1 /* Copyright (c) 2016 Thomas Graf <tgraf@tgraf.ch>
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.
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.
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>
20 #include <net/ip6_route.h>
23 struct bpf_prog *prog;
28 struct bpf_lwt_prog in;
29 struct bpf_lwt_prog out;
30 struct bpf_lwt_prog xmit;
34 #define MAX_PROG_NAME 256
36 static inline struct bpf_lwt *bpf_lwt_lwtunnel(struct lwtunnel_state *lwt)
38 return (struct bpf_lwt *)lwt->data;
41 #define NO_REDIRECT false
42 #define CAN_REDIRECT true
44 static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
45 struct dst_entry *dst, bool can_redirect)
49 /* Preempt disable is needed to protect per-cpu redirect_info between
50 * BPF prog and skb_do_redirect(). The call_rcu in bpf_prog_put() and
51 * access to maps strictly require a rcu_read_lock() for protection,
52 * mixing with BH RCU lock doesn't work.
55 bpf_compute_data_pointers(skb);
56 ret = bpf_prog_run_save_cb(lwt->prog, skb);
64 if (unlikely(!can_redirect)) {
65 pr_warn_once("Illegal redirect return code in prog %s\n",
66 lwt->name ? : "<unknown>");
69 skb_reset_mac_header(skb);
70 ret = skb_do_redirect(skb);
82 pr_warn_once("bpf-lwt: Illegal return value %u, expect packet loss\n", ret);
93 static int bpf_lwt_input_reroute(struct sk_buff *skb)
97 if (skb->protocol == htons(ETH_P_IP)) {
98 struct iphdr *iph = ip_hdr(skb);
100 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
101 iph->tos, skb_dst(skb)->dev);
102 } else if (skb->protocol == htons(ETH_P_IPV6)) {
103 err = ipv6_stub->ipv6_route_input(skb);
110 return dst_input(skb);
117 static int bpf_input(struct sk_buff *skb)
119 struct dst_entry *dst = skb_dst(skb);
123 bpf = bpf_lwt_lwtunnel(dst->lwtstate);
125 ret = run_lwt_bpf(skb, &bpf->in, dst, NO_REDIRECT);
128 if (ret == BPF_LWT_REROUTE)
129 return bpf_lwt_input_reroute(skb);
132 if (unlikely(!dst->lwtstate->orig_input)) {
137 return dst->lwtstate->orig_input(skb);
140 static int bpf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
142 struct dst_entry *dst = skb_dst(skb);
146 bpf = bpf_lwt_lwtunnel(dst->lwtstate);
148 ret = run_lwt_bpf(skb, &bpf->out, dst, NO_REDIRECT);
153 if (unlikely(!dst->lwtstate->orig_output)) {
154 pr_warn_once("orig_output not set on dst for prog %s\n",
160 return dst->lwtstate->orig_output(net, sk, skb);
163 static int xmit_check_hhlen(struct sk_buff *skb)
165 int hh_len = skb_dst(skb)->dev->hard_header_len;
167 if (skb_headroom(skb) < hh_len) {
168 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
170 if (pskb_expand_head(skb, nhead, 0, GFP_ATOMIC))
177 static int bpf_lwt_xmit_reroute(struct sk_buff *skb)
179 struct net_device *l3mdev = l3mdev_master_dev_rcu(skb_dst(skb)->dev);
180 int oif = l3mdev ? l3mdev->ifindex : 0;
181 struct dst_entry *dst = NULL;
182 int err = -EAFNOSUPPORT;
187 if (skb->protocol == htons(ETH_P_IP))
189 else if (skb->protocol == htons(ETH_P_IPV6))
194 sk = sk_to_full_sk(skb->sk);
196 if (sk->sk_bound_dev_if)
197 oif = sk->sk_bound_dev_if;
200 net = dev_net(skb_dst(skb)->dev);
204 struct iphdr *iph = ip_hdr(skb);
205 struct flowi4 fl4 = {};
208 fl4.flowi4_oif = oif;
209 fl4.flowi4_mark = skb->mark;
210 fl4.flowi4_uid = sock_net_uid(net, sk);
211 fl4.flowi4_tos = RT_TOS(iph->tos);
212 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
213 fl4.flowi4_proto = iph->protocol;
214 fl4.daddr = iph->daddr;
215 fl4.saddr = iph->saddr;
217 rt = ip_route_output_key(net, &fl4);
224 struct ipv6hdr *iph6 = ipv6_hdr(skb);
225 struct flowi6 fl6 = {};
227 fl6.flowi6_oif = oif;
228 fl6.flowi6_mark = skb->mark;
229 fl6.flowi6_uid = sock_net_uid(net, sk);
230 fl6.flowlabel = ip6_flowinfo(iph6);
231 fl6.flowi6_proto = iph6->nexthdr;
232 fl6.daddr = iph6->daddr;
233 fl6.saddr = iph6->saddr;
235 err = ipv6_stub->ipv6_dst_lookup(net, skb->sk, &dst, &fl6);
243 if (unlikely(dst->error)) {
249 /* Although skb header was reserved in bpf_lwt_push_ip_encap(), it
250 * was done for the previous dst, so we are doing it here again, in
251 * case the new dst needs much more space. The call below is a noop
252 * if there is enough header space in skb.
254 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
259 skb_dst_set(skb, dst);
261 err = dst_output(dev_net(skb_dst(skb)->dev), skb->sk, skb);
265 /* ip[6]_finish_output2 understand LWTUNNEL_XMIT_DONE */
266 return LWTUNNEL_XMIT_DONE;
273 static int bpf_xmit(struct sk_buff *skb)
275 struct dst_entry *dst = skb_dst(skb);
278 bpf = bpf_lwt_lwtunnel(dst->lwtstate);
279 if (bpf->xmit.prog) {
280 __be16 proto = skb->protocol;
283 ret = run_lwt_bpf(skb, &bpf->xmit, dst, CAN_REDIRECT);
286 /* If the header changed, e.g. via bpf_lwt_push_encap,
287 * BPF_LWT_REROUTE below should have been used if the
288 * protocol was also changed.
290 if (skb->protocol != proto) {
294 /* If the header was expanded, headroom might be too
295 * small for L2 header to come, expand as needed.
297 ret = xmit_check_hhlen(skb);
301 return LWTUNNEL_XMIT_CONTINUE;
303 return LWTUNNEL_XMIT_DONE;
304 case BPF_LWT_REROUTE:
305 return bpf_lwt_xmit_reroute(skb);
311 return LWTUNNEL_XMIT_CONTINUE;
314 static void bpf_lwt_prog_destroy(struct bpf_lwt_prog *prog)
317 bpf_prog_put(prog->prog);
322 static void bpf_destroy_state(struct lwtunnel_state *lwt)
324 struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
326 bpf_lwt_prog_destroy(&bpf->in);
327 bpf_lwt_prog_destroy(&bpf->out);
328 bpf_lwt_prog_destroy(&bpf->xmit);
331 static const struct nla_policy bpf_prog_policy[LWT_BPF_PROG_MAX + 1] = {
332 [LWT_BPF_PROG_FD] = { .type = NLA_U32, },
333 [LWT_BPF_PROG_NAME] = { .type = NLA_NUL_STRING,
334 .len = MAX_PROG_NAME },
337 static int bpf_parse_prog(struct nlattr *attr, struct bpf_lwt_prog *prog,
338 enum bpf_prog_type type)
340 struct nlattr *tb[LWT_BPF_PROG_MAX + 1];
345 ret = nla_parse_nested(tb, LWT_BPF_PROG_MAX, attr, bpf_prog_policy,
350 if (!tb[LWT_BPF_PROG_FD] || !tb[LWT_BPF_PROG_NAME])
353 prog->name = nla_memdup(tb[LWT_BPF_PROG_NAME], GFP_ATOMIC);
357 fd = nla_get_u32(tb[LWT_BPF_PROG_FD]);
358 p = bpf_prog_get_type(fd, type);
367 static const struct nla_policy bpf_nl_policy[LWT_BPF_MAX + 1] = {
368 [LWT_BPF_IN] = { .type = NLA_NESTED, },
369 [LWT_BPF_OUT] = { .type = NLA_NESTED, },
370 [LWT_BPF_XMIT] = { .type = NLA_NESTED, },
371 [LWT_BPF_XMIT_HEADROOM] = { .type = NLA_U32 },
374 static int bpf_build_state(struct nlattr *nla,
375 unsigned int family, const void *cfg,
376 struct lwtunnel_state **ts,
377 struct netlink_ext_ack *extack)
379 struct nlattr *tb[LWT_BPF_MAX + 1];
380 struct lwtunnel_state *newts;
384 if (family != AF_INET && family != AF_INET6)
385 return -EAFNOSUPPORT;
387 ret = nla_parse_nested(tb, LWT_BPF_MAX, nla, bpf_nl_policy, extack);
391 if (!tb[LWT_BPF_IN] && !tb[LWT_BPF_OUT] && !tb[LWT_BPF_XMIT])
394 newts = lwtunnel_state_alloc(sizeof(*bpf));
398 newts->type = LWTUNNEL_ENCAP_BPF;
399 bpf = bpf_lwt_lwtunnel(newts);
401 if (tb[LWT_BPF_IN]) {
402 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
403 ret = bpf_parse_prog(tb[LWT_BPF_IN], &bpf->in,
404 BPF_PROG_TYPE_LWT_IN);
409 if (tb[LWT_BPF_OUT]) {
410 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
411 ret = bpf_parse_prog(tb[LWT_BPF_OUT], &bpf->out,
412 BPF_PROG_TYPE_LWT_OUT);
417 if (tb[LWT_BPF_XMIT]) {
418 newts->flags |= LWTUNNEL_STATE_XMIT_REDIRECT;
419 ret = bpf_parse_prog(tb[LWT_BPF_XMIT], &bpf->xmit,
420 BPF_PROG_TYPE_LWT_XMIT);
425 if (tb[LWT_BPF_XMIT_HEADROOM]) {
426 u32 headroom = nla_get_u32(tb[LWT_BPF_XMIT_HEADROOM]);
428 if (headroom > LWT_BPF_MAX_HEADROOM) {
433 newts->headroom = headroom;
436 bpf->family = family;
442 bpf_destroy_state(newts);
447 static int bpf_fill_lwt_prog(struct sk_buff *skb, int attr,
448 struct bpf_lwt_prog *prog)
455 nest = nla_nest_start(skb, attr);
460 nla_put_string(skb, LWT_BPF_PROG_NAME, prog->name))
463 return nla_nest_end(skb, nest);
466 static int bpf_fill_encap_info(struct sk_buff *skb, struct lwtunnel_state *lwt)
468 struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
470 if (bpf_fill_lwt_prog(skb, LWT_BPF_IN, &bpf->in) < 0 ||
471 bpf_fill_lwt_prog(skb, LWT_BPF_OUT, &bpf->out) < 0 ||
472 bpf_fill_lwt_prog(skb, LWT_BPF_XMIT, &bpf->xmit) < 0)
478 static int bpf_encap_nlsize(struct lwtunnel_state *lwtstate)
480 int nest_len = nla_total_size(sizeof(struct nlattr)) +
481 nla_total_size(MAX_PROG_NAME) + /* LWT_BPF_PROG_NAME */
484 return nest_len + /* LWT_BPF_IN */
485 nest_len + /* LWT_BPF_OUT */
486 nest_len + /* LWT_BPF_XMIT */
490 static int bpf_lwt_prog_cmp(struct bpf_lwt_prog *a, struct bpf_lwt_prog *b)
493 * The LWT state is currently rebuilt for delete requests which
494 * results in a new bpf_prog instance. Comparing names for now.
496 if (!a->name && !b->name)
499 if (!a->name || !b->name)
502 return strcmp(a->name, b->name);
505 static int bpf_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
507 struct bpf_lwt *a_bpf = bpf_lwt_lwtunnel(a);
508 struct bpf_lwt *b_bpf = bpf_lwt_lwtunnel(b);
510 return bpf_lwt_prog_cmp(&a_bpf->in, &b_bpf->in) ||
511 bpf_lwt_prog_cmp(&a_bpf->out, &b_bpf->out) ||
512 bpf_lwt_prog_cmp(&a_bpf->xmit, &b_bpf->xmit);
515 static const struct lwtunnel_encap_ops bpf_encap_ops = {
516 .build_state = bpf_build_state,
517 .destroy_state = bpf_destroy_state,
519 .output = bpf_output,
521 .fill_encap = bpf_fill_encap_info,
522 .get_encap_size = bpf_encap_nlsize,
523 .cmp_encap = bpf_encap_cmp,
524 .owner = THIS_MODULE,
527 static int handle_gso_type(struct sk_buff *skb, unsigned int gso_type,
530 struct skb_shared_info *shinfo = skb_shinfo(skb);
532 gso_type |= SKB_GSO_DODGY;
533 shinfo->gso_type |= gso_type;
534 skb_decrease_gso_size(shinfo, encap_len);
535 shinfo->gso_segs = 0;
539 static int handle_gso_encap(struct sk_buff *skb, bool ipv4, int encap_len)
545 /* SCTP and UDP_L4 gso need more nuanced handling than what
546 * handle_gso_type() does above: skb_decrease_gso_size() is not enough.
547 * So at the moment only TCP GSO packets are let through.
549 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
553 protocol = ip_hdr(skb)->protocol;
554 next_hdr_offset = sizeof(struct iphdr);
555 next_hdr = skb_network_header(skb) + next_hdr_offset;
557 protocol = ipv6_hdr(skb)->nexthdr;
558 next_hdr_offset = sizeof(struct ipv6hdr);
559 next_hdr = skb_network_header(skb) + next_hdr_offset;
564 next_hdr_offset += sizeof(struct gre_base_hdr);
565 if (next_hdr_offset > encap_len)
568 if (((struct gre_base_hdr *)next_hdr)->flags & GRE_CSUM)
569 return handle_gso_type(skb, SKB_GSO_GRE_CSUM,
571 return handle_gso_type(skb, SKB_GSO_GRE, encap_len);
574 next_hdr_offset += sizeof(struct udphdr);
575 if (next_hdr_offset > encap_len)
578 if (((struct udphdr *)next_hdr)->check)
579 return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL_CSUM,
581 return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL, encap_len);
586 return handle_gso_type(skb, SKB_GSO_IPXIP4, encap_len);
588 return handle_gso_type(skb, SKB_GSO_IPXIP6, encap_len);
591 return -EPROTONOSUPPORT;
595 int bpf_lwt_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, bool ingress)
601 if (unlikely(len < sizeof(struct iphdr) || len > LWT_BPF_MAX_HEADROOM))
604 /* validate protocol and length */
605 iph = (struct iphdr *)hdr;
606 if (iph->version == 4) {
608 if (unlikely(len < iph->ihl * 4))
610 } else if (iph->version == 6) {
612 if (unlikely(len < sizeof(struct ipv6hdr)))
619 err = skb_cow_head(skb, len + skb->mac_len);
621 err = skb_cow_head(skb,
622 len + LL_RESERVED_SPACE(skb_dst(skb)->dev));
626 /* push the encap headers and fix pointers */
627 skb_reset_inner_headers(skb);
628 skb_reset_inner_mac_header(skb); /* mac header is not yet set */
629 skb_set_inner_protocol(skb, skb->protocol);
630 skb->encapsulation = 1;
633 skb_postpush_rcsum(skb, iph, len);
634 skb_reset_network_header(skb);
635 memcpy(skb_network_header(skb), hdr, len);
636 bpf_compute_data_pointers(skb);
640 skb->protocol = htons(ETH_P_IP);
644 iph->check = ip_fast_csum((unsigned char *)iph,
647 skb->protocol = htons(ETH_P_IPV6);
651 return handle_gso_encap(skb, ipv4, len);
656 static int __init bpf_lwt_init(void)
658 return lwtunnel_encap_add_ops(&bpf_encap_ops, LWTUNNEL_ENCAP_BPF);
661 subsys_initcall(bpf_lwt_init)