lkml.org 
[lkml]   [2020]   [Oct]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Litmus test for question from Al Viro
On Wed, Sep 30, 2020 at 09:51:16PM -0700, Paul E. McKenney wrote:
> Hello!
>
> Al Viro posted the following query:
>
> ------------------------------------------------------------------------
>
> <viro> fun question regarding barriers, if you have time for that
> <viro> V->A = V->B = 1;
> <viro>
> <viro> CPU1:
> <viro> to_free = NULL
> <viro> spin_lock(&LOCK)
> <viro> if (!smp_load_acquire(&V->B))
> <viro> to_free = V
> <viro> V->A = 0
> <viro> spin_unlock(&LOCK)
> <viro> kfree(to_free)
> <viro>
> <viro> CPU2:
> <viro> to_free = V;
> <viro> if (READ_ONCE(V->A)) {
> <viro> spin_lock(&LOCK)
> <viro> if (V->A)
> <viro> to_free = NULL
> <viro> smp_store_release(&V->B, 0);
> <viro> spin_unlock(&LOCK)
> <viro> }
> <viro> kfree(to_free);
> <viro> 1) is it guaranteed that V will be freed exactly once and that
> no accesses to *V will happen after freeing it?
> <viro> 2) do we need smp_store_release() there? I.e. will anything
> break if it's replaced with plain V->B = 0?

Here are my answers to Al's questions:

1) It is guaranteed that V will be freed exactly once. It is not
guaranteed that no accesses to *V will occur after it is freed, because
the test contains a data race. CPU1's plain "V->A = 0" write races with
CPU2's READ_ONCE; if the plain write were replaced with
"WRITE_ONCE(V->A, 0)" then the guarantee would hold. Equally well,
CPU1's smp_load_acquire could be replaced with a plain read while the
plain write is replaced with smp_store_release.

2) The smp_store_release in CPU2 is not needed. Replacing it with a
plain V->B = 0 will not break anything.

Analysis: Apart from the kfree calls themselves, the only access to a
shared variable outside of a critical section is CPU2's READ_ONCE of
V->A. So let's consider two possibilities:

1: The READ_ONCE returns 0. Then CPU2 doesn't execute its critical
section and does kfree(V). However, the fact that the READ_ONCE got 0
means that CPU1 has already entered its critical section, has already
written to V->A (but with a plain write!) and therefore has already seen
V->B = 1 (because of the smp_load_acquire), and therefore will not free
V. This case shows that the ordering we require is for CPU1 to read
V->B before it writes V->A. The ordering can be enforced by using
either a load-acquire (as in the litmus test) or a store-release.

2: The READ_ONCE returns 1. Then CPU2 does execute its critical
section, and we can simply treat this case the same as if the critical
section was executed unconditionally. Whichever CPU runs its critical
section second will free V, and the other CPU won't try to access V
after leaving its own critical section (and thus won't access V after it
has been freed).

> ------------------------------------------------------------------------
>
> Of course herd7 supports neither structures nor arrays, but I was
> crazy enough to try individual variables with made-up address and data
> dependencies. This litmus test must also detect use-after-free bugs,
> but a simple variable should be able to do that. So here is a
> prototype:
>
> ------------------------------------------------------------------------
>
> C C-viro-2020.09.29a
>
> {
> int a = 1;
> int b = 1;
> int v = 1;
> }

Not the way I would have done it, but okay. I would have modeled the
kfree by setting a and b both to some sentinel value.

> P0(int *a, int *b, int *v, spinlock_t *l)
> {
> int r0;
> int r1;
> int r2 = 2;
> int r8;
> int r9a = 2;
> int r9b = 2;
>
> r0 = 0;
> spin_lock(l);
> r9a = READ_ONCE(*v); // Use after free?
> r8 = r9a - r9a; // Restore address dependency
> r8 = b + r8;
> r1 = smp_load_acquire(r8);
> if (r1 == 0)
> r0 = 1;
> r9b = READ_ONCE(*v); // Use after free?
> WRITE_ONCE(*a, r9b - r9b); // Use data dependency
> spin_unlock(l);
> if (r0) {
> r2 = READ_ONCE(*v);
> WRITE_ONCE(*v, 0); /* kfree(). */
> }
> }
>
> P1(int *a, int *b, int *v, spinlock_t *l)
> {
> int r0;
> int r1;
> int r1a;
> int r2 = 2;
> int r8;
> int r9a = 2;
> int r9b = 2;
> int r9c = 2;
>
> r0 = READ_ONCE(*v);
> r9a = r0; // Use after free?

Wrong. This should be:

r0 = 1;
r9a = READ_ONCE(*v);

> r8 = r9a - r9a; // Restore address dependency
> r8 = a + r8;
> r1 = READ_ONCE(*r8);
> if (r1) {
> spin_lock(l);
> r9b = READ_ONCE(*v); // Use after free?
> r8 = r9b - r9b; // Restore address dependency
> r8 = a + r8;
> r1a = READ_ONCE(*r8);
> if (r1a)
> r0 = 0;
> r9c = READ_ONCE(*v); // Use after free?
> smp_store_release(b, r9c - rc9); // Use data dependency
> spin_unlock(l);
> }
> if (r0) {
> r2 = READ_ONCE(*v);
> WRITE_ONCE(*v, 0); /* kfree(). */
> }
> }
>
> locations [a;b;v;0:r1;0:r8;1:r1;1:r8]
> exists (0:r0=1:r0 \/ (* Both or neither did kfree(). *)
> v=1 \/ (* Neither did kfree, redundant check. *)
> 0:r2=0 \/ 1:r2=0 \/ (* Both did kfree, redundant check. *)
> 0:r9a=0 \/ 0:r9b=0 \/ 1:r9a=0 \/ (* CPU1 use after free. *)
> 1:r9b=0 \/ 1:r9c=0) (* CPU2 use after free. *)
>
> ------------------------------------------------------------------------
>
> This "exists" clause is satisfied:
>
> ------------------------------------------------------------------------
>
> $ herd7 -conf linux-kernel.cfg ~/paper/scalability/LWNLinuxMM/litmus/manual/kernel/C-viro-2020.09.29a.litmus
> Test C-viro-2020.09.29a Allowed
> States 5
> 0:r0=0; 0:r1=1; 0:r2=2; 0:r8=b; 0:r9a=0; 0:r9b=0; 1:r0=1; 1:r1=0; 1:r2=1; 1:r8=a; 1:r9a=1; 1:r9b=2; 1:r9c=2; a=0; b=1; v=0;
> 0:r0=0; 0:r1=1; 0:r2=2; 0:r8=b; 0:r9a=1; 0:r9b=0; 1:r0=1; 1:r1=0; 1:r2=1; 1:r8=a; 1:r9a=1; 1:r9b=2; 1:r9c=2; a=0; b=1; v=0;
> 0:r0=0; 0:r1=1; 0:r2=2; 0:r8=b; 0:r9a=1; 0:r9b=1; 1:r0=0; 1:r1=1; 1:r2=2; 1:r8=a; 1:r9a=1; 1:r9b=1; 1:r9c=1; a=0; b=1; v=1;

The values for this case don't make sense. I haven't checked the other
four cases. Printing a graph of the relations for this case (the only
state with v=1 at the end) might help.

> 0:r0=0; 0:r1=1; 0:r2=2; 0:r8=b; 0:r9a=1; 0:r9b=1; 1:r0=1; 1:r1=0; 1:r2=1; 1:r8=a; 1:r9a=1; 1:r9b=2; 1:r9c=2; a=0; b=1; v=0;
> 0:r0=0; 0:r1=1; 0:r2=2; 0:r8=b; 0:r9a=1; 0:r9b=1; 1:r0=1; 1:r1=1; 1:r2=1; 1:r8=a; 1:r9a=1; 1:r9b=1; 1:r9c=1; a=0; b=1; v=0;
> Ok
> Witnesses
> Positive: 3 Negative: 2
> Condition exists (0:r0=1:r0 \/ v=1 \/ 0:r2=0 \/ 1:r2=0 \/ 0:r9a=0 \/ 0:r9b=0 \/ 1:r9a=0 \/ 1:r9b=0 \/ 1:r9c=0)
> Observation C-viro-2020.09.29a Sometimes 3 2
> Time C-viro-2020.09.29a 14.33
> Hash=89f74abff4de682ee0bea8ee6dd53134

Why didn't this flag the data race?

> ------------------------------------------------------------------------
>
> So did we end up with herd7 not respecting "fake" dependencies like
> those shown above, or have I just messed up the translation from Al's
> example to the litmus test? (Given one thing and another over the past
> couple of days, my guess would be that I just messed up the translation,
> especially given that I don't see a reference to fake dependencies in
> the documentation, but I figured that I should ask.)

What do you get if you fix up the litmus test?

Alan

\
 
 \ /
  Last update: 2020-10-01 18:16    [W:0.155 / U:3.232 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site