Allow __free(iput) markings for easier cleanup on inode allocations.

Cc: Christian Brauner <brauner@kernel.org>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 include/linux/fs.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 98b7a7a8c42e..7c3eed3dd1bf 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2459,6 +2459,8 @@ extern int current_umask(void);
 
 extern void ihold(struct inode * inode);
 extern void iput(struct inode *);
+DEFINE_FREE(iput, struct inode *, if (_T) iput(_T))
+
 int inode_update_timestamps(struct inode *inode, int flags);
 int generic_update_time(struct inode *, int);
 
-- 
2.34.1

On Sat, Dec 02, 2023 at 01:22:13PM -0800, Kees Cook wrote:
> Allow __free(iput) markings for easier cleanup on inode allocations.

NAK.  That's a bloody awful idea for that particular data type, since
	1) ERR_PTR(...) is not uncommon and passing it to iput() is a bug.
	2) the common pattern is to have reference-consuming primitives,
with failure exits normally *not* having to do iput() at all.

Please, don't.

On Sat, Dec 02, 2023 at 09:28:46PM +0000, Al Viro wrote:
> On Sat, Dec 02, 2023 at 01:22:13PM -0800, Kees Cook wrote:
> > Allow __free(iput) markings for easier cleanup on inode allocations.
> 
> NAK.  That's a bloody awful idea for that particular data type, since
> 	1) ERR_PTR(...) is not uncommon and passing it to iput() is a bug.

Ah, sounds like instead of "if (_T)", you'd rather see
"if (!IS_ERR_OR_NULL(_T))" ?

> 	2) the common pattern is to have reference-consuming primitives,
> with failure exits normally *not* having to do iput() at all.

This I'm not following. If I make a call to "new_inode(sb)" that I end
up not using, I need to call "iput()" in it...

How should this patch be written to avoid the iput() on failure?
https://lore.kernel.org/all/20231202212217.243710-4-keescook@chromium.org/

-Kees

-- 
Kees Cook

On Sat, Dec 02, 2023 at 01:34:32PM -0800, Kees Cook wrote:
> On Sat, Dec 02, 2023 at 09:28:46PM +0000, Al Viro wrote:
> > On Sat, Dec 02, 2023 at 01:22:13PM -0800, Kees Cook wrote:
> > > Allow __free(iput) markings for easier cleanup on inode allocations.
> > 
> > NAK.  That's a bloody awful idea for that particular data type, since
> > 	1) ERR_PTR(...) is not uncommon and passing it to iput() is a bug.
> 
> Ah, sounds like instead of "if (_T)", you'd rather see
> "if (!IS_ERR_OR_NULL(_T))" ?
> 
> > 	2) the common pattern is to have reference-consuming primitives,
> > with failure exits normally *not* having to do iput() at all.
> 
> This I'm not following. If I make a call to "new_inode(sb)" that I end
> up not using, I need to call "iput()" in it...

If we wanted to do this properly then we would need to emulate consume
or move semantics like Rust has. So a cleanup function for inodes based
on scope for example and then another primitive that transfers/moves
ownership of that refcount to the consumer. Usually this is emulate by
stuff like TAKE_POINTER() and similar stuff in userspace. But I'm not
sure how pleasant it would be to do this cleanly.

On Sat, Dec 02, 2023 at 01:34:32PM -0800, Kees Cook wrote:
> On Sat, Dec 02, 2023 at 09:28:46PM +0000, Al Viro wrote:
> > On Sat, Dec 02, 2023 at 01:22:13PM -0800, Kees Cook wrote:
> > > Allow __free(iput) markings for easier cleanup on inode allocations.
> > 
> > NAK.  That's a bloody awful idea for that particular data type, since
> > 	1) ERR_PTR(...) is not uncommon and passing it to iput() is a bug.
> 
> Ah, sounds like instead of "if (_T)", you'd rather see
> "if (!IS_ERR_OR_NULL(_T))" ?

No.  I would rather *not* see IS_ERR_OR_NULL anywhere, but that's
a separate rant.

> > 	2) the common pattern is to have reference-consuming primitives,
> > with failure exits normally *not* having to do iput() at all.
> 
> This I'm not following. If I make a call to "new_inode(sb)" that I end
> up not using, I need to call "iput()" in it...
> 
> How should this patch be written to avoid the iput() on failure?
> https://lore.kernel.org/all/20231202212217.243710-4-keescook@chromium.org/

I'll poke around and see what I can suggest; said that, one thing I have
spotted there on the quick look is that you are exposing hashed dentry associated
with your inode before you set its ->i_private.  Have an open() hit just after
that d_add() and this
static int pstore_file_open(struct inode *inode, struct file *file)
{
        struct pstore_private *ps = inode->i_private;
        struct seq_file *sf;
        int err;
        const struct seq_operations *sops = NULL;

        if (ps->record->type == PSTORE_TYPE_FTRACE)
... with happily oops on you.

On Sat, Dec 02, 2023 at 09:42:12PM +0000, Al Viro wrote:

> I'll poke around and see what I can suggest; said that, one thing I have
> spotted there on the quick look is that you are exposing hashed dentry associated
> with your inode before you set its ->i_private.

... and on the second look, no, you do not do anything of that sort.
My apologies...