]> git.proxmox.com Git - ovs.git/blob - datapath/actions.c
projects / ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
dpctl: Fix dpctl process command parameter error.
[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), skb->csum);
182 }
183
184 hdr->h_proto = ethertype;
185 }
186
187 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
188 const struct ovs_action_push_mpls *mpls)
189 {
190 struct mpls_shim_hdr *new_mpls_lse;
191
192 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
193 if (skb->encapsulation)
194 return -ENOTSUPP;
195
196 if (skb_cow_head(skb, MPLS_HLEN) < 0)
197 return -ENOMEM;
198
199 if (!ovs_skb_get_inner_protocol(skb)) {
200 skb_set_inner_network_header(skb, skb->mac_len);
201 ovs_skb_set_inner_protocol(skb, skb->protocol);
202 }
203
204 skb_push(skb, MPLS_HLEN);
205 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
206 skb->mac_len);
207 skb_reset_mac_header(skb);
208 #ifdef MPLS_HEADER_IS_L3
209 skb_set_network_header(skb, skb->mac_len);
210 #endif
211
212 new_mpls_lse = mpls_hdr(skb);
213 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
214
215 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
216
217 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
218 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
219 skb->protocol = mpls->mpls_ethertype;
220
221 invalidate_flow_key(key);
222 return 0;
223 }
224
225 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
226 const __be16 ethertype)
227 {
228 int err;
229
230 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
231 if (unlikely(err))
232 return err;
233
234 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
235
236 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
237 skb->mac_len);
238
239 __skb_pull(skb, MPLS_HLEN);
240 skb_reset_mac_header(skb);
241 skb_set_network_header(skb, skb->mac_len);
242
243 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
244 struct ethhdr *hdr;
245
246 /* mpls_hdr() is used to locate the ethertype
247 * field correctly in the presence of VLAN tags.
248 */
249 hdr = (struct ethhdr *)((void*)mpls_hdr(skb) - ETH_HLEN);
250 update_ethertype(skb, hdr, ethertype);
251 }
252 if (eth_p_mpls(skb->protocol))
253 skb->protocol = ethertype;
254
255 invalidate_flow_key(key);
256 return 0;
257 }
258
259 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
260 const __be32 *mpls_lse, const __be32 *mask)
261 {
262 struct mpls_shim_hdr *stack;
263 __be32 lse;
264 int err;
265
266 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
267 if (unlikely(err))
268 return err;
269
270 stack = mpls_hdr(skb);
271 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
272 if (skb->ip_summed == CHECKSUM_COMPLETE) {
273 __be32 diff[] = { ~(stack->label_stack_entry), lse };
274
275 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
276 }
277
278 stack->label_stack_entry = lse;
279 flow_key->mpls.lse[0] = lse;
280 return 0;
281 }
282
283 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
284 {
285 int err;
286
287 err = skb_vlan_pop(skb);
288 if (skb_vlan_tag_present(skb)) {
289 invalidate_flow_key(key);
290 } else {
291 key->eth.vlan.tci = 0;
292 key->eth.vlan.tpid = 0;
293 }
294 return err;
295 }
296
297 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
298 const struct ovs_action_push_vlan *vlan)
299 {
300 if (skb_vlan_tag_present(skb)) {
301 invalidate_flow_key(key);
302 } else {
303 key->eth.vlan.tci = vlan->vlan_tci;
304 key->eth.vlan.tpid = vlan->vlan_tpid;
305 }
306 return skb_vlan_push(skb, vlan->vlan_tpid,
307 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
308 }
309
310 /* 'src' is already properly masked. */
311 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
312 {
313 u16 *dst = (u16 *)dst_;
314 const u16 *src = (const u16 *)src_;
315 const u16 *mask = (const u16 *)mask_;
316
317 OVS_SET_MASKED(dst[0], src[0], mask[0]);
318 OVS_SET_MASKED(dst[1], src[1], mask[1]);
319 OVS_SET_MASKED(dst[2], src[2], mask[2]);
320 }
321
322 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
323 const struct ovs_key_ethernet *key,
324 const struct ovs_key_ethernet *mask)
325 {
326 int err;
327
328 err = skb_ensure_writable(skb, ETH_HLEN);
329 if (unlikely(err))
330 return err;
331
332 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
333
334 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
335 mask->eth_src);
336 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
337 mask->eth_dst);
338
339 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
340
341 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
342 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
343 return 0;
344 }
345
346 /* pop_eth does not support VLAN packets as this action is never called
347 * for them.
348 */
349 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
350 {
351 skb_pull_rcsum(skb, ETH_HLEN);
352 skb_reset_mac_header(skb);
353 skb_reset_mac_len(skb);
354
355 /* safe right before invalidate_flow_key */
356 key->mac_proto = MAC_PROTO_NONE;
357 invalidate_flow_key(key);
358 return 0;
359 }
360
361 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
362 const struct ovs_action_push_eth *ethh)
363 {
364 struct ethhdr *hdr;
365
366 /* Add the new Ethernet header */
367 if (skb_cow_head(skb, ETH_HLEN) < 0)
368 return -ENOMEM;
369
370 skb_push(skb, ETH_HLEN);
371 skb_reset_mac_header(skb);
372 skb_reset_mac_len(skb);
373
374 hdr = eth_hdr(skb);
375 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
376 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
377 hdr->h_proto = skb->protocol;
378
379 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
380
381 /* safe right before invalidate_flow_key */
382 key->mac_proto = MAC_PROTO_ETHERNET;
383 invalidate_flow_key(key);
384 return 0;
385 }
386
387 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
388 const struct nshhdr *nh)
389 {
390 int err;
391
392 err = ovs_nsh_push(skb, nh);
393 if (err)
394 return err;
395
396 /* safe right before invalidate_flow_key */
397 key->mac_proto = MAC_PROTO_NONE;
398 invalidate_flow_key(key);
399 return 0;
400 }
401
402 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
403 {
404 int err;
405
406 err = ovs_nsh_pop(skb);
407 if (err)
408 return err;
409
410 /* safe right before invalidate_flow_key */
411 if (skb->protocol == htons(ETH_P_TEB))
412 key->mac_proto = MAC_PROTO_ETHERNET;
413 else
414 key->mac_proto = MAC_PROTO_NONE;
415 invalidate_flow_key(key);
416 return 0;
417 }
418
419 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
420 __be32 addr, __be32 new_addr)
421 {
422 int transport_len = skb->len - skb_transport_offset(skb);
423
424 if (nh->frag_off & htons(IP_OFFSET))
425 return;
426
427 if (nh->protocol == IPPROTO_TCP) {
428 if (likely(transport_len >= sizeof(struct tcphdr)))
429 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
430 addr, new_addr, true);
431 } else if (nh->protocol == IPPROTO_UDP) {
432 if (likely(transport_len >= sizeof(struct udphdr))) {
433 struct udphdr *uh = udp_hdr(skb);
434
435 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
436 inet_proto_csum_replace4(&uh->check, skb,
437 addr, new_addr, true);
438 if (!uh->check)
439 uh->check = CSUM_MANGLED_0;
440 }
441 }
442 }
443
444 }
445
446 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
447 __be32 *addr, __be32 new_addr)
448 {
449 update_ip_l4_checksum(skb, nh, *addr, new_addr);
450 csum_replace4(&nh->check, *addr, new_addr);
451 skb_clear_hash(skb);
452 *addr = new_addr;
453 }
454
455 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
456 __be32 addr[4], const __be32 new_addr[4])
457 {
458 int transport_len = skb->len - skb_transport_offset(skb);
459
460 if (l4_proto == NEXTHDR_TCP) {
461 if (likely(transport_len >= sizeof(struct tcphdr)))
462 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
463 addr, new_addr, true);
464 } else if (l4_proto == NEXTHDR_UDP) {
465 if (likely(transport_len >= sizeof(struct udphdr))) {
466 struct udphdr *uh = udp_hdr(skb);
467
468 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
469 inet_proto_csum_replace16(&uh->check, skb,
470 addr, new_addr, true);
471 if (!uh->check)
472 uh->check = CSUM_MANGLED_0;
473 }
474 }
475 } else if (l4_proto == NEXTHDR_ICMP) {
476 if (likely(transport_len >= sizeof(struct icmp6hdr)))
477 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
478 skb, addr, new_addr, true);
479 }
480 }
481
482 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
483 const __be32 mask[4], __be32 masked[4])
484 {
485 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
486 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
487 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
488 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
489 }
490
491 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
492 __be32 addr[4], const __be32 new_addr[4],
493 bool recalculate_csum)
494 {
495 if (likely(recalculate_csum))
496 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
497
498 skb_clear_hash(skb);
499 memcpy(addr, new_addr, sizeof(__be32[4]));
500 }
501
502 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
503 {
504 /* Bits 21-24 are always unmasked, so this retains their values. */
505 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
506 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
507 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
508 }
509
510 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
511 u8 mask)
512 {
513 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
514
515 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
516 nh->ttl = new_ttl;
517 }
518
519 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
520 const struct ovs_key_ipv4 *key,
521 const struct ovs_key_ipv4 *mask)
522 {
523 struct iphdr *nh;
524 __be32 new_addr;
525 int err;
526
527 err = skb_ensure_writable(skb, skb_network_offset(skb) +
528 sizeof(struct iphdr));
529 if (unlikely(err))
530 return err;
531
532 nh = ip_hdr(skb);
533
534 /* Setting an IP addresses is typically only a side effect of
535 * matching on them in the current userspace implementation, so it
536 * makes sense to check if the value actually changed.
537 */
538 if (mask->ipv4_src) {
539 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
540
541 if (unlikely(new_addr != nh->saddr)) {
542 set_ip_addr(skb, nh, &nh->saddr, new_addr);
543 flow_key->ipv4.addr.src = new_addr;
544 }
545 }
546 if (mask->ipv4_dst) {
547 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
548
549 if (unlikely(new_addr != nh->daddr)) {
550 set_ip_addr(skb, nh, &nh->daddr, new_addr);
551 flow_key->ipv4.addr.dst = new_addr;
552 }
553 }
554 if (mask->ipv4_tos) {
555 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
556 flow_key->ip.tos = nh->tos;
557 }
558 if (mask->ipv4_ttl) {
559 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
560 flow_key->ip.ttl = nh->ttl;
561 }
562
563 return 0;
564 }
565
566 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
567 {
568 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
569 }
570
571 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
572 const struct ovs_key_ipv6 *key,
573 const struct ovs_key_ipv6 *mask)
574 {
575 struct ipv6hdr *nh;
576 int err;
577
578 err = skb_ensure_writable(skb, skb_network_offset(skb) +
579 sizeof(struct ipv6hdr));
580 if (unlikely(err))
581 return err;
582
583 nh = ipv6_hdr(skb);
584
585 /* Setting an IP addresses is typically only a side effect of
586 * matching on them in the current userspace implementation, so it
587 * makes sense to check if the value actually changed.
588 */
589 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
590 __be32 *saddr = (__be32 *)&nh->saddr;
591 __be32 masked[4];
592
593 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
594
595 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
596 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
597 true);
598 memcpy(&flow_key->ipv6.addr.src, masked,
599 sizeof(flow_key->ipv6.addr.src));
600 }
601 }
602 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
603 unsigned int offset = 0;
604 int flags = IP6_FH_F_SKIP_RH;
605 bool recalc_csum = true;
606 __be32 *daddr = (__be32 *)&nh->daddr;
607 __be32 masked[4];
608
609 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
610
611 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
612 if (ipv6_ext_hdr(nh->nexthdr))
613 recalc_csum = (ipv6_find_hdr(skb, &offset,
614 NEXTHDR_ROUTING,
615 NULL, &flags)
616 != NEXTHDR_ROUTING);
617
618 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
619 recalc_csum);
620 memcpy(&flow_key->ipv6.addr.dst, masked,
621 sizeof(flow_key->ipv6.addr.dst));
622 }
623 }
624 if (mask->ipv6_tclass) {
625 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
626 flow_key->ip.tos = ipv6_get_dsfield(nh);
627 }
628 if (mask->ipv6_label) {
629 set_ipv6_fl(nh, ntohl(key->ipv6_label),
630 ntohl(mask->ipv6_label));
631 flow_key->ipv6.label =
632 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
633 }
634 if (mask->ipv6_hlimit) {
635 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
636 mask->ipv6_hlimit);
637 flow_key->ip.ttl = nh->hop_limit;
638 }
639 return 0;
640 }
641
642 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
643 const struct nlattr *a)
644 {
645 struct nshhdr *nh;
646 size_t length;
647 int err;
648 u8 flags;
649 u8 ttl;
650 int i;
651
652 struct ovs_key_nsh key;
653 struct ovs_key_nsh mask;
654
655 err = nsh_key_from_nlattr(a, &key, &mask);
656 if (err)
657 return err;
658
659 /* Make sure the NSH base header is there */
660 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
661 return -ENOMEM;
662
663 nh = nsh_hdr(skb);
664 length = nsh_hdr_len(nh);
665
666 /* Make sure the whole NSH header is there */
667 err = skb_ensure_writable(skb, skb_network_offset(skb) +
668 length);
669 if (unlikely(err))
670 return err;
671
672 nh = nsh_hdr(skb);
673 skb_postpull_rcsum(skb, nh, length);
674 flags = nsh_get_flags(nh);
675 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
676 flow_key->nsh.base.flags = flags;
677 ttl = nsh_get_ttl(nh);
678 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
679 flow_key->nsh.base.ttl = ttl;
680 nsh_set_flags_and_ttl(nh, flags, ttl);
681 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
682 mask.base.path_hdr);
683 flow_key->nsh.base.path_hdr = nh->path_hdr;
684 switch (nh->mdtype) {
685 case NSH_M_TYPE1:
686 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
687 nh->md1.context[i] =
688 OVS_MASKED(nh->md1.context[i], key.context[i],
689 mask.context[i]);
690 }
691 memcpy(flow_key->nsh.context, nh->md1.context,
692 sizeof(nh->md1.context));
693 break;
694 case NSH_M_TYPE2:
695 memset(flow_key->nsh.context, 0,
696 sizeof(flow_key->nsh.context));
697 break;
698 default:
699 return -EINVAL;
700 }
701 skb_postpush_rcsum(skb, nh, length);
702 return 0;
703 }
704
705 /* Must follow skb_ensure_writable() since that can move the skb data. */
706 static void set_tp_port(struct sk_buff *skb, __be16 *port,
707 __be16 new_port, __sum16 *check)
708 {
709 inet_proto_csum_replace2(check, skb, *port, new_port, false);
710 *port = new_port;
711 }
712
713 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
714 const struct ovs_key_udp *key,
715 const struct ovs_key_udp *mask)
716 {
717 struct udphdr *uh;
718 __be16 src, dst;
719 int err;
720
721 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
722 sizeof(struct udphdr));
723 if (unlikely(err))
724 return err;
725
726 uh = udp_hdr(skb);
727 /* Either of the masks is non-zero, so do not bother checking them. */
728 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
729 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
730
731 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
732 if (likely(src != uh->source)) {
733 set_tp_port(skb, &uh->source, src, &uh->check);
734 flow_key->tp.src = src;
735 }
736 if (likely(dst != uh->dest)) {
737 set_tp_port(skb, &uh->dest, dst, &uh->check);
738 flow_key->tp.dst = dst;
739 }
740
741 if (unlikely(!uh->check))
742 uh->check = CSUM_MANGLED_0;
743 } else {
744 uh->source = src;
745 uh->dest = dst;
746 flow_key->tp.src = src;
747 flow_key->tp.dst = dst;
748 }
749
750 skb_clear_hash(skb);
751
752 return 0;
753 }
754
755 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
756 const struct ovs_key_tcp *key,
757 const struct ovs_key_tcp *mask)
758 {
759 struct tcphdr *th;
760 __be16 src, dst;
761 int err;
762
763 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
764 sizeof(struct tcphdr));
765 if (unlikely(err))
766 return err;
767
768 th = tcp_hdr(skb);
769 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
770 if (likely(src != th->source)) {
771 set_tp_port(skb, &th->source, src, &th->check);
772 flow_key->tp.src = src;
773 }
774 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
775 if (likely(dst != th->dest)) {
776 set_tp_port(skb, &th->dest, dst, &th->check);
777 flow_key->tp.dst = dst;
778 }
779 skb_clear_hash(skb);
780
781 return 0;
782 }
783
784 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
785 const struct ovs_key_sctp *key,
786 const struct ovs_key_sctp *mask)
787 {
788 unsigned int sctphoff = skb_transport_offset(skb);
789 struct sctphdr *sh;
790 __le32 old_correct_csum, new_csum, old_csum;
791 int err;
792
793 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
794 if (unlikely(err))
795 return err;
796
797 sh = sctp_hdr(skb);
798 old_csum = sh->checksum;
799 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
800
801 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
802 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
803
804 new_csum = sctp_compute_cksum(skb, sctphoff);
805
806 /* Carry any checksum errors through. */
807 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
808
809 skb_clear_hash(skb);
810 flow_key->tp.src = sh->source;
811 flow_key->tp.dst = sh->dest;
812
813 return 0;
814 }
815
816 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS)
817 {
818 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
819 struct vport *vport = data->vport;
820
821 if (skb_cow_head(skb, data->l2_len) < 0) {
822 kfree_skb(skb);
823 return -ENOMEM;
824 }
825
826 __skb_dst_copy(skb, data->dst);
827 *OVS_GSO_CB(skb) = data->cb;
828 ovs_skb_set_inner_protocol(skb, data->inner_protocol);
829 if (data->vlan_tci & VLAN_CFI_MASK)
830 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
831 else
832 __vlan_hwaccel_clear_tag(skb);
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 if (skb_vlan_tag_present(skb))
877 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
878 else
879 data->vlan_tci = 0;
880 data->vlan_proto = skb->vlan_proto;
881 data->mac_proto = mac_proto;
882 data->l2_len = hlen;
883 memcpy(&data->l2_data, skb->data, hlen);
884
885 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
886 skb_pull(skb, hlen);
887 }
888
889 static void ovs_fragment(struct net *net, struct vport *vport,
890 struct sk_buff *skb, u16 mru,
891 struct sw_flow_key *key)
892 {
893 u16 orig_network_offset = 0;
894
895 if (eth_p_mpls(skb->protocol)) {
896 orig_network_offset = skb_network_offset(skb);
897 skb->network_header = skb->inner_network_header;
898 }
899
900 if (skb_network_offset(skb) > MAX_L2_LEN) {
901 OVS_NLERR(1, "L2 header too long to fragment");
902 goto err;
903 }
904
905 if (key->eth.type == htons(ETH_P_IP)) {
906 struct dst_entry ovs_dst;
907 unsigned long orig_dst;
908
909 prepare_frag(vport, skb, orig_network_offset,
910 ovs_key_mac_proto(key));
911 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
912 DST_OBSOLETE_NONE, DST_NOCOUNT);
913 ovs_dst.dev = vport->dev;
914
915 orig_dst = (unsigned long) skb_dst(skb);
916 skb_dst_set_noref(skb, &ovs_dst);
917 IPCB(skb)->frag_max_size = mru;
918
919 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
920 refdst_drop(orig_dst);
921 } else if (key->eth.type == htons(ETH_P_IPV6)) {
922 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
923 unsigned long orig_dst;
924 struct rt6_info ovs_rt;
925
926 if (!v6ops)
927 goto err;
928
929 prepare_frag(vport, skb, orig_network_offset,
930 ovs_key_mac_proto(key));
931 memset(&ovs_rt, 0, sizeof(ovs_rt));
932 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
933 DST_OBSOLETE_NONE, DST_NOCOUNT);
934 ovs_rt.dst.dev = vport->dev;
935
936 orig_dst = (unsigned long) skb_dst(skb);
937 skb_dst_set_noref(skb, &ovs_rt.dst);
938 IP6CB(skb)->frag_max_size = mru;
939 #ifdef HAVE_IP_LOCAL_OUT_TAKES_NET
940 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
941 #else
942 v6ops->fragment(skb->sk, skb, ovs_vport_output);
943 #endif
944 refdst_drop(orig_dst);
945 } else {
946 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
947 ovs_vport_name(vport), ntohs(key->eth.type), mru,
948 vport->dev->mtu);
949 goto err;
950 }
951
952 return;
953 err:
954 kfree_skb(skb);
955 }
956
957 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
958 struct sw_flow_key *key)
959 {
960 struct vport *vport = ovs_vport_rcu(dp, out_port);
961
962 if (likely(vport)) {
963 u16 mru = OVS_CB(skb)->mru;
964 u32 cutlen = OVS_CB(skb)->cutlen;
965
966 if (unlikely(cutlen > 0)) {
967 if (skb->len - cutlen > ovs_mac_header_len(key))
968 pskb_trim(skb, skb->len - cutlen);
969 else
970 pskb_trim(skb, ovs_mac_header_len(key));
971 }
972
973 if (likely(!mru ||
974 (skb->len <= mru + vport->dev->hard_header_len))) {
975 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
976 } else if (mru <= vport->dev->mtu) {
977 struct net *net = ovs_dp_get_net(dp);
978
979 ovs_fragment(net, vport, skb, mru, key);
980 } else {
981 OVS_NLERR(true, "Cannot fragment IP frames");
982 kfree_skb(skb);
983 }
984 } else {
985 kfree_skb(skb);
986 }
987 }
988
989 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
990 struct sw_flow_key *key, const struct nlattr *attr,
991 const struct nlattr *actions, int actions_len,
992 uint32_t cutlen)
993 {
994 struct dp_upcall_info upcall;
995 const struct nlattr *a;
996 int rem, err;
997
998 memset(&upcall, 0, sizeof(upcall));
999 upcall.cmd = OVS_PACKET_CMD_ACTION;
1000 upcall.mru = OVS_CB(skb)->mru;
1001
1002 SKB_INIT_FILL_METADATA_DST(skb);
1003 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
1004 a = nla_next(a, &rem)) {
1005 switch (nla_type(a)) {
1006 case OVS_USERSPACE_ATTR_USERDATA:
1007 upcall.userdata = a;
1008 break;
1009
1010 case OVS_USERSPACE_ATTR_PID:
1011 upcall.portid = nla_get_u32(a);
1012 break;
1013
1014 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1015 /* Get out tunnel info. */
1016 struct vport *vport;
1017
1018 vport = ovs_vport_rcu(dp, nla_get_u32(a));
1019 if (vport) {
1020 err = dev_fill_metadata_dst(vport->dev, skb);
1021 if (!err)
1022 upcall.egress_tun_info = skb_tunnel_info(skb);
1023 }
1024
1025 break;
1026 }
1027
1028 case OVS_USERSPACE_ATTR_ACTIONS: {
1029 /* Include actions. */
1030 upcall.actions = actions;
1031 upcall.actions_len = actions_len;
1032 break;
1033 }
1034
1035 } /* End of switch. */
1036 }
1037
1038 err = ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1039 SKB_RESTORE_FILL_METADATA_DST(skb);
1040 return err;
1041 }
1042
1043 /* When 'last' is true, sample() should always consume the 'skb'.
1044 * Otherwise, sample() should keep 'skb' intact regardless what
1045 * actions are executed within sample().
1046 */
1047 static int sample(struct datapath *dp, struct sk_buff *skb,
1048 struct sw_flow_key *key, const struct nlattr *attr,
1049 bool last)
1050 {
1051 struct nlattr *actions;
1052 struct nlattr *sample_arg;
1053 int rem = nla_len(attr);
1054 const struct sample_arg *arg;
1055 bool clone_flow_key;
1056
1057 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1058 sample_arg = nla_data(attr);
1059 arg = nla_data(sample_arg);
1060 actions = nla_next(sample_arg, &rem);
1061
1062 if ((arg->probability != U32_MAX) &&
1063 (!arg->probability || prandom_u32() > arg->probability)) {
1064 if (last)
1065 consume_skb(skb);
1066 return 0;
1067 }
1068
1069 clone_flow_key = !arg->exec;
1070 return clone_execute(dp, skb, key, 0, actions, rem, last,
1071 clone_flow_key);
1072 }
1073
1074 /* When 'last' is true, clone() should always consume the 'skb'.
1075 * Otherwise, clone() should keep 'skb' intact regardless what
1076 * actions are executed within clone().
1077 */
1078 static int clone(struct datapath *dp, struct sk_buff *skb,
1079 struct sw_flow_key *key, const struct nlattr *attr,
1080 bool last)
1081 {
1082 struct nlattr *actions;
1083 struct nlattr *clone_arg;
1084 int rem = nla_len(attr);
1085 bool dont_clone_flow_key;
1086
1087 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1088 clone_arg = nla_data(attr);
1089 dont_clone_flow_key = nla_get_u32(clone_arg);
1090 actions = nla_next(clone_arg, &rem);
1091
1092 return clone_execute(dp, skb, key, 0, actions, rem, last,
1093 !dont_clone_flow_key);
1094 }
1095
1096 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1097 const struct nlattr *attr)
1098 {
1099 struct ovs_action_hash *hash_act = nla_data(attr);
1100 u32 hash = 0;
1101
1102 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1103 hash = skb_get_hash(skb);
1104 hash = jhash_1word(hash, hash_act->hash_basis);
1105 if (!hash)
1106 hash = 0x1;
1107
1108 key->ovs_flow_hash = hash;
1109 }
1110
1111 static int execute_set_action(struct sk_buff *skb,
1112 struct sw_flow_key *flow_key,
1113 const struct nlattr *a)
1114 {
1115 /* Only tunnel set execution is supported without a mask. */
1116 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1117 struct ovs_tunnel_info *tun = nla_data(a);
1118
1119 ovs_skb_dst_drop(skb);
1120 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
1121 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1122 return 0;
1123 }
1124
1125 return -EINVAL;
1126 }
1127
1128 /* Mask is at the midpoint of the data. */
1129 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1130
1131 static int execute_masked_set_action(struct sk_buff *skb,
1132 struct sw_flow_key *flow_key,
1133 const struct nlattr *a)
1134 {
1135 int err = 0;
1136
1137 switch (nla_type(a)) {
1138 case OVS_KEY_ATTR_PRIORITY:
1139 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1140 *get_mask(a, u32 *));
1141 flow_key->phy.priority = skb->priority;
1142 break;
1143
1144 case OVS_KEY_ATTR_SKB_MARK:
1145 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1146 flow_key->phy.skb_mark = skb->mark;
1147 break;
1148
1149 case OVS_KEY_ATTR_TUNNEL_INFO:
1150 /* Masked data not supported for tunnel. */
1151 err = -EINVAL;
1152 break;
1153
1154 case OVS_KEY_ATTR_ETHERNET:
1155 err = set_eth_addr(skb, flow_key, nla_data(a),
1156 get_mask(a, struct ovs_key_ethernet *));
1157 break;
1158
1159 case OVS_KEY_ATTR_NSH:
1160 err = set_nsh(skb, flow_key, a);
1161 break;
1162
1163 case OVS_KEY_ATTR_IPV4:
1164 err = set_ipv4(skb, flow_key, nla_data(a),
1165 get_mask(a, struct ovs_key_ipv4 *));
1166 break;
1167
1168 case OVS_KEY_ATTR_IPV6:
1169 err = set_ipv6(skb, flow_key, nla_data(a),
1170 get_mask(a, struct ovs_key_ipv6 *));
1171 break;
1172
1173 case OVS_KEY_ATTR_TCP:
1174 err = set_tcp(skb, flow_key, nla_data(a),
1175 get_mask(a, struct ovs_key_tcp *));
1176 break;
1177
1178 case OVS_KEY_ATTR_UDP:
1179 err = set_udp(skb, flow_key, nla_data(a),
1180 get_mask(a, struct ovs_key_udp *));
1181 break;
1182
1183 case OVS_KEY_ATTR_SCTP:
1184 err = set_sctp(skb, flow_key, nla_data(a),
1185 get_mask(a, struct ovs_key_sctp *));
1186 break;
1187
1188 case OVS_KEY_ATTR_MPLS:
1189 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1190 __be32 *));
1191 break;
1192
1193 case OVS_KEY_ATTR_CT_STATE:
1194 case OVS_KEY_ATTR_CT_ZONE:
1195 case OVS_KEY_ATTR_CT_MARK:
1196 case OVS_KEY_ATTR_CT_LABELS:
1197 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1198 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1199 err = -EINVAL;
1200 break;
1201 }
1202
1203 return err;
1204 }
1205
1206 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1207 struct sw_flow_key *key,
1208 const struct nlattr *a, bool last)
1209 {
1210 u32 recirc_id;
1211
1212 if (!is_flow_key_valid(key)) {
1213 int err;
1214
1215 err = ovs_flow_key_update(skb, key);
1216 if (err)
1217 return err;
1218 }
1219 BUG_ON(!is_flow_key_valid(key));
1220
1221 recirc_id = nla_get_u32(a);
1222 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1223 }
1224
1225 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1226 struct sw_flow_key *key,
1227 const struct nlattr *attr, bool last)
1228 {
1229 const struct nlattr *actions, *cpl_arg;
1230 const struct check_pkt_len_arg *arg;
1231 int rem = nla_len(attr);
1232 bool clone_flow_key;
1233
1234 /* The first netlink attribute in 'attr' is always
1235 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1236 */
1237 cpl_arg = nla_data(attr);
1238 arg = nla_data(cpl_arg);
1239
1240 if (skb->len <= arg->pkt_len) {
1241 /* Second netlink attribute in 'attr' is always
1242 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1243 */
1244 actions = nla_next(cpl_arg, &rem);
1245 clone_flow_key = !arg->exec_for_lesser_equal;
1246 } else {
1247 /* Third netlink attribute in 'attr' is always
1248 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1249 */
1250 actions = nla_next(cpl_arg, &rem);
1251 actions = nla_next(actions, &rem);
1252 clone_flow_key = !arg->exec_for_greater;
1253 }
1254
1255 return clone_execute(dp, skb, key, 0, nla_data(actions),
1256 nla_len(actions), last, clone_flow_key);
1257 }
1258
1259 /* Execute a list of actions against 'skb'. */
1260 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1261 struct sw_flow_key *key,
1262 const struct nlattr *attr, int len)
1263 {
1264 const struct nlattr *a;
1265 int rem;
1266
1267 for (a = attr, rem = len; rem > 0;
1268 a = nla_next(a, &rem)) {
1269 int err = 0;
1270
1271 switch (nla_type(a)) {
1272 case OVS_ACTION_ATTR_OUTPUT: {
1273 int port = nla_get_u32(a);
1274 struct sk_buff *clone;
1275
1276 /* Every output action needs a separate clone
1277 * of 'skb', In case the output action is the
1278 * last action, cloning can be avoided.
1279 */
1280 if (nla_is_last(a, rem)) {
1281 do_output(dp, skb, port, key);
1282 /* 'skb' has been used for output.
1283 */
1284 return 0;
1285 }
1286
1287 clone = skb_clone(skb, GFP_ATOMIC);
1288 if (clone)
1289 do_output(dp, clone, port, key);
1290 OVS_CB(skb)->cutlen = 0;
1291 break;
1292 }
1293
1294 case OVS_ACTION_ATTR_TRUNC: {
1295 struct ovs_action_trunc *trunc = nla_data(a);
1296
1297 if (skb->len > trunc->max_len)
1298 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1299 break;
1300 }
1301
1302 case OVS_ACTION_ATTR_USERSPACE:
1303 output_userspace(dp, skb, key, a, attr,
1304 len, OVS_CB(skb)->cutlen);
1305 OVS_CB(skb)->cutlen = 0;
1306 break;
1307
1308 case OVS_ACTION_ATTR_HASH:
1309 execute_hash(skb, key, a);
1310 break;
1311
1312 case OVS_ACTION_ATTR_PUSH_MPLS:
1313 err = push_mpls(skb, key, nla_data(a));
1314 break;
1315
1316 case OVS_ACTION_ATTR_POP_MPLS:
1317 err = pop_mpls(skb, key, nla_get_be16(a));
1318 break;
1319
1320 case OVS_ACTION_ATTR_PUSH_VLAN:
1321 err = push_vlan(skb, key, nla_data(a));
1322 break;
1323
1324 case OVS_ACTION_ATTR_POP_VLAN:
1325 err = pop_vlan(skb, key);
1326 break;
1327
1328 case OVS_ACTION_ATTR_RECIRC: {
1329 bool last = nla_is_last(a, rem);
1330
1331 err = execute_recirc(dp, skb, key, a, last);
1332 if (last) {
1333 /* If this is the last action, the skb has
1334 * been consumed or freed.
1335 * Return immediately.
1336 */
1337 return err;
1338 }
1339 break;
1340 }
1341
1342 case OVS_ACTION_ATTR_SET:
1343 err = execute_set_action(skb, key, nla_data(a));
1344 break;
1345
1346 case OVS_ACTION_ATTR_SET_MASKED:
1347 case OVS_ACTION_ATTR_SET_TO_MASKED:
1348 err = execute_masked_set_action(skb, key, nla_data(a));
1349 break;
1350
1351 case OVS_ACTION_ATTR_SAMPLE: {
1352 bool last = nla_is_last(a, rem);
1353
1354 err = sample(dp, skb, key, a, last);
1355 if (last)
1356 return err;
1357
1358 break;
1359 }
1360
1361 case OVS_ACTION_ATTR_CT:
1362 if (!is_flow_key_valid(key)) {
1363 err = ovs_flow_key_update(skb, key);
1364 if (err)
1365 return err;
1366 }
1367
1368 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1369 nla_data(a));
1370
1371 /* Hide stolen IP fragments from user space. */
1372 if (err)
1373 return err == -EINPROGRESS ? 0 : err;
1374 break;
1375
1376 case OVS_ACTION_ATTR_CT_CLEAR:
1377 err = ovs_ct_clear(skb, key);
1378 break;
1379
1380 case OVS_ACTION_ATTR_PUSH_ETH:
1381 err = push_eth(skb, key, nla_data(a));
1382 break;
1383
1384 case OVS_ACTION_ATTR_POP_ETH:
1385 err = pop_eth(skb, key);
1386 break;
1387
1388 case OVS_ACTION_ATTR_PUSH_NSH: {
1389 u8 buffer[NSH_HDR_MAX_LEN];
1390 struct nshhdr *nh = (struct nshhdr *)buffer;
1391
1392 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1393 NSH_HDR_MAX_LEN);
1394 if (unlikely(err))
1395 break;
1396 err = push_nsh(skb, key, nh);
1397 break;
1398 }
1399
1400 case OVS_ACTION_ATTR_POP_NSH:
1401 err = pop_nsh(skb, key);
1402 break;
1403
1404 case OVS_ACTION_ATTR_METER:
1405 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1406 consume_skb(skb);
1407 return 0;
1408 }
1409 break;
1410
1411 case OVS_ACTION_ATTR_CLONE: {
1412 bool last = nla_is_last(a, rem);
1413
1414 err = clone(dp, skb, key, a, last);
1415 if (last)
1416 return err;
1417 break;
1418 }
1419
1420 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1421 bool last = nla_is_last(a, rem);
1422
1423 err = execute_check_pkt_len(dp, skb, key, a, last);
1424 if (last)
1425 return err;
1426
1427 break;
1428 }
1429 }
1430
1431 if (unlikely(err)) {
1432 kfree_skb(skb);
1433 return err;
1434 }
1435 }
1436
1437 consume_skb(skb);
1438 return 0;
1439 }
1440
1441 /* Execute the actions on the clone of the packet. The effect of the
1442 * execution does not affect the original 'skb' nor the original 'key'.
1443 *
1444 * The execution may be deferred in case the actions can not be executed
1445 * immediately.
1446 */
1447 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1448 struct sw_flow_key *key, u32 recirc_id,
1449 const struct nlattr *actions, int len,
1450 bool last, bool clone_flow_key)
1451 {
1452 struct deferred_action *da;
1453 struct sw_flow_key *clone;
1454
1455 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1456 if (!skb) {
1457 /* Out of memory, skip this action.
1458 */
1459 return 0;
1460 }
1461
1462 /* When clone_flow_key is false, the 'key' will not be change
1463 * by the actions, then the 'key' can be used directly.
1464 * Otherwise, try to clone key from the next recursion level of
1465 * 'flow_keys'. If clone is successful, execute the actions
1466 * without deferring.
1467 */
1468 clone = clone_flow_key ? clone_key(key) : key;
1469 if (clone) {
1470 int err = 0;
1471
1472 if (actions) { /* Sample action */
1473 if (clone_flow_key)
1474 __this_cpu_inc(exec_actions_level);
1475
1476 err = do_execute_actions(dp, skb, clone,
1477 actions, len);
1478
1479 if (clone_flow_key)
1480 __this_cpu_dec(exec_actions_level);
1481 } else { /* Recirc action */
1482 clone->recirc_id = recirc_id;
1483 ovs_dp_process_packet(skb, clone);
1484 }
1485 return err;
1486 }
1487
1488 /* Out of 'flow_keys' space. Defer actions */
1489 da = add_deferred_actions(skb, key, actions, len);
1490 if (da) {
1491 if (!actions) { /* Recirc action */
1492 key = &da->pkt_key;
1493 key->recirc_id = recirc_id;
1494 }
1495 } else {
1496 /* Out of per CPU action FIFO space. Drop the 'skb' and
1497 * log an error.
1498 */
1499 kfree_skb(skb);
1500
1501 if (net_ratelimit()) {
1502 if (actions) { /* Sample action */
1503 pr_warn("%s: deferred action limit reached, drop sample action\n",
1504 ovs_dp_name(dp));
1505 } else { /* Recirc action */
1506 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1507 ovs_dp_name(dp));
1508 }
1509 }
1510 }
1511 return 0;
1512 }
1513
1514 static void process_deferred_actions(struct datapath *dp)
1515 {
1516 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1517
1518 /* Do not touch the FIFO in case there is no deferred actions. */
1519 if (action_fifo_is_empty(fifo))
1520 return;
1521
1522 /* Finishing executing all deferred actions. */
1523 do {
1524 struct deferred_action *da = action_fifo_get(fifo);
1525 struct sk_buff *skb = da->skb;
1526 struct sw_flow_key *key = &da->pkt_key;
1527 const struct nlattr *actions = da->actions;
1528 int actions_len = da->actions_len;
1529
1530 if (actions)
1531 do_execute_actions(dp, skb, key, actions, actions_len);
1532 else
1533 ovs_dp_process_packet(skb, key);
1534 } while (!action_fifo_is_empty(fifo));
1535
1536 /* Reset FIFO for the next packet. */
1537 action_fifo_init(fifo);
1538 }
1539
1540 /* Execute a list of actions against 'skb'. */
1541 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1542 const struct sw_flow_actions *acts,
1543 struct sw_flow_key *key)
1544 {
1545 int err, level;
1546
1547 level = __this_cpu_inc_return(exec_actions_level);
1548 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1549 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1550 ovs_dp_name(dp));
1551 kfree_skb(skb);
1552 err = -ENETDOWN;
1553 goto out;
1554 }
1555
1556 OVS_CB(skb)->acts_origlen = acts->orig_len;
1557 err = do_execute_actions(dp, skb, key,
1558 acts->actions, acts->actions_len);
1559
1560 if (level == 1)
1561 process_deferred_actions(dp);
1562
1563 out:
1564 __this_cpu_dec(exec_actions_level);
1565 return err;
1566 }
1567
1568 int action_fifos_init(void)
1569 {
1570 action_fifos = alloc_percpu(struct action_fifo);
1571 if (!action_fifos)
1572 return -ENOMEM;
1573
1574 flow_keys = alloc_percpu(struct action_flow_keys);
1575 if (!flow_keys) {
1576 free_percpu(action_fifos);
1577 return -ENOMEM;
1578 }
1579
1580 return 0;
1581 }
1582
1583 void action_fifos_exit(void)
1584 {
1585 free_percpu(action_fifos);
1586 free_percpu(flow_keys);
1587 }
openvswitch packages for PVE