2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/skbuff.h>
24 #include <linux/openvswitch.h>
25 #include <linux/sctp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/in6.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_vlan.h>
33 #include <net/checksum.h>
34 #include <net/dsfield.h>
35 #include <net/sctp/checksum.h>
41 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
42 const struct nlattr *attr, int len);
44 static int make_writable(struct sk_buff *skb, int write_len)
46 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
49 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
52 /* remove VLAN header from packet and update csum accordingly. */
53 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
55 struct vlan_hdr *vhdr;
58 err = make_writable(skb, VLAN_ETH_HLEN);
62 if (skb->ip_summed == CHECKSUM_COMPLETE)
63 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
64 + (2 * ETH_ALEN), VLAN_HLEN, 0));
66 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
67 *current_tci = vhdr->h_vlan_TCI;
69 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
70 __skb_pull(skb, VLAN_HLEN);
72 vlan_set_encap_proto(skb, vhdr);
73 skb->mac_header += VLAN_HLEN;
74 skb_reset_mac_len(skb);
79 static int pop_vlan(struct sk_buff *skb)
84 if (likely(vlan_tx_tag_present(skb))) {
87 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
88 skb->len < VLAN_ETH_HLEN))
91 err = __pop_vlan_tci(skb, &tci);
95 /* move next vlan tag to hw accel tag */
96 if (likely(skb->protocol != htons(ETH_P_8021Q) ||
97 skb->len < VLAN_ETH_HLEN))
100 err = __pop_vlan_tci(skb, &tci);
104 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
108 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
110 if (unlikely(vlan_tx_tag_present(skb))) {
113 /* push down current VLAN tag */
114 current_tag = vlan_tx_tag_get(skb);
116 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
119 if (skb->ip_summed == CHECKSUM_COMPLETE)
120 skb->csum = csum_add(skb->csum, csum_partial(skb->data
121 + (2 * ETH_ALEN), VLAN_HLEN, 0));
124 __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
128 static int set_eth_addr(struct sk_buff *skb,
129 const struct ovs_key_ethernet *eth_key)
132 err = make_writable(skb, ETH_HLEN);
136 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
138 ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
139 ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
141 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
146 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
147 __be32 *addr, __be32 new_addr)
149 int transport_len = skb->len - skb_transport_offset(skb);
151 if (nh->protocol == IPPROTO_TCP) {
152 if (likely(transport_len >= sizeof(struct tcphdr)))
153 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
155 } else if (nh->protocol == IPPROTO_UDP) {
156 if (likely(transport_len >= sizeof(struct udphdr))) {
157 struct udphdr *uh = udp_hdr(skb);
159 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
160 inet_proto_csum_replace4(&uh->check, skb,
163 uh->check = CSUM_MANGLED_0;
168 csum_replace4(&nh->check, *addr, new_addr);
173 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
174 __be32 addr[4], const __be32 new_addr[4])
176 int transport_len = skb->len - skb_transport_offset(skb);
178 if (l4_proto == IPPROTO_TCP) {
179 if (likely(transport_len >= sizeof(struct tcphdr)))
180 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
182 } else if (l4_proto == IPPROTO_UDP) {
183 if (likely(transport_len >= sizeof(struct udphdr))) {
184 struct udphdr *uh = udp_hdr(skb);
186 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
187 inet_proto_csum_replace16(&uh->check, skb,
190 uh->check = CSUM_MANGLED_0;
196 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
197 __be32 addr[4], const __be32 new_addr[4],
198 bool recalculate_csum)
200 if (recalculate_csum)
201 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
204 memcpy(addr, new_addr, sizeof(__be32[4]));
207 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
209 nh->priority = tc >> 4;
210 nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
213 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
215 nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
216 nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
217 nh->flow_lbl[2] = fl & 0x000000FF;
220 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
222 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
226 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
231 err = make_writable(skb, skb_network_offset(skb) +
232 sizeof(struct iphdr));
238 if (ipv4_key->ipv4_src != nh->saddr)
239 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
241 if (ipv4_key->ipv4_dst != nh->daddr)
242 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
244 if (ipv4_key->ipv4_tos != nh->tos)
245 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
247 if (ipv4_key->ipv4_ttl != nh->ttl)
248 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
253 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
260 err = make_writable(skb, skb_network_offset(skb) +
261 sizeof(struct ipv6hdr));
266 saddr = (__be32 *)&nh->saddr;
267 daddr = (__be32 *)&nh->daddr;
269 if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
270 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
271 ipv6_key->ipv6_src, true);
273 if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
274 unsigned int offset = 0;
275 int flags = OVS_IP6T_FH_F_SKIP_RH;
276 bool recalc_csum = true;
278 if (ipv6_ext_hdr(nh->nexthdr))
279 recalc_csum = ipv6_find_hdr(skb, &offset,
280 NEXTHDR_ROUTING, NULL,
281 &flags) != NEXTHDR_ROUTING;
283 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
284 ipv6_key->ipv6_dst, recalc_csum);
287 set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
288 set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
289 nh->hop_limit = ipv6_key->ipv6_hlimit;
294 /* Must follow make_writable() since that can move the skb data. */
295 static void set_tp_port(struct sk_buff *skb, __be16 *port,
296 __be16 new_port, __sum16 *check)
298 inet_proto_csum_replace2(check, skb, *port, new_port, 0);
303 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
305 struct udphdr *uh = udp_hdr(skb);
307 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
308 set_tp_port(skb, port, new_port, &uh->check);
311 uh->check = CSUM_MANGLED_0;
318 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
323 err = make_writable(skb, skb_transport_offset(skb) +
324 sizeof(struct udphdr));
329 if (udp_port_key->udp_src != uh->source)
330 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
332 if (udp_port_key->udp_dst != uh->dest)
333 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
338 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
343 err = make_writable(skb, skb_transport_offset(skb) +
344 sizeof(struct tcphdr));
349 if (tcp_port_key->tcp_src != th->source)
350 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
352 if (tcp_port_key->tcp_dst != th->dest)
353 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
358 static int set_sctp(struct sk_buff *skb,
359 const struct ovs_key_sctp *sctp_port_key)
363 unsigned int sctphoff = skb_transport_offset(skb);
365 err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
370 if (sctp_port_key->sctp_src != sh->source ||
371 sctp_port_key->sctp_dst != sh->dest) {
372 __le32 old_correct_csum, new_csum, old_csum;
374 old_csum = sh->checksum;
375 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
377 sh->source = sctp_port_key->sctp_src;
378 sh->dest = sctp_port_key->sctp_dst;
380 new_csum = sctp_compute_cksum(skb, sctphoff);
382 /* Carry any checksum errors through. */
383 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
391 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
398 vport = ovs_vport_rcu(dp, out_port);
399 if (unlikely(!vport)) {
404 ovs_vport_send(vport, skb);
408 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
409 const struct nlattr *attr)
411 struct dp_upcall_info upcall;
412 const struct nlattr *a;
415 BUG_ON(!OVS_CB(skb)->pkt_key);
417 upcall.cmd = OVS_PACKET_CMD_ACTION;
418 upcall.key = OVS_CB(skb)->pkt_key;
419 upcall.userdata = NULL;
422 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
423 a = nla_next(a, &rem)) {
424 switch (nla_type(a)) {
425 case OVS_USERSPACE_ATTR_USERDATA:
429 case OVS_USERSPACE_ATTR_PID:
430 upcall.portid = nla_get_u32(a);
435 return ovs_dp_upcall(dp, skb, &upcall);
438 static bool last_action(const struct nlattr *a, int rem)
440 return a->nla_len == rem;
443 static int sample(struct datapath *dp, struct sk_buff *skb,
444 const struct nlattr *attr)
446 const struct nlattr *acts_list = NULL;
447 const struct nlattr *a;
448 struct sk_buff *sample_skb;
451 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
452 a = nla_next(a, &rem)) {
453 switch (nla_type(a)) {
454 case OVS_SAMPLE_ATTR_PROBABILITY:
455 if (prandom_u32() >= nla_get_u32(a))
459 case OVS_SAMPLE_ATTR_ACTIONS:
465 rem = nla_len(acts_list);
466 a = nla_data(acts_list);
468 /* Actions list is either empty or only contains a single user-space
469 * action, the latter being a special case as it is the only known
470 * usage of the sample action.
471 * In these special cases don't clone the skb as there are no
472 * side-effects in the nested actions.
473 * Otherwise, clone in case the nested actions have side effects. */
474 if (likely(rem == 0 ||
475 (nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
476 last_action(a, rem)))) {
480 sample_skb = skb_clone(skb, GFP_ATOMIC);
483 /* Note that do_execute_actions() never consumes skb.
484 * In the case where skb has been cloned above it is the clone that
485 * is consumed. Otherwise the skb_get(skb) call prevents
486 * consumption by do_execute_actions(). Thus, it is safe to simply
487 * return the error code and let the caller (also
488 * do_execute_actions()) free skb on error. */
489 return do_execute_actions(dp, sample_skb, a, rem);
492 static void execute_hash(struct sk_buff *skb, const struct nlattr *attr)
494 struct sw_flow_key *key = OVS_CB(skb)->pkt_key;
495 struct ovs_action_hash *hash_act = nla_data(attr);
498 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
499 hash = skb_get_hash(skb);
500 hash = jhash_1word(hash, hash_act->hash_basis);
504 key->ovs_flow_hash = hash;
507 static int execute_set_action(struct sk_buff *skb,
508 const struct nlattr *nested_attr)
512 switch (nla_type(nested_attr)) {
513 case OVS_KEY_ATTR_PRIORITY:
514 skb->priority = nla_get_u32(nested_attr);
517 case OVS_KEY_ATTR_SKB_MARK:
518 skb->mark = nla_get_u32(nested_attr);
521 case OVS_KEY_ATTR_IPV4_TUNNEL:
522 OVS_CB(skb)->tun_key = nla_data(nested_attr);
525 case OVS_KEY_ATTR_ETHERNET:
526 err = set_eth_addr(skb, nla_data(nested_attr));
529 case OVS_KEY_ATTR_IPV4:
530 err = set_ipv4(skb, nla_data(nested_attr));
533 case OVS_KEY_ATTR_IPV6:
534 err = set_ipv6(skb, nla_data(nested_attr));
537 case OVS_KEY_ATTR_TCP:
538 err = set_tcp(skb, nla_data(nested_attr));
541 case OVS_KEY_ATTR_UDP:
542 err = set_udp(skb, nla_data(nested_attr));
545 case OVS_KEY_ATTR_SCTP:
546 err = set_sctp(skb, nla_data(nested_attr));
553 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
554 const struct nlattr *a)
556 struct sw_flow_key recirc_key;
557 const struct vport *p = OVS_CB(skb)->input_vport;
558 uint32_t hash = OVS_CB(skb)->pkt_key->ovs_flow_hash;
561 err = ovs_flow_extract(skb, p->port_no, &recirc_key);
567 recirc_key.ovs_flow_hash = hash;
568 recirc_key.recirc_id = nla_get_u32(a);
570 ovs_dp_process_packet_with_key(skb, &recirc_key, true);
575 /* Execute a list of actions against 'skb'. */
576 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
577 const struct nlattr *attr, int len)
579 /* Every output action needs a separate clone of 'skb', but the common
580 * case is just a single output action, so that doing a clone and
581 * then freeing the original skbuff is wasteful. So the following code
582 * is slightly obscure just to avoid that. */
584 const struct nlattr *a;
587 for (a = attr, rem = len; rem > 0;
588 a = nla_next(a, &rem)) {
591 if (prev_port != -1) {
592 do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
596 switch (nla_type(a)) {
597 case OVS_ACTION_ATTR_OUTPUT:
598 prev_port = nla_get_u32(a);
601 case OVS_ACTION_ATTR_USERSPACE:
602 output_userspace(dp, skb, a);
605 case OVS_ACTION_ATTR_HASH:
606 execute_hash(skb, a);
609 case OVS_ACTION_ATTR_PUSH_VLAN:
610 err = push_vlan(skb, nla_data(a));
611 if (unlikely(err)) /* skb already freed. */
615 case OVS_ACTION_ATTR_POP_VLAN:
619 case OVS_ACTION_ATTR_RECIRC: {
620 struct sk_buff *recirc_skb;
622 if (!last_action(a, rem))
623 recirc_skb = skb_clone(skb, GFP_ATOMIC);
627 err = execute_recirc(dp, recirc_skb, a);
629 if (recirc_skb == skb)
635 case OVS_ACTION_ATTR_SET:
636 err = execute_set_action(skb, nla_data(a));
639 case OVS_ACTION_ATTR_SAMPLE:
640 err = sample(dp, skb, a);
651 do_output(dp, skb, prev_port);
658 /* We limit the number of times that we pass into execute_actions()
659 * to avoid blowing out the stack in the event that we have a loop.
661 * Each loop adds some (estimated) cost to the kernel stack.
662 * The loop terminates when the max cost is exceeded.
664 #define RECIRC_STACK_COST 1
665 #define DEFAULT_STACK_COST 4
666 /* Allow up to 4 regular services, and up to 3 recirculations */
667 #define MAX_STACK_COST (DEFAULT_STACK_COST * 4 + RECIRC_STACK_COST * 3)
669 struct loop_counter {
670 u8 stack_cost; /* loop stack cost. */
671 bool looping; /* Loop detected? */
674 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
676 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
679 pr_warn("%s: flow loop detected, dropping\n",
681 actions->actions_len = 0;
685 /* Execute a list of actions against 'skb'. */
686 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, bool recirc)
688 struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
689 const u8 stack_cost = recirc ? RECIRC_STACK_COST : DEFAULT_STACK_COST;
690 struct loop_counter *loop;
693 /* Check whether we've looped too much. */
694 loop = &__get_cpu_var(loop_counters);
695 loop->stack_cost += stack_cost;
696 if (unlikely(loop->stack_cost > MAX_STACK_COST))
697 loop->looping = true;
698 if (unlikely(loop->looping)) {
699 error = loop_suppress(dp, acts);
704 OVS_CB(skb)->tun_key = NULL;
705 error = do_execute_actions(dp, skb, acts->actions, acts->actions_len);
707 /* Check whether sub-actions looped too much. */
708 if (unlikely(loop->looping))
709 error = loop_suppress(dp, acts);
712 /* Decrement loop stack cost. */
713 loop->stack_cost -= stack_cost;
714 if (!loop->stack_cost)
715 loop->looping = false;