xref: /dragonfly/sys/kern/subr_diskslice.c (revision 6bd457ed)
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 1.11 2005/08/03 16:36:33 hmp 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 buf *bp);
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 int
136 dscheck(struct buf *bp, struct diskslices *ssp)
137 {
138 	daddr_t	blkno;
139 	u_long	endsecno;
140 	daddr_t	labelsect;
141 	struct disklabel *lp;
142 	char *msg;
143 	long nsec;
144 	struct partition *pp;
145 	daddr_t	secno;
146 	daddr_t	slicerel_secno;
147 	struct diskslice *sp;
148 
149 	blkno = bp->b_blkno;
150 	if (blkno < 0) {
151 		printf("dscheck(%s): negative b_blkno %ld\n",
152 		    devtoname(bp->b_dev), (long)blkno);
153 		bp->b_error = EINVAL;
154 		goto bad;
155 	}
156 	sp = &ssp->dss_slices[dkslice(bp->b_dev)];
157 	lp = sp->ds_label;
158 	if (ssp->dss_secmult == 1) {
159 		if (bp->b_bcount % (u_long)DEV_BSIZE)
160 			goto bad_bcount;
161 		secno = blkno;
162 		nsec = bp->b_bcount >> DEV_BSHIFT;
163 	} else if (ssp->dss_secshift != -1) {
164 		if (bp->b_bcount & (ssp->dss_secsize - 1))
165 			goto bad_bcount;
166 		if (blkno & (ssp->dss_secmult - 1))
167 			goto bad_blkno;
168 		secno = blkno >> ssp->dss_secshift;
169 		nsec = bp->b_bcount >> (DEV_BSHIFT + ssp->dss_secshift);
170 	} else {
171 		if (bp->b_bcount % ssp->dss_secsize)
172 			goto bad_bcount;
173 		if (blkno % ssp->dss_secmult)
174 			goto bad_blkno;
175 		secno = blkno / ssp->dss_secmult;
176 		nsec = bp->b_bcount / ssp->dss_secsize;
177 	}
178 	if (lp == NULL) {
179 		labelsect = -LABELSECTOR - 1;
180 		endsecno = sp->ds_size;
181 		slicerel_secno = secno;
182 	} else {
183 		labelsect = lp->d_partitions[LABEL_PART].p_offset;
184 		if (labelsect != 0)
185 			Debugger("labelsect != 0 in dscheck()");
186 		pp = &lp->d_partitions[dkpart(bp->b_dev)];
187 		endsecno = pp->p_size;
188 		slicerel_secno = pp->p_offset + secno;
189 	}
190 
191 	/* overwriting disk label ? */
192 	/* XXX should also protect bootstrap in first 8K */
193 	if (slicerel_secno <= LABELSECTOR + labelsect &&
194 #if LABELSECTOR != 0
195 	    slicerel_secno + nsec > LABELSECTOR + labelsect &&
196 #endif
197 	    (bp->b_flags & B_READ) == 0 && sp->ds_wlabel == 0) {
198 		bp->b_error = EROFS;
199 		goto bad;
200 	}
201 
202 #if defined(DOSBBSECTOR) && defined(notyet)
203 	/* overwriting master boot record? */
204 	if (slicerel_secno <= DOSBBSECTOR && (bp->b_flags & B_READ) == 0 &&
205 	    sp->ds_wlabel == 0) {
206 		bp->b_error = EROFS;
207 		goto bad;
208 	}
209 #endif
210 
211 	/* beyond partition? */
212 	if (secno + nsec > endsecno) {
213 		/* if exactly at end of disk, return an EOF */
214 		if (secno == endsecno) {
215 			bp->b_resid = bp->b_bcount;
216 			return (0);
217 		}
218 		/* or truncate if part of it fits */
219 		nsec = endsecno - secno;
220 		if (nsec <= 0) {
221 			bp->b_error = EINVAL;
222 			goto bad;
223 		}
224 		bp->b_bcount = nsec * ssp->dss_secsize;
225 	}
226 
227 	bp->b_pblkno = sp->ds_offset + slicerel_secno;
228 
229 	/*
230 	 * Snoop on label accesses if the slice offset is nonzero.  Fudge
231 	 * offsets in the label to keep the in-core label coherent with
232 	 * the on-disk one.
233 	 */
234 	if (slicerel_secno <= LABELSECTOR + labelsect
235 #if LABELSECTOR != 0
236 	    && slicerel_secno + nsec > LABELSECTOR + labelsect
237 #endif
238 	    && sp->ds_offset != 0) {
239 		struct iodone_chain *ic;
240 
241 		ic = malloc(sizeof *ic , M_DEVBUF, M_WAITOK);
242 		ic->ic_prev_flags = bp->b_flags;
243 		ic->ic_prev_iodone = bp->b_iodone;
244 		ic->ic_prev_iodone_chain = bp->b_iodone_chain;
245 		ic->ic_args[0].ia_long = (LABELSECTOR + labelsect -
246 		    slicerel_secno) * ssp->dss_secsize;
247 		ic->ic_args[1].ia_ptr = sp;
248 		bp->b_iodone = dsiodone;
249 		bp->b_iodone_chain = ic;
250 		if (!(bp->b_flags & B_READ)) {
251 			/*
252 			 * XXX even disklabel(8) writes directly so we need
253 			 * to adjust writes.  Perhaps we should drop support
254 			 * for DIOCWLABEL (always write protect labels) and
255 			 * require the use of DIOCWDINFO.
256 			 *
257 			 * XXX probably need to copy the data to avoid even
258 			 * temporarily corrupting the in-core copy.
259 			 */
260 			if (bp->b_vp != NULL) {
261 				crit_enter();
262 				bp->b_vp->v_numoutput++;
263 				crit_exit();
264 			}
265 			/* XXX need name here. */
266 			msg = fixlabel((char *)NULL, sp,
267 				       (struct disklabel *)
268 				       (bp->b_data + ic->ic_args[0].ia_long),
269 				       TRUE);
270 			if (msg != NULL) {
271 				printf("dscheck(%s): %s\n",
272 				    devtoname(bp->b_dev), msg);
273 				bp->b_error = EROFS;
274 				goto bad;
275 			}
276 		}
277 	}
278 	return (1);
279 
280 bad_bcount:
281 	printf(
282 	"dscheck(%s): b_bcount %ld is not on a sector boundary (ssize %d)\n",
283 	    devtoname(bp->b_dev), bp->b_bcount, ssp->dss_secsize);
284 	bp->b_error = EINVAL;
285 	goto bad;
286 
287 bad_blkno:
288 	printf(
289 	"dscheck(%s): b_blkno %ld is not on a sector boundary (ssize %d)\n",
290 	    devtoname(bp->b_dev), (long)blkno, 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 (-1);
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 buf *bp)
535 {
536 	struct iodone_chain *ic;
537 	char *msg;
538 
539 	ic = bp->b_iodone_chain;
540 	bp->b_flags = bp->b_flags & ~B_DONE;
541 	bp->b_iodone = ic->ic_prev_iodone;
542 	bp->b_iodone_chain = ic->ic_prev_iodone_chain;
543 	if (!(bp->b_flags & B_READ)
544 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
545 		msg = fixlabel((char *)NULL, ic->ic_args[1].ia_ptr,
546 			       (struct disklabel *)
547 			       (bp->b_data + ic->ic_args[0].ia_long),
548 			       FALSE);
549 		if (msg != NULL)
550 			printf("%s\n", msg);
551 	}
552 	free(ic, M_DEVBUF);
553 	biodone(bp);
554 }
555 
556 int
557 dsisopen(struct diskslices *ssp)
558 {
559 	int slice;
560 
561 	if (ssp == NULL)
562 		return (0);
563 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
564 		if (ssp->dss_slices[slice].ds_openmask)
565 			return (1);
566 	}
567 	return (0);
568 }
569 
570 /*
571  * Allocate a slices "struct" and initialize it to contain only an empty
572  * compatibility slice (pointing to itself), a whole disk slice (covering
573  * the disk as described by the label), and (nslices - BASE_SLICES) empty
574  * slices beginning at BASE_SLICE.
575  */
576 struct diskslices *
577 dsmakeslicestruct(int nslices, struct disklabel *lp)
578 {
579 	struct diskslice *sp;
580 	struct diskslices *ssp;
581 
582 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
583 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
584 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
585 	ssp->dss_nslices = nslices;
586 	ssp->dss_oflags = 0;
587 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
588 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
589 		ssp->dss_secshift = -1;
590 	else
591 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
592 	ssp->dss_secsize = lp->d_secsize;
593 	sp = &ssp->dss_slices[0];
594 	bzero(sp, nslices * sizeof *sp);
595 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
596 	return (ssp);
597 }
598 
599 char *
600 dsname(dev_t dev, int unit, int slice, int part, char *partname)
601 {
602 	static char name[32];
603 	const char *dname;
604 
605 	dname = dev_dname(dev);
606 	if (strlen(dname) > 16)
607 		dname = "nametoolong";
608 	snprintf(name, sizeof(name), "%s%d", dname, unit);
609 	partname[0] = '\0';
610 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
611 		partname[0] = 'a' + part;
612 		partname[1] = '\0';
613 		if (slice != COMPATIBILITY_SLICE) {
614 			snprintf(name + strlen(name),
615 			    sizeof(name) - strlen(name), "s%d", slice - 1);
616 		}
617 	}
618 	return (name);
619 }
620 
621 /*
622  * This should only be called when the unit is inactive and the strategy
623  * routine should not allow it to become active unless we call it.  Our
624  * strategy routine must be special to allow activity.
625  */
626 int
627 dsopen(dev_t dev, int mode, u_int flags,
628 	struct diskslices **sspp, struct disklabel *lp)
629 {
630 	dev_t dev1;
631 	int error;
632 	struct disklabel *lp1;
633 	char *msg;
634 	u_char mask;
635 	bool_t need_init;
636 	int part;
637 	char partname[2];
638 	int slice;
639 	char *sname;
640 	struct diskslice *sp;
641 	struct diskslices *ssp;
642 	int unit;
643 
644 	dev->si_bsize_phys = lp->d_secsize;
645 
646 	unit = dkunit(dev);
647 	if (lp->d_secsize % DEV_BSIZE) {
648 		printf("%s: invalid sector size %lu\n", devtoname(dev),
649 		    (u_long)lp->d_secsize);
650 		return (EINVAL);
651 	}
652 
653 	/*
654 	 * XXX reinitialize the slice table unless there is an open device
655 	 * on the unit.  This should only be done if the media has changed.
656 	 */
657 	ssp = *sspp;
658 	need_init = !dsisopen(ssp);
659 	if (ssp != NULL && need_init)
660 		dsgone(sspp);
661 	if (need_init) {
662 		/*
663 		 * Allocate a minimal slices "struct".  This will become
664 		 * the final slices "struct" if we don't want real slices
665 		 * or if we can't find any real slices.
666 		 */
667 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
668 
669 		if (!(flags & DSO_ONESLICE)) {
670 			TRACE(("dsinit\n"));
671 			error = dsinit(dev, lp, sspp);
672 			if (error != 0) {
673 				dsgone(sspp);
674 				return (error);
675 			}
676 		}
677 		ssp = *sspp;
678 		ssp->dss_oflags = flags;
679 
680 		/*
681 		 * If there are no real slices, then make the compatiblity
682 		 * slice cover the whole disk.
683 		 */
684 		if (ssp->dss_nslices == BASE_SLICE)
685 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
686 				= lp->d_secperunit;
687 
688 		/* Point the compatibility slice at the BSD slice, if any. */
689 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
690 			sp = &ssp->dss_slices[slice];
691 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
692 				ssp->dss_first_bsd_slice = slice;
693 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
694 					= sp->ds_offset;
695 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
696 					= sp->ds_size;
697 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
698 					= sp->ds_type;
699 				break;
700 			}
701 		}
702 
703 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
704 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
705 	}
706 
707 	/*
708 	 * Initialize secondary info for all slices.  It is needed for more
709 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
710 	 */
711 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
712 		sp = &ssp->dss_slices[slice];
713 		if (sp->ds_label != NULL)
714 			continue;
715 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
716 		sname = dsname(dev, unit, slice, RAW_PART, partname);
717 		/*
718 		 * XXX this should probably only be done for the need_init
719 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
720 		 */
721 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
722 		lp1 = clone_label(lp);
723 		TRACE(("readdisklabel\n"));
724 		if (flags & DSO_NOLABELS)
725 			msg = NULL;
726 		else {
727 			msg = readdisklabel(dev1, lp1);
728 
729 			/*
730 			 * readdisklabel() returns NULL for success, and an
731 			 * error string for failure.
732 			 *
733 			 * If there isn't a label on the disk, and if the
734 			 * DSO_COMPATLABEL is set, we want to use the
735 			 * faked-up label provided by the caller.
736 			 *
737 			 * So we set msg to NULL to indicate that there is
738 			 * no failure (since we have a faked-up label),
739 			 * free lp1, and then clone it again from lp.
740 			 * (In case readdisklabel() modified lp1.)
741 			 */
742 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
743 				msg = NULL;
744 				free(lp1, M_DEVBUF);
745 				lp1 = clone_label(lp);
746 			}
747 		}
748 		if (msg == NULL)
749 			msg = fixlabel(sname, sp, lp1, FALSE);
750 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
751 			msg = "inconsistent sector size";
752 		if (msg != NULL) {
753 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
754 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
755 				    sname, msg);
756 			free(lp1, M_DEVBUF);
757 			continue;
758 		}
759 		if (lp1->d_flags & D_BADSECT) {
760 			log(LOG_ERR, "%s: bad sector table not supported\n",
761 			    sname);
762 			free(lp1, M_DEVBUF);
763 			continue;
764 		}
765 		set_ds_label(ssp, slice, lp1);
766 		set_ds_wlabel(ssp, slice, FALSE);
767 	}
768 
769 	slice = dkslice(dev);
770 	if (slice >= ssp->dss_nslices)
771 		return (ENXIO);
772 	sp = &ssp->dss_slices[slice];
773 	part = dkpart(dev);
774 	if (part != RAW_PART
775 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
776 		return (EINVAL);	/* XXX needs translation */
777 	mask = 1 << part;
778 	sp->ds_openmask |= mask;
779 	return (0);
780 }
781 
782 int
783 dssize(dev_t dev, struct diskslices **sspp)
784 {
785 	struct disklabel *lp;
786 	int part;
787 	int slice;
788 	struct diskslices *ssp;
789 
790 	slice = dkslice(dev);
791 	part = dkpart(dev);
792 	ssp = *sspp;
793 	if (ssp == NULL || slice >= ssp->dss_nslices
794 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
795 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
796 			return (-1);
797 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
798 		ssp = *sspp;
799 	}
800 	lp = ssp->dss_slices[slice].ds_label;
801 	if (lp == NULL)
802 		return (-1);
803 	return ((int)lp->d_partitions[part].p_size);
804 }
805 
806 static void
807 free_ds_label(struct diskslices *ssp, int slice)
808 {
809 	struct disklabel *lp;
810 	struct diskslice *sp;
811 
812 	sp = &ssp->dss_slices[slice];
813 	lp = sp->ds_label;
814 	if (lp == NULL)
815 		return;
816 	free(lp, M_DEVBUF);
817 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
818 }
819 
820 static char *
821 fixlabel(char *sname, struct diskslice *sp, struct disklabel *lp, int writeflag)
822 {
823 	u_long end;
824 	u_long offset;
825 	int part;
826 	struct partition *pp;
827 	u_long start;
828 	bool_t warned;
829 
830 	/* These errors "can't happen" so don't bother reporting details. */
831 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
832 		return ("fixlabel: invalid magic");
833 	if (dkcksum(lp) != 0)
834 		return ("fixlabel: invalid checksum");
835 
836 	pp = &lp->d_partitions[RAW_PART];
837 	if (writeflag) {
838 		start = 0;
839 		offset = sp->ds_offset;
840 	} else {
841 		start = sp->ds_offset;
842 		offset = -sp->ds_offset;
843 	}
844 	if (pp->p_offset != start) {
845 		if (sname != NULL) {
846 			printf(
847 "%s: rejecting BSD label: raw partition offset != slice offset\n",
848 			       sname);
849 			slice_info(sname, sp);
850 			partition_info(sname, RAW_PART, pp);
851 		}
852 		return ("fixlabel: raw partition offset != slice offset");
853 	}
854 	if (pp->p_size != sp->ds_size) {
855 		if (sname != NULL) {
856 			printf("%s: raw partition size != slice size\n", sname);
857 			slice_info(sname, sp);
858 			partition_info(sname, RAW_PART, pp);
859 		}
860 		if (pp->p_size > sp->ds_size) {
861 			if (sname == NULL)
862 				return ("fixlabel: raw partition size > slice size");
863 			printf("%s: truncating raw partition\n", sname);
864 			pp->p_size = sp->ds_size;
865 		}
866 	}
867 	end = start + sp->ds_size;
868 	if (start > end)
869 		return ("fixlabel: slice wraps");
870 	if (lp->d_secpercyl <= 0)
871 		return ("fixlabel: d_secpercyl <= 0");
872 	pp -= RAW_PART;
873 	warned = FALSE;
874 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
875 		if (pp->p_offset != 0 || pp->p_size != 0) {
876 			if (pp->p_offset < start
877 			    || pp->p_offset + pp->p_size > end
878 			    || pp->p_offset + pp->p_size < pp->p_offset) {
879 				if (sname != NULL) {
880 					printf(
881 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
882 					       sname);
883 					if (!warned) {
884 						slice_info(sname, sp);
885 						warned = TRUE;
886 					}
887 					partition_info(sname, part, pp);
888 				}
889 				/* XXX else silently discard junk. */
890 				bzero(pp, sizeof *pp);
891 			} else
892 				pp->p_offset += offset;
893 		}
894 	}
895 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
896 	lp->d_secperunit = sp->ds_size;
897  	lp->d_checksum = 0;
898  	lp->d_checksum = dkcksum(lp);
899 	return (NULL);
900 }
901 
902 static void
903 partition_info(char *sname, int part, struct partition *pp)
904 {
905 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
906 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
907 	       (u_long)pp->p_size);
908 }
909 
910 static void
911 slice_info(char *sname, struct diskslice *sp)
912 {
913 	printf("%s: start %lu, end %lu, size %lu\n", sname,
914 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
915 }
916 
917 static void
918 set_ds_label(struct diskslices *ssp, int slice, struct disklabel *lp)
919 {
920 	ssp->dss_slices[slice].ds_label = lp;
921 	if (slice == COMPATIBILITY_SLICE)
922 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
923 	else if (slice == ssp->dss_first_bsd_slice)
924 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
925 }
926 
927 static void
928 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
929 {
930 	ssp->dss_slices[slice].ds_wlabel = wlabel;
931 	if (slice == COMPATIBILITY_SLICE)
932 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
933 	else if (slice == ssp->dss_first_bsd_slice)
934 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
935 }
936