projects / unix-history / blobdiff
? search:
summary | tags | clone url | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
more-or-less working with new proc & user structs
[unix-history] / usr / src / sys / kern / vfs_cluster.c
diff --git a/usr/src/sys/kern/vfs_cluster.c b/usr/src/sys/kern/vfs_cluster.c
index 9be3827..94a8d27 100644 (file)
--- a/usr/src/sys/kern/vfs_cluster.c
+++ b/usr/src/sys/kern/vfs_cluster.c
@@ -1,549 +1,578 @@
-/*     vfs_cluster.c   4.5     %G%     */
-
-#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, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)vfs_cluster.c       7.34 (Berkeley) %G%
  */
 
  */
 
-#define        BUFHSZ  63
-#define        BUFHASH(blkno)  (blkno % BUFHSZ)
-short  bufhash[BUFHSZ];
+#include "param.h"
+#include "user.h"
+#include "proc.h"
+#include "buf.h"
+#include "vnode.h"
+#include "specdev.h"
+#include "mount.h"
+#include "trace.h"
+#include "ucred.h"
 
 /*
 
 /*
- * Initialize hash links for buffers.
+ * Find the block in the buffer pool.
+ * If the buffer is not present, allocate a new buffer and load
+ * its contents according to the filesystem fill routine.
  */
  */
-bhinit()
-{
-       register int i;
-
-       for (i = 0; i < BUFHSZ; i++)
-               bufhash[i] = -1;
-}
-
-/* #define     DISKMON 1 */
-
-#ifdef DISKMON
-struct {
-       int     nbuf;
-       long    nread;
-       long    nreada;
-       long    ncache;
-       long    nwrite;
-       long    bufcount[NBUF];
-} io_info;
-#endif
-
-/*
- * 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[NSWBUF];
-short  swsize[NSWBUF];         /* CAN WE JUST USE B_BCOUNT? */
-int    swpf[NSWBUF];
-
-
-#ifdef FASTVAX
-#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;
+bread(vp, blkno, size, cred, bpp)
+       struct vnode *vp;
+       daddr_t blkno;
+       int size;
+       struct ucred *cred;
+       struct buf **bpp;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
 {
 {
+       struct proc *p = curproc;               /* XXX */
        register struct buf *bp;
 
        register struct buf *bp;
 
-       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
-               return(bp);
+       if (size == 0)
+               panic("bread: size 0");
+#ifdef SECSIZE
+       bp = getblk(dev, blkno, size, secsize);
+#else SECSIZE
+       *bpp = bp = getblk(vp, blkno, size);
+#endif SECSIZE
+       if (bp->b_flags & (B_DONE | B_DELWRI)) {
+               trace(TR_BREADHIT, pack(vp, size), blkno);
+               return (0);
        }
        bp->b_flags |= B_READ;
        }
        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 */
-       iowait(bp);
-       return(bp);
+       if (bp->b_bcount > bp->b_bufsize)
+               panic("bread");
+       if (bp->b_rcred == NOCRED && cred != NOCRED) {
+               crhold(cred);
+               bp->b_rcred = cred;
+       }
+       VOP_STRATEGY(bp);
+       trace(TR_BREADMISS, pack(vp, size), blkno);
+       p->p_stats->p_ru.ru_inblock++;          /* pay for read */
+       return (biowait(bp));
 }
 
 /*
 }
 
 /*
- * Read in the block, like bread, but also start I/O on the
- * read-ahead block (which is not allocated to the caller)
+ * Operates like bread, but also starts I/O on the specified
+ * read-ahead block.
  */
  */
-struct buf *
-breada(dev, blkno, rablkno)
-dev_t dev;
-daddr_t blkno, rablkno;
+breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
+       struct vnode *vp;
+       daddr_t blkno; int size;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
+       daddr_t rablkno; int rabsize;
+       struct ucred *cred;
+       struct buf **bpp;
 {
 {
+       struct proc *p = curproc;               /* XXX */
        register struct buf *bp, *rabp;
 
        bp = NULL;
        register struct buf *bp, *rabp;
 
        bp = NULL;
-       if (!incore(dev, blkno)) {
-               bp = getblk(dev, blkno);
-               if ((bp->b_flags&B_DONE) == 0) {
+       /*
+        * If the block is not memory resident,
+        * allocate a buffer and start I/O.
+        */
+       if (!incore(vp, blkno)) {
+               *bpp = bp = getblk(vp, blkno, size);
+#endif SECSIZE
+               if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) {
                        bp->b_flags |= B_READ;
                        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 (bp->b_bcount > bp->b_bufsize)
+                               panic("breada");
+                       if (bp->b_rcred == NOCRED && cred != NOCRED) {
+                               crhold(cred);
+                               bp->b_rcred = cred;
+                       }
+                       VOP_STRATEGY(bp);
+                       trace(TR_BREADMISS, pack(vp, size), blkno);
+                       p->p_stats->p_ru.ru_inblock++;  /* pay for read */
+               } else
+                       trace(TR_BREADHIT, pack(vp, size), blkno);
        }
        }
-       if (rablkno && !incore(dev, rablkno)) {
-               rabp = getblk(dev, rablkno);
-               if (rabp->b_flags & B_DONE) {
+
+       /*
+        * If there is a read-ahead block, start I/O on it too.
+        */
+       if (!incore(vp, rablkno)) {
+               rabp = getblk(vp, rablkno, rabsize);
+#endif SECSIZE
+               if (rabp->b_flags & (B_DONE | B_DELWRI)) {
                        brelse(rabp);
                        brelse(rabp);
-#ifdef EPAWNJ
-                       trace(TR_BREAD|TR_HIT|TR_RA, dev, blkno);
-#endif
+                       trace(TR_BREADHITRA, pack(vp, rabsize), rablkno);
                } else {
                } 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
-                       u.u_vm.vm_inblk++;              /* pay in advance */
+                       rabp->b_flags |= B_ASYNC | B_READ;
+                       if (rabp->b_bcount > rabp->b_bufsize)
+                               panic("breadrabp");
+                       if (rabp->b_rcred == NOCRED && cred != NOCRED) {
+                               crhold(cred);
+                               rabp->b_rcred = cred;
+                       }
+                       VOP_STRATEGY(rabp);
+                       trace(TR_BREADMISSRA, pack(vp, rabsize), rablkno);
+                       p->p_stats->p_ru.ru_inblock++;  /* pay in advance */
                }
        }
                }
        }
-       if(bp == NULL)
-               return(bread(dev, blkno));
-       iowait(bp);
-       return(bp);
+
+       /*
+        * If block was memory resident, let bread get it.
+        * If block was not memory resident, the read was
+        * started above, so just wait for the read to complete.
+        */
+       if (bp == NULL)
+#ifdef SECSIZE
+               return (bread(dev, blkno, size, secsize));
+#else SECSIZE
+               return (bread(vp, blkno, size, cred, bpp));
+       return (biowait(bp));
 }
 
 /*
 }
 
 /*
- * Write the buffer, waiting for completion.
- * Then release the buffer.
+ * Synchronous write.
+ * Release buffer on completion.
  */
 bwrite(bp)
  */
 bwrite(bp)
-register struct buf *bp;
+       register struct buf *bp;
 {
 {
-       register flag;
+       struct proc *p = curproc;               /* XXX */
+       register int flag;
+       int s, error;
 
        flag = bp->b_flags;
 
        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 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);
+       bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
+       if ((flag & B_DELWRI) == 0)
+               p->p_stats->p_ru.ru_oublock++;          /* no one paid yet */
+       else
+               reassignbuf(bp, bp->b_vp);
+       trace(TR_BWRITE, pack(bp->b_vp, bp->b_bcount), bp->b_lblkno);
+       if (bp->b_bcount > bp->b_bufsize)
+               panic("bwrite");
+       s = splbio();
+       bp->b_vp->v_numoutput++;
+       splx(s);
+       VOP_STRATEGY(bp);
+
+       /*
+        * If the write was synchronous, then await I/O completion.
+        * If the write was "delayed", then we put the buffer on
+        * the queue of blocks awaiting I/O completion status.
+        */
+       if ((flag & B_ASYNC) == 0) {
+               error = biowait(bp);
                brelse(bp);
                brelse(bp);
-       } else if (flag & B_DELWRI)
+       } else if (flag & B_DELWRI) {
                bp->b_flags |= B_AGE;
                bp->b_flags |= B_AGE;
-       else
-               geterror(bp);
+               error = 0;
+       }
+       return (error);
 }
 
 /*
 }
 
 /*
- * 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.
+ * Delayed write.
+ *
+ * The buffer is marked dirty, but is not queued for I/O.
+ * This routine should be used when the buffer is expected
+ * to be modified again soon, typically a small write that
+ * partially fills a buffer.
+ *
+ * NB: magnetic tapes cannot be delayed; they must be
+ * written in the order that the writes are requested.
  */
 bdwrite(bp)
  */
 bdwrite(bp)
-register struct buf *bp;
+       register struct buf *bp;
 {
 {
-       register struct buf *dp;
+       struct proc *p = curproc;               /* XXX */
 
 
-       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)
+       if ((bp->b_flags & B_DELWRI) == 0) {
+               bp->b_flags |= B_DELWRI;
+               reassignbuf(bp, bp->b_vp);
+               p->p_stats->p_ru.ru_oublock++;          /* no one paid yet */
+       }
+       /*
+        * If this is a tape drive, the write must be initiated.
+        */
+       if (bdevsw[major(bp->b_dev)].d_flags & B_TAPE)
                bawrite(bp);
                bawrite(bp);
-       else {
-               bp->b_flags |= B_DELWRI | B_DONE;
+       } else {
+               bp->b_flags |= (B_DONE | B_DELWRI);
                brelse(bp);
        }
 }
 
 /*
                brelse(bp);
        }
 }
 
 /*
- * Release the buffer, start I/O on it, but don't wait for completion.
+ * Asynchronous write.
+ * Start I/O on a buffer, but do not wait for it to complete.
+ * The buffer is released when the I/O completes.
  */
 bawrite(bp)
  */
 bawrite(bp)
-register struct buf *bp;
+       register struct buf *bp;
 {
 
 {
 
+       /*
+        * Setting the ASYNC flag causes bwrite to return
+        * after starting the I/O.
+        */
        bp->b_flags |= B_ASYNC;
        bp->b_flags |= B_ASYNC;
-       bwrite(bp);
+       (void) bwrite(bp);
 }
 
 /*
 }
 
 /*
- * release the buffer, with no I/O implied.
+ * Release a buffer.
+ * Even if the buffer is dirty, no I/O is started.
  */
 brelse(bp)
  */
 brelse(bp)
-register struct buf *bp;
+       register struct buf *bp;
 {
 {
-       register struct buf **backp;
-       register s;
+       register struct buf *flist;
+       int s;
 
 
-       if (bp->b_flags&B_WANTED)
+       trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
+       /*
+        * If a process is waiting for the buffer, or
+        * is waiting for a free buffer, awaken it.
+        */
+       if (bp->b_flags & B_WANTED)
                wakeup((caddr_t)bp);
                wakeup((caddr_t)bp);
-       if (bfreelist.b_flags&B_WANTED) {
-               bfreelist.b_flags &= ~B_WANTED;
-               wakeup((caddr_t)&bfreelist);
+       if (bfreelist[0].b_flags & B_WANTED) {
+               bfreelist[0].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 */
+       /*
+        * Retry I/O for locked buffers rather than invalidating them.
+        */
+       if ((bp->b_flags & B_ERROR) && (bp->b_flags & B_LOCKED))
+               bp->b_flags &= ~B_ERROR;
+       /*
+        * Disassociate buffers that are no longer valid.
+        */
+       if (bp->b_flags & (B_NOCACHE | B_ERROR))
+               bp->b_flags |= B_INVAL;
+       if ((bp->b_bufsize <= 0) || (bp->b_flags & (B_ERROR | B_INVAL))) {
+               if (bp->b_vp)
+                       brelvp(bp);
+               bp->b_flags &= ~B_DELWRI;
        }
        }
-       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;
+       /*
+        * Stick the buffer back on a free list.
+        */
+       s = splbio();
+       if (bp->b_bufsize <= 0) {
+               /* block has no buffer ... put at front of unused buffer list */
+               flist = &bfreelist[BQ_EMPTY];
+               binsheadfree(bp, flist);
+       } else if (bp->b_flags & (B_ERROR | B_INVAL)) {
+               /* block has no info ... put at front of most free list */
+               flist = &bfreelist[BQ_AGE];
+               binsheadfree(bp, flist);
        } else {
        } 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];
+               binstailfree(bp, flist);
        }
        }
-       bp->b_flags &= ~(B_WANTED|B_BUSY|B_ASYNC|B_AGE);
+       bp->b_flags &= ~(B_WANTED | B_BUSY | B_ASYNC | B_AGE | B_NOCACHE);
        splx(s);
 }
 
 /*
        splx(s);
 }
 
 /*
- * See if the block is associated with some buffer
- * (mainly to avoid getting hung up on a wait in breada)
+ * Check to see if a block is currently memory resident.
  */
  */
-incore(dev, blkno)
-dev_t dev;
-daddr_t blkno;
+incore(vp, blkno)
+       struct vnode *vp;
+       daddr_t blkno;
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-       register int dblkno = fsbtodb(blkno);
+       register struct buf *dp;
 
 
-       for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1];
-           bp = &buf[bp->b_hlink])
-               if (bp->b_blkno == dblkno && bp->b_dev == dev
-                                       && !(bp->b_flags & B_INVAL))
+       dp = BUFHASH(vp, blkno);
+       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
+               if (bp->b_lblkno == blkno && bp->b_vp == vp &&
+                   (bp->b_flags & B_INVAL) == 0)
                        return (1);
        return (0);
 }
 
                        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.
+ * Check to see if a block is currently memory resident.
+ * If it is resident, return it. If it is not resident,
+ * allocate a new buffer and assign it to the block.
  */
 struct buf *
  */
 struct buf *
-getblk(dev, blkno)
-dev_t dev;
-daddr_t blkno;
+#ifdef SECSIZE
+getblk(dev, blkno, size, secsize)
+#else SECSIZE
+getblk(vp, blkno, size)
+       register struct vnode *vp;
+       daddr_t blkno;
+       int size;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
 {
 {
-       register struct buf *bp, *dp, *ep;
-       register int i, x, dblkno;
-
-       if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT))
-               blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1);
-       dblkno = fsbtodb(blkno);
-    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
-                                       || bp->b_flags & B_INVAL)
+       register struct buf *bp, *dp;
+       int s;
+
+       if (size > MAXBSIZE)
+               panic("getblk: size too big");
+       /*
+        * Search the cache for the block. If the buffer is found,
+        * but it is currently locked, the we must wait for it to
+        * become available.
+        */
+       dp = BUFHASH(vp, blkno);
+loop:
+       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
+               if (bp->b_lblkno != blkno || bp->b_vp != vp ||
+                   (bp->b_flags & B_INVAL))
                        continue;
                        continue;
-               (void) spl6();
-               if (bp->b_flags&B_BUSY) {
+               s = splbio();
+               if (bp->b_flags & B_BUSY) {
                        bp->b_flags |= B_WANTED;
                        bp->b_flags |= B_WANTED;
-                       sleep((caddr_t)bp, PRIBIO+1);
+                       sleep((caddr_t)bp, PRIBIO + 1);
+                       splx(s);
                        goto loop;
                }
                        goto loop;
                }
-               (void) spl0();
-#ifdef DISKMON
-               i = 0;
-               dp = bp->av_forw;
-               while (dp != &bfreelist) {
-                       i++;
-                       dp = dp->av_forw;
+               bremfree(bp);
+               bp->b_flags |= B_BUSY;
+               splx(s);
+               if (bp->b_bcount != size) {
+                       printf("getblk: stray size");
+                       bp->b_flags |= B_INVAL;
+                       bwrite(bp);
+                       goto loop;
                }
                }
-               if (i<NBUF)
-                       io_info.bufcount[i]++;
-#endif
-               notavail(bp);
                bp->b_flags |= B_CACHE;
                bp->b_flags |= B_CACHE;
-               return(bp);
-       }
-       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);
-               goto loop;
+               return (bp);
        }
        }
-       (void) spl0();
-       bp = bfreelist.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) */
-#ifdef EPAWNJ
-       trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno));
-#endif
-       (void) spl6();
-       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:
-       (void) spl0();
-       /* END INLINE EXPANSION */
-       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;
-       i = BUFHASH(blkno);
-       bp->b_hlink = bufhash[i];
-       bufhash[i] = bp - buf;
-       return(bp);
+       bp = getnewbuf();
+       bremhash(bp);
+       bgetvp(vp, bp);
+       bp->b_bcount = 0;
+       bp->b_lblkno = blkno;
+#ifdef SECSIZE
+       bp->b_blksize = secsize;
+#endif SECSIZE
+       bp->b_blkno = blkno;
+       bp->b_error = 0;
+       bp->b_resid = 0;
+       binshash(bp, dp);
+       allocbuf(bp, size);
+       return (bp);
 }
 
 /*
 }
 
 /*
- * get an empty block,
- * not assigned to any particular device
+ * Allocate a buffer.
+ * The caller will assign it to a block.
  */
 struct buf *
  */
 struct buf *
-geteblk()
+geteblk(size)
+       int size;
 {
 {
-       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) spl0();
-       dp = &bfreelist;
-       bp = bfreelist.av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-       if (bp->b_dev != NODEV) {
-#ifdef EPAWNJ
-               trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno));
-#endif
-               bunhash(bp);
-       }
-       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_t)NODEV;
-       bp->b_hlink = -1;
-       return(bp);
+       register struct buf *bp, *flist;
+
+       if (size > MAXBSIZE)
+               panic("geteblk: size too big");
+       bp = getnewbuf();
+       bp->b_flags |= B_INVAL;
+       bremhash(bp);
+       flist = &bfreelist[BQ_AGE];
+       bp->b_bcount = 0;
+#ifdef SECSIZE
+       bp->b_blksize = DEV_BSIZE;
+#endif SECSIZE
+       bp->b_error = 0;
+       bp->b_resid = 0;
+       binshash(bp, flist);
+       allocbuf(bp, size);
+       return (bp);
 }
 
 }
 
-bunhash(bp)
-       register struct buf *bp;
+/*
+ * Expand or contract the actual memory allocated to a buffer.
+ * If no memory is available, release buffer and take error exit.
+ */
+allocbuf(tp, size)
+       register struct buf *tp;
+       int size;
 {
 {
-       register struct buf *ep;
-       register int i, x, s;
+       register struct buf *bp, *ep;
+       int sizealloc, take, s;
 
 
-       if (bp->b_dev == NODEV)
-               return;
-       s = spl6();
-       i = BUFHASH(dbtofsb(bp->b_blkno));
-       x = bp - buf;
-       if (bufhash[i] == x) {
-               bufhash[i] = bp->b_hlink;
-               goto ret;
+       sizealloc = roundup(size, CLBYTES);
+       /*
+        * Buffer size does not change
+        */
+       if (sizealloc == tp->b_bufsize)
+               goto out;
+       /*
+        * Buffer size is shrinking.
+        * Place excess space in a buffer header taken from the
+        * BQ_EMPTY buffer list and placed on the "most free" list.
+        * If no extra buffer headers are available, leave the
+        * extra space in the present buffer.
+        */
+       if (sizealloc < tp->b_bufsize) {
+               ep = bfreelist[BQ_EMPTY].av_forw;
+               if (ep == &bfreelist[BQ_EMPTY])
+                       goto out;
+               s = splbio();
+               bremfree(ep);
+               ep->b_flags |= B_BUSY;
+               splx(s);
+               pagemove(tp->b_un.b_addr + sizealloc, ep->b_un.b_addr,
+                   (int)tp->b_bufsize - sizealloc);
+               ep->b_bufsize = tp->b_bufsize - sizealloc;
+               tp->b_bufsize = sizealloc;
+               ep->b_flags |= B_INVAL;
+               ep->b_bcount = 0;
+               brelse(ep);
+               goto out;
        }
        }
-       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 ret;
+       /*
+        * More buffer space is needed. Get it out of buffers on
+        * the "most free" list, placing the empty headers on the
+        * BQ_EMPTY buffer header list.
+        */
+       while (tp->b_bufsize < sizealloc) {
+               take = sizealloc - tp->b_bufsize;
+               bp = getnewbuf();
+               if (take >= bp->b_bufsize)
+                       take = bp->b_bufsize;
+               pagemove(&bp->b_un.b_addr[bp->b_bufsize - take],
+                   &tp->b_un.b_addr[tp->b_bufsize], take);
+               tp->b_bufsize += take;
+               bp->b_bufsize = bp->b_bufsize - take;
+               if (bp->b_bcount > bp->b_bufsize)
+                       bp->b_bcount = bp->b_bufsize;
+               if (bp->b_bufsize <= 0) {
+                       bremhash(bp);
+                       binshash(bp, &bfreelist[BQ_EMPTY]);
+                       bp->b_dev = NODEV;
+                       bp->b_error = 0;
+                       bp->b_flags |= B_INVAL;
                }
                }
-       panic("bunhash");
-ret:
-       splx(s);
+               brelse(bp);
+       }
+out:
+       tp->b_bcount = size;
+       return (1);
 }
 
 /*
 }
 
 /*
- * Wait for I/O completion on the buffer; return errors
- * to the user.
+ * Find a buffer which is available for use.
+ * Select something from a free list.
+ * Preference is to AGE list, then LRU list.
  */
  */
-iowait(bp)
-register struct buf *bp;
+struct buf *
+getnewbuf()
 {
 {
+       register struct buf *bp, *dp;
+       register struct ucred *cred;
+       int s;
 
 
-       (void) spl6();
-       while ((bp->b_flags&B_DONE)==0)
-               sleep((caddr_t)bp, PRIBIO);
-       (void) spl0();
-       geterror(bp);
+loop:
+       s = splbio();
+       for (dp = &bfreelist[BQ_AGE]; 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);
+               splx(s);
+               goto loop;
+       }
+       bp = dp->av_forw;
+       bremfree(bp);
+       bp->b_flags |= B_BUSY;
+       splx(s);
+       if (bp->b_flags & B_DELWRI) {
+               (void) bawrite(bp);
+               goto loop;
+       }
+       trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
+       if (bp->b_vp)
+               brelvp(bp);
+       if (bp->b_rcred != NOCRED) {
+               cred = bp->b_rcred;
+               bp->b_rcred = NOCRED;
+               crfree(cred);
+       }
+       if (bp->b_wcred != NOCRED) {
+               cred = bp->b_wcred;
+               bp->b_wcred = NOCRED;
+               crfree(cred);
+       }
+       bp->b_flags = B_BUSY;
+       bp->b_dirtyoff = bp->b_dirtyend = 0;
+       return (bp);
 }
 
 }
 
-#ifndef FASTVAX
 /*
 /*
- * Unlink a buffer from the available list and mark it busy.
- * (internal interface)
+ * Wait for I/O to complete.
+ *
+ * Extract and return any errors associated with the I/O.
+ * If the error flag is set, but no specific error is
+ * given, return EIO.
  */
  */
-notavail(bp)
-register struct buf *bp;
+biowait(bp)
+       register struct buf *bp;
 {
 {
-       register s;
+       int s;
 
 
-       s = spl6();
-       bp->av_back->av_forw = bp->av_forw;
-       bp->av_forw->av_back = bp->av_back;
-       bp->b_flags |= B_BUSY;
+       s = splbio();
+       while ((bp->b_flags & B_DONE) == 0)
+               sleep((caddr_t)bp, PRIBIO);
        splx(s);
        splx(s);
+       if ((bp->b_flags & B_ERROR) == 0)
+               return (0);
+       if (bp->b_error)
+               return (bp->b_error);
+       return (EIO);
 }
 }
-#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.
+ * Mark I/O complete on a buffer.
+ *
+ * If a callback has been requested, e.g. the pageout
+ * daemon, do so. Otherwise, awaken waiting processes.
  */
  */
-iodone(bp)
-register struct buf *bp;
+biodone(bp)
+       register struct buf *bp;
 {
 {
-       register int s;
+       register struct vnode *vp;
 
        if (bp->b_flags & B_DONE)
 
        if (bp->b_flags & B_DONE)
-               panic("dup iodone");
+               panic("dup biodone");
        bp->b_flags |= B_DONE;
        bp->b_flags |= B_DONE;
-       if (bp->b_flags & B_DIRTY) {
-               if (bp->b_flags & B_ERROR)
-                       panic("IO err in push");
-               s = spl6();
-               cnt.v_pgout++;
-               bp->av_forw = bclnlist;
-               bp->b_bcount = swsize[bp - swbuf];
-               bp->b_pfcent = swpf[bp - swbuf];
-               bclnlist = bp;
-               if (bswlist.b_flags & B_WANTED)
-                       wakeup((caddr_t)&proc[2]);
-               splx(s);
+       if ((bp->b_flags & B_READ) == 0) {
+               bp->b_dirtyoff = bp->b_dirtyend = 0;
+               if (vp = bp->b_vp) {
+                       vp->v_numoutput--;
+                       if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
+                               if (vp->v_numoutput < 0)
+                                       panic("biodone: neg numoutput");
+                               vp->v_flag &= ~VBWAIT;
+                               wakeup((caddr_t)&vp->v_numoutput);
+                       }
+               }
+       }
+       if (bp->b_flags & B_CALL) {
+               bp->b_flags &= ~B_CALL;
+               (*bp->b_iodone)(bp);
                return;
        }
                return;
        }
-       if (bp->b_flags&B_ASYNC)
+       if (bp->b_flags & B_ASYNC)
                brelse(bp);
        else {
                bp->b_flags &= ~B_WANTED;
                brelse(bp);
        else {
                bp->b_flags &= ~B_WANTED;
@@ -552,270 +581,250 @@ register struct buf *bp;
 }
 
 /*
 }
 
 /*
- * Zero the core associated with a buffer.
+ * Make sure all write-behind blocks associated
+ * with mount point are flushed out (from sync).
  */
  */
-clrbuf(bp)
-struct buf *bp;
+mntflushbuf(mountp, flags)
+       struct mount *mountp;
+       int flags;
 {
 {
-       register *p;
-       register c;
-
-       p = bp->b_un.b_words;
-       c = BSIZE/sizeof(int);
-       do
-               *p++ = 0;
-       while (--c);
-       bp->b_resid = 0;
+       register struct vnode *vp;
+
+       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("mntflushbuf: not busy");
+loop:
+       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
+               if (vget(vp))
+                       goto loop;
+               vflushbuf(vp, flags);
+               vput(vp);
+               if (vp->v_mount != mountp)
+                       goto loop;
+       }
 }
 
 /*
 }
 
 /*
- * 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.
+ * Flush all dirty buffers associated with a vnode.
  */
  */
-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;
+vflushbuf(vp, flags)
+       register struct vnode *vp;
+       int flags;
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-       register int c;
-       int p2dp;
-       register struct pte *dpte, *vpte;
-
-       (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->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;
-               }
-               (*bdevsw[major(dev)].d_strategy)(bp);
-               if (flag & B_DIRTY) {
-                       if (c < nbytes)
-                               panic("big push");
-                       return;
-               }
-               (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");
-                       swkill(p, (char *)0);
+       struct buf *nbp;
+       int s;
+
+loop:
+       s = splbio();
+       for (bp = vp->v_dirtyblkhd; bp; bp = nbp) {
+               nbp = bp->b_blockf;
+               if ((bp->b_flags & B_BUSY))
+                       continue;
+               if ((bp->b_flags & B_DELWRI) == 0)
+                       panic("vflushbuf: not dirty");
+               bremfree(bp);
+               bp->b_flags |= B_BUSY;
+               splx(s);
+               /*
+                * Wait for I/O associated with indirect blocks to complete,
+                * since there is no way to quickly wait for them below.
+                * NB: This is really specific to ufs, but is done here
+                * as it is easier and quicker.
+                */
+               if (bp->b_vp == vp || (flags & B_SYNC) == 0) {
+                       (void) bawrite(bp);
+                       s = splbio();
+               } else {
+                       (void) bwrite(bp);
+                       goto loop;
                }
                }
-               nbytes -= c;
-               dblkno += btoc(c);
        }
        }
-       (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;
-       if (bswlist.b_flags & B_WANTED) {
-               bswlist.b_flags &= ~B_WANTED;
-               wakeup((caddr_t)&bswlist);
-               wakeup((caddr_t)&proc[2]);
+       splx(s);
+       if ((flags & B_SYNC) == 0)
+               return;
+       s = splbio();
+       while (vp->v_numoutput) {
+               vp->v_flag |= VBWAIT;
+               sleep((caddr_t)&vp->v_numoutput, PRIBIO + 1);
+       }
+       splx(s);
+       if (vp->v_dirtyblkhd) {
+               vprint("vflushbuf: dirty", vp);
+               goto loop;
        }
        }
-       (void) spl0();
 }
 
 /*
 }
 
 /*
- * If rout == 0 then killed on swap error, else
- * rout is the name of the routine where we ran out of
- * swap space.
+ * Invalidate in core blocks belonging to closed or umounted filesystem
+ *
+ * Go through the list of vnodes associated with the file system;
+ * for each vnode invalidate any buffers that it holds. Normally
+ * this routine is preceeded by a bflush call, so that on a quiescent
+ * filesystem there will be no dirty buffers when we are done. Binval
+ * returns the count of dirty buffers when it is finished.
  */
  */
-swkill(p, rout)
-       struct proc *p;
-       char *rout;
+mntinvalbuf(mountp)
+       struct mount *mountp;
 {
 {
+       register struct vnode *vp;
+       int dirty = 0;
 
 
-       printf("%d: ", p->p_pid);
-       if (rout)
-               printf("out of swap space in %s\n", rout);
-       else
-               printf("killed on swap error\n");
-       /*
-        * 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;
+       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("mntinvalbuf: not busy");
+loop:
+       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
+               if (vget(vp))
+                       goto loop;
+               dirty += vinvalbuf(vp, 1);
+               vput(vp);
+               if (vp->v_mount != mountp)
+                       goto loop;
+       }
+       return (dirty);
 }
 
 /*
 }
 
 /*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
+ * Flush out and invalidate all buffers associated with a vnode.
+ * Called with the underlying object locked.
  */
  */
-bflush(dev)
-dev_t dev;
+vinvalbuf(vp, save)
+       register struct vnode *vp;
+       int save;
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-
-loop:
-       (void) spl6();
-       for (bp = bfreelist.av_forw; bp != &bfreelist; 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;
+       struct buf *nbp, *blist;
+       int s, dirty = 0;
+
+       for (;;) {
+               if (blist = vp->v_dirtyblkhd)
+                       /* void */;
+               else if (blist = vp->v_cleanblkhd)
+                       /* void */;
+               else
+                       break;
+               for (bp = blist; bp; bp = nbp) {
+                       nbp = bp->b_blockf;
+                       s = splbio();
+                       if (bp->b_flags & B_BUSY) {
+                               bp->b_flags |= B_WANTED;
+                               sleep((caddr_t)bp, PRIBIO + 1);
+                               splx(s);
+                               break;
+                       }
+                       bremfree(bp);
+                       bp->b_flags |= B_BUSY;
+                       splx(s);
+                       if (save && (bp->b_flags & B_DELWRI)) {
+                               dirty++;
+                               (void) bwrite(bp);
+                               break;
+                       }
+                       if (bp->b_vp != vp)
+                               reassignbuf(bp, bp->b_vp);
+                       else
+                               bp->b_flags |= B_INVAL;
+                       brelse(bp);
                }
        }
                }
        }
-       (void) spl0();
+       if (vp->v_dirtyblkhd || vp->v_cleanblkhd)
+               panic("vinvalbuf: flush failed");
+       return (dirty);
 }
 
 /*
 }
 
 /*
- * 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.
+ * Associate a buffer with a vnode.
  */
  */
-physio(strat, bp, dev, rw, mincnt)
-int (*strat)(); 
-register struct buf *bp;
-unsigned (*mincnt)();
+bgetvp(vp, bp)
+       register struct vnode *vp;
+       register struct buf *bp;
 {
 {
-       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_error==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_vp)
+               panic("bgetvp: not free");
+       VHOLD(vp);
+       bp->b_vp = vp;
+       if (vp->v_type == VBLK || vp->v_type == VCHR)
+               bp->b_dev = vp->v_rdev;
+       else
+               bp->b_dev = NODEV;
+       /*
+        * Insert onto list for new vnode.
+        */
+       if (vp->v_cleanblkhd) {
+               bp->b_blockf = vp->v_cleanblkhd;
+               bp->b_blockb = &vp->v_cleanblkhd;
+               vp->v_cleanblkhd->b_blockb = &bp->b_blockf;
+               vp->v_cleanblkhd = bp;
+       } else {
+               vp->v_cleanblkhd = bp;
+               bp->b_blockb = &vp->v_cleanblkhd;
+               bp->b_blockf = NULL;
        }
        }
-       bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
-       u.u_count = bp->b_resid;
-       geterror(bp);
-}
-
-/*ARGSUSED*/
-unsigned
-minphys(bp)
-struct buf *bp;
-{
-
-       if (bp->b_bcount > 60 * 1024)
-               bp->b_bcount = 60 * 1024;
 }
 
 /*
 }
 
 /*
- * 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.
+ * Disassociate a buffer from a vnode.
  */
  */
-geterror(bp)
-register struct buf *bp;
+brelvp(bp)
+       register struct buf *bp;
 {
 {
+       struct buf *bq;
+       struct vnode *vp;
 
 
-       if (bp->b_flags&B_ERROR)
-               if ((u.u_error = bp->b_error)==0)
-                       u.u_error = EIO;
+       if (bp->b_vp == (struct vnode *) 0)
+               panic("brelvp: NULL");
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_blockb) {
+               if (bq = bp->b_blockf)
+                       bq->b_blockb = bp->b_blockb;
+               *bp->b_blockb = bq;
+               bp->b_blockf = NULL;
+               bp->b_blockb = NULL;
+       }
+       vp = bp->b_vp;
+       bp->b_vp = (struct vnode *) 0;
+       HOLDRELE(vp);
 }
 
 /*
 }
 
 /*
- * 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
+ * Reassign a buffer from one vnode to another.
+ * Used to assign file specific control information
+ * (indirect blocks) to the vnode to which they belong.
  */
  */
-binval(dev)
-dev_t dev;
+reassignbuf(bp, newvp)
+       register struct buf *bp;
+       register struct vnode *newvp;
 {
 {
-       register struct buf *bp, *dp;
+       register struct buf *bq, **listheadp;
 
 
-       dp = bdevsw[major(dev)].d_tab;
-
-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-               if (bp->b_dev == dev)
-                       bp->b_flags |= B_INVAL;
+       if (newvp == NULL)
+               panic("reassignbuf: NULL");
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_blockb) {
+               if (bq = bp->b_blockf)
+                       bq->b_blockb = bp->b_blockb;
+               *bp->b_blockb = bq;
+       }
+       /*
+        * If dirty, put on list of dirty buffers;
+        * otherwise insert onto list of clean buffers.
+        */
+       if (bp->b_flags & B_DELWRI)
+               listheadp = &newvp->v_dirtyblkhd;
+       else
+               listheadp = &newvp->v_cleanblkhd;
+       if (*listheadp) {
+               bp->b_blockf = *listheadp;
+               bp->b_blockb = listheadp;
+               bp->b_blockf->b_blockb = &bp->b_blockf;
+               *listheadp = bp;
+       } else {
+               *listheadp = bp;
+               bp->b_blockb = listheadp;
+               bp->b_blockf = NULL;
+       }
 }
 }
Continuous source code history from original PDP-7 UNIX to FreeBSD 2, the first fully unencumbered FreeBSD release.