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, skb_get_hash(skb), 231 &n_mask_hit); 232 if (unlikely(!flow)) { 233 struct dp_upcall_info upcall; 234 235 memset(&upcall, 0, sizeof(upcall)); 236 upcall.cmd = OVS_PACKET_CMD_MISS; 237 upcall.portid = ovs_vport_find_upcall_portid(p, skb); 238 upcall.mru = OVS_CB(skb)->mru; 239 error = ovs_dp_upcall(dp, skb, key, &upcall, 0); 240 if (unlikely(error)) 241 kfree_skb(skb); 242 else 243 consume_skb(skb); 244 stats_counter = &stats->n_missed; 245 goto out; 246 } 247 248 ovs_flow_stats_update(flow, key->tp.flags, skb); 249 sf_acts = rcu_dereference(flow->sf_acts); 250 error = ovs_execute_actions(dp, skb, sf_acts, key); 251 if (unlikely(error)) 252 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n", 253 ovs_dp_name(dp), error); 254 255 stats_counter = &stats->n_hit; 256 257 out: 258 /* Update datapath statistics. */ 259 u64_stats_update_begin(&stats->syncp); 260 (*stats_counter)++; 261 stats->n_mask_hit += n_mask_hit; 262 u64_stats_update_end(&stats->syncp); 263 } 264 265 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, 266 const struct sw_flow_key *key, 267 const struct dp_upcall_info *upcall_info, 268 uint32_t cutlen) 269 { 270 struct dp_stats_percpu *stats; 271 int err; 272 273 if (upcall_info->portid == 0) { 274 err = -ENOTCONN; 275 goto err; 276 } 277 278 if (!skb_is_gso(skb)) 279 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 280 else 281 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen); 282 if (err) 283 goto err; 284 285 return 0; 286 287 err: 288 stats = this_cpu_ptr(dp->stats_percpu); 289 290 u64_stats_update_begin(&stats->syncp); 291 stats->n_lost++; 292 u64_stats_update_end(&stats->syncp); 293 294 return err; 295 } 296 297 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb, 298 const struct sw_flow_key *key, 299 const struct dp_upcall_info *upcall_info, 300 uint32_t cutlen) 301 { 302 unsigned int gso_type = skb_shinfo(skb)->gso_type; 303 struct sw_flow_key later_key; 304 struct sk_buff *segs, *nskb; 305 int err; 306 307 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET); 308 segs = __skb_gso_segment(skb, NETIF_F_SG, false); 309 if (IS_ERR(segs)) 310 return PTR_ERR(segs); 311 if (segs == NULL) 312 return -EINVAL; 313 314 if (gso_type & SKB_GSO_UDP) { 315 /* The initial flow key extracted by ovs_flow_key_extract() 316 * in this case is for a first fragment, so we need to 317 * properly mark later fragments. 318 */ 319 later_key = *key; 320 later_key.ip.frag = OVS_FRAG_TYPE_LATER; 321 } 322 323 /* Queue all of the segments. */ 324 skb = segs; 325 do { 326 if (gso_type & SKB_GSO_UDP && skb != segs) 327 key = &later_key; 328 329 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 330 if (err) 331 break; 332 333 } while ((skb = skb->next)); 334 335 /* Free all of the segments. */ 336 skb = segs; 337 do { 338 nskb = skb->next; 339 if (err) 340 kfree_skb(skb); 341 else 342 consume_skb(skb); 343 } while ((skb = nskb)); 344 return err; 345 } 346 347 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info, 348 unsigned int hdrlen, int actions_attrlen) 349 { 350 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header)) 351 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */ 352 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */ 353 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */ 354 355 /* OVS_PACKET_ATTR_USERDATA */ 356 if (upcall_info->userdata) 357 size += NLA_ALIGN(upcall_info->userdata->nla_len); 358 359 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */ 360 if (upcall_info->egress_tun_info) 361 size += nla_total_size(ovs_tun_key_attr_size()); 362 363 /* OVS_PACKET_ATTR_ACTIONS */ 364 if (upcall_info->actions_len) 365 size += nla_total_size(actions_attrlen); 366 367 /* OVS_PACKET_ATTR_MRU */ 368 if (upcall_info->mru) 369 size += nla_total_size(sizeof(upcall_info->mru)); 370 371 return size; 372 } 373 374 static void pad_packet(struct datapath *dp, struct sk_buff *skb) 375 { 376 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) { 377 size_t plen = NLA_ALIGN(skb->len) - skb->len; 378 379 if (plen > 0) 380 skb_put_zero(skb, plen); 381 } 382 } 383 384 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb, 385 const struct sw_flow_key *key, 386 const struct dp_upcall_info *upcall_info, 387 uint32_t cutlen) 388 { 389 struct ovs_header *upcall; 390 struct sk_buff *nskb = NULL; 391 struct sk_buff *user_skb = NULL; /* to be queued to userspace */ 392 struct nlattr *nla; 393 size_t len; 394 unsigned int hlen; 395 int err, dp_ifindex; 396 397 dp_ifindex = get_dpifindex(dp); 398 if (!dp_ifindex) 399 return -ENODEV; 400 401 if (skb_vlan_tag_present(skb)) { 402 nskb = skb_clone(skb, GFP_ATOMIC); 403 if (!nskb) 404 return -ENOMEM; 405 406 nskb = __vlan_hwaccel_push_inside(nskb); 407 if (!nskb) 408 return -ENOMEM; 409 410 skb = nskb; 411 } 412 413 if (nla_attr_size(skb->len) > USHRT_MAX) { 414 err = -EFBIG; 415 goto out; 416 } 417 418 /* Complete checksum if needed */ 419 if (skb->ip_summed == CHECKSUM_PARTIAL && 420 (err = skb_csum_hwoffload_help(skb, 0))) 421 goto out; 422 423 /* Older versions of OVS user space enforce alignment of the last 424 * Netlink attribute to NLA_ALIGNTO which would require extensive 425 * padding logic. Only perform zerocopy if padding is not required. 426 */ 427 if (dp->user_features & OVS_DP_F_UNALIGNED) 428 hlen = skb_zerocopy_headlen(skb); 429 else 430 hlen = skb->len; 431 432 len = upcall_msg_size(upcall_info, hlen - cutlen, 433 OVS_CB(skb)->acts_origlen); 434 user_skb = genlmsg_new(len, GFP_ATOMIC); 435 if (!user_skb) { 436 err = -ENOMEM; 437 goto out; 438 } 439 440 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 441 0, upcall_info->cmd); 442 if (!upcall) { 443 err = -EINVAL; 444 goto out; 445 } 446 upcall->dp_ifindex = dp_ifindex; 447 448 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb); 449 if (err) 450 goto out; 451 452 if (upcall_info->userdata) 453 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA, 454 nla_len(upcall_info->userdata), 455 nla_data(upcall_info->userdata)); 456 457 if (upcall_info->egress_tun_info) { 458 nla = nla_nest_start_noflag(user_skb, 459 OVS_PACKET_ATTR_EGRESS_TUN_KEY); 460 if (!nla) { 461 err = -EMSGSIZE; 462 goto out; 463 } 464 err = ovs_nla_put_tunnel_info(user_skb, 465 upcall_info->egress_tun_info); 466 if (err) 467 goto out; 468 469 nla_nest_end(user_skb, nla); 470 } 471 472 if (upcall_info->actions_len) { 473 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS); 474 if (!nla) { 475 err = -EMSGSIZE; 476 goto out; 477 } 478 err = ovs_nla_put_actions(upcall_info->actions, 479 upcall_info->actions_len, 480 user_skb); 481 if (!err) 482 nla_nest_end(user_skb, nla); 483 else 484 nla_nest_cancel(user_skb, nla); 485 } 486 487 /* Add OVS_PACKET_ATTR_MRU */ 488 if (upcall_info->mru) { 489 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, 490 upcall_info->mru)) { 491 err = -ENOBUFS; 492 goto out; 493 } 494 pad_packet(dp, user_skb); 495 } 496 497 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */ 498 if (cutlen > 0) { 499 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, 500 skb->len)) { 501 err = -ENOBUFS; 502 goto out; 503 } 504 pad_packet(dp, user_skb); 505 } 506 507 /* Only reserve room for attribute header, packet data is added 508 * in skb_zerocopy() */ 509 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) { 510 err = -ENOBUFS; 511 goto out; 512 } 513 nla->nla_len = nla_attr_size(skb->len - cutlen); 514 515 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen); 516 if (err) 517 goto out; 518 519 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */ 520 pad_packet(dp, user_skb); 521 522 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; 523 524 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid); 525 user_skb = NULL; 526 out: 527 if (err) 528 skb_tx_error(skb); 529 kfree_skb(user_skb); 530 kfree_skb(nskb); 531 return err; 532 } 533 534 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) 535 { 536 struct ovs_header *ovs_header = info->userhdr; 537 struct net *net = sock_net(skb->sk); 538 struct nlattr **a = info->attrs; 539 struct sw_flow_actions *acts; 540 struct sk_buff *packet; 541 struct sw_flow *flow; 542 struct sw_flow_actions *sf_acts; 543 struct datapath *dp; 544 struct vport *input_vport; 545 u16 mru = 0; 546 int len; 547 int err; 548 bool log = !a[OVS_PACKET_ATTR_PROBE]; 549 550 err = -EINVAL; 551 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || 552 !a[OVS_PACKET_ATTR_ACTIONS]) 553 goto err; 554 555 len = nla_len(a[OVS_PACKET_ATTR_PACKET]); 556 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); 557 err = -ENOMEM; 558 if (!packet) 559 goto err; 560 skb_reserve(packet, NET_IP_ALIGN); 561 562 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len); 563 564 /* Set packet's mru */ 565 if (a[OVS_PACKET_ATTR_MRU]) { 566 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]); 567 packet->ignore_df = 1; 568 } 569 OVS_CB(packet)->mru = mru; 570 571 /* Build an sw_flow for sending this packet. */ 572 flow = ovs_flow_alloc(); 573 err = PTR_ERR(flow); 574 if (IS_ERR(flow)) 575 goto err_kfree_skb; 576 577 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY], 578 packet, &flow->key, log); 579 if (err) 580 goto err_flow_free; 581 582 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS], 583 &flow->key, &acts, log); 584 if (err) 585 goto err_flow_free; 586 587 rcu_assign_pointer(flow->sf_acts, acts); 588 packet->priority = flow->key.phy.priority; 589 packet->mark = flow->key.phy.skb_mark; 590 591 rcu_read_lock(); 592 dp = get_dp_rcu(net, ovs_header->dp_ifindex); 593 err = -ENODEV; 594 if (!dp) 595 goto err_unlock; 596 597 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port); 598 if (!input_vport) 599 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL); 600 601 if (!input_vport) 602 goto err_unlock; 603 604 packet->dev = input_vport->dev; 605 OVS_CB(packet)->input_vport = input_vport; 606 sf_acts = rcu_dereference(flow->sf_acts); 607 608 local_bh_disable(); 609 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key); 610 local_bh_enable(); 611 rcu_read_unlock(); 612 613 ovs_flow_free(flow, false); 614 return err; 615 616 err_unlock: 617 rcu_read_unlock(); 618 err_flow_free: 619 ovs_flow_free(flow, false); 620 err_kfree_skb: 621 kfree_skb(packet); 622 err: 623 return err; 624 } 625 626 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { 627 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN }, 628 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, 629 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, 630 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG }, 631 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 }, 632 }; 633 634 static const struct genl_ops dp_packet_genl_ops[] = { 635 { .cmd = OVS_PACKET_CMD_EXECUTE, 636 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 637 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 638 .doit = ovs_packet_cmd_execute 639 } 640 }; 641 642 static struct genl_family dp_packet_genl_family __ro_after_init = { 643 .hdrsize = sizeof(struct ovs_header), 644 .name = OVS_PACKET_FAMILY, 645 .version = OVS_PACKET_VERSION, 646 .maxattr = OVS_PACKET_ATTR_MAX, 647 .policy = packet_policy, 648 .netnsok = true, 649 .parallel_ops = true, 650 .ops = dp_packet_genl_ops, 651 .n_ops = ARRAY_SIZE(dp_packet_genl_ops), 652 .module = THIS_MODULE, 653 }; 654 655 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats, 656 struct ovs_dp_megaflow_stats *mega_stats) 657 { 658 int i; 659 660 memset(mega_stats, 0, sizeof(*mega_stats)); 661 662 stats->n_flows = ovs_flow_tbl_count(&dp->table); 663 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table); 664 665 stats->n_hit = stats->n_missed = stats->n_lost = 0; 666 667 for_each_possible_cpu(i) { 668 const struct dp_stats_percpu *percpu_stats; 669 struct dp_stats_percpu local_stats; 670 unsigned int start; 671 672 percpu_stats = per_cpu_ptr(dp->stats_percpu, i); 673 674 do { 675 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp); 676 local_stats = *percpu_stats; 677 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start)); 678 679 stats->n_hit += local_stats.n_hit; 680 stats->n_missed += local_stats.n_missed; 681 stats->n_lost += local_stats.n_lost; 682 mega_stats->n_mask_hit += local_stats.n_mask_hit; 683 } 684 } 685 686 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags) 687 { 688 return ovs_identifier_is_ufid(sfid) && 689 !(ufid_flags & OVS_UFID_F_OMIT_KEY); 690 } 691 692 static bool should_fill_mask(uint32_t ufid_flags) 693 { 694 return !(ufid_flags & OVS_UFID_F_OMIT_MASK); 695 } 696 697 static bool should_fill_actions(uint32_t ufid_flags) 698 { 699 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS); 700 } 701 702 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts, 703 const struct sw_flow_id *sfid, 704 uint32_t ufid_flags) 705 { 706 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header)); 707 708 /* OVS_FLOW_ATTR_UFID */ 709 if (sfid && ovs_identifier_is_ufid(sfid)) 710 len += nla_total_size(sfid->ufid_len); 711 712 /* OVS_FLOW_ATTR_KEY */ 713 if (!sfid || should_fill_key(sfid, ufid_flags)) 714 len += nla_total_size(ovs_key_attr_size()); 715 716 /* OVS_FLOW_ATTR_MASK */ 717 if (should_fill_mask(ufid_flags)) 718 len += nla_total_size(ovs_key_attr_size()); 719 720 /* OVS_FLOW_ATTR_ACTIONS */ 721 if (should_fill_actions(ufid_flags)) 722 len += nla_total_size(acts->orig_len); 723 724 return len 725 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */ 726 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */ 727 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */ 728 } 729 730 /* Called with ovs_mutex or RCU read lock. */ 731 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow, 732 struct sk_buff *skb) 733 { 734 struct ovs_flow_stats stats; 735 __be16 tcp_flags; 736 unsigned long used; 737 738 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags); 739 740 if (used && 741 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used), 742 OVS_FLOW_ATTR_PAD)) 743 return -EMSGSIZE; 744 745 if (stats.n_packets && 746 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS, 747 sizeof(struct ovs_flow_stats), &stats, 748 OVS_FLOW_ATTR_PAD)) 749 return -EMSGSIZE; 750 751 if ((u8)ntohs(tcp_flags) && 752 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags))) 753 return -EMSGSIZE; 754 755 return 0; 756 } 757 758 /* Called with ovs_mutex or RCU read lock. */ 759 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow, 760 struct sk_buff *skb, int skb_orig_len) 761 { 762 struct nlattr *start; 763 int err; 764 765 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if 766 * this is the first flow to be dumped into 'skb'. This is unusual for 767 * Netlink but individual action lists can be longer than 768 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. 769 * The userspace caller can always fetch the actions separately if it 770 * really wants them. (Most userspace callers in fact don't care.) 771 * 772 * This can only fail for dump operations because the skb is always 773 * properly sized for single flows. 774 */ 775 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS); 776 if (start) { 777 const struct sw_flow_actions *sf_acts; 778 779 sf_acts = rcu_dereference_ovsl(flow->sf_acts); 780 err = ovs_nla_put_actions(sf_acts->actions, 781 sf_acts->actions_len, skb); 782 783 if (!err) 784 nla_nest_end(skb, start); 785 else { 786 if (skb_orig_len) 787 return err; 788 789 nla_nest_cancel(skb, start); 790 } 791 } else if (skb_orig_len) { 792 return -EMSGSIZE; 793 } 794 795 return 0; 796 } 797 798 /* Called with ovs_mutex or RCU read lock. */ 799 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex, 800 struct sk_buff *skb, u32 portid, 801 u32 seq, u32 flags, u8 cmd, u32 ufid_flags) 802 { 803 const int skb_orig_len = skb->len; 804 struct ovs_header *ovs_header; 805 int err; 806 807 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, 808 flags, cmd); 809 if (!ovs_header) 810 return -EMSGSIZE; 811 812 ovs_header->dp_ifindex = dp_ifindex; 813 814 err = ovs_nla_put_identifier(flow, skb); 815 if (err) 816 goto error; 817 818 if (should_fill_key(&flow->id, ufid_flags)) { 819 err = ovs_nla_put_masked_key(flow, skb); 820 if (err) 821 goto error; 822 } 823 824 if (should_fill_mask(ufid_flags)) { 825 err = ovs_nla_put_mask(flow, skb); 826 if (err) 827 goto error; 828 } 829 830 err = ovs_flow_cmd_fill_stats(flow, skb); 831 if (err) 832 goto error; 833 834 if (should_fill_actions(ufid_flags)) { 835 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len); 836 if (err) 837 goto error; 838 } 839 840 genlmsg_end(skb, ovs_header); 841 return 0; 842 843 error: 844 genlmsg_cancel(skb, ovs_header); 845 return err; 846 } 847 848 /* May not be called with RCU read lock. */ 849 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts, 850 const struct sw_flow_id *sfid, 851 struct genl_info *info, 852 bool always, 853 uint32_t ufid_flags) 854 { 855 struct sk_buff *skb; 856 size_t len; 857 858 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0)) 859 return NULL; 860 861 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags); 862 skb = genlmsg_new(len, GFP_KERNEL); 863 if (!skb) 864 return ERR_PTR(-ENOMEM); 865 866 return skb; 867 } 868 869 /* Called with ovs_mutex. */ 870 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow, 871 int dp_ifindex, 872 struct genl_info *info, u8 cmd, 873 bool always, u32 ufid_flags) 874 { 875 struct sk_buff *skb; 876 int retval; 877 878 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), 879 &flow->id, info, always, ufid_flags); 880 if (IS_ERR_OR_NULL(skb)) 881 return skb; 882 883 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb, 884 info->snd_portid, info->snd_seq, 0, 885 cmd, ufid_flags); 886 BUG_ON(retval < 0); 887 return skb; 888 } 889 890 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info) 891 { 892 struct net *net = sock_net(skb->sk); 893 struct nlattr **a = info->attrs; 894 struct ovs_header *ovs_header = info->userhdr; 895 struct sw_flow *flow = NULL, *new_flow; 896 struct sw_flow_mask mask; 897 struct sk_buff *reply; 898 struct datapath *dp; 899 struct sw_flow_actions *acts; 900 struct sw_flow_match match; 901 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 902 int error; 903 bool log = !a[OVS_FLOW_ATTR_PROBE]; 904 905 /* Must have key and actions. */ 906 error = -EINVAL; 907 if (!a[OVS_FLOW_ATTR_KEY]) { 908 OVS_NLERR(log, "Flow key attr not present in new flow."); 909 goto error; 910 } 911 if (!a[OVS_FLOW_ATTR_ACTIONS]) { 912 OVS_NLERR(log, "Flow actions attr not present in new flow."); 913 goto error; 914 } 915 916 /* Most of the time we need to allocate a new flow, do it before 917 * locking. 918 */ 919 new_flow = ovs_flow_alloc(); 920 if (IS_ERR(new_flow)) { 921 error = PTR_ERR(new_flow); 922 goto error; 923 } 924 925 /* Extract key. */ 926 ovs_match_init(&match, &new_flow->key, false, &mask); 927 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 928 a[OVS_FLOW_ATTR_MASK], log); 929 if (error) 930 goto err_kfree_flow; 931 932 /* Extract flow identifier. */ 933 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID], 934 &new_flow->key, log); 935 if (error) 936 goto err_kfree_flow; 937 938 /* unmasked key is needed to match when ufid is not used. */ 939 if (ovs_identifier_is_key(&new_flow->id)) 940 match.key = new_flow->id.unmasked_key; 941 942 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask); 943 944 /* Validate actions. */ 945 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS], 946 &new_flow->key, &acts, log); 947 if (error) { 948 OVS_NLERR(log, "Flow actions may not be safe on all matching packets."); 949 goto err_kfree_flow; 950 } 951 952 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false, 953 ufid_flags); 954 if (IS_ERR(reply)) { 955 error = PTR_ERR(reply); 956 goto err_kfree_acts; 957 } 958 959 ovs_lock(); 960 dp = get_dp(net, ovs_header->dp_ifindex); 961 if (unlikely(!dp)) { 962 error = -ENODEV; 963 goto err_unlock_ovs; 964 } 965 966 /* Check if this is a duplicate flow */ 967 if (ovs_identifier_is_ufid(&new_flow->id)) 968 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id); 969 if (!flow) 970 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key); 971 if (likely(!flow)) { 972 rcu_assign_pointer(new_flow->sf_acts, acts); 973 974 /* Put flow in bucket. */ 975 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask); 976 if (unlikely(error)) { 977 acts = NULL; 978 goto err_unlock_ovs; 979 } 980 981 if (unlikely(reply)) { 982 error = ovs_flow_cmd_fill_info(new_flow, 983 ovs_header->dp_ifindex, 984 reply, info->snd_portid, 985 info->snd_seq, 0, 986 OVS_FLOW_CMD_NEW, 987 ufid_flags); 988 BUG_ON(error < 0); 989 } 990 ovs_unlock(); 991 } else { 992 struct sw_flow_actions *old_acts; 993 994 /* Bail out if we're not allowed to modify an existing flow. 995 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL 996 * because Generic Netlink treats the latter as a dump 997 * request. We also accept NLM_F_EXCL in case that bug ever 998 * gets fixed. 999 */ 1000 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE 1001 | NLM_F_EXCL))) { 1002 error = -EEXIST; 1003 goto err_unlock_ovs; 1004 } 1005 /* The flow identifier has to be the same for flow updates. 1006 * Look for any overlapping flow. 1007 */ 1008 if (unlikely(!ovs_flow_cmp(flow, &match))) { 1009 if (ovs_identifier_is_key(&flow->id)) 1010 flow = ovs_flow_tbl_lookup_exact(&dp->table, 1011 &match); 1012 else /* UFID matches but key is different */ 1013 flow = NULL; 1014 if (!flow) { 1015 error = -ENOENT; 1016 goto err_unlock_ovs; 1017 } 1018 } 1019 /* Update actions. */ 1020 old_acts = ovsl_dereference(flow->sf_acts); 1021 rcu_assign_pointer(flow->sf_acts, acts); 1022 1023 if (unlikely(reply)) { 1024 error = ovs_flow_cmd_fill_info(flow, 1025 ovs_header->dp_ifindex, 1026 reply, info->snd_portid, 1027 info->snd_seq, 0, 1028 OVS_FLOW_CMD_NEW, 1029 ufid_flags); 1030 BUG_ON(error < 0); 1031 } 1032 ovs_unlock(); 1033 1034 ovs_nla_free_flow_actions_rcu(old_acts); 1035 ovs_flow_free(new_flow, false); 1036 } 1037 1038 if (reply) 1039 ovs_notify(&dp_flow_genl_family, reply, info); 1040 return 0; 1041 1042 err_unlock_ovs: 1043 ovs_unlock(); 1044 kfree_skb(reply); 1045 err_kfree_acts: 1046 ovs_nla_free_flow_actions(acts); 1047 err_kfree_flow: 1048 ovs_flow_free(new_flow, false); 1049 error: 1050 return error; 1051 } 1052 1053 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */ 1054 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net, 1055 const struct nlattr *a, 1056 const struct sw_flow_key *key, 1057 const struct sw_flow_mask *mask, 1058 bool log) 1059 { 1060 struct sw_flow_actions *acts; 1061 struct sw_flow_key masked_key; 1062 int error; 1063 1064 ovs_flow_mask_key(&masked_key, key, true, mask); 1065 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log); 1066 if (error) { 1067 OVS_NLERR(log, 1068 "Actions may not be safe on all matching packets"); 1069 return ERR_PTR(error); 1070 } 1071 1072 return acts; 1073 } 1074 1075 /* Factor out match-init and action-copy to avoid 1076 * "Wframe-larger-than=1024" warning. Because mask is only 1077 * used to get actions, we new a function to save some 1078 * stack space. 1079 * 1080 * If there are not key and action attrs, we return 0 1081 * directly. In the case, the caller will also not use the 1082 * match as before. If there is action attr, we try to get 1083 * actions and save them to *acts. Before returning from 1084 * the function, we reset the match->mask pointer. Because 1085 * we should not to return match object with dangling reference 1086 * to mask. 1087 * */ 1088 static noinline_for_stack int 1089 ovs_nla_init_match_and_action(struct net *net, 1090 struct sw_flow_match *match, 1091 struct sw_flow_key *key, 1092 struct nlattr **a, 1093 struct sw_flow_actions **acts, 1094 bool log) 1095 { 1096 struct sw_flow_mask mask; 1097 int error = 0; 1098 1099 if (a[OVS_FLOW_ATTR_KEY]) { 1100 ovs_match_init(match, key, true, &mask); 1101 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY], 1102 a[OVS_FLOW_ATTR_MASK], log); 1103 if (error) 1104 goto error; 1105 } 1106 1107 if (a[OVS_FLOW_ATTR_ACTIONS]) { 1108 if (!a[OVS_FLOW_ATTR_KEY]) { 1109 OVS_NLERR(log, 1110 "Flow key attribute not present in set flow."); 1111 error = -EINVAL; 1112 goto error; 1113 } 1114 1115 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key, 1116 &mask, log); 1117 if (IS_ERR(*acts)) { 1118 error = PTR_ERR(*acts); 1119 goto error; 1120 } 1121 } 1122 1123 /* On success, error is 0. */ 1124 error: 1125 match->mask = NULL; 1126 return error; 1127 } 1128 1129 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info) 1130 { 1131 struct net *net = sock_net(skb->sk); 1132 struct nlattr **a = info->attrs; 1133 struct ovs_header *ovs_header = info->userhdr; 1134 struct sw_flow_key key; 1135 struct sw_flow *flow; 1136 struct sk_buff *reply = NULL; 1137 struct datapath *dp; 1138 struct sw_flow_actions *old_acts = NULL, *acts = NULL; 1139 struct sw_flow_match match; 1140 struct sw_flow_id sfid; 1141 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1142 int error = 0; 1143 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1144 bool ufid_present; 1145 1146 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log); 1147 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) { 1148 OVS_NLERR(log, 1149 "Flow set message rejected, Key attribute missing."); 1150 return -EINVAL; 1151 } 1152 1153 error = ovs_nla_init_match_and_action(net, &match, &key, a, 1154 &acts, log); 1155 if (error) 1156 goto error; 1157 1158 if (acts) { 1159 /* Can allocate before locking if have acts. */ 1160 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false, 1161 ufid_flags); 1162 if (IS_ERR(reply)) { 1163 error = PTR_ERR(reply); 1164 goto err_kfree_acts; 1165 } 1166 } 1167 1168 ovs_lock(); 1169 dp = get_dp(net, ovs_header->dp_ifindex); 1170 if (unlikely(!dp)) { 1171 error = -ENODEV; 1172 goto err_unlock_ovs; 1173 } 1174 /* Check that the flow exists. */ 1175 if (ufid_present) 1176 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid); 1177 else 1178 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1179 if (unlikely(!flow)) { 1180 error = -ENOENT; 1181 goto err_unlock_ovs; 1182 } 1183 1184 /* Update actions, if present. */ 1185 if (likely(acts)) { 1186 old_acts = ovsl_dereference(flow->sf_acts); 1187 rcu_assign_pointer(flow->sf_acts, acts); 1188 1189 if (unlikely(reply)) { 1190 error = ovs_flow_cmd_fill_info(flow, 1191 ovs_header->dp_ifindex, 1192 reply, info->snd_portid, 1193 info->snd_seq, 0, 1194 OVS_FLOW_CMD_SET, 1195 ufid_flags); 1196 BUG_ON(error < 0); 1197 } 1198 } else { 1199 /* Could not alloc without acts before locking. */ 1200 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, 1201 info, OVS_FLOW_CMD_SET, false, 1202 ufid_flags); 1203 1204 if (IS_ERR(reply)) { 1205 error = PTR_ERR(reply); 1206 goto err_unlock_ovs; 1207 } 1208 } 1209 1210 /* Clear stats. */ 1211 if (a[OVS_FLOW_ATTR_CLEAR]) 1212 ovs_flow_stats_clear(flow); 1213 ovs_unlock(); 1214 1215 if (reply) 1216 ovs_notify(&dp_flow_genl_family, reply, info); 1217 if (old_acts) 1218 ovs_nla_free_flow_actions_rcu(old_acts); 1219 1220 return 0; 1221 1222 err_unlock_ovs: 1223 ovs_unlock(); 1224 kfree_skb(reply); 1225 err_kfree_acts: 1226 ovs_nla_free_flow_actions(acts); 1227 error: 1228 return error; 1229 } 1230 1231 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) 1232 { 1233 struct nlattr **a = info->attrs; 1234 struct ovs_header *ovs_header = info->userhdr; 1235 struct net *net = sock_net(skb->sk); 1236 struct sw_flow_key key; 1237 struct sk_buff *reply; 1238 struct sw_flow *flow; 1239 struct datapath *dp; 1240 struct sw_flow_match match; 1241 struct sw_flow_id ufid; 1242 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1243 int err = 0; 1244 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1245 bool ufid_present; 1246 1247 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1248 if (a[OVS_FLOW_ATTR_KEY]) { 1249 ovs_match_init(&match, &key, true, NULL); 1250 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL, 1251 log); 1252 } else if (!ufid_present) { 1253 OVS_NLERR(log, 1254 "Flow get message rejected, Key attribute missing."); 1255 err = -EINVAL; 1256 } 1257 if (err) 1258 return err; 1259 1260 ovs_lock(); 1261 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1262 if (!dp) { 1263 err = -ENODEV; 1264 goto unlock; 1265 } 1266 1267 if (ufid_present) 1268 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1269 else 1270 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1271 if (!flow) { 1272 err = -ENOENT; 1273 goto unlock; 1274 } 1275 1276 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info, 1277 OVS_FLOW_CMD_GET, true, ufid_flags); 1278 if (IS_ERR(reply)) { 1279 err = PTR_ERR(reply); 1280 goto unlock; 1281 } 1282 1283 ovs_unlock(); 1284 return genlmsg_reply(reply, info); 1285 unlock: 1286 ovs_unlock(); 1287 return err; 1288 } 1289 1290 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) 1291 { 1292 struct nlattr **a = info->attrs; 1293 struct ovs_header *ovs_header = info->userhdr; 1294 struct net *net = sock_net(skb->sk); 1295 struct sw_flow_key key; 1296 struct sk_buff *reply; 1297 struct sw_flow *flow = NULL; 1298 struct datapath *dp; 1299 struct sw_flow_match match; 1300 struct sw_flow_id ufid; 1301 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1302 int err; 1303 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1304 bool ufid_present; 1305 1306 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1307 if (a[OVS_FLOW_ATTR_KEY]) { 1308 ovs_match_init(&match, &key, true, NULL); 1309 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 1310 NULL, log); 1311 if (unlikely(err)) 1312 return err; 1313 } 1314 1315 ovs_lock(); 1316 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1317 if (unlikely(!dp)) { 1318 err = -ENODEV; 1319 goto unlock; 1320 } 1321 1322 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) { 1323 err = ovs_flow_tbl_flush(&dp->table); 1324 goto unlock; 1325 } 1326 1327 if (ufid_present) 1328 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1329 else 1330 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1331 if (unlikely(!flow)) { 1332 err = -ENOENT; 1333 goto unlock; 1334 } 1335 1336 ovs_flow_tbl_remove(&dp->table, flow); 1337 ovs_unlock(); 1338 1339 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts, 1340 &flow->id, info, false, ufid_flags); 1341 if (likely(reply)) { 1342 if (!IS_ERR(reply)) { 1343 rcu_read_lock(); /*To keep RCU checker happy. */ 1344 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, 1345 reply, info->snd_portid, 1346 info->snd_seq, 0, 1347 OVS_FLOW_CMD_DEL, 1348 ufid_flags); 1349 rcu_read_unlock(); 1350 BUG_ON(err < 0); 1351 1352 ovs_notify(&dp_flow_genl_family, reply, info); 1353 } else { 1354 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply)); 1355 } 1356 } 1357 1358 ovs_flow_free(flow, true); 1359 return 0; 1360 unlock: 1361 ovs_unlock(); 1362 return err; 1363 } 1364 1365 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1366 { 1367 struct nlattr *a[__OVS_FLOW_ATTR_MAX]; 1368 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1369 struct table_instance *ti; 1370 struct datapath *dp; 1371 u32 ufid_flags; 1372 int err; 1373 1374 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a, 1375 OVS_FLOW_ATTR_MAX, flow_policy, NULL); 1376 if (err) 1377 return err; 1378 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1379 1380 rcu_read_lock(); 1381 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 1382 if (!dp) { 1383 rcu_read_unlock(); 1384 return -ENODEV; 1385 } 1386 1387 ti = rcu_dereference(dp->table.ti); 1388 for (;;) { 1389 struct sw_flow *flow; 1390 u32 bucket, obj; 1391 1392 bucket = cb->args[0]; 1393 obj = cb->args[1]; 1394 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj); 1395 if (!flow) 1396 break; 1397 1398 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb, 1399 NETLINK_CB(cb->skb).portid, 1400 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1401 OVS_FLOW_CMD_GET, ufid_flags) < 0) 1402 break; 1403 1404 cb->args[0] = bucket; 1405 cb->args[1] = obj; 1406 } 1407 rcu_read_unlock(); 1408 return skb->len; 1409 } 1410 1411 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { 1412 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, 1413 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED }, 1414 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, 1415 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, 1416 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG }, 1417 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 }, 1418 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 }, 1419 }; 1420 1421 static const struct genl_ops dp_flow_genl_ops[] = { 1422 { .cmd = OVS_FLOW_CMD_NEW, 1423 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1424 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1425 .doit = ovs_flow_cmd_new 1426 }, 1427 { .cmd = OVS_FLOW_CMD_DEL, 1428 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1429 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1430 .doit = ovs_flow_cmd_del 1431 }, 1432 { .cmd = OVS_FLOW_CMD_GET, 1433 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1434 .flags = 0, /* OK for unprivileged users. */ 1435 .doit = ovs_flow_cmd_get, 1436 .dumpit = ovs_flow_cmd_dump 1437 }, 1438 { .cmd = OVS_FLOW_CMD_SET, 1439 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1440 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1441 .doit = ovs_flow_cmd_set, 1442 }, 1443 }; 1444 1445 static struct genl_family dp_flow_genl_family __ro_after_init = { 1446 .hdrsize = sizeof(struct ovs_header), 1447 .name = OVS_FLOW_FAMILY, 1448 .version = OVS_FLOW_VERSION, 1449 .maxattr = OVS_FLOW_ATTR_MAX, 1450 .policy = flow_policy, 1451 .netnsok = true, 1452 .parallel_ops = true, 1453 .ops = dp_flow_genl_ops, 1454 .n_ops = ARRAY_SIZE(dp_flow_genl_ops), 1455 .mcgrps = &ovs_dp_flow_multicast_group, 1456 .n_mcgrps = 1, 1457 .module = THIS_MODULE, 1458 }; 1459 1460 static size_t ovs_dp_cmd_msg_size(void) 1461 { 1462 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header)); 1463 1464 msgsize += nla_total_size(IFNAMSIZ); 1465 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats)); 1466 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats)); 1467 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */ 1468 1469 return msgsize; 1470 } 1471 1472 /* Called with ovs_mutex. */ 1473 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, 1474 u32 portid, u32 seq, u32 flags, u8 cmd) 1475 { 1476 struct ovs_header *ovs_header; 1477 struct ovs_dp_stats dp_stats; 1478 struct ovs_dp_megaflow_stats dp_megaflow_stats; 1479 int err; 1480 1481 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family, 1482 flags, cmd); 1483 if (!ovs_header) 1484 goto error; 1485 1486 ovs_header->dp_ifindex = get_dpifindex(dp); 1487 1488 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); 1489 if (err) 1490 goto nla_put_failure; 1491 1492 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats); 1493 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), 1494 &dp_stats, OVS_DP_ATTR_PAD)) 1495 goto nla_put_failure; 1496 1497 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS, 1498 sizeof(struct ovs_dp_megaflow_stats), 1499 &dp_megaflow_stats, OVS_DP_ATTR_PAD)) 1500 goto nla_put_failure; 1501 1502 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features)) 1503 goto nla_put_failure; 1504 1505 genlmsg_end(skb, ovs_header); 1506 return 0; 1507 1508 nla_put_failure: 1509 genlmsg_cancel(skb, ovs_header); 1510 error: 1511 return -EMSGSIZE; 1512 } 1513 1514 static struct sk_buff *ovs_dp_cmd_alloc_info(void) 1515 { 1516 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL); 1517 } 1518 1519 /* Called with rcu_read_lock or ovs_mutex. */ 1520 static struct datapath *lookup_datapath(struct net *net, 1521 const struct ovs_header *ovs_header, 1522 struct nlattr *a[OVS_DP_ATTR_MAX + 1]) 1523 { 1524 struct datapath *dp; 1525 1526 if (!a[OVS_DP_ATTR_NAME]) 1527 dp = get_dp(net, ovs_header->dp_ifindex); 1528 else { 1529 struct vport *vport; 1530 1531 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); 1532 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; 1533 } 1534 return dp ? dp : ERR_PTR(-ENODEV); 1535 } 1536 1537 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info) 1538 { 1539 struct datapath *dp; 1540 1541 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1542 if (IS_ERR(dp)) 1543 return; 1544 1545 WARN(dp->user_features, "Dropping previously announced user features\n"); 1546 dp->user_features = 0; 1547 } 1548 1549 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support); 1550 1551 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[]) 1552 { 1553 u32 user_features = 0; 1554 1555 if (a[OVS_DP_ATTR_USER_FEATURES]) { 1556 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); 1557 1558 if (user_features & ~(OVS_DP_F_VPORT_PIDS | 1559 OVS_DP_F_UNALIGNED | 1560 OVS_DP_F_TC_RECIRC_SHARING)) 1561 return -EOPNOTSUPP; 1562 1563 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 1564 if (user_features & OVS_DP_F_TC_RECIRC_SHARING) 1565 return -EOPNOTSUPP; 1566 #endif 1567 } 1568 1569 dp->user_features = user_features; 1570 1571 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) 1572 static_branch_enable(&tc_recirc_sharing_support); 1573 else 1574 static_branch_disable(&tc_recirc_sharing_support); 1575 1576 return 0; 1577 } 1578 1579 static int ovs_dp_stats_init(struct datapath *dp) 1580 { 1581 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu); 1582 if (!dp->stats_percpu) 1583 return -ENOMEM; 1584 1585 return 0; 1586 } 1587 1588 static int ovs_dp_vport_init(struct datapath *dp) 1589 { 1590 int i; 1591 1592 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS, 1593 sizeof(struct hlist_head), 1594 GFP_KERNEL); 1595 if (!dp->ports) 1596 return -ENOMEM; 1597 1598 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 1599 INIT_HLIST_HEAD(&dp->ports[i]); 1600 1601 return 0; 1602 } 1603 1604 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) 1605 { 1606 struct nlattr **a = info->attrs; 1607 struct vport_parms parms; 1608 struct sk_buff *reply; 1609 struct datapath *dp; 1610 struct vport *vport; 1611 struct ovs_net *ovs_net; 1612 int err; 1613 1614 err = -EINVAL; 1615 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) 1616 goto err; 1617 1618 reply = ovs_dp_cmd_alloc_info(); 1619 if (!reply) 1620 return -ENOMEM; 1621 1622 err = -ENOMEM; 1623 dp = kzalloc(sizeof(*dp), GFP_KERNEL); 1624 if (dp == NULL) 1625 goto err_destroy_reply; 1626 1627 ovs_dp_set_net(dp, sock_net(skb->sk)); 1628 1629 /* Allocate table. */ 1630 err = ovs_flow_tbl_init(&dp->table); 1631 if (err) 1632 goto err_destroy_dp; 1633 1634 err = ovs_dp_stats_init(dp); 1635 if (err) 1636 goto err_destroy_table; 1637 1638 err = ovs_dp_vport_init(dp); 1639 if (err) 1640 goto err_destroy_stats; 1641 1642 err = ovs_meters_init(dp); 1643 if (err) 1644 goto err_destroy_ports; 1645 1646 /* Set up our datapath device. */ 1647 parms.name = nla_data(a[OVS_DP_ATTR_NAME]); 1648 parms.type = OVS_VPORT_TYPE_INTERNAL; 1649 parms.options = NULL; 1650 parms.dp = dp; 1651 parms.port_no = OVSP_LOCAL; 1652 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID]; 1653 1654 err = ovs_dp_change(dp, a); 1655 if (err) 1656 goto err_destroy_meters; 1657 1658 /* So far only local changes have been made, now need the lock. */ 1659 ovs_lock(); 1660 1661 vport = new_vport(&parms); 1662 if (IS_ERR(vport)) { 1663 err = PTR_ERR(vport); 1664 if (err == -EBUSY) 1665 err = -EEXIST; 1666 1667 if (err == -EEXIST) { 1668 /* An outdated user space instance that does not understand 1669 * the concept of user_features has attempted to create a new 1670 * datapath and is likely to reuse it. Drop all user features. 1671 */ 1672 if (info->genlhdr->version < OVS_DP_VER_FEATURES) 1673 ovs_dp_reset_user_features(skb, info); 1674 } 1675 1676 ovs_unlock(); 1677 goto err_destroy_meters; 1678 } 1679 1680 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1681 info->snd_seq, 0, OVS_DP_CMD_NEW); 1682 BUG_ON(err < 0); 1683 1684 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); 1685 list_add_tail_rcu(&dp->list_node, &ovs_net->dps); 1686 1687 ovs_unlock(); 1688 1689 ovs_notify(&dp_datapath_genl_family, reply, info); 1690 return 0; 1691 1692 err_destroy_meters: 1693 ovs_meters_exit(dp); 1694 err_destroy_ports: 1695 kfree(dp->ports); 1696 err_destroy_stats: 1697 free_percpu(dp->stats_percpu); 1698 err_destroy_table: 1699 ovs_flow_tbl_destroy(&dp->table); 1700 err_destroy_dp: 1701 kfree(dp); 1702 err_destroy_reply: 1703 kfree_skb(reply); 1704 err: 1705 return err; 1706 } 1707 1708 /* Called with ovs_mutex. */ 1709 static void __dp_destroy(struct datapath *dp) 1710 { 1711 int i; 1712 1713 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 1714 struct vport *vport; 1715 struct hlist_node *n; 1716 1717 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) 1718 if (vport->port_no != OVSP_LOCAL) 1719 ovs_dp_detach_port(vport); 1720 } 1721 1722 list_del_rcu(&dp->list_node); 1723 1724 /* OVSP_LOCAL is datapath internal port. We need to make sure that 1725 * all ports in datapath are destroyed first before freeing datapath. 1726 */ 1727 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); 1728 1729 /* RCU destroy the flow table */ 1730 call_rcu(&dp->rcu, destroy_dp_rcu); 1731 } 1732 1733 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) 1734 { 1735 struct sk_buff *reply; 1736 struct datapath *dp; 1737 int err; 1738 1739 reply = ovs_dp_cmd_alloc_info(); 1740 if (!reply) 1741 return -ENOMEM; 1742 1743 ovs_lock(); 1744 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1745 err = PTR_ERR(dp); 1746 if (IS_ERR(dp)) 1747 goto err_unlock_free; 1748 1749 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1750 info->snd_seq, 0, OVS_DP_CMD_DEL); 1751 BUG_ON(err < 0); 1752 1753 __dp_destroy(dp); 1754 ovs_unlock(); 1755 1756 ovs_notify(&dp_datapath_genl_family, reply, info); 1757 1758 return 0; 1759 1760 err_unlock_free: 1761 ovs_unlock(); 1762 kfree_skb(reply); 1763 return err; 1764 } 1765 1766 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) 1767 { 1768 struct sk_buff *reply; 1769 struct datapath *dp; 1770 int err; 1771 1772 reply = ovs_dp_cmd_alloc_info(); 1773 if (!reply) 1774 return -ENOMEM; 1775 1776 ovs_lock(); 1777 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1778 err = PTR_ERR(dp); 1779 if (IS_ERR(dp)) 1780 goto err_unlock_free; 1781 1782 err = ovs_dp_change(dp, info->attrs); 1783 if (err) 1784 goto err_unlock_free; 1785 1786 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1787 info->snd_seq, 0, OVS_DP_CMD_SET); 1788 BUG_ON(err < 0); 1789 1790 ovs_unlock(); 1791 ovs_notify(&dp_datapath_genl_family, reply, info); 1792 1793 return 0; 1794 1795 err_unlock_free: 1796 ovs_unlock(); 1797 kfree_skb(reply); 1798 return err; 1799 } 1800 1801 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) 1802 { 1803 struct sk_buff *reply; 1804 struct datapath *dp; 1805 int err; 1806 1807 reply = ovs_dp_cmd_alloc_info(); 1808 if (!reply) 1809 return -ENOMEM; 1810 1811 ovs_lock(); 1812 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1813 if (IS_ERR(dp)) { 1814 err = PTR_ERR(dp); 1815 goto err_unlock_free; 1816 } 1817 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1818 info->snd_seq, 0, OVS_DP_CMD_GET); 1819 BUG_ON(err < 0); 1820 ovs_unlock(); 1821 1822 return genlmsg_reply(reply, info); 1823 1824 err_unlock_free: 1825 ovs_unlock(); 1826 kfree_skb(reply); 1827 return err; 1828 } 1829 1830 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1831 { 1832 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); 1833 struct datapath *dp; 1834 int skip = cb->args[0]; 1835 int i = 0; 1836 1837 ovs_lock(); 1838 list_for_each_entry(dp, &ovs_net->dps, list_node) { 1839 if (i >= skip && 1840 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid, 1841 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1842 OVS_DP_CMD_GET) < 0) 1843 break; 1844 i++; 1845 } 1846 ovs_unlock(); 1847 1848 cb->args[0] = i; 1849 1850 return skb->len; 1851 } 1852 1853 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { 1854 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1855 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1856 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 }, 1857 }; 1858 1859 static const struct genl_ops dp_datapath_genl_ops[] = { 1860 { .cmd = OVS_DP_CMD_NEW, 1861 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1862 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1863 .doit = ovs_dp_cmd_new 1864 }, 1865 { .cmd = OVS_DP_CMD_DEL, 1866 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1867 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1868 .doit = ovs_dp_cmd_del 1869 }, 1870 { .cmd = OVS_DP_CMD_GET, 1871 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1872 .flags = 0, /* OK for unprivileged users. */ 1873 .doit = ovs_dp_cmd_get, 1874 .dumpit = ovs_dp_cmd_dump 1875 }, 1876 { .cmd = OVS_DP_CMD_SET, 1877 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1878 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1879 .doit = ovs_dp_cmd_set, 1880 }, 1881 }; 1882 1883 static struct genl_family dp_datapath_genl_family __ro_after_init = { 1884 .hdrsize = sizeof(struct ovs_header), 1885 .name = OVS_DATAPATH_FAMILY, 1886 .version = OVS_DATAPATH_VERSION, 1887 .maxattr = OVS_DP_ATTR_MAX, 1888 .policy = datapath_policy, 1889 .netnsok = true, 1890 .parallel_ops = true, 1891 .ops = dp_datapath_genl_ops, 1892 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops), 1893 .mcgrps = &ovs_dp_datapath_multicast_group, 1894 .n_mcgrps = 1, 1895 .module = THIS_MODULE, 1896 }; 1897 1898 /* Called with ovs_mutex or RCU read lock. */ 1899 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, 1900 struct net *net, u32 portid, u32 seq, 1901 u32 flags, u8 cmd, gfp_t gfp) 1902 { 1903 struct ovs_header *ovs_header; 1904 struct ovs_vport_stats vport_stats; 1905 int err; 1906 1907 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family, 1908 flags, cmd); 1909 if (!ovs_header) 1910 return -EMSGSIZE; 1911 1912 ovs_header->dp_ifindex = get_dpifindex(vport->dp); 1913 1914 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || 1915 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || 1916 nla_put_string(skb, OVS_VPORT_ATTR_NAME, 1917 ovs_vport_name(vport)) || 1918 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex)) 1919 goto nla_put_failure; 1920 1921 if (!net_eq(net, dev_net(vport->dev))) { 1922 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp); 1923 1924 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id)) 1925 goto nla_put_failure; 1926 } 1927 1928 ovs_vport_get_stats(vport, &vport_stats); 1929 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS, 1930 sizeof(struct ovs_vport_stats), &vport_stats, 1931 OVS_VPORT_ATTR_PAD)) 1932 goto nla_put_failure; 1933 1934 if (ovs_vport_get_upcall_portids(vport, skb)) 1935 goto nla_put_failure; 1936 1937 err = ovs_vport_get_options(vport, skb); 1938 if (err == -EMSGSIZE) 1939 goto error; 1940 1941 genlmsg_end(skb, ovs_header); 1942 return 0; 1943 1944 nla_put_failure: 1945 err = -EMSGSIZE; 1946 error: 1947 genlmsg_cancel(skb, ovs_header); 1948 return err; 1949 } 1950 1951 static struct sk_buff *ovs_vport_cmd_alloc_info(void) 1952 { 1953 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1954 } 1955 1956 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */ 1957 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net, 1958 u32 portid, u32 seq, u8 cmd) 1959 { 1960 struct sk_buff *skb; 1961 int retval; 1962 1963 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1964 if (!skb) 1965 return ERR_PTR(-ENOMEM); 1966 1967 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd, 1968 GFP_KERNEL); 1969 BUG_ON(retval < 0); 1970 1971 return skb; 1972 } 1973 1974 /* Called with ovs_mutex or RCU read lock. */ 1975 static struct vport *lookup_vport(struct net *net, 1976 const struct ovs_header *ovs_header, 1977 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) 1978 { 1979 struct datapath *dp; 1980 struct vport *vport; 1981 1982 if (a[OVS_VPORT_ATTR_IFINDEX]) 1983 return ERR_PTR(-EOPNOTSUPP); 1984 if (a[OVS_VPORT_ATTR_NAME]) { 1985 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); 1986 if (!vport) 1987 return ERR_PTR(-ENODEV); 1988 if (ovs_header->dp_ifindex && 1989 ovs_header->dp_ifindex != get_dpifindex(vport->dp)) 1990 return ERR_PTR(-ENODEV); 1991 return vport; 1992 } else if (a[OVS_VPORT_ATTR_PORT_NO]) { 1993 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1994 1995 if (port_no >= DP_MAX_PORTS) 1996 return ERR_PTR(-EFBIG); 1997 1998 dp = get_dp(net, ovs_header->dp_ifindex); 1999 if (!dp) 2000 return ERR_PTR(-ENODEV); 2001 2002 vport = ovs_vport_ovsl_rcu(dp, port_no); 2003 if (!vport) 2004 return ERR_PTR(-ENODEV); 2005 return vport; 2006 } else 2007 return ERR_PTR(-EINVAL); 2008 2009 } 2010 2011 static unsigned int ovs_get_max_headroom(struct datapath *dp) 2012 { 2013 unsigned int dev_headroom, max_headroom = 0; 2014 struct net_device *dev; 2015 struct vport *vport; 2016 int i; 2017 2018 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 2019 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 2020 dev = vport->dev; 2021 dev_headroom = netdev_get_fwd_headroom(dev); 2022 if (dev_headroom > max_headroom) 2023 max_headroom = dev_headroom; 2024 } 2025 } 2026 2027 return max_headroom; 2028 } 2029 2030 /* Called with ovs_mutex */ 2031 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom) 2032 { 2033 struct vport *vport; 2034 int i; 2035 2036 dp->max_headroom = new_headroom; 2037 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 2038 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) 2039 netdev_set_rx_headroom(vport->dev, new_headroom); 2040 } 2041 2042 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) 2043 { 2044 struct nlattr **a = info->attrs; 2045 struct ovs_header *ovs_header = info->userhdr; 2046 struct vport_parms parms; 2047 struct sk_buff *reply; 2048 struct vport *vport; 2049 struct datapath *dp; 2050 unsigned int new_headroom; 2051 u32 port_no; 2052 int err; 2053 2054 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || 2055 !a[OVS_VPORT_ATTR_UPCALL_PID]) 2056 return -EINVAL; 2057 if (a[OVS_VPORT_ATTR_IFINDEX]) 2058 return -EOPNOTSUPP; 2059 2060 port_no = a[OVS_VPORT_ATTR_PORT_NO] 2061 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0; 2062 if (port_no >= DP_MAX_PORTS) 2063 return -EFBIG; 2064 2065 reply = ovs_vport_cmd_alloc_info(); 2066 if (!reply) 2067 return -ENOMEM; 2068 2069 ovs_lock(); 2070 restart: 2071 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 2072 err = -ENODEV; 2073 if (!dp) 2074 goto exit_unlock_free; 2075 2076 if (port_no) { 2077 vport = ovs_vport_ovsl(dp, port_no); 2078 err = -EBUSY; 2079 if (vport) 2080 goto exit_unlock_free; 2081 } else { 2082 for (port_no = 1; ; port_no++) { 2083 if (port_no >= DP_MAX_PORTS) { 2084 err = -EFBIG; 2085 goto exit_unlock_free; 2086 } 2087 vport = ovs_vport_ovsl(dp, port_no); 2088 if (!vport) 2089 break; 2090 } 2091 } 2092 2093 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); 2094 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); 2095 parms.options = a[OVS_VPORT_ATTR_OPTIONS]; 2096 parms.dp = dp; 2097 parms.port_no = port_no; 2098 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2099 2100 vport = new_vport(&parms); 2101 err = PTR_ERR(vport); 2102 if (IS_ERR(vport)) { 2103 if (err == -EAGAIN) 2104 goto restart; 2105 goto exit_unlock_free; 2106 } 2107 2108 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2109 info->snd_portid, info->snd_seq, 0, 2110 OVS_VPORT_CMD_NEW, GFP_KERNEL); 2111 2112 new_headroom = netdev_get_fwd_headroom(vport->dev); 2113 2114 if (new_headroom > dp->max_headroom) 2115 ovs_update_headroom(dp, new_headroom); 2116 else 2117 netdev_set_rx_headroom(vport->dev, dp->max_headroom); 2118 2119 BUG_ON(err < 0); 2120 ovs_unlock(); 2121 2122 ovs_notify(&dp_vport_genl_family, reply, info); 2123 return 0; 2124 2125 exit_unlock_free: 2126 ovs_unlock(); 2127 kfree_skb(reply); 2128 return err; 2129 } 2130 2131 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) 2132 { 2133 struct nlattr **a = info->attrs; 2134 struct sk_buff *reply; 2135 struct vport *vport; 2136 int err; 2137 2138 reply = ovs_vport_cmd_alloc_info(); 2139 if (!reply) 2140 return -ENOMEM; 2141 2142 ovs_lock(); 2143 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2144 err = PTR_ERR(vport); 2145 if (IS_ERR(vport)) 2146 goto exit_unlock_free; 2147 2148 if (a[OVS_VPORT_ATTR_TYPE] && 2149 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) { 2150 err = -EINVAL; 2151 goto exit_unlock_free; 2152 } 2153 2154 if (a[OVS_VPORT_ATTR_OPTIONS]) { 2155 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); 2156 if (err) 2157 goto exit_unlock_free; 2158 } 2159 2160 2161 if (a[OVS_VPORT_ATTR_UPCALL_PID]) { 2162 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2163 2164 err = ovs_vport_set_upcall_portids(vport, ids); 2165 if (err) 2166 goto exit_unlock_free; 2167 } 2168 2169 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2170 info->snd_portid, info->snd_seq, 0, 2171 OVS_VPORT_CMD_SET, GFP_KERNEL); 2172 BUG_ON(err < 0); 2173 2174 ovs_unlock(); 2175 ovs_notify(&dp_vport_genl_family, reply, info); 2176 return 0; 2177 2178 exit_unlock_free: 2179 ovs_unlock(); 2180 kfree_skb(reply); 2181 return err; 2182 } 2183 2184 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) 2185 { 2186 bool update_headroom = false; 2187 struct nlattr **a = info->attrs; 2188 struct sk_buff *reply; 2189 struct datapath *dp; 2190 struct vport *vport; 2191 unsigned int new_headroom; 2192 int err; 2193 2194 reply = ovs_vport_cmd_alloc_info(); 2195 if (!reply) 2196 return -ENOMEM; 2197 2198 ovs_lock(); 2199 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2200 err = PTR_ERR(vport); 2201 if (IS_ERR(vport)) 2202 goto exit_unlock_free; 2203 2204 if (vport->port_no == OVSP_LOCAL) { 2205 err = -EINVAL; 2206 goto exit_unlock_free; 2207 } 2208 2209 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2210 info->snd_portid, info->snd_seq, 0, 2211 OVS_VPORT_CMD_DEL, GFP_KERNEL); 2212 BUG_ON(err < 0); 2213 2214 /* the vport deletion may trigger dp headroom update */ 2215 dp = vport->dp; 2216 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom) 2217 update_headroom = true; 2218 2219 netdev_reset_rx_headroom(vport->dev); 2220 ovs_dp_detach_port(vport); 2221 2222 if (update_headroom) { 2223 new_headroom = ovs_get_max_headroom(dp); 2224 2225 if (new_headroom < dp->max_headroom) 2226 ovs_update_headroom(dp, new_headroom); 2227 } 2228 ovs_unlock(); 2229 2230 ovs_notify(&dp_vport_genl_family, reply, info); 2231 return 0; 2232 2233 exit_unlock_free: 2234 ovs_unlock(); 2235 kfree_skb(reply); 2236 return err; 2237 } 2238 2239 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) 2240 { 2241 struct nlattr **a = info->attrs; 2242 struct ovs_header *ovs_header = info->userhdr; 2243 struct sk_buff *reply; 2244 struct vport *vport; 2245 int err; 2246 2247 reply = ovs_vport_cmd_alloc_info(); 2248 if (!reply) 2249 return -ENOMEM; 2250 2251 rcu_read_lock(); 2252 vport = lookup_vport(sock_net(skb->sk), ovs_header, a); 2253 err = PTR_ERR(vport); 2254 if (IS_ERR(vport)) 2255 goto exit_unlock_free; 2256 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2257 info->snd_portid, info->snd_seq, 0, 2258 OVS_VPORT_CMD_GET, GFP_ATOMIC); 2259 BUG_ON(err < 0); 2260 rcu_read_unlock(); 2261 2262 return genlmsg_reply(reply, info); 2263 2264 exit_unlock_free: 2265 rcu_read_unlock(); 2266 kfree_skb(reply); 2267 return err; 2268 } 2269 2270 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 2271 { 2272 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 2273 struct datapath *dp; 2274 int bucket = cb->args[0], skip = cb->args[1]; 2275 int i, j = 0; 2276 2277 rcu_read_lock(); 2278 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 2279 if (!dp) { 2280 rcu_read_unlock(); 2281 return -ENODEV; 2282 } 2283 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { 2284 struct vport *vport; 2285 2286 j = 0; 2287 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 2288 if (j >= skip && 2289 ovs_vport_cmd_fill_info(vport, skb, 2290 sock_net(skb->sk), 2291 NETLINK_CB(cb->skb).portid, 2292 cb->nlh->nlmsg_seq, 2293 NLM_F_MULTI, 2294 OVS_VPORT_CMD_GET, 2295 GFP_ATOMIC) < 0) 2296 goto out; 2297 2298 j++; 2299 } 2300 skip = 0; 2301 } 2302 out: 2303 rcu_read_unlock(); 2304 2305 cb->args[0] = i; 2306 cb->args[1] = j; 2307 2308 return skb->len; 2309 } 2310 2311 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { 2312 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 2313 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, 2314 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, 2315 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, 2316 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC }, 2317 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, 2318 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 }, 2319 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 }, 2320 }; 2321 2322 static const struct genl_ops dp_vport_genl_ops[] = { 2323 { .cmd = OVS_VPORT_CMD_NEW, 2324 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2325 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2326 .doit = ovs_vport_cmd_new 2327 }, 2328 { .cmd = OVS_VPORT_CMD_DEL, 2329 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2330 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2331 .doit = ovs_vport_cmd_del 2332 }, 2333 { .cmd = OVS_VPORT_CMD_GET, 2334 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2335 .flags = 0, /* OK for unprivileged users. */ 2336 .doit = ovs_vport_cmd_get, 2337 .dumpit = ovs_vport_cmd_dump 2338 }, 2339 { .cmd = OVS_VPORT_CMD_SET, 2340 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 2341 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2342 .doit = ovs_vport_cmd_set, 2343 }, 2344 }; 2345 2346 struct genl_family dp_vport_genl_family __ro_after_init = { 2347 .hdrsize = sizeof(struct ovs_header), 2348 .name = OVS_VPORT_FAMILY, 2349 .version = OVS_VPORT_VERSION, 2350 .maxattr = OVS_VPORT_ATTR_MAX, 2351 .policy = vport_policy, 2352 .netnsok = true, 2353 .parallel_ops = true, 2354 .ops = dp_vport_genl_ops, 2355 .n_ops = ARRAY_SIZE(dp_vport_genl_ops), 2356 .mcgrps = &ovs_dp_vport_multicast_group, 2357 .n_mcgrps = 1, 2358 .module = THIS_MODULE, 2359 }; 2360 2361 static struct genl_family * const dp_genl_families[] = { 2362 &dp_datapath_genl_family, 2363 &dp_vport_genl_family, 2364 &dp_flow_genl_family, 2365 &dp_packet_genl_family, 2366 &dp_meter_genl_family, 2367 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 2368 &dp_ct_limit_genl_family, 2369 #endif 2370 }; 2371 2372 static void dp_unregister_genl(int n_families) 2373 { 2374 int i; 2375 2376 for (i = 0; i < n_families; i++) 2377 genl_unregister_family(dp_genl_families[i]); 2378 } 2379 2380 static int __init dp_register_genl(void) 2381 { 2382 int err; 2383 int i; 2384 2385 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { 2386 2387 err = genl_register_family(dp_genl_families[i]); 2388 if (err) 2389 goto error; 2390 } 2391 2392 return 0; 2393 2394 error: 2395 dp_unregister_genl(i); 2396 return err; 2397 } 2398 2399 static int __net_init ovs_init_net(struct net *net) 2400 { 2401 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2402 2403 INIT_LIST_HEAD(&ovs_net->dps); 2404 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq); 2405 return ovs_ct_init(net); 2406 } 2407 2408 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet, 2409 struct list_head *head) 2410 { 2411 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2412 struct datapath *dp; 2413 2414 list_for_each_entry(dp, &ovs_net->dps, list_node) { 2415 int i; 2416 2417 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 2418 struct vport *vport; 2419 2420 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) { 2421 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL) 2422 continue; 2423 2424 if (dev_net(vport->dev) == dnet) 2425 list_add(&vport->detach_list, head); 2426 } 2427 } 2428 } 2429 } 2430 2431 static void __net_exit ovs_exit_net(struct net *dnet) 2432 { 2433 struct datapath *dp, *dp_next; 2434 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id); 2435 struct vport *vport, *vport_next; 2436 struct net *net; 2437 LIST_HEAD(head); 2438 2439 ovs_ct_exit(dnet); 2440 ovs_lock(); 2441 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) 2442 __dp_destroy(dp); 2443 2444 down_read(&net_rwsem); 2445 for_each_net(net) 2446 list_vports_from_net(net, dnet, &head); 2447 up_read(&net_rwsem); 2448 2449 /* Detach all vports from given namespace. */ 2450 list_for_each_entry_safe(vport, vport_next, &head, detach_list) { 2451 list_del(&vport->detach_list); 2452 ovs_dp_detach_port(vport); 2453 } 2454 2455 ovs_unlock(); 2456 2457 cancel_work_sync(&ovs_net->dp_notify_work); 2458 } 2459 2460 static struct pernet_operations ovs_net_ops = { 2461 .init = ovs_init_net, 2462 .exit = ovs_exit_net, 2463 .id = &ovs_net_id, 2464 .size = sizeof(struct ovs_net), 2465 }; 2466 2467 static int __init dp_init(void) 2468 { 2469 int err; 2470 2471 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb)); 2472 2473 pr_info("Open vSwitch switching datapath\n"); 2474 2475 err = action_fifos_init(); 2476 if (err) 2477 goto error; 2478 2479 err = ovs_internal_dev_rtnl_link_register(); 2480 if (err) 2481 goto error_action_fifos_exit; 2482 2483 err = ovs_flow_init(); 2484 if (err) 2485 goto error_unreg_rtnl_link; 2486 2487 err = ovs_vport_init(); 2488 if (err) 2489 goto error_flow_exit; 2490 2491 err = register_pernet_device(&ovs_net_ops); 2492 if (err) 2493 goto error_vport_exit; 2494 2495 err = register_netdevice_notifier(&ovs_dp_device_notifier); 2496 if (err) 2497 goto error_netns_exit; 2498 2499 err = ovs_netdev_init(); 2500 if (err) 2501 goto error_unreg_notifier; 2502 2503 err = dp_register_genl(); 2504 if (err < 0) 2505 goto error_unreg_netdev; 2506 2507 return 0; 2508 2509 error_unreg_netdev: 2510 ovs_netdev_exit(); 2511 error_unreg_notifier: 2512 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2513 error_netns_exit: 2514 unregister_pernet_device(&ovs_net_ops); 2515 error_vport_exit: 2516 ovs_vport_exit(); 2517 error_flow_exit: 2518 ovs_flow_exit(); 2519 error_unreg_rtnl_link: 2520 ovs_internal_dev_rtnl_link_unregister(); 2521 error_action_fifos_exit: 2522 action_fifos_exit(); 2523 error: 2524 return err; 2525 } 2526 2527 static void dp_cleanup(void) 2528 { 2529 dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); 2530 ovs_netdev_exit(); 2531 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2532 unregister_pernet_device(&ovs_net_ops); 2533 rcu_barrier(); 2534 ovs_vport_exit(); 2535 ovs_flow_exit(); 2536 ovs_internal_dev_rtnl_link_unregister(); 2537 action_fifos_exit(); 2538 } 2539 2540 module_init(dp_init); 2541 module_exit(dp_cleanup); 2542 2543 MODULE_DESCRIPTION("Open vSwitch switching datapath"); 2544 MODULE_LICENSE("GPL"); 2545 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY); 2546 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY); 2547 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY); 2548 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY); 2549 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY); 2550 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY); 2551