Re: GFS2 and DLM

[Ingo Molnar <mingo@elte.hu>]

* Christoph Hellwig <hch@infradead.org> wrote:

> ->follow_link needs exactly the same locking as ->readlink.  The whole 
> point of using generic_readlink is to avoid having the filesystem 
> reimplement almost the same code twice, once copying to a kernel 
> buffer and once to a user buffer.

yeah, you are right, i confused it with ->follow_link() and was wrong 
about the locking: generic_readlink() is just a wrapper around 
->follow_link() and vfs_readlink().

Still, as far as i can see the gfs2 implementation of readlink is faster 
(and hence a valid solution), because it knows the length of the symlink 
buffer and hence can avoid the strlen() call in vfs_readlink():

 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
 {
         int len;

         len = PTR_ERR(link);
         if (IS_ERR(link))
                 goto out;

         len = strlen(link); <============= [this one]

while gfs2 can do a straight copy to userspace:

        error = gfs2_readlinki(ip, &buf, &len);
        if (error)
                return error;

        if (user_size > len - 1)
                user_size = len - 1;

        if (copy_to_user(user_buf, buf, user_size))
                error = -EFAULT;
        else
                error = user_size;

btw., ocfs2 does not use generic_readlink() either.

> Please read the code before giving such useless comments.  

thank you for the encouragement to participate in VFS review activities, 
it's really appreciated! It's always a joy taking part in lkml 
discussions.

	Ingo