Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (c) 2007-2014 Nicira, Inc.
3 *
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.
7 *
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.
12 *
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
16 * 02110-1301, USA
17 */
18
19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21#include <linux/skbuff.h>
22#include <linux/in.h>
23#include <linux/ip.h>
24#include <linux/openvswitch.h>
25#include <linux/netfilter_ipv6.h>
26#include <linux/sctp.h>
27#include <linux/tcp.h>
28#include <linux/udp.h>
29#include <linux/in6.h>
30#include <linux/if_arp.h>
31#include <linux/if_vlan.h>
32
33#include <net/dst.h>
34#include <net/ip.h>
35#include <net/ipv6.h>
36#include <net/ip6_fib.h>
37#include <net/checksum.h>
38#include <net/dsfield.h>
39#include <net/mpls.h>
40#include <net/sctp/checksum.h>
41
42#include "datapath.h"
43#include "flow.h"
44#include "conntrack.h"
45#include "vport.h"
46
47static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
48 struct sw_flow_key *key,
49 const struct nlattr *attr, int len);
50
51struct deferred_action {
52 struct sk_buff *skb;
53 const struct nlattr *actions;
54
55 /* Store pkt_key clone when creating deferred action. */
56 struct sw_flow_key pkt_key;
57};
58
59#define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
60struct ovs_frag_data {
61 unsigned long dst;
62 struct vport *vport;
63 struct ovs_skb_cb cb;
64 __be16 inner_protocol;
65 u16 network_offset; /* valid only for MPLS */
66 u16 vlan_tci;
67 __be16 vlan_proto;
68 unsigned int l2_len;
69 u8 mac_proto;
70 u8 l2_data[MAX_L2_LEN];
71};
72
73static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
74
75#define DEFERRED_ACTION_FIFO_SIZE 10
76#define OVS_RECURSION_LIMIT 5
77#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
78struct action_fifo {
79 int head;
80 int tail;
81 /* Deferred action fifo queue storage. */
82 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
83};
84
85struct recirc_keys {
86 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
87};
88
89static struct action_fifo __percpu *action_fifos;
90static struct recirc_keys __percpu *recirc_keys;
91static DEFINE_PER_CPU(int, exec_actions_level);
92
93static void action_fifo_init(struct action_fifo *fifo)
94{
95 fifo->head = 0;
96 fifo->tail = 0;
97}
98
99static bool action_fifo_is_empty(const struct action_fifo *fifo)
100{
101 return (fifo->head == fifo->tail);
102}
103
104static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
105{
106 if (action_fifo_is_empty(fifo))
107 return NULL;
108
109 return &fifo->fifo[fifo->tail++];
110}
111
112static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
113{
114 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
115 return NULL;
116
117 return &fifo->fifo[fifo->head++];
118}
119
120/* Return true if fifo is not full */
121static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
122 const struct sw_flow_key *key,
123 const struct nlattr *attr)
124{
125 struct action_fifo *fifo;
126 struct deferred_action *da;
127
128 fifo = this_cpu_ptr(action_fifos);
129 da = action_fifo_put(fifo);
130 if (da) {
131 da->skb = skb;
132 da->actions = attr;
133 da->pkt_key = *key;
134 }
135
136 return da;
137}
138
139static void invalidate_flow_key(struct sw_flow_key *key)
140{
141 key->mac_proto |= SW_FLOW_KEY_INVALID;
142}
143
144static bool is_flow_key_valid(const struct sw_flow_key *key)
145{
146 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
147}
148
149static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
150 __be16 ethertype)
151{
152 if (skb->ip_summed == CHECKSUM_COMPLETE) {
153 __be16 diff[] = { ~(hdr->h_proto), ethertype };
154
155 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
156 ~skb->csum);
157 }
158
159 hdr->h_proto = ethertype;
160}
161
162static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
163 const struct ovs_action_push_mpls *mpls)
164{
165 struct mpls_shim_hdr *new_mpls_lse;
166
167 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
168 if (skb->encapsulation)
169 return -ENOTSUPP;
170
171 if (skb_cow_head(skb, MPLS_HLEN) < 0)
172 return -ENOMEM;
173
174 if (!skb->inner_protocol) {
175 skb_set_inner_network_header(skb, skb->mac_len);
176 skb_set_inner_protocol(skb, skb->protocol);
177 }
178
179 skb_push(skb, MPLS_HLEN);
180 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
181 skb->mac_len);
182 skb_reset_mac_header(skb);
183 skb_set_network_header(skb, skb->mac_len);
184
185 new_mpls_lse = mpls_hdr(skb);
186 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
187
188 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
189
190 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
191 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
192 skb->protocol = mpls->mpls_ethertype;
193
194 invalidate_flow_key(key);
195 return 0;
196}
197
198static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
199 const __be16 ethertype)
200{
201 int err;
202
203 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
204 if (unlikely(err))
205 return err;
206
207 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
208
209 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
210 skb->mac_len);
211
212 __skb_pull(skb, MPLS_HLEN);
213 skb_reset_mac_header(skb);
214 skb_set_network_header(skb, skb->mac_len);
215
216 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
217 struct ethhdr *hdr;
218
219 /* mpls_hdr() is used to locate the ethertype field correctly in the
220 * presence of VLAN tags.
221 */
222 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
223 update_ethertype(skb, hdr, ethertype);
224 }
225 if (eth_p_mpls(skb->protocol))
226 skb->protocol = ethertype;
227
228 invalidate_flow_key(key);
229 return 0;
230}
231
232static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
233 const __be32 *mpls_lse, const __be32 *mask)
234{
235 struct mpls_shim_hdr *stack;
236 __be32 lse;
237 int err;
238
239 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
240 if (unlikely(err))
241 return err;
242
243 stack = mpls_hdr(skb);
244 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
245 if (skb->ip_summed == CHECKSUM_COMPLETE) {
246 __be32 diff[] = { ~(stack->label_stack_entry), lse };
247
248 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
249 ~skb->csum);
250 }
251
252 stack->label_stack_entry = lse;
253 flow_key->mpls.top_lse = lse;
254 return 0;
255}
256
257static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
258{
259 int err;
260
261 err = skb_vlan_pop(skb);
262 if (skb_vlan_tag_present(skb)) {
263 invalidate_flow_key(key);
264 } else {
265 key->eth.vlan.tci = 0;
266 key->eth.vlan.tpid = 0;
267 }
268 return err;
269}
270
271static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
272 const struct ovs_action_push_vlan *vlan)
273{
274 if (skb_vlan_tag_present(skb)) {
275 invalidate_flow_key(key);
276 } else {
277 key->eth.vlan.tci = vlan->vlan_tci;
278 key->eth.vlan.tpid = vlan->vlan_tpid;
279 }
280 return skb_vlan_push(skb, vlan->vlan_tpid,
281 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
282}
283
284/* 'src' is already properly masked. */
285static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
286{
287 u16 *dst = (u16 *)dst_;
288 const u16 *src = (const u16 *)src_;
289 const u16 *mask = (const u16 *)mask_;
290
291 OVS_SET_MASKED(dst[0], src[0], mask[0]);
292 OVS_SET_MASKED(dst[1], src[1], mask[1]);
293 OVS_SET_MASKED(dst[2], src[2], mask[2]);
294}
295
296static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
297 const struct ovs_key_ethernet *key,
298 const struct ovs_key_ethernet *mask)
299{
300 int err;
301
302 err = skb_ensure_writable(skb, ETH_HLEN);
303 if (unlikely(err))
304 return err;
305
306 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
307
308 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
309 mask->eth_src);
310 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
311 mask->eth_dst);
312
313 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
314
315 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
316 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
317 return 0;
318}
319
320/* pop_eth does not support VLAN packets as this action is never called
321 * for them.
322 */
323static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
324{
325 skb_pull_rcsum(skb, ETH_HLEN);
326 skb_reset_mac_header(skb);
327 skb_reset_mac_len(skb);
328
329 /* safe right before invalidate_flow_key */
330 key->mac_proto = MAC_PROTO_NONE;
331 invalidate_flow_key(key);
332 return 0;
333}
334
335static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
336 const struct ovs_action_push_eth *ethh)
337{
338 struct ethhdr *hdr;
339
340 /* Add the new Ethernet header */
341 if (skb_cow_head(skb, ETH_HLEN) < 0)
342 return -ENOMEM;
343
344 skb_push(skb, ETH_HLEN);
345 skb_reset_mac_header(skb);
346 skb_reset_mac_len(skb);
347
348 hdr = eth_hdr(skb);
349 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
350 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
351 hdr->h_proto = skb->protocol;
352
353 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
354
355 /* safe right before invalidate_flow_key */
356 key->mac_proto = MAC_PROTO_ETHERNET;
357 invalidate_flow_key(key);
358 return 0;
359}
360
361static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
362 __be32 addr, __be32 new_addr)
363{
364 int transport_len = skb->len - skb_transport_offset(skb);
365
366 if (nh->frag_off & htons(IP_OFFSET))
367 return;
368
369 if (nh->protocol == IPPROTO_TCP) {
370 if (likely(transport_len >= sizeof(struct tcphdr)))
371 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
372 addr, new_addr, true);
373 } else if (nh->protocol == IPPROTO_UDP) {
374 if (likely(transport_len >= sizeof(struct udphdr))) {
375 struct udphdr *uh = udp_hdr(skb);
376
377 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
378 inet_proto_csum_replace4(&uh->check, skb,
379 addr, new_addr, true);
380 if (!uh->check)
381 uh->check = CSUM_MANGLED_0;
382 }
383 }
384 }
385}
386
387static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
388 __be32 *addr, __be32 new_addr)
389{
390 update_ip_l4_checksum(skb, nh, *addr, new_addr);
391 csum_replace4(&nh->check, *addr, new_addr);
392 skb_clear_hash(skb);
393 *addr = new_addr;
394}
395
396static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
397 __be32 addr[4], const __be32 new_addr[4])
398{
399 int transport_len = skb->len - skb_transport_offset(skb);
400
401 if (l4_proto == NEXTHDR_TCP) {
402 if (likely(transport_len >= sizeof(struct tcphdr)))
403 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
404 addr, new_addr, true);
405 } else if (l4_proto == NEXTHDR_UDP) {
406 if (likely(transport_len >= sizeof(struct udphdr))) {
407 struct udphdr *uh = udp_hdr(skb);
408
409 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
410 inet_proto_csum_replace16(&uh->check, skb,
411 addr, new_addr, true);
412 if (!uh->check)
413 uh->check = CSUM_MANGLED_0;
414 }
415 }
416 } else if (l4_proto == NEXTHDR_ICMP) {
417 if (likely(transport_len >= sizeof(struct icmp6hdr)))
418 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
419 skb, addr, new_addr, true);
420 }
421}
422
423static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
424 const __be32 mask[4], __be32 masked[4])
425{
426 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
427 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
428 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
429 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
430}
431
432static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
433 __be32 addr[4], const __be32 new_addr[4],
434 bool recalculate_csum)
435{
436 if (recalculate_csum)
437 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
438
439 skb_clear_hash(skb);
440 memcpy(addr, new_addr, sizeof(__be32[4]));
441}
442
443static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
444{
445 /* Bits 21-24 are always unmasked, so this retains their values. */
446 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
447 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
448 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
449}
450
451static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
452 u8 mask)
453{
454 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
455
456 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
457 nh->ttl = new_ttl;
458}
459
460static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
461 const struct ovs_key_ipv4 *key,
462 const struct ovs_key_ipv4 *mask)
463{
464 struct iphdr *nh;
465 __be32 new_addr;
466 int err;
467
468 err = skb_ensure_writable(skb, skb_network_offset(skb) +
469 sizeof(struct iphdr));
470 if (unlikely(err))
471 return err;
472
473 nh = ip_hdr(skb);
474
475 /* Setting an IP addresses is typically only a side effect of
476 * matching on them in the current userspace implementation, so it
477 * makes sense to check if the value actually changed.
478 */
479 if (mask->ipv4_src) {
480 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
481
482 if (unlikely(new_addr != nh->saddr)) {
483 set_ip_addr(skb, nh, &nh->saddr, new_addr);
484 flow_key->ipv4.addr.src = new_addr;
485 }
486 }
487 if (mask->ipv4_dst) {
488 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
489
490 if (unlikely(new_addr != nh->daddr)) {
491 set_ip_addr(skb, nh, &nh->daddr, new_addr);
492 flow_key->ipv4.addr.dst = new_addr;
493 }
494 }
495 if (mask->ipv4_tos) {
496 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
497 flow_key->ip.tos = nh->tos;
498 }
499 if (mask->ipv4_ttl) {
500 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
501 flow_key->ip.ttl = nh->ttl;
502 }
503
504 return 0;
505}
506
507static bool is_ipv6_mask_nonzero(const __be32 addr[4])
508{
509 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
510}
511
512static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
513 const struct ovs_key_ipv6 *key,
514 const struct ovs_key_ipv6 *mask)
515{
516 struct ipv6hdr *nh;
517 int err;
518
519 err = skb_ensure_writable(skb, skb_network_offset(skb) +
520 sizeof(struct ipv6hdr));
521 if (unlikely(err))
522 return err;
523
524 nh = ipv6_hdr(skb);
525
526 /* Setting an IP addresses is typically only a side effect of
527 * matching on them in the current userspace implementation, so it
528 * makes sense to check if the value actually changed.
529 */
530 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
531 __be32 *saddr = (__be32 *)&nh->saddr;
532 __be32 masked[4];
533
534 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
535
536 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
537 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
538 true);
539 memcpy(&flow_key->ipv6.addr.src, masked,
540 sizeof(flow_key->ipv6.addr.src));
541 }
542 }
543 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
544 unsigned int offset = 0;
545 int flags = IP6_FH_F_SKIP_RH;
546 bool recalc_csum = true;
547 __be32 *daddr = (__be32 *)&nh->daddr;
548 __be32 masked[4];
549
550 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
551
552 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
553 if (ipv6_ext_hdr(nh->nexthdr))
554 recalc_csum = (ipv6_find_hdr(skb, &offset,
555 NEXTHDR_ROUTING,
556 NULL, &flags)
557 != NEXTHDR_ROUTING);
558
559 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
560 recalc_csum);
561 memcpy(&flow_key->ipv6.addr.dst, masked,
562 sizeof(flow_key->ipv6.addr.dst));
563 }
564 }
565 if (mask->ipv6_tclass) {
566 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
567 flow_key->ip.tos = ipv6_get_dsfield(nh);
568 }
569 if (mask->ipv6_label) {
570 set_ipv6_fl(nh, ntohl(key->ipv6_label),
571 ntohl(mask->ipv6_label));
572 flow_key->ipv6.label =
573 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
574 }
575 if (mask->ipv6_hlimit) {
576 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
577 mask->ipv6_hlimit);
578 flow_key->ip.ttl = nh->hop_limit;
579 }
580 return 0;
581}
582
583/* Must follow skb_ensure_writable() since that can move the skb data. */
584static void set_tp_port(struct sk_buff *skb, __be16 *port,
585 __be16 new_port, __sum16 *check)
586{
587 inet_proto_csum_replace2(check, skb, *port, new_port, false);
588 *port = new_port;
589}
590
591static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
592 const struct ovs_key_udp *key,
593 const struct ovs_key_udp *mask)
594{
595 struct udphdr *uh;
596 __be16 src, dst;
597 int err;
598
599 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
600 sizeof(struct udphdr));
601 if (unlikely(err))
602 return err;
603
604 uh = udp_hdr(skb);
605 /* Either of the masks is non-zero, so do not bother checking them. */
606 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
607 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
608
609 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
610 if (likely(src != uh->source)) {
611 set_tp_port(skb, &uh->source, src, &uh->check);
612 flow_key->tp.src = src;
613 }
614 if (likely(dst != uh->dest)) {
615 set_tp_port(skb, &uh->dest, dst, &uh->check);
616 flow_key->tp.dst = dst;
617 }
618
619 if (unlikely(!uh->check))
620 uh->check = CSUM_MANGLED_0;
621 } else {
622 uh->source = src;
623 uh->dest = dst;
624 flow_key->tp.src = src;
625 flow_key->tp.dst = dst;
626 }
627
628 skb_clear_hash(skb);
629
630 return 0;
631}
632
633static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
634 const struct ovs_key_tcp *key,
635 const struct ovs_key_tcp *mask)
636{
637 struct tcphdr *th;
638 __be16 src, dst;
639 int err;
640
641 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
642 sizeof(struct tcphdr));
643 if (unlikely(err))
644 return err;
645
646 th = tcp_hdr(skb);
647 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
648 if (likely(src != th->source)) {
649 set_tp_port(skb, &th->source, src, &th->check);
650 flow_key->tp.src = src;
651 }
652 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
653 if (likely(dst != th->dest)) {
654 set_tp_port(skb, &th->dest, dst, &th->check);
655 flow_key->tp.dst = dst;
656 }
657 skb_clear_hash(skb);
658
659 return 0;
660}
661
662static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
663 const struct ovs_key_sctp *key,
664 const struct ovs_key_sctp *mask)
665{
666 unsigned int sctphoff = skb_transport_offset(skb);
667 struct sctphdr *sh;
668 __le32 old_correct_csum, new_csum, old_csum;
669 int err;
670
671 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
672 if (unlikely(err))
673 return err;
674
675 sh = sctp_hdr(skb);
676 old_csum = sh->checksum;
677 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
678
679 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
680 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
681
682 new_csum = sctp_compute_cksum(skb, sctphoff);
683
684 /* Carry any checksum errors through. */
685 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
686
687 skb_clear_hash(skb);
688 flow_key->tp.src = sh->source;
689 flow_key->tp.dst = sh->dest;
690
691 return 0;
692}
693
694static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
695{
696 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
697 struct vport *vport = data->vport;
698
699 if (skb_cow_head(skb, data->l2_len) < 0) {
700 kfree_skb(skb);
701 return -ENOMEM;
702 }
703
704 __skb_dst_copy(skb, data->dst);
705 *OVS_CB(skb) = data->cb;
706 skb->inner_protocol = data->inner_protocol;
707 skb->vlan_tci = data->vlan_tci;
708 skb->vlan_proto = data->vlan_proto;
709
710 /* Reconstruct the MAC header. */
711 skb_push(skb, data->l2_len);
712 memcpy(skb->data, &data->l2_data, data->l2_len);
713 skb_postpush_rcsum(skb, skb->data, data->l2_len);
714 skb_reset_mac_header(skb);
715
716 if (eth_p_mpls(skb->protocol)) {
717 skb->inner_network_header = skb->network_header;
718 skb_set_network_header(skb, data->network_offset);
719 skb_reset_mac_len(skb);
720 }
721
722 ovs_vport_send(vport, skb, data->mac_proto);
723 return 0;
724}
725
726static unsigned int
727ovs_dst_get_mtu(const struct dst_entry *dst)
728{
729 return dst->dev->mtu;
730}
731
732static struct dst_ops ovs_dst_ops = {
733 .family = AF_UNSPEC,
734 .mtu = ovs_dst_get_mtu,
735};
736
737/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
738 * ovs_vport_output(), which is called once per fragmented packet.
739 */
740static void prepare_frag(struct vport *vport, struct sk_buff *skb,
741 u16 orig_network_offset, u8 mac_proto)
742{
743 unsigned int hlen = skb_network_offset(skb);
744 struct ovs_frag_data *data;
745
746 data = this_cpu_ptr(&ovs_frag_data_storage);
747 data->dst = skb->_skb_refdst;
748 data->vport = vport;
749 data->cb = *OVS_CB(skb);
750 data->inner_protocol = skb->inner_protocol;
751 data->network_offset = orig_network_offset;
752 data->vlan_tci = skb->vlan_tci;
753 data->vlan_proto = skb->vlan_proto;
754 data->mac_proto = mac_proto;
755 data->l2_len = hlen;
756 memcpy(&data->l2_data, skb->data, hlen);
757
758 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
759 skb_pull(skb, hlen);
760}
761
762static void ovs_fragment(struct net *net, struct vport *vport,
763 struct sk_buff *skb, u16 mru,
764 struct sw_flow_key *key)
765{
766 u16 orig_network_offset = 0;
767
768 if (eth_p_mpls(skb->protocol)) {
769 orig_network_offset = skb_network_offset(skb);
770 skb->network_header = skb->inner_network_header;
771 }
772
773 if (skb_network_offset(skb) > MAX_L2_LEN) {
774 OVS_NLERR(1, "L2 header too long to fragment");
775 goto err;
776 }
777
778 if (key->eth.type == htons(ETH_P_IP)) {
779 struct dst_entry ovs_dst;
780 unsigned long orig_dst;
781
782 prepare_frag(vport, skb, orig_network_offset,
783 ovs_key_mac_proto(key));
784 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
785 DST_OBSOLETE_NONE, DST_NOCOUNT);
786 ovs_dst.dev = vport->dev;
787
788 orig_dst = skb->_skb_refdst;
789 skb_dst_set_noref(skb, &ovs_dst);
790 IPCB(skb)->frag_max_size = mru;
791
792 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
793 refdst_drop(orig_dst);
794 } else if (key->eth.type == htons(ETH_P_IPV6)) {
795 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
796 unsigned long orig_dst;
797 struct rt6_info ovs_rt;
798
799 if (!v6ops) {
800 goto err;
801 }
802
803 prepare_frag(vport, skb, orig_network_offset,
804 ovs_key_mac_proto(key));
805 memset(&ovs_rt, 0, sizeof(ovs_rt));
806 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
807 DST_OBSOLETE_NONE, DST_NOCOUNT);
808 ovs_rt.dst.dev = vport->dev;
809
810 orig_dst = skb->_skb_refdst;
811 skb_dst_set_noref(skb, &ovs_rt.dst);
812 IP6CB(skb)->frag_max_size = mru;
813
814 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
815 refdst_drop(orig_dst);
816 } else {
817 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
818 ovs_vport_name(vport), ntohs(key->eth.type), mru,
819 vport->dev->mtu);
820 goto err;
821 }
822
823 return;
824err:
825 kfree_skb(skb);
826}
827
828static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
829 struct sw_flow_key *key)
830{
831 struct vport *vport = ovs_vport_rcu(dp, out_port);
832
833 if (likely(vport)) {
834 u16 mru = OVS_CB(skb)->mru;
835 u32 cutlen = OVS_CB(skb)->cutlen;
836
837 if (unlikely(cutlen > 0)) {
838 if (skb->len - cutlen > ovs_mac_header_len(key))
839 pskb_trim(skb, skb->len - cutlen);
840 else
841 pskb_trim(skb, ovs_mac_header_len(key));
842 }
843
844 if (likely(!mru ||
845 (skb->len <= mru + vport->dev->hard_header_len))) {
846 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
847 } else if (mru <= vport->dev->mtu) {
848 struct net *net = read_pnet(&dp->net);
849
850 ovs_fragment(net, vport, skb, mru, key);
851 } else {
852 kfree_skb(skb);
853 }
854 } else {
855 kfree_skb(skb);
856 }
857}
858
859static int output_userspace(struct datapath *dp, struct sk_buff *skb,
860 struct sw_flow_key *key, const struct nlattr *attr,
861 const struct nlattr *actions, int actions_len,
862 uint32_t cutlen)
863{
864 struct dp_upcall_info upcall;
865 const struct nlattr *a;
866 int rem;
867
868 memset(&upcall, 0, sizeof(upcall));
869 upcall.cmd = OVS_PACKET_CMD_ACTION;
870 upcall.mru = OVS_CB(skb)->mru;
871
872 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
873 a = nla_next(a, &rem)) {
874 switch (nla_type(a)) {
875 case OVS_USERSPACE_ATTR_USERDATA:
876 upcall.userdata = a;
877 break;
878
879 case OVS_USERSPACE_ATTR_PID:
880 upcall.portid = nla_get_u32(a);
881 break;
882
883 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
884 /* Get out tunnel info. */
885 struct vport *vport;
886
887 vport = ovs_vport_rcu(dp, nla_get_u32(a));
888 if (vport) {
889 int err;
890
891 err = dev_fill_metadata_dst(vport->dev, skb);
892 if (!err)
893 upcall.egress_tun_info = skb_tunnel_info(skb);
894 }
895
896 break;
897 }
898
899 case OVS_USERSPACE_ATTR_ACTIONS: {
900 /* Include actions. */
901 upcall.actions = actions;
902 upcall.actions_len = actions_len;
903 break;
904 }
905
906 } /* End of switch. */
907 }
908
909 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
910}
911
912static int sample(struct datapath *dp, struct sk_buff *skb,
913 struct sw_flow_key *key, const struct nlattr *attr,
914 const struct nlattr *actions, int actions_len)
915{
916 const struct nlattr *acts_list = NULL;
917 const struct nlattr *a;
918 int rem;
919 u32 cutlen = 0;
920
921 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
922 a = nla_next(a, &rem)) {
923 u32 probability;
924
925 switch (nla_type(a)) {
926 case OVS_SAMPLE_ATTR_PROBABILITY:
927 probability = nla_get_u32(a);
928 if (!probability || prandom_u32() > probability)
929 return 0;
930 break;
931
932 case OVS_SAMPLE_ATTR_ACTIONS:
933 acts_list = a;
934 break;
935 }
936 }
937
938 rem = nla_len(acts_list);
939 a = nla_data(acts_list);
940
941 /* Actions list is empty, do nothing */
942 if (unlikely(!rem))
943 return 0;
944
945 /* The only known usage of sample action is having a single user-space
946 * action, or having a truncate action followed by a single user-space
947 * action. Treat this usage as a special case.
948 * The output_userspace() should clone the skb to be sent to the
949 * user space. This skb will be consumed by its caller.
950 */
951 if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
952 struct ovs_action_trunc *trunc = nla_data(a);
953
954 if (skb->len > trunc->max_len)
955 cutlen = skb->len - trunc->max_len;
956
957 a = nla_next(a, &rem);
958 }
959
960 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
961 nla_is_last(a, rem)))
962 return output_userspace(dp, skb, key, a, actions,
963 actions_len, cutlen);
964
965 skb = skb_clone(skb, GFP_ATOMIC);
966 if (!skb)
967 /* Skip the sample action when out of memory. */
968 return 0;
969
970 if (!add_deferred_actions(skb, key, a)) {
971 if (net_ratelimit())
972 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
973 ovs_dp_name(dp));
974
975 kfree_skb(skb);
976 }
977 return 0;
978}
979
980static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
981 const struct nlattr *attr)
982{
983 struct ovs_action_hash *hash_act = nla_data(attr);
984 u32 hash = 0;
985
986 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
987 hash = skb_get_hash(skb);
988 hash = jhash_1word(hash, hash_act->hash_basis);
989 if (!hash)
990 hash = 0x1;
991
992 key->ovs_flow_hash = hash;
993}
994
995static int execute_set_action(struct sk_buff *skb,
996 struct sw_flow_key *flow_key,
997 const struct nlattr *a)
998{
999 /* Only tunnel set execution is supported without a mask. */
1000 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1001 struct ovs_tunnel_info *tun = nla_data(a);
1002
1003 skb_dst_drop(skb);
1004 dst_hold((struct dst_entry *)tun->tun_dst);
1005 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1006 return 0;
1007 }
1008
1009 return -EINVAL;
1010}
1011
1012/* Mask is at the midpoint of the data. */
1013#define get_mask(a, type) ((const type)nla_data(a) + 1)
1014
1015static int execute_masked_set_action(struct sk_buff *skb,
1016 struct sw_flow_key *flow_key,
1017 const struct nlattr *a)
1018{
1019 int err = 0;
1020
1021 switch (nla_type(a)) {
1022 case OVS_KEY_ATTR_PRIORITY:
1023 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1024 *get_mask(a, u32 *));
1025 flow_key->phy.priority = skb->priority;
1026 break;
1027
1028 case OVS_KEY_ATTR_SKB_MARK:
1029 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1030 flow_key->phy.skb_mark = skb->mark;
1031 break;
1032
1033 case OVS_KEY_ATTR_TUNNEL_INFO:
1034 /* Masked data not supported for tunnel. */
1035 err = -EINVAL;
1036 break;
1037
1038 case OVS_KEY_ATTR_ETHERNET:
1039 err = set_eth_addr(skb, flow_key, nla_data(a),
1040 get_mask(a, struct ovs_key_ethernet *));
1041 break;
1042
1043 case OVS_KEY_ATTR_IPV4:
1044 err = set_ipv4(skb, flow_key, nla_data(a),
1045 get_mask(a, struct ovs_key_ipv4 *));
1046 break;
1047
1048 case OVS_KEY_ATTR_IPV6:
1049 err = set_ipv6(skb, flow_key, nla_data(a),
1050 get_mask(a, struct ovs_key_ipv6 *));
1051 break;
1052
1053 case OVS_KEY_ATTR_TCP:
1054 err = set_tcp(skb, flow_key, nla_data(a),
1055 get_mask(a, struct ovs_key_tcp *));
1056 break;
1057
1058 case OVS_KEY_ATTR_UDP:
1059 err = set_udp(skb, flow_key, nla_data(a),
1060 get_mask(a, struct ovs_key_udp *));
1061 break;
1062
1063 case OVS_KEY_ATTR_SCTP:
1064 err = set_sctp(skb, flow_key, nla_data(a),
1065 get_mask(a, struct ovs_key_sctp *));
1066 break;
1067
1068 case OVS_KEY_ATTR_MPLS:
1069 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1070 __be32 *));
1071 break;
1072
1073 case OVS_KEY_ATTR_CT_STATE:
1074 case OVS_KEY_ATTR_CT_ZONE:
1075 case OVS_KEY_ATTR_CT_MARK:
1076 case OVS_KEY_ATTR_CT_LABELS:
1077 err = -EINVAL;
1078 break;
1079 }
1080
1081 return err;
1082}
1083
1084static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1085 struct sw_flow_key *key,
1086 const struct nlattr *a, int rem)
1087{
1088 struct deferred_action *da;
1089 int level;
1090
1091 if (!is_flow_key_valid(key)) {
1092 int err;
1093
1094 err = ovs_flow_key_update(skb, key);
1095 if (err)
1096 return err;
1097 }
1098 BUG_ON(!is_flow_key_valid(key));
1099
1100 if (!nla_is_last(a, rem)) {
1101 /* Recirc action is the not the last action
1102 * of the action list, need to clone the skb.
1103 */
1104 skb = skb_clone(skb, GFP_ATOMIC);
1105
1106 /* Skip the recirc action when out of memory, but
1107 * continue on with the rest of the action list.
1108 */
1109 if (!skb)
1110 return 0;
1111 }
1112
1113 level = this_cpu_read(exec_actions_level);
1114 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
1115 struct recirc_keys *rks = this_cpu_ptr(recirc_keys);
1116 struct sw_flow_key *recirc_key = &rks->key[level - 1];
1117
1118 *recirc_key = *key;
1119 recirc_key->recirc_id = nla_get_u32(a);
1120 ovs_dp_process_packet(skb, recirc_key);
1121
1122 return 0;
1123 }
1124
1125 da = add_deferred_actions(skb, key, NULL);
1126 if (da) {
1127 da->pkt_key.recirc_id = nla_get_u32(a);
1128 } else {
1129 kfree_skb(skb);
1130
1131 if (net_ratelimit())
1132 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1133 ovs_dp_name(dp));
1134 }
1135
1136 return 0;
1137}
1138
1139/* Execute a list of actions against 'skb'. */
1140static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1141 struct sw_flow_key *key,
1142 const struct nlattr *attr, int len)
1143{
1144 /* Every output action needs a separate clone of 'skb', but the common
1145 * case is just a single output action, so that doing a clone and
1146 * then freeing the original skbuff is wasteful. So the following code
1147 * is slightly obscure just to avoid that.
1148 */
1149 int prev_port = -1;
1150 const struct nlattr *a;
1151 int rem;
1152
1153 for (a = attr, rem = len; rem > 0;
1154 a = nla_next(a, &rem)) {
1155 int err = 0;
1156
1157 if (unlikely(prev_port != -1)) {
1158 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1159
1160 if (out_skb)
1161 do_output(dp, out_skb, prev_port, key);
1162
1163 OVS_CB(skb)->cutlen = 0;
1164 prev_port = -1;
1165 }
1166
1167 switch (nla_type(a)) {
1168 case OVS_ACTION_ATTR_OUTPUT:
1169 prev_port = nla_get_u32(a);
1170 break;
1171
1172 case OVS_ACTION_ATTR_TRUNC: {
1173 struct ovs_action_trunc *trunc = nla_data(a);
1174
1175 if (skb->len > trunc->max_len)
1176 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1177 break;
1178 }
1179
1180 case OVS_ACTION_ATTR_USERSPACE:
1181 output_userspace(dp, skb, key, a, attr,
1182 len, OVS_CB(skb)->cutlen);
1183 OVS_CB(skb)->cutlen = 0;
1184 break;
1185
1186 case OVS_ACTION_ATTR_HASH:
1187 execute_hash(skb, key, a);
1188 break;
1189
1190 case OVS_ACTION_ATTR_PUSH_MPLS:
1191 err = push_mpls(skb, key, nla_data(a));
1192 break;
1193
1194 case OVS_ACTION_ATTR_POP_MPLS:
1195 err = pop_mpls(skb, key, nla_get_be16(a));
1196 break;
1197
1198 case OVS_ACTION_ATTR_PUSH_VLAN:
1199 err = push_vlan(skb, key, nla_data(a));
1200 break;
1201
1202 case OVS_ACTION_ATTR_POP_VLAN:
1203 err = pop_vlan(skb, key);
1204 break;
1205
1206 case OVS_ACTION_ATTR_RECIRC:
1207 err = execute_recirc(dp, skb, key, a, rem);
1208 if (nla_is_last(a, rem)) {
1209 /* If this is the last action, the skb has
1210 * been consumed or freed.
1211 * Return immediately.
1212 */
1213 return err;
1214 }
1215 break;
1216
1217 case OVS_ACTION_ATTR_SET:
1218 err = execute_set_action(skb, key, nla_data(a));
1219 break;
1220
1221 case OVS_ACTION_ATTR_SET_MASKED:
1222 case OVS_ACTION_ATTR_SET_TO_MASKED:
1223 err = execute_masked_set_action(skb, key, nla_data(a));
1224 break;
1225
1226 case OVS_ACTION_ATTR_SAMPLE:
1227 err = sample(dp, skb, key, a, attr, len);
1228 break;
1229
1230 case OVS_ACTION_ATTR_CT:
1231 if (!is_flow_key_valid(key)) {
1232 err = ovs_flow_key_update(skb, key);
1233 if (err)
1234 return err;
1235 }
1236
1237 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1238 nla_data(a));
1239
1240 /* Hide stolen IP fragments from user space. */
1241 if (err)
1242 return err == -EINPROGRESS ? 0 : err;
1243 break;
1244
1245 case OVS_ACTION_ATTR_PUSH_ETH:
1246 err = push_eth(skb, key, nla_data(a));
1247 break;
1248
1249 case OVS_ACTION_ATTR_POP_ETH:
1250 err = pop_eth(skb, key);
1251 break;
1252 }
1253
1254 if (unlikely(err)) {
1255 kfree_skb(skb);
1256 return err;
1257 }
1258 }
1259
1260 if (prev_port != -1)
1261 do_output(dp, skb, prev_port, key);
1262 else
1263 consume_skb(skb);
1264
1265 return 0;
1266}
1267
1268static void process_deferred_actions(struct datapath *dp)
1269{
1270 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1271
1272 /* Do not touch the FIFO in case there is no deferred actions. */
1273 if (action_fifo_is_empty(fifo))
1274 return;
1275
1276 /* Finishing executing all deferred actions. */
1277 do {
1278 struct deferred_action *da = action_fifo_get(fifo);
1279 struct sk_buff *skb = da->skb;
1280 struct sw_flow_key *key = &da->pkt_key;
1281 const struct nlattr *actions = da->actions;
1282
1283 if (actions)
1284 do_execute_actions(dp, skb, key, actions,
1285 nla_len(actions));
1286 else
1287 ovs_dp_process_packet(skb, key);
1288 } while (!action_fifo_is_empty(fifo));
1289
1290 /* Reset FIFO for the next packet. */
1291 action_fifo_init(fifo);
1292}
1293
1294/* Execute a list of actions against 'skb'. */
1295int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1296 const struct sw_flow_actions *acts,
1297 struct sw_flow_key *key)
1298{
1299 int err, level;
1300
1301 level = __this_cpu_inc_return(exec_actions_level);
1302 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1303 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1304 ovs_dp_name(dp));
1305 kfree_skb(skb);
1306 err = -ENETDOWN;
1307 goto out;
1308 }
1309
1310 err = do_execute_actions(dp, skb, key,
1311 acts->actions, acts->actions_len);
1312
1313 if (level == 1)
1314 process_deferred_actions(dp);
1315
1316out:
1317 __this_cpu_dec(exec_actions_level);
1318 return err;
1319}
1320
1321int action_fifos_init(void)
1322{
1323 action_fifos = alloc_percpu(struct action_fifo);
1324 if (!action_fifos)
1325 return -ENOMEM;
1326
1327 recirc_keys = alloc_percpu(struct recirc_keys);
1328 if (!recirc_keys) {
1329 free_percpu(action_fifos);
1330 return -ENOMEM;
1331 }
1332
1333 return 0;
1334}
1335
1336void action_fifos_exit(void)
1337{
1338 free_percpu(action_fifos);
1339 free_percpu(recirc_keys);
1340}