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         if (bpf_size != nla_len(tb[TCA_BPF_OPS])) {
 184                 ret = -EINVAL;
 185                 goto errout;
 186         }
 187
 188         bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
 189         if (bpf_ops == NULL) {
 190                 ret = -ENOMEM;
 191                 goto errout;
 192         }
 193
 194         memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
 195
 196         tmp.len = bpf_len;
 197         tmp.filter = bpf_ops;
 198
 199         ret = bpf_prog_create(&fp, &tmp);
 200         if (ret)
 201                 goto errout_free;
 202
 203         prog->bpf_len = bpf_len;
 204         prog->bpf_ops = bpf_ops;
 205         prog->filter = fp;
 206         prog->res.classid = classid;
 207
 208         tcf_bind_filter(tp, &prog->res, base);
 209         tcf_exts_change(tp, &prog->exts, &exts);
 210
 211         return 0;
 212 errout_free:
 213         kfree(bpf_ops);
 214 errout:
 215         tcf_exts_destroy(&exts);
 216         return ret;
 217 }
 218
 219 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
 220                                    struct cls_bpf_head *head)
 221 {
 222         unsigned int i = 0x80000000;
 223         u32 handle;
 224
 225         do {
 226                 if (++head->hgen == 0x7FFFFFFF)
 227                         head->hgen = 1;
 228         } while (--i > 0 && cls_bpf_get(tp, head->hgen));
 229
 230         if (unlikely(i == 0)) {
 231                 pr_err("Insufficient number of handles\n");
 232                 handle = 0;
 233         } else {
 234                 handle = head->hgen;
 235         }
 236
 237         return handle;
 238 }
 239
 240 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
 241                           struct tcf_proto *tp, unsigned long base,
 242                           u32 handle, struct nlattr **tca,
 243                           unsigned long *arg, bool ovr)
 244 {
 245         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 246         struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
 247         struct nlattr *tb[TCA_BPF_MAX + 1];
 248         struct cls_bpf_prog *prog;
 249         int ret;
 250
 251         if (tca[TCA_OPTIONS] == NULL)
 252                 return -EINVAL;
 253
 254         ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
 255         if (ret < 0)
 256                 return ret;
 257
 258         prog = kzalloc(sizeof(*prog), GFP_KERNEL);
 259         if (!prog)
 260                 return -ENOBUFS;
 261
 262         tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
 263
 264         if (oldprog) {
 265                 if (handle && oldprog->handle != handle) {
 266                         ret = -EINVAL;
 267                         goto errout;
 268                 }
 269         }
 270
 271         if (handle == 0)
 272                 prog->handle = cls_bpf_grab_new_handle(tp, head);
 273         else
 274                 prog->handle = handle;
 275         if (prog->handle == 0) {
 276                 ret = -EINVAL;
 277                 goto errout;
 278         }
 279
 280         ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
 281         if (ret < 0)
 282                 goto errout;
 283
 284         if (oldprog) {
 285                 list_replace_rcu(&prog->link, &oldprog->link);
 286                 tcf_unbind_filter(tp, &oldprog->res);
 287                 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
 288         } else {
 289                 list_add_rcu(&prog->link, &head->plist);
 290         }
 291
 292         *arg = (unsigned long) prog;
 293         return 0;
 294 errout:
 295         kfree(prog);
 296
 297         return ret;
 298 }
 299
 300 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 301                         struct sk_buff *skb, struct tcmsg *tm)
 302 {
 303         struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
 304         struct nlattr *nest, *nla;
 305
 306         if (prog == NULL)
 307                 return skb->len;
 308
 309         tm->tcm_handle = prog->handle;
 310
 311         nest = nla_nest_start(skb, TCA_OPTIONS);
 312         if (nest == NULL)
 313                 goto nla_put_failure;
 314
 315         if (nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
 316                 goto nla_put_failure;
 317         if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_len))
 318                 goto nla_put_failure;
 319
 320         nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_len *
 321                           sizeof(struct sock_filter));
 322         if (nla == NULL)
 323                 goto nla_put_failure;
 324
 325         memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
 326
 327         if (tcf_exts_dump(skb, &prog->exts) < 0)
 328                 goto nla_put_failure;
 329
 330         nla_nest_end(skb, nest);
 331
 332         if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
 333                 goto nla_put_failure;
 334
 335         return skb->len;
 336
 337 nla_put_failure:
 338         nla_nest_cancel(skb, nest);
 339         return -1;
 340 }
 341
 342 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 343 {
 344         struct cls_bpf_head *head = rtnl_dereference(tp->root);
 345         struct cls_bpf_prog *prog;
 346
 347         list_for_each_entry(prog, &head->plist, link) {
 348                 if (arg->count < arg->skip)
 349                         goto skip;
 350                 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
 351                         arg->stop = 1;
 352                         break;
 353                 }
 354 skip:
 355                 arg->count++;
 356         }
 357 }
 358
 359 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
 360         .kind           =       "bpf",
 361         .owner          =       THIS_MODULE,
 362         .classify       =       cls_bpf_classify,
 363         .init           =       cls_bpf_init,
 364         .destroy        =       cls_bpf_destroy,
 365         .get            =       cls_bpf_get,
 366         .change         =       cls_bpf_change,
 367         .delete         =       cls_bpf_delete,
 368         .walk           =       cls_bpf_walk,
 369         .dump           =       cls_bpf_dump,
 370 };
 371
 372 static int __init cls_bpf_init_mod(void)
 373 {
 374         return register_tcf_proto_ops(&cls_bpf_ops);
 375 }
 376
 377 static void __exit cls_bpf_exit_mod(void)
 378 {
 379         unregister_tcf_proto_ops(&cls_bpf_ops);
 380 }
 381
 382 module_init(cls_bpf_init_mod);
 383 module_exit(cls_bpf_exit_mod);