../machine => machine
[unix-history] / usr / src / sys / kern / kern_physio.c
index 3015942..79e2e03 100644 (file)
@@ -1,71 +1,30 @@
-/*     kern_physio.c   4.23    81/07/25        */
-
-#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"
-
 /*
 /*
- * The following several routines allocate and free
- * buffers with various side effects.  In general the
- * arguments to an allocate routine are a device and
- * a block number, and the value is a pointer to
- * to the buffer header; the buffer is marked "busy"
- * so that no one else can touch it.  If the block was
- * already in core, no I/O need be done; if it is
- * already busy, the process waits until it becomes free.
- * The following routines allocate a buffer:
- *     getblk
- *     bread
- *     breada
- *     baddr   (if it is incore)
- * Eventually the buffer must be released, possibly with the
- * side effect of writing it out, by using one of
- *     bwrite
- *     bdwrite
- *     bawrite
- *     brelse
+ * 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       7.5 (Berkeley) %G%
  */
 
  */
 
-struct buf bfreelist[BQUEUES];
-struct buf bswlist, *bclnlist;
-
-#define        BUFHSZ  63
-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 (bp = bufhash, i = 0; i < BUFHSZ; i++, bp++)
-               bp->b_forw = bp->b_back = (struct buf *)bp;
-}
-
-/* #define     DISKMON 1 */
-
-#ifdef DISKMON
-struct {
-       int     nbuf;
-       long    nread;
-       long    nreada;
-       long    ncache;
-       long    nwrite;
-       long    bufcount[64];
-} io_info;
-#endif
+#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"
+
+#include "machine/pte.h"
+#ifdef SECSIZE
+#include "file.h"
+#include "ioctl.h"
+#include "disklabel.h"
+#endif SECSIZE
 
 /*
  * Swap IO headers -
 
 /*
  * Swap IO headers -
@@ -78,462 +37,6 @@ struct {
  * in a list of cleaned pages to be processed by the pageout daemon.
  */
 struct buf *swbuf;
  * 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;
-
-
-#ifndef        UNFAST
-#define        notavail(bp) \
-{ \
-       int s = spl6(); \
-       (bp)->av_back->av_forw = (bp)->av_forw; \
-       (bp)->av_forw->av_back = (bp)->av_back; \
-       (bp)->b_flags |= B_BUSY; \
-       splx(s); \
-}
-#endif
-
-/*
- * Read in (if necessary) the block and return a buffer pointer.
- */
-struct buf *
-bread(dev, blkno)
-dev_t dev;
-daddr_t blkno;
-{
-       register struct buf *bp;
-
-       bp = getblk(dev, blkno);
-       if (bp->b_flags&B_DONE) {
-#ifdef TRACE
-               trace(TR_BREADHIT, dev, blkno);
-#endif
-#ifdef DISKMON
-               io_info.ncache++;
-#endif
-               return(bp);
-       }
-       bp->b_flags |= B_READ;
-       bp->b_bcount = BSIZE;
-       (*bdevsw[major(dev)].d_strategy)(bp);
-#ifdef TRACE
-       trace(TR_BREADMISS, dev, blkno);
-#endif
-#ifdef DISKMON
-       io_info.nread++;
-#endif
-       u.u_vm.vm_inblk++;              /* pay for read */
-       iowait(bp);
-       return(bp);
-}
-
-/*
- * Read in the block, like bread, but also start I/O on the
- * read-ahead block (which is not allocated to the caller)
- */
-struct buf *
-breada(dev, blkno, rablkno)
-dev_t dev;
-daddr_t blkno, rablkno;
-{
-       register struct buf *bp, *rabp;
-
-       bp = NULL;
-       if (!incore(dev, blkno)) {
-               bp = getblk(dev, blkno);
-               if ((bp->b_flags&B_DONE) == 0) {
-                       bp->b_flags |= B_READ;
-                       bp->b_bcount = BSIZE;
-                       (*bdevsw[major(dev)].d_strategy)(bp);
-#ifdef TRACE
-                       trace(TR_BREADMISS, dev, blkno);
-#endif
-#ifdef DISKMON
-                       io_info.nread++;
-#endif
-                       u.u_vm.vm_inblk++;              /* pay for read */
-               }
-#ifdef TRACE
-               else
-                       trace(TR_BREADHIT, dev, blkno);
-#endif
-       }
-       if (rablkno && !incore(dev, rablkno)) {
-               rabp = getblk(dev, rablkno);
-               if (rabp->b_flags & B_DONE) {
-                       brelse(rabp);
-#ifdef TRACE
-                       trace(TR_BREADHITRA, dev, blkno);
-#endif
-               } else {
-                       rabp->b_flags |= B_READ|B_ASYNC;
-                       rabp->b_bcount = BSIZE;
-                       (*bdevsw[major(dev)].d_strategy)(rabp);
-#ifdef TRACE
-                       trace(TR_BREADMISSRA, dev, rablock);
-#endif
-#ifdef DISKMON
-                       io_info.nreada++;
-#endif
-                       u.u_vm.vm_inblk++;              /* pay in advance */
-               }
-       }
-       if(bp == NULL)
-               return(bread(dev, blkno));
-       iowait(bp);
-       return(bp);
-}
-
-/*
- * Write the buffer, waiting for completion.
- * Then release the buffer.
- */
-bwrite(bp)
-register struct buf *bp;
-{
-       register flag;
-
-       flag = bp->b_flags;
-       bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI | B_AGE);
-       bp->b_bcount = BSIZE;
-#ifdef DISKMON
-       io_info.nwrite++;
-#endif
-       if ((flag&B_DELWRI) == 0)
-               u.u_vm.vm_oublk++;              /* noone paid yet */
-#ifdef TRACE
-       trace(TR_BWRITE, bp->b_dev, bp->b_blkno);
-#endif
-       (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
-       if ((flag&B_ASYNC) == 0) {
-               iowait(bp);
-               brelse(bp);
-       } else if (flag & B_DELWRI)
-               bp->b_flags |= B_AGE;
-       else
-               geterror(bp);
-}
-
-/*
- * Release the buffer, marking it so that if it is grabbed
- * for another purpose it will be written out before being
- * given up (e.g. when writing a partial block where it is
- * assumed that another write for the same block will soon follow).
- * This can't be done for magtape, since writes must be done
- * in the same order as requested.
- */
-bdwrite(bp)
-register struct buf *bp;
-{
-       register int flags;
-
-       if ((bp->b_flags&B_DELWRI) == 0)
-               u.u_vm.vm_oublk++;              /* noone paid yet */
-       flags = bdevsw[major(bp->b_dev)].d_flags;
-       if(flags & B_TAPE)
-               bawrite(bp);
-       else {
-               bp->b_flags |= B_DELWRI | B_DONE;
-               brelse(bp);
-       }
-}
-
-/*
- * Release the buffer, start I/O on it, but don't wait for completion.
- */
-bawrite(bp)
-register struct buf *bp;
-{
-
-       bp->b_flags |= B_ASYNC;
-       bwrite(bp);
-}
-
-/*
- * release the buffer, with no I/O implied.
- */
-brelse(bp)
-register struct buf *bp;
-{
-       register struct buf *flist;
-       register s;
-
-       if (bp->b_flags&B_WANTED)
-               wakeup((caddr_t)bp);
-       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_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 {
-               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);
-}
-
-/*
- * See if the block is associated with some buffer
- * (mainly to avoid getting hung up on a wait in breada)
- */
-incore(dev, blkno)
-dev_t dev;
-daddr_t blkno;
-{
-       register struct buf *bp;
-       register struct buf *dp;
-       register int dblkno = fsbtodb(blkno);
-
-       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);
-       return (0);
-}
-
-struct buf *
-baddr(dev, blkno)
-dev_t dev;
-daddr_t blkno;
-{
-
-       if (incore(dev, blkno))
-               return (bread(dev, blkno));
-       return (0);
-}
-
-/*
- * Assign a buffer for the given block.  If the appropriate
- * block is already associated, return it; otherwise search
- * for the oldest non-busy buffer and reassign it.
- */
-struct buf *
-getblk(dev, blkno)
-dev_t dev;
-daddr_t blkno;
-{
-       register struct buf *bp, *dp, *ep;
-       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 = dp->b_forw; bp != dp; bp = bp->b_forw) {
-               if (bp->b_blkno != dblkno || bp->b_dev != dev ||
-                   bp->b_flags&B_INVAL)
-                       continue;
-               (void) spl6();
-               if (bp->b_flags&B_BUSY) {
-                       bp->b_flags |= B_WANTED;
-                       sleep((caddr_t)bp, PRIBIO+1);
-                       goto loop;
-               }
-               (void) spl0();
-#ifdef DISKMON
-               i = 0;
-               dp = bp->av_forw;
-               while ((dp->b_flags & B_HEAD) == 0) {
-                       i++;
-                       dp = dp->av_forw;
-               }
-               if (i<64)
-                       io_info.bufcount[i]++;
-#endif
-               notavail(bp);
-               bp->b_flags |= B_CACHE;
-               return(bp);
-       }
-       if (major(dev) >= nblkdev)
-               panic("blkdev");
-       (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;
-       }
-       (void) spl0();
-       bp = ep->av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-#ifdef TRACE
-       trace(TR_BRELSE, bp->b_dev, 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;
-       bp->b_forw = dp->b_forw;
-       bp->b_back = dp;
-       dp->b_forw->b_back = bp;
-       dp->b_forw = bp;
-       bp->b_dev = dev;
-       bp->b_blkno = dblkno;
-       return(bp);
-}
-
-/*
- * get an empty block,
- * not assigned to any particular device
- */
-struct buf *
-geteblk()
-{
-       register struct buf *bp, *dp;
-
-loop:
-       (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();
-       bp = dp->av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-#ifdef TRACE
-       trace(TR_BRELSE, bp->b_dev, 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;
-       bp->b_back = dp;
-       dp->b_forw->b_back = bp;
-       dp->b_forw = bp;
-       bp->b_dev = (dev_t)NODEV;
-       return(bp);
-}
-
-/*
- * Wait for I/O completion on the buffer; return errors
- * to the user.
- */
-iowait(bp)
-register struct buf *bp;
-{
-
-       (void) spl6();
-       while ((bp->b_flags&B_DONE)==0)
-               sleep((caddr_t)bp, PRIBIO);
-       (void) spl0();
-       geterror(bp);
-}
-
-#ifdef UNFAST
-/*
- * Unlink a buffer from the available list and mark it busy.
- * (internal interface)
- */
-notavail(bp)
-register struct buf *bp;
-{
-       register s;
-
-       s = spl6();
-       bp->av_back->av_forw = bp->av_forw;
-       bp->av_forw->av_back = bp->av_back;
-       bp->b_flags |= B_BUSY;
-       splx(s);
-}
-#endif
-
-/*
- * Mark I/O complete on a buffer. If the header
- * indicates a dirty page push completion, the
- * header is inserted into the ``cleaned'' list
- * to be processed by the pageout daemon. Otherwise
- * release it if I/O is asynchronous, and wake 
- * up anyone waiting for it.
- */
-iodone(bp)
-register struct buf *bp;
-{
-       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)
-                       panic("IO err in push");
-               s = spl6();
-               bp->av_forw = bclnlist;
-               bp->b_bcount = swsize[bp - swbuf];
-               bp->b_pfcent = swpf[bp - swbuf];
-               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);
-               return;
-       }
-       if (bp->b_flags&B_ASYNC)
-               brelse(bp);
-       else {
-               bp->b_flags &= ~B_WANTED;
-               wakeup((caddr_t)bp);
-       }
-}
-
-/*
- * Zero the core associated with a buffer.
- */
-clrbuf(bp)
-struct buf *bp;
-{
-       register *p;
-       register c;
-
-       p = bp->b_un.b_words;
-       c = BSIZE/sizeof(int);
-       do
-               *p++ = 0;
-       while (--c);
-       bp->b_resid = 0;
-}
 
 /*
  * swap I/O -
 
 /*
  * swap I/O -
@@ -544,32 +47,29 @@ struct buf *bp;
  * 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,
  * 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
+ * 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;
  * 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;
+       int nbytes, rdflg, flag;
        dev_t dev;
        dev_t dev;
-       unsigned pfcent;
+       u_int pfcent;
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-       register int c;
-       int p2dp;
        register struct pte *dpte, *vpte;
        register struct pte *dpte, *vpte;
+       register u_int c;
+       int p2dp, s, error = 0;
+       struct buf *getswbuf();
+       int swdone();
 
 
-       (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();
-
+       bp = getswbuf(PSWP+1);
        bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
        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);
        if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
                if (rdflg == B_READ)
                        sum.v_pswpin += btoc(nbytes);
@@ -585,51 +85,67 @@ swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
                                panic("swap bad pte");
                        *dpte++ = *vpte++;
                }
                                panic("swap bad pte");
                        *dpte++ = *vpte++;
                }
-               bp->b_un.b_addr = (caddr_t)ctob(p2dp);
+               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) {
        } else
                bp->b_un.b_addr = addr;
        while (nbytes > 0) {
-               c = imin(ctob(120), nbytes);
-               bp->b_bcount = c;
+               bp->b_bcount = nbytes;
+               minphys(bp);
+               c = bp->b_bcount;
                bp->b_blkno = dblkno;
                bp->b_dev = dev;
                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);
 #ifdef TRACE
                trace(TR_SWAPIO, dev, bp->b_blkno);
 #endif
                (*bdevsw[major(dev)].d_strategy)(bp);
+               /* pageout daemon doesn't wait for pushed pages */
                if (flag & B_DIRTY) {
                        if (c < nbytes)
                                panic("big push");
                if (flag & B_DIRTY) {
                        if (c < nbytes)
                                panic("big push");
-                       return;
+                       return (0);
                }
                }
-               (void) spl6();
-               while((bp->b_flags&B_DONE)==0)
-                       sleep((caddr_t)bp, PSWP);
-               (void) spl0();
                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");
                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);
+                       swkill(p, "swap: read error from swap device");
+                       error = EIO;
                }
                nbytes -= c;
                }
                nbytes -= c;
-               dblkno += btoc(c);
+#ifdef SECSIZE
+               if (flag & B_PGIN && nbytes > 0)
+                       panic("big pgin");
+#endif SECSIZE
+               dblkno += btodb(c);
        }
        }
-       (void) spl6();
        bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
        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);
+       freeswbuf(bp);
+       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]);
                wakeup((caddr_t)&proc[2]);
-       }
-       (void) spl0();
+       splx(s);
 }
 
 /*
 }
 
 /*
@@ -641,14 +157,9 @@ swkill(p, rout)
        struct proc *p;
        char *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);
+       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
        /*
         * To be sure no looping (e.g. in vmsched trying to
         * swap out) mark process locked in core (as though
@@ -659,37 +170,12 @@ swkill(p, rout)
        p->p_flag |= SULOCK;
 }
 
        p->p_flag |= SULOCK;
 }
 
-/*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
- */
-bflush(dev)
-dev_t dev;
-{
-       register struct buf *bp;
-       register struct buf *flist;
-
-loop:
-       (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);
-                       bwrite(bp);
-                       goto loop;
-               }
-       }
-       (void) spl0();
-}
-
 /*
  * Raw I/O. The arguments are
  *     The strategy routine for the device
 /*
  * 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
+ *     A buffer, which will either be a special buffer header owned
+ *         exclusively by the device for this purpose, or NULL,
+ *         indicating that we should use a swap buffer
  *     The device number
  *     Read/write flag
  * Essentially all the work is computing physical addresses and
  *     The device number
  *     Read/write flag
  * Essentially all the work is computing physical addresses and
@@ -699,100 +185,156 @@ loop:
  * faulted and locked. After the completion of the I/O, the above pages
  * are unlocked.
  */
  * faulted and locked. After the completion of the I/O, the above pages
  * are unlocked.
  */
-physio(strat, bp, dev, rw, mincnt)
-int (*strat)(); 
-register struct buf *bp;
-unsigned (*mincnt)();
+physio(strat, bp, dev, rw, mincnt, uio)
+       int (*strat)(); 
+       register struct buf *bp;
+       dev_t dev;
+       int rw;
+       u_int (*mincnt)();
+       struct uio *uio;
 {
 {
+       register struct iovec *iov;
        register int c;
        char *a;
        register int c;
        char *a;
-
-       if (useracc(u.u_base,u.u_count,rw==B_READ?B_WRITE:B_READ) == NULL) {
-               u.u_error = EFAULT;
-               return;
-       }
-       (void) spl6();
-       while (bp->b_flags&B_BUSY) {
-               bp->b_flags |= B_WANTED;
-               sleep((caddr_t)bp, PRIBIO+1);
-       }
-       bp->b_error = 0;
-       bp->b_proc = u.u_procp;
-       bp->b_un.b_addr = u.u_base;
-       while (u.u_count != 0) {
-               bp->b_flags = B_BUSY | B_PHYS | rw;
-               bp->b_dev = dev;
-               bp->b_blkno = u.u_offset >> PGSHIFT;
-               bp->b_bcount = u.u_count;
-               (*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);
-               (void) spl0();
-               bp->b_un.b_addr += c;
-               u.u_count -= c;
-               u.u_offset += c;
-               if (bp->b_flags&B_ERROR)
-                       break;
+       int s, allocbuf = 0, error = 0;
+       struct buf *getswbuf();
+#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);
+               }
+               if (!allocbuf) {        /* only if sharing caller's buffer */
+                       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--;
        }
        }
-       bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
-       u.u_count = bp->b_resid;
-       geterror(bp);
+       if (allocbuf)
+               freeswbuf(bp);
+       return (error);
 }
 
 }
 
-/*ARGSUSED*/
-unsigned
+u_int
 minphys(bp)
 minphys(bp)
-struct buf *bp;
+       struct buf *bp;
 {
 {
+       if (bp->b_bcount > MAXPHYS)
+               bp->b_bcount = MAXPHYS;
+}
+
+static
+struct buf *
+getswbuf(prio)
+       int prio;
+{
+       int s;
+       struct buf *bp;
 
 
-       if (bp->b_bcount > 60 * 1024)
-               bp->b_bcount = 60 * 1024;
+       s = splbio();
+       while (bswlist.av_forw == NULL) {
+               bswlist.b_flags |= B_WANTED;
+               sleep((caddr_t)&bswlist, prio);
+       }
+       bp = bswlist.av_forw;
+       bswlist.av_forw = bp->av_forw;
+       splx(s);
+       return (bp);
 }
 
 }
 
-/*
- * Pick up the device's error number and pass it to the user;
- * if there is an error but the number is 0 set a generalized
- * code.  Actually the latter is always true because devices
- * don't yet return specific errors.
- */
-geterror(bp)
-register struct buf *bp;
+static
+freeswbuf(bp)
+       struct buf *bp;
 {
 {
+       int s;
 
 
-       if (bp->b_flags&B_ERROR)
-               if ((u.u_error = bp->b_error)==0)
-                       u.u_error = EIO;
+       s = splbio();
+       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);
 }
 
 }
 
-/*
- * 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;
+rawread(dev, uio)
+       dev_t dev;
+       struct uio *uio;
 {
 {
-       register struct buf *bp;
-       register struct bufhd *hp;
-#define dp ((struct buf *)hp)
+       return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
+           dev, B_READ, minphys, uio));
+}
 
 
-       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;
+rawwrite(dev, uio)
+       dev_t dev;
+       struct uio *uio;
+{
+       return (physio(cdevsw[major(dev)].d_strategy, (struct buf *)NULL,
+           dev, B_WRITE, minphys, uio));
 }
 }