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