~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/openvswitch/actions.c

Version: ~ [ linux-6.0-rc1 ] ~ [ linux-5.19.1 ] ~ [ linux-5.18.17 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.60 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.136 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.210 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.255 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.290 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.325 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/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 #include "flow_netlink.h"
 47 
 48 struct deferred_action {
 49         struct sk_buff *skb;
 50         const struct nlattr *actions;
 51         int actions_len;
 52 
 53         /* Store pkt_key clone when creating deferred action. */
 54         struct sw_flow_key pkt_key;
 55 };
 56 
 57 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
 58 struct ovs_frag_data {
 59         unsigned long dst;
 60         struct vport *vport;
 61         struct ovs_skb_cb cb;
 62         __be16 inner_protocol;
 63         u16 network_offset;     /* valid only for MPLS */
 64         u16 vlan_tci;
 65         __be16 vlan_proto;
 66         unsigned int l2_len;
 67         u8 mac_proto;
 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 #define OVS_RECURSION_LIMIT 5
 75 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
 76 struct action_fifo {
 77         int head;
 78         int tail;
 79         /* Deferred action fifo queue storage. */
 80         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
 81 };
 82 
 83 struct action_flow_keys {
 84         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
 85 };
 86 
 87 static struct action_fifo __percpu *action_fifos;
 88 static struct action_flow_keys __percpu *flow_keys;
 89 static DEFINE_PER_CPU(int, exec_actions_level);
 90 
 91 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
 92  * space. Return NULL if out of key spaces.
 93  */
 94 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
 95 {
 96         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
 97         int level = this_cpu_read(exec_actions_level);
 98         struct sw_flow_key *key = NULL;
 99 
100         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
101                 key = &keys->key[level - 1];
102                 *key = *key_;
103         }
104 
105         return key;
106 }
107 
108 static void action_fifo_init(struct action_fifo *fifo)
109 {
110         fifo->head = 0;
111         fifo->tail = 0;
112 }
113 
114 static bool action_fifo_is_empty(const struct action_fifo *fifo)
115 {
116         return (fifo->head == fifo->tail);
117 }
118 
119 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
120 {
121         if (action_fifo_is_empty(fifo))
122                 return NULL;
123 
124         return &fifo->fifo[fifo->tail++];
125 }
126 
127 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
128 {
129         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
130                 return NULL;
131 
132         return &fifo->fifo[fifo->head++];
133 }
134 
135 /* Return true if fifo is not full */
136 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
137                                     const struct sw_flow_key *key,
138                                     const struct nlattr *actions,
139                                     const int actions_len)
140 {
141         struct action_fifo *fifo;
142         struct deferred_action *da;
143 
144         fifo = this_cpu_ptr(action_fifos);
145         da = action_fifo_put(fifo);
146         if (da) {
147                 da->skb = skb;
148                 da->actions = actions;
149                 da->actions_len = actions_len;
150                 da->pkt_key = *key;
151         }
152 
153         return da;
154 }
155 
156 static void invalidate_flow_key(struct sw_flow_key *key)
157 {
158         key->mac_proto |= SW_FLOW_KEY_INVALID;
159 }
160 
161 static bool is_flow_key_valid(const struct sw_flow_key *key)
162 {
163         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
164 }
165 
166 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
167                          struct sw_flow_key *key,
168                          u32 recirc_id,
169                          const struct nlattr *actions, int len,
170                          bool last, bool clone_flow_key);
171 
172 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
173                              __be16 ethertype)
174 {
175         if (skb->ip_summed == CHECKSUM_COMPLETE) {
176                 __be16 diff[] = { ~(hdr->h_proto), ethertype };
177 
178                 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
179         }
180 
181         hdr->h_proto = ethertype;
182 }
183 
184 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
185                      const struct ovs_action_push_mpls *mpls)
186 {
187         struct mpls_shim_hdr *new_mpls_lse;
188 
189         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
190         if (skb->encapsulation)
191                 return -ENOTSUPP;
192 
193         if (skb_cow_head(skb, MPLS_HLEN) < 0)
194                 return -ENOMEM;
195 
196         if (!skb->inner_protocol) {
197                 skb_set_inner_network_header(skb, skb->mac_len);
198                 skb_set_inner_protocol(skb, skb->protocol);
199         }
200 
201         skb_push(skb, MPLS_HLEN);
202         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
203                 skb->mac_len);
204         skb_reset_mac_header(skb);
205         skb_set_network_header(skb, skb->mac_len);
206 
207         new_mpls_lse = mpls_hdr(skb);
208         new_mpls_lse->label_stack_entry = mpls->mpls_lse;
209 
210         skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
211 
212         if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
213                 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
214         skb->protocol = mpls->mpls_ethertype;
215 
216         invalidate_flow_key(key);
217         return 0;
218 }
219 
220 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
221                     const __be16 ethertype)
222 {
223         int err;
224 
225         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
226         if (unlikely(err))
227                 return err;
228 
229         skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
230 
231         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
232                 skb->mac_len);
233 
234         __skb_pull(skb, MPLS_HLEN);
235         skb_reset_mac_header(skb);
236         skb_set_network_header(skb, skb->mac_len);
237 
238         if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
239                 struct ethhdr *hdr;
240 
241                 /* mpls_hdr() is used to locate the ethertype field correctly in the
242                  * presence of VLAN tags.
243                  */
244                 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
245                 update_ethertype(skb, hdr, ethertype);
246         }
247         if (eth_p_mpls(skb->protocol))
248                 skb->protocol = ethertype;
249 
250         invalidate_flow_key(key);
251         return 0;
252 }
253 
254 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
255                     const __be32 *mpls_lse, const __be32 *mask)
256 {
257         struct mpls_shim_hdr *stack;
258         __be32 lse;
259         int err;
260 
261         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
262         if (unlikely(err))
263                 return err;
264 
265         stack = mpls_hdr(skb);
266         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
267         if (skb->ip_summed == CHECKSUM_COMPLETE) {
268                 __be32 diff[] = { ~(stack->label_stack_entry), lse };
269 
270                 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
271         }
272 
273         stack->label_stack_entry = lse;
274         flow_key->mpls.top_lse = lse;
275         return 0;
276 }
277 
278 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
279 {
280         int err;
281 
282         err = skb_vlan_pop(skb);
283         if (skb_vlan_tag_present(skb)) {
284                 invalidate_flow_key(key);
285         } else {
286                 key->eth.vlan.tci = 0;
287                 key->eth.vlan.tpid = 0;
288         }
289         return err;
290 }
291 
292 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
293                      const struct ovs_action_push_vlan *vlan)
294 {
295         if (skb_vlan_tag_present(skb)) {
296                 invalidate_flow_key(key);
297         } else {
298                 key->eth.vlan.tci = vlan->vlan_tci;
299                 key->eth.vlan.tpid = vlan->vlan_tpid;
300         }
301         return skb_vlan_push(skb, vlan->vlan_tpid,
302                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
303 }
304 
305 /* 'src' is already properly masked. */
306 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
307 {
308         u16 *dst = (u16 *)dst_;
309         const u16 *src = (const u16 *)src_;
310         const u16 *mask = (const u16 *)mask_;
311 
312         OVS_SET_MASKED(dst[0], src[0], mask[0]);
313         OVS_SET_MASKED(dst[1], src[1], mask[1]);
314         OVS_SET_MASKED(dst[2], src[2], mask[2]);
315 }
316 
317 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
318                         const struct ovs_key_ethernet *key,
319                         const struct ovs_key_ethernet *mask)
320 {
321         int err;
322 
323         err = skb_ensure_writable(skb, ETH_HLEN);
324         if (unlikely(err))
325                 return err;
326 
327         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
328 
329         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
330                                mask->eth_src);
331         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
332                                mask->eth_dst);
333 
334         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
335 
336         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
337         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
338         return 0;
339 }
340 
341 /* pop_eth does not support VLAN packets as this action is never called
342  * for them.
343  */
344 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
345 {
346         skb_pull_rcsum(skb, ETH_HLEN);
347         skb_reset_mac_header(skb);
348         skb_reset_mac_len(skb);
349 
350         /* safe right before invalidate_flow_key */
351         key->mac_proto = MAC_PROTO_NONE;
352         invalidate_flow_key(key);
353         return 0;
354 }
355 
356 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
357                     const struct ovs_action_push_eth *ethh)
358 {
359         struct ethhdr *hdr;
360 
361         /* Add the new Ethernet header */
362         if (skb_cow_head(skb, ETH_HLEN) < 0)
363                 return -ENOMEM;
364 
365         skb_push(skb, ETH_HLEN);
366         skb_reset_mac_header(skb);
367         skb_reset_mac_len(skb);
368 
369         hdr = eth_hdr(skb);
370         ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
371         ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
372         hdr->h_proto = skb->protocol;
373 
374         skb_postpush_rcsum(skb, hdr, ETH_HLEN);
375 
376         /* safe right before invalidate_flow_key */
377         key->mac_proto = MAC_PROTO_ETHERNET;
378         invalidate_flow_key(key);
379         return 0;
380 }
381 
382 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
383                     const struct nshhdr *nh)
384 {
385         int err;
386 
387         err = nsh_push(skb, nh);
388         if (err)
389                 return err;
390 
391         /* safe right before invalidate_flow_key */
392         key->mac_proto = MAC_PROTO_NONE;
393         invalidate_flow_key(key);
394         return 0;
395 }
396 
397 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
398 {
399         int err;
400 
401         err = nsh_pop(skb);
402         if (err)
403                 return err;
404 
405         /* safe right before invalidate_flow_key */
406         if (skb->protocol == htons(ETH_P_TEB))
407                 key->mac_proto = MAC_PROTO_ETHERNET;
408         else
409                 key->mac_proto = MAC_PROTO_NONE;
410         invalidate_flow_key(key);
411         return 0;
412 }
413 
414 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
415                                   __be32 addr, __be32 new_addr)
416 {
417         int transport_len = skb->len - skb_transport_offset(skb);
418 
419         if (nh->frag_off & htons(IP_OFFSET))
420                 return;
421 
422         if (nh->protocol == IPPROTO_TCP) {
423                 if (likely(transport_len >= sizeof(struct tcphdr)))
424                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
425                                                  addr, new_addr, true);
426         } else if (nh->protocol == IPPROTO_UDP) {
427                 if (likely(transport_len >= sizeof(struct udphdr))) {
428                         struct udphdr *uh = udp_hdr(skb);
429 
430                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
431                                 inet_proto_csum_replace4(&uh->check, skb,
432                                                          addr, new_addr, true);
433                                 if (!uh->check)
434                                         uh->check = CSUM_MANGLED_0;
435                         }
436                 }
437         }
438 }
439 
440 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
441                         __be32 *addr, __be32 new_addr)
442 {
443         update_ip_l4_checksum(skb, nh, *addr, new_addr);
444         csum_replace4(&nh->check, *addr, new_addr);
445         skb_clear_hash(skb);
446         *addr = new_addr;
447 }
448 
449 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
450                                  __be32 addr[4], const __be32 new_addr[4])
451 {
452         int transport_len = skb->len - skb_transport_offset(skb);
453 
454         if (l4_proto == NEXTHDR_TCP) {
455                 if (likely(transport_len >= sizeof(struct tcphdr)))
456                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
457                                                   addr, new_addr, true);
458         } else if (l4_proto == NEXTHDR_UDP) {
459                 if (likely(transport_len >= sizeof(struct udphdr))) {
460                         struct udphdr *uh = udp_hdr(skb);
461 
462                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
463                                 inet_proto_csum_replace16(&uh->check, skb,
464                                                           addr, new_addr, true);
465                                 if (!uh->check)
466                                         uh->check = CSUM_MANGLED_0;
467                         }
468                 }
469         } else if (l4_proto == NEXTHDR_ICMP) {
470                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
471                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
472                                                   skb, addr, new_addr, true);
473         }
474 }
475 
476 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
477                            const __be32 mask[4], __be32 masked[4])
478 {
479         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
480         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
481         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
482         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
483 }
484 
485 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
486                           __be32 addr[4], const __be32 new_addr[4],
487                           bool recalculate_csum)
488 {
489         if (recalculate_csum)
490                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
491 
492         skb_clear_hash(skb);
493         memcpy(addr, new_addr, sizeof(__be32[4]));
494 }
495 
496 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
497 {
498         /* Bits 21-24 are always unmasked, so this retains their values. */
499         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
500         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
501         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
502 }
503 
504 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
505                        u8 mask)
506 {
507         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
508 
509         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
510         nh->ttl = new_ttl;
511 }
512 
513 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
514                     const struct ovs_key_ipv4 *key,
515                     const struct ovs_key_ipv4 *mask)
516 {
517         struct iphdr *nh;
518         __be32 new_addr;
519         int err;
520 
521         err = skb_ensure_writable(skb, skb_network_offset(skb) +
522                                   sizeof(struct iphdr));
523         if (unlikely(err))
524                 return err;
525 
526         nh = ip_hdr(skb);
527 
528         /* Setting an IP addresses is typically only a side effect of
529          * matching on them in the current userspace implementation, so it
530          * makes sense to check if the value actually changed.
531          */
532         if (mask->ipv4_src) {
533                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
534 
535                 if (unlikely(new_addr != nh->saddr)) {
536                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
537                         flow_key->ipv4.addr.src = new_addr;
538                 }
539         }
540         if (mask->ipv4_dst) {
541                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
542 
543                 if (unlikely(new_addr != nh->daddr)) {
544                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
545                         flow_key->ipv4.addr.dst = new_addr;
546                 }
547         }
548         if (mask->ipv4_tos) {
549                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
550                 flow_key->ip.tos = nh->tos;
551         }
552         if (mask->ipv4_ttl) {
553                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
554                 flow_key->ip.ttl = nh->ttl;
555         }
556 
557         return 0;
558 }
559 
560 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
561 {
562         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
563 }
564 
565 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
566                     const struct ovs_key_ipv6 *key,
567                     const struct ovs_key_ipv6 *mask)
568 {
569         struct ipv6hdr *nh;
570         int err;
571 
572         err = skb_ensure_writable(skb, skb_network_offset(skb) +
573                                   sizeof(struct ipv6hdr));
574         if (unlikely(err))
575                 return err;
576 
577         nh = ipv6_hdr(skb);
578 
579         /* Setting an IP addresses is typically only a side effect of
580          * matching on them in the current userspace implementation, so it
581          * makes sense to check if the value actually changed.
582          */
583         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
584                 __be32 *saddr = (__be32 *)&nh->saddr;
585                 __be32 masked[4];
586 
587                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
588 
589                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
590                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
591                                       true);
592                         memcpy(&flow_key->ipv6.addr.src, masked,
593                                sizeof(flow_key->ipv6.addr.src));
594                 }
595         }
596         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
597                 unsigned int offset = 0;
598                 int flags = IP6_FH_F_SKIP_RH;
599                 bool recalc_csum = true;
600                 __be32 *daddr = (__be32 *)&nh->daddr;
601                 __be32 masked[4];
602 
603                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
604 
605                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
606                         if (ipv6_ext_hdr(nh->nexthdr))
607                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
608                                                              NEXTHDR_ROUTING,
609                                                              NULL, &flags)
610                                                != NEXTHDR_ROUTING);
611 
612                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
613                                       recalc_csum);
614                         memcpy(&flow_key->ipv6.addr.dst, masked,
615                                sizeof(flow_key->ipv6.addr.dst));
616                 }
617         }
618         if (mask->ipv6_tclass) {
619                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
620                 flow_key->ip.tos = ipv6_get_dsfield(nh);
621         }
622         if (mask->ipv6_label) {
623                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
624                             ntohl(mask->ipv6_label));
625                 flow_key->ipv6.label =
626                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
627         }
628         if (mask->ipv6_hlimit) {
629                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
630                                mask->ipv6_hlimit);
631                 flow_key->ip.ttl = nh->hop_limit;
632         }
633         return 0;
634 }
635 
636 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
637                    const struct nlattr *a)
638 {
639         struct nshhdr *nh;
640         size_t length;
641         int err;
642         u8 flags;
643         u8 ttl;
644         int i;
645 
646         struct ovs_key_nsh key;
647         struct ovs_key_nsh mask;
648 
649         err = nsh_key_from_nlattr(a, &key, &mask);
650         if (err)
651                 return err;
652 
653         /* Make sure the NSH base header is there */
654         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
655                 return -ENOMEM;
656 
657         nh = nsh_hdr(skb);
658         length = nsh_hdr_len(nh);
659 
660         /* Make sure the whole NSH header is there */
661         err = skb_ensure_writable(skb, skb_network_offset(skb) +
662                                        length);
663         if (unlikely(err))
664                 return err;
665 
666         nh = nsh_hdr(skb);
667         skb_postpull_rcsum(skb, nh, length);
668         flags = nsh_get_flags(nh);
669         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
670         flow_key->nsh.base.flags = flags;
671         ttl = nsh_get_ttl(nh);
672         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
673         flow_key->nsh.base.ttl = ttl;
674         nsh_set_flags_and_ttl(nh, flags, ttl);
675         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
676                                   mask.base.path_hdr);
677         flow_key->nsh.base.path_hdr = nh->path_hdr;
678         switch (nh->mdtype) {
679         case NSH_M_TYPE1:
680                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
681                         nh->md1.context[i] =
682                             OVS_MASKED(nh->md1.context[i], key.context[i],
683                                        mask.context[i]);
684                 }
685                 memcpy(flow_key->nsh.context, nh->md1.context,
686                        sizeof(nh->md1.context));
687                 break;
688         case NSH_M_TYPE2:
689                 memset(flow_key->nsh.context, 0,
690                        sizeof(flow_key->nsh.context));
691                 break;
692         default:
693                 return -EINVAL;
694         }
695         skb_postpush_rcsum(skb, nh, length);
696         return 0;
697 }
698 
699 /* Must follow skb_ensure_writable() since that can move the skb data. */
700 static void set_tp_port(struct sk_buff *skb, __be16 *port,
701                         __be16 new_port, __sum16 *check)
702 {
703         inet_proto_csum_replace2(check, skb, *port, new_port, false);
704         *port = new_port;
705 }
706 
707 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
708                    const struct ovs_key_udp *key,
709                    const struct ovs_key_udp *mask)
710 {
711         struct udphdr *uh;
712         __be16 src, dst;
713         int err;
714 
715         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
716                                   sizeof(struct udphdr));
717         if (unlikely(err))
718                 return err;
719 
720         uh = udp_hdr(skb);
721         /* Either of the masks is non-zero, so do not bother checking them. */
722         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
723         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
724 
725         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
726                 if (likely(src != uh->source)) {
727                         set_tp_port(skb, &uh->source, src, &uh->check);
728                         flow_key->tp.src = src;
729                 }
730                 if (likely(dst != uh->dest)) {
731                         set_tp_port(skb, &uh->dest, dst, &uh->check);
732                         flow_key->tp.dst = dst;
733                 }
734 
735                 if (unlikely(!uh->check))
736                         uh->check = CSUM_MANGLED_0;
737         } else {
738                 uh->source = src;
739                 uh->dest = dst;
740                 flow_key->tp.src = src;
741                 flow_key->tp.dst = dst;
742         }
743 
744         skb_clear_hash(skb);
745 
746         return 0;
747 }
748 
749 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
750                    const struct ovs_key_tcp *key,
751                    const struct ovs_key_tcp *mask)
752 {
753         struct tcphdr *th;
754         __be16 src, dst;
755         int err;
756 
757         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
758                                   sizeof(struct tcphdr));
759         if (unlikely(err))
760                 return err;
761 
762         th = tcp_hdr(skb);
763         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
764         if (likely(src != th->source)) {
765                 set_tp_port(skb, &th->source, src, &th->check);
766                 flow_key->tp.src = src;
767         }
768         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
769         if (likely(dst != th->dest)) {
770                 set_tp_port(skb, &th->dest, dst, &th->check);
771                 flow_key->tp.dst = dst;
772         }
773         skb_clear_hash(skb);
774 
775         return 0;
776 }
777 
778 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
779                     const struct ovs_key_sctp *key,
780                     const struct ovs_key_sctp *mask)
781 {
782         unsigned int sctphoff = skb_transport_offset(skb);
783         struct sctphdr *sh;
784         __le32 old_correct_csum, new_csum, old_csum;
785         int err;
786 
787         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
788         if (unlikely(err))
789                 return err;
790 
791         sh = sctp_hdr(skb);
792         old_csum = sh->checksum;
793         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
794 
795         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
796         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
797 
798         new_csum = sctp_compute_cksum(skb, sctphoff);
799 
800         /* Carry any checksum errors through. */
801         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
802 
803         skb_clear_hash(skb);
804         flow_key->tp.src = sh->source;
805         flow_key->tp.dst = sh->dest;
806 
807         return 0;
808 }
809 
810 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
811 {
812         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
813         struct vport *vport = data->vport;
814 
815         if (skb_cow_head(skb, data->l2_len) < 0) {
816                 kfree_skb(skb);
817                 return -ENOMEM;
818         }
819 
820         __skb_dst_copy(skb, data->dst);
821         *OVS_CB(skb) = data->cb;
822         skb->inner_protocol = data->inner_protocol;
823         if (data->vlan_tci & VLAN_CFI_MASK)
824                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
825         else
826                 __vlan_hwaccel_clear_tag(skb);
827 
828         /* Reconstruct the MAC header.  */
829         skb_push(skb, data->l2_len);
830         memcpy(skb->data, &data->l2_data, data->l2_len);
831         skb_postpush_rcsum(skb, skb->data, data->l2_len);
832         skb_reset_mac_header(skb);
833 
834         if (eth_p_mpls(skb->protocol)) {
835                 skb->inner_network_header = skb->network_header;
836                 skb_set_network_header(skb, data->network_offset);
837                 skb_reset_mac_len(skb);
838         }
839 
840         ovs_vport_send(vport, skb, data->mac_proto);
841         return 0;
842 }
843 
844 static unsigned int
845 ovs_dst_get_mtu(const struct dst_entry *dst)
846 {
847         return dst->dev->mtu;
848 }
849 
850 static struct dst_ops ovs_dst_ops = {
851         .family = AF_UNSPEC,
852         .mtu = ovs_dst_get_mtu,
853 };
854 
855 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
856  * ovs_vport_output(), which is called once per fragmented packet.
857  */
858 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
859                          u16 orig_network_offset, u8 mac_proto)
860 {
861         unsigned int hlen = skb_network_offset(skb);
862         struct ovs_frag_data *data;
863 
864         data = this_cpu_ptr(&ovs_frag_data_storage);
865         data->dst = skb->_skb_refdst;
866         data->vport = vport;
867         data->cb = *OVS_CB(skb);
868         data->inner_protocol = skb->inner_protocol;
869         data->network_offset = orig_network_offset;
870         if (skb_vlan_tag_present(skb))
871                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
872         else
873                 data->vlan_tci = 0;
874         data->vlan_proto = skb->vlan_proto;
875         data->mac_proto = mac_proto;
876         data->l2_len = hlen;
877         memcpy(&data->l2_data, skb->data, hlen);
878 
879         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
880         skb_pull(skb, hlen);
881 }
882 
883 static void ovs_fragment(struct net *net, struct vport *vport,
884                          struct sk_buff *skb, u16 mru,
885                          struct sw_flow_key *key)
886 {
887         u16 orig_network_offset = 0;
888 
889         if (eth_p_mpls(skb->protocol)) {
890                 orig_network_offset = skb_network_offset(skb);
891                 skb->network_header = skb->inner_network_header;
892         }
893 
894         if (skb_network_offset(skb) > MAX_L2_LEN) {
895                 OVS_NLERR(1, "L2 header too long to fragment");
896                 goto err;
897         }
898 
899         if (key->eth.type == htons(ETH_P_IP)) {
900                 struct dst_entry ovs_dst;
901                 unsigned long orig_dst;
902 
903                 prepare_frag(vport, skb, orig_network_offset,
904                              ovs_key_mac_proto(key));
905                 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
906                          DST_OBSOLETE_NONE, DST_NOCOUNT);
907                 ovs_dst.dev = vport->dev;
908 
909                 orig_dst = skb->_skb_refdst;
910                 skb_dst_set_noref(skb, &ovs_dst);
911                 IPCB(skb)->frag_max_size = mru;
912 
913                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
914                 refdst_drop(orig_dst);
915         } else if (key->eth.type == htons(ETH_P_IPV6)) {
916                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
917                 unsigned long orig_dst;
918                 struct rt6_info ovs_rt;
919 
920                 if (!v6ops)
921                         goto err;
922 
923                 prepare_frag(vport, skb, orig_network_offset,
924                              ovs_key_mac_proto(key));
925                 memset(&ovs_rt, 0, sizeof(ovs_rt));
926                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
927                          DST_OBSOLETE_NONE, DST_NOCOUNT);
928                 ovs_rt.dst.dev = vport->dev;
929 
930                 orig_dst = skb->_skb_refdst;
931                 skb_dst_set_noref(skb, &ovs_rt.dst);
932                 IP6CB(skb)->frag_max_size = mru;
933 
934                 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
935                 refdst_drop(orig_dst);
936         } else {
937                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
938                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
939                           vport->dev->mtu);
940                 goto err;
941         }
942 
943         return;
944 err:
945         kfree_skb(skb);
946 }
947 
948 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
949                       struct sw_flow_key *key)
950 {
951         struct vport *vport = ovs_vport_rcu(dp, out_port);
952 
953         if (likely(vport)) {
954                 u16 mru = OVS_CB(skb)->mru;
955                 u32 cutlen = OVS_CB(skb)->cutlen;
956 
957                 if (unlikely(cutlen > 0)) {
958                         if (skb->len - cutlen > ovs_mac_header_len(key))
959                                 pskb_trim(skb, skb->len - cutlen);
960                         else
961                                 pskb_trim(skb, ovs_mac_header_len(key));
962                 }
963 
964                 if (likely(!mru ||
965                            (skb->len <= mru + vport->dev->hard_header_len))) {
966                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
967                 } else if (mru <= vport->dev->mtu) {
968                         struct net *net = read_pnet(&dp->net);
969 
970                         ovs_fragment(net, vport, skb, mru, key);
971                 } else {
972                         kfree_skb(skb);
973                 }
974         } else {
975                 kfree_skb(skb);
976         }
977 }
978 
979 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
980                             struct sw_flow_key *key, const struct nlattr *attr,
981                             const struct nlattr *actions, int actions_len,
982                             uint32_t cutlen)
983 {
984         struct dp_upcall_info upcall;
985         const struct nlattr *a;
986         int rem;
987 
988         memset(&upcall, 0, sizeof(upcall));
989         upcall.cmd = OVS_PACKET_CMD_ACTION;
990         upcall.mru = OVS_CB(skb)->mru;
991 
992         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
993                  a = nla_next(a, &rem)) {
994                 switch (nla_type(a)) {
995                 case OVS_USERSPACE_ATTR_USERDATA:
996                         upcall.userdata = a;
997                         break;
998 
999                 case OVS_USERSPACE_ATTR_PID:
1000                         upcall.portid = nla_get_u32(a);
1001                         break;
1002 
1003                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1004                         /* Get out tunnel info. */
1005                         struct vport *vport;
1006 
1007                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
1008                         if (vport) {
1009                                 int err;
1010 
1011                                 err = dev_fill_metadata_dst(vport->dev, skb);
1012                                 if (!err)
1013                                         upcall.egress_tun_info = skb_tunnel_info(skb);
1014                         }
1015 
1016                         break;
1017                 }
1018 
1019                 case OVS_USERSPACE_ATTR_ACTIONS: {
1020                         /* Include actions. */
1021                         upcall.actions = actions;
1022                         upcall.actions_len = actions_len;
1023                         break;
1024                 }
1025 
1026                 } /* End of switch. */
1027         }
1028 
1029         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1030 }
1031 
1032 /* When 'last' is true, sample() should always consume the 'skb'.
1033  * Otherwise, sample() should keep 'skb' intact regardless what
1034  * actions are executed within sample().
1035  */
1036 static int sample(struct datapath *dp, struct sk_buff *skb,
1037                   struct sw_flow_key *key, const struct nlattr *attr,
1038                   bool last)
1039 {
1040         struct nlattr *actions;
1041         struct nlattr *sample_arg;
1042         int rem = nla_len(attr);
1043         const struct sample_arg *arg;
1044         bool clone_flow_key;
1045 
1046         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1047         sample_arg = nla_data(attr);
1048         arg = nla_data(sample_arg);
1049         actions = nla_next(sample_arg, &rem);
1050 
1051         if ((arg->probability != U32_MAX) &&
1052             (!arg->probability || prandom_u32() > arg->probability)) {
1053                 if (last)
1054                         consume_skb(skb);
1055                 return 0;
1056         }
1057 
1058         clone_flow_key = !arg->exec;
1059         return clone_execute(dp, skb, key, 0, actions, rem, last,
1060                              clone_flow_key);
1061 }
1062 
1063 /* When 'last' is true, clone() should always consume the 'skb'.
1064  * Otherwise, clone() should keep 'skb' intact regardless what
1065  * actions are executed within clone().
1066  */
1067 static int clone(struct datapath *dp, struct sk_buff *skb,
1068                  struct sw_flow_key *key, const struct nlattr *attr,
1069                  bool last)
1070 {
1071         struct nlattr *actions;
1072         struct nlattr *clone_arg;
1073         int rem = nla_len(attr);
1074         bool dont_clone_flow_key;
1075 
1076         /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1077         clone_arg = nla_data(attr);
1078         dont_clone_flow_key = nla_get_u32(clone_arg);
1079         actions = nla_next(clone_arg, &rem);
1080 
1081         return clone_execute(dp, skb, key, 0, actions, rem, last,
1082                              !dont_clone_flow_key);
1083 }
1084 
1085 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1086                          const struct nlattr *attr)
1087 {
1088         struct ovs_action_hash *hash_act = nla_data(attr);
1089         u32 hash = 0;
1090 
1091         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1092         hash = skb_get_hash(skb);
1093         hash = jhash_1word(hash, hash_act->hash_basis);
1094         if (!hash)
1095                 hash = 0x1;
1096 
1097         key->ovs_flow_hash = hash;
1098 }
1099 
1100 static int execute_set_action(struct sk_buff *skb,
1101                               struct sw_flow_key *flow_key,
1102                               const struct nlattr *a)
1103 {
1104         /* Only tunnel set execution is supported without a mask. */
1105         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1106                 struct ovs_tunnel_info *tun = nla_data(a);
1107 
1108                 skb_dst_drop(skb);
1109                 dst_hold((struct dst_entry *)tun->tun_dst);
1110                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1111                 return 0;
1112         }
1113 
1114         return -EINVAL;
1115 }
1116 
1117 /* Mask is at the midpoint of the data. */
1118 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1119 
1120 static int execute_masked_set_action(struct sk_buff *skb,
1121                                      struct sw_flow_key *flow_key,
1122                                      const struct nlattr *a)
1123 {
1124         int err = 0;
1125 
1126         switch (nla_type(a)) {
1127         case OVS_KEY_ATTR_PRIORITY:
1128                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1129                                *get_mask(a, u32 *));
1130                 flow_key->phy.priority = skb->priority;
1131                 break;
1132 
1133         case OVS_KEY_ATTR_SKB_MARK:
1134                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1135                 flow_key->phy.skb_mark = skb->mark;
1136                 break;
1137 
1138         case OVS_KEY_ATTR_TUNNEL_INFO:
1139                 /* Masked data not supported for tunnel. */
1140                 err = -EINVAL;
1141                 break;
1142 
1143         case OVS_KEY_ATTR_ETHERNET:
1144                 err = set_eth_addr(skb, flow_key, nla_data(a),
1145                                    get_mask(a, struct ovs_key_ethernet *));
1146                 break;
1147 
1148         case OVS_KEY_ATTR_NSH:
1149                 err = set_nsh(skb, flow_key, a);
1150                 break;
1151 
1152         case OVS_KEY_ATTR_IPV4:
1153                 err = set_ipv4(skb, flow_key, nla_data(a),
1154                                get_mask(a, struct ovs_key_ipv4 *));
1155                 break;
1156 
1157         case OVS_KEY_ATTR_IPV6:
1158                 err = set_ipv6(skb, flow_key, nla_data(a),
1159                                get_mask(a, struct ovs_key_ipv6 *));
1160                 break;
1161 
1162         case OVS_KEY_ATTR_TCP:
1163                 err = set_tcp(skb, flow_key, nla_data(a),
1164                               get_mask(a, struct ovs_key_tcp *));
1165                 break;
1166 
1167         case OVS_KEY_ATTR_UDP:
1168                 err = set_udp(skb, flow_key, nla_data(a),
1169                               get_mask(a, struct ovs_key_udp *));
1170                 break;
1171 
1172         case OVS_KEY_ATTR_SCTP:
1173                 err = set_sctp(skb, flow_key, nla_data(a),
1174                                get_mask(a, struct ovs_key_sctp *));
1175                 break;
1176 
1177         case OVS_KEY_ATTR_MPLS:
1178                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1179                                                                     __be32 *));
1180                 break;
1181 
1182         case OVS_KEY_ATTR_CT_STATE:
1183         case OVS_KEY_ATTR_CT_ZONE:
1184         case OVS_KEY_ATTR_CT_MARK:
1185         case OVS_KEY_ATTR_CT_LABELS:
1186         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1187         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1188                 err = -EINVAL;
1189                 break;
1190         }
1191 
1192         return err;
1193 }
1194 
1195 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1196                           struct sw_flow_key *key,
1197                           const struct nlattr *a, bool last)
1198 {
1199         u32 recirc_id;
1200 
1201         if (!is_flow_key_valid(key)) {
1202                 int err;
1203 
1204                 err = ovs_flow_key_update(skb, key);
1205                 if (err)
1206                         return err;
1207         }
1208         BUG_ON(!is_flow_key_valid(key));
1209 
1210         recirc_id = nla_get_u32(a);
1211         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1212 }
1213 
1214 /* Execute a list of actions against 'skb'. */
1215 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1216                               struct sw_flow_key *key,
1217                               const struct nlattr *attr, int len)
1218 {
1219         const struct nlattr *a;
1220         int rem;
1221 
1222         for (a = attr, rem = len; rem > 0;
1223              a = nla_next(a, &rem)) {
1224                 int err = 0;
1225 
1226                 switch (nla_type(a)) {
1227                 case OVS_ACTION_ATTR_OUTPUT: {
1228                         int port = nla_get_u32(a);
1229                         struct sk_buff *clone;
1230 
1231                         /* Every output action needs a separate clone
1232                          * of 'skb', In case the output action is the
1233                          * last action, cloning can be avoided.
1234                          */
1235                         if (nla_is_last(a, rem)) {
1236                                 do_output(dp, skb, port, key);
1237                                 /* 'skb' has been used for output.
1238                                  */
1239                                 return 0;
1240                         }
1241 
1242                         clone = skb_clone(skb, GFP_ATOMIC);
1243                         if (clone)
1244                                 do_output(dp, clone, port, key);
1245                         OVS_CB(skb)->cutlen = 0;
1246                         break;
1247                 }
1248 
1249                 case OVS_ACTION_ATTR_TRUNC: {
1250                         struct ovs_action_trunc *trunc = nla_data(a);
1251 
1252                         if (skb->len > trunc->max_len)
1253                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1254                         break;
1255                 }
1256 
1257                 case OVS_ACTION_ATTR_USERSPACE:
1258                         output_userspace(dp, skb, key, a, attr,
1259                                                      len, OVS_CB(skb)->cutlen);
1260                         OVS_CB(skb)->cutlen = 0;
1261                         break;
1262 
1263                 case OVS_ACTION_ATTR_HASH:
1264                         execute_hash(skb, key, a);
1265                         break;
1266 
1267                 case OVS_ACTION_ATTR_PUSH_MPLS:
1268                         err = push_mpls(skb, key, nla_data(a));
1269                         break;
1270 
1271                 case OVS_ACTION_ATTR_POP_MPLS:
1272                         err = pop_mpls(skb, key, nla_get_be16(a));
1273                         break;
1274 
1275                 case OVS_ACTION_ATTR_PUSH_VLAN:
1276                         err = push_vlan(skb, key, nla_data(a));
1277                         break;
1278 
1279                 case OVS_ACTION_ATTR_POP_VLAN:
1280                         err = pop_vlan(skb, key);
1281                         break;
1282 
1283                 case OVS_ACTION_ATTR_RECIRC: {
1284                         bool last = nla_is_last(a, rem);
1285 
1286                         err = execute_recirc(dp, skb, key, a, last);
1287                         if (last) {
1288                                 /* If this is the last action, the skb has
1289                                  * been consumed or freed.
1290                                  * Return immediately.
1291                                  */
1292                                 return err;
1293                         }
1294                         break;
1295                 }
1296 
1297                 case OVS_ACTION_ATTR_SET:
1298                         err = execute_set_action(skb, key, nla_data(a));
1299                         break;
1300 
1301                 case OVS_ACTION_ATTR_SET_MASKED:
1302                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1303                         err = execute_masked_set_action(skb, key, nla_data(a));
1304                         break;
1305 
1306                 case OVS_ACTION_ATTR_SAMPLE: {
1307                         bool last = nla_is_last(a, rem);
1308 
1309                         err = sample(dp, skb, key, a, last);
1310                         if (last)
1311                                 return err;
1312 
1313                         break;
1314                 }
1315 
1316                 case OVS_ACTION_ATTR_CT:
1317                         if (!is_flow_key_valid(key)) {
1318                                 err = ovs_flow_key_update(skb, key);
1319                                 if (err)
1320                                         return err;
1321                         }
1322 
1323                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1324                                              nla_data(a));
1325 
1326                         /* Hide stolen IP fragments from user space. */
1327                         if (err)
1328                                 return err == -EINPROGRESS ? 0 : err;
1329                         break;
1330 
1331                 case OVS_ACTION_ATTR_CT_CLEAR:
1332                         err = ovs_ct_clear(skb, key);
1333                         break;
1334 
1335                 case OVS_ACTION_ATTR_PUSH_ETH:
1336                         err = push_eth(skb, key, nla_data(a));
1337                         break;
1338 
1339                 case OVS_ACTION_ATTR_POP_ETH:
1340                         err = pop_eth(skb, key);
1341                         break;
1342 
1343                 case OVS_ACTION_ATTR_PUSH_NSH: {
1344                         u8 buffer[NSH_HDR_MAX_LEN];
1345                         struct nshhdr *nh = (struct nshhdr *)buffer;
1346 
1347                         err = nsh_hdr_from_nlattr(nla_data(a), nh,
1348                                                   NSH_HDR_MAX_LEN);
1349                         if (unlikely(err))
1350                                 break;
1351                         err = push_nsh(skb, key, nh);
1352                         break;
1353                 }
1354 
1355                 case OVS_ACTION_ATTR_POP_NSH:
1356                         err = pop_nsh(skb, key);
1357                         break;
1358 
1359                 case OVS_ACTION_ATTR_METER:
1360                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1361                                 consume_skb(skb);
1362                                 return 0;
1363                         }
1364                         break;
1365 
1366                 case OVS_ACTION_ATTR_CLONE: {
1367                         bool last = nla_is_last(a, rem);
1368 
1369                         err = clone(dp, skb, key, a, last);
1370                         if (last)
1371                                 return err;
1372 
1373                         break;
1374                 }
1375                 }
1376 
1377                 if (unlikely(err)) {
1378                         kfree_skb(skb);
1379                         return err;
1380                 }
1381         }
1382 
1383         consume_skb(skb);
1384         return 0;
1385 }
1386 
1387 /* Execute the actions on the clone of the packet. The effect of the
1388  * execution does not affect the original 'skb' nor the original 'key'.
1389  *
1390  * The execution may be deferred in case the actions can not be executed
1391  * immediately.
1392  */
1393 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1394                          struct sw_flow_key *key, u32 recirc_id,
1395                          const struct nlattr *actions, int len,
1396                          bool last, bool clone_flow_key)
1397 {
1398         struct deferred_action *da;
1399         struct sw_flow_key *clone;
1400 
1401         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1402         if (!skb) {
1403                 /* Out of memory, skip this action.
1404                  */
1405                 return 0;
1406         }
1407 
1408         /* When clone_flow_key is false, the 'key' will not be change
1409          * by the actions, then the 'key' can be used directly.
1410          * Otherwise, try to clone key from the next recursion level of
1411          * 'flow_keys'. If clone is successful, execute the actions
1412          * without deferring.
1413          */
1414         clone = clone_flow_key ? clone_key(key) : key;
1415         if (clone) {
1416                 int err = 0;
1417 
1418                 if (actions) { /* Sample action */
1419                         if (clone_flow_key)
1420                                 __this_cpu_inc(exec_actions_level);
1421 
1422                         err = do_execute_actions(dp, skb, clone,
1423                                                  actions, len);
1424 
1425                         if (clone_flow_key)
1426                                 __this_cpu_dec(exec_actions_level);
1427                 } else { /* Recirc action */
1428                         clone->recirc_id = recirc_id;
1429                         ovs_dp_process_packet(skb, clone);
1430                 }
1431                 return err;
1432         }
1433 
1434         /* Out of 'flow_keys' space. Defer actions */
1435         da = add_deferred_actions(skb, key, actions, len);
1436         if (da) {
1437                 if (!actions) { /* Recirc action */
1438                         key = &da->pkt_key;
1439                         key->recirc_id = recirc_id;
1440                 }
1441         } else {
1442                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1443                  * log an error.
1444                  */
1445                 kfree_skb(skb);
1446 
1447                 if (net_ratelimit()) {
1448                         if (actions) { /* Sample action */
1449                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1450                                         ovs_dp_name(dp));
1451                         } else {  /* Recirc action */
1452                                 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1453                                         ovs_dp_name(dp));
1454                         }
1455                 }
1456         }
1457         return 0;
1458 }
1459 
1460 static void process_deferred_actions(struct datapath *dp)
1461 {
1462         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1463 
1464         /* Do not touch the FIFO in case there is no deferred actions. */
1465         if (action_fifo_is_empty(fifo))
1466                 return;
1467 
1468         /* Finishing executing all deferred actions. */
1469         do {
1470                 struct deferred_action *da = action_fifo_get(fifo);
1471                 struct sk_buff *skb = da->skb;
1472                 struct sw_flow_key *key = &da->pkt_key;
1473                 const struct nlattr *actions = da->actions;
1474                 int actions_len = da->actions_len;
1475 
1476                 if (actions)
1477                         do_execute_actions(dp, skb, key, actions, actions_len);
1478                 else
1479                         ovs_dp_process_packet(skb, key);
1480         } while (!action_fifo_is_empty(fifo));
1481 
1482         /* Reset FIFO for the next packet.  */
1483         action_fifo_init(fifo);
1484 }
1485 
1486 /* Execute a list of actions against 'skb'. */
1487 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1488                         const struct sw_flow_actions *acts,
1489                         struct sw_flow_key *key)
1490 {
1491         int err, level;
1492 
1493         level = __this_cpu_inc_return(exec_actions_level);
1494         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1495                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1496                                      ovs_dp_name(dp));
1497                 kfree_skb(skb);
1498                 err = -ENETDOWN;
1499                 goto out;
1500         }
1501 
1502         OVS_CB(skb)->acts_origlen = acts->orig_len;
1503         err = do_execute_actions(dp, skb, key,
1504                                  acts->actions, acts->actions_len);
1505 
1506         if (level == 1)
1507                 process_deferred_actions(dp);
1508 
1509 out:
1510         __this_cpu_dec(exec_actions_level);
1511         return err;
1512 }
1513 
1514 int action_fifos_init(void)
1515 {
1516         action_fifos = alloc_percpu(struct action_fifo);
1517         if (!action_fifos)
1518                 return -ENOMEM;
1519 
1520         flow_keys = alloc_percpu(struct action_flow_keys);
1521         if (!flow_keys) {
1522                 free_percpu(action_fifos);
1523                 return -ENOMEM;
1524         }
1525 
1526         return 0;
1527 }
1528 
1529 void action_fifos_exit(void)
1530 {
1531         free_percpu(action_fifos);
1532         free_percpu(flow_keys);
1533 }
1534 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp