1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2007-2014 Nicira, Inc. 4 */ 5 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include <linux/init.h> 9 #include <linux/module.h> 10 #include <linux/if_arp.h> 11 #include <linux/if_vlan.h> 12 #include <linux/in.h> 13 #include <linux/ip.h> 14 #include <linux/jhash.h> 15 #include <linux/delay.h> 16 #include <linux/time.h> 17 #include <linux/etherdevice.h> 18 #include <linux/genetlink.h> 19 #include <linux/kernel.h> 20 #include <linux/kthread.h> 21 #include <linux/mutex.h> 22 #include <linux/percpu.h> 23 #include <linux/rcupdate.h> 24 #include <linux/tcp.h> 25 #include <linux/udp.h> 26 #include <linux/ethtool.h> 27 #include <linux/wait.h> 28 #include <asm/div64.h> 29 #include <linux/highmem.h> 30 #include <linux/netfilter_bridge.h> 31 #include <linux/netfilter_ipv4.h> 32 #include <linux/inetdevice.h> 33 #include <linux/list.h> 34 #include <linux/openvswitch.h> 35 #include <linux/rculist.h> 36 #include <linux/dmi.h> 37 #include <net/genetlink.h> 38 #include <net/net_namespace.h> 39 #include <net/netns/generic.h> 40 41 #include "datapath.h" 42 #include "flow.h" 43 #include "flow_table.h" 44 #include "flow_netlink.h" 45 #include "meter.h" 46 #include "vport-internal_dev.h" 47 #include "vport-netdev.h" 48 49 unsigned int ovs_net_id __read_mostly; 50 51 static struct genl_family dp_packet_genl_family; 52 static struct genl_family dp_flow_genl_family; 53 static struct genl_family dp_datapath_genl_family; 54 55 static const struct nla_policy flow_policy[]; 56 57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = { 58 .name = OVS_FLOW_MCGROUP, 59 }; 60 61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = { 62 .name = OVS_DATAPATH_MCGROUP, 63 }; 64 65 static const struct genl_multicast_group ovs_dp_vport_multicast_group = { 66 .name = OVS_VPORT_MCGROUP, 67 }; 68 69 /* Check if need to build a reply message. 70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */ 71 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info, 72 unsigned int group) 73 { 74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO || 75 genl_has_listeners(family, genl_info_net(info), group); 76 } 77 78 static void ovs_notify(struct genl_family *family, 79 struct sk_buff *skb, struct genl_info *info) 80 { 81 genl_notify(family, skb, info, 0, GFP_KERNEL); 82 } 83 84 /** 85 * DOC: Locking: 86 * 87 * All writes e.g. Writes to device state (add/remove datapath, port, set 88 * operations on vports, etc.), Writes to other state (flow table 89 * modifications, set miscellaneous datapath parameters, etc.) are protected 90 * by ovs_lock. 91 * 92 * Reads are protected by RCU. 93 * 94 * There are a few special cases (mostly stats) that have their own 95 * synchronization but they nest under all of above and don't interact with 96 * each other. 97 * 98 * The RTNL lock nests inside ovs_mutex. 99 */ 100 101 static DEFINE_MUTEX(ovs_mutex); 102 103 void ovs_lock(void) 104 { 105 mutex_lock(&ovs_mutex); 106 } 107 108 void ovs_unlock(void) 109 { 110 mutex_unlock(&ovs_mutex); 111 } 112 113 #ifdef CONFIG_LOCKDEP 114 int lockdep_ovsl_is_held(void) 115 { 116 if (debug_locks) 117 return lockdep_is_held(&ovs_mutex); 118 else 119 return 1; 120 } 121 #endif 122 123 static struct vport *new_vport(const struct vport_parms *); 124 static int queue_gso_packets(struct datapath *dp, struct sk_buff *, 125 const struct sw_flow_key *, 126 const struct dp_upcall_info *, 127 uint32_t cutlen); 128 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *, 129 const struct sw_flow_key *, 130 const struct dp_upcall_info *, 131 uint32_t cutlen); 132 133 /* Must be called with rcu_read_lock or ovs_mutex. */ 134 const char *ovs_dp_name(const struct datapath *dp) 135 { 136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL); 137 return ovs_vport_name(vport); 138 } 139 140 static int get_dpifindex(const struct datapath *dp) 141 { 142 struct vport *local; 143 int ifindex; 144 145 rcu_read_lock(); 146 147 local = ovs_vport_rcu(dp, OVSP_LOCAL); 148 if (local) 149 ifindex = local->dev->ifindex; 150 else 151 ifindex = 0; 152 153 rcu_read_unlock(); 154 155 return ifindex; 156 } 157 158 static void destroy_dp_rcu(struct rcu_head *rcu) 159 { 160 struct datapath *dp = container_of(rcu, struct datapath, rcu); 161 162 ovs_flow_tbl_destroy(&dp->table); 163 free_percpu(dp->stats_percpu); 164 kfree(dp->ports); 165 ovs_meters_exit(dp); 166 kfree(dp); 167 } 168 169 static struct hlist_head *vport_hash_bucket(const struct datapath *dp, 170 u16 port_no) 171 { 172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; 173 } 174 175 /* Called with ovs_mutex or RCU read lock. */ 176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no) 177 { 178 struct vport *vport; 179 struct hlist_head *head; 180 181 head = vport_hash_bucket(dp, port_no); 182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) { 183 if (vport->port_no == port_no) 184 return vport; 185 } 186 return NULL; 187 } 188 189 /* Called with ovs_mutex. */ 190 static struct vport *new_vport(const struct vport_parms *parms) 191 { 192 struct vport *vport; 193 194 vport = ovs_vport_add(parms); 195 if (!IS_ERR(vport)) { 196 struct datapath *dp = parms->dp; 197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no); 198 199 hlist_add_head_rcu(&vport->dp_hash_node, head); 200 } 201 return vport; 202 } 203 204 void ovs_dp_detach_port(struct vport *p) 205 { 206 ASSERT_OVSL(); 207 208 /* First drop references to device. */ 209 hlist_del_rcu(&p->dp_hash_node); 210 211 /* Then destroy it. */ 212 ovs_vport_del(p); 213 } 214 215 /* Must be called with rcu_read_lock. */ 216 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key) 217 { 218 const struct vport *p = OVS_CB(skb)->input_vport; 219 struct datapath *dp = p->dp; 220 struct sw_flow *flow; 221 struct sw_flow_actions *sf_acts; 222 struct dp_stats_percpu *stats; 223 u64 *stats_counter; 224 u32 n_mask_hit; 225 int error; 226 227 stats = this_cpu_ptr(dp->stats_percpu); 228 229 /* Look up flow. */ 230 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit); 231 if (unlikely(!flow)) { 232 struct dp_upcall_info upcall; 233 234 memset(&upcall, 0, sizeof(upcall)); 235 upcall.cmd = OVS_PACKET_CMD_MISS; 236 upcall.portid = ovs_vport_find_upcall_portid(p, skb); 237 upcall.mru = OVS_CB(skb)->mru; 238 error = ovs_dp_upcall(dp, skb, key, &upcall, 0); 239 if (unlikely(error)) 240 kfree_skb(skb); 241 else 242 consume_skb(skb); 243 stats_counter = &stats->n_missed; 244 goto out; 245 } 246 247 ovs_flow_stats_update(flow, key->tp.flags, skb); 248 sf_acts = rcu_dereference(flow->sf_acts); 249 error = ovs_execute_actions(dp, skb, sf_acts, key); 250 if (unlikely(error)) 251 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n", 252 ovs_dp_name(dp), error); 253 254 stats_counter = &stats->n_hit; 255 256 out: 257 /* Update datapath statistics. */ 258 u64_stats_update_begin(&stats->syncp); 259 (*stats_counter)++; 260 stats->n_mask_hit += n_mask_hit; 261 u64_stats_update_end(&stats->syncp); 262 } 263 264 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, 265 const struct sw_flow_key *key, 266 const struct dp_upcall_info *upcall_info, 267 uint32_t cutlen) 268 { 269 struct dp_stats_percpu *stats; 270 int err; 271 272 if (upcall_info->portid == 0) { 273 err = -ENOTCONN; 274 goto err; 275 } 276 277 if (!skb_is_gso(skb)) 278 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 279 else 280 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen); 281 if (err) 282 goto err; 283 284 return 0; 285 286 err: 287 stats = this_cpu_ptr(dp->stats_percpu); 288 289 u64_stats_update_begin(&stats->syncp); 290 stats->n_lost++; 291 u64_stats_update_end(&stats->syncp); 292 293 return err; 294 } 295 296 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb, 297 const struct sw_flow_key *key, 298 const struct dp_upcall_info *upcall_info, 299 uint32_t cutlen) 300 { 301 unsigned int gso_type = skb_shinfo(skb)->gso_type; 302 struct sw_flow_key later_key; 303 struct sk_buff *segs, *nskb; 304 int err; 305 306 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET); 307 segs = __skb_gso_segment(skb, NETIF_F_SG, false); 308 if (IS_ERR(segs)) 309 return PTR_ERR(segs); 310 if (segs == NULL) 311 return -EINVAL; 312 313 if (gso_type & SKB_GSO_UDP) { 314 /* The initial flow key extracted by ovs_flow_key_extract() 315 * in this case is for a first fragment, so we need to 316 * properly mark later fragments. 317 */ 318 later_key = *key; 319 later_key.ip.frag = OVS_FRAG_TYPE_LATER; 320 } 321 322 /* Queue all of the segments. */ 323 skb = segs; 324 do { 325 if (gso_type & SKB_GSO_UDP && skb != segs) 326 key = &later_key; 327 328 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 329 if (err) 330 break; 331 332 } while ((skb = skb->next)); 333 334 /* Free all of the segments. */ 335 skb = segs; 336 do { 337 nskb = skb->next; 338 if (err) 339 kfree_skb(skb); 340 else 341 consume_skb(skb); 342 } while ((skb = nskb)); 343 return err; 344 } 345 346 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info, 347 unsigned int hdrlen, int actions_attrlen) 348 { 349 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header)) 350 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */ 351 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */ 352 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */ 353 354 /* OVS_PACKET_ATTR_USERDATA */ 355 if (upcall_info->userdata) 356 size += NLA_ALIGN(upcall_info->userdata->nla_len); 357 358 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */ 359 if (upcall_info->egress_tun_info) 360 size += nla_total_size(ovs_tun_key_attr_size()); 361 362 /* OVS_PACKET_ATTR_ACTIONS */ 363 if (upcall_info->actions_len) 364 size += nla_total_size(actions_attrlen); 365 366 /* OVS_PACKET_ATTR_MRU */ 367 if (upcall_info->mru) 368 size += nla_total_size(sizeof(upcall_info->mru)); 369 370 return size; 371 } 372 373 static void pad_packet(struct datapath *dp, struct sk_buff *skb) 374 { 375 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) { 376 size_t plen = NLA_ALIGN(skb->len) - skb->len; 377 378 if (plen > 0) 379 skb_put_zero(skb, plen); 380 } 381 } 382 383 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb, 384 const struct sw_flow_key *key, 385 const struct dp_upcall_info *upcall_info, 386 uint32_t cutlen) 387 { 388 struct ovs_header *upcall; 389 struct sk_buff *nskb = NULL; 390 struct sk_buff *user_skb = NULL; /* to be queued to userspace */ 391 struct nlattr *nla; 392 size_t len; 393 unsigned int hlen; 394 int err, dp_ifindex; 395 396 dp_ifindex = get_dpifindex(dp); 397 if (!dp_ifindex) 398 return -ENODEV; 399 400 if (skb_vlan_tag_present(skb)) { 401 nskb = skb_clone(skb, GFP_ATOMIC); 402 if (!nskb) 403 return -ENOMEM; 404 405 nskb = __vlan_hwaccel_push_inside(nskb); 406 if (!nskb) 407 return -ENOMEM; 408 409 skb = nskb; 410 } 411 412 if (nla_attr_size(skb->len) > USHRT_MAX) { 413 err = -EFBIG; 414 goto out; 415 } 416 417 /* Complete checksum if needed */ 418 if (skb->ip_summed == CHECKSUM_PARTIAL && 419 (err = skb_csum_hwoffload_help(skb, 0))) 420 goto out; 421 422 /* Older versions of OVS user space enforce alignment of the last 423 * Netlink attribute to NLA_ALIGNTO which would require extensive 424 * padding logic. Only perform zerocopy if padding is not required. 425 */ 426 if (dp->user_features & OVS_DP_F_UNALIGNED) 427 hlen = skb_zerocopy_headlen(skb); 428 else 429 hlen = skb->len; 430 431 len = upcall_msg_size(upcall_info, hlen - cutlen, 432 OVS_CB(skb)->acts_origlen); 433 user_skb = genlmsg_new(len, GFP_ATOMIC); 434 if (!user_skb) { 435 err = -ENOMEM; 436 goto out; 437 } 438 439 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 440 0, upcall_info->cmd); 441 if (!upcall) { 442 err = -EINVAL; 443 goto out; 444 } 445 upcall->dp_ifindex = dp_ifindex; 446 447 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb); 448 if (err) 449 goto out; 450 451 if (upcall_info->userdata) 452 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA, 453 nla_len(upcall_info->userdata), 454 nla_data(upcall_info->userdata)); 455 456 if (upcall_info->egress_tun_info) { 457 nla = nla_nest_start_noflag(user_skb, 458 OVS_PACKET_ATTR_EGRESS_TUN_KEY); 459 if (!nla) { 460 err = -EMSGSIZE; 461 goto out; 462 } 463 err = ovs_nla_put_tunnel_info(user_skb, 464 upcall_info->egress_tun_info); 465 if (err) 466 goto out; 467 468 nla_nest_end(user_skb, nla); 469 } 470 471 if (upcall_info->actions_len) { 472 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS); 473 if (!nla) { 474 err = -EMSGSIZE; 475 goto out; 476 } 477 err = ovs_nla_put_actions(upcall_info->actions, 478 upcall_info->actions_len, 479 user_skb); 480 if (!err) 481 nla_nest_end(user_skb, nla); 482 else 483 nla_nest_cancel(user_skb, nla); 484 } 485 486 /* Add OVS_PACKET_ATTR_MRU */ 487 if (upcall_info->mru) { 488 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, 489 upcall_info->mru)) { 490 err = -ENOBUFS; 491 goto out; 492 } 493 pad_packet(dp, user_skb); 494 } 495 496 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */ 497 if (cutlen > 0) { 498 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, 499 skb->len)) { 500 err = -ENOBUFS; 501 goto out; 502 } 503 pad_packet(dp, user_skb); 504 } 505 506 /* Only reserve room for attribute header, packet data is added 507 * in skb_zerocopy() */ 508 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) { 509 err = -ENOBUFS; 510 goto out; 511 } 512 nla->nla_len = nla_attr_size(skb->len - cutlen); 513 514 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen); 515 if (err) 516 goto out; 517 518 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */ 519 pad_packet(dp, user_skb); 520 521 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; 522 523 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid); 524 user_skb = NULL; 525 out: 526 if (err) 527 skb_tx_error(skb); 528 kfree_skb(user_skb); 529 kfree_skb(nskb); 530 return err; 531 } 532 533 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) 534 { 535 struct ovs_header *ovs_header = info->userhdr; 536 struct net *net = sock_net(skb->sk); 537 struct nlattr **a = info->attrs; 538 struct sw_flow_actions *acts; 539 struct sk_buff *packet; 540 struct sw_flow *flow; 541 struct sw_flow_actions *sf_acts; 542 struct datapath *dp; 543 struct vport *input_vport; 544 u16 mru = 0; 545 int len; 546 int err; 547 bool log = !a[OVS_PACKET_ATTR_PROBE]; 548 549 err = -EINVAL; 550 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || 551 !a[OVS_PACKET_ATTR_ACTIONS]) 552 goto err; 553 554 len = nla_len(a[OVS_PACKET_ATTR_PACKET]); 555 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); 556 err = -ENOMEM; 557 if (!packet) 558 goto err; 559 skb_reserve(packet, NET_IP_ALIGN); 560 561 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len); 562 563 /* Set packet's mru */ 564 if (a[OVS_PACKET_ATTR_MRU]) { 565 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]); 566 packet->ignore_df = 1; 567 } 568 OVS_CB(packet)->mru = mru; 569 570 /* Build an sw_flow for sending this packet. */ 571 flow = ovs_flow_alloc(); 572 err = PTR_ERR(flow); 573 if (IS_ERR(flow)) 574 goto err_kfree_skb; 575 576 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY], 577 packet, &flow->key, log); 578 if (err) 579 goto err_flow_free; 580 581 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS], 582 &flow->key, &acts, log); 583 if (err) 584 goto err_flow_free; 585 586 rcu_assign_pointer(flow->sf_acts, acts); 587 packet->priority = flow->key.phy.priority; 588 packet->mark = flow->key.phy.skb_mark; 589 590 rcu_read_lock(); 591 dp = get_dp_rcu(net, ovs_header->dp_ifindex); 592 err = -ENODEV; 593 if (!dp) 594 goto err_unlock; 595 596 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port); 597 if (!input_vport) 598 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL); 599 600 if (!input_vport) 601 goto err_unlock; 602 603 packet->dev = input_vport->dev; 604 OVS_CB(packet)->input_vport = input_vport; 605 sf_acts = rcu_dereference(flow->sf_acts); 606 607 local_bh_disable(); 608 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key); 609 local_bh_enable(); 610 rcu_read_unlock(); 611 612 ovs_flow_free(flow, false); 613 return err; 614 615 err_unlock: 616 rcu_read_unlock(); 617 err_flow_free: 618 ovs_flow_free(flow, false); 619 err_kfree_skb: 620 kfree_skb(packet); 621 err: 622 return err; 623 } 624 625 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { 626 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN }, 627 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, 628 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, 629 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG }, 630 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 }, 631 }; 632 633 static const struct genl_ops dp_packet_genl_ops[] = { 634 { .cmd = OVS_PACKET_CMD_EXECUTE, 635 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 636 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 637 .doit = ovs_packet_cmd_execute 638 } 639 }; 640 641 static struct genl_family dp_packet_genl_family __ro_after_init = { 642 .hdrsize = sizeof(struct ovs_header), 643 .name = OVS_PACKET_FAMILY, 644 .version = OVS_PACKET_VERSION, 645 .maxattr = OVS_PACKET_ATTR_MAX, 646 .policy = packet_policy, 647 .netnsok = true, 648 .parallel_ops = true, 649 .ops = dp_packet_genl_ops, 650 .n_ops = ARRAY_SIZE(dp_packet_genl_ops), 651 .module = THIS_MODULE, 652 }; 653 654 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats, 655 struct ovs_dp_megaflow_stats *mega_stats) 656 { 657 int i; 658 659 memset(mega_stats, 0, sizeof(*mega_stats)); 660 661 stats->n_flows = ovs_flow_tbl_count(&dp->table); 662 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table); 663 664 stats->n_hit = stats->n_missed = stats->n_lost = 0; 665 666 for_each_possible_cpu(i) { 667 const struct dp_stats_percpu *percpu_stats; 668 struct dp_stats_percpu local_stats; 669 unsigned int start; 670 671 percpu_stats = per_cpu_ptr(dp->stats_percpu, i); 672 673 do { 674 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp); 675 local_stats = *percpu_stats; 676 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start)); 677 678 stats->n_hit += local_stats.n_hit; 679 stats->n_missed += local_stats.n_missed; 680 stats->n_lost += local_stats.n_lost; 681 mega_stats->n_mask_hit += local_stats.n_mask_hit; 682 } 683 } 684 685 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags) 686 { 687 return ovs_identifier_is_ufid(sfid) && 688 !(ufid_flags & OVS_UFID_F_OMIT_KEY); 689 } 690 691 static bool should_fill_mask(uint32_t ufid_flags) 692 { 693 return !(ufid_flags & OVS_UFID_F_OMIT_MASK); 694 } 695 696 static bool should_fill_actions(uint32_t ufid_flags) 697 { 698 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS); 699 } 700 701 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts, 702 const struct sw_flow_id *sfid, 703 uint32_t ufid_flags) 704 { 705 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header)); 706 707 /* OVS_FLOW_ATTR_UFID */ 708 if (sfid && ovs_identifier_is_ufid(sfid)) 709 len += nla_total_size(sfid->ufid_len); 710 711 /* OVS_FLOW_ATTR_KEY */ 712 if (!sfid || should_fill_key(sfid, ufid_flags)) 713 len += nla_total_size(ovs_key_attr_size()); 714 715 /* OVS_FLOW_ATTR_MASK */ 716 if (should_fill_mask(ufid_flags)) 717 len += nla_total_size(ovs_key_attr_size()); 718 719 /* OVS_FLOW_ATTR_ACTIONS */ 720 if (should_fill_actions(ufid_flags)) 721 len += nla_total_size(acts->orig_len); 722 723 return len 724 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */ 725 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */ 726 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */ 727 } 728 729 /* Called with ovs_mutex or RCU read lock. */ 730 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow, 731 struct sk_buff *skb) 732 { 733 struct ovs_flow_stats stats; 734 __be16 tcp_flags; 735 unsigned long used; 736 737 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags); 738 739 if (used && 740 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used), 741 OVS_FLOW_ATTR_PAD)) 742 return -EMSGSIZE; 743 744 if (stats.n_packets && 745 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS, 746 sizeof(struct ovs_flow_stats), &stats, 747 OVS_FLOW_ATTR_PAD)) 748 return -EMSGSIZE; 749 750 if ((u8)ntohs(tcp_flags) && 751 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags))) 752 return -EMSGSIZE; 753 754 return 0; 755 } 756 757 /* Called with ovs_mutex or RCU read lock. */ 758 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow, 759 struct sk_buff *skb, int skb_orig_len) 760 { 761 struct nlattr *start; 762 int err; 763 764 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if 765 * this is the first flow to be dumped into 'skb'. This is unusual for 766 * Netlink but individual action lists can be longer than 767 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. 768 * The userspace caller can always fetch the actions separately if it 769 * really wants them. (Most userspace callers in fact don't care.) 770 * 771 * This can only fail for dump operations because the skb is always 772 * properly sized for single flows. 773 */ 774 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS); 775 if (start) { 776 const struct sw_flow_actions *sf_acts; 777 778 sf_acts = rcu_dereference_ovsl(flow->sf_acts); 779 err = ovs_nla_put_actions(sf_acts->actions, 780 sf_acts->actions_len, skb); 781 782 if (!err) 783 nla_nest_end(skb, start); 784 else { 785 if (skb_orig_len) 786 return err; 787 788 nla_nest_cancel(skb, start); 789 } 790 } else if (skb_orig_len) { 791 return -EMSGSIZE; 792 } 793 794 return 0; 795 } 796 797 /* Called with ovs_mutex or RCU read lock. */ 798 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex, 799 struct sk_buff *skb, u32 portid, 800 u32 seq, u32 flags, u8 cmd, u32 ufid_flags) 801 { 802 const int skb_orig_len = skb->len; 803 struct ovs_header *ovs_header; 804 int err; 805 806 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, 807 flags, cmd); 808 if (!ovs_header) 809 return -EMSGSIZE; 810 811 ovs_header->dp_ifindex = dp_ifindex; 812 813 err = ovs_nla_put_identifier(flow, skb); 814 if (err) 815 goto error; 816 817 if (should_fill_key(&flow->id, ufid_flags)) { 818 err = ovs_nla_put_masked_key(flow, skb); 819 if (err) 820 goto error; 821 } 822 823 if (should_fill_mask(ufid_flags)) { 824 err = ovs_nla_put_mask(flow, skb); 825 if (err) 826 goto error; 827 } 828 829 err = ovs_flow_cmd_fill_stats(flow, skb); 830 if (err) 831 goto error; 832 833 if (should_fill_actions(ufid_flags)) { 834 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len); 835 if (err) 836 goto error; 837 } 838 839 genlmsg_end(skb, ovs_header); 840 return 0; 841 842 error: 843 genlmsg_cancel(skb, ovs_header); 844 return err; 845 } 846 847 /* May not be called with RCU read lock. */ 848 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts, 849 const struct sw_flow_id *sfid, 850 struct genl_info *info, 851 bool always, 852 uint32_t ufid_flags) 853 { 854 struct sk_buff *skb; 855 size_t len; 856 857 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0)) 858 return NULL; 859 860 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags); 861 skb = genlmsg_new(len, GFP_KERNEL); 862 if (!skb) 863 return ERR_PTR(-ENOMEM); 864 865 return skb; 866 } 867 868 /* Called with ovs_mutex. */ 869 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow, 870 int dp_ifindex, 871 struct genl_info *info, u8 cmd, 872 bool always, u32 ufid_flags) 873 { 874 struct sk_buff *skb; 875 int retval; 876 877 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), 878 &flow->id, info, always, ufid_flags); 879 if (IS_ERR_OR_NULL(skb)) 880 return skb; 881 882 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb, 883 info->snd_portid, info->snd_seq, 0, 884 cmd, ufid_flags); 885 BUG_ON(retval < 0); 886 return skb; 887 } 888 889 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info) 890 { 891 struct net *net = sock_net(skb->sk); 892 struct nlattr **a = info->attrs; 893 struct ovs_header *ovs_header = info->userhdr; 894 struct sw_flow *flow = NULL, *new_flow; 895 struct sw_flow_mask mask; 896 struct sk_buff *reply; 897 struct datapath *dp; 898 struct sw_flow_actions *acts; 899 struct sw_flow_match match; 900 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 901 int error; 902 bool log = !a[OVS_FLOW_ATTR_PROBE]; 903 904 /* Must have key and actions. */ 905 error = -EINVAL; 906 if (!a[OVS_FLOW_ATTR_KEY]) { 907 OVS_NLERR(log, "Flow key attr not present in new flow."); 908 goto error; 909 } 910 if (!a[OVS_FLOW_ATTR_ACTIONS]) { 911 OVS_NLERR(log, "Flow actions attr not present in new flow."); 912 goto error; 913 } 914 915 /* Most of the time we need to allocate a new flow, do it before 916 * locking. 917 */ 918 new_flow = ovs_flow_alloc(); 919 if (IS_ERR(new_flow)) { 920 error = PTR_ERR(new_flow); 921 goto error; 922 } 923 924 /* Extract key. */ 925 ovs_match_init(&match, &new_flow->key, false, &mask); 926 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 927 a[OVS_FLOW_ATTR_MASK], log); 928 if (error) 929 goto err_kfree_flow; 930 931 /* Extract flow identifier. */ 932 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID], 933 &new_flow->key, log); 934 if (error) 935 goto err_kfree_flow; 936 937 /* unmasked key is needed to match when ufid is not used. */ 938 if (ovs_identifier_is_key(&new_flow->id)) 939 match.key = new_flow->id.unmasked_key; 940 941 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask); 942 943 /* Validate actions. */ 944 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS], 945 &new_flow->key, &acts, log); 946 if (error) { 947 OVS_NLERR(log, "Flow actions may not be safe on all matching packets."); 948 goto err_kfree_flow; 949 } 950 951 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false, 952 ufid_flags); 953 if (IS_ERR(reply)) { 954 error = PTR_ERR(reply); 955 goto err_kfree_acts; 956 } 957 958 ovs_lock(); 959 dp = get_dp(net, ovs_header->dp_ifindex); 960 if (unlikely(!dp)) { 961 error = -ENODEV; 962 goto err_unlock_ovs; 963 } 964 965 /* Check if this is a duplicate flow */ 966 if (ovs_identifier_is_ufid(&new_flow->id)) 967 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id); 968 if (!flow) 969 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key); 970 if (likely(!flow)) { 971 rcu_assign_pointer(new_flow->sf_acts, acts); 972 973 /* Put flow in bucket. */ 974 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask); 975 if (unlikely(error)) { 976 acts = NULL; 977 goto err_unlock_ovs; 978 } 979 980 if (unlikely(reply)) { 981 error = ovs_flow_cmd_fill_info(new_flow, 982 ovs_header->dp_ifindex, 983 reply, info->snd_portid, 984 info->snd_seq, 0, 985 OVS_FLOW_CMD_NEW, 986 ufid_flags); 987 BUG_ON(error < 0); 988 } 989 ovs_unlock(); 990 } else { 991 struct sw_flow_actions *old_acts; 992 993 /* Bail out if we're not allowed to modify an existing flow. 994 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL 995 * because Generic Netlink treats the latter as a dump 996 * request. We also accept NLM_F_EXCL in case that bug ever 997 * gets fixed. 998 */ 999 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE 1000 | NLM_F_EXCL))) { 1001 error = -EEXIST; 1002 goto err_unlock_ovs; 1003 } 1004 /* The flow identifier has to be the same for flow updates. 1005 * Look for any overlapping flow. 1006 */ 1007 if (unlikely(!ovs_flow_cmp(flow, &match))) { 1008 if (ovs_identifier_is_key(&flow->id)) 1009 flow = ovs_flow_tbl_lookup_exact(&dp->table, 1010 &match); 1011 else /* UFID matches but key is different */ 1012 flow = NULL; 1013 if (!flow) { 1014 error = -ENOENT; 1015 goto err_unlock_ovs; 1016 } 1017 } 1018 /* Update actions. */ 1019 old_acts = ovsl_dereference(flow->sf_acts); 1020 rcu_assign_pointer(flow->sf_acts, acts); 1021 1022 if (unlikely(reply)) { 1023 error = ovs_flow_cmd_fill_info(flow, 1024 ovs_header->dp_ifindex, 1025 reply, info->snd_portid, 1026 info->snd_seq, 0, 1027 OVS_FLOW_CMD_NEW, 1028 ufid_flags); 1029 BUG_ON(error < 0); 1030 } 1031 ovs_unlock(); 1032 1033 ovs_nla_free_flow_actions_rcu(old_acts); 1034 ovs_flow_free(new_flow, false); 1035 } 1036 1037 if (reply) 1038 ovs_notify(&dp_flow_genl_family, reply, info); 1039 return 0; 1040 1041 err_unlock_ovs: 1042 ovs_unlock(); 1043 kfree_skb(reply); 1044 err_kfree_acts: 1045 ovs_nla_free_flow_actions(acts); 1046 err_kfree_flow: 1047 ovs_flow_free(new_flow, false); 1048 error: 1049 return error; 1050 } 1051 1052 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */ 1053 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net, 1054 const struct nlattr *a, 1055 const struct sw_flow_key *key, 1056 const struct sw_flow_mask *mask, 1057 bool log) 1058 { 1059 struct sw_flow_actions *acts; 1060 struct sw_flow_key masked_key; 1061 int error; 1062 1063 ovs_flow_mask_key(&masked_key, key, true, mask); 1064 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log); 1065 if (error) { 1066 OVS_NLERR(log, 1067 "Actions may not be safe on all matching packets"); 1068 return ERR_PTR(error); 1069 } 1070 1071 return acts; 1072 } 1073 1074 /* Factor out match-init and action-copy to avoid 1075 * "Wframe-larger-than=1024" warning. Because mask is only 1076 * used to get actions, we new a function to save some 1077 * stack space. 1078 * 1079 * If there are not key and action attrs, we return 0 1080 * directly. In the case, the caller will also not use the 1081 * match as before. If there is action attr, we try to get 1082 * actions and save them to *acts. Before returning from 1083 * the function, we reset the match->mask pointer. Because 1084 * we should not to return match object with dangling reference 1085 * to mask. 1086 * */ 1087 static noinline_for_stack int 1088 ovs_nla_init_match_and_action(struct net *net, 1089 struct sw_flow_match *match, 1090 struct sw_flow_key *key, 1091 struct nlattr **a, 1092 struct sw_flow_actions **acts, 1093 bool log) 1094 { 1095 struct sw_flow_mask mask; 1096 int error = 0; 1097 1098 if (a[OVS_FLOW_ATTR_KEY]) { 1099 ovs_match_init(match, key, true, &mask); 1100 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY], 1101 a[OVS_FLOW_ATTR_MASK], log); 1102 if (error) 1103 goto error; 1104 } 1105 1106 if (a[OVS_FLOW_ATTR_ACTIONS]) { 1107 if (!a[OVS_FLOW_ATTR_KEY]) { 1108 OVS_NLERR(log, 1109 "Flow key attribute not present in set flow."); 1110 error = -EINVAL; 1111 goto error; 1112 } 1113 1114 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key, 1115 &mask, log); 1116 if (IS_ERR(*acts)) { 1117 error = PTR_ERR(*acts); 1118 goto error; 1119 } 1120 } 1121 1122 /* On success, error is 0. */ 1123 error: 1124 match->mask = NULL; 1125 return error; 1126 } 1127 1128 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info) 1129 { 1130 struct net *net = sock_net(skb->sk); 1131 struct nlattr **a = info->attrs; 1132 struct ovs_header *ovs_header = info->userhdr; 1133 struct sw_flow_key key; 1134 struct sw_flow *flow; 1135 struct sk_buff *reply = NULL; 1136 struct datapath *dp; 1137 struct sw_flow_actions *old_acts = NULL, *acts = NULL; 1138 struct sw_flow_match match; 1139 struct sw_flow_id sfid; 1140 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1141 int error = 0; 1142 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1143 bool ufid_present; 1144 1145 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log); 1146 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) { 1147 OVS_NLERR(log, 1148 "Flow set message rejected, Key attribute missing."); 1149 return -EINVAL; 1150 } 1151 1152 error = ovs_nla_init_match_and_action(net, &match, &key, a, 1153 &acts, log); 1154 if (error) 1155 goto error; 1156 1157 if (acts) { 1158 /* Can allocate before locking if have acts. */ 1159 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false, 1160 ufid_flags); 1161 if (IS_ERR(reply)) { 1162 error = PTR_ERR(reply); 1163 goto err_kfree_acts; 1164 } 1165 } 1166 1167 ovs_lock(); 1168 dp = get_dp(net, ovs_header->dp_ifindex); 1169 if (unlikely(!dp)) { 1170 error = -ENODEV; 1171 goto err_unlock_ovs; 1172 } 1173 /* Check that the flow exists. */ 1174 if (ufid_present) 1175 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid); 1176 else 1177 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1178 if (unlikely(!flow)) { 1179 error = -ENOENT; 1180 goto err_unlock_ovs; 1181 } 1182 1183 /* Update actions, if present. */ 1184 if (likely(acts)) { 1185 old_acts = ovsl_dereference(flow->sf_acts); 1186 rcu_assign_pointer(flow->sf_acts, acts); 1187 1188 if (unlikely(reply)) { 1189 error = ovs_flow_cmd_fill_info(flow, 1190 ovs_header->dp_ifindex, 1191 reply, info->snd_portid, 1192 info->snd_seq, 0, 1193 OVS_FLOW_CMD_SET, 1194 ufid_flags); 1195 BUG_ON(error < 0); 1196 } 1197 } else { 1198 /* Could not alloc without acts before locking. */ 1199 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, 1200 info, OVS_FLOW_CMD_SET, false, 1201 ufid_flags); 1202 1203 if (IS_ERR(reply)) { 1204 error = PTR_ERR(reply); 1205 goto err_unlock_ovs; 1206 } 1207 } 1208 1209 /* Clear stats. */ 1210 if (a[OVS_FLOW_ATTR_CLEAR]) 1211 ovs_flow_stats_clear(flow); 1212 ovs_unlock(); 1213 1214 if (reply) 1215 ovs_notify(&dp_flow_genl_family, reply, info); 1216 if (old_acts) 1217 ovs_nla_free_flow_actions_rcu(old_acts); 1218 1219 return 0; 1220 1221 err_unlock_ovs: 1222 ovs_unlock(); 1223 kfree_skb(reply); 1224 err_kfree_acts: 1225 ovs_nla_free_flow_actions(acts); 1226 error: 1227 return error; 1228 } 1229 1230 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) 1231 { 1232 struct nlattr **a = info->attrs; 1233 struct ovs_header *ovs_header = info->userhdr; 1234 struct net *net = sock_net(skb->sk); 1235 struct sw_flow_key key; 1236 struct sk_buff *reply; 1237 struct sw_flow *flow; 1238 struct datapath *dp; 1239 struct sw_flow_match match; 1240 struct sw_flow_id ufid; 1241 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1242 int err = 0; 1243 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1244 bool ufid_present; 1245 1246 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1247 if (a[OVS_FLOW_ATTR_KEY]) { 1248 ovs_match_init(&match, &key, true, NULL); 1249 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL, 1250 log); 1251 } else if (!ufid_present) { 1252 OVS_NLERR(log, 1253 "Flow get message rejected, Key attribute missing."); 1254 err = -EINVAL; 1255 } 1256 if (err) 1257 return err; 1258 1259 ovs_lock(); 1260 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1261 if (!dp) { 1262 err = -ENODEV; 1263 goto unlock; 1264 } 1265 1266 if (ufid_present) 1267 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1268 else 1269 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1270 if (!flow) { 1271 err = -ENOENT; 1272 goto unlock; 1273 } 1274 1275 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info, 1276 OVS_FLOW_CMD_GET, true, ufid_flags); 1277 if (IS_ERR(reply)) { 1278 err = PTR_ERR(reply); 1279 goto unlock; 1280 } 1281 1282 ovs_unlock(); 1283 return genlmsg_reply(reply, info); 1284 unlock: 1285 ovs_unlock(); 1286 return err; 1287 } 1288 1289 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) 1290 { 1291 struct nlattr **a = info->attrs; 1292 struct ovs_header *ovs_header = info->userhdr; 1293 struct net *net = sock_net(skb->sk); 1294 struct sw_flow_key key; 1295 struct sk_buff *reply; 1296 struct sw_flow *flow = NULL; 1297 struct datapath *dp; 1298 struct sw_flow_match match; 1299 struct sw_flow_id ufid; 1300 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1301 int err; 1302 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1303 bool ufid_present; 1304 1305 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1306 if (a[OVS_FLOW_ATTR_KEY]) { 1307 ovs_match_init(&match, &key, true, NULL); 1308 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 1309 NULL, log); 1310 if (unlikely(err)) 1311 return err; 1312 } 1313 1314 ovs_lock(); 1315 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1316 if (unlikely(!dp)) { 1317 err = -ENODEV; 1318 goto unlock; 1319 } 1320 1321 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) { 1322 err = ovs_flow_tbl_flush(&dp->table); 1323 goto unlock; 1324 } 1325 1326 if (ufid_present) 1327 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1328 else 1329 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1330 if (unlikely(!flow)) { 1331 err = -ENOENT; 1332 goto unlock; 1333 } 1334 1335 ovs_flow_tbl_remove(&dp->table, flow); 1336 ovs_unlock(); 1337 1338 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts, 1339 &flow->id, info, false, ufid_flags); 1340 if (likely(reply)) { 1341 if (!IS_ERR(reply)) { 1342 rcu_read_lock(); /*To keep RCU checker happy. */ 1343 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, 1344 reply, info->snd_portid, 1345 info->snd_seq, 0, 1346 OVS_FLOW_CMD_DEL, 1347 ufid_flags); 1348 rcu_read_unlock(); 1349 BUG_ON(err < 0); 1350 1351 ovs_notify(&dp_flow_genl_family, reply, info); 1352 } else { 1353 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply)); 1354 } 1355 } 1356 1357 ovs_flow_free(flow, true); 1358 return 0; 1359 unlock: 1360 ovs_unlock(); 1361 return err; 1362 } 1363 1364 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1365 { 1366 struct nlattr *a[__OVS_FLOW_ATTR_MAX]; 1367 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1368 struct table_instance *ti; 1369 struct datapath *dp; 1370 u32 ufid_flags; 1371 int err; 1372 1373 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a, 1374 OVS_FLOW_ATTR_MAX, flow_policy, NULL); 1375 if (err) 1376 return err; 1377 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1378 1379 rcu_read_lock(); 1380 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 1381 if (!dp) { 1382 rcu_read_unlock(); 1383 return -ENODEV; 1384 } 1385 1386 ti = rcu_dereference(dp->table.ti); 1387 for (;;) { 1388 struct sw_flow *flow; 1389 u32 bucket, obj; 1390 1391 bucket = cb->args[0]; 1392 obj = cb->args[1]; 1393 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj); 1394 if (!flow) 1395 break; 1396 1397 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb, 1398 NETLINK_CB(cb->skb).portid, 1399 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1400 OVS_FLOW_CMD_GET, ufid_flags) < 0) 1401 break; 1402 1403 cb->args[0] = bucket; 1404 cb->args[1] = obj; 1405 } 1406 rcu_read_unlock(); 1407 return skb->len; 1408 } 1409 1410 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { 1411 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, 1412 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED }, 1413 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, 1414 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, 1415 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG }, 1416 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 }, 1417 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 }, 1418 }; 1419 1420 static const struct genl_ops dp_flow_genl_ops[] = { 1421 { .cmd = OVS_FLOW_CMD_NEW, 1422 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1423 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1424 .doit = ovs_flow_cmd_new 1425 }, 1426 { .cmd = OVS_FLOW_CMD_DEL, 1427 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1428 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1429 .doit = ovs_flow_cmd_del 1430 }, 1431 { .cmd = OVS_FLOW_CMD_GET, 1432 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1433 .flags = 0, /* OK for unprivileged users. */ 1434 .doit = ovs_flow_cmd_get, 1435 .dumpit = ovs_flow_cmd_dump 1436 }, 1437 { .cmd = OVS_FLOW_CMD_SET, 1438 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1439 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1440 .doit = ovs_flow_cmd_set, 1441 }, 1442 }; 1443 1444 static struct genl_family dp_flow_genl_family __ro_after_init = { 1445 .hdrsize = sizeof(struct ovs_header), 1446 .name = OVS_FLOW_FAMILY, 1447 .version = OVS_FLOW_VERSION, 1448 .maxattr = OVS_FLOW_ATTR_MAX, 1449 .policy = flow_policy, 1450 .netnsok = true, 1451 .parallel_ops = true, 1452 .ops = dp_flow_genl_ops, 1453 .n_ops = ARRAY_SIZE(dp_flow_genl_ops), 1454 .mcgrps = &ovs_dp_flow_multicast_group, 1455 .n_mcgrps = 1, 1456 .module = THIS_MODULE, 1457 }; 1458 1459 static size_t ovs_dp_cmd_msg_size(void) 1460 { 1461 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header)); 1462 1463 msgsize += nla_total_size(IFNAMSIZ); 1464 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats)); 1465 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats)); 1466 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */ 1467 1468 return msgsize; 1469 } 1470 1471 /* Called with ovs_mutex. */ 1472 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, 1473 u32 portid, u32 seq, u32 flags, u8 cmd) 1474 { 1475 struct ovs_header *ovs_header; 1476 struct ovs_dp_stats dp_stats; 1477 struct ovs_dp_megaflow_stats dp_megaflow_stats; 1478 int err; 1479 1480 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family, 1481 flags, cmd); 1482 if (!ovs_header) 1483 goto error; 1484 1485 ovs_header->dp_ifindex = get_dpifindex(dp); 1486 1487 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); 1488 if (err) 1489 goto nla_put_failure; 1490 1491 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats); 1492 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), 1493 &dp_stats, OVS_DP_ATTR_PAD)) 1494 goto nla_put_failure; 1495 1496 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS, 1497 sizeof(struct ovs_dp_megaflow_stats), 1498 &dp_megaflow_stats, OVS_DP_ATTR_PAD)) 1499 goto nla_put_failure; 1500 1501 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features)) 1502 goto nla_put_failure; 1503 1504 genlmsg_end(skb, ovs_header); 1505 return 0; 1506 1507 nla_put_failure: 1508 genlmsg_cancel(skb, ovs_header); 1509 error: 1510 return -EMSGSIZE; 1511 } 1512 1513 static struct sk_buff *ovs_dp_cmd_alloc_info(void) 1514 { 1515 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL); 1516 } 1517 1518 /* Called with rcu_read_lock or ovs_mutex. */ 1519 static struct datapath *lookup_datapath(struct net *net, 1520 const struct ovs_header *ovs_header, 1521 struct nlattr *a[OVS_DP_ATTR_MAX + 1]) 1522 { 1523 struct datapath *dp; 1524 1525 if (!a[OVS_DP_ATTR_NAME]) 1526 dp = get_dp(net, ovs_header->dp_ifindex); 1527 else { 1528 struct vport *vport; 1529 1530 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); 1531 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; 1532 } 1533 return dp ? dp : ERR_PTR(-ENODEV); 1534 } 1535 1536 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info) 1537 { 1538 struct datapath *dp; 1539 1540 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1541 if (IS_ERR(dp)) 1542 return; 1543 1544 WARN(dp->user_features, "Dropping previously announced user features\n"); 1545 dp->user_features = 0; 1546 } 1547 1548 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support); 1549 1550 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[]) 1551 { 1552 u32 user_features = 0; 1553 1554 if (a[OVS_DP_ATTR_USER_FEATURES]) { 1555 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); 1556 1557 if (user_features & ~(OVS_DP_F_VPORT_PIDS | 1558 OVS_DP_F_UNALIGNED | 1559 OVS_DP_F_TC_RECIRC_SHARING)) 1560 return -EOPNOTSUPP; 1561 1562 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 1563 if (user_features & OVS_DP_F_TC_RECIRC_SHARING) 1564 return -EOPNOTSUPP; 1565 #endif 1566 } 1567 1568 dp->user_features = user_features; 1569 1570 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) 1571 static_branch_enable(&tc_recirc_sharing_support); 1572 else 1573 static_branch_disable(&tc_recirc_sharing_support); 1574 1575 return 0; 1576 } 1577 1578 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) 1579 { 1580 struct nlattr **a = info->attrs; 1581 struct vport_parms parms; 1582 struct sk_buff *reply; 1583 struct datapath *dp; 1584 struct vport *vport; 1585 struct ovs_net *ovs_net; 1586 int err, i; 1587 1588 err = -EINVAL; 1589 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) 1590 goto err; 1591 1592 reply = ovs_dp_cmd_alloc_info(); 1593 if (!reply) 1594 return -ENOMEM; 1595 1596 err = -ENOMEM; 1597 dp = kzalloc(sizeof(*dp), GFP_KERNEL); 1598 if (dp == NULL) 1599 goto err_free_reply; 1600 1601 ovs_dp_set_net(dp, sock_net(skb->sk)); 1602 1603 /* Allocate table. */ 1604 err = ovs_flow_tbl_init(&dp->table); 1605 if (err) 1606 goto err_free_dp; 1607 1608 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu); 1609 if (!dp->stats_percpu) { 1610 err = -ENOMEM; 1611 goto err_destroy_table; 1612 } 1613 1614 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS, 1615 sizeof(struct hlist_head), 1616 GFP_KERNEL); 1617 if (!dp->ports) { 1618 err = -ENOMEM; 1619 goto err_destroy_percpu; 1620 } 1621 1622 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 1623 INIT_HLIST_HEAD(&dp->ports[i]); 1624 1625 err = ovs_meters_init(dp); 1626 if (err) 1627 goto err_destroy_ports_array; 1628 1629 /* Set up our datapath device. */ 1630 parms.name = nla_data(a[OVS_DP_ATTR_NAME]); 1631 parms.type = OVS_VPORT_TYPE_INTERNAL; 1632 parms.options = NULL; 1633 parms.dp = dp; 1634 parms.port_no = OVSP_LOCAL; 1635 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID]; 1636 1637 err = ovs_dp_change(dp, a); 1638 if (err) 1639 goto err_destroy_meters; 1640 1641 /* So far only local changes have been made, now need the lock. */ 1642 ovs_lock(); 1643 1644 vport = new_vport(&parms); 1645 if (IS_ERR(vport)) { 1646 err = PTR_ERR(vport); 1647 if (err == -EBUSY) 1648 err = -EEXIST; 1649 1650 if (err == -EEXIST) { 1651 /* An outdated user space instance that does not understand 1652 * the concept of user_features has attempted to create a new 1653 * datapath and is likely to reuse it. Drop all user features. 1654 */ 1655 if (info->genlhdr->version < OVS_DP_VER_FEATURES) 1656 ovs_dp_reset_user_features(skb, info); 1657 } 1658 1659 goto err_destroy_meters; 1660 } 1661 1662 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1663 info->snd_seq, 0, OVS_DP_CMD_NEW); 1664 BUG_ON(err < 0); 1665 1666 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); 1667 list_add_tail_rcu(&dp->list_node, &ovs_net->dps); 1668 1669 ovs_unlock(); 1670 1671 ovs_notify(&dp_datapath_genl_family, reply, info); 1672 return 0; 1673 1674 err_destroy_meters: 1675 ovs_unlock(); 1676 ovs_meters_exit(dp); 1677 err_destroy_ports_array: 1678 kfree(dp->ports); 1679 err_destroy_percpu: 1680 free_percpu(dp->stats_percpu); 1681 err_destroy_table: 1682 ovs_flow_tbl_destroy(&dp->table); 1683 err_free_dp: 1684 kfree(dp); 1685 err_free_reply: 1686 kfree_skb(reply); 1687 err: 1688 return err; 1689 } 1690 1691 /* Called with ovs_mutex. */ 1692 static void __dp_destroy(struct datapath *dp) 1693 { 1694 int i; 1695 1696 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 1697 struct vport *vport; 1698 struct hlist_node *n; 1699 1700 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) 1701 if (vport->port_no != OVSP_LOCAL) 1702 ovs_dp_detach_port(vport); 1703 } 1704 1705 list_del_rcu(&dp->list_node); 1706 1707 /* OVSP_LOCAL is datapath internal port. We need to make sure that 1708 * all ports in datapath are destroyed first before freeing datapath. 1709 */ 1710 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); 1711 1712 /* RCU destroy the flow table */ 1713 call_rcu(&dp->rcu, destroy_dp_rcu); 1714 } 1715 1716 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) 1717 { 1718 struct sk_buff *reply; 1719 struct datapath *dp; 1720 int err; 1721 1722 reply = ovs_dp_cmd_alloc_info(); 1723 if (!reply) 1724 return -ENOMEM; 1725 1726 ovs_lock(); 1727 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1728 err = PTR_ERR(dp); 1729 if (IS_ERR(dp)) 1730 goto err_unlock_free; 1731 1732 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1733 info->snd_seq, 0, OVS_DP_CMD_DEL); 1734 BUG_ON(err < 0); 1735 1736 __dp_destroy(dp); 1737 ovs_unlock(); 1738 1739 ovs_notify(&dp_datapath_genl_family, reply, info); 1740 1741 return 0; 1742 1743 err_unlock_free: 1744 ovs_unlock(); 1745 kfree_skb(reply); 1746 return err; 1747 } 1748 1749 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) 1750 { 1751 struct sk_buff *reply; 1752 struct datapath *dp; 1753 int err; 1754 1755 reply = ovs_dp_cmd_alloc_info(); 1756 if (!reply) 1757 return -ENOMEM; 1758 1759 ovs_lock(); 1760 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1761 err = PTR_ERR(dp); 1762 if (IS_ERR(dp)) 1763 goto err_unlock_free; 1764 1765 err = ovs_dp_change(dp, info->attrs); 1766 if (err) 1767 goto err_unlock_free; 1768 1769 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1770 info->snd_seq, 0, OVS_DP_CMD_SET); 1771 BUG_ON(err < 0); 1772 1773 ovs_unlock(); 1774 ovs_notify(&dp_datapath_genl_family, reply, info); 1775 1776 return 0; 1777 1778 err_unlock_free: 1779 ovs_unlock(); 1780 kfree_skb(reply); 1781 return err; 1782 } 1783 1784 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) 1785 { 1786 struct sk_buff *reply; 1787 struct datapath *dp; 1788 int err; 1789 1790 reply = ovs_dp_cmd_alloc_info(); 1791 if (!reply) 1792 return -ENOMEM; 1793 1794 ovs_lock(); 1795 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1796 if (IS_ERR(dp)) { 1797 err = PTR_ERR(dp); 1798 goto err_unlock_free; 1799 } 1800 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1801 info->snd_seq, 0, OVS_DP_CMD_GET); 1802 BUG_ON(err < 0); 1803 ovs_unlock(); 1804 1805 return genlmsg_reply(reply, info); 1806 1807 err_unlock_free: 1808 ovs_unlock(); 1809 kfree_skb(reply); 1810 return err; 1811 } 1812 1813 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1814 { 1815 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); 1816 struct datapath *dp; 1817 int skip = cb->args[0]; 1818 int i = 0; 1819 1820 ovs_lock(); 1821 list_for_each_entry(dp, &ovs_net->dps, list_node) { 1822 if (i >= skip && 1823 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid, 1824 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1825 OVS_DP_CMD_GET) < 0) 1826 break; 1827 i++; 1828 } 1829 ovs_unlock(); 1830 1831 cb->args[0] = i; 1832 1833 return skb->len; 1834 } 1835 1836 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { 1837 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1838 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1839 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 }, 1840 }; 1841 1842 static const struct genl_ops dp_datapath_genl_ops[] = { 1843 { .cmd = OVS_DP_CMD_NEW, 1844 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1845 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1846 .doit = ovs_dp_cmd_new 1847 }, 1848 { .cmd = OVS_DP_CMD_DEL, 1849 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1850 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1851 .doit = ovs_dp_cmd_del 1852 }, 1853 { .cmd = OVS_DP_CMD_GET, 1854 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1855 .flags = 0, /* OK for unprivileged users. */ 1856 .doit = ovs_dp_cmd_get, 1857 .dumpit = ovs_dp_cmd_dump 1858 }, 1859 { .cmd = OVS_DP_CMD_SET, 1860 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1861 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1862 .doit = ovs_dp_cmd_set, 1863 }, 1864 }; 1865 1866 static struct genl_family dp_datapath_genl_family __ro_after_init = { 1867 .hdrsize = sizeof(struct ovs_header), 1868 .name = OVS_DATAPATH_FAMILY, 1869 .version = OVS_DATAPATH_VERSION, 1870 .maxattr = OVS_DP_ATTR_MAX, 1871 .policy = datapath_policy, 1872 .netnsok = true, 1873 .parallel_ops = true, 1874 .ops = dp_datapath_genl_ops, 1875 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops), 1876 .mcgrps = &ovs_dp_datapath_multicast_group, 1877 .n_mcgrps = 1, 1878 .module = THIS_MODULE, 1879 }; 1880 1881 /* Called with ovs_mutex or RCU read lock. */ 1882 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, 1883 struct net *net, u32 portid, u32 seq, 1884 u32 flags, u8 cmd, gfp_t gfp) 1885 { 1886 struct ovs_header *ovs_header; 1887 struct ovs_vport_stats vport_stats; 1888 int err; 1889 1890 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family, 1891 flags, cmd); 1892 if (!ovs_header) 1893 return -EMSGSIZE; 1894 1895 ovs_header->dp_ifindex = get_dpifindex(vport->dp); 1896 1897 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || 1898 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || 1899 nla_put_string(skb, OVS_VPORT_ATTR_NAME, 1900 ovs_vport_name(vport)) || 1901 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex)) 1902 goto nla_put_failure; 1903 1904 if (!net_eq(net, dev_net(vport->dev))) { 1905 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp); 1906 1907 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id)) 1908 goto nla_put_failure; 1909 } 1910 1911 ovs_vport_get_stats(vport, &vport_stats); 1912 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS, 1913 sizeof(struct ovs_vport_stats), &vport_stats, 1914 OVS_VPORT_ATTR_PAD)) 1915 goto nla_put_failure; 1916 1917 if (ovs_vport_get_upcall_portids(vport, skb)) 1918 goto nla_put_failure; 1919 1920 err = ovs_vport_get_options(vport, skb); 1921 if (err == -EMSGSIZE) 1922 goto error; 1923 1924 genlmsg_end(skb, ovs_header); 1925 return 0; 1926 1927 nla_put_failure: 1928 err = -EMSGSIZE; 1929 error: 1930 genlmsg_cancel(skb, ovs_header); 1931 return err; 1932 } 1933 1934 static struct sk_buff *ovs_vport_cmd_alloc_info(void) 1935 { 1936 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1937 } 1938 1939 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */ 1940 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net, 1941 u32 portid, u32 seq, u8 cmd) 1942 { 1943 struct sk_buff *skb; 1944 int retval; 1945 1946 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1947 if (!skb) 1948 return ERR_PTR(-ENOMEM); 1949 1950 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd, 1951 GFP_KERNEL); 1952 BUG_ON(retval < 0); 1953 1954 return skb; 1955 } 1956 1957 /* Called with ovs_mutex or RCU read lock. */ 1958 static struct vport *lookup_vport(struct net *net, 1959 const struct ovs_header *ovs_header, 1960 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) 1961 { 1962 struct datapath *dp; 1963 struct vport *vport; 1964 1965 if (a[OVS_VPORT_ATTR_IFINDEX]) 1966 return ERR_PTR(-EOPNOTSUPP); 1967 if (a[OVS_VPORT_ATTR_NAME]) { 1968 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); 1969 if (!vport) 1970 return ERR_PTR(-ENODEV); 1971 if (ovs_header->dp_ifindex && 1972 ovs_header->dp_ifindex != get_dpifindex(vport->dp)) 1973 return ERR_PTR(-ENODEV); 1974 return vport; 1975 } else if (a[OVS_VPORT_ATTR_PORT_NO]) { 1976 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1977 1978 if (port_no >= DP_MAX_PORTS) 1979 return ERR_PTR(-EFBIG); 1980 1981 dp = get_dp(net, ovs_header->dp_ifindex); 1982 if (!dp) 1983 return ERR_PTR(-ENODEV); 1984 1985 vport = ovs_vport_ovsl_rcu(dp, port_no); 1986 if (!vport) 1987 return ERR_PTR(-ENODEV); 1988 return vport; 1989 } else 1990 return ERR_PTR(-EINVAL); 1991 1992 } 1993 1994 static unsigned int ovs_get_max_headroom(struct datapath *dp) 1995 { 1996 unsigned int dev_headroom, max_headroom = 0; 1997 struct net_device *dev; 1998 struct vport *vport; 1999 int i; 2000 2001 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 2002 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 2003 dev = vport->dev; 2004 dev_headroom = netdev_get_fwd_headroom(dev); 2005 if (dev_headroom > max_headroom) 2006 max_headroom = dev_headroom; 2007 } 2008 } 2009 2010 return max_headroom; 2011 } 2012 2013 /* Called with ovs_mutex */ 2014 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom) 2015 { 2016 struct vport *vport; 2017 int i; 2018 2019 dp->max_headroom = new_headroom; 2020 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 2021 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) 2022 netdev_set_rx_headroom(vport->dev, new_headroom); 2023 } 2024 2025 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) 2026 { 2027 struct nlattr **a = info->attrs; 2028 struct ovs_header *ovs_header = info->userhdr; 2029 struct vport_parms parms; 2030 struct sk_buff *reply; 2031 struct vport *vport; 2032 struct datapath *dp; 2033 unsigned int new_headroom; 2034 u32 port_no; 2035 int err; 2036 2037 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || 2038 !a[OVS_VPORT_ATTR_UPCALL_PID]) 2039 return -EINVAL; 2040 if (a[OVS_VPORT_ATTR_IFINDEX]) 2041 return -EOPNOTSUPP; 2042 2043 port_no = a[OVS_VPORT_ATTR_PORT_NO] 2044 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0; 2045 if (port_no >= DP_MAX_PORTS) 2046 return -EFBIG; 2047 2048 reply = ovs_vport_cmd_alloc_info(); 2049 if (!reply) 2050 return -ENOMEM; 2051 2052 ovs_lock(); 2053 restart: 2054 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 2055 err = -ENODEV; 2056 if (!dp) 2057 goto exit_unlock_free; 2058 2059 if (port_no) { 2060 vport = ovs_vport_ovsl(dp, port_no); 2061 err = -EBUSY; 2062 if (vport) 2063 goto exit_unlock_free; 2064 } else { 2065 for (port_no = 1; ; port_no++) { 2066 if (port_no >= DP_MAX_PORTS) { 2067 err = -EFBIG; 2068 goto exit_unlock_free; 2069 } 2070 vport = ovs_vport_ovsl(dp, port_no); 2071 if (!vport) 2072 break; 2073 } 2074 } 2075 2076 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); 2077 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); 2078 parms.options = a[OVS_VPORT_ATTR_OPTIONS]; 2079 parms.dp = dp; 2080 parms.port_no = port_no; 2081 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2082 2083 vport = new_vport(&parms); 2084 err = PTR_ERR(vport); 2085 if (IS_ERR(vport)) { 2086 if (err == -EAGAIN) 2087 goto restart; 2088 goto exit_unlock_free; 2089 } 2090 2091 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2092 info->snd_portid, info->snd_seq, 0, 2093 OVS_VPORT_CMD_NEW, GFP_KERNEL); 2094 2095 new_headroom = netdev_get_fwd_headroom(vport->dev); 2096 2097 if (new_headroom > dp->max_headroom) 2098 ovs_update_headroom(dp, new_headroom); 2099 else 2100 netdev_set_rx_headroom(vport->dev, dp->max_headroom); 2101 2102 BUG_ON(err < 0); 2103 ovs_unlock(); 2104 2105 ovs_notify(&dp_vport_genl_family, reply, info); 2106 return 0; 2107 2108 exit_unlock_free: 2109 ovs_unlock(); 2110 kfree_skb(reply); 2111 return err; 2112 } 2113 2114 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) 2115 { 2116 struct nlattr **a = info->attrs; 2117 struct sk_buff *reply; 2118 struct vport *vport; 2119 int err; 2120 2121 reply = ovs_vport_cmd_alloc_info(); 2122 if (!reply) 2123 return -ENOMEM; 2124 2125 ovs_lock(); 2126 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2127 err = PTR_ERR(vport); 2128 if (IS_ERR(vport)) 2129 goto exit_unlock_free; 2130 2131 if (a[OVS_VPORT_ATTR_TYPE] && 2132 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) { 2133 err = -EINVAL; 2134 goto exit_unlock_free; 2135 } 2136 2137 if (a[OVS_VPORT_ATTR_OPTIONS]) { 2138 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); 2139 if (err) 2140 goto exit_unlock_free; 2141 } 2142 2143 2144 if (a[OVS_VPORT_ATTR_UPCALL_PID]) { 2145 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2146 2147 err = ovs_vport_set_upcall_portids(vport, ids); 2148 if (err) 2149 goto exit_unlock_free; 2150 } 2151 2152 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2153 info->snd_portid, info->snd_seq, 0, 2154 OVS_VPORT_CMD_SET, GFP_KERNEL); 2155 BUG_ON(err < 0); 2156 2157 ovs_unlock(); 2158 ovs_notify(&dp_vport_genl_family, reply, info); 2159 return 0; 2160 2161 exit_unlock_free: 2162 ovs_unlock(); 2163 kfree_skb(reply); 2164 return err; 2165 } 2166 2167 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) 2168 { 2169 bool update_headroom = false; 2170 struct nlattr **a = info->attrs; 2171 struct sk_buff *reply; 2172 struct datapath *dp; 2173 struct vport *vport; 2174 unsigned int new_headroom; 2175 int err; 2176 2177 reply = ovs_vport_cmd_alloc_info(); 2178 if (!reply) 2179 return -ENOMEM; 2180 2181 ovs_lock(); 2182 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2183 err = PTR_ERR(vport); 2184 if (IS_ERR(vport)) 2185 goto exit_unlock_free; 2186 2187 if (vport->port_no == OVSP_LOCAL) { 2188 err = -EINVAL; 2189 goto exit_unlock_free; 2190 } 2191 2192 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2193 info->snd_portid, info->snd_seq, 0, 2194 OVS_VPORT_CMD_DEL, GFP_KERNEL); 2195 BUG_ON(err < 0); 2196 2197 /* the vport deletion may trigger dp headroom update */ 2198 dp = vport->dp; 2199 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom) 2200 update_headroom = true; 2201 2202 netdev_reset_rx_headroom(vport->dev); 2203 ovs_dp_detach_port(vport); 2204 2205 if (update_headroom) { 2206 new_headroom = ovs_get_max_headroom(dp); 2207 2208 if (new_headroom < dp->max_headroom) 2209 ovs_update_headroom(dp, new_headroom); 2210 } 2211 ovs_unlock(); 2212 2213 ovs_notify(&dp_vport_genl_family, reply, info); 2214 return 0; 2215 2216 exit_unlock_free: 2217 ovs_unlock(); 2218 kfree_skb(reply); 2219 return err; 2220 } 2221 2222 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) 2223 { 2224 struct nlattr **a = info->attrs; 2225 struct ovs_header *ovs_header = info->userhdr; 2226 struct sk_buff *reply; 2227 struct vport *vport; 2228 int err; 2229 2230 reply = ovs_vport_cmd_alloc_info(); 2231 if (!reply) 2232 return -ENOMEM; 2233 2234 rcu_read_lock(); 2235 vport = lookup_vport(sock_net(skb->sk), ovs_header, a); 2236 err = PTR_ERR(vport); 2237 if (IS_ERR(vport)) 2238 goto exit_unlock_free; 2239 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2240 info->snd_portid, info->snd_seq, 0, 2241 OVS_VPORT_CMD_GET, GFP_ATOMIC); 2242 BUG_ON(err < 0); 2243 rcu_read_unlock(); 2244 2245 return genlmsg_reply(reply, info); 2246 2247 exit_unlock_free: 2248 rcu_read_unlock(); 2249 kfree_skb(reply); 2250 return err; 2251 } 2252 2253 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 2254 { 2255 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 2256 struct datapath *dp; 2257 int bucket = cb->args[0], skip = cb->args[1]; 2258 int i, j = 0; 2259 2260 rcu_read_lock(); 2261 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 2262 if (!dp) { 2263 rcu_read_unlock(); 2264 return -ENODEV; 2265 } 2266 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { 2267 struct vport *vport; 2268 2269 j = 0; 2270 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 2271 if (j >= skip && 2272 ovs_vport_cmd_fill_info(vport, skb, 2273 sock_net(skb->sk), 2274 NETLINK_CB(cb->skb).portid, 2275 cb->nlh->nlmsg_seq, 2276 NLM_F_MULTI, 2277 OVS_VPORT_CMD_GET, 2278 GFP_ATOMIC) < 0) 2279 goto out; 2280 2281 j++; 2282 } 2283 skip = 0; 2284 } 2285 out: 2286 rcu_read_unlock(); 2287 2288 cb->args[0] = i; 2289 cb->args[1] = j; 2290 2291 return skb->len; 2292 } 2293 2294 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { 2295 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 2296 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, 2297 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, 2298 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, 2299 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC }, 2300 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, 2301 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 }, 2302 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 }, 2303 }; 2304 2305 static const struct genl_ops dp_vport_genl_ops[] = { 2306 { .cmd = OVS_VPORT_CMD_NEW, 2307 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2308 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2309 .doit = ovs_vport_cmd_new 2310 }, 2311 { .cmd = OVS_VPORT_CMD_DEL, 2312 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2313 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2314 .doit = ovs_vport_cmd_del 2315 }, 2316 { .cmd = OVS_VPORT_CMD_GET, 2317 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2318 .flags = 0, /* OK for unprivileged users. */ 2319 .doit = ovs_vport_cmd_get, 2320 .dumpit = ovs_vport_cmd_dump 2321 }, 2322 { .cmd = OVS_VPORT_CMD_SET, 2323 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2324 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2325 .doit = ovs_vport_cmd_set, 2326 }, 2327 }; 2328 2329 struct genl_family dp_vport_genl_family __ro_after_init = { 2330 .hdrsize = sizeof(struct ovs_header), 2331 .name = OVS_VPORT_FAMILY, 2332 .version = OVS_VPORT_VERSION, 2333 .maxattr = OVS_VPORT_ATTR_MAX, 2334 .policy = vport_policy, 2335 .netnsok = true, 2336 .parallel_ops = true, 2337 .ops = dp_vport_genl_ops, 2338 .n_ops = ARRAY_SIZE(dp_vport_genl_ops), 2339 .mcgrps = &ovs_dp_vport_multicast_group, 2340 .n_mcgrps = 1, 2341 .module = THIS_MODULE, 2342 }; 2343 2344 static struct genl_family * const dp_genl_families[] = { 2345 &dp_datapath_genl_family, 2346 &dp_vport_genl_family, 2347 &dp_flow_genl_family, 2348 &dp_packet_genl_family, 2349 &dp_meter_genl_family, 2350 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 2351 &dp_ct_limit_genl_family, 2352 #endif 2353 }; 2354 2355 static void dp_unregister_genl(int n_families) 2356 { 2357 int i; 2358 2359 for (i = 0; i < n_families; i++) 2360 genl_unregister_family(dp_genl_families[i]); 2361 } 2362 2363 static int __init dp_register_genl(void) 2364 { 2365 int err; 2366 int i; 2367 2368 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { 2369 2370 err = genl_register_family(dp_genl_families[i]); 2371 if (err) 2372 goto error; 2373 } 2374 2375 return 0; 2376 2377 error: 2378 dp_unregister_genl(i); 2379 return err; 2380 } 2381 2382 static int __net_init ovs_init_net(struct net *net) 2383 { 2384 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2385 2386 INIT_LIST_HEAD(&ovs_net->dps); 2387 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq); 2388 return ovs_ct_init(net); 2389 } 2390 2391 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet, 2392 struct list_head *head) 2393 { 2394 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2395 struct datapath *dp; 2396 2397 list_for_each_entry(dp, &ovs_net->dps, list_node) { 2398 int i; 2399 2400 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 2401 struct vport *vport; 2402 2403 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) { 2404 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL) 2405 continue; 2406 2407 if (dev_net(vport->dev) == dnet) 2408 list_add(&vport->detach_list, head); 2409 } 2410 } 2411 } 2412 } 2413 2414 static void __net_exit ovs_exit_net(struct net *dnet) 2415 { 2416 struct datapath *dp, *dp_next; 2417 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id); 2418 struct vport *vport, *vport_next; 2419 struct net *net; 2420 LIST_HEAD(head); 2421 2422 ovs_ct_exit(dnet); 2423 ovs_lock(); 2424 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) 2425 __dp_destroy(dp); 2426 2427 down_read(&net_rwsem); 2428 for_each_net(net) 2429 list_vports_from_net(net, dnet, &head); 2430 up_read(&net_rwsem); 2431 2432 /* Detach all vports from given namespace. */ 2433 list_for_each_entry_safe(vport, vport_next, &head, detach_list) { 2434 list_del(&vport->detach_list); 2435 ovs_dp_detach_port(vport); 2436 } 2437 2438 ovs_unlock(); 2439 2440 cancel_work_sync(&ovs_net->dp_notify_work); 2441 } 2442 2443 static struct pernet_operations ovs_net_ops = { 2444 .init = ovs_init_net, 2445 .exit = ovs_exit_net, 2446 .id = &ovs_net_id, 2447 .size = sizeof(struct ovs_net), 2448 }; 2449 2450 static int __init dp_init(void) 2451 { 2452 int err; 2453 2454 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb)); 2455 2456 pr_info("Open vSwitch switching datapath\n"); 2457 2458 err = action_fifos_init(); 2459 if (err) 2460 goto error; 2461 2462 err = ovs_internal_dev_rtnl_link_register(); 2463 if (err) 2464 goto error_action_fifos_exit; 2465 2466 err = ovs_flow_init(); 2467 if (err) 2468 goto error_unreg_rtnl_link; 2469 2470 err = ovs_vport_init(); 2471 if (err) 2472 goto error_flow_exit; 2473 2474 err = register_pernet_device(&ovs_net_ops); 2475 if (err) 2476 goto error_vport_exit; 2477 2478 err = register_netdevice_notifier(&ovs_dp_device_notifier); 2479 if (err) 2480 goto error_netns_exit; 2481 2482 err = ovs_netdev_init(); 2483 if (err) 2484 goto error_unreg_notifier; 2485 2486 err = dp_register_genl(); 2487 if (err < 0) 2488 goto error_unreg_netdev; 2489 2490 return 0; 2491 2492 error_unreg_netdev: 2493 ovs_netdev_exit(); 2494 error_unreg_notifier: 2495 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2496 error_netns_exit: 2497 unregister_pernet_device(&ovs_net_ops); 2498 error_vport_exit: 2499 ovs_vport_exit(); 2500 error_flow_exit: 2501 ovs_flow_exit(); 2502 error_unreg_rtnl_link: 2503 ovs_internal_dev_rtnl_link_unregister(); 2504 error_action_fifos_exit: 2505 action_fifos_exit(); 2506 error: 2507 return err; 2508 } 2509 2510 static void dp_cleanup(void) 2511 { 2512 dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); 2513 ovs_netdev_exit(); 2514 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2515 unregister_pernet_device(&ovs_net_ops); 2516 rcu_barrier(); 2517 ovs_vport_exit(); 2518 ovs_flow_exit(); 2519 ovs_internal_dev_rtnl_link_unregister(); 2520 action_fifos_exit(); 2521 } 2522 2523 module_init(dp_init); 2524 module_exit(dp_cleanup); 2525 2526 MODULE_DESCRIPTION("Open vSwitch switching datapath"); 2527 MODULE_LICENSE("GPL"); 2528 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY); 2529 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY); 2530 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY); 2531 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY); 2532 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY); 2533 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY); 2534