Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2007-2017 Nicira, Inc.
4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/skbuff.h>
9#include <linux/in.h>
10#include <linux/ip.h>
11#include <linux/openvswitch.h>
12#include <linux/netfilter_ipv6.h>
13#include <linux/sctp.h>
14#include <linux/tcp.h>
15#include <linux/udp.h>
16#include <linux/in6.h>
17#include <linux/if_arp.h>
18#include <linux/if_vlan.h>
19
20#include <net/dst.h>
21#include <net/ip.h>
22#include <net/ipv6.h>
23#include <net/ip6_fib.h>
24#include <net/checksum.h>
25#include <net/dsfield.h>
26#include <net/mpls.h>
27#include <net/sctp/checksum.h>
28
29#include "datapath.h"
30#include "flow.h"
31#include "conntrack.h"
32#include "vport.h"
33#include "flow_netlink.h"
34
35struct deferred_action {
36 struct sk_buff *skb;
37 const struct nlattr *actions;
38 int actions_len;
39
40 /* Store pkt_key clone when creating deferred action. */
41 struct sw_flow_key pkt_key;
42};
43
44#define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
45struct ovs_frag_data {
46 unsigned long dst;
47 struct vport *vport;
48 struct ovs_skb_cb cb;
49 __be16 inner_protocol;
50 u16 network_offset; /* valid only for MPLS */
51 u16 vlan_tci;
52 __be16 vlan_proto;
53 unsigned int l2_len;
54 u8 mac_proto;
55 u8 l2_data[MAX_L2_LEN];
56};
57
58static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
59
60#define DEFERRED_ACTION_FIFO_SIZE 10
61#define OVS_RECURSION_LIMIT 5
62#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
63struct action_fifo {
64 int head;
65 int tail;
66 /* Deferred action fifo queue storage. */
67 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
68};
69
70struct action_flow_keys {
71 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
72};
73
74static struct action_fifo __percpu *action_fifos;
75static struct action_flow_keys __percpu *flow_keys;
76static DEFINE_PER_CPU(int, exec_actions_level);
77
78/* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
79 * space. Return NULL if out of key spaces.
80 */
81static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
82{
83 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
84 int level = this_cpu_read(exec_actions_level);
85 struct sw_flow_key *key = NULL;
86
87 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
88 key = &keys->key[level - 1];
89 *key = *key_;
90 }
91
92 return key;
93}
94
95static void action_fifo_init(struct action_fifo *fifo)
96{
97 fifo->head = 0;
98 fifo->tail = 0;
99}
100
101static bool action_fifo_is_empty(const struct action_fifo *fifo)
102{
103 return (fifo->head == fifo->tail);
104}
105
106static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
107{
108 if (action_fifo_is_empty(fifo))
109 return NULL;
110
111 return &fifo->fifo[fifo->tail++];
112}
113
114static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
115{
116 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
117 return NULL;
118
119 return &fifo->fifo[fifo->head++];
120}
121
122/* Return true if fifo is not full */
123static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
124 const struct sw_flow_key *key,
125 const struct nlattr *actions,
126 const int actions_len)
127{
128 struct action_fifo *fifo;
129 struct deferred_action *da;
130
131 fifo = this_cpu_ptr(action_fifos);
132 da = action_fifo_put(fifo);
133 if (da) {
134 da->skb = skb;
135 da->actions = actions;
136 da->actions_len = actions_len;
137 da->pkt_key = *key;
138 }
139
140 return da;
141}
142
143static void invalidate_flow_key(struct sw_flow_key *key)
144{
145 key->mac_proto |= SW_FLOW_KEY_INVALID;
146}
147
148static bool is_flow_key_valid(const struct sw_flow_key *key)
149{
150 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
151}
152
153static int clone_execute(struct datapath *dp, struct sk_buff *skb,
154 struct sw_flow_key *key,
155 u32 recirc_id,
156 const struct nlattr *actions, int len,
157 bool last, bool clone_flow_key);
158
159static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
160 struct sw_flow_key *key,
161 const struct nlattr *attr, int len);
162
163static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
164 const struct ovs_action_push_mpls *mpls)
165{
166 int err;
167
168 err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
169 skb->mac_len);
170 if (err)
171 return err;
172
173 invalidate_flow_key(key);
174 return 0;
175}
176
177static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
178 const __be16 ethertype)
179{
180 int err;
181
182 err = skb_mpls_pop(skb, ethertype, skb->mac_len);
183 if (err)
184 return err;
185
186 invalidate_flow_key(key);
187 return 0;
188}
189
190static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
191 const __be32 *mpls_lse, const __be32 *mask)
192{
193 struct mpls_shim_hdr *stack;
194 __be32 lse;
195 int err;
196
197 stack = mpls_hdr(skb);
198 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
199 err = skb_mpls_update_lse(skb, lse);
200 if (err)
201 return err;
202
203 flow_key->mpls.top_lse = lse;
204 return 0;
205}
206
207static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
208{
209 int err;
210
211 err = skb_vlan_pop(skb);
212 if (skb_vlan_tag_present(skb)) {
213 invalidate_flow_key(key);
214 } else {
215 key->eth.vlan.tci = 0;
216 key->eth.vlan.tpid = 0;
217 }
218 return err;
219}
220
221static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
222 const struct ovs_action_push_vlan *vlan)
223{
224 if (skb_vlan_tag_present(skb)) {
225 invalidate_flow_key(key);
226 } else {
227 key->eth.vlan.tci = vlan->vlan_tci;
228 key->eth.vlan.tpid = vlan->vlan_tpid;
229 }
230 return skb_vlan_push(skb, vlan->vlan_tpid,
231 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
232}
233
234/* 'src' is already properly masked. */
235static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
236{
237 u16 *dst = (u16 *)dst_;
238 const u16 *src = (const u16 *)src_;
239 const u16 *mask = (const u16 *)mask_;
240
241 OVS_SET_MASKED(dst[0], src[0], mask[0]);
242 OVS_SET_MASKED(dst[1], src[1], mask[1]);
243 OVS_SET_MASKED(dst[2], src[2], mask[2]);
244}
245
246static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
247 const struct ovs_key_ethernet *key,
248 const struct ovs_key_ethernet *mask)
249{
250 int err;
251
252 err = skb_ensure_writable(skb, ETH_HLEN);
253 if (unlikely(err))
254 return err;
255
256 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
257
258 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
259 mask->eth_src);
260 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
261 mask->eth_dst);
262
263 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
264
265 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
266 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
267 return 0;
268}
269
270/* pop_eth does not support VLAN packets as this action is never called
271 * for them.
272 */
273static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
274{
275 skb_pull_rcsum(skb, ETH_HLEN);
276 skb_reset_mac_header(skb);
277 skb_reset_mac_len(skb);
278
279 /* safe right before invalidate_flow_key */
280 key->mac_proto = MAC_PROTO_NONE;
281 invalidate_flow_key(key);
282 return 0;
283}
284
285static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
286 const struct ovs_action_push_eth *ethh)
287{
288 struct ethhdr *hdr;
289
290 /* Add the new Ethernet header */
291 if (skb_cow_head(skb, ETH_HLEN) < 0)
292 return -ENOMEM;
293
294 skb_push(skb, ETH_HLEN);
295 skb_reset_mac_header(skb);
296 skb_reset_mac_len(skb);
297
298 hdr = eth_hdr(skb);
299 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
300 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
301 hdr->h_proto = skb->protocol;
302
303 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
304
305 /* safe right before invalidate_flow_key */
306 key->mac_proto = MAC_PROTO_ETHERNET;
307 invalidate_flow_key(key);
308 return 0;
309}
310
311static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
312 const struct nshhdr *nh)
313{
314 int err;
315
316 err = nsh_push(skb, nh);
317 if (err)
318 return err;
319
320 /* safe right before invalidate_flow_key */
321 key->mac_proto = MAC_PROTO_NONE;
322 invalidate_flow_key(key);
323 return 0;
324}
325
326static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
327{
328 int err;
329
330 err = nsh_pop(skb);
331 if (err)
332 return err;
333
334 /* safe right before invalidate_flow_key */
335 if (skb->protocol == htons(ETH_P_TEB))
336 key->mac_proto = MAC_PROTO_ETHERNET;
337 else
338 key->mac_proto = MAC_PROTO_NONE;
339 invalidate_flow_key(key);
340 return 0;
341}
342
343static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
344 __be32 addr, __be32 new_addr)
345{
346 int transport_len = skb->len - skb_transport_offset(skb);
347
348 if (nh->frag_off & htons(IP_OFFSET))
349 return;
350
351 if (nh->protocol == IPPROTO_TCP) {
352 if (likely(transport_len >= sizeof(struct tcphdr)))
353 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
354 addr, new_addr, true);
355 } else if (nh->protocol == IPPROTO_UDP) {
356 if (likely(transport_len >= sizeof(struct udphdr))) {
357 struct udphdr *uh = udp_hdr(skb);
358
359 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
360 inet_proto_csum_replace4(&uh->check, skb,
361 addr, new_addr, true);
362 if (!uh->check)
363 uh->check = CSUM_MANGLED_0;
364 }
365 }
366 }
367}
368
369static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
370 __be32 *addr, __be32 new_addr)
371{
372 update_ip_l4_checksum(skb, nh, *addr, new_addr);
373 csum_replace4(&nh->check, *addr, new_addr);
374 skb_clear_hash(skb);
375 *addr = new_addr;
376}
377
378static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
379 __be32 addr[4], const __be32 new_addr[4])
380{
381 int transport_len = skb->len - skb_transport_offset(skb);
382
383 if (l4_proto == NEXTHDR_TCP) {
384 if (likely(transport_len >= sizeof(struct tcphdr)))
385 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
386 addr, new_addr, true);
387 } else if (l4_proto == NEXTHDR_UDP) {
388 if (likely(transport_len >= sizeof(struct udphdr))) {
389 struct udphdr *uh = udp_hdr(skb);
390
391 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
392 inet_proto_csum_replace16(&uh->check, skb,
393 addr, new_addr, true);
394 if (!uh->check)
395 uh->check = CSUM_MANGLED_0;
396 }
397 }
398 } else if (l4_proto == NEXTHDR_ICMP) {
399 if (likely(transport_len >= sizeof(struct icmp6hdr)))
400 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
401 skb, addr, new_addr, true);
402 }
403}
404
405static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
406 const __be32 mask[4], __be32 masked[4])
407{
408 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
409 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
410 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
411 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
412}
413
414static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
415 __be32 addr[4], const __be32 new_addr[4],
416 bool recalculate_csum)
417{
418 if (recalculate_csum)
419 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
420
421 skb_clear_hash(skb);
422 memcpy(addr, new_addr, sizeof(__be32[4]));
423}
424
425static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
426{
427 /* Bits 21-24 are always unmasked, so this retains their values. */
428 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
429 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
430 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
431}
432
433static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
434 u8 mask)
435{
436 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
437
438 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
439 nh->ttl = new_ttl;
440}
441
442static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
443 const struct ovs_key_ipv4 *key,
444 const struct ovs_key_ipv4 *mask)
445{
446 struct iphdr *nh;
447 __be32 new_addr;
448 int err;
449
450 err = skb_ensure_writable(skb, skb_network_offset(skb) +
451 sizeof(struct iphdr));
452 if (unlikely(err))
453 return err;
454
455 nh = ip_hdr(skb);
456
457 /* Setting an IP addresses is typically only a side effect of
458 * matching on them in the current userspace implementation, so it
459 * makes sense to check if the value actually changed.
460 */
461 if (mask->ipv4_src) {
462 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
463
464 if (unlikely(new_addr != nh->saddr)) {
465 set_ip_addr(skb, nh, &nh->saddr, new_addr);
466 flow_key->ipv4.addr.src = new_addr;
467 }
468 }
469 if (mask->ipv4_dst) {
470 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
471
472 if (unlikely(new_addr != nh->daddr)) {
473 set_ip_addr(skb, nh, &nh->daddr, new_addr);
474 flow_key->ipv4.addr.dst = new_addr;
475 }
476 }
477 if (mask->ipv4_tos) {
478 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
479 flow_key->ip.tos = nh->tos;
480 }
481 if (mask->ipv4_ttl) {
482 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
483 flow_key->ip.ttl = nh->ttl;
484 }
485
486 return 0;
487}
488
489static bool is_ipv6_mask_nonzero(const __be32 addr[4])
490{
491 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
492}
493
494static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
495 const struct ovs_key_ipv6 *key,
496 const struct ovs_key_ipv6 *mask)
497{
498 struct ipv6hdr *nh;
499 int err;
500
501 err = skb_ensure_writable(skb, skb_network_offset(skb) +
502 sizeof(struct ipv6hdr));
503 if (unlikely(err))
504 return err;
505
506 nh = ipv6_hdr(skb);
507
508 /* Setting an IP addresses is typically only a side effect of
509 * matching on them in the current userspace implementation, so it
510 * makes sense to check if the value actually changed.
511 */
512 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
513 __be32 *saddr = (__be32 *)&nh->saddr;
514 __be32 masked[4];
515
516 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
517
518 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
519 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
520 true);
521 memcpy(&flow_key->ipv6.addr.src, masked,
522 sizeof(flow_key->ipv6.addr.src));
523 }
524 }
525 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
526 unsigned int offset = 0;
527 int flags = IP6_FH_F_SKIP_RH;
528 bool recalc_csum = true;
529 __be32 *daddr = (__be32 *)&nh->daddr;
530 __be32 masked[4];
531
532 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
533
534 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
535 if (ipv6_ext_hdr(nh->nexthdr))
536 recalc_csum = (ipv6_find_hdr(skb, &offset,
537 NEXTHDR_ROUTING,
538 NULL, &flags)
539 != NEXTHDR_ROUTING);
540
541 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
542 recalc_csum);
543 memcpy(&flow_key->ipv6.addr.dst, masked,
544 sizeof(flow_key->ipv6.addr.dst));
545 }
546 }
547 if (mask->ipv6_tclass) {
548 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
549 flow_key->ip.tos = ipv6_get_dsfield(nh);
550 }
551 if (mask->ipv6_label) {
552 set_ipv6_fl(nh, ntohl(key->ipv6_label),
553 ntohl(mask->ipv6_label));
554 flow_key->ipv6.label =
555 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
556 }
557 if (mask->ipv6_hlimit) {
558 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
559 mask->ipv6_hlimit);
560 flow_key->ip.ttl = nh->hop_limit;
561 }
562 return 0;
563}
564
565static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
566 const struct nlattr *a)
567{
568 struct nshhdr *nh;
569 size_t length;
570 int err;
571 u8 flags;
572 u8 ttl;
573 int i;
574
575 struct ovs_key_nsh key;
576 struct ovs_key_nsh mask;
577
578 err = nsh_key_from_nlattr(a, &key, &mask);
579 if (err)
580 return err;
581
582 /* Make sure the NSH base header is there */
583 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
584 return -ENOMEM;
585
586 nh = nsh_hdr(skb);
587 length = nsh_hdr_len(nh);
588
589 /* Make sure the whole NSH header is there */
590 err = skb_ensure_writable(skb, skb_network_offset(skb) +
591 length);
592 if (unlikely(err))
593 return err;
594
595 nh = nsh_hdr(skb);
596 skb_postpull_rcsum(skb, nh, length);
597 flags = nsh_get_flags(nh);
598 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
599 flow_key->nsh.base.flags = flags;
600 ttl = nsh_get_ttl(nh);
601 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
602 flow_key->nsh.base.ttl = ttl;
603 nsh_set_flags_and_ttl(nh, flags, ttl);
604 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
605 mask.base.path_hdr);
606 flow_key->nsh.base.path_hdr = nh->path_hdr;
607 switch (nh->mdtype) {
608 case NSH_M_TYPE1:
609 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
610 nh->md1.context[i] =
611 OVS_MASKED(nh->md1.context[i], key.context[i],
612 mask.context[i]);
613 }
614 memcpy(flow_key->nsh.context, nh->md1.context,
615 sizeof(nh->md1.context));
616 break;
617 case NSH_M_TYPE2:
618 memset(flow_key->nsh.context, 0,
619 sizeof(flow_key->nsh.context));
620 break;
621 default:
622 return -EINVAL;
623 }
624 skb_postpush_rcsum(skb, nh, length);
625 return 0;
626}
627
628/* Must follow skb_ensure_writable() since that can move the skb data. */
629static void set_tp_port(struct sk_buff *skb, __be16 *port,
630 __be16 new_port, __sum16 *check)
631{
632 inet_proto_csum_replace2(check, skb, *port, new_port, false);
633 *port = new_port;
634}
635
636static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
637 const struct ovs_key_udp *key,
638 const struct ovs_key_udp *mask)
639{
640 struct udphdr *uh;
641 __be16 src, dst;
642 int err;
643
644 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
645 sizeof(struct udphdr));
646 if (unlikely(err))
647 return err;
648
649 uh = udp_hdr(skb);
650 /* Either of the masks is non-zero, so do not bother checking them. */
651 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
652 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
653
654 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
655 if (likely(src != uh->source)) {
656 set_tp_port(skb, &uh->source, src, &uh->check);
657 flow_key->tp.src = src;
658 }
659 if (likely(dst != uh->dest)) {
660 set_tp_port(skb, &uh->dest, dst, &uh->check);
661 flow_key->tp.dst = dst;
662 }
663
664 if (unlikely(!uh->check))
665 uh->check = CSUM_MANGLED_0;
666 } else {
667 uh->source = src;
668 uh->dest = dst;
669 flow_key->tp.src = src;
670 flow_key->tp.dst = dst;
671 }
672
673 skb_clear_hash(skb);
674
675 return 0;
676}
677
678static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
679 const struct ovs_key_tcp *key,
680 const struct ovs_key_tcp *mask)
681{
682 struct tcphdr *th;
683 __be16 src, dst;
684 int err;
685
686 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
687 sizeof(struct tcphdr));
688 if (unlikely(err))
689 return err;
690
691 th = tcp_hdr(skb);
692 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
693 if (likely(src != th->source)) {
694 set_tp_port(skb, &th->source, src, &th->check);
695 flow_key->tp.src = src;
696 }
697 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
698 if (likely(dst != th->dest)) {
699 set_tp_port(skb, &th->dest, dst, &th->check);
700 flow_key->tp.dst = dst;
701 }
702 skb_clear_hash(skb);
703
704 return 0;
705}
706
707static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
708 const struct ovs_key_sctp *key,
709 const struct ovs_key_sctp *mask)
710{
711 unsigned int sctphoff = skb_transport_offset(skb);
712 struct sctphdr *sh;
713 __le32 old_correct_csum, new_csum, old_csum;
714 int err;
715
716 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
717 if (unlikely(err))
718 return err;
719
720 sh = sctp_hdr(skb);
721 old_csum = sh->checksum;
722 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
723
724 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
725 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
726
727 new_csum = sctp_compute_cksum(skb, sctphoff);
728
729 /* Carry any checksum errors through. */
730 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
731
732 skb_clear_hash(skb);
733 flow_key->tp.src = sh->source;
734 flow_key->tp.dst = sh->dest;
735
736 return 0;
737}
738
739static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
740{
741 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
742 struct vport *vport = data->vport;
743
744 if (skb_cow_head(skb, data->l2_len) < 0) {
745 kfree_skb(skb);
746 return -ENOMEM;
747 }
748
749 __skb_dst_copy(skb, data->dst);
750 *OVS_CB(skb) = data->cb;
751 skb->inner_protocol = data->inner_protocol;
752 if (data->vlan_tci & VLAN_CFI_MASK)
753 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
754 else
755 __vlan_hwaccel_clear_tag(skb);
756
757 /* Reconstruct the MAC header. */
758 skb_push(skb, data->l2_len);
759 memcpy(skb->data, &data->l2_data, data->l2_len);
760 skb_postpush_rcsum(skb, skb->data, data->l2_len);
761 skb_reset_mac_header(skb);
762
763 if (eth_p_mpls(skb->protocol)) {
764 skb->inner_network_header = skb->network_header;
765 skb_set_network_header(skb, data->network_offset);
766 skb_reset_mac_len(skb);
767 }
768
769 ovs_vport_send(vport, skb, data->mac_proto);
770 return 0;
771}
772
773static unsigned int
774ovs_dst_get_mtu(const struct dst_entry *dst)
775{
776 return dst->dev->mtu;
777}
778
779static struct dst_ops ovs_dst_ops = {
780 .family = AF_UNSPEC,
781 .mtu = ovs_dst_get_mtu,
782};
783
784/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
785 * ovs_vport_output(), which is called once per fragmented packet.
786 */
787static void prepare_frag(struct vport *vport, struct sk_buff *skb,
788 u16 orig_network_offset, u8 mac_proto)
789{
790 unsigned int hlen = skb_network_offset(skb);
791 struct ovs_frag_data *data;
792
793 data = this_cpu_ptr(&ovs_frag_data_storage);
794 data->dst = skb->_skb_refdst;
795 data->vport = vport;
796 data->cb = *OVS_CB(skb);
797 data->inner_protocol = skb->inner_protocol;
798 data->network_offset = orig_network_offset;
799 if (skb_vlan_tag_present(skb))
800 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
801 else
802 data->vlan_tci = 0;
803 data->vlan_proto = skb->vlan_proto;
804 data->mac_proto = mac_proto;
805 data->l2_len = hlen;
806 memcpy(&data->l2_data, skb->data, hlen);
807
808 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
809 skb_pull(skb, hlen);
810}
811
812static void ovs_fragment(struct net *net, struct vport *vport,
813 struct sk_buff *skb, u16 mru,
814 struct sw_flow_key *key)
815{
816 u16 orig_network_offset = 0;
817
818 if (eth_p_mpls(skb->protocol)) {
819 orig_network_offset = skb_network_offset(skb);
820 skb->network_header = skb->inner_network_header;
821 }
822
823 if (skb_network_offset(skb) > MAX_L2_LEN) {
824 OVS_NLERR(1, "L2 header too long to fragment");
825 goto err;
826 }
827
828 if (key->eth.type == htons(ETH_P_IP)) {
829 struct dst_entry ovs_dst;
830 unsigned long orig_dst;
831
832 prepare_frag(vport, skb, orig_network_offset,
833 ovs_key_mac_proto(key));
834 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
835 DST_OBSOLETE_NONE, DST_NOCOUNT);
836 ovs_dst.dev = vport->dev;
837
838 orig_dst = skb->_skb_refdst;
839 skb_dst_set_noref(skb, &ovs_dst);
840 IPCB(skb)->frag_max_size = mru;
841
842 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
843 refdst_drop(orig_dst);
844 } else if (key->eth.type == htons(ETH_P_IPV6)) {
845 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
846 unsigned long orig_dst;
847 struct rt6_info ovs_rt;
848
849 if (!v6ops)
850 goto err;
851
852 prepare_frag(vport, skb, orig_network_offset,
853 ovs_key_mac_proto(key));
854 memset(&ovs_rt, 0, sizeof(ovs_rt));
855 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
856 DST_OBSOLETE_NONE, DST_NOCOUNT);
857 ovs_rt.dst.dev = vport->dev;
858
859 orig_dst = skb->_skb_refdst;
860 skb_dst_set_noref(skb, &ovs_rt.dst);
861 IP6CB(skb)->frag_max_size = mru;
862
863 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
864 refdst_drop(orig_dst);
865 } else {
866 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
867 ovs_vport_name(vport), ntohs(key->eth.type), mru,
868 vport->dev->mtu);
869 goto err;
870 }
871
872 return;
873err:
874 kfree_skb(skb);
875}
876
877static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
878 struct sw_flow_key *key)
879{
880 struct vport *vport = ovs_vport_rcu(dp, out_port);
881
882 if (likely(vport)) {
883 u16 mru = OVS_CB(skb)->mru;
884 u32 cutlen = OVS_CB(skb)->cutlen;
885
886 if (unlikely(cutlen > 0)) {
887 if (skb->len - cutlen > ovs_mac_header_len(key))
888 pskb_trim(skb, skb->len - cutlen);
889 else
890 pskb_trim(skb, ovs_mac_header_len(key));
891 }
892
893 if (likely(!mru ||
894 (skb->len <= mru + vport->dev->hard_header_len))) {
895 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
896 } else if (mru <= vport->dev->mtu) {
897 struct net *net = read_pnet(&dp->net);
898
899 ovs_fragment(net, vport, skb, mru, key);
900 } else {
901 kfree_skb(skb);
902 }
903 } else {
904 kfree_skb(skb);
905 }
906}
907
908static int output_userspace(struct datapath *dp, struct sk_buff *skb,
909 struct sw_flow_key *key, const struct nlattr *attr,
910 const struct nlattr *actions, int actions_len,
911 uint32_t cutlen)
912{
913 struct dp_upcall_info upcall;
914 const struct nlattr *a;
915 int rem;
916
917 memset(&upcall, 0, sizeof(upcall));
918 upcall.cmd = OVS_PACKET_CMD_ACTION;
919 upcall.mru = OVS_CB(skb)->mru;
920
921 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
922 a = nla_next(a, &rem)) {
923 switch (nla_type(a)) {
924 case OVS_USERSPACE_ATTR_USERDATA:
925 upcall.userdata = a;
926 break;
927
928 case OVS_USERSPACE_ATTR_PID:
929 upcall.portid = nla_get_u32(a);
930 break;
931
932 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
933 /* Get out tunnel info. */
934 struct vport *vport;
935
936 vport = ovs_vport_rcu(dp, nla_get_u32(a));
937 if (vport) {
938 int err;
939
940 err = dev_fill_metadata_dst(vport->dev, skb);
941 if (!err)
942 upcall.egress_tun_info = skb_tunnel_info(skb);
943 }
944
945 break;
946 }
947
948 case OVS_USERSPACE_ATTR_ACTIONS: {
949 /* Include actions. */
950 upcall.actions = actions;
951 upcall.actions_len = actions_len;
952 break;
953 }
954
955 } /* End of switch. */
956 }
957
958 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
959}
960
961/* When 'last' is true, sample() should always consume the 'skb'.
962 * Otherwise, sample() should keep 'skb' intact regardless what
963 * actions are executed within sample().
964 */
965static int sample(struct datapath *dp, struct sk_buff *skb,
966 struct sw_flow_key *key, const struct nlattr *attr,
967 bool last)
968{
969 struct nlattr *actions;
970 struct nlattr *sample_arg;
971 int rem = nla_len(attr);
972 const struct sample_arg *arg;
973 bool clone_flow_key;
974
975 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
976 sample_arg = nla_data(attr);
977 arg = nla_data(sample_arg);
978 actions = nla_next(sample_arg, &rem);
979
980 if ((arg->probability != U32_MAX) &&
981 (!arg->probability || prandom_u32() > arg->probability)) {
982 if (last)
983 consume_skb(skb);
984 return 0;
985 }
986
987 clone_flow_key = !arg->exec;
988 return clone_execute(dp, skb, key, 0, actions, rem, last,
989 clone_flow_key);
990}
991
992/* When 'last' is true, clone() should always consume the 'skb'.
993 * Otherwise, clone() should keep 'skb' intact regardless what
994 * actions are executed within clone().
995 */
996static int clone(struct datapath *dp, struct sk_buff *skb,
997 struct sw_flow_key *key, const struct nlattr *attr,
998 bool last)
999{
1000 struct nlattr *actions;
1001 struct nlattr *clone_arg;
1002 int rem = nla_len(attr);
1003 bool dont_clone_flow_key;
1004
1005 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1006 clone_arg = nla_data(attr);
1007 dont_clone_flow_key = nla_get_u32(clone_arg);
1008 actions = nla_next(clone_arg, &rem);
1009
1010 return clone_execute(dp, skb, key, 0, actions, rem, last,
1011 !dont_clone_flow_key);
1012}
1013
1014static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1015 const struct nlattr *attr)
1016{
1017 struct ovs_action_hash *hash_act = nla_data(attr);
1018 u32 hash = 0;
1019
1020 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1021 hash = skb_get_hash(skb);
1022 hash = jhash_1word(hash, hash_act->hash_basis);
1023 if (!hash)
1024 hash = 0x1;
1025
1026 key->ovs_flow_hash = hash;
1027}
1028
1029static int execute_set_action(struct sk_buff *skb,
1030 struct sw_flow_key *flow_key,
1031 const struct nlattr *a)
1032{
1033 /* Only tunnel set execution is supported without a mask. */
1034 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1035 struct ovs_tunnel_info *tun = nla_data(a);
1036
1037 skb_dst_drop(skb);
1038 dst_hold((struct dst_entry *)tun->tun_dst);
1039 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1040 return 0;
1041 }
1042
1043 return -EINVAL;
1044}
1045
1046/* Mask is at the midpoint of the data. */
1047#define get_mask(a, type) ((const type)nla_data(a) + 1)
1048
1049static int execute_masked_set_action(struct sk_buff *skb,
1050 struct sw_flow_key *flow_key,
1051 const struct nlattr *a)
1052{
1053 int err = 0;
1054
1055 switch (nla_type(a)) {
1056 case OVS_KEY_ATTR_PRIORITY:
1057 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1058 *get_mask(a, u32 *));
1059 flow_key->phy.priority = skb->priority;
1060 break;
1061
1062 case OVS_KEY_ATTR_SKB_MARK:
1063 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1064 flow_key->phy.skb_mark = skb->mark;
1065 break;
1066
1067 case OVS_KEY_ATTR_TUNNEL_INFO:
1068 /* Masked data not supported for tunnel. */
1069 err = -EINVAL;
1070 break;
1071
1072 case OVS_KEY_ATTR_ETHERNET:
1073 err = set_eth_addr(skb, flow_key, nla_data(a),
1074 get_mask(a, struct ovs_key_ethernet *));
1075 break;
1076
1077 case OVS_KEY_ATTR_NSH:
1078 err = set_nsh(skb, flow_key, a);
1079 break;
1080
1081 case OVS_KEY_ATTR_IPV4:
1082 err = set_ipv4(skb, flow_key, nla_data(a),
1083 get_mask(a, struct ovs_key_ipv4 *));
1084 break;
1085
1086 case OVS_KEY_ATTR_IPV6:
1087 err = set_ipv6(skb, flow_key, nla_data(a),
1088 get_mask(a, struct ovs_key_ipv6 *));
1089 break;
1090
1091 case OVS_KEY_ATTR_TCP:
1092 err = set_tcp(skb, flow_key, nla_data(a),
1093 get_mask(a, struct ovs_key_tcp *));
1094 break;
1095
1096 case OVS_KEY_ATTR_UDP:
1097 err = set_udp(skb, flow_key, nla_data(a),
1098 get_mask(a, struct ovs_key_udp *));
1099 break;
1100
1101 case OVS_KEY_ATTR_SCTP:
1102 err = set_sctp(skb, flow_key, nla_data(a),
1103 get_mask(a, struct ovs_key_sctp *));
1104 break;
1105
1106 case OVS_KEY_ATTR_MPLS:
1107 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1108 __be32 *));
1109 break;
1110
1111 case OVS_KEY_ATTR_CT_STATE:
1112 case OVS_KEY_ATTR_CT_ZONE:
1113 case OVS_KEY_ATTR_CT_MARK:
1114 case OVS_KEY_ATTR_CT_LABELS:
1115 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1116 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1117 err = -EINVAL;
1118 break;
1119 }
1120
1121 return err;
1122}
1123
1124static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1125 struct sw_flow_key *key,
1126 const struct nlattr *a, bool last)
1127{
1128 u32 recirc_id;
1129
1130 if (!is_flow_key_valid(key)) {
1131 int err;
1132
1133 err = ovs_flow_key_update(skb, key);
1134 if (err)
1135 return err;
1136 }
1137 BUG_ON(!is_flow_key_valid(key));
1138
1139 recirc_id = nla_get_u32(a);
1140 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1141}
1142
1143static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1144 struct sw_flow_key *key,
1145 const struct nlattr *attr, bool last)
1146{
1147 const struct nlattr *actions, *cpl_arg;
1148 const struct check_pkt_len_arg *arg;
1149 int rem = nla_len(attr);
1150 bool clone_flow_key;
1151
1152 /* The first netlink attribute in 'attr' is always
1153 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1154 */
1155 cpl_arg = nla_data(attr);
1156 arg = nla_data(cpl_arg);
1157
1158 if (skb->len <= arg->pkt_len) {
1159 /* Second netlink attribute in 'attr' is always
1160 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1161 */
1162 actions = nla_next(cpl_arg, &rem);
1163 clone_flow_key = !arg->exec_for_lesser_equal;
1164 } else {
1165 /* Third netlink attribute in 'attr' is always
1166 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1167 */
1168 actions = nla_next(cpl_arg, &rem);
1169 actions = nla_next(actions, &rem);
1170 clone_flow_key = !arg->exec_for_greater;
1171 }
1172
1173 return clone_execute(dp, skb, key, 0, nla_data(actions),
1174 nla_len(actions), last, clone_flow_key);
1175}
1176
1177/* Execute a list of actions against 'skb'. */
1178static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1179 struct sw_flow_key *key,
1180 const struct nlattr *attr, int len)
1181{
1182 const struct nlattr *a;
1183 int rem;
1184
1185 for (a = attr, rem = len; rem > 0;
1186 a = nla_next(a, &rem)) {
1187 int err = 0;
1188
1189 switch (nla_type(a)) {
1190 case OVS_ACTION_ATTR_OUTPUT: {
1191 int port = nla_get_u32(a);
1192 struct sk_buff *clone;
1193
1194 /* Every output action needs a separate clone
1195 * of 'skb', In case the output action is the
1196 * last action, cloning can be avoided.
1197 */
1198 if (nla_is_last(a, rem)) {
1199 do_output(dp, skb, port, key);
1200 /* 'skb' has been used for output.
1201 */
1202 return 0;
1203 }
1204
1205 clone = skb_clone(skb, GFP_ATOMIC);
1206 if (clone)
1207 do_output(dp, clone, port, key);
1208 OVS_CB(skb)->cutlen = 0;
1209 break;
1210 }
1211
1212 case OVS_ACTION_ATTR_TRUNC: {
1213 struct ovs_action_trunc *trunc = nla_data(a);
1214
1215 if (skb->len > trunc->max_len)
1216 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1217 break;
1218 }
1219
1220 case OVS_ACTION_ATTR_USERSPACE:
1221 output_userspace(dp, skb, key, a, attr,
1222 len, OVS_CB(skb)->cutlen);
1223 OVS_CB(skb)->cutlen = 0;
1224 break;
1225
1226 case OVS_ACTION_ATTR_HASH:
1227 execute_hash(skb, key, a);
1228 break;
1229
1230 case OVS_ACTION_ATTR_PUSH_MPLS:
1231 err = push_mpls(skb, key, nla_data(a));
1232 break;
1233
1234 case OVS_ACTION_ATTR_POP_MPLS:
1235 err = pop_mpls(skb, key, nla_get_be16(a));
1236 break;
1237
1238 case OVS_ACTION_ATTR_PUSH_VLAN:
1239 err = push_vlan(skb, key, nla_data(a));
1240 break;
1241
1242 case OVS_ACTION_ATTR_POP_VLAN:
1243 err = pop_vlan(skb, key);
1244 break;
1245
1246 case OVS_ACTION_ATTR_RECIRC: {
1247 bool last = nla_is_last(a, rem);
1248
1249 err = execute_recirc(dp, skb, key, a, last);
1250 if (last) {
1251 /* If this is the last action, the skb has
1252 * been consumed or freed.
1253 * Return immediately.
1254 */
1255 return err;
1256 }
1257 break;
1258 }
1259
1260 case OVS_ACTION_ATTR_SET:
1261 err = execute_set_action(skb, key, nla_data(a));
1262 break;
1263
1264 case OVS_ACTION_ATTR_SET_MASKED:
1265 case OVS_ACTION_ATTR_SET_TO_MASKED:
1266 err = execute_masked_set_action(skb, key, nla_data(a));
1267 break;
1268
1269 case OVS_ACTION_ATTR_SAMPLE: {
1270 bool last = nla_is_last(a, rem);
1271
1272 err = sample(dp, skb, key, a, last);
1273 if (last)
1274 return err;
1275
1276 break;
1277 }
1278
1279 case OVS_ACTION_ATTR_CT:
1280 if (!is_flow_key_valid(key)) {
1281 err = ovs_flow_key_update(skb, key);
1282 if (err)
1283 return err;
1284 }
1285
1286 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1287 nla_data(a));
1288
1289 /* Hide stolen IP fragments from user space. */
1290 if (err)
1291 return err == -EINPROGRESS ? 0 : err;
1292 break;
1293
1294 case OVS_ACTION_ATTR_CT_CLEAR:
1295 err = ovs_ct_clear(skb, key);
1296 break;
1297
1298 case OVS_ACTION_ATTR_PUSH_ETH:
1299 err = push_eth(skb, key, nla_data(a));
1300 break;
1301
1302 case OVS_ACTION_ATTR_POP_ETH:
1303 err = pop_eth(skb, key);
1304 break;
1305
1306 case OVS_ACTION_ATTR_PUSH_NSH: {
1307 u8 buffer[NSH_HDR_MAX_LEN];
1308 struct nshhdr *nh = (struct nshhdr *)buffer;
1309
1310 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1311 NSH_HDR_MAX_LEN);
1312 if (unlikely(err))
1313 break;
1314 err = push_nsh(skb, key, nh);
1315 break;
1316 }
1317
1318 case OVS_ACTION_ATTR_POP_NSH:
1319 err = pop_nsh(skb, key);
1320 break;
1321
1322 case OVS_ACTION_ATTR_METER:
1323 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1324 consume_skb(skb);
1325 return 0;
1326 }
1327 break;
1328
1329 case OVS_ACTION_ATTR_CLONE: {
1330 bool last = nla_is_last(a, rem);
1331
1332 err = clone(dp, skb, key, a, last);
1333 if (last)
1334 return err;
1335
1336 break;
1337 }
1338
1339 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1340 bool last = nla_is_last(a, rem);
1341
1342 err = execute_check_pkt_len(dp, skb, key, a, last);
1343 if (last)
1344 return err;
1345
1346 break;
1347 }
1348 }
1349
1350 if (unlikely(err)) {
1351 kfree_skb(skb);
1352 return err;
1353 }
1354 }
1355
1356 consume_skb(skb);
1357 return 0;
1358}
1359
1360/* Execute the actions on the clone of the packet. The effect of the
1361 * execution does not affect the original 'skb' nor the original 'key'.
1362 *
1363 * The execution may be deferred in case the actions can not be executed
1364 * immediately.
1365 */
1366static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1367 struct sw_flow_key *key, u32 recirc_id,
1368 const struct nlattr *actions, int len,
1369 bool last, bool clone_flow_key)
1370{
1371 struct deferred_action *da;
1372 struct sw_flow_key *clone;
1373
1374 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1375 if (!skb) {
1376 /* Out of memory, skip this action.
1377 */
1378 return 0;
1379 }
1380
1381 /* When clone_flow_key is false, the 'key' will not be change
1382 * by the actions, then the 'key' can be used directly.
1383 * Otherwise, try to clone key from the next recursion level of
1384 * 'flow_keys'. If clone is successful, execute the actions
1385 * without deferring.
1386 */
1387 clone = clone_flow_key ? clone_key(key) : key;
1388 if (clone) {
1389 int err = 0;
1390
1391 if (actions) { /* Sample action */
1392 if (clone_flow_key)
1393 __this_cpu_inc(exec_actions_level);
1394
1395 err = do_execute_actions(dp, skb, clone,
1396 actions, len);
1397
1398 if (clone_flow_key)
1399 __this_cpu_dec(exec_actions_level);
1400 } else { /* Recirc action */
1401 clone->recirc_id = recirc_id;
1402 ovs_dp_process_packet(skb, clone);
1403 }
1404 return err;
1405 }
1406
1407 /* Out of 'flow_keys' space. Defer actions */
1408 da = add_deferred_actions(skb, key, actions, len);
1409 if (da) {
1410 if (!actions) { /* Recirc action */
1411 key = &da->pkt_key;
1412 key->recirc_id = recirc_id;
1413 }
1414 } else {
1415 /* Out of per CPU action FIFO space. Drop the 'skb' and
1416 * log an error.
1417 */
1418 kfree_skb(skb);
1419
1420 if (net_ratelimit()) {
1421 if (actions) { /* Sample action */
1422 pr_warn("%s: deferred action limit reached, drop sample action\n",
1423 ovs_dp_name(dp));
1424 } else { /* Recirc action */
1425 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1426 ovs_dp_name(dp));
1427 }
1428 }
1429 }
1430 return 0;
1431}
1432
1433static void process_deferred_actions(struct datapath *dp)
1434{
1435 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1436
1437 /* Do not touch the FIFO in case there is no deferred actions. */
1438 if (action_fifo_is_empty(fifo))
1439 return;
1440
1441 /* Finishing executing all deferred actions. */
1442 do {
1443 struct deferred_action *da = action_fifo_get(fifo);
1444 struct sk_buff *skb = da->skb;
1445 struct sw_flow_key *key = &da->pkt_key;
1446 const struct nlattr *actions = da->actions;
1447 int actions_len = da->actions_len;
1448
1449 if (actions)
1450 do_execute_actions(dp, skb, key, actions, actions_len);
1451 else
1452 ovs_dp_process_packet(skb, key);
1453 } while (!action_fifo_is_empty(fifo));
1454
1455 /* Reset FIFO for the next packet. */
1456 action_fifo_init(fifo);
1457}
1458
1459/* Execute a list of actions against 'skb'. */
1460int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1461 const struct sw_flow_actions *acts,
1462 struct sw_flow_key *key)
1463{
1464 int err, level;
1465
1466 level = __this_cpu_inc_return(exec_actions_level);
1467 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1468 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1469 ovs_dp_name(dp));
1470 kfree_skb(skb);
1471 err = -ENETDOWN;
1472 goto out;
1473 }
1474
1475 OVS_CB(skb)->acts_origlen = acts->orig_len;
1476 err = do_execute_actions(dp, skb, key,
1477 acts->actions, acts->actions_len);
1478
1479 if (level == 1)
1480 process_deferred_actions(dp);
1481
1482out:
1483 __this_cpu_dec(exec_actions_level);
1484 return err;
1485}
1486
1487int action_fifos_init(void)
1488{
1489 action_fifos = alloc_percpu(struct action_fifo);
1490 if (!action_fifos)
1491 return -ENOMEM;
1492
1493 flow_keys = alloc_percpu(struct action_flow_keys);
1494 if (!flow_keys) {
1495 free_percpu(action_fifos);
1496 return -ENOMEM;
1497 }
1498
1499 return 0;
1500}
1501
1502void action_fifos_exit(void)
1503{
1504 free_percpu(action_fifos);
1505 free_percpu(flow_keys);
1506}
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 vlan_tci;
66 __be16 vlan_proto;
67 unsigned int l2_len;
68 u8 l2_data[MAX_L2_LEN];
69};
70
71static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
72
73#define DEFERRED_ACTION_FIFO_SIZE 10
74struct action_fifo {
75 int head;
76 int tail;
77 /* Deferred action fifo queue storage. */
78 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
79};
80
81static struct action_fifo __percpu *action_fifos;
82static DEFINE_PER_CPU(int, exec_actions_level);
83
84static void action_fifo_init(struct action_fifo *fifo)
85{
86 fifo->head = 0;
87 fifo->tail = 0;
88}
89
90static bool action_fifo_is_empty(const struct action_fifo *fifo)
91{
92 return (fifo->head == fifo->tail);
93}
94
95static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
96{
97 if (action_fifo_is_empty(fifo))
98 return NULL;
99
100 return &fifo->fifo[fifo->tail++];
101}
102
103static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
104{
105 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
106 return NULL;
107
108 return &fifo->fifo[fifo->head++];
109}
110
111/* Return true if fifo is not full */
112static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
113 const struct sw_flow_key *key,
114 const struct nlattr *attr)
115{
116 struct action_fifo *fifo;
117 struct deferred_action *da;
118
119 fifo = this_cpu_ptr(action_fifos);
120 da = action_fifo_put(fifo);
121 if (da) {
122 da->skb = skb;
123 da->actions = attr;
124 da->pkt_key = *key;
125 }
126
127 return da;
128}
129
130static void invalidate_flow_key(struct sw_flow_key *key)
131{
132 key->eth.type = htons(0);
133}
134
135static bool is_flow_key_valid(const struct sw_flow_key *key)
136{
137 return !!key->eth.type;
138}
139
140static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
141 const struct ovs_action_push_mpls *mpls)
142{
143 __be32 *new_mpls_lse;
144 struct ethhdr *hdr;
145
146 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
147 if (skb->encapsulation)
148 return -ENOTSUPP;
149
150 if (skb_cow_head(skb, MPLS_HLEN) < 0)
151 return -ENOMEM;
152
153 skb_push(skb, MPLS_HLEN);
154 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
155 skb->mac_len);
156 skb_reset_mac_header(skb);
157
158 new_mpls_lse = (__be32 *)skb_mpls_header(skb);
159 *new_mpls_lse = mpls->mpls_lse;
160
161 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
162
163 hdr = eth_hdr(skb);
164 hdr->h_proto = mpls->mpls_ethertype;
165
166 if (!skb->inner_protocol)
167 skb_set_inner_protocol(skb, skb->protocol);
168 skb->protocol = mpls->mpls_ethertype;
169
170 invalidate_flow_key(key);
171 return 0;
172}
173
174static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
175 const __be16 ethertype)
176{
177 struct ethhdr *hdr;
178 int err;
179
180 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
181 if (unlikely(err))
182 return err;
183
184 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
185
186 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
187 skb->mac_len);
188
189 __skb_pull(skb, MPLS_HLEN);
190 skb_reset_mac_header(skb);
191
192 /* skb_mpls_header() is used to locate the ethertype
193 * field correctly in the presence of VLAN tags.
194 */
195 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
196 hdr->h_proto = ethertype;
197 if (eth_p_mpls(skb->protocol))
198 skb->protocol = ethertype;
199
200 invalidate_flow_key(key);
201 return 0;
202}
203
204static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
205 const __be32 *mpls_lse, const __be32 *mask)
206{
207 __be32 *stack;
208 __be32 lse;
209 int err;
210
211 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
212 if (unlikely(err))
213 return err;
214
215 stack = (__be32 *)skb_mpls_header(skb);
216 lse = OVS_MASKED(*stack, *mpls_lse, *mask);
217 if (skb->ip_summed == CHECKSUM_COMPLETE) {
218 __be32 diff[] = { ~(*stack), lse };
219
220 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
221 ~skb->csum);
222 }
223
224 *stack = lse;
225 flow_key->mpls.top_lse = lse;
226 return 0;
227}
228
229static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
230{
231 int err;
232
233 err = skb_vlan_pop(skb);
234 if (skb_vlan_tag_present(skb))
235 invalidate_flow_key(key);
236 else
237 key->eth.tci = 0;
238 return err;
239}
240
241static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
242 const struct ovs_action_push_vlan *vlan)
243{
244 if (skb_vlan_tag_present(skb))
245 invalidate_flow_key(key);
246 else
247 key->eth.tci = vlan->vlan_tci;
248 return skb_vlan_push(skb, vlan->vlan_tpid,
249 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
250}
251
252/* 'src' is already properly masked. */
253static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
254{
255 u16 *dst = (u16 *)dst_;
256 const u16 *src = (const u16 *)src_;
257 const u16 *mask = (const u16 *)mask_;
258
259 OVS_SET_MASKED(dst[0], src[0], mask[0]);
260 OVS_SET_MASKED(dst[1], src[1], mask[1]);
261 OVS_SET_MASKED(dst[2], src[2], mask[2]);
262}
263
264static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
265 const struct ovs_key_ethernet *key,
266 const struct ovs_key_ethernet *mask)
267{
268 int err;
269
270 err = skb_ensure_writable(skb, ETH_HLEN);
271 if (unlikely(err))
272 return err;
273
274 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
275
276 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
277 mask->eth_src);
278 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
279 mask->eth_dst);
280
281 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
282
283 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
284 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
285 return 0;
286}
287
288static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
289 __be32 addr, __be32 new_addr)
290{
291 int transport_len = skb->len - skb_transport_offset(skb);
292
293 if (nh->frag_off & htons(IP_OFFSET))
294 return;
295
296 if (nh->protocol == IPPROTO_TCP) {
297 if (likely(transport_len >= sizeof(struct tcphdr)))
298 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
299 addr, new_addr, true);
300 } else if (nh->protocol == IPPROTO_UDP) {
301 if (likely(transport_len >= sizeof(struct udphdr))) {
302 struct udphdr *uh = udp_hdr(skb);
303
304 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
305 inet_proto_csum_replace4(&uh->check, skb,
306 addr, new_addr, true);
307 if (!uh->check)
308 uh->check = CSUM_MANGLED_0;
309 }
310 }
311 }
312}
313
314static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
315 __be32 *addr, __be32 new_addr)
316{
317 update_ip_l4_checksum(skb, nh, *addr, new_addr);
318 csum_replace4(&nh->check, *addr, new_addr);
319 skb_clear_hash(skb);
320 *addr = new_addr;
321}
322
323static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
324 __be32 addr[4], const __be32 new_addr[4])
325{
326 int transport_len = skb->len - skb_transport_offset(skb);
327
328 if (l4_proto == NEXTHDR_TCP) {
329 if (likely(transport_len >= sizeof(struct tcphdr)))
330 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
331 addr, new_addr, true);
332 } else if (l4_proto == NEXTHDR_UDP) {
333 if (likely(transport_len >= sizeof(struct udphdr))) {
334 struct udphdr *uh = udp_hdr(skb);
335
336 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
337 inet_proto_csum_replace16(&uh->check, skb,
338 addr, new_addr, true);
339 if (!uh->check)
340 uh->check = CSUM_MANGLED_0;
341 }
342 }
343 } else if (l4_proto == NEXTHDR_ICMP) {
344 if (likely(transport_len >= sizeof(struct icmp6hdr)))
345 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
346 skb, addr, new_addr, true);
347 }
348}
349
350static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
351 const __be32 mask[4], __be32 masked[4])
352{
353 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
354 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
355 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
356 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
357}
358
359static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
360 __be32 addr[4], const __be32 new_addr[4],
361 bool recalculate_csum)
362{
363 if (recalculate_csum)
364 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
365
366 skb_clear_hash(skb);
367 memcpy(addr, new_addr, sizeof(__be32[4]));
368}
369
370static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
371{
372 /* Bits 21-24 are always unmasked, so this retains their values. */
373 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
374 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
375 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
376}
377
378static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
379 u8 mask)
380{
381 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
382
383 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
384 nh->ttl = new_ttl;
385}
386
387static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
388 const struct ovs_key_ipv4 *key,
389 const struct ovs_key_ipv4 *mask)
390{
391 struct iphdr *nh;
392 __be32 new_addr;
393 int err;
394
395 err = skb_ensure_writable(skb, skb_network_offset(skb) +
396 sizeof(struct iphdr));
397 if (unlikely(err))
398 return err;
399
400 nh = ip_hdr(skb);
401
402 /* Setting an IP addresses is typically only a side effect of
403 * matching on them in the current userspace implementation, so it
404 * makes sense to check if the value actually changed.
405 */
406 if (mask->ipv4_src) {
407 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
408
409 if (unlikely(new_addr != nh->saddr)) {
410 set_ip_addr(skb, nh, &nh->saddr, new_addr);
411 flow_key->ipv4.addr.src = new_addr;
412 }
413 }
414 if (mask->ipv4_dst) {
415 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
416
417 if (unlikely(new_addr != nh->daddr)) {
418 set_ip_addr(skb, nh, &nh->daddr, new_addr);
419 flow_key->ipv4.addr.dst = new_addr;
420 }
421 }
422 if (mask->ipv4_tos) {
423 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
424 flow_key->ip.tos = nh->tos;
425 }
426 if (mask->ipv4_ttl) {
427 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
428 flow_key->ip.ttl = nh->ttl;
429 }
430
431 return 0;
432}
433
434static bool is_ipv6_mask_nonzero(const __be32 addr[4])
435{
436 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
437}
438
439static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
440 const struct ovs_key_ipv6 *key,
441 const struct ovs_key_ipv6 *mask)
442{
443 struct ipv6hdr *nh;
444 int err;
445
446 err = skb_ensure_writable(skb, skb_network_offset(skb) +
447 sizeof(struct ipv6hdr));
448 if (unlikely(err))
449 return err;
450
451 nh = ipv6_hdr(skb);
452
453 /* Setting an IP addresses is typically only a side effect of
454 * matching on them in the current userspace implementation, so it
455 * makes sense to check if the value actually changed.
456 */
457 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
458 __be32 *saddr = (__be32 *)&nh->saddr;
459 __be32 masked[4];
460
461 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
462
463 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
464 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
465 true);
466 memcpy(&flow_key->ipv6.addr.src, masked,
467 sizeof(flow_key->ipv6.addr.src));
468 }
469 }
470 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
471 unsigned int offset = 0;
472 int flags = IP6_FH_F_SKIP_RH;
473 bool recalc_csum = true;
474 __be32 *daddr = (__be32 *)&nh->daddr;
475 __be32 masked[4];
476
477 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
478
479 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
480 if (ipv6_ext_hdr(nh->nexthdr))
481 recalc_csum = (ipv6_find_hdr(skb, &offset,
482 NEXTHDR_ROUTING,
483 NULL, &flags)
484 != NEXTHDR_ROUTING);
485
486 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
487 recalc_csum);
488 memcpy(&flow_key->ipv6.addr.dst, masked,
489 sizeof(flow_key->ipv6.addr.dst));
490 }
491 }
492 if (mask->ipv6_tclass) {
493 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
494 flow_key->ip.tos = ipv6_get_dsfield(nh);
495 }
496 if (mask->ipv6_label) {
497 set_ipv6_fl(nh, ntohl(key->ipv6_label),
498 ntohl(mask->ipv6_label));
499 flow_key->ipv6.label =
500 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
501 }
502 if (mask->ipv6_hlimit) {
503 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
504 mask->ipv6_hlimit);
505 flow_key->ip.ttl = nh->hop_limit;
506 }
507 return 0;
508}
509
510/* Must follow skb_ensure_writable() since that can move the skb data. */
511static void set_tp_port(struct sk_buff *skb, __be16 *port,
512 __be16 new_port, __sum16 *check)
513{
514 inet_proto_csum_replace2(check, skb, *port, new_port, false);
515 *port = new_port;
516}
517
518static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
519 const struct ovs_key_udp *key,
520 const struct ovs_key_udp *mask)
521{
522 struct udphdr *uh;
523 __be16 src, dst;
524 int err;
525
526 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
527 sizeof(struct udphdr));
528 if (unlikely(err))
529 return err;
530
531 uh = udp_hdr(skb);
532 /* Either of the masks is non-zero, so do not bother checking them. */
533 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
534 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
535
536 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
537 if (likely(src != uh->source)) {
538 set_tp_port(skb, &uh->source, src, &uh->check);
539 flow_key->tp.src = src;
540 }
541 if (likely(dst != uh->dest)) {
542 set_tp_port(skb, &uh->dest, dst, &uh->check);
543 flow_key->tp.dst = dst;
544 }
545
546 if (unlikely(!uh->check))
547 uh->check = CSUM_MANGLED_0;
548 } else {
549 uh->source = src;
550 uh->dest = dst;
551 flow_key->tp.src = src;
552 flow_key->tp.dst = dst;
553 }
554
555 skb_clear_hash(skb);
556
557 return 0;
558}
559
560static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
561 const struct ovs_key_tcp *key,
562 const struct ovs_key_tcp *mask)
563{
564 struct tcphdr *th;
565 __be16 src, dst;
566 int err;
567
568 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
569 sizeof(struct tcphdr));
570 if (unlikely(err))
571 return err;
572
573 th = tcp_hdr(skb);
574 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
575 if (likely(src != th->source)) {
576 set_tp_port(skb, &th->source, src, &th->check);
577 flow_key->tp.src = src;
578 }
579 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
580 if (likely(dst != th->dest)) {
581 set_tp_port(skb, &th->dest, dst, &th->check);
582 flow_key->tp.dst = dst;
583 }
584 skb_clear_hash(skb);
585
586 return 0;
587}
588
589static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
590 const struct ovs_key_sctp *key,
591 const struct ovs_key_sctp *mask)
592{
593 unsigned int sctphoff = skb_transport_offset(skb);
594 struct sctphdr *sh;
595 __le32 old_correct_csum, new_csum, old_csum;
596 int err;
597
598 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
599 if (unlikely(err))
600 return err;
601
602 sh = sctp_hdr(skb);
603 old_csum = sh->checksum;
604 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
605
606 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
607 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
608
609 new_csum = sctp_compute_cksum(skb, sctphoff);
610
611 /* Carry any checksum errors through. */
612 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
613
614 skb_clear_hash(skb);
615 flow_key->tp.src = sh->source;
616 flow_key->tp.dst = sh->dest;
617
618 return 0;
619}
620
621static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
622{
623 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
624 struct vport *vport = data->vport;
625
626 if (skb_cow_head(skb, data->l2_len) < 0) {
627 kfree_skb(skb);
628 return -ENOMEM;
629 }
630
631 __skb_dst_copy(skb, data->dst);
632 *OVS_CB(skb) = data->cb;
633 skb->inner_protocol = data->inner_protocol;
634 skb->vlan_tci = data->vlan_tci;
635 skb->vlan_proto = data->vlan_proto;
636
637 /* Reconstruct the MAC header. */
638 skb_push(skb, data->l2_len);
639 memcpy(skb->data, &data->l2_data, data->l2_len);
640 skb_postpush_rcsum(skb, skb->data, data->l2_len);
641 skb_reset_mac_header(skb);
642
643 ovs_vport_send(vport, skb);
644 return 0;
645}
646
647static unsigned int
648ovs_dst_get_mtu(const struct dst_entry *dst)
649{
650 return dst->dev->mtu;
651}
652
653static struct dst_ops ovs_dst_ops = {
654 .family = AF_UNSPEC,
655 .mtu = ovs_dst_get_mtu,
656};
657
658/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
659 * ovs_vport_output(), which is called once per fragmented packet.
660 */
661static void prepare_frag(struct vport *vport, struct sk_buff *skb)
662{
663 unsigned int hlen = skb_network_offset(skb);
664 struct ovs_frag_data *data;
665
666 data = this_cpu_ptr(&ovs_frag_data_storage);
667 data->dst = skb->_skb_refdst;
668 data->vport = vport;
669 data->cb = *OVS_CB(skb);
670 data->inner_protocol = skb->inner_protocol;
671 data->vlan_tci = skb->vlan_tci;
672 data->vlan_proto = skb->vlan_proto;
673 data->l2_len = hlen;
674 memcpy(&data->l2_data, skb->data, hlen);
675
676 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
677 skb_pull(skb, hlen);
678}
679
680static void ovs_fragment(struct net *net, struct vport *vport,
681 struct sk_buff *skb, u16 mru, __be16 ethertype)
682{
683 if (skb_network_offset(skb) > MAX_L2_LEN) {
684 OVS_NLERR(1, "L2 header too long to fragment");
685 goto err;
686 }
687
688 if (ethertype == htons(ETH_P_IP)) {
689 struct dst_entry ovs_dst;
690 unsigned long orig_dst;
691
692 prepare_frag(vport, skb);
693 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
694 DST_OBSOLETE_NONE, DST_NOCOUNT);
695 ovs_dst.dev = vport->dev;
696
697 orig_dst = skb->_skb_refdst;
698 skb_dst_set_noref(skb, &ovs_dst);
699 IPCB(skb)->frag_max_size = mru;
700
701 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
702 refdst_drop(orig_dst);
703 } else if (ethertype == htons(ETH_P_IPV6)) {
704 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
705 unsigned long orig_dst;
706 struct rt6_info ovs_rt;
707
708 if (!v6ops) {
709 goto err;
710 }
711
712 prepare_frag(vport, skb);
713 memset(&ovs_rt, 0, sizeof(ovs_rt));
714 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
715 DST_OBSOLETE_NONE, DST_NOCOUNT);
716 ovs_rt.dst.dev = vport->dev;
717
718 orig_dst = skb->_skb_refdst;
719 skb_dst_set_noref(skb, &ovs_rt.dst);
720 IP6CB(skb)->frag_max_size = mru;
721
722 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
723 refdst_drop(orig_dst);
724 } else {
725 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
726 ovs_vport_name(vport), ntohs(ethertype), mru,
727 vport->dev->mtu);
728 goto err;
729 }
730
731 return;
732err:
733 kfree_skb(skb);
734}
735
736static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
737 struct sw_flow_key *key)
738{
739 struct vport *vport = ovs_vport_rcu(dp, out_port);
740
741 if (likely(vport)) {
742 u16 mru = OVS_CB(skb)->mru;
743
744 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
745 ovs_vport_send(vport, skb);
746 } else if (mru <= vport->dev->mtu) {
747 struct net *net = read_pnet(&dp->net);
748 __be16 ethertype = key->eth.type;
749
750 if (!is_flow_key_valid(key)) {
751 if (eth_p_mpls(skb->protocol))
752 ethertype = skb->inner_protocol;
753 else
754 ethertype = vlan_get_protocol(skb);
755 }
756
757 ovs_fragment(net, vport, skb, mru, ethertype);
758 } else {
759 kfree_skb(skb);
760 }
761 } else {
762 kfree_skb(skb);
763 }
764}
765
766static int output_userspace(struct datapath *dp, struct sk_buff *skb,
767 struct sw_flow_key *key, const struct nlattr *attr,
768 const struct nlattr *actions, int actions_len)
769{
770 struct dp_upcall_info upcall;
771 const struct nlattr *a;
772 int rem;
773
774 memset(&upcall, 0, sizeof(upcall));
775 upcall.cmd = OVS_PACKET_CMD_ACTION;
776 upcall.mru = OVS_CB(skb)->mru;
777
778 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
779 a = nla_next(a, &rem)) {
780 switch (nla_type(a)) {
781 case OVS_USERSPACE_ATTR_USERDATA:
782 upcall.userdata = a;
783 break;
784
785 case OVS_USERSPACE_ATTR_PID:
786 upcall.portid = nla_get_u32(a);
787 break;
788
789 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
790 /* Get out tunnel info. */
791 struct vport *vport;
792
793 vport = ovs_vport_rcu(dp, nla_get_u32(a));
794 if (vport) {
795 int err;
796
797 err = dev_fill_metadata_dst(vport->dev, skb);
798 if (!err)
799 upcall.egress_tun_info = skb_tunnel_info(skb);
800 }
801
802 break;
803 }
804
805 case OVS_USERSPACE_ATTR_ACTIONS: {
806 /* Include actions. */
807 upcall.actions = actions;
808 upcall.actions_len = actions_len;
809 break;
810 }
811
812 } /* End of switch. */
813 }
814
815 return ovs_dp_upcall(dp, skb, key, &upcall);
816}
817
818static int sample(struct datapath *dp, struct sk_buff *skb,
819 struct sw_flow_key *key, const struct nlattr *attr,
820 const struct nlattr *actions, int actions_len)
821{
822 const struct nlattr *acts_list = NULL;
823 const struct nlattr *a;
824 int rem;
825
826 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
827 a = nla_next(a, &rem)) {
828 u32 probability;
829
830 switch (nla_type(a)) {
831 case OVS_SAMPLE_ATTR_PROBABILITY:
832 probability = nla_get_u32(a);
833 if (!probability || prandom_u32() > probability)
834 return 0;
835 break;
836
837 case OVS_SAMPLE_ATTR_ACTIONS:
838 acts_list = a;
839 break;
840 }
841 }
842
843 rem = nla_len(acts_list);
844 a = nla_data(acts_list);
845
846 /* Actions list is empty, do nothing */
847 if (unlikely(!rem))
848 return 0;
849
850 /* The only known usage of sample action is having a single user-space
851 * action. Treat this usage as a special case.
852 * The output_userspace() should clone the skb to be sent to the
853 * user space. This skb will be consumed by its caller.
854 */
855 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
856 nla_is_last(a, rem)))
857 return output_userspace(dp, skb, key, a, actions, actions_len);
858
859 skb = skb_clone(skb, GFP_ATOMIC);
860 if (!skb)
861 /* Skip the sample action when out of memory. */
862 return 0;
863
864 if (!add_deferred_actions(skb, key, a)) {
865 if (net_ratelimit())
866 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
867 ovs_dp_name(dp));
868
869 kfree_skb(skb);
870 }
871 return 0;
872}
873
874static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
875 const struct nlattr *attr)
876{
877 struct ovs_action_hash *hash_act = nla_data(attr);
878 u32 hash = 0;
879
880 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
881 hash = skb_get_hash(skb);
882 hash = jhash_1word(hash, hash_act->hash_basis);
883 if (!hash)
884 hash = 0x1;
885
886 key->ovs_flow_hash = hash;
887}
888
889static int execute_set_action(struct sk_buff *skb,
890 struct sw_flow_key *flow_key,
891 const struct nlattr *a)
892{
893 /* Only tunnel set execution is supported without a mask. */
894 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
895 struct ovs_tunnel_info *tun = nla_data(a);
896
897 skb_dst_drop(skb);
898 dst_hold((struct dst_entry *)tun->tun_dst);
899 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
900 return 0;
901 }
902
903 return -EINVAL;
904}
905
906/* Mask is at the midpoint of the data. */
907#define get_mask(a, type) ((const type)nla_data(a) + 1)
908
909static int execute_masked_set_action(struct sk_buff *skb,
910 struct sw_flow_key *flow_key,
911 const struct nlattr *a)
912{
913 int err = 0;
914
915 switch (nla_type(a)) {
916 case OVS_KEY_ATTR_PRIORITY:
917 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
918 *get_mask(a, u32 *));
919 flow_key->phy.priority = skb->priority;
920 break;
921
922 case OVS_KEY_ATTR_SKB_MARK:
923 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
924 flow_key->phy.skb_mark = skb->mark;
925 break;
926
927 case OVS_KEY_ATTR_TUNNEL_INFO:
928 /* Masked data not supported for tunnel. */
929 err = -EINVAL;
930 break;
931
932 case OVS_KEY_ATTR_ETHERNET:
933 err = set_eth_addr(skb, flow_key, nla_data(a),
934 get_mask(a, struct ovs_key_ethernet *));
935 break;
936
937 case OVS_KEY_ATTR_IPV4:
938 err = set_ipv4(skb, flow_key, nla_data(a),
939 get_mask(a, struct ovs_key_ipv4 *));
940 break;
941
942 case OVS_KEY_ATTR_IPV6:
943 err = set_ipv6(skb, flow_key, nla_data(a),
944 get_mask(a, struct ovs_key_ipv6 *));
945 break;
946
947 case OVS_KEY_ATTR_TCP:
948 err = set_tcp(skb, flow_key, nla_data(a),
949 get_mask(a, struct ovs_key_tcp *));
950 break;
951
952 case OVS_KEY_ATTR_UDP:
953 err = set_udp(skb, flow_key, nla_data(a),
954 get_mask(a, struct ovs_key_udp *));
955 break;
956
957 case OVS_KEY_ATTR_SCTP:
958 err = set_sctp(skb, flow_key, nla_data(a),
959 get_mask(a, struct ovs_key_sctp *));
960 break;
961
962 case OVS_KEY_ATTR_MPLS:
963 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
964 __be32 *));
965 break;
966
967 case OVS_KEY_ATTR_CT_STATE:
968 case OVS_KEY_ATTR_CT_ZONE:
969 case OVS_KEY_ATTR_CT_MARK:
970 case OVS_KEY_ATTR_CT_LABELS:
971 err = -EINVAL;
972 break;
973 }
974
975 return err;
976}
977
978static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
979 struct sw_flow_key *key,
980 const struct nlattr *a, int rem)
981{
982 struct deferred_action *da;
983
984 if (!is_flow_key_valid(key)) {
985 int err;
986
987 err = ovs_flow_key_update(skb, key);
988 if (err)
989 return err;
990 }
991 BUG_ON(!is_flow_key_valid(key));
992
993 if (!nla_is_last(a, rem)) {
994 /* Recirc action is the not the last action
995 * of the action list, need to clone the skb.
996 */
997 skb = skb_clone(skb, GFP_ATOMIC);
998
999 /* Skip the recirc action when out of memory, but
1000 * continue on with the rest of the action list.
1001 */
1002 if (!skb)
1003 return 0;
1004 }
1005
1006 da = add_deferred_actions(skb, key, NULL);
1007 if (da) {
1008 da->pkt_key.recirc_id = nla_get_u32(a);
1009 } else {
1010 kfree_skb(skb);
1011
1012 if (net_ratelimit())
1013 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1014 ovs_dp_name(dp));
1015 }
1016
1017 return 0;
1018}
1019
1020/* Execute a list of actions against 'skb'. */
1021static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1022 struct sw_flow_key *key,
1023 const struct nlattr *attr, int len)
1024{
1025 /* Every output action needs a separate clone of 'skb', but the common
1026 * case is just a single output action, so that doing a clone and
1027 * then freeing the original skbuff is wasteful. So the following code
1028 * is slightly obscure just to avoid that.
1029 */
1030 int prev_port = -1;
1031 const struct nlattr *a;
1032 int rem;
1033
1034 for (a = attr, rem = len; rem > 0;
1035 a = nla_next(a, &rem)) {
1036 int err = 0;
1037
1038 if (unlikely(prev_port != -1)) {
1039 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1040
1041 if (out_skb)
1042 do_output(dp, out_skb, prev_port, key);
1043
1044 prev_port = -1;
1045 }
1046
1047 switch (nla_type(a)) {
1048 case OVS_ACTION_ATTR_OUTPUT:
1049 prev_port = nla_get_u32(a);
1050 break;
1051
1052 case OVS_ACTION_ATTR_USERSPACE:
1053 output_userspace(dp, skb, key, a, attr, len);
1054 break;
1055
1056 case OVS_ACTION_ATTR_HASH:
1057 execute_hash(skb, key, a);
1058 break;
1059
1060 case OVS_ACTION_ATTR_PUSH_MPLS:
1061 err = push_mpls(skb, key, nla_data(a));
1062 break;
1063
1064 case OVS_ACTION_ATTR_POP_MPLS:
1065 err = pop_mpls(skb, key, nla_get_be16(a));
1066 break;
1067
1068 case OVS_ACTION_ATTR_PUSH_VLAN:
1069 err = push_vlan(skb, key, nla_data(a));
1070 break;
1071
1072 case OVS_ACTION_ATTR_POP_VLAN:
1073 err = pop_vlan(skb, key);
1074 break;
1075
1076 case OVS_ACTION_ATTR_RECIRC:
1077 err = execute_recirc(dp, skb, key, a, rem);
1078 if (nla_is_last(a, rem)) {
1079 /* If this is the last action, the skb has
1080 * been consumed or freed.
1081 * Return immediately.
1082 */
1083 return err;
1084 }
1085 break;
1086
1087 case OVS_ACTION_ATTR_SET:
1088 err = execute_set_action(skb, key, nla_data(a));
1089 break;
1090
1091 case OVS_ACTION_ATTR_SET_MASKED:
1092 case OVS_ACTION_ATTR_SET_TO_MASKED:
1093 err = execute_masked_set_action(skb, key, nla_data(a));
1094 break;
1095
1096 case OVS_ACTION_ATTR_SAMPLE:
1097 err = sample(dp, skb, key, a, attr, len);
1098 break;
1099
1100 case OVS_ACTION_ATTR_CT:
1101 if (!is_flow_key_valid(key)) {
1102 err = ovs_flow_key_update(skb, key);
1103 if (err)
1104 return err;
1105 }
1106
1107 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1108 nla_data(a));
1109
1110 /* Hide stolen IP fragments from user space. */
1111 if (err)
1112 return err == -EINPROGRESS ? 0 : err;
1113 break;
1114 }
1115
1116 if (unlikely(err)) {
1117 kfree_skb(skb);
1118 return err;
1119 }
1120 }
1121
1122 if (prev_port != -1)
1123 do_output(dp, skb, prev_port, key);
1124 else
1125 consume_skb(skb);
1126
1127 return 0;
1128}
1129
1130static void process_deferred_actions(struct datapath *dp)
1131{
1132 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1133
1134 /* Do not touch the FIFO in case there is no deferred actions. */
1135 if (action_fifo_is_empty(fifo))
1136 return;
1137
1138 /* Finishing executing all deferred actions. */
1139 do {
1140 struct deferred_action *da = action_fifo_get(fifo);
1141 struct sk_buff *skb = da->skb;
1142 struct sw_flow_key *key = &da->pkt_key;
1143 const struct nlattr *actions = da->actions;
1144
1145 if (actions)
1146 do_execute_actions(dp, skb, key, actions,
1147 nla_len(actions));
1148 else
1149 ovs_dp_process_packet(skb, key);
1150 } while (!action_fifo_is_empty(fifo));
1151
1152 /* Reset FIFO for the next packet. */
1153 action_fifo_init(fifo);
1154}
1155
1156/* Execute a list of actions against 'skb'. */
1157int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1158 const struct sw_flow_actions *acts,
1159 struct sw_flow_key *key)
1160{
1161 static const int ovs_recursion_limit = 5;
1162 int err, level;
1163
1164 level = __this_cpu_inc_return(exec_actions_level);
1165 if (unlikely(level > ovs_recursion_limit)) {
1166 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1167 ovs_dp_name(dp));
1168 kfree_skb(skb);
1169 err = -ENETDOWN;
1170 goto out;
1171 }
1172
1173 err = do_execute_actions(dp, skb, key,
1174 acts->actions, acts->actions_len);
1175
1176 if (level == 1)
1177 process_deferred_actions(dp);
1178
1179out:
1180 __this_cpu_dec(exec_actions_level);
1181 return err;
1182}
1183
1184int action_fifos_init(void)
1185{
1186 action_fifos = alloc_percpu(struct action_fifo);
1187 if (!action_fifos)
1188 return -ENOMEM;
1189
1190 return 0;
1191}
1192
1193void action_fifos_exit(void)
1194{
1195 free_percpu(action_fifos);
1196}