linux-kernel - Re: [PATCH 3/7] nfsd: remove redundant i

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Sun, 24 Jul 2016 16:21:50 -0400
From:	"J. Bruce Fields" <bfields@...ldses.org>
To:	Al Viro <viro@...IV.linux.org.uk>
Cc:	"J. Bruce Fields" <bfields@...hat.com>,
	Oleg Drokin <green@...uxhacker.ru>,
	Jeff Layton <jlayton@...chiereds.net>,
	linux-nfs@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/7] nfsd: remove redundant i_lookup check

On Sun, Jul 24, 2016 at 03:23:07PM +0100, Al Viro wrote:
> On Sun, Jul 24, 2016 at 08:10:14AM -0400, J. Bruce Fields wrote:
> > On Sun, Jul 24, 2016 at 01:22:06AM +0100, Al Viro wrote:
> > > On Fri, Jul 22, 2016 at 01:48:52PM -0400, J. Bruce Fields wrote:
> > > > From: "J. Bruce Fields" <bfields@...hat.com>
> > > > 
> > > > I'm not sure why this was added.  It doesn't seem necessary, and no
> > > > other caller does this.
> > > 
> > > lookup_one_len() will explode if you call it for non-directory (==
> > > !d_can_lookup(), i.e. something without ->lookup()).  So unless the callers
> > > do guarantee that check being true, it *is* needed.
> > 
> > Both callers call fh_verify(.,.,S_IFDIR,.), so at this point we know
> > that i_mode & S_IFMT == S_IFDIR.  Is there some odd case where that's
> > insufficient?  If so, I think there may be bugs elsewhere in nfsd.  If
> > not, I'll add a note to the changelog.
> 
> First of all, such objects do exist; they probably won't be encountered by
> nfsd and all instances I can think of are not writable, but...
> 
> > Thanks for reminding me to check this, I hadn't thought of that as an
> > "is this a directory" check, it makes more sense now.
> 
> I'd have turned that into d_can_lookup(fhp->fh_dentry), actually.

So would such a check mainly just protect developers from themselves if
they try to make a weird filesystems exportable?

If we need to catch this I'd rather do it in fh_verify, which would
cover some other operations, too.  Maybe like the below.  We could be
nicer and WARN()/error out instead of BUG.  But it's unclear to me
whether this case is worth checking for at all.

--b.

diff --git a/fs/nfsd/nfsfh.c b/fs/nfsd/nfsfh.c
index 27250e279c37..372747a00214 100644
--- a/fs/nfsd/nfsfh.c
+++ b/fs/nfsd/nfsfh.c
@@ -59,14 +59,17 @@ static int nfsd_acceptable(void *expv, struct dentry *dentry)
  * the write call).
  */
 static inline __be32
-nfsd_mode_check(struct svc_rqst *rqstp, umode_t mode, umode_t requested)
+nfsd_mode_check(struct svc_rqst *rqstp, struct dentry *dentry,
+		umode_t requested)
 {
-	mode &= S_IFMT;
+	umode_t mode = d_inode(dentry)->i_mode & S_IFMT;
 
 	if (requested == 0) /* the caller doesn't care */
 		return nfs_ok;
-	if (mode == requested)
+	if (mode == requested) {
+		BUG_ON(mode == S_IFDIR && !d_can_lookup(dentry));
 		return nfs_ok;
+	}
 	/*
 	 * v4 has an error more specific than err_notdir which we should
 	 * return in preference to err_notdir:
@@ -340,7 +343,7 @@ fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
 	if (error)
 		goto out;
 
-	error = nfsd_mode_check(rqstp, d_inode(dentry)->i_mode, type);
+	error = nfsd_mode_check(rqstp, dentry, type);
 	if (error)
 		goto out;