]>
Commit | Line | Data |
---|---|---|
a489b168 | 1 | /* |
4ea96698 | 2 | * Copyright (c) 2015-2019 Nicira, Inc. |
a489b168 DDP |
3 | * |
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: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
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. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
bd5e81a0 | 18 | #include <ctype.h> |
a489b168 | 19 | #include <errno.h> |
ff6aa424 | 20 | #include <sys/types.h> |
a489b168 DDP |
21 | #include <netinet/in.h> |
22 | #include <netinet/icmp6.h> | |
bd5e81a0 | 23 | #include <string.h> |
a489b168 DDP |
24 | |
25 | #include "bitmap.h" | |
bd5e81a0 | 26 | #include "conntrack.h" |
a489b168 DDP |
27 | #include "conntrack-private.h" |
28 | #include "coverage.h" | |
29 | #include "csum.h" | |
4d4e68ed | 30 | #include "ct-dpif.h" |
a489b168 DDP |
31 | #include "dp-packet.h" |
32 | #include "flow.h" | |
4ea96698 | 33 | #include "ipf.h" |
a489b168 DDP |
34 | #include "netdev.h" |
35 | #include "odp-netlink.h" | |
36 | #include "openvswitch/hmap.h" | |
37 | #include "openvswitch/vlog.h" | |
38 | #include "ovs-rcu.h" | |
e6ef6cc6 | 39 | #include "ovs-thread.h" |
fd016ae3 | 40 | #include "openvswitch/poll-loop.h" |
a489b168 DDP |
41 | #include "random.h" |
42 | #include "timeval.h" | |
43 | ||
44 | VLOG_DEFINE_THIS_MODULE(conntrack); | |
45 | ||
46 | COVERAGE_DEFINE(conntrack_full); | |
e6ef6cc6 | 47 | COVERAGE_DEFINE(conntrack_long_cleanup); |
a489b168 DDP |
48 | |
49 | struct conn_lookup_ctx { | |
50 | struct conn_key key; | |
51 | struct conn *conn; | |
52 | uint32_t hash; | |
53 | bool reply; | |
dbb597d3 | 54 | bool icmp_related; |
a489b168 DDP |
55 | }; |
56 | ||
bd5e81a0 DB |
57 | enum ftp_ctl_pkt { |
58 | /* Control packets with address and/or port specifiers. */ | |
59 | CT_FTP_CTL_INTEREST, | |
60 | /* Control packets without address and/or port specifiers. */ | |
61 | CT_FTP_CTL_OTHER, | |
62 | CT_FTP_CTL_INVALID, | |
63 | }; | |
64 | ||
65 | enum ct_alg_mode { | |
66 | CT_FTP_MODE_ACTIVE, | |
67 | CT_FTP_MODE_PASSIVE, | |
7be77cb0 | 68 | CT_TFTP_MODE, |
bd5e81a0 DB |
69 | }; |
70 | ||
94e71143 DB |
71 | enum ct_alg_ctl_type { |
72 | CT_ALG_CTL_NONE, | |
73 | CT_ALG_CTL_FTP, | |
74 | CT_ALG_CTL_TFTP, | |
be38342d DB |
75 | /* SIP is not enabled through Openflow and presently only used as |
76 | * an example of an alg that allows a wildcard src ip. */ | |
77 | CT_ALG_CTL_SIP, | |
94e71143 DB |
78 | }; |
79 | ||
a489b168 | 80 | static bool conn_key_extract(struct conntrack *, struct dp_packet *, |
66e4ad8a DDP |
81 | ovs_be16 dl_type, struct conn_lookup_ctx *, |
82 | uint16_t zone); | |
a489b168 DDP |
83 | static uint32_t conn_key_hash(const struct conn_key *, uint32_t basis); |
84 | static void conn_key_reverse(struct conn_key *); | |
85 | static void conn_key_lookup(struct conntrack_bucket *ctb, | |
86 | struct conn_lookup_ctx *ctx, | |
87 | long long now); | |
88 | static bool valid_new(struct dp_packet *pkt, struct conn_key *); | |
e6ef6cc6 DDP |
89 | static struct conn *new_conn(struct conntrack_bucket *, struct dp_packet *pkt, |
90 | struct conn_key *, long long now); | |
a489b168 | 91 | static void delete_conn(struct conn *); |
e6ef6cc6 DDP |
92 | static enum ct_update_res conn_update(struct conn *, |
93 | struct conntrack_bucket *ctb, | |
94 | struct dp_packet *, bool reply, | |
95 | long long now); | |
a489b168 DDP |
96 | static bool conn_expired(struct conn *, long long now); |
97 | static void set_mark(struct dp_packet *, struct conn *, | |
98 | uint32_t val, uint32_t mask); | |
99 | static void set_label(struct dp_packet *, struct conn *, | |
100 | const struct ovs_key_ct_labels *val, | |
101 | const struct ovs_key_ct_labels *mask); | |
e6ef6cc6 | 102 | static void *clean_thread_main(void *f_); |
a489b168 | 103 | |
286de272 DB |
104 | static struct nat_conn_key_node * |
105 | nat_conn_keys_lookup(struct hmap *nat_conn_keys, | |
106 | const struct conn_key *key, | |
107 | uint32_t basis); | |
108 | ||
80cee116 DB |
109 | static bool |
110 | nat_conn_keys_insert(struct hmap *nat_conn_keys, | |
111 | const struct conn *nat_conn, | |
112 | uint32_t hash_basis); | |
113 | ||
286de272 DB |
114 | static void |
115 | nat_conn_keys_remove(struct hmap *nat_conn_keys, | |
116 | const struct conn_key *key, | |
117 | uint32_t basis); | |
118 | ||
119 | static bool | |
120 | nat_select_range_tuple(struct conntrack *ct, const struct conn *conn, | |
121 | struct conn *nat_conn); | |
122 | ||
123 | static uint8_t | |
124 | reverse_icmp_type(uint8_t type); | |
125 | static uint8_t | |
126 | reverse_icmp6_type(uint8_t type); | |
127 | static inline bool | |
128 | extract_l3_ipv4(struct conn_key *key, const void *data, size_t size, | |
129 | const char **new_data, bool validate_checksum); | |
130 | static inline bool | |
131 | extract_l3_ipv6(struct conn_key *key, const void *data, size_t size, | |
132 | const char **new_data); | |
bd5e81a0 | 133 | static struct alg_exp_node * |
be38342d DB |
134 | expectation_lookup(struct hmap *alg_expectations, const struct conn_key *key, |
135 | uint32_t basis, bool src_ip_wc); | |
bd5e81a0 DB |
136 | |
137 | static int | |
138 | repl_ftp_v4_addr(struct dp_packet *pkt, ovs_be32 v4_addr_rep, | |
139 | char *ftp_data_v4_start, | |
cd7c99a6 | 140 | size_t addr_offset_from_ftp_data_start, size_t addr_size); |
bd5e81a0 DB |
141 | |
142 | static enum ftp_ctl_pkt | |
143 | process_ftp_ctl_v4(struct conntrack *ct, | |
144 | struct dp_packet *pkt, | |
145 | const struct conn *conn_for_expectation, | |
4417ca3d | 146 | ovs_be32 *v4_addr_rep, |
bd5e81a0 | 147 | char **ftp_data_v4_start, |
cd7c99a6 DB |
148 | size_t *addr_offset_from_ftp_data_start, |
149 | size_t *addr_size); | |
bd5e81a0 DB |
150 | |
151 | static enum ftp_ctl_pkt | |
152 | detect_ftp_ctl_type(const struct conn_lookup_ctx *ctx, | |
153 | struct dp_packet *pkt); | |
154 | ||
4417ca3d DB |
155 | static void |
156 | expectation_clean(struct conntrack *ct, const struct conn_key *master_key, | |
157 | uint32_t basis); | |
158 | ||
94e71143 DB |
159 | static struct ct_l4_proto *l4_protos[] = { |
160 | [IPPROTO_TCP] = &ct_proto_tcp, | |
161 | [IPPROTO_UDP] = &ct_proto_other, | |
162 | [IPPROTO_ICMP] = &ct_proto_icmp4, | |
163 | [IPPROTO_ICMPV6] = &ct_proto_icmp6, | |
164 | }; | |
165 | ||
bd5e81a0 DB |
166 | static void |
167 | handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
168 | struct dp_packet *pkt, | |
169 | const struct conn *conn_for_expectation, | |
170 | long long now, enum ftp_ctl_pkt ftp_ctl, bool nat); | |
171 | ||
7be77cb0 DB |
172 | static void |
173 | handle_tftp_ctl(struct conntrack *ct, | |
94e71143 | 174 | const struct conn_lookup_ctx *ctx OVS_UNUSED, |
be38342d | 175 | struct dp_packet *pkt, |
7be77cb0 | 176 | const struct conn *conn_for_expectation, |
4417ca3d DB |
177 | long long now OVS_UNUSED, |
178 | enum ftp_ctl_pkt ftp_ctl OVS_UNUSED, bool nat OVS_UNUSED); | |
94e71143 DB |
179 | |
180 | typedef void (*alg_helper)(struct conntrack *ct, | |
181 | const struct conn_lookup_ctx *ctx, | |
182 | struct dp_packet *pkt, | |
183 | const struct conn *conn_for_expectation, | |
184 | long long now, enum ftp_ctl_pkt ftp_ctl, | |
185 | bool nat); | |
186 | ||
187 | static alg_helper alg_helpers[] = { | |
188 | [CT_ALG_CTL_NONE] = NULL, | |
189 | [CT_ALG_CTL_FTP] = handle_ftp_ctl, | |
190 | [CT_ALG_CTL_TFTP] = handle_tftp_ctl, | |
a489b168 DDP |
191 | }; |
192 | ||
193 | long long ct_timeout_val[] = { | |
194 | #define CT_TIMEOUT(NAME, VAL) [CT_TM_##NAME] = VAL, | |
195 | CT_TIMEOUTS | |
196 | #undef CT_TIMEOUT | |
197 | }; | |
198 | ||
bd5e81a0 DB |
199 | /* The maximum TCP or UDP port number. */ |
200 | #define CT_MAX_L4_PORT 65535 | |
bd5e81a0 DB |
201 | /* String buffer used for parsing FTP string messages. |
202 | * This is sized about twice what is needed to leave some | |
203 | * margin of error. */ | |
204 | #define LARGEST_FTP_MSG_OF_INTEREST 128 | |
205 | /* FTP port string used in active mode. */ | |
206 | #define FTP_PORT_CMD "PORT" | |
207 | /* FTP pasv string used in passive mode. */ | |
208 | #define FTP_PASV_REPLY_CODE "227" | |
209 | /* Maximum decimal digits for port in FTP command. | |
210 | * The port is represented as two 3 digit numbers with the | |
211 | * high part a multiple of 256. */ | |
212 | #define MAX_FTP_PORT_DGTS 3 | |
213 | ||
214 | /* FTP extension EPRT string used for active mode. */ | |
215 | #define FTP_EPRT_CMD "EPRT" | |
216 | /* FTP extension EPSV string used for passive mode. */ | |
217 | #define FTP_EPSV_REPLY "EXTENDED PASSIVE" | |
218 | /* Maximum decimal digits for port in FTP extended command. */ | |
219 | #define MAX_EXT_FTP_PORT_DGTS 5 | |
220 | /* FTP extended command code for IPv6. */ | |
221 | #define FTP_AF_V6 '2' | |
222 | /* Used to indicate a wildcard L4 source port number for ALGs. | |
223 | * This is used for port numbers that we cannot predict in | |
224 | * expectations. */ | |
225 | #define ALG_WC_SRC_PORT 0 | |
226 | ||
a489b168 | 227 | /* If the total number of connections goes above this value, no new connections |
286de272 | 228 | * are accepted; this is for CT_CONN_TYPE_DEFAULT connections. */ |
a489b168 DDP |
229 | #define DEFAULT_N_CONN_LIMIT 3000000 |
230 | ||
5ed7a0b4 DB |
231 | /* Does a member by member comparison of two conn_keys; this |
232 | * function must be kept in sync with struct conn_key; returns 0 | |
233 | * if the keys are equal or 1 if the keys are not equal. */ | |
234 | static int | |
235 | conn_key_cmp(const struct conn_key *key1, const struct conn_key *key2) | |
236 | { | |
237 | if (!memcmp(&key1->src.addr, &key2->src.addr, sizeof key1->src.addr) && | |
238 | !memcmp(&key1->dst.addr, &key2->dst.addr, sizeof key1->dst.addr) && | |
239 | (key1->src.icmp_id == key2->src.icmp_id) && | |
240 | (key1->src.icmp_type == key2->src.icmp_type) && | |
241 | (key1->src.icmp_code == key2->src.icmp_code) && | |
242 | (key1->dst.icmp_id == key2->dst.icmp_id) && | |
243 | (key1->dst.icmp_type == key2->dst.icmp_type) && | |
244 | (key1->dst.icmp_code == key2->dst.icmp_code) && | |
245 | (key1->dl_type == key2->dl_type) && | |
246 | (key1->zone == key2->zone) && | |
247 | (key1->nw_proto == key2->nw_proto)) { | |
248 | ||
249 | return 0; | |
250 | } | |
251 | return 1; | |
252 | } | |
253 | ||
d8682ee5 | 254 | static void |
dec0dbbc DB |
255 | ct_print_conn_info(const struct conn *c, const char *log_msg, |
256 | enum vlog_level vll, bool force, bool rl_on) | |
66f400f5 DB |
257 | { |
258 | #define CT_VLOG(RL_ON, LEVEL, ...) \ | |
259 | do { \ | |
260 | if (RL_ON) { \ | |
261 | static struct vlog_rate_limit rl_ = VLOG_RATE_LIMIT_INIT(5, 5); \ | |
262 | vlog_rate_limit(&this_module, LEVEL, &rl_, __VA_ARGS__); \ | |
263 | } else { \ | |
264 | vlog(&this_module, LEVEL, __VA_ARGS__); \ | |
265 | } \ | |
266 | } while (0) | |
267 | ||
268 | if (OVS_UNLIKELY(force || vlog_is_enabled(&this_module, vll))) { | |
269 | if (c->key.dl_type == htons(ETH_TYPE_IP)) { | |
270 | CT_VLOG(rl_on, vll, "%s: src ip "IP_FMT" dst ip "IP_FMT" rev src " | |
271 | "ip "IP_FMT" rev dst ip "IP_FMT" src/dst ports " | |
272 | "%"PRIu16"/%"PRIu16" rev src/dst ports " | |
273 | "%"PRIu16"/%"PRIu16" zone/rev zone " | |
274 | "%"PRIu16"/%"PRIu16" nw_proto/rev nw_proto " | |
275 | "%"PRIu8"/%"PRIu8, log_msg, | |
cda1b109 DB |
276 | IP_ARGS(c->key.src.addr.ipv4), |
277 | IP_ARGS(c->key.dst.addr.ipv4), | |
278 | IP_ARGS(c->rev_key.src.addr.ipv4), | |
279 | IP_ARGS(c->rev_key.dst.addr.ipv4), | |
66f400f5 DB |
280 | ntohs(c->key.src.port), ntohs(c->key.dst.port), |
281 | ntohs(c->rev_key.src.port), ntohs(c->rev_key.dst.port), | |
282 | c->key.zone, c->rev_key.zone, c->key.nw_proto, | |
283 | c->rev_key.nw_proto); | |
284 | } else { | |
285 | char ip6_s[INET6_ADDRSTRLEN]; | |
286 | inet_ntop(AF_INET6, &c->key.src.addr.ipv6, ip6_s, sizeof ip6_s); | |
287 | char ip6_d[INET6_ADDRSTRLEN]; | |
288 | inet_ntop(AF_INET6, &c->key.dst.addr.ipv6, ip6_d, sizeof ip6_d); | |
289 | char ip6_rs[INET6_ADDRSTRLEN]; | |
290 | inet_ntop(AF_INET6, &c->rev_key.src.addr.ipv6, ip6_rs, | |
291 | sizeof ip6_rs); | |
292 | char ip6_rd[INET6_ADDRSTRLEN]; | |
293 | inet_ntop(AF_INET6, &c->rev_key.dst.addr.ipv6, ip6_rd, | |
294 | sizeof ip6_rd); | |
295 | ||
296 | CT_VLOG(rl_on, vll, "%s: src ip %s dst ip %s rev src ip %s" | |
297 | " rev dst ip %s src/dst ports %"PRIu16"/%"PRIu16 | |
298 | " rev src/dst ports %"PRIu16"/%"PRIu16" zone/rev zone " | |
299 | "%"PRIu16"/%"PRIu16" nw_proto/rev nw_proto " | |
300 | "%"PRIu8"/%"PRIu8, log_msg, ip6_s, ip6_d, ip6_rs, | |
301 | ip6_rd, ntohs(c->key.src.port), ntohs(c->key.dst.port), | |
302 | ntohs(c->rev_key.src.port), ntohs(c->rev_key.dst.port), | |
303 | c->key.zone, c->rev_key.zone, c->key.nw_proto, | |
304 | c->rev_key.nw_proto); | |
305 | } | |
306 | } | |
307 | } | |
308 | ||
a489b168 DDP |
309 | /* Initializes the connection tracker 'ct'. The caller is responsible for |
310 | * calling 'conntrack_destroy()', when the instance is not needed anymore */ | |
311 | void | |
312 | conntrack_init(struct conntrack *ct) | |
313 | { | |
e6ef6cc6 | 314 | long long now = time_msec(); |
a489b168 | 315 | |
8b934ced DB |
316 | ct_rwlock_init(&ct->resources_lock); |
317 | ct_rwlock_wrlock(&ct->resources_lock); | |
286de272 | 318 | hmap_init(&ct->nat_conn_keys); |
bd5e81a0 | 319 | hmap_init(&ct->alg_expectations); |
4417ca3d | 320 | hindex_init(&ct->alg_expectation_refs); |
bd5e81a0 | 321 | ovs_list_init(&ct->alg_exp_list); |
8b934ced | 322 | ct_rwlock_unlock(&ct->resources_lock); |
286de272 | 323 | |
dec0dbbc | 324 | for (unsigned i = 0; i < CONNTRACK_BUCKETS; i++) { |
a489b168 DDP |
325 | struct conntrack_bucket *ctb = &ct->buckets[i]; |
326 | ||
327 | ct_lock_init(&ctb->lock); | |
328 | ct_lock_lock(&ctb->lock); | |
329 | hmap_init(&ctb->connections); | |
dec0dbbc | 330 | for (unsigned j = 0; j < ARRAY_SIZE(ctb->exp_lists); j++) { |
e6ef6cc6 DDP |
331 | ovs_list_init(&ctb->exp_lists[j]); |
332 | } | |
a489b168 | 333 | ct_lock_unlock(&ctb->lock); |
e6ef6cc6 DDP |
334 | ovs_mutex_init(&ctb->cleanup_mutex); |
335 | ovs_mutex_lock(&ctb->cleanup_mutex); | |
336 | ctb->next_cleanup = now + CT_TM_MIN; | |
337 | ovs_mutex_unlock(&ctb->cleanup_mutex); | |
a489b168 DDP |
338 | } |
339 | ct->hash_basis = random_uint32(); | |
340 | atomic_count_init(&ct->n_conn, 0); | |
341 | atomic_init(&ct->n_conn_limit, DEFAULT_N_CONN_LIMIT); | |
e6ef6cc6 DDP |
342 | latch_init(&ct->clean_thread_exit); |
343 | ct->clean_thread = ovs_thread_create("ct_clean", clean_thread_main, ct); | |
4ea96698 | 344 | ct->ipf = ipf_init(); |
a489b168 DDP |
345 | } |
346 | ||
347 | /* Destroys the connection tracker 'ct' and frees all the allocated memory. */ | |
348 | void | |
349 | conntrack_destroy(struct conntrack *ct) | |
350 | { | |
e6ef6cc6 DDP |
351 | latch_set(&ct->clean_thread_exit); |
352 | pthread_join(ct->clean_thread, NULL); | |
353 | latch_destroy(&ct->clean_thread_exit); | |
dec0dbbc | 354 | for (unsigned i = 0; i < CONNTRACK_BUCKETS; i++) { |
a489b168 DDP |
355 | struct conntrack_bucket *ctb = &ct->buckets[i]; |
356 | struct conn *conn; | |
357 | ||
e6ef6cc6 | 358 | ovs_mutex_destroy(&ctb->cleanup_mutex); |
a489b168 | 359 | ct_lock_lock(&ctb->lock); |
bd5e81a0 | 360 | HMAP_FOR_EACH_POP (conn, node, &ctb->connections) { |
286de272 DB |
361 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
362 | atomic_count_dec(&ct->n_conn); | |
363 | } | |
a489b168 DDP |
364 | delete_conn(conn); |
365 | } | |
366 | hmap_destroy(&ctb->connections); | |
367 | ct_lock_unlock(&ctb->lock); | |
368 | ct_lock_destroy(&ctb->lock); | |
369 | } | |
8b934ced | 370 | ct_rwlock_wrlock(&ct->resources_lock); |
286de272 DB |
371 | struct nat_conn_key_node *nat_conn_key_node; |
372 | HMAP_FOR_EACH_POP (nat_conn_key_node, node, &ct->nat_conn_keys) { | |
373 | free(nat_conn_key_node); | |
374 | } | |
375 | hmap_destroy(&ct->nat_conn_keys); | |
bd5e81a0 DB |
376 | |
377 | struct alg_exp_node *alg_exp_node; | |
378 | HMAP_FOR_EACH_POP (alg_exp_node, node, &ct->alg_expectations) { | |
379 | free(alg_exp_node); | |
380 | } | |
4417ca3d | 381 | |
bd5e81a0 DB |
382 | ovs_list_poison(&ct->alg_exp_list); |
383 | hmap_destroy(&ct->alg_expectations); | |
4417ca3d | 384 | hindex_destroy(&ct->alg_expectation_refs); |
8b934ced DB |
385 | ct_rwlock_unlock(&ct->resources_lock); |
386 | ct_rwlock_destroy(&ct->resources_lock); | |
4ea96698 | 387 | ipf_destroy(ct->ipf); |
a489b168 DDP |
388 | } |
389 | \f | |
390 | static unsigned hash_to_bucket(uint32_t hash) | |
391 | { | |
392 | /* Extracts the most significant bits in hash. The least significant bits | |
393 | * are already used internally by the hmap implementation. */ | |
394 | BUILD_ASSERT(CONNTRACK_BUCKETS_SHIFT < 32 && CONNTRACK_BUCKETS_SHIFT >= 1); | |
395 | ||
396 | return (hash >> (32 - CONNTRACK_BUCKETS_SHIFT)) % CONNTRACK_BUCKETS; | |
397 | } | |
398 | ||
399 | static void | |
286de272 | 400 | write_ct_md(struct dp_packet *pkt, uint16_t zone, const struct conn *conn, |
bd5e81a0 | 401 | const struct conn_key *key, const struct alg_exp_node *alg_exp) |
a489b168 | 402 | { |
286de272 | 403 | pkt->md.ct_state |= CS_TRACKED; |
a489b168 | 404 | pkt->md.ct_zone = zone; |
daf4d3c1 JR |
405 | pkt->md.ct_mark = conn ? conn->mark : 0; |
406 | pkt->md.ct_label = conn ? conn->label : OVS_U128_ZERO; | |
407 | ||
408 | /* Use the original direction tuple if we have it. */ | |
409 | if (conn) { | |
bd5e81a0 DB |
410 | if (conn->alg_related) { |
411 | key = &conn->master_key; | |
412 | } else { | |
413 | key = &conn->key; | |
414 | } | |
415 | } else if (alg_exp) { | |
416 | pkt->md.ct_mark = alg_exp->master_mark; | |
417 | pkt->md.ct_label = alg_exp->master_label; | |
418 | key = &alg_exp->master_key; | |
daf4d3c1 | 419 | } |
dec0dbbc | 420 | |
daf4d3c1 | 421 | pkt->md.ct_orig_tuple_ipv6 = false; |
dec0dbbc | 422 | |
daf4d3c1 JR |
423 | if (key) { |
424 | if (key->dl_type == htons(ETH_TYPE_IP)) { | |
425 | pkt->md.ct_orig_tuple.ipv4 = (struct ovs_key_ct_tuple_ipv4) { | |
cda1b109 DB |
426 | key->src.addr.ipv4, |
427 | key->dst.addr.ipv4, | |
daf4d3c1 JR |
428 | key->nw_proto != IPPROTO_ICMP |
429 | ? key->src.port : htons(key->src.icmp_type), | |
430 | key->nw_proto != IPPROTO_ICMP | |
431 | ? key->dst.port : htons(key->src.icmp_code), | |
432 | key->nw_proto, | |
433 | }; | |
286de272 | 434 | } else { |
daf4d3c1 JR |
435 | pkt->md.ct_orig_tuple_ipv6 = true; |
436 | pkt->md.ct_orig_tuple.ipv6 = (struct ovs_key_ct_tuple_ipv6) { | |
cda1b109 DB |
437 | key->src.addr.ipv6, |
438 | key->dst.addr.ipv6, | |
daf4d3c1 JR |
439 | key->nw_proto != IPPROTO_ICMPV6 |
440 | ? key->src.port : htons(key->src.icmp_type), | |
441 | key->nw_proto != IPPROTO_ICMPV6 | |
442 | ? key->dst.port : htons(key->src.icmp_code), | |
443 | key->nw_proto, | |
444 | }; | |
445 | } | |
446 | } else { | |
447 | memset(&pkt->md.ct_orig_tuple, 0, sizeof pkt->md.ct_orig_tuple); | |
448 | } | |
bd5e81a0 DB |
449 | } |
450 | ||
451 | static uint8_t | |
452 | get_ip_proto(const struct dp_packet *pkt) | |
453 | { | |
454 | uint8_t ip_proto; | |
455 | struct eth_header *l2 = dp_packet_eth(pkt); | |
456 | if (l2->eth_type == htons(ETH_TYPE_IPV6)) { | |
457 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
458 | ip_proto = nh6->ip6_ctlun.ip6_un1.ip6_un1_nxt; | |
459 | } else { | |
460 | struct ip_header *l3_hdr = dp_packet_l3(pkt); | |
461 | ip_proto = l3_hdr->ip_proto; | |
462 | } | |
286de272 | 463 | |
bd5e81a0 DB |
464 | return ip_proto; |
465 | } | |
466 | ||
467 | static bool | |
94e71143 | 468 | is_ftp_ctl(const enum ct_alg_ctl_type ct_alg_ctl) |
bd5e81a0 | 469 | { |
94e71143 | 470 | return ct_alg_ctl == CT_ALG_CTL_FTP; |
bd5e81a0 DB |
471 | } |
472 | ||
94e71143 | 473 | static enum ct_alg_ctl_type |
bd7d93f8 DB |
474 | get_alg_ctl_type(const struct dp_packet *pkt, ovs_be16 tp_src, ovs_be16 tp_dst, |
475 | const char *helper) | |
7be77cb0 | 476 | { |
94e71143 DB |
477 | /* CT_IPPORT_FTP/TFTP is used because IPPORT_FTP/TFTP in not defined |
478 | * in OSX, at least in in.h. Since these values will never change, remove | |
7be77cb0 | 479 | * the external dependency. */ |
94e71143 DB |
480 | enum { CT_IPPORT_FTP = 21 }; |
481 | enum { CT_IPPORT_TFTP = 69 }; | |
bd7d93f8 DB |
482 | uint8_t ip_proto = get_ip_proto(pkt); |
483 | struct udp_header *uh = dp_packet_l4(pkt); | |
484 | struct tcp_header *th = dp_packet_l4(pkt); | |
485 | ovs_be16 ftp_src_port = htons(CT_IPPORT_FTP); | |
486 | ovs_be16 ftp_dst_port = htons(CT_IPPORT_FTP); | |
487 | ovs_be16 tftp_dst_port = htons(CT_IPPORT_TFTP); | |
488 | ||
489 | if (OVS_UNLIKELY(tp_dst)) { | |
490 | if (helper && !strncmp(helper, "ftp", strlen("ftp"))) { | |
491 | ftp_dst_port = tp_dst; | |
492 | } else if (helper && !strncmp(helper, "tftp", strlen("tftp"))) { | |
493 | tftp_dst_port = tp_dst; | |
494 | } | |
495 | } else if (OVS_UNLIKELY(tp_src)) { | |
496 | if (helper && !strncmp(helper, "ftp", strlen("ftp"))) { | |
497 | ftp_src_port = tp_src; | |
498 | } | |
499 | } | |
7be77cb0 | 500 | |
bd7d93f8 | 501 | if (ip_proto == IPPROTO_UDP && uh->udp_dst == tftp_dst_port) { |
94e71143 DB |
502 | return CT_ALG_CTL_TFTP; |
503 | } else if (ip_proto == IPPROTO_TCP && | |
bd7d93f8 | 504 | (th->tcp_src == ftp_src_port || th->tcp_dst == ftp_dst_port)) { |
94e71143 DB |
505 | return CT_ALG_CTL_FTP; |
506 | } | |
507 | return CT_ALG_CTL_NONE; | |
508 | } | |
509 | ||
be38342d DB |
510 | static bool |
511 | alg_src_ip_wc(enum ct_alg_ctl_type alg_ctl_type) | |
512 | { | |
513 | if (alg_ctl_type == CT_ALG_CTL_SIP) { | |
514 | return true; | |
515 | } | |
516 | return false; | |
517 | } | |
518 | ||
94e71143 DB |
519 | static void |
520 | handle_alg_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
521 | struct dp_packet *pkt, enum ct_alg_ctl_type ct_alg_ctl, | |
522 | const struct conn *conn, long long now, bool nat, | |
523 | const struct conn *conn_for_expectation) | |
524 | { | |
525 | /* ALG control packet handling with expectation creation. */ | |
3a2a425b | 526 | if (OVS_UNLIKELY(alg_helpers[ct_alg_ctl] && conn && conn->alg)) { |
94e71143 DB |
527 | alg_helpers[ct_alg_ctl](ct, ctx, pkt, conn_for_expectation, now, |
528 | CT_FTP_CTL_INTEREST, nat); | |
529 | } | |
7be77cb0 DB |
530 | } |
531 | ||
286de272 DB |
532 | static void |
533 | pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
534 | { | |
535 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
536 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
537 | struct tcp_header *th = dp_packet_l4(pkt); | |
538 | packet_set_tcp_port(pkt, conn->rev_key.dst.port, th->tcp_dst); | |
539 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
540 | struct udp_header *uh = dp_packet_l4(pkt); | |
541 | packet_set_udp_port(pkt, conn->rev_key.dst.port, uh->udp_dst); | |
542 | } | |
543 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
544 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
545 | struct tcp_header *th = dp_packet_l4(pkt); | |
546 | packet_set_tcp_port(pkt, th->tcp_src, conn->rev_key.src.port); | |
547 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
548 | struct udp_header *uh = dp_packet_l4(pkt); | |
549 | packet_set_udp_port(pkt, uh->udp_src, conn->rev_key.src.port); | |
550 | } | |
551 | } | |
552 | } | |
553 | ||
554 | static void | |
555 | nat_packet(struct dp_packet *pkt, const struct conn *conn, bool related) | |
556 | { | |
557 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
558 | pkt->md.ct_state |= CS_SRC_NAT; | |
559 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
560 | struct ip_header *nh = dp_packet_l3(pkt); | |
561 | packet_set_ipv4_addr(pkt, &nh->ip_src, | |
cda1b109 | 562 | conn->rev_key.dst.addr.ipv4); |
286de272 DB |
563 | } else { |
564 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
565 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
566 | nh6->ip6_src.be32, | |
cda1b109 | 567 | &conn->rev_key.dst.addr.ipv6, true); |
286de272 DB |
568 | } |
569 | if (!related) { | |
570 | pat_packet(pkt, conn); | |
571 | } | |
572 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
573 | pkt->md.ct_state |= CS_DST_NAT; | |
574 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
575 | struct ip_header *nh = dp_packet_l3(pkt); | |
576 | packet_set_ipv4_addr(pkt, &nh->ip_dst, | |
cda1b109 | 577 | conn->rev_key.src.addr.ipv4); |
286de272 DB |
578 | } else { |
579 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
580 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
581 | nh6->ip6_dst.be32, | |
cda1b109 | 582 | &conn->rev_key.src.addr.ipv6, true); |
286de272 DB |
583 | } |
584 | if (!related) { | |
585 | pat_packet(pkt, conn); | |
586 | } | |
587 | } | |
588 | } | |
589 | ||
590 | static void | |
591 | un_pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
592 | { | |
593 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
594 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
595 | struct tcp_header *th = dp_packet_l4(pkt); | |
596 | packet_set_tcp_port(pkt, th->tcp_src, conn->key.src.port); | |
597 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
598 | struct udp_header *uh = dp_packet_l4(pkt); | |
599 | packet_set_udp_port(pkt, uh->udp_src, conn->key.src.port); | |
600 | } | |
601 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
602 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
603 | struct tcp_header *th = dp_packet_l4(pkt); | |
604 | packet_set_tcp_port(pkt, conn->key.dst.port, th->tcp_dst); | |
605 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
606 | struct udp_header *uh = dp_packet_l4(pkt); | |
607 | packet_set_udp_port(pkt, conn->key.dst.port, uh->udp_dst); | |
608 | } | |
609 | } | |
610 | } | |
611 | ||
edd1bef4 DB |
612 | static void |
613 | reverse_pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
614 | { | |
615 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
616 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
617 | struct tcp_header *th_in = dp_packet_l4(pkt); | |
618 | packet_set_tcp_port(pkt, conn->key.src.port, | |
619 | th_in->tcp_dst); | |
620 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
621 | struct udp_header *uh_in = dp_packet_l4(pkt); | |
622 | packet_set_udp_port(pkt, conn->key.src.port, | |
623 | uh_in->udp_dst); | |
624 | } | |
625 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
626 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
627 | struct tcp_header *th_in = dp_packet_l4(pkt); | |
628 | packet_set_tcp_port(pkt, th_in->tcp_src, | |
629 | conn->key.dst.port); | |
630 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
631 | struct udp_header *uh_in = dp_packet_l4(pkt); | |
632 | packet_set_udp_port(pkt, uh_in->udp_src, | |
633 | conn->key.dst.port); | |
634 | } | |
635 | } | |
636 | } | |
637 | ||
638 | static void | |
639 | reverse_nat_packet(struct dp_packet *pkt, const struct conn *conn) | |
640 | { | |
641 | char *tail = dp_packet_tail(pkt); | |
642 | char pad = dp_packet_l2_pad_size(pkt); | |
643 | struct conn_key inner_key; | |
644 | const char *inner_l4 = NULL; | |
645 | uint16_t orig_l3_ofs = pkt->l3_ofs; | |
646 | uint16_t orig_l4_ofs = pkt->l4_ofs; | |
647 | ||
648 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
649 | struct ip_header *nh = dp_packet_l3(pkt); | |
650 | struct icmp_header *icmp = dp_packet_l4(pkt); | |
651 | struct ip_header *inner_l3 = (struct ip_header *) (icmp + 1); | |
bd5e81a0 DB |
652 | extract_l3_ipv4(&inner_key, inner_l3, tail - ((char *)inner_l3) - pad, |
653 | &inner_l4, false); | |
edd1bef4 DB |
654 | pkt->l3_ofs += (char *) inner_l3 - (char *) nh; |
655 | pkt->l4_ofs += inner_l4 - (char *) icmp; | |
656 | ||
657 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
658 | packet_set_ipv4_addr(pkt, &inner_l3->ip_src, | |
cda1b109 | 659 | conn->key.src.addr.ipv4); |
edd1bef4 DB |
660 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { |
661 | packet_set_ipv4_addr(pkt, &inner_l3->ip_dst, | |
cda1b109 | 662 | conn->key.dst.addr.ipv4); |
edd1bef4 | 663 | } |
dec0dbbc | 664 | |
edd1bef4 DB |
665 | reverse_pat_packet(pkt, conn); |
666 | icmp->icmp_csum = 0; | |
667 | icmp->icmp_csum = csum(icmp, tail - (char *) icmp - pad); | |
668 | } else { | |
669 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
670 | struct icmp6_error_header *icmp6 = dp_packet_l4(pkt); | |
671 | struct ovs_16aligned_ip6_hdr *inner_l3_6 = | |
672 | (struct ovs_16aligned_ip6_hdr *) (icmp6 + 1); | |
673 | extract_l3_ipv6(&inner_key, inner_l3_6, | |
674 | tail - ((char *)inner_l3_6) - pad, | |
675 | &inner_l4); | |
676 | pkt->l3_ofs += (char *) inner_l3_6 - (char *) nh6; | |
677 | pkt->l4_ofs += inner_l4 - (char *) icmp6; | |
678 | ||
679 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
680 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
681 | inner_l3_6->ip6_src.be32, | |
cda1b109 | 682 | &conn->key.src.addr.ipv6, true); |
edd1bef4 DB |
683 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { |
684 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
685 | inner_l3_6->ip6_dst.be32, | |
cda1b109 | 686 | &conn->key.dst.addr.ipv6, true); |
edd1bef4 DB |
687 | } |
688 | reverse_pat_packet(pkt, conn); | |
edd1bef4 | 689 | icmp6->icmp6_base.icmp6_cksum = 0; |
76d85771 DB |
690 | icmp6->icmp6_base.icmp6_cksum = packet_csum_upperlayer6(nh6, icmp6, |
691 | IPPROTO_ICMPV6, tail - (char *) icmp6 - pad); | |
edd1bef4 DB |
692 | } |
693 | pkt->l3_ofs = orig_l3_ofs; | |
694 | pkt->l4_ofs = orig_l4_ofs; | |
695 | } | |
696 | ||
286de272 DB |
697 | static void |
698 | un_nat_packet(struct dp_packet *pkt, const struct conn *conn, | |
699 | bool related) | |
700 | { | |
701 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
702 | pkt->md.ct_state |= CS_DST_NAT; | |
703 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
704 | struct ip_header *nh = dp_packet_l3(pkt); | |
705 | packet_set_ipv4_addr(pkt, &nh->ip_dst, | |
cda1b109 | 706 | conn->key.src.addr.ipv4); |
286de272 DB |
707 | } else { |
708 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
709 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
710 | nh6->ip6_dst.be32, | |
cda1b109 | 711 | &conn->key.src.addr.ipv6, true); |
286de272 | 712 | } |
edd1bef4 DB |
713 | |
714 | if (OVS_UNLIKELY(related)) { | |
715 | reverse_nat_packet(pkt, conn); | |
716 | } else { | |
286de272 DB |
717 | un_pat_packet(pkt, conn); |
718 | } | |
719 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
720 | pkt->md.ct_state |= CS_SRC_NAT; | |
721 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
722 | struct ip_header *nh = dp_packet_l3(pkt); | |
723 | packet_set_ipv4_addr(pkt, &nh->ip_src, | |
cda1b109 | 724 | conn->key.dst.addr.ipv4); |
286de272 DB |
725 | } else { |
726 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
727 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
728 | nh6->ip6_src.be32, | |
cda1b109 | 729 | &conn->key.dst.addr.ipv6, true); |
286de272 | 730 | } |
edd1bef4 DB |
731 | |
732 | if (OVS_UNLIKELY(related)) { | |
733 | reverse_nat_packet(pkt, conn); | |
734 | } else { | |
286de272 DB |
735 | un_pat_packet(pkt, conn); |
736 | } | |
737 | } | |
738 | } | |
739 | ||
740 | /* Typical usage of this helper is in non per-packet code; | |
741 | * this is because the bucket lock needs to be held for lookup | |
742 | * and a hash would have already been needed. Hence, this function | |
743 | * is just intended for code clarity. */ | |
744 | static struct conn * | |
bd5e81a0 | 745 | conn_lookup(struct conntrack *ct, const struct conn_key *key, long long now) |
286de272 DB |
746 | { |
747 | struct conn_lookup_ctx ctx; | |
748 | ctx.conn = NULL; | |
c3f6bae2 | 749 | memcpy(&ctx.key, key, sizeof ctx.key); |
286de272 DB |
750 | ctx.hash = conn_key_hash(key, ct->hash_basis); |
751 | unsigned bucket = hash_to_bucket(ctx.hash); | |
752 | conn_key_lookup(&ct->buckets[bucket], &ctx, now); | |
753 | return ctx.conn; | |
754 | } | |
755 | ||
bd5e81a0 DB |
756 | static void |
757 | conn_seq_skew_set(struct conntrack *ct, const struct conn_key *key, | |
758 | long long now, int seq_skew, bool seq_skew_dir) | |
759 | { | |
dec0dbbc | 760 | unsigned bucket = hash_to_bucket(conn_key_hash(key, ct->hash_basis)); |
bd5e81a0 DB |
761 | ct_lock_lock(&ct->buckets[bucket].lock); |
762 | struct conn *conn = conn_lookup(ct, key, now); | |
763 | if (conn && seq_skew) { | |
764 | conn->seq_skew = seq_skew; | |
765 | conn->seq_skew_dir = seq_skew_dir; | |
766 | } | |
767 | ct_lock_unlock(&ct->buckets[bucket].lock); | |
768 | } | |
769 | ||
286de272 DB |
770 | static void |
771 | nat_clean(struct conntrack *ct, struct conn *conn, | |
772 | struct conntrack_bucket *ctb) | |
773 | OVS_REQUIRES(ctb->lock) | |
774 | { | |
8b934ced | 775 | ct_rwlock_wrlock(&ct->resources_lock); |
286de272 | 776 | nat_conn_keys_remove(&ct->nat_conn_keys, &conn->rev_key, ct->hash_basis); |
8b934ced | 777 | ct_rwlock_unlock(&ct->resources_lock); |
286de272 | 778 | ct_lock_unlock(&ctb->lock); |
dec0dbbc DB |
779 | unsigned bucket_rev_conn = |
780 | hash_to_bucket(conn_key_hash(&conn->rev_key, ct->hash_basis)); | |
286de272 | 781 | ct_lock_lock(&ct->buckets[bucket_rev_conn].lock); |
8b934ced | 782 | ct_rwlock_wrlock(&ct->resources_lock); |
dec0dbbc | 783 | long long now = time_msec(); |
286de272 | 784 | struct conn *rev_conn = conn_lookup(ct, &conn->rev_key, now); |
286de272 DB |
785 | struct nat_conn_key_node *nat_conn_key_node = |
786 | nat_conn_keys_lookup(&ct->nat_conn_keys, &conn->rev_key, | |
787 | ct->hash_basis); | |
788 | ||
789 | /* In the unlikely event, rev conn was recreated, then skip | |
790 | * rev_conn cleanup. */ | |
791 | if (rev_conn && (!nat_conn_key_node || | |
5ed7a0b4 DB |
792 | conn_key_cmp(&nat_conn_key_node->value, |
793 | &rev_conn->rev_key))) { | |
286de272 DB |
794 | hmap_remove(&ct->buckets[bucket_rev_conn].connections, |
795 | &rev_conn->node); | |
796 | free(rev_conn); | |
797 | } | |
286de272 | 798 | |
dec0dbbc | 799 | delete_conn(conn); |
8b934ced | 800 | ct_rwlock_unlock(&ct->resources_lock); |
286de272 DB |
801 | ct_lock_unlock(&ct->buckets[bucket_rev_conn].lock); |
802 | ct_lock_lock(&ctb->lock); | |
803 | } | |
804 | ||
9e8f3960 | 805 | /* Must be called with 'CT_CONN_TYPE_DEFAULT' 'conn_type'. */ |
286de272 DB |
806 | static void |
807 | conn_clean(struct conntrack *ct, struct conn *conn, | |
808 | struct conntrack_bucket *ctb) | |
809 | OVS_REQUIRES(ctb->lock) | |
810 | { | |
9e8f3960 DB |
811 | ovs_assert(conn->conn_type == CT_CONN_TYPE_DEFAULT); |
812 | ||
4417ca3d DB |
813 | if (conn->alg) { |
814 | expectation_clean(ct, &conn->key, ct->hash_basis); | |
815 | } | |
286de272 DB |
816 | ovs_list_remove(&conn->exp_node); |
817 | hmap_remove(&ctb->connections, &conn->node); | |
818 | atomic_count_dec(&ct->n_conn); | |
819 | if (conn->nat_info) { | |
820 | nat_clean(ct, conn, ctb); | |
821 | } else { | |
822 | delete_conn(conn); | |
823 | } | |
a489b168 DDP |
824 | } |
825 | ||
3a2a425b DB |
826 | static bool |
827 | ct_verify_helper(const char *helper, enum ct_alg_ctl_type ct_alg_ctl) | |
828 | { | |
829 | if (ct_alg_ctl == CT_ALG_CTL_NONE) { | |
830 | return true; | |
831 | } else if (helper) { | |
832 | if ((ct_alg_ctl == CT_ALG_CTL_FTP) && | |
833 | !strncmp(helper, "ftp", strlen("ftp"))) { | |
834 | return true; | |
835 | } else if ((ct_alg_ctl == CT_ALG_CTL_TFTP) && | |
836 | !strncmp(helper, "tftp", strlen("tftp"))) { | |
837 | return true; | |
838 | } else { | |
839 | return false; | |
840 | } | |
841 | } else { | |
842 | return false; | |
843 | } | |
844 | } | |
845 | ||
ac6abe5f | 846 | /* This function is called with the bucket lock held. */ |
a489b168 DDP |
847 | static struct conn * |
848 | conn_not_found(struct conntrack *ct, struct dp_packet *pkt, | |
286de272 DB |
849 | struct conn_lookup_ctx *ctx, bool commit, long long now, |
850 | const struct nat_action_info_t *nat_action_info, | |
bd5e81a0 DB |
851 | struct conn *conn_for_un_nat_copy, |
852 | const char *helper, | |
3a2a425b DB |
853 | const struct alg_exp_node *alg_exp, |
854 | enum ct_alg_ctl_type ct_alg_ctl) | |
a489b168 | 855 | { |
a489b168 DDP |
856 | struct conn *nc = NULL; |
857 | ||
858 | if (!valid_new(pkt, &ctx->key)) { | |
286de272 | 859 | pkt->md.ct_state = CS_INVALID; |
a489b168 DDP |
860 | return nc; |
861 | } | |
dec0dbbc | 862 | |
286de272 | 863 | pkt->md.ct_state = CS_NEW; |
dec0dbbc | 864 | |
bd5e81a0 DB |
865 | if (alg_exp) { |
866 | pkt->md.ct_state |= CS_RELATED; | |
867 | } | |
a489b168 DDP |
868 | |
869 | if (commit) { | |
870 | unsigned int n_conn_limit; | |
a489b168 DDP |
871 | atomic_read_relaxed(&ct->n_conn_limit, &n_conn_limit); |
872 | ||
873 | if (atomic_count_get(&ct->n_conn) >= n_conn_limit) { | |
874 | COVERAGE_INC(conntrack_full); | |
875 | return nc; | |
876 | } | |
877 | ||
dec0dbbc | 878 | unsigned bucket = hash_to_bucket(ctx->hash); |
e6ef6cc6 | 879 | nc = new_conn(&ct->buckets[bucket], pkt, &ctx->key, now); |
286de272 DB |
880 | ctx->conn = nc; |
881 | nc->rev_key = nc->key; | |
882 | conn_key_reverse(&nc->rev_key); | |
a489b168 | 883 | |
3a2a425b DB |
884 | if (ct_verify_helper(helper, ct_alg_ctl)) { |
885 | nc->alg = nullable_xstrdup(helper); | |
bd5e81a0 DB |
886 | } |
887 | ||
888 | if (alg_exp) { | |
889 | nc->alg_related = true; | |
890 | nc->mark = alg_exp->master_mark; | |
891 | nc->label = alg_exp->master_label; | |
892 | nc->master_key = alg_exp->master_key; | |
893 | } | |
894 | ||
286de272 DB |
895 | if (nat_action_info) { |
896 | nc->nat_info = xmemdup(nat_action_info, sizeof *nc->nat_info); | |
a489b168 | 897 | |
bd5e81a0 | 898 | if (alg_exp) { |
be38342d | 899 | if (alg_exp->nat_rpl_dst) { |
bd5e81a0 DB |
900 | nc->rev_key.dst.addr = alg_exp->alg_nat_repl_addr; |
901 | nc->nat_info->nat_action = NAT_ACTION_SRC; | |
902 | } else { | |
903 | nc->rev_key.src.addr = alg_exp->alg_nat_repl_addr; | |
904 | nc->nat_info->nat_action = NAT_ACTION_DST; | |
905 | } | |
c3f6bae2 | 906 | memcpy(conn_for_un_nat_copy, nc, sizeof *conn_for_un_nat_copy); |
d8682ee5 DB |
907 | ct_rwlock_wrlock(&ct->resources_lock); |
908 | bool new_insert = nat_conn_keys_insert(&ct->nat_conn_keys, | |
909 | conn_for_un_nat_copy, | |
910 | ct->hash_basis); | |
911 | ct_rwlock_unlock(&ct->resources_lock); | |
912 | if (!new_insert) { | |
913 | char *log_msg = xasprintf("Pre-existing alg " | |
914 | "nat_conn_key"); | |
915 | ct_print_conn_info(conn_for_un_nat_copy, log_msg, VLL_INFO, | |
916 | true, false); | |
917 | free(log_msg); | |
918 | } | |
bd5e81a0 | 919 | } else { |
c3f6bae2 | 920 | memcpy(conn_for_un_nat_copy, nc, sizeof *conn_for_un_nat_copy); |
bd5e81a0 | 921 | ct_rwlock_wrlock(&ct->resources_lock); |
dec0dbbc DB |
922 | bool nat_res = nat_select_range_tuple(ct, nc, |
923 | conn_for_un_nat_copy); | |
286de272 | 924 | |
bd5e81a0 DB |
925 | if (!nat_res) { |
926 | goto nat_res_exhaustion; | |
927 | } | |
286de272 | 928 | |
bd5e81a0 DB |
929 | /* Update nc with nat adjustments made to |
930 | * conn_for_un_nat_copy by nat_select_range_tuple(). */ | |
286de272 | 931 | *nc = *conn_for_un_nat_copy; |
bd5e81a0 | 932 | ct_rwlock_unlock(&ct->resources_lock); |
286de272 | 933 | } |
bd5e81a0 DB |
934 | conn_for_un_nat_copy->conn_type = CT_CONN_TYPE_UN_NAT; |
935 | conn_for_un_nat_copy->nat_info = NULL; | |
936 | conn_for_un_nat_copy->alg = NULL; | |
dbb597d3 | 937 | nat_packet(pkt, nc, ctx->icmp_related); |
286de272 | 938 | } |
a489b168 DDP |
939 | hmap_insert(&ct->buckets[bucket].connections, &nc->node, ctx->hash); |
940 | atomic_count_inc(&ct->n_conn); | |
941 | } | |
bd5e81a0 | 942 | |
a489b168 | 943 | return nc; |
bd5e81a0 DB |
944 | |
945 | /* This would be a user error or a DOS attack. | |
946 | * A user error is prevented by allocating enough | |
947 | * combinations of NAT addresses when combined with | |
948 | * ephemeral ports. A DOS attack should be protected | |
949 | * against with firewall rules or a separate firewall. | |
950 | * Also using zone partitioning can limit DoS impact. */ | |
951 | nat_res_exhaustion: | |
d8c5a93b | 952 | ovs_list_remove(&nc->exp_node); |
bd5e81a0 DB |
953 | delete_conn(nc); |
954 | /* conn_for_un_nat_copy is a local variable in process_one; this | |
955 | * memset() serves to document that conn_for_un_nat_copy is from | |
956 | * this point on unused. */ | |
957 | memset(conn_for_un_nat_copy, 0, sizeof *conn_for_un_nat_copy); | |
958 | ct_rwlock_unlock(&ct->resources_lock); | |
959 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
960 | VLOG_WARN_RL(&rl, "Unable to NAT due to tuple space exhaustion - " | |
961 | "if DoS attack, use firewalling and/or zone partitioning."); | |
962 | return NULL; | |
a489b168 DDP |
963 | } |
964 | ||
286de272 DB |
965 | static bool |
966 | conn_update_state(struct conntrack *ct, struct dp_packet *pkt, | |
967 | struct conn_lookup_ctx *ctx, struct conn **conn, | |
968 | long long now, unsigned bucket) | |
969 | OVS_REQUIRES(ct->buckets[bucket].lock) | |
970 | { | |
971 | bool create_new_conn = false; | |
972 | ||
dbb597d3 | 973 | if (ctx->icmp_related) { |
286de272 DB |
974 | pkt->md.ct_state |= CS_RELATED; |
975 | if (ctx->reply) { | |
976 | pkt->md.ct_state |= CS_REPLY_DIR; | |
977 | } | |
978 | } else { | |
bd5e81a0 DB |
979 | if ((*conn)->alg_related) { |
980 | pkt->md.ct_state |= CS_RELATED; | |
981 | } | |
dec0dbbc | 982 | |
286de272 DB |
983 | enum ct_update_res res = conn_update(*conn, &ct->buckets[bucket], |
984 | pkt, ctx->reply, now); | |
985 | ||
986 | switch (res) { | |
987 | case CT_UPDATE_VALID: | |
988 | pkt->md.ct_state |= CS_ESTABLISHED; | |
989 | pkt->md.ct_state &= ~CS_NEW; | |
990 | if (ctx->reply) { | |
991 | pkt->md.ct_state |= CS_REPLY_DIR; | |
992 | } | |
993 | break; | |
994 | case CT_UPDATE_INVALID: | |
995 | pkt->md.ct_state = CS_INVALID; | |
996 | break; | |
997 | case CT_UPDATE_NEW: | |
998 | conn_clean(ct, *conn, &ct->buckets[bucket]); | |
999 | create_new_conn = true; | |
1000 | break; | |
1001 | default: | |
1002 | OVS_NOT_REACHED(); | |
1003 | } | |
1004 | } | |
1005 | return create_new_conn; | |
1006 | } | |
1007 | ||
1008 | static void | |
1009 | create_un_nat_conn(struct conntrack *ct, struct conn *conn_for_un_nat_copy, | |
bd5e81a0 | 1010 | long long now, bool alg_un_nat) |
286de272 DB |
1011 | { |
1012 | struct conn *nc = xmemdup(conn_for_un_nat_copy, sizeof *nc); | |
1013 | nc->key = conn_for_un_nat_copy->rev_key; | |
1014 | nc->rev_key = conn_for_un_nat_copy->key; | |
1015 | uint32_t un_nat_hash = conn_key_hash(&nc->key, ct->hash_basis); | |
1016 | unsigned un_nat_conn_bucket = hash_to_bucket(un_nat_hash); | |
1017 | ct_lock_lock(&ct->buckets[un_nat_conn_bucket].lock); | |
286de272 DB |
1018 | struct conn *rev_conn = conn_lookup(ct, &nc->key, now); |
1019 | ||
bd5e81a0 | 1020 | if (alg_un_nat) { |
d8682ee5 DB |
1021 | if (!rev_conn) { |
1022 | hmap_insert(&ct->buckets[un_nat_conn_bucket].connections, | |
1023 | &nc->node, un_nat_hash); | |
1024 | } else { | |
1025 | char *log_msg = xasprintf("Unusual condition for un_nat conn " | |
1026 | "create for alg: rev_conn %p", rev_conn); | |
1027 | ct_print_conn_info(nc, log_msg, VLL_INFO, true, false); | |
1028 | free(log_msg); | |
1029 | free(nc); | |
1030 | } | |
286de272 | 1031 | } else { |
bd5e81a0 DB |
1032 | ct_rwlock_rdlock(&ct->resources_lock); |
1033 | ||
1034 | struct nat_conn_key_node *nat_conn_key_node = | |
1035 | nat_conn_keys_lookup(&ct->nat_conn_keys, &nc->key, ct->hash_basis); | |
1036 | if (nat_conn_key_node && !conn_key_cmp(&nat_conn_key_node->value, | |
1037 | &nc->rev_key) && !rev_conn) { | |
bd5e81a0 DB |
1038 | hmap_insert(&ct->buckets[un_nat_conn_bucket].connections, |
1039 | &nc->node, un_nat_hash); | |
1040 | } else { | |
d8682ee5 DB |
1041 | char *log_msg = xasprintf("Unusual condition for un_nat conn " |
1042 | "create: nat_conn_key_node/rev_conn " | |
1043 | "%p/%p", nat_conn_key_node, rev_conn); | |
1044 | ct_print_conn_info(nc, log_msg, VLL_INFO, true, false); | |
1045 | free(log_msg); | |
bd5e81a0 DB |
1046 | free(nc); |
1047 | } | |
1048 | ct_rwlock_unlock(&ct->resources_lock); | |
286de272 | 1049 | } |
286de272 DB |
1050 | ct_lock_unlock(&ct->buckets[un_nat_conn_bucket].lock); |
1051 | } | |
1052 | ||
1053 | static void | |
1054 | handle_nat(struct dp_packet *pkt, struct conn *conn, | |
1055 | uint16_t zone, bool reply, bool related) | |
1056 | { | |
1057 | if (conn->nat_info && | |
1058 | (!(pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) || | |
1059 | (pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT) && | |
1060 | zone != pkt->md.ct_zone))) { | |
bd5e81a0 | 1061 | |
286de272 DB |
1062 | if (pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) { |
1063 | pkt->md.ct_state &= ~(CS_SRC_NAT | CS_DST_NAT); | |
1064 | } | |
1065 | if (reply) { | |
1066 | un_nat_packet(pkt, conn, related); | |
1067 | } else { | |
1068 | nat_packet(pkt, conn, related); | |
1069 | } | |
1070 | } | |
1071 | } | |
1072 | ||
f8016041 DB |
1073 | static bool |
1074 | check_orig_tuple(struct conntrack *ct, struct dp_packet *pkt, | |
1075 | struct conn_lookup_ctx *ctx_in, long long now, | |
1076 | unsigned *bucket, struct conn **conn, | |
1077 | const struct nat_action_info_t *nat_action_info) | |
1078 | OVS_REQUIRES(ct->buckets[*bucket].lock) | |
1079 | { | |
1080 | if ((ctx_in->key.dl_type == htons(ETH_TYPE_IP) && | |
1081 | !pkt->md.ct_orig_tuple.ipv4.ipv4_proto) || | |
1082 | (ctx_in->key.dl_type == htons(ETH_TYPE_IPV6) && | |
1083 | !pkt->md.ct_orig_tuple.ipv6.ipv6_proto) || | |
1084 | !(pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) || | |
1085 | nat_action_info) { | |
1086 | return false; | |
1087 | } | |
1088 | ||
1089 | ct_lock_unlock(&ct->buckets[*bucket].lock); | |
1090 | struct conn_lookup_ctx ctx; | |
1091 | memset(&ctx, 0 , sizeof ctx); | |
1092 | ctx.conn = NULL; | |
1093 | ||
1094 | if (ctx_in->key.dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 DB |
1095 | ctx.key.src.addr.ipv4 = pkt->md.ct_orig_tuple.ipv4.ipv4_src; |
1096 | ctx.key.dst.addr.ipv4 = pkt->md.ct_orig_tuple.ipv4.ipv4_dst; | |
f8016041 DB |
1097 | |
1098 | if (ctx_in->key.nw_proto == IPPROTO_ICMP) { | |
1099 | ctx.key.src.icmp_id = ctx_in->key.src.icmp_id; | |
1100 | ctx.key.dst.icmp_id = ctx_in->key.dst.icmp_id; | |
1101 | uint16_t src_port = ntohs(pkt->md.ct_orig_tuple.ipv4.src_port); | |
1102 | ctx.key.src.icmp_type = (uint8_t) src_port; | |
1103 | ctx.key.dst.icmp_type = reverse_icmp_type(ctx.key.src.icmp_type); | |
1104 | } else { | |
1105 | ctx.key.src.port = pkt->md.ct_orig_tuple.ipv4.src_port; | |
1106 | ctx.key.dst.port = pkt->md.ct_orig_tuple.ipv4.dst_port; | |
1107 | } | |
1108 | ctx.key.nw_proto = pkt->md.ct_orig_tuple.ipv4.ipv4_proto; | |
1109 | } else { | |
cda1b109 DB |
1110 | ctx.key.src.addr.ipv6 = pkt->md.ct_orig_tuple.ipv6.ipv6_src; |
1111 | ctx.key.dst.addr.ipv6 = pkt->md.ct_orig_tuple.ipv6.ipv6_dst; | |
f8016041 DB |
1112 | |
1113 | if (ctx_in->key.nw_proto == IPPROTO_ICMPV6) { | |
1114 | ctx.key.src.icmp_id = ctx_in->key.src.icmp_id; | |
1115 | ctx.key.dst.icmp_id = ctx_in->key.dst.icmp_id; | |
1116 | uint16_t src_port = ntohs(pkt->md.ct_orig_tuple.ipv6.src_port); | |
1117 | ctx.key.src.icmp_type = (uint8_t) src_port; | |
1118 | ctx.key.dst.icmp_type = reverse_icmp6_type(ctx.key.src.icmp_type); | |
1119 | } else { | |
1120 | ctx.key.src.port = pkt->md.ct_orig_tuple.ipv6.src_port; | |
1121 | ctx.key.dst.port = pkt->md.ct_orig_tuple.ipv6.dst_port; | |
1122 | } | |
1123 | ctx.key.nw_proto = pkt->md.ct_orig_tuple.ipv6.ipv6_proto; | |
1124 | } | |
1125 | ||
1126 | ctx.key.dl_type = ctx_in->key.dl_type; | |
1127 | ctx.key.zone = pkt->md.ct_zone; | |
f8016041 DB |
1128 | ctx.hash = conn_key_hash(&ctx.key, ct->hash_basis); |
1129 | *bucket = hash_to_bucket(ctx.hash); | |
1130 | ct_lock_lock(&ct->buckets[*bucket].lock); | |
1131 | conn_key_lookup(&ct->buckets[*bucket], &ctx, now); | |
1132 | *conn = ctx.conn; | |
f8016041 DB |
1133 | return *conn ? true : false; |
1134 | } | |
1135 | ||
bd5e81a0 DB |
1136 | static bool |
1137 | is_un_nat_conn_valid(const struct conn *un_nat_conn) | |
1138 | { | |
1139 | return un_nat_conn->conn_type == CT_CONN_TYPE_UN_NAT; | |
1140 | } | |
1141 | ||
94e71143 DB |
1142 | static bool |
1143 | conn_update_state_alg(struct conntrack *ct, struct dp_packet *pkt, | |
1144 | struct conn_lookup_ctx *ctx, struct conn *conn, | |
1145 | const struct nat_action_info_t *nat_action_info, | |
1146 | enum ct_alg_ctl_type ct_alg_ctl, long long now, | |
1147 | unsigned bucket, bool *create_new_conn) | |
1148 | OVS_REQUIRES(ct->buckets[bucket].lock) | |
1149 | { | |
1150 | if (is_ftp_ctl(ct_alg_ctl)) { | |
1151 | /* Keep sequence tracking in sync with the source of the | |
1152 | * sequence skew. */ | |
1153 | if (ctx->reply != conn->seq_skew_dir) { | |
1154 | handle_ftp_ctl(ct, ctx, pkt, conn, now, CT_FTP_CTL_OTHER, | |
1155 | !!nat_action_info); | |
1156 | *create_new_conn = conn_update_state(ct, pkt, ctx, &conn, now, | |
1157 | bucket); | |
1158 | } else { | |
1159 | *create_new_conn = conn_update_state(ct, pkt, ctx, &conn, now, | |
1160 | bucket); | |
030958a0 DB |
1161 | |
1162 | if (*create_new_conn == false) { | |
1163 | handle_ftp_ctl(ct, ctx, pkt, conn, now, CT_FTP_CTL_OTHER, | |
1164 | !!nat_action_info); | |
1165 | } | |
94e71143 DB |
1166 | } |
1167 | return true; | |
1168 | } | |
1169 | return false; | |
1170 | } | |
1171 | ||
286de272 | 1172 | static void |
a489b168 DDP |
1173 | process_one(struct conntrack *ct, struct dp_packet *pkt, |
1174 | struct conn_lookup_ctx *ctx, uint16_t zone, | |
286de272 DB |
1175 | bool force, bool commit, long long now, const uint32_t *setmark, |
1176 | const struct ovs_key_ct_labels *setlabel, | |
bd5e81a0 | 1177 | const struct nat_action_info_t *nat_action_info, |
bd7d93f8 | 1178 | ovs_be16 tp_src, ovs_be16 tp_dst, const char *helper) |
a489b168 | 1179 | { |
286de272 | 1180 | struct conn *conn; |
a489b168 | 1181 | unsigned bucket = hash_to_bucket(ctx->hash); |
286de272 DB |
1182 | ct_lock_lock(&ct->buckets[bucket].lock); |
1183 | conn_key_lookup(&ct->buckets[bucket], ctx, now); | |
1184 | conn = ctx->conn; | |
a489b168 | 1185 | |
a76a37ef JR |
1186 | /* Delete found entry if in wrong direction. 'force' implies commit. */ |
1187 | if (conn && force && ctx->reply) { | |
286de272 | 1188 | conn_clean(ct, conn, &ct->buckets[bucket]); |
a76a37ef JR |
1189 | conn = NULL; |
1190 | } | |
1191 | ||
286de272 DB |
1192 | if (OVS_LIKELY(conn)) { |
1193 | if (conn->conn_type == CT_CONN_TYPE_UN_NAT) { | |
a489b168 | 1194 | |
286de272 | 1195 | ctx->reply = true; |
a489b168 | 1196 | |
286de272 DB |
1197 | struct conn_lookup_ctx ctx2; |
1198 | ctx2.conn = NULL; | |
1199 | ctx2.key = conn->rev_key; | |
1200 | ctx2.hash = conn_key_hash(&conn->rev_key, ct->hash_basis); | |
1201 | ||
1202 | ct_lock_unlock(&ct->buckets[bucket].lock); | |
1203 | bucket = hash_to_bucket(ctx2.hash); | |
1204 | ||
1205 | ct_lock_lock(&ct->buckets[bucket].lock); | |
1206 | conn_key_lookup(&ct->buckets[bucket], &ctx2, now); | |
1207 | ||
1208 | if (ctx2.conn) { | |
1209 | conn = ctx2.conn; | |
1210 | } else { | |
1211 | /* It is a race condition where conn has timed out and removed | |
1212 | * between unlock of the rev_conn and lock of the forward conn; | |
1213 | * nothing to do. */ | |
1214 | pkt->md.ct_state |= CS_TRACKED | CS_INVALID; | |
1215 | ct_lock_unlock(&ct->buckets[bucket].lock); | |
1216 | return; | |
a489b168 DDP |
1217 | } |
1218 | } | |
286de272 DB |
1219 | } |
1220 | ||
1221 | bool create_new_conn = false; | |
1222 | struct conn conn_for_un_nat_copy; | |
1223 | conn_for_un_nat_copy.conn_type = CT_CONN_TYPE_DEFAULT; | |
94e71143 | 1224 | |
bd7d93f8 DB |
1225 | enum ct_alg_ctl_type ct_alg_ctl = get_alg_ctl_type(pkt, tp_src, tp_dst, |
1226 | helper); | |
bd5e81a0 | 1227 | |
286de272 | 1228 | if (OVS_LIKELY(conn)) { |
94e71143 DB |
1229 | if (OVS_LIKELY(!conn_update_state_alg(ct, pkt, ctx, conn, |
1230 | nat_action_info, | |
1231 | ct_alg_ctl, now, bucket, | |
1232 | &create_new_conn))) { | |
bd5e81a0 DB |
1233 | create_new_conn = conn_update_state(ct, pkt, ctx, &conn, now, |
1234 | bucket); | |
1235 | } | |
286de272 | 1236 | if (nat_action_info && !create_new_conn) { |
dbb597d3 | 1237 | handle_nat(pkt, conn, zone, ctx->reply, ctx->icmp_related); |
286de272 | 1238 | } |
bd5e81a0 | 1239 | |
dec0dbbc | 1240 | } else if (check_orig_tuple(ct, pkt, ctx, now, &bucket, &conn, |
bd5e81a0 | 1241 | nat_action_info)) { |
dec0dbbc | 1242 | create_new_conn = conn_update_state(ct, pkt, ctx, &conn, now, bucket); |
a489b168 | 1243 | } else { |
dbb597d3 | 1244 | if (ctx->icmp_related) { |
bd5e81a0 DB |
1245 | /* An icmp related conn should always be found; no new |
1246 | connection is created based on an icmp related packet. */ | |
286de272 | 1247 | pkt->md.ct_state = CS_INVALID; |
5c2e106b | 1248 | } else { |
286de272 | 1249 | create_new_conn = true; |
5c2e106b | 1250 | } |
a489b168 DDP |
1251 | } |
1252 | ||
bd5e81a0 | 1253 | const struct alg_exp_node *alg_exp = NULL; |
96bbcbf7 | 1254 | struct alg_exp_node alg_exp_entry; |
dec0dbbc | 1255 | |
286de272 | 1256 | if (OVS_UNLIKELY(create_new_conn)) { |
bd5e81a0 DB |
1257 | |
1258 | ct_rwlock_rdlock(&ct->resources_lock); | |
1259 | alg_exp = expectation_lookup(&ct->alg_expectations, &ctx->key, | |
be38342d DB |
1260 | ct->hash_basis, |
1261 | alg_src_ip_wc(ct_alg_ctl)); | |
bd5e81a0 | 1262 | if (alg_exp) { |
c3f6bae2 | 1263 | memcpy(&alg_exp_entry, alg_exp, sizeof alg_exp_entry); |
bd5e81a0 DB |
1264 | alg_exp = &alg_exp_entry; |
1265 | } | |
1266 | ct_rwlock_unlock(&ct->resources_lock); | |
1267 | ||
286de272 | 1268 | conn = conn_not_found(ct, pkt, ctx, commit, now, nat_action_info, |
3a2a425b DB |
1269 | &conn_for_un_nat_copy, helper, alg_exp, |
1270 | ct_alg_ctl); | |
286de272 DB |
1271 | } |
1272 | ||
bd5e81a0 DB |
1273 | write_ct_md(pkt, zone, conn, &ctx->key, alg_exp); |
1274 | ||
286de272 DB |
1275 | if (conn && setmark) { |
1276 | set_mark(pkt, conn, setmark[0], setmark[1]); | |
1277 | } | |
a489b168 | 1278 | |
286de272 DB |
1279 | if (conn && setlabel) { |
1280 | set_label(pkt, conn, &setlabel[0], &setlabel[1]); | |
1281 | } | |
1282 | ||
bd5e81a0 | 1283 | struct conn conn_for_expectation; |
94e71143 | 1284 | if (OVS_UNLIKELY((ct_alg_ctl != CT_ALG_CTL_NONE) && conn)) { |
bd5e81a0 DB |
1285 | conn_for_expectation = *conn; |
1286 | } | |
1287 | ||
286de272 DB |
1288 | ct_lock_unlock(&ct->buckets[bucket].lock); |
1289 | ||
bd5e81a0 DB |
1290 | if (is_un_nat_conn_valid(&conn_for_un_nat_copy)) { |
1291 | create_un_nat_conn(ct, &conn_for_un_nat_copy, now, !!alg_exp); | |
1292 | } | |
1293 | ||
94e71143 DB |
1294 | handle_alg_ctl(ct, ctx, pkt, ct_alg_ctl, conn, now, !!nat_action_info, |
1295 | &conn_for_expectation); | |
a489b168 DDP |
1296 | } |
1297 | ||
1298 | /* Sends the packets in '*pkt_batch' through the connection tracker 'ct'. All | |
51b9a533 | 1299 | * the packets must have the same 'dl_type' (IPv4 or IPv6) and should have |
4ea96698 DB |
1300 | * the l3 and and l4 offset properly set. Performs fragment reassembly with |
1301 | * the help of ipf_preprocess_conntrack(). | |
a489b168 DDP |
1302 | * |
1303 | * If 'commit' is true, the packets are allowed to create new entries in the | |
1304 | * connection tables. 'setmark', if not NULL, should point to a two | |
1305 | * elements array containing a value and a mask to set the connection mark. | |
1306 | * 'setlabel' behaves similarly for the connection label.*/ | |
1307 | int | |
1308 | conntrack_execute(struct conntrack *ct, struct dp_packet_batch *pkt_batch, | |
a76a37ef | 1309 | ovs_be16 dl_type, bool force, bool commit, uint16_t zone, |
66e4ad8a | 1310 | const uint32_t *setmark, |
a489b168 | 1311 | const struct ovs_key_ct_labels *setlabel, |
bd7d93f8 | 1312 | ovs_be16 tp_src, ovs_be16 tp_dst, const char *helper, |
94053e66 FA |
1313 | const struct nat_action_info_t *nat_action_info, |
1314 | long long now) | |
a489b168 | 1315 | { |
4ea96698 DB |
1316 | ipf_preprocess_conntrack(ct->ipf, pkt_batch, now, dl_type, zone, |
1317 | ct->hash_basis); | |
1318 | ||
43495c45 | 1319 | struct dp_packet *packet; |
61ce32b9 | 1320 | struct conn_lookup_ctx ctx; |
a489b168 | 1321 | |
e883448e | 1322 | DP_PACKET_BATCH_FOR_EACH (i, packet, pkt_batch) { |
4ea96698 DB |
1323 | if (packet->md.ct_state == CS_INVALID |
1324 | || !conn_key_extract(ct, packet, dl_type, &ctx, zone)) { | |
43495c45 BB |
1325 | packet->md.ct_state = CS_INVALID; |
1326 | write_ct_md(packet, zone, NULL, NULL, NULL); | |
a489b168 DDP |
1327 | continue; |
1328 | } | |
94e71143 | 1329 | process_one(ct, packet, &ctx, zone, force, commit, now, setmark, |
bd7d93f8 | 1330 | setlabel, nat_action_info, tp_src, tp_dst, helper); |
a489b168 DDP |
1331 | } |
1332 | ||
4ea96698 DB |
1333 | ipf_postprocess_conntrack(ct->ipf, pkt_batch, now, dl_type); |
1334 | ||
a489b168 DDP |
1335 | return 0; |
1336 | } | |
1337 | ||
1fe178d2 EG |
1338 | void |
1339 | conntrack_clear(struct dp_packet *packet) | |
1340 | { | |
1341 | /* According to pkt_metadata_init(), ct_state == 0 is enough to make all of | |
1342 | * the conntrack fields invalid. */ | |
1343 | packet->md.ct_state = 0; | |
1344 | } | |
1345 | ||
a489b168 DDP |
1346 | static void |
1347 | set_mark(struct dp_packet *pkt, struct conn *conn, uint32_t val, uint32_t mask) | |
1348 | { | |
bd5e81a0 DB |
1349 | if (conn->alg_related) { |
1350 | pkt->md.ct_mark = conn->mark; | |
1351 | } else { | |
1352 | pkt->md.ct_mark = val | (pkt->md.ct_mark & ~(mask)); | |
1353 | conn->mark = pkt->md.ct_mark; | |
1354 | } | |
a489b168 DDP |
1355 | } |
1356 | ||
1357 | static void | |
1358 | set_label(struct dp_packet *pkt, struct conn *conn, | |
1359 | const struct ovs_key_ct_labels *val, | |
1360 | const struct ovs_key_ct_labels *mask) | |
1361 | { | |
bd5e81a0 DB |
1362 | if (conn->alg_related) { |
1363 | pkt->md.ct_label = conn->label; | |
1364 | } else { | |
1365 | ovs_u128 v, m; | |
a489b168 | 1366 | |
bd5e81a0 DB |
1367 | memcpy(&v, val, sizeof v); |
1368 | memcpy(&m, mask, sizeof m); | |
a489b168 | 1369 | |
bd5e81a0 | 1370 | pkt->md.ct_label.u64.lo = v.u64.lo |
a489b168 | 1371 | | (pkt->md.ct_label.u64.lo & ~(m.u64.lo)); |
bd5e81a0 | 1372 | pkt->md.ct_label.u64.hi = v.u64.hi |
a489b168 | 1373 | | (pkt->md.ct_label.u64.hi & ~(m.u64.hi)); |
bd5e81a0 DB |
1374 | conn->label = pkt->md.ct_label; |
1375 | } | |
a489b168 | 1376 | } |
286de272 | 1377 | |
a489b168 | 1378 | \f |
e6ef6cc6 DDP |
1379 | /* Delete the expired connections from 'ctb', up to 'limit'. Returns the |
1380 | * earliest expiration time among the remaining connections in 'ctb'. Returns | |
1381 | * LLONG_MAX if 'ctb' is empty. The return value might be smaller than 'now', | |
1382 | * if 'limit' is reached */ | |
1383 | static long long | |
bd5e81a0 DB |
1384 | sweep_bucket(struct conntrack *ct, struct conntrack_bucket *ctb, |
1385 | long long now, size_t limit) | |
e6ef6cc6 DDP |
1386 | OVS_REQUIRES(ctb->lock) |
1387 | { | |
1388 | struct conn *conn, *next; | |
1389 | long long min_expiration = LLONG_MAX; | |
e6ef6cc6 DDP |
1390 | size_t count = 0; |
1391 | ||
dec0dbbc | 1392 | for (unsigned i = 0; i < N_CT_TM; i++) { |
e6ef6cc6 | 1393 | LIST_FOR_EACH_SAFE (conn, next, exp_node, &ctb->exp_lists[i]) { |
286de272 DB |
1394 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
1395 | if (!conn_expired(conn, now) || count >= limit) { | |
1396 | min_expiration = MIN(min_expiration, conn->expiration); | |
1397 | if (count >= limit) { | |
1398 | /* Do not check other lists. */ | |
1399 | COVERAGE_INC(conntrack_long_cleanup); | |
1400 | return min_expiration; | |
1401 | } | |
1402 | break; | |
e6ef6cc6 | 1403 | } |
286de272 DB |
1404 | conn_clean(ct, conn, ctb); |
1405 | count++; | |
e6ef6cc6 | 1406 | } |
e6ef6cc6 DDP |
1407 | } |
1408 | } | |
e6ef6cc6 DDP |
1409 | return min_expiration; |
1410 | } | |
1411 | ||
1412 | /* Cleans up old connection entries from 'ct'. Returns the time when the | |
1413 | * next expiration might happen. The return value might be smaller than | |
1414 | * 'now', meaning that an internal limit has been reached, and some expired | |
1415 | * connections have not been deleted. */ | |
1416 | static long long | |
1417 | conntrack_clean(struct conntrack *ct, long long now) | |
1418 | { | |
1419 | long long next_wakeup = now + CT_TM_MIN; | |
1420 | unsigned int n_conn_limit; | |
1421 | size_t clean_count = 0; | |
e6ef6cc6 DDP |
1422 | |
1423 | atomic_read_relaxed(&ct->n_conn_limit, &n_conn_limit); | |
1424 | ||
dec0dbbc | 1425 | for (unsigned i = 0; i < CONNTRACK_BUCKETS; i++) { |
e6ef6cc6 DDP |
1426 | struct conntrack_bucket *ctb = &ct->buckets[i]; |
1427 | size_t prev_count; | |
1428 | long long min_exp; | |
1429 | ||
1430 | ovs_mutex_lock(&ctb->cleanup_mutex); | |
1431 | if (ctb->next_cleanup > now) { | |
1432 | goto next_bucket; | |
1433 | } | |
1434 | ||
1435 | ct_lock_lock(&ctb->lock); | |
1436 | prev_count = hmap_count(&ctb->connections); | |
1437 | /* If the connections are well distributed among buckets, we want to | |
1438 | * limit to 10% of the global limit equally split among buckets. If | |
1439 | * the bucket is busier than the others, we limit to 10% of its | |
1440 | * current size. */ | |
1441 | min_exp = sweep_bucket(ct, ctb, now, | |
1442 | MAX(prev_count/10, n_conn_limit/(CONNTRACK_BUCKETS*10))); | |
1443 | clean_count += prev_count - hmap_count(&ctb->connections); | |
1444 | ||
1445 | if (min_exp > now) { | |
1446 | /* We call hmap_shrink() only if sweep_bucket() managed to delete | |
1447 | * every expired connection. */ | |
1448 | hmap_shrink(&ctb->connections); | |
1449 | } | |
1450 | ||
1451 | ct_lock_unlock(&ctb->lock); | |
1452 | ||
1453 | ctb->next_cleanup = MIN(min_exp, now + CT_TM_MIN); | |
1454 | ||
1455 | next_bucket: | |
1456 | next_wakeup = MIN(next_wakeup, ctb->next_cleanup); | |
1457 | ovs_mutex_unlock(&ctb->cleanup_mutex); | |
1458 | } | |
1459 | ||
1460 | VLOG_DBG("conntrack cleanup %"PRIuSIZE" entries in %lld msec", | |
1461 | clean_count, time_msec() - now); | |
1462 | ||
1463 | return next_wakeup; | |
1464 | } | |
1465 | ||
1466 | /* Cleanup: | |
e6ef6cc6 DDP |
1467 | * |
1468 | * We must call conntrack_clean() periodically. conntrack_clean() return | |
1469 | * value gives an hint on when the next cleanup must be done (either because | |
1470 | * there is an actual connection that expires, or because a new connection | |
1471 | * might be created with the minimum timeout). | |
1472 | * | |
1473 | * The logic below has two goals: | |
1474 | * | |
6c54734e DDP |
1475 | * - We want to reduce the number of wakeups and batch connection cleanup |
1476 | * when the load is not very high. CT_CLEAN_INTERVAL ensures that if we | |
1477 | * are coping with the current cleanup tasks, then we wait at least | |
1478 | * 5 seconds to do further cleanup. | |
e6ef6cc6 | 1479 | * |
6c54734e DDP |
1480 | * - We don't want to keep the buckets locked too long, as we might prevent |
1481 | * traffic from flowing. CT_CLEAN_MIN_INTERVAL ensures that if cleanup is | |
1482 | * behind, there is at least some 200ms blocks of time when buckets will be | |
1483 | * left alone, so the datapath can operate unhindered. | |
e6ef6cc6 DDP |
1484 | */ |
1485 | #define CT_CLEAN_INTERVAL 5000 /* 5 seconds */ | |
1486 | #define CT_CLEAN_MIN_INTERVAL 200 /* 0.2 seconds */ | |
1487 | ||
1488 | static void * | |
1489 | clean_thread_main(void *f_) | |
1490 | { | |
1491 | struct conntrack *ct = f_; | |
1492 | ||
1493 | while (!latch_is_set(&ct->clean_thread_exit)) { | |
1494 | long long next_wake; | |
1495 | long long now = time_msec(); | |
e6ef6cc6 DDP |
1496 | next_wake = conntrack_clean(ct, now); |
1497 | ||
1498 | if (next_wake < now) { | |
1499 | poll_timer_wait_until(now + CT_CLEAN_MIN_INTERVAL); | |
1500 | } else { | |
1501 | poll_timer_wait_until(MAX(next_wake, now + CT_CLEAN_INTERVAL)); | |
1502 | } | |
1503 | latch_wait(&ct->clean_thread_exit); | |
1504 | poll_block(); | |
1505 | } | |
1506 | ||
1507 | return NULL; | |
1508 | } | |
1509 | \f | |
e917d3ee DB |
1510 | /* 'Data' is a pointer to the beginning of the L3 header and 'new_data' is |
1511 | * used to store a pointer to the first byte after the L3 header. 'Size' is | |
1512 | * the size of the packet beyond the data pointer. */ | |
a489b168 DDP |
1513 | static inline bool |
1514 | extract_l3_ipv4(struct conn_key *key, const void *data, size_t size, | |
1515 | const char **new_data, bool validate_checksum) | |
1516 | { | |
e917d3ee DB |
1517 | if (OVS_UNLIKELY(size < IP_HEADER_LEN)) { |
1518 | return false; | |
a489b168 DDP |
1519 | } |
1520 | ||
dec0dbbc DB |
1521 | const struct ip_header *ip = data; |
1522 | size_t ip_len = IP_IHL(ip->ip_ihl_ver) * 4; | |
a489b168 | 1523 | |
e917d3ee DB |
1524 | if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) { |
1525 | return false; | |
1526 | } | |
a489b168 | 1527 | |
e917d3ee DB |
1528 | if (OVS_UNLIKELY(size < ip_len)) { |
1529 | return false; | |
1530 | } | |
a489b168 | 1531 | |
e917d3ee DB |
1532 | if (IP_IS_FRAGMENT(ip->ip_frag_off)) { |
1533 | return false; | |
a489b168 DDP |
1534 | } |
1535 | ||
1536 | if (validate_checksum && csum(data, ip_len) != 0) { | |
1537 | return false; | |
1538 | } | |
1539 | ||
e917d3ee DB |
1540 | if (new_data) { |
1541 | *new_data = (char *) data + ip_len; | |
1542 | } | |
1543 | ||
cda1b109 DB |
1544 | key->src.addr.ipv4 = get_16aligned_be32(&ip->ip_src); |
1545 | key->dst.addr.ipv4 = get_16aligned_be32(&ip->ip_dst); | |
a489b168 DDP |
1546 | key->nw_proto = ip->ip_proto; |
1547 | ||
1548 | return true; | |
1549 | } | |
1550 | ||
e917d3ee DB |
1551 | /* 'Data' is a pointer to the beginning of the L3 header and 'new_data' is |
1552 | * used to store a pointer to the first byte after the L3 header. 'Size' is | |
1553 | * the size of the packet beyond the data pointer. */ | |
a489b168 DDP |
1554 | static inline bool |
1555 | extract_l3_ipv6(struct conn_key *key, const void *data, size_t size, | |
1556 | const char **new_data) | |
1557 | { | |
1558 | const struct ovs_16aligned_ip6_hdr *ip6 = data; | |
286de272 | 1559 | |
e917d3ee DB |
1560 | if (OVS_UNLIKELY(size < sizeof *ip6)) { |
1561 | return false; | |
a489b168 DDP |
1562 | } |
1563 | ||
1564 | data = ip6 + 1; | |
1565 | size -= sizeof *ip6; | |
dec0dbbc DB |
1566 | uint8_t nw_proto = ip6->ip6_nxt; |
1567 | uint8_t nw_frag = 0; | |
a489b168 | 1568 | |
523464ab DB |
1569 | const struct ovs_16aligned_ip6_frag *frag_hdr; |
1570 | if (!parse_ipv6_ext_hdrs(&data, &size, &nw_proto, &nw_frag, &frag_hdr)) { | |
a489b168 DDP |
1571 | return false; |
1572 | } | |
1573 | ||
a489b168 DDP |
1574 | if (nw_frag) { |
1575 | return false; | |
1576 | } | |
1577 | ||
c8b1ad49 DB |
1578 | if (new_data) { |
1579 | *new_data = data; | |
1580 | } | |
1581 | ||
cda1b109 DB |
1582 | memcpy(&key->src.addr.ipv6, &ip6->ip6_src, sizeof key->src.addr); |
1583 | memcpy(&key->dst.addr.ipv6, &ip6->ip6_dst, sizeof key->dst.addr); | |
a489b168 DDP |
1584 | key->nw_proto = nw_proto; |
1585 | ||
1586 | return true; | |
1587 | } | |
1588 | ||
1589 | static inline bool | |
1590 | checksum_valid(const struct conn_key *key, const void *data, size_t size, | |
1591 | const void *l3) | |
1592 | { | |
a489b168 | 1593 | if (key->dl_type == htons(ETH_TYPE_IP)) { |
76d85771 DB |
1594 | uint32_t csum = packet_csum_pseudoheader(l3); |
1595 | return csum_finish(csum_continue(csum, data, size)) == 0; | |
a489b168 | 1596 | } else if (key->dl_type == htons(ETH_TYPE_IPV6)) { |
76d85771 | 1597 | return packet_csum_upperlayer6(l3, data, key->nw_proto, size) == 0; |
a489b168 DDP |
1598 | } else { |
1599 | return false; | |
1600 | } | |
a489b168 DDP |
1601 | } |
1602 | ||
1603 | static inline bool | |
1604 | check_l4_tcp(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1605 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1606 | { |
1607 | const struct tcp_header *tcp = data; | |
40225b0c BP |
1608 | if (size < sizeof *tcp) { |
1609 | return false; | |
1610 | } | |
a489b168 | 1611 | |
40225b0c | 1612 | size_t tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4; |
a489b168 DDP |
1613 | if (OVS_UNLIKELY(tcp_len < TCP_HEADER_LEN || tcp_len > size)) { |
1614 | return false; | |
1615 | } | |
1616 | ||
324459a3 | 1617 | return validate_checksum ? checksum_valid(key, data, size, l3) : true; |
a489b168 DDP |
1618 | } |
1619 | ||
1620 | static inline bool | |
1621 | check_l4_udp(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1622 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1623 | { |
1624 | const struct udp_header *udp = data; | |
40225b0c BP |
1625 | if (size < sizeof *udp) { |
1626 | return false; | |
1627 | } | |
a489b168 | 1628 | |
40225b0c | 1629 | size_t udp_len = ntohs(udp->udp_len); |
a489b168 DDP |
1630 | if (OVS_UNLIKELY(udp_len < UDP_HEADER_LEN || udp_len > size)) { |
1631 | return false; | |
1632 | } | |
1633 | ||
1634 | /* Validation must be skipped if checksum is 0 on IPv4 packets */ | |
1635 | return (udp->udp_csum == 0 && key->dl_type == htons(ETH_TYPE_IP)) | |
324459a3 | 1636 | || (validate_checksum ? checksum_valid(key, data, size, l3) : true); |
a489b168 DDP |
1637 | } |
1638 | ||
1639 | static inline bool | |
324459a3 | 1640 | check_l4_icmp(const void *data, size_t size, bool validate_checksum) |
a489b168 | 1641 | { |
324459a3 | 1642 | return validate_checksum ? csum(data, size) == 0 : true; |
a489b168 DDP |
1643 | } |
1644 | ||
1645 | static inline bool | |
1646 | check_l4_icmp6(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1647 | const void *l3, bool validate_checksum) |
a489b168 | 1648 | { |
324459a3 | 1649 | return validate_checksum ? checksum_valid(key, data, size, l3) : true; |
a489b168 DDP |
1650 | } |
1651 | ||
1652 | static inline bool | |
1653 | extract_l4_tcp(struct conn_key *key, const void *data, size_t size) | |
1654 | { | |
a489b168 DDP |
1655 | if (OVS_UNLIKELY(size < TCP_HEADER_LEN)) { |
1656 | return false; | |
1657 | } | |
1658 | ||
dec0dbbc | 1659 | const struct tcp_header *tcp = data; |
a489b168 DDP |
1660 | key->src.port = tcp->tcp_src; |
1661 | key->dst.port = tcp->tcp_dst; | |
1662 | ||
1663 | /* Port 0 is invalid */ | |
1664 | return key->src.port && key->dst.port; | |
1665 | } | |
1666 | ||
1667 | static inline bool | |
1668 | extract_l4_udp(struct conn_key *key, const void *data, size_t size) | |
1669 | { | |
a489b168 DDP |
1670 | if (OVS_UNLIKELY(size < UDP_HEADER_LEN)) { |
1671 | return false; | |
1672 | } | |
1673 | ||
dec0dbbc | 1674 | const struct udp_header *udp = data; |
a489b168 DDP |
1675 | key->src.port = udp->udp_src; |
1676 | key->dst.port = udp->udp_dst; | |
1677 | ||
1678 | /* Port 0 is invalid */ | |
1679 | return key->src.port && key->dst.port; | |
1680 | } | |
1681 | ||
1682 | static inline bool extract_l4(struct conn_key *key, const void *data, | |
324459a3 SC |
1683 | size_t size, bool *related, const void *l3, |
1684 | bool validate_checksum); | |
a489b168 | 1685 | |
b269a122 DDP |
1686 | static uint8_t |
1687 | reverse_icmp_type(uint8_t type) | |
1688 | { | |
1689 | switch (type) { | |
1690 | case ICMP4_ECHO_REQUEST: | |
1691 | return ICMP4_ECHO_REPLY; | |
1692 | case ICMP4_ECHO_REPLY: | |
1693 | return ICMP4_ECHO_REQUEST; | |
1694 | ||
1695 | case ICMP4_TIMESTAMP: | |
1696 | return ICMP4_TIMESTAMPREPLY; | |
1697 | case ICMP4_TIMESTAMPREPLY: | |
1698 | return ICMP4_TIMESTAMP; | |
1699 | ||
1700 | case ICMP4_INFOREQUEST: | |
1701 | return ICMP4_INFOREPLY; | |
1702 | case ICMP4_INFOREPLY: | |
1703 | return ICMP4_INFOREQUEST; | |
1704 | default: | |
1705 | OVS_NOT_REACHED(); | |
1706 | } | |
1707 | } | |
1708 | ||
a489b168 DDP |
1709 | /* If 'related' is not NULL and the function is processing an ICMP |
1710 | * error packet, extract the l3 and l4 fields from the nested header | |
1711 | * instead and set *related to true. If 'related' is NULL we're | |
1712 | * already processing a nested header and no such recursion is | |
1713 | * possible */ | |
1714 | static inline int | |
1715 | extract_l4_icmp(struct conn_key *key, const void *data, size_t size, | |
1716 | bool *related) | |
1717 | { | |
a489b168 DDP |
1718 | if (OVS_UNLIKELY(size < ICMP_HEADER_LEN)) { |
1719 | return false; | |
1720 | } | |
1721 | ||
dec0dbbc DB |
1722 | const struct icmp_header *icmp = data; |
1723 | ||
a489b168 DDP |
1724 | switch (icmp->icmp_type) { |
1725 | case ICMP4_ECHO_REQUEST: | |
1726 | case ICMP4_ECHO_REPLY: | |
1727 | case ICMP4_TIMESTAMP: | |
1728 | case ICMP4_TIMESTAMPREPLY: | |
1729 | case ICMP4_INFOREQUEST: | |
1730 | case ICMP4_INFOREPLY: | |
b269a122 DDP |
1731 | if (icmp->icmp_code != 0) { |
1732 | return false; | |
1733 | } | |
a489b168 | 1734 | /* Separate ICMP connection: identified using id */ |
b269a122 DDP |
1735 | key->src.icmp_id = key->dst.icmp_id = icmp->icmp_fields.echo.id; |
1736 | key->src.icmp_type = icmp->icmp_type; | |
1737 | key->dst.icmp_type = reverse_icmp_type(icmp->icmp_type); | |
a489b168 DDP |
1738 | break; |
1739 | case ICMP4_DST_UNREACH: | |
1740 | case ICMP4_TIME_EXCEEDED: | |
1741 | case ICMP4_PARAM_PROB: | |
1742 | case ICMP4_SOURCEQUENCH: | |
1743 | case ICMP4_REDIRECT: { | |
1744 | /* ICMP packet part of another connection. We should | |
1745 | * extract the key from embedded packet header */ | |
1746 | struct conn_key inner_key; | |
1747 | const char *l3 = (const char *) (icmp + 1); | |
1748 | const char *tail = (const char *) data + size; | |
1749 | const char *l4; | |
a489b168 DDP |
1750 | |
1751 | if (!related) { | |
1752 | return false; | |
1753 | } | |
1754 | ||
1755 | memset(&inner_key, 0, sizeof inner_key); | |
1756 | inner_key.dl_type = htons(ETH_TYPE_IP); | |
dec0dbbc | 1757 | bool ok = extract_l3_ipv4(&inner_key, l3, tail - l3, &l4, false); |
a489b168 DDP |
1758 | if (!ok) { |
1759 | return false; | |
1760 | } | |
1761 | ||
cda1b109 | 1762 | if (inner_key.src.addr.ipv4 != key->dst.addr.ipv4) { |
a489b168 DDP |
1763 | return false; |
1764 | } | |
1765 | ||
1766 | key->src = inner_key.src; | |
1767 | key->dst = inner_key.dst; | |
1768 | key->nw_proto = inner_key.nw_proto; | |
1769 | ||
324459a3 | 1770 | ok = extract_l4(key, l4, tail - l4, NULL, l3, false); |
a489b168 DDP |
1771 | if (ok) { |
1772 | conn_key_reverse(key); | |
1773 | *related = true; | |
1774 | } | |
1775 | return ok; | |
1776 | } | |
1777 | default: | |
1778 | return false; | |
1779 | } | |
1780 | ||
1781 | return true; | |
1782 | } | |
1783 | ||
b269a122 DDP |
1784 | static uint8_t |
1785 | reverse_icmp6_type(uint8_t type) | |
1786 | { | |
1787 | switch (type) { | |
1788 | case ICMP6_ECHO_REQUEST: | |
1789 | return ICMP6_ECHO_REPLY; | |
1790 | case ICMP6_ECHO_REPLY: | |
1791 | return ICMP6_ECHO_REQUEST; | |
1792 | default: | |
1793 | OVS_NOT_REACHED(); | |
1794 | } | |
1795 | } | |
1796 | ||
a489b168 DDP |
1797 | /* If 'related' is not NULL and the function is processing an ICMP |
1798 | * error packet, extract the l3 and l4 fields from the nested header | |
1799 | * instead and set *related to true. If 'related' is NULL we're | |
1800 | * already processing a nested header and no such recursion is | |
1801 | * possible */ | |
1802 | static inline bool | |
1803 | extract_l4_icmp6(struct conn_key *key, const void *data, size_t size, | |
1804 | bool *related) | |
1805 | { | |
1806 | const struct icmp6_header *icmp6 = data; | |
1807 | ||
1808 | /* All the messages that we support need at least 4 bytes after | |
1809 | * the header */ | |
1810 | if (size < sizeof *icmp6 + 4) { | |
1811 | return false; | |
1812 | } | |
1813 | ||
1814 | switch (icmp6->icmp6_type) { | |
1815 | case ICMP6_ECHO_REQUEST: | |
1816 | case ICMP6_ECHO_REPLY: | |
b269a122 DDP |
1817 | if (icmp6->icmp6_code != 0) { |
1818 | return false; | |
1819 | } | |
a489b168 | 1820 | /* Separate ICMP connection: identified using id */ |
b269a122 DDP |
1821 | key->src.icmp_id = key->dst.icmp_id = *(ovs_be16 *) (icmp6 + 1); |
1822 | key->src.icmp_type = icmp6->icmp6_type; | |
1823 | key->dst.icmp_type = reverse_icmp6_type(icmp6->icmp6_type); | |
a489b168 DDP |
1824 | break; |
1825 | case ICMP6_DST_UNREACH: | |
1826 | case ICMP6_PACKET_TOO_BIG: | |
1827 | case ICMP6_TIME_EXCEEDED: | |
1828 | case ICMP6_PARAM_PROB: { | |
1829 | /* ICMP packet part of another connection. We should | |
1830 | * extract the key from embedded packet header */ | |
1831 | struct conn_key inner_key; | |
1832 | const char *l3 = (const char *) icmp6 + 8; | |
1833 | const char *tail = (const char *) data + size; | |
1834 | const char *l4 = NULL; | |
a489b168 DDP |
1835 | |
1836 | if (!related) { | |
1837 | return false; | |
1838 | } | |
1839 | ||
1840 | memset(&inner_key, 0, sizeof inner_key); | |
1841 | inner_key.dl_type = htons(ETH_TYPE_IPV6); | |
dec0dbbc | 1842 | bool ok = extract_l3_ipv6(&inner_key, l3, tail - l3, &l4); |
a489b168 DDP |
1843 | if (!ok) { |
1844 | return false; | |
1845 | } | |
1846 | ||
1847 | /* pf doesn't do this, but it seems a good idea */ | |
cda1b109 DB |
1848 | if (!ipv6_addr_equals(&inner_key.src.addr.ipv6, |
1849 | &key->dst.addr.ipv6)) { | |
a489b168 DDP |
1850 | return false; |
1851 | } | |
1852 | ||
1853 | key->src = inner_key.src; | |
1854 | key->dst = inner_key.dst; | |
1855 | key->nw_proto = inner_key.nw_proto; | |
1856 | ||
324459a3 | 1857 | ok = extract_l4(key, l4, tail - l4, NULL, l3, false); |
a489b168 DDP |
1858 | if (ok) { |
1859 | conn_key_reverse(key); | |
1860 | *related = true; | |
1861 | } | |
1862 | return ok; | |
1863 | } | |
1864 | default: | |
1865 | return false; | |
1866 | } | |
1867 | ||
1868 | return true; | |
1869 | } | |
1870 | ||
1871 | /* Extract l4 fields into 'key', which must already contain valid l3 | |
1872 | * members. | |
1873 | * | |
1874 | * If 'related' is not NULL and an ICMP error packet is being | |
1875 | * processed, the function will extract the key from the packet nested | |
1401f6de | 1876 | * in the ICMP payload and set '*related' to true. |
a489b168 | 1877 | * |
9171c635 DB |
1878 | * 'size' here is the layer 4 size, which can be a nested size if parsing |
1879 | * an ICMP or ICMP6 header. | |
1880 | * | |
a489b168 DDP |
1881 | * If 'related' is NULL, it means that we're already parsing a header nested |
1882 | * in an ICMP error. In this case, we skip checksum and length validation. */ | |
1883 | static inline bool | |
1884 | extract_l4(struct conn_key *key, const void *data, size_t size, bool *related, | |
324459a3 | 1885 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1886 | { |
1887 | if (key->nw_proto == IPPROTO_TCP) { | |
324459a3 SC |
1888 | return (!related || check_l4_tcp(key, data, size, l3, |
1889 | validate_checksum)) && extract_l4_tcp(key, data, size); | |
a489b168 | 1890 | } else if (key->nw_proto == IPPROTO_UDP) { |
324459a3 SC |
1891 | return (!related || check_l4_udp(key, data, size, l3, |
1892 | validate_checksum)) && extract_l4_udp(key, data, size); | |
a489b168 DDP |
1893 | } else if (key->dl_type == htons(ETH_TYPE_IP) |
1894 | && key->nw_proto == IPPROTO_ICMP) { | |
324459a3 | 1895 | return (!related || check_l4_icmp(data, size, validate_checksum)) |
a489b168 DDP |
1896 | && extract_l4_icmp(key, data, size, related); |
1897 | } else if (key->dl_type == htons(ETH_TYPE_IPV6) | |
1898 | && key->nw_proto == IPPROTO_ICMPV6) { | |
324459a3 SC |
1899 | return (!related || check_l4_icmp6(key, data, size, l3, |
1900 | validate_checksum)) && extract_l4_icmp6(key, data, size, | |
1901 | related); | |
a489b168 DDP |
1902 | } else { |
1903 | return false; | |
1904 | } | |
1905 | } | |
1906 | ||
1907 | static bool | |
66e4ad8a | 1908 | conn_key_extract(struct conntrack *ct, struct dp_packet *pkt, ovs_be16 dl_type, |
a489b168 DDP |
1909 | struct conn_lookup_ctx *ctx, uint16_t zone) |
1910 | { | |
2482b0b0 | 1911 | const struct eth_header *l2 = dp_packet_eth(pkt); |
a489b168 DDP |
1912 | const struct ip_header *l3 = dp_packet_l3(pkt); |
1913 | const char *l4 = dp_packet_l4(pkt); | |
a489b168 DDP |
1914 | |
1915 | memset(ctx, 0, sizeof *ctx); | |
1916 | ||
1917 | if (!l2 || !l3 || !l4) { | |
1918 | return false; | |
1919 | } | |
1920 | ||
1921 | ctx->key.zone = zone; | |
1922 | ||
1923 | /* XXX In this function we parse the packet (again, it has already | |
1924 | * gone through miniflow_extract()) for two reasons: | |
1925 | * | |
1926 | * 1) To extract the l3 addresses and l4 ports. | |
1927 | * We already have the l3 and l4 headers' pointers. Extracting | |
1928 | * the l3 addresses and the l4 ports is really cheap, since they | |
1929 | * can be found at fixed locations. | |
66e4ad8a DDP |
1930 | * 2) To extract the l4 type. |
1931 | * Extracting the l4 types, for IPv6 can be quite expensive, because | |
1932 | * it's not at a fixed location. | |
a489b168 DDP |
1933 | * |
1934 | * Here's a way to avoid (2) with the help of the datapath. | |
66e4ad8a | 1935 | * The datapath doesn't keep the packet's extracted flow[1], so |
a489b168 | 1936 | * using that is not an option. We could use the packet's matching |
66e4ad8a DDP |
1937 | * megaflow, but we have to make sure that the l4 type (nw_proto) |
1938 | * is unwildcarded. This means either: | |
a489b168 | 1939 | * |
66e4ad8a DDP |
1940 | * a) dpif-netdev unwildcards the l4 type when a new flow is installed |
1941 | * if the actions contains ct(). | |
a489b168 | 1942 | * |
66e4ad8a DDP |
1943 | * b) ofproto-dpif-xlate unwildcards the l4 type when translating a ct() |
1944 | * action. This is already done in different actions, but it's | |
1945 | * unnecessary for the kernel. | |
a489b168 DDP |
1946 | * |
1947 | * --- | |
66e4ad8a | 1948 | * [1] The reasons for this are that keeping the flow increases |
a489b168 DDP |
1949 | * (slightly) the cache footprint and increases computation |
1950 | * time as we move the packet around. Most importantly, the flow | |
1951 | * should be updated by the actions and this can be slow, as | |
1952 | * we use a sparse representation (miniflow). | |
1953 | * | |
1954 | */ | |
dec0dbbc | 1955 | bool ok; |
66e4ad8a | 1956 | ctx->key.dl_type = dl_type; |
dec0dbbc | 1957 | |
a489b168 | 1958 | if (ctx->key.dl_type == htons(ETH_TYPE_IP)) { |
dec0dbbc | 1959 | bool hwol_bad_l3_csum = dp_packet_ip_checksum_bad(pkt); |
324459a3 SC |
1960 | if (hwol_bad_l3_csum) { |
1961 | ok = false; | |
1962 | } else { | |
dec0dbbc | 1963 | bool hwol_good_l3_csum = dp_packet_ip_checksum_valid(pkt); |
324459a3 | 1964 | /* Validate the checksum only when hwol is not supported. */ |
9171c635 | 1965 | ok = extract_l3_ipv4(&ctx->key, l3, dp_packet_l3_size(pkt), NULL, |
324459a3 SC |
1966 | !hwol_good_l3_csum); |
1967 | } | |
a489b168 | 1968 | } else if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
9171c635 | 1969 | ok = extract_l3_ipv6(&ctx->key, l3, dp_packet_l3_size(pkt), NULL); |
a489b168 DDP |
1970 | } else { |
1971 | ok = false; | |
1972 | } | |
1973 | ||
1974 | if (ok) { | |
324459a3 SC |
1975 | bool hwol_bad_l4_csum = dp_packet_l4_checksum_bad(pkt); |
1976 | if (!hwol_bad_l4_csum) { | |
1977 | bool hwol_good_l4_csum = dp_packet_l4_checksum_valid(pkt); | |
1978 | /* Validate the checksum only when hwol is not supported. */ | |
9171c635 DB |
1979 | if (extract_l4(&ctx->key, l4, dp_packet_l4_size(pkt), |
1980 | &ctx->icmp_related, l3, !hwol_good_l4_csum)) { | |
324459a3 SC |
1981 | ctx->hash = conn_key_hash(&ctx->key, ct->hash_basis); |
1982 | return true; | |
1983 | } | |
a489b168 DDP |
1984 | } |
1985 | } | |
1986 | ||
1987 | return false; | |
1988 | } | |
92edd073 DB |
1989 | |
1990 | static uint32_t | |
cda1b109 | 1991 | ct_addr_hash_add(uint32_t hash, const union ct_addr *addr) |
92edd073 DB |
1992 | { |
1993 | BUILD_ASSERT_DECL(sizeof *addr % 4 == 0); | |
1994 | return hash_add_bytes32(hash, (const uint32_t *) addr, sizeof *addr); | |
1995 | } | |
1996 | ||
1997 | static uint32_t | |
1998 | ct_endpoint_hash_add(uint32_t hash, const struct ct_endpoint *ep) | |
1999 | { | |
2000 | BUILD_ASSERT_DECL(sizeof *ep % 4 == 0); | |
2001 | return hash_add_bytes32(hash, (const uint32_t *) ep, sizeof *ep); | |
2002 | } | |
a489b168 DDP |
2003 | \f |
2004 | /* Symmetric */ | |
2005 | static uint32_t | |
2006 | conn_key_hash(const struct conn_key *key, uint32_t basis) | |
2007 | { | |
2008 | uint32_t hsrc, hdst, hash; | |
a489b168 | 2009 | hsrc = hdst = basis; |
6b1d4625 DB |
2010 | hsrc = ct_endpoint_hash_add(hsrc, &key->src); |
2011 | hdst = ct_endpoint_hash_add(hdst, &key->dst); | |
a489b168 DDP |
2012 | |
2013 | /* Even if source and destination are swapped the hash will be the same. */ | |
2014 | hash = hsrc ^ hdst; | |
2015 | ||
2016 | /* Hash the rest of the key(L3 and L4 types and zone). */ | |
763b40b0 | 2017 | return hash_words((uint32_t *) (&key->dst + 1), |
a489b168 DDP |
2018 | (uint32_t *) (key + 1) - (uint32_t *) (&key->dst + 1), |
2019 | hash); | |
a489b168 DDP |
2020 | } |
2021 | ||
2022 | static void | |
2023 | conn_key_reverse(struct conn_key *key) | |
2024 | { | |
dec0dbbc | 2025 | struct ct_endpoint tmp = key->src; |
a489b168 DDP |
2026 | key->src = key->dst; |
2027 | key->dst = tmp; | |
2028 | } | |
2029 | ||
286de272 | 2030 | static uint32_t |
cda1b109 | 2031 | nat_ipv6_addrs_delta(struct in6_addr *ipv6_min, struct in6_addr *ipv6_max) |
286de272 | 2032 | { |
cda1b109 DB |
2033 | uint8_t *ipv6_min_hi = &ipv6_min->s6_addr[0]; |
2034 | uint8_t *ipv6_min_lo = &ipv6_min->s6_addr[0] + sizeof(uint64_t); | |
2035 | uint8_t *ipv6_max_hi = &ipv6_max->s6_addr[0]; | |
2036 | uint8_t *ipv6_max_lo = &ipv6_max->s6_addr[0] + sizeof(uint64_t); | |
286de272 DB |
2037 | |
2038 | ovs_be64 addr6_64_min_hi; | |
2039 | ovs_be64 addr6_64_min_lo; | |
2040 | memcpy(&addr6_64_min_hi, ipv6_min_hi, sizeof addr6_64_min_hi); | |
2041 | memcpy(&addr6_64_min_lo, ipv6_min_lo, sizeof addr6_64_min_lo); | |
2042 | ||
2043 | ovs_be64 addr6_64_max_hi; | |
2044 | ovs_be64 addr6_64_max_lo; | |
2045 | memcpy(&addr6_64_max_hi, ipv6_max_hi, sizeof addr6_64_max_hi); | |
2046 | memcpy(&addr6_64_max_lo, ipv6_max_lo, sizeof addr6_64_max_lo); | |
2047 | ||
2048 | uint64_t diff; | |
dec0dbbc | 2049 | |
286de272 DB |
2050 | if (addr6_64_min_hi == addr6_64_max_hi && |
2051 | ntohll(addr6_64_min_lo) <= ntohll(addr6_64_max_lo)) { | |
2052 | diff = ntohll(addr6_64_max_lo) - ntohll(addr6_64_min_lo); | |
2053 | } else if (ntohll(addr6_64_min_hi) + 1 == ntohll(addr6_64_max_hi) && | |
2054 | ntohll(addr6_64_min_lo) > ntohll(addr6_64_max_lo)) { | |
2055 | diff = UINT64_MAX - (ntohll(addr6_64_min_lo) - | |
2056 | ntohll(addr6_64_max_lo) - 1); | |
2057 | } else { | |
2058 | /* Limit address delta supported to 32 bits or 4 billion approximately. | |
2059 | * Possibly, this should be visible to the user through a datapath | |
2060 | * support check, however the practical impact is probably nil. */ | |
2061 | diff = 0xfffffffe; | |
2062 | } | |
dec0dbbc | 2063 | |
286de272 DB |
2064 | if (diff > 0xfffffffe) { |
2065 | diff = 0xfffffffe; | |
2066 | } | |
2067 | return diff; | |
2068 | } | |
2069 | ||
2070 | /* This function must be used in tandem with nat_ipv6_addrs_delta(), which | |
2071 | * restricts the input parameters. */ | |
a489b168 | 2072 | static void |
cda1b109 | 2073 | nat_ipv6_addr_increment(struct in6_addr *ipv6, uint32_t increment) |
286de272 | 2074 | { |
cda1b109 DB |
2075 | uint8_t *ipv6_hi = &ipv6->s6_addr[0]; |
2076 | uint8_t *ipv6_lo = &ipv6->s6_addr[0] + sizeof(ovs_be64); | |
286de272 DB |
2077 | ovs_be64 addr6_64_hi; |
2078 | ovs_be64 addr6_64_lo; | |
2079 | memcpy(&addr6_64_hi, ipv6_hi, sizeof addr6_64_hi); | |
2080 | memcpy(&addr6_64_lo, ipv6_lo, sizeof addr6_64_lo); | |
2081 | ||
2082 | if (UINT64_MAX - increment >= ntohll(addr6_64_lo)) { | |
2083 | addr6_64_lo = htonll(increment + ntohll(addr6_64_lo)); | |
2084 | } else if (addr6_64_hi != OVS_BE64_MAX) { | |
2085 | addr6_64_hi = htonll(1 + ntohll(addr6_64_hi)); | |
2086 | addr6_64_lo = htonll(increment - (UINT64_MAX - | |
2087 | ntohll(addr6_64_lo) + 1)); | |
2088 | } else { | |
2089 | OVS_NOT_REACHED(); | |
2090 | } | |
2091 | ||
2092 | memcpy(ipv6_hi, &addr6_64_hi, sizeof addr6_64_hi); | |
2093 | memcpy(ipv6_lo, &addr6_64_lo, sizeof addr6_64_lo); | |
286de272 DB |
2094 | } |
2095 | ||
2096 | static uint32_t | |
2097 | nat_range_hash(const struct conn *conn, uint32_t basis) | |
2098 | { | |
2099 | uint32_t hash = basis; | |
286de272 | 2100 | |
92edd073 DB |
2101 | hash = ct_addr_hash_add(hash, &conn->nat_info->min_addr); |
2102 | hash = ct_addr_hash_add(hash, &conn->nat_info->max_addr); | |
2103 | hash = hash_add(hash, | |
2104 | (conn->nat_info->max_port << 16) | |
2105 | | conn->nat_info->min_port); | |
92edd073 DB |
2106 | hash = ct_endpoint_hash_add(hash, &conn->key.src); |
2107 | hash = ct_endpoint_hash_add(hash, &conn->key.dst); | |
286de272 DB |
2108 | hash = hash_add(hash, (OVS_FORCE uint32_t) conn->key.dl_type); |
2109 | hash = hash_add(hash, conn->key.nw_proto); | |
2110 | hash = hash_add(hash, conn->key.zone); | |
92edd073 DB |
2111 | |
2112 | /* The purpose of the second parameter is to distinguish hashes of data of | |
2113 | * different length; our data always has the same length so there is no | |
2114 | * value in counting. */ | |
2115 | return hash_finish(hash, 0); | |
286de272 DB |
2116 | } |
2117 | ||
2118 | static bool | |
2119 | nat_select_range_tuple(struct conntrack *ct, const struct conn *conn, | |
2120 | struct conn *nat_conn) | |
2121 | { | |
bd5e81a0 DB |
2122 | enum { MIN_NAT_EPHEMERAL_PORT = 1024, |
2123 | MAX_NAT_EPHEMERAL_PORT = 65535 }; | |
286de272 DB |
2124 | |
2125 | uint16_t min_port; | |
2126 | uint16_t max_port; | |
2127 | uint16_t first_port; | |
286de272 DB |
2128 | uint32_t hash = nat_range_hash(conn, ct->hash_basis); |
2129 | ||
2130 | if ((conn->nat_info->nat_action & NAT_ACTION_SRC) && | |
2131 | (!(conn->nat_info->nat_action & NAT_ACTION_SRC_PORT))) { | |
2132 | min_port = ntohs(conn->key.src.port); | |
2133 | max_port = ntohs(conn->key.src.port); | |
2134 | first_port = min_port; | |
2135 | } else if ((conn->nat_info->nat_action & NAT_ACTION_DST) && | |
2136 | (!(conn->nat_info->nat_action & NAT_ACTION_DST_PORT))) { | |
2137 | min_port = ntohs(conn->key.dst.port); | |
2138 | max_port = ntohs(conn->key.dst.port); | |
2139 | first_port = min_port; | |
2140 | } else { | |
2141 | uint16_t deltap = conn->nat_info->max_port - conn->nat_info->min_port; | |
2142 | uint32_t port_index = hash % (deltap + 1); | |
2143 | first_port = conn->nat_info->min_port + port_index; | |
2144 | min_port = conn->nat_info->min_port; | |
2145 | max_port = conn->nat_info->max_port; | |
2146 | } | |
2147 | ||
2148 | uint32_t deltaa = 0; | |
2149 | uint32_t address_index; | |
cda1b109 | 2150 | union ct_addr ct_addr; |
286de272 | 2151 | memset(&ct_addr, 0, sizeof ct_addr); |
cda1b109 | 2152 | union ct_addr max_ct_addr; |
286de272 DB |
2153 | memset(&max_ct_addr, 0, sizeof max_ct_addr); |
2154 | max_ct_addr = conn->nat_info->max_addr; | |
2155 | ||
2156 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 DB |
2157 | deltaa = ntohl(conn->nat_info->max_addr.ipv4) - |
2158 | ntohl(conn->nat_info->min_addr.ipv4); | |
286de272 | 2159 | address_index = hash % (deltaa + 1); |
cda1b109 DB |
2160 | ct_addr.ipv4 = htonl( |
2161 | ntohl(conn->nat_info->min_addr.ipv4) + address_index); | |
286de272 | 2162 | } else { |
cda1b109 DB |
2163 | deltaa = nat_ipv6_addrs_delta(&conn->nat_info->min_addr.ipv6, |
2164 | &conn->nat_info->max_addr.ipv6); | |
286de272 DB |
2165 | /* deltaa must be within 32 bits for full hash coverage. A 64 or |
2166 | * 128 bit hash is unnecessary and hence not used here. Most code | |
2167 | * is kept common with V4; nat_ipv6_addrs_delta() will do the | |
2168 | * enforcement via max_ct_addr. */ | |
2169 | max_ct_addr = conn->nat_info->min_addr; | |
cda1b109 | 2170 | nat_ipv6_addr_increment(&max_ct_addr.ipv6, deltaa); |
286de272 | 2171 | address_index = hash % (deltaa + 1); |
cda1b109 DB |
2172 | ct_addr.ipv6 = conn->nat_info->min_addr.ipv6; |
2173 | nat_ipv6_addr_increment(&ct_addr.ipv6, address_index); | |
286de272 DB |
2174 | } |
2175 | ||
2176 | uint16_t port = first_port; | |
2177 | bool all_ports_tried = false; | |
32b2c81f DB |
2178 | /* For DNAT or for specified port ranges, we don't use ephemeral ports. */ |
2179 | bool ephemeral_ports_tried | |
2180 | = conn->nat_info->nat_action & NAT_ACTION_DST || | |
2181 | conn->nat_info->nat_action & NAT_ACTION_SRC_PORT | |
2182 | ? true : false; | |
cda1b109 | 2183 | union ct_addr first_addr = ct_addr; |
4cd0481c DB |
2184 | bool pat_enabled = conn->key.nw_proto != IPPROTO_ICMP && |
2185 | conn->key.nw_proto != IPPROTO_ICMPV6; | |
286de272 DB |
2186 | |
2187 | while (true) { | |
2188 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
2189 | nat_conn->rev_key.dst.addr = ct_addr; | |
286de272 DB |
2190 | nat_conn->rev_key.dst.port = htons(port); |
2191 | } else { | |
1c8689d7 | 2192 | nat_conn->rev_key.src.addr = ct_addr; |
286de272 DB |
2193 | nat_conn->rev_key.src.port = htons(port); |
2194 | } | |
2195 | ||
80cee116 DB |
2196 | bool new_insert = nat_conn_keys_insert(&ct->nat_conn_keys, nat_conn, |
2197 | ct->hash_basis); | |
2198 | if (new_insert) { | |
286de272 | 2199 | return true; |
4cd0481c | 2200 | } else if (pat_enabled && !all_ports_tried) { |
286de272 DB |
2201 | if (min_port == max_port) { |
2202 | all_ports_tried = true; | |
2203 | } else if (port == max_port) { | |
2204 | port = min_port; | |
2205 | } else { | |
2206 | port++; | |
2207 | } | |
2208 | if (port == first_port) { | |
2209 | all_ports_tried = true; | |
2210 | } | |
2211 | } else { | |
2212 | if (memcmp(&ct_addr, &max_ct_addr, sizeof ct_addr)) { | |
2213 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2214 | ct_addr.ipv4 = htonl(ntohl(ct_addr.ipv4) + 1); |
286de272 | 2215 | } else { |
cda1b109 | 2216 | nat_ipv6_addr_increment(&ct_addr.ipv6, 1); |
286de272 DB |
2217 | } |
2218 | } else { | |
2219 | ct_addr = conn->nat_info->min_addr; | |
2220 | } | |
2221 | if (!memcmp(&ct_addr, &first_addr, sizeof ct_addr)) { | |
4cd0481c | 2222 | if (pat_enabled && !ephemeral_ports_tried) { |
ac04639a | 2223 | ephemeral_ports_tried = true; |
286de272 | 2224 | ct_addr = conn->nat_info->min_addr; |
8417e688 | 2225 | first_addr = ct_addr; |
286de272 DB |
2226 | min_port = MIN_NAT_EPHEMERAL_PORT; |
2227 | max_port = MAX_NAT_EPHEMERAL_PORT; | |
2228 | } else { | |
2229 | break; | |
2230 | } | |
2231 | } | |
2232 | first_port = min_port; | |
2233 | port = first_port; | |
2234 | all_ports_tried = false; | |
2235 | } | |
2236 | } | |
2237 | return false; | |
2238 | } | |
2239 | ||
ac6abe5f | 2240 | /* This function must be called with the ct->resources lock taken. */ |
286de272 DB |
2241 | static struct nat_conn_key_node * |
2242 | nat_conn_keys_lookup(struct hmap *nat_conn_keys, | |
2243 | const struct conn_key *key, | |
2244 | uint32_t basis) | |
2245 | { | |
2246 | struct nat_conn_key_node *nat_conn_key_node; | |
286de272 | 2247 | |
dec0dbbc DB |
2248 | HMAP_FOR_EACH_WITH_HASH (nat_conn_key_node, node, |
2249 | conn_key_hash(key, basis), nat_conn_keys) { | |
5ed7a0b4 | 2250 | if (!conn_key_cmp(&nat_conn_key_node->key, key)) { |
286de272 DB |
2251 | return nat_conn_key_node; |
2252 | } | |
2253 | } | |
2254 | return NULL; | |
2255 | } | |
2256 | ||
80cee116 DB |
2257 | /* This function must be called with the ct->resources lock taken. */ |
2258 | static bool | |
2259 | nat_conn_keys_insert(struct hmap *nat_conn_keys, const struct conn *nat_conn, | |
2260 | uint32_t basis) | |
2261 | { | |
2262 | struct nat_conn_key_node *nat_conn_key_node = | |
2263 | nat_conn_keys_lookup(nat_conn_keys, &nat_conn->rev_key, basis); | |
2264 | ||
2265 | if (!nat_conn_key_node) { | |
2266 | struct nat_conn_key_node *nat_conn_key = xzalloc(sizeof *nat_conn_key); | |
2267 | nat_conn_key->key = nat_conn->rev_key; | |
2268 | nat_conn_key->value = nat_conn->key; | |
dec0dbbc DB |
2269 | hmap_insert(nat_conn_keys, &nat_conn_key->node, |
2270 | conn_key_hash(&nat_conn_key->key, basis)); | |
80cee116 DB |
2271 | return true; |
2272 | } | |
2273 | return false; | |
2274 | } | |
2275 | ||
ac6abe5f | 2276 | /* This function must be called with the ct->resources write lock taken. */ |
286de272 | 2277 | static void |
bd5e81a0 DB |
2278 | nat_conn_keys_remove(struct hmap *nat_conn_keys, |
2279 | const struct conn_key *key, | |
286de272 DB |
2280 | uint32_t basis) |
2281 | { | |
2282 | struct nat_conn_key_node *nat_conn_key_node; | |
286de272 | 2283 | |
dec0dbbc DB |
2284 | HMAP_FOR_EACH_WITH_HASH (nat_conn_key_node, node, |
2285 | conn_key_hash(key, basis), nat_conn_keys) { | |
5ed7a0b4 | 2286 | if (!conn_key_cmp(&nat_conn_key_node->key, key)) { |
286de272 DB |
2287 | hmap_remove(nat_conn_keys, &nat_conn_key_node->node); |
2288 | free(nat_conn_key_node); | |
2289 | return; | |
2290 | } | |
2291 | } | |
2292 | } | |
2293 | ||
2294 | static void | |
2295 | conn_key_lookup(struct conntrack_bucket *ctb, struct conn_lookup_ctx *ctx, | |
a489b168 | 2296 | long long now) |
ac6abe5f | 2297 | OVS_REQUIRES(ctb->lock) |
a489b168 DDP |
2298 | { |
2299 | uint32_t hash = ctx->hash; | |
2300 | struct conn *conn; | |
2301 | ||
2302 | ctx->conn = NULL; | |
2303 | ||
2304 | HMAP_FOR_EACH_WITH_HASH (conn, node, hash, &ctb->connections) { | |
5ed7a0b4 | 2305 | if (!conn_key_cmp(&conn->key, &ctx->key) |
a489b168 DDP |
2306 | && !conn_expired(conn, now)) { |
2307 | ctx->conn = conn; | |
2308 | ctx->reply = false; | |
2309 | break; | |
2310 | } | |
5ed7a0b4 | 2311 | if (!conn_key_cmp(&conn->rev_key, &ctx->key) |
a489b168 DDP |
2312 | && !conn_expired(conn, now)) { |
2313 | ctx->conn = conn; | |
2314 | ctx->reply = true; | |
2315 | break; | |
2316 | } | |
2317 | } | |
2318 | } | |
2319 | ||
2320 | static enum ct_update_res | |
e6ef6cc6 DDP |
2321 | conn_update(struct conn *conn, struct conntrack_bucket *ctb, |
2322 | struct dp_packet *pkt, bool reply, long long now) | |
a489b168 | 2323 | { |
e6ef6cc6 DDP |
2324 | return l4_protos[conn->key.nw_proto]->conn_update(conn, ctb, pkt, |
2325 | reply, now); | |
a489b168 DDP |
2326 | } |
2327 | ||
2328 | static bool | |
2329 | conn_expired(struct conn *conn, long long now) | |
2330 | { | |
286de272 DB |
2331 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
2332 | return now >= conn->expiration; | |
2333 | } | |
2334 | return false; | |
a489b168 DDP |
2335 | } |
2336 | ||
2337 | static bool | |
2338 | valid_new(struct dp_packet *pkt, struct conn_key *key) | |
2339 | { | |
2340 | return l4_protos[key->nw_proto]->valid_new(pkt); | |
2341 | } | |
2342 | ||
2343 | static struct conn * | |
e6ef6cc6 DDP |
2344 | new_conn(struct conntrack_bucket *ctb, struct dp_packet *pkt, |
2345 | struct conn_key *key, long long now) | |
a489b168 | 2346 | { |
dec0dbbc | 2347 | struct conn *newconn = l4_protos[key->nw_proto]->new_conn(ctb, pkt, now); |
a489b168 DDP |
2348 | if (newconn) { |
2349 | newconn->key = *key; | |
2350 | } | |
2351 | ||
2352 | return newconn; | |
2353 | } | |
2354 | ||
2355 | static void | |
2356 | delete_conn(struct conn *conn) | |
2357 | { | |
286de272 | 2358 | free(conn->nat_info); |
bd5e81a0 | 2359 | free(conn->alg); |
a489b168 DDP |
2360 | free(conn); |
2361 | } | |
4d4e68ed | 2362 | \f |
271e48a0 YHW |
2363 | /* Convert a conntrack address 'a' into an IP address 'b' based on 'dl_type'. |
2364 | * | |
2365 | * Note that 'dl_type' should be either "ETH_TYPE_IP" or "ETH_TYPE_IPv6" | |
2366 | * in network-byte order. */ | |
4d4e68ed | 2367 | static void |
cda1b109 | 2368 | ct_endpoint_to_ct_dpif_inet_addr(const union ct_addr *a, |
4d4e68ed DDP |
2369 | union ct_dpif_inet_addr *b, |
2370 | ovs_be16 dl_type) | |
2371 | { | |
2372 | if (dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2373 | b->ip = a->ipv4; |
4d4e68ed | 2374 | } else if (dl_type == htons(ETH_TYPE_IPV6)){ |
cda1b109 | 2375 | b->in6 = a->ipv6; |
4d4e68ed DDP |
2376 | } |
2377 | } | |
2378 | ||
271e48a0 YHW |
2379 | /* Convert an IP address 'a' into a conntrack address 'b' based on 'dl_type'. |
2380 | * | |
2381 | * Note that 'dl_type' should be either "ETH_TYPE_IP" or "ETH_TYPE_IPv6" | |
2382 | * in network-byte order. */ | |
2383 | static void | |
2384 | ct_dpif_inet_addr_to_ct_endpoint(const union ct_dpif_inet_addr *a, | |
cda1b109 | 2385 | union ct_addr *b, ovs_be16 dl_type) |
271e48a0 YHW |
2386 | { |
2387 | if (dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2388 | b->ipv4 = a->ip; |
271e48a0 | 2389 | } else if (dl_type == htons(ETH_TYPE_IPV6)){ |
cda1b109 | 2390 | b->ipv6 = a->in6; |
271e48a0 YHW |
2391 | } |
2392 | } | |
2393 | ||
4d4e68ed DDP |
2394 | static void |
2395 | conn_key_to_tuple(const struct conn_key *key, struct ct_dpif_tuple *tuple) | |
2396 | { | |
2397 | if (key->dl_type == htons(ETH_TYPE_IP)) { | |
2398 | tuple->l3_type = AF_INET; | |
2399 | } else if (key->dl_type == htons(ETH_TYPE_IPV6)) { | |
2400 | tuple->l3_type = AF_INET6; | |
2401 | } | |
2402 | tuple->ip_proto = key->nw_proto; | |
2403 | ct_endpoint_to_ct_dpif_inet_addr(&key->src.addr, &tuple->src, | |
2404 | key->dl_type); | |
2405 | ct_endpoint_to_ct_dpif_inet_addr(&key->dst.addr, &tuple->dst, | |
2406 | key->dl_type); | |
2407 | ||
2408 | if (key->nw_proto == IPPROTO_ICMP || key->nw_proto == IPPROTO_ICMPV6) { | |
b269a122 DDP |
2409 | tuple->icmp_id = key->src.icmp_id; |
2410 | tuple->icmp_type = key->src.icmp_type; | |
2411 | tuple->icmp_code = key->src.icmp_code; | |
4d4e68ed DDP |
2412 | } else { |
2413 | tuple->src_port = key->src.port; | |
2414 | tuple->dst_port = key->dst.port; | |
2415 | } | |
2416 | } | |
2417 | ||
271e48a0 YHW |
2418 | static void |
2419 | tuple_to_conn_key(const struct ct_dpif_tuple *tuple, uint16_t zone, | |
2420 | struct conn_key *key) | |
2421 | { | |
2422 | if (tuple->l3_type == AF_INET) { | |
2423 | key->dl_type = htons(ETH_TYPE_IP); | |
2424 | } else if (tuple->l3_type == AF_INET6) { | |
2425 | key->dl_type = htons(ETH_TYPE_IPV6); | |
2426 | } | |
2427 | key->nw_proto = tuple->ip_proto; | |
2428 | ct_dpif_inet_addr_to_ct_endpoint(&tuple->src, &key->src.addr, | |
2429 | key->dl_type); | |
2430 | ct_dpif_inet_addr_to_ct_endpoint(&tuple->dst, &key->dst.addr, | |
2431 | key->dl_type); | |
2432 | ||
2433 | if (tuple->ip_proto == IPPROTO_ICMP || tuple->ip_proto == IPPROTO_ICMPV6) { | |
2434 | key->src.icmp_id = tuple->icmp_id; | |
2435 | key->src.icmp_type = tuple->icmp_type; | |
2436 | key->src.icmp_code = tuple->icmp_code; | |
2437 | key->dst.icmp_id = tuple->icmp_id; | |
2438 | key->dst.icmp_type = reverse_icmp_type(tuple->icmp_type); | |
2439 | key->dst.icmp_code = tuple->icmp_code; | |
2440 | } else { | |
2441 | key->src.port = tuple->src_port; | |
2442 | key->dst.port = tuple->dst_port; | |
2443 | } | |
2444 | key->zone = zone; | |
2445 | } | |
2446 | ||
4d4e68ed DDP |
2447 | static void |
2448 | conn_to_ct_dpif_entry(const struct conn *conn, struct ct_dpif_entry *entry, | |
ded30c74 | 2449 | long long now, int bkt) |
4d4e68ed | 2450 | { |
4d4e68ed DDP |
2451 | memset(entry, 0, sizeof *entry); |
2452 | conn_key_to_tuple(&conn->key, &entry->tuple_orig); | |
2453 | conn_key_to_tuple(&conn->rev_key, &entry->tuple_reply); | |
2454 | ||
2455 | entry->zone = conn->key.zone; | |
2456 | entry->mark = conn->mark; | |
2457 | ||
286de272 | 2458 | memcpy(&entry->labels, &conn->label, sizeof entry->labels); |
4d4e68ed DDP |
2459 | /* Not implemented yet */ |
2460 | entry->timestamp.start = 0; | |
2461 | entry->timestamp.stop = 0; | |
2462 | ||
dec0dbbc | 2463 | long long expiration = conn->expiration - now; |
4d4e68ed DDP |
2464 | entry->timeout = (expiration > 0) ? expiration / 1000 : 0; |
2465 | ||
dec0dbbc | 2466 | struct ct_l4_proto *class = l4_protos[conn->key.nw_proto]; |
4d4e68ed DDP |
2467 | if (class->conn_get_protoinfo) { |
2468 | class->conn_get_protoinfo(conn, &entry->protoinfo); | |
2469 | } | |
bd5e81a0 | 2470 | |
ded30c74 | 2471 | entry->bkt = bkt; |
bd5e81a0 DB |
2472 | |
2473 | if (conn->alg) { | |
2474 | /* Caller is responsible for freeing. */ | |
2475 | entry->helper.name = xstrdup(conn->alg); | |
2476 | } | |
4d4e68ed DDP |
2477 | } |
2478 | ||
4ea96698 DB |
2479 | struct ipf * |
2480 | conntrack_ipf_ctx(struct conntrack *ct) | |
2481 | { | |
2482 | return ct->ipf; | |
2483 | } | |
2484 | ||
4d4e68ed DDP |
2485 | int |
2486 | conntrack_dump_start(struct conntrack *ct, struct conntrack_dump *dump, | |
ded30c74 | 2487 | const uint16_t *pzone, int *ptot_bkts) |
4d4e68ed DDP |
2488 | { |
2489 | memset(dump, 0, sizeof(*dump)); | |
dec0dbbc | 2490 | |
4d4e68ed DDP |
2491 | if (pzone) { |
2492 | dump->zone = *pzone; | |
2493 | dump->filter_zone = true; | |
2494 | } | |
4d4e68ed | 2495 | |
dec0dbbc | 2496 | dump->ct = ct; |
ded30c74 | 2497 | *ptot_bkts = CONNTRACK_BUCKETS; |
4d4e68ed DDP |
2498 | return 0; |
2499 | } | |
2500 | ||
2501 | int | |
2502 | conntrack_dump_next(struct conntrack_dump *dump, struct ct_dpif_entry *entry) | |
2503 | { | |
2504 | struct conntrack *ct = dump->ct; | |
2505 | long long now = time_msec(); | |
2506 | ||
2507 | while (dump->bucket < CONNTRACK_BUCKETS) { | |
2508 | struct hmap_node *node; | |
2509 | ||
2510 | ct_lock_lock(&ct->buckets[dump->bucket].lock); | |
2511 | for (;;) { | |
2512 | struct conn *conn; | |
2513 | ||
2514 | node = hmap_at_position(&ct->buckets[dump->bucket].connections, | |
2515 | &dump->bucket_pos); | |
2516 | if (!node) { | |
2517 | break; | |
2518 | } | |
2519 | INIT_CONTAINER(conn, node, node); | |
286de272 DB |
2520 | if ((!dump->filter_zone || conn->key.zone == dump->zone) && |
2521 | (conn->conn_type != CT_CONN_TYPE_UN_NAT)) { | |
ded30c74 | 2522 | conn_to_ct_dpif_entry(conn, entry, now, dump->bucket); |
4d4e68ed DDP |
2523 | break; |
2524 | } | |
2525 | /* Else continue, until we find an entry in the appropriate zone | |
2526 | * or the bucket has been scanned completely. */ | |
2527 | } | |
2528 | ct_lock_unlock(&ct->buckets[dump->bucket].lock); | |
2529 | ||
2530 | if (!node) { | |
2531 | memset(&dump->bucket_pos, 0, sizeof dump->bucket_pos); | |
2532 | dump->bucket++; | |
2533 | } else { | |
2534 | return 0; | |
2535 | } | |
2536 | } | |
2537 | return EOF; | |
2538 | } | |
2539 | ||
2540 | int | |
2541 | conntrack_dump_done(struct conntrack_dump *dump OVS_UNUSED) | |
2542 | { | |
2543 | return 0; | |
2544 | } | |
5d9cbb4c DDP |
2545 | |
2546 | int | |
2547 | conntrack_flush(struct conntrack *ct, const uint16_t *zone) | |
2548 | { | |
dec0dbbc | 2549 | for (unsigned i = 0; i < CONNTRACK_BUCKETS; i++) { |
5d9cbb4c DDP |
2550 | struct conn *conn, *next; |
2551 | ||
2552 | ct_lock_lock(&ct->buckets[i].lock); | |
bd5e81a0 | 2553 | HMAP_FOR_EACH_SAFE (conn, next, node, &ct->buckets[i].connections) { |
286de272 DB |
2554 | if ((!zone || *zone == conn->key.zone) && |
2555 | (conn->conn_type == CT_CONN_TYPE_DEFAULT)) { | |
2556 | conn_clean(ct, conn, &ct->buckets[i]); | |
5d9cbb4c DDP |
2557 | } |
2558 | } | |
2559 | ct_lock_unlock(&ct->buckets[i].lock); | |
2560 | } | |
bd5e81a0 | 2561 | |
5d9cbb4c DDP |
2562 | return 0; |
2563 | } | |
bd5e81a0 | 2564 | |
271e48a0 YHW |
2565 | int |
2566 | conntrack_flush_tuple(struct conntrack *ct, const struct ct_dpif_tuple *tuple, | |
2567 | uint16_t zone) | |
2568 | { | |
2569 | struct conn_lookup_ctx ctx; | |
2570 | int error = 0; | |
2571 | ||
2572 | memset(&ctx, 0, sizeof(ctx)); | |
2573 | tuple_to_conn_key(tuple, zone, &ctx.key); | |
2574 | ctx.hash = conn_key_hash(&ctx.key, ct->hash_basis); | |
2575 | unsigned bucket = hash_to_bucket(ctx.hash); | |
2576 | ||
2577 | ct_lock_lock(&ct->buckets[bucket].lock); | |
2578 | conn_key_lookup(&ct->buckets[bucket], &ctx, time_msec()); | |
a1d5eeff | 2579 | if (ctx.conn && ctx.conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
271e48a0 YHW |
2580 | conn_clean(ct, ctx.conn, &ct->buckets[bucket]); |
2581 | } else { | |
a1d5eeff | 2582 | VLOG_WARN("Must flush tuple using the original pre-NATed tuple"); |
271e48a0 YHW |
2583 | error = ENOENT; |
2584 | } | |
2585 | ct_lock_unlock(&ct->buckets[bucket].lock); | |
2586 | return error; | |
2587 | } | |
2588 | ||
c92339ad DB |
2589 | int |
2590 | conntrack_set_maxconns(struct conntrack *ct, uint32_t maxconns) | |
2591 | { | |
2592 | atomic_store_relaxed(&ct->n_conn_limit, maxconns); | |
2593 | return 0; | |
2594 | } | |
2595 | ||
2596 | int | |
2597 | conntrack_get_maxconns(struct conntrack *ct, uint32_t *maxconns) | |
2598 | { | |
2599 | atomic_read_relaxed(&ct->n_conn_limit, maxconns); | |
2600 | return 0; | |
2601 | } | |
2602 | ||
875075b3 DB |
2603 | int |
2604 | conntrack_get_nconns(struct conntrack *ct, uint32_t *nconns) | |
2605 | { | |
2606 | *nconns = atomic_count_get(&ct->n_conn); | |
2607 | return 0; | |
2608 | } | |
2609 | ||
bd5e81a0 DB |
2610 | /* This function must be called with the ct->resources read lock taken. */ |
2611 | static struct alg_exp_node * | |
be38342d DB |
2612 | expectation_lookup(struct hmap *alg_expectations, const struct conn_key *key, |
2613 | uint32_t basis, bool src_ip_wc) | |
bd5e81a0 | 2614 | { |
c3f6bae2 DB |
2615 | struct conn_key check_key; |
2616 | memcpy(&check_key, key, sizeof check_key); | |
bd5e81a0 | 2617 | check_key.src.port = ALG_WC_SRC_PORT; |
dec0dbbc | 2618 | |
be38342d DB |
2619 | if (src_ip_wc) { |
2620 | memset(&check_key.src.addr, 0, sizeof check_key.src.addr); | |
2621 | } | |
dec0dbbc | 2622 | |
bd5e81a0 DB |
2623 | struct alg_exp_node *alg_exp_node; |
2624 | ||
bd5e81a0 | 2625 | HMAP_FOR_EACH_WITH_HASH (alg_exp_node, node, |
dec0dbbc | 2626 | conn_key_hash(&check_key, basis), |
bd5e81a0 DB |
2627 | alg_expectations) { |
2628 | if (!conn_key_cmp(&alg_exp_node->key, &check_key)) { | |
2629 | return alg_exp_node; | |
2630 | } | |
2631 | } | |
2632 | return NULL; | |
2633 | } | |
2634 | ||
4417ca3d DB |
2635 | /* This function must be called with the ct->resources write lock taken. */ |
2636 | static void | |
2637 | expectation_remove(struct hmap *alg_expectations, | |
2638 | const struct conn_key *key, uint32_t basis) | |
2639 | { | |
2640 | struct alg_exp_node *alg_exp_node; | |
2641 | ||
2642 | HMAP_FOR_EACH_WITH_HASH (alg_exp_node, node, conn_key_hash(key, basis), | |
2643 | alg_expectations) { | |
2644 | if (!conn_key_cmp(&alg_exp_node->key, key)) { | |
2645 | hmap_remove(alg_expectations, &alg_exp_node->node); | |
2646 | break; | |
2647 | } | |
2648 | } | |
2649 | } | |
2650 | ||
2651 | /* This function must be called with the ct->resources read lock taken. */ | |
2652 | static struct alg_exp_node * | |
2653 | expectation_ref_lookup_unique(const struct hindex *alg_expectation_refs, | |
2654 | const struct conn_key *master_key, | |
2655 | const struct conn_key *alg_exp_key, | |
2656 | uint32_t basis) | |
2657 | { | |
2658 | struct alg_exp_node *alg_exp_node; | |
2659 | ||
2660 | HINDEX_FOR_EACH_WITH_HASH (alg_exp_node, node_ref, | |
2661 | conn_key_hash(master_key, basis), | |
2662 | alg_expectation_refs) { | |
2663 | if (!conn_key_cmp(&alg_exp_node->master_key, master_key) && | |
2664 | !conn_key_cmp(&alg_exp_node->key, alg_exp_key)) { | |
2665 | return alg_exp_node; | |
2666 | } | |
2667 | } | |
2668 | return NULL; | |
2669 | } | |
2670 | ||
2671 | /* This function must be called with the ct->resources write lock taken. */ | |
2672 | static void | |
2673 | expectation_ref_create(struct hindex *alg_expectation_refs, | |
2674 | struct alg_exp_node *alg_exp_node, | |
2675 | uint32_t basis) | |
2676 | { | |
2677 | if (!expectation_ref_lookup_unique(alg_expectation_refs, | |
2678 | &alg_exp_node->master_key, | |
2679 | &alg_exp_node->key, basis)) { | |
2680 | hindex_insert(alg_expectation_refs, &alg_exp_node->node_ref, | |
2681 | conn_key_hash(&alg_exp_node->master_key, basis)); | |
2682 | } | |
2683 | } | |
2684 | ||
2685 | static void | |
2686 | expectation_clean(struct conntrack *ct, const struct conn_key *master_key, | |
2687 | uint32_t basis) | |
2688 | { | |
2689 | ct_rwlock_wrlock(&ct->resources_lock); | |
2690 | ||
2691 | struct alg_exp_node *node, *next; | |
2692 | HINDEX_FOR_EACH_WITH_HASH_SAFE (node, next, node_ref, | |
2693 | conn_key_hash(master_key, basis), | |
2694 | &ct->alg_expectation_refs) { | |
2695 | if (!conn_key_cmp(&node->master_key, master_key)) { | |
2696 | expectation_remove(&ct->alg_expectations, &node->key, basis); | |
2697 | hindex_remove(&ct->alg_expectation_refs, &node->node_ref); | |
2698 | free(node); | |
2699 | } | |
2700 | } | |
2701 | ||
2702 | ct_rwlock_unlock(&ct->resources_lock); | |
2703 | } | |
2704 | ||
bd5e81a0 | 2705 | static void |
be38342d DB |
2706 | expectation_create(struct conntrack *ct, ovs_be16 dst_port, |
2707 | const struct conn *master_conn, bool reply, bool src_ip_wc, | |
2708 | bool skip_nat) | |
bd5e81a0 | 2709 | { |
cda1b109 DB |
2710 | union ct_addr src_addr; |
2711 | union ct_addr dst_addr; | |
2712 | union ct_addr alg_nat_repl_addr; | |
be38342d | 2713 | struct alg_exp_node *alg_exp_node = xzalloc(sizeof *alg_exp_node); |
bd5e81a0 | 2714 | |
be38342d | 2715 | if (reply) { |
bd5e81a0 DB |
2716 | src_addr = master_conn->key.src.addr; |
2717 | dst_addr = master_conn->key.dst.addr; | |
efa29a89 | 2718 | alg_exp_node->nat_rpl_dst = true; |
be38342d DB |
2719 | if (skip_nat) { |
2720 | alg_nat_repl_addr = dst_addr; | |
efa29a89 DM |
2721 | } else if (master_conn->nat_info && |
2722 | master_conn->nat_info->nat_action & NAT_ACTION_DST) { | |
2723 | alg_nat_repl_addr = master_conn->rev_key.src.addr; | |
2724 | alg_exp_node->nat_rpl_dst = false; | |
be38342d DB |
2725 | } else { |
2726 | alg_nat_repl_addr = master_conn->rev_key.dst.addr; | |
2727 | } | |
be38342d DB |
2728 | } else { |
2729 | src_addr = master_conn->rev_key.src.addr; | |
2730 | dst_addr = master_conn->rev_key.dst.addr; | |
efa29a89 | 2731 | alg_exp_node->nat_rpl_dst = false; |
be38342d DB |
2732 | if (skip_nat) { |
2733 | alg_nat_repl_addr = src_addr; | |
efa29a89 DM |
2734 | } else if (master_conn->nat_info && |
2735 | master_conn->nat_info->nat_action & NAT_ACTION_DST) { | |
2736 | alg_nat_repl_addr = master_conn->key.dst.addr; | |
2737 | alg_exp_node->nat_rpl_dst = true; | |
be38342d DB |
2738 | } else { |
2739 | alg_nat_repl_addr = master_conn->key.src.addr; | |
2740 | } | |
be38342d DB |
2741 | } |
2742 | if (src_ip_wc) { | |
2743 | memset(&src_addr, 0, sizeof src_addr); | |
bd5e81a0 DB |
2744 | } |
2745 | ||
bd5e81a0 DB |
2746 | alg_exp_node->key.dl_type = master_conn->key.dl_type; |
2747 | alg_exp_node->key.nw_proto = master_conn->key.nw_proto; | |
2748 | alg_exp_node->key.zone = master_conn->key.zone; | |
2749 | alg_exp_node->key.src.addr = src_addr; | |
2750 | alg_exp_node->key.dst.addr = dst_addr; | |
2751 | alg_exp_node->key.src.port = ALG_WC_SRC_PORT; | |
2752 | alg_exp_node->key.dst.port = dst_port; | |
2753 | alg_exp_node->master_mark = master_conn->mark; | |
2754 | alg_exp_node->master_label = master_conn->label; | |
2755 | alg_exp_node->master_key = master_conn->key; | |
bd5e81a0 DB |
2756 | /* Take the write lock here because it is almost 100% |
2757 | * likely that the lookup will fail and | |
2758 | * expectation_create() will be called below. */ | |
2759 | ct_rwlock_wrlock(&ct->resources_lock); | |
2760 | struct alg_exp_node *alg_exp = expectation_lookup( | |
be38342d | 2761 | &ct->alg_expectations, &alg_exp_node->key, ct->hash_basis, src_ip_wc); |
bd5e81a0 DB |
2762 | if (alg_exp) { |
2763 | free(alg_exp_node); | |
2764 | ct_rwlock_unlock(&ct->resources_lock); | |
2765 | return; | |
2766 | } | |
2767 | ||
2768 | alg_exp_node->alg_nat_repl_addr = alg_nat_repl_addr; | |
4417ca3d | 2769 | hmap_insert(&ct->alg_expectations, &alg_exp_node->node, |
dec0dbbc | 2770 | conn_key_hash(&alg_exp_node->key, ct->hash_basis)); |
4417ca3d DB |
2771 | expectation_ref_create(&ct->alg_expectation_refs, alg_exp_node, |
2772 | ct->hash_basis); | |
bd5e81a0 DB |
2773 | ct_rwlock_unlock(&ct->resources_lock); |
2774 | } | |
2775 | ||
bd5e81a0 DB |
2776 | static void |
2777 | replace_substring(char *substr, uint8_t substr_size, | |
2778 | uint8_t total_size, char *rep_str, | |
2779 | uint8_t rep_str_size) | |
2780 | { | |
2781 | memmove(substr + rep_str_size, substr + substr_size, | |
2782 | total_size - substr_size); | |
2783 | memcpy(substr, rep_str, rep_str_size); | |
2784 | } | |
2785 | ||
cd7c99a6 DB |
2786 | static void |
2787 | repl_bytes(char *str, char c1, char c2) | |
2788 | { | |
2789 | while (*str) { | |
2790 | if (*str == c1) { | |
2791 | *str = c2; | |
2792 | } | |
2793 | str++; | |
2794 | } | |
2795 | } | |
2796 | ||
2797 | static void | |
2798 | modify_packet(struct dp_packet *pkt, char *pkt_str, size_t size, | |
2799 | char *repl_str, size_t repl_size, | |
2800 | uint32_t orig_used_size) | |
2801 | { | |
2802 | replace_substring(pkt_str, size, | |
2803 | (const char *) dp_packet_tail(pkt) - pkt_str, | |
2804 | repl_str, repl_size); | |
2805 | dp_packet_set_size(pkt, orig_used_size + (int) repl_size - (int) size); | |
2806 | } | |
2807 | ||
bd5e81a0 DB |
2808 | /* Replace IPV4 address in FTP message with NATed address. */ |
2809 | static int | |
2810 | repl_ftp_v4_addr(struct dp_packet *pkt, ovs_be32 v4_addr_rep, | |
2811 | char *ftp_data_start, | |
cd7c99a6 DB |
2812 | size_t addr_offset_from_ftp_data_start, |
2813 | size_t addr_size OVS_UNUSED) | |
bd5e81a0 DB |
2814 | { |
2815 | enum { MAX_FTP_V4_NAT_DELTA = 8 }; | |
2816 | ||
2817 | /* Do conservative check for pathological MTU usage. */ | |
2818 | uint32_t orig_used_size = dp_packet_size(pkt); | |
cd7c99a6 DB |
2819 | if (orig_used_size + MAX_FTP_V4_NAT_DELTA > |
2820 | dp_packet_get_allocated(pkt)) { | |
2821 | ||
bd5e81a0 | 2822 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
cd7c99a6 DB |
2823 | VLOG_WARN_RL(&rl, "Unsupported effective MTU %u used with FTP V4", |
2824 | dp_packet_get_allocated(pkt)); | |
bd5e81a0 DB |
2825 | return 0; |
2826 | } | |
2827 | ||
cd7c99a6 DB |
2828 | char v4_addr_str[INET_ADDRSTRLEN] = {0}; |
2829 | ovs_assert(inet_ntop(AF_INET, &v4_addr_rep, v4_addr_str, | |
2830 | sizeof v4_addr_str)); | |
2831 | repl_bytes(v4_addr_str, '.', ','); | |
2832 | modify_packet(pkt, ftp_data_start + addr_offset_from_ftp_data_start, | |
2833 | addr_size, v4_addr_str, strlen(v4_addr_str), | |
2834 | orig_used_size); | |
2835 | return (int) strlen(v4_addr_str) - (int) addr_size; | |
bd5e81a0 DB |
2836 | } |
2837 | ||
2838 | static char * | |
2839 | skip_non_digits(char *str) | |
2840 | { | |
2841 | while (!isdigit(*str) && *str != 0) { | |
2842 | str++; | |
2843 | } | |
2844 | return str; | |
2845 | } | |
2846 | ||
2847 | static char * | |
2848 | terminate_number_str(char *str, uint8_t max_digits) | |
2849 | { | |
2850 | uint8_t digits_found = 0; | |
2851 | while (isdigit(*str) && digits_found <= max_digits) { | |
2852 | str++; | |
2853 | digits_found++; | |
2854 | } | |
2855 | ||
2856 | *str = 0; | |
2857 | return str; | |
2858 | } | |
2859 | ||
2860 | ||
2861 | static void | |
2862 | get_ftp_ctl_msg(struct dp_packet *pkt, char *ftp_msg) | |
2863 | { | |
2864 | struct tcp_header *th = dp_packet_l4(pkt); | |
2865 | char *tcp_hdr = (char *) th; | |
2866 | uint32_t tcp_payload_len = tcp_payload_length(pkt); | |
2867 | size_t tcp_payload_of_interest = MIN(tcp_payload_len, | |
2868 | LARGEST_FTP_MSG_OF_INTEREST); | |
2869 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
2870 | ||
2871 | ovs_strlcpy(ftp_msg, tcp_hdr + tcp_hdr_len, | |
2872 | tcp_payload_of_interest); | |
2873 | } | |
2874 | ||
2875 | static enum ftp_ctl_pkt | |
2876 | detect_ftp_ctl_type(const struct conn_lookup_ctx *ctx, | |
2877 | struct dp_packet *pkt) | |
2878 | { | |
bd5e81a0 DB |
2879 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; |
2880 | get_ftp_ctl_msg(pkt, ftp_msg); | |
dec0dbbc | 2881 | |
bd5e81a0 DB |
2882 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
2883 | if (strncasecmp(ftp_msg, FTP_EPRT_CMD, strlen(FTP_EPRT_CMD)) && | |
2884 | !strcasestr(ftp_msg, FTP_EPSV_REPLY)) { | |
2885 | return CT_FTP_CTL_OTHER; | |
2886 | } | |
2887 | } else { | |
2888 | if (strncasecmp(ftp_msg, FTP_PORT_CMD, strlen(FTP_PORT_CMD)) && | |
2889 | strncasecmp(ftp_msg, FTP_PASV_REPLY_CODE, | |
2890 | strlen(FTP_PASV_REPLY_CODE))) { | |
2891 | return CT_FTP_CTL_OTHER; | |
2892 | } | |
2893 | } | |
2894 | ||
2895 | return CT_FTP_CTL_INTEREST; | |
2896 | } | |
2897 | ||
2898 | static enum ftp_ctl_pkt | |
2899 | process_ftp_ctl_v4(struct conntrack *ct, | |
2900 | struct dp_packet *pkt, | |
2901 | const struct conn *conn_for_expectation, | |
4417ca3d | 2902 | ovs_be32 *v4_addr_rep, |
bd5e81a0 | 2903 | char **ftp_data_v4_start, |
cd7c99a6 DB |
2904 | size_t *addr_offset_from_ftp_data_start, |
2905 | size_t *addr_size) | |
bd5e81a0 DB |
2906 | { |
2907 | struct tcp_header *th = dp_packet_l4(pkt); | |
2908 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
2909 | char *tcp_hdr = (char *) th; | |
2910 | *ftp_data_v4_start = tcp_hdr + tcp_hdr_len; | |
2911 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; | |
2912 | get_ftp_ctl_msg(pkt, ftp_msg); | |
bd5e81a0 DB |
2913 | char *ftp = ftp_msg; |
2914 | enum ct_alg_mode mode; | |
dec0dbbc | 2915 | |
23bea975 | 2916 | if (!strncasecmp(ftp, FTP_PORT_CMD, strlen(FTP_PORT_CMD))) { |
bd5e81a0 DB |
2917 | ftp = ftp_msg + strlen(FTP_PORT_CMD); |
2918 | mode = CT_FTP_MODE_ACTIVE; | |
2919 | } else { | |
2920 | ftp = ftp_msg + strlen(FTP_PASV_REPLY_CODE); | |
2921 | mode = CT_FTP_MODE_PASSIVE; | |
2922 | } | |
2923 | ||
2924 | /* Find first space. */ | |
2925 | ftp = strchr(ftp, ' '); | |
2926 | if (!ftp) { | |
2927 | return CT_FTP_CTL_INVALID; | |
2928 | } | |
2929 | ||
2930 | /* Find the first digit, after space. */ | |
2931 | ftp = skip_non_digits(ftp); | |
2932 | if (*ftp == 0) { | |
2933 | return CT_FTP_CTL_INVALID; | |
2934 | } | |
2935 | ||
2936 | char *ip_addr_start = ftp; | |
2937 | *addr_offset_from_ftp_data_start = ip_addr_start - ftp_msg; | |
bd5e81a0 | 2938 | |
dec0dbbc | 2939 | uint8_t comma_count = 0; |
bd5e81a0 DB |
2940 | while (comma_count < 4 && *ftp) { |
2941 | if (*ftp == ',') { | |
2942 | comma_count++; | |
2943 | if (comma_count == 4) { | |
2944 | *ftp = 0; | |
2945 | } else { | |
2946 | *ftp = '.'; | |
2947 | } | |
2948 | } | |
2949 | ftp++; | |
2950 | } | |
2951 | if (comma_count != 4) { | |
2952 | return CT_FTP_CTL_INVALID; | |
2953 | } | |
2954 | ||
2955 | struct in_addr ip_addr; | |
2956 | int rc2 = inet_pton(AF_INET, ip_addr_start, &ip_addr); | |
2957 | if (rc2 != 1) { | |
2958 | return CT_FTP_CTL_INVALID; | |
2959 | } | |
2960 | ||
cd7c99a6 | 2961 | *addr_size = ftp - ip_addr_start - 1; |
bd5e81a0 DB |
2962 | char *save_ftp = ftp; |
2963 | ftp = terminate_number_str(ftp, MAX_FTP_PORT_DGTS); | |
2964 | if (!ftp) { | |
2965 | return CT_FTP_CTL_INVALID; | |
2966 | } | |
2967 | int value; | |
2968 | if (!str_to_int(save_ftp, 10, &value)) { | |
2969 | return CT_FTP_CTL_INVALID; | |
2970 | } | |
2971 | ||
2972 | /* This is derived from the L4 port maximum is 65535. */ | |
2973 | if (value > 255) { | |
2974 | return CT_FTP_CTL_INVALID; | |
2975 | } | |
2976 | ||
2977 | uint16_t port_hs = value; | |
2978 | port_hs <<= 8; | |
2979 | ||
2980 | /* Skip over comma. */ | |
2981 | ftp++; | |
2982 | save_ftp = ftp; | |
2983 | bool digit_found = false; | |
2984 | while (isdigit(*ftp)) { | |
2985 | ftp++; | |
2986 | digit_found = true; | |
2987 | } | |
2988 | if (!digit_found) { | |
2989 | return CT_FTP_CTL_INVALID; | |
2990 | } | |
2991 | *ftp = 0; | |
2992 | if (!str_to_int(save_ftp, 10, &value)) { | |
2993 | return CT_FTP_CTL_INVALID; | |
2994 | } | |
2995 | ||
2996 | if (value > 255) { | |
2997 | return CT_FTP_CTL_INVALID; | |
2998 | } | |
2999 | ||
78a0b272 | 3000 | port_hs |= value; |
bd5e81a0 DB |
3001 | ovs_be16 port = htons(port_hs); |
3002 | ovs_be32 conn_ipv4_addr; | |
3003 | ||
3004 | switch (mode) { | |
3005 | case CT_FTP_MODE_ACTIVE: | |
cda1b109 DB |
3006 | *v4_addr_rep = conn_for_expectation->rev_key.dst.addr.ipv4; |
3007 | conn_ipv4_addr = conn_for_expectation->key.src.addr.ipv4; | |
bd5e81a0 DB |
3008 | break; |
3009 | case CT_FTP_MODE_PASSIVE: | |
cda1b109 DB |
3010 | *v4_addr_rep = conn_for_expectation->key.dst.addr.ipv4; |
3011 | conn_ipv4_addr = conn_for_expectation->rev_key.src.addr.ipv4; | |
bd5e81a0 | 3012 | break; |
7be77cb0 | 3013 | case CT_TFTP_MODE: |
bd5e81a0 DB |
3014 | default: |
3015 | OVS_NOT_REACHED(); | |
3016 | } | |
3017 | ||
3018 | ovs_be32 ftp_ipv4_addr; | |
3019 | ftp_ipv4_addr = ip_addr.s_addr; | |
3020 | /* Although most servers will block this exploit, there may be some | |
3021 | * less well managed. */ | |
3022 | if (ftp_ipv4_addr != conn_ipv4_addr && ftp_ipv4_addr != *v4_addr_rep) { | |
3023 | return CT_FTP_CTL_INVALID; | |
3024 | } | |
3025 | ||
be38342d DB |
3026 | expectation_create(ct, port, conn_for_expectation, |
3027 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
bd5e81a0 DB |
3028 | return CT_FTP_CTL_INTEREST; |
3029 | } | |
3030 | ||
3031 | static char * | |
3032 | skip_ipv6_digits(char *str) | |
3033 | { | |
3034 | while (isxdigit(*str) || *str == ':' || *str == '.') { | |
3035 | str++; | |
3036 | } | |
3037 | return str; | |
3038 | } | |
3039 | ||
3040 | static enum ftp_ctl_pkt | |
3041 | process_ftp_ctl_v6(struct conntrack *ct, | |
3042 | struct dp_packet *pkt, | |
3043 | const struct conn *conn_for_expectation, | |
cda1b109 | 3044 | union ct_addr *v6_addr_rep, char **ftp_data_start, |
bd5e81a0 DB |
3045 | size_t *addr_offset_from_ftp_data_start, |
3046 | size_t *addr_size, enum ct_alg_mode *mode) | |
3047 | { | |
3048 | struct tcp_header *th = dp_packet_l4(pkt); | |
3049 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
3050 | char *tcp_hdr = (char *) th; | |
3051 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; | |
bd5e81a0 DB |
3052 | get_ftp_ctl_msg(pkt, ftp_msg); |
3053 | *ftp_data_start = tcp_hdr + tcp_hdr_len; | |
bd5e81a0 DB |
3054 | char *ftp = ftp_msg; |
3055 | struct in6_addr ip6_addr; | |
dec0dbbc | 3056 | |
23bea975 | 3057 | if (!strncasecmp(ftp, FTP_EPRT_CMD, strlen(FTP_EPRT_CMD))) { |
bd5e81a0 DB |
3058 | ftp = ftp_msg + strlen(FTP_EPRT_CMD); |
3059 | ftp = skip_non_digits(ftp); | |
3060 | if (*ftp != FTP_AF_V6 || isdigit(ftp[1])) { | |
3061 | return CT_FTP_CTL_INVALID; | |
3062 | } | |
3063 | /* Jump over delimiter. */ | |
3064 | ftp += 2; | |
3065 | ||
bd5e81a0 | 3066 | memset(&ip6_addr, 0, sizeof ip6_addr); |
dec0dbbc | 3067 | char *ip_addr_start = ftp; |
bd5e81a0 DB |
3068 | *addr_offset_from_ftp_data_start = ip_addr_start - ftp_msg; |
3069 | ftp = skip_ipv6_digits(ftp); | |
3070 | *ftp = 0; | |
3071 | *addr_size = ftp - ip_addr_start; | |
3072 | int rc2 = inet_pton(AF_INET6, ip_addr_start, &ip6_addr); | |
3073 | if (rc2 != 1) { | |
3074 | return CT_FTP_CTL_INVALID; | |
3075 | } | |
3076 | ftp++; | |
3077 | *mode = CT_FTP_MODE_ACTIVE; | |
3078 | } else { | |
3079 | ftp = ftp_msg + strcspn(ftp_msg, "("); | |
3080 | ftp = skip_non_digits(ftp); | |
3081 | if (!isdigit(*ftp)) { | |
3082 | return CT_FTP_CTL_INVALID; | |
3083 | } | |
3084 | ||
3085 | /* Not used for passive mode. */ | |
3086 | *addr_offset_from_ftp_data_start = 0; | |
3087 | *addr_size = 0; | |
3088 | ||
3089 | *mode = CT_FTP_MODE_PASSIVE; | |
3090 | } | |
3091 | ||
3092 | char *save_ftp = ftp; | |
3093 | ftp = terminate_number_str(ftp, MAX_EXT_FTP_PORT_DGTS); | |
3094 | if (!ftp) { | |
3095 | return CT_FTP_CTL_INVALID; | |
3096 | } | |
dec0dbbc | 3097 | |
bd5e81a0 DB |
3098 | int value; |
3099 | if (!str_to_int(save_ftp, 10, &value)) { | |
3100 | return CT_FTP_CTL_INVALID; | |
3101 | } | |
3102 | if (value > CT_MAX_L4_PORT) { | |
3103 | return CT_FTP_CTL_INVALID; | |
3104 | } | |
3105 | ||
3106 | uint16_t port_hs = value; | |
3107 | ovs_be16 port = htons(port_hs); | |
3108 | ||
3109 | switch (*mode) { | |
3110 | case CT_FTP_MODE_ACTIVE: | |
3111 | *v6_addr_rep = conn_for_expectation->rev_key.dst.addr; | |
3112 | /* Although most servers will block this exploit, there may be some | |
3113 | * less well managed. */ | |
cda1b109 DB |
3114 | if (memcmp(&ip6_addr, &v6_addr_rep->ipv6, sizeof ip6_addr) && |
3115 | memcmp(&ip6_addr, &conn_for_expectation->key.src.addr.ipv6, | |
bd5e81a0 DB |
3116 | sizeof ip6_addr)) { |
3117 | return CT_FTP_CTL_INVALID; | |
3118 | } | |
3119 | break; | |
3120 | case CT_FTP_MODE_PASSIVE: | |
3121 | *v6_addr_rep = conn_for_expectation->key.dst.addr; | |
3122 | break; | |
7be77cb0 | 3123 | case CT_TFTP_MODE: |
bd5e81a0 DB |
3124 | default: |
3125 | OVS_NOT_REACHED(); | |
3126 | } | |
3127 | ||
be38342d DB |
3128 | expectation_create(ct, port, conn_for_expectation, |
3129 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
bd5e81a0 DB |
3130 | return CT_FTP_CTL_INTEREST; |
3131 | } | |
3132 | ||
3133 | static int | |
cda1b109 | 3134 | repl_ftp_v6_addr(struct dp_packet *pkt, union ct_addr v6_addr_rep, |
bd5e81a0 DB |
3135 | char *ftp_data_start, |
3136 | size_t addr_offset_from_ftp_data_start, | |
3137 | size_t addr_size, enum ct_alg_mode mode) | |
3138 | { | |
3139 | /* This is slightly bigger than really possible. */ | |
3140 | enum { MAX_FTP_V6_NAT_DELTA = 45 }; | |
3141 | ||
3142 | if (mode == CT_FTP_MODE_PASSIVE) { | |
3143 | return 0; | |
3144 | } | |
3145 | ||
3146 | /* Do conservative check for pathological MTU usage. */ | |
3147 | uint32_t orig_used_size = dp_packet_size(pkt); | |
cd7c99a6 DB |
3148 | if (orig_used_size + MAX_FTP_V6_NAT_DELTA > |
3149 | dp_packet_get_allocated(pkt)) { | |
3150 | ||
bd5e81a0 | 3151 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
cd7c99a6 DB |
3152 | VLOG_WARN_RL(&rl, "Unsupported effective MTU %u used with FTP V6", |
3153 | dp_packet_get_allocated(pkt)); | |
bd5e81a0 DB |
3154 | return 0; |
3155 | } | |
3156 | ||
298530b8 | 3157 | char v6_addr_str[INET6_ADDRSTRLEN] = {0}; |
cda1b109 | 3158 | ovs_assert(inet_ntop(AF_INET6, &v6_addr_rep.ipv6, v6_addr_str, |
298530b8 | 3159 | sizeof v6_addr_str)); |
cd7c99a6 DB |
3160 | modify_packet(pkt, ftp_data_start + addr_offset_from_ftp_data_start, |
3161 | addr_size, v6_addr_str, strlen(v6_addr_str), | |
3162 | orig_used_size); | |
3163 | return (int) strlen(v6_addr_str) - (int) addr_size; | |
bd5e81a0 DB |
3164 | } |
3165 | ||
d13d7115 DB |
3166 | /* Increment/decrement a TCP sequence number. */ |
3167 | static void | |
3168 | adj_seqnum(ovs_16aligned_be32 *val, int32_t inc) | |
3169 | { | |
3170 | put_16aligned_be32(val, htonl(ntohl(get_16aligned_be32(val)) + inc)); | |
3171 | } | |
3172 | ||
bd5e81a0 DB |
3173 | static void |
3174 | handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
253e4dc0 DM |
3175 | struct dp_packet *pkt, const struct conn *ec, long long now, |
3176 | enum ftp_ctl_pkt ftp_ctl, bool nat) | |
bd5e81a0 DB |
3177 | { |
3178 | struct ip_header *l3_hdr = dp_packet_l3(pkt); | |
3179 | ovs_be32 v4_addr_rep = 0; | |
cda1b109 | 3180 | union ct_addr v6_addr_rep; |
faa0826d | 3181 | size_t addr_offset_from_ftp_data_start = 0; |
bd5e81a0 DB |
3182 | size_t addr_size = 0; |
3183 | char *ftp_data_start; | |
bd5e81a0 DB |
3184 | enum ct_alg_mode mode = CT_FTP_MODE_ACTIVE; |
3185 | ||
3186 | if (detect_ftp_ctl_type(ctx, pkt) != ftp_ctl) { | |
3187 | return; | |
3188 | } | |
3189 | ||
bd5e81a0 DB |
3190 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); |
3191 | int64_t seq_skew = 0; | |
dec0dbbc | 3192 | |
253e4dc0 | 3193 | if (ftp_ctl == CT_FTP_CTL_INTEREST) { |
bd5e81a0 DB |
3194 | enum ftp_ctl_pkt rc; |
3195 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { | |
253e4dc0 | 3196 | rc = process_ftp_ctl_v6(ct, pkt, ec, |
4417ca3d | 3197 | &v6_addr_rep, &ftp_data_start, |
bd5e81a0 DB |
3198 | &addr_offset_from_ftp_data_start, |
3199 | &addr_size, &mode); | |
3200 | } else { | |
253e4dc0 | 3201 | rc = process_ftp_ctl_v4(ct, pkt, ec, |
4417ca3d | 3202 | &v4_addr_rep, &ftp_data_start, |
cd7c99a6 DB |
3203 | &addr_offset_from_ftp_data_start, |
3204 | &addr_size); | |
bd5e81a0 DB |
3205 | } |
3206 | if (rc == CT_FTP_CTL_INVALID) { | |
3207 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
3208 | VLOG_WARN_RL(&rl, "Invalid FTP control packet format"); | |
3209 | pkt->md.ct_state |= CS_TRACKED | CS_INVALID; | |
3210 | return; | |
3211 | } else if (rc == CT_FTP_CTL_INTEREST) { | |
3212 | uint16_t ip_len; | |
dec0dbbc | 3213 | |
bd5e81a0 | 3214 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
253e4dc0 DM |
3215 | if (nat) { |
3216 | seq_skew = repl_ftp_v6_addr(pkt, v6_addr_rep, | |
3217 | ftp_data_start, | |
3218 | addr_offset_from_ftp_data_start, | |
3219 | addr_size, mode); | |
3220 | } | |
3221 | ||
bd5e81a0 | 3222 | if (seq_skew) { |
253e4dc0 DM |
3223 | ip_len = ntohs(nh6->ip6_ctlun.ip6_un1.ip6_un1_plen) + |
3224 | seq_skew; | |
bd5e81a0 | 3225 | nh6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(ip_len); |
bd5e81a0 DB |
3226 | } |
3227 | } else { | |
253e4dc0 DM |
3228 | if (nat) { |
3229 | seq_skew = repl_ftp_v4_addr(pkt, v4_addr_rep, | |
3230 | ftp_data_start, | |
cd7c99a6 DB |
3231 | addr_offset_from_ftp_data_start, |
3232 | addr_size); | |
253e4dc0 | 3233 | } |
bd5e81a0 | 3234 | if (seq_skew) { |
253e4dc0 | 3235 | ip_len = ntohs(l3_hdr->ip_tot_len) + seq_skew; |
bd5e81a0 DB |
3236 | l3_hdr->ip_csum = recalc_csum16(l3_hdr->ip_csum, |
3237 | l3_hdr->ip_tot_len, htons(ip_len)); | |
3238 | l3_hdr->ip_tot_len = htons(ip_len); | |
bd5e81a0 DB |
3239 | } |
3240 | } | |
3241 | } else { | |
3242 | OVS_NOT_REACHED(); | |
3243 | } | |
bd5e81a0 DB |
3244 | } |
3245 | ||
3246 | struct tcp_header *th = dp_packet_l4(pkt); | |
dec0dbbc | 3247 | |
253e4dc0 | 3248 | if (nat && ec->seq_skew != 0) { |
d13d7115 DB |
3249 | ctx->reply != ec->seq_skew_dir ? |
3250 | adj_seqnum(&th->tcp_ack, -ec->seq_skew) : | |
3251 | adj_seqnum(&th->tcp_seq, ec->seq_skew); | |
bd5e81a0 DB |
3252 | } |
3253 | ||
bd5e81a0 | 3254 | th->tcp_csum = 0; |
bd5e81a0 | 3255 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
76d85771 DB |
3256 | th->tcp_csum = packet_csum_upperlayer6(nh6, th, ctx->key.nw_proto, |
3257 | dp_packet_l4_size(pkt)); | |
bd5e81a0 | 3258 | } else { |
76d85771 DB |
3259 | uint32_t tcp_csum = packet_csum_pseudoheader(l3_hdr); |
3260 | th->tcp_csum = csum_finish( | |
3261 | csum_continue(tcp_csum, th, dp_packet_l4_size(pkt))); | |
bd5e81a0 | 3262 | } |
253e4dc0 DM |
3263 | |
3264 | if (seq_skew) { | |
3265 | conn_seq_skew_set(ct, &ec->key, now, seq_skew + ec->seq_skew, | |
3266 | ctx->reply); | |
3267 | } | |
bd5e81a0 | 3268 | } |
7be77cb0 DB |
3269 | |
3270 | static void | |
3271 | handle_tftp_ctl(struct conntrack *ct, | |
94e71143 | 3272 | const struct conn_lookup_ctx *ctx OVS_UNUSED, |
be38342d | 3273 | struct dp_packet *pkt, |
7be77cb0 | 3274 | const struct conn *conn_for_expectation, |
4417ca3d DB |
3275 | long long now OVS_UNUSED, |
3276 | enum ftp_ctl_pkt ftp_ctl OVS_UNUSED, bool nat OVS_UNUSED) | |
7be77cb0 | 3277 | { |
be38342d DB |
3278 | expectation_create(ct, conn_for_expectation->key.src.port, |
3279 | conn_for_expectation, | |
3280 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
7be77cb0 | 3281 | } |