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

/*	kern_physio.c	6.2	83/09/09	*/

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

#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/vm.h"
#include "../h/trace.h"
#include "../h/map.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;

/*
 * 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, (char *)0);
		}
		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;
{
	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 struct iovec *iov = uio->uio_iov;
	register int c;
	char *a;
	int s, error = 0;

nextiov:
	if (uio->uio_iovcnt == 0)
		return (0);
	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
13849ba3	1	/* kern_physio.c 6.2 83/09/09 */
961945a8 SL	2
961945a8 SL	3	#include "../machine/pte.h"
663dbc72 BJ	4
	5	#include "../h/param.h"
	6	#include "../h/systm.h"
	7	#include "../h/dir.h"
	8	#include "../h/user.h"
	9	#include "../h/buf.h"
	10	#include "../h/conf.h"
	11	#include "../h/proc.h"
	12	#include "../h/seg.h"
663dbc72	13	#include "../h/vm.h"
973ecc4f	14	#include "../h/trace.h"
d668d9ba	15	#include "../h/map.h"
d6d7360b	16	#include "../h/uio.h"
663dbc72	17
663dbc72 BJ	18	/*
	19	* Swap IO headers -
	20	* They contain the necessary information for the swap I/O.
	21	* At any given time, a swap header can be in three
	22	* different lists. When free it is in the free list,
	23	* when allocated and the I/O queued, it is on the swap
	24	* device list, and finally, if the operation was a dirty
	25	* page push, when the I/O completes, it is inserted
	26	* in a list of cleaned pages to be processed by the pageout daemon.
	27	*/
4c05b581	28	struct buf *swbuf;
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;
39d536e6	46	int nbytes, rdflg, flag;
663dbc72	47	dev_t dev;
39d536e6	48	u_int pfcent;
663dbc72 BJ	49	{
663dbc72 BJ	50	register struct buf *bp;
e438ed8e	51	register u_int c;
663dbc72 BJ	52	int p2dp;
663dbc72 BJ	53	register struct pte dpte, vpte;
530d0032	54	int s;
d668d9ba	55	extern swdone();
663dbc72	56
530d0032	57	s = spl6();
663dbc72 BJ	58	while (bswlist.av_forw == NULL) {
	59	bswlist.b_flags \|= B_WANTED;
	60	sleep((caddr_t)&bswlist, PSWP+1);
	61	}
	62	bp = bswlist.av_forw;
	63	bswlist.av_forw = bp->av_forw;
530d0032	64	splx(s);
663dbc72 BJ	65
	66	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
	67	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	68	if (rdflg == B_READ)
	69	sum.v_pswpin += btoc(nbytes);
	70	else
	71	sum.v_pswpout += btoc(nbytes);
	72	bp->b_proc = p;
	73	if (flag & B_DIRTY) {
	74	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	75	dpte = dptopte(&proc[2], p2dp);
	76	vpte = vtopte(p, btop(addr));
	77	for (c = 0; c < nbytes; c += NBPG) {
	78	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	79	panic("swap bad pte");
	80	dpte++ = vpte++;
	81	}
d668d9ba SL	82	bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], p2dp));
	83	bp->b_flags \|= B_CALL;
	84	bp->b_iodone = swdone;
	85	bp->b_pfcent = pfcent;
663dbc72 BJ	86	} else
	87	bp->b_un.b_addr = addr;
	88	while (nbytes > 0) {
e438ed8e BJ	89	bp->b_bcount = nbytes;
	90	minphys(bp);
	91	c = bp->b_bcount;
663dbc72 BJ	92	bp->b_blkno = dblkno;
663dbc72 BJ	93	bp->b_dev = dev;
53f9ca20 BJ	94	#ifdef TRACE
	95	trace(TR_SWAPIO, dev, bp->b_blkno);
	96	#endif
ca1f746a	97	physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP);
663dbc72 BJ	98	if (flag & B_DIRTY) {
	99	if (c < nbytes)
	100	panic("big push");
663dbc72 BJ	101	return;
663dbc72 BJ	102	}
663dbc72 BJ	103	bp->b_un.b_addr += c;
	104	bp->b_flags &= ~B_DONE;
	105	if (bp->b_flags & B_ERROR) {
	106	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	107	panic("hard IO err in swap");
	108	swkill(p, (char *)0);
	109	}
	110	nbytes -= c;
919fe934	111	dblkno += btodb(c);
663dbc72	112	}
530d0032	113	s = spl6();
663dbc72 BJ	114	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	115	bp->av_forw = bswlist.av_forw;
	116	bswlist.av_forw = bp;
	117	if (bswlist.b_flags & B_WANTED) {
	118	bswlist.b_flags &= ~B_WANTED;
	119	wakeup((caddr_t)&bswlist);
	120	wakeup((caddr_t)&proc[2]);
	121	}
530d0032	122	splx(s);
663dbc72 BJ	123	}
663dbc72 BJ	124
d668d9ba SL	125	/*
	126	* Put a buffer on the clean list after I/O is done.
	127	* Called from biodone.
	128	*/
	129	swdone(bp)
	130	register struct buf *bp;
	131	{
	132	register int s;
	133
	134	if (bp->b_flags & B_ERROR)
	135	panic("IO err in push");
	136	s = spl6();
	137	bp->av_forw = bclnlist;
	138	cnt.v_pgout++;
	139	cnt.v_pgpgout += bp->b_bcount / NBPG;
	140	bclnlist = bp;
	141	if (bswlist.b_flags & B_WANTED)
	142	wakeup((caddr_t)&proc[2]);
	143	splx(s);
	144	}
	145
663dbc72 BJ	146	/*
	147	* If rout == 0 then killed on swap error, else
	148	* rout is the name of the routine where we ran out of
	149	* swap space.
	150	*/
	151	swkill(p, rout)
	152	struct proc *p;
	153	char *rout;
	154	{
444f631c	155	char *mesg;
663dbc72	156
444f631c	157	printf("pid %d: ", p->p_pid);
663dbc72	158	if (rout)
444f631c	159	printf(mesg = "killed due to no swap space\n");
663dbc72	160	else
444f631c BJ	161	printf(mesg = "killed on swap error\n");
444f631c BJ	162	uprintf("sorry, pid %d was %s", p->p_pid, mesg);
663dbc72 BJ	163	/*
	164	* To be sure no looping (e.g. in vmsched trying to
	165	* swap out) mark process locked in core (as though
	166	* done by user) after killing it so noone will try
	167	* to swap it out.
	168	*/
a30d2e97	169	psignal(p, SIGKILL);
663dbc72 BJ	170	p->p_flag \|= SULOCK;
	171	}
	172
663dbc72 BJ	173	/*
	174	* Raw I/O. The arguments are
	175	* The strategy routine for the device
	176	* A buffer, which will always be a special buffer
	177	* header owned exclusively by the device for this purpose
	178	* The device number
	179	* Read/write flag
	180	* Essentially all the work is computing physical addresses and
	181	* validating them.
	182	* If the user has the proper access privilidges, the process is
	183	* marked 'delayed unlock' and the pages involved in the I/O are
	184	* faulted and locked. After the completion of the I/O, the above pages
	185	* are unlocked.
	186	*/
d6d7360b BJ	187	physio(strat, bp, dev, rw, mincnt, uio)
	188	int (*strat)();
	189	register struct buf *bp;
	190	dev_t dev;
	191	int rw;
	192	unsigned (*mincnt)();
	193	struct uio *uio;
663dbc72	194	{
406ddcbe	195	register struct iovec *iov = uio->uio_iov;
663dbc72 BJ	196	register int c;
663dbc72 BJ	197	char *a;
d6d7360b	198	int s, error = 0;
663dbc72	199
d6d7360b	200	nextiov:
406ddcbe	201	if (uio->uio_iovcnt == 0)
d6d7360b	202	return (0);
406ddcbe	203	if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL)
d6d7360b	204	return (EFAULT);
530d0032	205	s = spl6();
663dbc72 BJ	206	while (bp->b_flags&B_BUSY) {
	207	bp->b_flags \|= B_WANTED;
	208	sleep((caddr_t)bp, PRIBIO+1);
	209	}
ef3b3d5a	210	splx(s);
663dbc72 BJ	211	bp->b_error = 0;
663dbc72 BJ	212	bp->b_proc = u.u_procp;
d6d7360b BJ	213	bp->b_un.b_addr = iov->iov_base;
d6d7360b BJ	214	while (iov->iov_len > 0) {
663dbc72 BJ	215	bp->b_flags = B_BUSY \| B_PHYS \| rw;
663dbc72 BJ	216	bp->b_dev = dev;
919fe934	217	bp->b_blkno = btodb(uio->uio_offset);
d6d7360b	218	bp->b_bcount = iov->iov_len;
663dbc72 BJ	219	(*mincnt)(bp);
	220	c = bp->b_bcount;
	221	u.u_procp->p_flag \|= SPHYSIO;
	222	vslock(a = bp->b_un.b_addr, c);
e438ed8e	223	physstrat(bp, strat, PRIBIO);
81263dba	224	(void) spl6();
663dbc72 BJ	225	vsunlock(a, c, rw);
	226	u.u_procp->p_flag &= ~SPHYSIO;
	227	if (bp->b_flags&B_WANTED)
	228	wakeup((caddr_t)bp);
530d0032	229	splx(s);
d6d7360b	230	c -= bp->b_resid;
663dbc72	231	bp->b_un.b_addr += c;
d6d7360b BJ	232	iov->iov_len -= c;
	233	uio->uio_resid -= c;
	234	uio->uio_offset += c;
961945a8	235	/* temp kludge for tape drives */
35a494b8	236	if (bp->b_resid \|\| (bp->b_flags&B_ERROR))
52a593fa	237	break;
663dbc72 BJ	238	}
663dbc72 BJ	239	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
d6d7360b	240	error = geterror(bp);
961945a8 SL	241	/* temp kludge for tape drives */
961945a8 SL	242	if (bp->b_resid \|\| error)
d6d7360b	243	return (error);
d6d7360b BJ	244	uio->uio_iov++;
	245	uio->uio_iovcnt--;
	246	goto nextiov;
663dbc72 BJ	247	}
663dbc72 BJ	248
35a494b8 SL	249	#define MAXPHYS (63 * 1024)
35a494b8 SL	250
663dbc72 BJ	251	unsigned
663dbc72 BJ	252	minphys(bp)
d6d7360b	253	struct buf *bp;
663dbc72 BJ	254	{
663dbc72 BJ	255
35a494b8 SL	256	if (bp->b_bcount > MAXPHYS)
35a494b8 SL	257	bp->b_bcount = MAXPHYS;
663dbc72	258	}