]> git.baikalelectronics.ru Git - kernel.git/commit
selinux: fix sidtab string cache locking
authorOndrej Mosnacek <omosnace@redhat.com>
Mon, 3 Feb 2020 08:50:23 +0000 (09:50 +0100)
committerPaul Moore <paul@paul-moore.com>
Wed, 5 Feb 2020 23:31:10 +0000 (18:31 -0500)
commitff581ae5218fdb6baeb342404727752d4289c957
tree5fbf40170029dc6e2883d3d73bd98c4805e92a16
parent9a728a73479cf3ef045b9f982d1290c89bea9ac1
selinux: fix sidtab string cache locking

Avoiding taking a lock in an IRQ context is not enough to prevent
deadlocks, as discovered by syzbot:

===
WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
5.5.0-syzkaller #0 Not tainted
-----------------------------------------------------
syz-executor.0/8927 [HC0[0]:SC0[2]:HE1:SE0] is trying to acquire:
ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:338 [inline]
ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: sidtab_sid2str_put.part.0+0x36/0x880 security/selinux/ss/sidtab.c:533

and this task is already holding:
ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: spin_lock include/linux/spinlock.h:338 [inline]
ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: nf_conntrack_lock+0x17/0x70 net/netfilter/nf_conntrack_core.c:91
which would create a new lock dependency:
 (&(&nf_conntrack_locks[i])->rlock){+.-.} -> (&(&s->cache_lock)->rlock){+.+.}

but this new dependency connects a SOFTIRQ-irq-safe lock:
 (&(&nf_conntrack_locks[i])->rlock){+.-.}

[...]

other info that might help us debug this:

 Possible interrupt unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&(&s->cache_lock)->rlock);
                               local_irq_disable();
                               lock(&(&nf_conntrack_locks[i])->rlock);
                               lock(&(&s->cache_lock)->rlock);
  <Interrupt>
    lock(&(&nf_conntrack_locks[i])->rlock);

 *** DEADLOCK ***
[...]
===

Fix this by simply locking with irqsave/irqrestore and stop giving up on
!in_task(). It makes the locking a bit slower, but it shouldn't make a
big difference in real workloads. Under the scenario from [1] (only
cache hits) it only increased the runtime overhead from the
security_secid_to_secctx() function from ~2% to ~3% (it was ~5-65%
before introducing the cache).

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1733259

Fixes: a880aefcb5ae ("selinux: cache the SID -> context string translation")
Reported-by: syzbot+61cba5033e2072d61806@syzkaller.appspotmail.com
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: Paul Moore <paul@paul-moore.com>
security/selinux/ss/sidtab.c