Re: [PATCH] vfs: set fs_context::user_ns for reconfigure

[Eric Biggers <ebiggers@kernel.org>]

On Fri, Aug 30, 2019 at 10:12:28PM -0500, Eric Biggers wrote:
> On Wed, Aug 21, 2019 at 10:16:33PM -0700, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > fs_context::user_ns is used by fuse_parse_param(), even during remount,
> > so it needs to be set to the existing value for reconfigure.
> > 
> > Reproducer:
> > 
> > 	#include <fcntl.h>
> > 	#include <sys/mount.h>
> > 
> > 	int main()
> > 	{
> > 		char opts[128];
> > 		int fd = open("/dev/fuse", O_RDWR);
> > 
> > 		sprintf(opts, "fd=%d,rootmode=040000,user_id=0,group_id=0", fd);
> > 		mkdir("mnt", 0777);
> > 		mount("foo",  "mnt", "fuse.foo", 0, opts);
> > 		mount("foo", "mnt", "fuse.foo", MS_REMOUNT, opts);
> > 	}
> > 
> > Crash:
> > 	BUG: kernel NULL pointer dereference, address: 0000000000000000
> > 	#PF: supervisor read access in kernel mode
> > 	#PF: error_code(0x0000) - not-present page
> > 	PGD 0 P4D 0
> > 	Oops: 0000 [#1] SMP
> > 	CPU: 0 PID: 129 Comm: syz_make_kuid Not tainted 5.3.0-rc5-next-20190821 #3
> > 	Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-20181126_142135-anatol 04/01/2014
> > 	RIP: 0010:map_id_range_down+0xb/0xc0 kernel/user_namespace.c:291
> > 	[...]
> > 	Call Trace:
> > 	 map_id_down kernel/user_namespace.c:312 [inline]
> > 	 make_kuid+0xe/0x10 kernel/user_namespace.c:389
> > 	 fuse_parse_param+0x116/0x210 fs/fuse/inode.c:523
> > 	 vfs_parse_fs_param+0xdb/0x1b0 fs/fs_context.c:145
> > 	 vfs_parse_fs_string+0x6a/0xa0 fs/fs_context.c:188
> > 	 generic_parse_monolithic+0x85/0xc0 fs/fs_context.c:228
> > 	 parse_monolithic_mount_data+0x1b/0x20 fs/fs_context.c:708
> > 	 do_remount fs/namespace.c:2525 [inline]
> > 	 do_mount+0x39a/0xa60 fs/namespace.c:3107
> > 	 ksys_mount+0x7d/0xd0 fs/namespace.c:3325
> > 	 __do_sys_mount fs/namespace.c:3339 [inline]
> > 	 __se_sys_mount fs/namespace.c:3336 [inline]
> > 	 __x64_sys_mount+0x20/0x30 fs/namespace.c:3336
> > 	 do_syscall_64+0x4a/0x1a0 arch/x86/entry/common.c:290
> > 	 entry_SYSCALL_64_after_hwframe+0x49/0xbe
> > 
> > Reported-by: syzbot+7d6a57304857423318a5@syzkaller.appspotmail.com
> > Fixes: 408cbe695350 ("vfs: Convert fuse to use the new mount API")
> > Cc: David Howells <dhowells@redhat.com>
> > Cc: Miklos Szeredi <miklos@szeredi.hu>
> > Signed-off-by: Eric Biggers <ebiggers@google.com>
> > ---
> >  fs/fs_context.c | 4 +---
> >  1 file changed, 1 insertion(+), 3 deletions(-)
> > 
> > diff --git a/fs/fs_context.c b/fs/fs_context.c
> > index cc61d305dc4b..44c4174b250a 100644
> > --- a/fs/fs_context.c
> > +++ b/fs/fs_context.c
> > @@ -279,10 +279,8 @@ static struct fs_context *alloc_fs_context(struct file_system_type *fs_type,
> >  		fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
> >  		break;
> >  	case FS_CONTEXT_FOR_RECONFIGURE:
> > -		/* We don't pin any namespaces as the superblock's
> > -		 * subscriptions cannot be changed at this point.
> > -		 */
> >  		atomic_inc(&reference->d_sb->s_active);
> > +		fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
> >  		fc->root = dget(reference);
> >  		break;
> >  	}
> > -- 
> > 2.22.1
> 
> Ping.
> 

Ping.  This is still broken in linux-next.

- Eric