procfs: get rid of ancient BS in pid_revalidate() uses

First of all, calling pid_revalidate() in the end of <pid>/* lookups
is *not* about closing any kind of races; that used to be true once
upon a time, but these days those comments are actively misleading.
Especially since pid_revalidate() doesn't even do d_drop() on
failure anymore.  It doesn't matter, anyway, since once
pid_revalidate() starts returning false, ->d_delete() of those
dentries starts saying "don't keep"; they won't get stuck in
dcache any longer than they are pinned.

These calls cannot be just removed, though - the side effect of
pid_revalidate() (updating i_uid/i_gid/etc.) is what we are calling
it for here.

Let's separate the "update ownership" into a new helper (pid_update_inode())
and use it, both in lookups and in pid_revalidate() itself.

The comments in pid_revalidate() are also out of date - they refer to
the time when pid_revalidate() used to call d_drop() directly...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

1 files changed, 3 insertions, 6 deletions

diff --git a/fs/proc/namespaces.c b/fs/proc/namespaces.c
index 59b17e509f46..ad1adce6541d 100644
--- a/fs/proc/namespaces.c
+++ b/fs/proc/namespaces.c
@@ -96,19 +96,16 @@ static int proc_ns_instantiate(struct inode *dir,
 
 	inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK | S_IRWXUGO);
 	if (!inode)
-		goto out;
+		return -ENOENT;
 
 	ei = PROC_I(inode);
 	inode->i_op = &proc_ns_link_inode_operations;
 	ei->ns_ops = ns_ops;
+	pid_update_inode(task, inode);
 
 	d_set_d_op(dentry, &pid_dentry_operations);
 	d_add(dentry, inode);
-	/* Close the race of the process dying before we return the dentry */
-	if (pid_revalidate(dentry, 0))
-		return 0;
-out:
-	return -ENOENT;
+	return 0;
 }
 
 static int proc_ns_dir_readdir(struct file *file, struct dir_context *ctx)