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

/*	kern_physio.c	4.31	82/08/13	*/

#include "../h/param.h"
#include "../h/systm.h"
#include "../h/dir.h"
#include "../h/user.h"
#include "../h/buf.h"
#include "../h/conf.h"
#include "../h/proc.h"
#include "../h/seg.h"
#include "../h/pte.h"
#include "../h/vm.h"
#include "../h/trace.h"
#include "../h/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;
short	*swsize;		/* CAN WE JUST USE B_BCOUNT? */
int	*swpf;

/*
 * 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 flag, nbytes;
	dev_t dev;
	unsigned pfcent;
{
	register struct buf *bp;
	register int c;
	int p2dp;
	register struct pte *dpte, *vpte;
	int s;

	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(p2dp);
	} else
		bp->b_un.b_addr = addr;
	while (nbytes > 0) {
		c = imin(ctob(120), nbytes);
		bp->b_bcount = c;
		bp->b_blkno = dblkno;
		bp->b_dev = dev;
		if (flag & B_DIRTY) {
			swpf[bp - swbuf] = pfcent;
			swsize[bp - swbuf] = nbytes;
		}
#ifdef TRACE
		trace(TR_SWAPIO, dev, bp->b_blkno);
#endif
		(*bdevsw[major(dev)].d_strategy)(bp);
		if (flag & B_DIRTY) {
			if (c < nbytes)
				panic("big push");
			return;
		}
		s = spl6();
		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, (char *)0);
		}
		nbytes -= c;
		dblkno += btoc(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);
}

/*
 * 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;
{
	char *mesg;

	printf("pid %d: ", p->p_pid);
	if (rout)
		printf(mesg = "killed due to no swap space\n");
	else
		printf(mesg = "killed on swap error\n");
	uprintf("sorry, pid %d was %s", p->p_pid, mesg);
	/*
	 * 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 int c;
	struct uio auio;
	register struct iovec *iov;
	struct iovec aiov;
	char *a;
	int s, error = 0;

	if (uio == 0) {
		uio = &auio;
		uio->uio_iov = &aiov;
		uio->uio_iovcnt = 1;
		uio->uio_offset = u.u_offset;
		uio->uio_segflg = u.u_segflg;
		iov = &aiov;
		iov->iov_base = u.u_base;
		iov->iov_len = u.u_count;
		uio->uio_resid = iov->iov_len;
	} else
		iov = uio->uio_iov;
nextiov:
	if (uio->uio_iovcnt == 0) {
		u.u_count = uio->uio_resid;
		return (0);
	}
	if (useracc(iov->iov_base,iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL) {
		u.u_count = uio->uio_resid;
		u.u_error = EFAULT;
		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 = uio->uio_offset >> PGSHIFT;
		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);
		(void) spl6();
		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)
			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;
		if (bp->b_flags&B_ERROR)
			break;
	}
	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
	error = geterror(bp);
	if (error) {
		u.u_error = error;
		u.u_count = uio->uio_resid;
		return (error);
	}
	uio->uio_iov++;
	uio->uio_iovcnt--;
	goto nextiov;
}

/*ARGSUSED*/
unsigned
minphys(bp)
	struct buf *bp;
{

	if (bp->b_bcount > 63 * 1024)
		bp->b_bcount = 63 * 1024;
}
Commit	Line	Data
d6d7360b	1	/* kern_physio.c 4.31 82/08/13 */
663dbc72 BJ	2
	3	#include "../h/param.h"
	4	#include "../h/systm.h"
	5	#include "../h/dir.h"
	6	#include "../h/user.h"
	7	#include "../h/buf.h"
	8	#include "../h/conf.h"
	9	#include "../h/proc.h"
	10	#include "../h/seg.h"
	11	#include "../h/pte.h"
	12	#include "../h/vm.h"
973ecc4f	13	#include "../h/trace.h"
d6d7360b	14	#include "../h/uio.h"
663dbc72	15
663dbc72 BJ	16	/*
	17	* Swap IO headers -
	18	* They contain the necessary information for the swap I/O.
	19	* At any given time, a swap header can be in three
	20	* different lists. When free it is in the free list,
	21	* when allocated and the I/O queued, it is on the swap
	22	* device list, and finally, if the operation was a dirty
	23	* page push, when the I/O completes, it is inserted
	24	* in a list of cleaned pages to be processed by the pageout daemon.
	25	*/
4c05b581 BJ	26	struct buf *swbuf;
	27	short swsize; / CAN WE JUST USE B_BCOUNT? */
	28	int *swpf;
663dbc72	29
663dbc72 BJ	30	/*
	31	* swap I/O -
	32	*
	33	* If the flag indicates a dirty page push initiated
	34	* by the pageout daemon, we map the page into the i th
	35	* virtual page of process 2 (the daemon itself) where i is
	36	* the index of the swap header that has been allocated.
	37	* We simply initialize the header and queue the I/O but
	38	* do not wait for completion. When the I/O completes,
	39	* iodone() will link the header to a list of cleaned
	40	* pages to be processed by the pageout daemon.
	41	*/
	42	swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	43	struct proc *p;
	44	swblk_t dblkno;
	45	caddr_t addr;
	46	int flag, nbytes;
	47	dev_t dev;
	48	unsigned pfcent;
	49	{
	50	register struct buf *bp;
	51	register int c;
	52	int p2dp;
	53	register struct pte dpte, vpte;
530d0032	54	int s;
663dbc72	55
530d0032	56	s = spl6();
663dbc72 BJ	57	while (bswlist.av_forw == NULL) {
	58	bswlist.b_flags \|= B_WANTED;
	59	sleep((caddr_t)&bswlist, PSWP+1);
	60	}
	61	bp = bswlist.av_forw;
	62	bswlist.av_forw = bp->av_forw;
530d0032	63	splx(s);
663dbc72 BJ	64
	65	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
	66	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	67	if (rdflg == B_READ)
	68	sum.v_pswpin += btoc(nbytes);
	69	else
	70	sum.v_pswpout += btoc(nbytes);
	71	bp->b_proc = p;
	72	if (flag & B_DIRTY) {
	73	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	74	dpte = dptopte(&proc[2], p2dp);
	75	vpte = vtopte(p, btop(addr));
	76	for (c = 0; c < nbytes; c += NBPG) {
	77	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	78	panic("swap bad pte");
	79	dpte++ = vpte++;
	80	}
	81	bp->b_un.b_addr = (caddr_t)ctob(p2dp);
	82	} else
	83	bp->b_un.b_addr = addr;
	84	while (nbytes > 0) {
	85	c = imin(ctob(120), nbytes);
	86	bp->b_bcount = c;
	87	bp->b_blkno = dblkno;
	88	bp->b_dev = dev;
d2f87136 BJ	89	if (flag & B_DIRTY) {
	90	swpf[bp - swbuf] = pfcent;
	91	swsize[bp - swbuf] = nbytes;
	92	}
53f9ca20 BJ	93	#ifdef TRACE
	94	trace(TR_SWAPIO, dev, bp->b_blkno);
	95	#endif
663dbc72 BJ	96	(*bdevsw[major(dev)].d_strategy)(bp);
	97	if (flag & B_DIRTY) {
	98	if (c < nbytes)
	99	panic("big push");
663dbc72 BJ	100	return;
663dbc72 BJ	101	}
530d0032	102	s = spl6();
663dbc72 BJ	103	while((bp->b_flags&B_DONE)==0)
663dbc72 BJ	104	sleep((caddr_t)bp, PSWP);
530d0032	105	splx(s);
663dbc72 BJ	106	bp->b_un.b_addr += c;
	107	bp->b_flags &= ~B_DONE;
	108	if (bp->b_flags & B_ERROR) {
	109	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	110	panic("hard IO err in swap");
	111	swkill(p, (char *)0);
	112	}
	113	nbytes -= c;
	114	dblkno += btoc(c);
	115	}
530d0032	116	s = spl6();
663dbc72 BJ	117	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	118	bp->av_forw = bswlist.av_forw;
	119	bswlist.av_forw = bp;
	120	if (bswlist.b_flags & B_WANTED) {
	121	bswlist.b_flags &= ~B_WANTED;
	122	wakeup((caddr_t)&bswlist);
	123	wakeup((caddr_t)&proc[2]);
	124	}
530d0032	125	splx(s);
663dbc72 BJ	126	}
	127
	128	/*
	129	* If rout == 0 then killed on swap error, else
	130	* rout is the name of the routine where we ran out of
	131	* swap space.
	132	*/
	133	swkill(p, rout)
	134	struct proc *p;
	135	char *rout;
	136	{
444f631c	137	char *mesg;
663dbc72	138
444f631c	139	printf("pid %d: ", p->p_pid);
663dbc72	140	if (rout)
444f631c	141	printf(mesg = "killed due to no swap space\n");
663dbc72	142	else
444f631c BJ	143	printf(mesg = "killed on swap error\n");
444f631c BJ	144	uprintf("sorry, pid %d was %s", p->p_pid, mesg);
663dbc72 BJ	145	/*
	146	* To be sure no looping (e.g. in vmsched trying to
	147	* swap out) mark process locked in core (as though
	148	* done by user) after killing it so noone will try
	149	* to swap it out.
	150	*/
a30d2e97	151	psignal(p, SIGKILL);
663dbc72 BJ	152	p->p_flag \|= SULOCK;
	153	}
	154
663dbc72 BJ	155	/*
	156	* Raw I/O. The arguments are
	157	* The strategy routine for the device
	158	* A buffer, which will always be a special buffer
	159	* header owned exclusively by the device for this purpose
	160	* The device number
	161	* Read/write flag
	162	* Essentially all the work is computing physical addresses and
	163	* validating them.
	164	* If the user has the proper access privilidges, the process is
	165	* marked 'delayed unlock' and the pages involved in the I/O are
	166	* faulted and locked. After the completion of the I/O, the above pages
	167	* are unlocked.
	168	*/
d6d7360b BJ	169	physio(strat, bp, dev, rw, mincnt, uio)
	170	int (*strat)();
	171	register struct buf *bp;
	172	dev_t dev;
	173	int rw;
	174	unsigned (*mincnt)();
	175	struct uio *uio;
663dbc72 BJ	176	{
663dbc72 BJ	177	register int c;
d6d7360b BJ	178	struct uio auio;
	179	register struct iovec *iov;
	180	struct iovec aiov;
663dbc72	181	char *a;
d6d7360b	182	int s, error = 0;
663dbc72	183
d6d7360b BJ	184	if (uio == 0) {
	185	uio = &auio;
	186	uio->uio_iov = &aiov;
	187	uio->uio_iovcnt = 1;
	188	uio->uio_offset = u.u_offset;
	189	uio->uio_segflg = u.u_segflg;
	190	iov = &aiov;
	191	iov->iov_base = u.u_base;
	192	iov->iov_len = u.u_count;
	193	uio->uio_resid = iov->iov_len;
	194	} else
	195	iov = uio->uio_iov;
	196	nextiov:
	197	if (uio->uio_iovcnt == 0) {
	198	u.u_count = uio->uio_resid;
	199	return (0);
	200	}
	201	if (useracc(iov->iov_base,iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL) {
	202	u.u_count = uio->uio_resid;
663dbc72	203	u.u_error = EFAULT;
d6d7360b	204	return (EFAULT);
663dbc72	205	}
530d0032	206	s = spl6();
663dbc72 BJ	207	while (bp->b_flags&B_BUSY) {
	208	bp->b_flags \|= B_WANTED;
	209	sleep((caddr_t)bp, PRIBIO+1);
	210	}
ef3b3d5a	211	splx(s);
663dbc72 BJ	212	bp->b_error = 0;
663dbc72 BJ	213	bp->b_proc = u.u_procp;
d6d7360b BJ	214	bp->b_un.b_addr = iov->iov_base;
d6d7360b BJ	215	while (iov->iov_len > 0) {
663dbc72 BJ	216	bp->b_flags = B_BUSY \| B_PHYS \| rw;
663dbc72 BJ	217	bp->b_dev = dev;
d6d7360b BJ	218	bp->b_blkno = uio->uio_offset >> PGSHIFT;
d6d7360b BJ	219	bp->b_bcount = iov->iov_len;
663dbc72 BJ	220	(*mincnt)(bp);
	221	c = bp->b_bcount;
	222	u.u_procp->p_flag \|= SPHYSIO;
	223	vslock(a = bp->b_un.b_addr, c);
	224	(*strat)(bp);
81263dba	225	(void) spl6();
663dbc72 BJ	226	while ((bp->b_flags&B_DONE) == 0)
	227	sleep((caddr_t)bp, PRIBIO);
	228	vsunlock(a, c, rw);
	229	u.u_procp->p_flag &= ~SPHYSIO;
	230	if (bp->b_flags&B_WANTED)
	231	wakeup((caddr_t)bp);
530d0032	232	splx(s);
d6d7360b	233	c -= bp->b_resid;
663dbc72	234	bp->b_un.b_addr += c;
d6d7360b BJ	235	iov->iov_len -= c;
	236	uio->uio_resid -= c;
	237	uio->uio_offset += c;
52a593fa BJ	238	if (bp->b_flags&B_ERROR)
52a593fa BJ	239	break;
663dbc72 BJ	240	}
663dbc72 BJ	241	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
d6d7360b BJ	242	error = geterror(bp);
	243	if (error) {
	244	u.u_error = error;
	245	u.u_count = uio->uio_resid;
	246	return (error);
	247	}
	248	uio->uio_iov++;
	249	uio->uio_iovcnt--;
	250	goto nextiov;
663dbc72 BJ	251	}
	252
	253	/ARGSUSED/
	254	unsigned
	255	minphys(bp)
d6d7360b	256	struct buf *bp;
663dbc72 BJ	257	{
663dbc72 BJ	258
2ec65c94 BJ	259	if (bp->b_bcount > 63 * 1024)
2ec65c94 BJ	260	bp->b_bcount = 63 * 1024;
663dbc72 BJ	261	}
663dbc72 BJ	262