Home | History | Annotate | Line # | Download | only in kern
subr_disk.c revision 1.93.10.1
      1 /*	$NetBSD: subr_disk.c,v 1.93.10.1 2009/04/04 17:49:21 snj Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1996, 1997, 1999, 2000, 2009 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     30  * POSSIBILITY OF SUCH DAMAGE.
     31  */
     32 
     33 /*
     34  * Copyright (c) 1982, 1986, 1988, 1993
     35  *	The Regents of the University of California.  All rights reserved.
     36  * (c) UNIX System Laboratories, Inc.
     37  * All or some portions of this file are derived from material licensed
     38  * to the University of California by American Telephone and Telegraph
     39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     40  * the permission of UNIX System Laboratories, Inc.
     41  *
     42  * Redistribution and use in source and binary forms, with or without
     43  * modification, are permitted provided that the following conditions
     44  * are met:
     45  * 1. Redistributions of source code must retain the above copyright
     46  *    notice, this list of conditions and the following disclaimer.
     47  * 2. Redistributions in binary form must reproduce the above copyright
     48  *    notice, this list of conditions and the following disclaimer in the
     49  *    documentation and/or other materials provided with the distribution.
     50  * 3. Neither the name of the University nor the names of its contributors
     51  *    may be used to endorse or promote products derived from this software
     52  *    without specific prior written permission.
     53  *
     54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     64  * SUCH DAMAGE.
     65  *
     66  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
     67  */
     68 
     69 #include <sys/cdefs.h>
     70 __KERNEL_RCSID(0, "$NetBSD: subr_disk.c,v 1.93.10.1 2009/04/04 17:49:21 snj Exp $");
     71 
     72 #include <sys/param.h>
     73 #include <sys/kernel.h>
     74 #include <sys/malloc.h>
     75 #include <sys/buf.h>
     76 #include <sys/syslog.h>
     77 #include <sys/disklabel.h>
     78 #include <sys/disk.h>
     79 #include <sys/sysctl.h>
     80 #include <lib/libkern/libkern.h>
     81 
     82 /*
     83  * Compute checksum for disk label.
     84  */
     85 u_int
     86 dkcksum(struct disklabel *lp)
     87 {
     88 	return dkcksum_sized(lp, lp->d_npartitions);
     89 }
     90 
     91 u_int
     92 dkcksum_sized(struct disklabel *lp, size_t npartitions)
     93 {
     94 	u_short *start, *end;
     95 	u_short sum = 0;
     96 
     97 	start = (u_short *)lp;
     98 	end = (u_short *)&lp->d_partitions[npartitions];
     99 	while (start < end)
    100 		sum ^= *start++;
    101 	return (sum);
    102 }
    103 
    104 /*
    105  * Disk error is the preface to plaintive error messages
    106  * about failing disk transfers.  It prints messages of the form
    107 
    108 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
    109 
    110  * if the offset of the error in the transfer and a disk label
    111  * are both available.  blkdone should be -1 if the position of the error
    112  * is unknown; the disklabel pointer may be null from drivers that have not
    113  * been converted to use them.  The message is printed with printf
    114  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
    115  * The message should be completed (with at least a newline) with printf
    116  * or addlog, respectively.  There is no trailing space.
    117  */
    118 #ifndef PRIdaddr
    119 #define PRIdaddr PRId64
    120 #endif
    121 void
    122 diskerr(const struct buf *bp, const char *dname, const char *what, int pri,
    123     int blkdone, const struct disklabel *lp)
    124 {
    125 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
    126 	void (*pr)(const char *, ...);
    127 	char partname = 'a' + part;
    128 	daddr_t sn;
    129 
    130 	if (/*CONSTCOND*/0)
    131 		/* Compiler will error this is the format is wrong... */
    132 		printf("%" PRIdaddr, bp->b_blkno);
    133 
    134 	if (pri != LOG_PRINTF) {
    135 		static const char fmt[] = "";
    136 		log(pri, fmt);
    137 		pr = addlog;
    138 	} else
    139 		pr = printf;
    140 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
    141 	    bp->b_flags & B_READ ? "read" : "writ");
    142 	sn = bp->b_blkno;
    143 	if (bp->b_bcount <= DEV_BSIZE)
    144 		(*pr)("%" PRIdaddr, sn);
    145 	else {
    146 		if (blkdone >= 0) {
    147 			sn += blkdone;
    148 			(*pr)("%" PRIdaddr " of ", sn);
    149 		}
    150 		(*pr)("%" PRIdaddr "-%" PRIdaddr "", bp->b_blkno,
    151 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
    152 	}
    153 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
    154 		sn += lp->d_partitions[part].p_offset;
    155 		(*pr)(" (%s%d bn %" PRIdaddr "; cn %" PRIdaddr "",
    156 		    dname, unit, sn, sn / lp->d_secpercyl);
    157 		sn %= lp->d_secpercyl;
    158 		(*pr)(" tn %" PRIdaddr " sn %" PRIdaddr ")",
    159 		    sn / lp->d_nsectors, sn % lp->d_nsectors);
    160 	}
    161 }
    162 
    163 /*
    164  * Searches the iostatlist for the disk corresponding to the
    165  * name provided.
    166  */
    167 struct disk *
    168 disk_find(const char *name)
    169 {
    170 	struct io_stats *stat;
    171 
    172 	stat = iostat_find(name);
    173 
    174 	if ((stat != NULL) && (stat->io_type == IOSTAT_DISK))
    175 		return stat->io_parent;
    176 
    177 	return (NULL);
    178 }
    179 
    180 void
    181 disk_init(struct disk *diskp, const char *name, const struct dkdriver *driver)
    182 {
    183 
    184 	/*
    185 	 * Initialize the wedge-related locks and other fields.
    186 	 */
    187 	mutex_init(&diskp->dk_rawlock, MUTEX_DEFAULT, IPL_NONE);
    188 	mutex_init(&diskp->dk_openlock, MUTEX_DEFAULT, IPL_NONE);
    189 	LIST_INIT(&diskp->dk_wedges);
    190 	diskp->dk_nwedges = 0;
    191 	diskp->dk_labelsector = LABELSECTOR;
    192 	disk_blocksize(diskp, DEV_BSIZE);
    193 	diskp->dk_name = name;
    194 	diskp->dk_driver = driver;
    195 }
    196 
    197 /*
    198  * Attach a disk.
    199  */
    200 void
    201 disk_attach(struct disk *diskp)
    202 {
    203 
    204 	/*
    205 	 * Allocate and initialize the disklabel structures.  Note that
    206 	 * it's not safe to sleep here, since we're probably going to be
    207 	 * called during autoconfiguration.
    208 	 */
    209 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT);
    210 	diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF,
    211 	    M_NOWAIT);
    212 	if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
    213 		panic("disk_attach: can't allocate storage for disklabel");
    214 
    215 	memset(diskp->dk_label, 0, sizeof(struct disklabel));
    216 	memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel));
    217 
    218 	/*
    219 	 * Set up the stats collection.
    220 	 */
    221 	diskp->dk_stats = iostat_alloc(IOSTAT_DISK, diskp, diskp->dk_name);
    222 }
    223 
    224 /*
    225  * Detach a disk.
    226  */
    227 void
    228 disk_detach(struct disk *diskp)
    229 {
    230 
    231 	/*
    232 	 * Remove from the drivelist.
    233 	 */
    234 	iostat_free(diskp->dk_stats);
    235 
    236 	/*
    237 	 * Release the disk-info dictionary.
    238 	 */
    239 	if (diskp->dk_info) {
    240 		prop_object_release(diskp->dk_info);
    241 		diskp->dk_info = NULL;
    242 	}
    243 
    244 	/*
    245 	 * Free the space used by the disklabel structures.
    246 	 */
    247 	free(diskp->dk_label, M_DEVBUF);
    248 	free(diskp->dk_cpulabel, M_DEVBUF);
    249 }
    250 
    251 void
    252 disk_destroy(struct disk *diskp)
    253 {
    254 
    255 	mutex_destroy(&diskp->dk_openlock);
    256 	mutex_destroy(&diskp->dk_rawlock);
    257 }
    258 
    259 /*
    260  * Mark the disk as busy for metrics collection.
    261  */
    262 void
    263 disk_busy(struct disk *diskp)
    264 {
    265 
    266 	iostat_busy(diskp->dk_stats);
    267 }
    268 
    269 /*
    270  * Finished disk operations, gather metrics.
    271  */
    272 void
    273 disk_unbusy(struct disk *diskp, long bcount, int read)
    274 {
    275 
    276 	iostat_unbusy(diskp->dk_stats, bcount, read);
    277 }
    278 
    279 /*
    280  * Return true if disk has an I/O operation in flight.
    281  */
    282 bool
    283 disk_isbusy(struct disk *diskp)
    284 {
    285 
    286 	return iostat_isbusy(diskp->dk_stats);
    287 }
    288 
    289 /*
    290  * Set the physical blocksize of a disk, in bytes.
    291  * Only necessary if blocksize != DEV_BSIZE.
    292  */
    293 void
    294 disk_blocksize(struct disk *diskp, int blocksize)
    295 {
    296 
    297 	diskp->dk_blkshift = DK_BSIZE2BLKSHIFT(blocksize);
    298 	diskp->dk_byteshift = DK_BSIZE2BYTESHIFT(blocksize);
    299 }
    300 
    301 /*
    302  * Bounds checking against the media size, used for the raw partition.
    303  * The sector size passed in should currently always be DEV_BSIZE,
    304  * and the media size the size of the device in DEV_BSIZE sectors.
    305  */
    306 int
    307 bounds_check_with_mediasize(struct buf *bp, int secsize, uint64_t mediasize)
    308 {
    309 	int64_t sz;
    310 
    311 	sz = howmany(bp->b_bcount, secsize);
    312 
    313 	if (bp->b_blkno + sz > mediasize) {
    314 		sz = mediasize - bp->b_blkno;
    315 		if (sz == 0) {
    316 			/* If exactly at end of disk, return EOF. */
    317 			bp->b_resid = bp->b_bcount;
    318 			return 0;
    319 		}
    320 		if (sz < 0) {
    321 			/* If past end of disk, return EINVAL. */
    322 			bp->b_error = EINVAL;
    323 			return 0;
    324 		}
    325 		/* Otherwise, truncate request. */
    326 		bp->b_bcount = sz << DEV_BSHIFT;
    327 	}
    328 
    329 	return 1;
    330 }
    331 
    332 /*
    333  * Determine the size of the transfer, and make sure it is
    334  * within the boundaries of the partition. Adjust transfer
    335  * if needed, and signal errors or early completion.
    336  */
    337 int
    338 bounds_check_with_label(struct disk *dk, struct buf *bp, int wlabel)
    339 {
    340 	struct disklabel *lp = dk->dk_label;
    341 	struct partition *p = lp->d_partitions + DISKPART(bp->b_dev);
    342 	uint64_t p_size, p_offset, labelsector;
    343 	int64_t sz;
    344 
    345 	/* Protect against division by zero. XXX: Should never happen?!?! */
    346 	if (lp->d_secpercyl == 0) {
    347 		bp->b_error = EINVAL;
    348 		return -1;
    349 	}
    350 
    351 	p_size = p->p_size << dk->dk_blkshift;
    352 	p_offset = p->p_offset << dk->dk_blkshift;
    353 #if RAW_PART == 3
    354 	labelsector = lp->d_partitions[2].p_offset;
    355 #else
    356 	labelsector = lp->d_partitions[RAW_PART].p_offset;
    357 #endif
    358 	labelsector = (labelsector + dk->dk_labelsector) << dk->dk_blkshift;
    359 
    360 	sz = howmany(bp->b_bcount, DEV_BSIZE);
    361 	if ((bp->b_blkno + sz) > p_size) {
    362 		sz = p_size - bp->b_blkno;
    363 		if (sz == 0) {
    364 			/* If exactly at end of disk, return EOF. */
    365 			bp->b_resid = bp->b_bcount;
    366 			return 0;
    367 		}
    368 		if (sz < 0) {
    369 			/* If past end of disk, return EINVAL. */
    370 			bp->b_error = EINVAL;
    371 			return -1;
    372 		}
    373 		/* Otherwise, truncate request. */
    374 		bp->b_bcount = sz << DEV_BSHIFT;
    375 	}
    376 
    377 	/* Overwriting disk label? */
    378 	if (bp->b_blkno + p_offset <= labelsector &&
    379 	    bp->b_blkno + p_offset + sz > labelsector &&
    380 	    (bp->b_flags & B_READ) == 0 && !wlabel) {
    381 		bp->b_error = EROFS;
    382 		return -1;
    383 	}
    384 
    385 	/* calculate cylinder for disksort to order transfers with */
    386 	bp->b_cylinder = (bp->b_blkno + p->p_offset) /
    387 	    (lp->d_secsize / DEV_BSIZE) / lp->d_secpercyl;
    388 	return 1;
    389 }
    390 
    391 int
    392 disk_read_sectors(void (*strat)(struct buf *), const struct disklabel *lp,
    393     struct buf *bp, unsigned int sector, int count)
    394 {
    395 	bp->b_blkno = sector;
    396 	bp->b_bcount = count * lp->d_secsize;
    397 	bp->b_flags = (bp->b_flags & ~B_WRITE) | B_READ;
    398 	bp->b_oflags &= ~BO_DONE;
    399 	bp->b_cylinder = sector / lp->d_secpercyl;
    400 	(*strat)(bp);
    401 	return biowait(bp);
    402 }
    403 
    404 const char *
    405 convertdisklabel(struct disklabel *lp, void (*strat)(struct buf *),
    406     struct buf *bp, uint32_t secperunit)
    407 {
    408 	struct partition rp, *altp, *p;
    409 	int geom_ok;
    410 
    411 	memset(&rp, 0, sizeof(rp));
    412 	rp.p_size = secperunit;
    413 	rp.p_fstype = FS_UNUSED;
    414 
    415 	/* If we can seek to d_secperunit - 1, believe the disk geometry. */
    416 	if (secperunit != 0 &&
    417 	    disk_read_sectors(strat, lp, bp, secperunit - 1, 1) == 0)
    418 		geom_ok = 1;
    419 	else
    420 		geom_ok = 0;
    421 
    422 #if 0
    423 	printf("%s: secperunit (%" PRIu32 ") %s\n", __func__,
    424 	    secperunit, geom_ok ? "ok" : "not ok");
    425 #endif
    426 
    427 	p = &lp->d_partitions[RAW_PART];
    428 	if (RAW_PART == 'c' - 'a')
    429 		altp = &lp->d_partitions['d' - 'a'];
    430 	else
    431 		altp = &lp->d_partitions['c' - 'a'];
    432 
    433 	if (lp->d_npartitions > RAW_PART && p->p_offset == 0 && p->p_size != 0)
    434 		;	/* already a raw partition */
    435 	else if (lp->d_npartitions > MAX('c', 'd') - 'a' &&
    436 		 altp->p_offset == 0 && altp->p_size != 0) {
    437 		/* alternate partition ('c' or 'd') is suitable for raw slot,
    438 		 * swap with 'd' or 'c'.
    439 		 */
    440 		rp = *p;
    441 		*p = *altp;
    442 		*altp = rp;
    443 	} else if (lp->d_npartitions <= RAW_PART &&
    444 	           lp->d_npartitions > 'c' - 'a') {
    445 		/* No raw partition is present, but the alternate is present.
    446 		 * Copy alternate to raw partition.
    447 		 */
    448 		lp->d_npartitions = RAW_PART + 1;
    449 		*p = *altp;
    450 	} else if (!geom_ok)
    451 		return "no raw partition and disk reports bad geometry";
    452 	else if (lp->d_npartitions <= RAW_PART) {
    453 		memset(&lp->d_partitions[lp->d_npartitions], 0,
    454 		    sizeof(struct partition) * (RAW_PART - lp->d_npartitions));
    455 		*p = rp;
    456 		lp->d_npartitions = RAW_PART + 1;
    457 	} else if (lp->d_npartitions < MAXPARTITIONS) {
    458 		memmove(p + 1, p,
    459 		    sizeof(struct partition) * (lp->d_npartitions - RAW_PART));
    460 		*p = rp;
    461 		lp->d_npartitions++;
    462 	} else
    463 		return "no raw partition and partition table is full";
    464 	return NULL;
    465 }
    466 
    467 /*
    468  * disk_ioctl --
    469  *	Generic disk ioctl handling.
    470  */
    471 int
    472 disk_ioctl(struct disk *diskp, u_long cmd, void *data, int flag,
    473 	   struct lwp *l)
    474 {
    475 	int error;
    476 
    477 	switch (cmd) {
    478 	case DIOCGDISKINFO:
    479 	    {
    480 		struct plistref *pref = (struct plistref *) data;
    481 
    482 		if (diskp->dk_info == NULL)
    483 			error = ENOTSUP;
    484 		else
    485 			error = prop_dictionary_copyout_ioctl(pref, cmd,
    486 							diskp->dk_info);
    487 		break;
    488 	    }
    489 
    490 	default:
    491 		error = EPASSTHROUGH;
    492 	}
    493 
    494 	return (error);
    495 }
    496