net: x86: bpf: don't forget to free sk_filter (v2)
authorAndrey Vagin <avagin@openvz.org>
Thu, 7 Nov 2013 04:35:12 +0000 (08:35 +0400)
committerDavid S. Miller <davem@davemloft.net>
Fri, 8 Nov 2013 00:06:52 +0000 (19:06 -0500)
sk_filter isn't freed if bpf_func is equal to sk_run_filter.

This memory leak was introduced by v3.12-rc3-224-gd45ed4a4
"net: fix unsafe set_memory_rw from softirq".

Before this patch sk_filter was freed in sk_filter_release_rcu,
now it should be freed in bpf_jit_free.

Here is output of kmemleak:
unreferenced object 0xffff8800b774eab0 (size 128):
  comm "systemd", pid 1, jiffies 4294669014 (age 124.062s)
  hex dump (first 32 bytes):
    00 00 00 00 0b 00 00 00 20 63 7f b7 00 88 ff ff  ........ c......
    60 d4 55 81 ff ff ff ff 30 d9 55 81 ff ff ff ff  `.U.....0.U.....
  backtrace:
    [<ffffffff816444be>] kmemleak_alloc+0x4e/0xb0
    [<ffffffff811845af>] __kmalloc+0xef/0x260
    [<ffffffff81534028>] sock_kmalloc+0x38/0x60
    [<ffffffff8155d4dd>] sk_attach_filter+0x5d/0x190
    [<ffffffff815378a1>] sock_setsockopt+0x991/0x9e0
    [<ffffffff81531bd6>] SyS_setsockopt+0xb6/0xd0
    [<ffffffff8165f3e9>] system_call_fastpath+0x16/0x1b
    [<ffffffffffffffff>] 0xffffffffffffffff

v2: add extra { } after else

Fixes: d45ed4a4e33a ("net: fix unsafe set_memory_rw from softirq")
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
arch/x86/net/bpf_jit_comp.c

index 516593e1ce33b92175195c8de2983258b723493c..26328e800869fc6ac1821c0ab3dcd2baa1506ec1 100644 (file)
@@ -788,5 +788,7 @@ void bpf_jit_free(struct sk_filter *fp)
        if (fp->bpf_func != sk_run_filter) {
                INIT_WORK(&fp->work, bpf_jit_free_deferred);
                schedule_work(&fp->work);
+       } else {
+               kfree(fp);
        }
 }