add USL's copyright notice
[unix-history] / usr / src / sys / kern / vfs_bio.c
index 8550de5..43123da 100644 (file)
-/*     vfs_bio.c       4.37    82/10/17        */
-
-#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"
-
-int bioprintfs = 0;
+/*-
+ * Copyright (c) 1986, 1989, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ * (c) UNIX System Laboratories, Inc.
+ * All or some portions of this file are derived from material licensed
+ * to the University of California by American Telephone and Telegraph
+ * Co. or Unix System Laboratories, Inc. and are reproduced herein with
+ * the permission of UNIX System Laboratories, Inc.
+ *
+ * This code is derived from software contributed to Berkeley by
+ * Berkeley Software Design Inc.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)vfs_bio.c   8.7 (Berkeley) %G%
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/buf.h>
+#include <sys/vnode.h>
+#include <sys/mount.h>
+#include <sys/trace.h>
+#include <sys/malloc.h>
+#include <sys/resourcevar.h>
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
 
 /*
 
 /*
- * Read in (if necessary) the block and return a buffer pointer.
+ * Definitions for the buffer hash lists.
  */
  */
-struct buf *
-bread(dev, blkno, size)
-       dev_t dev;
+#define        BUFHASH(dvp, lbn)       \
+       (&bufhashtbl[((int)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
+LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
+u_long bufhash;
+
+/*
+ * Insq/Remq for the buffer hash lists.
+ */
+#define        binshash(bp, dp)        LIST_INSERT_HEAD(dp, bp, b_hash)
+#define        bremhash(bp)            LIST_REMOVE(bp, b_hash)
+
+/*
+ * Definitions for the buffer free lists.
+ */
+#define        BQUEUES         4               /* number of free buffer queues */
+
+#define        BQ_LOCKED       0               /* super-blocks &c */
+#define        BQ_LRU          1               /* lru, useful buffers */
+#define        BQ_AGE          2               /* rubbish */
+#define        BQ_EMPTY        3               /* buffer headers with no memory */
+
+TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
+int needbuffer;
+
+/*
+ * Insq/Remq for the buffer free lists.
+ */
+#define        binsheadfree(bp, dp)    TAILQ_INSERT_HEAD(dp, bp, b_freelist)
+#define        binstailfree(bp, dp)    TAILQ_INSERT_TAIL(dp, bp, b_freelist)
+
+void
+bremfree(bp)
+       struct buf *bp;
+{
+       struct bqueues *dp = NULL;
+
+       /*
+        * We only calculate the head of the freelist when removing
+        * the last element of the list as that is the only time that
+        * it is needed (e.g. to reset the tail pointer).
+        *
+        * NB: This makes an assumption about how tailq's are implemented.
+        */
+       if (bp->b_freelist.tqe_next == NULL) {
+               for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
+                       if (dp->tqh_last == &bp->b_freelist.tqe_next)
+                               break;
+               if (dp == &bufqueues[BQUEUES])
+                       panic("bremfree: lost tail");
+       }
+       TAILQ_REMOVE(dp, bp, b_freelist);
+}
+
+/*
+ * Initialize buffers and hash links for buffers.
+ */
+void
+bufinit()
+{
+       register struct buf *bp;
+       struct bqueues *dp;
+       register int i;
+       int base, residual;
+
+       for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
+               TAILQ_INIT(dp);
+       bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash);
+       base = bufpages / nbuf;
+       residual = bufpages % nbuf;
+       for (i = 0; i < nbuf; i++) {
+               bp = &buf[i];
+               bzero((char *)bp, sizeof *bp);
+               bp->b_dev = NODEV;
+               bp->b_rcred = NOCRED;
+               bp->b_wcred = NOCRED;
+               bp->b_vnbufs.le_next = NOLIST;
+               bp->b_data = buffers + i * MAXBSIZE;
+               if (i < residual)
+                       bp->b_bufsize = (base + 1) * CLBYTES;
+               else
+                       bp->b_bufsize = base * CLBYTES;
+               bp->b_flags = B_INVAL;
+               dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY];
+               binsheadfree(bp, dp);
+               binshash(bp, &invalhash);
+       }
+}
+
+/*
+ * 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.
+ */
+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;
 
        register struct buf *bp;
 
-       bp = getblk(dev, blkno, size);
-       if (bp->b_flags&B_DONE) {
-               trace(TR_BREADHIT, dev, blkno);
-               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, 0, 0);
+#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;
-       (*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_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 N specified
+ * read-ahead blocks.
  */
  */
-struct buf *
-breada(dev, blkno, size, rablkno, rabsize)
-       dev_t dev;
+breadn(vp, blkno, size, rablkno, rabsize, num, cred, bpp)
+       struct vnode *vp;
        daddr_t blkno; int size;
        daddr_t blkno; int size;
-       daddr_t rablkno; int rabsize;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
+       daddr_t rablkno[]; int rabsize[];
+       int num;
+       struct ucred *cred;
+       struct buf **bpp;
 {
 {
+       struct proc *p = curproc;               /* XXX */
        register struct buf *bp, *rabp;
        register struct buf *bp, *rabp;
+       register int i;
 
        bp = NULL;
        /*
 
        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, 0, 0);
+#endif SECSIZE
+               if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) {
                        bp->b_flags |= B_READ;
                        bp->b_flags |= B_READ;
-                       (*bdevsw[major(dev)].d_strategy)(bp);
-                       trace(TR_BREADMISS, dev, blkno);
-                       u.u_ru.ru_inblock++;            /* pay for read */
-               } else
-                       trace(TR_BREADHIT, dev, blkno);
+                       if (bp->b_bcount > bp->b_bufsize)
+                               panic("breadn");
+                       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 there's a read-ahead block, start i/o
-        * on it also (as above).
+        * If there's read-ahead block(s), start I/O
+        * on them also (as above).
         */
         */
-       if (rablkno && !incore(dev, rablkno)) {
-               rabp = getblk(dev, rablkno, rabsize);
-               if (rabp->b_flags & B_DONE) {
+       for (i = 0; i < num; i++) {
+               if (incore(vp, rablkno[i]))
+                       continue;
+               rabp = getblk(vp, rablkno[i], rabsize[i], 0, 0);
+#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[i]), rablkno[i]);
                } else {
                } else {
-                       rabp->b_flags |= B_READ|B_ASYNC;
-                       (*bdevsw[major(dev)].d_strategy)(rabp);
-                       trace(TR_BREADMISSRA, dev, rablock);
-                       u.u_ru.ru_inblock++;            /* 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[i]), rablkno[i]);
+                       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 = 0;
 
 
+       if (bp->b_vp && (bp->b_vp->v_mount->mnt_flag & MNT_ASYNC))
+               bp->b_flags |= B_ASYNC;
        flag = bp->b_flags;
        flag = bp->b_flags;
-       bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI | B_AGE);
-       if ((flag&B_DELWRI) == 0)
-               u.u_ru.ru_oublock++;            /* noone paid yet */
-       trace(TR_BWRITE, bp->b_dev, bp->b_blkno);
-if (bioprintfs)
-printf("write %x blk %d count %d\n", bp->b_dev, bp->b_blkno, bp->b_bcount);
-       (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
+       bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
+       if (flag & B_ASYNC) {
+               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++;
+       bp->b_flags |= B_WRITEINPROG;
+       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.
-        * Otherwise, the i/o must be finished and we check for
-        * an error.
+        * the queue of blocks awaiting I/O completion status.
         */
         */
-       if ((flag&B_ASYNC) == 0) {
-               biowait(bp);
+       if ((flag & B_ASYNC) == 0) {
+               error = biowait(bp);
+               if ((flag&B_DELWRI) == 0)
+                       p->p_stats->p_ru.ru_oublock++;  /* no one paid yet */
+               else
+                       reassignbuf(bp, bp->b_vp);
+               if (bp->b_flags & B_EINTR) {
+                       bp->b_flags &= ~B_EINTR;
+                       error = EINTR;
+               }
                brelse(bp);
                brelse(bp);
-       } else if (flag & B_DELWRI)
+       } else if (flag & B_DELWRI) {
+               s = splbio();
                bp->b_flags |= B_AGE;
                bp->b_flags |= B_AGE;
-       else
-               u.u_error = geterror(bp);
+               splx(s);
+       }
+       return (error);
+}
+
+int
+vn_bwrite(ap)
+       struct vop_bwrite_args *ap;
+{
+       
+       return (bwrite(ap->a_bp));
 }
 
 }
 
+
 /*
 /*
- * Release the buffer, marking it so that if it is grabbed
- * for another purpose it will be written out before being
- * given up (e.g. when writing a partial block where it is
- * assumed that another write for the same block will soon follow).
- * This can't be done for magtape, since writes must be done
- * in the same order as requested.
+ * 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) VOP_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;
 {
  */
 brelse(bp)
        register struct buf *bp;
 {
-       register struct buf *flist;
-       register s;
+       register struct bqueues *flist;
+       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) {
-               bfreelist[0].b_flags &= ~B_WANTED;
-               wakeup((caddr_t)bfreelist);
+       if (needbuffer) {
+               needbuffer = 0;
+               wakeup((caddr_t)&needbuffer);
+       }
+       /*
+        * Retry I/O for locked buffers rather than invalidating them.
+        */
+       s = splbio();
+       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;
        }
        }
-       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 */
-
        /*
         * Stick the buffer back on a free list.
         */
        /*
         * Stick the buffer back on a free list.
         */
-       s = spl6();
-       if (bp->b_flags & (B_ERROR|B_INVAL)) {
+       if (bp->b_bufsize <= 0) {
+               /* block has no buffer ... put at front of unused buffer list */
+               flist = &bufqueues[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 */
                /* block has no info ... put at front of most free list */
-               flist = &bfreelist[BQUEUES-1];
+               flist = &bufqueues[BQ_AGE];
                binsheadfree(bp, flist);
        } else {
                if (bp->b_flags & B_LOCKED)
                binsheadfree(bp, flist);
        } else {
                if (bp->b_flags & B_LOCKED)
-                       flist = &bfreelist[BQ_LOCKED];
+                       flist = &bufqueues[BQ_LOCKED];
                else if (bp->b_flags & B_AGE)
                else if (bp->b_flags & B_AGE)
-                       flist = &bfreelist[BQ_AGE];
+                       flist = &bufqueues[BQ_AGE];
                else
                else
-                       flist = &bfreelist[BQ_LRU];
+                       flist = &bufqueues[BQ_LRU];
                binstailfree(bp, flist);
        }
                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;
+struct buf *
+incore(vp, blkno)
+       struct vnode *vp;
        daddr_t blkno;
 {
        register struct buf *bp;
        daddr_t blkno;
 {
        register struct buf *bp;
-       register struct buf *dp;
 
 
-       dp = BUFHASH(dev, blkno);
-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-               if (bp->b_blkno == blkno && bp->b_dev == dev &&
+       for (bp = BUFHASH(vp, blkno)->lh_first; bp; bp = bp->b_hash.le_next)
+               if (bp->b_lblkno == blkno && bp->b_vp == vp &&
                    (bp->b_flags & B_INVAL) == 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);
+                       return (bp);
+       return (NULL);
 }
 
 /*
 }
 
 /*
- * 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, slpflag, slptimeo)
+       register struct vnode *vp;
        daddr_t blkno;
        daddr_t blkno;
-       int size;
+       int size, slpflag, slptimeo;
+#ifdef SECSIZE
+       long secsize;
+#endif SECSIZE
 {
 {
-       register struct buf *bp, *dp, *ep;
-       int s;
+       register struct buf *bp;
+       struct bufhashhdr *dp;
+       int s, error;
 
 
-       if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT))
-               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:
 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)
+       for (bp = dp->lh_first; bp; bp = bp->b_hash.le_next) {
+               if (bp->b_lblkno != blkno || bp->b_vp != vp)
                        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);
+                       error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
+                               "getblk", slptimeo);
                        splx(s);
                        splx(s);
+                       if (error)
+                               return (NULL);
                        goto loop;
                }
                        goto loop;
                }
+               /*
+                * The test for B_INVAL is moved down here, since there
+                * are cases where B_INVAL is set before VOP_BWRITE() is
+                * called and for NFS, the process cannot be allowed to
+                * allocate a new buffer for the same block until the write
+                * back to the server has been completed. (ie. B_BUSY clears)
+                */
+               if (bp->b_flags & B_INVAL) {
+                       splx(s);
+                       continue;
+               }
+               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;
+                       VOP_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");
        /*
        /*
-        * Not found in the cache, select something from
-        * a free list.  Preference is to LRU list, then AGE list.
+        * The loop back to the top when getnewbuf() fails is because
+        * stateless filesystems like NFS have no node locks. Thus,
+        * there is a slight chance that more than one process will
+        * try and getnewbuf() for the same block concurrently when
+        * the first sleeps in getnewbuf(). So after a sleep, go back
+        * up to the top to check the hash lists again.
         */
         */
-       s = spl6();
-       for (ep = &bfreelist[BQUEUES-1]; ep > bfreelist; ep--)
-               if (ep->av_forw != ep)
-                       break;
-       if (ep == bfreelist) {          /* no free blocks at all */
-               ep->b_flags |= B_WANTED;
-               sleep((caddr_t)ep, PRIBIO+1);
-               splx(s);
+       if ((bp = getnewbuf(slpflag, slptimeo)) == 0)
                goto loop;
                goto loop;
-       }
-       splx(s);
-       bp = ep->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;
-       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_blkno = blkno;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
+       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 *
 geteblk(size)
        int size;
 {
  */
 struct buf *
 geteblk(size)
        int size;
 {
-       register struct buf *bp, *dp;
-       int s;
+       register struct buf *bp;
 
 
-loop:
-       s = spl6();
-       for (dp = &bfreelist[BQUEUES-1]; dp > bfreelist; dp--)
-               if (dp->av_forw != dp)
-                       break;
-       if (dp == bfreelist) {          /* no free blocks */
-               dp->b_flags |= B_WANTED;
-               sleep((caddr_t)dp, PRIBIO+1);
-               goto loop;
-       }
-       splx(s);
-       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);
+       if (size > MAXBSIZE)
+               panic("geteblk: size too big");
+       while ((bp = getnewbuf(0, 0)) == NULL)
+               /* void */;
+       bp->b_flags |= B_INVAL;
        bremhash(bp);
        bremhash(bp);
-       binshash(bp, dp);
-       bp->b_dev = (dev_t)NODEV;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
+       binshash(bp, &invalhash);
+       bp->b_bcount = 0;
+#ifdef SECSIZE
+       bp->b_blksize = DEV_BSIZE;
+#endif SECSIZE
+       bp->b_error = 0;
+       bp->b_resid = 0;
+       allocbuf(bp, size);
+       return (bp);
 }
 
 /*
 }
 
 /*
- * Allocate space associated with a buffer.
+ * Expand or contract the actual memory allocated to a buffer.
+ * If no memory is available, release buffer and take error exit.
  */
  */
-brealloc(bp, size)
-       register struct buf *bp;
+allocbuf(tp, size)
+       register struct buf *tp;
        int size;
 {
        int size;
 {
-       daddr_t start, last;
-       register struct buf *ep;
-       struct buf *dp;
-       int s;
+       register struct buf *bp, *ep;
+       int sizealloc, take, s;
 
 
+       sizealloc = roundup(size, CLBYTES);
        /*
        /*
-        * First need to make sure that all overlaping previous I/O
-        * is dispatched with.
+        * Buffer size does not change
         */
         */
-       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");
-               goto allocit;
-       }
-       bp->b_flags &= ~B_DONE;
-       if (bp->b_dev == NODEV)
-               goto allocit;
-
+       if (sizealloc == tp->b_bufsize)
+               goto out;
        /*
        /*
-        * 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.
+        * 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.
         */
         */
-       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;
-if (bioprintfs)
-if (ep->b_flags&B_BUSY)
-printf("sleeping on:dev 0x%x, blks %d-%d, flg 0%o allocing dev 0x%x, blks %d-%d, flg 0%o\n",
-ep->b_dev, ep->b_blkno, ep->b_blkno + (ep->b_bcount / DEV_BSIZE) - 1,
-ep->b_flags, bp->b_dev, start, last, bp->b_flags);
-               s = spl6();
-               if (ep->b_flags&B_BUSY) {
-                       ep->b_flags |= B_WANTED;
-                       sleep((caddr_t)ep, PRIBIO+1);
-                       (void) splx(s);
-                       goto loop;
-               }
-               (void) splx(s);
-               notavail(ep);
-               if (ep->b_flags & B_DELWRI) {
-if (bioprintfs)
-printf("DELWRI:dev 0x%x, blks %d-%d, flg 0%o allocing dev 0x%x, blks %d-%d, flg 0%o\n",
-ep->b_dev, ep->b_blkno, ep->b_blkno + (ep->b_bcount / DEV_BSIZE) - 1,
-ep->b_flags, bp->b_dev, start, last, bp->b_flags);
-                       bwrite(ep);
-                       goto loop;
-               }
+       if (sizealloc < tp->b_bufsize) {
+               if ((ep = bufqueues[BQ_EMPTY].tqh_first) == NULL)
+                       goto out;
+               s = splbio();
+               bremfree(ep);
+               ep->b_flags |= B_BUSY;
+               splx(s);
+               pagemove((char *)tp->b_data + sizealloc, ep->b_data,
+                   (int)tp->b_bufsize - sizealloc);
+               ep->b_bufsize = tp->b_bufsize - sizealloc;
+               tp->b_bufsize = sizealloc;
                ep->b_flags |= B_INVAL;
                ep->b_flags |= B_INVAL;
+               ep->b_bcount = 0;
                brelse(ep);
                brelse(ep);
+               goto out;
        }
        }
-allocit:
        /*
        /*
-        * Here the buffer is already available, so all we
-        * need to do is set the size. Someday a better memory
-        * management scheme will be implemented.
+        * 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.
         */
         */
-       bp->b_bcount = size;
+       while (tp->b_bufsize < sizealloc) {
+               take = sizealloc - tp->b_bufsize;
+               while ((bp = getnewbuf(0, 0)) == NULL)
+                       /* void */;
+               if (take >= bp->b_bufsize)
+                       take = bp->b_bufsize;
+               pagemove(&((char *)bp->b_data)[bp->b_bufsize - take],
+                   &((char *)tp->b_data)[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, &invalhash);
+                       bp->b_dev = NODEV;
+                       bp->b_error = 0;
+                       bp->b_flags |= B_INVAL;
+               }
+               brelse(bp);
+       }
+out:
+       tp->b_bcount = size;
        return (1);
 }
 
 /*
        return (1);
 }
 
 /*
- * Release space associated with a buffer.
+ * Find a buffer which is available for use.
+ * Select something from a free list.
+ * Preference is to AGE list, then LRU list.
  */
  */
-bfree(bp)
-       struct buf *bp;
+struct buf *
+getnewbuf(slpflag, slptimeo)
+       int slpflag, slptimeo;
 {
 {
-       /*
-        * Here the buffer does not change, so all we
-        * need to do is set the size. Someday a better memory
-        * management scheme will be implemented.
-        */
-       bp->b_bcount = 0;
+       register struct buf *bp;
+       register struct bqueues *dp;
+       register struct ucred *cred;
+       int s;
+       struct buf *abp;
+       static int losecnt = 0;
+
+loop:
+       s = splbio();
+       abp = NULL;
+       for (dp = &bufqueues[BQ_AGE]; dp > bufqueues; dp--) {
+               for (bp = dp->qe_next; bp; bp = bp->b_freelist.qe_next) {
+                       if (abp == NULL)
+                               abp = bp;
+                       if ((bp->b_flags & B_DELWRI) &&
+                           bp->b_vp && VOP_ISLOCKED(bp->b_vp))
+                               continue;
+                       goto found;
+               }
+       }
+       if (dp == bufqueues) {          /* no free blocks */
+               if (abp) {
+                       bp = abp;
+                       bp->b_flags |= B_XXX;
+                       if (losecnt++ < 20) {
+                               vprint("skipping blkno check", bp->b_vp);
+                               printf("\tlblkno %d, blkno %d\n",
+                                  bp->b_lblkno, bp->b_blkno);
+                       }
+                       goto found;
+               }
+               needbuffer = 1;
+               (void) tsleep((caddr_t)&needbuffer, slpflag | (PRIBIO + 1),
+                       "getnewbuf", slptimeo);
+               splx(s);
+               return (NULL);
+       }
+found:
+       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;
+       bp->b_validoff = bp->b_validend = 0;
+       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);
-       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.
  */
  */
+void
 biodone(bp)
        register struct buf *bp;
 {
 biodone(bp)
        register struct buf *bp;
 {
-       register int s;
 
        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_DIRTY) {
-               if (bp->b_flags & B_ERROR)
-                       panic("IO err in push");
-               s = spl6();
-               bp->av_forw = bclnlist;
-               bp->b_bcount = swsize[bp - swbuf];
-               bp->b_pfcent = swpf[bp - swbuf];
-               cnt.v_pgout++;
-               cnt.v_pgpgout += bp->b_bcount / NBPG;
-               bclnlist = bp;
-               if (bswlist.b_flags & B_WANTED)
-                       wakeup((caddr_t)&proc[2]);
-               splx(s);
+       if ((bp->b_flags & B_READ) == 0)
+               vwakeup(bp);
+       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;
@@ -516,73 +746,48 @@ biodone(bp)
        }
 }
 
        }
 }
 
-/*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
- * (and temporarily pagein)
- */
-bflush(dev)
-       dev_t dev;
+int
+count_lock_queue()
 {
        register struct buf *bp;
 {
        register struct buf *bp;
-       register struct buf *flist;
-       int s;
-
-loop:
-       s = spl6();
-       for (flist = bfreelist; flist < &bfreelist[BQUEUES]; flist++)
-       for (bp = flist->av_forw; bp != flist; bp = bp->av_forw) {
-               if ((bp->b_flags & B_DELWRI) == 0)
-                       continue;
-               if (dev == NODEV || dev == bp->b_dev) {
-                       bp->b_flags |= B_ASYNC;
-                       notavail(bp);
-                       bwrite(bp);
-                       goto loop;
-               }
-       }
-       splx(s);
-}
-
-/*
- * Pick up the device's error number and pass it to the user;
- * if there is an error but the number is 0 set a generalized
- * code.  Actually the latter is always true because devices
- * don't yet return specific errors.
- */
-geterror(bp)
-       register struct buf *bp;
-{
-       int error = 0;
+       register int ret;
 
 
-       if (bp->b_flags&B_ERROR)
-               if ((error = bp->b_error)==0)
-                       return (EIO);
-       return (error);
+       for (ret = 0, bp = (struct buf *)bufqueues[BQ_LOCKED].tqh_first;
+           bp; bp = (struct buf *)bp->b_freelist.tqe_next)
+               ++ret;
+       return(ret);
 }
 
 }
 
+#ifdef DIAGNOSTIC
 /*
 /*
- * 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
+ * Print out statistics on the current allocation of the buffer pool.
+ * Can be enabled to print out on every ``sync'' by setting "syncprt"
+ * in vfs_syscalls.c using sysctl.
  */
  */
-binval(dev)
-       dev_t dev;
+void
+vfs_bufstats()
 {
 {
+       int s, i, j, count;
        register struct buf *bp;
        register struct buf *bp;
-       register struct bufhd *hp;
-#define dp ((struct buf *)hp)
-
-       for (hp = bufhash; hp < &bufhash[BUFHSZ]; hp++)
-               for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-                       if (bp->b_dev == dev)
-                               bp->b_flags |= B_INVAL;
+       register struct bqueues *dp;
+       int counts[MAXBSIZE/CLBYTES+1];
+       static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" };
+
+       for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) {
+               count = 0;
+               for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
+                       counts[j] = 0;
+               s = splbio();
+               for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) {
+                       counts[bp->b_bufsize/CLBYTES]++;
+                       count++;
+               }
+               splx(s);
+               printf("%s: total-%d", bname[i], count);
+               for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
+                       if (counts[j] != 0)
+                               printf(", %d-%d", j * CLBYTES, counts[j]);
+               printf("\n");
+       }
 }
 }
+#endif /* DIAGNOSTIC */