When a new block is allocated to a directory, an fsync of a file

whose name is within that block must ensure not only that the block
containing the file name has been written, but also that the on-disk
directory inode references that block. When a new directory block
is created, we allocate a newdirblk structure which is linked to
the associated allocdirect (on its ad_newdirblk list). When the
allocdirect has been satisfied, the newdirblk structure is moved
to the inodedep id_bufwait list of its directory to await the inode
being written.  When the inode is written, the directory entries
are fully committed and can be deleted from their pagedep->id_pendinghd
and inodedep->id_pendinghd lists.

2 files changed, 28 insertions, 8 deletions

diff --git a/sys/ufs/ufs/ufs_extern.h b/sys/ufs/ufs/ufs_extern.h
index e08ee8d..d95013b 100644
--- a/sys/ufs/ufs/ufs_extern.h
+++ b/sys/ufs/ufs/ufs_extern.h
@@ -95,8 +95,8 @@ int	 ufs_vinit __P((struct mount *, vop_t **, vop_t **, struct vnode **));
 /*
  * Soft update function prototypes.
  */
-void	softdep_setup_directory_add __P((struct buf *, struct inode *, off_t,
-	    long, struct buf *));
+int	softdep_setup_directory_add __P((struct buf *, struct inode *, off_t,
+	    long, struct buf *, int));
 void	softdep_change_directoryentry_offset __P((struct inode *, caddr_t,
 	    caddr_t, caddr_t, int));
 void	softdep_setup_remove __P((struct buf *,struct inode *, struct inode *,
diff --git a/sys/ufs/ufs/ufs_lookup.c b/sys/ufs/ufs/ufs_lookup.c
index 5acc85d..61a620d 100644
--- a/sys/ufs/ufs/ufs_lookup.c
+++ b/sys/ufs/ufs/ufs_lookup.c
@@ -745,10 +745,29 @@ ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
 				   (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
 				blkoff += DIRBLKSIZ;
 			}
-			softdep_setup_directory_add(bp, dp, dp->i_offset,
-			    dirp->d_ino, newdirbp);
-			bdwrite(bp);
-			return (UFS_UPDATE(dvp, 0));
+			if (softdep_setup_directory_add(bp, dp, dp->i_offset,
+			    dirp->d_ino, newdirbp, 1) == 0) {
+				bdwrite(bp);
+				return (UFS_UPDATE(dvp, 0));
+			}
+			/* We have just allocated a directory block in an
+			 * indirect block. Rather than tracking when it gets
+			 * claimed by the inode, we simply do a VOP_FSYNC
+			 * now to ensure that it is there (in case the user
+			 * does a future fsync). Note that we have to unlock
+			 * the inode for the entry that we just entered, as
+			 * the VOP_FSYNC may need to lock other inodes which
+			 * can lead to deadlock if we also hold a lock on
+			 * the newly entered node.
+			 */
+			if ((error = BUF_WRITE(bp)))
+				return (error);
+			if (tvp != NULL)
+				VOP_UNLOCK(tvp, 0, p);
+			error = VOP_FSYNC(dvp, p->p_ucred, MNT_WAIT, p);
+			if (tvp != NULL)
+				vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
+			return (error);
 		}
 		if (DOINGASYNC(dvp)) {
 			bdwrite(bp);
@@ -836,8 +855,9 @@ ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
 	bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
 
 	if (DOINGSOFTDEP(dvp)) {
-		softdep_setup_directory_add(bp, dp,
-		    dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
+		(void) softdep_setup_directory_add(bp, dp,
+		    dp->i_offset + (caddr_t)ep - dirbuf,
+		    dirp->d_ino, newdirbp, 0);
 		bdwrite(bp);
 	} else {
 		if (DOINGASYNC(dvp)) {