1 /* 2 * Copyright (c) 2007-2014 Nicira, Inc. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of version 2 of the GNU General Public 6 * License as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, but 9 * WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public License 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 16 * 02110-1301, USA 17 */ 18 19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 20 21 #include <linux/init.h> 22 #include <linux/module.h> 23 #include <linux/if_arp.h> 24 #include <linux/if_vlan.h> 25 #include <linux/in.h> 26 #include <linux/ip.h> 27 #include <linux/jhash.h> 28 #include <linux/delay.h> 29 #include <linux/time.h> 30 #include <linux/etherdevice.h> 31 #include <linux/genetlink.h> 32 #include <linux/kernel.h> 33 #include <linux/kthread.h> 34 #include <linux/mutex.h> 35 #include <linux/percpu.h> 36 #include <linux/rcupdate.h> 37 #include <linux/tcp.h> 38 #include <linux/udp.h> 39 #include <linux/ethtool.h> 40 #include <linux/wait.h> 41 #include <asm/div64.h> 42 #include <linux/highmem.h> 43 #include <linux/netfilter_bridge.h> 44 #include <linux/netfilter_ipv4.h> 45 #include <linux/inetdevice.h> 46 #include <linux/list.h> 47 #include <linux/openvswitch.h> 48 #include <linux/rculist.h> 49 #include <linux/dmi.h> 50 #include <net/genetlink.h> 51 #include <net/net_namespace.h> 52 #include <net/netns/generic.h> 53 54 #include "datapath.h" 55 #include "flow.h" 56 #include "flow_table.h" 57 #include "flow_netlink.h" 58 #include "meter.h" 59 #include "vport-internal_dev.h" 60 #include "vport-netdev.h" 61 62 unsigned int ovs_net_id __read_mostly; 63 64 static struct genl_family dp_packet_genl_family; 65 static struct genl_family dp_flow_genl_family; 66 static struct genl_family dp_datapath_genl_family; 67 68 static const struct nla_policy flow_policy[]; 69 70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = { 71 .name = OVS_FLOW_MCGROUP, 72 }; 73 74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = { 75 .name = OVS_DATAPATH_MCGROUP, 76 }; 77 78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = { 79 .name = OVS_VPORT_MCGROUP, 80 }; 81 82 /* Check if need to build a reply message. 83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */ 84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info, 85 unsigned int group) 86 { 87 return info->nlhdr->nlmsg_flags & NLM_F_ECHO || 88 genl_has_listeners(family, genl_info_net(info), group); 89 } 90 91 static void ovs_notify(struct genl_family *family, 92 struct sk_buff *skb, struct genl_info *info) 93 { 94 genl_notify(family, skb, info, 0, GFP_KERNEL); 95 } 96 97 /** 98 * DOC: Locking: 99 * 100 * All writes e.g. Writes to device state (add/remove datapath, port, set 101 * operations on vports, etc.), Writes to other state (flow table 102 * modifications, set miscellaneous datapath parameters, etc.) are protected 103 * by ovs_lock. 104 * 105 * Reads are protected by RCU. 106 * 107 * There are a few special cases (mostly stats) that have their own 108 * synchronization but they nest under all of above and don't interact with 109 * each other. 110 * 111 * The RTNL lock nests inside ovs_mutex. 112 */ 113 114 static DEFINE_MUTEX(ovs_mutex); 115 116 void ovs_lock(void) 117 { 118 mutex_lock(&ovs_mutex); 119 } 120 121 void ovs_unlock(void) 122 { 123 mutex_unlock(&ovs_mutex); 124 } 125 126 #ifdef CONFIG_LOCKDEP 127 int lockdep_ovsl_is_held(void) 128 { 129 if (debug_locks) 130 return lockdep_is_held(&ovs_mutex); 131 else 132 return 1; 133 } 134 #endif 135 136 static struct vport *new_vport(const struct vport_parms *); 137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *, 138 const struct sw_flow_key *, 139 const struct dp_upcall_info *, 140 uint32_t cutlen); 141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *, 142 const struct sw_flow_key *, 143 const struct dp_upcall_info *, 144 uint32_t cutlen); 145 146 /* Must be called with rcu_read_lock or ovs_mutex. */ 147 const char *ovs_dp_name(const struct datapath *dp) 148 { 149 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL); 150 return ovs_vport_name(vport); 151 } 152 153 static int get_dpifindex(const struct datapath *dp) 154 { 155 struct vport *local; 156 int ifindex; 157 158 rcu_read_lock(); 159 160 local = ovs_vport_rcu(dp, OVSP_LOCAL); 161 if (local) 162 ifindex = local->dev->ifindex; 163 else 164 ifindex = 0; 165 166 rcu_read_unlock(); 167 168 return ifindex; 169 } 170 171 static void destroy_dp_rcu(struct rcu_head *rcu) 172 { 173 struct datapath *dp = container_of(rcu, struct datapath, rcu); 174 175 ovs_flow_tbl_destroy(&dp->table); 176 free_percpu(dp->stats_percpu); 177 kfree(dp->ports); 178 ovs_meters_exit(dp); 179 kfree(dp); 180 } 181 182 static struct hlist_head *vport_hash_bucket(const struct datapath *dp, 183 u16 port_no) 184 { 185 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; 186 } 187 188 /* Called with ovs_mutex or RCU read lock. */ 189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no) 190 { 191 struct vport *vport; 192 struct hlist_head *head; 193 194 head = vport_hash_bucket(dp, port_no); 195 hlist_for_each_entry_rcu(vport, head, dp_hash_node) { 196 if (vport->port_no == port_no) 197 return vport; 198 } 199 return NULL; 200 } 201 202 /* Called with ovs_mutex. */ 203 static struct vport *new_vport(const struct vport_parms *parms) 204 { 205 struct vport *vport; 206 207 vport = ovs_vport_add(parms); 208 if (!IS_ERR(vport)) { 209 struct datapath *dp = parms->dp; 210 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no); 211 212 hlist_add_head_rcu(&vport->dp_hash_node, head); 213 } 214 return vport; 215 } 216 217 void ovs_dp_detach_port(struct vport *p) 218 { 219 ASSERT_OVSL(); 220 221 /* First drop references to device. */ 222 hlist_del_rcu(&p->dp_hash_node); 223 224 /* Then destroy it. */ 225 ovs_vport_del(p); 226 } 227 228 /* Must be called with rcu_read_lock. */ 229 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key) 230 { 231 const struct vport *p = OVS_CB(skb)->input_vport; 232 struct datapath *dp = p->dp; 233 struct sw_flow *flow; 234 struct sw_flow_actions *sf_acts; 235 struct dp_stats_percpu *stats; 236 u64 *stats_counter; 237 u32 n_mask_hit; 238 239 stats = this_cpu_ptr(dp->stats_percpu); 240 241 /* Look up flow. */ 242 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit); 243 if (unlikely(!flow)) { 244 struct dp_upcall_info upcall; 245 int error; 246 247 memset(&upcall, 0, sizeof(upcall)); 248 upcall.cmd = OVS_PACKET_CMD_MISS; 249 upcall.portid = ovs_vport_find_upcall_portid(p, skb); 250 upcall.mru = OVS_CB(skb)->mru; 251 error = ovs_dp_upcall(dp, skb, key, &upcall, 0); 252 if (unlikely(error)) 253 kfree_skb(skb); 254 else 255 consume_skb(skb); 256 stats_counter = &stats->n_missed; 257 goto out; 258 } 259 260 ovs_flow_stats_update(flow, key->tp.flags, skb); 261 sf_acts = rcu_dereference(flow->sf_acts); 262 ovs_execute_actions(dp, skb, sf_acts, key); 263 264 stats_counter = &stats->n_hit; 265 266 out: 267 /* Update datapath statistics. */ 268 u64_stats_update_begin(&stats->syncp); 269 (*stats_counter)++; 270 stats->n_mask_hit += n_mask_hit; 271 u64_stats_update_end(&stats->syncp); 272 } 273 274 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, 275 const struct sw_flow_key *key, 276 const struct dp_upcall_info *upcall_info, 277 uint32_t cutlen) 278 { 279 struct dp_stats_percpu *stats; 280 int err; 281 282 if (upcall_info->portid == 0) { 283 err = -ENOTCONN; 284 goto err; 285 } 286 287 if (!skb_is_gso(skb)) 288 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 289 else 290 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen); 291 if (err) 292 goto err; 293 294 return 0; 295 296 err: 297 stats = this_cpu_ptr(dp->stats_percpu); 298 299 u64_stats_update_begin(&stats->syncp); 300 stats->n_lost++; 301 u64_stats_update_end(&stats->syncp); 302 303 return err; 304 } 305 306 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb, 307 const struct sw_flow_key *key, 308 const struct dp_upcall_info *upcall_info, 309 uint32_t cutlen) 310 { 311 unsigned int gso_type = skb_shinfo(skb)->gso_type; 312 struct sw_flow_key later_key; 313 struct sk_buff *segs, *nskb; 314 int err; 315 316 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET); 317 segs = __skb_gso_segment(skb, NETIF_F_SG, false); 318 if (IS_ERR(segs)) 319 return PTR_ERR(segs); 320 if (segs == NULL) 321 return -EINVAL; 322 323 if (gso_type & SKB_GSO_UDP) { 324 /* The initial flow key extracted by ovs_flow_key_extract() 325 * in this case is for a first fragment, so we need to 326 * properly mark later fragments. 327 */ 328 later_key = *key; 329 later_key.ip.frag = OVS_FRAG_TYPE_LATER; 330 } 331 332 /* Queue all of the segments. */ 333 skb = segs; 334 do { 335 if (gso_type & SKB_GSO_UDP && skb != segs) 336 key = &later_key; 337 338 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen); 339 if (err) 340 break; 341 342 } while ((skb = skb->next)); 343 344 /* Free all of the segments. */ 345 skb = segs; 346 do { 347 nskb = skb->next; 348 if (err) 349 kfree_skb(skb); 350 else 351 consume_skb(skb); 352 } while ((skb = nskb)); 353 return err; 354 } 355 356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info, 357 unsigned int hdrlen, int actions_attrlen) 358 { 359 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header)) 360 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */ 361 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */ 362 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */ 363 364 /* OVS_PACKET_ATTR_USERDATA */ 365 if (upcall_info->userdata) 366 size += NLA_ALIGN(upcall_info->userdata->nla_len); 367 368 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */ 369 if (upcall_info->egress_tun_info) 370 size += nla_total_size(ovs_tun_key_attr_size()); 371 372 /* OVS_PACKET_ATTR_ACTIONS */ 373 if (upcall_info->actions_len) 374 size += nla_total_size(actions_attrlen); 375 376 /* OVS_PACKET_ATTR_MRU */ 377 if (upcall_info->mru) 378 size += nla_total_size(sizeof(upcall_info->mru)); 379 380 return size; 381 } 382 383 static void pad_packet(struct datapath *dp, struct sk_buff *skb) 384 { 385 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) { 386 size_t plen = NLA_ALIGN(skb->len) - skb->len; 387 388 if (plen > 0) 389 skb_put_zero(skb, plen); 390 } 391 } 392 393 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb, 394 const struct sw_flow_key *key, 395 const struct dp_upcall_info *upcall_info, 396 uint32_t cutlen) 397 { 398 struct ovs_header *upcall; 399 struct sk_buff *nskb = NULL; 400 struct sk_buff *user_skb = NULL; /* to be queued to userspace */ 401 struct nlattr *nla; 402 size_t len; 403 unsigned int hlen; 404 int err, dp_ifindex; 405 406 dp_ifindex = get_dpifindex(dp); 407 if (!dp_ifindex) 408 return -ENODEV; 409 410 if (skb_vlan_tag_present(skb)) { 411 nskb = skb_clone(skb, GFP_ATOMIC); 412 if (!nskb) 413 return -ENOMEM; 414 415 nskb = __vlan_hwaccel_push_inside(nskb); 416 if (!nskb) 417 return -ENOMEM; 418 419 skb = nskb; 420 } 421 422 if (nla_attr_size(skb->len) > USHRT_MAX) { 423 err = -EFBIG; 424 goto out; 425 } 426 427 /* Complete checksum if needed */ 428 if (skb->ip_summed == CHECKSUM_PARTIAL && 429 (err = skb_csum_hwoffload_help(skb, 0))) 430 goto out; 431 432 /* Older versions of OVS user space enforce alignment of the last 433 * Netlink attribute to NLA_ALIGNTO which would require extensive 434 * padding logic. Only perform zerocopy if padding is not required. 435 */ 436 if (dp->user_features & OVS_DP_F_UNALIGNED) 437 hlen = skb_zerocopy_headlen(skb); 438 else 439 hlen = skb->len; 440 441 len = upcall_msg_size(upcall_info, hlen - cutlen, 442 OVS_CB(skb)->acts_origlen); 443 user_skb = genlmsg_new(len, GFP_ATOMIC); 444 if (!user_skb) { 445 err = -ENOMEM; 446 goto out; 447 } 448 449 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 450 0, upcall_info->cmd); 451 if (!upcall) { 452 err = -EINVAL; 453 goto out; 454 } 455 upcall->dp_ifindex = dp_ifindex; 456 457 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb); 458 BUG_ON(err); 459 460 if (upcall_info->userdata) 461 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA, 462 nla_len(upcall_info->userdata), 463 nla_data(upcall_info->userdata)); 464 465 if (upcall_info->egress_tun_info) { 466 nla = nla_nest_start_noflag(user_skb, 467 OVS_PACKET_ATTR_EGRESS_TUN_KEY); 468 if (!nla) { 469 err = -EMSGSIZE; 470 goto out; 471 } 472 err = ovs_nla_put_tunnel_info(user_skb, 473 upcall_info->egress_tun_info); 474 BUG_ON(err); 475 nla_nest_end(user_skb, nla); 476 } 477 478 if (upcall_info->actions_len) { 479 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS); 480 if (!nla) { 481 err = -EMSGSIZE; 482 goto out; 483 } 484 err = ovs_nla_put_actions(upcall_info->actions, 485 upcall_info->actions_len, 486 user_skb); 487 if (!err) 488 nla_nest_end(user_skb, nla); 489 else 490 nla_nest_cancel(user_skb, nla); 491 } 492 493 /* Add OVS_PACKET_ATTR_MRU */ 494 if (upcall_info->mru) { 495 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, 496 upcall_info->mru)) { 497 err = -ENOBUFS; 498 goto out; 499 } 500 pad_packet(dp, user_skb); 501 } 502 503 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */ 504 if (cutlen > 0) { 505 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, 506 skb->len)) { 507 err = -ENOBUFS; 508 goto out; 509 } 510 pad_packet(dp, user_skb); 511 } 512 513 /* Only reserve room for attribute header, packet data is added 514 * in skb_zerocopy() */ 515 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) { 516 err = -ENOBUFS; 517 goto out; 518 } 519 nla->nla_len = nla_attr_size(skb->len - cutlen); 520 521 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen); 522 if (err) 523 goto out; 524 525 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */ 526 pad_packet(dp, user_skb); 527 528 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; 529 530 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid); 531 user_skb = NULL; 532 out: 533 if (err) 534 skb_tx_error(skb); 535 kfree_skb(user_skb); 536 kfree_skb(nskb); 537 return err; 538 } 539 540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) 541 { 542 struct ovs_header *ovs_header = info->userhdr; 543 struct net *net = sock_net(skb->sk); 544 struct nlattr **a = info->attrs; 545 struct sw_flow_actions *acts; 546 struct sk_buff *packet; 547 struct sw_flow *flow; 548 struct sw_flow_actions *sf_acts; 549 struct datapath *dp; 550 struct vport *input_vport; 551 u16 mru = 0; 552 int len; 553 int err; 554 bool log = !a[OVS_PACKET_ATTR_PROBE]; 555 556 err = -EINVAL; 557 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || 558 !a[OVS_PACKET_ATTR_ACTIONS]) 559 goto err; 560 561 len = nla_len(a[OVS_PACKET_ATTR_PACKET]); 562 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); 563 err = -ENOMEM; 564 if (!packet) 565 goto err; 566 skb_reserve(packet, NET_IP_ALIGN); 567 568 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len); 569 570 /* Set packet's mru */ 571 if (a[OVS_PACKET_ATTR_MRU]) { 572 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]); 573 packet->ignore_df = 1; 574 } 575 OVS_CB(packet)->mru = mru; 576 577 /* Build an sw_flow for sending this packet. */ 578 flow = ovs_flow_alloc(); 579 err = PTR_ERR(flow); 580 if (IS_ERR(flow)) 581 goto err_kfree_skb; 582 583 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY], 584 packet, &flow->key, log); 585 if (err) 586 goto err_flow_free; 587 588 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS], 589 &flow->key, &acts, log); 590 if (err) 591 goto err_flow_free; 592 593 rcu_assign_pointer(flow->sf_acts, acts); 594 packet->priority = flow->key.phy.priority; 595 packet->mark = flow->key.phy.skb_mark; 596 597 rcu_read_lock(); 598 dp = get_dp_rcu(net, ovs_header->dp_ifindex); 599 err = -ENODEV; 600 if (!dp) 601 goto err_unlock; 602 603 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port); 604 if (!input_vport) 605 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL); 606 607 if (!input_vport) 608 goto err_unlock; 609 610 packet->dev = input_vport->dev; 611 OVS_CB(packet)->input_vport = input_vport; 612 sf_acts = rcu_dereference(flow->sf_acts); 613 614 local_bh_disable(); 615 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key); 616 local_bh_enable(); 617 rcu_read_unlock(); 618 619 ovs_flow_free(flow, false); 620 return err; 621 622 err_unlock: 623 rcu_read_unlock(); 624 err_flow_free: 625 ovs_flow_free(flow, false); 626 err_kfree_skb: 627 kfree_skb(packet); 628 err: 629 return err; 630 } 631 632 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { 633 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN }, 634 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, 635 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, 636 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG }, 637 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 }, 638 }; 639 640 static const struct genl_ops dp_packet_genl_ops[] = { 641 { .cmd = OVS_PACKET_CMD_EXECUTE, 642 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 643 .doit = ovs_packet_cmd_execute 644 } 645 }; 646 647 static struct genl_family dp_packet_genl_family __ro_after_init = { 648 .hdrsize = sizeof(struct ovs_header), 649 .name = OVS_PACKET_FAMILY, 650 .version = OVS_PACKET_VERSION, 651 .maxattr = OVS_PACKET_ATTR_MAX, 652 .policy = packet_policy, 653 .netnsok = true, 654 .parallel_ops = true, 655 .ops = dp_packet_genl_ops, 656 .n_ops = ARRAY_SIZE(dp_packet_genl_ops), 657 .module = THIS_MODULE, 658 }; 659 660 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats, 661 struct ovs_dp_megaflow_stats *mega_stats) 662 { 663 int i; 664 665 memset(mega_stats, 0, sizeof(*mega_stats)); 666 667 stats->n_flows = ovs_flow_tbl_count(&dp->table); 668 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table); 669 670 stats->n_hit = stats->n_missed = stats->n_lost = 0; 671 672 for_each_possible_cpu(i) { 673 const struct dp_stats_percpu *percpu_stats; 674 struct dp_stats_percpu local_stats; 675 unsigned int start; 676 677 percpu_stats = per_cpu_ptr(dp->stats_percpu, i); 678 679 do { 680 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp); 681 local_stats = *percpu_stats; 682 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start)); 683 684 stats->n_hit += local_stats.n_hit; 685 stats->n_missed += local_stats.n_missed; 686 stats->n_lost += local_stats.n_lost; 687 mega_stats->n_mask_hit += local_stats.n_mask_hit; 688 } 689 } 690 691 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags) 692 { 693 return ovs_identifier_is_ufid(sfid) && 694 !(ufid_flags & OVS_UFID_F_OMIT_KEY); 695 } 696 697 static bool should_fill_mask(uint32_t ufid_flags) 698 { 699 return !(ufid_flags & OVS_UFID_F_OMIT_MASK); 700 } 701 702 static bool should_fill_actions(uint32_t ufid_flags) 703 { 704 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS); 705 } 706 707 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts, 708 const struct sw_flow_id *sfid, 709 uint32_t ufid_flags) 710 { 711 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header)); 712 713 /* OVS_FLOW_ATTR_UFID */ 714 if (sfid && ovs_identifier_is_ufid(sfid)) 715 len += nla_total_size(sfid->ufid_len); 716 717 /* OVS_FLOW_ATTR_KEY */ 718 if (!sfid || should_fill_key(sfid, ufid_flags)) 719 len += nla_total_size(ovs_key_attr_size()); 720 721 /* OVS_FLOW_ATTR_MASK */ 722 if (should_fill_mask(ufid_flags)) 723 len += nla_total_size(ovs_key_attr_size()); 724 725 /* OVS_FLOW_ATTR_ACTIONS */ 726 if (should_fill_actions(ufid_flags)) 727 len += nla_total_size(acts->orig_len); 728 729 return len 730 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */ 731 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */ 732 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */ 733 } 734 735 /* Called with ovs_mutex or RCU read lock. */ 736 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow, 737 struct sk_buff *skb) 738 { 739 struct ovs_flow_stats stats; 740 __be16 tcp_flags; 741 unsigned long used; 742 743 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags); 744 745 if (used && 746 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used), 747 OVS_FLOW_ATTR_PAD)) 748 return -EMSGSIZE; 749 750 if (stats.n_packets && 751 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS, 752 sizeof(struct ovs_flow_stats), &stats, 753 OVS_FLOW_ATTR_PAD)) 754 return -EMSGSIZE; 755 756 if ((u8)ntohs(tcp_flags) && 757 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags))) 758 return -EMSGSIZE; 759 760 return 0; 761 } 762 763 /* Called with ovs_mutex or RCU read lock. */ 764 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow, 765 struct sk_buff *skb, int skb_orig_len) 766 { 767 struct nlattr *start; 768 int err; 769 770 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if 771 * this is the first flow to be dumped into 'skb'. This is unusual for 772 * Netlink but individual action lists can be longer than 773 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. 774 * The userspace caller can always fetch the actions separately if it 775 * really wants them. (Most userspace callers in fact don't care.) 776 * 777 * This can only fail for dump operations because the skb is always 778 * properly sized for single flows. 779 */ 780 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS); 781 if (start) { 782 const struct sw_flow_actions *sf_acts; 783 784 sf_acts = rcu_dereference_ovsl(flow->sf_acts); 785 err = ovs_nla_put_actions(sf_acts->actions, 786 sf_acts->actions_len, skb); 787 788 if (!err) 789 nla_nest_end(skb, start); 790 else { 791 if (skb_orig_len) 792 return err; 793 794 nla_nest_cancel(skb, start); 795 } 796 } else if (skb_orig_len) { 797 return -EMSGSIZE; 798 } 799 800 return 0; 801 } 802 803 /* Called with ovs_mutex or RCU read lock. */ 804 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex, 805 struct sk_buff *skb, u32 portid, 806 u32 seq, u32 flags, u8 cmd, u32 ufid_flags) 807 { 808 const int skb_orig_len = skb->len; 809 struct ovs_header *ovs_header; 810 int err; 811 812 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, 813 flags, cmd); 814 if (!ovs_header) 815 return -EMSGSIZE; 816 817 ovs_header->dp_ifindex = dp_ifindex; 818 819 err = ovs_nla_put_identifier(flow, skb); 820 if (err) 821 goto error; 822 823 if (should_fill_key(&flow->id, ufid_flags)) { 824 err = ovs_nla_put_masked_key(flow, skb); 825 if (err) 826 goto error; 827 } 828 829 if (should_fill_mask(ufid_flags)) { 830 err = ovs_nla_put_mask(flow, skb); 831 if (err) 832 goto error; 833 } 834 835 err = ovs_flow_cmd_fill_stats(flow, skb); 836 if (err) 837 goto error; 838 839 if (should_fill_actions(ufid_flags)) { 840 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len); 841 if (err) 842 goto error; 843 } 844 845 genlmsg_end(skb, ovs_header); 846 return 0; 847 848 error: 849 genlmsg_cancel(skb, ovs_header); 850 return err; 851 } 852 853 /* May not be called with RCU read lock. */ 854 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts, 855 const struct sw_flow_id *sfid, 856 struct genl_info *info, 857 bool always, 858 uint32_t ufid_flags) 859 { 860 struct sk_buff *skb; 861 size_t len; 862 863 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0)) 864 return NULL; 865 866 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags); 867 skb = genlmsg_new(len, GFP_KERNEL); 868 if (!skb) 869 return ERR_PTR(-ENOMEM); 870 871 return skb; 872 } 873 874 /* Called with ovs_mutex. */ 875 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow, 876 int dp_ifindex, 877 struct genl_info *info, u8 cmd, 878 bool always, u32 ufid_flags) 879 { 880 struct sk_buff *skb; 881 int retval; 882 883 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), 884 &flow->id, info, always, ufid_flags); 885 if (IS_ERR_OR_NULL(skb)) 886 return skb; 887 888 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb, 889 info->snd_portid, info->snd_seq, 0, 890 cmd, ufid_flags); 891 BUG_ON(retval < 0); 892 return skb; 893 } 894 895 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info) 896 { 897 struct net *net = sock_net(skb->sk); 898 struct nlattr **a = info->attrs; 899 struct ovs_header *ovs_header = info->userhdr; 900 struct sw_flow *flow = NULL, *new_flow; 901 struct sw_flow_mask mask; 902 struct sk_buff *reply; 903 struct datapath *dp; 904 struct sw_flow_actions *acts; 905 struct sw_flow_match match; 906 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 907 int error; 908 bool log = !a[OVS_FLOW_ATTR_PROBE]; 909 910 /* Must have key and actions. */ 911 error = -EINVAL; 912 if (!a[OVS_FLOW_ATTR_KEY]) { 913 OVS_NLERR(log, "Flow key attr not present in new flow."); 914 goto error; 915 } 916 if (!a[OVS_FLOW_ATTR_ACTIONS]) { 917 OVS_NLERR(log, "Flow actions attr not present in new flow."); 918 goto error; 919 } 920 921 /* Most of the time we need to allocate a new flow, do it before 922 * locking. 923 */ 924 new_flow = ovs_flow_alloc(); 925 if (IS_ERR(new_flow)) { 926 error = PTR_ERR(new_flow); 927 goto error; 928 } 929 930 /* Extract key. */ 931 ovs_match_init(&match, &new_flow->key, false, &mask); 932 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 933 a[OVS_FLOW_ATTR_MASK], log); 934 if (error) 935 goto err_kfree_flow; 936 937 /* Extract flow identifier. */ 938 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID], 939 &new_flow->key, log); 940 if (error) 941 goto err_kfree_flow; 942 943 /* unmasked key is needed to match when ufid is not used. */ 944 if (ovs_identifier_is_key(&new_flow->id)) 945 match.key = new_flow->id.unmasked_key; 946 947 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask); 948 949 /* Validate actions. */ 950 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS], 951 &new_flow->key, &acts, log); 952 if (error) { 953 OVS_NLERR(log, "Flow actions may not be safe on all matching packets."); 954 goto err_kfree_flow; 955 } 956 957 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false, 958 ufid_flags); 959 if (IS_ERR(reply)) { 960 error = PTR_ERR(reply); 961 goto err_kfree_acts; 962 } 963 964 ovs_lock(); 965 dp = get_dp(net, ovs_header->dp_ifindex); 966 if (unlikely(!dp)) { 967 error = -ENODEV; 968 goto err_unlock_ovs; 969 } 970 971 /* Check if this is a duplicate flow */ 972 if (ovs_identifier_is_ufid(&new_flow->id)) 973 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id); 974 if (!flow) 975 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key); 976 if (likely(!flow)) { 977 rcu_assign_pointer(new_flow->sf_acts, acts); 978 979 /* Put flow in bucket. */ 980 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask); 981 if (unlikely(error)) { 982 acts = NULL; 983 goto err_unlock_ovs; 984 } 985 986 if (unlikely(reply)) { 987 error = ovs_flow_cmd_fill_info(new_flow, 988 ovs_header->dp_ifindex, 989 reply, info->snd_portid, 990 info->snd_seq, 0, 991 OVS_FLOW_CMD_NEW, 992 ufid_flags); 993 BUG_ON(error < 0); 994 } 995 ovs_unlock(); 996 } else { 997 struct sw_flow_actions *old_acts; 998 999 /* Bail out if we're not allowed to modify an existing flow. 1000 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL 1001 * because Generic Netlink treats the latter as a dump 1002 * request. We also accept NLM_F_EXCL in case that bug ever 1003 * gets fixed. 1004 */ 1005 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE 1006 | NLM_F_EXCL))) { 1007 error = -EEXIST; 1008 goto err_unlock_ovs; 1009 } 1010 /* The flow identifier has to be the same for flow updates. 1011 * Look for any overlapping flow. 1012 */ 1013 if (unlikely(!ovs_flow_cmp(flow, &match))) { 1014 if (ovs_identifier_is_key(&flow->id)) 1015 flow = ovs_flow_tbl_lookup_exact(&dp->table, 1016 &match); 1017 else /* UFID matches but key is different */ 1018 flow = NULL; 1019 if (!flow) { 1020 error = -ENOENT; 1021 goto err_unlock_ovs; 1022 } 1023 } 1024 /* Update actions. */ 1025 old_acts = ovsl_dereference(flow->sf_acts); 1026 rcu_assign_pointer(flow->sf_acts, acts); 1027 1028 if (unlikely(reply)) { 1029 error = ovs_flow_cmd_fill_info(flow, 1030 ovs_header->dp_ifindex, 1031 reply, info->snd_portid, 1032 info->snd_seq, 0, 1033 OVS_FLOW_CMD_NEW, 1034 ufid_flags); 1035 BUG_ON(error < 0); 1036 } 1037 ovs_unlock(); 1038 1039 ovs_nla_free_flow_actions_rcu(old_acts); 1040 ovs_flow_free(new_flow, false); 1041 } 1042 1043 if (reply) 1044 ovs_notify(&dp_flow_genl_family, reply, info); 1045 return 0; 1046 1047 err_unlock_ovs: 1048 ovs_unlock(); 1049 kfree_skb(reply); 1050 err_kfree_acts: 1051 ovs_nla_free_flow_actions(acts); 1052 err_kfree_flow: 1053 ovs_flow_free(new_flow, false); 1054 error: 1055 return error; 1056 } 1057 1058 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */ 1059 static struct sw_flow_actions *get_flow_actions(struct net *net, 1060 const struct nlattr *a, 1061 const struct sw_flow_key *key, 1062 const struct sw_flow_mask *mask, 1063 bool log) 1064 { 1065 struct sw_flow_actions *acts; 1066 struct sw_flow_key masked_key; 1067 int error; 1068 1069 ovs_flow_mask_key(&masked_key, key, true, mask); 1070 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log); 1071 if (error) { 1072 OVS_NLERR(log, 1073 "Actions may not be safe on all matching packets"); 1074 return ERR_PTR(error); 1075 } 1076 1077 return acts; 1078 } 1079 1080 /* Factor out match-init and action-copy to avoid 1081 * "Wframe-larger-than=1024" warning. Because mask is only 1082 * used to get actions, we new a function to save some 1083 * stack space. 1084 * 1085 * If there are not key and action attrs, we return 0 1086 * directly. In the case, the caller will also not use the 1087 * match as before. If there is action attr, we try to get 1088 * actions and save them to *acts. Before returning from 1089 * the function, we reset the match->mask pointer. Because 1090 * we should not to return match object with dangling reference 1091 * to mask. 1092 * */ 1093 static int ovs_nla_init_match_and_action(struct net *net, 1094 struct sw_flow_match *match, 1095 struct sw_flow_key *key, 1096 struct nlattr **a, 1097 struct sw_flow_actions **acts, 1098 bool log) 1099 { 1100 struct sw_flow_mask mask; 1101 int error = 0; 1102 1103 if (a[OVS_FLOW_ATTR_KEY]) { 1104 ovs_match_init(match, key, true, &mask); 1105 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY], 1106 a[OVS_FLOW_ATTR_MASK], log); 1107 if (error) 1108 goto error; 1109 } 1110 1111 if (a[OVS_FLOW_ATTR_ACTIONS]) { 1112 if (!a[OVS_FLOW_ATTR_KEY]) { 1113 OVS_NLERR(log, 1114 "Flow key attribute not present in set flow."); 1115 error = -EINVAL; 1116 goto error; 1117 } 1118 1119 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key, 1120 &mask, log); 1121 if (IS_ERR(*acts)) { 1122 error = PTR_ERR(*acts); 1123 goto error; 1124 } 1125 } 1126 1127 /* On success, error is 0. */ 1128 error: 1129 match->mask = NULL; 1130 return error; 1131 } 1132 1133 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info) 1134 { 1135 struct net *net = sock_net(skb->sk); 1136 struct nlattr **a = info->attrs; 1137 struct ovs_header *ovs_header = info->userhdr; 1138 struct sw_flow_key key; 1139 struct sw_flow *flow; 1140 struct sk_buff *reply = NULL; 1141 struct datapath *dp; 1142 struct sw_flow_actions *old_acts = NULL, *acts = NULL; 1143 struct sw_flow_match match; 1144 struct sw_flow_id sfid; 1145 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1146 int error = 0; 1147 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1148 bool ufid_present; 1149 1150 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log); 1151 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) { 1152 OVS_NLERR(log, 1153 "Flow set message rejected, Key attribute missing."); 1154 return -EINVAL; 1155 } 1156 1157 error = ovs_nla_init_match_and_action(net, &match, &key, a, 1158 &acts, log); 1159 if (error) 1160 goto error; 1161 1162 if (acts) { 1163 /* Can allocate before locking if have acts. */ 1164 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false, 1165 ufid_flags); 1166 if (IS_ERR(reply)) { 1167 error = PTR_ERR(reply); 1168 goto err_kfree_acts; 1169 } 1170 } 1171 1172 ovs_lock(); 1173 dp = get_dp(net, ovs_header->dp_ifindex); 1174 if (unlikely(!dp)) { 1175 error = -ENODEV; 1176 goto err_unlock_ovs; 1177 } 1178 /* Check that the flow exists. */ 1179 if (ufid_present) 1180 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid); 1181 else 1182 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1183 if (unlikely(!flow)) { 1184 error = -ENOENT; 1185 goto err_unlock_ovs; 1186 } 1187 1188 /* Update actions, if present. */ 1189 if (likely(acts)) { 1190 old_acts = ovsl_dereference(flow->sf_acts); 1191 rcu_assign_pointer(flow->sf_acts, acts); 1192 1193 if (unlikely(reply)) { 1194 error = ovs_flow_cmd_fill_info(flow, 1195 ovs_header->dp_ifindex, 1196 reply, info->snd_portid, 1197 info->snd_seq, 0, 1198 OVS_FLOW_CMD_SET, 1199 ufid_flags); 1200 BUG_ON(error < 0); 1201 } 1202 } else { 1203 /* Could not alloc without acts before locking. */ 1204 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, 1205 info, OVS_FLOW_CMD_SET, false, 1206 ufid_flags); 1207 1208 if (IS_ERR(reply)) { 1209 error = PTR_ERR(reply); 1210 goto err_unlock_ovs; 1211 } 1212 } 1213 1214 /* Clear stats. */ 1215 if (a[OVS_FLOW_ATTR_CLEAR]) 1216 ovs_flow_stats_clear(flow); 1217 ovs_unlock(); 1218 1219 if (reply) 1220 ovs_notify(&dp_flow_genl_family, reply, info); 1221 if (old_acts) 1222 ovs_nla_free_flow_actions_rcu(old_acts); 1223 1224 return 0; 1225 1226 err_unlock_ovs: 1227 ovs_unlock(); 1228 kfree_skb(reply); 1229 err_kfree_acts: 1230 ovs_nla_free_flow_actions(acts); 1231 error: 1232 return error; 1233 } 1234 1235 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) 1236 { 1237 struct nlattr **a = info->attrs; 1238 struct ovs_header *ovs_header = info->userhdr; 1239 struct net *net = sock_net(skb->sk); 1240 struct sw_flow_key key; 1241 struct sk_buff *reply; 1242 struct sw_flow *flow; 1243 struct datapath *dp; 1244 struct sw_flow_match match; 1245 struct sw_flow_id ufid; 1246 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1247 int err = 0; 1248 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1249 bool ufid_present; 1250 1251 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1252 if (a[OVS_FLOW_ATTR_KEY]) { 1253 ovs_match_init(&match, &key, true, NULL); 1254 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL, 1255 log); 1256 } else if (!ufid_present) { 1257 OVS_NLERR(log, 1258 "Flow get message rejected, Key attribute missing."); 1259 err = -EINVAL; 1260 } 1261 if (err) 1262 return err; 1263 1264 ovs_lock(); 1265 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1266 if (!dp) { 1267 err = -ENODEV; 1268 goto unlock; 1269 } 1270 1271 if (ufid_present) 1272 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1273 else 1274 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1275 if (!flow) { 1276 err = -ENOENT; 1277 goto unlock; 1278 } 1279 1280 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info, 1281 OVS_FLOW_CMD_GET, true, ufid_flags); 1282 if (IS_ERR(reply)) { 1283 err = PTR_ERR(reply); 1284 goto unlock; 1285 } 1286 1287 ovs_unlock(); 1288 return genlmsg_reply(reply, info); 1289 unlock: 1290 ovs_unlock(); 1291 return err; 1292 } 1293 1294 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) 1295 { 1296 struct nlattr **a = info->attrs; 1297 struct ovs_header *ovs_header = info->userhdr; 1298 struct net *net = sock_net(skb->sk); 1299 struct sw_flow_key key; 1300 struct sk_buff *reply; 1301 struct sw_flow *flow = NULL; 1302 struct datapath *dp; 1303 struct sw_flow_match match; 1304 struct sw_flow_id ufid; 1305 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1306 int err; 1307 bool log = !a[OVS_FLOW_ATTR_PROBE]; 1308 bool ufid_present; 1309 1310 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log); 1311 if (a[OVS_FLOW_ATTR_KEY]) { 1312 ovs_match_init(&match, &key, true, NULL); 1313 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], 1314 NULL, log); 1315 if (unlikely(err)) 1316 return err; 1317 } 1318 1319 ovs_lock(); 1320 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 1321 if (unlikely(!dp)) { 1322 err = -ENODEV; 1323 goto unlock; 1324 } 1325 1326 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) { 1327 err = ovs_flow_tbl_flush(&dp->table); 1328 goto unlock; 1329 } 1330 1331 if (ufid_present) 1332 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid); 1333 else 1334 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); 1335 if (unlikely(!flow)) { 1336 err = -ENOENT; 1337 goto unlock; 1338 } 1339 1340 ovs_flow_tbl_remove(&dp->table, flow); 1341 ovs_unlock(); 1342 1343 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts, 1344 &flow->id, info, false, ufid_flags); 1345 if (likely(reply)) { 1346 if (likely(!IS_ERR(reply))) { 1347 rcu_read_lock(); /*To keep RCU checker happy. */ 1348 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, 1349 reply, info->snd_portid, 1350 info->snd_seq, 0, 1351 OVS_FLOW_CMD_DEL, 1352 ufid_flags); 1353 rcu_read_unlock(); 1354 BUG_ON(err < 0); 1355 1356 ovs_notify(&dp_flow_genl_family, reply, info); 1357 } else { 1358 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply)); 1359 } 1360 } 1361 1362 ovs_flow_free(flow, true); 1363 return 0; 1364 unlock: 1365 ovs_unlock(); 1366 return err; 1367 } 1368 1369 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1370 { 1371 struct nlattr *a[__OVS_FLOW_ATTR_MAX]; 1372 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 1373 struct table_instance *ti; 1374 struct datapath *dp; 1375 u32 ufid_flags; 1376 int err; 1377 1378 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a, 1379 OVS_FLOW_ATTR_MAX, flow_policy, NULL); 1380 if (err) 1381 return err; 1382 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]); 1383 1384 rcu_read_lock(); 1385 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 1386 if (!dp) { 1387 rcu_read_unlock(); 1388 return -ENODEV; 1389 } 1390 1391 ti = rcu_dereference(dp->table.ti); 1392 for (;;) { 1393 struct sw_flow *flow; 1394 u32 bucket, obj; 1395 1396 bucket = cb->args[0]; 1397 obj = cb->args[1]; 1398 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj); 1399 if (!flow) 1400 break; 1401 1402 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb, 1403 NETLINK_CB(cb->skb).portid, 1404 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1405 OVS_FLOW_CMD_GET, ufid_flags) < 0) 1406 break; 1407 1408 cb->args[0] = bucket; 1409 cb->args[1] = obj; 1410 } 1411 rcu_read_unlock(); 1412 return skb->len; 1413 } 1414 1415 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { 1416 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, 1417 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED }, 1418 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, 1419 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, 1420 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG }, 1421 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 }, 1422 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 }, 1423 }; 1424 1425 static const struct genl_ops dp_flow_genl_ops[] = { 1426 { .cmd = OVS_FLOW_CMD_NEW, 1427 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1428 .doit = ovs_flow_cmd_new 1429 }, 1430 { .cmd = OVS_FLOW_CMD_DEL, 1431 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1432 .doit = ovs_flow_cmd_del 1433 }, 1434 { .cmd = OVS_FLOW_CMD_GET, 1435 .flags = 0, /* OK for unprivileged users. */ 1436 .doit = ovs_flow_cmd_get, 1437 .dumpit = ovs_flow_cmd_dump 1438 }, 1439 { .cmd = OVS_FLOW_CMD_SET, 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 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[]) 1550 { 1551 if (a[OVS_DP_ATTR_USER_FEATURES]) 1552 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); 1553 } 1554 1555 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) 1556 { 1557 struct nlattr **a = info->attrs; 1558 struct vport_parms parms; 1559 struct sk_buff *reply; 1560 struct datapath *dp; 1561 struct vport *vport; 1562 struct ovs_net *ovs_net; 1563 int err, i; 1564 1565 err = -EINVAL; 1566 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) 1567 goto err; 1568 1569 reply = ovs_dp_cmd_alloc_info(); 1570 if (!reply) 1571 return -ENOMEM; 1572 1573 err = -ENOMEM; 1574 dp = kzalloc(sizeof(*dp), GFP_KERNEL); 1575 if (dp == NULL) 1576 goto err_free_reply; 1577 1578 ovs_dp_set_net(dp, sock_net(skb->sk)); 1579 1580 /* Allocate table. */ 1581 err = ovs_flow_tbl_init(&dp->table); 1582 if (err) 1583 goto err_free_dp; 1584 1585 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu); 1586 if (!dp->stats_percpu) { 1587 err = -ENOMEM; 1588 goto err_destroy_table; 1589 } 1590 1591 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS, 1592 sizeof(struct hlist_head), 1593 GFP_KERNEL); 1594 if (!dp->ports) { 1595 err = -ENOMEM; 1596 goto err_destroy_percpu; 1597 } 1598 1599 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 1600 INIT_HLIST_HEAD(&dp->ports[i]); 1601 1602 err = ovs_meters_init(dp); 1603 if (err) 1604 goto err_destroy_ports_array; 1605 1606 /* Set up our datapath device. */ 1607 parms.name = nla_data(a[OVS_DP_ATTR_NAME]); 1608 parms.type = OVS_VPORT_TYPE_INTERNAL; 1609 parms.options = NULL; 1610 parms.dp = dp; 1611 parms.port_no = OVSP_LOCAL; 1612 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID]; 1613 1614 ovs_dp_change(dp, a); 1615 1616 /* So far only local changes have been made, now need the lock. */ 1617 ovs_lock(); 1618 1619 vport = new_vport(&parms); 1620 if (IS_ERR(vport)) { 1621 err = PTR_ERR(vport); 1622 if (err == -EBUSY) 1623 err = -EEXIST; 1624 1625 if (err == -EEXIST) { 1626 /* An outdated user space instance that does not understand 1627 * the concept of user_features has attempted to create a new 1628 * datapath and is likely to reuse it. Drop all user features. 1629 */ 1630 if (info->genlhdr->version < OVS_DP_VER_FEATURES) 1631 ovs_dp_reset_user_features(skb, info); 1632 } 1633 1634 goto err_destroy_meters; 1635 } 1636 1637 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1638 info->snd_seq, 0, OVS_DP_CMD_NEW); 1639 BUG_ON(err < 0); 1640 1641 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); 1642 list_add_tail_rcu(&dp->list_node, &ovs_net->dps); 1643 1644 ovs_unlock(); 1645 1646 ovs_notify(&dp_datapath_genl_family, reply, info); 1647 return 0; 1648 1649 err_destroy_meters: 1650 ovs_unlock(); 1651 ovs_meters_exit(dp); 1652 err_destroy_ports_array: 1653 kfree(dp->ports); 1654 err_destroy_percpu: 1655 free_percpu(dp->stats_percpu); 1656 err_destroy_table: 1657 ovs_flow_tbl_destroy(&dp->table); 1658 err_free_dp: 1659 kfree(dp); 1660 err_free_reply: 1661 kfree_skb(reply); 1662 err: 1663 return err; 1664 } 1665 1666 /* Called with ovs_mutex. */ 1667 static void __dp_destroy(struct datapath *dp) 1668 { 1669 int i; 1670 1671 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 1672 struct vport *vport; 1673 struct hlist_node *n; 1674 1675 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) 1676 if (vport->port_no != OVSP_LOCAL) 1677 ovs_dp_detach_port(vport); 1678 } 1679 1680 list_del_rcu(&dp->list_node); 1681 1682 /* OVSP_LOCAL is datapath internal port. We need to make sure that 1683 * all ports in datapath are destroyed first before freeing datapath. 1684 */ 1685 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); 1686 1687 /* RCU destroy the flow table */ 1688 call_rcu(&dp->rcu, destroy_dp_rcu); 1689 } 1690 1691 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) 1692 { 1693 struct sk_buff *reply; 1694 struct datapath *dp; 1695 int err; 1696 1697 reply = ovs_dp_cmd_alloc_info(); 1698 if (!reply) 1699 return -ENOMEM; 1700 1701 ovs_lock(); 1702 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1703 err = PTR_ERR(dp); 1704 if (IS_ERR(dp)) 1705 goto err_unlock_free; 1706 1707 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1708 info->snd_seq, 0, OVS_DP_CMD_DEL); 1709 BUG_ON(err < 0); 1710 1711 __dp_destroy(dp); 1712 ovs_unlock(); 1713 1714 ovs_notify(&dp_datapath_genl_family, reply, info); 1715 1716 return 0; 1717 1718 err_unlock_free: 1719 ovs_unlock(); 1720 kfree_skb(reply); 1721 return err; 1722 } 1723 1724 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) 1725 { 1726 struct sk_buff *reply; 1727 struct datapath *dp; 1728 int err; 1729 1730 reply = ovs_dp_cmd_alloc_info(); 1731 if (!reply) 1732 return -ENOMEM; 1733 1734 ovs_lock(); 1735 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1736 err = PTR_ERR(dp); 1737 if (IS_ERR(dp)) 1738 goto err_unlock_free; 1739 1740 ovs_dp_change(dp, info->attrs); 1741 1742 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1743 info->snd_seq, 0, OVS_DP_CMD_SET); 1744 BUG_ON(err < 0); 1745 1746 ovs_unlock(); 1747 ovs_notify(&dp_datapath_genl_family, reply, info); 1748 1749 return 0; 1750 1751 err_unlock_free: 1752 ovs_unlock(); 1753 kfree_skb(reply); 1754 return err; 1755 } 1756 1757 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) 1758 { 1759 struct sk_buff *reply; 1760 struct datapath *dp; 1761 int err; 1762 1763 reply = ovs_dp_cmd_alloc_info(); 1764 if (!reply) 1765 return -ENOMEM; 1766 1767 ovs_lock(); 1768 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); 1769 if (IS_ERR(dp)) { 1770 err = PTR_ERR(dp); 1771 goto err_unlock_free; 1772 } 1773 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, 1774 info->snd_seq, 0, OVS_DP_CMD_GET); 1775 BUG_ON(err < 0); 1776 ovs_unlock(); 1777 1778 return genlmsg_reply(reply, info); 1779 1780 err_unlock_free: 1781 ovs_unlock(); 1782 kfree_skb(reply); 1783 return err; 1784 } 1785 1786 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 1787 { 1788 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); 1789 struct datapath *dp; 1790 int skip = cb->args[0]; 1791 int i = 0; 1792 1793 ovs_lock(); 1794 list_for_each_entry(dp, &ovs_net->dps, list_node) { 1795 if (i >= skip && 1796 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid, 1797 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1798 OVS_DP_CMD_GET) < 0) 1799 break; 1800 i++; 1801 } 1802 ovs_unlock(); 1803 1804 cb->args[0] = i; 1805 1806 return skb->len; 1807 } 1808 1809 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { 1810 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 1811 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 1812 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 }, 1813 }; 1814 1815 static const struct genl_ops dp_datapath_genl_ops[] = { 1816 { .cmd = OVS_DP_CMD_NEW, 1817 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1818 .doit = ovs_dp_cmd_new 1819 }, 1820 { .cmd = OVS_DP_CMD_DEL, 1821 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1822 .doit = ovs_dp_cmd_del 1823 }, 1824 { .cmd = OVS_DP_CMD_GET, 1825 .flags = 0, /* OK for unprivileged users. */ 1826 .doit = ovs_dp_cmd_get, 1827 .dumpit = ovs_dp_cmd_dump 1828 }, 1829 { .cmd = OVS_DP_CMD_SET, 1830 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 1831 .doit = ovs_dp_cmd_set, 1832 }, 1833 }; 1834 1835 static struct genl_family dp_datapath_genl_family __ro_after_init = { 1836 .hdrsize = sizeof(struct ovs_header), 1837 .name = OVS_DATAPATH_FAMILY, 1838 .version = OVS_DATAPATH_VERSION, 1839 .maxattr = OVS_DP_ATTR_MAX, 1840 .policy = datapath_policy, 1841 .netnsok = true, 1842 .parallel_ops = true, 1843 .ops = dp_datapath_genl_ops, 1844 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops), 1845 .mcgrps = &ovs_dp_datapath_multicast_group, 1846 .n_mcgrps = 1, 1847 .module = THIS_MODULE, 1848 }; 1849 1850 /* Called with ovs_mutex or RCU read lock. */ 1851 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, 1852 struct net *net, u32 portid, u32 seq, 1853 u32 flags, u8 cmd) 1854 { 1855 struct ovs_header *ovs_header; 1856 struct ovs_vport_stats vport_stats; 1857 int err; 1858 1859 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family, 1860 flags, cmd); 1861 if (!ovs_header) 1862 return -EMSGSIZE; 1863 1864 ovs_header->dp_ifindex = get_dpifindex(vport->dp); 1865 1866 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || 1867 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || 1868 nla_put_string(skb, OVS_VPORT_ATTR_NAME, 1869 ovs_vport_name(vport)) || 1870 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex)) 1871 goto nla_put_failure; 1872 1873 if (!net_eq(net, dev_net(vport->dev))) { 1874 int id = peernet2id_alloc(net, dev_net(vport->dev)); 1875 1876 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id)) 1877 goto nla_put_failure; 1878 } 1879 1880 ovs_vport_get_stats(vport, &vport_stats); 1881 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS, 1882 sizeof(struct ovs_vport_stats), &vport_stats, 1883 OVS_VPORT_ATTR_PAD)) 1884 goto nla_put_failure; 1885 1886 if (ovs_vport_get_upcall_portids(vport, skb)) 1887 goto nla_put_failure; 1888 1889 err = ovs_vport_get_options(vport, skb); 1890 if (err == -EMSGSIZE) 1891 goto error; 1892 1893 genlmsg_end(skb, ovs_header); 1894 return 0; 1895 1896 nla_put_failure: 1897 err = -EMSGSIZE; 1898 error: 1899 genlmsg_cancel(skb, ovs_header); 1900 return err; 1901 } 1902 1903 static struct sk_buff *ovs_vport_cmd_alloc_info(void) 1904 { 1905 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1906 } 1907 1908 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */ 1909 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net, 1910 u32 portid, u32 seq, u8 cmd) 1911 { 1912 struct sk_buff *skb; 1913 int retval; 1914 1915 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 1916 if (!skb) 1917 return ERR_PTR(-ENOMEM); 1918 1919 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd); 1920 BUG_ON(retval < 0); 1921 1922 return skb; 1923 } 1924 1925 /* Called with ovs_mutex or RCU read lock. */ 1926 static struct vport *lookup_vport(struct net *net, 1927 const struct ovs_header *ovs_header, 1928 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) 1929 { 1930 struct datapath *dp; 1931 struct vport *vport; 1932 1933 if (a[OVS_VPORT_ATTR_IFINDEX]) 1934 return ERR_PTR(-EOPNOTSUPP); 1935 if (a[OVS_VPORT_ATTR_NAME]) { 1936 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); 1937 if (!vport) 1938 return ERR_PTR(-ENODEV); 1939 if (ovs_header->dp_ifindex && 1940 ovs_header->dp_ifindex != get_dpifindex(vport->dp)) 1941 return ERR_PTR(-ENODEV); 1942 return vport; 1943 } else if (a[OVS_VPORT_ATTR_PORT_NO]) { 1944 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); 1945 1946 if (port_no >= DP_MAX_PORTS) 1947 return ERR_PTR(-EFBIG); 1948 1949 dp = get_dp(net, ovs_header->dp_ifindex); 1950 if (!dp) 1951 return ERR_PTR(-ENODEV); 1952 1953 vport = ovs_vport_ovsl_rcu(dp, port_no); 1954 if (!vport) 1955 return ERR_PTR(-ENODEV); 1956 return vport; 1957 } else 1958 return ERR_PTR(-EINVAL); 1959 1960 } 1961 1962 /* Called with ovs_mutex */ 1963 static void update_headroom(struct datapath *dp) 1964 { 1965 unsigned dev_headroom, max_headroom = 0; 1966 struct net_device *dev; 1967 struct vport *vport; 1968 int i; 1969 1970 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 1971 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 1972 dev = vport->dev; 1973 dev_headroom = netdev_get_fwd_headroom(dev); 1974 if (dev_headroom > max_headroom) 1975 max_headroom = dev_headroom; 1976 } 1977 } 1978 1979 dp->max_headroom = max_headroom; 1980 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) 1981 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) 1982 netdev_set_rx_headroom(vport->dev, max_headroom); 1983 } 1984 1985 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) 1986 { 1987 struct nlattr **a = info->attrs; 1988 struct ovs_header *ovs_header = info->userhdr; 1989 struct vport_parms parms; 1990 struct sk_buff *reply; 1991 struct vport *vport; 1992 struct datapath *dp; 1993 u32 port_no; 1994 int err; 1995 1996 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || 1997 !a[OVS_VPORT_ATTR_UPCALL_PID]) 1998 return -EINVAL; 1999 if (a[OVS_VPORT_ATTR_IFINDEX]) 2000 return -EOPNOTSUPP; 2001 2002 port_no = a[OVS_VPORT_ATTR_PORT_NO] 2003 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0; 2004 if (port_no >= DP_MAX_PORTS) 2005 return -EFBIG; 2006 2007 reply = ovs_vport_cmd_alloc_info(); 2008 if (!reply) 2009 return -ENOMEM; 2010 2011 ovs_lock(); 2012 restart: 2013 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); 2014 err = -ENODEV; 2015 if (!dp) 2016 goto exit_unlock_free; 2017 2018 if (port_no) { 2019 vport = ovs_vport_ovsl(dp, port_no); 2020 err = -EBUSY; 2021 if (vport) 2022 goto exit_unlock_free; 2023 } else { 2024 for (port_no = 1; ; port_no++) { 2025 if (port_no >= DP_MAX_PORTS) { 2026 err = -EFBIG; 2027 goto exit_unlock_free; 2028 } 2029 vport = ovs_vport_ovsl(dp, port_no); 2030 if (!vport) 2031 break; 2032 } 2033 } 2034 2035 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); 2036 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); 2037 parms.options = a[OVS_VPORT_ATTR_OPTIONS]; 2038 parms.dp = dp; 2039 parms.port_no = port_no; 2040 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2041 2042 vport = new_vport(&parms); 2043 err = PTR_ERR(vport); 2044 if (IS_ERR(vport)) { 2045 if (err == -EAGAIN) 2046 goto restart; 2047 goto exit_unlock_free; 2048 } 2049 2050 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2051 info->snd_portid, info->snd_seq, 0, 2052 OVS_VPORT_CMD_NEW); 2053 2054 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom) 2055 update_headroom(dp); 2056 else 2057 netdev_set_rx_headroom(vport->dev, dp->max_headroom); 2058 2059 BUG_ON(err < 0); 2060 ovs_unlock(); 2061 2062 ovs_notify(&dp_vport_genl_family, reply, info); 2063 return 0; 2064 2065 exit_unlock_free: 2066 ovs_unlock(); 2067 kfree_skb(reply); 2068 return err; 2069 } 2070 2071 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) 2072 { 2073 struct nlattr **a = info->attrs; 2074 struct sk_buff *reply; 2075 struct vport *vport; 2076 int err; 2077 2078 reply = ovs_vport_cmd_alloc_info(); 2079 if (!reply) 2080 return -ENOMEM; 2081 2082 ovs_lock(); 2083 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2084 err = PTR_ERR(vport); 2085 if (IS_ERR(vport)) 2086 goto exit_unlock_free; 2087 2088 if (a[OVS_VPORT_ATTR_TYPE] && 2089 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) { 2090 err = -EINVAL; 2091 goto exit_unlock_free; 2092 } 2093 2094 if (a[OVS_VPORT_ATTR_OPTIONS]) { 2095 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); 2096 if (err) 2097 goto exit_unlock_free; 2098 } 2099 2100 2101 if (a[OVS_VPORT_ATTR_UPCALL_PID]) { 2102 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID]; 2103 2104 err = ovs_vport_set_upcall_portids(vport, ids); 2105 if (err) 2106 goto exit_unlock_free; 2107 } 2108 2109 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2110 info->snd_portid, info->snd_seq, 0, 2111 OVS_VPORT_CMD_SET); 2112 BUG_ON(err < 0); 2113 2114 ovs_unlock(); 2115 ovs_notify(&dp_vport_genl_family, reply, info); 2116 return 0; 2117 2118 exit_unlock_free: 2119 ovs_unlock(); 2120 kfree_skb(reply); 2121 return err; 2122 } 2123 2124 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) 2125 { 2126 bool must_update_headroom = false; 2127 struct nlattr **a = info->attrs; 2128 struct sk_buff *reply; 2129 struct datapath *dp; 2130 struct vport *vport; 2131 int err; 2132 2133 reply = ovs_vport_cmd_alloc_info(); 2134 if (!reply) 2135 return -ENOMEM; 2136 2137 ovs_lock(); 2138 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); 2139 err = PTR_ERR(vport); 2140 if (IS_ERR(vport)) 2141 goto exit_unlock_free; 2142 2143 if (vport->port_no == OVSP_LOCAL) { 2144 err = -EINVAL; 2145 goto exit_unlock_free; 2146 } 2147 2148 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2149 info->snd_portid, info->snd_seq, 0, 2150 OVS_VPORT_CMD_DEL); 2151 BUG_ON(err < 0); 2152 2153 /* the vport deletion may trigger dp headroom update */ 2154 dp = vport->dp; 2155 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom) 2156 must_update_headroom = true; 2157 netdev_reset_rx_headroom(vport->dev); 2158 ovs_dp_detach_port(vport); 2159 2160 if (must_update_headroom) 2161 update_headroom(dp); 2162 ovs_unlock(); 2163 2164 ovs_notify(&dp_vport_genl_family, reply, info); 2165 return 0; 2166 2167 exit_unlock_free: 2168 ovs_unlock(); 2169 kfree_skb(reply); 2170 return err; 2171 } 2172 2173 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) 2174 { 2175 struct nlattr **a = info->attrs; 2176 struct ovs_header *ovs_header = info->userhdr; 2177 struct sk_buff *reply; 2178 struct vport *vport; 2179 int err; 2180 2181 reply = ovs_vport_cmd_alloc_info(); 2182 if (!reply) 2183 return -ENOMEM; 2184 2185 rcu_read_lock(); 2186 vport = lookup_vport(sock_net(skb->sk), ovs_header, a); 2187 err = PTR_ERR(vport); 2188 if (IS_ERR(vport)) 2189 goto exit_unlock_free; 2190 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info), 2191 info->snd_portid, info->snd_seq, 0, 2192 OVS_VPORT_CMD_GET); 2193 BUG_ON(err < 0); 2194 rcu_read_unlock(); 2195 2196 return genlmsg_reply(reply, info); 2197 2198 exit_unlock_free: 2199 rcu_read_unlock(); 2200 kfree_skb(reply); 2201 return err; 2202 } 2203 2204 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) 2205 { 2206 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); 2207 struct datapath *dp; 2208 int bucket = cb->args[0], skip = cb->args[1]; 2209 int i, j = 0; 2210 2211 rcu_read_lock(); 2212 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); 2213 if (!dp) { 2214 rcu_read_unlock(); 2215 return -ENODEV; 2216 } 2217 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { 2218 struct vport *vport; 2219 2220 j = 0; 2221 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { 2222 if (j >= skip && 2223 ovs_vport_cmd_fill_info(vport, skb, 2224 sock_net(skb->sk), 2225 NETLINK_CB(cb->skb).portid, 2226 cb->nlh->nlmsg_seq, 2227 NLM_F_MULTI, 2228 OVS_VPORT_CMD_GET) < 0) 2229 goto out; 2230 2231 j++; 2232 } 2233 skip = 0; 2234 } 2235 out: 2236 rcu_read_unlock(); 2237 2238 cb->args[0] = i; 2239 cb->args[1] = j; 2240 2241 return skb->len; 2242 } 2243 2244 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { 2245 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, 2246 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, 2247 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, 2248 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, 2249 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, 2250 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, 2251 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 }, 2252 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 }, 2253 }; 2254 2255 static const struct genl_ops dp_vport_genl_ops[] = { 2256 { .cmd = OVS_VPORT_CMD_NEW, 2257 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2258 .doit = ovs_vport_cmd_new 2259 }, 2260 { .cmd = OVS_VPORT_CMD_DEL, 2261 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2262 .doit = ovs_vport_cmd_del 2263 }, 2264 { .cmd = OVS_VPORT_CMD_GET, 2265 .flags = 0, /* OK for unprivileged users. */ 2266 .doit = ovs_vport_cmd_get, 2267 .dumpit = ovs_vport_cmd_dump 2268 }, 2269 { .cmd = OVS_VPORT_CMD_SET, 2270 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ 2271 .doit = ovs_vport_cmd_set, 2272 }, 2273 }; 2274 2275 struct genl_family dp_vport_genl_family __ro_after_init = { 2276 .hdrsize = sizeof(struct ovs_header), 2277 .name = OVS_VPORT_FAMILY, 2278 .version = OVS_VPORT_VERSION, 2279 .maxattr = OVS_VPORT_ATTR_MAX, 2280 .policy = vport_policy, 2281 .netnsok = true, 2282 .parallel_ops = true, 2283 .ops = dp_vport_genl_ops, 2284 .n_ops = ARRAY_SIZE(dp_vport_genl_ops), 2285 .mcgrps = &ovs_dp_vport_multicast_group, 2286 .n_mcgrps = 1, 2287 .module = THIS_MODULE, 2288 }; 2289 2290 static struct genl_family * const dp_genl_families[] = { 2291 &dp_datapath_genl_family, 2292 &dp_vport_genl_family, 2293 &dp_flow_genl_family, 2294 &dp_packet_genl_family, 2295 &dp_meter_genl_family, 2296 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT) 2297 &dp_ct_limit_genl_family, 2298 #endif 2299 }; 2300 2301 static void dp_unregister_genl(int n_families) 2302 { 2303 int i; 2304 2305 for (i = 0; i < n_families; i++) 2306 genl_unregister_family(dp_genl_families[i]); 2307 } 2308 2309 static int __init dp_register_genl(void) 2310 { 2311 int err; 2312 int i; 2313 2314 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { 2315 2316 err = genl_register_family(dp_genl_families[i]); 2317 if (err) 2318 goto error; 2319 } 2320 2321 return 0; 2322 2323 error: 2324 dp_unregister_genl(i); 2325 return err; 2326 } 2327 2328 static int __net_init ovs_init_net(struct net *net) 2329 { 2330 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2331 2332 INIT_LIST_HEAD(&ovs_net->dps); 2333 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq); 2334 return ovs_ct_init(net); 2335 } 2336 2337 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet, 2338 struct list_head *head) 2339 { 2340 struct ovs_net *ovs_net = net_generic(net, ovs_net_id); 2341 struct datapath *dp; 2342 2343 list_for_each_entry(dp, &ovs_net->dps, list_node) { 2344 int i; 2345 2346 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { 2347 struct vport *vport; 2348 2349 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) { 2350 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL) 2351 continue; 2352 2353 if (dev_net(vport->dev) == dnet) 2354 list_add(&vport->detach_list, head); 2355 } 2356 } 2357 } 2358 } 2359 2360 static void __net_exit ovs_exit_net(struct net *dnet) 2361 { 2362 struct datapath *dp, *dp_next; 2363 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id); 2364 struct vport *vport, *vport_next; 2365 struct net *net; 2366 LIST_HEAD(head); 2367 2368 ovs_ct_exit(dnet); 2369 ovs_lock(); 2370 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) 2371 __dp_destroy(dp); 2372 2373 down_read(&net_rwsem); 2374 for_each_net(net) 2375 list_vports_from_net(net, dnet, &head); 2376 up_read(&net_rwsem); 2377 2378 /* Detach all vports from given namespace. */ 2379 list_for_each_entry_safe(vport, vport_next, &head, detach_list) { 2380 list_del(&vport->detach_list); 2381 ovs_dp_detach_port(vport); 2382 } 2383 2384 ovs_unlock(); 2385 2386 cancel_work_sync(&ovs_net->dp_notify_work); 2387 } 2388 2389 static struct pernet_operations ovs_net_ops = { 2390 .init = ovs_init_net, 2391 .exit = ovs_exit_net, 2392 .id = &ovs_net_id, 2393 .size = sizeof(struct ovs_net), 2394 }; 2395 2396 static int __init dp_init(void) 2397 { 2398 int err; 2399 2400 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb)); 2401 2402 pr_info("Open vSwitch switching datapath\n"); 2403 2404 err = action_fifos_init(); 2405 if (err) 2406 goto error; 2407 2408 err = ovs_internal_dev_rtnl_link_register(); 2409 if (err) 2410 goto error_action_fifos_exit; 2411 2412 err = ovs_flow_init(); 2413 if (err) 2414 goto error_unreg_rtnl_link; 2415 2416 err = ovs_vport_init(); 2417 if (err) 2418 goto error_flow_exit; 2419 2420 err = register_pernet_device(&ovs_net_ops); 2421 if (err) 2422 goto error_vport_exit; 2423 2424 err = register_netdevice_notifier(&ovs_dp_device_notifier); 2425 if (err) 2426 goto error_netns_exit; 2427 2428 err = ovs_netdev_init(); 2429 if (err) 2430 goto error_unreg_notifier; 2431 2432 err = dp_register_genl(); 2433 if (err < 0) 2434 goto error_unreg_netdev; 2435 2436 return 0; 2437 2438 error_unreg_netdev: 2439 ovs_netdev_exit(); 2440 error_unreg_notifier: 2441 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2442 error_netns_exit: 2443 unregister_pernet_device(&ovs_net_ops); 2444 error_vport_exit: 2445 ovs_vport_exit(); 2446 error_flow_exit: 2447 ovs_flow_exit(); 2448 error_unreg_rtnl_link: 2449 ovs_internal_dev_rtnl_link_unregister(); 2450 error_action_fifos_exit: 2451 action_fifos_exit(); 2452 error: 2453 return err; 2454 } 2455 2456 static void dp_cleanup(void) 2457 { 2458 dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); 2459 ovs_netdev_exit(); 2460 unregister_netdevice_notifier(&ovs_dp_device_notifier); 2461 unregister_pernet_device(&ovs_net_ops); 2462 rcu_barrier(); 2463 ovs_vport_exit(); 2464 ovs_flow_exit(); 2465 ovs_internal_dev_rtnl_link_unregister(); 2466 action_fifos_exit(); 2467 } 2468 2469 module_init(dp_init); 2470 module_exit(dp_cleanup); 2471 2472 MODULE_DESCRIPTION("Open vSwitch switching datapath"); 2473 MODULE_LICENSE("GPL"); 2474 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY); 2475 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY); 2476 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY); 2477 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY); 2478 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY); 2479 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY); 2480