xref: /dragonfly/sys/kern/subr_diskmbr.c (revision 0cfebe3d)
1 /*-
2  * Copyright (c) 1994 Bruce D. Evans.
3  * All rights reserved.
4  *
5  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)ufs_disksubr.c	7.16 (Berkeley) 5/4/91
37  *	from: ufs_disksubr.c,v 1.8 1994/06/07 01:21:39 phk Exp $
38  * $FreeBSD: src/sys/kern/subr_diskmbr.c,v 1.45 2000/01/28 10:22:07 bde Exp $
39  * $DragonFly: src/sys/kern/subr_diskmbr.c,v 1.26 2007/06/19 06:07:57 dillon Exp $
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/buf.h>
45 #include <sys/conf.h>
46 #include <sys/diskslice.h>
47 #define	DOSPTYP_EXTENDED	5
48 #define	DOSPTYP_EXTENDEDX	15
49 #define	DOSPTYP_ONTRACK		84
50 #include <sys/diskslice.h>
51 #include <sys/diskmbr.h>
52 #include <sys/disk.h>
53 #include <sys/malloc.h>
54 #include <sys/syslog.h>
55 #include <sys/device.h>
56 
57 #define TRACE(str)	do { if (dsi_debug) kprintf str; } while (0)
58 
59 static volatile u_char dsi_debug;
60 
61 /*
62  * This is what we have embedded in every boot1 for supporting the bogus
63  * "Dangerously Dedicated" mode. However, the old table is broken because
64  * it has an illegal geometry in it - it specifies 256 heads (heads = end
65  * head + 1) which causes nasty stuff when that wraps to zero in bios code.
66  * eg: divide by zero etc. This caused the dead-thinkpad problem, numerous
67  * SCSI bios crashes, EFI to crash, etc.
68  *
69  * We still have to recognize the old table though, even though we stopped
70  * inflicting it upon the world.
71  */
72 static struct dos_partition historical_bogus_partition_table[NDOSPART] = {
73 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
74 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
75 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
76 	{ 0x80, 0, 1, 0, DOSPTYP_386BSD, 255, 255, 255, 0, 50000, },
77 };
78 static struct dos_partition historical_bogus_partition_table_fixed[NDOSPART] = {
79 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
80 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
81 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
82 	{ 0x80, 0, 1, 0, DOSPTYP_386BSD, 254, 255, 255, 0, 50000, },
83 };
84 
85 static int check_part (char *sname, struct dos_partition *dp,
86 			   u_int64_t offset, int nsectors, int ntracks,
87 			   u_int64_t mbr_offset);
88 static void mbr_extended (cdev_t dev, struct disk_info *info,
89 			      struct diskslices *ssp, u_int64_t ext_offset,
90 			      u_int64_t ext_size, u_int64_t base_ext_offset,
91 			      int nsectors, int ntracks, u_int64_t mbr_offset,
92 			      int level);
93 static int mbr_setslice (char *sname, struct disk_info *info,
94 			     struct diskslice *sp, struct dos_partition *dp,
95 			     u_int64_t br_offset);
96 
97 
98 int
99 mbrinit(cdev_t dev, struct disk_info *info, struct diskslices **sspp)
100 {
101 	struct buf *bp;
102 	u_char	*cp;
103 	int	dospart;
104 	struct dos_partition *dp;
105 	struct dos_partition *dp0;
106 	struct dos_partition dpcopy[NDOSPART];
107 	int	error;
108 	int	max_ncyls;
109 	int	max_nsectors;
110 	int	max_ntracks;
111 	u_int64_t mbr_offset;
112 	char	partname[2];
113 	u_long	secpercyl;
114 	char	*sname;
115 	struct diskslice *sp;
116 	struct diskslices *ssp;
117 	cdev_t wdev;
118 
119 	mbr_offset = DOSBBSECTOR;
120 reread_mbr:
121 	if (info->d_media_blksize & DEV_BMASK)
122 		return (EIO);
123 	/* Read master boot record. */
124 	wdev = dkmodpart(dkmodslice(dev, WHOLE_DISK_SLICE), WHOLE_SLICE_PART);
125 	bp = geteblk((int)info->d_media_blksize);
126 	bp->b_bio1.bio_offset = (off_t)mbr_offset * info->d_media_blksize;
127 	bp->b_bcount = info->d_media_blksize;
128 	bp->b_cmd = BUF_CMD_READ;
129 	dev_dstrategy(wdev, &bp->b_bio1);
130 	if (biowait(bp) != 0) {
131 		diskerr(&bp->b_bio1, wdev,
132 			"reading primary partition table: error",
133 			LOG_PRINTF, 0);
134 		kprintf("\n");
135 		error = EIO;
136 		goto done;
137 	}
138 
139 	/* Weakly verify it. */
140 	cp = bp->b_data;
141 	sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, WHOLE_SLICE_PART, partname);
142 	if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
143 		if (bootverbose)
144 			kprintf("%s: invalid primary partition table: no magic\n",
145 			       sname);
146 		error = EINVAL;
147 		goto done;
148 	}
149 
150 	/* Make a copy of the partition table to avoid alignment problems. */
151 	memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
152 
153 	dp0 = &dpcopy[0];
154 
155 	/*
156 	 * Check for "Ontrack Diskmanager" or GPT.  If a GPT is found in
157 	 * the first dos partition, ignore the rest of the MBR and go
158 	 * to GPT processing.
159 	 */
160 	for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
161 		if (dospart == 0 &&
162 		    (dp->dp_typ == DOSPTYP_PMBR || dp->dp_typ == DOSPTYP_GPT)) {
163 			if (bootverbose)
164 				kprintf(
165 	    "%s: Found GPT in slice #%d\n", sname, dospart + 1);
166 			error = gptinit(dev, info, sspp);
167 			goto done;
168 		}
169 
170 		if (dp->dp_typ == DOSPTYP_ONTRACK) {
171 			if (bootverbose)
172 				kprintf(
173 	    "%s: Found \"Ontrack Disk Manager\" on this disk.\n", sname);
174 			bp->b_flags |= B_INVAL | B_AGE;
175 			brelse(bp);
176 			mbr_offset = 63;
177 			goto reread_mbr;
178 		}
179 	}
180 
181 	if (bcmp(dp0, historical_bogus_partition_table,
182 		 sizeof historical_bogus_partition_table) == 0 ||
183 	    bcmp(dp0, historical_bogus_partition_table_fixed,
184 		 sizeof historical_bogus_partition_table_fixed) == 0) {
185 #if 0
186 		TRACE(("%s: invalid primary partition table: historical\n",
187 		       sname));
188 #endif /* 0 */
189 		if (bootverbose)
190 			kprintf(
191      "%s: invalid primary partition table: Dangerously Dedicated (ignored)\n",
192 			       sname);
193 		error = EINVAL;
194 		goto done;
195 	}
196 
197 	/* Guess the geometry. */
198 	/*
199 	 * TODO:
200 	 * Perhaps skip entries with 0 size.
201 	 * Perhaps only look at entries of type DOSPTYP_386BSD.
202 	 */
203 	max_ncyls = 0;
204 	max_nsectors = 0;
205 	max_ntracks = 0;
206 	for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
207 		int	ncyls;
208 		int	nsectors;
209 		int	ntracks;
210 
211 		ncyls = DPCYL(dp->dp_ecyl, dp->dp_esect) + 1;
212 		if (max_ncyls < ncyls)
213 			max_ncyls = ncyls;
214 		nsectors = DPSECT(dp->dp_esect);
215 		if (max_nsectors < nsectors)
216 			max_nsectors = nsectors;
217 		ntracks = dp->dp_ehd + 1;
218 		if (max_ntracks < ntracks)
219 			max_ntracks = ntracks;
220 	}
221 
222 	/*
223 	 * Check that we have guessed the geometry right by checking the
224 	 * partition entries.
225 	 */
226 	/*
227 	 * TODO:
228 	 * As above.
229 	 * Check for overlaps.
230 	 * Check against d_secperunit if the latter is reliable.
231 	 */
232 	error = 0;
233 	for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
234 		if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
235 		    && dp->dp_start == 0 && dp->dp_size == 0)
236 			continue;
237 		sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
238 			       WHOLE_SLICE_PART, partname);
239 
240 		/*
241 		 * Temporarily ignore errors from this check.  We could
242 		 * simplify things by accepting the table eariler if we
243 		 * always ignore errors here.  Perhaps we should always
244 		 * accept the table if the magic is right but not let
245 		 * bad entries affect the geometry.
246 		 */
247 		check_part(sname, dp, mbr_offset, max_nsectors, max_ntracks,
248 			   mbr_offset);
249 	}
250 	if (error != 0)
251 		goto done;
252 
253 	/*
254 	 * Accept the DOS partition table.
255 	 *
256 	 * Adjust the disk information structure with updated CHS
257 	 * conversion parameters, but only use values extracted from
258 	 * the primary partition table.
259 	 *
260 	 * NOTE!  Regardless of our having to deal with this old cruft,
261 	 * we do not screw around with the info->d_media* parameters.
262 	 */
263 	secpercyl = (u_long)max_nsectors * max_ntracks;
264 	if (secpercyl != 0 && mbr_offset == DOSBBSECTOR) {
265 		info->d_secpertrack = max_nsectors;
266 		info->d_nheads = max_ntracks;
267 		info->d_secpercyl = secpercyl;
268 		info->d_ncylinders = info->d_media_blocks / secpercyl;
269 	}
270 
271 	/*
272 	 * We are passed a pointer to a suitably initialized minimal
273 	 * slices "struct" with no dangling pointers in it.  Replace it
274 	 * by a maximal one.  This usually oversizes the "struct", but
275 	 * enlarging it while searching for logical drives would be
276 	 * inconvenient.
277 	 */
278 	kfree(*sspp, M_DEVBUF);
279 	ssp = dsmakeslicestruct(MAX_SLICES, info);
280 	*sspp = ssp;
281 
282 	/* Initialize normal slices. */
283 	sp = &ssp->dss_slices[BASE_SLICE];
284 	for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++, sp++) {
285 		sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
286 			       WHOLE_SLICE_PART, partname);
287 		(void)mbr_setslice(sname, info, sp, dp, mbr_offset);
288 	}
289 	ssp->dss_nslices = BASE_SLICE + NDOSPART;
290 
291 	/* Handle extended partitions. */
292 	sp -= NDOSPART;
293 	for (dospart = 0; dospart < NDOSPART; dospart++, sp++) {
294 		if (sp->ds_type == DOSPTYP_EXTENDED ||
295 		    sp->ds_type == DOSPTYP_EXTENDEDX) {
296 			mbr_extended(wdev, info, ssp,
297 				     sp->ds_offset, sp->ds_size, sp->ds_offset,
298 				     max_nsectors, max_ntracks, mbr_offset, 1);
299 		}
300 	}
301 
302 	/*
303 	 * mbr_extended() abuses ssp->dss_nslices for the number of slices
304 	 * that would be found if there were no limit on the number of slices
305 	 * in *ssp.  Cut it back now.
306 	 */
307 	if (ssp->dss_nslices > MAX_SLICES)
308 		ssp->dss_nslices = MAX_SLICES;
309 
310 done:
311 	bp->b_flags |= B_INVAL | B_AGE;
312 	brelse(bp);
313 	if (error == EINVAL)
314 		error = 0;
315 	return (error);
316 }
317 
318 static int
319 check_part(char *sname, struct dos_partition *dp, u_int64_t offset,
320 	    int nsectors, int ntracks, u_int64_t mbr_offset)
321 {
322 	int	chs_ecyl;
323 	int	chs_esect;
324 	int	chs_scyl;
325 	int	chs_ssect;
326 	int	error;
327 	u_long	secpercyl;
328 	u_int64_t esector;
329 	u_int64_t esector1;
330 	u_int64_t ssector;
331 	u_int64_t ssector1;
332 
333 	secpercyl = (u_long)nsectors * ntracks;
334 	chs_scyl = DPCYL(dp->dp_scyl, dp->dp_ssect);
335 	chs_ssect = DPSECT(dp->dp_ssect);
336 	ssector = chs_ssect - 1 + dp->dp_shd * nsectors + chs_scyl * secpercyl
337 		  + mbr_offset;
338 	ssector1 = offset + dp->dp_start;
339 
340 	/*
341 	 * If ssector1 is on a cylinder >= 1024, then ssector can't be right.
342 	 * Allow the C/H/S for it to be 1023/ntracks-1/nsectors, or correct
343 	 * apart from the cylinder being reduced modulo 1024.  Always allow
344 	 * 1023/255/63, because this is the official way to represent
345 	 * pure-LBA for the starting position.
346 	 */
347 	if ((ssector < ssector1
348 	     && ((chs_ssect == nsectors && dp->dp_shd == ntracks - 1
349 		  && chs_scyl == 1023)
350 		 || (secpercyl != 0
351 		     && (ssector1 - ssector) % (1024 * secpercyl) == 0)))
352 	    || (dp->dp_scyl == 255 && dp->dp_shd == 255
353 		&& dp->dp_ssect == 255)) {
354 		TRACE(("%s: C/H/S start %d/%d/%d, start %llu: allow\n",
355 		       sname, chs_scyl, dp->dp_shd, chs_ssect, ssector1));
356 		ssector = ssector1;
357 	}
358 
359 	chs_ecyl = DPCYL(dp->dp_ecyl, dp->dp_esect);
360 	chs_esect = DPSECT(dp->dp_esect);
361 	esector = chs_esect - 1 + dp->dp_ehd * nsectors + chs_ecyl * secpercyl
362 		  + mbr_offset;
363 	esector1 = ssector1 + dp->dp_size - 1;
364 
365 	/*
366 	 * Allow certain bogus C/H/S values for esector, as above. However,
367 	 * heads == 255 isn't really legal and causes some BIOS crashes. The
368 	 * correct value to indicate a pure-LBA end is 1023/heads-1/sectors -
369 	 * usually 1023/254/63. "heads" is base 0, "sectors" is base 1.
370 	 */
371 	if ((esector < esector1
372 	     && ((chs_esect == nsectors && dp->dp_ehd == ntracks - 1
373 		  && chs_ecyl == 1023)
374 		 || (secpercyl != 0
375 		     && (esector1 - esector) % (1024 * secpercyl) == 0)))
376 	    || (dp->dp_ecyl == 255 && dp->dp_ehd == 255
377 		&& dp->dp_esect == 255)) {
378 		TRACE(("%s: C/H/S end %d/%d/%d, end %llu: allow\n",
379 		       sname, chs_ecyl, dp->dp_ehd, chs_esect, esector1));
380 		esector = esector1;
381 	}
382 
383 	error = (ssector == ssector1 && esector == esector1) ? 0 : EINVAL;
384 	if (bootverbose)
385 		kprintf("%s: type 0x%x, start %llu, end = %llu, size %lu %s\n",
386 		       sname, dp->dp_typ, ssector1, esector1,
387 		       (u_long)dp->dp_size, error ? "" : ": OK");
388 	if (ssector != ssector1 && bootverbose)
389 		kprintf("%s: C/H/S start %d/%d/%d (%llu) != start %llu: invalid\n",
390 		       sname, chs_scyl, dp->dp_shd, chs_ssect,
391 		       ssector, ssector1);
392 	if (esector != esector1 && bootverbose)
393 		kprintf("%s: C/H/S end %d/%d/%d (%llu) != end %llu: invalid\n",
394 		       sname, chs_ecyl, dp->dp_ehd, chs_esect,
395 		       esector, esector1);
396 	return (error);
397 }
398 
399 static
400 void
401 mbr_extended(cdev_t dev, struct disk_info *info, struct diskslices *ssp,
402 	    u_int64_t ext_offset, u_int64_t ext_size, u_int64_t base_ext_offset,
403 	    int nsectors, int ntracks, u_int64_t mbr_offset, int level)
404 {
405 	struct buf *bp;
406 	u_char	*cp;
407 	int	dospart;
408 	struct dos_partition *dp;
409 	struct dos_partition dpcopy[NDOSPART];
410 	u_int64_t ext_offsets[NDOSPART];
411 	u_int64_t ext_sizes[NDOSPART];
412 	char	partname[2];
413 	int	slice;
414 	char	*sname;
415 	struct diskslice *sp;
416 
417 	if (level >= 16) {
418 		kprintf(
419 	"%s: excessive recursion in search for slices; aborting search\n",
420 		       devtoname(dev));
421 		return;
422 	}
423 
424 	/* Read extended boot record. */
425 	bp = geteblk((int)info->d_media_blksize);
426 	bp->b_bio1.bio_offset = (off_t)ext_offset * info->d_media_blksize;
427 	bp->b_bcount = info->d_media_blksize;
428 	bp->b_cmd = BUF_CMD_READ;
429 	dev_dstrategy(dev, &bp->b_bio1);
430 	if (biowait(bp) != 0) {
431 		diskerr(&bp->b_bio1, dev,
432 			"reading extended partition table: error",
433 			LOG_PRINTF, 0);
434 		kprintf("\n");
435 		goto done;
436 	}
437 
438 	/* Weakly verify it. */
439 	cp = bp->b_data;
440 	if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
441 		sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, WHOLE_SLICE_PART,
442 			       partname);
443 		if (bootverbose)
444 			kprintf("%s: invalid extended partition table: no magic\n",
445 			       sname);
446 		goto done;
447 	}
448 
449 	/* Make a copy of the partition table to avoid alignment problems. */
450 	memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
451 
452 	slice = ssp->dss_nslices;
453 	for (dospart = 0, dp = &dpcopy[0]; dospart < NDOSPART;
454 	    dospart++, dp++) {
455 		ext_sizes[dospart] = 0;
456 		if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
457 		    && dp->dp_start == 0 && dp->dp_size == 0)
458 			continue;
459 		if (dp->dp_typ == DOSPTYP_EXTENDED ||
460 		    dp->dp_typ == DOSPTYP_EXTENDEDX) {
461 			static char buf[32];
462 
463 			sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE,
464 				       WHOLE_SLICE_PART, partname);
465 			ksnprintf(buf, sizeof(buf), "%s", sname);
466 			if (strlen(buf) < sizeof buf - 11)
467 				strcat(buf, "<extended>");
468 			check_part(buf, dp, base_ext_offset, nsectors,
469 				   ntracks, mbr_offset);
470 			ext_offsets[dospart] = base_ext_offset + dp->dp_start;
471 			ext_sizes[dospart] = dp->dp_size;
472 		} else {
473 			sname = dsname(dev, dkunit(dev), slice, WHOLE_SLICE_PART,
474 				       partname);
475 			check_part(sname, dp, ext_offset, nsectors, ntracks,
476 				   mbr_offset);
477 			if (slice >= MAX_SLICES) {
478 				kprintf("%s: too many slices\n", sname);
479 				slice++;
480 				continue;
481 			}
482 			sp = &ssp->dss_slices[slice];
483 			if (mbr_setslice(sname, info, sp, dp, ext_offset) != 0)
484 				continue;
485 			slice++;
486 		}
487 	}
488 	ssp->dss_nslices = slice;
489 
490 	/* If we found any more slices, recursively find all the subslices. */
491 	for (dospart = 0; dospart < NDOSPART; dospart++) {
492 		if (ext_sizes[dospart] != 0) {
493 			mbr_extended(dev, info, ssp, ext_offsets[dospart],
494 				     ext_sizes[dospart], base_ext_offset,
495 				     nsectors, ntracks, mbr_offset, ++level);
496 		}
497 	}
498 
499 done:
500 	bp->b_flags |= B_INVAL | B_AGE;
501 	brelse(bp);
502 }
503 
504 static int
505 mbr_setslice(char *sname, struct disk_info *info, struct diskslice *sp,
506 	    struct dos_partition *dp, u_int64_t br_offset)
507 {
508 	u_int64_t	offset;
509 	u_int64_t	size;
510 
511 	offset = br_offset + dp->dp_start;
512 	if (offset > info->d_media_blocks || offset < br_offset) {
513 		kprintf(
514 		"%s: slice starts beyond end of the disk: rejecting it\n",
515 		       sname);
516 		return (1);
517 	}
518 	size = info->d_media_blocks - offset;
519 	if (size >= dp->dp_size)
520 		size = dp->dp_size;
521 	else
522 		kprintf(
523 "%s: slice extends beyond end of disk: truncating from %lu to %llu sectors\n",
524 		       sname, (u_long)dp->dp_size, size);
525 	sp->ds_offset = offset;
526 	sp->ds_size = size;
527 	sp->ds_type = dp->dp_typ;
528 	bzero(&sp->ds_type_uuid, sizeof(sp->ds_type_uuid));
529 	bzero(&sp->ds_stor_uuid, sizeof(sp->ds_type_uuid));
530 
531 	/*
532 	 * Slices do not overlap with the parent (if any).
533 	 */
534 	sp->ds_reserved = 0;
535 	return (0);
536 }
537