Re: [git pull] vfs fixes

[Al Viro <viro@ZenIV.linux.org.uk>]

On Mon, Apr 03, 2017 at 01:30:45AM +0100, Al Viro wrote:

> Currently true and almost certainly will remain so.  Point taken, what you
> are suggesting is better.  Actually, the invariant to watch for is
> "no d_can_lookup() withtout DCACHE_RCUACCESS" and that we can trivially
> enforce by one-liner change in d_flags_for_inode() -
> s/DCACHE_DIRECTORY_TYPE/& | DCACHE_RCUACCESS/
> 
> OK...

Do you have any objections against the following (still untested) variant?
I don't see any point in checking for flags & LOOKUP_RCU in case of !*s -
flags is in register at that point, so...

diff --git a/fs/dcache.c b/fs/dcache.c
index 95d71eda8142..05550139a8a6 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -1757,7 +1757,13 @@ static unsigned d_flags_for_inode(struct inode *inode)
 		return DCACHE_MISS_TYPE;
 
 	if (S_ISDIR(inode->i_mode)) {
-		add_flags = DCACHE_DIRECTORY_TYPE;
+		/*
+		 * Any potential starting point of lookup should have
+		 * DCACHE_RCUACCESS; currently directory dentries
+		 * come from d_alloc() anyway, but it costs us nothing
+		 * to enforce it here.
+		 */
+		add_flags = DCACHE_DIRECTORY_TYPE | DCACHE_RCUACCESS;
 		if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
 			if (unlikely(!inode->i_op->lookup))
 				add_flags = DCACHE_AUTODIR_TYPE;
diff --git a/fs/namei.c b/fs/namei.c
index d41fab78798b..19dcf62133cc 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2145,6 +2145,9 @@ static const char *path_init(struct nameidata *nd, unsigned flags)
 	int retval = 0;
 	const char *s = nd->name->name;
 
+	if (!*s)
+		flags &= ~LOOKUP_RCU;
+
 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
 	nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
 	nd->depth = 0;