[PATCH net] powerpc/bpf: fix tail call implementation

[Eric Dumazet <edumazet@google.com>]

We have seen many crashes on powerpc hosts while loading bpf programs.

The problem here is that bpf_int_jit_compile() does a first pass
to compute the program length.

Then it allocates memory to store the generated program and
calls bpf_jit_build_body() a second time (and a third time
later)

What I have observed is that the second bpf_jit_build_body()
could end up using few more words than expected.

If bpf_jit_binary_alloc() put the space for the program
at the end of the allocated page, we then write on
a non mapped memory.

It appears that bpf_jit_emit_tail_call() calls
bpf_jit_emit_common_epilogue() while ctx->seen might not
be stable.

Only after the second pass we can be sure ctx->seen wont be changed.

Trying to avoid a second pass seems quite complex and probably
not worth it.

Fixes: ce0761419faef ("powerpc/bpf: Implement support for tail calls")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Cc: netdev@vger.kernel.org
Cc: bpf@vger.kernel.org
---
 arch/powerpc/net/bpf_jit_comp64.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
index 02a59946a78af46416e9d949047ad3fefe3fc159..be3517ef0574d0911f6d63b530b33954c8ca343d 100644
--- a/arch/powerpc/net/bpf_jit_comp64.c
+++ b/arch/powerpc/net/bpf_jit_comp64.c
@@ -1141,6 +1141,19 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 		goto out_addrs;
 	}
 
+	/*
+	 * If we have seen a tail call, we need a second pass.
+	 * This is because bpf_jit_emit_common_epilogue() is called
+	 * from bpf_jit_emit_tail_call() with a not yet stable ctx->seen.
+	 */
+	if (cgctx.seen & SEEN_TAILCALL) {
+		cgctx.idx = 0;
+		if (bpf_jit_build_body(fp, 0, &cgctx, addrs, false)) {
+			fp = org_fp;
+			goto out_addrs;
+		}
+	}
+
 	/*
 	 * Pretend to build prologue, given the features we've seen.  This will
 	 * update ctgtx.idx as it pretends to output instructions, then we can
-- 
2.24.0.rc1.363.gb1bccd3e3d-goog