-/* vfs_bio.c 3.2 %H% */
+/* vfs_bio.c 4.14 %G% */
-int distrust = 1; /* TEST */
#include "../h/param.h"
#include "../h/systm.h"
#include "../h/dir.h"
#include "../h/seg.h"
#include "../h/pte.h"
#include "../h/vm.h"
+#include "../h/trace.h"
/*
* The following several routines allocate and free
*/
#define BUFHSZ 63
-#define BUFHASH(blkno) (blkno % BUFHSZ)
-short bufhash[BUFHSZ];
+struct bufhd bufhash[BUFHSZ];
+#define BUFHASH(dev, dblkno) \
+ ((struct buf *)&bufhash[((int)(dev)+(int)(dblkno)) % BUFHSZ])
/*
* Initialize hash links for buffers.
bhinit()
{
register int i;
+ register struct bufhd *bp;
- for (i = 0; i < BUFHSZ; i++)
- bufhash[i] = -1;
+ for (bp = bufhash, i = 0; i < BUFHSZ; i++, bp++)
+ bp->b_forw = bp->b_back = (struct buf *)bp;
}
/* #define DISKMON 1 */
long nreada;
long ncache;
long nwrite;
- long bufcount[NBUF];
+ long bufcount[64];
} io_info;
#endif
* 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[NSWBUF];
-short swsize[NSWBUF]; /* CAN WE JUST USE B_BCOUNT? */
-int swpf[NSWBUF];
+struct buf *swbuf;
+short *swsize; /* CAN WE JUST USE B_BCOUNT? */
+int *swpf;
-#ifdef FASTVAX
+#ifndef UNFAST
#define notavail(bp) \
{ \
int s = spl6(); \
bp = getblk(dev, blkno);
if (bp->b_flags&B_DONE) {
+#ifdef EPAWNJ
+ trace(TR_BREAD|TR_HIT, dev, blkno);
+#endif
#ifdef DISKMON
io_info.ncache++;
#endif
bp->b_flags |= B_READ;
bp->b_bcount = BSIZE;
(*bdevsw[major(dev)].d_strategy)(bp);
+#ifdef EPAWNJ
+ trace(TR_BREAD|TR_MISS, dev, blkno);
+#endif
#ifdef DISKMON
io_info.nread++;
#endif
bp->b_flags |= B_READ;
bp->b_bcount = BSIZE;
(*bdevsw[major(dev)].d_strategy)(bp);
+#ifdef EPAWNJ
+ trace(TR_BREAD|TR_MISS, dev, blkno);
+#endif
#ifdef DISKMON
io_info.nread++;
#endif
u.u_vm.vm_inblk++; /* pay for read */
}
+#ifdef EPAWNJ
+ else
+ trace(TR_BREAD|TR_HIT, dev, blkno);
+#endif
}
if (rablkno && !incore(dev, rablkno)) {
rabp = getblk(dev, rablkno);
- if (rabp->b_flags & B_DONE)
+ if (rabp->b_flags & B_DONE) {
brelse(rabp);
- else {
+#ifdef EPAWNJ
+ trace(TR_BREAD|TR_HIT|TR_RA, dev, blkno);
+#endif
+ } else {
rabp->b_flags |= B_READ|B_ASYNC;
rabp->b_bcount = BSIZE;
(*bdevsw[major(dev)].d_strategy)(rabp);
+#ifdef EPAWNJ
+ trace(TR_BREAD|TR_MISS|TR_RA, dev, rablock);
+#endif
#ifdef DISKMON
io_info.nreada++;
#endif
#endif
if ((flag&B_DELWRI) == 0)
u.u_vm.vm_oublk++; /* noone paid yet */
+#ifdef EPAWNJ
+ trace(TR_BWRITE, bp->b_dev, dbtofsb(bp->b_blkno));
+#endif
(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
if ((flag&B_ASYNC) == 0) {
iowait(bp);
bdwrite(bp)
register struct buf *bp;
{
- register struct buf *dp;
+ register int flags;
if ((bp->b_flags&B_DELWRI) == 0)
u.u_vm.vm_oublk++; /* noone paid yet */
- dp = bdevsw[major(bp->b_dev)].d_tab;
- if(dp->b_flags & B_TAPE)
+ flags = bdevsw[major(bp->b_dev)].d_flags;
+ if(flags & B_TAPE)
bawrite(bp);
else {
bp->b_flags |= B_DELWRI | B_DONE;
brelse(bp)
register struct buf *bp;
{
- register struct buf **backp;
+ register struct buf *flist;
register s;
if (bp->b_flags&B_WANTED)
wakeup((caddr_t)bp);
- if (bfreelist.b_flags&B_WANTED) {
- bfreelist.b_flags &= ~B_WANTED;
- wakeup((caddr_t)&bfreelist);
- }
- if ((bp->b_flags&B_ERROR) && bp->b_dev != NODEV) {
- bunhash(bp);
- bp->b_dev = NODEV; /* no assoc. on error */
+ if (bfreelist[0].b_flags&B_WANTED) {
+ bfreelist[0].b_flags &= ~B_WANTED;
+ wakeup((caddr_t)bfreelist);
}
+ if (bp->b_flags&B_ERROR)
+ if (bp->b_flags & B_LOCKED)
+ bp->b_flags &= ~B_ERROR; /* try again later */
+ else
+ bp->b_dev = NODEV; /* no assoc */
s = spl6();
- if(bp->b_flags & (B_AGE|B_ERROR)) {
- backp = &bfreelist.av_forw;
- (*backp)->av_back = bp;
- bp->av_forw = *backp;
- *backp = bp;
- bp->av_back = &bfreelist;
+ if (bp->b_flags & (B_ERROR|B_INVAL)) {
+ /* block has no info ... put at front of most free list */
+ flist = &bfreelist[BQUEUES-1];
+ flist->av_forw->av_back = bp;
+ bp->av_forw = flist->av_forw;
+ flist->av_forw = bp;
+ bp->av_back = flist;
} else {
- backp = &bfreelist.av_back;
- (*backp)->av_forw = bp;
- bp->av_back = *backp;
- *backp = bp;
- bp->av_forw = &bfreelist;
+ if (bp->b_flags & B_LOCKED)
+ flist = &bfreelist[BQ_LOCKED];
+ else if (bp->b_flags & B_AGE)
+ flist = &bfreelist[BQ_AGE];
+ else
+ flist = &bfreelist[BQ_LRU];
+ flist->av_back->av_forw = bp;
+ bp->av_back = flist->av_back;
+ flist->av_back = bp;
+ bp->av_forw = flist;
}
bp->b_flags &= ~(B_WANTED|B_BUSY|B_ASYNC|B_AGE);
splx(s);
}
-/* HASHING IS A GUN LIKE CHANGE, THIS IS THE SAFETY */
-struct buf *
-oincore(dev, blkno)
- dev_t dev;
- daddr_t blkno;
-{
- register struct buf *bp;
- register struct buf *dp;
- register int dblkno = fsbtodb(blkno);
-
- dp = bdevsw[major(dev)].d_tab;
- for (bp=dp->b_forw; bp != dp; bp = bp->b_forw)
- if (bp->b_blkno==dblkno && bp->b_dev==dev &&
- bp >= buf && bp < &buf[NBUF])
- return (bp);
- return ((struct buf *)0);
-}
-
/*
* See if the block is associated with some buffer
* (mainly to avoid getting hung up on a wait in breada)
daddr_t blkno;
{
register struct buf *bp;
+ register struct buf *dp;
register int dblkno = fsbtodb(blkno);
- for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1];
- bp = &buf[bp->b_hlink])
- if (bp->b_blkno == dblkno && bp->b_dev == dev) {
- if (distrust)
- if (oincore(dev, blkno) != bp) /* TEST */
- panic("incore 1"); /* TEST */
+ dp = BUFHASH(dev, dblkno);
+ for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
+ if (bp->b_blkno == dblkno && bp->b_dev == dev &&
+ !(bp->b_flags & B_INVAL))
return (1);
- }
- if (distrust)
- if (oincore(dev, blkno)) /* TEST */
- panic("incore 2"); /* TEST */
return (0);
}
daddr_t blkno;
{
register struct buf *bp, *dp, *ep;
- register int i, x;
register int dblkno = fsbtodb(blkno);
+#ifdef DISKMON
+ register int i;
+#endif
+ if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT))
+ blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1);
+ dblkno = fsbtodb(blkno);
+ dp = BUFHASH(dev, dblkno);
loop:
- VOID spl0();
- for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1];
- bp = &buf[bp->b_hlink]) {
- if (bp->b_blkno != dblkno || bp->b_dev != dev)
+ (void) spl0();
+ for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
+ if (bp->b_blkno != dblkno || bp->b_dev != dev ||
+ bp->b_flags&B_INVAL)
continue;
- if (distrust)
- if (bp != oincore(dev, blkno)) /* TEST */
- panic("getblk 1"); /* TEST */
- VOID spl6();
+ (void) spl6();
if (bp->b_flags&B_BUSY) {
bp->b_flags |= B_WANTED;
sleep((caddr_t)bp, PRIBIO+1);
goto loop;
}
- VOID spl0();
+ (void) spl0();
#ifdef DISKMON
i = 0;
dp = bp->av_forw;
- while (dp != &bfreelist) {
+ while ((dp->b_flags & B_HEAD) == 0) {
i++;
dp = dp->av_forw;
}
- if (i<NBUF)
+ if (i<64)
io_info.bufcount[i]++;
#endif
notavail(bp);
bp->b_flags |= B_CACHE;
return(bp);
}
- if (distrust)
- if (oincore(dev, blkno)) /* TEST */
- panic("getblk 2"); /* TEST */
if (major(dev) >= nblkdev)
panic("blkdev");
- dp = bdevsw[major(dev)].d_tab;
- if (dp == NULL)
- panic("devtab");
- VOID spl6();
- if (bfreelist.av_forw == &bfreelist) {
- bfreelist.b_flags |= B_WANTED;
- sleep((caddr_t)&bfreelist, PRIBIO+1);
+ (void) spl6();
+ for (ep = &bfreelist[BQUEUES-1]; ep > bfreelist; ep--)
+ if (ep->av_forw != ep)
+ break;
+ if (ep == bfreelist) { /* no free blocks at all */
+ ep->b_flags |= B_WANTED;
+ sleep((caddr_t)ep, PRIBIO+1);
goto loop;
}
- spl0();
- bp = bfreelist.av_forw;
+ (void) spl0();
+ bp = ep->av_forw;
notavail(bp);
if (bp->b_flags & B_DELWRI) {
bp->b_flags |= B_ASYNC;
bwrite(bp);
goto loop;
}
- if (bp->b_dev == NODEV)
- goto done;
- /* INLINE EXPANSION OF bunhash(bp) */
- i = BUFHASH(dbtofsb(bp->b_blkno));
- x = bp - buf;
- if (bufhash[i] == x) {
- bufhash[i] = bp->b_hlink;
- } else {
- for (ep = &buf[bufhash[i]]; ep != &buf[-1];
- ep = &buf[ep->b_hlink])
- if (ep->b_hlink == x) {
- ep->b_hlink = bp->b_hlink;
- goto done;
- }
- panic("getblk");
- }
-done:
- /* END INLINE EXPANSION */
+#ifdef EPAWNJ
+ trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno));
+#endif
bp->b_flags = B_BUSY;
bp->b_back->b_forw = bp->b_forw;
bp->b_forw->b_back = bp->b_back;
dp->b_forw = bp;
bp->b_dev = dev;
bp->b_blkno = dblkno;
- i = BUFHASH(blkno);
- bp->b_hlink = bufhash[i];
- bufhash[i] = bp - buf;
return(bp);
}
struct buf *
geteblk()
{
- register struct buf *bp, *dp, *ep;
- register int i, x;
+ register struct buf *bp, *dp;
loop:
- VOID spl6();
- while (bfreelist.av_forw == &bfreelist) {
- bfreelist.b_flags |= B_WANTED;
- sleep((caddr_t)&bfreelist, PRIBIO+1);
+ (void) spl6();
+ for (dp = &bfreelist[BQUEUES-1]; dp > bfreelist; dp--)
+ if (dp->av_forw != dp)
+ break;
+ if (dp == bfreelist) { /* no free blocks */
+ dp->b_flags |= B_WANTED;
+ sleep((caddr_t)dp, PRIBIO+1);
+ goto loop;
}
- VOID spl0();
- dp = &bfreelist;
- bp = bfreelist.av_forw;
+ (void) spl0();
+ bp = dp->av_forw;
notavail(bp);
if (bp->b_flags & B_DELWRI) {
bp->b_flags |= B_ASYNC;
bwrite(bp);
goto loop;
}
- if (bp->b_dev != NODEV)
- bunhash(bp);
- bp->b_flags = B_BUSY;
+#ifdef EPAWNJ
+ trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno));
+#endif
+ bp->b_flags = B_BUSY|B_INVAL;
bp->b_back->b_forw = bp->b_forw;
bp->b_forw->b_back = bp->b_back;
bp->b_forw = dp->b_forw;
dp->b_forw->b_back = bp;
dp->b_forw = bp;
bp->b_dev = (dev_t)NODEV;
- bp->b_hlink = -1;
return(bp);
}
-bunhash(bp)
- register struct buf *bp;
-{
- register struct buf *ep;
- register int i, x;
-
- if (bp->b_dev == NODEV)
- return;
- i = BUFHASH(dbtofsb(bp->b_blkno));
- x = bp - buf;
- if (bufhash[i] == x) {
- bufhash[i] = bp->b_hlink;
- return;
- }
- for (ep = &buf[bufhash[i]]; ep != &buf[-1];
- ep = &buf[ep->b_hlink])
- if (ep->b_hlink == x) {
- ep->b_hlink = bp->b_hlink;
- return;
- }
- panic("bunhash");
-}
-
/*
* Wait for I/O completion on the buffer; return errors
* to the user.
register struct buf *bp;
{
- VOID spl6();
+ (void) spl6();
while ((bp->b_flags&B_DONE)==0)
sleep((caddr_t)bp, PRIBIO);
- VOID spl0();
+ (void) spl0();
geterror(bp);
}
-#ifndef FASTVAX
+#ifdef UNFAST
/*
* Unlink a buffer from the available list and mark it busy.
* (internal interface)
{
register int s;
+ if (bp->b_flags & B_DONE)
+ panic("dup iodone");
bp->b_flags |= B_DONE;
if (bp->b_flags & B_DIRTY) {
if (bp->b_flags & B_ERROR)
if (bswlist.b_flags & B_WANTED)
wakeup((caddr_t)&proc[2]);
splx(s);
+ return;
}
if (bp->b_flags&B_ASYNC)
brelse(bp);
int p2dp;
register struct pte *dpte, *vpte;
- VOID spl6();
+ (void) 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;
- VOID spl0();
+ (void) spl0();
bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
bp->b_bcount = c;
bp->b_blkno = dblkno;
bp->b_dev = dev;
- if (dev == swapdev)
- bp->b_blkno += swplo;
+ if (flag & B_DIRTY) {
+ swpf[bp - swbuf] = pfcent;
+ swsize[bp - swbuf] = nbytes;
+ }
(*bdevsw[major(dev)].d_strategy)(bp);
if (flag & B_DIRTY) {
if (c < nbytes)
panic("big push");
- swsize[bp - swbuf] = nbytes;
- swpf[bp - swbuf] = pfcent;
return;
}
- VOID spl6();
+ (void) spl6();
while((bp->b_flags&B_DONE)==0)
sleep((caddr_t)bp, PSWP);
- VOID spl0();
+ (void) spl0();
bp->b_un.b_addr += c;
bp->b_flags &= ~B_DONE;
if (bp->b_flags & B_ERROR) {
nbytes -= c;
dblkno += btoc(c);
}
- VOID spl6();
+ (void) 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;
wakeup((caddr_t)&bswlist);
wakeup((caddr_t)&proc[2]);
}
- VOID spl0();
+ (void) spl0();
}
/*
* done by user) after killing it so noone will try
* to swap it out.
*/
- psignal(p, SIGKIL);
+ psignal(p, SIGKILL);
p->p_flag |= SULOCK;
}
dev_t dev;
{
register struct buf *bp;
+ register struct buf *flist;
loop:
- VOID spl6();
- for (bp = bfreelist.av_forw; bp != &bfreelist; bp = bp->av_forw) {
+ (void) spl6();
+ for (flist = bfreelist; flist < &bfreelist[BQUEUES]; flist++)
+ for (bp = flist->av_forw; bp != flist; bp = bp->av_forw) {
if (bp->b_flags&B_DELWRI && (dev == NODEV||dev==bp->b_dev)) {
bp->b_flags |= B_ASYNC;
notavail(bp);
goto loop;
}
}
- VOID spl0();
+ (void) spl0();
}
/*
u.u_error = EFAULT;
return;
}
- VOID spl6();
+ (void) spl6();
while (bp->b_flags&B_BUSY) {
bp->b_flags |= B_WANTED;
sleep((caddr_t)bp, PRIBIO+1);
u.u_procp->p_flag |= SPHYSIO;
vslock(a = bp->b_un.b_addr, c);
(*strat)(bp);
- VOID spl6();
+ (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);
- VOID spl0();
+ (void) spl0();
bp->b_un.b_addr += c;
u.u_count -= c;
u.u_offset += c;
if ((u.u_error = bp->b_error)==0)
u.u_error = EIO;
}
+
+/*
+ * Invalidate in core blocks belonging to closed or umounted filesystem
+ *
+ * This is not nicely done at all - the buffer ought to be removed from the
+ * hash chains & have its dev/blkno fields clobbered, but unfortunately we
+ * can't do that here, as it is quite possible that the block is still
+ * being used for i/o. Eventually, all disc drivers should be forced to
+ * have a close routine, which ought ensure that the queue is empty, then
+ * properly flush the queues. Until that happy day, this suffices for
+ * correctness. ... kre
+ */
+binval(dev)
+dev_t dev;
+{
+ register struct buf *bp;
+ register struct bufhd *hp;
+#define dp ((struct buf *)hp)
+
+ for (hp = bufhash; hp < &bufhash[BUFHSZ]; hp++)
+ for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
+ if (bp->b_dev == dev)
+ bp->b_flags |= B_INVAL;
+}