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

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

Version: ~ [ linux-5.16-rc1 ] ~ [ linux-5.15.2 ] ~ [ linux-5.14.18 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.79 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.159 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.217 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.255 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.290 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.292 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * Copyright (c) 2007-2017 Nicira, Inc.
  4  */
  5 
  6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  7 
  8 #include <linux/skbuff.h>
  9 #include <linux/in.h>
 10 #include <linux/ip.h>
 11 #include <linux/openvswitch.h>
 12 #include <linux/sctp.h>
 13 #include <linux/tcp.h>
 14 #include <linux/udp.h>
 15 #include <linux/in6.h>
 16 #include <linux/if_arp.h>
 17 #include <linux/if_vlan.h>
 18 
 19 #include <net/dst.h>
 20 #include <net/ip.h>
 21 #include <net/ipv6.h>
 22 #include <net/ip6_fib.h>
 23 #include <net/checksum.h>
 24 #include <net/dsfield.h>
 25 #include <net/mpls.h>
 26 #include <net/sctp/checksum.h>
 27 
 28 #include "datapath.h"
 29 #include "flow.h"
 30 #include "conntrack.h"
 31 #include "vport.h"
 32 #include "flow_netlink.h"
 33 #include "openvswitch_trace.h"
 34 
 35 struct deferred_action {
 36         struct sk_buff *skb;
 37         const struct nlattr *actions;
 38         int actions_len;
 39 
 40         /* Store pkt_key clone when creating deferred action. */
 41         struct sw_flow_key pkt_key;
 42 };
 43 
 44 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
 45 struct ovs_frag_data {
 46         unsigned long dst;
 47         struct vport *vport;
 48         struct ovs_skb_cb cb;
 49         __be16 inner_protocol;
 50         u16 network_offset;     /* valid only for MPLS */
 51         u16 vlan_tci;
 52         __be16 vlan_proto;
 53         unsigned int l2_len;
 54         u8 mac_proto;
 55         u8 l2_data[MAX_L2_LEN];
 56 };
 57 
 58 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
 59 
 60 #define DEFERRED_ACTION_FIFO_SIZE 10
 61 #define OVS_RECURSION_LIMIT 5
 62 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
 63 struct action_fifo {
 64         int head;
 65         int tail;
 66         /* Deferred action fifo queue storage. */
 67         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
 68 };
 69 
 70 struct action_flow_keys {
 71         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
 72 };
 73 
 74 static struct action_fifo __percpu *action_fifos;
 75 static struct action_flow_keys __percpu *flow_keys;
 76 static DEFINE_PER_CPU(int, exec_actions_level);
 77 
 78 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
 79  * space. Return NULL if out of key spaces.
 80  */
 81 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
 82 {
 83         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
 84         int level = this_cpu_read(exec_actions_level);
 85         struct sw_flow_key *key = NULL;
 86 
 87         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
 88                 key = &keys->key[level - 1];
 89                 *key = *key_;
 90         }
 91 
 92         return key;
 93 }
 94 
 95 static void action_fifo_init(struct action_fifo *fifo)
 96 {
 97         fifo->head = 0;
 98         fifo->tail = 0;
 99 }
100 
101 static bool action_fifo_is_empty(const struct action_fifo *fifo)
102 {
103         return (fifo->head == fifo->tail);
104 }
105 
106 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
107 {
108         if (action_fifo_is_empty(fifo))
109                 return NULL;
110 
111         return &fifo->fifo[fifo->tail++];
112 }
113 
114 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
115 {
116         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
117                 return NULL;
118 
119         return &fifo->fifo[fifo->head++];
120 }
121 
122 /* Return true if fifo is not full */
123 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
124                                     const struct sw_flow_key *key,
125                                     const struct nlattr *actions,
126                                     const int actions_len)
127 {
128         struct action_fifo *fifo;
129         struct deferred_action *da;
130 
131         fifo = this_cpu_ptr(action_fifos);
132         da = action_fifo_put(fifo);
133         if (da) {
134                 da->skb = skb;
135                 da->actions = actions;
136                 da->actions_len = actions_len;
137                 da->pkt_key = *key;
138         }
139 
140         return da;
141 }
142 
143 static void invalidate_flow_key(struct sw_flow_key *key)
144 {
145         key->mac_proto |= SW_FLOW_KEY_INVALID;
146 }
147 
148 static bool is_flow_key_valid(const struct sw_flow_key *key)
149 {
150         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
151 }
152 
153 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
154                          struct sw_flow_key *key,
155                          u32 recirc_id,
156                          const struct nlattr *actions, int len,
157                          bool last, bool clone_flow_key);
158 
159 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
160                               struct sw_flow_key *key,
161                               const struct nlattr *attr, int len);
162 
163 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
164                      __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
165 {
166         int err;
167 
168         err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
169         if (err)
170                 return err;
171 
172         if (!mac_len)
173                 key->mac_proto = MAC_PROTO_NONE;
174 
175         invalidate_flow_key(key);
176         return 0;
177 }
178 
179 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
180                     const __be16 ethertype)
181 {
182         int err;
183 
184         err = skb_mpls_pop(skb, ethertype, skb->mac_len,
185                            ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
186         if (err)
187                 return err;
188 
189         if (ethertype == htons(ETH_P_TEB))
190                 key->mac_proto = MAC_PROTO_ETHERNET;
191 
192         invalidate_flow_key(key);
193         return 0;
194 }
195 
196 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
197                     const __be32 *mpls_lse, const __be32 *mask)
198 {
199         struct mpls_shim_hdr *stack;
200         __be32 lse;
201         int err;
202 
203         if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
204                 return -ENOMEM;
205 
206         stack = mpls_hdr(skb);
207         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
208         err = skb_mpls_update_lse(skb, lse);
209         if (err)
210                 return err;
211 
212         flow_key->mpls.lse[0] = lse;
213         return 0;
214 }
215 
216 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
217 {
218         int err;
219 
220         err = skb_vlan_pop(skb);
221         if (skb_vlan_tag_present(skb)) {
222                 invalidate_flow_key(key);
223         } else {
224                 key->eth.vlan.tci = 0;
225                 key->eth.vlan.tpid = 0;
226         }
227         return err;
228 }
229 
230 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
231                      const struct ovs_action_push_vlan *vlan)
232 {
233         if (skb_vlan_tag_present(skb)) {
234                 invalidate_flow_key(key);
235         } else {
236                 key->eth.vlan.tci = vlan->vlan_tci;
237                 key->eth.vlan.tpid = vlan->vlan_tpid;
238         }
239         return skb_vlan_push(skb, vlan->vlan_tpid,
240                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
241 }
242 
243 /* 'src' is already properly masked. */
244 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
245 {
246         u16 *dst = (u16 *)dst_;
247         const u16 *src = (const u16 *)src_;
248         const u16 *mask = (const u16 *)mask_;
249 
250         OVS_SET_MASKED(dst[0], src[0], mask[0]);
251         OVS_SET_MASKED(dst[1], src[1], mask[1]);
252         OVS_SET_MASKED(dst[2], src[2], mask[2]);
253 }
254 
255 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
256                         const struct ovs_key_ethernet *key,
257                         const struct ovs_key_ethernet *mask)
258 {
259         int err;
260 
261         err = skb_ensure_writable(skb, ETH_HLEN);
262         if (unlikely(err))
263                 return err;
264 
265         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
266 
267         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
268                                mask->eth_src);
269         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
270                                mask->eth_dst);
271 
272         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
273 
274         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
275         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
276         return 0;
277 }
278 
279 /* pop_eth does not support VLAN packets as this action is never called
280  * for them.
281  */
282 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
283 {
284         int err;
285 
286         err = skb_eth_pop(skb);
287         if (err)
288                 return err;
289 
290         /* safe right before invalidate_flow_key */
291         key->mac_proto = MAC_PROTO_NONE;
292         invalidate_flow_key(key);
293         return 0;
294 }
295 
296 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
297                     const struct ovs_action_push_eth *ethh)
298 {
299         int err;
300 
301         err = skb_eth_push(skb, ethh->addresses.eth_dst,
302                            ethh->addresses.eth_src);
303         if (err)
304                 return err;
305 
306         /* safe right before invalidate_flow_key */
307         key->mac_proto = MAC_PROTO_ETHERNET;
308         invalidate_flow_key(key);
309         return 0;
310 }
311 
312 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
313                     const struct nshhdr *nh)
314 {
315         int err;
316 
317         err = nsh_push(skb, nh);
318         if (err)
319                 return err;
320 
321         /* safe right before invalidate_flow_key */
322         key->mac_proto = MAC_PROTO_NONE;
323         invalidate_flow_key(key);
324         return 0;
325 }
326 
327 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
328 {
329         int err;
330 
331         err = nsh_pop(skb);
332         if (err)
333                 return err;
334 
335         /* safe right before invalidate_flow_key */
336         if (skb->protocol == htons(ETH_P_TEB))
337                 key->mac_proto = MAC_PROTO_ETHERNET;
338         else
339                 key->mac_proto = MAC_PROTO_NONE;
340         invalidate_flow_key(key);
341         return 0;
342 }
343 
344 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
345                                   __be32 addr, __be32 new_addr)
346 {
347         int transport_len = skb->len - skb_transport_offset(skb);
348 
349         if (nh->frag_off & htons(IP_OFFSET))
350                 return;
351 
352         if (nh->protocol == IPPROTO_TCP) {
353                 if (likely(transport_len >= sizeof(struct tcphdr)))
354                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
355                                                  addr, new_addr, true);
356         } else if (nh->protocol == IPPROTO_UDP) {
357                 if (likely(transport_len >= sizeof(struct udphdr))) {
358                         struct udphdr *uh = udp_hdr(skb);
359 
360                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
361                                 inet_proto_csum_replace4(&uh->check, skb,
362                                                          addr, new_addr, true);
363                                 if (!uh->check)
364                                         uh->check = CSUM_MANGLED_0;
365                         }
366                 }
367         }
368 }
369 
370 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
371                         __be32 *addr, __be32 new_addr)
372 {
373         update_ip_l4_checksum(skb, nh, *addr, new_addr);
374         csum_replace4(&nh->check, *addr, new_addr);
375         skb_clear_hash(skb);
376         *addr = new_addr;
377 }
378 
379 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
380                                  __be32 addr[4], const __be32 new_addr[4])
381 {
382         int transport_len = skb->len - skb_transport_offset(skb);
383 
384         if (l4_proto == NEXTHDR_TCP) {
385                 if (likely(transport_len >= sizeof(struct tcphdr)))
386                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
387                                                   addr, new_addr, true);
388         } else if (l4_proto == NEXTHDR_UDP) {
389                 if (likely(transport_len >= sizeof(struct udphdr))) {
390                         struct udphdr *uh = udp_hdr(skb);
391 
392                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
393                                 inet_proto_csum_replace16(&uh->check, skb,
394                                                           addr, new_addr, true);
395                                 if (!uh->check)
396                                         uh->check = CSUM_MANGLED_0;
397                         }
398                 }
399         } else if (l4_proto == NEXTHDR_ICMP) {
400                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
401                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
402                                                   skb, addr, new_addr, true);
403         }
404 }
405 
406 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
407                            const __be32 mask[4], __be32 masked[4])
408 {
409         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
410         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
411         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
412         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
413 }
414 
415 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
416                           __be32 addr[4], const __be32 new_addr[4],
417                           bool recalculate_csum)
418 {
419         if (recalculate_csum)
420                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
421 
422         skb_clear_hash(skb);
423         memcpy(addr, new_addr, sizeof(__be32[4]));
424 }
425 
426 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
427 {
428         /* Bits 21-24 are always unmasked, so this retains their values. */
429         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
430         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
431         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
432 }
433 
434 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
435                        u8 mask)
436 {
437         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
438 
439         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
440         nh->ttl = new_ttl;
441 }
442 
443 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
444                     const struct ovs_key_ipv4 *key,
445                     const struct ovs_key_ipv4 *mask)
446 {
447         struct iphdr *nh;
448         __be32 new_addr;
449         int err;
450 
451         err = skb_ensure_writable(skb, skb_network_offset(skb) +
452                                   sizeof(struct iphdr));
453         if (unlikely(err))
454                 return err;
455 
456         nh = ip_hdr(skb);
457 
458         /* Setting an IP addresses is typically only a side effect of
459          * matching on them in the current userspace implementation, so it
460          * makes sense to check if the value actually changed.
461          */
462         if (mask->ipv4_src) {
463                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
464 
465                 if (unlikely(new_addr != nh->saddr)) {
466                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
467                         flow_key->ipv4.addr.src = new_addr;
468                 }
469         }
470         if (mask->ipv4_dst) {
471                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
472 
473                 if (unlikely(new_addr != nh->daddr)) {
474                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
475                         flow_key->ipv4.addr.dst = new_addr;
476                 }
477         }
478         if (mask->ipv4_tos) {
479                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
480                 flow_key->ip.tos = nh->tos;
481         }
482         if (mask->ipv4_ttl) {
483                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
484                 flow_key->ip.ttl = nh->ttl;
485         }
486 
487         return 0;
488 }
489 
490 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
491 {
492         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
493 }
494 
495 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
496                     const struct ovs_key_ipv6 *key,
497                     const struct ovs_key_ipv6 *mask)
498 {
499         struct ipv6hdr *nh;
500         int err;
501 
502         err = skb_ensure_writable(skb, skb_network_offset(skb) +
503                                   sizeof(struct ipv6hdr));
504         if (unlikely(err))
505                 return err;
506 
507         nh = ipv6_hdr(skb);
508 
509         /* Setting an IP addresses is typically only a side effect of
510          * matching on them in the current userspace implementation, so it
511          * makes sense to check if the value actually changed.
512          */
513         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
514                 __be32 *saddr = (__be32 *)&nh->saddr;
515                 __be32 masked[4];
516 
517                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
518 
519                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
520                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
521                                       true);
522                         memcpy(&flow_key->ipv6.addr.src, masked,
523                                sizeof(flow_key->ipv6.addr.src));
524                 }
525         }
526         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
527                 unsigned int offset = 0;
528                 int flags = IP6_FH_F_SKIP_RH;
529                 bool recalc_csum = true;
530                 __be32 *daddr = (__be32 *)&nh->daddr;
531                 __be32 masked[4];
532 
533                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
534 
535                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
536                         if (ipv6_ext_hdr(nh->nexthdr))
537                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
538                                                              NEXTHDR_ROUTING,
539                                                              NULL, &flags)
540                                                != NEXTHDR_ROUTING);
541 
542                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
543                                       recalc_csum);
544                         memcpy(&flow_key->ipv6.addr.dst, masked,
545                                sizeof(flow_key->ipv6.addr.dst));
546                 }
547         }
548         if (mask->ipv6_tclass) {
549                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
550                 flow_key->ip.tos = ipv6_get_dsfield(nh);
551         }
552         if (mask->ipv6_label) {
553                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
554                             ntohl(mask->ipv6_label));
555                 flow_key->ipv6.label =
556                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
557         }
558         if (mask->ipv6_hlimit) {
559                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
560                                mask->ipv6_hlimit);
561                 flow_key->ip.ttl = nh->hop_limit;
562         }
563         return 0;
564 }
565 
566 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
567                    const struct nlattr *a)
568 {
569         struct nshhdr *nh;
570         size_t length;
571         int err;
572         u8 flags;
573         u8 ttl;
574         int i;
575 
576         struct ovs_key_nsh key;
577         struct ovs_key_nsh mask;
578 
579         err = nsh_key_from_nlattr(a, &key, &mask);
580         if (err)
581                 return err;
582 
583         /* Make sure the NSH base header is there */
584         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
585                 return -ENOMEM;
586 
587         nh = nsh_hdr(skb);
588         length = nsh_hdr_len(nh);
589 
590         /* Make sure the whole NSH header is there */
591         err = skb_ensure_writable(skb, skb_network_offset(skb) +
592                                        length);
593         if (unlikely(err))
594                 return err;
595 
596         nh = nsh_hdr(skb);
597         skb_postpull_rcsum(skb, nh, length);
598         flags = nsh_get_flags(nh);
599         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
600         flow_key->nsh.base.flags = flags;
601         ttl = nsh_get_ttl(nh);
602         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
603         flow_key->nsh.base.ttl = ttl;
604         nsh_set_flags_and_ttl(nh, flags, ttl);
605         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
606                                   mask.base.path_hdr);
607         flow_key->nsh.base.path_hdr = nh->path_hdr;
608         switch (nh->mdtype) {
609         case NSH_M_TYPE1:
610                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
611                         nh->md1.context[i] =
612                             OVS_MASKED(nh->md1.context[i], key.context[i],
613                                        mask.context[i]);
614                 }
615                 memcpy(flow_key->nsh.context, nh->md1.context,
616                        sizeof(nh->md1.context));
617                 break;
618         case NSH_M_TYPE2:
619                 memset(flow_key->nsh.context, 0,
620                        sizeof(flow_key->nsh.context));
621                 break;
622         default:
623                 return -EINVAL;
624         }
625         skb_postpush_rcsum(skb, nh, length);
626         return 0;
627 }
628 
629 /* Must follow skb_ensure_writable() since that can move the skb data. */
630 static void set_tp_port(struct sk_buff *skb, __be16 *port,
631                         __be16 new_port, __sum16 *check)
632 {
633         inet_proto_csum_replace2(check, skb, *port, new_port, false);
634         *port = new_port;
635 }
636 
637 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
638                    const struct ovs_key_udp *key,
639                    const struct ovs_key_udp *mask)
640 {
641         struct udphdr *uh;
642         __be16 src, dst;
643         int err;
644 
645         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
646                                   sizeof(struct udphdr));
647         if (unlikely(err))
648                 return err;
649 
650         uh = udp_hdr(skb);
651         /* Either of the masks is non-zero, so do not bother checking them. */
652         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
653         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
654 
655         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
656                 if (likely(src != uh->source)) {
657                         set_tp_port(skb, &uh->source, src, &uh->check);
658                         flow_key->tp.src = src;
659                 }
660                 if (likely(dst != uh->dest)) {
661                         set_tp_port(skb, &uh->dest, dst, &uh->check);
662                         flow_key->tp.dst = dst;
663                 }
664 
665                 if (unlikely(!uh->check))
666                         uh->check = CSUM_MANGLED_0;
667         } else {
668                 uh->source = src;
669                 uh->dest = dst;
670                 flow_key->tp.src = src;
671                 flow_key->tp.dst = dst;
672         }
673 
674         skb_clear_hash(skb);
675 
676         return 0;
677 }
678 
679 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
680                    const struct ovs_key_tcp *key,
681                    const struct ovs_key_tcp *mask)
682 {
683         struct tcphdr *th;
684         __be16 src, dst;
685         int err;
686 
687         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
688                                   sizeof(struct tcphdr));
689         if (unlikely(err))
690                 return err;
691 
692         th = tcp_hdr(skb);
693         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
694         if (likely(src != th->source)) {
695                 set_tp_port(skb, &th->source, src, &th->check);
696                 flow_key->tp.src = src;
697         }
698         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
699         if (likely(dst != th->dest)) {
700                 set_tp_port(skb, &th->dest, dst, &th->check);
701                 flow_key->tp.dst = dst;
702         }
703         skb_clear_hash(skb);
704 
705         return 0;
706 }
707 
708 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
709                     const struct ovs_key_sctp *key,
710                     const struct ovs_key_sctp *mask)
711 {
712         unsigned int sctphoff = skb_transport_offset(skb);
713         struct sctphdr *sh;
714         __le32 old_correct_csum, new_csum, old_csum;
715         int err;
716 
717         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
718         if (unlikely(err))
719                 return err;
720 
721         sh = sctp_hdr(skb);
722         old_csum = sh->checksum;
723         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
724 
725         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
726         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
727 
728         new_csum = sctp_compute_cksum(skb, sctphoff);
729 
730         /* Carry any checksum errors through. */
731         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
732 
733         skb_clear_hash(skb);
734         flow_key->tp.src = sh->source;
735         flow_key->tp.dst = sh->dest;
736 
737         return 0;
738 }
739 
740 static int ovs_vport_output(struct net *net, struct sock *sk,
741                             struct sk_buff *skb)
742 {
743         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
744         struct vport *vport = data->vport;
745 
746         if (skb_cow_head(skb, data->l2_len) < 0) {
747                 kfree_skb(skb);
748                 return -ENOMEM;
749         }
750 
751         __skb_dst_copy(skb, data->dst);
752         *OVS_CB(skb) = data->cb;
753         skb->inner_protocol = data->inner_protocol;
754         if (data->vlan_tci & VLAN_CFI_MASK)
755                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
756         else
757                 __vlan_hwaccel_clear_tag(skb);
758 
759         /* Reconstruct the MAC header.  */
760         skb_push(skb, data->l2_len);
761         memcpy(skb->data, &data->l2_data, data->l2_len);
762         skb_postpush_rcsum(skb, skb->data, data->l2_len);
763         skb_reset_mac_header(skb);
764 
765         if (eth_p_mpls(skb->protocol)) {
766                 skb->inner_network_header = skb->network_header;
767                 skb_set_network_header(skb, data->network_offset);
768                 skb_reset_mac_len(skb);
769         }
770 
771         ovs_vport_send(vport, skb, data->mac_proto);
772         return 0;
773 }
774 
775 static unsigned int
776 ovs_dst_get_mtu(const struct dst_entry *dst)
777 {
778         return dst->dev->mtu;
779 }
780 
781 static struct dst_ops ovs_dst_ops = {
782         .family = AF_UNSPEC,
783         .mtu = ovs_dst_get_mtu,
784 };
785 
786 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
787  * ovs_vport_output(), which is called once per fragmented packet.
788  */
789 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
790                          u16 orig_network_offset, u8 mac_proto)
791 {
792         unsigned int hlen = skb_network_offset(skb);
793         struct ovs_frag_data *data;
794 
795         data = this_cpu_ptr(&ovs_frag_data_storage);
796         data->dst = skb->_skb_refdst;
797         data->vport = vport;
798         data->cb = *OVS_CB(skb);
799         data->inner_protocol = skb->inner_protocol;
800         data->network_offset = orig_network_offset;
801         if (skb_vlan_tag_present(skb))
802                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
803         else
804                 data->vlan_tci = 0;
805         data->vlan_proto = skb->vlan_proto;
806         data->mac_proto = mac_proto;
807         data->l2_len = hlen;
808         memcpy(&data->l2_data, skb->data, hlen);
809 
810         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
811         skb_pull(skb, hlen);
812 }
813 
814 static void ovs_fragment(struct net *net, struct vport *vport,
815                          struct sk_buff *skb, u16 mru,
816                          struct sw_flow_key *key)
817 {
818         u16 orig_network_offset = 0;
819 
820         if (eth_p_mpls(skb->protocol)) {
821                 orig_network_offset = skb_network_offset(skb);
822                 skb->network_header = skb->inner_network_header;
823         }
824 
825         if (skb_network_offset(skb) > MAX_L2_LEN) {
826                 OVS_NLERR(1, "L2 header too long to fragment");
827                 goto err;
828         }
829 
830         if (key->eth.type == htons(ETH_P_IP)) {
831                 struct rtable ovs_rt = { 0 };
832                 unsigned long orig_dst;
833 
834                 prepare_frag(vport, skb, orig_network_offset,
835                              ovs_key_mac_proto(key));
836                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
837                          DST_OBSOLETE_NONE, DST_NOCOUNT);
838                 ovs_rt.dst.dev = vport->dev;
839 
840                 orig_dst = skb->_skb_refdst;
841                 skb_dst_set_noref(skb, &ovs_rt.dst);
842                 IPCB(skb)->frag_max_size = mru;
843 
844                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
845                 refdst_drop(orig_dst);
846         } else if (key->eth.type == htons(ETH_P_IPV6)) {
847                 unsigned long orig_dst;
848                 struct rt6_info ovs_rt;
849 
850                 prepare_frag(vport, skb, orig_network_offset,
851                              ovs_key_mac_proto(key));
852                 memset(&ovs_rt, 0, sizeof(ovs_rt));
853                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
854                          DST_OBSOLETE_NONE, DST_NOCOUNT);
855                 ovs_rt.dst.dev = vport->dev;
856 
857                 orig_dst = skb->_skb_refdst;
858                 skb_dst_set_noref(skb, &ovs_rt.dst);
859                 IP6CB(skb)->frag_max_size = mru;
860 
861                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
862                 refdst_drop(orig_dst);
863         } else {
864                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
865                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
866                           vport->dev->mtu);
867                 goto err;
868         }
869 
870         return;
871 err:
872         kfree_skb(skb);
873 }
874 
875 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
876                       struct sw_flow_key *key)
877 {
878         struct vport *vport = ovs_vport_rcu(dp, out_port);
879 
880         if (likely(vport)) {
881                 u16 mru = OVS_CB(skb)->mru;
882                 u32 cutlen = OVS_CB(skb)->cutlen;
883 
884                 if (unlikely(cutlen > 0)) {
885                         if (skb->len - cutlen > ovs_mac_header_len(key))
886                                 pskb_trim(skb, skb->len - cutlen);
887                         else
888                                 pskb_trim(skb, ovs_mac_header_len(key));
889                 }
890 
891                 if (likely(!mru ||
892                            (skb->len <= mru + vport->dev->hard_header_len))) {
893                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
894                 } else if (mru <= vport->dev->mtu) {
895                         struct net *net = read_pnet(&dp->net);
896 
897                         ovs_fragment(net, vport, skb, mru, key);
898                 } else {
899                         kfree_skb(skb);
900                 }
901         } else {
902                 kfree_skb(skb);
903         }
904 }
905 
906 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
907                             struct sw_flow_key *key, const struct nlattr *attr,
908                             const struct nlattr *actions, int actions_len,
909                             uint32_t cutlen)
910 {
911         struct dp_upcall_info upcall;
912         const struct nlattr *a;
913         int rem;
914 
915         memset(&upcall, 0, sizeof(upcall));
916         upcall.cmd = OVS_PACKET_CMD_ACTION;
917         upcall.mru = OVS_CB(skb)->mru;
918 
919         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
920              a = nla_next(a, &rem)) {
921                 switch (nla_type(a)) {
922                 case OVS_USERSPACE_ATTR_USERDATA:
923                         upcall.userdata = a;
924                         break;
925 
926                 case OVS_USERSPACE_ATTR_PID:
927                         if (dp->user_features &
928                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
929                                 upcall.portid =
930                                   ovs_dp_get_upcall_portid(dp,
931                                                            smp_processor_id());
932                         else
933                                 upcall.portid = nla_get_u32(a);
934                         break;
935 
936                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
937                         /* Get out tunnel info. */
938                         struct vport *vport;
939 
940                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
941                         if (vport) {
942                                 int err;
943 
944                                 err = dev_fill_metadata_dst(vport->dev, skb);
945                                 if (!err)
946                                         upcall.egress_tun_info = skb_tunnel_info(skb);
947                         }
948 
949                         break;
950                 }
951 
952                 case OVS_USERSPACE_ATTR_ACTIONS: {
953                         /* Include actions. */
954                         upcall.actions = actions;
955                         upcall.actions_len = actions_len;
956                         break;
957                 }
958 
959                 } /* End of switch. */
960         }
961 
962         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
963 }
964 
965 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
966                                      struct sw_flow_key *key,
967                                      const struct nlattr *attr)
968 {
969         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
970         struct nlattr *actions = nla_data(attr);
971 
972         if (nla_len(actions))
973                 return clone_execute(dp, skb, key, 0, nla_data(actions),
974                                      nla_len(actions), true, false);
975 
976         consume_skb(skb);
977         return 0;
978 }
979 
980 /* When 'last' is true, sample() should always consume the 'skb'.
981  * Otherwise, sample() should keep 'skb' intact regardless what
982  * actions are executed within sample().
983  */
984 static int sample(struct datapath *dp, struct sk_buff *skb,
985                   struct sw_flow_key *key, const struct nlattr *attr,
986                   bool last)
987 {
988         struct nlattr *actions;
989         struct nlattr *sample_arg;
990         int rem = nla_len(attr);
991         const struct sample_arg *arg;
992         bool clone_flow_key;
993 
994         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
995         sample_arg = nla_data(attr);
996         arg = nla_data(sample_arg);
997         actions = nla_next(sample_arg, &rem);
998 
999         if ((arg->probability != U32_MAX) &&
1000             (!arg->probability || prandom_u32() > arg->probability)) {
1001                 if (last)
1002                         consume_skb(skb);
1003                 return 0;
1004         }
1005 
1006         clone_flow_key = !arg->exec;
1007         return clone_execute(dp, skb, key, 0, actions, rem, last,
1008                              clone_flow_key);
1009 }
1010 
1011 /* When 'last' is true, clone() should always consume the 'skb'.
1012  * Otherwise, clone() should keep 'skb' intact regardless what
1013  * actions are executed within clone().
1014  */
1015 static int clone(struct datapath *dp, struct sk_buff *skb,
1016                  struct sw_flow_key *key, const struct nlattr *attr,
1017                  bool last)
1018 {
1019         struct nlattr *actions;
1020         struct nlattr *clone_arg;
1021         int rem = nla_len(attr);
1022         bool dont_clone_flow_key;
1023 
1024         /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1025         clone_arg = nla_data(attr);
1026         dont_clone_flow_key = nla_get_u32(clone_arg);
1027         actions = nla_next(clone_arg, &rem);
1028 
1029         return clone_execute(dp, skb, key, 0, actions, rem, last,
1030                              !dont_clone_flow_key);
1031 }
1032 
1033 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1034                          const struct nlattr *attr)
1035 {
1036         struct ovs_action_hash *hash_act = nla_data(attr);
1037         u32 hash = 0;
1038 
1039         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1040         hash = skb_get_hash(skb);
1041         hash = jhash_1word(hash, hash_act->hash_basis);
1042         if (!hash)
1043                 hash = 0x1;
1044 
1045         key->ovs_flow_hash = hash;
1046 }
1047 
1048 static int execute_set_action(struct sk_buff *skb,
1049                               struct sw_flow_key *flow_key,
1050                               const struct nlattr *a)
1051 {
1052         /* Only tunnel set execution is supported without a mask. */
1053         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1054                 struct ovs_tunnel_info *tun = nla_data(a);
1055 
1056                 skb_dst_drop(skb);
1057                 dst_hold((struct dst_entry *)tun->tun_dst);
1058                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1059                 return 0;
1060         }
1061 
1062         return -EINVAL;
1063 }
1064 
1065 /* Mask is at the midpoint of the data. */
1066 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1067 
1068 static int execute_masked_set_action(struct sk_buff *skb,
1069                                      struct sw_flow_key *flow_key,
1070                                      const struct nlattr *a)
1071 {
1072         int err = 0;
1073 
1074         switch (nla_type(a)) {
1075         case OVS_KEY_ATTR_PRIORITY:
1076                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1077                                *get_mask(a, u32 *));
1078                 flow_key->phy.priority = skb->priority;
1079                 break;
1080 
1081         case OVS_KEY_ATTR_SKB_MARK:
1082                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1083                 flow_key->phy.skb_mark = skb->mark;
1084                 break;
1085 
1086         case OVS_KEY_ATTR_TUNNEL_INFO:
1087                 /* Masked data not supported for tunnel. */
1088                 err = -EINVAL;
1089                 break;
1090 
1091         case OVS_KEY_ATTR_ETHERNET:
1092                 err = set_eth_addr(skb, flow_key, nla_data(a),
1093                                    get_mask(a, struct ovs_key_ethernet *));
1094                 break;
1095 
1096         case OVS_KEY_ATTR_NSH:
1097                 err = set_nsh(skb, flow_key, a);
1098                 break;
1099 
1100         case OVS_KEY_ATTR_IPV4:
1101                 err = set_ipv4(skb, flow_key, nla_data(a),
1102                                get_mask(a, struct ovs_key_ipv4 *));
1103                 break;
1104 
1105         case OVS_KEY_ATTR_IPV6:
1106                 err = set_ipv6(skb, flow_key, nla_data(a),
1107                                get_mask(a, struct ovs_key_ipv6 *));
1108                 break;
1109 
1110         case OVS_KEY_ATTR_TCP:
1111                 err = set_tcp(skb, flow_key, nla_data(a),
1112                               get_mask(a, struct ovs_key_tcp *));
1113                 break;
1114 
1115         case OVS_KEY_ATTR_UDP:
1116                 err = set_udp(skb, flow_key, nla_data(a),
1117                               get_mask(a, struct ovs_key_udp *));
1118                 break;
1119 
1120         case OVS_KEY_ATTR_SCTP:
1121                 err = set_sctp(skb, flow_key, nla_data(a),
1122                                get_mask(a, struct ovs_key_sctp *));
1123                 break;
1124 
1125         case OVS_KEY_ATTR_MPLS:
1126                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1127                                                                     __be32 *));
1128                 break;
1129 
1130         case OVS_KEY_ATTR_CT_STATE:
1131         case OVS_KEY_ATTR_CT_ZONE:
1132         case OVS_KEY_ATTR_CT_MARK:
1133         case OVS_KEY_ATTR_CT_LABELS:
1134         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1135         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1136                 err = -EINVAL;
1137                 break;
1138         }
1139 
1140         return err;
1141 }
1142 
1143 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1144                           struct sw_flow_key *key,
1145                           const struct nlattr *a, bool last)
1146 {
1147         u32 recirc_id;
1148 
1149         if (!is_flow_key_valid(key)) {
1150                 int err;
1151 
1152                 err = ovs_flow_key_update(skb, key);
1153                 if (err)
1154                         return err;
1155         }
1156         BUG_ON(!is_flow_key_valid(key));
1157 
1158         recirc_id = nla_get_u32(a);
1159         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1160 }
1161 
1162 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1163                                  struct sw_flow_key *key,
1164                                  const struct nlattr *attr, bool last)
1165 {
1166         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1167         const struct nlattr *actions, *cpl_arg;
1168         int len, max_len, rem = nla_len(attr);
1169         const struct check_pkt_len_arg *arg;
1170         bool clone_flow_key;
1171 
1172         /* The first netlink attribute in 'attr' is always
1173          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1174          */
1175         cpl_arg = nla_data(attr);
1176         arg = nla_data(cpl_arg);
1177 
1178         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1179         max_len = arg->pkt_len;
1180 
1181         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1182             len <= max_len) {
1183                 /* Second netlink attribute in 'attr' is always
1184                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1185                  */
1186                 actions = nla_next(cpl_arg, &rem);
1187                 clone_flow_key = !arg->exec_for_lesser_equal;
1188         } else {
1189                 /* Third netlink attribute in 'attr' is always
1190                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1191                  */
1192                 actions = nla_next(cpl_arg, &rem);
1193                 actions = nla_next(actions, &rem);
1194                 clone_flow_key = !arg->exec_for_greater;
1195         }
1196 
1197         return clone_execute(dp, skb, key, 0, nla_data(actions),
1198                              nla_len(actions), last, clone_flow_key);
1199 }
1200 
1201 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1202 {
1203         int err;
1204 
1205         if (skb->protocol == htons(ETH_P_IPV6)) {
1206                 struct ipv6hdr *nh;
1207 
1208                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1209                                           sizeof(*nh));
1210                 if (unlikely(err))
1211                         return err;
1212 
1213                 nh = ipv6_hdr(skb);
1214 
1215                 if (nh->hop_limit <= 1)
1216                         return -EHOSTUNREACH;
1217 
1218                 key->ip.ttl = --nh->hop_limit;
1219         } else if (skb->protocol == htons(ETH_P_IP)) {
1220                 struct iphdr *nh;
1221                 u8 old_ttl;
1222 
1223                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1224                                           sizeof(*nh));
1225                 if (unlikely(err))
1226                         return err;
1227 
1228                 nh = ip_hdr(skb);
1229                 if (nh->ttl <= 1)
1230                         return -EHOSTUNREACH;
1231 
1232                 old_ttl = nh->ttl--;
1233                 csum_replace2(&nh->check, htons(old_ttl << 8),
1234                               htons(nh->ttl << 8));
1235                 key->ip.ttl = nh->ttl;
1236         }
1237         return 0;
1238 }
1239 
1240 /* Execute a list of actions against 'skb'. */
1241 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1242                               struct sw_flow_key *key,
1243                               const struct nlattr *attr, int len)
1244 {
1245         const struct nlattr *a;
1246         int rem;
1247 
1248         for (a = attr, rem = len; rem > 0;
1249              a = nla_next(a, &rem)) {
1250                 int err = 0;
1251 
1252                 if (trace_ovs_do_execute_action_enabled())
1253                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1254 
1255                 switch (nla_type(a)) {
1256                 case OVS_ACTION_ATTR_OUTPUT: {
1257                         int port = nla_get_u32(a);
1258                         struct sk_buff *clone;
1259 
1260                         /* Every output action needs a separate clone
1261                          * of 'skb', In case the output action is the
1262                          * last action, cloning can be avoided.
1263                          */
1264                         if (nla_is_last(a, rem)) {
1265                                 do_output(dp, skb, port, key);
1266                                 /* 'skb' has been used for output.
1267                                  */
1268                                 return 0;
1269                         }
1270 
1271                         clone = skb_clone(skb, GFP_ATOMIC);
1272                         if (clone)
1273                                 do_output(dp, clone, port, key);
1274                         OVS_CB(skb)->cutlen = 0;
1275                         break;
1276                 }
1277 
1278                 case OVS_ACTION_ATTR_TRUNC: {
1279                         struct ovs_action_trunc *trunc = nla_data(a);
1280 
1281                         if (skb->len > trunc->max_len)
1282                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1283                         break;
1284                 }
1285 
1286                 case OVS_ACTION_ATTR_USERSPACE:
1287                         output_userspace(dp, skb, key, a, attr,
1288                                                      len, OVS_CB(skb)->cutlen);
1289                         OVS_CB(skb)->cutlen = 0;
1290                         break;
1291 
1292                 case OVS_ACTION_ATTR_HASH:
1293                         execute_hash(skb, key, a);
1294                         break;
1295 
1296                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1297                         struct ovs_action_push_mpls *mpls = nla_data(a);
1298 
1299                         err = push_mpls(skb, key, mpls->mpls_lse,
1300                                         mpls->mpls_ethertype, skb->mac_len);
1301                         break;
1302                 }
1303                 case OVS_ACTION_ATTR_ADD_MPLS: {
1304                         struct ovs_action_add_mpls *mpls = nla_data(a);
1305                         __u16 mac_len = 0;
1306 
1307                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1308                                 mac_len = skb->mac_len;
1309 
1310                         err = push_mpls(skb, key, mpls->mpls_lse,
1311                                         mpls->mpls_ethertype, mac_len);
1312                         break;
1313                 }
1314                 case OVS_ACTION_ATTR_POP_MPLS:
1315                         err = pop_mpls(skb, key, nla_get_be16(a));
1316                         break;
1317 
1318                 case OVS_ACTION_ATTR_PUSH_VLAN:
1319                         err = push_vlan(skb, key, nla_data(a));
1320                         break;
1321 
1322                 case OVS_ACTION_ATTR_POP_VLAN:
1323                         err = pop_vlan(skb, key);
1324                         break;
1325 
1326                 case OVS_ACTION_ATTR_RECIRC: {
1327                         bool last = nla_is_last(a, rem);
1328 
1329                         err = execute_recirc(dp, skb, key, a, last);
1330                         if (last) {
1331                                 /* If this is the last action, the skb has
1332                                  * been consumed or freed.
1333                                  * Return immediately.
1334                                  */
1335                                 return err;
1336                         }
1337                         break;
1338                 }
1339 
1340                 case OVS_ACTION_ATTR_SET:
1341                         err = execute_set_action(skb, key, nla_data(a));
1342                         break;
1343 
1344                 case OVS_ACTION_ATTR_SET_MASKED:
1345                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1346                         err = execute_masked_set_action(skb, key, nla_data(a));
1347                         break;
1348 
1349                 case OVS_ACTION_ATTR_SAMPLE: {
1350                         bool last = nla_is_last(a, rem);
1351 
1352                         err = sample(dp, skb, key, a, last);
1353                         if (last)
1354                                 return err;
1355 
1356                         break;
1357                 }
1358 
1359                 case OVS_ACTION_ATTR_CT:
1360                         if (!is_flow_key_valid(key)) {
1361                                 err = ovs_flow_key_update(skb, key);
1362                                 if (err)
1363                                         return err;
1364                         }
1365 
1366                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1367                                              nla_data(a));
1368 
1369                         /* Hide stolen IP fragments from user space. */
1370                         if (err)
1371                                 return err == -EINPROGRESS ? 0 : err;
1372                         break;
1373 
1374                 case OVS_ACTION_ATTR_CT_CLEAR:
1375                         err = ovs_ct_clear(skb, key);
1376                         break;
1377 
1378                 case OVS_ACTION_ATTR_PUSH_ETH:
1379                         err = push_eth(skb, key, nla_data(a));
1380                         break;
1381 
1382                 case OVS_ACTION_ATTR_POP_ETH:
1383                         err = pop_eth(skb, key);
1384                         break;
1385 
1386                 case OVS_ACTION_ATTR_PUSH_NSH: {
1387                         u8 buffer[NSH_HDR_MAX_LEN];
1388                         struct nshhdr *nh = (struct nshhdr *)buffer;
1389 
1390                         err = nsh_hdr_from_nlattr(nla_data(a), nh,
1391                                                   NSH_HDR_MAX_LEN);
1392                         if (unlikely(err))
1393                                 break;
1394                         err = push_nsh(skb, key, nh);
1395                         break;
1396                 }
1397 
1398                 case OVS_ACTION_ATTR_POP_NSH:
1399                         err = pop_nsh(skb, key);
1400                         break;
1401 
1402                 case OVS_ACTION_ATTR_METER:
1403                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1404                                 consume_skb(skb);
1405                                 return 0;
1406                         }
1407                         break;
1408 
1409                 case OVS_ACTION_ATTR_CLONE: {
1410                         bool last = nla_is_last(a, rem);
1411 
1412                         err = clone(dp, skb, key, a, last);
1413                         if (last)
1414                                 return err;
1415 
1416                         break;
1417                 }
1418 
1419                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1420                         bool last = nla_is_last(a, rem);
1421 
1422                         err = execute_check_pkt_len(dp, skb, key, a, last);
1423                         if (last)
1424                                 return err;
1425 
1426                         break;
1427                 }
1428 
1429                 case OVS_ACTION_ATTR_DEC_TTL:
1430                         err = execute_dec_ttl(skb, key);
1431                         if (err == -EHOSTUNREACH)
1432                                 return dec_ttl_exception_handler(dp, skb,
1433                                                                  key, a);
1434                         break;
1435                 }
1436 
1437                 if (unlikely(err)) {
1438                         kfree_skb(skb);
1439                         return err;
1440                 }
1441         }
1442 
1443         consume_skb(skb);
1444         return 0;
1445 }
1446 
1447 /* Execute the actions on the clone of the packet. The effect of the
1448  * execution does not affect the original 'skb' nor the original 'key'.
1449  *
1450  * The execution may be deferred in case the actions can not be executed
1451  * immediately.
1452  */
1453 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1454                          struct sw_flow_key *key, u32 recirc_id,
1455                          const struct nlattr *actions, int len,
1456                          bool last, bool clone_flow_key)
1457 {
1458         struct deferred_action *da;
1459         struct sw_flow_key *clone;
1460 
1461         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1462         if (!skb) {
1463                 /* Out of memory, skip this action.
1464                  */
1465                 return 0;
1466         }
1467 
1468         /* When clone_flow_key is false, the 'key' will not be change
1469          * by the actions, then the 'key' can be used directly.
1470          * Otherwise, try to clone key from the next recursion level of
1471          * 'flow_keys'. If clone is successful, execute the actions
1472          * without deferring.
1473          */
1474         clone = clone_flow_key ? clone_key(key) : key;
1475         if (clone) {
1476                 int err = 0;
1477 
1478                 if (actions) { /* Sample action */
1479                         if (clone_flow_key)
1480                                 __this_cpu_inc(exec_actions_level);
1481 
1482                         err = do_execute_actions(dp, skb, clone,
1483                                                  actions, len);
1484 
1485                         if (clone_flow_key)
1486                                 __this_cpu_dec(exec_actions_level);
1487                 } else { /* Recirc action */
1488                         clone->recirc_id = recirc_id;
1489                         ovs_dp_process_packet(skb, clone);
1490                 }
1491                 return err;
1492         }
1493 
1494         /* Out of 'flow_keys' space. Defer actions */
1495         da = add_deferred_actions(skb, key, actions, len);
1496         if (da) {
1497                 if (!actions) { /* Recirc action */
1498                         key = &da->pkt_key;
1499                         key->recirc_id = recirc_id;
1500                 }
1501         } else {
1502                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1503                  * log an error.
1504                  */
1505                 kfree_skb(skb);
1506 
1507                 if (net_ratelimit()) {
1508                         if (actions) { /* Sample action */
1509                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1510                                         ovs_dp_name(dp));
1511                         } else {  /* Recirc action */
1512                                 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1513                                         ovs_dp_name(dp));
1514                         }
1515                 }
1516         }
1517         return 0;
1518 }
1519 
1520 static void process_deferred_actions(struct datapath *dp)
1521 {
1522         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1523 
1524         /* Do not touch the FIFO in case there is no deferred actions. */
1525         if (action_fifo_is_empty(fifo))
1526                 return;
1527 
1528         /* Finishing executing all deferred actions. */
1529         do {
1530                 struct deferred_action *da = action_fifo_get(fifo);
1531                 struct sk_buff *skb = da->skb;
1532                 struct sw_flow_key *key = &da->pkt_key;
1533                 const struct nlattr *actions = da->actions;
1534                 int actions_len = da->actions_len;
1535 
1536                 if (actions)
1537                         do_execute_actions(dp, skb, key, actions, actions_len);
1538                 else
1539                         ovs_dp_process_packet(skb, key);
1540         } while (!action_fifo_is_empty(fifo));
1541 
1542         /* Reset FIFO for the next packet.  */
1543         action_fifo_init(fifo);
1544 }
1545 
1546 /* Execute a list of actions against 'skb'. */
1547 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1548                         const struct sw_flow_actions *acts,
1549                         struct sw_flow_key *key)
1550 {
1551         int err, level;
1552 
1553         level = __this_cpu_inc_return(exec_actions_level);
1554         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1555                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1556                                      ovs_dp_name(dp));
1557                 kfree_skb(skb);
1558                 err = -ENETDOWN;
1559                 goto out;
1560         }
1561 
1562         OVS_CB(skb)->acts_origlen = acts->orig_len;
1563         err = do_execute_actions(dp, skb, key,
1564                                  acts->actions, acts->actions_len);
1565 
1566         if (level == 1)
1567                 process_deferred_actions(dp);
1568 
1569 out:
1570         __this_cpu_dec(exec_actions_level);
1571         return err;
1572 }
1573 
1574 int action_fifos_init(void)
1575 {
1576         action_fifos = alloc_percpu(struct action_fifo);
1577         if (!action_fifos)
1578                 return -ENOMEM;
1579 
1580         flow_keys = alloc_percpu(struct action_flow_keys);
1581         if (!flow_keys) {
1582                 free_percpu(action_fifos);
1583                 return -ENOMEM;
1584         }
1585 
1586         return 0;
1587 }
1588 
1589 void action_fifos_exit(void)
1590 {
1591         free_percpu(action_fifos);
1592         free_percpu(flow_keys);
1593 }
1594 

~ [ 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