/*
 * Copyright (c) 1982, 1986 Regents of the University of California.
 * All rights reserved.  The Berkeley software License Agreement
 * specifies the terms and conditions for redistribution.
 *
 *	@(#)kern_physio.c	7.7 (Berkeley) 06/27/89
 */

#include "param.h"
#include "systm.h"
#include "user.h"
#include "buf.h"
#include "conf.h"
#include "proc.h"
#include "seg.h"
#include "vm.h"
#include "trace.h"
#include "map.h"
#include "vnode.h"

#include "machine/pte.h"

/*
 * Swap IO headers -
 * They contain the necessary information for the swap I/O.
 * At any given time, a swap header can be in three
 * different lists. When free it is in the free list, 
 * when allocated and the I/O queued, it is on the swap 
 * device list, and finally, if the operation was a dirty 
 * page push, when the I/O completes, it is inserted 
 * in a list of cleaned pages to be processed by the pageout daemon.
 */
struct	buf *swbuf;

/*
 * swap I/O -
 *
 * If the flag indicates a dirty page push initiated
 * by the pageout daemon, we map the page into the i th
 * virtual page of process 2 (the daemon itself) where i is
 * the index of the swap header that has been allocated.
 * We simply initialize the header and queue the I/O but
 * do not wait for completion. When the I/O completes,
 * biodone() will link the header to a list of cleaned
 * pages to be processed by the pageout daemon.
 */
swap(p, dblkno, addr, nbytes, rdflg, flag, vp, pfcent)
	struct proc *p;
	swblk_t dblkno;
	caddr_t addr;
	int nbytes, rdflg, flag;
	struct vnode *vp;
	u_int pfcent;
{
	register struct buf *bp;
	register struct pte *dpte, *vpte;
	register u_int c;
	int p2dp, s, error = 0;
	struct buf *getswbuf();
	int swdone();

	bp = getswbuf(PSWP+1);
	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
		if (rdflg == B_READ)
			sum.v_pswpin += btoc(nbytes);
		else
			sum.v_pswpout += btoc(nbytes);
	bp->b_proc = p;
	if (flag & B_DIRTY) {
		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
		dpte = dptopte(&proc[2], p2dp);
		vpte = vtopte(p, btop(addr));
		for (c = 0; c < nbytes; c += NBPG) {
			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
				panic("swap bad pte");
			*dpte++ = *vpte++;
		}
		bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
		bp->b_flags |= B_CALL;
		bp->b_iodone = swdone;
		bp->b_pfcent = pfcent;
	} else
		bp->b_un.b_addr = addr;
	while (nbytes > 0) {
		bp->b_blkno = dblkno;
		bp->b_dev = vp->v_rdev;
		if (bp->b_vp)
			brelvp(bp);
		VREF(vp);
		bp->b_vp = vp;
		bp->b_bcount = nbytes;
		minphys(bp);
		c = bp->b_bcount;
#ifdef TRACE
		trace(TR_SWAPIO, vp, bp->b_blkno);
#endif
		VOP_STRATEGY(bp);
		/* pageout daemon doesn't wait for pushed pages */
		if (flag & B_DIRTY) {
			if (c < nbytes)
				panic("big push");
			return (0);
		} else {
			s = splbio();
			while ((bp->b_flags & B_DONE) == 0)
				sleep((caddr_t)bp, PSWP);
			splx(s);
		}
		bp->b_un.b_addr += c;
		bp->b_flags &= ~B_DONE;
		if (bp->b_flags & B_ERROR) {
			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
				panic("hard IO err in swap");
			swkill(p, "swap: read error from swap device");
			error = EIO;
		}
		nbytes -= c;
		dblkno += btodb(c);
	}
	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
	freeswbuf(bp);
	return (error);
}

/*
 * Put a buffer on the clean list after I/O is done.
 * Called from biodone.
 */
swdone(bp)
	register struct buf *bp;
{
	register int s;

	if (bp->b_flags & B_ERROR)
		panic("IO err in push");
	s = splbio();
	bp->av_forw = bclnlist;
	cnt.v_pgout++;
	cnt.v_pgpgout += bp->b_bcount / NBPG;
	bclnlist = bp;
	if (bswlist.b_flags & B_WANTED)
		wakeup((caddr_t)&proc[2]);
	splx(s);
}

/*
 * If rout == 0 then killed on swap error, else
 * rout is the name of the routine where we ran out of
 * swap space.
 */
swkill(p, rout)
	struct proc *p;
	char *rout;
{

	printf("pid %d: %s\n", p->p_pid, rout);
	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
	/*
	 * To be sure no looping (e.g. in vmsched trying to
	 * swap out) mark process locked in core (as though
	 * done by user) after killing it so noone will try
	 * to swap it out.
	 */
	psignal(p, SIGKILL);
	p->p_flag |= SULOCK;
}

/*
 * Raw I/O. The arguments are
 *	The strategy routine for the device
 *	A buffer, which will either be a special buffer header owned
 *	    exclusively by the device for this purpose, or NULL,
 *	    indicating that we should use a swap buffer
 *	The device number
 *	Read/write flag
 * Essentially all the work is computing physical addresses and
 * validating them.
 * If the user has the proper access privilidges, the process is
 * marked 'delayed unlock' and the pages involved in the I/O are
 * faulted and locked. After the completion of the I/O, the above pages
 * are unlocked.
 */
physio(strat, bp, dev, rw, mincnt, uio)
	int (*strat)(); 
	register struct buf *bp;
	dev_t dev;
	int rw;
	u_int (*mincnt)();
	struct uio *uio;
{
	register struct iovec *iov;
	register int c;
	char *a;
	int s, allocbuf = 0, error = 0;
	struct buf *getswbuf();

	if (bp == NULL) {
		allocbuf = 1;
		bp = getswbuf(PRIBIO+1);
	}
	for (; uio->uio_iovcnt; uio->uio_iov++, uio->uio_iovcnt--) {
		iov = uio->uio_iov;
		if (!useracc(iov->iov_base, (u_int)iov->iov_len,
		    rw == B_READ ? B_WRITE : B_READ)) {
			error = EFAULT;
			break;
		}
		if (!allocbuf) {	/* only if sharing caller's buffer */
			s = splbio();
			while (bp->b_flags&B_BUSY) {
				bp->b_flags |= B_WANTED;
				sleep((caddr_t)bp, PRIBIO+1);
			}
			splx(s);
		}
		bp->b_error = 0;
		bp->b_proc = u.u_procp;
		bp->b_un.b_addr = iov->iov_base;
		while (iov->iov_len > 0) {
			bp->b_flags = B_BUSY | B_PHYS | B_RAW | rw;
			bp->b_dev = dev;
			bp->b_blkno = btodb(uio->uio_offset);
			bp->b_bcount = iov->iov_len;
			(*mincnt)(bp);
			c = bp->b_bcount;
			u.u_procp->p_flag |= SPHYSIO;
			vslock(a = bp->b_un.b_addr, c);
			(*strat)(bp);
			s = splbio();
			while ((bp->b_flags & B_DONE) == 0)
				sleep((caddr_t)bp, PRIBIO);
			vsunlock(a, c, rw);
			u.u_procp->p_flag &= ~SPHYSIO;
			if (bp->b_flags&B_WANTED)	/* rare */
				wakeup((caddr_t)bp);
			splx(s);
			c -= bp->b_resid;
			bp->b_un.b_addr += c;
			iov->iov_len -= c;
			uio->uio_resid -= c;
			uio->uio_offset += c;
			/* temp kludge for tape drives */
			if (bp->b_resid || (bp->b_flags&B_ERROR))
				break;
		}
		bp->b_flags &= ~(B_BUSY | B_WANTED | B_PHYS | B_RAW);
		error = biowait(bp);
		/* temp kludge for tape drives */
		if (bp->b_resid || error)
			break;
	}
	if (allocbuf)
		freeswbuf(bp);
	return (error);
}

u_int
minphys(bp)
	struct buf *bp;
{
	if (bp->b_bcount > MAXPHYS)
		bp->b_bcount = MAXPHYS;
}

static
struct buf *
getswbuf(prio)
	int prio;
{
	int s;
	struct buf *bp;

	s = splbio();
	while (bswlist.av_forw == NULL) {
		bswlist.b_flags |= B_WANTED;
		sleep((caddr_t)&bswlist, prio);
	}
	bp = bswlist.av_forw;
	bswlist.av_forw = bp->av_forw;
	splx(s);
	return (bp);
}

static
freeswbuf(bp)
	struct buf *bp;
{
	int s;

	s = splbio();
	bp->av_forw = bswlist.av_forw;
	bswlist.av_forw = bp;
	if (bswlist.b_flags & B_WANTED) {
		bswlist.b_flags &= ~B_WANTED;
		wakeup((caddr_t)&bswlist);
		wakeup((caddr_t)&proc[2]);
	}
	splx(s);
}

rawread(dev, uio)
	dev_t dev;
	struct uio *uio;
{
	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
	    dev, B_READ, minphys, uio));
}

rawwrite(dev, uio)
	dev_t dev;
	struct uio *uio;
{
	return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
	    dev, B_WRITE, minphys, uio));
}