[unix-history] / usr / src / sys / kern / kern_physio.c

/*
 * 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	7.3 (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"
#ifdef SECSIZE
#include "file.h"
#include "ioctl.h"
#include "disklabel.h"
#endif SECSIZE

/*
 * 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, 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;
#ifdef SECSIZE
	bp->b_blksize = DEV_BSIZE;
#endif SECSIZE
	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 (0);
		}
		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;
#ifdef SECSIZE
		if (flag & B_PGIN && nbytes > 0)
			panic("big pgin");
#endif SECSIZE
		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;
#ifdef SECSIZE
	int bsize;
	struct partinfo dpart;
#endif SECSIZE

#ifdef SECSIZE
	if ((unsigned)major(dev) < nchrdev &&
	    (*cdevsw[major(dev)].d_ioctl)(dev, DIOCGPART, (caddr_t)&dpart,
	    FREAD) == 0)
		bsize = dpart.disklab->d_secsize;
	else
		bsize = DEV_BSIZE;
#endif SECSIZE
	for (;;) {
		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;
#ifdef SECSIZE
		bp->b_blksize = bsize;
#endif SECSIZE
		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;
#ifdef SECSIZE
			bp->b_blkno = uio->uio_offset / bsize;
#else SECSIZE
			bp->b_blkno = btodb(uio->uio_offset);
#endif SECSIZE
			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--;
	}
}

unsigned
minphys(bp)
	struct buf *bp;
{

	if (bp->b_bcount > MAXPHYS)
		bp->b_bcount = MAXPHYS;
}
Commit	Line	Data
da7c5cc6	1	/*
0880b18e	2	* Copyright (c) 1982, 1986 Regents of the University of California.
da7c5cc6 KM	3	* All rights reserved. The Berkeley software License Agreement
	4	* specifies the terms and conditions for redistribution.
	5	*
609e7cfa	6	* @(#)kern_physio.c 7.3 (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"
ec67a3ce MK	23	#ifdef SECSIZE
	24	#include "file.h"
	25	#include "ioctl.h"
	26	#include "disklabel.h"
	27	#endif SECSIZE
663dbc72	28
663dbc72 BJ	29	/*
	30	* Swap IO headers -
	31	* They contain the necessary information for the swap I/O.
	32	* At any given time, a swap header can be in three
	33	* different lists. When free it is in the free list,
	34	* when allocated and the I/O queued, it is on the swap
	35	* device list, and finally, if the operation was a dirty
	36	* page push, when the I/O completes, it is inserted
	37	* in a list of cleaned pages to be processed by the pageout daemon.
	38	*/
4c05b581	39	struct buf *swbuf;
663dbc72	40
663dbc72 BJ	41	/*
	42	* swap I/O -
	43	*
	44	* If the flag indicates a dirty page push initiated
	45	* by the pageout daemon, we map the page into the i th
	46	* virtual page of process 2 (the daemon itself) where i is
	47	* the index of the swap header that has been allocated.
	48	* We simply initialize the header and queue the I/O but
	49	* do not wait for completion. When the I/O completes,
ec67a3ce	50	* biodone() will link the header to a list of cleaned
663dbc72 BJ	51	* pages to be processed by the pageout daemon.
	52	*/
	53	swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	54	struct proc *p;
	55	swblk_t dblkno;
	56	caddr_t addr;
39d536e6	57	int nbytes, rdflg, flag;
663dbc72	58	dev_t dev;
39d536e6	59	u_int pfcent;
663dbc72 BJ	60	{
663dbc72 BJ	61	register struct buf *bp;
e438ed8e	62	register u_int c;
663dbc72 BJ	63	int p2dp;
663dbc72 BJ	64	register struct pte dpte, vpte;
530d0032	65	int s;
d668d9ba	66	extern swdone();
699e2902	67	int error = 0;
663dbc72	68
d95fc990	69	s = splbio();
663dbc72 BJ	70	while (bswlist.av_forw == NULL) {
	71	bswlist.b_flags \|= B_WANTED;
	72	sleep((caddr_t)&bswlist, PSWP+1);
	73	}
	74	bp = bswlist.av_forw;
	75	bswlist.av_forw = bp->av_forw;
530d0032	76	splx(s);
663dbc72 BJ	77
663dbc72 BJ	78	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
ec67a3ce MK	79	#ifdef SECSIZE
	80	bp->b_blksize = DEV_BSIZE;
	81	#endif SECSIZE
663dbc72 BJ	82	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	83	if (rdflg == B_READ)
	84	sum.v_pswpin += btoc(nbytes);
	85	else
	86	sum.v_pswpout += btoc(nbytes);
	87	bp->b_proc = p;
	88	if (flag & B_DIRTY) {
	89	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	90	dpte = dptopte(&proc[2], p2dp);
	91	vpte = vtopte(p, btop(addr));
	92	for (c = 0; c < nbytes; c += NBPG) {
	93	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	94	panic("swap bad pte");
	95	dpte++ = vpte++;
	96	}
d668d9ba SL	97	bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
	98	bp->b_flags \|= B_CALL;
	99	bp->b_iodone = swdone;
	100	bp->b_pfcent = pfcent;
663dbc72 BJ	101	} else
	102	bp->b_un.b_addr = addr;
	103	while (nbytes > 0) {
e438ed8e BJ	104	bp->b_bcount = nbytes;
	105	minphys(bp);
	106	c = bp->b_bcount;
663dbc72 BJ	107	bp->b_blkno = dblkno;
663dbc72 BJ	108	bp->b_dev = dev;
53f9ca20 BJ	109	#ifdef TRACE
	110	trace(TR_SWAPIO, dev, bp->b_blkno);
	111	#endif
ca1f746a	112	physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
663dbc72 BJ	113	if (flag & B_DIRTY) {
	114	if (c < nbytes)
	115	panic("big push");
ec67a3ce	116	return (0);
663dbc72	117	}
663dbc72 BJ	118	bp->b_un.b_addr += c;
	119	bp->b_flags &= ~B_DONE;
	120	if (bp->b_flags & B_ERROR) {
	121	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	122	panic("hard IO err in swap");
d03b3d84	123	swkill(p, "swap: read error from swap device");
699e2902	124	error = EIO;
663dbc72 BJ	125	}
663dbc72 BJ	126	nbytes -= c;
ec67a3ce MK	127	#ifdef SECSIZE
	128	if (flag & B_PGIN && nbytes > 0)
	129	panic("big pgin");
	130	#endif SECSIZE
919fe934	131	dblkno += btodb(c);
663dbc72	132	}
d95fc990	133	s = splbio();
663dbc72 BJ	134	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	135	bp->av_forw = bswlist.av_forw;
	136	bswlist.av_forw = bp;
	137	if (bswlist.b_flags & B_WANTED) {
	138	bswlist.b_flags &= ~B_WANTED;
	139	wakeup((caddr_t)&bswlist);
	140	wakeup((caddr_t)&proc[2]);
	141	}
530d0032	142	splx(s);
699e2902	143	return (error);
663dbc72 BJ	144	}
663dbc72 BJ	145
d668d9ba SL	146	/*
	147	* Put a buffer on the clean list after I/O is done.
	148	* Called from biodone.
	149	*/
	150	swdone(bp)
	151	register struct buf *bp;
	152	{
	153	register int s;
	154
	155	if (bp->b_flags & B_ERROR)
	156	panic("IO err in push");
d95fc990	157	s = splbio();
d668d9ba SL	158	bp->av_forw = bclnlist;
	159	cnt.v_pgout++;
	160	cnt.v_pgpgout += bp->b_bcount / NBPG;
	161	bclnlist = bp;
	162	if (bswlist.b_flags & B_WANTED)
	163	wakeup((caddr_t)&proc[2]);
	164	splx(s);
	165	}
	166
663dbc72 BJ	167	/*
	168	* If rout == 0 then killed on swap error, else
	169	* rout is the name of the routine where we ran out of
	170	* swap space.
	171	*/
	172	swkill(p, rout)
	173	struct proc *p;
	174	char *rout;
	175	{
	176
7cd10076 JB	177	printf("pid %d: %s\n", p->p_pid, rout);
7cd10076 JB	178	uprintf("sorry, pid %d was killed in %s\n", p->p_pid, rout);
663dbc72 BJ	179	/*
	180	* To be sure no looping (e.g. in vmsched trying to
	181	* swap out) mark process locked in core (as though
	182	* done by user) after killing it so noone will try
	183	* to swap it out.
	184	*/
a30d2e97	185	psignal(p, SIGKILL);
663dbc72 BJ	186	p->p_flag \|= SULOCK;
	187	}
	188
663dbc72 BJ	189	/*
	190	* Raw I/O. The arguments are
	191	* The strategy routine for the device
	192	* A buffer, which will always be a special buffer
	193	* header owned exclusively by the device for this purpose
	194	* The device number
	195	* Read/write flag
	196	* Essentially all the work is computing physical addresses and
	197	* validating them.
	198	* If the user has the proper access privilidges, the process is
	199	* marked 'delayed unlock' and the pages involved in the I/O are
	200	* faulted and locked. After the completion of the I/O, the above pages
	201	* are unlocked.
	202	*/
d6d7360b BJ	203	physio(strat, bp, dev, rw, mincnt, uio)
	204	int (*strat)();
	205	register struct buf *bp;
	206	dev_t dev;
	207	int rw;
	208	unsigned (*mincnt)();
	209	struct uio *uio;
663dbc72	210	{
a196746e	211	register struct iovec *iov;
663dbc72 BJ	212	register int c;
663dbc72 BJ	213	char *a;
d6d7360b	214	int s, error = 0;
ec67a3ce MK	215	#ifdef SECSIZE
	216	int bsize;
	217	struct partinfo dpart;
	218	#endif SECSIZE
663dbc72	219
ec67a3ce MK	220	#ifdef SECSIZE
	221	if ((unsigned)major(dev) < nchrdev &&
	222	(*cdevsw[major(dev)].d_ioctl)(dev, DIOCGPART, (caddr_t)&dpart,
	223	FREAD) == 0)
	224	bsize = dpart.disklab->d_secsize;
	225	else
	226	bsize = DEV_BSIZE;
	227	#endif SECSIZE
	228	for (;;) {
	229	if (uio->uio_iovcnt == 0)
	230	return (0);
	231	iov = uio->uio_iov;
	232	if (useracc(iov->iov_base, (u_int)iov->iov_len,
	233	rw==B_READ? B_WRITE : B_READ) == NULL)
	234	return (EFAULT);
	235	s = splbio();
	236	while (bp->b_flags&B_BUSY) {
	237	bp->b_flags \|= B_WANTED;
	238	sleep((caddr_t)bp, PRIBIO+1);
	239	}
530d0032	240	splx(s);
ec67a3ce MK	241	bp->b_error = 0;
	242	bp->b_proc = u.u_procp;
	243	#ifdef SECSIZE
	244	bp->b_blksize = bsize;
	245	#endif SECSIZE
	246	bp->b_un.b_addr = iov->iov_base;
	247	while (iov->iov_len > 0) {
	248	bp->b_flags = B_BUSY \| B_PHYS \| rw;
	249	bp->b_dev = dev;
	250	#ifdef SECSIZE
	251	bp->b_blkno = uio->uio_offset / bsize;
	252	#else SECSIZE
	253	bp->b_blkno = btodb(uio->uio_offset);
	254	#endif SECSIZE
	255	bp->b_bcount = iov->iov_len;
	256	(*mincnt)(bp);
	257	c = bp->b_bcount;
	258	u.u_procp->p_flag \|= SPHYSIO;
	259	vslock(a = bp->b_un.b_addr, c);
	260	physstrat(bp, strat, PRIBIO);
	261	(void) splbio();
	262	vsunlock(a, c, rw);
	263	u.u_procp->p_flag &= ~SPHYSIO;
	264	if (bp->b_flags&B_WANTED)
	265	wakeup((caddr_t)bp);
	266	splx(s);
	267	c -= bp->b_resid;
	268	bp->b_un.b_addr += c;
	269	iov->iov_len -= c;
	270	uio->uio_resid -= c;
	271	uio->uio_offset += c;
	272	/* temp kludge for tape drives */
	273	if (bp->b_resid \|\| (bp->b_flags&B_ERROR))
	274	break;
	275	}
	276	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
	277	error = geterror(bp);
961945a8	278	/* temp kludge for tape drives */
ec67a3ce MK	279	if (bp->b_resid \|\| error)
	280	return (error);
	281	uio->uio_iov++;
	282	uio->uio_iovcnt--;
663dbc72	283	}
663dbc72 BJ	284	}
663dbc72 BJ	285
663dbc72 BJ	286	unsigned
663dbc72 BJ	287	minphys(bp)
d6d7360b	288	struct buf *bp;
663dbc72 BJ	289	{
663dbc72 BJ	290
35a494b8 SL	291	if (bp->b_bcount > MAXPHYS)
35a494b8 SL	292	bp->b_bcount = MAXPHYS;
663dbc72	293	}