datapath.c revision 8aa51d64c1f526e43b1e7f89fb8b98c2fd583f4b
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 */ 77static LIST_HEAD(dps); 78 79#define REHASH_FLOW_INTERVAL (10 * 60 * HZ) 80static void rehash_flow_table(struct work_struct *work); 81static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table); 82 83static struct vport *new_vport(const struct vport_parms *); 84static int queue_gso_packets(int dp_ifindex, struct sk_buff *, 85 const struct dp_upcall_info *); 86static 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. */ 90static 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. */ 108const 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 114static 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 132static 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. */ 142static 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. */ 158void 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. */ 171void 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 211out: 212 /* Update datapath statistics. */ 213 u64_stats_update_begin(&stats->sync); 214 (*stats_counter)++; 215 u64_stats_update_end(&stats->sync); 216} 217 218static 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 226int 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 253err: 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 263static 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 308static 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 366out: 367 kfree_skb(nskb); 368 return err; 369} 370 371/* Called with genl_mutex. */ 372static 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 393static int validate_actions(const struct nlattr *attr, 394 const struct sw_flow_key *key, int depth); 395 396static 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 424static 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 437static 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 493static 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 514static 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 591static 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 599static 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 683err_unlock: 684 rcu_read_unlock(); 685err_flow_free: 686 ovs_flow_free(flow); 687err_kfree_skb: 688 kfree_skb(packet); 689err: 690 return err; 691} 692 693static 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 699static 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 707static 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 733static 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 739static 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 747static struct genl_multicast_group ovs_dp_flow_multicast_group = { 748 .name = OVS_FLOW_MCGROUP 749}; 750 751/* Called with genl_lock. */ 752static 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)); 791 792 if (stats.n_packets) 793 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, 794 sizeof(struct ovs_flow_stats), &stats); 795 796 if (tcp_flags) 797 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags); 798 799 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if 800 * this is the first flow to be dumped into 'skb'. This is unusual for 801 * Netlink but individual action lists can be longer than 802 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. 803 * The userspace caller can always fetch the actions separately if it 804 * really wants them. (Most userspace callers in fact don't care.) 805 * 806 * This can only fail for dump operations because the skb is always 807 * properly sized for single flows. 808 */ 809 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len, 810 sf_acts->actions); 811 if (err < 0 && skb_orig_len) 812 goto error; 813 814 return genlmsg_end(skb, ovs_header); 815 816nla_put_failure: 817 err = -EMSGSIZE; 818error: 819 genlmsg_cancel(skb, ovs_header); 820 return err; 821} 822 823static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow) 824{ 825 const struct sw_flow_actions *sf_acts; 826 int len; 827 828 sf_acts = rcu_dereference_protected(flow->sf_acts, 829 lockdep_genl_is_held()); 830 831 /* OVS_FLOW_ATTR_KEY */ 832 len = nla_total_size(FLOW_BUFSIZE); 833 /* OVS_FLOW_ATTR_ACTIONS */ 834 len += nla_total_size(sf_acts->actions_len); 835 /* OVS_FLOW_ATTR_STATS */ 836 len += nla_total_size(sizeof(struct ovs_flow_stats)); 837 /* OVS_FLOW_ATTR_TCP_FLAGS */ 838 len += nla_total_size(1); 839 /* OVS_FLOW_ATTR_USED */ 840 len += nla_total_size(8); 841 842 len += NLMSG_ALIGN(sizeof(struct ovs_header)); 843 844 return genlmsg_new(len, GFP_KERNEL); 845} 846 847static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, 848 struct datapath *dp, 849 u32 pid, u32 seq, u8 cmd) 850{ 851 struct sk_buff *skb; 852 int retval; 853 854 skb = ovs_flow_cmd_alloc_info(flow); 855 if (!skb) 856 return ERR_PTR(-ENOMEM); 857 858 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); 859 BUG_ON(retval < 0); 860 return skb; 861} 862 863static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) 864{ 865 struct nlattr **a = info->attrs; 866 struct ovs_header *ovs_header = info->userhdr; 867 struct sw_flow_key key; 868 struct sw_flow *flow; 869 struct sk_buff *reply; 870 struct datapath *dp; 871 struct flow_table *table; 872 int error; 873 int key_len; 874 875 /* Extract key. */ 876 error = -EINVAL; 877 if (!a[OVS_FLOW_ATTR_KEY]) 878 goto error; 879 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 880 if (error) 881 goto error; 882 883 /* Validate actions. */ 884 if (a[OVS_FLOW_ATTR_ACTIONS]) { 885 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0); 886 if (error) 887 goto error; 888 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) { 889 error = -EINVAL; 890 goto error; 891 } 892 893 dp = get_dp(ovs_header->dp_ifindex); 894 error = -ENODEV; 895 if (!dp) 896 goto error; 897 898 table = genl_dereference(dp->table); 899 flow = ovs_flow_tbl_lookup(table, &key, key_len); 900 if (!flow) { 901 struct sw_flow_actions *acts; 902 903 /* Bail out if we're not allowed to create a new flow. */ 904 error = -ENOENT; 905 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET) 906 goto error; 907 908 /* Expand table, if necessary, to make room. */ 909 if (ovs_flow_tbl_need_to_expand(table)) { 910 struct flow_table *new_table; 911 912 new_table = ovs_flow_tbl_expand(table); 913 if (!IS_ERR(new_table)) { 914 rcu_assign_pointer(dp->table, new_table); 915 ovs_flow_tbl_deferred_destroy(table); 916 table = genl_dereference(dp->table); 917 } 918 } 919 920 /* Allocate flow. */ 921 flow = ovs_flow_alloc(); 922 if (IS_ERR(flow)) { 923 error = PTR_ERR(flow); 924 goto error; 925 } 926 flow->key = key; 927 clear_stats(flow); 928 929 /* Obtain actions. */ 930 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]); 931 error = PTR_ERR(acts); 932 if (IS_ERR(acts)) 933 goto error_free_flow; 934 rcu_assign_pointer(flow->sf_acts, acts); 935 936 /* Put flow in bucket. */ 937 flow->hash = ovs_flow_hash(&key, key_len); 938 ovs_flow_tbl_insert(table, flow); 939 940 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 941 info->snd_seq, 942 OVS_FLOW_CMD_NEW); 943 } else { 944 /* We found a matching flow. */ 945 struct sw_flow_actions *old_acts; 946 struct nlattr *acts_attrs; 947 948 /* Bail out if we're not allowed to modify an existing flow. 949 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL 950 * because Generic Netlink treats the latter as a dump 951 * request. We also accept NLM_F_EXCL in case that bug ever 952 * gets fixed. 953 */ 954 error = -EEXIST; 955 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW && 956 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) 957 goto error; 958 959 /* Update actions. */ 960 old_acts = rcu_dereference_protected(flow->sf_acts, 961 lockdep_genl_is_held()); 962 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS]; 963 if (acts_attrs && 964 (old_acts->actions_len != nla_len(acts_attrs) || 965 memcmp(old_acts->actions, nla_data(acts_attrs), 966 old_acts->actions_len))) { 967 struct sw_flow_actions *new_acts; 968 969 new_acts = ovs_flow_actions_alloc(acts_attrs); 970 error = PTR_ERR(new_acts); 971 if (IS_ERR(new_acts)) 972 goto error; 973 974 rcu_assign_pointer(flow->sf_acts, new_acts); 975 ovs_flow_deferred_free_acts(old_acts); 976 } 977 978 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 979 info->snd_seq, OVS_FLOW_CMD_NEW); 980 981 /* Clear stats. */ 982 if (a[OVS_FLOW_ATTR_CLEAR]) { 983 spin_lock_bh(&flow->lock); 984 clear_stats(flow); 985 spin_unlock_bh(&flow->lock); 986 } 987 } 988 989 if (!IS_ERR(reply)) 990 genl_notify(reply, genl_info_net(info), info->snd_pid, 991 ovs_dp_flow_multicast_group.id, info->nlhdr, 992 GFP_KERNEL); 993 else 994 netlink_set_err(init_net.genl_sock, 0, 995 ovs_dp_flow_multicast_group.id, PTR_ERR(reply)); 996 return 0; 997 998error_free_flow: 999 ovs_flow_free(flow); 1000error: 1001 return error; 1002} 1003 1004static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) 1005{ 1006 struct nlattr **a = info->attrs; 1007 struct ovs_header *ovs_header = info->userhdr; 1008 struct sw_flow_key key; 1009 struct sk_buff *reply; 1010 struct sw_flow *flow; 1011 struct datapath *dp; 1012 struct flow_table *table; 1013 int err; 1014 int key_len; 1015 1016 if (!a[OVS_FLOW_ATTR_KEY]) 1017 return -EINVAL; 1018 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 1019 if (err) 1020 return err; 1021 1022 dp = get_dp(ovs_header->dp_ifindex); 1023 if (!dp) 1024 return -ENODEV; 1025 1026 table = genl_dereference(dp->table); 1027 flow = ovs_flow_tbl_lookup(table, &key, key_len); 1028 if (!flow) 1029 return -ENOENT; 1030 1031 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, 1032 info->snd_seq, OVS_FLOW_CMD_NEW); 1033 if (IS_ERR(reply)) 1034 return PTR_ERR(reply); 1035 1036 return genlmsg_reply(reply, info); 1037} 1038 1039static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) 1040{ 1041 struct nlattr **a = info->attrs; 1042 struct ovs_header *ovs_header = info->userhdr; 1043 struct sw_flow_key key; 1044 struct sk_buff *reply; 1045 struct sw_flow *flow; 1046 struct datapath *dp; 1047 struct flow_table *table; 1048 int err; 1049 int key_len; 1050 1051 if (!a[OVS_FLOW_ATTR_KEY]) 1052 return flush_flows(ovs_header->dp_ifindex); 1053 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); 1054 if (err) 1055 return err; 1056 1057 dp = get_dp(ovs_header->dp_ifindex); 1058 if (!dp) 1059 return -ENODEV; 1060 1061 table = genl_dereference(dp->table); 1062 flow = ovs_flow_tbl_lookup(table, &key, key_len); 1063 if (!flow) 1064 return -ENOENT; 1065 1066 reply = ovs_flow_cmd_alloc_info(flow); 1067 if (!reply) 1068 return -ENOMEM; 1069 1070 ovs_flow_tbl_remove(table, flow); 1071 1072 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, 1073 info->snd_seq, 0, OVS_FLOW_CMD_DEL); 1074 BUG_ON(err < 0); 1075 1076 ovs_flow_deferred_free(flow); 1077 1078 genl_notify(reply, genl_info_net(info), info->snd_pid, 1079 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); 1080 return 0; 1081} 1082 1083static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1084{ 1085 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1086 struct datapath *dp; 1087 struct flow_table *table; 1088 1089 dp = get_dp(ovs_header->dp_ifindex); 1090 if (!dp) 1091 return -ENODEV; 1092 1093 table = genl_dereference(dp->table); 1094 1095 for (;;) { 1096 struct sw_flow *flow; 1097 u32 bucket, obj; 1098 1099 bucket = cb->args[0]; 1100 obj = cb->args[1]; 1101 flow = ovs_flow_tbl_next(table, &bucket, &obj); 1102 if (!flow) 1103 break; 1104 1105 if (ovs_flow_cmd_fill_info(flow, dp, skb, 1106 NETLINK_CB(cb->skb).pid, 1107 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1108 OVS_FLOW_CMD_NEW) < 0) 1109 break; 1110 1111 cb->args[0] = bucket; 1112 cb->args[1] = obj; 1113 } 1114 return skb->len; 1115} 1116 1117static struct genl_ops dp_flow_genl_ops[] = { 1118 { .cmd = OVS_FLOW_CMD_NEW, 1119 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1120 .policy = flow_policy, 1121 .doit = ovs_flow_cmd_new_or_set 1122 }, 1123 { .cmd = OVS_FLOW_CMD_DEL, 1124 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1125 .policy = flow_policy, 1126 .doit = ovs_flow_cmd_del 1127 }, 1128 { .cmd = OVS_FLOW_CMD_GET, 1129 .flags = 0, /* OK for unprivileged users. */ 1130 .policy = flow_policy, 1131 .doit = ovs_flow_cmd_get, 1132 .dumpit = ovs_flow_cmd_dump 1133 }, 1134 { .cmd = OVS_FLOW_CMD_SET, 1135 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1136 .policy = flow_policy, 1137 .doit = ovs_flow_cmd_new_or_set, 1138 }, 1139}; 1140 1141static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { 1142 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1143 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1144}; 1145 1146static struct genl_family dp_datapath_genl_family = { 1147 .id = GENL_ID_GENERATE, 1148 .hdrsize = sizeof(struct ovs_header), 1149 .name = OVS_DATAPATH_FAMILY, 1150 .version = OVS_DATAPATH_VERSION, 1151 .maxattr = OVS_DP_ATTR_MAX 1152}; 1153 1154static struct genl_multicast_group ovs_dp_datapath_multicast_group = { 1155 .name = OVS_DATAPATH_MCGROUP 1156}; 1157 1158static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, 1159 u32 pid, u32 seq, u32 flags, u8 cmd) 1160{ 1161 struct ovs_header *ovs_header; 1162 struct ovs_dp_stats dp_stats; 1163 int err; 1164 1165 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, 1166 flags, cmd); 1167 if (!ovs_header) 1168 goto error; 1169 1170 ovs_header->dp_ifindex = get_dpifindex(dp); 1171 1172 rcu_read_lock(); 1173 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); 1174 rcu_read_unlock(); 1175 if (err) 1176 goto nla_put_failure; 1177 1178 get_dp_stats(dp, &dp_stats); 1179 NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats); 1180 1181 return genlmsg_end(skb, ovs_header); 1182 1183nla_put_failure: 1184 genlmsg_cancel(skb, ovs_header); 1185error: 1186 return -EMSGSIZE; 1187} 1188 1189static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid, 1190 u32 seq, u8 cmd) 1191{ 1192 struct sk_buff *skb; 1193 int retval; 1194 1195 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1196 if (!skb) 1197 return ERR_PTR(-ENOMEM); 1198 1199 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); 1200 if (retval < 0) { 1201 kfree_skb(skb); 1202 return ERR_PTR(retval); 1203 } 1204 return skb; 1205} 1206 1207/* Called with genl_mutex and optionally with RTNL lock also. */ 1208static struct datapath *lookup_datapath(struct ovs_header *ovs_header, 1209 struct nlattr *a[OVS_DP_ATTR_MAX + 1]) 1210{ 1211 struct datapath *dp; 1212 1213 if (!a[OVS_DP_ATTR_NAME]) 1214 dp = get_dp(ovs_header->dp_ifindex); 1215 else { 1216 struct vport *vport; 1217 1218 rcu_read_lock(); 1219 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME])); 1220 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; 1221 rcu_read_unlock(); 1222 } 1223 return dp ? dp : ERR_PTR(-ENODEV); 1224} 1225 1226static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) 1227{ 1228 struct nlattr **a = info->attrs; 1229 struct vport_parms parms; 1230 struct sk_buff *reply; 1231 struct datapath *dp; 1232 struct vport *vport; 1233 int err; 1234 1235 err = -EINVAL; 1236 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) 1237 goto err; 1238 1239 rtnl_lock(); 1240 err = -ENODEV; 1241 if (!try_module_get(THIS_MODULE)) 1242 goto err_unlock_rtnl; 1243 1244 err = -ENOMEM; 1245 dp = kzalloc(sizeof(*dp), GFP_KERNEL); 1246 if (dp == NULL) 1247 goto err_put_module; 1248 INIT_LIST_HEAD(&dp->port_list); 1249 1250 /* Allocate table. */ 1251 err = -ENOMEM; 1252 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS)); 1253 if (!dp->table) 1254 goto err_free_dp; 1255 1256 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); 1257 if (!dp->stats_percpu) { 1258 err = -ENOMEM; 1259 goto err_destroy_table; 1260 } 1261 1262 /* Set up our datapath device. */ 1263 parms.name = nla_data(a[OVS_DP_ATTR_NAME]); 1264 parms.type = OVS_VPORT_TYPE_INTERNAL; 1265 parms.options = NULL; 1266 parms.dp = dp; 1267 parms.port_no = OVSP_LOCAL; 1268 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]); 1269 1270 vport = new_vport(&parms); 1271 if (IS_ERR(vport)) { 1272 err = PTR_ERR(vport); 1273 if (err == -EBUSY) 1274 err = -EEXIST; 1275 1276 goto err_destroy_percpu; 1277 } 1278 1279 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1280 info->snd_seq, OVS_DP_CMD_NEW); 1281 err = PTR_ERR(reply); 1282 if (IS_ERR(reply)) 1283 goto err_destroy_local_port; 1284 1285 list_add_tail(&dp->list_node, &dps); 1286 rtnl_unlock(); 1287 1288 genl_notify(reply, genl_info_net(info), info->snd_pid, 1289 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1290 GFP_KERNEL); 1291 return 0; 1292 1293err_destroy_local_port: 1294 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL])); 1295err_destroy_percpu: 1296 free_percpu(dp->stats_percpu); 1297err_destroy_table: 1298 ovs_flow_tbl_destroy(genl_dereference(dp->table)); 1299err_free_dp: 1300 kfree(dp); 1301err_put_module: 1302 module_put(THIS_MODULE); 1303err_unlock_rtnl: 1304 rtnl_unlock(); 1305err: 1306 return err; 1307} 1308 1309static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) 1310{ 1311 struct vport *vport, *next_vport; 1312 struct sk_buff *reply; 1313 struct datapath *dp; 1314 int err; 1315 1316 rtnl_lock(); 1317 dp = lookup_datapath(info->userhdr, info->attrs); 1318 err = PTR_ERR(dp); 1319 if (IS_ERR(dp)) 1320 goto exit_unlock; 1321 1322 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1323 info->snd_seq, OVS_DP_CMD_DEL); 1324 err = PTR_ERR(reply); 1325 if (IS_ERR(reply)) 1326 goto exit_unlock; 1327 1328 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node) 1329 if (vport->port_no != OVSP_LOCAL) 1330 ovs_dp_detach_port(vport); 1331 1332 list_del(&dp->list_node); 1333 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL])); 1334 1335 /* rtnl_unlock() will wait until all the references to devices that 1336 * are pending unregistration have been dropped. We do it here to 1337 * ensure that any internal devices (which contain DP pointers) are 1338 * fully destroyed before freeing the datapath. 1339 */ 1340 rtnl_unlock(); 1341 1342 call_rcu(&dp->rcu, destroy_dp_rcu); 1343 module_put(THIS_MODULE); 1344 1345 genl_notify(reply, genl_info_net(info), info->snd_pid, 1346 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1347 GFP_KERNEL); 1348 1349 return 0; 1350 1351exit_unlock: 1352 rtnl_unlock(); 1353 return err; 1354} 1355 1356static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) 1357{ 1358 struct sk_buff *reply; 1359 struct datapath *dp; 1360 int err; 1361 1362 dp = lookup_datapath(info->userhdr, info->attrs); 1363 if (IS_ERR(dp)) 1364 return PTR_ERR(dp); 1365 1366 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1367 info->snd_seq, OVS_DP_CMD_NEW); 1368 if (IS_ERR(reply)) { 1369 err = PTR_ERR(reply); 1370 netlink_set_err(init_net.genl_sock, 0, 1371 ovs_dp_datapath_multicast_group.id, err); 1372 return 0; 1373 } 1374 1375 genl_notify(reply, genl_info_net(info), info->snd_pid, 1376 ovs_dp_datapath_multicast_group.id, info->nlhdr, 1377 GFP_KERNEL); 1378 1379 return 0; 1380} 1381 1382static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) 1383{ 1384 struct sk_buff *reply; 1385 struct datapath *dp; 1386 1387 dp = lookup_datapath(info->userhdr, info->attrs); 1388 if (IS_ERR(dp)) 1389 return PTR_ERR(dp); 1390 1391 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, 1392 info->snd_seq, OVS_DP_CMD_NEW); 1393 if (IS_ERR(reply)) 1394 return PTR_ERR(reply); 1395 1396 return genlmsg_reply(reply, info); 1397} 1398 1399static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1400{ 1401 struct datapath *dp; 1402 int skip = cb->args[0]; 1403 int i = 0; 1404 1405 list_for_each_entry(dp, &dps, list_node) { 1406 if (i >= skip && 1407 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid, 1408 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1409 OVS_DP_CMD_NEW) < 0) 1410 break; 1411 i++; 1412 } 1413 1414 cb->args[0] = i; 1415 1416 return skb->len; 1417} 1418 1419static struct genl_ops dp_datapath_genl_ops[] = { 1420 { .cmd = OVS_DP_CMD_NEW, 1421 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1422 .policy = datapath_policy, 1423 .doit = ovs_dp_cmd_new 1424 }, 1425 { .cmd = OVS_DP_CMD_DEL, 1426 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1427 .policy = datapath_policy, 1428 .doit = ovs_dp_cmd_del 1429 }, 1430 { .cmd = OVS_DP_CMD_GET, 1431 .flags = 0, /* OK for unprivileged users. */ 1432 .policy = datapath_policy, 1433 .doit = ovs_dp_cmd_get, 1434 .dumpit = ovs_dp_cmd_dump 1435 }, 1436 { .cmd = OVS_DP_CMD_SET, 1437 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1438 .policy = datapath_policy, 1439 .doit = ovs_dp_cmd_set, 1440 }, 1441}; 1442 1443static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { 1444 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1445 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, 1446 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, 1447 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, 1448 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1449 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, 1450}; 1451 1452static struct genl_family dp_vport_genl_family = { 1453 .id = GENL_ID_GENERATE, 1454 .hdrsize = sizeof(struct ovs_header), 1455 .name = OVS_VPORT_FAMILY, 1456 .version = OVS_VPORT_VERSION, 1457 .maxattr = OVS_VPORT_ATTR_MAX 1458}; 1459 1460struct genl_multicast_group ovs_dp_vport_multicast_group = { 1461 .name = OVS_VPORT_MCGROUP 1462}; 1463 1464/* Called with RTNL lock or RCU read lock. */ 1465static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, 1466 u32 pid, u32 seq, u32 flags, u8 cmd) 1467{ 1468 struct ovs_header *ovs_header; 1469 struct ovs_vport_stats vport_stats; 1470 int err; 1471 1472 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, 1473 flags, cmd); 1474 if (!ovs_header) 1475 return -EMSGSIZE; 1476 1477 ovs_header->dp_ifindex = get_dpifindex(vport->dp); 1478 1479 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no); 1480 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type); 1481 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)); 1482 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid); 1483 1484 ovs_vport_get_stats(vport, &vport_stats); 1485 NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats), 1486 &vport_stats); 1487 1488 err = ovs_vport_get_options(vport, skb); 1489 if (err == -EMSGSIZE) 1490 goto error; 1491 1492 return genlmsg_end(skb, ovs_header); 1493 1494nla_put_failure: 1495 err = -EMSGSIZE; 1496error: 1497 genlmsg_cancel(skb, ovs_header); 1498 return err; 1499} 1500 1501/* Called with RTNL lock or RCU read lock. */ 1502struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid, 1503 u32 seq, u8 cmd) 1504{ 1505 struct sk_buff *skb; 1506 int retval; 1507 1508 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 1509 if (!skb) 1510 return ERR_PTR(-ENOMEM); 1511 1512 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd); 1513 if (retval < 0) { 1514 kfree_skb(skb); 1515 return ERR_PTR(retval); 1516 } 1517 return skb; 1518} 1519 1520/* Called with RTNL lock or RCU read lock. */ 1521static struct vport *lookup_vport(struct ovs_header *ovs_header, 1522 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) 1523{ 1524 struct datapath *dp; 1525 struct vport *vport; 1526 1527 if (a[OVS_VPORT_ATTR_NAME]) { 1528 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME])); 1529 if (!vport) 1530 return ERR_PTR(-ENODEV); 1531 if (ovs_header->dp_ifindex && 1532 ovs_header->dp_ifindex != get_dpifindex(vport->dp)) 1533 return ERR_PTR(-ENODEV); 1534 return vport; 1535 } else if (a[OVS_VPORT_ATTR_PORT_NO]) { 1536 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1537 1538 if (port_no >= DP_MAX_PORTS) 1539 return ERR_PTR(-EFBIG); 1540 1541 dp = get_dp(ovs_header->dp_ifindex); 1542 if (!dp) 1543 return ERR_PTR(-ENODEV); 1544 1545 vport = rcu_dereference_rtnl(dp->ports[port_no]); 1546 if (!vport) 1547 return ERR_PTR(-ENOENT); 1548 return vport; 1549 } else 1550 return ERR_PTR(-EINVAL); 1551} 1552 1553static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) 1554{ 1555 struct nlattr **a = info->attrs; 1556 struct ovs_header *ovs_header = info->userhdr; 1557 struct vport_parms parms; 1558 struct sk_buff *reply; 1559 struct vport *vport; 1560 struct datapath *dp; 1561 u32 port_no; 1562 int err; 1563 1564 err = -EINVAL; 1565 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || 1566 !a[OVS_VPORT_ATTR_UPCALL_PID]) 1567 goto exit; 1568 1569 rtnl_lock(); 1570 dp = get_dp(ovs_header->dp_ifindex); 1571 err = -ENODEV; 1572 if (!dp) 1573 goto exit_unlock; 1574 1575 if (a[OVS_VPORT_ATTR_PORT_NO]) { 1576 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1577 1578 err = -EFBIG; 1579 if (port_no >= DP_MAX_PORTS) 1580 goto exit_unlock; 1581 1582 vport = rtnl_dereference(dp->ports[port_no]); 1583 err = -EBUSY; 1584 if (vport) 1585 goto exit_unlock; 1586 } else { 1587 for (port_no = 1; ; port_no++) { 1588 if (port_no >= DP_MAX_PORTS) { 1589 err = -EFBIG; 1590 goto exit_unlock; 1591 } 1592 vport = rtnl_dereference(dp->ports[port_no]); 1593 if (!vport) 1594 break; 1595 } 1596 } 1597 1598 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); 1599 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); 1600 parms.options = a[OVS_VPORT_ATTR_OPTIONS]; 1601 parms.dp = dp; 1602 parms.port_no = port_no; 1603 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); 1604 1605 vport = new_vport(&parms); 1606 err = PTR_ERR(vport); 1607 if (IS_ERR(vport)) 1608 goto exit_unlock; 1609 1610 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1611 OVS_VPORT_CMD_NEW); 1612 if (IS_ERR(reply)) { 1613 err = PTR_ERR(reply); 1614 ovs_dp_detach_port(vport); 1615 goto exit_unlock; 1616 } 1617 genl_notify(reply, genl_info_net(info), info->snd_pid, 1618 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1619 1620exit_unlock: 1621 rtnl_unlock(); 1622exit: 1623 return err; 1624} 1625 1626static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) 1627{ 1628 struct nlattr **a = info->attrs; 1629 struct sk_buff *reply; 1630 struct vport *vport; 1631 int err; 1632 1633 rtnl_lock(); 1634 vport = lookup_vport(info->userhdr, a); 1635 err = PTR_ERR(vport); 1636 if (IS_ERR(vport)) 1637 goto exit_unlock; 1638 1639 err = 0; 1640 if (a[OVS_VPORT_ATTR_TYPE] && 1641 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) 1642 err = -EINVAL; 1643 1644 if (!err && a[OVS_VPORT_ATTR_OPTIONS]) 1645 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); 1646 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID]) 1647 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); 1648 1649 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1650 OVS_VPORT_CMD_NEW); 1651 if (IS_ERR(reply)) { 1652 netlink_set_err(init_net.genl_sock, 0, 1653 ovs_dp_vport_multicast_group.id, PTR_ERR(reply)); 1654 goto exit_unlock; 1655 } 1656 1657 genl_notify(reply, genl_info_net(info), info->snd_pid, 1658 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1659 1660exit_unlock: 1661 rtnl_unlock(); 1662 return err; 1663} 1664 1665static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) 1666{ 1667 struct nlattr **a = info->attrs; 1668 struct sk_buff *reply; 1669 struct vport *vport; 1670 int err; 1671 1672 rtnl_lock(); 1673 vport = lookup_vport(info->userhdr, a); 1674 err = PTR_ERR(vport); 1675 if (IS_ERR(vport)) 1676 goto exit_unlock; 1677 1678 if (vport->port_no == OVSP_LOCAL) { 1679 err = -EINVAL; 1680 goto exit_unlock; 1681 } 1682 1683 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1684 OVS_VPORT_CMD_DEL); 1685 err = PTR_ERR(reply); 1686 if (IS_ERR(reply)) 1687 goto exit_unlock; 1688 1689 ovs_dp_detach_port(vport); 1690 1691 genl_notify(reply, genl_info_net(info), info->snd_pid, 1692 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); 1693 1694exit_unlock: 1695 rtnl_unlock(); 1696 return err; 1697} 1698 1699static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) 1700{ 1701 struct nlattr **a = info->attrs; 1702 struct ovs_header *ovs_header = info->userhdr; 1703 struct sk_buff *reply; 1704 struct vport *vport; 1705 int err; 1706 1707 rcu_read_lock(); 1708 vport = lookup_vport(ovs_header, a); 1709 err = PTR_ERR(vport); 1710 if (IS_ERR(vport)) 1711 goto exit_unlock; 1712 1713 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, 1714 OVS_VPORT_CMD_NEW); 1715 err = PTR_ERR(reply); 1716 if (IS_ERR(reply)) 1717 goto exit_unlock; 1718 1719 rcu_read_unlock(); 1720 1721 return genlmsg_reply(reply, info); 1722 1723exit_unlock: 1724 rcu_read_unlock(); 1725 return err; 1726} 1727 1728static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1729{ 1730 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1731 struct datapath *dp; 1732 u32 port_no; 1733 int retval; 1734 1735 dp = get_dp(ovs_header->dp_ifindex); 1736 if (!dp) 1737 return -ENODEV; 1738 1739 rcu_read_lock(); 1740 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) { 1741 struct vport *vport; 1742 1743 vport = rcu_dereference(dp->ports[port_no]); 1744 if (!vport) 1745 continue; 1746 1747 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid, 1748 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1749 OVS_VPORT_CMD_NEW) < 0) 1750 break; 1751 } 1752 rcu_read_unlock(); 1753 1754 cb->args[0] = port_no; 1755 retval = skb->len; 1756 1757 return retval; 1758} 1759 1760static void rehash_flow_table(struct work_struct *work) 1761{ 1762 struct datapath *dp; 1763 1764 genl_lock(); 1765 1766 list_for_each_entry(dp, &dps, list_node) { 1767 struct flow_table *old_table = genl_dereference(dp->table); 1768 struct flow_table *new_table; 1769 1770 new_table = ovs_flow_tbl_rehash(old_table); 1771 if (!IS_ERR(new_table)) { 1772 rcu_assign_pointer(dp->table, new_table); 1773 ovs_flow_tbl_deferred_destroy(old_table); 1774 } 1775 } 1776 1777 genl_unlock(); 1778 1779 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); 1780} 1781 1782static struct genl_ops dp_vport_genl_ops[] = { 1783 { .cmd = OVS_VPORT_CMD_NEW, 1784 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1785 .policy = vport_policy, 1786 .doit = ovs_vport_cmd_new 1787 }, 1788 { .cmd = OVS_VPORT_CMD_DEL, 1789 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1790 .policy = vport_policy, 1791 .doit = ovs_vport_cmd_del 1792 }, 1793 { .cmd = OVS_VPORT_CMD_GET, 1794 .flags = 0, /* OK for unprivileged users. */ 1795 .policy = vport_policy, 1796 .doit = ovs_vport_cmd_get, 1797 .dumpit = ovs_vport_cmd_dump 1798 }, 1799 { .cmd = OVS_VPORT_CMD_SET, 1800 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1801 .policy = vport_policy, 1802 .doit = ovs_vport_cmd_set, 1803 }, 1804}; 1805 1806struct genl_family_and_ops { 1807 struct genl_family *family; 1808 struct genl_ops *ops; 1809 int n_ops; 1810 struct genl_multicast_group *group; 1811}; 1812 1813static const struct genl_family_and_ops dp_genl_families[] = { 1814 { &dp_datapath_genl_family, 1815 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops), 1816 &ovs_dp_datapath_multicast_group }, 1817 { &dp_vport_genl_family, 1818 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops), 1819 &ovs_dp_vport_multicast_group }, 1820 { &dp_flow_genl_family, 1821 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops), 1822 &ovs_dp_flow_multicast_group }, 1823 { &dp_packet_genl_family, 1824 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), 1825 NULL }, 1826}; 1827 1828static void dp_unregister_genl(int n_families) 1829{ 1830 int i; 1831 1832 for (i = 0; i < n_families; i++) 1833 genl_unregister_family(dp_genl_families[i].family); 1834} 1835 1836static int dp_register_genl(void) 1837{ 1838 int n_registered; 1839 int err; 1840 int i; 1841 1842 n_registered = 0; 1843 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { 1844 const struct genl_family_and_ops *f = &dp_genl_families[i]; 1845 1846 err = genl_register_family_with_ops(f->family, f->ops, 1847 f->n_ops); 1848 if (err) 1849 goto error; 1850 n_registered++; 1851 1852 if (f->group) { 1853 err = genl_register_mc_group(f->family, f->group); 1854 if (err) 1855 goto error; 1856 } 1857 } 1858 1859 return 0; 1860 1861error: 1862 dp_unregister_genl(n_registered); 1863 return err; 1864} 1865 1866static int __init dp_init(void) 1867{ 1868 struct sk_buff *dummy_skb; 1869 int err; 1870 1871 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb)); 1872 1873 pr_info("Open vSwitch switching datapath\n"); 1874 1875 err = ovs_flow_init(); 1876 if (err) 1877 goto error; 1878 1879 err = ovs_vport_init(); 1880 if (err) 1881 goto error_flow_exit; 1882 1883 err = register_netdevice_notifier(&ovs_dp_device_notifier); 1884 if (err) 1885 goto error_vport_exit; 1886 1887 err = dp_register_genl(); 1888 if (err < 0) 1889 goto error_unreg_notifier; 1890 1891 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); 1892 1893 return 0; 1894 1895error_unreg_notifier: 1896 unregister_netdevice_notifier(&ovs_dp_device_notifier); 1897error_vport_exit: 1898 ovs_vport_exit(); 1899error_flow_exit: 1900 ovs_flow_exit(); 1901error: 1902 return err; 1903} 1904 1905static void dp_cleanup(void) 1906{ 1907 cancel_delayed_work_sync(&rehash_flow_wq); 1908 rcu_barrier(); 1909 dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); 1910 unregister_netdevice_notifier(&ovs_dp_device_notifier); 1911 ovs_vport_exit(); 1912 ovs_flow_exit(); 1913} 1914 1915module_init(dp_init); 1916module_exit(dp_cleanup); 1917 1918MODULE_DESCRIPTION("Open vSwitch switching datapath"); 1919MODULE_LICENSE("GPL"); 1920