Home | History | Annotate | Line # | Download | only in kern
subr_disk_mbr.c revision 1.1
      1 /*	$NetBSD: subr_disk_mbr.c,v 1.1 2003/07/07 13:28:48 dsl Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)ufs_disksubr.c	7.16 (Berkeley) 5/4/91
     36  */
     37 
     38 /*
     39  * Code to find a NetBSD label on a disk that contains an i386 style MBR.
     40  * The first NetBSD label found in the 2nd sector of a NetBSD partition
     41  * is used.  If there isn't a NetBSD partition then a label is faked up
     42  * from the MBR.
     43  *
     44  * If there isn't a disklabel or anything in the MBR then partition a
     45  * is set to cover the whole disk.
     46  * Useful for files that contain single filesystems (etc).
     47  *
     48  * This code will read host endian netbsd labels from little endian MBR.
     49  *
     50  * Based on the i386 disksubr.c
     51  *
     52  * XXX There are potential problems writing labels to disks where there
     53  * is only space for 8 netbsd partitions but this code has been compiled
     54  * with MAXPARTITIONS=16.
     55  */
     56 
     57 #include <sys/cdefs.h>
     58 __KERNEL_RCSID(0, "$NetBSD: subr_disk_mbr.c,v 1.1 2003/07/07 13:28:48 dsl Exp $");
     59 
     60 #include <sys/param.h>
     61 #include <sys/systm.h>
     62 #include <sys/buf.h>
     63 #include <sys/disklabel.h>
     64 #include <sys/disk.h>
     65 #include <sys/syslog.h>
     66 
     67 #include "opt_mbr.h"
     68 
     69 #define MBR_LABELSECTOR	1
     70 
     71 #define SCAN_CONTINUE	0
     72 #define SCAN_FOUND	1
     73 #define SCAN_ERROR	2
     74 
     75 typedef struct mbr_args {
     76 	struct disklabel *lp;
     77 	void		(*strat)(struct buf *);
     78 	struct buf	*bp;
     79 	const char	*msg;
     80 	int		error;
     81 } mbr_args_t;
     82 
     83 #define READ_LABEL	1
     84 #define WRITE_LABEL	2
     85 static int validate_label(mbr_args_t *, uint, int);
     86 static int look_netbsd_part(mbr_args_t *, mbr_partition_t *, int, uint);
     87 static int write_netbsd_label(mbr_args_t *, mbr_partition_t *, int, uint);
     88 
     89 static int
     90 read_sector(mbr_args_t *a, uint sector)
     91 {
     92 	struct buf *bp = a->bp;
     93 	int error;
     94 
     95 	bp->b_blkno = sector;
     96 	bp->b_bcount = a->lp->d_secsize;
     97 	bp->b_flags = (bp->b_flags & ~(B_WRITE | B_DONE)) | B_READ;
     98 	bp->b_cylinder = sector / a->lp->d_secpercyl;
     99 	(*a->strat)(bp);
    100 	error = biowait(bp);
    101 	if (error != 0)
    102 		a->error = error;
    103 	return error;
    104 }
    105 
    106 /*
    107  * Scan MBR for partitions, call 'action' routine for each.
    108  */
    109 
    110 static int
    111 scan_mbr(mbr_args_t *a, int (*actn)(mbr_args_t *, mbr_partition_t *, int, uint))
    112 {
    113 	mbr_partition_t ptns[NMBRPART];
    114 	mbr_partition_t *dp;
    115 	mbr_sector_t *mbr;
    116 	uint ext_base, this_ext, next_ext;
    117 	int rval;
    118 	int i;
    119 #ifdef COMPAT_386BSD_MBRPART
    120 	int dp_386bsd = -1;
    121 #endif
    122 
    123 	ext_base = 0;
    124 	this_ext = 0;
    125 	for (;;) {
    126 		if (read_sector(a, this_ext)) {
    127 			a->msg = "dos partition I/O error";
    128 			return SCAN_ERROR;
    129 		}
    130 
    131 		/* Note: Magic number is little-endian. */
    132 		mbr = (void *)a->bp->b_data;
    133 		if (mbr->mbr_signature != htole16(MBR_MAGIC)) {
    134 			/* Don't set an error message here.... */
    135 			return SCAN_ERROR;
    136 		}
    137 
    138 		/* Copy data out of buffer so action can use bp */
    139 		memcpy(ptns, &mbr->mbr_parts, sizeof ptns);
    140 
    141 		/* look for NetBSD partition */
    142 		next_ext = 0;
    143 		dp = ptns;
    144 		for (i = 0; i < NMBRPART; i++, dp++) {
    145 			if (dp->mbrp_typ == 0)
    146 				continue;
    147 			if (MBR_IS_EXTENDED(dp->mbrp_typ)) {
    148 				next_ext = le32toh(dp->mbrp_start);
    149 				continue;
    150 			}
    151 #ifdef COMPAT_386BSD_MBRPART
    152 			if (dp->mbrp_typ == MBR_PTYPE_386BSD) {
    153 				/*
    154 				 * If more than one matches, take last,
    155 				 * as NetBSD install tool does.
    156 				 */
    157 				if (this_ext == 0)
    158 					dp_386bsd = i;
    159 				continue;
    160 			}
    161 #endif
    162 			rval = (*actn)(a, dp, i, this_ext);
    163 			if (rval != SCAN_CONTINUE)
    164 				return rval;
    165 		}
    166 		if (next_ext == 0)
    167 			break;
    168 		if (ext_base == 0) {
    169 			ext_base = next_ext;
    170 			next_ext = 0;
    171 		}
    172 		next_ext += ext_base;
    173 		if (next_ext <= this_ext)
    174 			break;
    175 		this_ext = next_ext;
    176 	}
    177 #ifdef COMPAT_386BSD_MBRPART
    178 	if (this_ext == 0 && dp_386bsd != -1)
    179 		return (*actn)(a, &ptns[dp_386bsd], dp_386bsd, 0);
    180 #endif
    181 	return SCAN_CONTINUE;
    182 }
    183 
    184 /*
    185  * Attempt to read a disk label from a device
    186  * using the indicated strategy routine.
    187  * The label must be partly set up before this:
    188  * secpercyl, secsize and anything required for a block i/o read
    189  * operation in the driver's strategy/start routines
    190  * must be filled in before calling us.
    191  *
    192  * If dos partition table requested, attempt to load it and
    193  * find disklabel inside a DOS partition. Also, if bad block
    194  * table needed, attempt to extract it as well. Return buffer
    195  * for use in signalling errors if requested.
    196  *
    197  * Returns null on success and an error string on failure.
    198  */
    199 const char *
    200 readdisklabel(dev, strat, lp, osdep)
    201 	dev_t dev;
    202 	void (*strat)(struct buf *);
    203 	struct disklabel *lp;
    204 	struct cpu_disklabel *osdep;
    205 {
    206 	struct dkbad *bdp;
    207 	int rval;
    208 	int i;
    209 	mbr_args_t a;
    210 
    211 	a.lp = lp;
    212 	a.strat = strat;
    213 	a.msg = NULL;
    214 	a.error = 0;
    215 
    216 	/* minimal requirements for architypal disk label */
    217 	if (lp->d_secsize == 0)
    218 		lp->d_secsize = DEV_BSIZE;
    219 	if (lp->d_secperunit == 0)
    220 		lp->d_secperunit = 0x1fffffff;
    221 	lp->d_npartitions = RAW_PART + 1;
    222 	for (i = 0; i < RAW_PART; i++) {
    223 		lp->d_partitions[i].p_size = 0;
    224 		lp->d_partitions[i].p_offset = 0;
    225 	}
    226 	if (lp->d_partitions[RAW_PART].p_size == 0)
    227 		lp->d_partitions[RAW_PART].p_size = 0x1fffffff;
    228 	lp->d_partitions[RAW_PART].p_offset = 0;
    229 
    230 	/*
    231 	 * Set partition 'a' to be the whole disk.
    232 	 * Cleared if we find an mbr or a netbsd label.
    233 	 */
    234 	lp->d_partitions[0].p_size = lp->d_partitions[RAW_PART].p_size;
    235 	lp->d_partitions[0].p_fstype = FS_BSDFFS;
    236 
    237 	/* get a buffer and initialize it */
    238 	a.bp = geteblk((int)lp->d_secsize);
    239 	a.bp->b_dev = dev;
    240 
    241 	if (osdep)
    242 		/*
    243 		 * Scan mbr searching for netbsd partition and saving
    244 		 * bios partition information to use if the netbsd one
    245 		 * is absent.
    246 		 */
    247 		rval = scan_mbr(&a, look_netbsd_part);
    248 	else
    249 		rval = SCAN_CONTINUE;
    250 
    251 	if (rval == SCAN_CONTINUE || (rval == SCAN_ERROR && a.msg == NULL)) {
    252 		/* Look at start of disk */
    253 		rval = validate_label(&a, LABELSECTOR, READ_LABEL);
    254 		if (LABELSECTOR != 0 && rval == SCAN_CONTINUE)
    255 			rval = validate_label(&a, 0, READ_LABEL);
    256 	}
    257 
    258 	/* Obtain bad sector table if requested and present */
    259 	if (rval == SCAN_FOUND && osdep && (lp->d_flags & D_BADSECT)) {
    260 		struct dkbad *db;
    261 		int blkno;
    262 
    263 		bdp = &osdep->bad;
    264 		i = 0;
    265 		rval = SCAN_ERROR;
    266 		do {
    267 			/* read a bad sector table */
    268 			blkno = lp->d_secperunit - lp->d_nsectors + i;
    269 			if (lp->d_secsize > DEV_BSIZE)
    270 				blkno *= lp->d_secsize / DEV_BSIZE;
    271 			else
    272 				blkno /= DEV_BSIZE / lp->d_secsize;
    273 			/* if successful, validate, otherwise try another */
    274 			if (read_sector(&a, blkno)) {
    275 				a.msg = "bad sector table I/O error";
    276 				continue;
    277 			}
    278 			db = (struct dkbad *)(a.bp->b_data);
    279 #define DKBAD_MAGIC 0x4321
    280 			if (db->bt_mbz != 0 || db->bt_flag != DKBAD_MAGIC) {
    281 				a.msg = "bad sector table corrupted";
    282 				continue;
    283 			}
    284 			rval = SCAN_FOUND;
    285 			*bdp = *db;
    286 			break;
    287 		} while ((a.bp->b_flags & B_ERROR) && (i += 2) < 10 &&
    288 			i < lp->d_nsectors);
    289 	}
    290 
    291 	brelse(a.bp);
    292 	if (rval != SCAN_ERROR)
    293 		return NULL;
    294 	return a.msg;
    295 }
    296 
    297 static int
    298 look_netbsd_part(mbr_args_t *a, mbr_partition_t *dp, int slot, uint ext_base)
    299 {
    300 	struct partition *pp;
    301 	int ptn_base = ext_base + le32toh(dp->mbrp_start);
    302 	int rval;
    303 
    304 	if (
    305 #ifdef COMPAT_386BSD_MBRPART
    306 	    dp->mbrp_typ == MBR_PTYPE_386BSD ||
    307 #endif
    308 	    dp->mbrp_typ == MBR_PTYPE_NETBSD) {
    309 		rval = validate_label(a, ptn_base + MBR_LABELSECTOR, READ_LABEL);
    310 
    311 		/* Put actual location where we found the label into ptn 2 */
    312 		if (RAW_PART != 2 && (rval == SCAN_FOUND ||
    313 					a->lp->d_partitions[2].p_size == 0)) {
    314 			a->lp->d_partitions[2].p_size = le32toh(dp->mbrp_size);
    315 			a->lp->d_partitions[2].p_offset = ptn_base;
    316 		}
    317 
    318 		/* If we got a netbsd label look no further */
    319 		if (rval == SCAN_FOUND)
    320 			return rval;
    321 	}
    322 
    323 	/* Install main partitions into e..h and extended into i+ */
    324 	if (ext_base == 0)
    325 		slot += 4;
    326 	else {
    327 		slot = 4 + NMBRPART;
    328 		pp = &a->lp->d_partitions[slot];
    329 		for (; slot < MAXPARTITIONS; pp++, slot++) {
    330 			/* This gets called twice - avoid duplicates */
    331 			if (pp->p_offset == ptn_base &&
    332 			    pp->p_size == le32toh(dp->mbrp_size))
    333 				break;
    334 			if (pp->p_size == 0)
    335 				break;
    336 		}
    337 	}
    338 
    339 	if (slot < MAXPARTITIONS) {
    340 		/* Stop 'a' being the entire disk */
    341 		a->lp->d_partitions[0].p_size = 0;
    342 		a->lp->d_partitions[0].p_fstype = 0;
    343 
    344 		/* save partition info */
    345 		pp = &a->lp->d_partitions[slot];
    346 		pp->p_offset = ptn_base;
    347 		pp->p_size = le32toh(dp->mbrp_size);
    348 		pp->p_fstype = xlat_mbr_fstype(dp->mbrp_typ);
    349 
    350 		if (slot >= a->lp->d_npartitions)
    351 			a->lp->d_npartitions = slot + 1;
    352 	}
    353 
    354 	return SCAN_CONTINUE;
    355 }
    356 
    357 
    358 static int
    359 validate_label(mbr_args_t *a, uint label_sector, int action)
    360 {
    361 	struct disklabel *dlp, *dlp_lim;
    362 	int error;
    363 
    364 	/* Next, dig out disk label */
    365 	if (read_sector(a, label_sector)) {
    366 		a->msg =  "disk label read failed";
    367 		return SCAN_ERROR;
    368 	}
    369 
    370 	/* Locate disk label within block and validate */
    371 	/*
    372 	 * XXX (dsl) This search may be a waste of time, a lot of other i386
    373 	 * code assumes the label is at offset LABELOFFSET (=0) in the sector.
    374 	 *
    375 	 * If we want to support diska from other netbsd ports, then the
    376 	 * code should also allow for a shorter label nearer the end of
    377 	 * the disk sector, and (IIRC) labels within 8k of the disk start.
    378 	 */
    379 	dlp_lim = (void *)(a->bp->b_data + a->lp->d_secsize - sizeof(*dlp));
    380 	for (dlp = (void *)a->bp->b_data; dlp <= dlp_lim;
    381 	    dlp = (void *)((char *)dlp + sizeof(long))) {
    382 		if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC)
    383 			continue;
    384 		if (dlp->d_npartitions > MAXPARTITIONS || dkcksum(dlp) != 0) {
    385 			a->msg = "disk label corrupted";
    386 			continue;
    387 		}
    388 		switch (action) {
    389 		case READ_LABEL:
    390 			*a->lp = *dlp;
    391 			break;
    392 		case WRITE_LABEL:
    393 			*dlp = *a->lp;
    394 			a->bp->b_flags &= ~(B_READ|B_DONE);
    395 			a->bp->b_flags |= B_WRITE;
    396 			(*a->strat)(a->bp);
    397 			error = biowait(a->bp);
    398 			if (error != 0) {
    399 				a->error = error;
    400 				a->msg = "disk label write failed";
    401 				return SCAN_ERROR;
    402 			}
    403 			break;
    404 		}
    405 		return SCAN_FOUND;
    406 	}
    407 
    408 	return SCAN_CONTINUE;
    409 }
    410 
    411 /*
    412  * Check new disk label for sensibility
    413  * before setting it.
    414  */
    415 int
    416 setdisklabel(olp, nlp, openmask, osdep)
    417 	struct disklabel *olp, *nlp;
    418 	u_long openmask;
    419 	struct cpu_disklabel *osdep;
    420 {
    421 	int i;
    422 	struct partition *opp, *npp;
    423 
    424 	/* sanity clause */
    425 	if (nlp->d_secpercyl == 0 || nlp->d_secsize == 0
    426 		|| (nlp->d_secsize % DEV_BSIZE) != 0)
    427 			return (EINVAL);
    428 
    429 	/* special case to allow disklabel to be invalidated */
    430 	if (nlp->d_magic == 0xffffffff) {
    431 		*olp = *nlp;
    432 		return (0);
    433 	}
    434 
    435 	if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC ||
    436 	    dkcksum(nlp) != 0)
    437 		return (EINVAL);
    438 
    439 	/* XXX missing check if other dos partitions will be overwritten */
    440 
    441 	while (openmask != 0) {
    442 		i = ffs(openmask) - 1;
    443 		openmask &= ~(1 << i);
    444 		if (i > nlp->d_npartitions)
    445 			return (EBUSY);
    446 		opp = &olp->d_partitions[i];
    447 		npp = &nlp->d_partitions[i];
    448 		/*
    449 		 * Copy internally-set partition information
    450 		 * if new label doesn't include it.		XXX
    451 		 */
    452 		if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) {
    453 			*npp = *opp;
    454 			continue;
    455 		}
    456 		if (npp->p_offset != opp->p_offset || npp->p_size < opp->p_size)
    457 			return (EBUSY);
    458 	}
    459  	nlp->d_checksum = 0;
    460  	nlp->d_checksum = dkcksum(nlp);
    461 	*olp = *nlp;
    462 	return (0);
    463 }
    464 
    465 
    466 /*
    467  * Write disk label back to device after modification.
    468  */
    469 int
    470 writedisklabel(dev, strat, lp, osdep)
    471 	dev_t dev;
    472 	void (*strat)(struct buf *);
    473 	struct disklabel *lp;
    474 	struct cpu_disklabel *osdep;
    475 {
    476 	mbr_args_t a;
    477 	int rval;
    478 
    479 	a.lp = lp;
    480 	a.strat = strat;
    481 	a.msg = NULL;
    482 	a.error = 0;
    483 
    484 	/* get a buffer and initialize it */
    485 	a.bp = geteblk((int)lp->d_secsize);
    486 	a.bp->b_dev = dev;
    487 
    488 	if (osdep)
    489 		rval = scan_mbr(&a, write_netbsd_label);
    490 	else
    491 		rval = validate_label(&a, LABELSECTOR, WRITE_LABEL);
    492 
    493 	if (rval != SCAN_FOUND && a.error == 0)
    494 		a.error = ESRCH;
    495 
    496 	brelse(a.bp);
    497 	return a.error;
    498 }
    499 
    500 static int
    501 write_netbsd_label(mbr_args_t *a, mbr_partition_t *dp, int slot, uint ext_base)
    502 {
    503 	int ptn_base = ext_base + le32toh(dp->mbrp_start);
    504 	int rval;
    505 
    506 	if (dp->mbrp_typ != MBR_PTYPE_NETBSD)
    507 		return SCAN_CONTINUE;
    508 
    509 	rval = validate_label(a, ptn_base + MBR_LABELSECTOR, WRITE_LABEL);
    510 #ifdef notyet 	/* because the error reporting is wrong */
    511 	/* write label to all netbsd partitions */
    512 	if (rval == SCAN_FOUND)
    513 		return SCAN_CONTINUE;
    514 #endif
    515 	return rval;
    516 }
    517 
    518 
    519 /*
    520  * Determine the size of the transfer, and make sure it is
    521  * within the boundaries of the partition. Adjust transfer
    522  * if needed, and signal errors or early completion.
    523  */
    524 int
    525 bounds_check_with_label(dk, bp, wlabel)
    526 	struct disk *dk;
    527 	struct buf *bp;
    528 	int wlabel;
    529 {
    530 	struct disklabel *lp = dk->dk_label;
    531 	struct partition *p = lp->d_partitions + DISKPART(bp->b_dev);
    532 	int labelsector = lp->d_partitions[2].p_offset + LABELSECTOR;
    533 	int sz;
    534 
    535 	sz = howmany(bp->b_bcount, lp->d_secsize);
    536 
    537 	if (bp->b_blkno + sz > p->p_size) {
    538 		sz = p->p_size - bp->b_blkno;
    539 		if (sz == 0) {
    540 			/* If exactly at end of disk, return EOF. */
    541 			bp->b_resid = bp->b_bcount;
    542 			return (0);
    543 		}
    544 		if (sz < 0) {
    545 			/* If past end of disk, return EINVAL. */
    546 			bp->b_error = EINVAL;
    547 			goto bad;
    548 		}
    549 		/* Otherwise, truncate request. */
    550 		bp->b_bcount = sz << DEV_BSHIFT;
    551 	}
    552 
    553 	/* Overwriting disk label? */
    554 	if (bp->b_blkno + p->p_offset <= labelsector &&
    555 	    bp->b_blkno + p->p_offset + sz > labelsector &&
    556 	    (bp->b_flags & B_READ) == 0 && !wlabel) {
    557 		bp->b_error = EROFS;
    558 		goto bad;
    559 	}
    560 
    561 	/* calculate cylinder for disksort to order transfers with */
    562 	bp->b_cylinder = (bp->b_blkno + p->p_offset) /
    563 	    (lp->d_secsize / DEV_BSIZE) / lp->d_secpercyl;
    564 	return (1);
    565 
    566 bad:
    567 	bp->b_flags |= B_ERROR;
    568 	return (-1);
    569 }
    570