[PATCH 3/3] pipefs and sockfs respin -- make .drop_inode = generic_delete_inode

[PATCH 3/3] pipefs and sockfs respin -- make .drop_inode = generic_delete_inode

[Jeff Layton]

A little late since I sent some patches to akpm already, but I'm thinking
that this might be the better way to go. Rather than trying to have the
filesystems manage i_nlink, just make drop_inode for the filesystems be
generic_delete_inode. It seems that some filesystems already do that, and it
seems like that should do the right thing here too. This seems to be a little
easier to follow than messing with i_nlink.

Does this seem reasonable? If so, I'll send this on to akpm and ask him to
commit this rather than the original pipefs patch I sent.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

diff --git a/fs/pipe.c b/fs/pipe.c
index 68090e8..9b3cb34 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -871,6 +871,8 @@ static struct inode * get_pipe_inode(void)
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	inode->i_ino = iunique(pipe_mnt->mnt_sb, 1);
+	insert_inode_hash(inode);
 
 	return inode;
 
@@ -1003,6 +1005,11 @@ int do_pipe(int *fd)
 	return error;
 }
 
+static struct super_operations pipefs_sops = {
+	.statfs		= simple_statfs,
+	.drop_inode	= generic_delete_inode,
+};
+
 /*
  * pipefs should _never_ be mounted by userland - too much of security hassle,
  * no real gain from having the whole whorehouse mounted. So we don't need
@@ -1013,7 +1020,7 @@ static int pipefs_get_sb(struct file_system_type *fs_type,
 			 int flags, const char *dev_name, void *data,
 			 struct vfsmount *mnt)
 {
-	return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
+	return get_sb_pseudo(fs_type, "pipe:", &pipefs_sops, PIPEFS_MAGIC, mnt);
 }
 
 static struct file_system_type pipe_fs_type = {
diff --git a/net/socket.c b/net/socket.c
index 4e39631..20193a8 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -284,6 +284,7 @@ static int init_inodecache(void)
 static struct super_operations sockfs_ops = {
 	.alloc_inode =	sock_alloc_inode,
 	.destroy_inode =sock_destroy_inode,
+	.drop_inode =	generic_delete_inode,
 	.statfs =	simple_statfs,
 };
 
@@ -495,6 +496,8 @@ static struct socket *sock_alloc(void)
 	inode->i_mode = S_IFSOCK | S_IRWXUGO;
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
+	inode->i_ino = iunique(sock_mnt->mnt_sb, 1);
+	insert_inode_hash(inode);
 
 	get_cpu_var(sockets_in_use)++;
 	put_cpu_var(sockets_in_use);

Re: [PATCH 3/3] pipefs and sockfs respin -- make .drop_inode = generic_delete_inode

[Christoph Hellwig]

On Mon, Jan 22, 2007 at 09:16:06AM -0500, Jeff Layton wrote:
> A little late since I sent some patches to akpm already, but I'm thinking
> that this might be the better way to go. Rather than trying to have the
> filesystems manage i_nlink, just make drop_inode for the filesystems be
> generic_delete_inode. It seems that some filesystems already do that, and it
> seems like that should do the right thing here too. This seems to be a little
> easier to follow than messing with i_nlink.
> 
> Does this seem reasonable? If so, I'll send this on to akpm and ask him to
> commit this rather than the original pipefs patch I sent.

Sounds good to me.