net/sched/cls_bpf.c

   1 /*
   2  * Berkeley Packet Filter based traffic classifier
   3  *
   4  * Might be used to classify traffic through flexible, user-defined and
   5  * possibly JIT-ed BPF filters for traffic control as an alternative to
   6  * ematches.
   7  *
   8  * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/module.h>
  16 #include <linux/types.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/filter.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/pkt_cls.h>
  21 #include <net/sock.h>
  22
  23 MODULE_LICENSE("GPL");
  24 MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
  25 MODULE_DESCRIPTION("TC BPF based classifier");
  26
  27 struct cls_bpf_head {
  28         struct list_head plist;
  29         u32 hgen;
  30         struct rcu_head rcu;
  31 };
  32
  33 struct cls_bpf_prog {
  34         struct bpf_prog *filter;
  35         struct sock_filter *bpf_ops;
  36         struct tcf_exts exts;
  37         struct tcf_result res;
  38         struct list_head link;
  39         u32 handle;
  40         u16 bpf_len;
  41         struct tcf_proto *tp;
  42         struct rcu_head rcu;
  43 };
  44
  45 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
  46         [TCA_BPF_CLASSID]       = { .type = NLA_U32 },
  47         [TCA_BPF_OPS_LEN]       = { .type = NLA_U16 },
  48         [TCA_BPF_OPS]           = { .type = NLA_BINARY,
  49                                     .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
  50 };
  51
  52 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
  53                             struct tcf_result *res)
  54 {
  55         struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
  56         struct cls_bpf_prog *prog;
  57         int ret;
  58
  59         list_for_each_entry_rcu(prog, &head->plist, link) {
  60                 int filter_res = BPF_PROG_RUN(prog->filter, skb);
  61
  62                 if (filter_res == 0)
  63                         continue;
  64
  65                 *res = prog->res;
  66                 if (filter_res != -1)
  67                         res->classid = filter_res;
  68
  69                 ret = tcf_exts_exec(skb, &prog->exts, res);
  70                 if (ret < 0)
  71                         continue;
  72
  73                 return ret;
  74         }
  75
  76         return -1;
  77 }
  78
  79 static int cls_bpf_init(struct tcf_proto *tp)
  80 {
  81         struct cls_bpf_head *head;
  82
  83         head = kzalloc(sizeof(*head), GFP_KERNEL);
  84         if (head == NULL)
  85                 return -ENOBUFS;
  86
  87         INIT_LIST_HEAD_RCU(&head->plist);
  88         rcu_assign_pointer(tp->root, head);
  89
  90         return 0;
  91 }
  92
  93 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
  94 {
  95         tcf_exts_destroy(&prog->exts);
  96
  97         bpf_prog_destroy(prog->filter);
  98
  99         kfree(prog->bpf_ops);
 100         kfree(prog);
 101 }
 102
 103 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
 104 {
 105         struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
 106
 107         cls_bpf_delete_prog(prog->tp, prog);
 108 }
 109
 110 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
 111 {
 112         struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
 113
 114         list_del_rcu(&prog->link);
 115         tcf_unbind_filter(tp, &prog->res);
 116         call_rcu(&prog->rcu, __cls_bpf_delete_prog);
 117         return 0;
 118 }
 119
 120 static void cls_bpf_destroy(struct tcf_proto *tp)
 121 {
 122         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 123         struct cls_bpf_prog *prog, *tmp;
 124
 125         list_for_each_entry_safe(prog, tmp, &head->plist, link) {
 126                 list_del_rcu(&prog->link);
 127                 tcf_unbind_filter(tp, &prog->res);
 128                 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
 129         }
 130
 131         RCU_INIT_POINTER(tp->root, NULL);
 132         kfree_rcu(head, rcu);
 133 }
 134
 135 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
 136 {
 137         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 138         struct cls_bpf_prog *prog;
 139         unsigned long ret = 0UL;
 140
 141         if (head == NULL)
 142                 return 0UL;
 143
 144         list_for_each_entry(prog, &head->plist, link) {
 145                 if (prog->handle == handle) {
 146                         ret = (unsigned long) prog;
 147                         break;
 148                 }
 149         }
 150
 151         return ret;
 152 }
 153
 154 static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
 155                                    struct cls_bpf_prog *prog,
 156                                    unsigned long base, struct nlattr **tb,
 157                                    struct nlattr *est, bool ovr)
 158 {
 159         struct sock_filter *bpf_ops;
 160         struct tcf_exts exts;
 161         struct sock_fprog_kern tmp;
 162         struct bpf_prog *fp;
 163         u16 bpf_size, bpf_len;
 164         u32 classid;
 165         int ret;
 166
 167         if (!tb[TCA_BPF_OPS_LEN] || !tb[TCA_BPF_OPS] || !tb[TCA_BPF_CLASSID])
 168                 return -EINVAL;
 169
 170         tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
 171         ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
 172         if (ret < 0)
 173                 return ret;
 174
 175         classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
 176         bpf_len = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
 177         if (bpf_len > BPF_MAXINSNS || bpf_len == 0) {
 178                 ret = -EINVAL;
 179                 goto errout;
 180         }
 181
 182         bpf_size = bpf_len * sizeof(*bpf_ops);
 183         bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
 184         if (bpf_ops == NULL) {
 185                 ret = -ENOMEM;
 186                 goto errout;
 187         }
 188
 189         memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
 190
 191         tmp.len = bpf_len;
 192         tmp.filter = bpf_ops;
 193
 194         ret = bpf_prog_create(&fp, &tmp);
 195         if (ret)
 196                 goto errout_free;
 197
 198         prog->bpf_len = bpf_len;
 199         prog->bpf_ops = bpf_ops;
 200         prog->filter = fp;
 201         prog->res.classid = classid;
 202
 203         tcf_bind_filter(tp, &prog->res, base);
 204         tcf_exts_change(tp, &prog->exts, &exts);
 205
 206         return 0;
 207 errout_free:
 208         kfree(bpf_ops);
 209 errout:
 210         tcf_exts_destroy(&exts);
 211         return ret;
 212 }
 213
 214 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
 215                                    struct cls_bpf_head *head)
 216 {
 217         unsigned int i = 0x80000000;
 218
 219         do {
 220                 if (++head->hgen == 0x7FFFFFFF)
 221                         head->hgen = 1;
 222         } while (--i > 0 && cls_bpf_get(tp, head->hgen));
 223         if (i == 0)
 224                 pr_err("Insufficient number of handles\n");
 225
 226         return i;
 227 }
 228
 229 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
 230                           struct tcf_proto *tp, unsigned long base,
 231                           u32 handle, struct nlattr **tca,
 232                           unsigned long *arg, bool ovr)
 233 {
 234         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 235         struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
 236         struct nlattr *tb[TCA_BPF_MAX + 1];
 237         struct cls_bpf_prog *prog;
 238         int ret;
 239
 240         if (tca[TCA_OPTIONS] == NULL)
 241                 return -EINVAL;
 242
 243         ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
 244         if (ret < 0)
 245                 return ret;
 246
 247         prog = kzalloc(sizeof(*prog), GFP_KERNEL);
 248         if (!prog)
 249                 return -ENOBUFS;
 250
 251         tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
 252
 253         if (oldprog) {
 254                 if (handle && oldprog->handle != handle) {
 255                         ret = -EINVAL;
 256                         goto errout;
 257                 }
 258         }
 259
 260         if (handle == 0)
 261                 prog->handle = cls_bpf_grab_new_handle(tp, head);
 262         else
 263                 prog->handle = handle;
 264         if (prog->handle == 0) {
 265                 ret = -EINVAL;
 266                 goto errout;
 267         }
 268
 269         ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
 270         if (ret < 0)
 271                 goto errout;
 272
 273         if (oldprog) {
 274                 list_replace_rcu(&prog->link, &oldprog->link);
 275                 tcf_unbind_filter(tp, &oldprog->res);
 276                 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
 277         } else {
 278                 list_add_rcu(&prog->link, &head->plist);
 279         }
 280
 281         *arg = (unsigned long) prog;
 282         return 0;
 283 errout:
 284         kfree(prog);
 285
 286         return ret;
 287 }
 288
 289 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 290                         struct sk_buff *skb, struct tcmsg *tm)
 291 {
 292         struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
 293         struct nlattr *nest, *nla;
 294
 295         if (prog == NULL)
 296                 return skb->len;
 297
 298         tm->tcm_handle = prog->handle;
 299
 300         nest = nla_nest_start(skb, TCA_OPTIONS);
 301         if (nest == NULL)
 302                 goto nla_put_failure;
 303
 304         if (nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
 305                 goto nla_put_failure;
 306         if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_len))
 307                 goto nla_put_failure;
 308
 309         nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_len *
 310                           sizeof(struct sock_filter));
 311         if (nla == NULL)
 312                 goto nla_put_failure;
 313
 314         memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
 315
 316         if (tcf_exts_dump(skb, &prog->exts) < 0)
 317                 goto nla_put_failure;
 318
 319         nla_nest_end(skb, nest);
 320
 321         if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
 322                 goto nla_put_failure;
 323
 324         return skb->len;
 325
 326 nla_put_failure:
 327         nla_nest_cancel(skb, nest);
 328         return -1;
 329 }
 330
 331 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 332 {
 333         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 334         struct cls_bpf_prog *prog;
 335
 336         list_for_each_entry(prog, &head->plist, link) {
 337                 if (arg->count < arg->skip)
 338                         goto skip;
 339                 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
 340                         arg->stop = 1;
 341                         break;
 342                 }
 343 skip:
 344                 arg->count++;
 345         }
 346 }
 347
 348 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
 349         .kind           =       "bpf",
 350         .owner          =       THIS_MODULE,
 351         .classify       =       cls_bpf_classify,
 352         .init           =       cls_bpf_init,
 353         .destroy        =       cls_bpf_destroy,
 354         .get            =       cls_bpf_get,
 355         .change         =       cls_bpf_change,
 356         .delete         =       cls_bpf_delete,
 357         .walk           =       cls_bpf_walk,
 358         .dump           =       cls_bpf_dump,
 359 };
 360
 361 static int __init cls_bpf_init_mod(void)
 362 {
 363         return register_tcf_proto_ops(&cls_bpf_ops);
 364 }
 365
 366 static void __exit cls_bpf_exit_mod(void)
 367 {
 368         unregister_tcf_proto_ops(&cls_bpf_ops);
 369 }
 370
 371 module_init(cls_bpf_init_mod);
 372 module_exit(cls_bpf_exit_mod);