2 * Copyright (c) 2015, 2016 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "conntrack.h"
21 #include <sys/types.h>
22 #include <netinet/in.h>
23 #include <netinet/icmp6.h>
26 #include "conntrack-private.h"
30 #include "dp-packet.h"
33 #include "odp-netlink.h"
34 #include "openvswitch/hmap.h"
35 #include "openvswitch/vlog.h"
37 #include "ovs-thread.h"
38 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(conntrack
);
45 COVERAGE_DEFINE(conntrack_full
);
46 COVERAGE_DEFINE(conntrack_long_cleanup
);
48 struct conn_lookup_ctx
{
56 static bool conn_key_extract(struct conntrack
*, struct dp_packet
*,
57 ovs_be16 dl_type
, struct conn_lookup_ctx
*,
59 static uint32_t conn_key_hash(const struct conn_key
*, uint32_t basis
);
60 static void conn_key_reverse(struct conn_key
*);
61 static void conn_key_lookup(struct conntrack_bucket
*ctb
,
62 struct conn_lookup_ctx
*ctx
,
64 static bool valid_new(struct dp_packet
*pkt
, struct conn_key
*);
65 static struct conn
*new_conn(struct conntrack_bucket
*, struct dp_packet
*pkt
,
66 struct conn_key
*, long long now
);
67 static void delete_conn(struct conn
*);
68 static enum ct_update_res
conn_update(struct conn
*,
69 struct conntrack_bucket
*ctb
,
70 struct dp_packet
*, bool reply
,
72 static bool conn_expired(struct conn
*, long long now
);
73 static void set_mark(struct dp_packet
*, struct conn
*,
74 uint32_t val
, uint32_t mask
);
75 static void set_label(struct dp_packet
*, struct conn
*,
76 const struct ovs_key_ct_labels
*val
,
77 const struct ovs_key_ct_labels
*mask
);
78 static void *clean_thread_main(void *f_
);
80 static struct nat_conn_key_node
*
81 nat_conn_keys_lookup(struct hmap
*nat_conn_keys
,
82 const struct conn_key
*key
,
86 nat_conn_keys_remove(struct hmap
*nat_conn_keys
,
87 const struct conn_key
*key
,
91 nat_select_range_tuple(struct conntrack
*ct
, const struct conn
*conn
,
92 struct conn
*nat_conn
);
95 reverse_icmp_type(uint8_t type
);
97 reverse_icmp6_type(uint8_t type
);
99 extract_l3_ipv4(struct conn_key
*key
, const void *data
, size_t size
,
100 const char **new_data
, bool validate_checksum
);
102 extract_l3_ipv6(struct conn_key
*key
, const void *data
, size_t size
,
103 const char **new_data
);
105 static struct ct_l4_proto
*l4_protos
[] = {
106 [IPPROTO_TCP
] = &ct_proto_tcp
,
107 [IPPROTO_UDP
] = &ct_proto_other
,
108 [IPPROTO_ICMP
] = &ct_proto_icmp4
,
109 [IPPROTO_ICMPV6
] = &ct_proto_icmp6
,
112 long long ct_timeout_val
[] = {
113 #define CT_TIMEOUT(NAME, VAL) [CT_TM_##NAME] = VAL,
118 /* If the total number of connections goes above this value, no new connections
119 * are accepted; this is for CT_CONN_TYPE_DEFAULT connections. */
120 #define DEFAULT_N_CONN_LIMIT 3000000
122 /* Initializes the connection tracker 'ct'. The caller is responsible for
123 * calling 'conntrack_destroy()', when the instance is not needed anymore */
125 conntrack_init(struct conntrack
*ct
)
128 long long now
= time_msec();
130 ct_rwlock_init(&ct
->nat_resources_lock
);
131 ct_rwlock_wrlock(&ct
->nat_resources_lock
);
132 hmap_init(&ct
->nat_conn_keys
);
133 ct_rwlock_unlock(&ct
->nat_resources_lock
);
135 for (i
= 0; i
< CONNTRACK_BUCKETS
; i
++) {
136 struct conntrack_bucket
*ctb
= &ct
->buckets
[i
];
138 ct_lock_init(&ctb
->lock
);
139 ct_lock_lock(&ctb
->lock
);
140 hmap_init(&ctb
->connections
);
141 for (j
= 0; j
< ARRAY_SIZE(ctb
->exp_lists
); j
++) {
142 ovs_list_init(&ctb
->exp_lists
[j
]);
144 ct_lock_unlock(&ctb
->lock
);
145 ovs_mutex_init(&ctb
->cleanup_mutex
);
146 ovs_mutex_lock(&ctb
->cleanup_mutex
);
147 ctb
->next_cleanup
= now
+ CT_TM_MIN
;
148 ovs_mutex_unlock(&ctb
->cleanup_mutex
);
150 ct
->hash_basis
= random_uint32();
151 atomic_count_init(&ct
->n_conn
, 0);
152 atomic_init(&ct
->n_conn_limit
, DEFAULT_N_CONN_LIMIT
);
153 latch_init(&ct
->clean_thread_exit
);
154 ct
->clean_thread
= ovs_thread_create("ct_clean", clean_thread_main
, ct
);
157 /* Destroys the connection tracker 'ct' and frees all the allocated memory. */
159 conntrack_destroy(struct conntrack
*ct
)
163 latch_set(&ct
->clean_thread_exit
);
164 pthread_join(ct
->clean_thread
, NULL
);
165 latch_destroy(&ct
->clean_thread_exit
);
166 for (i
= 0; i
< CONNTRACK_BUCKETS
; i
++) {
167 struct conntrack_bucket
*ctb
= &ct
->buckets
[i
];
170 ovs_mutex_destroy(&ctb
->cleanup_mutex
);
171 ct_lock_lock(&ctb
->lock
);
172 HMAP_FOR_EACH_POP(conn
, node
, &ctb
->connections
) {
173 if (conn
->conn_type
== CT_CONN_TYPE_DEFAULT
) {
174 atomic_count_dec(&ct
->n_conn
);
178 hmap_destroy(&ctb
->connections
);
179 ct_lock_unlock(&ctb
->lock
);
180 ct_lock_destroy(&ctb
->lock
);
182 ct_rwlock_wrlock(&ct
->nat_resources_lock
);
183 struct nat_conn_key_node
*nat_conn_key_node
;
184 HMAP_FOR_EACH_POP (nat_conn_key_node
, node
, &ct
->nat_conn_keys
) {
185 free(nat_conn_key_node
);
187 hmap_destroy(&ct
->nat_conn_keys
);
188 ct_rwlock_unlock(&ct
->nat_resources_lock
);
189 ct_rwlock_destroy(&ct
->nat_resources_lock
);
192 static unsigned hash_to_bucket(uint32_t hash
)
194 /* Extracts the most significant bits in hash. The least significant bits
195 * are already used internally by the hmap implementation. */
196 BUILD_ASSERT(CONNTRACK_BUCKETS_SHIFT
< 32 && CONNTRACK_BUCKETS_SHIFT
>= 1);
198 return (hash
>> (32 - CONNTRACK_BUCKETS_SHIFT
)) % CONNTRACK_BUCKETS
;
202 write_ct_md(struct dp_packet
*pkt
, uint16_t zone
, const struct conn
*conn
,
203 const struct conn_key
*key
)
205 pkt
->md
.ct_state
|= CS_TRACKED
;
206 pkt
->md
.ct_zone
= zone
;
207 pkt
->md
.ct_mark
= conn
? conn
->mark
: 0;
208 pkt
->md
.ct_label
= conn
? conn
->label
: OVS_U128_ZERO
;
210 /* Use the original direction tuple if we have it. */
214 pkt
->md
.ct_orig_tuple_ipv6
= false;
216 if (key
->dl_type
== htons(ETH_TYPE_IP
)) {
217 pkt
->md
.ct_orig_tuple
.ipv4
= (struct ovs_key_ct_tuple_ipv4
) {
218 key
->src
.addr
.ipv4_aligned
,
219 key
->dst
.addr
.ipv4_aligned
,
220 key
->nw_proto
!= IPPROTO_ICMP
221 ? key
->src
.port
: htons(key
->src
.icmp_type
),
222 key
->nw_proto
!= IPPROTO_ICMP
223 ? key
->dst
.port
: htons(key
->src
.icmp_code
),
227 pkt
->md
.ct_orig_tuple_ipv6
= true;
228 pkt
->md
.ct_orig_tuple
.ipv6
= (struct ovs_key_ct_tuple_ipv6
) {
229 key
->src
.addr
.ipv6_aligned
,
230 key
->dst
.addr
.ipv6_aligned
,
231 key
->nw_proto
!= IPPROTO_ICMPV6
232 ? key
->src
.port
: htons(key
->src
.icmp_type
),
233 key
->nw_proto
!= IPPROTO_ICMPV6
234 ? key
->dst
.port
: htons(key
->src
.icmp_code
),
239 memset(&pkt
->md
.ct_orig_tuple
, 0, sizeof pkt
->md
.ct_orig_tuple
);
245 pat_packet(struct dp_packet
*pkt
, const struct conn
*conn
)
247 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
248 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
249 struct tcp_header
*th
= dp_packet_l4(pkt
);
250 packet_set_tcp_port(pkt
, conn
->rev_key
.dst
.port
, th
->tcp_dst
);
251 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
252 struct udp_header
*uh
= dp_packet_l4(pkt
);
253 packet_set_udp_port(pkt
, conn
->rev_key
.dst
.port
, uh
->udp_dst
);
255 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
256 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
257 struct tcp_header
*th
= dp_packet_l4(pkt
);
258 packet_set_tcp_port(pkt
, th
->tcp_src
, conn
->rev_key
.src
.port
);
259 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
260 struct udp_header
*uh
= dp_packet_l4(pkt
);
261 packet_set_udp_port(pkt
, uh
->udp_src
, conn
->rev_key
.src
.port
);
267 nat_packet(struct dp_packet
*pkt
, const struct conn
*conn
, bool related
)
269 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
270 pkt
->md
.ct_state
|= CS_SRC_NAT
;
271 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
272 struct ip_header
*nh
= dp_packet_l3(pkt
);
273 packet_set_ipv4_addr(pkt
, &nh
->ip_src
,
274 conn
->rev_key
.dst
.addr
.ipv4_aligned
);
276 struct ovs_16aligned_ip6_hdr
*nh6
= dp_packet_l3(pkt
);
277 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
279 &conn
->rev_key
.dst
.addr
.ipv6_aligned
,
283 pat_packet(pkt
, conn
);
285 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
286 pkt
->md
.ct_state
|= CS_DST_NAT
;
287 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
288 struct ip_header
*nh
= dp_packet_l3(pkt
);
289 packet_set_ipv4_addr(pkt
, &nh
->ip_dst
,
290 conn
->rev_key
.src
.addr
.ipv4_aligned
);
292 struct ovs_16aligned_ip6_hdr
*nh6
= dp_packet_l3(pkt
);
293 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
295 &conn
->rev_key
.src
.addr
.ipv6_aligned
,
299 pat_packet(pkt
, conn
);
305 un_pat_packet(struct dp_packet
*pkt
, const struct conn
*conn
)
307 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
308 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
309 struct tcp_header
*th
= dp_packet_l4(pkt
);
310 packet_set_tcp_port(pkt
, th
->tcp_src
, conn
->key
.src
.port
);
311 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
312 struct udp_header
*uh
= dp_packet_l4(pkt
);
313 packet_set_udp_port(pkt
, uh
->udp_src
, conn
->key
.src
.port
);
315 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
316 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
317 struct tcp_header
*th
= dp_packet_l4(pkt
);
318 packet_set_tcp_port(pkt
, conn
->key
.dst
.port
, th
->tcp_dst
);
319 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
320 struct udp_header
*uh
= dp_packet_l4(pkt
);
321 packet_set_udp_port(pkt
, conn
->key
.dst
.port
, uh
->udp_dst
);
327 reverse_pat_packet(struct dp_packet
*pkt
, const struct conn
*conn
)
329 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
330 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
331 struct tcp_header
*th_in
= dp_packet_l4(pkt
);
332 packet_set_tcp_port(pkt
, conn
->key
.src
.port
,
334 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
335 struct udp_header
*uh_in
= dp_packet_l4(pkt
);
336 packet_set_udp_port(pkt
, conn
->key
.src
.port
,
339 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
340 if (conn
->key
.nw_proto
== IPPROTO_TCP
) {
341 struct tcp_header
*th_in
= dp_packet_l4(pkt
);
342 packet_set_tcp_port(pkt
, th_in
->tcp_src
,
344 } else if (conn
->key
.nw_proto
== IPPROTO_UDP
) {
345 struct udp_header
*uh_in
= dp_packet_l4(pkt
);
346 packet_set_udp_port(pkt
, uh_in
->udp_src
,
353 reverse_nat_packet(struct dp_packet
*pkt
, const struct conn
*conn
)
355 char *tail
= dp_packet_tail(pkt
);
356 char pad
= dp_packet_l2_pad_size(pkt
);
357 struct conn_key inner_key
;
358 const char *inner_l4
= NULL
;
359 uint16_t orig_l3_ofs
= pkt
->l3_ofs
;
360 uint16_t orig_l4_ofs
= pkt
->l4_ofs
;
362 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
363 struct ip_header
*nh
= dp_packet_l3(pkt
);
364 struct icmp_header
*icmp
= dp_packet_l4(pkt
);
365 struct ip_header
*inner_l3
= (struct ip_header
*) (icmp
+ 1);
366 extract_l3_ipv4(&inner_key
, inner_l3
, tail
- ((char *)inner_l3
)
367 -pad
, &inner_l4
, false);
369 pkt
->l3_ofs
+= (char *) inner_l3
- (char *) nh
;
370 pkt
->l4_ofs
+= inner_l4
- (char *) icmp
;
372 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
373 packet_set_ipv4_addr(pkt
, &inner_l3
->ip_src
,
374 conn
->key
.src
.addr
.ipv4_aligned
);
375 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
376 packet_set_ipv4_addr(pkt
, &inner_l3
->ip_dst
,
377 conn
->key
.dst
.addr
.ipv4_aligned
);
379 reverse_pat_packet(pkt
, conn
);
381 icmp
->icmp_csum
= csum(icmp
, tail
- (char *) icmp
- pad
);
383 struct ovs_16aligned_ip6_hdr
*nh6
= dp_packet_l3(pkt
);
384 struct icmp6_error_header
*icmp6
= dp_packet_l4(pkt
);
385 struct ovs_16aligned_ip6_hdr
*inner_l3_6
=
386 (struct ovs_16aligned_ip6_hdr
*) (icmp6
+ 1);
387 extract_l3_ipv6(&inner_key
, inner_l3_6
,
388 tail
- ((char *)inner_l3_6
) - pad
,
390 pkt
->l3_ofs
+= (char *) inner_l3_6
- (char *) nh6
;
391 pkt
->l4_ofs
+= inner_l4
- (char *) icmp6
;
393 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
394 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
395 inner_l3_6
->ip6_src
.be32
,
396 &conn
->key
.src
.addr
.ipv6_aligned
,
398 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
399 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
400 inner_l3_6
->ip6_dst
.be32
,
401 &conn
->key
.dst
.addr
.ipv6_aligned
,
404 reverse_pat_packet(pkt
, conn
);
405 uint32_t icmp6_csum
= packet_csum_pseudoheader6(nh6
);
406 icmp6
->icmp6_base
.icmp6_cksum
= 0;
407 icmp6
->icmp6_base
.icmp6_cksum
= csum_finish(
408 csum_continue(icmp6_csum
, icmp6
, tail
- (char *) icmp6
- pad
));
410 pkt
->l3_ofs
= orig_l3_ofs
;
411 pkt
->l4_ofs
= orig_l4_ofs
;
415 un_nat_packet(struct dp_packet
*pkt
, const struct conn
*conn
,
418 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
419 pkt
->md
.ct_state
|= CS_DST_NAT
;
420 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
421 struct ip_header
*nh
= dp_packet_l3(pkt
);
422 packet_set_ipv4_addr(pkt
, &nh
->ip_dst
,
423 conn
->key
.src
.addr
.ipv4_aligned
);
425 struct ovs_16aligned_ip6_hdr
*nh6
= dp_packet_l3(pkt
);
426 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
428 &conn
->key
.src
.addr
.ipv6_aligned
, true);
431 if (OVS_UNLIKELY(related
)) {
432 reverse_nat_packet(pkt
, conn
);
434 un_pat_packet(pkt
, conn
);
436 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_DST
) {
437 pkt
->md
.ct_state
|= CS_SRC_NAT
;
438 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
439 struct ip_header
*nh
= dp_packet_l3(pkt
);
440 packet_set_ipv4_addr(pkt
, &nh
->ip_src
,
441 conn
->key
.dst
.addr
.ipv4_aligned
);
443 struct ovs_16aligned_ip6_hdr
*nh6
= dp_packet_l3(pkt
);
444 packet_set_ipv6_addr(pkt
, conn
->key
.nw_proto
,
446 &conn
->key
.dst
.addr
.ipv6_aligned
, true);
449 if (OVS_UNLIKELY(related
)) {
450 reverse_nat_packet(pkt
, conn
);
452 un_pat_packet(pkt
, conn
);
457 /* Typical usage of this helper is in non per-packet code;
458 * this is because the bucket lock needs to be held for lookup
459 * and a hash would have already been needed. Hence, this function
460 * is just intended for code clarity. */
462 conn_lookup(struct conntrack
*ct
, struct conn_key
*key
, long long now
)
464 struct conn_lookup_ctx ctx
;
467 ctx
.hash
= conn_key_hash(key
, ct
->hash_basis
);
468 unsigned bucket
= hash_to_bucket(ctx
.hash
);
469 conn_key_lookup(&ct
->buckets
[bucket
], &ctx
, now
);
474 nat_clean(struct conntrack
*ct
, struct conn
*conn
,
475 struct conntrack_bucket
*ctb
)
476 OVS_REQUIRES(ctb
->lock
)
478 long long now
= time_msec();
479 ct_rwlock_wrlock(&ct
->nat_resources_lock
);
480 nat_conn_keys_remove(&ct
->nat_conn_keys
, &conn
->rev_key
, ct
->hash_basis
);
481 ct_rwlock_unlock(&ct
->nat_resources_lock
);
482 ct_lock_unlock(&ctb
->lock
);
484 uint32_t hash_rev_conn
= conn_key_hash(&conn
->rev_key
, ct
->hash_basis
);
485 unsigned bucket_rev_conn
= hash_to_bucket(hash_rev_conn
);
487 ct_lock_lock(&ct
->buckets
[bucket_rev_conn
].lock
);
488 ct_rwlock_wrlock(&ct
->nat_resources_lock
);
490 struct conn
*rev_conn
= conn_lookup(ct
, &conn
->rev_key
, now
);
492 struct nat_conn_key_node
*nat_conn_key_node
=
493 nat_conn_keys_lookup(&ct
->nat_conn_keys
, &conn
->rev_key
,
496 /* In the unlikely event, rev conn was recreated, then skip
497 * rev_conn cleanup. */
498 if (rev_conn
&& (!nat_conn_key_node
||
499 memcmp(&nat_conn_key_node
->value
, &rev_conn
->rev_key
,
500 sizeof nat_conn_key_node
->value
))) {
501 hmap_remove(&ct
->buckets
[bucket_rev_conn
].connections
,
507 ct_rwlock_unlock(&ct
->nat_resources_lock
);
508 ct_lock_unlock(&ct
->buckets
[bucket_rev_conn
].lock
);
509 ct_lock_lock(&ctb
->lock
);
513 conn_clean(struct conntrack
*ct
, struct conn
*conn
,
514 struct conntrack_bucket
*ctb
)
515 OVS_REQUIRES(ctb
->lock
)
517 ovs_list_remove(&conn
->exp_node
);
518 hmap_remove(&ctb
->connections
, &conn
->node
);
519 atomic_count_dec(&ct
->n_conn
);
520 if (conn
->nat_info
) {
521 nat_clean(ct
, conn
, ctb
);
528 conn_not_found(struct conntrack
*ct
, struct dp_packet
*pkt
,
529 struct conn_lookup_ctx
*ctx
, bool commit
, long long now
,
530 const struct nat_action_info_t
*nat_action_info
,
531 struct conn
*conn_for_un_nat_copy
)
533 unsigned bucket
= hash_to_bucket(ctx
->hash
);
534 struct conn
*nc
= NULL
;
536 if (!valid_new(pkt
, &ctx
->key
)) {
537 pkt
->md
.ct_state
= CS_INVALID
;
540 pkt
->md
.ct_state
= CS_NEW
;
543 unsigned int n_conn_limit
;
545 atomic_read_relaxed(&ct
->n_conn_limit
, &n_conn_limit
);
547 if (atomic_count_get(&ct
->n_conn
) >= n_conn_limit
) {
548 COVERAGE_INC(conntrack_full
);
552 nc
= new_conn(&ct
->buckets
[bucket
], pkt
, &ctx
->key
, now
);
554 nc
->rev_key
= nc
->key
;
555 conn_key_reverse(&nc
->rev_key
);
557 if (nat_action_info
) {
558 nc
->nat_info
= xmemdup(nat_action_info
, sizeof *nc
->nat_info
);
559 ct_rwlock_wrlock(&ct
->nat_resources_lock
);
561 bool nat_res
= nat_select_range_tuple(ct
, nc
,
562 conn_for_un_nat_copy
);
568 ct_rwlock_unlock(&ct
->nat_resources_lock
);
572 if (conn_for_un_nat_copy
&&
573 nc
->conn_type
== CT_CONN_TYPE_DEFAULT
) {
574 *nc
= *conn_for_un_nat_copy
;
575 conn_for_un_nat_copy
->conn_type
= CT_CONN_TYPE_UN_NAT
;
577 ct_rwlock_unlock(&ct
->nat_resources_lock
);
579 nat_packet(pkt
, nc
, ctx
->related
);
581 hmap_insert(&ct
->buckets
[bucket
].connections
, &nc
->node
, ctx
->hash
);
582 atomic_count_inc(&ct
->n_conn
);
588 conn_update_state(struct conntrack
*ct
, struct dp_packet
*pkt
,
589 struct conn_lookup_ctx
*ctx
, struct conn
**conn
,
590 long long now
, unsigned bucket
)
591 OVS_REQUIRES(ct
->buckets
[bucket
].lock
)
593 bool create_new_conn
= false;
596 pkt
->md
.ct_state
|= CS_RELATED
;
598 pkt
->md
.ct_state
|= CS_REPLY_DIR
;
601 enum ct_update_res res
= conn_update(*conn
, &ct
->buckets
[bucket
],
602 pkt
, ctx
->reply
, now
);
605 case CT_UPDATE_VALID
:
606 pkt
->md
.ct_state
|= CS_ESTABLISHED
;
607 pkt
->md
.ct_state
&= ~CS_NEW
;
609 pkt
->md
.ct_state
|= CS_REPLY_DIR
;
612 case CT_UPDATE_INVALID
:
613 pkt
->md
.ct_state
= CS_INVALID
;
616 conn_clean(ct
, *conn
, &ct
->buckets
[bucket
]);
617 create_new_conn
= true;
623 return create_new_conn
;
627 create_un_nat_conn(struct conntrack
*ct
, struct conn
*conn_for_un_nat_copy
,
630 struct conn
*nc
= xmemdup(conn_for_un_nat_copy
, sizeof *nc
);
631 nc
->key
= conn_for_un_nat_copy
->rev_key
;
632 nc
->rev_key
= conn_for_un_nat_copy
->key
;
633 uint32_t un_nat_hash
= conn_key_hash(&nc
->key
, ct
->hash_basis
);
634 unsigned un_nat_conn_bucket
= hash_to_bucket(un_nat_hash
);
635 ct_lock_lock(&ct
->buckets
[un_nat_conn_bucket
].lock
);
636 ct_rwlock_rdlock(&ct
->nat_resources_lock
);
638 struct conn
*rev_conn
= conn_lookup(ct
, &nc
->key
, now
);
640 struct nat_conn_key_node
*nat_conn_key_node
=
641 nat_conn_keys_lookup(&ct
->nat_conn_keys
, &nc
->key
, ct
->hash_basis
);
642 if (nat_conn_key_node
643 && !memcmp(&nat_conn_key_node
->value
, &nc
->rev_key
,
644 sizeof nat_conn_key_node
->value
)
646 hmap_insert(&ct
->buckets
[un_nat_conn_bucket
].connections
,
647 &nc
->node
, un_nat_hash
);
651 ct_rwlock_unlock(&ct
->nat_resources_lock
);
652 ct_lock_unlock(&ct
->buckets
[un_nat_conn_bucket
].lock
);
656 handle_nat(struct dp_packet
*pkt
, struct conn
*conn
,
657 uint16_t zone
, bool reply
, bool related
)
659 if (conn
->nat_info
&&
660 (!(pkt
->md
.ct_state
& (CS_SRC_NAT
| CS_DST_NAT
)) ||
661 (pkt
->md
.ct_state
& (CS_SRC_NAT
| CS_DST_NAT
) &&
662 zone
!= pkt
->md
.ct_zone
))) {
663 if (pkt
->md
.ct_state
& (CS_SRC_NAT
| CS_DST_NAT
)) {
664 pkt
->md
.ct_state
&= ~(CS_SRC_NAT
| CS_DST_NAT
);
667 un_nat_packet(pkt
, conn
, related
);
669 nat_packet(pkt
, conn
, related
);
675 process_one(struct conntrack
*ct
, struct dp_packet
*pkt
,
676 struct conn_lookup_ctx
*ctx
, uint16_t zone
,
677 bool force
, bool commit
, long long now
, const uint32_t *setmark
,
678 const struct ovs_key_ct_labels
*setlabel
,
679 const struct nat_action_info_t
*nat_action_info
)
682 unsigned bucket
= hash_to_bucket(ctx
->hash
);
683 ct_lock_lock(&ct
->buckets
[bucket
].lock
);
684 conn_key_lookup(&ct
->buckets
[bucket
], ctx
, now
);
687 /* Delete found entry if in wrong direction. 'force' implies commit. */
688 if (conn
&& force
&& ctx
->reply
) {
689 conn_clean(ct
, conn
, &ct
->buckets
[bucket
]);
693 if (OVS_LIKELY(conn
)) {
694 if (conn
->conn_type
== CT_CONN_TYPE_UN_NAT
) {
698 struct conn_lookup_ctx ctx2
;
700 ctx2
.key
= conn
->rev_key
;
701 ctx2
.hash
= conn_key_hash(&conn
->rev_key
, ct
->hash_basis
);
703 ct_lock_unlock(&ct
->buckets
[bucket
].lock
);
704 bucket
= hash_to_bucket(ctx2
.hash
);
706 ct_lock_lock(&ct
->buckets
[bucket
].lock
);
707 conn_key_lookup(&ct
->buckets
[bucket
], &ctx2
, now
);
712 /* It is a race condition where conn has timed out and removed
713 * between unlock of the rev_conn and lock of the forward conn;
715 pkt
->md
.ct_state
|= CS_TRACKED
| CS_INVALID
;
716 ct_lock_unlock(&ct
->buckets
[bucket
].lock
);
722 bool create_new_conn
= false;
723 struct conn conn_for_un_nat_copy
;
724 conn_for_un_nat_copy
.conn_type
= CT_CONN_TYPE_DEFAULT
;
725 if (OVS_LIKELY(conn
)) {
726 create_new_conn
= conn_update_state(ct
, pkt
, ctx
, &conn
, now
, bucket
);
727 if (nat_action_info
&& !create_new_conn
) {
728 handle_nat(pkt
, conn
, zone
, ctx
->reply
, ctx
->related
);
732 pkt
->md
.ct_state
= CS_INVALID
;
734 create_new_conn
= true;
738 if (OVS_UNLIKELY(create_new_conn
)) {
739 conn
= conn_not_found(ct
, pkt
, ctx
, commit
, now
, nat_action_info
,
740 &conn_for_un_nat_copy
);
743 write_ct_md(pkt
, zone
, conn
, &ctx
->key
);
744 if (conn
&& setmark
) {
745 set_mark(pkt
, conn
, setmark
[0], setmark
[1]);
748 if (conn
&& setlabel
) {
749 set_label(pkt
, conn
, &setlabel
[0], &setlabel
[1]);
752 ct_lock_unlock(&ct
->buckets
[bucket
].lock
);
754 if (conn_for_un_nat_copy
.conn_type
== CT_CONN_TYPE_UN_NAT
) {
755 create_un_nat_conn(ct
, &conn_for_un_nat_copy
, now
);
759 /* Sends the packets in '*pkt_batch' through the connection tracker 'ct'. All
760 * the packets should have the same 'dl_type' (IPv4 or IPv6) and should have
761 * the l3 and and l4 offset properly set.
763 * If 'commit' is true, the packets are allowed to create new entries in the
764 * connection tables. 'setmark', if not NULL, should point to a two
765 * elements array containing a value and a mask to set the connection mark.
766 * 'setlabel' behaves similarly for the connection label.*/
768 conntrack_execute(struct conntrack
*ct
, struct dp_packet_batch
*pkt_batch
,
769 ovs_be16 dl_type
, bool force
, bool commit
, uint16_t zone
,
770 const uint32_t *setmark
,
771 const struct ovs_key_ct_labels
*setlabel
,
773 const struct nat_action_info_t
*nat_action_info
)
775 struct dp_packet
**pkts
= pkt_batch
->packets
;
776 size_t cnt
= pkt_batch
->count
;
777 long long now
= time_msec();
778 struct conn_lookup_ctx ctx
;
781 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 5);
783 VLOG_WARN_RL(&rl
, "ALG helper \"%s\" not supported", helper
);
784 /* Continue without the helper */
787 for (size_t i
= 0; i
< cnt
; i
++) {
788 if (!conn_key_extract(ct
, pkts
[i
], dl_type
, &ctx
, zone
)) {
789 pkts
[i
]->md
.ct_state
= CS_INVALID
;
790 write_ct_md(pkts
[i
], zone
, NULL
, NULL
);
793 process_one(ct
, pkts
[i
], &ctx
, zone
, force
, commit
,
794 now
, setmark
, setlabel
, nat_action_info
);
801 set_mark(struct dp_packet
*pkt
, struct conn
*conn
, uint32_t val
, uint32_t mask
)
803 pkt
->md
.ct_mark
= val
| (pkt
->md
.ct_mark
& ~(mask
));
804 conn
->mark
= pkt
->md
.ct_mark
;
808 set_label(struct dp_packet
*pkt
, struct conn
*conn
,
809 const struct ovs_key_ct_labels
*val
,
810 const struct ovs_key_ct_labels
*mask
)
814 memcpy(&v
, val
, sizeof v
);
815 memcpy(&m
, mask
, sizeof m
);
817 pkt
->md
.ct_label
.u64
.lo
= v
.u64
.lo
818 | (pkt
->md
.ct_label
.u64
.lo
& ~(m
.u64
.lo
));
819 pkt
->md
.ct_label
.u64
.hi
= v
.u64
.hi
820 | (pkt
->md
.ct_label
.u64
.hi
& ~(m
.u64
.hi
));
821 conn
->label
= pkt
->md
.ct_label
;
825 /* Delete the expired connections from 'ctb', up to 'limit'. Returns the
826 * earliest expiration time among the remaining connections in 'ctb'. Returns
827 * LLONG_MAX if 'ctb' is empty. The return value might be smaller than 'now',
828 * if 'limit' is reached */
830 sweep_bucket(struct conntrack
*ct
, struct conntrack_bucket
*ctb
, long long now
,
832 OVS_REQUIRES(ctb
->lock
)
834 struct conn
*conn
, *next
;
835 long long min_expiration
= LLONG_MAX
;
839 for (i
= 0; i
< N_CT_TM
; i
++) {
840 LIST_FOR_EACH_SAFE (conn
, next
, exp_node
, &ctb
->exp_lists
[i
]) {
841 if (conn
->conn_type
== CT_CONN_TYPE_DEFAULT
) {
842 if (!conn_expired(conn
, now
) || count
>= limit
) {
843 min_expiration
= MIN(min_expiration
, conn
->expiration
);
844 if (count
>= limit
) {
845 /* Do not check other lists. */
846 COVERAGE_INC(conntrack_long_cleanup
);
847 return min_expiration
;
851 conn_clean(ct
, conn
, ctb
);
857 return min_expiration
;
860 /* Cleans up old connection entries from 'ct'. Returns the time when the
861 * next expiration might happen. The return value might be smaller than
862 * 'now', meaning that an internal limit has been reached, and some expired
863 * connections have not been deleted. */
865 conntrack_clean(struct conntrack
*ct
, long long now
)
867 long long next_wakeup
= now
+ CT_TM_MIN
;
868 unsigned int n_conn_limit
;
869 size_t clean_count
= 0;
872 atomic_read_relaxed(&ct
->n_conn_limit
, &n_conn_limit
);
874 for (i
= 0; i
< CONNTRACK_BUCKETS
; i
++) {
875 struct conntrack_bucket
*ctb
= &ct
->buckets
[i
];
879 ovs_mutex_lock(&ctb
->cleanup_mutex
);
880 if (ctb
->next_cleanup
> now
) {
884 ct_lock_lock(&ctb
->lock
);
885 prev_count
= hmap_count(&ctb
->connections
);
886 /* If the connections are well distributed among buckets, we want to
887 * limit to 10% of the global limit equally split among buckets. If
888 * the bucket is busier than the others, we limit to 10% of its
890 min_exp
= sweep_bucket(ct
, ctb
, now
,
891 MAX(prev_count
/10, n_conn_limit
/(CONNTRACK_BUCKETS
*10)));
892 clean_count
+= prev_count
- hmap_count(&ctb
->connections
);
895 /* We call hmap_shrink() only if sweep_bucket() managed to delete
896 * every expired connection. */
897 hmap_shrink(&ctb
->connections
);
900 ct_lock_unlock(&ctb
->lock
);
902 ctb
->next_cleanup
= MIN(min_exp
, now
+ CT_TM_MIN
);
905 next_wakeup
= MIN(next_wakeup
, ctb
->next_cleanup
);
906 ovs_mutex_unlock(&ctb
->cleanup_mutex
);
909 VLOG_DBG("conntrack cleanup %"PRIuSIZE
" entries in %lld msec",
910 clean_count
, time_msec() - now
);
917 * We must call conntrack_clean() periodically. conntrack_clean() return
918 * value gives an hint on when the next cleanup must be done (either because
919 * there is an actual connection that expires, or because a new connection
920 * might be created with the minimum timeout).
922 * The logic below has two goals:
924 * - We want to reduce the number of wakeups and batch connection cleanup
925 * when the load is not very high. CT_CLEAN_INTERVAL ensures that if we
926 * are coping with the current cleanup tasks, then we wait at least
927 * 5 seconds to do further cleanup.
929 * - We don't want to keep the buckets locked too long, as we might prevent
930 * traffic from flowing. CT_CLEAN_MIN_INTERVAL ensures that if cleanup is
931 * behind, there is at least some 200ms blocks of time when buckets will be
932 * left alone, so the datapath can operate unhindered.
934 #define CT_CLEAN_INTERVAL 5000 /* 5 seconds */
935 #define CT_CLEAN_MIN_INTERVAL 200 /* 0.2 seconds */
938 clean_thread_main(void *f_
)
940 struct conntrack
*ct
= f_
;
942 while (!latch_is_set(&ct
->clean_thread_exit
)) {
944 long long now
= time_msec();
946 next_wake
= conntrack_clean(ct
, now
);
948 if (next_wake
< now
) {
949 poll_timer_wait_until(now
+ CT_CLEAN_MIN_INTERVAL
);
951 poll_timer_wait_until(MAX(next_wake
, now
+ CT_CLEAN_INTERVAL
));
953 latch_wait(&ct
->clean_thread_exit
);
962 /* The function stores a pointer to the first byte after the header in
963 * '*new_data', if 'new_data' is not NULL. If it is NULL, the caller is
964 * not interested in the header's tail, meaning that the header has
965 * already been parsed (e.g. by flow_extract): we take this as a hint to
966 * save a few checks. If 'validate_checksum' is true, the function returns
967 * false if the IPv4 checksum is invalid. */
969 extract_l3_ipv4(struct conn_key
*key
, const void *data
, size_t size
,
970 const char **new_data
, bool validate_checksum
)
972 const struct ip_header
*ip
= data
;
976 if (OVS_UNLIKELY(size
< IP_HEADER_LEN
)) {
981 ip_len
= IP_IHL(ip
->ip_ihl_ver
) * 4;
984 if (OVS_UNLIKELY(ip_len
< IP_HEADER_LEN
)) {
987 if (OVS_UNLIKELY(size
< ip_len
)) {
991 *new_data
= (char *) data
+ ip_len
;
994 if (IP_IS_FRAGMENT(ip
->ip_frag_off
)) {
998 if (validate_checksum
&& csum(data
, ip_len
) != 0) {
1002 key
->src
.addr
.ipv4
= ip
->ip_src
;
1003 key
->dst
.addr
.ipv4
= ip
->ip_dst
;
1004 key
->nw_proto
= ip
->ip_proto
;
1009 /* The function stores a pointer to the first byte after the header in
1010 * '*new_data', if 'new_data' is not NULL. If it is NULL, the caller is
1011 * not interested in the header's tail, meaning that the header has
1012 * already been parsed (e.g. by flow_extract): we take this as a hint to
1013 * save a few checks. */
1015 extract_l3_ipv6(struct conn_key
*key
, const void *data
, size_t size
,
1016 const char **new_data
)
1018 const struct ovs_16aligned_ip6_hdr
*ip6
= data
;
1021 if (OVS_UNLIKELY(size
< sizeof *ip6
)) {
1026 uint8_t nw_proto
= ip6
->ip6_nxt
;
1027 uint8_t nw_frag
= 0;
1030 size
-= sizeof *ip6
;
1032 if (!parse_ipv6_ext_hdrs(&data
, &size
, &nw_proto
, &nw_frag
)) {
1044 key
->src
.addr
.ipv6
= ip6
->ip6_src
;
1045 key
->dst
.addr
.ipv6
= ip6
->ip6_dst
;
1046 key
->nw_proto
= nw_proto
;
1052 checksum_valid(const struct conn_key
*key
, const void *data
, size_t size
,
1057 if (key
->dl_type
== htons(ETH_TYPE_IP
)) {
1058 csum
= packet_csum_pseudoheader(l3
);
1059 } else if (key
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1060 csum
= packet_csum_pseudoheader6(l3
);
1065 csum
= csum_continue(csum
, data
, size
);
1067 return csum_finish(csum
) == 0;
1071 check_l4_tcp(const struct conn_key
*key
, const void *data
, size_t size
,
1074 const struct tcp_header
*tcp
= data
;
1075 if (size
< sizeof *tcp
) {
1079 size_t tcp_len
= TCP_OFFSET(tcp
->tcp_ctl
) * 4;
1080 if (OVS_UNLIKELY(tcp_len
< TCP_HEADER_LEN
|| tcp_len
> size
)) {
1084 return checksum_valid(key
, data
, size
, l3
);
1088 check_l4_udp(const struct conn_key
*key
, const void *data
, size_t size
,
1091 const struct udp_header
*udp
= data
;
1092 if (size
< sizeof *udp
) {
1096 size_t udp_len
= ntohs(udp
->udp_len
);
1097 if (OVS_UNLIKELY(udp_len
< UDP_HEADER_LEN
|| udp_len
> size
)) {
1101 /* Validation must be skipped if checksum is 0 on IPv4 packets */
1102 return (udp
->udp_csum
== 0 && key
->dl_type
== htons(ETH_TYPE_IP
))
1103 || checksum_valid(key
, data
, size
, l3
);
1107 check_l4_icmp(const void *data
, size_t size
)
1109 return csum(data
, size
) == 0;
1113 check_l4_icmp6(const struct conn_key
*key
, const void *data
, size_t size
,
1116 return checksum_valid(key
, data
, size
, l3
);
1120 extract_l4_tcp(struct conn_key
*key
, const void *data
, size_t size
)
1122 const struct tcp_header
*tcp
= data
;
1124 if (OVS_UNLIKELY(size
< TCP_HEADER_LEN
)) {
1128 key
->src
.port
= tcp
->tcp_src
;
1129 key
->dst
.port
= tcp
->tcp_dst
;
1131 /* Port 0 is invalid */
1132 return key
->src
.port
&& key
->dst
.port
;
1136 extract_l4_udp(struct conn_key
*key
, const void *data
, size_t size
)
1138 const struct udp_header
*udp
= data
;
1140 if (OVS_UNLIKELY(size
< UDP_HEADER_LEN
)) {
1144 key
->src
.port
= udp
->udp_src
;
1145 key
->dst
.port
= udp
->udp_dst
;
1147 /* Port 0 is invalid */
1148 return key
->src
.port
&& key
->dst
.port
;
1151 static inline bool extract_l4(struct conn_key
*key
, const void *data
,
1152 size_t size
, bool *related
, const void *l3
);
1155 reverse_icmp_type(uint8_t type
)
1158 case ICMP4_ECHO_REQUEST
:
1159 return ICMP4_ECHO_REPLY
;
1160 case ICMP4_ECHO_REPLY
:
1161 return ICMP4_ECHO_REQUEST
;
1163 case ICMP4_TIMESTAMP
:
1164 return ICMP4_TIMESTAMPREPLY
;
1165 case ICMP4_TIMESTAMPREPLY
:
1166 return ICMP4_TIMESTAMP
;
1168 case ICMP4_INFOREQUEST
:
1169 return ICMP4_INFOREPLY
;
1170 case ICMP4_INFOREPLY
:
1171 return ICMP4_INFOREQUEST
;
1177 /* If 'related' is not NULL and the function is processing an ICMP
1178 * error packet, extract the l3 and l4 fields from the nested header
1179 * instead and set *related to true. If 'related' is NULL we're
1180 * already processing a nested header and no such recursion is
1183 extract_l4_icmp(struct conn_key
*key
, const void *data
, size_t size
,
1186 const struct icmp_header
*icmp
= data
;
1188 if (OVS_UNLIKELY(size
< ICMP_HEADER_LEN
)) {
1192 switch (icmp
->icmp_type
) {
1193 case ICMP4_ECHO_REQUEST
:
1194 case ICMP4_ECHO_REPLY
:
1195 case ICMP4_TIMESTAMP
:
1196 case ICMP4_TIMESTAMPREPLY
:
1197 case ICMP4_INFOREQUEST
:
1198 case ICMP4_INFOREPLY
:
1199 if (icmp
->icmp_code
!= 0) {
1202 /* Separate ICMP connection: identified using id */
1203 key
->src
.icmp_id
= key
->dst
.icmp_id
= icmp
->icmp_fields
.echo
.id
;
1204 key
->src
.icmp_type
= icmp
->icmp_type
;
1205 key
->dst
.icmp_type
= reverse_icmp_type(icmp
->icmp_type
);
1207 case ICMP4_DST_UNREACH
:
1208 case ICMP4_TIME_EXCEEDED
:
1209 case ICMP4_PARAM_PROB
:
1210 case ICMP4_SOURCEQUENCH
:
1211 case ICMP4_REDIRECT
: {
1212 /* ICMP packet part of another connection. We should
1213 * extract the key from embedded packet header */
1214 struct conn_key inner_key
;
1215 const char *l3
= (const char *) (icmp
+ 1);
1216 const char *tail
= (const char *) data
+ size
;
1224 memset(&inner_key
, 0, sizeof inner_key
);
1225 inner_key
.dl_type
= htons(ETH_TYPE_IP
);
1226 ok
= extract_l3_ipv4(&inner_key
, l3
, tail
- l3
, &l4
, false);
1231 if (inner_key
.src
.addr
.ipv4_aligned
!= key
->dst
.addr
.ipv4_aligned
1232 || inner_key
.dst
.addr
.ipv4_aligned
!= key
->src
.addr
.ipv4_aligned
) {
1236 key
->src
= inner_key
.src
;
1237 key
->dst
= inner_key
.dst
;
1238 key
->nw_proto
= inner_key
.nw_proto
;
1240 ok
= extract_l4(key
, l4
, tail
- l4
, NULL
, l3
);
1242 conn_key_reverse(key
);
1255 reverse_icmp6_type(uint8_t type
)
1258 case ICMP6_ECHO_REQUEST
:
1259 return ICMP6_ECHO_REPLY
;
1260 case ICMP6_ECHO_REPLY
:
1261 return ICMP6_ECHO_REQUEST
;
1267 /* If 'related' is not NULL and the function is processing an ICMP
1268 * error packet, extract the l3 and l4 fields from the nested header
1269 * instead and set *related to true. If 'related' is NULL we're
1270 * already processing a nested header and no such recursion is
1273 extract_l4_icmp6(struct conn_key
*key
, const void *data
, size_t size
,
1276 const struct icmp6_header
*icmp6
= data
;
1278 /* All the messages that we support need at least 4 bytes after
1280 if (size
< sizeof *icmp6
+ 4) {
1284 switch (icmp6
->icmp6_type
) {
1285 case ICMP6_ECHO_REQUEST
:
1286 case ICMP6_ECHO_REPLY
:
1287 if (icmp6
->icmp6_code
!= 0) {
1290 /* Separate ICMP connection: identified using id */
1291 key
->src
.icmp_id
= key
->dst
.icmp_id
= *(ovs_be16
*) (icmp6
+ 1);
1292 key
->src
.icmp_type
= icmp6
->icmp6_type
;
1293 key
->dst
.icmp_type
= reverse_icmp6_type(icmp6
->icmp6_type
);
1295 case ICMP6_DST_UNREACH
:
1296 case ICMP6_PACKET_TOO_BIG
:
1297 case ICMP6_TIME_EXCEEDED
:
1298 case ICMP6_PARAM_PROB
: {
1299 /* ICMP packet part of another connection. We should
1300 * extract the key from embedded packet header */
1301 struct conn_key inner_key
;
1302 const char *l3
= (const char *) icmp6
+ 8;
1303 const char *tail
= (const char *) data
+ size
;
1304 const char *l4
= NULL
;
1311 memset(&inner_key
, 0, sizeof inner_key
);
1312 inner_key
.dl_type
= htons(ETH_TYPE_IPV6
);
1313 ok
= extract_l3_ipv6(&inner_key
, l3
, tail
- l3
, &l4
);
1318 /* pf doesn't do this, but it seems a good idea */
1319 if (!ipv6_addr_equals(&inner_key
.src
.addr
.ipv6_aligned
,
1320 &key
->dst
.addr
.ipv6_aligned
)
1321 || !ipv6_addr_equals(&inner_key
.dst
.addr
.ipv6_aligned
,
1322 &key
->src
.addr
.ipv6_aligned
)) {
1326 key
->src
= inner_key
.src
;
1327 key
->dst
= inner_key
.dst
;
1328 key
->nw_proto
= inner_key
.nw_proto
;
1330 ok
= extract_l4(key
, l4
, tail
- l4
, NULL
, l3
);
1332 conn_key_reverse(key
);
1344 /* Extract l4 fields into 'key', which must already contain valid l3
1347 * If 'related' is not NULL and an ICMP error packet is being
1348 * processed, the function will extract the key from the packet nested
1349 * in the ICMP paylod and set '*related' to true.
1351 * If 'related' is NULL, it means that we're already parsing a header nested
1352 * in an ICMP error. In this case, we skip checksum and length validation. */
1354 extract_l4(struct conn_key
*key
, const void *data
, size_t size
, bool *related
,
1357 if (key
->nw_proto
== IPPROTO_TCP
) {
1358 return (!related
|| check_l4_tcp(key
, data
, size
, l3
))
1359 && extract_l4_tcp(key
, data
, size
);
1360 } else if (key
->nw_proto
== IPPROTO_UDP
) {
1361 return (!related
|| check_l4_udp(key
, data
, size
, l3
))
1362 && extract_l4_udp(key
, data
, size
);
1363 } else if (key
->dl_type
== htons(ETH_TYPE_IP
)
1364 && key
->nw_proto
== IPPROTO_ICMP
) {
1365 return (!related
|| check_l4_icmp(data
, size
))
1366 && extract_l4_icmp(key
, data
, size
, related
);
1367 } else if (key
->dl_type
== htons(ETH_TYPE_IPV6
)
1368 && key
->nw_proto
== IPPROTO_ICMPV6
) {
1369 return (!related
|| check_l4_icmp6(key
, data
, size
, l3
))
1370 && extract_l4_icmp6(key
, data
, size
, related
);
1377 conn_key_extract(struct conntrack
*ct
, struct dp_packet
*pkt
, ovs_be16 dl_type
,
1378 struct conn_lookup_ctx
*ctx
, uint16_t zone
)
1380 const struct eth_header
*l2
= dp_packet_eth(pkt
);
1381 const struct ip_header
*l3
= dp_packet_l3(pkt
);
1382 const char *l4
= dp_packet_l4(pkt
);
1383 const char *tail
= dp_packet_tail(pkt
);
1386 memset(ctx
, 0, sizeof *ctx
);
1388 if (!l2
|| !l3
|| !l4
) {
1392 ctx
->key
.zone
= zone
;
1394 /* XXX In this function we parse the packet (again, it has already
1395 * gone through miniflow_extract()) for two reasons:
1397 * 1) To extract the l3 addresses and l4 ports.
1398 * We already have the l3 and l4 headers' pointers. Extracting
1399 * the l3 addresses and the l4 ports is really cheap, since they
1400 * can be found at fixed locations.
1401 * 2) To extract the l4 type.
1402 * Extracting the l4 types, for IPv6 can be quite expensive, because
1403 * it's not at a fixed location.
1405 * Here's a way to avoid (2) with the help of the datapath.
1406 * The datapath doesn't keep the packet's extracted flow[1], so
1407 * using that is not an option. We could use the packet's matching
1408 * megaflow, but we have to make sure that the l4 type (nw_proto)
1409 * is unwildcarded. This means either:
1411 * a) dpif-netdev unwildcards the l4 type when a new flow is installed
1412 * if the actions contains ct().
1414 * b) ofproto-dpif-xlate unwildcards the l4 type when translating a ct()
1415 * action. This is already done in different actions, but it's
1416 * unnecessary for the kernel.
1419 * [1] The reasons for this are that keeping the flow increases
1420 * (slightly) the cache footprint and increases computation
1421 * time as we move the packet around. Most importantly, the flow
1422 * should be updated by the actions and this can be slow, as
1423 * we use a sparse representation (miniflow).
1426 ctx
->key
.dl_type
= dl_type
;
1427 if (ctx
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
1428 ok
= extract_l3_ipv4(&ctx
->key
, l3
, tail
- (char *) l3
, NULL
, true);
1429 } else if (ctx
->key
.dl_type
== htons(ETH_TYPE_IPV6
)) {
1430 ok
= extract_l3_ipv6(&ctx
->key
, l3
, tail
- (char *) l3
, NULL
);
1436 if (extract_l4(&ctx
->key
, l4
, tail
- l4
, &ctx
->related
, l3
)) {
1437 ctx
->hash
= conn_key_hash(&ctx
->key
, ct
->hash_basis
);
1447 conn_key_hash(const struct conn_key
*key
, uint32_t basis
)
1449 uint32_t hsrc
, hdst
, hash
;
1452 hsrc
= hdst
= basis
;
1454 for (i
= 0; i
< sizeof(key
->src
) / sizeof(uint32_t); i
++) {
1455 hsrc
= hash_add(hsrc
, ((uint32_t *) &key
->src
)[i
]);
1456 hdst
= hash_add(hdst
, ((uint32_t *) &key
->dst
)[i
]);
1459 /* Even if source and destination are swapped the hash will be the same. */
1462 /* Hash the rest of the key(L3 and L4 types and zone). */
1463 hash
= hash_words((uint32_t *) (&key
->dst
+ 1),
1464 (uint32_t *) (key
+ 1) - (uint32_t *) (&key
->dst
+ 1),
1471 conn_key_reverse(struct conn_key
*key
)
1473 struct ct_endpoint tmp
;
1476 key
->src
= key
->dst
;
1481 nat_ipv6_addrs_delta(struct in6_addr
*ipv6_aligned_min
,
1482 struct in6_addr
*ipv6_aligned_max
)
1484 uint8_t *ipv6_min_hi
= &ipv6_aligned_min
->s6_addr
[0];
1485 uint8_t *ipv6_min_lo
= &ipv6_aligned_min
->s6_addr
[0] + sizeof(uint64_t);
1486 uint8_t *ipv6_max_hi
= &ipv6_aligned_max
->s6_addr
[0];
1487 uint8_t *ipv6_max_lo
= &ipv6_aligned_max
->s6_addr
[0] + sizeof(uint64_t);
1489 ovs_be64 addr6_64_min_hi
;
1490 ovs_be64 addr6_64_min_lo
;
1491 memcpy(&addr6_64_min_hi
, ipv6_min_hi
, sizeof addr6_64_min_hi
);
1492 memcpy(&addr6_64_min_lo
, ipv6_min_lo
, sizeof addr6_64_min_lo
);
1494 ovs_be64 addr6_64_max_hi
;
1495 ovs_be64 addr6_64_max_lo
;
1496 memcpy(&addr6_64_max_hi
, ipv6_max_hi
, sizeof addr6_64_max_hi
);
1497 memcpy(&addr6_64_max_lo
, ipv6_max_lo
, sizeof addr6_64_max_lo
);
1500 if (addr6_64_min_hi
== addr6_64_max_hi
&&
1501 ntohll(addr6_64_min_lo
) <= ntohll(addr6_64_max_lo
)) {
1502 diff
= ntohll(addr6_64_max_lo
) - ntohll(addr6_64_min_lo
);
1503 } else if (ntohll(addr6_64_min_hi
) + 1 == ntohll(addr6_64_max_hi
) &&
1504 ntohll(addr6_64_min_lo
) > ntohll(addr6_64_max_lo
)) {
1505 diff
= UINT64_MAX
- (ntohll(addr6_64_min_lo
) -
1506 ntohll(addr6_64_max_lo
) - 1);
1508 /* Limit address delta supported to 32 bits or 4 billion approximately.
1509 * Possibly, this should be visible to the user through a datapath
1510 * support check, however the practical impact is probably nil. */
1513 if (diff
> 0xfffffffe) {
1519 /* This function must be used in tandem with nat_ipv6_addrs_delta(), which
1520 * restricts the input parameters. */
1522 nat_ipv6_addr_increment(struct in6_addr
*ipv6_aligned
, uint32_t increment
)
1524 uint8_t *ipv6_hi
= &ipv6_aligned
->s6_addr
[0];
1525 uint8_t *ipv6_lo
= &ipv6_aligned
->s6_addr
[0] + sizeof(ovs_be64
);
1526 ovs_be64 addr6_64_hi
;
1527 ovs_be64 addr6_64_lo
;
1528 memcpy(&addr6_64_hi
, ipv6_hi
, sizeof addr6_64_hi
);
1529 memcpy(&addr6_64_lo
, ipv6_lo
, sizeof addr6_64_lo
);
1531 if (UINT64_MAX
- increment
>= ntohll(addr6_64_lo
)) {
1532 addr6_64_lo
= htonll(increment
+ ntohll(addr6_64_lo
));
1533 } else if (addr6_64_hi
!= OVS_BE64_MAX
) {
1534 addr6_64_hi
= htonll(1 + ntohll(addr6_64_hi
));
1535 addr6_64_lo
= htonll(increment
- (UINT64_MAX
-
1536 ntohll(addr6_64_lo
) + 1));
1541 memcpy(ipv6_hi
, &addr6_64_hi
, sizeof addr6_64_hi
);
1542 memcpy(ipv6_lo
, &addr6_64_lo
, sizeof addr6_64_lo
);
1548 nat_range_hash(const struct conn
*conn
, uint32_t basis
)
1550 uint32_t hash
= basis
;
1555 i
< sizeof(conn
->nat_info
->min_addr
) / sizeof(uint32_t);
1557 hash
= hash_add(hash
, ((uint32_t *) &conn
->nat_info
->min_addr
)[i
]);
1558 hash
= hash_add(hash
, ((uint32_t *) &conn
->nat_info
->max_addr
)[i
]);
1561 memcpy(&port
, &conn
->nat_info
->min_port
, sizeof port
);
1562 hash
= hash_add(hash
, port
);
1564 for (i
= 0; i
< sizeof(conn
->key
.src
.addr
) / sizeof(uint32_t); i
++) {
1565 hash
= hash_add(hash
, ((uint32_t *) &conn
->key
.src
)[i
]);
1566 hash
= hash_add(hash
, ((uint32_t *) &conn
->key
.dst
)[i
]);
1569 memcpy(&port
, &conn
->key
.src
.port
, sizeof port
);
1570 hash
= hash_add(hash
, port
);
1571 memcpy(&port
, &conn
->key
.dst
.port
, sizeof port
);
1572 hash
= hash_add(hash
, port
);
1574 hash
= hash_add(hash
, (OVS_FORCE
uint32_t) conn
->key
.dl_type
);
1575 hash
= hash_add(hash
, conn
->key
.nw_proto
);
1576 hash
= hash_add(hash
, conn
->key
.zone
);
1581 nat_select_range_tuple(struct conntrack
*ct
, const struct conn
*conn
,
1582 struct conn
*nat_conn
)
1584 #define MIN_NAT_EPHEMERAL_PORT 1024
1585 #define MAX_NAT_EPHEMERAL_PORT 65535
1589 uint16_t first_port
;
1591 uint32_t hash
= nat_range_hash(conn
, ct
->hash_basis
);
1593 if ((conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) &&
1594 (!(conn
->nat_info
->nat_action
& NAT_ACTION_SRC_PORT
))) {
1595 min_port
= ntohs(conn
->key
.src
.port
);
1596 max_port
= ntohs(conn
->key
.src
.port
);
1597 first_port
= min_port
;
1598 } else if ((conn
->nat_info
->nat_action
& NAT_ACTION_DST
) &&
1599 (!(conn
->nat_info
->nat_action
& NAT_ACTION_DST_PORT
))) {
1600 min_port
= ntohs(conn
->key
.dst
.port
);
1601 max_port
= ntohs(conn
->key
.dst
.port
);
1602 first_port
= min_port
;
1604 uint16_t deltap
= conn
->nat_info
->max_port
- conn
->nat_info
->min_port
;
1605 uint32_t port_index
= hash
% (deltap
+ 1);
1606 first_port
= conn
->nat_info
->min_port
+ port_index
;
1607 min_port
= conn
->nat_info
->min_port
;
1608 max_port
= conn
->nat_info
->max_port
;
1611 uint32_t deltaa
= 0;
1612 uint32_t address_index
;
1613 struct ct_addr ct_addr
;
1614 memset(&ct_addr
, 0, sizeof ct_addr
);
1615 struct ct_addr max_ct_addr
;
1616 memset(&max_ct_addr
, 0, sizeof max_ct_addr
);
1617 max_ct_addr
= conn
->nat_info
->max_addr
;
1619 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
1620 deltaa
= ntohl(conn
->nat_info
->max_addr
.ipv4_aligned
) -
1621 ntohl(conn
->nat_info
->min_addr
.ipv4_aligned
);
1622 address_index
= hash
% (deltaa
+ 1);
1623 ct_addr
.ipv4_aligned
= htonl(
1624 ntohl(conn
->nat_info
->min_addr
.ipv4_aligned
) + address_index
);
1626 deltaa
= nat_ipv6_addrs_delta(&conn
->nat_info
->min_addr
.ipv6_aligned
,
1627 &conn
->nat_info
->max_addr
.ipv6_aligned
);
1628 /* deltaa must be within 32 bits for full hash coverage. A 64 or
1629 * 128 bit hash is unnecessary and hence not used here. Most code
1630 * is kept common with V4; nat_ipv6_addrs_delta() will do the
1631 * enforcement via max_ct_addr. */
1632 max_ct_addr
= conn
->nat_info
->min_addr
;
1633 nat_ipv6_addr_increment(&max_ct_addr
.ipv6_aligned
, deltaa
);
1635 address_index
= hash
% (deltaa
+ 1);
1636 ct_addr
.ipv6_aligned
= conn
->nat_info
->min_addr
.ipv6_aligned
;
1637 nat_ipv6_addr_increment(&ct_addr
.ipv6_aligned
, address_index
);
1640 uint16_t port
= first_port
;
1641 bool all_ports_tried
= false;
1642 bool original_ports_tried
= false;
1643 struct ct_addr first_addr
= ct_addr
;
1647 if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
1648 nat_conn
->rev_key
.dst
.addr
= ct_addr
;
1650 nat_conn
->rev_key
.src
.addr
= ct_addr
;
1653 if ((conn
->key
.nw_proto
== IPPROTO_ICMP
) ||
1654 (conn
->key
.nw_proto
== IPPROTO_ICMPV6
)) {
1655 all_ports_tried
= true;
1656 } else if (conn
->nat_info
->nat_action
& NAT_ACTION_SRC
) {
1657 nat_conn
->rev_key
.dst
.port
= htons(port
);
1659 nat_conn
->rev_key
.src
.port
= htons(port
);
1662 struct nat_conn_key_node
*nat_conn_key_node
=
1663 nat_conn_keys_lookup(&ct
->nat_conn_keys
, &nat_conn
->rev_key
,
1666 if (!nat_conn_key_node
) {
1667 struct nat_conn_key_node
*nat_conn_key
=
1668 xzalloc(sizeof *nat_conn_key
);
1669 nat_conn_key
->key
= nat_conn
->rev_key
;
1670 nat_conn_key
->value
= nat_conn
->key
;
1671 uint32_t nat_conn_key_hash
= conn_key_hash(&nat_conn_key
->key
,
1673 hmap_insert(&ct
->nat_conn_keys
, &nat_conn_key
->node
,
1676 } else if (!all_ports_tried
) {
1677 if (min_port
== max_port
) {
1678 all_ports_tried
= true;
1679 } else if (port
== max_port
) {
1684 if (port
== first_port
) {
1685 all_ports_tried
= true;
1688 if (memcmp(&ct_addr
, &max_ct_addr
, sizeof ct_addr
)) {
1689 if (conn
->key
.dl_type
== htons(ETH_TYPE_IP
)) {
1690 ct_addr
.ipv4_aligned
= htonl(
1691 ntohl(ct_addr
.ipv4_aligned
) + 1);
1693 nat_ipv6_addr_increment(&ct_addr
.ipv6_aligned
, 1);
1696 ct_addr
= conn
->nat_info
->min_addr
;
1698 if (!memcmp(&ct_addr
, &first_addr
, sizeof ct_addr
)) {
1699 if (!original_ports_tried
) {
1700 original_ports_tried
= true;
1701 ct_addr
= conn
->nat_info
->min_addr
;
1702 min_port
= MIN_NAT_EPHEMERAL_PORT
;
1703 max_port
= MAX_NAT_EPHEMERAL_PORT
;
1708 first_port
= min_port
;
1710 all_ports_tried
= false;
1716 static struct nat_conn_key_node
*
1717 nat_conn_keys_lookup(struct hmap
*nat_conn_keys
,
1718 const struct conn_key
*key
,
1721 struct nat_conn_key_node
*nat_conn_key_node
;
1722 uint32_t nat_conn_key_hash
= conn_key_hash(key
, basis
);
1724 HMAP_FOR_EACH_WITH_HASH (nat_conn_key_node
, node
, nat_conn_key_hash
,
1726 if (!memcmp(&nat_conn_key_node
->key
, key
,
1727 sizeof nat_conn_key_node
->key
)) {
1728 return nat_conn_key_node
;
1735 nat_conn_keys_remove(struct hmap
*nat_conn_keys
, const struct conn_key
*key
,
1738 struct nat_conn_key_node
*nat_conn_key_node
;
1739 uint32_t nat_conn_key_hash
= conn_key_hash(key
, basis
);
1741 HMAP_FOR_EACH_WITH_HASH (nat_conn_key_node
, node
, nat_conn_key_hash
,
1743 if (!memcmp(&nat_conn_key_node
->key
, key
,
1744 sizeof nat_conn_key_node
->key
)) {
1745 hmap_remove(nat_conn_keys
, &nat_conn_key_node
->node
);
1746 free(nat_conn_key_node
);
1753 conn_key_lookup(struct conntrack_bucket
*ctb
, struct conn_lookup_ctx
*ctx
,
1756 uint32_t hash
= ctx
->hash
;
1761 HMAP_FOR_EACH_WITH_HASH (conn
, node
, hash
, &ctb
->connections
) {
1762 if (!memcmp(&conn
->key
, &ctx
->key
, sizeof conn
->key
)
1763 && !conn_expired(conn
, now
)) {
1768 if (!memcmp(&conn
->rev_key
, &ctx
->key
, sizeof conn
->rev_key
)
1769 && !conn_expired(conn
, now
)) {
1777 static enum ct_update_res
1778 conn_update(struct conn
*conn
, struct conntrack_bucket
*ctb
,
1779 struct dp_packet
*pkt
, bool reply
, long long now
)
1781 return l4_protos
[conn
->key
.nw_proto
]->conn_update(conn
, ctb
, pkt
,
1786 conn_expired(struct conn
*conn
, long long now
)
1788 if (conn
->conn_type
== CT_CONN_TYPE_DEFAULT
) {
1789 return now
>= conn
->expiration
;
1795 valid_new(struct dp_packet
*pkt
, struct conn_key
*key
)
1797 return l4_protos
[key
->nw_proto
]->valid_new(pkt
);
1800 static struct conn
*
1801 new_conn(struct conntrack_bucket
*ctb
, struct dp_packet
*pkt
,
1802 struct conn_key
*key
, long long now
)
1804 struct conn
*newconn
;
1806 newconn
= l4_protos
[key
->nw_proto
]->new_conn(ctb
, pkt
, now
);
1809 newconn
->key
= *key
;
1816 delete_conn(struct conn
*conn
)
1818 free(conn
->nat_info
);
1823 ct_endpoint_to_ct_dpif_inet_addr(const struct ct_addr
*a
,
1824 union ct_dpif_inet_addr
*b
,
1827 if (dl_type
== htons(ETH_TYPE_IP
)) {
1828 b
->ip
= a
->ipv4_aligned
;
1829 } else if (dl_type
== htons(ETH_TYPE_IPV6
)){
1830 b
->in6
= a
->ipv6_aligned
;
1835 conn_key_to_tuple(const struct conn_key
*key
, struct ct_dpif_tuple
*tuple
)
1837 if (key
->dl_type
== htons(ETH_TYPE_IP
)) {
1838 tuple
->l3_type
= AF_INET
;
1839 } else if (key
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1840 tuple
->l3_type
= AF_INET6
;
1842 tuple
->ip_proto
= key
->nw_proto
;
1843 ct_endpoint_to_ct_dpif_inet_addr(&key
->src
.addr
, &tuple
->src
,
1845 ct_endpoint_to_ct_dpif_inet_addr(&key
->dst
.addr
, &tuple
->dst
,
1848 if (key
->nw_proto
== IPPROTO_ICMP
|| key
->nw_proto
== IPPROTO_ICMPV6
) {
1849 tuple
->icmp_id
= key
->src
.icmp_id
;
1850 tuple
->icmp_type
= key
->src
.icmp_type
;
1851 tuple
->icmp_code
= key
->src
.icmp_code
;
1853 tuple
->src_port
= key
->src
.port
;
1854 tuple
->dst_port
= key
->dst
.port
;
1859 conn_to_ct_dpif_entry(const struct conn
*conn
, struct ct_dpif_entry
*entry
,
1862 struct ct_l4_proto
*class;
1863 long long expiration
;
1864 memset(entry
, 0, sizeof *entry
);
1865 conn_key_to_tuple(&conn
->key
, &entry
->tuple_orig
);
1866 conn_key_to_tuple(&conn
->rev_key
, &entry
->tuple_reply
);
1868 entry
->zone
= conn
->key
.zone
;
1869 entry
->mark
= conn
->mark
;
1871 memcpy(&entry
->labels
, &conn
->label
, sizeof entry
->labels
);
1872 /* Not implemented yet */
1873 entry
->timestamp
.start
= 0;
1874 entry
->timestamp
.stop
= 0;
1876 expiration
= conn
->expiration
- now
;
1877 entry
->timeout
= (expiration
> 0) ? expiration
/ 1000 : 0;
1879 class = l4_protos
[conn
->key
.nw_proto
];
1880 if (class->conn_get_protoinfo
) {
1881 class->conn_get_protoinfo(conn
, &entry
->protoinfo
);
1886 conntrack_dump_start(struct conntrack
*ct
, struct conntrack_dump
*dump
,
1887 const uint16_t *pzone
)
1889 memset(dump
, 0, sizeof(*dump
));
1891 dump
->zone
= *pzone
;
1892 dump
->filter_zone
= true;
1900 conntrack_dump_next(struct conntrack_dump
*dump
, struct ct_dpif_entry
*entry
)
1902 struct conntrack
*ct
= dump
->ct
;
1903 long long now
= time_msec();
1905 while (dump
->bucket
< CONNTRACK_BUCKETS
) {
1906 struct hmap_node
*node
;
1908 ct_lock_lock(&ct
->buckets
[dump
->bucket
].lock
);
1912 node
= hmap_at_position(&ct
->buckets
[dump
->bucket
].connections
,
1917 INIT_CONTAINER(conn
, node
, node
);
1918 if ((!dump
->filter_zone
|| conn
->key
.zone
== dump
->zone
) &&
1919 (conn
->conn_type
!= CT_CONN_TYPE_UN_NAT
)) {
1920 conn_to_ct_dpif_entry(conn
, entry
, now
);
1923 /* Else continue, until we find an entry in the appropriate zone
1924 * or the bucket has been scanned completely. */
1926 ct_lock_unlock(&ct
->buckets
[dump
->bucket
].lock
);
1929 memset(&dump
->bucket_pos
, 0, sizeof dump
->bucket_pos
);
1939 conntrack_dump_done(struct conntrack_dump
*dump OVS_UNUSED
)
1945 conntrack_flush(struct conntrack
*ct
, const uint16_t *zone
)
1949 for (i
= 0; i
< CONNTRACK_BUCKETS
; i
++) {
1950 struct conn
*conn
, *next
;
1952 ct_lock_lock(&ct
->buckets
[i
].lock
);
1953 HMAP_FOR_EACH_SAFE(conn
, next
, node
, &ct
->buckets
[i
].connections
) {
1954 if ((!zone
|| *zone
== conn
->key
.zone
) &&
1955 (conn
->conn_type
== CT_CONN_TYPE_DEFAULT
)) {
1956 conn_clean(ct
, conn
, &ct
->buckets
[i
]);
1959 ct_lock_unlock(&ct
->buckets
[i
].lock
);