]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
x86: bpf_jit_comp: add pkt_type support
authorEric Dumazet <edumazet@google.com>
Thu, 31 Jan 2013 01:51:44 +0000 (17:51 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 31 Jan 2013 03:38:34 +0000 (22:38 -0500)
Supporting access to skb->pkt_type is a bit tricky if we want
to have a generic code, allowing pkt_type to be moved in struct sk_buff

pkt_type is a bit field, so compiler cannot really help us to find
its offset. Let's use a helper for this : It will throw a one time
message if pkt_type no longer starts at a byte boundary or is
no longer a 3bit field.

Reported-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
arch/x86/net/bpf_jit_comp.c

index d11a47099d330d1dcdc21261df571212551f3f03..3cbe45381bbb12de387729b7f070822207b3d6dc 100644 (file)
@@ -1,6 +1,6 @@
 /* bpf_jit_comp.c : BPF JIT compiler
  *
- * Copyright (C) 2011 Eric Dumazet (eric.dumazet@gmail.com)
+ * Copyright (C) 2011-2013 Eric Dumazet (eric.dumazet@gmail.com)
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
@@ -124,6 +124,26 @@ static inline void bpf_flush_icache(void *start, void *end)
 #define CHOOSE_LOAD_FUNC(K, func) \
        ((int)K < 0 ? ((int)K >= SKF_LL_OFF ? func##_negative_offset : func) : func##_positive_offset)
 
+/* Helper to find the offset of pkt_type in sk_buff
+ * We want to make sure its still a 3bit field starting at a byte boundary.
+ */
+#define PKT_TYPE_MAX 7
+static int pkt_type_offset(void)
+{
+       struct sk_buff skb_probe = {
+               .pkt_type = ~0,
+       };
+       char *ct = (char *)&skb_probe;
+       unsigned int off;
+
+       for (off = 0; off < sizeof(struct sk_buff); off++) {
+               if (ct[off] == PKT_TYPE_MAX)
+                       return off;
+       }
+       pr_err_once("Please fix pkt_type_offset(), as pkt_type couldn't be found\n");
+       return -1;
+}
+
 void bpf_jit_compile(struct sk_filter *fp)
 {
        u8 temp[64];
@@ -216,6 +236,7 @@ void bpf_jit_compile(struct sk_filter *fp)
                case BPF_S_ANC_VLAN_TAG:
                case BPF_S_ANC_VLAN_TAG_PRESENT:
                case BPF_S_ANC_QUEUE:
+               case BPF_S_ANC_PKTTYPE:
                case BPF_S_LD_W_ABS:
                case BPF_S_LD_H_ABS:
                case BPF_S_LD_B_ABS:
@@ -536,6 +557,23 @@ void bpf_jit_compile(struct sk_filter *fp)
                                        EMIT3(0x83, 0xe0, 0x01); /* and    $0x1,%eax */
                                }
                                break;
+                       case BPF_S_ANC_PKTTYPE:
+                       {
+                               int off = pkt_type_offset();
+
+                               if (off < 0)
+                                       goto out;
+                               if (is_imm8(off)) {
+                                       /* movzbl off8(%rdi),%eax */
+                                       EMIT4(0x0f, 0xb6, 0x47, off);
+                               } else {
+                                       /* movbl off32(%rdi),%eax */
+                                       EMIT3(0x0f, 0xb6, 0x87);
+                                       EMIT(off, 4);
+                               }
+                               EMIT3(0x83, 0xe0, PKT_TYPE_MAX); /* and    $0x7,%eax */
+                               break;
+                       }
                        case BPF_S_LD_W_ABS:
                                func = CHOOSE_LOAD_FUNC(K, sk_load_word);
 common_load:                   seen |= SEEN_DATAREF;