net/sched/act_bpf.c

   1 /*
   2  * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/init.h>
  12 #include <linux/kernel.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/rtnetlink.h>
  15 #include <linux/filter.h>
  16 #include <net/netlink.h>
  17 #include <net/pkt_sched.h>
  18
  19 #include <linux/tc_act/tc_bpf.h>
  20 #include <net/tc_act/tc_bpf.h>
  21
  22 #define BPF_TAB_MASK     15
  23
  24 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *a,
  25                    struct tcf_result *res)
  26 {
  27         struct tcf_bpf *b = a->priv;
  28         int action, filter_res;
  29
  30         spin_lock(&b->tcf_lock);
  31
  32         b->tcf_tm.lastuse = jiffies;
  33         bstats_update(&b->tcf_bstats, skb);
  34
  35         filter_res = BPF_PROG_RUN(b->filter, skb);
  36
  37         /* A BPF program may overwrite the default action opcode.
  38          * Similarly as in cls_bpf, if filter_res == -1 we use the
  39          * default action specified from tc.
  40          *
  41          * In case a different well-known TC_ACT opcode has been
  42          * returned, it will overwrite the default one.
  43          *
  44          * For everything else that is unkown, TC_ACT_UNSPEC is
  45          * returned.
  46          */
  47         switch (filter_res) {
  48         case TC_ACT_PIPE:
  49         case TC_ACT_RECLASSIFY:
  50         case TC_ACT_OK:
  51                 action = filter_res;
  52                 break;
  53         case TC_ACT_SHOT:
  54                 action = filter_res;
  55                 b->tcf_qstats.drops++;
  56                 break;
  57         case TC_ACT_UNSPEC:
  58                 action = b->tcf_action;
  59                 break;
  60         default:
  61                 action = TC_ACT_UNSPEC;
  62                 break;
  63         }
  64
  65         spin_unlock(&b->tcf_lock);
  66         return action;
  67 }
  68
  69 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *a,
  70                         int bind, int ref)
  71 {
  72         unsigned char *tp = skb_tail_pointer(skb);
  73         struct tcf_bpf *b = a->priv;
  74         struct tc_act_bpf opt = {
  75                 .index    = b->tcf_index,
  76                 .refcnt   = b->tcf_refcnt - ref,
  77                 .bindcnt  = b->tcf_bindcnt - bind,
  78                 .action   = b->tcf_action,
  79         };
  80         struct tcf_t t;
  81         struct nlattr *nla;
  82
  83         if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
  84                 goto nla_put_failure;
  85
  86         if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, b->bpf_num_ops))
  87                 goto nla_put_failure;
  88
  89         nla = nla_reserve(skb, TCA_ACT_BPF_OPS, b->bpf_num_ops *
  90                           sizeof(struct sock_filter));
  91         if (!nla)
  92                 goto nla_put_failure;
  93
  94         memcpy(nla_data(nla), b->bpf_ops, nla_len(nla));
  95
  96         t.install = jiffies_to_clock_t(jiffies - b->tcf_tm.install);
  97         t.lastuse = jiffies_to_clock_t(jiffies - b->tcf_tm.lastuse);
  98         t.expires = jiffies_to_clock_t(b->tcf_tm.expires);
  99         if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(t), &t))
 100                 goto nla_put_failure;
 101         return skb->len;
 102
 103 nla_put_failure:
 104         nlmsg_trim(skb, tp);
 105         return -1;
 106 }
 107
 108 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
 109         [TCA_ACT_BPF_PARMS]     = { .len = sizeof(struct tc_act_bpf) },
 110         [TCA_ACT_BPF_OPS_LEN]   = { .type = NLA_U16 },
 111         [TCA_ACT_BPF_OPS]       = { .type = NLA_BINARY,
 112                                     .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
 113 };
 114
 115 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
 116                         struct nlattr *est, struct tc_action *a,
 117                         int ovr, int bind)
 118 {
 119         struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
 120         struct tc_act_bpf *parm;
 121         struct tcf_bpf *b;
 122         u16 bpf_size, bpf_num_ops;
 123         struct sock_filter *bpf_ops;
 124         struct sock_fprog_kern tmp;
 125         struct bpf_prog *fp;
 126         int ret;
 127
 128         if (!nla)
 129                 return -EINVAL;
 130
 131         ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
 132         if (ret < 0)
 133                 return ret;
 134
 135         if (!tb[TCA_ACT_BPF_PARMS] ||
 136             !tb[TCA_ACT_BPF_OPS_LEN] || !tb[TCA_ACT_BPF_OPS])
 137                 return -EINVAL;
 138         parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
 139
 140         bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
 141         if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
 142                 return -EINVAL;
 143
 144         bpf_size = bpf_num_ops * sizeof(*bpf_ops);
 145         if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
 146                 return -EINVAL;
 147
 148         bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
 149         if (!bpf_ops)
 150                 return -ENOMEM;
 151
 152         memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
 153
 154         tmp.len = bpf_num_ops;
 155         tmp.filter = bpf_ops;
 156
 157         ret = bpf_prog_create(&fp, &tmp);
 158         if (ret)
 159                 goto free_bpf_ops;
 160
 161         if (!tcf_hash_check(parm->index, a, bind)) {
 162                 ret = tcf_hash_create(parm->index, est, a, sizeof(*b), bind);
 163                 if (ret)
 164                         goto destroy_fp;
 165
 166                 ret = ACT_P_CREATED;
 167         } else {
 168                 if (bind)
 169                         goto destroy_fp;
 170                 tcf_hash_release(a, bind);
 171                 if (!ovr) {
 172                         ret = -EEXIST;
 173                         goto destroy_fp;
 174                 }
 175         }
 176
 177         b = to_bpf(a);
 178         spin_lock_bh(&b->tcf_lock);
 179         b->tcf_action = parm->action;
 180         b->bpf_num_ops = bpf_num_ops;
 181         b->bpf_ops = bpf_ops;
 182         b->filter = fp;
 183         spin_unlock_bh(&b->tcf_lock);
 184
 185         if (ret == ACT_P_CREATED)
 186                 tcf_hash_insert(a);
 187         return ret;
 188
 189 destroy_fp:
 190         bpf_prog_destroy(fp);
 191 free_bpf_ops:
 192         kfree(bpf_ops);
 193         return ret;
 194 }
 195
 196 static void tcf_bpf_cleanup(struct tc_action *a, int bind)
 197 {
 198         struct tcf_bpf *b = a->priv;
 199
 200         bpf_prog_destroy(b->filter);
 201 }
 202
 203 static struct tc_action_ops act_bpf_ops = {
 204         .kind =         "bpf",
 205         .type =         TCA_ACT_BPF,
 206         .owner =        THIS_MODULE,
 207         .act =          tcf_bpf,
 208         .dump =         tcf_bpf_dump,
 209         .cleanup =      tcf_bpf_cleanup,
 210         .init =         tcf_bpf_init,
 211 };
 212
 213 static int __init bpf_init_module(void)
 214 {
 215         return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK);
 216 }
 217
 218 static void __exit bpf_cleanup_module(void)
 219 {
 220         tcf_unregister_action(&act_bpf_ops);
 221 }
 222
 223 module_init(bpf_init_module);
 224 module_exit(bpf_cleanup_module);
 225
 226 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
 227 MODULE_DESCRIPTION("TC BPF based action");
 228 MODULE_LICENSE("GPL v2");