]>
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" | |
33 | #include "netdev.h" | |
34 | #include "odp-netlink.h" | |
35 | #include "openvswitch/hmap.h" | |
36 | #include "openvswitch/vlog.h" | |
37 | #include "ovs-rcu.h" | |
e6ef6cc6 | 38 | #include "ovs-thread.h" |
fd016ae3 | 39 | #include "openvswitch/poll-loop.h" |
a489b168 DDP |
40 | #include "random.h" |
41 | #include "timeval.h" | |
42 | ||
43 | VLOG_DEFINE_THIS_MODULE(conntrack); | |
44 | ||
45 | COVERAGE_DEFINE(conntrack_full); | |
e6ef6cc6 | 46 | COVERAGE_DEFINE(conntrack_long_cleanup); |
a489b168 DDP |
47 | |
48 | struct conn_lookup_ctx { | |
49 | struct conn_key key; | |
50 | struct conn *conn; | |
51 | uint32_t hash; | |
52 | bool reply; | |
dbb597d3 | 53 | bool icmp_related; |
a489b168 DDP |
54 | }; |
55 | ||
bd5e81a0 DB |
56 | enum ftp_ctl_pkt { |
57 | /* Control packets with address and/or port specifiers. */ | |
58 | CT_FTP_CTL_INTEREST, | |
59 | /* Control packets without address and/or port specifiers. */ | |
60 | CT_FTP_CTL_OTHER, | |
61 | CT_FTP_CTL_INVALID, | |
62 | }; | |
63 | ||
64 | enum ct_alg_mode { | |
65 | CT_FTP_MODE_ACTIVE, | |
66 | CT_FTP_MODE_PASSIVE, | |
7be77cb0 | 67 | CT_TFTP_MODE, |
bd5e81a0 DB |
68 | }; |
69 | ||
94e71143 DB |
70 | enum ct_alg_ctl_type { |
71 | CT_ALG_CTL_NONE, | |
72 | CT_ALG_CTL_FTP, | |
73 | CT_ALG_CTL_TFTP, | |
be38342d DB |
74 | /* SIP is not enabled through Openflow and presently only used as |
75 | * an example of an alg that allows a wildcard src ip. */ | |
76 | CT_ALG_CTL_SIP, | |
94e71143 DB |
77 | }; |
78 | ||
a489b168 | 79 | static bool conn_key_extract(struct conntrack *, struct dp_packet *, |
66e4ad8a DDP |
80 | ovs_be16 dl_type, struct conn_lookup_ctx *, |
81 | uint16_t zone); | |
a489b168 DDP |
82 | static uint32_t conn_key_hash(const struct conn_key *, uint32_t basis); |
83 | static void conn_key_reverse(struct conn_key *); | |
a489b168 | 84 | static bool valid_new(struct dp_packet *pkt, struct conn_key *); |
967bb5c5 | 85 | static struct conn *new_conn(struct conntrack *ct, struct dp_packet *pkt, |
e6ef6cc6 | 86 | struct conn_key *, long long now); |
967bb5c5 | 87 | static void delete_conn_cmn(struct conn *); |
a489b168 | 88 | static void delete_conn(struct conn *); |
967bb5c5 DB |
89 | static void delete_conn_one(struct conn *conn); |
90 | static enum ct_update_res conn_update(struct conntrack *ct, struct conn *conn, | |
91 | struct dp_packet *pkt, | |
92 | struct conn_lookup_ctx *ctx, | |
e6ef6cc6 | 93 | long long now); |
a489b168 DDP |
94 | static bool conn_expired(struct conn *, long long now); |
95 | static void set_mark(struct dp_packet *, struct conn *, | |
96 | uint32_t val, uint32_t mask); | |
97 | static void set_label(struct dp_packet *, struct conn *, | |
98 | const struct ovs_key_ct_labels *val, | |
99 | const struct ovs_key_ct_labels *mask); | |
e6ef6cc6 | 100 | static void *clean_thread_main(void *f_); |
a489b168 | 101 | |
286de272 DB |
102 | static bool |
103 | nat_select_range_tuple(struct conntrack *ct, const struct conn *conn, | |
104 | struct conn *nat_conn); | |
105 | ||
106 | static uint8_t | |
107 | reverse_icmp_type(uint8_t type); | |
108 | static uint8_t | |
109 | reverse_icmp6_type(uint8_t type); | |
110 | static inline bool | |
111 | extract_l3_ipv4(struct conn_key *key, const void *data, size_t size, | |
112 | const char **new_data, bool validate_checksum); | |
113 | static inline bool | |
114 | extract_l3_ipv6(struct conn_key *key, const void *data, size_t size, | |
115 | const char **new_data); | |
bd5e81a0 | 116 | static struct alg_exp_node * |
be38342d DB |
117 | expectation_lookup(struct hmap *alg_expectations, const struct conn_key *key, |
118 | uint32_t basis, bool src_ip_wc); | |
bd5e81a0 DB |
119 | |
120 | static int | |
121 | repl_ftp_v4_addr(struct dp_packet *pkt, ovs_be32 v4_addr_rep, | |
122 | char *ftp_data_v4_start, | |
cd7c99a6 | 123 | size_t addr_offset_from_ftp_data_start, size_t addr_size); |
bd5e81a0 DB |
124 | |
125 | static enum ftp_ctl_pkt | |
126 | process_ftp_ctl_v4(struct conntrack *ct, | |
127 | struct dp_packet *pkt, | |
128 | const struct conn *conn_for_expectation, | |
4417ca3d | 129 | ovs_be32 *v4_addr_rep, |
bd5e81a0 | 130 | char **ftp_data_v4_start, |
cd7c99a6 DB |
131 | size_t *addr_offset_from_ftp_data_start, |
132 | size_t *addr_size); | |
bd5e81a0 DB |
133 | |
134 | static enum ftp_ctl_pkt | |
135 | detect_ftp_ctl_type(const struct conn_lookup_ctx *ctx, | |
136 | struct dp_packet *pkt); | |
137 | ||
4417ca3d | 138 | static void |
967bb5c5 | 139 | expectation_clean(struct conntrack *ct, const struct conn_key *master_key); |
4417ca3d | 140 | |
94e71143 DB |
141 | static struct ct_l4_proto *l4_protos[] = { |
142 | [IPPROTO_TCP] = &ct_proto_tcp, | |
143 | [IPPROTO_UDP] = &ct_proto_other, | |
144 | [IPPROTO_ICMP] = &ct_proto_icmp4, | |
145 | [IPPROTO_ICMPV6] = &ct_proto_icmp6, | |
146 | }; | |
147 | ||
bd5e81a0 DB |
148 | static void |
149 | handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
967bb5c5 DB |
150 | struct dp_packet *pkt, struct conn *ec, long long now, |
151 | enum ftp_ctl_pkt ftp_ctl, bool nat); | |
bd5e81a0 | 152 | |
7be77cb0 DB |
153 | static void |
154 | handle_tftp_ctl(struct conntrack *ct, | |
94e71143 | 155 | const struct conn_lookup_ctx *ctx OVS_UNUSED, |
967bb5c5 DB |
156 | struct dp_packet *pkt, struct conn *conn_for_expectation, |
157 | long long now OVS_UNUSED, enum ftp_ctl_pkt ftp_ctl OVS_UNUSED, | |
158 | bool nat OVS_UNUSED); | |
94e71143 DB |
159 | |
160 | typedef void (*alg_helper)(struct conntrack *ct, | |
161 | const struct conn_lookup_ctx *ctx, | |
162 | struct dp_packet *pkt, | |
967bb5c5 | 163 | struct conn *conn_for_expectation, |
94e71143 DB |
164 | long long now, enum ftp_ctl_pkt ftp_ctl, |
165 | bool nat); | |
166 | ||
167 | static alg_helper alg_helpers[] = { | |
168 | [CT_ALG_CTL_NONE] = NULL, | |
169 | [CT_ALG_CTL_FTP] = handle_ftp_ctl, | |
170 | [CT_ALG_CTL_TFTP] = handle_tftp_ctl, | |
a489b168 DDP |
171 | }; |
172 | ||
173 | long long ct_timeout_val[] = { | |
174 | #define CT_TIMEOUT(NAME, VAL) [CT_TM_##NAME] = VAL, | |
175 | CT_TIMEOUTS | |
176 | #undef CT_TIMEOUT | |
177 | }; | |
178 | ||
bd5e81a0 DB |
179 | /* The maximum TCP or UDP port number. */ |
180 | #define CT_MAX_L4_PORT 65535 | |
bd5e81a0 DB |
181 | /* String buffer used for parsing FTP string messages. |
182 | * This is sized about twice what is needed to leave some | |
183 | * margin of error. */ | |
184 | #define LARGEST_FTP_MSG_OF_INTEREST 128 | |
185 | /* FTP port string used in active mode. */ | |
186 | #define FTP_PORT_CMD "PORT" | |
187 | /* FTP pasv string used in passive mode. */ | |
188 | #define FTP_PASV_REPLY_CODE "227" | |
189 | /* Maximum decimal digits for port in FTP command. | |
190 | * The port is represented as two 3 digit numbers with the | |
191 | * high part a multiple of 256. */ | |
192 | #define MAX_FTP_PORT_DGTS 3 | |
193 | ||
194 | /* FTP extension EPRT string used for active mode. */ | |
195 | #define FTP_EPRT_CMD "EPRT" | |
196 | /* FTP extension EPSV string used for passive mode. */ | |
197 | #define FTP_EPSV_REPLY "EXTENDED PASSIVE" | |
198 | /* Maximum decimal digits for port in FTP extended command. */ | |
199 | #define MAX_EXT_FTP_PORT_DGTS 5 | |
200 | /* FTP extended command code for IPv6. */ | |
201 | #define FTP_AF_V6 '2' | |
202 | /* Used to indicate a wildcard L4 source port number for ALGs. | |
203 | * This is used for port numbers that we cannot predict in | |
204 | * expectations. */ | |
205 | #define ALG_WC_SRC_PORT 0 | |
206 | ||
a489b168 | 207 | /* If the total number of connections goes above this value, no new connections |
286de272 | 208 | * are accepted; this is for CT_CONN_TYPE_DEFAULT connections. */ |
a489b168 DDP |
209 | #define DEFAULT_N_CONN_LIMIT 3000000 |
210 | ||
5ed7a0b4 DB |
211 | /* Does a member by member comparison of two conn_keys; this |
212 | * function must be kept in sync with struct conn_key; returns 0 | |
213 | * if the keys are equal or 1 if the keys are not equal. */ | |
214 | static int | |
215 | conn_key_cmp(const struct conn_key *key1, const struct conn_key *key2) | |
216 | { | |
217 | if (!memcmp(&key1->src.addr, &key2->src.addr, sizeof key1->src.addr) && | |
218 | !memcmp(&key1->dst.addr, &key2->dst.addr, sizeof key1->dst.addr) && | |
219 | (key1->src.icmp_id == key2->src.icmp_id) && | |
220 | (key1->src.icmp_type == key2->src.icmp_type) && | |
221 | (key1->src.icmp_code == key2->src.icmp_code) && | |
222 | (key1->dst.icmp_id == key2->dst.icmp_id) && | |
223 | (key1->dst.icmp_type == key2->dst.icmp_type) && | |
224 | (key1->dst.icmp_code == key2->dst.icmp_code) && | |
225 | (key1->dl_type == key2->dl_type) && | |
226 | (key1->zone == key2->zone) && | |
227 | (key1->nw_proto == key2->nw_proto)) { | |
228 | ||
229 | return 0; | |
230 | } | |
231 | return 1; | |
232 | } | |
233 | ||
d8682ee5 | 234 | static void |
dec0dbbc DB |
235 | ct_print_conn_info(const struct conn *c, const char *log_msg, |
236 | enum vlog_level vll, bool force, bool rl_on) | |
66f400f5 DB |
237 | { |
238 | #define CT_VLOG(RL_ON, LEVEL, ...) \ | |
239 | do { \ | |
240 | if (RL_ON) { \ | |
241 | static struct vlog_rate_limit rl_ = VLOG_RATE_LIMIT_INIT(5, 5); \ | |
242 | vlog_rate_limit(&this_module, LEVEL, &rl_, __VA_ARGS__); \ | |
243 | } else { \ | |
244 | vlog(&this_module, LEVEL, __VA_ARGS__); \ | |
245 | } \ | |
246 | } while (0) | |
247 | ||
248 | if (OVS_UNLIKELY(force || vlog_is_enabled(&this_module, vll))) { | |
249 | if (c->key.dl_type == htons(ETH_TYPE_IP)) { | |
250 | CT_VLOG(rl_on, vll, "%s: src ip "IP_FMT" dst ip "IP_FMT" rev src " | |
251 | "ip "IP_FMT" rev dst ip "IP_FMT" src/dst ports " | |
252 | "%"PRIu16"/%"PRIu16" rev src/dst ports " | |
253 | "%"PRIu16"/%"PRIu16" zone/rev zone " | |
254 | "%"PRIu16"/%"PRIu16" nw_proto/rev nw_proto " | |
255 | "%"PRIu8"/%"PRIu8, log_msg, | |
cda1b109 DB |
256 | IP_ARGS(c->key.src.addr.ipv4), |
257 | IP_ARGS(c->key.dst.addr.ipv4), | |
258 | IP_ARGS(c->rev_key.src.addr.ipv4), | |
259 | IP_ARGS(c->rev_key.dst.addr.ipv4), | |
66f400f5 DB |
260 | ntohs(c->key.src.port), ntohs(c->key.dst.port), |
261 | ntohs(c->rev_key.src.port), ntohs(c->rev_key.dst.port), | |
262 | c->key.zone, c->rev_key.zone, c->key.nw_proto, | |
263 | c->rev_key.nw_proto); | |
264 | } else { | |
265 | char ip6_s[INET6_ADDRSTRLEN]; | |
266 | inet_ntop(AF_INET6, &c->key.src.addr.ipv6, ip6_s, sizeof ip6_s); | |
267 | char ip6_d[INET6_ADDRSTRLEN]; | |
268 | inet_ntop(AF_INET6, &c->key.dst.addr.ipv6, ip6_d, sizeof ip6_d); | |
269 | char ip6_rs[INET6_ADDRSTRLEN]; | |
270 | inet_ntop(AF_INET6, &c->rev_key.src.addr.ipv6, ip6_rs, | |
271 | sizeof ip6_rs); | |
272 | char ip6_rd[INET6_ADDRSTRLEN]; | |
273 | inet_ntop(AF_INET6, &c->rev_key.dst.addr.ipv6, ip6_rd, | |
274 | sizeof ip6_rd); | |
275 | ||
276 | CT_VLOG(rl_on, vll, "%s: src ip %s dst ip %s rev src ip %s" | |
277 | " rev dst ip %s src/dst ports %"PRIu16"/%"PRIu16 | |
278 | " rev src/dst ports %"PRIu16"/%"PRIu16" zone/rev zone " | |
279 | "%"PRIu16"/%"PRIu16" nw_proto/rev nw_proto " | |
280 | "%"PRIu8"/%"PRIu8, log_msg, ip6_s, ip6_d, ip6_rs, | |
281 | ip6_rd, ntohs(c->key.src.port), ntohs(c->key.dst.port), | |
282 | ntohs(c->rev_key.src.port), ntohs(c->rev_key.dst.port), | |
283 | c->key.zone, c->rev_key.zone, c->key.nw_proto, | |
284 | c->rev_key.nw_proto); | |
285 | } | |
286 | } | |
287 | } | |
288 | ||
a489b168 DDP |
289 | /* Initializes the connection tracker 'ct'. The caller is responsible for |
290 | * calling 'conntrack_destroy()', when the instance is not needed anymore */ | |
57593fd2 DB |
291 | struct conntrack * |
292 | conntrack_init(void) | |
a489b168 | 293 | { |
57593fd2 DB |
294 | struct conntrack *ct = xzalloc(sizeof *ct); |
295 | ||
967bb5c5 DB |
296 | ovs_rwlock_init(&ct->resources_lock); |
297 | ovs_rwlock_wrlock(&ct->resources_lock); | |
bd5e81a0 | 298 | hmap_init(&ct->alg_expectations); |
4417ca3d | 299 | hindex_init(&ct->alg_expectation_refs); |
967bb5c5 | 300 | ovs_rwlock_unlock(&ct->resources_lock); |
a489b168 | 301 | |
967bb5c5 DB |
302 | ovs_mutex_init_adaptive(&ct->ct_lock); |
303 | ovs_mutex_lock(&ct->ct_lock); | |
304 | cmap_init(&ct->conns); | |
305 | for (unsigned i = 0; i < ARRAY_SIZE(ct->exp_lists); i++) { | |
306 | ovs_list_init(&ct->exp_lists[i]); | |
a489b168 | 307 | } |
967bb5c5 DB |
308 | ovs_mutex_unlock(&ct->ct_lock); |
309 | ||
a489b168 DDP |
310 | ct->hash_basis = random_uint32(); |
311 | atomic_count_init(&ct->n_conn, 0); | |
312 | atomic_init(&ct->n_conn_limit, DEFAULT_N_CONN_LIMIT); | |
e6ef6cc6 DDP |
313 | latch_init(&ct->clean_thread_exit); |
314 | ct->clean_thread = ovs_thread_create("ct_clean", clean_thread_main, ct); | |
4ea96698 | 315 | ct->ipf = ipf_init(); |
57593fd2 DB |
316 | |
317 | return ct; | |
a489b168 DDP |
318 | } |
319 | ||
967bb5c5 DB |
320 | static void |
321 | conn_clean_cmn(struct conntrack *ct, struct conn *conn) | |
322 | OVS_REQUIRES(ct->ct_lock) | |
323 | { | |
324 | if (conn->alg) { | |
325 | expectation_clean(ct, &conn->key); | |
326 | } | |
327 | ||
328 | uint32_t hash = conn_key_hash(&conn->key, ct->hash_basis); | |
329 | cmap_remove(&ct->conns, &conn->cm_node, hash); | |
330 | } | |
331 | ||
332 | /* Must be called with 'conn' of 'conn_type' CT_CONN_TYPE_DEFAULT. Also | |
333 | * removes the associated nat 'conn' from the lookup datastructures. */ | |
334 | static void | |
335 | conn_clean(struct conntrack *ct, struct conn *conn) | |
336 | OVS_REQUIRES(ct->ct_lock) | |
337 | { | |
338 | ovs_assert(conn->conn_type == CT_CONN_TYPE_DEFAULT); | |
339 | ||
340 | conn_clean_cmn(ct, conn); | |
341 | if (conn->nat_conn) { | |
342 | uint32_t hash = conn_key_hash(&conn->nat_conn->key, ct->hash_basis); | |
343 | cmap_remove(&ct->conns, &conn->nat_conn->cm_node, hash); | |
344 | } | |
345 | ovs_list_remove(&conn->exp_node); | |
5f918a8a | 346 | conn->cleaned = true; |
967bb5c5 DB |
347 | ovsrcu_postpone(delete_conn, conn); |
348 | atomic_count_dec(&ct->n_conn); | |
349 | } | |
350 | ||
351 | static void | |
352 | conn_clean_one(struct conntrack *ct, struct conn *conn) | |
353 | OVS_REQUIRES(ct->ct_lock) | |
354 | { | |
355 | conn_clean_cmn(ct, conn); | |
356 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { | |
357 | ovs_list_remove(&conn->exp_node); | |
5f918a8a | 358 | conn->cleaned = true; |
967bb5c5 DB |
359 | atomic_count_dec(&ct->n_conn); |
360 | } | |
361 | ovsrcu_postpone(delete_conn_one, conn); | |
362 | } | |
363 | ||
364 | /* Destroys the connection tracker 'ct' and frees all the allocated memory. | |
365 | * The caller of this function must already have shut down packet input | |
366 | * and PMD threads (which would have been quiesced). */ | |
a489b168 DDP |
367 | void |
368 | conntrack_destroy(struct conntrack *ct) | |
369 | { | |
967bb5c5 | 370 | struct conn *conn; |
e6ef6cc6 DDP |
371 | latch_set(&ct->clean_thread_exit); |
372 | pthread_join(ct->clean_thread, NULL); | |
373 | latch_destroy(&ct->clean_thread_exit); | |
a489b168 | 374 | |
967bb5c5 DB |
375 | ovs_mutex_lock(&ct->ct_lock); |
376 | CMAP_FOR_EACH (conn, cm_node, &ct->conns) { | |
377 | conn_clean_one(ct, conn); | |
a489b168 | 378 | } |
967bb5c5 DB |
379 | cmap_destroy(&ct->conns); |
380 | ovs_mutex_unlock(&ct->ct_lock); | |
381 | ovs_mutex_destroy(&ct->ct_lock); | |
bd5e81a0 | 382 | |
967bb5c5 | 383 | ovs_rwlock_wrlock(&ct->resources_lock); |
bd5e81a0 DB |
384 | struct alg_exp_node *alg_exp_node; |
385 | HMAP_FOR_EACH_POP (alg_exp_node, node, &ct->alg_expectations) { | |
386 | free(alg_exp_node); | |
387 | } | |
bd5e81a0 | 388 | hmap_destroy(&ct->alg_expectations); |
4417ca3d | 389 | hindex_destroy(&ct->alg_expectation_refs); |
967bb5c5 DB |
390 | ovs_rwlock_unlock(&ct->resources_lock); |
391 | ovs_rwlock_destroy(&ct->resources_lock); | |
392 | ||
4ea96698 | 393 | ipf_destroy(ct->ipf); |
21ffe409 | 394 | free(ct); |
a489b168 DDP |
395 | } |
396 | \f | |
967bb5c5 DB |
397 | |
398 | static bool | |
399 | conn_key_lookup(struct conntrack *ct, const struct conn_key *key, | |
400 | uint32_t hash, long long now, struct conn **conn_out, | |
401 | bool *reply) | |
a489b168 | 402 | { |
967bb5c5 DB |
403 | struct conn *conn; |
404 | bool found = false; | |
405 | ||
406 | CMAP_FOR_EACH_WITH_HASH (conn, cm_node, hash, &ct->conns) { | |
407 | if (!conn_key_cmp(&conn->key, key) && !conn_expired(conn, now)) { | |
408 | found = true; | |
409 | if (reply) { | |
410 | *reply = false; | |
411 | } | |
412 | break; | |
413 | } | |
414 | if (!conn_key_cmp(&conn->rev_key, key) && !conn_expired(conn, now)) { | |
415 | found = true; | |
416 | if (reply) { | |
417 | *reply = true; | |
418 | } | |
419 | break; | |
420 | } | |
421 | } | |
a489b168 | 422 | |
967bb5c5 DB |
423 | if (found && conn_out) { |
424 | *conn_out = conn; | |
425 | } else if (conn_out) { | |
426 | *conn_out = NULL; | |
427 | } | |
428 | return found; | |
a489b168 DDP |
429 | } |
430 | ||
431 | static void | |
286de272 | 432 | write_ct_md(struct dp_packet *pkt, uint16_t zone, const struct conn *conn, |
bd5e81a0 | 433 | const struct conn_key *key, const struct alg_exp_node *alg_exp) |
a489b168 | 434 | { |
286de272 | 435 | pkt->md.ct_state |= CS_TRACKED; |
a489b168 | 436 | pkt->md.ct_zone = zone; |
967bb5c5 DB |
437 | |
438 | if (conn) { | |
439 | ovs_mutex_lock(&conn->lock); | |
440 | pkt->md.ct_mark = conn->mark; | |
441 | pkt->md.ct_label = conn->label; | |
442 | ovs_mutex_unlock(&conn->lock); | |
443 | } else { | |
444 | pkt->md.ct_mark = 0; | |
445 | pkt->md.ct_label = OVS_U128_ZERO; | |
446 | } | |
daf4d3c1 JR |
447 | |
448 | /* Use the original direction tuple if we have it. */ | |
449 | if (conn) { | |
bd5e81a0 DB |
450 | if (conn->alg_related) { |
451 | key = &conn->master_key; | |
452 | } else { | |
453 | key = &conn->key; | |
454 | } | |
455 | } else if (alg_exp) { | |
456 | pkt->md.ct_mark = alg_exp->master_mark; | |
457 | pkt->md.ct_label = alg_exp->master_label; | |
458 | key = &alg_exp->master_key; | |
daf4d3c1 | 459 | } |
dec0dbbc | 460 | |
daf4d3c1 | 461 | pkt->md.ct_orig_tuple_ipv6 = false; |
dec0dbbc | 462 | |
daf4d3c1 JR |
463 | if (key) { |
464 | if (key->dl_type == htons(ETH_TYPE_IP)) { | |
465 | pkt->md.ct_orig_tuple.ipv4 = (struct ovs_key_ct_tuple_ipv4) { | |
cda1b109 DB |
466 | key->src.addr.ipv4, |
467 | key->dst.addr.ipv4, | |
daf4d3c1 JR |
468 | key->nw_proto != IPPROTO_ICMP |
469 | ? key->src.port : htons(key->src.icmp_type), | |
470 | key->nw_proto != IPPROTO_ICMP | |
471 | ? key->dst.port : htons(key->src.icmp_code), | |
472 | key->nw_proto, | |
473 | }; | |
286de272 | 474 | } else { |
daf4d3c1 JR |
475 | pkt->md.ct_orig_tuple_ipv6 = true; |
476 | pkt->md.ct_orig_tuple.ipv6 = (struct ovs_key_ct_tuple_ipv6) { | |
cda1b109 DB |
477 | key->src.addr.ipv6, |
478 | key->dst.addr.ipv6, | |
daf4d3c1 JR |
479 | key->nw_proto != IPPROTO_ICMPV6 |
480 | ? key->src.port : htons(key->src.icmp_type), | |
481 | key->nw_proto != IPPROTO_ICMPV6 | |
482 | ? key->dst.port : htons(key->src.icmp_code), | |
483 | key->nw_proto, | |
484 | }; | |
485 | } | |
486 | } else { | |
487 | memset(&pkt->md.ct_orig_tuple, 0, sizeof pkt->md.ct_orig_tuple); | |
488 | } | |
bd5e81a0 DB |
489 | } |
490 | ||
491 | static uint8_t | |
492 | get_ip_proto(const struct dp_packet *pkt) | |
493 | { | |
494 | uint8_t ip_proto; | |
495 | struct eth_header *l2 = dp_packet_eth(pkt); | |
496 | if (l2->eth_type == htons(ETH_TYPE_IPV6)) { | |
497 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
498 | ip_proto = nh6->ip6_ctlun.ip6_un1.ip6_un1_nxt; | |
499 | } else { | |
500 | struct ip_header *l3_hdr = dp_packet_l3(pkt); | |
501 | ip_proto = l3_hdr->ip_proto; | |
502 | } | |
286de272 | 503 | |
bd5e81a0 DB |
504 | return ip_proto; |
505 | } | |
506 | ||
507 | static bool | |
94e71143 | 508 | is_ftp_ctl(const enum ct_alg_ctl_type ct_alg_ctl) |
bd5e81a0 | 509 | { |
94e71143 | 510 | return ct_alg_ctl == CT_ALG_CTL_FTP; |
bd5e81a0 DB |
511 | } |
512 | ||
94e71143 | 513 | static enum ct_alg_ctl_type |
bd7d93f8 DB |
514 | get_alg_ctl_type(const struct dp_packet *pkt, ovs_be16 tp_src, ovs_be16 tp_dst, |
515 | const char *helper) | |
7be77cb0 | 516 | { |
94e71143 DB |
517 | /* CT_IPPORT_FTP/TFTP is used because IPPORT_FTP/TFTP in not defined |
518 | * in OSX, at least in in.h. Since these values will never change, remove | |
7be77cb0 | 519 | * the external dependency. */ |
94e71143 DB |
520 | enum { CT_IPPORT_FTP = 21 }; |
521 | enum { CT_IPPORT_TFTP = 69 }; | |
bd7d93f8 DB |
522 | uint8_t ip_proto = get_ip_proto(pkt); |
523 | struct udp_header *uh = dp_packet_l4(pkt); | |
524 | struct tcp_header *th = dp_packet_l4(pkt); | |
525 | ovs_be16 ftp_src_port = htons(CT_IPPORT_FTP); | |
526 | ovs_be16 ftp_dst_port = htons(CT_IPPORT_FTP); | |
527 | ovs_be16 tftp_dst_port = htons(CT_IPPORT_TFTP); | |
528 | ||
529 | if (OVS_UNLIKELY(tp_dst)) { | |
530 | if (helper && !strncmp(helper, "ftp", strlen("ftp"))) { | |
531 | ftp_dst_port = tp_dst; | |
532 | } else if (helper && !strncmp(helper, "tftp", strlen("tftp"))) { | |
533 | tftp_dst_port = tp_dst; | |
534 | } | |
535 | } else if (OVS_UNLIKELY(tp_src)) { | |
536 | if (helper && !strncmp(helper, "ftp", strlen("ftp"))) { | |
537 | ftp_src_port = tp_src; | |
538 | } | |
539 | } | |
7be77cb0 | 540 | |
bd7d93f8 | 541 | if (ip_proto == IPPROTO_UDP && uh->udp_dst == tftp_dst_port) { |
94e71143 DB |
542 | return CT_ALG_CTL_TFTP; |
543 | } else if (ip_proto == IPPROTO_TCP && | |
bd7d93f8 | 544 | (th->tcp_src == ftp_src_port || th->tcp_dst == ftp_dst_port)) { |
94e71143 DB |
545 | return CT_ALG_CTL_FTP; |
546 | } | |
547 | return CT_ALG_CTL_NONE; | |
548 | } | |
549 | ||
be38342d DB |
550 | static bool |
551 | alg_src_ip_wc(enum ct_alg_ctl_type alg_ctl_type) | |
552 | { | |
553 | if (alg_ctl_type == CT_ALG_CTL_SIP) { | |
554 | return true; | |
555 | } | |
556 | return false; | |
557 | } | |
558 | ||
94e71143 DB |
559 | static void |
560 | handle_alg_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
561 | struct dp_packet *pkt, enum ct_alg_ctl_type ct_alg_ctl, | |
967bb5c5 | 562 | struct conn *conn, long long now, bool nat) |
94e71143 DB |
563 | { |
564 | /* ALG control packet handling with expectation creation. */ | |
3a2a425b | 565 | if (OVS_UNLIKELY(alg_helpers[ct_alg_ctl] && conn && conn->alg)) { |
967bb5c5 DB |
566 | ovs_mutex_lock(&conn->lock); |
567 | alg_helpers[ct_alg_ctl](ct, ctx, pkt, conn, now, CT_FTP_CTL_INTEREST, | |
568 | nat); | |
569 | ovs_mutex_unlock(&conn->lock); | |
94e71143 | 570 | } |
7be77cb0 DB |
571 | } |
572 | ||
286de272 DB |
573 | static void |
574 | pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
575 | { | |
576 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
577 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
578 | struct tcp_header *th = dp_packet_l4(pkt); | |
579 | packet_set_tcp_port(pkt, conn->rev_key.dst.port, th->tcp_dst); | |
580 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
581 | struct udp_header *uh = dp_packet_l4(pkt); | |
582 | packet_set_udp_port(pkt, conn->rev_key.dst.port, uh->udp_dst); | |
583 | } | |
584 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
585 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
586 | struct tcp_header *th = dp_packet_l4(pkt); | |
587 | packet_set_tcp_port(pkt, th->tcp_src, conn->rev_key.src.port); | |
588 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
589 | struct udp_header *uh = dp_packet_l4(pkt); | |
590 | packet_set_udp_port(pkt, uh->udp_src, conn->rev_key.src.port); | |
591 | } | |
592 | } | |
593 | } | |
594 | ||
595 | static void | |
596 | nat_packet(struct dp_packet *pkt, const struct conn *conn, bool related) | |
597 | { | |
598 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
599 | pkt->md.ct_state |= CS_SRC_NAT; | |
600 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
601 | struct ip_header *nh = dp_packet_l3(pkt); | |
602 | packet_set_ipv4_addr(pkt, &nh->ip_src, | |
cda1b109 | 603 | conn->rev_key.dst.addr.ipv4); |
286de272 DB |
604 | } else { |
605 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
606 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
607 | nh6->ip6_src.be32, | |
cda1b109 | 608 | &conn->rev_key.dst.addr.ipv6, true); |
286de272 DB |
609 | } |
610 | if (!related) { | |
611 | pat_packet(pkt, conn); | |
612 | } | |
613 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
614 | pkt->md.ct_state |= CS_DST_NAT; | |
615 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
616 | struct ip_header *nh = dp_packet_l3(pkt); | |
617 | packet_set_ipv4_addr(pkt, &nh->ip_dst, | |
cda1b109 | 618 | conn->rev_key.src.addr.ipv4); |
286de272 DB |
619 | } else { |
620 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
621 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
622 | nh6->ip6_dst.be32, | |
cda1b109 | 623 | &conn->rev_key.src.addr.ipv6, true); |
286de272 DB |
624 | } |
625 | if (!related) { | |
626 | pat_packet(pkt, conn); | |
627 | } | |
628 | } | |
629 | } | |
630 | ||
631 | static void | |
632 | un_pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
633 | { | |
634 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
635 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
636 | struct tcp_header *th = dp_packet_l4(pkt); | |
637 | packet_set_tcp_port(pkt, th->tcp_src, conn->key.src.port); | |
638 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
639 | struct udp_header *uh = dp_packet_l4(pkt); | |
640 | packet_set_udp_port(pkt, uh->udp_src, conn->key.src.port); | |
641 | } | |
642 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
643 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
644 | struct tcp_header *th = dp_packet_l4(pkt); | |
645 | packet_set_tcp_port(pkt, conn->key.dst.port, th->tcp_dst); | |
646 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
647 | struct udp_header *uh = dp_packet_l4(pkt); | |
648 | packet_set_udp_port(pkt, conn->key.dst.port, uh->udp_dst); | |
649 | } | |
650 | } | |
651 | } | |
652 | ||
edd1bef4 DB |
653 | static void |
654 | reverse_pat_packet(struct dp_packet *pkt, const struct conn *conn) | |
655 | { | |
656 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
657 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
658 | struct tcp_header *th_in = dp_packet_l4(pkt); | |
659 | packet_set_tcp_port(pkt, conn->key.src.port, | |
660 | th_in->tcp_dst); | |
661 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
662 | struct udp_header *uh_in = dp_packet_l4(pkt); | |
663 | packet_set_udp_port(pkt, conn->key.src.port, | |
664 | uh_in->udp_dst); | |
665 | } | |
666 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
667 | if (conn->key.nw_proto == IPPROTO_TCP) { | |
668 | struct tcp_header *th_in = dp_packet_l4(pkt); | |
669 | packet_set_tcp_port(pkt, th_in->tcp_src, | |
670 | conn->key.dst.port); | |
671 | } else if (conn->key.nw_proto == IPPROTO_UDP) { | |
672 | struct udp_header *uh_in = dp_packet_l4(pkt); | |
673 | packet_set_udp_port(pkt, uh_in->udp_src, | |
674 | conn->key.dst.port); | |
675 | } | |
676 | } | |
677 | } | |
678 | ||
679 | static void | |
680 | reverse_nat_packet(struct dp_packet *pkt, const struct conn *conn) | |
681 | { | |
682 | char *tail = dp_packet_tail(pkt); | |
683 | char pad = dp_packet_l2_pad_size(pkt); | |
684 | struct conn_key inner_key; | |
685 | const char *inner_l4 = NULL; | |
686 | uint16_t orig_l3_ofs = pkt->l3_ofs; | |
687 | uint16_t orig_l4_ofs = pkt->l4_ofs; | |
688 | ||
689 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
690 | struct ip_header *nh = dp_packet_l3(pkt); | |
691 | struct icmp_header *icmp = dp_packet_l4(pkt); | |
692 | struct ip_header *inner_l3 = (struct ip_header *) (icmp + 1); | |
bd5e81a0 DB |
693 | extract_l3_ipv4(&inner_key, inner_l3, tail - ((char *)inner_l3) - pad, |
694 | &inner_l4, false); | |
edd1bef4 DB |
695 | pkt->l3_ofs += (char *) inner_l3 - (char *) nh; |
696 | pkt->l4_ofs += inner_l4 - (char *) icmp; | |
697 | ||
698 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
699 | packet_set_ipv4_addr(pkt, &inner_l3->ip_src, | |
cda1b109 | 700 | conn->key.src.addr.ipv4); |
edd1bef4 DB |
701 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { |
702 | packet_set_ipv4_addr(pkt, &inner_l3->ip_dst, | |
cda1b109 | 703 | conn->key.dst.addr.ipv4); |
edd1bef4 | 704 | } |
dec0dbbc | 705 | |
edd1bef4 DB |
706 | reverse_pat_packet(pkt, conn); |
707 | icmp->icmp_csum = 0; | |
708 | icmp->icmp_csum = csum(icmp, tail - (char *) icmp - pad); | |
709 | } else { | |
710 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
711 | struct icmp6_error_header *icmp6 = dp_packet_l4(pkt); | |
712 | struct ovs_16aligned_ip6_hdr *inner_l3_6 = | |
713 | (struct ovs_16aligned_ip6_hdr *) (icmp6 + 1); | |
714 | extract_l3_ipv6(&inner_key, inner_l3_6, | |
715 | tail - ((char *)inner_l3_6) - pad, | |
716 | &inner_l4); | |
717 | pkt->l3_ofs += (char *) inner_l3_6 - (char *) nh6; | |
718 | pkt->l4_ofs += inner_l4 - (char *) icmp6; | |
719 | ||
720 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
721 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
722 | inner_l3_6->ip6_src.be32, | |
cda1b109 | 723 | &conn->key.src.addr.ipv6, true); |
edd1bef4 DB |
724 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { |
725 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
726 | inner_l3_6->ip6_dst.be32, | |
cda1b109 | 727 | &conn->key.dst.addr.ipv6, true); |
edd1bef4 DB |
728 | } |
729 | reverse_pat_packet(pkt, conn); | |
edd1bef4 | 730 | icmp6->icmp6_base.icmp6_cksum = 0; |
76d85771 DB |
731 | icmp6->icmp6_base.icmp6_cksum = packet_csum_upperlayer6(nh6, icmp6, |
732 | IPPROTO_ICMPV6, tail - (char *) icmp6 - pad); | |
edd1bef4 DB |
733 | } |
734 | pkt->l3_ofs = orig_l3_ofs; | |
735 | pkt->l4_ofs = orig_l4_ofs; | |
736 | } | |
737 | ||
286de272 DB |
738 | static void |
739 | un_nat_packet(struct dp_packet *pkt, const struct conn *conn, | |
740 | bool related) | |
741 | { | |
742 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
743 | pkt->md.ct_state |= CS_DST_NAT; | |
744 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
745 | struct ip_header *nh = dp_packet_l3(pkt); | |
746 | packet_set_ipv4_addr(pkt, &nh->ip_dst, | |
cda1b109 | 747 | conn->key.src.addr.ipv4); |
286de272 DB |
748 | } else { |
749 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
750 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
751 | nh6->ip6_dst.be32, | |
cda1b109 | 752 | &conn->key.src.addr.ipv6, true); |
286de272 | 753 | } |
edd1bef4 DB |
754 | |
755 | if (OVS_UNLIKELY(related)) { | |
756 | reverse_nat_packet(pkt, conn); | |
757 | } else { | |
286de272 DB |
758 | un_pat_packet(pkt, conn); |
759 | } | |
760 | } else if (conn->nat_info->nat_action & NAT_ACTION_DST) { | |
761 | pkt->md.ct_state |= CS_SRC_NAT; | |
762 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
763 | struct ip_header *nh = dp_packet_l3(pkt); | |
764 | packet_set_ipv4_addr(pkt, &nh->ip_src, | |
cda1b109 | 765 | conn->key.dst.addr.ipv4); |
286de272 DB |
766 | } else { |
767 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); | |
768 | packet_set_ipv6_addr(pkt, conn->key.nw_proto, | |
769 | nh6->ip6_src.be32, | |
cda1b109 | 770 | &conn->key.dst.addr.ipv6, true); |
286de272 | 771 | } |
edd1bef4 DB |
772 | |
773 | if (OVS_UNLIKELY(related)) { | |
774 | reverse_nat_packet(pkt, conn); | |
775 | } else { | |
286de272 DB |
776 | un_pat_packet(pkt, conn); |
777 | } | |
778 | } | |
779 | } | |
780 | ||
bd5e81a0 | 781 | static void |
967bb5c5 | 782 | conn_seq_skew_set(struct conntrack *ct, const struct conn *conn_in, |
bd5e81a0 | 783 | long long now, int seq_skew, bool seq_skew_dir) |
967bb5c5 | 784 | OVS_NO_THREAD_SAFETY_ANALYSIS |
bd5e81a0 | 785 | { |
967bb5c5 DB |
786 | struct conn *conn; |
787 | bool reply; | |
788 | uint32_t hash = conn_key_hash(&conn_in->key, ct->hash_basis); | |
789 | ovs_mutex_unlock(&conn_in->lock); | |
790 | conn_key_lookup(ct, &conn_in->key, hash, now, &conn, &reply); | |
791 | ovs_mutex_lock(&conn_in->lock); | |
792 | ||
bd5e81a0 DB |
793 | if (conn && seq_skew) { |
794 | conn->seq_skew = seq_skew; | |
795 | conn->seq_skew_dir = seq_skew_dir; | |
796 | } | |
a720a7fa DB |
797 | } |
798 | ||
3a2a425b DB |
799 | static bool |
800 | ct_verify_helper(const char *helper, enum ct_alg_ctl_type ct_alg_ctl) | |
801 | { | |
802 | if (ct_alg_ctl == CT_ALG_CTL_NONE) { | |
803 | return true; | |
804 | } else if (helper) { | |
805 | if ((ct_alg_ctl == CT_ALG_CTL_FTP) && | |
806 | !strncmp(helper, "ftp", strlen("ftp"))) { | |
807 | return true; | |
808 | } else if ((ct_alg_ctl == CT_ALG_CTL_TFTP) && | |
809 | !strncmp(helper, "tftp", strlen("tftp"))) { | |
810 | return true; | |
811 | } else { | |
812 | return false; | |
813 | } | |
814 | } else { | |
815 | return false; | |
816 | } | |
817 | } | |
818 | ||
a489b168 DDP |
819 | static struct conn * |
820 | conn_not_found(struct conntrack *ct, struct dp_packet *pkt, | |
286de272 DB |
821 | struct conn_lookup_ctx *ctx, bool commit, long long now, |
822 | const struct nat_action_info_t *nat_action_info, | |
967bb5c5 | 823 | const char *helper, const struct alg_exp_node *alg_exp, |
3a2a425b | 824 | enum ct_alg_ctl_type ct_alg_ctl) |
967bb5c5 | 825 | OVS_REQUIRES(ct->ct_lock) |
a489b168 | 826 | { |
a489b168 | 827 | struct conn *nc = NULL; |
967bb5c5 | 828 | struct conn *nat_conn = NULL; |
a489b168 DDP |
829 | |
830 | if (!valid_new(pkt, &ctx->key)) { | |
286de272 | 831 | pkt->md.ct_state = CS_INVALID; |
a489b168 DDP |
832 | return nc; |
833 | } | |
dec0dbbc | 834 | |
286de272 | 835 | pkt->md.ct_state = CS_NEW; |
dec0dbbc | 836 | |
bd5e81a0 DB |
837 | if (alg_exp) { |
838 | pkt->md.ct_state |= CS_RELATED; | |
839 | } | |
a489b168 DDP |
840 | |
841 | if (commit) { | |
842 | unsigned int n_conn_limit; | |
a489b168 | 843 | atomic_read_relaxed(&ct->n_conn_limit, &n_conn_limit); |
a489b168 DDP |
844 | if (atomic_count_get(&ct->n_conn) >= n_conn_limit) { |
845 | COVERAGE_INC(conntrack_full); | |
846 | return nc; | |
847 | } | |
848 | ||
967bb5c5 | 849 | nc = new_conn(ct, pkt, &ctx->key, now); |
a720a7fa | 850 | memcpy(&nc->key, &ctx->key, sizeof nc->key); |
82b9ac94 | 851 | memcpy(&nc->rev_key, &nc->key, sizeof nc->rev_key); |
286de272 | 852 | conn_key_reverse(&nc->rev_key); |
a489b168 | 853 | |
3a2a425b DB |
854 | if (ct_verify_helper(helper, ct_alg_ctl)) { |
855 | nc->alg = nullable_xstrdup(helper); | |
bd5e81a0 DB |
856 | } |
857 | ||
858 | if (alg_exp) { | |
859 | nc->alg_related = true; | |
860 | nc->mark = alg_exp->master_mark; | |
861 | nc->label = alg_exp->master_label; | |
862 | nc->master_key = alg_exp->master_key; | |
863 | } | |
864 | ||
286de272 DB |
865 | if (nat_action_info) { |
866 | nc->nat_info = xmemdup(nat_action_info, sizeof *nc->nat_info); | |
967bb5c5 | 867 | nat_conn = xzalloc(sizeof *nat_conn); |
a489b168 | 868 | |
bd5e81a0 | 869 | if (alg_exp) { |
be38342d | 870 | if (alg_exp->nat_rpl_dst) { |
bd5e81a0 DB |
871 | nc->rev_key.dst.addr = alg_exp->alg_nat_repl_addr; |
872 | nc->nat_info->nat_action = NAT_ACTION_SRC; | |
873 | } else { | |
874 | nc->rev_key.src.addr = alg_exp->alg_nat_repl_addr; | |
875 | nc->nat_info->nat_action = NAT_ACTION_DST; | |
876 | } | |
bd5e81a0 | 877 | } else { |
967bb5c5 DB |
878 | memcpy(nat_conn, nc, sizeof *nat_conn); |
879 | bool nat_res = nat_select_range_tuple(ct, nc, nat_conn); | |
286de272 | 880 | |
bd5e81a0 DB |
881 | if (!nat_res) { |
882 | goto nat_res_exhaustion; | |
883 | } | |
286de272 | 884 | |
967bb5c5 DB |
885 | /* Update nc with nat adjustments made to nat_conn by |
886 | * nat_select_range_tuple(). */ | |
887 | memcpy(nc, nat_conn, sizeof *nc); | |
286de272 | 888 | } |
967bb5c5 | 889 | |
dbb597d3 | 890 | nat_packet(pkt, nc, ctx->icmp_related); |
967bb5c5 DB |
891 | memcpy(&nat_conn->key, &nc->rev_key, sizeof nat_conn->key); |
892 | memcpy(&nat_conn->rev_key, &nc->key, sizeof nat_conn->rev_key); | |
893 | nat_conn->conn_type = CT_CONN_TYPE_UN_NAT; | |
894 | nat_conn->nat_info = NULL; | |
895 | nat_conn->alg = NULL; | |
896 | nat_conn->nat_conn = NULL; | |
897 | uint32_t nat_hash = conn_key_hash(&nat_conn->key, ct->hash_basis); | |
898 | cmap_insert(&ct->conns, &nat_conn->cm_node, nat_hash); | |
899 | } | |
900 | ||
901 | nc->nat_conn = nat_conn; | |
902 | ovs_mutex_init_adaptive(&nc->lock); | |
903 | nc->conn_type = CT_CONN_TYPE_DEFAULT; | |
904 | cmap_insert(&ct->conns, &nc->cm_node, ctx->hash); | |
a489b168 | 905 | atomic_count_inc(&ct->n_conn); |
967bb5c5 | 906 | ctx->conn = nc; /* For completeness. */ |
a489b168 | 907 | } |
bd5e81a0 | 908 | |
a489b168 | 909 | return nc; |
bd5e81a0 | 910 | |
967bb5c5 DB |
911 | /* This would be a user error or a DOS attack. A user error is prevented |
912 | * by allocating enough combinations of NAT addresses when combined with | |
913 | * ephemeral ports. A DOS attack should be protected against with | |
914 | * firewall rules or a separate firewall. Also using zone partitioning | |
915 | * can limit DoS impact. */ | |
bd5e81a0 | 916 | nat_res_exhaustion: |
967bb5c5 DB |
917 | free(nat_conn); |
918 | ovs_list_remove(&nc->exp_node); | |
919 | delete_conn_cmn(nc); | |
bd5e81a0 DB |
920 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
921 | VLOG_WARN_RL(&rl, "Unable to NAT due to tuple space exhaustion - " | |
922 | "if DoS attack, use firewalling and/or zone partitioning."); | |
923 | return NULL; | |
a489b168 DDP |
924 | } |
925 | ||
286de272 DB |
926 | static bool |
927 | conn_update_state(struct conntrack *ct, struct dp_packet *pkt, | |
967bb5c5 DB |
928 | struct conn_lookup_ctx *ctx, struct conn *conn, |
929 | long long now) | |
286de272 | 930 | { |
967bb5c5 | 931 | ovs_assert(conn->conn_type == CT_CONN_TYPE_DEFAULT); |
286de272 DB |
932 | bool create_new_conn = false; |
933 | ||
dbb597d3 | 934 | if (ctx->icmp_related) { |
286de272 DB |
935 | pkt->md.ct_state |= CS_RELATED; |
936 | if (ctx->reply) { | |
937 | pkt->md.ct_state |= CS_REPLY_DIR; | |
938 | } | |
939 | } else { | |
967bb5c5 | 940 | if (conn->alg_related) { |
bd5e81a0 DB |
941 | pkt->md.ct_state |= CS_RELATED; |
942 | } | |
dec0dbbc | 943 | |
967bb5c5 | 944 | enum ct_update_res res = conn_update(ct, conn, pkt, ctx, now); |
286de272 DB |
945 | |
946 | switch (res) { | |
947 | case CT_UPDATE_VALID: | |
948 | pkt->md.ct_state |= CS_ESTABLISHED; | |
949 | pkt->md.ct_state &= ~CS_NEW; | |
950 | if (ctx->reply) { | |
951 | pkt->md.ct_state |= CS_REPLY_DIR; | |
952 | } | |
953 | break; | |
954 | case CT_UPDATE_INVALID: | |
955 | pkt->md.ct_state = CS_INVALID; | |
956 | break; | |
957 | case CT_UPDATE_NEW: | |
967bb5c5 DB |
958 | ovs_mutex_lock(&ct->ct_lock); |
959 | conn_clean(ct, conn); | |
960 | ovs_mutex_unlock(&ct->ct_lock); | |
286de272 DB |
961 | create_new_conn = true; |
962 | break; | |
963 | default: | |
964 | OVS_NOT_REACHED(); | |
965 | } | |
966 | } | |
967 | return create_new_conn; | |
968 | } | |
969 | ||
286de272 DB |
970 | static void |
971 | handle_nat(struct dp_packet *pkt, struct conn *conn, | |
972 | uint16_t zone, bool reply, bool related) | |
973 | { | |
974 | if (conn->nat_info && | |
975 | (!(pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) || | |
976 | (pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT) && | |
977 | zone != pkt->md.ct_zone))) { | |
bd5e81a0 | 978 | |
286de272 DB |
979 | if (pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) { |
980 | pkt->md.ct_state &= ~(CS_SRC_NAT | CS_DST_NAT); | |
981 | } | |
982 | if (reply) { | |
983 | un_nat_packet(pkt, conn, related); | |
984 | } else { | |
985 | nat_packet(pkt, conn, related); | |
986 | } | |
987 | } | |
988 | } | |
989 | ||
f8016041 DB |
990 | static bool |
991 | check_orig_tuple(struct conntrack *ct, struct dp_packet *pkt, | |
992 | struct conn_lookup_ctx *ctx_in, long long now, | |
967bb5c5 | 993 | struct conn **conn, |
f8016041 | 994 | const struct nat_action_info_t *nat_action_info) |
f8016041 DB |
995 | { |
996 | if ((ctx_in->key.dl_type == htons(ETH_TYPE_IP) && | |
997 | !pkt->md.ct_orig_tuple.ipv4.ipv4_proto) || | |
998 | (ctx_in->key.dl_type == htons(ETH_TYPE_IPV6) && | |
999 | !pkt->md.ct_orig_tuple.ipv6.ipv6_proto) || | |
1000 | !(pkt->md.ct_state & (CS_SRC_NAT | CS_DST_NAT)) || | |
1001 | nat_action_info) { | |
1002 | return false; | |
1003 | } | |
1004 | ||
967bb5c5 DB |
1005 | struct conn_key key; |
1006 | memset(&key, 0 , sizeof key); | |
f8016041 DB |
1007 | |
1008 | if (ctx_in->key.dl_type == htons(ETH_TYPE_IP)) { | |
967bb5c5 DB |
1009 | key.src.addr.ipv4 = pkt->md.ct_orig_tuple.ipv4.ipv4_src; |
1010 | key.dst.addr.ipv4 = pkt->md.ct_orig_tuple.ipv4.ipv4_dst; | |
f8016041 DB |
1011 | |
1012 | if (ctx_in->key.nw_proto == IPPROTO_ICMP) { | |
967bb5c5 DB |
1013 | key.src.icmp_id = ctx_in->key.src.icmp_id; |
1014 | key.dst.icmp_id = ctx_in->key.dst.icmp_id; | |
f8016041 | 1015 | uint16_t src_port = ntohs(pkt->md.ct_orig_tuple.ipv4.src_port); |
967bb5c5 DB |
1016 | key.src.icmp_type = (uint8_t) src_port; |
1017 | key.dst.icmp_type = reverse_icmp_type(key.src.icmp_type); | |
f8016041 | 1018 | } else { |
967bb5c5 DB |
1019 | key.src.port = pkt->md.ct_orig_tuple.ipv4.src_port; |
1020 | key.dst.port = pkt->md.ct_orig_tuple.ipv4.dst_port; | |
f8016041 | 1021 | } |
967bb5c5 | 1022 | key.nw_proto = pkt->md.ct_orig_tuple.ipv4.ipv4_proto; |
f8016041 | 1023 | } else { |
967bb5c5 DB |
1024 | key.src.addr.ipv6 = pkt->md.ct_orig_tuple.ipv6.ipv6_src; |
1025 | key.dst.addr.ipv6 = pkt->md.ct_orig_tuple.ipv6.ipv6_dst; | |
f8016041 DB |
1026 | |
1027 | if (ctx_in->key.nw_proto == IPPROTO_ICMPV6) { | |
967bb5c5 DB |
1028 | key.src.icmp_id = ctx_in->key.src.icmp_id; |
1029 | key.dst.icmp_id = ctx_in->key.dst.icmp_id; | |
f8016041 | 1030 | uint16_t src_port = ntohs(pkt->md.ct_orig_tuple.ipv6.src_port); |
967bb5c5 DB |
1031 | key.src.icmp_type = (uint8_t) src_port; |
1032 | key.dst.icmp_type = reverse_icmp6_type(key.src.icmp_type); | |
f8016041 | 1033 | } else { |
967bb5c5 DB |
1034 | key.src.port = pkt->md.ct_orig_tuple.ipv6.src_port; |
1035 | key.dst.port = pkt->md.ct_orig_tuple.ipv6.dst_port; | |
f8016041 | 1036 | } |
967bb5c5 | 1037 | key.nw_proto = pkt->md.ct_orig_tuple.ipv6.ipv6_proto; |
f8016041 DB |
1038 | } |
1039 | ||
967bb5c5 DB |
1040 | key.dl_type = ctx_in->key.dl_type; |
1041 | key.zone = pkt->md.ct_zone; | |
1042 | uint32_t hash = conn_key_hash(&key, ct->hash_basis); | |
1043 | bool reply; | |
1044 | conn_key_lookup(ct, &key, hash, now, conn, &reply); | |
f8016041 DB |
1045 | return *conn ? true : false; |
1046 | } | |
1047 | ||
94e71143 DB |
1048 | static bool |
1049 | conn_update_state_alg(struct conntrack *ct, struct dp_packet *pkt, | |
1050 | struct conn_lookup_ctx *ctx, struct conn *conn, | |
1051 | const struct nat_action_info_t *nat_action_info, | |
1052 | enum ct_alg_ctl_type ct_alg_ctl, long long now, | |
967bb5c5 | 1053 | bool *create_new_conn) |
94e71143 DB |
1054 | { |
1055 | if (is_ftp_ctl(ct_alg_ctl)) { | |
1056 | /* Keep sequence tracking in sync with the source of the | |
1057 | * sequence skew. */ | |
967bb5c5 | 1058 | ovs_mutex_lock(&conn->lock); |
94e71143 DB |
1059 | if (ctx->reply != conn->seq_skew_dir) { |
1060 | handle_ftp_ctl(ct, ctx, pkt, conn, now, CT_FTP_CTL_OTHER, | |
1061 | !!nat_action_info); | |
967bb5c5 DB |
1062 | /* conn_update_state locks for unrelated fields, so unlock. */ |
1063 | ovs_mutex_unlock(&conn->lock); | |
1064 | *create_new_conn = conn_update_state(ct, pkt, ctx, conn, now); | |
94e71143 | 1065 | } else { |
967bb5c5 DB |
1066 | /* conn_update_state locks for unrelated fields, so unlock. */ |
1067 | ovs_mutex_unlock(&conn->lock); | |
1068 | *create_new_conn = conn_update_state(ct, pkt, ctx, conn, now); | |
1069 | ovs_mutex_lock(&conn->lock); | |
030958a0 DB |
1070 | if (*create_new_conn == false) { |
1071 | handle_ftp_ctl(ct, ctx, pkt, conn, now, CT_FTP_CTL_OTHER, | |
1072 | !!nat_action_info); | |
1073 | } | |
967bb5c5 | 1074 | ovs_mutex_unlock(&conn->lock); |
94e71143 DB |
1075 | } |
1076 | return true; | |
1077 | } | |
1078 | return false; | |
1079 | } | |
1080 | ||
286de272 | 1081 | static void |
a489b168 DDP |
1082 | process_one(struct conntrack *ct, struct dp_packet *pkt, |
1083 | struct conn_lookup_ctx *ctx, uint16_t zone, | |
286de272 DB |
1084 | bool force, bool commit, long long now, const uint32_t *setmark, |
1085 | const struct ovs_key_ct_labels *setlabel, | |
bd5e81a0 | 1086 | const struct nat_action_info_t *nat_action_info, |
bd7d93f8 | 1087 | ovs_be16 tp_src, ovs_be16 tp_dst, const char *helper) |
a489b168 | 1088 | { |
967bb5c5 DB |
1089 | bool create_new_conn = false; |
1090 | conn_key_lookup(ct, &ctx->key, ctx->hash, now, &ctx->conn, &ctx->reply); | |
1091 | struct conn *conn = ctx->conn; | |
a489b168 | 1092 | |
a76a37ef | 1093 | /* Delete found entry if in wrong direction. 'force' implies commit. */ |
a720a7fa | 1094 | if (OVS_UNLIKELY(force && ctx->reply && conn)) { |
967bb5c5 DB |
1095 | ovs_mutex_lock(&ct->ct_lock); |
1096 | conn_clean(ct, conn); | |
1097 | ovs_mutex_unlock(&ct->ct_lock); | |
a76a37ef JR |
1098 | conn = NULL; |
1099 | } | |
1100 | ||
286de272 DB |
1101 | if (OVS_LIKELY(conn)) { |
1102 | if (conn->conn_type == CT_CONN_TYPE_UN_NAT) { | |
a489b168 | 1103 | |
286de272 | 1104 | ctx->reply = true; |
967bb5c5 DB |
1105 | struct conn *rev_conn = conn; /* Save for debugging. */ |
1106 | uint32_t hash = conn_key_hash(&conn->rev_key, ct->hash_basis); | |
1107 | conn_key_lookup(ct, &ctx->key, hash, now, &conn, &ctx->reply); | |
a489b168 | 1108 | |
967bb5c5 | 1109 | if (!conn) { |
286de272 | 1110 | pkt->md.ct_state |= CS_TRACKED | CS_INVALID; |
967bb5c5 DB |
1111 | char *log_msg = xasprintf("Missing master conn %p", rev_conn); |
1112 | ct_print_conn_info(conn, log_msg, VLL_INFO, true, true); | |
1113 | free(log_msg); | |
286de272 | 1114 | return; |
a489b168 DDP |
1115 | } |
1116 | } | |
286de272 DB |
1117 | } |
1118 | ||
bd7d93f8 DB |
1119 | enum ct_alg_ctl_type ct_alg_ctl = get_alg_ctl_type(pkt, tp_src, tp_dst, |
1120 | helper); | |
bd5e81a0 | 1121 | |
286de272 | 1122 | if (OVS_LIKELY(conn)) { |
94e71143 DB |
1123 | if (OVS_LIKELY(!conn_update_state_alg(ct, pkt, ctx, conn, |
1124 | nat_action_info, | |
967bb5c5 | 1125 | ct_alg_ctl, now, |
94e71143 | 1126 | &create_new_conn))) { |
967bb5c5 | 1127 | create_new_conn = conn_update_state(ct, pkt, ctx, conn, now); |
bd5e81a0 | 1128 | } |
286de272 | 1129 | if (nat_action_info && !create_new_conn) { |
dbb597d3 | 1130 | handle_nat(pkt, conn, zone, ctx->reply, ctx->icmp_related); |
286de272 | 1131 | } |
bd5e81a0 | 1132 | |
967bb5c5 DB |
1133 | } else if (check_orig_tuple(ct, pkt, ctx, now, &conn, nat_action_info)) { |
1134 | create_new_conn = conn_update_state(ct, pkt, ctx, conn, now); | |
a489b168 | 1135 | } else { |
dbb597d3 | 1136 | if (ctx->icmp_related) { |
bd5e81a0 DB |
1137 | /* An icmp related conn should always be found; no new |
1138 | connection is created based on an icmp related packet. */ | |
286de272 | 1139 | pkt->md.ct_state = CS_INVALID; |
5c2e106b | 1140 | } else { |
286de272 | 1141 | create_new_conn = true; |
5c2e106b | 1142 | } |
a489b168 DDP |
1143 | } |
1144 | ||
bd5e81a0 | 1145 | const struct alg_exp_node *alg_exp = NULL; |
96bbcbf7 | 1146 | struct alg_exp_node alg_exp_entry; |
dec0dbbc | 1147 | |
286de272 | 1148 | if (OVS_UNLIKELY(create_new_conn)) { |
bd5e81a0 | 1149 | |
967bb5c5 | 1150 | ovs_rwlock_rdlock(&ct->resources_lock); |
bd5e81a0 | 1151 | alg_exp = expectation_lookup(&ct->alg_expectations, &ctx->key, |
be38342d DB |
1152 | ct->hash_basis, |
1153 | alg_src_ip_wc(ct_alg_ctl)); | |
bd5e81a0 | 1154 | if (alg_exp) { |
c3f6bae2 | 1155 | memcpy(&alg_exp_entry, alg_exp, sizeof alg_exp_entry); |
bd5e81a0 DB |
1156 | alg_exp = &alg_exp_entry; |
1157 | } | |
967bb5c5 | 1158 | ovs_rwlock_unlock(&ct->resources_lock); |
bd5e81a0 | 1159 | |
967bb5c5 | 1160 | ovs_mutex_lock(&ct->ct_lock); |
286de272 | 1161 | conn = conn_not_found(ct, pkt, ctx, commit, now, nat_action_info, |
967bb5c5 DB |
1162 | helper, alg_exp, ct_alg_ctl); |
1163 | ovs_mutex_unlock(&ct->ct_lock); | |
286de272 DB |
1164 | } |
1165 | ||
bd5e81a0 DB |
1166 | write_ct_md(pkt, zone, conn, &ctx->key, alg_exp); |
1167 | ||
286de272 DB |
1168 | if (conn && setmark) { |
1169 | set_mark(pkt, conn, setmark[0], setmark[1]); | |
1170 | } | |
a489b168 | 1171 | |
286de272 DB |
1172 | if (conn && setlabel) { |
1173 | set_label(pkt, conn, &setlabel[0], &setlabel[1]); | |
1174 | } | |
1175 | ||
967bb5c5 | 1176 | handle_alg_ctl(ct, ctx, pkt, ct_alg_ctl, conn, now, !!nat_action_info); |
a489b168 DDP |
1177 | } |
1178 | ||
1179 | /* Sends the packets in '*pkt_batch' through the connection tracker 'ct'. All | |
51b9a533 | 1180 | * the packets must have the same 'dl_type' (IPv4 or IPv6) and should have |
4ea96698 DB |
1181 | * the l3 and and l4 offset properly set. Performs fragment reassembly with |
1182 | * the help of ipf_preprocess_conntrack(). | |
a489b168 DDP |
1183 | * |
1184 | * If 'commit' is true, the packets are allowed to create new entries in the | |
1185 | * connection tables. 'setmark', if not NULL, should point to a two | |
1186 | * elements array containing a value and a mask to set the connection mark. | |
1187 | * 'setlabel' behaves similarly for the connection label.*/ | |
1188 | int | |
1189 | conntrack_execute(struct conntrack *ct, struct dp_packet_batch *pkt_batch, | |
a76a37ef | 1190 | ovs_be16 dl_type, bool force, bool commit, uint16_t zone, |
66e4ad8a | 1191 | const uint32_t *setmark, |
a489b168 | 1192 | const struct ovs_key_ct_labels *setlabel, |
bd7d93f8 | 1193 | ovs_be16 tp_src, ovs_be16 tp_dst, const char *helper, |
94053e66 FA |
1194 | const struct nat_action_info_t *nat_action_info, |
1195 | long long now) | |
a489b168 | 1196 | { |
4ea96698 DB |
1197 | ipf_preprocess_conntrack(ct->ipf, pkt_batch, now, dl_type, zone, |
1198 | ct->hash_basis); | |
1199 | ||
43495c45 | 1200 | struct dp_packet *packet; |
61ce32b9 | 1201 | struct conn_lookup_ctx ctx; |
a489b168 | 1202 | |
e883448e | 1203 | DP_PACKET_BATCH_FOR_EACH (i, packet, pkt_batch) { |
4ea96698 DB |
1204 | if (packet->md.ct_state == CS_INVALID |
1205 | || !conn_key_extract(ct, packet, dl_type, &ctx, zone)) { | |
43495c45 BB |
1206 | packet->md.ct_state = CS_INVALID; |
1207 | write_ct_md(packet, zone, NULL, NULL, NULL); | |
a489b168 DDP |
1208 | continue; |
1209 | } | |
94e71143 | 1210 | process_one(ct, packet, &ctx, zone, force, commit, now, setmark, |
bd7d93f8 | 1211 | setlabel, nat_action_info, tp_src, tp_dst, helper); |
a489b168 DDP |
1212 | } |
1213 | ||
4ea96698 DB |
1214 | ipf_postprocess_conntrack(ct->ipf, pkt_batch, now, dl_type); |
1215 | ||
a489b168 DDP |
1216 | return 0; |
1217 | } | |
1218 | ||
1fe178d2 EG |
1219 | void |
1220 | conntrack_clear(struct dp_packet *packet) | |
1221 | { | |
1222 | /* According to pkt_metadata_init(), ct_state == 0 is enough to make all of | |
1223 | * the conntrack fields invalid. */ | |
1224 | packet->md.ct_state = 0; | |
1225 | } | |
1226 | ||
a489b168 DDP |
1227 | static void |
1228 | set_mark(struct dp_packet *pkt, struct conn *conn, uint32_t val, uint32_t mask) | |
1229 | { | |
967bb5c5 | 1230 | ovs_mutex_lock(&conn->lock); |
bd5e81a0 DB |
1231 | if (conn->alg_related) { |
1232 | pkt->md.ct_mark = conn->mark; | |
1233 | } else { | |
1234 | pkt->md.ct_mark = val | (pkt->md.ct_mark & ~(mask)); | |
1235 | conn->mark = pkt->md.ct_mark; | |
1236 | } | |
967bb5c5 | 1237 | ovs_mutex_unlock(&conn->lock); |
a489b168 DDP |
1238 | } |
1239 | ||
1240 | static void | |
1241 | set_label(struct dp_packet *pkt, struct conn *conn, | |
1242 | const struct ovs_key_ct_labels *val, | |
1243 | const struct ovs_key_ct_labels *mask) | |
1244 | { | |
967bb5c5 | 1245 | ovs_mutex_lock(&conn->lock); |
bd5e81a0 DB |
1246 | if (conn->alg_related) { |
1247 | pkt->md.ct_label = conn->label; | |
1248 | } else { | |
1249 | ovs_u128 v, m; | |
a489b168 | 1250 | |
bd5e81a0 DB |
1251 | memcpy(&v, val, sizeof v); |
1252 | memcpy(&m, mask, sizeof m); | |
a489b168 | 1253 | |
bd5e81a0 | 1254 | pkt->md.ct_label.u64.lo = v.u64.lo |
a489b168 | 1255 | | (pkt->md.ct_label.u64.lo & ~(m.u64.lo)); |
bd5e81a0 | 1256 | pkt->md.ct_label.u64.hi = v.u64.hi |
a489b168 | 1257 | | (pkt->md.ct_label.u64.hi & ~(m.u64.hi)); |
bd5e81a0 DB |
1258 | conn->label = pkt->md.ct_label; |
1259 | } | |
967bb5c5 | 1260 | ovs_mutex_unlock(&conn->lock); |
a489b168 | 1261 | } |
286de272 | 1262 | |
a489b168 | 1263 | \f |
e6ef6cc6 DDP |
1264 | /* Delete the expired connections from 'ctb', up to 'limit'. Returns the |
1265 | * earliest expiration time among the remaining connections in 'ctb'. Returns | |
1266 | * LLONG_MAX if 'ctb' is empty. The return value might be smaller than 'now', | |
1267 | * if 'limit' is reached */ | |
1268 | static long long | |
967bb5c5 | 1269 | ct_sweep(struct conntrack *ct, long long now, size_t limit) |
e6ef6cc6 DDP |
1270 | { |
1271 | struct conn *conn, *next; | |
1272 | long long min_expiration = LLONG_MAX; | |
e6ef6cc6 DDP |
1273 | size_t count = 0; |
1274 | ||
967bb5c5 DB |
1275 | ovs_mutex_lock(&ct->ct_lock); |
1276 | ||
dec0dbbc | 1277 | for (unsigned i = 0; i < N_CT_TM; i++) { |
967bb5c5 DB |
1278 | LIST_FOR_EACH_SAFE (conn, next, exp_node, &ct->exp_lists[i]) { |
1279 | ovs_mutex_lock(&conn->lock); | |
1280 | if (now < conn->expiration || count >= limit) { | |
a720a7fa | 1281 | min_expiration = MIN(min_expiration, conn->expiration); |
967bb5c5 | 1282 | ovs_mutex_unlock(&conn->lock); |
a720a7fa DB |
1283 | if (count >= limit) { |
1284 | /* Do not check other lists. */ | |
1285 | COVERAGE_INC(conntrack_long_cleanup); | |
967bb5c5 | 1286 | goto out; |
e6ef6cc6 | 1287 | } |
a720a7fa | 1288 | break; |
967bb5c5 DB |
1289 | } else { |
1290 | ovs_mutex_unlock(&conn->lock); | |
1291 | conn_clean(ct, conn); | |
e6ef6cc6 | 1292 | } |
a720a7fa | 1293 | count++; |
e6ef6cc6 DDP |
1294 | } |
1295 | } | |
967bb5c5 DB |
1296 | |
1297 | out: | |
1298 | VLOG_DBG("conntrack cleanup %"PRIuSIZE" entries in %lld msec", count, | |
1299 | time_msec() - now); | |
1300 | ovs_mutex_unlock(&ct->ct_lock); | |
e6ef6cc6 DDP |
1301 | return min_expiration; |
1302 | } | |
1303 | ||
1304 | /* Cleans up old connection entries from 'ct'. Returns the time when the | |
1305 | * next expiration might happen. The return value might be smaller than | |
1306 | * 'now', meaning that an internal limit has been reached, and some expired | |
1307 | * connections have not been deleted. */ | |
1308 | static long long | |
1309 | conntrack_clean(struct conntrack *ct, long long now) | |
1310 | { | |
e6ef6cc6 | 1311 | unsigned int n_conn_limit; |
e6ef6cc6 | 1312 | atomic_read_relaxed(&ct->n_conn_limit, &n_conn_limit); |
967bb5c5 DB |
1313 | size_t clean_max = n_conn_limit > 10 ? n_conn_limit / 10 : 1; |
1314 | long long min_exp = ct_sweep(ct, now, clean_max); | |
1315 | long long next_wakeup = MIN(min_exp, now + CT_TM_MIN); | |
e6ef6cc6 DDP |
1316 | |
1317 | return next_wakeup; | |
1318 | } | |
1319 | ||
1320 | /* Cleanup: | |
e6ef6cc6 DDP |
1321 | * |
1322 | * We must call conntrack_clean() periodically. conntrack_clean() return | |
1323 | * value gives an hint on when the next cleanup must be done (either because | |
1324 | * there is an actual connection that expires, or because a new connection | |
1325 | * might be created with the minimum timeout). | |
1326 | * | |
1327 | * The logic below has two goals: | |
1328 | * | |
6c54734e DDP |
1329 | * - We want to reduce the number of wakeups and batch connection cleanup |
1330 | * when the load is not very high. CT_CLEAN_INTERVAL ensures that if we | |
1331 | * are coping with the current cleanup tasks, then we wait at least | |
1332 | * 5 seconds to do further cleanup. | |
e6ef6cc6 | 1333 | * |
967bb5c5 | 1334 | * - We don't want to keep the map locked too long, as we might prevent |
6c54734e | 1335 | * traffic from flowing. CT_CLEAN_MIN_INTERVAL ensures that if cleanup is |
967bb5c5 | 1336 | * behind, there is at least some 200ms blocks of time when the map will be |
6c54734e | 1337 | * left alone, so the datapath can operate unhindered. |
e6ef6cc6 DDP |
1338 | */ |
1339 | #define CT_CLEAN_INTERVAL 5000 /* 5 seconds */ | |
1340 | #define CT_CLEAN_MIN_INTERVAL 200 /* 0.2 seconds */ | |
1341 | ||
1342 | static void * | |
1343 | clean_thread_main(void *f_) | |
1344 | { | |
1345 | struct conntrack *ct = f_; | |
1346 | ||
1347 | while (!latch_is_set(&ct->clean_thread_exit)) { | |
1348 | long long next_wake; | |
1349 | long long now = time_msec(); | |
e6ef6cc6 DDP |
1350 | next_wake = conntrack_clean(ct, now); |
1351 | ||
1352 | if (next_wake < now) { | |
1353 | poll_timer_wait_until(now + CT_CLEAN_MIN_INTERVAL); | |
1354 | } else { | |
1355 | poll_timer_wait_until(MAX(next_wake, now + CT_CLEAN_INTERVAL)); | |
1356 | } | |
1357 | latch_wait(&ct->clean_thread_exit); | |
1358 | poll_block(); | |
1359 | } | |
1360 | ||
1361 | return NULL; | |
1362 | } | |
1363 | \f | |
e917d3ee DB |
1364 | /* 'Data' is a pointer to the beginning of the L3 header and 'new_data' is |
1365 | * used to store a pointer to the first byte after the L3 header. 'Size' is | |
1366 | * the size of the packet beyond the data pointer. */ | |
a489b168 DDP |
1367 | static inline bool |
1368 | extract_l3_ipv4(struct conn_key *key, const void *data, size_t size, | |
1369 | const char **new_data, bool validate_checksum) | |
1370 | { | |
e917d3ee DB |
1371 | if (OVS_UNLIKELY(size < IP_HEADER_LEN)) { |
1372 | return false; | |
a489b168 DDP |
1373 | } |
1374 | ||
dec0dbbc DB |
1375 | const struct ip_header *ip = data; |
1376 | size_t ip_len = IP_IHL(ip->ip_ihl_ver) * 4; | |
a489b168 | 1377 | |
e917d3ee DB |
1378 | if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) { |
1379 | return false; | |
1380 | } | |
a489b168 | 1381 | |
e917d3ee DB |
1382 | if (OVS_UNLIKELY(size < ip_len)) { |
1383 | return false; | |
1384 | } | |
a489b168 | 1385 | |
e917d3ee DB |
1386 | if (IP_IS_FRAGMENT(ip->ip_frag_off)) { |
1387 | return false; | |
a489b168 DDP |
1388 | } |
1389 | ||
1390 | if (validate_checksum && csum(data, ip_len) != 0) { | |
1391 | return false; | |
1392 | } | |
1393 | ||
e917d3ee DB |
1394 | if (new_data) { |
1395 | *new_data = (char *) data + ip_len; | |
1396 | } | |
1397 | ||
cda1b109 DB |
1398 | key->src.addr.ipv4 = get_16aligned_be32(&ip->ip_src); |
1399 | key->dst.addr.ipv4 = get_16aligned_be32(&ip->ip_dst); | |
a489b168 DDP |
1400 | key->nw_proto = ip->ip_proto; |
1401 | ||
1402 | return true; | |
1403 | } | |
1404 | ||
e917d3ee DB |
1405 | /* 'Data' is a pointer to the beginning of the L3 header and 'new_data' is |
1406 | * used to store a pointer to the first byte after the L3 header. 'Size' is | |
1407 | * the size of the packet beyond the data pointer. */ | |
a489b168 DDP |
1408 | static inline bool |
1409 | extract_l3_ipv6(struct conn_key *key, const void *data, size_t size, | |
1410 | const char **new_data) | |
1411 | { | |
1412 | const struct ovs_16aligned_ip6_hdr *ip6 = data; | |
286de272 | 1413 | |
e917d3ee DB |
1414 | if (OVS_UNLIKELY(size < sizeof *ip6)) { |
1415 | return false; | |
a489b168 DDP |
1416 | } |
1417 | ||
1418 | data = ip6 + 1; | |
1419 | size -= sizeof *ip6; | |
dec0dbbc DB |
1420 | uint8_t nw_proto = ip6->ip6_nxt; |
1421 | uint8_t nw_frag = 0; | |
a489b168 | 1422 | |
523464ab DB |
1423 | const struct ovs_16aligned_ip6_frag *frag_hdr; |
1424 | if (!parse_ipv6_ext_hdrs(&data, &size, &nw_proto, &nw_frag, &frag_hdr)) { | |
a489b168 DDP |
1425 | return false; |
1426 | } | |
1427 | ||
a489b168 DDP |
1428 | if (nw_frag) { |
1429 | return false; | |
1430 | } | |
1431 | ||
c8b1ad49 DB |
1432 | if (new_data) { |
1433 | *new_data = data; | |
1434 | } | |
1435 | ||
cda1b109 DB |
1436 | memcpy(&key->src.addr.ipv6, &ip6->ip6_src, sizeof key->src.addr); |
1437 | memcpy(&key->dst.addr.ipv6, &ip6->ip6_dst, sizeof key->dst.addr); | |
a489b168 DDP |
1438 | key->nw_proto = nw_proto; |
1439 | ||
1440 | return true; | |
1441 | } | |
1442 | ||
1443 | static inline bool | |
1444 | checksum_valid(const struct conn_key *key, const void *data, size_t size, | |
1445 | const void *l3) | |
1446 | { | |
a489b168 | 1447 | if (key->dl_type == htons(ETH_TYPE_IP)) { |
76d85771 DB |
1448 | uint32_t csum = packet_csum_pseudoheader(l3); |
1449 | return csum_finish(csum_continue(csum, data, size)) == 0; | |
a489b168 | 1450 | } else if (key->dl_type == htons(ETH_TYPE_IPV6)) { |
76d85771 | 1451 | return packet_csum_upperlayer6(l3, data, key->nw_proto, size) == 0; |
a489b168 DDP |
1452 | } else { |
1453 | return false; | |
1454 | } | |
a489b168 DDP |
1455 | } |
1456 | ||
1457 | static inline bool | |
1458 | check_l4_tcp(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1459 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1460 | { |
1461 | const struct tcp_header *tcp = data; | |
40225b0c BP |
1462 | if (size < sizeof *tcp) { |
1463 | return false; | |
1464 | } | |
a489b168 | 1465 | |
40225b0c | 1466 | size_t tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4; |
a489b168 DDP |
1467 | if (OVS_UNLIKELY(tcp_len < TCP_HEADER_LEN || tcp_len > size)) { |
1468 | return false; | |
1469 | } | |
1470 | ||
324459a3 | 1471 | return validate_checksum ? checksum_valid(key, data, size, l3) : true; |
a489b168 DDP |
1472 | } |
1473 | ||
1474 | static inline bool | |
1475 | check_l4_udp(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1476 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1477 | { |
1478 | const struct udp_header *udp = data; | |
40225b0c BP |
1479 | if (size < sizeof *udp) { |
1480 | return false; | |
1481 | } | |
a489b168 | 1482 | |
40225b0c | 1483 | size_t udp_len = ntohs(udp->udp_len); |
a489b168 DDP |
1484 | if (OVS_UNLIKELY(udp_len < UDP_HEADER_LEN || udp_len > size)) { |
1485 | return false; | |
1486 | } | |
1487 | ||
1488 | /* Validation must be skipped if checksum is 0 on IPv4 packets */ | |
1489 | return (udp->udp_csum == 0 && key->dl_type == htons(ETH_TYPE_IP)) | |
324459a3 | 1490 | || (validate_checksum ? checksum_valid(key, data, size, l3) : true); |
a489b168 DDP |
1491 | } |
1492 | ||
1493 | static inline bool | |
324459a3 | 1494 | check_l4_icmp(const void *data, size_t size, bool validate_checksum) |
a489b168 | 1495 | { |
324459a3 | 1496 | return validate_checksum ? csum(data, size) == 0 : true; |
a489b168 DDP |
1497 | } |
1498 | ||
1499 | static inline bool | |
1500 | check_l4_icmp6(const struct conn_key *key, const void *data, size_t size, | |
324459a3 | 1501 | const void *l3, bool validate_checksum) |
a489b168 | 1502 | { |
324459a3 | 1503 | return validate_checksum ? checksum_valid(key, data, size, l3) : true; |
a489b168 DDP |
1504 | } |
1505 | ||
1506 | static inline bool | |
1507 | extract_l4_tcp(struct conn_key *key, const void *data, size_t size) | |
1508 | { | |
a489b168 DDP |
1509 | if (OVS_UNLIKELY(size < TCP_HEADER_LEN)) { |
1510 | return false; | |
1511 | } | |
1512 | ||
dec0dbbc | 1513 | const struct tcp_header *tcp = data; |
a489b168 DDP |
1514 | key->src.port = tcp->tcp_src; |
1515 | key->dst.port = tcp->tcp_dst; | |
1516 | ||
1517 | /* Port 0 is invalid */ | |
1518 | return key->src.port && key->dst.port; | |
1519 | } | |
1520 | ||
1521 | static inline bool | |
1522 | extract_l4_udp(struct conn_key *key, const void *data, size_t size) | |
1523 | { | |
a489b168 DDP |
1524 | if (OVS_UNLIKELY(size < UDP_HEADER_LEN)) { |
1525 | return false; | |
1526 | } | |
1527 | ||
dec0dbbc | 1528 | const struct udp_header *udp = data; |
a489b168 DDP |
1529 | key->src.port = udp->udp_src; |
1530 | key->dst.port = udp->udp_dst; | |
1531 | ||
1532 | /* Port 0 is invalid */ | |
1533 | return key->src.port && key->dst.port; | |
1534 | } | |
1535 | ||
1536 | static inline bool extract_l4(struct conn_key *key, const void *data, | |
324459a3 SC |
1537 | size_t size, bool *related, const void *l3, |
1538 | bool validate_checksum); | |
a489b168 | 1539 | |
b269a122 DDP |
1540 | static uint8_t |
1541 | reverse_icmp_type(uint8_t type) | |
1542 | { | |
1543 | switch (type) { | |
1544 | case ICMP4_ECHO_REQUEST: | |
1545 | return ICMP4_ECHO_REPLY; | |
1546 | case ICMP4_ECHO_REPLY: | |
1547 | return ICMP4_ECHO_REQUEST; | |
1548 | ||
1549 | case ICMP4_TIMESTAMP: | |
1550 | return ICMP4_TIMESTAMPREPLY; | |
1551 | case ICMP4_TIMESTAMPREPLY: | |
1552 | return ICMP4_TIMESTAMP; | |
1553 | ||
1554 | case ICMP4_INFOREQUEST: | |
1555 | return ICMP4_INFOREPLY; | |
1556 | case ICMP4_INFOREPLY: | |
1557 | return ICMP4_INFOREQUEST; | |
1558 | default: | |
1559 | OVS_NOT_REACHED(); | |
1560 | } | |
1561 | } | |
1562 | ||
a489b168 DDP |
1563 | /* If 'related' is not NULL and the function is processing an ICMP |
1564 | * error packet, extract the l3 and l4 fields from the nested header | |
1565 | * instead and set *related to true. If 'related' is NULL we're | |
1566 | * already processing a nested header and no such recursion is | |
1567 | * possible */ | |
1568 | static inline int | |
1569 | extract_l4_icmp(struct conn_key *key, const void *data, size_t size, | |
1570 | bool *related) | |
1571 | { | |
a489b168 DDP |
1572 | if (OVS_UNLIKELY(size < ICMP_HEADER_LEN)) { |
1573 | return false; | |
1574 | } | |
1575 | ||
dec0dbbc DB |
1576 | const struct icmp_header *icmp = data; |
1577 | ||
a489b168 DDP |
1578 | switch (icmp->icmp_type) { |
1579 | case ICMP4_ECHO_REQUEST: | |
1580 | case ICMP4_ECHO_REPLY: | |
1581 | case ICMP4_TIMESTAMP: | |
1582 | case ICMP4_TIMESTAMPREPLY: | |
1583 | case ICMP4_INFOREQUEST: | |
1584 | case ICMP4_INFOREPLY: | |
b269a122 DDP |
1585 | if (icmp->icmp_code != 0) { |
1586 | return false; | |
1587 | } | |
a489b168 | 1588 | /* Separate ICMP connection: identified using id */ |
b269a122 DDP |
1589 | key->src.icmp_id = key->dst.icmp_id = icmp->icmp_fields.echo.id; |
1590 | key->src.icmp_type = icmp->icmp_type; | |
1591 | key->dst.icmp_type = reverse_icmp_type(icmp->icmp_type); | |
a489b168 DDP |
1592 | break; |
1593 | case ICMP4_DST_UNREACH: | |
1594 | case ICMP4_TIME_EXCEEDED: | |
1595 | case ICMP4_PARAM_PROB: | |
1596 | case ICMP4_SOURCEQUENCH: | |
1597 | case ICMP4_REDIRECT: { | |
1598 | /* ICMP packet part of another connection. We should | |
1599 | * extract the key from embedded packet header */ | |
1600 | struct conn_key inner_key; | |
1601 | const char *l3 = (const char *) (icmp + 1); | |
1602 | const char *tail = (const char *) data + size; | |
1603 | const char *l4; | |
a489b168 DDP |
1604 | |
1605 | if (!related) { | |
1606 | return false; | |
1607 | } | |
1608 | ||
1609 | memset(&inner_key, 0, sizeof inner_key); | |
1610 | inner_key.dl_type = htons(ETH_TYPE_IP); | |
dec0dbbc | 1611 | bool ok = extract_l3_ipv4(&inner_key, l3, tail - l3, &l4, false); |
a489b168 DDP |
1612 | if (!ok) { |
1613 | return false; | |
1614 | } | |
1615 | ||
cda1b109 | 1616 | if (inner_key.src.addr.ipv4 != key->dst.addr.ipv4) { |
a489b168 DDP |
1617 | return false; |
1618 | } | |
1619 | ||
1620 | key->src = inner_key.src; | |
1621 | key->dst = inner_key.dst; | |
1622 | key->nw_proto = inner_key.nw_proto; | |
1623 | ||
324459a3 | 1624 | ok = extract_l4(key, l4, tail - l4, NULL, l3, false); |
a489b168 DDP |
1625 | if (ok) { |
1626 | conn_key_reverse(key); | |
1627 | *related = true; | |
1628 | } | |
1629 | return ok; | |
1630 | } | |
1631 | default: | |
1632 | return false; | |
1633 | } | |
1634 | ||
1635 | return true; | |
1636 | } | |
1637 | ||
b269a122 DDP |
1638 | static uint8_t |
1639 | reverse_icmp6_type(uint8_t type) | |
1640 | { | |
1641 | switch (type) { | |
1642 | case ICMP6_ECHO_REQUEST: | |
1643 | return ICMP6_ECHO_REPLY; | |
1644 | case ICMP6_ECHO_REPLY: | |
1645 | return ICMP6_ECHO_REQUEST; | |
1646 | default: | |
1647 | OVS_NOT_REACHED(); | |
1648 | } | |
1649 | } | |
1650 | ||
a489b168 DDP |
1651 | /* If 'related' is not NULL and the function is processing an ICMP |
1652 | * error packet, extract the l3 and l4 fields from the nested header | |
1653 | * instead and set *related to true. If 'related' is NULL we're | |
1654 | * already processing a nested header and no such recursion is | |
1655 | * possible */ | |
1656 | static inline bool | |
1657 | extract_l4_icmp6(struct conn_key *key, const void *data, size_t size, | |
1658 | bool *related) | |
1659 | { | |
1660 | const struct icmp6_header *icmp6 = data; | |
1661 | ||
1662 | /* All the messages that we support need at least 4 bytes after | |
1663 | * the header */ | |
1664 | if (size < sizeof *icmp6 + 4) { | |
1665 | return false; | |
1666 | } | |
1667 | ||
1668 | switch (icmp6->icmp6_type) { | |
1669 | case ICMP6_ECHO_REQUEST: | |
1670 | case ICMP6_ECHO_REPLY: | |
b269a122 DDP |
1671 | if (icmp6->icmp6_code != 0) { |
1672 | return false; | |
1673 | } | |
a489b168 | 1674 | /* Separate ICMP connection: identified using id */ |
b269a122 DDP |
1675 | key->src.icmp_id = key->dst.icmp_id = *(ovs_be16 *) (icmp6 + 1); |
1676 | key->src.icmp_type = icmp6->icmp6_type; | |
1677 | key->dst.icmp_type = reverse_icmp6_type(icmp6->icmp6_type); | |
a489b168 DDP |
1678 | break; |
1679 | case ICMP6_DST_UNREACH: | |
1680 | case ICMP6_PACKET_TOO_BIG: | |
1681 | case ICMP6_TIME_EXCEEDED: | |
1682 | case ICMP6_PARAM_PROB: { | |
1683 | /* ICMP packet part of another connection. We should | |
1684 | * extract the key from embedded packet header */ | |
1685 | struct conn_key inner_key; | |
1686 | const char *l3 = (const char *) icmp6 + 8; | |
1687 | const char *tail = (const char *) data + size; | |
1688 | const char *l4 = NULL; | |
a489b168 DDP |
1689 | |
1690 | if (!related) { | |
1691 | return false; | |
1692 | } | |
1693 | ||
1694 | memset(&inner_key, 0, sizeof inner_key); | |
1695 | inner_key.dl_type = htons(ETH_TYPE_IPV6); | |
dec0dbbc | 1696 | bool ok = extract_l3_ipv6(&inner_key, l3, tail - l3, &l4); |
a489b168 DDP |
1697 | if (!ok) { |
1698 | return false; | |
1699 | } | |
1700 | ||
1701 | /* pf doesn't do this, but it seems a good idea */ | |
cda1b109 DB |
1702 | if (!ipv6_addr_equals(&inner_key.src.addr.ipv6, |
1703 | &key->dst.addr.ipv6)) { | |
a489b168 DDP |
1704 | return false; |
1705 | } | |
1706 | ||
1707 | key->src = inner_key.src; | |
1708 | key->dst = inner_key.dst; | |
1709 | key->nw_proto = inner_key.nw_proto; | |
1710 | ||
324459a3 | 1711 | ok = extract_l4(key, l4, tail - l4, NULL, l3, false); |
a489b168 DDP |
1712 | if (ok) { |
1713 | conn_key_reverse(key); | |
1714 | *related = true; | |
1715 | } | |
1716 | return ok; | |
1717 | } | |
1718 | default: | |
1719 | return false; | |
1720 | } | |
1721 | ||
1722 | return true; | |
1723 | } | |
1724 | ||
1725 | /* Extract l4 fields into 'key', which must already contain valid l3 | |
1726 | * members. | |
1727 | * | |
1728 | * If 'related' is not NULL and an ICMP error packet is being | |
1729 | * processed, the function will extract the key from the packet nested | |
1401f6de | 1730 | * in the ICMP payload and set '*related' to true. |
a489b168 | 1731 | * |
9171c635 DB |
1732 | * 'size' here is the layer 4 size, which can be a nested size if parsing |
1733 | * an ICMP or ICMP6 header. | |
1734 | * | |
a489b168 DDP |
1735 | * If 'related' is NULL, it means that we're already parsing a header nested |
1736 | * in an ICMP error. In this case, we skip checksum and length validation. */ | |
1737 | static inline bool | |
1738 | extract_l4(struct conn_key *key, const void *data, size_t size, bool *related, | |
324459a3 | 1739 | const void *l3, bool validate_checksum) |
a489b168 DDP |
1740 | { |
1741 | if (key->nw_proto == IPPROTO_TCP) { | |
324459a3 SC |
1742 | return (!related || check_l4_tcp(key, data, size, l3, |
1743 | validate_checksum)) && extract_l4_tcp(key, data, size); | |
a489b168 | 1744 | } else if (key->nw_proto == IPPROTO_UDP) { |
324459a3 SC |
1745 | return (!related || check_l4_udp(key, data, size, l3, |
1746 | validate_checksum)) && extract_l4_udp(key, data, size); | |
a489b168 DDP |
1747 | } else if (key->dl_type == htons(ETH_TYPE_IP) |
1748 | && key->nw_proto == IPPROTO_ICMP) { | |
324459a3 | 1749 | return (!related || check_l4_icmp(data, size, validate_checksum)) |
a489b168 DDP |
1750 | && extract_l4_icmp(key, data, size, related); |
1751 | } else if (key->dl_type == htons(ETH_TYPE_IPV6) | |
1752 | && key->nw_proto == IPPROTO_ICMPV6) { | |
324459a3 SC |
1753 | return (!related || check_l4_icmp6(key, data, size, l3, |
1754 | validate_checksum)) && extract_l4_icmp6(key, data, size, | |
1755 | related); | |
a489b168 DDP |
1756 | } else { |
1757 | return false; | |
1758 | } | |
1759 | } | |
1760 | ||
1761 | static bool | |
66e4ad8a | 1762 | conn_key_extract(struct conntrack *ct, struct dp_packet *pkt, ovs_be16 dl_type, |
a489b168 DDP |
1763 | struct conn_lookup_ctx *ctx, uint16_t zone) |
1764 | { | |
2482b0b0 | 1765 | const struct eth_header *l2 = dp_packet_eth(pkt); |
a489b168 DDP |
1766 | const struct ip_header *l3 = dp_packet_l3(pkt); |
1767 | const char *l4 = dp_packet_l4(pkt); | |
a489b168 DDP |
1768 | |
1769 | memset(ctx, 0, sizeof *ctx); | |
1770 | ||
1771 | if (!l2 || !l3 || !l4) { | |
1772 | return false; | |
1773 | } | |
1774 | ||
1775 | ctx->key.zone = zone; | |
1776 | ||
1777 | /* XXX In this function we parse the packet (again, it has already | |
1778 | * gone through miniflow_extract()) for two reasons: | |
1779 | * | |
1780 | * 1) To extract the l3 addresses and l4 ports. | |
1781 | * We already have the l3 and l4 headers' pointers. Extracting | |
1782 | * the l3 addresses and the l4 ports is really cheap, since they | |
1783 | * can be found at fixed locations. | |
66e4ad8a DDP |
1784 | * 2) To extract the l4 type. |
1785 | * Extracting the l4 types, for IPv6 can be quite expensive, because | |
1786 | * it's not at a fixed location. | |
a489b168 DDP |
1787 | * |
1788 | * Here's a way to avoid (2) with the help of the datapath. | |
66e4ad8a | 1789 | * The datapath doesn't keep the packet's extracted flow[1], so |
a489b168 | 1790 | * using that is not an option. We could use the packet's matching |
66e4ad8a DDP |
1791 | * megaflow, but we have to make sure that the l4 type (nw_proto) |
1792 | * is unwildcarded. This means either: | |
a489b168 | 1793 | * |
66e4ad8a DDP |
1794 | * a) dpif-netdev unwildcards the l4 type when a new flow is installed |
1795 | * if the actions contains ct(). | |
a489b168 | 1796 | * |
66e4ad8a DDP |
1797 | * b) ofproto-dpif-xlate unwildcards the l4 type when translating a ct() |
1798 | * action. This is already done in different actions, but it's | |
1799 | * unnecessary for the kernel. | |
a489b168 DDP |
1800 | * |
1801 | * --- | |
66e4ad8a | 1802 | * [1] The reasons for this are that keeping the flow increases |
a489b168 DDP |
1803 | * (slightly) the cache footprint and increases computation |
1804 | * time as we move the packet around. Most importantly, the flow | |
1805 | * should be updated by the actions and this can be slow, as | |
1806 | * we use a sparse representation (miniflow). | |
1807 | * | |
1808 | */ | |
dec0dbbc | 1809 | bool ok; |
66e4ad8a | 1810 | ctx->key.dl_type = dl_type; |
dec0dbbc | 1811 | |
a489b168 | 1812 | if (ctx->key.dl_type == htons(ETH_TYPE_IP)) { |
dec0dbbc | 1813 | bool hwol_bad_l3_csum = dp_packet_ip_checksum_bad(pkt); |
324459a3 SC |
1814 | if (hwol_bad_l3_csum) { |
1815 | ok = false; | |
1816 | } else { | |
dec0dbbc | 1817 | bool hwol_good_l3_csum = dp_packet_ip_checksum_valid(pkt); |
324459a3 | 1818 | /* Validate the checksum only when hwol is not supported. */ |
9171c635 | 1819 | ok = extract_l3_ipv4(&ctx->key, l3, dp_packet_l3_size(pkt), NULL, |
324459a3 SC |
1820 | !hwol_good_l3_csum); |
1821 | } | |
a489b168 | 1822 | } else if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
9171c635 | 1823 | ok = extract_l3_ipv6(&ctx->key, l3, dp_packet_l3_size(pkt), NULL); |
a489b168 DDP |
1824 | } else { |
1825 | ok = false; | |
1826 | } | |
1827 | ||
1828 | if (ok) { | |
324459a3 SC |
1829 | bool hwol_bad_l4_csum = dp_packet_l4_checksum_bad(pkt); |
1830 | if (!hwol_bad_l4_csum) { | |
1831 | bool hwol_good_l4_csum = dp_packet_l4_checksum_valid(pkt); | |
1832 | /* Validate the checksum only when hwol is not supported. */ | |
9171c635 DB |
1833 | if (extract_l4(&ctx->key, l4, dp_packet_l4_size(pkt), |
1834 | &ctx->icmp_related, l3, !hwol_good_l4_csum)) { | |
324459a3 SC |
1835 | ctx->hash = conn_key_hash(&ctx->key, ct->hash_basis); |
1836 | return true; | |
1837 | } | |
a489b168 DDP |
1838 | } |
1839 | } | |
1840 | ||
1841 | return false; | |
1842 | } | |
92edd073 DB |
1843 | |
1844 | static uint32_t | |
cda1b109 | 1845 | ct_addr_hash_add(uint32_t hash, const union ct_addr *addr) |
92edd073 DB |
1846 | { |
1847 | BUILD_ASSERT_DECL(sizeof *addr % 4 == 0); | |
1848 | return hash_add_bytes32(hash, (const uint32_t *) addr, sizeof *addr); | |
1849 | } | |
1850 | ||
1851 | static uint32_t | |
1852 | ct_endpoint_hash_add(uint32_t hash, const struct ct_endpoint *ep) | |
1853 | { | |
1854 | BUILD_ASSERT_DECL(sizeof *ep % 4 == 0); | |
1855 | return hash_add_bytes32(hash, (const uint32_t *) ep, sizeof *ep); | |
1856 | } | |
a489b168 DDP |
1857 | \f |
1858 | /* Symmetric */ | |
1859 | static uint32_t | |
1860 | conn_key_hash(const struct conn_key *key, uint32_t basis) | |
1861 | { | |
1862 | uint32_t hsrc, hdst, hash; | |
a489b168 | 1863 | hsrc = hdst = basis; |
6b1d4625 DB |
1864 | hsrc = ct_endpoint_hash_add(hsrc, &key->src); |
1865 | hdst = ct_endpoint_hash_add(hdst, &key->dst); | |
a489b168 DDP |
1866 | |
1867 | /* Even if source and destination are swapped the hash will be the same. */ | |
1868 | hash = hsrc ^ hdst; | |
1869 | ||
1870 | /* Hash the rest of the key(L3 and L4 types and zone). */ | |
763b40b0 | 1871 | return hash_words((uint32_t *) (&key->dst + 1), |
a489b168 DDP |
1872 | (uint32_t *) (key + 1) - (uint32_t *) (&key->dst + 1), |
1873 | hash); | |
a489b168 DDP |
1874 | } |
1875 | ||
1876 | static void | |
1877 | conn_key_reverse(struct conn_key *key) | |
1878 | { | |
dec0dbbc | 1879 | struct ct_endpoint tmp = key->src; |
a489b168 DDP |
1880 | key->src = key->dst; |
1881 | key->dst = tmp; | |
1882 | } | |
1883 | ||
286de272 | 1884 | static uint32_t |
cda1b109 | 1885 | nat_ipv6_addrs_delta(struct in6_addr *ipv6_min, struct in6_addr *ipv6_max) |
286de272 | 1886 | { |
cda1b109 DB |
1887 | uint8_t *ipv6_min_hi = &ipv6_min->s6_addr[0]; |
1888 | uint8_t *ipv6_min_lo = &ipv6_min->s6_addr[0] + sizeof(uint64_t); | |
1889 | uint8_t *ipv6_max_hi = &ipv6_max->s6_addr[0]; | |
1890 | uint8_t *ipv6_max_lo = &ipv6_max->s6_addr[0] + sizeof(uint64_t); | |
286de272 DB |
1891 | |
1892 | ovs_be64 addr6_64_min_hi; | |
1893 | ovs_be64 addr6_64_min_lo; | |
1894 | memcpy(&addr6_64_min_hi, ipv6_min_hi, sizeof addr6_64_min_hi); | |
1895 | memcpy(&addr6_64_min_lo, ipv6_min_lo, sizeof addr6_64_min_lo); | |
1896 | ||
1897 | ovs_be64 addr6_64_max_hi; | |
1898 | ovs_be64 addr6_64_max_lo; | |
1899 | memcpy(&addr6_64_max_hi, ipv6_max_hi, sizeof addr6_64_max_hi); | |
1900 | memcpy(&addr6_64_max_lo, ipv6_max_lo, sizeof addr6_64_max_lo); | |
1901 | ||
1902 | uint64_t diff; | |
dec0dbbc | 1903 | |
286de272 DB |
1904 | if (addr6_64_min_hi == addr6_64_max_hi && |
1905 | ntohll(addr6_64_min_lo) <= ntohll(addr6_64_max_lo)) { | |
1906 | diff = ntohll(addr6_64_max_lo) - ntohll(addr6_64_min_lo); | |
1907 | } else if (ntohll(addr6_64_min_hi) + 1 == ntohll(addr6_64_max_hi) && | |
1908 | ntohll(addr6_64_min_lo) > ntohll(addr6_64_max_lo)) { | |
1909 | diff = UINT64_MAX - (ntohll(addr6_64_min_lo) - | |
1910 | ntohll(addr6_64_max_lo) - 1); | |
1911 | } else { | |
1912 | /* Limit address delta supported to 32 bits or 4 billion approximately. | |
1913 | * Possibly, this should be visible to the user through a datapath | |
1914 | * support check, however the practical impact is probably nil. */ | |
1915 | diff = 0xfffffffe; | |
1916 | } | |
dec0dbbc | 1917 | |
286de272 DB |
1918 | if (diff > 0xfffffffe) { |
1919 | diff = 0xfffffffe; | |
1920 | } | |
1921 | return diff; | |
1922 | } | |
1923 | ||
1924 | /* This function must be used in tandem with nat_ipv6_addrs_delta(), which | |
1925 | * restricts the input parameters. */ | |
a489b168 | 1926 | static void |
cda1b109 | 1927 | nat_ipv6_addr_increment(struct in6_addr *ipv6, uint32_t increment) |
286de272 | 1928 | { |
cda1b109 DB |
1929 | uint8_t *ipv6_hi = &ipv6->s6_addr[0]; |
1930 | uint8_t *ipv6_lo = &ipv6->s6_addr[0] + sizeof(ovs_be64); | |
286de272 DB |
1931 | ovs_be64 addr6_64_hi; |
1932 | ovs_be64 addr6_64_lo; | |
1933 | memcpy(&addr6_64_hi, ipv6_hi, sizeof addr6_64_hi); | |
1934 | memcpy(&addr6_64_lo, ipv6_lo, sizeof addr6_64_lo); | |
1935 | ||
1936 | if (UINT64_MAX - increment >= ntohll(addr6_64_lo)) { | |
1937 | addr6_64_lo = htonll(increment + ntohll(addr6_64_lo)); | |
1938 | } else if (addr6_64_hi != OVS_BE64_MAX) { | |
1939 | addr6_64_hi = htonll(1 + ntohll(addr6_64_hi)); | |
1940 | addr6_64_lo = htonll(increment - (UINT64_MAX - | |
1941 | ntohll(addr6_64_lo) + 1)); | |
1942 | } else { | |
1943 | OVS_NOT_REACHED(); | |
1944 | } | |
1945 | ||
1946 | memcpy(ipv6_hi, &addr6_64_hi, sizeof addr6_64_hi); | |
1947 | memcpy(ipv6_lo, &addr6_64_lo, sizeof addr6_64_lo); | |
286de272 DB |
1948 | } |
1949 | ||
1950 | static uint32_t | |
1951 | nat_range_hash(const struct conn *conn, uint32_t basis) | |
1952 | { | |
1953 | uint32_t hash = basis; | |
286de272 | 1954 | |
92edd073 DB |
1955 | hash = ct_addr_hash_add(hash, &conn->nat_info->min_addr); |
1956 | hash = ct_addr_hash_add(hash, &conn->nat_info->max_addr); | |
1957 | hash = hash_add(hash, | |
1958 | (conn->nat_info->max_port << 16) | |
1959 | | conn->nat_info->min_port); | |
92edd073 DB |
1960 | hash = ct_endpoint_hash_add(hash, &conn->key.src); |
1961 | hash = ct_endpoint_hash_add(hash, &conn->key.dst); | |
286de272 DB |
1962 | hash = hash_add(hash, (OVS_FORCE uint32_t) conn->key.dl_type); |
1963 | hash = hash_add(hash, conn->key.nw_proto); | |
1964 | hash = hash_add(hash, conn->key.zone); | |
92edd073 DB |
1965 | |
1966 | /* The purpose of the second parameter is to distinguish hashes of data of | |
1967 | * different length; our data always has the same length so there is no | |
1968 | * value in counting. */ | |
1969 | return hash_finish(hash, 0); | |
286de272 DB |
1970 | } |
1971 | ||
1972 | static bool | |
1973 | nat_select_range_tuple(struct conntrack *ct, const struct conn *conn, | |
1974 | struct conn *nat_conn) | |
1975 | { | |
bd5e81a0 DB |
1976 | enum { MIN_NAT_EPHEMERAL_PORT = 1024, |
1977 | MAX_NAT_EPHEMERAL_PORT = 65535 }; | |
286de272 DB |
1978 | |
1979 | uint16_t min_port; | |
1980 | uint16_t max_port; | |
1981 | uint16_t first_port; | |
286de272 DB |
1982 | uint32_t hash = nat_range_hash(conn, ct->hash_basis); |
1983 | ||
1984 | if ((conn->nat_info->nat_action & NAT_ACTION_SRC) && | |
1985 | (!(conn->nat_info->nat_action & NAT_ACTION_SRC_PORT))) { | |
1986 | min_port = ntohs(conn->key.src.port); | |
1987 | max_port = ntohs(conn->key.src.port); | |
1988 | first_port = min_port; | |
1989 | } else if ((conn->nat_info->nat_action & NAT_ACTION_DST) && | |
1990 | (!(conn->nat_info->nat_action & NAT_ACTION_DST_PORT))) { | |
1991 | min_port = ntohs(conn->key.dst.port); | |
1992 | max_port = ntohs(conn->key.dst.port); | |
1993 | first_port = min_port; | |
1994 | } else { | |
1995 | uint16_t deltap = conn->nat_info->max_port - conn->nat_info->min_port; | |
1996 | uint32_t port_index = hash % (deltap + 1); | |
1997 | first_port = conn->nat_info->min_port + port_index; | |
1998 | min_port = conn->nat_info->min_port; | |
1999 | max_port = conn->nat_info->max_port; | |
2000 | } | |
2001 | ||
2002 | uint32_t deltaa = 0; | |
2003 | uint32_t address_index; | |
cda1b109 | 2004 | union ct_addr ct_addr; |
286de272 | 2005 | memset(&ct_addr, 0, sizeof ct_addr); |
cda1b109 | 2006 | union ct_addr max_ct_addr; |
286de272 DB |
2007 | memset(&max_ct_addr, 0, sizeof max_ct_addr); |
2008 | max_ct_addr = conn->nat_info->max_addr; | |
2009 | ||
2010 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 DB |
2011 | deltaa = ntohl(conn->nat_info->max_addr.ipv4) - |
2012 | ntohl(conn->nat_info->min_addr.ipv4); | |
286de272 | 2013 | address_index = hash % (deltaa + 1); |
cda1b109 DB |
2014 | ct_addr.ipv4 = htonl( |
2015 | ntohl(conn->nat_info->min_addr.ipv4) + address_index); | |
286de272 | 2016 | } else { |
cda1b109 DB |
2017 | deltaa = nat_ipv6_addrs_delta(&conn->nat_info->min_addr.ipv6, |
2018 | &conn->nat_info->max_addr.ipv6); | |
286de272 DB |
2019 | /* deltaa must be within 32 bits for full hash coverage. A 64 or |
2020 | * 128 bit hash is unnecessary and hence not used here. Most code | |
2021 | * is kept common with V4; nat_ipv6_addrs_delta() will do the | |
2022 | * enforcement via max_ct_addr. */ | |
2023 | max_ct_addr = conn->nat_info->min_addr; | |
cda1b109 | 2024 | nat_ipv6_addr_increment(&max_ct_addr.ipv6, deltaa); |
286de272 | 2025 | address_index = hash % (deltaa + 1); |
cda1b109 DB |
2026 | ct_addr.ipv6 = conn->nat_info->min_addr.ipv6; |
2027 | nat_ipv6_addr_increment(&ct_addr.ipv6, address_index); | |
286de272 DB |
2028 | } |
2029 | ||
2030 | uint16_t port = first_port; | |
2031 | bool all_ports_tried = false; | |
32b2c81f DB |
2032 | /* For DNAT or for specified port ranges, we don't use ephemeral ports. */ |
2033 | bool ephemeral_ports_tried | |
2034 | = conn->nat_info->nat_action & NAT_ACTION_DST || | |
2035 | conn->nat_info->nat_action & NAT_ACTION_SRC_PORT | |
2036 | ? true : false; | |
cda1b109 | 2037 | union ct_addr first_addr = ct_addr; |
4cd0481c DB |
2038 | bool pat_enabled = conn->key.nw_proto != IPPROTO_ICMP && |
2039 | conn->key.nw_proto != IPPROTO_ICMPV6; | |
286de272 DB |
2040 | |
2041 | while (true) { | |
2042 | if (conn->nat_info->nat_action & NAT_ACTION_SRC) { | |
2043 | nat_conn->rev_key.dst.addr = ct_addr; | |
286de272 DB |
2044 | nat_conn->rev_key.dst.port = htons(port); |
2045 | } else { | |
1c8689d7 | 2046 | nat_conn->rev_key.src.addr = ct_addr; |
286de272 DB |
2047 | nat_conn->rev_key.src.port = htons(port); |
2048 | } | |
2049 | ||
967bb5c5 DB |
2050 | uint32_t conn_hash = conn_key_hash(&nat_conn->rev_key, |
2051 | ct->hash_basis); | |
2052 | bool found = conn_key_lookup(ct, &nat_conn->rev_key, conn_hash, | |
2053 | time_msec(), NULL, NULL); | |
2054 | if (!found) { | |
286de272 | 2055 | return true; |
4cd0481c | 2056 | } else if (pat_enabled && !all_ports_tried) { |
286de272 DB |
2057 | if (min_port == max_port) { |
2058 | all_ports_tried = true; | |
2059 | } else if (port == max_port) { | |
2060 | port = min_port; | |
2061 | } else { | |
2062 | port++; | |
2063 | } | |
2064 | if (port == first_port) { | |
2065 | all_ports_tried = true; | |
2066 | } | |
2067 | } else { | |
2068 | if (memcmp(&ct_addr, &max_ct_addr, sizeof ct_addr)) { | |
2069 | if (conn->key.dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2070 | ct_addr.ipv4 = htonl(ntohl(ct_addr.ipv4) + 1); |
286de272 | 2071 | } else { |
cda1b109 | 2072 | nat_ipv6_addr_increment(&ct_addr.ipv6, 1); |
286de272 DB |
2073 | } |
2074 | } else { | |
2075 | ct_addr = conn->nat_info->min_addr; | |
2076 | } | |
2077 | if (!memcmp(&ct_addr, &first_addr, sizeof ct_addr)) { | |
4cd0481c | 2078 | if (pat_enabled && !ephemeral_ports_tried) { |
ac04639a | 2079 | ephemeral_ports_tried = true; |
286de272 | 2080 | ct_addr = conn->nat_info->min_addr; |
8417e688 | 2081 | first_addr = ct_addr; |
286de272 DB |
2082 | min_port = MIN_NAT_EPHEMERAL_PORT; |
2083 | max_port = MAX_NAT_EPHEMERAL_PORT; | |
2084 | } else { | |
2085 | break; | |
2086 | } | |
2087 | } | |
2088 | first_port = min_port; | |
2089 | port = first_port; | |
2090 | all_ports_tried = false; | |
2091 | } | |
2092 | } | |
2093 | return false; | |
2094 | } | |
2095 | ||
a489b168 | 2096 | static enum ct_update_res |
967bb5c5 DB |
2097 | conn_update(struct conntrack *ct, struct conn *conn, struct dp_packet *pkt, |
2098 | struct conn_lookup_ctx *ctx, long long now) | |
a489b168 | 2099 | { |
967bb5c5 DB |
2100 | ovs_mutex_lock(&conn->lock); |
2101 | enum ct_update_res update_res = | |
2102 | l4_protos[conn->key.nw_proto]->conn_update(ct, conn, pkt, ctx->reply, | |
2103 | now); | |
2104 | ovs_mutex_unlock(&conn->lock); | |
2105 | return update_res; | |
a489b168 DDP |
2106 | } |
2107 | ||
2108 | static bool | |
2109 | conn_expired(struct conn *conn, long long now) | |
2110 | { | |
286de272 | 2111 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
967bb5c5 DB |
2112 | ovs_mutex_lock(&conn->lock); |
2113 | bool expired = now >= conn->expiration ? true : false; | |
2114 | ovs_mutex_unlock(&conn->lock); | |
2115 | return expired; | |
286de272 DB |
2116 | } |
2117 | return false; | |
a489b168 DDP |
2118 | } |
2119 | ||
2120 | static bool | |
2121 | valid_new(struct dp_packet *pkt, struct conn_key *key) | |
2122 | { | |
2123 | return l4_protos[key->nw_proto]->valid_new(pkt); | |
2124 | } | |
2125 | ||
2126 | static struct conn * | |
967bb5c5 DB |
2127 | new_conn(struct conntrack *ct, struct dp_packet *pkt, struct conn_key *key, |
2128 | long long now) | |
a489b168 | 2129 | { |
967bb5c5 | 2130 | return l4_protos[key->nw_proto]->new_conn(ct, pkt, now); |
a489b168 DDP |
2131 | } |
2132 | ||
2133 | static void | |
967bb5c5 | 2134 | delete_conn_cmn(struct conn *conn) |
a489b168 | 2135 | { |
286de272 | 2136 | free(conn->nat_info); |
bd5e81a0 | 2137 | free(conn->alg); |
a489b168 DDP |
2138 | free(conn); |
2139 | } | |
967bb5c5 DB |
2140 | |
2141 | static void | |
2142 | delete_conn(struct conn *conn) | |
2143 | { | |
2144 | ovs_assert(conn->conn_type == CT_CONN_TYPE_DEFAULT); | |
2145 | ovs_mutex_destroy(&conn->lock); | |
2146 | free(conn->nat_conn); | |
2147 | delete_conn_cmn(conn); | |
2148 | } | |
2149 | ||
2150 | /* Only used by conn_clean_one(). */ | |
2151 | static void | |
2152 | delete_conn_one(struct conn *conn) | |
2153 | { | |
2154 | if (conn->conn_type == CT_CONN_TYPE_DEFAULT) { | |
2155 | ovs_mutex_destroy(&conn->lock); | |
2156 | } | |
2157 | delete_conn_cmn(conn); | |
2158 | } | |
4d4e68ed | 2159 | \f |
271e48a0 YHW |
2160 | /* Convert a conntrack address 'a' into an IP address 'b' based on 'dl_type'. |
2161 | * | |
2162 | * Note that 'dl_type' should be either "ETH_TYPE_IP" or "ETH_TYPE_IPv6" | |
2163 | * in network-byte order. */ | |
4d4e68ed | 2164 | static void |
cda1b109 | 2165 | ct_endpoint_to_ct_dpif_inet_addr(const union ct_addr *a, |
4d4e68ed DDP |
2166 | union ct_dpif_inet_addr *b, |
2167 | ovs_be16 dl_type) | |
2168 | { | |
2169 | if (dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2170 | b->ip = a->ipv4; |
4d4e68ed | 2171 | } else if (dl_type == htons(ETH_TYPE_IPV6)){ |
cda1b109 | 2172 | b->in6 = a->ipv6; |
4d4e68ed DDP |
2173 | } |
2174 | } | |
2175 | ||
271e48a0 YHW |
2176 | /* Convert an IP address 'a' into a conntrack address 'b' based on 'dl_type'. |
2177 | * | |
2178 | * Note that 'dl_type' should be either "ETH_TYPE_IP" or "ETH_TYPE_IPv6" | |
2179 | * in network-byte order. */ | |
2180 | static void | |
2181 | ct_dpif_inet_addr_to_ct_endpoint(const union ct_dpif_inet_addr *a, | |
cda1b109 | 2182 | union ct_addr *b, ovs_be16 dl_type) |
271e48a0 YHW |
2183 | { |
2184 | if (dl_type == htons(ETH_TYPE_IP)) { | |
cda1b109 | 2185 | b->ipv4 = a->ip; |
271e48a0 | 2186 | } else if (dl_type == htons(ETH_TYPE_IPV6)){ |
cda1b109 | 2187 | b->ipv6 = a->in6; |
271e48a0 YHW |
2188 | } |
2189 | } | |
2190 | ||
4d4e68ed DDP |
2191 | static void |
2192 | conn_key_to_tuple(const struct conn_key *key, struct ct_dpif_tuple *tuple) | |
2193 | { | |
2194 | if (key->dl_type == htons(ETH_TYPE_IP)) { | |
2195 | tuple->l3_type = AF_INET; | |
2196 | } else if (key->dl_type == htons(ETH_TYPE_IPV6)) { | |
2197 | tuple->l3_type = AF_INET6; | |
2198 | } | |
2199 | tuple->ip_proto = key->nw_proto; | |
2200 | ct_endpoint_to_ct_dpif_inet_addr(&key->src.addr, &tuple->src, | |
2201 | key->dl_type); | |
2202 | ct_endpoint_to_ct_dpif_inet_addr(&key->dst.addr, &tuple->dst, | |
2203 | key->dl_type); | |
2204 | ||
2205 | if (key->nw_proto == IPPROTO_ICMP || key->nw_proto == IPPROTO_ICMPV6) { | |
b269a122 DDP |
2206 | tuple->icmp_id = key->src.icmp_id; |
2207 | tuple->icmp_type = key->src.icmp_type; | |
2208 | tuple->icmp_code = key->src.icmp_code; | |
4d4e68ed DDP |
2209 | } else { |
2210 | tuple->src_port = key->src.port; | |
2211 | tuple->dst_port = key->dst.port; | |
2212 | } | |
2213 | } | |
2214 | ||
271e48a0 YHW |
2215 | static void |
2216 | tuple_to_conn_key(const struct ct_dpif_tuple *tuple, uint16_t zone, | |
2217 | struct conn_key *key) | |
2218 | { | |
2219 | if (tuple->l3_type == AF_INET) { | |
2220 | key->dl_type = htons(ETH_TYPE_IP); | |
2221 | } else if (tuple->l3_type == AF_INET6) { | |
2222 | key->dl_type = htons(ETH_TYPE_IPV6); | |
2223 | } | |
2224 | key->nw_proto = tuple->ip_proto; | |
2225 | ct_dpif_inet_addr_to_ct_endpoint(&tuple->src, &key->src.addr, | |
2226 | key->dl_type); | |
2227 | ct_dpif_inet_addr_to_ct_endpoint(&tuple->dst, &key->dst.addr, | |
2228 | key->dl_type); | |
2229 | ||
2230 | if (tuple->ip_proto == IPPROTO_ICMP || tuple->ip_proto == IPPROTO_ICMPV6) { | |
2231 | key->src.icmp_id = tuple->icmp_id; | |
2232 | key->src.icmp_type = tuple->icmp_type; | |
2233 | key->src.icmp_code = tuple->icmp_code; | |
2234 | key->dst.icmp_id = tuple->icmp_id; | |
2235 | key->dst.icmp_type = reverse_icmp_type(tuple->icmp_type); | |
2236 | key->dst.icmp_code = tuple->icmp_code; | |
2237 | } else { | |
2238 | key->src.port = tuple->src_port; | |
2239 | key->dst.port = tuple->dst_port; | |
2240 | } | |
2241 | key->zone = zone; | |
2242 | } | |
2243 | ||
4d4e68ed DDP |
2244 | static void |
2245 | conn_to_ct_dpif_entry(const struct conn *conn, struct ct_dpif_entry *entry, | |
f1a0469e | 2246 | long long now) |
4d4e68ed | 2247 | { |
4d4e68ed DDP |
2248 | memset(entry, 0, sizeof *entry); |
2249 | conn_key_to_tuple(&conn->key, &entry->tuple_orig); | |
2250 | conn_key_to_tuple(&conn->rev_key, &entry->tuple_reply); | |
2251 | ||
2252 | entry->zone = conn->key.zone; | |
4d4e68ed | 2253 | |
967bb5c5 DB |
2254 | ovs_mutex_lock(&conn->lock); |
2255 | entry->mark = conn->mark; | |
286de272 | 2256 | memcpy(&entry->labels, &conn->label, sizeof entry->labels); |
4d4e68ed | 2257 | |
dec0dbbc | 2258 | long long expiration = conn->expiration - now; |
4d4e68ed | 2259 | |
dec0dbbc | 2260 | struct ct_l4_proto *class = l4_protos[conn->key.nw_proto]; |
4d4e68ed DDP |
2261 | if (class->conn_get_protoinfo) { |
2262 | class->conn_get_protoinfo(conn, &entry->protoinfo); | |
2263 | } | |
f1a0469e | 2264 | ovs_mutex_unlock(&conn->lock); |
bd5e81a0 | 2265 | |
f1a0469e | 2266 | entry->timeout = (expiration > 0) ? expiration / 1000 : 0; |
bd5e81a0 DB |
2267 | |
2268 | if (conn->alg) { | |
2269 | /* Caller is responsible for freeing. */ | |
2270 | entry->helper.name = xstrdup(conn->alg); | |
2271 | } | |
4d4e68ed DDP |
2272 | } |
2273 | ||
4ea96698 DB |
2274 | struct ipf * |
2275 | conntrack_ipf_ctx(struct conntrack *ct) | |
2276 | { | |
2277 | return ct->ipf; | |
2278 | } | |
2279 | ||
4d4e68ed DDP |
2280 | int |
2281 | conntrack_dump_start(struct conntrack *ct, struct conntrack_dump *dump, | |
ded30c74 | 2282 | const uint16_t *pzone, int *ptot_bkts) |
4d4e68ed DDP |
2283 | { |
2284 | memset(dump, 0, sizeof(*dump)); | |
dec0dbbc | 2285 | |
4d4e68ed DDP |
2286 | if (pzone) { |
2287 | dump->zone = *pzone; | |
2288 | dump->filter_zone = true; | |
2289 | } | |
4d4e68ed | 2290 | |
dec0dbbc | 2291 | dump->ct = ct; |
967bb5c5 | 2292 | *ptot_bkts = 1; /* Need to clean up the callers. */ |
4d4e68ed DDP |
2293 | return 0; |
2294 | } | |
2295 | ||
2296 | int | |
2297 | conntrack_dump_next(struct conntrack_dump *dump, struct ct_dpif_entry *entry) | |
2298 | { | |
2299 | struct conntrack *ct = dump->ct; | |
2300 | long long now = time_msec(); | |
2301 | ||
967bb5c5 DB |
2302 | for (;;) { |
2303 | struct cmap_node *cm_node = cmap_next_position(&ct->conns, | |
2304 | &dump->cm_pos); | |
2305 | if (!cm_node) { | |
2306 | break; | |
4d4e68ed | 2307 | } |
967bb5c5 DB |
2308 | struct conn *conn; |
2309 | INIT_CONTAINER(conn, cm_node, cm_node); | |
2310 | if ((!dump->filter_zone || conn->key.zone == dump->zone) && | |
2311 | (conn->conn_type != CT_CONN_TYPE_UN_NAT)) { | |
f1a0469e | 2312 | conn_to_ct_dpif_entry(conn, entry, now); |
4d4e68ed DDP |
2313 | return 0; |
2314 | } | |
2315 | } | |
967bb5c5 | 2316 | |
4d4e68ed DDP |
2317 | return EOF; |
2318 | } | |
2319 | ||
2320 | int | |
2321 | conntrack_dump_done(struct conntrack_dump *dump OVS_UNUSED) | |
2322 | { | |
2323 | return 0; | |
2324 | } | |
5d9cbb4c DDP |
2325 | |
2326 | int | |
2327 | conntrack_flush(struct conntrack *ct, const uint16_t *zone) | |
2328 | { | |
967bb5c5 DB |
2329 | struct conn *conn; |
2330 | ||
2331 | ovs_mutex_lock(&ct->ct_lock); | |
2332 | CMAP_FOR_EACH (conn, cm_node, &ct->conns) { | |
2333 | if (!zone || *zone == conn->key.zone) { | |
2334 | conn_clean_one(ct, conn); | |
5d9cbb4c | 2335 | } |
5d9cbb4c | 2336 | } |
967bb5c5 | 2337 | ovs_mutex_unlock(&ct->ct_lock); |
bd5e81a0 | 2338 | |
5d9cbb4c DDP |
2339 | return 0; |
2340 | } | |
bd5e81a0 | 2341 | |
271e48a0 YHW |
2342 | int |
2343 | conntrack_flush_tuple(struct conntrack *ct, const struct ct_dpif_tuple *tuple, | |
2344 | uint16_t zone) | |
2345 | { | |
2346 | struct conn_lookup_ctx ctx; | |
2347 | int error = 0; | |
2348 | ||
2349 | memset(&ctx, 0, sizeof(ctx)); | |
2350 | tuple_to_conn_key(tuple, zone, &ctx.key); | |
2351 | ctx.hash = conn_key_hash(&ctx.key, ct->hash_basis); | |
967bb5c5 DB |
2352 | ovs_mutex_lock(&ct->ct_lock); |
2353 | conn_key_lookup(ct, &ctx.key, ctx.hash, time_msec(), &ctx.conn, | |
2354 | &ctx.reply); | |
271e48a0 | 2355 | |
a1d5eeff | 2356 | if (ctx.conn && ctx.conn->conn_type == CT_CONN_TYPE_DEFAULT) { |
967bb5c5 | 2357 | conn_clean(ct, ctx.conn); |
271e48a0 | 2358 | } else { |
a1d5eeff | 2359 | VLOG_WARN("Must flush tuple using the original pre-NATed tuple"); |
271e48a0 YHW |
2360 | error = ENOENT; |
2361 | } | |
967bb5c5 DB |
2362 | |
2363 | ovs_mutex_unlock(&ct->ct_lock); | |
271e48a0 YHW |
2364 | return error; |
2365 | } | |
2366 | ||
c92339ad DB |
2367 | int |
2368 | conntrack_set_maxconns(struct conntrack *ct, uint32_t maxconns) | |
2369 | { | |
2370 | atomic_store_relaxed(&ct->n_conn_limit, maxconns); | |
2371 | return 0; | |
2372 | } | |
2373 | ||
2374 | int | |
2375 | conntrack_get_maxconns(struct conntrack *ct, uint32_t *maxconns) | |
2376 | { | |
2377 | atomic_read_relaxed(&ct->n_conn_limit, maxconns); | |
2378 | return 0; | |
2379 | } | |
2380 | ||
875075b3 DB |
2381 | int |
2382 | conntrack_get_nconns(struct conntrack *ct, uint32_t *nconns) | |
2383 | { | |
2384 | *nconns = atomic_count_get(&ct->n_conn); | |
2385 | return 0; | |
2386 | } | |
2387 | ||
bd5e81a0 DB |
2388 | /* This function must be called with the ct->resources read lock taken. */ |
2389 | static struct alg_exp_node * | |
be38342d DB |
2390 | expectation_lookup(struct hmap *alg_expectations, const struct conn_key *key, |
2391 | uint32_t basis, bool src_ip_wc) | |
bd5e81a0 | 2392 | { |
c3f6bae2 DB |
2393 | struct conn_key check_key; |
2394 | memcpy(&check_key, key, sizeof check_key); | |
bd5e81a0 | 2395 | check_key.src.port = ALG_WC_SRC_PORT; |
dec0dbbc | 2396 | |
be38342d DB |
2397 | if (src_ip_wc) { |
2398 | memset(&check_key.src.addr, 0, sizeof check_key.src.addr); | |
2399 | } | |
dec0dbbc | 2400 | |
bd5e81a0 DB |
2401 | struct alg_exp_node *alg_exp_node; |
2402 | ||
bd5e81a0 | 2403 | HMAP_FOR_EACH_WITH_HASH (alg_exp_node, node, |
dec0dbbc | 2404 | conn_key_hash(&check_key, basis), |
bd5e81a0 DB |
2405 | alg_expectations) { |
2406 | if (!conn_key_cmp(&alg_exp_node->key, &check_key)) { | |
2407 | return alg_exp_node; | |
2408 | } | |
2409 | } | |
2410 | return NULL; | |
2411 | } | |
2412 | ||
4417ca3d DB |
2413 | /* This function must be called with the ct->resources write lock taken. */ |
2414 | static void | |
2415 | expectation_remove(struct hmap *alg_expectations, | |
2416 | const struct conn_key *key, uint32_t basis) | |
2417 | { | |
2418 | struct alg_exp_node *alg_exp_node; | |
2419 | ||
2420 | HMAP_FOR_EACH_WITH_HASH (alg_exp_node, node, conn_key_hash(key, basis), | |
2421 | alg_expectations) { | |
2422 | if (!conn_key_cmp(&alg_exp_node->key, key)) { | |
2423 | hmap_remove(alg_expectations, &alg_exp_node->node); | |
2424 | break; | |
2425 | } | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | /* This function must be called with the ct->resources read lock taken. */ | |
2430 | static struct alg_exp_node * | |
2431 | expectation_ref_lookup_unique(const struct hindex *alg_expectation_refs, | |
2432 | const struct conn_key *master_key, | |
2433 | const struct conn_key *alg_exp_key, | |
2434 | uint32_t basis) | |
2435 | { | |
2436 | struct alg_exp_node *alg_exp_node; | |
2437 | ||
2438 | HINDEX_FOR_EACH_WITH_HASH (alg_exp_node, node_ref, | |
2439 | conn_key_hash(master_key, basis), | |
2440 | alg_expectation_refs) { | |
2441 | if (!conn_key_cmp(&alg_exp_node->master_key, master_key) && | |
2442 | !conn_key_cmp(&alg_exp_node->key, alg_exp_key)) { | |
2443 | return alg_exp_node; | |
2444 | } | |
2445 | } | |
2446 | return NULL; | |
2447 | } | |
2448 | ||
2449 | /* This function must be called with the ct->resources write lock taken. */ | |
2450 | static void | |
2451 | expectation_ref_create(struct hindex *alg_expectation_refs, | |
2452 | struct alg_exp_node *alg_exp_node, | |
2453 | uint32_t basis) | |
2454 | { | |
2455 | if (!expectation_ref_lookup_unique(alg_expectation_refs, | |
2456 | &alg_exp_node->master_key, | |
2457 | &alg_exp_node->key, basis)) { | |
2458 | hindex_insert(alg_expectation_refs, &alg_exp_node->node_ref, | |
2459 | conn_key_hash(&alg_exp_node->master_key, basis)); | |
2460 | } | |
2461 | } | |
2462 | ||
2463 | static void | |
967bb5c5 | 2464 | expectation_clean(struct conntrack *ct, const struct conn_key *master_key) |
4417ca3d | 2465 | { |
967bb5c5 | 2466 | ovs_rwlock_wrlock(&ct->resources_lock); |
4417ca3d DB |
2467 | |
2468 | struct alg_exp_node *node, *next; | |
2469 | HINDEX_FOR_EACH_WITH_HASH_SAFE (node, next, node_ref, | |
967bb5c5 | 2470 | conn_key_hash(master_key, ct->hash_basis), |
4417ca3d DB |
2471 | &ct->alg_expectation_refs) { |
2472 | if (!conn_key_cmp(&node->master_key, master_key)) { | |
967bb5c5 DB |
2473 | expectation_remove(&ct->alg_expectations, &node->key, |
2474 | ct->hash_basis); | |
4417ca3d DB |
2475 | hindex_remove(&ct->alg_expectation_refs, &node->node_ref); |
2476 | free(node); | |
2477 | } | |
2478 | } | |
2479 | ||
967bb5c5 | 2480 | ovs_rwlock_unlock(&ct->resources_lock); |
4417ca3d DB |
2481 | } |
2482 | ||
bd5e81a0 | 2483 | static void |
be38342d DB |
2484 | expectation_create(struct conntrack *ct, ovs_be16 dst_port, |
2485 | const struct conn *master_conn, bool reply, bool src_ip_wc, | |
2486 | bool skip_nat) | |
bd5e81a0 | 2487 | { |
cda1b109 DB |
2488 | union ct_addr src_addr; |
2489 | union ct_addr dst_addr; | |
2490 | union ct_addr alg_nat_repl_addr; | |
be38342d | 2491 | struct alg_exp_node *alg_exp_node = xzalloc(sizeof *alg_exp_node); |
bd5e81a0 | 2492 | |
be38342d | 2493 | if (reply) { |
bd5e81a0 DB |
2494 | src_addr = master_conn->key.src.addr; |
2495 | dst_addr = master_conn->key.dst.addr; | |
efa29a89 | 2496 | alg_exp_node->nat_rpl_dst = true; |
be38342d DB |
2497 | if (skip_nat) { |
2498 | alg_nat_repl_addr = dst_addr; | |
efa29a89 DM |
2499 | } else if (master_conn->nat_info && |
2500 | master_conn->nat_info->nat_action & NAT_ACTION_DST) { | |
2501 | alg_nat_repl_addr = master_conn->rev_key.src.addr; | |
2502 | alg_exp_node->nat_rpl_dst = false; | |
be38342d DB |
2503 | } else { |
2504 | alg_nat_repl_addr = master_conn->rev_key.dst.addr; | |
2505 | } | |
be38342d DB |
2506 | } else { |
2507 | src_addr = master_conn->rev_key.src.addr; | |
2508 | dst_addr = master_conn->rev_key.dst.addr; | |
efa29a89 | 2509 | alg_exp_node->nat_rpl_dst = false; |
be38342d DB |
2510 | if (skip_nat) { |
2511 | alg_nat_repl_addr = src_addr; | |
efa29a89 DM |
2512 | } else if (master_conn->nat_info && |
2513 | master_conn->nat_info->nat_action & NAT_ACTION_DST) { | |
2514 | alg_nat_repl_addr = master_conn->key.dst.addr; | |
2515 | alg_exp_node->nat_rpl_dst = true; | |
be38342d DB |
2516 | } else { |
2517 | alg_nat_repl_addr = master_conn->key.src.addr; | |
2518 | } | |
be38342d DB |
2519 | } |
2520 | if (src_ip_wc) { | |
2521 | memset(&src_addr, 0, sizeof src_addr); | |
bd5e81a0 DB |
2522 | } |
2523 | ||
bd5e81a0 DB |
2524 | alg_exp_node->key.dl_type = master_conn->key.dl_type; |
2525 | alg_exp_node->key.nw_proto = master_conn->key.nw_proto; | |
2526 | alg_exp_node->key.zone = master_conn->key.zone; | |
2527 | alg_exp_node->key.src.addr = src_addr; | |
2528 | alg_exp_node->key.dst.addr = dst_addr; | |
2529 | alg_exp_node->key.src.port = ALG_WC_SRC_PORT; | |
2530 | alg_exp_node->key.dst.port = dst_port; | |
2531 | alg_exp_node->master_mark = master_conn->mark; | |
2532 | alg_exp_node->master_label = master_conn->label; | |
82b9ac94 DB |
2533 | memcpy(&alg_exp_node->master_key, &master_conn->key, |
2534 | sizeof alg_exp_node->master_key); | |
bd5e81a0 DB |
2535 | /* Take the write lock here because it is almost 100% |
2536 | * likely that the lookup will fail and | |
2537 | * expectation_create() will be called below. */ | |
967bb5c5 | 2538 | ovs_rwlock_wrlock(&ct->resources_lock); |
bd5e81a0 | 2539 | struct alg_exp_node *alg_exp = expectation_lookup( |
be38342d | 2540 | &ct->alg_expectations, &alg_exp_node->key, ct->hash_basis, src_ip_wc); |
bd5e81a0 DB |
2541 | if (alg_exp) { |
2542 | free(alg_exp_node); | |
967bb5c5 | 2543 | ovs_rwlock_unlock(&ct->resources_lock); |
bd5e81a0 DB |
2544 | return; |
2545 | } | |
2546 | ||
2547 | alg_exp_node->alg_nat_repl_addr = alg_nat_repl_addr; | |
4417ca3d | 2548 | hmap_insert(&ct->alg_expectations, &alg_exp_node->node, |
dec0dbbc | 2549 | conn_key_hash(&alg_exp_node->key, ct->hash_basis)); |
4417ca3d DB |
2550 | expectation_ref_create(&ct->alg_expectation_refs, alg_exp_node, |
2551 | ct->hash_basis); | |
967bb5c5 | 2552 | ovs_rwlock_unlock(&ct->resources_lock); |
bd5e81a0 DB |
2553 | } |
2554 | ||
bd5e81a0 DB |
2555 | static void |
2556 | replace_substring(char *substr, uint8_t substr_size, | |
2557 | uint8_t total_size, char *rep_str, | |
2558 | uint8_t rep_str_size) | |
2559 | { | |
2560 | memmove(substr + rep_str_size, substr + substr_size, | |
2561 | total_size - substr_size); | |
2562 | memcpy(substr, rep_str, rep_str_size); | |
2563 | } | |
2564 | ||
cd7c99a6 DB |
2565 | static void |
2566 | repl_bytes(char *str, char c1, char c2) | |
2567 | { | |
2568 | while (*str) { | |
2569 | if (*str == c1) { | |
2570 | *str = c2; | |
2571 | } | |
2572 | str++; | |
2573 | } | |
2574 | } | |
2575 | ||
2576 | static void | |
2577 | modify_packet(struct dp_packet *pkt, char *pkt_str, size_t size, | |
2578 | char *repl_str, size_t repl_size, | |
2579 | uint32_t orig_used_size) | |
2580 | { | |
2581 | replace_substring(pkt_str, size, | |
2582 | (const char *) dp_packet_tail(pkt) - pkt_str, | |
2583 | repl_str, repl_size); | |
2584 | dp_packet_set_size(pkt, orig_used_size + (int) repl_size - (int) size); | |
2585 | } | |
2586 | ||
bd5e81a0 DB |
2587 | /* Replace IPV4 address in FTP message with NATed address. */ |
2588 | static int | |
2589 | repl_ftp_v4_addr(struct dp_packet *pkt, ovs_be32 v4_addr_rep, | |
2590 | char *ftp_data_start, | |
cd7c99a6 DB |
2591 | size_t addr_offset_from_ftp_data_start, |
2592 | size_t addr_size OVS_UNUSED) | |
bd5e81a0 DB |
2593 | { |
2594 | enum { MAX_FTP_V4_NAT_DELTA = 8 }; | |
2595 | ||
2596 | /* Do conservative check for pathological MTU usage. */ | |
2597 | uint32_t orig_used_size = dp_packet_size(pkt); | |
cd7c99a6 DB |
2598 | if (orig_used_size + MAX_FTP_V4_NAT_DELTA > |
2599 | dp_packet_get_allocated(pkt)) { | |
2600 | ||
bd5e81a0 | 2601 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
cd7c99a6 DB |
2602 | VLOG_WARN_RL(&rl, "Unsupported effective MTU %u used with FTP V4", |
2603 | dp_packet_get_allocated(pkt)); | |
bd5e81a0 DB |
2604 | return 0; |
2605 | } | |
2606 | ||
cd7c99a6 DB |
2607 | char v4_addr_str[INET_ADDRSTRLEN] = {0}; |
2608 | ovs_assert(inet_ntop(AF_INET, &v4_addr_rep, v4_addr_str, | |
2609 | sizeof v4_addr_str)); | |
2610 | repl_bytes(v4_addr_str, '.', ','); | |
2611 | modify_packet(pkt, ftp_data_start + addr_offset_from_ftp_data_start, | |
2612 | addr_size, v4_addr_str, strlen(v4_addr_str), | |
2613 | orig_used_size); | |
2614 | return (int) strlen(v4_addr_str) - (int) addr_size; | |
bd5e81a0 DB |
2615 | } |
2616 | ||
2617 | static char * | |
2618 | skip_non_digits(char *str) | |
2619 | { | |
2620 | while (!isdigit(*str) && *str != 0) { | |
2621 | str++; | |
2622 | } | |
2623 | return str; | |
2624 | } | |
2625 | ||
2626 | static char * | |
2627 | terminate_number_str(char *str, uint8_t max_digits) | |
2628 | { | |
2629 | uint8_t digits_found = 0; | |
2630 | while (isdigit(*str) && digits_found <= max_digits) { | |
2631 | str++; | |
2632 | digits_found++; | |
2633 | } | |
2634 | ||
2635 | *str = 0; | |
2636 | return str; | |
2637 | } | |
2638 | ||
2639 | ||
2640 | static void | |
2641 | get_ftp_ctl_msg(struct dp_packet *pkt, char *ftp_msg) | |
2642 | { | |
2643 | struct tcp_header *th = dp_packet_l4(pkt); | |
2644 | char *tcp_hdr = (char *) th; | |
2645 | uint32_t tcp_payload_len = tcp_payload_length(pkt); | |
2646 | size_t tcp_payload_of_interest = MIN(tcp_payload_len, | |
2647 | LARGEST_FTP_MSG_OF_INTEREST); | |
2648 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
2649 | ||
2650 | ovs_strlcpy(ftp_msg, tcp_hdr + tcp_hdr_len, | |
2651 | tcp_payload_of_interest); | |
2652 | } | |
2653 | ||
2654 | static enum ftp_ctl_pkt | |
2655 | detect_ftp_ctl_type(const struct conn_lookup_ctx *ctx, | |
2656 | struct dp_packet *pkt) | |
2657 | { | |
bd5e81a0 DB |
2658 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; |
2659 | get_ftp_ctl_msg(pkt, ftp_msg); | |
dec0dbbc | 2660 | |
bd5e81a0 DB |
2661 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
2662 | if (strncasecmp(ftp_msg, FTP_EPRT_CMD, strlen(FTP_EPRT_CMD)) && | |
2663 | !strcasestr(ftp_msg, FTP_EPSV_REPLY)) { | |
2664 | return CT_FTP_CTL_OTHER; | |
2665 | } | |
2666 | } else { | |
2667 | if (strncasecmp(ftp_msg, FTP_PORT_CMD, strlen(FTP_PORT_CMD)) && | |
2668 | strncasecmp(ftp_msg, FTP_PASV_REPLY_CODE, | |
2669 | strlen(FTP_PASV_REPLY_CODE))) { | |
2670 | return CT_FTP_CTL_OTHER; | |
2671 | } | |
2672 | } | |
2673 | ||
2674 | return CT_FTP_CTL_INTEREST; | |
2675 | } | |
2676 | ||
2677 | static enum ftp_ctl_pkt | |
2678 | process_ftp_ctl_v4(struct conntrack *ct, | |
2679 | struct dp_packet *pkt, | |
2680 | const struct conn *conn_for_expectation, | |
4417ca3d | 2681 | ovs_be32 *v4_addr_rep, |
bd5e81a0 | 2682 | char **ftp_data_v4_start, |
cd7c99a6 DB |
2683 | size_t *addr_offset_from_ftp_data_start, |
2684 | size_t *addr_size) | |
bd5e81a0 DB |
2685 | { |
2686 | struct tcp_header *th = dp_packet_l4(pkt); | |
2687 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
2688 | char *tcp_hdr = (char *) th; | |
2689 | *ftp_data_v4_start = tcp_hdr + tcp_hdr_len; | |
2690 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; | |
2691 | get_ftp_ctl_msg(pkt, ftp_msg); | |
bd5e81a0 DB |
2692 | char *ftp = ftp_msg; |
2693 | enum ct_alg_mode mode; | |
dec0dbbc | 2694 | |
23bea975 | 2695 | if (!strncasecmp(ftp, FTP_PORT_CMD, strlen(FTP_PORT_CMD))) { |
bd5e81a0 DB |
2696 | ftp = ftp_msg + strlen(FTP_PORT_CMD); |
2697 | mode = CT_FTP_MODE_ACTIVE; | |
2698 | } else { | |
2699 | ftp = ftp_msg + strlen(FTP_PASV_REPLY_CODE); | |
2700 | mode = CT_FTP_MODE_PASSIVE; | |
2701 | } | |
2702 | ||
2703 | /* Find first space. */ | |
2704 | ftp = strchr(ftp, ' '); | |
2705 | if (!ftp) { | |
2706 | return CT_FTP_CTL_INVALID; | |
2707 | } | |
2708 | ||
2709 | /* Find the first digit, after space. */ | |
2710 | ftp = skip_non_digits(ftp); | |
2711 | if (*ftp == 0) { | |
2712 | return CT_FTP_CTL_INVALID; | |
2713 | } | |
2714 | ||
2715 | char *ip_addr_start = ftp; | |
2716 | *addr_offset_from_ftp_data_start = ip_addr_start - ftp_msg; | |
bd5e81a0 | 2717 | |
dec0dbbc | 2718 | uint8_t comma_count = 0; |
bd5e81a0 DB |
2719 | while (comma_count < 4 && *ftp) { |
2720 | if (*ftp == ',') { | |
2721 | comma_count++; | |
2722 | if (comma_count == 4) { | |
2723 | *ftp = 0; | |
2724 | } else { | |
2725 | *ftp = '.'; | |
2726 | } | |
2727 | } | |
2728 | ftp++; | |
2729 | } | |
2730 | if (comma_count != 4) { | |
2731 | return CT_FTP_CTL_INVALID; | |
2732 | } | |
2733 | ||
2734 | struct in_addr ip_addr; | |
2735 | int rc2 = inet_pton(AF_INET, ip_addr_start, &ip_addr); | |
2736 | if (rc2 != 1) { | |
2737 | return CT_FTP_CTL_INVALID; | |
2738 | } | |
2739 | ||
cd7c99a6 | 2740 | *addr_size = ftp - ip_addr_start - 1; |
bd5e81a0 DB |
2741 | char *save_ftp = ftp; |
2742 | ftp = terminate_number_str(ftp, MAX_FTP_PORT_DGTS); | |
2743 | if (!ftp) { | |
2744 | return CT_FTP_CTL_INVALID; | |
2745 | } | |
2746 | int value; | |
2747 | if (!str_to_int(save_ftp, 10, &value)) { | |
2748 | return CT_FTP_CTL_INVALID; | |
2749 | } | |
2750 | ||
2751 | /* This is derived from the L4 port maximum is 65535. */ | |
2752 | if (value > 255) { | |
2753 | return CT_FTP_CTL_INVALID; | |
2754 | } | |
2755 | ||
2756 | uint16_t port_hs = value; | |
2757 | port_hs <<= 8; | |
2758 | ||
2759 | /* Skip over comma. */ | |
2760 | ftp++; | |
2761 | save_ftp = ftp; | |
2762 | bool digit_found = false; | |
2763 | while (isdigit(*ftp)) { | |
2764 | ftp++; | |
2765 | digit_found = true; | |
2766 | } | |
2767 | if (!digit_found) { | |
2768 | return CT_FTP_CTL_INVALID; | |
2769 | } | |
2770 | *ftp = 0; | |
2771 | if (!str_to_int(save_ftp, 10, &value)) { | |
2772 | return CT_FTP_CTL_INVALID; | |
2773 | } | |
2774 | ||
2775 | if (value > 255) { | |
2776 | return CT_FTP_CTL_INVALID; | |
2777 | } | |
2778 | ||
78a0b272 | 2779 | port_hs |= value; |
bd5e81a0 DB |
2780 | ovs_be16 port = htons(port_hs); |
2781 | ovs_be32 conn_ipv4_addr; | |
2782 | ||
2783 | switch (mode) { | |
2784 | case CT_FTP_MODE_ACTIVE: | |
cda1b109 DB |
2785 | *v4_addr_rep = conn_for_expectation->rev_key.dst.addr.ipv4; |
2786 | conn_ipv4_addr = conn_for_expectation->key.src.addr.ipv4; | |
bd5e81a0 DB |
2787 | break; |
2788 | case CT_FTP_MODE_PASSIVE: | |
cda1b109 DB |
2789 | *v4_addr_rep = conn_for_expectation->key.dst.addr.ipv4; |
2790 | conn_ipv4_addr = conn_for_expectation->rev_key.src.addr.ipv4; | |
bd5e81a0 | 2791 | break; |
7be77cb0 | 2792 | case CT_TFTP_MODE: |
bd5e81a0 DB |
2793 | default: |
2794 | OVS_NOT_REACHED(); | |
2795 | } | |
2796 | ||
2797 | ovs_be32 ftp_ipv4_addr; | |
2798 | ftp_ipv4_addr = ip_addr.s_addr; | |
2799 | /* Although most servers will block this exploit, there may be some | |
2800 | * less well managed. */ | |
2801 | if (ftp_ipv4_addr != conn_ipv4_addr && ftp_ipv4_addr != *v4_addr_rep) { | |
2802 | return CT_FTP_CTL_INVALID; | |
2803 | } | |
2804 | ||
be38342d DB |
2805 | expectation_create(ct, port, conn_for_expectation, |
2806 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
bd5e81a0 DB |
2807 | return CT_FTP_CTL_INTEREST; |
2808 | } | |
2809 | ||
2810 | static char * | |
2811 | skip_ipv6_digits(char *str) | |
2812 | { | |
2813 | while (isxdigit(*str) || *str == ':' || *str == '.') { | |
2814 | str++; | |
2815 | } | |
2816 | return str; | |
2817 | } | |
2818 | ||
2819 | static enum ftp_ctl_pkt | |
2820 | process_ftp_ctl_v6(struct conntrack *ct, | |
2821 | struct dp_packet *pkt, | |
2822 | const struct conn *conn_for_expectation, | |
cda1b109 | 2823 | union ct_addr *v6_addr_rep, char **ftp_data_start, |
bd5e81a0 DB |
2824 | size_t *addr_offset_from_ftp_data_start, |
2825 | size_t *addr_size, enum ct_alg_mode *mode) | |
2826 | { | |
2827 | struct tcp_header *th = dp_packet_l4(pkt); | |
2828 | size_t tcp_hdr_len = TCP_OFFSET(th->tcp_ctl) * 4; | |
2829 | char *tcp_hdr = (char *) th; | |
2830 | char ftp_msg[LARGEST_FTP_MSG_OF_INTEREST + 1] = {0}; | |
bd5e81a0 DB |
2831 | get_ftp_ctl_msg(pkt, ftp_msg); |
2832 | *ftp_data_start = tcp_hdr + tcp_hdr_len; | |
bd5e81a0 DB |
2833 | char *ftp = ftp_msg; |
2834 | struct in6_addr ip6_addr; | |
dec0dbbc | 2835 | |
23bea975 | 2836 | if (!strncasecmp(ftp, FTP_EPRT_CMD, strlen(FTP_EPRT_CMD))) { |
bd5e81a0 DB |
2837 | ftp = ftp_msg + strlen(FTP_EPRT_CMD); |
2838 | ftp = skip_non_digits(ftp); | |
2839 | if (*ftp != FTP_AF_V6 || isdigit(ftp[1])) { | |
2840 | return CT_FTP_CTL_INVALID; | |
2841 | } | |
2842 | /* Jump over delimiter. */ | |
2843 | ftp += 2; | |
2844 | ||
bd5e81a0 | 2845 | memset(&ip6_addr, 0, sizeof ip6_addr); |
dec0dbbc | 2846 | char *ip_addr_start = ftp; |
bd5e81a0 DB |
2847 | *addr_offset_from_ftp_data_start = ip_addr_start - ftp_msg; |
2848 | ftp = skip_ipv6_digits(ftp); | |
2849 | *ftp = 0; | |
2850 | *addr_size = ftp - ip_addr_start; | |
2851 | int rc2 = inet_pton(AF_INET6, ip_addr_start, &ip6_addr); | |
2852 | if (rc2 != 1) { | |
2853 | return CT_FTP_CTL_INVALID; | |
2854 | } | |
2855 | ftp++; | |
2856 | *mode = CT_FTP_MODE_ACTIVE; | |
2857 | } else { | |
2858 | ftp = ftp_msg + strcspn(ftp_msg, "("); | |
2859 | ftp = skip_non_digits(ftp); | |
2860 | if (!isdigit(*ftp)) { | |
2861 | return CT_FTP_CTL_INVALID; | |
2862 | } | |
2863 | ||
2864 | /* Not used for passive mode. */ | |
2865 | *addr_offset_from_ftp_data_start = 0; | |
2866 | *addr_size = 0; | |
2867 | ||
2868 | *mode = CT_FTP_MODE_PASSIVE; | |
2869 | } | |
2870 | ||
2871 | char *save_ftp = ftp; | |
2872 | ftp = terminate_number_str(ftp, MAX_EXT_FTP_PORT_DGTS); | |
2873 | if (!ftp) { | |
2874 | return CT_FTP_CTL_INVALID; | |
2875 | } | |
dec0dbbc | 2876 | |
bd5e81a0 DB |
2877 | int value; |
2878 | if (!str_to_int(save_ftp, 10, &value)) { | |
2879 | return CT_FTP_CTL_INVALID; | |
2880 | } | |
2881 | if (value > CT_MAX_L4_PORT) { | |
2882 | return CT_FTP_CTL_INVALID; | |
2883 | } | |
2884 | ||
2885 | uint16_t port_hs = value; | |
2886 | ovs_be16 port = htons(port_hs); | |
2887 | ||
2888 | switch (*mode) { | |
2889 | case CT_FTP_MODE_ACTIVE: | |
2890 | *v6_addr_rep = conn_for_expectation->rev_key.dst.addr; | |
2891 | /* Although most servers will block this exploit, there may be some | |
2892 | * less well managed. */ | |
cda1b109 DB |
2893 | if (memcmp(&ip6_addr, &v6_addr_rep->ipv6, sizeof ip6_addr) && |
2894 | memcmp(&ip6_addr, &conn_for_expectation->key.src.addr.ipv6, | |
bd5e81a0 DB |
2895 | sizeof ip6_addr)) { |
2896 | return CT_FTP_CTL_INVALID; | |
2897 | } | |
2898 | break; | |
2899 | case CT_FTP_MODE_PASSIVE: | |
2900 | *v6_addr_rep = conn_for_expectation->key.dst.addr; | |
2901 | break; | |
7be77cb0 | 2902 | case CT_TFTP_MODE: |
bd5e81a0 DB |
2903 | default: |
2904 | OVS_NOT_REACHED(); | |
2905 | } | |
2906 | ||
be38342d DB |
2907 | expectation_create(ct, port, conn_for_expectation, |
2908 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
bd5e81a0 DB |
2909 | return CT_FTP_CTL_INTEREST; |
2910 | } | |
2911 | ||
2912 | static int | |
cda1b109 | 2913 | repl_ftp_v6_addr(struct dp_packet *pkt, union ct_addr v6_addr_rep, |
bd5e81a0 DB |
2914 | char *ftp_data_start, |
2915 | size_t addr_offset_from_ftp_data_start, | |
2916 | size_t addr_size, enum ct_alg_mode mode) | |
2917 | { | |
2918 | /* This is slightly bigger than really possible. */ | |
2919 | enum { MAX_FTP_V6_NAT_DELTA = 45 }; | |
2920 | ||
2921 | if (mode == CT_FTP_MODE_PASSIVE) { | |
2922 | return 0; | |
2923 | } | |
2924 | ||
2925 | /* Do conservative check for pathological MTU usage. */ | |
2926 | uint32_t orig_used_size = dp_packet_size(pkt); | |
cd7c99a6 DB |
2927 | if (orig_used_size + MAX_FTP_V6_NAT_DELTA > |
2928 | dp_packet_get_allocated(pkt)) { | |
2929 | ||
bd5e81a0 | 2930 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
cd7c99a6 DB |
2931 | VLOG_WARN_RL(&rl, "Unsupported effective MTU %u used with FTP V6", |
2932 | dp_packet_get_allocated(pkt)); | |
bd5e81a0 DB |
2933 | return 0; |
2934 | } | |
2935 | ||
298530b8 | 2936 | char v6_addr_str[INET6_ADDRSTRLEN] = {0}; |
cda1b109 | 2937 | ovs_assert(inet_ntop(AF_INET6, &v6_addr_rep.ipv6, v6_addr_str, |
298530b8 | 2938 | sizeof v6_addr_str)); |
cd7c99a6 DB |
2939 | modify_packet(pkt, ftp_data_start + addr_offset_from_ftp_data_start, |
2940 | addr_size, v6_addr_str, strlen(v6_addr_str), | |
2941 | orig_used_size); | |
2942 | return (int) strlen(v6_addr_str) - (int) addr_size; | |
bd5e81a0 DB |
2943 | } |
2944 | ||
d13d7115 DB |
2945 | /* Increment/decrement a TCP sequence number. */ |
2946 | static void | |
2947 | adj_seqnum(ovs_16aligned_be32 *val, int32_t inc) | |
2948 | { | |
2949 | put_16aligned_be32(val, htonl(ntohl(get_16aligned_be32(val)) + inc)); | |
2950 | } | |
2951 | ||
bd5e81a0 DB |
2952 | static void |
2953 | handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx, | |
967bb5c5 | 2954 | struct dp_packet *pkt, struct conn *ec, long long now, |
253e4dc0 | 2955 | enum ftp_ctl_pkt ftp_ctl, bool nat) |
bd5e81a0 DB |
2956 | { |
2957 | struct ip_header *l3_hdr = dp_packet_l3(pkt); | |
2958 | ovs_be32 v4_addr_rep = 0; | |
cda1b109 | 2959 | union ct_addr v6_addr_rep; |
faa0826d | 2960 | size_t addr_offset_from_ftp_data_start = 0; |
bd5e81a0 DB |
2961 | size_t addr_size = 0; |
2962 | char *ftp_data_start; | |
bd5e81a0 DB |
2963 | enum ct_alg_mode mode = CT_FTP_MODE_ACTIVE; |
2964 | ||
2965 | if (detect_ftp_ctl_type(ctx, pkt) != ftp_ctl) { | |
2966 | return; | |
2967 | } | |
2968 | ||
bd5e81a0 DB |
2969 | struct ovs_16aligned_ip6_hdr *nh6 = dp_packet_l3(pkt); |
2970 | int64_t seq_skew = 0; | |
dec0dbbc | 2971 | |
253e4dc0 | 2972 | if (ftp_ctl == CT_FTP_CTL_INTEREST) { |
bd5e81a0 DB |
2973 | enum ftp_ctl_pkt rc; |
2974 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { | |
253e4dc0 | 2975 | rc = process_ftp_ctl_v6(ct, pkt, ec, |
4417ca3d | 2976 | &v6_addr_rep, &ftp_data_start, |
bd5e81a0 DB |
2977 | &addr_offset_from_ftp_data_start, |
2978 | &addr_size, &mode); | |
2979 | } else { | |
253e4dc0 | 2980 | rc = process_ftp_ctl_v4(ct, pkt, ec, |
4417ca3d | 2981 | &v4_addr_rep, &ftp_data_start, |
cd7c99a6 DB |
2982 | &addr_offset_from_ftp_data_start, |
2983 | &addr_size); | |
bd5e81a0 DB |
2984 | } |
2985 | if (rc == CT_FTP_CTL_INVALID) { | |
2986 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
2987 | VLOG_WARN_RL(&rl, "Invalid FTP control packet format"); | |
2988 | pkt->md.ct_state |= CS_TRACKED | CS_INVALID; | |
2989 | return; | |
2990 | } else if (rc == CT_FTP_CTL_INTEREST) { | |
2991 | uint16_t ip_len; | |
dec0dbbc | 2992 | |
bd5e81a0 | 2993 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
253e4dc0 DM |
2994 | if (nat) { |
2995 | seq_skew = repl_ftp_v6_addr(pkt, v6_addr_rep, | |
2996 | ftp_data_start, | |
2997 | addr_offset_from_ftp_data_start, | |
2998 | addr_size, mode); | |
2999 | } | |
3000 | ||
bd5e81a0 | 3001 | if (seq_skew) { |
253e4dc0 DM |
3002 | ip_len = ntohs(nh6->ip6_ctlun.ip6_un1.ip6_un1_plen) + |
3003 | seq_skew; | |
bd5e81a0 | 3004 | nh6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(ip_len); |
bd5e81a0 DB |
3005 | } |
3006 | } else { | |
253e4dc0 DM |
3007 | if (nat) { |
3008 | seq_skew = repl_ftp_v4_addr(pkt, v4_addr_rep, | |
3009 | ftp_data_start, | |
cd7c99a6 DB |
3010 | addr_offset_from_ftp_data_start, |
3011 | addr_size); | |
253e4dc0 | 3012 | } |
bd5e81a0 | 3013 | if (seq_skew) { |
253e4dc0 | 3014 | ip_len = ntohs(l3_hdr->ip_tot_len) + seq_skew; |
bd5e81a0 DB |
3015 | l3_hdr->ip_csum = recalc_csum16(l3_hdr->ip_csum, |
3016 | l3_hdr->ip_tot_len, htons(ip_len)); | |
3017 | l3_hdr->ip_tot_len = htons(ip_len); | |
bd5e81a0 DB |
3018 | } |
3019 | } | |
3020 | } else { | |
3021 | OVS_NOT_REACHED(); | |
3022 | } | |
bd5e81a0 DB |
3023 | } |
3024 | ||
3025 | struct tcp_header *th = dp_packet_l4(pkt); | |
dec0dbbc | 3026 | |
253e4dc0 | 3027 | if (nat && ec->seq_skew != 0) { |
d13d7115 DB |
3028 | ctx->reply != ec->seq_skew_dir ? |
3029 | adj_seqnum(&th->tcp_ack, -ec->seq_skew) : | |
3030 | adj_seqnum(&th->tcp_seq, ec->seq_skew); | |
bd5e81a0 DB |
3031 | } |
3032 | ||
bd5e81a0 | 3033 | th->tcp_csum = 0; |
bd5e81a0 | 3034 | if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) { |
76d85771 DB |
3035 | th->tcp_csum = packet_csum_upperlayer6(nh6, th, ctx->key.nw_proto, |
3036 | dp_packet_l4_size(pkt)); | |
bd5e81a0 | 3037 | } else { |
76d85771 DB |
3038 | uint32_t tcp_csum = packet_csum_pseudoheader(l3_hdr); |
3039 | th->tcp_csum = csum_finish( | |
3040 | csum_continue(tcp_csum, th, dp_packet_l4_size(pkt))); | |
bd5e81a0 | 3041 | } |
253e4dc0 DM |
3042 | |
3043 | if (seq_skew) { | |
967bb5c5 | 3044 | conn_seq_skew_set(ct, ec, now, seq_skew + ec->seq_skew, |
253e4dc0 DM |
3045 | ctx->reply); |
3046 | } | |
bd5e81a0 | 3047 | } |
7be77cb0 DB |
3048 | |
3049 | static void | |
3050 | handle_tftp_ctl(struct conntrack *ct, | |
94e71143 | 3051 | const struct conn_lookup_ctx *ctx OVS_UNUSED, |
967bb5c5 DB |
3052 | struct dp_packet *pkt, struct conn *conn_for_expectation, |
3053 | long long now OVS_UNUSED, enum ftp_ctl_pkt ftp_ctl OVS_UNUSED, | |
3054 | bool nat OVS_UNUSED) | |
7be77cb0 | 3055 | { |
be38342d DB |
3056 | expectation_create(ct, conn_for_expectation->key.src.port, |
3057 | conn_for_expectation, | |
3058 | !!(pkt->md.ct_state & CS_REPLY_DIR), false, false); | |
7be77cb0 | 3059 | } |