net/core/lwt_bpf.c

   1 /* Copyright (c) 2016 Thomas Graf <tgraf@tgraf.ch>
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  *
   7  * This program is distributed in the hope that it will be useful, but
   8  * WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  10  * General Public License for more details.
  11  */
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/types.h>
  17 #include <linux/bpf.h>
  18 #include <net/lwtunnel.h>
  19
  20 struct bpf_lwt_prog {
  21         struct bpf_prog *prog;
  22         char *name;
  23 };
  24
  25 struct bpf_lwt {
  26         struct bpf_lwt_prog in;
  27         struct bpf_lwt_prog out;
  28         struct bpf_lwt_prog xmit;
  29         int family;
  30 };
  31
  32 #define MAX_PROG_NAME 256
  33
  34 static inline struct bpf_lwt *bpf_lwt_lwtunnel(struct lwtunnel_state *lwt)
  35 {
  36         return (struct bpf_lwt *)lwt->data;
  37 }
  38
  39 #define NO_REDIRECT false
  40 #define CAN_REDIRECT true
  41
  42 static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
  43                        struct dst_entry *dst, bool can_redirect)
  44 {
  45         int ret;
  46
  47         /* Preempt disable is needed to protect per-cpu redirect_info between
  48          * BPF prog and skb_do_redirect(). The call_rcu in bpf_prog_put() and
  49          * access to maps strictly require a rcu_read_lock() for protection,
  50          * mixing with BH RCU lock doesn't work.
  51          */
  52         preempt_disable();
  53         rcu_read_lock();
  54         bpf_compute_data_end(skb);
  55         ret = bpf_prog_run_save_cb(lwt->prog, skb);
  56         rcu_read_unlock();
  57
  58         switch (ret) {
  59         case BPF_OK:
  60                 break;
  61
  62         case BPF_REDIRECT:
  63                 if (unlikely(!can_redirect)) {
  64                         pr_warn_once("Illegal redirect return code in prog %s\n",
  65                                      lwt->name ? : "<unknown>");
  66                         ret = BPF_OK;
  67                 } else {
  68                         ret = skb_do_redirect(skb);
  69                         if (ret == 0)
  70                                 ret = BPF_REDIRECT;
  71                 }
  72                 break;
  73
  74         case BPF_DROP:
  75                 kfree_skb(skb);
  76                 ret = -EPERM;
  77                 break;
  78
  79         default:
  80                 pr_warn_once("bpf-lwt: Illegal return value %u, expect packet loss\n", ret);
  81                 kfree_skb(skb);
  82                 ret = -EINVAL;
  83                 break;
  84         }
  85
  86         preempt_enable();
  87
  88         return ret;
  89 }
  90
  91 static int bpf_input(struct sk_buff *skb)
  92 {
  93         struct dst_entry *dst = skb_dst(skb);
  94         struct bpf_lwt *bpf;
  95         int ret;
  96
  97         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
  98         if (bpf->in.prog) {
  99                 ret = run_lwt_bpf(skb, &bpf->in, dst, NO_REDIRECT);
 100                 if (ret < 0)
 101                         return ret;
 102         }
 103
 104         if (unlikely(!dst->lwtstate->orig_input)) {
 105                 pr_warn_once("orig_input not set on dst for prog %s\n",
 106                              bpf->out.name);
 107                 kfree_skb(skb);
 108                 return -EINVAL;
 109         }
 110
 111         return dst->lwtstate->orig_input(skb);
 112 }
 113
 114 static int bpf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 115 {
 116         struct dst_entry *dst = skb_dst(skb);
 117         struct bpf_lwt *bpf;
 118         int ret;
 119
 120         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 121         if (bpf->out.prog) {
 122                 ret = run_lwt_bpf(skb, &bpf->out, dst, NO_REDIRECT);
 123                 if (ret < 0)
 124                         return ret;
 125         }
 126
 127         if (unlikely(!dst->lwtstate->orig_output)) {
 128                 pr_warn_once("orig_output not set on dst for prog %s\n",
 129                              bpf->out.name);
 130                 kfree_skb(skb);
 131                 return -EINVAL;
 132         }
 133
 134         return dst->lwtstate->orig_output(net, sk, skb);
 135 }
 136
 137 static int xmit_check_hhlen(struct sk_buff *skb)
 138 {
 139         int hh_len = skb_dst(skb)->dev->hard_header_len;
 140
 141         if (skb_headroom(skb) < hh_len) {
 142                 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
 143
 144                 if (pskb_expand_head(skb, nhead, 0, GFP_ATOMIC))
 145                         return -ENOMEM;
 146         }
 147
 148         return 0;
 149 }
 150
 151 static int bpf_xmit(struct sk_buff *skb)
 152 {
 153         struct dst_entry *dst = skb_dst(skb);
 154         struct bpf_lwt *bpf;
 155
 156         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 157         if (bpf->xmit.prog) {
 158                 int ret;
 159
 160                 ret = run_lwt_bpf(skb, &bpf->xmit, dst, CAN_REDIRECT);
 161                 switch (ret) {
 162                 case BPF_OK:
 163                         /* If the header was expanded, headroom might be too
 164                          * small for L2 header to come, expand as needed.
 165                          */
 166                         ret = xmit_check_hhlen(skb);
 167                         if (unlikely(ret))
 168                                 return ret;
 169
 170                         return LWTUNNEL_XMIT_CONTINUE;
 171                 case BPF_REDIRECT:
 172                         return LWTUNNEL_XMIT_DONE;
 173                 default:
 174                         return ret;
 175                 }
 176         }
 177
 178         return LWTUNNEL_XMIT_CONTINUE;
 179 }
 180
 181 static void bpf_lwt_prog_destroy(struct bpf_lwt_prog *prog)
 182 {
 183         if (prog->prog)
 184                 bpf_prog_put(prog->prog);
 185
 186         kfree(prog->name);
 187 }
 188
 189 static void bpf_destroy_state(struct lwtunnel_state *lwt)
 190 {
 191         struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
 192
 193         bpf_lwt_prog_destroy(&bpf->in);
 194         bpf_lwt_prog_destroy(&bpf->out);
 195         bpf_lwt_prog_destroy(&bpf->xmit);
 196 }
 197
 198 static const struct nla_policy bpf_prog_policy[LWT_BPF_PROG_MAX + 1] = {
 199         [LWT_BPF_PROG_FD]   = { .type = NLA_U32, },
 200         [LWT_BPF_PROG_NAME] = { .type = NLA_NUL_STRING,
 201                                 .len = MAX_PROG_NAME },
 202 };
 203
 204 static int bpf_parse_prog(struct nlattr *attr, struct bpf_lwt_prog *prog,
 205                           enum bpf_prog_type type)
 206 {
 207         struct nlattr *tb[LWT_BPF_PROG_MAX + 1];
 208         struct bpf_prog *p;
 209         int ret;
 210         u32 fd;
 211
 212         ret = nla_parse_nested(tb, LWT_BPF_PROG_MAX, attr, bpf_prog_policy,
 213                                NULL);
 214         if (ret < 0)
 215                 return ret;
 216
 217         if (!tb[LWT_BPF_PROG_FD] || !tb[LWT_BPF_PROG_NAME])
 218                 return -EINVAL;
 219
 220         prog->name = nla_memdup(tb[LWT_BPF_PROG_NAME], GFP_KERNEL);
 221         if (!prog->name)
 222                 return -ENOMEM;
 223
 224         fd = nla_get_u32(tb[LWT_BPF_PROG_FD]);
 225         p = bpf_prog_get_type(fd, type);
 226         if (IS_ERR(p))
 227                 return PTR_ERR(p);
 228
 229         prog->prog = p;
 230
 231         return 0;
 232 }
 233
 234 static const struct nla_policy bpf_nl_policy[LWT_BPF_MAX + 1] = {
 235         [LWT_BPF_IN]            = { .type = NLA_NESTED, },
 236         [LWT_BPF_OUT]           = { .type = NLA_NESTED, },
 237         [LWT_BPF_XMIT]          = { .type = NLA_NESTED, },
 238         [LWT_BPF_XMIT_HEADROOM] = { .type = NLA_U32 },
 239 };
 240
 241 static int bpf_build_state(struct nlattr *nla,
 242                            unsigned int family, const void *cfg,
 243                            struct lwtunnel_state **ts)
 244 {
 245         struct nlattr *tb[LWT_BPF_MAX + 1];
 246         struct lwtunnel_state *newts;
 247         struct bpf_lwt *bpf;
 248         int ret;
 249
 250         if (family != AF_INET && family != AF_INET6)
 251                 return -EAFNOSUPPORT;
 252
 253         ret = nla_parse_nested(tb, LWT_BPF_MAX, nla, bpf_nl_policy, NULL);
 254         if (ret < 0)
 255                 return ret;
 256
 257         if (!tb[LWT_BPF_IN] && !tb[LWT_BPF_OUT] && !tb[LWT_BPF_XMIT])
 258                 return -EINVAL;
 259
 260         newts = lwtunnel_state_alloc(sizeof(*bpf));
 261         if (!newts)
 262                 return -ENOMEM;
 263
 264         newts->type = LWTUNNEL_ENCAP_BPF;
 265         bpf = bpf_lwt_lwtunnel(newts);
 266
 267         if (tb[LWT_BPF_IN]) {
 268                 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
 269                 ret = bpf_parse_prog(tb[LWT_BPF_IN], &bpf->in,
 270                                      BPF_PROG_TYPE_LWT_IN);
 271                 if (ret  < 0)
 272                         goto errout;
 273         }
 274
 275         if (tb[LWT_BPF_OUT]) {
 276                 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
 277                 ret = bpf_parse_prog(tb[LWT_BPF_OUT], &bpf->out,
 278                                      BPF_PROG_TYPE_LWT_OUT);
 279                 if (ret < 0)
 280                         goto errout;
 281         }
 282
 283         if (tb[LWT_BPF_XMIT]) {
 284                 newts->flags |= LWTUNNEL_STATE_XMIT_REDIRECT;
 285                 ret = bpf_parse_prog(tb[LWT_BPF_XMIT], &bpf->xmit,
 286                                      BPF_PROG_TYPE_LWT_XMIT);
 287                 if (ret < 0)
 288                         goto errout;
 289         }
 290
 291         if (tb[LWT_BPF_XMIT_HEADROOM]) {
 292                 u32 headroom = nla_get_u32(tb[LWT_BPF_XMIT_HEADROOM]);
 293
 294                 if (headroom > LWT_BPF_MAX_HEADROOM) {
 295                         ret = -ERANGE;
 296                         goto errout;
 297                 }
 298
 299                 newts->headroom = headroom;
 300         }
 301
 302         bpf->family = family;
 303         *ts = newts;
 304
 305         return 0;
 306
 307 errout:
 308         bpf_destroy_state(newts);
 309         kfree(newts);
 310         return ret;
 311 }
 312
 313 static int bpf_fill_lwt_prog(struct sk_buff *skb, int attr,
 314                              struct bpf_lwt_prog *prog)
 315 {
 316         struct nlattr *nest;
 317
 318         if (!prog->prog)
 319                 return 0;
 320
 321         nest = nla_nest_start(skb, attr);
 322         if (!nest)
 323                 return -EMSGSIZE;
 324
 325         if (prog->name &&
 326             nla_put_string(skb, LWT_BPF_PROG_NAME, prog->name))
 327                 return -EMSGSIZE;
 328
 329         return nla_nest_end(skb, nest);
 330 }
 331
 332 static int bpf_fill_encap_info(struct sk_buff *skb, struct lwtunnel_state *lwt)
 333 {
 334         struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
 335
 336         if (bpf_fill_lwt_prog(skb, LWT_BPF_IN, &bpf->in) < 0 ||
 337             bpf_fill_lwt_prog(skb, LWT_BPF_OUT, &bpf->out) < 0 ||
 338             bpf_fill_lwt_prog(skb, LWT_BPF_XMIT, &bpf->xmit) < 0)
 339                 return -EMSGSIZE;
 340
 341         return 0;
 342 }
 343
 344 static int bpf_encap_nlsize(struct lwtunnel_state *lwtstate)
 345 {
 346         int nest_len = nla_total_size(sizeof(struct nlattr)) +
 347                        nla_total_size(MAX_PROG_NAME) + /* LWT_BPF_PROG_NAME */
 348                        0;
 349
 350         return nest_len + /* LWT_BPF_IN */
 351                nest_len + /* LWT_BPF_OUT */
 352                nest_len + /* LWT_BPF_XMIT */
 353                0;
 354 }
 355
 356 static int bpf_lwt_prog_cmp(struct bpf_lwt_prog *a, struct bpf_lwt_prog *b)
 357 {
 358         /* FIXME:
 359          * The LWT state is currently rebuilt for delete requests which
 360          * results in a new bpf_prog instance. Comparing names for now.
 361          */
 362         if (!a->name && !b->name)
 363                 return 0;
 364
 365         if (!a->name || !b->name)
 366                 return 1;
 367
 368         return strcmp(a->name, b->name);
 369 }
 370
 371 static int bpf_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
 372 {
 373         struct bpf_lwt *a_bpf = bpf_lwt_lwtunnel(a);
 374         struct bpf_lwt *b_bpf = bpf_lwt_lwtunnel(b);
 375
 376         return bpf_lwt_prog_cmp(&a_bpf->in, &b_bpf->in) ||
 377                bpf_lwt_prog_cmp(&a_bpf->out, &b_bpf->out) ||
 378                bpf_lwt_prog_cmp(&a_bpf->xmit, &b_bpf->xmit);
 379 }
 380
 381 static const struct lwtunnel_encap_ops bpf_encap_ops = {
 382         .build_state    = bpf_build_state,
 383         .destroy_state  = bpf_destroy_state,
 384         .input          = bpf_input,
 385         .output         = bpf_output,
 386         .xmit           = bpf_xmit,
 387         .fill_encap     = bpf_fill_encap_info,
 388         .get_encap_size = bpf_encap_nlsize,
 389         .cmp_encap      = bpf_encap_cmp,
 390         .owner          = THIS_MODULE,
 391 };
 392
 393 static int __init bpf_lwt_init(void)
 394 {
 395         return lwtunnel_encap_add_ops(&bpf_encap_ops, LWTUNNEL_ENCAP_BPF);
 396 }
 397
 398 subsys_initcall(bpf_lwt_init)