Re: [External] Re: [PATCH] ovl: Do not override fsuid and fsgid in ovl_link()

From: Christian Brauner <brauner@kernel.org>
To: 天赐张 <zhangtianci.1997@bytedance.com>
Cc: Amir Goldstein <amir73il@gmail.com>,
	Miklos Szeredi <miklos@szeredi.hu>,
	overlayfs <linux-unionfs@vger.kernel.org>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>
Subject: Re: [External] Re: [PATCH] ovl: Do not override fsuid and fsgid in ovl_link()
Date: Wed, 17 Aug 2022 14:29:15 +0200	[thread overview]
Message-ID: <20220817122915.k6aps55sjr4hnm5j@wittgenstein> (raw)
In-Reply-To: <20220817115638.2etj6ruuutjurgjv@wittgenstein>

On Wed, Aug 17, 2022 at 01:56:43PM +0200, Christian Brauner wrote:
> On Wed, Aug 17, 2022 at 07:36:44PM +0800, 天赐张 wrote:
> > On Wed, Aug 17, 2022 at 6:29 PM Christian Brauner <brauner@kernel.org> wrote:
> > >
> > > On Wed, Aug 17, 2022 at 12:27:27PM +0200, Christian Brauner wrote:
> > > > On Wed, Aug 17, 2022 at 12:55:22PM +0300, Amir Goldstein wrote:
> > > > > On Wed, Aug 17, 2022 at 12:53 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > >
> > > > > > On Wed, Aug 17, 2022 at 12:11 PM 天赐张 <zhangtianci.1997@bytedance.com> wrote:
> > > > > > >
> > > > > > > On Wed, Aug 17, 2022 at 3:36 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > > >
> > > > > > > > On Wed, Aug 17, 2022 at 6:49 AM Zhang Tianci
> > > > > > > > <zhangtianci.1997@bytedance.com> wrote:
> > > > > > > > >
> > > > > > > > > ovl_link() did not create a new inode after commit
> > > > > > > > > 51f7e52dc943 ("ovl: share inode for hard link"), so
> > > > > > > > > in ovl_create_or_link() we should not override cred's
> > > > > > > > > fsuid and fsgid when called by ovl_link().
> > > > > > > > >
> > > > > > > > > Signed-off-by: Zhang Tianci <zhangtianci.1997@bytedance.com>
> > > > > > > > > Signed-off-by: Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>
> > > > > > > > > ---
> > > > > > > > >  fs/overlayfs/dir.c | 4 ++--
> > > > > > > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > > > > > >
> > > > > > > > > diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
> > > > > > > > > index 6b03457f72bb..568d338032db 100644
> > > > > > > > > --- a/fs/overlayfs/dir.c
> > > > > > > > > +++ b/fs/overlayfs/dir.c
> > > > > > > > > @@ -595,9 +595,9 @@ static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
> > > > > > > > >         err = -ENOMEM;
> > > > > > > > >         override_cred = prepare_creds();
> > > > > > > > >         if (override_cred) {
> > > > > > > > > -               override_cred->fsuid = inode->i_uid;
> > > > > > > > > -               override_cred->fsgid = inode->i_gid;
> > > > > > > > >                 if (!attr->hardlink) {
> > > > > > > > > +                       override_cred->fsuid = inode->i_uid;
> > > > > > > > > +                       override_cred->fsgid = inode->i_gid;
> > > > > > > > >                         err = security_dentry_create_files_as(dentry,
> > > > > > > > >                                         attr->mode, &dentry->d_name, old_cred,
> > > > > > > > >                                         override_cred);
> > > > > > > > > --
> > > > > > > >
> > > > > > > > This change looks incorrect.
> > > > > > > > Unless I am missing something, fsuid/fsgid still need to
> > > > > > > > be overridden for calling link() on underlying fs.
> > > > > > > > What made you do this change?
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > Amir.
> > > > > > >
> > > > > > > Hi Amir,
> > > > > > >
> > > > > > > I ran into an error when I tested overlay on fuse:
> > > > > > >   $ mkdir /lower /fuse /merge
> > > > > > >   $ mount -t fuse /fuse
> > > > > > >   $ mkdir /fuse/upper /fuse/work
> > > > > > >   $ mount -t overlay /merge -o lowerdir=/lower,upperdir=/fuse/upper,workdir=work
> > > > > > >   $ touch /merge/file
> > > > > > >   $ chown bin.bin /merge/file // the file's caller becomes "bin"
> > > > > > >   $ ln /merge/file /merge/lnkfile
> > > > > > >
> > > > > > > Then I got an error(EACCES) because fuse daemon checks the link()'s
> > > > > > > caller is "bin", it denied this request.
> > > > > > > I browsed the changing history of ovl_link(). There are two key commits:
> > > > > > > The first is commit bb0d2b8ad296 ("ovl: fix sgid on directory") which
> > > > > > > overrides the cred's fsuid/fsgid using the new inode. The new inode's
> > > > > > > owner is initialized by inode_init_owner(), and inode->fsuid is
> > > > > > > assigned to the current user. So the override fsuid becomes the
> > > > > > > current user. We know link() is actually modifying the directory, so
> > > > > > > the caller must have the MAY_WRITE permission on the directory. The
> > > > > > > current caller may should have this permission. I think this is
> > > > > > > acceptable to use the caller's fsuid(But I still feel a little
> > > > > > > conflicted with the overlay's design).
> > > > > > > The second is commit 51f7e52dc943 ("ovl: share inode for hard link")
> > > > > > > which removed the inode creation in ovl_link(). This commit move
> > > > > > > inode_init_owner() into ovl_create_object(), so the ovl_link() just
> > > > > > > give the old inode to ovl_create_or_link(). Then the override fsuid
> > > > > > > becomes the old inode's fsuid, neither the caller nor the overlay's
> > > > > > > creator! So I think this is incorrect.
> > > > > > > I think the link() should be like unlink(), overlay fs should just use
> > > > > > > the creator cred to do underlying fs's operations.
> > > > > > >
> > > > > >
> > > > > > I see. The reproducer and explanation belong in the commit message.
> > > > > >
> > > > > > Your argument makes sense to me, but CC Christian to make
> > > > > > sure I am not missing anything related to ACLs and what not.
> > > > >
> > > > > Once again with correct email address...
> > > >
> > > > So we have:
> > > >
> > > > ovl_create_object()
> > > > -> ovl_override_creds(ovl_sb)
> > > > -> ovl_new_inode()
> > > >    -> inode_init_owner()
> > > >       {
> > > >               inode->i_uid = current_fsuid();
> > > >               inode->i_gid = current_fsgid();
> > 
> > In inode_init_owner(), the inode->i_gid may inherit from parent dir.
> > And this is the main purpose of the commit bb0d2b8ad296 ("ovl: fix
> > sgid on directory").
> > 
> > > >       }
> > > > -> ovl_create_or_link(inode, ...)
> > > > -> prepare_creds() // Copy of caller's creds
> > >
> > > s/caller's/creator's/
> > >
> > > > {
> > > >         override_creds->fsuid = inode->i_uid;
> > > >         override_creds->fsgid = inode->i_gid;
> > > > }
> > > > -> revert_creds()
> > > >
> > > > which afaict means that the mounter's credentials are used apart from
> > > > the fs{g,u}id which is taken from inode->i_{g,u}id which should
> > > > correspond to current_fs{g,u}id().
> > > >
> > > > The commit that is pointed out in the patch
> > > > 51f7e52dc943 ("ovl: share inode for hard link")
> > > > seems to have broken that assumption.
> > > >
> > > > Given that the intention was to use the creator's creds _with the
> > > > caller's fs{g,u}id_ wouldn't it make more sense to simply ensure that
> > > > the caller's fs{g,u}id are always used instead of using the full
> > > > creator's creds just for the link operation? So something like this
> > > > (untested):
> > > >
> > > > diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
> > > > index 6b03457f72bb..4a3ee16a6d70 100644
> > > > --- a/fs/overlayfs/dir.c
> > > > +++ b/fs/overlayfs/dir.c
> > > > @@ -575,6 +575,9 @@ static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
> > > >         const struct cred *old_cred;
> > > >         struct cred *override_cred;
> > > >         struct dentry *parent = dentry->d_parent;
> > > > +       /* Retrieve caller's fs{g,u}id before we override creds below. */
> > > > +       kuid_t caller_fsuid = current_fsuid();
> > > > +       kgid_t caller_fsgid = current_fsgid();
> > > >
> > > >         err = ovl_copy_up(parent);
> > > >         if (err)
> > > > @@ -595,8 +598,8 @@ static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
> > > >         err = -ENOMEM;
> > > >         override_cred = prepare_creds();
> > > >         if (override_cred) {
> > > > -               override_cred->fsuid = inode->i_uid;
> > > > -               override_cred->fsgid = inode->i_gid;
> > > > +               override_cred->fsuid = caller_fsuid;
> > > > +               override_cred->fsgid = caller_fsgid;
> > 
> > So the override_cred->fsgid should be inode->i_gid if the inode is a new inode.
> > 
> > > >                 if (!attr->hardlink) {
> > > >                         err = security_dentry_create_files_as(dentry,
> > > >                                         attr->mode, &dentry->d_name, old_cred,
> > 
> > So your meaning should be like this:
> > 
> > diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
> > index 6b03457f72bb..9aead6ddc071 100644
> > --- a/fs/overlayfs/dir.c
> > +++ b/fs/overlayfs/dir.c
> > @@ -575,6 +575,8 @@ static int ovl_create_or_link(struct dentry
> > *dentry, struct inode *inode,
> >         const struct cred *old_cred;
> >         struct cred *override_cred;
> >         struct dentry *parent = dentry->d_parent;
> > +       kuid_t caller_fsuid = current_fsuid();
> > +       kgid_t caller_fsgid = current_fsgid();
> > 
> >         err = ovl_copy_up(parent);
> >         if (err)
> > @@ -595,9 +597,9 @@ static int ovl_create_or_link(struct dentry
> > *dentry, struct inode *inode,
> >         err = -ENOMEM;
> >         override_cred = prepare_creds();
> >         if (override_cred) {
> > -               override_cred->fsuid = inode->i_uid;
> > -               override_cred->fsgid = inode->i_gid;
> >                 if (!attr->hardlink) {
> > +                       override_cred->fsuid = inode->i_uid;
> > +                       override_cred->fsgid = inode->i_gid;
> >                         err = security_dentry_create_files_as(dentry,
> >                                         attr->mode, &dentry->d_name, old_cred,
> >                                         override_cred);
> > @@ -605,6 +607,9 @@ static int ovl_create_or_link(struct dentry
> > *dentry, struct inode *inode,
> >                                 put_cred(override_cred);
> >                                 goto out_revert_creds;
> >                         }
> > +               } else {
> > +                       override_cred->fsuid = caller_fsuid;
> > +                       override_cred->fsgid = caller_fsgid;
> >                 }
> >                 put_cred(override_creds(override_cred));
> >                 put_cred(override_cred);
> 
> Hah, wait. I had a pretty obvious braino when I did that. I forgot to
> account for setgid handling in inode_init_owner(). Let me take another
> close look...

Ok, so if we look at all the callchains:

(1) .create = ovl_create()
    -> ovl_create_object()
       -> ovl_create_or_link()
   
(2) .mkdir = ovl_mkdir()
    -> ovl_create_object()
       -> ovl_create_or_link()
   
(3) .mknod = ovl_mknod()
    -> ovl_create_object()
       -> ovl_create_or_link()
   
(4) .symlink = ovl_symlink()
    -> ovl_create_object()
       -> ovl_create_or_link()

(5) .link = ovl_link()
     -> ovl_create_or_link()

we see that (1) to (4) set the caller's fs{g,u}id to the
inode->i_{g,u}id. That was instantiated by inode_init_owner(). This will
be current_fs{g,u}id() of the caller unless the directory the file is
created in was setgid in which case inode->i_gid can be different from
current_fsgid().

So iiuc, taking the inode->i_{g,u}id is done to get fsgid set to
inode->i_gid when the parent directory is setgid.

But for (5) the inode->i_{g,u}id should be irrelevant because it's a
hardlink and so we should just use the caller's fs{g,u}id. So with all
that in mind we should be doing:

diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
index 6b03457f72bb..74b3813eda47 100644
--- a/fs/overlayfs/dir.c
+++ b/fs/overlayfs/dir.c
@@ -646,6 +646,8 @@ static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
        inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
        attr.mode = inode->i_mode;

+       attr.fsuid = inode->i_uid;
+       attr.fsgid = inode->i_gid;
        err = ovl_create_or_link(dentry, inode, &attr, false);
        /* Did we end up using the preallocated inode? */
        if (inode != d_inode(dentry))
@@ -702,6 +704,7 @@ static int ovl_link(struct dentry *old, struct inode *newdir,
 {
        int err;
        struct inode *inode;
+       struct ovl_cattr attr;

        err = ovl_want_write(old);
        if (err)
@@ -728,9 +731,12 @@ static int ovl_link(struct dentry *old, struct inode *newdir,
        inode = d_inode(old);
        ihold(inode);

-       err = ovl_create_or_link(new, inode,
-                       &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
-                       ovl_type_origin(old));
+       attr = (struct ovl_cattr){
+               .hardlink = ovl_dentry_upper(old),
+               .fsuid = current_fsuid(),
+               .fsgid = current_fsgid(),
+       };
+       err = ovl_create_or_link(new, inode, &attr, ovl_type_origin(old));
        if (err)
                iput(inode);

diff --git a/fs/overlayfs/overlayfs.h b/fs/overlayfs/overlayfs.h
index 87759165d32b..85043123a103 100644
--- a/fs/overlayfs/overlayfs.h
+++ b/fs/overlayfs/overlayfs.h
@@ -655,6 +655,8 @@ struct ovl_cattr {
        umode_t mode;
        const char *link;
        struct dentry *hardlink;
+       kuid_t fsuid;
+       kgid_t fsgid;
 };

 #define OVL_CATTR(m) (&(struct ovl_cattr) { .mode = (m) })

Using the full creator's credentials for just hardlinks seems odd to me
when all creation requests use the caller's fs{g,u}id...
