ccd.c

基于组件方式开发操作系统的OSKIT源代码

	/*
	 * Allocate an interleave table.
	 * Chances are this is too big, but we don't care.
	 */
	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF, M_WAITOK);
	bzero((caddr_t)cs->sc_itable, size);

	/*
	 * Trivial case: no interleave (actually interleave of disk size).
	 * Each table entry represents a single component in its entirety.
	 */
	if (cs->sc_ileave == 0) {
		bn = 0;
		ii = cs->sc_itable;

		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
			/* Allocate space for ii_index. */
			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
			ii->ii_ndisk = 1;
			ii->ii_startblk = bn;
			ii->ii_startoff = 0;
			ii->ii_index[0] = ix;
			bn += cs->sc_cinfo[ix].ci_size;
			ii++;
		}
		ii->ii_ndisk = 0;
#ifdef DEBUG
		if (ccddebug & CCDB_INIT)
			printiinfo(cs->sc_itable);
#endif
		return;
	}

	/*
	 * The following isn't fast or pretty; it doesn't have to be.
	 */
	size = 0;
	bn = lbn = 0;
	for (ii = cs->sc_itable; ; ii++) {
		/* Allocate space for ii_index. */
		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
		    M_DEVBUF, M_WAITOK);

		/*
		 * Locate the smallest of the remaining components
		 */
		smallci = NULL;
		for (ci = cs->sc_cinfo;
		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
			if (ci->ci_size > size &&
			    (smallci == NULL ||
			     ci->ci_size < smallci->ci_size))
				smallci = ci;

		/*
		 * Nobody left, all done
		 */
		if (smallci == NULL) {
			ii->ii_ndisk = 0;
			break;
		}

		/*
		 * Record starting logical block and component offset
		 */
		ii->ii_startblk = bn / cs->sc_ileave;
		ii->ii_startoff = lbn;

		/*
		 * Determine how many disks take part in this interleave
		 * and record their indices.
		 */
		ix = 0;
		for (ci = cs->sc_cinfo;
		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
			if (ci->ci_size >= smallci->ci_size)
				ii->ii_index[ix++] = ci - cs->sc_cinfo;
		ii->ii_ndisk = ix;
		bn += ix * (smallci->ci_size - size);
		lbn = smallci->ci_size / cs->sc_ileave;
		size = smallci->ci_size;
	}
#ifdef DEBUG
	if (ccddebug & CCDB_INIT)
		printiinfo(cs->sc_itable);
#endif
}

/* ARGSUSED */
static int
ccdopen(dev, flags, fmt, p)
	dev_t dev;
	int flags, fmt;
	struct proc *p;
{
	int unit = ccdunit(dev);
	struct ccd_softc *cs;
	struct disklabel *lp;
	int error = 0, part, pmask;

#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdopen(%x, %x)\n", dev, flags);
#endif
	if (unit >= numccd)
		return (ENXIO);
	cs = &ccd_softc[unit];

	if (error = ccdlock(cs))
		return (error);

	lp = &cs->sc_dkdev.dk_label;

	part = ccdpart(dev);
	pmask = (1 << part);

	/*
	 * If we're initialized, check to see if there are any other
	 * open partitions.  If not, then it's safe to update
	 * the in-core disklabel.
	 */
	if ((cs->sc_flags & CCDF_INITED) && (cs->sc_dkdev.dk_openmask == 0))
		ccdgetdisklabel(dev);

	/* Check that the partition exists. */
	if (part != RAW_PART && ((part >= lp->d_npartitions) ||
	    (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
		error = ENXIO;
		goto done;
	}

	/* Prevent our unit from being unconfigured while open. */
	switch (fmt) {
	case S_IFCHR:
		cs->sc_dkdev.dk_copenmask |= pmask;
		break;

	case S_IFBLK:
		cs->sc_dkdev.dk_bopenmask |= pmask;
		break;
	}
	cs->sc_dkdev.dk_openmask =
	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;

 done:
	ccdunlock(cs);
	return (0);
}

/* ARGSUSED */
static int
ccdclose(dev, flags, fmt, p)
	dev_t dev;
	int flags, fmt;
	struct proc *p;
{
	int unit = ccdunit(dev);
	struct ccd_softc *cs;
	int error = 0, part;

#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdclose(%x, %x)\n", dev, flags);
#endif

	if (unit >= numccd)
		return (ENXIO);
	cs = &ccd_softc[unit];

	if (error = ccdlock(cs))
		return (error);

	part = ccdpart(dev);

	/* ...that much closer to allowing unconfiguration... */
	switch (fmt) {
	case S_IFCHR:
		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
		break;

	case S_IFBLK:
		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
		break;
	}
	cs->sc_dkdev.dk_openmask =
	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;

	ccdunlock(cs);
	return (0);
}

static int
ccdread(dev_t dev, struct uio *uio, int ioflag)
{
	return (physio(ccdstrategy, NULL, dev, 1, minphys, uio));
}

static int
ccdwrite(dev_t dev, struct uio *uio, int ioflag)
{
	return (physio(ccdstrategy, NULL, dev, 0, minphys, uio));
}

static void
ccdstrategy(bp)
	register struct buf *bp;
{
	register int unit = ccdunit(bp->b_dev);
	register struct ccd_softc *cs = &ccd_softc[unit];
	register int s;
	int wlabel;
	struct disklabel *lp;

#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdstrategy(%x): unit %d\n", bp, unit);
#endif
	if ((cs->sc_flags & CCDF_INITED) == 0) {
		bp->b_error = ENXIO;
		bp->b_flags |= B_ERROR;
		goto done;
	}

	/* If it's a nil transfer, wake up the top half now. */
	if (bp->b_bcount == 0)
		goto done;

	lp = &cs->sc_dkdev.dk_label;

	/*
	 * Do bounds checking and adjust transfer.  If there's an
	 * error, the bounds check will flag that for us.
	 */
	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
	if (ccdpart(bp->b_dev) != RAW_PART)
		if (bounds_check_with_label(bp, lp, wlabel) <= 0)
			goto done;

	bp->b_resid = bp->b_bcount;

	/*
	 * "Start" the unit.
	 */
	s = splbio();
	ccdstart(cs, bp);
	splx(s);
	return;
done:
	biodone(bp);
}

static void
ccdstart(cs, bp)
	register struct ccd_softc *cs;
	register struct buf *bp;
{
	register long bcount, rcount;
	struct ccdbuf *cbp[4];
	/* XXX! : 2 reads and 2 writes for RAID 4/5 */
	caddr_t addr;
	daddr_t bn;
	struct partition *pp;

#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdstart(%x, %x)\n", cs, bp);
#endif

	/* Record the transaction start  */
	devstat_start_transaction(&cs->device_stats);

	/*
	 * Translate the partition-relative block number to an absolute.
	 */
	bn = bp->b_blkno;
	if (ccdpart(bp->b_dev) != RAW_PART) {
		pp = &cs->sc_dkdev.dk_label.d_partitions[ccdpart(bp->b_dev)];
		bn += pp->p_offset;
	}

	/*
	 * Allocate component buffers and fire off the requests
	 */
	addr = bp->b_data;
	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
		ccdbuffer(cbp, cs, bp, bn, addr, bcount);
		rcount = cbp[0]->cb_buf.b_bcount;
		if ((cbp[0]->cb_buf.b_flags & B_READ) == 0)
			cbp[0]->cb_buf.b_vp->v_numoutput++;
		VOP_STRATEGY(cbp[0]->cb_buf.b_vp, &cbp[0]->cb_buf);
		if (cs->sc_cflags & CCDF_MIRROR &&
		    (cbp[0]->cb_buf.b_flags & B_READ) == 0) {
			/* mirror, start another write */
			cbp[1]->cb_buf.b_vp->v_numoutput++;
			VOP_STRATEGY(cbp[1]->cb_buf.b_vp, &cbp[1]->cb_buf);
		}
		bn += btodb(rcount);
		addr += rcount;
	}
}

/*
 * Build a component buffer header.
 */
static void
ccdbuffer(cb, cs, bp, bn, addr, bcount)
	register struct ccdbuf **cb;
	register struct ccd_softc *cs;
	struct buf *bp;
	daddr_t bn;
	caddr_t addr;
	long bcount;
{
	register struct ccdcinfo *ci, *ci2 = NULL;	/* XXX */
	register struct ccdbuf *cbp;
	register daddr_t cbn, cboff;
      register off_t cbc;

#ifdef DEBUG
	if (ccddebug & CCDB_IO)
		printf("ccdbuffer(%x, %x, %d, %x, %d)\n",
		       cs, bp, bn, addr, bcount);
#endif
	/*
	 * Determine which component bn falls in.
	 */
	cbn = bn;
	cboff = 0;

	/*
	 * Serially concatenated
	 */
	if (cs->sc_ileave == 0) {
		register daddr_t sblk;

		sblk = 0;
		for (ci = cs->sc_cinfo; cbn >= sblk + ci->ci_size; ci++)
			sblk += ci->ci_size;
		cbn -= sblk;
	}
	/*
	 * Interleaved
	 */
	else {
		register struct ccdiinfo *ii;
		int ccdisk, off;

		cboff = cbn % cs->sc_ileave;
		cbn /= cs->sc_ileave;
		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
			if (ii->ii_startblk > cbn)
				break;
		ii--;
		off = cbn - ii->ii_startblk;
		if (ii->ii_ndisk == 1) {
			ccdisk = ii->ii_index[0];
			cbn = ii->ii_startoff + off;
		} else {
			if (cs->sc_cflags & CCDF_MIRROR) {
				ccdisk = ii->ii_index[off % (ii->ii_ndisk/2)];
				cbn = ii->ii_startoff + off / (ii->ii_ndisk/2);
				/* mirrored data */
				ci2 = &cs->sc_cinfo[ccdisk + ii->ii_ndisk/2];
			}
			else if (cs->sc_cflags & CCDF_PARITY) {
				ccdisk = ii->ii_index[off % (ii->ii_ndisk-1)];
				cbn = ii->ii_startoff + off / (ii->ii_ndisk-1);
				if (cbn % ii->ii_ndisk <= ccdisk)
					ccdisk++;
			}
			else {
				ccdisk = ii->ii_index[off % ii->ii_ndisk];
				cbn = ii->ii_startoff + off / ii->ii_ndisk;
			}
		}
		cbn *= cs->sc_ileave;
		ci = &cs->sc_cinfo[ccdisk];
	}

	/*
	 * Fill in the component buf structure.
	 */
	cbp = getccdbuf();
	bzero(cbp, sizeof (struct ccdbuf));
	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
	cbp->cb_buf.b_iodone = (void (*)(struct buf *))ccdiodone;
	cbp->cb_buf.b_proc = bp->b_proc;
	cbp->cb_buf.b_dev = ci->ci_dev;		/* XXX */
	cbp->cb_buf.b_blkno = cbn + cboff + CCD_OFFSET;
	cbp->cb_buf.b_offset = dbtob(cbn + cboff + CCD_OFFSET);
	cbp->cb_buf.b_data = addr;
	cbp->cb_buf.b_vp = ci->ci_vp;
	LIST_INIT(&cbp->cb_buf.b_dep);
	cbp->cb_buf.b_resid = 0;
	if (cs->sc_ileave == 0)
              cbc = dbtob((off_t)(ci->ci_size - cbn));
	else
              cbc = dbtob((off_t)(cs->sc_ileave - cboff));
      cbp->cb_buf.b_bcount = (cbc < bcount) ? cbc : bcount;

 	cbp->cb_buf.b_bufsize = cbp->cb_buf.b_bcount;

	/*
	 * context for ccdiodone
	 */
	cbp->cb_obp = bp;
	cbp->cb_unit = cs - ccd_softc;
	cbp->cb_comp = ci - cs->sc_cinfo;

#ifdef DEBUG
	if (ccddebug & CCDB_IO)
		printf(" dev %x(u%d): cbp %x bn %d addr %x bcnt %d\n",
		       ci->ci_dev, ci-cs->sc_cinfo, cbp, cbp->cb_buf.b_blkno,
		       cbp->cb_buf.b_data, cbp->cb_buf.b_bcount);
#endif
	cb[0] = cbp;
	if (cs->sc_cflags & CCDF_MIRROR &&
	    (cbp->cb_buf.b_flags & B_READ) == 0) {
		/* mirror, start one more write */
		cbp = getccdbuf();
		bzero(cbp, sizeof (struct ccdbuf));
		*cbp = *cb[0];
		cbp->cb_buf.b_dev = ci2->ci_dev;
		cbp->cb_buf.b_vp = ci2->ci_vp;
		LIST_INIT(&cbp->cb_buf.b_dep);
		cbp->cb_comp = ci2 - cs->sc_cinfo;
		cb[1] = cbp;
		/* link together the ccdbuf's and clear "mirror done" flag */
		cb[0]->cb_mirror = cb[1];
		cb[1]->cb_mirror = cb[0];
		cb[0]->cb_pflags &= ~CCDPF_MIRROR_DONE;
		cb[1]->cb_pflags &= ~CCDPF_MIRROR_DONE;
	}
}

static void
ccdintr(cs, bp)
	register struct ccd_softc *cs;
	register struct buf *bp;
{
#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdintr(%x, %x)\n", cs, bp);
#endif
	/*
	 * Request is done for better or worse, wakeup the top half.
	 */
	/* Record device statistics */
	devstat_end_transaction(&cs->device_stats,
				bp->b_bcount - bp->b_resid,
				(bp->b_flags & B_ORDERED) ?
				DEVSTAT_TAG_ORDERED : DEVSTAT_TAG_SIMPLE,
				(bp->b_flags & B_READ) ? DEVSTAT_READ :
				DEVSTAT_WRITE);

	if (bp->b_flags & B_ERROR)
		bp->b_resid = bp->b_bcount;
	biodone(bp);
}

/*
 * Called at interrupt time.
 * Mark the component as done and if all components are done,
 * take a ccd interrupt.
 */
static void
ccdiodone(cbp)
	struct ccdbuf *cbp;
{
	register struct buf *bp = cbp->cb_obp;
	register int unit = cbp->cb_unit;
	int count, s;

	s = splbio();
#ifdef DEBUG
	if (ccddebug & CCDB_FOLLOW)
		printf("ccdiodone(%x)\n", cbp);
	if (ccddebug & CCDB_IO) {
		printf("ccdiodone: bp %x bcount %d resid %d\n",
		       bp, bp->b_bcount, bp->b_resid);
		printf(" dev %x(u%d), cbp %x bn %d addr %x bcnt %d\n",
		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
		       cbp->cb_buf.b_bcount);
	}
#endif

	if (cbp->cb_buf.b_flags & B_ERROR) {
		bp->b_flags |= B_ERROR;
		bp->b_error = cbp->cb_buf.b_error ? cbp->cb_buf.b_error : EIO;
#ifdef DEBUG
		printf("ccd%d: error %d on component %d\n",
		       unit, bp->b_error, cbp->cb_comp);
#endif
	}

	if (ccd_softc[unit].sc_cflags & CCDF_MIRROR &&
	    (cbp->cb_buf.b_flags & B_READ) == 0)
		if ((cbp->cb_pflags & CCDPF_MIRROR_DONE) == 0) {
			/* I'm done before my counterpart, so just set
			   partner's flag and return */
			cbp->cb_mirror->cb_pflags |= CCDPF_MIRROR_DONE;
			putccdbuf(cbp);
			splx(s);
			return;
		}
		
	count = cbp->cb_buf.b_bcount;
	putccdbuf(cbp);