xref: /csrg-svn/sys/ufs/ffs/ufs_disksubr.c (revision 34637)
123397Smckusick /*
234535Skarels  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
323397Smckusick  * All rights reserved.  The Berkeley software License Agreement
423397Smckusick  * specifies the terms and conditions for redistribution.
523397Smckusick  *
6*34637Skarels  *	@(#)ufs_disksubr.c	7.9 (Berkeley) 06/04/88
723397Smckusick  */
816Sbill 
930533Skarels #include "param.h"
1030533Skarels #include "systm.h"
1130533Skarels #include "buf.h"
1230533Skarels #include "disklabel.h"
1334535Skarels #include "syslog.h"
1430533Skarels 
1530740Skarels #include "dir.h"
1630740Skarels #include "user.h"
1730740Skarels 
1816Sbill /*
192626Swnj  * Seek sort for disks.  We depend on the driver
202626Swnj  * which calls us using b_resid as the current cylinder number.
212626Swnj  *
222626Swnj  * The argument dp structure holds a b_actf activity chain pointer
232626Swnj  * on which we keep two queues, sorted in ascending cylinder order.
242626Swnj  * The first queue holds those requests which are positioned after
252626Swnj  * the current cylinder (in the first request); the second holds
262626Swnj  * requests which came in after their cylinder number was passed.
272626Swnj  * Thus we implement a one way scan, retracting after reaching the
282626Swnj  * end of the drive to the first request on the second queue,
292626Swnj  * at which time it becomes the first queue.
302626Swnj  *
312626Swnj  * A one-way scan is natural because of the way UNIX read-ahead
322626Swnj  * blocks are allocated.
3316Sbill  */
3416Sbill 
3516Sbill #define	b_cylin	b_resid
3616Sbill 
3716Sbill disksort(dp, bp)
382626Swnj 	register struct buf *dp, *bp;
3916Sbill {
4016Sbill 	register struct buf *ap;
4116Sbill 
422626Swnj 	/*
432626Swnj 	 * If nothing on the activity queue, then
442626Swnj 	 * we become the only thing.
452626Swnj 	 */
4616Sbill 	ap = dp->b_actf;
4716Sbill 	if(ap == NULL) {
4816Sbill 		dp->b_actf = bp;
4916Sbill 		dp->b_actl = bp;
5016Sbill 		bp->av_forw = NULL;
5116Sbill 		return;
5216Sbill 	}
532626Swnj 	/*
542626Swnj 	 * If we lie after the first (currently active)
552626Swnj 	 * request, then we must locate the second request list
562626Swnj 	 * and add ourselves to it.
572626Swnj 	 */
582626Swnj 	if (bp->b_cylin < ap->b_cylin) {
592626Swnj 		while (ap->av_forw) {
602626Swnj 			/*
612626Swnj 			 * Check for an ``inversion'' in the
622626Swnj 			 * normally ascending cylinder numbers,
632626Swnj 			 * indicating the start of the second request list.
642626Swnj 			 */
652626Swnj 			if (ap->av_forw->b_cylin < ap->b_cylin) {
662626Swnj 				/*
672626Swnj 				 * Search the second request list
682626Swnj 				 * for the first request at a larger
692626Swnj 				 * cylinder number.  We go before that;
702626Swnj 				 * if there is no such request, we go at end.
712626Swnj 				 */
722626Swnj 				do {
732626Swnj 					if (bp->b_cylin < ap->av_forw->b_cylin)
742626Swnj 						goto insert;
7532573Skarels 					if (bp->b_cylin == ap->av_forw->b_cylin &&
7632573Skarels 					    bp->b_blkno < ap->av_forw->b_blkno)
7732573Skarels 						goto insert;
782626Swnj 					ap = ap->av_forw;
792626Swnj 				} while (ap->av_forw);
802626Swnj 				goto insert;		/* after last */
812626Swnj 			}
822626Swnj 			ap = ap->av_forw;
8316Sbill 		}
842626Swnj 		/*
852626Swnj 		 * No inversions... we will go after the last, and
862626Swnj 		 * be the first request in the second request list.
872626Swnj 		 */
882626Swnj 		goto insert;
8916Sbill 	}
902626Swnj 	/*
912626Swnj 	 * Request is at/after the current request...
922626Swnj 	 * sort in the first request list.
932626Swnj 	 */
942626Swnj 	while (ap->av_forw) {
952626Swnj 		/*
962626Swnj 		 * We want to go after the current request
972626Swnj 		 * if there is an inversion after it (i.e. it is
982626Swnj 		 * the end of the first request list), or if
992626Swnj 		 * the next request is a larger cylinder than our request.
1002626Swnj 		 */
1012626Swnj 		if (ap->av_forw->b_cylin < ap->b_cylin ||
10232573Skarels 		    bp->b_cylin < ap->av_forw->b_cylin ||
10332573Skarels 		    (bp->b_cylin == ap->av_forw->b_cylin &&
10432573Skarels 		    bp->b_blkno < ap->av_forw->b_blkno))
1052626Swnj 			goto insert;
1062626Swnj 		ap = ap->av_forw;
1072626Swnj 	}
1082626Swnj 	/*
1092626Swnj 	 * Neither a second list nor a larger
1102626Swnj 	 * request... we go at the end of the first list,
1112626Swnj 	 * which is the same as the end of the whole schebang.
1122626Swnj 	 */
1132626Swnj insert:
1142626Swnj 	bp->av_forw = ap->av_forw;
1152626Swnj 	ap->av_forw = bp;
1162626Swnj 	if (ap == dp->b_actl)
11716Sbill 		dp->b_actl = bp;
11816Sbill }
11930533Skarels 
12030533Skarels /*
12130740Skarels  * Attempt to read a disk label from a device
12230740Skarels  * using the indicated stategy routine.
12330740Skarels  * The label must be partly set up before this:
12430740Skarels  * secpercyl and anything required in the strategy routine
12530740Skarels  * (e.g., sector size) must be filled in before calling us.
12630740Skarels  * Returns null on success and an error string on failure.
12730740Skarels  */
12830740Skarels char *
12930740Skarels readdisklabel(dev, strat, lp)
13030740Skarels 	dev_t dev;
13130740Skarels 	int (*strat)();
13230740Skarels 	register struct disklabel *lp;
13330740Skarels {
13430740Skarels 	register struct buf *bp;
13530740Skarels 	struct disklabel *dlp;
13630740Skarels 	char *msg = NULL;
13730740Skarels 
13830740Skarels 	if (lp->d_secperunit == 0)
13930740Skarels 		lp->d_secperunit = 0x1fffffff;
14030740Skarels 	lp->d_npartitions = 1;
14130740Skarels 	if (lp->d_partitions[0].p_size == 0)
14230740Skarels 		lp->d_partitions[0].p_size = 0x1fffffff;
14330740Skarels 	lp->d_partitions[0].p_offset = 0;
14430740Skarels 
14534102Skarels 	bp = geteblk((int)lp->d_secsize);
14630740Skarels 	bp->b_dev = dev;
14730740Skarels 	bp->b_blkno = LABELSECTOR;
14832068Skarels 	bp->b_bcount = lp->d_secsize;
14930740Skarels 	bp->b_flags = B_BUSY | B_READ;
15030740Skarels 	bp->b_cylin = LABELSECTOR / lp->d_secpercyl;
15130740Skarels 	(*strat)(bp);
15230740Skarels 	biowait(bp);
15330740Skarels 	if (bp->b_flags & B_ERROR) {
15430740Skarels 		u.u_error = 0;		/* XXX */
15530740Skarels 		msg = "I/O error";
15632068Skarels 	} else for (dlp = (struct disklabel *)bp->b_un.b_addr;
15732068Skarels 	    dlp <= (struct disklabel *)(bp->b_un.b_addr+DEV_BSIZE-sizeof(*dlp));
15832068Skarels 	    dlp = (struct disklabel *)((char *)dlp + sizeof(long))) {
15932068Skarels 		if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC) {
16032068Skarels 			if (msg == NULL)
16132068Skarels 				msg = "no disk label";
16232068Skarels 		} else if (dkcksum(dlp) != 0)
16330740Skarels 			msg = "disk label corrupted";
16432068Skarels 		else {
16530740Skarels 			*lp = *dlp;
16632068Skarels 			msg = NULL;
16732068Skarels 			break;
16832068Skarels 		}
16930740Skarels 	}
17032068Skarels 	if (lp->d_npartitions > MAXPARTITIONS)
17132068Skarels 		lp->d_npartitions = MAXPARTITIONS;
17230740Skarels 	bp->b_flags = B_INVAL | B_AGE;
17330740Skarels 	brelse(bp);
17430740Skarels 	return (msg);
17530740Skarels }
17630740Skarels 
17730740Skarels /*
17832573Skarels  * Check new disk label for sensibility
17932573Skarels  * before setting it.
18032573Skarels  */
18132573Skarels setdisklabel(olp, nlp, openmask)
18232573Skarels 	register struct disklabel *olp, *nlp;
18332573Skarels 	u_long openmask;
18432573Skarels {
18532573Skarels 	register i;
18632573Skarels 	register struct partition *opp, *npp;
18732573Skarels 
18832573Skarels 	if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC ||
18932573Skarels 	    dkcksum(nlp) != 0)
19032573Skarels 		return (EINVAL);
19134473Smckusick 	while ((i = ffs((long)openmask)) != 0) {
19232573Skarels 		i--;
19332573Skarels 		openmask &= ~(1 << i);
19432573Skarels 		if (nlp->d_npartitions <= i)
19532573Skarels 			return (EBUSY);
19632573Skarels 		opp = &olp->d_partitions[i];
19732573Skarels 		npp = &nlp->d_partitions[i];
19832573Skarels 		if (npp->p_offset != opp->p_offset || npp->p_size < opp->p_size)
19932573Skarels 			return (EBUSY);
20032573Skarels 		/*
20132573Skarels 		 * Copy internally-set partition information
20232573Skarels 		 * if new label doesn't include it.		XXX
20332573Skarels 		 */
20432573Skarels 		if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) {
20532573Skarels 			npp->p_fstype = opp->p_fstype;
20632573Skarels 			npp->p_fsize = opp->p_fsize;
20732573Skarels 			npp->p_frag = opp->p_frag;
20832573Skarels 			npp->p_cpg = opp->p_cpg;
20932573Skarels 		}
21032573Skarels 	}
21134102Skarels  	nlp->d_checksum = 0;
21234102Skarels  	nlp->d_checksum = dkcksum(nlp);
21332573Skarels 	*olp = *nlp;
21432573Skarels 	return (0);
21532573Skarels }
21632573Skarels 
21732573Skarels /* encoding of disk minor numbers, should be elsewhere... */
21832573Skarels #define dkunit(dev)		(minor(dev) >> 3)
21932573Skarels #define dkpart(dev)		(minor(dev) & 07)
22032573Skarels #define dkminor(unit, part)	(((unit) << 3) | (part))
22132573Skarels 
22232573Skarels /*
22332573Skarels  * Write disk label back to device after modification.
22432573Skarels  */
22532573Skarels writedisklabel(dev, strat, lp)
22632573Skarels 	dev_t dev;
22732573Skarels 	int (*strat)();
22832573Skarels 	register struct disklabel *lp;
22932573Skarels {
23032573Skarels 	struct buf *bp;
23132573Skarels 	struct disklabel *dlp;
23232573Skarels 	int labelpart;
23332573Skarels 	int error = 0;
23432573Skarels 
23532573Skarels 	labelpart = dkpart(dev);
23632573Skarels 	if (lp->d_partitions[labelpart].p_offset != 0) {
23732573Skarels 		if (lp->d_partitions[0].p_offset != 0)
23832573Skarels 			return (EXDEV);			/* not quite right */
23932573Skarels 		labelpart = 0;
24032573Skarels 	}
24134102Skarels 	bp = geteblk((int)lp->d_secsize);
24232573Skarels 	bp->b_dev = makedev(major(dev), dkminor(dkunit(dev), labelpart));
24332573Skarels 	bp->b_blkno = LABELSECTOR;
24432573Skarels 	bp->b_bcount = lp->d_secsize;
24532573Skarels 	bp->b_flags = B_READ;
24632573Skarels 	(*strat)(bp);
24732573Skarels 	biowait(bp);
24832573Skarels 	if (bp->b_flags & B_ERROR) {
24932573Skarels 		error = u.u_error;		/* XXX */
25032573Skarels 		u.u_error = 0;
25134102Skarels 		goto done;
25232573Skarels 	}
25334102Skarels 	for (dlp = (struct disklabel *)bp->b_un.b_addr;
25434102Skarels 	    dlp <= (struct disklabel *)
25534102Skarels 	      (bp->b_un.b_addr + lp->d_secsize - sizeof(*dlp));
25634102Skarels 	    dlp = (struct disklabel *)((char *)dlp + sizeof(long))) {
25734102Skarels 		if (dlp->d_magic == DISKMAGIC && dlp->d_magic2 == DISKMAGIC &&
25834102Skarels 		    dkcksum(dlp) == 0) {
25934102Skarels 			*dlp = *lp;
26034102Skarels 			bp->b_flags = B_WRITE;
26134102Skarels 			(*strat)(bp);
26234102Skarels 			biowait(bp);
26334102Skarels 			if (bp->b_flags & B_ERROR) {
26434102Skarels 				error = u.u_error;		/* XXX */
26534102Skarels 				u.u_error = 0;
26634102Skarels 			}
26734102Skarels 			goto done;
26834102Skarels 		}
26932573Skarels 	}
27034102Skarels 	error = ESRCH;
27134102Skarels done:
27232573Skarels 	brelse(bp);
27332573Skarels 	return (error);
27432573Skarels }
27532573Skarels 
27632573Skarels /*
27730533Skarels  * Compute checksum for disk label.
27830533Skarels  */
27930533Skarels dkcksum(lp)
28030533Skarels 	register struct disklabel *lp;
28130533Skarels {
28230533Skarels 	register u_short *start, *end;
28330533Skarels 	register u_short sum = 0;
28430533Skarels 
28530533Skarels 	start = (u_short *)lp;
28630533Skarels 	end = (u_short *)&lp->d_partitions[lp->d_npartitions];
28730533Skarels 	while (start < end)
28830533Skarels 		sum ^= *start++;
28930533Skarels 	return (sum);
29030533Skarels }
29134535Skarels 
29234535Skarels /*
29334535Skarels  * Disk error is the preface to plaintive error messages
29434535Skarels  * about failing disk transfers.  It prints messages of the form
295*34637Skarels 
296*34637Skarels hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
297*34637Skarels 
29834535Skarels  * if the offset of the error in the transfer and a disk label
29934535Skarels  * are both available.  blkdone should be -1 if the position of the error
30034535Skarels  * is unknown; the disklabel pointer may be null from drivers that have not
30134535Skarels  * been converted to use them.  The message is printed with printf
30234535Skarels  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
30334535Skarels  * The message should be completed (with at least a newline) with printf
30434535Skarels  * or addlog, respectively.  There is no trailing space.
30534535Skarels  */
30634535Skarels diskerr(bp, dname, what, pri, blkdone, lp)
30734535Skarels 	register struct buf *bp;
30834535Skarels 	char *dname, *what;
30934535Skarels 	int pri, blkdone;
31034535Skarels 	register struct disklabel *lp;
31134535Skarels {
31234535Skarels 	int unit = dkunit(bp->b_dev), part = dkpart(bp->b_dev);
31334535Skarels 	register int (*pr)(), sn;
31434535Skarels 	char partname = 'a' + part;
31534535Skarels 	extern printf(), addlog();
31634535Skarels 
31734535Skarels 	if (pri != LOG_PRINTF) {
31834535Skarels 		log(pri, "");
31934535Skarels 		pr = addlog;
32034535Skarels 	} else
32134535Skarels 		pr = printf;
32234535Skarels 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
32334535Skarels 	    bp->b_flags & B_READ ? "read" : "writ");
32434535Skarels 	sn = bp->b_blkno;
32534535Skarels 	if (bp->b_bcount <= DEV_BSIZE)
32634535Skarels 		(*pr)("%d", sn);
32734535Skarels 	else {
32834535Skarels 		if (blkdone >= 0) {
32934535Skarels 			sn += blkdone;
33034535Skarels 			(*pr)("%d of ", sn);
33134535Skarels 		}
33234535Skarels 		(*pr)("%d-%d", bp->b_blkno,
33334535Skarels 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
33434535Skarels 	}
33534535Skarels 	if (lp && (blkdone >= 0 || bp->b_bcount <= DEV_BSIZE)) {
33634535Skarels 		sn += lp->d_partitions[part].p_offset;
337*34637Skarels 		(*pr)(" (%s%d bn %d; cn %d", dname, unit, sn,
338*34637Skarels 		    sn / lp->d_secpercyl);
339*34637Skarels 		sn %= lp->d_secpercyl;
340*34637Skarels 		(*pr)(" tn %d sn %d)", sn / lp->d_ntracks, sn % lp->d_ntracks);
34134535Skarels 	}
34234535Skarels }
343