[unix-history] / 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	6.7 (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();

	s = spl6();
	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;
		}
		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");
		}
		nbytes -= c;
		dblkno += btodb(c);
	}
	s = spl6();
	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);
}

/*
 * 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 = spl6();
	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 = spl6();
	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) spl6();
		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
da7c5cc6 KM	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	*
7cd10076	6	* @(#)kern_physio.c 6.7 (Berkeley) %G%
da7c5cc6	7	*/
961945a8 SL	8
961945a8 SL	9	#include "../machine/pte.h"
663dbc72	10
94368568 JB	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"
663dbc72	23
663dbc72 BJ	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	*/
4c05b581	34	struct buf *swbuf;
663dbc72	35
663dbc72 BJ	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;
39d536e6	52	int nbytes, rdflg, flag;
663dbc72	53	dev_t dev;
39d536e6	54	u_int pfcent;
663dbc72 BJ	55	{
663dbc72 BJ	56	register struct buf *bp;
e438ed8e	57	register u_int c;
663dbc72 BJ	58	int p2dp;
663dbc72 BJ	59	register struct pte dpte, vpte;
530d0032	60	int s;
d668d9ba	61	extern swdone();
663dbc72	62
530d0032	63	s = spl6();
663dbc72 BJ	64	while (bswlist.av_forw == NULL) {
	65	bswlist.b_flags \|= B_WANTED;
	66	sleep((caddr_t)&bswlist, PSWP+1);
	67	}
	68	bp = bswlist.av_forw;
	69	bswlist.av_forw = bp->av_forw;
530d0032	70	splx(s);
663dbc72 BJ	71
	72	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
	73	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	74	if (rdflg == B_READ)
	75	sum.v_pswpin += btoc(nbytes);
	76	else
	77	sum.v_pswpout += btoc(nbytes);
	78	bp->b_proc = p;
	79	if (flag & B_DIRTY) {
	80	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	81	dpte = dptopte(&proc[2], p2dp);
	82	vpte = vtopte(p, btop(addr));
	83	for (c = 0; c < nbytes; c += NBPG) {
	84	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	85	panic("swap bad pte");
	86	dpte++ = vpte++;
	87	}
d668d9ba SL	88	bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
	89	bp->b_flags \|= B_CALL;
	90	bp->b_iodone = swdone;
	91	bp->b_pfcent = pfcent;
663dbc72 BJ	92	} else
	93	bp->b_un.b_addr = addr;
	94	while (nbytes > 0) {
e438ed8e BJ	95	bp->b_bcount = nbytes;
	96	minphys(bp);
	97	c = bp->b_bcount;
663dbc72 BJ	98	bp->b_blkno = dblkno;
663dbc72 BJ	99	bp->b_dev = dev;
53f9ca20 BJ	100	#ifdef TRACE
	101	trace(TR_SWAPIO, dev, bp->b_blkno);
	102	#endif
ca1f746a	103	physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
663dbc72 BJ	104	if (flag & B_DIRTY) {
	105	if (c < nbytes)
	106	panic("big push");
663dbc72 BJ	107	return;
663dbc72 BJ	108	}
663dbc72 BJ	109	bp->b_un.b_addr += c;
	110	bp->b_flags &= ~B_DONE;
	111	if (bp->b_flags & B_ERROR) {
	112	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	113	panic("hard IO err in swap");
d03b3d84	114	swkill(p, "swap: read error from swap device");
663dbc72 BJ	115	}
663dbc72 BJ	116	nbytes -= c;
919fe934	117	dblkno += btodb(c);
663dbc72	118	}
530d0032	119	s = spl6();
663dbc72 BJ	120	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	121	bp->av_forw = bswlist.av_forw;
	122	bswlist.av_forw = bp;
	123	if (bswlist.b_flags & B_WANTED) {
	124	bswlist.b_flags &= ~B_WANTED;
	125	wakeup((caddr_t)&bswlist);
	126	wakeup((caddr_t)&proc[2]);
	127	}
530d0032	128	splx(s);
663dbc72 BJ	129	}
663dbc72 BJ	130
d668d9ba SL	131	/*
	132	* Put a buffer on the clean list after I/O is done.
	133	* Called from biodone.
	134	*/
	135	swdone(bp)
	136	register struct buf *bp;
	137	{
	138	register int s;
	139
	140	if (bp->b_flags & B_ERROR)
	141	panic("IO err in push");
	142	s = spl6();
	143	bp->av_forw = bclnlist;
	144	cnt.v_pgout++;
	145	cnt.v_pgpgout += bp->b_bcount / NBPG;
	146	bclnlist = bp;
	147	if (bswlist.b_flags & B_WANTED)
	148	wakeup((caddr_t)&proc[2]);
	149	splx(s);
	150	}
	151
663dbc72 BJ	152	/*
	153	* If rout == 0 then killed on swap error, else
	154	* rout is the name of the routine where we ran out of
	155	* swap space.
	156	*/
	157	swkill(p, rout)
	158	struct proc *p;
	159	char *rout;
	160	{
	161
7cd10076 JB	162	printf("pid %d: %s\n", p->p_pid, rout);
7cd10076 JB	163	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
663dbc72 BJ	164	/*
	165	* To be sure no looping (e.g. in vmsched trying to
	166	* swap out) mark process locked in core (as though
	167	* done by user) after killing it so noone will try
	168	* to swap it out.
	169	*/
a30d2e97	170	psignal(p, SIGKILL);
663dbc72 BJ	171	p->p_flag \|= SULOCK;
	172	}
	173
663dbc72 BJ	174	/*
	175	* Raw I/O. The arguments are
	176	* The strategy routine for the device
	177	* A buffer, which will always be a special buffer
	178	* header owned exclusively by the device for this purpose
	179	* The device number
	180	* Read/write flag
	181	* Essentially all the work is computing physical addresses and
	182	* validating them.
	183	* If the user has the proper access privilidges, the process is
	184	* marked 'delayed unlock' and the pages involved in the I/O are
	185	* faulted and locked. After the completion of the I/O, the above pages
	186	* are unlocked.
	187	*/
d6d7360b BJ	188	physio(strat, bp, dev, rw, mincnt, uio)
	189	int (*strat)();
	190	register struct buf *bp;
	191	dev_t dev;
	192	int rw;
	193	unsigned (*mincnt)();
	194	struct uio *uio;
663dbc72	195	{
a196746e	196	register struct iovec *iov;
663dbc72 BJ	197	register int c;
663dbc72 BJ	198	char *a;
d6d7360b	199	int s, error = 0;
663dbc72	200
d6d7360b	201	nextiov:
406ddcbe	202	if (uio->uio_iovcnt == 0)
d6d7360b	203	return (0);
a196746e	204	iov = uio->uio_iov;
406ddcbe	205	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
d6d7360b	206	return (EFAULT);
530d0032	207	s = spl6();
663dbc72 BJ	208	while (bp->b_flags&B_BUSY) {
	209	bp->b_flags \|= B_WANTED;
	210	sleep((caddr_t)bp, PRIBIO+1);
	211	}
ef3b3d5a	212	splx(s);
663dbc72 BJ	213	bp->b_error = 0;
663dbc72 BJ	214	bp->b_proc = u.u_procp;
d6d7360b BJ	215	bp->b_un.b_addr = iov->iov_base;
d6d7360b BJ	216	while (iov->iov_len > 0) {
663dbc72 BJ	217	bp->b_flags = B_BUSY \| B_PHYS \| rw;
663dbc72 BJ	218	bp->b_dev = dev;
919fe934	219	bp->b_blkno = btodb(uio->uio_offset);
d6d7360b	220	bp->b_bcount = iov->iov_len;
663dbc72 BJ	221	(*mincnt)(bp);
	222	c = bp->b_bcount;
	223	u.u_procp->p_flag \|= SPHYSIO;
	224	vslock(a = bp->b_un.b_addr, c);
e438ed8e	225	physstrat(bp, strat, PRIBIO);
81263dba	226	(void) spl6();
663dbc72 BJ	227	vsunlock(a, c, rw);
	228	u.u_procp->p_flag &= ~SPHYSIO;
	229	if (bp->b_flags&B_WANTED)
	230	wakeup((caddr_t)bp);
530d0032	231	splx(s);
d6d7360b	232	c -= bp->b_resid;
663dbc72	233	bp->b_un.b_addr += c;
d6d7360b BJ	234	iov->iov_len -= c;
	235	uio->uio_resid -= c;
	236	uio->uio_offset += c;
961945a8	237	/* temp kludge for tape drives */
35a494b8	238	if (bp->b_resid \|\| (bp->b_flags&B_ERROR))
52a593fa	239	break;
663dbc72 BJ	240	}
663dbc72 BJ	241	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
d6d7360b	242	error = geterror(bp);
961945a8 SL	243	/* temp kludge for tape drives */
961945a8 SL	244	if (bp->b_resid \|\| error)
d6d7360b	245	return (error);
d6d7360b BJ	246	uio->uio_iov++;
	247	uio->uio_iovcnt--;
	248	goto nextiov;
663dbc72 BJ	249	}
663dbc72 BJ	250
35a494b8 SL	251	#define MAXPHYS (63 * 1024)
35a494b8 SL	252
663dbc72 BJ	253	unsigned
663dbc72 BJ	254	minphys(bp)
d6d7360b	255	struct buf *bp;
663dbc72 BJ	256	{
663dbc72 BJ	257
35a494b8 SL	258	if (bp->b_bcount > MAXPHYS)
35a494b8 SL	259	bp->b_bcount = MAXPHYS;
663dbc72	260	}