xref: /dragonfly/sys/kern/subr_diskslice.c (revision 1bf4b486)
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 1.10 2005/06/06 15:02:28 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 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_flags |= B_CALL;
249 		bp->b_iodone = dsiodone;
250 		bp->b_iodone_chain = ic;
251 		if (!(bp->b_flags & B_READ)) {
252 			/*
253 			 * XXX even disklabel(8) writes directly so we need
254 			 * to adjust writes.  Perhaps we should drop support
255 			 * for DIOCWLABEL (always write protect labels) and
256 			 * require the use of DIOCWDINFO.
257 			 *
258 			 * XXX probably need to copy the data to avoid even
259 			 * temporarily corrupting the in-core copy.
260 			 */
261 			if (bp->b_vp != NULL) {
262 				crit_enter();
263 				bp->b_vp->v_numoutput++;
264 				crit_exit();
265 			}
266 			/* XXX need name here. */
267 			msg = fixlabel((char *)NULL, sp,
268 				       (struct disklabel *)
269 				       (bp->b_data + ic->ic_args[0].ia_long),
270 				       TRUE);
271 			if (msg != NULL) {
272 				printf("dscheck(%s): %s\n",
273 				    devtoname(bp->b_dev), msg);
274 				bp->b_error = EROFS;
275 				goto bad;
276 			}
277 		}
278 	}
279 	return (1);
280 
281 bad_bcount:
282 	printf(
283 	"dscheck(%s): b_bcount %ld is not on a sector boundary (ssize %d)\n",
284 	    devtoname(bp->b_dev), bp->b_bcount, ssp->dss_secsize);
285 	bp->b_error = EINVAL;
286 	goto bad;
287 
288 bad_blkno:
289 	printf(
290 	"dscheck(%s): b_blkno %ld is not on a sector boundary (ssize %d)\n",
291 	    devtoname(bp->b_dev), (long)blkno, 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 (-1);
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 buf *bp)
536 {
537 	struct iodone_chain *ic;
538 	char *msg;
539 
540 	ic = bp->b_iodone_chain;
541 	bp->b_flags = (ic->ic_prev_flags & B_CALL)
542 		      | (bp->b_flags & ~(B_CALL | B_DONE));
543 	bp->b_iodone = ic->ic_prev_iodone;
544 	bp->b_iodone_chain = ic->ic_prev_iodone_chain;
545 	if (!(bp->b_flags & B_READ)
546 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
547 		msg = fixlabel((char *)NULL, ic->ic_args[1].ia_ptr,
548 			       (struct disklabel *)
549 			       (bp->b_data + ic->ic_args[0].ia_long),
550 			       FALSE);
551 		if (msg != NULL)
552 			printf("%s\n", msg);
553 	}
554 	free(ic, M_DEVBUF);
555 	biodone(bp);
556 }
557 
558 int
559 dsisopen(struct diskslices *ssp)
560 {
561 	int slice;
562 
563 	if (ssp == NULL)
564 		return (0);
565 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
566 		if (ssp->dss_slices[slice].ds_openmask)
567 			return (1);
568 	}
569 	return (0);
570 }
571 
572 /*
573  * Allocate a slices "struct" and initialize it to contain only an empty
574  * compatibility slice (pointing to itself), a whole disk slice (covering
575  * the disk as described by the label), and (nslices - BASE_SLICES) empty
576  * slices beginning at BASE_SLICE.
577  */
578 struct diskslices *
579 dsmakeslicestruct(int nslices, struct disklabel *lp)
580 {
581 	struct diskslice *sp;
582 	struct diskslices *ssp;
583 
584 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
585 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
586 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
587 	ssp->dss_nslices = nslices;
588 	ssp->dss_oflags = 0;
589 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
590 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
591 		ssp->dss_secshift = -1;
592 	else
593 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
594 	ssp->dss_secsize = lp->d_secsize;
595 	sp = &ssp->dss_slices[0];
596 	bzero(sp, nslices * sizeof *sp);
597 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
598 	return (ssp);
599 }
600 
601 char *
602 dsname(dev_t dev, int unit, int slice, int part, char *partname)
603 {
604 	static char name[32];
605 	const char *dname;
606 
607 	dname = dev_dname(dev);
608 	if (strlen(dname) > 16)
609 		dname = "nametoolong";
610 	snprintf(name, sizeof(name), "%s%d", dname, unit);
611 	partname[0] = '\0';
612 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
613 		partname[0] = 'a' + part;
614 		partname[1] = '\0';
615 		if (slice != COMPATIBILITY_SLICE) {
616 			snprintf(name + strlen(name),
617 			    sizeof(name) - strlen(name), "s%d", slice - 1);
618 		}
619 	}
620 	return (name);
621 }
622 
623 /*
624  * This should only be called when the unit is inactive and the strategy
625  * routine should not allow it to become active unless we call it.  Our
626  * strategy routine must be special to allow activity.
627  */
628 int
629 dsopen(dev_t dev, int mode, u_int flags,
630 	struct diskslices **sspp, struct disklabel *lp)
631 {
632 	dev_t dev1;
633 	int error;
634 	struct disklabel *lp1;
635 	char *msg;
636 	u_char mask;
637 	bool_t need_init;
638 	int part;
639 	char partname[2];
640 	int slice;
641 	char *sname;
642 	struct diskslice *sp;
643 	struct diskslices *ssp;
644 	int unit;
645 
646 	dev->si_bsize_phys = lp->d_secsize;
647 
648 	unit = dkunit(dev);
649 	if (lp->d_secsize % DEV_BSIZE) {
650 		printf("%s: invalid sector size %lu\n", devtoname(dev),
651 		    (u_long)lp->d_secsize);
652 		return (EINVAL);
653 	}
654 
655 	/*
656 	 * XXX reinitialize the slice table unless there is an open device
657 	 * on the unit.  This should only be done if the media has changed.
658 	 */
659 	ssp = *sspp;
660 	need_init = !dsisopen(ssp);
661 	if (ssp != NULL && need_init)
662 		dsgone(sspp);
663 	if (need_init) {
664 		/*
665 		 * Allocate a minimal slices "struct".  This will become
666 		 * the final slices "struct" if we don't want real slices
667 		 * or if we can't find any real slices.
668 		 */
669 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
670 
671 		if (!(flags & DSO_ONESLICE)) {
672 			TRACE(("dsinit\n"));
673 			error = dsinit(dev, lp, sspp);
674 			if (error != 0) {
675 				dsgone(sspp);
676 				return (error);
677 			}
678 		}
679 		ssp = *sspp;
680 		ssp->dss_oflags = flags;
681 
682 		/*
683 		 * If there are no real slices, then make the compatiblity
684 		 * slice cover the whole disk.
685 		 */
686 		if (ssp->dss_nslices == BASE_SLICE)
687 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
688 				= lp->d_secperunit;
689 
690 		/* Point the compatibility slice at the BSD slice, if any. */
691 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
692 			sp = &ssp->dss_slices[slice];
693 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
694 				ssp->dss_first_bsd_slice = slice;
695 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
696 					= sp->ds_offset;
697 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
698 					= sp->ds_size;
699 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
700 					= sp->ds_type;
701 				break;
702 			}
703 		}
704 
705 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
706 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
707 	}
708 
709 	/*
710 	 * Initialize secondary info for all slices.  It is needed for more
711 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
712 	 */
713 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
714 		sp = &ssp->dss_slices[slice];
715 		if (sp->ds_label != NULL)
716 			continue;
717 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
718 		sname = dsname(dev, unit, slice, RAW_PART, partname);
719 		/*
720 		 * XXX this should probably only be done for the need_init
721 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
722 		 */
723 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
724 		lp1 = clone_label(lp);
725 		TRACE(("readdisklabel\n"));
726 		if (flags & DSO_NOLABELS)
727 			msg = NULL;
728 		else {
729 			msg = readdisklabel(dev1, lp1);
730 
731 			/*
732 			 * readdisklabel() returns NULL for success, and an
733 			 * error string for failure.
734 			 *
735 			 * If there isn't a label on the disk, and if the
736 			 * DSO_COMPATLABEL is set, we want to use the
737 			 * faked-up label provided by the caller.
738 			 *
739 			 * So we set msg to NULL to indicate that there is
740 			 * no failure (since we have a faked-up label),
741 			 * free lp1, and then clone it again from lp.
742 			 * (In case readdisklabel() modified lp1.)
743 			 */
744 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
745 				msg = NULL;
746 				free(lp1, M_DEVBUF);
747 				lp1 = clone_label(lp);
748 			}
749 		}
750 		if (msg == NULL)
751 			msg = fixlabel(sname, sp, lp1, FALSE);
752 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
753 			msg = "inconsistent sector size";
754 		if (msg != NULL) {
755 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
756 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
757 				    sname, msg);
758 			free(lp1, M_DEVBUF);
759 			continue;
760 		}
761 		if (lp1->d_flags & D_BADSECT) {
762 			log(LOG_ERR, "%s: bad sector table not supported\n",
763 			    sname);
764 			free(lp1, M_DEVBUF);
765 			continue;
766 		}
767 		set_ds_label(ssp, slice, lp1);
768 		set_ds_wlabel(ssp, slice, FALSE);
769 	}
770 
771 	slice = dkslice(dev);
772 	if (slice >= ssp->dss_nslices)
773 		return (ENXIO);
774 	sp = &ssp->dss_slices[slice];
775 	part = dkpart(dev);
776 	if (part != RAW_PART
777 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
778 		return (EINVAL);	/* XXX needs translation */
779 	mask = 1 << part;
780 	sp->ds_openmask |= mask;
781 	return (0);
782 }
783 
784 int
785 dssize(dev_t dev, struct diskslices **sspp)
786 {
787 	struct disklabel *lp;
788 	int part;
789 	int slice;
790 	struct diskslices *ssp;
791 
792 	slice = dkslice(dev);
793 	part = dkpart(dev);
794 	ssp = *sspp;
795 	if (ssp == NULL || slice >= ssp->dss_nslices
796 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
797 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
798 			return (-1);
799 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
800 		ssp = *sspp;
801 	}
802 	lp = ssp->dss_slices[slice].ds_label;
803 	if (lp == NULL)
804 		return (-1);
805 	return ((int)lp->d_partitions[part].p_size);
806 }
807 
808 static void
809 free_ds_label(struct diskslices *ssp, int slice)
810 {
811 	struct disklabel *lp;
812 	struct diskslice *sp;
813 
814 	sp = &ssp->dss_slices[slice];
815 	lp = sp->ds_label;
816 	if (lp == NULL)
817 		return;
818 	free(lp, M_DEVBUF);
819 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
820 }
821 
822 static char *
823 fixlabel(char *sname, struct diskslice *sp, struct disklabel *lp, int writeflag)
824 {
825 	u_long end;
826 	u_long offset;
827 	int part;
828 	struct partition *pp;
829 	u_long start;
830 	bool_t warned;
831 
832 	/* These errors "can't happen" so don't bother reporting details. */
833 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
834 		return ("fixlabel: invalid magic");
835 	if (dkcksum(lp) != 0)
836 		return ("fixlabel: invalid checksum");
837 
838 	pp = &lp->d_partitions[RAW_PART];
839 	if (writeflag) {
840 		start = 0;
841 		offset = sp->ds_offset;
842 	} else {
843 		start = sp->ds_offset;
844 		offset = -sp->ds_offset;
845 	}
846 	if (pp->p_offset != start) {
847 		if (sname != NULL) {
848 			printf(
849 "%s: rejecting BSD label: raw partition offset != slice offset\n",
850 			       sname);
851 			slice_info(sname, sp);
852 			partition_info(sname, RAW_PART, pp);
853 		}
854 		return ("fixlabel: raw partition offset != slice offset");
855 	}
856 	if (pp->p_size != sp->ds_size) {
857 		if (sname != NULL) {
858 			printf("%s: raw partition size != slice size\n", sname);
859 			slice_info(sname, sp);
860 			partition_info(sname, RAW_PART, pp);
861 		}
862 		if (pp->p_size > sp->ds_size) {
863 			if (sname == NULL)
864 				return ("fixlabel: raw partition size > slice size");
865 			printf("%s: truncating raw partition\n", sname);
866 			pp->p_size = sp->ds_size;
867 		}
868 	}
869 	end = start + sp->ds_size;
870 	if (start > end)
871 		return ("fixlabel: slice wraps");
872 	if (lp->d_secpercyl <= 0)
873 		return ("fixlabel: d_secpercyl <= 0");
874 	pp -= RAW_PART;
875 	warned = FALSE;
876 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
877 		if (pp->p_offset != 0 || pp->p_size != 0) {
878 			if (pp->p_offset < start
879 			    || pp->p_offset + pp->p_size > end
880 			    || pp->p_offset + pp->p_size < pp->p_offset) {
881 				if (sname != NULL) {
882 					printf(
883 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
884 					       sname);
885 					if (!warned) {
886 						slice_info(sname, sp);
887 						warned = TRUE;
888 					}
889 					partition_info(sname, part, pp);
890 				}
891 				/* XXX else silently discard junk. */
892 				bzero(pp, sizeof *pp);
893 			} else
894 				pp->p_offset += offset;
895 		}
896 	}
897 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
898 	lp->d_secperunit = sp->ds_size;
899  	lp->d_checksum = 0;
900  	lp->d_checksum = dkcksum(lp);
901 	return (NULL);
902 }
903 
904 static void
905 partition_info(char *sname, int part, struct partition *pp)
906 {
907 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
908 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
909 	       (u_long)pp->p_size);
910 }
911 
912 static void
913 slice_info(char *sname, struct diskslice *sp)
914 {
915 	printf("%s: start %lu, end %lu, size %lu\n", sname,
916 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
917 }
918 
919 static void
920 set_ds_label(struct diskslices *ssp, int slice, struct disklabel *lp)
921 {
922 	ssp->dss_slices[slice].ds_label = lp;
923 	if (slice == COMPATIBILITY_SLICE)
924 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
925 	else if (slice == ssp->dss_first_bsd_slice)
926 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
927 }
928 
929 static void
930 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
931 {
932 	ssp->dss_slices[slice].ds_wlabel = wlabel;
933 	if (slice == COMPATIBILITY_SLICE)
934 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
935 	else if (slice == ssp->dss_first_bsd_slice)
936 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
937 }
938