]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/actions.c
projects / mirror_ovs.git / blob
? search:


8abe70aa51095da4c1bf9b629e255e1aa8919d5f
[mirror_ovs.git] / datapath / actions.c
1 /*
2 * Copyright (c) 2007-2017 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/checksum.h>
37 #include <net/dsfield.h>
38 #include <net/mpls.h>
39 #include <net/sctp/checksum.h>
40
41 #include "datapath.h"
42 #include "conntrack.h"
43 #include "gso.h"
44 #include "vport.h"
45 #include "flow_netlink.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 int actions_len;
55
56 /* Store pkt_key clone when creating deferred action. */
57 struct sw_flow_key pkt_key;
58 };
59
60 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
61 struct ovs_frag_data {
62 unsigned long dst;
63 struct vport *vport;
64 struct ovs_gso_cb cb;
65 __be16 inner_protocol;
66 u16 network_offset; /* valid only for MPLS */
67 u16 vlan_tci;
68 __be16 vlan_proto;
69 unsigned int l2_len;
70 u8 mac_proto;
71 u8 l2_data[MAX_L2_LEN];
72 };
73
74 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
75
76 #define DEFERRED_ACTION_FIFO_SIZE 10
77 #define OVS_RECURSION_LIMIT 4
78 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
79 struct action_fifo {
80 int head;
81 int tail;
82 /* Deferred action fifo queue storage. */
83 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
84 };
85
86 struct action_flow_keys {
87 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
88 };
89
90 static struct action_fifo __percpu *action_fifos;
91 static struct action_flow_keys __percpu *flow_keys;
92 static DEFINE_PER_CPU(int, exec_actions_level);
93
94 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
95 * space. Return NULL if out of key spaces.
96 */
97 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
98 {
99 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
100 int level = this_cpu_read(exec_actions_level);
101 struct sw_flow_key *key = NULL;
102
103 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
104 key = &keys->key[level - 1];
105 *key = *key_;
106 }
107
108 return key;
109 }
110
111 static void action_fifo_init(struct action_fifo *fifo)
112 {
113 fifo->head = 0;
114 fifo->tail = 0;
115 }
116
117 static bool action_fifo_is_empty(const struct action_fifo *fifo)
118 {
119 return (fifo->head == fifo->tail);
120 }
121
122 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
123 {
124 if (action_fifo_is_empty(fifo))
125 return NULL;
126
127 return &fifo->fifo[fifo->tail++];
128 }
129
130 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
131 {
132 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
133 return NULL;
134
135 return &fifo->fifo[fifo->head++];
136 }
137
138 /* Return queue entry if fifo is not full */
139 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
140 const struct sw_flow_key *key,
141 const struct nlattr *actions,
142 const int actions_len)
143 {
144 struct action_fifo *fifo;
145 struct deferred_action *da;
146
147 fifo = this_cpu_ptr(action_fifos);
148 da = action_fifo_put(fifo);
149 if (da) {
150 da->skb = skb;
151 da->actions = actions;
152 da->actions_len = actions_len;
153 da->pkt_key = *key;
154 }
155
156 return da;
157 }
158
159 static void invalidate_flow_key(struct sw_flow_key *key)
160 {
161 key->mac_proto |= SW_FLOW_KEY_INVALID;
162 }
163
164 static bool is_flow_key_valid(const struct sw_flow_key *key)
165 {
166 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
167 }
168
169 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
170 struct sw_flow_key *key,
171 u32 recirc_id,
172 const struct nlattr *actions, int len,
173 bool last, bool clone_flow_key);
174
175 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
176 __be16 ethertype)
177 {
178 if (skb->ip_summed == CHECKSUM_COMPLETE) {
179 __be16 diff[] = { ~(hdr->h_proto), ethertype };
180
181 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
182 ~skb->csum);
183 }
184
185 hdr->h_proto = ethertype;
186 }
187
188 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
189 const struct ovs_action_push_mpls *mpls)
190 {
191 struct mpls_shim_hdr *new_mpls_lse;
192
193 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
194 if (skb->encapsulation)
195 return -ENOTSUPP;
196
197 if (skb_cow_head(skb, MPLS_HLEN) < 0)
198 return -ENOMEM;
199
200 if (!ovs_skb_get_inner_protocol(skb)) {
201 skb_set_inner_network_header(skb, skb->mac_len);
202 ovs_skb_set_inner_protocol(skb, skb->protocol);
203 }
204
205 skb_push(skb, MPLS_HLEN);
206 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
207 skb->mac_len);
208 skb_reset_mac_header(skb);
209 #ifdef MPLS_HEADER_IS_L3
210 skb_set_network_header(skb, skb->mac_len);
211 #endif
212
213 new_mpls_lse = mpls_hdr(skb);
214 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
215
216 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
217
218 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
219 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
220 skb->protocol = mpls->mpls_ethertype;
221
222 invalidate_flow_key(key);
223 return 0;
224 }
225
226 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
227 const __be16 ethertype)
228 {
229 int err;
230
231 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
232 if (unlikely(err))
233 return err;
234
235 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
236
237 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
238 skb->mac_len);
239
240 __skb_pull(skb, MPLS_HLEN);
241 skb_reset_mac_header(skb);
242 skb_set_network_header(skb, skb->mac_len);
243
244 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
245 struct ethhdr *hdr;
246
247 /* mpls_hdr() is used to locate the ethertype
248 * field correctly in the presence of VLAN tags.
249 */
250 hdr = (struct ethhdr *)((void*)mpls_hdr(skb) - ETH_HLEN);
251 update_ethertype(skb, hdr, ethertype);
252 }
253 if (eth_p_mpls(skb->protocol))
254 skb->protocol = ethertype;
255
256 invalidate_flow_key(key);
257 return 0;
258 }
259
260 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
261 const __be32 *mpls_lse, const __be32 *mask)
262 {
263 struct mpls_shim_hdr *stack;
264 __be32 lse;
265 int err;
266
267 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
268 if (unlikely(err))
269 return err;
270
271 stack = mpls_hdr(skb);
272 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
273 if (skb->ip_summed == CHECKSUM_COMPLETE) {
274 __be32 diff[] = { ~(stack->label_stack_entry), lse };
275
276 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
277 ~skb->csum);
278 }
279
280 stack->label_stack_entry = lse;
281 flow_key->mpls.top_lse = lse;
282 return 0;
283 }
284
285 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
286 {
287 int err;
288
289 err = skb_vlan_pop(skb);
290 if (skb_vlan_tag_present(skb)) {
291 invalidate_flow_key(key);
292 } else {
293 key->eth.vlan.tci = 0;
294 key->eth.vlan.tpid = 0;
295 }
296 return err;
297 }
298
299 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
300 const struct ovs_action_push_vlan *vlan)
301 {
302 if (skb_vlan_tag_present(skb)) {
303 invalidate_flow_key(key);
304 } else {
305 key->eth.vlan.tci = vlan->vlan_tci;
306 key->eth.vlan.tpid = vlan->vlan_tpid;
307 }
308 return skb_vlan_push(skb, vlan->vlan_tpid,
309 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
310 }
311
312 /* 'src' is already properly masked. */
313 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
314 {
315 u16 *dst = (u16 *)dst_;
316 const u16 *src = (const u16 *)src_;
317 const u16 *mask = (const u16 *)mask_;
318
319 OVS_SET_MASKED(dst[0], src[0], mask[0]);
320 OVS_SET_MASKED(dst[1], src[1], mask[1]);
321 OVS_SET_MASKED(dst[2], src[2], mask[2]);
322 }
323
324 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
325 const struct ovs_key_ethernet *key,
326 const struct ovs_key_ethernet *mask)
327 {
328 int err;
329
330 err = skb_ensure_writable(skb, ETH_HLEN);
331 if (unlikely(err))
332 return err;
333
334 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
335
336 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
337 mask->eth_src);
338 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
339 mask->eth_dst);
340
341 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
342
343 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
344 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
345 return 0;
346 }
347
348 /* pop_eth does not support VLAN packets as this action is never called
349 * for them.
350 */
351 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
352 {
353 skb_pull_rcsum(skb, ETH_HLEN);
354 skb_reset_mac_header(skb);
355 skb_reset_mac_len(skb);
356
357 /* safe right before invalidate_flow_key */
358 key->mac_proto = MAC_PROTO_NONE;
359 invalidate_flow_key(key);
360 return 0;
361 }
362
363 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
364 const struct ovs_action_push_eth *ethh)
365 {
366 struct ethhdr *hdr;
367
368 /* Add the new Ethernet header */
369 if (skb_cow_head(skb, ETH_HLEN) < 0)
370 return -ENOMEM;
371
372 skb_push(skb, ETH_HLEN);
373 skb_reset_mac_header(skb);
374 skb_reset_mac_len(skb);
375
376 hdr = eth_hdr(skb);
377 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
378 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
379 hdr->h_proto = skb->protocol;
380
381 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
382
383 /* safe right before invalidate_flow_key */
384 key->mac_proto = MAC_PROTO_ETHERNET;
385 invalidate_flow_key(key);
386 return 0;
387 }
388
389 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
390 const struct nshhdr *nh)
391 {
392 int err;
393
394 err = ovs_nsh_push(skb, nh);
395 if (err)
396 return err;
397
398 /* safe right before invalidate_flow_key */
399 key->mac_proto = MAC_PROTO_NONE;
400 invalidate_flow_key(key);
401 return 0;
402 }
403
404 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
405 {
406 int err;
407
408 err = ovs_nsh_pop(skb);
409 if (err)
410 return err;
411
412 /* safe right before invalidate_flow_key */
413 if (skb->protocol == htons(ETH_P_TEB))
414 key->mac_proto = MAC_PROTO_ETHERNET;
415 else
416 key->mac_proto = MAC_PROTO_NONE;
417 invalidate_flow_key(key);
418 return 0;
419 }
420
421 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
422 __be32 addr, __be32 new_addr)
423 {
424 int transport_len = skb->len - skb_transport_offset(skb);
425
426 if (nh->frag_off & htons(IP_OFFSET))
427 return;
428
429 if (nh->protocol == IPPROTO_TCP) {
430 if (likely(transport_len >= sizeof(struct tcphdr)))
431 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
432 addr, new_addr, true);
433 } else if (nh->protocol == IPPROTO_UDP) {
434 if (likely(transport_len >= sizeof(struct udphdr))) {
435 struct udphdr *uh = udp_hdr(skb);
436
437 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
438 inet_proto_csum_replace4(&uh->check, skb,
439 addr, new_addr, true);
440 if (!uh->check)
441 uh->check = CSUM_MANGLED_0;
442 }
443 }
444 }
445
446 }
447
448 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
449 __be32 *addr, __be32 new_addr)
450 {
451 update_ip_l4_checksum(skb, nh, *addr, new_addr);
452 csum_replace4(&nh->check, *addr, new_addr);
453 skb_clear_hash(skb);
454 *addr = new_addr;
455 }
456
457 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
458 __be32 addr[4], const __be32 new_addr[4])
459 {
460 int transport_len = skb->len - skb_transport_offset(skb);
461
462 if (l4_proto == NEXTHDR_TCP) {
463 if (likely(transport_len >= sizeof(struct tcphdr)))
464 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
465 addr, new_addr, true);
466 } else if (l4_proto == NEXTHDR_UDP) {
467 if (likely(transport_len >= sizeof(struct udphdr))) {
468 struct udphdr *uh = udp_hdr(skb);
469
470 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
471 inet_proto_csum_replace16(&uh->check, skb,
472 addr, new_addr, true);
473 if (!uh->check)
474 uh->check = CSUM_MANGLED_0;
475 }
476 }
477 } else if (l4_proto == NEXTHDR_ICMP) {
478 if (likely(transport_len >= sizeof(struct icmp6hdr)))
479 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
480 skb, addr, new_addr, true);
481 }
482 }
483
484 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
485 const __be32 mask[4], __be32 masked[4])
486 {
487 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
488 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
489 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
490 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
491 }
492
493 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
494 __be32 addr[4], const __be32 new_addr[4],
495 bool recalculate_csum)
496 {
497 if (likely(recalculate_csum))
498 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
499
500 skb_clear_hash(skb);
501 memcpy(addr, new_addr, sizeof(__be32[4]));
502 }
503
504 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
505 {
506 /* Bits 21-24 are always unmasked, so this retains their values. */
507 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
508 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
509 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
510 }
511
512 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
513 u8 mask)
514 {
515 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
516
517 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
518 nh->ttl = new_ttl;
519 }
520
521 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
522 const struct ovs_key_ipv4 *key,
523 const struct ovs_key_ipv4 *mask)
524 {
525 struct iphdr *nh;
526 __be32 new_addr;
527 int err;
528
529 err = skb_ensure_writable(skb, skb_network_offset(skb) +
530 sizeof(struct iphdr));
531 if (unlikely(err))
532 return err;
533
534 nh = ip_hdr(skb);
535
536 /* Setting an IP addresses is typically only a side effect of
537 * matching on them in the current userspace implementation, so it
538 * makes sense to check if the value actually changed.
539 */
540 if (mask->ipv4_src) {
541 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
542
543 if (unlikely(new_addr != nh->saddr)) {
544 set_ip_addr(skb, nh, &nh->saddr, new_addr);
545 flow_key->ipv4.addr.src = new_addr;
546 }
547 }
548 if (mask->ipv4_dst) {
549 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
550
551 if (unlikely(new_addr != nh->daddr)) {
552 set_ip_addr(skb, nh, &nh->daddr, new_addr);
553 flow_key->ipv4.addr.dst = new_addr;
554 }
555 }
556 if (mask->ipv4_tos) {
557 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
558 flow_key->ip.tos = nh->tos;
559 }
560 if (mask->ipv4_ttl) {
561 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
562 flow_key->ip.ttl = nh->ttl;
563 }
564
565 return 0;
566 }
567
568 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
569 {
570 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
571 }
572
573 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
574 const struct ovs_key_ipv6 *key,
575 const struct ovs_key_ipv6 *mask)
576 {
577 struct ipv6hdr *nh;
578 int err;
579
580 err = skb_ensure_writable(skb, skb_network_offset(skb) +
581 sizeof(struct ipv6hdr));
582 if (unlikely(err))
583 return err;
584
585 nh = ipv6_hdr(skb);
586
587 /* Setting an IP addresses is typically only a side effect of
588 * matching on them in the current userspace implementation, so it
589 * makes sense to check if the value actually changed.
590 */
591 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
592 __be32 *saddr = (__be32 *)&nh->saddr;
593 __be32 masked[4];
594
595 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
596
597 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
598 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
599 true);
600 memcpy(&flow_key->ipv6.addr.src, masked,
601 sizeof(flow_key->ipv6.addr.src));
602 }
603 }
604 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
605 unsigned int offset = 0;
606 int flags = IP6_FH_F_SKIP_RH;
607 bool recalc_csum = true;
608 __be32 *daddr = (__be32 *)&nh->daddr;
609 __be32 masked[4];
610
611 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
612
613 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
614 if (ipv6_ext_hdr(nh->nexthdr))
615 recalc_csum = (ipv6_find_hdr(skb, &offset,
616 NEXTHDR_ROUTING,
617 NULL, &flags)
618 != NEXTHDR_ROUTING);
619
620 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
621 recalc_csum);
622 memcpy(&flow_key->ipv6.addr.dst, masked,
623 sizeof(flow_key->ipv6.addr.dst));
624 }
625 }
626 if (mask->ipv6_tclass) {
627 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
628 flow_key->ip.tos = ipv6_get_dsfield(nh);
629 }
630 if (mask->ipv6_label) {
631 set_ipv6_fl(nh, ntohl(key->ipv6_label),
632 ntohl(mask->ipv6_label));
633 flow_key->ipv6.label =
634 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
635 }
636 if (mask->ipv6_hlimit) {
637 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
638 mask->ipv6_hlimit);
639 flow_key->ip.ttl = nh->hop_limit;
640 }
641 return 0;
642 }
643
644 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
645 const struct nlattr *a)
646 {
647 struct nshhdr *nh;
648 size_t length;
649 int err;
650 u8 flags;
651 u8 ttl;
652 int i;
653
654 struct ovs_key_nsh key;
655 struct ovs_key_nsh mask;
656
657 err = nsh_key_from_nlattr(a, &key, &mask);
658 if (err)
659 return err;
660
661 /* Make sure the NSH base header is there */
662 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
663 return -ENOMEM;
664
665 nh = nsh_hdr(skb);
666 length = nsh_hdr_len(nh);
667
668 /* Make sure the whole NSH header is there */
669 err = skb_ensure_writable(skb, skb_network_offset(skb) +
670 length);
671 if (unlikely(err))
672 return err;
673
674 nh = nsh_hdr(skb);
675 skb_postpull_rcsum(skb, nh, length);
676 flags = nsh_get_flags(nh);
677 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
678 flow_key->nsh.base.flags = flags;
679 ttl = nsh_get_ttl(nh);
680 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
681 flow_key->nsh.base.ttl = ttl;
682 nsh_set_flags_and_ttl(nh, flags, ttl);
683 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
684 mask.base.path_hdr);
685 flow_key->nsh.base.path_hdr = nh->path_hdr;
686 switch (nh->mdtype) {
687 case NSH_M_TYPE1:
688 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
689 nh->md1.context[i] =
690 OVS_MASKED(nh->md1.context[i], key.context[i],
691 mask.context[i]);
692 }
693 memcpy(flow_key->nsh.context, nh->md1.context,
694 sizeof(nh->md1.context));
695 break;
696 case NSH_M_TYPE2:
697 memset(flow_key->nsh.context, 0,
698 sizeof(flow_key->nsh.context));
699 break;
700 default:
701 return -EINVAL;
702 }
703 skb_postpush_rcsum(skb, nh, length);
704 return 0;
705 }
706
707 /* Must follow skb_ensure_writable() since that can move the skb data. */
708 static void set_tp_port(struct sk_buff *skb, __be16 *port,
709 __be16 new_port, __sum16 *check)
710 {
711 inet_proto_csum_replace2(check, skb, *port, new_port, false);
712 *port = new_port;
713 }
714
715 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
716 const struct ovs_key_udp *key,
717 const struct ovs_key_udp *mask)
718 {
719 struct udphdr *uh;
720 __be16 src, dst;
721 int err;
722
723 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
724 sizeof(struct udphdr));
725 if (unlikely(err))
726 return err;
727
728 uh = udp_hdr(skb);
729 /* Either of the masks is non-zero, so do not bother checking them. */
730 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
731 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
732
733 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
734 if (likely(src != uh->source)) {
735 set_tp_port(skb, &uh->source, src, &uh->check);
736 flow_key->tp.src = src;
737 }
738 if (likely(dst != uh->dest)) {
739 set_tp_port(skb, &uh->dest, dst, &uh->check);
740 flow_key->tp.dst = dst;
741 }
742
743 if (unlikely(!uh->check))
744 uh->check = CSUM_MANGLED_0;
745 } else {
746 uh->source = src;
747 uh->dest = dst;
748 flow_key->tp.src = src;
749 flow_key->tp.dst = dst;
750 }
751
752 skb_clear_hash(skb);
753
754 return 0;
755 }
756
757 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
758 const struct ovs_key_tcp *key,
759 const struct ovs_key_tcp *mask)
760 {
761 struct tcphdr *th;
762 __be16 src, dst;
763 int err;
764
765 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
766 sizeof(struct tcphdr));
767 if (unlikely(err))
768 return err;
769
770 th = tcp_hdr(skb);
771 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
772 if (likely(src != th->source)) {
773 set_tp_port(skb, &th->source, src, &th->check);
774 flow_key->tp.src = src;
775 }
776 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
777 if (likely(dst != th->dest)) {
778 set_tp_port(skb, &th->dest, dst, &th->check);
779 flow_key->tp.dst = dst;
780 }
781 skb_clear_hash(skb);
782
783 return 0;
784 }
785
786 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
787 const struct ovs_key_sctp *key,
788 const struct ovs_key_sctp *mask)
789 {
790 unsigned int sctphoff = skb_transport_offset(skb);
791 struct sctphdr *sh;
792 __le32 old_correct_csum, new_csum, old_csum;
793 int err;
794
795 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
796 if (unlikely(err))
797 return err;
798
799 sh = sctp_hdr(skb);
800 old_csum = sh->checksum;
801 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
802
803 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
804 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
805
806 new_csum = sctp_compute_cksum(skb, sctphoff);
807
808 /* Carry any checksum errors through. */
809 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
810
811 skb_clear_hash(skb);
812 flow_key->tp.src = sh->source;
813 flow_key->tp.dst = sh->dest;
814
815 return 0;
816 }
817
818 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS)
819 {
820 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
821 struct vport *vport = data->vport;
822
823 if (skb_cow_head(skb, data->l2_len) < 0) {
824 kfree_skb(skb);
825 return -ENOMEM;
826 }
827
828 __skb_dst_copy(skb, data->dst);
829 *OVS_GSO_CB(skb) = data->cb;
830 ovs_skb_set_inner_protocol(skb, data->inner_protocol);
831 skb->vlan_tci = data->vlan_tci;
832 skb->vlan_proto = data->vlan_proto;
833
834 /* Reconstruct the MAC header. */
835 skb_push(skb, data->l2_len);
836 memcpy(skb->data, &data->l2_data, data->l2_len);
837 skb_postpush_rcsum(skb, skb->data, data->l2_len);
838 skb_reset_mac_header(skb);
839
840 if (eth_p_mpls(skb->protocol)) {
841 skb->inner_network_header = skb->network_header;
842 skb_set_network_header(skb, data->network_offset);
843 skb_reset_mac_len(skb);
844 }
845
846 ovs_vport_send(vport, skb, data->mac_proto);
847 return 0;
848 }
849
850 static unsigned int
851 ovs_dst_get_mtu(const struct dst_entry *dst)
852 {
853 return dst->dev->mtu;
854 }
855
856 static struct dst_ops ovs_dst_ops = {
857 .family = AF_UNSPEC,
858 .mtu = ovs_dst_get_mtu,
859 };
860
861 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
862 * ovs_vport_output(), which is called once per fragmented packet.
863 */
864 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
865 u16 orig_network_offset, u8 mac_proto)
866 {
867 unsigned int hlen = skb_network_offset(skb);
868 struct ovs_frag_data *data;
869
870 data = this_cpu_ptr(&ovs_frag_data_storage);
871 data->dst = (unsigned long) skb_dst(skb);
872 data->vport = vport;
873 data->cb = *OVS_GSO_CB(skb);
874 data->inner_protocol = ovs_skb_get_inner_protocol(skb);
875 data->network_offset = orig_network_offset;
876 data->vlan_tci = skb->vlan_tci;
877 data->vlan_proto = skb->vlan_proto;
878 data->mac_proto = mac_proto;
879 data->l2_len = hlen;
880 memcpy(&data->l2_data, skb->data, hlen);
881
882 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
883 skb_pull(skb, hlen);
884 }
885
886 static void ovs_fragment(struct net *net, struct vport *vport,
887 struct sk_buff *skb, u16 mru,
888 struct sw_flow_key *key)
889 {
890 u16 orig_network_offset = 0;
891
892 if (eth_p_mpls(skb->protocol)) {
893 orig_network_offset = skb_network_offset(skb);
894 skb->network_header = skb->inner_network_header;
895 }
896
897 if (skb_network_offset(skb) > MAX_L2_LEN) {
898 OVS_NLERR(1, "L2 header too long to fragment");
899 goto err;
900 }
901
902 if (key->eth.type == htons(ETH_P_IP)) {
903 struct dst_entry ovs_dst;
904 unsigned long orig_dst;
905
906 prepare_frag(vport, skb, orig_network_offset,
907 ovs_key_mac_proto(key));
908 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
909 DST_OBSOLETE_NONE, DST_NOCOUNT);
910 ovs_dst.dev = vport->dev;
911
912 orig_dst = (unsigned long) skb_dst(skb);
913 skb_dst_set_noref(skb, &ovs_dst);
914 IPCB(skb)->frag_max_size = mru;
915
916 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
917 refdst_drop(orig_dst);
918 } else if (key->eth.type == htons(ETH_P_IPV6)) {
919 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
920 unsigned long orig_dst;
921 struct rt6_info ovs_rt;
922
923 if (!v6ops)
924 goto err;
925
926 prepare_frag(vport, skb, orig_network_offset,
927 ovs_key_mac_proto(key));
928 memset(&ovs_rt, 0, sizeof(ovs_rt));
929 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
930 DST_OBSOLETE_NONE, DST_NOCOUNT);
931 ovs_rt.dst.dev = vport->dev;
932
933 orig_dst = (unsigned long) skb_dst(skb);
934 skb_dst_set_noref(skb, &ovs_rt.dst);
935 IP6CB(skb)->frag_max_size = mru;
936 #ifdef HAVE_IP_LOCAL_OUT_TAKES_NET
937 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
938 #else
939 v6ops->fragment(skb->sk, skb, ovs_vport_output);
940 #endif
941 refdst_drop(orig_dst);
942 } else {
943 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
944 ovs_vport_name(vport), ntohs(key->eth.type), mru,
945 vport->dev->mtu);
946 goto err;
947 }
948
949 return;
950 err:
951 kfree_skb(skb);
952 }
953
954 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
955 struct sw_flow_key *key)
956 {
957 struct vport *vport = ovs_vport_rcu(dp, out_port);
958
959 if (likely(vport)) {
960 u16 mru = OVS_CB(skb)->mru;
961 u32 cutlen = OVS_CB(skb)->cutlen;
962
963 if (unlikely(cutlen > 0)) {
964 if (skb->len - cutlen > ovs_mac_header_len(key))
965 pskb_trim(skb, skb->len - cutlen);
966 else
967 pskb_trim(skb, ovs_mac_header_len(key));
968 }
969
970 if (likely(!mru ||
971 (skb->len <= mru + vport->dev->hard_header_len))) {
972 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
973 } else if (mru <= vport->dev->mtu) {
974 struct net *net = ovs_dp_get_net(dp);
975
976 ovs_fragment(net, vport, skb, mru, key);
977 } else {
978 OVS_NLERR(true, "Cannot fragment IP frames");
979 kfree_skb(skb);
980 }
981 } else {
982 kfree_skb(skb);
983 }
984 }
985
986 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
987 struct sw_flow_key *key, const struct nlattr *attr,
988 const struct nlattr *actions, int actions_len,
989 uint32_t cutlen)
990 {
991 struct dp_upcall_info upcall;
992 const struct nlattr *a;
993 int rem, err;
994
995 memset(&upcall, 0, sizeof(upcall));
996 upcall.cmd = OVS_PACKET_CMD_ACTION;
997 upcall.mru = OVS_CB(skb)->mru;
998
999 SKB_INIT_FILL_METADATA_DST(skb);
1000 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
1001 a = nla_next(a, &rem)) {
1002 switch (nla_type(a)) {
1003 case OVS_USERSPACE_ATTR_USERDATA:
1004 upcall.userdata = a;
1005 break;
1006
1007 case OVS_USERSPACE_ATTR_PID:
1008 upcall.portid = nla_get_u32(a);
1009 break;
1010
1011 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1012 /* Get out tunnel info. */
1013 struct vport *vport;
1014
1015 vport = ovs_vport_rcu(dp, nla_get_u32(a));
1016 if (vport) {
1017 err = dev_fill_metadata_dst(vport->dev, skb);
1018 if (!err)
1019 upcall.egress_tun_info = skb_tunnel_info(skb);
1020 }
1021
1022 break;
1023 }
1024
1025 case OVS_USERSPACE_ATTR_ACTIONS: {
1026 /* Include actions. */
1027 upcall.actions = actions;
1028 upcall.actions_len = actions_len;
1029 break;
1030 }
1031
1032 } /* End of switch. */
1033 }
1034
1035 err = ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1036 SKB_RESTORE_FILL_METADATA_DST(skb);
1037 return err;
1038 }
1039
1040 /* When 'last' is true, sample() should always consume the 'skb'.
1041 * Otherwise, sample() should keep 'skb' intact regardless what
1042 * actions are executed within sample().
1043 */
1044 static int sample(struct datapath *dp, struct sk_buff *skb,
1045 struct sw_flow_key *key, const struct nlattr *attr,
1046 bool last)
1047 {
1048 struct nlattr *actions;
1049 struct nlattr *sample_arg;
1050 int rem = nla_len(attr);
1051 const struct sample_arg *arg;
1052 bool clone_flow_key;
1053
1054 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1055 sample_arg = nla_data(attr);
1056 arg = nla_data(sample_arg);
1057 actions = nla_next(sample_arg, &rem);
1058
1059 if ((arg->probability != U32_MAX) &&
1060 (!arg->probability || prandom_u32() > arg->probability)) {
1061 if (last)
1062 consume_skb(skb);
1063 return 0;
1064 }
1065
1066 clone_flow_key = !arg->exec;
1067 return clone_execute(dp, skb, key, 0, actions, rem, last,
1068 clone_flow_key);
1069 }
1070
1071 /* When 'last' is true, clone() should always consume the 'skb'.
1072 * Otherwise, clone() should keep 'skb' intact regardless what
1073 * actions are executed within clone().
1074 */
1075 static int clone(struct datapath *dp, struct sk_buff *skb,
1076 struct sw_flow_key *key, const struct nlattr *attr,
1077 bool last)
1078 {
1079 struct nlattr *actions;
1080 struct nlattr *clone_arg;
1081 int rem = nla_len(attr);
1082 bool dont_clone_flow_key;
1083
1084 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1085 clone_arg = nla_data(attr);
1086 dont_clone_flow_key = nla_get_u32(clone_arg);
1087 actions = nla_next(clone_arg, &rem);
1088
1089 return clone_execute(dp, skb, key, 0, actions, rem, last,
1090 !dont_clone_flow_key);
1091 }
1092
1093 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1094 const struct nlattr *attr)
1095 {
1096 struct ovs_action_hash *hash_act = nla_data(attr);
1097 u32 hash = 0;
1098
1099 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1100 hash = skb_get_hash(skb);
1101 hash = jhash_1word(hash, hash_act->hash_basis);
1102 if (!hash)
1103 hash = 0x1;
1104
1105 key->ovs_flow_hash = hash;
1106 }
1107
1108 static int execute_set_action(struct sk_buff *skb,
1109 struct sw_flow_key *flow_key,
1110 const struct nlattr *a)
1111 {
1112 /* Only tunnel set execution is supported without a mask. */
1113 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1114 struct ovs_tunnel_info *tun = nla_data(a);
1115
1116 ovs_skb_dst_drop(skb);
1117 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
1118 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1119 return 0;
1120 }
1121
1122 return -EINVAL;
1123 }
1124
1125 /* Mask is at the midpoint of the data. */
1126 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1127
1128 static int execute_masked_set_action(struct sk_buff *skb,
1129 struct sw_flow_key *flow_key,
1130 const struct nlattr *a)
1131 {
1132 int err = 0;
1133
1134 switch (nla_type(a)) {
1135 case OVS_KEY_ATTR_PRIORITY:
1136 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1137 *get_mask(a, u32 *));
1138 flow_key->phy.priority = skb->priority;
1139 break;
1140
1141 case OVS_KEY_ATTR_SKB_MARK:
1142 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1143 flow_key->phy.skb_mark = skb->mark;
1144 break;
1145
1146 case OVS_KEY_ATTR_TUNNEL_INFO:
1147 /* Masked data not supported for tunnel. */
1148 err = -EINVAL;
1149 break;
1150
1151 case OVS_KEY_ATTR_ETHERNET:
1152 err = set_eth_addr(skb, flow_key, nla_data(a),
1153 get_mask(a, struct ovs_key_ethernet *));
1154 break;
1155
1156 case OVS_KEY_ATTR_NSH:
1157 err = set_nsh(skb, flow_key, a);
1158 break;
1159
1160 case OVS_KEY_ATTR_IPV4:
1161 err = set_ipv4(skb, flow_key, nla_data(a),
1162 get_mask(a, struct ovs_key_ipv4 *));
1163 break;
1164
1165 case OVS_KEY_ATTR_IPV6:
1166 err = set_ipv6(skb, flow_key, nla_data(a),
1167 get_mask(a, struct ovs_key_ipv6 *));
1168 break;
1169
1170 case OVS_KEY_ATTR_TCP:
1171 err = set_tcp(skb, flow_key, nla_data(a),
1172 get_mask(a, struct ovs_key_tcp *));
1173 break;
1174
1175 case OVS_KEY_ATTR_UDP:
1176 err = set_udp(skb, flow_key, nla_data(a),
1177 get_mask(a, struct ovs_key_udp *));
1178 break;
1179
1180 case OVS_KEY_ATTR_SCTP:
1181 err = set_sctp(skb, flow_key, nla_data(a),
1182 get_mask(a, struct ovs_key_sctp *));
1183 break;
1184
1185 case OVS_KEY_ATTR_MPLS:
1186 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1187 __be32 *));
1188 break;
1189
1190 case OVS_KEY_ATTR_CT_STATE:
1191 case OVS_KEY_ATTR_CT_ZONE:
1192 case OVS_KEY_ATTR_CT_MARK:
1193 case OVS_KEY_ATTR_CT_LABELS:
1194 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1195 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1196 err = -EINVAL;
1197 break;
1198 }
1199
1200 return err;
1201 }
1202
1203 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1204 struct sw_flow_key *key,
1205 const struct nlattr *a, bool last)
1206 {
1207 u32 recirc_id;
1208
1209 if (!is_flow_key_valid(key)) {
1210 int err;
1211
1212 err = ovs_flow_key_update(skb, key);
1213 if (err)
1214 return err;
1215 }
1216 BUG_ON(!is_flow_key_valid(key));
1217
1218 recirc_id = nla_get_u32(a);
1219 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1220 }
1221
1222 /* Execute a list of actions against 'skb'. */
1223 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1224 struct sw_flow_key *key,
1225 const struct nlattr *attr, int len)
1226 {
1227 const struct nlattr *a;
1228 int rem;
1229
1230 for (a = attr, rem = len; rem > 0;
1231 a = nla_next(a, &rem)) {
1232 int err = 0;
1233
1234 switch (nla_type(a)) {
1235 case OVS_ACTION_ATTR_OUTPUT: {
1236 int port = nla_get_u32(a);
1237 struct sk_buff *clone;
1238
1239 /* Every output action needs a separate clone
1240 * of 'skb', In case the output action is the
1241 * last action, cloning can be avoided.
1242 */
1243 if (nla_is_last(a, rem)) {
1244 do_output(dp, skb, port, key);
1245 /* 'skb' has been used for output.
1246 */
1247 return 0;
1248 }
1249
1250 clone = skb_clone(skb, GFP_ATOMIC);
1251 if (clone)
1252 do_output(dp, clone, port, key);
1253 OVS_CB(skb)->cutlen = 0;
1254 break;
1255 }
1256
1257 case OVS_ACTION_ATTR_TRUNC: {
1258 struct ovs_action_trunc *trunc = nla_data(a);
1259
1260 if (skb->len > trunc->max_len)
1261 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1262 break;
1263 }
1264
1265 case OVS_ACTION_ATTR_USERSPACE:
1266 output_userspace(dp, skb, key, a, attr,
1267 len, OVS_CB(skb)->cutlen);
1268 OVS_CB(skb)->cutlen = 0;
1269 break;
1270
1271 case OVS_ACTION_ATTR_HASH:
1272 execute_hash(skb, key, a);
1273 break;
1274
1275 case OVS_ACTION_ATTR_PUSH_MPLS:
1276 err = push_mpls(skb, key, nla_data(a));
1277 break;
1278
1279 case OVS_ACTION_ATTR_POP_MPLS:
1280 err = pop_mpls(skb, key, nla_get_be16(a));
1281 break;
1282
1283 case OVS_ACTION_ATTR_PUSH_VLAN:
1284 err = push_vlan(skb, key, nla_data(a));
1285 break;
1286
1287 case OVS_ACTION_ATTR_POP_VLAN:
1288 err = pop_vlan(skb, key);
1289 break;
1290
1291 case OVS_ACTION_ATTR_RECIRC: {
1292 bool last = nla_is_last(a, rem);
1293
1294 err = execute_recirc(dp, skb, key, a, last);
1295 if (last) {
1296 /* If this is the last action, the skb has
1297 * been consumed or freed.
1298 * Return immediately.
1299 */
1300 return err;
1301 }
1302 break;
1303 }
1304
1305 case OVS_ACTION_ATTR_SET:
1306 err = execute_set_action(skb, key, nla_data(a));
1307 break;
1308
1309 case OVS_ACTION_ATTR_SET_MASKED:
1310 case OVS_ACTION_ATTR_SET_TO_MASKED:
1311 err = execute_masked_set_action(skb, key, nla_data(a));
1312 break;
1313
1314 case OVS_ACTION_ATTR_SAMPLE: {
1315 bool last = nla_is_last(a, rem);
1316
1317 err = sample(dp, skb, key, a, last);
1318 if (last)
1319 return err;
1320
1321 break;
1322 }
1323
1324 case OVS_ACTION_ATTR_CT:
1325 if (!is_flow_key_valid(key)) {
1326 err = ovs_flow_key_update(skb, key);
1327 if (err)
1328 return err;
1329 }
1330
1331 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1332 nla_data(a));
1333
1334 /* Hide stolen IP fragments from user space. */
1335 if (err)
1336 return err == -EINPROGRESS ? 0 : err;
1337 break;
1338
1339 case OVS_ACTION_ATTR_CT_CLEAR:
1340 err = ovs_ct_clear(skb, key);
1341 break;
1342
1343 case OVS_ACTION_ATTR_PUSH_ETH:
1344 err = push_eth(skb, key, nla_data(a));
1345 break;
1346
1347 case OVS_ACTION_ATTR_POP_ETH:
1348 err = pop_eth(skb, key);
1349 break;
1350
1351 case OVS_ACTION_ATTR_PUSH_NSH: {
1352 u8 buffer[NSH_HDR_MAX_LEN];
1353 struct nshhdr *nh = (struct nshhdr *)buffer;
1354
1355 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1356 NSH_HDR_MAX_LEN);
1357 if (unlikely(err))
1358 break;
1359 err = push_nsh(skb, key, nh);
1360 break;
1361 }
1362
1363 case OVS_ACTION_ATTR_POP_NSH:
1364 err = pop_nsh(skb, key);
1365 break;
1366
1367 case OVS_ACTION_ATTR_METER:
1368 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1369 consume_skb(skb);
1370 return 0;
1371 }
1372 break;
1373
1374 case OVS_ACTION_ATTR_CLONE: {
1375 bool last = nla_is_last(a, rem);
1376
1377 err = clone(dp, skb, key, a, last);
1378 if (last)
1379 return err;
1380 break;
1381 }
1382 }
1383
1384 if (unlikely(err)) {
1385 kfree_skb(skb);
1386 return err;
1387 }
1388 }
1389
1390 consume_skb(skb);
1391 return 0;
1392 }
1393
1394 /* Execute the actions on the clone of the packet. The effect of the
1395 * execution does not affect the original 'skb' nor the original 'key'.
1396 *
1397 * The execution may be deferred in case the actions can not be executed
1398 * immediately.
1399 */
1400 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1401 struct sw_flow_key *key, u32 recirc_id,
1402 const struct nlattr *actions, int len,
1403 bool last, bool clone_flow_key)
1404 {
1405 struct deferred_action *da;
1406 struct sw_flow_key *clone;
1407
1408 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1409 if (!skb) {
1410 /* Out of memory, skip this action.
1411 */
1412 return 0;
1413 }
1414
1415 /* When clone_flow_key is false, the 'key' will not be change
1416 * by the actions, then the 'key' can be used directly.
1417 * Otherwise, try to clone key from the next recursion level of
1418 * 'flow_keys'. If clone is successful, execute the actions
1419 * without deferring.
1420 */
1421 clone = clone_flow_key ? clone_key(key) : key;
1422 if (clone) {
1423 int err = 0;
1424
1425 if (actions) { /* Sample action */
1426 if (clone_flow_key)
1427 __this_cpu_inc(exec_actions_level);
1428
1429 err = do_execute_actions(dp, skb, clone,
1430 actions, len);
1431
1432 if (clone_flow_key)
1433 __this_cpu_dec(exec_actions_level);
1434 } else { /* Recirc action */
1435 clone->recirc_id = recirc_id;
1436 ovs_dp_process_packet(skb, clone);
1437 }
1438 return err;
1439 }
1440
1441 /* Out of 'flow_keys' space. Defer actions */
1442 da = add_deferred_actions(skb, key, actions, len);
1443 if (da) {
1444 if (!actions) { /* Recirc action */
1445 key = &da->pkt_key;
1446 key->recirc_id = recirc_id;
1447 }
1448 } else {
1449 /* Out of per CPU action FIFO space. Drop the 'skb' and
1450 * log an error.
1451 */
1452 kfree_skb(skb);
1453
1454 if (net_ratelimit()) {
1455 if (actions) { /* Sample action */
1456 pr_warn("%s: deferred action limit reached, drop sample action\n",
1457 ovs_dp_name(dp));
1458 } else { /* Recirc action */
1459 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1460 ovs_dp_name(dp));
1461 }
1462 }
1463 }
1464 return 0;
1465 }
1466
1467 static void process_deferred_actions(struct datapath *dp)
1468 {
1469 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1470
1471 /* Do not touch the FIFO in case there is no deferred actions. */
1472 if (action_fifo_is_empty(fifo))
1473 return;
1474
1475 /* Finishing executing all deferred actions. */
1476 do {
1477 struct deferred_action *da = action_fifo_get(fifo);
1478 struct sk_buff *skb = da->skb;
1479 struct sw_flow_key *key = &da->pkt_key;
1480 const struct nlattr *actions = da->actions;
1481 int actions_len = da->actions_len;
1482
1483 if (actions)
1484 do_execute_actions(dp, skb, key, actions, actions_len);
1485 else
1486 ovs_dp_process_packet(skb, key);
1487 } while (!action_fifo_is_empty(fifo));
1488
1489 /* Reset FIFO for the next packet. */
1490 action_fifo_init(fifo);
1491 }
1492
1493 /* Execute a list of actions against 'skb'. */
1494 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1495 const struct sw_flow_actions *acts,
1496 struct sw_flow_key *key)
1497 {
1498 int err, level;
1499
1500 level = __this_cpu_inc_return(exec_actions_level);
1501 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1502 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1503 ovs_dp_name(dp));
1504 kfree_skb(skb);
1505 err = -ENETDOWN;
1506 goto out;
1507 }
1508
1509 OVS_CB(skb)->acts_origlen = acts->orig_len;
1510 err = do_execute_actions(dp, skb, key,
1511 acts->actions, acts->actions_len);
1512
1513 if (level == 1)
1514 process_deferred_actions(dp);
1515
1516 out:
1517 __this_cpu_dec(exec_actions_level);
1518 return err;
1519 }
1520
1521 int action_fifos_init(void)
1522 {
1523 action_fifos = alloc_percpu(struct action_fifo);
1524 if (!action_fifos)
1525 return -ENOMEM;
1526
1527 flow_keys = alloc_percpu(struct action_flow_keys);
1528 if (!flow_keys) {
1529 free_percpu(action_fifos);
1530 return -ENOMEM;
1531 }
1532
1533 return 0;
1534 }
1535
1536 void action_fifos_exit(void)
1537 {
1538 free_percpu(action_fifos);
1539 free_percpu(flow_keys);
1540 }
openvswitch mirror