[Devel] [PATCH] binfmt_misc: fix mount after umount in CT

[Alexander Mikhalitsyn]

On Tue, 19 Oct 2021 13:59:34 +0300
Alexander Mikhalitsyn <alexander.mikhalitsyn at virtuozzo.com> wrote:

> On Mon, 18 Oct 2021 19:51:49 +0300
> Nikita Yushchenko <nikita.yushchenko at virtuozzo.com> wrote:
> 
> > The assumption that bm_fill_super() is not called for the second time
> > for CT is wrong: umount operation clears sb->s_root, which causes
> > vfs_get_super() to call fill_super again on the next mount.
> > 
> > Make bm_fill_super() handle multiple-calls corrently:
> > - initialize bm_data and set ve->binfmt_misc only if it is not done
> >   before,
> > - delay desctruction of it up to CT destruction.
> > 
> > https://jira.sw.ru/browse/PSBM-133968
> > Fixes: 8250ff41d190 ("ve/fs/binfmt: clean bm_data reference from ve on err path")
> > Signed-off-by: Nikita Yushchenko <nikita.yushchenko at virtuozzo.com>
> > ---
> >  fs/binfmt_misc.c | 37 +++++++++++++------------------------
> >  1 file changed, 13 insertions(+), 24 deletions(-)
> > 
> > diff --git a/fs/binfmt_misc.c b/fs/binfmt_misc.c
> > index 0946e7e6caa5..5f4e90c1ade2 100644
> > --- a/fs/binfmt_misc.c
> > +++ b/fs/binfmt_misc.c
> > @@ -853,38 +853,25 @@ static int bm_fill_super(struct super_block *sb, struct fs_context *fc)
> >  	};
> >  
> >  	struct ve_struct *ve = get_exec_env();
> > -	struct binfmt_misc *bm_data;
> > +	struct binfmt_misc *bm_data = ve->binfmt_misc;
> >  
> > -	/*
> > -	 * bm_get_tree()
> > -	 *  get_tree_keyed(fc, bm_fill_super, get_ve(ve))
> > -	 *   fc->s_fs_info = current VE
> > -	 *   vfs_get_super(fc, vfs_get_keyed_super, bm_fill_super)
> > -	 *    sb = sget_fc(fc, test, set_anon_super_fc)
> > -	 *    if (!sb->s_root) {
> > -	 *		err = bm_fill_super(sb, fc);
> > -	 *
> > -	 * => we should never get here with initialized ve->binfmt_misc.
> > -	 */
> > -	if (WARN_ON_ONCE(ve->binfmt_misc))
> > -		return -EEXIST;
> > +	if (!bm_data) {
> > +		bm_data = kzalloc(sizeof(struct binfmt_misc), GFP_KERNEL);
> > +		if (!bm_data)
> > +			return -ENOMEM;
> >  
> > -	bm_data = kzalloc(sizeof(struct binfmt_misc), GFP_KERNEL);
> > -	if (!bm_data)
> > -		return -ENOMEM;
> > +		INIT_LIST_HEAD(&bm_data->entries);
> > +		rwlock_init(&bm_data->entries_lock);
> >  
> > -	INIT_LIST_HEAD(&bm_data->entries);
> > -	rwlock_init(&bm_data->entries_lock);
> > +		ve->binfmt_misc = bm_data;
> 
> Isn't it better to move ve->binfmt_misc assignment to the
> end of function where we know that all operations was successful?
> 
> > +		/* this will be cleared by ve_binfmt_fini() */
> > +	}
> >  
> >  	err = simple_fill_super(sb, BINFMTFS_MAGIC, bm_files);
> > -	if (err) {
> > -		kfree(bm_data);
> 
> If we have ve->binfmt_misc assignment in the upper part of code, then
> we need to do ve->binfmt_misc = NULL here.

please ignore. misread ;)

> 
> > +	if (err)
> >  		return err;
> > -	}
> >  
> >  	sb->s_op = &s_ops;
> > -
> > -	ve->binfmt_misc = bm_data;
> see above
> 
> >  	bm_data->enabled = 1;
> >  
> >  	return 0;
> > @@ -971,6 +958,8 @@ static void ve_binfmt_fini(void *data)
> >  	while (!list_empty(&bm_data->entries))
> >  		kill_node(bm_data, list_first_entry(
> >  			&bm_data->entries, Node, list));
> > +
> > +	kfree(bm_data);
> 
> We have kfree in ve_destroy (kernel/ve/ve.c) already.

this is actual

> 
> >  }
> >  
> >  static struct ve_hook ve_binfmt_hook = {
> > -- 
> > 2.30.2
> > 
> 
> 
> _______________________________________________
> Devel mailing list
> Devel at openvz.org
> https://lists.openvz.org/mailman/listinfo/devel