Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1428845 > unrolled thread

[PATCH 5/8] fuse: don't use ->d_time

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH 5/8] fuse: don't use ->d_time Miklos Szeredi <mszeredi@redhat.com> - 2016-06-22 16:40 +0200
    Re: [PATCH 5/8] fuse: don't use ->d_time Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-22 18:50 +0200

#1428845 — [PATCH 5/8] fuse: don't use ->d_time

Store in memory pointed to by ->d_fsdata.  Use ->d_allocate() to allocate
the storage.

We could cast ->d_fsdata directly on 64bit archs, but I don't think this is
worth the extra complexity.

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
 fs/fuse/dir.c | 36 ++++++++++++++++--------------------
 1 file changed, 16 insertions(+), 20 deletions(-)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index ccd4971cc6c1..1a1dabb72036 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -37,37 +37,20 @@ static void fuse_advise_use_readdirplus(struct inode *dir)
 	set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
 }
 
-#if BITS_PER_LONG >= 64
 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
 {
-	entry->d_time = time;
+	*(u64 *) entry->d_fsdata = time;
 }
 
 static inline u64 fuse_dentry_time(struct dentry *entry)
 {
-	return entry->d_time;
+	return *(u64 *) entry->d_fsdata;
 }
-#else
-/*
- * On 32 bit archs store the high 32 bits of time in d_fsdata
- */
-static void fuse_dentry_settime(struct dentry *entry, u64 time)
-{
-	entry->d_time = time;
-	entry->d_fsdata = (void *) (unsigned long) (time >> 32);
-}
-
-static u64 fuse_dentry_time(struct dentry *entry)
-{
-	return (u64) entry->d_time +
-		((u64) (unsigned long) entry->d_fsdata << 32);
-}
-#endif
 
 /*
  * FUSE caches dentries and attributes with separate timeout.  The
  * time in jiffies until the dentry/attributes are valid is stored in
- * dentry->d_time and fuse_inode->i_time respectively.
+ * dentry->d_fsdata and fuse_inode->i_time respectively.
  */
 
 /*
@@ -272,8 +255,21 @@ static int invalid_nodeid(u64 nodeid)
 	return !nodeid || nodeid == FUSE_ROOT_ID;
 }
 
+static int fuse_dentry_allocate(struct dentry *dentry)
+{
+	dentry->d_fsdata = kzalloc(sizeof(u64), GFP_KERNEL);
+
+	return dentry->d_fsdata ? 0 : -ENOMEM;
+}
+static void fuse_dentry_release(struct dentry *dentry)
+{
+	kfree(dentry->d_fsdata);
+}
+
 const struct dentry_operations fuse_dentry_operations = {
 	.d_revalidate	= fuse_dentry_revalidate,
+	.d_allocate	= fuse_dentry_allocate,
+	.d_release	= fuse_dentry_release,
 };
 
 int fuse_valid_type(int m)
-- 
2.5.5

[toc] | [next] | [standalone]

#1428974

On Wed, Jun 22, 2016 at 04:35:07PM +0200, Miklos Szeredi wrote:
> Store in memory pointed to by ->d_fsdata.  Use ->d_allocate() to allocate
> the storage.
> 
> We could cast ->d_fsdata directly on 64bit archs, but I don't think this is
> worth the extra complexity.

Now, _that_ is interesting:

> +static void fuse_dentry_release(struct dentry *dentry)
> +{
> +	kfree(dentry->d_fsdata);
> +}

What happens to fuse_dentry_revalidate() called on dentry in process of
getting dropped?  Unlike freeing struct dentry itself, ->d_release() is
not RCU-delayed.  So you are risking dereference of ->d_fsdata after
kfree(); at the very least, it needs RCU-delayed freeing...

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web