1 /* 2 * Copyright (c) 2007-2012 Nicira Networks. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of version 2 of the GNU General Public 6 * License as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, but 9 * WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public License 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 16 * 02110-1301, USA 17 */ 18 19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 20 21 #include <linux/init.h> 22 #include <linux/module.h> 23 #include <linux/if_arp.h> 24 #include <linux/if_vlan.h> 25 #include <linux/in.h> 26 #include <linux/ip.h> 27 #include <linux/jhash.h> 28 #include <linux/delay.h> 29 #include <linux/time.h> 30 #include <linux/etherdevice.h> 31 #include <linux/genetlink.h> 32 #include <linux/kernel.h> 33 #include <linux/kthread.h> 34 #include <linux/mutex.h> 35 #include <linux/percpu.h> 36 #include <linux/rcupdate.h> 37 #include <linux/tcp.h> 38 #include <linux/udp.h> 39 #include <linux/ethtool.h> 40 #include <linux/wait.h> 41 #include <asm/div64.h> 42 #include <linux/highmem.h> 43 #include <linux/netfilter_bridge.h> 44 #include <linux/netfilter_ipv4.h> 45 #include <linux/inetdevice.h> 46 #include <linux/list.h> 47 #include <linux/openvswitch.h> 48 #include <linux/rculist.h> 49 #include <linux/dmi.h> 50 #include <linux/workqueue.h> 51 #include <net/genetlink.h> 52 53 #include "datapath.h" 54 #include "flow.h" 55 #include "vport-internal_dev.h" 56 57 /** 58 * DOC: Locking: 59 * 60 * Writes to device state (add/remove datapath, port, set operations on vports, 61 * etc.) are protected by RTNL. 62 * 63 * Writes to other state (flow table modifications, set miscellaneous datapath 64 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside 65 * genl_mutex. 66 * 67 * Reads are protected by RCU. 68 * 69 * There are a few special cases (mostly stats) that have their own 70 * synchronization but they nest under all of above and don't interact with 71 * each other. 72 */ 73 74 /* Global list of datapaths to enable dumping them all out. 75 * Protected by genl_mutex. 76 */ 77 static LIST_HEAD(dps); 78 79 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ) 80 static void rehash_flow_table(struct work_struct *work); 81 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table); 82 83 static struct vport *new_vport(const struct vport_parms *); 84 static int queue_gso_packets(int dp_ifindex, struct sk_buff *, 85 const struct dp_upcall_info *); 86 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *, 87 const struct dp_upcall_info *); 88 89 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */ 90 static struct datapath *get_dp(int dp_ifindex) 91 { 92 struct datapath *dp = NULL; 93 struct net_device *dev; 94 95 rcu_read_lock(); 96 dev = dev_get_by_index_rcu(&init_net, dp_ifindex); 97 if (dev) { 98 struct vport *vport = ovs_internal_dev_get_vport(dev); 99 if (vport) 100 dp = vport->dp; 101 } 102 rcu_read_unlock(); 103 104 return dp; 105 } 106 107 /* Must be called with rcu_read_lock or RTNL lock. */ 108 const char *ovs_dp_name(const struct datapath *dp) 109 { 110 struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]); 111 return vport->ops->get_name(vport); 112 } 113 114 static int get_dpifindex(struct datapath *dp) 115 { 116 struct vport *local; 117 int ifindex; 118 119 rcu_read_lock(); 120 121 local = rcu_dereference(dp->ports[OVSP_LOCAL]); 122 if (local) 123 ifindex = local->ops->get_ifindex(local); 124 else 125 ifindex = 0; 126 127 rcu_read_unlock(); 128 129 return ifindex; 130 } 131 132 static void destroy_dp_rcu(struct rcu_head *rcu) 133 { 134 struct datapath *dp = container_of(rcu, struct datapath, rcu); 135 136 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table); 137 free_percpu(dp->stats_percpu); 138 kfree(dp); 139 } 140 141 /* Called with RTNL lock and genl_lock. */ 142 static struct vport *new_vport(const struct vport_parms *parms) 143 { 144 struct vport *vport; 145 146 vport = ovs_vport_add(parms); 147 if (!IS_ERR(vport)) { 148 struct datapath *dp = parms->dp; 149 150 rcu_assign_pointer(dp->ports[parms->port_no], vport); 151 list_add(&vport->node, &dp->port_list); 152 } 153 154 return vport; 155 } 156 157 /* Called with RTNL lock. */ 158 void ovs_dp_detach_port(struct vport *p) 159 { 160 ASSERT_RTNL(); 161 162 /* First drop references to device. */ 163 list_del(&p->node); 164 rcu_assign_pointer(p->dp->ports[p->port_no], NULL); 165 166 /* Then destroy it. */ 167 ovs_vport_del(p); 168 } 169 170 /* Must be called with rcu_read_lock. */ 171 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb) 172 { 173 struct datapath *dp = p->dp; 174 struct sw_flow *flow; 175 struct dp_stats_percpu *stats; 176 struct sw_flow_key key; 177 u64 *stats_counter; 178 int error; 179 int key_len; 180 181 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); 182 183 /* Extract flow from 'skb' into 'key'. */ 184 error = ovs_flow_extract(skb, p->port_no, &key, &key_len); 185 if (unlikely(error)) { 186 kfree_skb(skb); 187 return; 188 } 189 190 /* Look up flow. */ 191 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len); 192 if (unlikely(!flow)) { 193 struct dp_upcall_info upcall; 194 195 upcall.cmd = OVS_PACKET_CMD_MISS; 196 upcall.key = &key; 197 upcall.userdata = NULL; 198 upcall.pid = p->upcall_pid; 199 ovs_dp_upcall(dp, skb, &upcall); 200 consume_skb(skb); 201 stats_counter = &stats->n_missed; 202 goto out; 203 } 204 205 OVS_CB(skb)->flow = flow; 206 207 stats_counter = &stats->n_hit; 208 ovs_flow_used(OVS_CB(skb)->flow, skb); 209 ovs_execute_actions(dp, skb); 210 211 out: 212 /* Update datapath statistics. */ 213 u64_stats_update_begin(&stats->sync); 214 (*stats_counter)++; 215 u64_stats_update_end(&stats->sync); 216 } 217 218 static struct genl_family dp_packet_genl_family = { 219 .id = GENL_ID_GENERATE, 220 .hdrsize = sizeof(struct ovs_header), 221 .name = OVS_PACKET_FAMILY, 222 .version = OVS_PACKET_VERSION, 223 .maxattr = OVS_PACKET_ATTR_MAX 224 }; 225 226 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, 227 const struct dp_upcall_info *upcall_info) 228 { 229 struct dp_stats_percpu *stats; 230 int dp_ifindex; 231 int err; 232 233 if (upcall_info->pid == 0) { 234 err = -ENOTCONN; 235 goto err; 236 } 237 238 dp_ifindex = get_dpifindex(dp); 239 if (!dp_ifindex) { 240 err = -ENODEV; 241 goto err; 242 } 243 244 if (!skb_is_gso(skb)) 245 err = queue_userspace_packet(dp_ifindex, skb, upcall_info); 246 else 247 err = queue_gso_packets(dp_ifindex, skb, upcall_info); 248 if (err) 249 goto err; 250 251 return 0; 252 253 err: 254 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); 255 256 u64_stats_update_begin(&stats->sync); 257 stats->n_lost++; 258 u64_stats_update_end(&stats->sync); 259 260 return err; 261 } 262 263 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb, 264 const struct dp_upcall_info *upcall_info) 265 { 266 struct dp_upcall_info later_info; 267 struct sw_flow_key later_key; 268 struct sk_buff *segs, *nskb; 269 int err; 270 271 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); 272 if (IS_ERR(skb)) 273 return PTR_ERR(skb); 274 275 /* Queue all of the segments. */ 276 skb = segs; 277 do { 278 err = queue_userspace_packet(dp_ifindex, skb, upcall_info); 279 if (err) 280 break; 281 282 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) { 283 /* The initial flow key extracted by ovs_flow_extract() 284 * in this case is for a first fragment, so we need to 285 * properly mark later fragments. 286 */ 287 later_key = *upcall_info->key; 288 later_key.ip.frag = OVS_FRAG_TYPE_LATER; 289 290 later_info = *upcall_info; 291 later_info.key = &later_key; 292 upcall_info = &later_info; 293 } 294 } while ((skb = skb->next)); 295 296 /* Free all of the segments. */ 297 skb = segs; 298 do { 299 nskb = skb->next; 300 if (err) 301 kfree_skb(skb); 302 else 303 consume_skb(skb); 304 } while ((skb = nskb)); 305 return err; 306 } 307 308 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb, 309 const struct dp_upcall_info *upcall_info) 310 { 311 struct ovs_header *upcall; 312 struct sk_buff *nskb = NULL; 313 struct sk_buff *user_skb; /* to be queued to userspace */ 314 struct nlattr *nla; 315 unsigned int len; 316 int err; 317 318 if (vlan_tx_tag_present(skb)) { 319 nskb = skb_clone(skb, GFP_ATOMIC); 320 if (!nskb) 321 return -ENOMEM; 322 323 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb)); 324 if (!nskb) 325 return -ENOMEM; 326 327 nskb->vlan_tci = 0; 328 skb = nskb; 329 } 330 331 if (nla_attr_size(skb->len) > USHRT_MAX) { 332 err = -EFBIG; 333 goto out; 334 } 335 336 len = sizeof(struct ovs_header); 337 len += nla_total_size(skb->len); 338 len += nla_total_size(FLOW_BUFSIZE); 339 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION) 340 len += nla_total_size(8); 341 342 user_skb = genlmsg_new(len, GFP_ATOMIC); 343 if (!user_skb) { 344 err = -ENOMEM; 345 goto out; 346 } 347 348 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 349 0, upcall_info->cmd); 350 upcall->dp_ifindex = dp_ifindex; 351 352 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY); 353 ovs_flow_to_nlattrs(upcall_info->key, user_skb); 354 nla_nest_end(user_skb, nla); 355 356 if (upcall_info->userdata) 357 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, 358 nla_get_u64(upcall_info->userdata)); 359 360 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len); 361 362 skb_copy_and_csum_dev(skb, nla_data(nla)); 363 364 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid); 365 366 out: 367 kfree_skb(nskb); 368 return err; 369 } 370 371 /* Called with genl_mutex. */ 372 static int flush_flows(int dp_ifindex) 373 { 374 struct flow_table *old_table; 375 struct flow_table *new_table; 376 struct datapath *dp; 377 378 dp = get_dp(dp_ifindex); 379 if (!dp) 380 return -ENODEV; 381 382 old_table = genl_dereference(dp->table); 383 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS); 384 if (!new_table) 385 return -ENOMEM; 386 387 rcu_assign_pointer(dp->table, new_table); 388 389 ovs_flow_tbl_deferred_destroy(old_table); 390 return 0; 391 } 392 393 static int validate_actions(const struct nlattr *attr, 394 const struct sw_flow_key *key, int depth); 395 396 static int validate_sample(const struct nlattr *attr, 397 const struct sw_flow_key *key, int depth) 398 { 399 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1]; 400 const struct nlattr *probability, *actions; 401 const struct nlattr *a; 402 int rem; 403 404 memset(attrs, 0, sizeof(attrs)); 405 nla_for_each_nested(a, attr, rem) { 406 int type = nla_type(a); 407 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type]) 408 return -EINVAL; 409 attrs[type] = a; 410 } 411 if (rem) 412 return -EINVAL; 413 414 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY]; 415 if (!probability || nla_len(probability) != sizeof(u32)) 416 return -EINVAL; 417 418 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS]; 419 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) 420 return -EINVAL; 421 return validate_actions(actions, key, depth + 1); 422 } 423 424 static int validate_tp_port(const struct sw_flow_key *flow_key) 425 { 426 if (flow_key->eth.type == htons(ETH_P_IP)) { 427 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst) 428 return 0; 429 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) { 430 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst) 431 return 0; 432 } 433 434 return -EINVAL; 435 } 436 437 static int validate_set(const struct nlattr *a, 438 const struct sw_flow_key *flow_key) 439 { 440 const struct nlattr *ovs_key = nla_data(a); 441 int key_type = nla_type(ovs_key); 442 443 /* There can be only one key in a action */ 444 if (nla_total_size(nla_len(ovs_key)) != nla_len(a)) 445 return -EINVAL; 446 447 if (key_type > OVS_KEY_ATTR_MAX || 448 nla_len(ovs_key) != ovs_key_lens[key_type]) 449 return -EINVAL; 450 451 switch (key_type) { 452 const struct ovs_key_ipv4 *ipv4_key; 453 454 case OVS_KEY_ATTR_PRIORITY: 455 case OVS_KEY_ATTR_ETHERNET: 456 break; 457 458 case OVS_KEY_ATTR_IPV4: 459 if (flow_key->eth.type != htons(ETH_P_IP)) 460 return -EINVAL; 461 462 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst) 463 return -EINVAL; 464 465 ipv4_key = nla_data(ovs_key); 466 if (ipv4_key->ipv4_proto != flow_key->ip.proto) 467 return -EINVAL; 468 469 if (ipv4_key->ipv4_frag != flow_key->ip.frag) 470 return -EINVAL; 471 472 break; 473 474 case OVS_KEY_ATTR_TCP: 475 if (flow_key->ip.proto != IPPROTO_TCP) 476 return -EINVAL; 477 478 return validate_tp_port(flow_key); 479 480 case OVS_KEY_ATTR_UDP: 481 if (flow_key->ip.proto != IPPROTO_UDP) 482 return -EINVAL; 483 484 return validate_tp_port(flow_key); 485 486 default: 487 return -EINVAL; 488 } 489 490 return 0; 491 } 492 493 static int validate_userspace(const struct nlattr *attr) 494 { 495 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = { 496 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 }, 497 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 }, 498 }; 499 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1]; 500 int error; 501 502 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, 503 attr, userspace_policy); 504 if (error) 505 return error; 506 507 if (!a[OVS_USERSPACE_ATTR_PID] || 508 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID])) 509 return -EINVAL; 510 511 return 0; 512 } 513 514 static int validate_actions(const struct nlattr *attr, 515 const struct sw_flow_key *key, int depth) 516 { 517 const struct nlattr *a; 518 int rem, err; 519 520 if (depth >= SAMPLE_ACTION_DEPTH) 521 return -EOVERFLOW; 522 523 nla_for_each_nested(a, attr, rem) { 524 /* Expected argument lengths, (u32)-1 for variable length. */ 525 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = { 526 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32), 527 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1, 528 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan), 529 [OVS_ACTION_ATTR_POP_VLAN] = 0, 530 [OVS_ACTION_ATTR_SET] = (u32)-1, 531 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1 532 }; 533 const struct ovs_action_push_vlan *vlan; 534 int type = nla_type(a); 535 536 if (type > OVS_ACTION_ATTR_MAX || 537 (action_lens[type] != nla_len(a) && 538 action_lens[type] != (u32)-1)) 539 return -EINVAL; 540 541 switch (type) { 542 case OVS_ACTION_ATTR_UNSPEC: 543 return -EINVAL; 544 545 case OVS_ACTION_ATTR_USERSPACE: 546 err = validate_userspace(a); 547 if (err) 548 return err; 549 break; 550 551 case OVS_ACTION_ATTR_OUTPUT: 552 if (nla_get_u32(a) >= DP_MAX_PORTS) 553 return -EINVAL; 554 break; 555 556 557 case OVS_ACTION_ATTR_POP_VLAN: 558 break; 559 560 case OVS_ACTION_ATTR_PUSH_VLAN: 561 vlan = nla_data(a); 562 if (vlan->vlan_tpid != htons(ETH_P_8021Q)) 563 return -EINVAL; 564 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT))) 565 return -EINVAL; 566 break; 567 568 case OVS_ACTION_ATTR_SET: 569 err = validate_set(a, key); 570 if (err) 571 return err; 572 break; 573 574 case OVS_ACTION_ATTR_SAMPLE: 575 err = validate_sample(a, key, depth); 576 if (err) 577 return err; 578 break; 579 580 default: 581 return -EINVAL; 582 } 583 } 584 585 if (rem > 0) 586 return -EINVAL; 587 588 return 0; 589 } 590 591 static void clear_stats(struct sw_flow *flow) 592 { 593 flow->used = 0; 594 flow->tcp_flags = 0; 595 flow->packet_count = 0; 596 flow->byte_count = 0; 597 } 598 599 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) 600 { 601 struct ovs_header *ovs_header = info->userhdr; 602 struct nlattr **a = info->attrs; 603 struct sw_flow_actions *acts; 604 struct sk_buff *packet; 605 struct sw_flow *flow; 606 struct datapath *dp; 607 struct ethhdr *eth; 608 int len; 609 int err; 610 int key_len; 611 612 err = -EINVAL; 613 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || 614 !a[OVS_PACKET_ATTR_ACTIONS] || 615 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN) 616 goto err; 617 618 len = nla_len(a[OVS_PACKET_ATTR_PACKET]); 619 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); 620 err = -ENOMEM; 621 if (!packet) 622 goto err; 623 skb_reserve(packet, NET_IP_ALIGN); 624 625 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len); 626 627 skb_reset_mac_header(packet); 628 eth = eth_hdr(packet); 629 630 /* Normally, setting the skb 'protocol' field would be handled by a 631 * call to eth_type_trans(), but it assumes there's a sending 632 * device, which we may not have. */ 633 if (ntohs(eth->h_proto) >= 1536) 634 packet->protocol = eth->h_proto; 635 else 636 packet->protocol = htons(ETH_P_802_2); 637 638 /* Build an sw_flow for sending this packet. */ 639 flow = ovs_flow_alloc(); 640 err = PTR_ERR(flow); 641 if (IS_ERR(flow)) 642 goto err_kfree_skb; 643 644 err = ovs_flow_extract(packet, -1, &flow->key, &key_len); 645 if (err) 646 goto err_flow_free; 647 648 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority, 649 &flow->key.phy.in_port, 650 a[OVS_PACKET_ATTR_KEY]); 651 if (err) 652 goto err_flow_free; 653 654 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0); 655 if (err) 656 goto err_flow_free; 657 658 flow->hash = ovs_flow_hash(&flow->key, key_len); 659 660 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]); 661 err = PTR_ERR(acts); 662 if (IS_ERR(acts)) 663 goto err_flow_free; 664 rcu_assign_pointer(flow->sf_acts, acts); 665 666 OVS_CB(packet)->flow = flow; 667 packet->priority = flow->key.phy.priority; 668 669 rcu_read_lock(); 670 dp = get_dp(ovs_header->dp_ifindex); 671 err = -ENODEV; 672 if (!dp) 673 goto err_unlock; 674 675 local_bh_disable(); 676 err = ovs_execute_actions(dp, packet); 677 local_bh_enable(); 678 rcu_read_unlock(); 679 680 ovs_flow_free(flow); 681 return err; 682 683 err_unlock: 684 rcu_read_unlock(); 685 err_flow_free: 686 ovs_flow_free(flow); 687 err_kfree_skb: 688 kfree_skb(packet); 689 err: 690 return err; 691 } 692 693 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { 694 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, 695 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, 696 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, 697 }; 698 699 static struct genl_ops dp_packet_genl_ops[] = { 700 { .cmd = OVS_PACKET_CMD_EXECUTE, 701 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 702 .policy = packet_policy, 703 .doit = ovs_packet_cmd_execute 704 } 705 }; 706 707 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats) 708 { 709 int i; 710 struct flow_table *table = genl_dereference(dp->table); 711 712 stats->n_flows = ovs_flow_tbl_count(table); 713 714 stats->n_hit = stats->n_missed = stats->n_lost = 0; 715 for_each_possible_cpu(i) { 716 const struct dp_stats_percpu *percpu_stats; 717 struct dp_stats_percpu local_stats; 718 unsigned int start; 719 720 percpu_stats = per_cpu_ptr(dp->stats_percpu, i); 721 722 do { 723 start = u64_stats_fetch_begin_bh(&percpu_stats->sync); 724 local_stats = *percpu_stats; 725 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start)); 726 727 stats->n_hit += local_stats.n_hit; 728 stats->n_missed += local_stats.n_missed; 729 stats->n_lost += local_stats.n_lost; 730 } 731 } 732 733 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { 734 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, 735 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, 736 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, 737 }; 738 739 static struct genl_family dp_flow_genl_family = { 740 .id = GENL_ID_GENERATE, 741 .hdrsize = sizeof(struct ovs_header), 742 .name = OVS_FLOW_FAMILY, 743 .version = OVS_FLOW_VERSION, 744 .maxattr = OVS_FLOW_ATTR_MAX 745 }; 746 747 static struct genl_multicast_group ovs_dp_flow_multicast_group = { 748 .name = OVS_FLOW_MCGROUP 749 }; 750 751 /* Called with genl_lock. */ 752 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, 753 struct sk_buff *skb, u32 pid, 754 u32 seq, u32 flags, u8 cmd) 755 { 756 const int skb_orig_len = skb->len; 757 const struct sw_flow_actions *sf_acts; 758 struct ovs_flow_stats stats; 759 struct ovs_header *ovs_header; 760 struct nlattr *nla; 761 unsigned long used; 762 u8 tcp_flags; 763 int err; 764 765 sf_acts = rcu_dereference_protected(flow->sf_acts, 766 lockdep_genl_is_held()); 767 768 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd); 769 if (!ovs_header) 770 return -EMSGSIZE; 771 772 ovs_header->dp_ifindex = get_dpifindex(dp); 773 774 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY); 775 if (!nla) 776 goto nla_put_failure; 777 err = ovs_flow_to_nlattrs(&flow->key, skb); 778 if (err) 779 goto error; 780 nla_nest_end(skb, nla); 781 782 spin_lock_bh(&flow->lock); 783 used = flow->used; 784 stats.n_packets = flow->packet_count; 785 stats.n_bytes = flow->byte_count; 786 tcp_flags = flow->tcp_flags; 787 spin_unlock_bh(&flow->lock); 788 789 if (used && 790 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used))) 791 goto nla_put_failure; 792 793 if (stats.n_packets && 794 nla_put(skb, OVS_FLOW_ATTR_STATS, 795 sizeof(struct ovs_flow_stats), &stats)) 796 goto nla_put_failure; 797 798 if (tcp_flags && 799 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags)) 800 goto nla_put_failure; 801 802 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if 803 * this is the first flow to be dumped into 'skb'. This is unusual for 804 * Netlink but individual action lists can be longer than 805 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. 806 * The userspace caller can always fetch the actions separately if it 807 * really wants them. (Most userspace callers in fact don't care.) 808 * 809 * This can only fail for dump operations because the skb is always 810 * properly sized for single flows. 811 */ 812 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len, 813 sf_acts->actions); 814 if (err < 0 && skb_orig_len) 815 goto error; 816 817 return genlmsg_end(skb, ovs_header); 818 819 nla_put_failure: 820 err = -EMSGSIZE; 821 error: 822 genlmsg_cancel(skb, ovs_header); 823 return err; 824 } 825 826 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow) 827 { 828 const struct sw_flow_actions *sf_acts; 829 int len; 830 831 sf_acts = rcu_dereference_protected(flow->sf_acts, 832 lockdep_genl_is_held()); 833 834 /* OVS_FLOW_ATTR_KEY */ 835 len = nla_total_size(FLOW_BUFSIZE); 836 /* OVS_FLOW_ATTR_ACTIONS */ 837 len += nla_total_size(sf_acts->actions_len); 838 /* OVS_FLOW_ATTR_STATS */ 839 len += nla_total_size(sizeof(struct ovs_flow_stats)); 840 /* OVS_FLOW_ATTR_TCP_FLAGS */ 841 len += nla_total_size(1); 842 /* OVS_FLOW_ATTR_USED */ 843 len += nla_total_size(8); 844 845 len += NLMSG_ALIGN(sizeof(struct ovs_header)); 846 847 return genlmsg_new(len, GFP_KERNEL); 848 } 849 850 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, 851 struct datapath *dp, 852 u32 pid, u32 seq, u8 cmd) 853 { 854 struct sk_buff *skb; 855 int retval; 856 857 skb = ovs_flow_cmd_alloc_info(flow); 858 if (!skb) 859 return ERR_PTR(-ENOMEM); 860 861 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); 862 BUG_ON(retval < 0); 863 return skb; 864 } 865 866 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) 867 { 868 struct nlattr **a = info->attrs; 869 struct ovs_header *ovs_header = info->userhdr; 870 struct sw_flow_key key; 871 struct sw_flow *flow; 872 struct sk_buff *reply; 873 struct datapath *dp; 874 struct flow_table *table; 875 int error; 876 int key_len; 877 878 /* Extract key. */ 879 error = -EINVAL; 880 if (!a[OVS_FLOW_ATTR_KEY]) 881 goto error; 882 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 883 if (error) 884 goto error; 885 886 /* Validate actions. */ 887 if (a[OVS_FLOW_ATTR_ACTIONS]) { 888 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0); 889 if (error) 890 goto error; 891 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) { 892 error = -EINVAL; 893 goto error; 894 } 895 896 dp = get_dp(ovs_header->dp_ifindex); 897 error = -ENODEV; 898 if (!dp) 899 goto error; 900 901 table = genl_dereference(dp->table); 902 flow = ovs_flow_tbl_lookup(table, &key, key_len); 903 if (!flow) { 904 struct sw_flow_actions *acts; 905 906 /* Bail out if we're not allowed to create a new flow. */ 907 error = -ENOENT; 908 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET) 909 goto error; 910 911 /* Expand table, if necessary, to make room. */ 912 if (ovs_flow_tbl_need_to_expand(table)) { 913 struct flow_table *new_table; 914 915 new_table = ovs_flow_tbl_expand(table); 916 if (!IS_ERR(new_table)) { 917 rcu_assign_pointer(dp->table, new_table); 918 ovs_flow_tbl_deferred_destroy(table); 919 table = genl_dereference(dp->table); 920 } 921 } 922 923 /* Allocate flow. */ 924 flow = ovs_flow_alloc(); 925 if (IS_ERR(flow)) { 926 error = PTR_ERR(flow); 927 goto error; 928 } 929 flow->key = key; 930 clear_stats(flow); 931 932 /* Obtain actions. */ 933 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]); 934 error = PTR_ERR(acts); 935 if (IS_ERR(acts)) 936 goto error_free_flow; 937 rcu_assign_pointer(flow->sf_acts, acts); 938 939 /* Put flow in bucket. */ 940 flow->hash = ovs_flow_hash(&key, key_len); 941 ovs_flow_tbl_insert(table, flow); 942 943 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 944 info->snd_seq, 945 OVS_FLOW_CMD_NEW); 946 } else { 947 /* We found a matching flow. */ 948 struct sw_flow_actions *old_acts; 949 struct nlattr *acts_attrs; 950 951 /* Bail out if we're not allowed to modify an existing flow. 952 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL 953 * because Generic Netlink treats the latter as a dump 954 * request. We also accept NLM_F_EXCL in case that bug ever 955 * gets fixed. 956 */ 957 error = -EEXIST; 958 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW && 959 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) 960 goto error; 961 962 /* Update actions. */ 963 old_acts = rcu_dereference_protected(flow->sf_acts, 964 lockdep_genl_is_held()); 965 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS]; 966 if (acts_attrs && 967 (old_acts->actions_len != nla_len(acts_attrs) || 968 memcmp(old_acts->actions, nla_data(acts_attrs), 969 old_acts->actions_len))) { 970 struct sw_flow_actions *new_acts; 971 972 new_acts = ovs_flow_actions_alloc(acts_attrs); 973 error = PTR_ERR(new_acts); 974 if (IS_ERR(new_acts)) 975 goto error; 976 977 rcu_assign_pointer(flow->sf_acts, new_acts); 978 ovs_flow_deferred_free_acts(old_acts); 979 } 980 981 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 982 info->snd_seq, OVS_FLOW_CMD_NEW); 983 984 /* Clear stats. */ 985 if (a[OVS_FLOW_ATTR_CLEAR]) { 986 spin_lock_bh(&flow->lock); 987 clear_stats(flow); 988 spin_unlock_bh(&flow->lock); 989 } 990 } 991 992 if (!IS_ERR(reply)) 993 genl_notify(reply, genl_info_net(info), info->snd_pid, 994 ovs_dp_flow_multicast_group.id, info->nlhdr, 995 GFP_KERNEL); 996 else 997 netlink_set_err(init_net.genl_sock, 0, 998 ovs_dp_flow_multicast_group.id, PTR_ERR(reply)); 999 return 0; 1000 1001 error_free_flow: 1002 ovs_flow_free(flow); 1003 error: 1004 return error; 1005 } 1006 1007 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) 1008 { 1009 struct nlattr **a = info->attrs; 1010 struct ovs_header *ovs_header = info->userhdr; 1011 struct sw_flow_key key; 1012 struct sk_buff *reply; 1013 struct sw_flow *flow; 1014 struct datapath *dp; 1015 struct flow_table *table; 1016 int err; 1017 int key_len; 1018 1019 if (!a[OVS_FLOW_ATTR_KEY]) 1020 return -EINVAL; 1021 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 1022 if (err) 1023 return err; 1024 1025 dp = get_dp(ovs_header->dp_ifindex); 1026 if (!dp) 1027 return -ENODEV; 1028 1029 table = genl_dereference(dp->table); 1030 flow = ovs_flow_tbl_lookup(table, &key, key_len); 1031 if (!flow) 1032 return -ENOENT; 1033 1034 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 1035 info->snd_seq, OVS_FLOW_CMD_NEW); 1036 if (IS_ERR(reply)) 1037 return PTR_ERR(reply); 1038 1039 return genlmsg_reply(reply, info); 1040 } 1041 1042 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) 1043 { 1044 struct nlattr **a = info->attrs; 1045 struct ovs_header *ovs_header = info->userhdr; 1046 struct sw_flow_key key; 1047 struct sk_buff *reply; 1048 struct sw_flow *flow; 1049 struct datapath *dp; 1050 struct flow_table *table; 1051 int err; 1052 int key_len; 1053 1054 if (!a[OVS_FLOW_ATTR_KEY]) 1055 return flush_flows(ovs_header->dp_ifindex); 1056 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 1057 if (err) 1058 return err; 1059 1060 dp = get_dp(ovs_header->dp_ifindex); 1061 if (!dp) 1062 return -ENODEV; 1063 1064 table = genl_dereference(dp->table); 1065 flow = ovs_flow_tbl_lookup(table, &key, key_len); 1066 if (!flow) 1067 return -ENOENT; 1068 1069 reply = ovs_flow_cmd_alloc_info(flow); 1070 if (!reply) 1071 return -ENOMEM; 1072 1073 ovs_flow_tbl_remove(table, flow); 1074 1075 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, 1076 info->snd_seq, 0, OVS_FLOW_CMD_DEL); 1077 BUG_ON(err < 0); 1078 1079 ovs_flow_deferred_free(flow); 1080 1081 genl_notify(reply, genl_info_net(info), info->snd_pid, 1082 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); 1083 return 0; 1084 } 1085 1086 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1087 { 1088 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1089 struct datapath *dp; 1090 struct flow_table *table; 1091 1092 dp = get_dp(ovs_header->dp_ifindex); 1093 if (!dp) 1094 return -ENODEV; 1095 1096 table = genl_dereference(dp->table); 1097 1098 for (;;) { 1099 struct sw_flow *flow; 1100 u32 bucket, obj; 1101 1102 bucket = cb->args[0]; 1103 obj = cb->args[1]; 1104 flow = ovs_flow_tbl_next(table, &bucket, &obj); 1105 if (!flow) 1106 break; 1107 1108 if (ovs_flow_cmd_fill_info(flow, dp, skb, 1109 NETLINK_CB(cb->skb).pid, 1110 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1111 OVS_FLOW_CMD_NEW) < 0) 1112 break; 1113 1114 cb->args[0] = bucket; 1115 cb->args[1] = obj; 1116 } 1117 return skb->len; 1118 } 1119 1120 static struct genl_ops dp_flow_genl_ops[] = { 1121 { .cmd = OVS_FLOW_CMD_NEW, 1122 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1123 .policy = flow_policy, 1124 .doit = ovs_flow_cmd_new_or_set 1125 }, 1126 { .cmd = OVS_FLOW_CMD_DEL, 1127 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1128 .policy = flow_policy, 1129 .doit = ovs_flow_cmd_del 1130 }, 1131 { .cmd = OVS_FLOW_CMD_GET, 1132 .flags = 0, /* OK for unprivileged users. */ 1133 .policy = flow_policy, 1134 .doit = ovs_flow_cmd_get, 1135 .dumpit = ovs_flow_cmd_dump 1136 }, 1137 { .cmd = OVS_FLOW_CMD_SET, 1138 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1139 .policy = flow_policy, 1140 .doit = ovs_flow_cmd_new_or_set, 1141 }, 1142 }; 1143 1144 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { 1145 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1146 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1147 }; 1148 1149 static struct genl_family dp_datapath_genl_family = { 1150 .id = GENL_ID_GENERATE, 1151 .hdrsize = sizeof(struct ovs_header), 1152 .name = OVS_DATAPATH_FAMILY, 1153 .version = OVS_DATAPATH_VERSION, 1154 .maxattr = OVS_DP_ATTR_MAX 1155 }; 1156 1157 static struct genl_multicast_group ovs_dp_datapath_multicast_group = { 1158 .name = OVS_DATAPATH_MCGROUP 1159 }; 1160 1161 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, 1162 u32 pid, u32 seq, u32 flags, u8 cmd) 1163 { 1164 struct ovs_header *ovs_header; 1165 struct ovs_dp_stats dp_stats; 1166 int err; 1167 1168 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, 1169 flags, cmd); 1170 if (!ovs_header) 1171 goto error; 1172 1173 ovs_header->dp_ifindex = get_dpifindex(dp); 1174 1175 rcu_read_lock(); 1176 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); 1177 rcu_read_unlock(); 1178 if (err) 1179 goto nla_put_failure; 1180 1181 get_dp_stats(dp, &dp_stats); 1182 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats)) 1183 goto nla_put_failure; 1184 1185 return genlmsg_end(skb, ovs_header); 1186 1187 nla_put_failure: 1188 genlmsg_cancel(skb, ovs_header); 1189 error: 1190 return -EMSGSIZE; 1191 } 1192 1193 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid, 1194 u32 seq, u8 cmd) 1195 { 1196 struct sk_buff *skb; 1197 int retval; 1198 1199 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1200 if (!skb) 1201 return ERR_PTR(-ENOMEM); 1202 1203 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); 1204 if (retval < 0) { 1205 kfree_skb(skb); 1206 return ERR_PTR(retval); 1207 } 1208 return skb; 1209 } 1210 1211 /* Called with genl_mutex and optionally with RTNL lock also. */ 1212 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, 1213 struct nlattr *a[OVS_DP_ATTR_MAX + 1]) 1214 { 1215 struct datapath *dp; 1216 1217 if (!a[OVS_DP_ATTR_NAME]) 1218 dp = get_dp(ovs_header->dp_ifindex); 1219 else { 1220 struct vport *vport; 1221 1222 rcu_read_lock(); 1223 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME])); 1224 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; 1225 rcu_read_unlock(); 1226 } 1227 return dp ? dp : ERR_PTR(-ENODEV); 1228 } 1229 1230 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) 1231 { 1232 struct nlattr **a = info->attrs; 1233 struct vport_parms parms; 1234 struct sk_buff *reply; 1235 struct datapath *dp; 1236 struct vport *vport; 1237 int err; 1238 1239 err = -EINVAL; 1240 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) 1241 goto err; 1242 1243 rtnl_lock(); 1244 err = -ENODEV; 1245 if (!try_module_get(THIS_MODULE)) 1246 goto err_unlock_rtnl; 1247 1248 err = -ENOMEM; 1249 dp = kzalloc(sizeof(*dp), GFP_KERNEL); 1250 if (dp == NULL) 1251 goto err_put_module; 1252 INIT_LIST_HEAD(&dp->port_list); 1253 1254 /* Allocate table. */ 1255 err = -ENOMEM; 1256 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS)); 1257 if (!dp->table) 1258 goto err_free_dp; 1259 1260 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); 1261 if (!dp->stats_percpu) { 1262 err = -ENOMEM; 1263 goto err_destroy_table; 1264 } 1265 1266 /* Set up our datapath device. */ 1267 parms.name = nla_data(a[OVS_DP_ATTR_NAME]); 1268 parms.type = OVS_VPORT_TYPE_INTERNAL; 1269 parms.options = NULL; 1270 parms.dp = dp; 1271 parms.port_no = OVSP_LOCAL; 1272 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]); 1273 1274 vport = new_vport(&parms); 1275 if (IS_ERR(vport)) { 1276 err = PTR_ERR(vport); 1277 if (err == -EBUSY) 1278 err = -EEXIST; 1279 1280 goto err_destroy_percpu; 1281 } 1282 1283 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1284 info->snd_seq, OVS_DP_CMD_NEW); 1285 err = PTR_ERR(reply); 1286 if (IS_ERR(reply)) 1287 goto err_destroy_local_port; 1288 1289 list_add_tail(&dp->list_node, &dps); 1290 rtnl_unlock(); 1291 1292 genl_notify(reply, genl_info_net(info), info->snd_pid, 1293 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1294 GFP_KERNEL); 1295 return 0; 1296 1297 err_destroy_local_port: 1298 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL])); 1299 err_destroy_percpu: 1300 free_percpu(dp->stats_percpu); 1301 err_destroy_table: 1302 ovs_flow_tbl_destroy(genl_dereference(dp->table)); 1303 err_free_dp: 1304 kfree(dp); 1305 err_put_module: 1306 module_put(THIS_MODULE); 1307 err_unlock_rtnl: 1308 rtnl_unlock(); 1309 err: 1310 return err; 1311 } 1312 1313 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) 1314 { 1315 struct vport *vport, *next_vport; 1316 struct sk_buff *reply; 1317 struct datapath *dp; 1318 int err; 1319 1320 rtnl_lock(); 1321 dp = lookup_datapath(info->userhdr, info->attrs); 1322 err = PTR_ERR(dp); 1323 if (IS_ERR(dp)) 1324 goto exit_unlock; 1325 1326 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1327 info->snd_seq, OVS_DP_CMD_DEL); 1328 err = PTR_ERR(reply); 1329 if (IS_ERR(reply)) 1330 goto exit_unlock; 1331 1332 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node) 1333 if (vport->port_no != OVSP_LOCAL) 1334 ovs_dp_detach_port(vport); 1335 1336 list_del(&dp->list_node); 1337 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL])); 1338 1339 /* rtnl_unlock() will wait until all the references to devices that 1340 * are pending unregistration have been dropped. We do it here to 1341 * ensure that any internal devices (which contain DP pointers) are 1342 * fully destroyed before freeing the datapath. 1343 */ 1344 rtnl_unlock(); 1345 1346 call_rcu(&dp->rcu, destroy_dp_rcu); 1347 module_put(THIS_MODULE); 1348 1349 genl_notify(reply, genl_info_net(info), info->snd_pid, 1350 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1351 GFP_KERNEL); 1352 1353 return 0; 1354 1355 exit_unlock: 1356 rtnl_unlock(); 1357 return err; 1358 } 1359 1360 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) 1361 { 1362 struct sk_buff *reply; 1363 struct datapath *dp; 1364 int err; 1365 1366 dp = lookup_datapath(info->userhdr, info->attrs); 1367 if (IS_ERR(dp)) 1368 return PTR_ERR(dp); 1369 1370 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1371 info->snd_seq, OVS_DP_CMD_NEW); 1372 if (IS_ERR(reply)) { 1373 err = PTR_ERR(reply); 1374 netlink_set_err(init_net.genl_sock, 0, 1375 ovs_dp_datapath_multicast_group.id, err); 1376 return 0; 1377 } 1378 1379 genl_notify(reply, genl_info_net(info), info->snd_pid, 1380 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1381 GFP_KERNEL); 1382 1383 return 0; 1384 } 1385 1386 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) 1387 { 1388 struct sk_buff *reply; 1389 struct datapath *dp; 1390 1391 dp = lookup_datapath(info->userhdr, info->attrs); 1392 if (IS_ERR(dp)) 1393 return PTR_ERR(dp); 1394 1395 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1396 info->snd_seq, OVS_DP_CMD_NEW); 1397 if (IS_ERR(reply)) 1398 return PTR_ERR(reply); 1399 1400 return genlmsg_reply(reply, info); 1401 } 1402 1403 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1404 { 1405 struct datapath *dp; 1406 int skip = cb->args[0]; 1407 int i = 0; 1408 1409 list_for_each_entry(dp, &dps, list_node) { 1410 if (i >= skip && 1411 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid, 1412 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1413 OVS_DP_CMD_NEW) < 0) 1414 break; 1415 i++; 1416 } 1417 1418 cb->args[0] = i; 1419 1420 return skb->len; 1421 } 1422 1423 static struct genl_ops dp_datapath_genl_ops[] = { 1424 { .cmd = OVS_DP_CMD_NEW, 1425 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1426 .policy = datapath_policy, 1427 .doit = ovs_dp_cmd_new 1428 }, 1429 { .cmd = OVS_DP_CMD_DEL, 1430 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1431 .policy = datapath_policy, 1432 .doit = ovs_dp_cmd_del 1433 }, 1434 { .cmd = OVS_DP_CMD_GET, 1435 .flags = 0, /* OK for unprivileged users. */ 1436 .policy = datapath_policy, 1437 .doit = ovs_dp_cmd_get, 1438 .dumpit = ovs_dp_cmd_dump 1439 }, 1440 { .cmd = OVS_DP_CMD_SET, 1441 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1442 .policy = datapath_policy, 1443 .doit = ovs_dp_cmd_set, 1444 }, 1445 }; 1446 1447 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { 1448 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1449 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, 1450 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, 1451 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, 1452 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1453 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, 1454 }; 1455 1456 static struct genl_family dp_vport_genl_family = { 1457 .id = GENL_ID_GENERATE, 1458 .hdrsize = sizeof(struct ovs_header), 1459 .name = OVS_VPORT_FAMILY, 1460 .version = OVS_VPORT_VERSION, 1461 .maxattr = OVS_VPORT_ATTR_MAX 1462 }; 1463 1464 struct genl_multicast_group ovs_dp_vport_multicast_group = { 1465 .name = OVS_VPORT_MCGROUP 1466 }; 1467 1468 /* Called with RTNL lock or RCU read lock. */ 1469 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, 1470 u32 pid, u32 seq, u32 flags, u8 cmd) 1471 { 1472 struct ovs_header *ovs_header; 1473 struct ovs_vport_stats vport_stats; 1474 int err; 1475 1476 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, 1477 flags, cmd); 1478 if (!ovs_header) 1479 return -EMSGSIZE; 1480 1481 ovs_header->dp_ifindex = get_dpifindex(vport->dp); 1482 1483 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || 1484 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || 1485 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) || 1486 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid)) 1487 goto nla_put_failure; 1488 1489 ovs_vport_get_stats(vport, &vport_stats); 1490 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats), 1491 &vport_stats)) 1492 goto nla_put_failure; 1493 1494 err = ovs_vport_get_options(vport, skb); 1495 if (err == -EMSGSIZE) 1496 goto error; 1497 1498 return genlmsg_end(skb, ovs_header); 1499 1500 nla_put_failure: 1501 err = -EMSGSIZE; 1502 error: 1503 genlmsg_cancel(skb, ovs_header); 1504 return err; 1505 } 1506 1507 /* Called with RTNL lock or RCU read lock. */ 1508 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid, 1509 u32 seq, u8 cmd) 1510 { 1511 struct sk_buff *skb; 1512 int retval; 1513 1514 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 1515 if (!skb) 1516 return ERR_PTR(-ENOMEM); 1517 1518 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd); 1519 if (retval < 0) { 1520 kfree_skb(skb); 1521 return ERR_PTR(retval); 1522 } 1523 return skb; 1524 } 1525 1526 /* Called with RTNL lock or RCU read lock. */ 1527 static struct vport *lookup_vport(struct ovs_header *ovs_header, 1528 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) 1529 { 1530 struct datapath *dp; 1531 struct vport *vport; 1532 1533 if (a[OVS_VPORT_ATTR_NAME]) { 1534 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME])); 1535 if (!vport) 1536 return ERR_PTR(-ENODEV); 1537 if (ovs_header->dp_ifindex && 1538 ovs_header->dp_ifindex != get_dpifindex(vport->dp)) 1539 return ERR_PTR(-ENODEV); 1540 return vport; 1541 } else if (a[OVS_VPORT_ATTR_PORT_NO]) { 1542 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1543 1544 if (port_no >= DP_MAX_PORTS) 1545 return ERR_PTR(-EFBIG); 1546 1547 dp = get_dp(ovs_header->dp_ifindex); 1548 if (!dp) 1549 return ERR_PTR(-ENODEV); 1550 1551 vport = rcu_dereference_rtnl(dp->ports[port_no]); 1552 if (!vport) 1553 return ERR_PTR(-ENOENT); 1554 return vport; 1555 } else 1556 return ERR_PTR(-EINVAL); 1557 } 1558 1559 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) 1560 { 1561 struct nlattr **a = info->attrs; 1562 struct ovs_header *ovs_header = info->userhdr; 1563 struct vport_parms parms; 1564 struct sk_buff *reply; 1565 struct vport *vport; 1566 struct datapath *dp; 1567 u32 port_no; 1568 int err; 1569 1570 err = -EINVAL; 1571 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || 1572 !a[OVS_VPORT_ATTR_UPCALL_PID]) 1573 goto exit; 1574 1575 rtnl_lock(); 1576 dp = get_dp(ovs_header->dp_ifindex); 1577 err = -ENODEV; 1578 if (!dp) 1579 goto exit_unlock; 1580 1581 if (a[OVS_VPORT_ATTR_PORT_NO]) { 1582 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1583 1584 err = -EFBIG; 1585 if (port_no >= DP_MAX_PORTS) 1586 goto exit_unlock; 1587 1588 vport = rtnl_dereference(dp->ports[port_no]); 1589 err = -EBUSY; 1590 if (vport) 1591 goto exit_unlock; 1592 } else { 1593 for (port_no = 1; ; port_no++) { 1594 if (port_no >= DP_MAX_PORTS) { 1595 err = -EFBIG; 1596 goto exit_unlock; 1597 } 1598 vport = rtnl_dereference(dp->ports[port_no]); 1599 if (!vport) 1600 break; 1601 } 1602 } 1603 1604 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); 1605 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); 1606 parms.options = a[OVS_VPORT_ATTR_OPTIONS]; 1607 parms.dp = dp; 1608 parms.port_no = port_no; 1609 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); 1610 1611 vport = new_vport(&parms); 1612 err = PTR_ERR(vport); 1613 if (IS_ERR(vport)) 1614 goto exit_unlock; 1615 1616 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1617 OVS_VPORT_CMD_NEW); 1618 if (IS_ERR(reply)) { 1619 err = PTR_ERR(reply); 1620 ovs_dp_detach_port(vport); 1621 goto exit_unlock; 1622 } 1623 genl_notify(reply, genl_info_net(info), info->snd_pid, 1624 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1625 1626 exit_unlock: 1627 rtnl_unlock(); 1628 exit: 1629 return err; 1630 } 1631 1632 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) 1633 { 1634 struct nlattr **a = info->attrs; 1635 struct sk_buff *reply; 1636 struct vport *vport; 1637 int err; 1638 1639 rtnl_lock(); 1640 vport = lookup_vport(info->userhdr, a); 1641 err = PTR_ERR(vport); 1642 if (IS_ERR(vport)) 1643 goto exit_unlock; 1644 1645 err = 0; 1646 if (a[OVS_VPORT_ATTR_TYPE] && 1647 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) 1648 err = -EINVAL; 1649 1650 if (!err && a[OVS_VPORT_ATTR_OPTIONS]) 1651 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); 1652 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID]) 1653 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); 1654 1655 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1656 OVS_VPORT_CMD_NEW); 1657 if (IS_ERR(reply)) { 1658 netlink_set_err(init_net.genl_sock, 0, 1659 ovs_dp_vport_multicast_group.id, PTR_ERR(reply)); 1660 goto exit_unlock; 1661 } 1662 1663 genl_notify(reply, genl_info_net(info), info->snd_pid, 1664 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1665 1666 exit_unlock: 1667 rtnl_unlock(); 1668 return err; 1669 } 1670 1671 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) 1672 { 1673 struct nlattr **a = info->attrs; 1674 struct sk_buff *reply; 1675 struct vport *vport; 1676 int err; 1677 1678 rtnl_lock(); 1679 vport = lookup_vport(info->userhdr, a); 1680 err = PTR_ERR(vport); 1681 if (IS_ERR(vport)) 1682 goto exit_unlock; 1683 1684 if (vport->port_no == OVSP_LOCAL) { 1685 err = -EINVAL; 1686 goto exit_unlock; 1687 } 1688 1689 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1690 OVS_VPORT_CMD_DEL); 1691 err = PTR_ERR(reply); 1692 if (IS_ERR(reply)) 1693 goto exit_unlock; 1694 1695 ovs_dp_detach_port(vport); 1696 1697 genl_notify(reply, genl_info_net(info), info->snd_pid, 1698 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1699 1700 exit_unlock: 1701 rtnl_unlock(); 1702 return err; 1703 } 1704 1705 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) 1706 { 1707 struct nlattr **a = info->attrs; 1708 struct ovs_header *ovs_header = info->userhdr; 1709 struct sk_buff *reply; 1710 struct vport *vport; 1711 int err; 1712 1713 rcu_read_lock(); 1714 vport = lookup_vport(ovs_header, a); 1715 err = PTR_ERR(vport); 1716 if (IS_ERR(vport)) 1717 goto exit_unlock; 1718 1719 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1720 OVS_VPORT_CMD_NEW); 1721 err = PTR_ERR(reply); 1722 if (IS_ERR(reply)) 1723 goto exit_unlock; 1724 1725 rcu_read_unlock(); 1726 1727 return genlmsg_reply(reply, info); 1728 1729 exit_unlock: 1730 rcu_read_unlock(); 1731 return err; 1732 } 1733 1734 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1735 { 1736 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1737 struct datapath *dp; 1738 u32 port_no; 1739 int retval; 1740 1741 dp = get_dp(ovs_header->dp_ifindex); 1742 if (!dp) 1743 return -ENODEV; 1744 1745 rcu_read_lock(); 1746 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) { 1747 struct vport *vport; 1748 1749 vport = rcu_dereference(dp->ports[port_no]); 1750 if (!vport) 1751 continue; 1752 1753 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid, 1754 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1755 OVS_VPORT_CMD_NEW) < 0) 1756 break; 1757 } 1758 rcu_read_unlock(); 1759 1760 cb->args[0] = port_no; 1761 retval = skb->len; 1762 1763 return retval; 1764 } 1765 1766 static void rehash_flow_table(struct work_struct *work) 1767 { 1768 struct datapath *dp; 1769 1770 genl_lock(); 1771 1772 list_for_each_entry(dp, &dps, list_node) { 1773 struct flow_table *old_table = genl_dereference(dp->table); 1774 struct flow_table *new_table; 1775 1776 new_table = ovs_flow_tbl_rehash(old_table); 1777 if (!IS_ERR(new_table)) { 1778 rcu_assign_pointer(dp->table, new_table); 1779 ovs_flow_tbl_deferred_destroy(old_table); 1780 } 1781 } 1782 1783 genl_unlock(); 1784 1785 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); 1786 } 1787 1788 static struct genl_ops dp_vport_genl_ops[] = { 1789 { .cmd = OVS_VPORT_CMD_NEW, 1790 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1791 .policy = vport_policy, 1792 .doit = ovs_vport_cmd_new 1793 }, 1794 { .cmd = OVS_VPORT_CMD_DEL, 1795 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1796 .policy = vport_policy, 1797 .doit = ovs_vport_cmd_del 1798 }, 1799 { .cmd = OVS_VPORT_CMD_GET, 1800 .flags = 0, /* OK for unprivileged users. */ 1801 .policy = vport_policy, 1802 .doit = ovs_vport_cmd_get, 1803 .dumpit = ovs_vport_cmd_dump 1804 }, 1805 { .cmd = OVS_VPORT_CMD_SET, 1806 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1807 .policy = vport_policy, 1808 .doit = ovs_vport_cmd_set, 1809 }, 1810 }; 1811 1812 struct genl_family_and_ops { 1813 struct genl_family *family; 1814 struct genl_ops *ops; 1815 int n_ops; 1816 struct genl_multicast_group *group; 1817 }; 1818 1819 static const struct genl_family_and_ops dp_genl_families[] = { 1820 { &dp_datapath_genl_family, 1821 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops), 1822 &ovs_dp_datapath_multicast_group }, 1823 { &dp_vport_genl_family, 1824 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops), 1825 &ovs_dp_vport_multicast_group }, 1826 { &dp_flow_genl_family, 1827 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops), 1828 &ovs_dp_flow_multicast_group }, 1829 { &dp_packet_genl_family, 1830 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), 1831 NULL }, 1832 }; 1833 1834 static void dp_unregister_genl(int n_families) 1835 { 1836 int i; 1837 1838 for (i = 0; i < n_families; i++) 1839 genl_unregister_family(dp_genl_families[i].family); 1840 } 1841 1842 static int dp_register_genl(void) 1843 { 1844 int n_registered; 1845 int err; 1846 int i; 1847 1848 n_registered = 0; 1849 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { 1850 const struct genl_family_and_ops *f = &dp_genl_families[i]; 1851 1852 err = genl_register_family_with_ops(f->family, f->ops, 1853 f->n_ops); 1854 if (err) 1855 goto error; 1856 n_registered++; 1857 1858 if (f->group) { 1859 err = genl_register_mc_group(f->family, f->group); 1860 if (err) 1861 goto error; 1862 } 1863 } 1864 1865 return 0; 1866 1867 error: 1868 dp_unregister_genl(n_registered); 1869 return err; 1870 } 1871 1872 static int __init dp_init(void) 1873 { 1874 struct sk_buff *dummy_skb; 1875 int err; 1876 1877 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb)); 1878 1879 pr_info("Open vSwitch switching datapath\n"); 1880 1881 err = ovs_flow_init(); 1882 if (err) 1883 goto error; 1884 1885 err = ovs_vport_init(); 1886 if (err) 1887 goto error_flow_exit; 1888 1889 err = register_netdevice_notifier(&ovs_dp_device_notifier); 1890 if (err) 1891 goto error_vport_exit; 1892 1893 err = dp_register_genl(); 1894 if (err < 0) 1895 goto error_unreg_notifier; 1896 1897 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); 1898 1899 return 0; 1900 1901 error_unreg_notifier: 1902 unregister_netdevice_notifier(&ovs_dp_device_notifier); 1903 error_vport_exit: 1904 ovs_vport_exit(); 1905 error_flow_exit: 1906 ovs_flow_exit(); 1907 error: 1908 return err; 1909 } 1910 1911 static void dp_cleanup(void) 1912 { 1913 cancel_delayed_work_sync(&rehash_flow_wq); 1914 rcu_barrier(); 1915 dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); 1916 unregister_netdevice_notifier(&ovs_dp_device_notifier); 1917 ovs_vport_exit(); 1918 ovs_flow_exit(); 1919 } 1920 1921 module_init(dp_init); 1922 module_exit(dp_cleanup); 1923 1924 MODULE_DESCRIPTION("Open vSwitch switching datapath"); 1925 MODULE_LICENSE("GPL"); 1926