In older kernels, attempting to fetch that system.posix_acl_access on
NFSv4 would return -EOPNOTSUPP, but in more recent kernels that returns
-ENODATA.

Most filesystems that don't support POSIX ACLs leave the SB_POSIXACL
flag clear, which cues the VFS to return -EOPNOTSUPP in this situation.
We can't do that with NFSv4 since that flag also cues the VFS to avoid
applying the umask early.

Fix this by adding a stub get_acl handler for NFSv4 that always returns
-EOPNOTSUPP.

Reported-by: Ondrej Valousek <ondrej.valousek.xm@renesas.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
I suspect that this problem popped in due to some VFS layer changes. I
haven't identified the patch that broke it, but I think this is probably
the least invasive way to fix it.

Another alternative would be to return -EOPNOTSUPP on filesystems that
set SB_POSIXACL but that don't set get_acl or get_inode_acl.

Thoughts?
---
 fs/nfs/nfs4proc.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 794343790ea8..93f3caf4ec08 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -10621,6 +10621,12 @@ static void nfs4_disable_swap(struct inode *inode)
 	nfs4_schedule_state_manager(clp);
 }
 
+static struct posix_acl *nfs4_get_posix_acl(struct mnt_idmap *idmap, struct dentry *dentry,
+					    int type)
+{
+	return ERR_PTR(-EOPNOTSUPP);
+}
+
 static const struct inode_operations nfs4_dir_inode_operations = {
 	.create		= nfs_create,
 	.lookup		= nfs_lookup,
@@ -10636,6 +10642,7 @@ static const struct inode_operations nfs4_dir_inode_operations = {
 	.getattr	= nfs_getattr,
 	.setattr	= nfs_setattr,
 	.listxattr	= nfs4_listxattr,
+	.get_acl	= nfs4_get_posix_acl,
 };
 
 static const struct inode_operations nfs4_file_inode_operations = {
@@ -10643,6 +10650,7 @@ static const struct inode_operations nfs4_file_inode_operations = {
 	.getattr	= nfs_getattr,
 	.setattr	= nfs_setattr,
 	.listxattr	= nfs4_listxattr,
+	.get_acl	= nfs4_get_posix_acl,
 };
 
 const struct nfs_rpc_ops nfs_v4_clientops = {

---
base-commit: c0894ef10b846fd4a74a670cbec2483246030e06
change-id: 20230907-kdevops-d311e57bf5d2

Best regards,
-- 
Jeff Layton <jlayton@kernel.org>

On Thu, Sep 07, 2023 at 01:32:36PM -0400, Jeff Layton wrote:
> In older kernels, attempting to fetch that system.posix_acl_access on
> NFSv4 would return -EOPNOTSUPP, but in more recent kernels that returns
> -ENODATA.
> 
> Most filesystems that don't support POSIX ACLs leave the SB_POSIXACL
> flag clear, which cues the VFS to return -EOPNOTSUPP in this situation.
> We can't do that with NFSv4 since that flag also cues the VFS to avoid
> applying the umask early.
> 
> Fix this by adding a stub get_acl handler for NFSv4 that always returns
> -EOPNOTSUPP.
> 
> Reported-by: Ondrej Valousek <ondrej.valousek.xm@renesas.com>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
> I suspect that this problem popped in due to some VFS layer changes. I
> haven't identified the patch that broke it, but I think this is probably
> the least invasive way to fix it.
> 
> Another alternative would be to return -EOPNOTSUPP on filesystems that
> set SB_POSIXACL but that don't set get_acl or get_inode_acl.
> 
> Thoughts?

Yes: I hate POSIX ACLs. ;)

Before the VFS rework to only rely on i_op->*acl* methods POSIX ACLs
were set using sb->s_xattr handlers. So when a filesystem raised
SB_POSIXACL but didn't set sb->s_xattr handlers for POSIX ACLs we would:

__vfs_getxattr()
-> xattr_resolve_name()
  // no match so return EOPNOTSUPP

No we have

vfs_get_acl()
-> __get_acl()
   -> i_op->get_acl
   // no get_acl inode method return ENODATA

So as a bugfix to backport I think you should do exactly what you do
here because I'm not sure if some fs relies on ENODATA to be returned if
no get_acl inode method is set. There's a lot of quirkiness everywhere.
But we should look through all callers and if nothing relies on EINVAL
just start returning EOPNOTSUPP if no get_acl i_op is set.

Looks good to me,
Acked-by: Christian Brauner <brauner@kernel.org>

On Fri, 2023-09-08 at 14:55 +0200, Christian Brauner wrote:
> On Thu, Sep 07, 2023 at 01:32:36PM -0400, Jeff Layton wrote:
> > In older kernels, attempting to fetch that system.posix_acl_access on
> > NFSv4 would return -EOPNOTSUPP, but in more recent kernels that returns
> > -ENODATA.
> > 
> > Most filesystems that don't support POSIX ACLs leave the SB_POSIXACL
> > flag clear, which cues the VFS to return -EOPNOTSUPP in this situation.
> > We can't do that with NFSv4 since that flag also cues the VFS to avoid
> > applying the umask early.
> > 
> > Fix this by adding a stub get_acl handler for NFSv4 that always returns
> > -EOPNOTSUPP.
> > 
> > Reported-by: Ondrej Valousek <ondrej.valousek.xm@renesas.com>
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > ---
> > I suspect that this problem popped in due to some VFS layer changes. I
> > haven't identified the patch that broke it, but I think this is probably
> > the least invasive way to fix it.
> > 
> > Another alternative would be to return -EOPNOTSUPP on filesystems that
> > set SB_POSIXACL but that don't set get_acl or get_inode_acl.
> > 
> > Thoughts?
> 
> Yes: I hate POSIX ACLs. ;)

The API is quite weird, yes.

> Before the VFS rework to only rely on i_op->*acl* methods POSIX ACLs
> were set using sb->s_xattr handlers. So when a filesystem raised
> SB_POSIXACL but didn't set sb->s_xattr handlers for POSIX ACLs we would:
> 
> __vfs_getxattr()
> -> xattr_resolve_name()
>   // no match so return EOPNOTSUPP
> 
> No we have
> 
> vfs_get_acl()
> -> __get_acl()
>    -> i_op->get_acl
>    // no get_acl inode method return ENODATA
> 
> So as a bugfix to backport I think you should do exactly what you do
> here because I'm not sure if some fs relies on ENODATA to be returned if
> no get_acl inode method is set. There's a lot of quirkiness everywhere.
> But we should look through all callers and if nothing relies on EINVAL
> just start returning EOPNOTSUPP if no get_acl i_op is set.
> 
> Looks good to me,
> Acked-by: Christian Brauner <brauner@kernel.org>

Thanks. I did some rudimentary git grepping, and I don't see any that
set SB_POSIXACL but don't set either get_acl or get_inode_acl. I'll spin
up a patch and we can get it into -next for a while and see if anything
shakes out.
-- 
Jeff Layton <jlayton@kernel.org>