Re: 64fa03de33: BUG:Dentry_still_in_use

["Serge E. Hallyn" <serge@hallyn.com>]

Quoting Masami Ichikawa (masami256@gmail.com):
> On Mon, May 8, 2017 at 1:44 PM, Serge E. Hallyn <serge@hallyn.com> wrote:
> > From 6a3fb632f67f8425c6e76c65dad8115f1550d2a0 Mon Sep 17 00:00:00 2001
> > From: Serge Hallyn <serge@hallyn.com>
> > Date: Sun, 7 May 2017 23:40:42 -0500
> > Subject: [PATCH 1/1] cap_inode_getsecurity: don't pin dentry (fold up)
> >
> > This should fix the "Dentry_still_in_use" reported by the kernel
> > test robot.
> >
> > Signed-off-by: Serge Hallyn <serge@hallyn.com>
> > ---
> >  security/commoncap.c | 27 +++++++++++++++------------
> >  1 file changed, 15 insertions(+), 12 deletions(-)
> >
> > diff --git a/security/commoncap.c b/security/commoncap.c
> > index a1a2935..c970b71 100644
> > --- a/security/commoncap.c
> > +++ b/security/commoncap.c
> > @@ -406,21 +406,21 @@ int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
> >                                  &tmpbuf, size, GFP_NOFS);
> >
> >         if (ret < 0)
> > -               return ret;
> > +               goto out;
> >
> >         fs_ns = inode->i_sb->s_user_ns;
> >         cap = (struct vfs_cap_data *) tmpbuf;
> >         if (is_v2header(ret, cap->magic_etc)) {
> >                 /* If this is sizeof(vfs_cap_data) then we're ok with the
> >                  * on-disk value, so return that.  */
> > -               if (alloc)
> > +               if (alloc) {
> >                         *buffer = tmpbuf;
> > -               else
> > -                       kfree(tmpbuf);
> > -               return ret;
> > +                       tmpbuf = NULL;
> > +               }
> > +               goto out;
> >         } else if (!is_v3header(ret, cap->magic_etc)) {
> > -               kfree(tmpbuf);
> > -               return -EINVAL;
> > +               ret = -EINVAL;
> > +               goto out;
> >         }
> >
> >         nscap = (struct vfs_ns_cap_data *) tmpbuf;
> > @@ -434,14 +434,14 @@ int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
> >                 if (alloc) {
> >                         *buffer = tmpbuf;
> >                         nscap->rootid = cpu_to_le32(mappedroot);
> > -               } else
> > -                       kfree(tmpbuf);
> > -               return size;
> > +                       tmpbuf = NULL;
> > +               }
> > +               goto out;
> >         }
> >
> >         if (!rootid_owns_currentns(kroot)) {
> > -               kfree(tmpbuf);
> > -               return -EOPNOTSUPP;
> > +               ret = -EOPNOTSUPP;
> > +               goto out;
> >         }
> >
> >         /* This comes from a parent namespace.  Return as a v2 capability */
> > @@ -459,6 +459,9 @@ int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
> >                         cap->magic_etc = cpu_to_le32(magic);
> >                 }
> >         }
> > +
> > +out:
> > +       dput(dentry);
> >         kfree(tmpbuf);
> >         return size;
> 
> If ret is set to some error code, e.g. -EINVAL, then jump to out
> label, this function should return error code, doesn't it?

D'oh, yes.

Sorry, I'd wanted to ack the fact that there was a real bug before I went to bed.

I'll send a proper patch tonight.