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 <linux/bpf.h>
  17
  18 #include <net/netlink.h>
  19 #include <net/pkt_sched.h>
  20
  21 #include <linux/tc_act/tc_bpf.h>
  22 #include <net/tc_act/tc_bpf.h>
  23
  24 #define BPF_TAB_MASK            15
  25 #define ACT_BPF_NAME_LEN        256
  26
  27 struct tcf_bpf_cfg {
  28         struct bpf_prog *filter;
  29         struct sock_filter *bpf_ops;
  30         char *bpf_name;
  31         u32 bpf_fd;
  32         u16 bpf_num_ops;
  33 };
  34
  35 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
  36                    struct tcf_result *res)
  37 {
  38         struct tcf_bpf *prog = act->priv;
  39         int action, filter_res;
  40         bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
  41
  42         if (unlikely(!skb_mac_header_was_set(skb)))
  43                 return TC_ACT_UNSPEC;
  44
  45         spin_lock(&prog->tcf_lock);
  46
  47         prog->tcf_tm.lastuse = jiffies;
  48         bstats_update(&prog->tcf_bstats, skb);
  49
  50         /* Needed here for accessing maps. */
  51         rcu_read_lock();
  52         if (at_ingress) {
  53                 __skb_push(skb, skb->mac_len);
  54                 filter_res = BPF_PROG_RUN(prog->filter, skb);
  55                 __skb_pull(skb, skb->mac_len);
  56         } else {
  57                 filter_res = BPF_PROG_RUN(prog->filter, skb);
  58         }
  59         rcu_read_unlock();
  60
  61         /* A BPF program may overwrite the default action opcode.
  62          * Similarly as in cls_bpf, if filter_res == -1 we use the
  63          * default action specified from tc.
  64          *
  65          * In case a different well-known TC_ACT opcode has been
  66          * returned, it will overwrite the default one.
  67          *
  68          * For everything else that is unkown, TC_ACT_UNSPEC is
  69          * returned.
  70          */
  71         switch (filter_res) {
  72         case TC_ACT_PIPE:
  73         case TC_ACT_RECLASSIFY:
  74         case TC_ACT_OK:
  75                 action = filter_res;
  76                 break;
  77         case TC_ACT_SHOT:
  78                 action = filter_res;
  79                 prog->tcf_qstats.drops++;
  80                 break;
  81         case TC_ACT_UNSPEC:
  82                 action = prog->tcf_action;
  83                 break;
  84         default:
  85                 action = TC_ACT_UNSPEC;
  86                 break;
  87         }
  88
  89         spin_unlock(&prog->tcf_lock);
  90         return action;
  91 }
  92
  93 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
  94 {
  95         return !prog->bpf_ops;
  96 }
  97
  98 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
  99                                  struct sk_buff *skb)
 100 {
 101         struct nlattr *nla;
 102
 103         if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
 104                 return -EMSGSIZE;
 105
 106         nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
 107                           sizeof(struct sock_filter));
 108         if (nla == NULL)
 109                 return -EMSGSIZE;
 110
 111         memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
 112
 113         return 0;
 114 }
 115
 116 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
 117                                   struct sk_buff *skb)
 118 {
 119         if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
 120                 return -EMSGSIZE;
 121
 122         if (prog->bpf_name &&
 123             nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
 124                 return -EMSGSIZE;
 125
 126         return 0;
 127 }
 128
 129 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
 130                         int bind, int ref)
 131 {
 132         unsigned char *tp = skb_tail_pointer(skb);
 133         struct tcf_bpf *prog = act->priv;
 134         struct tc_act_bpf opt = {
 135                 .index   = prog->tcf_index,
 136                 .refcnt  = prog->tcf_refcnt - ref,
 137                 .bindcnt = prog->tcf_bindcnt - bind,
 138                 .action  = prog->tcf_action,
 139         };
 140         struct tcf_t tm;
 141         int ret;
 142
 143         if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
 144                 goto nla_put_failure;
 145
 146         if (tcf_bpf_is_ebpf(prog))
 147                 ret = tcf_bpf_dump_ebpf_info(prog, skb);
 148         else
 149                 ret = tcf_bpf_dump_bpf_info(prog, skb);
 150         if (ret)
 151                 goto nla_put_failure;
 152
 153         tm.install = jiffies_to_clock_t(jiffies - prog->tcf_tm.install);
 154         tm.lastuse = jiffies_to_clock_t(jiffies - prog->tcf_tm.lastuse);
 155         tm.expires = jiffies_to_clock_t(prog->tcf_tm.expires);
 156
 157         if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm))
 158                 goto nla_put_failure;
 159
 160         return skb->len;
 161
 162 nla_put_failure:
 163         nlmsg_trim(skb, tp);
 164         return -1;
 165 }
 166
 167 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
 168         [TCA_ACT_BPF_PARMS]     = { .len = sizeof(struct tc_act_bpf) },
 169         [TCA_ACT_BPF_FD]        = { .type = NLA_U32 },
 170         [TCA_ACT_BPF_NAME]      = { .type = NLA_NUL_STRING, .len = ACT_BPF_NAME_LEN },
 171         [TCA_ACT_BPF_OPS_LEN]   = { .type = NLA_U16 },
 172         [TCA_ACT_BPF_OPS]       = { .type = NLA_BINARY,
 173                                     .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
 174 };
 175
 176 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
 177 {
 178         struct sock_filter *bpf_ops;
 179         struct sock_fprog_kern fprog_tmp;
 180         struct bpf_prog *fp;
 181         u16 bpf_size, bpf_num_ops;
 182         int ret;
 183
 184         bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
 185         if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
 186                 return -EINVAL;
 187
 188         bpf_size = bpf_num_ops * sizeof(*bpf_ops);
 189         if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
 190                 return -EINVAL;
 191
 192         bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
 193         if (bpf_ops == NULL)
 194                 return -ENOMEM;
 195
 196         memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
 197
 198         fprog_tmp.len = bpf_num_ops;
 199         fprog_tmp.filter = bpf_ops;
 200
 201         ret = bpf_prog_create(&fp, &fprog_tmp);
 202         if (ret < 0) {
 203                 kfree(bpf_ops);
 204                 return ret;
 205         }
 206
 207         cfg->bpf_ops = bpf_ops;
 208         cfg->bpf_num_ops = bpf_num_ops;
 209         cfg->filter = fp;
 210
 211         return 0;
 212 }
 213
 214 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
 215 {
 216         struct bpf_prog *fp;
 217         char *name = NULL;
 218         u32 bpf_fd;
 219
 220         bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
 221
 222         fp = bpf_prog_get(bpf_fd);
 223         if (IS_ERR(fp))
 224                 return PTR_ERR(fp);
 225
 226         if (fp->type != BPF_PROG_TYPE_SCHED_ACT) {
 227                 bpf_prog_put(fp);
 228                 return -EINVAL;
 229         }
 230
 231         if (tb[TCA_ACT_BPF_NAME]) {
 232                 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
 233                                nla_len(tb[TCA_ACT_BPF_NAME]),
 234                                GFP_KERNEL);
 235                 if (!name) {
 236                         bpf_prog_put(fp);
 237                         return -ENOMEM;
 238                 }
 239         }
 240
 241         cfg->bpf_fd = bpf_fd;
 242         cfg->bpf_name = name;
 243         cfg->filter = fp;
 244
 245         return 0;
 246 }
 247
 248 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
 249                         struct nlattr *est, struct tc_action *act,
 250                         int replace, int bind)
 251 {
 252         struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
 253         struct tc_act_bpf *parm;
 254         struct tcf_bpf *prog;
 255         struct tcf_bpf_cfg cfg;
 256         bool is_bpf, is_ebpf;
 257         int ret;
 258
 259         if (!nla)
 260                 return -EINVAL;
 261
 262         ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
 263         if (ret < 0)
 264                 return ret;
 265
 266         is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
 267         is_ebpf = tb[TCA_ACT_BPF_FD];
 268
 269         if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf) ||
 270             !tb[TCA_ACT_BPF_PARMS])
 271                 return -EINVAL;
 272
 273         parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
 274
 275         memset(&cfg, 0, sizeof(cfg));
 276
 277         ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
 278                        tcf_bpf_init_from_efd(tb, &cfg);
 279         if (ret < 0)
 280                 return ret;
 281
 282         if (!tcf_hash_check(parm->index, act, bind)) {
 283                 ret = tcf_hash_create(parm->index, est, act,
 284                                       sizeof(*prog), bind);
 285                 if (ret < 0)
 286                         goto destroy_fp;
 287
 288                 ret = ACT_P_CREATED;
 289         } else {
 290                 /* Don't override defaults. */
 291                 if (bind)
 292                         goto destroy_fp;
 293
 294                 tcf_hash_release(act, bind);
 295                 if (!replace) {
 296                         ret = -EEXIST;
 297                         goto destroy_fp;
 298                 }
 299         }
 300
 301         prog = to_bpf(act);
 302         spin_lock_bh(&prog->tcf_lock);
 303
 304         prog->bpf_ops = cfg.bpf_ops;
 305         prog->bpf_name = cfg.bpf_name;
 306
 307         if (cfg.bpf_num_ops)
 308                 prog->bpf_num_ops = cfg.bpf_num_ops;
 309         if (cfg.bpf_fd)
 310                 prog->bpf_fd = cfg.bpf_fd;
 311
 312         prog->tcf_action = parm->action;
 313         prog->filter = cfg.filter;
 314
 315         spin_unlock_bh(&prog->tcf_lock);
 316
 317         if (ret == ACT_P_CREATED)
 318                 tcf_hash_insert(act);
 319
 320         return ret;
 321
 322 destroy_fp:
 323         if (is_ebpf)
 324                 bpf_prog_put(cfg.filter);
 325         else
 326                 bpf_prog_destroy(cfg.filter);
 327
 328         kfree(cfg.bpf_ops);
 329         kfree(cfg.bpf_name);
 330
 331         return ret;
 332 }
 333
 334 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
 335 {
 336         const struct tcf_bpf *prog = act->priv;
 337
 338         if (tcf_bpf_is_ebpf(prog))
 339                 bpf_prog_put(prog->filter);
 340         else
 341                 bpf_prog_destroy(prog->filter);
 342 }
 343
 344 static struct tc_action_ops act_bpf_ops __read_mostly = {
 345         .kind           =       "bpf",
 346         .type           =       TCA_ACT_BPF,
 347         .owner          =       THIS_MODULE,
 348         .act            =       tcf_bpf,
 349         .dump           =       tcf_bpf_dump,
 350         .cleanup        =       tcf_bpf_cleanup,
 351         .init           =       tcf_bpf_init,
 352 };
 353
 354 static int __init bpf_init_module(void)
 355 {
 356         return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK);
 357 }
 358
 359 static void __exit bpf_cleanup_module(void)
 360 {
 361         tcf_unregister_action(&act_bpf_ops);
 362 }
 363
 364 module_init(bpf_init_module);
 365 module_exit(bpf_cleanup_module);
 366
 367 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
 368 MODULE_DESCRIPTION("TC BPF based action");
 369 MODULE_LICENSE("GPL v2");