Re: [syzbot] KCSAN: data-race in assoc_array_apply_edit / search_nested_keyrings

[Eric Biggers]

On Tue, May 04, 2021 at 01:38:20PM +0200, 'Marco Elver' via syzkaller-bugs wrote:
> Hello,
> 
> On Tue, 4 May 2021 at 13:33, syzbot
> <syzbot+e4eb6db47eb0f80308c6@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> > HEAD commit:    5e321ded Merge tag 'for-5.13/parisc' of git://git.kernel.o..
> > git tree:       upstream
> > console output: https://syzkaller.appspot.com/x/log.txt?x=111cafb9d00000
> > kernel config:  https://syzkaller.appspot.com/x/.config?x=a4da2ebcb6e8f526
> > dashboard link: https://syzkaller.appspot.com/bug?extid=e4eb6db47eb0f80308c6
> > compiler:       Debian clang version 11.0.1-2
> >
> > Unfortunately, I don't have any reproducer for this issue yet.
> >
> > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > Reported-by: syzbot+e4eb6db47eb0f80308c6@xxxxxxxxxxxxxxxxxxxxxxxxx
> >
> > ==================================================================
> > BUG: KCSAN: data-race in assoc_array_apply_edit / search_nested_keyrings
> >
> > write to 0xffff8881065ffc10 of 8 bytes by task 30966 on cpu 1:
> >  assoc_array_apply_edit+0x3e/0x660 lib/assoc_array.c:1357
> >  __key_link+0x8a/0xc0 security/keys/keyring.c:1372
> >  __key_instantiate_and_link+0x15b/0x290 security/keys/key.c:459
> >  key_create_or_update+0x750/0x990 security/keys/key.c:941
> >  __do_sys_add_key security/keys/keyctl.c:134 [inline]
> >  __se_sys_add_key+0x26f/0x300 security/keys/keyctl.c:74
> >  __x64_sys_add_key+0x63/0x70 security/keys/keyctl.c:74
> >  do_syscall_64+0x4a/0x90 arch/x86/entry/common.c:47
> >  entry_SYSCALL_64_after_hwframe+0x44/0xae
> >
> > read to 0xffff8881065ffc10 of 8 bytes by task 30971 on cpu 0:
> >  search_nested_keyrings+0x34f/0x920 security/keys/keyring.c:751
> >  keyring_search_rcu+0xf4/0x180 security/keys/keyring.c:922
> >  search_cred_keyrings_rcu+0x135/0x240 security/keys/process_keys.c:480
> >  search_process_keyrings_rcu security/keys/process_keys.c:544 [inline]
> >  lookup_user_key+0xab6/0xd40 security/keys/process_keys.c:762
> >  __do_sys_add_key security/keys/keyctl.c:126 [inline]
> >  __se_sys_add_key+0x23a/0x300 security/keys/keyctl.c:74
> >  __x64_sys_add_key+0x63/0x70 security/keys/keyctl.c:74
> >  do_syscall_64+0x4a/0x90 arch/x86/entry/common.c:47
> >  entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> I found this interesting because the code around
> security/keys/keyring.c:751 is diligent in marking concurrency
> accesses with READ_ONCE(). So on the off-chance there is unexpected
> concurrency here, I thought it'd be worth double-checking as I wasn't
> able to conclude if this is just missing a READ_ONCE().
> 
> Thank you!
> 

It looks more like the problem is missing WRITE_ONCE(), not READ_ONCE().

smp_store_release() would also handle this properly, without the need for the
explicit smp_wmb().

- Eric