Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[deliverable/linux.git] / net / sched / act_bpf.c
index e9e923a8c74754a4d6bb062f33164c1b1c0f78a0..aaae8e83bf189cc99fb624bf581979df7ce6fce5 100644 (file)
 struct tcf_bpf_cfg {
        struct bpf_prog *filter;
        struct sock_filter *bpf_ops;
-       char *bpf_name;
+       const char *bpf_name;
        u32 bpf_fd;
        u16 bpf_num_ops;
+       bool is_ebpf;
 };
 
 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
@@ -207,6 +208,7 @@ static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
        cfg->bpf_ops = bpf_ops;
        cfg->bpf_num_ops = bpf_num_ops;
        cfg->filter = fp;
+       cfg->is_ebpf = false;
 
        return 0;
 }
@@ -241,18 +243,40 @@ static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
        cfg->bpf_fd = bpf_fd;
        cfg->bpf_name = name;
        cfg->filter = fp;
+       cfg->is_ebpf = true;
 
        return 0;
 }
 
+static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
+{
+       if (cfg->is_ebpf)
+               bpf_prog_put(cfg->filter);
+       else
+               bpf_prog_destroy(cfg->filter);
+
+       kfree(cfg->bpf_ops);
+       kfree(cfg->bpf_name);
+}
+
+static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
+                                 struct tcf_bpf_cfg *cfg)
+{
+       cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
+       cfg->filter = prog->filter;
+
+       cfg->bpf_ops = prog->bpf_ops;
+       cfg->bpf_name = prog->bpf_name;
+}
+
 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
                        struct nlattr *est, struct tc_action *act,
                        int replace, int bind)
 {
        struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
+       struct tcf_bpf_cfg cfg, old;
        struct tc_act_bpf *parm;
        struct tcf_bpf *prog;
-       struct tcf_bpf_cfg cfg;
        bool is_bpf, is_ebpf;
        int ret;
 
@@ -301,6 +325,9 @@ static int tcf_bpf_init(struct net *net, struct nlattr *nla,
        prog = to_bpf(act);
        spin_lock_bh(&prog->tcf_lock);
 
+       if (ret != ACT_P_CREATED)
+               tcf_bpf_prog_fill_cfg(prog, &old);
+
        prog->bpf_ops = cfg.bpf_ops;
        prog->bpf_name = cfg.bpf_name;
 
@@ -316,32 +343,22 @@ static int tcf_bpf_init(struct net *net, struct nlattr *nla,
 
        if (ret == ACT_P_CREATED)
                tcf_hash_insert(act);
+       else
+               tcf_bpf_cfg_cleanup(&old);
 
        return ret;
 
 destroy_fp:
-       if (is_ebpf)
-               bpf_prog_put(cfg.filter);
-       else
-               bpf_prog_destroy(cfg.filter);
-
-       kfree(cfg.bpf_ops);
-       kfree(cfg.bpf_name);
-
+       tcf_bpf_cfg_cleanup(&cfg);
        return ret;
 }
 
 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
 {
-       const struct tcf_bpf *prog = act->priv;
-
-       if (tcf_bpf_is_ebpf(prog))
-               bpf_prog_put(prog->filter);
-       else
-               bpf_prog_destroy(prog->filter);
+       struct tcf_bpf_cfg tmp;
 
-       kfree(prog->bpf_ops);
-       kfree(prog->bpf_name);
+       tcf_bpf_prog_fill_cfg(act->priv, &tmp);
+       tcf_bpf_cfg_cleanup(&tmp);
 }
 
 static struct tc_action_ops act_bpf_ops __read_mostly = {
This page took 0.025123 seconds and 5 git commands to generate.