LFS version 2; minor cleanups, nothing important

[unix-history] / usr / src / sys / ufs / lfs / lfs_balloc.c

diff --git a/usr/src/sys/ufs/lfs/lfs_balloc.c b/usr/src/sys/ufs/lfs/lfs_balloc.c
index 27477df..ca75b9b 100644 (file)
--- a/usr/src/sys/ufs/lfs/lfs_balloc.c
+++ b/usr/src/sys/ufs/lfs/lfs_balloc.c
@@ -1,206 +1,176 @@
-/*     lfs_balloc.c    5.3     82/11/13        */
+/*
+ * Copyright (c) 1989, 1991 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)lfs_balloc.c        7.22 (Berkeley) %G%
+ */
+
+#include <sys/param.h>
+#include <sys/buf.h>
+#include <sys/proc.h>
+#include <sys/vnode.h>
+#include <sys/mount.h>
+#include <sys/resourcevar.h>
+#include <sys/specdev.h>
+#include <sys/trace.h>

 
 

-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/conf.h"
-#include "../h/inode.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/buf.h"
-#include "../h/proc.h"
-#include "../h/fs.h"
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
+#include <ufs/ufs/ufsmount.h>
+
+#include <ufs/lfs/lfs.h>
+#include <ufs/lfs/lfs_extern.h>

 
 /*
 
 /*

- * Bmap defines the structure of file system storage
- * by returning the physical block number on a device given the
- * inode and the logical block number in a file.
- * When convenient, it also leaves the physical
- * block number of the next block of the file in rablock
- * for use in read-ahead.
+ * Bmap converts a the logical block number of a file to its physical block
+ * number on the disk. The conversion is done by using the logical block
+ * number to index into the array of block pointers described by the dinode.
+ *
+ * LFS has a different version of bmap from FFS because of a naming conflict.
+ * In FFS, meta blocks are given real disk addresses at allocation time, and
+ * are linked into the device vnode, using a logical block number which is
+ * the same as the physical block number.  This can't be done by LFS because
+ * blocks aren't given disk addresses until they're written, so there's no
+ * way to distinguish the meta-data blocks for one file from any other file.
+ * This means that meta-data blocks have to be on the vnode for the file so
+ * they can be found, and have to have "names" different from the standard
+ * data blocks.  To do this, we divide the name space into positive and
+ * negative block numbers, and give the meta-data blocks negative logical
+ * numbers.  Indirect blocks are addressed by the negative address of the
+ * first data block to which they point.  Double indirect blocks are addressed
+ * by one less than the address of the first indirect block to which they
+ * point.  Triple indirect blocks are addressed by one less than the address
+ * of the first double indirect block to which they point.

  */
  */

-/*VARARGS3*/
-daddr_t
-bmap(ip, bn, rwflg, size)
-       register struct inode *ip;
-       daddr_t bn;
-       int rwflg;
-       int size;       /* supplied only when rwflg == B_WRITE */
+int
+lfs_bmap(vp, bn, vpp, bnp)
+       struct vnode *vp;
+       register daddr_t bn;
+       struct vnode **vpp;
+       daddr_t *bnp;

 {
 {

-       register int i;
-       int osize, nsize;
-       struct buf *bp, *nbp;
-       struct fs *fs;
-       int j, sh;
-       daddr_t nb, lbn, *bap, pref, blkpref();
-
-       if (bn < 0) {
-               u.u_error = EFBIG;
-               return ((daddr_t)0);
-       }
-       fs = ip->i_fs;
-       rablock = 0;
-       rasize = 0;             /* conservative */
+       register struct inode *ip;
+       register struct lfs *fs;
+       register daddr_t nb;
+       struct buf *bp;
+       struct vnode *devvp;
+       daddr_t *bap, daddr, metalbn;
+       long realbn;
+       int error, j, off, sh;

 
        /*
 
        /*

-        * If the next write will extend the file into a new block,
-        * and the file is currently composed of a fragment
-        * this fragment has to be extended to be a full block.
-        */
-       nb = lblkno(fs, ip->i_size);
-       if (rwflg == B_WRITE && nb < NDADDR && nb < bn) {
-               osize = blksize(fs, ip, nb);
-               if (osize < fs->fs_bsize && osize > 0) {
-                       bp = realloccg(ip, ip->i_db[nb],
-                               blkpref(ip, nb, (int)nb, &ip->i_db[0]),
-                               osize, (int)fs->fs_bsize);
-                       ip->i_size = (nb + 1) * fs->fs_bsize;
-                       ip->i_db[nb] = dbtofsb(fs, bp->b_blkno);
-                       ip->i_flag |= IUPD|ICHG;
-                       bdwrite(bp);
-               }
-       }
-       /*
-        * The first NDADDR blocks are direct blocks
+        * Check for underlying vnode requests and ensure that logical
+        * to physical mapping is requested.

         */
         */

+       ip = VTOI(vp);
+       if (vpp != NULL)
+               *vpp = ip->i_devvp;
+       if (bnp == NULL)
+               return (0);
+
+#ifdef VERBOSE
+printf("lfs_bmap: block number %d, inode %d\n", bn, ip->i_number);
+#endif
+       realbn = bn;
+       if ((long)bn < 0)
+               bn = -(long)bn;
+
+       /* The first NDADDR blocks are direct blocks. */

        if (bn < NDADDR) {
                nb = ip->i_db[bn];
        if (bn < NDADDR) {
                nb = ip->i_db[bn];

-               if (rwflg == B_READ) {
-                       if (nb == 0)
-                               return ((daddr_t)-1);
-                       goto gotit;
-               }
-               if (nb == 0 || ip->i_size < (bn + 1) * fs->fs_bsize) {
-                       if (nb != 0) {
-                               /* consider need to reallocate a frag */
-                               osize = fragroundup(fs, blkoff(fs, ip->i_size));
-                               nsize = fragroundup(fs, size);
-                               if (nsize <= osize)
-                                       goto gotit;
-                               bp = realloccg(ip, nb,
-                                       blkpref(ip, bn, (int)bn, &ip->i_db[0]),
-                                       osize, nsize);
-                       } else {
-                               if (ip->i_size < (bn + 1) * fs->fs_bsize)
-                                       nsize = fragroundup(fs, size);
-                               else
-                                       nsize = fs->fs_bsize;
-                               bp = alloc(ip,
-                                       blkpref(ip, bn, (int)bn, &ip->i_db[0]),
-                                       nsize);
-                       }
-                       if (bp == NULL)
-                               return ((daddr_t)-1);
-                       nb = dbtofsb(fs, bp->b_blkno);
-                       if ((ip->i_mode&IFMT) == IFDIR)
-                               /*
-                                * Write directory blocks synchronously
-                                * so they never appear with garbage in
-                                * them on the disk.
-                                */
-                               bwrite(bp);
-                       else
-                               bdwrite(bp);
-                       ip->i_db[bn] = nb;
-                       ip->i_flag |= IUPD|ICHG;
-               }
-gotit:
-               if (bn < NDADDR - 1) {
-                       rablock = fsbtodb(fs, ip->i_db[bn + 1]);
-                       rasize = blksize(fs, ip, bn + 1);
+               if (nb == 0) {
+                       *bnp = UNASSIGNED;
+                       return (0);

                }
                }

-               return (nb);
+               *bnp = nb;
+               return (0);

        }
 
        }
 

-       /*
-        * Determine how many levels of indirection.
+       /* 
+        * Determine the number of levels of indirection.  After this loop
+        * is done, sh indicates the number of data blocks possible at the
+        * given level of indirection, and NIADDR - j is the number of levels
+        * of indirection needed to locate the requested block.

         */
         */

-       pref = 0;
-       sh = 1;
-       lbn = bn;

        bn -= NDADDR;
        bn -= NDADDR;

-       for (j = NIADDR; j>0; j--) {
+       fs = ip->i_lfs;
+       sh = 1;
+       for (j = NIADDR; j > 0; j--) {

                sh *= NINDIR(fs);
                if (bn < sh)
                        break;
                bn -= sh;
        }
                sh *= NINDIR(fs);
                if (bn < sh)
                        break;
                bn -= sh;
        }

-       if (j == 0) {
-               u.u_error = EFBIG;
-               return ((daddr_t)0);
-       }
+       if (j == 0)
+               return (EFBIG);

 
 

-       /*
-        * fetch the first indirect block
+       /* Calculate the address of the first meta-block. */
+       if (realbn >= 0)
+               metalbn = -(realbn - bn + NIADDR - j);
+       else
+               metalbn = -(-realbn - bn + NIADDR - j);
+
+       /* 
+        * Fetch through the indirect blocks.  At each iteration, off is the
+        * offset into the bap array which is an array of disk addresses at
+        * the current level of indirection.

         */
         */

-       nb = ip->i_ib[NIADDR - j];
-       if (nb == 0) {
-               if (rwflg == B_READ)
-                       return ((daddr_t)-1);
-               pref = blkpref(ip, lbn, 0, (daddr_t *)0);
-               bp = alloc(ip, pref, (int)fs->fs_bsize);
-               if (bp == NULL)
-                       return ((daddr_t)-1);
-               nb = dbtofsb(fs, bp->b_blkno);
+       bp = NULL;
+       devvp = VFSTOUFS(vp->v_mount)->um_devvp;
+       for (off = NIADDR - j, bap = ip->i_ib; j <= NIADDR; j++) {

                /*
                /*

-                * Write synchronously so that indirect blocks
-                * never point at garbage.
+                * In LFS, it's possible to have a block appended to a file
+                * for which the meta-blocks have not yet been allocated.
+                * This is a win if the file never gets written or if the
+                * file's growing.

                 */
                 */

-               bwrite(bp);
-               ip->i_ib[NIADDR - j] = nb;
-               ip->i_flag |= IUPD|ICHG;
-       }
+               if ((daddr = bap[off]) == 0) {
+                       daddr = UNASSIGNED;
+                       break;
+               }

 
 

-       /*
-        * fetch through the indirect blocks
-        */
-       for (; j <= NIADDR; j++) {
-               bp = bread(ip->i_dev, fsbtodb(fs, nb), (int)fs->fs_bsize);
-               if (bp->b_flags & B_ERROR) {
+               /* If searching for a meta-data block, quit when found. */
+               if (metalbn == realbn)
+                       break;
+
+               /*
+                * Read in the appropriate indirect block.  LFS can't do a
+                * bread because bread knows that FFS will hand it the device
+                * vnode, not the file vnode, so the b_dev and b_blkno would
+                * be wrong.
+                *
+                * XXX
+                * This REALLY needs to be fixed, at the very least it needs
+                * to be rethought when the buffer cache goes away.
+                */
+               if (bp)

                        brelse(bp);
                        brelse(bp);

-                       return ((daddr_t)0);
+               bp = getblk(vp, metalbn, fs->lfs_bsize);
+               if (bp->b_flags & (B_DONE | B_DELWRI)) {
+                       trace(TR_BREADHIT, pack(vp, size), metalbn);
+               } else {
+                       trace(TR_BREADMISS, pack(vp, size), metalbn);
+                       bp->b_blkno = daddr;
+                       bp->b_flags |= B_READ;
+                       bp->b_dev = devvp->v_rdev;
+                       (devvp->v_op->vop_strategy)(bp);
+                       curproc->p_stats->p_ru.ru_inblock++;    /* XXX */
+                       if (error = biowait(bp)) {
+                               brelse(bp);
+                               return (error);
+                       }

                }
                }

+

                bap = bp->b_un.b_daddr;
                sh /= NINDIR(fs);
                bap = bp->b_un.b_daddr;
                sh /= NINDIR(fs);

-               i = (bn / sh) % NINDIR(fs);
-               nb = bap[i];
-               if (nb == 0) {
-                       if (rwflg==B_READ) {
-                               brelse(bp);
-                               return ((daddr_t)-1);
-                       }
-                       if (pref == 0)
-                               if (j < NIADDR)
-                                       pref = blkpref(ip, lbn, 0,
-                                               (daddr_t *)0);
-                               else
-                                       pref = blkpref(ip, lbn, i, &bap[0]);
-                       nbp = alloc(ip, pref, (int)fs->fs_bsize);
-                       if (nbp == NULL) {
-                               brelse(bp);
-                               return ((daddr_t)-1);
-                       }
-                       nb = dbtofsb(fs, nbp->b_blkno);
-                       if (j < NIADDR || (ip->i_mode&IFMT) == IFDIR)
-                               /*
-                                * Write synchronously so indirect blocks
-                                * never point at garbage and blocks
-                                * in directories never contain garbage.
-                                */
-                               bwrite(nbp);
-                       else
-                               bdwrite(nbp);
-                       bap[i] = nb;
-                       bdwrite(bp);
-               } else
-                       brelse(bp);
+               off = (bn / sh) % NINDIR(fs);
+               metalbn -= -1 + off * sh;

        }
        }

+       if (bp)
+               brelse(bp);

 
 

-       /*
-        * calculate read-ahead.
-        */
-       if (i < NINDIR(fs) - 1) {
-               rablock = fsbtodb(fs, bap[i+1]);
-               rasize = fs->fs_bsize;
-       }
-       return (nb);
+       *bnp = daddr;
+       return (0);

 }
 }