On 10/10, Oleg Nesterov wrote:
>
> On 10/10, Eryu Guan wrote:
> >
> > inode->i_private is assigned by a Node pointer only after
> > registering a new binary format, so it could be NULL if we only
> > mount binfmt_misc but don't register any format,
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> Not really, I 'think... I mean, the problem is that e == NULL when inode
> is root_inode, or "status", or "register", created by bm_fill_super()...
>
> Yes, you need to unregister all formats to hit this problem.
>
>
> Oh, and I'm afraid I intoduced another problem, bm_register_write() error
> paths can call iput() with MISC_FMT_OPEN_FILE set but e->interpreter == NULL.
> I'll re-check and send another patch today.
Ah, no, I was wrong again... If bm_register_write() fails, bm_evict_inode()
will hit the same ->i_private == NULL problem fixed by your patch.
So the changelog can be updated to explain that i_private == NULL is possible
if inode was created by bm_fill_super(), or iput() was called by the error path
in bm_register_write().
But the patch is obviously fine, thanks again.
> Thanks a lot!
>
>
> Acked-by: Oleg Nesterov <oleg@xxxxxxxxxx>
>
>
> > and this results in
> > NULL pointer dereference at umount time. e.g.
> >
> > mount -t binfmt_misc binfmt_misc /proc/sys/fs/binfmt_misc
> > umount /proc/sys/fs/binfmt_misc
> >
> > [ 9379.678259] BUG: unable to handle kernel NULL pointer dereference at 0000000000000013
> > [ 9379.985952] IP: bm_evict_inode+0x16/0x40 [binfmt_misc]
> > ...
> > [ 9380.964911] Call Trace:
> > [ 9380.977633] evict+0xd3/0x1a0
> > [ 9380.994449] iput+0x17d/0x1d0
> > [ 9381.010306] dentry_unlink_inode+0xb9/0xf0
> > [ 9381.034046] __dentry_kill+0xc7/0x170
> > [ 9381.055145] shrink_dentry_list+0x122/0x280
> > [ 9381.078908] shrink_dcache_parent+0x39/0x90
> > [ 9381.103082] do_one_tree+0x12/0x40
> > [ 9381.122005] shrink_dcache_for_umount+0x2d/0x90
> > [ 9381.146517] generic_shutdown_super+0x1f/0x120
> > [ 9381.171644] kill_litter_super+0x29/0x40
> > [ 9381.193513] deactivate_locked_super+0x43/0x70
> > [ 9381.219177] deactivate_super+0x45/0x60
> > [ 9381.240130] cleanup_mnt+0x3f/0x70
> > [ 9381.259064] __cleanup_mnt+0x12/0x20
> > [ 9381.279802] task_work_run+0x86/0xa0
> > [ 9381.299612] exit_to_usermode_loop+0x6d/0x99
> > [ 9381.323872] syscall_return_slowpath+0xba/0xf0
> > [ 9381.350464] entry_SYSCALL_64_fastpath+0xa3/0xa
> >
> > Fix it by making sure Node (e) is not NULL.
> >
> > Cc: Oleg Nesterov <oleg@xxxxxxxxxx>
> > Fixes: 83f918274e4b ("exec: binfmt_misc: shift filp_close(interp_file) from kill_node() to bm_evict_inode()")
> > Signed-off-by: Eryu Guan <eguan@xxxxxxxxxx>
> > ---
> > fs/binfmt_misc.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/fs/binfmt_misc.c b/fs/binfmt_misc.c
> > index 2a46762def31..a7c5a9861bef 100644
> > --- a/fs/binfmt_misc.c
> > +++ b/fs/binfmt_misc.c
> > @@ -596,7 +596,7 @@ static void bm_evict_inode(struct inode *inode)
> > {
> > Node *e = inode->i_private;
> >
> > - if (e->flags & MISC_FMT_OPEN_FILE)
> > + if (e && e->flags & MISC_FMT_OPEN_FILE)
> > filp_close(e->interp_file, NULL);
> >
> > clear_inode(inode);
> > --
> > 2.13.6
> >
