xref: /dflybsd-src/sys/kern/subr_diskslice.c (revision 41871674d0079dec70d55eb824f39d07dc7b3310)
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 1.15 2006/04/03 02:02:35 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/dinode.h>	/* XXX used only for fs.h */
66 #include <vfs/ufs/fs.h>		/* XXX used only to get BBSIZE/SBSIZE */
67 
68 #define TRACE(str)	do { if (ds_debug) printf str; } while (0)
69 
70 typedef	u_char	bool_t;
71 
72 static volatile bool_t ds_debug;
73 
74 static struct disklabel *clone_label (struct disklabel *lp);
75 static void dsiodone (struct bio *bio);
76 static char *fixlabel (char *sname, struct diskslice *sp,
77 			   struct disklabel *lp, int writeflag);
78 static void free_ds_label (struct diskslices *ssp, int slice);
79 static void partition_info (char *sname, int part, struct partition *pp);
80 static void slice_info (char *sname, struct diskslice *sp);
81 static void set_ds_label (struct diskslices *ssp, int slice,
82 			      struct disklabel *lp);
83 static void set_ds_wlabel (struct diskslices *ssp, int slice, int wlabel);
84 
85 /*
86  * Duplicate a label for the whole disk, and initialize defaults in the
87  * copy for fields that are not already initialized.  The caller only
88  * needs to initialize d_secsize and d_secperunit, and zero the fields
89  * that are to be defaulted.
90  */
91 static struct disklabel *
92 clone_label(struct disklabel *lp)
93 {
94 	struct disklabel *lp1;
95 
96 	lp1 = malloc(sizeof *lp1, M_DEVBUF, M_WAITOK);
97 	*lp1 = *lp;
98 	lp = NULL;
99 	if (lp1->d_typename[0] == '\0')
100 		strncpy(lp1->d_typename, "amnesiac", sizeof(lp1->d_typename));
101 	if (lp1->d_packname[0] == '\0')
102 		strncpy(lp1->d_packname, "fictitious", sizeof(lp1->d_packname));
103 	if (lp1->d_nsectors == 0)
104 		lp1->d_nsectors = 32;
105 	if (lp1->d_ntracks == 0)
106 		lp1->d_ntracks = 64;
107 	lp1->d_secpercyl = lp1->d_nsectors * lp1->d_ntracks;
108 	lp1->d_ncylinders = lp1->d_secperunit / lp1->d_secpercyl;
109 	if (lp1->d_rpm == 0)
110 		lp1->d_rpm = 3600;
111 	if (lp1->d_interleave == 0)
112 		lp1->d_interleave = 1;
113 	if (lp1->d_npartitions < RAW_PART + 1)
114 		lp1->d_npartitions = MAXPARTITIONS;
115 	if (lp1->d_bbsize == 0)
116 		lp1->d_bbsize = BBSIZE;
117 	if (lp1->d_sbsize == 0)
118 		lp1->d_sbsize = SBSIZE;
119 	lp1->d_partitions[RAW_PART].p_size = lp1->d_secperunit;
120 	lp1->d_magic = DISKMAGIC;
121 	lp1->d_magic2 = DISKMAGIC;
122 	lp1->d_checksum = dkcksum(lp1);
123 	return (lp1);
124 }
125 
126 /*
127  * Determine the size of the transfer, and make sure it is
128  * within the boundaries of the partition. Adjust transfer
129  * if needed, and signal errors or early completion.
130  *
131  * XXX TODO:
132  *	o Split buffers that are too big for the device.
133  *	o Check for overflow.
134  *	o Finish cleaning this up.
135  *
136  * This function returns 1 on success, 0 if transfer equates
137  * to EOF (end of disk) or -1 on failure.  The appropriate
138  * 'errno' value is also set in bp->b_error and bp->b_flags
139  * is marked with B_ERROR.
140  */
141 struct bio *
142 dscheck(dev_t dev, struct bio *bio, struct diskslices *ssp)
143 {
144 	struct buf *bp = bio->bio_buf;
145 	struct bio *nbio;
146 	u_long	endsecno;
147 	daddr_t	labelsect;
148 	struct disklabel *lp;
149 	char *msg;
150 	long nsec;
151 	struct partition *pp;
152 	daddr_t	secno;
153 	daddr_t	slicerel_secno;
154 	struct diskslice *sp;
155 	int shift;
156 	int mask;
157 
158 	if (bio->bio_offset < 0) {
159 		printf("dscheck(%s): negative bio_offset %lld\n",
160 		    devtoname(dev), bio->bio_offset);
161 		bp->b_error = EINVAL;
162 		goto bad;
163 	}
164 	sp = &ssp->dss_slices[dkslice(dev)];
165 	lp = sp->ds_label;
166 
167 	if (ssp->dss_secmult == 1) {
168 		shift = DEV_BSHIFT;
169 		goto doshift;
170 	} else if (ssp->dss_secshift != -1) {
171 		shift = DEV_BSHIFT + ssp->dss_secshift;
172 doshift:
173 		mask = (1 << shift) - 1;
174 		if ((int)bp->b_bcount & mask)
175 			goto bad_bcount;
176 		if ((int)bio->bio_offset & mask)
177 			goto bad_blkno;
178 		secno = (daddr_t)(bio->bio_offset >> shift);
179 		nsec = bp->b_bcount >> shift;
180 	} else {
181 		if (bp->b_bcount % ssp->dss_secsize)
182 			goto bad_bcount;
183 		if (bio->bio_offset % ssp->dss_secsize)
184 			goto bad_blkno;
185 		secno = (daddr_t)(bio->bio_offset / ssp->dss_secsize);
186 		nsec = bp->b_bcount / ssp->dss_secsize;
187 	}
188 	if (lp == NULL) {
189 		labelsect = -LABELSECTOR - 1;
190 		endsecno = sp->ds_size;
191 		slicerel_secno = secno;
192 	} else {
193 		labelsect = lp->d_partitions[LABEL_PART].p_offset;
194 		if (labelsect != 0)
195 			Debugger("labelsect != 0 in dscheck()");
196 		pp = &lp->d_partitions[dkpart(dev)];
197 		endsecno = pp->p_size;
198 		slicerel_secno = pp->p_offset + secno;
199 	}
200 
201 	/* overwriting disk label ? */
202 	/* XXX should also protect bootstrap in first 8K */
203 	if (slicerel_secno <= LABELSECTOR + labelsect &&
204 #if LABELSECTOR != 0
205 	    slicerel_secno + nsec > LABELSECTOR + labelsect &&
206 #endif
207 	    (bp->b_flags & B_READ) == 0 && sp->ds_wlabel == 0) {
208 		bp->b_error = EROFS;
209 		goto bad;
210 	}
211 
212 #if defined(DOSBBSECTOR) && defined(notyet)
213 	/* overwriting master boot record? */
214 	if (slicerel_secno <= DOSBBSECTOR && (bp->b_flags & B_READ) == 0 &&
215 	    sp->ds_wlabel == 0) {
216 		bp->b_error = EROFS;
217 		goto bad;
218 	}
219 #endif
220 
221 	/* beyond partition? */
222 	if (secno + nsec > endsecno) {
223 		/* if exactly at end of disk, return an EOF */
224 		if (secno == endsecno) {
225 			bp->b_resid = bp->b_bcount;
226 			return (0);
227 		}
228 		/* or truncate if part of it fits */
229 		nsec = endsecno - secno;
230 		if (nsec <= 0) {
231 			bp->b_error = EINVAL;
232 			goto bad;
233 		}
234 		bp->b_bcount = nsec * ssp->dss_secsize;
235 	}
236 
237 	nbio = push_bio(bio);
238 	nbio->bio_offset = (off_t)(sp->ds_offset + slicerel_secno) * ssp->dss_secsize;
239 
240 	/*
241 	 * Snoop on label accesses if the slice offset is nonzero.  Fudge
242 	 * offsets in the label to keep the in-core label coherent with
243 	 * the on-disk one.
244 	 */
245 	if (slicerel_secno <= LABELSECTOR + labelsect
246 #if LABELSECTOR != 0
247 	    && slicerel_secno + nsec > LABELSECTOR + labelsect
248 #endif
249 	    && sp->ds_offset != 0) {
250 		nbio->bio_done = dsiodone;
251 		nbio->bio_caller_info1.ptr = sp;
252 		nbio->bio_caller_info2.offset = (off_t)(LABELSECTOR + labelsect -
253 					 slicerel_secno) * ssp->dss_secsize;
254 		if ((bp->b_flags & B_READ) == 0) {
255 			/*
256 			 * XXX even disklabel(8) writes directly so we need
257 			 * to adjust writes.  Perhaps we should drop support
258 			 * for DIOCWLABEL (always write protect labels) and
259 			 * require the use of DIOCWDINFO.
260 			 *
261 			 * XXX probably need to copy the data to avoid even
262 			 * temporarily corrupting the in-core copy.
263 			 */
264 			/* XXX need name here. */
265 			msg = fixlabel(
266 				NULL, sp,
267 			       (struct disklabel *)
268 			       (bp->b_data + (int)nbio->bio_caller_info2.offset),
269 			       TRUE);
270 			if (msg != NULL) {
271 				printf("dscheck(%s): %s\n",
272 				    devtoname(dev), msg);
273 				bp->b_error = EROFS;
274 				pop_bio(nbio);
275 				goto bad;
276 			}
277 		}
278 	}
279 	return (nbio);
280 
281 bad_bcount:
282 	printf(
283 	"dscheck(%s): b_bcount %d is not on a sector boundary (ssize %d)\n",
284 	    devtoname(dev), bp->b_bcount, ssp->dss_secsize);
285 	bp->b_error = EINVAL;
286 	goto bad;
287 
288 bad_blkno:
289 	printf(
290 	"dscheck(%s): bio_offset %lld is not on a sector boundary (ssize %d)\n",
291 	    devtoname(dev), bio->bio_offset, ssp->dss_secsize);
292 	bp->b_error = EINVAL;
293 	goto bad;
294 
295 bad:
296 	bp->b_resid = bp->b_bcount;
297 	bp->b_flags |= B_ERROR;
298 	return (NULL);
299 }
300 
301 void
302 dsclose(dev_t dev, int mode, struct diskslices *ssp)
303 {
304 	u_char mask;
305 	struct diskslice *sp;
306 
307 	sp = &ssp->dss_slices[dkslice(dev)];
308 	mask = 1 << dkpart(dev);
309 	sp->ds_openmask &= ~mask;
310 }
311 
312 void
313 dsgone(struct diskslices **sspp)
314 {
315 	int slice;
316 	struct diskslice *sp;
317 	struct diskslices *ssp;
318 
319 	for (slice = 0, ssp = *sspp; slice < ssp->dss_nslices; slice++) {
320 		sp = &ssp->dss_slices[slice];
321 		free_ds_label(ssp, slice);
322 	}
323 	free(ssp, M_DEVBUF);
324 	*sspp = NULL;
325 }
326 
327 /*
328  * For the "write" commands (DIOCSDINFO and DIOCWDINFO), this
329  * is subject to the same restriction as dsopen().
330  */
331 int
332 dsioctl(dev_t dev, u_long cmd, caddr_t data,
333 	int flags, struct diskslices **sspp)
334 {
335 	int error;
336 	struct disklabel *lp;
337 	int old_wlabel;
338 	u_char openmask;
339 	int part;
340 	int slice;
341 	struct diskslice *sp;
342 	struct diskslices *ssp;
343 	struct partition *pp;
344 
345 	slice = dkslice(dev);
346 	ssp = *sspp;
347 	sp = &ssp->dss_slices[slice];
348 	lp = sp->ds_label;
349 	switch (cmd) {
350 
351 	case DIOCGDVIRGIN:
352 		lp = (struct disklabel *)data;
353 		if (ssp->dss_slices[WHOLE_DISK_SLICE].ds_label) {
354 			*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
355 		} else {
356 			bzero(lp, sizeof(struct disklabel));
357 		}
358 
359 		lp->d_magic = DISKMAGIC;
360 		lp->d_magic2 = DISKMAGIC;
361 		pp = &lp->d_partitions[RAW_PART];
362 		pp->p_offset = 0;
363 		pp->p_size = sp->ds_size;
364 
365 		lp->d_npartitions = MAXPARTITIONS;
366 		if (lp->d_interleave == 0)
367 			lp->d_interleave = 1;
368 		if (lp->d_rpm == 0)
369 			lp->d_rpm = 3600;
370 		if (lp->d_nsectors == 0)
371 			lp->d_nsectors = 32;
372 		if (lp->d_ntracks == 0)
373 			lp->d_ntracks = 64;
374 
375 		lp->d_bbsize = BBSIZE;
376 		lp->d_sbsize = SBSIZE;
377 		lp->d_secpercyl = lp->d_nsectors * lp->d_ntracks;
378 		lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
379 		lp->d_secperunit = sp->ds_size;
380 		lp->d_checksum = 0;
381 		lp->d_checksum = dkcksum(lp);
382 		return (0);
383 
384 	case DIOCGDINFO:
385 		if (lp == NULL)
386 			return (EINVAL);
387 		*(struct disklabel *)data = *lp;
388 		return (0);
389 
390 #ifdef notyet
391 	case DIOCGDINFOP:
392 		if (lp == NULL)
393 			return (EINVAL);
394 		*(struct disklabel **)data = lp;
395 		return (0);
396 #endif
397 
398 	case DIOCGPART:
399 		if (lp == NULL)
400 			return (EINVAL);
401 		((struct partinfo *)data)->disklab = lp;
402 		((struct partinfo *)data)->part
403 			= &lp->d_partitions[dkpart(dev)];
404 		return (0);
405 
406 	case DIOCGSLICEINFO:
407 		bcopy(ssp, data, (char *)&ssp->dss_slices[ssp->dss_nslices] -
408 				 (char *)ssp);
409 		return (0);
410 
411 	case DIOCSDINFO:
412 		if (slice == WHOLE_DISK_SLICE)
413 			return (ENODEV);
414 		if (!(flags & FWRITE))
415 			return (EBADF);
416 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
417 		if (sp->ds_label == NULL)
418 			bzero(lp, sizeof *lp);
419 		else
420 			bcopy(sp->ds_label, lp, sizeof *lp);
421 		if (sp->ds_label == NULL)
422 			openmask = 0;
423 		else {
424 			openmask = sp->ds_openmask;
425 			if (slice == COMPATIBILITY_SLICE)
426 				openmask |= ssp->dss_slices[
427 				    ssp->dss_first_bsd_slice].ds_openmask;
428 			else if (slice == ssp->dss_first_bsd_slice)
429 				openmask |= ssp->dss_slices[
430 				    COMPATIBILITY_SLICE].ds_openmask;
431 		}
432 		error = setdisklabel(lp, (struct disklabel *)data,
433 				     (u_long)openmask);
434 		/* XXX why doesn't setdisklabel() check this? */
435 		if (error == 0 && lp->d_partitions[RAW_PART].p_offset != 0)
436 			error = EXDEV;
437 		if (error == 0) {
438 			if (lp->d_secperunit > sp->ds_size)
439 				error = ENOSPC;
440 			for (part = 0; part < lp->d_npartitions; part++)
441 				if (lp->d_partitions[part].p_size > sp->ds_size)
442 					error = ENOSPC;
443 		}
444 		if (error != 0) {
445 			free(lp, M_DEVBUF);
446 			return (error);
447 		}
448 		free_ds_label(ssp, slice);
449 		set_ds_label(ssp, slice, lp);
450 		return (0);
451 
452 	case DIOCSYNCSLICEINFO:
453 		if (slice != WHOLE_DISK_SLICE || dkpart(dev) != RAW_PART)
454 			return (EINVAL);
455 		if (!*(int *)data)
456 			for (slice = 0; slice < ssp->dss_nslices; slice++) {
457 				openmask = ssp->dss_slices[slice].ds_openmask;
458 				if (openmask
459 				    && (slice != WHOLE_DISK_SLICE
460 					|| openmask & ~(1 << RAW_PART)))
461 					return (EBUSY);
462 			}
463 
464 		/*
465 		 * Temporarily forget the current slices struct and read
466 		 * the current one.
467 		 * XXX should wait for current accesses on this disk to
468 		 * complete, then lock out future accesses and opens.
469 		 */
470 		*sspp = NULL;
471 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
472 		*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
473 		error = dsopen(dev, S_IFCHR, ssp->dss_oflags, sspp, lp);
474 		if (error != 0) {
475 			free(lp, M_DEVBUF);
476 			*sspp = ssp;
477 			return (error);
478 		}
479 
480 		/*
481 		 * Reopen everything.  This is a no-op except in the "force"
482 		 * case and when the raw bdev and cdev are both open.  Abort
483 		 * if anything fails.
484 		 */
485 		for (slice = 0; slice < ssp->dss_nslices; slice++) {
486 			for (openmask = ssp->dss_slices[slice].ds_openmask,
487 			     part = 0; openmask; openmask >>= 1, part++) {
488 				if (!(openmask & 1))
489 					continue;
490 				error = dsopen(dkmodslice(dkmodpart(dev, part),
491 							  slice),
492 					       S_IFCHR, ssp->dss_oflags, sspp,
493 					       lp);
494 				if (error != 0) {
495 					free(lp, M_DEVBUF);
496 					*sspp = ssp;
497 					return (EBUSY);
498 				}
499 			}
500 		}
501 
502 		free(lp, M_DEVBUF);
503 		dsgone(&ssp);
504 		return (0);
505 
506 	case DIOCWDINFO:
507 		error = dsioctl(dev, DIOCSDINFO, data, flags, &ssp);
508 		if (error != 0)
509 			return (error);
510 		/*
511 		 * XXX this used to hack on dk_openpart to fake opening
512 		 * partition 0 in case that is used instead of dkpart(dev).
513 		 */
514 		old_wlabel = sp->ds_wlabel;
515 		set_ds_wlabel(ssp, slice, TRUE);
516 		error = writedisklabel(dev, sp->ds_label);
517 		/* XXX should invalidate in-core label if write failed. */
518 		set_ds_wlabel(ssp, slice, old_wlabel);
519 		return (error);
520 
521 	case DIOCWLABEL:
522 		if (slice == WHOLE_DISK_SLICE)
523 			return (ENODEV);
524 		if (!(flags & FWRITE))
525 			return (EBADF);
526 		set_ds_wlabel(ssp, slice, *(int *)data != 0);
527 		return (0);
528 
529 	default:
530 		return (ENOIOCTL);
531 	}
532 }
533 
534 static void
535 dsiodone(struct bio *bio)
536 {
537 	struct buf *bp = bio->bio_buf;
538 	char *msg;
539 
540 	bp->b_flags = bp->b_flags & ~B_DONE;
541 	if (!(bp->b_flags & B_READ)
542 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
543 		msg = fixlabel(NULL, bio->bio_caller_info1.ptr,
544 			       (struct disklabel *)
545 			       (bp->b_data + (int)bio->bio_caller_info2.offset),
546 			       FALSE);
547 		if (msg != NULL)
548 			printf("%s\n", msg);
549 	}
550 	biodone(bio->bio_prev);
551 }
552 
553 int
554 dsisopen(struct diskslices *ssp)
555 {
556 	int slice;
557 
558 	if (ssp == NULL)
559 		return (0);
560 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
561 		if (ssp->dss_slices[slice].ds_openmask)
562 			return (1);
563 	}
564 	return (0);
565 }
566 
567 /*
568  * Allocate a slices "struct" and initialize it to contain only an empty
569  * compatibility slice (pointing to itself), a whole disk slice (covering
570  * the disk as described by the label), and (nslices - BASE_SLICES) empty
571  * slices beginning at BASE_SLICE.
572  */
573 struct diskslices *
574 dsmakeslicestruct(int nslices, struct disklabel *lp)
575 {
576 	struct diskslice *sp;
577 	struct diskslices *ssp;
578 
579 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
580 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
581 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
582 	ssp->dss_nslices = nslices;
583 	ssp->dss_oflags = 0;
584 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
585 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
586 		ssp->dss_secshift = -1;
587 	else
588 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
589 	ssp->dss_secsize = lp->d_secsize;
590 	sp = &ssp->dss_slices[0];
591 	bzero(sp, nslices * sizeof *sp);
592 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
593 	return (ssp);
594 }
595 
596 char *
597 dsname(dev_t dev, int unit, int slice, int part, char *partname)
598 {
599 	static char name[32];
600 	const char *dname;
601 
602 	dname = dev_dname(dev);
603 	if (strlen(dname) > 16)
604 		dname = "nametoolong";
605 	snprintf(name, sizeof(name), "%s%d", dname, unit);
606 	partname[0] = '\0';
607 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
608 		partname[0] = 'a' + part;
609 		partname[1] = '\0';
610 		if (slice != COMPATIBILITY_SLICE) {
611 			snprintf(name + strlen(name),
612 			    sizeof(name) - strlen(name), "s%d", slice - 1);
613 		}
614 	}
615 	return (name);
616 }
617 
618 /*
619  * This should only be called when the unit is inactive and the strategy
620  * routine should not allow it to become active unless we call it.  Our
621  * strategy routine must be special to allow activity.
622  */
623 int
624 dsopen(dev_t dev, int mode, u_int flags,
625 	struct diskslices **sspp, struct disklabel *lp)
626 {
627 	dev_t dev1;
628 	int error;
629 	struct disklabel *lp1;
630 	char *msg;
631 	u_char mask;
632 	bool_t need_init;
633 	int part;
634 	char partname[2];
635 	int slice;
636 	char *sname;
637 	struct diskslice *sp;
638 	struct diskslices *ssp;
639 	int unit;
640 
641 	dev->si_bsize_phys = lp->d_secsize;
642 
643 	unit = dkunit(dev);
644 	if (lp->d_secsize % DEV_BSIZE) {
645 		printf("%s: invalid sector size %lu\n", devtoname(dev),
646 		    (u_long)lp->d_secsize);
647 		return (EINVAL);
648 	}
649 
650 	/*
651 	 * XXX reinitialize the slice table unless there is an open device
652 	 * on the unit.  This should only be done if the media has changed.
653 	 */
654 	ssp = *sspp;
655 	need_init = !dsisopen(ssp);
656 	if (ssp != NULL && need_init)
657 		dsgone(sspp);
658 	if (need_init) {
659 		/*
660 		 * Allocate a minimal slices "struct".  This will become
661 		 * the final slices "struct" if we don't want real slices
662 		 * or if we can't find any real slices.
663 		 */
664 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
665 
666 		if (!(flags & DSO_ONESLICE)) {
667 			TRACE(("dsinit\n"));
668 			error = dsinit(dev, lp, sspp);
669 			if (error != 0) {
670 				dsgone(sspp);
671 				return (error);
672 			}
673 		}
674 		ssp = *sspp;
675 		ssp->dss_oflags = flags;
676 
677 		/*
678 		 * If there are no real slices, then make the compatiblity
679 		 * slice cover the whole disk.
680 		 */
681 		if (ssp->dss_nslices == BASE_SLICE)
682 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
683 				= lp->d_secperunit;
684 
685 		/* Point the compatibility slice at the BSD slice, if any. */
686 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
687 			sp = &ssp->dss_slices[slice];
688 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
689 				ssp->dss_first_bsd_slice = slice;
690 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
691 					= sp->ds_offset;
692 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
693 					= sp->ds_size;
694 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
695 					= sp->ds_type;
696 				break;
697 			}
698 		}
699 
700 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
701 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
702 	}
703 
704 	/*
705 	 * Initialize secondary info for all slices.  It is needed for more
706 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
707 	 */
708 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
709 		sp = &ssp->dss_slices[slice];
710 		if (sp->ds_label != NULL)
711 			continue;
712 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
713 		sname = dsname(dev, unit, slice, RAW_PART, partname);
714 		/*
715 		 * XXX this should probably only be done for the need_init
716 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
717 		 */
718 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
719 		lp1 = clone_label(lp);
720 		TRACE(("readdisklabel\n"));
721 		if (flags & DSO_NOLABELS)
722 			msg = NULL;
723 		else {
724 			msg = readdisklabel(dev1, lp1);
725 
726 			/*
727 			 * readdisklabel() returns NULL for success, and an
728 			 * error string for failure.
729 			 *
730 			 * If there isn't a label on the disk, and if the
731 			 * DSO_COMPATLABEL is set, we want to use the
732 			 * faked-up label provided by the caller.
733 			 *
734 			 * So we set msg to NULL to indicate that there is
735 			 * no failure (since we have a faked-up label),
736 			 * free lp1, and then clone it again from lp.
737 			 * (In case readdisklabel() modified lp1.)
738 			 */
739 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
740 				msg = NULL;
741 				free(lp1, M_DEVBUF);
742 				lp1 = clone_label(lp);
743 			}
744 		}
745 		if (msg == NULL)
746 			msg = fixlabel(sname, sp, lp1, FALSE);
747 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
748 			msg = "inconsistent sector size";
749 		if (msg != NULL) {
750 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
751 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
752 				    sname, msg);
753 			free(lp1, M_DEVBUF);
754 			continue;
755 		}
756 		if (lp1->d_flags & D_BADSECT) {
757 			log(LOG_ERR, "%s: bad sector table not supported\n",
758 			    sname);
759 			free(lp1, M_DEVBUF);
760 			continue;
761 		}
762 		set_ds_label(ssp, slice, lp1);
763 		set_ds_wlabel(ssp, slice, FALSE);
764 	}
765 
766 	slice = dkslice(dev);
767 	if (slice >= ssp->dss_nslices)
768 		return (ENXIO);
769 	sp = &ssp->dss_slices[slice];
770 	part = dkpart(dev);
771 	if (part != RAW_PART
772 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
773 		return (EINVAL);	/* XXX needs translation */
774 	mask = 1 << part;
775 	sp->ds_openmask |= mask;
776 	return (0);
777 }
778 
779 int
780 dssize(dev_t dev, struct diskslices **sspp)
781 {
782 	struct disklabel *lp;
783 	int part;
784 	int slice;
785 	struct diskslices *ssp;
786 
787 	slice = dkslice(dev);
788 	part = dkpart(dev);
789 	ssp = *sspp;
790 	if (ssp == NULL || slice >= ssp->dss_nslices
791 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
792 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
793 			return (-1);
794 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
795 		ssp = *sspp;
796 	}
797 	lp = ssp->dss_slices[slice].ds_label;
798 	if (lp == NULL)
799 		return (-1);
800 	return ((int)lp->d_partitions[part].p_size);
801 }
802 
803 static void
804 free_ds_label(struct diskslices *ssp, int slice)
805 {
806 	struct disklabel *lp;
807 	struct diskslice *sp;
808 
809 	sp = &ssp->dss_slices[slice];
810 	lp = sp->ds_label;
811 	if (lp == NULL)
812 		return;
813 	free(lp, M_DEVBUF);
814 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
815 }
816 
817 static char *
818 fixlabel(char *sname, struct diskslice *sp, struct disklabel *lp, int writeflag)
819 {
820 	u_long end;
821 	u_long offset;
822 	int part;
823 	struct partition *pp;
824 	u_long start;
825 	bool_t warned;
826 
827 	/* These errors "can't happen" so don't bother reporting details. */
828 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
829 		return ("fixlabel: invalid magic");
830 	if (dkcksum(lp) != 0)
831 		return ("fixlabel: invalid checksum");
832 
833 	pp = &lp->d_partitions[RAW_PART];
834 	if (writeflag) {
835 		start = 0;
836 		offset = sp->ds_offset;
837 	} else {
838 		start = sp->ds_offset;
839 		offset = -sp->ds_offset;
840 	}
841 	if (pp->p_offset != start) {
842 		if (sname != NULL) {
843 			printf(
844 "%s: rejecting BSD label: raw partition offset != slice offset\n",
845 			       sname);
846 			slice_info(sname, sp);
847 			partition_info(sname, RAW_PART, pp);
848 		}
849 		return ("fixlabel: raw partition offset != slice offset");
850 	}
851 	if (pp->p_size != sp->ds_size) {
852 		if (sname != NULL) {
853 			printf("%s: raw partition size != slice size\n", sname);
854 			slice_info(sname, sp);
855 			partition_info(sname, RAW_PART, pp);
856 		}
857 		if (pp->p_size > sp->ds_size) {
858 			if (sname == NULL)
859 				return ("fixlabel: raw partition size > slice size");
860 			printf("%s: truncating raw partition\n", sname);
861 			pp->p_size = sp->ds_size;
862 		}
863 	}
864 	end = start + sp->ds_size;
865 	if (start > end)
866 		return ("fixlabel: slice wraps");
867 	if (lp->d_secpercyl <= 0)
868 		return ("fixlabel: d_secpercyl <= 0");
869 	pp -= RAW_PART;
870 	warned = FALSE;
871 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
872 		if (pp->p_offset != 0 || pp->p_size != 0) {
873 			if (pp->p_offset < start
874 			    || pp->p_offset + pp->p_size > end
875 			    || pp->p_offset + pp->p_size < pp->p_offset) {
876 				if (sname != NULL) {
877 					printf(
878 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
879 					       sname);
880 					if (!warned) {
881 						slice_info(sname, sp);
882 						warned = TRUE;
883 					}
884 					partition_info(sname, part, pp);
885 				}
886 				/* XXX else silently discard junk. */
887 				bzero(pp, sizeof *pp);
888 			} else
889 				pp->p_offset += offset;
890 		}
891 	}
892 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
893 	lp->d_secperunit = sp->ds_size;
894  	lp->d_checksum = 0;
895  	lp->d_checksum = dkcksum(lp);
896 	return (NULL);
897 }
898 
899 static void
900 partition_info(char *sname, int part, struct partition *pp)
901 {
902 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
903 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
904 	       (u_long)pp->p_size);
905 }
906 
907 static void
908 slice_info(char *sname, struct diskslice *sp)
909 {
910 	printf("%s: start %lu, end %lu, size %lu\n", sname,
911 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
912 }
913 
914 static void
915 set_ds_label(struct diskslices *ssp, int slice, struct disklabel *lp)
916 {
917 	ssp->dss_slices[slice].ds_label = lp;
918 	if (slice == COMPATIBILITY_SLICE)
919 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
920 	else if (slice == ssp->dss_first_bsd_slice)
921 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
922 }
923 
924 static void
925 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
926 {
927 	ssp->dss_slices[slice].ds_wlabel = wlabel;
928 	if (slice == COMPATIBILITY_SLICE)
929 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
930 	else if (slice == ssp->dss_first_bsd_slice)
931 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
932 }
933