1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Checksum updating actions
8 #include <linux/types.h>
9 #include <linux/init.h>
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/spinlock.h>
14 #include <linux/netlink.h>
15 #include <net/netlink.h>
16 #include <linux/rtnetlink.h>
18 #include <linux/skbuff.h>
23 #include <linux/icmpv6.h>
24 #include <linux/igmp.h>
27 #include <net/ip6_checksum.h>
28 #include <net/sctp/checksum.h>
30 #include <net/act_api.h>
31 #include <net/pkt_cls.h>
33 #include <linux/tc_act/tc_csum.h>
34 #include <net/tc_act/tc_csum.h>
36 static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
37 [TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
40 static unsigned int csum_net_id;
41 static struct tc_action_ops act_csum_ops;
43 static int tcf_csum_init(struct net *net, struct nlattr *nla,
44 struct nlattr *est, struct tc_action **a, int ovr,
45 int bind, bool rtnl_held, struct tcf_proto *tp,
46 struct netlink_ext_ack *extack)
48 struct tc_action_net *tn = net_generic(net, csum_net_id);
49 struct tcf_csum_params *params_new;
50 struct nlattr *tb[TCA_CSUM_MAX + 1];
51 struct tcf_chain *goto_ch = NULL;
60 err = nla_parse_nested_deprecated(tb, TCA_CSUM_MAX, nla, csum_policy,
65 if (tb[TCA_CSUM_PARMS] == NULL)
67 parm = nla_data(tb[TCA_CSUM_PARMS]);
69 err = tcf_idr_check_alloc(tn, &index, a, bind);
71 ret = tcf_idr_create(tn, index, est, a,
72 &act_csum_ops, bind, true);
74 tcf_idr_cleanup(tn, index);
79 if (bind)/* dont override defaults */
82 tcf_idr_release(*a, bind);
89 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
95 params_new = kzalloc(sizeof(*params_new), GFP_KERNEL);
96 if (unlikely(!params_new)) {
100 params_new->update_flags = parm->update_flags;
102 spin_lock_bh(&p->tcf_lock);
103 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
104 rcu_swap_protected(p->params, params_new,
105 lockdep_is_held(&p->tcf_lock));
106 spin_unlock_bh(&p->tcf_lock);
109 tcf_chain_put_by_act(goto_ch);
111 kfree_rcu(params_new, rcu);
113 if (ret == ACT_P_CREATED)
114 tcf_idr_insert(tn, *a);
119 tcf_chain_put_by_act(goto_ch);
121 tcf_idr_release(*a, bind);
126 * tcf_csum_skb_nextlayer - Get next layer pointer
127 * @skb: sk_buff to use
128 * @ihl: previous summed headers length
129 * @ipl: complete packet length
130 * @jhl: next header length
132 * Check the expected next layer availability in the specified sk_buff.
133 * Return the next layer pointer if pass, NULL otherwise.
135 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
136 unsigned int ihl, unsigned int ipl,
139 int ntkoff = skb_network_offset(skb);
142 if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
143 skb_try_make_writable(skb, hl + ntkoff))
146 return (void *)(skb_network_header(skb) + ihl);
149 static int tcf_csum_ipv4_icmp(struct sk_buff *skb, unsigned int ihl,
152 struct icmphdr *icmph;
154 icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
159 skb->csum = csum_partial(icmph, ipl - ihl, 0);
160 icmph->checksum = csum_fold(skb->csum);
162 skb->ip_summed = CHECKSUM_NONE;
167 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
168 unsigned int ihl, unsigned int ipl)
170 struct igmphdr *igmph;
172 igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
177 skb->csum = csum_partial(igmph, ipl - ihl, 0);
178 igmph->csum = csum_fold(skb->csum);
180 skb->ip_summed = CHECKSUM_NONE;
185 static int tcf_csum_ipv6_icmp(struct sk_buff *skb, unsigned int ihl,
188 struct icmp6hdr *icmp6h;
189 const struct ipv6hdr *ip6h;
191 icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
195 ip6h = ipv6_hdr(skb);
196 icmp6h->icmp6_cksum = 0;
197 skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
198 icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
199 ipl - ihl, IPPROTO_ICMPV6,
202 skb->ip_summed = CHECKSUM_NONE;
207 static int tcf_csum_ipv4_tcp(struct sk_buff *skb, unsigned int ihl,
211 const struct iphdr *iph;
213 if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
216 tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
222 skb->csum = csum_partial(tcph, ipl - ihl, 0);
223 tcph->check = tcp_v4_check(ipl - ihl,
224 iph->saddr, iph->daddr, skb->csum);
226 skb->ip_summed = CHECKSUM_NONE;
231 static int tcf_csum_ipv6_tcp(struct sk_buff *skb, unsigned int ihl,
235 const struct ipv6hdr *ip6h;
237 if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
240 tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
244 ip6h = ipv6_hdr(skb);
246 skb->csum = csum_partial(tcph, ipl - ihl, 0);
247 tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
248 ipl - ihl, IPPROTO_TCP,
251 skb->ip_summed = CHECKSUM_NONE;
256 static int tcf_csum_ipv4_udp(struct sk_buff *skb, unsigned int ihl,
257 unsigned int ipl, int udplite)
260 const struct iphdr *iph;
263 if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
267 * Support both UDP and UDPLITE checksum algorithms, Don't use
268 * udph->len to get the real length without any protocol check,
269 * UDPLITE uses udph->len for another thing,
270 * Use iph->tot_len, or just ipl.
273 udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
278 ul = ntohs(udph->len);
280 if (udplite || udph->check) {
286 skb->csum = csum_partial(udph, ipl - ihl, 0);
287 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
288 skb->csum = csum_partial(udph, ul, 0);
290 goto ignore_obscure_skb;
293 goto ignore_obscure_skb;
295 skb->csum = csum_partial(udph, ul, 0);
298 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
303 udph->check = CSUM_MANGLED_0;
306 skb->ip_summed = CHECKSUM_NONE;
312 static int tcf_csum_ipv6_udp(struct sk_buff *skb, unsigned int ihl,
313 unsigned int ipl, int udplite)
316 const struct ipv6hdr *ip6h;
319 if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
323 * Support both UDP and UDPLITE checksum algorithms, Don't use
324 * udph->len to get the real length without any protocol check,
325 * UDPLITE uses udph->len for another thing,
326 * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
329 udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
333 ip6h = ipv6_hdr(skb);
334 ul = ntohs(udph->len);
340 skb->csum = csum_partial(udph, ipl - ihl, 0);
342 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
343 skb->csum = csum_partial(udph, ul, 0);
346 goto ignore_obscure_skb;
349 goto ignore_obscure_skb;
351 skb->csum = csum_partial(udph, ul, 0);
354 udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
355 udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
359 udph->check = CSUM_MANGLED_0;
361 skb->ip_summed = CHECKSUM_NONE;
367 static int tcf_csum_sctp(struct sk_buff *skb, unsigned int ihl,
370 struct sctphdr *sctph;
372 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
375 sctph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*sctph));
379 sctph->checksum = sctp_compute_cksum(skb,
380 skb_network_offset(skb) + ihl);
381 skb->ip_summed = CHECKSUM_NONE;
382 skb->csum_not_inet = 0;
387 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
389 const struct iphdr *iph;
392 ntkoff = skb_network_offset(skb);
394 if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
399 switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
401 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
402 if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
403 ntohs(iph->tot_len)))
407 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
408 if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
409 ntohs(iph->tot_len)))
413 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
414 if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
415 ntohs(iph->tot_len)))
419 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
420 if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
421 ntohs(iph->tot_len), 0))
424 case IPPROTO_UDPLITE:
425 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
426 if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
427 ntohs(iph->tot_len), 1))
431 if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
432 !tcf_csum_sctp(skb, iph->ihl * 4, ntohs(iph->tot_len)))
437 if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
438 if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
441 ip_send_check(ip_hdr(skb));
450 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh, unsigned int ixhl,
453 int off, len, optlen;
454 unsigned char *xh = (void *)ip6xh;
456 off = sizeof(*ip6xh);
465 optlen = xh[off + 1] + 2;
466 if (optlen != 6 || len < 6 || (off & 3) != 2)
467 /* wrong jumbo option length/alignment */
469 *pl = ntohl(*(__be32 *)(xh + off + 2));
472 optlen = xh[off + 1] + 2;
474 /* ignore obscure options */
486 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
488 struct ipv6hdr *ip6h;
489 struct ipv6_opt_hdr *ip6xh;
490 unsigned int hl, ixhl;
495 ntkoff = skb_network_offset(skb);
499 if (!pskb_may_pull(skb, hl + ntkoff))
502 ip6h = ipv6_hdr(skb);
504 pl = ntohs(ip6h->payload_len);
505 nexthdr = ip6h->nexthdr;
509 case NEXTHDR_FRAGMENT:
511 case NEXTHDR_ROUTING:
514 if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
516 ip6xh = (void *)(skb_network_header(skb) + hl);
517 ixhl = ipv6_optlen(ip6xh);
518 if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
520 ip6xh = (void *)(skb_network_header(skb) + hl);
521 if ((nexthdr == NEXTHDR_HOP) &&
522 !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
524 nexthdr = ip6xh->nexthdr;
528 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
529 if (!tcf_csum_ipv6_icmp(skb,
530 hl, pl + sizeof(*ip6h)))
534 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
535 if (!tcf_csum_ipv6_tcp(skb,
536 hl, pl + sizeof(*ip6h)))
540 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
541 if (!tcf_csum_ipv6_udp(skb, hl,
542 pl + sizeof(*ip6h), 0))
545 case IPPROTO_UDPLITE:
546 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
547 if (!tcf_csum_ipv6_udp(skb, hl,
548 pl + sizeof(*ip6h), 1))
552 if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
553 !tcf_csum_sctp(skb, hl, pl + sizeof(*ip6h)))
559 } while (pskb_may_pull(skb, hl + 1 + ntkoff));
569 static int tcf_csum_act(struct sk_buff *skb, const struct tc_action *a,
570 struct tcf_result *res)
572 struct tcf_csum *p = to_tcf_csum(a);
573 bool orig_vlan_tag_present = false;
574 unsigned int vlan_hdr_count = 0;
575 struct tcf_csum_params *params;
580 params = rcu_dereference_bh(p->params);
582 tcf_lastuse_update(&p->tcf_tm);
583 bstats_cpu_update(this_cpu_ptr(p->common.cpu_bstats), skb);
585 action = READ_ONCE(p->tcf_action);
586 if (unlikely(action == TC_ACT_SHOT))
589 update_flags = params->update_flags;
590 protocol = tc_skb_protocol(skb);
593 case cpu_to_be16(ETH_P_IP):
594 if (!tcf_csum_ipv4(skb, update_flags))
597 case cpu_to_be16(ETH_P_IPV6):
598 if (!tcf_csum_ipv6(skb, update_flags))
601 case cpu_to_be16(ETH_P_8021AD): /* fall through */
602 case cpu_to_be16(ETH_P_8021Q):
603 if (skb_vlan_tag_present(skb) && !orig_vlan_tag_present) {
604 protocol = skb->protocol;
605 orig_vlan_tag_present = true;
607 struct vlan_hdr *vlan = (struct vlan_hdr *)skb->data;
609 protocol = vlan->h_vlan_encapsulated_proto;
610 skb_pull(skb, VLAN_HLEN);
611 skb_reset_network_header(skb);
618 /* Restore the skb for the pulled VLAN tags */
619 while (vlan_hdr_count--) {
620 skb_push(skb, VLAN_HLEN);
621 skb_reset_network_header(skb);
627 qstats_drop_inc(this_cpu_ptr(p->common.cpu_qstats));
628 action = TC_ACT_SHOT;
632 static int tcf_csum_dump(struct sk_buff *skb, struct tc_action *a, int bind,
635 unsigned char *b = skb_tail_pointer(skb);
636 struct tcf_csum *p = to_tcf_csum(a);
637 struct tcf_csum_params *params;
638 struct tc_csum opt = {
639 .index = p->tcf_index,
640 .refcnt = refcount_read(&p->tcf_refcnt) - ref,
641 .bindcnt = atomic_read(&p->tcf_bindcnt) - bind,
645 spin_lock_bh(&p->tcf_lock);
646 params = rcu_dereference_protected(p->params,
647 lockdep_is_held(&p->tcf_lock));
648 opt.action = p->tcf_action;
649 opt.update_flags = params->update_flags;
651 if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
652 goto nla_put_failure;
654 tcf_tm_dump(&t, &p->tcf_tm);
655 if (nla_put_64bit(skb, TCA_CSUM_TM, sizeof(t), &t, TCA_CSUM_PAD))
656 goto nla_put_failure;
657 spin_unlock_bh(&p->tcf_lock);
662 spin_unlock_bh(&p->tcf_lock);
667 static void tcf_csum_cleanup(struct tc_action *a)
669 struct tcf_csum *p = to_tcf_csum(a);
670 struct tcf_csum_params *params;
672 params = rcu_dereference_protected(p->params, 1);
674 kfree_rcu(params, rcu);
677 static int tcf_csum_walker(struct net *net, struct sk_buff *skb,
678 struct netlink_callback *cb, int type,
679 const struct tc_action_ops *ops,
680 struct netlink_ext_ack *extack)
682 struct tc_action_net *tn = net_generic(net, csum_net_id);
684 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
687 static int tcf_csum_search(struct net *net, struct tc_action **a, u32 index)
689 struct tc_action_net *tn = net_generic(net, csum_net_id);
691 return tcf_idr_search(tn, a, index);
694 static size_t tcf_csum_get_fill_size(const struct tc_action *act)
696 return nla_total_size(sizeof(struct tc_csum));
699 static struct tc_action_ops act_csum_ops = {
702 .owner = THIS_MODULE,
704 .dump = tcf_csum_dump,
705 .init = tcf_csum_init,
706 .cleanup = tcf_csum_cleanup,
707 .walk = tcf_csum_walker,
708 .lookup = tcf_csum_search,
709 .get_fill_size = tcf_csum_get_fill_size,
710 .size = sizeof(struct tcf_csum),
713 static __net_init int csum_init_net(struct net *net)
715 struct tc_action_net *tn = net_generic(net, csum_net_id);
717 return tc_action_net_init(net, tn, &act_csum_ops);
720 static void __net_exit csum_exit_net(struct list_head *net_list)
722 tc_action_net_exit(net_list, csum_net_id);
725 static struct pernet_operations csum_net_ops = {
726 .init = csum_init_net,
727 .exit_batch = csum_exit_net,
729 .size = sizeof(struct tc_action_net),
732 MODULE_DESCRIPTION("Checksum updating actions");
733 MODULE_LICENSE("GPL");
735 static int __init csum_init_module(void)
737 return tcf_register_action(&act_csum_ops, &csum_net_ops);
740 static void __exit csum_cleanup_module(void)
742 tcf_unregister_action(&act_csum_ops, &csum_net_ops);
745 module_init(csum_init_module);
746 module_exit(csum_cleanup_module);