xref: /dragonfly/sys/kern/subr_diskslice.c (revision 4e7eb5cc)
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 1.7 2003/11/10 06:12:13 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_labeldevs (dev_t dev, struct diskslices *ssp);
82 static void set_ds_wlabel (struct diskslices *ssp, int slice,
83 			       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(lp)
93 	struct disklabel *lp;
94 {
95 	struct disklabel *lp1;
96 
97 	lp1 = malloc(sizeof *lp1, M_DEVBUF, M_WAITOK);
98 	*lp1 = *lp;
99 	lp = NULL;
100 	if (lp1->d_typename[0] == '\0')
101 		strncpy(lp1->d_typename, "amnesiac", sizeof(lp1->d_typename));
102 	if (lp1->d_packname[0] == '\0')
103 		strncpy(lp1->d_packname, "fictitious", sizeof(lp1->d_packname));
104 	if (lp1->d_nsectors == 0)
105 		lp1->d_nsectors = 32;
106 	if (lp1->d_ntracks == 0)
107 		lp1->d_ntracks = 64;
108 	lp1->d_secpercyl = lp1->d_nsectors * lp1->d_ntracks;
109 	lp1->d_ncylinders = lp1->d_secperunit / lp1->d_secpercyl;
110 	if (lp1->d_rpm == 0)
111 		lp1->d_rpm = 3600;
112 	if (lp1->d_interleave == 0)
113 		lp1->d_interleave = 1;
114 	if (lp1->d_npartitions < RAW_PART + 1)
115 		lp1->d_npartitions = MAXPARTITIONS;
116 	if (lp1->d_bbsize == 0)
117 		lp1->d_bbsize = BBSIZE;
118 	if (lp1->d_sbsize == 0)
119 		lp1->d_sbsize = SBSIZE;
120 	lp1->d_partitions[RAW_PART].p_size = lp1->d_secperunit;
121 	lp1->d_magic = DISKMAGIC;
122 	lp1->d_magic2 = DISKMAGIC;
123 	lp1->d_checksum = dkcksum(lp1);
124 	return (lp1);
125 }
126 
127 /*
128  * Determine the size of the transfer, and make sure it is
129  * within the boundaries of the partition. Adjust transfer
130  * if needed, and signal errors or early completion.
131  *
132  * XXX TODO:
133  *	o Split buffers that are too big for the device.
134  *	o Check for overflow.
135  *	o Finish cleaning this up.
136  */
137 int
138 dscheck(bp, ssp)
139 	struct buf *bp;
140 	struct diskslices *ssp;
141 {
142 	daddr_t	blkno;
143 	u_long	endsecno;
144 	daddr_t	labelsect;
145 	struct disklabel *lp;
146 	char *msg;
147 	long	nsec;
148 	struct partition *pp;
149 	daddr_t	secno;
150 	daddr_t	slicerel_secno;
151 	struct diskslice *sp;
152 	int s;
153 
154 	blkno = bp->b_blkno;
155 	if (blkno < 0) {
156 		printf("dscheck(%s): negative b_blkno %ld\n",
157 		    devtoname(bp->b_dev), (long)blkno);
158 		bp->b_error = EINVAL;
159 		goto bad;
160 	}
161 	sp = &ssp->dss_slices[dkslice(bp->b_dev)];
162 	lp = sp->ds_label;
163 	if (ssp->dss_secmult == 1) {
164 		if (bp->b_bcount % (u_long)DEV_BSIZE)
165 			goto bad_bcount;
166 		secno = blkno;
167 		nsec = bp->b_bcount >> DEV_BSHIFT;
168 	} else if (ssp->dss_secshift != -1) {
169 		if (bp->b_bcount & (ssp->dss_secsize - 1))
170 			goto bad_bcount;
171 		if (blkno & (ssp->dss_secmult - 1))
172 			goto bad_blkno;
173 		secno = blkno >> ssp->dss_secshift;
174 		nsec = bp->b_bcount >> (DEV_BSHIFT + ssp->dss_secshift);
175 	} else {
176 		if (bp->b_bcount % ssp->dss_secsize)
177 			goto bad_bcount;
178 		if (blkno % ssp->dss_secmult)
179 			goto bad_blkno;
180 		secno = blkno / ssp->dss_secmult;
181 		nsec = bp->b_bcount / ssp->dss_secsize;
182 	}
183 	if (lp == NULL) {
184 		labelsect = -LABELSECTOR - 1;
185 		endsecno = sp->ds_size;
186 		slicerel_secno = secno;
187 	} else {
188 		labelsect = lp->d_partitions[LABEL_PART].p_offset;
189 if (labelsect != 0) Debugger("labelsect != 0 in dscheck()");
190 		pp = &lp->d_partitions[dkpart(bp->b_dev)];
191 		endsecno = pp->p_size;
192 		slicerel_secno = pp->p_offset + secno;
193 	}
194 
195 	/* overwriting disk label ? */
196 	/* XXX should also protect bootstrap in first 8K */
197 	if (slicerel_secno <= LABELSECTOR + labelsect &&
198 #if LABELSECTOR != 0
199 	    slicerel_secno + nsec > LABELSECTOR + labelsect &&
200 #endif
201 	    (bp->b_flags & B_READ) == 0 && sp->ds_wlabel == 0) {
202 		bp->b_error = EROFS;
203 		goto bad;
204 	}
205 
206 #if defined(DOSBBSECTOR) && defined(notyet)
207 	/* overwriting master boot record? */
208 	if (slicerel_secno <= DOSBBSECTOR && (bp->b_flags & B_READ) == 0 &&
209 	    sp->ds_wlabel == 0) {
210 		bp->b_error = EROFS;
211 		goto bad;
212 	}
213 #endif
214 
215 	/* beyond partition? */
216 	if (secno + nsec > endsecno) {
217 		/* if exactly at end of disk, return an EOF */
218 		if (secno == endsecno) {
219 			bp->b_resid = bp->b_bcount;
220 			return (0);
221 		}
222 		/* or truncate if part of it fits */
223 		nsec = endsecno - secno;
224 		if (nsec <= 0) {
225 			bp->b_error = EINVAL;
226 			goto bad;
227 		}
228 		bp->b_bcount = nsec * ssp->dss_secsize;
229 	}
230 
231 	bp->b_pblkno = sp->ds_offset + slicerel_secno;
232 
233 	/*
234 	 * Snoop on label accesses if the slice offset is nonzero.  Fudge
235 	 * offsets in the label to keep the in-core label coherent with
236 	 * the on-disk one.
237 	 */
238 	if (slicerel_secno <= LABELSECTOR + labelsect
239 #if LABELSECTOR != 0
240 	    && slicerel_secno + nsec > LABELSECTOR + labelsect
241 #endif
242 	    && sp->ds_offset != 0) {
243 		struct iodone_chain *ic;
244 
245 		ic = malloc(sizeof *ic , M_DEVBUF, M_WAITOK);
246 		ic->ic_prev_flags = bp->b_flags;
247 		ic->ic_prev_iodone = bp->b_iodone;
248 		ic->ic_prev_iodone_chain = bp->b_iodone_chain;
249 		ic->ic_args[0].ia_long = (LABELSECTOR + labelsect -
250 		    slicerel_secno) * ssp->dss_secsize;
251 		ic->ic_args[1].ia_ptr = sp;
252 		bp->b_flags |= B_CALL;
253 		bp->b_iodone = dsiodone;
254 		bp->b_iodone_chain = ic;
255 		if (!(bp->b_flags & B_READ)) {
256 			/*
257 			 * XXX even disklabel(8) writes directly so we need
258 			 * to adjust writes.  Perhaps we should drop support
259 			 * for DIOCWLABEL (always write protect labels) and
260 			 * require the use of DIOCWDINFO.
261 			 *
262 			 * XXX probably need to copy the data to avoid even
263 			 * temporarily corrupting the in-core copy.
264 			 */
265 			if (bp->b_vp != NULL) {
266 				s = splbio();
267 				bp->b_vp->v_numoutput++;
268 				splx(s);
269 			}
270 			/* XXX need name here. */
271 			msg = fixlabel((char *)NULL, sp,
272 				       (struct disklabel *)
273 				       (bp->b_data + ic->ic_args[0].ia_long),
274 				       TRUE);
275 			if (msg != NULL) {
276 				printf("dscheck(%s): %s\n",
277 				    devtoname(bp->b_dev), msg);
278 				bp->b_error = EROFS;
279 				goto bad;
280 			}
281 		}
282 	}
283 	return (1);
284 
285 bad_bcount:
286 	printf(
287 	"dscheck(%s): b_bcount %ld is not on a sector boundary (ssize %d)\n",
288 	    devtoname(bp->b_dev), bp->b_bcount, ssp->dss_secsize);
289 	bp->b_error = EINVAL;
290 	goto bad;
291 
292 bad_blkno:
293 	printf(
294 	"dscheck(%s): b_blkno %ld is not on a sector boundary (ssize %d)\n",
295 	    devtoname(bp->b_dev), (long)blkno, ssp->dss_secsize);
296 	bp->b_error = EINVAL;
297 	goto bad;
298 
299 bad:
300 	bp->b_resid = bp->b_bcount;
301 	bp->b_flags |= B_ERROR;
302 	return (-1);
303 }
304 
305 void
306 dsclose(dev, mode, ssp)
307 	dev_t	dev;
308 	int	mode;
309 	struct diskslices *ssp;
310 {
311 	u_char	mask;
312 	struct diskslice *sp;
313 
314 	sp = &ssp->dss_slices[dkslice(dev)];
315 	mask = 1 << dkpart(dev);
316 	sp->ds_openmask &= ~mask;
317 }
318 
319 void
320 dsgone(sspp)
321 	struct diskslices **sspp;
322 {
323 	int	slice;
324 	struct diskslice *sp;
325 	struct diskslices *ssp;
326 
327 	for (slice = 0, ssp = *sspp; slice < ssp->dss_nslices; slice++) {
328 		sp = &ssp->dss_slices[slice];
329 		free_ds_label(ssp, slice);
330 	}
331 	free(ssp, M_DEVBUF);
332 	*sspp = NULL;
333 }
334 
335 /*
336  * For the "write" commands (DIOCSDINFO and DIOCWDINFO), this
337  * is subject to the same restriction as dsopen().
338  */
339 int
340 dsioctl(dev, cmd, data, flags, sspp)
341 	dev_t	dev;
342 	u_long	cmd;
343 	caddr_t	data;
344 	int	flags;
345 	struct diskslices **sspp;
346 {
347 	int	error;
348 	struct disklabel *lp;
349 	int	old_wlabel;
350 	u_char	openmask;
351 	int	part;
352 	int	slice;
353 	struct diskslice *sp;
354 	struct diskslices *ssp;
355 	struct partition *pp;
356 
357 	slice = dkslice(dev);
358 	ssp = *sspp;
359 	sp = &ssp->dss_slices[slice];
360 	lp = sp->ds_label;
361 	switch (cmd) {
362 
363 	case DIOCGDVIRGIN:
364 		lp = (struct disklabel *)data;
365 		if (ssp->dss_slices[WHOLE_DISK_SLICE].ds_label) {
366 			*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
367 		} else {
368 			bzero(lp, sizeof(struct disklabel));
369 		}
370 
371 		lp->d_magic = DISKMAGIC;
372 		lp->d_magic2 = DISKMAGIC;
373 		pp = &lp->d_partitions[RAW_PART];
374 		pp->p_offset = 0;
375 		pp->p_size = sp->ds_size;
376 
377 		lp->d_npartitions = MAXPARTITIONS;
378 		if (lp->d_interleave == 0)
379 			lp->d_interleave = 1;
380 		if (lp->d_rpm == 0)
381 			lp->d_rpm = 3600;
382 		if (lp->d_nsectors == 0)
383 			lp->d_nsectors = 32;
384 		if (lp->d_ntracks == 0)
385 			lp->d_ntracks = 64;
386 
387 		lp->d_bbsize = BBSIZE;
388 		lp->d_sbsize = SBSIZE;
389 		lp->d_secpercyl = lp->d_nsectors * lp->d_ntracks;
390 		lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
391 		lp->d_secperunit = sp->ds_size;
392 		lp->d_checksum = 0;
393 		lp->d_checksum = dkcksum(lp);
394 		return (0);
395 
396 	case DIOCGDINFO:
397 		if (lp == NULL)
398 			return (EINVAL);
399 		*(struct disklabel *)data = *lp;
400 		return (0);
401 
402 #ifdef notyet
403 	case DIOCGDINFOP:
404 		if (lp == NULL)
405 			return (EINVAL);
406 		*(struct disklabel **)data = lp;
407 		return (0);
408 #endif
409 
410 	case DIOCGPART:
411 		if (lp == NULL)
412 			return (EINVAL);
413 		((struct partinfo *)data)->disklab = lp;
414 		((struct partinfo *)data)->part
415 			= &lp->d_partitions[dkpart(dev)];
416 		return (0);
417 
418 	case DIOCGSLICEINFO:
419 		bcopy(ssp, data, (char *)&ssp->dss_slices[ssp->dss_nslices] -
420 				 (char *)ssp);
421 		return (0);
422 
423 	case DIOCSDINFO:
424 		if (slice == WHOLE_DISK_SLICE)
425 			return (ENODEV);
426 		if (!(flags & FWRITE))
427 			return (EBADF);
428 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
429 		if (sp->ds_label == NULL)
430 			bzero(lp, sizeof *lp);
431 		else
432 			bcopy(sp->ds_label, lp, sizeof *lp);
433 		if (sp->ds_label == NULL)
434 			openmask = 0;
435 		else {
436 			openmask = sp->ds_openmask;
437 			if (slice == COMPATIBILITY_SLICE)
438 				openmask |= ssp->dss_slices[
439 				    ssp->dss_first_bsd_slice].ds_openmask;
440 			else if (slice == ssp->dss_first_bsd_slice)
441 				openmask |= ssp->dss_slices[
442 				    COMPATIBILITY_SLICE].ds_openmask;
443 		}
444 		error = setdisklabel(lp, (struct disklabel *)data,
445 				     (u_long)openmask);
446 		/* XXX why doesn't setdisklabel() check this? */
447 		if (error == 0 && lp->d_partitions[RAW_PART].p_offset != 0)
448 			error = EXDEV;
449 		if (error == 0) {
450 			if (lp->d_secperunit > sp->ds_size)
451 				error = ENOSPC;
452 			for (part = 0; part < lp->d_npartitions; part++)
453 				if (lp->d_partitions[part].p_size > sp->ds_size)
454 					error = ENOSPC;
455 		}
456 		if (error != 0) {
457 			free(lp, M_DEVBUF);
458 			return (error);
459 		}
460 		free_ds_label(ssp, slice);
461 		set_ds_label(ssp, slice, lp);
462 		set_ds_labeldevs(dev, ssp);
463 		return (0);
464 
465 	case DIOCSYNCSLICEINFO:
466 		if (slice != WHOLE_DISK_SLICE || dkpart(dev) != RAW_PART)
467 			return (EINVAL);
468 		if (!*(int *)data)
469 			for (slice = 0; slice < ssp->dss_nslices; slice++) {
470 				openmask = ssp->dss_slices[slice].ds_openmask;
471 				if (openmask
472 				    && (slice != WHOLE_DISK_SLICE
473 					|| openmask & ~(1 << RAW_PART)))
474 					return (EBUSY);
475 			}
476 
477 		/*
478 		 * Temporarily forget the current slices struct and read
479 		 * the current one.
480 		 * XXX should wait for current accesses on this disk to
481 		 * complete, then lock out future accesses and opens.
482 		 */
483 		*sspp = NULL;
484 		lp = malloc(sizeof *lp, M_DEVBUF, M_WAITOK);
485 		*lp = *ssp->dss_slices[WHOLE_DISK_SLICE].ds_label;
486 		error = dsopen(dev, S_IFCHR, ssp->dss_oflags, sspp, lp);
487 		if (error != 0) {
488 			free(lp, M_DEVBUF);
489 			*sspp = ssp;
490 			return (error);
491 		}
492 
493 		/*
494 		 * Reopen everything.  This is a no-op except in the "force"
495 		 * case and when the raw bdev and cdev are both open.  Abort
496 		 * if anything fails.
497 		 */
498 		for (slice = 0; slice < ssp->dss_nslices; slice++) {
499 			for (openmask = ssp->dss_slices[slice].ds_openmask,
500 			     part = 0; openmask; openmask >>= 1, part++) {
501 				if (!(openmask & 1))
502 					continue;
503 				error = dsopen(dkmodslice(dkmodpart(dev, part),
504 							  slice),
505 					       S_IFCHR, ssp->dss_oflags, sspp,
506 					       lp);
507 				if (error != 0) {
508 					free(lp, M_DEVBUF);
509 					*sspp = ssp;
510 					return (EBUSY);
511 				}
512 			}
513 		}
514 
515 		free(lp, M_DEVBUF);
516 		dsgone(&ssp);
517 		return (0);
518 
519 	case DIOCWDINFO:
520 		error = dsioctl(dev, DIOCSDINFO, data, flags, &ssp);
521 		if (error != 0)
522 			return (error);
523 		/*
524 		 * XXX this used to hack on dk_openpart to fake opening
525 		 * partition 0 in case that is used instead of dkpart(dev).
526 		 */
527 		old_wlabel = sp->ds_wlabel;
528 		set_ds_wlabel(ssp, slice, TRUE);
529 		error = writedisklabel(dev, sp->ds_label);
530 		/* XXX should invalidate in-core label if write failed. */
531 		set_ds_wlabel(ssp, slice, old_wlabel);
532 		return (error);
533 
534 	case DIOCWLABEL:
535 #ifndef __alpha__
536 		if (slice == WHOLE_DISK_SLICE)
537 			return (ENODEV);
538 #endif
539 		if (!(flags & FWRITE))
540 			return (EBADF);
541 		set_ds_wlabel(ssp, slice, *(int *)data != 0);
542 		return (0);
543 
544 	default:
545 		return (ENOIOCTL);
546 	}
547 }
548 
549 static void
550 dsiodone(bp)
551 	struct buf *bp;
552 {
553 	struct iodone_chain *ic;
554 	char *msg;
555 
556 	ic = bp->b_iodone_chain;
557 	bp->b_flags = (ic->ic_prev_flags & B_CALL)
558 		      | (bp->b_flags & ~(B_CALL | B_DONE));
559 	bp->b_iodone = ic->ic_prev_iodone;
560 	bp->b_iodone_chain = ic->ic_prev_iodone_chain;
561 	if (!(bp->b_flags & B_READ)
562 	    || (!(bp->b_flags & B_ERROR) && bp->b_error == 0)) {
563 		msg = fixlabel((char *)NULL, ic->ic_args[1].ia_ptr,
564 			       (struct disklabel *)
565 			       (bp->b_data + ic->ic_args[0].ia_long),
566 			       FALSE);
567 		if (msg != NULL)
568 			printf("%s\n", msg);
569 	}
570 	free(ic, M_DEVBUF);
571 	biodone(bp);
572 }
573 
574 int
575 dsisopen(ssp)
576 	struct diskslices *ssp;
577 {
578 	int	slice;
579 
580 	if (ssp == NULL)
581 		return (0);
582 	for (slice = 0; slice < ssp->dss_nslices; slice++)
583 		if (ssp->dss_slices[slice].ds_openmask)
584 			return (1);
585 	return (0);
586 }
587 
588 /*
589  * Allocate a slices "struct" and initialize it to contain only an empty
590  * compatibility slice (pointing to itself), a whole disk slice (covering
591  * the disk as described by the label), and (nslices - BASE_SLICES) empty
592  * slices beginning at BASE_SLICE.
593  */
594 struct diskslices *
595 dsmakeslicestruct(nslices, lp)
596 	int nslices;
597 	struct disklabel *lp;
598 {
599 	struct diskslice *sp;
600 	struct diskslices *ssp;
601 
602 	ssp = malloc(offsetof(struct diskslices, dss_slices) +
603 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
604 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
605 	ssp->dss_nslices = nslices;
606 	ssp->dss_oflags = 0;
607 	ssp->dss_secmult = lp->d_secsize / DEV_BSIZE;
608 	if (ssp->dss_secmult & (ssp->dss_secmult - 1))
609 		ssp->dss_secshift = -1;
610 	else
611 		ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
612 	ssp->dss_secsize = lp->d_secsize;
613 	sp = &ssp->dss_slices[0];
614 	bzero(sp, nslices * sizeof *sp);
615 	sp[WHOLE_DISK_SLICE].ds_size = lp->d_secperunit;
616 	return (ssp);
617 }
618 
619 char *
620 dsname(dev, unit, slice, part, partname)
621 	dev_t	dev;
622 	int	unit;
623 	int	slice;
624 	int	part;
625 	char	*partname;
626 {
627 	static char name[32];
628 	const char *dname;
629 
630 	dname = dev_dname(dev);
631 	if (strlen(dname) > 16)
632 		dname = "nametoolong";
633 	snprintf(name, sizeof(name), "%s%d", dname, unit);
634 	partname[0] = '\0';
635 	if (slice != WHOLE_DISK_SLICE || part != RAW_PART) {
636 		partname[0] = 'a' + part;
637 		partname[1] = '\0';
638 		if (slice != COMPATIBILITY_SLICE)
639 			snprintf(name + strlen(name),
640 			    sizeof(name) - strlen(name), "s%d", slice - 1);
641 	}
642 	return (name);
643 }
644 
645 /*
646  * This should only be called when the unit is inactive and the strategy
647  * routine should not allow it to become active unless we call it.  Our
648  * strategy routine must be special to allow activity.
649  */
650 int
651 dsopen(dev, mode, flags, sspp, lp)
652 	dev_t	dev;
653 	int	mode;
654 	u_int	flags;
655 	struct diskslices **sspp;
656 	struct disklabel *lp;
657 {
658 	dev_t	dev1;
659 	int	error;
660 	struct disklabel *lp1;
661 	char	*msg;
662 	u_char	mask;
663 	bool_t	need_init;
664 	int	part;
665 	char	partname[2];
666 	int	slice;
667 	char	*sname;
668 	struct diskslice *sp;
669 	struct diskslices *ssp;
670 	int	unit;
671 
672 	dev->si_bsize_phys = lp->d_secsize;
673 
674 	unit = dkunit(dev);
675 	if (lp->d_secsize % DEV_BSIZE) {
676 		printf("%s: invalid sector size %lu\n", devtoname(dev),
677 		    (u_long)lp->d_secsize);
678 		return (EINVAL);
679 	}
680 
681 	/*
682 	 * XXX reinitialize the slice table unless there is an open device
683 	 * on the unit.  This should only be done if the media has changed.
684 	 */
685 	ssp = *sspp;
686 	need_init = !dsisopen(ssp);
687 	if (ssp != NULL && need_init)
688 		dsgone(sspp);
689 	if (need_init) {
690 		/*
691 		 * Allocate a minimal slices "struct".  This will become
692 		 * the final slices "struct" if we don't want real slices
693 		 * or if we can't find any real slices.
694 		 */
695 		*sspp = dsmakeslicestruct(BASE_SLICE, lp);
696 
697 		if (!(flags & DSO_ONESLICE)) {
698 			TRACE(("dsinit\n"));
699 			error = dsinit(dev, lp, sspp);
700 			if (error != 0) {
701 				dsgone(sspp);
702 				return (error);
703 			}
704 		}
705 		ssp = *sspp;
706 		ssp->dss_oflags = flags;
707 
708 		/*
709 		 * If there are no real slices, then make the compatiblity
710 		 * slice cover the whole disk.
711 		 */
712 		if (ssp->dss_nslices == BASE_SLICE)
713 			ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
714 				= lp->d_secperunit;
715 
716 		/* Point the compatibility slice at the BSD slice, if any. */
717 		for (slice = BASE_SLICE; slice < ssp->dss_nslices; slice++) {
718 			sp = &ssp->dss_slices[slice];
719 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
720 				ssp->dss_first_bsd_slice = slice;
721 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_offset
722 					= sp->ds_offset;
723 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_size
724 					= sp->ds_size;
725 				ssp->dss_slices[COMPATIBILITY_SLICE].ds_type
726 					= sp->ds_type;
727 				break;
728 			}
729 		}
730 
731 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_label = clone_label(lp);
732 		ssp->dss_slices[WHOLE_DISK_SLICE].ds_wlabel = TRUE;
733 	}
734 
735 	/*
736 	 * Initialize secondary info for all slices.  It is needed for more
737 	 * than the current slice in the DEVFS case.  XXX DEVFS is no more.
738 	 */
739 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
740 		sp = &ssp->dss_slices[slice];
741 		if (sp->ds_label != NULL
742 #ifdef __alpha__
743 		    && slice != WHOLE_DISK_SLICE
744 #endif
745 		    )
746 			continue;
747 		dev1 = dkmodslice(dkmodpart(dev, RAW_PART), slice);
748 		sname = dsname(dev, unit, slice, RAW_PART, partname);
749 		/*
750 		 * XXX this should probably only be done for the need_init
751 		 * case, but there may be a problem with DIOCSYNCSLICEINFO.
752 		 */
753 		set_ds_wlabel(ssp, slice, TRUE);	/* XXX invert */
754 		lp1 = clone_label(lp);
755 		TRACE(("readdisklabel\n"));
756 		if (flags & DSO_NOLABELS)
757 			msg = NULL;
758 		else {
759 			msg = readdisklabel(dev1, lp1);
760 
761 			/*
762 			 * readdisklabel() returns NULL for success, and an
763 			 * error string for failure.
764 			 *
765 			 * If there isn't a label on the disk, and if the
766 			 * DSO_COMPATLABEL is set, we want to use the
767 			 * faked-up label provided by the caller.
768 			 *
769 			 * So we set msg to NULL to indicate that there is
770 			 * no failure (since we have a faked-up label),
771 			 * free lp1, and then clone it again from lp.
772 			 * (In case readdisklabel() modified lp1.)
773 			 */
774 			if (msg != NULL && (flags & DSO_COMPATLABEL)) {
775 				msg = NULL;
776 				free(lp1, M_DEVBUF);
777 				lp1 = clone_label(lp);
778 			}
779 		}
780 		if (msg == NULL)
781 			msg = fixlabel(sname, sp, lp1, FALSE);
782 		if (msg == NULL && lp1->d_secsize != ssp->dss_secsize)
783 			msg = "inconsistent sector size";
784 		if (msg != NULL) {
785 			if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
786 				log(LOG_WARNING, "%s: cannot find label (%s)\n",
787 				    sname, msg);
788 			free(lp1, M_DEVBUF);
789 			continue;
790 		}
791 		if (lp1->d_flags & D_BADSECT) {
792 			log(LOG_ERR, "%s: bad sector table not supported\n",
793 			    sname);
794 			free(lp1, M_DEVBUF);
795 			continue;
796 		}
797 		set_ds_label(ssp, slice, lp1);
798 		set_ds_labeldevs(dev1, ssp);
799 		set_ds_wlabel(ssp, slice, FALSE);
800 	}
801 
802 	slice = dkslice(dev);
803 	if (slice >= ssp->dss_nslices)
804 		return (ENXIO);
805 	sp = &ssp->dss_slices[slice];
806 	part = dkpart(dev);
807 	if (part != RAW_PART
808 	    && (sp->ds_label == NULL || part >= sp->ds_label->d_npartitions))
809 		return (EINVAL);	/* XXX needs translation */
810 	mask = 1 << part;
811 	sp->ds_openmask |= mask;
812 	return (0);
813 }
814 
815 int
816 dssize(dev, sspp)
817 	dev_t	dev;
818 	struct diskslices **sspp;
819 {
820 	struct disklabel *lp;
821 	int	part;
822 	int	slice;
823 	struct diskslices *ssp;
824 
825 	slice = dkslice(dev);
826 	part = dkpart(dev);
827 	ssp = *sspp;
828 	if (ssp == NULL || slice >= ssp->dss_nslices
829 	    || !(ssp->dss_slices[slice].ds_openmask & (1 << part))) {
830 		if (dev_dopen(dev, FREAD, S_IFCHR, NULL) != 0)
831 			return (-1);
832 		dev_dclose(dev, FREAD, S_IFCHR, NULL);
833 		ssp = *sspp;
834 	}
835 	lp = ssp->dss_slices[slice].ds_label;
836 	if (lp == NULL)
837 		return (-1);
838 	return ((int)lp->d_partitions[part].p_size);
839 }
840 
841 static void
842 free_ds_label(ssp, slice)
843 	struct diskslices *ssp;
844 	int	slice;
845 {
846 	struct disklabel *lp;
847 	struct diskslice *sp;
848 
849 	sp = &ssp->dss_slices[slice];
850 	lp = sp->ds_label;
851 	if (lp == NULL)
852 		return;
853 	free(lp, M_DEVBUF);
854 	set_ds_label(ssp, slice, (struct disklabel *)NULL);
855 }
856 
857 static char *
858 fixlabel(sname, sp, lp, writeflag)
859 	char	*sname;
860 	struct diskslice *sp;
861 	struct disklabel *lp;
862 	int	writeflag;
863 {
864 	u_long	end;
865 	u_long	offset;
866 	int	part;
867 	struct partition *pp;
868 	u_long	start;
869 	bool_t	warned;
870 
871 	/* These errors "can't happen" so don't bother reporting details. */
872 	if (lp->d_magic != DISKMAGIC || lp->d_magic2 != DISKMAGIC)
873 		return ("fixlabel: invalid magic");
874 	if (dkcksum(lp) != 0)
875 		return ("fixlabel: invalid checksum");
876 
877 	pp = &lp->d_partitions[RAW_PART];
878 	if (writeflag) {
879 		start = 0;
880 		offset = sp->ds_offset;
881 	} else {
882 		start = sp->ds_offset;
883 		offset = -sp->ds_offset;
884 	}
885 	if (pp->p_offset != start) {
886 		if (sname != NULL) {
887 			printf(
888 "%s: rejecting BSD label: raw partition offset != slice offset\n",
889 			       sname);
890 			slice_info(sname, sp);
891 			partition_info(sname, RAW_PART, pp);
892 		}
893 		return ("fixlabel: raw partition offset != slice offset");
894 	}
895 	if (pp->p_size != sp->ds_size) {
896 		if (sname != NULL) {
897 			printf("%s: raw partition size != slice size\n", sname);
898 			slice_info(sname, sp);
899 			partition_info(sname, RAW_PART, pp);
900 		}
901 		if (pp->p_size > sp->ds_size) {
902 			if (sname == NULL)
903 				return ("fixlabel: raw partition size > slice size");
904 			printf("%s: truncating raw partition\n", sname);
905 			pp->p_size = sp->ds_size;
906 		}
907 	}
908 	end = start + sp->ds_size;
909 	if (start > end)
910 		return ("fixlabel: slice wraps");
911 	if (lp->d_secpercyl <= 0)
912 		return ("fixlabel: d_secpercyl <= 0");
913 	pp -= RAW_PART;
914 	warned = FALSE;
915 	for (part = 0; part < lp->d_npartitions; part++, pp++) {
916 		if (pp->p_offset != 0 || pp->p_size != 0) {
917 			if (pp->p_offset < start
918 			    || pp->p_offset + pp->p_size > end
919 			    || pp->p_offset + pp->p_size < pp->p_offset) {
920 				if (sname != NULL) {
921 					printf(
922 "%s: rejecting partition in BSD label: it isn't entirely within the slice\n",
923 					       sname);
924 					if (!warned) {
925 						slice_info(sname, sp);
926 						warned = TRUE;
927 					}
928 					partition_info(sname, part, pp);
929 				}
930 				/* XXX else silently discard junk. */
931 				bzero(pp, sizeof *pp);
932 			} else
933 				pp->p_offset += offset;
934 		}
935 	}
936 	lp->d_ncylinders = sp->ds_size / lp->d_secpercyl;
937 	lp->d_secperunit = sp->ds_size;
938  	lp->d_checksum = 0;
939  	lp->d_checksum = dkcksum(lp);
940 	return (NULL);
941 }
942 
943 static void
944 partition_info(sname, part, pp)
945 	char	*sname;
946 	int	part;
947 	struct partition *pp;
948 {
949 	printf("%s%c: start %lu, end %lu, size %lu\n", sname, 'a' + part,
950 	       (u_long)pp->p_offset, (u_long)(pp->p_offset + pp->p_size - 1),
951 	       (u_long)pp->p_size);
952 }
953 
954 static void
955 slice_info(sname, sp)
956 	char	*sname;
957 	struct diskslice *sp;
958 {
959 	printf("%s: start %lu, end %lu, size %lu\n", sname,
960 	       sp->ds_offset, sp->ds_offset + sp->ds_size - 1, sp->ds_size);
961 }
962 
963 static void
964 set_ds_label(ssp, slice, lp)
965 	struct diskslices *ssp;
966 	int	slice;
967 	struct disklabel *lp;
968 {
969 	ssp->dss_slices[slice].ds_label = lp;
970 	if (slice == COMPATIBILITY_SLICE)
971 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_label = lp;
972 	else if (slice == ssp->dss_first_bsd_slice)
973 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_label = lp;
974 }
975 
976 /* XXX remove this? */
977 static void
978 set_ds_labeldevs(dev, ssp)
979 	dev_t	dev;
980 	struct diskslices *ssp;
981 {
982 }
983 
984 static void
985 set_ds_wlabel(ssp, slice, wlabel)
986 	struct diskslices *ssp;
987 	int	slice;
988 	int	wlabel;
989 {
990 	ssp->dss_slices[slice].ds_wlabel = wlabel;
991 	if (slice == COMPATIBILITY_SLICE)
992 		ssp->dss_slices[ssp->dss_first_bsd_slice].ds_wlabel = wlabel;
993 	else if (slice == ssp->dss_first_bsd_slice)
994 		ssp->dss_slices[COMPATIBILITY_SLICE].ds_wlabel = wlabel;
995 }
996