Re: [Bug] WARNING: zero-size vmalloc in sel_write_load

[Paul Moore]

On Mon, Mar 4, 2024 at 3:11 PM Christian Göttsche
<cgzones@xxxxxxxxxxxxxx> wrote:
> On Mon, 4 Mar 2024 at 20:19, Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> > On Mon, Mar 4, 2024 at 9:08 AM Sam Sun <samsun1006219@xxxxxxxxx> wrote:
> > >
> > > Dear developers and maintainers,
> > >
> > > We encountered a warning in function sel_write_load(). It is tested on
> > > kernel 6.8.0-rc7. Bug report is listed below.
> > >
> > > ```
> > > WARNING: CPU: 1 PID: 8109 at mm/vmalloc.c:3247
> > > __vmalloc_node_range+0x10a0/0x1490 mm/vmalloc.c:3247
> > > Modules linked in:
> > > CPU: 1 PID: 8109 Comm: syz-executor370 Not tainted 6.7.0-rc7 #1
> > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
> > > RIP: 0010:__vmalloc_node_range+0x10a0/0x1490 mm/vmalloc.c:3247
> > > Code: 65 48 2b 04 25 28 00 00 00 0f 85 98 02 00 00 48 81 c4 70 01 00
> > > 00 4c 89 e0 5b 5d 41 5c 41 5d 41 5e 41 5f c3 e8 71 43 b7 ff 90 <0f> 0b
> > > 90 45 31 e4 eb a1 e8 63 43 b7 ff 48 8b 4c 24 40 31 f6 45 31
> > > RSP: 0018:ffffc90002adf9c0 EFLAGS: 00010293
> > > RAX: 0000000000000000 RBX: dffffc0000000000 RCX: ffffffff81cdc194
> > > RDX: ffff888022124ec0 RSI: ffffffff81cdd16f RDI: 0000000000000007
> > > RBP: 0000000000000000 R08: 0000000000000007 R09: 0000000000000000
> > > R10: 0000000000000000 R11: 0000000000000001 R12: ffff888107373a48
> > > R13: dffffc0000000000 R14: 0000000000000000 R15: ffffc90002adfec8
> > > FS:  00005555560953c0(0000) GS:ffff888135c00000(0000) knlGS:0000000000000000
> > > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > CR2: 0000000020000010 CR3: 000000010503d000 CR4: 0000000000750ef0
> > > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > > PKRU: 55555554
> > > Call Trace:
> > >  <TASK>
> > >  __vmalloc_node mm/vmalloc.c:3385 [inline]
> > >  vmalloc+0x6b/0x80 mm/vmalloc.c:3418
> > >  sel_write_load+0x27f/0x19d0 security/selinux/selinuxfs.c:603
> > >  vfs_write+0x2a9/0xd80 fs/read_write.c:582
> > >  ksys_pwrite64 fs/read_write.c:699 [inline]
> > >  __do_sys_pwrite64 fs/read_write.c:709 [inline]
> > >  __se_sys_pwrite64 fs/read_write.c:706 [inline]
> > >  __x64_sys_pwrite64+0x1f3/0x250 fs/read_write.c:706
> > >  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
> > >  do_syscall_64+0x40/0x110 arch/x86/entry/common.c:83
> > >  entry_SYSCALL_64_after_hwframe+0x63/0x6b
> > > RIP: 0033:0x7f40e7728f8d
> > > Code: 28 c3 e8 46 1e 00 00 66 0f 1f 44 00 00 f3 0f 1e fa 48 89 f8 48
> > > 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d
> > > 01 f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48
> > > RSP: 002b:00007fff5bf39508 EFLAGS: 00000246 ORIG_RAX: 0000000000000012
> > > RAX: ffffffffffffffda RBX: 00007fff5bf39708 RCX: 00007f40e7728f8d
> > > RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000003
> > > RBP: 0000000000000001 R08: 0000000000000000 R09: 00007fff5bf39708
> > > R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000001
> > > R13: 00007fff5bf396f8 R14: 00007f40e77a6530 R15: 0000000000000001
> > >  </TASK>
> > > ```
> > >
> > > The cause of this bug is that in sel_write_load(), parameter "count"
> > > is controlled by user, which could be zero. It is passed to vmalloc()
> > > as an argument.
> > >
> > > If you have any questions, please contact us.
> > > Reported by: Yue Sun <samsun1006219@xxxxxxxxx>
> > > Reported by: xingwei lee <xrivendell7@xxxxxxxxx>
> >
> > Everything appears to be working as expected, vmalloc() caught the
> > zero-length allocation request, emitted the warning, returned NULL to
> > sel_write_load(), and sel_write_load() handled the error condition
> > triggered by vmalloc(0) returning NULL.  Did you see any unexpected
> > behavior beyond this warning message above?
>
> Probably because kernel warnings should not be reachable from
> userspace ...

My question was asking if the reporter was seeing any unexpected
behavior *beyond* the warning message.  I wanted to make sure they
weren't seeing anything else on their system that we should also take
into account.

> ... although in this case loading a policy is a highly
> privileged operation - , because they mostly signal incorrect internal
> state and can lead with the enabled option of panic_on_warn to system
> halts.
>
> I have two suggestions:
>
> I. Can the documentation of vmalloc() mention that passing a size of 0
> is discouraged?

One could always submit a patch and see what happens, that's almost
always the best way to get feedback.

> II. Can the global SELinux state mutex in sel_write_load() be acquired
> after the avc permission check, so that rouge processes with write
> access to /load but not granted security { load_policy } can not pound
> on that mutex?

We need to make sure that there is not a window between the
avc_has_perm() permission check and the actual policy load in
security_load_policy() where another task could race and cause some
unexpected behavior.  For that reason I think we need to take the
mutex before the avc_has_perm() call, and we likely want to keep the
vmalloc()/copy_from_user() after the permission check for the same
reason you wanted to delay taking the mutex.

We probably could consider moving the @ppos and @count sanity checks
before the mutex.

-- 
paul-moore.com