On Tue, Sep 29, 2020 at 10:21:19AM +0200, Dmitry Vyukov wrote: > On Tue, Sep 29, 2020 at 10:06 AM David Laight <David.Laight@xxxxxxxxxx> wrote: > > > > From: Christoph Hellwig > > > Sent: 29 September 2020 07:56 > > > > > > On Mon, Sep 28, 2020 at 11:46:48PM -0700, Eric Biggers wrote: > > > > > Linus asked for it. What is the call chain that we hit it with? > > > > > > > > Call Trace: > > > > kernel_read+0x52/0x70 fs/read_write.c:471 > > > > kernel_read_file fs/exec.c:989 [inline] > > > > kernel_read_file+0x2e5/0x620 fs/exec.c:952 > > > > kernel_read_file_from_fd+0x56/0xa0 fs/exec.c:1076 > > > > __do_sys_finit_module+0xe6/0x190 kernel/module.c:4066 > > > > do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46 > > > > entry_SYSCALL_64_after_hwframe+0x44/0xa9 > > > > > > > > See the email from syzbot for the full details: > > > > https://lkml.kernel.org/linux-fsdevel/000000000000da992305b02e9a51@xxxxxxxxxx > > > > > > Passing a fs without read permissions definitively looks bogus for > > > the finit_module syscall. So I think all we need is an extra check > > > to validate the fd. > > > > The sysbot test looked like it didn't even have a regular file. > > I thought I saw a test for that - but it might be in a different path. > > > > You do need to ensure that 'exec' doesn't need read access. > > The test tried to load a module from /dev/input/mouse > > r2 = syz_open_dev$mouse(&(0x7f0000000000)='/dev/input/mouse#\x00', > 0x101, 0x109887) > finit_module(r2, 0x0, 0x0) > > because... why not? Everything is a file! :) Yes, syzbot is fine here. It is the modules code that needs to better verify the fd.
