]> git.baikalelectronics.ru Git - kernel.git/commitdiff
powerpc/bpf: Fix tail call implementation
authorEric Dumazet <edumazet@google.com>
Fri, 1 Nov 2019 03:34:44 +0000 (20:34 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Fri, 1 Nov 2019 23:32:26 +0000 (00:32 +0100)
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: 43db670a33cd1 ("powerpc/bpf: Implement support for tail calls")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
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: Martin KaFai Lau <kafai@fb.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20191101033444.143741-1-edumazet@google.com
arch/powerpc/net/bpf_jit_comp64.c

index 02a59946a78af46416e9d949047ad3fefe3fc159..be3517ef0574d0911f6d63b530b33954c8ca343d 100644 (file)
@@ -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