bpf: verifier: add checks for BPF_ABS | BPF_IND instructions
authorAlexei Starovoitov <ast@plumgrid.com>
Mon, 1 Dec 2014 23:06:34 +0000 (15:06 -0800)
committerDavid S. Miller <davem@davemloft.net>
Sat, 6 Dec 2014 05:47:32 +0000 (21:47 -0800)
introduce program type BPF_PROG_TYPE_SOCKET_FILTER that is used
for attaching programs to sockets where ctx == skb.

add verifier checks for ABS/IND instructions which can only be seen
in socket filters, therefore the check:
  if (env->prog->aux->prog_type != BPF_PROG_TYPE_SOCKET_FILTER)
    verbose("BPF_LD_ABS|IND instructions are only allowed in socket filters\n");

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/uapi/linux/bpf.h
kernel/bpf/verifier.c

index 4a3d0f8..45da7ec 100644 (file)
@@ -117,6 +117,7 @@ enum bpf_map_type {
 
 enum bpf_prog_type {
        BPF_PROG_TYPE_UNSPEC,
+       BPF_PROG_TYPE_SOCKET_FILTER,
 };
 
 /* flags for BPF_MAP_UPDATE_ELEM command */
index b6a1f7c..a28e09c 100644 (file)
@@ -1172,6 +1172,70 @@ static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
        return 0;
 }
 
+/* verify safety of LD_ABS|LD_IND instructions:
+ * - they can only appear in the programs where ctx == skb
+ * - since they are wrappers of function calls, they scratch R1-R5 registers,
+ *   preserve R6-R9, and store return value into R0
+ *
+ * Implicit input:
+ *   ctx == skb == R6 == CTX
+ *
+ * Explicit input:
+ *   SRC == any register
+ *   IMM == 32-bit immediate
+ *
+ * Output:
+ *   R0 - 8/16/32-bit skb data converted to cpu endianness
+ */
+static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
+{
+       struct reg_state *regs = env->cur_state.regs;
+       u8 mode = BPF_MODE(insn->code);
+       struct reg_state *reg;
+       int i, err;
+
+       if (env->prog->aux->prog_type != BPF_PROG_TYPE_SOCKET_FILTER) {
+               verbose("BPF_LD_ABS|IND instructions are only allowed in socket filters\n");
+               return -EINVAL;
+       }
+
+       if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
+           (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
+               verbose("BPF_LD_ABS uses reserved fields\n");
+               return -EINVAL;
+       }
+
+       /* check whether implicit source operand (register R6) is readable */
+       err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
+       if (err)
+               return err;
+
+       if (regs[BPF_REG_6].type != PTR_TO_CTX) {
+               verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
+               return -EINVAL;
+       }
+
+       if (mode == BPF_IND) {
+               /* check explicit source operand */
+               err = check_reg_arg(regs, insn->src_reg, SRC_OP);
+               if (err)
+                       return err;
+       }
+
+       /* reset caller saved regs to unreadable */
+       for (i = 0; i < CALLER_SAVED_REGS; i++) {
+               reg = regs + caller_saved[i];
+               reg->type = NOT_INIT;
+               reg->imm = 0;
+       }
+
+       /* mark destination R0 register as readable, since it contains
+        * the value fetched from the packet
+        */
+       regs[BPF_REG_0].type = UNKNOWN_VALUE;
+       return 0;
+}
+
 /* non-recursive DFS pseudo code
  * 1  procedure DFS-iterative(G,v):
  * 2      label v as discovered
@@ -1677,8 +1741,10 @@ process_bpf_exit:
                        u8 mode = BPF_MODE(insn->code);
 
                        if (mode == BPF_ABS || mode == BPF_IND) {
-                               verbose("LD_ABS is not supported yet\n");
-                               return -EINVAL;
+                               err = check_ld_abs(env, insn);
+                               if (err)
+                                       return err;
+
                        } else if (mode == BPF_IMM) {
                                err = check_ld_imm(env, insn);
                                if (err)