BSD 4_4_Lite1 release

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

diff --git a/usr/src/sys/ufs/lfs/lfs_segment.c b/usr/src/sys/ufs/lfs/lfs_segment.c
index 10b3c8a..249d59d 100644 (file)
--- a/usr/src/sys/ufs/lfs/lfs_segment.c
+++ b/usr/src/sys/ufs/lfs/lfs_segment.c
@@ -1,10 +1,36 @@
 /*
 /*

- * Copyright (c) 1991 Regents of the University of California.
- * All rights reserved.
+ * Copyright (c) 1991, 1993
+ *     The Regents of the University of California.  All rights reserved.

  *
  *

- * %sccs.include.redist.c%
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.

  *
  *

- *     @(#)lfs_segment.c       7.20 (Berkeley) %G%
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)lfs_segment.c       8.5 (Berkeley) 1/4/94

  */
 
 #include <sys/param.h>
  */
 
 #include <sys/param.h>


@@ -18,34 +44,24 @@
 #include <sys/proc.h>
 #include <sys/conf.h>
 #include <sys/vnode.h>
 #include <sys/proc.h>
 #include <sys/conf.h>
 #include <sys/vnode.h>

-#include <sys/specdev.h>
-#include <sys/fifo.h>

 #include <sys/malloc.h>
 #include <sys/mount.h>
 
 #include <sys/malloc.h>
 #include <sys/mount.h>
 

+#include <miscfs/specfs/specdev.h>
+#include <miscfs/fifofs/fifo.h>
+

 #include <ufs/ufs/quota.h>
 #include <ufs/ufs/inode.h>
 #include <ufs/ufs/dir.h>
 #include <ufs/ufs/ufsmount.h>
 #include <ufs/ufs/quota.h>
 #include <ufs/ufs/inode.h>
 #include <ufs/ufs/dir.h>
 #include <ufs/ufs/ufsmount.h>

+#include <ufs/ufs/ufs_extern.h>

 
 #include <ufs/lfs/lfs.h>
 #include <ufs/lfs/lfs_extern.h>
 
 
 #include <ufs/lfs/lfs.h>
 #include <ufs/lfs/lfs_extern.h>
 

-/* In-memory description of a segment about to be written. */
-struct segment {
-       struct buf      **bpp;          /* pointer to buffer array */
-       struct buf      **cbpp;         /* pointer to next available bp */
-       struct buf      *ibp;           /* buffer pointer to inode page */
-       struct finfo    *fip;           /* current fileinfo pointer */
-       void    *segsum;                /* segment summary info */
-       u_long  ninodes;                /* number of inodes in this segment */
-       u_long  seg_bytes_left;         /* bytes left in segment */
-       u_long  sum_bytes_left;         /* bytes left in summary block */
-       u_long  seg_number;             /* number of this segment */
-#define        SEGM_CKP        0x01            /* doing a checkpoint */
-       u_long  seg_flags;              /* run-time flags for this segment */
-};
+extern int count_lock_queue __P((void));

 
 

+#define MAX_ACTIVE     10

 /*
  * Determine if it's OK to start a partial in this segment, or if we need
  * to go on to a new segment.
 /*
  * Determine if it's OK to start a partial in this segment, or if we need
  * to go on to a new segment.


@@ -57,27 +73,36 @@ struct segment {
 void    lfs_callback __P((struct buf *));
 void    lfs_gather __P((struct lfs *, struct segment *,
             struct vnode *, int (*) __P((struct lfs *, struct buf *))));
 void    lfs_callback __P((struct buf *));
 void    lfs_gather __P((struct lfs *, struct segment *,
             struct vnode *, int (*) __P((struct lfs *, struct buf *))));

-void    lfs_initseg __P((struct lfs *, struct segment *));
+int     lfs_gatherblock __P((struct segment *, struct buf *, int *));

 void    lfs_iset __P((struct inode *, daddr_t, time_t));
 int     lfs_match_data __P((struct lfs *, struct buf *));
 int     lfs_match_dindir __P((struct lfs *, struct buf *));
 int     lfs_match_indir __P((struct lfs *, struct buf *));
 int     lfs_match_tindir __P((struct lfs *, struct buf *));
 void    lfs_iset __P((struct inode *, daddr_t, time_t));
 int     lfs_match_data __P((struct lfs *, struct buf *));
 int     lfs_match_dindir __P((struct lfs *, struct buf *));
 int     lfs_match_indir __P((struct lfs *, struct buf *));
 int     lfs_match_tindir __P((struct lfs *, struct buf *));

-struct buf *
-        lfs_newbuf __P((struct lfs *, daddr_t, size_t));

 void    lfs_newseg __P((struct lfs *));
 void    lfs_shellsort __P((struct buf **, daddr_t *, register int));
 void    lfs_newseg __P((struct lfs *));
 void    lfs_shellsort __P((struct buf **, daddr_t *, register int));

-void    lfs_updatemeta __P((struct lfs *,
-           struct segment *, struct vnode *, daddr_t *, struct buf **, int));
+void    lfs_supercallback __P((struct buf *));
+void    lfs_updatemeta __P((struct segment *));
+int     lfs_vref __P((struct vnode *));
+void    lfs_vunref __P((struct vnode *));

 void    lfs_writefile __P((struct lfs *, struct segment *, struct vnode *));
 int     lfs_writeinode __P((struct lfs *, struct segment *, struct inode *));
 int     lfs_writeseg __P((struct lfs *, struct segment *));
 void    lfs_writefile __P((struct lfs *, struct segment *, struct vnode *));
 int     lfs_writeinode __P((struct lfs *, struct segment *, struct inode *));
 int     lfs_writeseg __P((struct lfs *, struct segment *));

-void    lfs_writesuper __P((struct lfs *, struct segment *));
+void    lfs_writesuper __P((struct lfs *));

 void    lfs_writevnodes __P((struct lfs *fs, struct mount *mp,
            struct segment *sp, int dirops));
 
 int    lfs_allclean_wakeup;            /* Cleaner wakeup address. */
 
 void    lfs_writevnodes __P((struct lfs *fs, struct mount *mp,
            struct segment *sp, int dirops));
 
 int    lfs_allclean_wakeup;            /* Cleaner wakeup address. */
 

+/* Statistics Counters */
+#define DOSTATS
+struct lfs_stats lfs_stats;
+
+/* op values to lfs_writevnodes */
+#define        VN_REG  0
+#define        VN_DIROP        1
+#define        VN_EMPTY        2
+

 /*
  * Ifile and meta data blocks are not marked busy, so segment writes MUST be
  * single threaded.  Currently, there are two paths into lfs_segwrite, sync()
 /*
  * Ifile and meta data blocks are not marked busy, so segment writes MUST be
  * single threaded.  Currently, there are two paths into lfs_segwrite, sync()


@@ -91,89 +116,52 @@ lfs_vflush(vp)
 {
        struct inode *ip;
        struct lfs *fs;
 {
        struct inode *ip;
        struct lfs *fs;

-       struct mount *mp;

        struct segment *sp;
        struct segment *sp;

-       int error, s;

 
 

-#ifdef VERBOSE
-       printf("lfs_vflush\n");
-#endif
-       mp = vp->v_mount;
-       fs = VFSTOUFS(mp)->um_lfs;
+       fs = VFSTOUFS(vp->v_mount)->um_lfs;
+       if (fs->lfs_nactive > MAX_ACTIVE)
+               return(lfs_segwrite(vp->v_mount, SEGM_SYNC|SEGM_CKP));
+       lfs_seglock(fs, SEGM_SYNC);
+       sp = fs->lfs_sp;

 
 

-       /*
-        * XXX
-        * check flags?
-        * mp->mnt_flag & (MNT_MLOCK|MNT_RDONLY|MNT_MPBUSY) ||
-        */
-       if (vfs_busy(mp))
-               return (0);

 
 

-       /*
-        * Allocate a segment structure and enough space to hold pointers to
-        * the maximum possible number of buffers which can be described in a
-        * single summary block.
-        */
-       sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK);
-       sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) /
-           sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK);
-       sp->seg_flags = SEGM_CKP;
-       lfs_initseg(fs, sp);
-
-       /*
-        * Keep a cumulative count of the outstanding I/O operations.  If the
-        * disk drive catches up with us it could go to zero before we finish,
-        * so we artificially increment it by one until we've scheduled all of
-        * the writes we intend to do.
-        */
-       s = splbio();
-       ++fs->lfs_iocount;
-       splx(s);
-
-       if (vp->v_dirtyblkhd != NULL)
-               lfs_writefile(fs, sp, vp);

        ip = VTOI(vp);
        ip = VTOI(vp);

-       (void) lfs_writeinode(fs, sp, ip);
-       ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG);
-
-       (void) lfs_writeseg(fs, sp);
+       if (vp->v_dirtyblkhd.lh_first == NULL)
+               lfs_writevnodes(fs, vp->v_mount, sp, VN_EMPTY);

 
 

-       /*
-        * If the I/O count is non-zero, sleep until it reaches zero.  At the
-        * moment, the user's process hangs around so we can sleep.
-        */
-       s = splbio();
-       if (--fs->lfs_iocount && (error =
-           tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs vflush", 0))) {
-               free(sp->bpp, M_SEGMENT);
-               free(sp, M_SEGMENT);
-               return (error);
-       }
-       splx(s);
-       vfs_unbusy(mp);
+       do {
+               do {
+                       if (vp->v_dirtyblkhd.lh_first != NULL)
+                               lfs_writefile(fs, sp, vp);
+               } while (lfs_writeinode(fs, sp, ip));

 
 

-       /*
-        * XXX
-        * Should be writing a checkpoint?
-        */
-       free(sp->bpp, M_SEGMENT);
-       free(sp, M_SEGMENT);
+       } while (lfs_writeseg(fs, sp) && ip->i_number == LFS_IFILE_INUM);

 
 

+#ifdef DOSTATS
+       ++lfs_stats.nwrites;
+       if (sp->seg_flags & SEGM_SYNC)
+               ++lfs_stats.nsync_writes;
+       if (sp->seg_flags & SEGM_CKP)
+               ++lfs_stats.ncheckpoints;
+#endif
+       lfs_segunlock(fs);

        return (0);
 }
 
 void
        return (0);
 }
 
 void

-lfs_writevnodes(fs, mp, sp, dirops)
+lfs_writevnodes(fs, mp, sp, op)

        struct lfs *fs;
        struct mount *mp;
        struct segment *sp;
        struct lfs *fs;
        struct mount *mp;
        struct segment *sp;

-       int dirops;
+       int op;

 {
        struct inode *ip;
        struct vnode *vp;
 {
        struct inode *ip;
        struct vnode *vp;

-       int error, s;

 
 

-loop:  for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) {
+loop:
+       for (vp = mp->mnt_vnodelist.lh_first;
+            vp != NULL;
+            vp = vp->v_mntvnodes.le_next) {

                /*
                 * If the vnode that we are about to sync is no longer
                 * associated with this mount point, start over.
                /*
                 * If the vnode that we are about to sync is no longer
                 * associated with this mount point, start over.


@@ -181,108 +169,127 @@ loop:   for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) {
                if (vp->v_mount != mp)
                        goto loop;
 
                if (vp->v_mount != mp)
                        goto loop;
 

-               if (dirops && !(vp->v_flag & VDIROP) ||
-                   !dirops && (vp->v_flag & VDIROP))
+               /* XXX ignore dirops for now
+               if (op == VN_DIROP && !(vp->v_flag & VDIROP) ||
+                   op != VN_DIROP && (vp->v_flag & VDIROP))
+                       continue;
+               */
+
+               if (op == VN_EMPTY && vp->v_dirtyblkhd.lh_first)
+                       continue;
+
+               if (vp->v_type == VNON)
+                       continue;
+
+               if (lfs_vref(vp))

                        continue;
                        continue;

-               /*
-                * XXX
-                * Up the ref count so we don't get tossed out of
-                * memory.
-                */
-               VREF(vp);

 
                /*
                 * Write the inode/file if dirty and it's not the
                 * the IFILE.
                 */
                ip = VTOI(vp);
 
                /*
                 * Write the inode/file if dirty and it's not the
                 * the IFILE.
                 */
                ip = VTOI(vp);

-               if ((ip->i_flag & (IMOD | IACC | IUPD | ICHG) ||
-                   vp->v_dirtyblkhd != NULL) &&
+               if ((ip->i_flag &
+                   (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE) ||
+                   vp->v_dirtyblkhd.lh_first != NULL) &&

                    ip->i_number != LFS_IFILE_INUM) {
                    ip->i_number != LFS_IFILE_INUM) {

-                       if (vp->v_dirtyblkhd != NULL)
+                       if (vp->v_dirtyblkhd.lh_first != NULL)

                                lfs_writefile(fs, sp, vp);
                        (void) lfs_writeinode(fs, sp, ip);
                                lfs_writefile(fs, sp, vp);
                        (void) lfs_writeinode(fs, sp, ip);

-                       ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG);

                }
                vp->v_flag &= ~VDIROP;
                }
                vp->v_flag &= ~VDIROP;

-               vrele(vp);
+               lfs_vunref(vp);

        }
 }
 
 int
        }
 }
 
 int

-lfs_segwrite(mp, do_ckp)
+lfs_segwrite(mp, flags)

        struct mount *mp;
        struct mount *mp;

-       int do_ckp;                     /* Do a checkpoint. */
+       int flags;                      /* Do a checkpoint. */

 {
 {

-       USES_VOP_ISLOCKED;
+       struct buf *bp;

        struct inode *ip;
        struct lfs *fs;
        struct segment *sp;
        struct vnode *vp;
        struct inode *ip;
        struct lfs *fs;
        struct segment *sp;
        struct vnode *vp;

-       int error, s;
+       SEGUSE *segusep;
+       daddr_t ibno;
+       CLEANERINFO *cip;
+       int clean, do_ckp, error, i;

 
 

-#ifdef VERBOSE
-       printf("lfs_segwrite\n");
-#endif

        fs = VFSTOUFS(mp)->um_lfs;
 
        fs = VFSTOUFS(mp)->um_lfs;
 

+       /*
+        * If we have fewer than 2 clean segments, wait until cleaner
+        * writes.
+        */
+       do {
+               LFS_CLEANERINFO(cip, fs, bp);
+               clean = cip->clean;
+               brelse(bp);
+               if (clean <= 2) {
+                       printf ("segs clean: %d\n", clean);
+                       wakeup(&lfs_allclean_wakeup);
+                       if (error = tsleep(&fs->lfs_avail, PRIBIO + 1,
+                           "lfs writer", 0))
+                               return (error);
+               }
+       } while (clean <= 2 );
+

        /*
         * Allocate a segment structure and enough space to hold pointers to
         * the maximum possible number of buffers which can be described in a
         * single summary block.
         */
        /*
         * Allocate a segment structure and enough space to hold pointers to
         * the maximum possible number of buffers which can be described in a
         * single summary block.
         */

-       sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK);
-       sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) /
-           sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK);
-       sp->seg_flags = do_ckp ? SEGM_CKP : 0;
-       lfs_initseg(fs, sp);
+       do_ckp = flags & SEGM_CKP || fs->lfs_nactive > MAX_ACTIVE;
+       lfs_seglock(fs, flags | (do_ckp ? SEGM_CKP : 0));
+       sp = fs->lfs_sp;

 
 

-       /*
-        * Keep a cumulative count of the outstanding I/O operations.  If the
-        * disk drive catches up with us it could go to zero before we finish,
-        * so we artificially increment it by one until we've scheduled all of
-        * the writes we intend to do.  If not a checkpoint, we never do the
-        * final decrement, avoiding the wakeup in the callback routine.
-        */
-       s = splbio();
-       fs->lfs_iocount++;
-       splx(s);
+       lfs_writevnodes(fs, mp, sp, VN_REG);

 
 

-       lfs_writevnodes(fs, mp, sp, 0);
-       s = splbio();
+       /* XXX ignore ordering of dirops for now */
+       /* XXX

        fs->lfs_writer = 1;
        if (fs->lfs_dirops && (error =
            tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs writer", 0))) {
                free(sp->bpp, M_SEGMENT);
                free(sp, M_SEGMENT); 
                fs->lfs_writer = 0;
        fs->lfs_writer = 1;
        if (fs->lfs_dirops && (error =
            tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs writer", 0))) {
                free(sp->bpp, M_SEGMENT);
                free(sp, M_SEGMENT); 
                fs->lfs_writer = 0;

-               splx(s);
-               return(error);
+               return (error);

        }
        }

-       splx(s);

 
 

-       lfs_writevnodes(fs, mp, sp, 1);
+       lfs_writevnodes(fs, mp, sp, VN_DIROP);
+       */

 
        /*
 
        /*

-        * If this is a checkpoint, we need to loop on both the ifile and
-        * the writeseg to make sure that we don't end up with any dirty
-        * buffers left when this is all over.
+        * If we are doing a checkpoint, mark everything since the
+        * last checkpoint as no longer ACTIVE.

         */
         */

+       if (do_ckp)
+               for (ibno = fs->lfs_cleansz + fs->lfs_segtabsz;
+                    --ibno >= fs->lfs_cleansz; ) {
+                       if (bread(fs->lfs_ivnode, ibno, fs->lfs_bsize,
+                           NOCRED, &bp))
+
+                               panic("lfs: ifile read");
+                       segusep = (SEGUSE *)bp->b_data;
+                       for (i = fs->lfs_sepb; i--; segusep++)
+                               segusep->su_flags &= ~SEGUSE_ACTIVE;
+                               
+                       error = VOP_BWRITE(bp);
+               }
+

        if (do_ckp || fs->lfs_doifile) {
 redo:
                vp = fs->lfs_ivnode;
        if (do_ckp || fs->lfs_doifile) {
 redo:
                vp = fs->lfs_ivnode;

-               while (vget(vp));
+               while (vget(vp, 1));

                ip = VTOI(vp);
                ip = VTOI(vp);

-               do {
-                       if (vp->v_dirtyblkhd != NULL)
-                               lfs_writefile(fs, sp, vp);
-               } while (lfs_writeinode(fs, sp, ip) && do_ckp);
-               ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG);
+               if (vp->v_dirtyblkhd.lh_first != NULL)
+                       lfs_writefile(fs, sp, vp);
+               (void)lfs_writeinode(fs, sp, ip);

                vput(vp);
                vput(vp);

-               if (lfs_writeseg(fs, sp) && do_ckp) {
-                       lfs_initseg(fs, sp);
+               if (lfs_writeseg(fs, sp) && do_ckp)

                        goto redo;
                        goto redo;

-               }

        } else
                (void) lfs_writeseg(fs, sp);
 
        } else
                (void) lfs_writeseg(fs, sp);
 


@@ -290,27 +297,20 @@ redo:
         * If the I/O count is non-zero, sleep until it reaches zero.  At the
         * moment, the user's process hangs around so we can sleep.
         */
         * If the I/O count is non-zero, sleep until it reaches zero.  At the
         * moment, the user's process hangs around so we can sleep.
         */

-       s = splbio();
-       --fs->lfs_iocount;
+       /* XXX ignore dirops for now

        fs->lfs_writer = 0;
        fs->lfs_doifile = 0;
        wakeup(&fs->lfs_dirops);
        fs->lfs_writer = 0;
        fs->lfs_doifile = 0;
        wakeup(&fs->lfs_dirops);

-
-       if (do_ckp) {
-               if (fs->lfs_iocount && (error =
-                   tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs sync", 0))) {
-                       free(sp->bpp, M_SEGMENT);
-                       free(sp, M_SEGMENT);
-                       return (error);
-               }
-               splx(s);
-               lfs_writesuper(fs, sp);
-       } else 
-               splx(s);
-
-       free(sp->bpp, M_SEGMENT);
-       free(sp, M_SEGMENT);
-
+       */
+
+#ifdef DOSTATS
+       ++lfs_stats.nwrites;
+       if (sp->seg_flags & SEGM_SYNC)
+               ++lfs_stats.nsync_writes;
+       if (sp->seg_flags & SEGM_CKP)
+               ++lfs_stats.ncheckpoints;
+#endif
+       lfs_segunlock(fs);

        return (0);
 }
 
        return (0);
 }
 


@@ -327,15 +327,12 @@ lfs_writefile(fs, sp, vp)
        struct finfo *fip;
        IFILE *ifp;
 
        struct finfo *fip;
        IFILE *ifp;
 

-#ifdef VERBOSE
-       printf("lfs_writefile\n");
-#endif

        if (sp->seg_bytes_left < fs->lfs_bsize ||
        if (sp->seg_bytes_left < fs->lfs_bsize ||

-           sp->sum_bytes_left < sizeof(struct finfo)) {
+           sp->sum_bytes_left < sizeof(struct finfo))

                (void) lfs_writeseg(fs, sp);
                (void) lfs_writeseg(fs, sp);

-               lfs_initseg(fs, sp);
-       }
+

        sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(daddr_t);
        sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(daddr_t);

+       ++((SEGSUM *)(sp->segsum))->ss_nfinfo;

 
        fip = sp->fip;
        fip->fi_nblocks = 0;
 
        fip = sp->fip;
        fip->fi_nblocks = 0;


@@ -357,16 +354,15 @@ lfs_writefile(fs, sp, vp)
 #endif
 
        fip = sp->fip;
 #endif
 
        fip = sp->fip;

-#ifdef META
-       printf("lfs_writefile: adding %d blocks\n", fip->fi_nblocks);
-#endif

        if (fip->fi_nblocks != 0) {
        if (fip->fi_nblocks != 0) {

-               ++((SEGSUM *)(sp->segsum))->ss_nfinfo;

                sp->fip =
                    (struct finfo *)((caddr_t)fip + sizeof(struct finfo) +
                    sizeof(daddr_t) * (fip->fi_nblocks - 1));
                sp->fip =
                    (struct finfo *)((caddr_t)fip + sizeof(struct finfo) +
                    sizeof(daddr_t) * (fip->fi_nblocks - 1));

-       } else
+               sp->start_lbp = &sp->fip->fi_blocks[0];
+       } else {

                sp->sum_bytes_left += sizeof(struct finfo) - sizeof(daddr_t);
                sp->sum_bytes_left += sizeof(struct finfo) - sizeof(daddr_t);

+               --((SEGSUM *)(sp->segsum))->ss_nfinfo;
+       }

 }
 
 int
 }
 
 int


@@ -380,27 +376,30 @@ lfs_writeinode(fs, sp, ip)
        SEGUSE *sup;
        daddr_t daddr;
        ino_t ino;
        SEGUSE *sup;
        daddr_t daddr;
        ino_t ino;

-       int ndx;
+       int error, i, ndx;

        int redo_ifile = 0;
 
        int redo_ifile = 0;
 

-#ifdef VERBOSE
-       printf("lfs_writeinode\n");
-#endif
+       if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)))
+               return(0);
+

        /* Allocate a new inode block if necessary. */
        if (sp->ibp == NULL) {
                /* Allocate a new segment if necessary. */
                if (sp->seg_bytes_left < fs->lfs_bsize ||
        /* Allocate a new inode block if necessary. */
        if (sp->ibp == NULL) {
                /* Allocate a new segment if necessary. */
                if (sp->seg_bytes_left < fs->lfs_bsize ||

-                   sp->sum_bytes_left < sizeof(daddr_t)) {
+                   sp->sum_bytes_left < sizeof(daddr_t))

                        (void) lfs_writeseg(fs, sp);
                        (void) lfs_writeseg(fs, sp);

-                       lfs_initseg(fs, sp);
-               }

 
                /* Get next inode block. */
                daddr = fs->lfs_offset;
                fs->lfs_offset += fsbtodb(fs, 1);
                sp->ibp = *sp->cbpp++ =
 
                /* Get next inode block. */
                daddr = fs->lfs_offset;
                fs->lfs_offset += fsbtodb(fs, 1);
                sp->ibp = *sp->cbpp++ =

-                   lfs_newbuf(fs, daddr, fs->lfs_bsize);
-
+                   lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, daddr,
+                   fs->lfs_bsize);
+               /* Zero out inode numbers */
+               for (i = 0; i < INOPB(fs); ++i)
+                       ((struct dinode *)sp->ibp->b_data)[i].di_inumber = 0;
+               ++sp->start_bpp;
+               fs->lfs_avail -= fsbtodb(fs, 1);

                /* Set remaining space counters. */
                sp->seg_bytes_left -= fs->lfs_bsize;
                sp->sum_bytes_left -= sizeof(daddr_t);
                /* Set remaining space counters. */
                sp->seg_bytes_left -= fs->lfs_bsize;
                sp->sum_bytes_left -= sizeof(daddr_t);


@@ -410,10 +409,12 @@ lfs_writeinode(fs, sp, ip)
        }
 
        /* Update the inode times and copy the inode onto the inode page. */
        }
 
        /* Update the inode times and copy the inode onto the inode page. */

+       if (ip->i_flag & IN_MODIFIED)
+               --fs->lfs_uinodes;

        ITIMES(ip, &time, &time);
        ITIMES(ip, &time, &time);

+       ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);

        bp = sp->ibp;
        bp = sp->ibp;

-       bp->b_un.b_dino[sp->ninodes % INOPB(fs)] = ip->i_din;
-
+       ((struct dinode *)bp->b_data)[sp->ninodes % INOPB(fs)] = ip->i_din;

        /* Increment inode count in segment summary block. */
        ++((SEGSUM *)(sp->segsum))->ss_ninos;
 
        /* Increment inode count in segment summary block. */
        ++((SEGSUM *)(sp->segsum))->ss_ninos;
 


@@ -426,21 +427,22 @@ lfs_writeinode(fs, sp, ip)
         * address and access times for this inode in the ifile.
         */
        ino = ip->i_number;
         * address and access times for this inode in the ifile.
         */
        ino = ip->i_number;

-       if (ino == LFS_IFILE_INUM)
+       if (ino == LFS_IFILE_INUM) {
+               daddr = fs->lfs_idaddr;

                fs->lfs_idaddr = bp->b_blkno;
                fs->lfs_idaddr = bp->b_blkno;

-
-       LFS_IENTRY(ifp, fs, ino, ibp);
-       daddr = ifp->if_daddr;
-       ifp->if_daddr = bp->b_blkno;
-       LFS_UBWRITE(ibp);
-       redo_ifile = (ino == LFS_IFILE_INUM && !(ibp->b_flags & B_GATHERED));
+       } else {
+               LFS_IENTRY(ifp, fs, ino, ibp);
+               daddr = ifp->if_daddr;
+               ifp->if_daddr = bp->b_blkno;
+               error = VOP_BWRITE(ibp);
+       }

 
        /*
         * No need to update segment usage if there was no former inode address
         * or if the last inode address is in the current partial segment.
         */
        if (daddr != LFS_UNUSED_DADDR && 
 
        /*
         * No need to update segment usage if there was no former inode address
         * or if the last inode address is in the current partial segment.
         */
        if (daddr != LFS_UNUSED_DADDR && 

-           !(daddr >= fs->lfs_curseg && daddr <= ifp->if_daddr) ) {
+           !(daddr >= fs->lfs_lastpseg && daddr <= bp->b_blkno)) {

                LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp);
 #ifdef DIAGNOSTIC
                if (sup->su_nbytes < sizeof(struct dinode)) {
                LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp);
 #ifdef DIAGNOSTIC
                if (sup->su_nbytes < sizeof(struct dinode)) {


@@ -451,11 +453,60 @@ lfs_writeinode(fs, sp, ip)
                }
 #endif
                sup->su_nbytes -= sizeof(struct dinode);
                }
 #endif
                sup->su_nbytes -= sizeof(struct dinode);

-               LFS_UBWRITE(bp);
-               redo_ifile |= 
+               redo_ifile =

                    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
                    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));

+               error = VOP_BWRITE(bp);
+       }
+       return (redo_ifile);
+}
+
+int
+lfs_gatherblock(sp, bp, sptr)
+       struct segment *sp;
+       struct buf *bp;
+       int *sptr;
+{
+       struct lfs *fs;
+       int version;
+
+       /*
+        * If full, finish this segment.  We may be doing I/O, so
+        * release and reacquire the splbio().
+        */
+#ifdef DIAGNOSTIC
+       if (sp->vp == NULL)
+               panic ("lfs_gatherblock: Null vp in segment");
+#endif
+       fs = sp->fs;
+       if (sp->sum_bytes_left < sizeof(daddr_t) ||
+           sp->seg_bytes_left < fs->lfs_bsize) {
+               if (sptr)
+                       splx(*sptr);
+               lfs_updatemeta(sp);
+
+               version = sp->fip->fi_version;
+               (void) lfs_writeseg(fs, sp);
+
+               sp->fip->fi_version = version;
+               sp->fip->fi_ino = VTOI(sp->vp)->i_number;
+               /* Add the current file to the segment summary. */
+               ++((SEGSUM *)(sp->segsum))->ss_nfinfo;
+               sp->sum_bytes_left -= 
+                   sizeof(struct finfo) - sizeof(daddr_t);
+
+               if (sptr)
+                       *sptr = splbio();
+               return(1);

        }
        }

-       return(redo_ifile);
+
+       /* Insert into the buffer list, update the FINFO block. */
+       bp->b_flags |= B_GATHERED;
+       *sp->cbpp++ = bp;
+       sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno;
+
+       sp->sum_bytes_left -= sizeof(daddr_t);
+       sp->seg_bytes_left -= fs->lfs_bsize;
+       return(0);

 }
 
 void
 }
 
 void


@@ -465,25 +516,12 @@ lfs_gather(fs, sp, vp, match)
        struct vnode *vp;
        int (*match) __P((struct lfs *, struct buf *));
 {
        struct vnode *vp;
        int (*match) __P((struct lfs *, struct buf *));
 {

-       struct buf **bpp, *bp;
-struct buf *lastbp;
-       struct finfo *fip;
-       struct inode *ip;
-       daddr_t *lbp, *start_lbp;
-       u_long version;
+       struct buf *bp;

        int s;
 
        int s;
 

-#ifdef VERBOSE
-       printf("lfs_gather\n");
-#endif
-       ip = VTOI(vp);
-       bpp = sp->cbpp;
-       fip = sp->fip;
-       start_lbp = lbp = &fip->fi_blocks[fip->fi_nblocks];
-
-loop:  s = splbio();
-       lastbp = NULL;
-       for (bp = vp->v_dirtyblkhd; bp; lastbp = bp, bp = bp->b_blockf) {
+       sp->vp = vp;
+       s = splbio();
+loop:  for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = bp->b_vnbufs.le_next) {

                if (bp->b_flags & B_BUSY || !match(fs, bp) ||
                    bp->b_flags & B_GATHERED)
                        continue;
                if (bp->b_flags & B_BUSY || !match(fs, bp) ||
                    bp->b_flags & B_GATHERED)
                        continue;


@@ -493,90 +531,54 @@ loop:     s = splbio();
                if (!(bp->b_flags & B_LOCKED))
                        panic("lfs_gather: bp not B_LOCKED");
 #endif
                if (!(bp->b_flags & B_LOCKED))
                        panic("lfs_gather: bp not B_LOCKED");
 #endif

-               /*
-                * If full, finish this segment.  We may be doing I/O, so
-                * release and reacquire the splbio().
-                */
-               if (sp->sum_bytes_left < sizeof(daddr_t) ||
-                   sp->seg_bytes_left < fs->lfs_bsize) {
-                       splx(s);
-                       lfs_updatemeta(fs,
-                           sp, vp, start_lbp, bpp, lbp - start_lbp);
-
-                       /* Add the current file to the segment summary. */
-                       ++((SEGSUM *)(sp->segsum))->ss_nfinfo;
-
-                       version = fip->fi_version;
-                       (void) lfs_writeseg(fs, sp);
-                       lfs_initseg(fs, sp);
-
-                       fip = sp->fip;
-                       fip->fi_version = version;
-                       fip->fi_ino = ip->i_number;
-                       start_lbp = lbp = fip->fi_blocks;
-
-                       sp->sum_bytes_left -= 
-                           sizeof(struct finfo) - sizeof(daddr_t);
-
-                       bpp = sp->cbpp;
+               if (lfs_gatherblock(sp, bp, &s))

                        goto loop;
                        goto loop;

-               }
-
-               /* Insert into the buffer list, update the FINFO block. */
-               bp->b_flags |= B_GATHERED;
-               *sp->cbpp++ = bp;
-               ++fip->fi_nblocks;
-               *lbp++ = bp->b_lblkno;
-
-               sp->sum_bytes_left -= sizeof(daddr_t);
-               sp->seg_bytes_left -= bp->b_bufsize;

        }
        splx(s);
        }
        splx(s);

-       lfs_updatemeta(fs, sp, vp, start_lbp, bpp, lbp - start_lbp);
+       lfs_updatemeta(sp);
+       sp->vp = NULL;

 }
 
 }
 

+

 /*
  * Update the metadata that points to the blocks listed in the FINFO
  * array.
  */
 void
 /*
  * Update the metadata that points to the blocks listed in the FINFO
  * array.
  */
 void

-lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks)
-       struct lfs *fs;
+lfs_updatemeta(sp)

        struct segment *sp;
        struct segment *sp;

-       struct vnode *vp;
-       daddr_t *lbp;
-       struct buf **bpp;
-       int nblocks;

 {
 {

-       USES_VOP_BWRITE;

        SEGUSE *sup;
        struct buf *bp;
        SEGUSE *sup;
        struct buf *bp;

-       INDIR a[NIADDR], *ap;
+       struct lfs *fs;
+       struct vnode *vp;
+       struct indir a[NIADDR + 2], *ap;

        struct inode *ip;
        daddr_t daddr, lbn, off;
        struct inode *ip;
        daddr_t daddr, lbn, off;

-       int db_per_fsb, error, i, num;
+       int db_per_fsb, error, i, nblocks, num;

 
 

-#ifdef VERBOSE
-       printf("lfs_updatemeta\n");
-#endif
-       if (nblocks == 0)
+       vp = sp->vp;
+       nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
+       if (vp == NULL || nblocks == 0) 

                return;
 
        /* Sort the blocks. */
                return;
 
        /* Sort the blocks. */

-       lfs_shellsort(bpp, lbp, nblocks);
+       if (!(sp->seg_flags & SEGM_CLEAN))
+               lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks);

 
        /*
         * Assign disk addresses, and update references to the logical
         * block and the segment usage information.
         */
 
        /*
         * Assign disk addresses, and update references to the logical
         * block and the segment usage information.
         */

+       fs = sp->fs;

        db_per_fsb = fsbtodb(fs, 1);
        db_per_fsb = fsbtodb(fs, 1);

-       for (i = nblocks; i--; ++bpp) {
-               lbn = *lbp++;
-               (*bpp)->b_blkno = off = fs->lfs_offset;
+       for (i = nblocks; i--; ++sp->start_bpp) {
+               lbn = *sp->start_lbp++;
+               (*sp->start_bpp)->b_blkno = off = fs->lfs_offset;

                fs->lfs_offset += db_per_fsb;
 
                fs->lfs_offset += db_per_fsb;
 

-               if (error = lfs_bmaparray(vp, lbn, &daddr, a, &num))
-                       panic("lfs_updatemeta: lfs_bmaparray %d", error);
+               if (error = ufs_bmaparray(vp, lbn, &daddr, a, &num, NULL))
+                       panic("lfs_updatemeta: ufs_bmaparray %d", error);

                ip = VTOI(vp);
                switch (num) {
                case 0:
                ip = VTOI(vp);
                switch (num) {
                case 0:


@@ -590,12 +592,22 @@ lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks)
                        if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp))
                                panic("lfs_updatemeta: bread bno %d",
                                    ap->in_lbn);
                        if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp))
                                panic("lfs_updatemeta: bread bno %d",
                                    ap->in_lbn);

-                       bp->b_un.b_daddr[ap->in_off] = off;
+                       /*
+                        * Bread may create a new indirect block which needs
+                        * to get counted for the inode.
+                        */
+                       if (bp->b_blkno == -1 && !(bp->b_flags & B_CACHE)) {
+printf ("Updatemeta allocating indirect block: shouldn't happen\n");
+                               ip->i_blocks += btodb(fs->lfs_bsize);
+                               fs->lfs_bfree -= btodb(fs->lfs_bsize);
+                       }
+                       ((daddr_t *)bp->b_data)[ap->in_off] = off;

                        VOP_BWRITE(bp);
                }
 
                /* Update segment usage information. */
                        VOP_BWRITE(bp);
                }
 
                /* Update segment usage information. */

-               if (daddr != UNASSIGNED) {
+               if (daddr != UNASSIGNED &&
+                   !(daddr >= fs->lfs_lastpseg && daddr <= off)) {

                        LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp);
 #ifdef DIAGNOSTIC
                        if (sup->su_nbytes < fs->lfs_bsize) {
                        LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp);
 #ifdef DIAGNOSTIC
                        if (sup->su_nbytes < fs->lfs_bsize) {


@@ -606,7 +618,7 @@ lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks)
                        }
 #endif
                        sup->su_nbytes -= fs->lfs_bsize;
                        }
 #endif
                        sup->su_nbytes -= fs->lfs_bsize;

-                       LFS_UBWRITE(bp);
+                       error = VOP_BWRITE(bp);

                }
        }
 }
                }
        }
 }


@@ -614,26 +626,26 @@ lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks)
 /*
  * Start a new segment.
  */
 /*
  * Start a new segment.
  */

-void
-lfs_initseg(fs, sp)
+int
+lfs_initseg(fs)

        struct lfs *fs;
        struct lfs *fs;

-       struct segment *sp;

 {
 {

+       struct segment *sp;

        SEGUSE *sup;
        SEGSUM *ssp;
        struct buf *bp;
        SEGUSE *sup;
        SEGSUM *ssp;
        struct buf *bp;

-       daddr_t lbn, *lbnp;
+       int repeat;

 
 

-#ifdef VERBOSE
-       printf("lfs_initseg\n");
-#endif
+       sp = fs->lfs_sp;
+
+       repeat = 0;

        /* Advance to the next segment. */
        if (!LFS_PARTIAL_FITS(fs)) {
                /* Wake up any cleaning procs waiting on this file system. */
        /* Advance to the next segment. */
        if (!LFS_PARTIAL_FITS(fs)) {
                /* Wake up any cleaning procs waiting on this file system. */

-               wakeup(&fs->lfs_nextseg);

                wakeup(&lfs_allclean_wakeup);
 
                lfs_newseg(fs);
                wakeup(&lfs_allclean_wakeup);
 
                lfs_newseg(fs);

+               repeat = 1;

                fs->lfs_offset = fs->lfs_curseg;
                sp->seg_number = datosn(fs, fs->lfs_curseg);
                sp->seg_bytes_left = fs->lfs_dbpseg * DEV_BSIZE;
                fs->lfs_offset = fs->lfs_curseg;
                sp->seg_number = datosn(fs, fs->lfs_curseg);
                sp->seg_bytes_left = fs->lfs_dbpseg * DEV_BSIZE;


@@ -655,14 +667,17 @@ lfs_initseg(fs, sp)
        }
        fs->lfs_lastpseg = fs->lfs_offset;
 
        }
        fs->lfs_lastpseg = fs->lfs_offset;
 

+       sp->fs = fs;

        sp->ibp = NULL;
        sp->ninodes = 0;
 
        /* Get a new buffer for SEGSUM and enter it into the buffer list. */
        sp->cbpp = sp->bpp;
        sp->ibp = NULL;
        sp->ninodes = 0;
 
        /* Get a new buffer for SEGSUM and enter it into the buffer list. */
        sp->cbpp = sp->bpp;

-       *sp->cbpp = lfs_newbuf(fs, fs->lfs_offset, LFS_SUMMARY_SIZE);
-       sp->segsum = (*sp->cbpp)->b_un.b_addr;
-       ++sp->cbpp;
+       *sp->cbpp = lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, fs->lfs_offset,
+            LFS_SUMMARY_SIZE);
+       sp->segsum = (*sp->cbpp)->b_data;
+       bzero(sp->segsum, LFS_SUMMARY_SIZE);
+       sp->start_bpp = ++sp->cbpp;

        fs->lfs_offset += LFS_SUMMARY_SIZE / DEV_BSIZE;
 
        /* Set point to SEGSUM, initialize it. */
        fs->lfs_offset += LFS_SUMMARY_SIZE / DEV_BSIZE;
 
        /* Set point to SEGSUM, initialize it. */


@@ -673,9 +688,12 @@ lfs_initseg(fs, sp)
        /* Set pointer to first FINFO, initialize it. */
        sp->fip = (struct finfo *)(sp->segsum + sizeof(SEGSUM));
        sp->fip->fi_nblocks = 0;
        /* Set pointer to first FINFO, initialize it. */
        sp->fip = (struct finfo *)(sp->segsum + sizeof(SEGSUM));
        sp->fip->fi_nblocks = 0;

+       sp->start_lbp = &sp->fip->fi_blocks[0];

 
        sp->seg_bytes_left -= LFS_SUMMARY_SIZE;
        sp->sum_bytes_left = LFS_SUMMARY_SIZE - sizeof(SEGSUM);
 
        sp->seg_bytes_left -= LFS_SUMMARY_SIZE;
        sp->sum_bytes_left = LFS_SUMMARY_SIZE - sizeof(SEGSUM);

+
+       return(repeat);

 }
 
 /*
 }
 
 /*


@@ -690,27 +708,17 @@ lfs_newseg(fs)
        struct buf *bp;
        int curseg, isdirty, sn;
 
        struct buf *bp;
        int curseg, isdirty, sn;
 

-#ifdef VERBOSE
-       printf("lfs_newseg\n");
-#endif
-       /*
-        * Turn off the active bit for the current segment, turn on the
-        * active and dirty bits for the next segment, update the cleaner
-        * info.  Set the current segment to the next segment, get a new
-        * next segment.
-        */
-       LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_curseg), bp);
-       sup->su_flags &= ~SEGUSE_ACTIVE;
-       LFS_UBWRITE(bp);
-
-       LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_nextseg), bp);
-       sup->su_flags |= SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_LIVELOG;
-       LFS_UBWRITE(bp);
+        LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_nextseg), bp);
+        sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
+       sup->su_nbytes = 0;
+       sup->su_nsums = 0;
+       sup->su_ninos = 0;
+        (void) VOP_BWRITE(bp);

 
        LFS_CLEANERINFO(cip, fs, bp);
        --cip->clean;
        ++cip->dirty;
 
        LFS_CLEANERINFO(cip, fs, bp);
        --cip->clean;
        ++cip->dirty;

-       LFS_UBWRITE(bp);
+       (void) VOP_BWRITE(bp);

 
        fs->lfs_lastseg = fs->lfs_curseg;
        fs->lfs_curseg = fs->lfs_nextseg;
 
        fs->lfs_lastseg = fs->lfs_curseg;
        fs->lfs_curseg = fs->lfs_nextseg;


@@ -724,7 +732,12 @@ lfs_newseg(fs)
                if (!isdirty)
                        break;
        }
                if (!isdirty)
                        break;
        }

+
+       ++fs->lfs_nactive;

        fs->lfs_nextseg = sntoda(fs, sn);
        fs->lfs_nextseg = sntoda(fs, sn);

+#ifdef DOSTATS
+       ++lfs_stats.segsused;
+#endif

 }
 
 int
 }
 
 int


@@ -732,7 +745,7 @@ lfs_writeseg(fs, sp)
        struct lfs *fs;
        struct segment *sp;
 {
        struct lfs *fs;
        struct segment *sp;
 {

-       USES_VOP_STRATEGY;
+       extern int locked_queue_count;

        struct buf **bpp, *bp, *cbp;
        SEGUSE *sup;
        SEGSUM *ssp;
        struct buf **bpp, *bp, *cbp;
        SEGUSE *sup;
        SEGSUM *ssp;


@@ -741,15 +754,31 @@ lfs_writeseg(fs, sp)
        u_long *datap, *dp;
        int ch_per_blk, do_again, i, nblocks, num, s;
        int (*strategy)__P((struct vop_strategy_args *));
        u_long *datap, *dp;
        int ch_per_blk, do_again, i, nblocks, num, s;
        int (*strategy)__P((struct vop_strategy_args *));

+       struct vop_strategy_args vop_strategy_a;
+       u_short ninos;

        char *p;
 
        char *p;
 

-#ifdef VERBOSE
-       printf("lfs_writeseg\n");
-#endif
-       /* Checkpoint always writes superblock, even if no data blocks. */
-       if ((nblocks = sp->cbpp - sp->bpp) == 0 && !(sp->seg_flags & SEGM_CKP))
-               return;
+       /*
+        * If there are no buffers other than the segment summary to write
+        * and it is not a checkpoint, don't do anything.  On a checkpoint,
+        * even if there aren't any buffers, you need to write the superblock.
+        */
+       if ((nblocks = sp->cbpp - sp->bpp) == 1)
+               return (0);
+
+       ssp = (SEGSUM *)sp->segsum;

 
 

+       /* Update the segment usage information. */
+       LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
+       ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs);
+       sup->su_nbytes += nblocks - 1 - ninos << fs->lfs_bshift;
+       sup->su_nbytes += ssp->ss_ninos * sizeof(struct dinode);
+       sup->su_nbytes += LFS_SUMMARY_SIZE;
+       sup->su_lastmod = time.tv_sec;
+       sup->su_ninos += ninos;
+       ++sup->su_nsums;
+       do_again = !(bp->b_flags & B_GATHERED);
+       (void)VOP_BWRITE(bp);

        /*
         * Compute checksum across data and then across summary; the first
         * block (the summary block) is skipped.  Set the create time here
        /*
         * Compute checksum across data and then across summary; the first
         * block (the summary block) is skipped.  Set the create time here


@@ -759,23 +788,23 @@ lfs_writeseg(fs, sp)
         * Fix this to do it inline, instead of malloc/copy.
         */
        datap = dp = malloc(nblocks * sizeof(u_long), M_SEGMENT, M_WAITOK);
         * Fix this to do it inline, instead of malloc/copy.
         */
        datap = dp = malloc(nblocks * sizeof(u_long), M_SEGMENT, M_WAITOK);

-       for (bpp = sp->bpp, i = nblocks - 1; i--;)
-               *dp++ = (*++bpp)->b_un.b_words[0];
-       ssp = (SEGSUM *)sp->segsum;
+       for (bpp = sp->bpp, i = nblocks - 1; i--;) {
+               if ((*++bpp)->b_flags & B_INVAL) {
+                       if (copyin((*bpp)->b_saveaddr, dp++, sizeof(u_long)))
+                               panic("lfs_writeseg: copyin failed");
+               } else
+                       *dp++ = ((u_long *)(*bpp)->b_data)[0];
+       }

        ssp->ss_create = time.tv_sec;
        ssp->ss_create = time.tv_sec;

-       ssp->ss_datasum = cksum(datap, nblocks * sizeof(u_long));
+       ssp->ss_datasum = cksum(datap, (nblocks - 1) * sizeof(u_long));

        ssp->ss_sumsum =
            cksum(&ssp->ss_datasum, LFS_SUMMARY_SIZE - sizeof(ssp->ss_sumsum));
        free(datap, M_SEGMENT);
        ssp->ss_sumsum =
            cksum(&ssp->ss_datasum, LFS_SUMMARY_SIZE - sizeof(ssp->ss_sumsum));
        free(datap, M_SEGMENT);

-
-       /* Update the segment usage information. */
-       LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
-       sup->su_nbytes += nblocks - 1 - 
-           (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs) << fs->lfs_bshift;
-       sup->su_nbytes += ssp->ss_ninos * sizeof(struct dinode);
-       sup->su_lastmod = time.tv_sec;
-       LFS_UBWRITE(bp);
-       do_again = !(bp->b_flags & B_GATHERED);
+#ifdef DIAGNOSTIC
+       if (fs->lfs_bfree < fsbtodb(fs, ninos) + LFS_SUMMARY_SIZE / DEV_BSIZE)
+               panic("lfs_writeseg: No diskspace for summary");
+#endif
+       fs->lfs_bfree -= (fsbtodb(fs, ninos) + LFS_SUMMARY_SIZE / DEV_BSIZE);

 
        i_dev = VTOI(fs->lfs_ivnode)->i_dev;
        strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)];
 
        i_dev = VTOI(fs->lfs_ivnode)->i_dev;
        strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)];


@@ -783,8 +812,8 @@ lfs_writeseg(fs, sp)
        /*
         * When we simply write the blocks we lose a rotation for every block
         * written.  To avoid this problem, we allocate memory in chunks, copy
        /*
         * When we simply write the blocks we lose a rotation for every block
         * written.  To avoid this problem, we allocate memory in chunks, copy

-        * the buffers into the chunk and write the chunk.  56K was chosen as
-        * some driver/controllers can't handle unsigned 16 bit transfers.
+        * the buffers into the chunk and write the chunk.  MAXPHYS is the
+        * largest size I/O devices can handle.

         * When the data is copied to the chunk, turn off the the B_LOCKED bit
         * and brelse the buffer (which will move them to the LRU list).  Add
         * the B_CALL flag to the buffer header so we can count I/O's for the
         * When the data is copied to the chunk, turn off the the B_LOCKED bit
         * and brelse the buffer (which will move them to the LRU list).  Add
         * the B_CALL flag to the buffer header so we can count I/O's for the


@@ -795,8 +824,7 @@ lfs_writeseg(fs, sp)
         * easily make the buffers contiguous in kernel memory and if that's
         * fast enough.
         */
         * easily make the buffers contiguous in kernel memory and if that's
         * fast enough.
         */

-#define        LFS_CHUNKSIZE   (56 * 1024)
-       ch_per_blk = LFS_CHUNKSIZE / fs->lfs_bsize;
+       ch_per_blk = MAXPHYS / fs->lfs_bsize;

        for (bpp = sp->bpp, i = nblocks; i;) {
                num = ch_per_blk;
                if (num > i)
        for (bpp = sp->bpp, i = nblocks; i;) {
                num = ch_per_blk;
                if (num > i)


@@ -804,70 +832,112 @@ lfs_writeseg(fs, sp)
                i -= num;
                size = num * fs->lfs_bsize;
 
                i -= num;
                size = num * fs->lfs_bsize;
 

-               cbp = lfs_newbuf(fs, (*bpp)->b_blkno, 0);
+               cbp = lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp,
+                   (*bpp)->b_blkno, size);

                cbp->b_dev = i_dev;
                cbp->b_dev = i_dev;

-               cbp->b_flags = B_ASYNC | B_BUSY | B_CALL;
-               cbp->b_iodone = lfs_callback;
-               cbp->b_saveaddr = cbp->b_un.b_addr;
-               cbp->b_un.b_addr = malloc(size, M_SEGMENT, M_WAITOK);
+               cbp->b_flags |= B_ASYNC | B_BUSY;

 
                s = splbio();
                ++fs->lfs_iocount;
 
                s = splbio();
                ++fs->lfs_iocount;

-               for (p = cbp->b_un.b_addr; num--;) {
+               for (p = cbp->b_data; num--;) {

                        bp = *bpp++;
                        bp = *bpp++;

-                       bcopy(bp->b_un.b_addr, p, bp->b_bcount);
+                       /*
+                        * Fake buffers from the cleaner are marked as B_INVAL.
+                        * We need to copy the data from user space rather than
+                        * from the buffer indicated.
+                        * XXX == what do I do on an error?
+                        */
+                       if (bp->b_flags & B_INVAL) {
+                               if (copyin(bp->b_saveaddr, p, bp->b_bcount))
+                                       panic("lfs_writeseg: copyin failed");
+                       } else
+                               bcopy(bp->b_data, p, bp->b_bcount);

                        p += bp->b_bcount;
                        p += bp->b_bcount;

-                       bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI |
+                       if (bp->b_flags & B_LOCKED)
+                               --locked_queue_count;
+                       bp->b_flags &= ~(B_ERROR | B_READ | B_DELWRI |

                             B_LOCKED | B_GATHERED);
                             B_LOCKED | B_GATHERED);

-                       if (!(bp->b_flags & (B_NOCACHE | B_INVAL))) {
+                       if (bp->b_flags & B_CALL) {
+                               /* if B_CALL, it was created with newbuf */
+                               brelvp(bp);
+                               if (!(bp->b_flags & B_INVAL))
+                                       free(bp->b_data, M_SEGMENT);
+                               free(bp, M_SEGMENT);
+                       } else {

                                bremfree(bp);
                                bremfree(bp);

+                               bp->b_flags |= B_DONE;

                                reassignbuf(bp, bp->b_vp);
                                reassignbuf(bp, bp->b_vp);

+                               brelse(bp);

                        }
                        }

-                       brelse(bp);

                }
                }

+               ++cbp->b_vp->v_numoutput;

                splx(s);
                splx(s);

-               cbp->b_bcount = p - cbp->b_un.b_addr;
+               cbp->b_bcount = p - (char *)cbp->b_data;
+               /*
+                * XXXX This is a gross and disgusting hack.  Since these
+                * buffers are physically addressed, they hang off the
+                * device vnode (devvp).  As a result, they have no way
+                * of getting to the LFS superblock or lfs structure to
+                * keep track of the number of I/O's pending.  So, I am
+                * going to stuff the fs into the saveaddr field of
+                * the buffer (yuk).
+                */
+               cbp->b_saveaddr = (caddr_t)fs;

                vop_strategy_a.a_desc = VDESC(vop_strategy);
                vop_strategy_a.a_bp = cbp;
                (strategy)(&vop_strategy_a);
        }
                vop_strategy_a.a_desc = VDESC(vop_strategy);
                vop_strategy_a.a_bp = cbp;
                (strategy)(&vop_strategy_a);
        }

-       return(do_again);
+       /*
+        * XXX
+        * Vinvalbuf can move locked buffers off the locked queue
+        * and we have no way of knowing about this.  So, after
+        * doing a big write, we recalculate how many bufers are
+        * really still left on the locked queue.
+        */
+       locked_queue_count = count_lock_queue();
+       wakeup(&locked_queue_count);
+#ifdef DOSTATS
+       ++lfs_stats.psegwrites;
+       lfs_stats.blocktot += nblocks - 1;
+       if (fs->lfs_sp->seg_flags & SEGM_SYNC)
+               ++lfs_stats.psyncwrites;
+       if (fs->lfs_sp->seg_flags & SEGM_CLEAN) {
+               ++lfs_stats.pcleanwrites;
+               lfs_stats.cleanblocks += nblocks - 1;
+       }
+#endif
+       return (lfs_initseg(fs) || do_again);

 }
 
 void
 }
 
 void

-lfs_writesuper(fs, sp)
+lfs_writesuper(fs)

        struct lfs *fs;
        struct lfs *fs;

-       struct segment *sp;

 {
 {

-       USES_VOP_STRATEGY;

        struct buf *bp;
        dev_t i_dev;
        int (*strategy) __P((struct vop_strategy_args *));
        struct buf *bp;
        dev_t i_dev;
        int (*strategy) __P((struct vop_strategy_args *));

+       int s;
+       struct vop_strategy_args vop_strategy_a;

 
 

-#ifdef VERBOSE
-       printf("lfs_writesuper\n");
-#endif

        i_dev = VTOI(fs->lfs_ivnode)->i_dev;
        strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)];
 
        /* Checksum the superblock and copy it into a buffer. */
        fs->lfs_cksum = cksum(fs, sizeof(struct lfs) - sizeof(fs->lfs_cksum));
        i_dev = VTOI(fs->lfs_ivnode)->i_dev;
        strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)];
 
        /* Checksum the superblock and copy it into a buffer. */
        fs->lfs_cksum = cksum(fs, sizeof(struct lfs) - sizeof(fs->lfs_cksum));

-       bp = lfs_newbuf(fs, fs->lfs_sboffs[0], LFS_SBPAD);
-       *bp->b_un.b_lfs = *fs;
+       bp = lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, fs->lfs_sboffs[0],
+           LFS_SBPAD);
+       *(struct lfs *)bp->b_data = *fs;

 
 

-       /* Write the first superblock (wait). */
+       /* XXX Toggle between first two superblocks; for now just write first */

        bp->b_dev = i_dev;
        bp->b_dev = i_dev;

-       bp->b_flags |= B_BUSY;
+       bp->b_flags |= B_BUSY | B_CALL | B_ASYNC;

        bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI);
        bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI);

+       bp->b_iodone = lfs_supercallback;

        vop_strategy_a.a_desc = VDESC(vop_strategy);
        vop_strategy_a.a_bp = bp;
        vop_strategy_a.a_desc = VDESC(vop_strategy);
        vop_strategy_a.a_bp = bp;

-       (strategy)(&vop_strategy_a);
-       biowait(bp);
-
-       /* Write the second superblock (don't wait). */
-       bp->b_blkno = bp->b_lblkno = fs->lfs_sboffs[1];
-       bp->b_flags |= B_ASYNC | B_BUSY;
-       bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI);
+       s = splbio();
+       ++bp->b_vp->v_numoutput;
+       splx(s);

        (strategy)(&vop_strategy_a);
 }
 
        (strategy)(&vop_strategy_a);
 }
 


@@ -920,29 +990,28 @@ lfs_match_tindir(fs, bp)
  * Allocate a new buffer header.
  */
 struct buf *
  * Allocate a new buffer header.
  */
 struct buf *

-lfs_newbuf(fs, daddr, size)
-       struct lfs *fs;
+lfs_newbuf(vp, daddr, size)
+       struct vnode *vp;

        daddr_t daddr;
        size_t size;
 {
        struct buf *bp;
        daddr_t daddr;
        size_t size;
 {
        struct buf *bp;

-
-#ifdef VERBOSE
-       printf("lfs_newbuf\n");
-#endif
-       bp = getnewbuf();
-       bremhash(bp);
-       bgetvp(fs->lfs_ivnode, bp);
-       bp->b_bcount = 0;
+       size_t nbytes;
+
+       nbytes = roundup(size, DEV_BSIZE);
+       bp = malloc(sizeof(struct buf), M_SEGMENT, M_WAITOK);
+       bzero(bp, sizeof(struct buf));
+       if (nbytes)
+               bp->b_data = malloc(nbytes, M_SEGMENT, M_WAITOK);
+       bgetvp(vp, bp);
+       bp->b_bufsize = size;
+       bp->b_bcount = size;

        bp->b_lblkno = daddr;
        bp->b_blkno = daddr;
        bp->b_error = 0;
        bp->b_resid = 0;
        bp->b_lblkno = daddr;
        bp->b_blkno = daddr;
        bp->b_error = 0;
        bp->b_resid = 0;

-       if (size)
-               allocbuf(bp, size);
-       bp->b_flags |= B_NOCACHE;
-       bp->b_saveaddr = NULL;
-       binshash(bp, &bfreelist[BQ_AGE]);
+       bp->b_iodone = lfs_callback;
+       bp->b_flags |= B_BUSY | B_CALL | B_NOCACHE;

        return (bp);
 }
 
        return (bp);
 }
 


@@ -952,7 +1021,7 @@ lfs_callback(bp)
 {
        struct lfs *fs;
 
 {
        struct lfs *fs;
 

-       fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs;
+       fs = (struct lfs *)bp->b_saveaddr;

 #ifdef DIAGNOSTIC
        if (fs->lfs_iocount == 0)
                panic("lfs_callback: zero iocount\n");
 #ifdef DIAGNOSTIC
        if (fs->lfs_iocount == 0)
                panic("lfs_callback: zero iocount\n");


@@ -960,12 +1029,18 @@ lfs_callback(bp)
        if (--fs->lfs_iocount == 0)
                wakeup(&fs->lfs_iocount);
 
        if (--fs->lfs_iocount == 0)
                wakeup(&fs->lfs_iocount);
 

-       if (bp->b_saveaddr) {
-               free(bp->b_un.b_addr, M_SEGMENT);
-               bp->b_un.b_addr = bp->b_saveaddr;
-               bp->b_saveaddr = NULL;
-       }
-       brelse(bp);
+       brelvp(bp);
+       free(bp->b_data, M_SEGMENT);
+       free(bp, M_SEGMENT);
+}
+
+void
+lfs_supercallback(bp)
+       struct buf *bp;
+{
+       brelvp(bp);
+       free(bp->b_data, M_SEGMENT);
+       free(bp, M_SEGMENT);

 }
 
 /*
 }
 
 /*


@@ -1006,3 +1081,31 @@ lfs_shellsort(bp_array, lb_array, nmemb)
                                } else
                                        break;
 }
                                } else
                                        break;
 }

+
+/*
+ * Check VXLOCK.  Return 1 if the vnode is locked.  Otherwise, vget it.
+ */
+lfs_vref(vp)
+       register struct vnode *vp;
+{
+
+       if (vp->v_flag & VXLOCK)
+               return(1);
+       return (vget(vp, 0));
+}
+
+void
+lfs_vunref(vp)
+       register struct vnode *vp;
+{
+       extern int lfs_no_inactive;
+
+       /*
+        * This is vrele except that we do not want to VOP_INACTIVE
+        * this vnode. Rather than inline vrele here, we use a global
+        * flag to tell lfs_inactive not to run. Yes, its gross.
+        */
+       lfs_no_inactive = 1;
+       vrele(vp);
+       lfs_no_inactive = 0;
+}