usr/src/sys/kern/kern_physio.c

/*
 * Copyright (c) 1982 Regents of the University of California.
 * All rights reserved.  The Berkeley software License Agreement
 * specifies the terms and conditions for redistribution.
 *
 *	@(#)kern_physio.c	6.9 (Berkeley) %G%
 */

#include "../machine/pte.h"

#include "param.h"
#include "systm.h"
#include "dir.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 "uio.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,
 * iodone() will link the header to a list of cleaned
 * pages to be processed by the pageout daemon.
 */
swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	struct proc *p;
	swblk_t dblkno;
	caddr_t addr;
	int nbytes, rdflg, flag;
	dev_t dev;
	u_int pfcent;
{
	register struct buf *bp;
	register u_int c;
	int p2dp;
	register struct pte *dpte, *vpte;
	int s;
	extern swdone();
	int error = 0;

	s = splbio();
	while (bswlist.av_forw == NULL) {
		bswlist.b_flags |= B_WANTED;
		sleep((caddr_t)&bswlist, PSWP+1);
	}
	bp = bswlist.av_forw;
	bswlist.av_forw = bp->av_forw;
	splx(s);

	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_bcount = nbytes;
		minphys(bp);
		c = bp->b_bcount;
		bp->b_blkno = dblkno;
		bp->b_dev = dev;
#ifdef TRACE
		trace(TR_SWAPIO, dev, bp->b_blkno);
#endif
		physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
		if (flag & B_DIRTY) {
			if (c < nbytes)
				panic("big push");
			return (error);
		}
		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);
	}
	s = splbio();
	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
	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);
	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 always be a special buffer
 *	  header owned exclusively by the device for this purpose
 *	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;
	unsigned (*mincnt)();
	struct uio *uio;
{
	register struct iovec *iov;
	register int c;
	char *a;
	int s, error = 0;

nextiov:
	if (uio->uio_iovcnt == 0)
		return (0);
	iov = uio->uio_iov;
	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
		return (EFAULT);
	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 | 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);
		physstrat(bp, strat, PRIBIO);
		(void) splbio();
		vsunlock(a, c, rw);
		u.u_procp->p_flag &= ~SPHYSIO;
		if (bp->b_flags&B_WANTED)
			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);
	error = geterror(bp);
	/* temp kludge for tape drives */
	if (bp->b_resid || error)
		return (error);
	uio->uio_iov++;
	uio->uio_iovcnt--;
	goto nextiov;
}

#define	MAXPHYS	(63 * 1024)

unsigned
minphys(bp)
	struct buf *bp;
{

	if (bp->b_bcount > MAXPHYS)
		bp->b_bcount = MAXPHYS;
}
Commit	Line	Data
	1	/*
	2	* Copyright (c) 1982 Regents of the University of California.
	3	* All rights reserved. The Berkeley software License Agreement
	4	* specifies the terms and conditions for redistribution.
	5	*
	6	* @(#)kern_physio.c 6.9 (Berkeley) %G%
	7	*/
	8
	9	#include "../machine/pte.h"
	10
	11	#include "param.h"
	12	#include "systm.h"
	13	#include "dir.h"
	14	#include "user.h"
	15	#include "buf.h"
	16	#include "conf.h"
	17	#include "proc.h"
	18	#include "seg.h"
	19	#include "vm.h"
	20	#include "trace.h"
	21	#include "map.h"
	22	#include "uio.h"
	23
	24	/*
	25	* Swap IO headers -
	26	* They contain the necessary information for the swap I/O.
	27	* At any given time, a swap header can be in three
	28	* different lists. When free it is in the free list,
	29	* when allocated and the I/O queued, it is on the swap
	30	* device list, and finally, if the operation was a dirty
	31	* page push, when the I/O completes, it is inserted
	32	* in a list of cleaned pages to be processed by the pageout daemon.
	33	*/
	34	struct buf *swbuf;
	35
	36	/*
	37	* swap I/O -
	38	*
	39	* If the flag indicates a dirty page push initiated
	40	* by the pageout daemon, we map the page into the i th
	41	* virtual page of process 2 (the daemon itself) where i is
	42	* the index of the swap header that has been allocated.
	43	* We simply initialize the header and queue the I/O but
	44	* do not wait for completion. When the I/O completes,
	45	* iodone() will link the header to a list of cleaned
	46	* pages to be processed by the pageout daemon.
	47	*/
	48	swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	49	struct proc *p;
	50	swblk_t dblkno;
	51	caddr_t addr;
	52	int nbytes, rdflg, flag;
	53	dev_t dev;
	54	u_int pfcent;
	55	{
	56	register struct buf *bp;
	57	register u_int c;
	58	int p2dp;
	59	register struct pte dpte, vpte;
	60	int s;
	61	extern swdone();
	62	int error = 0;
	63
	64	s = splbio();
	65	while (bswlist.av_forw == NULL) {
	66	bswlist.b_flags \|= B_WANTED;
	67	sleep((caddr_t)&bswlist, PSWP+1);
	68	}
	69	bp = bswlist.av_forw;
	70	bswlist.av_forw = bp->av_forw;
	71	splx(s);
	72
	73	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
	74	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	75	if (rdflg == B_READ)
	76	sum.v_pswpin += btoc(nbytes);
	77	else
	78	sum.v_pswpout += btoc(nbytes);
	79	bp->b_proc = p;
	80	if (flag & B_DIRTY) {
	81	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	82	dpte = dptopte(&proc[2], p2dp);
	83	vpte = vtopte(p, btop(addr));
	84	for (c = 0; c < nbytes; c += NBPG) {
	85	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	86	panic("swap bad pte");
	87	dpte++ = vpte++;
	88	}
	89	bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
	90	bp->b_flags \|= B_CALL;
	91	bp->b_iodone = swdone;
	92	bp->b_pfcent = pfcent;
	93	} else
	94	bp->b_un.b_addr = addr;
	95	while (nbytes > 0) {
	96	bp->b_bcount = nbytes;
	97	minphys(bp);
	98	c = bp->b_bcount;
	99	bp->b_blkno = dblkno;
	100	bp->b_dev = dev;
	101	#ifdef TRACE
	102	trace(TR_SWAPIO, dev, bp->b_blkno);
	103	#endif
	104	physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
	105	if (flag & B_DIRTY) {
	106	if (c < nbytes)
	107	panic("big push");
	108	return (error);
	109	}
	110	bp->b_un.b_addr += c;
	111	bp->b_flags &= ~B_DONE;
	112	if (bp->b_flags & B_ERROR) {
	113	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	114	panic("hard IO err in swap");
	115	swkill(p, "swap: read error from swap device");
	116	error = EIO;
	117	}
	118	nbytes -= c;
	119	dblkno += btodb(c);
	120	}
	121	s = splbio();
	122	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	123	bp->av_forw = bswlist.av_forw;
	124	bswlist.av_forw = bp;
	125	if (bswlist.b_flags & B_WANTED) {
	126	bswlist.b_flags &= ~B_WANTED;
	127	wakeup((caddr_t)&bswlist);
	128	wakeup((caddr_t)&proc[2]);
	129	}
	130	splx(s);
	131	return (error);
	132	}
	133
	134	/*
	135	* Put a buffer on the clean list after I/O is done.
	136	* Called from biodone.
	137	*/
	138	swdone(bp)
	139	register struct buf *bp;
	140	{
	141	register int s;
	142
	143	if (bp->b_flags & B_ERROR)
	144	panic("IO err in push");
	145	s = splbio();
	146	bp->av_forw = bclnlist;
	147	cnt.v_pgout++;
	148	cnt.v_pgpgout += bp->b_bcount / NBPG;
	149	bclnlist = bp;
	150	if (bswlist.b_flags & B_WANTED)
	151	wakeup((caddr_t)&proc[2]);
	152	splx(s);
	153	}
	154
	155	/*
	156	* If rout == 0 then killed on swap error, else
	157	* rout is the name of the routine where we ran out of
	158	* swap space.
	159	*/
	160	swkill(p, rout)
	161	struct proc *p;
	162	char *rout;
	163	{
	164
	165	printf("pid %d: %s\n", p->p_pid, rout);
	166	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
	167	/*
	168	* To be sure no looping (e.g. in vmsched trying to
	169	* swap out) mark process locked in core (as though
	170	* done by user) after killing it so noone will try
	171	* to swap it out.
	172	*/
	173	psignal(p, SIGKILL);
	174	p->p_flag \|= SULOCK;
	175	}
	176
	177	/*
	178	* Raw I/O. The arguments are
	179	* The strategy routine for the device
	180	* A buffer, which will always be a special buffer
	181	* header owned exclusively by the device for this purpose
	182	* The device number
	183	* Read/write flag
	184	* Essentially all the work is computing physical addresses and
	185	* validating them.
	186	* If the user has the proper access privilidges, the process is
	187	* marked 'delayed unlock' and the pages involved in the I/O are
	188	* faulted and locked. After the completion of the I/O, the above pages
	189	* are unlocked.
	190	*/
	191	physio(strat, bp, dev, rw, mincnt, uio)
	192	int (*strat)();
	193	register struct buf *bp;
	194	dev_t dev;
	195	int rw;
	196	unsigned (*mincnt)();
	197	struct uio *uio;
	198	{
	199	register struct iovec *iov;
	200	register int c;
	201	char *a;
	202	int s, error = 0;
	203
	204	nextiov:
	205	if (uio->uio_iovcnt == 0)
	206	return (0);
	207	iov = uio->uio_iov;
	208	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
	209	return (EFAULT);
	210	s = splbio();
	211	while (bp->b_flags&B_BUSY) {
	212	bp->b_flags \|= B_WANTED;
	213	sleep((caddr_t)bp, PRIBIO+1);
	214	}
	215	splx(s);
	216	bp->b_error = 0;
	217	bp->b_proc = u.u_procp;
	218	bp->b_un.b_addr = iov->iov_base;
	219	while (iov->iov_len > 0) {
	220	bp->b_flags = B_BUSY \| B_PHYS \| rw;
	221	bp->b_dev = dev;
	222	bp->b_blkno = btodb(uio->uio_offset);
	223	bp->b_bcount = iov->iov_len;
	224	(*mincnt)(bp);
	225	c = bp->b_bcount;
	226	u.u_procp->p_flag \|= SPHYSIO;
	227	vslock(a = bp->b_un.b_addr, c);
	228	physstrat(bp, strat, PRIBIO);
	229	(void) splbio();
	230	vsunlock(a, c, rw);
	231	u.u_procp->p_flag &= ~SPHYSIO;
	232	if (bp->b_flags&B_WANTED)
	233	wakeup((caddr_t)bp);
	234	splx(s);
	235	c -= bp->b_resid;
	236	bp->b_un.b_addr += c;
	237	iov->iov_len -= c;
	238	uio->uio_resid -= c;
	239	uio->uio_offset += c;
	240	/* temp kludge for tape drives */
	241	if (bp->b_resid \|\| (bp->b_flags&B_ERROR))
	242	break;
	243	}
	244	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
	245	error = geterror(bp);
	246	/* temp kludge for tape drives */
	247	if (bp->b_resid \|\| error)
	248	return (error);
	249	uio->uio_iov++;
	250	uio->uio_iovcnt--;
	251	goto nextiov;
	252	}
	253
	254	#define MAXPHYS (63 * 1024)
	255
	256	unsigned
	257	minphys(bp)
	258	struct buf *bp;
	259	{
	260
	261	if (bp->b_bcount > MAXPHYS)
	262	bp->b_bcount = MAXPHYS;
	263	}