 			       sname);
845 			slice_info(sname, sp);
846 			partition_info(sname, RAW_PART, pp);
847 		}
848 		return ("fixlabel: raw partition offset != slice offset");
849 	}
850 	if (pp->p_size != sp->ds_size) {
851 		if (sname != NULL) {
852 			printf("%s: raw partition size != slice size\n", sname);
853 			slice_info(sname, sp);
854 			partition_info(sname, RAW_PART, pp);
855 		}
856 		if (pp->p_size > sp->ds_size) {
857 			if (sname == NULL)
858 				return ("fixlabel: raw partition size > slice size");
859 			printf("%s: truncating raw partition\n", sname);
860 			pp->p_size = sp->ds_size;
861 		}
862 	}
863 	end = start + sp->ds_size;
864 	if (start > end)
865 		return ("fixlabel: slice wraps");
866 	if (lp->d_secpercyl <= 0)
867 		return ("fixlabel: d_secpercyl <= 0");
868 	pp -= RAW_PART;
869 	warned = FALSE;
870 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
871 		if (pp->p_offset != 0 || pp->p_size != 0) {
872 			if (pp->p_offset < start
873 			    || pp->p_offset + pp->p_size > end
874 			    || pp->p_offset + pp->p_size < pp->p_offset) {
875 				if (sname != NULL) {
876 					printf(
877 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
878 					       sname);
879 					if (!warned) {
880 						slice_info(sname, sp);
881 						warned = TRUE;
882 					}
883 					partition_info(sname, part, pp);
884 				}
885 				/* XXX else silently discard junk. */
886 				bzero(pp, sizeof *pp);
887 			} else
888 				pp->p_offset += offset;
889 		}
890 	}
891 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
892 	lp->d_secperunit = sp->ds_size;
893  	lp->d_checksum = 0;
894  	lp->d_checksum = dkcksum(lp);
895 	return (NULL);
896 }
897 
898 static void
899 partition_info(char *sname, int part, struct partition *pp)
900 {
901 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
902 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
903 	       (u_long)pp->p_size);
904 }
905 
906 static void
907 slice_info(char *sname, struct diskslice *sp)
908 {
909 	printf("%s: start %lu, end %lu, size %lu\n", sname,
910 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
911 }
912 
913 static void
914 set_ds_label(struct diskslices *ssp, int slice, struct disklabel *lp)
915 {
916 	ssp->dss_slices[slice].ds_label = lp;
917 	if (slice == COMPATIBILITY_SLICE)
918 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
919 	else if (slice == ssp->dss_first_bsd_slice)
920 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
921 }
922 
923 static void
924 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
925 {
926 	ssp->dss_slices[slice].ds_wlabel = wlabel;
927 	if (slice == COMPATIBILITY_SLICE)
928 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
929 	else if (slice == ssp->dss_first_bsd_slice)
930 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
931 }
932