projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branches 'x86/amd', 'x86/vt-d', 'arm/exynos', 'arm/mediatek' and 'arm/renesas...
[deliverable/linux.git] / net / openvswitch / actions.c
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
47 static 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
51 struct 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)
60 struct 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
71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
72
73 #define DEFERRED_ACTION_FIFO_SIZE 10
74 struct 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
81 static struct action_fifo __percpu *action_fifos;
82 static DEFINE_PER_CPU(int, exec_actions_level);
83
84 static void action_fifo_init(struct action_fifo *fifo)
85 {
86 fifo->head = 0;
87 fifo->tail = 0;
88 }
89
90 static bool action_fifo_is_empty(const struct action_fifo *fifo)
91 {
92 return (fifo->head == fifo->tail);
93 }
94
95 static 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
103 static 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 */
112 static 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
130 static void invalidate_flow_key(struct sw_flow_key *key)
131 {
132 key->eth.type = htons(0);
133 }
134
135 static bool is_flow_key_valid(const struct sw_flow_key *key)
136 {
137 return !!key->eth.type;
138 }
139
140 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
141 __be16 ethertype)
142 {
143 if (skb->ip_summed == CHECKSUM_COMPLETE) {
144 __be16 diff[] = { ~(hdr->h_proto), ethertype };
145
146 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
147 ~skb->csum);
148 }
149
150 hdr->h_proto = ethertype;
151 }
152
153 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
154 const struct ovs_action_push_mpls *mpls)
155 {
156 __be32 *new_mpls_lse;
157
158 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
159 if (skb->encapsulation)
160 return -ENOTSUPP;
161
162 if (skb_cow_head(skb, MPLS_HLEN) < 0)
163 return -ENOMEM;
164
165 skb_push(skb, MPLS_HLEN);
166 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
167 skb->mac_len);
168 skb_reset_mac_header(skb);
169
170 new_mpls_lse = (__be32 *)skb_mpls_header(skb);
171 *new_mpls_lse = mpls->mpls_lse;
172
173 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
174
175 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
176 if (!skb->inner_protocol)
177 skb_set_inner_protocol(skb, skb->protocol);
178 skb->protocol = mpls->mpls_ethertype;
179
180 invalidate_flow_key(key);
181 return 0;
182 }
183
184 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
185 const __be16 ethertype)
186 {
187 struct ethhdr *hdr;
188 int err;
189
190 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
191 if (unlikely(err))
192 return err;
193
194 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
195
196 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
197 skb->mac_len);
198
199 __skb_pull(skb, MPLS_HLEN);
200 skb_reset_mac_header(skb);
201
202 /* skb_mpls_header() is used to locate the ethertype
203 * field correctly in the presence of VLAN tags.
204 */
205 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
206 update_ethertype(skb, hdr, ethertype);
207 if (eth_p_mpls(skb->protocol))
208 skb->protocol = ethertype;
209
210 invalidate_flow_key(key);
211 return 0;
212 }
213
214 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
215 const __be32 *mpls_lse, const __be32 *mask)
216 {
217 __be32 *stack;
218 __be32 lse;
219 int err;
220
221 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
222 if (unlikely(err))
223 return err;
224
225 stack = (__be32 *)skb_mpls_header(skb);
226 lse = OVS_MASKED(*stack, *mpls_lse, *mask);
227 if (skb->ip_summed == CHECKSUM_COMPLETE) {
228 __be32 diff[] = { ~(*stack), lse };
229
230 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
231 ~skb->csum);
232 }
233
234 *stack = lse;
235 flow_key->mpls.top_lse = lse;
236 return 0;
237 }
238
239 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
240 {
241 int err;
242
243 err = skb_vlan_pop(skb);
244 if (skb_vlan_tag_present(skb))
245 invalidate_flow_key(key);
246 else
247 key->eth.tci = 0;
248 return err;
249 }
250
251 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
252 const struct ovs_action_push_vlan *vlan)
253 {
254 if (skb_vlan_tag_present(skb))
255 invalidate_flow_key(key);
256 else
257 key->eth.tci = vlan->vlan_tci;
258 return skb_vlan_push(skb, vlan->vlan_tpid,
259 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
260 }
261
262 /* 'src' is already properly masked. */
263 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
264 {
265 u16 *dst = (u16 *)dst_;
266 const u16 *src = (const u16 *)src_;
267 const u16 *mask = (const u16 *)mask_;
268
269 OVS_SET_MASKED(dst[0], src[0], mask[0]);
270 OVS_SET_MASKED(dst[1], src[1], mask[1]);
271 OVS_SET_MASKED(dst[2], src[2], mask[2]);
272 }
273
274 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
275 const struct ovs_key_ethernet *key,
276 const struct ovs_key_ethernet *mask)
277 {
278 int err;
279
280 err = skb_ensure_writable(skb, ETH_HLEN);
281 if (unlikely(err))
282 return err;
283
284 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
285
286 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
287 mask->eth_src);
288 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
289 mask->eth_dst);
290
291 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
292
293 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
294 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
295 return 0;
296 }
297
298 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
299 __be32 addr, __be32 new_addr)
300 {
301 int transport_len = skb->len - skb_transport_offset(skb);
302
303 if (nh->frag_off & htons(IP_OFFSET))
304 return;
305
306 if (nh->protocol == IPPROTO_TCP) {
307 if (likely(transport_len >= sizeof(struct tcphdr)))
308 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
309 addr, new_addr, true);
310 } else if (nh->protocol == IPPROTO_UDP) {
311 if (likely(transport_len >= sizeof(struct udphdr))) {
312 struct udphdr *uh = udp_hdr(skb);
313
314 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
315 inet_proto_csum_replace4(&uh->check, skb,
316 addr, new_addr, true);
317 if (!uh->check)
318 uh->check = CSUM_MANGLED_0;
319 }
320 }
321 }
322 }
323
324 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
325 __be32 *addr, __be32 new_addr)
326 {
327 update_ip_l4_checksum(skb, nh, *addr, new_addr);
328 csum_replace4(&nh->check, *addr, new_addr);
329 skb_clear_hash(skb);
330 *addr = new_addr;
331 }
332
333 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
334 __be32 addr[4], const __be32 new_addr[4])
335 {
336 int transport_len = skb->len - skb_transport_offset(skb);
337
338 if (l4_proto == NEXTHDR_TCP) {
339 if (likely(transport_len >= sizeof(struct tcphdr)))
340 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
341 addr, new_addr, true);
342 } else if (l4_proto == NEXTHDR_UDP) {
343 if (likely(transport_len >= sizeof(struct udphdr))) {
344 struct udphdr *uh = udp_hdr(skb);
345
346 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
347 inet_proto_csum_replace16(&uh->check, skb,
348 addr, new_addr, true);
349 if (!uh->check)
350 uh->check = CSUM_MANGLED_0;
351 }
352 }
353 } else if (l4_proto == NEXTHDR_ICMP) {
354 if (likely(transport_len >= sizeof(struct icmp6hdr)))
355 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
356 skb, addr, new_addr, true);
357 }
358 }
359
360 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
361 const __be32 mask[4], __be32 masked[4])
362 {
363 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
364 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
365 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
366 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
367 }
368
369 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
370 __be32 addr[4], const __be32 new_addr[4],
371 bool recalculate_csum)
372 {
373 if (recalculate_csum)
374 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
375
376 skb_clear_hash(skb);
377 memcpy(addr, new_addr, sizeof(__be32[4]));
378 }
379
380 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
381 {
382 /* Bits 21-24 are always unmasked, so this retains their values. */
383 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
384 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
385 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
386 }
387
388 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
389 u8 mask)
390 {
391 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
392
393 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
394 nh->ttl = new_ttl;
395 }
396
397 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
398 const struct ovs_key_ipv4 *key,
399 const struct ovs_key_ipv4 *mask)
400 {
401 struct iphdr *nh;
402 __be32 new_addr;
403 int err;
404
405 err = skb_ensure_writable(skb, skb_network_offset(skb) +
406 sizeof(struct iphdr));
407 if (unlikely(err))
408 return err;
409
410 nh = ip_hdr(skb);
411
412 /* Setting an IP addresses is typically only a side effect of
413 * matching on them in the current userspace implementation, so it
414 * makes sense to check if the value actually changed.
415 */
416 if (mask->ipv4_src) {
417 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
418
419 if (unlikely(new_addr != nh->saddr)) {
420 set_ip_addr(skb, nh, &nh->saddr, new_addr);
421 flow_key->ipv4.addr.src = new_addr;
422 }
423 }
424 if (mask->ipv4_dst) {
425 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
426
427 if (unlikely(new_addr != nh->daddr)) {
428 set_ip_addr(skb, nh, &nh->daddr, new_addr);
429 flow_key->ipv4.addr.dst = new_addr;
430 }
431 }
432 if (mask->ipv4_tos) {
433 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
434 flow_key->ip.tos = nh->tos;
435 }
436 if (mask->ipv4_ttl) {
437 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
438 flow_key->ip.ttl = nh->ttl;
439 }
440
441 return 0;
442 }
443
444 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
445 {
446 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
447 }
448
449 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
450 const struct ovs_key_ipv6 *key,
451 const struct ovs_key_ipv6 *mask)
452 {
453 struct ipv6hdr *nh;
454 int err;
455
456 err = skb_ensure_writable(skb, skb_network_offset(skb) +
457 sizeof(struct ipv6hdr));
458 if (unlikely(err))
459 return err;
460
461 nh = ipv6_hdr(skb);
462
463 /* Setting an IP addresses is typically only a side effect of
464 * matching on them in the current userspace implementation, so it
465 * makes sense to check if the value actually changed.
466 */
467 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
468 __be32 *saddr = (__be32 *)&nh->saddr;
469 __be32 masked[4];
470
471 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
472
473 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
474 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
475 true);
476 memcpy(&flow_key->ipv6.addr.src, masked,
477 sizeof(flow_key->ipv6.addr.src));
478 }
479 }
480 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
481 unsigned int offset = 0;
482 int flags = IP6_FH_F_SKIP_RH;
483 bool recalc_csum = true;
484 __be32 *daddr = (__be32 *)&nh->daddr;
485 __be32 masked[4];
486
487 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
488
489 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
490 if (ipv6_ext_hdr(nh->nexthdr))
491 recalc_csum = (ipv6_find_hdr(skb, &offset,
492 NEXTHDR_ROUTING,
493 NULL, &flags)
494 != NEXTHDR_ROUTING);
495
496 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
497 recalc_csum);
498 memcpy(&flow_key->ipv6.addr.dst, masked,
499 sizeof(flow_key->ipv6.addr.dst));
500 }
501 }
502 if (mask->ipv6_tclass) {
503 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
504 flow_key->ip.tos = ipv6_get_dsfield(nh);
505 }
506 if (mask->ipv6_label) {
507 set_ipv6_fl(nh, ntohl(key->ipv6_label),
508 ntohl(mask->ipv6_label));
509 flow_key->ipv6.label =
510 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
511 }
512 if (mask->ipv6_hlimit) {
513 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
514 mask->ipv6_hlimit);
515 flow_key->ip.ttl = nh->hop_limit;
516 }
517 return 0;
518 }
519
520 /* Must follow skb_ensure_writable() since that can move the skb data. */
521 static void set_tp_port(struct sk_buff *skb, __be16 *port,
522 __be16 new_port, __sum16 *check)
523 {
524 inet_proto_csum_replace2(check, skb, *port, new_port, false);
525 *port = new_port;
526 }
527
528 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
529 const struct ovs_key_udp *key,
530 const struct ovs_key_udp *mask)
531 {
532 struct udphdr *uh;
533 __be16 src, dst;
534 int err;
535
536 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
537 sizeof(struct udphdr));
538 if (unlikely(err))
539 return err;
540
541 uh = udp_hdr(skb);
542 /* Either of the masks is non-zero, so do not bother checking them. */
543 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
544 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
545
546 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
547 if (likely(src != uh->source)) {
548 set_tp_port(skb, &uh->source, src, &uh->check);
549 flow_key->tp.src = src;
550 }
551 if (likely(dst != uh->dest)) {
552 set_tp_port(skb, &uh->dest, dst, &uh->check);
553 flow_key->tp.dst = dst;
554 }
555
556 if (unlikely(!uh->check))
557 uh->check = CSUM_MANGLED_0;
558 } else {
559 uh->source = src;
560 uh->dest = dst;
561 flow_key->tp.src = src;
562 flow_key->tp.dst = dst;
563 }
564
565 skb_clear_hash(skb);
566
567 return 0;
568 }
569
570 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
571 const struct ovs_key_tcp *key,
572 const struct ovs_key_tcp *mask)
573 {
574 struct tcphdr *th;
575 __be16 src, dst;
576 int err;
577
578 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
579 sizeof(struct tcphdr));
580 if (unlikely(err))
581 return err;
582
583 th = tcp_hdr(skb);
584 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
585 if (likely(src != th->source)) {
586 set_tp_port(skb, &th->source, src, &th->check);
587 flow_key->tp.src = src;
588 }
589 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
590 if (likely(dst != th->dest)) {
591 set_tp_port(skb, &th->dest, dst, &th->check);
592 flow_key->tp.dst = dst;
593 }
594 skb_clear_hash(skb);
595
596 return 0;
597 }
598
599 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
600 const struct ovs_key_sctp *key,
601 const struct ovs_key_sctp *mask)
602 {
603 unsigned int sctphoff = skb_transport_offset(skb);
604 struct sctphdr *sh;
605 __le32 old_correct_csum, new_csum, old_csum;
606 int err;
607
608 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
609 if (unlikely(err))
610 return err;
611
612 sh = sctp_hdr(skb);
613 old_csum = sh->checksum;
614 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
615
616 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
617 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
618
619 new_csum = sctp_compute_cksum(skb, sctphoff);
620
621 /* Carry any checksum errors through. */
622 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
623
624 skb_clear_hash(skb);
625 flow_key->tp.src = sh->source;
626 flow_key->tp.dst = sh->dest;
627
628 return 0;
629 }
630
631 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
632 {
633 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
634 struct vport *vport = data->vport;
635
636 if (skb_cow_head(skb, data->l2_len) < 0) {
637 kfree_skb(skb);
638 return -ENOMEM;
639 }
640
641 __skb_dst_copy(skb, data->dst);
642 *OVS_CB(skb) = data->cb;
643 skb->inner_protocol = data->inner_protocol;
644 skb->vlan_tci = data->vlan_tci;
645 skb->vlan_proto = data->vlan_proto;
646
647 /* Reconstruct the MAC header. */
648 skb_push(skb, data->l2_len);
649 memcpy(skb->data, &data->l2_data, data->l2_len);
650 skb_postpush_rcsum(skb, skb->data, data->l2_len);
651 skb_reset_mac_header(skb);
652
653 ovs_vport_send(vport, skb);
654 return 0;
655 }
656
657 static unsigned int
658 ovs_dst_get_mtu(const struct dst_entry *dst)
659 {
660 return dst->dev->mtu;
661 }
662
663 static struct dst_ops ovs_dst_ops = {
664 .family = AF_UNSPEC,
665 .mtu = ovs_dst_get_mtu,
666 };
667
668 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
669 * ovs_vport_output(), which is called once per fragmented packet.
670 */
671 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
672 {
673 unsigned int hlen = skb_network_offset(skb);
674 struct ovs_frag_data *data;
675
676 data = this_cpu_ptr(&ovs_frag_data_storage);
677 data->dst = skb->_skb_refdst;
678 data->vport = vport;
679 data->cb = *OVS_CB(skb);
680 data->inner_protocol = skb->inner_protocol;
681 data->vlan_tci = skb->vlan_tci;
682 data->vlan_proto = skb->vlan_proto;
683 data->l2_len = hlen;
684 memcpy(&data->l2_data, skb->data, hlen);
685
686 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
687 skb_pull(skb, hlen);
688 }
689
690 static void ovs_fragment(struct net *net, struct vport *vport,
691 struct sk_buff *skb, u16 mru, __be16 ethertype)
692 {
693 if (skb_network_offset(skb) > MAX_L2_LEN) {
694 OVS_NLERR(1, "L2 header too long to fragment");
695 goto err;
696 }
697
698 if (ethertype == htons(ETH_P_IP)) {
699 struct dst_entry ovs_dst;
700 unsigned long orig_dst;
701
702 prepare_frag(vport, skb);
703 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
704 DST_OBSOLETE_NONE, DST_NOCOUNT);
705 ovs_dst.dev = vport->dev;
706
707 orig_dst = skb->_skb_refdst;
708 skb_dst_set_noref(skb, &ovs_dst);
709 IPCB(skb)->frag_max_size = mru;
710
711 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
712 refdst_drop(orig_dst);
713 } else if (ethertype == htons(ETH_P_IPV6)) {
714 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
715 unsigned long orig_dst;
716 struct rt6_info ovs_rt;
717
718 if (!v6ops) {
719 goto err;
720 }
721
722 prepare_frag(vport, skb);
723 memset(&ovs_rt, 0, sizeof(ovs_rt));
724 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
725 DST_OBSOLETE_NONE, DST_NOCOUNT);
726 ovs_rt.dst.dev = vport->dev;
727
728 orig_dst = skb->_skb_refdst;
729 skb_dst_set_noref(skb, &ovs_rt.dst);
730 IP6CB(skb)->frag_max_size = mru;
731
732 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
733 refdst_drop(orig_dst);
734 } else {
735 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
736 ovs_vport_name(vport), ntohs(ethertype), mru,
737 vport->dev->mtu);
738 goto err;
739 }
740
741 return;
742 err:
743 kfree_skb(skb);
744 }
745
746 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
747 struct sw_flow_key *key)
748 {
749 struct vport *vport = ovs_vport_rcu(dp, out_port);
750
751 if (likely(vport)) {
752 u16 mru = OVS_CB(skb)->mru;
753 u32 cutlen = OVS_CB(skb)->cutlen;
754
755 if (unlikely(cutlen > 0)) {
756 if (skb->len - cutlen > ETH_HLEN)
757 pskb_trim(skb, skb->len - cutlen);
758 else
759 pskb_trim(skb, ETH_HLEN);
760 }
761
762 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
763 ovs_vport_send(vport, skb);
764 } else if (mru <= vport->dev->mtu) {
765 struct net *net = read_pnet(&dp->net);
766 __be16 ethertype = key->eth.type;
767
768 if (!is_flow_key_valid(key)) {
769 if (eth_p_mpls(skb->protocol))
770 ethertype = skb->inner_protocol;
771 else
772 ethertype = vlan_get_protocol(skb);
773 }
774
775 ovs_fragment(net, vport, skb, mru, ethertype);
776 } else {
777 kfree_skb(skb);
778 }
779 } else {
780 kfree_skb(skb);
781 }
782 }
783
784 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
785 struct sw_flow_key *key, const struct nlattr *attr,
786 const struct nlattr *actions, int actions_len,
787 uint32_t cutlen)
788 {
789 struct dp_upcall_info upcall;
790 const struct nlattr *a;
791 int rem;
792
793 memset(&upcall, 0, sizeof(upcall));
794 upcall.cmd = OVS_PACKET_CMD_ACTION;
795 upcall.mru = OVS_CB(skb)->mru;
796
797 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
798 a = nla_next(a, &rem)) {
799 switch (nla_type(a)) {
800 case OVS_USERSPACE_ATTR_USERDATA:
801 upcall.userdata = a;
802 break;
803
804 case OVS_USERSPACE_ATTR_PID:
805 upcall.portid = nla_get_u32(a);
806 break;
807
808 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
809 /* Get out tunnel info. */
810 struct vport *vport;
811
812 vport = ovs_vport_rcu(dp, nla_get_u32(a));
813 if (vport) {
814 int err;
815
816 err = dev_fill_metadata_dst(vport->dev, skb);
817 if (!err)
818 upcall.egress_tun_info = skb_tunnel_info(skb);
819 }
820
821 break;
822 }
823
824 case OVS_USERSPACE_ATTR_ACTIONS: {
825 /* Include actions. */
826 upcall.actions = actions;
827 upcall.actions_len = actions_len;
828 break;
829 }
830
831 } /* End of switch. */
832 }
833
834 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
835 }
836
837 static int sample(struct datapath *dp, struct sk_buff *skb,
838 struct sw_flow_key *key, const struct nlattr *attr,
839 const struct nlattr *actions, int actions_len)
840 {
841 const struct nlattr *acts_list = NULL;
842 const struct nlattr *a;
843 int rem;
844 u32 cutlen = 0;
845
846 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
847 a = nla_next(a, &rem)) {
848 u32 probability;
849
850 switch (nla_type(a)) {
851 case OVS_SAMPLE_ATTR_PROBABILITY:
852 probability = nla_get_u32(a);
853 if (!probability || prandom_u32() > probability)
854 return 0;
855 break;
856
857 case OVS_SAMPLE_ATTR_ACTIONS:
858 acts_list = a;
859 break;
860 }
861 }
862
863 rem = nla_len(acts_list);
864 a = nla_data(acts_list);
865
866 /* Actions list is empty, do nothing */
867 if (unlikely(!rem))
868 return 0;
869
870 /* The only known usage of sample action is having a single user-space
871 * action, or having a truncate action followed by a single user-space
872 * action. Treat this usage as a special case.
873 * The output_userspace() should clone the skb to be sent to the
874 * user space. This skb will be consumed by its caller.
875 */
876 if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
877 struct ovs_action_trunc *trunc = nla_data(a);
878
879 if (skb->len > trunc->max_len)
880 cutlen = skb->len - trunc->max_len;
881
882 a = nla_next(a, &rem);
883 }
884
885 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
886 nla_is_last(a, rem)))
887 return output_userspace(dp, skb, key, a, actions,
888 actions_len, cutlen);
889
890 skb = skb_clone(skb, GFP_ATOMIC);
891 if (!skb)
892 /* Skip the sample action when out of memory. */
893 return 0;
894
895 if (!add_deferred_actions(skb, key, a)) {
896 if (net_ratelimit())
897 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
898 ovs_dp_name(dp));
899
900 kfree_skb(skb);
901 }
902 return 0;
903 }
904
905 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
906 const struct nlattr *attr)
907 {
908 struct ovs_action_hash *hash_act = nla_data(attr);
909 u32 hash = 0;
910
911 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
912 hash = skb_get_hash(skb);
913 hash = jhash_1word(hash, hash_act->hash_basis);
914 if (!hash)
915 hash = 0x1;
916
917 key->ovs_flow_hash = hash;
918 }
919
920 static int execute_set_action(struct sk_buff *skb,
921 struct sw_flow_key *flow_key,
922 const struct nlattr *a)
923 {
924 /* Only tunnel set execution is supported without a mask. */
925 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
926 struct ovs_tunnel_info *tun = nla_data(a);
927
928 skb_dst_drop(skb);
929 dst_hold((struct dst_entry *)tun->tun_dst);
930 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
931 return 0;
932 }
933
934 return -EINVAL;
935 }
936
937 /* Mask is at the midpoint of the data. */
938 #define get_mask(a, type) ((const type)nla_data(a) + 1)
939
940 static int execute_masked_set_action(struct sk_buff *skb,
941 struct sw_flow_key *flow_key,
942 const struct nlattr *a)
943 {
944 int err = 0;
945
946 switch (nla_type(a)) {
947 case OVS_KEY_ATTR_PRIORITY:
948 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
949 *get_mask(a, u32 *));
950 flow_key->phy.priority = skb->priority;
951 break;
952
953 case OVS_KEY_ATTR_SKB_MARK:
954 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
955 flow_key->phy.skb_mark = skb->mark;
956 break;
957
958 case OVS_KEY_ATTR_TUNNEL_INFO:
959 /* Masked data not supported for tunnel. */
960 err = -EINVAL;
961 break;
962
963 case OVS_KEY_ATTR_ETHERNET:
964 err = set_eth_addr(skb, flow_key, nla_data(a),
965 get_mask(a, struct ovs_key_ethernet *));
966 break;
967
968 case OVS_KEY_ATTR_IPV4:
969 err = set_ipv4(skb, flow_key, nla_data(a),
970 get_mask(a, struct ovs_key_ipv4 *));
971 break;
972
973 case OVS_KEY_ATTR_IPV6:
974 err = set_ipv6(skb, flow_key, nla_data(a),
975 get_mask(a, struct ovs_key_ipv6 *));
976 break;
977
978 case OVS_KEY_ATTR_TCP:
979 err = set_tcp(skb, flow_key, nla_data(a),
980 get_mask(a, struct ovs_key_tcp *));
981 break;
982
983 case OVS_KEY_ATTR_UDP:
984 err = set_udp(skb, flow_key, nla_data(a),
985 get_mask(a, struct ovs_key_udp *));
986 break;
987
988 case OVS_KEY_ATTR_SCTP:
989 err = set_sctp(skb, flow_key, nla_data(a),
990 get_mask(a, struct ovs_key_sctp *));
991 break;
992
993 case OVS_KEY_ATTR_MPLS:
994 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
995 __be32 *));
996 break;
997
998 case OVS_KEY_ATTR_CT_STATE:
999 case OVS_KEY_ATTR_CT_ZONE:
1000 case OVS_KEY_ATTR_CT_MARK:
1001 case OVS_KEY_ATTR_CT_LABELS:
1002 err = -EINVAL;
1003 break;
1004 }
1005
1006 return err;
1007 }
1008
1009 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1010 struct sw_flow_key *key,
1011 const struct nlattr *a, int rem)
1012 {
1013 struct deferred_action *da;
1014
1015 if (!is_flow_key_valid(key)) {
1016 int err;
1017
1018 err = ovs_flow_key_update(skb, key);
1019 if (err)
1020 return err;
1021 }
1022 BUG_ON(!is_flow_key_valid(key));
1023
1024 if (!nla_is_last(a, rem)) {
1025 /* Recirc action is the not the last action
1026 * of the action list, need to clone the skb.
1027 */
1028 skb = skb_clone(skb, GFP_ATOMIC);
1029
1030 /* Skip the recirc action when out of memory, but
1031 * continue on with the rest of the action list.
1032 */
1033 if (!skb)
1034 return 0;
1035 }
1036
1037 da = add_deferred_actions(skb, key, NULL);
1038 if (da) {
1039 da->pkt_key.recirc_id = nla_get_u32(a);
1040 } else {
1041 kfree_skb(skb);
1042
1043 if (net_ratelimit())
1044 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1045 ovs_dp_name(dp));
1046 }
1047
1048 return 0;
1049 }
1050
1051 /* Execute a list of actions against 'skb'. */
1052 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1053 struct sw_flow_key *key,
1054 const struct nlattr *attr, int len)
1055 {
1056 /* Every output action needs a separate clone of 'skb', but the common
1057 * case is just a single output action, so that doing a clone and
1058 * then freeing the original skbuff is wasteful. So the following code
1059 * is slightly obscure just to avoid that.
1060 */
1061 int prev_port = -1;
1062 const struct nlattr *a;
1063 int rem;
1064
1065 for (a = attr, rem = len; rem > 0;
1066 a = nla_next(a, &rem)) {
1067 int err = 0;
1068
1069 if (unlikely(prev_port != -1)) {
1070 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1071
1072 if (out_skb)
1073 do_output(dp, out_skb, prev_port, key);
1074
1075 OVS_CB(skb)->cutlen = 0;
1076 prev_port = -1;
1077 }
1078
1079 switch (nla_type(a)) {
1080 case OVS_ACTION_ATTR_OUTPUT:
1081 prev_port = nla_get_u32(a);
1082 break;
1083
1084 case OVS_ACTION_ATTR_TRUNC: {
1085 struct ovs_action_trunc *trunc = nla_data(a);
1086
1087 if (skb->len > trunc->max_len)
1088 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1089 break;
1090 }
1091
1092 case OVS_ACTION_ATTR_USERSPACE:
1093 output_userspace(dp, skb, key, a, attr,
1094 len, OVS_CB(skb)->cutlen);
1095 OVS_CB(skb)->cutlen = 0;
1096 break;
1097
1098 case OVS_ACTION_ATTR_HASH:
1099 execute_hash(skb, key, a);
1100 break;
1101
1102 case OVS_ACTION_ATTR_PUSH_MPLS:
1103 err = push_mpls(skb, key, nla_data(a));
1104 break;
1105
1106 case OVS_ACTION_ATTR_POP_MPLS:
1107 err = pop_mpls(skb, key, nla_get_be16(a));
1108 break;
1109
1110 case OVS_ACTION_ATTR_PUSH_VLAN:
1111 err = push_vlan(skb, key, nla_data(a));
1112 break;
1113
1114 case OVS_ACTION_ATTR_POP_VLAN:
1115 err = pop_vlan(skb, key);
1116 break;
1117
1118 case OVS_ACTION_ATTR_RECIRC:
1119 err = execute_recirc(dp, skb, key, a, rem);
1120 if (nla_is_last(a, rem)) {
1121 /* If this is the last action, the skb has
1122 * been consumed or freed.
1123 * Return immediately.
1124 */
1125 return err;
1126 }
1127 break;
1128
1129 case OVS_ACTION_ATTR_SET:
1130 err = execute_set_action(skb, key, nla_data(a));
1131 break;
1132
1133 case OVS_ACTION_ATTR_SET_MASKED:
1134 case OVS_ACTION_ATTR_SET_TO_MASKED:
1135 err = execute_masked_set_action(skb, key, nla_data(a));
1136 break;
1137
1138 case OVS_ACTION_ATTR_SAMPLE:
1139 err = sample(dp, skb, key, a, attr, len);
1140 break;
1141
1142 case OVS_ACTION_ATTR_CT:
1143 if (!is_flow_key_valid(key)) {
1144 err = ovs_flow_key_update(skb, key);
1145 if (err)
1146 return err;
1147 }
1148
1149 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1150 nla_data(a));
1151
1152 /* Hide stolen IP fragments from user space. */
1153 if (err)
1154 return err == -EINPROGRESS ? 0 : err;
1155 break;
1156 }
1157
1158 if (unlikely(err)) {
1159 kfree_skb(skb);
1160 return err;
1161 }
1162 }
1163
1164 if (prev_port != -1)
1165 do_output(dp, skb, prev_port, key);
1166 else
1167 consume_skb(skb);
1168
1169 return 0;
1170 }
1171
1172 static void process_deferred_actions(struct datapath *dp)
1173 {
1174 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1175
1176 /* Do not touch the FIFO in case there is no deferred actions. */
1177 if (action_fifo_is_empty(fifo))
1178 return;
1179
1180 /* Finishing executing all deferred actions. */
1181 do {
1182 struct deferred_action *da = action_fifo_get(fifo);
1183 struct sk_buff *skb = da->skb;
1184 struct sw_flow_key *key = &da->pkt_key;
1185 const struct nlattr *actions = da->actions;
1186
1187 if (actions)
1188 do_execute_actions(dp, skb, key, actions,
1189 nla_len(actions));
1190 else
1191 ovs_dp_process_packet(skb, key);
1192 } while (!action_fifo_is_empty(fifo));
1193
1194 /* Reset FIFO for the next packet. */
1195 action_fifo_init(fifo);
1196 }
1197
1198 /* Execute a list of actions against 'skb'. */
1199 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1200 const struct sw_flow_actions *acts,
1201 struct sw_flow_key *key)
1202 {
1203 static const int ovs_recursion_limit = 5;
1204 int err, level;
1205
1206 level = __this_cpu_inc_return(exec_actions_level);
1207 if (unlikely(level > ovs_recursion_limit)) {
1208 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1209 ovs_dp_name(dp));
1210 kfree_skb(skb);
1211 err = -ENETDOWN;
1212 goto out;
1213 }
1214
1215 err = do_execute_actions(dp, skb, key,
1216 acts->actions, acts->actions_len);
1217
1218 if (level == 1)
1219 process_deferred_actions(dp);
1220
1221 out:
1222 __this_cpu_dec(exec_actions_level);
1223 return err;
1224 }
1225
1226 int action_fifos_init(void)
1227 {
1228 action_fifos = alloc_percpu(struct action_fifo);
1229 if (!action_fifos)
1230 return -ENOMEM;
1231
1232 return 0;
1233 }
1234
1235 void action_fifos_exit(void)
1236 {
1237 free_percpu(action_fifos);
1238 }
Linux kernel - Deliverables
This page took 0.071178 seconds and 5 git commands to generate.