1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2015 Nicira, Inc. 4 */ 5 6 #include <linux/module.h> 7 #include <linux/openvswitch.h> 8 #include <linux/tcp.h> 9 #include <linux/udp.h> 10 #include <linux/sctp.h> 11 #include <linux/static_key.h> 12 #include <net/ip.h> 13 #include <net/genetlink.h> 14 #include <net/netfilter/nf_conntrack_core.h> 15 #include <net/netfilter/nf_conntrack_count.h> 16 #include <net/netfilter/nf_conntrack_helper.h> 17 #include <net/netfilter/nf_conntrack_labels.h> 18 #include <net/netfilter/nf_conntrack_seqadj.h> 19 #include <net/netfilter/nf_conntrack_timeout.h> 20 #include <net/netfilter/nf_conntrack_zones.h> 21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h> 22 #include <net/ipv6_frag.h> 23 24 #if IS_ENABLED(CONFIG_NF_NAT) 25 #include <net/netfilter/nf_nat.h> 26 #endif 27 28 #include <net/netfilter/nf_conntrack_act_ct.h> 29 30 #include "datapath.h" 31 #include "conntrack.h" 32 #include "flow.h" 33 #include "flow_netlink.h" 34 35 struct ovs_ct_len_tbl { 36 int maxlen; 37 int minlen; 38 }; 39 40 /* Metadata mark for masked write to conntrack mark */ 41 struct md_mark { 42 u32 value; 43 u32 mask; 44 }; 45 46 /* Metadata label for masked write to conntrack label. */ 47 struct md_labels { 48 struct ovs_key_ct_labels value; 49 struct ovs_key_ct_labels mask; 50 }; 51 52 enum ovs_ct_nat { 53 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */ 54 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */ 55 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */ 56 }; 57 58 /* Conntrack action context for execution. */ 59 struct ovs_conntrack_info { 60 struct nf_conntrack_helper *helper; 61 struct nf_conntrack_zone zone; 62 struct nf_conn *ct; 63 u8 commit : 1; 64 u8 nat : 3; /* enum ovs_ct_nat */ 65 u8 force : 1; 66 u8 have_eventmask : 1; 67 u16 family; 68 u32 eventmask; /* Mask of 1 << IPCT_*. */ 69 struct md_mark mark; 70 struct md_labels labels; 71 char timeout[CTNL_TIMEOUT_NAME_MAX]; 72 struct nf_ct_timeout *nf_ct_timeout; 73 #if IS_ENABLED(CONFIG_NF_NAT) 74 struct nf_nat_range2 range; /* Only present for SRC NAT and DST NAT. */ 75 #endif 76 }; 77 78 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 79 #define OVS_CT_LIMIT_UNLIMITED 0 80 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED 81 #define CT_LIMIT_HASH_BUCKETS 512 82 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled); 83 84 struct ovs_ct_limit { 85 /* Elements in ovs_ct_limit_info->limits hash table */ 86 struct hlist_node hlist_node; 87 struct rcu_head rcu; 88 u16 zone; 89 u32 limit; 90 }; 91 92 struct ovs_ct_limit_info { 93 u32 default_limit; 94 struct hlist_head *limits; 95 struct nf_conncount_data *data; 96 }; 97 98 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = { 99 [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, }, 100 }; 101 #endif 102 103 static bool labels_nonzero(const struct ovs_key_ct_labels *labels); 104 105 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info); 106 107 static u16 key_to_nfproto(const struct sw_flow_key *key) 108 { 109 switch (ntohs(key->eth.type)) { 110 case ETH_P_IP: 111 return NFPROTO_IPV4; 112 case ETH_P_IPV6: 113 return NFPROTO_IPV6; 114 default: 115 return NFPROTO_UNSPEC; 116 } 117 } 118 119 /* Map SKB connection state into the values used by flow definition. */ 120 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo) 121 { 122 u8 ct_state = OVS_CS_F_TRACKED; 123 124 switch (ctinfo) { 125 case IP_CT_ESTABLISHED_REPLY: 126 case IP_CT_RELATED_REPLY: 127 ct_state |= OVS_CS_F_REPLY_DIR; 128 break; 129 default: 130 break; 131 } 132 133 switch (ctinfo) { 134 case IP_CT_ESTABLISHED: 135 case IP_CT_ESTABLISHED_REPLY: 136 ct_state |= OVS_CS_F_ESTABLISHED; 137 break; 138 case IP_CT_RELATED: 139 case IP_CT_RELATED_REPLY: 140 ct_state |= OVS_CS_F_RELATED; 141 break; 142 case IP_CT_NEW: 143 ct_state |= OVS_CS_F_NEW; 144 break; 145 default: 146 break; 147 } 148 149 return ct_state; 150 } 151 152 static u32 ovs_ct_get_mark(const struct nf_conn *ct) 153 { 154 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 155 return ct ? ct->mark : 0; 156 #else 157 return 0; 158 #endif 159 } 160 161 /* Guard against conntrack labels max size shrinking below 128 bits. */ 162 #if NF_CT_LABELS_MAX_SIZE < 16 163 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes 164 #endif 165 166 static void ovs_ct_get_labels(const struct nf_conn *ct, 167 struct ovs_key_ct_labels *labels) 168 { 169 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL; 170 171 if (cl) 172 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN); 173 else 174 memset(labels, 0, OVS_CT_LABELS_LEN); 175 } 176 177 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key, 178 const struct nf_conntrack_tuple *orig, 179 u8 icmp_proto) 180 { 181 key->ct_orig_proto = orig->dst.protonum; 182 if (orig->dst.protonum == icmp_proto) { 183 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type); 184 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code); 185 } else { 186 key->ct.orig_tp.src = orig->src.u.all; 187 key->ct.orig_tp.dst = orig->dst.u.all; 188 } 189 } 190 191 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state, 192 const struct nf_conntrack_zone *zone, 193 const struct nf_conn *ct) 194 { 195 key->ct_state = state; 196 key->ct_zone = zone->id; 197 key->ct.mark = ovs_ct_get_mark(ct); 198 ovs_ct_get_labels(ct, &key->ct.labels); 199 200 if (ct) { 201 const struct nf_conntrack_tuple *orig; 202 203 /* Use the master if we have one. */ 204 if (ct->master) 205 ct = ct->master; 206 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; 207 208 /* IP version must match with the master connection. */ 209 if (key->eth.type == htons(ETH_P_IP) && 210 nf_ct_l3num(ct) == NFPROTO_IPV4) { 211 key->ipv4.ct_orig.src = orig->src.u3.ip; 212 key->ipv4.ct_orig.dst = orig->dst.u3.ip; 213 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP); 214 return; 215 } else if (key->eth.type == htons(ETH_P_IPV6) && 216 !sw_flow_key_is_nd(key) && 217 nf_ct_l3num(ct) == NFPROTO_IPV6) { 218 key->ipv6.ct_orig.src = orig->src.u3.in6; 219 key->ipv6.ct_orig.dst = orig->dst.u3.in6; 220 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP); 221 return; 222 } 223 } 224 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack 225 * original direction key fields. 226 */ 227 key->ct_orig_proto = 0; 228 } 229 230 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has 231 * previously sent the packet to conntrack via the ct action. If 232 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are 233 * initialized from the connection status. 234 */ 235 static void ovs_ct_update_key(const struct sk_buff *skb, 236 const struct ovs_conntrack_info *info, 237 struct sw_flow_key *key, bool post_ct, 238 bool keep_nat_flags) 239 { 240 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt; 241 enum ip_conntrack_info ctinfo; 242 struct nf_conn *ct; 243 u8 state = 0; 244 245 ct = nf_ct_get(skb, &ctinfo); 246 if (ct) { 247 state = ovs_ct_get_state(ctinfo); 248 /* All unconfirmed entries are NEW connections. */ 249 if (!nf_ct_is_confirmed(ct)) 250 state |= OVS_CS_F_NEW; 251 /* OVS persists the related flag for the duration of the 252 * connection. 253 */ 254 if (ct->master) 255 state |= OVS_CS_F_RELATED; 256 if (keep_nat_flags) { 257 state |= key->ct_state & OVS_CS_F_NAT_MASK; 258 } else { 259 if (ct->status & IPS_SRC_NAT) 260 state |= OVS_CS_F_SRC_NAT; 261 if (ct->status & IPS_DST_NAT) 262 state |= OVS_CS_F_DST_NAT; 263 } 264 zone = nf_ct_zone(ct); 265 } else if (post_ct) { 266 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID; 267 if (info) 268 zone = &info->zone; 269 } 270 __ovs_ct_update_key(key, state, zone, ct); 271 } 272 273 /* This is called to initialize CT key fields possibly coming in from the local 274 * stack. 275 */ 276 void ovs_ct_fill_key(const struct sk_buff *skb, 277 struct sw_flow_key *key, 278 bool post_ct) 279 { 280 ovs_ct_update_key(skb, NULL, key, post_ct, false); 281 } 282 283 int ovs_ct_put_key(const struct sw_flow_key *swkey, 284 const struct sw_flow_key *output, struct sk_buff *skb) 285 { 286 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state)) 287 return -EMSGSIZE; 288 289 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 290 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone)) 291 return -EMSGSIZE; 292 293 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 294 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark)) 295 return -EMSGSIZE; 296 297 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 298 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels), 299 &output->ct.labels)) 300 return -EMSGSIZE; 301 302 if (swkey->ct_orig_proto) { 303 if (swkey->eth.type == htons(ETH_P_IP)) { 304 struct ovs_key_ct_tuple_ipv4 orig; 305 306 memset(&orig, 0, sizeof(orig)); 307 orig.ipv4_src = output->ipv4.ct_orig.src; 308 orig.ipv4_dst = output->ipv4.ct_orig.dst; 309 orig.src_port = output->ct.orig_tp.src; 310 orig.dst_port = output->ct.orig_tp.dst; 311 orig.ipv4_proto = output->ct_orig_proto; 312 313 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4, 314 sizeof(orig), &orig)) 315 return -EMSGSIZE; 316 } else if (swkey->eth.type == htons(ETH_P_IPV6)) { 317 struct ovs_key_ct_tuple_ipv6 orig; 318 319 memset(&orig, 0, sizeof(orig)); 320 memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32, 321 sizeof(orig.ipv6_src)); 322 memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32, 323 sizeof(orig.ipv6_dst)); 324 orig.src_port = output->ct.orig_tp.src; 325 orig.dst_port = output->ct.orig_tp.dst; 326 orig.ipv6_proto = output->ct_orig_proto; 327 328 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6, 329 sizeof(orig), &orig)) 330 return -EMSGSIZE; 331 } 332 } 333 334 return 0; 335 } 336 337 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key, 338 u32 ct_mark, u32 mask) 339 { 340 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 341 u32 new_mark; 342 343 new_mark = ct_mark | (ct->mark & ~(mask)); 344 if (ct->mark != new_mark) { 345 ct->mark = new_mark; 346 if (nf_ct_is_confirmed(ct)) 347 nf_conntrack_event_cache(IPCT_MARK, ct); 348 key->ct.mark = new_mark; 349 } 350 351 return 0; 352 #else 353 return -ENOTSUPP; 354 #endif 355 } 356 357 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct) 358 { 359 struct nf_conn_labels *cl; 360 361 cl = nf_ct_labels_find(ct); 362 if (!cl) { 363 nf_ct_labels_ext_add(ct); 364 cl = nf_ct_labels_find(ct); 365 } 366 367 return cl; 368 } 369 370 /* Initialize labels for a new, yet to be committed conntrack entry. Note that 371 * since the new connection is not yet confirmed, and thus no-one else has 372 * access to it's labels, we simply write them over. 373 */ 374 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key, 375 const struct ovs_key_ct_labels *labels, 376 const struct ovs_key_ct_labels *mask) 377 { 378 struct nf_conn_labels *cl, *master_cl; 379 bool have_mask = labels_nonzero(mask); 380 381 /* Inherit master's labels to the related connection? */ 382 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL; 383 384 if (!master_cl && !have_mask) 385 return 0; /* Nothing to do. */ 386 387 cl = ovs_ct_get_conn_labels(ct); 388 if (!cl) 389 return -ENOSPC; 390 391 /* Inherit the master's labels, if any. */ 392 if (master_cl) 393 *cl = *master_cl; 394 395 if (have_mask) { 396 u32 *dst = (u32 *)cl->bits; 397 int i; 398 399 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++) 400 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) | 401 (labels->ct_labels_32[i] 402 & mask->ct_labels_32[i]); 403 } 404 405 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the 406 * IPCT_LABEL bit is set in the event cache. 407 */ 408 nf_conntrack_event_cache(IPCT_LABEL, ct); 409 410 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN); 411 412 return 0; 413 } 414 415 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key, 416 const struct ovs_key_ct_labels *labels, 417 const struct ovs_key_ct_labels *mask) 418 { 419 struct nf_conn_labels *cl; 420 int err; 421 422 cl = ovs_ct_get_conn_labels(ct); 423 if (!cl) 424 return -ENOSPC; 425 426 err = nf_connlabels_replace(ct, labels->ct_labels_32, 427 mask->ct_labels_32, 428 OVS_CT_LABELS_LEN_32); 429 if (err) 430 return err; 431 432 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN); 433 434 return 0; 435 } 436 437 /* 'skb' should already be pulled to nh_ofs. */ 438 static int ovs_ct_helper(struct sk_buff *skb, u16 proto) 439 { 440 const struct nf_conntrack_helper *helper; 441 const struct nf_conn_help *help; 442 enum ip_conntrack_info ctinfo; 443 unsigned int protoff; 444 struct nf_conn *ct; 445 int err; 446 447 ct = nf_ct_get(skb, &ctinfo); 448 if (!ct || ctinfo == IP_CT_RELATED_REPLY) 449 return NF_ACCEPT; 450 451 help = nfct_help(ct); 452 if (!help) 453 return NF_ACCEPT; 454 455 helper = rcu_dereference(help->helper); 456 if (!helper) 457 return NF_ACCEPT; 458 459 switch (proto) { 460 case NFPROTO_IPV4: 461 protoff = ip_hdrlen(skb); 462 break; 463 case NFPROTO_IPV6: { 464 u8 nexthdr = ipv6_hdr(skb)->nexthdr; 465 __be16 frag_off; 466 int ofs; 467 468 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, 469 &frag_off); 470 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) { 471 pr_debug("proto header not found\n"); 472 return NF_ACCEPT; 473 } 474 protoff = ofs; 475 break; 476 } 477 default: 478 WARN_ONCE(1, "helper invoked on non-IP family!"); 479 return NF_DROP; 480 } 481 482 err = helper->help(skb, protoff, ct, ctinfo); 483 if (err != NF_ACCEPT) 484 return err; 485 486 /* Adjust seqs after helper. This is needed due to some helpers (e.g., 487 * FTP with NAT) adusting the TCP payload size when mangling IP 488 * addresses and/or port numbers in the text-based control connection. 489 */ 490 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) && 491 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff)) 492 return NF_DROP; 493 return NF_ACCEPT; 494 } 495 496 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero 497 * value if 'skb' is freed. 498 */ 499 static int handle_fragments(struct net *net, struct sw_flow_key *key, 500 u16 zone, struct sk_buff *skb) 501 { 502 struct ovs_skb_cb ovs_cb = *OVS_CB(skb); 503 int err; 504 505 if (key->eth.type == htons(ETH_P_IP)) { 506 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; 507 508 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 509 err = ip_defrag(net, skb, user); 510 if (err) 511 return err; 512 513 ovs_cb.mru = IPCB(skb)->frag_max_size; 514 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) 515 } else if (key->eth.type == htons(ETH_P_IPV6)) { 516 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; 517 518 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); 519 err = nf_ct_frag6_gather(net, skb, user); 520 if (err) { 521 if (err != -EINPROGRESS) 522 kfree_skb(skb); 523 return err; 524 } 525 526 key->ip.proto = ipv6_hdr(skb)->nexthdr; 527 ovs_cb.mru = IP6CB(skb)->frag_max_size; 528 #endif 529 } else { 530 kfree_skb(skb); 531 return -EPFNOSUPPORT; 532 } 533 534 /* The key extracted from the fragment that completed this datagram 535 * likely didn't have an L4 header, so regenerate it. 536 */ 537 ovs_flow_key_update_l3l4(skb, key); 538 539 key->ip.frag = OVS_FRAG_TYPE_NONE; 540 skb_clear_hash(skb); 541 skb->ignore_df = 1; 542 *OVS_CB(skb) = ovs_cb; 543 544 return 0; 545 } 546 547 static struct nf_conntrack_expect * 548 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone, 549 u16 proto, const struct sk_buff *skb) 550 { 551 struct nf_conntrack_tuple tuple; 552 struct nf_conntrack_expect *exp; 553 554 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple)) 555 return NULL; 556 557 exp = __nf_ct_expect_find(net, zone, &tuple); 558 if (exp) { 559 struct nf_conntrack_tuple_hash *h; 560 561 /* Delete existing conntrack entry, if it clashes with the 562 * expectation. This can happen since conntrack ALGs do not 563 * check for clashes between (new) expectations and existing 564 * conntrack entries. nf_conntrack_in() will check the 565 * expectations only if a conntrack entry can not be found, 566 * which can lead to OVS finding the expectation (here) in the 567 * init direction, but which will not be removed by the 568 * nf_conntrack_in() call, if a matching conntrack entry is 569 * found instead. In this case all init direction packets 570 * would be reported as new related packets, while reply 571 * direction packets would be reported as un-related 572 * established packets. 573 */ 574 h = nf_conntrack_find_get(net, zone, &tuple); 575 if (h) { 576 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); 577 578 nf_ct_delete(ct, 0, 0); 579 nf_ct_put(ct); 580 } 581 } 582 583 return exp; 584 } 585 586 /* This replicates logic from nf_conntrack_core.c that is not exported. */ 587 static enum ip_conntrack_info 588 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h) 589 { 590 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); 591 592 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) 593 return IP_CT_ESTABLISHED_REPLY; 594 /* Once we've had two way comms, always ESTABLISHED. */ 595 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 596 return IP_CT_ESTABLISHED; 597 if (test_bit(IPS_EXPECTED_BIT, &ct->status)) 598 return IP_CT_RELATED; 599 return IP_CT_NEW; 600 } 601 602 /* Find an existing connection which this packet belongs to without 603 * re-attributing statistics or modifying the connection state. This allows an 604 * skb->_nfct lost due to an upcall to be recovered during actions execution. 605 * 606 * Must be called with rcu_read_lock. 607 * 608 * On success, populates skb->_nfct and returns the connection. Returns NULL 609 * if there is no existing entry. 610 */ 611 static struct nf_conn * 612 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone, 613 u8 l3num, struct sk_buff *skb, bool natted) 614 { 615 struct nf_conntrack_tuple tuple; 616 struct nf_conntrack_tuple_hash *h; 617 struct nf_conn *ct; 618 619 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num, 620 net, &tuple)) { 621 pr_debug("ovs_ct_find_existing: Can't get tuple\n"); 622 return NULL; 623 } 624 625 /* Must invert the tuple if skb has been transformed by NAT. */ 626 if (natted) { 627 struct nf_conntrack_tuple inverse; 628 629 if (!nf_ct_invert_tuple(&inverse, &tuple)) { 630 pr_debug("ovs_ct_find_existing: Inversion failed!\n"); 631 return NULL; 632 } 633 tuple = inverse; 634 } 635 636 /* look for tuple match */ 637 h = nf_conntrack_find_get(net, zone, &tuple); 638 if (!h) 639 return NULL; /* Not found. */ 640 641 ct = nf_ct_tuplehash_to_ctrack(h); 642 643 /* Inverted packet tuple matches the reverse direction conntrack tuple, 644 * select the other tuplehash to get the right 'ctinfo' bits for this 645 * packet. 646 */ 647 if (natted) 648 h = &ct->tuplehash[!h->tuple.dst.dir]; 649 650 nf_ct_set(skb, ct, ovs_ct_get_info(h)); 651 return ct; 652 } 653 654 static 655 struct nf_conn *ovs_ct_executed(struct net *net, 656 const struct sw_flow_key *key, 657 const struct ovs_conntrack_info *info, 658 struct sk_buff *skb, 659 bool *ct_executed) 660 { 661 struct nf_conn *ct = NULL; 662 663 /* If no ct, check if we have evidence that an existing conntrack entry 664 * might be found for this skb. This happens when we lose a skb->_nfct 665 * due to an upcall, or if the direction is being forced. If the 666 * connection was not confirmed, it is not cached and needs to be run 667 * through conntrack again. 668 */ 669 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) && 670 !(key->ct_state & OVS_CS_F_INVALID) && 671 (key->ct_zone == info->zone.id); 672 673 if (*ct_executed || (!key->ct_state && info->force)) { 674 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb, 675 !!(key->ct_state & 676 OVS_CS_F_NAT_MASK)); 677 } 678 679 return ct; 680 } 681 682 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 683 static bool skb_nfct_cached(struct net *net, 684 const struct sw_flow_key *key, 685 const struct ovs_conntrack_info *info, 686 struct sk_buff *skb) 687 { 688 enum ip_conntrack_info ctinfo; 689 struct nf_conn *ct; 690 bool ct_executed = true; 691 692 ct = nf_ct_get(skb, &ctinfo); 693 if (!ct) 694 ct = ovs_ct_executed(net, key, info, skb, &ct_executed); 695 696 if (ct) 697 nf_ct_get(skb, &ctinfo); 698 else 699 return false; 700 701 if (!net_eq(net, read_pnet(&ct->ct_net))) 702 return false; 703 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct))) 704 return false; 705 if (info->helper) { 706 struct nf_conn_help *help; 707 708 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); 709 if (help && rcu_access_pointer(help->helper) != info->helper) 710 return false; 711 } 712 if (info->nf_ct_timeout) { 713 struct nf_conn_timeout *timeout_ext; 714 715 timeout_ext = nf_ct_timeout_find(ct); 716 if (!timeout_ext || info->nf_ct_timeout != 717 rcu_dereference(timeout_ext->timeout)) 718 return false; 719 } 720 /* Force conntrack entry direction to the current packet? */ 721 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 722 /* Delete the conntrack entry if confirmed, else just release 723 * the reference. 724 */ 725 if (nf_ct_is_confirmed(ct)) 726 nf_ct_delete(ct, 0, 0); 727 728 nf_ct_put(ct); 729 nf_ct_set(skb, NULL, 0); 730 return false; 731 } 732 733 return ct_executed; 734 } 735 736 #if IS_ENABLED(CONFIG_NF_NAT) 737 static void ovs_nat_update_key(struct sw_flow_key *key, 738 const struct sk_buff *skb, 739 enum nf_nat_manip_type maniptype) 740 { 741 if (maniptype == NF_NAT_MANIP_SRC) { 742 __be16 src; 743 744 key->ct_state |= OVS_CS_F_SRC_NAT; 745 if (key->eth.type == htons(ETH_P_IP)) 746 key->ipv4.addr.src = ip_hdr(skb)->saddr; 747 else if (key->eth.type == htons(ETH_P_IPV6)) 748 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr, 749 sizeof(key->ipv6.addr.src)); 750 else 751 return; 752 753 if (key->ip.proto == IPPROTO_UDP) 754 src = udp_hdr(skb)->source; 755 else if (key->ip.proto == IPPROTO_TCP) 756 src = tcp_hdr(skb)->source; 757 else if (key->ip.proto == IPPROTO_SCTP) 758 src = sctp_hdr(skb)->source; 759 else 760 return; 761 762 key->tp.src = src; 763 } else { 764 __be16 dst; 765 766 key->ct_state |= OVS_CS_F_DST_NAT; 767 if (key->eth.type == htons(ETH_P_IP)) 768 key->ipv4.addr.dst = ip_hdr(skb)->daddr; 769 else if (key->eth.type == htons(ETH_P_IPV6)) 770 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr, 771 sizeof(key->ipv6.addr.dst)); 772 else 773 return; 774 775 if (key->ip.proto == IPPROTO_UDP) 776 dst = udp_hdr(skb)->dest; 777 else if (key->ip.proto == IPPROTO_TCP) 778 dst = tcp_hdr(skb)->dest; 779 else if (key->ip.proto == IPPROTO_SCTP) 780 dst = sctp_hdr(skb)->dest; 781 else 782 return; 783 784 key->tp.dst = dst; 785 } 786 } 787 788 /* Modelled after nf_nat_ipv[46]_fn(). 789 * range is only used for new, uninitialized NAT state. 790 * Returns either NF_ACCEPT or NF_DROP. 791 */ 792 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct, 793 enum ip_conntrack_info ctinfo, 794 const struct nf_nat_range2 *range, 795 enum nf_nat_manip_type maniptype, struct sw_flow_key *key) 796 { 797 int hooknum, nh_off, err = NF_ACCEPT; 798 799 nh_off = skb_network_offset(skb); 800 skb_pull_rcsum(skb, nh_off); 801 802 /* See HOOK2MANIP(). */ 803 if (maniptype == NF_NAT_MANIP_SRC) 804 hooknum = NF_INET_LOCAL_IN; /* Source NAT */ 805 else 806 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */ 807 808 switch (ctinfo) { 809 case IP_CT_RELATED: 810 case IP_CT_RELATED_REPLY: 811 if (IS_ENABLED(CONFIG_NF_NAT) && 812 skb->protocol == htons(ETH_P_IP) && 813 ip_hdr(skb)->protocol == IPPROTO_ICMP) { 814 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo, 815 hooknum)) 816 err = NF_DROP; 817 goto push; 818 } else if (IS_ENABLED(CONFIG_IPV6) && 819 skb->protocol == htons(ETH_P_IPV6)) { 820 __be16 frag_off; 821 u8 nexthdr = ipv6_hdr(skb)->nexthdr; 822 int hdrlen = ipv6_skip_exthdr(skb, 823 sizeof(struct ipv6hdr), 824 &nexthdr, &frag_off); 825 826 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) { 827 if (!nf_nat_icmpv6_reply_translation(skb, ct, 828 ctinfo, 829 hooknum, 830 hdrlen)) 831 err = NF_DROP; 832 goto push; 833 } 834 } 835 /* Non-ICMP, fall thru to initialize if needed. */ 836 fallthrough; 837 case IP_CT_NEW: 838 /* Seen it before? This can happen for loopback, retrans, 839 * or local packets. 840 */ 841 if (!nf_nat_initialized(ct, maniptype)) { 842 /* Initialize according to the NAT action. */ 843 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS) 844 /* Action is set up to establish a new 845 * mapping. 846 */ 847 ? nf_nat_setup_info(ct, range, maniptype) 848 : nf_nat_alloc_null_binding(ct, hooknum); 849 if (err != NF_ACCEPT) 850 goto push; 851 } 852 break; 853 854 case IP_CT_ESTABLISHED: 855 case IP_CT_ESTABLISHED_REPLY: 856 break; 857 858 default: 859 err = NF_DROP; 860 goto push; 861 } 862 863 err = nf_nat_packet(ct, ctinfo, hooknum, skb); 864 push: 865 skb_push_rcsum(skb, nh_off); 866 867 /* Update the flow key if NAT successful. */ 868 if (err == NF_ACCEPT) 869 ovs_nat_update_key(key, skb, maniptype); 870 871 return err; 872 } 873 874 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */ 875 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key, 876 const struct ovs_conntrack_info *info, 877 struct sk_buff *skb, struct nf_conn *ct, 878 enum ip_conntrack_info ctinfo) 879 { 880 enum nf_nat_manip_type maniptype; 881 int err; 882 883 /* Add NAT extension if not confirmed yet. */ 884 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct)) 885 return NF_ACCEPT; /* Can't NAT. */ 886 887 /* Determine NAT type. 888 * Check if the NAT type can be deduced from the tracked connection. 889 * Make sure new expected connections (IP_CT_RELATED) are NATted only 890 * when committing. 891 */ 892 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW && 893 ct->status & IPS_NAT_MASK && 894 (ctinfo != IP_CT_RELATED || info->commit)) { 895 /* NAT an established or related connection like before. */ 896 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) 897 /* This is the REPLY direction for a connection 898 * for which NAT was applied in the forward 899 * direction. Do the reverse NAT. 900 */ 901 maniptype = ct->status & IPS_SRC_NAT 902 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC; 903 else 904 maniptype = ct->status & IPS_SRC_NAT 905 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST; 906 } else if (info->nat & OVS_CT_SRC_NAT) { 907 maniptype = NF_NAT_MANIP_SRC; 908 } else if (info->nat & OVS_CT_DST_NAT) { 909 maniptype = NF_NAT_MANIP_DST; 910 } else { 911 return NF_ACCEPT; /* Connection is not NATed. */ 912 } 913 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype, key); 914 915 if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) { 916 if (ct->status & IPS_SRC_NAT) { 917 if (maniptype == NF_NAT_MANIP_SRC) 918 maniptype = NF_NAT_MANIP_DST; 919 else 920 maniptype = NF_NAT_MANIP_SRC; 921 922 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, 923 maniptype, key); 924 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) { 925 err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL, 926 NF_NAT_MANIP_SRC, key); 927 } 928 } 929 930 return err; 931 } 932 #else /* !CONFIG_NF_NAT */ 933 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key, 934 const struct ovs_conntrack_info *info, 935 struct sk_buff *skb, struct nf_conn *ct, 936 enum ip_conntrack_info ctinfo) 937 { 938 return NF_ACCEPT; 939 } 940 #endif 941 942 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if 943 * not done already. Update key with new CT state after passing the packet 944 * through conntrack. 945 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be 946 * set to NULL and 0 will be returned. 947 */ 948 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key, 949 const struct ovs_conntrack_info *info, 950 struct sk_buff *skb) 951 { 952 /* If we are recirculating packets to match on conntrack fields and 953 * committing with a separate conntrack action, then we don't need to 954 * actually run the packet through conntrack twice unless it's for a 955 * different zone. 956 */ 957 bool cached = skb_nfct_cached(net, key, info, skb); 958 enum ip_conntrack_info ctinfo; 959 struct nf_conn *ct; 960 961 if (!cached) { 962 struct nf_hook_state state = { 963 .hook = NF_INET_PRE_ROUTING, 964 .pf = info->family, 965 .net = net, 966 }; 967 struct nf_conn *tmpl = info->ct; 968 int err; 969 970 /* Associate skb with specified zone. */ 971 if (tmpl) { 972 ct = nf_ct_get(skb, &ctinfo); 973 nf_ct_put(ct); 974 nf_conntrack_get(&tmpl->ct_general); 975 nf_ct_set(skb, tmpl, IP_CT_NEW); 976 } 977 978 err = nf_conntrack_in(skb, &state); 979 if (err != NF_ACCEPT) 980 return -ENOENT; 981 982 /* Clear CT state NAT flags to mark that we have not yet done 983 * NAT after the nf_conntrack_in() call. We can actually clear 984 * the whole state, as it will be re-initialized below. 985 */ 986 key->ct_state = 0; 987 988 /* Update the key, but keep the NAT flags. */ 989 ovs_ct_update_key(skb, info, key, true, true); 990 } 991 992 ct = nf_ct_get(skb, &ctinfo); 993 if (ct) { 994 bool add_helper = false; 995 996 /* Packets starting a new connection must be NATted before the 997 * helper, so that the helper knows about the NAT. We enforce 998 * this by delaying both NAT and helper calls for unconfirmed 999 * connections until the committing CT action. For later 1000 * packets NAT and Helper may be called in either order. 1001 * 1002 * NAT will be done only if the CT action has NAT, and only 1003 * once per packet (per zone), as guarded by the NAT bits in 1004 * the key->ct_state. 1005 */ 1006 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) && 1007 (nf_ct_is_confirmed(ct) || info->commit) && 1008 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) { 1009 return -EINVAL; 1010 } 1011 1012 /* Userspace may decide to perform a ct lookup without a helper 1013 * specified followed by a (recirculate and) commit with one, 1014 * or attach a helper in a later commit. Therefore, for 1015 * connections which we will commit, we may need to attach 1016 * the helper here. 1017 */ 1018 if (info->commit && info->helper && !nfct_help(ct)) { 1019 int err = __nf_ct_try_assign_helper(ct, info->ct, 1020 GFP_ATOMIC); 1021 if (err) 1022 return err; 1023 add_helper = true; 1024 1025 /* helper installed, add seqadj if NAT is required */ 1026 if (info->nat && !nfct_seqadj(ct)) { 1027 if (!nfct_seqadj_ext_add(ct)) 1028 return -EINVAL; 1029 } 1030 } 1031 1032 /* Call the helper only if: 1033 * - nf_conntrack_in() was executed above ("!cached") or a 1034 * helper was just attached ("add_helper") for a confirmed 1035 * connection, or 1036 * - When committing an unconfirmed connection. 1037 */ 1038 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper : 1039 info->commit) && 1040 ovs_ct_helper(skb, info->family) != NF_ACCEPT) { 1041 return -EINVAL; 1042 } 1043 1044 if (nf_ct_protonum(ct) == IPPROTO_TCP && 1045 nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) { 1046 /* Be liberal for tcp packets so that out-of-window 1047 * packets are not marked invalid. 1048 */ 1049 nf_ct_set_tcp_be_liberal(ct); 1050 } 1051 1052 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 1053 } 1054 1055 return 0; 1056 } 1057 1058 /* Lookup connection and read fields into key. */ 1059 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key, 1060 const struct ovs_conntrack_info *info, 1061 struct sk_buff *skb) 1062 { 1063 struct nf_conntrack_expect *exp; 1064 1065 /* If we pass an expected packet through nf_conntrack_in() the 1066 * expectation is typically removed, but the packet could still be 1067 * lost in upcall processing. To prevent this from happening we 1068 * perform an explicit expectation lookup. Expected connections are 1069 * always new, and will be passed through conntrack only when they are 1070 * committed, as it is OK to remove the expectation at that time. 1071 */ 1072 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb); 1073 if (exp) { 1074 u8 state; 1075 1076 /* NOTE: New connections are NATted and Helped only when 1077 * committed, so we are not calling into NAT here. 1078 */ 1079 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED; 1080 __ovs_ct_update_key(key, state, &info->zone, exp->master); 1081 } else { 1082 struct nf_conn *ct; 1083 int err; 1084 1085 err = __ovs_ct_lookup(net, key, info, skb); 1086 if (err) 1087 return err; 1088 1089 ct = (struct nf_conn *)skb_nfct(skb); 1090 if (ct) 1091 nf_ct_deliver_cached_events(ct); 1092 } 1093 1094 return 0; 1095 } 1096 1097 static bool labels_nonzero(const struct ovs_key_ct_labels *labels) 1098 { 1099 size_t i; 1100 1101 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++) 1102 if (labels->ct_labels_32[i]) 1103 return true; 1104 1105 return false; 1106 } 1107 1108 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 1109 static struct hlist_head *ct_limit_hash_bucket( 1110 const struct ovs_ct_limit_info *info, u16 zone) 1111 { 1112 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)]; 1113 } 1114 1115 /* Call with ovs_mutex */ 1116 static void ct_limit_set(const struct ovs_ct_limit_info *info, 1117 struct ovs_ct_limit *new_ct_limit) 1118 { 1119 struct ovs_ct_limit *ct_limit; 1120 struct hlist_head *head; 1121 1122 head = ct_limit_hash_bucket(info, new_ct_limit->zone); 1123 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) { 1124 if (ct_limit->zone == new_ct_limit->zone) { 1125 hlist_replace_rcu(&ct_limit->hlist_node, 1126 &new_ct_limit->hlist_node); 1127 kfree_rcu(ct_limit, rcu); 1128 return; 1129 } 1130 } 1131 1132 hlist_add_head_rcu(&new_ct_limit->hlist_node, head); 1133 } 1134 1135 /* Call with ovs_mutex */ 1136 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone) 1137 { 1138 struct ovs_ct_limit *ct_limit; 1139 struct hlist_head *head; 1140 struct hlist_node *n; 1141 1142 head = ct_limit_hash_bucket(info, zone); 1143 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) { 1144 if (ct_limit->zone == zone) { 1145 hlist_del_rcu(&ct_limit->hlist_node); 1146 kfree_rcu(ct_limit, rcu); 1147 return; 1148 } 1149 } 1150 } 1151 1152 /* Call with RCU read lock */ 1153 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone) 1154 { 1155 struct ovs_ct_limit *ct_limit; 1156 struct hlist_head *head; 1157 1158 head = ct_limit_hash_bucket(info, zone); 1159 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) { 1160 if (ct_limit->zone == zone) 1161 return ct_limit->limit; 1162 } 1163 1164 return info->default_limit; 1165 } 1166 1167 static int ovs_ct_check_limit(struct net *net, 1168 const struct ovs_conntrack_info *info, 1169 const struct nf_conntrack_tuple *tuple) 1170 { 1171 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 1172 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; 1173 u32 per_zone_limit, connections; 1174 u32 conncount_key; 1175 1176 conncount_key = info->zone.id; 1177 1178 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id); 1179 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED) 1180 return 0; 1181 1182 connections = nf_conncount_count(net, ct_limit_info->data, 1183 &conncount_key, tuple, &info->zone); 1184 if (connections > per_zone_limit) 1185 return -ENOMEM; 1186 1187 return 0; 1188 } 1189 #endif 1190 1191 /* Lookup connection and confirm if unconfirmed. */ 1192 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key, 1193 const struct ovs_conntrack_info *info, 1194 struct sk_buff *skb) 1195 { 1196 enum ip_conntrack_info ctinfo; 1197 struct nf_conn *ct; 1198 int err; 1199 1200 err = __ovs_ct_lookup(net, key, info, skb); 1201 if (err) 1202 return err; 1203 1204 /* The connection could be invalid, in which case this is a no-op.*/ 1205 ct = nf_ct_get(skb, &ctinfo); 1206 if (!ct) 1207 return 0; 1208 1209 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 1210 if (static_branch_unlikely(&ovs_ct_limit_enabled)) { 1211 if (!nf_ct_is_confirmed(ct)) { 1212 err = ovs_ct_check_limit(net, info, 1213 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 1214 if (err) { 1215 net_warn_ratelimited("openvswitch: zone: %u " 1216 "exceeds conntrack limit\n", 1217 info->zone.id); 1218 return err; 1219 } 1220 } 1221 } 1222 #endif 1223 1224 /* Set the conntrack event mask if given. NEW and DELETE events have 1225 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener 1226 * typically would receive many kinds of updates. Setting the event 1227 * mask allows those events to be filtered. The set event mask will 1228 * remain in effect for the lifetime of the connection unless changed 1229 * by a further CT action with both the commit flag and the eventmask 1230 * option. */ 1231 if (info->have_eventmask) { 1232 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct); 1233 1234 if (cache) 1235 cache->ctmask = info->eventmask; 1236 } 1237 1238 /* Apply changes before confirming the connection so that the initial 1239 * conntrack NEW netlink event carries the values given in the CT 1240 * action. 1241 */ 1242 if (info->mark.mask) { 1243 err = ovs_ct_set_mark(ct, key, info->mark.value, 1244 info->mark.mask); 1245 if (err) 1246 return err; 1247 } 1248 if (!nf_ct_is_confirmed(ct)) { 1249 err = ovs_ct_init_labels(ct, key, &info->labels.value, 1250 &info->labels.mask); 1251 if (err) 1252 return err; 1253 1254 nf_conn_act_ct_ext_add(ct); 1255 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1256 labels_nonzero(&info->labels.mask)) { 1257 err = ovs_ct_set_labels(ct, key, &info->labels.value, 1258 &info->labels.mask); 1259 if (err) 1260 return err; 1261 } 1262 /* This will take care of sending queued events even if the connection 1263 * is already confirmed. 1264 */ 1265 if (nf_conntrack_confirm(skb) != NF_ACCEPT) 1266 return -EINVAL; 1267 1268 return 0; 1269 } 1270 1271 /* Trim the skb to the length specified by the IP/IPv6 header, 1272 * removing any trailing lower-layer padding. This prepares the skb 1273 * for higher-layer processing that assumes skb->len excludes padding 1274 * (such as nf_ip_checksum). The caller needs to pull the skb to the 1275 * network header, and ensure ip_hdr/ipv6_hdr points to valid data. 1276 */ 1277 static int ovs_skb_network_trim(struct sk_buff *skb) 1278 { 1279 unsigned int len; 1280 int err; 1281 1282 switch (skb->protocol) { 1283 case htons(ETH_P_IP): 1284 len = ntohs(ip_hdr(skb)->tot_len); 1285 break; 1286 case htons(ETH_P_IPV6): 1287 len = sizeof(struct ipv6hdr) 1288 + ntohs(ipv6_hdr(skb)->payload_len); 1289 break; 1290 default: 1291 len = skb->len; 1292 } 1293 1294 err = pskb_trim_rcsum(skb, len); 1295 if (err) 1296 kfree_skb(skb); 1297 1298 return err; 1299 } 1300 1301 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero 1302 * value if 'skb' is freed. 1303 */ 1304 int ovs_ct_execute(struct net *net, struct sk_buff *skb, 1305 struct sw_flow_key *key, 1306 const struct ovs_conntrack_info *info) 1307 { 1308 int nh_ofs; 1309 int err; 1310 1311 /* The conntrack module expects to be working at L3. */ 1312 nh_ofs = skb_network_offset(skb); 1313 skb_pull_rcsum(skb, nh_ofs); 1314 1315 err = ovs_skb_network_trim(skb); 1316 if (err) 1317 return err; 1318 1319 if (key->ip.frag != OVS_FRAG_TYPE_NONE) { 1320 err = handle_fragments(net, key, info->zone.id, skb); 1321 if (err) 1322 return err; 1323 } 1324 1325 if (info->commit) 1326 err = ovs_ct_commit(net, key, info, skb); 1327 else 1328 err = ovs_ct_lookup(net, key, info, skb); 1329 1330 skb_push_rcsum(skb, nh_ofs); 1331 if (err) 1332 kfree_skb(skb); 1333 return err; 1334 } 1335 1336 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key) 1337 { 1338 enum ip_conntrack_info ctinfo; 1339 struct nf_conn *ct; 1340 1341 ct = nf_ct_get(skb, &ctinfo); 1342 1343 nf_ct_put(ct); 1344 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 1345 1346 if (key) 1347 ovs_ct_fill_key(skb, key, false); 1348 1349 return 0; 1350 } 1351 1352 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name, 1353 const struct sw_flow_key *key, bool log) 1354 { 1355 struct nf_conntrack_helper *helper; 1356 struct nf_conn_help *help; 1357 int ret = 0; 1358 1359 helper = nf_conntrack_helper_try_module_get(name, info->family, 1360 key->ip.proto); 1361 if (!helper) { 1362 OVS_NLERR(log, "Unknown helper \"%s\"", name); 1363 return -EINVAL; 1364 } 1365 1366 help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL); 1367 if (!help) { 1368 nf_conntrack_helper_put(helper); 1369 return -ENOMEM; 1370 } 1371 1372 #if IS_ENABLED(CONFIG_NF_NAT) 1373 if (info->nat) { 1374 ret = nf_nat_helper_try_module_get(name, info->family, 1375 key->ip.proto); 1376 if (ret) { 1377 nf_conntrack_helper_put(helper); 1378 OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d", 1379 name, ret); 1380 return ret; 1381 } 1382 } 1383 #endif 1384 rcu_assign_pointer(help->helper, helper); 1385 info->helper = helper; 1386 return ret; 1387 } 1388 1389 #if IS_ENABLED(CONFIG_NF_NAT) 1390 static int parse_nat(const struct nlattr *attr, 1391 struct ovs_conntrack_info *info, bool log) 1392 { 1393 struct nlattr *a; 1394 int rem; 1395 bool have_ip_max = false; 1396 bool have_proto_max = false; 1397 bool ip_vers = (info->family == NFPROTO_IPV6); 1398 1399 nla_for_each_nested(a, attr, rem) { 1400 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = { 1401 [OVS_NAT_ATTR_SRC] = {0, 0}, 1402 [OVS_NAT_ATTR_DST] = {0, 0}, 1403 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr), 1404 sizeof(struct in6_addr)}, 1405 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr), 1406 sizeof(struct in6_addr)}, 1407 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)}, 1408 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)}, 1409 [OVS_NAT_ATTR_PERSISTENT] = {0, 0}, 1410 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0}, 1411 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0}, 1412 }; 1413 int type = nla_type(a); 1414 1415 if (type > OVS_NAT_ATTR_MAX) { 1416 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)", 1417 type, OVS_NAT_ATTR_MAX); 1418 return -EINVAL; 1419 } 1420 1421 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) { 1422 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)", 1423 type, nla_len(a), 1424 ovs_nat_attr_lens[type][ip_vers]); 1425 return -EINVAL; 1426 } 1427 1428 switch (type) { 1429 case OVS_NAT_ATTR_SRC: 1430 case OVS_NAT_ATTR_DST: 1431 if (info->nat) { 1432 OVS_NLERR(log, "Only one type of NAT may be specified"); 1433 return -ERANGE; 1434 } 1435 info->nat |= OVS_CT_NAT; 1436 info->nat |= ((type == OVS_NAT_ATTR_SRC) 1437 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT); 1438 break; 1439 1440 case OVS_NAT_ATTR_IP_MIN: 1441 nla_memcpy(&info->range.min_addr, a, 1442 sizeof(info->range.min_addr)); 1443 info->range.flags |= NF_NAT_RANGE_MAP_IPS; 1444 break; 1445 1446 case OVS_NAT_ATTR_IP_MAX: 1447 have_ip_max = true; 1448 nla_memcpy(&info->range.max_addr, a, 1449 sizeof(info->range.max_addr)); 1450 info->range.flags |= NF_NAT_RANGE_MAP_IPS; 1451 break; 1452 1453 case OVS_NAT_ATTR_PROTO_MIN: 1454 info->range.min_proto.all = htons(nla_get_u16(a)); 1455 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1456 break; 1457 1458 case OVS_NAT_ATTR_PROTO_MAX: 1459 have_proto_max = true; 1460 info->range.max_proto.all = htons(nla_get_u16(a)); 1461 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1462 break; 1463 1464 case OVS_NAT_ATTR_PERSISTENT: 1465 info->range.flags |= NF_NAT_RANGE_PERSISTENT; 1466 break; 1467 1468 case OVS_NAT_ATTR_PROTO_HASH: 1469 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM; 1470 break; 1471 1472 case OVS_NAT_ATTR_PROTO_RANDOM: 1473 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY; 1474 break; 1475 1476 default: 1477 OVS_NLERR(log, "Unknown nat attribute (%d)", type); 1478 return -EINVAL; 1479 } 1480 } 1481 1482 if (rem > 0) { 1483 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem); 1484 return -EINVAL; 1485 } 1486 if (!info->nat) { 1487 /* Do not allow flags if no type is given. */ 1488 if (info->range.flags) { 1489 OVS_NLERR(log, 1490 "NAT flags may be given only when NAT range (SRC or DST) is also specified." 1491 ); 1492 return -EINVAL; 1493 } 1494 info->nat = OVS_CT_NAT; /* NAT existing connections. */ 1495 } else if (!info->commit) { 1496 OVS_NLERR(log, 1497 "NAT attributes may be specified only when CT COMMIT flag is also specified." 1498 ); 1499 return -EINVAL; 1500 } 1501 /* Allow missing IP_MAX. */ 1502 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) { 1503 memcpy(&info->range.max_addr, &info->range.min_addr, 1504 sizeof(info->range.max_addr)); 1505 } 1506 /* Allow missing PROTO_MAX. */ 1507 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED && 1508 !have_proto_max) { 1509 info->range.max_proto.all = info->range.min_proto.all; 1510 } 1511 return 0; 1512 } 1513 #endif 1514 1515 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = { 1516 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 }, 1517 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 }, 1518 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16), 1519 .maxlen = sizeof(u16) }, 1520 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark), 1521 .maxlen = sizeof(struct md_mark) }, 1522 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels), 1523 .maxlen = sizeof(struct md_labels) }, 1524 [OVS_CT_ATTR_HELPER] = { .minlen = 1, 1525 .maxlen = NF_CT_HELPER_NAME_LEN }, 1526 #if IS_ENABLED(CONFIG_NF_NAT) 1527 /* NAT length is checked when parsing the nested attributes. */ 1528 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX }, 1529 #endif 1530 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32), 1531 .maxlen = sizeof(u32) }, 1532 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1, 1533 .maxlen = CTNL_TIMEOUT_NAME_MAX }, 1534 }; 1535 1536 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info, 1537 const char **helper, bool log) 1538 { 1539 struct nlattr *a; 1540 int rem; 1541 1542 nla_for_each_nested(a, attr, rem) { 1543 int type = nla_type(a); 1544 int maxlen; 1545 int minlen; 1546 1547 if (type > OVS_CT_ATTR_MAX) { 1548 OVS_NLERR(log, 1549 "Unknown conntrack attr (type=%d, max=%d)", 1550 type, OVS_CT_ATTR_MAX); 1551 return -EINVAL; 1552 } 1553 1554 maxlen = ovs_ct_attr_lens[type].maxlen; 1555 minlen = ovs_ct_attr_lens[type].minlen; 1556 if (nla_len(a) < minlen || nla_len(a) > maxlen) { 1557 OVS_NLERR(log, 1558 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)", 1559 type, nla_len(a), maxlen); 1560 return -EINVAL; 1561 } 1562 1563 switch (type) { 1564 case OVS_CT_ATTR_FORCE_COMMIT: 1565 info->force = true; 1566 fallthrough; 1567 case OVS_CT_ATTR_COMMIT: 1568 info->commit = true; 1569 break; 1570 #ifdef CONFIG_NF_CONNTRACK_ZONES 1571 case OVS_CT_ATTR_ZONE: 1572 info->zone.id = nla_get_u16(a); 1573 break; 1574 #endif 1575 #ifdef CONFIG_NF_CONNTRACK_MARK 1576 case OVS_CT_ATTR_MARK: { 1577 struct md_mark *mark = nla_data(a); 1578 1579 if (!mark->mask) { 1580 OVS_NLERR(log, "ct_mark mask cannot be 0"); 1581 return -EINVAL; 1582 } 1583 info->mark = *mark; 1584 break; 1585 } 1586 #endif 1587 #ifdef CONFIG_NF_CONNTRACK_LABELS 1588 case OVS_CT_ATTR_LABELS: { 1589 struct md_labels *labels = nla_data(a); 1590 1591 if (!labels_nonzero(&labels->mask)) { 1592 OVS_NLERR(log, "ct_labels mask cannot be 0"); 1593 return -EINVAL; 1594 } 1595 info->labels = *labels; 1596 break; 1597 } 1598 #endif 1599 case OVS_CT_ATTR_HELPER: 1600 *helper = nla_data(a); 1601 if (!memchr(*helper, '\0', nla_len(a))) { 1602 OVS_NLERR(log, "Invalid conntrack helper"); 1603 return -EINVAL; 1604 } 1605 break; 1606 #if IS_ENABLED(CONFIG_NF_NAT) 1607 case OVS_CT_ATTR_NAT: { 1608 int err = parse_nat(a, info, log); 1609 1610 if (err) 1611 return err; 1612 break; 1613 } 1614 #endif 1615 case OVS_CT_ATTR_EVENTMASK: 1616 info->have_eventmask = true; 1617 info->eventmask = nla_get_u32(a); 1618 break; 1619 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT 1620 case OVS_CT_ATTR_TIMEOUT: 1621 memcpy(info->timeout, nla_data(a), nla_len(a)); 1622 if (!memchr(info->timeout, '\0', nla_len(a))) { 1623 OVS_NLERR(log, "Invalid conntrack timeout"); 1624 return -EINVAL; 1625 } 1626 break; 1627 #endif 1628 1629 default: 1630 OVS_NLERR(log, "Unknown conntrack attr (%d)", 1631 type); 1632 return -EINVAL; 1633 } 1634 } 1635 1636 #ifdef CONFIG_NF_CONNTRACK_MARK 1637 if (!info->commit && info->mark.mask) { 1638 OVS_NLERR(log, 1639 "Setting conntrack mark requires 'commit' flag."); 1640 return -EINVAL; 1641 } 1642 #endif 1643 #ifdef CONFIG_NF_CONNTRACK_LABELS 1644 if (!info->commit && labels_nonzero(&info->labels.mask)) { 1645 OVS_NLERR(log, 1646 "Setting conntrack labels requires 'commit' flag."); 1647 return -EINVAL; 1648 } 1649 #endif 1650 if (rem > 0) { 1651 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem); 1652 return -EINVAL; 1653 } 1654 1655 return 0; 1656 } 1657 1658 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr) 1659 { 1660 if (attr == OVS_KEY_ATTR_CT_STATE) 1661 return true; 1662 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1663 attr == OVS_KEY_ATTR_CT_ZONE) 1664 return true; 1665 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1666 attr == OVS_KEY_ATTR_CT_MARK) 1667 return true; 1668 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1669 attr == OVS_KEY_ATTR_CT_LABELS) { 1670 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 1671 1672 return ovs_net->xt_label; 1673 } 1674 1675 return false; 1676 } 1677 1678 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr, 1679 const struct sw_flow_key *key, 1680 struct sw_flow_actions **sfa, bool log) 1681 { 1682 struct ovs_conntrack_info ct_info; 1683 const char *helper = NULL; 1684 u16 family; 1685 int err; 1686 1687 family = key_to_nfproto(key); 1688 if (family == NFPROTO_UNSPEC) { 1689 OVS_NLERR(log, "ct family unspecified"); 1690 return -EINVAL; 1691 } 1692 1693 memset(&ct_info, 0, sizeof(ct_info)); 1694 ct_info.family = family; 1695 1696 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID, 1697 NF_CT_DEFAULT_ZONE_DIR, 0); 1698 1699 err = parse_ct(attr, &ct_info, &helper, log); 1700 if (err) 1701 return err; 1702 1703 /* Set up template for tracking connections in specific zones. */ 1704 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL); 1705 if (!ct_info.ct) { 1706 OVS_NLERR(log, "Failed to allocate conntrack template"); 1707 return -ENOMEM; 1708 } 1709 1710 if (ct_info.timeout[0]) { 1711 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto, 1712 ct_info.timeout)) 1713 pr_info_ratelimited("Failed to associated timeout " 1714 "policy `%s'\n", ct_info.timeout); 1715 else 1716 ct_info.nf_ct_timeout = rcu_dereference( 1717 nf_ct_timeout_find(ct_info.ct)->timeout); 1718 1719 } 1720 1721 if (helper) { 1722 err = ovs_ct_add_helper(&ct_info, helper, key, log); 1723 if (err) 1724 goto err_free_ct; 1725 } 1726 1727 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info, 1728 sizeof(ct_info), log); 1729 if (err) 1730 goto err_free_ct; 1731 1732 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status); 1733 return 0; 1734 err_free_ct: 1735 __ovs_ct_free_action(&ct_info); 1736 return err; 1737 } 1738 1739 #if IS_ENABLED(CONFIG_NF_NAT) 1740 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info, 1741 struct sk_buff *skb) 1742 { 1743 struct nlattr *start; 1744 1745 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT); 1746 if (!start) 1747 return false; 1748 1749 if (info->nat & OVS_CT_SRC_NAT) { 1750 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC)) 1751 return false; 1752 } else if (info->nat & OVS_CT_DST_NAT) { 1753 if (nla_put_flag(skb, OVS_NAT_ATTR_DST)) 1754 return false; 1755 } else { 1756 goto out; 1757 } 1758 1759 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) { 1760 if (IS_ENABLED(CONFIG_NF_NAT) && 1761 info->family == NFPROTO_IPV4) { 1762 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN, 1763 info->range.min_addr.ip) || 1764 (info->range.max_addr.ip 1765 != info->range.min_addr.ip && 1766 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX, 1767 info->range.max_addr.ip)))) 1768 return false; 1769 } else if (IS_ENABLED(CONFIG_IPV6) && 1770 info->family == NFPROTO_IPV6) { 1771 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN, 1772 &info->range.min_addr.in6) || 1773 (memcmp(&info->range.max_addr.in6, 1774 &info->range.min_addr.in6, 1775 sizeof(info->range.max_addr.in6)) && 1776 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX, 1777 &info->range.max_addr.in6)))) 1778 return false; 1779 } else { 1780 return false; 1781 } 1782 } 1783 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED && 1784 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN, 1785 ntohs(info->range.min_proto.all)) || 1786 (info->range.max_proto.all != info->range.min_proto.all && 1787 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX, 1788 ntohs(info->range.max_proto.all))))) 1789 return false; 1790 1791 if (info->range.flags & NF_NAT_RANGE_PERSISTENT && 1792 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT)) 1793 return false; 1794 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM && 1795 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH)) 1796 return false; 1797 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY && 1798 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM)) 1799 return false; 1800 out: 1801 nla_nest_end(skb, start); 1802 1803 return true; 1804 } 1805 #endif 1806 1807 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info, 1808 struct sk_buff *skb) 1809 { 1810 struct nlattr *start; 1811 1812 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT); 1813 if (!start) 1814 return -EMSGSIZE; 1815 1816 if (ct_info->commit && nla_put_flag(skb, ct_info->force 1817 ? OVS_CT_ATTR_FORCE_COMMIT 1818 : OVS_CT_ATTR_COMMIT)) 1819 return -EMSGSIZE; 1820 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1821 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id)) 1822 return -EMSGSIZE; 1823 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask && 1824 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark), 1825 &ct_info->mark)) 1826 return -EMSGSIZE; 1827 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1828 labels_nonzero(&ct_info->labels.mask) && 1829 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels), 1830 &ct_info->labels)) 1831 return -EMSGSIZE; 1832 if (ct_info->helper) { 1833 if (nla_put_string(skb, OVS_CT_ATTR_HELPER, 1834 ct_info->helper->name)) 1835 return -EMSGSIZE; 1836 } 1837 if (ct_info->have_eventmask && 1838 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask)) 1839 return -EMSGSIZE; 1840 if (ct_info->timeout[0]) { 1841 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout)) 1842 return -EMSGSIZE; 1843 } 1844 1845 #if IS_ENABLED(CONFIG_NF_NAT) 1846 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb)) 1847 return -EMSGSIZE; 1848 #endif 1849 nla_nest_end(skb, start); 1850 1851 return 0; 1852 } 1853 1854 void ovs_ct_free_action(const struct nlattr *a) 1855 { 1856 struct ovs_conntrack_info *ct_info = nla_data(a); 1857 1858 __ovs_ct_free_action(ct_info); 1859 } 1860 1861 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info) 1862 { 1863 if (ct_info->helper) { 1864 #if IS_ENABLED(CONFIG_NF_NAT) 1865 if (ct_info->nat) 1866 nf_nat_helper_put(ct_info->helper); 1867 #endif 1868 nf_conntrack_helper_put(ct_info->helper); 1869 } 1870 if (ct_info->ct) { 1871 if (ct_info->timeout[0]) 1872 nf_ct_destroy_timeout(ct_info->ct); 1873 nf_ct_tmpl_free(ct_info->ct); 1874 } 1875 } 1876 1877 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 1878 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net) 1879 { 1880 int i, err; 1881 1882 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info), 1883 GFP_KERNEL); 1884 if (!ovs_net->ct_limit_info) 1885 return -ENOMEM; 1886 1887 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT; 1888 ovs_net->ct_limit_info->limits = 1889 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head), 1890 GFP_KERNEL); 1891 if (!ovs_net->ct_limit_info->limits) { 1892 kfree(ovs_net->ct_limit_info); 1893 return -ENOMEM; 1894 } 1895 1896 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++) 1897 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]); 1898 1899 ovs_net->ct_limit_info->data = 1900 nf_conncount_init(net, NFPROTO_INET, sizeof(u32)); 1901 1902 if (IS_ERR(ovs_net->ct_limit_info->data)) { 1903 err = PTR_ERR(ovs_net->ct_limit_info->data); 1904 kfree(ovs_net->ct_limit_info->limits); 1905 kfree(ovs_net->ct_limit_info); 1906 pr_err("openvswitch: failed to init nf_conncount %d\n", err); 1907 return err; 1908 } 1909 return 0; 1910 } 1911 1912 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net) 1913 { 1914 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info; 1915 int i; 1916 1917 nf_conncount_destroy(net, NFPROTO_INET, info->data); 1918 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) { 1919 struct hlist_head *head = &info->limits[i]; 1920 struct ovs_ct_limit *ct_limit; 1921 1922 hlist_for_each_entry_rcu(ct_limit, head, hlist_node, 1923 lockdep_ovsl_is_held()) 1924 kfree_rcu(ct_limit, rcu); 1925 } 1926 kfree(info->limits); 1927 kfree(info); 1928 } 1929 1930 static struct sk_buff * 1931 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd, 1932 struct ovs_header **ovs_reply_header) 1933 { 1934 struct ovs_header *ovs_header = info->userhdr; 1935 struct sk_buff *skb; 1936 1937 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1938 if (!skb) 1939 return ERR_PTR(-ENOMEM); 1940 1941 *ovs_reply_header = genlmsg_put(skb, info->snd_portid, 1942 info->snd_seq, 1943 &dp_ct_limit_genl_family, 0, cmd); 1944 1945 if (!*ovs_reply_header) { 1946 nlmsg_free(skb); 1947 return ERR_PTR(-EMSGSIZE); 1948 } 1949 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex; 1950 1951 return skb; 1952 } 1953 1954 static bool check_zone_id(int zone_id, u16 *pzone) 1955 { 1956 if (zone_id >= 0 && zone_id <= 65535) { 1957 *pzone = (u16)zone_id; 1958 return true; 1959 } 1960 return false; 1961 } 1962 1963 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, 1964 struct ovs_ct_limit_info *info) 1965 { 1966 struct ovs_zone_limit *zone_limit; 1967 int rem; 1968 u16 zone; 1969 1970 rem = NLA_ALIGN(nla_len(nla_zone_limit)); 1971 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit); 1972 1973 while (rem >= sizeof(*zone_limit)) { 1974 if (unlikely(zone_limit->zone_id == 1975 OVS_ZONE_LIMIT_DEFAULT_ZONE)) { 1976 ovs_lock(); 1977 info->default_limit = zone_limit->limit; 1978 ovs_unlock(); 1979 } else if (unlikely(!check_zone_id( 1980 zone_limit->zone_id, &zone))) { 1981 OVS_NLERR(true, "zone id is out of range"); 1982 } else { 1983 struct ovs_ct_limit *ct_limit; 1984 1985 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL); 1986 if (!ct_limit) 1987 return -ENOMEM; 1988 1989 ct_limit->zone = zone; 1990 ct_limit->limit = zone_limit->limit; 1991 1992 ovs_lock(); 1993 ct_limit_set(info, ct_limit); 1994 ovs_unlock(); 1995 } 1996 rem -= NLA_ALIGN(sizeof(*zone_limit)); 1997 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit + 1998 NLA_ALIGN(sizeof(*zone_limit))); 1999 } 2000 2001 if (rem) 2002 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem); 2003 2004 return 0; 2005 } 2006 2007 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit, 2008 struct ovs_ct_limit_info *info) 2009 { 2010 struct ovs_zone_limit *zone_limit; 2011 int rem; 2012 u16 zone; 2013 2014 rem = NLA_ALIGN(nla_len(nla_zone_limit)); 2015 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit); 2016 2017 while (rem >= sizeof(*zone_limit)) { 2018 if (unlikely(zone_limit->zone_id == 2019 OVS_ZONE_LIMIT_DEFAULT_ZONE)) { 2020 ovs_lock(); 2021 info->default_limit = OVS_CT_LIMIT_DEFAULT; 2022 ovs_unlock(); 2023 } else if (unlikely(!check_zone_id( 2024 zone_limit->zone_id, &zone))) { 2025 OVS_NLERR(true, "zone id is out of range"); 2026 } else { 2027 ovs_lock(); 2028 ct_limit_del(info, zone); 2029 ovs_unlock(); 2030 } 2031 rem -= NLA_ALIGN(sizeof(*zone_limit)); 2032 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit + 2033 NLA_ALIGN(sizeof(*zone_limit))); 2034 } 2035 2036 if (rem) 2037 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem); 2038 2039 return 0; 2040 } 2041 2042 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info, 2043 struct sk_buff *reply) 2044 { 2045 struct ovs_zone_limit zone_limit = { 2046 .zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE, 2047 .limit = info->default_limit, 2048 }; 2049 2050 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit); 2051 } 2052 2053 static int __ovs_ct_limit_get_zone_limit(struct net *net, 2054 struct nf_conncount_data *data, 2055 u16 zone_id, u32 limit, 2056 struct sk_buff *reply) 2057 { 2058 struct nf_conntrack_zone ct_zone; 2059 struct ovs_zone_limit zone_limit; 2060 u32 conncount_key = zone_id; 2061 2062 zone_limit.zone_id = zone_id; 2063 zone_limit.limit = limit; 2064 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0); 2065 2066 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL, 2067 &ct_zone); 2068 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit); 2069 } 2070 2071 static int ovs_ct_limit_get_zone_limit(struct net *net, 2072 struct nlattr *nla_zone_limit, 2073 struct ovs_ct_limit_info *info, 2074 struct sk_buff *reply) 2075 { 2076 struct ovs_zone_limit *zone_limit; 2077 int rem, err; 2078 u32 limit; 2079 u16 zone; 2080 2081 rem = NLA_ALIGN(nla_len(nla_zone_limit)); 2082 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit); 2083 2084 while (rem >= sizeof(*zone_limit)) { 2085 if (unlikely(zone_limit->zone_id == 2086 OVS_ZONE_LIMIT_DEFAULT_ZONE)) { 2087 err = ovs_ct_limit_get_default_limit(info, reply); 2088 if (err) 2089 return err; 2090 } else if (unlikely(!check_zone_id(zone_limit->zone_id, 2091 &zone))) { 2092 OVS_NLERR(true, "zone id is out of range"); 2093 } else { 2094 rcu_read_lock(); 2095 limit = ct_limit_get(info, zone); 2096 rcu_read_unlock(); 2097 2098 err = __ovs_ct_limit_get_zone_limit( 2099 net, info->data, zone, limit, reply); 2100 if (err) 2101 return err; 2102 } 2103 rem -= NLA_ALIGN(sizeof(*zone_limit)); 2104 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit + 2105 NLA_ALIGN(sizeof(*zone_limit))); 2106 } 2107 2108 if (rem) 2109 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem); 2110 2111 return 0; 2112 } 2113 2114 static int ovs_ct_limit_get_all_zone_limit(struct net *net, 2115 struct ovs_ct_limit_info *info, 2116 struct sk_buff *reply) 2117 { 2118 struct ovs_ct_limit *ct_limit; 2119 struct hlist_head *head; 2120 int i, err = 0; 2121 2122 err = ovs_ct_limit_get_default_limit(info, reply); 2123 if (err) 2124 return err; 2125 2126 rcu_read_lock(); 2127 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) { 2128 head = &info->limits[i]; 2129 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) { 2130 err = __ovs_ct_limit_get_zone_limit(net, info->data, 2131 ct_limit->zone, ct_limit->limit, reply); 2132 if (err) 2133 goto exit_err; 2134 } 2135 } 2136 2137 exit_err: 2138 rcu_read_unlock(); 2139 return err; 2140 } 2141 2142 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) 2143 { 2144 struct nlattr **a = info->attrs; 2145 struct sk_buff *reply; 2146 struct ovs_header *ovs_reply_header; 2147 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); 2148 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; 2149 int err; 2150 2151 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET, 2152 &ovs_reply_header); 2153 if (IS_ERR(reply)) 2154 return PTR_ERR(reply); 2155 2156 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) { 2157 err = -EINVAL; 2158 goto exit_err; 2159 } 2160 2161 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], 2162 ct_limit_info); 2163 if (err) 2164 goto exit_err; 2165 2166 static_branch_enable(&ovs_ct_limit_enabled); 2167 2168 genlmsg_end(reply, ovs_reply_header); 2169 return genlmsg_reply(reply, info); 2170 2171 exit_err: 2172 nlmsg_free(reply); 2173 return err; 2174 } 2175 2176 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) 2177 { 2178 struct nlattr **a = info->attrs; 2179 struct sk_buff *reply; 2180 struct ovs_header *ovs_reply_header; 2181 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); 2182 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; 2183 int err; 2184 2185 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL, 2186 &ovs_reply_header); 2187 if (IS_ERR(reply)) 2188 return PTR_ERR(reply); 2189 2190 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) { 2191 err = -EINVAL; 2192 goto exit_err; 2193 } 2194 2195 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], 2196 ct_limit_info); 2197 if (err) 2198 goto exit_err; 2199 2200 genlmsg_end(reply, ovs_reply_header); 2201 return genlmsg_reply(reply, info); 2202 2203 exit_err: 2204 nlmsg_free(reply); 2205 return err; 2206 } 2207 2208 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) 2209 { 2210 struct nlattr **a = info->attrs; 2211 struct nlattr *nla_reply; 2212 struct sk_buff *reply; 2213 struct ovs_header *ovs_reply_header; 2214 struct net *net = sock_net(skb->sk); 2215 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2216 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; 2217 int err; 2218 2219 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET, 2220 &ovs_reply_header); 2221 if (IS_ERR(reply)) 2222 return PTR_ERR(reply); 2223 2224 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT); 2225 if (!nla_reply) { 2226 err = -EMSGSIZE; 2227 goto exit_err; 2228 } 2229 2230 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) { 2231 err = ovs_ct_limit_get_zone_limit( 2232 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info, 2233 reply); 2234 if (err) 2235 goto exit_err; 2236 } else { 2237 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info, 2238 reply); 2239 if (err) 2240 goto exit_err; 2241 } 2242 2243 nla_nest_end(reply, nla_reply); 2244 genlmsg_end(reply, ovs_reply_header); 2245 return genlmsg_reply(reply, info); 2246 2247 exit_err: 2248 nlmsg_free(reply); 2249 return err; 2250 } 2251 2252 static const struct genl_small_ops ct_limit_genl_ops[] = { 2253 { .cmd = OVS_CT_LIMIT_CMD_SET, 2254 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2255 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN 2256 * privilege. */ 2257 .doit = ovs_ct_limit_cmd_set, 2258 }, 2259 { .cmd = OVS_CT_LIMIT_CMD_DEL, 2260 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2261 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN 2262 * privilege. */ 2263 .doit = ovs_ct_limit_cmd_del, 2264 }, 2265 { .cmd = OVS_CT_LIMIT_CMD_GET, 2266 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2267 .flags = 0, /* OK for unprivileged users. */ 2268 .doit = ovs_ct_limit_cmd_get, 2269 }, 2270 }; 2271 2272 static const struct genl_multicast_group ovs_ct_limit_multicast_group = { 2273 .name = OVS_CT_LIMIT_MCGROUP, 2274 }; 2275 2276 struct genl_family dp_ct_limit_genl_family __ro_after_init = { 2277 .hdrsize = sizeof(struct ovs_header), 2278 .name = OVS_CT_LIMIT_FAMILY, 2279 .version = OVS_CT_LIMIT_VERSION, 2280 .maxattr = OVS_CT_LIMIT_ATTR_MAX, 2281 .policy = ct_limit_policy, 2282 .netnsok = true, 2283 .parallel_ops = true, 2284 .small_ops = ct_limit_genl_ops, 2285 .n_small_ops = ARRAY_SIZE(ct_limit_genl_ops), 2286 .mcgrps = &ovs_ct_limit_multicast_group, 2287 .n_mcgrps = 1, 2288 .module = THIS_MODULE, 2289 }; 2290 #endif 2291 2292 int ovs_ct_init(struct net *net) 2293 { 2294 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE; 2295 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2296 2297 if (nf_connlabels_get(net, n_bits - 1)) { 2298 ovs_net->xt_label = false; 2299 OVS_NLERR(true, "Failed to set connlabel length"); 2300 } else { 2301 ovs_net->xt_label = true; 2302 } 2303 2304 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 2305 return ovs_ct_limit_init(net, ovs_net); 2306 #else 2307 return 0; 2308 #endif 2309 } 2310 2311 void ovs_ct_exit(struct net *net) 2312 { 2313 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2314 2315 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 2316 ovs_ct_limit_exit(net, ovs_net); 2317 #endif 2318 2319 if (ovs_net->xt_label) 2320 nf_connlabels_put(net); 2321 } 2322