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 accb42b..94a8d27 100644 (file)
--- a/usr/src/sys/kern/vfs_cluster.c
+++ b/usr/src/sys/kern/vfs_cluster.c
@@ -1,206 +1,274 @@
-/*     vfs_cluster.c   4.45    83/05/18        */
-
-#include "../machine/pte.h"
+/*
+ * 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%
+ */
 
 
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/buf.h"
-#include "../h/conf.h"
-#include "../h/proc.h"
-#include "../h/seg.h"
-#include "../h/vm.h"
-#include "../h/trace.h"
+#include "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"
 
 /*
 
 /*
- * Read in (if necessary) the block and return a buffer pointer.
+ * 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.
  */
  */
-struct buf *
-bread(dev, blkno, size)
-       dev_t dev;
+bread(vp, blkno, size, cred, bpp)
+       struct vnode *vp;
        daddr_t blkno;
        int size;
        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;
 
        if (size == 0)
                panic("bread: size 0");
        register struct buf *bp;
 
        if (size == 0)
                panic("bread: size 0");
-       bp = getblk(dev, blkno, size);
-       if (bp->b_flags&B_DONE) {
-               trace(TR_BREADHIT, dev, blkno);
-               return(bp);
+#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;
        if (bp->b_bcount > bp->b_bufsize)
                panic("bread");
        }
        bp->b_flags |= B_READ;
        if (bp->b_bcount > bp->b_bufsize)
                panic("bread");
-       (*bdevsw[major(dev)].d_strategy)(bp);
-       trace(TR_BREADMISS, dev, blkno);
-       u.u_ru.ru_inblock++;            /* pay for read */
-       biowait(bp);
-       return(bp);
+       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, size, rablkno, rabsize)
-       dev_t dev;
+breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
+       struct vnode *vp;
        daddr_t blkno; int size;
        daddr_t blkno; int size;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
        daddr_t rablkno; int rabsize;
        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 the block isn't in core, then allocate
-        * a buffer and initiate i/o (getblk checks
-        * for a cache hit).
+        * If the block is not memory resident,
+        * allocate a buffer and start I/O.
         */
         */
-       if (!incore(dev, blkno)) {
-               bp = getblk(dev, blkno, size);
-               if ((bp->b_flags&B_DONE) == 0) {
+       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;
                        if (bp->b_bcount > bp->b_bufsize)
                                panic("breada");
                        bp->b_flags |= B_READ;
                        if (bp->b_bcount > bp->b_bufsize)
                                panic("breada");
-                       (*bdevsw[major(dev)].d_strategy)(bp);
-                       trace(TR_BREADMISS, dev, blkno);
-                       u.u_ru.ru_inblock++;            /* pay for read */
+                       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
                } else
-                       trace(TR_BREADHIT, dev, blkno);
+                       trace(TR_BREADHIT, pack(vp, size), blkno);
        }
 
        /*
        }
 
        /*
-        * If there's a read-ahead block, start i/o
-        * on it also (as above).
+        * If there is a read-ahead block, start I/O on it too.
         */
         */
-       if (rablkno && !incore(dev, rablkno)) {
-               rabp = getblk(dev, rablkno, rabsize);
-               if (rabp->b_flags & B_DONE) {
+       if (!incore(vp, rablkno)) {
+               rabp = getblk(vp, rablkno, rabsize);
+#endif SECSIZE
+               if (rabp->b_flags & (B_DONE | B_DELWRI)) {
                        brelse(rabp);
                        brelse(rabp);
-                       trace(TR_BREADHITRA, dev, blkno);
+                       trace(TR_BREADHITRA, pack(vp, rabsize), rablkno);
                } else {
                } else {
-                       rabp->b_flags |= B_READ|B_ASYNC;
+                       rabp->b_flags |= B_ASYNC | B_READ;
                        if (rabp->b_bcount > rabp->b_bufsize)
                                panic("breadrabp");
                        if (rabp->b_bcount > rabp->b_bufsize)
                                panic("breadrabp");
-                       (*bdevsw[major(dev)].d_strategy)(rabp);
-                       trace(TR_BREADMISSRA, dev, rablock);
-                       u.u_ru.ru_inblock++;            /* pay in advance */
+                       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 block was in core, let bread get it.
-        * If block wasn't in core, then the read was started
-        * above, and just wait for it.
+        * 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)
         */
        if (bp == NULL)
-               return (bread(dev, blkno, size));
-       biowait(bp);
-       return (bp);
+#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)
        register struct buf *bp;
 {
  */
 bwrite(bp)
        register struct buf *bp;
 {
-       register flag;
+       struct proc *p = curproc;               /* XXX */
+       register int flag;
+       int s, error;
 
        flag = bp->b_flags;
        bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
 
        flag = bp->b_flags;
        bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
-       if ((flag&B_DELWRI) == 0)
-               u.u_ru.ru_oublock++;            /* noone paid yet */
-       trace(TR_BWRITE, bp->b_dev, bp->b_blkno);
+       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");
        if (bp->b_bcount > bp->b_bufsize)
                panic("bwrite");
-       (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
+       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 synchronous, then await I/O completion.
         * If the write was "delayed", then we put the buffer on
         * If the write was "delayed", then we put the buffer on
-        * the q of blocks awaiting i/o completion status.
+        * the queue of blocks awaiting I/O completion status.
         */
         */
-       if ((flag&B_ASYNC) == 0) {
-               biowait(bp);
+       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;
+               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)
        register struct buf *bp;
 {
  */
 bdwrite(bp)
        register struct buf *bp;
 {
-       register int flags;
+       struct proc *p = curproc;               /* XXX */
 
 
-       if ((bp->b_flags&B_DELWRI) == 0)
-               u.u_ru.ru_oublock++;            /* noone paid yet */
-       flags = bdevsw[major(bp->b_dev)].d_flags;
-       if(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)
        register struct buf *bp;
 {
 
  */
 bawrite(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)
        register struct buf *bp;
 {
        register struct buf *flist;
  */
 brelse(bp)
        register struct buf *bp;
 {
        register struct buf *flist;
-       register s;
+       int s;
 
 
+       trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
        /*
        /*
-        * If someone's waiting for the buffer, or
-        * is waiting for a buffer wake 'em up.
+        * If a process is waiting for the buffer, or
+        * is waiting for a free buffer, awaken it.
         */
         */
-       if (bp->b_flags&B_WANTED)
+       if (bp->b_flags & B_WANTED)
                wakeup((caddr_t)bp);
                wakeup((caddr_t)bp);
-       if (bfreelist[0].b_flags&B_WANTED) {
+       if (bfreelist[0].b_flags & B_WANTED) {
                bfreelist[0].b_flags &= ~B_WANTED;
                wakeup((caddr_t)bfreelist);
        }
                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 */
-
+       /*
+        * 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;
+       }
        /*
         * Stick the buffer back on a free list.
         */
        /*
         * Stick the buffer back on a free list.
         */
-       s = spl6();
+       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);
        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)) {
+       } 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);
                /* block has no info ... put at front of most free list */
                flist = &bfreelist[BQ_AGE];
                binsheadfree(bp, flist);
@@ -213,103 +281,99 @@ brelse(bp)
                        flist = &bfreelist[BQ_LRU];
                binstailfree(bp, flist);
        }
                        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;
+incore(vp, blkno)
+       struct vnode *vp;
        daddr_t blkno;
 {
        register struct buf *bp;
        register struct buf *dp;
 
        daddr_t blkno;
 {
        register struct buf *bp;
        register struct buf *dp;
 
-       dp = BUFHASH(dev, blkno);
+       dp = BUFHASH(vp, blkno);
        for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
        for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-               if (bp->b_blkno == blkno && bp->b_dev == dev &&
+               if (bp->b_lblkno == blkno && bp->b_vp == vp &&
                    (bp->b_flags & B_INVAL) == 0)
                        return (1);
        return (0);
 }
 
                    (bp->b_flags & B_INVAL) == 0)
                        return (1);
        return (0);
 }
 
-struct buf *
-baddr(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       int size;
-{
-
-       if (incore(dev, blkno))
-               return (bread(dev, blkno, size));
-       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.
- *
- * We use splx here because this routine may be called
- * on the interrupt stack during a dump, and we don't
- * want to lower the ipl back to 0.
+ * 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, size)
-       dev_t dev;
+#ifdef SECSIZE
+getblk(dev, blkno, size, secsize)
+#else SECSIZE
+getblk(vp, blkno, size)
+       register struct vnode *vp;
        daddr_t blkno;
        int size;
        daddr_t blkno;
        int size;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
 {
        register struct buf *bp, *dp;
        int s;
 
 {
        register struct buf *bp, *dp;
        int s;
 
-       if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT)) /* XXX */
-               blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1);
+       if (size > MAXBSIZE)
+               panic("getblk: size too big");
        /*
        /*
-        * Search the cache for the block.  If we hit, but
-        * the buffer is in use for i/o, then we wait until
-        * the i/o has completed.
+        * 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(dev, blkno);
+       dp = BUFHASH(vp, blkno);
 loop:
        for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
 loop:
        for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
-               if (bp->b_blkno != blkno || bp->b_dev != dev ||
-                   bp->b_flags&B_INVAL)
+               if (bp->b_lblkno != blkno || bp->b_vp != vp ||
+                   (bp->b_flags & B_INVAL))
                        continue;
                        continue;
-               s = 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;
                }
                        splx(s);
                        goto loop;
                }
+               bremfree(bp);
+               bp->b_flags |= B_BUSY;
                splx(s);
                splx(s);
-               notavail(bp);
-               if (brealloc(bp, size) == 0)
+               if (bp->b_bcount != size) {
+                       printf("getblk: stray size");
+                       bp->b_flags |= B_INVAL;
+                       bwrite(bp);
                        goto loop;
                        goto loop;
+               }
                bp->b_flags |= B_CACHE;
                bp->b_flags |= B_CACHE;
-               return(bp);
+               return (bp);
        }
        }
-       if (major(dev) >= nblkdev)
-               panic("blkdev");
        bp = getnewbuf();
        bp = getnewbuf();
-       bfree(bp);
        bremhash(bp);
        bremhash(bp);
-       binshash(bp, dp);
-       bp->b_dev = dev;
+       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_blkno = blkno;
        bp->b_error = 0;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
+       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 *
 geteblk(size)
  */
 struct buf *
 geteblk(size)
@@ -317,116 +381,40 @@ geteblk(size)
 {
        register struct buf *bp, *flist;
 
 {
        register struct buf *bp, *flist;
 
-loop:
+       if (size > MAXBSIZE)
+               panic("geteblk: size too big");
        bp = getnewbuf();
        bp->b_flags |= B_INVAL;
        bp = getnewbuf();
        bp->b_flags |= B_INVAL;
-       bfree(bp);
        bremhash(bp);
        flist = &bfreelist[BQ_AGE];
        bremhash(bp);
        flist = &bfreelist[BQ_AGE];
-       binshash(bp, flist);
-       bp->b_dev = (dev_t)NODEV;
+       bp->b_bcount = 0;
+#ifdef SECSIZE
+       bp->b_blksize = DEV_BSIZE;
+#endif SECSIZE
        bp->b_error = 0;
        bp->b_error = 0;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
-}
-
-/*
- * Allocate space associated with a buffer.
- * If can't get space, buffer is released
- */
-brealloc(bp, size)
-       register struct buf *bp;
-       int size;
-{
-       daddr_t start, last;
-       register struct buf *ep;
-       struct buf *dp;
-       int s;
-
-       /*
-        * First need to make sure that all overlaping previous I/O
-        * is dispatched with.
-        */
-       if (size == bp->b_bcount)
-               return (1);
-       if (size < bp->b_bcount) { 
-               if (bp->b_flags & B_DELWRI) {
-                       bwrite(bp);
-                       return (0);
-               }
-               if (bp->b_flags & B_LOCKED)
-                       panic("brealloc");
-               return (allocbuf(bp, size));
-       }
-       bp->b_flags &= ~B_DONE;
-       if (bp->b_dev == NODEV)
-               return (allocbuf(bp, size));
-
-       /*
-        * Search cache for any buffers that overlap the one that we
-        * are trying to allocate. Overlapping buffers must be marked
-        * invalid, after being written out if they are dirty. (indicated
-        * by B_DELWRI) A disk block must be mapped by at most one buffer
-        * at any point in time. Care must be taken to avoid deadlocking
-        * when two buffer are trying to get the same set of disk blocks.
-        */
-       start = bp->b_blkno;
-       last = start + (size / DEV_BSIZE) - 1;
-       dp = BUFHASH(bp->b_dev, bp->b_blkno);
-loop:
-       for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
-               if (ep == bp || ep->b_dev != bp->b_dev || (ep->b_flags&B_INVAL))
-                       continue;
-               /* look for overlap */
-               if (ep->b_bcount == 0 || ep->b_blkno > last ||
-                   ep->b_blkno + (ep->b_bcount / DEV_BSIZE) <= start)
-                       continue;
-               s = spl6();
-               if (ep->b_flags&B_BUSY) {
-                       ep->b_flags |= B_WANTED;
-                       sleep((caddr_t)ep, PRIBIO+1);
-                       splx(s);
-                       goto loop;
-               }
-               splx(s);
-               notavail(ep);
-               if (ep->b_flags & B_DELWRI) {
-                       bwrite(ep);
-                       goto loop;
-               }
-               ep->b_flags |= B_INVAL;
-               brelse(ep);
-       }
-       return (allocbuf(bp, size));
+       bp->b_resid = 0;
+       binshash(bp, flist);
+       allocbuf(bp, size);
+       return (bp);
 }
 
 /*
  * Expand or contract the actual memory allocated to a buffer.
 }
 
 /*
  * Expand or contract the actual memory allocated to a buffer.
- * If no memory is available, release buffer and take error exit
+ * If no memory is available, release buffer and take error exit.
  */
 allocbuf(tp, size)
        register struct buf *tp;
        int size;
 {
        register struct buf *bp, *ep;
  */
 allocbuf(tp, size)
        register struct buf *tp;
        int size;
 {
        register struct buf *bp, *ep;
-       int sizealloc, take;
-#ifdef sun
-       register char *a;
-       int osize;
-#endif
+       int sizealloc, take, s;
 
 
-#ifndef sun
        sizealloc = roundup(size, CLBYTES);
        sizealloc = roundup(size, CLBYTES);
-#else
-       sizealloc = roundup(size, BUFALLOCSIZE);
-#endif
        /*
         * Buffer size does not change
         */
        if (sizealloc == tp->b_bufsize)
                goto out;
        /*
         * Buffer size does not change
         */
        if (sizealloc == tp->b_bufsize)
                goto out;
-#ifndef sun
        /*
         * Buffer size is shrinking.
         * Place excess space in a buffer header taken from the
        /*
         * Buffer size is shrinking.
         * Place excess space in a buffer header taken from the
@@ -438,7 +426,10 @@ allocbuf(tp, size)
                ep = bfreelist[BQ_EMPTY].av_forw;
                if (ep == &bfreelist[BQ_EMPTY])
                        goto out;
                ep = bfreelist[BQ_EMPTY].av_forw;
                if (ep == &bfreelist[BQ_EMPTY])
                        goto out;
-               notavail(ep);
+               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;
                pagemove(tp->b_un.b_addr + sizealloc, ep->b_un.b_addr,
                    (int)tp->b_bufsize - sizealloc);
                ep->b_bufsize = tp->b_bufsize - sizealloc;
@@ -467,115 +458,17 @@ allocbuf(tp, size)
                if (bp->b_bufsize <= 0) {
                        bremhash(bp);
                        binshash(bp, &bfreelist[BQ_EMPTY]);
                if (bp->b_bufsize <= 0) {
                        bremhash(bp);
                        binshash(bp, &bfreelist[BQ_EMPTY]);
-                       bp->b_dev = (dev_t)NODEV;
+                       bp->b_dev = NODEV;
                        bp->b_error = 0;
                        bp->b_flags |= B_INVAL;
                }
                brelse(bp);
        }
                        bp->b_error = 0;
                        bp->b_flags |= B_INVAL;
                }
                brelse(bp);
        }
-#else
-       /*
-        * Buffer size is shrinking
-        * Just put the tail end back in the map
-        */
-       if (sizealloc < tp->b_bufsize) {
-               rmfree(buffermap, (long)(tp->b_bufsize - sizealloc),
-                       (long)(tp->b_un.b_addr + sizealloc));
-               tp->b_bufsize = sizealloc;
-               goto out;
-       }
-       /*
-        * Buffer is being expanded or created
-        * If being expanded, attempt to get contiguous
-        * section, otherwise get a new chunk and copy.
-        * If no space, free up a buffer on the AGE list
-        * and try again.
-        */
-       do {
-               if ((osize = tp->b_bufsize)) {
-                       a = (char *)rmget(buffermap, (long)(sizealloc-osize),
-                               (long)(tp->b_un.b_addr + osize));
-                       if (a == 0) {
-                               a = (char *)rmalloc(buffermap, (long)sizealloc);
-                               if (a != 0) {
-                                       bcopy(tp->b_un.b_addr, a, osize);
-                                       rmfree(buffermap, (long)osize,
-                                               (long)tp->b_un.b_addr);
-                                       tp->b_un.b_addr = a;
-                               }
-                       }
-               } else {
-                       a = (char *)rmalloc(buffermap, (long)sizealloc);
-                       if (a != 0)
-                               tp->b_un.b_addr = a;
-               }
-       } while (a == 0 && bfreemem());
-       if (a == 0) {
-               brelse(tp);
-               return (0);
-       }
-       tp->b_bufsize = sizealloc;
-#endif
 out:
        tp->b_bcount = size;
        return (1);
 }
 
 out:
        tp->b_bcount = size;
        return (1);
 }
 
-/*
- * Release space associated with a buffer.
- */
-bfree(bp)
-       struct buf *bp;
-{
-#ifdef sun
-       if (bp->b_bufsize) {
-               rmfree(buffermap, (long)bp->b_bufsize, (long)bp->b_un.b_addr);
-               bp->b_bufsize = 0;
-       }
-#endif
-       bp->b_bcount = 0;
-}
-
-#ifdef sun
-/*
- * Attempt to free up buffer space by flushing
- * something in the free list.
- * Don't wait for something, that could cause deadlocks
- * We start with BQ_AGE because we know BQ_EMPTY take no memory.
- */
-bfreemem()
-{
-       register struct buf *bp, *dp;
-       int s;
-
-loop:
-       s = spl6();
-       for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
-               if (dp->av_forw != dp)
-                       break;
-       splx(s);
-       if (dp == bfreelist) {          /* no free blocks */
-               return (0);
-       }
-       bp = dp->av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-       trace(TR_BRELSE, bp->b_dev, bp->b_blkno);
-       bp->b_flags = B_BUSY | B_INVAL;
-       bfree(bp);
-       bremhash(bp);
-       binshash(bp, &bfreelist[BQ_EMPTY]);
-       bp->b_dev = (dev_t)NODEV;
-       bp->b_error = 0;
-       brelse(bp);
-       return (1);
-}
-#endif
-
 /*
  * Find a buffer which is available for use.
  * Select something from a free list.
 /*
  * Find a buffer which is available for use.
  * Select something from a free list.
@@ -585,74 +478,101 @@ struct buf *
 getnewbuf()
 {
        register struct buf *bp, *dp;
 getnewbuf()
 {
        register struct buf *bp, *dp;
+       register struct ucred *cred;
        int s;
 
 loop:
        int s;
 
 loop:
-       s = spl6();
-#ifndef sun
+       s = splbio();
        for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
        for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
-#else
-       for (dp = &bfreelist[BQ_EMPTY]; dp > bfreelist; dp--)
-#endif
                if (dp->av_forw != dp)
                        break;
        if (dp == bfreelist) {          /* no free blocks */
                dp->b_flags |= B_WANTED;
                if (dp->av_forw != dp)
                        break;
        if (dp == bfreelist) {          /* no free blocks */
                dp->b_flags |= B_WANTED;
-               sleep((caddr_t)dp, PRIBIO+1);
+               sleep((caddr_t)dp, PRIBIO + 1);
                splx(s);
                goto loop;
        }
                splx(s);
                goto loop;
        }
-       splx(s);
        bp = dp->av_forw;
        bp = dp->av_forw;
-       notavail(bp);
+       bremfree(bp);
+       bp->b_flags |= B_BUSY;
+       splx(s);
        if (bp->b_flags & B_DELWRI) {
        if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
+               (void) bawrite(bp);
                goto loop;
        }
                goto loop;
        }
-       trace(TR_BRELSE, bp->b_dev, bp->b_blkno);
+       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_flags = B_BUSY;
+       bp->b_dirtyoff = bp->b_dirtyend = 0;
        return (bp);
 }
 
 /*
        return (bp);
 }
 
 /*
- * Wait for I/O completion on the buffer; return errors
- * to the user.
+ * 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.
  */
 biowait(bp)
        register struct buf *bp;
 {
        int s;
 
  */
 biowait(bp)
        register struct buf *bp;
 {
        int s;
 
-       s = spl6();
-       while ((bp->b_flags&B_DONE)==0)
+       s = splbio();
+       while ((bp->b_flags & B_DONE) == 0)
                sleep((caddr_t)bp, PRIBIO);
        splx(s);
                sleep((caddr_t)bp, PRIBIO);
        splx(s);
-       if (u.u_error == 0)                     /* XXX */
-               u.u_error = geterror(bp);
+       if ((bp->b_flags & B_ERROR) == 0)
+               return (0);
+       if (bp->b_error)
+               return (bp->b_error);
+       return (EIO);
 }
 
 /*
 }
 
 /*
- * 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.
  */
 biodone(bp)
        register struct buf *bp;
 {
  */
 biodone(bp)
        register struct buf *bp;
 {
+       register struct vnode *vp;
 
        if (bp->b_flags & B_DONE)
                panic("dup biodone");
        bp->b_flags |= B_DONE;
 
        if (bp->b_flags & B_DONE)
                panic("dup biodone");
        bp->b_flags |= B_DONE;
+       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;
        }
        if (bp->b_flags & B_CALL) {
                bp->b_flags &= ~B_CALL;
                (*bp->b_iodone)(bp);
                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;
@@ -661,114 +581,250 @@ biodone(bp)
 }
 
 /*
 }
 
 /*
- * Insure that no part of a specified block is in an incore buffer.
+ * Make sure all write-behind blocks associated
+ * with mount point are flushed out (from sync).
  */
  */
-blkflush(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       long size;
+mntflushbuf(mountp, flags)
+       struct mount *mountp;
+       int flags;
 {
 {
-       register struct buf *ep;
-       struct buf *dp;
-       daddr_t start, last;
-       int s;
+       register struct vnode *vp;
 
 
-       start = blkno;
-       last = start + (size / DEV_BSIZE) - 1;
-       dp = BUFHASH(dev, blkno);
+       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("mntflushbuf: not busy");
 loop:
 loop:
-       for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
-               if (ep->b_dev != dev || (ep->b_flags&B_INVAL))
-                       continue;
-               /* look for overlap */
-               if (ep->b_bcount == 0 || ep->b_blkno > last ||
-                   ep->b_blkno + (ep->b_bcount / DEV_BSIZE) <= start)
-                       continue;
-               s = spl6();
-               if (ep->b_flags&B_BUSY) {
-                       ep->b_flags |= B_WANTED;
-                       sleep((caddr_t)ep, PRIBIO+1);
-                       splx(s);
+       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
+               if (vget(vp))
                        goto loop;
                        goto loop;
-               }
-               if (ep->b_flags & B_DELWRI) {
-                       splx(s);
-                       notavail(ep);
-                       bwrite(ep);
+               vflushbuf(vp, flags);
+               vput(vp);
+               if (vp->v_mount != mountp)
                        goto loop;
                        goto loop;
-               }
-               splx(s);
        }
 }
 
 /*
        }
 }
 
 /*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
- * (and temporarily pagein)
+ * Flush all dirty buffers associated with a vnode.
  */
  */
-bflush(dev)
-       dev_t dev;
+vflushbuf(vp, flags)
+       register struct vnode *vp;
+       int flags;
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-       register struct buf *flist;
+       struct buf *nbp;
        int s;
 
 loop:
        int s;
 
 loop:
-       s = spl6();
-       for (flist = bfreelist; flist < &bfreelist[BQ_EMPTY]; flist++)
-       for (bp = flist->av_forw; bp != flist; bp = bp->av_forw) {
-               if ((bp->b_flags & B_DELWRI) == 0)
+       s = splbio();
+       for (bp = vp->v_dirtyblkhd; bp; bp = nbp) {
+               nbp = bp->b_blockf;
+               if ((bp->b_flags & B_BUSY))
                        continue;
                        continue;
-               if (dev == NODEV || dev == bp->b_dev) {
-                       bp->b_flags |= B_ASYNC;
-                       notavail(bp);
-                       bwrite(bp);
-                       splx(s);
+               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;
                }
        }
        splx(s);
                        goto loop;
                }
        }
        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;
+       }
+}
+
+/*
+ * 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.
+ */
+mntinvalbuf(mountp)
+       struct mount *mountp;
+{
+       register struct vnode *vp;
+       int dirty = 0;
+
+       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);
 }
 
 /*
 }
 
 /*
- * 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.
+ * Flush out and invalidate all buffers associated with a vnode.
+ * Called with the underlying object locked.
  */
  */
-geterror(bp)
+vinvalbuf(vp, save)
+       register struct vnode *vp;
+       int save;
+{
+       register struct buf *bp;
+       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);
+               }
+       }
+       if (vp->v_dirtyblkhd || vp->v_cleanblkhd)
+               panic("vinvalbuf: flush failed");
+       return (dirty);
+}
+
+/*
+ * Associate a buffer with a vnode.
+ */
+bgetvp(vp, bp)
+       register struct vnode *vp;
        register struct buf *bp;
 {
        register struct buf *bp;
 {
-       int error = 0;
 
 
-       if (bp->b_flags&B_ERROR)
-               if ((error = bp->b_error)==0)
-                       return (EIO);
-       return (error);
+       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;
+       }
 }
 
 /*
 }
 
 /*
- * 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
+ * Disassociate a buffer from a vnode.
  */
  */
-binval(dev)
-       dev_t dev;
+brelvp(bp)
+       register struct buf *bp;
 {
 {
+       struct buf *bq;
+       struct vnode *vp;
+
+       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);
+}
+
+/*
+ * Reassign a buffer from one vnode to another.
+ * Used to assign file specific control information
+ * (indirect blocks) to the vnode to which they belong.
+ */
+reassignbuf(bp, newvp)
        register struct buf *bp;
        register struct buf *bp;
-       register struct bufhd *hp;
-#define dp ((struct buf *)hp)
+       register struct vnode *newvp;
+{
+       register struct buf *bq, **listheadp;
 
 
-       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;
+       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.