xref: /dragonfly/sys/kern/subr_diskslice.c (revision 71126e33)
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 1.8 2004/06/02 17:18:43 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 
64 #include <vfs/ufs/fs.h>
65 
66 #define TRACE(str)	do { if (ds_debug) printf str; } while (0)
67 
68 typedef	u_char	bool_t;
69 
70 static volatile bool_t ds_debug;
71 
72 static struct disklabel *clone_label (struct disklabel *lp);
73 static void dsiodone (struct buf *bp);
74 static char *fixlabel (char *sname, struct diskslice *sp,
75 			   struct disklabel *lp, int writeflag);
76 static void free_ds_label (struct diskslices *ssp, int slice);
77 static void partition_info (char *sname, int part, struct partition *pp);
78 static void slice_info (char *sname, struct diskslice *sp);
79 static void set_ds_label (struct diskslices *ssp, int slice,
80 			      struct disklabel *lp);
81 static void set_ds_wlabel (struct diskslices *ssp, int slice, int wlabel);
82 
83 /*
84  * Duplicate a label for the whole disk, and initialize defaults in the
85  * copy for fields that are not already initialized.  The caller only
86  * needs to initialize d_secsize and d_secperunit, and zero the fields
87  * that are to be defaulted.
88  */
89 static struct disklabel *
90 clone_label(struct disklabel *lp)
91 {
92 	struct disklabel *lp1;
93 
94 	lp1 = malloc(sizeof *lp1, M_DEVBUF, M_WAITOK);
95 	*lp1 = *lp;
96 	lp = NULL;
97 	if (lp1->d_typename[0] == '\0')
98 		strncpy(lp1->d_typename, "amnesiac", sizeof(lp1->d_typename));
99 	if (lp1->d_packname[0] == '\0')
100 		strncpy(lp1->d_packname, "fictitious", sizeof(lp1->d_packname));
101 	if (lp1->d_nsectors == 0)
102 		lp1->d_nsectors = 32;
103 	if (lp1->d_ntracks == 0)
104 		lp1->d_ntracks = 64;
105 	lp1->d_secpercyl = lp1->d_nsectors * lp1->d_ntracks;
106 	lp1->d_ncylinders = lp1->d_secperunit / lp1->d_secpercyl;
107 	if (lp1->d_rpm == 0)
108 		lp1->d_rpm = 3600;
109 	if (lp1->d_interleave == 0)
110 		lp1->d_interleave = 1;
111 	if (lp1->d_npartitions < RAW_PART + 1)
112 		lp1->d_npartitions = MAXPARTITIONS;
113 	if (lp1->d_bbsize == 0)
114 		lp1->d_bbsize = BBSIZE;
115 	if (lp1->d_sbsize == 0)
116 		lp1->d_sbsize = SBSIZE;
117 	lp1->d_partitions[RAW_PART].p_size = lp1->d_secperunit;
118 	lp1->d_magic = DISKMAGIC;
119 	lp1->d_magic2 = DISKMAGIC;
120 	lp1->d_checksum = dkcksum(lp1);
121 	return (lp1);
122 }
123 
124 /*
125  * Determine the size of the transfer, and make sure it is
126  * within the boundaries of the partition. Adjust transfer
127  * if needed, and signal errors or early completion.
128  *
129  * XXX TODO:
130  *	o Split buffers that are too big for the device.
131  *	o Check for overflow.
132  *	o Finish cleaning this up.
133  */
134 int
135 dscheck(struct buf *bp, struct diskslices *ssp)
136 {
137 	daddr_t	blkno;
138 	u_long	endsecno;
139 	daddr_t	labelsect;
140 	struct disklabel *lp;
141 	char *msg;
142 	long nsec;
143 	struct partition *pp;
144 	daddr_t	secno;
145 	daddr_t	slicerel_secno;
146 	struct diskslice *sp;
147 	int s;
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 				s = splbio();
263 				bp->b_vp->v_numoutput++;
264 				splx(s);
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 #ifndef __alpha__
523 		if (slice == WHOLE_DISK_SLICE)
524 			return (ENODEV);
525 #endif
526 		if (!(flags & FWRITE))
527 			return (EBADF);
528 		set_ds_wlabel(ssp, slice, *(int *)data != 0);
529 		return (0);
530 
531 	default:
532 		return (ENOIOCTL);
533 	}
534 }
535 
536 static void
537 dsiodone(struct buf *bp)
538 {
539 	struct iodone_chain *ic;
540 	char *msg;
541 
542 	ic = bp->b_iodone_chain;
543 	bp->b_flags = (ic->ic_prev_flags & B_CALL)
544 		      | (bp->b_flags & ~(B_CALL | B_DONE));
545 	bp->b_iodone = ic->ic_prev_iodone;
546 	bp->b_iodone_chain = ic->ic_prev_iodone_chain;
547 	if (!(bp->b_flags & B_READ)
548 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
549 		msg = fixlabel((char *)NULL, ic->ic_args[1].ia_ptr,
550 			       (struct disklabel *)
551 			       (bp->b_data + ic->ic_args[0].ia_long),
552 			       FALSE);
553 		if (msg != NULL)
554 			printf("%s\n", msg);
555 	}
556 	free(ic, M_DEVBUF);
557 	biodone(bp);
558 }
559 
560 int
561 dsisopen(struct diskslices *ssp)
562 {
563 	int slice;
564 
565 	if (ssp == NULL)
566 		return (0);
567 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
568 		if (ssp->dss_slices[slice].ds_openmask)
569 			return (1);
570 	}
571 	return (0);
572 }
573 
574 /*
575  * Allocate a slices "struct" and initialize it to contain only an empty
576  * compatibility slice (pointing to itself), a whole disk slice (covering
577  * the disk as described by the label), and (nslices - BASE_SLICES) empty
578  * slices beginning at BASE_SLICE.
579  */
580 struct diskslices *
581 dsmakeslicestruct(int nslices, struct disklabel *lp)
582 {
583 	struct diskslice *sp;
584 	struct diskslices *ssp;
585 
586 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
587 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
588 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
589 	ssp->dss_nslices = nslices;
590 	ssp->dss_oflags = 0;
591 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
592 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
593 		ssp->dss_secshift = -1;
594 	else
595 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
596 	ssp->dss_secsize = lp->d_secsize;
597 	sp = &ssp->dss_slices[0];
598 	bzero(sp, nslices * sizeof *sp);
599 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
600 	return (ssp);
601 }
602 
603 char *
604 dsname(dev_t dev, int unit, int slice, int part, char *partname)
605 {
606 	static char name[32];
607 	const char *dname;
608 
609 	dname = dev_dname(dev);
610 	if (strlen(dname) > 16)
611 		dname = "nametoolong";
612 	snprintf(name, sizeof(name), "%s%d", dname, unit);
613 	partname[0] = '\0';
614 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
615 		partname[0] = 'a' + part;
616 		partname[1] = '\0';
617 		if (slice != COMPATIBILITY_SLICE) {
618 			snprintf(name + strlen(name),
619 			    sizeof(name) - strlen(name), "s%d", slice - 1);
620 		}
621 	}
622 	return (name);
623 }
624 
625 /*
626  * This should only be called when the unit is inactive and the strategy
627  * routine should not allow it to become active unless we call it.  Our
628  * strategy routine must be special to allow activity.
629  */
630 int
631 dsopen(dev_t dev, int mode, u_int flags,
632 	struct diskslices **sspp, struct disklabel *lp)
633 {
634 	dev_t dev1;
635 	int error;
636 	struct disklabel *lp1;
637 	char *msg;
638 	u_char mask;
639 	bool_t need_init;
640 	int part;
641 	char partname[2];
642 	int slice;
643 	char *sname;
644 	struct diskslice *sp;
645 	struct diskslices *ssp;
646 	int unit;
647 
648 	dev->si_bsize_phys = lp->d_secsize;
649 
650 	unit = dkunit(dev);
651 	if (lp->d_secsize % DEV_BSIZE) {
652 		printf("%s: invalid sector size %lu\n", devtoname(dev),
653 		    (u_long)lp->d_secsize);
654 		return (EINVAL);
655 	}
656 
657 	/*
658 	 * XXX reinitialize the slice table unless there is an open device
659 	 * on the unit.  This should only be done if the media has changed.
660 	 */
661 	ssp = *sspp;
662 	need_init = !dsisopen(ssp);
663 	if (ssp != NULL && need_init)
664 		dsgone(sspp);
665 	if (need_init) {
666 		/*
667 		 * Allocate a minimal slices "struct".  This will become
668 		 * the final slices "struct" if we don't want real slices
669 		 * or if we can't find any real slices.
670 		 */
671 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
672 
673 		if (!(flags & DSO_ONESLICE)) {
674 			TRACE(("dsinit\n"));
675 			error = dsinit(dev, lp, sspp);
676 			if (error != 0) {
677 				dsgone(sspp);
678 				return (error);
679 			}
680 		}
681 		ssp = *sspp;
682 		ssp->dss_oflags = flags;
683 
684 		/*
685 		 * If there are no real slices, then make the compatiblity
686 		 * slice cover the whole disk.
687 		 */
688 		if (ssp->dss_nslices == BASE_SLICE)
689 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
690 				= lp->d_secperunit;
691 
692 		/* Point the compatibility slice at the BSD slice, if any. */
693 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
694 			sp = &ssp->dss_slices[slice];
695 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
696 				ssp->dss_first_bsd_slice = slice;
697 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
698 					= sp->ds_offset;
699 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
700 					= sp->ds_size;
701 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
702 					= sp->ds_type;
703 				break;
704 			}
705 		}
706 
707 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
708 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
709 	}
710 
711 	/*
712 	 * Initialize secondary info for all slices.  It is needed for more
713 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
714 	 */
715 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
716 		sp = &ssp->dss_slices[slice];
717 		if (sp->ds_label != NULL
718 #ifdef __alpha__
719 		    && slice != WHOLE_DISK_SLICE
720 #endif
721 		    )
722 			continue;
723 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
724 		sname = dsname(dev, unit, slice, RAW_PART, partname);
725 		/*
726 		 * XXX this should probably only be done for the need_init
727 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
728 		 */
729 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
730 		lp1 = clone_label(lp);
731 		TRACE(("readdisklabel\n"));
732 		if (flags & DSO_NOLABELS)
733 			msg = NULL;
734 		else {
735 			msg = readdisklabel(dev1, lp1);
736 
737 			/*
738 			 * readdisklabel() returns NULL for success, and an
739 			 * error string for failure.
740 			 *
741 			 * If there isn't a label on the disk, and if the
742 			 * DSO_COMPATLABEL is set, we want to use the
743 			 * faked-up label provided by the caller.
744 			 *
745 			 * So we set msg to NULL to indicate that there is
746 			 * no failure (since we have a faked-up label),
747 			 * free lp1, and then clone it again from lp.
748 			 * (In case readdisklabel() modified lp1.)
749 			 */
750 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
751 				msg = NULL;
752 				free(lp1, M_DEVBUF);
753 				lp1 = clone_label(lp);
754 			}
755 		}
756 		if (msg == NULL)
757 			msg = fixlabel(sname, sp, lp1, FALSE);
758 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
759 			msg = "inconsistent sector size";
760 		if (msg != NULL) {
761 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
762 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
763 				    sname, msg);
764 			free(lp1, M_DEVBUF);
765 			continue;
766 		}
767 		if (lp1->d_flags & D_BADSECT) {
768 			log(LOG_ERR, "%s: bad sector table not supported\n",
769 			    sname);
770 			free(lp1, M_DEVBUF);
771 			continue;
772 		}
773 		set_ds_label(ssp, slice, lp1);
774 		set_ds_wlabel(ssp, slice, FALSE);
775 	}
776 
777 	slice = dkslice(dev);
778 	if (slice >= ssp->dss_nslices)
779 		return (ENXIO);
780 	sp = &ssp->dss_slices[slice];
781 	part = dkpart(dev);
782 	if (part != RAW_PART
783 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
784 		return (EINVAL);	/* XXX needs translation */
785 	mask = 1 << part;
786 	sp->ds_openmask |= mask;
787 	return (0);
788 }
789 
790 int
791 dssize(dev_t dev, struct diskslices **sspp)
792 {
793 	struct disklabel *lp;
794 	int part;
795 	int slice;
796 	struct diskslices *ssp;
797 
798 	slice = dkslice(dev);
799 	part = dkpart(dev);
800 	ssp = *sspp;
801 	if (ssp == NULL || slice >= ssp->dss_nslices
802 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
803 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
804 			return (-1);
805 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
806 		ssp = *sspp;
807 	}
808 	lp = ssp->dss_slices[slice].ds_label;
809 	if (lp == NULL)
810 		return (-1);
811 	return ((int)lp->d_partitions[part].p_size);
812 }
813 
814 static void
815 free_ds_label(struct diskslices *ssp, int slice)
816 {
817 	struct disklabel *lp;
818 	struct diskslice *sp;
819 
820 	sp = &ssp->dss_slices[slice];
821 	lp = sp->ds_label;
822 	if (lp == NULL)
823 		return;
824 	free(lp, M_DEVBUF);
825 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
826 }
827 
828 static char *
829 fixlabel(char *sname, struct diskslice *sp, struct disklabel *lp, int writeflag)
830 {
831 	u_long end;
832 	u_long offset;
833 	int part;
834 	struct partition *pp;
835 	u_long start;
836 	bool_t warned;
837 
838 	/* These errors "can't happen" so don't bother reporting details. */
839 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
840 		return ("fixlabel: invalid magic");
841 	if (dkcksum(lp) != 0)
842 		return ("fixlabel: invalid checksum");
843 
844 	pp = &lp->d_partitions[RAW_PART];
845 	if (writeflag) {
846 		start = 0;
847 		offset = sp->ds_offset;
848 	} else {
849 		start = sp->ds_offset;
850 		offset = -sp->ds_offset;
851 	}
852 	if (pp->p_offset != start) {
853 		if (sname != NULL) {
854 			printf(
855 "%s: rejecting BSD label: raw partition offset != slice offset\n",
856 			       sname);
857 			slice_info(sname, sp);
858 			partition_info(sname, RAW_PART, pp);
859 		}
860 		return ("fixlabel: raw partition offset != slice offset");
861 	}
862 	if (pp->p_size != sp->ds_size) {
863 		if (sname != NULL) {
864 			printf("%s: raw partition size != slice size\n", sname);
865 			slice_info(sname, sp);
866 			partition_info(sname, RAW_PART, pp);
867 		}
868 		if (pp->p_size > sp->ds_size) {
869 			if (sname == NULL)
870 				return ("fixlabel: raw partition size > slice size");
871 			printf("%s: truncating raw partition\n", sname);
872 			pp->p_size = sp->ds_size;
873 		}
874 	}
875 	end = start + sp->ds_size;
876 	if (start > end)
877 		return ("fixlabel: slice wraps");
878 	if (lp->d_secpercyl <= 0)
879 		return ("fixlabel: d_secpercyl <= 0");
880 	pp -= RAW_PART;
881 	warned = FALSE;
882 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
883 		if (pp->p_offset != 0 || pp->p_size != 0) {
884 			if (pp->p_offset < start
885 			    || pp->p_offset + pp->p_size > end
886 			    || pp->p_offset + pp->p_size < pp->p_offset) {
887 				if (sname != NULL) {
888 					printf(
889 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
890 					       sname);
891 					if (!warned) {
892 						slice_info(sname, sp);
893 						warned = TRUE;
894 					}
895 					partition_info(sname, part, pp);
896 				}
897 				/* XXX else silently discard junk. */
898 				bzero(pp, sizeof *pp);
899 			} else
900 				pp->p_offset += offset;
901 		}
902 	}
903 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
904 	lp->d_secperunit = sp->ds_size;
905  	lp->d_checksum = 0;
906  	lp->d_checksum = dkcksum(lp);
907 	return (NULL);
908 }
909 
910 static void
911 partition_info(char *sname, int part, struct partition *pp)
912 {
913 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
914 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
915 	       (u_long)pp->p_size);
916 }
917 
918 static void
919 slice_info(char *sname, struct diskslice *sp)
920 {
921 	printf("%s: start %lu, end %lu, size %lu\n", sname,
922 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
923 }
924 
925 static void
926 set_ds_label(struct diskslices *ssp, int slice, struct disklabel *lp)
927 {
928 	ssp->dss_slices[slice].ds_label = lp;
929 	if (slice == COMPATIBILITY_SLICE)
930 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
931 	else if (slice == ssp->dss_first_bsd_slice)
932 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
933 }
934 
935 static void
936 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
937 {
938 	ssp->dss_slices[slice].ds_wlabel = wlabel;
939 	if (slice == COMPATIBILITY_SLICE)
940 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
941 	else if (slice == ssp->dss_first_bsd_slice)
942 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
943 }
944