Re: [gobo-l]Re: [PATCH] gobohide: avoid null pointer accesses

[Felipe W Damasio <felipewd@terra.com.br>]

	Greetings,

Lucas Correia Villa Real wrote:
> Oops, sorry. 2.4.20 and/or 2.4.21.

	Ok, things are much clearer now :)

	But I still have some doubts..CC'ing LKML to see if any of those 
hackers can help us.

> int vfs_unlink(struct inode *dir, struct dentry *dentry)
> {
> 	...
> 	down(&dir->i_zombie);
> 	error = may_delete(dir, dentry, 0);
> 	if (!error) {
> 		...
> 		if (!error) {
> 			if (dentry->d_inode && S_ISLNK (dentry->d_inode->i_mode))
> 				if (gobolinux_hide(dentry->d_inode->i_ino) > 0)
> 					gobolinux_inode_del(dentry->d_inode->i_ino)


	Yeah, ok...but I still don't get when a dentry doesn't have a valid 
d_inode why we don't return ENOENT like in sys_unlink:


slashes:
         error = !dentry->d_inode ? -ENOENT :
                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;


	Which, by the way, would be called _instead_ of calling vfs_unlink...so 
should we assume that the dentry _should_ have a valid dinode?

	You said that the kernel oops'ed when unlinking a symlink in a NFS 
partition, right?

	Does anybody know if, in this case (a symlink inside a NFS partition), 
the dentry really doesn't have a valid d_inode entry?

	Thanks,

Felipe
-- 
It's most certainly GNU/Linux, not Linux. Read more at
http://www.gnu.org/gnu/why-gnu-linux.html