bpf: prevent memory disambiguation attack

Detect code patterns where malicious 'speculative store bypass' can be used and sanitize such patterns.

39: (bf) r3 = r10
40: (07) r3 += -216

 41: (79) r8 = *(u64 *)(r7 +0)   // slow read
 42: (7a) *(u64 *)(r10 -72) = 0  // verifier inserts this instruction
 43: (7b) *(u64 *)(r8 +0) = r3   // this store becomes slow due to r8
 44: (79) r1 = *(u64 *)(r6 +0)   // cpu speculatively executes this load
 45: (71) r2 = *(u8 *)(r1 +0)    // speculatively arbitrary 'load byte'
                                 // is now sanitized

Above code after x86 JIT becomes:

 e5: mov    %rbp,%rdx
 e8: add    $0xffffffffffffff28,%rdx
 ef: mov    0x0(%r13),%r14
 f3: movq   $0x0,-0x48(%rbp)
 fb: mov    %rdx,0x0(%r14)
 ff: mov    0x0(%rbx),%rdi

103: movzbq 0x0(%rdi),%rsi

---
 include/linux/bpf_verifier.h |  1 +
 kernel/bpf/verifier.c        | 59 +++++++++++++++++++++++++++++++++++++++++---
 2 files changed, 57 insertions(+), 3 deletions(-)

index 7e61c395fddf..65cfc2f59db9 100644
@@ -146,6 +146,7 @@ struct bpf_insn_aux_data {
 		s32 call_imm;			/* saved imm field of call insn */
 	};
 	int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
+	int sanitize_stack_off; /* stack slot to be cleared */
 	bool seen; /* this insn was processed by the verifier */
 };
 

2.9.5

(No further action needed since already in Linus tree [1]. This went via the
batch of x86 fixes on the speculative store buffer bypass from today [2].)

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=af86ca4e3088fe5eacf2f7e58c01fa68ca067672
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=3b78ce4a34b761c7fe13520de822984019ff1a8f