]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
[PATCH] autofs4: simplify expire tree traversal
authorIan Kent <raven@themaw.net>
Mon, 27 Mar 2006 09:14:45 +0000 (01:14 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Mon, 27 Mar 2006 16:44:39 +0000 (08:44 -0800)
Simplify the expire tree traversal code by using a function from namespace.c
to calculate the next entry in the top down tree traversals carried out during
the expire operation.

Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
fs/autofs4/expire.c

index bcc17f533699728658ea6833a4ec0e160f5ce65d..165fe9e2d5705916689029d2b04e80b90b90c8a2 100644 (file)
@@ -72,6 +72,27 @@ done:
        return status;
 }
 
+/*
+ * Calculate next entry in top down tree traversal.
+ * From next_mnt in namespace.c - elegant.
+ */
+static struct dentry *next_dentry(struct dentry *p, struct dentry *root)
+{
+       struct list_head *next = p->d_subdirs.next;
+
+       if (next == &p->d_subdirs) {
+               while (1) {
+                       if (p == root)
+                               return NULL;
+                       next = p->d_u.d_child.next;
+                       if (next != &p->d_parent->d_subdirs)
+                               break;
+                       p = p->d_parent;
+               }
+       }
+       return list_entry(next, struct dentry, d_u.d_child);
+}
+
 /* Check a directory tree of mount points for busyness
  * The tree is not busy iff no mountpoints are busy
  */
@@ -80,8 +101,7 @@ static int autofs4_tree_busy(struct vfsmount *mnt,
                             unsigned long timeout,
                             int do_now)
 {
-       struct dentry *this_parent = top;
-       struct list_head *next;
+       struct dentry *p;
 
        DPRINTK("top %p %.*s",
                top, (int)top->d_name.len, top->d_name.name);
@@ -99,49 +119,28 @@ static int autofs4_tree_busy(struct vfsmount *mnt,
                return 1;
 
        spin_lock(&dcache_lock);
-repeat:
-       next = this_parent->d_subdirs.next;
-resume:
-       while (next != &this_parent->d_subdirs) {
-               struct dentry *dentry = list_entry(next, struct dentry, d_u.d_child);
-
+       for (p = top; p; p = next_dentry(p, top)) {
                /* Negative dentry - give up */
-               if (!simple_positive(dentry)) {
-                       next = next->next;
+               if (!simple_positive(p))
                        continue;
-               }
 
                DPRINTK("dentry %p %.*s",
-                       dentry, (int)dentry->d_name.len, dentry->d_name.name);
-
-               if (!simple_empty_nolock(dentry)) {
-                       this_parent = dentry;
-                       goto repeat;
-               }
+                       p, (int) p->d_name.len, p->d_name.name);
 
-               dentry = dget(dentry);
+               p = dget(p);
                spin_unlock(&dcache_lock);
 
-               if (d_mountpoint(dentry)) {
+               if (d_mountpoint(p)) {
                        /* First busy => tree busy */
-                       if (autofs4_mount_busy(mnt, dentry)) {
-                               dput(dentry);
+                       if (autofs4_mount_busy(mnt, p)) {
+                               dput(p);
                                return 1;
                        }
                }
-
-               dput(dentry);
+               dput(p);
                spin_lock(&dcache_lock);
-               next = next->next;
-       }
-
-       if (this_parent != top) {
-               next = this_parent->d_u.d_child.next;
-               this_parent = this_parent->d_parent;
-               goto resume;
        }
        spin_unlock(&dcache_lock);
-
        return 0;
 }
 
@@ -150,59 +149,38 @@ static struct dentry *autofs4_check_leaves(struct vfsmount *mnt,
                                           unsigned long timeout,
                                           int do_now)
 {
-       struct dentry *this_parent = parent;
-       struct list_head *next;
+       struct dentry *p;
 
        DPRINTK("parent %p %.*s",
                parent, (int)parent->d_name.len, parent->d_name.name);
 
        spin_lock(&dcache_lock);
-repeat:
-       next = this_parent->d_subdirs.next;
-resume:
-       while (next != &this_parent->d_subdirs) {
-               struct dentry *dentry = list_entry(next, struct dentry, d_u.d_child);
-
+       for (p = parent; p; p = next_dentry(p, parent)) {
                /* Negative dentry - give up */
-               if (!simple_positive(dentry)) {
-                       next = next->next;
+               if (!simple_positive(p))
                        continue;
-               }
 
                DPRINTK("dentry %p %.*s",
-                       dentry, (int)dentry->d_name.len, dentry->d_name.name);
+                       p, (int) p->d_name.len, p->d_name.name);
 
-               if (!list_empty(&dentry->d_subdirs)) {
-                       this_parent = dentry;
-                       goto repeat;
-               }
-
-               dentry = dget(dentry);
+               p = dget(p);
                spin_unlock(&dcache_lock);
 
-               if (d_mountpoint(dentry)) {
+               if (d_mountpoint(p)) {
                        /* Can we expire this guy */
-                       if (!autofs4_can_expire(dentry, timeout, do_now))
+                       if (!autofs4_can_expire(p, timeout, do_now))
                                goto cont;
 
                        /* Can we umount this guy */
-                       if (!autofs4_mount_busy(mnt, dentry))
-                               return dentry;
+                       if (!autofs4_mount_busy(mnt, p))
+                               return p;
 
                }
 cont:
-               dput(dentry);
+               dput(p);
                spin_lock(&dcache_lock);
-               next = next->next;
-       }
-
-       if (this_parent != parent) {
-               next = this_parent->d_u.d_child.next;
-               this_parent = this_parent->d_parent;
-               goto resume;
        }
        spin_unlock(&dcache_lock);
-
        return NULL;
 }